diff --git a/preprint/preprint__0d397fac72f0ff0c7a01d139e71d10c74fc08d546e5a9dce6d59c01b0601b529/images_list.json b/preprint/preprint__0d397fac72f0ff0c7a01d139e71d10c74fc08d546e5a9dce6d59c01b0601b529/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..1432b6c6a9e1053baa930ef39ea041c1f6adcca1 --- /dev/null +++ b/preprint/preprint__0d397fac72f0ff0c7a01d139e71d10c74fc08d546e5a9dce6d59c01b0601b529/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Schematic representation of the structures obtained under high temperature synthetic conditions (300 \\(^\\circ \\mathrm{C}\\) for 6 hours) for MUV-24( \\(\\mathrm{coi}\\) ), \\(\\mathbf{dg}\\) -MUV-29, and MUV-28. In the diagram, tetrahedral \\(\\mathrm{Fe}^{\\mathrm{II}}\\) centers are illustrated as orange tetrahedra. The imidazolate ligands are shown in light blue, and the benzimidazolate ligands in dark blue. The corresponding PXRD pattern for each material is presented at the bottom, together with the calculated patterns of the crystalline materials.", + "footnote": [], + "bbox": [ + [ + 144, + 490, + 868, + 757 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. a) Thermogravimetric analysis (TGA) and differential scanning calorimetry (DSC) curves of dg-MUV-29-bim \\(_{0.48}\\) b) DSC of the materials obtained via high-temperature synthesis with variable \\(\\mathrm{im}^{-}\\) to \\(\\mathrm{bin}^{-}\\) ratios. In all cases, an initial heating cycle up to \\(200^{\\circ}\\mathrm{C}\\) was conducted to remove organic impurities. c) Optical images of various members of dg-MUV-29.", + "footnote": [], + "bbox": [ + [ + 140, + 342, + 860, + 531 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. a) X-ray PDF in the form of \\(D(\\mathrm{r})\\) of \\(\\mathbf{dg}\\) -MUV-29-bim \\(_{0.48}\\) , \\(\\mathbf{dg}\\) -Fe-MUV-29-bim \\(_{0.18}\\) , \\(\\mathbf{a}_{\\mathbf{g}}\\) -MUV-24 and IMIDFE. b) Schematic representation of the benzimidazolate bridge coordinated to two \\(\\mathrm{Fe^{II}}\\) centers and the connectivity between tetrahedral \\(\\mathrm{Fe^{II}}\\) ions with benzimidazolate linkers. (c-e) Mössbauer spectra of \\(\\mathbf{dg}\\) -MUV-29-bim \\(_{0.48}\\) (c) \\(\\mathbf{a}_{\\mathbf{g}}\\) -MUV-24 (d) and IMIDFE (e) samples, displayed with lines overlaid on the experimental data. These lines represent the sum of doublets and sextets, or distributions of QS (see Table S2), and are slightly shifted for clarity.", + "footnote": [], + "bbox": [ + [ + 145, + 237, + 850, + 760 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. a) Optical images of different \\(\\mathbf{dg}\\) -MUV-29-X monolithic glass. b) X-ray powder diffractograms of \\(\\mathbf{dg}\\) -MUV-29-X. c) DSC first upscan of \\(\\mathbf{dg}\\) -MUV-29-X family. To calculate the \\(T_{g}\\) , the material was preheated to \\(200^{\\circ}\\mathrm{C}\\) to remove impurities. In all cases, the amount of bim incorporated is \\(x = 0.5\\) .", + "footnote": [], + "bbox": [ + [ + 160, + 488, + 808, + 817 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. a) Thermal dependence of magnetic susceptibility in the temperature range 2–300 K of dg-MUV-29. The data have been fitted (red dotted line) following the Curie-Weiss model. b) Temperature dependence of the in-phase dynamic a.c. susceptibility of dg-MUV-29 measured at different frequencies. c) Mossbauer measurements at different temperatures of dg-MUV-29. The lines over the experimental points are the estimated distributions of quadrupole splittings. d) \\(I(Q)\\) data, offset by 5000; dots represent the experimental data of dg-MUV-29, and the line represents the Monte Carlo (MC) simulation. The MC simulation is matched to temperature, with 1.8 K data plotted against 13 K. An empirical scale, offset, and linear background have been applied to all MC data, fixed to the lowest temperature, and distances are scaled by the experimentally determined metal-metal distance \\(r_0 = 6.20 \\text{Å}\\) . e) Real space spin correlation function derived from MC simulations showing antiferromagnetic alternation that decays with increasing distance, \\(r\\) .", + "footnote": [], + "bbox": [ + [ + 128, + 83, + 860, + 426 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. a) Scheme of the perovskite photodetector and b) a photo of the device laminated. c) current vs. applied field under illumination (solid line) and in the dark (dotted line). d) Photocurrent response of the device at an applied field of \\(0.4\\mathrm{V / \\mu m}\\) when exposed to consecutive 40 s light pulses.", + "footnote": [], + "bbox": [ + [ + 230, + 266, + 744, + 556 + ] + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/images_list.json b/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ec832bc2afb059ed113c10fe4ada420794bbb7ba --- /dev/null +++ b/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1| Historical and future deposition rate of BC and dust. a,h Historical (1995-2014) and b,e,i,l future (2081-2100) spatial patterns of aerosol deposition rates and c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP585. d,g,k,n Time series of the average deposition rate of BC and dust for snow-covered regions over the NH where the average SWE from December to May exceeds \\(5\\mathrm{mm}\\) . Historical and future deposition rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May. In (a-c,e-f,h-j,i-m), grids with an average SWE from December to May smaller than \\(5\\mathrm{mm}\\) are masked. In (c,f,j,m), the black dots represent regions with statistically significant trends ( \\(\\mathrm{p< 0.05}\\) ) using the MK test. In (d,g,k,n), the line and background shading represent the mean and standard deviation of deposition rates, respectively, based on the seven CMIP6 models. The p values from the MK test of statistical significance of the temporal trends from 2015-2100 are shown inside each panel; and the vertical dashed line indicates year 2015 when SSP scenarios start. We use \\(\\mathrm{ngm^{-2}s^{-1}}\\) and \\(\\mu \\mathrm{gm}^{-2}\\mathrm{s}^{-1}\\) as BC and dust deposition rate units, respectively.", + "footnote": [], + "bbox": [ + [ + 120, + 241, + 707, + 680 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2| Spatial patterns of historical and future surface radiative forcings (RF) of BC, dust and LAP (the sum of BC and dust) in snow covered regions over the NH. a,d,g Historical (1995-2014) spatial patterns of RF. b,c,e,f,h,i The difference between future (2081-2100) and historical RF under SSP126 and SSP585. Historical and future RF are calculated based on ELM outputs from December to May. In each panel, grids where the average SWE during December to May is smaller than 5 mm are masked.", + "footnote": [], + "bbox": [ + [ + 135, + 90, + 864, + 696 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig 3| Historical and future average surface radiative forcing (RF) induced by LAP and SWE over the snow-covered TP regions (where the average SWE exceeds 5 mm in the historical period of 1995-2014) under different model configurations. a,b. The average RF during December to May under SSP126 and SSP585. c,d April SWE under SSP126 and SSP585. For each panel, Climatehist+LAPhist represents the historical (1995-2014) simulations with historical LAP depositions, while Climatefuture+LAPfuture and Climatefuture+LAPhist represent future (2081-2100) simulations with and without a future change of LAP depositions, respectively. The Climatefuture+LAPhist simulations used the historical average LAP depositions from 1995-2014. The horizontal axis labels represent different model configurations (see Methods), where Control has the ELM default settings and the others represent major adjustments made from the Control case. Specifically, PP assumes that the terrain is flat and neglects topographic effects on solar radiation; Koch assumes a non-spherical snow grain shape (Koch snowflake); extBC assumes external mixing between hydrophilic BC and snow grains; intDust assumes internal mixing between dust and snow grains; noLULCC has no land use and land cover change; MSE_high assumes high melt-water scavenging efficiency (MSE = 2, much higher than the default value of 0.2) of hydrophilic BC; and MSE_low assumes a low MSE (0.02) of hydrophilic BC. In (c,d), the contribution (δLAP) of future LAP change that mitigates snowpack loss under each ELM configuration is noted as a percentage and is calculated as the ratio of the SWE difference (ΔSWELAP) between Climatefuture+LAPfuture and Climatefuture+LAPhist to the SWE difference (ΔSWELimate) between Climatehist+LAPhist and Climatefuture+LAPfuture. The geographical coverage of the TP is shown in Fig. 4.", + "footnote": [], + "bbox": [ + [ + 115, + 117, + 872, + 333 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4| Future SWE in April and contributions of LAP change to future SWE. a Historical (1995-2014) and b,c future (2081-2100) spatial patterns of SWE in April under SSP126 and SSP585. d,e The differences (ΔSWELAP) of future (2081-2100) April SWEs with and without LAP change under SSP126 and SSP585. f,g Time series of relative ΔSWELAP (calculated as the ratio of ΔSWELAP to projected SWE without LAP change) over snow-covered regions (where the average SWE exceeds 5 mm in the historical period) in the TP. In (a-e), grids with an average April SWE smaller than 5 mm in the historical period are masked. In (f,g), BCfuture+Dustfuture, BCfuture+Dusthist, and BChist+Dustfuture represent different combinations of BC and dust depositions, where the subscripts of future and hist represent future and historical average depositions, respectively.", + "footnote": [], + "bbox": [ + [ + 117, + 83, + 880, + 590 + ] + ], + "page_idx": 8 + } +] \ No newline at end of file diff --git a/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc_det.mmd b/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..0f8783f703eb11e40bcdfbacfd18a316ae1fa700 --- /dev/null +++ b/preprint/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc/preprint__0d43f0eca1869c78649d9772ac1e085dccab5fd758ce08115c5a8105cc2bfccc_det.mmd @@ -0,0 +1,466 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 787, 177]]<|/det|> +# A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming + +<|ref|>text<|/ref|><|det|>[[44, 195, 748, 236]]<|/det|> +Dalei Hao ( \(\boxed{\boxed{\pi}}\) dalei.hao@pnnl.gov) Pacific Northwest National Laboratory https://orcid.org/0000- 0001- 7154- 6332 + +<|ref|>text<|/ref|><|det|>[[44, 242, 389, 283]]<|/det|> +Gautam Bisht Pacific Northwest National Laboratory + +<|ref|>text<|/ref|><|det|>[[44, 290, 750, 331]]<|/det|> +Hailong Wang Pacific Northwest National Laboratory https://orcid.org/0000- 0002- 1994- 4402 + +<|ref|>text<|/ref|><|det|>[[44, 336, 750, 377]]<|/det|> +Donghui Xu Pacific Northwest National Laboratory https://orcid.org/0000- 0002- 2859- 2664 + +<|ref|>text<|/ref|><|det|>[[44, 382, 389, 423]]<|/det|> +Huilin Huang Pacific Northwest National Laboratory + +<|ref|>text<|/ref|><|det|>[[44, 429, 750, 470]]<|/det|> +Yun Qian Pacific Northwest National Laboratory https://orcid.org/0000- 0003- 4821- 1934 + +<|ref|>text<|/ref|><|det|>[[44, 475, 750, 516]]<|/det|> +L. Leung Pacific Northwest National Laboratory https://orcid.org/0000- 0002- 3221- 9467 + +<|ref|>text<|/ref|><|det|>[[44, 556, 102, 573]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 594, 137, 612]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 632, 348, 651]]<|/det|> +Posted Date: November 23rd, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 670, 475, 690]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2230945/v1 + +<|ref|>text<|/ref|><|det|>[[44, 707, 910, 750]]<|/det|> +License: \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 768, 531, 788]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 823, 931, 867]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on October 2nd, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 41732- 6. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[72, 90, 845, 133]]<|/det|> +# A cleaner snow future mitigates Northern Hemisphere snowpack loss from warming + +<|ref|>text<|/ref|><|det|>[[72, 150, 88, 163]]<|/det|> +3 + +<|ref|>text<|/ref|><|det|>[[72, 175, 835, 210]]<|/det|> +4 Dalei Hao1\*, Gautam Bisht1, Hailong Wang1, Donghui Xu1, Huilin Huang1, Yun Qian1 and L. Ruby Leung1\* + +<|ref|>text<|/ref|><|det|>[[72, 220, 860, 255]]<|/det|> +6 'Atmospheric Sciences and Global Change Division, Pacific Northwest National Laboratory, Richland, WA, USA + +<|ref|>text<|/ref|><|det|>[[72, 270, 88, 282]]<|/det|> +8 + +<|ref|>text<|/ref|><|det|>[[72, 293, 574, 310]]<|/det|> +9 Correspondence to: dalei.hao@pnnl.gov; ruby.leung@pnnl.gov + +<|ref|>sub_title<|/ref|><|det|>[[68, 350, 185, 365]]<|/det|> +## 11 Abstract + +<|ref|>text<|/ref|><|det|>[[68, 375, 880, 545]]<|/det|> +12 Light- absorbing particles (LAP) such as black carbon and dust deposited on seasonal snowpack can result 13 in snow darkening, earlier snowmelt, and regional climate change. However, the future deposition and 14 surface radiative forcing of LAP in snow and their contributions to snowpack change remain unclear. 15 Here, using Earth System Model simulations, we show significant reduction in black carbon deposition in 16 the future. The reduced deposition decreases the December- May average LAP- induced radiative forcing 17 in snow over the Northern Hemisphere from \(1.3\mathrm{Wm}^{- 2}\) during 1995- 2014 to \(0.65\mathrm{Wm}^{- 2}\) and \(0.49\mathrm{Wm}^{- 2}\) 18 by 2081- 2100 under the SSP126 and SSP585 scenarios, respectively. The reduced black carbon 19 contamination in snow over the Tibetan Plateau will alleviate future snowpack loss due to climate change 20 by \(52.1\pm 8.0\%\) and \(8.0\pm 1.1\%\) for the two scenarios. Our findings highlight a cleaner snow future and its 21 benefits for future water availability from snowmelt. + +<|ref|>sub_title<|/ref|><|det|>[[68, 586, 158, 600]]<|/det|> +## 23 Main + +<|ref|>text<|/ref|><|det|>[[68, 611, 880, 787]]<|/det|> +24 Seasonal snow plays a critical role in Earth's energy budget and water cycle1, with snowmelt over 25 mountainous regions providing an important source of freshwater for two billion people globally2, 3. 26 However, anthropogenic climate change is projected to reduce global snowfall and cause earlier 27 snowmelt4. The high- mountain Asia (HMA), known as the water tower of Asia, has been experiencing an 28 overall decrease in snow water equivalent (SWE) in the last 30 years5, 6, 7. The Western United States 29 (WUS) is also expecting a low- to- no snow future, with a SWE decline of \(\sim 25\%\) by \(2050^{\mathrm{g}}\) . Projections 30 show a continuous decline in SWE for nearly all global high- mountain regions throughout the 21st 31 century under both the low and high Representative Concentration Pathway (i.e., RCP2.6 and RCP8.5) 32 greenhouse gas emission scenarios9. Future snowpack change under global warming will significantly 33 alter downstream runoff as well as the amount and timing10 of freshwater supply from snowmelt for both 34 humans and ecosystems. + +<|ref|>text<|/ref|><|det|>[[68, 803, 875, 905]]<|/det|> +35 Light- absorbing particles (LAP), such as black carbon (BC) and dust, can darken snow surface, reduce 36 snow albedo, and accelerate snowmelt processes11, 12, 13. BC originates from the incomplete burning of 37 fossil fuels, biomass, or bio- fuels14, while dust can be produced from the natural wind erosion of soil or 38 anthropogenic industrial emissions15. Snow darkening due to BC has significantly contributed to climate 39 change and glacier retreats16. Dust was found to have larger darkening impacts than BC over HMA11. 40 However, most climate models used in future snow projection neglect or oversimplify the effects of snow + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 879, 263]]<|/det|> +darkening due to BC and dust deposition9. A clean or dirty snow future could have a different impact on snowmelt and snow water resources. Future BC emissions are projected to decrease in both low and high emission scenarios, but the spatial and temporal patterns are expected to vary among different scenarios17. Under RCP8.5, most regions in the Northern Hemisphere (NH) are projected to show a significant decrease in BC deposition by \(2050^{18}\) . The intensification of human land-use and drought induced by climate change will affect future dust emissions and depositions19. The projected 25 to \(40\%\) loss of biological soil crusts (i.e. biocrusts) will lead to around a \(5 - 15\%\) increase of global dust emission and deposition by \(2070^{20}\) . Understanding the future changes in BC and dust depositions over snow- covered regions and how they affect snowpack is critical for constraining projections of downstream freshwater availability from snowmelt. + +<|ref|>text<|/ref|><|det|>[[112, 272, 875, 479]]<|/det|> +Here we first use simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to analyze the evolution of BC and dust deposition from 2015 to 2100 over the NH. We explore two contrasting Shared Socioeconomic Pathway and Representative Concentration Pathway (SSP- RCP) scenarios (SSP126 and SSP585). We then examine the future spatio- temporal characteristics of LAP mass, snow albedo reduction, and surface radiative forcing (RF, a measure of the net change in surface radiative fluxes due to the change of a forcing agent) induced by BC and dust. We use the state- of- the- art Energy Exascale Earth System Model (E3SM) Land Model (ELM) driven by meteorological and LAP changes simulated by a CMIP6 model (see Methods). We also quantify the contributions of BC and dust evolution to future snowpack change to better understand their implications for future snow water resources. Our analyses mainly focus on December to May because the impacts of LAP on snow are generally larger in mid- latitude mountains than high- latitude regions13 and these regions have no or low snow cover from June to November + +<|ref|>sub_title<|/ref|><|det|>[[113, 489, 448, 506]]<|/det|> +## Future evolution of BC and dust deposition + +<|ref|>text<|/ref|><|det|>[[112, 515, 874, 809]]<|/det|> +Compared to the historical period (1995- 2014), BC depositions over the entire snow- covered regions of the NH are projected to significantly decrease by 2081- 2100 under both SSP126 and SSP585 (Fig. 1a- g). The historical BC deposition rates are higher in relatively lower latitude regions and show the largest values over the Southern border of the Tibetan Plateau (TP) and East Asia (Fig. 1a) because of the significant fossil fuel combustion, traditional biomass usage, and transport- related activity in the Asian regions17, 21. The projected future BC deposition rates exhibit a similar spatial pattern as historical rates, but with significantly smaller magnitudes (Fig. 1b,e). Compared to the historical period, SSP126 shows larger BC deposition rate decreases than SSP585, especially in western Asia and eastern North America. There are significant decreasing trends in BC deposition from 2015 to 2100 over nearly all the snow- covered regions of the NH (Fig. 1c,f and Fig. S1a,b). The average BC deposition rates show a significant decreasing trend (p<0.05 based on the Mann- Kendall (MK) test) from 2015 to 2100, with higher decreasing rates from 2015- 2040 (Fig. 1d). Compared to SSP126, SSP585 also shows a significant decreasing trend (p<0.05 from the MK test) but at a more uniform rate throughout the century (Fig. 1g). The relatively small standard deviations of the seven CMIP6 models that provide the deposition data for both BC and dust indicate similar decreasing trends across the models (shaded regions in Fig. 1d,g). The significant decrease of projected BC deposition can be attributed to socio- economic development and technological progress that have reduced BC emissions worldwide17. + +<|ref|>text<|/ref|><|det|>[[112, 820, 880, 907]]<|/det|> +Compared to BC, overall future dust deposition is projected to be larger under both future scenarios from 2081- 2100 than the historical period, especially over the TP (Fig. 1h- n). The spatial patterns of dust deposition rates are similar under the historical and future scenarios. The northern and western TP regions show large values because of their proximity to drylands, the main source of dust emission22. Most regions show no significant trends in dust deposition under SSP126, while many Asian regions show a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 872, 227]]<|/det|> +significant increasing trend (MK test: \(\mathrm{p< 0.05}\) ) under SSP585 (Fig. S1c,d). The average dust deposition rates over snow- covered NH regions shows large interannual variability and has a small increasing trend (MK test: \(\mathrm{p< 0.05}\) ) from 2015 to 2100 under both scenarios (Fig. 1k,n). The large inter- model differences (shaded regions in Fig. 1k,n) can be attributed to different parameterizations of near- surface winds, soil erodibility, and/or vegetation evolution (prescribed vs dynamic vegetation) as well as diverse treatments of size of emitted dust particles in ESMs \(^{22,23}\) . Despite the different and potentially opposite trends of future BC and dust depositions, the significant reduction of BC deposition is expected to have a larger effect on snow than dust changes. + +<|ref|>image<|/ref|><|det|>[[120, 241, 707, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 688, 875, 896]]<|/det|> +
Fig. 1| Historical and future deposition rate of BC and dust. a,h Historical (1995-2014) and b,e,i,l future (2081-2100) spatial patterns of aerosol deposition rates and c,f,j,m their differences for (a-c,e-f) BC and (h-j,l-m) dust under SSP126 and SSP585. d,g,k,n Time series of the average deposition rate of BC and dust for snow-covered regions over the NH where the average SWE from December to May exceeds \(5\mathrm{mm}\) . Historical and future deposition rates are calculated based on the ensemble mean of seven CMIP6 model outputs from December to May. In (a-c,e-f,h-j,i-m), grids with an average SWE from December to May smaller than \(5\mathrm{mm}\) are masked. In (c,f,j,m), the black dots represent regions with statistically significant trends ( \(\mathrm{p< 0.05}\) ) using the MK test. In (d,g,k,n), the line and background shading represent the mean and standard deviation of deposition rates, respectively, based on the seven CMIP6 models. The p values from the MK test of statistical significance of the temporal trends from 2015-2100 are shown inside each panel; and the vertical dashed line indicates year 2015 when SSP scenarios start. We use \(\mathrm{ngm^{-2}s^{-1}}\) and \(\mu \mathrm{gm}^{-2}\mathrm{s}^{-1}\) as BC and dust deposition rate units, respectively.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 117, 575, 134]]<|/det|> +## Future surface radiative forcings from BC and dust in snow + +<|ref|>text<|/ref|><|det|>[[112, 143, 880, 317]]<|/det|> +We analyze the RF of BC, dust, and LAP (i.e., the sum of BC and dust) in snow simulated by ELM (see Methods). Northern Asia and TP have large RF of BC, dust, and LAP during the historical period (Fig. 2a,d,g). Due to reduced BC deposition, the projected BC RF over the whole NH decreases significantly by 2081- 2100 (Fig. 2a- c) compared to the historical period (Fig. 1). However, the change of dust RF is sensitive to the emission scenarios, with an increase in SSP126 but a decrease in SSP585 (Fig. 2d- f). Due to the dominating change in BC RF, the LAP- induced RF significantly decreases over the NH under both scenarios (Fig. 2g- i; Table S1). Although BC has a larger RF than dust in the historical period, the significant reduction of BC deposition leads to a larger RF of dust than BC in the future (Table S1). These RF changes are consistent with the spatio- temporal distribution of LAP mass in the top snow layer (Fig. S2) and LAP- induced snow albedo reduction (Fig. S3). + +<|ref|>text<|/ref|><|det|>[[111, 326, 881, 704]]<|/det|> +TP, a hotspot of climate change, is projected to experience a continuing decrease in LAP- induced RF from 2015- 2100 (Fig. 3 and Fig. S4; Table S1). In the Control ELM simulations (see Methods), the historical (1995- 2014) average LAP- induced RF over the TP is \(5.1 \mathrm{W m}^{- 2}\) , while the future (2081- 2100) average RF is \(2.4 \mathrm{W m}^{- 2}\) under SSP126 (Fig. 3a). This is likely due to the reduced LAP deposition and snowpack under climate change. Compared to the historical period, assuming no change in future LAP deposition (see Methods) would result in a slightly decreasing average RF over the TP of \(4.1 \mathrm{W m}^{- 2}\) during 2081- 2100 under SSP126 due to the faster snowmelt and reduced snowpack in warmer temperatures. For SSP585, the average LAP- induced RF over the TP with and without future change of LAP deposition are \(1.5 \mathrm{W m}^{- 2}\) and \(2.3 \mathrm{W m}^{- 2}\) , respectively (Fig. 3b). Although ELM model configurations can affect the magnitude of simulated LAP- induced RF, their impacts on the relative difference caused by future LAP changes are small (Fig. 3a,b). Future LAP changes can account for \(60.5 \pm 1.9\%\) and \(21.2 \pm 1.1\%\) of the decrease of future LAP- induced RF relative to the historical period for SSP126 and SSP585, respectively. Temporally, the average BC- induced RF in the ELM simulations shows a significant decreasing trend ( \(p< 0.05\) from the MK test) for both SSP126 and SSP585 scenarios. The trend is reduced under the assumption that future BC remains at the historical level (Fig. S4a). The dust- induced RF in SSP126 shows a slight increasing trend ( \(p< 0.05\) from the MK test) that switches to a slight decreasing trend ( \(p< 0.05\) from the MK test) if future dust deposition remains at the historical value (Fig. S4b). However, dust- induced RF shows a slight decreasing trend ( \(p< 0.05\) from the MK test) in SSP585 that becomes stronger if future dust deposition remains unchanged (Fig. S4b). For both future scenarios, the LAP- induced RF has a significant decreasing trend that is larger than either individual trend from BC or dust (Fig. S4c). The significant reduction of future LAP- induced RF is expected to contribute to a slowdown of future snowpack loss. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 90, 864, 696]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 705, 880, 790]]<|/det|> +
Fig. 2| Spatial patterns of historical and future surface radiative forcings (RF) of BC, dust and LAP (the sum of BC and dust) in snow covered regions over the NH. a,d,g Historical (1995-2014) spatial patterns of RF. b,c,e,f,h,i The difference between future (2081-2100) and historical RF under SSP126 and SSP585. Historical and future RF are calculated based on ELM outputs from December to May. In each panel, grids where the average SWE during December to May is smaller than 5 mm are masked.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 117, 872, 333]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 345, 879, 672]]<|/det|> +
Fig 3| Historical and future average surface radiative forcing (RF) induced by LAP and SWE over the snow-covered TP regions (where the average SWE exceeds 5 mm in the historical period of 1995-2014) under different model configurations. a,b. The average RF during December to May under SSP126 and SSP585. c,d April SWE under SSP126 and SSP585. For each panel, Climatehist+LAPhist represents the historical (1995-2014) simulations with historical LAP depositions, while Climatefuture+LAPfuture and Climatefuture+LAPhist represent future (2081-2100) simulations with and without a future change of LAP depositions, respectively. The Climatefuture+LAPhist simulations used the historical average LAP depositions from 1995-2014. The horizontal axis labels represent different model configurations (see Methods), where Control has the ELM default settings and the others represent major adjustments made from the Control case. Specifically, PP assumes that the terrain is flat and neglects topographic effects on solar radiation; Koch assumes a non-spherical snow grain shape (Koch snowflake); extBC assumes external mixing between hydrophilic BC and snow grains; intDust assumes internal mixing between dust and snow grains; noLULCC has no land use and land cover change; MSE_high assumes high melt-water scavenging efficiency (MSE = 2, much higher than the default value of 0.2) of hydrophilic BC; and MSE_low assumes a low MSE (0.02) of hydrophilic BC. In (c,d), the contribution (δLAP) of future LAP change that mitigates snowpack loss under each ELM configuration is noted as a percentage and is calculated as the ratio of the SWE difference (ΔSWELAP) between Climatefuture+LAPfuture and Climatefuture+LAPhist to the SWE difference (ΔSWELimate) between Climatehist+LAPhist and Climatefuture+LAPfuture. The geographical coverage of the TP is shown in Fig. 4.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 710, 561, 727]]<|/det|> +## Contributions of LAP changes to future snowpack change + +<|ref|>text<|/ref|><|det|>[[115, 737, 876, 893]]<|/det|> +We further quantify contributions of LAP changes to future snowpack changes using ELM (Fig. 4). Overall, the future SWE in April over the NH is projected to decrease compared to the historical period, especially under SSP585 (Fig. 4a-c and Fig. S5). WUS will have a significant decline in SWE under both low and high emission scenarios, identical to the findings of a low-to-no snow future noted in previous studies8, 24. However, enhanced precipitation and sub-freezing temperatures will lead to an increase in SWE over the North American Arctic and high-latitude regions of Asia25. Consistent with previous findings26, TP is projected to have an overall decrease of seasonal snowpack throughout the 21st century. Due to its reliance on wintertime snowfall and initial cooler summertime temperatures, SWE in Karakoram, located in the northwestern TP, increases or stays relatively stable in the future27. We define + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 870, 228]]<|/det|> +the contribution of LAP change to snowpack change ( \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) ) as the difference between the projected SWE with and without future LAP change (see Methods). \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) is positive over the NH, especially under SSP126, and TP has the largest \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) (Fig. 4d- e). This highlights that future LAP change will contribute to a slowdown of warming- driven snowpack loss. The relative \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) (calculated as the ratio of \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) to projected SWE without LAP change) over the TP significantly increases (MK test: \(\mathrm{p}< 0.05\) ) from 0 to over \(10\%\) (SSP126) and \(20\%\) (SSP585) during 2015- 2100 (Fig. 4f- g). The future reduction of BC concentration dominates the change of relative \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) (Fig. S4f- g), accounting for most of the slowdown of SWE reduction. + +<|ref|>text<|/ref|><|det|>[[112, 238, 879, 494]]<|/det|> +We further analyze the relative contribution of climate change (i.e., temperature and precipitation) and LAP change as well as the impacts of ELM model configurations (Fig. 3c- d and Fig. S6) on SWE. Compared to the historical period, SSP585 shows a stronger decrease of SWE than SSP126 regardless of model configuration (Fig. 3c,d). Although topography, snow grain shape, mixing states of LAP- snow, land use and land cover changes, and melt- water scavenging efficiency can affect the magnitude of SWE, they have small impacts on the relative change of SWE (Fig. 3c,d). For example, the larger snow albedo of non- spherical snow grain shape can lead to larger SWE28, but snow grain shape has little impact on the relative change of SWE. We further estimate climate change impacts ( \(\Delta \mathrm{SWE}_{\mathrm{Climate}}\) ) on SWE as the difference between the historical SWE and future SWE without LAP change. We define the contribution ( \(\delta_{\mathrm{LAP}}\) ) of future LAP change that mitigates snowpack loss as the ratio of \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) to \(\Delta \mathrm{SWE}_{\mathrm{Climate}}\) (see Methods). The spatial distribution of \(\delta_{\mathrm{LAP}}\) shows that TP is the most sensitive to LAP change (Fig. S6). Under different model configurations, \(\delta_{\mathrm{LAP}}\) of TP is \(52.1 \pm 8.0\%\) (SSP126) and \(8.0 \pm 1.1\%\) (SSP585) (Fig. 3c,d). These results suggest that under SSP585, climate change dominates the snowpack loss and only \(8\%\) of this loss can be mitigated by LAP change, while under SSP126, LAP change can offset around \(52.1\%\) of the impacts of climate change on the TP snowpack. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 83, 880, 590]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 600, 877, 755]]<|/det|> +
Fig. 4| Future SWE in April and contributions of LAP change to future SWE. a Historical (1995-2014) and b,c future (2081-2100) spatial patterns of SWE in April under SSP126 and SSP585. d,e The differences (ΔSWELAP) of future (2081-2100) April SWEs with and without LAP change under SSP126 and SSP585. f,g Time series of relative ΔSWELAP (calculated as the ratio of ΔSWELAP to projected SWE without LAP change) over snow-covered regions (where the average SWE exceeds 5 mm in the historical period) in the TP. In (a-e), grids with an average April SWE smaller than 5 mm in the historical period are masked. In (f,g), BCfuture+Dustfuture, BCfuture+Dusthist, and BChist+Dustfuture represent different combinations of BC and dust depositions, where the subscripts of future and hist represent future and historical average depositions, respectively.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 820, 198, 835]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[115, 845, 881, 895]]<|/det|> +Research has shown that LAP contribute to the melting of snow and the retreat of glacier21, 29. We show decreasing trends of LAP deposition, mass, and RF under both low (SSP126) and high (SSP585) emission scenarios using simulation results from CMIP6 and ELM. Our study highlights a cleaner snow future due + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 320]]<|/det|> +to the significant reduction of BC deposition on snow surface. The projected cleaner snow will help alleviate future snowpack loss induced by warmer climates, especially over the TP. This is consistent with a recent study30 showing that BC deposition decrease since the 1980s has moderated the influence of climate change in the decline of snow cover over the French Alps and the Pyrenees. The compensating effects of decreasing BC will also contribute to the shift in snowmelt timing and substantially influence melt water runoff30. Considering the important role of the TP in Asia's freshwater supply26, the increased SWE due to cleaner snow will be beneficial for the municipal, hydropower31 and agricultural32 sectors in Asian regions as well as vegetation growth33 and animal survival34. Cleaner snow can mitigate over half of the snowpack loss caused by climate change under the optimized emission scenario (i.e., SSP126). These results stress the importance of reducing combustion aerosol emissions by developing clean, renewable energy and negative- emission technologies, in addition to mitigating climate change. However, climate factors dominate snowpack loss under the worst emission scenario (i.e., SSP585). The change in the relative role of reduced LAP highlights the necessity of constraining global warming levels to mitigate snowpack loss. + +<|ref|>text<|/ref|><|det|>[[112, 344, 881, 500]]<|/det|> +Beyond the significant decrease of RF of BC over the TP, the RF of dust will increase under SSP126. This is potentially due to increasing drought frequency, duration, and intensity linked to climate change as well as continued land use and land cover change35. The dust RF is projected to have a small decrease under SSP585 (Fig. 2). Consequently, dust is projected to account for a larger portion of LAP- induced RF than BC in the future under both SSP126 and SSP585, while BC accounts for more in the historical period (Fig. 2 and Fig. S4; Table S1). The increasing contributions of dust to snowmelt will be more significant over the high- altitude HMA, as previous work demonstrated that dust dominates high- altitude snow darkening and melt over HMA11. These highlight the importance of mitigating soil disturbance and stabilizing soil surface in dust source regions to reduce both natural and anthropogenic dust emissions36. + +<|ref|>text<|/ref|><|det|>[[112, 536, 880, 907]]<|/det|> +However, there remain uncertainties in simulating the LAP darkening effects on snow. First, accurately characterizing the emission, transport, chemistry, and deposition of LAP under changing climate is challenging. The emission, cycling, and persistence of BC are still under- represented in ESMs37. Although reduced fossil fuel burning in developing countries can decrease future BC emissions, increased wildfire intensity and frequency due to climate change and land- use change may potentially increase the BC emissions38, 39. However, ESMs still have large uncertainties when representing fire ignition, suppression, spread, and particularly emission, given our incomplete understanding of the complex and interacting controls on wildfire activities40, 41. The uncertainties of wildfire simulations are expected to have only minor impacts on our results, as we focus on the snow season from December to May when wildfire activities are less frequent42. Although different ESMs show relatively similar magnitudes and spatio- temporal patterns of BC deposition (Fig. 1d,g), the uncertainties associated with the emission inventory, transport and deposition modeling in ESMs should be quantified and reduced43. Furthermore, while ESMs reproduce the global spatial patterns and seasonal variations of dust distribution, there remain large differences in deposition rates (Fig. 1k,n) among ESMs due to uncertainties in simulating the dust life cycle (i.e., emission, transport, deposition). Such inconsistencies have been widely reported and are attributed to uncertainties of simulated land surface properties and atmospheric states22, 23, 44, 45, 46. Prior work shows that ESMs underestimate the amount of coarse dust with diameter \(\geq 5 \mu \mathrm{m}^{47}\). Biological soil crusts (biocrusts), which are rarely considered in ESMs, have been found to have large effects on regional and global dust cycling20. These limitations could lead to an underestimation of dust impurity effects in both historical period and future scenarios. However, our analyses are primarily based on relative differences rather than absolute values, mitigating the impact of the previously mentioned uncertainties. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 861, 141]]<|/det|> +Advanced remote sensing observations, e.g., NASA's Earth Surface Mineral Dust Source Investigation (EMIT), are promising for providing potential constraints of the sign and magnitude for projecting dust deposition under climate scenarios48. + +<|ref|>text<|/ref|><|det|>[[112, 152, 880, 496]]<|/det|> +Second, representing the evolutions of LAP and their mixing states with snow affects estimates of the LAP darkening effect on snow. The Snow, Ice, and Aerosol Radiative (SNICAR) model used in ELM (see Methods) can model the snow darkening effects of BC and dust well and has been widely used in snow- related studies49. The Control ELM simulations assumed spherical snow grain shape, internal mixing of hydrophilic BC- snow, and external mixing of dust- snow (see Methods). Although nonspherical snow grain shape and the mixing state of LAP- snow can affect the magnitude of LAP- induced RF, their impacts on the relative contribution of future LAP change to RF is within \(\pm 2\%\) (Fig. 3). More observations are needed to better constrain the irregular snow grain shape and space- and time- varying mixing states of LAP- snow in ESMs28. Although LAP scavenging processes via melting water can regulate LAP concentration after deposition, using either low or high melt- water scavenging efficiencies for hydrophilic BC leads to similar results in our simulations (Fig. 3). Apart from BC and dust, brown carbon also has large darkening effects on snow50. However, it is not represented in most ESMs due to the large variations and high uncertainties of its chemical composition and optical properties. Snow algae also play important roles in snow melt and glacier retreat51, but nearly all ESMs neglect snow algae effects on snow. Although snow algae blooming and distributions have been successfully implemented in the Minimal Advanced Treatments of Surface Interaction and Runoff land surface model (MATSIRO)52, more observations and modeling are needed to evaluate and improve MATSIRO performance before further applications. Better constraining the projections LAP impacts will benefit from the long- term field measurements and hyperspectral remote sensing satellite missions, e.g., the Surface Biology and Geology mission led by NASA53. + +<|ref|>text<|/ref|><|det|>[[112, 533, 876, 860]]<|/det|> +Third, projecting future SWE change is sensitive to meteorological forcings and model configurations. Precipitation and temperature largely determine the snowfall and snowmelt rates, producing large effects on snowpack4. However, temperature and precipitation of ESM simulations still have systematic biases54. The Community Earth System Model Version 2 (CESM2) forcing data used in this study shows an overall consistent trend of future precipitation and air temperature projections over the TP with the ensemble mean of the used CMIP6 models (Fig. S7a- d). Snowpack simulations are affected by the representations of various snow processes55. For example, topography affects the solar radiation received at the surface and snow processes56, while land use and land cover change can also affect the snow accumulation and melting processes57. Despite their importance, both of these factors have small impacts on our analysis (Fig. 3). Although model uncertainties may influence SWE projections, our sensitivity experiments with different ELM model configurations show only small impacts on the relative contributions of LAP. Our previous study showed that ELM can well capture the snow distribution in the TP, compared to the MODIS remote sensing data, supporting the reliability of the results28. ELM simulations can reproduce the spatio- temporal pattern, interannual variability and elevation gradient for different snow properties over the WUS58. The simulations are in line with Snow Telemetry field measurements, MODIS remote sensing products, and data assimilation products. The projected SWE trends of ELM simulations under both SSP126 and SSP585 are consistent with the ensemble mean of the seven CMIP6 models (Fig. S7). These model sensitivity experiments, model evaluations, and intercomparisons provide confidence in simulating future SWE change using ELM. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 866, 142]]<|/det|> +With a focus on the future evolution of LAP deposition and RF over the NH, we identify a cleaner snow future and reduced snowpack loss, especially over the TP. The broader implications of reduced LAP pollution in climate change need further analysis via coupled ESM experiments. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 185, 106]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 118, 266, 134]]<|/det|> +## CMIP6 simulations + +<|ref|>text<|/ref|><|det|>[[115, 144, 879, 250]]<|/det|> +Monthly aerosol deposition data during 1995- 2100, including deposition rates of BC and dust from seven ESMs participating in CMIP6 (Table S2), are used in the study. We select the CMIP6 models that provide data for both BC and dust deposition with the variant of "r1i1p1f1". Two future scenarios of SSP126 and SSP585 are included in the analysis. All the data are remapped to a spatial resolution of \(0.94^{\circ} \times 1.25^{\circ}\) in latitude and longitude and are aggregated to the annual scale by calculating average values during December to May. + +<|ref|>text<|/ref|><|det|>[[115, 260, 877, 362]]<|/det|> +Only CESM2 provides aerosol deposition data with different BC and dust categories (i.e., hydrophilic BC, hydrophobic BC, dust particle size). The other six models just provided the total deposition rates of BC and dust. Therefore, we used the historical and future atmospheric \(\mathrm{CO_2}\) concentration, meteorological forcing data (e.g., precipitation, air temperature, humidity, wind speed, downward solar radiation, and longwave radiation), and aerosol deposition data from CESM2 to drive the ELM to further investigate future trends of LAP mass in snow. + +<|ref|>sub_title<|/ref|><|det|>[[115, 402, 264, 418]]<|/det|> +## E3SM Land Model + +<|ref|>text<|/ref|><|det|>[[113, 428, 875, 722]]<|/det|> +The E3SM is a state- of- the- art fully- coupled ESM supported by the U.S. Department of Energy, which aims to improve and enhance actionable predictions of Earth system variability and change59. ELM, the land component of E3SM, originated from the Community Land Model Version 4.5 (CLM4.5)60. ELM can mechanistically simulate snow processes from snow accumulation to snow evolution to snow melt based on a multi- layer scheme. ELM can also prognostically simulate the change of LAP concentrations at different snow layers after deposition and how these changes affect snow albedo. Specifically, ELM uses a hybrid mode of SNICAR and the delta- Eddington adding- doubling radiative transfer solver to calculate the shortwave radiative characteristics of snow at different spectral bands61. The new SNICAR scheme in ELM treats both external mixing and internal mixing (within- hydrometeor) of BC and snow grains as well as size- dependent BC optical properties62. We recently extended ELM's ability to consider both external mixing and internal mixing of between dust and snow grains as well as the impact of non- spherical snow grain shape on snow albedo28. In addition, we implemented a new parameterization that considers sub- grid topographic effects on solar radiation56. The model enhancements allow ELM to be used to investigate uncertainties related to model configurations on future snow projections. Our previous study showed that ELM can well capture the spatio- temporal distribution and interannual variability of snow properties and timing compared to field measurements, remote sensing observations, and data assimilated products28, 58. These confirm the effectiveness of simulating snow processes in ELM. + +<|ref|>sub_title<|/ref|><|det|>[[115, 760, 456, 777]]<|/det|> +## Simulating the mass and RF of LAP in snow + +<|ref|>text<|/ref|><|det|>[[115, 787, 879, 908]]<|/det|> +To analyze future trends of the mass and RF of LAP, we conducted a series of offline ELM simulations at \(0.5^{\circ}\) spatial resolution over the NH from 1950 to 2100. The simulations were driven by CESM historical and future (SSP126 and SSP585) meteorological forcing and aerosol deposition data. We used the prescribed satellite phenology mode and downscaled the forcing data temporally and spatially to half- hourly and \(0.5^{\circ}\) resolution with bilinear interpolation methods. For each simulation, ELM was run at a half- hour time step with a monthly output frequency. The first 45 years (i.e., 1950- 1994) were treated as model spin- up time and the remaining 106 years (i.e. 1995- 2100) are used in the analysis. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 872, 211]]<|/det|> +To quantify the contribution of future LAP changes, we conducted a suite of simulations from 2015 to 2100 under the SSP126 and SSP585 scenarios with separate configurations of: 1) monthly LAP deposition averaged in the historical period (1995- 2014) (i.e., without interannual variation and trends in LAP); 2) future projected LAP deposition (i.e., with interannual variation and trends in LAP); 3) historical average BC deposition and future projected dust deposition; and 4) future projected BC deposition and historical average dust deposition. These simulations were carried out with the default ELM model settings, except that the simulations considered topographic effects on solar radiation. + +<|ref|>text<|/ref|><|det|>[[113, 221, 879, 290]]<|/det|> +To quantify uncertainties associated with model configurations, we also conducted ELM simulations with different model configurations related to snow grain shape, mixing state of LAP- snow, land use and land cover change, topographic effects on solar radiation, and melt- water scavenging efficiency of hydrophilic BC (Table S3). + +<|ref|>sub_title<|/ref|><|det|>[[113, 299, 627, 317]]<|/det|> +## Quantifying contributions of LAP changes to future snowpack loss + +<|ref|>text<|/ref|><|det|>[[113, 325, 880, 429]]<|/det|> +Based on the ELM simulations, the difference ( \(\Delta \mathrm{SWE}_{\mathrm{Climate}}\) ) between the historical (1995- 2014) SWE and projected future SWE (2081- 2100) without LAP changes is used to characterize the impact of climate change (i.e., temperature and precipitation) (see equation 1). The SWE difference ( \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) ) between future SWE with and without LAP changes is used to represent the impact of future LAP changes (see equation 2). We further define the ratio of \(\Delta \mathrm{SWE}_{\mathrm{LAP}}\) to \(\Delta \mathrm{SWE}_{\mathrm{Climate}}\) as the relative contribution of future LAP changes to the slowdown in snowpack loss ( \(\delta_{\mathrm{LAP}}\) ), as shown in equation 3. + +<|ref|>equation<|/ref|><|det|>[[205, 437, 880, 459]]<|/det|> +\[\Delta \mathrm{SWE}_{\mathrm{Climate}} = SWE_{\mathrm{Climate}_{\mathrm{his}}} + LAP_{\mathrm{hist}} - SWE_{\mathrm{Climate}_{\mathrm{future}} + LAP_{\mathrm{hist}}} \quad (1)\] + +<|ref|>equation<|/ref|><|det|>[[208, 470, 880, 491]]<|/det|> +\[\Delta \mathrm{SWE}_{\mathrm{LAP}} = SWE_{\mathrm{Climate}_{\mathrm{future}} + LAP_{\mathrm{future}}} - SWE_{\mathrm{Climate}_{\mathrm{future}} + LA_{\mathrm{hist}}} \quad (2)\] + +<|ref|>equation<|/ref|><|det|>[[384, 500, 880, 530]]<|/det|> +\[\delta_{\mathrm{LAP}} = \frac{\Delta \mathrm{SWE}_{\mathrm{LAP}}}{\Delta \mathrm{SWE}_{\mathrm{Climate}}} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[113, 540, 855, 575]]<|/det|> +The mean and standard deviation of \(\delta_{\mathrm{LAP}}\) under different ELM configurations are used to represent the mean LAP effect and the corresponding uncertainty. + +<|ref|>sub_title<|/ref|><|det|>[[113, 585, 259, 601]]<|/det|> +## Statistical analysis + +<|ref|>text<|/ref|><|det|>[[113, 611, 857, 681]]<|/det|> +All analyses are conducted using MATLAB R2019b (MathWorks Inc.). We use the non- parametric Mann- Kendall (MK) Tau test and Sen's slope to detect the monotonic trend of time series data. Trends with \(p< 0.05\) are considered to be statistically significant in this study. Specifically, we use the 'ktaub' function \(^{63}\) in this study. + +<|ref|>sub_title<|/ref|><|det|>[[113, 722, 244, 737]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[113, 747, 870, 815]]<|/det|> +Except for the CESM2 data, the aerosol deposition data in CMIP6 simulations can be freely downloaded from https://esgf- node.llnl.gov/search/cmip6/. All CESM2 data can be acquired using NCAR's data sharing service. The ELM outputs from this study are openly available at https://github.com/daleihao/snow_SSP. + +<|ref|>sub_title<|/ref|><|det|>[[113, 826, 247, 842]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[113, 853, 857, 888]]<|/det|> +The ELM code is publicly available at https://github.com/daleihao/E3SM. Codes to generate all results and plot all figures are available at https://github.com/daleihao/snow_SSP. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 118, 266, 133]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[113, 144, 880, 316]]<|/det|> +This research has been supported by the U.S. Department of Energy (DOE), Office of Science, Biological and Environmental Research (BER) program, Earth System Model Development program area, as part of the Climate Process Team projects. H.W. acknowledges support by the Regional and Global Model Analysis program area, as part of the HiLAT project. This research was conducted at Pacific Northwest National Laboratory (PNNL), which is operated for the U.S. DOE by Battelle Memorial Institute under contract DEAC05- 76RL01830. This research used resources from Cori and Perlmutter supercomputers of the National Energy Research Scientific Computing Center (NERSC), a User Facility supported by the Office of Science of the U.S. DOE under contract no. DE- AC02- 15 05CH11231. The reported research also used DOE BER Earth and Environmental Systems Modeling program's Compy computing cluster located at PNNL. We thank Beth Mundy and Ben Bond- Lamberty for their valuable suggestions. + +<|ref|>sub_title<|/ref|><|det|>[[115, 327, 281, 343]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 354, 870, 405]]<|/det|> +D.H. and G.B. conceived the study. D.H. processed the data, conducted the simulations, performed the analysis and drafted the original manuscript. All authors made suggestions to the design of the study and the analysis of the results, and contributed to improving the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[115, 416, 272, 432]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[115, 444, 430, 461]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 499, 201, 514]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[112, 525, 875, 576]]<|/det|> +1. Flanner MG, Shell KM, Barlage M, Perovich DK, Tschudi MA. Radiative forcing and albedo feedback from the Northern Hemisphere cryosphere between 1979 and 2008. Nature Geoscience 2011, 4(3): 151-155. + +<|ref|>text<|/ref|><|det|>[[112, 598, 858, 647]]<|/det|> +2. Biemans H, Siderius C, Lutz AF, Nepal S, Ahmad B, Hassan T, et al. Importance of snow and glacier meltwater for agriculture on the Indo-Gangetic Plain. Nature Sustainability 2019, 2(7): 594-601. + +<|ref|>text<|/ref|><|det|>[[112, 670, 796, 690]]<|/det|> +3. Simpkins G. Snow-related water woes. Nature Climate Change 2018, 8(11): 945-945. + +<|ref|>text<|/ref|><|det|>[[112, 712, 812, 747]]<|/det|> +4. Barnett TP, Adam JC, Lettenmaier DP. Potential impacts of a warming climate on water availability in snow-dominated regions. Nature 2005, 438(7066): 303-309. + +<|ref|>text<|/ref|><|det|>[[112, 771, 810, 806]]<|/det|> +5. Smith T, Bookhagen B. Changes in seasonal snow water equivalent distribution in High Mountain Asia (1987 to 2009). Science Advances 2018, 4(1): e1701550. + +<|ref|>text<|/ref|><|det|>[[112, 829, 840, 863]]<|/det|> +6. Bormann KJ, Brown RD, Derksen C, Painter TH. Estimating snow-cover trends from space. Nature Climate Change 2018, 8(11): 924-928. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 88, 880, 125]]<|/det|> +7. Kraaijenbrink PDA, Stigter EE, Yao T, Immerzeel WW. Climate change decisive for Asia's snow meltwater supply. Nature Climate Change 2021, 11(7): 591-597. + +<|ref|>text<|/ref|><|det|>[[60, 145, 872, 197]]<|/det|> +8. Siirila-Woodburn ER, Rhoades AM, Hatchett BJ, Huning LS, Szinai J, Tague C, et al. A low-to-no snow future and its impacts on water resources in the western United States. Nature Reviews Earth & Environment 2021, 2(11): 800-819. + +<|ref|>text<|/ref|><|det|>[[60, 220, 850, 271]]<|/det|> +9. Hock R, Rasul G, Adler C, Cáceres B, Gruber S, Hirabayashi Y, et al. High Mountain Areas. InIPCC Special Report on the Ocean and Cryosphere in a Changing Climate; Pörtner, H.-O., Roberts, DC. Masson-Delmotte, V., Zhai, P., Tignor, M., Poloczanska, E., Mintenbeck, K ... + +<|ref|>text<|/ref|><|det|>[[60, 293, 790, 328]]<|/det|> +10. Xu D, Ivanov VY, Li X, Troy TJ. Peak Runoff Timing Is Linked to Global Warming Trajectories. Earth's Future 2021, 9(8): e2021EF002083. + +<|ref|>text<|/ref|><|det|>[[60, 350, 877, 402]]<|/det|> +11. Sarangi C, Qian Y, Rittger K, Ruby Leung L, Chand D, Bormann KJ, et al. Dust dominates high-altitude snow darkening and melt over high-mountain Asia. Nature Climate Change 2020, 10(11): 1045-1051. + +<|ref|>text<|/ref|><|det|>[[60, 424, 859, 460]]<|/det|> +12. Hadley OL, Kirchstetter TW. Black-carbon reduction of snow albedo. Nature Climate Change 2012, 2(6): 437-440. + +<|ref|>text<|/ref|><|det|>[[60, 482, 860, 517]]<|/det|> +13. Skiles SM, Flanner M, Cook JM, Dumont M, Painter TH. Radiative forcing by light-absorbing particles in snow. Nature Climate Change 2018, 8(11): 964-971. + +<|ref|>text<|/ref|><|det|>[[60, 539, 852, 590]]<|/det|> +14. Bond TC, Doherty SJ, Fahey DW, Forster PM, Berntsen T, DeAngelo BJ, et al. Bounding the role of black carbon in the climate system: A scientific assessment. Journal of Geophysical Research: Atmospheres 2013, 118(11): 5380-5552. + +<|ref|>text<|/ref|><|det|>[[60, 613, 875, 648]]<|/det|> +15. Tanaka TY, Chiba M. A numerical study of the contributions of dust source regions to the global dust budget. Global and Planetary Change 2006, 52(1): 88-104. + +<|ref|>text<|/ref|><|det|>[[60, 670, 830, 706]]<|/det|> +16. Hansen J, Nazarenko L. Soot climate forcing via snow and ice albedos. Proceedings of the National Academy of Sciences 2004, 101(2): 423-428. + +<|ref|>text<|/ref|><|det|>[[60, 728, 875, 779]]<|/det|> +17. Gidden MJ, Riahi K, Smith SJ, Fujimori S, Luderer G, Kriegler E, et al. Global emissions pathways under different socioeconomic scenarios for use in CMIP6: a dataset of harmonized emissions trajectories through the end of the century. Geosci Model Dev 2019, 12(4): 1443-1475. + +<|ref|>text<|/ref|><|det|>[[60, 802, 866, 853]]<|/det|> +18. Ménégoz M, Krinner G, Balkanski Y, Cozic A, Boucher O, Ciais P. Boreal and temperate snow cover variations induced by black carbon emissions in the middle of the 21st century. The Cryosphere 2013, 7(2): 537-554. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 870, 139]]<|/det|> +19. Neff JC, Ballantyne AP, Farmer GL, Mahowald NM, Conroy JL, Landry CC, et al. Increasing eolian dust deposition in the western United States linked to human activity. Nature Geoscience 2008, 1(3): 189-195. + +<|ref|>text<|/ref|><|det|>[[110, 163, 880, 196]]<|/det|> +20. Rodriguez-Caballero E, Stanelle T, Egerer S, Cheng Y, Su H, Canton Y, et al. Global cycling and climate effects of aeolian dust controlled by biological soil crusts. Nature Geoscience 2022. + +<|ref|>text<|/ref|><|det|>[[110, 220, 850, 254]]<|/det|> +21. Li C, Bosch C, Kang S, Andersson A, Chen P, Zhang Q, et al. Sources of black carbon to the Himalayan-Tibetan Plateau glaciers. Nature Communications 2016, 7(1): 12574. + +<|ref|>text<|/ref|><|det|>[[110, 278, 860, 312]]<|/det|> +22. Zhao A, Ryder CL, Wilcox LJ. How well do the CMIP6 models simulate dust aerosols? Atmos Chem Phys 2022, 22(3): 2095-2119. + +<|ref|>text<|/ref|><|det|>[[110, 336, 825, 369]]<|/det|> +23. Evan AT, Flamant C, Fiedler S, Doherty O. An analysis of aeolian dust in climate models. Geophysical Research Letters 2014, 41(16): 5996-6001. + +<|ref|>text<|/ref|><|det|>[[110, 393, 881, 427]]<|/det|> +24. Fyfe JC, Derksen C, Mudryk L, Flato GM, Santer BD, Swart NC, et al. Large near-term projected snowpack loss over the western United States. Nature Communications 2017, 8(1): 14996. + +<|ref|>text<|/ref|><|det|>[[110, 451, 844, 485]]<|/det|> +25. Krasting JP, Broccoli AJ, Dixon KW, Lanzante JR. Future Changes in Northern Hemisphere Snowfall. Journal of Climate 2013, 26(20): 7813-7828. + +<|ref|>text<|/ref|><|det|>[[110, 509, 870, 542]]<|/det|> +26. Yao T, Thompson L, Chen D, Chettri N. A Scientific Assessment of the Third Pole Environment, 2022. + +<|ref|>text<|/ref|><|det|>[[110, 567, 850, 617]]<|/det|> +27. Kapnick SB, Delworth TL, Ashfaq M, Malyshev S, Milly PCD. Snowfall less sensitive to warming in Karakoram than in Himalayas due to a unique seasonal cycle. Nature Geoscience 2014, 7(11): 834-840. + +<|ref|>text<|/ref|><|det|>[[110, 641, 852, 691]]<|/det|> +28. Hao D, Bisht G, He C, Bair E, Huang H, Dang C, et al. Improving snow albedo modeling in E3SM land model (version 2.0) and assessing its impacts on snow and surface fluxes over the Tibetan Plateau. Geosci Model Dev Discuss 2022, 2022: 1-31. + +<|ref|>text<|/ref|><|det|>[[110, 715, 870, 748]]<|/det|> +29. Ramanathan V, Carmichael G. Global and regional climate changes due to black carbon. Nature Geoscience 2008, 1(4): 221-227. + +<|ref|>text<|/ref|><|det|>[[110, 772, 870, 822]]<|/det|> +30. Réveillet M, Dumont M, Gascoin S, Lafaysse M, Nabat P, Ribes A, et al. Black carbon and dust alter the response of mountain snow cover under climate change. Nature Communications 2022, 13(1): 5279. + +<|ref|>text<|/ref|><|det|>[[110, 847, 880, 896]]<|/det|> +31. Li D, Lu X, Walling DE, Zhang T, Steiner JF, Wasson RJ, et al. High Mountain Asia hydropower systems threatened by climate-driven landscape instability. Nature Geoscience 2022, 15(7): 520-530. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 104, 857, 139]]<|/det|> +32. Qin Y, Abatzoglou JT, Siebert S, Huning LS, AghaKouchak A, Mankin JS, et al. Agricultural risks from changing snowmelt. Nature Climate Change 2020, 10(5): 459-465. + +<|ref|>text<|/ref|><|det|>[[112, 163, 866, 213]]<|/det|> +33. Shen M, Piao S, Dorji T, Liu Q, Cong N, Chen X, et al. Plant phenological responses to climate change on the Tibetan Plateau: research status and challenges. National Science Review 2015, 2(4): 454-467. + +<|ref|>text<|/ref|><|det|>[[112, 238, 872, 271]]<|/det|> +34. Luo Z, Jiang Z, Tang S. Impacts of climate change on distributions and diversity of ungulates on the Tibetan Plateau. Ecological Applications 2015, 25(1): 24-38. + +<|ref|>text<|/ref|><|det|>[[112, 295, 852, 328]]<|/det|> +35. Kok JF, Storelvmo T, Karydis V, Adebiyi AA, Mahowald NM, Evan A, et al. The impacts of mineral dust aerosols on global climate and climate change. 2022. + +<|ref|>text<|/ref|><|det|>[[112, 352, 870, 401]]<|/det|> +36. Painter TH, Deems JS, Belnap J, Hamlet AF, Landry CC, Udall B. Response of Colorado River runoff to dust radiative forcing in snow. Proceedings of the National Academy of Sciences 2010, 107(40): 17125-17130. + +<|ref|>text<|/ref|><|det|>[[112, 426, 850, 459]]<|/det|> +37. Coppola AI, Wagner S, Lennartz ST, Seidel M, Ward ND, Dittmar T, et al. The black carbon cycle and its role in the Earth system. Nature Reviews Earth & Environment 2022. + +<|ref|>text<|/ref|><|det|>[[112, 484, 848, 516]]<|/det|> +38. Sullivan A, Baker E, Kurvits T. Spreading Like Wildfire: The Rising Threat of Extraordinary Landscape Fires. 2022. + +<|ref|>text<|/ref|><|det|>[[112, 542, 875, 590]]<|/det|> +39. Jones MW, Abatzoglou JT, Veraverbeke S, Andela N, Lasslop G, Forkel M, et al. Global and Regional Trends and Drivers of Fire Under Climate Change. Reviews of Geophysics 2022, 60(3): e2020RG000726. + +<|ref|>text<|/ref|><|det|>[[112, 616, 857, 649]]<|/det|> +40. Kloster S, Lasslop G. Historical and future fire occurrence (1850 to 2100) simulated in CMIP5 Earth System Models. Global and Planetary Change 2017, 150: 58-69. + +<|ref|>text<|/ref|><|det|>[[112, 674, 825, 707]]<|/det|> +41. Hantson S, Arneth A, Harrison SP, Kelley DI, Prentice IC, Rabin SS, et al. The status and challenge of global fire modelling. Biogeosciences 2016, 13(11): 3359-3375. + +<|ref|>text<|/ref|><|det|>[[112, 732, 855, 781]]<|/det|> +42. Giglio L, Randerson JT, van der Werf GR. Analysis of daily, monthly, and annual burned area using the fourth-generation global fire emissions database (GFED4). Journal of Geophysical Research: Biogeosciences 2013, 118(1): 317-328. + +<|ref|>text<|/ref|><|det|>[[112, 806, 868, 839]]<|/det|> +43. Kang S, Zhang Y, Qian Y, Wang H. A review of black carbon in snow and ice and its impact on the cryosphere. Earth-Science Reviews 2020, 210: 103346. + +<|ref|>text<|/ref|><|det|>[[112, 864, 866, 897]]<|/det|> +44. Huneeus N, Schulz M, Balkanski Y, Griesfeller J, Prospero J, Kinne S, et al. Global dust model intercomparison in AeroCom phase I. Atmos Chem Phys 2011, 11(15): 7781-7816. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 103, 820, 139]]<|/det|> +45. Wu C, Lin Z, Liu X. The global dust cycle and uncertainty in CMIP5 (Coupled Model Intercomparison Project phase 5) models. Atmos Chem Phys 2020, 20(17): 10401-10425. + +<|ref|>text<|/ref|><|det|>[[113, 163, 876, 197]]<|/det|> +46. Aryal YN, Evans S. Global Dust Variability Explained by Drought Sensitivity in CMIP6 Models. Journal of Geophysical Research: Earth Surface 2021, 126(6): e2021JF006073. + +<|ref|>text<|/ref|><|det|>[[113, 221, 855, 255]]<|/det|> +47. Adebiyi AA, Kok JF. Climate models miss most of the coarse dust in the atmosphere. Science Advances 2020, 6(15): eaaz9507. + +<|ref|>text<|/ref|><|det|>[[113, 279, 876, 328]]<|/det|> +48. Green RO, Mahowald N, Ung C, Thompson DR, Bator L, Bennet M, et al. The Earth Surface Mineral Dust Source Investigation: An Earth Science Imaging Spectroscopy Mission. 2020 IEEE Aerospace Conference; 2020 7-14 March 2020; 2020. p. 1-15. + +<|ref|>text<|/ref|><|det|>[[113, 352, 876, 386]]<|/det|> +49. Flanner MG, Arnheim JB, Cook JM, Dang C, He C, Huang X, et al. SNICAR-ADv3: a community tool for modeling spectral snow albedo. Geosci Model Dev 2021, 14(12): 7673-7704. + +<|ref|>text<|/ref|><|det|>[[113, 410, 880, 460]]<|/det|> +50. Brown H, Wang H, Flanner M, Liu X, Singh B, Zhang R, et al. Brown Carbon Fuel and Emission Source Attributions to Global Snow Darkening Effect. Journal of Advances in Modeling Earth Systems 2022, 14(4): e2021MS002768. + +<|ref|>text<|/ref|><|det|>[[113, 484, 841, 518]]<|/det|> +51. Ganey GQ, Loso MG, Burgess AB, Dial RJ. The role of microbes in snowmelt and radiative forcing on an Alaskan icefield. Nature Geoscience 2017, 10(10): 754-759. + +<|ref|>text<|/ref|><|det|>[[113, 542, 880, 592]]<|/det|> +52. Onuma Y, Yoshimura K, Takeuchi N. Global Simulation of Snow Algal Blooming by Coupling a Land Surface and Newly Developed Snow Algae Models. Journal of Geophysical Research: Biogeosciences 2022, 127(2): e2021JG006339. + +<|ref|>text<|/ref|><|det|>[[113, 616, 860, 666]]<|/det|> +53. Cawse-Nicholson K, Townsend PA, Schimel D, Assiri AM, Blake PL, Buongiorno MF, et al. NASA's surface biology and geology designated observable: A perspective on surface imaging algorithms. Remote Sensing of Environment 2021, 257: 112349. + +<|ref|>text<|/ref|><|det|>[[113, 690, 872, 740]]<|/det|> +54. Chai Y, Yue Y, Slater LJ, Yin J, Borthwick AGL, Chen T, et al. Constrained CMIP6 projections indicate less warming and a slower increase in water availability across Asia. Nature Communications 2022, 13(1): 4124. + +<|ref|>text<|/ref|><|det|>[[113, 764, 879, 813]]<|/det|> +55. Krinner G, Derksen C, Essery R, Flanner M, Hagemann S, Clark M, et al. ESM-SnowMIP: assessing snow models and quantifying snow-related climate feedbacks. Geosci Model Dev 2018, 11(12): 5027-5049. + +<|ref|>text<|/ref|><|det|>[[113, 838, 864, 888]]<|/det|> +56. Hao D, Bisht G, Gu Y, Lee WL, Liou KN, Leung LR. A parameterization of sub-grid topographical effects on solar radiation in the E3SM Land Model (version 1.0): implementation and evaluation over the Tibetan Plateau. Geosci Model Dev 2021, 14(10): 6273-6289. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 99, 880, 135]]<|/det|> +57. Varhola A, Coops NC, Weiler M, Moore RD. Forest canopy effects on snow accumulation and ablation: An integrative review of empirical results. Journal of Hydrology 2010, 392(3): 219-233. + +<|ref|>text<|/ref|><|det|>[[57, 163, 861, 198]]<|/det|> +58. Hao D, Bisht G, Rittger K, Stillinger T, Bair E, Gu Y, et al. Evaluation of snow processes over the Western United States in E3SM land model. EGUsphere 2022, 2022: 1-38. + +<|ref|>text<|/ref|><|det|>[[57, 220, 872, 270]]<|/det|> +59. Leung LR, Bader DC, Taylor MA, McCoy RB. An Introduction to the E3SM Special Collection: Goals, Science Drivers, Development, and Analysis. Journal of Advances in Modeling Earth Systems 2020, 12(11): e2019MS001821. + +<|ref|>text<|/ref|><|det|>[[57, 295, 864, 345]]<|/det|> +60. Golaz J-C, Caldwell PM, Van Roekel LP, Petersen MR, Tang Q, Wolfe JD, et al. The DOE E3SM Coupled Model Version 1: Overview and Evaluation at Standard Resolution. Journal of Advances in Modeling Earth Systems 2019, 11(7): 2089-2129. + +<|ref|>text<|/ref|><|det|>[[57, 368, 876, 418]]<|/det|> +61. Dang C, Zender CS, Flanner MG. Intercomparison and improvement of two-stream shortwave radiative transfer schemes in Earth system models for a unified treatment of cryospheric surfaces. The Cryosphere 2019, 13(9): 2325-2343. + +<|ref|>text<|/ref|><|det|>[[57, 441, 878, 491]]<|/det|> +62. Wang H, Easter RC, Zhang R, Ma P-L, Singh B, Zhang K, et al. Aerosols in the E3SM Version 1: New Developments and Their Impacts on Radiative Forcing. Journal of Advances in Modeling Earth Systems 2020, 12(1): e2019MS001851. + +<|ref|>text<|/ref|><|det|>[[57, 515, 673, 533]]<|/det|> +63. Burkey J. Mann-Kendall Tau-b with Sen's Method (enhanced). 2022. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 131, 149, 150]]<|/det|> +SM.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/images_list.json b/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..2fa366489f6509d16bd0c247e344b95eaa8e33c8 --- /dev/null +++ b/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/images_list.json @@ -0,0 +1,131 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig.1. Epigenetic remodelling after centromere formation. (A) Schematic detailing experimental model system. Left; human lymphoblastoid cells (Parental) harbouring a canonical human chromosome 3 (HSA3; purple frame) and a chromosome 3 with a neocentromere located at 3q24 (Neo3; orange frame) were genetically manipulated to isolate Neo3 in a human-hamster hybrid (HybNeo3) for comparison to a human-hamster hybrid (GM10253A) with a single HSA3 chromosome. Middle, DNA FISH with a-satellite specific probe (red) and a fosmid probe (green) for 3q24. The primary constriction (arrow) coincides with the a-satellite array in HSA3 but is located at 3q24 in Neo3. Chromosomes were counterstained with DAPI. Bar is 5 μm. Right, chromosome 3 ideogram to indicate centromere and a-satellite locations on HSA3 and Neo3. (B) Distribution of active (H3K4me2, H3K27ac) and repressive (H3K9me3, H3K9me2) epigenetic marks measured by ChIP-chip at the 3q24 neocentromere region on HSA3 and Neo3 chromosomes. Neocentromere is marked in solid blue, equivalent position in HSA3 is marked in open blue box and remodelled pericentromeric heterochromatin domain is marked in yellow. Bottom, position of genes at 3q24 locus. (C) Detailed view of active (H4K20me1) and repressive (H3K9me2/3) epigenetic marks surrounding the neocentromere.", + "footnote": [], + "bbox": [ + [ + 77, + 45, + 899, + 714 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. RNA pol II binding at a functioning human centromere.", + "footnote": [], + "bbox": [ + [ + 188, + 40, + 812, + 732 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig.3. Facultative heterochromatin and convergent genes mark transcriptionally silenced pericentromeric domain.", + "footnote": [], + "bbox": [ + [ + 95, + 45, + 916, + 539 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig.4. Centromeric chromatin has a transcription dependent underwound and disrupted chromatin fibre structure.", + "footnote": [], + "bbox": [ + [ + 100, + 50, + 884, + 526 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Compact heterochromatin in pericentromeric domain", + "footnote": [], + "bbox": [ + [ + 115, + 45, + 895, + 820 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig.6. Large-scale neocentromere chromatin fibre decompaction.", + "footnote": [], + "bbox": [ + [ + 137, + 45, + 855, + 525 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig.7. Neocentromere chromosome instability.", + "footnote": [], + "bbox": [ + [ + 145, + 78, + 861, + 556 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_2.jpg", + "caption": "Supplementary Fig. 2. Transcription repression at the pericentromeric domain (Relates to Figure 2) Top, diagram showing individual genes at 3q24, yellow block corresponds to pericentromere domain, blue is the neocentromere. Bottom, RT-qPCR expression data for genes within and bordering the pericentromeric heterochromatic domain, showing expression from HSA3 and gene silencing on Neo3.", + "footnote": [], + "bbox": [], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_4.jpg", + "caption": "Supplementary Fig. 4 Legend over page", + "footnote": [], + "bbox": [], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_6.jpg", + "caption": "Supplementary Fig. 6. Neocentromere associated genome instability (Relates to Figure 7) (A) Left, representative FISH images of Neo3 metaphase chromosomes hybridised to a human chromosome 3 paint (green) and BAC (red) located at the neocentromere. Chromosome morphology was scored as normal or showing instability: deletions, fusions or duplications. Bar is \\(2\\mu \\mathrm{m}\\) . Right, quantification \\((\\%)\\) of different chromosome morphologies with increasing passage number (low \\(\\sim 10\\) , high \\(\\sim 100\\) ) over time. P values are for a \\(\\chi^2\\) test compared to low passage. (B) High passage Neo3 Chromosome copy number (RPKM), with a zoom in of the 3 Mb region around the neocentromere (blue). (C) High passage Neo3 Chromosome copy number (RPKM), with a zoom in of the 7 Mb region around the neocentromere (blue). Locations of DNA FISH probes are shown, neocentromere BAC probe (green) and border fosmid (red). (D) Left, representative FISH images of Neo3 metaphase chromosomes hybridised to a BAC (green) located at the neocentromere and a fosmid (red) located at the border. Bar is \\(2\\mu \\mathrm{m}\\) . Right, chromosome morphology was scored and quantified \\((\\%)\\) with increasing passage number over time. Loss of the border probe signal was coincident with fusion of human Neo3 fragment (light grey) to a hamster chromosome (dark grey). Bar is \\(2\\mu \\mathrm{m}\\) .", + "footnote": [], + "bbox": [], + "page_idx": 29 + } +] \ No newline at end of file diff --git a/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4.mmd b/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9f107d8bd94de1ccbf0b569548722415313d6f51 --- /dev/null +++ b/preprint/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4/preprint__0d5d236ca2db4837436a59c800cfcd8d4634bf4e1db39c00ce4a07df353e91c4.mmd @@ -0,0 +1,357 @@ + +# Human centromere formation activates transcription and opens chromatin fibre structure + +Nick Gilbert ( nick.gilbert@ed.ac.uk ) University of Edinburgh https://orcid.org/0000- 0003- 0505- 6081 + +Catherine Naughton The University of Edinburgh + +Covadonga Huidobro MRC Human Genetics Unit at The University of Edinburgh https://orcid.org/0000- 0001- 9823- 3846 + +Claudia Catacchio University of Bari, Department of Biology + +Adam Buckle MRC Human Genetics Unit at The University of Edinburgh + +Graeme Grimes MRC Human Genetics Unit at The University of Edinburgh + +Ryu- Suke Nozawa MRC Human Genetics Unit at The University of Edinburgh + +Stefania Purgato University of Bologna, Department of Pharmacy and Biotechnology + +Mariano Rocchi University of Bari, Department of Biology + +Article + +Keywords: + +Posted Date: January 18th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1061218/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on September 24th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33426- 2. + +<--- Page Split ---> + +## Human centromere formation activates transcription and opens chromatin fibre structure + +Catherine Naughton \(^{1}\) , Covadonga Huidobro \(^{1}\) , Claudia R. Catacchio \(^{1,2}\) , Adam Buckle \(^{1}\) , Graeme R. Grimes, Ryu- Suke Nozawa \(^{1}\) , Stefania Purgato \(^{1,3}\) , Mariano Rocchi \(^{2}\) , Nick Gilbert \(^{1*}\) + +\(^{1}\) MRC Human Genetics Unit, The University of Edinburgh, Crewe Rd, Edinburgh, EH4 2XU \(^{2}\) Department of Biology, University of Bari, Via Orabona 4, 70125 Bari, Italy \(^{3}\) Department of Pharmacy and Biotechnology, University of Bologna, 40126 Bologna, Italy + +\*Nick.Gilbert@ed.ac.uk + +## Abstract + +Human centromeres appear as constrictions on mitotic chromosomes and form a platform for kinetochore assembly in mitosis. Biophysical experiments led to a suggestion that repetitive DNA at centromeric regions form a compact scaffold necessary for function, but this was revised when neocentromeres were discovered on non- repetitive DNA. To test whether centromeres have a special chromatin structure we have analysed the architecture of a neocentromere. Centromere formation is accompanied by RNA pol II recruitment and active transcription to form a decompacted, negatively supercoiled domain enriched in 'open' chromatin fibres. In contrast, centromerisation causes a spreading of repressive epigenetic marks to surrounding regions, delimited by H3K27me3 polycomb boundaries and divergent genes. This flanking domain is transcriptionally silent and partially remodelled to form 'compact' chromatin, similar to satellite- containing DNA sequences, and exhibits genomic instability. We suggest transcription disrupts chromatin to provide a foundation for kinetochore formation whilst compact pericentromeric heterochromatin generates mechanical rigidity. + +<--- Page Split ---> + +## Introduction + +Centromeres are highly specialized genomic loci necessary to maintain genome stability. Cytogenetically they are the primary constriction of a metaphase chromosome and functionally provide an assembly site for the kinetochore, a multiprotein structure that forms attachments to the microtubules of the mitotic and meiotic spindles'. In mitosis the kinetochore is composed of a trilaminar structure with an outer layer binding to microtubules, but the architecture of the underlying chromatin fibre is unknown?. + +Human centromeric chromatin is assembled from CENP- A nucleosomes and repetitive \(\alpha\) - satellite DNA sequences that span 250- 5000 \(\mathrm{Kb}^{4,5}\) . Mouse acrocentric chromosomes have a similar organisation but, in this case, a small centromeric domain of minor satellite is flanked by a larger region of major satellite, which in interphase coalesces to form large dense chromocentres, enriched in heterochromatic marks and HP1 protein. A prevailing hypothesis is that repetitive satellite sequences at centromeres form compact heterochromatin which provides a stable scaffold for the kinetochore. This idea is supported by biophysical experiments: (i) analysis of satellite containing mouse and human centromeric chromatin by sucrose gradient sedimentation shows that it sediments more rapidly than expected for its size, indicating that is has a compact chromatin structure, analogous to a rigid rod'; (ii) in different species nucleosomes are positioned regularly on satellite sequences consistent with the assembly of chromatin fibres having a regular and stable structure8,9; (ii) in vitro pulling experiments indicate that regularly folded chromatin has the biophysics properties of a stiff spring10. + +In contrast to biophysical data that indicates satellite containing centromeric chromatin has a uniform compact architecture, immunofluorescence analysis on extended interphase chromatin fibers11- 12 show that it is divided into core and pericentromeric domains. The centromere core domain is enriched in active histone modifications indicative of transcription, whilst the surrounding pericentromeric regions are marked by repressive histone marks12. The centromere core, which is epigenetically defined by the variant histone CENP- A, interacts with CENP- C through the LEEGLG motif at the extreme C terminus14 and RNA15,16 to form an anchor for kinetochore formation, whilst the pericentromere recruits cohesin and condensin to regulate chromatin stiffness17. Whilst mechanisms for CENP- A recruitment are slightly different between species, it appears that a function of transcription at the core is to facilitate the incorporation of CENP- A containing nucleosomes18. Furthermore, studies in S. pombe indicate that chromatin remodelers spread from the centromeric core to surrounding pericentromeric regions19. This two- domain organisation appears critical for centromere stability, as experiments disrupting either transcription levels, or heterochromatic marks, affect chromatin compaction and result in mitotic defects20- 25. + +Higher eukaryotic centromeres are typically located on repetitive DNA sequences, but they can also be found at euchromatic sequences26. These neocentromeres often form in response to chromosome instability in cancer which deletes the canonical centromere, but can also occur congenitally. Neocentromeres that have been inherited through generations and become fixed in the population are described as evolutionary new centromeres (ENC's) and have often been inferred from studying the chromosome architecture between similarly related species27. Historically molecular analyses of centromeres have been challenging due to the repetitive nature of the underlying DNA, however as neocentromeres are formed on unique DNA sequences they provide a useful model to interrogate centromeric chromatin structure and provide insight into the properties of canonical centromeres. For example, examination of ENCs imply \(\alpha\) - satellite DNA may be acquired over time at neocentromeres as they mature28- 30, neocentromeres lacking pericentromeric heterochromatin in cis may establish interactions to distal heterochromatin in trans31; neocentromere formation promotes H3K9me3 loss and RNA polymerase II accumulation at the CENP- A core32, suggesting that chromatin is remodelled to accommodate a functional centromere. + +To reconcile results from biophysical and imaging based studies we have used a neocentromere as a model system to determine whether centromeres have a special chromatin structure. Centromere formation is accompanied by epigenetic and chromatin fibre remodelling: the CENP- A defined core becomes enriched in active epigenetic marks, RNA polymerase, and negatively supercoiled DNA, consistent with transcription. To examine the biophysical properties of the chromatin fibre, sedimentation analysis shows that it has a transcription dependent disrupted chromatin fibre structure. These structural changes of core centromeric chromatin further affect the large- scale chromatin fibre folding of this region which becomes decompacted in a transcription dependent manner. Strikingly, there is pronounced epigenetic remodelling and transcriptional silencing of a large 5 Mb region surrounding the centromeric core. Although there is no concomitant change in nucleosome positioning at the centromere there is evidence for partial remodelling of the flanking + +<--- Page Split ---> + +pericentromeric heterochromatin to form 'compact' chromatin. As this region is genomically unstable we propose that further remodelling of the pericentromeric region to form compact heterochromatin occurs as the neocentromere matures. Overall, our data indicates that centromeres are remodelled to have a special chromatin structure: chromatin fibres at the centromere core have a disrupted structure that we suggest provides a suitable foundation to attach the kinetochore components whilst flanking sequences form a compact heterochromatin- like structure that has mechanical rigidity. + +<--- Page Split ---> + +## Results Epigenetic remodelling at a human neocentromere + +To understand how new centromeres are accommodated in chromosomes and to investigate whether centromeres have a special chromatin structure required to form a stable kinetochore we used a previously identified neocentromere at 3q24 as a model system33. This neocentromere has been propagated across multiple generations, indicating it is stable through the germline, and is located in the vicinity of two genes but within a relatively gene poor segment of the genome. As the parental lymphoblastoid cells are heterozygous for the neocentromere at 3q24 the chromosome harbouring the neocentromere, Neo3, was genetically isolated from the normal counterpart in a human- hamster hybrid cell line (HybNeo3; Fig 1A) and compared to a human- hamster hybrid cell line, GM10253A, which has a single normal human chromosome 3, termed HSA3. No genetic changes were apparent in Neo3 and there was no evidence for repetitive DNA at 3q24 by deep sequencing, whilst as reported for other neocentromeres34- 36 \(\alpha\) - satellite persisted at the original centromere location (Fig. 1A). + +The position of the neocentromere was confirmed by using DNA fluorescence in situ hybridization (FISH) with probes to 3q24 (Supplementary Fig. 1A- C) and high- resolution mapped using ChIP with antibodies to CENP- A and CENP- C in the parental cells (Supplementary Fig. 1D), revealing a centromere core domain of 130 kb, similar to other synthetically derived neocentromeres32. In the derivatized human- hamster HybNeo3 cell line the centromere had drifted \(\approx 30\mathrm{kb}\) away from the telomere and had spread to encompass a \(\approx 190\mathrm{kb}\) domain. Centromere drift is apparent in horse and fission yeast37,38, and may represent a natural event controlled in part by the constitutive centromere- associated network (CCAN) and buffered by repetitive satellite DNA39. + +We set out to investigate how the chromatin fibre is remodelled in response to centromere formation and reasoned there could be two distinct possibilities (i) neocentromeres form at a genomic location that already has the features required for centromere function or (ii) neocentromeres have the capacity to remodel the local epigenetic environment. To discriminate between these two scenarios we examined the epigenetic repertoire of 3q24 using ChIP for active (H3K27ac, H3K4me2, H4K20me1) and repressive (H3K9me3, H3K9me2) epigenetic marks. A small block of GC- rich DNA in the vicinity of the neocentromere amplified aberrantly and was blacklisted (Supplementary Fig. 1E). + +The canonical 3q24 locus on HSA3 was decorated with active marks (Fig. 1B- C) coincident with actively transcribed genes in a euchromatic region, whilst repressive heterochromatic marks were absent. In contrast, after neocentromerisation, a large 5 Mb heterochromatin domain marked by H3K9me2/3 formed around the centromere on Neo3 (Fig. 1B, yellow box). Focal active marks in the vicinity of genes were absent and there was a significant loss of H3K27ac, a marker of CBP/P300 activity40. The upstream and downstream pericentromeric regions had the epigenetic hallmarks of heterochromatin consistent with the idea that centromeres are remodelled into a repressive state, even in the absence of repetitive DNA, demonstrating that special DNA sequences are not required for heterochromatin formation. At the centromeric core (Fig. 1B- C, blue box), coincident with CENP- C binding, the chromatin was remodelled to a state distinct from the flanking pericentromeric domains, devoid of heterochromatin marks and enriched for H4K20me1. This data suggests that instead of neocentromeres adopting the local epigenetic landscape they can remodel the local chromatin environment32 to form distinct centromeric and pericentromeric domains (Fig. 1B, blue and yellow boxes, respectively). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig.1. Epigenetic remodelling after centromere formation. (A) Schematic detailing experimental model system. Left; human lymphoblastoid cells (Parental) harbouring a canonical human chromosome 3 (HSA3; purple frame) and a chromosome 3 with a neocentromere located at 3q24 (Neo3; orange frame) were genetically manipulated to isolate Neo3 in a human-hamster hybrid (HybNeo3) for comparison to a human-hamster hybrid (GM10253A) with a single HSA3 chromosome. Middle, DNA FISH with a-satellite specific probe (red) and a fosmid probe (green) for 3q24. The primary constriction (arrow) coincides with the a-satellite array in HSA3 but is located at 3q24 in Neo3. Chromosomes were counterstained with DAPI. Bar is 5 μm. Right, chromosome 3 ideogram to indicate centromere and a-satellite locations on HSA3 and Neo3. (B) Distribution of active (H3K4me2, H3K27ac) and repressive (H3K9me3, H3K9me2) epigenetic marks measured by ChIP-chip at the 3q24 neocentromere region on HSA3 and Neo3 chromosomes. Neocentromere is marked in solid blue, equivalent position in HSA3 is marked in open blue box and remodelled pericentromeric heterochromatin domain is marked in yellow. Bottom, position of genes at 3q24 locus. (C) Detailed view of active (H4K20me1) and repressive (H3K9me2/3) epigenetic marks surrounding the neocentromere.
+ +<--- Page Split ---> + +## Transcriptional landscape at a neocentromere + +To understand the molecular basis for the distinct centromeric and pericentromeric domains, patterns of transcription were examined. By RT- qPCR active genes at 3q24 on HSA3 were all silenced upon centromere formation, as far as the distal DIPK2A and HLTF genes (Supplementary Fig. 2), suggestive of a spreading activity emanating from the centromere core domain. RNA sequencing was used to further explore the landscape and showed transcriptional repression over a 5 Mb pericentromeric domain on Neo3 (Fig. 2A). Recent data has indicated that (neo)centromeres are transcriptionally active in mitosis32,41 and in interphase in model organisms18,19,42. ChIP for RNA polymerase II showed it was absent from the pericentromeric domain in Neo3, but statistically significant levels of polymerase were apparent at the centromeric core in both interphase (Fig. 2B) and metaphase (Fig. 2C) cells. However, no transcripts were detected within the centromeric core domain on Neo3 even after exosome knockdown (Fig. 2D- E). Similarly, very deep sequencing for either short RNA transcripts, long RNA transcripts or nascent transcripts could not identify specific RNA species. This led us to speculate that transcription was at a very low level and dispersed across the centromeric core domain with multiple transcription initiation sites, but sufficient to remodel the local chromatin landscape. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. RNA pol II binding at a functioning human centromere.
+ +(A) Normalised transcription across HSA3 and Neo3 analysed by RNA-seq. (B) RNA pol II (CTD domain, Diagenode) (green) and CENP-C (gift, W. Earnshaw) (red) immunofluorescence staining on a metaphase spread from parental cells (HSA3, purple; Neo3, orange). Bar is 5 μm. (C) Western blot confirming EXOSC3 protein (arrow) knockdown following 72 h RNAi treatment in GM10253A cells. (D) Distribution of nascent transcripts (normalised RPKM) in 600 kb window around neocentromere (blue), mapped using TT-seq. Top panels correspond to RNAi control, bottom panels are for exosome RNAi knockdown. (E) Distribution of RNA pol II binding at 3q24 for Neo3 (orange) and HSA3 (purple) chromosomes. Top, RNA pol II CTD antibody (D2, Diagenode), bottom, RNA pol II CTD antibody from Hiroshi Kimura (HK1). Horizontal line (arrows) corresponds to a sampled background model with \(p = 0.01\). Significant peaks (\(p<0.001\)) called using the Ringo package for Neo3 (orange) and HSA3 (purple) are shown below tracks. + +<--- Page Split ---> + +## Distinct pericentromeric domain boundaries + +Centromerisation triggered epigenetic remodelling to form a repressive pericentromeric domain (Fig. 1B). To characterise how the domain was delimited we examined facultative epigenetic marks across the locus and identified strong H3K27me3 enrichment at the telomeric end of the domain and a weak enrichment for H3K36me3 at the other end (Fig. 3A). H3K27me3 is a mark indicative of polycomb activity, whilst H3K36me3 is added by Set2 at active genes but is also reported as a facultative mark at heterochromatin43. These features suggest that the repressive pericentromere activity spreads until it reaches these boundaries, but raises the question as to what defines them? CTCF is an abundant protein associated with GC- rich DNA sequences and provides boundary activity for marking chromatin interaction domains measured by 3C techniques44,45. Although CTCF was present throughout the pericentromeric domain there were strong peaks located at both ends of the region (Fig. 3B). More strikingly and consistent with a recent study in budding yeast46 the pericentromeric domain was also delimited by the first convergently transcribed gene encountered moving away from the centromere (Fig. 3B). These results reveal a pronounced two domain structure at centromeres with a core and pericentromeric domain flanked by boundary sites defined by convergent genes and CTCF binding. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig.3. Facultative heterochromatin and convergent genes mark transcriptionally silenced pericentromeric domain.
+ +(A) Distribution of facultative heterochromatin (H3K27me3 and H3K36me3) marks measured by ChIP-chip on HSA3 and Neo3 chromosomes. (B) Top, ChIP-chip showing distribution of CTCF at pericentromeric boundaries on Neo3 chromosomes. Bottom, schematic showing gene orientation at 3q24, convergent gene boundary marked in red. Vertical blue line corresponds to the neocentromere and repressive pericentromeric heterochromatin domain in yellow. + +<--- Page Split ---> + +## Local chromatin fibre remodelling after centromere formation + +In gene- rich euchromatin nucleosomal DNA is packaged into chromatin fibres which have a disrupted or 'open' configuration47, a structure that is particularly conspicuous at transcription start sites48. In contrast, the specialised chromatin found at centromeres is formed from alpha satellite 4and CENP- A containing nucleosomes3. Alpha satellite makes up \(3\%\) of the human genome and positions nucleosomes precisely in vivo8 and in vitro9. Sedimentation studies to examine the biophysical properties of satellite containing chromatin indicate that it has the characteristics of a rigid rod- like particle which may enable it to fold into an ordered or crystalline array7,49. To establish the biophysical properties of centromeric chromatin (Fig. 4A) soluble chromatin was prepared from nuclei containing HSA3 and Neo3 chromosomes and fractionated by sucrose gradient sedimentation and pulsed- field gel electrophoresis47 (PFGE) (Supplementary Fig. 3A). Subsequently DNA corresponding to 'open' or disrupted chromatin was isolated from the agarose gel (Supplementary Fig. 3B) and used to map the chromatin fibre structure in the centromeric domain. Chromatin fibres located at the neocentromere were substantially remodelled to have a pronounced 'open' configuration (Fig. 4B) which was restricted to the CENP- C containing core, and is similar to the characteristics observed at transcription start sites48. Due to the presence of low- level RNA polymerase in the centromeric domain (Fig. 2A) we speculated that chromatin remodelling was linked to transcription, as has been observed in model organisms18,19,42. Concomitantly transcription inhibition completely abrogated the formation of disrupted chromatin fibres (Fig. 4B) demonstrating that centromeric chromatin is remodelled to have a transcription dependent 'open' structure. + +Previous studies have suggested that CENP- A containing chromatin is folded differently and this could be linked to DNA supercoiling50. If DNA is twisted in a right- handed direction it becomes over- wound (positive supercoiling) whilst twisting in the opposite direction it adopts an under- wound (negatively supercoiled) configuration51. Our earlier work showed that the level of supercoiling is transcription dependent52 so we hypothesised that low level RNA polymerase II activity could impact the local DNA configuration. Using biotinylated 4,5,8- trimethylpsoralen (bTMP) as a DNA structure probe52 centromeric chromatin was found to be enriched in negatively supercoiled DNA (Fig. 4C), in a transcription dependent manner. These data further indicate that RNA polymerase is not present as a static component but is engaged in active transcription that remodels DNA and local chromatin fibre structure. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig.4. Centromeric chromatin has a transcription dependent underwound and disrupted chromatin fibre structure.
+ +(A) Top, schematic indicating the lack of understanding of chromatin structure at canonical centromeric chromatin (HSA3) and after centromerisation (Neo3). (B) Distribution of disrupted chromatin fibres across a 1.2 Mb region on HSA3 and Neo3, in the presence or absence of transcription inhibition (5 h a-amanitin treatment). (C) Organisation of negative (under-wound) and positive (over-wound) supercoils mapped by bTMP binding before or after transcription inhibition (5 h a-amanitin treatment). + +<--- Page Split ---> + +## Centromere formation is not accompanied by nucleosome repositioning + +Satellite- containing pericentromeric chromatin fibres found at canonical centromeres have a rigid rod- like structure? that may facilitate kinetochore formation and increase the fidelity of chromosome segregation. In contrast, neocentromeres do not have repetitive \(\alpha\) - satellite DNA sequences34- 36 to precisely position nucleosomes, so it was important to ask if centromerisation, per se, affected nucleosome positioning within the core or flanking DNA sequences. Mono and di- nucleosome fragments prepared from HSA3- and Neo3- containing nuclei using DFF nuclease (Supplementary Fig. 4A- B) were selected using biotinylated baits (Supplementary Fig. 4C) and deep sequenced. After centromerisation the size of the nucleosomal fragments did not change, despite centromeric nucleosomes being enriched in CENP- A (Supplementary Fig. 4D- F) and no difference in nucleosome positioning (Fig. 5A) or periodicity (Supplementary Fig. 4G) was observed. + +Despite no apparent change in the nucleosomal arrangement we speculated that over time pericentromeric chromatin may be remodelled to adopt a more compact configuration (scenario 2; Fig. 5B), analogous to the structure observed at canonical pericentromeres7. Consistent with this idea a 250 kb region at the pericentromeric boundary had a compact chromatin fibre structure (Fig. 5B) coincident with H3K36me3 (Fig. 3A) a mark that has previously been observed at constitutive and facultative heterochromatin43. This indicates that H3K9me2/3 heterochromatin marks are not sufficient to generate a compact chromatin fibre structure, but pericentromeric chromatin can be remodelled to form a structure analogous to canonical satellite- containing pericentromeric chromatin. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5. Compact heterochromatin in pericentromeric domain
+ +(A) Nucleosome dyad coverage on the HSA3 and Neo3 chromosome at the neocentromere domain (blue) with enlarged region (below) (B) Top, diagram showing potential scenarios for chromatin remodelling at pericentromere after centromerisation. Bottom, distribution of disrupted chromatin fibres, across the 3q24 locus in the HSA3 and Neo3 chromosomes. Blue bar corresponds to neocentromere and the yellow domain marks the silenced pericentromeric region. + +<--- Page Split ---> + +## Decompacted large-scale centromeric chromatin + +As previous studies indicated an inter- relationship between different levels of chromatin organisation47,48,52, we speculated that after centromere formation local changes in chromatin structure might be propagated and influence large- scale chromatin compaction53. To directly test this hypothesis 3D DNA FISH, with pairs of differentially labelled fosmid probes \(\approx 300\) kb apart (Supplementary Table 1), was used to ascertain large- scale chromatin compaction at the core and pericentromeric chromatin domains (Fig. 6A). In the hamster- hybrid cells harbouring HSA3 and Neo3 chromosomes there was no apparent change in compaction in pericentromeric regions but a pronounced decompaction at the centromeric core (Fig. 6B). To ensure this difference was not a consequence of comparing different cell lines the analysis was repeated in the parental cells using CENP- C immuno- FISH to discriminate between the Neo3 and HSA3 chromosomes. This similarly revealed a significant large- scale chromatin decompaction after centromerisation (Supplementary Fig. 5A- B) showing that remodelling occurs at multiple levels of centromere organisation. Centromeric large- scale chromatin structure was also transcription dependent, with both \(\alpha\) -amanitin and flavopiridol treatment causing chromatin compaction (Fig. 6C; Supplementary Fig. 5C). As hemomycin treatment (introduces nicks) also caused large- scale chromatin to compact (Fig. 6C), this suggested the fibres were under topological strain, consistent with being negatively supercoiled (Fig. 3C). Although we were unable to find evidence for centromeric derived transcripts (Fig. 2D) we speculated that transcripts may act locally to impact chromatin structure54. Consistent with this idea, RNase H treatment (hydrolyses RNA in the context of a DNA/RNA hybrid) compacted centromeric but not pericentromeric chromatin structure (Fig. 6D) suggesting that a transcription- dependent RNA component stabilised decompacted centromeric chromatin. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig.6. Large-scale neocentromere chromatin fibre decompaction.
+ +(A) Diagram showing fosmid FISH probes (grey circles) surrounding the CENPC-marked centromeric core domain (blue). (B) Top, representative images of 3D-FISH on HSA3 and Neo3 chromosomes hybridized to probe B (red) and C (green) in single chromosome human-hamster hybrid nuclei, counterstained with DAPI. Bar is 5 μm. Bottom, boxplot showing interprobe distance measurements (μm) for pairs of fosmid probes. Horizontal line, median; whiskers, 1.5 interquartile range; p-values are for a Wilcoxon test; *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001. (C) Boxplot showing interprobe distance (μm) between the BC (centromere) fosmid probes in HSA3 (white) and Neo3 (grey) chromosomes in GM10253A and HybNeo3 cell lines respectively, after treatment with a-amanitin (5 h), flavopiridol (3 h) or bleomycin (10 min). (D) Boxplot showing interprobe distance (μm) for the BC (centromere) and DG (pericentromere) pairs of fosmid probes on HSA3 or Neo3 chromosomes following RNase H treatment. + +<--- Page Split ---> + +## Inherent genome instability at a neocentromere + +Whilst neocentromeres form fully functional kinetochores that are stably propagated55, they are still associated with higher chromosome mis- segregation rates and mitotic errors56,57. To quantify neocentromere stability we examined the chromosome architecture and copy number of the centromeric and pericentromeric domains in cells propagated for different amounts of time (Fig. 7A- B). At low passage almost all chromosomes had a normal structure (Fig. 7B and Supplementary Fig. 6A) but after approximately 50 passages the pericentromeric region upstream of the neocentromere had undergone break and or fusion events in \(70\%\) of cells. Copy number analysis was used to estimate the position of breakage events (Supplementary Fig. 6B) which were predominantly in the vicinity of the centromere. However, a region of pronounced DNA amplification was visible near the DIPK2A gene located proximal to the pericentromere boundary (Supplementary Fig. 6C). FISH of the cell population with the border fosmid indicated this region was amplified and located on other chromosomes but strikingly was rarely visible \((2\%)\) after chromosome breakage and fusion events (Supplementary Fig. 6D) indicating that breaks occurred in the pericentromeric chromatin domain upstream of the neocentromere. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Fig.7. Neocentromere chromosome instability.
+ +(A) Neo3 ideogram and BAC probe used to analyse chromosome stability. (B) Left, Representative 2D DNA FISH images of Neo3 chromosome architecture. Neo3 chromosomes were hybridized with a human chromosome 3 paint (green) and neocentromere BAC probe (red) and scored as normal or abnormal (displaying neocentromere instability in the form of breaks, fusions to hamster chromosomes (dark grey) or duplications). Right, graph quantifying loss in chromosome stability as passage number increased. Bar is 2 μm. (C) Model showing transcription dependent centromere remodelling to a disrupted euchromatin state (grey) and pericentromeric chromatin repression to form heterochromatin (yellow). We suggest that disrupted euchromatin provides a suitable foundation for a high-fidelity kinetochore whilst heterochromatin and the accumulation of satellite sequences generates surrounding mechanical rigidity. + +<--- Page Split ---> + +## Discussion + +For efficient and accurate chromosome segregation kinetochores must assemble onto CENP- A- containing chromatin. This happens in two stages, initially the constitutive centromere- associated network (CCAN)58 binds to centromeric chromatin via CENP- C14. Then, in mitosis, the complete kinetochore is assembled to provide an attachment site for the microtubules. To achieve these two steps with high fidelity it has been speculated that the underlying chromatin must adopt a special or distinct chromatin structure2 (Fig 7C). + +Previous studies have indicated that centromeric chromatin is associated with histone marks that are reflective of actively transcribed chromatin12,59. After centromere formation we observe a strong enrichment of active histone marks (Fig. 1) and a significant recruitment of RNA polymerase (Fig. 2). Concomitantly the chromatin fibre is remodelled to form a disrupted or 'open' structure (Fig. 4A); but what are the mechanisms for forming disrupted chromatin and what role might it play in kinetochore formation and function? Gene- rich47 and transcriptionally active48 chromatin are reported to form disrupted chromatin fibre structures, through a combination of mechanisms. At typical euchromatic regions irregularly positioned nucleosomes are less able to fold into an organised chromatin structure, but at centromeric regions it is known that satellite DNA sequences position nucleosomes regularly and form a rigid chromatin fibre7,10. After centromere formation there was no apparent remodelling of nucleosomes (Fig. 5A) suggesting that this is not the basis for the disrupted chromatin fibre. Alternatively, it is reported that CENP- A nucleosome tails bind DNA less tightly to form more dynamic nucleosomal structures and may also interfere with linker histone binding, to promote chromatin fibre opening60. Although CENP- A nucleosome properties might influence chromatin fibre folding it appears that the disrupted chromatin structure is strongly transcription dependent (Figure 4B). We therefore speculate that transcription could disrupt nucleosome positioning through the activity of RNA polymerase and recruitment of chromatin remodelling machines. + +At centromeres disrupted chromatin fibres may serve two purposes. Firstly, a disrupted structure might increase the likelihood of proteinaceous components of the CCAN, such as CENP- C, bind to CENP- A- containing centromeric nucleosomes14 but may also facilitate other structural components such as RNA to interact due to increased access to histone proteins. Our data indicates that centromere chromatin structure is RNA dependent demonstrating that additional nucleic acids may play a structural role (Fig. 6). This is consistent with previous studies which show that RNA can interact with HP1 to facilitate heterochromatin folding54. Secondly, depending on the nature of centromeric chromatin, a flexible fibre might be able to adopt a structure that is able to form a better scaffold for kinetochore formation. For example, in one model it has been suggested that centromeric chromatin might form a layered configuration termed a boustrophedon 61 whilst in an alternate model the centromeric chromatin might be folded into small loops 3, creating an invaginated structure that CCAN proteins can securely attach to. Presumably both large- scale structures would form more readily from a flexible chromatin fibre. + +Another recent idea posits that heterochromatin can undergo liquid- liquid phase separation (LLPS)62,63 to form a gel- like microenvironment64 that could facilitate kinetochore assembly. LLPS often occurs through non- covalent interactions that can be modulated by the local concentration of RNA and proteinaceous components such as HP1 or the long tails of histones65,66, so might be facilitated by a disrupted underlying chromatin structure. + +This study indicates that the centromeric chromatin core has a flexible disrupted structure (Fig 1 and 4) flanked by transcriptionally repressed pericentromeric chromatin (Fig 1 and 5), to form a two- domain model (Fig. 7C). At a newly formed (neo)centromere these flanking regions are epigenetically remodelled and transcriptionally repressed (Fig 1), presumably by an activity emanating from the core, but epigenetic remodelling was insufficient to completely compact chromatin fibre structure (Fig. 5), as observed for satellite- containing heterochromatin7. Evidence from evolutionary new centromeres (ENCs) indicate that satellite sequences accumulate over a long period28- 30. Consistently only a small region of pericentromeric chromatin had a compact structure (Fig 5), showing that the neocentromere at 3q24 is young and has not yet matured to adopt a compact structure. Concomitantly, it exhibited a low level of aneuploidy suggesting that ENCs recruit satellite DNA sequences over time to progressively form a stable chromatin platform7 (Fig. 7C). We therefore suggest that centromere formation is accompanied by significant transcriptional, epigenetic and chromatin fibre remodelling to form a suitable environment for kinetochore assembly, and that over time the chromatin fibre structure matures to support high fidelity chromosome segregation in mitosis. + +<--- Page Split ---> + +## Acknowledgements + +We would like to thank all members of our groups for useful discussions and to Alison Pidoux and Jim Allan for critical comments on the manuscript. This work was funded by the UK Medical Research Council (MR/ J00913X/1; MC_UU_00007/13). + +## Author Contributions + +C.N., C.H., C.R.C., A.B., R-S.N., and S.P. undertook experiments; C.N., G.R.G. and N.G. analysed data; M.R. and N.G. designed experiments; C.N. and N.G. wrote the manuscript with input from all authors. + +## References + +1. Cleveland, D. W., Mao, Y. & Sullivan, K. F. Centromeres and kinetochores: From epigenetics to mitotic checkpoint signaling. Cell 112, 407–421 (2003). +2. Nagpal, H. & Fierz, B. The Elusive Structure of Centro-Chromatin: Molecular Order or Dynamic Heterogeneity? Journal of Molecular Biology vol. 433 (2021). +3. Black, B. E. & Cleveland, D. W. Epigenetic centromere propagation and the nature of CENP-A nucleosomes. Cell 144, 471–479 (2011). +4. Willard, H. F. Chromosome-Specific Organization of Human Alpha Satellite Dna. American Journal Human Genetics 37, 524–532 (1985). +5. Miga, K. H. Centromere studies in the era of ‘telomere-to-telomere’ genomics. Experimental Cell Research vol. 394 (2020). +6. Guenatri, M., Bailly, D., Maison, C. & Almouzni, G. Mouse centric and pericentric satellite repeats form distinct functional heterochromatin. Journal of Cell Biology 166, 493–505 (2004). +7. Gilbert, N. & Allan, J. Distinctive higher-order chromatin structure at mammalian centromeres. Proceedings of the National Academy of Sciences of the United States of America 98, 11949–11954 (2001). +8. Horz, W., Fittler, F. & Zachau, H. G. Sequence specific cleavage of African green monkey alpha-satellite DNA by micrococcal nuclease. Nucleic acids research 11, 4275–4285 (1983). +9. Linxweiler, W. & Horz, W. Reconstitution experiments show that sequence-specific histone-DNA interactions are the basis for nucleosome phasing on mouse satellite DNA. Cell 42, 281–290 (1985). +10. Kruithof, M. et al. Single-molecule force spectroscopy reveals a highly compliant helical folding for the 30-nm chromatin fiber. 16, 534–540 (2009). +11. Lam, A. L., Boivin, C. D., Bonney, C. F. M., Rudd, K. & Sullivan, B. A. Human centromeric chromatin is a dynamic chromosomal domain that can spread over noncentromeric DNA. Proceedings of the National Academy of Sciences 103, (2006). +12. Sullivan, B. A. & Karpen, G. H. Centromeric chromatin exhibits a histone modification pattern that is distinct from both euchromatin and heterochromatin. Nature Structural and Molecular Biology 11, (2004). +13. Blower, M. D., Sullivan, B. A. & Karpen, G. H. Conserved organization of centromeric chromatin in flies and humans. Developmental Cell 2, 319–330 (2002). +14. Carroll, C. W., Milks, K. J. & Straight, A. F. Dual recognition of CENP-A nucleosomes is required for centromere assembly. Journal of Cell Biology 189, (2010). +15. Du, Y., Topp, C. N. & Dawe, R. K. DNA binding of centromere protein C (CENPC) is stabilized by single-stranded RNA. PLoS Genetics 6, (2010). +16. McNulty, S. M., Sullivan, L. L. & Sullivan, B. A. Human Centromeres Produce Chromosome-Specific and Array-Specific Alpha Satellite Transcripts that Are Complexed with CENP-A and CENP-C. Developmental Cell 42, (2017). +17. Ribeiro, S. A. et al. Condensin regulates the stiffness of vertebrate centromeres. Molecular Biology of the Cell 20, (2009). +18. Bobkov, G. O. M., Gilbert, N. & Heun, P. Centromere transcription allows CENP-A to transit from chromatin association to stable incorporation. Journal of Cell Biology 217, 1957–1972 (2018). +19. Singh, P. P. et al. Hap2-Ino80-facilitated transcription promotes de novo establishment of CENP-A chromatin. Genes and Development 34, 226–238 (2020). +20. Bergmann, J. H. et al. Epigenetic engineering: Histone H3K9 acetylation is compatible with kinetochore structure and function. Journal of Cell Science 125, (2012). +21. Bergmann, J. H. et al. Epigenetic engineering shows H3K4me2 is required for HJURP targeting and + +<--- Page Split ---> + +CENP- A assembly on a synthetic human kinetochore. EMBO Journal 30, (2011).22. Nakano, M. et al. Inactivation of a Human Kinetochore by Specific Targeting of Chromatin Modifiers. Developmental Cell 14, (2008).23. Molina, O. et al. Epigenetic engineering reveals a balance between histone modifications and transcription in kinetochore maintenance. Nature Communications 7, (2016).24. Ohzeki, J. I. et al. Breaking the HAC Barrier: Histone H3K9 acetyl/methyl balance regulates CENP- A assembly. EMBO Journal 31, (2012).25. Decombe, S. et al. Epigenetic rewriting at centromeric DNA repeats leads to increased chromatin accessibility and chromosomal instability. Epigenetics and Chromatin 14, (2021).26. Naughton, C. & Gilbert, N. Centromere chromatin structure – Lessons from neocentromeres. Experimental Cell Research 111899 (2020) doi:10.1016/j.yexcr.2020.111899.27. Rocchi, M., Archidiacono, N., Schempp, W., Capozzi, O. & Stanyon, R. Centromere repositioning in mammals. Heredity vol. 108 (2012).28. Ventura, M. et al. Evolutionary formation of new centromeres in macaque. Science 316, (2007).29. Locke, D. P. et al. Comparative and demographic analysis of orang- utan genomes. Nature 469, (2011).30. Wade, C. M. et al. Genome sequence, comparative analysis, and population genetics of the domestic horse. Science 326, (2009).31. Nishimura, K., Komiya, M., Hori, T., Itoh, T. & Fukagawa, T. 3D genomic architecture reveals that neocentromeres associate with heterochromatin regions. Journal of Cell Biology 218, (2019).32. Murillo- Pineda, M. et al. Induction of spontaneous human neocentromere formation and long- term maturation. Journal of Cell Biology 220, (2021).33. Ventura, M. et al. Recurrent sites for new centromere seeding. Genome Research 14, 1696–1703 (2004).34. Voullaire, L. E., Slater, H. R., Petrovic, V. & Choo, K. H. A. A functional marker centromere with no detectable alpha- satellite, satellite III, or CENP- B protein: Activation of a latent centromere? American Journal of Human Genetics 52, 1153–1163 (1993).35. Depinet, T. W. et al. Characterization of neo- centromeres in marker chromosomes lacking detectable alpha- satellite DNA. Human molecular genetics 6, 1195–1204 (1997).36. Wandall, A., Tranebjærg, L. & Tommerup, N. A neocentromere on human chromosome 3 without detectable alpha- satellite DNA forms morphologically normal kinetochores. Chromosoma 107, 359–365 (1998).37. Purgato, S. et al. Centromere sliding on a mammalian chromosome. Chromosoma 124, 277–287 (2015).38. Yao, J. et al. Plasticity and epigenetic inheritance of centromere- specific histone H3 (CENP- A)- containing nucleosome positioning in the fission yeast. The Journal of biological chemistry 288, 19184–96 (2013).39. Hori, T. et al. Constitutive centromere- associated network controls centromere drift in vertebrate cells. Journal of Cell Biology 216, 101–113 (2017).40. Raiser, R. et al. Enhancer Activity Requires CBP/P300 Bromodomain- Dependent Histone H3K27 Acetylation. Cell Reports 24, 1722–1729 (2018).41. Chan, F. L. et al. Active transcription and essential role of RNA polymerase II at the centromere during mitosis. Proceedings of the National Academy of Sciences of the United States of America 109, 1979–1984 (2012).42. Shukla, M. et al. Centromere DNA Destabilizes H3 Nucleosomes to Promote CENP- A Deposition during the Cell Cycle. Current Biology 28, 3924–3936 (2018).43. Chantalat, S. et al. Histone H3 trimethylation at lysine 36 is associated with constitutive and facultative heterochromatin. Genome research 21, 1426–1437 (2011).44. Cuddapah, S. et al. Global analysis of the insulator binding protein CTCF in chromatin barrier regions reveals demarcation of active and repressive domains. Genome research 19, 24–32 (2009).45. Narendra, V. et al. CTCF establishes discrete functional chromatin domains at the Hox clusters during differentiation. Science (New York, N.Y.) 347, 1017–1021 (2015).46. Paldi, F. et al. Convergent genes shape budding yeast pericentromeres. Nature 582, 119–123 (2020).47. Gilbert, N. et al. Chromatin architecture of the human genome: Gene- rich domains are enriched in open chromatin fibers. Cell 118, 555–566 (2004).48. Naughton, C., Sproul, D., Hamilton, C. & Gilbert, N. Analysis of Active and Inactive X Chromosome Architecture Reveals the Independent Organization of 30 nm and Large- Scale Chromatin Structures. Molecular Cell 40, 397–409 (2010).49. Ghirlando, R. & Felsenfeld, G. Hydrodynamic studies on defined heterochromatin fragments support a + +<--- Page Split ---> + +30- nm fiber having six nucleosomes per turn. Journal of molecular biology 376, 1417- 1425 (2008). 50. Furuyama, T. & Henikoff, S. Centromeric Nucleosomes Induce Positive DNA Supercoils. Cell 138, 104- 113 (2009). 51. Gilbert, N. & Allan, J. Supercoiling in DNA and chromatin. Current Opinion in Genetics and Development 25, 15- 21 (2014). 52. Naughton, C. et al. Transcription forms and remodels supercoiling domains unfolding large- scale chromatin structures. Nature Structural and Molecular Biology 20, 387- 395 (2013). 53. Tolomeo, D. et al. Epigenetic origin of evolutionary novel centromeres. Scientific Reports 7, 41980- 41980 (2017). 54. Maison, C. et al. Higher- order structure in pericentric heterochromatin involves a distinct pattern of histone modification and an RNA component. 30, 329- 334 (2002). 55. Saffery, R. et al. Human centromeres and neocentromeres show identical distribution patterns of \(>20\) functionally important kinetochore- associated proteins. Human Molecular Genetics 9, 175- 185 (2000). 56. Fachinetti, D. et al. DNA Sequence- Specific Binding of CENP- B Enhances the Fidelity of Human Centromere Function. Developmental Cell 33, 314- 327 (2015). 57. Bassett, E. A. et al. Epigenetic centromere specification directs aurora B accumulation but is insufficient to efficiently correct mitotic errors. Journal of Cell Biology 190, 177- 185 (2010). 58. Hara, M. & Fukagawa, T. Critical Foundation of the Kinetochore: The Constitutive Centromere- Associated Network (CCAN). Progress in molecular and subcellular biology vol. 56 (2017). 59. Hori, T. et al. Histone H4 Lys 20 Monomethylation of the CENP- A Nucleosome Is Essential for Kinetochore Assembly. Developmental Cell 29, 740- 749 (2014). 60. Roulland, Y. et al. The Flexible Ends of CENP- A Nucleosome Are Required for Mitotic Fidelity. Molecular Cell 63, (2016). 61. Ribeiro, S. A. et al. A super- resolution map of the vertebrate kinetochore. Proceedings of the National Academy of Sciences of the United States of America 107, (2010). 62. Larson, A. G. et al. Liquid droplet formation by HP1α suggests a role for phase separation in heterochromatin. Nature 547, 236- 240 (2017). 63. Strom, A. R. et al. Phase separation drives heterochromatin domain formation. Nature 547, 241- 245 (2017). 64. Michieletto, D. & Gilbert, N. Role of nuclear RNA in regulating chromatin structure and transcription. Current Opinion in Cell Biology 58, 120- 125 (2019). 65. Gibson, B. A. et al. Organization of Chromatin by Intrinsic and Regulated Phase Separation. Cell 179, 470- 484. e21 (2019). 66. Turner, A. L. et al. Highly disordered histone H1- DNA model complexes and their condensates. Proceedings of the National Academy of Sciences of the United States of America 115, 11964- 11969 (2018). 67. Mitchell, A. R., Gosden, J. R. & Miller, D. A. A cloned sequence, p82H, of the alphoid repeated DNA family found at the centromeres of all human chromosomes. Chromosoma 92, 369- 377 (1985). 68. Kimura, H., Hayashi- Takanaka, Y., Goto, Y., Takizawa, N. & Nozaki, N. The organization of histone H3 modifications as revealed by a panel of specific monoclonal antibodies. Cell structure and function 33, 61- 73 (2008). 69. Trazzi, S. et al. The C- terminal domain of CENP- C displays multiple and critical functions for mammalian centromere formation. PloS one 4, e5832 (2009). 70. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics (Oxford, England) 25, 2078- 2079 (2009). 71. Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics (Oxford, England) 26, 841- 842 (2010). 72. Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics (Oxford, England) 30, 923- 930 (2014). + +<--- Page Split ---> + +## Methods + +## Cell lines + +The human parental lymphoblastoid cell line was grown in RPMI 1640 with L- Glutamine (Life Technologies) supplemented with \(20\%\) FCS, penicillin (100 U.ml- 1) and streptomycin (100 \(\mu \mathrm{g}.\mathrm{ml}^{- 1}\) ). Human/Hamster hybrid cell lines GM10253A and HybNeo3, harbouring HSA3 and Neo3 respectively, were grown in the same media but with \(10\%\) FCS. All cells were maintained at \(37^{\circ}C\) in an atmosphere of \(5\%\) CO2 and subjected to regular mycoplasma testing. Transcription was blocked by adding \(\alpha\) - amanitin (50 \(\mu \mathrm{g}.\mathrm{ml}^{- 1}\) ) or flavopiridol (100 \(\mu \mathrm{M}\) ) to cells for the times indicated. + +## DNA fluorescence in situ hybridization (FISH) + +FISH was performed on both metaphase chromosome spreads and interphase nuclei. Metaphase chromosomes were prepared by treating cells with \(0.1\mu \mathrm{g}\cdot \mathrm{m}\mathrm{l}^{- 1}\) Colcemid (Life Technologies, Cat No 15210- 040) for \(30\mathrm{min}\) (hybrid cells) or \(4\mathrm{hr}\) (parental lymphoblastoid cells) prior to harvest to induce mitotic arrest and increase the number of mitotic cells. Cells were recovered by trypsin treatment and washed in PBS. Hypotonic solution, containing \(75\mathrm{mM}\) KCl was added drop wise to a final \(5\mathrm{ml}\) volume. Hypotonic treatment was performed at room temperature for \(10\mathrm{min}\) , after which cells were pelleted by centrifugation at \(1200\mathrm{rpm}\) (250g) for \(5\mathrm{min}\) and fixed three times in \(5\mathrm{ml}\) of a freshly prepared solution of \(3:1\) ratio (v/v) methanol: acetic acid (MAA). The MAA fixative was added to the cell pellet dropwise with constant agitation. Chromosome preparations were stored at \(- 20^{\circ}C\) . To prepare slides with metaphase spreads, metaphase chromosome preparations were dropped onto glass slides. The glass slides were pre- treated in a dilute solution of HCl in Ethanol for at least one hour prior to use. The chromosome preparations were pelleted by centrifugation at \(1500\mathrm{rpm}\) (350g) for \(5\mathrm{min}\) and resuspended in freshly prepared MAA solution until the suspension became cloudy. Two drops of the suspension were dropped onto a pre- treated glass slide from a height of \(20\mathrm{cm}\) and dried at room temperature overnight before staining or hybridization. + +For 3D FISH on interphase nuclei hybrid cells were grown overnight on glass slides whilst parental non- adherent lymphoblastoid cells \((3\times 10^{4}\) cells) were cytospun onto glass slides at \(600\mathrm{rpm}\) (50g) for \(10\mathrm{min}\) . Slides were rinsed with PBS and fixed in \(4\%\) paraformaldehyde (PFA) for \(10\mathrm{min}\) . Slides were then rinsed with PBS and cells were permeabilized for \(10\mathrm{min}\) on ice with PBS supplemented with \(0.2\%\) triton. After rinsing, slides were stored in PBS (for immunohistochemistry) or \(70\%\) ethanol (for FISH) at \(4^{\circ}C\) . For chromatin nicking and RNase H treatment cells were grown on slides overnight, rinsed gently whilst still in the slide tray three times with PBS and then treated with bleomycin ( \(100\mu \mathrm{M}\) ) in PBS or RNase H ( \(1000\mu \mathrm{m}\) , NEB MO297S) in PBS supplemented with \(0.1\%\) triton, \(1\mathrm{mM}\) \(\mathrm{Ca^{2 + }}\) and \(1\mathrm{mM}\) \(\mathrm{Mg^{2 + }}\) for \(10\mathrm{min}\) at \(37^{\circ}C\) . Slides were then rinsed with PBS and PFA fixed as before. + +FISH was carried out as described52 except that MAA fixed metaphase spreads were denatured for \(1\mathrm{min}\) in \(70\%\) formamide in \(2\times\) SSC, pH 7.5, at \(70^{\circ}C\) and interphase cells (grown on glass slides or cytospun) were \(4\%\) PFA fixed and were denatured for \(45\mathrm{min}\) at \(80^{\circ}C\) . Following denaturation, slides were submerged in ice- cold \(70\%\) ethanol for \(2\mathrm{min}\) and then dehydrated through \(90\%\) and \(100\%\) ethanol for \(2\mathrm{min}\) each at room temperature. Fosmid and BAC clones (BACPAC Genomics) were labelled by nick translation with digoxigenin- 11- dUTP (Roche, #11093070910) or biotin- 16- dUTP (Roche, #11093088910) for antibody based detection as previously described or alternatively directly labelled with Green 500 dUTP (Enzo- 42845) or red- dUTP (ChromaTide Alexa Fluor 594- 5- dUTP C11400). \(\alpha\) - satellite probe \(\mathrm{p82H^{67}}\) was labelled by nick translation. For hybridization, \(150\mathrm{ng}\) of labelled probe was combined with \(5\mu \mathrm{g}\) salmon sperm and \(10\mu \mathrm{g}\) human \(\mathrm{C_{10}t1}\) DNA (Invitrogen, Cat No 15279011). Two volumes of ethanol were added and the probe mix was collected by centrifugation and dried. Dried probes were resuspended in \(10\mu \mathrm{l}\) of hybridization buffer containing \(50\%\) formamide (v/v), \(1\%\) Tween- 20 and \(10\%\) dextran sulfate (Sigma Aldrich, Cat No D8906- 100G) in \(2\times\) SSC. Chromosome 3 paint (XCP3 Green, Metasystems) was supplied already labelled with a green fluorophore and dissolved in hybridization solution and ready to use. Probes were denatured at \(70^{\circ}C\) for \(5\mathrm{min}\) and re- annealed at \(37^{\circ}C\) for \(15\mathrm{min}\) and chilled on ice. Probes were pipetted onto slides and hybridization was performed at \(37^{\circ}C\) overnight. Coverslips were removed and slides were washed four times in \(2\times\) SSC at \(45^{\circ}C\) for \(3\mathrm{min}\) and four times in \(0.1\times\) SSC at \(60^{\circ}C\) for \(3\mathrm{min}\) . Slides were blocked in \(5\%\) milk in \(4\times\) SSC for \(5\mathrm{min}\) at RT. Detection of biotin label was performed with sequential layers of fluorescein (FITC)- conjugated avidin, biotinylated anti- avidin and a further layer of FITC- avidin. Digoxigenin was detected with sequential layers of Rhodamine- conjugated anti- digoxigenin and Texas- Red (TR) - conjugated anti- sheep IgG. Slides were DAPI stained, mounted in Vectashield (Vector Laboratories, Cat No H- 1000) Epifluorescent images were acquired using a Photometrics Coolsnap HQ2 CCD camera on a Zeiss Axioplan II fluorescence microscope + +<--- Page Split ---> + +with a Plan- neofluar/apochromat 100x objective (Carl Zeiss, Cambridge, UK), a Mercury Halide fluorescent light source (Exfo Excite 120, Excelitas Technologies) and Chroma #83000 triple band pass filter set (Chroma Technology Corp., Rockingham, VT) with the single excitation and emission filters installed in motorised filter wheels (Prior Scientific Instruments, Cambridge, UK). Data was collected using Micromanager software and analyzed using custom scripts in iVision. + +For four- color 3D Immuno- FISH immunocytochemistry for CENP- C was performed prior to FISH. Slides stored in PBS were blocked in \(5\%\) horse serum then incubated overnight with anti- CENP- C antibody (1:200, gift, W. Earnshaw) before 1 hr incubation with Texas Red labelled anti- rabbit (1:100, Jackson ImmunoResearch Laboratories) secondary antibody. CENP- C signal was fixed with \(4\%\) PFA for 45 min followed by denaturation with \(50\%\) formamide in \(2 \times\) SSC, pH 7.5, at \(80^{\circ}\mathrm{C}\) for 45 min. Slides were then dipped briefly in \(2 \times\) SSC followed by incubation overnight at \(37^{\circ}\mathrm{C}\) with pairs of labelled fosmid probes. Slides were then washed and processed as above. Epifluorescent images were acquired using a Photometrics Coolsnap HQ2 CCD camera and a Zeiss Axiolmager A1 fluorescence microscope with a Plan Apochromat \(100 \times\) 1.4NA objective, a Nikon Intensilight Mercury based light source (Nikon UK Ltd, Kingston- on- Thames, UK) and a Chroma 89000ET single excitation and emission filters (Chroma Technology Corp., Rockingham, VT) with the excitation and emission filters installed in Sutter motorised filter wheels (Sutter Instrument, Novato, CA). A piezoelectrically driven objective mount (PIFOC model P- 721, Physik Instruments GmbH & Co, Karlsruhe) was used to control movement in the z dimension. Hardware control, image capture and analysis were performed using Nikon Nis- Elements software (Nikon UK Ltd, Kingston- on- Thames, UK) and Velocity (Perkin- elmer, Inc.). Images were deconvolved using a calculated point spread function with the constrained iterative algorithm of Velocity. Image analysis was carried out using Imaris software that calculate the distance between two fosmid probe signals. The significance of compaction between pairs of probes was tested using the nonparametric Wilcoxon test for paired samples, \(P < 0.05\) was considered significant. FISH probes are described in Supplementary Table 1. + +## Immunocytochemistry + +Metaphase chromosome spreads derived from parental cells were rinsed in PBS, blocked in \(5\%\) horse serum then incubated overnight with anti- CENP- C antibody (1:200, gift from W. Earnshaw) and anti- RNA pol II (1:1000, Abcam Ab24758) antibody. Secondary antibodies were FITC- conjugated anti- mouse and Texas Red- conjugated anti- rabbit antibodies (1:150, Jackson ImmunoResearch Laboratories). Slides were DAPI stained, mounted in Vectashield (Vector Laboratories, Cat No H- 1000) and imaged on a Zeiss epifluorescence microscope using a \(100 \times\) objective. + +## Chromatin immunoprecipitation + +ChIP was done as described except that a Soniprep 150 (Sanyo) was used for sonication. In brief, cells (5- 6 x \(10^{6}\) in 10 cm dishes) were cross- linked with \(10 \mathrm{ml} 1\%\) formaldehyde (Sigma) in medium for 5 min at room temperature and then incubated in \(10 \mathrm{ml} 200 \mathrm{mM}\) glycine in medium for 5 min. Cells were rinsed twice with PBS and incubated with \(7 \mathrm{ml}\) lysis buffer ( \(10 \mathrm{mM}\) Tris- HCl (pH 7.5), \(10 \mathrm{mM}\) NaCl and \(0.5\%\) NP- 40) for 10 min at room temperature with mild rotation. This lysis buffer was then aspirated off and cells were scraped into 1 ml lysis buffer and centrifuged at \(3000 \mathrm{rpm}\) for 3 min at \(4^{\circ}\mathrm{C}\) . Cell pellet was resuspended in \(100 \mu \mathrm{l}\) SDS lysis buffer ( \(50 \mathrm{mM}\) Tris- HCl (pH 7.5), \(10 \mathrm{mM}\) EDTA and \(1\%\) SDS) and mixed by pipetting. \(400 \mu \mathrm{l}\) ChIP dilution buffer ( \(50 \mathrm{mM}\) Tris- HCl (pH 7.5), \(167 \mathrm{mM}\) NaCl, \(1.1\%\) Triton X- 100, \(0.11\%\) sodium deoxycholate and protease inhibitor cocktail (complete EDTA- free; Roche)) was added before sonication (fifteen times for 20 seconds at \(2 \mu \mathrm{m}\) ). After centrifugation at \(13,000 \mathrm{rpm}\) for 15 min at \(4^{\circ}\mathrm{C}\) to remove the insoluble material the supernatant was removed to a new \(1.5 \mathrm{ml}\) tube and the volume made up to \(500 \mu \mathrm{l}\) with ChIP dilution buffer. \(50 \mu \mathrm{l}\) was removed as input for ChIP and the rest of the sample was incubated with antibody- bound Dynabeads overnight at \(4^{\circ}\mathrm{C}\) with rotation. Dynabeads were prepared in advance by taking \(50 \mu \mathrm{l}\) of beads and washing three times with \(500 \mu \mathrm{l}\) cold RIPA- 150 mM NaCl buffer ( \(50 \mathrm{mM}\) Tris- HCl (pH 7.5), \(150 \mathrm{mM}\) NaCl, \(1 \mathrm{mM}\) EDTA, \(1\%\) Triton X- 100, \(0.1\%\) SDS, \(0.11\%\) sodium deoxycholate and protease inhibitor cocktail). Beads were then incubated with \(500 \mu \mathrm{l}\) cold RIPA- 150 mM NaCl buffer plus antibody for \(2 \mathrm{hr}\) at \(4^{\circ}\mathrm{C}\) with rotation. Beads were then washed three times with \(500 \mu \mathrm{l}\) cold RIPA- 150 mM NaCl buffer and were then ready for overnight incubation with the ChIP sample. Beads were washed sequentially with \(1 \mathrm{ml}\) cold RIPA- 500 mM NaCl ( \(50 \mathrm{mM}\) Tris- HCl (pH 7.5), \(500 \mathrm{mM}\) NaCl, \(1 \mathrm{mM}\) EDTA, \(1\%\) Triton X- 100, \(0.1\%\) SDS, \(0.11\%\) sodium deoxycholate and protease inhibitor cocktail) and twice with \(1 \mathrm{ml}\) TE ( \(10 \mathrm{mM}\) Tris- HCl (pH 8.0) and \(1 \mathrm{mM}\) EDTA). DNA was eluted by the addition of \(200 \mu \mathrm{l}\) ChIP direct elution buffer ( \(10 \mathrm{mM}\) Tris- HCl (pH 8.0), \(300 \mathrm{mM}\) NaCl, \(5 \mathrm{mM}\) EDTA and \(0.5\%\) SDS) and incubated overnight at \(65^{\circ}\mathrm{C}\) . Samples were then treated with DNase- free RNase (Roche; \(5 \mu \mathrm{g} \cdot \mathrm{ml}^{-1}\) ; \(37^{\circ}\mathrm{C}\) ; \(30 \mathrm{min}\) ) and proteinase K ( \(250 \mu \mathrm{g} \cdot \mathrm{ml}^{-1}\) ; \(55^{\circ}\mathrm{C}\) ; \(1 \mathrm{hr}\) ). DNA was extracted with phenol- chloroform- isoamyl + +<--- Page Split ---> + +alcohol (25:24:1) and ethanol precipitated with carrier (1μl glycogen, Invitrogen) on dry ice for 30 min. Following \(70\%\) ethanol wash the DNA pellet was resuspended in \(20\mu l\) water and quantified using a NanoDrop. For microarray hybridization, immunoprecipitated DNA was amplified using whole-genome amplification (Sigma). + +Magnetic sheep anti- mouse IgG beads (Dynabeads, Invitrogen, 11201D) were used for mouse antibodies and protein G beads were used for rabbit antibodies (Dynabeads, Invitrogen, 10004D). Antibodies used were to CENP- A69 and CENP- C (gift from W. Earnshaw), H3K27ac (Abcam, Ab4729), H3K4me2 (Millipore, 07- 030), H3K9me2 (Millipore, 07- 212), H3K9me3 (Abcam, Ab8898), H3K27me3 (Abcam, Ab6002), H3K36me3 (Abcam, Ab9050), CTCF (D31H2) (Cell Signaling Technology, 3418), RNA Polymerase II (Diagenode, C15100055), RNA Polymerase II (gift from H. Kimura). All antibodies were characterized using western blots, and ChIP was optimized using quantitative PCR assays. + +## Analyzing changes in DNA supercoiling + +Biotinylated psoralen (bTMP) uptake was used to analyse DNA supercoiling as previous described52. Cells were treated with \(500\mu g\cdot ml^{- 1}\) of bTMP in PBS for 20 min at room temperature in the dark. bTMP was UV cross- linked to DNA at 360 nm for 10 min. DNA was purified from cells using SDS and proteinase K digestion and extracted using phenol- chloroform- isoamyl alcohol (25:24:1). DNA was fragmented by sonication (thirteen times for 30s at 2 \(\mu m\) ). Biotin incorporation into DNA was detected by dot blotting using alkaline phosphatase- conjugated avidin as a probe. The bTMP- DNA complex in TE was immunoprecipitated using avidin conjugated to magnetic beads (Dynabeads MyOne Streptavidin Invitrogen, 65001) for 2 h at room temperature and then overnight at \(4^{\circ}C\) . Beads were washed sequentially for 5 min each at room temperature with TSE I (20 mM Tris- HCl, pH 8.1, 2 mM EDTA, 150 mM NaCl, 1% Triton X- 100 and 0.1% SDS), TSE II (20 mM Tris- HCl, pH 8.1, 2 mM EDTA, 500 mM NaCl, 1% Triton X- 100 and 0.1% SDS) and buffer III (10 mM Tris- HCl, pH 8.1, 0.25 M LiCl, 1 mM EDTA, 1% NP40 and 1% deoxycholate). Beads were then washed twice with TE buffer for 5 min. To extract DNA and to release psoralen adducts, the samples were boiled for 10 min at \(90^{\circ}C\) in 50 \(\mu l\) of 95% formamide with 10 mM EDTA. Samples were then made up to 200 \(\mu l\) with water, and the DNA was purified using a Qiagen MinElute PCR purification kit. bTMP bound DNA was amplified using whole- genome amplification (Sigma) prior to microarray hybridization. + +## Chromatin Fractionation + +Disrupted or 'open' chromatin was isolated as described previously48. In brief, cell nuclei were digested with micrococcal nuclease and soluble chromatin released overnight followed by fractionation on a 6- 40% isokinetic sucrose gradient in 80 mM NaCl, 0.1 mM EDTA, 0.1 mM EGTA and 250 \(\mu M\) PMSF. DNA purified from gradient fractions was analyzed by electrophoresis through 0.7% agarose in \(1 \times \mathrm{TPE}\) buffer (90 mM Tris- phosphate, 2 mM EDTA) with buffer circulation. Preparative fractionation of DNA from gradient fractions was carried out by pulsed- field gel electrophoresis (PFGE) (CHEF system, Biorad) through 1% low melting point agarose in \(0.5 \times \mathrm{TBE}\) , at 180 V, for 40 hr, with a 0.1- 2 s switching time. Size markers were 1 kb (Promega) and \(\lambda\) - HindIII (NEB) DNA ladders. EtBr- stained gels were scanned using a 473 nm laser and a 580 nm band- pass filter on a Fuji FLA- 3000. DNA of \(\sim 20\) kb, corresponding to "open" chromatin, was isolated by \(\beta\) - agarase (NEB) digestion and amplified by whole genome amplification (Sigma) prior to microarray hybridization. + +## Microarray hybridization, data processing and analysis + +Whole- genome amplified DNA (ChIP/bTMP/open' chromatin) was labelled and hybridised as previously (NSMB) to custom 180K Agilent microarrays ) (7 Mb spanning the neocentromere domain (chr3:142781158- 149782213; GRCh38 (hg38). In brief, 500 ng DNA was random prime labeled (ENZO) with Cy3 (Sample DNA) or Cy5 (Input DNA) and purified on a MinElute PCR purification column (Qiagen). Labeled DNA was diluted in hybridization buffer (Agilent) and hybridized to arrays for 24 h at \(65^{\circ}C\) . Slides were washed according to the manufacturer's instructions and scanned on a Nimblegen Microarray scanner at \(2\mu m\) resolution generating a TIFF file. + +Spot signal intensity was extracted from the TIFF files using Agilent Feature Extraction software and were pre- processed in R using the RINGO bioconductor package to give the raw Cy5 and Cy3 signal intensities for each spot. Individual Cy5 and Cy3 channels were normalized to each other and between arrays using a variance stabilizing algorithm (for bTMP arrays) and loess, vsn (for ChIP arrays) or nimblegen ("open" chromatin arrays) normalized and scaled, using the standard Bioconductor LIMMA package. All arrays were quality controlled by checking array hybridization patterns, analyzing signal profiles and using MA plots. For data analysis log2(sample/input) data was loaded in to the ZOO package in R and for display the data was + +<--- Page Split ---> + +smoothed using a rolling median. + +## RNA extraction and RNA-Seq + +Total RNA was extracted from cells using RNeasy mini kit (Qiagen) with on- column DNase I digestion (RNAse- Free DNase Set, Qiagen). For RT- qPCR RNAs were reverse transcribed (Superscript II, Invitro- gen) using random primers and quantified by qPCR (Fast start SYBR green, Roche). Primer sequences are described below. Ribosomal RNA was depleted using Illumina® Ribo- Zero Plus rRNA Depletion Kit (Illumina, 20040526) following the manufacturer's instructions and libraries for RNA- seq were prepared and indexed using NEBNext® Ultra™ II DNA Library Prep Kit for Illumina® (NEB #E7645L) and NEBNext Singleplex Oligos for Illumina (NEB #E7335, E7500) following the manufacturer's instructions. Libraries were sized and quality controlled on a D1000 Tapestation tape (Agilent). Single- end RNA- seq of 50- bp read length was performed on Illumina Hi- Seq 2000 (UMC, Amsterdam). FASTQ sequence files were obtained and the RNA- seq reads were aligned to the human reference genome (hg38) using TopHat v2 and processed with Samtools v1.670 and the Bedtools "genome coverage" tool71. Aligned BAM files were processed with the Subread v1.5 "feature counts" tool72 to generate FPKM scores against hg38 RefSeq genes. + +## TT-seq + +Nascent RNA was labelled by adding \(500~\mu \mathrm{M}\) 4- thiouridine (4sU)(Sigma, T4509) to cells harbouring HSA3 and Neo3 in T75 flasks and incubating at \(37^{\circ}C\) for 10 min. Media was aspirated and RNA extraction was performed with TR1zol (Invitrogen) following the manufacturers' instructions. After DNase treatment (Turbo DNase, Thermo Fisher Scientific) RNA concentration and purity were determined using a NanoDrop. RNA (70 \(\mu \mathrm{g}\) ) was fragmented in \(100~\mu \mathrm{H}_2\mathrm{O}\) to \(< 1.5\mathrm{kb}\) by 20 cycles of 30 seconds on/30 seconds off at high power in a Biorputer plus and RNA size assessed by agarose gel electrophoresis. Fragmented 4sU labelled RNA was biotinylated by adding \(140~\mu \mathrm{l}\) of EZ- Link Biotin- HPDP (1mg.ml- 1 in dimethylformamide; Pierce, 21341), \(70~\mu \mathrm{l}\) of \(10\times\) biotinylation buffer (100 mM Tris- HCl pH 7.5, \(10~\mathrm{mM}\) EDTA) and \(H_2O\) to a final volume of \(700~\mu \mathrm{l}\) . This was incubated at room temperature for 1.5 h with rotation. Unincorporated biotin- HPDP was removed by two rounds of chloroform extraction with 2 ml Phase lock gel heavy tubes (Eppendorf). RNA was precipitated with 1/10 volume of 5M NaCl and an equal volume of Isopropanol. This was inverted to mix and incubated at room temperature for 10 min followed by centrifugation at \(10,000\mathrm{g}\) for 20 min at room temperature. RNA pellet was washed with \(80\%\) EtOH and centrifuged at 13,000 rpm for 10 min at \(4^{\circ}C\) . RNA was resuspended in \(100~\mu \mathrm{l}\) \(H_2O\) and dissolved by heating to \(40^{\circ}C\) for 10 min with agitation. RNA was then immediately placed on ice and RNA concentration determined by Nanodrop spectrophotometer. Biotinylated 4sU labelled RNAs were then recovered using μMACS Streptavidin MicroBeads (Miltenyi, 130- 074- 101) and separation on a μMACS Separator. For the concentration of total RNA in \(\mu \mathrm{g}\) per sample an equal amount in \(\mu \mathrm{l}\) of Streptavidin microbeads was added. This was incubated at room temperature for 15 min with rotation. μMacs columns were equilibrated with \(900~\mu \mathrm{l}\) room temperature washing buffer (100 mM Tris- HCl pH 7.5, \(10~\mathrm{mM}\) EDTA, \(1\mathrm{M}\) NaCl, \(0.1\%\) Tween20). The RNA/streptavidin bead solution was then applied to the column followed by three washes with \(900~\mu \mathrm{l}\) of washing buffer at \(65^{\circ}C\) and three washes with \(900~\mu \mathrm{l}\) of washing buffer at room temperature. RNA was eluted with \(2\times 100~\mu \mathrm{l}\) of fresh elution buffer (100 mM dithiothreitol in RNase- free \(H_2O\) ) directly into 2 ml lobind tubes (Eppendorf) containing \(700\mu \mathrm{l}\) Buffer RLT (RNeasy MinElute Cleanup Kit, Qiagen). \(500~\mu \mathrm{l}\) of \(100\%\) ethanol was added to the RNA solution, and mixed thoroughly by pipetting before RNA was purified through RNAeasy MinElute Spin Columns. RNA concentration was determined using a Nanodrop and libraries for RNA- seq were prepared and indexed using NEBNext® Ultra™ II Directional RNA Library Prep Kit for Illumina® (NEB #E7645L) and NEBNext Singleplex Oligos for Illumina (NEB #E7335, E7500) following the manufacturer's instructions. Libraries were sized and quality controlled on a D1000 Tapestation tape (Agilent) and Illumina sequencing (paired- end RNA- seq of 50- bp read length) was performed on NovaSeq S1 (Edinburgh Genomics). FASTQ sequence files were obtained and the RNA- seq reads were aligned to a Human(hg38) /Hamster (GCA_003668045.1) hybrid reference genome using Bowtie2 and processed with Samtools v1.670, and the deepTools "bamCoverage" tool71 with RPKM normalisation. + +## Exosome RNA interference + +For siRNA treatment, cells (GM10253A and HybNeo3; \(10\% - 20\%\) confluent) were transfected with \(10~\mathrm{nM}\) Silencer Select Pre- designed siRNA targeting EXOSC3 (Ambion, Life Technologies) using Lipofectamine RNAi MAX (ThermoFisher) \(24\mathrm{h}\) after seeding and again \(48\mathrm{h}\) later. After a further \(48\mathrm{h}\) exosome knockdown was confirmed by western blotting and TTseq was performed. Silencer Select RNA sequence for EXOSC3 were GAGATATATTCAAAGTTGA, part number s83102. The control RNA was Stealth RNAi siRNA Negative Control (ThermoFisher). For western blotting cells were suspended in NuPAGE LDS sample buffer (ThermoFisher) with \(10\mathrm{mM}\) DTT, incubated at \(100^{\circ}C\) for \(5\mathrm{min}\) and sonicated briefly. Protein samples were + +<--- Page Split ---> + +resolved on \(12\%\) bis- tris gels (ThermoFisher) and transferred to Immobilon- P PVDF \(0.45 \mathrm{mm}\) membrane (Merck Millipore) by wet transfer. Membranes were probed with anti- EXOSC3 antibody (Abcam, Ab156683) using standard techniques and detected by enhanced chemiluminescence. + +## Shallow DNA Sequencing + +The hybrid HSA3 and Neo3 cell lines were shallow sequenced to confirm copy number. Genomic DNA was prepared from cells and \(500 \mathrm{ng}\) DNA was fragmented using a Covaris sonicator. Genomic DNA libraries were prepared using Illumina TruSeq Nano DNA LT sample prep kit as per manufacturer's instructions and Illumina sequencing (50- bp, single end reads) was performed on Illumina Hiseq 4000 (VUMC Cancer Centre, Amsterdam). FASTQ sequence files were obtained and reads were aligned to the human reference genome (hg19) using BWA and processed with Samtools v1.6. In R the BAM files were loaded into the Bioconductor package QDNAseq for copy number analysis. Human reference genome HG19 was used here as QDNAseq has pre- calculated bin annotations for genome build hg19. + +## Neocentromere capture DNA sequencing + +NimbleGen Sequence Capture technologies were employed for targeted deep sequencing of the neocentromere domain. Capture probes tiling a 1.5Mb domain across the neocentromere were designed using Nimblegen capture design software and sequence capture performed using this custom SeqCap EZ Choice probe pool and SeqCap EZ HE- Oligo Kit A and SeqCap EZ Accessory Kit (Nimblegen) according to manufacturer's instructions. In brief, genomic DNA (gDNA) from the parental lymphoblastoid cell line was fragmented to \(\sim 200 - 500 \mathrm{bp}\) with 20 cycles of 30 seconds on/30 seconds off on a Biorutter. \(1 \mu \mathrm{g}\) gDNA was then used to prepare the gDNA sample library using NEBNext® Ultra™ II DNA Library Prep Kit for Illumina® (NEB #E7645L) following the manufacturer's instructions. The neocentromere domain was captured by hybridising this gDNA library with the biotinylated SeqCap EZ library. \(342 \mathrm{ng}\) of gDNA library was mixed with \(5 \mu \mathrm{g} \mathrm{C}_{0} \mathrm{t}1\) DNA, 1000 pmol of SeqCap HE Universal Oligo 1 and 1000 pmol SeqCap HE Index Oligo, \(7.5 \mu \mathrm{l} 2 \mathrm{X}\) Hybridisation Buffer and \(3 \mu \mathrm{l}\) Hybridisation Component A. This was vortexed for 10 seconds and centrifuged at maximum speed for 10 seconds before denaturing at \(95^{\circ} \mathrm{C}\) for 10 min. This gDNA/C0t1/Oligo/Hybridisation cocktail was then combined with the SeqCap EZ library (provided as \(4.5 \mu \mathrm{l}\) single- use aliquots in \(0.2 \mathrm{ml}\) tubes), vortexed for 3 seconds and centrifuged at maximum speed for 10 seconds. Hybridization was then performed on a thermocycler at \(47^{\circ} \mathrm{C}\) and incubated for 70 hours. Each hybridization reaction was then bound to streptavidin beads from SeqCap EZ Pure Capture Bead Kit and washed with SeqCap EZ Hybridization and Wash Kit (Nimblegen), following the manufacture's protocol. Captured libraries were re- amplified using Post LM- PCR oligos (Nimblegen) and Q5 High- Fidelity DNA polymerase (NEB) directly from the beads. A mastermix consisting of \(65 \mu \mathrm{l} 2 \times\) NEBNext Ultra II Q5 Master Mix NEB, \(50 \mu \mathrm{l}\) captured library (beads in H20), \(13 \mu \mathrm{l}\) LM- PCR Oligo mix (Oligo 1 & 2, \(2 \mu \mathrm{M}\) final concentration of each)(Nimblegen) was made up to \(50 \mu \mathrm{l}\) with H20, vortexed to mix and then split into \(2 \times 65 \mu \mathrm{l}\) samples for PCR using the following PCR cycling conditions. Initial incubation at \(98^{\circ} \mathrm{C}\) for 30 seconds, 14 cycles of \(98^{\circ} \mathrm{C}\) for 10 seconds, \(65^{\circ} \mathrm{C}\) for 30 seconds and \(72^{\circ} \mathrm{C}\) for 30 seconds. Final incubation of \(72^{\circ} \mathrm{C}\) for 5 min and hold at \(4^{\circ} \mathrm{C}\) . The two PCR reactions were recombined and the captured DNA was purified using 1.8:1 AMPure XP Beads: DNA ratio. Capture efficiency was determined to be between 94 and 117 fold using a Nimblegen Sequence Capture control locus qPCR assay. Neocentromere captured DNA libraries were sized and quality controlled on a D1000 Tapestation tape (Agilent), and paired- end sequenced (50 bp) on an Illumina MiSeq (Edinburgh Genomics). + +## Nucleosome positioning + +Nuclei were extracted from cells carrying HSA3 and Neo3 as described and resuspended in NB- R (85 mM KCl, \(10 \mathrm{mM}\) Tris- HCl [pH 7.6], \(5.5\%\) (w/v) sucrose, \(1.5 \mathrm{mM} \mathrm{CaCl}_{2}\) , \(3 \mathrm{mM} \mathrm{MgCl}_{2}\) , \(250 \mu \mathrm{M} \mathrm{PMSF}\) ). Nuclei (800 \(\mu \mathrm{l}\) at SA260) were digested with DFF nuclease (PMID: 17626049) for increasing amounts of time (100 \(\mu \mathrm{l}\) of digested nuclei were removed to a new tube after 1, 2, 4, 8, 16 and 32 min digestion) at room temperature in the presence of \(100 \mu \mathrm{g} \mathrm{ml}^{- 1}\) RNaseA. Digestion was stopped by adding EDTA to \(10 \mathrm{mM}\) . DNA was purified with SDS/Proteinase K digestion, phenol/chloroform extraction and ethanol precipitation. Agarose gel electrophoresis of DFF digested nuclei confirmed digestion to mono and di- nucleosomes. The 4 min and 8 min samples and the 16 min and 32 min samples were pooled and ran on \(2.5\%\) LMP GTG agarose in \(1 \times\) Sybr Safe (Thermo Fisher) dye. Mono and di- nucleosome DNA bands were excised from the gel and purified by \(\beta\) - agarase (NEB) digestion followed by phenol/chloroform extraction and ethanol precipitation. \(500 \mathrm{ng}\) DNA samples (50ng from the 4 and 8 min DFF digestion pool plus 450 ng of the 16 and 32 min DFF digestion pool) were concentrated to \(55 \mu \mathrm{l}\) volume using 2:1 AMPure XP Beads: DNA ratio for the mono nucleosome samples and 1.6:1 AMPure XP Beads: DNA ratio for the di nucleosome samples. Genomic DNA sample + +<--- Page Split ---> + +libraries were then prepared using NEBNext® Ultra™ II DNA Library Prep Kit for Illumina® (NEB #E7645L) following the manufacturer's instructions. This included adaptor- ligated DNA size selection of between 100- 200 bp for the mono nucleosome samples and 300- 400 bp for the di nucleosome samples. DNA library yield was increased by a further round of PCR with Post LM- PCR oligos (Nimblegen) and Q5 High- Fidelity DNA polymerase (NEB). The PCR reaction consisted of 30 μl of the mono or di nucleosomal DNA libraries (150- 270 ng), 50 μl 2 × NEBNext Ultra II Q5 Master Mix NEB, 5 μl Post LM- PCR oligo mix (Oligo 1 & 2, 2 μM final concentration of each; Nimblegen) and 15 μl H₂O. PCR cycling conditions were an initial incubation at 98°C for 30 seconds, 5 cycles of 98°C for 10 seconds and 65°C for 75 seconds and a final incubation of 65°C for 5 min. DNA was purified using 1.8:1 AMPure XP Beads: DNA ratio and libraries were quantified and sized on a D1000 Tapestation tape (Agilent). Libraries, representative of mono and di nucleosome positions throughout the genome, were pooled in equimolar amounts (150 nM) and then subjected to neocentromere capture (as above) to examine nucleosome positioning across the neocentromere region. 1.25 μg of the mono and di nucleosome library pool was mixed with 5μg C₁₁D DNA, 1000 pmol of SeqCap HE Universal Oligo 1 and 1000 pmol SeqCap HE Index Oligos 14, 16, 18 and 19, 7.5 μl 2 × Hybridisation Buffer and 3μl Hybridisation Component A. This was vortexed for 10 seconds and centrifuged at maximum speed for 10 seconds before denaturing at 95°C for 10 min. This gDNA/C₁₁t/Oligo/Hybridisation cocktail was then combined with the SeqCap EZ library (provided as 4.5μl single- use aliquots in 0.2 ml tubes), vortexed for 3 seconds and centrifuged at maximum speed for 10 seconds. Hybridization was then performed on a thermocycler at 47°C and incubated for 70 hours. Each hybridization reaction was then bound to streptavidin beads from SeqCap EZ Pure Capture Bead Kit and washed with SeqCap EZ Hybridization and Wash Kit (Nimblegen), following the manufacture's protocol. Captured libraries were re- amplified using Post LM- PCR oligos (Nimblegen) and KAPA High- Fidelity DNA polymerase (KAPA Biosystems) directly from the beads. 25 μl 2 × KAPA HiFi Hot- Start ReadyMix, and 5 μl LM- PCR Oligo mix (Oligo 1 & 2, 2μM final concentration of each; Nimblegen) was added to the 20 μl captured library (beads in H₂O), vortexed to mix and PCR amplified using the following PCR cycling conditions. Initial incubation at 98°C for 45 seconds, 14 cycles of 98°C for 15 seconds, 60°C for 30 seconds and 72°C for 30 seconds. Final incubation of 72°C for 1 min and hold at 4°C. Captured DNA was purified using 1.8:1 AMPure XP Beads: DNA ratio. Capture efficiency was determined to be between 120 and 240 fold using a Nimblegen Sequence Capture control locus qPCR assay. Neocentromere captured DNA libraries were sized and quality controlled on a D1000 Tapestation tape (Agilent), and paired- end sequenced (50 bp) on an Illumina MiSeq (Edinburgh Genomics). + +## Nucleosome positioning analysis + +Paired end sequencing reads were mapped to hg38 using Bowtie 2 with high quality (mq > 20) and paired reads selected for further analysis. Start and end positions of reads were extracted from bamfiles using bedtools bamtobed function and analysed in R. The NucleR package was used to calculated nucleosomal dyad positions with 40 bp trimming and coverage, data was formatted in the ZOO package and plotted using the lattice package. The acf function in R was used to calculate nucleosome autocorrelation. + +<--- Page Split ---> +![](images/Supplementary_Figure_2.jpg) + + +![](images/Supplementary_Figure_4.jpg) + + +Supplementary Fig. 1. Cell line model system used to interrogate centromeric chromatin structure and mapping of neocentromere mapping to 3q24 (relates to Figure 1). + +(A) Parental human lymphoblastoid cells with one canonical chromosome 3 (HSA3, purple frame) and one chromosome 3 with the centromere relocated to form a neocentromere at 3q24 (Neo3, orange frame). DNA FISH with a-satellite specific probe (red) and a 3q24 fosmid probe (green). (B) The Neo3 chromosome was retained after fusion of the parental cell line with a hamster cell to create a hybrid line called "HybNeo3". Metaphase spread of HybNeo3 hybridized with human C6t-1 DNA (green) identifying the seven human chromosomes present in this human/hamster hybrid (4,6,8,11,13,18, X and Neo3 (orange frame)) and a human a-satellite specific FISH probe (red). (C) The GM10253A hybrid cell line has a single canonical human chromosome 3 (HSA3). Metaphase spread of GM10253A hybridized with a human chromosome 3 paint (green; purple frame). Nuclei are counterstained with DAPI. Bar is 5 μm. (D) Top, ideogram depicting chromosomes HSA3 and Neo3. Bottom, distribution of CENP-A and CENP-C ChIP signal in parental and hybrid cells at 3q24. (E) Top, signal for control IgG ChIP-chip and bottom microarray probe GC composition (%). Blacklisted region (chr3: 147324413-147482213; hg38) is marked by red dashed lines. Neocentromere core (chr3: 147400413-147591023) defined from CENP-C (panel A) is marked in blue. + +<--- Page Split ---> +![](images/Supplementary_Figure_6.jpg) + +
Supplementary Fig. 2. Transcription repression at the pericentromeric domain (Relates to Figure 2) Top, diagram showing individual genes at 3q24, yellow block corresponds to pericentromere domain, blue is the neocentromere. Bottom, RT-qPCR expression data for genes within and bordering the pericentromeric heterochromatic domain, showing expression from HSA3 and gene silencing on Neo3.
+ +<--- Page Split ---> +![PLACEHOLDER_30_0] + + +Supplementary Fig. 3. Isolation of "open" chromatin probes (Relates to Figure 4). Soluble chromatin isolated from hybrid cell nuclei by micrococcal digestion was fractionated by size and structure on a sucrose gradient. (A) Top, agarose gel electrophoresis of DNA purified from sucrose gradient fractions. (B) DNA from gradient fractions was size selected by PFGE. DNA fragments \(\approx 10\) kb longer than the bulk of the DNA signal, corresponding to "open" or disrupted chromatin, was purified from gel slices (yellow boxes). + +<--- Page Split ---> +![PLACEHOLDER_31_0] + +
Supplementary Fig. 4 Legend over page
+ +<--- Page Split ---> + +## Supplementary Fig. 4. Arrangement of nucleosomes around neocentromere (Relates to Figure 5). + +(A) Agarose gel electrophoresis of DFF digested nuclei isolated from HSA3 and Neo3 containing cells. Mono and di-nucleosome fragments were excised, and the DNA extracted using \(\beta\) -agarase. (B) Agarose gel electrophoresis of purified mono- and di-nucleosomes fragments used for nucleosome mapping. (C) Top, ideogram depicting HSA3 and Neo3 chromosomes with enlargement of 3 Mb region around the neocentromere (marked by CENP-C). Bottom, genomic location of the capture baits used to enrich for 1.5 Mb of neocentromeric region. (D) Size distribution of mono and di nucleosomes (bp) isolated from HSA3 and Neo3 cells for 1.5 Mb around the region corresponding to the neocentromere. (E) Mono nucleosome size distribution in 10 kb windows across the 1.5 Mb captured domain (neocentromere marked in blue). (F) Left, mono nucleosome size (median) in 1 kb windows across the 1.5 Mb captured domain and focussed region covering the neocentromere (marked in blue). Right, variance (interquartile range) in mono nucleosome size in 1 kb windows across the 1.5 Mb captured domain and focussed region covering the neocentromere (marked in blue). (G) Autocorrelation of nucleosome dyad coverage at left flank, centromere core and right flank for different lag (bp). + +<--- Page Split ---> +![PLACEHOLDER_33_0] + + +Supplementary Fig. 5. Large scale chromatin is decompacted at the neocentromere in the parental cell line (Relates to Figure 6). + +(A) Top, chromosome 3 ideogram indicating CENP-C immunofluorescence signal (yellow) and the neocentromere specific FISH probes (green and red). Bottom right, representative image of 4 colour 3D immuno-FISH for identifying the chromosome 3 harbouring a neocentromere at 3q24 due to the presence of the CENP-C signal proximal to one pair of fosmid probes. Below left, three colour representation of the same image used for measuring interprobe distance. (B) Boxplot showing interprobe distance measurements (μm) between the pair of fosmid probes (B and C, see Fig 6A) for the HSA3 and Neo3 chromosomes in the parental lymphoblastoid cells. (C) RT-qPCR expression data for genes in the pericentromere region flanking the neocentromere domain in cells carrying the HSA3 and Neo3 chromosomes, following transcription inhibition with a-amanitin (5h) or flavopiridol (3h). + +<--- Page Split ---> +![PLACEHOLDER_34_0] + +
Supplementary Fig. 6. Neocentromere associated genome instability (Relates to Figure 7) (A) Left, representative FISH images of Neo3 metaphase chromosomes hybridised to a human chromosome 3 paint (green) and BAC (red) located at the neocentromere. Chromosome morphology was scored as normal or showing instability: deletions, fusions or duplications. Bar is \(2\mu \mathrm{m}\) . Right, quantification \((\%)\) of different chromosome morphologies with increasing passage number (low \(\sim 10\) , high \(\sim 100\) ) over time. P values are for a \(\chi^2\) test compared to low passage. (B) High passage Neo3 Chromosome copy number (RPKM), with a zoom in of the 3 Mb region around the neocentromere (blue). (C) High passage Neo3 Chromosome copy number (RPKM), with a zoom in of the 7 Mb region around the neocentromere (blue). Locations of DNA FISH probes are shown, neocentromere BAC probe (green) and border fosmid (red). (D) Left, representative FISH images of Neo3 metaphase chromosomes hybridised to a BAC (green) located at the neocentromere and a fosmid (red) located at the border. Bar is \(2\mu \mathrm{m}\) . Right, chromosome morphology was scored and quantified \((\%)\) with increasing passage number over time. Loss of the border probe signal was coincident with fusion of human Neo3 fragment (light grey) to a hamster chromosome (dark grey). Bar is \(2\mu \mathrm{m}\) .
+ +<--- Page Split ---> diff --git a/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/images_list.json b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..fcfe87ad325a5a3a55cf052b86347987a1d8a460 --- /dev/null +++ b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/images_list.json @@ -0,0 +1,182 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 LRBs are required for blue light responses. a, 6-day-old seedlings grown in darkness, red light (20 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ), far-red light (6 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) , long day (16h light / 8h dark), or short day (8h light / 16h dark). b, Hypocotyl length of indicated genotypes of (a). c, 6-day-old seedlings grown in darkness or blue light (10 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ). d, Hypocotyl length of indicated genotypes grown under blue light of different intensities (0, 5, 10, 20, 40 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ) for 6 days. e-g, 6-day old seedlings grown in darkness or blue light (20 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ). f-h, Hypocotyl length of indicated genotypes as in (e) and (g). Standard deviations (n≥20) are shown.", + "footnote": [], + "bbox": [ + [ + 144, + 110, + 850, + 540 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2,", + "footnote": [], + "bbox": [ + [ + 145, + 117, + 853, + 610 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 147, + 290, + 848, + 558 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 LRBs interact with CRY2 in vivo. Confocal microscopy images showing", + "footnote": [], + "bbox": [ + [ + 115, + 115, + 850, + 440 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 LRBs and COP1 are both required for CRY2 ubiquitination.", + "footnote": [], + "bbox": [ + [ + 140, + 152, + 845, + 747 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 The VP motif of CRY2 is required for the CRY2-COP1 interaction but not the CRY2-LRB interaction. a, A schematic diagrams depicting the CRY2P532L mutation. PHR, photolyase homologous region; CCE, CRY C-terminal extension; DQQVPSAV, VP motif in CRY2; numbers, amino acid positions. b, Hypocotyl length of 6-day-old seedlings of indicated genotypes grown under blue light of different light intensities (0, 10, 30, 60, 100 \\(\\mu \\mathrm{mol} \\mathrm{m}^{-2} \\mathrm{s}^{-1}\\) ). FGFP-CRY2 and FGFP-CRY2P532L were constitutively expressed in cry1cry2rdr6. Standard deviations (n≥20) are shown. c, Immunoblots showing degradation of FGFP-CRY2 or FGFP-CRY2P532L in 7-day-old etiolated transgenic seedlings", + "footnote": [], + "bbox": [ + [ + 150, + 110, + 854, + 635 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 68, + 92, + 780, + 530 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 55, + 45, + 765, + 547 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 63, + 630, + 763, + 901 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 65, + 68, + 750, + 395 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 50, + 52, + 688, + 592 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 44, + 45, + 683, + 520 + ] + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a.mmd b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a.mmd new file mode 100644 index 0000000000000000000000000000000000000000..037af723e2a96042fcf5aaf36fef3521abc8bd07 --- /dev/null +++ b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a.mmd @@ -0,0 +1,360 @@ + +# Regulation of Arabidopsis photoreceptor CRY2 by two distinct E3 ubiquitin ligases + +Yadi ChenFujian Agriculture and Forestry University + +Xiaohua HuFujian Agriculture and Forestry University + +Siyuan LiuFujian Agriculture and Forestry University + +Tiantian SuUniversity of California, Los Angeles + +Hsiaochi HuangUniversity of California, Los Angeles + +Huibo RenFujian Agriculture and Forestry University + +Zhensheng GaoFujian Agriculture and Forestry University + +Xu WangUniversity of California, LA + +Deshu LinFujian Agriculture and Forestry University + +Qin Wang ( qinwangCRY@163. com) Fujian Agriculture and Forestry University https://orcid.org/0000- 0002- 9202- 8005 + +Chentao LinUniversity of California, Los Angeles https://orcid.org/0000- 0002- 0004- 8480 + +## Article + +Keywords: cryptochromes, Arabidopsis, E3 ubiquitin ligases + +Posted Date: December 2nd, 2020 + +DOI: https://doi.org/10.21203/rs.3.rs- 110165/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on April 12th, 2021. See the published version at https://doi.org/10.1038/s41467-021-22410-x. + +<--- Page Split ---> + +# Regulation of Arabidopsis photoreceptor CRY2 by two distinct E3 ubiquitin ligases + +Yadi Chen \(^{1,3}\) , Xiaohua Hu \(^{1,3}\) , Siyuan Liu \(^{1}\) , Tiantian Su \(^{2}\) , Hsiaochi Huang \(^{2}\) , Huibo Ren \(^{1}\) , Zhensheng Gao \(^{1}\) , Xu Wang \(^{1,2}\) , Deshu Lin \(^{1}\) , Qin Wang \(^{1,*}\) , and Chentao Lin \(^{2}\) + +1. College of Life Sciences, Basic Forestry and Proteomics Research Center, Fujian Agriculture and Forestry University, Fuzhou 350002, China; 2. Department of Molecular, Cell & Developmental Biology, University of California, Los Angeles, CA 90095, USA; 3. These authors contributed equally to this work. \* For correspondence: qinwangCRY@163. com. + +## Abstract + +Cryptochromes (CRYs) are photoreceptors or components of the molecular clock in various evolutionary lineages, and they are commonly regulated by polyubiquitination and proteolysis. Multiple E3 ubiquitin ligases regulate CRYs in animal models, and previous genetics study also suggest existence of multiple E3 ubiquitin ligases for plant CRYs. However, only one E3 ligase, Cul4COP1-SPAs, has been reported for plant CRYs so far. Here we show that Cul3LRBs is the second E3 ligase of CRY2 in Arabidopsis. We demonstrated the blue light-specific and CRY-dependent activity of LRBs (Light-Response Bric-a-Brack/Tramtrack/Broad 1, 2 & 3) in blue-light regulation of hypocotyl elongation. LRBs physically interact with photoexcited and phosphorylated CRY2 to facilitate polyubiquitination and degradation of CRY2 in response to blue light. We propose that Cul4COP1-SPAs and Cul3LRBs E3 ligases interact with CRY2 via different structure elements to regulate the abundance of CRY2 photoreceptor under different light conditions, facilitating optimal photoresponses of plants grown in nature. + +CRYs are photolyase- like flavoproteins that act as photoreceptors in plants or core components of the molecular clock in mammals \(^{1 - 3}\) . Regulation of CRY abundance is an important mechanism to control cellular photo- responses and chrono- responses. For example, two cullin 1 family E3 ubiquitin ligases, SCFFBXL3 and SCFFBXL21, regulate + +<--- Page Split ---> + +ubiquitination and degradation of mCRYs to govern the circadian clock in mammals4-8. Similarly, a cullin 1- based E3 ubiquitin ligase receptor, Jetleg, and a cullin 4- based E3 ubiquitin ligase receptor, Brwd3, bind to dCRY to regulate ubiquitination of Tim and dCRY in Drosophila8,10. + +Arabidopsis cryptochrome 2 (CRY2) is one of the best studied plant CRYs that mediate blue light inhibition of cell elongation and photoperiodic promotion of floral initiation11,12. These physiological activities of CRY2 are regulated by at least three blue light- dependent mechanisms. Firstly, CRY2 undergoes blue light- dependent oligomerization to become active tetramers13-17. The BIC1 and BIC2 (Blue- light Inhibitors of Cryptochromes 1 and 2) proteins interact with photoexcited CRY2 to negatively regulate CRY2 photo- oligomerization in light, whereas the active CRY2 homooligomers may also undergo thermal relaxation to become inactive monomers in the absence of light14-17. Secondly, the activity of CRY2 homooligomers are positively regulated by protein phosphorylation reactions catalyzed by four related protein kinases PPKs (Photoregulatory Protein Kinases 1- 4)18-20. Thirdly, the photoexcited CRY2 proteins undergo polyubiquitination catalyzed by the E3 ubiquitin ligase Cul4COP1- SPAs and subsequently degraded by the 26S proteasome11,21,22. Like animal CRYs, the abundance and overall cellular activity of plant CRYs are regulated by phosphorylation, ubiquitination, and proteolysis3. However, only a cullin 4 family E3 ubiquitin ligase, Cul4COP1- SPAs, is presently known to regulate ubiquitination and degradation of plant CRYs3, raising the question how the highly conserved CRYs are differentially regulated in different evolutionary lineages. + +Under nature light conditions, plants rely on the coaction of blue light receptors CRYs and the red/far- red light receptors phytochrome (phys)23,24 to achieve the optimal photoresponses25. Mechanistically, the CRY- phy coaction could be achieved by different photoreceptors physically complexing with the same signaling proteins, such as bHLH transcription factors Phytochrome Interacting Factors (PIF1- 8)26, Photoregulatory Protein Kinases (PPK1- 4)20,27, the substrate receptor of a Cul4 E3 ligase Constitutive Photomorphogenic 1 (COP1)28, and Suppressors of PHYA- 105 1 family members (SPA1- 4) that act as COP1 activators29. For example, CRY2 interacts with PPKs, which catalyzes blue light- dependent phosphorylation of CRY2 to positively regulate the + +<--- Page Split ---> + +functions, polyubiquitination, and degradation of CRY220. On the other hand, PPKs interact with the phyB- PIF3 complex to negatively regulate the function of phyB by the so- called “mutually assured destruction” mechanism27,30. According to this hypothesis, the photoexcited phyB interacts with PIF3, recruiting PPKs to phosphorylate PIF3; phosphorylated PIF3 interact with the substrate receptors of the Cul3LRBs, Light- Response Bric- a- Brack/Tramtrack/Broad (LRB 1- 3), which catalyze polyubiquitination and degradation of PIF3 that also leads to phyB degradation27. Arabidopsis genome encodes three LRBs, LRB1, LRB2, and LRB330,31. LRB1 and LRB2 share higher homology and higher levels of mRNA expression than LRB3, and they act redundantly to suppress phyB- dependent photoresponses31. It has not been reported whether Cul3LRBs play other roles in addition to its important function regulating PIF3 ubiquitination and phyB- dependent red- light responses. + +## Results + +## LRBs are required for the CRY-dependent blue light responses + +We have previously shown that the blue light- dependent ubiquitination and proteolysis of CRY2 is diminished but not completely abolished in the cop1 null mutant22, suggesting the existence of another E3 ubiquitin ligase in addition to Cul4COP1-SPAS. Because LRB2 was identified as a putative CRY2- associated protein in our prior IP- MS (Immunoprecipitation- Mass Spectrometry) analyses of CRY2 complexes purified from transgenic Arabidopsis plants overexpressing GFP- CRY220 (Supplementary Table 1), we first tested whether LRBs play a role in blue light- dependent seedling development. As reported previously30, the Irb123 triple mutant showed little apparent abnormality when grown in darkness or far- red light, but it exhibited a dramatic short- hypocotyl phenotype when grown in continuous red light (Fig. 1a- b). The Irb123 triple mutant also showed short- hypocotyl phenotype when grown in both long- day and short- day photoperiods illuminated with white light (Fig. 1a- b). These results are consistent with the established function of LRBs regulating red light- dependent activities of phyB30,31. When grown in continuous blue light, the cry1cry2 mutant developed long hypocotyl phenotype as previously reported32, whereas the Irb123 showed hypocotyls modestly shorter than + +<--- Page Split ---> + +that of the wild- type seedings (Fig. 1c- d). The short hypocotyl phenotype of Irb123 mutant in blue light can be rescued by constitutively overexpressing LRB2 protein in Irb123 mutant (Supplementary Fig. 1). The short- hypocotyl phenotype of the Irb123 mutant seedlings (6- day- old) is more apparent in seedlings grown in blue light of the light intensities higher than \(10 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) (Fig. 1d), which is consistent with the previous report that showed hardly distinguished phenotype of the wild- type and the Irb12 double mutant (4- day- old) grown in continuous blue light with the light intensities of \(10 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) or lower \(^{31}\) . These results argue for a light intensity- dependent function of LRBs in blue light. To examine the functional relationship between CRYs and LRBs, we prepared the cry1cry2lrb123 quintuple mutant and compared the hypocotyl phenotype of the quintuple mutant with its parents (Fig. 1e- f). When grown under continuous blue light, the cry1cry2 mutant and the Irb123 mutant exhibit long or short hypocotyl phenotype, respectively. However, the cry1cry2lrb123 quintuple mutant shows the long hypocotyl phenotype indistinguishable to that of the cry1cry2 mutant parent in blue light (Fig. 1e- f), but not in other light conditions (Fig. 1a- b). This result clearly demonstrates that the function of LRBs in blue light is dependent on the activity of the blue- light receptor CRYs. Comparing to the Irb123 parent, the Irb123cop1 quadruple mutant shows de- etiolated phenotype of the cop1 parent in darkness but a slight additive effect of both parents in blue light (Fig. 1g- h). Taken together, these results argue that LRBs regulate plant development by controlling not only the red light- dependent activity of phytochromes but also the blue light- dependent activity of cryptochromes. + +## LRBs are required for the blue light-dependent CRY2 degradation + +We next tested how LRBs regulate cryptochrome activity by examining the blue light- dependent degradation of CRY2 in the Irb123 mutants. Results shown in Fig. 2 demonstrate that LRBs are required for the blue light- induced degradation of CRY2. The CRY2 protein level decreased by at least \(70\%\) (in 30 minutes) to \(85\%\) (in 60 minutes) in etiolated wild- type seedlings transferred to blue light ( \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) ) (Fig. 2a- b). In contrast, levels of the CRY2 protein exhibited little change in one Irb123 triple mutant allele (Irb1Irb2- 1Irb3) or less than \(50\%\) decrease in the second Irb123 triple mutant allele (Irb1Irb2- 2Irb3) under the same condition (Fig. 2a- b). This result indicates that + +<--- Page Split ---> + +Irb1/Irb2- 1/Irb3 is a slightly stronger allele in comparison to Irb1/Irb2- 2/Irb3. We then compared the blue light- induced CRY2 degradation in Irb123, cop1, and Irb123cop1 quadruple mutant in seedlings grown in continuous dark or blue light (Fig. 2c- d). Results of this experiment suggest that COP1, but not LRBs, determines abundance of the CRY2 protein in this steady- state condition, suggesting that LRBs are responsible for the rapid blue light response of CRY2 proteolysis but not the prolonged photoresponse of CRY2 proteolysis, whereas COP1 is required for the prolonged photoresponse of CRY2 proteolysis. To test this hypothesis, we compared the photoresponsive CRY2 proteolysis in etiolated seedlings transferred to blue light from 30 minutes to 14 hours (Fig. 2e- f). Results of this experiment show that the Irb123 mutants are clearly defective for the blue light- induced CRY2 degradation within two hours of light exposure, whereas the cop1 mutant shows defect in CRY2 degradation primarily during prolonged light exposure. As expected, little CRY2 degradation was detected in the Irb123cop1 quadruple mutant for both short or long exposure of blue light. We concluded that LRBs and COP1 are both required for the blue light- induced CRY2 degradation. + +## LRBs interact with phosphorylated CRY2 to catalyze its polyubiquitination + +Given that a substrate receptor of E3 ubiquitin ligases must physically interact with the substrate, we examined the CRY2- LRB interaction (Fig. 3), using the co- immunoprecipitation (co- IP) assays of proteins co- expressed in the heterologous HEK293T (Human Embryo Kidney) cells as we previously reported14. Because we have previously shown that the blue light- induced phosphorylation of CRY2 is required for CRY2 ubiquitination and degradation, and that PPK1 is one of the four related protein kinases that specifically phosphorylate photoexcited CRY218,20, we tested LRB- CRY2 interaction in the presence of the wild- type PPK1 in response to blue light in HEK293T cells. LRB1 and LRB2 preferentially interact with phosphorylated CRY2 in HEK293T cells co- expressing PPK1, but only when cells were exposed to blue light (Fig. 3a- b). The light dependence of LRB- CRY2 interaction is confirmed by the result that LRB2 failed to interact with the photo- insensitive CRY2D387A mutant that does not bind the FAD (Flavin Adenine Dinucleotide) chromophore33 (Fig. 3c). The phosphorylation dependence of LRB- CRY2 interaction was confirmed by the result that LRB2 failed to + +<--- Page Split ---> + +interact with CRY2 in light- treated HEK293T cells co- expressing the catalytically inactive PPK1D267N (Fig. 3d). These results demonstrate that LRBs directly and specifically interact with photoexcited and phosphorylated CRY2. We next examined LRB1- CRY2 and LRB2- CRY2 interactions in vivo, using the BiFC (Bimolecular Fluorescence Complementation) assay in \*Arabidopsis\* cells34. In this experiment, \*Arabidopsis\* leaves were transiently transformed by infiltration of \*Agrobacterium\* cells harboring plasmids expressing the indicated BiFC pairs of recombinant proteins34, and the direct protein- protein interactions between the BiFC pairs were analyzed by the BiFC assay. Figure 4 shows strong BiFC signals between CRY2 and LRB1 or CRY2 and LRB2. In contrast, no BiFC signal was detected between the photo- insensitive CRY2D387A and LRB1 or CRY2D387A and LRB2. These results clearly demonstrate the direct interaction of CRY2 and LRB proteins \*in vivo\*. + +To further test the hypothesis that the Cul3LRBs and Cul4COP1-SPAs E3 ubiquitin ligases catalyze blue light- dependent CRY2 ubiquitination, we performed three independent experiments to investigate the activity of LRBs and COP1 in the blue light- induced polyubiquitination of CRY2, using the IP/immunoblot assays. In the first two experiments, we prepared transgenic plants overexpressing Flag- GFP- tagged CRY2 (FGFP- CRY2) in the wild- type (WT), the strong (\*lrb1lrb2- 1lrb3\*) or weak (\*lrb1lrb2- 2lrb3\*) alleles of the \*lrb123\* triple mutant (Fig. 5a- b, d), and the \*cop1\* mutant backgrounds (Fig. 5c- d). In the third experiment, we prepared double- tagged transgenic lines that overexpress FGFP- CRY2 with either the Myc- tagged LRB (Myc- LRB1, Myc- LRB2) or the Myc- tagged COP1 (Myc- COP1) (Fig. 5e). In the first experiment, polyubiquitinated proteins of samples were indiscriminately precipitated using the TUBE2 (Tandem Ubiquitin Binding Entity 2)- conjugated beads35, and the TUBE2- enriched proteins were analyzed by immunoblots probed with the anti- ubiquitin or anti- CRY2 antibodies (Fig. 5a- c). This experiment detected a high level of polyubiquitination of CRY2 in blue light- treated wild- type background seedlings (Fig. 5a- c). However, little ubiquitinated CRY2 was detected in etiolated seedlings or in blue light- treated \*lrb123\* or \*cop1\* mutant background seedlings (Fig. 5a- c). In the other two experiments, FGFP- CRY2 were immunoprecipitated by the GFP- trap beads, and the level of polyubiquitination of FGFP- CRY2 was analyzed by immunoblots probed against anti- ubiquitin antibody (Fig. + +<--- Page Split ---> + +5d- e). In etiolated seedlings exposed to blue light, a high level of ubiquitinated FGFP- CRY2 was detected in the wild- type seedlings overexpressing FGFP- CRY2. In contrast, a lower level of ubiquitinated FGFP- CRY2 was detected in lrb1lrb2- 2lrb3 weak allele, whereas almost no ubiquitinated FGFP- CRY2 was detected in the lrb1lrb2- 1lrb3 strong allele or the cop1 mutant seedlings transgenically overexpressing FGFP- CRY2 (Fig. 5d). In comparison to FGFP- CRY2 overexpressed in the wild- type background, the blue light- induced polyubiquitination of GFP- CRY2 is enhanced in plants co- overexpressing LRB1, or LRB2, or COP1 (Fig. 5e). Taken together, results of those three independent experiments all support the hypothesis that two distinct E3 ubiquitin ligases, Cul3LRB8 and Cul4COP1-SPAS, can both catalyze blue light- dependent polyubiquitination of CRY2 in vivo. + +## COP1 and LRBs interact with different structure elements of CRY2 + +Because the LRBs and COP1 proteins share little sequence or structural similarity, we reasoned that they are unlikely to interact with the identical structure elements of CRY2. The VP motif is a conserved COP1- binding motif of many substrates of the Cul4COP1-SPAS E3 ligase, including CRY236,37. We previously reported that the CRY2P532L mutation impaired at the VP motif (Fig. 6a) loses all the physiological activities tested but retains many photochemical properties of CRY2, including blue light- induced oligomerization, phosphorylation, ubiquitination, and degradation15. We first compared the relative activity of FGFP- CRY2 and FGFP- CRY2P532L, using the hypocotyl inhibition assay in transgenic plants expressing the respective recombinant protein at similar levels in the cry1cry2 mutant background (Supplementary Fig. 2a). Transgenic seedlings expressing FGFP- CRY2P532L showed no obvious activity mediating light inhibition of hypocotyl elongation under all light intensities of blue light tested (Fig. 6b, Supplementary Fig. 2b). This result confirmed our previous report that CRY2P532L is a loss- of- function mutation15. We then compared the kinetics of blue light- induced degradation of FGFP- CRY2 and FGFP- CRY2P532L, using the same transgenic lines (Fig. 6c- d). Although FGFP- CRY2P532L is degraded in response to blue light as we previously reported15, this recombinant CRY2 mutant protein appears to degrade slower than its wild- type counterpart (Fig. 6c- d). The apparent half- life of FGFP- CRY2P532L is more than + +<--- Page Split ---> + +twice that of FGFP- CRY2 (Fig. 6d). Because the LRBs only interact with phosphorylated CRY2 whereas COP1 interacts with both phosphorylated and unphosphorylated CRY2 (Fig. 6e), we examined the phosphorylation activity of the CRY2P532L mutant protein in HEK293T cells co- expressing the CRY2 kinase PPK1. CRY2P532L mutant undergoes blue light- and PPK1- dependent phosphorylation (Fig. 6e- f). Neither the unphosphorylated CRY2D387A control or the phosphorylated CRY2P532L mutant protein interacts with COP1 or SPA1 in response to blue light (Fig. 6e). Because the CRY2P532L mutant does not interact with COP1, its blue light- induced degradation is most likely dependent on another E3 ligase, such as LRBs. Consistent with this hypothesis, the photoexcited and phosphorylated CRY2P532L mutant protein interacts with LRB2 with no discernable difference in comparison to the wild- type CRY2 (Fig. 6f). This result explains why CRY2P532L that does not interact with COP1 or SPA1 can still be degraded in plants (Fig. 6c). Taken together, the results of our experiments argue that two types of substrate receptors, COP1 and LRBs, interact with CRY2 via distinct structural elements of CRY2 to enable the blue light- induced CRY2 ubiquitination and degradation by the respective E3 ubiquitin ligases, Cul3LRBs and Cul4COP1-SPAs. + +## Discussion + +In this study, we demonstrate that, in addition to Cul4COP1-SPAs, the E3 ubiquitin ligases Cul3LRBs is required for blue light regulation and function of CRY2. We demonstrate that Cul3LRBs is responsible for the rapid ubiquitination and degradation of CRY2, whereas Cul4COP1-SPAs is responsible for the slow or prolonged ubiquitination and degradation of CRY2. We show that LRBs and COP1 interact with CRY2 via different structural elements: LRBs interacts with photoexcited and phosphorylated CRY2, but COP1 interacts with CRY2 regardless of its phosphorylation states. It is interesting that plants evolved with two distinct E3 ubiquitin ligases to regulate a CRY photoreceptor by different modes of interaction under different light conditions. In mammals, two related Cul1- based E3 ligases, SCFFBXL3 and SCFFBXL21, act to promote ubiquitination and degradation of CRYs in the nucleus and cytoplasm, respectively, to control the period of the circadian clock8. In Drosophila, Jetleg of a Cul1- based E3 ligase and Brwd3 of a + +<--- Page Split ---> + +Cul4- based E3 ligase regulate ubiquitination and degradation of dCRY- dependent circadian rhythm \(^{9,10}\) . Cul4 \(^{Brwd3}\) mediates light- dependent ubiquitination and degradation of dCRY, whereas Cul1 \(^{Jettag}\) interacts with dCRY to catalyze ubiquitination of the dCRY- interacting protein TIM \(^{10,38}\) . Decreased TIM may enhance interaction of dCRY with its E3 ligases to accelerate its degradation in light. These studies demonstrate that multiple E3 ligases are needed to regulate the activity of a CRY to control the circadian clock. The results of this study argue that plant CRY2 is also controlled by two distinct E3 ligases, and that the delicate control of the abundance of CRYs is evolutionarily conserved for not only circadian rhythms in animals but also photomorphogenesis in plants. Our finding that Cul3 \(^{LRBs}\) or Cul4 \(^{COP1 - SPAs}\) catalyze CRY2 ubiquitination in different light conditions to promote rapid or slow degradation of CRY2, respectively, highlights the importance of photosensitivity to photomorphogenesis of plants. The complex mechanism regulating CRY2 turnover is likely evolved to optimize photomorphogenesis under natural light conditions. In this regard, it is particularly interesting that Cul3 \(^{LRBs}\) has been previously shown to be the E3 ligase of PIF3 and critical for the function of phyB and phytochrome- dependent photoresponses \(^{30,31}\) . Our finding that Cul3 \(^{LRBs}\) also catalyzes blue light- dependent ubiquitination of CRY2 to regulate blue light- dependent photomorphogenesis argues for a previously unrecognized mechanism toward a better understanding of how plants respond to light in the nature white light conditions. + +## Methods + +## Plasmid construction + +Plasmid constructionPlasmid constructs in this study were generated by In- Fusion Cloning methods. The insertions were all verified by Sanger sequencing. The mammalian cell expression vectors used in this study were pQCMV- Flag- GFP, pQCMV- GFP and pCMV- Myc. pQCMV- Flag- GFP vector, driven by CMV promoter and with 1x Flag and GFP tags, was modified from pEGFP- N1 (Clontech). pQCMV- GFP vector, driven by CMV promoter and with a GFP tag, was modified from pQCMV- Flag- GFP. pCMV- Myc vector, driven by CMV promoter and with a 1x Myc tag, was described previously \(^{20}\) . To generate plasmids expressing Flag- LRB1, Flag- LRB2, Flag- CRY2 and Flag- CRY2 \(^{D387A}\) , the CDS regions of + +<--- Page Split ---> + +respective genes were amplified either from Arabidopsis cDNA or previous plasmids, and in- fusion into Spel/Kpnl- digested pQCMV- Flag- GFP vector, resulting a Flag tag at the N- terminus of genes. Myc- LRB1, Myc- LRB2, Myc- CRY2, Myc- SPA1 and Myc- COP1 plasmids were prepared by cloning CDS into the BamHI site of pCMV- Myc vector, resulting a Myc tag at the N- terminus of genes. The CDS of PPK1 and PPK1D267N were amplified from previous plasmids by using the primers containing 2x HA coding sequences at the 5' end of forward primers, and in- fusion into Spel/Kpnl- digested pQCMV- GFP vector to generate HA- PPK1 and HA- PPK1D267N. + +For BiFC constructs, the vectors used in this study were described previously39. The coding sequences of LRB1 (AT2G46260), LRB2 (AT3G61600), CRY2 (AT1G04400) and CRY2D387A were amplified and cloned into pDONR/Zeo entry vector. The entry constructs were then introduced into the destination vector pX- nYFP (N- terminus of YFP fused to C- terminus of genes) or pcCFP- X (C- terminus of CFP fused to N- terminus of genes) by LR reaction. + +pFGFP and pDT1H binary vectors were used for creating overexpression transgenic lines. The pFGFP binary vector, driven by Actin2 promoter and with 2x Flag and GFP tags, was modified from pCambia330120. The pDT1H binary vector, possessing two expression cassettes which can sequentially insert two genes into one vector, was modified from previously published vector pDT140 by replacing the BAR gene with HPT gene which converts basta resistance to hygromycin resistance in plants. The CDS of LRB2, CRY2 and CRY2P532L were cloned into BamHI- digested pFGFP vector to generate FGFP- LRB2, FGFP- CRY2, FGFP- CRY2P532L with 2x Flag and GFP tags fused to N- terminus of the genes. The CDS regions of LRB1, LRB2 and COP1 (AT2G32950) were cloned into the BamHI site of pDT1H vector to produce Myc- LRB1, Myc- LRB2 and Myc- COP1 with 4x Myc tags fused to N- terminus of the genes. + +## Plant materials and growth conditions + +All mutants used in this study are in the Arabidopsis thaliana Columbia ecotype background. The cry1cry2 double mutant were described as previously32. The lrb1lrb2- 1lrb3 (lrb1, Salk_145146; lrb2- 1, Salk_001013; lrb3, Salk_082868) and lrb1lrb2- 2lrb3 (lrb2- 2, Salk- 044446) triple mutants were gifts from Dr. Peter Quail and as described previously30,31. cop1- 4 and cop1- 6 are weak mutant alleles of COP1 as + +<--- Page Split ---> + +previously described41. cry1cry2lrb1lrb2- 2lrb3 and lrb1lrb2- 2lrb3cop1- 4 mutants were generated by crossing. The genotypes of lrb123 mutants were verified by PCR using the primers listed in Supplementary Table 2. cop1- 4 mutant was verified by PCR using the primers listed in Supplementary Table 2, followed by Sanger sequencing to confirm the point mutation of COP1. cry1cry2 mutant was verified by western blots using antibodies against CRY1 and CRY2 proteins. + +All transgenic lines were generated via Agrobacterium tumefaciens- mediated floral- dip method42. The wild- type plants used for transformation in this study are rdr6- 11, which suppresses gene silencing43. For in vivo ubiquitination study, FGFP- CRY2 was introduced into rdr6- 11, lrb1lrb2- 1lrb3, lrb1lrb2- 2lrb3, and cop1- 6 background. The FGFP- CRY2/Myc- LRB1 double- overexpression lines were prepared by introducing FGFP- CRY2 into Myc- LRB1/rdr6- 11 plants. The transgenic T1 populations were screened on MS plates containing 25 mg/L Glufosinate- ammonium (cat # CP6420, Bomei Biotechnology) and 25 mg/L hygromycin (cat # 10843555001, Roche), and western blots were performed to confirm the expression of both proteins. The same method was used for generating FGFP- CRY2/Myc- LRB2 and FGFP- CRY2/Myc- COP1 double- overexpression lines. For experiments comparing the hypocotyl phenotype and protein degradation kinetics of FGFP- CRY2 and FGFP- CRY2P532L, FGFP- CRY2 and FGFP- CRY2P532L were introduced into cry1cry2rdr6 background. The cry1cry2rdr6 were generated by crossing cry1- 30432, cry2- 112, and rdr6- 1143. The transgenic lines were screened on MS plates with 25 mg/L Glufosinate- ammonium, and lines with similar protein expression level of FGFP- CRY2 and FGFP- CRY2P532L were used for analysis. For lrb123 mutant blue- light hypersensitivity phenotype rescue experiments, FGFP- LRB2 and Myc- LRB2 were transformed into lrb1lrb2- 2lrb3 background. + +For routine maintenance, Arabidopsis thaliana were grown under long day conditions (16 hours light / 8 hours dark) at 22°C. For hypocotyl phenotype analysis, seedlings were grown on MS plates with 3% sucrose at 20- 22°C for 6 days under different light conditions. Light- emitting diode (LED) was used to obtain monochromatic light (blue light, peak 465 nm, half- bandwidth of 25 nm; red light, peak 660 nm, half- bandwidth of 20 nm; far- red light, peak 735 nm, half- bandwidth of 21 nm). For endogenous CRY2 degradation analysis in WT, lrb123, cop1 and lrb123cop1 mutants, seedlings were grown in darkness + +<--- Page Split ---> + +on MS plates with \(3\%\) sucrose for 7 days, then subjected to \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for the indicated time. For FGFP- CRY2 and FGFP- CRY2 \(^{P532L}\) degradation analysis, seedlings were grown in darkness on MS plates with \(3\%\) sucrose for 7 days, then subjected to \(100 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for the indicated time. For immunoprecipitation of polyubiquitinated proteins, 7- day- old etiolated seedlings grown on MS medium containing \(3\%\) sucrose were treated with \(50 \mu \mathrm{M} \mathrm{MG132}\) (Cat # S2619, Selleck) in liquid MS in the dark overnight with gentle shaking, and then moved to \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes before harvest. + +## Blue-light-induced CRY2 degradation assays + +For endogenous CRY2 degradation analysis, seedlings were grown in the dark for seven days and then treated with \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) of blue light for indicated time. For FGFP- CRY2 and FGFP- CRY2 \(^{P532L}\) degradation analysis, seedlings were treated with \(100 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) of blue light. Tissues were homogenized by TissueLyser (QIAGEN). Proteins were extracted in equal tissue volume of protein extraction buffer [120mM Tris pH 6.8, 100mM EDTA, \(4\%\) w/v SDS, \(10\%\) v/v beta- mercaptoethanol, \(5\%\) v/v Glycerol, \(0.05\%\) w/v Bromophenol blue] \(^{44}\) , heated at \(100^{\circ} \mathrm{C}\) for 8 minutes, centrifuged at 16000 rcf for 10 minutes, and analyzed by western blot. Proteins were separated in \(10\%\) SDS- PAGE gels, and transferred to nitrocellulose membranes (Pall Life Sciences). + +For immunoblot signals detected by enhance conventional chemiluminescent (ECL) method, membranes were blocked with \(5\%\) non- fat milk in PBST for 1 hour, blotted with anti- CRY2 primary antibody (1:3000, homemade) in PBST for 1.5 hours, washed 8 minutes x 3 times with PBST, blotted with anti- Rabbit- HRP (1:10000, cat # 31460, ThermoFisher) secondary antibody in PBST for 1.5 hours, washed 8 minutes x 3 times with PBST and then incubated with ECL solution for X- ray film development. The membranes were then stripped with stripping buffer [0.2 M Glycine, pH 2.5] and re- probed with anti- HSP82 (1:10000, cat # AbM51099- 31- PU, Beijing Protein Innovation) primary antibody and anti- Mouse- HRP (1:10000, cat # 31430, ThermoFisher) secondary antibody. Anti- HSP82 antibody from Oryza sativa can also recognized Arabidopsis HSP90 protein \(^{45}\) . Three independent blots were performed in parallel for quantification analysis. The quantification of protein intensity of ECL immunoblots was performed by ImageJ. + +<--- Page Split ---> + +For immunoblot signals detected by Odyssey CLx Infrared Imaging System (Li- COR), membranes were blocked with \(0.5\%\) casein in PBS for 1 hour, blotted with anti- CRY2 (1:3000) and anti- HSP82 (1:10000) mixed primary antibodies in PBST with \(0.5\%\) casein for 1.5 hours, washed 8 minutes x 3 times with PBST, blotted with Donkey anti-rabbit 790 (1:15000, cat # A11374, ThermoFisher) and Donkey anti- mouse 680 (1:15000, cat # A10038, ThermoFisher) mixed secondary antibodies in PBST with \(0.5\%\) casein for 1.5 hours, washed 8 minutes x 3 times with PBST and then signals were captured with Odyssey CLx by Image Studio Lite software. Three independent blots were performed in parallel for quantification analysis. Quantification of signals were processed with Image Studio Lite software. CRY2 degradation curves were indicated by CRY2 (B/D) ratio, calculated as CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark, and analyzed with one phase decay of nonlinear regression. + +## Protein expression and co-immunoprecipitation in HEK293T cells + +Human embryonic kidney (HEK) 293T cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with \(10\%\) (v/v) FBS, \(100 \text{mg/L}\) streptomycin and \(100 \text{IU penicillin}\) , in humidified \(5\%\) (v/v) CO2 air at \(37^{\circ}\text{C}\) . Cells were seeded at a density of approximately \(8 \times 10^{5}\) cells/6- cm plate and the transfection were carried out with a calcium phosphate precipitation protocol. Briefly, different combinations of plasmid DNA (2\~5 μg / construct) were mixed with \(30 \mu \text{l}\) 2.5 M CaCl2 and ddH2O to a total volume of \(300 \mu \text{l}\) , then \(300 \mu \text{l}\) of 2x HeBS [250 mM NaCl, 10 mM KCl, 1.5 mM Na2HPO4, 12 mM Dextrose and 50 mM HEPES, adjust the pH of the final solution to 7.05] was added drop by drop with vortex, and kept at room temperature for 5 minutes before applying to cells. The media were aspirated from each plate, DNA mixtures were gently added into plates and the plates were rotated gently to allow the mixtures to coat the entire plate. 3 ml of fresh media containing 25 μM chloriquine (cat # C6628, Sigma) were added to each plate and the plates were kept in the CO2 incubator overnight. The next day, the media were changed with 3 ml of fresh media without chloriquine. 36- 48 hours after transfection, the cells were subjected to blue light treatment for indicated time, washed twice with cold PBS buffer and then harvested in liquid nitrogen for co- immunoprecipitation. + +Cells transfected with different plasmid DNA were lysed in \(800 \mu \text{l}\) 1% Brij buffer [1% Brij- 35, 50 mM Tris- HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 1x Protease inhibitor cocktail, + +<--- Page Split ---> + +and 1x phosphatase inhibitor PhosSTOP] with rotating at \(4^{\circ}C\) for 20 minutes. Cell lysates were centrifuged at 16,000 rcf for 10 minutes at \(4^{\circ}C\) , and the supernatants were incubated with 20 \(\mu\)l Anti- FLAG M2 affinity gel (Cat. # F2426, Sigma) at \(4^{\circ}C\) for 2 hours with rotation. Beads were washed with \(1\%\) Brij buffer for 5 times. Proteins were competed from the beads with \(35\mu\) 3xFlag peptide solution [500 ng/ul in \(1\%\) Brij buffer] for 30 minutes at room temperature with mixing. Elution was transferred to a new tube, mixed with 4XSDS sample buffer, denatured at \(100^{\circ}C\) for 4 minutes and subjected to western blot analysis. Western blots were performed as described above. Primary antibodies used here were anti- Flag (1:1500, cat # F1804, Sigma), anti- Myc (1:5000, cat # 05- 724, Millipore) and anti- HA (1:5000, cat # 12013819001, Roche), and secondary antibodies used were anti- Mouse- HRP (1:10000, cat # 31430, ThermoFisher) and anti- Rabbit- HRP (1:10000, cat # 31460, ThermoFisher). + +## Bimolecular fluorescence complementation (BiFC) assay + +BiFC assays in Arabidopsis plants were performed as previously described methods with minor modifications34. Briefly, Agrobacteria AGL0 transformed with BiFC plasmids were grown in LB medium with \(40\mu\) M Acetosyringone and \(1\%\) Glucose in \(28^{\circ}C\) shaker overnight. Agrobacteria were centrifuged at 5000 rcf for 10 minutes, washed once with washing buffer [10 mM MgCl2, 100 \(\mu\) M Acetosyringone] and resuspended in infiltration buffer [1/4 MS pH 6.0, \(1\%\) sucrose, 100 \(\mu\) M Acetosyringone and \(0.01\%\) Silwet L- 77] to 0.5 of OD600. The agrobacteria were infiltrated into 3- 4 weeks old Arabidopsis leaves with syringe. Each BiFC assay was performed with at least 3 independent plants with 2- 3 leaves infiltrated for each. Plant leaves were dried before kept in the dark for 24 hours, and then moved back to long day conditions (16 h light/8 h dark) to grow for two more days. The infiltrated leaf samples were analyzed under a Leica TCS SP8X confocal microscope. Hoechst 33342 was used for nuclei staining. The quantification of fluorescence intensity was performed by ImageJ. Briefly, the images of GFP and Hoechst 33342 channels from the same field were stacked first, then the regions of nuclei were selected on the image of Hoechst 33342 channel. The integrated intensity of the selected regions over the entire stack were measured. The BiFC ratio of the selected regions were calculated as [GFP intensity]nuclei / [Hoechst 33342 intensity]nuclei for each + +<--- Page Split ---> + +image. Four to nine images ( \(\sim 40 - 100\) nuclei in total) for each BiFC assay were used for quantification and the BiFC ratios were presented as mean \(\pm\) SD (standard deviation). + +## Immunoprecipitation of ubiquitinated proteins in Arabidopsis + +7- day- old etiolated seedlings were treated with \(50 \mu \mathrm{M}\) MG132 in liquid MS in the dark overnight with gentle shaking, and then treated with \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes before harvest. For each immunoprecipitation, about 3 g of seedlings were used. Seedlings were ground in liquid nitrogen and homogenized in 1.5x tissue volume of IP buffer [50 mM Tris- HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X- 100, 20 mM NaF and 1x Protease inhibitor cocktail] at \(4^{\circ} \mathrm{C}\) for 20 minutes with mixing. Lysates were centrifuged at 16000 rcf for 15 minutes at \(4^{\circ} \mathrm{C}\) . \(100 \mu \mathrm{l}\) of supernatant were saved as inputs. + +For total ubiquitinated protein purification, the supernatant were incubated with \(30 \mu \mathrm{l}\) Agarose- TUBE2 beads (cat # UM402M, LifeSensors) for 2 hours at \(4^{\circ} \mathrm{C}\) with rotating. After binding, beads were pelleted and washed 4 times with ice- cold IP buffer (without inhibitors). Total ubiquitinated proteins were eluted with 2XSDS sample buffer, denatured at \(100^{\circ} \mathrm{C}\) for 4 minutes and subjected to western blot analysis. Anti- ubiquitin antibody ( \(\alpha\) - Ubaq, cat # 14- 6078- 80, Thermofisher) was used to detect total ubiquitinated proteins, and homemade anti- CRY2 antibody was used to detect CRY2 protein. + +For FGFP- CRY2 protein purification, the supernatant were incubated with \(50 \mu \mathrm{l}\) GFP- trap agarose beads (homemade or cat # gta- 20, Chromotek) and rotated at \(4^{\circ} \mathrm{C}\) for 2 hours. Beads were pelleted and washed 4 times with IP buffer (without inhibitors). Proteins were eluted with 2XSDS sample buffer by heating at \(100^{\circ} \mathrm{C}\) for 4 minutes and subjected to western blot analysis. Anti- Flag antibody was used to detect FGFP- CRY2 (IP) fusion protein, and anti- ubiquitin antibody was used to detect ubiquitinated CRY2. + +## References + +1 Cashmore, A. R. Cryptochromes: enabling plants and animals to determine circadian time. Cell 114, 537- 543 (2003). + +<--- Page Split ---> + +Sancar, A. Cryptochrome: the second photoactive pigment in the eye and its role in circadian photoreception. Annu Rev Biochem 69, 31- 67 (2000). + +Wang, Q. & Lin, C. Mechanisms of Cryptochrome- Mediated Photoresponses in Plants. Annu Rev Plant Biol 71, 103- 129, doi:10.1146/annurev- arplant- 050718- 100300 (2020). + +Busino, L. et al. SCFFbx13 controls the oscillation of the circadian clock by directing the degradation of cryptochrome proteins. Science 316, 900- 904 (2007). + +Godinho, S. I. et al. The after- hours mutant reveals a role for Fbx13 in determining mammalian circadian period. Science 316, 897- 900 (2007). + +Mammalian circadian period. Science 316, 897- 900 (2007). + +Siepka, S. M. et al. Circadian mutant Overtime reveals F- box protein FBXL3 regulation of cryptochrome and period gene expression. Cell 129, 1011- 1023 (2007). + +Hirano, A. et al. FBXL21 Regulates Oscillation of the Circadian Clock through Ubiquitination and Stabilization of Cryptochromes. Cell 152, 1106- 1118, doi:10.1016/j.cell.2013.01.054 (2013). + +Yoo, S.- H. et al. Competing E3 Ubiquitin Ligases Govern Circadian Periodicity by Degradation of CRY in Nucleus and Cytoplasm. Cell 152, 1091- 1105, doi:http://dx.doi.org/10.1016/j.cell.2013.01.055 (2013). + +doi:http://dx.doi.org/10.1016/j.cell.2013.01.055 (2013). + +Peschel, N., Chen, K. F., Szabo, G. & Stanewsky, R. Light- dependent interactions between the Drosophila circadian clock factors cryptochrome, jetlag, and timeless. + +Curr Biol 19, 241- 247 (2009). + +Ozturk, N., VanVickle- Chavez, S. J., Akileswaran, L., Van Gelder, R. N. & Sancar, + +A. Ramshackle (Brwd3) promotes light-induced ubiquitylation of Drosophila Cryptochrome by DDB1-CUL4-ROC1 E3 ligase complex. Proceedings of the National Academy of Sciences 110, 4980-4985, doi:10.1073/pnas.1303234110 (2013). + +Lin, C. et al. Enhancement of blue-light sensitivity of Arabidopsis seedlings by a blue light receptor cryptochrome 2. Proc Natl Acad Sci U S A 95, 2686- 2690 (1998). + +Lin, C. et al. Enhancement of blue- light sensitivity of Arabidopsis seedlings by a blue light receptor cryptochrome 2. Proc Natl Acad Sci U S A 95, 2686- 2690 (1998).Guo, H., Yang, H., Mockler, T. C. & Lin, C. Regulation of flowering time by Arabidopsis photoreceptors. Science 279, 1360- 1363 (1998). + +Guo, H., Yang, H., Mockler, T. C. & Lin, C. Regulation of flowering time by + +Arabidopsis photoreceptors. Science 279, 1360- 1363 (1998). + +<--- Page Split ---> + +496 13 Sang, Y. et al. N-terminal domain-mediated homodimerization is required for photoreceptor activity of Arabidopsis CRYPTOCHROME 1. Plant Cell 17, 1569- 1584 (2005). 499 14 Wang, Q. et al. Photoactivation and inactivation of Arabidopsis cryptochrome 2. Science 354, 343- 347, doi:10.1126/science.aaf9030 (2016). 501 15 Liu, Q. et al. Photooligomerization Determines Photosensitivity and Photoreactivity of Plant Cryptochromes. Mol Plant 13, 398- 413, doi:10.1016/j.molp.2020.01.002 (2020). 504 16 Ma, L. et al. Structural insights into BIC- mediated inactivation of Arabidopsis cryptochrome 2. Nat Struct Mol Biol 27, 472- 479, doi:10.1038/s41594- 020- 0410- z (2020). 507 17 Shao, K. et al. The oligomeric structures of plant cryptochromes. Nat Struct Mol Biol 27, 480- 488, doi:10.1038/s41594- 020- 0420- x (2020). 509 18 Shalitin, D. et al. Regulation of Arabidopsis cryptochrome 2 by blue- light- dependent phosphorylation. Nature 417, 763- 767. (2002). 511 19 Wang, Q. et al. The Blue Light- Dependent Phosphorylation of the CCE Domain Determines the Photosensitivity of Arabidopsis CRY2. Molecular Plant 8, 631- 643, doi:http://dx.doi.org/10.1016/j.molp.2015.03.005 (2015). 514 20 Liu, Q. et al. Molecular basis for blue light- dependent phosphorylation of Arabidopsis cryptochrome 2. Nat Commun 8, 15234, doi:10.1038/ncomms15234 (2017). 516 21 Yu, X. et al. Arabidopsis cryptochrome 2 completes its posttranslational life cycle in the nucleus. Plant Cell 19, 3146- 3156 (2007). 519 22 Liu, Q. et al. The Blue Light- Dependent Polyubiquitination and Degradation of Arabidopsis Cryptochrome2 Requires Multiple E3 Ubiquitin Ligases. Plant Cell Physiol 57, 2175- 2186, doi:10.1093/pcp/pcw134 (2016). 522 23 Quail, P. H. Phytochromes. Current biology : CB 20, R504- 507, doi:10.1016/j.cub.2010.04.014 (2010). 524 24 Legris, M., Ince, Y. C. & Fankhauser, C. Molecular mechanisms underlying phytochrome- controlled morphogenesis in plants. Nat Commun 10, 5219, doi:10.1038/s41467- 019- 13045- 0 (2019). + +<--- Page Split ---> + +Wang, Q. et al. New insights into the mechanisms of phytochrome- cryptochrome coaction. New Phytol 217, 547- 551, doi:10.1111/nph.14886 (2018).Pham, V. N., Kathare, P. K. & Huq, E. Phyochromes and Phytochrome Interacting Factors. Plant Physiol 176, 1025- 1038, doi:10.1104/pp.17.01384 (2018).Ni, W. et al. PPKs mediate direct signal transfer from phytochrome photoreceptors to transcription factor PIF3. Nat Commun 8, 15236, doi:10.1038/ncomms15236 (2017).Han, X., Huang, X. & Deng, X. W. The Photomorphogenic Central Repressor COP1: Conservation and Functional Diversification during Evolution. Plant Communications 1, 1- 13, doi:doi.org/10.1016/j.xplc.2020.100044 (2020).Hoecher, U. The activities of the E3 ubiquitin ligase COP1/SPA, a key repressor in light signaling. Curr Opin Plant Biol 37, 63- 69, doi:10.1016/j.pbi.2017.03.015 (2017).Ni, W. et al. A mutually assured destruction mechanism attenuates light signaling in Arabidopsis. Science 344, 1160- 1164, doi:10.1126/science.1250778 (2014).Christians, M. J., Gingerich, D. J., Hua, Z., Lauer, T. D. & Vierstra, R. D. The light- response BTB1 and BTB2 proteins assemble nuclear ubiquitin ligases that modify phytochrome B and D signaling in Arabidopsis. Plant Physiol 160, 118- 134, doi:10.1104/pp.112.199109 (2012).Mockler, T. C., Guo, H., Yang, H., Duong, H. & Lin, C. Antagonistic actions of Arabidopsis cryptochromes and phytochrome B in the regulation of floral induction. Development 126, 2073- 2082 (1999).Liu, H. et al. Photoexcited CRY2 interacts with CIB1 to regulate transcription and floral initiation in Arabidopsis. Science 322, 1535- 1539 (2008).Zhang, Y. et al. A Highly Efficient Agrobacterium- Mediated Method for Transient Gene Expression and Functional Studies in Multiple Plant Species. Plant Communications 1, 1- 12, doi:doi.org/10.1016/j.xplc.2020.100028 (2020). + +<--- Page Split ---> + +Hjerpe, R. et al. Efficient protection and isolation of ubiquitylated proteins using tandem ubiquitin-binding entities. EMBO Rep 10, 1250- 1258, doi:10.1038/embor.2009.192 (2009).Lau, K., Podolec, R., Chappuis, R., Ulm, R. & Hothorn, M. Plant photoreceptors and their signaling components compete for COP1 binding via VP peptide motifs. EMBO J 38, e102140, doi:10.15252/embj.2019102140 (2019).Ponnu, J., Riedel, T., Penner, E., Schrader, A. & Hoecker, U. Cryptochrome 2 competes with COP1 substrates to repress COP1 ubiquitin ligase activity during Arabidopsis photomorphogenesis. Proc Natl Acad Sci U S A, doi:10.1073/pnas.1909181116 (2019).Koh, K., Zheng, X. & Sehgal, A. JETLAG resets the Drosophila circadian clock by promoting light- induced degradation of TIMELESS. Science 312, 1809- 1812 (2006).Gampala, S. S. et al. An essential role for 14- 3- 3 proteins in brassinosteroid signal transduction in Arabidopsis. Dev Cell 13, 177- 189, doi:10.1016/j.devcel.2007.06.009 (2007).He, Z. et al. Construction and Validation of a Dual- Transgene Vector System for Stable Transformation in Plants. J Genet Genomics 43, 199- 207, doi:10.1016/j.jgg.2016.02.005 (2016).McNellis, T. W. et al. Genetic and molecular analysis of an allelic series of cop1 mutants suggests functional roles for the multiple protein domains. Plant Cell 6, 487- 500, doi:10.1105/tpc.6.4.487 (1994).Clough, S. J. Floral dip: agrobacterium- mediated germ line transformation. Methods Mol Biol 286, 91- 102, doi:10.1385/1- 59259- 827- 7:091 (2005).Peragine, A., Yoshikawa, M., Wu, G., Albrecht, H. L. & Poethig, R. S. SGS3 and SGS2/SDE1/RDR6 are required for juvenile development and the production of trans- acting siRNAs in Arabidopsis. Genes Dev 18, 2368- 2379 (2004).Tsugama, D., Liu, S. & Takano, T. A rapid chemical method for lysing Arabidopsis cells for protein analysis. Plant Methods 7, 22, doi:10.1186/1746- 4811- 7- 22 (2011). + +<--- Page Split ---> + +45 Li, X. et al. Identification and validation of rice reference proteins for western blotting. J Exp Bot 62, 4763- 4772, doi:10.1093/jxb/err084 (2011). + +## Acknowledgments + +The authors thank Dr. Peter Quail for providing the Irb123 mutant alleles. Works in the authors' laboratories are supported in part by the National Natural Science Foundation of China (31970265 to Q.W.), Natural Science Foundation of Fujian Province (2019J06014 to Q.W.), FAFU- ICE fund (KXGH17011 to Q.W.) and the National Institutes of Health (GM56265 to C.L.). The UCLA- FAFU Joint Research Center on Plant Proteomics provided the institutional supports. + +## Author contributions + +Y.C. and X.H. conducted most of the experiments. S.L. and T.S. performed some biochemical experiments in HEK293T. H.H. and H.R. prepared some genetic materials. S.L. and Z.G. helped with BiFC experiments. X.W. and D.L. provided critical feedback. C.L. and Q.W. conceived the project, designed the experiments, and wrote the manuscript. + +<--- Page Split ---> + +## Figures + +Fig. 1 LRBs are required for blue light responses. + +Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2, respectively. + +Fig. 3 LRBs interact with phosphorylated Arabidopsis CRY2 expressed in the HEK293T cells. + +Fig. 4 LRBs interact with CRY2 in vivo. + +Fig. 5 LRBs and COP1 are both required for CRY2 ubiquitination. + +Fig. 6 The VP motif of CRY2 is required for the CRY2- COP1 interaction but not the CRY2- LRB interaction. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 LRBs are required for blue light responses. a, 6-day-old seedlings grown in darkness, red light (20 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ), far-red light (6 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) , long day (16h light / 8h dark), or short day (8h light / 16h dark). b, Hypocotyl length of indicated genotypes of (a). c, 6-day-old seedlings grown in darkness or blue light (10 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). d, Hypocotyl length of indicated genotypes grown under blue light of different intensities (0, 5, 10, 20, 40 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ) for 6 days. e-g, 6-day old seedlings grown in darkness or blue light (20 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). f-h, Hypocotyl length of indicated genotypes as in (e) and (g). Standard deviations (n≥20) are shown.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2,
+ +Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2, respectively. a- b, Representative immunoblots (a) and a quantitative analysis (b, n=3) showing the endogenous CRY2 in the 7-day-old etiolated Irb123 mutants irradiated with \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) blue light for the indicated time. c- d, Representative immunoblots (c) and a quantitative analysis (d, n=3) showing the endogenous CRY2 in seedlings of indicated genotypes grown in darkness (D) or continuous blue light (cB, \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). e, Representative immunoblots showing degradation of the endogenous CRY2 in 7-day-old etiolated seedlings irradiated with \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) of blue light. f, Quantitative results of CRY2 degradation from + +<--- Page Split ---> + +immunoblots in (e), analyzed with one phase decay of nonlinear regression. CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark. CRY2 and HSP90 were probed with anti- CRY2 antibody and anti- HSP90 antibody. HSP90 is a loading control. Arrows indicate phosphorylated CRY2. + +![](images/Figure_3.jpg) + +
Figure 3
+ +Fig. 3 LRBs interact with phosphorylated Arabidopsis CRY2 expressed in the HEK293T cells. a- d, Co- immunoprecipitation results showing the blue light- dependent interaction of LRBs and phosphorylated CRY2 in heterologous HEK293T cells. HEK293T cells co- transfected with indicated plasmids were either kept in the dark (-) or treated with 100 \(\mu \mathrm{m} \mathrm{m}^{2} \mathrm{s}^{-1}\) blue light for 2 hours (+). Immunoprecipitations were performed with Flag- conjugated beads. CRY2D387A and PPK1D267N are inactive mutants of CRY2 and PPK1, respectively. Anti- Flag antibody, anti- Myc antibody or anti- HA antibody were used for detecting respective tagged proteins. Arrows show phosphorylated CRY2. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 LRBs interact with CRY2 in vivo. Confocal microscopy images showing
+ +BiFC of indicated protein pairs transiently expressed in Arabidopsis plants. The Hoechst 33342 (nuclei) and GFP (BiFC) signals are shown. The relative activity of CRY2- LRB interaction was presented as BiFC ratio, calculated as \(\text{BiFC ratio} = [\text{GFP intensity}]^{\text{nuclei}} / [\text{Hoechst 33342 intensity}]^{\text{nuclei}}\) , quantified from each image (n≥4). Total number of quantified images and nuclei in the images is indicated underneath. CRY2D387A inactive mutant is used as the negative control. Standard deviations (n≥4 images) are shown. Scale bars, 10 μm. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 LRBs and COP1 are both required for CRY2 ubiquitination.
+ +Immunoblots showing the ubiquitination of FGFP- CRY2 in indicated genotypes. 7- day- old etiolated seedlings constitutively expressing FGFP- CRY2 in indicated genotypes, pretreated with MG132, were kept in the dark (D) or exposed to 30 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes (B) before harvest for IP assays. + +<--- Page Split ---> + +a- c, Total ubiquitinated proteins were purified by TUBE2- conjugated beads. Immunoprecipitated proteins were analyzed by immunoblots probed with anti- ubiquitin antibody (a- Ubq) or anti- CRY2 antibody (a- CRY2). The extent of CRY2 ubiquitination were shown underneath, calculated as [CRY2- Ubq intensity]IP / [Ubq intensity]IP. d- e, FGFP- CRY2 proteins were purified with GFP- trap beads. Immunoprecipitated proteins were analyzed by immunoblots probed with anti- ubiquitin antibody (a- Ubq), anti- Flag antibody (a- Flag) or anti- Myc antibody (a- Myc) for respective epitope- tagged proteins. The extent of CRY2 ubiquitination is calculated by [CRY2- Ubq intensity]IP / [Flag intensity]IP with the short exposure immunoblots and shown underneath. S. exp or L. exp: short or long chemiluminescence exposures of immunoblots. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6 The VP motif of CRY2 is required for the CRY2-COP1 interaction but not the CRY2-LRB interaction. a, A schematic diagrams depicting the CRY2P532L mutation. PHR, photolyase homologous region; CCE, CRY C-terminal extension; DQQVPSAV, VP motif in CRY2; numbers, amino acid positions. b, Hypocotyl length of 6-day-old seedlings of indicated genotypes grown under blue light of different light intensities (0, 10, 30, 60, 100 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). FGFP-CRY2 and FGFP-CRY2P532L were constitutively expressed in cry1cry2rdr6. Standard deviations (n≥20) are shown. c, Immunoblots showing degradation of FGFP-CRY2 or FGFP-CRY2P532L in 7-day-old etiolated transgenic seedlings
+ +<--- Page Split ---> + +irradiated with blue light (100 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) ) for the indicated time. Immunoblots were probed with anti- CRY2 and anti- HSP90 antibodies. HSP90 is a loading control. d, Quantitative results of CRY2 degradation from immunoblots in (c), analyzed with one phase decay of nonlinear regression. CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark. \(\mathrm{T}_{1 / 2}\) indicates the time required for 50% degradation. e- f, Co- IP assays showing the lack of CRY2 \(^{P532L}\) - COP1 interaction (e) and the the blue light- and phosphorylation- dependent CRY2 \(^{P532L}\) - LRB2 interaction (f) in heterologous HEK293T cells. HEK293T cells co- transfected with indicated plasmids were either kept in the dark (-) or treated with 100 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 2 hours (+). Immunoprecipitations were performed with Flag- conjugated beads. CRY2 \(^{D387A}\) is a negative control. Anti- Flag antibody, anti- Myc antibody or anti- HA antibody were used for detecting respective tagged proteins. Arrows indicate phosphorylated CRY2. + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +[Please see the manuscript file to view the figure caption.] + +![](images/Figure_3.jpg) + +
Figure 3
+ +<--- Page Split ---> + +[Please see the manuscript file to view the figure caption.] + +![](images/Figure_4.jpg) + +
Figure 4
+ +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5
+ +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6
+ +[Please see the manuscript file to view the figure caption.] + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- LRBsup.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a_det.mmd b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d9938c2fccf88a561f6446ee5cfe97691640ab8c --- /dev/null +++ b/preprint/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a/preprint__0d62870c95b96539354e1342ce7b4541a37c8df2c8571a011c7dc063baa3034a_det.mmd @@ -0,0 +1,456 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 910, 175]]<|/det|> +# Regulation of Arabidopsis photoreceptor CRY2 by two distinct E3 ubiquitin ligases + +<|ref|>text<|/ref|><|det|>[[44, 195, 415, 236]]<|/det|> +Yadi ChenFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 242, 415, 283]]<|/det|> +Xiaohua HuFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 290, 415, 330]]<|/det|> +Siyuan LiuFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 336, 373, 377]]<|/det|> +Tiantian SuUniversity of California, Los Angeles + +<|ref|>text<|/ref|><|det|>[[44, 383, 373, 423]]<|/det|> +Hsiaochi HuangUniversity of California, Los Angeles + +<|ref|>text<|/ref|><|det|>[[44, 429, 415, 470]]<|/det|> +Huibo RenFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 475, 415, 516]]<|/det|> +Zhensheng GaoFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 521, 290, 562]]<|/det|> +Xu WangUniversity of California, LA + +<|ref|>text<|/ref|><|det|>[[44, 568, 415, 608]]<|/det|> +Deshu LinFujian Agriculture and Forestry University + +<|ref|>text<|/ref|><|det|>[[44, 614, 771, 655]]<|/det|> +Qin Wang ( qinwangCRY@163. com) Fujian Agriculture and Forestry University https://orcid.org/0000- 0002- 9202- 8005 + +<|ref|>text<|/ref|><|det|>[[44, 660, 730, 701]]<|/det|> +Chentao LinUniversity of California, Los Angeles https://orcid.org/0000- 0002- 0004- 8480 + +<|ref|>sub_title<|/ref|><|det|>[[44, 742, 102, 760]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 780, 560, 800]]<|/det|> +Keywords: cryptochromes, Arabidopsis, E3 ubiquitin ligases + +<|ref|>text<|/ref|><|det|>[[44, 818, 340, 837]]<|/det|> +Posted Date: December 2nd, 2020 + +<|ref|>text<|/ref|><|det|>[[44, 856, 463, 875]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 110165/v1 + +<|ref|>text<|/ref|><|det|>[[44, 894, 910, 936]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 908, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on April 12th, 2021. See the published version at https://doi.org/10.1038/s41467-021-22410-x. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[137, 88, 860, 142]]<|/det|> +# Regulation of Arabidopsis photoreceptor CRY2 by two distinct E3 ubiquitin ligases + +<|ref|>text<|/ref|><|det|>[[125, 174, 872, 220]]<|/det|> +Yadi Chen \(^{1,3}\) , Xiaohua Hu \(^{1,3}\) , Siyuan Liu \(^{1}\) , Tiantian Su \(^{2}\) , Hsiaochi Huang \(^{2}\) , Huibo Ren \(^{1}\) , Zhensheng Gao \(^{1}\) , Xu Wang \(^{1,2}\) , Deshu Lin \(^{1}\) , Qin Wang \(^{1,*}\) , and Chentao Lin \(^{2}\) + +<|ref|>text<|/ref|><|det|>[[115, 252, 888, 325]]<|/det|> +1. College of Life Sciences, Basic Forestry and Proteomics Research Center, Fujian Agriculture and Forestry University, Fuzhou 350002, China; 2. Department of Molecular, Cell & Developmental Biology, University of California, Los Angeles, CA 90095, USA; 3. These authors contributed equally to this work. \* For correspondence: qinwangCRY@163. com. + +<|ref|>sub_title<|/ref|><|det|>[[457, 360, 540, 378]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 408, 883, 744]]<|/det|> +Cryptochromes (CRYs) are photoreceptors or components of the molecular clock in various evolutionary lineages, and they are commonly regulated by polyubiquitination and proteolysis. Multiple E3 ubiquitin ligases regulate CRYs in animal models, and previous genetics study also suggest existence of multiple E3 ubiquitin ligases for plant CRYs. However, only one E3 ligase, Cul4COP1-SPAs, has been reported for plant CRYs so far. Here we show that Cul3LRBs is the second E3 ligase of CRY2 in Arabidopsis. We demonstrated the blue light-specific and CRY-dependent activity of LRBs (Light-Response Bric-a-Brack/Tramtrack/Broad 1, 2 & 3) in blue-light regulation of hypocotyl elongation. LRBs physically interact with photoexcited and phosphorylated CRY2 to facilitate polyubiquitination and degradation of CRY2 in response to blue light. We propose that Cul4COP1-SPAs and Cul3LRBs E3 ligases interact with CRY2 via different structure elements to regulate the abundance of CRY2 photoreceptor under different light conditions, facilitating optimal photoresponses of plants grown in nature. + +<|ref|>text<|/ref|><|det|>[[112, 802, 875, 901]]<|/det|> +CRYs are photolyase- like flavoproteins that act as photoreceptors in plants or core components of the molecular clock in mammals \(^{1 - 3}\) . Regulation of CRY abundance is an important mechanism to control cellular photo- responses and chrono- responses. For example, two cullin 1 family E3 ubiquitin ligases, SCFFBXL3 and SCFFBXL21, regulate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 866, 187]]<|/det|> +ubiquitination and degradation of mCRYs to govern the circadian clock in mammals4-8. Similarly, a cullin 1- based E3 ubiquitin ligase receptor, Jetleg, and a cullin 4- based E3 ubiquitin ligase receptor, Brwd3, bind to dCRY to regulate ubiquitination of Tim and dCRY in Drosophila8,10. + +<|ref|>text<|/ref|><|det|>[[111, 193, 883, 660]]<|/det|> +Arabidopsis cryptochrome 2 (CRY2) is one of the best studied plant CRYs that mediate blue light inhibition of cell elongation and photoperiodic promotion of floral initiation11,12. These physiological activities of CRY2 are regulated by at least three blue light- dependent mechanisms. Firstly, CRY2 undergoes blue light- dependent oligomerization to become active tetramers13-17. The BIC1 and BIC2 (Blue- light Inhibitors of Cryptochromes 1 and 2) proteins interact with photoexcited CRY2 to negatively regulate CRY2 photo- oligomerization in light, whereas the active CRY2 homooligomers may also undergo thermal relaxation to become inactive monomers in the absence of light14-17. Secondly, the activity of CRY2 homooligomers are positively regulated by protein phosphorylation reactions catalyzed by four related protein kinases PPKs (Photoregulatory Protein Kinases 1- 4)18-20. Thirdly, the photoexcited CRY2 proteins undergo polyubiquitination catalyzed by the E3 ubiquitin ligase Cul4COP1- SPAs and subsequently degraded by the 26S proteasome11,21,22. Like animal CRYs, the abundance and overall cellular activity of plant CRYs are regulated by phosphorylation, ubiquitination, and proteolysis3. However, only a cullin 4 family E3 ubiquitin ligase, Cul4COP1- SPAs, is presently known to regulate ubiquitination and degradation of plant CRYs3, raising the question how the highly conserved CRYs are differentially regulated in different evolutionary lineages. + +<|ref|>text<|/ref|><|det|>[[112, 664, 884, 893]]<|/det|> +Under nature light conditions, plants rely on the coaction of blue light receptors CRYs and the red/far- red light receptors phytochrome (phys)23,24 to achieve the optimal photoresponses25. Mechanistically, the CRY- phy coaction could be achieved by different photoreceptors physically complexing with the same signaling proteins, such as bHLH transcription factors Phytochrome Interacting Factors (PIF1- 8)26, Photoregulatory Protein Kinases (PPK1- 4)20,27, the substrate receptor of a Cul4 E3 ligase Constitutive Photomorphogenic 1 (COP1)28, and Suppressors of PHYA- 105 1 family members (SPA1- 4) that act as COP1 activators29. For example, CRY2 interacts with PPKs, which catalyzes blue light- dependent phosphorylation of CRY2 to positively regulate the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 881, 397]]<|/det|> +functions, polyubiquitination, and degradation of CRY220. On the other hand, PPKs interact with the phyB- PIF3 complex to negatively regulate the function of phyB by the so- called “mutually assured destruction” mechanism27,30. According to this hypothesis, the photoexcited phyB interacts with PIF3, recruiting PPKs to phosphorylate PIF3; phosphorylated PIF3 interact with the substrate receptors of the Cul3LRBs, Light- Response Bric- a- Brack/Tramtrack/Broad (LRB 1- 3), which catalyze polyubiquitination and degradation of PIF3 that also leads to phyB degradation27. Arabidopsis genome encodes three LRBs, LRB1, LRB2, and LRB330,31. LRB1 and LRB2 share higher homology and higher levels of mRNA expression than LRB3, and they act redundantly to suppress phyB- dependent photoresponses31. It has not been reported whether Cul3LRBs play other roles in addition to its important function regulating PIF3 ubiquitination and phyB- dependent red- light responses. + +<|ref|>sub_title<|/ref|><|det|>[[460, 429, 535, 448]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 483, 703, 503]]<|/det|> +## LRBs are required for the CRY-dependent blue light responses + +<|ref|>text<|/ref|><|det|>[[111, 508, 886, 895]]<|/det|> +We have previously shown that the blue light- dependent ubiquitination and proteolysis of CRY2 is diminished but not completely abolished in the cop1 null mutant22, suggesting the existence of another E3 ubiquitin ligase in addition to Cul4COP1-SPAS. Because LRB2 was identified as a putative CRY2- associated protein in our prior IP- MS (Immunoprecipitation- Mass Spectrometry) analyses of CRY2 complexes purified from transgenic Arabidopsis plants overexpressing GFP- CRY220 (Supplementary Table 1), we first tested whether LRBs play a role in blue light- dependent seedling development. As reported previously30, the Irb123 triple mutant showed little apparent abnormality when grown in darkness or far- red light, but it exhibited a dramatic short- hypocotyl phenotype when grown in continuous red light (Fig. 1a- b). The Irb123 triple mutant also showed short- hypocotyl phenotype when grown in both long- day and short- day photoperiods illuminated with white light (Fig. 1a- b). These results are consistent with the established function of LRBs regulating red light- dependent activities of phyB30,31. When grown in continuous blue light, the cry1cry2 mutant developed long hypocotyl phenotype as previously reported32, whereas the Irb123 showed hypocotyls modestly shorter than + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 888, 640]]<|/det|> +that of the wild- type seedings (Fig. 1c- d). The short hypocotyl phenotype of Irb123 mutant in blue light can be rescued by constitutively overexpressing LRB2 protein in Irb123 mutant (Supplementary Fig. 1). The short- hypocotyl phenotype of the Irb123 mutant seedlings (6- day- old) is more apparent in seedlings grown in blue light of the light intensities higher than \(10 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) (Fig. 1d), which is consistent with the previous report that showed hardly distinguished phenotype of the wild- type and the Irb12 double mutant (4- day- old) grown in continuous blue light with the light intensities of \(10 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) or lower \(^{31}\) . These results argue for a light intensity- dependent function of LRBs in blue light. To examine the functional relationship between CRYs and LRBs, we prepared the cry1cry2lrb123 quintuple mutant and compared the hypocotyl phenotype of the quintuple mutant with its parents (Fig. 1e- f). When grown under continuous blue light, the cry1cry2 mutant and the Irb123 mutant exhibit long or short hypocotyl phenotype, respectively. However, the cry1cry2lrb123 quintuple mutant shows the long hypocotyl phenotype indistinguishable to that of the cry1cry2 mutant parent in blue light (Fig. 1e- f), but not in other light conditions (Fig. 1a- b). This result clearly demonstrates that the function of LRBs in blue light is dependent on the activity of the blue- light receptor CRYs. Comparing to the Irb123 parent, the Irb123cop1 quadruple mutant shows de- etiolated phenotype of the cop1 parent in darkness but a slight additive effect of both parents in blue light (Fig. 1g- h). Taken together, these results argue that LRBs regulate plant development by controlling not only the red light- dependent activity of phytochromes but also the blue light- dependent activity of cryptochromes. + +<|ref|>sub_title<|/ref|><|det|>[[115, 666, 728, 687]]<|/det|> +## LRBs are required for the blue light-dependent CRY2 degradation + +<|ref|>text<|/ref|><|det|>[[112, 692, 886, 899]]<|/det|> +We next tested how LRBs regulate cryptochrome activity by examining the blue light- dependent degradation of CRY2 in the Irb123 mutants. Results shown in Fig. 2 demonstrate that LRBs are required for the blue light- induced degradation of CRY2. The CRY2 protein level decreased by at least \(70\%\) (in 30 minutes) to \(85\%\) (in 60 minutes) in etiolated wild- type seedlings transferred to blue light ( \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) ) (Fig. 2a- b). In contrast, levels of the CRY2 protein exhibited little change in one Irb123 triple mutant allele (Irb1Irb2- 1Irb3) or less than \(50\%\) decrease in the second Irb123 triple mutant allele (Irb1Irb2- 2Irb3) under the same condition (Fig. 2a- b). This result indicates that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 886, 476]]<|/det|> +Irb1/Irb2- 1/Irb3 is a slightly stronger allele in comparison to Irb1/Irb2- 2/Irb3. We then compared the blue light- induced CRY2 degradation in Irb123, cop1, and Irb123cop1 quadruple mutant in seedlings grown in continuous dark or blue light (Fig. 2c- d). Results of this experiment suggest that COP1, but not LRBs, determines abundance of the CRY2 protein in this steady- state condition, suggesting that LRBs are responsible for the rapid blue light response of CRY2 proteolysis but not the prolonged photoresponse of CRY2 proteolysis, whereas COP1 is required for the prolonged photoresponse of CRY2 proteolysis. To test this hypothesis, we compared the photoresponsive CRY2 proteolysis in etiolated seedlings transferred to blue light from 30 minutes to 14 hours (Fig. 2e- f). Results of this experiment show that the Irb123 mutants are clearly defective for the blue light- induced CRY2 degradation within two hours of light exposure, whereas the cop1 mutant shows defect in CRY2 degradation primarily during prolonged light exposure. As expected, little CRY2 degradation was detected in the Irb123cop1 quadruple mutant for both short or long exposure of blue light. We concluded that LRBs and COP1 are both required for the blue light- induced CRY2 degradation. + +<|ref|>sub_title<|/ref|><|det|>[[113, 505, 810, 527]]<|/det|> +## LRBs interact with phosphorylated CRY2 to catalyze its polyubiquitination + +<|ref|>text<|/ref|><|det|>[[110, 531, 884, 894]]<|/det|> +Given that a substrate receptor of E3 ubiquitin ligases must physically interact with the substrate, we examined the CRY2- LRB interaction (Fig. 3), using the co- immunoprecipitation (co- IP) assays of proteins co- expressed in the heterologous HEK293T (Human Embryo Kidney) cells as we previously reported14. Because we have previously shown that the blue light- induced phosphorylation of CRY2 is required for CRY2 ubiquitination and degradation, and that PPK1 is one of the four related protein kinases that specifically phosphorylate photoexcited CRY218,20, we tested LRB- CRY2 interaction in the presence of the wild- type PPK1 in response to blue light in HEK293T cells. LRB1 and LRB2 preferentially interact with phosphorylated CRY2 in HEK293T cells co- expressing PPK1, but only when cells were exposed to blue light (Fig. 3a- b). The light dependence of LRB- CRY2 interaction is confirmed by the result that LRB2 failed to interact with the photo- insensitive CRY2D387A mutant that does not bind the FAD (Flavin Adenine Dinucleotide) chromophore33 (Fig. 3c). The phosphorylation dependence of LRB- CRY2 interaction was confirmed by the result that LRB2 failed to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 886, 397]]<|/det|> +interact with CRY2 in light- treated HEK293T cells co- expressing the catalytically inactive PPK1D267N (Fig. 3d). These results demonstrate that LRBs directly and specifically interact with photoexcited and phosphorylated CRY2. We next examined LRB1- CRY2 and LRB2- CRY2 interactions in vivo, using the BiFC (Bimolecular Fluorescence Complementation) assay in \*Arabidopsis\* cells34. In this experiment, \*Arabidopsis\* leaves were transiently transformed by infiltration of \*Agrobacterium\* cells harboring plasmids expressing the indicated BiFC pairs of recombinant proteins34, and the direct protein- protein interactions between the BiFC pairs were analyzed by the BiFC assay. Figure 4 shows strong BiFC signals between CRY2 and LRB1 or CRY2 and LRB2. In contrast, no BiFC signal was detected between the photo- insensitive CRY2D387A and LRB1 or CRY2D387A and LRB2. These results clearly demonstrate the direct interaction of CRY2 and LRB proteins \*in vivo\*. + +<|ref|>text<|/ref|><|det|>[[110, 400, 886, 897]]<|/det|> +To further test the hypothesis that the Cul3LRBs and Cul4COP1-SPAs E3 ubiquitin ligases catalyze blue light- dependent CRY2 ubiquitination, we performed three independent experiments to investigate the activity of LRBs and COP1 in the blue light- induced polyubiquitination of CRY2, using the IP/immunoblot assays. In the first two experiments, we prepared transgenic plants overexpressing Flag- GFP- tagged CRY2 (FGFP- CRY2) in the wild- type (WT), the strong (\*lrb1lrb2- 1lrb3\*) or weak (\*lrb1lrb2- 2lrb3\*) alleles of the \*lrb123\* triple mutant (Fig. 5a- b, d), and the \*cop1\* mutant backgrounds (Fig. 5c- d). In the third experiment, we prepared double- tagged transgenic lines that overexpress FGFP- CRY2 with either the Myc- tagged LRB (Myc- LRB1, Myc- LRB2) or the Myc- tagged COP1 (Myc- COP1) (Fig. 5e). In the first experiment, polyubiquitinated proteins of samples were indiscriminately precipitated using the TUBE2 (Tandem Ubiquitin Binding Entity 2)- conjugated beads35, and the TUBE2- enriched proteins were analyzed by immunoblots probed with the anti- ubiquitin or anti- CRY2 antibodies (Fig. 5a- c). This experiment detected a high level of polyubiquitination of CRY2 in blue light- treated wild- type background seedlings (Fig. 5a- c). However, little ubiquitinated CRY2 was detected in etiolated seedlings or in blue light- treated \*lrb123\* or \*cop1\* mutant background seedlings (Fig. 5a- c). In the other two experiments, FGFP- CRY2 were immunoprecipitated by the GFP- trap beads, and the level of polyubiquitination of FGFP- CRY2 was analyzed by immunoblots probed against anti- ubiquitin antibody (Fig. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 881, 370]]<|/det|> +5d- e). In etiolated seedlings exposed to blue light, a high level of ubiquitinated FGFP- CRY2 was detected in the wild- type seedlings overexpressing FGFP- CRY2. In contrast, a lower level of ubiquitinated FGFP- CRY2 was detected in lrb1lrb2- 2lrb3 weak allele, whereas almost no ubiquitinated FGFP- CRY2 was detected in the lrb1lrb2- 1lrb3 strong allele or the cop1 mutant seedlings transgenically overexpressing FGFP- CRY2 (Fig. 5d). In comparison to FGFP- CRY2 overexpressed in the wild- type background, the blue light- induced polyubiquitination of GFP- CRY2 is enhanced in plants co- overexpressing LRB1, or LRB2, or COP1 (Fig. 5e). Taken together, results of those three independent experiments all support the hypothesis that two distinct E3 ubiquitin ligases, Cul3LRB8 and Cul4COP1-SPAS, can both catalyze blue light- dependent polyubiquitination of CRY2 in vivo. + +<|ref|>sub_title<|/ref|><|det|>[[115, 401, 736, 422]]<|/det|> +## COP1 and LRBs interact with different structure elements of CRY2 + +<|ref|>text<|/ref|><|det|>[[111, 427, 886, 894]]<|/det|> +Because the LRBs and COP1 proteins share little sequence or structural similarity, we reasoned that they are unlikely to interact with the identical structure elements of CRY2. The VP motif is a conserved COP1- binding motif of many substrates of the Cul4COP1-SPAS E3 ligase, including CRY236,37. We previously reported that the CRY2P532L mutation impaired at the VP motif (Fig. 6a) loses all the physiological activities tested but retains many photochemical properties of CRY2, including blue light- induced oligomerization, phosphorylation, ubiquitination, and degradation15. We first compared the relative activity of FGFP- CRY2 and FGFP- CRY2P532L, using the hypocotyl inhibition assay in transgenic plants expressing the respective recombinant protein at similar levels in the cry1cry2 mutant background (Supplementary Fig. 2a). Transgenic seedlings expressing FGFP- CRY2P532L showed no obvious activity mediating light inhibition of hypocotyl elongation under all light intensities of blue light tested (Fig. 6b, Supplementary Fig. 2b). This result confirmed our previous report that CRY2P532L is a loss- of- function mutation15. We then compared the kinetics of blue light- induced degradation of FGFP- CRY2 and FGFP- CRY2P532L, using the same transgenic lines (Fig. 6c- d). Although FGFP- CRY2P532L is degraded in response to blue light as we previously reported15, this recombinant CRY2 mutant protein appears to degrade slower than its wild- type counterpart (Fig. 6c- d). The apparent half- life of FGFP- CRY2P532L is more than + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 886, 504]]<|/det|> +twice that of FGFP- CRY2 (Fig. 6d). Because the LRBs only interact with phosphorylated CRY2 whereas COP1 interacts with both phosphorylated and unphosphorylated CRY2 (Fig. 6e), we examined the phosphorylation activity of the CRY2P532L mutant protein in HEK293T cells co- expressing the CRY2 kinase PPK1. CRY2P532L mutant undergoes blue light- and PPK1- dependent phosphorylation (Fig. 6e- f). Neither the unphosphorylated CRY2D387A control or the phosphorylated CRY2P532L mutant protein interacts with COP1 or SPA1 in response to blue light (Fig. 6e). Because the CRY2P532L mutant does not interact with COP1, its blue light- induced degradation is most likely dependent on another E3 ligase, such as LRBs. Consistent with this hypothesis, the photoexcited and phosphorylated CRY2P532L mutant protein interacts with LRB2 with no discernable difference in comparison to the wild- type CRY2 (Fig. 6f). This result explains why CRY2P532L that does not interact with COP1 or SPA1 can still be degraded in plants (Fig. 6c). Taken together, the results of our experiments argue that two types of substrate receptors, COP1 and LRBs, interact with CRY2 via distinct structural elements of CRY2 to enable the blue light- induced CRY2 ubiquitination and degradation by the respective E3 ubiquitin ligases, Cul3LRBs and Cul4COP1-SPAs. + +<|ref|>sub_title<|/ref|><|det|>[[444, 533, 551, 552]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 583, 881, 895]]<|/det|> +In this study, we demonstrate that, in addition to Cul4COP1-SPAs, the E3 ubiquitin ligases Cul3LRBs is required for blue light regulation and function of CRY2. We demonstrate that Cul3LRBs is responsible for the rapid ubiquitination and degradation of CRY2, whereas Cul4COP1-SPAs is responsible for the slow or prolonged ubiquitination and degradation of CRY2. We show that LRBs and COP1 interact with CRY2 via different structural elements: LRBs interacts with photoexcited and phosphorylated CRY2, but COP1 interacts with CRY2 regardless of its phosphorylation states. It is interesting that plants evolved with two distinct E3 ubiquitin ligases to regulate a CRY photoreceptor by different modes of interaction under different light conditions. In mammals, two related Cul1- based E3 ligases, SCFFBXL3 and SCFFBXL21, act to promote ubiquitination and degradation of CRYs in the nucleus and cytoplasm, respectively, to control the period of the circadian clock8. In Drosophila, Jetleg of a Cul1- based E3 ligase and Brwd3 of a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 886, 583]]<|/det|> +Cul4- based E3 ligase regulate ubiquitination and degradation of dCRY- dependent circadian rhythm \(^{9,10}\) . Cul4 \(^{Brwd3}\) mediates light- dependent ubiquitination and degradation of dCRY, whereas Cul1 \(^{Jettag}\) interacts with dCRY to catalyze ubiquitination of the dCRY- interacting protein TIM \(^{10,38}\) . Decreased TIM may enhance interaction of dCRY with its E3 ligases to accelerate its degradation in light. These studies demonstrate that multiple E3 ligases are needed to regulate the activity of a CRY to control the circadian clock. The results of this study argue that plant CRY2 is also controlled by two distinct E3 ligases, and that the delicate control of the abundance of CRYs is evolutionarily conserved for not only circadian rhythms in animals but also photomorphogenesis in plants. Our finding that Cul3 \(^{LRBs}\) or Cul4 \(^{COP1 - SPAs}\) catalyze CRY2 ubiquitination in different light conditions to promote rapid or slow degradation of CRY2, respectively, highlights the importance of photosensitivity to photomorphogenesis of plants. The complex mechanism regulating CRY2 turnover is likely evolved to optimize photomorphogenesis under natural light conditions. In this regard, it is particularly interesting that Cul3 \(^{LRBs}\) has been previously shown to be the E3 ligase of PIF3 and critical for the function of phyB and phytochrome- dependent photoresponses \(^{30,31}\) . Our finding that Cul3 \(^{LRBs}\) also catalyzes blue light- dependent ubiquitination of CRY2 to regulate blue light- dependent photomorphogenesis argues for a previously unrecognized mechanism toward a better understanding of how plants respond to light in the nature white light conditions. + +<|ref|>sub_title<|/ref|><|det|>[[456, 610, 540, 629]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 663, 317, 682]]<|/det|> +## Plasmid construction + +<|ref|>text<|/ref|><|det|>[[112, 688, 886, 894]]<|/det|> +Plasmid constructionPlasmid constructs in this study were generated by In- Fusion Cloning methods. The insertions were all verified by Sanger sequencing. The mammalian cell expression vectors used in this study were pQCMV- Flag- GFP, pQCMV- GFP and pCMV- Myc. pQCMV- Flag- GFP vector, driven by CMV promoter and with 1x Flag and GFP tags, was modified from pEGFP- N1 (Clontech). pQCMV- GFP vector, driven by CMV promoter and with a GFP tag, was modified from pQCMV- Flag- GFP. pCMV- Myc vector, driven by CMV promoter and with a 1x Myc tag, was described previously \(^{20}\) . To generate plasmids expressing Flag- LRB1, Flag- LRB2, Flag- CRY2 and Flag- CRY2 \(^{D387A}\) , the CDS regions of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 881, 293]]<|/det|> +respective genes were amplified either from Arabidopsis cDNA or previous plasmids, and in- fusion into Spel/Kpnl- digested pQCMV- Flag- GFP vector, resulting a Flag tag at the N- terminus of genes. Myc- LRB1, Myc- LRB2, Myc- CRY2, Myc- SPA1 and Myc- COP1 plasmids were prepared by cloning CDS into the BamHI site of pCMV- Myc vector, resulting a Myc tag at the N- terminus of genes. The CDS of PPK1 and PPK1D267N were amplified from previous plasmids by using the primers containing 2x HA coding sequences at the 5' end of forward primers, and in- fusion into Spel/Kpnl- digested pQCMV- GFP vector to generate HA- PPK1 and HA- PPK1D267N. + +<|ref|>text<|/ref|><|det|>[[112, 297, 884, 450]]<|/det|> +For BiFC constructs, the vectors used in this study were described previously39. The coding sequences of LRB1 (AT2G46260), LRB2 (AT3G61600), CRY2 (AT1G04400) and CRY2D387A were amplified and cloned into pDONR/Zeo entry vector. The entry constructs were then introduced into the destination vector pX- nYFP (N- terminus of YFP fused to C- terminus of genes) or pcCFP- X (C- terminus of CFP fused to N- terminus of genes) by LR reaction. + +<|ref|>text<|/ref|><|det|>[[112, 454, 881, 736]]<|/det|> +pFGFP and pDT1H binary vectors were used for creating overexpression transgenic lines. The pFGFP binary vector, driven by Actin2 promoter and with 2x Flag and GFP tags, was modified from pCambia330120. The pDT1H binary vector, possessing two expression cassettes which can sequentially insert two genes into one vector, was modified from previously published vector pDT140 by replacing the BAR gene with HPT gene which converts basta resistance to hygromycin resistance in plants. The CDS of LRB2, CRY2 and CRY2P532L were cloned into BamHI- digested pFGFP vector to generate FGFP- LRB2, FGFP- CRY2, FGFP- CRY2P532L with 2x Flag and GFP tags fused to N- terminus of the genes. The CDS regions of LRB1, LRB2 and COP1 (AT2G32950) were cloned into the BamHI site of pDT1H vector to produce Myc- LRB1, Myc- LRB2 and Myc- COP1 with 4x Myc tags fused to N- terminus of the genes. + +<|ref|>sub_title<|/ref|><|det|>[[115, 742, 474, 762]]<|/det|> +## Plant materials and growth conditions + +<|ref|>text<|/ref|><|det|>[[113, 767, 881, 893]]<|/det|> +All mutants used in this study are in the Arabidopsis thaliana Columbia ecotype background. The cry1cry2 double mutant were described as previously32. The lrb1lrb2- 1lrb3 (lrb1, Salk_145146; lrb2- 1, Salk_001013; lrb3, Salk_082868) and lrb1lrb2- 2lrb3 (lrb2- 2, Salk- 044446) triple mutants were gifts from Dr. Peter Quail and as described previously30,31. cop1- 4 and cop1- 6 are weak mutant alleles of COP1 as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 884, 240]]<|/det|> +previously described41. cry1cry2lrb1lrb2- 2lrb3 and lrb1lrb2- 2lrb3cop1- 4 mutants were generated by crossing. The genotypes of lrb123 mutants were verified by PCR using the primers listed in Supplementary Table 2. cop1- 4 mutant was verified by PCR using the primers listed in Supplementary Table 2, followed by Sanger sequencing to confirm the point mutation of COP1. cry1cry2 mutant was verified by western blots using antibodies against CRY1 and CRY2 proteins. + +<|ref|>text<|/ref|><|det|>[[111, 245, 881, 710]]<|/det|> +All transgenic lines were generated via Agrobacterium tumefaciens- mediated floral- dip method42. The wild- type plants used for transformation in this study are rdr6- 11, which suppresses gene silencing43. For in vivo ubiquitination study, FGFP- CRY2 was introduced into rdr6- 11, lrb1lrb2- 1lrb3, lrb1lrb2- 2lrb3, and cop1- 6 background. The FGFP- CRY2/Myc- LRB1 double- overexpression lines were prepared by introducing FGFP- CRY2 into Myc- LRB1/rdr6- 11 plants. The transgenic T1 populations were screened on MS plates containing 25 mg/L Glufosinate- ammonium (cat # CP6420, Bomei Biotechnology) and 25 mg/L hygromycin (cat # 10843555001, Roche), and western blots were performed to confirm the expression of both proteins. The same method was used for generating FGFP- CRY2/Myc- LRB2 and FGFP- CRY2/Myc- COP1 double- overexpression lines. For experiments comparing the hypocotyl phenotype and protein degradation kinetics of FGFP- CRY2 and FGFP- CRY2P532L, FGFP- CRY2 and FGFP- CRY2P532L were introduced into cry1cry2rdr6 background. The cry1cry2rdr6 were generated by crossing cry1- 30432, cry2- 112, and rdr6- 1143. The transgenic lines were screened on MS plates with 25 mg/L Glufosinate- ammonium, and lines with similar protein expression level of FGFP- CRY2 and FGFP- CRY2P532L were used for analysis. For lrb123 mutant blue- light hypersensitivity phenotype rescue experiments, FGFP- LRB2 and Myc- LRB2 were transformed into lrb1lrb2- 2lrb3 background. + +<|ref|>text<|/ref|><|det|>[[112, 715, 888, 894]]<|/det|> +For routine maintenance, Arabidopsis thaliana were grown under long day conditions (16 hours light / 8 hours dark) at 22°C. For hypocotyl phenotype analysis, seedlings were grown on MS plates with 3% sucrose at 20- 22°C for 6 days under different light conditions. Light- emitting diode (LED) was used to obtain monochromatic light (blue light, peak 465 nm, half- bandwidth of 25 nm; red light, peak 660 nm, half- bandwidth of 20 nm; far- red light, peak 735 nm, half- bandwidth of 21 nm). For endogenous CRY2 degradation analysis in WT, lrb123, cop1 and lrb123cop1 mutants, seedlings were grown in darkness + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 292]]<|/det|> +on MS plates with \(3\%\) sucrose for 7 days, then subjected to \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for the indicated time. For FGFP- CRY2 and FGFP- CRY2 \(^{P532L}\) degradation analysis, seedlings were grown in darkness on MS plates with \(3\%\) sucrose for 7 days, then subjected to \(100 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for the indicated time. For immunoprecipitation of polyubiquitinated proteins, 7- day- old etiolated seedlings grown on MS medium containing \(3\%\) sucrose were treated with \(50 \mu \mathrm{M} \mathrm{MG132}\) (Cat # S2619, Selleck) in liquid MS in the dark overnight with gentle shaking, and then moved to \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes before harvest. + +<|ref|>sub_title<|/ref|><|det|>[[115, 298, 537, 319]]<|/det|> +## Blue-light-induced CRY2 degradation assays + +<|ref|>text<|/ref|><|det|>[[111, 323, 884, 555]]<|/det|> +For endogenous CRY2 degradation analysis, seedlings were grown in the dark for seven days and then treated with \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) of blue light for indicated time. For FGFP- CRY2 and FGFP- CRY2 \(^{P532L}\) degradation analysis, seedlings were treated with \(100 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) of blue light. Tissues were homogenized by TissueLyser (QIAGEN). Proteins were extracted in equal tissue volume of protein extraction buffer [120mM Tris pH 6.8, 100mM EDTA, \(4\%\) w/v SDS, \(10\%\) v/v beta- mercaptoethanol, \(5\%\) v/v Glycerol, \(0.05\%\) w/v Bromophenol blue] \(^{44}\) , heated at \(100^{\circ} \mathrm{C}\) for 8 minutes, centrifuged at 16000 rcf for 10 minutes, and analyzed by western blot. Proteins were separated in \(10\%\) SDS- PAGE gels, and transferred to nitrocellulose membranes (Pall Life Sciences). + +<|ref|>text<|/ref|><|det|>[[111, 559, 884, 894]]<|/det|> +For immunoblot signals detected by enhance conventional chemiluminescent (ECL) method, membranes were blocked with \(5\%\) non- fat milk in PBST for 1 hour, blotted with anti- CRY2 primary antibody (1:3000, homemade) in PBST for 1.5 hours, washed 8 minutes x 3 times with PBST, blotted with anti- Rabbit- HRP (1:10000, cat # 31460, ThermoFisher) secondary antibody in PBST for 1.5 hours, washed 8 minutes x 3 times with PBST and then incubated with ECL solution for X- ray film development. The membranes were then stripped with stripping buffer [0.2 M Glycine, pH 2.5] and re- probed with anti- HSP82 (1:10000, cat # AbM51099- 31- PU, Beijing Protein Innovation) primary antibody and anti- Mouse- HRP (1:10000, cat # 31430, ThermoFisher) secondary antibody. Anti- HSP82 antibody from Oryza sativa can also recognized Arabidopsis HSP90 protein \(^{45}\) . Three independent blots were performed in parallel for quantification analysis. The quantification of protein intensity of ECL immunoblots was performed by ImageJ. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 397]]<|/det|> +For immunoblot signals detected by Odyssey CLx Infrared Imaging System (Li- COR), membranes were blocked with \(0.5\%\) casein in PBS for 1 hour, blotted with anti- CRY2 (1:3000) and anti- HSP82 (1:10000) mixed primary antibodies in PBST with \(0.5\%\) casein for 1.5 hours, washed 8 minutes x 3 times with PBST, blotted with Donkey anti-rabbit 790 (1:15000, cat # A11374, ThermoFisher) and Donkey anti- mouse 680 (1:15000, cat # A10038, ThermoFisher) mixed secondary antibodies in PBST with \(0.5\%\) casein for 1.5 hours, washed 8 minutes x 3 times with PBST and then signals were captured with Odyssey CLx by Image Studio Lite software. Three independent blots were performed in parallel for quantification analysis. Quantification of signals were processed with Image Studio Lite software. CRY2 degradation curves were indicated by CRY2 (B/D) ratio, calculated as CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark, and analyzed with one phase decay of nonlinear regression. + +<|ref|>sub_title<|/ref|><|det|>[[115, 402, 728, 422]]<|/det|> +## Protein expression and co-immunoprecipitation in HEK293T cells + +<|ref|>text<|/ref|><|det|>[[111, 428, 888, 841]]<|/det|> +Human embryonic kidney (HEK) 293T cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with \(10\%\) (v/v) FBS, \(100 \text{mg/L}\) streptomycin and \(100 \text{IU penicillin}\) , in humidified \(5\%\) (v/v) CO2 air at \(37^{\circ}\text{C}\) . Cells were seeded at a density of approximately \(8 \times 10^{5}\) cells/6- cm plate and the transfection were carried out with a calcium phosphate precipitation protocol. Briefly, different combinations of plasmid DNA (2\~5 μg / construct) were mixed with \(30 \mu \text{l}\) 2.5 M CaCl2 and ddH2O to a total volume of \(300 \mu \text{l}\) , then \(300 \mu \text{l}\) of 2x HeBS [250 mM NaCl, 10 mM KCl, 1.5 mM Na2HPO4, 12 mM Dextrose and 50 mM HEPES, adjust the pH of the final solution to 7.05] was added drop by drop with vortex, and kept at room temperature for 5 minutes before applying to cells. The media were aspirated from each plate, DNA mixtures were gently added into plates and the plates were rotated gently to allow the mixtures to coat the entire plate. 3 ml of fresh media containing 25 μM chloriquine (cat # C6628, Sigma) were added to each plate and the plates were kept in the CO2 incubator overnight. The next day, the media were changed with 3 ml of fresh media without chloriquine. 36- 48 hours after transfection, the cells were subjected to blue light treatment for indicated time, washed twice with cold PBS buffer and then harvested in liquid nitrogen for co- immunoprecipitation. + +<|ref|>text<|/ref|><|det|>[[115, 846, 888, 893]]<|/det|> +Cells transfected with different plasmid DNA were lysed in \(800 \mu \text{l}\) 1% Brij buffer [1% Brij- 35, 50 mM Tris- HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 1x Protease inhibitor cocktail, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 884, 397]]<|/det|> +and 1x phosphatase inhibitor PhosSTOP] with rotating at \(4^{\circ}C\) for 20 minutes. Cell lysates were centrifuged at 16,000 rcf for 10 minutes at \(4^{\circ}C\) , and the supernatants were incubated with 20 \(\mu\)l Anti- FLAG M2 affinity gel (Cat. # F2426, Sigma) at \(4^{\circ}C\) for 2 hours with rotation. Beads were washed with \(1\%\) Brij buffer for 5 times. Proteins were competed from the beads with \(35\mu\) 3xFlag peptide solution [500 ng/ul in \(1\%\) Brij buffer] for 30 minutes at room temperature with mixing. Elution was transferred to a new tube, mixed with 4XSDS sample buffer, denatured at \(100^{\circ}C\) for 4 minutes and subjected to western blot analysis. Western blots were performed as described above. Primary antibodies used here were anti- Flag (1:1500, cat # F1804, Sigma), anti- Myc (1:5000, cat # 05- 724, Millipore) and anti- HA (1:5000, cat # 12013819001, Roche), and secondary antibodies used were anti- Mouse- HRP (1:10000, cat # 31430, ThermoFisher) and anti- Rabbit- HRP (1:10000, cat # 31460, ThermoFisher). + +<|ref|>sub_title<|/ref|><|det|>[[115, 402, 649, 422]]<|/det|> +## Bimolecular fluorescence complementation (BiFC) assay + +<|ref|>text<|/ref|><|det|>[[110, 428, 884, 867]]<|/det|> +BiFC assays in Arabidopsis plants were performed as previously described methods with minor modifications34. Briefly, Agrobacteria AGL0 transformed with BiFC plasmids were grown in LB medium with \(40\mu\) M Acetosyringone and \(1\%\) Glucose in \(28^{\circ}C\) shaker overnight. Agrobacteria were centrifuged at 5000 rcf for 10 minutes, washed once with washing buffer [10 mM MgCl2, 100 \(\mu\) M Acetosyringone] and resuspended in infiltration buffer [1/4 MS pH 6.0, \(1\%\) sucrose, 100 \(\mu\) M Acetosyringone and \(0.01\%\) Silwet L- 77] to 0.5 of OD600. The agrobacteria were infiltrated into 3- 4 weeks old Arabidopsis leaves with syringe. Each BiFC assay was performed with at least 3 independent plants with 2- 3 leaves infiltrated for each. Plant leaves were dried before kept in the dark for 24 hours, and then moved back to long day conditions (16 h light/8 h dark) to grow for two more days. The infiltrated leaf samples were analyzed under a Leica TCS SP8X confocal microscope. Hoechst 33342 was used for nuclei staining. The quantification of fluorescence intensity was performed by ImageJ. Briefly, the images of GFP and Hoechst 33342 channels from the same field were stacked first, then the regions of nuclei were selected on the image of Hoechst 33342 channel. The integrated intensity of the selected regions over the entire stack were measured. The BiFC ratio of the selected regions were calculated as [GFP intensity]nuclei / [Hoechst 33342 intensity]nuclei for each + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 876, 137]]<|/det|> +image. Four to nine images ( \(\sim 40 - 100\) nuclei in total) for each BiFC assay were used for quantification and the BiFC ratios were presented as mean \(\pm\) SD (standard deviation). + +<|ref|>sub_title<|/ref|><|det|>[[115, 142, 690, 163]]<|/det|> +## Immunoprecipitation of ubiquitinated proteins in Arabidopsis + +<|ref|>text<|/ref|><|det|>[[112, 167, 882, 370]]<|/det|> +7- day- old etiolated seedlings were treated with \(50 \mu \mathrm{M}\) MG132 in liquid MS in the dark overnight with gentle shaking, and then treated with \(30 \mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes before harvest. For each immunoprecipitation, about 3 g of seedlings were used. Seedlings were ground in liquid nitrogen and homogenized in 1.5x tissue volume of IP buffer [50 mM Tris- HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Triton X- 100, 20 mM NaF and 1x Protease inhibitor cocktail] at \(4^{\circ} \mathrm{C}\) for 20 minutes with mixing. Lysates were centrifuged at 16000 rcf for 15 minutes at \(4^{\circ} \mathrm{C}\) . \(100 \mu \mathrm{l}\) of supernatant were saved as inputs. + +<|ref|>text<|/ref|><|det|>[[112, 375, 880, 554]]<|/det|> +For total ubiquitinated protein purification, the supernatant were incubated with \(30 \mu \mathrm{l}\) Agarose- TUBE2 beads (cat # UM402M, LifeSensors) for 2 hours at \(4^{\circ} \mathrm{C}\) with rotating. After binding, beads were pelleted and washed 4 times with ice- cold IP buffer (without inhibitors). Total ubiquitinated proteins were eluted with 2XSDS sample buffer, denatured at \(100^{\circ} \mathrm{C}\) for 4 minutes and subjected to western blot analysis. Anti- ubiquitin antibody ( \(\alpha\) - Ubaq, cat # 14- 6078- 80, Thermofisher) was used to detect total ubiquitinated proteins, and homemade anti- CRY2 antibody was used to detect CRY2 protein. + +<|ref|>text<|/ref|><|det|>[[112, 558, 882, 710]]<|/det|> +For FGFP- CRY2 protein purification, the supernatant were incubated with \(50 \mu \mathrm{l}\) GFP- trap agarose beads (homemade or cat # gta- 20, Chromotek) and rotated at \(4^{\circ} \mathrm{C}\) for 2 hours. Beads were pelleted and washed 4 times with IP buffer (without inhibitors). Proteins were eluted with 2XSDS sample buffer by heating at \(100^{\circ} \mathrm{C}\) for 4 minutes and subjected to western blot analysis. Anti- Flag antibody was used to detect FGFP- CRY2 (IP) fusion protein, and anti- ubiquitin antibody was used to detect ubiquitinated CRY2. + +<|ref|>sub_title<|/ref|><|det|>[[115, 770, 223, 787]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[115, 812, 830, 857]]<|/det|> +1 Cashmore, A. R. Cryptochromes: enabling plants and animals to determine circadian time. Cell 114, 537- 543 (2003). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[168, 88, 884, 137]]<|/det|> +Sancar, A. Cryptochrome: the second photoactive pigment in the eye and its role in circadian photoreception. Annu Rev Biochem 69, 31- 67 (2000). + +<|ref|>text<|/ref|><|det|>[[168, 141, 866, 191]]<|/det|> +Wang, Q. & Lin, C. Mechanisms of Cryptochrome- Mediated Photoresponses in Plants. Annu Rev Plant Biol 71, 103- 129, doi:10.1146/annurev- arplant- 050718- 100300 (2020). + +<|ref|>text<|/ref|><|det|>[[168, 196, 886, 245]]<|/det|> +Busino, L. et al. SCFFbx13 controls the oscillation of the circadian clock by directing the degradation of cryptochrome proteins. Science 316, 900- 904 (2007). + +<|ref|>text<|/ref|><|det|>[[168, 249, 884, 270]]<|/det|> +Godinho, S. I. et al. The after- hours mutant reveals a role for Fbx13 in determining mammalian circadian period. Science 316, 897- 900 (2007). + +<|ref|>text<|/ref|><|det|>[[168, 275, 886, 323]]<|/det|> +Mammalian circadian period. Science 316, 897- 900 (2007). + +<|ref|>text<|/ref|><|det|>[[168, 327, 847, 398]]<|/det|> +Siepka, S. M. et al. Circadian mutant Overtime reveals F- box protein FBXL3 regulation of cryptochrome and period gene expression. Cell 129, 1011- 1023 (2007). + +<|ref|>text<|/ref|><|det|>[[168, 402, 850, 451]]<|/det|> +Hirano, A. et al. FBXL21 Regulates Oscillation of the Circadian Clock through Ubiquitination and Stabilization of Cryptochromes. Cell 152, 1106- 1118, doi:10.1016/j.cell.2013.01.054 (2013). + +<|ref|>text<|/ref|><|det|>[[168, 456, 884, 505]]<|/det|> +Yoo, S.- H. et al. Competing E3 Ubiquitin Ligases Govern Circadian Periodicity by Degradation of CRY in Nucleus and Cytoplasm. Cell 152, 1091- 1105, doi:http://dx.doi.org/10.1016/j.cell.2013.01.055 (2013). + +<|ref|>text<|/ref|><|det|>[[168, 509, 646, 530]]<|/det|> +doi:http://dx.doi.org/10.1016/j.cell.2013.01.055 (2013). + +<|ref|>text<|/ref|><|det|>[[168, 535, 888, 584]]<|/det|> +Peschel, N., Chen, K. F., Szabo, G. & Stanewsky, R. Light- dependent interactions between the Drosophila circadian clock factors cryptochrome, jetlag, and timeless. + +<|ref|>text<|/ref|><|det|>[[168, 589, 430, 609]]<|/det|> +Curr Biol 19, 241- 247 (2009). + +<|ref|>text<|/ref|><|det|>[[168, 614, 884, 635]]<|/det|> +Ozturk, N., VanVickle- Chavez, S. J., Akileswaran, L., Van Gelder, R. N. & Sancar, + +<|ref|>text<|/ref|><|det|>[[168, 640, 884, 690]]<|/det|> +A. Ramshackle (Brwd3) promotes light-induced ubiquitylation of Drosophila Cryptochrome by DDB1-CUL4-ROC1 E3 ligase complex. Proceedings of the National Academy of Sciences 110, 4980-4985, doi:10.1073/pnas.1303234110 (2013). + +<|ref|>text<|/ref|><|det|>[[168, 694, 864, 765]]<|/det|> +Lin, C. et al. Enhancement of blue-light sensitivity of Arabidopsis seedlings by a blue light receptor cryptochrome 2. Proc Natl Acad Sci U S A 95, 2686- 2690 (1998). + +<|ref|>text<|/ref|><|det|>[[168, 769, 866, 842]]<|/det|> +Lin, C. et al. Enhancement of blue- light sensitivity of Arabidopsis seedlings by a blue light receptor cryptochrome 2. Proc Natl Acad Sci U S A 95, 2686- 2690 (1998).Guo, H., Yang, H., Mockler, T. C. & Lin, C. Regulation of flowering time by Arabidopsis photoreceptors. Science 279, 1360- 1363 (1998). + +<|ref|>text<|/ref|><|det|>[[168, 846, 758, 867]]<|/det|> +Guo, H., Yang, H., Mockler, T. C. & Lin, C. Regulation of flowering time by + +<|ref|>text<|/ref|><|det|>[[168, 872, 707, 893]]<|/det|> +Arabidopsis photoreceptors. Science 279, 1360- 1363 (1998). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 88, 888, 899]]<|/det|> +496 13 Sang, Y. et al. N-terminal domain-mediated homodimerization is required for photoreceptor activity of Arabidopsis CRYPTOCHROME 1. Plant Cell 17, 1569- 1584 (2005). 499 14 Wang, Q. et al. Photoactivation and inactivation of Arabidopsis cryptochrome 2. Science 354, 343- 347, doi:10.1126/science.aaf9030 (2016). 501 15 Liu, Q. et al. Photooligomerization Determines Photosensitivity and Photoreactivity of Plant Cryptochromes. Mol Plant 13, 398- 413, doi:10.1016/j.molp.2020.01.002 (2020). 504 16 Ma, L. et al. Structural insights into BIC- mediated inactivation of Arabidopsis cryptochrome 2. Nat Struct Mol Biol 27, 472- 479, doi:10.1038/s41594- 020- 0410- z (2020). 507 17 Shao, K. et al. The oligomeric structures of plant cryptochromes. Nat Struct Mol Biol 27, 480- 488, doi:10.1038/s41594- 020- 0420- x (2020). 509 18 Shalitin, D. et al. Regulation of Arabidopsis cryptochrome 2 by blue- light- dependent phosphorylation. Nature 417, 763- 767. (2002). 511 19 Wang, Q. et al. The Blue Light- Dependent Phosphorylation of the CCE Domain Determines the Photosensitivity of Arabidopsis CRY2. Molecular Plant 8, 631- 643, doi:http://dx.doi.org/10.1016/j.molp.2015.03.005 (2015). 514 20 Liu, Q. et al. Molecular basis for blue light- dependent phosphorylation of Arabidopsis cryptochrome 2. Nat Commun 8, 15234, doi:10.1038/ncomms15234 (2017). 516 21 Yu, X. et al. Arabidopsis cryptochrome 2 completes its posttranslational life cycle in the nucleus. Plant Cell 19, 3146- 3156 (2007). 519 22 Liu, Q. et al. The Blue Light- Dependent Polyubiquitination and Degradation of Arabidopsis Cryptochrome2 Requires Multiple E3 Ubiquitin Ligases. Plant Cell Physiol 57, 2175- 2186, doi:10.1093/pcp/pcw134 (2016). 522 23 Quail, P. H. Phytochromes. Current biology : CB 20, R504- 507, doi:10.1016/j.cub.2010.04.014 (2010). 524 24 Legris, M., Ince, Y. C. & Fankhauser, C. Molecular mechanisms underlying phytochrome- controlled morphogenesis in plants. Nat Commun 10, 5219, doi:10.1038/s41467- 019- 13045- 0 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 87, 888, 848]]<|/det|> +Wang, Q. et al. New insights into the mechanisms of phytochrome- cryptochrome coaction. New Phytol 217, 547- 551, doi:10.1111/nph.14886 (2018).Pham, V. N., Kathare, P. K. & Huq, E. Phyochromes and Phytochrome Interacting Factors. Plant Physiol 176, 1025- 1038, doi:10.1104/pp.17.01384 (2018).Ni, W. et al. PPKs mediate direct signal transfer from phytochrome photoreceptors to transcription factor PIF3. Nat Commun 8, 15236, doi:10.1038/ncomms15236 (2017).Han, X., Huang, X. & Deng, X. W. The Photomorphogenic Central Repressor COP1: Conservation and Functional Diversification during Evolution. Plant Communications 1, 1- 13, doi:doi.org/10.1016/j.xplc.2020.100044 (2020).Hoecher, U. The activities of the E3 ubiquitin ligase COP1/SPA, a key repressor in light signaling. Curr Opin Plant Biol 37, 63- 69, doi:10.1016/j.pbi.2017.03.015 (2017).Ni, W. et al. A mutually assured destruction mechanism attenuates light signaling in Arabidopsis. Science 344, 1160- 1164, doi:10.1126/science.1250778 (2014).Christians, M. J., Gingerich, D. J., Hua, Z., Lauer, T. D. & Vierstra, R. D. The light- response BTB1 and BTB2 proteins assemble nuclear ubiquitin ligases that modify phytochrome B and D signaling in Arabidopsis. Plant Physiol 160, 118- 134, doi:10.1104/pp.112.199109 (2012).Mockler, T. C., Guo, H., Yang, H., Duong, H. & Lin, C. Antagonistic actions of Arabidopsis cryptochromes and phytochrome B in the regulation of floral induction. Development 126, 2073- 2082 (1999).Liu, H. et al. Photoexcited CRY2 interacts with CIB1 to regulate transcription and floral initiation in Arabidopsis. Science 322, 1535- 1539 (2008).Zhang, Y. et al. A Highly Efficient Agrobacterium- Mediated Method for Transient Gene Expression and Functional Studies in Multiple Plant Species. Plant Communications 1, 1- 12, doi:doi.org/10.1016/j.xplc.2020.100028 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[0, 88, 888, 870]]<|/det|> +Hjerpe, R. et al. Efficient protection and isolation of ubiquitylated proteins using tandem ubiquitin-binding entities. EMBO Rep 10, 1250- 1258, doi:10.1038/embor.2009.192 (2009).Lau, K., Podolec, R., Chappuis, R., Ulm, R. & Hothorn, M. Plant photoreceptors and their signaling components compete for COP1 binding via VP peptide motifs. EMBO J 38, e102140, doi:10.15252/embj.2019102140 (2019).Ponnu, J., Riedel, T., Penner, E., Schrader, A. & Hoecker, U. Cryptochrome 2 competes with COP1 substrates to repress COP1 ubiquitin ligase activity during Arabidopsis photomorphogenesis. Proc Natl Acad Sci U S A, doi:10.1073/pnas.1909181116 (2019).Koh, K., Zheng, X. & Sehgal, A. JETLAG resets the Drosophila circadian clock by promoting light- induced degradation of TIMELESS. Science 312, 1809- 1812 (2006).Gampala, S. S. et al. An essential role for 14- 3- 3 proteins in brassinosteroid signal transduction in Arabidopsis. Dev Cell 13, 177- 189, doi:10.1016/j.devcel.2007.06.009 (2007).He, Z. et al. Construction and Validation of a Dual- Transgene Vector System for Stable Transformation in Plants. J Genet Genomics 43, 199- 207, doi:10.1016/j.jgg.2016.02.005 (2016).McNellis, T. W. et al. Genetic and molecular analysis of an allelic series of cop1 mutants suggests functional roles for the multiple protein domains. Plant Cell 6, 487- 500, doi:10.1105/tpc.6.4.487 (1994).Clough, S. J. Floral dip: agrobacterium- mediated germ line transformation. Methods Mol Biol 286, 91- 102, doi:10.1385/1- 59259- 827- 7:091 (2005).Peragine, A., Yoshikawa, M., Wu, G., Albrecht, H. L. & Poethig, R. S. SGS3 and SGS2/SDE1/RDR6 are required for juvenile development and the production of trans- acting siRNAs in Arabidopsis. Genes Dev 18, 2368- 2379 (2004).Tsugama, D., Liu, S. & Takano, T. A rapid chemical method for lysing Arabidopsis cells for protein analysis. Plant Methods 7, 22, doi:10.1186/1746- 4811- 7- 22 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 839, 135]]<|/det|> +45 Li, X. et al. Identification and validation of rice reference proteins for western blotting. J Exp Bot 62, 4763- 4772, doi:10.1093/jxb/err084 (2011). + +<|ref|>sub_title<|/ref|><|det|>[[115, 194, 291, 212]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[113, 219, 886, 371]]<|/det|> +The authors thank Dr. Peter Quail for providing the Irb123 mutant alleles. Works in the authors' laboratories are supported in part by the National Natural Science Foundation of China (31970265 to Q.W.), Natural Science Foundation of Fujian Province (2019J06014 to Q.W.), FAFU- ICE fund (KXGH17011 to Q.W.) and the National Institutes of Health (GM56265 to C.L.). The UCLA- FAFU Joint Research Center on Plant Proteomics provided the institutional supports. + +<|ref|>sub_title<|/ref|><|det|>[[115, 429, 313, 448]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[113, 454, 875, 579]]<|/det|> +Y.C. and X.H. conducted most of the experiments. S.L. and T.S. performed some biochemical experiments in HEK293T. H.H. and H.R. prepared some genetic materials. S.L. and Z.G. helped with BiFC experiments. X.W. and D.L. provided critical feedback. C.L. and Q.W. conceived the project, designed the experiments, and wrote the manuscript. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 125, 218, 144]]<|/det|> +## Figures + +<|ref|>text<|/ref|><|det|>[[144, 184, 574, 204]]<|/det|> +Fig. 1 LRBs are required for blue light responses. + +<|ref|>text<|/ref|><|det|>[[144, 241, 824, 262]]<|/det|> +Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2, respectively. + +<|ref|>text<|/ref|><|det|>[[144, 300, 816, 350]]<|/det|> +Fig. 3 LRBs interact with phosphorylated Arabidopsis CRY2 expressed in the HEK293T cells. + +<|ref|>text<|/ref|><|det|>[[144, 389, 483, 409]]<|/det|> +Fig. 4 LRBs interact with CRY2 in vivo. + +<|ref|>text<|/ref|><|det|>[[144, 447, 712, 468]]<|/det|> +Fig. 5 LRBs and COP1 are both required for CRY2 ubiquitination. + +<|ref|>text<|/ref|><|det|>[[144, 505, 832, 555]]<|/det|> +Fig. 6 The VP motif of CRY2 is required for the CRY2- COP1 interaction but not the CRY2- LRB interaction. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 110, 850, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 548, 852, 775]]<|/det|> +
Fig. 1 LRBs are required for blue light responses. a, 6-day-old seedlings grown in darkness, red light (20 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ), far-red light (6 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) , long day (16h light / 8h dark), or short day (8h light / 16h dark). b, Hypocotyl length of indicated genotypes of (a). c, 6-day-old seedlings grown in darkness or blue light (10 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). d, Hypocotyl length of indicated genotypes grown under blue light of different intensities (0, 5, 10, 20, 40 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ) for 6 days. e-g, 6-day old seedlings grown in darkness or blue light (20 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). f-h, Hypocotyl length of indicated genotypes as in (e) and (g). Standard deviations (n≥20) are shown.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[145, 117, 853, 610]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 625, 747, 647]]<|/det|> +
Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2,
+ +<|ref|>text<|/ref|><|det|>[[140, 653, 857, 884]]<|/det|> +Fig. 2 LRBs and COP1 regulate fast or slow proteolysis of CRY2, respectively. a- b, Representative immunoblots (a) and a quantitative analysis (b, n=3) showing the endogenous CRY2 in the 7-day-old etiolated Irb123 mutants irradiated with \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) blue light for the indicated time. c- d, Representative immunoblots (c) and a quantitative analysis (d, n=3) showing the endogenous CRY2 in seedlings of indicated genotypes grown in darkness (D) or continuous blue light (cB, \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). e, Representative immunoblots showing degradation of the endogenous CRY2 in 7-day-old etiolated seedlings irradiated with \(30 \mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) of blue light. f, Quantitative results of CRY2 degradation from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 94, 822, 204]]<|/det|> +immunoblots in (e), analyzed with one phase decay of nonlinear regression. CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark. CRY2 and HSP90 were probed with anti- CRY2 antibody and anti- HSP90 antibody. HSP90 is a loading control. Arrows indicate phosphorylated CRY2. + +<|ref|>image<|/ref|><|det|>[[147, 290, 848, 558]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[456, 266, 539, 285]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[142, 565, 855, 824]]<|/det|> +Fig. 3 LRBs interact with phosphorylated Arabidopsis CRY2 expressed in the HEK293T cells. a- d, Co- immunoprecipitation results showing the blue light- dependent interaction of LRBs and phosphorylated CRY2 in heterologous HEK293T cells. HEK293T cells co- transfected with indicated plasmids were either kept in the dark (-) or treated with 100 \(\mu \mathrm{m} \mathrm{m}^{2} \mathrm{s}^{-1}\) blue light for 2 hours (+). Immunoprecipitations were performed with Flag- conjugated beads. CRY2D387A and PPK1D267N are inactive mutants of CRY2 and PPK1, respectively. Anti- Flag antibody, anti- Myc antibody or anti- HA antibody were used for detecting respective tagged proteins. Arrows show phosphorylated CRY2. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 115, 850, 440]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 449, 850, 472]]<|/det|> +
Fig. 4 LRBs interact with CRY2 in vivo. Confocal microscopy images showing
+ +<|ref|>text<|/ref|><|det|>[[140, 480, 857, 678]]<|/det|> +BiFC of indicated protein pairs transiently expressed in Arabidopsis plants. The Hoechst 33342 (nuclei) and GFP (BiFC) signals are shown. The relative activity of CRY2- LRB interaction was presented as BiFC ratio, calculated as \(\text{BiFC ratio} = [\text{GFP intensity}]^{\text{nuclei}} / [\text{Hoechst 33342 intensity}]^{\text{nuclei}}\) , quantified from each image (n≥4). Total number of quantified images and nuclei in the images is indicated underneath. CRY2D387A inactive mutant is used as the negative control. Standard deviations (n≥4 images) are shown. Scale bars, 10 μm. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 152, 845, 747]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 753, 750, 774]]<|/det|> +
Fig. 5 LRBs and COP1 are both required for CRY2 ubiquitination.
+ +<|ref|>text<|/ref|><|det|>[[141, 783, 839, 895]]<|/det|> +Immunoblots showing the ubiquitination of FGFP- CRY2 in indicated genotypes. 7- day- old etiolated seedlings constitutively expressing FGFP- CRY2 in indicated genotypes, pretreated with MG132, were kept in the dark (D) or exposed to 30 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 5, 10 and 15 minutes (B) before harvest for IP assays. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 92, 851, 413]]<|/det|> +a- c, Total ubiquitinated proteins were purified by TUBE2- conjugated beads. Immunoprecipitated proteins were analyzed by immunoblots probed with anti- ubiquitin antibody (a- Ubq) or anti- CRY2 antibody (a- CRY2). The extent of CRY2 ubiquitination were shown underneath, calculated as [CRY2- Ubq intensity]IP / [Ubq intensity]IP. d- e, FGFP- CRY2 proteins were purified with GFP- trap beads. Immunoprecipitated proteins were analyzed by immunoblots probed with anti- ubiquitin antibody (a- Ubq), anti- Flag antibody (a- Flag) or anti- Myc antibody (a- Myc) for respective epitope- tagged proteins. The extent of CRY2 ubiquitination is calculated by [CRY2- Ubq intensity]IP / [Flag intensity]IP with the short exposure immunoblots and shown underneath. S. exp or L. exp: short or long chemiluminescence exposures of immunoblots. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 110, 854, 635]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 646, 855, 905]]<|/det|> +
Fig. 6 The VP motif of CRY2 is required for the CRY2-COP1 interaction but not the CRY2-LRB interaction. a, A schematic diagrams depicting the CRY2P532L mutation. PHR, photolyase homologous region; CCE, CRY C-terminal extension; DQQVPSAV, VP motif in CRY2; numbers, amino acid positions. b, Hypocotyl length of 6-day-old seedlings of indicated genotypes grown under blue light of different light intensities (0, 10, 30, 60, 100 \(\mu \mathrm{mol} \mathrm{m}^{-2} \mathrm{s}^{-1}\) ). FGFP-CRY2 and FGFP-CRY2P532L were constitutively expressed in cry1cry2rdr6. Standard deviations (n≥20) are shown. c, Immunoblots showing degradation of FGFP-CRY2 or FGFP-CRY2P532L in 7-day-old etiolated transgenic seedlings
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 92, 852, 472]]<|/det|> +irradiated with blue light (100 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) ) for the indicated time. Immunoblots were probed with anti- CRY2 and anti- HSP90 antibodies. HSP90 is a loading control. d, Quantitative results of CRY2 degradation from immunoblots in (c), analyzed with one phase decay of nonlinear regression. CRY2 (B/D) = [CRY2/HSP90]blue / [CRY2/HSP90]dark. \(\mathrm{T}_{1 / 2}\) indicates the time required for 50% degradation. e- f, Co- IP assays showing the lack of CRY2 \(^{P532L}\) - COP1 interaction (e) and the the blue light- and phosphorylation- dependent CRY2 \(^{P532L}\) - LRB2 interaction (f) in heterologous HEK293T cells. HEK293T cells co- transfected with indicated plasmids were either kept in the dark (-) or treated with 100 \(\mu \mathrm{mol} \mathrm{m}^{- 2} \mathrm{s}^{- 1}\) blue light for 2 hours (+). Immunoprecipitations were performed with Flag- conjugated beads. CRY2 \(^{D387A}\) is a negative control. Anti- Flag antibody, anti- Myc antibody or anti- HA antibody were used for detecting respective tagged proteins. Arrows indicate phosphorylated CRY2. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[68, 92, 780, 530]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 548, 115, 567]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 590, 540, 610]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 45, 765, 547]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 569, 117, 587]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[44, 610, 543, 629]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<|ref|>image<|/ref|><|det|>[[63, 630, 763, 901]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 928, 117, 946]]<|/det|> +
Figure 3
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 540, 64]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<|ref|>image<|/ref|><|det|>[[65, 68, 750, 395]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 419, 117, 439]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 460, 540, 480]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 52, 688, 592]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 612, 117, 631]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 655, 540, 675]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 45, 683, 520]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 550, 116, 568]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[42, 591, 540, 610]]<|/det|> +[Please see the manuscript file to view the figure caption.] + +<|ref|>sub_title<|/ref|><|det|>[[44, 636, 310, 662]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 686, 764, 705]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 724, 187, 742]]<|/det|> +- LRBsup.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/images_list.json b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..4f98f635ae56d94b9f54db01bd4702650a6e3fd8 --- /dev/null +++ b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/images_list.json @@ -0,0 +1,78 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 100, + 12, + 870, + 800 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "b", + "footnote": [], + "bbox": [ + [ + 199, + 20, + 728, + 199 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 104, + 50, + 800, + 936 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 0, + 0, + 999, + 999 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "3. Chr2 initiation", + "footnote": [], + "bbox": [], + "page_idx": 27 + } +] \ No newline at end of file diff --git a/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54.mmd b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54.mmd new file mode 100644 index 0000000000000000000000000000000000000000..ed918f6d0d05c6d404d12539720723192b5deaab --- /dev/null +++ b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54.mmd @@ -0,0 +1,362 @@ + +# Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae + +Marie-Eve Val + +marie-eve.kennedy-val@pasteur.fr + +Institut Pasteur, Université Paris Cité, CNRS UMR3525 https://orcid.org/0000-0001-7097-9072 + +Theophile Niault +Institut Pasteur + +Ariel Talavera +Université libre de Bruxelles + +Eric LeCam +Institut Gustave Roussy, Université Paris Saclay, CNRS UMR9019 + +Ole Skovgaard +Roskilde University https://orcid.org/0000-0002-4860-1847 + +Florian Fournes +DMF, UNIL + +Léa Wagner +Institut Pasteur, Université Paris Cité, CNRS UMR3525 + +Hedvig Tamman +Université libre de Bruxelles https://orcid.org/0000-0003-4453-7814 + +Andrew Thompson +SOLEIL Synchrotron + +Abel Garcia Pino +Université libre de Bruxelles + +Didier Mazel + +Institut Pasteur, Unité Plasticité du Génome Bactérien, UMR3525, CNRS https://orcid.org/0000-0001-6482-6002 + +Article + +Keywords: + +Posted Date: February 12th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs-3878828/v1 + +<--- Page Split ---> + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on January 8th, 2025. See the published version at https://doi.org/10.1038/s41467-024-55598-9. + +<--- Page Split ---> + +# Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae + +Théophile Niault1,2, Ariel Talavera3, Eric Le Cam4, Ole Skovgaard5, Florian Fournes1, Léa Wagner1, Hedvig Tamman3, Andrew Thompson6, Abel Garcia Pino3, Didier Mazel1 and Marie-Eve Val1 + +1 Institut Pasteur, Université Paris Cité, CNRS UMR3525, Unité Plasticité du Génome Bactérien, Département Génomes et Génétique, Paris 75015, France. + +2 Sorbonne Université, Collège Doctoral, Paris 75005, France + +3 Cellular and Molecular Microbiology, Faculté des Sciences, Université libre de Bruxelles (ULB), Boulevard du Triomphe, Brussels, Belgium. + +4 Genome Integrity and Cancer UMR 9019 CNRS, Université Paris Saclay, Gustave Roussy, Villejuif 94805, France + +5 Department of Science, Systems and Models, Roskilde University, Roskilde DK-4000, Denmark. + +6 SOLEIL Synchrotron, Saint-Aubin - BP48, Gif sur Yvette, France + +© Corresponding authors + +<--- Page Split ---> + +## ABSTRACT + +Over \(10\%\) of bacteria have expanded their genomes through the domestication of megalasmidis, which subsequently evolved into secondary chromosomes encoding core functions. A fundamental challenge of this genomic expansion is coordinating the replication of multiple replicons within a single cell cycle. In Vibrio cholerae, the replication of the secondary chromosome (Chr2) is intricately linked to the replication of the primary chromosome (Chr1) via a unique checkpoint sequence, crtS. This sequence binds to the initiator of Chr2, RctB. While crtS replication on Chr1 triggers Chr2 replication, the specific molecular dynamics remain elusive. To investigate this, we conducted a comprehensive genome- wide analysis of RctB binding patterns in V. cholerae across various cell cycle stages. Our findings show that RctB primarily binds to sites inhibiting replication initiation at the Chr2 origin (ori2). This inhibitory effect is counteracted when crtS replicates on Chr1, causing a shift in RctB binding to sites that activate replication at ori2. Structural analyses support the formation of diverse oligomeric states of RctB, coupled to the allosteric effect of DNA, which determine ori2 accessibility. We propose a synchronization model where, upon replication, crtS locally destabilizes the RctB inhibition complex, releasing the Chr2 replication origin. + +<--- Page Split ---> + +2 3 Replication initiation is a crucial step in the bacterial life cycle, subject to complex regulatory controls to 4 adapt to fluctuating growth conditions 1,2. This complexity increases further for bacteria with 5 multipartite genomes, which require an additional layer of control to replicate multiple chromosomes 6 simultaneously 3. Although substantial evidence across a variety of bacterial species supports the idea 7 that the replication of multiple chromosomes must be coordinated 4- 9, Vibrio cholerae is the only species 8 where a specific mechanism synchronizing the replication of two chromosomes has been identified 10- 12. In this pathogen, a replication checkpoint sequence, called crtS, ensures the synchronous replication 10 and termination of the primary (Chr1) and secondary chromosome (Chr2) within a single replication 11 cycle 12. The objective of the present study is to elucidate the molecular mechanisms by which crtS 12 orchestrates this synchronization process. 13 V. cholerae Chr2 originates from an iteron-type plasmid, evident from its replication origin (ori2) and 14 initiator (RctB) 13. RctB is divided into four structural regions (I - IV) 14 (Supplementary Fig. 1a). Its two 15 central domains (II, III) are structurally similar to the Rep iteron plasmid initiators domains (WH1, WH2) 16 while domains (I, IV) are unique to RctB 14. Domains, I, II and III, interact with DNA via Helix-Turn-Helix 17 (HTH) motifs and are essential for ori2 replication 14. The C-terminal domain (IV) is required to down- 18 regulate Chr2 initiation and to coordinate Chr1 and Chr2 replication 15,16. RctB recognizes three types of 19 double-stranded DNA sites: iterons, 29/39m and crtS. To promote replication initiation at ori2, RctB 20 binds to an array of six regularly spaced iterons (six 12-bp repeats), which prompts DNA unwinding at 21 an adjacent AT-rich region known as the DNA Unwinding Element (DUE) (Supplementary Fig. 1b) 17. At 22 this location, RctB engages with single strand of the open DUE on six regularly spaced direct repeats (5'- 23 ATCA) 18. A nucleoprotein complex, composed of RctB, IHF and DnaA, enables the recruitment of the 24 replicative helicase DnaB, similar to what occurs in iteron plasmids 19. Iterons sites each contain a GATC 25 motif which is specifically recognized and methylated by the Dam methylase. Iteron sites must be 26 methylated on both DNA strand for RctB to bind 20. RctB recognizes another type of sites, named for 27 their size as 29-mer or 39-mer (29/39m), which differ from iterons in both sequence and function 28 (Supplementary Fig. 1b). The 29/39m sites play a regulatory role in inhibiting ori2 initiation 21. Two 39m 29 sites and one 29m site flanking the minimal ori2 (containing the six iteron array and the DUE) 30 participates in RctB-mediated iterons handcuffing (pairing via initiator bridges) 21,22. The 29m site is also 31 found in the rctB gene promoter and contributes to rctB self-repression 22. Additionally, RctB binds to a 32 unique 62 bp sequence on Chr1, named crtS 11,12,23. This sequence is crucial for synchronizing the 33 replication of Chr1 and Chr2 12. The replication of crtS triggers the initiation of Chr2, and its position 34 relative to Chr1 origin (ori1) sets the replication timing for Chr2 12. If crtS is deleted, V. cholerae 35 undergoes lethal loss of Chr2 due to under-initiation at ori2 12. Although the action of crtS has been + +<--- Page Split ---> + +1 indirectly linked to the inhibitory function of the 29/39m sites 16, the exact molecular dynamics that 2 guide the coordination of Chr1 and Chr2 replication by crtS remains largely undefined. 3 In this study, we conducted a comprehensive, genome- wide analysis of RctB binding, unveiling key 4 elements of the crtS- triggered mechanism for Chr2 replication initiation at various stages of the V. 5 cholera cell cycle. We identified novel RctB binding sites and established a dynamic pattern of RctB 6 binding throughout the cell cycle. Our findings reveal that RctB predominantly binds to the 29/39m sites 7 within ori2, effectively preventing the initiation of Chr2 replication for a large portion of the cell cycle. 8 Using transmission electron microscopy (TEM), we observed that RctB forms large nucleoprotein 9 complexes at ori2, linking the 29/39m sites together. We show that RctB Domain IV can mediate 10 alternative dimerization interfaces, suggesting their role in the assembly of oligomeric bridging 11 structures when bound to 29/39m sites. Following crtS replication, we observed a marked shift in RctB 12 binding preferences at ori2, transitioning towards the iterons and the DUE, thereby enabling Chr2 13 replication initiation. Using an integrative approach combining ChIP- seq, live cell microscopy and 14 structural studies, we propose a model of the firing of Chr2 replication in which one crtS duplication 15 event directly triggers the activation of one ori2 by destabilizing an inhibition complex. + +## RESULTS + +## Genome-wide binding profile of RctB in V. cholera + +We conducted a ChIP- seq of RctB in V. cholera to investigate its genome- wide binding profile within an exponentially growing population. In comparison to a previous ChIP- on- chip study 11, we found that RctB binds to more regions across both chromosomes. These include 8 regions on Chr1 and 6 regions on Chr2, one of which encompasses ori2 (Fig. 1a, Supplementary Table 1). Most of these regions contain either iterons- like sites with a 5'- TGATCA inverted repeat or 29/39m- like sites with a 5'- TTACGG motif, as revealed by MEME analysis 24 (Fig. 1b). We also identified three regions that do not share any sequence similarity with these motifs (Supplementary Table 1), including crtS in the intergenic region between VC0764 and VC0765 on Chr1 (Fig. 1a, Supplementary Fig. 2) 23. Given that our ChIP- seq was executed on non- synchronized, exponentially growing bacterial cultures, the heights of the peaks offer valuable insights into the frequency with which RctB binds at various sites throughout the cell cycle. To assess the impact of the newly discovered RctB binding sites on Chr2 replication, we used a plasmid- based reporter system, pORI2, that exclusively relies on ori2 for replication 23. Each RctB binding regions (spanning 250bp on each side and centered around the peak) was inserted into the lacZ gene of Escherichia coli chromosome, and pORI2 copy number was measured using quantitative digital PCR (dPCR). Our results show that crtS stands out as the only strong enhancer of ori2 initiation (Fig. 1c). Mutations in three of the most prominent binding sites (VC0643, VC01643, and VC1042- VC1043) had + +<--- Page Split ---> + +1 minimal effects on Chr2 replication in V. cholera; although the 39m- like site within VC0643 did display 2 a weak, yet statistically significant, negative effect on Chr2 copy number (Fig. 1d). Given RctB's ability to 3 repress its own transcription by binding to its promoter 22, we explored whether RctB could repress 4 other genes. We monitored the transcription levels of VC1042 and VC1803 genes in V. cholera, both 5 of which have an RctB binding site within their promoters. RT- dPCR analyses revealed minimal to no 6 differences in the expression of these genes, regardless of RctB presence (Fig. 1e). Although the RctB 7 binding in the VC1803 promoter exhibited a weak, yet statistically significant, positive regulatory effect 8 on VC1803 transcription. VC1803 is a component of Vibrio pathogenicity island 2 (VPI- 2) and contains a 9 Cro/C1 repressor- like HTH domain profile commonly found in transcriptional repressors of temperate 10 bacteriophages, suggesting potential gene regulation by RctB within the pathogenicity island. In 11 conclusion, while crtS plays a pivotal role in regulating Chr2 replication, the functional relevance of the 12 newly discovered RctB binding sites on Chr1 remains uncertain. RctB appears to regulate its own 13 transcription exclusively, in contrast to the initiator DnaA, which acts more broadly as a general 14 transcription factor 25. + +## RctB binds predominantly to 29/39m inhibitory sites within ori2 + +Upon closer examination of the ori2 region, we observed that RctB ChIP signal was about 10- fold higher than at other chromosomal loci (Supplementary Fig. 2). RctB showed a strong preference for binding to the 29/39m sites (Fig. 2a). This predominant occupancy of RctB at inhibitory sites suggests that initiation of Chr2 replication is hindered for most of the cell cycle. To understand RctB binding dynamics in non- replicating cells, we performed a ChIP- seq of RctB during stationary phase. The RctB binding pattern at ori2 mirrored that observed during the exponential phase (Supplementary Fig. 3), suggesting that RctB binding to 29/39m inhibits initiation at ori2 during both replicative and non- replicative phases. In both conditions, the signal at the 29m site was approximately 3- fold higher than that at the two 39m sites. We hypothesized that this discrepancy might arise from the ParB2 protein, essential for Chr2 segregation, competing with RctB for binding to both 39m sites, as previously reported 26. Confirming our suspicion, our ChIP- seq of ParB2 on ori2 (Supplementary Fig. 4) revealed that ParB2 does indeed compete with RctB at both 39m sites and not at the 29m site. To further investigate the binding requirements of RctB at ori2, we performed a ChIP- seq analysis of RctB in a strain deleted for dam methylase (Δdam#4). RctB cannot bind to iterons in this mutant because they are not methylated, and so there is no initiation at ori2 20,27. The Δdam#4 mutant is viable due to the integration of Chr2 into Chr1, enabling its replication from ori1 as the DnaA initiator does not require Dam methylation 27 (Supplementary Fig. 5a). In the Δdam#4 mutant, the RctB binding pattern at ori2 was identical to that of the wild- type strain (Supplementary Fig. 5b). This result demonstrates that RctB molecules can form + +<--- Page Split ---> + +a complex at ori2 strictly through their interaction with 29/39m sites, without any engagement with iterons. It also confirms that RctB binding to crtS is independent of Dam as shown in 23. + +## RctB binding to 29/39m sites generates DNA loops in ori2 and prevents its binding to iterons + +To directly visualize the complex formed by RctB bound to 29/39m on ori2, we used positive- staining transmission electron microscopy (TEM) 28. An unmethylated DNA fragment containing ori2 was chosen to allow RctB to bind exclusively to the 29/39m sites. Our observations revealed that RctB formed large nucleoprotein complexes bridging the 29m and 39m sites together, causing looping within the origin (Fig. 2b, white arrows). Such loops in ori2 could create a steric hindrance preventing the initiation of replication by interfering with RctB binding to the iteron array and the proper unwinding of the DNA strands. To test this hypothesis, we performed a ChIP- seq of RctB in a mutant with the \(C_{21} > A\) mutation at the 29m site \((29m^{*}(C / A))\) known to disrupt RctB binding 16. The binding pattern of RctB at ori2 was drastically altered (Fig. 2c). RctB still bound to both 39m sites but not to the 29m. Additionally, we observed a clear binding of RctB to the array of six iterons and the adjacent DUE. Given that initiation of ori2 replication requires RctB interaction with both the iteron array and the DUE 18, our observations suggest that preventing RctB binding to 29m effectively alleviates the constraint on ori2 initiation. + +## RctBIV harbors two dimerization interfaces + +The C- terminal region of RctB, named domain IV (RctBIV), is critical to downregulate Chr2 copy number through its involvement directly or indirectly in handcuffing activities and binding to 29/39m 15,16,29. It is also crucial for crtS- mediated control of Chr2 replication 16. In vitro, RctB is able to self- oligomerize when bound to DNA through interactions mediated by domain IV 16,23. Thus, we hypothesized that the coordination of Chr1 and Chr2 replication likely depends on the ability of RctB to oligomerize on ori2 via domain IV. While RctB has remained refining to structural studies, likely due to the presence of intrinsically unstructured regions in the protein, the structure of domain I and domains II-III have been solved 14,30. Therefore, we used a truncated version of RctB to determine the structure of domain IV 14,30 and generated the stable fragment RctBIV (amino acids 532 to 658) that could be used in structural biology. The structure of RctBIV reveals that it is organized into two structural subdomains connected by an extended linker or hinge region (Fig. 3a). The N- terminal subdomain consists of a 3- stranded \(\beta\) - sheet stacking two \(\alpha\) - helices and the C- terminal subdomain has 4 \(\alpha\) - helices in a bundle (Fig. 3b). In the crystal, RctBIV retains a dimeric association as predicted by Size Exclusion Chromatography (SEC) (Supplementary Fig. 6). This dimerization is primarily mediated by a \(\beta\) - sheet ( \(\beta\) 3) spanning residue 545 to 549 and involves an interface of 392.4 Å2 (Fig. 3a- b, primary interface). The crystal lattice contacts of the C- terminal subdomain suggest the presence of a secondary interaction interface involving 315.4 Å2, which could mediate further oligomerization via this region (Fig. 3c, secondary interface). A DALI 31 + +<--- Page Split ---> + +search confirmed the C- terminal subdomain as a protein recognition domain capable of forming transient homodimeric interfaces, with Z- scores between 3.7 and 5.4. This secondary interface closely resembles a representative structural homologue detected by DALI (a putative transcription regulator from Neisseria gonorrhoeae) \(^{32}\) . These findings suggest that RctB \(^{IV}\) may associate in multiple ways and potentially form higher- order oligomeric structures when recruited to DNA. The presence of multiple intermolecular interfaces in RctB \(^{IV}\) inferred from the X- ray structure, together with class averages of RctB obtained by cryo electron microscopy (Fig. 3d) are consistent with the presence of dimeric and tetrameric forms RctB. These findings suggest that RctB \(^{IV}\) may associate in multiple ways and potentially form higher- order oligomeric structures when recruited to DNA. + +## Domain IV dimerization interfaces mediate RctB oligomerization + +To further characterize these interfaces, we used the bacterial two- hybrid assay in E. coli \(^{33}\) . We performed a structure- guided point mutation analysis to probe the potential interfaces involved in the oligomerization of RctB \(^{IV}\) . Given that RctB has a separate dimerization interface in domain II \(^{14}\) , we substituted residue D314 with a proline (D314P) to prevent domain II- mediated interactions that would obscure RctB \(^{IV}\) - mediated interactions (as done in \(^{16}\) ). In Fig. 3e, blue colonies (D314P/D314P) indicated that RctB could interact with other RctB through another domain, as already reported to be via domain IV \(^{16}\) . Our present results showed that substitutions in the \(\beta 3\) strand of RctB \(^{IV}\) caused the formation of white colonies (D314P/D314P- I546G- I546G), indicating that \(\beta 3\) is important for RctB interactions while substitutions that destabilize the \(\alpha 1\) helix appear to have little impact on dimerization as evidenced by the formation of blue colonies (D314P/D314P- A565P). Deletion of the linker connecting the two subdomains (591- 596), gives blue colonies (D314P/D314P- \(\Delta_{591 - 596}\) ) but to a lesser extent than the A565P substitution, suggesting that RctB self- association may be sensitive to local structural rearrangements between the two subdomains. The I625P and L651P substitutions in two different \(\alpha\) - helices (α3 and α6) of the C- terminal subdomain also abrogate RctB interactions confirming our previous results \(^{16}\) . Thus, RctB \(^{IV}\) can form homodimers via two different regions: at a primary interface formed between the \(\beta 3\) sheets and at the secondary interface between \(\alpha\) - helices. Collectively these results suggest that RctB \(^{IV}\) could be an anchor not only for dimerization but also for the formation of transient oligomeric structures when bound to DNA or even to bridge two chromosomal regions. + +## RctB oligomerization is required for the binding to inhibitory sites at ori2 + +To better understand the role of RctB \(^{IV}\) - mediated interactions on the genome- wide in vivo binding of RctB, we performed ChIP- seq of RctB- L651P (Fig. 3f). These analyses revealed a completely different binding profile to ori2 compared with RctB wt (Fig. 2a). RctB- L651P binding to ori2 is fully shifted to the iterons array and the DUE and is absent from the 29/39m sites. However, we observed the same binding + +<--- Page Split ---> + +1 profile at crtS between RctB- L65sP and RctB wt (Supplementary Fig. 7a). This suggests that, although not 2 involved in DNA binding, RctBv likely mediates the differential recognition of 29/39m inhibitory sites 3 relative to iterons or crtS, probably via the formation of alternative oligomeric structures. As a control, 4 we performed ChIP- seq of RctB- L65sP in a \(\Delta d a m\) context. As expected, in the absence of Dam 5 methylation, no binding of RctB- L65sP to ori2 was observed, although RctB still bound to crtS 6 (Supplementary Fig. 7b). Based on these results, we propose that intermolecular domain IV mediated 7 RctB interactions are crucial for the inhibition of replication at ori2 through the bridging of 29/39m sites. + +## crtS exclusively affects the RctB inhibition complex at ori2 + +The triggering mechanism of ori2 replication by crtS remains poorly understood. In a \(\Delta crtS\) strain, Chr2 is under- initiated, resulting in cell filamentation and lower Chr2 copy number than Chr1 12. To better understand the consequences of the loss of crtS, we conducted a ChIP- seq analysis of RctB in that strain (Fig. 4a). In this background, RctB also predominantly binds to the 29/39m inhibitory sites at ori2, but the ChIP signal was approximately 10- fold higher, indicating a much stronger inhibition of replication in the absence of crtS. Based on our new findings, we propose that crtS could trigger the initiation of Chr2 replication by destabilizing the nucleoprotein complex that impedes ori2 initiation. When RctB carried the L651P mutation, there was no difference in its binding profile at ori2, irrespective of the presence or absence of crtS, indicating that crtS solely impacts the binding of RctB to the 29/39m (Supplementary Fig. 7a). Given that RctB binds to other 39m- like sites outside of ori2 (Supplementary Table 1), we hypothesized that the absence of crtS could enhance the binding of RctB to these sites as well. However, we observed no increase in RctB binding to the other 39m in the \(\Delta crtS\) mutant (Supplementary Fig. 8). This suggests that crtS has an exclusive activity on ori2. To investigate this further, we examined RctB binding on crtS- containing DNA fragments by TEM. The binding of RctB on crtS was restricted to the site and induced a kink in the DNA (Supplementary Fig. 9a). When we mixed DNA substrates containing unmethylated ori2 and crtS with RctB, we detected sporadic RctB- mediated contacts between ori2 and crtS (Supplementary Fig. 9b), suggesting the possibility of direct interactions between crtS and the 29/39m inhibitory sites in ori2. These findings align with the high frequency of contacts between chromosomal regions proximal to crtS and ori2, observed in vivo by chromosome conformation capture 12. + +## The coupled replication of crtS and activation of ori2 is stoichiometric + +Past studies 12,23,34 demonstrated that introducing an additional crtS copy into Chr1 increases Chr2 copy number. To delve deeper into this relationship, we focused on strains with two crtS sites. First, we inserted one or two copies of crtS into the attTn7 site (located near VC0487 on Chr1) and deleted the endogenous crtS site. As anticipated, the strain with two crtS exhibited a higher Chr2 copy number + +<--- Page Split ---> + +compared to the strain with a single crtS when measured on stationary phase culture (1.7 vs. 1) (Supplementary Fig. 10a). However, when comparing the ChIP- seq profiles of RctB at ori2 between these strains, we did not observe any significant differences; in both cases, RctB mainly binds to the 29/39m sites (Supplementary Fig. 10b). This observation indicates that having two copies of crtS does not alter the binding of RctB to activate replication at ori2. Consequently, any activation of ori2 after crtS replication probably occurs transiently. To gain a temporal perspective, we used fluorescence microscopy to observe a V. cholerae mutant with two distant crtS sites 12. This strain contains the endogenous site (crtSwt) and a second site added next to ori1 (crtSori1)—thus delaying the replication timing of both crtS sites (Fig. 4b- d). We followed the dynamics of both crtS and ori2 loci using three different DNA binding protein fluorescent proteins and their cognate sites inserted in the genome: ParBri- CFP (parSpri site near crtSori1), ParBwri1- yGFP (parSwri1 site near crtSwi), LacI- RFP (lacO array near ori2) 35. Compared to the wild type, which contains only one or two foci of ori2, the two- crtS containing mutant has up to four ori2 foci (Fig. 4b). We tracked the longitudinal position of ori2 foci relative to the Chr1 sites in growing cells (Fig. 4d vs. Fig. 4c) and observed that the two- crtS containing cells are born with two ori2 foci. As cell size increases, we observed a sequential increase in the number of ori2 foci, going from two to three and then to four ori2. These results suggest that the early replication of crtSori1 triggers the replication of only one of the two ori2. The replication of the second site crtSwt activates the initiation of the second ori2. + +We took a closer look at the stoichiometry by Marker Frequency Analysis (MFA) on three exponentially growing strains with different crtS contents: the wild type (crtSwt), one mutant with crtS relocated at ori1 (crtSori1), and the two- crtS containing mutant (crtSwt, crtSori1). First, we confirm that the location of crtS on Chr1 controls the time of initiation of Chr2 and thereby the copy number of Chr2, as in crtSwt it is 0.484 and in crtSori1 it is 0.601 (Fig. 4e, left and center) as observed before 12. Then, the strain with two crtS sites (crtSwt and crtSori1) have an ori2 copy number of 0.900 approximating the sum of ori2 copy number in the single crtS strains: crtSwt and crtSori1 (0.484 + 0.601). These findings align with a model in which one crtS duplication event specifically activates one single ori2. This model is illustrated in Supplementary movie 1 (crtSwt), movie 2 (crtSori1) and movie 3 (crtSwt, crtSori1). + +## crtS replication induces a temporal shift in RctB binding at ori2 + +Based on these results, we hypothesized that to activate replication at ori2, RctB should interact with the iterons in a transient way, while remaining bound to the inhibitory 29/39m sites for most of the cell cycle. Our previous approach did not allow us to capture short- lived events. Therefore, we implemented the ChIP- seq of RctB in a replication- synchronized population to monitor the RctB binding profile over the course of the cell cycle. Similar to E. coli, an exponentially growing population of V. cholerae can be synchronized using serine hydroxamate (SHX, a serine analogue that chemically stimulates the stringent + +<--- Page Split ---> + +response) 36- 38. During treatment with SHX, replication forks can still progress but the initiation of new rounds of replication is hindered. Transfer of these cells to a fresh medium (without SHX) allows a theoretical synchronous restart of replication. Indeed, previous work showed that after SHX treatment, V. choleraer Chr1 re- initiates first and then Chr2 37 presumably due to control by crtS 12. In our hands, the addition of SHX to 1.5mg/mL final concentration to an exponentially growing culture of V. choleraer caused an instantaneous growth arrest (Fig. 5a). To ensure completion of ongoing replication, we measured the ratio of loci at the origin and terminus of Chr1 (ori1/ter1) by dPCR (Fig. 5b). After 120min of SHX treatment, all cells successfully completed replication with a ratio of ori1/ter1 = 1 meaning that Chr1 was fully replicated. Once SHX was removed, the cells restart to replicate within a time window of 20- 30 minutes, similar to what was observed 38. While synchronization by SHX is not as perfect in V. choleraer as it is in other bacteria, it provided a reasonable approximation to a synchronized population. Thus, we decided to use it to perform ChIP- seq of RctB at different time points after removal of SHX (15, 30, 45, 60 minutes) to follow the binding cycle of RctB. + +Using reads from the input file (control without immunoprecipitation), we performed MFA to track replication fork progression on both chromosomes (Fig. 5c- 5f, top panel). Within the 15 minutes after removal of SHX (Fig. 5c), replication has not restarted, as shown by the flat MFA profile of Chr1 and Chr2, RctB predominantly binds to 29/39m on ori2. After 30 minutes (Fig. 5d), Chr1 starts replicating and at 45 minutes (Fig. 5e) the crtS site seems to be just replicated (the Ter region of Chr1 still appears not replicated), while Chr2 replication has not yet started. At both time points, RctB ChIP- seq patterns at ori2 is nearly unchanged (Fig. 5d- 5e). At 60 minutes (Fig. 5f), Chr2 is finally replicating. On the ChIP- seq profile, two additional peaks appear: one on the six- iteron array and one on the DUE (Fig. 5f, black arrows). Here, we were able to capture a release of RctB from the inhibitory sites (29/39m) towards the activating sites (iteron) and towards the DUE. This temporal shift in the binding pattern of RctB from inhibitory to activating sites on ori2 underscored the opening of ori2 that resulted in the initiation of Chr2 replication. As expected, ori2 activation follows crtS replication on Chr1. We did not observe fluctuations in the binding of RctB at crtS before or after the passage of the replication fork, nor during stationary versus exponential phase of growth (Supplementary Fig. 3 & 11). Thus, RctB binds similarly to crtS throughout the cell cycle. + +## DISCUSSION + +Our work provides cell- cycle- integrated insights into the synchronized replication of Vibrio's two chromosomes. We identified an inhibition complex at ori2 that prevents replication initiation for a significant portion of the cell cycle. We also highlighted the pivotal role of crtS in temporarily disrupting this complex, enabling Chr2 replication. Based on the binding dynamics of RctB throughout the cell cycle, we propose a model for the regulated control of Chr2 replication. In this model, ori2 is inherently + +<--- Page Split ---> + +1 bistable, with the replication of crtS acting as a decisive ON switch. Before crtS replication, ori2 is sequestered by an inhibition complex, with RctB bridging the 29/39m sites, and inducing DNA loops that effectively keep ori2 in an OFF state. The replication of crtS destabilizes this complex, causing transient exposure of ori2 to an ON state compatible with replication initiation. This shift promotes the recruitment of RctB to the six iterons, the DUE, and ultimately, the initiation of Chr2 replication. A crucial feature of the stabilization of the OFF state is the oligomerization properties of RctB via the domain IV which defines the formation of the 29/39m bridges and preclude Chr2 replication. In a broader context, analogous types of nucleoprotein complexes have been observed intermolecularly in the regulation of iteron plasmid copy number, involving the initiators RepApp51039 and PiRGK40. In our model, we propose that initiation silencing occurs through intramolecular DNA looping at the origin. This is akin to the mechanism in the F plasmid, where RepE initiators establish a bridge between the iterons of the negative control element (incC) and those of the origin (oriF) that further inhibit the formation of an open complex41. Vibrio distinguishes itself by having tailored specific 29/39m sites to serve this regulatory role. + +The replication of crtS triggers a shift in RctB binding, favoring its binding with the iterons and DUE, conditions that are compatible with ori2 initiation. Further mechanistic research is needed to explore the molecular interactions between RctB and DNA, as well as how crtS disrupts the inhibition complex. However, our model is consistent with the bistable state of ori2 being influenced by crtS replication, the structural properties of RctB and the dramatic changes in the DNA structure and organization induced by RctB. Given that crtS specifically influences RctB binding at ori2—without affecting other genomic sites—and considering the stoichiometric relationship between crtS replication and ori2 activation, our data strongly suggest a direct physical interaction that leads to the activation of ori2 by crtS. This is consistent with our previous observations of frequent inter-chromosomal interactions proximal to crtS and ori212. Moreover, we observed that the regulatory control exerted by crtS over ori2 is lost when a single 29m or 39m site is mutated16. This synergistic effect implies that crtS likely disrupts the 29/39m bridging, rather than affecting each site individually. We propose that each newly duplicated crtS site have RctB bound to it. Both crtS-RctB complexes can each interact with one 29/39m-RctB complex. These interactions destabilize the DNA loop which inhibit ori2 initiation. Once ori2 is exposed, the iteron array and DUE become accessible, so that free RctB molecules can bind and initiate replication (Fig. 6). In contrast to plasmids, where initiators bind solely to iterons, RctB capacity to interact with multiple sites offers a more refined control over replication initiation. This multifaceted binding ability is determined by factors such as methylation-dependent binding to iterons20, domain IV-oligomerization-dependent binding to 29/39m (this study), and Lrp-enhanced binding to crtS42,43. All these factors introduce dynamic binding behaviors throughout the cell cycle, ensuring that Chr2 replication is finely tuned and synchronized with other cellular processes. + +<--- Page Split ---> + +## METHODS + +## Strains and growth conditions + +Vibrio cholerae N16961 El tor (7PET) is the strain studied here. As a heterologous system for dPCR experiments Escherichia coli K12 MG1655 was also used. Different E.coli strains were used for cloning DH5α, TOP10, P3813 and β3914 \(^{44}\) for conjugation. LB (Luria Broth) was used as standard rich medium for both E. coli and V. cholerae. + +## Strains and plasmids constructions + +Both V. cholerae and E. coli strains were modified using conditionally replicative suicide vectors as described in \(^{45}\) . These vectors were generated by Gibson assembly \(^{46}\) , and in cases requiring specific point mutations, they were introduced via PCR using oligonucleotides containing the desired mutations at the 5' end. For ChIP-seq experiment (on RctB and on ParB2), we replaced the wild-type alleles on the chromosome with a C-terminal 3xFLAG through allelic exchange. For the dPCR experiments, we introduced RctB secondary binding sites into the lacZ gene of MG1655, we took an extending of 250 base pairs on each side of the peak detected from ChIP-seq analysis. Subsequently, these modified strains were transformed with a pORI2 plasmid, following the procedure outlined in \(^{23}\) . Plasmids and strains used in this study are listed in Supplementary Tables 2 and 3. + +## High Resolution ChIP-seq + +ChIP-seq were performed as described in \(^{47}\) in V. cholerae expressing protein- 3xFLAG under its native promoter. We confirmed that 3xFLAG did not impair protein function by dPCR measurement of Chr2 copy number. Cells were grown in 100mL Mueller Hinton (MH) media with shaking at \(37^{\circ}C\) until they reach mid- log growth phase (OD \(_{600}\) 0.5). Then cells were fixed by crosslinking with \(1\%\) formaldehyde for 30 min at room temperature under agitation, followed by quenching with 0.5M glycine for 15 min. Bacteria were harvested, washed two times in ice- cold Tris- Buffered saline (20 mM Tris/HCl pH 7.5, 150 mM NaCl) and pellets were stored at - 80°C. Pellets were lyzed in buffer containing lysozyme and protease inhibitor pill (cOmplete Protease Inhibitor, Roche). Chromatin was then fragmented by sonication (Covaris S220) for in milliTUBE (1ml with AFA Fiber). Correct DNA fragmentation centered around 300bp was confirmed using \(1.8\%\) agarose gel electrophoresis. Following sonication, protein- DNA complexes were immunoprecipitated with anti- FLAG magnetic beads (M8823 Sigma). Correct immunoprecipitation was confirmed by western blot using anti- FLAG antibodies (F7425 Sigma). Once recovered, immunoprecipitated DNA was de- crosslink at \(65^{\circ}C\) overnight. Then DNA was blunt- ended, A- tailed, ligated to adaptors, amplified by PCR (13 cycles) and purified using TruSeq ChIP Library Preparation Kit (illumina) and AMPureXP beads (Beckman Coulter). ChIP- seq libraries quality was + +<--- Page Split ---> + +measured by Agilent Bioanalyzer 2100 instrument using a DNA High Sensitivity chip. Finally, libraries were sequenced using a MiniSeq Mid Output Kit on a Miniseq sequencing machine (Illumina). Paired- end reads from two independent RctB ChIP- seq experiments were mapped to Vibrio cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) with Bowtie2 using galaxy. pasteur platform. Peak calling was performed using MACS2, where aligned reads were significantly enriched compared to a control (strain without FLAG). Identified peaks were checked by hand, using IGV genome browser to confirm correct peak assignment. MEME- ChIP (version 5.4.1) online browser was used for motif enrichment finding. For normalization, the IP and INPUT files were first adjusted to the same number of reads using R. Then the coverage value for each position in the IP file was divided by the average coverage value of a 2kb sliding window from the INPUT file. Data available at the European Nucleotide Archive (Supplementary Information). + +The genomic replication pattern of the different strains was examined using a computational approach based on Marker Frequency Analysis (MFA). The reads from the INPUT sample (not immunoprecipitated DNA) of the ChIP- seq data were mapped to V. cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) using BOWTIE2, the resultant mapping was saved in two separate Binary Alignment Map (BAM) files corresponding to the two chromosomes. Subsequently, a R script (available on demand) was used for data processing and visualization. The depth of coverage was extracted from the BAM files using the Rsamtools package. The genome was divided into 10 kbp bins, and the average coverage for each bin was computed. Coverage values were then normalized with respect to the average coverage of the termination (ter) region, which is located at the midpoint of the first chromosome (+/- 5kb). Coverage values more than three standard deviations from the mean were considered outliers and removed from the dataset. For visualization, a ggplot2- based plot was generated, showing the normalized coverage for each bin against the genomic position in Mbp. Linear regression analysis was performed separately on each half of the chromosomes, and the best- fit lines were superimposed on the plot (Supplementary Fig. 12). + +## Marker Frequency Analysis + +Grow conditions, DNA extraction, sequencing and analysis for MFA was done as described previously 12. Data available at the European Nucleotide Archive (Supplementary Information). + +## Digital PCR quantification (dPCR) + +Quantifications of (ori1 and ori2) in V. cholerae and (oriC and pORI2) in E. coli were performed as described in 16 using multiplex dPCR (Stilla Technologies). dPCR was performed directly on cell lysate: 1mL of overnight culture was washed twice with 1mL PBS, pellets were then frozen at - 20°C and resuspended in 200uL PBS, samples were then boiled for 10min. Samples were centrifuge for 10min, + +<--- Page Split ---> + +1 supernatant was recovered, and DNA content was measured using Qubit device (ThermoFisher). PCR reactions were performed with 0.1 ng of DNA using the PerfeCTa MultiPlex qPCR ToughMix (Quantabio) on a Sapphire chip (Stilla Technologies). Digital PCR was conducted on a Naica Geode and Image acquisition on the Naica Prism3 reader. Images were analyzed Crystal Miner software (Stilla Technologies). The dPCR run was performed using the following steps: droplet partition (40°C, atmospheric pressure AP to + 950mbar, 12 min), initial denaturation (95°C, +950 mbar, for 2 min), followed by 45 cycles at (95°C for 10 s and 60°C for 30 s), droplet release (down 25°C, down to AP, 33 min). + +## Digital PCR quantification of RNA (RT-dPCR) + +Quantification of RctB mRNA was performed in V. cholerae from exponentially growing cultures (OD600 0.5) using multiplex digital RT- dPCR (Stilla Technologies) 48. Primers and probes are listed in Supplementary Table 4. V. cholerae RNA was prepared as described in 49. PCR reactions were performed with 1 ng of RNA using the qScriptTM XLT One- Step RT- qPCR ToughMix® (Quantabio). RT- dPCR and data analysis were conducted as described above for dPCR. The RT- dPCR run was performed in the following steps: droplet partition (40°C, AP to +950 mbar, 12 min), cDNA synthesis (50°C, +950 mbar, 10 min), initial denaturation (95°C, +950 mbar, for 1 min), followed by 45 cycles at (95°C for 10 s and 60°C for 15 s), droplet release (down 25°C, down to AP, 33 min). Expression values were normalized to the expression of the housekeeping gene gyrA as described in 49. + +## SHX Vibrio cholerae synchronization + +V. cholerae expressing RctB-3xFLAG from its natural promoter was streaked from a -80°C stock onto an MH agar plate. An isolated colony was then picked and cultured overnight in M9 media containing 0.2% casamino acids and 0.4% glucose at 30°C. Cells from this overnight culture were subsequently diluted 100-fold in fresh media and grown in 40 mL of M9 media with 0.2% casamino acids and 0.4% glucose at 30°C. When the cells reached an OD600nm of 0.5, they were treated with SHX (DL-serine hydroxamate, Sigma) at a final concentration of 1.5 mg/mL for 120 minutes. This treatment prevented new rounds of initiation and allowed cells to complete ongoing replication. Cells stalled in G1 were then centrifuged at 6,000g for 15 minutes, and the pellets were resuspended in 40 mL of fresh M9 media before being returned to 30°C for growth. Cells were fixed with 1% formaldehyde at various time points: 15, 30, 45, and 60 minutes after the wash. Finally, samples were processed according to the ChIP-seq protocol. + +## Transmission electron microscopy + +The ori2 DNA probe (1864bp) containing the three 29/39mer sites was produced by PCR from pFF149 plasmid using MV109/MV226 primers (Supplementary Table 4). The crtS probe (944bp) was produced + +<--- Page Split ---> + +1 by PCR from a pUC18::crtS plasmid using MV577/MV485 primers. DNA fragments were then purified using AMPure XP beads (Beckman Coulter). RctB, Dnak and Dnal proteins were produced and purified like in PMC6212839. RctB was first treated with Dnak/J chaperones to enhance RctB activity. For this, RctB (1200nM) was incubated with Dnak (100nM) and Dnal (100nM) in 20uL of 10 mM Tris- HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer for 10min at 4°C. Then, for the DNA- protein complexes formation 2uL of treated RctB was mixed with 4nM of DNA probes (ori2 probe, crtS probe or both) in a 10 mM Tris- HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer (20uL final volume) for 10min at room temperature. For observation of DNA- protein complexes, 5 uL drop of the incubation was deposited on a 600- mesh copper grid previously covered with a thin carbon film and preactivated by glow discharge in the presence of amylamine (Sigma- Aldrich, France). The grids were rinsed and positively stained with 2% (w/v) aqueous uranyl acetate, carefully dried with filter paper, and observed in annular dark- field mode in lossless filtered imaging using a Zeiss 902 transmission electron microscope. Images were captured at 85,000× magnification with a Veleta MegaviewIII CCD camera and analyzed with iTEM software (Olympus Soft Imaging Solution). + +## Expression and purification of RctB and RctBIV for crystallization and biophysical studies + +RctB and were RctBIV cloned into a pET- 24b plasmid downstream an N- terminal 10xHis_SUMO tag. BL21 E.coli cells were transformed with these constructs for expression and purification. Transformed cells were first grown at 37°C until reaching an OD\~0.6 then expression was induced with 0.25 mM Isopropyl \(\beta\) - D- 1- thiogalactopyranoside (IPTG) overnight at 16°C. Afterwards cells were harvested by centrifugation at 3000g. + +For purification, the cell lysate supernatants were loaded onto a Ni- Sepharose column. In both cases, proteins were eluted using a step gradient of imidazole. The fractions containing the protein of interest were then loaded into a Citiva 60/160 Seperdex 200 size exclusion chromatography column, equilibrated in 50 mM Hepes pH 7.5, 500 mM NaCl, 500 mM KCl, 2 mM MgCl2 and 1 mM TCEP. The 10xHis- SUMO tag was removed by incubating overnight the protein with the protease Ulpl in a 1:100 molar ratio. The tag- free RctB and RctBIV were then isolated from the cleaved tag and Ulpl by loading the mixture into a Co- sepharose column and collecting the flow- through that contained the tag- free samples. For further studies RctB and RctBIV were extensively dialyzed against 50 mM Hepes pH 7.5, 200 mM NaCl, 2 mM MgCl2 and 1 mM TCEP. + +## RctBIV Analytical Size Exclusion Chromatography (SEC) + +Analytical SEC of RctBIV was carried out on a Superdex 200 10/300 column (Citiva) equilibrated in 50 mM Hepes, 200mM NaCl, 2 mM MgCl2 and 1 mM TCEP. The flow rate was 1 ml/min and the injection volume 0.5 ml. The calibration of the column was performed by running the Biorad gel filtration standards (Cat. + +<--- Page Split ---> + +1 No. 151- 1901), containing a mixture of bovine thyroglobulin (670 kDa), bovine \(\gamma\) - globulin (158 kDa), chicken ovalbumin (44 kDa), equine myoglobin (17 kDa) and vitamin B12 (1.35 kDa). 3 RctBIV crystallization and structure determination 5 The screening of crystallization conditions of RctBIV was carried out using the sitting- drop vapor- diffusion method with highly pure samples of RctBIV at a concentration of 8 mgmL- 1. The drops were set up in Swiss (MRC) 96- well two- drop UVP sitting- drop plates using the Mosquito HTS system (TTP Labtech). Drops of 0.1 μL protein and 0.1 μL precipitant solution were equilibrated to 80 μL precipitant solution in the reservoir. Commercially available screens PACT premier, LMB and SG1 (Molecular Dimensions) were used to test crystallization conditions. The condition resulting in protein crystals (PACTpr screen position E3: 200 mM NaI, 20 % PEG 3350) were repeated as 2 μL drops. Crystals were harvested using suitable cryo- protecting solutions (consisting of the mother liquor supplemented with 20% to 25% of glycerol) and vitrified in liquid \(\mathbb{N}_2\) for transport and storage before X- ray exposure. 14 X- ray diffraction data was collected at the SOLEIL synchrotron (Gif- sur- Yvette, Paris, France) on the Proxima 1 (PX1) beamlines using an Eiger- X 16M detector. Native crystals typically diffracted to 1.8 Å – 2.0 Å resolution. We screened a collection of heavy atoms for experimental phasing and observed that the combination of quick soaks (under 30s) with NaBr at 1.5M, and longer soaks (20 min to 30 min) in AgNO3 (10 mM) and GdCl (10 mM) provided sufficient isomorphous signal for experimental phasing. 19 For this, data were collected at the peak and inflexion point of the Br edge and Ag and Gd K- edge. After phase improvement with Pirate and density modification with Parrot, the maps were used initially used in Buccaneer for model building and then as starting point for the MR- Rosetta 50 suit which completed one third of the structure. Further cycles of automated model building with AutoBuild from the Phenix package 51 extended the model to 90%. After several iterations of manual building with Coot 52 and maximum likelihood refinement as implemented in Buster/TNT 53, the model was extended to cover all the residues (R/Rfree of 19.2%/26.7 %). Supplementary Table 5 details all the X- ray data collection and refinement statistics. 27 Single particle cryo- electron microscopy (cryo- EM) 29 For the cryo- EM experiments RctB was purified in 50 mM Hepes, 200mM NaCl, 2 mM MgCl2 and 1 mM TCEP. RctB (3 μL) was applied to gold Quantifoil grids (UltraFoil 1.2/1.3 Au 300) under 100 % relative humidity at 10 °C at 0.25 mg/ml. Two grids were blotted for 2 and 3 seconds, respectively, and plunged into liquid ethane using a Vitrobot IV (ThermoFisher Scientific). Particles were imaged from the aforementioned grids on a Glacios microscope (ThermoFisher Scientific) operating at 200 kV with a Falcon 4i direct electron detector. In total, 3800 micrographs, with 0.96 Å/pxl, were collected using a defocus range from -1 μm to -3 μm and total dose of 50 e/Å2. The micrographs movies were motion + +<--- Page Split ---> + +1 corrected using MotionCor2 \(^{54}\) , and contrast transfer function, CTF, were estimated using CTFFind- 4.1 \(^{55}\) . Particles were picked using Topaz \(^{56}\) and subjected to reference free 2D classification using cryoSPARC \(^{57}\) . Several rounds of 2D classifications rendered a total of 133108 particles. + +## Live-cell Fluorescence Microscopy + +The genes encoding for the CFP- ParBp1, yGFP- ParBpmT1 and LacI- RFP- T fluorescent DNA binding proteins were inserted into the V. cholera chromosome \(^{35,58,59}\) . Their cognate binding sites were inserted near ori1 (parSp1), near VC783 (parSpMT1), and near ori2 (lacO array). For microscopy observations, cultures were prepared following the procedures described in \(^{12}\) . Initially, strains were streaked onto MH plates from a - 80°C stock and then grown overnight in MH rich liquid media. The following day, cultures were diluted at a ratio of 1:1000 in M9 with 1% fructose and grown to exponential phase (OD600 = 0.2). A 2μL aliquot of this culture was spotted on an agar pad (M9, 1% fructose, 1% agarose), for microscopy observation. Cell imaging was performed using an Axio Observer 7 inverted videomicroscope (Zeiss). Analysis of the acquired snapshots was conducted using Fiji software with the MicrobeJ plugin \(^{60}\) . + +## Bacterial Two Hybrid + +BTH101 cells were co- transformed with plasmids encoding T18 and T25 fused to RctB variants following the method described in \(^{33,61}\) and plated on LB agar plates containing kanamycin (25μg/mL) and carbenicillin (100μg/mL). Approximately 500 co- transformants were pooled, resuspended in PBS, diluted to OD600=1 and spotted (10μL) onto LB agar plates containing kanamycin (25μg/mL), carbenicillin (100μg/mL), Xgal (40μg/mL) and IPTG (0.5mM). Plates were incubated for 48h at 30°C, followed by an additional 24h at 4°C. Blue spots indicate an interaction between the tested proteins and white spots indicate no interaction. Empty pKT25 and pUT18C vectors were used as negative control, and pKT25T- Zip/pUT18C- Zip pair was used as positive control. + +## ACKNOWLEDGEMENTS + +We acknowledge Olivier Espeli for his critical discussions on the project and valuable advice, which were instrumental in resolving numerous issues. We would like to express our gratitude to Jakub Czarnecki, Julia Bos, Zeynep Baharoglu and Dhruba Chattoraj for their engaging and insightful discussions. For technical assistance and help with the analysis of the ChIP- seq data we are grateful to Jean- Yves Bouet (CBI), Stéphane Duigou, Christophe Possoz, F.- X. Barre (I2BC), Thomas Cokelaer, and Etienne Kornobis from the Bioinformatics and Biostatistics HUB (IP), as well as Marc Monot and Juliana Pipoli Da Fonseca from the Biomics Sequencing Platform (IP). We thank Tristan Piolot and Julien Dumont from the Orion technical core (CIRB) for their support with the widefield microscope. We thank Matthijn Vos, Eduard Baquero Salazar and Stéphane Tachon for their support at the Nanoimaging Core facility (IP). We thank + +<--- Page Split ---> + +1 Ankur Dalia, Tove Atlung and Flemming Hansen for generously sharing and/or helping access fluorescent 2 DNA- binding proteins needed for three- color fluorescent microscopy in V. cholera. 3 4 FUNDING 5 This work was supported by the Institut Pasteur; Institut National de la Santé et de la Recherche 6 Médicale (INSERM); Centre National de la Recherche Scientifique [CNRS- UMR 3525]; French National 7 Research Agency, Jeunes Chercheurs [ANR- 19CE12- 0001]; Laboratoires d’Excellence [ANR- 10- LABX62- 8 IBEID] ; Fondation pour la Recherche Médicale [Equipe FRM EQU202103012569] ; Fonds National de 9 Recherche Scientifique (FNRS J.0065.23F; FNRS- EQP UN.025.19; and PDR T.0090.22 to AGP); ERC (CoG 10 DiStRes, n° 864311 to AGP) and Fonds Jean Brachet and the Fondation Van Buuren (AGP). T.N. was 11 supported by the Ministère de l’Enseignement Supérieur et de la Recherche and [ANR- 19- CE12- 0001]; 12 F.F. was supported by [ANR- 10- LABX- 62IBEID] and [ANR- 19- CE12- 0001]. + +<--- Page Split ---> + +## FIGURES + +Figure 1: Chr2 initiator RctB binds preferentially to inhibitory sites on ori2. + +a. RctB ChIP signal plotted on the circular map of V. cholerae N16961 using shinyCircos \(^{62}\) with reference sequences of Chr1 (CP028827.1) and Chr2 (CP028828.1) \(^{63}\) , genes annotation (VCNNNN and VCANNNN) is based on the Heidelberg et al. genome \(^{64}\) . The RctB ChIP signal is shown in blue. Significant enrichment peaks are marked with a blue dot with the proximal genes indicated in red (39m-like motifs), green (iteron-like motifs), light blue (crtS) and purple (no similarity to known motifs). b. ChIP peak pattern analysis using the MEME suite \(^{24}\) . Two significantly enriched motifs correspond to known RctB binding sequences: iterons (left) and 39m (right). c. pORI2 copy number (pORI2/oriC) in E. coli containing different binding regions of RctB in attTn7 site. d. Chr2 copy number relative to Chr1 (ori2/ori1) in non-replicating V. cholerae. For VC0643 and VC1643 contained in CDS RctB binding sites have been inactivated without perturbing the amino acid sequence, for VC1042-1043 the intergenic region containing RctB has been deleted. e. Expression of VC1042 and VC1803 genes relative to the housekeeping gene gyrA in the N16961 wild-type strain (WT), and MCH2 (ΔrctB) V. cholerae mutant carrying fused chromosomes \(^{45}\) . To report the statistical significance of the results, a Student's t-test was performed, with non-significant differences denoted by "ns" and statistically significant differences denoted by an asterisk (*), with a significance level set at \(p < 0.05\) + +<--- Page Split ---> +![](images/Figure_1.jpg) + + + +
Figure 1
+ +<--- Page Split ---> + +Figure 2 : RctB predominantly binds to 29/39m sites, forming a nucleoprotein complex that introduces DNA loops into ori2 preventing binding to iterons and DUE. + +a. ChIP-seq of RctB in V. cholerae N16961 focused on ori2. The y-axis represents the normalized ChIP signal of RctB (IP/input coverage), the x-axis displays a 2Kbp window centered on ori2, with genomic coordinates of CP028828.1. The genetic context and RctB binding sites are depicted above the plot (Supplementary Fig. 1b for more details on ori2). b. TEM observations of nucleoprotein complexes formed by RctB binding to an un-methylated ori2 DNA (1864bp). DNA-bound RctB forms loops within ori2 connecting the 29m and 39m sites (white arrows). c. ChIP-seq of RctB on ori2 in a mutant having a point mutation in the 29m site (TTGGAACTATAGTGATATTA[C>A]GGTAAGTG) preventing RctB binding at this site. Same legend as 2a. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
b
+ +![](images/Figure_2.jpg) + + +![](images/Figure_3.jpg) + +
Figure 2
+ +<--- Page Split ---> + +Figure 3 : RctB domain IV structure reveals two dimerization interface involved in 29/39m- mediated inhibition. a. Crystal structure of the RctBIV dimer. Each monomer is composed of an αβ N- terminal subdomain (dark blue) connected to a four α- helices bundle (light blue) via a hinge region. b. Topological representation of RctBIV dimer. The primary dimer interface (blue shadow) is formed by the extension of the central β- sheet of each monomer through β3 and the antiparallel interactions of the stacking α1. c. Crystallographic tetramer formed through lattice contacts mediated by the C- terminal subdomain of neighboring dimers. Important residues contributing to the primary interface are shown in dark green. The secondary interface (yellow shadow) with the I625 and L651 functionally relevant residues shown in red. d. 2D class averages of RctB obtained from cryo- EM e. Bacterial- Two- hybrid of RctB- D314P (DP) vs. RctB domain IV mutants. Empty (negative control), Zip (positive control). f. ChIP- seq of the RctB- L651P mutant at ori2. Same legend as for Fig. 2a. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 3
+ +<--- Page Split ---> + +a. ChIP-seq of RctB at ori2 in a V. cholera ΔcrtS mutant compared to wt. b. Violin plot showing the distribution of cells with 1, 2, 3, and 4 foci. n=Total number of cells analyzed. c-d. Live fluorescence microscopy in V. cholera with 1 crtS at the native locus (c) and 2 crtS at the native locus and near ori1 (d). Binding sites for fluorescent proteins were inserted near ori1, VC783 (near crtS) and ori2. The x-axis represents cell length (μm). The y-axis represents the longitudinal position of ori1, VC783, and ori2 foci within the cells relative to the old pole (0 being the old pole and 1 the new pole). The old pole of the cell was defined as the closest pole to one ori1 focus. n=Total number of cells analyzed. p: percentage of cells. (e) MFA of exponentially growing V. cholera cultures using a corrected reference sequence of Chr1 12. For comparison, MFA from wt (crtSWT) and mutant with relocated crtS near ori1 (crtSor1) from 12 are displayed. Sequencing data from mutant with two crtS sites (crtSWT - crtSor1) are from this study. Log2 of the number of reads starting at each base (normalized against reads from a stationary phase wt control) is plotted against their relative position on Chr1 (in blue) and Chr2 (in yellow). Positions of ori1 and ori2 are set to 0. Any window containing repeated sequences is omitted; thus, the large gap observed in the right arm of Chr2 consists of filtered repeated sequences. Blue dots (Chr1) and yellow dots (Chr2) indicate the averages of 1000-bp windows; black dots indicate the averages of 10,000-bp windows. Dark blue, light blue, orange, and yellow lines indicate ori1, ter1, ori2, and ter2 numbers of reads, respectively. The dashed black lines indicate the number of reads of the loci where the crtS sites are located. + +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
Figure 4
+ +<--- Page Split ---> + +Figure 5: RctB binding pattern on ori2 shift toward iterons after crtS replication. a. V. cholerae growth curve in M9 glucose (0.2%) casamino acids (0.4%) at \(30^{\circ}C\) . At \(\mathrm{OD}_{600} = 0.5\) , DL- serine hydroxamate (SHX) was added to \(1.5mg / mL\) final concentration to prevent new rounds of initiation \(^{38}\) . After 120min of SHX treatment, cells were washed with fresh M9 medium in order to restart replication. b. ori1/ter1 ratio measurement by dPCR at different time points during SHX treatment and removal. c, d, e, f. ChIP- seq experiments on a synchronized population at different time points after SHX removal (R+15min, R+30min, R+45min, R+60min). Top panel: Marker frequency analysis of input samples, coverage on y- axis, genomic coordinates of Chr1 (CP028827.1) and Chr2 (CP028828.1) on x- axis. Curves smoothed with 10kb sliding window. Bottom panel: ChIP- seq RctB at ori2. Same legend as Fig. 2a. + +<--- Page Split ---> +![PLACEHOLDER_29_0] + + +<--- Page Split ---> + +## Figure 6: Model for the Cell-Cycle-Integrated Control of Chr2 Replication in V. cholera + +1. Prior crtS replication, ori2 is sequestered by an inhibition complex, formed by RctB multimeters bridging 29/39m sites. RctB oligomerization domain IV is involved in the formation of this nucleoprotein complex. +2. Upon crtS replication, we hypothesize that the second copy of crtS destabilizes the inhibition complex, possibly at the domain IV secondary interface. 3. Ori2 is thus released and the iteron array becomes accessible for RctB binding. This leads to the opening of the DUE, as shown 18. + +<--- Page Split ---> + +## 1. Prior crtS replication + +Inhibition complex at ori2 + +RctB multimerization through domain IV primary and secondary interfaces + +![PLACEHOLDER_31_0] + + +![PLACEHOLDER_31_1] + + +![PLACEHOLDER_31_2] + + +![PLACEHOLDER_31_3] + +
3. Chr2 initiation
+ +<--- Page Split ---> + +## REFERENCES + +<--- Page Split ---> + +20 Demarre, G. & Chattoraj, D. K. DNA adenine methylation is required to replicate both Vibrio cholerae chromosomes once per cell cycle. PLoS genetics 6, e1000939, doi:10.1371/journal.pgen.1000939 (2010).21 Venkova- Canova, T. & Chattoraj, D. K. Transition from a plasmid to a chromosomal mode of replication entails additional regulators. Proceedings of the National Academy of Sciences of the United States of America 108, 6199- 6204, doi:10.1073/pnas.1013244108 (2011).22 Venkova- Canova, T., Saha, A. & Chattoraj, D. K. A 29- mer site regulates transcription of the initiator gene as well as function of the replication origin of Vibrio cholerae chromosome II. Plasmid 67, 102- 110, doi:10.1016/j.plasmid.2011.12.009 (2012).23 de Lemos Martins, F., Fournes, F., Mazzuoli, M. V., Mazel, D. & Val, M. E. Vibrio cholerae chromosome 2 copy number is controlled by the methylation- independent binding of its monomeric initiator to the chromosome 1 crtS site. Nucleic acids research 46, 10145- 10156, doi:10.1093/nar/gky790 (2018).24 Bailey, T. L., Johnson, J., Grant, C. E. & Noble, W. S. The MEME Suite. Nucleic acids research 43, W39- 49, doi:10.1093/nar/gkv416 (2015).25 Menikpurge, I. P., Woo, K. & Mera, P. E. Transcriptional Activity of the Bacterial Replication Initiator DnaA. Frontiers in microbiology 12, 662317, doi:10.3389/fmicb.2021.662317 (2021).26 Venkova- Canova, T., Baek, J. H., Fitzgerald, P. C., Blokesch, M. & Chattoraj, D. K. Evidence for two different regulatory mechanisms linking replication and segregation of vibrio cholerae chromosome II. PLoS genetics 9, e1003579, doi:10.1371/journal.pgen.1003579 (2013).27 Val, M. E. et al. Fuse or die: how to survive the loss of Dam in Vibrio cholerae. Molecular microbiology 91, 665- 678, doi:10.1111/mmi.12483 (2014).28 Benureau, Y. et al. Method combining BAC film and positive staining for the characterization of DNA intermediates by dark- field electron microscopy. Biol Methods Protoc 5, bpaa012, doi:10.1093/biomethods/bpaa012 (2020).29 Jha, J. K., Ghirlando, R. & Chattoraj, D. K. Initiator protein dimerization plays a key role in replication control of Vibrio cholerae chromosome 2. Nucleic acids research 42, 10538- 10549, doi:10.1093/nar/gku771 (2014).30 Jha, J. K. et al. The DnaK Chaperone Uses Different Mechanisms To Promote and Inhibit Replication of Vibrio cholerae Chromosome 2. mBio 8, doi:10.1128/mBio.00427- 17 (2017).31 Holm, L., Laiho, A., Toronen, P. & Salgado, M. DALI shines a light on remote homologs: One hundred discoveries. Protein science : a publication of the Protein Society 32, e4519, doi:10.1002/pro.4519 (2023).32 Das, D. et al. The structure of the first representative of Pfam family PF09836 reveals a two- domain organization and suggests involvement in transcriptional regulation. Acta Crystallogr Sect F Struct Biol Cryst Commun 66, 1174- 1181, doi:10.1107/S1744309109022672 (2010).33 Karimova, G., Pidoux, J., Ullmann, A. & Ladant, D. A bacterial two- hybrid system based on a reconstituted signal transduction pathway. Proceedings of the National Academy of Sciences of the United States of America 95, 5752- 5756, doi:10.1073/pnas.95.10.5752 (1998).34 Ramachandran, R., Ciaccia, P. N., Filsuf, T. A., Jha, J. K. & Chattoraj, D. K. Chromosome 1 licenses chromosome 2 replication in Vibrio cholerae by doubling the crtS gene dosage. PLoS genetics 14, e1007426, doi:10.1371/journal.pgen.1007426 (2018).35 Dalia, A. B. & Dalia, T. N. Spatiotemporal Analysis of DNA Integration during Natural Transformation Reveals a Mode of Nongenetic Inheritance in Bacteria. Cell 179, 1499- 1511 e1410, doi:10.1016/j.cell.2019.11.021 (2019).36 Ferullo, D. J., Cooper, D. L., Moore, H. R. & Lovett, S. T. Cell cycle synchronization of Escherichia coli using the stringent response, with fluorescence labeling assays for DNA content and replication. Methods 48, 8- 13, doi:10.1016/j.ymeth.2009.02.010 (2009).37 Kemter, F. S. et al. Synchronous termination of replication of the two chromosomes is an evolutionary selected feature in Vibrianceae. PLoS genetics 14, e1007251, doi:10.1371/journal.pgen.1007251 (2018). + +<--- Page Split ---> + +38 Kemter, F. S. et al. Stringent response leads to continued cell division and a temporal restart of DNA replication after initial shutdown in Vibrio cholerae. Molecular microbiology 111, 1617- 1637, doi:10.1111/mmi.14241 (2019).39 Molina- Garcia, L. et al. Functional amyloids as inhibitors of plasmid DNA replication. Scientific reports 6, 25425, doi:10.1038/srep25425 (2016).40 McCahern, M. J., Bott, M. A., Tooker, P. A. & Helinski, D. R. Negative control of plasmid R6K replication: possible role of intermolecular coupling of replication origins. Proceedings of the National Academy of Sciences of the United States of America 86, 7942- 7946, doi:10.1073/pnas.86.20.7942 (1989).41 Zaman, S. & Bastia, D. Oligomeric initiator protein- mediated DNA looping negatively regulates plasmid replication in vitro by preventing origin melting. Molecular cell 20, 833- 843, doi:10.1016/j.molcel.2005.10.037 (2005).42 Ciaccia, P. N., Ramachandran, R. & Chattoraj, D. K. A Requirement for Global Transcription Factor Lrp in Licensing Replication of Vibrio cholerae Chromosome 2. Frontiers in microbiology 9, 2103, doi:10.3389/fmicb.2018.02103 (2018).43 Doan, A. et al. The replication enhancer crt5 depends on transcription factor Lrp for modulating binding of initiator RctB to ori2 of Vibrio cholerae. Nucleic acids research, doi:10.1093/nar/gkad1111 (2023).44 Le Roux, F., Binesse, J., Saulnier, D. & Mazel, D. Construction of a Vibrio splendidus mutant lacking the metalloprotease gene vsm by use of a novel counterselectable suicide vector. Applied and environmental microbiology 73, 777- 784, doi:10.1128/AEM.02147- 06 (2007).45 Val, M. E., Skovgaard, O., Ducos- Galand, M., Bland, M. J. & Mazel, D. Genome engineering in Vibrio cholerae: a feasible approach to address biological issues. PLoS genetics 8, e1002472, doi:10.1371/journal.pgen.1002472 (2012).46 Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. Nature methods 6, 343- 345, doi:10.1038/nmeth.1318 (2009).47 Diaz, R. E., Sanchez, A., Anton Le Berre, V. & Bouet, J. Y. High- Resolution Chromatin Immunoprecipitation: ChIP- Sequencing. Methods in molecular biology 1624, 61- 73, doi:10.1007/978- 1- 4939- 7098- 8_6 (2017).48 Madic, J. et al. Three- color crystal digital PCR. Biomolecular detection and quantification 10, 34- 46, doi:10.1016/j.bdq.2016.10.002 (2016).49 Lo Scrudato, M. & Blokesch, M. The regulatory network of natural competence and transformation of Vibrio cholerae. PLoS genetics 8, e1002778, doi:10.1371/journal.pgen.1002778 (2012).50 Terwilliger, T. C. et al. phenix.mr_rosetta: molecular replacement and model rebuilding with Phenix and Rosetta. J Struct Funct Genomics 13, 81- 90, doi:10.1007/s10969- 012- 9129- 3 (2012).51 Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta crystallographica. Section D, Biological crystallography 68, 352- 367, doi:10.1107/S0907444912001308 (2012).52 Emsley, P. & Cowtan, K. Coot: model- building tools for molecular graphics. Acta crystallographica. Section D, Biological crystallography 60, 2126- 2132, doi:10.1107/S0907444904019158 (2004).53 Smart, O. S. et al. Exploiting structure similarity in refinement: automated NCS and target- structure restraints in BUSTER. Acta crystallographica. Section D, Biological crystallography 68, 368- 380, doi:10.1107/S0907444911056058 (2012).54 Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nature methods 14, 331- 332, doi:10.1038/nmeth.4193 (2017).55 Rohou, A. & Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. Journal of structural biology 192, 216- 221, doi:10.1016/j.jsb.2015.08.008 (2015).56 Bepler, T. et al. Positive- unlabeled convolutional neural networks for particle picking in cryo- electron micrographs. Nature methods 16, 1153- 1160, doi:10.1038/s41592- 019- 0575- 8 (2019). + +<--- Page Split ---> + +57 Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nature methods 14, 290- 296, doi:10.1038/nmeth.4169 (2017).58 David, A. et al. The two Cis- acting sites, parS1 and oriC1, contribute to the longitudinal organisation of Vibrio cholerae chromosome I. PLoS genetics 10, e1004448, doi:10.1371/journal.pgen.1004448 (2014).59 Woldringh, C. L., Hansen, F. G., Vischer, N. O. & Atlung, T. Segregation of chromosome arms in growing and non- growing Escherichia coli cells. Frontiers in microbiology 6, 448, doi:10.3389/fmicb.2015.00448 (2015).60 Ducret, A., Quardokus, E. M. & Brun, Y. V. MicrobeJ, a tool for high throughput bacterial cell detection and quantitative analysis. Nature microbiology 1, 16077, doi:10.1038/nmicrobiol.2016.77 (2016).61 Battesti, A. & Bouveret, E. The bacterial two- hybrid system based on adenylate cyclase reconstitution in Escherichia coli. Methods 58, 325- 334, doi:10.1016/j.ymeth.2012.07.018 (2012).62 Yu, Y., Ouyang, Y. & Yao, W. shinyCircos: an R/Shiny application for interactive creation of Circos plot. Bioinformatics 34, 1229- 1231, doi:10.1093/bioinformatics/btx763 (2018).63 Matthey, N., Drebes Dorr, N. C. & Blokesch, M. Long- Read- Based Genome Sequences of Pandemic and Environmental Vibrio cholerae Strains. Microbiology resource announcements 7, doi:10.1128/MRA.01574- 18 (2018).64 Heidelberg, J. F. et al. DNA sequence of both chromosomes of the cholera pathogen Vibrio cholerae. Nature 406, 477- 483, doi:10.1038/35020000 (2000). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementarymovie1.mp4 Supplementarymovie2.mp4 Supplementarymovie3.mp4 NiaultSupplinformationFigures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54_det.mmd b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9e29fcd3e468c7a75690be5198fb6aecd6156a5b --- /dev/null +++ b/preprint/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54/preprint__0d8682061fae4e5d7c97c59890a3bed6def94782946a05226f210f3e15996d54_det.mmd @@ -0,0 +1,446 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 792, 175]]<|/det|> +# Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae + +<|ref|>text<|/ref|><|det|>[[44, 197, 164, 213]]<|/det|> +Marie-Eve Val + +<|ref|>text<|/ref|><|det|>[[53, 223, 410, 239]]<|/det|> +marie-eve.kennedy-val@pasteur.fr + +<|ref|>text<|/ref|><|det|>[[53, 270, 884, 288]]<|/det|> +Institut Pasteur, Université Paris Cité, CNRS UMR3525 https://orcid.org/0000-0001-7097-9072 + +<|ref|>text<|/ref|><|det|>[[44, 295, 193, 333]]<|/det|> +Theophile Niault +Institut Pasteur + +<|ref|>text<|/ref|><|det|>[[44, 341, 300, 380]]<|/det|> +Ariel Talavera +Université libre de Bruxelles + +<|ref|>text<|/ref|><|det|>[[44, 388, 618, 427]]<|/det|> +Eric LeCam +Institut Gustave Roussy, Université Paris Saclay, CNRS UMR9019 + +<|ref|>text<|/ref|><|det|>[[44, 434, 583, 473]]<|/det|> +Ole Skovgaard +Roskilde University https://orcid.org/0000-0002-4860-1847 + +<|ref|>text<|/ref|><|det|>[[44, 480, 184, 518]]<|/det|> +Florian Fournes +DMF, UNIL + +<|ref|>text<|/ref|><|det|>[[44, 526, 525, 565]]<|/det|> +Léa Wagner +Institut Pasteur, Université Paris Cité, CNRS UMR3525 + +<|ref|>text<|/ref|><|det|>[[44, 572, 658, 611]]<|/det|> +Hedvig Tamman +Université libre de Bruxelles https://orcid.org/0000-0003-4453-7814 + +<|ref|>text<|/ref|><|det|>[[44, 618, 233, 657]]<|/det|> +Andrew Thompson +SOLEIL Synchrotron + +<|ref|>text<|/ref|><|det|>[[44, 665, 300, 703]]<|/det|> +Abel Garcia Pino +Université libre de Bruxelles + +<|ref|>text<|/ref|><|det|>[[44, 711, 164, 728]]<|/det|> +Didier Mazel + +<|ref|>text<|/ref|><|det|>[[44, 733, 940, 772]]<|/det|> +Institut Pasteur, Unité Plasticité du Génome Bactérien, UMR3525, CNRS https://orcid.org/0000-0001-6482-6002 + +<|ref|>text<|/ref|><|det|>[[44, 817, 105, 833]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 855, 138, 872]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 893, 338, 910]]<|/det|> +Posted Date: February 12th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 931, 475, 948]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs-3878828/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 916, 88]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 106, 535, 126]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 161, 933, 205]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 8th, 2025. See the published version at https://doi.org/10.1038/s41467-024-55598-9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[115, 83, 878, 103]]<|/det|> +# Cell cycle-regulated release of a replication inhibition complex in Vibrio cholerae + +<|ref|>text<|/ref|><|det|>[[115, 138, 881, 178]]<|/det|> +Théophile Niault1,2, Ariel Talavera3, Eric Le Cam4, Ole Skovgaard5, Florian Fournes1, Léa Wagner1, Hedvig Tamman3, Andrew Thompson6, Abel Garcia Pino3, Didier Mazel1 and Marie-Eve Val1 + +<|ref|>text<|/ref|><|det|>[[115, 210, 881, 250]]<|/det|> +1 Institut Pasteur, Université Paris Cité, CNRS UMR3525, Unité Plasticité du Génome Bactérien, Département Génomes et Génétique, Paris 75015, France. + +<|ref|>text<|/ref|><|det|>[[115, 259, 567, 273]]<|/det|> +2 Sorbonne Université, Collège Doctoral, Paris 75005, France + +<|ref|>text<|/ref|><|det|>[[115, 283, 881, 321]]<|/det|> +3 Cellular and Molecular Microbiology, Faculté des Sciences, Université libre de Bruxelles (ULB), Boulevard du Triomphe, Brussels, Belgium. + +<|ref|>text<|/ref|><|det|>[[115, 330, 881, 368]]<|/det|> +4 Genome Integrity and Cancer UMR 9019 CNRS, Université Paris Saclay, Gustave Roussy, Villejuif 94805, France + +<|ref|>text<|/ref|><|det|>[[115, 378, 825, 392]]<|/det|> +5 Department of Science, Systems and Models, Roskilde University, Roskilde DK-4000, Denmark. + +<|ref|>text<|/ref|><|det|>[[115, 401, 585, 415]]<|/det|> +6 SOLEIL Synchrotron, Saint-Aubin - BP48, Gif sur Yvette, France + +<|ref|>text<|/ref|><|det|>[[115, 425, 305, 439]]<|/det|> +© Corresponding authors + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 83, 197, 98]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[115, 122, 885, 452]]<|/det|> +Over \(10\%\) of bacteria have expanded their genomes through the domestication of megalasmidis, which subsequently evolved into secondary chromosomes encoding core functions. A fundamental challenge of this genomic expansion is coordinating the replication of multiple replicons within a single cell cycle. In Vibrio cholerae, the replication of the secondary chromosome (Chr2) is intricately linked to the replication of the primary chromosome (Chr1) via a unique checkpoint sequence, crtS. This sequence binds to the initiator of Chr2, RctB. While crtS replication on Chr1 triggers Chr2 replication, the specific molecular dynamics remain elusive. To investigate this, we conducted a comprehensive genome- wide analysis of RctB binding patterns in V. cholerae across various cell cycle stages. Our findings show that RctB primarily binds to sites inhibiting replication initiation at the Chr2 origin (ori2). This inhibitory effect is counteracted when crtS replicates on Chr1, causing a shift in RctB binding to sites that activate replication at ori2. Structural analyses support the formation of diverse oligomeric states of RctB, coupled to the allosteric effect of DNA, which determine ori2 accessibility. We propose a synchronization model where, upon replication, crtS locally destabilizes the RctB inhibition complex, releasing the Chr2 replication origin. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 100, 888, 920]]<|/det|> +2 3 Replication initiation is a crucial step in the bacterial life cycle, subject to complex regulatory controls to 4 adapt to fluctuating growth conditions 1,2. This complexity increases further for bacteria with 5 multipartite genomes, which require an additional layer of control to replicate multiple chromosomes 6 simultaneously 3. Although substantial evidence across a variety of bacterial species supports the idea 7 that the replication of multiple chromosomes must be coordinated 4- 9, Vibrio cholerae is the only species 8 where a specific mechanism synchronizing the replication of two chromosomes has been identified 10- 12. In this pathogen, a replication checkpoint sequence, called crtS, ensures the synchronous replication 10 and termination of the primary (Chr1) and secondary chromosome (Chr2) within a single replication 11 cycle 12. The objective of the present study is to elucidate the molecular mechanisms by which crtS 12 orchestrates this synchronization process. 13 V. cholerae Chr2 originates from an iteron-type plasmid, evident from its replication origin (ori2) and 14 initiator (RctB) 13. RctB is divided into four structural regions (I - IV) 14 (Supplementary Fig. 1a). Its two 15 central domains (II, III) are structurally similar to the Rep iteron plasmid initiators domains (WH1, WH2) 16 while domains (I, IV) are unique to RctB 14. Domains, I, II and III, interact with DNA via Helix-Turn-Helix 17 (HTH) motifs and are essential for ori2 replication 14. The C-terminal domain (IV) is required to down- 18 regulate Chr2 initiation and to coordinate Chr1 and Chr2 replication 15,16. RctB recognizes three types of 19 double-stranded DNA sites: iterons, 29/39m and crtS. To promote replication initiation at ori2, RctB 20 binds to an array of six regularly spaced iterons (six 12-bp repeats), which prompts DNA unwinding at 21 an adjacent AT-rich region known as the DNA Unwinding Element (DUE) (Supplementary Fig. 1b) 17. At 22 this location, RctB engages with single strand of the open DUE on six regularly spaced direct repeats (5'- 23 ATCA) 18. A nucleoprotein complex, composed of RctB, IHF and DnaA, enables the recruitment of the 24 replicative helicase DnaB, similar to what occurs in iteron plasmids 19. Iterons sites each contain a GATC 25 motif which is specifically recognized and methylated by the Dam methylase. Iteron sites must be 26 methylated on both DNA strand for RctB to bind 20. RctB recognizes another type of sites, named for 27 their size as 29-mer or 39-mer (29/39m), which differ from iterons in both sequence and function 28 (Supplementary Fig. 1b). The 29/39m sites play a regulatory role in inhibiting ori2 initiation 21. Two 39m 29 sites and one 29m site flanking the minimal ori2 (containing the six iteron array and the DUE) 30 participates in RctB-mediated iterons handcuffing (pairing via initiator bridges) 21,22. The 29m site is also 31 found in the rctB gene promoter and contributes to rctB self-repression 22. Additionally, RctB binds to a 32 unique 62 bp sequence on Chr1, named crtS 11,12,23. This sequence is crucial for synchronizing the 33 replication of Chr1 and Chr2 12. The replication of crtS triggers the initiation of Chr2, and its position 34 relative to Chr1 origin (ori1) sets the replication timing for Chr2 12. If crtS is deleted, V. cholerae 35 undergoes lethal loss of Chr2 due to under-initiation at ori2 12. Although the action of crtS has been + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 80, 885, 437]]<|/det|> +1 indirectly linked to the inhibitory function of the 29/39m sites 16, the exact molecular dynamics that 2 guide the coordination of Chr1 and Chr2 replication by crtS remains largely undefined. 3 In this study, we conducted a comprehensive, genome- wide analysis of RctB binding, unveiling key 4 elements of the crtS- triggered mechanism for Chr2 replication initiation at various stages of the V. 5 cholera cell cycle. We identified novel RctB binding sites and established a dynamic pattern of RctB 6 binding throughout the cell cycle. Our findings reveal that RctB predominantly binds to the 29/39m sites 7 within ori2, effectively preventing the initiation of Chr2 replication for a large portion of the cell cycle. 8 Using transmission electron microscopy (TEM), we observed that RctB forms large nucleoprotein 9 complexes at ori2, linking the 29/39m sites together. We show that RctB Domain IV can mediate 10 alternative dimerization interfaces, suggesting their role in the assembly of oligomeric bridging 11 structures when bound to 29/39m sites. Following crtS replication, we observed a marked shift in RctB 12 binding preferences at ori2, transitioning towards the iterons and the DUE, thereby enabling Chr2 13 replication initiation. Using an integrative approach combining ChIP- seq, live cell microscopy and 14 structural studies, we propose a model of the firing of Chr2 replication in which one crtS duplication 15 event directly triggers the activation of one ori2 by destabilizing an inhibition complex. + +<|ref|>sub_title<|/ref|><|det|>[[115, 466, 184, 481]]<|/det|> +## RESULTS + +<|ref|>sub_title<|/ref|><|det|>[[115, 504, 501, 521]]<|/det|> +## Genome-wide binding profile of RctB in V. cholera + +<|ref|>text<|/ref|><|det|>[[110, 528, 885, 907]]<|/det|> +We conducted a ChIP- seq of RctB in V. cholera to investigate its genome- wide binding profile within an exponentially growing population. In comparison to a previous ChIP- on- chip study 11, we found that RctB binds to more regions across both chromosomes. These include 8 regions on Chr1 and 6 regions on Chr2, one of which encompasses ori2 (Fig. 1a, Supplementary Table 1). Most of these regions contain either iterons- like sites with a 5'- TGATCA inverted repeat or 29/39m- like sites with a 5'- TTACGG motif, as revealed by MEME analysis 24 (Fig. 1b). We also identified three regions that do not share any sequence similarity with these motifs (Supplementary Table 1), including crtS in the intergenic region between VC0764 and VC0765 on Chr1 (Fig. 1a, Supplementary Fig. 2) 23. Given that our ChIP- seq was executed on non- synchronized, exponentially growing bacterial cultures, the heights of the peaks offer valuable insights into the frequency with which RctB binds at various sites throughout the cell cycle. To assess the impact of the newly discovered RctB binding sites on Chr2 replication, we used a plasmid- based reporter system, pORI2, that exclusively relies on ori2 for replication 23. Each RctB binding regions (spanning 250bp on each side and centered around the peak) was inserted into the lacZ gene of Escherichia coli chromosome, and pORI2 copy number was measured using quantitative digital PCR (dPCR). Our results show that crtS stands out as the only strong enhancer of ori2 initiation (Fig. 1c). Mutations in three of the most prominent binding sites (VC0643, VC01643, and VC1042- VC1043) had + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 81, 885, 411]]<|/det|> +1 minimal effects on Chr2 replication in V. cholera; although the 39m- like site within VC0643 did display 2 a weak, yet statistically significant, negative effect on Chr2 copy number (Fig. 1d). Given RctB's ability to 3 repress its own transcription by binding to its promoter 22, we explored whether RctB could repress 4 other genes. We monitored the transcription levels of VC1042 and VC1803 genes in V. cholera, both 5 of which have an RctB binding site within their promoters. RT- dPCR analyses revealed minimal to no 6 differences in the expression of these genes, regardless of RctB presence (Fig. 1e). Although the RctB 7 binding in the VC1803 promoter exhibited a weak, yet statistically significant, positive regulatory effect 8 on VC1803 transcription. VC1803 is a component of Vibrio pathogenicity island 2 (VPI- 2) and contains a 9 Cro/C1 repressor- like HTH domain profile commonly found in transcriptional repressors of temperate 10 bacteriophages, suggesting potential gene regulation by RctB within the pathogenicity island. In 11 conclusion, while crtS plays a pivotal role in regulating Chr2 replication, the functional relevance of the 12 newly discovered RctB binding sites on Chr1 remains uncertain. RctB appears to regulate its own 13 transcription exclusively, in contrast to the initiator DnaA, which acts more broadly as a general 14 transcription factor 25. + +<|ref|>sub_title<|/ref|><|det|>[[67, 442, 593, 459]]<|/det|> +## RctB binds predominantly to 29/39m inhibitory sites within ori2 + +<|ref|>text<|/ref|><|det|>[[66, 464, 885, 890]]<|/det|> +Upon closer examination of the ori2 region, we observed that RctB ChIP signal was about 10- fold higher than at other chromosomal loci (Supplementary Fig. 2). RctB showed a strong preference for binding to the 29/39m sites (Fig. 2a). This predominant occupancy of RctB at inhibitory sites suggests that initiation of Chr2 replication is hindered for most of the cell cycle. To understand RctB binding dynamics in non- replicating cells, we performed a ChIP- seq of RctB during stationary phase. The RctB binding pattern at ori2 mirrored that observed during the exponential phase (Supplementary Fig. 3), suggesting that RctB binding to 29/39m inhibits initiation at ori2 during both replicative and non- replicative phases. In both conditions, the signal at the 29m site was approximately 3- fold higher than that at the two 39m sites. We hypothesized that this discrepancy might arise from the ParB2 protein, essential for Chr2 segregation, competing with RctB for binding to both 39m sites, as previously reported 26. Confirming our suspicion, our ChIP- seq of ParB2 on ori2 (Supplementary Fig. 4) revealed that ParB2 does indeed compete with RctB at both 39m sites and not at the 29m site. To further investigate the binding requirements of RctB at ori2, we performed a ChIP- seq analysis of RctB in a strain deleted for dam methylase (Δdam#4). RctB cannot bind to iterons in this mutant because they are not methylated, and so there is no initiation at ori2 20,27. The Δdam#4 mutant is viable due to the integration of Chr2 into Chr1, enabling its replication from ori1 as the DnaA initiator does not require Dam methylation 27 (Supplementary Fig. 5a). In the Δdam#4 mutant, the RctB binding pattern at ori2 was identical to that of the wild- type strain (Supplementary Fig. 5b). This result demonstrates that RctB molecules can form + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 81, 884, 125]]<|/det|> +a complex at ori2 strictly through their interaction with 29/39m sites, without any engagement with iterons. It also confirms that RctB binding to crtS is independent of Dam as shown in 23. + +<|ref|>sub_title<|/ref|><|det|>[[111, 153, 803, 171]]<|/det|> +## RctB binding to 29/39m sites generates DNA loops in ori2 and prevents its binding to iterons + +<|ref|>text<|/ref|><|det|>[[110, 177, 885, 460]]<|/det|> +To directly visualize the complex formed by RctB bound to 29/39m on ori2, we used positive- staining transmission electron microscopy (TEM) 28. An unmethylated DNA fragment containing ori2 was chosen to allow RctB to bind exclusively to the 29/39m sites. Our observations revealed that RctB formed large nucleoprotein complexes bridging the 29m and 39m sites together, causing looping within the origin (Fig. 2b, white arrows). Such loops in ori2 could create a steric hindrance preventing the initiation of replication by interfering with RctB binding to the iteron array and the proper unwinding of the DNA strands. To test this hypothesis, we performed a ChIP- seq of RctB in a mutant with the \(C_{21} > A\) mutation at the 29m site \((29m^{*}(C / A))\) known to disrupt RctB binding 16. The binding pattern of RctB at ori2 was drastically altered (Fig. 2c). RctB still bound to both 39m sites but not to the 29m. Additionally, we observed a clear binding of RctB to the array of six iterons and the adjacent DUE. Given that initiation of ori2 replication requires RctB interaction with both the iteron array and the DUE 18, our observations suggest that preventing RctB binding to 29m effectively alleviates the constraint on ori2 initiation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 489, 433, 506]]<|/det|> +## RctBIV harbors two dimerization interfaces + +<|ref|>text<|/ref|><|det|>[[110, 512, 885, 914]]<|/det|> +The C- terminal region of RctB, named domain IV (RctBIV), is critical to downregulate Chr2 copy number through its involvement directly or indirectly in handcuffing activities and binding to 29/39m 15,16,29. It is also crucial for crtS- mediated control of Chr2 replication 16. In vitro, RctB is able to self- oligomerize when bound to DNA through interactions mediated by domain IV 16,23. Thus, we hypothesized that the coordination of Chr1 and Chr2 replication likely depends on the ability of RctB to oligomerize on ori2 via domain IV. While RctB has remained refining to structural studies, likely due to the presence of intrinsically unstructured regions in the protein, the structure of domain I and domains II-III have been solved 14,30. Therefore, we used a truncated version of RctB to determine the structure of domain IV 14,30 and generated the stable fragment RctBIV (amino acids 532 to 658) that could be used in structural biology. The structure of RctBIV reveals that it is organized into two structural subdomains connected by an extended linker or hinge region (Fig. 3a). The N- terminal subdomain consists of a 3- stranded \(\beta\) - sheet stacking two \(\alpha\) - helices and the C- terminal subdomain has 4 \(\alpha\) - helices in a bundle (Fig. 3b). In the crystal, RctBIV retains a dimeric association as predicted by Size Exclusion Chromatography (SEC) (Supplementary Fig. 6). This dimerization is primarily mediated by a \(\beta\) - sheet ( \(\beta\) 3) spanning residue 545 to 549 and involves an interface of 392.4 Å2 (Fig. 3a- b, primary interface). The crystal lattice contacts of the C- terminal subdomain suggest the presence of a secondary interaction interface involving 315.4 Å2, which could mediate further oligomerization via this region (Fig. 3c, secondary interface). A DALI 31 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 884, 291]]<|/det|> +search confirmed the C- terminal subdomain as a protein recognition domain capable of forming transient homodimeric interfaces, with Z- scores between 3.7 and 5.4. This secondary interface closely resembles a representative structural homologue detected by DALI (a putative transcription regulator from Neisseria gonorrhoeae) \(^{32}\) . These findings suggest that RctB \(^{IV}\) may associate in multiple ways and potentially form higher- order oligomeric structures when recruited to DNA. The presence of multiple intermolecular interfaces in RctB \(^{IV}\) inferred from the X- ray structure, together with class averages of RctB obtained by cryo electron microscopy (Fig. 3d) are consistent with the presence of dimeric and tetrameric forms RctB. These findings suggest that RctB \(^{IV}\) may associate in multiple ways and potentially form higher- order oligomeric structures when recruited to DNA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 321, 595, 338]]<|/det|> +## Domain IV dimerization interfaces mediate RctB oligomerization + +<|ref|>text<|/ref|><|det|>[[110, 345, 884, 771]]<|/det|> +To further characterize these interfaces, we used the bacterial two- hybrid assay in E. coli \(^{33}\) . We performed a structure- guided point mutation analysis to probe the potential interfaces involved in the oligomerization of RctB \(^{IV}\) . Given that RctB has a separate dimerization interface in domain II \(^{14}\) , we substituted residue D314 with a proline (D314P) to prevent domain II- mediated interactions that would obscure RctB \(^{IV}\) - mediated interactions (as done in \(^{16}\) ). In Fig. 3e, blue colonies (D314P/D314P) indicated that RctB could interact with other RctB through another domain, as already reported to be via domain IV \(^{16}\) . Our present results showed that substitutions in the \(\beta 3\) strand of RctB \(^{IV}\) caused the formation of white colonies (D314P/D314P- I546G- I546G), indicating that \(\beta 3\) is important for RctB interactions while substitutions that destabilize the \(\alpha 1\) helix appear to have little impact on dimerization as evidenced by the formation of blue colonies (D314P/D314P- A565P). Deletion of the linker connecting the two subdomains (591- 596), gives blue colonies (D314P/D314P- \(\Delta_{591 - 596}\) ) but to a lesser extent than the A565P substitution, suggesting that RctB self- association may be sensitive to local structural rearrangements between the two subdomains. The I625P and L651P substitutions in two different \(\alpha\) - helices (α3 and α6) of the C- terminal subdomain also abrogate RctB interactions confirming our previous results \(^{16}\) . Thus, RctB \(^{IV}\) can form homodimers via two different regions: at a primary interface formed between the \(\beta 3\) sheets and at the secondary interface between \(\alpha\) - helices. Collectively these results suggest that RctB \(^{IV}\) could be an anchor not only for dimerization but also for the formation of transient oligomeric structures when bound to DNA or even to bridge two chromosomal regions. + +<|ref|>sub_title<|/ref|><|det|>[[115, 800, 660, 818]]<|/det|> +## RctB oligomerization is required for the binding to inhibitory sites at ori2 + +<|ref|>text<|/ref|><|det|>[[115, 824, 884, 913]]<|/det|> +To better understand the role of RctB \(^{IV}\) - mediated interactions on the genome- wide in vivo binding of RctB, we performed ChIP- seq of RctB- L651P (Fig. 3f). These analyses revealed a completely different binding profile to ori2 compared with RctB wt (Fig. 2a). RctB- L651P binding to ori2 is fully shifted to the iterons array and the DUE and is absent from the 29/39m sites. However, we observed the same binding + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 80, 885, 250]]<|/det|> +1 profile at crtS between RctB- L65sP and RctB wt (Supplementary Fig. 7a). This suggests that, although not 2 involved in DNA binding, RctBv likely mediates the differential recognition of 29/39m inhibitory sites 3 relative to iterons or crtS, probably via the formation of alternative oligomeric structures. As a control, 4 we performed ChIP- seq of RctB- L65sP in a \(\Delta d a m\) context. As expected, in the absence of Dam 5 methylation, no binding of RctB- L65sP to ori2 was observed, although RctB still bound to crtS 6 (Supplementary Fig. 7b). Based on these results, we propose that intermolecular domain IV mediated 7 RctB interactions are crucial for the inhibition of replication at ori2 through the bridging of 29/39m sites. + +<|ref|>sub_title<|/ref|><|det|>[[117, 274, 549, 291]]<|/det|> +## crtS exclusively affects the RctB inhibition complex at ori2 + +<|ref|>text<|/ref|><|det|>[[110, 297, 885, 768]]<|/det|> +The triggering mechanism of ori2 replication by crtS remains poorly understood. In a \(\Delta crtS\) strain, Chr2 is under- initiated, resulting in cell filamentation and lower Chr2 copy number than Chr1 12. To better understand the consequences of the loss of crtS, we conducted a ChIP- seq analysis of RctB in that strain (Fig. 4a). In this background, RctB also predominantly binds to the 29/39m inhibitory sites at ori2, but the ChIP signal was approximately 10- fold higher, indicating a much stronger inhibition of replication in the absence of crtS. Based on our new findings, we propose that crtS could trigger the initiation of Chr2 replication by destabilizing the nucleoprotein complex that impedes ori2 initiation. When RctB carried the L651P mutation, there was no difference in its binding profile at ori2, irrespective of the presence or absence of crtS, indicating that crtS solely impacts the binding of RctB to the 29/39m (Supplementary Fig. 7a). Given that RctB binds to other 39m- like sites outside of ori2 (Supplementary Table 1), we hypothesized that the absence of crtS could enhance the binding of RctB to these sites as well. However, we observed no increase in RctB binding to the other 39m in the \(\Delta crtS\) mutant (Supplementary Fig. 8). This suggests that crtS has an exclusive activity on ori2. To investigate this further, we examined RctB binding on crtS- containing DNA fragments by TEM. The binding of RctB on crtS was restricted to the site and induced a kink in the DNA (Supplementary Fig. 9a). When we mixed DNA substrates containing unmethylated ori2 and crtS with RctB, we detected sporadic RctB- mediated contacts between ori2 and crtS (Supplementary Fig. 9b), suggesting the possibility of direct interactions between crtS and the 29/39m inhibitory sites in ori2. These findings align with the high frequency of contacts between chromosomal regions proximal to crtS and ori2, observed in vivo by chromosome conformation capture 12. + +<|ref|>sub_title<|/ref|><|det|>[[115, 800, 635, 817]]<|/det|> +## The coupled replication of crtS and activation of ori2 is stoichiometric + +<|ref|>text<|/ref|><|det|>[[115, 823, 884, 913]]<|/det|> +Past studies 12,23,34 demonstrated that introducing an additional crtS copy into Chr1 increases Chr2 copy number. To delve deeper into this relationship, we focused on strains with two crtS sites. First, we inserted one or two copies of crtS into the attTn7 site (located near VC0487 on Chr1) and deleted the endogenous crtS site. As anticipated, the strain with two crtS exhibited a higher Chr2 copy number + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 80, 885, 505]]<|/det|> +compared to the strain with a single crtS when measured on stationary phase culture (1.7 vs. 1) (Supplementary Fig. 10a). However, when comparing the ChIP- seq profiles of RctB at ori2 between these strains, we did not observe any significant differences; in both cases, RctB mainly binds to the 29/39m sites (Supplementary Fig. 10b). This observation indicates that having two copies of crtS does not alter the binding of RctB to activate replication at ori2. Consequently, any activation of ori2 after crtS replication probably occurs transiently. To gain a temporal perspective, we used fluorescence microscopy to observe a V. cholerae mutant with two distant crtS sites 12. This strain contains the endogenous site (crtSwt) and a second site added next to ori1 (crtSori1)—thus delaying the replication timing of both crtS sites (Fig. 4b- d). We followed the dynamics of both crtS and ori2 loci using three different DNA binding protein fluorescent proteins and their cognate sites inserted in the genome: ParBri- CFP (parSpri site near crtSori1), ParBwri1- yGFP (parSwri1 site near crtSwi), LacI- RFP (lacO array near ori2) 35. Compared to the wild type, which contains only one or two foci of ori2, the two- crtS containing mutant has up to four ori2 foci (Fig. 4b). We tracked the longitudinal position of ori2 foci relative to the Chr1 sites in growing cells (Fig. 4d vs. Fig. 4c) and observed that the two- crtS containing cells are born with two ori2 foci. As cell size increases, we observed a sequential increase in the number of ori2 foci, going from two to three and then to four ori2. These results suggest that the early replication of crtSori1 triggers the replication of only one of the two ori2. The replication of the second site crtSwt activates the initiation of the second ori2. + +<|ref|>text<|/ref|><|det|>[[112, 510, 885, 722]]<|/det|> +We took a closer look at the stoichiometry by Marker Frequency Analysis (MFA) on three exponentially growing strains with different crtS contents: the wild type (crtSwt), one mutant with crtS relocated at ori1 (crtSori1), and the two- crtS containing mutant (crtSwt, crtSori1). First, we confirm that the location of crtS on Chr1 controls the time of initiation of Chr2 and thereby the copy number of Chr2, as in crtSwt it is 0.484 and in crtSori1 it is 0.601 (Fig. 4e, left and center) as observed before 12. Then, the strain with two crtS sites (crtSwt and crtSori1) have an ori2 copy number of 0.900 approximating the sum of ori2 copy number in the single crtS strains: crtSwt and crtSori1 (0.484 + 0.601). These findings align with a model in which one crtS duplication event specifically activates one single ori2. This model is illustrated in Supplementary movie 1 (crtSwt), movie 2 (crtSori1) and movie 3 (crtSwt, crtSori1). + +<|ref|>sub_title<|/ref|><|det|>[[115, 752, 585, 768]]<|/det|> +## crtS replication induces a temporal shift in RctB binding at ori2 + +<|ref|>text<|/ref|><|det|>[[113, 775, 884, 913]]<|/det|> +Based on these results, we hypothesized that to activate replication at ori2, RctB should interact with the iterons in a transient way, while remaining bound to the inhibitory 29/39m sites for most of the cell cycle. Our previous approach did not allow us to capture short- lived events. Therefore, we implemented the ChIP- seq of RctB in a replication- synchronized population to monitor the RctB binding profile over the course of the cell cycle. Similar to E. coli, an exponentially growing population of V. cholerae can be synchronized using serine hydroxamate (SHX, a serine analogue that chemically stimulates the stringent + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 78, 886, 386]]<|/det|> +response) 36- 38. During treatment with SHX, replication forks can still progress but the initiation of new rounds of replication is hindered. Transfer of these cells to a fresh medium (without SHX) allows a theoretical synchronous restart of replication. Indeed, previous work showed that after SHX treatment, V. choleraer Chr1 re- initiates first and then Chr2 37 presumably due to control by crtS 12. In our hands, the addition of SHX to 1.5mg/mL final concentration to an exponentially growing culture of V. choleraer caused an instantaneous growth arrest (Fig. 5a). To ensure completion of ongoing replication, we measured the ratio of loci at the origin and terminus of Chr1 (ori1/ter1) by dPCR (Fig. 5b). After 120min of SHX treatment, all cells successfully completed replication with a ratio of ori1/ter1 = 1 meaning that Chr1 was fully replicated. Once SHX was removed, the cells restart to replicate within a time window of 20- 30 minutes, similar to what was observed 38. While synchronization by SHX is not as perfect in V. choleraer as it is in other bacteria, it provided a reasonable approximation to a synchronized population. Thus, we decided to use it to perform ChIP- seq of RctB at different time points after removal of SHX (15, 30, 45, 60 minutes) to follow the binding cycle of RctB. + +<|ref|>text<|/ref|><|det|>[[101, 390, 886, 744]]<|/det|> +Using reads from the input file (control without immunoprecipitation), we performed MFA to track replication fork progression on both chromosomes (Fig. 5c- 5f, top panel). Within the 15 minutes after removal of SHX (Fig. 5c), replication has not restarted, as shown by the flat MFA profile of Chr1 and Chr2, RctB predominantly binds to 29/39m on ori2. After 30 minutes (Fig. 5d), Chr1 starts replicating and at 45 minutes (Fig. 5e) the crtS site seems to be just replicated (the Ter region of Chr1 still appears not replicated), while Chr2 replication has not yet started. At both time points, RctB ChIP- seq patterns at ori2 is nearly unchanged (Fig. 5d- 5e). At 60 minutes (Fig. 5f), Chr2 is finally replicating. On the ChIP- seq profile, two additional peaks appear: one on the six- iteron array and one on the DUE (Fig. 5f, black arrows). Here, we were able to capture a release of RctB from the inhibitory sites (29/39m) towards the activating sites (iteron) and towards the DUE. This temporal shift in the binding pattern of RctB from inhibitory to activating sites on ori2 underscored the opening of ori2 that resulted in the initiation of Chr2 replication. As expected, ori2 activation follows crtS replication on Chr1. We did not observe fluctuations in the binding of RctB at crtS before or after the passage of the replication fork, nor during stationary versus exponential phase of growth (Supplementary Fig. 3 & 11). Thus, RctB binds similarly to crtS throughout the cell cycle. + +<|ref|>sub_title<|/ref|><|det|>[[115, 777, 210, 792]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[115, 800, 884, 913]]<|/det|> +Our work provides cell- cycle- integrated insights into the synchronized replication of Vibrio's two chromosomes. We identified an inhibition complex at ori2 that prevents replication initiation for a significant portion of the cell cycle. We also highlighted the pivotal role of crtS in temporarily disrupting this complex, enabling Chr2 replication. Based on the binding dynamics of RctB throughout the cell cycle, we propose a model for the regulated control of Chr2 replication. In this model, ori2 is inherently + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 78, 886, 410]]<|/det|> +1 bistable, with the replication of crtS acting as a decisive ON switch. Before crtS replication, ori2 is sequestered by an inhibition complex, with RctB bridging the 29/39m sites, and inducing DNA loops that effectively keep ori2 in an OFF state. The replication of crtS destabilizes this complex, causing transient exposure of ori2 to an ON state compatible with replication initiation. This shift promotes the recruitment of RctB to the six iterons, the DUE, and ultimately, the initiation of Chr2 replication. A crucial feature of the stabilization of the OFF state is the oligomerization properties of RctB via the domain IV which defines the formation of the 29/39m bridges and preclude Chr2 replication. In a broader context, analogous types of nucleoprotein complexes have been observed intermolecularly in the regulation of iteron plasmid copy number, involving the initiators RepApp51039 and PiRGK40. In our model, we propose that initiation silencing occurs through intramolecular DNA looping at the origin. This is akin to the mechanism in the F plasmid, where RepE initiators establish a bridge between the iterons of the negative control element (incC) and those of the origin (oriF) that further inhibit the formation of an open complex41. Vibrio distinguishes itself by having tailored specific 29/39m sites to serve this regulatory role. + +<|ref|>text<|/ref|><|det|>[[110, 415, 885, 912]]<|/det|> +The replication of crtS triggers a shift in RctB binding, favoring its binding with the iterons and DUE, conditions that are compatible with ori2 initiation. Further mechanistic research is needed to explore the molecular interactions between RctB and DNA, as well as how crtS disrupts the inhibition complex. However, our model is consistent with the bistable state of ori2 being influenced by crtS replication, the structural properties of RctB and the dramatic changes in the DNA structure and organization induced by RctB. Given that crtS specifically influences RctB binding at ori2—without affecting other genomic sites—and considering the stoichiometric relationship between crtS replication and ori2 activation, our data strongly suggest a direct physical interaction that leads to the activation of ori2 by crtS. This is consistent with our previous observations of frequent inter-chromosomal interactions proximal to crtS and ori212. Moreover, we observed that the regulatory control exerted by crtS over ori2 is lost when a single 29m or 39m site is mutated16. This synergistic effect implies that crtS likely disrupts the 29/39m bridging, rather than affecting each site individually. We propose that each newly duplicated crtS site have RctB bound to it. Both crtS-RctB complexes can each interact with one 29/39m-RctB complex. These interactions destabilize the DNA loop which inhibit ori2 initiation. Once ori2 is exposed, the iteron array and DUE become accessible, so that free RctB molecules can bind and initiate replication (Fig. 6). In contrast to plasmids, where initiators bind solely to iterons, RctB capacity to interact with multiple sites offers a more refined control over replication initiation. This multifaceted binding ability is determined by factors such as methylation-dependent binding to iterons20, domain IV-oligomerization-dependent binding to 29/39m (this study), and Lrp-enhanced binding to crtS42,43. All these factors introduce dynamic binding behaviors throughout the cell cycle, ensuring that Chr2 replication is finely tuned and synchronized with other cellular processes. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 84, 197, 99]]<|/det|> +## METHODS + +<|ref|>sub_title<|/ref|><|det|>[[115, 125, 342, 140]]<|/det|> +## Strains and growth conditions + +<|ref|>text<|/ref|><|det|>[[115, 147, 883, 236]]<|/det|> +Vibrio cholerae N16961 El tor (7PET) is the strain studied here. As a heterologous system for dPCR experiments Escherichia coli K12 MG1655 was also used. Different E.coli strains were used for cloning DH5α, TOP10, P3813 and β3914 \(^{44}\) for conjugation. LB (Luria Broth) was used as standard rich medium for both E. coli and V. cholerae. + +<|ref|>sub_title<|/ref|><|det|>[[115, 268, 376, 283]]<|/det|> +## Strains and plasmids constructions + +<|ref|>text<|/ref|><|det|>[[110, 290, 884, 500]]<|/det|> +Both V. cholerae and E. coli strains were modified using conditionally replicative suicide vectors as described in \(^{45}\) . These vectors were generated by Gibson assembly \(^{46}\) , and in cases requiring specific point mutations, they were introduced via PCR using oligonucleotides containing the desired mutations at the 5' end. For ChIP-seq experiment (on RctB and on ParB2), we replaced the wild-type alleles on the chromosome with a C-terminal 3xFLAG through allelic exchange. For the dPCR experiments, we introduced RctB secondary binding sites into the lacZ gene of MG1655, we took an extending of 250 base pairs on each side of the peak detected from ChIP-seq analysis. Subsequently, these modified strains were transformed with a pORI2 plasmid, following the procedure outlined in \(^{23}\) . Plasmids and strains used in this study are listed in Supplementary Tables 2 and 3. + +<|ref|>sub_title<|/ref|><|det|>[[115, 532, 305, 548]]<|/det|> +## High Resolution ChIP-seq + +<|ref|>text<|/ref|><|det|>[[110, 555, 884, 907]]<|/det|> +ChIP-seq were performed as described in \(^{47}\) in V. cholerae expressing protein- 3xFLAG under its native promoter. We confirmed that 3xFLAG did not impair protein function by dPCR measurement of Chr2 copy number. Cells were grown in 100mL Mueller Hinton (MH) media with shaking at \(37^{\circ}C\) until they reach mid- log growth phase (OD \(_{600}\) 0.5). Then cells were fixed by crosslinking with \(1\%\) formaldehyde for 30 min at room temperature under agitation, followed by quenching with 0.5M glycine for 15 min. Bacteria were harvested, washed two times in ice- cold Tris- Buffered saline (20 mM Tris/HCl pH 7.5, 150 mM NaCl) and pellets were stored at - 80°C. Pellets were lyzed in buffer containing lysozyme and protease inhibitor pill (cOmplete Protease Inhibitor, Roche). Chromatin was then fragmented by sonication (Covaris S220) for in milliTUBE (1ml with AFA Fiber). Correct DNA fragmentation centered around 300bp was confirmed using \(1.8\%\) agarose gel electrophoresis. Following sonication, protein- DNA complexes were immunoprecipitated with anti- FLAG magnetic beads (M8823 Sigma). Correct immunoprecipitation was confirmed by western blot using anti- FLAG antibodies (F7425 Sigma). Once recovered, immunoprecipitated DNA was de- crosslink at \(65^{\circ}C\) overnight. Then DNA was blunt- ended, A- tailed, ligated to adaptors, amplified by PCR (13 cycles) and purified using TruSeq ChIP Library Preparation Kit (illumina) and AMPureXP beads (Beckman Coulter). ChIP- seq libraries quality was + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 81, 885, 340]]<|/det|> +measured by Agilent Bioanalyzer 2100 instrument using a DNA High Sensitivity chip. Finally, libraries were sequenced using a MiniSeq Mid Output Kit on a Miniseq sequencing machine (Illumina). Paired- end reads from two independent RctB ChIP- seq experiments were mapped to Vibrio cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) with Bowtie2 using galaxy. pasteur platform. Peak calling was performed using MACS2, where aligned reads were significantly enriched compared to a control (strain without FLAG). Identified peaks were checked by hand, using IGV genome browser to confirm correct peak assignment. MEME- ChIP (version 5.4.1) online browser was used for motif enrichment finding. For normalization, the IP and INPUT files were first adjusted to the same number of reads using R. Then the coverage value for each position in the IP file was divided by the average coverage value of a 2kb sliding window from the INPUT file. Data available at the European Nucleotide Archive (Supplementary Information). + +<|ref|>text<|/ref|><|det|>[[111, 345, 885, 675]]<|/det|> +The genomic replication pattern of the different strains was examined using a computational approach based on Marker Frequency Analysis (MFA). The reads from the INPUT sample (not immunoprecipitated DNA) of the ChIP- seq data were mapped to V. cholerae N16961 reference genome (Chr1: CP028827.1 and Chr2: CP028828.1) using BOWTIE2, the resultant mapping was saved in two separate Binary Alignment Map (BAM) files corresponding to the two chromosomes. Subsequently, a R script (available on demand) was used for data processing and visualization. The depth of coverage was extracted from the BAM files using the Rsamtools package. The genome was divided into 10 kbp bins, and the average coverage for each bin was computed. Coverage values were then normalized with respect to the average coverage of the termination (ter) region, which is located at the midpoint of the first chromosome (+/- 5kb). Coverage values more than three standard deviations from the mean were considered outliers and removed from the dataset. For visualization, a ggplot2- based plot was generated, showing the normalized coverage for each bin against the genomic position in Mbp. Linear regression analysis was performed separately on each half of the chromosomes, and the best- fit lines were superimposed on the plot (Supplementary Fig. 12). + +<|ref|>sub_title<|/ref|><|det|>[[115, 705, 320, 721]]<|/det|> +## Marker Frequency Analysis + +<|ref|>text<|/ref|><|det|>[[115, 727, 881, 770]]<|/det|> +Grow conditions, DNA extraction, sequencing and analysis for MFA was done as described previously 12. Data available at the European Nucleotide Archive (Supplementary Information). + +<|ref|>sub_title<|/ref|><|det|>[[115, 800, 361, 817]]<|/det|> +## Digital PCR quantification (dPCR) + +<|ref|>text<|/ref|><|det|>[[115, 823, 884, 913]]<|/det|> +Quantifications of (ori1 and ori2) in V. cholerae and (oriC and pORI2) in E. coli were performed as described in 16 using multiplex dPCR (Stilla Technologies). dPCR was performed directly on cell lysate: 1mL of overnight culture was washed twice with 1mL PBS, pellets were then frozen at - 20°C and resuspended in 200uL PBS, samples were then boiled for 10min. Samples were centrifuge for 10min, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 81, 885, 269]]<|/det|> +1 supernatant was recovered, and DNA content was measured using Qubit device (ThermoFisher). PCR reactions were performed with 0.1 ng of DNA using the PerfeCTa MultiPlex qPCR ToughMix (Quantabio) on a Sapphire chip (Stilla Technologies). Digital PCR was conducted on a Naica Geode and Image acquisition on the Naica Prism3 reader. Images were analyzed Crystal Miner software (Stilla Technologies). The dPCR run was performed using the following steps: droplet partition (40°C, atmospheric pressure AP to + 950mbar, 12 min), initial denaturation (95°C, +950 mbar, for 2 min), followed by 45 cycles at (95°C for 10 s and 60°C for 30 s), droplet release (down 25°C, down to AP, 33 min). + +<|ref|>sub_title<|/ref|><|det|>[[67, 298, 444, 315]]<|/det|> +## Digital PCR quantification of RNA (RT-dPCR) + +<|ref|>text<|/ref|><|det|>[[66, 321, 885, 530]]<|/det|> +Quantification of RctB mRNA was performed in V. cholerae from exponentially growing cultures (OD600 0.5) using multiplex digital RT- dPCR (Stilla Technologies) 48. Primers and probes are listed in Supplementary Table 4. V. cholerae RNA was prepared as described in 49. PCR reactions were performed with 1 ng of RNA using the qScriptTM XLT One- Step RT- qPCR ToughMix® (Quantabio). RT- dPCR and data analysis were conducted as described above for dPCR. The RT- dPCR run was performed in the following steps: droplet partition (40°C, AP to +950 mbar, 12 min), cDNA synthesis (50°C, +950 mbar, 10 min), initial denaturation (95°C, +950 mbar, for 1 min), followed by 45 cycles at (95°C for 10 s and 60°C for 15 s), droplet release (down 25°C, down to AP, 33 min). Expression values were normalized to the expression of the housekeeping gene gyrA as described in 49. + +<|ref|>sub_title<|/ref|><|det|>[[67, 560, 388, 576]]<|/det|> +## SHX Vibrio cholerae synchronization + +<|ref|>text<|/ref|><|det|>[[66, 584, 885, 817]]<|/det|> +V. cholerae expressing RctB-3xFLAG from its natural promoter was streaked from a -80°C stock onto an MH agar plate. An isolated colony was then picked and cultured overnight in M9 media containing 0.2% casamino acids and 0.4% glucose at 30°C. Cells from this overnight culture were subsequently diluted 100-fold in fresh media and grown in 40 mL of M9 media with 0.2% casamino acids and 0.4% glucose at 30°C. When the cells reached an OD600nm of 0.5, they were treated with SHX (DL-serine hydroxamate, Sigma) at a final concentration of 1.5 mg/mL for 120 minutes. This treatment prevented new rounds of initiation and allowed cells to complete ongoing replication. Cells stalled in G1 were then centrifuged at 6,000g for 15 minutes, and the pellets were resuspended in 40 mL of fresh M9 media before being returned to 30°C for growth. Cells were fixed with 1% formaldehyde at various time points: 15, 30, 45, and 60 minutes after the wash. Finally, samples were processed according to the ChIP-seq protocol. + +<|ref|>sub_title<|/ref|><|det|>[[67, 848, 373, 864]]<|/det|> +## Transmission electron microscopy + +<|ref|>text<|/ref|><|det|>[[66, 871, 884, 913]]<|/det|> +The ori2 DNA probe (1864bp) containing the three 29/39mer sites was produced by PCR from pFF149 plasmid using MV109/MV226 primers (Supplementary Table 4). The crtS probe (944bp) was produced + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[100, 81, 885, 411]]<|/det|> +1 by PCR from a pUC18::crtS plasmid using MV577/MV485 primers. DNA fragments were then purified using AMPure XP beads (Beckman Coulter). RctB, Dnak and Dnal proteins were produced and purified like in PMC6212839. RctB was first treated with Dnak/J chaperones to enhance RctB activity. For this, RctB (1200nM) was incubated with Dnak (100nM) and Dnal (100nM) in 20uL of 10 mM Tris- HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer for 10min at 4°C. Then, for the DNA- protein complexes formation 2uL of treated RctB was mixed with 4nM of DNA probes (ori2 probe, crtS probe or both) in a 10 mM Tris- HCl pH 7.5, 50 mM NaCl, 1mM MgCl2 buffer (20uL final volume) for 10min at room temperature. For observation of DNA- protein complexes, 5 uL drop of the incubation was deposited on a 600- mesh copper grid previously covered with a thin carbon film and preactivated by glow discharge in the presence of amylamine (Sigma- Aldrich, France). The grids were rinsed and positively stained with 2% (w/v) aqueous uranyl acetate, carefully dried with filter paper, and observed in annular dark- field mode in lossless filtered imaging using a Zeiss 902 transmission electron microscope. Images were captured at 85,000× magnification with a Veleta MegaviewIII CCD camera and analyzed with iTEM software (Olympus Soft Imaging Solution). + +<|ref|>sub_title<|/ref|><|det|>[[115, 440, 765, 459]]<|/det|> +## Expression and purification of RctB and RctBIV for crystallization and biophysical studies + +<|ref|>text<|/ref|><|det|>[[110, 463, 885, 580]]<|/det|> +RctB and were RctBIV cloned into a pET- 24b plasmid downstream an N- terminal 10xHis_SUMO tag. BL21 E.coli cells were transformed with these constructs for expression and purification. Transformed cells were first grown at 37°C until reaching an OD\~0.6 then expression was induced with 0.25 mM Isopropyl \(\beta\) - D- 1- thiogalactopyranoside (IPTG) overnight at 16°C. Afterwards cells were harvested by centrifugation at 3000g. + +<|ref|>text<|/ref|><|det|>[[110, 585, 885, 794]]<|/det|> +For purification, the cell lysate supernatants were loaded onto a Ni- Sepharose column. In both cases, proteins were eluted using a step gradient of imidazole. The fractions containing the protein of interest were then loaded into a Citiva 60/160 Seperdex 200 size exclusion chromatography column, equilibrated in 50 mM Hepes pH 7.5, 500 mM NaCl, 500 mM KCl, 2 mM MgCl2 and 1 mM TCEP. The 10xHis- SUMO tag was removed by incubating overnight the protein with the protease Ulpl in a 1:100 molar ratio. The tag- free RctB and RctBIV were then isolated from the cleaved tag and Ulpl by loading the mixture into a Co- sepharose column and collecting the flow- through that contained the tag- free samples. For further studies RctB and RctBIV were extensively dialyzed against 50 mM Hepes pH 7.5, 200 mM NaCl, 2 mM MgCl2 and 1 mM TCEP. + +<|ref|>sub_title<|/ref|><|det|>[[115, 825, 518, 843]]<|/det|> +## RctBIV Analytical Size Exclusion Chromatography (SEC) + +<|ref|>text<|/ref|><|det|>[[110, 848, 884, 914]]<|/det|> +Analytical SEC of RctBIV was carried out on a Superdex 200 10/300 column (Citiva) equilibrated in 50 mM Hepes, 200mM NaCl, 2 mM MgCl2 and 1 mM TCEP. The flow rate was 1 ml/min and the injection volume 0.5 ml. The calibration of the column was performed by running the Biorad gel filtration standards (Cat. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 80, 886, 920]]<|/det|> +1 No. 151- 1901), containing a mixture of bovine thyroglobulin (670 kDa), bovine \(\gamma\) - globulin (158 kDa), chicken ovalbumin (44 kDa), equine myoglobin (17 kDa) and vitamin B12 (1.35 kDa). 3 RctBIV crystallization and structure determination 5 The screening of crystallization conditions of RctBIV was carried out using the sitting- drop vapor- diffusion method with highly pure samples of RctBIV at a concentration of 8 mgmL- 1. The drops were set up in Swiss (MRC) 96- well two- drop UVP sitting- drop plates using the Mosquito HTS system (TTP Labtech). Drops of 0.1 μL protein and 0.1 μL precipitant solution were equilibrated to 80 μL precipitant solution in the reservoir. Commercially available screens PACT premier, LMB and SG1 (Molecular Dimensions) were used to test crystallization conditions. The condition resulting in protein crystals (PACTpr screen position E3: 200 mM NaI, 20 % PEG 3350) were repeated as 2 μL drops. Crystals were harvested using suitable cryo- protecting solutions (consisting of the mother liquor supplemented with 20% to 25% of glycerol) and vitrified in liquid \(\mathbb{N}_2\) for transport and storage before X- ray exposure. 14 X- ray diffraction data was collected at the SOLEIL synchrotron (Gif- sur- Yvette, Paris, France) on the Proxima 1 (PX1) beamlines using an Eiger- X 16M detector. Native crystals typically diffracted to 1.8 Å – 2.0 Å resolution. We screened a collection of heavy atoms for experimental phasing and observed that the combination of quick soaks (under 30s) with NaBr at 1.5M, and longer soaks (20 min to 30 min) in AgNO3 (10 mM) and GdCl (10 mM) provided sufficient isomorphous signal for experimental phasing. 19 For this, data were collected at the peak and inflexion point of the Br edge and Ag and Gd K- edge. After phase improvement with Pirate and density modification with Parrot, the maps were used initially used in Buccaneer for model building and then as starting point for the MR- Rosetta 50 suit which completed one third of the structure. Further cycles of automated model building with AutoBuild from the Phenix package 51 extended the model to 90%. After several iterations of manual building with Coot 52 and maximum likelihood refinement as implemented in Buster/TNT 53, the model was extended to cover all the residues (R/Rfree of 19.2%/26.7 %). Supplementary Table 5 details all the X- ray data collection and refinement statistics. 27 Single particle cryo- electron microscopy (cryo- EM) 29 For the cryo- EM experiments RctB was purified in 50 mM Hepes, 200mM NaCl, 2 mM MgCl2 and 1 mM TCEP. RctB (3 μL) was applied to gold Quantifoil grids (UltraFoil 1.2/1.3 Au 300) under 100 % relative humidity at 10 °C at 0.25 mg/ml. Two grids were blotted for 2 and 3 seconds, respectively, and plunged into liquid ethane using a Vitrobot IV (ThermoFisher Scientific). Particles were imaged from the aforementioned grids on a Glacios microscope (ThermoFisher Scientific) operating at 200 kV with a Falcon 4i direct electron detector. In total, 3800 micrographs, with 0.96 Å/pxl, were collected using a defocus range from -1 μm to -3 μm and total dose of 50 e/Å2. The micrographs movies were motion + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 80, 885, 149]]<|/det|> +1 corrected using MotionCor2 \(^{54}\) , and contrast transfer function, CTF, were estimated using CTFFind- 4.1 \(^{55}\) . Particles were picked using Topaz \(^{56}\) and subjected to reference free 2D classification using cryoSPARC \(^{57}\) . Several rounds of 2D classifications rendered a total of 133108 particles. + +<|ref|>sub_title<|/ref|><|det|>[[117, 179, 373, 195]]<|/det|> +## Live-cell Fluorescence Microscopy + +<|ref|>text<|/ref|><|det|>[[110, 201, 885, 410]]<|/det|> +The genes encoding for the CFP- ParBp1, yGFP- ParBpmT1 and LacI- RFP- T fluorescent DNA binding proteins were inserted into the V. cholera chromosome \(^{35,58,59}\) . Their cognate binding sites were inserted near ori1 (parSp1), near VC783 (parSpMT1), and near ori2 (lacO array). For microscopy observations, cultures were prepared following the procedures described in \(^{12}\) . Initially, strains were streaked onto MH plates from a - 80°C stock and then grown overnight in MH rich liquid media. The following day, cultures were diluted at a ratio of 1:1000 in M9 with 1% fructose and grown to exponential phase (OD600 = 0.2). A 2μL aliquot of this culture was spotted on an agar pad (M9, 1% fructose, 1% agarose), for microscopy observation. Cell imaging was performed using an Axio Observer 7 inverted videomicroscope (Zeiss). Analysis of the acquired snapshots was conducted using Fiji software with the MicrobeJ plugin \(^{60}\) . + +<|ref|>sub_title<|/ref|><|det|>[[116, 440, 274, 456]]<|/det|> +## Bacterial Two Hybrid + +<|ref|>text<|/ref|><|det|>[[110, 463, 885, 650]]<|/det|> +BTH101 cells were co- transformed with plasmids encoding T18 and T25 fused to RctB variants following the method described in \(^{33,61}\) and plated on LB agar plates containing kanamycin (25μg/mL) and carbenicillin (100μg/mL). Approximately 500 co- transformants were pooled, resuspended in PBS, diluted to OD600=1 and spotted (10μL) onto LB agar plates containing kanamycin (25μg/mL), carbenicillin (100μg/mL), Xgal (40μg/mL) and IPTG (0.5mM). Plates were incubated for 48h at 30°C, followed by an additional 24h at 4°C. Blue spots indicate an interaction between the tested proteins and white spots indicate no interaction. Empty pKT25 and pUT18C vectors were used as negative control, and pKT25T- Zip/pUT18C- Zip pair was used as positive control. + +<|ref|>sub_title<|/ref|><|det|>[[116, 682, 292, 697]]<|/det|> +## ACKNOWLEDGEMENTS + +<|ref|>text<|/ref|><|det|>[[110, 704, 885, 914]]<|/det|> +We acknowledge Olivier Espeli for his critical discussions on the project and valuable advice, which were instrumental in resolving numerous issues. We would like to express our gratitude to Jakub Czarnecki, Julia Bos, Zeynep Baharoglu and Dhruba Chattoraj for their engaging and insightful discussions. For technical assistance and help with the analysis of the ChIP- seq data we are grateful to Jean- Yves Bouet (CBI), Stéphane Duigou, Christophe Possoz, F.- X. Barre (I2BC), Thomas Cokelaer, and Etienne Kornobis from the Bioinformatics and Biostatistics HUB (IP), as well as Marc Monot and Juliana Pipoli Da Fonseca from the Biomics Sequencing Platform (IP). We thank Tristan Piolot and Julien Dumont from the Orion technical core (CIRB) for their support with the widefield microscope. We thank Matthijn Vos, Eduard Baquero Salazar and Stéphane Tachon for their support at the Nanoimaging Core facility (IP). We thank + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 81, 885, 370]]<|/det|> +1 Ankur Dalia, Tove Atlung and Flemming Hansen for generously sharing and/or helping access fluorescent 2 DNA- binding proteins needed for three- color fluorescent microscopy in V. cholera. 3 4 FUNDING 5 This work was supported by the Institut Pasteur; Institut National de la Santé et de la Recherche 6 Médicale (INSERM); Centre National de la Recherche Scientifique [CNRS- UMR 3525]; French National 7 Research Agency, Jeunes Chercheurs [ANR- 19CE12- 0001]; Laboratoires d’Excellence [ANR- 10- LABX62- 8 IBEID] ; Fondation pour la Recherche Médicale [Equipe FRM EQU202103012569] ; Fonds National de 9 Recherche Scientifique (FNRS J.0065.23F; FNRS- EQP UN.025.19; and PDR T.0090.22 to AGP); ERC (CoG 10 DiStRes, n° 864311 to AGP) and Fonds Jean Brachet and the Fondation Van Buuren (AGP). T.N. was 11 supported by the Ministère de l’Enseignement Supérieur et de la Recherche and [ANR- 19- CE12- 0001]; 12 F.F. was supported by [ANR- 10- LABX- 62IBEID] and [ANR- 19- CE12- 0001]. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 107, 184, 122]]<|/det|> +## FIGURES + +<|ref|>text<|/ref|><|det|>[[115, 153, 668, 170]]<|/det|> +Figure 1: Chr2 initiator RctB binds preferentially to inhibitory sites on ori2. + +<|ref|>text<|/ref|><|det|>[[114, 175, 885, 532]]<|/det|> +a. RctB ChIP signal plotted on the circular map of V. cholerae N16961 using shinyCircos \(^{62}\) with reference sequences of Chr1 (CP028827.1) and Chr2 (CP028828.1) \(^{63}\) , genes annotation (VCNNNN and VCANNNN) is based on the Heidelberg et al. genome \(^{64}\) . The RctB ChIP signal is shown in blue. Significant enrichment peaks are marked with a blue dot with the proximal genes indicated in red (39m-like motifs), green (iteron-like motifs), light blue (crtS) and purple (no similarity to known motifs). b. ChIP peak pattern analysis using the MEME suite \(^{24}\) . Two significantly enriched motifs correspond to known RctB binding sequences: iterons (left) and 39m (right). c. pORI2 copy number (pORI2/oriC) in E. coli containing different binding regions of RctB in attTn7 site. d. Chr2 copy number relative to Chr1 (ori2/ori1) in non-replicating V. cholerae. For VC0643 and VC1643 contained in CDS RctB binding sites have been inactivated without perturbing the amino acid sequence, for VC1042-1043 the intergenic region containing RctB has been deleted. e. Expression of VC1042 and VC1803 genes relative to the housekeeping gene gyrA in the N16961 wild-type strain (WT), and MCH2 (ΔrctB) V. cholerae mutant carrying fused chromosomes \(^{45}\) . To report the statistical significance of the results, a Student's t-test was performed, with non-significant differences denoted by "ns" and statistically significant differences denoted by an asterisk (*), with a significance level set at \(p < 0.05\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 12, 870, 800]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[492, 861, 551, 877]]<|/det|> +
Figure 1
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 106, 883, 147]]<|/det|> +Figure 2 : RctB predominantly binds to 29/39m sites, forming a nucleoprotein complex that introduces DNA loops into ori2 preventing binding to iterons and DUE. + +<|ref|>text<|/ref|><|det|>[[114, 153, 885, 339]]<|/det|> +a. ChIP-seq of RctB in V. cholerae N16961 focused on ori2. The y-axis represents the normalized ChIP signal of RctB (IP/input coverage), the x-axis displays a 2Kbp window centered on ori2, with genomic coordinates of CP028828.1. The genetic context and RctB binding sites are depicted above the plot (Supplementary Fig. 1b for more details on ori2). b. TEM observations of nucleoprotein complexes formed by RctB binding to an un-methylated ori2 DNA (1864bp). DNA-bound RctB forms loops within ori2 connecting the 29m and 39m sites (white arrows). c. ChIP-seq of RctB on ori2 in a mutant having a point mutation in the 29m site (TTGGAACTATAGTGATATTA[C>A]GGTAAGTG) preventing RctB binding at this site. Same legend as 2a. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[199, 20, 728, 199]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[200, 275, 213, 287]]<|/det|> +
b
+ +<|ref|>image<|/ref|><|det|>[[200, 280, 740, 408]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[200, 450, 722, 670]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[450, 817, 508, 831]]<|/det|> +
Figure 2
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 885, 339]]<|/det|> +Figure 3 : RctB domain IV structure reveals two dimerization interface involved in 29/39m- mediated inhibition. a. Crystal structure of the RctBIV dimer. Each monomer is composed of an αβ N- terminal subdomain (dark blue) connected to a four α- helices bundle (light blue) via a hinge region. b. Topological representation of RctBIV dimer. The primary dimer interface (blue shadow) is formed by the extension of the central β- sheet of each monomer through β3 and the antiparallel interactions of the stacking α1. c. Crystallographic tetramer formed through lattice contacts mediated by the C- terminal subdomain of neighboring dimers. Important residues contributing to the primary interface are shown in dark green. The secondary interface (yellow shadow) with the I625 and L651 functionally relevant residues shown in red. d. 2D class averages of RctB obtained from cryo- EM e. Bacterial- Two- hybrid of RctB- D314P (DP) vs. RctB domain IV mutants. Empty (negative control), Zip (positive control). f. ChIP- seq of the RctB- L651P mutant at ori2. Same legend as for Fig. 2a. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[104, 50, 800, 936]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[447, 976, 504, 990]]<|/det|> +
Figure 3
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 111, 885, 538]]<|/det|> +a. ChIP-seq of RctB at ori2 in a V. cholera ΔcrtS mutant compared to wt. b. Violin plot showing the distribution of cells with 1, 2, 3, and 4 foci. n=Total number of cells analyzed. c-d. Live fluorescence microscopy in V. cholera with 1 crtS at the native locus (c) and 2 crtS at the native locus and near ori1 (d). Binding sites for fluorescent proteins were inserted near ori1, VC783 (near crtS) and ori2. The x-axis represents cell length (μm). The y-axis represents the longitudinal position of ori1, VC783, and ori2 foci within the cells relative to the old pole (0 being the old pole and 1 the new pole). The old pole of the cell was defined as the closest pole to one ori1 focus. n=Total number of cells analyzed. p: percentage of cells. (e) MFA of exponentially growing V. cholera cultures using a corrected reference sequence of Chr1 12. For comparison, MFA from wt (crtSWT) and mutant with relocated crtS near ori1 (crtSor1) from 12 are displayed. Sequencing data from mutant with two crtS sites (crtSWT - crtSor1) are from this study. Log2 of the number of reads starting at each base (normalized against reads from a stationary phase wt control) is plotted against their relative position on Chr1 (in blue) and Chr2 (in yellow). Positions of ori1 and ori2 are set to 0. Any window containing repeated sequences is omitted; thus, the large gap observed in the right arm of Chr2 consists of filtered repeated sequences. Blue dots (Chr1) and yellow dots (Chr2) indicate the averages of 1000-bp windows; black dots indicate the averages of 10,000-bp windows. Dark blue, light blue, orange, and yellow lines indicate ori1, ter1, ori2, and ter2 numbers of reads, respectively. The dashed black lines indicate the number of reads of the loci where the crtS sites are located. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[500, 980, 560, 991]]<|/det|> +
Figure 4
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 82, 885, 291]]<|/det|> +Figure 5: RctB binding pattern on ori2 shift toward iterons after crtS replication. a. V. cholerae growth curve in M9 glucose (0.2%) casamino acids (0.4%) at \(30^{\circ}C\) . At \(\mathrm{OD}_{600} = 0.5\) , DL- serine hydroxamate (SHX) was added to \(1.5mg / mL\) final concentration to prevent new rounds of initiation \(^{38}\) . After 120min of SHX treatment, cells were washed with fresh M9 medium in order to restart replication. b. ori1/ter1 ratio measurement by dPCR at different time points during SHX treatment and removal. c, d, e, f. ChIP- seq experiments on a synchronized population at different time points after SHX removal (R+15min, R+30min, R+45min, R+60min). Top panel: Marker frequency analysis of input samples, coverage on y- axis, genomic coordinates of Chr1 (CP028827.1) and Chr2 (CP028828.1) on x- axis. Curves smoothed with 10kb sliding window. Bottom panel: ChIP- seq RctB at ori2. Same legend as Fig. 2a. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 763, 100]]<|/det|> +## Figure 6: Model for the Cell-Cycle-Integrated Control of Chr2 Replication in V. cholera + +<|ref|>text<|/ref|><|det|>[[115, 106, 884, 220]]<|/det|> +1. Prior crtS replication, ori2 is sequestered by an inhibition complex, formed by RctB multimeters bridging 29/39m sites. RctB oligomerization domain IV is involved in the formation of this nucleoprotein complex. +2. Upon crtS replication, we hypothesize that the second copy of crtS destabilizes the inhibition complex, possibly at the domain IV secondary interface. 3. Ori2 is thus released and the iteron array becomes accessible for RctB binding. This leads to the opening of the DUE, as shown 18. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[36, 38, 197, 48]]<|/det|> +## 1. Prior crtS replication + +<|ref|>text<|/ref|><|det|>[[36, 56, 168, 66]]<|/det|> +Inhibition complex at ori2 + +<|ref|>text<|/ref|><|det|>[[36, 74, 247, 92]]<|/det|> +RctB multimerization through domain IV primary and secondary interfaces + +<|ref|>image<|/ref|><|det|>[[270, 8, 576, 130]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[270, 140, 575, 198]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[270, 220, 575, 315]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[36, 380, 911, 460]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[36, 388, 145, 399]]<|/det|> +
3. Chr2 initiation
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 213, 98]]<|/det|> +## REFERENCES + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 886, 900]]<|/det|> +20 Demarre, G. & Chattoraj, D. K. DNA adenine methylation is required to replicate both Vibrio cholerae chromosomes once per cell cycle. PLoS genetics 6, e1000939, doi:10.1371/journal.pgen.1000939 (2010).21 Venkova- Canova, T. & Chattoraj, D. K. Transition from a plasmid to a chromosomal mode of replication entails additional regulators. Proceedings of the National Academy of Sciences of the United States of America 108, 6199- 6204, doi:10.1073/pnas.1013244108 (2011).22 Venkova- Canova, T., Saha, A. & Chattoraj, D. K. A 29- mer site regulates transcription of the initiator gene as well as function of the replication origin of Vibrio cholerae chromosome II. Plasmid 67, 102- 110, doi:10.1016/j.plasmid.2011.12.009 (2012).23 de Lemos Martins, F., Fournes, F., Mazzuoli, M. V., Mazel, D. & Val, M. E. Vibrio cholerae chromosome 2 copy number is controlled by the methylation- independent binding of its monomeric initiator to the chromosome 1 crtS site. Nucleic acids research 46, 10145- 10156, doi:10.1093/nar/gky790 (2018).24 Bailey, T. L., Johnson, J., Grant, C. E. & Noble, W. S. The MEME Suite. Nucleic acids research 43, W39- 49, doi:10.1093/nar/gkv416 (2015).25 Menikpurge, I. P., Woo, K. & Mera, P. E. Transcriptional Activity of the Bacterial Replication Initiator DnaA. Frontiers in microbiology 12, 662317, doi:10.3389/fmicb.2021.662317 (2021).26 Venkova- Canova, T., Baek, J. H., Fitzgerald, P. C., Blokesch, M. & Chattoraj, D. K. Evidence for two different regulatory mechanisms linking replication and segregation of vibrio cholerae chromosome II. PLoS genetics 9, e1003579, doi:10.1371/journal.pgen.1003579 (2013).27 Val, M. E. et al. Fuse or die: how to survive the loss of Dam in Vibrio cholerae. Molecular microbiology 91, 665- 678, doi:10.1111/mmi.12483 (2014).28 Benureau, Y. et al. Method combining BAC film and positive staining for the characterization of DNA intermediates by dark- field electron microscopy. Biol Methods Protoc 5, bpaa012, doi:10.1093/biomethods/bpaa012 (2020).29 Jha, J. K., Ghirlando, R. & Chattoraj, D. K. Initiator protein dimerization plays a key role in replication control of Vibrio cholerae chromosome 2. Nucleic acids research 42, 10538- 10549, doi:10.1093/nar/gku771 (2014).30 Jha, J. K. et al. The DnaK Chaperone Uses Different Mechanisms To Promote and Inhibit Replication of Vibrio cholerae Chromosome 2. mBio 8, doi:10.1128/mBio.00427- 17 (2017).31 Holm, L., Laiho, A., Toronen, P. & Salgado, M. DALI shines a light on remote homologs: One hundred discoveries. Protein science : a publication of the Protein Society 32, e4519, doi:10.1002/pro.4519 (2023).32 Das, D. et al. The structure of the first representative of Pfam family PF09836 reveals a two- domain organization and suggests involvement in transcriptional regulation. Acta Crystallogr Sect F Struct Biol Cryst Commun 66, 1174- 1181, doi:10.1107/S1744309109022672 (2010).33 Karimova, G., Pidoux, J., Ullmann, A. & Ladant, D. A bacterial two- hybrid system based on a reconstituted signal transduction pathway. Proceedings of the National Academy of Sciences of the United States of America 95, 5752- 5756, doi:10.1073/pnas.95.10.5752 (1998).34 Ramachandran, R., Ciaccia, P. N., Filsuf, T. A., Jha, J. K. & Chattoraj, D. K. Chromosome 1 licenses chromosome 2 replication in Vibrio cholerae by doubling the crtS gene dosage. PLoS genetics 14, e1007426, doi:10.1371/journal.pgen.1007426 (2018).35 Dalia, A. B. & Dalia, T. N. Spatiotemporal Analysis of DNA Integration during Natural Transformation Reveals a Mode of Nongenetic Inheritance in Bacteria. Cell 179, 1499- 1511 e1410, doi:10.1016/j.cell.2019.11.021 (2019).36 Ferullo, D. J., Cooper, D. L., Moore, H. R. & Lovett, S. T. Cell cycle synchronization of Escherichia coli using the stringent response, with fluorescence labeling assays for DNA content and replication. Methods 48, 8- 13, doi:10.1016/j.ymeth.2009.02.010 (2009).37 Kemter, F. S. et al. Synchronous termination of replication of the two chromosomes is an evolutionary selected feature in Vibrianceae. PLoS genetics 14, e1007251, doi:10.1371/journal.pgen.1007251 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 886, 900]]<|/det|> +38 Kemter, F. S. et al. Stringent response leads to continued cell division and a temporal restart of DNA replication after initial shutdown in Vibrio cholerae. Molecular microbiology 111, 1617- 1637, doi:10.1111/mmi.14241 (2019).39 Molina- Garcia, L. et al. Functional amyloids as inhibitors of plasmid DNA replication. Scientific reports 6, 25425, doi:10.1038/srep25425 (2016).40 McCahern, M. J., Bott, M. A., Tooker, P. A. & Helinski, D. R. Negative control of plasmid R6K replication: possible role of intermolecular coupling of replication origins. Proceedings of the National Academy of Sciences of the United States of America 86, 7942- 7946, doi:10.1073/pnas.86.20.7942 (1989).41 Zaman, S. & Bastia, D. Oligomeric initiator protein- mediated DNA looping negatively regulates plasmid replication in vitro by preventing origin melting. Molecular cell 20, 833- 843, doi:10.1016/j.molcel.2005.10.037 (2005).42 Ciaccia, P. N., Ramachandran, R. & Chattoraj, D. K. A Requirement for Global Transcription Factor Lrp in Licensing Replication of Vibrio cholerae Chromosome 2. Frontiers in microbiology 9, 2103, doi:10.3389/fmicb.2018.02103 (2018).43 Doan, A. et al. The replication enhancer crt5 depends on transcription factor Lrp for modulating binding of initiator RctB to ori2 of Vibrio cholerae. Nucleic acids research, doi:10.1093/nar/gkad1111 (2023).44 Le Roux, F., Binesse, J., Saulnier, D. & Mazel, D. Construction of a Vibrio splendidus mutant lacking the metalloprotease gene vsm by use of a novel counterselectable suicide vector. Applied and environmental microbiology 73, 777- 784, doi:10.1128/AEM.02147- 06 (2007).45 Val, M. E., Skovgaard, O., Ducos- Galand, M., Bland, M. J. & Mazel, D. Genome engineering in Vibrio cholerae: a feasible approach to address biological issues. PLoS genetics 8, e1002472, doi:10.1371/journal.pgen.1002472 (2012).46 Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. Nature methods 6, 343- 345, doi:10.1038/nmeth.1318 (2009).47 Diaz, R. E., Sanchez, A., Anton Le Berre, V. & Bouet, J. Y. High- Resolution Chromatin Immunoprecipitation: ChIP- Sequencing. Methods in molecular biology 1624, 61- 73, doi:10.1007/978- 1- 4939- 7098- 8_6 (2017).48 Madic, J. et al. Three- color crystal digital PCR. Biomolecular detection and quantification 10, 34- 46, doi:10.1016/j.bdq.2016.10.002 (2016).49 Lo Scrudato, M. & Blokesch, M. The regulatory network of natural competence and transformation of Vibrio cholerae. PLoS genetics 8, e1002778, doi:10.1371/journal.pgen.1002778 (2012).50 Terwilliger, T. C. et al. phenix.mr_rosetta: molecular replacement and model rebuilding with Phenix and Rosetta. J Struct Funct Genomics 13, 81- 90, doi:10.1007/s10969- 012- 9129- 3 (2012).51 Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine. Acta crystallographica. Section D, Biological crystallography 68, 352- 367, doi:10.1107/S0907444912001308 (2012).52 Emsley, P. & Cowtan, K. Coot: model- building tools for molecular graphics. Acta crystallographica. Section D, Biological crystallography 60, 2126- 2132, doi:10.1107/S0907444904019158 (2004).53 Smart, O. S. et al. Exploiting structure similarity in refinement: automated NCS and target- structure restraints in BUSTER. Acta crystallographica. Section D, Biological crystallography 68, 368- 380, doi:10.1107/S0907444911056058 (2012).54 Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nature methods 14, 331- 332, doi:10.1038/nmeth.4193 (2017).55 Rohou, A. & Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. Journal of structural biology 192, 216- 221, doi:10.1016/j.jsb.2015.08.008 (2015).56 Bepler, T. et al. Positive- unlabeled convolutional neural networks for particle picking in cryo- electron micrographs. Nature methods 16, 1153- 1160, doi:10.1038/s41592- 019- 0575- 8 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 884, 435]]<|/det|> +57 Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nature methods 14, 290- 296, doi:10.1038/nmeth.4169 (2017).58 David, A. et al. The two Cis- acting sites, parS1 and oriC1, contribute to the longitudinal organisation of Vibrio cholerae chromosome I. PLoS genetics 10, e1004448, doi:10.1371/journal.pgen.1004448 (2014).59 Woldringh, C. L., Hansen, F. G., Vischer, N. O. & Atlung, T. Segregation of chromosome arms in growing and non- growing Escherichia coli cells. Frontiers in microbiology 6, 448, doi:10.3389/fmicb.2015.00448 (2015).60 Ducret, A., Quardokus, E. M. & Brun, Y. V. MicrobeJ, a tool for high throughput bacterial cell detection and quantitative analysis. Nature microbiology 1, 16077, doi:10.1038/nmicrobiol.2016.77 (2016).61 Battesti, A. & Bouveret, E. The bacterial two- hybrid system based on adenylate cyclase reconstitution in Escherichia coli. Methods 58, 325- 334, doi:10.1016/j.ymeth.2012.07.018 (2012).62 Yu, Y., Ouyang, Y. & Yao, W. shinyCircos: an R/Shiny application for interactive creation of Circos plot. Bioinformatics 34, 1229- 1231, doi:10.1093/bioinformatics/btx763 (2018).63 Matthey, N., Drebes Dorr, N. C. & Blokesch, M. Long- Read- Based Genome Sequences of Pandemic and Environmental Vibrio cholerae Strains. Microbiology resource announcements 7, doi:10.1128/MRA.01574- 18 (2018).64 Heidelberg, J. F. et al. DNA sequence of both chromosomes of the cholera pathogen Vibrio cholerae. Nature 406, 477- 483, doi:10.1038/35020000 (2000). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 382, 230]]<|/det|> +Supplementarymovie1.mp4 Supplementarymovie2.mp4 Supplementarymovie3.mp4 NiaultSupplinformationFigures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/images_list.json b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..20c7aeb759f3ee83964d0fff8b46793681c7b1da --- /dev/null +++ b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 CryoSieve is capable of maintaining resolutions after removing the majority of particles in the final stacks. For all six experimental datasets, density maps of the CryoSieve-retained particles (steel blue) and all particles in the final stack (medium purple) were compared, obtained from CryoSPARC's ab initio reconstruction after discarding the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein-from-noise effect. The density maps were first FSC-weighted (based on FSCs given by CryoSPARC), and then B-factor sharpened using equivalent B-factors for the same protein: \\(-90\\mathrm{\\AA}^{2}\\) for TRPA1, \\(-180\\mathrm{\\AA}^{2}\\) for hemagglutinin, \\(-100\\mathrm{\\AA}^{2}\\) for LAT1, \\(-60\\mathrm{\\AA}^{2}\\) for pfCRT, \\(-70\\mathrm{\\AA}^{2}\\) for TSHR-Gs, and \\(-80\\mathrm{\\AA}^{2}\\) for TRPM8. The central bars indicate the proportions of the retained and removed particles.", + "footnote": [], + "bbox": [ + [ + 156, + 82, + 841, + 469 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 CryoSieve outperformed other algorithms in terms of FSC-based resolutions and Q-scores. We compared the density maps reconstructed from retained particles obtained by CryoSieve (indigo) and NCC (green), along with random (orange) as the baseline, at different retention ratios. Density maps were ab initio reconstructed by CryoSPARC after discarding the published refined Euler angles deposited on EMPIAR. Four metrics were employed for measuring density map quality: FSC-based resolutions, including model-to-map (solid lines with squares) and two-half-maps (solid lines with diamonds) resolutions, are shown in the first column; meanwhile, Q-scores of the raw maps (dashed lines with circles) and the sharpened maps (solid lines with circles) are shown in the second column, as Q-score varies by the B-factor used for sharpening. B-factors for sharpening were self-determined by CryoSPARC. The iterations where CryoSieve obtained the finest subset, determined by comprehending these metrics, are labeled with hatched bars.", + "footnote": [], + "bbox": [ + [ + 303, + 115, + 690, + 700 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 The two-dimensional resolution distribution between retained and removed particles was compared. The third iteration of CryoSieve achieved a retention ratio of \\(51.2\\%\\) and a removal ratio of \\(48.8\\%\\) , resulting in a similar number of particles for retention and removal. These two categories underwent CryoSPARC 2D clustering and averaging, i.e., 2D classification, with the number of 2D classes set to 50. All six experimental datasets were tested. CryoSPARC reported the 2D resolution of each 2D class, along with the number of particle images belonging to it. We statistically analyzed the number of particles belonging to each 2D resolution and plotted histograms, demonstrating the difference between retained (steel blue) and removed (crimson) particles in terms of 2D resolution distribution.", + "footnote": [], + "bbox": [ + [ + 160, + 277, + 840, + 568 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 CryoSieve prioritizes the removal of radiation-damaged particles. (a), InSilicoTEM generated simulated particle images with different levels of radiation damage. The parameter \\(M_{b}\\) in InSilicoTEM, ranging from 0 to 4, mimicked different levels of radiation damage, with higher \\(M_{b}\\) leading to weaker high-frequency signals and blurrier particles (clean, upper row). Noise, also generated by InSilicoTEM through physical process simulation, was then added to the clean particles to obtain noisy particles for testing (noisy, lower row). (b), The proportions of particles with varying levels of radiation damage (distinguished by colors) in retained particles at different retention ratios (labeled on the left) are shown. The retained particles obtained by CryoSieve (left horizontal bars) and NCC (right horizontal bars) were compared. NCC performed acceptably for particles with high radiation damage, but was unable to distinguish particles with relatively low radiation damage. Meanwhile, CryoSieve sequentially sieved out particles from high to low radiation damage. The accuracy (the proportion of zero radiation damage particles) of CryoSieve was 91.7% at iteration 7, while that of NCC was only 61.5%. (c), Side chains of density maps reconstructed by cryoSPARC using retained particles obtained by CryoSieve (indigo) and NCC (green) were compared. The atomic model (PDB 6PCQ) was fitted into the density maps. The two reconstruction maps are compared at an identical contour. Red arrows emphasize differences between the two maps. (d) Model-to-map FSCs of reconstructed density maps using retained particles were compared, with retention ratio of 21.0% (iteration 7). Retention particles were obtained by CryoSieve (indigo), NCC (green) and random (baseline method, orange), respectively. The FSC threshold (FSC = 0.5) was depicted as a horizontal dashed line.", + "footnote": [], + "bbox": [ + [ + 150, + 106, + 845, + 565 + ] + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff.mmd b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a0e33e2fd7c7be7f3ee31005167fd0e3827acfc0 --- /dev/null +++ b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff.mmd @@ -0,0 +1,367 @@ + +# Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM + +Jianying Zhu Tsinghua university + +Qi Zhang Tsinghua university + +Hui Zhang Yanqi Lake Beijing Institue of Mathematical Sciences and Applications + +Zuoqiang Shi Tsinghua university + +Mingxu Hu Tsinghua University https://orcid.org/0000- 0003- 3603- 3966 Chenglong Bao ( \(\square\) clbao@tsinghua.edu.cn ) Tsinghua University https://orcid.org/0000- 0002- 1201- 1212 + +## Article + +Keywords: cyro- EM, single particle analysis, particle sorting, final stacks + +Posted Date: May 29th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2921474/v1 + +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on December 10th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43555- x. + +<--- Page Split ---> + +# Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM + +Jianying Zhu \(^{1\dagger}\) , Qi Zhang \(^{3,4,5,6\dagger}\) , Hui Zhang \(^{2}\) , Zuoqiang Shi \(^{1,2*}\) , Mingxu Hu \(^{3,4,5,6*}\) and Chenglong Bao \(^{1,2*}\) + +\(^{1}\) Yau Mathematical Sciences Center, Tsinghua University, Beijing, China. \(^{2}\) Yanqi Lake Beijing Institute of Mathematical Sciences and Applications, Beijing, China. \(^{3}\) Key Laboratory of Protein Sciences (Tsinghua University), Ministry of Education, Beijing, China. \(^{4}\) School of Life Science, Tsinghua University, Beijing, China. \(^{5}\) Beijing Advanced Innovation Center for Structural Biology, Beijing, China. \(^{6}\) Beijing Frontier Research Center for Biological Structure, Beijing, China. + +\*Corresponding author(s). E- mail(s): zqshi@tsinghua.edu.cn; humingxu@mail.tsinghua.edu.cn; clbao@tsinghua.edu.cn; †These authors contributed equally to this work. + +## Abstract + +Cryo- electron microscopy (cryo- EM) is widely used to determine near- atomic resolution structures of biological macromolecules. Due to the extremely low signal- to- noise ratio, cryo- EM relies on averaging many images. However, a crucial question in the field of cryo- EM remains unanswered: how close can we get to the minimum number of particles required to reach a specific resolution in practice? The absence of an answer to this question has impeded progress in understanding sample behavior and the performance of sample preparation methods. To address this issue, we have developed a new iterative particle sorting and/or sieving method called CryoSieve. Extensive experiments demonstrate that CryoSieve outperforms other cryo- EM particle sorting algorithms and reveals that most particles are unnecessary in final + +<--- Page Split ---> + +stacks. The minority of particles remaining in the final stacks yield superior high- resolution amplitude in reconstructed density maps. For some datasets, the size of the finest subset approaches the theoretical limit. + +Keywords: cyro- EM, single particle analysis, particle sorting, final stacks + +## 1 Introduction + +The transformative impact of cryo- EM single- particle analysis (SPA) on the field of structural biology has been widely recognized by the scientific community [1]. Cryo- EM has advanced significantly due to a series of technological innovations [2- 7], enabling the technique to provide macromolecular structures with up to atomic resolution at an unprecedented rate. This technological progress is commonly referred to as the resolution revolution [8]. Cryo- EM involves using electron microscopy images of biomolecules embedded in vitreous, glass- like ice [9], which are then combined to generate three- dimensional (3D) density maps. These maps provide valuable insights into the function of macromolecules and their role in biological processes. + +The stability and electron- optical performance of electron microscopes do not hinder the use of cryo- EM [10]. However, biological samples studied in cryo- EM are radiation- sensitive [11, 12]. Therefore, a trade- off must be made between improving the signal- to- noise ratio (SNR) and limiting radiation damage [13, 14]. It was concluded that statistically well- defined three- dimensional (3D) structures could not be obtained from individual biological macromolecules at atomic resolution [15, 16]. Instead, increasing the SNR by averaging image data from many identical macromolecules is the only way to progress [13, 17, 18]. Over two decades ago, Henderson estimated that structures could be determined at a resolution of near 3 Å by merging data from approximately 12,000 particles, even for particles as small as approximately 40 kDa [19]. Later, Rosenthal and Henderson argued that the electron microscopy community should adopt the same threshold criterion for structure factor quality as the X- ray protein crystallography community, which was set at a figure- of- merit of 0.5 corresponding to a phase error of 60 degrees [16]. The theoretical limit of the minimum number of particle images required to achieve a specific resolution can be calculated using the theory proposed by Henderson and Rosenthal [16, 19], given the B- factor of the instrument (e.g., electron microscopy and camera) [13, 14, 20]. In practice, the final stacks of cryo- EM still far fall short of the theoretical limit, indicating a considerable gap between what can be accomplished and the physical limit of what cryo- EM can do [21]. The initial particle datasets obtained by particle picking from micrographs undergo multiple rounds of laborious 2D and 3D classification to generate the final stack for model determination. The final stacks, which yield atomic or sub- atomic resolution density maps, typically comprise several orders of magnitude fewer particles than the original datasets. Therefore, the + +<--- Page Split ---> + +cryo- EM field faces the long- standing question of how close we can approach the theoretical limit in practice. The lack of an answer to this open question has hindered the quantification of the performance of various underdeveloped sample preparation methods and impeded the investigation of trends and the understanding of the underlying mechanisms of sample behavior. + +To answer the question of how close cryo- EM can approach its theoretical limit, it is crucial to determine the minimum number of particles required to achieve a high- resolution 3D reconstruction within a given dataset. In this study, we introduce CryoSieve, an iterative particle sorting and/or sieving algorithm that identifies the smallest subset of particles necessary to generate high- resolution density maps, which we call the finest subset. CryoSieve compares the high- frequency components of synthetic and observed particle images. A higher CryoSieve score indicates superior quality rather than typical cryo- EM damage or artifacts (Section 2.1). Extensive experiments show that CryoSieve outperforms other particle sorting algorithms in various metrics and reveals that most particles in final stacks are futile (Section 2.2). The finest subsets generate 3D density maps with better high- resolution amplitude, using much fewer particles than the final stacks (Section 2.3). We propose that CryoSieve removes radiation- damaged particles within cryo- EM datasets, supported by experiments on simulated radiation damage particles (Section 2.4). Lastly, we compare the minimum particles required in theory with the size of the finest subsets obtained by CryoSieve, finding that some datasets come close to the theoretical limit after being sieved by CryoSieve (Section 2.5). We conclude that, for these datasets, the primary opportunity for further improvement lies in generating fewer futile particles during sample preparation rather than further improving the quality of the particle images that constitute the finest subset. + +## 2 Results + +### 2.1 Design of CryoSieve. + +We have developed a particle sorting and/or sieving model called CryoSieve that alternatively performs 3D reconstruction and particle selection, eliminating futile particles during each iteration. In CryoSieve, a cryo- EM single- particle reconstruction software selected by the user is used to reconstruct a new density map with the retained particle images, which is then used in the subsequent iteration. The retained particle images in each iteration form a subset of those from the previous iteration, as shown in the following formula: + +\[\left\{i_{1}^{(k)},i_{2}^{(k)}\dots ,i_{n^{(k)}}^{(k)}\right\} \subset \left\{i_{1}^{(k - 1)},i_{2}^{(k - 1)},\dots ,i_{n^{(k - 1)}}^{(k - 1)}\right\} ,\] + +where \(n^{(k - 1)}\) represents the number of retained particles. At each iteration, let \(b_{j}\) be the \(j\) - th particle image, \(A_{j}\) be its forward operator defined by the estimated parameters and \(x^{(k - 1)}\) be the reconstructed density map from the + +<--- Page Split ---> + +retained particle images in the previous iteration, particles are sieved out based on their CryoSieve score, which is defined as follows: + +\[g_{j}:= \left\| H^{(k)}b_{j}\right\|_{2}^{2} - \left\| H^{(k)}(b_{j} - A_{j}x^{(k - 1)})\right\|_{2}^{2},\quad j\in \left\{i_{1}^{(k - 1)},i_{2}^{(k - 1)},\dots ,i_{n^{(k - 1)}}^{(k - 1)}\right\} .\] + +Here, \(H^{(k)}\) is the high- pass operator at the \(k\) - th iteration, and its threshold frequency increases as the iteration progresses. + +The CryoSieve score estimates the similarity between a particle and a reference projection above a given frequency. A higher CryoSieve score indicates that the particle and the reference projection share a higher proportion of signal energy, indicating better particle quality. As radiation damage mainly affects the high- frequency range, the CryoSieve score includes a high- pass operator to extract the high- frequency part. We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high- frequency range through theoretical analysis and simulation verification. Assuming that noise in particles follows a Gaussian distribution, we have shown that, with high probability, the CryoSieve score is an ideal indicator of particle image quality, distinguishing it from typical cryo- EM damage or artifacts (Supplementary Material I). Furthermore, when simulating radiation damage as high- frequency random phasing, the CryoSieve score exhibits remarkable accuracy in selecting particles even with a very low signal- to- noise ratio (approximately 0.001), achieving a precision rate of around 90% (Supplementary Material I). + +### 2.2 Majority of the particles are futile in final stacks. + +We demonstrate the versatility of our method by applying it to six experimental datasets (Table 1). The first dataset is derived from the human TRPA1 ion channel (EMPIAR- 10024) [22]. The second dataset is from influenza hemagglutinin trimer (EMPIAR- 10097) [23], of which the preferred orientation necessitated 40- degree tilts during data acquisition. The third dataset involves LAT1- CD98hc bound to MEM- 108 Fab (EMPIAR- 10264) [24], while the fourth features membrane- bound pFCRT complexed with Fab (EMPIAR- 10330) [25]. Both of these datasets utilized signal subtraction during data processing. The fifth dataset is from CS- 17 Fab- bound TSHR- Gs (EMPIAR- 11120) [26], and the sixth is from TRPM8 bound to calcium (EMPIAR- 11233) [27]. All datasets were obtained using a voltage of 300 kV and an amplitude contrast of 0.07 or 0.1. The TEM systems and electron detectors used in the experiments are listed in Table 1, along with additional metadata such as the number of particles in the final stacks, spherical aberration and molecular symmetry. + +All of the datasets are deposited in the Electron Microscopy Public Image Archive (EMPIAR) as final stacks. These final stacks, which also contain + +<--- Page Split ---> + +Table 1 Microscopic imaging parameters of six experimental datasets along with their associated metadata. + +
datasetTEMelectron
detector		number of
particles		spherical
aberration (mm)		symmetry
TRPA1TF30 PolaraGatan K2 Summit43,5852.0\(C_{4}\)
hemagglutininTitan KriosGatan K2 Summit130,0002.7\(C_{3}\)
LAT1Titan KriosFEI FALCON III250,7122.7\(C_{1}\)
pfCRTTitan KriosGatan K2 Qutuamn16,9050.001\(C_{1}\)
TSHR-GsTitan KriosGatan K3 Qutuamn41,0542.7\(C_{1}\)
TRPM8Titan KriosGatan K2 Summit42,0402.6\(C_{4}\)
+ +Springer Nature 2021 LXFEX template + +<--- Page Split ---> + +the corresponding refined Euler angles, were used to generate the final published reconstructions. The final stacks are generated by manually selecting significantly smaller subsets through multiple rounds of 2D/3D classification, resulting in a substantial reduction in the number of particles compared to the original particle stacks. For example, the final stack of the CS- 17 Fab- bound TSHR- Gs dataset consists of 41,054 particles selected from a pool of 2,742,080 particles, representing a reduction by a factor of approximately 67 [26]. + +We employed CryoSieve to process the six experimental datasets. CryoSieve removed \(20\%\) of the particles in each iteration, resulting in a retaining ratio of \(80.0\%\) , \(64.0\%\) , \(51.2\%\) , and so on. The retained particles in different iterations were then used for ab initio reconstruction to determine the finest subset of particles. The finest subset only contained \(26.2\%\) to \(32.8\%\) of the particles in the final stack. However, the quality of the reconstructed map from the finest subset was consistent with that obtained from all particles in the final stack, as demonstrated in Figure 1. These results indicate that CryoSieve can effectively eliminate over half of the particles with unreliable high- frequency signals without negatively affecting the final reconstruction. Therefore, CryoSieve is highly effective in selecting the most informative particles. + +We performed a comparative analysis of CryoSieve with other cryo- EM particle sorting criteria or software currently used in the field, including the normalized cross- correlation (NCC) method [28] and the angular graph consistency (AGC) approach [29]. In our experiments, we used final stacks composed of relatively high- quality particles. NCC retains an equal number of particles compared to CryoSieve at each iteration, while AGC's retaining ratio is self- determined. However, AGC's retaining ratio was mainly over \(90\%\) , resulting in only a small fraction of particles being removed. Thus the quality of the reconstructed map using the retained particles did not improve or worsen (Supplementary Table 1), as these tested final stacks are composed of relatively high- quality particles. Additionally, we randomly selected the same number of particles from the tested final stacks at each iteration to observe the baseline effect of particle number reduction. + +For all the aforementioned methods (CryoSieve, NCC, AGC, and random), we discarded the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein- from- noise effect [30]. Instead, the retained particles were used for ab initio reconstruction by CryoSPARC to obtain refreshed sets of Euler angles and density maps. Several metrics, including FSC- based resolution [16] and Q- score [31], were used to measure the quality of the refreshed density maps. Comprehending these metrics, our analyses reveal that CryoSieve effectively sieves out \(67.2\%\) to \(73.8\%\) (varying based on datasets) of particles from the final stacks without deteriorating the yielded density maps (Figure 2). Meanwhile, subsets of equal size retained by the other methods failed to reconstruct density maps of the same quality as the original (Figure 2). Therefore, CryoSieve significantly outperforms other particle sorting algorithms, demonstrating that majority of the particles are dispensable in the final stacks. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 CryoSieve is capable of maintaining resolutions after removing the majority of particles in the final stacks. For all six experimental datasets, density maps of the CryoSieve-retained particles (steel blue) and all particles in the final stack (medium purple) were compared, obtained from CryoSPARC's ab initio reconstruction after discarding the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein-from-noise effect. The density maps were first FSC-weighted (based on FSCs given by CryoSPARC), and then B-factor sharpened using equivalent B-factors for the same protein: \(-90\mathrm{\AA}^{2}\) for TRPA1, \(-180\mathrm{\AA}^{2}\) for hemagglutinin, \(-100\mathrm{\AA}^{2}\) for LAT1, \(-60\mathrm{\AA}^{2}\) for pfCRT, \(-70\mathrm{\AA}^{2}\) for TSHR-Gs, and \(-80\mathrm{\AA}^{2}\) for TRPM8. The central bars indicate the proportions of the retained and removed particles.
+ +cisTEM [5] is capable of reporting a score for each single particle image after 3D reconstruction, though it is not a particle sorting criterion. Due to differences in alignment and other image processing workflows between cisTEM and cryoSPARC, cisTEM cannot be strictly compared with CryoSieve. Therefore, we compared Cryosieve with cisTEM by sorting single particles by cisTEM score and retaining the same number of particle images for ab initio reconstruction using cryoSPARC. CryoSieve outperformed cisTEM in all six experimental datasets (Supplementary Material II and Supplementary Figure 2). + +Moreover, we analyzed the differences between the particle images retained and removed using CryoSieve by performing 2D classification of the particles into 50 classes using CryoSPARC. To ensure a comparable number of particles for both retained and removed groups, iteration 3 was chosen for such analysis, with a retention ratio of \(51.2\%\) and a removal ratio of \(48.8\%\) . CryoSPARC + +<--- Page Split ---> + +reported the 2D resolution of each class, along with the number of particle images belonging to it. The particles retained by CryoSieve (Figure 3, steel blue) were distributed at a higher resolution compared to those removed by it (Figure 3, crimson). In five out of the six datasets, particle images with the highest resolution, i.e., 8.5–9.6 Å in hemagglutinin, 6.6–8.2 Å in LAT1, 7.2–11.6 Å in pfCRT, 7.2–8.5 Å in TSGH-Gs, and 11.6–7.5 Å in TRPM8, were entirely retained by CryoSieve. Therefore, our analysis suggests that CryoSieve selectively retained the higher- quality particle images in the final stack while discarding lower- quality ones. + +### 2.3 Better high-resolution amplitude with much fewer particles. + +B- factors, also known as Debye- Waller factors or temperature factors, reflect the rate at which the amplitude of high- resolution information decreases [16]. Lower B- factors indicate that the high- resolution signal has been better preserved during sample preparation, imaging, and image processing, implying that the particle images are of higher quality. B- factors are widely used to measure image quality in cryo- EM quantitatively [32–36]. In our six experimental datasets, the finest subset, comprising only 26.2% to 32.8% of particles in the final stack, yields 3D density maps with B- factors reduced by \(18.0\mathring{\mathrm{A}}^{2}\) to \(71.2\mathring{\mathrm{A}}^{2}\) compared to those produced by the original final stacks (Table 2, column B and C). In other words, the density maps reconstructed from the finest subset have a better high- resolution amplitude, meaning they contain a greater high- resolution intensity, despite the fact that the finest subsets only contain a small fraction of particles in the final stack. This indicates that CryoSieve significantly reduced the temperature factor and alleviated the amplitude contrast decay, suggesting that high- quality particles contribute to the density map and can be effectively selected by CryoSieve. + +### 2.4 CryoSieve can effectively detect radiation-damaged particles. + +We hypothesize that some particle images in the final stacks have been subject to some degree of radiation damage and cannot be screened out by conventional methods. These particles do not contribute positively to the reconstructed density map. To verify the possibility of this conjecture, we generated simulated particle images using InSilicoTEM [37], with varying degrees of radiation damage, and screened them using CryoSieve. + +The atomic model of \(E\) . coli 50S ribosome bound to VM2 (PDB 6PCQ) [38] was employed to generate particle images, with defocus ranges from 1800 nm to 2800 nm, an electron dose of \(40 \mathrm{e}^{- \mathring{\mathrm{A}}^{- 2}}\) , a Falcon I detector, a box size of \(450 \times 450\) , and a pixel size of \(1.04\mathring{\mathrm{A}}\) . Radiation damage was simulated by varying the motion blur factor from 0 to 4 (Figure 4a). Each set of the simulated single particles contained 8,000 particles, and the projection directions followed a random distribution. + +<--- Page Split ---> + + +Table 2 The finest subsets alleviate high-resolution amplitude decay, along with a comparison to their theoretical number of particle limit. + +
datasetA†B*C*DEF
TRPA13.90141.978.1(63.8-)52111,426(51.9×)43,585(83.7×)
hemagglutinin3.62232.0160.8(71.2-)97534,078(35.6×)130,000(133.3×)
LAT13.11132.696.0(36.6-)6,69765,687(9.8×)250,712(37.4×)
pfCRT3.3785.149.5(35.6-)4,2124,429(1.01×)16,905(4.0×)
TSHR-Gs2.9692.961.7(31.2-)9,20513,465(1.46×)41,054(4.5×)
TRPM82.9894.776.7(18.0-)2,20013,789(6.3×)42,040(19.1×)
+ +\(^{\dagger}\) (A), half- maps resolution of the CryoSieve- retained particles (A); (B), B- factor obtained from all particles in the final stacks (A²); (C), B- factor obtained from the CryoSieve- retained particles with temperature decrease (compared with all particles) in brackets (A²); (D), theoretical number of particles limit at \(B = 50\mathrm{\AA}^{2}\) ; (E), number of the CryoSieve- retained particles with folds of theoretical limit in brackets.; (F), number of particles in final stacks with folds of theoretical limit in brackets. \* B- factors were reported by cryoSPARC auto- postprocessing. + +We compared the retention behavior of CryoSieve and NCC for particles with different simulated degrees of radiation damage, with random retention as the baseline. As the number of iterations increased, the retention rate decreased, and CryoSieve demonstrated a greater ability to screen out particles with higher levels of radiation damage (Figure 4b). While NCC outperformed the random retention baseline, it was still inferior to CryoSieve in terms of screening accuracy (Figure 4b). Additionally, we compared CryoSieve with cisTEM and found that although cisTEM performed acceptably for particles with high radiation damage, it still exhibited inferior performance to CryoSieve for particles with relatively low radiation damage. (Supplementary Material II and Supplementary Figure 2). These results imply that CryoSieve could effectively removes radiation- damaged particles to improve the quality of the final density maps. + +### 2.5 The finest subsets may be close to the theoretical number of particles limit. + +The theoretical number limit of particle images, given by Rosenthal and Henderson [16], is + +\[\mathrm{N}_{\mathrm{particles}} = \frac{1}{N_{\mathrm{asym}}}\frac{S^2}{N_e\sigma_e d}\exp \left(\frac{B}{2d^2}\right),\] + +where \(N_{\mathrm{asym}}\) , \(S\) , \(N_{e}\) , \(\sigma_{e}\) , \(d\) , \(B\) stand for the number of asymmetric units, the signal- to- noise threshold criteria of the resolution, the electron dose, the elastic cross- section for carbon, the resolution, and the overall temperature factor, respectively. In the above formula, \(\frac{S}{N} = \frac{1}{\sqrt{3}}\) , which is equivalent to phase error of 60 degrees or 0.143- threshold of half- maps FSC [16]. Meanwhile, \(N_{e} = 5e^{- \mathrm{\AA}^{- 2}}\) , which is believed to be the limiting dose due to radiation + +<--- Page Split ---> + +damage for features near- atomic resolution [16, 19, 39, 40]. The electron dose used in practice is typically a fold higher than the limiting dose. Although the additional dose does not contribute to the structure factor amplitudes at near atomic resolution, it may have increased the signal up to the resolution limit of the final map, thus making the determination of particle parameters easier [16]. This conjecture agrees with the observation in the study of micrograph movie stack dose weighting, which found that only the initial few frames, not the subsequent frames, contribute to near atomic features [41- 43]. Finally, \(\sigma_{e} = 0.004\mathrm{\AA}^{2}\) is the elastic cross- section for carbon at \(300\mathrm{kV}\) [44]. + +The overall temperature factor, or Rosenthal and Henderson's B- factor, is the dominant factor in estimating the theoretical limit. Here, we proposed a simplified and crude assumption that limits only exist on instruments (TEM and electron detector) and that no other resolution- limiting factors exist. In other words, we assumed that all other procedures or techniques were ideal. For example, vitrified non- amorphous ice is perfectly flat and of ideal thickness, there is no beam- induced motion, and orientations of particles follow a uniform distribution, and there is no electron- charging effect. Therefore, B- factor represents a summary of all resolution- limiting factors of a given electron microscope, and describes the overall quality of the instrumental setup. Holger Stark and his colleagues have summarized the current knowledge on existing state- of- the- art commercial EM hardware and their B- factors [45]. For the standard Titan Krios, they concluded that its B- factor is \(50\mathrm{\AA}^{2}\) , which was determined by re- evaluating data from EMPIAR- 10216 as described by [46], with modifications to account for off- axial aberrations by splitting the micro- graphs into nine subsets [47]. Therefore, we computed the theoretical number of particle limits at \(B = 50\mathrm{\AA}^{- 2}\) (Table 2, column D). The sizes of the finest subsets obtained by CryoSieve were compared with such theoretical limits (Table 2, column E). + +Out of the six datasets examined, two (pfCRT and TSHR- Gs) were found to be close to their theoretical limits (Table 2, column E, emphasized by bold font). However, the TRPA1 dataset fell short of the theoretical limit by approximately 52. This could be due to the lower resolution capabilities of the TF30 Polara TEM used in the study compared to more advanced models like the Titan Krios. It is possible that the assumed B- factor of \(50\mathrm{\AA}^{2}\) for the TF30 Polara is relatively low and does not accurately reflect the properties of the TEM. Additionally, the sample preparation techniques used in 2015 when the TRPA1 study was conducted may not have been fully optimized to achieve the highest resolution possible. Hemagglutinin also fell short of the theoretical limit by a factor of approximately 36 due to using a tilt- collection strategy to account for preferred orientation, resulting in a larger effective ice thickness and a decrease in the quality of particle images. Finally, LAT1 and TRPM8 fell short of the theoretical limit by a factor of 9.8 and 6.3, respectively, indicating that there is still room for improvement in sample preparation for these datasets. + +<--- Page Split ---> + +## 3 Discussion + +In this study, we introduced the CryoSieve algorithm, which has the ability to estimate the minimum number of particles in a dataset, referred to as the finest subset. CryoSieve demonstrated that most particles in the final stacks are superfluous and do not contribute to reconstructing density maps. On the other hand, the minority of particles that remain in the final stacks yields superior high- resolution amplitude. We also discovered that for some datasets, the size of the finest subset comes close to the theoretical limit. Therefore, CryoSieve can, to some degree, provide insight into a long- standing question in the cryo- EM field: how close can we approach the theoretical limit in practice? + +Despite technical advancements that have made cryo- EM more accessible to structural biologists, sample preparation remains a major bottleneck in the workflow. Scientists and engineers are thus focusing their efforts on this challenge [48]. In single- particle analysis (SPA), sample preparation consists of two crucial steps: sample optimization and grid preparation. The former involves purifying the specimen while maintaining its optimal biochemical state. The latter entails preparing the sample for analysis in the microscope, including chemical or plasma treatment of the grid, sample deposition, and vitrification. Numerous techniques have been proposed to address macromolecular instability, but the efficacy of one approach over another depends on the sample's characteristics [48, 49]. Currently, grid preparation results are heavily influenced by the user's expertise and experience, which can make the process time- consuming and challenging [21, 50]. The numerous variables encountered in sample and grid preparation pose challenges in establishing cause- and- effect relationships, as researchers can only assess the sample at the molecular level using the microscope. As a result, quantitative statistics from comparisons of different sample and grid preparation protocols are still lacking, and a systematic approach is necessary to investigate trends and comprehend the fundamental mechanisms of sample behavior [51]. + +CryoSieve has the potential to establish a metric for the quantitative evaluation of various sample preparation techniques by measuring image quality based on the gaps between the theoretical limits and the size of the finest subsets. One of the possible future directions is to address the variables encountered during sample and grid preparation and establish cause- and- effect relationships. Resolving these issues, among others, cryo- EM could become a more versatile and an even more influential technology in structural biology, potentially addressing new research questions and aiding the growth of novel methodologies as the field advances [52]. + +## Acknowledgements + +This work was supported by the National Key R&D Program of China (No.2021YFA1001300) (to C.B.), the National Natural Science Foundation of China (No.12271291) (to C.B.), the Advanced Innovation Center for Structural Biology (to M.H.), the Beijing Frontier Research Center for Biological + +<--- Page Split ---> + +Structure (to M.H.), and the National Natural Science Foundation of China (No.12071244) (to Z.S.). We also would like to express our gratitude to Shouqing Li and Ranhao Zhang for generously sharing their expertise and experiences in particle selection and density map reconstruction in Cryo- EM. + +## Data availability + +The raw final stack datasets analyzed in this study were downloaded from the EMPIAR repository (EMPIAR- 10024, EMPIAR- 11233, EMPIAR- 10097, EMPIAR- 11120, EMPIAR- 10264, EMPIAR- 10330). Atomic coordinates from Protein Data Bank 6PCQ were used for the generation of simulated particles using InSilicoTEM. The finest subsets of these six experimental datasets obtained by CryoSieve can be found as Supplementary Material III. + +## Code availability + +CryoSieve will be open- source upon publication and is also available upon request during the review process. + +## Contribution + +C.B., M.H., and Z.S. initiated the project. M.H., Q.Z., and J.Z. developed CryoSieve and carried out testing. H.Z. provided support in using InSilicoTEM. J.Z. and M.H. analyzed the data. M.H., J.Z., and C.B. wrote the manuscript. + +## Competing interests + +All other authors declare no competing interests. + +## References + +[1] Nogales, E.: The development of cryo- EM into a mainstream structural biology technique. Nature Methods 13(1), 24- 27 (2016). https://doi.org/10.1038/nmeth.3694. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +[2] Bai, X.- c., Fernandez, I.S., McMullan, G., Scheres, S.H.: Ribosome structures to near- atomic resolution from thirty thousand cryo- EM particles. eLife 2, 00461 (2013). https://doi.org/10.7554/eLife.00461. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 06 + +[3] Campbell, M.G., Cheng, A., Brilot, A.F., Moeller, A., Lyumkis, D., Veesler, D., Pan, J., Harrison, S.C., Potter, C.S., Carragher, B., Grigorieff, N.: Movies of Ice- Embedded Particles Enhance Resolution in Electron + +<--- Page Split ---> + +Cryo- Microscopy. Structure 20(11), 1823- 1828 (2012). https://doi.org/ 10.1016/j.str.2012.08.026. Accessed 2023- 04- 06 + +[4] Li, X., Mooney, P., Zheng, S., Booth, C.R., Braunfeld, M.B., Gubbens, S., Agard, D.A., Cheng, Y.: Electron counting and beam- induced motion correction enable near- atomic- resolution single- particle cryo- EM. Nature Methods 10(6), 584- 590 (2013). https://doi.org/10.1038/nmeth.2472. Number: 6 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +[5] Grant, T., Rohou, A., Grigorieff, N.: cisTEM, user- friendly software for single- particle image processing. eLife 7, 35383 (2018). https://doi.org/ 10.7554/eLife.35383. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 06 + +[6] Scheres, S.H.W.: RELION: Implementation of a Bayesian approach to cryo- EM structure determination. Journal of Structural Biology 180(3), 519- 530 (2012). https://doi.org/10.1016/j.jsb.2012.09.006. Accessed 2023- 04- 06 + +[7] Punjani, A., Rubinstein, J.L., Fleet, D.J., Brubaker, M.A.: cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nature Methods 14(3), 290- 296 (2017). https://doi.org/10.1038/nmeth. 4169. Number: 3 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +[8] Kühlbrandt, W.: The Resolution Revolution. Science 343(6178), 1443- 1444 (2014). https://doi.org/10.1126/science.1251652. Publisher: American Association for the Advancement of Science. Accessed 2022- 11- 19 + +[9] Dubochet, J., McDowall, A.W.: Vitrification of pure water for electron microscopy. Journal of Microscopy 124(DEC), 3- 4 (1981). Publisher: Wiley- Blackwell. Accessed 2023- 04- 06 + +[10] Glaeser, R.M.: How Good Can Single- Particle Cryo- EM Become? What Remains Before It Approaches Its Physical Limits? Annual Review of Biophysics 48(1), 45- 61 (2019). https://doi.org/10.1146/annurev- biophys- 070317- 032828. - eprint: https://doi.org/10.1146/annurev- biophys- 0703- 032828. Accessed 2022- 12- 22 + +[11] Baker, L.A., Smith, E.A., Bueler, S.A., Rubinstein, J.L.: The resolution dependence of optimal exposures in liquid nitrogen temperature electron cryomicroscopy of catalase crystals. Journal of Structural Biology 169(3), 431- 437 (2010). https://doi.org/10.1016/j.jsb.2009.11.014. Accessed 2023- 04- 06 + +[12] Bammes, B.E., Jakana, J., Schmid, M.F., Chiu, W.: Radiation damage + +<--- Page Split ---> + +effects at four specimen temperatures from 4 to 100K. Journal of Structural Biology 169(3), 331- 341 (2010). https://doi.org/10.1016/j.jsb.2009.11.001. Accessed 2023- 04- 06 + +[13] Glaeser, R.M.: Limitations to significant information in biological electron microscopy as a result of radiation damage. Journal of Ultrastructure Research 36(3), 466- 482 (1971). https://doi.org/10.1016/S0022- 5320(71)80118- 1. Accessed 2023- 04- 06 + +[14] Glaeser, R.M.: Retrospective: Radiation damage and its associated "Information Limitations". Journal of Structural Biology 163(3), 271- 276 (2008). https://doi.org/10.1016/j.jsb.2008.06.001. Accessed 2023- 04- 06 + +[15] Breedlove, J.R., Trammell, G.T.: Molecular Microscopy: Fundamental Limitations. Science 170(3964), 1310- 1313 (1970). https://doi.org/10.1126/science.170.3964.1310. Publisher: American Association for the Advancement of Science. Accessed 2023- 04- 06 + +[16] Rosenthal, P.B., Henderson, R.: Optimal Determination of Particle Orientation, Absolute Hand, and Contrast Loss in Single- particle Electron Cryomicroscopy. Journal of Molecular Biology 333(4), 721- 745 (2003). https://doi.org/10.1016/j.jmb.2003.07.013. Accessed 2023- 01- 23 + +[17] Kuo, I.A.M., Glaeser, R.M.: Development of methodology for low exposure, high resolution electron microscopy of biological specimens. Ultramicroscopy 1(1), 53- 66 (1975). https://doi.org/10.1016/S0304- 3991(75)80007- 6. Accessed 2023- 04- 06 + +[18] Unwin, P.N.T., Henderson, R.: Molecular structure determination by electron microscopy of unstained crystalline specimens. Journal of Molecular Biology 94(3), 425- 440 (1975). https://doi.org/10.1016/S0022- 2836(75)90212- 0. Accessed 2023- 04- 06 + +[19] Henderson, R.: The potential and limitations of neutrons, electrons and X- rays for atomic resolution microscopy of unstained biological molecules. Quarterly Reviews of Biophysics 28(2), 171- 193 (1995). https://doi.org/10.1017/S003358350000305X. Publisher: Cambridge University Press. Accessed 2023- 03- 23 + +[20] Glaeser, R.M.: Review: Electron Crystallography: Present Excitement, a Nod to the Past, Anticipating the Future. Journal of Structural Biology 128(1), 3- 14 (1999). https://doi.org/10.1006/jsbj.1999.4172. Accessed 2023- 04- 06 + +[21] Glaeser, R.M.: How good can cryo- EM become? Nature Methods 13(1), 28- 32 (2016). https://doi.org/10.1038/nmeth.3695. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 04 + +<--- Page Split ---> + +[22] Paulsen, C.E., Armache, J.- P., Gao, Y., Cheng, Y., Julius, D.: Structure of the TRPA1 ion channel suggests regulatory mechanisms. Nature 520(7548), 511–517 (2015). https://doi.org/10.1038/nature14367. Number: 7548 Publisher: Nature Publishing Group. Accessed 2023- 02- 20 + +[23] Tan, Y.Z., Baldwin, P.R., Davis, J.H., Williamson, J.R., Potter, C.S., Carragher, B., Lyumkis, D.: Addressing preferred specimen orientation in single- particle cryo- EM through tilting. Nature Methods 14(8), 793–796 (2017). https://doi.org/10.1038/nmeth.4347. Number: 8 Publisher: Nature Publishing Group. Accessed 2023- 02- 21 + +[24] Lee, Y., Wiriyasermkul, P., Jin, C., Quan, L., Ohgaki, R., Okuda, S., Kusakizako, T., Nishizawa, T., Oda, K., Ishitani, R., Yokoyama, T., Nakane, T., Shirouzu, M., Endou, H., Nagamori, S., Kanai, Y., Nureki, O.: Cryo- EM structure of the human L- type amino acid transporter 1 in complex with glycoprotein CD98hc. Nature Structural & Molecular Biology 26(6), 510–517 (2019). https://doi.org/10.1038/s41594- 019- 0237- 7. Number: 6 Publisher: Nature Publishing Group. Accessed 2023- 02- 27 + +[25] Kim, J., Tan, Y.Z., Wicht, K.J., Erramilli, S.K., Dhingra, S.K., Okombo, J., Vendome, J., Hagenah, L.M., Giacometti, S.I., Warren, A.L., Nosol, K., Roope, P.D., Potter, C.S., Carragher, B., Kossiakoff, A.A., Quick, M., Fidock, D.A., Mancia, F.: Structure and drug resistance of the Plasmodium falciparum transporter PfCRT. Nature 576(7786), 315–320 (2019). https://doi.org/10.1038/s41586- 019- 1795- x. Number: 7786 Publisher: Nature Publishing Group. Accessed 2023- 02- 27 + +[26] Faust, B., Billesbølle, C.B., Suomivuori, C.- M., Singh, I., Zhang, K., Hoppe, N., Pinto, A.F.M., Diedrich, J.K., Muftuoglu, Y., Szkudlinski, M.W., Saghatelian, A., Dror, R.O., Cheng, Y., Manglik, A.: Autoantibody mimicry of hormone action at the thyrotropin receptor. Nature 609(7928), 846–853 (2022). https://doi.org/10.1038/s41586- 022- 05159- 1. Number: 7928 Publisher: Nature Publishing Group. Accessed 2023- 02- 21 + +[27] Diver, M.M., Cheng, Y., Julius, D.: Structural insights into TRPM8 inhibition and desensitization. Science 365(6460), 1434–1440 (2019). https://doi.org/10.1126/science.aax6672. Publisher: American Association for the Advancement of Science. Accessed 2023- 02- 21 + +[28] Zhou, Y., Moscovich, A., Bendory, T., Bartesaghi, A.: Unsupervised particle sorting for high- resolution single- particle cryo- EM. Inverse Problems 36(4), 044002 (2020). https://doi.org/10.1088/1361- 6420/ab5ec8. Publisher: IOP Publishing. Accessed 2022- 12- 21 + +[29] Méndez, J., Garduño, E., Carazo, J.M., S. Sorzano, C.O.: Identification of incorrectly oriented particles in cryo- EM single particle analysis. Journal of Structural Biology 213(3), 107771 (2021). https://doi.org/10.1016/j. + +<--- Page Split ---> + +jsb.2021.107771. Accessed 2023- 02- 23 + +[30] Henderson, R.: Avoiding the pitfalls of single particle cryo- electron microscopy: Einstein from noise. Proceedings of the National Academy of Sciences 110(45), 18037- 18041 (2013). https://doi.org/10.1073/pnas.1314449110. Publisher: Proceedings of the National Academy of Sciences. Accessed 2023- 02- 25 + +[31] Pintilie, G., Zhang, K., Su, Z., Li, S., Schmid, M.F., Chiu, W.: Measurement of atom resolvability in cryo- EM maps with Q- scores. Nature Methods 17(3), 328- 334 (2020). https://doi.org/10.1038/s41592- 020- 0731- 1. Number: 3 Publisher: Nature Publishing Group. Accessed 2023- 02- 25 + +[32] Zheng, L., Liu, N., Gao, X., Zhu, W., Liu, K., Wu, C., Yan, R., Zhang, J., Gao, X., Yao, Y., Deng, B., Xu, J., Lu, Y., Liu, Z., Li, M., Wei, X., Wang, H.- W., Peng, H.: Uniform thin ice on ultraflat graphene for high- resolution cryo- EM. Nature Methods 20(1), 123- 130 (2023). https://doi.org/10.1038/s41592- 022- 01693- y. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 20 + +[33] Gyobu, N., Tani, K., Hiroaki, Y., Kamegawa, A., Mitsuoka, K., Fujiyoshi, Y.: Improved specimen preparation for cryo- electron microscopy using a symmetric carbon sandwich technique. Journal of Structural Biology 146(3), 325- 333 (2004). https://doi.org/10.1016/j.jsb.2004.01.012 + +[34] Bock, L.V., Grubmüller, H.: Effects of cryo- EM cooling on structural ensembles. Nature Communications 13(1), 1709 (2022). https://doi.org/10.1038/s41467- 022- 29332- 2. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +[35] Ma, H., Jia, X., Zhang, K., Su, Z.: Cryo- EM advances in RNA structure determination. Signal Transduction and Targeted Therapy 7(1), 1- 6 (2022). https://doi.org/10.1038/s41392- 022- 00916- 0. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +[36] Zhang, K., Pintilie, G.D., Li, S., Schmid, M.F., Chiu, W.: Resolving individual atoms of protein complex by cryo- electron microscopy. Cell Research 30(12), 1136- 1139 (2020). https://doi.org/10.1038/s41422- 020- 00432- 2. Number: 12 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +[37] Vulović, M., Ravelli, R.B.G., van Vliet, L.J., Koster, A.J., Lazić, I., Lücken, U., Rullgard, H., Öktem, O., Rieger, B.: Image formation modeling in cryo- electron microscopy. Journal of Structural Biology 183(1), 19- 32 (2013). https://doi.org/10.1016/j.jsb.2013.05.008. Accessed 2023- 02- 25 + +<--- Page Split ---> + +[38] Li, Q., Pellegrino, J., Lee, D.J., Tran, A.A., Chaires, H.A., Wang, R., Park, J.E., Ji, K., Chow, D., Zhang, N., Brilot, A.F., Biel, J.T., van Zundert, G., Borrelli, K., Shinabarger, D., Wolfe, C., Murray, B., Jacobson, M.P., Mühle, E., Chesneau, O., Fraser, J.S., Seiple, I.B.: Synthetic group A streptogramin antibiotics that overcome Vat resistance. Nature 586(7827), 145–150 (2020). https://doi.org/10.1038/s41586-020-2761-3. Number: 7827 Publisher: Nature Publishing Group. Accessed 2023- 02- 25 + +[39] Henderson, R., Glaeser, R.M.: Quantitative analysis of image contrast in electron micrographs of beam- sensitive crystals. Ultramicroscopy 16(2), 139–150 (1985). https://doi.org/10.1016/0304- 3991(85)90069- 5. Accessed 2023- 03- 23 + +[40] Henderson, R.: Image contrast in high- resolution electron microscopy of biological macromolecules: TMV in ice. Ultramicroscopy 46(1), 1–18 (1992). https://doi.org/10.1016/0304- 3991(92)90003- 3. Accessed 2023- 04- 04 + +[41] Zivanov, J., Nakane, T., Scheres, S.H.W.: A Bayesian approach to beam- induced motion correction in cryo- EM single- particle analysis. IUCrJ 6(1), 5–17 (2019). https://doi.org/10.1107/S205225251801463X. Publisher: International Union of Crystallography. Accessed 2023- 04- 04 + +[42] Grant, T., Grigorieff, N.: Measuring the optimal exposure for single particle cryo- EM using a 2.6 Å reconstruction of rotavirus VP6. eLife 4, 06980 (2015). https://doi.org/10.7554/eLife.06980. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 04 + +[43] Zheng, S.Q., Palovcak, E., Armache, J.- P., Verba, K.A., Cheng, Y., Agard, D.A.: MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nature Methods 14(4), 331–332 (2017). https://doi.org/10.1038/nmeth.4193. Number: 4 Publisher: Nature Publishing Group. Accessed 2023- 04- 04 + +[44] Reimer, L.: Transmission Electron Microscopy: Physics of Image Formation and Microanalysis. Springer, ??? (2013). Google- Books- ID: CUFSCAAAQBAJ + +[45] Yip, K.M., Fischer, N., Paknia, E., Chari, A., Stark, H.: Breaking the next Cryo- EM resolution barrier – Atomic resolution determination of proteins! bioRxiv. Pages: 2020.05.21.106740 Section: New Results (2020). https://doi.org/10.1101/2020.05.21.106740. https://www.biorxiv.org/content/10.1101/2020.05.21.106740v1 Accessed 2023- 04- 15 + +[46] Danev, R., Yanagisawa, H., Kikkawa, M.: Cryo- Electron Microscopy Methodology: Current Aspects and Future Directions. Trends in Biochemical Sciences 44(10), 837–848 (2019). https://doi.org/10.1016/j.tibs.2019. + +<--- Page Split ---> + +04.008. Publisher: Elsevier. Accessed 2023- 04- 21 + +[47] Zivanov, J., Nakane, T., Scheres, S.H.W.: Estimation of high- order aberrations and anisotropic magnification from cryo- EM data sets in RELION- 3.1. IUCrJ 7(2), 253- 267 (2020). https://doi.org/10.1107/S2052252520000081. Publisher: International Union of Crystallography. Accessed 2023- 04- 21 + +[48] Carragher, B., Cheng, Y., Frost, A., Glaeser, R.m., Lander, G.c., Nogales, E., Wang, H.- W.: Current outcomes when optimizing 'standard' sample preparation for single- particle cryo- EM. Journal of Microscopy 276(1), 39- 45 (2019). https://doi.org/10.1111/jmi.12834. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/jmi.12834. Accessed 2023- 04- 06 + +[49] Drulyte, I., Johnson, R.M., Hesketh, E.L., Hurdiss, D.L., Scarff, C.A., Porav, S.A., Ranson, N.A., Muench, S.P., Thompson, R.F.: Approaches to altering particle distributions in cryo- electron microscopy sample preparation. Acta Crystallographica Section D: Structural Biology 74(6), 560- 571 (2018). https://doi.org/10.1107/S2059798318006496. Number: 6 Publisher: International Union of Crystallography. Accessed 2023- 04- 06 + +[50] Kim, L.Y., Rice, W.J., Eng, E.T., Kopylov, M., Cheng, A., Raczkowski, A.M., Jordan, K.D., Bobe, D., Potter, C.S., Carragher, B.: Benchmarking cryo- EM Single Particle Analysis Workflow. Frontiers in Molecular Biosciences 5 (2018). Accessed 2023- 04- 06 + +[51] Weissenberger, G., Henderikx, R.J.M., Peters, P.J.: Understanding the invisible hands of sample preparation for cryo- EM. Nature Methods 18(5), 463- 471 (2021). https://doi.org/10.1038/s41592- 021- 01130- 6. Number: 5 Publisher: Nature Publishing Group. Accessed 2022- 11- 14 + +[52] Callaway, E.: The revolution will not be crystallized: a new method sweeps through structural biology. Nature 525(7568), 172- 174 (2015). https://doi.org/10.1038/525172a. Number: 7568 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 CryoSieve outperformed other algorithms in terms of FSC-based resolutions and Q-scores. We compared the density maps reconstructed from retained particles obtained by CryoSieve (indigo) and NCC (green), along with random (orange) as the baseline, at different retention ratios. Density maps were ab initio reconstructed by CryoSPARC after discarding the published refined Euler angles deposited on EMPIAR. Four metrics were employed for measuring density map quality: FSC-based resolutions, including model-to-map (solid lines with squares) and two-half-maps (solid lines with diamonds) resolutions, are shown in the first column; meanwhile, Q-scores of the raw maps (dashed lines with circles) and the sharpened maps (solid lines with circles) are shown in the second column, as Q-score varies by the B-factor used for sharpening. B-factors for sharpening were self-determined by CryoSPARC. The iterations where CryoSieve obtained the finest subset, determined by comprehending these metrics, are labeled with hatched bars.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 The two-dimensional resolution distribution between retained and removed particles was compared. The third iteration of CryoSieve achieved a retention ratio of \(51.2\%\) and a removal ratio of \(48.8\%\) , resulting in a similar number of particles for retention and removal. These two categories underwent CryoSPARC 2D clustering and averaging, i.e., 2D classification, with the number of 2D classes set to 50. All six experimental datasets were tested. CryoSPARC reported the 2D resolution of each 2D class, along with the number of particle images belonging to it. We statistically analyzed the number of particles belonging to each 2D resolution and plotted histograms, demonstrating the difference between retained (steel blue) and removed (crimson) particles in terms of 2D resolution distribution.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 CryoSieve prioritizes the removal of radiation-damaged particles. (a), InSilicoTEM generated simulated particle images with different levels of radiation damage. The parameter \(M_{b}\) in InSilicoTEM, ranging from 0 to 4, mimicked different levels of radiation damage, with higher \(M_{b}\) leading to weaker high-frequency signals and blurrier particles (clean, upper row). Noise, also generated by InSilicoTEM through physical process simulation, was then added to the clean particles to obtain noisy particles for testing (noisy, lower row). (b), The proportions of particles with varying levels of radiation damage (distinguished by colors) in retained particles at different retention ratios (labeled on the left) are shown. The retained particles obtained by CryoSieve (left horizontal bars) and NCC (right horizontal bars) were compared. NCC performed acceptably for particles with high radiation damage, but was unable to distinguish particles with relatively low radiation damage. Meanwhile, CryoSieve sequentially sieved out particles from high to low radiation damage. The accuracy (the proportion of zero radiation damage particles) of CryoSieve was 91.7% at iteration 7, while that of NCC was only 61.5%. (c), Side chains of density maps reconstructed by cryoSPARC using retained particles obtained by CryoSieve (indigo) and NCC (green) were compared. The atomic model (PDB 6PCQ) was fitted into the density maps. The two reconstruction maps are compared at an identical contour. Red arrows emphasize differences between the two maps. (d) Model-to-map FSCs of reconstructed density maps using retained particles were compared, with retention ratio of 21.0% (iteration 7). Retention particles were obtained by CryoSieve (indigo), NCC (green) and random (baseline method, orange), respectively. The FSC threshold (FSC = 0.5) was depicted as a horizontal dashed line.
+ +<--- Page Split ---> + +## Methods + +## Details of comparing the performance of particle sorting algorithms. + +Since cryo- EM single- particle image processing software has experienced rapid development in the past few years, some of the final stacks deposited in EMPIAR can be better processed by state- of- the- art algorithms. To eliminate effects from different refinement software and their versions, ensuring fair comparisons between various particle sorting algorithms, the final stacks deposited on EMPIAR were reprocessed under a standard workflow using CryoSPARC v4.1.0 following a standard workflow. For hemagglutinin, the initial model was generated by low- pass filtering its atomic model to 30A, while for the other proteins, initial models were generated by arbitrary random initialization using CryoSPARC. Then, uniform refinement was applied for TRPA1, TRPM8, hemagglutinin and LAT1, while non- uniform refinement was applied for pfCRT and TSHR- Gs. + +To enable unbiased comparisons of density maps before and after particle sorting, the retained particles obtained from each particle sorting algorithm underwent identical refinement procedures, as previously described using CryoSPARC v4.1.0 in the standard workflow. The reconstructed density maps were used for subsequent measurements. To avoid the potential influence of the "Einstein- from- noise" effect, the former Euler angles were discarded, and new sets of Euler angles were determined through refinement of the retained particles. Moreover, in order to maintain independence between the two half sets and ensure that the Fourier Shell Correlation (FSC) served as the golden standard, half- set splits were preserved throughout the subsequent procedure by turning off the option "Force re- do GS split". + +The reconstructed density maps were evaluated by several metrics, including FSC- based resolution and Q- score. CryoSPARC produced two raw half maps and an auto- postprocessed density map (FSC- weighted, B- factor sharpened, two half sets averaged), accompanied by reporting half- maps FSC. + +FSC- based metric includes half- maps FSC (directly reported by CryoSPARC) and model- to- map FSC. Map- to- model FSC resolution was calculated using the following procedure, with the auto- postprocessed density map as input. The corresponding atomic model of the dataset was converted to the ground- truth density map by the molmap function of Chimera at Nyquist resolution. The mask was generated from the ground- truth density map (after low- pass filtering to 8 Å, extending by 4 pixels and applying a cosine- edge of 4 pixels) using Relion. Model- to- map FSC curves were determined between the input density map (obscured by the mask) and the ground- truth density map. The resolution threshold of the map- to- model FSC was set to 0.5. + +As Q- score is sensitive to B- factor sharpening, the Q- scores of both the raw maps and the auto- postprocessed maps were measured. The auto- postprocessed maps were directly provided by CryoSPARC, while the raw maps were obtained by first averaging the two raw half maps provided by + +<--- Page Split ---> + +CryoSPARC, then low- pass filtering them to an appropriate resolution, in order to eliminate the impact of varying noise intensities on the density maps. The low- pass filtering threshold frequency ranged from \(0.3\mathrm{\AA}\) to \(0.5\mathrm{\AA}\) higher than the CryoSPARC reported half- maps FSC resolution, thus ensuring the retention of useful signals. Specifically, the threshold frequency for TRPA1 was \(3.5\mathrm{\AA}\) , for TRPM8 and TSHR- Gs it was \(2.7\mathrm{\AA}\) , for hemagglutinin it was \(3.4\mathrm{\AA}\) , for pfCRT it was \(3.0\mathrm{\AA}\) , and for LAT it was \(2.8\mathrm{\AA}\) . Q- score was calculated using the MAPQ plugin for UCSF Chimera, with all parameters set to their default values. + +All conversions between CryoSPARC and Relion were performed using the pyem script. + +## CryoSieve's parameters. + +CryoSieve is iteratively performs 3D reconstruction and particle sieving, while maintaining independence between two half sets by independently sieving each set of particles. 3D reconstructions of each subset were performed using Relion v4.0- beta- 2, with the option "—subset" to preserve the half- set splitting. A mask, generated from the atomic model using RELION (low- pass filtered to 8 Å), was applied to the reconstructed raw density map to obtain \(x^{(k - 1)}\) in Equation 2.1 of the CryoSieve score. The same mask was applied to other particle sorting algorithms such as NCC and AGC, to ensure fair comparisons. Subsequently, particles were sieved out based on the ascending order of the CryoSieve score. In total, nine iterations were carried out, with each iteration retaining \(80\%\) of the particles from the previous iteration. The cutoff frequency of the highpass operator \(H^{(k)}\) increased linearly as the iteration progressed. For all datasets, except for LAT1, the initial cutoff frequency was set at \(40\mathrm{\AA}\) , and the final cutoff frequency was \(3\mathrm{\AA}\) . For LAT1, the initial cutoff frequency was \(50\mathrm{\AA}\) , and the final cutoff frequency was also \(3\mathrm{\AA}\) . + +## AGC's particle removal. + +Incorrectly oriented particle removal by AGC was carried out using Scipion v3.0.12, following the instructions provided in its user manual. The initial volumes required for Scipion's AGC algorithms were the reconstructed maps of all particles in final stack using Relion v4.0- beta- 2, applied with a mask identical to that used in CryoSieve and NCC. The "symmetry group" parameter of AGC was set to the corresponding symmetry of the structure, while the remaining parameters were set to their default values. Since the number of particles in the LAT1 dataset is too large to perform AGC without reporting error, we partitioned them into quarter- datasets and performed AGC algorithm within each subset. The remaining particles in the quarter- datasets were used for ab initio reconstruction in cryoSPARC. Particle removal of other datasets were performed using all particles in final stacks as a whole. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementalmaterial.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff_det.mmd b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..bc1601e081c21b23c4aa34fdd096d38ab317f9d6 --- /dev/null +++ b/preprint/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff/preprint__0db09a99923f418955497d1d6f55a2130a934dc119e9b2f2f18766ecb87dc1ff_det.mmd @@ -0,0 +1,498 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 827, 178]]<|/det|> +# Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM + +<|ref|>text<|/ref|><|det|>[[44, 196, 230, 238]]<|/det|> +Jianying Zhu Tsinghua university + +<|ref|>text<|/ref|><|det|>[[44, 243, 230, 284]]<|/det|> +Qi Zhang Tsinghua university + +<|ref|>text<|/ref|><|det|>[[44, 290, 668, 331]]<|/det|> +Hui Zhang Yanqi Lake Beijing Institue of Mathematical Sciences and Applications + +<|ref|>text<|/ref|><|det|>[[44, 336, 230, 377]]<|/det|> +Zuoqiang Shi Tsinghua university + +<|ref|>text<|/ref|><|det|>[[44, 383, 588, 470]]<|/det|> +Mingxu Hu Tsinghua University https://orcid.org/0000- 0003- 3603- 3966 Chenglong Bao ( \(\square\) clbao@tsinghua.edu.cn ) Tsinghua University https://orcid.org/0000- 0002- 1201- 1212 + +<|ref|>sub_title<|/ref|><|det|>[[44, 510, 102, 527]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 547, 662, 568]]<|/det|> +Keywords: cyro- EM, single particle analysis, particle sorting, final stacks + +<|ref|>text<|/ref|><|det|>[[44, 586, 297, 605]]<|/det|> +Posted Date: May 29th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 624, 474, 643]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2921474/v1 + +<|ref|>text<|/ref|><|det|>[[44, 661, 910, 704]]<|/det|> +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 722, 531, 742]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 777, 955, 821]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on December 10th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43555- x. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[140, 170, 857, 237]]<|/det|> +# Not final yet: a minority of final stacks yields superior amplitude in single-particle cryo-EM + +<|ref|>text<|/ref|><|det|>[[170, 260, 822, 308]]<|/det|> +Jianying Zhu \(^{1\dagger}\) , Qi Zhang \(^{3,4,5,6\dagger}\) , Hui Zhang \(^{2}\) , Zuoqiang Shi \(^{1,2*}\) , Mingxu Hu \(^{3,4,5,6*}\) and Chenglong Bao \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[168, 316, 825, 540]]<|/det|> +\(^{1}\) Yau Mathematical Sciences Center, Tsinghua University, Beijing, China. \(^{2}\) Yanqi Lake Beijing Institute of Mathematical Sciences and Applications, Beijing, China. \(^{3}\) Key Laboratory of Protein Sciences (Tsinghua University), Ministry of Education, Beijing, China. \(^{4}\) School of Life Science, Tsinghua University, Beijing, China. \(^{5}\) Beijing Advanced Innovation Center for Structural Biology, Beijing, China. \(^{6}\) Beijing Frontier Research Center for Biological Structure, Beijing, China. + +<|ref|>text<|/ref|><|det|>[[165, 575, 830, 636]]<|/det|> +\*Corresponding author(s). E- mail(s): zqshi@tsinghua.edu.cn; humingxu@mail.tsinghua.edu.cn; clbao@tsinghua.edu.cn; †These authors contributed equally to this work. + +<|ref|>sub_title<|/ref|><|det|>[[451, 670, 544, 686]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[165, 690, 830, 888]]<|/det|> +Cryo- electron microscopy (cryo- EM) is widely used to determine near- atomic resolution structures of biological macromolecules. Due to the extremely low signal- to- noise ratio, cryo- EM relies on averaging many images. However, a crucial question in the field of cryo- EM remains unanswered: how close can we get to the minimum number of particles required to reach a specific resolution in practice? The absence of an answer to this question has impeded progress in understanding sample behavior and the performance of sample preparation methods. To address this issue, we have developed a new iterative particle sorting and/or sieving method called CryoSieve. Extensive experiments demonstrate that CryoSieve outperforms other cryo- EM particle sorting algorithms and reveals that most particles are unnecessary in final + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[166, 82, 830, 133]]<|/det|> +stacks. The minority of particles remaining in the final stacks yield superior high- resolution amplitude in reconstructed density maps. For some datasets, the size of the finest subset approaches the theoretical limit. + +<|ref|>text<|/ref|><|det|>[[166, 147, 797, 163]]<|/det|> +Keywords: cyro- EM, single particle analysis, particle sorting, final stacks + +<|ref|>sub_title<|/ref|><|det|>[[115, 217, 353, 241]]<|/det|> +## 1 Introduction + +<|ref|>text<|/ref|><|det|>[[113, 253, 884, 433]]<|/det|> +The transformative impact of cryo- EM single- particle analysis (SPA) on the field of structural biology has been widely recognized by the scientific community [1]. Cryo- EM has advanced significantly due to a series of technological innovations [2- 7], enabling the technique to provide macromolecular structures with up to atomic resolution at an unprecedented rate. This technological progress is commonly referred to as the resolution revolution [8]. Cryo- EM involves using electron microscopy images of biomolecules embedded in vitreous, glass- like ice [9], which are then combined to generate three- dimensional (3D) density maps. These maps provide valuable insights into the function of macromolecules and their role in biological processes. + +<|ref|>text<|/ref|><|det|>[[111, 434, 884, 902]]<|/det|> +The stability and electron- optical performance of electron microscopes do not hinder the use of cryo- EM [10]. However, biological samples studied in cryo- EM are radiation- sensitive [11, 12]. Therefore, a trade- off must be made between improving the signal- to- noise ratio (SNR) and limiting radiation damage [13, 14]. It was concluded that statistically well- defined three- dimensional (3D) structures could not be obtained from individual biological macromolecules at atomic resolution [15, 16]. Instead, increasing the SNR by averaging image data from many identical macromolecules is the only way to progress [13, 17, 18]. Over two decades ago, Henderson estimated that structures could be determined at a resolution of near 3 Å by merging data from approximately 12,000 particles, even for particles as small as approximately 40 kDa [19]. Later, Rosenthal and Henderson argued that the electron microscopy community should adopt the same threshold criterion for structure factor quality as the X- ray protein crystallography community, which was set at a figure- of- merit of 0.5 corresponding to a phase error of 60 degrees [16]. The theoretical limit of the minimum number of particle images required to achieve a specific resolution can be calculated using the theory proposed by Henderson and Rosenthal [16, 19], given the B- factor of the instrument (e.g., electron microscopy and camera) [13, 14, 20]. In practice, the final stacks of cryo- EM still far fall short of the theoretical limit, indicating a considerable gap between what can be accomplished and the physical limit of what cryo- EM can do [21]. The initial particle datasets obtained by particle picking from micrographs undergo multiple rounds of laborious 2D and 3D classification to generate the final stack for model determination. The final stacks, which yield atomic or sub- atomic resolution density maps, typically comprise several orders of magnitude fewer particles than the original datasets. Therefore, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 80, 884, 172]]<|/det|> +cryo- EM field faces the long- standing question of how close we can approach the theoretical limit in practice. The lack of an answer to this open question has hindered the quantification of the performance of various underdeveloped sample preparation methods and impeded the investigation of trends and the understanding of the underlying mechanisms of sample behavior. + +<|ref|>text<|/ref|><|det|>[[111, 172, 884, 567]]<|/det|> +To answer the question of how close cryo- EM can approach its theoretical limit, it is crucial to determine the minimum number of particles required to achieve a high- resolution 3D reconstruction within a given dataset. In this study, we introduce CryoSieve, an iterative particle sorting and/or sieving algorithm that identifies the smallest subset of particles necessary to generate high- resolution density maps, which we call the finest subset. CryoSieve compares the high- frequency components of synthetic and observed particle images. A higher CryoSieve score indicates superior quality rather than typical cryo- EM damage or artifacts (Section 2.1). Extensive experiments show that CryoSieve outperforms other particle sorting algorithms in various metrics and reveals that most particles in final stacks are futile (Section 2.2). The finest subsets generate 3D density maps with better high- resolution amplitude, using much fewer particles than the final stacks (Section 2.3). We propose that CryoSieve removes radiation- damaged particles within cryo- EM datasets, supported by experiments on simulated radiation damage particles (Section 2.4). Lastly, we compare the minimum particles required in theory with the size of the finest subsets obtained by CryoSieve, finding that some datasets come close to the theoretical limit after being sieved by CryoSieve (Section 2.5). We conclude that, for these datasets, the primary opportunity for further improvement lies in generating fewer futile particles during sample preparation rather than further improving the quality of the particle images that constitute the finest subset. + +<|ref|>sub_title<|/ref|><|det|>[[111, 587, 270, 609]]<|/det|> +## 2 Results + +<|ref|>sub_title<|/ref|><|det|>[[111, 624, 446, 645]]<|/det|> +### 2.1 Design of CryoSieve. + +<|ref|>text<|/ref|><|det|>[[111, 653, 884, 781]]<|/det|> +We have developed a particle sorting and/or sieving model called CryoSieve that alternatively performs 3D reconstruction and particle selection, eliminating futile particles during each iteration. In CryoSieve, a cryo- EM single- particle reconstruction software selected by the user is used to reconstruct a new density map with the retained particle images, which is then used in the subsequent iteration. The retained particle images in each iteration form a subset of those from the previous iteration, as shown in the following formula: + +<|ref|>equation<|/ref|><|det|>[[243, 792, 752, 822]]<|/det|> +\[\left\{i_{1}^{(k)},i_{2}^{(k)}\dots ,i_{n^{(k)}}^{(k)}\right\} \subset \left\{i_{1}^{(k - 1)},i_{2}^{(k - 1)},\dots ,i_{n^{(k - 1)}}^{(k - 1)}\right\} ,\] + +<|ref|>text<|/ref|><|det|>[[111, 826, 884, 881]]<|/det|> +where \(n^{(k - 1)}\) represents the number of retained particles. At each iteration, let \(b_{j}\) be the \(j\) - th particle image, \(A_{j}\) be its forward operator defined by the estimated parameters and \(x^{(k - 1)}\) be the reconstructed density map from the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 80, 884, 118]]<|/det|> +retained particle images in the previous iteration, particles are sieved out based on their CryoSieve score, which is defined as follows: + +<|ref|>equation<|/ref|><|det|>[[111, 147, 884, 186]]<|/det|> +\[g_{j}\coloneqq \left\| H^{(k)}b_{j}\right\|_{2}^{2} - \left\| H^{(k)}(b_{j} - A_{j}x^{(k - 1)})\right\|_{2}^{2},\quad j\in \left\{i_{1}^{(k - 1)},i_{2}^{(k - 1)},\dots ,i_{n^{(k - 1)}}^{(k - 1)}\right\} .\] + +<|ref|>text<|/ref|><|det|>[[111, 200, 884, 238]]<|/det|> +Here, \(H^{(k)}\) is the high- pass operator at the \(k\) - th iteration, and its threshold frequency increases as the iteration progresses. + +<|ref|>text<|/ref|><|det|>[[111, 237, 884, 526]]<|/det|> +The CryoSieve score estimates the similarity between a particle and a reference projection above a given frequency. A higher CryoSieve score indicates that the particle and the reference projection share a higher proportion of signal energy, indicating better particle quality. As radiation damage mainly affects the high- frequency range, the CryoSieve score includes a high- pass operator to extract the high- frequency part. We have demonstrated that the CryoSieve score can identify particles with incorrect pose parameters or components in the high- frequency range through theoretical analysis and simulation verification. Assuming that noise in particles follows a Gaussian distribution, we have shown that, with high probability, the CryoSieve score is an ideal indicator of particle image quality, distinguishing it from typical cryo- EM damage or artifacts (Supplementary Material I). Furthermore, when simulating radiation damage as high- frequency random phasing, the CryoSieve score exhibits remarkable accuracy in selecting particles even with a very low signal- to- noise ratio (approximately 0.001), achieving a precision rate of around 90% (Supplementary Material I). + +<|ref|>sub_title<|/ref|><|det|>[[111, 543, 830, 565]]<|/det|> +### 2.2 Majority of the particles are futile in final stacks. + +<|ref|>text<|/ref|><|det|>[[111, 572, 884, 843]]<|/det|> +We demonstrate the versatility of our method by applying it to six experimental datasets (Table 1). The first dataset is derived from the human TRPA1 ion channel (EMPIAR- 10024) [22]. The second dataset is from influenza hemagglutinin trimer (EMPIAR- 10097) [23], of which the preferred orientation necessitated 40- degree tilts during data acquisition. The third dataset involves LAT1- CD98hc bound to MEM- 108 Fab (EMPIAR- 10264) [24], while the fourth features membrane- bound pFCRT complexed with Fab (EMPIAR- 10330) [25]. Both of these datasets utilized signal subtraction during data processing. The fifth dataset is from CS- 17 Fab- bound TSHR- Gs (EMPIAR- 11120) [26], and the sixth is from TRPM8 bound to calcium (EMPIAR- 11233) [27]. All datasets were obtained using a voltage of 300 kV and an amplitude contrast of 0.07 or 0.1. The TEM systems and electron detectors used in the experiments are listed in Table 1, along with additional metadata such as the number of particles in the final stacks, spherical aberration and molecular symmetry. + +<|ref|>text<|/ref|><|det|>[[111, 843, 884, 880]]<|/det|> +All of the datasets are deposited in the Electron Microscopy Public Image Archive (EMPIAR) as final stacks. These final stacks, which also contain + +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[88, 339, 699, 360]]<|/det|> +Table 1 Microscopic imaging parameters of six experimental datasets along with their associated metadata. + +<|ref|>table<|/ref|><|det|>[[87, 370, 917, 630]]<|/det|> + +
datasetTEMelectron
detector		number of
particles		spherical
aberration (mm)		symmetry
TRPA1TF30 PolaraGatan K2 Summit43,5852.0\(C_{4}\)
hemagglutininTitan KriosGatan K2 Summit130,0002.7\(C_{3}\)
LAT1Titan KriosFEI FALCON III250,7122.7\(C_{1}\)
pfCRTTitan KriosGatan K2 Qutuamn16,9050.001\(C_{1}\)
TSHR-GsTitan KriosGatan K3 Qutuamn41,0542.7\(C_{1}\)
TRPM8Titan KriosGatan K2 Summit42,0402.6\(C_{4}\)
+ +<|ref|>text<|/ref|><|det|>[[980, 312, 998, 680]]<|/det|> +Springer Nature 2021 LXFEX template + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 208]]<|/det|> +the corresponding refined Euler angles, were used to generate the final published reconstructions. The final stacks are generated by manually selecting significantly smaller subsets through multiple rounds of 2D/3D classification, resulting in a substantial reduction in the number of particles compared to the original particle stacks. For example, the final stack of the CS- 17 Fab- bound TSHR- Gs dataset consists of 41,054 particles selected from a pool of 2,742,080 particles, representing a reduction by a factor of approximately 67 [26]. + +<|ref|>text<|/ref|><|det|>[[111, 207, 884, 405]]<|/det|> +We employed CryoSieve to process the six experimental datasets. CryoSieve removed \(20\%\) of the particles in each iteration, resulting in a retaining ratio of \(80.0\%\) , \(64.0\%\) , \(51.2\%\) , and so on. The retained particles in different iterations were then used for ab initio reconstruction to determine the finest subset of particles. The finest subset only contained \(26.2\%\) to \(32.8\%\) of the particles in the final stack. However, the quality of the reconstructed map from the finest subset was consistent with that obtained from all particles in the final stack, as demonstrated in Figure 1. These results indicate that CryoSieve can effectively eliminate over half of the particles with unreliable high- frequency signals without negatively affecting the final reconstruction. Therefore, CryoSieve is highly effective in selecting the most informative particles. + +<|ref|>text<|/ref|><|det|>[[111, 405, 884, 639]]<|/det|> +We performed a comparative analysis of CryoSieve with other cryo- EM particle sorting criteria or software currently used in the field, including the normalized cross- correlation (NCC) method [28] and the angular graph consistency (AGC) approach [29]. In our experiments, we used final stacks composed of relatively high- quality particles. NCC retains an equal number of particles compared to CryoSieve at each iteration, while AGC's retaining ratio is self- determined. However, AGC's retaining ratio was mainly over \(90\%\) , resulting in only a small fraction of particles being removed. Thus the quality of the reconstructed map using the retained particles did not improve or worsen (Supplementary Table 1), as these tested final stacks are composed of relatively high- quality particles. Additionally, we randomly selected the same number of particles from the tested final stacks at each iteration to observe the baseline effect of particle number reduction. + +<|ref|>text<|/ref|><|det|>[[111, 639, 884, 875]]<|/det|> +For all the aforementioned methods (CryoSieve, NCC, AGC, and random), we discarded the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein- from- noise effect [30]. Instead, the retained particles were used for ab initio reconstruction by CryoSPARC to obtain refreshed sets of Euler angles and density maps. Several metrics, including FSC- based resolution [16] and Q- score [31], were used to measure the quality of the refreshed density maps. Comprehending these metrics, our analyses reveal that CryoSieve effectively sieves out \(67.2\%\) to \(73.8\%\) (varying based on datasets) of particles from the final stacks without deteriorating the yielded density maps (Figure 2). Meanwhile, subsets of equal size retained by the other methods failed to reconstruct density maps of the same quality as the original (Figure 2). Therefore, CryoSieve significantly outperforms other particle sorting algorithms, demonstrating that majority of the particles are dispensable in the final stacks. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[156, 82, 841, 469]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 479, 884, 630]]<|/det|> +
Fig. 1 CryoSieve is capable of maintaining resolutions after removing the majority of particles in the final stacks. For all six experimental datasets, density maps of the CryoSieve-retained particles (steel blue) and all particles in the final stack (medium purple) were compared, obtained from CryoSPARC's ab initio reconstruction after discarding the published refined Euler angles deposited on EMPIAR to avoid the Eisenstein-from-noise effect. The density maps were first FSC-weighted (based on FSCs given by CryoSPARC), and then B-factor sharpened using equivalent B-factors for the same protein: \(-90\mathrm{\AA}^{2}\) for TRPA1, \(-180\mathrm{\AA}^{2}\) for hemagglutinin, \(-100\mathrm{\AA}^{2}\) for LAT1, \(-60\mathrm{\AA}^{2}\) for pfCRT, \(-70\mathrm{\AA}^{2}\) for TSHR-Gs, and \(-80\mathrm{\AA}^{2}\) for TRPM8. The central bars indicate the proportions of the retained and removed particles.
+ +<|ref|>text<|/ref|><|det|>[[111, 650, 884, 815]]<|/det|> +cisTEM [5] is capable of reporting a score for each single particle image after 3D reconstruction, though it is not a particle sorting criterion. Due to differences in alignment and other image processing workflows between cisTEM and cryoSPARC, cisTEM cannot be strictly compared with CryoSieve. Therefore, we compared Cryosieve with cisTEM by sorting single particles by cisTEM score and retaining the same number of particle images for ab initio reconstruction using cryoSPARC. CryoSieve outperformed cisTEM in all six experimental datasets (Supplementary Material II and Supplementary Figure 2). + +<|ref|>text<|/ref|><|det|>[[111, 814, 884, 904]]<|/det|> +Moreover, we analyzed the differences between the particle images retained and removed using CryoSieve by performing 2D classification of the particles into 50 classes using CryoSPARC. To ensure a comparable number of particles for both retained and removed groups, iteration 3 was chosen for such analysis, with a retention ratio of \(51.2\%\) and a removal ratio of \(48.8\%\) . CryoSPARC + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 245]]<|/det|> +reported the 2D resolution of each class, along with the number of particle images belonging to it. The particles retained by CryoSieve (Figure 3, steel blue) were distributed at a higher resolution compared to those removed by it (Figure 3, crimson). In five out of the six datasets, particle images with the highest resolution, i.e., 8.5–9.6 Å in hemagglutinin, 6.6–8.2 Å in LAT1, 7.2–11.6 Å in pfCRT, 7.2–8.5 Å in TSGH-Gs, and 11.6–7.5 Å in TRPM8, were entirely retained by CryoSieve. Therefore, our analysis suggests that CryoSieve selectively retained the higher- quality particle images in the final stack while discarding lower- quality ones. + +<|ref|>sub_title<|/ref|><|det|>[[111, 262, 842, 305]]<|/det|> +### 2.3 Better high-resolution amplitude with much fewer particles. + +<|ref|>text<|/ref|><|det|>[[111, 312, 884, 604]]<|/det|> +B- factors, also known as Debye- Waller factors or temperature factors, reflect the rate at which the amplitude of high- resolution information decreases [16]. Lower B- factors indicate that the high- resolution signal has been better preserved during sample preparation, imaging, and image processing, implying that the particle images are of higher quality. B- factors are widely used to measure image quality in cryo- EM quantitatively [32–36]. In our six experimental datasets, the finest subset, comprising only 26.2% to 32.8% of particles in the final stack, yields 3D density maps with B- factors reduced by \(18.0\mathring{\mathrm{A}}^{2}\) to \(71.2\mathring{\mathrm{A}}^{2}\) compared to those produced by the original final stacks (Table 2, column B and C). In other words, the density maps reconstructed from the finest subset have a better high- resolution amplitude, meaning they contain a greater high- resolution intensity, despite the fact that the finest subsets only contain a small fraction of particles in the final stack. This indicates that CryoSieve significantly reduced the temperature factor and alleviated the amplitude contrast decay, suggesting that high- quality particles contribute to the density map and can be effectively selected by CryoSieve. + +<|ref|>sub_title<|/ref|><|det|>[[111, 622, 856, 666]]<|/det|> +### 2.4 CryoSieve can effectively detect radiation-damaged particles. + +<|ref|>text<|/ref|><|det|>[[111, 671, 884, 781]]<|/det|> +We hypothesize that some particle images in the final stacks have been subject to some degree of radiation damage and cannot be screened out by conventional methods. These particles do not contribute positively to the reconstructed density map. To verify the possibility of this conjecture, we generated simulated particle images using InSilicoTEM [37], with varying degrees of radiation damage, and screened them using CryoSieve. + +<|ref|>text<|/ref|><|det|>[[111, 780, 884, 911]]<|/det|> +The atomic model of \(E\) . coli 50S ribosome bound to VM2 (PDB 6PCQ) [38] was employed to generate particle images, with defocus ranges from 1800 nm to 2800 nm, an electron dose of \(40 \mathrm{e}^{- \mathring{\mathrm{A}}^{- 2}}\) , a Falcon I detector, a box size of \(450 \times 450\) , and a pixel size of \(1.04\mathring{\mathrm{A}}\) . Radiation damage was simulated by varying the motion blur factor from 0 to 4 (Figure 4a). Each set of the simulated single particles contained 8,000 particles, and the projection directions followed a random distribution. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[110, 120, 890, 252]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[110, 78, 821, 109]]<|/det|> +Table 2 The finest subsets alleviate high-resolution amplitude decay, along with a comparison to their theoretical number of particle limit. + +
datasetA†B*C*DEF
TRPA13.90141.978.1(63.8-)52111,426(51.9×)43,585(83.7×)
hemagglutinin3.62232.0160.8(71.2-)97534,078(35.6×)130,000(133.3×)
LAT13.11132.696.0(36.6-)6,69765,687(9.8×)250,712(37.4×)
pfCRT3.3785.149.5(35.6-)4,2124,429(1.01×)16,905(4.0×)
TSHR-Gs2.9692.961.7(31.2-)9,20513,465(1.46×)41,054(4.5×)
TRPM82.9894.776.7(18.0-)2,20013,789(6.3×)42,040(19.1×)
+ +<|ref|>text<|/ref|><|det|>[[130, 256, 885, 360]]<|/det|> +\(^{\dagger}\) (A), half- maps resolution of the CryoSieve- retained particles (A); (B), B- factor obtained from all particles in the final stacks (A²); (C), B- factor obtained from the CryoSieve- retained particles with temperature decrease (compared with all particles) in brackets (A²); (D), theoretical number of particles limit at \(B = 50\mathrm{\AA}^{2}\) ; (E), number of the CryoSieve- retained particles with folds of theoretical limit in brackets.; (F), number of particles in final stacks with folds of theoretical limit in brackets. \* B- factors were reported by cryoSPARC auto- postprocessing. + +<|ref|>text<|/ref|><|det|>[[111, 384, 885, 621]]<|/det|> +We compared the retention behavior of CryoSieve and NCC for particles with different simulated degrees of radiation damage, with random retention as the baseline. As the number of iterations increased, the retention rate decreased, and CryoSieve demonstrated a greater ability to screen out particles with higher levels of radiation damage (Figure 4b). While NCC outperformed the random retention baseline, it was still inferior to CryoSieve in terms of screening accuracy (Figure 4b). Additionally, we compared CryoSieve with cisTEM and found that although cisTEM performed acceptably for particles with high radiation damage, it still exhibited inferior performance to CryoSieve for particles with relatively low radiation damage. (Supplementary Material II and Supplementary Figure 2). These results imply that CryoSieve could effectively removes radiation- damaged particles to improve the quality of the final density maps. + +<|ref|>sub_title<|/ref|><|det|>[[111, 638, 833, 680]]<|/det|> +### 2.5 The finest subsets may be close to the theoretical number of particles limit. + +<|ref|>text<|/ref|><|det|>[[111, 688, 884, 725]]<|/det|> +The theoretical number limit of particle images, given by Rosenthal and Henderson [16], is + +<|ref|>equation<|/ref|><|det|>[[305, 737, 688, 782]]<|/det|> +\[\mathrm{N}_{\mathrm{particles}} = \frac{1}{N_{\mathrm{asym}}}\frac{S^2}{N_e\sigma_e d}\exp \left(\frac{B}{2d^2}\right),\] + +<|ref|>text<|/ref|><|det|>[[111, 785, 885, 902]]<|/det|> +where \(N_{\mathrm{asym}}\) , \(S\) , \(N_{e}\) , \(\sigma_{e}\) , \(d\) , \(B\) stand for the number of asymmetric units, the signal- to- noise threshold criteria of the resolution, the electron dose, the elastic cross- section for carbon, the resolution, and the overall temperature factor, respectively. In the above formula, \(\frac{S}{N} = \frac{1}{\sqrt{3}}\) , which is equivalent to phase error of 60 degrees or 0.143- threshold of half- maps FSC [16]. Meanwhile, \(N_{e} = 5e^{- \mathrm{\AA}^{- 2}}\) , which is believed to be the limiting dose due to radiation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 250]]<|/det|> +damage for features near- atomic resolution [16, 19, 39, 40]. The electron dose used in practice is typically a fold higher than the limiting dose. Although the additional dose does not contribute to the structure factor amplitudes at near atomic resolution, it may have increased the signal up to the resolution limit of the final map, thus making the determination of particle parameters easier [16]. This conjecture agrees with the observation in the study of micrograph movie stack dose weighting, which found that only the initial few frames, not the subsequent frames, contribute to near atomic features [41- 43]. Finally, \(\sigma_{e} = 0.004\mathrm{\AA}^{2}\) is the elastic cross- section for carbon at \(300\mathrm{kV}\) [44]. + +<|ref|>text<|/ref|><|det|>[[111, 248, 884, 596]]<|/det|> +The overall temperature factor, or Rosenthal and Henderson's B- factor, is the dominant factor in estimating the theoretical limit. Here, we proposed a simplified and crude assumption that limits only exist on instruments (TEM and electron detector) and that no other resolution- limiting factors exist. In other words, we assumed that all other procedures or techniques were ideal. For example, vitrified non- amorphous ice is perfectly flat and of ideal thickness, there is no beam- induced motion, and orientations of particles follow a uniform distribution, and there is no electron- charging effect. Therefore, B- factor represents a summary of all resolution- limiting factors of a given electron microscope, and describes the overall quality of the instrumental setup. Holger Stark and his colleagues have summarized the current knowledge on existing state- of- the- art commercial EM hardware and their B- factors [45]. For the standard Titan Krios, they concluded that its B- factor is \(50\mathrm{\AA}^{2}\) , which was determined by re- evaluating data from EMPIAR- 10216 as described by [46], with modifications to account for off- axial aberrations by splitting the micro- graphs into nine subsets [47]. Therefore, we computed the theoretical number of particle limits at \(B = 50\mathrm{\AA}^{- 2}\) (Table 2, column D). The sizes of the finest subsets obtained by CryoSieve were compared with such theoretical limits (Table 2, column E). + +<|ref|>text<|/ref|><|det|>[[111, 596, 884, 888]]<|/det|> +Out of the six datasets examined, two (pfCRT and TSHR- Gs) were found to be close to their theoretical limits (Table 2, column E, emphasized by bold font). However, the TRPA1 dataset fell short of the theoretical limit by approximately 52. This could be due to the lower resolution capabilities of the TF30 Polara TEM used in the study compared to more advanced models like the Titan Krios. It is possible that the assumed B- factor of \(50\mathrm{\AA}^{2}\) for the TF30 Polara is relatively low and does not accurately reflect the properties of the TEM. Additionally, the sample preparation techniques used in 2015 when the TRPA1 study was conducted may not have been fully optimized to achieve the highest resolution possible. Hemagglutinin also fell short of the theoretical limit by a factor of approximately 36 due to using a tilt- collection strategy to account for preferred orientation, resulting in a larger effective ice thickness and a decrease in the quality of particle images. Finally, LAT1 and TRPM8 fell short of the theoretical limit by a factor of 9.8 and 6.3, respectively, indicating that there is still room for improvement in sample preparation for these datasets. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 77, 319, 100]]<|/det|> +## 3 Discussion + +<|ref|>text<|/ref|><|det|>[[113, 112, 884, 277]]<|/det|> +In this study, we introduced the CryoSieve algorithm, which has the ability to estimate the minimum number of particles in a dataset, referred to as the finest subset. CryoSieve demonstrated that most particles in the final stacks are superfluous and do not contribute to reconstructing density maps. On the other hand, the minority of particles that remain in the final stacks yields superior high- resolution amplitude. We also discovered that for some datasets, the size of the finest subset comes close to the theoretical limit. Therefore, CryoSieve can, to some degree, provide insight into a long- standing question in the cryo- EM field: how close can we approach the theoretical limit in practice? + +<|ref|>text<|/ref|><|det|>[[112, 276, 884, 618]]<|/det|> +Despite technical advancements that have made cryo- EM more accessible to structural biologists, sample preparation remains a major bottleneck in the workflow. Scientists and engineers are thus focusing their efforts on this challenge [48]. In single- particle analysis (SPA), sample preparation consists of two crucial steps: sample optimization and grid preparation. The former involves purifying the specimen while maintaining its optimal biochemical state. The latter entails preparing the sample for analysis in the microscope, including chemical or plasma treatment of the grid, sample deposition, and vitrification. Numerous techniques have been proposed to address macromolecular instability, but the efficacy of one approach over another depends on the sample's characteristics [48, 49]. Currently, grid preparation results are heavily influenced by the user's expertise and experience, which can make the process time- consuming and challenging [21, 50]. The numerous variables encountered in sample and grid preparation pose challenges in establishing cause- and- effect relationships, as researchers can only assess the sample at the molecular level using the microscope. As a result, quantitative statistics from comparisons of different sample and grid preparation protocols are still lacking, and a systematic approach is necessary to investigate trends and comprehend the fundamental mechanisms of sample behavior [51]. + +<|ref|>text<|/ref|><|det|>[[112, 617, 883, 780]]<|/det|> +CryoSieve has the potential to establish a metric for the quantitative evaluation of various sample preparation techniques by measuring image quality based on the gaps between the theoretical limits and the size of the finest subsets. One of the possible future directions is to address the variables encountered during sample and grid preparation and establish cause- and- effect relationships. Resolving these issues, among others, cryo- EM could become a more versatile and an even more influential technology in structural biology, potentially addressing new research questions and aiding the growth of novel methodologies as the field advances [52]. + +<|ref|>sub_title<|/ref|><|det|>[[113, 799, 420, 823]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[112, 835, 883, 909]]<|/det|> +This work was supported by the National Key R&D Program of China (No.2021YFA1001300) (to C.B.), the National Natural Science Foundation of China (No.12271291) (to C.B.), the Advanced Innovation Center for Structural Biology (to M.H.), the Beijing Frontier Research Center for Biological + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 80, 885, 155]]<|/det|> +Structure (to M.H.), and the National Natural Science Foundation of China (No.12071244) (to Z.S.). We also would like to express our gratitude to Shouqing Li and Ranhao Zhang for generously sharing their expertise and experiences in particle selection and density map reconstruction in Cryo- EM. + +<|ref|>sub_title<|/ref|><|det|>[[112, 173, 379, 198]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[111, 209, 884, 319]]<|/det|> +The raw final stack datasets analyzed in this study were downloaded from the EMPIAR repository (EMPIAR- 10024, EMPIAR- 11233, EMPIAR- 10097, EMPIAR- 11120, EMPIAR- 10264, EMPIAR- 10330). Atomic coordinates from Protein Data Bank 6PCQ were used for the generation of simulated particles using InSilicoTEM. The finest subsets of these six experimental datasets obtained by CryoSieve can be found as Supplementary Material III. + +<|ref|>sub_title<|/ref|><|det|>[[111, 336, 384, 361]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[110, 372, 884, 410]]<|/det|> +CryoSieve will be open- source upon publication and is also available upon request during the review process. + +<|ref|>sub_title<|/ref|><|det|>[[111, 426, 323, 451]]<|/det|> +## Contribution + +<|ref|>text<|/ref|><|det|>[[110, 462, 884, 537]]<|/det|> +C.B., M.H., and Z.S. initiated the project. M.H., Q.Z., and J.Z. developed CryoSieve and carried out testing. H.Z. provided support in using InSilicoTEM. J.Z. and M.H. analyzed the data. M.H., J.Z., and C.B. wrote the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[110, 555, 440, 580]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[110, 591, 600, 611]]<|/det|> +All other authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[110, 629, 286, 652]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[123, 664, 884, 737]]<|/det|> +[1] Nogales, E.: The development of cryo- EM into a mainstream structural biology technique. Nature Methods 13(1), 24- 27 (2016). https://doi.org/10.1038/nmeth.3694. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 752, 884, 825]]<|/det|> +[2] Bai, X.- c., Fernandez, I.S., McMullan, G., Scheres, S.H.: Ribosome structures to near- atomic resolution from thirty thousand cryo- EM particles. eLife 2, 00461 (2013). https://doi.org/10.7554/eLife.00461. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 840, 884, 895]]<|/det|> +[3] Campbell, M.G., Cheng, A., Brilot, A.F., Moeller, A., Lyumkis, D., Veesler, D., Pan, J., Harrison, S.C., Potter, C.S., Carragher, B., Grigorieff, N.: Movies of Ice- Embedded Particles Enhance Resolution in Electron + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[160, 81, 884, 119]]<|/det|> +Cryo- Microscopy. Structure 20(11), 1823- 1828 (2012). https://doi.org/ 10.1016/j.str.2012.08.026. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 131, 884, 223]]<|/det|> +[4] Li, X., Mooney, P., Zheng, S., Booth, C.R., Braunfeld, M.B., Gubbens, S., Agard, D.A., Cheng, Y.: Electron counting and beam- induced motion correction enable near- atomic- resolution single- particle cryo- EM. Nature Methods 10(6), 584- 590 (2013). https://doi.org/10.1038/nmeth.2472. Number: 6 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 237, 884, 310]]<|/det|> +[5] Grant, T., Rohou, A., Grigorieff, N.: cisTEM, user- friendly software for single- particle image processing. eLife 7, 35383 (2018). https://doi.org/ 10.7554/eLife.35383. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 324, 884, 397]]<|/det|> +[6] Scheres, S.H.W.: RELION: Implementation of a Bayesian approach to cryo- EM structure determination. Journal of Structural Biology 180(3), 519- 530 (2012). https://doi.org/10.1016/j.jsb.2012.09.006. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 410, 884, 503]]<|/det|> +[7] Punjani, A., Rubinstein, J.L., Fleet, D.J., Brubaker, M.A.: cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nature Methods 14(3), 290- 296 (2017). https://doi.org/10.1038/nmeth. 4169. Number: 3 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 516, 884, 572]]<|/det|> +[8] Kühlbrandt, W.: The Resolution Revolution. Science 343(6178), 1443- 1444 (2014). https://doi.org/10.1126/science.1251652. Publisher: American Association for the Advancement of Science. Accessed 2022- 11- 19 + +<|ref|>text<|/ref|><|det|>[[123, 585, 884, 641]]<|/det|> +[9] Dubochet, J., McDowall, A.W.: Vitrification of pure water for electron microscopy. Journal of Microscopy 124(DEC), 3- 4 (1981). Publisher: Wiley- Blackwell. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[123, 654, 884, 764]]<|/det|> +[10] Glaeser, R.M.: How Good Can Single- Particle Cryo- EM Become? What Remains Before It Approaches Its Physical Limits? Annual Review of Biophysics 48(1), 45- 61 (2019). https://doi.org/10.1146/annurev- biophys- 070317- 032828. - eprint: https://doi.org/10.1146/annurev- biophys- 0703- 032828. Accessed 2022- 12- 22 + +<|ref|>text<|/ref|><|det|>[[123, 778, 884, 870]]<|/det|> +[11] Baker, L.A., Smith, E.A., Bueler, S.A., Rubinstein, J.L.: The resolution dependence of optimal exposures in liquid nitrogen temperature electron cryomicroscopy of catalase crystals. Journal of Structural Biology 169(3), 431- 437 (2010). https://doi.org/10.1016/j.jsb.2009.11.014. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[90, 883, 884, 903]]<|/det|> +[12] Bammes, B.E., Jakana, J., Schmid, M.F., Chiu, W.: Radiation damage + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[159, 81, 884, 135]]<|/det|> +effects at four specimen temperatures from 4 to 100K. Journal of Structural Biology 169(3), 331- 341 (2010). https://doi.org/10.1016/j.jsb.2009.11.001. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 149, 884, 222]]<|/det|> +[13] Glaeser, R.M.: Limitations to significant information in biological electron microscopy as a result of radiation damage. Journal of Ultrastructure Research 36(3), 466- 482 (1971). https://doi.org/10.1016/S0022- 5320(71)80118- 1. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 236, 884, 293]]<|/det|> +[14] Glaeser, R.M.: Retrospective: Radiation damage and its associated "Information Limitations". Journal of Structural Biology 163(3), 271- 276 (2008). https://doi.org/10.1016/j.jsb.2008.06.001. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 306, 884, 379]]<|/det|> +[15] Breedlove, J.R., Trammell, G.T.: Molecular Microscopy: Fundamental Limitations. Science 170(3964), 1310- 1313 (1970). https://doi.org/10.1126/science.170.3964.1310. Publisher: American Association for the Advancement of Science. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 392, 884, 467]]<|/det|> +[16] Rosenthal, P.B., Henderson, R.: Optimal Determination of Particle Orientation, Absolute Hand, and Contrast Loss in Single- particle Electron Cryomicroscopy. Journal of Molecular Biology 333(4), 721- 745 (2003). https://doi.org/10.1016/j.jmb.2003.07.013. Accessed 2023- 01- 23 + +<|ref|>text<|/ref|><|det|>[[66, 480, 884, 554]]<|/det|> +[17] Kuo, I.A.M., Glaeser, R.M.: Development of methodology for low exposure, high resolution electron microscopy of biological specimens. Ultramicroscopy 1(1), 53- 66 (1975). https://doi.org/10.1016/S0304- 3991(75)80007- 6. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 567, 884, 641]]<|/det|> +[18] Unwin, P.N.T., Henderson, R.: Molecular structure determination by electron microscopy of unstained crystalline specimens. Journal of Molecular Biology 94(3), 425- 440 (1975). https://doi.org/10.1016/S0022- 2836(75)90212- 0. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 654, 884, 747]]<|/det|> +[19] Henderson, R.: The potential and limitations of neutrons, electrons and X- rays for atomic resolution microscopy of unstained biological molecules. Quarterly Reviews of Biophysics 28(2), 171- 193 (1995). https://doi.org/10.1017/S003358350000305X. Publisher: Cambridge University Press. Accessed 2023- 03- 23 + +<|ref|>text<|/ref|><|det|>[[66, 760, 884, 833]]<|/det|> +[20] Glaeser, R.M.: Review: Electron Crystallography: Present Excitement, a Nod to the Past, Anticipating the Future. Journal of Structural Biology 128(1), 3- 14 (1999). https://doi.org/10.1006/jsbj.1999.4172. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[66, 846, 884, 903]]<|/det|> +[21] Glaeser, R.M.: How good can cryo- EM become? Nature Methods 13(1), 28- 32 (2016). https://doi.org/10.1038/nmeth.3695. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 04 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 80, 884, 155]]<|/det|> +[22] Paulsen, C.E., Armache, J.- P., Gao, Y., Cheng, Y., Julius, D.: Structure of the TRPA1 ion channel suggests regulatory mechanisms. Nature 520(7548), 511–517 (2015). https://doi.org/10.1038/nature14367. Number: 7548 Publisher: Nature Publishing Group. Accessed 2023- 02- 20 + +<|ref|>text<|/ref|><|det|>[[70, 168, 884, 259]]<|/det|> +[23] Tan, Y.Z., Baldwin, P.R., Davis, J.H., Williamson, J.R., Potter, C.S., Carragher, B., Lyumkis, D.: Addressing preferred specimen orientation in single- particle cryo- EM through tilting. Nature Methods 14(8), 793–796 (2017). https://doi.org/10.1038/nmeth.4347. Number: 8 Publisher: Nature Publishing Group. Accessed 2023- 02- 21 + +<|ref|>text<|/ref|><|det|>[[70, 272, 884, 400]]<|/det|> +[24] Lee, Y., Wiriyasermkul, P., Jin, C., Quan, L., Ohgaki, R., Okuda, S., Kusakizako, T., Nishizawa, T., Oda, K., Ishitani, R., Yokoyama, T., Nakane, T., Shirouzu, M., Endou, H., Nagamori, S., Kanai, Y., Nureki, O.: Cryo- EM structure of the human L- type amino acid transporter 1 in complex with glycoprotein CD98hc. Nature Structural & Molecular Biology 26(6), 510–517 (2019). https://doi.org/10.1038/s41594- 019- 0237- 7. Number: 6 Publisher: Nature Publishing Group. Accessed 2023- 02- 27 + +<|ref|>text<|/ref|><|det|>[[70, 413, 884, 541]]<|/det|> +[25] Kim, J., Tan, Y.Z., Wicht, K.J., Erramilli, S.K., Dhingra, S.K., Okombo, J., Vendome, J., Hagenah, L.M., Giacometti, S.I., Warren, A.L., Nosol, K., Roope, P.D., Potter, C.S., Carragher, B., Kossiakoff, A.A., Quick, M., Fidock, D.A., Mancia, F.: Structure and drug resistance of the Plasmodium falciparum transporter PfCRT. Nature 576(7786), 315–320 (2019). https://doi.org/10.1038/s41586- 019- 1795- x. Number: 7786 Publisher: Nature Publishing Group. Accessed 2023- 02- 27 + +<|ref|>text<|/ref|><|det|>[[70, 555, 884, 664]]<|/det|> +[26] Faust, B., Billesbølle, C.B., Suomivuori, C.- M., Singh, I., Zhang, K., Hoppe, N., Pinto, A.F.M., Diedrich, J.K., Muftuoglu, Y., Szkudlinski, M.W., Saghatelian, A., Dror, R.O., Cheng, Y., Manglik, A.: Autoantibody mimicry of hormone action at the thyrotropin receptor. Nature 609(7928), 846–853 (2022). https://doi.org/10.1038/s41586- 022- 05159- 1. Number: 7928 Publisher: Nature Publishing Group. Accessed 2023- 02- 21 + +<|ref|>text<|/ref|><|det|>[[70, 678, 884, 750]]<|/det|> +[27] Diver, M.M., Cheng, Y., Julius, D.: Structural insights into TRPM8 inhibition and desensitization. Science 365(6460), 1434–1440 (2019). https://doi.org/10.1126/science.aax6672. Publisher: American Association for the Advancement of Science. Accessed 2023- 02- 21 + +<|ref|>text<|/ref|><|det|>[[70, 765, 884, 837]]<|/det|> +[28] Zhou, Y., Moscovich, A., Bendory, T., Bartesaghi, A.: Unsupervised particle sorting for high- resolution single- particle cryo- EM. Inverse Problems 36(4), 044002 (2020). https://doi.org/10.1088/1361- 6420/ab5ec8. Publisher: IOP Publishing. Accessed 2022- 12- 21 + +<|ref|>text<|/ref|><|det|>[[70, 852, 884, 907]]<|/det|> +[29] Méndez, J., Garduño, E., Carazo, J.M., S. Sorzano, C.O.: Identification of incorrectly oriented particles in cryo- EM single particle analysis. Journal of Structural Biology 213(3), 107771 (2021). https://doi.org/10.1016/j. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[157, 81, 533, 100]]<|/det|> +jsb.2021.107771. Accessed 2023- 02- 23 + +<|ref|>text<|/ref|><|det|>[[110, 113, 884, 204]]<|/det|> +[30] Henderson, R.: Avoiding the pitfalls of single particle cryo- electron microscopy: Einstein from noise. Proceedings of the National Academy of Sciences 110(45), 18037- 18041 (2013). https://doi.org/10.1073/pnas.1314449110. Publisher: Proceedings of the National Academy of Sciences. Accessed 2023- 02- 25 + +<|ref|>text<|/ref|><|det|>[[110, 218, 884, 293]]<|/det|> +[31] Pintilie, G., Zhang, K., Su, Z., Li, S., Schmid, M.F., Chiu, W.: Measurement of atom resolvability in cryo- EM maps with Q- scores. Nature Methods 17(3), 328- 334 (2020). https://doi.org/10.1038/s41592- 020- 0731- 1. Number: 3 Publisher: Nature Publishing Group. Accessed 2023- 02- 25 + +<|ref|>text<|/ref|><|det|>[[110, 305, 884, 416]]<|/det|> +[32] Zheng, L., Liu, N., Gao, X., Zhu, W., Liu, K., Wu, C., Yan, R., Zhang, J., Gao, X., Yao, Y., Deng, B., Xu, J., Lu, Y., Liu, Z., Li, M., Wei, X., Wang, H.- W., Peng, H.: Uniform thin ice on ultraflat graphene for high- resolution cryo- EM. Nature Methods 20(1), 123- 130 (2023). https://doi.org/10.1038/s41592- 022- 01693- y. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 20 + +<|ref|>text<|/ref|><|det|>[[110, 428, 884, 504]]<|/det|> +[33] Gyobu, N., Tani, K., Hiroaki, Y., Kamegawa, A., Mitsuoka, K., Fujiyoshi, Y.: Improved specimen preparation for cryo- electron microscopy using a symmetric carbon sandwich technique. Journal of Structural Biology 146(3), 325- 333 (2004). https://doi.org/10.1016/j.jsb.2004.01.012 + +<|ref|>text<|/ref|><|det|>[[110, 515, 884, 590]]<|/det|> +[34] Bock, L.V., Grubmüller, H.: Effects of cryo- EM cooling on structural ensembles. Nature Communications 13(1), 1709 (2022). https://doi.org/10.1038/s41467- 022- 29332- 2. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +<|ref|>text<|/ref|><|det|>[[110, 602, 884, 677]]<|/det|> +[35] Ma, H., Jia, X., Zhang, K., Su, Z.: Cryo- EM advances in RNA structure determination. Signal Transduction and Targeted Therapy 7(1), 1- 6 (2022). https://doi.org/10.1038/s41392- 022- 00916- 0. Number: 1 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +<|ref|>text<|/ref|><|det|>[[110, 689, 884, 781]]<|/det|> +[36] Zhang, K., Pintilie, G.D., Li, S., Schmid, M.F., Chiu, W.: Resolving individual atoms of protein complex by cryo- electron microscopy. Cell Research 30(12), 1136- 1139 (2020). https://doi.org/10.1038/s41422- 020- 00432- 2. Number: 12 Publisher: Nature Publishing Group. Accessed 2023- 04- 21 + +<|ref|>text<|/ref|><|det|>[[110, 794, 884, 886]]<|/det|> +[37] Vulović, M., Ravelli, R.B.G., van Vliet, L.J., Koster, A.J., Lazić, I., Lücken, U., Rullgard, H., Öktem, O., Rieger, B.: Image formation modeling in cryo- electron microscopy. Journal of Structural Biology 183(1), 19- 32 (2013). https://doi.org/10.1016/j.jsb.2013.05.008. Accessed 2023- 02- 25 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 80, 884, 210]]<|/det|> +[38] Li, Q., Pellegrino, J., Lee, D.J., Tran, A.A., Chaires, H.A., Wang, R., Park, J.E., Ji, K., Chow, D., Zhang, N., Brilot, A.F., Biel, J.T., van Zundert, G., Borrelli, K., Shinabarger, D., Wolfe, C., Murray, B., Jacobson, M.P., Mühle, E., Chesneau, O., Fraser, J.S., Seiple, I.B.: Synthetic group A streptogramin antibiotics that overcome Vat resistance. Nature 586(7827), 145–150 (2020). https://doi.org/10.1038/s41586-020-2761-3. Number: 7827 Publisher: Nature Publishing Group. Accessed 2023- 02- 25 + +<|ref|>text<|/ref|><|det|>[[70, 222, 884, 295]]<|/det|> +[39] Henderson, R., Glaeser, R.M.: Quantitative analysis of image contrast in electron micrographs of beam- sensitive crystals. Ultramicroscopy 16(2), 139–150 (1985). https://doi.org/10.1016/0304- 3991(85)90069- 5. Accessed 2023- 03- 23 + +<|ref|>text<|/ref|><|det|>[[70, 309, 884, 381]]<|/det|> +[40] Henderson, R.: Image contrast in high- resolution electron microscopy of biological macromolecules: TMV in ice. Ultramicroscopy 46(1), 1–18 (1992). https://doi.org/10.1016/0304- 3991(92)90003- 3. Accessed 2023- 04- 04 + +<|ref|>text<|/ref|><|det|>[[70, 395, 884, 470]]<|/det|> +[41] Zivanov, J., Nakane, T., Scheres, S.H.W.: A Bayesian approach to beam- induced motion correction in cryo- EM single- particle analysis. IUCrJ 6(1), 5–17 (2019). https://doi.org/10.1107/S205225251801463X. Publisher: International Union of Crystallography. Accessed 2023- 04- 04 + +<|ref|>text<|/ref|><|det|>[[70, 483, 884, 556]]<|/det|> +[42] Grant, T., Grigorieff, N.: Measuring the optimal exposure for single particle cryo- EM using a 2.6 Å reconstruction of rotavirus VP6. eLife 4, 06980 (2015). https://doi.org/10.7554/eLife.06980. Publisher: eLife Sciences Publications, Ltd. Accessed 2023- 04- 04 + +<|ref|>text<|/ref|><|det|>[[70, 570, 884, 661]]<|/det|> +[43] Zheng, S.Q., Palovcak, E., Armache, J.- P., Verba, K.A., Cheng, Y., Agard, D.A.: MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nature Methods 14(4), 331–332 (2017). https://doi.org/10.1038/nmeth.4193. Number: 4 Publisher: Nature Publishing Group. Accessed 2023- 04- 04 + +<|ref|>text<|/ref|><|det|>[[70, 675, 884, 730]]<|/det|> +[44] Reimer, L.: Transmission Electron Microscopy: Physics of Image Formation and Microanalysis. Springer, ??? (2013). Google- Books- ID: CUFSCAAAQBAJ + +<|ref|>text<|/ref|><|det|>[[70, 744, 884, 835]]<|/det|> +[45] Yip, K.M., Fischer, N., Paknia, E., Chari, A., Stark, H.: Breaking the next Cryo- EM resolution barrier – Atomic resolution determination of proteins! bioRxiv. Pages: 2020.05.21.106740 Section: New Results (2020). https://doi.org/10.1101/2020.05.21.106740. https://www.biorxiv.org/content/10.1101/2020.05.21.106740v1 Accessed 2023- 04- 15 + +<|ref|>text<|/ref|><|det|>[[70, 850, 884, 905]]<|/det|> +[46] Danev, R., Yanagisawa, H., Kikkawa, M.: Cryo- Electron Microscopy Methodology: Current Aspects and Future Directions. Trends in Biochemical Sciences 44(10), 837–848 (2019). https://doi.org/10.1016/j.tibs.2019. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[157, 81, 639, 100]]<|/det|> +04.008. Publisher: Elsevier. Accessed 2023- 04- 21 + +<|ref|>text<|/ref|><|det|>[[110, 113, 885, 204]]<|/det|> +[47] Zivanov, J., Nakane, T., Scheres, S.H.W.: Estimation of high- order aberrations and anisotropic magnification from cryo- EM data sets in RELION- 3.1. IUCrJ 7(2), 253- 267 (2020). https://doi.org/10.1107/S2052252520000081. Publisher: International Union of Crystallography. Accessed 2023- 04- 21 + +<|ref|>text<|/ref|><|det|>[[110, 218, 885, 328]]<|/det|> +[48] Carragher, B., Cheng, Y., Frost, A., Glaeser, R.m., Lander, G.c., Nogales, E., Wang, H.- W.: Current outcomes when optimizing 'standard' sample preparation for single- particle cryo- EM. Journal of Microscopy 276(1), 39- 45 (2019). https://doi.org/10.1111/jmi.12834. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/jmi.12834. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[110, 341, 885, 453]]<|/det|> +[49] Drulyte, I., Johnson, R.M., Hesketh, E.L., Hurdiss, D.L., Scarff, C.A., Porav, S.A., Ranson, N.A., Muench, S.P., Thompson, R.F.: Approaches to altering particle distributions in cryo- electron microscopy sample preparation. Acta Crystallographica Section D: Structural Biology 74(6), 560- 571 (2018). https://doi.org/10.1107/S2059798318006496. Number: 6 Publisher: International Union of Crystallography. Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[110, 465, 885, 539]]<|/det|> +[50] Kim, L.Y., Rice, W.J., Eng, E.T., Kopylov, M., Cheng, A., Raczkowski, A.M., Jordan, K.D., Bobe, D., Potter, C.S., Carragher, B.: Benchmarking cryo- EM Single Particle Analysis Workflow. Frontiers in Molecular Biosciences 5 (2018). Accessed 2023- 04- 06 + +<|ref|>text<|/ref|><|det|>[[110, 551, 885, 626]]<|/det|> +[51] Weissenberger, G., Henderikx, R.J.M., Peters, P.J.: Understanding the invisible hands of sample preparation for cryo- EM. Nature Methods 18(5), 463- 471 (2021). https://doi.org/10.1038/s41592- 021- 01130- 6. Number: 5 Publisher: Nature Publishing Group. Accessed 2022- 11- 14 + +<|ref|>text<|/ref|><|det|>[[110, 639, 885, 713]]<|/det|> +[52] Callaway, E.: The revolution will not be crystallized: a new method sweeps through structural biology. Nature 525(7568), 172- 174 (2015). https://doi.org/10.1038/525172a. Number: 7568 Publisher: Nature Publishing Group. Accessed 2023- 04- 06 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[303, 115, 690, 700]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 707, 884, 878]]<|/det|> +
Fig. 2 CryoSieve outperformed other algorithms in terms of FSC-based resolutions and Q-scores. We compared the density maps reconstructed from retained particles obtained by CryoSieve (indigo) and NCC (green), along with random (orange) as the baseline, at different retention ratios. Density maps were ab initio reconstructed by CryoSPARC after discarding the published refined Euler angles deposited on EMPIAR. Four metrics were employed for measuring density map quality: FSC-based resolutions, including model-to-map (solid lines with squares) and two-half-maps (solid lines with diamonds) resolutions, are shown in the first column; meanwhile, Q-scores of the raw maps (dashed lines with circles) and the sharpened maps (solid lines with circles) are shown in the second column, as Q-score varies by the B-factor used for sharpening. B-factors for sharpening were self-determined by CryoSPARC. The iterations where CryoSieve obtained the finest subset, determined by comprehending these metrics, are labeled with hatched bars.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[160, 277, 840, 568]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 576, 884, 720]]<|/det|> +
Fig. 3 The two-dimensional resolution distribution between retained and removed particles was compared. The third iteration of CryoSieve achieved a retention ratio of \(51.2\%\) and a removal ratio of \(48.8\%\) , resulting in a similar number of particles for retention and removal. These two categories underwent CryoSPARC 2D clustering and averaging, i.e., 2D classification, with the number of 2D classes set to 50. All six experimental datasets were tested. CryoSPARC reported the 2D resolution of each 2D class, along with the number of particle images belonging to it. We statistically analyzed the number of particles belonging to each 2D resolution and plotted histograms, demonstrating the difference between retained (steel blue) and removed (crimson) particles in terms of 2D resolution distribution.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 106, 845, 565]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 576, 884, 888]]<|/det|> +
Fig. 4 CryoSieve prioritizes the removal of radiation-damaged particles. (a), InSilicoTEM generated simulated particle images with different levels of radiation damage. The parameter \(M_{b}\) in InSilicoTEM, ranging from 0 to 4, mimicked different levels of radiation damage, with higher \(M_{b}\) leading to weaker high-frequency signals and blurrier particles (clean, upper row). Noise, also generated by InSilicoTEM through physical process simulation, was then added to the clean particles to obtain noisy particles for testing (noisy, lower row). (b), The proportions of particles with varying levels of radiation damage (distinguished by colors) in retained particles at different retention ratios (labeled on the left) are shown. The retained particles obtained by CryoSieve (left horizontal bars) and NCC (right horizontal bars) were compared. NCC performed acceptably for particles with high radiation damage, but was unable to distinguish particles with relatively low radiation damage. Meanwhile, CryoSieve sequentially sieved out particles from high to low radiation damage. The accuracy (the proportion of zero radiation damage particles) of CryoSieve was 91.7% at iteration 7, while that of NCC was only 61.5%. (c), Side chains of density maps reconstructed by cryoSPARC using retained particles obtained by CryoSieve (indigo) and NCC (green) were compared. The atomic model (PDB 6PCQ) was fitted into the density maps. The two reconstruction maps are compared at an identical contour. Red arrows emphasize differences between the two maps. (d) Model-to-map FSCs of reconstructed density maps using retained particles were compared, with retention ratio of 21.0% (iteration 7). Retention particles were obtained by CryoSieve (indigo), NCC (green) and random (baseline method, orange), respectively. The FSC threshold (FSC = 0.5) was depicted as a horizontal dashed line.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 78, 256, 100]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[113, 113, 872, 157]]<|/det|> +## Details of comparing the performance of particle sorting algorithms. + +<|ref|>text<|/ref|><|det|>[[112, 164, 884, 380]]<|/det|> +Since cryo- EM single- particle image processing software has experienced rapid development in the past few years, some of the final stacks deposited in EMPIAR can be better processed by state- of- the- art algorithms. To eliminate effects from different refinement software and their versions, ensuring fair comparisons between various particle sorting algorithms, the final stacks deposited on EMPIAR were reprocessed under a standard workflow using CryoSPARC v4.1.0 following a standard workflow. For hemagglutinin, the initial model was generated by low- pass filtering its atomic model to 30A, while for the other proteins, initial models were generated by arbitrary random initialization using CryoSPARC. Then, uniform refinement was applied for TRPA1, TRPM8, hemagglutinin and LAT1, while non- uniform refinement was applied for pfCRT and TSHR- Gs. + +<|ref|>text<|/ref|><|det|>[[111, 380, 884, 577]]<|/det|> +To enable unbiased comparisons of density maps before and after particle sorting, the retained particles obtained from each particle sorting algorithm underwent identical refinement procedures, as previously described using CryoSPARC v4.1.0 in the standard workflow. The reconstructed density maps were used for subsequent measurements. To avoid the potential influence of the "Einstein- from- noise" effect, the former Euler angles were discarded, and new sets of Euler angles were determined through refinement of the retained particles. Moreover, in order to maintain independence between the two half sets and ensure that the Fourier Shell Correlation (FSC) served as the golden standard, half- set splits were preserved throughout the subsequent procedure by turning off the option "Force re- do GS split". + +<|ref|>text<|/ref|><|det|>[[111, 578, 883, 650]]<|/det|> +The reconstructed density maps were evaluated by several metrics, including FSC- based resolution and Q- score. CryoSPARC produced two raw half maps and an auto- postprocessed density map (FSC- weighted, B- factor sharpened, two half sets averaged), accompanied by reporting half- maps FSC. + +<|ref|>text<|/ref|><|det|>[[110, 650, 883, 830]]<|/det|> +FSC- based metric includes half- maps FSC (directly reported by CryoSPARC) and model- to- map FSC. Map- to- model FSC resolution was calculated using the following procedure, with the auto- postprocessed density map as input. The corresponding atomic model of the dataset was converted to the ground- truth density map by the molmap function of Chimera at Nyquist resolution. The mask was generated from the ground- truth density map (after low- pass filtering to 8 Å, extending by 4 pixels and applying a cosine- edge of 4 pixels) using Relion. Model- to- map FSC curves were determined between the input density map (obscured by the mask) and the ground- truth density map. The resolution threshold of the map- to- model FSC was set to 0.5. + +<|ref|>text<|/ref|><|det|>[[110, 831, 882, 903]]<|/det|> +As Q- score is sensitive to B- factor sharpening, the Q- scores of both the raw maps and the auto- postprocessed maps were measured. The auto- postprocessed maps were directly provided by CryoSPARC, while the raw maps were obtained by first averaging the two raw half maps provided by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 245]]<|/det|> +CryoSPARC, then low- pass filtering them to an appropriate resolution, in order to eliminate the impact of varying noise intensities on the density maps. The low- pass filtering threshold frequency ranged from \(0.3\mathrm{\AA}\) to \(0.5\mathrm{\AA}\) higher than the CryoSPARC reported half- maps FSC resolution, thus ensuring the retention of useful signals. Specifically, the threshold frequency for TRPA1 was \(3.5\mathrm{\AA}\) , for TRPM8 and TSHR- Gs it was \(2.7\mathrm{\AA}\) , for hemagglutinin it was \(3.4\mathrm{\AA}\) , for pfCRT it was \(3.0\mathrm{\AA}\) , and for LAT it was \(2.8\mathrm{\AA}\) . Q- score was calculated using the MAPQ plugin for UCSF Chimera, with all parameters set to their default values. + +<|ref|>text<|/ref|><|det|>[[111, 244, 884, 281]]<|/det|> +All conversions between CryoSPARC and Relion were performed using the pyem script. + +<|ref|>sub_title<|/ref|><|det|>[[112, 300, 436, 320]]<|/det|> +## CryoSieve's parameters. + +<|ref|>text<|/ref|><|det|>[[111, 328, 884, 598]]<|/det|> +CryoSieve is iteratively performs 3D reconstruction and particle sieving, while maintaining independence between two half sets by independently sieving each set of particles. 3D reconstructions of each subset were performed using Relion v4.0- beta- 2, with the option "—subset" to preserve the half- set splitting. A mask, generated from the atomic model using RELION (low- pass filtered to 8 Å), was applied to the reconstructed raw density map to obtain \(x^{(k - 1)}\) in Equation 2.1 of the CryoSieve score. The same mask was applied to other particle sorting algorithms such as NCC and AGC, to ensure fair comparisons. Subsequently, particles were sieved out based on the ascending order of the CryoSieve score. In total, nine iterations were carried out, with each iteration retaining \(80\%\) of the particles from the previous iteration. The cutoff frequency of the highpass operator \(H^{(k)}\) increased linearly as the iteration progressed. For all datasets, except for LAT1, the initial cutoff frequency was set at \(40\mathrm{\AA}\) , and the final cutoff frequency was \(3\mathrm{\AA}\) . For LAT1, the initial cutoff frequency was \(50\mathrm{\AA}\) , and the final cutoff frequency was also \(3\mathrm{\AA}\) . + +<|ref|>sub_title<|/ref|><|det|>[[112, 617, 436, 637]]<|/det|> +## AGC's particle removal. + +<|ref|>text<|/ref|><|det|>[[111, 645, 884, 862]]<|/det|> +Incorrectly oriented particle removal by AGC was carried out using Scipion v3.0.12, following the instructions provided in its user manual. The initial volumes required for Scipion's AGC algorithms were the reconstructed maps of all particles in final stack using Relion v4.0- beta- 2, applied with a mask identical to that used in CryoSieve and NCC. The "symmetry group" parameter of AGC was set to the corresponding symmetry of the structure, while the remaining parameters were set to their default values. Since the number of particles in the LAT1 dataset is too large to perform AGC without reporting error, we partitioned them into quarter- datasets and performed AGC algorithm within each subset. The remaining particles in the quarter- datasets were used for ab initio reconstruction in cryoSPARC. Particle removal of other datasets were performed using all particles in final stacks as a whole. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 312, 149]]<|/det|> +Supplementalmaterial.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/images_list.json b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..5b77b4c6515848412505966b816cf9c385ec06c4 --- /dev/null +++ b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/images_list.json @@ -0,0 +1,167 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. (a) Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 1) measured under high pressures. The inset is an enlarged view of the resistance curve, showing the superconducting transition in detail. (b) The resistance curves for Ti Sample 2. (c) The resistance curve measured at 248 GPa, where the derivative of the resistance with respect to temperature dR/dT is plotted to clearly show the onset Tc.", + "footnote": [], + "bbox": [ + [ + 310, + 108, + 683, + 740 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. (a) Temperature dependence of the electrical resistance of Ti metal measured under different magnetic fields at the fixed pressure of 248 GPa. (b) Upper critical field versus superconducting transition temperature of \\(\\mathrm{Tc}^{2\\mathrm{ero}}\\) . The line is a fit obtained using the Ginzburg–Landau function.", + "footnote": [], + "bbox": [ + [ + 316, + 112, + 655, + 515 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. (a,b) Hall resistance as a function of magnetic field measured under different pressure. (c-e) Carrier density, resistance at fixed temperature, and onset Tc versus pressure, respectively. The yellow and blue areas mark the anomaly of these data, indicating the critical pressures where the phase transitions occur.", + "footnote": [], + "bbox": [ + [ + 305, + 115, + 702, + 606 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. The superconducting critical transition temperature (Tc) of Ti metal under compression up to 320 GPa plotted against the background of stability fields of the indicated structural phases.", + "footnote": [], + "bbox": [ + [ + 206, + 125, + 752, + 430 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Calculated electronic band structures of three representative Ti phases: (a) Ti-ω at 20 GPa, (b) Ti-γ at 100 GPa and (c) Ti-δ at 180 GPa. The contributions from the s-electron and d-electron states are shown by red and gray circles, respectively, and the circle areas are proportional to the weights of the corresponding band states. Energy is measured relative to the Fermi energy \\(\\mathrm{E}_{\\mathrm{F}}\\) .", + "footnote": [], + "bbox": [ + [ + 195, + 555, + 861, + 680 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Fig.S1 The photo showing sample assembly on the top of diamond cullet..", + "footnote": [], + "bbox": [ + [ + 275, + 247, + 690, + 517 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "Fig.S2 Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 5) measured under high pressures.", + "footnote": [], + "bbox": [ + [ + 218, + 135, + 777, + 410 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_2.jpg", + "caption": "Fig. S3 (a) The superconducting transitions measured under different magnetic fields at 154 GPa. (b) at 180 GPa.", + "footnote": [], + "bbox": [ + [ + 170, + 518, + 825, + 688 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_3.jpg", + "caption": "Fig. S4 The in-situ high pressure X-ray diffraction patterns with the highest experimental pressure of 225 GPa, using the wavelength of 0.3344 Å. The pressure induced phase transition sequence from Ti-α phase, to Ti-ω phase, to Ti-γ phase, and then to Ti-δ phase has been observed, which is consistent with the previously reported results (Y. Akahama et al. 2002).", + "footnote": [], + "bbox": [ + [ + 300, + 115, + 741, + 366 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_4.jpg", + "caption": "Fig. S5. The temperature dependence of resistance measured at 18 GPa. The resistance in the low temperature range of 6-70 K is fitted (red line) with the formula of \\(R = R_0 + AT^n\\) , where \\(R_0\\) is the residual resistance, \\(A\\) is the coefficient of the power law, and \\(n = 3.1\\) is the extracted exponent.", + "footnote": [], + "bbox": [ + [ + 266, + 508, + 690, + 737 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_5.jpg", + "caption": "Fig. S6. The superconducting critical transition temperature (Tc) of the indicated Ti phases in their stability fields in the [20, 220] GPa hydrostatic pressure range. The calculated crystal structure of Ti is sensitive to the pressure conditions, and under hydrostatic pressures at or above 160 GPa, the Ti-δ phase essentially relaxes to the Ti-β (bcc) phase without considering any stress anisotropy.", + "footnote": [], + "bbox": [ + [ + 295, + 108, + 688, + 318 + ] + ], + "page_idx": 19 + } +] \ No newline at end of file diff --git a/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b.mmd b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f73144c29d90f526d02c8d90256b0ad136d58425 --- /dev/null +++ b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b.mmd @@ -0,0 +1,260 @@ + +# Robust record high Tc elemental superconductivity in titanium + +Changling Zhang Chinese Academy of Sciences + +Xin He Chinese Academy of Sciences + +Chang Liu Jilin University https://orcid.org/0000- 0003- 0824- 1098 + +Zhiwen Li Chinese Academy of Sciences + +Ke Lu Chinese Academy of Sciences + +Sijia Zhang Chinese Academy of Sciences + +Shaomin Feng Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China + +Xiancheng WANG Institute of Physics https://orcid.org/0000- 0001- 6263- 4963 + +Yi Peng Beijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences + +Youwen Long Institute of Physics https://orcid.org/0000- 0002- 8587- 7818 + +Richeng Yu Institute of Physics, Chinese Academy of Sciences + +Luhong Wang Harbin Institute of Technology + +Vitali Prakapenka University of Chicago + +Stella Chariton The University of Chicago + +Quan Li Jilin University https://orcid.org/0000- 0002- 7724- 1289 + +Haozhe Liu + +<--- Page Split ---> + +Center for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 5720- 2031 + +Changfeng Chen Department of Physics and Astronomy, University of Nevada Changqing Jin ( \(\square\) Jin@iphy.ac.cn ) Chinese Academy of Sciences https://orcid.org/0000- 0001- 8097- 9156 + +## Article + +# Keywords: + +Posted Date: May 18th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1630556/v1 + +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on September 15th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33077- 3. + +<--- Page Split ---> + +# Robust record high Tc elemental superconductivity in titanium + +C. +L. Zhanga1,2, +X. Hea1,2,3, +C. Liua4, +Z. +W. Li1,2, +K. Lu1,2, +S. +J. Zhang1, +S. +M. Feng1, +X. +C. Wanga1,2, +Y. Peng1,2, +Y. +W. Long1,2,3, +R.C.Yu1,2, +L. +H. Wang5, +V. +B. Prakapenka6, +S. Chariton6, +Q. Li4, +H. +Z. Liu7, +C. +F. Chen8, +C. +Q. Jina1,2,3 + +1Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China 2School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China 3 Songshan Lake Materials Laboratory, Dongguan 523808, China 4International Center for Computational Method and Software, College of Physics, and International Center of Future Science, Jilin University, Changchun 130012, China 5Harbin Institute of Technology, Harbin 130001, China 6Center for Advanced Radiations Sources, University of Chicago, Chicago, Illinois 60637, USA 7Center for High Pressure Science & Technology Advanced Research, Beijing 100094, China 8Department of Physics and Astronomy, University of Nevada, Las Vegas, Nevada 89154, USA + +Searching for superconductivity in single element with high Tc is of great interest both for fundamental science & potential applications due to the easy recipe. It is especially interesting to explore high Tc superconductivity in transition metal elements wherein the conventional BCS theories suggest that \(d\) electrons are not favorable to form high Tc superconducting state. Here we report discovery of surprisingly robust superconductivity up to 310 GPa with almost constant Tc above 20K in a wide pressure range from 108GPa to 240 GPa in titanium. The maximum Tconset above 26 K and zero- resistance Tczero of 21 K are record- high values hitherto achieved among elemental superconductors. This finding is in sharp contrast to the theoretical predications based on the pristine electron- phonon coupling scenario. Measured normal state conductivity reveals significant \(s\) - \(d\) scattering. Quantum calculations show strong \(s\) - \(d\) transfer and \(d\) - band dominance, indicating correlation driven contributions to the robust high- Tc superconductivity seen in dense titanium. Different from most high- Tc superconductors that are either complex ceramic or intermetallic compounds, titanium metal is much easier to process and hosts notably higher and more robust Tc and critical magnetic field Hc compared to the widely used superconducting NbTi alloy. This study opens a fresh promising avenue for rational design and discovery of high- Tc superconductors among simple materials via pressure tuned unconventional mechanism. + +<--- Page Split ---> + +Titanium (Ti) metal has long attracted tremendous scientific interests because of its combined properties of light weight, high strength and corrosion resistance. As an advanced metallic structural material, Ti and its alloys find wide use in the fields of aerospace, biomedicine and at extreme conditions (1- 3). High pressure can modify crystal structures which, in turn, may lead to new functionalities. At ambient pressure and room temperature, Ti crystalizes in a hexagonal close- packed (hcp) structure (Ti- \(\alpha\) phase) (4). Under applied pressure, Ti undergoes structural transitions in the sequence of Ti- \(\alpha\) , Ti- \(\alpha\) , Ti- \(\gamma\) , Ti- \(\delta\) and Ti- \(\beta\) phases, where Ti- \(\alpha\) phase is a hexagonal structure, Ti- \(\gamma\) and Ti- \(\delta\) phases are orthorhombic and Ti- \(\beta\) phase is body- centered cubic (5- 9). The \(\alpha\) - to- \(\alpha\) transition occurs around 8 GPa (5, 6), and the Ti- \(\alpha\) phase is stable up to about 100 GPa, then transforms into Ti- \(\gamma\) phase (6, 10), which further transforms into Ti- \(\delta\) phase at \(\sim 140\) GPa (6), before cubic Ti- \(\beta\) phase stabilizes at 243 GPa (9). + +Superconductivity (SC) in high- pressure phases of Ti metal was previously reported to have a measured maximal critical temperature (Tc) of 3.5 K at 56 GPa (11); early theoretical calculations based on the electron- phonon coupling mechanism predicate that the Tc for Ti metal is capped at about 5 K for all the known high pressure phases (12). Generally, compression of crystal lattice has markedly different effects on the 4s and 3d bands, prompting notable \(s\) - \(d\) electron transfer. The narrower \(d\) - bands possess stronger correlation characters, while the \(s\) - \(d\) transfer tends to enhance electronic density of state (DOS) near the Fermi level favoring SC (13- 17). Here, we report a surprising experimental observation of dramatic pressure enhanced SC in Ti over a wide pressure range with the maximal Tc above 26 K, setting a new record among elemental superconductors. Measured normal- state conductivity results and analysis of electronic and superconducting properties indicate prominent influence of the s- d transfer and d- band driven correlation effects on the significantly enhanced and unusually robust SC in highly compressed Ti metal. + +The electrical resistance and Hall effect measurements were performed using the four- probe method in Van der Pauw way as shown in Fig. S1. The applied current is \(100 \mu \mathrm{A}\) . Diamond anvil + +<--- Page Split ---> + +cells were used to produce high pressures. A variant of anvils with beveled culet size of \(20 / 140 / 300\mu \mathrm{m}\) , \(30 / 140 / 300\mu \mathrm{m}\) or \(50 / 140\) \(300\mu \mathrm{m}\) are adopted in the experiments. A plate of T301 stainless steel that is covered with mixture of cBN powder & epoxy as insulate layer was used as the gasket. A hole of approximately 15 to \(30\mu \mathrm{m}\) in diameter depending on top culet size was drilled in the center of the gasket to serve as ample chamber. The hBN powder was generally used as pressure- transmitting medium that filled in the high- pressure chamber. We used the ATHENA procedure to produce the specimen assembly (18). Four trapezoidal Pt foils with thickness approximately \(0.5\mu \mathrm{m}\) as the inner electrode were deposited on the culet surface. Cross- shaped Ti specimens with side lengths \(\sim 10\mu \mathrm{m}*10\mu \mathrm{m}\) and thickness of \(1\mu \mathrm{m}\) were stacked on the electrodes. Pressure was calibrated by the shift of the first- order Raman edge frequency from the diamond cutlet. Diamond anvil cells were put in a MagLab system that provides synergetic extreme environments with temperatures from \(300\mathrm{K}\) to \(1.5\mathrm{K}\) and magnetic fields up to \(9\mathrm{T}\) for the transport measurements (18- 21). + +In- situ high- pressure angle- dispersive X- ray diffraction data were collected at room temperature at GSECARS of Advanced Photon Source at the Argonne National Laboratory. The x- ray with the wavelength \(\lambda = 0.3344\mathrm{\AA}\) was focused down to a spot of \(\sim 3\mu \mathrm{m}\) in diameter on the sample. A symmetric diamond anvil cell with beveled anvil (50/300 \(\mu \mathrm{m}\) ) was used. Rhenium steel gasket was pre- pressed to a thickness of \(20\mu \mathrm{m}\) , and a hole of diameter of \(15\mu \mathrm{m}\) was drilled at the center to serve as sample chamber, which was then filled with Ti power mixed with Pt. Pressure was calibrated using the equitation of state of both Re and Pt. + +To assess structural, electronic, and phonon- mediated superconducting properties of Ti metal under pressure, we employed the latest computational techniques to determine the total energy, lattice dynamics and electron- phonon coupling using the QUANTUM ESPRESSO code (22), with improved description over previously reported results (12). Superconducting critical temperature Tc has been evaluated based on the Eliashberg theory of superconductivity (23, 24), using the + +<--- Page Split ---> + +following formula that McMillan derived (25) and later modified by Allen and Dynes (26), + +\[T_{c} = \frac{\omega_{log}}{1.20} exp[-\frac{1.04(1 + \lambda)}{\lambda - \mu^{*}(1 + 0.62\lambda)} ]\] + +where \(\omega_{log}\) is a logarithmically averaged characteristic phonon frequency, and \(\mu^{*}\) is the Coulomb pseudopotential which describes the effective electron- electron repulsion (27). This equation is generally accurate for materials with EPC parameter \(\lambda\) at 1.5 or less (28- 31), which is satisfied in the present study. The Coulomb pseudopotential \(\mu^{*}\) is often treated as an adjustable parameter with values within a narrow range around 0.1 for most materials, making this formulism highly robust (26- 30), and compares well with the latest ab initio Eliashberg theory (31). In this work, the commonly used value of \(\mu^{*} = 0.13\) is adopted for all the reported calculations. We also calculated electronic band structures to assess the evolution of the \(s\) - and \(d\) - bands under pressure. + +Fig. 1a shows the experimentally measured temperature dependence of electrical resistance under high pressure up to 180 GPa for Ti metal in Sample 1. The results show that Ti metal becomes superconductive with Tc above 2 K at 18 GPa and rises slightly to \(\sim 3.5 \mathrm{K}\) at 54 GPa, which agrees with previously reported results(11). The value of Tc increases at an enhanced pace from 54 to 99 GPa, then undergoes a steep rise from 10.2 K at 99 GPa to 20.3 K at 108 GPa, and Tc is further enhanced to 22 K at 180 GPa. For Sample 2, the pressure range was increased, with the highest pressure reaching 310 GPa. The resistance curves under pressure are shown in Fig. 1b. With further increasing pressure the Tc reaches a maximum of 26 K at 248 GPa, as shown in Fig. 1c. The onset critical temperature (Tc onset), the midpoint critical temperature (Tc mid) and the zero- resistance critical temperature (Tc zero) of superconducting transition are determined by the derivative of resistance with respect to temperature \(dR / dT\) as shown in the inset of Fig. 1c. It is noted that Tc stays almost constant over a wide pressure range up to at least 240 GPa, and such robust superconducting behaviors are consistently seen in different specimens (Fig. S2). At 108 GPa where Tc jumps up, the resistance exhibits a two- step superconducting transition behavior with the lower Tc of \(\sim 11 \mathrm{K}\) that is comparable to the Tc value at 99 GPa, indicating a phase + +<--- Page Split ---> + +transition near 108 GPa, which is close to the pressure for the reported \(\omega - \gamma\) phase transition (6,10). In the Ti- \(\omega\) phase, Tc rises smoothly with pressure below 56 GPa with a slope of 0.07 K/GPa; assuming this slope keeps unchanged, Tc should reach 8.7 K before the \(\omega - \gamma\) phase transition (at 128 GPa) (11). However, our results show that the slope \(\mathrm{dTc / dP}\) increases significantly between 54- 99 GPa, which leads to the much higher measured Tc at 99 GPa. + +To further probe the pressure driven SC phase of Ti metal, we have examined the effect of magnetic field on the SC transition behavior. Fig. 2a presents the electrical resistance measured at 248 GPa and under different magnetic fields. It is seen that the transition is gradually suppressed by the magnetic field. We have plotted onset Tc versus magnetic field in Fig. 2b, from which the upper critical field at zero temperature \(\mu_0H_{c2}(0)\) can be estimated. The \(\mu_0H_{c2}(T)\) date were fitted to the Ginzburg Landau (GL) function \(\mu_0H_{c2}(T) = \mu_0H_{c2}(0)(1 - (T / T_c)^2)\) , which gives a value of \(\mu_0H_{c2}(0) = 32\) T. At other pressures where Tc is above 20 K, the estimated upper critical field has been obtained to be near 30 T, (Fig. S3(a,b)), which is larger than that of the most commonly used low- temperature NbTi superconductor (\(\mu_0H_{c2}(T) = 15\) T). Using the \(\mu_0H_{c2}(T)\) value of 32 T, the GL coherence length was calculated to be \(\xi = 32\) Å via \(\mu_0H_{c2}(0) = \Phi_0 / 2\pi \xi^2\) , where \(\Phi_0 = 2.067 \times 10^{- 15}\) Web is the magnetic flux quantum. + +Fig. 3(a,b) presents the measured Hall resistance at room temperature and different pressures. The Hall resistance is negative and decreases linearly with magnetic field, indicating that the dominant carriers are electrons. The carrier density \(n\) as a function pressure estimated from the Hall resistance is presented in Fig. 3(c), with a value of about \(2.5 \times 10^{22} / \mathrm{cm}^3\) that is little changed in the low- pressure range. Above 108 GPa, \(n\) increases dramatically and is enhanced by more than one order of magnitude to \(3.1 \times 10^{23} / \mathrm{cm}^3\) at 137 GPa. Further increasing pressure leads to reduced carrier density of \(4.5 \times 10^{22} / \mathrm{cm}^3\) at 144 GPa. The changes of the carrier density indicate phase transitions at pressures of about 108 GPa and 144 GPa, respectively. To see more clearly these phase transitions, the pressure dependence of resistance \(R(P)\) at fixed temperature is plotted in Fig. + +<--- Page Split ---> + +3(d). The \(R(P)\) curve shows two peaks near the critical pressures, which corresponds to the \(\omega - \gamma\) phase and \(\gamma - \delta\) phase transitions reported by Y. Akahama (6) and Y. K. Vohra (10), respectively. + +We have carried out the high- pressure X- ray diffraction experiments, and the results are shown in Fig. S4. Combining the phase transition reported by previous works and our transport experiments, we plot the superconducting and structural phase diagram shown in Fig. 4. Up to 310 GPa, five different crystal structures, in the sequence of Ti- \(\alpha\) , Ti- \(\omega\) , Ti- \(\gamma\) , Ti- \(\delta\) and Ti- \(\beta\) are identified. The Ti- \(\alpha\) phase \((P = 0 - 9\) GPa) hosts SC with Tc below 2 K; the Ti- \(\omega\) phase \((P = 9 - 108\) GPa) sees a monotonously increasing Tc with pressure to 12 K at \(\sim 108\) GPa; while in the high- pressure phases of Ti- \(\gamma\) \((P = 108 - 144\) GPa) and Ti- \(\delta\) \((P = 144 - 240\) GPa), Tc stays above 20 K with the maximum Tc \(= 26\) K occurring at the boundary of Ti- \(\delta\) and Ti- \(\beta\) phases. It is remarkable that the Tc of compressed Ti metal stays robust above 20 K in a very wide range of pressures. This behavior is at odds with the expectation of conventional phonon- mediated superconducting theory, which predicts sensitive pressure dependence and descending Tc at very high pressures due to phonon stiffening. Interestingly, it is noted that the phenomenon of robust Tc values extended over a wide pressure range has been found in high- Tc cuprates wherein superconductivity is driven by correlation effects (32a), and this analogy indicates that mechanisms of novel many- body origin may be driving the unexpectedly robust superconductivity in dense Ti metal at ultrahigh pressures. + +Among all elemental solids, only a few have been reported to exhibit SC with Tc near 20 K at high pressures, including alkali metal Li (32, 33), alkali- earth metal Ca(34) and rare- earth metals of Y (16). The high Tc values of Li, which is the lightest elemental metal containing only simple 2s- electrons, is attributed to the enhancement of the electron- phonon coupling due to the phonon modes softening under high pressure (35, 36); while the Tc enhancements in Ca and Y are mainly ascribed to pressure induced electron transfer from the \(s\) to \(d\) shell. + +The resistance of Ti as a function of pressure measured at 18 GPa (Fig. S5) exhibits interesting scaling behavior. Upon fitting with the formula \(R = R_0 + A^* T^n\) , where \(R_0\) is the residual resistance, \(A\) + +<--- Page Split ---> + +is the coefficient of the power law, and \(n\) is the exponent, the resulting scaling exponent \(n = 3.1\) is different from the value expected for systems dominated by either electron-phonon scattering \((n = 5)\) or pure electron-electron scattering \((n = 2)\) . In fact, the \(n\) value near 3 implies that the \(s - d\) interband scattering dominates the electron transport with contributions from electron correlation effects, as observed in 1T- TiSe \(_2\) (37) or Ta \(_4\) Pd \(_3\) Te \(_{16}\) (38). Such \(s - d\) interband scattering and the impact of the \(d\) electron correlation effects on the formation of Cooper pairs require further exploration. + +We have examined relative energetic stability of the high- pressure phases of Ti metal from first- principles calculations, and the results are consistent with the experimentally measured phase stability and transformation sequence. Adopting the determined crystal structures of the Ti metal phases, we have calculated their electronic, phonon and electron- phonon coupling properties, which are used as input to determine the SC critical temperature Tc. The obtained Tc data for the Ti- \(\omega\) , Ti- \(\gamma\) , and Ti- \(\delta\) phases (Fig. S6) reveal the following trends: (i) Tc increases with rising pressure monotonically in the Ti- \(\omega\) phase over its entire stability range; (ii) Tc is significantly enhanced upon the phase transition into the Ti- \(\gamma\) phase over the relatively narrow pressure range where this transition occurs; (iii) Tc undergoes an even larger jump when the structure enters the Ti- \(\delta\) phase. These findings are in overall agreement with the experimental results up to about 160 GPa, but large discrepancies exist at higher pressures where the record- setting Tc is observed. + +The experimentally observed superconducting properties of the densely compressed Ti metal indicate clear inadequacy of the conventional phonon- mediated SC mechanism for describing the unexpectedly high values of Tc and anomalously robust superconducting state over a very wide pressure range. The electron- electron correlation effects associated with the \(d\) - bands in Ti are expected to have a strong influence on transport and superconducting properties. Calculated electronic band structure of Ti- \(\omega\) phase at 20 GPa (Fig. 5a) shows significant overlap of the \(s\) - and \(d\) - electron states near the Fermi level, which corroborates our measured normal- state resistivity results (Fig. S5) indicating strong \(s - d\) scattering contribution to the resistivity (37,38). At higher + +<--- Page Split ---> + +pressure of 100 GPa, the band structure of Ti- \(\gamma\) phase (Fig. 5b) shows that the \(s\) - states move up in energy while still have large overlap with the \(d\) - bands, which is conducive to significant \(s\) - \(d\) scattering. At further increased pressure of 180 GPa, the electronic states in Ti- \(\delta\) phase (Fig. 5c) near the Fermi level are dominated by the \(d\) - electron states. There are flat \(d\) - bands below the Fermi level that can host significant correlation effects. These \(d\) - bands will likely rise much closer to the Fermi level due to the electron- electron interactions when properly treated by many- body theory. Such correlation effects could greatly enhance the \(d\) - band derived electronic DOS at the Fermi level favorable for increasing Tc and also drive additional non- phonon- mediated mechanisms for further strengthening the superconducting state with higher Tc and maintaining its robust presence over a wide pressure range. This study raises intriguing possibility of major impact by many- body effects on superconductivity in dense Ti, which needs in- depth study by sophisticated many- body theoretical treatments to explore pertinent novel processes and underlying mechanisms. + +The present study unveils unexpectedly robust superconductivity in compressed Ti metal with record- setting Tc among elemental superconductors. The Tc and Hc values of Ti are notably higher than those of the widely used superconducting NbTi alloy. Our discovery raises the possibility of finding more materials via pressure driven correlation effects stemming from the contributions of \(d\) electrons, leading to superconductivity with much higher Tc than previously believed achievable, and such materials may be stabilized at lower pressures via mechanical strain or chemical pressure. This intriguing scenario calls for further research into the impact on superconductivity by the \(s\) - \(d\) interaction and \(d\) - electron dominated correlation effects in highly compressed \(d\) - band metals for evaluation of diverse materials, from elemental solids to alloys, in search of hitherto unknown and unexplored superconducting materials that could improve fundamental understanding of broader varieties of superconductors. Equally important, the present findings open new avenues for expanding the scope of superconductors with notably enhanced and robust Tc and Hc that are more adaptive and suitable for applications in diverse and demanding implementation settings. + +<--- Page Split ---> + +Acknowledgments: This work is supported by NSF & MOST of China through research projects. + +## Author Contributions + +Research Design & Supervision: C.Q.J.; high pressure synthesis and in situ resistance measurements: C.L.Z., X.H., Z.W. L., K.L., S.J.Z., S.M.F., X.C.W., Y.W.L. R.C.Y, Y.P., C.Q.J; in situ synchrotron experiments: L. H. W., V.B.P., S.C., H.Z.L., First principle calculations: C.L., Q.L., C.F.C., manuscript writing: C.L.Z., X.C.W. C.Q.J & C.F.C. All authors contributed to the discussions. + +Competing interests: The authors declare no competing interests as defined by Nature Portfolio, or other interests that might be perceived to influence the results and/or discussion reported in this paper. + +## References + +1. B. H. Kear, E. R. Thompson, Science 208, 847-856 (1980). +2. R. R. Boyer, Mat Sci Eng a-Struct 213, 103-114 (1996). +3. M. E. Coon, S. B. Stephan, V. Gupta, C. P. Kealey, M. T. Stephan, Nat Biomed Eng 4, 195-206 (2020). +4. A. Dewaele, V. Stutzmann, J. Bouchet, F. Bottin, F. Occelli, et al., Phys. Rev. B 91, 134108 (2015). +5. H. Xia, G. Parthasarathy, H. Luo, Y. K. Vohra, A. L. Ruoff, Phys. Rev. B 42, 6736-6738 (1990). +6. Y. Akahama, H. Kawamura, T. Le Bihan, Phys. Rev. Lett. 87, 275503 (2001). +7. Y. Akahama, H. Kawamura, T. Le Bihan, J. Phys.: Condens. Matter 14, 10583-10588 (2002). +8. A. L. Kutepov, S. G. Kutepova, Phys. Rev. B 67, 132102 (2003). +9. Y. Akahama, S. Kawaguchi, N. Hirao, Y. Ohishi, J. Appl. Phys. 128, 035901 (2020). +10. Y. K. Vohra, P. T. Spencer, Phys. Rev. Lett. 86, 3068-3071 (2001). +11. I. O. Bashkin, V. G. Tissen, M. V. Nefedova, E. G. Ponyatovsky, Physica C-Superconductivity and Its Applications 453, 12-14 (2007). +12. K. Iyakutti, C. N. Louis, S. Anuratha, S. Mahalakshmi, Int. J. Mod. Phys. B 23, 723-741 (2009). +13. R. Ahuja, J. M. Wills, B. Johansson, O. Eriksson, Phys. Rev. B 48, 16269-16279 (1993). +14. H. L. Skriver, Phys. Rev. B 31, 1909-1923 (1985). +15. T. Yabuuchi, T. Matsuoka, Y. Nakamoto, K. Shimizu, J. Phys. Soc. Jpn. 75, 083703 (2006). +16. J. J. Hamlin, V. G. Tissen, J. S. Schilling, Physica C-Superconductivity and Its Applications 451, 82-85 (2007). +17. M. Ishizuka, M. Iketani, S. Endo, Phys. Rev. B 61, R3823-R3825 (2000). +18. Y. T. Jia, X. He, S. M. Feng, S. J. Zhang, C. L. Zhang, et al., Crystals 10, 1116 (2020). +19. J. L. Zhang, S. J. Zhang, H. M. Weng, W. Zhang, L. X. Yang, et al., Proc Natl Acad Sci U S A/ PNAS 108, 24-28 (2011). +20. Z. W. Li, X. He, C. L. Zhang, S. J. Zhang, S. M. Feng, et al., arXiv:2103.16917 (2021). + +<--- Page Split ---> + +21. Y. T. Jia, C. S. Gong, Y. X. Liu, J. F. Zhao, C. Dong, et al., Chin. Phys. Lett. 37, 097404 (2020). +22. P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, et al., J. Phys.: Condens. Matter 21, 395502 (2009). +23. G. M. Eliashberg, Soviet Physics Jetp-Ussr 11, 696-702 (1960). +24. D. J. Scalapino, Schriefer Jr, J. W. Wilkins, Phys. Rev. 148, 263-279 (1966). +25. W. L. Mcmillan, Phys. Rev. 167, 331-344 (1968). +26. P. B. Allen, R. C. Dynes, Phys. Rev. B 12, 905-922 (1975). +27. P. Morel, P. W. Anderson, Phys. Rev. 125, 1263-1271 (1962). +28. W. L. Mcmillan, J. M. Rowell, (Marcel Dekker New York) (1969). +29. J. P. Carbotte, Rev. Mod. Phys. 62, 1027-1157 (1990). +30. J. P. Carbotte, F. Marsiglio, (Springer-Verlag Berlin Heidelberg Germany) (2003). +31. A. Sanna, J. A. Flores-Livas, A. Davydov, G. Profeta, K. Dewhurst, et al., J. Phys. Soc. Jpn. 87, 041012 (2018). +32a. C. W. Chu, L. Gao, F. Chen, Z. J. Huang, R. L. Meng, and Y. Y. Xue, Nature 365, 323-325(1993) +32. K. Shimizu, H. Ishikawa, D. Takao, T. Yagi, K. Amaya, Nature 419, 597-599 (2002). +33. V. V. Struzhkin, M. I. Eremets, W. Gan, H. K. Mao, R. J. Hemley, Science 298, 1213-1215 (2002). +34. M. Sakata, Y. Nakamoto, K. Shimizu, T. Matsuoka, Y. Ohishi, Phys. Rev. B 83, 220512 (2011). +35. J. S. Tse, Y. Ma, H. M. Tutuncu, Journal of Physics: Condensed Matter 17, S911-S920 (2005). +36. D. Kasinathan, J. Kuneš, A. Lazicki, H. Rosner, C. S. Yoo, et al., Phys. Rev. Lett. 96, 047004 (2006). +37. A. F. Kusmartseva, B. Sipos, H. Berger, L. Forro, E. Tutis, Phys. Rev. Lett. 103, 236401 (2009). +38. J. Pan, W. H. Jiao, X. C. Hong, Z. Zhang, L. P. He, et al., Phys. Rev. B 92, 180505 (2015). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. (a) Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 1) measured under high pressures. The inset is an enlarged view of the resistance curve, showing the superconducting transition in detail. (b) The resistance curves for Ti Sample 2. (c) The resistance curve measured at 248 GPa, where the derivative of the resistance with respect to temperature dR/dT is plotted to clearly show the onset Tc.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. (a) Temperature dependence of the electrical resistance of Ti metal measured under different magnetic fields at the fixed pressure of 248 GPa. (b) Upper critical field versus superconducting transition temperature of \(\mathrm{Tc}^{2\mathrm{ero}}\) . The line is a fit obtained using the Ginzburg–Landau function.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3. (a,b) Hall resistance as a function of magnetic field measured under different pressure. (c-e) Carrier density, resistance at fixed temperature, and onset Tc versus pressure, respectively. The yellow and blue areas mark the anomaly of these data, indicating the critical pressures where the phase transitions occur.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4. The superconducting critical transition temperature (Tc) of Ti metal under compression up to 320 GPa plotted against the background of stability fields of the indicated structural phases.
+ +![](images/Figure_5.jpg) + +
Fig. 5. Calculated electronic band structures of three representative Ti phases: (a) Ti-ω at 20 GPa, (b) Ti-γ at 100 GPa and (c) Ti-δ at 180 GPa. The contributions from the s-electron and d-electron states are shown by red and gray circles, respectively, and the circle areas are proportional to the weights of the corresponding band states. Energy is measured relative to the Fermi energy \(\mathrm{E}_{\mathrm{F}}\) .
+ +<--- Page Split ---> + +# Supplementary Materials + +![](images/Figure_unknown_0.jpg) + +
Fig.S1 The photo showing sample assembly on the top of diamond cullet..
+ +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
Fig.S2 Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 5) measured under high pressures.
+ +![](images/Figure_unknown_2.jpg) + +
Fig. S3 (a) The superconducting transitions measured under different magnetic fields at 154 GPa. (b) at 180 GPa.
+ +<--- Page Split ---> +![](images/Figure_unknown_3.jpg) + +
Fig. S4 The in-situ high pressure X-ray diffraction patterns with the highest experimental pressure of 225 GPa, using the wavelength of 0.3344 Å. The pressure induced phase transition sequence from Ti-α phase, to Ti-ω phase, to Ti-γ phase, and then to Ti-δ phase has been observed, which is consistent with the previously reported results (Y. Akahama et al. 2002).
+ +![](images/Figure_unknown_4.jpg) + +
Fig. S5. The temperature dependence of resistance measured at 18 GPa. The resistance in the low temperature range of 6-70 K is fitted (red line) with the formula of \(R = R_0 + AT^n\) , where \(R_0\) is the residual resistance, \(A\) is the coefficient of the power law, and \(n = 3.1\) is the extracted exponent.
+ +<--- Page Split ---> +![](images/Figure_unknown_5.jpg) + +
Fig. S6. The superconducting critical transition temperature (Tc) of the indicated Ti phases in their stability fields in the [20, 220] GPa hydrostatic pressure range. The calculated crystal structure of Ti is sensitive to the pressure conditions, and under hydrostatic pressures at or above 160 GPa, the Ti-δ phase essentially relaxes to the Ti-β (bcc) phase without considering any stress anisotropy.
+ +<--- Page Split ---> diff --git a/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b_det.mmd b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..992206eb74b640db960287e9e7c203a285ad4f55 --- /dev/null +++ b/preprint/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b/preprint__0db3dd52880b0f8f89e506d1a22a10e2665c7cd4a3bfd16ccaf255c2e532d33b_det.mmd @@ -0,0 +1,306 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 933, 175]]<|/det|> +# Robust record high Tc elemental superconductivity in titanium + +<|ref|>text<|/ref|><|det|>[[44, 196, 320, 238]]<|/det|> +Changling Zhang Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 244, 320, 284]]<|/det|> +Xin He Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 290, 543, 330]]<|/det|> +Chang Liu Jilin University https://orcid.org/0000- 0003- 0824- 1098 + +<|ref|>text<|/ref|><|det|>[[44, 336, 320, 376]]<|/det|> +Zhiwen Li Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 383, 320, 423]]<|/det|> +Ke Lu Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 429, 320, 469]]<|/det|> +Sijia Zhang Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 475, 922, 540]]<|/det|> +Shaomin Feng Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China + +<|ref|>text<|/ref|><|det|>[[44, 544, 581, 585]]<|/det|> +Xiancheng WANG Institute of Physics https://orcid.org/0000- 0001- 6263- 4963 + +<|ref|>text<|/ref|><|det|>[[44, 590, 955, 654]]<|/det|> +Yi Peng Beijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 659, 581, 700]]<|/det|> +Youwen Long Institute of Physics https://orcid.org/0000- 0002- 8587- 7818 + +<|ref|>text<|/ref|><|det|>[[44, 705, 494, 746]]<|/det|> +Richeng Yu Institute of Physics, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 752, 320, 792]]<|/det|> +Luhong Wang Harbin Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 798, 241, 838]]<|/det|> +Vitali Prakapenka University of Chicago + +<|ref|>text<|/ref|><|det|>[[44, 844, 280, 885]]<|/det|> +Stella Chariton The University of Chicago + +<|ref|>text<|/ref|><|det|>[[44, 891, 541, 931]]<|/det|> +Quan Li Jilin University https://orcid.org/0000- 0002- 7724- 1289 + +<|ref|>text<|/ref|><|det|>[[44, 937, 144, 954]]<|/det|> +Haozhe Liu + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 45, 925, 88]]<|/det|> +Center for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 5720- 2031 + +<|ref|>text<|/ref|><|det|>[[44, 93, 676, 181]]<|/det|> +Changfeng Chen Department of Physics and Astronomy, University of Nevada Changqing Jin ( \(\square\) Jin@iphy.ac.cn ) Chinese Academy of Sciences https://orcid.org/0000- 0001- 8097- 9156 + +<|ref|>sub_title<|/ref|><|det|>[[44, 222, 102, 240]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 259, 136, 278]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 297, 297, 317]]<|/det|> +Posted Date: May 18th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 335, 475, 355]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1630556/v1 + +<|ref|>text<|/ref|><|det|>[[42, 372, 910, 415]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 450, 914, 494]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 15th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33077- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[238, 75, 758, 92]]<|/det|> +# Robust record high Tc elemental superconductivity in titanium + +<|ref|>text<|/ref|><|det|>[[156, 110, 841, 166]]<|/det|> +C. +L. Zhanga1,2, +X. Hea1,2,3, +C. Liua4, +Z. +W. Li1,2, +K. Lu1,2, +S. +J. Zhang1, +S. +M. Feng1, +X. +C. Wanga1,2, +Y. Peng1,2, +Y. +W. Long1,2,3, +R.C.Yu1,2, +L. +H. Wang5, +V. +B. Prakapenka6, +S. Chariton6, +Q. Li4, +H. +Z. Liu7, +C. +F. Chen8, +C. +Q. Jina1,2,3 + +<|ref|>text<|/ref|><|det|>[[153, 186, 844, 371]]<|/det|> +1Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China 2School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100190, China 3 Songshan Lake Materials Laboratory, Dongguan 523808, China 4International Center for Computational Method and Software, College of Physics, and International Center of Future Science, Jilin University, Changchun 130012, China 5Harbin Institute of Technology, Harbin 130001, China 6Center for Advanced Radiations Sources, University of Chicago, Chicago, Illinois 60637, USA 7Center for High Pressure Science & Technology Advanced Research, Beijing 100094, China 8Department of Physics and Astronomy, University of Nevada, Las Vegas, Nevada 89154, USA + +<|ref|>text<|/ref|><|det|>[[147, 393, 852, 775]]<|/det|> +Searching for superconductivity in single element with high Tc is of great interest both for fundamental science & potential applications due to the easy recipe. It is especially interesting to explore high Tc superconductivity in transition metal elements wherein the conventional BCS theories suggest that \(d\) electrons are not favorable to form high Tc superconducting state. Here we report discovery of surprisingly robust superconductivity up to 310 GPa with almost constant Tc above 20K in a wide pressure range from 108GPa to 240 GPa in titanium. The maximum Tconset above 26 K and zero- resistance Tczero of 21 K are record- high values hitherto achieved among elemental superconductors. This finding is in sharp contrast to the theoretical predications based on the pristine electron- phonon coupling scenario. Measured normal state conductivity reveals significant \(s\) - \(d\) scattering. Quantum calculations show strong \(s\) - \(d\) transfer and \(d\) - band dominance, indicating correlation driven contributions to the robust high- Tc superconductivity seen in dense titanium. Different from most high- Tc superconductors that are either complex ceramic or intermetallic compounds, titanium metal is much easier to process and hosts notably higher and more robust Tc and critical magnetic field Hc compared to the widely used superconducting NbTi alloy. This study opens a fresh promising avenue for rational design and discovery of high- Tc superconductors among simple materials via pressure tuned unconventional mechanism. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 852, 439]]<|/det|> +Titanium (Ti) metal has long attracted tremendous scientific interests because of its combined properties of light weight, high strength and corrosion resistance. As an advanced metallic structural material, Ti and its alloys find wide use in the fields of aerospace, biomedicine and at extreme conditions (1- 3). High pressure can modify crystal structures which, in turn, may lead to new functionalities. At ambient pressure and room temperature, Ti crystalizes in a hexagonal close- packed (hcp) structure (Ti- \(\alpha\) phase) (4). Under applied pressure, Ti undergoes structural transitions in the sequence of Ti- \(\alpha\) , Ti- \(\alpha\) , Ti- \(\gamma\) , Ti- \(\delta\) and Ti- \(\beta\) phases, where Ti- \(\alpha\) phase is a hexagonal structure, Ti- \(\gamma\) and Ti- \(\delta\) phases are orthorhombic and Ti- \(\beta\) phase is body- centered cubic (5- 9). The \(\alpha\) - to- \(\alpha\) transition occurs around 8 GPa (5, 6), and the Ti- \(\alpha\) phase is stable up to about 100 GPa, then transforms into Ti- \(\gamma\) phase (6, 10), which further transforms into Ti- \(\delta\) phase at \(\sim 140\) GPa (6), before cubic Ti- \(\beta\) phase stabilizes at 243 GPa (9). + +<|ref|>text<|/ref|><|det|>[[147, 454, 852, 838]]<|/det|> +Superconductivity (SC) in high- pressure phases of Ti metal was previously reported to have a measured maximal critical temperature (Tc) of 3.5 K at 56 GPa (11); early theoretical calculations based on the electron- phonon coupling mechanism predicate that the Tc for Ti metal is capped at about 5 K for all the known high pressure phases (12). Generally, compression of crystal lattice has markedly different effects on the 4s and 3d bands, prompting notable \(s\) - \(d\) electron transfer. The narrower \(d\) - bands possess stronger correlation characters, while the \(s\) - \(d\) transfer tends to enhance electronic density of state (DOS) near the Fermi level favoring SC (13- 17). Here, we report a surprising experimental observation of dramatic pressure enhanced SC in Ti over a wide pressure range with the maximal Tc above 26 K, setting a new record among elemental superconductors. Measured normal- state conductivity results and analysis of electronic and superconducting properties indicate prominent influence of the s- d transfer and d- band driven correlation effects on the significantly enhanced and unusually robust SC in highly compressed Ti metal. + +<|ref|>text<|/ref|><|det|>[[148, 853, 850, 904]]<|/det|> +The electrical resistance and Hall effect measurements were performed using the four- probe method in Van der Pauw way as shown in Fig. S1. The applied current is \(100 \mu \mathrm{A}\) . Diamond anvil + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 853, 505]]<|/det|> +cells were used to produce high pressures. A variant of anvils with beveled culet size of \(20 / 140 / 300\mu \mathrm{m}\) , \(30 / 140 / 300\mu \mathrm{m}\) or \(50 / 140\) \(300\mu \mathrm{m}\) are adopted in the experiments. A plate of T301 stainless steel that is covered with mixture of cBN powder & epoxy as insulate layer was used as the gasket. A hole of approximately 15 to \(30\mu \mathrm{m}\) in diameter depending on top culet size was drilled in the center of the gasket to serve as ample chamber. The hBN powder was generally used as pressure- transmitting medium that filled in the high- pressure chamber. We used the ATHENA procedure to produce the specimen assembly (18). Four trapezoidal Pt foils with thickness approximately \(0.5\mu \mathrm{m}\) as the inner electrode were deposited on the culet surface. Cross- shaped Ti specimens with side lengths \(\sim 10\mu \mathrm{m}*10\mu \mathrm{m}\) and thickness of \(1\mu \mathrm{m}\) were stacked on the electrodes. Pressure was calibrated by the shift of the first- order Raman edge frequency from the diamond cutlet. Diamond anvil cells were put in a MagLab system that provides synergetic extreme environments with temperatures from \(300\mathrm{K}\) to \(1.5\mathrm{K}\) and magnetic fields up to \(9\mathrm{T}\) for the transport measurements (18- 21). + +<|ref|>text<|/ref|><|det|>[[147, 520, 852, 736]]<|/det|> +In- situ high- pressure angle- dispersive X- ray diffraction data were collected at room temperature at GSECARS of Advanced Photon Source at the Argonne National Laboratory. The x- ray with the wavelength \(\lambda = 0.3344\mathrm{\AA}\) was focused down to a spot of \(\sim 3\mu \mathrm{m}\) in diameter on the sample. A symmetric diamond anvil cell with beveled anvil (50/300 \(\mu \mathrm{m}\) ) was used. Rhenium steel gasket was pre- pressed to a thickness of \(20\mu \mathrm{m}\) , and a hole of diameter of \(15\mu \mathrm{m}\) was drilled at the center to serve as sample chamber, which was then filled with Ti power mixed with Pt. Pressure was calibrated using the equitation of state of both Re and Pt. + +<|ref|>text<|/ref|><|det|>[[147, 754, 852, 903]]<|/det|> +To assess structural, electronic, and phonon- mediated superconducting properties of Ti metal under pressure, we employed the latest computational techniques to determine the total energy, lattice dynamics and electron- phonon coupling using the QUANTUM ESPRESSO code (22), with improved description over previously reported results (12). Superconducting critical temperature Tc has been evaluated based on the Eliashberg theory of superconductivity (23, 24), using the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 792, 105]]<|/det|> +following formula that McMillan derived (25) and later modified by Allen and Dynes (26), + +<|ref|>equation<|/ref|><|det|>[[366, 111, 630, 143]]<|/det|> +\[T_{c} = \frac{\omega_{log}}{1.20} exp[-\frac{1.04(1 + \lambda)}{\lambda - \mu^{*}(1 + 0.62\lambda)} ]\] + +<|ref|>text<|/ref|><|det|>[[147, 158, 852, 408]]<|/det|> +where \(\omega_{log}\) is a logarithmically averaged characteristic phonon frequency, and \(\mu^{*}\) is the Coulomb pseudopotential which describes the effective electron- electron repulsion (27). This equation is generally accurate for materials with EPC parameter \(\lambda\) at 1.5 or less (28- 31), which is satisfied in the present study. The Coulomb pseudopotential \(\mu^{*}\) is often treated as an adjustable parameter with values within a narrow range around 0.1 for most materials, making this formulism highly robust (26- 30), and compares well with the latest ab initio Eliashberg theory (31). In this work, the commonly used value of \(\mu^{*} = 0.13\) is adopted for all the reported calculations. We also calculated electronic band structures to assess the evolution of the \(s\) - and \(d\) - bands under pressure. + +<|ref|>text<|/ref|><|det|>[[147, 422, 852, 910]]<|/det|> +Fig. 1a shows the experimentally measured temperature dependence of electrical resistance under high pressure up to 180 GPa for Ti metal in Sample 1. The results show that Ti metal becomes superconductive with Tc above 2 K at 18 GPa and rises slightly to \(\sim 3.5 \mathrm{K}\) at 54 GPa, which agrees with previously reported results(11). The value of Tc increases at an enhanced pace from 54 to 99 GPa, then undergoes a steep rise from 10.2 K at 99 GPa to 20.3 K at 108 GPa, and Tc is further enhanced to 22 K at 180 GPa. For Sample 2, the pressure range was increased, with the highest pressure reaching 310 GPa. The resistance curves under pressure are shown in Fig. 1b. With further increasing pressure the Tc reaches a maximum of 26 K at 248 GPa, as shown in Fig. 1c. The onset critical temperature (Tc onset), the midpoint critical temperature (Tc mid) and the zero- resistance critical temperature (Tc zero) of superconducting transition are determined by the derivative of resistance with respect to temperature \(dR / dT\) as shown in the inset of Fig. 1c. It is noted that Tc stays almost constant over a wide pressure range up to at least 240 GPa, and such robust superconducting behaviors are consistently seen in different specimens (Fig. S2). At 108 GPa where Tc jumps up, the resistance exhibits a two- step superconducting transition behavior with the lower Tc of \(\sim 11 \mathrm{K}\) that is comparable to the Tc value at 99 GPa, indicating a phase + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 852, 238]]<|/det|> +transition near 108 GPa, which is close to the pressure for the reported \(\omega - \gamma\) phase transition (6,10). In the Ti- \(\omega\) phase, Tc rises smoothly with pressure below 56 GPa with a slope of 0.07 K/GPa; assuming this slope keeps unchanged, Tc should reach 8.7 K before the \(\omega - \gamma\) phase transition (at 128 GPa) (11). However, our results show that the slope \(\mathrm{dTc / dP}\) increases significantly between 54- 99 GPa, which leads to the much higher measured Tc at 99 GPa. + +<|ref|>text<|/ref|><|det|>[[146, 255, 853, 602]]<|/det|> +To further probe the pressure driven SC phase of Ti metal, we have examined the effect of magnetic field on the SC transition behavior. Fig. 2a presents the electrical resistance measured at 248 GPa and under different magnetic fields. It is seen that the transition is gradually suppressed by the magnetic field. We have plotted onset Tc versus magnetic field in Fig. 2b, from which the upper critical field at zero temperature \(\mu_0H_{c2}(0)\) can be estimated. The \(\mu_0H_{c2}(T)\) date were fitted to the Ginzburg Landau (GL) function \(\mu_0H_{c2}(T) = \mu_0H_{c2}(0)(1 - (T / T_c)^2)\) , which gives a value of \(\mu_0H_{c2}(0) = 32\) T. At other pressures where Tc is above 20 K, the estimated upper critical field has been obtained to be near 30 T, (Fig. S3(a,b)), which is larger than that of the most commonly used low- temperature NbTi superconductor (\(\mu_0H_{c2}(T) = 15\) T). Using the \(\mu_0H_{c2}(T)\) value of 32 T, the GL coherence length was calculated to be \(\xi = 32\) Å via \(\mu_0H_{c2}(0) = \Phi_0 / 2\pi \xi^2\) , where \(\Phi_0 = 2.067 \times 10^{- 15}\) Web is the magnetic flux quantum. + +<|ref|>text<|/ref|><|det|>[[147, 618, 852, 904]]<|/det|> +Fig. 3(a,b) presents the measured Hall resistance at room temperature and different pressures. The Hall resistance is negative and decreases linearly with magnetic field, indicating that the dominant carriers are electrons. The carrier density \(n\) as a function pressure estimated from the Hall resistance is presented in Fig. 3(c), with a value of about \(2.5 \times 10^{22} / \mathrm{cm}^3\) that is little changed in the low- pressure range. Above 108 GPa, \(n\) increases dramatically and is enhanced by more than one order of magnitude to \(3.1 \times 10^{23} / \mathrm{cm}^3\) at 137 GPa. Further increasing pressure leads to reduced carrier density of \(4.5 \times 10^{22} / \mathrm{cm}^3\) at 144 GPa. The changes of the carrier density indicate phase transitions at pressures of about 108 GPa and 144 GPa, respectively. To see more clearly these phase transitions, the pressure dependence of resistance \(R(P)\) at fixed temperature is plotted in Fig. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 850, 140]]<|/det|> +3(d). The \(R(P)\) curve shows two peaks near the critical pressures, which corresponds to the \(\omega - \gamma\) phase and \(\gamma - \delta\) phase transitions reported by Y. Akahama (6) and Y. K. Vohra (10), respectively. + +<|ref|>text<|/ref|><|det|>[[147, 155, 852, 640]]<|/det|> +We have carried out the high- pressure X- ray diffraction experiments, and the results are shown in Fig. S4. Combining the phase transition reported by previous works and our transport experiments, we plot the superconducting and structural phase diagram shown in Fig. 4. Up to 310 GPa, five different crystal structures, in the sequence of Ti- \(\alpha\) , Ti- \(\omega\) , Ti- \(\gamma\) , Ti- \(\delta\) and Ti- \(\beta\) are identified. The Ti- \(\alpha\) phase \((P = 0 - 9\) GPa) hosts SC with Tc below 2 K; the Ti- \(\omega\) phase \((P = 9 - 108\) GPa) sees a monotonously increasing Tc with pressure to 12 K at \(\sim 108\) GPa; while in the high- pressure phases of Ti- \(\gamma\) \((P = 108 - 144\) GPa) and Ti- \(\delta\) \((P = 144 - 240\) GPa), Tc stays above 20 K with the maximum Tc \(= 26\) K occurring at the boundary of Ti- \(\delta\) and Ti- \(\beta\) phases. It is remarkable that the Tc of compressed Ti metal stays robust above 20 K in a very wide range of pressures. This behavior is at odds with the expectation of conventional phonon- mediated superconducting theory, which predicts sensitive pressure dependence and descending Tc at very high pressures due to phonon stiffening. Interestingly, it is noted that the phenomenon of robust Tc values extended over a wide pressure range has been found in high- Tc cuprates wherein superconductivity is driven by correlation effects (32a), and this analogy indicates that mechanisms of novel many- body origin may be driving the unexpectedly robust superconductivity in dense Ti metal at ultrahigh pressures. + +<|ref|>text<|/ref|><|det|>[[147, 653, 852, 837]]<|/det|> +Among all elemental solids, only a few have been reported to exhibit SC with Tc near 20 K at high pressures, including alkali metal Li (32, 33), alkali- earth metal Ca(34) and rare- earth metals of Y (16). The high Tc values of Li, which is the lightest elemental metal containing only simple 2s- electrons, is attributed to the enhancement of the electron- phonon coupling due to the phonon modes softening under high pressure (35, 36); while the Tc enhancements in Ca and Y are mainly ascribed to pressure induced electron transfer from the \(s\) to \(d\) shell. + +<|ref|>text<|/ref|><|det|>[[148, 852, 850, 904]]<|/det|> +The resistance of Ti as a function of pressure measured at 18 GPa (Fig. S5) exhibits interesting scaling behavior. Upon fitting with the formula \(R = R_0 + A^* T^n\) , where \(R_0\) is the residual resistance, \(A\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 855, 272]]<|/det|> +is the coefficient of the power law, and \(n\) is the exponent, the resulting scaling exponent \(n = 3.1\) is different from the value expected for systems dominated by either electron-phonon scattering \((n = 5)\) or pure electron-electron scattering \((n = 2)\) . In fact, the \(n\) value near 3 implies that the \(s - d\) interband scattering dominates the electron transport with contributions from electron correlation effects, as observed in 1T- TiSe \(_2\) (37) or Ta \(_4\) Pd \(_3\) Te \(_{16}\) (38). Such \(s - d\) interband scattering and the impact of the \(d\) electron correlation effects on the formation of Cooper pairs require further exploration. + +<|ref|>text<|/ref|><|det|>[[147, 288, 853, 638]]<|/det|> +We have examined relative energetic stability of the high- pressure phases of Ti metal from first- principles calculations, and the results are consistent with the experimentally measured phase stability and transformation sequence. Adopting the determined crystal structures of the Ti metal phases, we have calculated their electronic, phonon and electron- phonon coupling properties, which are used as input to determine the SC critical temperature Tc. The obtained Tc data for the Ti- \(\omega\) , Ti- \(\gamma\) , and Ti- \(\delta\) phases (Fig. S6) reveal the following trends: (i) Tc increases with rising pressure monotonically in the Ti- \(\omega\) phase over its entire stability range; (ii) Tc is significantly enhanced upon the phase transition into the Ti- \(\gamma\) phase over the relatively narrow pressure range where this transition occurs; (iii) Tc undergoes an even larger jump when the structure enters the Ti- \(\delta\) phase. These findings are in overall agreement with the experimental results up to about 160 GPa, but large discrepancies exist at higher pressures where the record- setting Tc is observed. + +<|ref|>text<|/ref|><|det|>[[147, 654, 853, 904]]<|/det|> +The experimentally observed superconducting properties of the densely compressed Ti metal indicate clear inadequacy of the conventional phonon- mediated SC mechanism for describing the unexpectedly high values of Tc and anomalously robust superconducting state over a very wide pressure range. The electron- electron correlation effects associated with the \(d\) - bands in Ti are expected to have a strong influence on transport and superconducting properties. Calculated electronic band structure of Ti- \(\omega\) phase at 20 GPa (Fig. 5a) shows significant overlap of the \(s\) - and \(d\) - electron states near the Fermi level, which corroborates our measured normal- state resistivity results (Fig. S5) indicating strong \(s - d\) scattering contribution to the resistivity (37,38). At higher + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 852, 473]]<|/det|> +pressure of 100 GPa, the band structure of Ti- \(\gamma\) phase (Fig. 5b) shows that the \(s\) - states move up in energy while still have large overlap with the \(d\) - bands, which is conducive to significant \(s\) - \(d\) scattering. At further increased pressure of 180 GPa, the electronic states in Ti- \(\delta\) phase (Fig. 5c) near the Fermi level are dominated by the \(d\) - electron states. There are flat \(d\) - bands below the Fermi level that can host significant correlation effects. These \(d\) - bands will likely rise much closer to the Fermi level due to the electron- electron interactions when properly treated by many- body theory. Such correlation effects could greatly enhance the \(d\) - band derived electronic DOS at the Fermi level favorable for increasing Tc and also drive additional non- phonon- mediated mechanisms for further strengthening the superconducting state with higher Tc and maintaining its robust presence over a wide pressure range. This study raises intriguing possibility of major impact by many- body effects on superconductivity in dense Ti, which needs in- depth study by sophisticated many- body theoretical treatments to explore pertinent novel processes and underlying mechanisms. + +<|ref|>text<|/ref|><|det|>[[147, 488, 853, 904]]<|/det|> +The present study unveils unexpectedly robust superconductivity in compressed Ti metal with record- setting Tc among elemental superconductors. The Tc and Hc values of Ti are notably higher than those of the widely used superconducting NbTi alloy. Our discovery raises the possibility of finding more materials via pressure driven correlation effects stemming from the contributions of \(d\) electrons, leading to superconductivity with much higher Tc than previously believed achievable, and such materials may be stabilized at lower pressures via mechanical strain or chemical pressure. This intriguing scenario calls for further research into the impact on superconductivity by the \(s\) - \(d\) interaction and \(d\) - electron dominated correlation effects in highly compressed \(d\) - band metals for evaluation of diverse materials, from elemental solids to alloys, in search of hitherto unknown and unexplored superconducting materials that could improve fundamental understanding of broader varieties of superconductors. Equally important, the present findings open new avenues for expanding the scope of superconductors with notably enhanced and robust Tc and Hc that are more adaptive and suitable for applications in diverse and demanding implementation settings. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 113, 835, 130]]<|/det|> +Acknowledgments: This work is supported by NSF & MOST of China through research projects. + +<|ref|>sub_title<|/ref|><|det|>[[150, 163, 386, 183]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[148, 194, 852, 298]]<|/det|> +Research Design & Supervision: C.Q.J.; high pressure synthesis and in situ resistance measurements: C.L.Z., X.H., Z.W. L., K.L., S.J.Z., S.M.F., X.C.W., Y.W.L. R.C.Y, Y.P., C.Q.J; in situ synchrotron experiments: L. H. W., V.B.P., S.C., H.Z.L., First principle calculations: C.L., Q.L., C.F.C., manuscript writing: C.L.Z., X.C.W. C.Q.J & C.F.C. All authors contributed to the discussions. + +<|ref|>text<|/ref|><|det|>[[148, 346, 851, 420]]<|/det|> +Competing interests: The authors declare no competing interests as defined by Nature Portfolio, or other interests that might be perceived to influence the results and/or discussion reported in this paper. + +<|ref|>sub_title<|/ref|><|det|>[[148, 444, 234, 458]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[147, 463, 852, 907]]<|/det|> +1. B. H. Kear, E. R. Thompson, Science 208, 847-856 (1980). +2. R. R. Boyer, Mat Sci Eng a-Struct 213, 103-114 (1996). +3. M. E. Coon, S. B. Stephan, V. Gupta, C. P. Kealey, M. T. Stephan, Nat Biomed Eng 4, 195-206 (2020). +4. A. Dewaele, V. Stutzmann, J. Bouchet, F. Bottin, F. Occelli, et al., Phys. Rev. B 91, 134108 (2015). +5. H. Xia, G. Parthasarathy, H. Luo, Y. K. Vohra, A. L. Ruoff, Phys. Rev. B 42, 6736-6738 (1990). +6. Y. Akahama, H. Kawamura, T. Le Bihan, Phys. Rev. Lett. 87, 275503 (2001). +7. Y. Akahama, H. Kawamura, T. Le Bihan, J. Phys.: Condens. Matter 14, 10583-10588 (2002). +8. A. L. Kutepov, S. G. Kutepova, Phys. Rev. B 67, 132102 (2003). +9. Y. Akahama, S. Kawaguchi, N. Hirao, Y. Ohishi, J. Appl. Phys. 128, 035901 (2020). +10. Y. K. Vohra, P. T. Spencer, Phys. Rev. Lett. 86, 3068-3071 (2001). +11. I. O. Bashkin, V. G. Tissen, M. V. Nefedova, E. G. Ponyatovsky, Physica C-Superconductivity and Its Applications 453, 12-14 (2007). +12. K. Iyakutti, C. N. Louis, S. Anuratha, S. Mahalakshmi, Int. J. Mod. Phys. B 23, 723-741 (2009). +13. R. Ahuja, J. M. Wills, B. Johansson, O. Eriksson, Phys. Rev. B 48, 16269-16279 (1993). +14. H. L. Skriver, Phys. Rev. B 31, 1909-1923 (1985). +15. T. Yabuuchi, T. Matsuoka, Y. Nakamoto, K. Shimizu, J. Phys. Soc. Jpn. 75, 083703 (2006). +16. J. J. Hamlin, V. G. Tissen, J. S. Schilling, Physica C-Superconductivity and Its Applications 451, 82-85 (2007). +17. M. Ishizuka, M. Iketani, S. Endo, Phys. Rev. B 61, R3823-R3825 (2000). +18. Y. T. Jia, X. He, S. M. Feng, S. J. Zhang, C. L. Zhang, et al., Crystals 10, 1116 (2020). +19. J. L. Zhang, S. J. Zhang, H. M. Weng, W. Zhang, L. X. Yang, et al., Proc Natl Acad Sci U S A/ PNAS 108, 24-28 (2011). +20. Z. W. Li, X. He, C. L. Zhang, S. J. Zhang, S. M. Feng, et al., arXiv:2103.16917 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 74, 852, 485]]<|/det|> +21. Y. T. Jia, C. S. Gong, Y. X. Liu, J. F. Zhao, C. Dong, et al., Chin. Phys. Lett. 37, 097404 (2020). +22. P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, et al., J. Phys.: Condens. Matter 21, 395502 (2009). +23. G. M. Eliashberg, Soviet Physics Jetp-Ussr 11, 696-702 (1960). +24. D. J. Scalapino, Schriefer Jr, J. W. Wilkins, Phys. Rev. 148, 263-279 (1966). +25. W. L. Mcmillan, Phys. Rev. 167, 331-344 (1968). +26. P. B. Allen, R. C. Dynes, Phys. Rev. B 12, 905-922 (1975). +27. P. Morel, P. W. Anderson, Phys. Rev. 125, 1263-1271 (1962). +28. W. L. Mcmillan, J. M. Rowell, (Marcel Dekker New York) (1969). +29. J. P. Carbotte, Rev. Mod. Phys. 62, 1027-1157 (1990). +30. J. P. Carbotte, F. Marsiglio, (Springer-Verlag Berlin Heidelberg Germany) (2003). +31. A. Sanna, J. A. Flores-Livas, A. Davydov, G. Profeta, K. Dewhurst, et al., J. Phys. Soc. Jpn. 87, 041012 (2018). +32a. C. W. Chu, L. Gao, F. Chen, Z. J. Huang, R. L. Meng, and Y. Y. Xue, Nature 365, 323-325(1993) +32. K. Shimizu, H. Ishikawa, D. Takao, T. Yagi, K. Amaya, Nature 419, 597-599 (2002). +33. V. V. Struzhkin, M. I. Eremets, W. Gan, H. K. Mao, R. J. Hemley, Science 298, 1213-1215 (2002). +34. M. Sakata, Y. Nakamoto, K. Shimizu, T. Matsuoka, Y. Ohishi, Phys. Rev. B 83, 220512 (2011). +35. J. S. Tse, Y. Ma, H. M. Tutuncu, Journal of Physics: Condensed Matter 17, S911-S920 (2005). +36. D. Kasinathan, J. Kuneš, A. Lazicki, H. Rosner, C. S. Yoo, et al., Phys. Rev. Lett. 96, 047004 (2006). +37. A. F. Kusmartseva, B. Sipos, H. Berger, L. Forro, E. Tutis, Phys. Rev. Lett. 103, 236401 (2009). +38. J. Pan, W. H. Jiao, X. C. Hong, Z. Zhang, L. P. He, et al., Phys. Rev. B 92, 180505 (2015). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[310, 108, 683, 740]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 770, 851, 861]]<|/det|> +
Fig. 1. (a) Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 1) measured under high pressures. The inset is an enlarged view of the resistance curve, showing the superconducting transition in detail. (b) The resistance curves for Ti Sample 2. (c) The resistance curve measured at 248 GPa, where the derivative of the resistance with respect to temperature dR/dT is plotted to clearly show the onset Tc.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[316, 112, 655, 515]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 530, 850, 601]]<|/det|> +
Fig. 2. (a) Temperature dependence of the electrical resistance of Ti metal measured under different magnetic fields at the fixed pressure of 248 GPa. (b) Upper critical field versus superconducting transition temperature of \(\mathrm{Tc}^{2\mathrm{ero}}\) . The line is a fit obtained using the Ginzburg–Landau function.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[305, 115, 702, 606]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 622, 850, 694]]<|/det|> +
Fig. 3. (a,b) Hall resistance as a function of magnetic field measured under different pressure. (c-e) Carrier density, resistance at fixed temperature, and onset Tc versus pressure, respectively. The yellow and blue areas mark the anomaly of these data, indicating the critical pressures where the phase transitions occur.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[206, 125, 752, 430]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 455, 850, 491]]<|/det|> +
Fig. 4. The superconducting critical transition temperature (Tc) of Ti metal under compression up to 320 GPa plotted against the background of stability fields of the indicated structural phases.
+ +<|ref|>image<|/ref|><|det|>[[195, 555, 861, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 715, 852, 789]]<|/det|> +
Fig. 5. Calculated electronic band structures of three representative Ti phases: (a) Ti-ω at 20 GPa, (b) Ti-γ at 100 GPa and (c) Ti-δ at 180 GPa. The contributions from the s-electron and d-electron states are shown by red and gray circles, respectively, and the circle areas are proportional to the weights of the corresponding band states. Energy is measured relative to the Fermi energy \(\mathrm{E}_{\mathrm{F}}\) .
+ +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[376, 93, 673, 115]]<|/det|> +# Supplementary Materials + +<|ref|>image<|/ref|><|det|>[[275, 247, 690, 517]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 642, 668, 658]]<|/det|> +
Fig.S1 The photo showing sample assembly on the top of diamond cullet..
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[218, 135, 777, 410]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 437, 848, 472]]<|/det|> +
Fig.S2 Temperature dependence of the electrical resistance of elemental metallic Ti (Sample 5) measured under high pressures.
+ +<|ref|>image<|/ref|><|det|>[[170, 518, 825, 688]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 698, 848, 732]]<|/det|> +
Fig. S3 (a) The superconducting transitions measured under different magnetic fields at 154 GPa. (b) at 180 GPa.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[300, 115, 741, 366]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 382, 851, 455]]<|/det|> +
Fig. S4 The in-situ high pressure X-ray diffraction patterns with the highest experimental pressure of 225 GPa, using the wavelength of 0.3344 Å. The pressure induced phase transition sequence from Ti-α phase, to Ti-ω phase, to Ti-γ phase, and then to Ti-δ phase has been observed, which is consistent with the previously reported results (Y. Akahama et al. 2002).
+ +<|ref|>image<|/ref|><|det|>[[266, 508, 690, 737]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 771, 850, 825]]<|/det|> +
Fig. S5. The temperature dependence of resistance measured at 18 GPa. The resistance in the low temperature range of 6-70 K is fitted (red line) with the formula of \(R = R_0 + AT^n\) , where \(R_0\) is the residual resistance, \(A\) is the coefficient of the power law, and \(n = 3.1\) is the extracted exponent.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[295, 108, 688, 318]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 326, 852, 399]]<|/det|> +
Fig. S6. The superconducting critical transition temperature (Tc) of the indicated Ti phases in their stability fields in the [20, 220] GPa hydrostatic pressure range. The calculated crystal structure of Ti is sensitive to the pressure conditions, and under hydrostatic pressures at or above 160 GPa, the Ti-δ phase essentially relaxes to the Ti-β (bcc) phase without considering any stress anisotropy.
+ +<--- Page Split ---> diff --git a/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/images_list.json b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..4847392d29a7959f9b872dba7c5e9371f6582e1d --- /dev/null +++ b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/images_list.json @@ -0,0 +1,18 @@ +[ + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2", + "footnote": [], + "bbox": [], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig.4", + "footnote": [], + "bbox": [], + "page_idx": 49 + } +] \ No newline at end of file diff --git a/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f.mmd b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f.mmd new file mode 100644 index 0000000000000000000000000000000000000000..335ff2b4ebdbe577d5dcba7761e9e4907d4c1247 --- /dev/null +++ b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f.mmd @@ -0,0 +1,791 @@ + +# The mitochondrial mRNA stabilizing protein, SLIRP, regulates skeletal muscle mitochondrial structure and respiration by exercise-recoverable mechanisms + +Lykke Sylow Lykkesylow@sund.ku.dk + +University of Copenhagen https://orcid.org/0000- 0003- 0905- 5932 + +Tang Cam Phung Pham University of Copenhagen + +Steffen Raun University of Copenhagen + +Essi Havula University of Helsinki + +Carlos Henriquez- Olguin University of Copenhagen https://orcid.org/0000- 0002- 9315- 9365 + +Diana Rubalcava- Gracia Karolinska Institute https://orcid.org/0000- 0002- 4615- 7375 + +Emma Frank University of Copenhagen + +Andreas Fritzen University of Copenhagen + +Paulo Jannig Karolinska Institutet https://orcid.org/0000- 0002- 2569- 4091 + +Nicoline Andersen University of Copenhagen + +Rikke Kruse Odense University Hospital + +Mona Ali University of Copenhagen + +Jens Halling University of Copenhagen + +Stine Ringholm University of Copenhagen + +<--- Page Split ---> + +Elise Needham University of Cambridge Solvejg Hansen University of Copenhagen Anders Lemminger University of Copenhagen Peter Schjerling Bispebjerg Hospital https://orcid.org/0000-0001-7138-3211 Maria Petersen Odense University Hospital Martin de Almeida University of Southern Denmark https://orcid.org/0000-0003-1794-6137 Thomas Jensen University of Copenhagen https://orcid.org/0000-0001-6139-8268 Bente Kiens University of Copenhagen https://orcid.org/0000-0001-5705-5625 Morten Hostrup University of Copenhagen https://orcid.org/0000-0002-6201-2483 Steen Larsen University of Copenhagen https://orcid.org/0000-0002-5170-4337 Niels Ørtenblad University of Southern Denmark https://orcid.org/0000-0001-9060-4139 Kurt Højlund Odense University Hospital Michael Kjær Bispebjerg Hospital Jorge Ruas Karolinska Institutet Aleksandra Trifunovic University of Cologne https://orcid.org/0000-0002-5472-3517 Jørgen Wojtaszewski University of Copenhagen https://orcid.org/0000-0001-8185-3408 Joachim Nielsen University of Southern Denmark https://orcid.org/0000-0003-1730-3094 Klaus Qvortrup University of Copenhagen https://orcid.org/0000-0001-8811-0260 Henriette Pilegaard University of Copenhagen https://orcid.org/0000-0002-1071-0327 Erik Richter + +<--- Page Split ---> + +## Article + +Keywords: SRA stem- loop interacting RNA- binding protein, mitochondrial dysfunction, skeletal muscle, exercise training, Drosophila + +Posted Date: March 1st, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 3900062/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on November 13th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 54183- 4. + +<--- Page Split ---> + +1 The mitochondrial mRNA stabilizing protein, SLIRP, regulates skeletal muscle mitochondrial structure and respiration by exercise-recoverable mechanisms + +3 Tang Cam Phung Pham \(^{1,2}\) , Steffen Henning Raun \(^{2}\) , Essi Havula \(^{3}\) , Carlos Henriquez- Olguin \(^{1}\) , Diana Rubalcava- Gracia \(^{4}\) , Emma Frank \(^{2}\) , Andreas Mæchel Fritzen \(^{1,2}\) , Paulo R. Jannig \(^{5}\) , Nicoline Resen Andersen \(^{1}\) , Rikke Kruse \(^{6}\) , Mona Sadek Ali \(^{2}\) , Jens Frey Halling \(^{7}\) , Stine Ringholm \(^{7}\) , Elise J. Needham \(^{8,9}\) , Solvejg Hansen \(^{1}\) , Anders Krogh Lemminger \(^{1}\) , Peter Schjerling \(^{10,11}\) , Maria Houborg Petersen \(^{6}\) , Martin Eisemann de Almeida \(^{6,12}\) , Thomas Elbenhardt Jensen \(^{1}\) , Bente Kiens \(^{1}\) , Morten Hostrup \(^{1}\) , Steen Larsen \(^{2,10,11,13}\) , Niels Ørtenblad \(^{12}\) , Kurt Højlund \(^{6,14}\) , Michael Kjær \(^{10,11}\) , Jorge L. Ruas \(^{5}\) , Aleksandra Trifunovic \(^{15}\) , Jørgen Frank Pind Wojtaszewski \(^{1}\) , Joachim Nielsen \(^{12}\) , Klaus Qvortrup \(^{2}\) , Henriette Pilegaard \(^{7}\) , Erik Arne Richter \(^{1}\) , Lykke Sylow \(^{1,2*}\) + +11 ORCID: TCPP, 0000- 0003- 3476- 6891; SHR, 0000- 0003- 2050- 505X; EH, 0000- 0002- 9253- 5816; 12 DRG, 0000- 0002- 4615- 7375; JN, 0000- 0003- 1730- 3094; AMF, 0000- 0001- 8793- 9739; PS, 0000- 13 0001- 7138- 3211; NRA 0000- 0002- 1512- 2727; MSA, 0000- 0003- 1152- 3256; EAR, 0000- 0002- 6850- 14 3056; LS, 0000- 0003- 0905- 5932; MK, 0000- 0002- 4582- 8755; EF: 0000- 0002- 2294- 952X; AKL, 15 0000- 0002- 7929- 7147; MH, 0000- 0002- 6201- 2483; SR, 0000- 0001- 8436- 559X; HP, 0000- 0002- 16 1071- 0327; SL, 0000- 0002- 5170- 4337; PRJ, 0000- 0002- 2569- 4091; JLR, 0000- 0002- 1110- 2606; JW, 17 0000- 0001- 8185- 3408) + +18 1 Department of Nutrition, Exercise and Sports, Faculty of Science, University of Copenhagen, Copenhagen, Denmark 20 2 Department of Biomedical Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark 21 3 Stem Cells and Metabolism Research Program, Faculty of Medicine, University of Helsinki, Finland 22 4 Division of Molecular Metabolism, Department of Medical Biochemistry and Biophysics, Karolinska Institutet, Stockholm, Sweden 25 5 Molecular and Cellular Exercise Physiology, Department of Physiology and Pharmacology, Karolinska Institutet, SE- 17177 Stockholm, Sweden 27 6 Steno Diabetes Center Odense, Odense University Hospital, Odense, Denmark 28 7 Department of Biology, University of Copenhagen, Copenhagen, Denmark + +<--- Page Split ---> + +29 'British Heart Foundation Cardiovascular Epidemiology Unit, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK + +31 'Victor Phillip Dahdaleh Heart and Lung Research Institute, University of Cambridge, Cambridge, UK + +33 'Institute of Sports Medicine Copenhagen, Department of Orthopaedic Surgery M, Bispebjerg Hospital, Copenhagen, Denmark + +35 'Center for Healthy Aging, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark + +37 'Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark + +39 'Clinical Research Centre, Medical University of Bialystok, Bialystok, Poland + +40 'Department of Clinical Research, University of Southern Denmark, Odense, Denmark + +41 'Institute for Mitochondrial Diseases and Aging, Cologne Excellence Cluster on Cellular Stress + +42 Responses in Aging- Associated Diseases (CECAD) and Center for Molecular Medicine (CMMC), + +43 Medical Faculty, University of Cologne, D- 50931 Cologne, Germany + +44 + +45 Running title: Exercise reverses SLIRP- induced mitochondrial defects + +46 Keywords: SRA stem- loop interacting RNA- binding protein, mitochondrial dysfunction, skeletal + +47 muscle, exercise training, Drosophila + +48 \*Corresponding author: Lykke Sylow + +49 Lykkesylow@sund.ku.dk + +50 Competing Interests. The authors declare no competing interests. + +<--- Page Split ---> + +## Summary and graphical abstract + +Summary and graphical abstractDecline in mitochondrial function associates with decreased muscle mass and strength in multiple conditions, including sarcopenia and type 2 diabetes. Optimal treatment could include improving mitochondrial function, however, there are limited and equivocal data regarding the molecular cues controlling muscle mitochondrial plasticity. Here we uncover the mitochondrial- mRNA- stabilizing protein SLIRP, in complex with LRPPRC, as a PGC- 1α target that regulates mitochondrial structure, respiration, and mitochondrially- encoded- mRNA pools in skeletal muscle. Exercise training effectively counteracted mitochondrial defects induced by loss of LRPPRC/SLIRP, despite sustained low mitochondrially- encoded- mRNA pools, via increased mitoribosome translation capacity. In humans, exercise training robustly increased muscle SLIRP and LRPPRC protein content across exercise modalities and sexes, yet this increase was less prominent in subjects with type 2 diabetes. Our work identifies a mechanism of post- transcriptional mitochondrial regulation in skeletal muscle through mitochondrial mRNA stabilization. It emphasizes exercise as an effective approach to alleviate mitochondrial defects by possibly increasing mitoribosome capacity. + +![](images/Figure_2.jpg) + + +<--- Page Split ---> + +## Introduction + +Mitochondrial homeostasis and function are vital for skeletal muscle physiology, primarily influenced by the ability to meet energy demands through oxidative phosphorylation (OXPHOS) in mitochondria1. The decline in mitochondrial function is associated with decreased muscle mass and strength in multiple conditions, including sarcopenia, type 2 diabetes (T2D), and cancer. Reduced muscle function adversely impacts health and quality of life1- 9 and is associated with increased all- cause mortality. Mounting evidence suggests that improvements in mitochondrial metabolism contribute to the exercise- induced functional benefits, as mitochondrial preservation protects against the age- associated decline in skeletal muscle mass and performance10, 11. The beneficial effects of exercise training are more potent than any drug in preserving muscle mass and function, positioning exercise training as a frontline strategy for the prevention and treatment of sarcopenia, T2D, and cancer12- 14. Viewing mitochondrial control through the lens of exercise biology is a strategy to gain new mechanistic insights into mitochondrial regulation, which is currently poorly understood. With an aging population and no FDA/EMA- approved drugs to treat muscle functional decline, there is an urgent unmet need to identify potential treatment targets. + +OXPHOS proteins are of dual origin and transcribed in spatially distinct cellular compartments. Nuclear (n)DNA encodes for the majority of OXPHOS proteins. Yet, 13 essential OXPHOS proteins are encoded by the small circular high- copy number mitochondrial DNA (mtDNA)15, 16. In non- muscle systems, a protein complex comprised of steroid receptor RNA activator stem- loop interacting RNA- binding protein (SLIRP) and the leucine- rich pentatricopeptide repeat containing protein (LRPPRC) mediates mitochondrial- mRNA (mt- mRNA) stability and polyadenylation of most mtDNA- encoded OXPHOS transcripts17- 21. Stabilization of mt- mRNA is critical for protein translation by the mitochondrial ribosome (mitoribosome). Yet, the mitoribosome comprises 82 mitoribosomal proteins encoded by nuclear genes, and 12S and 16S mt- rRNAs, highlighting the requirement for intricate coordination between the processes of cytosolic and mitochondrial protein synthesis18, 19. Despite the high mitochondrial abundance in skeletal muscle, the complex interaction between LRPPRC/SLIRP- mediated posttranscriptional processes, mitoribosomal translation, and mitochondrial function has not + +<--- Page Split ---> + +been studied before in skeletal muscle. Elucidating the underlying mechanisms for stabilizing mt- mRNA could not only facilitate our understanding of the fundamental energy metabolism in muscle, but also aid intervention strategies for diseases associated with skeletal muscle mitochondrial defects. Endurance exercise training potently increases mitochondrial mass in skeletal muscle, in part due to the increased content of nDNA- and mtDNA- encoded mitochondrial proteins1, 22. Importantly, exercise training retains its ability to improve oxidative capacity not only in healthy individuals but also in patients with mtDNA mutations, with common PGC- 1α- mediated mitochondrial adaptive responses shared between both groups13. These results suggest that exercise training induces adaptations to reinforce the mitochondria's ability to efficiently respond to increased energy demands, even in the presence of mtDNA mutations, that may negatively affect mitochondrial transcription and translation. However, a notable knowledge gap persists in understanding the role of mitochondrial posttranscriptional processes, specifically in mt- mRNA stabilization and translation, in skeletal muscle biology and following exercise training. Yet, that knowledge is needed to gain a comprehensive understanding of the adaptive responses to exercise training. + +Since SLIRP, a key player in mitochondrial posttranscriptional gene expression, is markedly upregulated in mouse skeletal muscle by exercise training23, we hypothesized that SLIRP would regulate mitochondrial function in skeletal muscle at rest and in response to exercise training. Our findings illuminate SLIRP's role in regulating mt- mRNA transcript levels in skeletal muscle, downstream of PGC- 1α1. Knockout (KO) of SLIRP led to damaged and fragmented mitochondria alongside lowered respiration. Intriguingly, exercise training could compensate for the absence of SLIRP, leading to improvements in mitochondrial integrity and respiratory capacity. Our findings imply the activation of complex exercise- induced molecular signaling in skeletal muscle which intriguingly bypasses mt- mRNA defects, possibly through enhanced mitoribosomal translation. + +<--- Page Split ---> + +## Results + +Slirp knockout caused defects in mitochondrial structure and respiratory capacity in mouse muscle. + +Protein profiling of five different skeletal muscle tissues showed that SLIRP content was highest in the oxidative soleus compared to glycolytic extensor digitorum longus (EDL, Fig. 1A, Supplementary Fig. 1A), in accordance with the potential critical role for SLIRP in oxidation. Moreover, SLIRP was present across a diverse array of tissues in mice and was highly abundant in energy- demanding tissues such as liver, kidney, brown adipose tissue, heart, and skeletal muscle (Fig. 1A, Supplementary Fig. 1A), in line with previously reported Slirp mRNA levels24. + +In skeletal muscle of global Slirp KO mice, SLIRP protein was undetectable. Moreover, SLIRP's binding partner LRPPRC showed \(90\%\) reduction in protein content (Fig. 1B), indicative of their co- stabilization. To confirm co- stabilization, we administered intramuscular injections of recombinant adeno- associated viral serotype 6 (rAAV6): Slirp. As expected, the absence of concomitant upregulation of LRPPRC, failed to increase total SLIRP protein content due to downregulation of endogenous SLIRP protein (Fig. 1C). These findings show that SLIRP and LRPPRC stabilize each other in skeletal muscle, aligning with findings from non- muscle tissue17, 18, 20. + +We next determined the role of SLIRP for mitochondrial structure in skeletal muscle, using the membrane potential probe, tetramethylrhodamine ethyl ester (TMRE+), in intact flexor digitorum brevis (FDB) muscle fibers of Slirp KO and littermate control wild- type (WT) mice. Interestingly, the finely interconnected mitochondrial network, obvious in WT fibers, was disrupted in Slirp KO muscle fibers, evident by a \(30\%\) increased mitochondrial fragmentation index (Fig. 1D). Transmission electron microscopy (TEM) analysis of gastrocnemius muscle provided further insights into the ultrastructural changes of mitochondrial morphology in Slirp KO (Fig. 1E). Half of the KO gastrocnemius muscles displayed reduced density of the matrix, disarray of the cristae, and substantial enlargement and vacuolation of intermyofibrillar (IMF) mitochondria (Fig. 1E). Quantitative analysis of the mitochondria showed that the percentage of damaged IMF mitochondria was increased (Fig. 1F). + +<--- Page Split ---> + +However, there were no significant alterations in the total number of IMF mitochondria (Fig. 1G), number of damaged or the total number of subsarcolemmal mitochondria (Fig. 1G), nor any changes in number of damaged or total number of subsarcolemmal mitochondria (not shown). The percentage of damaged mitochondria in Slirp KO muscle varied between \(2 - 20\%\) of total mitochondria compared to \(< 2\%\) of damaged mitochondria in WT muscle (Fig. 1E, F). Other structural parameters of IMF mitochondria, such as mitochondrial area, aspect ratio, elongation, convexity, perimeter, sphericity, and diameter did not display any significant changes in Slirp KO muscles, likely due to heterogeneity in the extent of damage (Fig. 1F, Supplementary Fig. 1B- H). + +In agreement with the observable abnormalities in mitochondrial structure, maximal respiratory capacity was reduced in permeabilized gastrocnemius muscle fibers of Slirp KO mice compared to WT (Fig. 1H). This was particularly evident when adding glutamate to assess maximal complex I linked respiratory activity (- 24%) and succinate to assess complex I+II linked respiratory capacity (- 21%). These results indicate an important role for SLIRP in skeletal muscle respiration, which contrast findings in isolated mitochondria from liver and heart tissues, where Slirp KO did not compromise mitochondrial respiration18. In agreement with lower respiration, fatty acid oxidation tended (p=0.0718) to be lower in in intact incubated soleus muscle of Slirp KO mice (Fig. 1I). However, electrically-induced muscle contraction increased fatty acid oxidation similarly in both genotypes, suggesting that fatty acid utilization for fuel during muscle contraction does not depend on SLIRP (Fig. 1I). + +With the suggested role of SLIRP as an mt-mRNA stabilizing protein in non-muscle cells18, we determined mtDNA- encoded mRNA transcripts. Intriguingly, we observed a 60- 80% reduction in mt- Nd1, mt- Nd5/Nd6 (Complex I), mt- CytB (Complex III), mt- Col (Complex IV), and mt- Atp6 (ATP synthase) in gastrocnemius muscle (Fig. 1J), suggesting that SLIRP stabilizes mt- mRNA in skeletal muscle. + +Together, these results suggest that mt- mRNA stabilization via SLIRP is required for proper mitochondrial network structure and morphology, and respiration in mouse muscle. + +<--- Page Split ---> + +## SLIRP knockdown impaired locomotion and reduced lifespan in flies. + +To determine the long- term consequences of SLIRP deficiency on the whole organism, we utilized the UAS- GAL4 system \(^{25}\) for muscle- specific (Mef2- GAL4> ) knock- down (KD) of SLIRP1 and SLIRP2 in Drosophila (D.) melanogaster. Only one SLIRP gene is present in mouse and human, while two SLIRP genes exist in D. melanogaster (Flybase annotation symbols (http://flybase.org): CG33714 and CG8021) likely to have originated from gene duplication events \(^{26}\) . This consequently gives rise to two fly orthologue proteins of the human and mouse SLIRP, denoted SLIRP1 (CG33714) and SLIRP2 (CG8021) \(^{26}\) . To test physical functionality, we subjected the flies to a negative geotaxis (climbing) assay \(^{27}\) and found that SLIRP2 KD, but not SLIRP1 KD, flies climbed 23% slower than control flies, indicative of impaired muscle function (Fig. 1K). The detrimental effects of SLIRP KD in the muscle became further apparent following starvation. Median fasting survival of SLIRP1 KD flies was 41 h, compared with 44 h for control and SLIRP2 KD flies (Fig. 1L, p=0.0681). Muscle- specific SLIRP deficiency, irrespective of orthologue, had a deleterious effect on lifespan. While the median survival of control flies was 77 days, median survival of SLIRP1 and SLIRP2 KD flies was only 55 and 63 days, respectively, and thus both SLIRP1 and SLIRP2 KD significantly reduced lifespan (Fig. 1M). + +Taken together these findings in the flies show that SLIRP KD- induced mitochondrial defects can have detrimental long- term consequences such as impaired muscle function, starvation intolerance, and reduced life span. + +## Muscle SLIRP protein content increases in response to exercise training, under the regulation of PGC-1α1. + +Having established critical functional roles of SLIRP in skeletal muscle mitochondrial morphology and respiratory capacity with detrimental effects on lifespan, we next investigated the upstream regulation of SLIRP protein content. Our recent work pinpointed SLIRP as a protein responsive to exercise training \(^{23}\) ; yet the mechanisms governing its induction and functions remained elusive. Knowing that exercise potently increases PGC- 1α protein and mRNA levels in both mouse and human skeletal muscle \(^{28,31}\) , and that PGC- 1α is an important regulator of mitochondrial biogenesis, respiration, and + +<--- Page Split ---> + +quality control \(^{32}\) , we explored the regulation of SLIRP by PGC- 1α. The Ppargc1a gene encodes several PGC- 1α isoforms, including isoform 1 (PGC- 1α1) and isoform 4 (PGC- 1α4), with distinct regulation and biological functions \(^{33, 34}\) . Overexpression (OE) of PGC- 1α1 in mouse skeletal muscle \(^{33, 35}\) , associating with endurance exercise training- like adaptations \(^{33, 35, 36}\) , resulted in approximately 8.3- fold higher muscle SLIRP protein content, without a concomitant change in mRNA levels (Fig. 2A). On the contrary, PGC- 1α4 OE \(^{37}\) , inducing resistance- type exercise training adaptations such as muscle hypertrophy and strength \(^{33}\) , had no effect on SLIRP protein content in mouse gastrocnemius muscle (Fig. 2B). This suggests that SLIRP is an unrecognized player in PGC- 1α1- regulated oxidative metabolism. + +To test whether PGC- 1α regulates Slirp mRNA in skeletal muscle following exercise, we determined Slirp and Ppargc1a (Exon 3- 5) mRNA levels in tibialis anterior (TA) muscle 3 h into recovery from one exercise bout in mice lacking PGC- 1α in skeletal muscle (PGC- 1α mKO) and WT control littermates. Slirp mRNA was not upregulated 3 h post- exercise (Fig. 2C), in contrast to the expected \(^{31}\) increase in Ppargc1a (+6.8- fold; Fig. 2C) in WT mice. Interestingly, Slirp mRNA levels were only modestly 14% reduced in PGC- 1α KO skeletal muscle (Fig. 2C), in line with the dissociation between Slirp mRNA levels and SLIRP protein levels observed in PGC- 1α1 OE skeletal muscle (Fig. 2A). + +These findings align with reports describing only moderate correlations between mRNA and protein levels following exercise \(^{38 - 40}\) . Yet, with the interpretative limitations of a single post- exercise time point, we aimed to acquire a more detailed record of the temporal changes in Slirp transcripts in recovery from exercise. We subjected WT mice to an acute exercise bout at 60% of their maximal running capacity and harvested quadriceps muscles at rest, immediately after, 2h, 6h, and 24h post- exercise. The phosphorylation of AMP- activated protein kinase (AMPK) at T172 served as a validation of elevated metabolic stress during the exercise bout (Fig. 2E). We found that Slirp mRNA levels were 1.4- fold elevated 6h into the recovery period in quadriceps muscle (Fig. 2D), a time- point not included in our prior cohort (Fig. 2C), and partially concurrent with PGC- 1α1 mRNA levels that were elevated 2h (+3.0- fold) and 6h post- exercise (+3.3- fold; Fig. 2D). Following a single bout of exercise, we observed + +<--- Page Split ---> + +no changes to SLIRP protein content (Fig. 2E). These findings indicate that SLIRP underlies PGC- 1α1- regulated control following acute exercise. + +Given that SLIRP might be regulated by PGC- 1α, we next turned to investigate the long- term dependency of exercise training- induced regulation of SLIRP protein by PGC- 1α. In contrast to acute exercise, exercise training increased SLIRP protein content 4.8- fold in WT mice (Fig. 2F), recapitulating our earlier observations in mice engaged in voluntary wheel running23. The effect of exercise training on SLIRP protein content was markedly blunted by 70% in mice lacking PGC- 1α in muscle compared to trained WT mice (Fig. 2F). Expectedly, in WT mice, exercise training potently upregulated the steady- state levels of multiple OXPHOS proteins including SDHB, UQCRC2, MTCO1 in skeletal muscle (Supplementary Fig. 1I- K). In mice lacking PGC- 1α in muscle, SDHB (Supplementary Fig. 1I) and MTCO1 protein (Supplementary Fig. 1K) was 50- 60% decreased in SED mice. Moreover, the effect of exercise training on SDHB (Supplementary Fig. 1I) and UQCRC2 protein content (Supplementary Fig. 1J) was blunted by 30% and 65%, respectively, in PGC- 1α mKO compared to WT mice. In contrast to PGC- 1α, the exercise- sensitive metabolic sensor AMPK41 did not seem to regulate SLIRP expression, as SLIRP protein abundance was comparable to controls in muscle- specific double KO of AMPKα1 and AMPKα2 and WT mice42, 43 treated with or without the AMPK- activator AICAR for 4 weeks (Fig. 2G). + +PGC- 1α has marked effects on muscle endurance, partially attributed to its pivotal role in mitochondrial biogenesis, and respiration35, 44. Our results suggest SLIRP as an unrecognized downstream target of PGC- 1α in such exercise adaptations and oxidative metabolism. + +SLIRP is dispensable for organismal adaptations to exercise training, yet, needed for improving blood glucose regulation after exercise training in male mice. + +Having identified SLIRP as an exercise training responsive protein downstream of PGC- 1α, we next determined SLIRP's mechanistic involvement in exercise training- mediated organismal adaptations. To this end, we conducted a 10- week voluntary wheel- running training study in Slirp KO and WT + +<--- Page Split ---> + +littlerate mice of both sexes commencing at \(\sim 18\) weeks of age. The experimental design is shown in Fig. 3A. + +The daily running distance recorded over 6 weeks of the 10- week study was \(2.4\mathrm{km / day}\) on average for male mice of both genotypes, whereas it was \(4.8\mathrm{km / day}\) for female exercise- trained (ET) WT mice and \(3.0\mathrm{km / day}\) for female KO ET mice (Fig. 3B). Of note, the running distance observed in our study of 18- week old mice is shorter than those reported in other exercise training studies using \(\sim 10\) - week- old mice \(^{23,45,46}\) . Despite the higher food intake in ET mice (Fig. 3C), exercise training lowered body weight in both trained genotypes (Supplementary Fig. 2A). Moreover, irrespective of genotype, exercise training reduced fat mass/body weight ratio (Supplementary Fig. 2B) and increased lean mass/body weight ratio compared to sedentary (SED) groups (Supplementary Fig. 2C). Exercise training had no effect on gastrocnemius or TA muscle weight relative to tibia length in either sex (Supplementary Fig. 2D). However, we noted that female Slirp KO SED and both ET groups had elevated heart weight relative to WT SED (Supplementary Fig. 2D). + +Exercise capacity increased similarly with exercise training, independently of genotype (+1.2- fold for WT ET, +1.3- fold for KO ET; Supplementary Fig. 2E) and sex (females shown in Supplementary Fig. 2G). Post- exercise blood lactate levels tended to be reduced (- 1.7- fold; \(\mathrm{p} = 0.067\) ) by ET in WT male mice (Supplementary Fig. 2F). The training- induced lowering of blood lactate post- exercise was absent in male and female Slirp KO mice (Supplementary Fig. 2F, H). + +Exercise training is a powerful preventative treatment against many metabolic disorders \(^{12}\) , partially due to its remarkable effects in improving glucose tolerance and insulin sensitivity documented in both animal models and humans \(^{23,45,47}\) . Accordingly, male WT ET mice displayed reduced fasting blood glucose (Fig. 3D) and improved blood glucose control evidenced by a downshifted glucose response curve compared to WT SED mice. This effect was not observed in trained male Slirp KO mice (Fig. 3E). Thus, trained Slirp KO mice displayed similar glucose tolerance as both SED groups. The incremental area under the curve (iAUC) of blood glucose was similar between groups (Fig. 3F), suggesting that the lowered plasma glucose excursion during the GTT was driven by lower basal glucose levels. Yet, exercise training reduced the levels of plasma insulin in both WT and Slirp KO + +<--- Page Split ---> + +mice, suggesting that insulin sensitivity was increased by exercise training in both groups (Fig. 3G). There was no effect of exercise training or genotype on fasting glucose, glucose tolerance or insulin levels in female mice (Fig. 3H- K). + +Our results demonstrate that SLIRP is largely dispensable for organismal adaptations to exercise training, yet, in male mice required for training- induced improvement in blood glucose regulation. + +## Exercise training reverses Slirp KO-induced defects in muscle mitochondrial structure and function. + +Exercise training can counteract mitochondrial damage, arising from excessive accumulation of reactive oxygen species (ROS) or impaired assembly of OXPHOS complex following mtDNA mutations11, 13, 48- 50. As we established that loss of SLIRP had marked negative implications for mitochondrial structure and respiration in skeletal muscle (Fig. 1), we asked whether exercise training could rescue these defects. + +Skeletal muscle mitochondrial network structure was qualitatively investigated in TMRE+ stained FDB fibers by confocal microscopy. Remarkably, exercise training completely rescued the derangements in mitochondrial network structure and mitochondrial fragmentation observed in Slirp KO SED mice, quantitatively corroborated by restoration of the fragmentation index (Fig. 4A). + +Concomitantly with improved mitochondrial structure, exercise training restored the respiratory flux in Slirp KO mice to the level of WT muscle (Fig. 4B). WT mice did not improve respiratory flux with exercise training, which was consistent with no effect of exercise training on fatty acid oxidation of contracting isolated WT soleus muscle (Supplementary Fig. 2I) and aligns with other studies51. Exercise training in Slirp KO mice improved fatty acid oxidation in contracting Slirp KO soleus muscles (Supplementary Fig. 2I), in alignment with improved respiratory flux in Slirp KO muscles by exercise training. Thus, exercise training counteracted the structural alterations in Slirp KO SED mice and had a positive impact on mitochondrial respiratory capacity, suggesting restored mitochondrial function. + +<--- Page Split ---> + +Together, these results support that exercise training exploits the remarkable adaptive plasticity that mitochondria retain even in the event of mitochondrial network disruptions and impaired respiratory flux in skeletal muscle. + +## Sustained reductions of mitochondrial transcript levels can be bypassed by exercise training. + +We next aimed to investigate the molecular underpinnings for the defects in mitochondrial structure and function in Slirp KO and their correction by exercise training. First, we verified that both SLIRP (+1.4- fold; Fig. 4C) and concomitantly LRPPRC (+1.4- fold; Fig. 4D) were upregulated by exercise training in the gastrocnemius muscle in WT, but not Slirp KO mice (Fig. 4C, D). These findings reinforce the co- stabilizing relationship of SLIRP and LRPPRC existing not only at baseline17, 18, 20, 52 but also during exercise training conditions. + +SLIRP, together with LRPPRC, has been shown to maintain mt- mRNA stability and aid mitoribosomal translation in non- muscle tissues17- 20, 52. Indeed, the marked downregulation of mt- mRNA transcripts in SED Slirp KO muscle, also shown in Fig. 1J, remained reduced in ET Slirp KO mice relative to WT ET mice (Fig. 4F). In contrast, mtDNA copy number, measured using quantitative PCR and primers for mt- Ndl (Supplementary Fig. 2J) or mt- Nds/mt- Nds (Supplementary Fig. 2K), were unaltered across groups or elevated only in Slirp KO SED mice, suggesting a compensatory response to mitigate mitochondrial dysfunction by increasing mtDNA copy number53. Thus, these findings suggest that SLIRP is crucial for mitochondrial transcript stability in skeletal muscle. We were intrigued to find that exercise training still restored protein content of mtDNA- encoded OXPHOS subunits, MT- ND3, MT- CO1 and MT- ATP6 (Fig. 4F). Thus, exercise training completely counteracted the sustained reduction in the mt- mRNA levels of these proteins in Slirp KO ET mice. The mild reduction of MT- ND3 protein, a subunit of Complex I, seen in Slirp KO muscle was similarly rescued by exercise training (Fig. 4F). Interestingly, although the nuclear- encoded Complex I subunit NDUFB8 was downregulated by 57% in the Slirp KO SED mice, exercise training markedly upregulated NDUFB8 protein content in Slirp KO mice, yet, below NDUFB8 protein content in WT ET mice (Fig. 4G). Other nuclear- encoded OXPHOS subunits, such as SDHB (Complex II), UQCRC2 (Complex III) and ATP5A (ATP synthase) were similarly upregulated by exercise training, independently of Slirp KO (Fig. 4G, representative + +<--- Page Split ---> + +blots shown in Fig. 4H). Citrate synthase (CS) activity correlates with mitochondrial content in human54 and mouse skeletal muscle55. CS activity, indicative of mitochondrial content, was increased similarly in Slirp KO and WT mice after exercise training (Supplementary Fig. 2L), suggesting that improvements in mitochondrial content following exercise training are independent of SLIRP. + +These results underline a substantial skeletal muscle mitochondrial plasticity, which is retained despite a \(60 - 80\%\) reduction in mt- mRNA pools. These findings point towards unidentified mechanisms by which ET improves translational efficiency to dictate final protein content. + +## Exercise training increases mitochondrial protein translation capacity by elevating mitoribosome mass. + +We next sought to determine the potential mechanism by which sustained reductions of mitochondrial transcript levels can be bypassed by exercise training. Enhanced mitochondrial protein translation mediates exercise training adaptations in muscle mitochondrial function in both young and elderly individuals39. Thus, we assessed pathways associated with protein translation in mitochondria, facilitated by the resident mitoribomes. We observed training- induced increases in 12S ribosomal RNA (12S rRNA) and MRPL11 protein content (Fig. 5A, B), essential components of mitoribosome translation56, 57. MRPL11 protein content was \(36\%\) more elevated in response to exercise training in Slirp KO ET than WT ET mice, mainly in male mice (Fig. 5B). Conversely, exercise training did not induce changes in cytosolic ribosomal protein S6 protein content (rpS6, Fig. 5C; representative blots in Fig. 5D), indicating a compartmentalized ribosomal response to exercise training. + +Given the predominant impact of Slirp KO on mitochondria, training- induced adaptive changes might also manifest directly within the affected cellular compartment. Here, we investigated a mitochondrial quality control system involving a mitochondrial scavenger enzyme, peroxiredoxin 3 (PRDX3). PRDX3 plays a role in balancing redox environment in mitochondria58, 59, possibly contributing to the preservation of factors critical for translation during conditions of mitochondrial dysfunction and exercise training. Interestingly, PRDX3 protein content was \(60\%\) lower in Slirp KO SED compared to WT SED mice, indicative of a lower mitochondrial oxidative stress defense capacity in Slirp KO + +<--- Page Split ---> + +muscle. This was restored by exercise training, bringing PRDX3 up to levels observed in trained WT mice (Fig. 5E). We further noted that ET lowered the PRDX3 dimer to monomer ratio in both genotypes, indicative of a lower basal mitochondrial- derived peroxide accumulation likely due to an enhancement of scavenger activity (Fig. 5F). The cytosolic peroxidoxin 2 (PRDX2) was unaffected by genotype and exercise training (Fig. 5G, representative blots in Fig. 5H), further pointing towards highly mitochondria- selective effects of exercise training to circumvent SLIRP deficiency. + +Collectively, our findings indicate a complex interplay of spatially distinct molecular muscle adaptations in response to Slirp KO and ET. Our findings suggest that exercise training induces mechanisms to increase mitochondrial protein translation capacity. Mechanistically, exercise training elevated mitoribosome mass and restored oxidative stress defense systems to powerfully circumvent the Slirp KO- induced depletion of mtDNA- encoded transcript levels. + +## Elevation of muscle SLIRP and LRPPRC protein content in response to ET is conserved in human skeletal muscle. + +Our intriguing results from mouse and fly models prompted us to test whether our results provide translational value for humans in health and disease. SLIRP and LRPPRC protein content were determined in human skeletal muscle in four independent exercise cohorts employing different exercise modalities in conjunction with or without aging or diabetes and in males and females. Participant characteristics and study designs for the different ET modalities have been published previously for all the clinical studies60-63. + +Skeletal muscle SLIRP and LRPPRC protein abundances were 70% increased following 14- weeks of controlled and supervised aerobic and strength exercise training intervention in healthy young women60 (Fig. 6A). Also, a 6- week high- intensity interval training (HIIT) intervention elevated muscle protein content of SLIRP (+80%, Fig. 6B) and LRPPRC (+50%, Fig. 6B) in healthy young men63. Thus, the exercise training response of SLIRP was highly consistent, independent of sex, and responsive to various exercise modalities. The 12- week progressive resistance exercise training61 increased SLIRP and LRPPRC in the entire group of young and old individuals but had no significant effect on LRPPRC + +<--- Page Split ---> + +abundance in either young or elderly subjects separately (Fig. 6C). The mtDNA- encoded transcript levels of MT- ND6, MT- CYB, MT- CO1 and MT- ATP6 were comparable across all groups (Fig. 6D), indicating that mitochondrial gene expression was not significantly affected by either aging or the training regimen in this study. + +Exercise training- mediated improvements in mitochondrial content and function may be associated with improved glucose metabolism in individuals with obesity and T2D64,65, although this association is not always clear66. The training response of SLIRP and LRPPRC to HIIT62 was conserved in individuals who were lean (+150% and +54%) or obese (+85% and +45%, Fig. 6E). However, the exercise response for LRPPRC, but not SLIRP, was blunted in patients with T2D in comparison to lean individuals (Fig. 6E), suggesting impaired sensitivity of muscle to some mitochondrial adaptations to ET in T2D. At the mRNA level, mt- DNA encoded transcripts of MT- ND1, MT- CYB, MT- CO1 and MT- ATP6 were reduced with HIIT across all groups (Fig. 6F). Yet, protein content of MT- CO2 (Fig. 6G, representative blot in Supplementary Fig. 3A) and other nuclear- encoded OXPHOS proteins (Supplementary Fig. 3A) were significantly upregulated with HIIT. In line with previous findings39, the upregulation in OXPHOS protein abundance occurred despite lower levels of mRNA, and demonstrate a disassociation between the transcriptome and proteome67. + +Together, we show that SLIRP and LRPPRC are robustly upregulated in skeletal muscle across various exercise training modalities in men and women, lean and obese. However, the exercise training response of LRPPRC was blunted in patients with T2D. This suggests that T2D is associated with decreased sensitivity to some of the adaptive mitochondrial responses normally elicited by exercise training, possibly influenced by exercise intensity or volume. Thus, our clinical data in humans provide strong translational value of the SLIRP- deficient fly and mouse models and point towards a conserved requirement of SLIRP for skeletal muscle health, and potentially the SLIRP- independent mechanisms that compensate during exercise training. + +<--- Page Split ---> + +## Discussion + +Our investigation led to six key findings that bridge substantial knowledge gaps concerning the role of mitochondrial posttranscriptional mechanisms on skeletal muscle biology and their contributions to exercise training adaptations. Our findings not only unravel a delicate interplay between the mt- mRNA- stabilizing protein SLIRP, and the integrity of skeletal muscle mitochondria structure and performance, but also underscore the potential of exercise training in promoting selective mitochondrial translation capacity to circumvent these mechanisms. + +First, lack of Slirp led to mitochondrial network disruption, mitochondrial fragmentation, and reduced respiratory capacity in skeletal muscle in mice. Second, SLIRP deficiency impaired muscle functionality, and reduced lifespan in flies. Third, SLIRP was identified as an exercise training- responsive downstream target of PGC1- \(\alpha\) . Fourth, SLIRP was needed for improving blood glucose regulation after exercise training in male mice. Fifth, exercise training restored the mitochondrial defects elicited by the lack of SLIRP on mitochondrial structure, respiration, and mtDNA- encoded OXPHOS, via mechanisms likely involving the upregulation of mitoribosome content and enhanced antioxidant defense. Finally, we established a translational foundation for our findings by substantiating the conservation of SLIRP's response to exercise training in four independent human cohorts integrating different exercise modalities, sex, health, and age conditions. + +Our first finding establishes SLIRP as an hitherto unrecognized player in regulating mitochondria content, structure, and respiration in skeletal muscle. Our results are partially consistent with reports in heart, liver, and kidney Slirp KO17, 18, 52 showing that SLIRP forms a complex with LRPPRC and is crucial for the maintenance of the mt- mRNA reservoir. Accordingly, our observation of mitochondria with abnormal cristae in Slirp KO muscle aligns with transmission electron micrographs of Lrpprc KO heart tissue19. However, despite the stark similarities, mitochondrial respiration was compromised in Slirp KO skeletal muscle, but not in heart or liver18, underscoring SLIRP's functional importance within skeletal muscle relative to other tissues. In alignment, previous research further showed that Slirp KO skeletal muscle fibers have reduced sarcoplasmic reticulum Ca2+ storage capacity68. Thus, our present findings underscore the importance of SLIRP across various tissue types, with emphasis on its + +<--- Page Split ---> + +importance to skeletal muscle mitochondrial morphology and respiration, and overall mitochondrial health. + +Our second finding that SLIRP depletion compromised climbing ability, impaired starvation tolerance, and reduced lifespan in flies, illustrates the detrimental functional consequences of lacking SLIRP in the muscle tissue. The effects of mitochondrial proteins on modulating life span are equivocal. Some report that the depletion of specific OXPHOS genes reduces life span, whereas others report no effect or an increase on life span in model organisms69- 71. The "Mitochondrial Threshold Effect Theory"72 posits that mild mitochondrial dysfunction allows normal physiology in model organisms until a threshold is reached. Beyond that, severe mitochondrial dysfunction can lead to premature aging, developmental arrest or even death73- 75. In support, heterozygous depletion of Mclk1, which encodes an enzyme necessary for biosynthesis of the electron carrier coenzyme Q of OXPHOS, even increased lifespan in mice, but homozygous depletion was embryonically lethal75. Our findings support an important role for SLIRP and mitochondrial function in maintaining normal lifespan because we observed reduced survival in muscle- specific SLIRP KD flies. + +Our third finding was that PGC- 1α1 regulated SLIRP protein, providing evidence for a new training- responsive PGC- 1α target in skeletal muscle. Interestingly, training- induced adaptations in mitochondrial proteins can occur independently of PGC- 1α1 as the main regulator76. In agreement with that, our fifth finding was that the training- induced improvements in mitochondrial function could bypass SLIRP deletion. These findings agree with other studies showing that exercise training can rescue mitochondrial dysfunction and elicit improvements in glucose homeostasis in mice and humans13, 48, 76, 77. Interestingly, SLIRP was required for the training- induced improvements in blood glucose regulation in trained male Slirp KO mice. While not explored in our study, this could be due to decreased intramyocellular insulin signalling78, reduced insulin- independent glucose uptake14, capillarization, and/or blood flow79. Female mice did not improve their glucose tolerance in response to exercise training. Sexual dimorphism in response to metabolic challenges, including exercise training, has been frequently reported in mouse models80, 81. Yet, in both sexes, SLIRP is largely + +<--- Page Split ---> + +dispensable for exercise training- induced mitochondrial and metabolic adaptations in skeletal muscle, even though SLIRP is crucial for mitochondrial transcript stability in the basal state. + +We were intrigued by the observation that exercise training could circumvent even \(60 - 80\%\) loss in mitochondrial- mRNA pools to upregulate mtDNA- encoded OXPHOS proteins. In other words, the beneficial effect of exercise training on mitochondrial oxidative phosphorylation and ATP provision is independent of a correction of mt- mRNA transcript levels and LRPPRC/SLIRP protein content. These findings give rise to the possibility that mt- mRNAs are produced in excess in vivo18. Excess mt- mRNA may allow for rapid engagement of mitochondrial protein synthesis in the event of sudden changes in energy demand. In response to exercise training, the mitoribosomal components, MRPL11 and 12S rRNA were upregulated, indicating increased capacity for mitochondrial protein synthesis. These adaptations to exercise training may help sustain protein translation, even in the presence of low mt- mRNA abundance. Exercise training also upregulated PRDX3, linked to peroxide scavenging. While MRPL11, 12S rRNA and PRDX3 may not directly control translation efficiency, their roles in the mitoribosomal makeup and improved redox state can influence the overall cellular environment where protein translation occurs. Albeit limited by not directly measuring mitochondrial translational rate, this working hypothesis is supported by findings in human skeletal muscle showing that mitochondrial protein translation is at the core of exercise training- induced benefits39. + +Finally, our results likely are relevant to humans, as SLIRP and LRPPRC were consistently upregulated in human skeletal muscle in response to multiple exercise training modalities in both sexes. The effect of resistance exercise training was blunted for LRPPRC in patients with T2D. Interestingly, the age- related decline in muscle mitochondrial protein synthesis82 can be reversed by exercise training in elderly39, corroborating our findings that ET bypasses SLIRP to increase mitoribosomal translation. Importantly, our findings demonstrate that the synthesis of mtDNA- encoded OXPHOS proteins is not limited by transcript abundance in skeletal muscle – a feature that seems to be conserved as a fundamental mechanism26, 39. + +Taken together, our findings not only unravel a new mechanism for the regulation of basal mitochondrial function in skeletal muscle, but also highlight an incredible exercise training – stimulated + +<--- Page Split ---> + +plasticity of mitochondria in skeletal muscle facing mitochondrial defects. Our findings underscore ET as a therapeutic intervention to combat mitochondrial dysfunction in genetic and lifestyle- induced muscle pathologies, including T2D. + +## Conclusions + +In this work, we uncover a conserved role for mitochondrial transcript stability via SLIRP in basal mitochondrial structure and respiratory capacity in skeletal muscle. We identify SLIRP as a novel exercise- responsive downstream target of PGC- 1α that is conserved in human skeletal muscle. Because the defects in Slirp KO muscle could be reversed by ET, these findings add to the clinical relevance and add impetus to the important health- promoting role of ET in pathologies characterized by mitochondrial defects. + +## Figure Legends + +## Fig 1. Slirp knockout led to defects in mitochondrial structure and function, and reduced + +## lifespan + +(A) SLIRP protein content across different wild-type tissues (n=4, female C57BL/6J, 12 weeks of age) assessed by immunoblotting. + +(B) SLIRP and LRPPRC protein content in tibialis anterior muscle of Slirp knockout (KO) and littermate wildtype (WT) mice, and recombinant adeno-associated virus serotype 6 encoding SLIRP (rAAV6:SLIRP) and rAAV6:eGFP as control mice. + +(C) SLIRP and LRPPRC protein content in tibialis anterior muscle of recombinant adeno-associated virus serotype 6 encoding SLIRP (rAAV6:SLIRP) and rAAV6:eGFP in the contralateral leg as control. +(D) Confocal microscopy of mitochondrial network structure using a mitochondrial membrane potential probe (TMRE+) in flexor digitorum brevis muscle fibers of Slirp KO and WT mice (n=4-5, 7-10 fibers per mouse). Corresponding fragmentation index are presented as super plots83; small symbols represent each fiber, large symbols represent the mean of fibers from each mouse that are color- and symbol- + +<--- Page Split ---> + +505 coded for each sex (○ (blue) male, ◇ (red) female). + +506 (E) Transmission electron microscopy images of Slirp KO and WT gastrocnemius muscle. + +507 (F and G) Quantification of percentage of damaged intramyofibrillar (IMF) mitochondria within total mitochondrial and relative volume of mitochondria. Small symbols refer to damaged or total mitochondria per fiber, large symbols of identical symbol refer to average of fibers per biological replicate (female Slirp KO and WT gastrocnemius muscle, n=4/group). + +511 (H) Mitochondrial respiration in of Slirp KO and WT gastrocnemius muscle (n=4- 5/group, ○ (blue) male, ◇ (red) female). + +513 (I) Fatty acid oxidation in isolated Slirp KO and WT soleus muscle at rest and in response to contraction (n=3- 6/group, ○ (blue) male, ◇ (red) female). + +515 (J) RT- qPCR analysis of mitochondrial transcript levels in gastrocnemius of male Slirp KO and WT mice (n=6- 7/group). + +517 (K- M) Climbing time during climbing assay, starvation tolerance and life span of control (Mef2- GAL4>GD control 6000), SLIRP1 (Mef2- GAL4>SLIRP1 RNAi 51019 GD) and SLIRP2 (Mef2- GAL4>SLIRP2 RNAi 23675 GD) knockdown flies. + +520 Data are shown as mean ± SEM, including individual values, where applicable. + +521 Geno, main effect of genotype; substrate, main effect of substrate addition; Substrate X Geno, interaction between genotype and substrate; Contraction, main effect of contraction. \*p<0.05, \*\*p<0.01, \*\*\*p<0.001, as per Unpaired Student's t test (D, J), Mann Whitney test (F, G), Two- way RM ANOVA with Šidák's multiple comparisons test (H, I), ordinary one- way ANOVA with Dunnett's multiple comparisons test (K), Log- rank (Mantel- Cox) test (L, M). + +## Fig 2. SLIRP is upregulated by ET and is a target of PGC-1α + +527 (A) RT- qPCR analysis of Slirp transcript levels relative to Actb levels and Western blot analysis of SLIRP protein abundance in quadriceps with skeletal- muscle- specific transgenic expression of PGC- + +<--- Page Split ---> + +1α1 (α1 OE; n=8- 11/group) + +(B) Western blot analysis of SLIRP protein abundance in gastrocnemius with skeletal-muscle-specific transgenic expression of PGC-1α4 (α4 OE; n=6-9/group). + +(C) RT-qPCR analysis of Slirp and Ppargc1α transcript levels relative to Hprt levels in gastrocnemius of muscle-specific PGC-1α knockout (PGC-1α mKO) and littermate control (WT) mice at rest and 3h after acute exercise bout (n=5-8/group). + +(D) RT-qPCR analysis of Slirp and Ppargc1α transcript levels relative to Actb levels in WT quadriceps at rest, immediately after, 2 h, 6 h or 24 h after acute exercise bout (n=8/group). + +(E) Western blot analysis of SLIRP and pAMPK T172 protein abundance in WT quadriceps at rest, immediately after, 2 h, 6 h or 24 h after acute exercise bout (n=8/group). + +(F) Western blot analysis of SLIRP protein abundance in quadriceps of sedentary (SED) and 12-week ET PGC-1α mKO and WT mice (n=9-10/group). + +(G) Western blot analysis of SLIRP protein abundance in quadriceps of muscle-specific AMPKα1 and -α2 double KO (mDKO) and WT mice treated with/without AICAR (n=4-7/group). + +Data are means ± SEM, including individual values. Geno, main effect of genotype; Ex, main effect of acute exercise; Geno x Ex, interaction between genotype and acute exercise. \*p<0.05, \*\*\*p<0.001, as per Unpaired Student's t test (A), Two-way ANOVA with Šidák's multiple comparisons test (C, F, G), and ordinary one-way ANOVA with Dunnett's multiple comparisons test (D, E). + +## Fig. 3. SLIRP is dispensable for organismal adaptations to ET, yet, needed for improving blood glucose regulation after ET in male mice + +(A) Experimental design of 10-week exercise training (ET) intervention. +(B) Average running distance of male and female ET Slirp KO or littermate control (WT) mice measured for 6 weeks (n=6-11/group, ○ (blue) male, ◇ (red) female) + +(C) Average food intake of male and female sedentary (SED) and ET Slirp KO or WT mice measured + +<--- Page Split ---> + +for 2 weeks after 4 weeks of running (n=3- 8, ○ (blue) male, ◇ (red) female). + +(D- K) Blood glucose levels following a 4- hour fasting period before the glucose tolerance test (GTT). Glucose tolerance of male and female SED and ET Slirp KO or WT mice after 7- weeks of ET. iAUC of glycemic excursion in response to bolus of glucose \(2\mathrm{gkg}^{- 1}\) body weight (BW). Insulin response before (0 min) and following (20 min) the oral glucose challenge (n=6- 11/group, ○ (blue) male, ◇ (red) female). + +Data are means ± SEM, including individual values where applicable. + +Geno, main effect of genotype; ET, main effect of exercise training; Acute exercise, main effect of acute exercise but; Time X Geno ET, interaction between glucose bolus and genotype in the ET groups. \(^{*}\mathrm{p}< 0.05\) , \(^{**} \mathrm{p}< 0.01\) ; (E) WT ET vs. KO ET, \(^{*}\mathrm{p}< 0.05\) ; WT SED vs. WT ET, \(^{\#}\mathrm{p}< 0.05\) , as per Unpaired Student's t test (B), Two- way RM ANOVA with Sidak's multiple comparisons test in ET groups (E, I), Two- way ANOVA (C, D, F, G, H, J, K). + +## Fig 4. Exercise training reverses SLIRP-induced defects in muscle mitochondrial structure and respiration despite sustained reductions of mitochondrial transcripts + +(A) Confocal microscopy of mitochondrial network structure using a mitochondrial membrane potential probe (TMRE+) in flexor digitorum brevis muscle fibers of sedentary (SED) and 10-week ET Slirp knockout (KO) mice and littermate controls (WT) (n=3-4, 7-10 fibers per mouse), and corresponding fragmentation index. SED data is also depicted in Fig. 1D. + +(B) Mitochondrial respiration measured by Oroboros respirometry system in gastrocnemius of SED and ET Slirp KO and WT (n=2-6/per group, ○ male, ◇ female). SED data is also depicted in Fig. 1H. + +(C) Western blot analysis of SLIRP and LRPPRC protein abundance in gastrocnemius of SED and 10-week ET Slirp KO and WT mice (n=6-11/group, ○ (blue) male, ◇ (red) female). + +(D) Schematic of mtDNA- and nuclear DNA-encoded oxidative phosphorylation (OXPHOS) proteins analysed. + +(E) RT-qPCR analysis of mitochondrial transcript levels in gastrocnemius of male SED and ET Slirp + +<--- Page Split ---> + +578 KO and WT mice (n=6- 11/group). SED data is also depicted in Fig. 1J. + +(F-H) Western blot analysis of (F) mtDNA-encoded proteins, MT-ND1 (CI), MT-ND3 (CI), MT-CO1 (CIV), MT-ATP6 (CV), and (G) nDNA-encoded proteins NDUFB8 (CI), SDHB (CII), UQCRC2 (CIII) and ATP5A (CV), and (H) representative blots in gastrocnemius of SED and ET Slirp KO and WT mice (n=6-11/group, ○ (blue) male, ◇ (red) female). + +Data are means ± SEM, including individual values. Geno, main effect of genotype; ET, main effect of exercise training; Geno x ET, interaction between genotype and exercise training; Substrate, main effect of substrate. *p<0.05, **p<0.01, ***p<0.01 as per Two-way ANOVA with Šidák's multiple comparisons test (A, B, C, F, G). + +## Fig 5. Exercise training increases mitochondrial protein translation capacity by elevating + +### mitoribosome mass + +(A-H) RT-qPCR analysis of 12S rRNA (A) and Western blot analysis of MRPL11 (B), rpS6 (C), PRDX3 (E), PRDX3 dimer/monomer ratio (F), PRDX2 (G), and representative blots in gastrocnemius of sedentary (SED) and exercise trained (ET) Slirp knockout (KO) and control littermate (WT) mice (n=6-11/group, ○ (blue) male, ◇ (red) female). Data are means ± SEM, including individual values. Geno, main effect of genotype; ET, main effect of exercise training; *p<0.05, **p<0.01, as per Two-way ANOVA (A-C, E-G). + +## Fig 6. In human skeletal muscle SLIRP and LRPPRC protein content are increased by exercise training + +(A) Western blot analysis of SLIRP and LRPPRC protein in vastus lateralis of healthy young women (n=9/group, ○ Pre ET, □ Post ET)) after a 14-week controlled aerobic and strength exercise training (ET) intervention60. + +(B) Immunoblotting of SLIRP and LRPPRC protein in vastus lateralis of healthy young men (n=6/group, ○ Pre ET, □ Post ET)) after a 6-week high intensity interval training63. + +(C, D) Immunoblotting of SLIRP and LRPPRC protein (C) and RT-qPCR analysis of mitochondrial + +<--- Page Split ---> + +transcript levels (D) in vastus lateralis of male young and older individuals after 12- week progressive resistance training \(^{61}\) (n=6- 11/group, (○ Pre ET, □ Post ET)). + +(E- G) Immunoblotting of SLIRP, LRPPRC (E), and MT- CO2 (G) protein and RT- qPCR analysis of mitochondrial transcript levels (F) in vastus lateralis of male individuals, who were lean, obese or type 2 diabetic after high- intensity interval training \(^{62}\) (n=13- 16/group○ pre ET, □ post ET). Con, control; OB, obese; T2D, Type 2 diabetes. + +Data presented as individual before- and- after values. ET, main effect of exercise training, ET x T2D, Interaction between exercise training and presence of type 2 diabetes in Con and T2D group; \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per paired Student's t test (A- C), Mixed- effects model (D, F), Two- way RM ANOVA with Šidák's multiple comparisons test (E, G). + +## Supplementary Fig. 1 + +(A) Quantification of SLIRP protein content across different WT tissues (WT samples of n=4, 12 weeks of age) assessed by immunoblotting. + +(B- H) Transmission electron microscopy quantification of the following parameters of mitochondrial area in the intermyofibrillar (IMF) region (B), aspect ratio (C), elongation (D), convexity (E), perimeter (F), sphericity (G), and diameter (H). Small circles refer to each fiber, large squares in similar color refer to average of fibers per biological replicate, (female, n=4). + +(I- M) Western blot analysis of SDHB (CII), UQCRC2 (CIII), MT- CO1 (CIV), and ATP5A (CV) protein abundance in quadriceps of sedentary (SED) and 12- week exercise- training (ET) PGC- 1α mKO and control littermates (WT) mice (subset of total cohort, male, n=3). Representative blots. Coomassie was used as loading control. + +Data are means ± SEM, including individual values. ET, main effect of exercise training; Geno, main effect of genotype. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per Unpaired Student's t test (B- D, F- H), Mann Whitney test (E), Two- way ANOVA (I- L). + +## Supplementary Fig. 2 + +<--- Page Split ---> + +(A-C) Body weight, fat mass relative to body weight, lean mass relative to body weight of male and female SED and ET Slirp KO or WT mice after 6-weeks of ET (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +(D) Gastrocnemius, tibialis anterior (TA) and heart weight relative to tibia length of female SED and ET Slirp KO or WT mice after 10-weeks of ET (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +(E-H) Exercise capacity of male and female SED and ET Slirp KO or WT mice after 8-weeks of ET and corresponding blood lactate levels before and after exercise bout (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +(I) Fatty acid oxidation in isolated soleus of male and female SED and ET Slirp KO or WT mice after 10-weeks of ET. WT SED male/female, n=3/4, WT ET male/female, n=4/3, KO SED male/female, n=4/9, KO ET male/female, n=4/7. SED data is also depicted in Fig. 1H. + +(J-L) RT-qPCR analysis of mtDNA levels and citrate synthase activity in gastrocnemius of SED and ET Slirp KO and WT mice (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +Data are means \(\pm\) SEM, including individual values where applicable. ET, main effect of exercise training; Geno, main effect of genotype; Acute exercise, main effect of acute exercise; Geno in ET group, main effect of genotype within exercise training group; Contraction, main effect of contraction. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per Two-way ANOVA within sex (A-D), Two-way ANOVA (E, G, J-L), Two-way RM ANOVA with Sidak's multiple comparisons test (F, H, I). + +## Supplementary Fig. 3 + +(A) Western blot analysis of NDUFB8 (CI), SDHB (CII), UQCRC2 (CIII), ATP5A (CV) and ATP5B (CV) protein in vastus lateralis of male individuals, who were lean (pre/post, n=16/16), obese (pre/post, n=15/15) or type 2 diabetic (pre/post n=13/13) after high-intensity interval training62, and representative images. Coomassie was used as control. Data presented as individual before-and-after values. ET, main effect of exercise training. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , Two-way RM ANOVA (A). + +<--- Page Split ---> + +## Fly maintenance + +In standard conditions, the flies were maintained on a SYA medium containing \(0.5\%\) (w/v) agar, \(2.4\%\) (v/v) nipagin, \(0.7\%\) (v/v) propionic acid, \(10\%\) (w/v) dry baker's yeast and \(5\%\) (w/v) sucrose. The experiments took place at \(+25^{\circ}C\) , \(65\%\) humidity under a 12h:12h light:dark cycle. RNAi lines for SLIRP1 (51019 GD), SLIRP2 (23675 GD), and the GD library control line (w1118, 60000 GD) were obtained from Vienna Drosophila Resource Center. Mef2- GAL4 line was obtained from Bloomington Stock Center to generate muscle- specific SLIRP KD flies. + +## Negative geotaxis (climbing) assay + +Drosophila has a natural tendency to climb upwards, also known as negative geotaxis. The negative geotaxis, or climbing, assay was performed as previously described84. Briefly, 20 male Drosophila flies aged 5- 6 days were transferred into empty vials (25 x 95mm), and another empty vial was taped on top to create a \(25 \times 190\mathrm{mm}\) cylinder. Six cylinders were set side by side into an apparatus. The apparatus was tapped down 5 times to make all the flies fall to the bottom of the cylinder. The climbing assay was measured in technical triplicate for each biological replicate, with 90 seconds allowed for each climb. The assay was recorded with a camera and videos were analysed to determine the time for half of the flies in the vial to reach halfway point (95mm) in the cylinder. + +## Starvation assay using the Drosophila Activity Monitoring System (DAMS) + +Four- day old male flies were individually housed in monitor tubes with an outside diameter of 5mm (PPT5x65, Trikinetics, Waltham, MA). For starvation analysis, the tubes contained \(1\%\) agar in water. 32 tubes per monitor (DAM2 Drosophila Activity Monitor, Trikinetics, Waltham, MA) were set up, one monitor per genotype. For the starvation analysis, flies were kept in the DAM system until the last fly had died. + +## Lifespan assay + +<--- Page Split ---> + +For lifespan assays, flies were mated at a controlled density of 25 female and 10 male flies, respectively. One- day old male flies were collected into vials (10 male flies per vial, with 10 vials per genotype). The flies were kept on a SYA diet in Drosflippers (http://www.drosflipper.com/). Flies were flipped onto fresh food every second day and deaths were scored during the transfer. + +## Mice + +## PGC-1α1 transgenic mouse (PGC-1α1 OE) + +The generation of the model, quality, and specificity of the overexpression has been described previously35, of which we analyzed quadriceps muscle samples. + +## PGC-1α4 transgenic mouse (PGC-1α4 OE) + +The generation of the model, quality, and specificity of the overexpression has been described previously33, of which we analyzed gastrocnemius muscle samples. + +## Skeletal muscle-specific PGC-1α knockout (PGC-1α mKO) + +The generation of the model, quality, and specificity of the KO has been described previously44. Male muscle- specific PGC- 1α MKO mice and littermate control mice homozygous for loxP inserts (lox/lox) were generated by crossbreeding myogenin- Cre mice with loxP flanked- Pgc- 1α mice. + +For the acute exercise bout, Tibialis anterior muscles of PGC- 1α MKO mice and littermate control mice were collected 3 hours after one bout of equal distance treadmill running (1.4 km) at \(10^{\circ}\) incline and \(60\%\) of their individual maximal running speed achieved by a graded treadmill running test85. + +For the ET intervention, mice were single- housed with or without access to in- cage running wheels from 8 to 20 weeks old. Running distance and duration were monitored by a regular cycle computer. PGC- 1α MKO mice tended to run less than lox/lox, thus running wheels of lox/lox mice were occasionally blocked to ensure equal running distance. On average, mice ran 25 km/week. The running wheels were blocked for all mice 24 hours prior to euthanization and harvest of quadriceps muscle in the morning in the fed state. + +<--- Page Split ---> + +## AAV6 Vector construction and preparation + +The AAV6 vector for SLIRP overexpression and ultra- purified eGFP control AAV6 virus were manufactured by VectorBuilder Inc. (Shenandoah, Texas, USA; AAV6SP(VB190219- 1010jvy)- C). + +## Transfection of tibialis anterior muscle using rAAV6 vector + +Viral particles were diluted in Gelofusine (B. Braun, Germany) to a dosage of \(5 \times 10^{9}\) vector genomes at a volume of \(30 \mathrm{uL}\) per injection. For muscle- specific delivery of rAAV6 vectors, 12- week- old mice were placed under general anaesthesia (2% isoflourane in \(\mathrm{O_2}\) ) and mice were injected intramuscularly with rAAV6:SLIRP in the TA muscle of one leg and control rAAV6:eGFP (empty vector) in the contralateral leg. Muscles were harvested 4 weeks after rAAV6 administration. + +## Time course of acute exercise + +Twelve- week old female C57BL/6J mice (n=8 for each time point) were subjected to an acute exercise bout (1h running, 60% of maximal running intensity, 15° incline). Quadriceps muscle was harvested from rested and exercised mice immediately after, 2h, 6h, and 24 h after the acute exercise bout. + +## SLIRP knockout mouse (Slirp KO mice) + +The generation of the model, quality, and specificity of the KO has been described previously18. Sperm of homozygous Slirp KO mice kindly provided by Nils- Göran Larsson and re- derived offspring was backcrossed to C57BL/6N background in our own animal facilities. + +All mice were maintained under a 12:12 h light/dark photocycle at \(22 \pm 2^{\circ}\mathrm{C}\) with nesting material. The female mice were group- housed (except during the ET intervention), whereas the male mice were single- housed. All mice received a rodent chow diet (Altromin no. 1324; Chr. Pedersen, Denmark) and water ad libitum. + +Mouse genotyping of Slirp KO and WT mice was performed as previously described86 using qPCR on DNA from ear punches with the following primers: WT; 5'- AGAAGGGAAT CCACAGGATA GGACA- 3' and 5'- GCTTTATTCC TAGTCTGGC CTTGTT- 3', KO; 5'- AGAAGGGAAT CCACAGGATA GGACA- 3' and 5'- CGCCGTATAA TGTATGCTAT ACGAAGTT- 3'. + +<--- Page Split ---> + +## 10-week ET intervention in Slirp KO mice + +For 10- week voluntary wheel- running exercise training interventions, 16- 18- week- old Slirp KO and littermate mice were randomized into test groups with or without access to in- cage running wheels (Tecniplast activity cage, wheel diameter: \(23\mathrm{cm}\) ; Tecniplast, Buguggiate VA, Italy). The experimental design is schematically illustrated in Fig. 3A. Running distance was monitored by a regular cycle computer prior to the metabolic tests for 6 weeks. Running wheels were locked \(12\mathrm{h}\) prior to terminal procedures to avoid effects of acute exercise bouts. + +## Body composition + +Total, fat, and lean body mass was measured by nuclear magnetic resonance using an EchoMRITM (USA). + +## Glucose tolerance test (GTT) and plasma insulin analysis + +We subjected the Slirp KO and WT mice to a GTT at 7 weeks of the training intervention study (as illustrated in Fig. 3A). The GTT was executed after a \(5\mathrm{h}\) fasting period (7:00 a.m.- 12:00 p.m.). Resting blood samples were taken from the tail \(30\mathrm{min}\) prior to the intraperitoneal injection of d- mono- glucose (2 g/kg body weight). Tail blood glucose was measured after 0, 20, 40, 60, and 90 min of injection. + +To measure glucose- stimulated plasma insulin concentration at time- point (min) 0 and 20, tail vein blood samples were collected in a capillary- tube (50 \(\mu \mathrm{l}\) ), centrifuged at \(14,200\mathrm{g}\) for \(5\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) , plasma collected and stored at \(- 80^{\circ}\mathrm{C}\) . Insulin concentration was determined in duplicates using the Ultra- Sensitive Mouse Insulin ELISA Kit (#80- INSTRU- E10; ALPCO Diagnostics) to the manufacturer's instructions. The incremental area under the curve (iAUC) from the basal blood glucose concentration was determined using the trapezoid rule. + +## Exercise capacity tests + +Slirp KO and WT mice were acclimatized to the treadmill on 3 consecutive days by running at a speed of \(0.16\mathrm{m / s}\) and incline of \(10^{\circ}\) for \(5\mathrm{min}\) the first day and \(10\mathrm{min}\) the following days. Prior to the running test and with a \(1\mathrm{h}\) delay, 2 blood samples were drawn from the tail before the test for pre- exercise blood + +<--- Page Split ---> + +glucose and blood lactate measurements. Afterwards, the mice ran at \(0.16\mathrm{m / s}\) for 5 minutes followed by a gradual increase in speed every minute with \(0.02\mathrm{m / s}\) until exhaustion. Once the mouse reached its maximal running capacity, 2 blood samples were immediately drawn from the tail for post- exercise blood glucose and blood lactate measurements. The test was stopped when the mouse failed to keep up with the treadmill despite motivational efforts by the researcher. + +## TMRE staining in live fibers + +Flexor digitorum brevis muscles (FDB) were dissected and then incubated for 2 hours in serum- free \(\alpha\) - MEM (22571- 020, Gibco) containing \(1.9\mathrm{mg / ml}\) collagenase type 1 from clostridium histolyticum (C0130, Sigma- Aldrich) and \(1\mathrm{mg / ml}\) bovine serum albumin (9048- 46- 8, Sigma- Aldrich) at \(37^{\circ}\mathrm{C}\) on a rotator. After collagenase treatment, muscles were incubated in \(\alpha\) - MEM containing \(10\%\) fetal bovine serum (26050- 70, Gibco) and subjected to mechanical dissociation using fire- polished Pasteur pipettes. Single muscle fibers were seeded in \(35\times 14\mathrm{mm}\) glass- bottom microwell dishes (P35G- 1.5- 14- C, MatTek Corporation) coated with \(4\mathrm{uL}\) of Engelbreth- Holm- Swarm murine sarcoma ECM gel (E1270, Merck). Muscle fibers were kept in \(\alpha\) - MEM containing \(5\%\) fetal bovine serum in a cell incubator ( \(37^{\circ}\mathrm{C}\) , \(5\%\) CO2) for at least \(16\mathrm{h}\) prior to the experiments. + +For determination of mitochondrial membrane potential ( \(\Delta \Psi\) mitochondrial) and mitochondrial network morphology, the fibers were incubated in \(20\mathrm{mM}\) tetramethylrhodamine and ethyl ester (TMRE+, Life Technologies) dissolved in Krebs Ringer buffer ( \(145\mathrm{mMNaCl}\) , \(5\mathrm{mM}\) KCl, \(1\mathrm{mM}\) CaCl2, \(1\mathrm{mM}\) MgCl2, \(5.6\mathrm{mM}\) glucose, \(20\mathrm{mM}\) HEPES, pH 7.4) for 30 minutes before imaging. Confocal images were collected using a C- Apochromat \(\times 40\) , \(1.2\mathrm{NA}\) water immersion objective lens on an LSM 980 confocal microscope (Zeiss) driven Zeiss Zen Blue 3. + +Mitochondrial network analysis was performed semi- automatically in ImageJ (National Institute of Health, USA). At least two nucleus- free regions per fiber were analyzed blindly. Before segmentation, the background was subtracted, and the pixels were averaged using a value of mean=2. The images were segmented, and particle analyses revealed the relative area of the TMRE+ signal compared to the total area (% mitochondrial area). The fragmentation index was calculated as the number of objects in + +<--- Page Split ---> + +relation to the total area covered by the dye. The ImageJ 'Red Hot' lookup table was used to visualize the images. + +## Transmission electron microscopy analysis + +A small longitudinal section ( \(< 2\mathrm{mm}\) ) of the red portion of the gastrocnemius muscle tissue was fixed by immersion \(2\%\) glutaraldehyde in \(0.05\mathrm{M}\) Phosphatebuffer, pH 7.4 and stored at \(4^{\circ}\mathrm{C}\) . The samples were rinsed four times in \(0.1\mathrm{M}\) sodium cacodylate buffer, pH 7.4, and post- fixed with \(1\%\) osmium tetroxide \(\mathrm{(OsO_4)}\) and \(1.5\%\) potassium ferrocyanide \(\mathrm{[K_4Fe(CN)_6]}\) in \(0.1\mathrm{M}\) sodium cacodylate buffer, pH 7.4 for \(90\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) . The samples were then rinsed twice and dehydrated through a graded mixture of alcohol at \(4 - 20^{\circ}\mathrm{C}\) , infiltrated with graded mixtures of propylene oxide and Epon at \(20^{\circ}\mathrm{C}\) , and embedded in \(100\%\) Epon at \(30^{\circ}\mathrm{C}\) , as previously described (Nielsen et al., 2011). The samples were cut in \(60\mathrm{nm}\) longitudinal sections using a Leica Ultracut UCT ultramicrotome. The sections were contrasted with uranyl acetate and lead citrate, and subsequently examined and image recorded in a CM100 TEM (Philips, Eindhoven, The Netherlands) equipped with a Veleta camera and the iTem software package (Olympus, Hamburg, Germany) with a resolution of \(2048\times 2048\) pixels. + +A mean of eight fibers per sample (7- 9) was included from the sectioned samples, and from each fiber, 24 images were obtained at x13,500 magnification. The imaging was performed in a randomized systematic order including 12 images from the subsarcolemmal (SS) region, and 6 from both the superficial and central region of the intermyofibrillar (IMF) space. The Z- disc width was measured once on all IMF images and the mean from all fibers within one sample was calculated to determine the fiber type. The categorization of fiber types was based on previous reports from observations \(^{43}\) . The images were analyzed by two genotype- blinded investigators and all further analyses were performed by the same blinded investigator. The quantification of mitochondrial morphology was done using the Radius EM Imaging Software (Emsis GmbH, Radius 2.0). + +Several measurements were included in this study: 1) Mitochondrial number: Total number of mitochondrial profiles (count). 2) Mitochondrial volume fraction: IMF mitochondrial volume is + +<--- Page Split ---> + +annotated as percentage of mitochondria covering the IMF space. 3) Damaged mitochondria: Mitochondria were categorized as damaged when they were swollen, vacuolated, or empty. + +## Mitochondrial respiration in gastrocnemius muscle + +In a subset of mice, mitochondrial respiratory capacity was measured in permeabilized gastrocnemius skeletal muscle fibers as previously described87. In brief, gastrocnemius muscle was rinsed from fat and connective tissue and separated into small fiber bundles. Fiber bundles were permeabilized with saponin (50 \(\mu \mathrm{g} / \mathrm{ml}\) ) in BIOPS buffer for 30 min, followed by a 20 min wash in MiR05 buffer on ice. Mitochondrial respiration was measured in duplicate under hyperoxic conditions at \(37^{\circ}\mathrm{C}\) using high resolution respirometry (Oxygraph- 2k, Oroboros Instruments, Innsbruck, Austria). The following protocol was applied: Leak respiration was assessed by addition of malate (5 mM) and pyruvate (5 mM), followed by adding three different concentrations of ADP (0.01 mM, 0.25 mM and 5 mM; for the final concentration of ADP, 3 mM of magnesium (Mg) was added as well) to measure complex I linked respiratory capacity. 10 mM Glutamate was added to measure maximal complex I linked respiratory capacity followed by 10 mM succinate to measure complex I+II linked respiratory capacity. Finally, 5 \(\mu \mathrm{M}\) Antimycin A was added to inhibit complex III in the electron transport chain. Pooled data for both sexes are shown as no sex- specific differences were detected. + +## Contraction-stimulated fatty acid oxidation in isolated soleus muscles + +Contraction- stimulated exogenous palmitate oxidation in isolated soleus muscle from Slirp KO and WT mice was measured as previously described88. In brief, excised soleus muscles from mice anesthetized with pentobarbital were mounted at resting tension ( \(\sim 5 \mathrm{mN}\) ) in 15 ml vertical incubation chambers with a force transducer (Radnoti, Monrovia, CA) containing \(30^{\circ}\mathrm{C}\) carbonated (95% \(\mathrm{O}_2\) and 5% \(\mathrm{CO}_2\) ) Krebs- Henselieit Ringer buffer (KRB), \(\mathrm{pH} = 7.4\) , supplemented with 5 mM glucose, 2% fatty acid- free BSA, and 0.5 mM palmitate. After \(\sim 20 \mathrm{min}\) of pre- incubation, the incubation buffer was refreshed with KRB additionally containing [1- 14C]- palmitate (0.0044 MBq/ml; Amersham BioSciences, Buckinghamshire, U.K.). To seal the incubation chambers, mineral oil (Cat. No. M5904, Sigma- Aldrich) was added on top. Exogenous palmitate oxidation was measured simultaneously at rest and during 25 min contractions (18 trains/min, 0.6 s pulses, 30 Hz, 60 V). After incubation, incubation + +<--- Page Split ---> + +buffer and muscles were collected to determine the rate of palmitate oxidation as previously described87. 89, 90. Palmitate oxidation was determined as \(\mathrm{CO}_2\) production (complete FA oxidation) and acid-soluble metabolites (representing incomplete FA oxidation). As no difference was observed in complete and incomplete FA oxidation between genotypes, palmitate oxidation is presented as a sum of these two forms. + +## Lysate preparation and immunoblotting + +Lysate preparation and immunoblotting of vastus lateralis muscles samples of HIIT in young, healthy men were performed as described by Hostrup et al.91, whereas lysate preparation and immunoblotting of vastus lateralis muscles of HIIT in individuals, who were lean, obese or T2D were performed as described by Kruse et al92. + +To preserve and assess PRDX3 dimer to monomer ratio, freshly harvested mouse gastrocnemius muscle tissue was incubated for 10 min in ice- cold \(100\mathrm{mM}\) N- Ethylmaleimide (NEM) diluted in PBS. NEM was subsequently aspirated and the tissue homogenized for 1 min at \(30\mathrm{Hz}\) using a TissueLyser II bead mill (QIAGEN, USA) in ice- cold homogenization buffer [10% glycerol, 1% NP- 40, \(20\mathrm{mM}\) sodium pyrophosphate, \(150\mathrm{mM}\) NaCl, \(50\mathrm{mM}\) Hepes (pH 7.5), \(20\mathrm{mM}\) \(\beta\) - glycerophosphate, \(10\mathrm{mM}\) NaF, 2 mM phenylmethylsulfonyl fluoride, \(1\mathrm{mM}\) EDTA (pH 8.0), \(1\mathrm{mM}\) EGTA (pH 8.0), \(2\mathrm{mM}\) \(\mathrm{Na}_3\mathrm{VO}_4\) , leupeptin ( \(10\mu \mathrm{g}\mathrm{ml}^{- 1}\) ), aprotinin ( \(10\mu \mathrm{g}\mathrm{ml}^{- 1}\) ), and \(3\mathrm{mM}\) benzamidine]. Lysate preparation and immunoblotting of all remaining mouse tissues and human vastus lateralis muscle samples of aerobic and strength ET in women60, and resistance training in the young and elderly cohorts93 were performed as described in23. Immunoblotting of the NEM- treated samples was performed under non- denaturing conditions. The primary antibodies used are listed in Supplementary Table 2. + +Coomassie Brilliant Blue staining was used as a control to assess total protein loading and transfer efficiency94 by quantifying the whole lane and for each sample set, a representative membrane from the immunoblotting is shown. The same Coomassie brilliant blue staining is presented for proteins analyzed when derived from the same sample set. Band densitometry was carried out using Image Lab (version 4.0). For each set of samples, a standard curve was loaded to ensure quantification within the linear + +<--- Page Split ---> + +855 range for each protein probed for. For immunoblotting in Fig. 5, pooled data for both sexes are shown + +856 as no sex-specific differences were detected, unless specified. + +Supplementary Table 2: List of primary antibodies used during Western blotting analysis + +
ProteinManufacturerIdentifierDilution
p-AMPK T172Cell SignalingCat# 2531; RRID:AB_330330
Cat# A-21351;
RRID:AB_221512		1:1000,2%
Skim milk
1:2000,2%
Skim milk
1:1000,2%
Skim milk
ATP5F1BThermoFisherCat# 2867; RRID:AB_22329461:1000,2%
Skim milk
Hexokinase IICell SignalingCat# ab97505;
RRID:AB_10688419		1:1000,2%
Skim milk
1:1000,2%
LRPPRCAbcam1:1000,2%
Skim milk
MRPL11Cell SignalingCat# 2066; RRID:AB_21455981:1000,2%
MT-ATP6ThermoFisher
Scientific		Cat# PA5-109432;
RRID:AB_2854843
Cat# ab110413;
RRID:AB_2629281
Cat# ab181848;
RRID:AB_2687504
Cat# ab192306;
RRID:AB_2943449
Cat#ab109367;		1:1000,3% BSA
1:5000,3%
Skim milk
1:1000,2%
Skim milk
1:1000,2%
1:1000,2%
Skim milk
1:100,2%
Skim milk
MT-CO2Abcam1:1000,2%
MT-ND1AbcamSkim milk
MT-ND3Abcam1:1000,2%
PRDX2AbcamRRID:AB_10862524
Cat#ab73349;
RRID:AB_1860862		1:1000,2%
Skim milk
1:1000,2%
PRDX3Abcam1:1000,2%
rpS6Cell Signaling
Santa Cruz		Cat# 2217, RRID:AB_331355
Cat# sc-514508; RRID:
AB_2943449		Skim milk
1:500,2% Skim
SLIRP (mouse)BiotechnologyCat# 703632;milk
SLIRP (human)Thermo Fisher
Scientific		RRID:AB_2784591
Cat# ab110411;
RRID:AB_2756818		1:2000,2%
Skim milk
1:1000,2%
Total OXPHOS Human
WB Antibody Cocktail		AbcamSkim milk
Total OXPHOS Rodent
WB Antibody Cocktail		AbcamCat# ab110413;
RRID:AB_2629281		1:5000,2%
Skim milk
+ +857 + +858 + +# RNA extraction and RT-qPCR + +Fig. 1J, 4F: RNA was extracted from ~20 mg of pulverized whole mouse gastrocnemius muscle using + +TRIzolTM reagent (Invitrogen) following the manufacturer's instructions (tissue homogenization was + +performed using the MP Bio lysis system) and treated with the TURBO DNA-freeTM Kit (Ambion) to + +<--- Page Split ---> + +remove contaminating DNA. For RT- qPCR expression analysis, cDNA was reversed transcribed from \(0.85 \mu \mathrm{g}\) total RNA using the High- Capacity cDNA Reverse Transcription Kit (Applied Biosystems) in the presence of RNase Block (Agilent). The qPCR was performed in a QuantStudio 6 Flex Real- Time PCR System (Life Technologies), using TaqMan™ Universal Master Mix II, with UNG (Applied Biosystems) to quantify mitochondrial transcripts (mitochondrial- rRNAs and mt- mRNAs). The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method, comparing the Ct values of mitochondrial transcripts to that of the beta- actin reference gene for normalization. + +Fig. 2A, C, D: Total RNA was extracted from pulverized quadriceps muscle using an adapted guanidinium thiocyanate- phenol- chloroform extraction method. Reverse transcription to cDNA was performed as previously described46. Real- time qPCR was performed in triplicate with QuantStudio 7 Flex Real- Time PCR System (Applied Biosystems, Waltham, MA). Cycle threshold (Ct) was converted to a relative amount using a standard curve derived from a serial dilution of a representative pooled samples run together with the samples of interest. Beta- actin or Hprt mRNA was used for normalization of target mRNA levels. + +Fig. 6J- M: Total RNA extraction from vastus lateralis muscle and reverse transcription to cDNA was performed as previously described61. The qPCR was performed in a QuantStudio 6 Flex Real- Time PCR System (Life Technologies), using TaqMan™ Universal Master Mix II (Applied Biosystems) to quantify mt- mRNAs. The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method, comparing the Ct values of mitochondrial transcripts to that of the 18S rRNA reference gene for normalization. + +Fig. 6Q- T: Total RNA was extracted from skeletal muscle biopsies using TRI Reagent (Sigma- Aldrich) following the manufacturer's instructions. cDNA was reversed transcribed using the High- Capacity cDNA Reverse Transcription Kit (Applied Biosystems), while the qPCR was performed on an Aria Mx (Agilent) using a TaqMan™ Universal Master Mix II (Applied Biosystems). The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method with the mRNA levels being normalized to the geometric mean of PPIA and B2M. + +<--- Page Split ---> + +888 Primers or Taqman probes or Taqman expression assays used for mRNA levels measured in Fig. 2, + +889 Fig. 4 and Fig. 6 are shown in Supplementary Table 3. + +Supplementary Table 3: List of primers or Taqman probes used for RT-qPCR + +
Primers used for Fig. 2C
PGC-1a_Ex3-5_FAGCCGTGACCACTGACAACGAG
PGC-1a_Ex3-5_RGCTGCATGGTTCTGAGTGCTAAG
Slirp_FGTCGGCCTATTGCGTTTGTC
Slirp_RATGTTCTCTCAGCTCACTCGC
Hprt FAGTCCCAGCGTCTGGATTAG
Hprt RTTTCCAAATCCTCGGCATAATGA
+ +
Primers/ Taqman gene expression assays used for Fig. 2A, D; Fig. 4F
SlirpMm01296845_m1
Full length Ppargc1 FTCAAGCCAAACCAACAACTTTATCT
Full length Ppargc1 RGGTTCGCTCAATAGTCTTGTTCTCA
Ppargc1 probeCACCAAATGACCCCAAGGGTTCCC
12S rRNAAJBJWSP
18S rRNAMm03928990_g1
ActBMm01205647_g1
mt-Nd1Mm04225274_s1
mt-Nd6Mm04225325_g1
mt-CytBMm04225271_g1
mt-ColMm04225243_g1
mt-Atp6Mm03649417_g1
Taqman gene expression assays used for Fig. 6
MT-ND1Hs02596873_s1
MT-ND6Hs02596879_g1
MT-CYBHs02596867_s1
MT-CO1Hs02596864_g1
MT-ATP6Hs02596862_g1
PPIAHs99999904_m1
B2MHs99999907_m1
+ +# 890 DNA isolation and mtDNA quantification + +891 Total DNA was extracted from ~20 mg mouse gastrocnemius muscle using the Dneasy Blood and + +892 Tissue Kit (Qiagen) according to the manufacturer's instructions and treated with RNase A. Levels of + +893 mtDNA were measured by qPCR using 2.5 ng of DNA in a QuantStudio 6 Flex Real-Time PCR System + +<--- Page Split ---> + +using TaqMan™ Universal Master Mix II, with UNG. The mt- Nd1 and mt- Nd6/Nd5 TaqMan gene expression assays were used. The 18S probe was used for normalization. + +## Aerobic and strength ET intervention in healthy young women + +A detailed description of the study participants, research design and methods has been previously published61. In the present study, we included vastus lateralis muscle lysate for immunoblotting obtained from 9 healthy young women (age, \(33 \pm 6\) years; body mass index (BMI), \(23.2 \pm 2.6 \mathrm{kg} \mathrm{m}^{- 2}\) ) before and after 14- week of aerobic and strength exercise training. + +## High-intensity interval training intervention in healthy young men + +A detailed description of the study participants, research design and methods has been previously published63. In the present study, we included vastus lateralis muscle lysate for immunoblotting obtained from 6 healthy young men before and after 6- week HIIT training. + +## Progressive resistance exercise training in male young and older individuals + +A detailed description of the study participants, research design and methods has been previously published45. The original study reported no additive effect of vitamin D intake during the 12 weeks of resistance exercise training on muscle hypertrophy or muscle strength60. Accordingly, in the present study, samples from young or older participants were considered as one group, irrespective of vitamin D intake. We included vastus lateralis muscle samples ( \(\sim 10 \mathrm{mg}\) wet weight) for immunoblotting obtained from a subset of the original sample set due to lack of sample material. For the young participants, we included 10 samples before and 6 samples after resistance exercise training. For the older participants we included 11 before and 9 samples after resistance exercise training. + +## High-intensity interval training in middle-aged male individuals, who were lean, obese or obese with type 2 diabetes + +A detailed description of the study participants, research design and methods has been previously published62. In short, 15 middle- aged men with T2D and obesity, 15 age- matched glucose- tolerant men with obesity, and 18 age- matched glucose- tolerant lean men were recruited. All participants, except + +<--- Page Split ---> + +four (two men with T2D and obesity and two lean men) completed the study. The training protocol consisted of 8- weeks with three weekly training sessions consisting of supervised HIIT in combination with biking and rowing. HIIT- sessions consisted of \(5 \times 1\) min exercise blocks interspersed with 1 min rest, and shifted between blocks on cycle and rowing ergometers. The volume was increased from two to five blocks during the 8 weeks. In the present study, we included vastus lateralis muscle samples that were obtained 4- 5 days after the last HIIT session but 48 hours after the last physical activity (a \(\mathrm{VO}_2\) max test). The number of samples included for immunoblotting as follows: 16 before and after HIIT from lean glucose- tolerant men, 15 before and after HIIT from glucose- tolerant men with obesity, and 13 before and after HIIT from men with T2D and obesity. + +## Ethical committee approval + +Clinical experiments were approved by the Ethics Committee of Copenhagen or by the Regional Scientific Ethical Committees for Southern Denmark and performed in accordance with the Helsinki Declaration II. Studies including human study participants are described in Ref60- 63. All mouse experiments complied with the European Convention for the protection of vertebrate animals used for experimental and other scientific purposes (No. 123, Strasbourg, France, 1985; EU Directive 2010/63/EU for animal experiments) and were approved by the Danish Animal Experimental Inspectorate (License number: 2016- 15- 0201- 01043). + +## Graphical illustrations + +Graphical illustrations were created in ©BioRender - biorender.com, as indicated. + +## Statistical methods + +Statistical analyses were performed using GraphPad Prism 9 (version 9.3.1). Results are presented as mean ± SEM with individual values shown, when feasible. Statistical differences were analyzed by ordinary one- way ANOVA, repeated/ordinary two- way ANOVA, or Log- rank (Mantel- Cox) test, and Mann- Whitney test as applicable. Dunnett's multiple comparisons test or Šidák's multiple comparisons test were used to evaluate significant interactions in ANOVAs. As statistical tests varied according to + +<--- Page Split ---> + +the dataset being analyzed, the respective tests utilized are specified within the figure legends. \(0.05 \leq p < 0.1\) were considered a tendency and p values \(< 0.05\) were considered significant. + +## Acknowledgements + +We acknowledge the technical assistance of Betina Bolmgren and Irene Nielsen, Martin Thomassen, and Roberto Meneses- Valdes (Department of Nutrition, Exercise and Sports, Faculty of Science, University of Copenhagen, Denmark), Anja Jokipii- Utzon (Institute of Sports Medicine, Bispebjerg Hospital, Copenhagen, Denmark), and Michala Carlsson (Department of Biomedical Sciences, University of Copenhagen, Denmark). We thank Vivian Shang for her assistance with the Drosophila experiments (Charles Perkins Centre, The University of Sydney, Australia). We acknowledge Cristiano di Benedetto for preparing muscle samples for TEM analysis (Core Facility for Integrated Microscopy, Faculty of Health and Medical Sciences, University of Copenhagen, Denmark). + +## Funding + +The study was supported by the Novo Nordisk Foundation (grant NNF16OC0023418, NNF18OC0032082, and NNF20OC0063577 to L.S.; grant NNF22OC0074110 to A.M.F.), by Independent Research Fund Denmark to L.S. (#0169- 00013B), by the European Union's Horizon 2020 research and innovation programme (Marie Sklodowska- Curie grant agreement No 801199 to T.C.P.P. and E.A.R.), and grants by Danish Council for Independent Research - Medical Sciences (4181- 00078) and the Augustinus Foundation to H.P. + +## CRediT authorship contribution statement + +Tang Cam Phung Pham: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft, Visualization, Project administration, Funding acquisition; Steffen Henning Raun: Conceptualization; Investigation, Writing - Review & Editing; Essi Havula: Methodology, Investigation, Writing - Review & Editing; Carlos Henriquez- Olguin: Methodology, Investigation, Writing - Review & Editing; Diana Rubalcava- Gracia: Methodology, Investigation, Writing - Review & Editing; Emma Frank: Methodology, Investigation, Writing - Review & Editing; Andreas Mæchel Fritzen: Methodology, Investigation, Writing - Review & Editing; Paulo Jannig: Methodology, + +<--- Page Split ---> + +970 Investigation, Writing - Review & Editing; Nicoline Resen Andersen: Investigation, Writing - Review & Editing; Rikke Kruse: Investigation, Writing - Review & Editing; Mona Sadek Ali: Investigation, Writing - Review & Editing; Jens Frey Halling: Resources, Writing - Review & Editing; Stine Ringholm: Resources, Writing - Review & Editing; Solvejg Hansen: Resources, Writing - Review & Editing; Anders Krogh Lemminger: Resources, Writing - Review & Editing; Maria Houborg Petersen: Resources, Writing - Review & Editing; Martin Eisemann de Almedia: Resources, Writing - Review & Editing; Thomas E. Jensen: Resources, Writing - Review & Editing; Bente Kiens: Resources, Writing - Review & Editing; Morten Hostrup: Resources, Writing - Review & Editing; Steen Larsen: Methodology, Investigation, Writing - Review & Editing; Niels Ortenblad: Resources, Writing - Review & Editing; Kurt Højlund: Resources, Writing - Review & Editing; Peter Schjerling: Methodology, Investigation, Resources, Writing - Review & Editing; Michael Kjær: Resources, Writing - Review & Editing; Jorge Ruas: Resources, Writing - Review & Editing; Aleksandra Trifunovic: Resources, Writing - Review & Editing; Jørgen Wojtaszewski: Resources, Writing - Review & Editing; Joachim Nielsen: Methodology, Investigation, Writing - Review & Editing; Klaus Qvortrup: Resources, Writing - Review & Editing; Henriette Pilegaard: Resources, Writing - Review & Editing; Erik Arne Richter: Resources, Investigation, Writing - Review & Editing, Supervision, Funding acquisition; Lykke Sylow: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft, Project administration, Supervision, Funding acquisition. + +## References + +991 1. Romanello, V. & Sandri, M. The connection between the dynamic remodeling of the mitochondrial network and the regulation of muscle mass. Cell Mol Life Sci 78, 1305-1328 (2021). 992 2. Distefano, G. et al. Physical activity unveils the relationship between mitochondrial energetics, muscle quality, and physical function in older adults. J Cachexia Sarcopenia Muscle 9, 279-294 (2018). 993 3. Gonzalez- Freire, M. et al. Reconsidering the Role of Mitochondria in Aging. J Gerontol A Biol Sci Med Sci 70, 1334-1342 (2015). + +<--- Page Split ---> + +999 4. Janssen, I., Heymsfield, S.B. & Ross, R. Low relative skeletal muscle mass (sarcopenia) in 1000 older persons is associated with functional impairment and physical disability. J Am Geriatr 1001 Soc 50, 889- 896 (2002). 1002 5. Short, K.R. et al. Decline in skeletal muscle mitochondrial function with aging in humans. Proc Natl Acad Sci U S A 102, 5618- 5623 (2005). 1003 6. Haines, M.S., Leong, A., Porneala, B.C., Meigs, J.B. & Miller, K.K. Association between muscle mass and diabetes prevalence independent of body fat distribution in adults under 50 years 1006 old. Nutr Diabetes 12, 29 (2022). 1007 7. Sugimoto, K. et al. Hyperglycemia in non- obese patients with type 2 diabetes is associated with low muscle mass: The Multicenter Study for Clarifying Evidence for Sarcopenia in 1009 Patients with Diabetes Mellitus. J Diabetes Investig 10, 1471- 1479 (2019). 1010 8. Yoon, J.W. et al. Hyperglycemia is Associated with Impaired Muscle Quality in Older Men 1011 with Diabetes: The Korean Longitudinal Study on Health and Aging. Diabetes Metab J 40, 140- 146 (2016). 1013 9. Ebhardt, H.A. et al. Comprehensive proteome analysis of human skeletal muscle in cachexia 1014 and sarcopenia: a pilot study. J Cachexia Sarcopenia Muscle 8, 567- 582 (2017). 1015 10. Romanello, V. & Sandri, M. Mitochondrial Quality Control and Muscle Mass Maintenance. 1016 Front Physiol 6, 422 (2015). 1017 11. Safdar, A. et al. Endurance exercise rescues progeroid aging and induces systemic 1018 mitochondrial rejuvenation in mtDNA mutator mice. Proc Natl Acad Sci U S A 108, 4135- 4140 (2011). 1020 12. Pedersen, B.K. & Saltin, B. Exercise as medicine - evidence for prescribing exercise as therapy 1021 in 26 different chronic diseases. Scand J Med Sci Sports 25 Suppl 3, 1- 72 (2015). 1022 13. Adhihety, P.J., Taivassalo, T., Haller, R.G., Walkinshaw, D.R. & Hood, D.A. The effect of 1023 training on the expression of mitochondrial biogenesis- and apoptosis- related proteins in 1024 skeletal muscle of patients with mtDNA defects. Am J Physiol Endocrinol Metab 293, E672- 1025 680 (2007). 1026 14. Sylow, L., Kleinert, M., Richter, E.A. & Jensen, T.E. Exercise- stimulated glucose uptake - 1027 regulation and implications for glycaemic control. Nat Rev Endocrinol 13, 133- 148 (2017). 1028 15. Boengler, K., Heusch, G. & Schulz, R. Nuclear- encoded mitochondrial proteins and their role 1029 in cardioprotection. Biochim Biophys Acta 1813, 1286- 1294 (2011). 1030 16. Couvillion, M.T., Soto, I.C., Shipkovenska, G. & Churchman, L.S. Synchronized mitochondrial 1031 and cytosolic translation programs. Nature 533, 499- 503 (2016). 1032 17. Sasarman, F. et al. LRPpRC and SLIRP interact in a ribonucleoprotein complex that regulates 1033 posttranscriptional gene expression in mitochondria. Mol Biol Cell 21, 1315- 1323 (2010). 1034 18. Lagouge, M. et al. SLIRP Regulates the Rate of Mitochondrial Protein Synthesis and Protects 1035 LRPpRC from Degradation. PLoS Genet 11, e1005423 (2015). 1036 19. Ruzzenente, B. et al. LRPpRC is necessary for polyadenylation and coordination of translation 1037 of mitochondrial mRNAs. EMBO J 31, 443- 456 (2012). 1038 20. Chujo, T. et al. LRPpRC/SLIRP suppresses PNPase- mediated mRNA decay and promotes 1039 polyadenylation in human mitochondria. Nucleic Acids Res 40, 8033- 8047 (2012). 1040 21. Baughman, J.M. et al. A computational screen for regulators of oxidative phosphorylation 1041 implicates SLIRP in mitochondrial RNA homeostasis. PLoS Genet 5, e1000590 (2009). 1042 22. Holloszy, J.O. Biochemical adaptations in muscle. Effects of exercise on mitochondrial 1043 oxygen uptake and respiratory enzyme activity in skeletal muscle. J Biol Chem 242, 2278- 2282 (1967). 1045 23. Kleinert, M. et al. Quantitative proteomic characterization of cellular pathways associated 1046 with altered insulin sensitivity in skeletal muscle following high- fat diet feeding and exercise 1047 training. Sci Rep 8, 10723 (2018). 1048 24. Hatchell, E.C. et al. SLIRP, a small SRA binding protein, is a nuclear receptor corepressor. Mol 1049 Cell 22, 657- 668 (2006). + +<--- Page Split ---> + +1050 25. Brand, A.H. & Perrimon, N. Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development 118, 401- 415 (1993). 1051 26. Baggio, F. et al. Drosophila melanogaster LRPPRC2 is involved in coordination of mitochondrial translation. Nucleic Acids Res 42, 13920- 13938 (2014). 1053 27. Inagaki, H.K., Kamikouchi, A. & Ito, K. Methods for quantifying simple gravity sensing in Drosophila melanogaster. Nat Protoc 5, 20- 25 (2010). 1056 28. Baar, K. et al. Adaptations of skeletal muscle to exercise: rapid increase in the transcriptional coactivator PGC- 1. FASEB J 16, 1879- 1886 (2002). 1058 29. Chinsomboon, J. et al. The transcriptional coactivator PGC- 1alpha mediates exercise- induced angiogenesis in skeletal muscle. Proc Natl Acad Sci U S A 106, 21401- 21406 (2009). 1059 30. Pilegaard, H., Saltin, B. & Neufer, P.D. Exercise induces transient transcriptional activation of the PGC- 1alpha gene in human skeletal muscle. J Physiol 546, 851- 858 (2003). 1061 31. Kjobsted, R. et al. Intact Regulation of the AMPK Signaling Network in Response to Exercise and Insulin in Skeletal Muscle of Male Patients With Type 2 Diabetes: Illumination of AMPK Activation in Recovery From Exercise. Diabetes 65, 1219- 1230 (2016). 1064 32. Halling, J.F. & Pilegaard, H. PGC- 1alpha- mediated regulation of mitochondrial function and physiological implications. Appl Physiol Nutr Metab 45, 927- 936 (2020). 1067 33. Ruas, J.L. et al. A PGC- 1alpha isoform induced by resistance training regulates skeletal muscle hypertrophy. Cell 151, 1319- 1331 (2012). 1069 34. Jannig, P.R., Dumesic, P.A., Spiegelman, B.M. & Ruas, J.L. SnapShot: Regulation and biology of PGC- 1alpha. Cell 185, 1444- 1444 e1441 (2022). 1071 35. Lin, J. et al. Transcriptional co- activator PGC- 1 alpha drives the formation of slow- twitch muscle fibres. Nature 418, 797- 801 (2002). 1073 36. Arany, Z. PGC- 1 coactivators and skeletal muscle adaptations in health and disease. Curr Opin Genet Dev 18, 426- 434 (2008). 1075 37. Handschin, C. et al. Abnormal glucose homeostasis in skeletal muscle- specific PGC- 1alpha knockout mice reveals skeletal muscle- pancreatic beta cell crosstalk. J Clin Invest 117, 3463- 3474 (2007). 1078 38. Miller, B.F., Konopka, A.R. & Hamilton, K.L. The rigorous study of exercise adaptations: why mRNA might not be enough. J Appl Physiol (1985) 121, 594- 596 (2016). 1080 39. Robinson, M.M. et al. Enhanced Protein Translation Underlies Improved Metabolic and Physical Adaptations to Different Exercise Training Modes in Young and Old Humans. Cell Metab 25, 581- 592 (2017). 1083 40. Liu, Y., Beyer, A. & Aebersold, R. On the Dependency of Cellular Protein Levels on mRNA Abundance. Cell 165, 535- 550 (2016). 1085 41. Richter, E.A. & Ruderman, N.B. AMPK and the biochemistry of exercise: implications for human health and disease. Biochem J 418, 261- 275 (2009). 1087 42. Fentz, J. et al. AMPKalpha is critical for enhancing skeletal muscle fatty acid utilization during in vivo exercise in mice. FASEB J 29, 1725- 1738 (2015). 1089 43. Lantier, L. et al. AMPK controls exercise endurance, mitochondrial oxidative capacity, and skeletal muscle integrity. FASEB J 28, 3211- 3224 (2014). 1091 44. Handschin, C. et al. Skeletal muscle fiber- type switching, exercise intolerance, and myopathy in PGC- 1alpha muscle- specific knock- out animals. J Biol Chem 282, 30014- 30021 (2007). 1093 45. Raun, S.H. et al. Housing temperature influences exercise training adaptations in mice. Nat Commun 11, 1560 (2020). 1095 46. Brandt, N. et al. The impact of exercise training and resveratrol supplementation on gut microbiota composition in high- fat diet feed mice. Physiol Rep 6, e13881 (2018). 1097 47. Wojtaszewski, J.F. & Richter, E.A. Effects of acute exercise and training on insulin action and sensitivity: focus on molecular mechanisms in muscle. Essays Biochem 42, 31- 46 (2006). 1099 48. Memme, J.M. & Hood, D.A. Molecular Basis for the Therapeutic Effects of Exercise on Mitochondrial Defects. Front Physiol 11, 615038 (2020). + +<--- Page Split ---> + +1101 49. Carter, H.N., Pauly, M., Tryon, L.D. & Hood, D.A. Effect of contractile activity on PGC-1alpha transcription in young and aged skeletal muscle. J Appl Physiol (1985) 124, 1605-1615 (2018).1104 50. Lanza, I.R. et al. Endurance exercise as a countermeasure for aging. Diabetes 57, 2933-2942 (2008).1106 51. Schaefer, P.M. et al. Mitochondrial mutations alter endurance exercise response and determinants in mice. Proc Natl Acad Sci U S A 119, e2200549119 (2022).1108 52. Spahr, H. et al. SLIRP stabilizes LRPPRC via an RRM-PPR protein interface. Nucleic Acids Res 44, 6868-6882 (2016).1110 53. Filograna, R., Mennuni, M., Alsina, D. & Larsson, N.G. Mitochondrial DNA copy number in human disease: the more the better? FEBS Lett 595, 976-1002 (2021).1112 54. Larsen, S. et al. Biomarkers of mitochondrial content in skeletal muscle of healthy young human subjects. J Physiol 590, 3349-3360 (2012).1114 55. Halling, J.F., Ringholm, S., Olesen, J., Prats, C. & Pilegaard, H. Exercise training protects against aging-induced mitochondrial fragmentation in mouse skeletal muscle in a PGC-11alpha dependent manner. Exp Gerontol 96, 1-6 (2017).1117 56. Metodiev, M.D. et al. Methylation of 125 rRNA is necessary for in vivo stability of the small subunit of the mammalian mitochondrial ribosome. Cell Metab 9, 386-397 (2009).1119 57. Yang, Y. et al. NAD+-dependent deacetylase SIRT3 regulates mitochondrial protein synthesis by deacetylation of the ribosomal protein MRPL10. J Biol Chem 285, 7417-7429 (2010).1121 58. Cox, A.G., Winterbourn, C.C. & Hampton, M.B. Mitochondrial peroxiredoxin involvement in antioxidant defence and redox signalling. Biochem J 425, 313-325 (2009).1123 59. Lee, K.P. et al. Peroxiredoxin 3 has a crucial role in the contractile function of skeletal muscle by regulating mitochondrial homeostasis. Free Radic Biol Med 77, 298-306 (2014).1125 60. Hansen, S.L. et al. Mechanisms Underlying Absent Training-Induced Improvement in Insulin Action in Lean, Hyperandrogenic Women With Polycystic Ovary Syndrome. Diabetes 69, 2267-2280 (2020).1127 61. Agergaard, J. et al. Does vitamin-D intake during resistance training improve the skeletal muscle hypertrophic and strength response in young and elderly men? - a randomized controlled trial. Nutr Metab (Lond) 12, 32 (2015).1121 62. Petersen, M.H. et al. High-intensity interval training combining rowing and cycling efficiently improves insulin sensitivity, body composition and VO(2)max in men with obesity and type 2 diabetes. Front Endocrinol (Lausanne) 13, 1032235 (2022).1134 63. Hostrup, M. et al. High-Intensity Training Represses FXYD5 and Glycosylates Na,K-ATPase in Type II Muscle Fibres, Which Are Linked with Improved Muscle K(+) Handling and Performance. Int J Mol Sci 24 (2023).1137 64. Hey-Mogensen, M. et al. Effect of physical training on mitochondrial respiration and reactive oxygen species release in skeletal muscle in patients with obesity and type 2 diabetes. Diabetologia 53, 1976-1985 (2010).1140 65. Phielix, E., Meeex, R., Moonen-Kornips, E., Hesselink, M.K. & Schrauwen, P. Exercise training increases mitochondrial content and ex vivo mitochondrial function similarly in patients with type 2 diabetes and in control individuals. Diabetologia 53, 1714-1721 (2010).1143 66. Karakelides, H., Irving, B.A., Short, K.R., O'Brien, P. & Nair, K.S. Age, obesity, and sex effects on insulin sensitivity and skeletal muscle mitochondrial function. Diabetes 59, 89-97 (2010).1145 67. Schwanhausser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337-342 (2011).1147 68. Gineste, C. et al. Enzymatically dissociated muscle fibers display rapid dedifferentiation and impaired mitochondrial calcium control. iScience 25, 105654 (2022).1149 69. Copeland, J.M. et al. Extension of Drosophila life span by RNAi of the mitochondrial respiratory chain. Curr Biol 19, 1591-1598 (2009). + +<--- Page Split ---> + +1151 70. Dell'agnello, C. et al. Increased longevity and refractoriness to Ca(2+) - dependent neurodegeneration in Surf1 knockout mice. Hum Mol Genet 16, 431- 444 (2007).1153 71. Dillin, A. et al. Rates of behavior and aging specified by mitochondrial function during development. Science 298, 2398- 2401 (2002).1155 72. Rossignol, R. et al. Mitochondrial threshold effects. Biochem J 370, 751- 762 (2003).1156 73. Hwang, A.B., Jeong, D.E. & Lee, S.J. Mitochondria and organismal longevity. Curr Genomics 113, 519- 532 (2012).1158 74. Xu, C. et al. Genetic inhibition of an ATP synthase subunit extends lifespan in C. elegans. Sci Rep 8, 14836 (2018).1159 75. Liu, X. et al. Evolutionary conservation of the clk- 1- dependent mechanism of longevity: loss of mclk1 increases cellular fitness and lifespan in mice. Genes Dev 19, 2424- 2434 (2005).1162 76. Leick, L. et al. PGC- 1alpha is not mandatory for exercise- and training- induced adaptive gene responses in mouse skeletal muscle. Am J Physiol Endocrinol Metab 294, E463- 474 (2008).1164 77. Larsen, S., Skaaby, S., Helge, J.W. & Dela, F. Effects of exercise training on mitochondrial function in patients with type 2 diabetes. World J Diabetes 5, 482- 492 (2014).1166 78. Sylow, L., Tokarz, V.L., Richter, E.A. & Klip, A. The many actions of insulin in skeletal muscle, the paramount tissue determining glycemia. Cell Metab 33, 758- 780 (2021).1168 79. Richter, E.A., Sylow, L. & Hargreaves, M. Interactions between insulin and exercise. Biochem J 478, 3827- 3846 (2021).1170 80. Cartee, G.D. Sexual Dimorphic Effects of Exercise Training on Subcutaneous White Adipose Tissue of Mice. Diabetes 70, 1242- 1243 (2021).1171 81. Holcomb, L.E., Rowe, P., O'Neill, C.C., DeWitt, E.A. & Kolwicz, S.C., Jr. Sex differences in endurance exercise capacity and skeletal muscle lipid metabolism in mice. Physiol Rep 10, e15174 (2022).1175 82. Rooyackers, O.E., Adey, D.B., Ades, P.A. & Nair, K.S. Effect of age on in vivo rates of mitochondrial protein synthesis in human skeletal muscle. Proc Natl Acad Sci U S A 93, 15364- 15369 (1996).1178 83. Lord, S.J., Velle, K.B., Mullins, R.D. & Fritz- Laylin, L.K. SuperPlots: Communicating reproducibility and variability in cell biology. J Cell Biol 219 (2020).1180 84. Ganetzky, B. & Flanagan, J.R. On the relationship between senescence and age- related changes in two wild- type strains of Drosophila melanogaster. Exp Gerontol 13, 189- 196 (1978).1183 85. Correia, J.C. et al. Muscle- secreted neururin couples myofiber oxidative metabolism and slow motor neuron identity. Cell Metab 33, 2215- 2230 e2218 (2021).1185 86. Geng, T. et al. PGC- 1alpha plays a functional role in exercise- induced mitochondrial biogenesis and angiogenesis but not fiber- type transformation in mouse skeletal muscle. Am J Physiol Cell Physiol 298, C572- 579 (2010).1187 87. Jeppesen, J. et al. LKB1 regulates lipid oxidation during exercise independently of AMPK. Diabetes 62, 1490- 1499 (2013).1189 88. Eisenberg, B.R. Quantitative ultrastructure of mammalian skeletal muscle, in Handbook of Physiology, Skeletal Muscle. (eds. L.D. Peachey, R.H. Adrian & S.R. Geiger) p. 73- 112 (American Physiological Society, Bethesda, MD; 1983).1193 89. Pesta, D. & Gnaiger, E. High- resolution respirometry: OXPHOS protocols for human cells and permeabilized fibers from small biopsies of human muscle. Methods Mol Biol 810, 25- 58 (2012).1196 90. Dzamko, N. et al. AMPK- independent pathways regulate skeletal muscle fatty acid oxidation. J Physiol 586, 5819- 5831 (2008).1198 91. Hostrup, M. et al. High- intensity interval training remodels the proteome and acetylome of human skeletal muscle. Elife 11 (2022).1200 92. Kruse, R. et al. Intact initiation of autophagy and mitochondrial fission by acute exercise in skeletal muscle of patients with Type 2 diabetes. Clin Sci (Lond) 131, 37- 47 (2017). + +<--- Page Split ---> + +1202 93. Kruse, R. et al. Effect of long-term testosterone therapy on molecular regulators of skeletal muscle mass and fibre-type distribution in aging men with subnormal testosterone. 1203 Metabolism 112, 154347 (2020). 1204 94. Welinder, C. & Ekblad, L. Coomassie staining as loading control in Western blot analysis. J Proteome Res 10, 1416-1419 (2011). 1207 + +<--- Page Split ---> +![](images/Figure_4.jpg) + + +![PLACEHOLDER_49_1] + + +![PLACEHOLDER_49_2] + + +<--- Page Split ---> +![PLACEHOLDER_50_0] + + + +
Fig. 2
+ +<--- Page Split ---> +![PLACEHOLDER_51_0] + + + +![PLACEHOLDER_51_1] + + + +![PLACEHOLDER_51_2] + + + +![PLACEHOLDER_51_3] + + + +![PLACEHOLDER_51_4] + + + +![PLACEHOLDER_51_5] + + + +![PLACEHOLDER_51_6] + + + +![PLACEHOLDER_51_7] + + + +![PLACEHOLDER_51_8] + + +<--- Page Split ---> +![PLACEHOLDER_52_0] + + + +![PLACEHOLDER_52_1] + + + +![PLACEHOLDER_52_2] + + + +![PLACEHOLDER_52_3] + + + +![PLACEHOLDER_52_4] + + + +
Fig.4
+ +<--- Page Split ---> +![PLACEHOLDER_53_0] + + +<--- Page Split ---> +![PLACEHOLDER_54_0] + + +<--- Page Split ---> +![PLACEHOLDER_55_0] + + +![PLACEHOLDER_55_1] + + +![PLACEHOLDER_55_2] + + +![PLACEHOLDER_55_3] + + +<--- Page Split ---> +![PLACEHOLDER_56_0] + + +<--- Page Split ---> + +# HIIT in individuals with a lean body, obesity, and T2D + +![PLACEHOLDER_57_0] + + +<--- Page Split ---> diff --git a/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f_det.mmd b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..4e45f4e2073c7d653e02aecf1c141d04934217cc --- /dev/null +++ b/preprint/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f/preprint__0db7a7b09a74a339902dc4b181b4ce940d0055363f3a89dccc6cc3d69449e21f_det.mmd @@ -0,0 +1,1010 @@ +<|ref|>title<|/ref|><|det|>[[42, 106, 927, 241]]<|/det|> +# The mitochondrial mRNA stabilizing protein, SLIRP, regulates skeletal muscle mitochondrial structure and respiration by exercise-recoverable mechanisms + +<|ref|>text<|/ref|><|det|>[[42, 263, 240, 310]]<|/det|> +Lykke Sylow Lykkesylow@sund.ku.dk + +<|ref|>text<|/ref|><|det|>[[42, 336, 640, 356]]<|/det|> +University of Copenhagen https://orcid.org/0000- 0003- 0905- 5932 + +<|ref|>text<|/ref|><|det|>[[42, 361, 281, 401]]<|/det|> +Tang Cam Phung Pham University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 406, 281, 447]]<|/det|> +Steffen Raun University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 453, 640, 495]]<|/det|> +Essi Havula University of Helsinki + +<|ref|>text<|/ref|><|det|>[[42, 500, 640, 541]]<|/det|> +Carlos Henriquez- Olguin University of Copenhagen https://orcid.org/0000- 0002- 9315- 9365 + +<|ref|>text<|/ref|><|det|>[[42, 545, 586, 586]]<|/det|> +Diana Rubalcava- Gracia Karolinska Institute https://orcid.org/0000- 0002- 4615- 7375 + +<|ref|>text<|/ref|><|det|>[[42, 592, 281, 633]]<|/det|> +Emma Frank University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 638, 281, 679]]<|/det|> +Andreas Fritzen University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 685, 590, 726]]<|/det|> +Paulo Jannig Karolinska Institutet https://orcid.org/0000- 0002- 2569- 4091 + +<|ref|>text<|/ref|><|det|>[[42, 731, 281, 772]]<|/det|> +Nicoline Andersen University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 778, 291, 819]]<|/det|> +Rikke Kruse Odense University Hospital + +<|ref|>text<|/ref|><|det|>[[42, 824, 281, 865]]<|/det|> +Mona Ali University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 870, 281, 911]]<|/det|> +Jens Halling University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[42, 916, 281, 957]]<|/det|> +Stine Ringholm University of Copenhagen + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 42, 692, 940]]<|/det|> +Elise Needham University of Cambridge Solvejg Hansen University of Copenhagen Anders Lemminger University of Copenhagen Peter Schjerling Bispebjerg Hospital https://orcid.org/0000-0001-7138-3211 Maria Petersen Odense University Hospital Martin de Almeida University of Southern Denmark https://orcid.org/0000-0003-1794-6137 Thomas Jensen University of Copenhagen https://orcid.org/0000-0001-6139-8268 Bente Kiens University of Copenhagen https://orcid.org/0000-0001-5705-5625 Morten Hostrup University of Copenhagen https://orcid.org/0000-0002-6201-2483 Steen Larsen University of Copenhagen https://orcid.org/0000-0002-5170-4337 Niels Ørtenblad University of Southern Denmark https://orcid.org/0000-0001-9060-4139 Kurt Højlund Odense University Hospital Michael Kjær Bispebjerg Hospital Jorge Ruas Karolinska Institutet Aleksandra Trifunovic University of Cologne https://orcid.org/0000-0002-5472-3517 Jørgen Wojtaszewski University of Copenhagen https://orcid.org/0000-0001-8185-3408 Joachim Nielsen University of Southern Denmark https://orcid.org/0000-0003-1730-3094 Klaus Qvortrup University of Copenhagen https://orcid.org/0000-0001-8811-0260 Henriette Pilegaard University of Copenhagen https://orcid.org/0000-0002-1071-0327 Erik Richter + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 106, 103, 124]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 144, 932, 187]]<|/det|> +Keywords: SRA stem- loop interacting RNA- binding protein, mitochondrial dysfunction, skeletal muscle, exercise training, Drosophila + +<|ref|>text<|/ref|><|det|>[[44, 205, 303, 223]]<|/det|> +Posted Date: March 1st, 2024 + +<|ref|>text<|/ref|><|det|>[[42, 243, 475, 262]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3900062/v1 + +<|ref|>text<|/ref|><|det|>[[42, 280, 914, 323]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 341, 535, 360]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 395, 916, 438]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 13th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 54183- 4. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 83, 880, 130]]<|/det|> +1 The mitochondrial mRNA stabilizing protein, SLIRP, regulates skeletal muscle mitochondrial structure and respiration by exercise-recoverable mechanisms + +<|ref|>text<|/ref|><|det|>[[70, 150, 880, 390]]<|/det|> +3 Tang Cam Phung Pham \(^{1,2}\) , Steffen Henning Raun \(^{2}\) , Essi Havula \(^{3}\) , Carlos Henriquez- Olguin \(^{1}\) , Diana Rubalcava- Gracia \(^{4}\) , Emma Frank \(^{2}\) , Andreas Mæchel Fritzen \(^{1,2}\) , Paulo R. Jannig \(^{5}\) , Nicoline Resen Andersen \(^{1}\) , Rikke Kruse \(^{6}\) , Mona Sadek Ali \(^{2}\) , Jens Frey Halling \(^{7}\) , Stine Ringholm \(^{7}\) , Elise J. Needham \(^{8,9}\) , Solvejg Hansen \(^{1}\) , Anders Krogh Lemminger \(^{1}\) , Peter Schjerling \(^{10,11}\) , Maria Houborg Petersen \(^{6}\) , Martin Eisemann de Almeida \(^{6,12}\) , Thomas Elbenhardt Jensen \(^{1}\) , Bente Kiens \(^{1}\) , Morten Hostrup \(^{1}\) , Steen Larsen \(^{2,10,11,13}\) , Niels Ørtenblad \(^{12}\) , Kurt Højlund \(^{6,14}\) , Michael Kjær \(^{10,11}\) , Jorge L. Ruas \(^{5}\) , Aleksandra Trifunovic \(^{15}\) , Jørgen Frank Pind Wojtaszewski \(^{1}\) , Joachim Nielsen \(^{12}\) , Klaus Qvortrup \(^{2}\) , Henriette Pilegaard \(^{7}\) , Erik Arne Richter \(^{1}\) , Lykke Sylow \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[67, 403, 880, 556]]<|/det|> +11 ORCID: TCPP, 0000- 0003- 3476- 6891; SHR, 0000- 0003- 2050- 505X; EH, 0000- 0002- 9253- 5816; 12 DRG, 0000- 0002- 4615- 7375; JN, 0000- 0003- 1730- 3094; AMF, 0000- 0001- 8793- 9739; PS, 0000- 13 0001- 7138- 3211; NRA 0000- 0002- 1512- 2727; MSA, 0000- 0003- 1152- 3256; EAR, 0000- 0002- 6850- 14 3056; LS, 0000- 0003- 0905- 5932; MK, 0000- 0002- 4582- 8755; EF: 0000- 0002- 2294- 952X; AKL, 15 0000- 0002- 7929- 7147; MH, 0000- 0002- 6201- 2483; SR, 0000- 0001- 8436- 559X; HP, 0000- 0002- 16 1071- 0327; SL, 0000- 0002- 5170- 4337; PRJ, 0000- 0002- 2569- 4091; JLR, 0000- 0002- 1110- 2606; JW, 17 0000- 0001- 8185- 3408) + +<|ref|>text<|/ref|><|det|>[[67, 569, 877, 870]]<|/det|> +18 1 Department of Nutrition, Exercise and Sports, Faculty of Science, University of Copenhagen, Copenhagen, Denmark 20 2 Department of Biomedical Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark 21 3 Stem Cells and Metabolism Research Program, Faculty of Medicine, University of Helsinki, Finland 22 4 Division of Molecular Metabolism, Department of Medical Biochemistry and Biophysics, Karolinska Institutet, Stockholm, Sweden 25 5 Molecular and Cellular Exercise Physiology, Department of Physiology and Pharmacology, Karolinska Institutet, SE- 17177 Stockholm, Sweden 27 6 Steno Diabetes Center Odense, Odense University Hospital, Odense, Denmark 28 7 Department of Biology, University of Copenhagen, Copenhagen, Denmark + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 84, 833, 123]]<|/det|> +29 'British Heart Foundation Cardiovascular Epidemiology Unit, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK + +<|ref|>text<|/ref|><|det|>[[67, 138, 850, 177]]<|/det|> +31 'Victor Phillip Dahdaleh Heart and Lung Research Institute, University of Cambridge, Cambridge, UK + +<|ref|>text<|/ref|><|det|>[[67, 193, 831, 231]]<|/det|> +33 'Institute of Sports Medicine Copenhagen, Department of Orthopaedic Surgery M, Bispebjerg Hospital, Copenhagen, Denmark + +<|ref|>text<|/ref|><|det|>[[67, 237, 840, 275]]<|/det|> +35 'Center for Healthy Aging, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark + +<|ref|>text<|/ref|><|det|>[[67, 281, 820, 319]]<|/det|> +37 'Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark + +<|ref|>text<|/ref|><|det|>[[67, 325, 712, 343]]<|/det|> +39 'Clinical Research Centre, Medical University of Bialystok, Bialystok, Poland + +<|ref|>text<|/ref|><|det|>[[67, 349, 775, 366]]<|/det|> +40 'Department of Clinical Research, University of Southern Denmark, Odense, Denmark + +<|ref|>text<|/ref|><|det|>[[67, 372, 844, 390]]<|/det|> +41 'Institute for Mitochondrial Diseases and Aging, Cologne Excellence Cluster on Cellular Stress + +<|ref|>text<|/ref|><|det|>[[67, 395, 850, 413]]<|/det|> +42 Responses in Aging- Associated Diseases (CECAD) and Center for Molecular Medicine (CMMC), + +<|ref|>text<|/ref|><|det|>[[67, 419, 636, 436]]<|/det|> +43 Medical Faculty, University of Cologne, D- 50931 Cologne, Germany + +<|ref|>text<|/ref|><|det|>[[67, 443, 650, 460]]<|/det|> +44 + +<|ref|>text<|/ref|><|det|>[[67, 466, 650, 483]]<|/det|> +45 Running title: Exercise reverses SLIRP- induced mitochondrial defects + +<|ref|>text<|/ref|><|det|>[[67, 504, 880, 522]]<|/det|> +46 Keywords: SRA stem- loop interacting RNA- binding protein, mitochondrial dysfunction, skeletal + +<|ref|>text<|/ref|><|det|>[[67, 535, 412, 551]]<|/det|> +47 muscle, exercise training, Drosophila + +<|ref|>text<|/ref|><|det|>[[67, 575, 440, 591]]<|/det|> +48 \*Corresponding author: Lykke Sylow + +<|ref|>text<|/ref|><|det|>[[67, 614, 530, 630]]<|/det|> +49 Lykkesylow@sund.ku.dk + +<|ref|>text<|/ref|><|det|>[[67, 653, 612, 670]]<|/det|> +50 Competing Interests. The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 383, 101]]<|/det|> +## Summary and graphical abstract + +<|ref|>text<|/ref|><|det|>[[115, 111, 883, 494]]<|/det|> +Summary and graphical abstractDecline in mitochondrial function associates with decreased muscle mass and strength in multiple conditions, including sarcopenia and type 2 diabetes. Optimal treatment could include improving mitochondrial function, however, there are limited and equivocal data regarding the molecular cues controlling muscle mitochondrial plasticity. Here we uncover the mitochondrial- mRNA- stabilizing protein SLIRP, in complex with LRPPRC, as a PGC- 1α target that regulates mitochondrial structure, respiration, and mitochondrially- encoded- mRNA pools in skeletal muscle. Exercise training effectively counteracted mitochondrial defects induced by loss of LRPPRC/SLIRP, despite sustained low mitochondrially- encoded- mRNA pools, via increased mitoribosome translation capacity. In humans, exercise training robustly increased muscle SLIRP and LRPPRC protein content across exercise modalities and sexes, yet this increase was less prominent in subjects with type 2 diabetes. Our work identifies a mechanism of post- transcriptional mitochondrial regulation in skeletal muscle through mitochondrial mRNA stabilization. It emphasizes exercise as an effective approach to alleviate mitochondrial defects by possibly increasing mitoribosome capacity. + +<|ref|>image<|/ref|><|det|>[[115, 530, 880, 920]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 85, 222, 100]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[113, 121, 883, 533]]<|/det|> +Mitochondrial homeostasis and function are vital for skeletal muscle physiology, primarily influenced by the ability to meet energy demands through oxidative phosphorylation (OXPHOS) in mitochondria1. The decline in mitochondrial function is associated with decreased muscle mass and strength in multiple conditions, including sarcopenia, type 2 diabetes (T2D), and cancer. Reduced muscle function adversely impacts health and quality of life1- 9 and is associated with increased all- cause mortality. Mounting evidence suggests that improvements in mitochondrial metabolism contribute to the exercise- induced functional benefits, as mitochondrial preservation protects against the age- associated decline in skeletal muscle mass and performance10, 11. The beneficial effects of exercise training are more potent than any drug in preserving muscle mass and function, positioning exercise training as a frontline strategy for the prevention and treatment of sarcopenia, T2D, and cancer12- 14. Viewing mitochondrial control through the lens of exercise biology is a strategy to gain new mechanistic insights into mitochondrial regulation, which is currently poorly understood. With an aging population and no FDA/EMA- approved drugs to treat muscle functional decline, there is an urgent unmet need to identify potential treatment targets. + +<|ref|>text<|/ref|><|det|>[[113, 553, 883, 901]]<|/det|> +OXPHOS proteins are of dual origin and transcribed in spatially distinct cellular compartments. Nuclear (n)DNA encodes for the majority of OXPHOS proteins. Yet, 13 essential OXPHOS proteins are encoded by the small circular high- copy number mitochondrial DNA (mtDNA)15, 16. In non- muscle systems, a protein complex comprised of steroid receptor RNA activator stem- loop interacting RNA- binding protein (SLIRP) and the leucine- rich pentatricopeptide repeat containing protein (LRPPRC) mediates mitochondrial- mRNA (mt- mRNA) stability and polyadenylation of most mtDNA- encoded OXPHOS transcripts17- 21. Stabilization of mt- mRNA is critical for protein translation by the mitochondrial ribosome (mitoribosome). Yet, the mitoribosome comprises 82 mitoribosomal proteins encoded by nuclear genes, and 12S and 16S mt- rRNAs, highlighting the requirement for intricate coordination between the processes of cytosolic and mitochondrial protein synthesis18, 19. Despite the high mitochondrial abundance in skeletal muscle, the complex interaction between LRPPRC/SLIRP- mediated posttranscriptional processes, mitoribosomal translation, and mitochondrial function has not + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 82, 883, 504]]<|/det|> +been studied before in skeletal muscle. Elucidating the underlying mechanisms for stabilizing mt- mRNA could not only facilitate our understanding of the fundamental energy metabolism in muscle, but also aid intervention strategies for diseases associated with skeletal muscle mitochondrial defects. Endurance exercise training potently increases mitochondrial mass in skeletal muscle, in part due to the increased content of nDNA- and mtDNA- encoded mitochondrial proteins1, 22. Importantly, exercise training retains its ability to improve oxidative capacity not only in healthy individuals but also in patients with mtDNA mutations, with common PGC- 1α- mediated mitochondrial adaptive responses shared between both groups13. These results suggest that exercise training induces adaptations to reinforce the mitochondria's ability to efficiently respond to increased energy demands, even in the presence of mtDNA mutations, that may negatively affect mitochondrial transcription and translation. However, a notable knowledge gap persists in understanding the role of mitochondrial posttranscriptional processes, specifically in mt- mRNA stabilization and translation, in skeletal muscle biology and following exercise training. Yet, that knowledge is needed to gain a comprehensive understanding of the adaptive responses to exercise training. + +<|ref|>text<|/ref|><|det|>[[111, 522, 883, 781]]<|/det|> +Since SLIRP, a key player in mitochondrial posttranscriptional gene expression, is markedly upregulated in mouse skeletal muscle by exercise training23, we hypothesized that SLIRP would regulate mitochondrial function in skeletal muscle at rest and in response to exercise training. Our findings illuminate SLIRP's role in regulating mt- mRNA transcript levels in skeletal muscle, downstream of PGC- 1α1. Knockout (KO) of SLIRP led to damaged and fragmented mitochondria alongside lowered respiration. Intriguingly, exercise training could compensate for the absence of SLIRP, leading to improvements in mitochondrial integrity and respiratory capacity. Our findings imply the activation of complex exercise- induced molecular signaling in skeletal muscle which intriguingly bypasses mt- mRNA defects, possibly through enhanced mitoribosomal translation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 180, 100]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[117, 123, 880, 171]]<|/det|> +Slirp knockout caused defects in mitochondrial structure and respiratory capacity in mouse muscle. + +<|ref|>text<|/ref|><|det|>[[115, 193, 881, 361]]<|/det|> +Protein profiling of five different skeletal muscle tissues showed that SLIRP content was highest in the oxidative soleus compared to glycolytic extensor digitorum longus (EDL, Fig. 1A, Supplementary Fig. 1A), in accordance with the potential critical role for SLIRP in oxidation. Moreover, SLIRP was present across a diverse array of tissues in mice and was highly abundant in energy- demanding tissues such as liver, kidney, brown adipose tissue, heart, and skeletal muscle (Fig. 1A, Supplementary Fig. 1A), in line with previously reported Slirp mRNA levels24. + +<|ref|>text<|/ref|><|det|>[[115, 382, 881, 580]]<|/det|> +In skeletal muscle of global Slirp KO mice, SLIRP protein was undetectable. Moreover, SLIRP's binding partner LRPPRC showed \(90\%\) reduction in protein content (Fig. 1B), indicative of their co- stabilization. To confirm co- stabilization, we administered intramuscular injections of recombinant adeno- associated viral serotype 6 (rAAV6): Slirp. As expected, the absence of concomitant upregulation of LRPPRC, failed to increase total SLIRP protein content due to downregulation of endogenous SLIRP protein (Fig. 1C). These findings show that SLIRP and LRPPRC stabilize each other in skeletal muscle, aligning with findings from non- muscle tissue17, 18, 20. + +<|ref|>text<|/ref|><|det|>[[115, 602, 881, 890]]<|/det|> +We next determined the role of SLIRP for mitochondrial structure in skeletal muscle, using the membrane potential probe, tetramethylrhodamine ethyl ester (TMRE+), in intact flexor digitorum brevis (FDB) muscle fibers of Slirp KO and littermate control wild- type (WT) mice. Interestingly, the finely interconnected mitochondrial network, obvious in WT fibers, was disrupted in Slirp KO muscle fibers, evident by a \(30\%\) increased mitochondrial fragmentation index (Fig. 1D). Transmission electron microscopy (TEM) analysis of gastrocnemius muscle provided further insights into the ultrastructural changes of mitochondrial morphology in Slirp KO (Fig. 1E). Half of the KO gastrocnemius muscles displayed reduced density of the matrix, disarray of the cristae, and substantial enlargement and vacuolation of intermyofibrillar (IMF) mitochondria (Fig. 1E). Quantitative analysis of the mitochondria showed that the percentage of damaged IMF mitochondria was increased (Fig. 1F). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 883, 313]]<|/det|> +However, there were no significant alterations in the total number of IMF mitochondria (Fig. 1G), number of damaged or the total number of subsarcolemmal mitochondria (Fig. 1G), nor any changes in number of damaged or total number of subsarcolemmal mitochondria (not shown). The percentage of damaged mitochondria in Slirp KO muscle varied between \(2 - 20\%\) of total mitochondria compared to \(< 2\%\) of damaged mitochondria in WT muscle (Fig. 1E, F). Other structural parameters of IMF mitochondria, such as mitochondrial area, aspect ratio, elongation, convexity, perimeter, sphericity, and diameter did not display any significant changes in Slirp KO muscles, likely due to heterogeneity in the extent of damage (Fig. 1F, Supplementary Fig. 1B- H). + +<|ref|>text<|/ref|><|det|>[[115, 333, 883, 620]]<|/det|> +In agreement with the observable abnormalities in mitochondrial structure, maximal respiratory capacity was reduced in permeabilized gastrocnemius muscle fibers of Slirp KO mice compared to WT (Fig. 1H). This was particularly evident when adding glutamate to assess maximal complex I linked respiratory activity (- 24%) and succinate to assess complex I+II linked respiratory capacity (- 21%). These results indicate an important role for SLIRP in skeletal muscle respiration, which contrast findings in isolated mitochondria from liver and heart tissues, where Slirp KO did not compromise mitochondrial respiration18. In agreement with lower respiration, fatty acid oxidation tended (p=0.0718) to be lower in in intact incubated soleus muscle of Slirp KO mice (Fig. 1I). However, electrically-induced muscle contraction increased fatty acid oxidation similarly in both genotypes, suggesting that fatty acid utilization for fuel during muscle contraction does not depend on SLIRP (Fig. 1I). + +<|ref|>text<|/ref|><|det|>[[115, 641, 882, 780]]<|/det|> +With the suggested role of SLIRP as an mt-mRNA stabilizing protein in non-muscle cells18, we determined mtDNA- encoded mRNA transcripts. Intriguingly, we observed a 60- 80% reduction in mt- Nd1, mt- Nd5/Nd6 (Complex I), mt- CytB (Complex III), mt- Col (Complex IV), and mt- Atp6 (ATP synthase) in gastrocnemius muscle (Fig. 1J), suggesting that SLIRP stabilizes mt- mRNA in skeletal muscle. + +<|ref|>text<|/ref|><|det|>[[115, 803, 880, 849]]<|/det|> +Together, these results suggest that mt- mRNA stabilization via SLIRP is required for proper mitochondrial network structure and morphology, and respiration in mouse muscle. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 84, 668, 101]]<|/det|> +## SLIRP knockdown impaired locomotion and reduced lifespan in flies. + +<|ref|>text<|/ref|><|det|>[[115, 120, 883, 530]]<|/det|> +To determine the long- term consequences of SLIRP deficiency on the whole organism, we utilized the UAS- GAL4 system \(^{25}\) for muscle- specific (Mef2- GAL4> ) knock- down (KD) of SLIRP1 and SLIRP2 in Drosophila (D.) melanogaster. Only one SLIRP gene is present in mouse and human, while two SLIRP genes exist in D. melanogaster (Flybase annotation symbols (http://flybase.org): CG33714 and CG8021) likely to have originated from gene duplication events \(^{26}\) . This consequently gives rise to two fly orthologue proteins of the human and mouse SLIRP, denoted SLIRP1 (CG33714) and SLIRP2 (CG8021) \(^{26}\) . To test physical functionality, we subjected the flies to a negative geotaxis (climbing) assay \(^{27}\) and found that SLIRP2 KD, but not SLIRP1 KD, flies climbed 23% slower than control flies, indicative of impaired muscle function (Fig. 1K). The detrimental effects of SLIRP KD in the muscle became further apparent following starvation. Median fasting survival of SLIRP1 KD flies was 41 h, compared with 44 h for control and SLIRP2 KD flies (Fig. 1L, p=0.0681). Muscle- specific SLIRP deficiency, irrespective of orthologue, had a deleterious effect on lifespan. While the median survival of control flies was 77 days, median survival of SLIRP1 and SLIRP2 KD flies was only 55 and 63 days, respectively, and thus both SLIRP1 and SLIRP2 KD significantly reduced lifespan (Fig. 1M). + +<|ref|>text<|/ref|><|det|>[[116, 553, 881, 629]]<|/det|> +Taken together these findings in the flies show that SLIRP KD- induced mitochondrial defects can have detrimental long- term consequences such as impaired muscle function, starvation intolerance, and reduced life span. + +<|ref|>sub_title<|/ref|><|det|>[[116, 652, 880, 698]]<|/det|> +## Muscle SLIRP protein content increases in response to exercise training, under the regulation of PGC-1α1. + +<|ref|>text<|/ref|><|det|>[[115, 720, 882, 890]]<|/det|> +Having established critical functional roles of SLIRP in skeletal muscle mitochondrial morphology and respiratory capacity with detrimental effects on lifespan, we next investigated the upstream regulation of SLIRP protein content. Our recent work pinpointed SLIRP as a protein responsive to exercise training \(^{23}\) ; yet the mechanisms governing its induction and functions remained elusive. Knowing that exercise potently increases PGC- 1α protein and mRNA levels in both mouse and human skeletal muscle \(^{28,31}\) , and that PGC- 1α is an important regulator of mitochondrial biogenesis, respiration, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 883, 344]]<|/det|> +quality control \(^{32}\) , we explored the regulation of SLIRP by PGC- 1α. The Ppargc1a gene encodes several PGC- 1α isoforms, including isoform 1 (PGC- 1α1) and isoform 4 (PGC- 1α4), with distinct regulation and biological functions \(^{33, 34}\) . Overexpression (OE) of PGC- 1α1 in mouse skeletal muscle \(^{33, 35}\) , associating with endurance exercise training- like adaptations \(^{33, 35, 36}\) , resulted in approximately 8.3- fold higher muscle SLIRP protein content, without a concomitant change in mRNA levels (Fig. 2A). On the contrary, PGC- 1α4 OE \(^{37}\) , inducing resistance- type exercise training adaptations such as muscle hypertrophy and strength \(^{33}\) , had no effect on SLIRP protein content in mouse gastrocnemius muscle (Fig. 2B). This suggests that SLIRP is an unrecognized player in PGC- 1α1- regulated oxidative metabolism. + +<|ref|>text<|/ref|><|det|>[[115, 363, 883, 562]]<|/det|> +To test whether PGC- 1α regulates Slirp mRNA in skeletal muscle following exercise, we determined Slirp and Ppargc1a (Exon 3- 5) mRNA levels in tibialis anterior (TA) muscle 3 h into recovery from one exercise bout in mice lacking PGC- 1α in skeletal muscle (PGC- 1α mKO) and WT control littermates. Slirp mRNA was not upregulated 3 h post- exercise (Fig. 2C), in contrast to the expected \(^{31}\) increase in Ppargc1a (+6.8- fold; Fig. 2C) in WT mice. Interestingly, Slirp mRNA levels were only modestly 14% reduced in PGC- 1α KO skeletal muscle (Fig. 2C), in line with the dissociation between Slirp mRNA levels and SLIRP protein levels observed in PGC- 1α1 OE skeletal muscle (Fig. 2A). + +<|ref|>text<|/ref|><|det|>[[115, 582, 883, 872]]<|/det|> +These findings align with reports describing only moderate correlations between mRNA and protein levels following exercise \(^{38 - 40}\) . Yet, with the interpretative limitations of a single post- exercise time point, we aimed to acquire a more detailed record of the temporal changes in Slirp transcripts in recovery from exercise. We subjected WT mice to an acute exercise bout at 60% of their maximal running capacity and harvested quadriceps muscles at rest, immediately after, 2h, 6h, and 24h post- exercise. The phosphorylation of AMP- activated protein kinase (AMPK) at T172 served as a validation of elevated metabolic stress during the exercise bout (Fig. 2E). We found that Slirp mRNA levels were 1.4- fold elevated 6h into the recovery period in quadriceps muscle (Fig. 2D), a time- point not included in our prior cohort (Fig. 2C), and partially concurrent with PGC- 1α1 mRNA levels that were elevated 2h (+3.0- fold) and 6h post- exercise (+3.3- fold; Fig. 2D). Following a single bout of exercise, we observed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 130]]<|/det|> +no changes to SLIRP protein content (Fig. 2E). These findings indicate that SLIRP underlies PGC- 1α1- regulated control following acute exercise. + +<|ref|>text<|/ref|><|det|>[[115, 152, 883, 592]]<|/det|> +Given that SLIRP might be regulated by PGC- 1α, we next turned to investigate the long- term dependency of exercise training- induced regulation of SLIRP protein by PGC- 1α. In contrast to acute exercise, exercise training increased SLIRP protein content 4.8- fold in WT mice (Fig. 2F), recapitulating our earlier observations in mice engaged in voluntary wheel running23. The effect of exercise training on SLIRP protein content was markedly blunted by 70% in mice lacking PGC- 1α in muscle compared to trained WT mice (Fig. 2F). Expectedly, in WT mice, exercise training potently upregulated the steady- state levels of multiple OXPHOS proteins including SDHB, UQCRC2, MTCO1 in skeletal muscle (Supplementary Fig. 1I- K). In mice lacking PGC- 1α in muscle, SDHB (Supplementary Fig. 1I) and MTCO1 protein (Supplementary Fig. 1K) was 50- 60% decreased in SED mice. Moreover, the effect of exercise training on SDHB (Supplementary Fig. 1I) and UQCRC2 protein content (Supplementary Fig. 1J) was blunted by 30% and 65%, respectively, in PGC- 1α mKO compared to WT mice. In contrast to PGC- 1α, the exercise- sensitive metabolic sensor AMPK41 did not seem to regulate SLIRP expression, as SLIRP protein abundance was comparable to controls in muscle- specific double KO of AMPKα1 and AMPKα2 and WT mice42, 43 treated with or without the AMPK- activator AICAR for 4 weeks (Fig. 2G). + +<|ref|>text<|/ref|><|det|>[[115, 613, 881, 689]]<|/det|> +PGC- 1α has marked effects on muscle endurance, partially attributed to its pivotal role in mitochondrial biogenesis, and respiration35, 44. Our results suggest SLIRP as an unrecognized downstream target of PGC- 1α in such exercise adaptations and oxidative metabolism. + +<|ref|>text<|/ref|><|det|>[[115, 712, 881, 759]]<|/det|> +SLIRP is dispensable for organismal adaptations to exercise training, yet, needed for improving blood glucose regulation after exercise training in male mice. + +<|ref|>text<|/ref|><|det|>[[115, 782, 881, 858]]<|/det|> +Having identified SLIRP as an exercise training responsive protein downstream of PGC- 1α, we next determined SLIRP's mechanistic involvement in exercise training- mediated organismal adaptations. To this end, we conducted a 10- week voluntary wheel- running training study in Slirp KO and WT + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 130]]<|/det|> +littlerate mice of both sexes commencing at \(\sim 18\) weeks of age. The experimental design is shown in Fig. 3A. + +<|ref|>text<|/ref|><|det|>[[115, 153, 883, 473]]<|/det|> +The daily running distance recorded over 6 weeks of the 10- week study was \(2.4\mathrm{km / day}\) on average for male mice of both genotypes, whereas it was \(4.8\mathrm{km / day}\) for female exercise- trained (ET) WT mice and \(3.0\mathrm{km / day}\) for female KO ET mice (Fig. 3B). Of note, the running distance observed in our study of 18- week old mice is shorter than those reported in other exercise training studies using \(\sim 10\) - week- old mice \(^{23,45,46}\) . Despite the higher food intake in ET mice (Fig. 3C), exercise training lowered body weight in both trained genotypes (Supplementary Fig. 2A). Moreover, irrespective of genotype, exercise training reduced fat mass/body weight ratio (Supplementary Fig. 2B) and increased lean mass/body weight ratio compared to sedentary (SED) groups (Supplementary Fig. 2C). Exercise training had no effect on gastrocnemius or TA muscle weight relative to tibia length in either sex (Supplementary Fig. 2D). However, we noted that female Slirp KO SED and both ET groups had elevated heart weight relative to WT SED (Supplementary Fig. 2D). + +<|ref|>text<|/ref|><|det|>[[115, 493, 881, 631]]<|/det|> +Exercise capacity increased similarly with exercise training, independently of genotype (+1.2- fold for WT ET, +1.3- fold for KO ET; Supplementary Fig. 2E) and sex (females shown in Supplementary Fig. 2G). Post- exercise blood lactate levels tended to be reduced (- 1.7- fold; \(\mathrm{p} = 0.067\) ) by ET in WT male mice (Supplementary Fig. 2F). The training- induced lowering of blood lactate post- exercise was absent in male and female Slirp KO mice (Supplementary Fig. 2F, H). + +<|ref|>text<|/ref|><|det|>[[115, 653, 883, 911]]<|/det|> +Exercise training is a powerful preventative treatment against many metabolic disorders \(^{12}\) , partially due to its remarkable effects in improving glucose tolerance and insulin sensitivity documented in both animal models and humans \(^{23,45,47}\) . Accordingly, male WT ET mice displayed reduced fasting blood glucose (Fig. 3D) and improved blood glucose control evidenced by a downshifted glucose response curve compared to WT SED mice. This effect was not observed in trained male Slirp KO mice (Fig. 3E). Thus, trained Slirp KO mice displayed similar glucose tolerance as both SED groups. The incremental area under the curve (iAUC) of blood glucose was similar between groups (Fig. 3F), suggesting that the lowered plasma glucose excursion during the GTT was driven by lower basal glucose levels. Yet, exercise training reduced the levels of plasma insulin in both WT and Slirp KO + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 160]]<|/det|> +mice, suggesting that insulin sensitivity was increased by exercise training in both groups (Fig. 3G). There was no effect of exercise training or genotype on fasting glucose, glucose tolerance or insulin levels in female mice (Fig. 3H- K). + +<|ref|>text<|/ref|><|det|>[[115, 183, 880, 230]]<|/det|> +Our results demonstrate that SLIRP is largely dispensable for organismal adaptations to exercise training, yet, in male mice required for training- induced improvement in blood glucose regulation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 252, 880, 299]]<|/det|> +## Exercise training reverses Slirp KO-induced defects in muscle mitochondrial structure and function. + +<|ref|>text<|/ref|><|det|>[[115, 321, 881, 459]]<|/det|> +Exercise training can counteract mitochondrial damage, arising from excessive accumulation of reactive oxygen species (ROS) or impaired assembly of OXPHOS complex following mtDNA mutations11, 13, 48- 50. As we established that loss of SLIRP had marked negative implications for mitochondrial structure and respiration in skeletal muscle (Fig. 1), we asked whether exercise training could rescue these defects. + +<|ref|>text<|/ref|><|det|>[[115, 482, 881, 589]]<|/det|> +Skeletal muscle mitochondrial network structure was qualitatively investigated in TMRE+ stained FDB fibers by confocal microscopy. Remarkably, exercise training completely rescued the derangements in mitochondrial network structure and mitochondrial fragmentation observed in Slirp KO SED mice, quantitatively corroborated by restoration of the fragmentation index (Fig. 4A). + +<|ref|>text<|/ref|><|det|>[[115, 611, 883, 840]]<|/det|> +Concomitantly with improved mitochondrial structure, exercise training restored the respiratory flux in Slirp KO mice to the level of WT muscle (Fig. 4B). WT mice did not improve respiratory flux with exercise training, which was consistent with no effect of exercise training on fatty acid oxidation of contracting isolated WT soleus muscle (Supplementary Fig. 2I) and aligns with other studies51. Exercise training in Slirp KO mice improved fatty acid oxidation in contracting Slirp KO soleus muscles (Supplementary Fig. 2I), in alignment with improved respiratory flux in Slirp KO muscles by exercise training. Thus, exercise training counteracted the structural alterations in Slirp KO SED mice and had a positive impact on mitochondrial respiratory capacity, suggesting restored mitochondrial function. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 160]]<|/det|> +Together, these results support that exercise training exploits the remarkable adaptive plasticity that mitochondria retain even in the event of mitochondrial network disruptions and impaired respiratory flux in skeletal muscle. + +<|ref|>sub_title<|/ref|><|det|>[[115, 183, 850, 201]]<|/det|> +## Sustained reductions of mitochondrial transcript levels can be bypassed by exercise training. + +<|ref|>text<|/ref|><|det|>[[115, 222, 881, 391]]<|/det|> +We next aimed to investigate the molecular underpinnings for the defects in mitochondrial structure and function in Slirp KO and their correction by exercise training. First, we verified that both SLIRP (+1.4- fold; Fig. 4C) and concomitantly LRPPRC (+1.4- fold; Fig. 4D) were upregulated by exercise training in the gastrocnemius muscle in WT, but not Slirp KO mice (Fig. 4C, D). These findings reinforce the co- stabilizing relationship of SLIRP and LRPPRC existing not only at baseline17, 18, 20, 52 but also during exercise training conditions. + +<|ref|>text<|/ref|><|det|>[[114, 411, 881, 912]]<|/det|> +SLIRP, together with LRPPRC, has been shown to maintain mt- mRNA stability and aid mitoribosomal translation in non- muscle tissues17- 20, 52. Indeed, the marked downregulation of mt- mRNA transcripts in SED Slirp KO muscle, also shown in Fig. 1J, remained reduced in ET Slirp KO mice relative to WT ET mice (Fig. 4F). In contrast, mtDNA copy number, measured using quantitative PCR and primers for mt- Ndl (Supplementary Fig. 2J) or mt- Nds/mt- Nds (Supplementary Fig. 2K), were unaltered across groups or elevated only in Slirp KO SED mice, suggesting a compensatory response to mitigate mitochondrial dysfunction by increasing mtDNA copy number53. Thus, these findings suggest that SLIRP is crucial for mitochondrial transcript stability in skeletal muscle. We were intrigued to find that exercise training still restored protein content of mtDNA- encoded OXPHOS subunits, MT- ND3, MT- CO1 and MT- ATP6 (Fig. 4F). Thus, exercise training completely counteracted the sustained reduction in the mt- mRNA levels of these proteins in Slirp KO ET mice. The mild reduction of MT- ND3 protein, a subunit of Complex I, seen in Slirp KO muscle was similarly rescued by exercise training (Fig. 4F). Interestingly, although the nuclear- encoded Complex I subunit NDUFB8 was downregulated by 57% in the Slirp KO SED mice, exercise training markedly upregulated NDUFB8 protein content in Slirp KO mice, yet, below NDUFB8 protein content in WT ET mice (Fig. 4G). Other nuclear- encoded OXPHOS subunits, such as SDHB (Complex II), UQCRC2 (Complex III) and ATP5A (ATP synthase) were similarly upregulated by exercise training, independently of Slirp KO (Fig. 4G, representative + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 191]]<|/det|> +blots shown in Fig. 4H). Citrate synthase (CS) activity correlates with mitochondrial content in human54 and mouse skeletal muscle55. CS activity, indicative of mitochondrial content, was increased similarly in Slirp KO and WT mice after exercise training (Supplementary Fig. 2L), suggesting that improvements in mitochondrial content following exercise training are independent of SLIRP. + +<|ref|>text<|/ref|><|det|>[[115, 213, 881, 290]]<|/det|> +These results underline a substantial skeletal muscle mitochondrial plasticity, which is retained despite a \(60 - 80\%\) reduction in mt- mRNA pools. These findings point towards unidentified mechanisms by which ET improves translational efficiency to dictate final protein content. + +<|ref|>sub_title<|/ref|><|det|>[[115, 313, 880, 359]]<|/det|> +## Exercise training increases mitochondrial protein translation capacity by elevating mitoribosome mass. + +<|ref|>text<|/ref|><|det|>[[115, 381, 881, 672]]<|/det|> +We next sought to determine the potential mechanism by which sustained reductions of mitochondrial transcript levels can be bypassed by exercise training. Enhanced mitochondrial protein translation mediates exercise training adaptations in muscle mitochondrial function in both young and elderly individuals39. Thus, we assessed pathways associated with protein translation in mitochondria, facilitated by the resident mitoribomes. We observed training- induced increases in 12S ribosomal RNA (12S rRNA) and MRPL11 protein content (Fig. 5A, B), essential components of mitoribosome translation56, 57. MRPL11 protein content was \(36\%\) more elevated in response to exercise training in Slirp KO ET than WT ET mice, mainly in male mice (Fig. 5B). Conversely, exercise training did not induce changes in cytosolic ribosomal protein S6 protein content (rpS6, Fig. 5C; representative blots in Fig. 5D), indicating a compartmentalized ribosomal response to exercise training. + +<|ref|>text<|/ref|><|det|>[[115, 694, 881, 890]]<|/det|> +Given the predominant impact of Slirp KO on mitochondria, training- induced adaptive changes might also manifest directly within the affected cellular compartment. Here, we investigated a mitochondrial quality control system involving a mitochondrial scavenger enzyme, peroxiredoxin 3 (PRDX3). PRDX3 plays a role in balancing redox environment in mitochondria58, 59, possibly contributing to the preservation of factors critical for translation during conditions of mitochondrial dysfunction and exercise training. Interestingly, PRDX3 protein content was \(60\%\) lower in Slirp KO SED compared to WT SED mice, indicative of a lower mitochondrial oxidative stress defense capacity in Slirp KO + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 252]]<|/det|> +muscle. This was restored by exercise training, bringing PRDX3 up to levels observed in trained WT mice (Fig. 5E). We further noted that ET lowered the PRDX3 dimer to monomer ratio in both genotypes, indicative of a lower basal mitochondrial- derived peroxide accumulation likely due to an enhancement of scavenger activity (Fig. 5F). The cytosolic peroxidoxin 2 (PRDX2) was unaffected by genotype and exercise training (Fig. 5G, representative blots in Fig. 5H), further pointing towards highly mitochondria- selective effects of exercise training to circumvent SLIRP deficiency. + +<|ref|>text<|/ref|><|det|>[[115, 273, 881, 411]]<|/det|> +Collectively, our findings indicate a complex interplay of spatially distinct molecular muscle adaptations in response to Slirp KO and ET. Our findings suggest that exercise training induces mechanisms to increase mitochondrial protein translation capacity. Mechanistically, exercise training elevated mitoribosome mass and restored oxidative stress defense systems to powerfully circumvent the Slirp KO- induced depletion of mtDNA- encoded transcript levels. + +<|ref|>sub_title<|/ref|><|det|>[[115, 433, 880, 481]]<|/det|> +## Elevation of muscle SLIRP and LRPPRC protein content in response to ET is conserved in human skeletal muscle. + +<|ref|>text<|/ref|><|det|>[[115, 503, 881, 670]]<|/det|> +Our intriguing results from mouse and fly models prompted us to test whether our results provide translational value for humans in health and disease. SLIRP and LRPPRC protein content were determined in human skeletal muscle in four independent exercise cohorts employing different exercise modalities in conjunction with or without aging or diabetes and in males and females. Participant characteristics and study designs for the different ET modalities have been published previously for all the clinical studies60-63. + +<|ref|>text<|/ref|><|det|>[[115, 692, 881, 891]]<|/det|> +Skeletal muscle SLIRP and LRPPRC protein abundances were 70% increased following 14- weeks of controlled and supervised aerobic and strength exercise training intervention in healthy young women60 (Fig. 6A). Also, a 6- week high- intensity interval training (HIIT) intervention elevated muscle protein content of SLIRP (+80%, Fig. 6B) and LRPPRC (+50%, Fig. 6B) in healthy young men63. Thus, the exercise training response of SLIRP was highly consistent, independent of sex, and responsive to various exercise modalities. The 12- week progressive resistance exercise training61 increased SLIRP and LRPPRC in the entire group of young and old individuals but had no significant effect on LRPPRC + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 192]]<|/det|> +abundance in either young or elderly subjects separately (Fig. 6C). The mtDNA- encoded transcript levels of MT- ND6, MT- CYB, MT- CO1 and MT- ATP6 were comparable across all groups (Fig. 6D), indicating that mitochondrial gene expression was not significantly affected by either aging or the training regimen in this study. + +<|ref|>text<|/ref|><|det|>[[115, 213, 882, 561]]<|/det|> +Exercise training- mediated improvements in mitochondrial content and function may be associated with improved glucose metabolism in individuals with obesity and T2D64,65, although this association is not always clear66. The training response of SLIRP and LRPPRC to HIIT62 was conserved in individuals who were lean (+150% and +54%) or obese (+85% and +45%, Fig. 6E). However, the exercise response for LRPPRC, but not SLIRP, was blunted in patients with T2D in comparison to lean individuals (Fig. 6E), suggesting impaired sensitivity of muscle to some mitochondrial adaptations to ET in T2D. At the mRNA level, mt- DNA encoded transcripts of MT- ND1, MT- CYB, MT- CO1 and MT- ATP6 were reduced with HIIT across all groups (Fig. 6F). Yet, protein content of MT- CO2 (Fig. 6G, representative blot in Supplementary Fig. 3A) and other nuclear- encoded OXPHOS proteins (Supplementary Fig. 3A) were significantly upregulated with HIIT. In line with previous findings39, the upregulation in OXPHOS protein abundance occurred despite lower levels of mRNA, and demonstrate a disassociation between the transcriptome and proteome67. + +<|ref|>text<|/ref|><|det|>[[115, 583, 882, 811]]<|/det|> +Together, we show that SLIRP and LRPPRC are robustly upregulated in skeletal muscle across various exercise training modalities in men and women, lean and obese. However, the exercise training response of LRPPRC was blunted in patients with T2D. This suggests that T2D is associated with decreased sensitivity to some of the adaptive mitochondrial responses normally elicited by exercise training, possibly influenced by exercise intensity or volume. Thus, our clinical data in humans provide strong translational value of the SLIRP- deficient fly and mouse models and point towards a conserved requirement of SLIRP for skeletal muscle health, and potentially the SLIRP- independent mechanisms that compensate during exercise training. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 85, 204, 100]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[115, 123, 882, 291]]<|/det|> +Our investigation led to six key findings that bridge substantial knowledge gaps concerning the role of mitochondrial posttranscriptional mechanisms on skeletal muscle biology and their contributions to exercise training adaptations. Our findings not only unravel a delicate interplay between the mt- mRNA- stabilizing protein SLIRP, and the integrity of skeletal muscle mitochondria structure and performance, but also underscore the potential of exercise training in promoting selective mitochondrial translation capacity to circumvent these mechanisms. + +<|ref|>text<|/ref|><|det|>[[115, 312, 882, 600]]<|/det|> +First, lack of Slirp led to mitochondrial network disruption, mitochondrial fragmentation, and reduced respiratory capacity in skeletal muscle in mice. Second, SLIRP deficiency impaired muscle functionality, and reduced lifespan in flies. Third, SLIRP was identified as an exercise training- responsive downstream target of PGC1- \(\alpha\) . Fourth, SLIRP was needed for improving blood glucose regulation after exercise training in male mice. Fifth, exercise training restored the mitochondrial defects elicited by the lack of SLIRP on mitochondrial structure, respiration, and mtDNA- encoded OXPHOS, via mechanisms likely involving the upregulation of mitoribosome content and enhanced antioxidant defense. Finally, we established a translational foundation for our findings by substantiating the conservation of SLIRP's response to exercise training in four independent human cohorts integrating different exercise modalities, sex, health, and age conditions. + +<|ref|>text<|/ref|><|det|>[[115, 621, 882, 911]]<|/det|> +Our first finding establishes SLIRP as an hitherto unrecognized player in regulating mitochondria content, structure, and respiration in skeletal muscle. Our results are partially consistent with reports in heart, liver, and kidney Slirp KO17, 18, 52 showing that SLIRP forms a complex with LRPPRC and is crucial for the maintenance of the mt- mRNA reservoir. Accordingly, our observation of mitochondria with abnormal cristae in Slirp KO muscle aligns with transmission electron micrographs of Lrpprc KO heart tissue19. However, despite the stark similarities, mitochondrial respiration was compromised in Slirp KO skeletal muscle, but not in heart or liver18, underscoring SLIRP's functional importance within skeletal muscle relative to other tissues. In alignment, previous research further showed that Slirp KO skeletal muscle fibers have reduced sarcoplasmic reticulum Ca2+ storage capacity68. Thus, our present findings underscore the importance of SLIRP across various tissue types, with emphasis on its + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 130]]<|/det|> +importance to skeletal muscle mitochondrial morphology and respiration, and overall mitochondrial health. + +<|ref|>text<|/ref|><|det|>[[115, 152, 883, 503]]<|/det|> +Our second finding that SLIRP depletion compromised climbing ability, impaired starvation tolerance, and reduced lifespan in flies, illustrates the detrimental functional consequences of lacking SLIRP in the muscle tissue. The effects of mitochondrial proteins on modulating life span are equivocal. Some report that the depletion of specific OXPHOS genes reduces life span, whereas others report no effect or an increase on life span in model organisms69- 71. The "Mitochondrial Threshold Effect Theory"72 posits that mild mitochondrial dysfunction allows normal physiology in model organisms until a threshold is reached. Beyond that, severe mitochondrial dysfunction can lead to premature aging, developmental arrest or even death73- 75. In support, heterozygous depletion of Mclk1, which encodes an enzyme necessary for biosynthesis of the electron carrier coenzyme Q of OXPHOS, even increased lifespan in mice, but homozygous depletion was embryonically lethal75. Our findings support an important role for SLIRP and mitochondrial function in maintaining normal lifespan because we observed reduced survival in muscle- specific SLIRP KD flies. + +<|ref|>text<|/ref|><|det|>[[115, 522, 883, 872]]<|/det|> +Our third finding was that PGC- 1α1 regulated SLIRP protein, providing evidence for a new training- responsive PGC- 1α target in skeletal muscle. Interestingly, training- induced adaptations in mitochondrial proteins can occur independently of PGC- 1α1 as the main regulator76. In agreement with that, our fifth finding was that the training- induced improvements in mitochondrial function could bypass SLIRP deletion. These findings agree with other studies showing that exercise training can rescue mitochondrial dysfunction and elicit improvements in glucose homeostasis in mice and humans13, 48, 76, 77. Interestingly, SLIRP was required for the training- induced improvements in blood glucose regulation in trained male Slirp KO mice. While not explored in our study, this could be due to decreased intramyocellular insulin signalling78, reduced insulin- independent glucose uptake14, capillarization, and/or blood flow79. Female mice did not improve their glucose tolerance in response to exercise training. Sexual dimorphism in response to metabolic challenges, including exercise training, has been frequently reported in mouse models80, 81. Yet, in both sexes, SLIRP is largely + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 131]]<|/det|> +dispensable for exercise training- induced mitochondrial and metabolic adaptations in skeletal muscle, even though SLIRP is crucial for mitochondrial transcript stability in the basal state. + +<|ref|>text<|/ref|><|det|>[[115, 152, 883, 592]]<|/det|> +We were intrigued by the observation that exercise training could circumvent even \(60 - 80\%\) loss in mitochondrial- mRNA pools to upregulate mtDNA- encoded OXPHOS proteins. In other words, the beneficial effect of exercise training on mitochondrial oxidative phosphorylation and ATP provision is independent of a correction of mt- mRNA transcript levels and LRPPRC/SLIRP protein content. These findings give rise to the possibility that mt- mRNAs are produced in excess in vivo18. Excess mt- mRNA may allow for rapid engagement of mitochondrial protein synthesis in the event of sudden changes in energy demand. In response to exercise training, the mitoribosomal components, MRPL11 and 12S rRNA were upregulated, indicating increased capacity for mitochondrial protein synthesis. These adaptations to exercise training may help sustain protein translation, even in the presence of low mt- mRNA abundance. Exercise training also upregulated PRDX3, linked to peroxide scavenging. While MRPL11, 12S rRNA and PRDX3 may not directly control translation efficiency, their roles in the mitoribosomal makeup and improved redox state can influence the overall cellular environment where protein translation occurs. Albeit limited by not directly measuring mitochondrial translational rate, this working hypothesis is supported by findings in human skeletal muscle showing that mitochondrial protein translation is at the core of exercise training- induced benefits39. + +<|ref|>text<|/ref|><|det|>[[115, 612, 883, 841]]<|/det|> +Finally, our results likely are relevant to humans, as SLIRP and LRPPRC were consistently upregulated in human skeletal muscle in response to multiple exercise training modalities in both sexes. The effect of resistance exercise training was blunted for LRPPRC in patients with T2D. Interestingly, the age- related decline in muscle mitochondrial protein synthesis82 can be reversed by exercise training in elderly39, corroborating our findings that ET bypasses SLIRP to increase mitoribosomal translation. Importantly, our findings demonstrate that the synthesis of mtDNA- encoded OXPHOS proteins is not limited by transcript abundance in skeletal muscle – a feature that seems to be conserved as a fundamental mechanism26, 39. + +<|ref|>text<|/ref|><|det|>[[115, 863, 881, 910]]<|/det|> +Taken together, our findings not only unravel a new mechanism for the regulation of basal mitochondrial function in skeletal muscle, but also highlight an incredible exercise training – stimulated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 161]]<|/det|> +plasticity of mitochondria in skeletal muscle facing mitochondrial defects. Our findings underscore ET as a therapeutic intervention to combat mitochondrial dysfunction in genetic and lifestyle- induced muscle pathologies, including T2D. + +<|ref|>sub_title<|/ref|><|det|>[[117, 184, 217, 200]]<|/det|> +## Conclusions + +<|ref|>text<|/ref|><|det|>[[115, 222, 882, 390]]<|/det|> +In this work, we uncover a conserved role for mitochondrial transcript stability via SLIRP in basal mitochondrial structure and respiratory capacity in skeletal muscle. We identify SLIRP as a novel exercise- responsive downstream target of PGC- 1α that is conserved in human skeletal muscle. Because the defects in Slirp KO muscle could be reversed by ET, these findings add to the clinical relevance and add impetus to the important health- promoting role of ET in pathologies characterized by mitochondrial defects. + +<|ref|>sub_title<|/ref|><|det|>[[117, 453, 244, 469]]<|/det|> +## Figure Legends + +<|ref|>sub_title<|/ref|><|det|>[[117, 492, 820, 509]]<|/det|> +## Fig 1. Slirp knockout led to defects in mitochondrial structure and function, and reduced + +<|ref|>sub_title<|/ref|><|det|>[[117, 522, 184, 538]]<|/det|> +## lifespan + +<|ref|>text<|/ref|><|det|>[[115, 560, 880, 608]]<|/det|> +(A) SLIRP protein content across different wild-type tissues (n=4, female C57BL/6J, 12 weeks of age) assessed by immunoblotting. + +<|ref|>text<|/ref|><|det|>[[115, 629, 881, 707]]<|/det|> +(B) SLIRP and LRPPRC protein content in tibialis anterior muscle of Slirp knockout (KO) and littermate wildtype (WT) mice, and recombinant adeno-associated virus serotype 6 encoding SLIRP (rAAV6:SLIRP) and rAAV6:eGFP as control mice. + +<|ref|>text<|/ref|><|det|>[[115, 729, 881, 910]]<|/det|> +(C) SLIRP and LRPPRC protein content in tibialis anterior muscle of recombinant adeno-associated virus serotype 6 encoding SLIRP (rAAV6:SLIRP) and rAAV6:eGFP in the contralateral leg as control. +(D) Confocal microscopy of mitochondrial network structure using a mitochondrial membrane potential probe (TMRE+) in flexor digitorum brevis muscle fibers of Slirp KO and WT mice (n=4-5, 7-10 fibers per mouse). Corresponding fragmentation index are presented as super plots83; small symbols represent each fiber, large symbols represent the mean of fibers from each mouse that are color- and symbol- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 84, 500, 100]]<|/det|> +505 coded for each sex (○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[67, 123, 790, 141]]<|/det|> +506 (E) Transmission electron microscopy images of Slirp KO and WT gastrocnemius muscle. + +<|ref|>text<|/ref|><|det|>[[67, 162, 881, 270]]<|/det|> +507 (F and G) Quantification of percentage of damaged intramyofibrillar (IMF) mitochondria within total mitochondrial and relative volume of mitochondria. Small symbols refer to damaged or total mitochondria per fiber, large symbols of identical symbol refer to average of fibers per biological replicate (female Slirp KO and WT gastrocnemius muscle, n=4/group). + +<|ref|>text<|/ref|><|det|>[[67, 291, 881, 339]]<|/det|> +511 (H) Mitochondrial respiration in of Slirp KO and WT gastrocnemius muscle (n=4- 5/group, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[67, 361, 880, 409]]<|/det|> +513 (I) Fatty acid oxidation in isolated Slirp KO and WT soleus muscle at rest and in response to contraction (n=3- 6/group, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[67, 430, 880, 478]]<|/det|> +515 (J) RT- qPCR analysis of mitochondrial transcript levels in gastrocnemius of male Slirp KO and WT mice (n=6- 7/group). + +<|ref|>text<|/ref|><|det|>[[67, 500, 880, 577]]<|/det|> +517 (K- M) Climbing time during climbing assay, starvation tolerance and life span of control (Mef2- GAL4>GD control 6000), SLIRP1 (Mef2- GAL4>SLIRP1 RNAi 51019 GD) and SLIRP2 (Mef2- GAL4>SLIRP2 RNAi 23675 GD) knockdown flies. + +<|ref|>text<|/ref|><|det|>[[67, 600, 707, 618]]<|/det|> +520 Data are shown as mean ± SEM, including individual values, where applicable. + +<|ref|>text<|/ref|><|det|>[[67, 640, 881, 779]]<|/det|> +521 Geno, main effect of genotype; substrate, main effect of substrate addition; Substrate X Geno, interaction between genotype and substrate; Contraction, main effect of contraction. \*p<0.05, \*\*p<0.01, \*\*\*p<0.001, as per Unpaired Student's t test (D, J), Mann Whitney test (F, G), Two- way RM ANOVA with Šidák's multiple comparisons test (H, I), ordinary one- way ANOVA with Dunnett's multiple comparisons test (K), Log- rank (Mantel- Cox) test (L, M). + +<|ref|>sub_title<|/ref|><|det|>[[67, 800, 601, 818]]<|/det|> +## Fig 2. SLIRP is upregulated by ET and is a target of PGC-1α + +<|ref|>text<|/ref|><|det|>[[67, 840, 880, 886]]<|/det|> +527 (A) RT- qPCR analysis of Slirp transcript levels relative to Actb levels and Western blot analysis of SLIRP protein abundance in quadriceps with skeletal- muscle- specific transgenic expression of PGC- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 327, 101]]<|/det|> +1α1 (α1 OE; n=8- 11/group) + +<|ref|>text<|/ref|><|det|>[[115, 123, 880, 171]]<|/det|> +(B) Western blot analysis of SLIRP protein abundance in gastrocnemius with skeletal-muscle-specific transgenic expression of PGC-1α4 (α4 OE; n=6-9/group). + +<|ref|>text<|/ref|><|det|>[[115, 193, 880, 270]]<|/det|> +(C) RT-qPCR analysis of Slirp and Ppargc1α transcript levels relative to Hprt levels in gastrocnemius of muscle-specific PGC-1α knockout (PGC-1α mKO) and littermate control (WT) mice at rest and 3h after acute exercise bout (n=5-8/group). + +<|ref|>text<|/ref|><|det|>[[115, 292, 880, 340]]<|/det|> +(D) RT-qPCR analysis of Slirp and Ppargc1α transcript levels relative to Actb levels in WT quadriceps at rest, immediately after, 2 h, 6 h or 24 h after acute exercise bout (n=8/group). + +<|ref|>text<|/ref|><|det|>[[115, 362, 880, 410]]<|/det|> +(E) Western blot analysis of SLIRP and pAMPK T172 protein abundance in WT quadriceps at rest, immediately after, 2 h, 6 h or 24 h after acute exercise bout (n=8/group). + +<|ref|>text<|/ref|><|det|>[[115, 432, 880, 479]]<|/det|> +(F) Western blot analysis of SLIRP protein abundance in quadriceps of sedentary (SED) and 12-week ET PGC-1α mKO and WT mice (n=9-10/group). + +<|ref|>text<|/ref|><|det|>[[115, 501, 880, 548]]<|/det|> +(G) Western blot analysis of SLIRP protein abundance in quadriceps of muscle-specific AMPKα1 and -α2 double KO (mDKO) and WT mice treated with/without AICAR (n=4-7/group). + +<|ref|>text<|/ref|><|det|>[[115, 570, 881, 678]]<|/det|> +Data are means ± SEM, including individual values. Geno, main effect of genotype; Ex, main effect of acute exercise; Geno x Ex, interaction between genotype and acute exercise. \*p<0.05, \*\*\*p<0.001, as per Unpaired Student's t test (A), Two-way ANOVA with Šidák's multiple comparisons test (C, F, G), and ordinary one-way ANOVA with Dunnett's multiple comparisons test (D, E). + +<|ref|>sub_title<|/ref|><|det|>[[115, 700, 872, 747]]<|/det|> +## Fig. 3. SLIRP is dispensable for organismal adaptations to ET, yet, needed for improving blood glucose regulation after ET in male mice + +<|ref|>text<|/ref|><|det|>[[115, 769, 880, 864]]<|/det|> +(A) Experimental design of 10-week exercise training (ET) intervention. +(B) Average running distance of male and female ET Slirp KO or littermate control (WT) mice measured for 6 weeks (n=6-11/group, ○ (blue) male, ◇ (red) female) + +<|ref|>text<|/ref|><|det|>[[115, 878, 880, 896]]<|/det|> +(C) Average food intake of male and female sedentary (SED) and ET Slirp KO or WT mice measured + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 82, 688, 100]]<|/det|> +for 2 weeks after 4 weeks of running (n=3- 8, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[66, 121, 881, 262]]<|/det|> +(D- K) Blood glucose levels following a 4- hour fasting period before the glucose tolerance test (GTT). Glucose tolerance of male and female SED and ET Slirp KO or WT mice after 7- weeks of ET. iAUC of glycemic excursion in response to bolus of glucose \(2\mathrm{gkg}^{- 1}\) body weight (BW). Insulin response before (0 min) and following (20 min) the oral glucose challenge (n=6- 11/group, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[67, 282, 636, 300]]<|/det|> +Data are means ± SEM, including individual values where applicable. + +<|ref|>text<|/ref|><|det|>[[66, 321, 881, 460]]<|/det|> +Geno, main effect of genotype; ET, main effect of exercise training; Acute exercise, main effect of acute exercise but; Time X Geno ET, interaction between glucose bolus and genotype in the ET groups. \(^{*}\mathrm{p}< 0.05\) , \(^{**} \mathrm{p}< 0.01\) ; (E) WT ET vs. KO ET, \(^{*}\mathrm{p}< 0.05\) ; WT SED vs. WT ET, \(^{\#}\mathrm{p}< 0.05\) , as per Unpaired Student's t test (B), Two- way RM ANOVA with Sidak's multiple comparisons test in ET groups (E, I), Two- way ANOVA (C, D, F, G, H, J, K). + +<|ref|>sub_title<|/ref|><|det|>[[67, 481, 870, 529]]<|/det|> +## Fig 4. Exercise training reverses SLIRP-induced defects in muscle mitochondrial structure and respiration despite sustained reductions of mitochondrial transcripts + +<|ref|>text<|/ref|><|det|>[[66, 549, 881, 660]]<|/det|> +(A) Confocal microscopy of mitochondrial network structure using a mitochondrial membrane potential probe (TMRE+) in flexor digitorum brevis muscle fibers of sedentary (SED) and 10-week ET Slirp knockout (KO) mice and littermate controls (WT) (n=3-4, 7-10 fibers per mouse), and corresponding fragmentation index. SED data is also depicted in Fig. 1D. + +<|ref|>text<|/ref|><|det|>[[66, 680, 881, 728]]<|/det|> +(B) Mitochondrial respiration measured by Oroboros respirometry system in gastrocnemius of SED and ET Slirp KO and WT (n=2-6/per group, ○ male, ◇ female). SED data is also depicted in Fig. 1H. + +<|ref|>text<|/ref|><|det|>[[66, 749, 880, 797]]<|/det|> +(C) Western blot analysis of SLIRP and LRPPRC protein abundance in gastrocnemius of SED and 10-week ET Slirp KO and WT mice (n=6-11/group, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[66, 818, 880, 866]]<|/det|> +(D) Schematic of mtDNA- and nuclear DNA-encoded oxidative phosphorylation (OXPHOS) proteins analysed. + +<|ref|>text<|/ref|><|det|>[[66, 887, 880, 906]]<|/det|> +(E) RT-qPCR analysis of mitochondrial transcript levels in gastrocnemius of male SED and ET Slirp + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 82, 652, 101]]<|/det|> +578 KO and WT mice (n=6- 11/group). SED data is also depicted in Fig. 1J. + +<|ref|>text<|/ref|><|det|>[[117, 123, 880, 230]]<|/det|> +(F-H) Western blot analysis of (F) mtDNA-encoded proteins, MT-ND1 (CI), MT-ND3 (CI), MT-CO1 (CIV), MT-ATP6 (CV), and (G) nDNA-encoded proteins NDUFB8 (CI), SDHB (CII), UQCRC2 (CIII) and ATP5A (CV), and (H) representative blots in gastrocnemius of SED and ET Slirp KO and WT mice (n=6-11/group, ○ (blue) male, ◇ (red) female). + +<|ref|>text<|/ref|><|det|>[[117, 255, 881, 362]]<|/det|> +Data are means ± SEM, including individual values. Geno, main effect of genotype; ET, main effect of exercise training; Geno x ET, interaction between genotype and exercise training; Substrate, main effect of substrate. *p<0.05, **p<0.01, ***p<0.01 as per Two-way ANOVA with Šidák's multiple comparisons test (A, B, C, F, G). + +<|ref|>sub_title<|/ref|><|det|>[[117, 384, 822, 401]]<|/det|> +## Fig 5. Exercise training increases mitochondrial protein translation capacity by elevating + +<|ref|>sub_title<|/ref|><|det|>[[117, 415, 277, 429]]<|/det|> +### mitoribosome mass + +<|ref|>text<|/ref|><|det|>[[117, 453, 880, 620]]<|/det|> +(A-H) RT-qPCR analysis of 12S rRNA (A) and Western blot analysis of MRPL11 (B), rpS6 (C), PRDX3 (E), PRDX3 dimer/monomer ratio (F), PRDX2 (G), and representative blots in gastrocnemius of sedentary (SED) and exercise trained (ET) Slirp knockout (KO) and control littermate (WT) mice (n=6-11/group, ○ (blue) male, ◇ (red) female). Data are means ± SEM, including individual values. Geno, main effect of genotype; ET, main effect of exercise training; *p<0.05, **p<0.01, as per Two-way ANOVA (A-C, E-G). + +<|ref|>sub_title<|/ref|><|det|>[[117, 644, 870, 689]]<|/det|> +## Fig 6. In human skeletal muscle SLIRP and LRPPRC protein content are increased by exercise training + +<|ref|>text<|/ref|><|det|>[[117, 713, 880, 789]]<|/det|> +(A) Western blot analysis of SLIRP and LRPPRC protein in vastus lateralis of healthy young women (n=9/group, ○ Pre ET, □ Post ET)) after a 14-week controlled aerobic and strength exercise training (ET) intervention60. + +<|ref|>text<|/ref|><|det|>[[117, 813, 880, 858]]<|/det|> +(B) Immunoblotting of SLIRP and LRPPRC protein in vastus lateralis of healthy young men (n=6/group, ○ Pre ET, □ Post ET)) after a 6-week high intensity interval training63. + +<|ref|>text<|/ref|><|det|>[[117, 882, 880, 897]]<|/det|> +(C, D) Immunoblotting of SLIRP and LRPPRC protein (C) and RT-qPCR analysis of mitochondrial + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 130]]<|/det|> +transcript levels (D) in vastus lateralis of male young and older individuals after 12- week progressive resistance training \(^{61}\) (n=6- 11/group, (○ Pre ET, □ Post ET)). + +<|ref|>text<|/ref|><|det|>[[115, 153, 881, 260]]<|/det|> +(E- G) Immunoblotting of SLIRP, LRPPRC (E), and MT- CO2 (G) protein and RT- qPCR analysis of mitochondrial transcript levels (F) in vastus lateralis of male individuals, who were lean, obese or type 2 diabetic after high- intensity interval training \(^{62}\) (n=13- 16/group○ pre ET, □ post ET). Con, control; OB, obese; T2D, Type 2 diabetes. + +<|ref|>text<|/ref|><|det|>[[115, 283, 881, 389]]<|/det|> +Data presented as individual before- and- after values. ET, main effect of exercise training, ET x T2D, Interaction between exercise training and presence of type 2 diabetes in Con and T2D group; \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per paired Student's t test (A- C), Mixed- effects model (D, F), Two- way RM ANOVA with Šidák's multiple comparisons test (E, G). + +<|ref|>sub_title<|/ref|><|det|>[[116, 414, 291, 430]]<|/det|> +## Supplementary Fig. 1 + +<|ref|>text<|/ref|><|det|>[[115, 453, 880, 500]]<|/det|> +(A) Quantification of SLIRP protein content across different WT tissues (WT samples of n=4, 12 weeks of age) assessed by immunoblotting. + +<|ref|>text<|/ref|><|det|>[[115, 522, 881, 626]]<|/det|> +(B- H) Transmission electron microscopy quantification of the following parameters of mitochondrial area in the intermyofibrillar (IMF) region (B), aspect ratio (C), elongation (D), convexity (E), perimeter (F), sphericity (G), and diameter (H). Small circles refer to each fiber, large squares in similar color refer to average of fibers per biological replicate, (female, n=4). + +<|ref|>text<|/ref|><|det|>[[115, 650, 881, 757]]<|/det|> +(I- M) Western blot analysis of SDHB (CII), UQCRC2 (CIII), MT- CO1 (CIV), and ATP5A (CV) protein abundance in quadriceps of sedentary (SED) and 12- week exercise- training (ET) PGC- 1α mKO and control littermates (WT) mice (subset of total cohort, male, n=3). Representative blots. Coomassie was used as loading control. + +<|ref|>text<|/ref|><|det|>[[115, 781, 880, 856]]<|/det|> +Data are means ± SEM, including individual values. ET, main effect of exercise training; Geno, main effect of genotype. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per Unpaired Student's t test (B- D, F- H), Mann Whitney test (E), Two- way ANOVA (I- L). + +<|ref|>sub_title<|/ref|><|det|>[[116, 881, 291, 897]]<|/det|> +## Supplementary Fig. 2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 160]]<|/det|> +(A-C) Body weight, fat mass relative to body weight, lean mass relative to body weight of male and female SED and ET Slirp KO or WT mice after 6-weeks of ET (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +<|ref|>text<|/ref|><|det|>[[115, 182, 880, 230]]<|/det|> +(D) Gastrocnemius, tibialis anterior (TA) and heart weight relative to tibia length of female SED and ET Slirp KO or WT mice after 10-weeks of ET (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +<|ref|>text<|/ref|><|det|>[[115, 251, 880, 329]]<|/det|> +(E-H) Exercise capacity of male and female SED and ET Slirp KO or WT mice after 8-weeks of ET and corresponding blood lactate levels before and after exercise bout (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +<|ref|>text<|/ref|><|det|>[[115, 351, 880, 430]]<|/det|> +(I) Fatty acid oxidation in isolated soleus of male and female SED and ET Slirp KO or WT mice after 10-weeks of ET. WT SED male/female, n=3/4, WT ET male/female, n=4/3, KO SED male/female, n=4/9, KO ET male/female, n=4/7. SED data is also depicted in Fig. 1H. + +<|ref|>text<|/ref|><|det|>[[115, 451, 880, 500]]<|/det|> +(J-L) RT-qPCR analysis of mtDNA levels and citrate synthase activity in gastrocnemius of SED and ET Slirp KO and WT mice (n=6-11/group, \(\circ\) (blue) male, \(\diamond\) (red) female). + +<|ref|>text<|/ref|><|det|>[[115, 521, 881, 660]]<|/det|> +Data are means \(\pm\) SEM, including individual values where applicable. ET, main effect of exercise training; Geno, main effect of genotype; Acute exercise, main effect of acute exercise; Geno in ET group, main effect of genotype within exercise training group; Contraction, main effect of contraction. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , as per Two-way ANOVA within sex (A-D), Two-way ANOVA (E, G, J-L), Two-way RM ANOVA with Sidak's multiple comparisons test (F, H, I). + +<|ref|>sub_title<|/ref|><|det|>[[116, 682, 292, 699]]<|/det|> +## Supplementary Fig. 3 + +<|ref|>text<|/ref|><|det|>[[115, 720, 881, 858]]<|/det|> +(A) Western blot analysis of NDUFB8 (CI), SDHB (CII), UQCRC2 (CIII), ATP5A (CV) and ATP5B (CV) protein in vastus lateralis of male individuals, who were lean (pre/post, n=16/16), obese (pre/post, n=15/15) or type 2 diabetic (pre/post n=13/13) after high-intensity interval training62, and representative images. Coomassie was used as control. Data presented as individual before-and-after values. ET, main effect of exercise training. \(^{*}\mathrm{p}< 0.05\) , \(^{**}\mathrm{p}< 0.01\) , \(^{***}\mathrm{p}< 0.001\) , Two-way RM ANOVA (A). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 125, 252, 141]]<|/det|> +## Fly maintenance + +<|ref|>text<|/ref|><|det|>[[115, 162, 881, 330]]<|/det|> +In standard conditions, the flies were maintained on a SYA medium containing \(0.5\%\) (w/v) agar, \(2.4\%\) (v/v) nipagin, \(0.7\%\) (v/v) propionic acid, \(10\%\) (w/v) dry baker's yeast and \(5\%\) (w/v) sucrose. The experiments took place at \(+25^{\circ}C\) , \(65\%\) humidity under a 12h:12h light:dark cycle. RNAi lines for SLIRP1 (51019 GD), SLIRP2 (23675 GD), and the GD library control line (w1118, 60000 GD) were obtained from Vienna Drosophila Resource Center. Mef2- GAL4 line was obtained from Bloomington Stock Center to generate muscle- specific SLIRP KD flies. + +<|ref|>sub_title<|/ref|><|det|>[[118, 353, 390, 370]]<|/det|> +## Negative geotaxis (climbing) assay + +<|ref|>text<|/ref|><|det|>[[115, 390, 882, 619]]<|/det|> +Drosophila has a natural tendency to climb upwards, also known as negative geotaxis. The negative geotaxis, or climbing, assay was performed as previously described84. Briefly, 20 male Drosophila flies aged 5- 6 days were transferred into empty vials (25 x 95mm), and another empty vial was taped on top to create a \(25 \times 190\mathrm{mm}\) cylinder. Six cylinders were set side by side into an apparatus. The apparatus was tapped down 5 times to make all the flies fall to the bottom of the cylinder. The climbing assay was measured in technical triplicate for each biological replicate, with 90 seconds allowed for each climb. The assay was recorded with a camera and videos were analysed to determine the time for half of the flies in the vial to reach halfway point (95mm) in the cylinder. + +<|ref|>sub_title<|/ref|><|det|>[[118, 641, 713, 659]]<|/det|> +## Starvation assay using the Drosophila Activity Monitoring System (DAMS) + +<|ref|>text<|/ref|><|det|>[[115, 680, 881, 818]]<|/det|> +Four- day old male flies were individually housed in monitor tubes with an outside diameter of 5mm (PPT5x65, Trikinetics, Waltham, MA). For starvation analysis, the tubes contained \(1\%\) agar in water. 32 tubes per monitor (DAM2 Drosophila Activity Monitor, Trikinetics, Waltham, MA) were set up, one monitor per genotype. For the starvation analysis, flies were kept in the DAM system until the last fly had died. + +<|ref|>sub_title<|/ref|><|det|>[[118, 841, 236, 857]]<|/det|> +## Lifespan assay + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 192]]<|/det|> +For lifespan assays, flies were mated at a controlled density of 25 female and 10 male flies, respectively. One- day old male flies were collected into vials (10 male flies per vial, with 10 vials per genotype). The flies were kept on a SYA diet in Drosflippers (http://www.drosflipper.com/). Flies were flipped onto fresh food every second day and deaths were scored during the transfer. + +<|ref|>sub_title<|/ref|><|det|>[[115, 215, 161, 230]]<|/det|> +## Mice + +<|ref|>sub_title<|/ref|><|det|>[[115, 253, 460, 270]]<|/det|> +## PGC-1α1 transgenic mouse (PGC-1α1 OE) + +<|ref|>text<|/ref|><|det|>[[115, 293, 880, 340]]<|/det|> +The generation of the model, quality, and specificity of the overexpression has been described previously35, of which we analyzed quadriceps muscle samples. + +<|ref|>sub_title<|/ref|><|det|>[[115, 361, 460, 378]]<|/det|> +## PGC-1α4 transgenic mouse (PGC-1α4 OE) + +<|ref|>text<|/ref|><|det|>[[115, 401, 880, 448]]<|/det|> +The generation of the model, quality, and specificity of the overexpression has been described previously33, of which we analyzed gastrocnemius muscle samples. + +<|ref|>sub_title<|/ref|><|det|>[[115, 470, 586, 487]]<|/det|> +## Skeletal muscle-specific PGC-1α knockout (PGC-1α mKO) + +<|ref|>text<|/ref|><|det|>[[115, 509, 881, 586]]<|/det|> +The generation of the model, quality, and specificity of the KO has been described previously44. Male muscle- specific PGC- 1α MKO mice and littermate control mice homozygous for loxP inserts (lox/lox) were generated by crossbreeding myogenin- Cre mice with loxP flanked- Pgc- 1α mice. + +<|ref|>text<|/ref|><|det|>[[115, 608, 881, 685]]<|/det|> +For the acute exercise bout, Tibialis anterior muscles of PGC- 1α MKO mice and littermate control mice were collected 3 hours after one bout of equal distance treadmill running (1.4 km) at \(10^{\circ}\) incline and \(60\%\) of their individual maximal running speed achieved by a graded treadmill running test85. + +<|ref|>text<|/ref|><|det|>[[115, 708, 881, 877]]<|/det|> +For the ET intervention, mice were single- housed with or without access to in- cage running wheels from 8 to 20 weeks old. Running distance and duration were monitored by a regular cycle computer. PGC- 1α MKO mice tended to run less than lox/lox, thus running wheels of lox/lox mice were occasionally blocked to ensure equal running distance. On average, mice ran 25 km/week. The running wheels were blocked for all mice 24 hours prior to euthanization and harvest of quadriceps muscle in the morning in the fed state. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 465, 101]]<|/det|> +## AAV6 Vector construction and preparation + +<|ref|>text<|/ref|><|det|>[[117, 123, 880, 170]]<|/det|> +The AAV6 vector for SLIRP overexpression and ultra- purified eGFP control AAV6 virus were manufactured by VectorBuilder Inc. (Shenandoah, Texas, USA; AAV6SP(VB190219- 1010jvy)- C). + +<|ref|>sub_title<|/ref|><|det|>[[117, 193, 590, 209]]<|/det|> +## Transfection of tibialis anterior muscle using rAAV6 vector + +<|ref|>text<|/ref|><|det|>[[117, 231, 881, 370]]<|/det|> +Viral particles were diluted in Gelofusine (B. Braun, Germany) to a dosage of \(5 \times 10^{9}\) vector genomes at a volume of \(30 \mathrm{uL}\) per injection. For muscle- specific delivery of rAAV6 vectors, 12- week- old mice were placed under general anaesthesia (2% isoflourane in \(\mathrm{O_2}\) ) and mice were injected intramuscularly with rAAV6:SLIRP in the TA muscle of one leg and control rAAV6:eGFP (empty vector) in the contralateral leg. Muscles were harvested 4 weeks after rAAV6 administration. + +<|ref|>sub_title<|/ref|><|det|>[[118, 393, 352, 409]]<|/det|> +## Time course of acute exercise + +<|ref|>text<|/ref|><|det|>[[117, 431, 881, 507]]<|/det|> +Twelve- week old female C57BL/6J mice (n=8 for each time point) were subjected to an acute exercise bout (1h running, 60% of maximal running intensity, 15° incline). Quadriceps muscle was harvested from rested and exercised mice immediately after, 2h, 6h, and 24 h after the acute exercise bout. + +<|ref|>sub_title<|/ref|><|det|>[[117, 531, 439, 547]]<|/det|> +## SLIRP knockout mouse (Slirp KO mice) + +<|ref|>text<|/ref|><|det|>[[117, 570, 881, 646]]<|/det|> +The generation of the model, quality, and specificity of the KO has been described previously18. Sperm of homozygous Slirp KO mice kindly provided by Nils- Göran Larsson and re- derived offspring was backcrossed to C57BL/6N background in our own animal facilities. + +<|ref|>text<|/ref|><|det|>[[117, 670, 881, 777]]<|/det|> +All mice were maintained under a 12:12 h light/dark photocycle at \(22 \pm 2^{\circ}\mathrm{C}\) with nesting material. The female mice were group- housed (except during the ET intervention), whereas the male mice were single- housed. All mice received a rodent chow diet (Altromin no. 1324; Chr. Pedersen, Denmark) and water ad libitum. + +<|ref|>text<|/ref|><|det|>[[117, 800, 881, 907]]<|/det|> +Mouse genotyping of Slirp KO and WT mice was performed as previously described86 using qPCR on DNA from ear punches with the following primers: WT; 5'- AGAAGGGAAT CCACAGGATA GGACA- 3' and 5'- GCTTTATTCC TAGTCTGGC CTTGTT- 3', KO; 5'- AGAAGGGAAT CCACAGGATA GGACA- 3' and 5'- CGCCGTATAA TGTATGCTAT ACGAAGTT- 3'. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[67, 84, 454, 101]]<|/det|> +## 10-week ET intervention in Slirp KO mice + +<|ref|>text<|/ref|><|det|>[[115, 122, 881, 291]]<|/det|> +For 10- week voluntary wheel- running exercise training interventions, 16- 18- week- old Slirp KO and littermate mice were randomized into test groups with or without access to in- cage running wheels (Tecniplast activity cage, wheel diameter: \(23\mathrm{cm}\) ; Tecniplast, Buguggiate VA, Italy). The experimental design is schematically illustrated in Fig. 3A. Running distance was monitored by a regular cycle computer prior to the metabolic tests for 6 weeks. Running wheels were locked \(12\mathrm{h}\) prior to terminal procedures to avoid effects of acute exercise bouts. + +<|ref|>sub_title<|/ref|><|det|>[[67, 314, 262, 330]]<|/det|> +## Body composition + +<|ref|>text<|/ref|><|det|>[[115, 353, 880, 399]]<|/det|> +Total, fat, and lean body mass was measured by nuclear magnetic resonance using an EchoMRITM (USA). + +<|ref|>sub_title<|/ref|><|det|>[[67, 422, 570, 439]]<|/det|> +## Glucose tolerance test (GTT) and plasma insulin analysis + +<|ref|>text<|/ref|><|det|>[[115, 460, 881, 567]]<|/det|> +We subjected the Slirp KO and WT mice to a GTT at 7 weeks of the training intervention study (as illustrated in Fig. 3A). The GTT was executed after a \(5\mathrm{h}\) fasting period (7:00 a.m.- 12:00 p.m.). Resting blood samples were taken from the tail \(30\mathrm{min}\) prior to the intraperitoneal injection of d- mono- glucose (2 g/kg body weight). Tail blood glucose was measured after 0, 20, 40, 60, and 90 min of injection. + +<|ref|>text<|/ref|><|det|>[[115, 590, 881, 758]]<|/det|> +To measure glucose- stimulated plasma insulin concentration at time- point (min) 0 and 20, tail vein blood samples were collected in a capillary- tube (50 \(\mu \mathrm{l}\) ), centrifuged at \(14,200\mathrm{g}\) for \(5\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) , plasma collected and stored at \(- 80^{\circ}\mathrm{C}\) . Insulin concentration was determined in duplicates using the Ultra- Sensitive Mouse Insulin ELISA Kit (#80- INSTRU- E10; ALPCO Diagnostics) to the manufacturer's instructions. The incremental area under the curve (iAUC) from the basal blood glucose concentration was determined using the trapezoid rule. + +<|ref|>sub_title<|/ref|><|det|>[[117, 781, 297, 797]]<|/det|> +## Exercise capacity tests + +<|ref|>text<|/ref|><|det|>[[115, 820, 881, 896]]<|/det|> +Slirp KO and WT mice were acclimatized to the treadmill on 3 consecutive days by running at a speed of \(0.16\mathrm{m / s}\) and incline of \(10^{\circ}\) for \(5\mathrm{min}\) the first day and \(10\mathrm{min}\) the following days. Prior to the running test and with a \(1\mathrm{h}\) delay, 2 blood samples were drawn from the tail before the test for pre- exercise blood + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 222]]<|/det|> +glucose and blood lactate measurements. Afterwards, the mice ran at \(0.16\mathrm{m / s}\) for 5 minutes followed by a gradual increase in speed every minute with \(0.02\mathrm{m / s}\) until exhaustion. Once the mouse reached its maximal running capacity, 2 blood samples were immediately drawn from the tail for post- exercise blood glucose and blood lactate measurements. The test was stopped when the mouse failed to keep up with the treadmill despite motivational efforts by the researcher. + +<|ref|>sub_title<|/ref|><|det|>[[118, 244, 346, 260]]<|/det|> +## TMRE staining in live fibers + +<|ref|>text<|/ref|><|det|>[[115, 281, 882, 542]]<|/det|> +Flexor digitorum brevis muscles (FDB) were dissected and then incubated for 2 hours in serum- free \(\alpha\) - MEM (22571- 020, Gibco) containing \(1.9\mathrm{mg / ml}\) collagenase type 1 from clostridium histolyticum (C0130, Sigma- Aldrich) and \(1\mathrm{mg / ml}\) bovine serum albumin (9048- 46- 8, Sigma- Aldrich) at \(37^{\circ}\mathrm{C}\) on a rotator. After collagenase treatment, muscles were incubated in \(\alpha\) - MEM containing \(10\%\) fetal bovine serum (26050- 70, Gibco) and subjected to mechanical dissociation using fire- polished Pasteur pipettes. Single muscle fibers were seeded in \(35\times 14\mathrm{mm}\) glass- bottom microwell dishes (P35G- 1.5- 14- C, MatTek Corporation) coated with \(4\mathrm{uL}\) of Engelbreth- Holm- Swarm murine sarcoma ECM gel (E1270, Merck). Muscle fibers were kept in \(\alpha\) - MEM containing \(5\%\) fetal bovine serum in a cell incubator ( \(37^{\circ}\mathrm{C}\) , \(5\%\) CO2) for at least \(16\mathrm{h}\) prior to the experiments. + +<|ref|>text<|/ref|><|det|>[[115, 562, 882, 731]]<|/det|> +For determination of mitochondrial membrane potential ( \(\Delta \Psi\) mitochondrial) and mitochondrial network morphology, the fibers were incubated in \(20\mathrm{mM}\) tetramethylrhodamine and ethyl ester (TMRE+, Life Technologies) dissolved in Krebs Ringer buffer ( \(145\mathrm{mMNaCl}\) , \(5\mathrm{mM}\) KCl, \(1\mathrm{mM}\) CaCl2, \(1\mathrm{mM}\) MgCl2, \(5.6\mathrm{mM}\) glucose, \(20\mathrm{mM}\) HEPES, pH 7.4) for 30 minutes before imaging. Confocal images were collected using a C- Apochromat \(\times 40\) , \(1.2\mathrm{NA}\) water immersion objective lens on an LSM 980 confocal microscope (Zeiss) driven Zeiss Zen Blue 3. + +<|ref|>text<|/ref|><|det|>[[115, 752, 882, 890]]<|/det|> +Mitochondrial network analysis was performed semi- automatically in ImageJ (National Institute of Health, USA). At least two nucleus- free regions per fiber were analyzed blindly. Before segmentation, the background was subtracted, and the pixels were averaged using a value of mean=2. The images were segmented, and particle analyses revealed the relative area of the TMRE+ signal compared to the total area (% mitochondrial area). The fragmentation index was calculated as the number of objects in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 131]]<|/det|> +relation to the total area covered by the dye. The ImageJ 'Red Hot' lookup table was used to visualize the images. + +<|ref|>sub_title<|/ref|><|det|>[[117, 154, 459, 171]]<|/det|> +## Transmission electron microscopy analysis + +<|ref|>text<|/ref|><|det|>[[115, 191, 883, 512]]<|/det|> +A small longitudinal section ( \(< 2\mathrm{mm}\) ) of the red portion of the gastrocnemius muscle tissue was fixed by immersion \(2\%\) glutaraldehyde in \(0.05\mathrm{M}\) Phosphatebuffer, pH 7.4 and stored at \(4^{\circ}\mathrm{C}\) . The samples were rinsed four times in \(0.1\mathrm{M}\) sodium cacodylate buffer, pH 7.4, and post- fixed with \(1\%\) osmium tetroxide \(\mathrm{(OsO_4)}\) and \(1.5\%\) potassium ferrocyanide \(\mathrm{[K_4Fe(CN)_6]}\) in \(0.1\mathrm{M}\) sodium cacodylate buffer, pH 7.4 for \(90\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) . The samples were then rinsed twice and dehydrated through a graded mixture of alcohol at \(4 - 20^{\circ}\mathrm{C}\) , infiltrated with graded mixtures of propylene oxide and Epon at \(20^{\circ}\mathrm{C}\) , and embedded in \(100\%\) Epon at \(30^{\circ}\mathrm{C}\) , as previously described (Nielsen et al., 2011). The samples were cut in \(60\mathrm{nm}\) longitudinal sections using a Leica Ultracut UCT ultramicrotome. The sections were contrasted with uranyl acetate and lead citrate, and subsequently examined and image recorded in a CM100 TEM (Philips, Eindhoven, The Netherlands) equipped with a Veleta camera and the iTem software package (Olympus, Hamburg, Germany) with a resolution of \(2048\times 2048\) pixels. + +<|ref|>text<|/ref|><|det|>[[115, 531, 883, 791]]<|/det|> +A mean of eight fibers per sample (7- 9) was included from the sectioned samples, and from each fiber, 24 images were obtained at x13,500 magnification. The imaging was performed in a randomized systematic order including 12 images from the subsarcolemmal (SS) region, and 6 from both the superficial and central region of the intermyofibrillar (IMF) space. The Z- disc width was measured once on all IMF images and the mean from all fibers within one sample was calculated to determine the fiber type. The categorization of fiber types was based on previous reports from observations \(^{43}\) . The images were analyzed by two genotype- blinded investigators and all further analyses were performed by the same blinded investigator. The quantification of mitochondrial morphology was done using the Radius EM Imaging Software (Emsis GmbH, Radius 2.0). + +<|ref|>text<|/ref|><|det|>[[115, 812, 881, 859]]<|/det|> +Several measurements were included in this study: 1) Mitochondrial number: Total number of mitochondrial profiles (count). 2) Mitochondrial volume fraction: IMF mitochondrial volume is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 131]]<|/det|> +annotated as percentage of mitochondria covering the IMF space. 3) Damaged mitochondria: Mitochondria were categorized as damaged when they were swollen, vacuolated, or empty. + +<|ref|>sub_title<|/ref|><|det|>[[117, 154, 525, 171]]<|/det|> +## Mitochondrial respiration in gastrocnemius muscle + +<|ref|>text<|/ref|><|det|>[[115, 183, 883, 562]]<|/det|> +In a subset of mice, mitochondrial respiratory capacity was measured in permeabilized gastrocnemius skeletal muscle fibers as previously described87. In brief, gastrocnemius muscle was rinsed from fat and connective tissue and separated into small fiber bundles. Fiber bundles were permeabilized with saponin (50 \(\mu \mathrm{g} / \mathrm{ml}\) ) in BIOPS buffer for 30 min, followed by a 20 min wash in MiR05 buffer on ice. Mitochondrial respiration was measured in duplicate under hyperoxic conditions at \(37^{\circ}\mathrm{C}\) using high resolution respirometry (Oxygraph- 2k, Oroboros Instruments, Innsbruck, Austria). The following protocol was applied: Leak respiration was assessed by addition of malate (5 mM) and pyruvate (5 mM), followed by adding three different concentrations of ADP (0.01 mM, 0.25 mM and 5 mM; for the final concentration of ADP, 3 mM of magnesium (Mg) was added as well) to measure complex I linked respiratory capacity. 10 mM Glutamate was added to measure maximal complex I linked respiratory capacity followed by 10 mM succinate to measure complex I+II linked respiratory capacity. Finally, 5 \(\mu \mathrm{M}\) Antimycin A was added to inhibit complex III in the electron transport chain. Pooled data for both sexes are shown as no sex- specific differences were detected. + +<|ref|>sub_title<|/ref|><|det|>[[117, 575, 666, 591]]<|/det|> +## Contraction-stimulated fatty acid oxidation in isolated soleus muscles + +<|ref|>text<|/ref|><|det|>[[115, 603, 883, 893]]<|/det|> +Contraction- stimulated exogenous palmitate oxidation in isolated soleus muscle from Slirp KO and WT mice was measured as previously described88. In brief, excised soleus muscles from mice anesthetized with pentobarbital were mounted at resting tension ( \(\sim 5 \mathrm{mN}\) ) in 15 ml vertical incubation chambers with a force transducer (Radnoti, Monrovia, CA) containing \(30^{\circ}\mathrm{C}\) carbonated (95% \(\mathrm{O}_2\) and 5% \(\mathrm{CO}_2\) ) Krebs- Henselieit Ringer buffer (KRB), \(\mathrm{pH} = 7.4\) , supplemented with 5 mM glucose, 2% fatty acid- free BSA, and 0.5 mM palmitate. After \(\sim 20 \mathrm{min}\) of pre- incubation, the incubation buffer was refreshed with KRB additionally containing [1- 14C]- palmitate (0.0044 MBq/ml; Amersham BioSciences, Buckinghamshire, U.K.). To seal the incubation chambers, mineral oil (Cat. No. M5904, Sigma- Aldrich) was added on top. Exogenous palmitate oxidation was measured simultaneously at rest and during 25 min contractions (18 trains/min, 0.6 s pulses, 30 Hz, 60 V). After incubation, incubation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 222]]<|/det|> +buffer and muscles were collected to determine the rate of palmitate oxidation as previously described87. 89, 90. Palmitate oxidation was determined as \(\mathrm{CO}_2\) production (complete FA oxidation) and acid-soluble metabolites (representing incomplete FA oxidation). As no difference was observed in complete and incomplete FA oxidation between genotypes, palmitate oxidation is presented as a sum of these two forms. + +<|ref|>sub_title<|/ref|><|det|>[[118, 235, 440, 252]]<|/det|> +## Lysate preparation and immunoblotting + +<|ref|>text<|/ref|><|det|>[[115, 273, 881, 381]]<|/det|> +Lysate preparation and immunoblotting of vastus lateralis muscles samples of HIIT in young, healthy men were performed as described by Hostrup et al.91, whereas lysate preparation and immunoblotting of vastus lateralis muscles of HIIT in individuals, who were lean, obese or T2D were performed as described by Kruse et al92. + +<|ref|>text<|/ref|><|det|>[[115, 402, 883, 722]]<|/det|> +To preserve and assess PRDX3 dimer to monomer ratio, freshly harvested mouse gastrocnemius muscle tissue was incubated for 10 min in ice- cold \(100\mathrm{mM}\) N- Ethylmaleimide (NEM) diluted in PBS. NEM was subsequently aspirated and the tissue homogenized for 1 min at \(30\mathrm{Hz}\) using a TissueLyser II bead mill (QIAGEN, USA) in ice- cold homogenization buffer [10% glycerol, 1% NP- 40, \(20\mathrm{mM}\) sodium pyrophosphate, \(150\mathrm{mM}\) NaCl, \(50\mathrm{mM}\) Hepes (pH 7.5), \(20\mathrm{mM}\) \(\beta\) - glycerophosphate, \(10\mathrm{mM}\) NaF, 2 mM phenylmethylsulfonyl fluoride, \(1\mathrm{mM}\) EDTA (pH 8.0), \(1\mathrm{mM}\) EGTA (pH 8.0), \(2\mathrm{mM}\) \(\mathrm{Na}_3\mathrm{VO}_4\) , leupeptin ( \(10\mu \mathrm{g}\mathrm{ml}^{- 1}\) ), aprotinin ( \(10\mu \mathrm{g}\mathrm{ml}^{- 1}\) ), and \(3\mathrm{mM}\) benzamidine]. Lysate preparation and immunoblotting of all remaining mouse tissues and human vastus lateralis muscle samples of aerobic and strength ET in women60, and resistance training in the young and elderly cohorts93 were performed as described in23. Immunoblotting of the NEM- treated samples was performed under non- denaturing conditions. The primary antibodies used are listed in Supplementary Table 2. + +<|ref|>text<|/ref|><|det|>[[115, 742, 881, 881]]<|/det|> +Coomassie Brilliant Blue staining was used as a control to assess total protein loading and transfer efficiency94 by quantifying the whole lane and for each sample set, a representative membrane from the immunoblotting is shown. The same Coomassie brilliant blue staining is presented for proteins analyzed when derived from the same sample set. Band densitometry was carried out using Image Lab (version 4.0). For each set of samples, a standard curve was loaded to ensure quantification within the linear + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 85, 880, 101]]<|/det|> +855 range for each protein probed for. For immunoblotting in Fig. 5, pooled data for both sexes are shown + +<|ref|>text<|/ref|><|det|>[[60, 117, 578, 129]]<|/det|> +856 as no sex-specific differences were detected, unless specified. + +<|ref|>table_caption<|/ref|><|det|>[[123, 157, 837, 170]]<|/det|> +Supplementary Table 2: List of primary antibodies used during Western blotting analysis + +<|ref|>table<|/ref|><|det|>[[115, 171, 860, 712]]<|/det|> + +
ProteinManufacturerIdentifierDilution
p-AMPK T172Cell SignalingCat# 2531; RRID:AB_330330
Cat# A-21351;
RRID:AB_221512		1:1000,2%
Skim milk
1:2000,2%
Skim milk
1:1000,2%
Skim milk
ATP5F1BThermoFisherCat# 2867; RRID:AB_22329461:1000,2%
Skim milk
Hexokinase IICell SignalingCat# ab97505;
RRID:AB_10688419		1:1000,2%
Skim milk
1:1000,2%
LRPPRCAbcam1:1000,2%
Skim milk
MRPL11Cell SignalingCat# 2066; RRID:AB_21455981:1000,2%
MT-ATP6ThermoFisher
Scientific		Cat# PA5-109432;
RRID:AB_2854843
Cat# ab110413;
RRID:AB_2629281
Cat# ab181848;
RRID:AB_2687504
Cat# ab192306;
RRID:AB_2943449
Cat#ab109367;		1:1000,3% BSA
1:5000,3%
Skim milk
1:1000,2%
Skim milk
1:1000,2%
1:1000,2%
Skim milk
1:100,2%
Skim milk
MT-CO2Abcam1:1000,2%
MT-ND1AbcamSkim milk
MT-ND3Abcam1:1000,2%
PRDX2AbcamRRID:AB_10862524
Cat#ab73349;
RRID:AB_1860862		1:1000,2%
Skim milk
1:1000,2%
PRDX3Abcam1:1000,2%
rpS6Cell Signaling
Santa Cruz		Cat# 2217, RRID:AB_331355
Cat# sc-514508; RRID:
AB_2943449		Skim milk
1:500,2% Skim
SLIRP (mouse)BiotechnologyCat# 703632;milk
SLIRP (human)Thermo Fisher
Scientific		RRID:AB_2784591
Cat# ab110411;
RRID:AB_2756818		1:2000,2%
Skim milk
1:1000,2%
Total OXPHOS Human
WB Antibody Cocktail		AbcamSkim milk
Total OXPHOS Rodent
WB Antibody Cocktail		AbcamCat# ab110413;
RRID:AB_2629281		1:5000,2%
Skim milk
+ +<|ref|>text<|/ref|><|det|>[[60, 710, 92, 720]]<|/det|> +857 + +<|ref|>text<|/ref|><|det|>[[60, 752, 92, 761]]<|/det|> +858 + +<|ref|>title<|/ref|><|det|>[[118, 789, 369, 801]]<|/det|> +# RNA extraction and RT-qPCR + +<|ref|>text<|/ref|><|det|>[[118, 828, 880, 841]]<|/det|> +Fig. 1J, 4F: RNA was extracted from ~20 mg of pulverized whole mouse gastrocnemius muscle using + +<|ref|>text<|/ref|><|det|>[[118, 858, 880, 871]]<|/det|> +TRIzolTM reagent (Invitrogen) following the manufacturer's instructions (tissue homogenization was + +<|ref|>text<|/ref|><|det|>[[118, 888, 880, 901]]<|/det|> +performed using the MP Bio lysis system) and treated with the TURBO DNA-freeTM Kit (Ambion) to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 281]]<|/det|> +remove contaminating DNA. For RT- qPCR expression analysis, cDNA was reversed transcribed from \(0.85 \mu \mathrm{g}\) total RNA using the High- Capacity cDNA Reverse Transcription Kit (Applied Biosystems) in the presence of RNase Block (Agilent). The qPCR was performed in a QuantStudio 6 Flex Real- Time PCR System (Life Technologies), using TaqMan™ Universal Master Mix II, with UNG (Applied Biosystems) to quantify mitochondrial transcripts (mitochondrial- rRNAs and mt- mRNAs). The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method, comparing the Ct values of mitochondrial transcripts to that of the beta- actin reference gene for normalization. + +<|ref|>text<|/ref|><|det|>[[115, 303, 882, 501]]<|/det|> +Fig. 2A, C, D: Total RNA was extracted from pulverized quadriceps muscle using an adapted guanidinium thiocyanate- phenol- chloroform extraction method. Reverse transcription to cDNA was performed as previously described46. Real- time qPCR was performed in triplicate with QuantStudio 7 Flex Real- Time PCR System (Applied Biosystems, Waltham, MA). Cycle threshold (Ct) was converted to a relative amount using a standard curve derived from a serial dilution of a representative pooled samples run together with the samples of interest. Beta- actin or Hprt mRNA was used for normalization of target mRNA levels. + +<|ref|>text<|/ref|><|det|>[[115, 522, 882, 660]]<|/det|> +Fig. 6J- M: Total RNA extraction from vastus lateralis muscle and reverse transcription to cDNA was performed as previously described61. The qPCR was performed in a QuantStudio 6 Flex Real- Time PCR System (Life Technologies), using TaqMan™ Universal Master Mix II (Applied Biosystems) to quantify mt- mRNAs. The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method, comparing the Ct values of mitochondrial transcripts to that of the 18S rRNA reference gene for normalization. + +<|ref|>text<|/ref|><|det|>[[115, 682, 882, 850]]<|/det|> +Fig. 6Q- T: Total RNA was extracted from skeletal muscle biopsies using TRI Reagent (Sigma- Aldrich) following the manufacturer's instructions. cDNA was reversed transcribed using the High- Capacity cDNA Reverse Transcription Kit (Applied Biosystems), while the qPCR was performed on an Aria Mx (Agilent) using a TaqMan™ Universal Master Mix II (Applied Biosystems). The gene expression levels were determined using the \(\Delta \Delta \mathrm{Ct}\) method with the mRNA levels being normalized to the geometric mean of PPIA and B2M. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 87, 878, 101]]<|/det|> +888 Primers or Taqman probes or Taqman expression assays used for mRNA levels measured in Fig. 2, + +<|ref|>text<|/ref|><|det|>[[60, 118, 525, 130]]<|/det|> +889 Fig. 4 and Fig. 6 are shown in Supplementary Table 3. + +<|ref|>table_caption<|/ref|><|det|>[[125, 157, 750, 170]]<|/det|> +Supplementary Table 3: List of primers or Taqman probes used for RT-qPCR + +<|ref|>table<|/ref|><|det|>[[120, 180, 851, 305]]<|/det|> + +
Primers used for Fig. 2C
PGC-1a_Ex3-5_FAGCCGTGACCACTGACAACGAG
PGC-1a_Ex3-5_RGCTGCATGGTTCTGAGTGCTAAG
Slirp_FGTCGGCCTATTGCGTTTGTC
Slirp_RATGTTCTCTCAGCTCACTCGC
Hprt FAGTCCCAGCGTCTGGATTAG
Hprt RTTTCCAAATCCTCGGCATAATGA
+ +<|ref|>table<|/ref|><|det|>[[120, 321, 851, 756]]<|/det|> + +
Primers/ Taqman gene expression assays used for Fig. 2A, D; Fig. 4F
SlirpMm01296845_m1
Full length Ppargc1 FTCAAGCCAAACCAACAACTTTATCT
Full length Ppargc1 RGGTTCGCTCAATAGTCTTGTTCTCA
Ppargc1 probeCACCAAATGACCCCAAGGGTTCCC
12S rRNAAJBJWSP
18S rRNAMm03928990_g1
ActBMm01205647_g1
mt-Nd1Mm04225274_s1
mt-Nd6Mm04225325_g1
mt-CytBMm04225271_g1
mt-ColMm04225243_g1
mt-Atp6Mm03649417_g1
Taqman gene expression assays used for Fig. 6
MT-ND1Hs02596873_s1
MT-ND6Hs02596879_g1
MT-CYBHs02596867_s1
MT-CO1Hs02596864_g1
MT-ATP6Hs02596862_g1
PPIAHs99999904_m1
B2MHs99999907_m1
+ +<|ref|>title<|/ref|><|det|>[[60, 796, 450, 808]]<|/det|> +# 890 DNA isolation and mtDNA quantification + +<|ref|>text<|/ref|><|det|>[[60, 835, 880, 848]]<|/det|> +891 Total DNA was extracted from ~20 mg mouse gastrocnemius muscle using the Dneasy Blood and + +<|ref|>text<|/ref|><|det|>[[60, 865, 880, 877]]<|/det|> +892 Tissue Kit (Qiagen) according to the manufacturer's instructions and treated with RNase A. Levels of + +<|ref|>text<|/ref|><|det|>[[60, 895, 880, 908]]<|/det|> +893 mtDNA were measured by qPCR using 2.5 ng of DNA in a QuantStudio 6 Flex Real-Time PCR System + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 130]]<|/det|> +using TaqMan™ Universal Master Mix II, with UNG. The mt- Nd1 and mt- Nd6/Nd5 TaqMan gene expression assays were used. The 18S probe was used for normalization. + +<|ref|>sub_title<|/ref|><|det|>[[115, 153, 616, 171]]<|/det|> +## Aerobic and strength ET intervention in healthy young women + +<|ref|>text<|/ref|><|det|>[[115, 193, 881, 300]]<|/det|> +A detailed description of the study participants, research design and methods has been previously published61. In the present study, we included vastus lateralis muscle lysate for immunoblotting obtained from 9 healthy young women (age, \(33 \pm 6\) years; body mass index (BMI), \(23.2 \pm 2.6 \mathrm{kg} \mathrm{m}^{- 2}\) ) before and after 14- week of aerobic and strength exercise training. + +<|ref|>sub_title<|/ref|><|det|>[[116, 323, 644, 340]]<|/det|> +## High-intensity interval training intervention in healthy young men + +<|ref|>text<|/ref|><|det|>[[115, 362, 881, 440]]<|/det|> +A detailed description of the study participants, research design and methods has been previously published63. In the present study, we included vastus lateralis muscle lysate for immunoblotting obtained from 6 healthy young men before and after 6- week HIIT training. + +<|ref|>sub_title<|/ref|><|det|>[[116, 462, 717, 479]]<|/det|> +## Progressive resistance exercise training in male young and older individuals + +<|ref|>text<|/ref|><|det|>[[115, 501, 882, 727]]<|/det|> +A detailed description of the study participants, research design and methods has been previously published45. The original study reported no additive effect of vitamin D intake during the 12 weeks of resistance exercise training on muscle hypertrophy or muscle strength60. Accordingly, in the present study, samples from young or older participants were considered as one group, irrespective of vitamin D intake. We included vastus lateralis muscle samples ( \(\sim 10 \mathrm{mg}\) wet weight) for immunoblotting obtained from a subset of the original sample set due to lack of sample material. For the young participants, we included 10 samples before and 6 samples after resistance exercise training. For the older participants we included 11 before and 9 samples after resistance exercise training. + +<|ref|>sub_title<|/ref|><|det|>[[115, 750, 880, 797]]<|/det|> +## High-intensity interval training in middle-aged male individuals, who were lean, obese or obese with type 2 diabetes + +<|ref|>text<|/ref|><|det|>[[115, 820, 881, 896]]<|/det|> +A detailed description of the study participants, research design and methods has been previously published62. In short, 15 middle- aged men with T2D and obesity, 15 age- matched glucose- tolerant men with obesity, and 18 age- matched glucose- tolerant lean men were recruited. All participants, except + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 343]]<|/det|> +four (two men with T2D and obesity and two lean men) completed the study. The training protocol consisted of 8- weeks with three weekly training sessions consisting of supervised HIIT in combination with biking and rowing. HIIT- sessions consisted of \(5 \times 1\) min exercise blocks interspersed with 1 min rest, and shifted between blocks on cycle and rowing ergometers. The volume was increased from two to five blocks during the 8 weeks. In the present study, we included vastus lateralis muscle samples that were obtained 4- 5 days after the last HIIT session but 48 hours after the last physical activity (a \(\mathrm{VO}_2\) max test). The number of samples included for immunoblotting as follows: 16 before and after HIIT from lean glucose- tolerant men, 15 before and after HIIT from glucose- tolerant men with obesity, and 13 before and after HIIT from men with T2D and obesity. + +<|ref|>sub_title<|/ref|><|det|>[[118, 364, 340, 381]]<|/det|> +## Ethical committee approval + +<|ref|>text<|/ref|><|det|>[[115, 403, 882, 600]]<|/det|> +Clinical experiments were approved by the Ethics Committee of Copenhagen or by the Regional Scientific Ethical Committees for Southern Denmark and performed in accordance with the Helsinki Declaration II. Studies including human study participants are described in Ref60- 63. All mouse experiments complied with the European Convention for the protection of vertebrate animals used for experimental and other scientific purposes (No. 123, Strasbourg, France, 1985; EU Directive 2010/63/EU for animal experiments) and were approved by the Danish Animal Experimental Inspectorate (License number: 2016- 15- 0201- 01043). + +<|ref|>sub_title<|/ref|><|det|>[[118, 623, 300, 640]]<|/det|> +## Graphical illustrations + +<|ref|>text<|/ref|><|det|>[[118, 661, 725, 679]]<|/det|> +Graphical illustrations were created in ©BioRender - biorender.com, as indicated. + +<|ref|>sub_title<|/ref|><|det|>[[118, 703, 270, 718]]<|/det|> +## Statistical methods + +<|ref|>text<|/ref|><|det|>[[115, 740, 882, 879]]<|/det|> +Statistical analyses were performed using GraphPad Prism 9 (version 9.3.1). Results are presented as mean ± SEM with individual values shown, when feasible. Statistical differences were analyzed by ordinary one- way ANOVA, repeated/ordinary two- way ANOVA, or Log- rank (Mantel- Cox) test, and Mann- Whitney test as applicable. Dunnett's multiple comparisons test or Šidák's multiple comparisons test were used to evaluate significant interactions in ANOVAs. As statistical tests varied according to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 879, 131]]<|/det|> +the dataset being analyzed, the respective tests utilized are specified within the figure legends. \(0.05 \leq p < 0.1\) were considered a tendency and p values \(< 0.05\) were considered significant. + +<|ref|>sub_title<|/ref|><|det|>[[118, 154, 275, 170]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[115, 193, 882, 421]]<|/det|> +We acknowledge the technical assistance of Betina Bolmgren and Irene Nielsen, Martin Thomassen, and Roberto Meneses- Valdes (Department of Nutrition, Exercise and Sports, Faculty of Science, University of Copenhagen, Denmark), Anja Jokipii- Utzon (Institute of Sports Medicine, Bispebjerg Hospital, Copenhagen, Denmark), and Michala Carlsson (Department of Biomedical Sciences, University of Copenhagen, Denmark). We thank Vivian Shang for her assistance with the Drosophila experiments (Charles Perkins Centre, The University of Sydney, Australia). We acknowledge Cristiano di Benedetto for preparing muscle samples for TEM analysis (Core Facility for Integrated Microscopy, Faculty of Health and Medical Sciences, University of Copenhagen, Denmark). + +<|ref|>sub_title<|/ref|><|det|>[[118, 444, 189, 460]]<|/det|> +## Funding + +<|ref|>text<|/ref|><|det|>[[115, 483, 882, 650]]<|/det|> +The study was supported by the Novo Nordisk Foundation (grant NNF16OC0023418, NNF18OC0032082, and NNF20OC0063577 to L.S.; grant NNF22OC0074110 to A.M.F.), by Independent Research Fund Denmark to L.S. (#0169- 00013B), by the European Union's Horizon 2020 research and innovation programme (Marie Sklodowska- Curie grant agreement No 801199 to T.C.P.P. and E.A.R.), and grants by Danish Council for Independent Research - Medical Sciences (4181- 00078) and the Augustinus Foundation to H.P. + +<|ref|>sub_title<|/ref|><|det|>[[118, 673, 460, 690]]<|/det|> +## CRediT authorship contribution statement + +<|ref|>text<|/ref|><|det|>[[115, 704, 882, 900]]<|/det|> +Tang Cam Phung Pham: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft, Visualization, Project administration, Funding acquisition; Steffen Henning Raun: Conceptualization; Investigation, Writing - Review & Editing; Essi Havula: Methodology, Investigation, Writing - Review & Editing; Carlos Henriquez- Olguin: Methodology, Investigation, Writing - Review & Editing; Diana Rubalcava- Gracia: Methodology, Investigation, Writing - Review & Editing; Emma Frank: Methodology, Investigation, Writing - Review & Editing; Andreas Mæchel Fritzen: Methodology, Investigation, Writing - Review & Editing; Paulo Jannig: Methodology, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 80, 884, 644]]<|/det|> +970 Investigation, Writing - Review & Editing; Nicoline Resen Andersen: Investigation, Writing - Review & Editing; Rikke Kruse: Investigation, Writing - Review & Editing; Mona Sadek Ali: Investigation, Writing - Review & Editing; Jens Frey Halling: Resources, Writing - Review & Editing; Stine Ringholm: Resources, Writing - Review & Editing; Solvejg Hansen: Resources, Writing - Review & Editing; Anders Krogh Lemminger: Resources, Writing - Review & Editing; Maria Houborg Petersen: Resources, Writing - Review & Editing; Martin Eisemann de Almedia: Resources, Writing - Review & Editing; Thomas E. Jensen: Resources, Writing - Review & Editing; Bente Kiens: Resources, Writing - Review & Editing; Morten Hostrup: Resources, Writing - Review & Editing; Steen Larsen: Methodology, Investigation, Writing - Review & Editing; Niels Ortenblad: Resources, Writing - Review & Editing; Kurt Højlund: Resources, Writing - Review & Editing; Peter Schjerling: Methodology, Investigation, Resources, Writing - Review & Editing; Michael Kjær: Resources, Writing - Review & Editing; Jorge Ruas: Resources, Writing - Review & Editing; Aleksandra Trifunovic: Resources, Writing - Review & Editing; Jørgen Wojtaszewski: Resources, Writing - Review & Editing; Joachim Nielsen: Methodology, Investigation, Writing - Review & Editing; Klaus Qvortrup: Resources, Writing - Review & Editing; Henriette Pilegaard: Resources, Writing - Review & Editing; Erik Arne Richter: Resources, Investigation, Writing - Review & Editing, Supervision, Funding acquisition; Lykke Sylow: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft, Project administration, Supervision, Funding acquisition. + +<|ref|>sub_title<|/ref|><|det|>[[118, 705, 207, 720]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[115, 744, 864, 870]]<|/det|> +991 1. Romanello, V. & Sandri, M. The connection between the dynamic remodeling of the mitochondrial network and the regulation of muscle mass. Cell Mol Life Sci 78, 1305-1328 (2021). 992 2. Distefano, G. et al. Physical activity unveils the relationship between mitochondrial energetics, muscle quality, and physical function in older adults. J Cachexia Sarcopenia Muscle 9, 279-294 (2018). 993 3. Gonzalez- Freire, M. et al. Reconsidering the Role of Mitochondria in Aging. J Gerontol A Biol Sci Med Sci 70, 1334-1342 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 80, 880, 900]]<|/det|> +999 4. Janssen, I., Heymsfield, S.B. & Ross, R. Low relative skeletal muscle mass (sarcopenia) in 1000 older persons is associated with functional impairment and physical disability. J Am Geriatr 1001 Soc 50, 889- 896 (2002). 1002 5. Short, K.R. et al. Decline in skeletal muscle mitochondrial function with aging in humans. Proc Natl Acad Sci U S A 102, 5618- 5623 (2005). 1003 6. Haines, M.S., Leong, A., Porneala, B.C., Meigs, J.B. & Miller, K.K. Association between muscle mass and diabetes prevalence independent of body fat distribution in adults under 50 years 1006 old. Nutr Diabetes 12, 29 (2022). 1007 7. Sugimoto, K. et al. Hyperglycemia in non- obese patients with type 2 diabetes is associated with low muscle mass: The Multicenter Study for Clarifying Evidence for Sarcopenia in 1009 Patients with Diabetes Mellitus. J Diabetes Investig 10, 1471- 1479 (2019). 1010 8. Yoon, J.W. et al. Hyperglycemia is Associated with Impaired Muscle Quality in Older Men 1011 with Diabetes: The Korean Longitudinal Study on Health and Aging. Diabetes Metab J 40, 140- 146 (2016). 1013 9. Ebhardt, H.A. et al. Comprehensive proteome analysis of human skeletal muscle in cachexia 1014 and sarcopenia: a pilot study. J Cachexia Sarcopenia Muscle 8, 567- 582 (2017). 1015 10. Romanello, V. & Sandri, M. Mitochondrial Quality Control and Muscle Mass Maintenance. 1016 Front Physiol 6, 422 (2015). 1017 11. Safdar, A. et al. Endurance exercise rescues progeroid aging and induces systemic 1018 mitochondrial rejuvenation in mtDNA mutator mice. Proc Natl Acad Sci U S A 108, 4135- 4140 (2011). 1020 12. Pedersen, B.K. & Saltin, B. Exercise as medicine - evidence for prescribing exercise as therapy 1021 in 26 different chronic diseases. Scand J Med Sci Sports 25 Suppl 3, 1- 72 (2015). 1022 13. Adhihety, P.J., Taivassalo, T., Haller, R.G., Walkinshaw, D.R. & Hood, D.A. The effect of 1023 training on the expression of mitochondrial biogenesis- and apoptosis- related proteins in 1024 skeletal muscle of patients with mtDNA defects. Am J Physiol Endocrinol Metab 293, E672- 1025 680 (2007). 1026 14. Sylow, L., Kleinert, M., Richter, E.A. & Jensen, T.E. Exercise- stimulated glucose uptake - 1027 regulation and implications for glycaemic control. Nat Rev Endocrinol 13, 133- 148 (2017). 1028 15. Boengler, K., Heusch, G. & Schulz, R. Nuclear- encoded mitochondrial proteins and their role 1029 in cardioprotection. Biochim Biophys Acta 1813, 1286- 1294 (2011). 1030 16. Couvillion, M.T., Soto, I.C., Shipkovenska, G. & Churchman, L.S. Synchronized mitochondrial 1031 and cytosolic translation programs. Nature 533, 499- 503 (2016). 1032 17. Sasarman, F. et al. LRPpRC and SLIRP interact in a ribonucleoprotein complex that regulates 1033 posttranscriptional gene expression in mitochondria. Mol Biol Cell 21, 1315- 1323 (2010). 1034 18. Lagouge, M. et al. SLIRP Regulates the Rate of Mitochondrial Protein Synthesis and Protects 1035 LRPpRC from Degradation. PLoS Genet 11, e1005423 (2015). 1036 19. Ruzzenente, B. et al. LRPpRC is necessary for polyadenylation and coordination of translation 1037 of mitochondrial mRNAs. EMBO J 31, 443- 456 (2012). 1038 20. Chujo, T. et al. LRPpRC/SLIRP suppresses PNPase- mediated mRNA decay and promotes 1039 polyadenylation in human mitochondria. Nucleic Acids Res 40, 8033- 8047 (2012). 1040 21. Baughman, J.M. et al. A computational screen for regulators of oxidative phosphorylation 1041 implicates SLIRP in mitochondrial RNA homeostasis. PLoS Genet 5, e1000590 (2009). 1042 22. Holloszy, J.O. Biochemical adaptations in muscle. Effects of exercise on mitochondrial 1043 oxygen uptake and respiratory enzyme activity in skeletal muscle. J Biol Chem 242, 2278- 2282 (1967). 1045 23. Kleinert, M. et al. Quantitative proteomic characterization of cellular pathways associated 1046 with altered insulin sensitivity in skeletal muscle following high- fat diet feeding and exercise 1047 training. Sci Rep 8, 10723 (2018). 1048 24. Hatchell, E.C. et al. SLIRP, a small SRA binding protein, is a nuclear receptor corepressor. Mol 1049 Cell 22, 657- 668 (2006). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 80, 875, 900]]<|/det|> +1050 25. Brand, A.H. & Perrimon, N. Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development 118, 401- 415 (1993). 1051 26. Baggio, F. et al. Drosophila melanogaster LRPPRC2 is involved in coordination of mitochondrial translation. Nucleic Acids Res 42, 13920- 13938 (2014). 1053 27. Inagaki, H.K., Kamikouchi, A. & Ito, K. Methods for quantifying simple gravity sensing in Drosophila melanogaster. Nat Protoc 5, 20- 25 (2010). 1056 28. Baar, K. et al. Adaptations of skeletal muscle to exercise: rapid increase in the transcriptional coactivator PGC- 1. FASEB J 16, 1879- 1886 (2002). 1058 29. Chinsomboon, J. et al. The transcriptional coactivator PGC- 1alpha mediates exercise- induced angiogenesis in skeletal muscle. Proc Natl Acad Sci U S A 106, 21401- 21406 (2009). 1059 30. Pilegaard, H., Saltin, B. & Neufer, P.D. Exercise induces transient transcriptional activation of the PGC- 1alpha gene in human skeletal muscle. J Physiol 546, 851- 858 (2003). 1061 31. Kjobsted, R. et al. Intact Regulation of the AMPK Signaling Network in Response to Exercise and Insulin in Skeletal Muscle of Male Patients With Type 2 Diabetes: Illumination of AMPK Activation in Recovery From Exercise. Diabetes 65, 1219- 1230 (2016). 1064 32. Halling, J.F. & Pilegaard, H. PGC- 1alpha- mediated regulation of mitochondrial function and physiological implications. Appl Physiol Nutr Metab 45, 927- 936 (2020). 1067 33. Ruas, J.L. et al. A PGC- 1alpha isoform induced by resistance training regulates skeletal muscle hypertrophy. Cell 151, 1319- 1331 (2012). 1069 34. Jannig, P.R., Dumesic, P.A., Spiegelman, B.M. & Ruas, J.L. SnapShot: Regulation and biology of PGC- 1alpha. Cell 185, 1444- 1444 e1441 (2022). 1071 35. Lin, J. et al. Transcriptional co- activator PGC- 1 alpha drives the formation of slow- twitch muscle fibres. Nature 418, 797- 801 (2002). 1073 36. Arany, Z. PGC- 1 coactivators and skeletal muscle adaptations in health and disease. Curr Opin Genet Dev 18, 426- 434 (2008). 1075 37. Handschin, C. et al. Abnormal glucose homeostasis in skeletal muscle- specific PGC- 1alpha knockout mice reveals skeletal muscle- pancreatic beta cell crosstalk. J Clin Invest 117, 3463- 3474 (2007). 1078 38. Miller, B.F., Konopka, A.R. & Hamilton, K.L. The rigorous study of exercise adaptations: why mRNA might not be enough. J Appl Physiol (1985) 121, 594- 596 (2016). 1080 39. Robinson, M.M. et al. Enhanced Protein Translation Underlies Improved Metabolic and Physical Adaptations to Different Exercise Training Modes in Young and Old Humans. Cell Metab 25, 581- 592 (2017). 1083 40. Liu, Y., Beyer, A. & Aebersold, R. On the Dependency of Cellular Protein Levels on mRNA Abundance. Cell 165, 535- 550 (2016). 1085 41. Richter, E.A. & Ruderman, N.B. AMPK and the biochemistry of exercise: implications for human health and disease. Biochem J 418, 261- 275 (2009). 1087 42. Fentz, J. et al. AMPKalpha is critical for enhancing skeletal muscle fatty acid utilization during in vivo exercise in mice. FASEB J 29, 1725- 1738 (2015). 1089 43. Lantier, L. et al. AMPK controls exercise endurance, mitochondrial oxidative capacity, and skeletal muscle integrity. FASEB J 28, 3211- 3224 (2014). 1091 44. Handschin, C. et al. Skeletal muscle fiber- type switching, exercise intolerance, and myopathy in PGC- 1alpha muscle- specific knock- out animals. J Biol Chem 282, 30014- 30021 (2007). 1093 45. Raun, S.H. et al. Housing temperature influences exercise training adaptations in mice. Nat Commun 11, 1560 (2020). 1095 46. Brandt, N. et al. The impact of exercise training and resveratrol supplementation on gut microbiota composition in high- fat diet feed mice. Physiol Rep 6, e13881 (2018). 1097 47. Wojtaszewski, J.F. & Richter, E.A. Effects of acute exercise and training on insulin action and sensitivity: focus on molecular mechanisms in muscle. Essays Biochem 42, 31- 46 (2006). 1099 48. Memme, J.M. & Hood, D.A. Molecular Basis for the Therapeutic Effects of Exercise on Mitochondrial Defects. Front Physiol 11, 615038 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 87, 878, 884]]<|/det|> +1101 49. Carter, H.N., Pauly, M., Tryon, L.D. & Hood, D.A. Effect of contractile activity on PGC-1alpha transcription in young and aged skeletal muscle. J Appl Physiol (1985) 124, 1605-1615 (2018).1104 50. Lanza, I.R. et al. Endurance exercise as a countermeasure for aging. Diabetes 57, 2933-2942 (2008).1106 51. Schaefer, P.M. et al. Mitochondrial mutations alter endurance exercise response and determinants in mice. Proc Natl Acad Sci U S A 119, e2200549119 (2022).1108 52. Spahr, H. et al. SLIRP stabilizes LRPPRC via an RRM-PPR protein interface. Nucleic Acids Res 44, 6868-6882 (2016).1110 53. Filograna, R., Mennuni, M., Alsina, D. & Larsson, N.G. Mitochondrial DNA copy number in human disease: the more the better? FEBS Lett 595, 976-1002 (2021).1112 54. Larsen, S. et al. Biomarkers of mitochondrial content in skeletal muscle of healthy young human subjects. J Physiol 590, 3349-3360 (2012).1114 55. Halling, J.F., Ringholm, S., Olesen, J., Prats, C. & Pilegaard, H. Exercise training protects against aging-induced mitochondrial fragmentation in mouse skeletal muscle in a PGC-11alpha dependent manner. Exp Gerontol 96, 1-6 (2017).1117 56. Metodiev, M.D. et al. Methylation of 125 rRNA is necessary for in vivo stability of the small subunit of the mammalian mitochondrial ribosome. Cell Metab 9, 386-397 (2009).1119 57. Yang, Y. et al. NAD+-dependent deacetylase SIRT3 regulates mitochondrial protein synthesis by deacetylation of the ribosomal protein MRPL10. J Biol Chem 285, 7417-7429 (2010).1121 58. Cox, A.G., Winterbourn, C.C. & Hampton, M.B. Mitochondrial peroxiredoxin involvement in antioxidant defence and redox signalling. Biochem J 425, 313-325 (2009).1123 59. Lee, K.P. et al. Peroxiredoxin 3 has a crucial role in the contractile function of skeletal muscle by regulating mitochondrial homeostasis. Free Radic Biol Med 77, 298-306 (2014).1125 60. Hansen, S.L. et al. Mechanisms Underlying Absent Training-Induced Improvement in Insulin Action in Lean, Hyperandrogenic Women With Polycystic Ovary Syndrome. Diabetes 69, 2267-2280 (2020).1127 61. Agergaard, J. et al. Does vitamin-D intake during resistance training improve the skeletal muscle hypertrophic and strength response in young and elderly men? - a randomized controlled trial. Nutr Metab (Lond) 12, 32 (2015).1121 62. Petersen, M.H. et al. High-intensity interval training combining rowing and cycling efficiently improves insulin sensitivity, body composition and VO(2)max in men with obesity and type 2 diabetes. Front Endocrinol (Lausanne) 13, 1032235 (2022).1134 63. Hostrup, M. et al. High-Intensity Training Represses FXYD5 and Glycosylates Na,K-ATPase in Type II Muscle Fibres, Which Are Linked with Improved Muscle K(+) Handling and Performance. Int J Mol Sci 24 (2023).1137 64. Hey-Mogensen, M. et al. Effect of physical training on mitochondrial respiration and reactive oxygen species release in skeletal muscle in patients with obesity and type 2 diabetes. Diabetologia 53, 1976-1985 (2010).1140 65. Phielix, E., Meeex, R., Moonen-Kornips, E., Hesselink, M.K. & Schrauwen, P. Exercise training increases mitochondrial content and ex vivo mitochondrial function similarly in patients with type 2 diabetes and in control individuals. Diabetologia 53, 1714-1721 (2010).1143 66. Karakelides, H., Irving, B.A., Short, K.R., O'Brien, P. & Nair, K.S. Age, obesity, and sex effects on insulin sensitivity and skeletal muscle mitochondrial function. Diabetes 59, 89-97 (2010).1145 67. Schwanhausser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337-342 (2011).1147 68. Gineste, C. et al. Enzymatically dissociated muscle fibers display rapid dedifferentiation and impaired mitochondrial calcium control. iScience 25, 105654 (2022).1149 69. Copeland, J.M. et al. Extension of Drosophila life span by RNAi of the mitochondrial respiratory chain. Curr Biol 19, 1591-1598 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 87, 876, 900]]<|/det|> +1151 70. Dell'agnello, C. et al. Increased longevity and refractoriness to Ca(2+) - dependent neurodegeneration in Surf1 knockout mice. Hum Mol Genet 16, 431- 444 (2007).1153 71. Dillin, A. et al. Rates of behavior and aging specified by mitochondrial function during development. Science 298, 2398- 2401 (2002).1155 72. Rossignol, R. et al. Mitochondrial threshold effects. Biochem J 370, 751- 762 (2003).1156 73. Hwang, A.B., Jeong, D.E. & Lee, S.J. Mitochondria and organismal longevity. Curr Genomics 113, 519- 532 (2012).1158 74. Xu, C. et al. Genetic inhibition of an ATP synthase subunit extends lifespan in C. elegans. Sci Rep 8, 14836 (2018).1159 75. Liu, X. et al. Evolutionary conservation of the clk- 1- dependent mechanism of longevity: loss of mclk1 increases cellular fitness and lifespan in mice. Genes Dev 19, 2424- 2434 (2005).1162 76. Leick, L. et al. PGC- 1alpha is not mandatory for exercise- and training- induced adaptive gene responses in mouse skeletal muscle. Am J Physiol Endocrinol Metab 294, E463- 474 (2008).1164 77. Larsen, S., Skaaby, S., Helge, J.W. & Dela, F. Effects of exercise training on mitochondrial function in patients with type 2 diabetes. World J Diabetes 5, 482- 492 (2014).1166 78. Sylow, L., Tokarz, V.L., Richter, E.A. & Klip, A. The many actions of insulin in skeletal muscle, the paramount tissue determining glycemia. Cell Metab 33, 758- 780 (2021).1168 79. Richter, E.A., Sylow, L. & Hargreaves, M. Interactions between insulin and exercise. Biochem J 478, 3827- 3846 (2021).1170 80. Cartee, G.D. Sexual Dimorphic Effects of Exercise Training on Subcutaneous White Adipose Tissue of Mice. Diabetes 70, 1242- 1243 (2021).1171 81. Holcomb, L.E., Rowe, P., O'Neill, C.C., DeWitt, E.A. & Kolwicz, S.C., Jr. Sex differences in endurance exercise capacity and skeletal muscle lipid metabolism in mice. Physiol Rep 10, e15174 (2022).1175 82. Rooyackers, O.E., Adey, D.B., Ades, P.A. & Nair, K.S. Effect of age on in vivo rates of mitochondrial protein synthesis in human skeletal muscle. Proc Natl Acad Sci U S A 93, 15364- 15369 (1996).1178 83. Lord, S.J., Velle, K.B., Mullins, R.D. & Fritz- Laylin, L.K. SuperPlots: Communicating reproducibility and variability in cell biology. J Cell Biol 219 (2020).1180 84. Ganetzky, B. & Flanagan, J.R. On the relationship between senescence and age- related changes in two wild- type strains of Drosophila melanogaster. Exp Gerontol 13, 189- 196 (1978).1183 85. Correia, J.C. et al. Muscle- secreted neururin couples myofiber oxidative metabolism and slow motor neuron identity. Cell Metab 33, 2215- 2230 e2218 (2021).1185 86. Geng, T. et al. PGC- 1alpha plays a functional role in exercise- induced mitochondrial biogenesis and angiogenesis but not fiber- type transformation in mouse skeletal muscle. Am J Physiol Cell Physiol 298, C572- 579 (2010).1187 87. Jeppesen, J. et al. LKB1 regulates lipid oxidation during exercise independently of AMPK. Diabetes 62, 1490- 1499 (2013).1189 88. Eisenberg, B.R. Quantitative ultrastructure of mammalian skeletal muscle, in Handbook of Physiology, Skeletal Muscle. (eds. L.D. Peachey, R.H. Adrian & S.R. Geiger) p. 73- 112 (American Physiological Society, Bethesda, MD; 1983).1193 89. Pesta, D. & Gnaiger, E. High- resolution respirometry: OXPHOS protocols for human cells and permeabilized fibers from small biopsies of human muscle. Methods Mol Biol 810, 25- 58 (2012).1196 90. Dzamko, N. et al. AMPK- independent pathways regulate skeletal muscle fatty acid oxidation. J Physiol 586, 5819- 5831 (2008).1198 91. Hostrup, M. et al. High- intensity interval training remodels the proteome and acetylome of human skeletal muscle. Elife 11 (2022).1200 92. Kruse, R. et al. Intact initiation of autophagy and mitochondrial fission by acute exercise in skeletal muscle of patients with Type 2 diabetes. Clin Sci (Lond) 131, 37- 47 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 85, 864, 170]]<|/det|> +1202 93. Kruse, R. et al. Effect of long-term testosterone therapy on molecular regulators of skeletal muscle mass and fibre-type distribution in aging men with subnormal testosterone. 1203 Metabolism 112, 154347 (2020). 1204 94. Welinder, C. & Ekblad, L. Coomassie staining as loading control in Western blot analysis. J Proteome Res 10, 1416-1419 (2011). 1207 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[30, 7, 950, 240]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[30, 247, 944, 520]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[30, 536, 960, 936]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 25, 999, 817]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[920, 825, 990, 848]]<|/det|> +
Fig. 2
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[12, 45, 325, 310]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[12, 365, 325, 610]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[12, 671, 325, 925]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[340, 45, 655, 310]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[340, 365, 655, 630]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[340, 671, 655, 925]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[670, 45, 965, 310]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[670, 365, 965, 630]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[670, 671, 965, 925]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[18, 20, 978, 225]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[18, 238, 978, 444]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[18, 448, 978, 560]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[18, 585, 978, 747]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[18, 770, 978, 944]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[920, 966, 985, 988]]<|/det|> +
Fig.4
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[10, 50, 864, 920]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[20, 24, 940, 475]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[20, 465, 940, 650]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[24, 666, 465, 870]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[500, 465, 940, 650]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[10, 10, 884, 952]]<|/det|> + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[40, 23, 456, 52]]<|/det|> +# HIIT in individuals with a lean body, obesity, and T2D + +<|ref|>image<|/ref|><|det|>[[18, 100, 955, 911]]<|/det|> + +<--- Page Split ---> diff --git a/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/images_list.json b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..cad8ea4a6f3ecbe9c17dd2e409f900f71b27d77a --- /dev/null +++ b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig 1. Interploidy and interspecies hybrid seeds share similar phenotypes and deregulated genes.", + "footnote": [], + "bbox": [ + [ + 150, + 75, + 728, + 730 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Interspecies hybridization causes chromatin condensation defects in endosperm nuclei.", + "footnote": [], + "bbox": [ + [ + 150, + 84, + 810, + 528 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. Maternal loss of NRPD1 mimics paternal-excess hybrid phenotypes.", + "footnote": [], + "bbox": [ + [ + 144, + 75, + 748, + 565 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Maternal sirenRNAs are depleted in hybrid endosperm.", + "footnote": [], + "bbox": [ + [ + 144, + 78, + 760, + 715 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Loss of DNA methylation at trans sirenRNA targets and increased dosage of paternal AGLs contributes to seed incompatibility.", + "footnote": [], + "bbox": [ + [ + 147, + 78, + 692, + 700 + ] + ], + "page_idx": 28 + } +] \ No newline at end of file diff --git a/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0.mmd b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0.mmd new file mode 100644 index 0000000000000000000000000000000000000000..c82d2af801ebbc8db35aa026ee99bfb992501e1e --- /dev/null +++ b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0.mmd @@ -0,0 +1,289 @@ + +# Dosage sensitive maternal siRNAs determine hybridization success in Capsella + +Katarzyna Dziasek https://orcid.org/0000- 0002- 7279- 1417 Juan Santos- González Swedish University of Agricultural Sciences Yichun Qiu Max Planck Institute of Molecular Plant Physiology Diana Rigola KeyGene (Netherlands) Koen Nijbroek Keygene Claudia Köhler + +koehler@mpimp- golm.mpg.de + +Max Planck Institute of Molecular Plant Physiology https://orcid.org/0000- 0002- 2619- 4857 + +## Article + +# Keywords: + +Posted Date: December 6th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3705839/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Plants on November 11th, 2024. See the published version at https://doi.org/10.1038/s41477- 024- 01844- 3. + +<--- Page Split ---> + +# Dosage sensitive maternal siRNAs determine hybridization success in Capsella + +Katarzyna Dziasek \(^{1}\) , Juan Santos- González \(^{1}\) , Yichun Qiu \(^{2}\) , Diana Rigola \(^{3}\) , Koen Nijbroek \(^{3}\) , Claudia Köhler \(^{1,2*}\) + +1 Department of Plant Reproductive Biology and Epigenetics, Max Planck Institute of Molecular Plant Physiology, Potsdam, 14476, Germany. + +2 Department of Plant Biology, Uppsala BioCenter, Swedish University of Agricultural Sciences and Linnean Centre for Plant Biology, 75007, Uppsala, Sweden. + +3 Keygene N.V., Agro Business Park 90, 6708 PW Wageningen, The Netherlands + +\*Corresponding author. Email: koehler@mpimp- golm.mpg.de. + +<--- Page Split ---> + +## Abstract + +Hybrid seed failure arising from wide crosses between plant species is a recurring obstacle in plant breeding, impeding the transfer of desirable traits. This postzygotic reproductive barrier primarily occurs in the endosperm, a dosage- sensitive tissue that nurtures the embryo and functions analogously to the placenta in mammals. Here, we show that the endosperm in incompatible seeds loses DNA methylation and chromatin condensation, resembling the endosperm of seeds depleted for maternal RNA Polymerase IV (Pol IV) function. This similarity in phenotype corresponds with a concurrent reduction in small interfering RNAs in the endosperm (sirenRNAs), maternal Pol IV- dependent siRNAs governing DNA methylation and regulating target genes. Notably, several AGAMOUS- LIKE MADS- BOX transcription factors (AGLs), known regulators of endosperm development, are targeted by sirenRNAs in cis and in trans. This finding aligns with the enrichment of AGL target genes among deregulated genes. Together, we propose that the response to interspecies hybridizations is triggered by a combination of reduced maternally derived sirenRNAs and heightened expression of AGLs targeting hypomethylated regions in the hybrid endosperm. + +<--- Page Split ---> + +Hybrid seed failure resulting from wide crosses between plant species poses a recurring challenge in plant breeding, obstructing the transfer of desirable traits between species. This postzygotic reproductive barrier predominantly manifests in the endosperm – an embryo nourishing tissue serving a similar function as the placenta in mammals \(^{1}\) . In most flowering plants, the endosperm is triploid, comprising two maternal and one paternal genome. This specific genomic ratio is pivotal for endosperm viability, as evidenced by seed abortion upon hybridization of plants with differing ploidy \(^{2 - 4}\) . Much like failures in interploidy hybridization, crosses between closely related species with equal ploidy often result in seed arrest. The term "effective ploidy" has been used to characterize this phenomenon, wherein certain species exhibit behavior reminiscent of polyploid species, despite their actual diploid nature \(^{5}\) . Remarkably, seeds arrested from both interploidy and interspecies crosses exhibit similar phenotypic and molecular abnormalities \(^{6 - 8}\) , implying the existence of a dosage- sensitive element contributing to this reproductive barrier within the endosperm. However, the precise nature of this component remains elusive. + +In the Capsella genus, interspecies hybridizations between the inbreeder C. rubella and the outbreeder C. grandiflora lead to a loss of chromatin condensation and DNA methylation in the endosperm, particularly in the CHG and CHH contexts (H representing all bases except G). This chromatin change is associated with an accumulation of deregulated genes in pericentromeric regions, many of which are targeted by type I AGAMOUS- LIKE MADS- box transcription factors (AGLs) \(^{9}\) . Type I AGLs belonging to the \(\mathsf{M}_{\gamma}\) and \(\mathsf{M}_{\gamma}\) - interacting \(\mathsf{M}\alpha^{*}\) clades are key regulators of endosperm proliferation and cellularization \(^{10}\) . This data suggests that hypomethylation in the hybrid endosperm facilitates the access of AGLs to hypomethylated regions, resulting in detrimental gene activation. Similarly, in Arabidopsis, interploidy hybridizations of diploid (2x) maternal plants with tetraploid (4x) paternal plants causes endosperm hypomethylation in CHG and CHH context \(^{11}\) , which has been linked to a disruption of the RNA- directed DNA methylation (RdDM) pathway in the hybrid endosperm \(^{12}\) . This pathway initiates with the plant- specific DNA- dependent RNA Polymerase IV (Pol IV), generating short non- coding transcripts that are transformed into double- stranded RNA by RNA- DEPENDENT RNA POLYMEREASE 2 (RDR2) and further processed into 24 nt short interfering (si) RNAs by DICER- LIKE3 (DCL3). These siRNAs, when incorporated into ARGONAUTE (AGO) proteins, guide DNA + +<--- Page Split ---> + +methyltransferases to target sequences, inducing DNA methylation13. Notably, both ovules and endosperm accumulate highly abundant siRNAs over a modest number of distinct loci, commonly referred to as sirenRNAs (small- interfering RNA in endosperm) that are primarily of maternal origin14- 16. In \*Arabidopsis\*, sirenRNAs target \*AGLs\* and their abundance is reduced in triploid seeds derived from 2x \*x4x\* crosses17. Nonetheless, whether sirenRNAs constitute the dosage- sensitive component determining hybridization success remains an open question. + +Previous work revealed that loss of paternal Pol IV function can suppress the interplay hybridization barrier in \*Arabidopsis\*11,14. Interestingly, affected genes in the endosperm differ depending on whether the \*nrdp1\* mutation is maternally or paternally inherited, suggesting that maternal and paternal Pol IV- dependent siRNAs have distinct effects on endosperm development for reasons that remain to be fully elucidated11,18. + +In this study, we demonstrate that the loss of maternal Pol IV function mirrors the loss of chromatin condensation and DNA methylation observed in \*Capsella\* hybrids. This resemblance in phenotype is correlated with a shared decrease in sirenRNAs, which regulate target genes in \*cis\* and \*trans\* by guiding DNA methylation. Among these sirenRNA targets are several \*AGLs\*, aligning with the enrichment of AGL target genes among deregulated genes. Based on this data we propose that the response to interspecies and interplay hybridizations is instigated by two factors: the depletion of maternally derived sirenRNAs and the heightened expression of \*AGLs\* targeting hypomethylated regions in hybrid endosperm. + +## Results + +## \*Capsella\* interplay and interspecies hybrids share similar phenotypic and morphological defects + +To test whether interplay and interspecies hybridization share a common molecular basis, we analyzed the cross between diploid (2x) maternal and tetraploid (4x) paternal \*C. rubella\* (Cr) plants (referred to as \*Cr \* \*4xCr\*, by convention the female parent is always mentioned first). Like previously reported for paternal- excess interplay crosses in \*Arabidopsis\*3, the resulting 3x Cr seeds were dark and shriveled and failed to germinate (Fig. 1A, Supplementary Fig. 1A; the term paternal- excess will be used + +<--- Page Split ---> + +to refer to crosses where the paternal parent has higher numerical or effective ploidy compared to the maternal parent). Defects of \(3x Cr\) seeds related to failed endosperm cellularization (Supplementary Fig. 1B), resembling previously reported defects of interspecies paternal- excess hybrid seeds in Capsella, Arabidopsis and other species \(^{3,6,19,20}\) . As in the case of interspecific hybrids \(^{21}\) , \(3x\) embryos could be rescued by removing them from the seeds and growing them in vitro (Supplementary Fig. 1C), supporting the view that the failure of \(3x\) embryo survival is a consequence of a defect in endosperm development. + +To determine if phenotypic similarities in interplay and interspecies hybrids are reflected by a similar molecular signature, we generated transcriptome data of hybrid seeds (6 DAP) resulting from incompatible crosses. These included \(Cr \times 4xCr\) interplay crosses and previously characterized paternal- excess interspecies crosses of \(Cr \times C\) . grandiflora (Cg) and C. orientalis (Co) \(\times Cr^{21}\) with the corresponding intraspecific crosses. We furthermore included the cross of C. bursa pastoris (Cbp) \(\times Cg\) (Supplementary Table 1). Despite Cbp being an allotetraploid species, when utilized as the maternal parent in crosses with Cg, Cbp exhibited strikingly similar behavior to Cr. The resultant seeds exhibited a paternal excess phenotype, characterized by shrinkage and abortion, with embryos arrested at the torpedo stage (Fig. 1A). Correspondingly, reciprocal crosses of \(Cr \times Cbp\) gave rise to normal, viable seeds (Fig. 1B), revealing that Cbp, despite being a tetraploid species, behaved like a diploid. Based on the cross- incompatibilities between the species, we assigned an effective ploidy order of \(Co < Cr = Cpb < Cg = 4xCr\) . Consistent with theoretical predictions \(^{22}\) , this order aligned with the breeding mode of the species; while Cg is an outbreeder, Cr, Co and Cbp are inbreeding species. However, while Co became an inbreeder already \(\sim 900.000\) years ago, Cr and Cbp made the transition to inbreeding only \(\sim 200.000\) years back \(^{23,24}\) . + +We identified deregulated genes in the hybrids by comparing the hybrid transcriptome with that of the corresponding maternal parent. Our data revealed that all tested hybrids shared a large set of commonly upregulated genes, indicative of a common genetic basis (Fig. 1 C, D). This conclusion is consistent with earlier work showing that increased ploidy of the maternal Cr parent can restore viability of hybrid \(Cr \times Cg\) seeds \(^{21}\) . To test the molecular consequences of increased maternal ploidy, we profiled the transcriptome of \(4xCr \times Cg\) seeds and found that most genes that were upregulated + +<--- Page Split ---> + +in \(Cr \times Cg\) hybrid seeds were either downregulated or unchanged in \(4xCr \times Cg\) seeds in comparison to the maternal parent (Fig. 1C). This data supports the idea that a dosage- sensitive component generated by the maternal parent determines hybridization success. To further challenge this idea, we tested whether we could suppress \(Co \times Cr\) hybrid seed defects by increasing the ploidy of the maternal Co parent. Seeds derived from \(Co \times Cr\) crosses aborted with similar morphological and transcriptional defects as \(Cr \times Cg\) , \(Cbp \times Cg\) and \(Cr \times 4xCr\) hybrids \(^{21}\) and (Fig. 1A, C, D), while seeds obtained from crosses of \(4xCo \times Cr\) were phenotypically like non- hybrid seeds and able to germinate (Fig. 1E, F). + +Together, this data strongly suggests that interplay and interspecies hybridization barriers share a common molecular basis. Moreover, the fact that increased ploidy of the maternal parent can alleviate paternal- excess hybrid incompatibility, implies the existence of a quantitative maternal factor determining hybridization success. + +## Incompatible hybrids have decondensed pericentromeric regions + +Previous work revealed that \(Cr \times Cg\) hybrid endosperm has decondensed chromocenters, which relates to a preferential expression of genes localized in pericentromeric regions \(^{9}\) . To test whether this phenomenon is a general pattern connected with hybrid incompatibility in the Capsella genus, we analyzed the location of upregulated genes in paternal- excess crosses in relation to their position on the chromosomes. Strikingly, we found that in all interspecies and interplay hybrids there was a preferential location of upregulated genes close to pericentromeric regions (Fig. 2A). We tested whether this phenomenon corresponded to a loss of chromosome condensation and analyzed chromocenter condensation in a range of paternal- excess incompatible hybrids. Consistent with the similar molecular phenotype of interspecies and interplay paternal- excess hybrid seeds, we found that all hybrid nuclei had significantly reduced chromocenter condensation compared to non- hybrid nuclei (Fig. 2B, C). Importantly, chromocenter condensation defects, DNA hypomethylation and gene deregulation was restored in nuclei of \(4xCr \times Cg\) hybrids, revealing that a dosage- sensitive maternal factor regulates chromatin condensation in the endosperm (Fig. 2A- E, Supplementary Fig. 2A, B). + +<--- Page Split ---> + +Decondensation of pericentromeric regions in \(Cr \times Cg\) hybrid endosperm co- occurred with a loss of CHG and CHH methylation9. Similarly, loss of chromatin condensation in nuclei of interploidy hybrids related to reduced CHG and CHH methylation on genes and TEs (Supplementary Fig. 2C- F). Together, this data reveals that interploidy and interspecies hybridization cause similar molecular defects consistent with the similar phenotypes of hybrid seeds. + +## Maternal loss of Pol IV function causes chromatin decondensation and loss of CHG and CHH methylation + +The observed reduction of CHG and CHH methylation upon interploidy and interspecies hybridization9 (Supplementary Fig. 2C- F) suggests a role of the RdDM pathway in establishing hybridization barriers in the endosperm. Pol IV is required to produce precursor RNAs that are converted into 24 nt siRNAs guiding de novo DNA methylation in all sequence contexts13. We thus tested whether a mutant in NRPD1, encoding the large subunit of Pol IV, would cause a similar effect on chromatin condensation and loss of DNA methylation. Indeed, we found that maternal loss of NRPD1 reduced chromatin condensation in the endosperm, while loss of paternal NRPD1 had no effect and loss of maternal and paternal NRPD1 function did not enhance chromatin decondensation (Fig. 3A, B). Chromatin decondensation upon loss of maternal NRPD1 function was associated with loss of CHG and CHH methylation on genes and TEs in the endosperm, with similar regions being affected in interploidy and interspecies hybrids and seeds depleted of maternal NRPD1 function (Fig. 3C, D, Supplementary Fig. 3A- D). This was also reflected by a large set of similarly deregulated genes in the endosperm of seeds lacking maternal NRPD1 function and interploidy and interspecies hybrids (Fig. 3E, Supplementary Table 1). Previous studies found no evidence of NRPD1 being imprinted in Capsella21,25, suggesting that maternal Pol IV- dependent siRNAs are generated in maternal sporophytic tissues, consistent with previous work16,26. Together, we conclude that reduced dosage of maternal Pol IV- dependent siRNAs affects DNA methylation, chromatin condensation and gene expression in a similar manner as interploidy and interspecies hybridizations. This data raises the hypothesis that interploidy and interspecies hybrid endosperms are depleted of maternal Pol IV- dependent siRNAs, causing loss of DNA methylation, decondensation of pericentromeric regions and activation of genes. + +<--- Page Split ---> + +## Maternal sirenRNAs are depleted in hybrid endosperm + +To understand the role of maternal NRPD1 in hybrid incompatibility, we sequenced sRNAs in manually dissected endosperm of 6 DAP seeds derived from crosses of \(Cr \times Cr\) , \(Cg \times Cg\) , \(Cr \times Cg\) , \(nrpd1 \times Cr\) and \(nrpd1 \times nrpd1\) (Supplementary Table 2). Profiles of sRNAs correlated well among biological replicates of the same genotype but were clearly distinct between genotypes (Supplementary Fig. 4A, B). Consistent with previous work14, we found that 24 nt sRNAs were the predominant sRNA species in the endosperm (Supplementary Fig. 4C). Using ShortStack27 we identified 13957 clusters accumulating 24 nt sRNAs in the endosperm of \(Cr \times Cr\) seeds. Out of those, only a few loci produced high levels of sRNAs in \(Cr \times Cr\) endosperm, resembling the characteristics of siren loci15,16 (Fig. 4A, Supplementary Table 3). We identified 1389 loci giving rise to 90% of total sRNAs in \(Cr \times Cr\) endosperm that we will refer to as siren loci. As previously reported16, siren loci were longer than other loci expressing sRNAs (Supplementary Fig. 4D) and overlapped TEs, intergenic regions and genes (Supplementary Fig. 4E). SirenRNA loci were enriched for 24 nt sRNAs (Supplementary Fig. 4F) with an adenine bias at the 5' nucleotide (Supplementary Fig. 4G), suggesting interaction with AGO4- related AGO proteins28, consistent with previous observations in Brassica16. SirenRNA loci are distributed over the length of the chromosomes with an enrichment in pericentromeric regions (Supplementary Fig. 4H, I), but without a pronounced preference for any specific TE family (Supplementary Fig. 4J). Abundance of sirenRNAs was strongly depleted upon loss of maternal NRPD1 function and decreased further in nrpd1 homozygous endosperm (Fig. 4B), indicating that sirenRNAs are produced in maternal sporophytic tissues and in the endosperm. There are conflicting data on the origin of sirenRNAs; while they have been proposed to be predominantly maternally produced in Brassica16,29, biparental production was proposed in Arabidopsis18. Our data suggests that most likely both, the maternal seed coat and the endosperm are the source of sirenRNAs in the endosperm, with a predominant fraction being maternally derived. Like in maternal nrpd1 endosperm, in \(Cr \times Cg\) hybrid endosperm sirenRNAs abundance was strongly depleted (Fig. 4B). Nearly all loci losing sRNAs in \(Cr \times Cg\) endosperm corresponded to siren loci (Fig. 4C, D), strongly suggesting that maternally produced sirenRNAs are the dosage sensitive component that becomes depleted in hybrid endosperm. + +<--- Page Split ---> + +Mapping reads from \(Cr \times Cr\) and \(Cg \times Cg\) to their respective genomes revealed variations in the level and specificity of sirenRNAs. Specifically, when \(Cg \times Cg\) reads were mapped to the \(Cr\) genome, there was an apparent decrease in sirenRNA levels, which was also observed when mapping \(Cr \times Cr\) reads to the \(Cg\) genome. This data reveals that at least some sirenRNAs are species- specific and that \(Cg\) accumulates substantially higher levels of sirenRNAs compared to \(Cr\) (Fig. 4B, D). + +Siren loci were heavily methylated, corresponding to the production of high levels of sirenRNAs (Fig. 4D, E). Depletion of sirenRNAs in \(Cr \times Cg\) and \(nrpd1 \times Cr\) endosperm corresponded to increased mRNA level of many genes overlapping those loci (Fig. 4F, G), indicating that sirenRNAs negatively regulate gene expression, consistent with previous work29,30. Transcriptional changes related to a loss of CHG and CHH methylation that however were not significant in the hybrid endosperm (Fig. 4D, H). + +Depletion of maternal 24 nt siRNAs in the endosperm of hybrid tomato seeds was related to decreased expression of Pol IV subunits encoding genes \(NRPD1\) and \(NRPD2^{31}\) . We tested expression of RdDM components in hybrid seed transcriptomes and found significantly reduced \(NRPD1\) expression in \(Cr \times Cg\) , \(Co \times Cr\) and \(Cr \times 4x Cr\) endosperm (Supplementary Fig. 5), which may connect to loss of sirenRNAs in the hybrid. + +## Depletion of sirenRNAs in hybrid endosperm causes loss of methylation at trans targets + +Previous work revealed that sirenRNAs can methylate other genomic sequences in trans29. We identified sirenRNA trans targets by mapping sirenRNAs to the genome masked for sirenRNA producing loci by allowing two mismatches. Indeed, we found that loci accumulating higher levels of non- perfectly matching sirenRNAs had higher methylation levels in CHG and CHH context (Fig. 5A). SirenRNAs targeted genes and TEs, with dosage- dependent effects on DNA methylation being most prominent on TEs (Fig. 5B). Together, this data strongly suggests that sirenRNAs guide DNA methylation in trans in the endosperm, like their proposed function in ovules29. Consistent with this idea, we found a decline of CHG and CHH methylation at trans targets in \(Cr \times Cg\) hybrid endosperm that correlated with the dosage of depleted sirenRNAs (Fig. 5C, D). We next asked whether there is a connection between sirenRNAs, loss of DNA + +<--- Page Split ---> + +methylation at trans targets and changes in gene expression in the hybrid \(Cr \times Cg\) endosperm. Consistent with sirenRNAs most prominently targeting TEs (Fig. 5B), we found that genes upregulated in \(Cr \times Cg\) hybrids were closer to a TE upstream of their transcription start site than other genes (Fig. 5E). Moreover, the TEs in the promoters of those genes had reduced CHG and CHH methylation in \(Cr \times Cg\) hybrids (Fig. 5F). Among the genes that were targeted by trans- acting sirenRNAs and lost DNA methylation we found many upregulated AGLs (Supplementary Fig. 6A, B and Supplementary Table 4), consistent with previous findings reporting targeting of AGLs by maternal Pol IV- dependent siRNAs30. + +Based on this data we conclude that sirenRNAs in the Capsella endosperm drive non- CG DNA methylation in trans. Depletion of sirenRNAs in the \(Cr \times Cg\) hybrid leads to loss of DNA methylation and increased expression of several trans targets, among them several genes with potential function in establishing hybridization barriers (Supplementary Table 4). These include SUVH7, which is involved in the triploid block in Arabidopsis32, YUC10, an auxin biosynthesis gene33 and ARF21 – a centromeric ARF regulating endosperm cellularization34. Moreover, among those genes are nine AGLs including AGL38 – the ortholog of PHE1 (Supplementary Fig. 6), which activates key regulators of the triploid block in Arabidopsis35. + +## Upregulated genes in hybrid endosperm are enriched for putative AGL targets + +Previous work revealed that target sequence motifs of the AGL PHE1 are frequently located in helitron TEs35. Since helitrons in Arabidopsis and other Brassicaceae are concentrated in pericentromeric regions36,37, we tested whether pericentromeric regions in Capsella were enriched for PHE1 binding motifs. We found indeed that the frequency of genes containing PHE1 binding motifs was significantly higher in pericentromeric regions compared to chromosome arms (Fig. 5G). Importantly, in all analyzed incompatible crosses, the promoter regions of upregulated genes located in pericentromeric regions were significantly enriched for PHE1 binding sites (Fig. 5G). Based on this data we propose that increased expression of genes in pericentromeric regions in hybrid endosperm is a consequence of increased expression of AGLs and their ability to target pericentromeric regions due to loss of DNA methylation9. DNA methylation was shown to antagonize binding of PHE135, supporting the proposed antagonism of DNA methylation and AGL binding. Our data (Supplementary Fig. 6A, + +<--- Page Split ---> + +B) and previously published data show that sirenRNAs negatively regulate AGLs in the endosperm30. Nevertheless, loss of maternal Pol IV function does not cause endosperm cellularization failure in \(Cr^{38}\) , suggesting an additional component causing this phenotype in paternal-excess hybrids. It was shown that many AGLs have higher expression levels in Cg compared to \(Cr^{6}\) . Consistently, the transcriptome data generated in this study revealed substantially higher expression of many PHE1-like (γ-type) AGLs in 4xCr, resembling the profile of Cg AGL genes (Fig. 5H). The same pattern applied for \(\alpha *\) like AGLs (Fig. 5I), which encode potential heterodimerization partners of \(\gamma\) type AGLs10. Based on this data we propose that increased expression of type I AGL genes in Cg and 4xCr causes increased expression of AGL targets in hypomethylated regions (Fig. 5G). The allopolyploid Cbp behaved like Cr in crosses with Cg (Fig. 1A, C), which was reflected by reduced expression of several AGLs in Cbp compared to Cg. This data supports the idea that increased dosage of AGLs in hybrid endosperm is a critical determinant for hybrid seed arrest. Similarly, the expression of \(\gamma\) and \(\alpha *\) type AGLs in Co was lower compared to Cg, resembling that of Cr. While several AGLs were higher expressed in Co compared to Cr, the CrPHE1 ortholog Carubv10020903m.g9 was nearly twice as strongly expressed in Cr (Fig. 5H), possibly connecting to the paternal-excess effect of Cr when crossed to Co (Fig. 1A). Furthermore, the majority of AGLs were upregulated to substantially higher levels in paternal excess hybrid seeds compared to nrpd1 seeds, supporting the idea that increased expression of AGLs determines the phenotypic difference between paternal excess hybrid seeds and seeds lacking maternal NRPD1 function (Fig. 5J). + +Together, based on our data we propose that seed abortion in response to paternal- excess interspecies and interploidy hybridizations is triggered by two factors, depletion of maternally derived sirenRNAs and increased expression of AGLs targeting hypomethylated regions in hybrid endosperm. + +## Discussion + +In this study, we reveal that the dosage of maternal Pol IV- dependent sirenRNAs plays a pivotal role in determining the success of hybridization in Capsella. Our findings demonstrate that sirenRNAs establish chromatin condensation and DNA methylation in the endosperm, critical processes that are disrupted in maternal nrpd1 mutant endosperm and paternal- excess hybrid endosperm. Consistent with the similar + +<--- Page Split ---> + +molecular and chromatin phenotypes observed in hybrid paternal- excess and maternal nrpd1 mutant endosperm, there was a strong depletion of sirenRNAs in the hybrid, underscoring the impact of hybridization on sirenRNA production. This depletion of sirenRNAs in both hybrid and maternal nrpd1 mutant endosperm correlated with a substantial increase in mRNA levels of genes overlapping siren loci, highlighting sirenRNAs' role as negative regulators of gene expression. Siren loci exhibit extensive methylation in wild- type endosperm (Fig. 4D, E). The transcriptional alterations observed in nrpd1 and hybrid endosperm align with a reduction of DNA methylation at siren loci, providing strong support for the role of sirenRNAs in directing DNA methylation. Our data also suggests, in line with previous findings in male meiotic and ovules29,39, that sirenRNAs may act in trans by guiding DNA methylation to non- perfectly matching TEs and protein- coding genes. Genes targeted by sirenRNAs that undergo upregulation in hybrid endosperm encompass several genes with potential involvement in establishing hybridization barriers, including SUVH7, YUC10, ARF21, and PHE1 orthologs32,34,35. This strongly implies a direct connection between the depletion of sirenRNAs and the arrest of hybrid seed development. + +SirenRNAs, originating predominantly from the maternal source (16 and our own data), likely guide methylation in the endosperm post- fertilization. Since heightened maternal genome dosage can ameliorate hybrid seed defects while concurrently inducing DNA hypermethylation (Fig. 2 D, E), we posit that dosage of maternally provided sirenRNAs is insufficient to repress sirenRNA targets in paternal excess hybrids. Consistently, interploidy \(Cr \times 4x Cr\) seeds exhibit similar molecular aberrations to \(Cr \times Cg\) hybrids, strongly corroborating the notion that the relative dosage of maternal sirenRNAs to their targets is a key determinant of hybridization success. Loss of RDR2 that together with Pol IV is required for siRNA production13,40 causes strong seed defects in \(Cg\) , differing from mild defects of \(rdr2\) mutants in \(Cr^{41}\) . This data aligns with our finding that \(Cg\) produces higher levels of sirenRNAs than \(Cr\) , suggesting that sirenRNA levels adapt to the expression strength of their targets. Among those targets are \(AGL\) genes that we found to scale in expression with the effective ploidy of \(Capsella\) species \((Co< Cr = Cpb< Cg = 4xCr)\) . Differential expression level of \(AGLs\) in \(Capsella\) is likely driven by mating system divergence. Based on theoretical predictions, genes promoting endosperm growth are stronger expressed in outbreeding compared to inbreeding species22. Consistent with these predictions, \(Cg\) is an outbreeder and has + +<--- Page Split ---> + +strongest \(AGL\) expression levels, followed by the recent inbreeders \(Cr\) and \(Cbp\) and concluded by the ancient inbreeder Co (Fig 5H). + +The maternal \(nrpd1\) mutant exhibited a phenotype mirroring that of paternal- excess hybrids in terms of alterations in DNA methylation and chromatin condensation. However, in contrast to hybrid seeds, the \(nrpd1\) mutant seeds did not undergo abortion38. One probably decisive difference between hybrid and \(nrpd1\) endosperm is the extent of \(AGL\) deregulation, which may account for the difference in phenotype. While \(AGL\) s targeted by sirenRNAs were also upregulated in \(nrpd1\) endosperm, most experienced higher upregulation in the hybrid endosperm (Fig. 5J). This discrepancy is likely attributed to the amplified \(AGL\) expression in the paternal parents of hybrids, inducing a more pronounced response in hybrid endosperm compared to the \(nrpd1\) mutant. + +Given that \(AGL\) target genes are notably concentrated in pericentromeric regions, the combined effects of diminished DNA methylation and increased \(AGL\) activity likely accounts for the heightened expression of \(AGL\) targets in paternal- excess hybrids. The same scenario applies to interploidy \(Cr \times 4x\) \(Cr\) seeds, which exhibit a substantial overlap in hyperactivated \(AGL\) target genes with paternal excess hybrids. Mutations in several \(AGL\) s were shown to weaken barriers associated with interploidy and interspecies hybridization35,42,43, providing further support for this scenario. Taken together, based on our findings we propose that maternal sirenRNAs serve as a dosage- sensitive factor that decisively influences the success of hybridization between plant species. + +## Methods + +## Plant material and growth conditions + +The following accessions of different Capsella species have been used in this study: \(Cr 48.21, 4xCr 48.21, Cbp 27.4, Cg 23.5\) and Co 1719. Tetraploid Co seeds were kindly provided by Martin Lascoux44. Seeds were surface sterilized with \(30\%\) bleach and \(70\%\) ethanol, rinsed with distilled water and sown on agar plates containing \(1/2\) Murashige and Skoog (MS) medium and \(1\%\) sucrose. Seeds were then stratified for 2 days in the + +<--- Page Split ---> + +darkness at \(4^{\circ}C\) and moved into a growth chamber with long- day photoperiod (16 h and \(22^{\circ}C\) light, \(8h\) and \(19^{\circ}C\) darkness) with a light intensity of \(110\mu \mathrm{E}\) . 7 day old seedlings were transferred to pots filled with sterile soil and plants were grown in a growth chamber with \(60\%\) humidity and daily cycles of \(16h\) light at \(21^{\circ}C\) and \(8h\) darkness at \(18^{\circ}C\) with a light intensity of \(150\mu \mathrm{E}\) . For germination tests, dry seeds were stored for 30 days to break dormancy. Seeds were then surface sterilized and sown on agar plates as described above. Seeds were stratified for 2 days at \(4^{\circ}C\) , then moved to a growth chamber and scored after 7 days for seedling establishment. + +## Histological and fluorescence analyses + +Manually pollinated siliques were harvested after \(4 - 7\) DAP and processed for clearing and Feulgen staining as previously described in \(^{33}\) . + +## Capsella grandiflora sequencing and genome assembly + +DNA was extracted from young leaf material of \(Cg\) using the CTAB method \(^{45}\) . DNA was sized on the BluePippin system (Sage science, Beverly, USA), to remove small fragments, and sequenced on one Oxford Nanopore PromethION flow cell. The pore version used was R9.4.1 and the PromethION release version was 19.05.1. A total of 49.2 Gb was sequenced, corresponding to about \(200x\) coverage of the estimated genome size of approximately \(250\mathrm{Mb}\) (roughly calculated by flow cytometry analysis). Additionally, \(Cg\) DNA was used to prepare an Illumina overlap library. DNA was sheared using a Covaris M220 (Covaris, Inc., Massachusetts) to \(450\mathrm{bp}\) . The sheared DNA was sized using a BluePippin prep system and used to prepare a PCR- free Illumina sequencing library. The library was sequenced on the Illumina NovaSeq SP platform using an SP flow cell and \(2\times 250\mathrm{bp}\) protocol. + +A selection of the longest Oxford nanopore PromethION sequence reads, together representing \(60x\) haploid genome coverage, were assembled using Minimap2 (v2.16- r922, with settings \(- m1600 - K2G - I8G)^{46}\) and Miniasm (v0.2- r137- dirty, with settings \(- R - c2 - m500 - s4000)^{47}\) . A consensus sequence was generated through three iterations of Racon (v.1.4.10, default settings) \(^{48}\) , using all sequence reads. The Illumina read pairs were then used to polish the consensus through three rounds of BWA- mem (v0.7.17; default parameters) \(^{49}\) and Pilon (v.1.22, default parameters) \(^{50}\) . + +<--- Page Split ---> + +Finally, purge_dups \(^{51}\) was used (v15082019, default parameters) to generate a haploid representation of the heterozygous Cg assembly. + +## RNA sequencing + +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and collected in \(20 \mu \mathrm{l}\) of RNA later solution (Sigma- Aldrich). Each sample consisted of 10- 15 siliques. RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) according to manufacturer's instruction. 500 ng of RNA was used for RNA Sequencing library preparation using the NEBNext Poly(A)mRNA Magnetic Isolation Module and NEBNext Ultra RNA LibraryPrep Kit for Illumina. Three biological replicates were generated for each genotype. Libraries were sequenced on an Illumina HiSeq X machine in 150 bp paired- end mode. + +## RNA sequencing analysis + +Adapter trimming was performed using Trim galore with the following parameters: -- three_prime_clip_R1 15 --three_prime_clip_R2 15 --clip_R1 10 --clip_R2 10. Sequencing reads were aligned to the \(Cr\) genome v. 1.0 (Phytozome) using HISAT \(^{52}\) . Reads were assigned to genes with featureCounts from the Bioconductor Rsubread package \(^{53}\) . Differentially regulated genes were detected using DESEQ \(^{54}\) . Genes were considered as upregulated based on log2(fold change) \(>1\) and padj \(< 0.05\) in comparison to the maternal parent (for incompatible hybrids) or to \(Cr \times Cr\) (for crosses with the nrp1 mutant). Pericentromeric regions were defined as in \(^{55}\) . + +## Bisulfite sequencing + +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and the endosperm was manually dissected as previously described \(^{25}\) . Samples were stored at - 20 °C and then used for genomic DNA isolation using the DNeasy Plant Mini Kit (Qiagen). Biological duplicates were generated for each genotype. Libraries were prepared with the Accel- NGS Methyl- Seq DNA Library Kit from Illumina, and the sequencing was performed on an Illumina NovaSeq 6000 machine in 150- bp paired- end mode. + +<--- Page Split ---> + +## Bioinformatic analysis of BS-Seq data + +For DNA methylation analysis, the 150- bp- long paired- end reads were first quality trimmed by removing the first 5 bases from the 5' end and the last 15 bases from the 3' end. Reads were mapped to the \(Cr\) reference genome in paired- end mode (- score_min L,0- 0.6) using Bismark v0.16.356. Mapped reads were deduplicated and cytosine methylation values calculated using the Bismark Methylation Extractor. + +Differentially methylated regions (DMRs) in CG, CHG and CHH contexts were calculated using 50bp windows across the genome as units. Only hypomethylated regions \((Cr\times Cr > Cr\times Cg,Cr\times Cr > Cr\times 4xCr,Cr\times Cr > nrpd1\times Cr)\) were considered. Windows with differences in fractional methylation below the 1st decile (Fisher's exact test p- value \(< 0.01\) ) were selected and these were merged if they occurred within 300 bp. + +## Small RNAs sequencing + +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and squeezed on facial tissue paper to release the endosperm. Seed coats and embryos were removed with forceps. A drop of RNA later solution (Sigma- Aldrich) was added to the tissue containing the absorbed endosperm, transferred to an Eppendorf tube and frozen in liquid nitrogen. Each sample consisted of endosperm form about 600 seeds. Small RNAs were extracted using the mirVana miRNA Isolation Kit (ThermoFisher Scientific) following the manufacturer instruction for sRNA. Libraries were prepared using the NEBNext® Multiplex Small RNA Library Prep Set for Illumina® with 50 - 100 ng of input for each sample. Size selection was performed on a 6% polyacrylamide gel and bands corresponding to about 150 bp were cut and purified from the gel for further analysis. Sequencing was performed on a NovaSeq 6000 machine in 150- bp paired- end mode. + +## Bioinformatic analyses of small RNA-seq data + +Adapters were removed from the first read of the 150 bp long read pair of each library using cutadapt and the resulting 18- 25 bp long reads were selected. Reads belonging to chloroplast, mitochondria and structural noncoding RNAs (tRNAs, snRNAs, rRNAs, or snoRNAs) were removed using bowtie (- v 1). + +<--- Page Split ---> + +Remaining reads were mapped to the \(Cr\) or \(Cg\) genome (see above) and sRNA loci were annotated with ShortStack \(^{27,57}\) using filtered reads from all generated libraries and sRNA from leaves \(^{38}\) . Options used for ShortStack were –mismatches 0, –mmap u, –mincov 0.5 rpm, and –pad 75. Replicates were checked for consistency by principal component analysis using the vegan 2.6- 4 package in \(R^{58}\) . Clusters with differential accumulation of sRNAs were identified with DESEQ \(^{254}\) . Genes and TEs were considered as depleted of sRNAs based on \(\log 2\) (fold change) \(< 0\) and \(\text{padj} < 0.1\) in comparison to the maternal parent (for incompatible hybrids) or to \(Cr \times Cr\) (for crosses with the \(nrpd1\) mutant). + +## Endosperm nuclei spreading + +Endosperm nuclei spreading was performed as previously described \(^{9}\) . + +## Quantification of chromocenter condensation + +Each endosperm nucleus was saved as single .tif file and then analyzed in Fiji \(^{59}\) . Particle Analysis was run after threshold adjustment with MaxEntropy. Only particles larger than \(0.05 \mu \text{m}^2\) were considered. Circularity of the particles was used as a proxy for chromocenter condensation; with values ranging between 0 and 1 (1 corresponds to a perfect circle). Each nucleus was represented by the mean circularity of all chromocenters. For each genotype, around 50 nuclei were analyzed. + +## Motif search + +PHE1 binding motifs were previously identified in \(^{35}\) . The motif file was used for scanning promoter sequences (1kb upstream of the ATG) of Capsella genes with findMotifsGenome.pl function from HOMER \(^{60}\) . + +## Data availability + +The original data files for RNA seq and DNA bisulfite sequencing can be obtained from the NCBI Gene Expression Omnibus (GSE246468). The assembled \(Cg\) genome sequence has been deposited at DDBJ/ENA/GenBank under the accession JAVXYZ000000000. + +<--- Page Split ---> + +## Acknowledgements + +We thank Tainlin Duan and Martin Lascoux from Uppsala University for sharing tetraploid Co seeds. We are grateful to Heinrich Bente for useful suggestions about smallRNA library preparation and to Marco Incarbone and members of the Köhler group for critical reading and helpful comments on the manuscript. This research was supported by grants from the Knut and Alice Wallenberg Foundation (2018- 0206 and 2019- 0062 to CK), and the Max Planck Society. + +<--- Page Split ---> + +## References + +1. Coughlan, J.M. The role of hybrid seed inviability in angiosperm speciation. Am J Bot 110, 1-14 (2023).2. Sekine, D. et al. Dissection of two major components of the post-zygotic hybridization barrier in rice endosperm. Plant J 76, 792-9 (2013).3. Scott, R.J., Spielman, M., Bailey, J. & Dickinson, H.G. Parent-of-origin effects on seed development in Arabidopsis thaliana. Development 125, 3329-3341 (1998).4. Pennington, P.D., Costa, L.M., Gutierrez-Marcos, J.F., Greenland, A.J. & Dickinson, H.G. When genomes collide: aberrant seed development following maize interploidy crosses. Ann Bot 101, 833-43 (2008).5. Johnston, S.A. & Hanneman Jr., R.E. Manipulations of endosperm balance number overcome crossing barriers between diploid Solanum species. Science 217, 446-448 (1982).6. Rebernig, C.A., Lafon-Placette, C., Hatorangan, M.R., Slotte, T. & Kohler, C. Non-reciprocal interspecies hybridization barriers in the Capsella genus are established in the endosperm. PLoS Genet 11, e1005295 (2015).7. Roth, M., Florez-Rueda, A.M. & Stadler, T. Differences in effective ploidy drive genome-wide endosperm expression polarization and seed failure in wild tomato hybrids. Genetics 212, 141-152 (2019).8. Tonosaki, K. et al. Overcoming the species hybridization barrier by ploidy manipulation in the genus Oryza. Plant J 93, 534-544 (2018).9. Dziasek, K. et al. Hybrid seed incompatibility in Capsella is connected to chromatin condensation defects in the endosperm. PLoS Genet 17, e1009370 (2021).10. Qiu, Y. & Kohler, C. Endosperm evolution by duplicated and neofunctionalized type I MADS-box transcription factors. Mol Biol Evol 39, msab355 (2022).11. Martinez, G. et al. Paternal easiRNAs regulate parental genome dosage in Arabidopsis. Nat Genet 50, 193-198 (2018).12. Satyaki, P.R.V. & Gehring, M. Paternally acting canonical RNA-directed DNA methylation pathway genes sensitize Arabidopsis endosperm to paternal genome dosage. Plant Cell 31, 1563-1578 (2019).13. Cuerda-Gil, D. & Slotkin, R.K. Non-canonical RNA-directed DNA methylation. Nat Plants 2, 16163 (2016).14. Erdmann, R.M., Satyaki, P.R.V., Klosinska, M. & Gehring, M. A small RNA pathway mediates allelic dosage in endosperm. Cell Rep 21, 3364-3372 (2017).15. Rodrigues, J.A. et al. Imprinted expression of genes and small RNA is associated with localized hypomethylation of the maternal genome in rice endosperm. Proc Natl Acad Sci U S A 110, 7934-9 (2013).16. Grover, J.W. et al. Abundant expression of maternal siRNAs is a conserved feature of seed development. Proc Natl Acad Sci U S A 117, 15305-15315 (2020).17. Lu, J., Zhang, C., Baulcombe, D.C. & Chen, Z.J. Maternal siRNAs as regulators of parental genome imbalance and gene expression in endosperm of Arabidopsis seeds. Proc Natl Acad Sci U S A 109, 5529-34 (2012).18. Satyaki, P.R.V. & Gehring, M. RNA Pol IV induces antagonistic parent-of-origin effects on Arabidopsis endosperm. PLoS Biol 20, e3001602 (2022).19. Ishikawa, R. et al. Rice interspecies hybrids show precocious or delayed developmental transitions in the endosperm without change to the rate of syncytial nuclear division. Plant J 65, 798-806 (2010).20. Lafon-Placette, C. & Kohler, C. Endosperm-based postzygotic hybridization barriers: developmental mechanisms and evolutionary drivers. Mol Ecol 25, 2620-2629 (2016).21. Lafon-Placette, C. et al. Paternally expressed imprinted genes associate with hybridization barriers in Capsella. Nat Plants 4, 352-357 (2018). + +<--- Page Split ---> + +22. Brandvain, Y. & Haig, D. Divergent mating systems and parental conflict as a barrier to hybridization in flowering plants. Am Nat 166, 330-8 (2005).23. Douglas, G.M. et al. Hybrid origins and the earliest stages of diploidization in the highly successful recent polyploid Capsella bursa-pastoris. Proc Natl Acad Sci U S A 112, 2806-11 (2015).24. Slotte, T. et al. The Capsella rubella genome and the genomic consequences of rapid mating system evolution. Nat Genet 45, 831-835 (2013).25. Hatorangan, M.R., Laenen, B., Steige, K.A., Slotte, T. & Kohler, C. Rapid Evolution of Genomic Imprinting in Two Species of the Brassicaceae. Plant Cell 28, 1815-27 (2016).26. Grover, J.W. et al. Maternal components of RNA-directed DNA methylation are required for seed development in Brassica rapa. Plant J 94, 582 (2018).27. Axtell, M.J. ShortStack: comprehensive annotation and quantification of small RNA genes. RNA 19, 740-51 (2013).28. Mi, S. et al. Sorting of small RNAs into Arabidopsis organate complexes is directed by the 5' terminal nucleotide. Cell 133, 116-27 (2008).29. Burgess, D., Chow, H.T., Grover, J.W., Freeling, M. & Mosher, R.A. Ovule siRNAs methylate protein-coding genes in trans. The Plant Cell 34, 3647-3664 (2022).30. Kirkbride, R.C. et al. Maternal small RNAs mediate spatial-temporal regulation of gene expression, imprinting, and seed development in Arabidopsis. Proc Natl Acad Sci U S A 116, 2761-2766 (2019).31. Florez-Rueda, A.M., Fiscalini, F., Roth, M., Grossniklaus, U. & Stadler, T. Endosperm and seed transcriptomes reveal possible roles for small RNA pathways in wild tomato hybrid seed failure. Genome Biol Evol 13, evab107 (2021).32. Wolff, P., Jiang, H., Wang, G., Santos-Gonzalez, J. & Kohler, C. Paternally expressed imprinted genes establish postzygotic hybridization barriers in Arabidopsis thaliana. Elife 4(2015).33. Batista, R.A., Figueiredo, D.D., Santos-Gonzalez, J. & Kohler, C. Auxin regulates endosperm cellularization in Arabidopsis. Genes Dev 33, 466-476 (2019).34. Butel, N., Qiu, Y., Xu, W., Santos-Gonzalez, J. & Kohler, C. The molecular basis of parental conflict driven regulation of endosperm cellularization. Preprint at: https://www.biorxiv.org/content/10.1101/2023.06.22.546051v1 (2023).35. Batista, R.A. et al. The MADS-box transcription factor PHERES1 controls imprinting in the endosperm by binding to domesticated transposons. Elife 8, e50541 (2019).36. Yang, L. & Bennetzen, J.L. Structure-based discovery and description of plant and animal Helitrons. Proc Natl Acad Sci U S A 106, 12832-7 (2009).37. Hu, K. et al. Helitron distribution in Brassicaceae and whole genome helitron density as a character for distinguishing plant species. BMC Bioinformatics 20, 354 (2019).38. Wang, Z. et al. Polymerase IV plays a crucial role in pollen development in Capsella. Plant Cell 32, 950-966 (2020).39. Long, J. et al. Nurse cell--derived small RNAs define paternal epigenetic inheritance in Arabidopsis. Science 373(2021).40. Law, J.A. & Jacobsen, S.E. Establishing, maintaining and modifying DNA methylation patterns in plants and animals. Nat Rev Genet 11, 204-20 (2010).41. Dew-Budd, K.J. et al. Mating system is associated with seed phenotypes upon loss of RNA-directed DNA methylation in Brassicaceae. Plant Physiol (2023).42. Hehenberger, E., Kradolfer, D. & Köhler, C. Endosperm cellularization defines an important developmental transition for embryo development. Development 139, 2031-2039 (2012).43. Josefsson, C., Dilkes, B. & Comai, L. Parent-dependent loss of gene silencing during interspecies hybridization. Curr Biol 16, 1322-1328 (2006).44. Duan, T., Sicard, A., Glemin, S. & Lascoux, M. Separating phases of allopolyploid evolution with resynthesized and natural Capsella bursa-pastoris. Elife 12, RP88398 (2023).45. Doyle, J.J. & Doyle, J.L. Isolation of plant DNA from fresh tissue. Focus (Madison) 12, 13-15 (1990). + +<--- Page Split ---> + +46. Li, H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 34, 3094-3100 (2018).47. Li, H. Minimap and miniasm: fast mapping and de novo assembly for noisy long sequences. Bioinformatics 32, 2103-10 (2016).48. Magoc, T. & Salzberg, S.L. FLASH: fast length adjustment of short reads to improve genome assemblies. Bioinformatics 27, 2957-63 (2011).49. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754-60 (2009).50. Walker, B.J. et al. Pilon: an integrated tool for comprehensive microbial variant detection and genome assembly improvement. PLoS One 9, e112963 (2014).51. Guan, D. et al. Identifying and removing haplotypic duplication in primary genome assemblies. Bioinformatics 36, 2896-2898 (2020).52. Kim, D., Paggi, J.M., Park, C., Bennett, C. & Salzberg, S.L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol 37, 907-915 (2019).53. Liao, Y., Smyth, G.K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923-30 (2014).54. Love, M.I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014).55. Koenig, D. et al. Long-term balancing selection drives evolution of immunity genes in Capsella. Elife 8, e43606 (2019).56. Krueger, F. & Andrews, S.R. Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 27, 1571-2 (2011).57. Johnson, N.R., Yeoh, J.M., Coruh, C. & Axtell, M.J. Improved Placement of Multi-mapping Small RNAs. G3 (Bethesda) 6, 2103-11 (2016).58. Oksanen, J. et al. vegan: Community ecology package. R package. 2.6-2 edn (2022).59. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676-82 (2012).60. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38, 576-89 (2010). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig 1. Interploidy and interspecies hybrid seeds share similar phenotypes and deregulated genes.
+ +A. Cleared hybrid Capsella seeds at 4 to 6 days after pollination (DAP). Corresponding mature seeds are shown on the right side of each panel. Scale bars correspond to \(100 \mu \mathrm{m}\) and \(1 \mathrm{mm}\) for cleared and mature seeds, respectively. Color code reflects effective ploidy as indicated in the figure. B. Mature seed phenotypes of crosses \(Cr \times Cbp\) and \(Cbp \times Cr\). C. Heatmap showing log2 fold changes (compared to the corresponding maternal parent) of genes upregulated in + +<--- Page Split ---> + +the analyzed hybrids. D. Upset plot showing the overlap of upregulated genes between interploidy \((Cr \times 4x Cr)\) and interspecies crosses of different Capsella species. E. Pictures of seeds from \(Co \times Cr\) and \(4xCo \times Cr\) . Scale bars represent 1mm. F. Corresponding germination data to crosses shown in panel E. Shown is the fraction of germinating seeds from each cross, with each filled circle representing one biological replicate. Numbers on top represent total seed numbers. Asterisks represent statistical significance as calculated by ANOVA with post hoc Tukey's HSD test (\\*\\* \(p < 0.001\) ). + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. Interspecies hybridization causes chromatin condensation defects in endosperm nuclei.
+ +A. Percent of genes in pericentromeric and non-pericentromeric regions that are upregulated in different incompatible Capsella hybrids. Asterisks represent statistical significance as calculated by the chi-square test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). B. DAPI stained chromocenters from endosperm nuclei at 4 DAP of different Capsella genotypes. Scale bar, 5 μm. C. Quantification of chromatin condensation given as mean circularity of chromocenters (see Methods). Numbers below boxes indicate number of analyzed nuclei. Asterisks represent statistical significance as calculated by the Wilcoxon test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). D. Metagene plots show DNA methylation levels of TEs in the endosperm of \(Cg \times Cg\) , \(4xCr \times Cg\) , and \(4xCr \times 4xCr\) 6 DAP seeds. D. Boxplots showing methylation level of TEs in the endosperm of 6 DAP seeds of the indicated genotypes. Boxes show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3. Maternal loss of NRPD1 mimics paternal-excess hybrid phenotypes.
+ +A. DAPI stained chromocenters from endosperm nuclei from \(Cr \times Cr\) , \(Cr \times nrdp1\) , \(nrdp1 \times Cr\) and \(nrdp1 \times nrdp1\) . Scale bar, 5 μm. B. Quantification of chromatin condensation given as mean circularity of chromocenters (see Methods). Numbers below boxes indicate number of analyzed nuclei. Asterisks represent statistical significance as calculated by the Wilcoxon test (*** p-value < 0.001). C. Metagene plots show DNA methylation levels of TEs in the endosperm of \(Cr \times Cr\) , \(nrdp1 \times nrdp1\) , \(Cr \times nrdp1\) , and \(nrdp1 \times Cr\) 6 DAP seeds. D. Boxplots showing methylation level of TEs in the endosperm of 6 DAP seeds of the indicated genotypes. Boxes show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (*** p-value < 0.001, ** p-value < 0.01, * p-value < 0.05). E. Upset plot showing the overlap of upregulated genes among \(nrdp1 \times nrdp1\) , \(nrdp1 \times Cr\) and \(Cr \times Cg\) derived seeds. Genes were considered as upregulated based on log2(fold change) >1 and padj <0.05 in comparison to \(Cr \times Cr\) . + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4. Maternal sirenRNAs are depleted in hybrid endosperm.
+ +A. Cumulative expression plot of 24 nt-dominant siRNA loci in endosperm and leaves. Only loci with expression \(\geq 2\) RPM in combined replicates were analyzed (n = 13957 in endosperm and 13461 in leaves). In endosperm and leaves, 0.99% versus 27.6% of highest expressed 24 nt-dominant loci account for 90% of siRNAs, respectively. B. Mean siRNA accumulation at siren loci in the endosperm of indicated genotypes and Cr leaves. siRNA abundance at each + +<--- Page Split ---> + +locus is normalized by locus length (in kb), and library size. Mean of two replicates is presented. Black vertical lines represent median. Individual measurements are shown as the rug below each density. Left panel shows data obtained by mapping reads to the \(Cr\) genome, right panel shows data obtained by mapping reads to the \(Cg\) genome. C. Venn diagram showing overlap among siren loci, loci with siRNAs depletion in \(Cr \times Cg\) hybrid and loci with siRNAs depletion in \(nrpd1 \times Cr\) (detected by DESeq2, log2(fold change) \(< 0\) , padj \(< 0.1\) ). D. siRNA accumulation and CHH methylation in the endosperm of indicated genotypes on selected genes overlapping siren loci (marked with violet lines): \(Carubv10025009m.g\) (AGL33), \(Carubv10011966m.g\) (YUC10), \(Carubv10028580m.g\) (AGL62), \(Carubv10010156m.g\) (AGL28), \(Carubv10002555m.g\) and \(Carubv10025621m.g\) (AGL33). Blue boxes represent genes, violet boxes represent TEs. E. Comparison of DNA methylation levels of genes and TEs overlapping siren loci with all genes and TEs in \(Cr\) genome. Boxes on this and following panels show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). F. Expression of genes overlapping siren loci in \(Cr \times Cr\) , \(Cr \times Cg\) and \(nrpd1 \times Cr\) seeds at 6 DAP. G. Heatmap showing log2 fold changes (compared to the corresponding maternal parent) of genes overlapping siren loci. H. Comparison of DNA methylation in \(Cr \times Cr\) , \(nrpd1 \times Cr\) and \(Cr \times Cg\) endosperm on upregulated genes overlapping siren loci. Asterisks indicate statistically significant differences as calculated by a one- sided Wilcoxon test (\\* p- value \(< 0.05\) ). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5. Loss of DNA methylation at trans sirenRNA targets and increased dosage of paternal AGLs contributes to seed incompatibility.
+ +A. CHG and CHH DNA methylation at sirenRNA trans targets (50 bp genomic windows, with \(\geq 1\) RPM realigning sirenRNAs and \(\leq 2\) mismatches) in \(Cr \times Cr\) endosperm. Barplot shows number of windows in each category, breaking points correspond to 0.7, 0.9, 0.98 and 0.995 quantiles. B. CHG and CHH DNA methylation at sirenRNA trans targets overlapping genes (coding region +2kb promoters (upper panel)) and transposable elements (TEs, lower panel). Barplot shows number of windows in each category, breaking points correspond to 0.5, 0.8 + +<--- Page Split ---> + +and 0.9 quantiles. C. Comparison of CHG and CHH DNA methylation levels in 50 bp windows differing in sirenRNA accumulation at trans targets. X- axis shows difference in DNA methylation \((Cr\times Cr - Cr\times Cg)\) and Y- axis difference in sirenRNA accumulation in trans \((Cr\) \(\times Cr - Cr\times Cg)\) . Barplot shows number of windows in each category, breaking points correspond to 0.6, 0.85 and 0.95 quantiles. D. Examples of genes with reduced sirenRNA accumulation at trans targets in \(Cr\times Cg\) , \(nrpd1\times Cr\) and \(nrpd1\times nrpd1\) compared to \(Cr\times Cr\) endosperm. Tracks show levels of sirenRNAs and CHH methylation at Carubv10021959m.g (F- box/RNI- like/FBD- like domain- containing protein), Carubv10007644m.g (MADS- box transcription factor), Carubv10020348m.g (F- box family protein) and Carubv10027768m.g. Blue boxes represent genes, violet boxes represent TEs. E. Plot shows distance to the nearest TE upstream of the transcriptional start site of upregulated genes in \(Cr\times Cg\) compared to non- upregulated genes. Boxes in this and following panels show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). F. Methylation level of TEs in promoters of upregulated genes in the \(Cr\times Cg\) hybrid. G. Distribution of PHE1 binding motifs in the Capsella genome in different groups of genes: all genes, genes upregulated in \(Cr\times Cr\) , genes upregulated in \(Cbp\times Cg\) , genes upregulated in \(Co\times Cr\) , genes upregulated in \(Cr\times 4xCr\) , genes upregulated in \(4xCr\times Cg\) . Values of pericentromeric/non- pericentromeric genes upregulated in different genotypes are compared to all pericentromeric/non- pericentromeric genes. NPC - non- pericentromeric region, PC - pericentromeric region. Asterisks indicate statistically significant differences calculated by Chi- square test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). H, I. Heatmaps showing expression (RPKM) of type I \(\gamma\) - MADS box (H) and \(\alpha^{*}\) - MADS box (I) transcription factors in different Capsella genotypes. J. Heatmap showing \(\log 2\) (fold change) (in comparison to maternal parent) of \(\gamma\) and \(\alpha^{*}\) type I MADS box TFs in different Capsella hybrids. Genes marked in bold are overlapping siren loci, genes marked in purple are targeted by trans- acting sirenRNAs. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryFigures.pdf SupplementaryTable1.xlsx SupplementaryTable2.xlsx SupplementaryTable3.xlsx SupplementaryTable4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0_det.mmd b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..98e18fd737fafdd341fbb14dc8ef78579e24fdca --- /dev/null +++ b/preprint/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0/preprint__0dbed8082b3128cfea9dd46422854e9f5373ff73f84cde909891329bd47eaff0_det.mmd @@ -0,0 +1,369 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 833, 176]]<|/det|> +# Dosage sensitive maternal siRNAs determine hybridization success in Capsella + +<|ref|>text<|/ref|><|det|>[[44, 196, 500, 460]]<|/det|> +Katarzyna Dziasek https://orcid.org/0000- 0002- 7279- 1417 Juan Santos- González Swedish University of Agricultural Sciences Yichun Qiu Max Planck Institute of Molecular Plant Physiology Diana Rigola KeyGene (Netherlands) Koen Nijbroek Keygene Claudia Köhler + +<|ref|>text<|/ref|><|det|>[[55, 470, 333, 487]]<|/det|> +koehler@mpimp- golm.mpg.de + +<|ref|>text<|/ref|><|det|>[[55, 500, 857, 522]]<|/det|> +Max Planck Institute of Molecular Plant Physiology https://orcid.org/0000- 0002- 2619- 4857 + +<|ref|>sub_title<|/ref|><|det|>[[44, 561, 103, 579]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 599, 135, 617]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 636, 337, 655]]<|/det|> +Posted Date: December 6th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 675, 475, 694]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3705839/v1 + +<|ref|>text<|/ref|><|det|>[[44, 712, 914, 755]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 772, 534, 792]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 828, 921, 872]]<|/det|> +Version of Record: A version of this preprint was published at Nature Plants on November 11th, 2024. See the published version at https://doi.org/10.1038/s41477- 024- 01844- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[191, 173, 810, 216]]<|/det|> +# Dosage sensitive maternal siRNAs determine hybridization success in Capsella + +<|ref|>text<|/ref|><|det|>[[175, 237, 821, 280]]<|/det|> +Katarzyna Dziasek \(^{1}\) , Juan Santos- González \(^{1}\) , Yichun Qiu \(^{2}\) , Diana Rigola \(^{3}\) , Koen Nijbroek \(^{3}\) , Claudia Köhler \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[149, 300, 844, 344]]<|/det|> +1 Department of Plant Reproductive Biology and Epigenetics, Max Planck Institute of Molecular Plant Physiology, Potsdam, 14476, Germany. + +<|ref|>text<|/ref|><|det|>[[149, 364, 846, 408]]<|/det|> +2 Department of Plant Biology, Uppsala BioCenter, Swedish University of Agricultural Sciences and Linnean Centre for Plant Biology, 75007, Uppsala, Sweden. + +<|ref|>text<|/ref|><|det|>[[147, 427, 820, 447]]<|/det|> +3 Keygene N.V., Agro Business Park 90, 6708 PW Wageningen, The Netherlands + +<|ref|>text<|/ref|><|det|>[[148, 467, 655, 486]]<|/det|> +\*Corresponding author. Email: koehler@mpimp- golm.mpg.de. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[149, 84, 227, 100]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[147, 120, 852, 485]]<|/det|> +Hybrid seed failure arising from wide crosses between plant species is a recurring obstacle in plant breeding, impeding the transfer of desirable traits. This postzygotic reproductive barrier primarily occurs in the endosperm, a dosage- sensitive tissue that nurtures the embryo and functions analogously to the placenta in mammals. Here, we show that the endosperm in incompatible seeds loses DNA methylation and chromatin condensation, resembling the endosperm of seeds depleted for maternal RNA Polymerase IV (Pol IV) function. This similarity in phenotype corresponds with a concurrent reduction in small interfering RNAs in the endosperm (sirenRNAs), maternal Pol IV- dependent siRNAs governing DNA methylation and regulating target genes. Notably, several AGAMOUS- LIKE MADS- BOX transcription factors (AGLs), known regulators of endosperm development, are targeted by sirenRNAs in cis and in trans. This finding aligns with the enrichment of AGL target genes among deregulated genes. Together, we propose that the response to interspecies hybridizations is triggered by a combination of reduced maternally derived sirenRNAs and heightened expression of AGLs targeting hypomethylated regions in the hybrid endosperm. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 445]]<|/det|> +Hybrid seed failure resulting from wide crosses between plant species poses a recurring challenge in plant breeding, obstructing the transfer of desirable traits between species. This postzygotic reproductive barrier predominantly manifests in the endosperm – an embryo nourishing tissue serving a similar function as the placenta in mammals \(^{1}\) . In most flowering plants, the endosperm is triploid, comprising two maternal and one paternal genome. This specific genomic ratio is pivotal for endosperm viability, as evidenced by seed abortion upon hybridization of plants with differing ploidy \(^{2 - 4}\) . Much like failures in interploidy hybridization, crosses between closely related species with equal ploidy often result in seed arrest. The term "effective ploidy" has been used to characterize this phenomenon, wherein certain species exhibit behavior reminiscent of polyploid species, despite their actual diploid nature \(^{5}\) . Remarkably, seeds arrested from both interploidy and interspecies crosses exhibit similar phenotypic and molecular abnormalities \(^{6 - 8}\) , implying the existence of a dosage- sensitive element contributing to this reproductive barrier within the endosperm. However, the precise nature of this component remains elusive. + +<|ref|>text<|/ref|><|det|>[[147, 464, 852, 904]]<|/det|> +In the Capsella genus, interspecies hybridizations between the inbreeder C. rubella and the outbreeder C. grandiflora lead to a loss of chromatin condensation and DNA methylation in the endosperm, particularly in the CHG and CHH contexts (H representing all bases except G). This chromatin change is associated with an accumulation of deregulated genes in pericentromeric regions, many of which are targeted by type I AGAMOUS- LIKE MADS- box transcription factors (AGLs) \(^{9}\) . Type I AGLs belonging to the \(\mathsf{M}_{\gamma}\) and \(\mathsf{M}_{\gamma}\) - interacting \(\mathsf{M}\alpha^{*}\) clades are key regulators of endosperm proliferation and cellularization \(^{10}\) . This data suggests that hypomethylation in the hybrid endosperm facilitates the access of AGLs to hypomethylated regions, resulting in detrimental gene activation. Similarly, in Arabidopsis, interploidy hybridizations of diploid (2x) maternal plants with tetraploid (4x) paternal plants causes endosperm hypomethylation in CHG and CHH context \(^{11}\) , which has been linked to a disruption of the RNA- directed DNA methylation (RdDM) pathway in the hybrid endosperm \(^{12}\) . This pathway initiates with the plant- specific DNA- dependent RNA Polymerase IV (Pol IV), generating short non- coding transcripts that are transformed into double- stranded RNA by RNA- DEPENDENT RNA POLYMEREASE 2 (RDR2) and further processed into 24 nt short interfering (si) RNAs by DICER- LIKE3 (DCL3). These siRNAs, when incorporated into ARGONAUTE (AGO) proteins, guide DNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 81, 852, 249]]<|/det|> +methyltransferases to target sequences, inducing DNA methylation13. Notably, both ovules and endosperm accumulate highly abundant siRNAs over a modest number of distinct loci, commonly referred to as sirenRNAs (small- interfering RNA in endosperm) that are primarily of maternal origin14- 16. In \*Arabidopsis\*, sirenRNAs target \*AGLs\* and their abundance is reduced in triploid seeds derived from 2x \*x4x\* crosses17. Nonetheless, whether sirenRNAs constitute the dosage- sensitive component determining hybridization success remains an open question. + +<|ref|>text<|/ref|><|det|>[[148, 268, 852, 409]]<|/det|> +Previous work revealed that loss of paternal Pol IV function can suppress the interplay hybridization barrier in \*Arabidopsis\*11,14. Interestingly, affected genes in the endosperm differ depending on whether the \*nrdp1\* mutation is maternally or paternally inherited, suggesting that maternal and paternal Pol IV- dependent siRNAs have distinct effects on endosperm development for reasons that remain to be fully elucidated11,18. + +<|ref|>text<|/ref|><|det|>[[147, 429, 852, 645]]<|/det|> +In this study, we demonstrate that the loss of maternal Pol IV function mirrors the loss of chromatin condensation and DNA methylation observed in \*Capsella\* hybrids. This resemblance in phenotype is correlated with a shared decrease in sirenRNAs, which regulate target genes in \*cis\* and \*trans\* by guiding DNA methylation. Among these sirenRNA targets are several \*AGLs\*, aligning with the enrichment of AGL target genes among deregulated genes. Based on this data we propose that the response to interspecies and interplay hybridizations is instigated by two factors: the depletion of maternally derived sirenRNAs and the heightened expression of \*AGLs\* targeting hypomethylated regions in hybrid endosperm. + +<|ref|>sub_title<|/ref|><|det|>[[153, 666, 223, 682]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[149, 704, 849, 747]]<|/det|> +## \*Capsella\* interplay and interspecies hybrids share similar phenotypic and morphological defects + +<|ref|>text<|/ref|><|det|>[[148, 767, 852, 910]]<|/det|> +To test whether interplay and interspecies hybridization share a common molecular basis, we analyzed the cross between diploid (2x) maternal and tetraploid (4x) paternal \*C. rubella\* (Cr) plants (referred to as \*Cr \* \*4xCr\*, by convention the female parent is always mentioned first). Like previously reported for paternal- excess interplay crosses in \*Arabidopsis\*3, the resulting 3x Cr seeds were dark and shriveled and failed to germinate (Fig. 1A, Supplementary Fig. 1A; the term paternal- excess will be used + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 273]]<|/det|> +to refer to crosses where the paternal parent has higher numerical or effective ploidy compared to the maternal parent). Defects of \(3x Cr\) seeds related to failed endosperm cellularization (Supplementary Fig. 1B), resembling previously reported defects of interspecies paternal- excess hybrid seeds in Capsella, Arabidopsis and other species \(^{3,6,19,20}\) . As in the case of interspecific hybrids \(^{21}\) , \(3x\) embryos could be rescued by removing them from the seeds and growing them in vitro (Supplementary Fig. 1C), supporting the view that the failure of \(3x\) embryo survival is a consequence of a defect in endosperm development. + +<|ref|>text<|/ref|><|det|>[[146, 291, 852, 730]]<|/det|> +To determine if phenotypic similarities in interplay and interspecies hybrids are reflected by a similar molecular signature, we generated transcriptome data of hybrid seeds (6 DAP) resulting from incompatible crosses. These included \(Cr \times 4xCr\) interplay crosses and previously characterized paternal- excess interspecies crosses of \(Cr \times C\) . grandiflora (Cg) and C. orientalis (Co) \(\times Cr^{21}\) with the corresponding intraspecific crosses. We furthermore included the cross of C. bursa pastoris (Cbp) \(\times Cg\) (Supplementary Table 1). Despite Cbp being an allotetraploid species, when utilized as the maternal parent in crosses with Cg, Cbp exhibited strikingly similar behavior to Cr. The resultant seeds exhibited a paternal excess phenotype, characterized by shrinkage and abortion, with embryos arrested at the torpedo stage (Fig. 1A). Correspondingly, reciprocal crosses of \(Cr \times Cbp\) gave rise to normal, viable seeds (Fig. 1B), revealing that Cbp, despite being a tetraploid species, behaved like a diploid. Based on the cross- incompatibilities between the species, we assigned an effective ploidy order of \(Co < Cr = Cpb < Cg = 4xCr\) . Consistent with theoretical predictions \(^{22}\) , this order aligned with the breeding mode of the species; while Cg is an outbreeder, Cr, Co and Cbp are inbreeding species. However, while Co became an inbreeder already \(\sim 900.000\) years ago, Cr and Cbp made the transition to inbreeding only \(\sim 200.000\) years back \(^{23,24}\) . + +<|ref|>text<|/ref|><|det|>[[147, 748, 852, 917]]<|/det|> +We identified deregulated genes in the hybrids by comparing the hybrid transcriptome with that of the corresponding maternal parent. Our data revealed that all tested hybrids shared a large set of commonly upregulated genes, indicative of a common genetic basis (Fig. 1 C, D). This conclusion is consistent with earlier work showing that increased ploidy of the maternal Cr parent can restore viability of hybrid \(Cr \times Cg\) seeds \(^{21}\) . To test the molecular consequences of increased maternal ploidy, we profiled the transcriptome of \(4xCr \times Cg\) seeds and found that most genes that were upregulated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 298]]<|/det|> +in \(Cr \times Cg\) hybrid seeds were either downregulated or unchanged in \(4xCr \times Cg\) seeds in comparison to the maternal parent (Fig. 1C). This data supports the idea that a dosage- sensitive component generated by the maternal parent determines hybridization success. To further challenge this idea, we tested whether we could suppress \(Co \times Cr\) hybrid seed defects by increasing the ploidy of the maternal Co parent. Seeds derived from \(Co \times Cr\) crosses aborted with similar morphological and transcriptional defects as \(Cr \times Cg\) , \(Cbp \times Cg\) and \(Cr \times 4xCr\) hybrids \(^{21}\) and (Fig. 1A, C, D), while seeds obtained from crosses of \(4xCo \times Cr\) were phenotypically like non- hybrid seeds and able to germinate (Fig. 1E, F). + +<|ref|>text<|/ref|><|det|>[[148, 317, 852, 410]]<|/det|> +Together, this data strongly suggests that interplay and interspecies hybridization barriers share a common molecular basis. Moreover, the fact that increased ploidy of the maternal parent can alleviate paternal- excess hybrid incompatibility, implies the existence of a quantitative maternal factor determining hybridization success. + +<|ref|>sub_title<|/ref|><|det|>[[149, 429, 724, 449]]<|/det|> +## Incompatible hybrids have decondensed pericentromeric regions + +<|ref|>text<|/ref|><|det|>[[147, 468, 852, 858]]<|/det|> +Previous work revealed that \(Cr \times Cg\) hybrid endosperm has decondensed chromocenters, which relates to a preferential expression of genes localized in pericentromeric regions \(^{9}\) . To test whether this phenomenon is a general pattern connected with hybrid incompatibility in the Capsella genus, we analyzed the location of upregulated genes in paternal- excess crosses in relation to their position on the chromosomes. Strikingly, we found that in all interspecies and interplay hybrids there was a preferential location of upregulated genes close to pericentromeric regions (Fig. 2A). We tested whether this phenomenon corresponded to a loss of chromosome condensation and analyzed chromocenter condensation in a range of paternal- excess incompatible hybrids. Consistent with the similar molecular phenotype of interspecies and interplay paternal- excess hybrid seeds, we found that all hybrid nuclei had significantly reduced chromocenter condensation compared to non- hybrid nuclei (Fig. 2B, C). Importantly, chromocenter condensation defects, DNA hypomethylation and gene deregulation was restored in nuclei of \(4xCr \times Cg\) hybrids, revealing that a dosage- sensitive maternal factor regulates chromatin condensation in the endosperm (Fig. 2A- E, Supplementary Fig. 2A, B). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 82, 852, 223]]<|/det|> +Decondensation of pericentromeric regions in \(Cr \times Cg\) hybrid endosperm co- occurred with a loss of CHG and CHH methylation9. Similarly, loss of chromatin condensation in nuclei of interploidy hybrids related to reduced CHG and CHH methylation on genes and TEs (Supplementary Fig. 2C- F). Together, this data reveals that interploidy and interspecies hybridization cause similar molecular defects consistent with the similar phenotypes of hybrid seeds. + +<|ref|>sub_title<|/ref|><|det|>[[149, 243, 850, 287]]<|/det|> +## Maternal loss of Pol IV function causes chromatin decondensation and loss of CHG and CHH methylation + +<|ref|>text<|/ref|><|det|>[[147, 300, 852, 894]]<|/det|> +The observed reduction of CHG and CHH methylation upon interploidy and interspecies hybridization9 (Supplementary Fig. 2C- F) suggests a role of the RdDM pathway in establishing hybridization barriers in the endosperm. Pol IV is required to produce precursor RNAs that are converted into 24 nt siRNAs guiding de novo DNA methylation in all sequence contexts13. We thus tested whether a mutant in NRPD1, encoding the large subunit of Pol IV, would cause a similar effect on chromatin condensation and loss of DNA methylation. Indeed, we found that maternal loss of NRPD1 reduced chromatin condensation in the endosperm, while loss of paternal NRPD1 had no effect and loss of maternal and paternal NRPD1 function did not enhance chromatin decondensation (Fig. 3A, B). Chromatin decondensation upon loss of maternal NRPD1 function was associated with loss of CHG and CHH methylation on genes and TEs in the endosperm, with similar regions being affected in interploidy and interspecies hybrids and seeds depleted of maternal NRPD1 function (Fig. 3C, D, Supplementary Fig. 3A- D). This was also reflected by a large set of similarly deregulated genes in the endosperm of seeds lacking maternal NRPD1 function and interploidy and interspecies hybrids (Fig. 3E, Supplementary Table 1). Previous studies found no evidence of NRPD1 being imprinted in Capsella21,25, suggesting that maternal Pol IV- dependent siRNAs are generated in maternal sporophytic tissues, consistent with previous work16,26. Together, we conclude that reduced dosage of maternal Pol IV- dependent siRNAs affects DNA methylation, chromatin condensation and gene expression in a similar manner as interploidy and interspecies hybridizations. This data raises the hypothesis that interploidy and interspecies hybrid endosperms are depleted of maternal Pol IV- dependent siRNAs, causing loss of DNA methylation, decondensation of pericentromeric regions and activation of genes. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[152, 82, 633, 101]]<|/det|> +## Maternal sirenRNAs are depleted in hybrid endosperm + +<|ref|>text<|/ref|><|det|>[[146, 120, 852, 880]]<|/det|> +To understand the role of maternal NRPD1 in hybrid incompatibility, we sequenced sRNAs in manually dissected endosperm of 6 DAP seeds derived from crosses of \(Cr \times Cr\) , \(Cg \times Cg\) , \(Cr \times Cg\) , \(nrpd1 \times Cr\) and \(nrpd1 \times nrpd1\) (Supplementary Table 2). Profiles of sRNAs correlated well among biological replicates of the same genotype but were clearly distinct between genotypes (Supplementary Fig. 4A, B). Consistent with previous work14, we found that 24 nt sRNAs were the predominant sRNA species in the endosperm (Supplementary Fig. 4C). Using ShortStack27 we identified 13957 clusters accumulating 24 nt sRNAs in the endosperm of \(Cr \times Cr\) seeds. Out of those, only a few loci produced high levels of sRNAs in \(Cr \times Cr\) endosperm, resembling the characteristics of siren loci15,16 (Fig. 4A, Supplementary Table 3). We identified 1389 loci giving rise to 90% of total sRNAs in \(Cr \times Cr\) endosperm that we will refer to as siren loci. As previously reported16, siren loci were longer than other loci expressing sRNAs (Supplementary Fig. 4D) and overlapped TEs, intergenic regions and genes (Supplementary Fig. 4E). SirenRNA loci were enriched for 24 nt sRNAs (Supplementary Fig. 4F) with an adenine bias at the 5' nucleotide (Supplementary Fig. 4G), suggesting interaction with AGO4- related AGO proteins28, consistent with previous observations in Brassica16. SirenRNA loci are distributed over the length of the chromosomes with an enrichment in pericentromeric regions (Supplementary Fig. 4H, I), but without a pronounced preference for any specific TE family (Supplementary Fig. 4J). Abundance of sirenRNAs was strongly depleted upon loss of maternal NRPD1 function and decreased further in nrpd1 homozygous endosperm (Fig. 4B), indicating that sirenRNAs are produced in maternal sporophytic tissues and in the endosperm. There are conflicting data on the origin of sirenRNAs; while they have been proposed to be predominantly maternally produced in Brassica16,29, biparental production was proposed in Arabidopsis18. Our data suggests that most likely both, the maternal seed coat and the endosperm are the source of sirenRNAs in the endosperm, with a predominant fraction being maternally derived. Like in maternal nrpd1 endosperm, in \(Cr \times Cg\) hybrid endosperm sirenRNAs abundance was strongly depleted (Fig. 4B). Nearly all loci losing sRNAs in \(Cr \times Cg\) endosperm corresponded to siren loci (Fig. 4C, D), strongly suggesting that maternally produced sirenRNAs are the dosage sensitive component that becomes depleted in hybrid endosperm. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 224]]<|/det|> +Mapping reads from \(Cr \times Cr\) and \(Cg \times Cg\) to their respective genomes revealed variations in the level and specificity of sirenRNAs. Specifically, when \(Cg \times Cg\) reads were mapped to the \(Cr\) genome, there was an apparent decrease in sirenRNA levels, which was also observed when mapping \(Cr \times Cr\) reads to the \(Cg\) genome. This data reveals that at least some sirenRNAs are species- specific and that \(Cg\) accumulates substantially higher levels of sirenRNAs compared to \(Cr\) (Fig. 4B, D). + +<|ref|>text<|/ref|><|det|>[[147, 243, 852, 387]]<|/det|> +Siren loci were heavily methylated, corresponding to the production of high levels of sirenRNAs (Fig. 4D, E). Depletion of sirenRNAs in \(Cr \times Cg\) and \(nrpd1 \times Cr\) endosperm corresponded to increased mRNA level of many genes overlapping those loci (Fig. 4F, G), indicating that sirenRNAs negatively regulate gene expression, consistent with previous work29,30. Transcriptional changes related to a loss of CHG and CHH methylation that however were not significant in the hybrid endosperm (Fig. 4D, H). + +<|ref|>text<|/ref|><|det|>[[147, 404, 852, 546]]<|/det|> +Depletion of maternal 24 nt siRNAs in the endosperm of hybrid tomato seeds was related to decreased expression of Pol IV subunits encoding genes \(NRPD1\) and \(NRPD2^{31}\) . We tested expression of RdDM components in hybrid seed transcriptomes and found significantly reduced \(NRPD1\) expression in \(Cr \times Cg\) , \(Co \times Cr\) and \(Cr \times 4x Cr\) endosperm (Supplementary Fig. 5), which may connect to loss of sirenRNAs in the hybrid. + +<|ref|>sub_title<|/ref|><|det|>[[148, 566, 850, 610]]<|/det|> +## Depletion of sirenRNAs in hybrid endosperm causes loss of methylation at trans targets + +<|ref|>text<|/ref|><|det|>[[147, 629, 852, 896]]<|/det|> +Previous work revealed that sirenRNAs can methylate other genomic sequences in trans29. We identified sirenRNA trans targets by mapping sirenRNAs to the genome masked for sirenRNA producing loci by allowing two mismatches. Indeed, we found that loci accumulating higher levels of non- perfectly matching sirenRNAs had higher methylation levels in CHG and CHH context (Fig. 5A). SirenRNAs targeted genes and TEs, with dosage- dependent effects on DNA methylation being most prominent on TEs (Fig. 5B). Together, this data strongly suggests that sirenRNAs guide DNA methylation in trans in the endosperm, like their proposed function in ovules29. Consistent with this idea, we found a decline of CHG and CHH methylation at trans targets in \(Cr \times Cg\) hybrid endosperm that correlated with the dosage of depleted sirenRNAs (Fig. 5C, D). We next asked whether there is a connection between sirenRNAs, loss of DNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 298]]<|/det|> +methylation at trans targets and changes in gene expression in the hybrid \(Cr \times Cg\) endosperm. Consistent with sirenRNAs most prominently targeting TEs (Fig. 5B), we found that genes upregulated in \(Cr \times Cg\) hybrids were closer to a TE upstream of their transcription start site than other genes (Fig. 5E). Moreover, the TEs in the promoters of those genes had reduced CHG and CHH methylation in \(Cr \times Cg\) hybrids (Fig. 5F). Among the genes that were targeted by trans- acting sirenRNAs and lost DNA methylation we found many upregulated AGLs (Supplementary Fig. 6A, B and Supplementary Table 4), consistent with previous findings reporting targeting of AGLs by maternal Pol IV- dependent siRNAs30. + +<|ref|>text<|/ref|><|det|>[[147, 317, 852, 533]]<|/det|> +Based on this data we conclude that sirenRNAs in the Capsella endosperm drive non- CG DNA methylation in trans. Depletion of sirenRNAs in the \(Cr \times Cg\) hybrid leads to loss of DNA methylation and increased expression of several trans targets, among them several genes with potential function in establishing hybridization barriers (Supplementary Table 4). These include SUVH7, which is involved in the triploid block in Arabidopsis32, YUC10, an auxin biosynthesis gene33 and ARF21 – a centromeric ARF regulating endosperm cellularization34. Moreover, among those genes are nine AGLs including AGL38 – the ortholog of PHE1 (Supplementary Fig. 6), which activates key regulators of the triploid block in Arabidopsis35. + +<|ref|>sub_title<|/ref|><|det|>[[150, 552, 835, 572]]<|/det|> +## Upregulated genes in hybrid endosperm are enriched for putative AGL targets + +<|ref|>text<|/ref|><|det|>[[147, 590, 852, 907]]<|/det|> +Previous work revealed that target sequence motifs of the AGL PHE1 are frequently located in helitron TEs35. Since helitrons in Arabidopsis and other Brassicaceae are concentrated in pericentromeric regions36,37, we tested whether pericentromeric regions in Capsella were enriched for PHE1 binding motifs. We found indeed that the frequency of genes containing PHE1 binding motifs was significantly higher in pericentromeric regions compared to chromosome arms (Fig. 5G). Importantly, in all analyzed incompatible crosses, the promoter regions of upregulated genes located in pericentromeric regions were significantly enriched for PHE1 binding sites (Fig. 5G). Based on this data we propose that increased expression of genes in pericentromeric regions in hybrid endosperm is a consequence of increased expression of AGLs and their ability to target pericentromeric regions due to loss of DNA methylation9. DNA methylation was shown to antagonize binding of PHE135, supporting the proposed antagonism of DNA methylation and AGL binding. Our data (Supplementary Fig. 6A, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 78, 852, 626]]<|/det|> +B) and previously published data show that sirenRNAs negatively regulate AGLs in the endosperm30. Nevertheless, loss of maternal Pol IV function does not cause endosperm cellularization failure in \(Cr^{38}\) , suggesting an additional component causing this phenotype in paternal-excess hybrids. It was shown that many AGLs have higher expression levels in Cg compared to \(Cr^{6}\) . Consistently, the transcriptome data generated in this study revealed substantially higher expression of many PHE1-like (γ-type) AGLs in 4xCr, resembling the profile of Cg AGL genes (Fig. 5H). The same pattern applied for \(\alpha *\) like AGLs (Fig. 5I), which encode potential heterodimerization partners of \(\gamma\) type AGLs10. Based on this data we propose that increased expression of type I AGL genes in Cg and 4xCr causes increased expression of AGL targets in hypomethylated regions (Fig. 5G). The allopolyploid Cbp behaved like Cr in crosses with Cg (Fig. 1A, C), which was reflected by reduced expression of several AGLs in Cbp compared to Cg. This data supports the idea that increased dosage of AGLs in hybrid endosperm is a critical determinant for hybrid seed arrest. Similarly, the expression of \(\gamma\) and \(\alpha *\) type AGLs in Co was lower compared to Cg, resembling that of Cr. While several AGLs were higher expressed in Co compared to Cr, the CrPHE1 ortholog Carubv10020903m.g9 was nearly twice as strongly expressed in Cr (Fig. 5H), possibly connecting to the paternal-excess effect of Cr when crossed to Co (Fig. 1A). Furthermore, the majority of AGLs were upregulated to substantially higher levels in paternal excess hybrid seeds compared to nrpd1 seeds, supporting the idea that increased expression of AGLs determines the phenotypic difference between paternal excess hybrid seeds and seeds lacking maternal NRPD1 function (Fig. 5J). + +<|ref|>text<|/ref|><|det|>[[148, 642, 850, 736]]<|/det|> +Together, based on our data we propose that seed abortion in response to paternal- excess interspecies and interploidy hybridizations is triggered by two factors, depletion of maternally derived sirenRNAs and increased expression of AGLs targeting hypomethylated regions in hybrid endosperm. + +<|ref|>sub_title<|/ref|><|det|>[[149, 756, 250, 773]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[148, 793, 852, 912]]<|/det|> +In this study, we reveal that the dosage of maternal Pol IV- dependent sirenRNAs plays a pivotal role in determining the success of hybridization in Capsella. Our findings demonstrate that sirenRNAs establish chromatin condensation and DNA methylation in the endosperm, critical processes that are disrupted in maternal nrpd1 mutant endosperm and paternal- excess hybrid endosperm. Consistent with the similar + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 470]]<|/det|> +molecular and chromatin phenotypes observed in hybrid paternal- excess and maternal nrpd1 mutant endosperm, there was a strong depletion of sirenRNAs in the hybrid, underscoring the impact of hybridization on sirenRNA production. This depletion of sirenRNAs in both hybrid and maternal nrpd1 mutant endosperm correlated with a substantial increase in mRNA levels of genes overlapping siren loci, highlighting sirenRNAs' role as negative regulators of gene expression. Siren loci exhibit extensive methylation in wild- type endosperm (Fig. 4D, E). The transcriptional alterations observed in nrpd1 and hybrid endosperm align with a reduction of DNA methylation at siren loci, providing strong support for the role of sirenRNAs in directing DNA methylation. Our data also suggests, in line with previous findings in male meiotic and ovules29,39, that sirenRNAs may act in trans by guiding DNA methylation to non- perfectly matching TEs and protein- coding genes. Genes targeted by sirenRNAs that undergo upregulation in hybrid endosperm encompass several genes with potential involvement in establishing hybridization barriers, including SUVH7, YUC10, ARF21, and PHE1 orthologs32,34,35. This strongly implies a direct connection between the depletion of sirenRNAs and the arrest of hybrid seed development. + +<|ref|>text<|/ref|><|det|>[[147, 488, 852, 903]]<|/det|> +SirenRNAs, originating predominantly from the maternal source (16 and our own data), likely guide methylation in the endosperm post- fertilization. Since heightened maternal genome dosage can ameliorate hybrid seed defects while concurrently inducing DNA hypermethylation (Fig. 2 D, E), we posit that dosage of maternally provided sirenRNAs is insufficient to repress sirenRNA targets in paternal excess hybrids. Consistently, interploidy \(Cr \times 4x Cr\) seeds exhibit similar molecular aberrations to \(Cr \times Cg\) hybrids, strongly corroborating the notion that the relative dosage of maternal sirenRNAs to their targets is a key determinant of hybridization success. Loss of RDR2 that together with Pol IV is required for siRNA production13,40 causes strong seed defects in \(Cg\) , differing from mild defects of \(rdr2\) mutants in \(Cr^{41}\) . This data aligns with our finding that \(Cg\) produces higher levels of sirenRNAs than \(Cr\) , suggesting that sirenRNA levels adapt to the expression strength of their targets. Among those targets are \(AGL\) genes that we found to scale in expression with the effective ploidy of \(Capsella\) species \((Co< Cr = Cpb< Cg = 4xCr)\) . Differential expression level of \(AGLs\) in \(Capsella\) is likely driven by mating system divergence. Based on theoretical predictions, genes promoting endosperm growth are stronger expressed in outbreeding compared to inbreeding species22. Consistent with these predictions, \(Cg\) is an outbreeder and has + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[149, 81, 850, 126]]<|/det|> +strongest \(AGL\) expression levels, followed by the recent inbreeders \(Cr\) and \(Cbp\) and concluded by the ancient inbreeder Co (Fig 5H). + +<|ref|>text<|/ref|><|det|>[[147, 144, 852, 386]]<|/det|> +The maternal \(nrpd1\) mutant exhibited a phenotype mirroring that of paternal- excess hybrids in terms of alterations in DNA methylation and chromatin condensation. However, in contrast to hybrid seeds, the \(nrpd1\) mutant seeds did not undergo abortion38. One probably decisive difference between hybrid and \(nrpd1\) endosperm is the extent of \(AGL\) deregulation, which may account for the difference in phenotype. While \(AGL\) s targeted by sirenRNAs were also upregulated in \(nrpd1\) endosperm, most experienced higher upregulation in the hybrid endosperm (Fig. 5J). This discrepancy is likely attributed to the amplified \(AGL\) expression in the paternal parents of hybrids, inducing a more pronounced response in hybrid endosperm compared to the \(nrpd1\) mutant. + +<|ref|>text<|/ref|><|det|>[[147, 404, 852, 646]]<|/det|> +Given that \(AGL\) target genes are notably concentrated in pericentromeric regions, the combined effects of diminished DNA methylation and increased \(AGL\) activity likely accounts for the heightened expression of \(AGL\) targets in paternal- excess hybrids. The same scenario applies to interploidy \(Cr \times 4x\) \(Cr\) seeds, which exhibit a substantial overlap in hyperactivated \(AGL\) target genes with paternal excess hybrids. Mutations in several \(AGL\) s were shown to weaken barriers associated with interploidy and interspecies hybridization35,42,43, providing further support for this scenario. Taken together, based on our findings we propose that maternal sirenRNAs serve as a dosage- sensitive factor that decisively influences the success of hybridization between plant species. + +<|ref|>sub_title<|/ref|><|det|>[[149, 700, 226, 717]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[149, 738, 475, 756]]<|/det|> +## Plant material and growth conditions + +<|ref|>text<|/ref|><|det|>[[148, 776, 852, 895]]<|/det|> +The following accessions of different Capsella species have been used in this study: \(Cr 48.21, 4xCr 48.21, Cbp 27.4, Cg 23.5\) and Co 1719. Tetraploid Co seeds were kindly provided by Martin Lascoux44. Seeds were surface sterilized with \(30\%\) bleach and \(70\%\) ethanol, rinsed with distilled water and sown on agar plates containing \(1/2\) Murashige and Skoog (MS) medium and \(1\%\) sucrose. Seeds were then stratified for 2 days in the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 273]]<|/det|> +darkness at \(4^{\circ}C\) and moved into a growth chamber with long- day photoperiod (16 h and \(22^{\circ}C\) light, \(8h\) and \(19^{\circ}C\) darkness) with a light intensity of \(110\mu \mathrm{E}\) . 7 day old seedlings were transferred to pots filled with sterile soil and plants were grown in a growth chamber with \(60\%\) humidity and daily cycles of \(16h\) light at \(21^{\circ}C\) and \(8h\) darkness at \(18^{\circ}C\) with a light intensity of \(150\mu \mathrm{E}\) . For germination tests, dry seeds were stored for 30 days to break dormancy. Seeds were then surface sterilized and sown on agar plates as described above. Seeds were stratified for 2 days at \(4^{\circ}C\) , then moved to a growth chamber and scored after 7 days for seedling establishment. + +<|ref|>sub_title<|/ref|><|det|>[[149, 293, 495, 312]]<|/det|> +## Histological and fluorescence analyses + +<|ref|>text<|/ref|><|det|>[[149, 331, 850, 375]]<|/det|> +Manually pollinated siliques were harvested after \(4 - 7\) DAP and processed for clearing and Feulgen staining as previously described in \(^{33}\) . + +<|ref|>sub_title<|/ref|><|det|>[[149, 395, 640, 414]]<|/det|> +## Capsella grandiflora sequencing and genome assembly + +<|ref|>text<|/ref|><|det|>[[147, 428, 852, 693]]<|/det|> +DNA was extracted from young leaf material of \(Cg\) using the CTAB method \(^{45}\) . DNA was sized on the BluePippin system (Sage science, Beverly, USA), to remove small fragments, and sequenced on one Oxford Nanopore PromethION flow cell. The pore version used was R9.4.1 and the PromethION release version was 19.05.1. A total of 49.2 Gb was sequenced, corresponding to about \(200x\) coverage of the estimated genome size of approximately \(250\mathrm{Mb}\) (roughly calculated by flow cytometry analysis). Additionally, \(Cg\) DNA was used to prepare an Illumina overlap library. DNA was sheared using a Covaris M220 (Covaris, Inc., Massachusetts) to \(450\mathrm{bp}\) . The sheared DNA was sized using a BluePippin prep system and used to prepare a PCR- free Illumina sequencing library. The library was sequenced on the Illumina NovaSeq SP platform using an SP flow cell and \(2\times 250\mathrm{bp}\) protocol. + +<|ref|>text<|/ref|><|det|>[[147, 713, 852, 880]]<|/det|> +A selection of the longest Oxford nanopore PromethION sequence reads, together representing \(60x\) haploid genome coverage, were assembled using Minimap2 (v2.16- r922, with settings \(- m1600 - K2G - I8G)^{46}\) and Miniasm (v0.2- r137- dirty, with settings \(- R - c2 - m500 - s4000)^{47}\) . A consensus sequence was generated through three iterations of Racon (v.1.4.10, default settings) \(^{48}\) , using all sequence reads. The Illumina read pairs were then used to polish the consensus through three rounds of BWA- mem (v0.7.17; default parameters) \(^{49}\) and Pilon (v.1.22, default parameters) \(^{50}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 82, 850, 125]]<|/det|> +Finally, purge_dups \(^{51}\) was used (v15082019, default parameters) to generate a haploid representation of the heterozygous Cg assembly. + +<|ref|>sub_title<|/ref|><|det|>[[149, 145, 300, 164]]<|/det|> +## RNA sequencing + +<|ref|>text<|/ref|><|det|>[[148, 184, 852, 375]]<|/det|> +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and collected in \(20 \mu \mathrm{l}\) of RNA later solution (Sigma- Aldrich). Each sample consisted of 10- 15 siliques. RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) according to manufacturer's instruction. 500 ng of RNA was used for RNA Sequencing library preparation using the NEBNext Poly(A)mRNA Magnetic Isolation Module and NEBNext Ultra RNA LibraryPrep Kit for Illumina. Three biological replicates were generated for each genotype. Libraries were sequenced on an Illumina HiSeq X machine in 150 bp paired- end mode. + +<|ref|>sub_title<|/ref|><|det|>[[150, 395, 377, 414]]<|/det|> +## RNA sequencing analysis + +<|ref|>text<|/ref|><|det|>[[148, 433, 852, 625]]<|/det|> +Adapter trimming was performed using Trim galore with the following parameters: -- three_prime_clip_R1 15 --three_prime_clip_R2 15 --clip_R1 10 --clip_R2 10. Sequencing reads were aligned to the \(Cr\) genome v. 1.0 (Phytozome) using HISAT \(^{52}\) . Reads were assigned to genes with featureCounts from the Bioconductor Rsubread package \(^{53}\) . Differentially regulated genes were detected using DESEQ \(^{54}\) . Genes were considered as upregulated based on log2(fold change) \(>1\) and padj \(< 0.05\) in comparison to the maternal parent (for incompatible hybrids) or to \(Cr \times Cr\) (for crosses with the nrp1 mutant). Pericentromeric regions were defined as in \(^{55}\) . + +<|ref|>sub_title<|/ref|><|det|>[[150, 645, 332, 664]]<|/det|> +## Bisulfite sequencing + +<|ref|>text<|/ref|><|det|>[[148, 683, 852, 848]]<|/det|> +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and the endosperm was manually dissected as previously described \(^{25}\) . Samples were stored at - 20 °C and then used for genomic DNA isolation using the DNeasy Plant Mini Kit (Qiagen). Biological duplicates were generated for each genotype. Libraries were prepared with the Accel- NGS Methyl- Seq DNA Library Kit from Illumina, and the sequencing was performed on an Illumina NovaSeq 6000 machine in 150- bp paired- end mode. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[149, 82, 484, 101]]<|/det|> +## Bioinformatic analysis of BS-Seq data + +<|ref|>text<|/ref|><|det|>[[148, 120, 852, 238]]<|/det|> +For DNA methylation analysis, the 150- bp- long paired- end reads were first quality trimmed by removing the first 5 bases from the 5' end and the last 15 bases from the 3' end. Reads were mapped to the \(Cr\) reference genome in paired- end mode (- score_min L,0- 0.6) using Bismark v0.16.356. Mapped reads were deduplicated and cytosine methylation values calculated using the Bismark Methylation Extractor. + +<|ref|>text<|/ref|><|det|>[[148, 257, 852, 400]]<|/det|> +Differentially methylated regions (DMRs) in CG, CHG and CHH contexts were calculated using 50bp windows across the genome as units. Only hypomethylated regions \((Cr\times Cr > Cr\times Cg,Cr\times Cr > Cr\times 4xCr,Cr\times Cr > nrpd1\times Cr)\) were considered. Windows with differences in fractional methylation below the 1st decile (Fisher's exact test p- value \(< 0.01\) ) were selected and these were merged if they occurred within 300 bp. + +<|ref|>sub_title<|/ref|><|det|>[[150, 420, 364, 439]]<|/det|> +## Small RNAs sequencing + +<|ref|>text<|/ref|><|det|>[[147, 457, 852, 748]]<|/det|> +Seeds derived after manual pollination were dissected out of siliques at 6 DAP and squeezed on facial tissue paper to release the endosperm. Seed coats and embryos were removed with forceps. A drop of RNA later solution (Sigma- Aldrich) was added to the tissue containing the absorbed endosperm, transferred to an Eppendorf tube and frozen in liquid nitrogen. Each sample consisted of endosperm form about 600 seeds. Small RNAs were extracted using the mirVana miRNA Isolation Kit (ThermoFisher Scientific) following the manufacturer instruction for sRNA. Libraries were prepared using the NEBNext® Multiplex Small RNA Library Prep Set for Illumina® with 50 - 100 ng of input for each sample. Size selection was performed on a 6% polyacrylamide gel and bands corresponding to about 150 bp were cut and purified from the gel for further analysis. Sequencing was performed on a NovaSeq 6000 machine in 150- bp paired- end mode. + +<|ref|>sub_title<|/ref|><|det|>[[149, 767, 555, 786]]<|/det|> +## Bioinformatic analyses of small RNA-seq data + +<|ref|>text<|/ref|><|det|>[[148, 806, 851, 898]]<|/det|> +Adapters were removed from the first read of the 150 bp long read pair of each library using cutadapt and the resulting 18- 25 bp long reads were selected. Reads belonging to chloroplast, mitochondria and structural noncoding RNAs (tRNAs, snRNAs, rRNAs, or snoRNAs) were removed using bowtie (- v 1). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 298]]<|/det|> +Remaining reads were mapped to the \(Cr\) or \(Cg\) genome (see above) and sRNA loci were annotated with ShortStack \(^{27,57}\) using filtered reads from all generated libraries and sRNA from leaves \(^{38}\) . Options used for ShortStack were –mismatches 0, –mmap u, –mincov 0.5 rpm, and –pad 75. Replicates were checked for consistency by principal component analysis using the vegan 2.6- 4 package in \(R^{58}\) . Clusters with differential accumulation of sRNAs were identified with DESEQ \(^{254}\) . Genes and TEs were considered as depleted of sRNAs based on \(\log 2\) (fold change) \(< 0\) and \(\text{padj} < 0.1\) in comparison to the maternal parent (for incompatible hybrids) or to \(Cr \times Cr\) (for crosses with the \(nrpd1\) mutant). + +<|ref|>sub_title<|/ref|><|det|>[[150, 318, 405, 336]]<|/det|> +## Endosperm nuclei spreading + +<|ref|>text<|/ref|><|det|>[[150, 355, 720, 374]]<|/det|> +Endosperm nuclei spreading was performed as previously described \(^{9}\) . + +<|ref|>sub_title<|/ref|><|det|>[[149, 395, 550, 413]]<|/det|> +## Quantification of chromocenter condensation + +<|ref|>text<|/ref|><|det|>[[148, 432, 852, 576]]<|/det|> +Each endosperm nucleus was saved as single .tif file and then analyzed in Fiji \(^{59}\) . Particle Analysis was run after threshold adjustment with MaxEntropy. Only particles larger than \(0.05 \mu \text{m}^2\) were considered. Circularity of the particles was used as a proxy for chromocenter condensation; with values ranging between 0 and 1 (1 corresponds to a perfect circle). Each nucleus was represented by the mean circularity of all chromocenters. For each genotype, around 50 nuclei were analyzed. + +<|ref|>sub_title<|/ref|><|det|>[[149, 596, 260, 614]]<|/det|> +## Motif search + +<|ref|>text<|/ref|><|det|>[[149, 633, 851, 701]]<|/det|> +PHE1 binding motifs were previously identified in \(^{35}\) . The motif file was used for scanning promoter sequences (1kb upstream of the ATG) of Capsella genes with findMotifsGenome.pl function from HOMER \(^{60}\) . + +<|ref|>sub_title<|/ref|><|det|>[[149, 722, 290, 740]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[148, 759, 851, 853]]<|/det|> +The original data files for RNA seq and DNA bisulfite sequencing can be obtained from the NCBI Gene Expression Omnibus (GSE246468). The assembled \(Cg\) genome sequence has been deposited at DDBJ/ENA/GenBank under the accession JAVXYZ000000000. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[149, 83, 324, 100]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[147, 119, 852, 263]]<|/det|> +We thank Tainlin Duan and Martin Lascoux from Uppsala University for sharing tetraploid Co seeds. We are grateful to Heinrich Bente for useful suggestions about smallRNA library preparation and to Marco Incarbone and members of the Köhler group for critical reading and helpful comments on the manuscript. This research was supported by grants from the Knut and Alice Wallenberg Foundation (2018- 0206 and 2019- 0062 to CK), and the Max Planck Society. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[149, 83, 250, 100]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[144, 108, 850, 893]]<|/det|> +1. Coughlan, J.M. The role of hybrid seed inviability in angiosperm speciation. Am J Bot 110, 1-14 (2023).2. Sekine, D. et al. Dissection of two major components of the post-zygotic hybridization barrier in rice endosperm. Plant J 76, 792-9 (2013).3. Scott, R.J., Spielman, M., Bailey, J. & Dickinson, H.G. Parent-of-origin effects on seed development in Arabidopsis thaliana. Development 125, 3329-3341 (1998).4. Pennington, P.D., Costa, L.M., Gutierrez-Marcos, J.F., Greenland, A.J. & Dickinson, H.G. When genomes collide: aberrant seed development following maize interploidy crosses. Ann Bot 101, 833-43 (2008).5. Johnston, S.A. & Hanneman Jr., R.E. Manipulations of endosperm balance number overcome crossing barriers between diploid Solanum species. Science 217, 446-448 (1982).6. Rebernig, C.A., Lafon-Placette, C., Hatorangan, M.R., Slotte, T. & Kohler, C. Non-reciprocal interspecies hybridization barriers in the Capsella genus are established in the endosperm. PLoS Genet 11, e1005295 (2015).7. Roth, M., Florez-Rueda, A.M. & Stadler, T. Differences in effective ploidy drive genome-wide endosperm expression polarization and seed failure in wild tomato hybrids. Genetics 212, 141-152 (2019).8. Tonosaki, K. et al. Overcoming the species hybridization barrier by ploidy manipulation in the genus Oryza. Plant J 93, 534-544 (2018).9. Dziasek, K. et al. Hybrid seed incompatibility in Capsella is connected to chromatin condensation defects in the endosperm. PLoS Genet 17, e1009370 (2021).10. Qiu, Y. & Kohler, C. Endosperm evolution by duplicated and neofunctionalized type I MADS-box transcription factors. Mol Biol Evol 39, msab355 (2022).11. Martinez, G. et al. Paternal easiRNAs regulate parental genome dosage in Arabidopsis. Nat Genet 50, 193-198 (2018).12. Satyaki, P.R.V. & Gehring, M. Paternally acting canonical RNA-directed DNA methylation pathway genes sensitize Arabidopsis endosperm to paternal genome dosage. Plant Cell 31, 1563-1578 (2019).13. Cuerda-Gil, D. & Slotkin, R.K. Non-canonical RNA-directed DNA methylation. Nat Plants 2, 16163 (2016).14. Erdmann, R.M., Satyaki, P.R.V., Klosinska, M. & Gehring, M. A small RNA pathway mediates allelic dosage in endosperm. Cell Rep 21, 3364-3372 (2017).15. Rodrigues, J.A. et al. Imprinted expression of genes and small RNA is associated with localized hypomethylation of the maternal genome in rice endosperm. Proc Natl Acad Sci U S A 110, 7934-9 (2013).16. Grover, J.W. et al. Abundant expression of maternal siRNAs is a conserved feature of seed development. Proc Natl Acad Sci U S A 117, 15305-15315 (2020).17. Lu, J., Zhang, C., Baulcombe, D.C. & Chen, Z.J. Maternal siRNAs as regulators of parental genome imbalance and gene expression in endosperm of Arabidopsis seeds. Proc Natl Acad Sci U S A 109, 5529-34 (2012).18. Satyaki, P.R.V. & Gehring, M. RNA Pol IV induces antagonistic parent-of-origin effects on Arabidopsis endosperm. PLoS Biol 20, e3001602 (2022).19. Ishikawa, R. et al. Rice interspecies hybrids show precocious or delayed developmental transitions in the endosperm without change to the rate of syncytial nuclear division. Plant J 65, 798-806 (2010).20. Lafon-Placette, C. & Kohler, C. Endosperm-based postzygotic hybridization barriers: developmental mechanisms and evolutionary drivers. Mol Ecol 25, 2620-2629 (2016).21. Lafon-Placette, C. et al. Paternally expressed imprinted genes associate with hybridization barriers in Capsella. Nat Plants 4, 352-357 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 78, 850, 920]]<|/det|> +22. Brandvain, Y. & Haig, D. Divergent mating systems and parental conflict as a barrier to hybridization in flowering plants. Am Nat 166, 330-8 (2005).23. Douglas, G.M. et al. Hybrid origins and the earliest stages of diploidization in the highly successful recent polyploid Capsella bursa-pastoris. Proc Natl Acad Sci U S A 112, 2806-11 (2015).24. Slotte, T. et al. The Capsella rubella genome and the genomic consequences of rapid mating system evolution. Nat Genet 45, 831-835 (2013).25. Hatorangan, M.R., Laenen, B., Steige, K.A., Slotte, T. & Kohler, C. Rapid Evolution of Genomic Imprinting in Two Species of the Brassicaceae. Plant Cell 28, 1815-27 (2016).26. Grover, J.W. et al. Maternal components of RNA-directed DNA methylation are required for seed development in Brassica rapa. Plant J 94, 582 (2018).27. Axtell, M.J. ShortStack: comprehensive annotation and quantification of small RNA genes. RNA 19, 740-51 (2013).28. Mi, S. et al. Sorting of small RNAs into Arabidopsis organate complexes is directed by the 5' terminal nucleotide. Cell 133, 116-27 (2008).29. Burgess, D., Chow, H.T., Grover, J.W., Freeling, M. & Mosher, R.A. Ovule siRNAs methylate protein-coding genes in trans. The Plant Cell 34, 3647-3664 (2022).30. Kirkbride, R.C. et al. Maternal small RNAs mediate spatial-temporal regulation of gene expression, imprinting, and seed development in Arabidopsis. Proc Natl Acad Sci U S A 116, 2761-2766 (2019).31. Florez-Rueda, A.M., Fiscalini, F., Roth, M., Grossniklaus, U. & Stadler, T. Endosperm and seed transcriptomes reveal possible roles for small RNA pathways in wild tomato hybrid seed failure. Genome Biol Evol 13, evab107 (2021).32. Wolff, P., Jiang, H., Wang, G., Santos-Gonzalez, J. & Kohler, C. Paternally expressed imprinted genes establish postzygotic hybridization barriers in Arabidopsis thaliana. Elife 4(2015).33. Batista, R.A., Figueiredo, D.D., Santos-Gonzalez, J. & Kohler, C. Auxin regulates endosperm cellularization in Arabidopsis. Genes Dev 33, 466-476 (2019).34. Butel, N., Qiu, Y., Xu, W., Santos-Gonzalez, J. & Kohler, C. The molecular basis of parental conflict driven regulation of endosperm cellularization. Preprint at: https://www.biorxiv.org/content/10.1101/2023.06.22.546051v1 (2023).35. Batista, R.A. et al. The MADS-box transcription factor PHERES1 controls imprinting in the endosperm by binding to domesticated transposons. Elife 8, e50541 (2019).36. Yang, L. & Bennetzen, J.L. Structure-based discovery and description of plant and animal Helitrons. Proc Natl Acad Sci U S A 106, 12832-7 (2009).37. Hu, K. et al. Helitron distribution in Brassicaceae and whole genome helitron density as a character for distinguishing plant species. BMC Bioinformatics 20, 354 (2019).38. Wang, Z. et al. Polymerase IV plays a crucial role in pollen development in Capsella. Plant Cell 32, 950-966 (2020).39. Long, J. et al. Nurse cell--derived small RNAs define paternal epigenetic inheritance in Arabidopsis. Science 373(2021).40. Law, J.A. & Jacobsen, S.E. Establishing, maintaining and modifying DNA methylation patterns in plants and animals. Nat Rev Genet 11, 204-20 (2010).41. Dew-Budd, K.J. et al. Mating system is associated with seed phenotypes upon loss of RNA-directed DNA methylation in Brassicaceae. Plant Physiol (2023).42. Hehenberger, E., Kradolfer, D. & Köhler, C. Endosperm cellularization defines an important developmental transition for embryo development. Development 139, 2031-2039 (2012).43. Josefsson, C., Dilkes, B. & Comai, L. Parent-dependent loss of gene silencing during interspecies hybridization. Curr Biol 16, 1322-1328 (2006).44. Duan, T., Sicard, A., Glemin, S. & Lascoux, M. Separating phases of allopolyploid evolution with resynthesized and natural Capsella bursa-pastoris. Elife 12, RP88398 (2023).45. Doyle, J.J. & Doyle, J.L. Isolation of plant DNA from fresh tissue. Focus (Madison) 12, 13-15 (1990). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 81, 850, 565]]<|/det|> +46. Li, H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 34, 3094-3100 (2018).47. Li, H. Minimap and miniasm: fast mapping and de novo assembly for noisy long sequences. Bioinformatics 32, 2103-10 (2016).48. Magoc, T. & Salzberg, S.L. FLASH: fast length adjustment of short reads to improve genome assemblies. Bioinformatics 27, 2957-63 (2011).49. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754-60 (2009).50. Walker, B.J. et al. Pilon: an integrated tool for comprehensive microbial variant detection and genome assembly improvement. PLoS One 9, e112963 (2014).51. Guan, D. et al. Identifying and removing haplotypic duplication in primary genome assemblies. Bioinformatics 36, 2896-2898 (2020).52. Kim, D., Paggi, J.M., Park, C., Bennett, C. & Salzberg, S.L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol 37, 907-915 (2019).53. Liao, Y., Smyth, G.K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923-30 (2014).54. Love, M.I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014).55. Koenig, D. et al. Long-term balancing selection drives evolution of immunity genes in Capsella. Elife 8, e43606 (2019).56. Krueger, F. & Andrews, S.R. Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 27, 1571-2 (2011).57. Johnson, N.R., Yeoh, J.M., Coruh, C. & Axtell, M.J. Improved Placement of Multi-mapping Small RNAs. G3 (Bethesda) 6, 2103-11 (2016).58. Oksanen, J. et al. vegan: Community ecology package. R package. 2.6-2 edn (2022).59. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676-82 (2012).60. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38, 576-89 (2010). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 75, 728, 730]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 747, 850, 787]]<|/det|> +
Fig 1. Interploidy and interspecies hybrid seeds share similar phenotypes and deregulated genes.
+ +<|ref|>text<|/ref|><|det|>[[147, 805, 852, 916]]<|/det|> +A. Cleared hybrid Capsella seeds at 4 to 6 days after pollination (DAP). Corresponding mature seeds are shown on the right side of each panel. Scale bars correspond to \(100 \mu \mathrm{m}\) and \(1 \mathrm{mm}\) for cleared and mature seeds, respectively. Color code reflects effective ploidy as indicated in the figure. B. Mature seed phenotypes of crosses \(Cr \times Cbp\) and \(Cbp \times Cr\). C. Heatmap showing log2 fold changes (compared to the corresponding maternal parent) of genes upregulated in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 81, 852, 235]]<|/det|> +the analyzed hybrids. D. Upset plot showing the overlap of upregulated genes between interploidy \((Cr \times 4x Cr)\) and interspecies crosses of different Capsella species. E. Pictures of seeds from \(Co \times Cr\) and \(4xCo \times Cr\) . Scale bars represent 1mm. F. Corresponding germination data to crosses shown in panel E. Shown is the fraction of germinating seeds from each cross, with each filled circle representing one biological replicate. Numbers on top represent total seed numbers. Asterisks represent statistical significance as calculated by ANOVA with post hoc Tukey's HSD test (\\*\\* \(p < 0.001\) ). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 84, 810, 528]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[148, 544, 850, 583]]<|/det|> +
Fig. 2. Interspecies hybridization causes chromatin condensation defects in endosperm nuclei.
+ +<|ref|>text<|/ref|><|det|>[[147, 602, 852, 892]]<|/det|> +A. Percent of genes in pericentromeric and non-pericentromeric regions that are upregulated in different incompatible Capsella hybrids. Asterisks represent statistical significance as calculated by the chi-square test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). B. DAPI stained chromocenters from endosperm nuclei at 4 DAP of different Capsella genotypes. Scale bar, 5 μm. C. Quantification of chromatin condensation given as mean circularity of chromocenters (see Methods). Numbers below boxes indicate number of analyzed nuclei. Asterisks represent statistical significance as calculated by the Wilcoxon test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). D. Metagene plots show DNA methylation levels of TEs in the endosperm of \(Cg \times Cg\) , \(4xCr \times Cg\) , and \(4xCr \times 4xCr\) 6 DAP seeds. D. Boxplots showing methylation level of TEs in the endosperm of 6 DAP seeds of the indicated genotypes. Boxes show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (*** p-value \(< 0.001\) , ** p-value \(< 0.01\) , * p-value \(< 0.05\) ). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 75, 748, 565]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[146, 583, 750, 601]]<|/det|> +
Fig. 3. Maternal loss of NRPD1 mimics paternal-excess hybrid phenotypes.
+ +<|ref|>text<|/ref|><|det|>[[146, 618, 852, 907]]<|/det|> +A. DAPI stained chromocenters from endosperm nuclei from \(Cr \times Cr\) , \(Cr \times nrdp1\) , \(nrdp1 \times Cr\) and \(nrdp1 \times nrdp1\) . Scale bar, 5 μm. B. Quantification of chromatin condensation given as mean circularity of chromocenters (see Methods). Numbers below boxes indicate number of analyzed nuclei. Asterisks represent statistical significance as calculated by the Wilcoxon test (*** p-value < 0.001). C. Metagene plots show DNA methylation levels of TEs in the endosperm of \(Cr \times Cr\) , \(nrdp1 \times nrdp1\) , \(Cr \times nrdp1\) , and \(nrdp1 \times Cr\) 6 DAP seeds. D. Boxplots showing methylation level of TEs in the endosperm of 6 DAP seeds of the indicated genotypes. Boxes show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (*** p-value < 0.001, ** p-value < 0.01, * p-value < 0.05). E. Upset plot showing the overlap of upregulated genes among \(nrdp1 \times nrdp1\) , \(nrdp1 \times Cr\) and \(Cr \times Cg\) derived seeds. Genes were considered as upregulated based on log2(fold change) >1 and padj <0.05 in comparison to \(Cr \times Cr\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 78, 760, 715]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[148, 758, 646, 776]]<|/det|> +
Fig. 4. Maternal sirenRNAs are depleted in hybrid endosperm.
+ +<|ref|>text<|/ref|><|det|>[[148, 794, 852, 904]]<|/det|> +A. Cumulative expression plot of 24 nt-dominant siRNA loci in endosperm and leaves. Only loci with expression \(\geq 2\) RPM in combined replicates were analyzed (n = 13957 in endosperm and 13461 in leaves). In endosperm and leaves, 0.99% versus 27.6% of highest expressed 24 nt-dominant loci account for 90% of siRNAs, respectively. B. Mean siRNA accumulation at siren loci in the endosperm of indicated genotypes and Cr leaves. siRNA abundance at each + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 78, 852, 530]]<|/det|> +locus is normalized by locus length (in kb), and library size. Mean of two replicates is presented. Black vertical lines represent median. Individual measurements are shown as the rug below each density. Left panel shows data obtained by mapping reads to the \(Cr\) genome, right panel shows data obtained by mapping reads to the \(Cg\) genome. C. Venn diagram showing overlap among siren loci, loci with siRNAs depletion in \(Cr \times Cg\) hybrid and loci with siRNAs depletion in \(nrpd1 \times Cr\) (detected by DESeq2, log2(fold change) \(< 0\) , padj \(< 0.1\) ). D. siRNA accumulation and CHH methylation in the endosperm of indicated genotypes on selected genes overlapping siren loci (marked with violet lines): \(Carubv10025009m.g\) (AGL33), \(Carubv10011966m.g\) (YUC10), \(Carubv10028580m.g\) (AGL62), \(Carubv10010156m.g\) (AGL28), \(Carubv10002555m.g\) and \(Carubv10025621m.g\) (AGL33). Blue boxes represent genes, violet boxes represent TEs. E. Comparison of DNA methylation levels of genes and TEs overlapping siren loci with all genes and TEs in \(Cr\) genome. Boxes on this and following panels show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). F. Expression of genes overlapping siren loci in \(Cr \times Cr\) , \(Cr \times Cg\) and \(nrpd1 \times Cr\) seeds at 6 DAP. G. Heatmap showing log2 fold changes (compared to the corresponding maternal parent) of genes overlapping siren loci. H. Comparison of DNA methylation in \(Cr \times Cr\) , \(nrpd1 \times Cr\) and \(Cr \times Cg\) endosperm on upregulated genes overlapping siren loci. Asterisks indicate statistically significant differences as calculated by a one- sided Wilcoxon test (\\* p- value \(< 0.05\) ). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[147, 78, 692, 700]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[148, 717, 850, 757]]<|/det|> +
Fig. 5. Loss of DNA methylation at trans sirenRNA targets and increased dosage of paternal AGLs contributes to seed incompatibility.
+ +<|ref|>text<|/ref|><|det|>[[148, 776, 852, 907]]<|/det|> +A. CHG and CHH DNA methylation at sirenRNA trans targets (50 bp genomic windows, with \(\geq 1\) RPM realigning sirenRNAs and \(\leq 2\) mismatches) in \(Cr \times Cr\) endosperm. Barplot shows number of windows in each category, breaking points correspond to 0.7, 0.9, 0.98 and 0.995 quantiles. B. CHG and CHH DNA methylation at sirenRNA trans targets overlapping genes (coding region +2kb promoters (upper panel)) and transposable elements (TEs, lower panel). Barplot shows number of windows in each category, breaking points correspond to 0.5, 0.8 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 77, 852, 688]]<|/det|> +and 0.9 quantiles. C. Comparison of CHG and CHH DNA methylation levels in 50 bp windows differing in sirenRNA accumulation at trans targets. X- axis shows difference in DNA methylation \((Cr\times Cr - Cr\times Cg)\) and Y- axis difference in sirenRNA accumulation in trans \((Cr\) \(\times Cr - Cr\times Cg)\) . Barplot shows number of windows in each category, breaking points correspond to 0.6, 0.85 and 0.95 quantiles. D. Examples of genes with reduced sirenRNA accumulation at trans targets in \(Cr\times Cg\) , \(nrpd1\times Cr\) and \(nrpd1\times nrpd1\) compared to \(Cr\times Cr\) endosperm. Tracks show levels of sirenRNAs and CHH methylation at Carubv10021959m.g (F- box/RNI- like/FBD- like domain- containing protein), Carubv10007644m.g (MADS- box transcription factor), Carubv10020348m.g (F- box family protein) and Carubv10027768m.g. Blue boxes represent genes, violet boxes represent TEs. E. Plot shows distance to the nearest TE upstream of the transcriptional start site of upregulated genes in \(Cr\times Cg\) compared to non- upregulated genes. Boxes in this and following panels show medians and the interquartile range, and error bars show the full range excluding outliers. Asterisks indicate statistically significant differences as calculated by the Wilcoxon test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). F. Methylation level of TEs in promoters of upregulated genes in the \(Cr\times Cg\) hybrid. G. Distribution of PHE1 binding motifs in the Capsella genome in different groups of genes: all genes, genes upregulated in \(Cr\times Cr\) , genes upregulated in \(Cbp\times Cg\) , genes upregulated in \(Co\times Cr\) , genes upregulated in \(Cr\times 4xCr\) , genes upregulated in \(4xCr\times Cg\) . Values of pericentromeric/non- pericentromeric genes upregulated in different genotypes are compared to all pericentromeric/non- pericentromeric genes. NPC - non- pericentromeric region, PC - pericentromeric region. Asterisks indicate statistically significant differences calculated by Chi- square test (\\*\\*\\* p- value \(< 0.001\) , \\*\\* p- value \(< 0.01\) , \\* p- value \(< 0.05\) ). H, I. Heatmaps showing expression (RPKM) of type I \(\gamma\) - MADS box (H) and \(\alpha^{*}\) - MADS box (I) transcription factors in different Capsella genotypes. J. Heatmap showing \(\log 2\) (fold change) (in comparison to maternal parent) of \(\gamma\) and \(\alpha^{*}\) type I MADS box TFs in different Capsella hybrids. Genes marked in bold are overlapping siren loci, genes marked in purple are targeted by trans- acting sirenRNAs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 114]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 317, 256]]<|/det|> +SupplementaryFigures.pdf SupplementaryTable1.xlsx SupplementaryTable2.xlsx SupplementaryTable3.xlsx SupplementaryTable4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/images_list.json b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..28adae8f5a9f07a3dfc33cefc6ffc61480eab589 --- /dev/null +++ b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/images_list.json @@ -0,0 +1,138 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1: Overall sensitivity analysis of vaccine effectiveness, based on UK data.", + "footnote": [], + "bbox": [ + [ + 135, + 120, + 860, + 636 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: Optimal dose allocation.", + "footnote": [], + "bbox": [ + [ + 115, + 112, + 880, + 380 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: Dose allocation thresholds in different countries.", + "footnote": [], + "bbox": [ + [ + 120, + 155, + 730, + 533 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_1.jpg", + "caption": "Supplementary Fig. 1: Impact of vaccine delivery speed, dose split, and dose availability on vaccine effectiveness as a measure of reduction in clinical cases.", + "footnote": [], + "bbox": [ + [ + 120, + 152, + 870, + 510 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_3.jpg", + "caption": "Supplementary Fig. 3: Detailed sensitivity analysis of vaccine effectiveness for the most sensitive parameters, based on UK data.", + "footnote": [], + "bbox": [], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_5.jpg", + "caption": "Supplementary Fig. 5: Impact of vaccine booster dose interval on vaccine effectiveness as a measure of reduction in deaths.", + "footnote": [], + "bbox": [], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_6.jpg", + "caption": "Supplementary Fig. 6: Vaccine delivery speed.", + "footnote": [], + "bbox": [ + [ + 120, + 130, + 583, + 420 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Figures", + "footnote": [], + "bbox": [ + [ + 55, + 102, + 765, + 650 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 44, + 45, + 789, + 345 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 44, + 92, + 644, + 512 + ] + ], + "page_idx": 28 + } +] \ No newline at end of file diff --git a/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b.mmd b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b.mmd new file mode 100644 index 0000000000000000000000000000000000000000..16106de36cd6ca50497be836f9af28d0e2bc763a --- /dev/null +++ b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b.mmd @@ -0,0 +1,375 @@ + +# An analysis of the potential global impact of dosing regimen and rollout options for the ChAdOx1 nCoV-19 vaccine + +Ricardo Aguas University of Oxford + +Anouska Bharath University of Oxford + +Lisa White University of Oxford https://orcid.org/0000- 0002- 6523- 185X + +Bo Gao University of Oxford + +Merryn Voysey Oxford Vaccine Group, Department of Paediatrics, University of Oxford, United Kingdom + +Andrew Pollard University of Oxford https://orcid.org/0000- 0001- 7361- 719X + +Rima Shretta ( \(\boxed{\bullet}\) rima.shretta@ndm.ox.ac.uk) University of Oxford https://orcid.org/0000- 0001- 5011- 5998 + +## Article + +Keywords: COVID- 19, coronavirus, vaccine + +Posted Date: March 11th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 296726/v1 + +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on November 4th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26449- 8. + +<--- Page Split ---> + +# An analysis of the potential global impact of dosing regimen and rollout options for the ChAdOx1 nCoV-19 vaccine + +Ricardo Aguas1, Anouska Bharath1, Lisa J White1, Bo Gao1, Andrew J Pollard2, Merryn Voysey2, Rima Shretta\*1 + +1 Nuffield Department of Medicine, University of Oxford: R. Aguas PhD, A. Barath PhD, L.J. White PhD, B. Gao PhD, R. Shretta PhD2 Oxford Vaccine Group, Department of Paediatrics, University of Oxford: A. J. Pollard FMedSci, M. Voysey DPhil. + +# \*Corresponding author + +Rima Shretta Nuffield Department of Medicine University of Oxford Old Road Campus Headington, Oxford OX3 7LF United Kingdom rima.shretta@ndm.ox.ac.uk + +<--- Page Split ---> + +## Summary + +## Background + +The ongoing COVID- 19 pandemic has placed an unprecedented health and economic burden on countries at all levels of socioeconomic development, emphasizing the need to evaluate the most effective vaccination strategy in multiple, diverse environments. The high reported efficacy, low cost, and long shelf- life of the ChAdOx1 nCoV- 19 vaccine positions it well for evaluation in different settings. + +## Methods + +Using data from the ongoing ChAdOx1 nCoV- 19 clinical trials, an individual- based model was constructed to predict the 6- month population- level impact of vaccine deployment. A detailed probabilistic sensitivity analysis (PSA) was developed to evaluate the importance of epidemiological, demographic, immunological, and logistical factors in determining vaccine effectiveness. Using representative countries, logistical plans for vaccination rollout at various levels of vaccine availability and delivery speed, conditional on vaccine efficacy profiles (efficacy of the booster dose, time interval between doses, and relative efficacy of the first dose) were explored. + +## Findings and Interpretation + +Our results highlight how expedient vaccine delivery to high- risk groups is critical in mitigating COVID- 19 disease and mortality. In scenarios where the number of vaccine doses available is insufficient for high- risk groups (those aged more than 65 years) to receive two vaccine doses, administration of a single dose of vaccine is optimal. This effect is consistent even when vaccine efficacy after one dose is just \(75\%\) of the levels achieved after two doses. These findings offer a nuanced perspective of the critical drivers of COVID- 19 vaccination effectiveness and can inform optimal allocation strategies. These are relevant to high- income countries with a large high- risk group population as well as to low- income countries with younger populations, where the cost and logistical challenges of procuring and delivering two doses for each citizen represent a significant challenge. + +<--- Page Split ---> + +## Funding + +Bill and Melinda Gates Foundation (OPP1193472); Li Ka Shing Foundation; Oxford University COVID- 19 Research Response Fund (ref: 0009280). + +The funders played no role in the design or outcomes of this work. + +## Contributors + +RA, AB, LJW, and RS conceived the paper and contributed to the analysis. RA developed the model and wrote the code. BG ran the PSA and contributed to data processing. MV and AP provided the data, discussed the analysis, and commented on the draft. All authors have read, contributed to, and approved the final version of the manuscript. + +## Declaration of interests + +The authors declare no conflict of interest. + +## Data Sharing + +The data used to inform this analysis have been published and are available in the public domain. The code used is available at: https://github.com/ricardoaguas/como- ChAdOx1- vaccine-. + +## Acknowledgements + +LJW is funded by the Li Ka Shing Foundation. RA is funded by the Bill and Melinda Gates Foundation (OPP1193472). The Covid- 19 Modelling Consortium has support from the Oxford University COVID- 19 Research Response Fund (ref: 0009280). The authors have not been paid to write this article by a pharmaceutical company or other agency. The authors are grateful to Adam Bodley for proofreading the final draft. + +<--- Page Split ---> + +## Introduction + +As of March \(2^{\text{nd}}\) 2021, almost 115 million people have been diagnosed with COVID- 19 worldwide, and in excess of 2.5 million confirmed deaths have been reported \(^{2,3}\) . Vaccination is a critical strategy to control the spread of SARS- CoV- 2, the virus that causes COVID- 19, and to reduce the severity of symptomatic disease. Three vaccines have already received emergency use authorization in the United Kingdom (UK). The developers of two of these vaccines have reported efficacies of \(95\%\) for their vaccines in their respective Phase 3 trials (Pfizer/BioNTech and Moderna) \(^{4}\) . The third vaccine, ChAdOx1 nCoV- 19, jointly developed by Oxford University and AstraZeneca, demonstrated an acceptable safety profile and efficacy against symptomatic COVID- 19, with no hospital admissions or severe cases of disease reported in the intervention arm during Phase 3 trials conducted in three countries. This vaccine can be stored and distributed at \(2 - 8^{\circ}C\) and will be made available at a lower cost than the other vaccines, making it suitable for global access, particularly in low- and middle- income countries (LMICs) \(^{5 - 7}\) . + +While clinical trials have validated the efficacy of the ChAdOx1 nCoV- 19 vaccine in reducing symptomatic infection, appropriate national vaccination strategies across the world must consider heterogeneity among populations as well as the diverse demographic and socioeconomic environments of affected countries. In particular, the younger population typically present in LMICs justifies the need to assess the effects of associated behaviours and health profiles on vaccine effectiveness. These countries exhibit competing health, social, and economic challenges owing to inadequate healthcare infrastructure and a high prevalence of immunocompromising and infectious diseases. In these settings, individuals could also suffer complex vaccine responses when compared with responses in individuals in more developed economies \(^{8,9}\) . At the same time, many LMICs have been unable to secure vaccine doses in advance from potential suppliers and thus are likely to have incomplete coverage of their populations, particularly in the short- term. The global COVID- 19 vaccine alliance, COVAX, has pledged to procure and distribute vaccines equitably to LMICs; however, this will cover a maximum of \(20\%\) of the total population in each country \(^{10}\) . Although the University of Oxford and AstraZeneca have made the largest supply commitment to LMICs at more affordable prices than other vaccine manufacturers, there is a need to evaluate the impact of a range of factors on vaccine effectiveness \(^{11 - 13}\) . + +<--- Page Split ---> + +Due to shortages in supply, the UK government has instituted a policy of administering the booster dose of the vaccine at up to 12 weeks following the initial dose, prompting a debate among scientists, manufacturers, and governments on optimal dosing intervals for COVID- 19 vaccines14,15. The purpose of this analysis is therefore to evaluate the efficacy of the ChAdOx1 nCoV- 19 vaccine in countries with different demographic profiles, as a function of vaccine efficacy, dosage regime (interval between initial and booster doses, or no booster at all), coverage, and immunity wave rate. Given the differences in healthcare infrastructure and vaccine access around the world, decision- makers should consider the effect of these factors on population- level impact to determine the most effective strategy for their context13. + +Where vaccination programmes have begun, priority has so far been given to older age groups, individuals with co- morbidities, and frontline medical staff. The model developed therefore considers a simplified system where the vaccine is delivered to age groups in descending order while supplies are available. As there is limited evidence of indirect effects, that is, the potential for reductions in transmission, this vaccine effect was assumed to be negligible for the purposes of this analysis. + +## Methods + +The methodology employed was very specifically tailored to the research question and its context. Vaccine production rates are always going to be insufficient to meet the demand generated by a global pandemic. In a context of limited vaccine dose availability, it is imperative to prioritize those individuals who would yield the greatest epidemiological benefit. Assuming the most pressing need is to reduce hospitalization rates and deaths, the initial targeting of those at higher risk for these outcomes seems logical, given that the alternative of immunizing sufficient people at lower risk for the indirect benefits to outweigh the direct benefits of a vaccine targeted at those at higher risk is not feasible with the number of vaccines available in the short- term. Even the UK, where mass production of the AztraZeneca vaccine has enabled \(20\%\) of the population to be vaccinated within 3 months, opted to prioritize the high- risk groups (those aged more than 65 years), partially because of the uncertainty around vaccine efficacy against infection. The ChAdOx1 nCoV- 19 vaccine + +<--- Page Split ---> + +clinical trials were the only Phase 3 trials in which infection was evaluated as an outcome. No evidence was found for a transmission reduction effect (VE = 3.8% [- 72.4 to 46.3])7, but important questions were raised about how to allocate a limited number of doses to optimize the impact on symptomatic disease, given that a single- dose regimen could offer prolonged protection and thus a delay of the second dose could be warranted. + +We began from the premise laid out above and implemented an individual- based, age- dependent, static transmission model to predict the number of infections, clinical cases, and deaths expected to occur within 6 months of vaccination programme rollout. Individuals are simulated as autonomous systems, each with a set of attributes, informing their serostatus, vaccination uptake history (number of doses and dosing interval), and age. Box 1 details how the dynamics processes inherent to disease transmission and vaccination campaign logistics are considered in the model. + +<--- Page Split ---> + +Box 1. Consider a hypothetical scenario where the number of SARS- CoV- 2 infections over 6 months in an unvaccinated population is 100,000. Policymakers could opt for one of two alternative options with very different direct benefit outlooks when deciding on how to allocate the limited number of vaccine doses available to them. + +Option 1: Vaccinate high contact groups (aged 25- 40 years) (36% of all infections, 10% of all hospitalizations, 1% of all deaths) \(^{1}\) + +Predicted infections: + +100,000\*36%\*[1- vaccine direct effect on infection] + 100,000\*64%\*[1- vaccine indirect effect on infection] + +Predicted hospitalizations: + +Predicted infections\*10%\*[1- vaccine direct effect on hospitalization] + Predicted infections \*90%\*[1- vaccine Indirect effect on hospitalizations] + +Predicted deaths: + +Predicted infections\*1%\*[1- vaccine direct effect on deaths] + Predicted infections\*99%\*[1- vaccine indirect effect on deaths] + +Option 2: Vaccinate high risk groups (aged >65 years) (10% of all infections, 66% of all hospitalizations, 90% of all deaths) \(^{1}\) + +Predicted infections: + +100,000\*10%\*[1- vaccine direct effect on infection] + 100,000\*90%\*[1- vaccine indirect effect on infection] + +Predicted hospitalizations: + +Predicted infections \*66%\*[1- vaccine direct effect on hospitalization] + Predicted infections \*34%\*[1- vaccine indirect effect on hospitalizations] + +Predicted deaths: + +Predicted infections\*90%\*[1- vaccine direct effect on deaths] + Predicted infections\*10%\*[1- vaccine indirect effect on deaths] + +Thus, vaccines targeting high- contact groups would have to provide indirect effects in the order of 80% (80% reduction in risk in the untargeted population) to prevent an approximate number of deaths similar to that which would be provided by targeting the high- risk group with a direct vaccine effect against death of 85%. + +<--- Page Split ---> + +## Transmission and clinical cascade + +The spread of COVID- 19 is sensitive to the underlying network of contacts between infectious and susceptible individuals in their various societal spheres (home, work, public transport, etc). For a given population, we can summarize the number of contacts per day as an age- dependent force of infection \(\lambda (a)\) , i.e. a daily risk of acquiring an infection given age \(a\) . The age- dependent risk of infection can then be defined as: + +\[\lambda (a) = k_{\lambda}\frac{\sum_{j = 1}^{N}c_{ij}P_{j}}{\sum_{i = 1}^{N}(\sum_{j = 1}^{N}c_{ij}P_{j})}\] + +Where \(c_{ij}\) is the daily number of contacts between age groups \(i\) and \(j\) for a particular country, \(P_{j}\) is the population age distribution, \(N\) is the total number of age categories, and \(k_{\lambda}\) is the overall daily risk of infection (which is informed by the number of infectious people in the population). Here, we decided to simplify the transmission process by making the daily risk of infection constant over time, thus having a static transmission model. When evaluating different epidemiological scenarios with different levels of population attack size over the 6- month period explored here, we simply changed the daily risk of infection parameter until the correct attack rate was obtained. Without this constraint, computation of the very high number of simulations required to perform the sensitivity analyses presented here would be virtually impossible. + +The risk of developing severe disease and possibly dying as a consequence of infection was informed by age- dependent infection hospitalization (IHR) and hospitalization fatality (HFR) ratios, published in16. Thus, the modelled daily risk that an individual will develop severe disease is given by \(\lambda (a)IHR(a)\) , whereas the risk of dying is approximately \(\lambda (a)IHR(a)HFR(a)\) . The timing of these events and the lag between infection and clinical outcome are not relevant, as we are only making comparisons between synthetic populations, as detailed below. + +## Vaccination delivery and vaccine efficacy + +Different vaccine dose allocation schemes were simulated, by limiting the number of doses distributed in 6 months, as well as allowing for different dosing intervals (delaying the second + +<--- Page Split ---> + +dose) and dosing splits (giving one dose vs two doses). The allocation of doses was always prioritized to the oldest age groups. Individuals were assigned vaccine doses in descending order of age until the maximum number of doses had been allocated. + +Given a fixed number of available doses, one can calculate the target recipient population by looking at the dose- split proposal. If all doses are given as single doses in 6 months, then the target population for vaccination is equal to the number of available doses. At the other extreme, where all vaccines receive two doses, the number of recipients would be half the number of available doses. Within the group that is meant to receive two doses of the vaccine, a \(5\%\) dropout rate (vaccine refusal) was imposed, and a range of booster dose intervals was explored. + +We implemented three different logistical implementations of a vaccine campaign rollout: constant effort, frontloaded, and backloaded. The distinction was in the speed at which the target population received vaccine doses during the initial 2 months. As individuals were assigned a vaccine, the number of doses received would be determined by a draw from a uniform distribution according to the desired dose split. Individuals given two doses would be assigned a booster dose interval following a beta distribution with \(\alpha = 0.15\) and \(\beta = 0.95\) . + +Although vaccine efficacy was explored in the sensitivity analyses presented here, we centred the explored ranges around the point estimates presented in \(^{17}\) . Vaccine efficacy, \(V_{e}(t)\) , was treated as a direct modulator of the risk of infection, clinical disease, and death; it was then defined for each individual, at each timestep of the simulation, as: + +\[V_{e}(t) = V_{i}^{j}e^{-\delta t}\] + +, where V is the vaccine efficacy in an individual with baseline status \(i\) that received dose number \(j\) a \(t\) number of days ago, while \(\delta\) is the rate of loss of vaccine- induced immunity. + +Throughout this paper, we present a sensitivity analysis of the post- dose two maximum efficacy, the relative efficacy of dose one vs dose two, and the booster dose interval. While doing so, we constrain vaccine efficacy against clinical disease to be the same as that against death, while vaccine efficacy against infection is fixed at \(5\%\) . We also imposed a stepwise increase in post- dose two vaccine efficacy across an 8- week booster dose interval, as observed in the clinical trial \(^{17}\) . This means that giving the second vaccine dose less than 8 + +<--- Page Split ---> + +weeks after the first dose will result in a \(25\%\) lower post- dose two efficacy relative to the maximum assumed vaccine efficacy. + +## Vaccine effectiveness + +The vaccination campaign population impact is referred to throughout as vaccine effectiveness and was defined as: + +\[V_{eff} = 100\frac{AR_v - AR_u}{AR_u}\] + +, where \(AR_{v}\) is the attack rate (over 6 months) in the vaccinated population, and \(AR_{u}\) is the attack rate in a population that mirrors the vaccinated population in all aspects except vaccination. We thus have a pair of populations for each parameter set in our analyses and calculate the expected vaccine effectiveness for each parameter set as the relative difference in occurrence of each of the disease endpoints (infection, clinical symptoms, and death). + +## Results + +We conducted an extensive initial sensitivity analysis to determine how the impact of rolling out a COVID- 19 vaccination campaign in the UK depends on epidemiological, logistical, and immunological factors. The sensitivity of the modelled vaccine effectiveness to the variables explored is illustrated in Figure 1. It is clear that the prospects for vaccine impact are most sensitive to the number of vaccine doses available within 6 months, the speed of delivery within the same timeframe, and the vaccine efficacy (both the maximum efficacy post- dose two and the relative efficacy of dose one compared with dose two). Interestingly, for the same inputs, the median expected vaccine effectiveness is greater for deaths than it is for clinical cases. + +From this first exploratory analysis one could immediately suggest that, for maximum effectiveness, a vaccine campaign should aim to vaccinate as many people as possible (thus governments/policymakers should procure the maximum number of doses possible), in the shortest time possible. These are by far the two variables the model outputs are most sensitive to, as can clearly be seen in Supplementary Figures 1 and 2. These figures also reveal + +<--- Page Split ---> + +a very interesting interaction between the vaccine efficacy profile and an actionable decision of how the first available doses are to be distributed. In the frontloaded scenario, where the speed of vaccine delivery is maximal during the early stages of the 6- month vaccination campaign, a single- dose regimen is expected to perform significantly worse if vaccine efficacy post- dose one is \(50\%\) lower than vaccine efficacy post- dose two (top row). However, if the vaccine efficacy after both doses is the same (bottom row), a single- dose regimen can actually be preferable, especially if the number of vaccine doses available is small. + +These initial results prompted further investigation of the possible interactions and trade- offs between the vaccine efficacy profiles and logistical implementation variables. In this detailed analysis, the population attack size was fixed at \(12\%\) , delivery speed to frontloaded, and vaccine- induced protection to last 360 days. The results are summarized in Supplementary Figure 3. As determined by the initial sensitivity analysis, vaccine effectiveness is quite sensitive to the number of available doses, the maximum post- dose two efficacy, and the efficacy of the first dose relative to the second. + +Two interesting results pertain to the sensitivity of the model to changes in the dose number split and the interval in days between doses for the two- dose regimen recipients. While giving everyone two doses, irrespective of all other variables, seems to be preferable to the single- dose option, the length of the whiskers suggests there might be a parameter space for which the single- dose option is optimal, as seen in Supplementary Figures 1 and 2. Increasing the time interval between doses generally produces improved vaccine effectiveness, although a slight decrease in median effectiveness can be observed after an 8- week (56 day) interval. This is further investigated in Supplementary Figures 4 and 5, revealing an interesting trade- off, with a large increase in predicted effectiveness after 7 to 8 weeks, followed by a small decrease as the booster dose interval expands, but only if the efficacy of the first dose is low. If the efficacy of the first dose is similar to the efficacy of the second, increasing the interval between doses up to 12 weeks does not decrease vaccine effectiveness. + +Interestingly, we find a non- linear increase in effectiveness for large values of dose availability, which can be explained by the markedly non- linear risk of severe disease and death with age. As the number of available doses increases, a larger proportion of the population will receive a vaccine dose. However, since vaccines are allocated in descending order of age, as a larger proportion of the population is reached, more and more low- risk + +<--- Page Split ---> + +individuals are vaccinated, for whom the vaccine accrued benefits are smaller and smaller. It is then advisable to investigate these relationships for different settings. + +We proceeded to investigate what factors could potentially influence the decision- making process regarding the distribution of doses during the first 6 months of vaccine programme rollout. We thus evaluated the relative predicted effectiveness of the single- dose versus the double- dose regimen, for different countries with potentially different dose availability, and assuming different vaccine efficacy profiles (Figure 2). For very high levels of dose allocation (high y- axis values), a two- dose regimen is clearly optimal. This starts to become less evident for scenarios where the protection conferred by the first dose gets closer to the post- dose two efficacy (moving right along the x- axis). Interestingly, when the number of available doses is small, the single- dose regimen will become more effective than the double- dose regimen, as the thick black line is crossed. The parameter combinations defining the line where there is a shift in strategy positioning are country dependent, with countries with older populations having a larger parameter space in which a single- dose option is preferable, as shown in Figure 3. + +## Discussion + +The SARS- CoV- 2/COVID- 19 pandemic has created an unprecedented public health challenge, spurring a global race to develop and distribute viable vaccines. A vaccine that creates broad immunity against the SARS- CoV- 2 virus could be the only effective means to control the pandemic and allow a return to "normalcy". To have a significant impact on the disease, a critical mass of the global population at risk will need to be vaccinated. However, many high- income countries have secured more than half of the available vaccine doses for themselves, leaving LMICs, which make up more than 85% of the global population, to find their own solutions18. To address the problem of equitable access, WHO, Gavi, and the Coalition for Epidemic Preparedness Innovations (CEPI) established COVAX, a global alliance that has pledged to pool investment and allocate and distribute COVID- 19 vaccines equitably, particularly in LMICs10. However, COVAX is currently under- resourced and the doses secured are insufficient to achieve the coverage levels needed19. Supply constraints and new variants of SARS- CoV- 2 are steering countries towards strategies that counter low + +<--- Page Split ---> + +access with dosing patterns or volumes to maximize the impact of the vaccines. Data from the ChAdOx1 nCoV- 19 vaccine trials have allowed us to explore potential strategies to inform optimal allocation programmes, particularly in contexts where the cost and logistics of implementing multiple doses within a short timeframe may be challenging. + +Our findings indicate that vaccine effectiveness is dependent upon (i) the country context, which includes the demographic profile, the attack rate of the virus, and the amount of vaccine that is available (which influences the proportion of the population that is vaccinated); ii) the characteristics of the vaccine, which include the efficacy of a single dose relative to a double dose and the waning of efficacy over time; and iii) the proportion of the population receiving the second dose, the time interval between doses, and the delivery speed. + +Our analysis demonstrates that in scenarios where the number of vaccine doses available is insufficient for the highest risk groups (aged \(>65\) years) to receive two doses, the allocation of a single vaccine dose is optimal. This effect is consistent even when the vaccine efficacy of a single dose is just \(75\%\) of the levels achieved after a double dose, until allocation drops to a population coverage of \(10\%\) , after which vaccinating only the high- risk individuals, with two doses, is more effective. In scenarios where the number of doses available to the country is sufficiently high, or if the relative single- dose efficacy is low ( \(50\%\) or less), providing a booster dose within 8 weeks would be preferable. Apart from these specific conditions, the results indicate that providing individuals with two doses of vaccine would have a similar effectiveness to the use of a single dose given to twice the number of individuals. + +The speed at which the high- risk population is vaccinated greatly influences the expected vaccine effectiveness in preventing clinical cases and death. This is particularly true if the transmission rate is high, with faster vaccination reducing the number of infections in groups awaiting their first dose during the rollout. Distributing the vaccine very slowly provides an effectiveness of less than \(10\%\) , regardless of the number of doses and allocations. The impact of allocation on outcomes is also greater when the population is vaccinated rapidly over a six- month period. In both of these scenarios, providing a single dose is preferable. + +<--- Page Split ---> + +An interesting trade- off was found between the booster dose interval and the relative vaccine efficacy of a single dose. For vaccines with large differences between first and second dose efficacy, delaying the booster dose interval past 8 weeks after the first dose was found to be detrimental. However, if a single dose provided at least \(75\%\) of the protection conferred by a double dose, delaying the booster dose interval to 12 weeks had a negligible impact on the number of cases and deaths. Given the similar reported efficacies of single and double doses of ChAdOx1 nCoV- 19, a 12- week interval is the optimal scenario for this vaccine17. However, this finding may not be applicable to other COVID- 19 vaccines. + +These differences are more profound when considering the demographic characteristics of a population. In high- income countries, which have a larger older population (>65 years), a single- dose regimen will allow the vaccination of more individuals more quickly, with a correspondingly greater impact on cases and deaths. In the UK, the six- month allocation threshold above which a two- dose regimen would be preferred was found to be about \(16.5\%\) . + +The six- month allocation threshold above which a two- dose regimen would be preferred is much lower in LMICs, mainly due to mortality in the younger population. In these contexts, decision- makers will need to consider the affordability, availability, and logistical constraints and feasibility of implementing a single or a double dose, the dosage intervals, and delivery speed. Most LMICs lack the digital databases necessary to manage patient data, reliably track vaccine inventories, keep track of who has received which vaccine, and inform people where and when they are due for a booster. Governments would also need to ensure that they reserve sufficient stocks to allow the administration of booster doses. In these cases, a robust cost- benefit analysis of each option will need to be considered. + +The dosing interval for COVID- 19 vaccines has been a subject of debate among scientists, regulators, and governments around the world following the UK government's decision to prioritize administering the first dose of vaccine to as many at- risk people as possible and increasing the interval between the two doses to up to 12 weeks14,15,20. A one- dose vaccine regimen or a two- dose regimen with longer time intervals may be sufficient to reduce symptoms of COVID- 19 in the most vulnerable individuals and ultimately slow the pandemic, given that the time difference between first and second doses was shown to have a negligible effect on overall vaccine effectiveness (clinical cases, infections, and + +<--- Page Split ---> + +deaths). Indeed, a recent WHO notification stated that some countries are facing "exceptional circumstances" and may want to delay second doses to "allow for a higher initial coverage". Other exceptional circumstances may involve trade-offs around the relative size of the highest risk population in a country and the currently unknown potential for a vaccine to reduce transmission, which may lead to some countries targeting high- contact groups to benefit from any potential indirect effects. + +Nevertheless, these thresholds are likely to differ depending on the country context. For example, smaller countries may be able to rapidly rollout the vaccine to a higher percentage of their population compared with the speed at which larger countries can do this. It should be noted that an implementation strategy is determined at a country level. The assumptions made in this work are based on the association of certain parameters with the health infrastructure and existing population of certain country groups defined by income level. + +Published clinical data were used to inform the parameters used in the model described in this paper. These data provide an aggregate efficacy of the ChAdOx1 nCoV- 19 vaccine among people of a wide range of ages living in different countries. However, there were limited data available for assessing the effects of certain parameters (such as the effect of the dosing interval on post- dose two efficacy) on vaccine efficacy, which begat the need to conduct the post hoc exploratory analysis presented here. + +## Conclusion + +This analysis demonstrates that in scenarios where the number of vaccine doses available is insufficient for the highest risk groups (>65 years of age) to receive two vaccine doses, allocation of a single vaccine dose to twice the number of individuals or extending the time interval between doses may be more optimal strategies. In contexts without supply constraints, or if the single- dose efficacy is low, providing a booster dose would be preferable. Apart from these specific conditions, the results indicate that providing individuals with two doses of vaccine would have a similar effectiveness to the use of a single dose in twice the number of individuals. In an ideal world, decisions about vaccination strategies would be made within the exact parameters of the trials that have been + +<--- Page Split ---> + +conducted. However, the limited availability of resources, and specific country contexts, may require decision- makers to consider alternative strategies. + +<--- Page Split ---> + +# Tables and Figures + +Table 1: Model parameters. + +
ParameterModel termRangeDescription
Population attack size (% of the population)ATT(4; 12; 20)This is the percentage of the population
infected within the 6-month study period
Vaccine allocation (% of the
population during study
period)		TRG(5; 10; 20; 30)Allocation range was based on the assumed administration speed. Using current data,
we assumed that higher-income countries
could reach a maximum speed, allowing
30% coverage of the population within 6
months
Second dose administered
(% of the vaccinated
population administered a
second dose)		DSP(0; 25; 50; 75;
100)		This is the percentage of the vaccinated
population that are administered a second (booster) dose
Interval between first dose and booster doseBTI(4 weeks; 7
weeks; 12 weeks)		The interval between doses can affect
vaccine efficacy; the range chosen was
based on available clinical trial data
Vaccine delivery speedDEL(fixed;
frontloaded;
backloaded)		The speed of vaccine delivery to the
population - see Supplementary figure 6
Vaccine efficacy after the
second dose		PD2(65; 75; 85)Maximum efficacy following the second
dose
Vaccine efficacy of the first
dose compared with the
second dose (%)		D2B(50; 75; 100)Effect of the first dose compared with the
second dose
Immunity wane rate (days
following last dose)		VCW(90; 180; 360;
540)		Vaccine protection decay post-last dose
+ +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1: Overall sensitivity analysis of vaccine effectiveness, based on UK data.
+ +The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. The full list of parameters explored and their descriptions can be found in Table 1. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: Optimal dose allocation.
+ +
Efficacy of Single dose vs Double dose regimen
+ +The coloured surfaces and respective contour lines indicate the ratio between the predicted vaccine effectiveness for a double- dose regimen vs a single- dose regimen. This ratio is a mean ratio, obtained by averaging out the ratios obtained in all model runs assuming the corresponding x and y parameter values and thus are not expected to be regular. Contour line 1 (thicker black line) indicates the parameter combinations for which there is no expected difference between giving everyone a single dose vs giving everyone two doses. For values greater than 1 (hot colours), a two- dose regimen is preferable, and for values less than 1 (cold colours), a single- dose regimen is preferable. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: Dose allocation thresholds in different countries.
+ +The figure illustrates the parameter combinations that define the allocation threshold above which a two- dose regimen would be preferred over a single- dose regimen. The areas under the curves are \(16.5\%\) , \(8\%\) , and \(3.8\%\) for the UK, Brazil, and Uganda, respectively, which correlates almost perfectly with the proportion of the population above the age of 65 years in those countries. + +<--- Page Split ---> +![](images/Supplementary_Figure_1.jpg) + +
Supplementary Fig. 1: Impact of vaccine delivery speed, dose split, and dose availability on vaccine effectiveness as a measure of reduction in clinical cases.
+ +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> + +Supplementary Fig. 2: Impact of vaccine delivery speed, dose split, and dose availability on vaccine effectiveness as a measure of reduction in deaths. + +![](images/Supplementary_Figure_3.jpg) + + +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +![](images/Supplementary_Figure_5.jpg) + +
Supplementary Fig. 3: Detailed sensitivity analysis of vaccine effectiveness for the most sensitive parameters, based on UK data.
+ +The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. + +<--- Page Split ---> + +Supplementary Fig. 4: Impact of vaccine booster dose interval on vaccine effectiveness as a measure of reduction in clinical cases. + +![](images/Supplementary_Figure_6.jpg) + + +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Supplementary Fig. 5: Impact of vaccine booster dose interval on vaccine effectiveness as a measure of reduction in deaths.
+ +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Supplementary Fig. 6: Vaccine delivery speed.
+ +The figure shows the number of vaccine doses administered over the course of the vaccination campaign (6 months). + +<--- Page Split ---> + +## References + +1. Office for National Statistics. Coronavirus (COVID-19) weekly insights: latest health indicators in England, 12 February 2021. February 12, 2021 ed. United Kingdom; 2021. +2. Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time. Lancet Infect Dis 2020; 20(5): 533-4. +3. Johns Hopkins University. COVID-19 Dashboard. 2020. https://coronavirus.jhu.edu/map.html (accessed June 2020. +4. Pfizer. Pfizer and BioNtech announce vaccine candidate against Covid-19 achieved success in first interim analysis from phase 3 study. 2020. +5. Folegatti PM, Ewer KJ, Aley PK, et al. Safety and immunogenicity of the ChAdOx1 nCoV-19 vaccine against SARS-CoV-2: a preliminary report of a phase 1/2, single-blind, randomised controlled trial. Lancet 2020; 396(10249): 467-78. +6. Ramasamy MN, Minassian AM, Ewer KJ, et al. Safety and immunogenicity of ChAdOx1 nCoV-19 vaccine administered in a prime-boost regimen in young and old adults (COV002): a single-blind, randomised, controlled, phase 2/3 trial. Lancet 2021; 396(10267): 1979-93. +7. Voysey M, Clemens SAC, Madhi SA, et al. Safety and efficacy of the ChAdOx1 nCoV-19 vaccine (AZD1222) against SARS-CoV-2: an interim analysis of four randomised controlled trials in Brazil, South Africa, and the UK. Lancet 2021; 397(10269): 99-111. +8. MacLennan CA. Vaccines for low-income countries. Seminars in Immunology 2013; 25(2): 114-23. +9. Rodrigues CMC, Plotkin SA. Impact of Vaccines; Health, Economic and Social Perspectives. Frontiers in Microbiology 2020; 11(1526). +10. WHO. Fair allocation mechanism for COVID-19 vaccines through the COVAX Facility. Geneva: WHO, 2020. +11. Davies NG, Klepac P, Liu Y, Prem K, Jit M, Eggo RM. Age-dependent effects in the transmission and control of COVID-19 epidemics. Nat Med 2020; 26(8): 1205-11. +12. Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. The Lancet Digital Health 2020; 2(12): e638-e49. +13. Wang W, Wu Q, Yang J, et al. Global, regional, and national estimates of target population sizes for covid-19 vaccination: descriptive study. BMJ 2020; 371: m4704. +14. Iacobucci G, Mahase E. Covid-19 vaccination: What's the evidence for extending the dosing interval? BMJ 2021; 372: n18. +15. Mahase E. Covid-19: Medical community split over vaccine interval policy as WHO recommends six weeks. BMJ 2021; 372: n226. +16. Brazeau N, Verity R, Jenks S, et al. COVID-19 Infection Fatality Ratio Estimates from Seroprevalence. London: Imperial College London, 2020. +17. Voysey M, Clemens SAC, Madhi SA, et al. Single dose administration, and the influence of the timing of the booster dose on immunogenicity and efficacy of ChAdOx1 nCoV-19 (AZD1222) vaccine Lancet 2021. +18. Twohey M, Collins K, Thomas K. With First Dibs on Vaccines, Rich Countries Have 'Cleared the Shelves'. New York Times. 2020 December 15, 2020. +19. UN News. COVID-19 vaccines: donors urged to step up funding for needy countries. New York: UN News; 2020. +20. Davis N. Vaccine experts call for clarity on UK's 12-week Covid jab interval. The Guardian. 2021 January 25, 2021. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figures
+ +Overall sensitivity analysis of vaccine effectiveness, based on UK data. The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. The full list of parameters explored and their descriptions can be found in Table 1. + +<--- Page Split ---> +![PLACEHOLDER_29_0] + +
Figure 2
+ +Optimal dose allocation. The coloured surfaces and respective contour lines indicate the ratio between the predicted vaccine effectiveness for a double- dose regimen vs a single- dose regimen. This ratio is a mean ratio, obtained by averaging out the ratios obtained in all model runs assuming the corresponding \(x\) and \(y\) parameter values and thus are not expected to be regular. Contour line 1 (thicker black line) indicates the parameter combinations for which there is no expected difference between giving everyone a single dose vs giving everyone two doses. For values greater than 1 (hot colours), a two- dose regimen is preferable, and for values less than 1 (cold colours), a single- dose regimen is preferable. + +<--- Page Split ---> +![PLACEHOLDER_30_0] + +
Figure 3
+ +Dose allocation thresholds in different countries. The figure illustrates the parameter combinations that define the allocation threshold above which a two- dose regimen would be preferred over a single- dose regimen. The areas under the curves are \(16.5\%\) , \(8\%\) , and \(3.8\%\) for the UK, Brazil, and Uganda, respectively, which correlates almost perfectly with the proportion of the population above the age of 65 years in those countries. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementary Figures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b_det.mmd b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..dec744b481d892d7f5eaace13c52d4434b1cda30 --- /dev/null +++ b/preprint/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b/preprint__0dbf2481fcf07fc6fd0e01821e9109b6f6dfa78793e31b5946d85a21c8c5a94b_det.mmd @@ -0,0 +1,481 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 940, 208]]<|/det|> +# An analysis of the potential global impact of dosing regimen and rollout options for the ChAdOx1 nCoV-19 vaccine + +<|ref|>text<|/ref|><|det|>[[44, 230, 230, 271]]<|/det|> +Ricardo Aguas University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 277, 230, 317]]<|/det|> +Anouska Bharath University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 323, 590, 365]]<|/det|> +Lisa White University of Oxford https://orcid.org/0000- 0002- 6523- 185X + +<|ref|>text<|/ref|><|det|>[[44, 370, 230, 410]]<|/det|> +Bo Gao University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 416, 814, 458]]<|/det|> +Merryn Voysey Oxford Vaccine Group, Department of Paediatrics, University of Oxford, United Kingdom + +<|ref|>text<|/ref|><|det|>[[44, 463, 590, 504]]<|/det|> +Andrew Pollard University of Oxford https://orcid.org/0000- 0001- 7361- 719X + +<|ref|>text<|/ref|><|det|>[[44, 508, 590, 550]]<|/det|> +Rima Shretta ( \(\boxed{\bullet}\) rima.shretta@ndm.ox.ac.uk) University of Oxford https://orcid.org/0000- 0001- 5011- 5998 + +<|ref|>sub_title<|/ref|><|det|>[[44, 591, 102, 608]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 628, 411, 647]]<|/det|> +Keywords: COVID- 19, coronavirus, vaccine + +<|ref|>text<|/ref|><|det|>[[44, 667, 312, 686]]<|/det|> +Posted Date: March 11th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 704, 463, 724]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 296726/v1 + +<|ref|>text<|/ref|><|det|>[[44, 741, 910, 784]]<|/det|> +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 820, 946, 863]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 4th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26449- 8. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[117, 84, 850, 128]]<|/det|> +# An analysis of the potential global impact of dosing regimen and rollout options for the ChAdOx1 nCoV-19 vaccine + +<|ref|>text<|/ref|><|det|>[[117, 162, 815, 207]]<|/det|> +Ricardo Aguas1, Anouska Bharath1, Lisa J White1, Bo Gao1, Andrew J Pollard2, Merryn Voysey2, Rima Shretta\*1 + +<|ref|>text<|/ref|><|det|>[[117, 241, 856, 338]]<|/det|> +1 Nuffield Department of Medicine, University of Oxford: R. Aguas PhD, A. Barath PhD, L.J. White PhD, B. Gao PhD, R. Shretta PhD2 Oxford Vaccine Group, Department of Paediatrics, University of Oxford: A. J. Pollard FMedSci, M. Voysey DPhil. + +<|ref|>title<|/ref|><|det|>[[120, 372, 314, 389]]<|/det|> +# \*Corresponding author + +<|ref|>text<|/ref|><|det|>[[117, 399, 395, 574]]<|/det|> +Rima Shretta Nuffield Department of Medicine University of Oxford Old Road Campus Headington, Oxford OX3 7LF United Kingdom rima.shretta@ndm.ox.ac.uk + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 86, 201, 101]]<|/det|> +## Summary + +<|ref|>sub_title<|/ref|><|det|>[[118, 121, 221, 137]]<|/det|> +## Background + +<|ref|>text<|/ref|><|det|>[[117, 155, 881, 279]]<|/det|> +The ongoing COVID- 19 pandemic has placed an unprecedented health and economic burden on countries at all levels of socioeconomic development, emphasizing the need to evaluate the most effective vaccination strategy in multiple, diverse environments. The high reported efficacy, low cost, and long shelf- life of the ChAdOx1 nCoV- 19 vaccine positions it well for evaluation in different settings. + +<|ref|>sub_title<|/ref|><|det|>[[118, 297, 197, 313]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[117, 330, 880, 532]]<|/det|> +Using data from the ongoing ChAdOx1 nCoV- 19 clinical trials, an individual- based model was constructed to predict the 6- month population- level impact of vaccine deployment. A detailed probabilistic sensitivity analysis (PSA) was developed to evaluate the importance of epidemiological, demographic, immunological, and logistical factors in determining vaccine effectiveness. Using representative countries, logistical plans for vaccination rollout at various levels of vaccine availability and delivery speed, conditional on vaccine efficacy profiles (efficacy of the booster dose, time interval between doses, and relative efficacy of the first dose) were explored. + +<|ref|>sub_title<|/ref|><|det|>[[119, 550, 350, 567]]<|/det|> +## Findings and Interpretation + +<|ref|>text<|/ref|><|det|>[[117, 584, 880, 839]]<|/det|> +Our results highlight how expedient vaccine delivery to high- risk groups is critical in mitigating COVID- 19 disease and mortality. In scenarios where the number of vaccine doses available is insufficient for high- risk groups (those aged more than 65 years) to receive two vaccine doses, administration of a single dose of vaccine is optimal. This effect is consistent even when vaccine efficacy after one dose is just \(75\%\) of the levels achieved after two doses. These findings offer a nuanced perspective of the critical drivers of COVID- 19 vaccination effectiveness and can inform optimal allocation strategies. These are relevant to high- income countries with a large high- risk group population as well as to low- income countries with younger populations, where the cost and logistical challenges of procuring and delivering two doses for each citizen represent a significant challenge. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 86, 189, 101]]<|/det|> +## Funding + +<|ref|>text<|/ref|><|det|>[[118, 120, 875, 165]]<|/det|> +Bill and Melinda Gates Foundation (OPP1193472); Li Ka Shing Foundation; Oxford University COVID- 19 Research Response Fund (ref: 0009280). + +<|ref|>text<|/ref|><|det|>[[118, 182, 664, 200]]<|/det|> +The funders played no role in the design or outcomes of this work. + +<|ref|>sub_title<|/ref|><|det|>[[118, 254, 227, 270]]<|/det|> +## Contributors + +<|ref|>text<|/ref|><|det|>[[117, 288, 872, 386]]<|/det|> +RA, AB, LJW, and RS conceived the paper and contributed to the analysis. RA developed the model and wrote the code. BG ran the PSA and contributed to data processing. MV and AP provided the data, discussed the analysis, and commented on the draft. All authors have read, contributed to, and approved the final version of the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[118, 439, 317, 455]]<|/det|> +## Declaration of interests + +<|ref|>text<|/ref|><|det|>[[118, 474, 469, 491]]<|/det|> +The authors declare no conflict of interest. + +<|ref|>sub_title<|/ref|><|det|>[[118, 535, 228, 551]]<|/det|> +## Data Sharing + +<|ref|>text<|/ref|><|det|>[[118, 560, 850, 630]]<|/det|> +The data used to inform this analysis have been published and are available in the public domain. The code used is available at: https://github.com/ricardoaguas/como- ChAdOx1- vaccine-. + +<|ref|>sub_title<|/ref|><|det|>[[118, 666, 286, 682]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[117, 690, 866, 815]]<|/det|> +LJW is funded by the Li Ka Shing Foundation. RA is funded by the Bill and Melinda Gates Foundation (OPP1193472). The Covid- 19 Modelling Consortium has support from the Oxford University COVID- 19 Research Response Fund (ref: 0009280). The authors have not been paid to write this article by a pharmaceutical company or other agency. The authors are grateful to Adam Bodley for proofreading the final draft. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 86, 225, 101]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[115, 118, 879, 453]]<|/det|> +As of March \(2^{\text{nd}}\) 2021, almost 115 million people have been diagnosed with COVID- 19 worldwide, and in excess of 2.5 million confirmed deaths have been reported \(^{2,3}\) . Vaccination is a critical strategy to control the spread of SARS- CoV- 2, the virus that causes COVID- 19, and to reduce the severity of symptomatic disease. Three vaccines have already received emergency use authorization in the United Kingdom (UK). The developers of two of these vaccines have reported efficacies of \(95\%\) for their vaccines in their respective Phase 3 trials (Pfizer/BioNTech and Moderna) \(^{4}\) . The third vaccine, ChAdOx1 nCoV- 19, jointly developed by Oxford University and AstraZeneca, demonstrated an acceptable safety profile and efficacy against symptomatic COVID- 19, with no hospital admissions or severe cases of disease reported in the intervention arm during Phase 3 trials conducted in three countries. This vaccine can be stored and distributed at \(2 - 8^{\circ}C\) and will be made available at a lower cost than the other vaccines, making it suitable for global access, particularly in low- and middle- income countries (LMICs) \(^{5 - 7}\) . + +<|ref|>text<|/ref|><|det|>[[115, 468, 877, 906]]<|/det|> +While clinical trials have validated the efficacy of the ChAdOx1 nCoV- 19 vaccine in reducing symptomatic infection, appropriate national vaccination strategies across the world must consider heterogeneity among populations as well as the diverse demographic and socioeconomic environments of affected countries. In particular, the younger population typically present in LMICs justifies the need to assess the effects of associated behaviours and health profiles on vaccine effectiveness. These countries exhibit competing health, social, and economic challenges owing to inadequate healthcare infrastructure and a high prevalence of immunocompromising and infectious diseases. In these settings, individuals could also suffer complex vaccine responses when compared with responses in individuals in more developed economies \(^{8,9}\) . At the same time, many LMICs have been unable to secure vaccine doses in advance from potential suppliers and thus are likely to have incomplete coverage of their populations, particularly in the short- term. The global COVID- 19 vaccine alliance, COVAX, has pledged to procure and distribute vaccines equitably to LMICs; however, this will cover a maximum of \(20\%\) of the total population in each country \(^{10}\) . Although the University of Oxford and AstraZeneca have made the largest supply commitment to LMICs at more affordable prices than other vaccine manufacturers, there is a need to evaluate the impact of a range of factors on vaccine effectiveness \(^{11 - 13}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 83, 880, 338]]<|/det|> +Due to shortages in supply, the UK government has instituted a policy of administering the booster dose of the vaccine at up to 12 weeks following the initial dose, prompting a debate among scientists, manufacturers, and governments on optimal dosing intervals for COVID- 19 vaccines14,15. The purpose of this analysis is therefore to evaluate the efficacy of the ChAdOx1 nCoV- 19 vaccine in countries with different demographic profiles, as a function of vaccine efficacy, dosage regime (interval between initial and booster doses, or no booster at all), coverage, and immunity wave rate. Given the differences in healthcare infrastructure and vaccine access around the world, decision- makers should consider the effect of these factors on population- level impact to determine the most effective strategy for their context13. + +<|ref|>text<|/ref|><|det|>[[116, 354, 875, 504]]<|/det|> +Where vaccination programmes have begun, priority has so far been given to older age groups, individuals with co- morbidities, and frontline medical staff. The model developed therefore considers a simplified system where the vaccine is delivered to age groups in descending order while supplies are available. As there is limited evidence of indirect effects, that is, the potential for reductions in transmission, this vaccine effect was assumed to be negligible for the purposes of this analysis. + +<|ref|>sub_title<|/ref|><|det|>[[118, 558, 196, 574]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[116, 591, 883, 899]]<|/det|> +The methodology employed was very specifically tailored to the research question and its context. Vaccine production rates are always going to be insufficient to meet the demand generated by a global pandemic. In a context of limited vaccine dose availability, it is imperative to prioritize those individuals who would yield the greatest epidemiological benefit. Assuming the most pressing need is to reduce hospitalization rates and deaths, the initial targeting of those at higher risk for these outcomes seems logical, given that the alternative of immunizing sufficient people at lower risk for the indirect benefits to outweigh the direct benefits of a vaccine targeted at those at higher risk is not feasible with the number of vaccines available in the short- term. Even the UK, where mass production of the AztraZeneca vaccine has enabled \(20\%\) of the population to be vaccinated within 3 months, opted to prioritize the high- risk groups (those aged more than 65 years), partially because of the uncertainty around vaccine efficacy against infection. The ChAdOx1 nCoV- 19 vaccine + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 882, 207]]<|/det|> +clinical trials were the only Phase 3 trials in which infection was evaluated as an outcome. No evidence was found for a transmission reduction effect (VE = 3.8% [- 72.4 to 46.3])7, but important questions were raised about how to allocate a limited number of doses to optimize the impact on symptomatic disease, given that a single- dose regimen could offer prolonged protection and thus a delay of the second dose could be warranted. + +<|ref|>text<|/ref|><|det|>[[117, 224, 883, 399]]<|/det|> +We began from the premise laid out above and implemented an individual- based, age- dependent, static transmission model to predict the number of infections, clinical cases, and deaths expected to occur within 6 months of vaccination programme rollout. Individuals are simulated as autonomous systems, each with a set of attributes, informing their serostatus, vaccination uptake history (number of doses and dosing interval), and age. Box 1 details how the dynamics processes inherent to disease transmission and vaccination campaign logistics are considered in the model. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[128, 92, 875, 148]]<|/det|> +Box 1. Consider a hypothetical scenario where the number of SARS- CoV- 2 infections over 6 months in an unvaccinated population is 100,000. Policymakers could opt for one of two alternative options with very different direct benefit outlooks when deciding on how to allocate the limited number of vaccine doses available to them. + +<|ref|>text<|/ref|><|det|>[[128, 160, 875, 196]]<|/det|> +Option 1: Vaccinate high contact groups (aged 25- 40 years) (36% of all infections, 10% of all hospitalizations, 1% of all deaths) \(^{1}\) + +<|ref|>text<|/ref|><|det|>[[189, 209, 320, 222]]<|/det|> +Predicted infections: + +<|ref|>text<|/ref|><|det|>[[189, 237, 860, 253]]<|/det|> +100,000\*36%\*[1- vaccine direct effect on infection] + 100,000\*64%\*[1- vaccine indirect effect on infection] + +<|ref|>text<|/ref|><|det|>[[190, 268, 356, 281]]<|/det|> +Predicted hospitalizations: + +<|ref|>text<|/ref|><|det|>[[190, 296, 874, 330]]<|/det|> +Predicted infections\*10%\*[1- vaccine direct effect on hospitalization] + Predicted infections \*90%\*[1- vaccine Indirect effect on hospitalizations] + +<|ref|>text<|/ref|><|det|>[[190, 345, 301, 358]]<|/det|> +Predicted deaths: + +<|ref|>text<|/ref|><|det|>[[190, 373, 874, 408]]<|/det|> +Predicted infections\*1%\*[1- vaccine direct effect on deaths] + Predicted infections\*99%\*[1- vaccine indirect effect on deaths] + +<|ref|>text<|/ref|><|det|>[[128, 421, 874, 456]]<|/det|> +Option 2: Vaccinate high risk groups (aged >65 years) (10% of all infections, 66% of all hospitalizations, 90% of all deaths) \(^{1}\) + +<|ref|>text<|/ref|><|det|>[[190, 472, 320, 485]]<|/det|> +Predicted infections: + +<|ref|>text<|/ref|><|det|>[[190, 500, 860, 515]]<|/det|> +100,000\*10%\*[1- vaccine direct effect on infection] + 100,000\*90%\*[1- vaccine indirect effect on infection] + +<|ref|>text<|/ref|><|det|>[[190, 530, 356, 543]]<|/det|> +Predicted hospitalizations: + +<|ref|>text<|/ref|><|det|>[[190, 558, 874, 592]]<|/det|> +Predicted infections \*66%\*[1- vaccine direct effect on hospitalization] + Predicted infections \*34%\*[1- vaccine indirect effect on hospitalizations] + +<|ref|>text<|/ref|><|det|>[[190, 607, 301, 620]]<|/det|> +Predicted deaths: + +<|ref|>text<|/ref|><|det|>[[190, 635, 874, 666]]<|/det|> +Predicted infections\*90%\*[1- vaccine direct effect on deaths] + Predicted infections\*10%\*[1- vaccine indirect effect on deaths] + +<|ref|>text<|/ref|><|det|>[[117, 761, 881, 860]]<|/det|> +Thus, vaccines targeting high- contact groups would have to provide indirect effects in the order of 80% (80% reduction in risk in the untargeted population) to prevent an approximate number of deaths similar to that which would be provided by targeting the high- risk group with a direct vaccine effect against death of 85%. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 400, 102]]<|/det|> +## Transmission and clinical cascade + +<|ref|>text<|/ref|><|det|>[[116, 118, 883, 244]]<|/det|> +The spread of COVID- 19 is sensitive to the underlying network of contacts between infectious and susceptible individuals in their various societal spheres (home, work, public transport, etc). For a given population, we can summarize the number of contacts per day as an age- dependent force of infection \(\lambda (a)\) , i.e. a daily risk of acquiring an infection given age \(a\) . The age- dependent risk of infection can then be defined as: + +<|ref|>equation<|/ref|><|det|>[[380, 257, 616, 305]]<|/det|> +\[\lambda (a) = k_{\lambda}\frac{\sum_{j = 1}^{N}c_{ij}P_{j}}{\sum_{i = 1}^{N}(\sum_{j = 1}^{N}c_{ij}P_{j})}\] + +<|ref|>text<|/ref|><|det|>[[115, 319, 883, 576]]<|/det|> +Where \(c_{ij}\) is the daily number of contacts between age groups \(i\) and \(j\) for a particular country, \(P_{j}\) is the population age distribution, \(N\) is the total number of age categories, and \(k_{\lambda}\) is the overall daily risk of infection (which is informed by the number of infectious people in the population). Here, we decided to simplify the transmission process by making the daily risk of infection constant over time, thus having a static transmission model. When evaluating different epidemiological scenarios with different levels of population attack size over the 6- month period explored here, we simply changed the daily risk of infection parameter until the correct attack rate was obtained. Without this constraint, computation of the very high number of simulations required to perform the sensitivity analyses presented here would be virtually impossible. + +<|ref|>text<|/ref|><|det|>[[115, 591, 883, 768]]<|/det|> +The risk of developing severe disease and possibly dying as a consequence of infection was informed by age- dependent infection hospitalization (IHR) and hospitalization fatality (HFR) ratios, published in16. Thus, the modelled daily risk that an individual will develop severe disease is given by \(\lambda (a)IHR(a)\) , whereas the risk of dying is approximately \(\lambda (a)IHR(a)HFR(a)\) . The timing of these events and the lag between infection and clinical outcome are not relevant, as we are only making comparisons between synthetic populations, as detailed below. + +<|ref|>sub_title<|/ref|><|det|>[[118, 820, 463, 838]]<|/det|> +## Vaccination delivery and vaccine efficacy + +<|ref|>text<|/ref|><|det|>[[118, 855, 881, 900]]<|/det|> +Different vaccine dose allocation schemes were simulated, by limiting the number of doses distributed in 6 months, as well as allowing for different dosing intervals (delaying the second + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 881, 155]]<|/det|> +dose) and dosing splits (giving one dose vs two doses). The allocation of doses was always prioritized to the oldest age groups. Individuals were assigned vaccine doses in descending order of age until the maximum number of doses had been allocated. + +<|ref|>text<|/ref|><|det|>[[117, 171, 883, 347]]<|/det|> +Given a fixed number of available doses, one can calculate the target recipient population by looking at the dose- split proposal. If all doses are given as single doses in 6 months, then the target population for vaccination is equal to the number of available doses. At the other extreme, where all vaccines receive two doses, the number of recipients would be half the number of available doses. Within the group that is meant to receive two doses of the vaccine, a \(5\%\) dropout rate (vaccine refusal) was imposed, and a range of booster dose intervals was explored. + +<|ref|>text<|/ref|><|det|>[[117, 363, 883, 514]]<|/det|> +We implemented three different logistical implementations of a vaccine campaign rollout: constant effort, frontloaded, and backloaded. The distinction was in the speed at which the target population received vaccine doses during the initial 2 months. As individuals were assigned a vaccine, the number of doses received would be determined by a draw from a uniform distribution according to the desired dose split. Individuals given two doses would be assigned a booster dose interval following a beta distribution with \(\alpha = 0.15\) and \(\beta = 0.95\) . + +<|ref|>text<|/ref|><|det|>[[117, 529, 882, 628]]<|/det|> +Although vaccine efficacy was explored in the sensitivity analyses presented here, we centred the explored ranges around the point estimates presented in \(^{17}\) . Vaccine efficacy, \(V_{e}(t)\) , was treated as a direct modulator of the risk of infection, clinical disease, and death; it was then defined for each individual, at each timestep of the simulation, as: + +<|ref|>equation<|/ref|><|det|>[[431, 644, 565, 667]]<|/det|> +\[V_{e}(t) = V_{i}^{j}e^{-\delta t}\] + +<|ref|>text<|/ref|><|det|>[[117, 682, 881, 728]]<|/det|> +, where V is the vaccine efficacy in an individual with baseline status \(i\) that received dose number \(j\) a \(t\) number of days ago, while \(\delta\) is the rate of loss of vaccine- induced immunity. + +<|ref|>text<|/ref|><|det|>[[117, 744, 883, 894]]<|/det|> +Throughout this paper, we present a sensitivity analysis of the post- dose two maximum efficacy, the relative efficacy of dose one vs dose two, and the booster dose interval. While doing so, we constrain vaccine efficacy against clinical disease to be the same as that against death, while vaccine efficacy against infection is fixed at \(5\%\) . We also imposed a stepwise increase in post- dose two vaccine efficacy across an 8- week booster dose interval, as observed in the clinical trial \(^{17}\) . This means that giving the second vaccine dose less than 8 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 880, 129]]<|/det|> +weeks after the first dose will result in a \(25\%\) lower post- dose two efficacy relative to the maximum assumed vaccine efficacy. + +<|ref|>sub_title<|/ref|><|det|>[[118, 182, 300, 199]]<|/det|> +## Vaccine effectiveness + +<|ref|>text<|/ref|><|det|>[[118, 217, 880, 262]]<|/det|> +The vaccination campaign population impact is referred to throughout as vaccine effectiveness and was defined as: + +<|ref|>equation<|/ref|><|det|>[[395, 277, 599, 315]]<|/det|> +\[V_{eff} = 100\frac{AR_v - AR_u}{AR_u}\] + +<|ref|>text<|/ref|><|det|>[[117, 330, 883, 456]]<|/det|> +, where \(AR_{v}\) is the attack rate (over 6 months) in the vaccinated population, and \(AR_{u}\) is the attack rate in a population that mirrors the vaccinated population in all aspects except vaccination. We thus have a pair of populations for each parameter set in our analyses and calculate the expected vaccine effectiveness for each parameter set as the relative difference in occurrence of each of the disease endpoints (infection, clinical symptoms, and death). + +<|ref|>sub_title<|/ref|><|det|>[[118, 508, 182, 524]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[117, 541, 883, 770]]<|/det|> +We conducted an extensive initial sensitivity analysis to determine how the impact of rolling out a COVID- 19 vaccination campaign in the UK depends on epidemiological, logistical, and immunological factors. The sensitivity of the modelled vaccine effectiveness to the variables explored is illustrated in Figure 1. It is clear that the prospects for vaccine impact are most sensitive to the number of vaccine doses available within 6 months, the speed of delivery within the same timeframe, and the vaccine efficacy (both the maximum efficacy post- dose two and the relative efficacy of dose one compared with dose two). Interestingly, for the same inputs, the median expected vaccine effectiveness is greater for deaths than it is for clinical cases. + +<|ref|>text<|/ref|><|det|>[[117, 787, 883, 911]]<|/det|> +From this first exploratory analysis one could immediately suggest that, for maximum effectiveness, a vaccine campaign should aim to vaccinate as many people as possible (thus governments/policymakers should procure the maximum number of doses possible), in the shortest time possible. These are by far the two variables the model outputs are most sensitive to, as can clearly be seen in Supplementary Figures 1 and 2. These figures also reveal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 883, 260]]<|/det|> +a very interesting interaction between the vaccine efficacy profile and an actionable decision of how the first available doses are to be distributed. In the frontloaded scenario, where the speed of vaccine delivery is maximal during the early stages of the 6- month vaccination campaign, a single- dose regimen is expected to perform significantly worse if vaccine efficacy post- dose one is \(50\%\) lower than vaccine efficacy post- dose two (top row). However, if the vaccine efficacy after both doses is the same (bottom row), a single- dose regimen can actually be preferable, especially if the number of vaccine doses available is small. + +<|ref|>text<|/ref|><|det|>[[117, 276, 883, 451]]<|/det|> +These initial results prompted further investigation of the possible interactions and trade- offs between the vaccine efficacy profiles and logistical implementation variables. In this detailed analysis, the population attack size was fixed at \(12\%\) , delivery speed to frontloaded, and vaccine- induced protection to last 360 days. The results are summarized in Supplementary Figure 3. As determined by the initial sensitivity analysis, vaccine effectiveness is quite sensitive to the number of available doses, the maximum post- dose two efficacy, and the efficacy of the first dose relative to the second. + +<|ref|>text<|/ref|><|det|>[[117, 468, 883, 775]]<|/det|> +Two interesting results pertain to the sensitivity of the model to changes in the dose number split and the interval in days between doses for the two- dose regimen recipients. While giving everyone two doses, irrespective of all other variables, seems to be preferable to the single- dose option, the length of the whiskers suggests there might be a parameter space for which the single- dose option is optimal, as seen in Supplementary Figures 1 and 2. Increasing the time interval between doses generally produces improved vaccine effectiveness, although a slight decrease in median effectiveness can be observed after an 8- week (56 day) interval. This is further investigated in Supplementary Figures 4 and 5, revealing an interesting trade- off, with a large increase in predicted effectiveness after 7 to 8 weeks, followed by a small decrease as the booster dose interval expands, but only if the efficacy of the first dose is low. If the efficacy of the first dose is similar to the efficacy of the second, increasing the interval between doses up to 12 weeks does not decrease vaccine effectiveness. + +<|ref|>text<|/ref|><|det|>[[117, 790, 883, 914]]<|/det|> +Interestingly, we find a non- linear increase in effectiveness for large values of dose availability, which can be explained by the markedly non- linear risk of severe disease and death with age. As the number of available doses increases, a larger proportion of the population will receive a vaccine dose. However, since vaccines are allocated in descending order of age, as a larger proportion of the population is reached, more and more low- risk + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 880, 129]]<|/det|> +individuals are vaccinated, for whom the vaccine accrued benefits are smaller and smaller. It is then advisable to investigate these relationships for different settings. + +<|ref|>text<|/ref|><|det|>[[116, 145, 883, 476]]<|/det|> +We proceeded to investigate what factors could potentially influence the decision- making process regarding the distribution of doses during the first 6 months of vaccine programme rollout. We thus evaluated the relative predicted effectiveness of the single- dose versus the double- dose regimen, for different countries with potentially different dose availability, and assuming different vaccine efficacy profiles (Figure 2). For very high levels of dose allocation (high y- axis values), a two- dose regimen is clearly optimal. This starts to become less evident for scenarios where the protection conferred by the first dose gets closer to the post- dose two efficacy (moving right along the x- axis). Interestingly, when the number of available doses is small, the single- dose regimen will become more effective than the double- dose regimen, as the thick black line is crossed. The parameter combinations defining the line where there is a shift in strategy positioning are country dependent, with countries with older populations having a larger parameter space in which a single- dose option is preferable, as shown in Figure 3. + +<|ref|>sub_title<|/ref|><|det|>[[118, 531, 208, 547]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[116, 565, 880, 899]]<|/det|> +The SARS- CoV- 2/COVID- 19 pandemic has created an unprecedented public health challenge, spurring a global race to develop and distribute viable vaccines. A vaccine that creates broad immunity against the SARS- CoV- 2 virus could be the only effective means to control the pandemic and allow a return to "normalcy". To have a significant impact on the disease, a critical mass of the global population at risk will need to be vaccinated. However, many high- income countries have secured more than half of the available vaccine doses for themselves, leaving LMICs, which make up more than 85% of the global population, to find their own solutions18. To address the problem of equitable access, WHO, Gavi, and the Coalition for Epidemic Preparedness Innovations (CEPI) established COVAX, a global alliance that has pledged to pool investment and allocate and distribute COVID- 19 vaccines equitably, particularly in LMICs10. However, COVAX is currently under- resourced and the doses secured are insufficient to achieve the coverage levels needed19. Supply constraints and new variants of SARS- CoV- 2 are steering countries towards strategies that counter low + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 870, 182]]<|/det|> +access with dosing patterns or volumes to maximize the impact of the vaccines. Data from the ChAdOx1 nCoV- 19 vaccine trials have allowed us to explore potential strategies to inform optimal allocation programmes, particularly in contexts where the cost and logistics of implementing multiple doses within a short timeframe may be challenging. + +<|ref|>text<|/ref|><|det|>[[117, 198, 875, 374]]<|/det|> +Our findings indicate that vaccine effectiveness is dependent upon (i) the country context, which includes the demographic profile, the attack rate of the virus, and the amount of vaccine that is available (which influences the proportion of the population that is vaccinated); ii) the characteristics of the vaccine, which include the efficacy of a single dose relative to a double dose and the waning of efficacy over time; and iii) the proportion of the population receiving the second dose, the time interval between doses, and the delivery speed. + +<|ref|>text<|/ref|><|det|>[[116, 390, 880, 670]]<|/det|> +Our analysis demonstrates that in scenarios where the number of vaccine doses available is insufficient for the highest risk groups (aged \(>65\) years) to receive two doses, the allocation of a single vaccine dose is optimal. This effect is consistent even when the vaccine efficacy of a single dose is just \(75\%\) of the levels achieved after a double dose, until allocation drops to a population coverage of \(10\%\) , after which vaccinating only the high- risk individuals, with two doses, is more effective. In scenarios where the number of doses available to the country is sufficiently high, or if the relative single- dose efficacy is low ( \(50\%\) or less), providing a booster dose within 8 weeks would be preferable. Apart from these specific conditions, the results indicate that providing individuals with two doses of vaccine would have a similar effectiveness to the use of a single dose given to twice the number of individuals. + +<|ref|>text<|/ref|><|det|>[[117, 687, 853, 888]]<|/det|> +The speed at which the high- risk population is vaccinated greatly influences the expected vaccine effectiveness in preventing clinical cases and death. This is particularly true if the transmission rate is high, with faster vaccination reducing the number of infections in groups awaiting their first dose during the rollout. Distributing the vaccine very slowly provides an effectiveness of less than \(10\%\) , regardless of the number of doses and allocations. The impact of allocation on outcomes is also greater when the population is vaccinated rapidly over a six- month period. In both of these scenarios, providing a single dose is preferable. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 288]]<|/det|> +An interesting trade- off was found between the booster dose interval and the relative vaccine efficacy of a single dose. For vaccines with large differences between first and second dose efficacy, delaying the booster dose interval past 8 weeks after the first dose was found to be detrimental. However, if a single dose provided at least \(75\%\) of the protection conferred by a double dose, delaying the booster dose interval to 12 weeks had a negligible impact on the number of cases and deaths. Given the similar reported efficacies of single and double doses of ChAdOx1 nCoV- 19, a 12- week interval is the optimal scenario for this vaccine17. However, this finding may not be applicable to other COVID- 19 vaccines. + +<|ref|>text<|/ref|><|det|>[[116, 302, 881, 451]]<|/det|> +These differences are more profound when considering the demographic characteristics of a population. In high- income countries, which have a larger older population (>65 years), a single- dose regimen will allow the vaccination of more individuals more quickly, with a correspondingly greater impact on cases and deaths. In the UK, the six- month allocation threshold above which a two- dose regimen would be preferred was found to be about \(16.5\%\) . + +<|ref|>text<|/ref|><|det|>[[116, 467, 878, 696]]<|/det|> +The six- month allocation threshold above which a two- dose regimen would be preferred is much lower in LMICs, mainly due to mortality in the younger population. In these contexts, decision- makers will need to consider the affordability, availability, and logistical constraints and feasibility of implementing a single or a double dose, the dosage intervals, and delivery speed. Most LMICs lack the digital databases necessary to manage patient data, reliably track vaccine inventories, keep track of who has received which vaccine, and inform people where and when they are due for a booster. Governments would also need to ensure that they reserve sufficient stocks to allow the administration of booster doses. In these cases, a robust cost- benefit analysis of each option will need to be considered. + +<|ref|>text<|/ref|><|det|>[[116, 712, 870, 914]]<|/det|> +The dosing interval for COVID- 19 vaccines has been a subject of debate among scientists, regulators, and governments around the world following the UK government's decision to prioritize administering the first dose of vaccine to as many at- risk people as possible and increasing the interval between the two doses to up to 12 weeks14,15,20. A one- dose vaccine regimen or a two- dose regimen with longer time intervals may be sufficient to reduce symptoms of COVID- 19 in the most vulnerable individuals and ultimately slow the pandemic, given that the time difference between first and second doses was shown to have a negligible effect on overall vaccine effectiveness (clinical cases, infections, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 875, 234]]<|/det|> +deaths). Indeed, a recent WHO notification stated that some countries are facing "exceptional circumstances" and may want to delay second doses to "allow for a higher initial coverage". Other exceptional circumstances may involve trade-offs around the relative size of the highest risk population in a country and the currently unknown potential for a vaccine to reduce transmission, which may lead to some countries targeting high- contact groups to benefit from any potential indirect effects. + +<|ref|>text<|/ref|><|det|>[[115, 250, 880, 400]]<|/det|> +Nevertheless, these thresholds are likely to differ depending on the country context. For example, smaller countries may be able to rapidly rollout the vaccine to a higher percentage of their population compared with the speed at which larger countries can do this. It should be noted that an implementation strategy is determined at a country level. The assumptions made in this work are based on the association of certain parameters with the health infrastructure and existing population of certain country groups defined by income level. + +<|ref|>text<|/ref|><|det|>[[116, 415, 867, 566]]<|/det|> +Published clinical data were used to inform the parameters used in the model described in this paper. These data provide an aggregate efficacy of the ChAdOx1 nCoV- 19 vaccine among people of a wide range of ages living in different countries. However, there were limited data available for assessing the effects of certain parameters (such as the effect of the dosing interval on post- dose two efficacy) on vaccine efficacy, which begat the need to conduct the post hoc exploratory analysis presented here. + +<|ref|>sub_title<|/ref|><|det|>[[118, 618, 212, 634]]<|/det|> +## Conclusion + +<|ref|>text<|/ref|><|det|>[[116, 653, 880, 880]]<|/det|> +This analysis demonstrates that in scenarios where the number of vaccine doses available is insufficient for the highest risk groups (>65 years of age) to receive two vaccine doses, allocation of a single vaccine dose to twice the number of individuals or extending the time interval between doses may be more optimal strategies. In contexts without supply constraints, or if the single- dose efficacy is low, providing a booster dose would be preferable. Apart from these specific conditions, the results indicate that providing individuals with two doses of vaccine would have a similar effectiveness to the use of a single dose in twice the number of individuals. In an ideal world, decisions about vaccination strategies would be made within the exact parameters of the trials that have been + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 842, 129]]<|/det|> +conducted. However, the limited availability of resources, and specific country contexts, may require decision- makers to consider alternative strategies. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[118, 86, 275, 100]]<|/det|> +# Tables and Figures + +<|ref|>table_caption<|/ref|><|det|>[[118, 146, 335, 158]]<|/det|> +Table 1: Model parameters. + +<|ref|>table<|/ref|><|det|>[[118, 185, 900, 776]]<|/det|> + +
ParameterModel termRangeDescription
Population attack size (% of the population)ATT(4; 12; 20)This is the percentage of the population
infected within the 6-month study period
Vaccine allocation (% of the
population during study
period)		TRG(5; 10; 20; 30)Allocation range was based on the assumed administration speed. Using current data,
we assumed that higher-income countries
could reach a maximum speed, allowing
30% coverage of the population within 6
months
Second dose administered
(% of the vaccinated
population administered a
second dose)		DSP(0; 25; 50; 75;
100)		This is the percentage of the vaccinated
population that are administered a second (booster) dose
Interval between first dose and booster doseBTI(4 weeks; 7
weeks; 12 weeks)		The interval between doses can affect
vaccine efficacy; the range chosen was
based on available clinical trial data
Vaccine delivery speedDEL(fixed;
frontloaded;
backloaded)		The speed of vaccine delivery to the
population - see Supplementary figure 6
Vaccine efficacy after the
second dose		PD2(65; 75; 85)Maximum efficacy following the second
dose
Vaccine efficacy of the first
dose compared with the
second dose (%)		D2B(50; 75; 100)Effect of the first dose compared with the
second dose
Immunity wane rate (days
following last dose)		VCW(90; 180; 360;
540)		Vaccine protection decay post-last dose
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 120, 860, 636]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 85, 707, 100]]<|/det|> +
Fig. 1: Overall sensitivity analysis of vaccine effectiveness, based on UK data.
+ +<|ref|>text<|/ref|><|det|>[[118, 656, 881, 742]]<|/det|> +The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. The full list of parameters explored and their descriptions can be found in Table 1. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 112, 880, 380]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 86, 357, 101]]<|/det|> +
Fig. 2: Optimal dose allocation.
+<|ref|>image_caption<|/ref|><|det|>[[366, 383, 631, 396]]<|/det|> +
Efficacy of Single dose vs Double dose regimen
+ +<|ref|>text<|/ref|><|det|>[[118, 408, 881, 529]]<|/det|> +The coloured surfaces and respective contour lines indicate the ratio between the predicted vaccine effectiveness for a double- dose regimen vs a single- dose regimen. This ratio is a mean ratio, obtained by averaging out the ratios obtained in all model runs assuming the corresponding x and y parameter values and thus are not expected to be regular. Contour line 1 (thicker black line) indicates the parameter combinations for which there is no expected difference between giving everyone a single dose vs giving everyone two doses. For values greater than 1 (hot colours), a two- dose regimen is preferable, and for values less than 1 (cold colours), a single- dose regimen is preferable. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 155, 730, 533]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 85, 545, 100]]<|/det|> +
Fig. 3: Dose allocation thresholds in different countries.
+ +<|ref|>text<|/ref|><|det|>[[118, 580, 881, 648]]<|/det|> +The figure illustrates the parameter combinations that define the allocation threshold above which a two- dose regimen would be preferred over a single- dose regimen. The areas under the curves are \(16.5\%\) , \(8\%\) , and \(3.8\%\) for the UK, Brazil, and Uganda, respectively, which correlates almost perfectly with the proportion of the population above the age of 65 years in those countries. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 152, 870, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 110, 875, 144]]<|/det|> +
Supplementary Fig. 1: Impact of vaccine delivery speed, dose split, and dose availability on vaccine effectiveness as a measure of reduction in clinical cases.
+ +<|ref|>text<|/ref|><|det|>[[118, 523, 881, 610]]<|/det|> +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 875, 117]]<|/det|> +Supplementary Fig. 2: Impact of vaccine delivery speed, dose split, and dose availability on vaccine effectiveness as a measure of reduction in deaths. + +<|ref|>image<|/ref|><|det|>[[120, 152, 875, 510]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[117, 550, 881, 636]]<|/det|> +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 137, 864, 653]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 84, 863, 117]]<|/det|> +
Supplementary Fig. 3: Detailed sensitivity analysis of vaccine effectiveness for the most sensitive parameters, based on UK data.
+ +<|ref|>text<|/ref|><|det|>[[118, 673, 850, 742]]<|/det|> +The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 85, 820, 117]]<|/det|> +Supplementary Fig. 4: Impact of vaccine booster dose interval on vaccine effectiveness as a measure of reduction in clinical cases. + +<|ref|>image<|/ref|><|det|>[[118, 127, 878, 483]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[118, 496, 881, 584]]<|/det|> +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 125, 879, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 85, 820, 117]]<|/det|> +
Supplementary Fig. 5: Impact of vaccine booster dose interval on vaccine effectiveness as a measure of reduction in deaths.
+ +<|ref|>text<|/ref|><|det|>[[117, 523, 881, 610]]<|/det|> +The white number on the black background in each panel defines the vaccine efficacy of the first dose relative to the second dose. Lines represent the mean effectiveness calculated using all runs, where the parameters are those defined by each figure, irrespective of all remaining parameters. Lines are coloured according to the dose split, i.e. the proportion of individuals receiving two vaccine doses. These results are based on the UK population structure. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 130, 583, 420]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 85, 470, 100]]<|/det|> +
Supplementary Fig. 6: Vaccine delivery speed.
+ +<|ref|>text<|/ref|><|det|>[[117, 457, 831, 491]]<|/det|> +The figure shows the number of vaccine doses administered over the course of the vaccination campaign (6 months). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 213, 101]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[113, 110, 880, 833]]<|/det|> +1. Office for National Statistics. Coronavirus (COVID-19) weekly insights: latest health indicators in England, 12 February 2021. February 12, 2021 ed. United Kingdom; 2021. +2. Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time. Lancet Infect Dis 2020; 20(5): 533-4. +3. Johns Hopkins University. COVID-19 Dashboard. 2020. https://coronavirus.jhu.edu/map.html (accessed June 2020. +4. Pfizer. Pfizer and BioNtech announce vaccine candidate against Covid-19 achieved success in first interim analysis from phase 3 study. 2020. +5. Folegatti PM, Ewer KJ, Aley PK, et al. Safety and immunogenicity of the ChAdOx1 nCoV-19 vaccine against SARS-CoV-2: a preliminary report of a phase 1/2, single-blind, randomised controlled trial. Lancet 2020; 396(10249): 467-78. +6. Ramasamy MN, Minassian AM, Ewer KJ, et al. Safety and immunogenicity of ChAdOx1 nCoV-19 vaccine administered in a prime-boost regimen in young and old adults (COV002): a single-blind, randomised, controlled, phase 2/3 trial. Lancet 2021; 396(10267): 1979-93. +7. Voysey M, Clemens SAC, Madhi SA, et al. Safety and efficacy of the ChAdOx1 nCoV-19 vaccine (AZD1222) against SARS-CoV-2: an interim analysis of four randomised controlled trials in Brazil, South Africa, and the UK. Lancet 2021; 397(10269): 99-111. +8. MacLennan CA. Vaccines for low-income countries. Seminars in Immunology 2013; 25(2): 114-23. +9. Rodrigues CMC, Plotkin SA. Impact of Vaccines; Health, Economic and Social Perspectives. Frontiers in Microbiology 2020; 11(1526). +10. WHO. Fair allocation mechanism for COVID-19 vaccines through the COVAX Facility. Geneva: WHO, 2020. +11. Davies NG, Klepac P, Liu Y, Prem K, Jit M, Eggo RM. Age-dependent effects in the transmission and control of COVID-19 epidemics. Nat Med 2020; 26(8): 1205-11. +12. Pullano G, Valdano E, Scarpa N, Rubrichi S, Colizza V. Evaluating the effect of demographic factors, socioeconomic factors, and risk aversion on mobility during the COVID-19 epidemic in France under lockdown: a population-based study. The Lancet Digital Health 2020; 2(12): e638-e49. +13. Wang W, Wu Q, Yang J, et al. Global, regional, and national estimates of target population sizes for covid-19 vaccination: descriptive study. BMJ 2020; 371: m4704. +14. Iacobucci G, Mahase E. Covid-19 vaccination: What's the evidence for extending the dosing interval? BMJ 2021; 372: n18. +15. Mahase E. Covid-19: Medical community split over vaccine interval policy as WHO recommends six weeks. BMJ 2021; 372: n226. +16. Brazeau N, Verity R, Jenks S, et al. COVID-19 Infection Fatality Ratio Estimates from Seroprevalence. London: Imperial College London, 2020. +17. Voysey M, Clemens SAC, Madhi SA, et al. Single dose administration, and the influence of the timing of the booster dose on immunogenicity and efficacy of ChAdOx1 nCoV-19 (AZD1222) vaccine Lancet 2021. +18. Twohey M, Collins K, Thomas K. With First Dibs on Vaccines, Rich Countries Have 'Cleared the Shelves'. New York Times. 2020 December 15, 2020. +19. UN News. COVID-19 vaccines: donors urged to step up funding for needy countries. New York: UN News; 2020. +20. Davis N. Vaccine experts call for clarity on UK's 12-week Covid jab interval. The Guardian. 2021 January 25, 2021. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 102, 765, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 46, 143, 71]]<|/det|> +
Figures
+ +<|ref|>text<|/ref|><|det|>[[42, 728, 949, 839]]<|/det|> +Overall sensitivity analysis of vaccine effectiveness, based on UK data. The boxplots show the median and interquartile ranges of the predicted vaccine effectiveness on each of the outcomes for specific parameters. They were generated by aggregating all model simulations for each of the parameters, with each boxplot summarizing the variance in predicted vaccine efficacy for all possible combinations of the other parameters. The full list of parameters explored and their descriptions can be found in Table 1. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 45, 789, 345]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 368, 118, 388]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[41, 410, 958, 570]]<|/det|> +Optimal dose allocation. The coloured surfaces and respective contour lines indicate the ratio between the predicted vaccine effectiveness for a double- dose regimen vs a single- dose regimen. This ratio is a mean ratio, obtained by averaging out the ratios obtained in all model runs assuming the corresponding \(x\) and \(y\) parameter values and thus are not expected to be regular. Contour line 1 (thicker black line) indicates the parameter combinations for which there is no expected difference between giving everyone a single dose vs giving everyone two doses. For values greater than 1 (hot colours), a two- dose regimen is preferable, and for values less than 1 (cold colours), a single- dose regimen is preferable. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 92, 644, 512]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 553, 117, 572]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[41, 593, 953, 705]]<|/det|> +Dose allocation thresholds in different countries. The figure illustrates the parameter combinations that define the allocation threshold above which a two- dose regimen would be preferred over a single- dose regimen. The areas under the curves are \(16.5\%\) , \(8\%\) , and \(3.8\%\) for the UK, Brazil, and Uganda, respectively, which correlates almost perfectly with the proportion of the population above the age of 65 years in those countries. + +<|ref|>sub_title<|/ref|><|det|>[[44, 728, 310, 755]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 778, 765, 799]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 817, 315, 837]]<|/det|> +Supplementary Figures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/images_list.json b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..abcc1e3d8b93764e064895ce0f6c9d91d9af8460 --- /dev/null +++ b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Workflow of the SpaTalk method and visualization. a Overview of SpaTalk, including the input, intermediate process of decoding spatially resolved cell–cell communications, and output. b Conceptual framework of cell-type decomposition with SpaTalk. Five different spatial technologies and datasets were selected and analyzed: spot-based ST data (Slide-seq and 10X Visium) and single-cell ST data (STARmap, MERFISH, and seqFISH+). NNLM was used to dissect the optimal proportion of cell types for the projection of cells from scRNA-seq reference data onto the spatial cells/spots, generating single-cell ST data with known cell types. c Schematic representation of SpaTalk to infer spatially resolved cell–cell communications", + "footnote": [], + "bbox": [ + [ + 152, + 230, + 840, + 750 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Superior performance of SpaTalk over existing methods. a Schematic diagram for generating simulated spot data. Cells were split according to the fixed spatial distance and then merged for the single-cell ST data with known cell types. b Performance comparison of SpaTalk with other existing cell-type deconvolution methods (RCTD, Seurat, SPOTlight, deconvSeq, Stereoscope). The asterisk represents the top-ranked method for each dataset. NA, not available. c Schematic illustration of the procedure and rationale for single-cell ST data to evaluate predicted LRIs that mediate spatially resolved cell–cell communications. d and e Performance comparison of SpaTalk with existing cell–cell communication inference methods (Giotto, SpaOTsc, NicheNet, CytoTalk, and CellCall) on the STARmap and seqFISH+ datasets. The \\(P\\) value", + "footnote": [], + "bbox": [ + [ + 147, + 325, + 844, + 710 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Identification of spatial inhibitory signal transmission among neurons and non-neuronal cells. a STARmap single-cell ST dataset of the mouse visual cortex involving 973 cells and 1020 genes. Astro, astrocytes; eL2/3, eL4, eL5, eL6, excitatory neuron subtypes; Endo, endothelial cells; HPC, hippocampus; Micro, microglia; Oligo, oligodendrocytes; SMC, smooth muscle cells; cc, corpus callosum. b Significantly enriched LRIs that mediate cell–cell communications among neurons and non-neuronal cells inferred by SpaTalk with \\(P< 0.05\\) . The \\(P\\) value represents the significance of spatial proximity of LRIs using the permutation test. c Sanky plot of the associations among ligands, receptors, and biological processes or pathways in the KEGG and Reactome databases that mediate cell–cell communications in the central nervous system. d Spatial distribution and intra-cellular signaling pathways of the Cort–Sstr2 pairs between the eL5 senders and Astro receivers. \\(P\\) values were calculated with the Wilcoxon test. e Co-expression of target genes in receivers and the percentage of expressed cells for target genes. f Significantly enriched biological processes and pathways with the ligand-receptor-target genes using the Fisher exact test. g Slide-seq spot-based ST dataset of the mouse visual cortex involving 42,550 spots and 22,542 genes. OPC, oligodendrocyte progenitor cell. h Communications of eL5–Astro mediated by the Cort–Sstr2 interaction in space and the intra-cellular signal pathway", + "footnote": [], + "bbox": [ + [ + 149, + 88, + 846, + 536 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Modulation of periportal hepatocytes on the metabolic microenvironment sensed by pericentral hepatocytes. a Slide-seq spot-based ST dataset of the mouse liver involving 25,595 spots and 17,545 genes. b Cell-type decomposition by SpaTalk. PC, pericentral; PP, periportal; Hep, hepatocytes. c Scaled Pearson's correlation coefficients between the expression of known marker genes and the percent for each cell type. DC, dendritic cell; Macro, macrophages. d Enriched LRIs that mediate cell–cell communications between pericentral and periportal hepatocytes. e Significant differentially expressed genes (DEGs) between periportal and pericentral hepatocytes assessed with the Wilcoxon test and the corresponding significantly enriched biological processes and pathways determined with the Metascape web tool. Representative DEGs are labeled beside the heatmap. f Significantly activated pathways in pericentral (up) and periportal (down) hepatocytes determined by Gene Set Enrichment Analysis (GSEA). g Communications from periportal hepatocytes to", + "footnote": [], + "bbox": [ + [ + 150, + 191, + 844, + 636 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Spatial characterization of tumor and stromal cells in human squamous cell", + "footnote": [], + "bbox": [ + [ + 150, + 400, + 839, + 875 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 Reconstruction of cell–cell communications between TSK subpopulations and stromal cells in space with SpaTalk. a Top 20 inferred LRIs that mediate cell–cell communication from the TSK senders to the FB and Endo receivers. Colored blocks represent the known ligand–receptor pairs in CellTalkDB. The asterisk represents the significantly enriched LRIs determined by SpaTalk. b Top 20 inferred LRIs that mediate cell–cell communication from the FB and Endo senders to TSK receivers. c Region of interest (ROI) covering 42 spatial spots in space with a high total score of TSK, FB, and Endo according to their signature genes. The point plot shows the decomposed single-cell ST atlas determined by SpaTalk. d Expression of known cancer-associated fibroblast (CAF) markers across TSKs, FB, Endo, and other cells. e Differentially expressed genes (DEGs) between EMT-like and EMT-unlike TSKs and the corresponding", + "footnote": [], + "bbox": [ + [ + 155, + 188, + 840, + 675 + ] + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5.mmd b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6f202c6611241b1b51dc5487f5a625f8ec59cd84 --- /dev/null +++ b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5.mmd @@ -0,0 +1,371 @@ + +# Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk + +Xin Shao Zhejiang University https://orcid.org/0000- 0002- 1928- 3878 Chengyu Li Zhejiang University https://orcid.org/0000- 0003- 3144- 9460 Haihong Yang Zhejiang University Xiaoyan Lu Zhejiang University Jie Liao Zhejiang University https://orcid.org/0000- 0002- 6697- 8998 Jingyang Qian Zhejiang University Kai Wang Zhejiang University Junyun Cheng Zhejiang University Penghui Yang Zhejiang University Huajun Chen Zhejiang University Xiao Xu Zhejiang University Xiaohui Fan ( \(\boxed{\pmb{\times}}\) fanxh@zju.edu.cn ) Zhejiang University https://orcid.org/0000- 0002- 6336- 3007 + +## Article + +# Keywords: + +Posted Date: April 22nd, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1576678/v1 + +<--- Page Split ---> + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on July 30th, 2022. See the published version at https://doi.org/10.1038/s41467-022-32111-8. + +<--- Page Split ---> + +# Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk + +Xin Shao1, 3, †, Chengyu Li3, †, Haihong Yang2, 4, †, Xiaoyan Lu3, Jie Liao3, Jingyang Qian3, Kai Wang1, Junyun Cheng3, Penghui Yang3, Huajun Chen2, 4, *, Xiao Xu1, 5, *, Xiaohui Fan1, 3, 5, *1 Key Laboratory of Integrated Oncology and Intelligent Medicine of Zhejiang Province, Department of Hepatobiliary and Pancreatic Surgery, Affiliated Hangzhou First People's Hospital, Zhejiang University School of Medicine, Hangzhou 310006, China. 2 Hangzhou Innovation Center, Zhejiang University, Hangzhou 310058, China 3 Pharmaceutical Informatics Institute, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, China 4 College of Computer Science and Technology, Zhejiang University, Hangzhou 310027, China 5 Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou 310003, China. 1 These authors should be regarded as Joint First Authors 1 Corresponding author: Dr. Huajun Chen (huajunsir@zju.edu.cn), Dr. Xiao Xu (E-mail: zjxu@zju.edu.cn) and Dr. Xiaohui Fan (E-mail: fanxh@zju.edu.cn, Tel/Fax: 86-571- 88208596) + +<--- Page Split ---> + +## Abstract + +Spatially resolved transcriptomics (ST) provides genetic information in space toward elucidation of the spatial architecture in intact organs and the spatially resolved cell- cell communications mediating tissue homeostasis, development, and disease. To facilitate inference of spatially resolved cell- cell communications from ST data, we here present SpaTalk, which relies on a graph network and knowledge graph to model and score the ligand- receptor- target signaling network between spatially proximal cells, decomposed from ST data through a non- negative linear model and spatial mapping between single- cell RNA- sequencing and ST data. The performance of SpaTalk benchmarked on public single- cell ST datasets was superior to that of existing cell- cell communication inference methods. SpaTalk was then applied to STARmap, Slide- seq, and 10X Visium data, revealing the in- depth communicative mechanisms underlying normal and disease tissues with spatial structure. SpaTalk can uncover spatially resolved cell- cell communications for single- cell and spot- based ST data universally, providing new insights into spatial inter- cellular dynamics. + +<--- Page Split ---> + +## Introduction + +Cell- cell communications via secreting and receiving ligands frequently occur in multicellular organisms, which is a vital feature involving numerous biological processes1. Standard algorithms for inferring cell- cell communications mediated by ligand- receptor interactions (LRIs) primarily incorporate a database of known LRIs and single- cell transcriptomic data by delineating cell populations and their lineage relationships2, 3. One common strategy is to integrate the abundance of ligands and receptors for the inference of signals from senders to receivers based on the premise that highly co- expressed ligands and receptors are likely to mediate inter- cellular communications4, 5. Another strategy applies the downstream targets triggered by LRIs in receivers to enrich and score the ligand- receptor- target (LRT) signaling network6- 8. Although single- cell transcriptomic data can provide information on the genes contributing to cell- cell communications, the spatial information of cells is inevitably lost when dissociating tissues into single cells, thereby hindering the extension of current tools to investigate cell- cell communications in tissues with spatial structure9. + +Recent technological advances in spatially resolved transcriptomics (ST) benefiting from spatial barcoding and imaging- based approaches have enabled the measurement of whole or mostly whole transcriptomes while retaining the spatial information10, 11, which have been increasingly adopted to generate new insights in the biological and biomedical domains, with dramatically improved accuracy and reliability in the inference of spatially proximal cell- cell communications12. Given the space- constrained nature of juxtacrine and paracrine signaling, such spatial gene expression information is vital to understand cell- cell communications mediating tissue homeostasis, development, and disease13, 14. + +<--- Page Split ---> + +Several methods have recently emerged to decode the mechanisms of cell- cell communications in space15. For example, Giotto utilizes preferential cell neighbors over single- cell ST datasets for each pair of cell types with an enrichment test to evaluate the likelihood of a given LRI based on proximal co- expressing cells and infer cell- cell communication in space16. SpaOTsc applies structured optimal transport mapping between scRNA- seq and ST data to assign a spatial position for each cell, resulting in a cell- cell distance as a transport cost to infer the ligand- receptor signaling network that mediates space- constrained cell- cell communication17. However, Giotto and SpaOTsc are limited to infer inter- cellular communications over single- cell ST data rather than the spot- based ST data and between paired cell types rather than paired cells. It still lacks of methods that can infer and visualize spatially resolved cell- cell communications at single- cell resolution over ST data to date, posing a great challenge for decoding spatial inter- cellular dynamics underlying disease pathology. + +To address this challenge, we herein proposed SpaTalk, a spatially resolved cell- cell communication inference method by creatively integrating the principles of the ligand- receptor proximity and ligand- receptor- target (LRT) co- expression to model and score the LRT signaling network between spatially proximal cells relying on the graph network and knowledge graph approaches18. The performance of SpaTalk was evaluated on benchmarked datasets with remarkable superiority over other methods. By applying to STARmap19, Slide-seq20, 21, and 10X Visium22 datasets, SpaTalk revealed the in- depth communicative mechanisms underlying normal and disease tissues with spatial structure. Collectively, these results demonstrate SpaTalk as a useful and universal method that can help to uncover spatially resolved cell- cell communications for both single- cell and spot- based ST data, providing insights into the understanding + +<--- Page Split ---> + +of spatial inter- cellular dynamics in tissues. + +## Results + +Overview of the SpaTalk method. Fig. 1 provides an overview of the workflow for developing and testing SpaTalk, comprising two main components: (1) dissect the cell- type composition of ST data and (2) infer spatially resolved cell- cell communications over decomposed single- cell ST data (Fig. 1a). In the first component, the non- negative linear model (NNLM) \(^{23 - 25}\) was applied to decode the cell- type composition for a single- cell or spot- based ST data matrix using the scRNA- seq data matrix with k cell types as the underlying reference. By incorporating Lee's multiplicative iteration algorithm and relative entropy loss \(^{25}\) , the model was trained with default hyperparameters until convergence, producing a weight matrix representing the optimal proportion of cell types for each cell/spot. For single- cell ST data, the cell type with the maximum weight was assigned to label each cell. For spot- based ST data, the cell types with different weights were used as the reference to project the cells from scRNA- seq data onto the spatial spot (Fig. 1b). Through random sampling and deep iteration processes, the optimal cellular combination that most resembled the spatial spot was refined to reconstruct the single- cell ST data for spot- based ST data. + +The second component of SpaTalk is to infer spatially resolved cell- cell communications and downstream signal pathways. To identify possible communications among cells mediated by LRIs, the principles of ligand- receptor proximity and ligand- receptor- target (LRT) co- expression were incorporated based on a recent review \(^{12}\) . In detail, the KNN algorithm is first applied to each cell in space to construct the cell graph network. For the ligand of the sender (cell type A) and the + +<--- Page Split ---> + +receptor of the receiver (cell type B), the number of LRI pairs is obtained from the graph network by counting the 1- hop neighbor nodes of receivers for each sender. A permutation test filters and scores the significantly enriched LRIs, generating the inter- cellular score (Fig. 1c). + +![](images/Figure_1.jpg) + +
Fig. 1 Workflow of the SpaTalk method and visualization. a Overview of SpaTalk, including the input, intermediate process of decoding spatially resolved cell–cell communications, and output. b Conceptual framework of cell-type decomposition with SpaTalk. Five different spatial technologies and datasets were selected and analyzed: spot-based ST data (Slide-seq and 10X Visium) and single-cell ST data (STARmap, MERFISH, and seqFISH+). NNLM was used to dissect the optimal proportion of cell types for the projection of cells from scRNA-seq reference data onto the spatial cells/spots, generating single-cell ST data with known cell types. c Schematic representation of SpaTalk to infer spatially resolved cell–cell communications
+ +<--- Page Split ---> + +mediated by LRIs. The inter- cellular and intra- cellular scores were obtained and combined from the cell- cell graph network and the LRT- knowledge graph (KG), respectively, by integrating the KNN, permutation test, and random walk algorithms. L, ligand; R, receptor; TF, transcription factor; T, target. d Visualization of spatially resolved cell- cell communications, including a heatmap, Sankey plot, and diagram of the LRI from senders to receivers in space, as well as ligand- receptor- target (LRT) signaling pathways over the reconstructed single- cell ST data. + +The knowledge graph (KG) was then introduced to model the intracellular signal propagation process. In practice, LRIs from CellTalkDB26, pathways from Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome, and TFs from AnimalTFDB27 were integrated to construct the LRT- KG, wherein the weight between entities represents the co- expressed coefficient. Taking the receptor as the query node, we incorporated the random walk algorithm28 into the LRT- KG to filter and score the downstream activated TFs and calculate the intracellular score of the LRI from senders to receivers. Inter- cellular and intra- cellular scores are combined to rank the LRIs that mediate spatially resolved cell- cell communications. + +SpaTalk also includes numerous visualization functions to characterize the cell- type composition and spatially resolved cell- cell communications, such as the diagram of the LRI from senders to receivers in space and LRT signaling pathways, over the reconstructed single- cell ST data (Fig. 1d). Five broad ranges of different spatial technologies and corresponding representative datasets were analyzed and visualized: spot- based ST data (Slide- seq20, 21 and 10X Visium22) and single- cell ST data (STARmap19, MERFISH29, and seqFISH+30). + +Performance comparison of SpaTalk with other methods. The cell- type decomposition by SpaTalk is the foundation for subsequent analyses. To evaluate its performance, four single- cell ST datasets from the mouse cortex, hypothalamus, + +<--- Page Split ---> + +olfactory bulb, and sub- ventricular zone were utilized (Supplementary Fig. 1a). All cells were split according to the fixed spatial distance and then merged into simulated spots as the benchmark datasets (Fig. 2a). The quality of predicted cell- type decompositions and expression profiles was evaluated by Pearson's correlation coefficient and the root mean square error (RMSE) based on the ground truth, wherein SpaTalk exhibited fantastic performance over the benchmark datasets (Supplementary Fig. 1b, c). Although the majority of existing cell- type deconvolution methods (RCTD31, Seurat32, SPOTlight33, deconvSeq34, and Stereoscope35) can achieve a decent correlation coefficient and low RMSE on spot deconvolution, SpaTalk outperformed these methods on most benchmark datasets with the top- ranked performance, except for the MERFISH dataset (Fig. 2b). The MERFISH dataset only includes 155 genes, whereas the STARmap and seqFISH+ datasets cover 1020 and 10,000 genes per cell, respectively, suggesting that SpaTalk is potentially more effective for spatial data with higher gene coverage. + +We next compared the performance of SpaTalk with that of existing cell- cell communication inference methods (Supplementary Fig. 2a). Consequently, most methods exhibited a large fraction of overlapped predictions with the rest of the methods despite the different number of inferred cell- cell communications (Supplementary Fig. 2b, c), indicating the reproducible inference across these methods. Regarding the inferred LRIs (Supplementary Fig. 2d), we reasoned that the spatial distances of the inferred LRI between sender- receiver pairs will be shorter than those between all cell- cell pairs and thus the inferred LRI will be more co- expressed in local space as cells that are close are more likely to signal (Fig. 2c). The one- sided Wilcoxon test was performed to evaluate the spatial proximity significance of the inferred LRIs, + +<--- Page Split ---> + +and the co- expressed percentage of the LRI was calculated as the co- expression level using the cell- cell graph network. Although most LRIs inferred by other methods showed significantly closer spatial distances between sender- receiver pairs than that between all cell- cell pairs, superior performance of SpaTalk was observed, ranking first for both evaluation indices for STARmap datasets (Fig. 2d). Similarly, SpaTalk obtained a higher median \(- \log_{10}P\) value and co- expression percent on the seqFISH+ OB and SVZ datasets but not for SpaOTsc (Fig. 2e). + +![](images/Figure_2.jpg) + +
Fig. 2 Superior performance of SpaTalk over existing methods. a Schematic diagram for generating simulated spot data. Cells were split according to the fixed spatial distance and then merged for the single-cell ST data with known cell types. b Performance comparison of SpaTalk with other existing cell-type deconvolution methods (RCTD, Seurat, SPOTlight, deconvSeq, Stereoscope). The asterisk represents the top-ranked method for each dataset. NA, not available. c Schematic illustration of the procedure and rationale for single-cell ST data to evaluate predicted LRIs that mediate spatially resolved cell–cell communications. d and e Performance comparison of SpaTalk with existing cell–cell communication inference methods (Giotto, SpaOTsc, NicheNet, CytoTalk, and CellCall) on the STARmap and seqFISH+ datasets. The \(P\) value
+ +<--- Page Split ---> + +represents the difference of spatial distances between sender- receiver and all cell- cell pairs assessed with the Wilcoxon test. f Schematic illustration of the procedure and rationale for single- cell ST data to evaluate predicted downstream target and pathways underlying LRIs. g Performance comparison of SpaTalk on the inferred downstream targets with other methods (NicheNet, CytoTalk, and CellCall) over the STARmap and seqFISH+ datasets. The \(P\) value represents the significance of enriched pathways or biological processes from the KEGG and Reactome databases using inferred downstream targets with the Fisher exact test. + +We compared the performance of SpaTalk for inference of intra- cellular signal pathways of the receiver cell type triggered by the LRI with those of NicheNet6, CytoTalk7, and CellCall8 that also infer the downstream targets of LRIs. We reasoned that a more accurate method would be more likely to enrich the receptor- related biological processes or pathways using the inferred downstream target genes in the receiver cell type (Fig. 2f); hence, the Fisher- exact test was adopted for pathway enrichment analysis with the KEGG and Reactome databases on target genes in receivers. The target genes inferred by all methods enriched the most intra- cellular pathways or biological processes triggered by the inter- cellular LRI (Fig. 2g). Nevertheless, SpaTalk exhibited the top- ranked performance over three benchmarked datasets, exceeding other existing methods in inference of the LRT signal network. + +Identification of signal transmission among neurons and non- neuronal cells. SpaTalk was first applied to investigate and visualize the cell- cell communications over the STARmap ST dataset of the mouse visual cortex (Fig. 3a), including data of 1020 sequenced genes for 973 cells in space covering the spatial axis of excitatory neurons (eL2/3, eL4, eL5, eL6, annotated by anatomic cortical layers); Pvalb, Reln, Sst, and Vip- expressing neurons; and non- neuronal cells, including astrocytes (Astro), endothelial cells (Endo), microglia (Micro), oligodendrocytes (Oligo), and smooth muscle cells + +<--- Page Split ---> + +(SMCs) from layers L1 though L6 to the corpus callosum and hippocampus. As shown in Fig. 3b, SpaTalk identified the spatial signal transmission among excitatory and inhibitory neurons and non- neuronal cells such as Sst- Sstr2, Inhba- Acvr1c, and Cort- Sstr2. For example, the direct cell- cell communication mediated by the Sst- Sstr2 interaction was observed between Sst- expressing neurons and Pvalb- expressing neurons (Supplementary Fig. 3a), which are known to regulate neuron activity via nonsynaptic, inter- neuronal communication36, 37. Concordantly, these identified LRIs among neurons and non- neuronal cells are associated with multiple biological processes and pathways that play vital roles in the regulation of physiological neural development and the balance of excitatory/inhibitory transmission in the central nervous system38, 39, including growth hormone synthesis, secretion, and action; neuroactive ligand- receptor interaction; signaling by activin; and the transforming growth factor (TGF)- beta signaling pathway (Fig. 3c). + +Notably, Astro were relatively abundant across space of the L5 layer and colocalized with numerous excitatory neurons, exhibiting direct cell- cell communication with eL2/3, eL5, and eL6 (Supplementary Fig. 3b). Given the Cort- Sstr2 interaction between eL5 and Astro, eL5 highly expressed and secreted the CORT ligand to interact with the SSTR2 receptor on Astro with a highly overlapped distribution density of eL5 and Astro in the constrained space, wherein the spatial distances of Cort and Sstr2 in eL5- Astro pairs were significantly closer than those in all cell- cell pairs (Fig. 3d). Focusing on the intra- cellular signal pathway of the Cort- Sstr2 interaction reconstructed by SpaTalk, two downstream TFs were identified, Egr1 and Smad3, which are involved in canonical TGF- beta signaling, in line with previous findings40. Despite the relatively low- rise co- expression of target genes in receivers, the intra + +<--- Page Split ---> + +cellular signal triggered by the Cort- Sstr2 interaction successfully propagated to the target genes, reaching a high percentage of cells expressing most target genes (Fig. 3e). The Fisher exact test showed significant enrichment with receptor- related pathways (Fig. 3f), indicating the reliability of eL5- Astro communications mediated by the Cort- Sstr2 interaction. + +We then applied SpaTalk to another spot- based ST dataset of the mouse cortex (Slide- seq v2), including 17,545 unique genes among 42,550 spots in space, reaching up to 4000 expressed genes per spot. Leveraging previously published adult mouse cortical cell taxonomy by scRNA- seq data41 (Supplementary Fig. 3c), SpaTalk reconstructed the spatial transcriptomics atlas at the single- cell resolution for Slide- seq data, which showed consistent spatial localization of neurons and non- neuronal cells across different layers, including the major excitatory and inhibitory neurons and Oligo (Fig. 3g). Compared to STARmap data, oligodendrocyte progenitor cells, and Igtp-, Ndnf-, and Smad3- expressing neurons were also observed in the Slide- seq data. Concordantly, most cell- cell communications in STARmap data were also found in Slide- seq data except for those of distinct cell types in the two datasets (Supplementary Fig. 3d). For example, the eL5- Astro communications mediated by the Cort- Sstr2 interaction in space and the downstream targets such as SMAD3 were also observed in Slide- seq data (Fig. 3h), suggesting the universality of the spatially resolved cell- cell communications inferred by SpaTalk. In addition, the direct communications among Pvalb neurons, eL6, and Oligo were also significantly enriched, in accordance with the fact that neuron- oligo communication controls the oligodendrocyte function and myelin biogenesis (Supplementary Fig. 3e). + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Identification of spatial inhibitory signal transmission among neurons and non-neuronal cells. a STARmap single-cell ST dataset of the mouse visual cortex involving 973 cells and 1020 genes. Astro, astrocytes; eL2/3, eL4, eL5, eL6, excitatory neuron subtypes; Endo, endothelial cells; HPC, hippocampus; Micro, microglia; Oligo, oligodendrocytes; SMC, smooth muscle cells; cc, corpus callosum. b Significantly enriched LRIs that mediate cell–cell communications among neurons and non-neuronal cells inferred by SpaTalk with \(P< 0.05\) . The \(P\) value represents the significance of spatial proximity of LRIs using the permutation test. c Sanky plot of the associations among ligands, receptors, and biological processes or pathways in the KEGG and Reactome databases that mediate cell–cell communications in the central nervous system. d Spatial distribution and intra-cellular signaling pathways of the Cort–Sstr2 pairs between the eL5 senders and Astro receivers. \(P\) values were calculated with the Wilcoxon test. e Co-expression of target genes in receivers and the percentage of expressed cells for target genes. f Significantly enriched biological processes and pathways with the ligand-receptor-target genes using the Fisher exact test. g Slide-seq spot-based ST dataset of the mouse visual cortex involving 42,550 spots and 22,542 genes. OPC, oligodendrocyte progenitor cell. h Communications of eL5–Astro mediated by the Cort–Sstr2 interaction in space and the intra-cellular signal pathway
+ +<--- Page Split ---> + +inferred by SpaTalk over Slide- seq data. + +Metabolic modulation of periportal hepatocytes on pericentral hepatocytes. We applied SpaTalk to the Slide- seq ST dataset of the mouse liver covering 17,545 unique genes among 25,595 spots in space (Fig. 4a). To explore the cell- type composition of Slide- seq data, a mouse liver scRNA- seq reference integrating the non- parenchymal cells from the Mouse Cell Atlas (MCA) \(^{42}\) and the parenchymal hepatic cells from the GSE125688 dataset \(^{43}\) were utilized (Supplementary Fig. 4a), containing 6029 cells, including major immune cells such as macrophages (Macro), and the pericentral and periportal hepatocytes. The reconstructed single- cell ST atlas was perfectly accordant with the original outcome obtained by Slide- seq (Fig. 4b), wherein the expression of known marker genes \(^{44}\) and the percent for each cell type were highly correlated across spots (Supplementary Fig. 4b, c, d), such as the pericentrally and periportally zonated genes Cyp2e1 and Pck1 (Fig. 4c). Immune cells were hardly observed in each spot, with pericentral and periportal hepatocytes accounting for the major proportion across spots (Supplementary Fig. 4e); the same phenomenon was observed in recently published ST data of the healthy liver \(^{45}\) . + +The cell- cell communications between pericentral and periportal hepatocytes were further explored by SpaTalk (Fig. 4d). Both hepatocyte types secrete and receive multiple ligands for their communication, forming spatially distributed metabolic cascades to cooperatively optimize the metabolic environment. For instance, with the gradient expression of enzymes in sequential lobule layers, pericentral hepatocytes perform the primary steroid, alcohol, and lipid metabolic processes, while periportal hepatocyte mainly carry out the small- molecule and monosaccharide biosynthetic processes, amino acid and triglyceride metabolic processes, and gluconeogenesis (Fig. + +<--- Page Split ---> + +4e, f), in line with the variable functions of zonated hepatocytes residing in the central and portal veins46. Notably, the periportal hepatocytes substantially expressed more ligands, including epidermal growth factor (Egf), transforming growth factor alpha (Tgfa), heparin-binding EGF-like growth factor (Hbegf), insulin-like growth factor 1 (Igf1), and vascular endothelial growth factor A (Vegfa), to promote the growth of pericentral hepatocytes. As the blood flows from the portal vein toward the central vein, this could reflect the fact that periportal hepatocytes respire most of the oxygen, leading to decreased oxygen concentrations along the lobule axis; thus, periportal hepatocytes secrete numerous growth factors via paracrine signaling to modulate the pericentral hepatocytes for the prevention of hypoxia and the maintenance and amelioration of pericentrally metabolic functions46. + +Taking the LRI of Apob- Cd36 as an example, the spatially resolved cell- cell communications between periportal and pericentral hepatocytes mainly occurred across the mid-lobule layers (Fig. 4g). The gene product of Apob is an apolipoprotein of chylomicrons and low- density lipoproteins highly involved with the regulation of lipids and fatty acids metabolism through CD36, indicating the modulation of periportal hepatocytes on the metabolic microenvironment sensed by pericentral hepatocytes. From the reconstructed intra- cellular signal propagation network triggered by the Apob- Cd36 interaction (Fig. 4h), sequential target TFs were activated, including Ahr that regulates xenobiotic- metabolizing enzymes such as cytochrome P450, and Nr1h4 that regulates the expression of genes involved in bile acid synthesis and transport, in agreement with the corresponding module score of pericentral hepatocytes in space (Fig. 4i). The LRT network was also remarkably enriched in the AMPK and PPAR signaling pathways, which play crucial roles in the regulation of energy + +<--- Page Split ---> + +334 and metabolic homeostasis, suggesting the spatially fine- tuned cell- cell communications along the portal- central lobule axis for minimizing risks to pericentral hepatocytes. + +![](images/Figure_4.jpg) + +
Fig. 4 Modulation of periportal hepatocytes on the metabolic microenvironment sensed by pericentral hepatocytes. a Slide-seq spot-based ST dataset of the mouse liver involving 25,595 spots and 17,545 genes. b Cell-type decomposition by SpaTalk. PC, pericentral; PP, periportal; Hep, hepatocytes. c Scaled Pearson's correlation coefficients between the expression of known marker genes and the percent for each cell type. DC, dendritic cell; Macro, macrophages. d Enriched LRIs that mediate cell–cell communications between pericentral and periportal hepatocytes. e Significant differentially expressed genes (DEGs) between periportal and pericentral hepatocytes assessed with the Wilcoxon test and the corresponding significantly enriched biological processes and pathways determined with the Metascape web tool. Representative DEGs are labeled beside the heatmap. f Significantly activated pathways in pericentral (up) and periportal (down) hepatocytes determined by Gene Set Enrichment Analysis (GSEA). g Communications from periportal hepatocytes to
+ +<--- Page Split ---> + +pericentral hepatocytes mediated by the Apob- Cd36 interaction in space. h Intracellular signal pathway inferred by SpaTalk over Slide- seq data and the significantly enriched pathways over the ligand- receptor- target network. i Inferred module score of each hepatocyte type over the pathway signatures determined with Seurat. + +Spatial characterization of cell types over 10X Visium data. Given the widely used 10X Visium tool in ST studies, we applied SpaTalk to a human skin squamous cell carcinoma (SCC) ST dataset published by Ji et al.22, who profiled SCC and matched normal tissues via 10X scRNA- seq and used Visium to identify a tumor- specific keratinocyte (TSK) in the tumor (Fig. 5a). Using the matched SCC scRNA- seq data of patient 2 as reference, the optimal cell- type composition for each spot was deconvoluted by SpaTalk, which exhibited a similar characterization with that histologically assessed from hematoxylin and eosin- stained frozen sections (Fig. 5b). The percent of TSK inferred by our method was compared to the TSK score based on markers (e.g., MMP10, PTHLH, LAMC2, and IL24) defined by Ji et al. across 646 spots (Fig. 5c). A high correlation between the TSK percent and score was observed (Fig. 5d). Moreover, the percent of inferred cell types was prominently associated with the expression of known marker genes, indicating the accuracy of SpaTalk for cell- type decomposition (Supplementary Fig. 5a, b). + +Next, we reconstructed the single- cell ST profile by assuming a total of 30 cells in each spot according to a recent review12, which covered the main epithelial cells, including differentiating, cycling, and basal keratinocytes; melanocytes; fibroblasts (FB); Endo; natural killer (NK) cells; and T cells (Fig. 5e). Despite the asymmetrical distribution for most cell types, TSK, FB, and Endo showed specific patterns of locations in space, which were highly adjacent in some tumor areas (Fig. 5f), forming direct cell- cell communications in the tumor microenvironment (TME). By filtering cells from the TSK leading spots (score \(\geq 0.8\) ), we found that TSKs reside within a fibrovascular niche, + +<--- Page Split ---> + +resulting in high colocalization of TSKs, FB, and Endo at the TSK leading spots (Fig. 5g), in line with the previous findings22. Moreover, we used SpaTalk to investigate the cell- type composition over the ST data of another SCC patient. Despite a low percentage across 621 spatial spots, most TSKs centered on a handful of corner spots in space, exhibiting a highly consistent distribution with TSK scores (Fig. 5h and Supplementary Fig. 5c). Unsurprisingly, the fibrovascular niche was also observed in the TSK leading spots of patient 10 with clear spatial co- localization (Supplementary Fig. 5d), indicating the close cell- cell communication among TSKs, FB, and Endo in the TME underlying the occurrence and development of SCC. + +![](images/Figure_5.jpg) + +
Fig. 5 Spatial characterization of tumor and stromal cells in human squamous cell
+ +<--- Page Split ---> + +carcinoma Visium data with SpaTalk. a Visium spot- based ST dataset of human skin SCC in patient 2 with the matched scRNA- seq dataset involving the main keratinocytes (KC), stromal cells, and immune cells. b Cell- type decomposition by SpaTalk. Cyc, cycling; Diff, differentiating; NK, natural killer; FB, fibroblast. c TSK percent and TSK score across spatial spots. The expression of known TSK markers is plotted. d Pearson's correlation coefficient between the TSK percent and TSK score. f Contour plot of TSK, FB, and Endo based on the reconstructed single- cell ST atlas by SpaTalk. g TSK leading spots with a TSK score \(>0.8\) in space. The bar chart represents the number of different cell types and the line chart represents the number of neighbors adjacent to TSKs among the TSK leading spots. h Visium spot- based ST dataset of human skin SCC in patient 10 and the cell- type decomposition by SpaTalk showing the percent of TSKs across 621 spatial spots. + +Reconstruction of TSK- stroma communications in space. To dissect the underlying LRI mediating the spatially resolved cell- cell communications between TSKs and stromal cells of the fibrovascular niche in the TME, we applied SpaTalk to infer the communications between TSK- FB and TSK- Endo pairs over the decomposed single- cell ST data of SCC in patient 2, including the top- ranked 20 LRIs based on the integrated inter- cellular and intra- cellular scores (Fig. 6a). Consistent with a TSK- fibrovascular niche, prominent TSK signaling to FB and Endo was mediated by several common ligand- receptor pairs, including VEGFA- NPR1, VEGFB- NPR1, PGF- SDC1, and CDH1- ITGAE, associated with tumor angiogenesis. Additionally, TSKs modulate FB through secreting matrix metallopeptidase (MMP)1 and MMP9, which are linked to tumor metastasis via cellular movement and extracellular matrix (ECM) disassembly. Conversely, FB and Endo prominently co- expressed numerous ligands such as MDK, HGF, HMGB1, and THBS1, matching TSK receptors that promote the proliferation and differentiation of TSKs (Fig. 6b). Further supporting TSKs as an epithelial mesenchymal transition (EMT)- like population, SpaTalk predicted that the widely expressed TGFB1 regulates TSKs. TSK receptors corresponding to additional ligands from FB and Endo + +<--- Page Split ---> + +included several integrins (e.g., ITGA5 and ITGB6) and nectins (e.g., NECTIN1 and NECTIN2), highlighting other pathways associated with EMT and epithelial tumor invasion \(^{47,48}\) . + +![](images/Figure_6.jpg) + +
Fig. 6 Reconstruction of cell–cell communications between TSK subpopulations and stromal cells in space with SpaTalk. a Top 20 inferred LRIs that mediate cell–cell communication from the TSK senders to the FB and Endo receivers. Colored blocks represent the known ligand–receptor pairs in CellTalkDB. The asterisk represents the significantly enriched LRIs determined by SpaTalk. b Top 20 inferred LRIs that mediate cell–cell communication from the FB and Endo senders to TSK receivers. c Region of interest (ROI) covering 42 spatial spots in space with a high total score of TSK, FB, and Endo according to their signature genes. The point plot shows the decomposed single-cell ST atlas determined by SpaTalk. d Expression of known cancer-associated fibroblast (CAF) markers across TSKs, FB, Endo, and other cells. e Differentially expressed genes (DEGs) between EMT-like and EMT-unlike TSKs and the corresponding
+ +<--- Page Split ---> + +enriched biological processes and pathways. Representative DEGs are labeled beside the point. f Number of cell- cell pairs over the LRIs from the EMT- like and EMT- unlike TSKs to CAFs. g Communications from EMT- like and EMT- unlike TSKs to CAFs mediated by the MMP1- CD44 interaction in space. h Comparison of communications among CAFs, Endo, EMT- like TSKs, and EMT- unlike TSKs with respect to the number of cell- cell pairs in space over the matched LRIs evaluated with a paired t test. + +Next, we focused on a region of interest (ROI) covering 42 spatial spots in space for in- depth exploration of TSK- stroma communications, which exhibited a high total score of TSK, FB, and Endo according to their signature genes, occupying the major part in the ROI (Fig. 6c). By mapping cancer- associated fibroblast (CAF) markers to the cells in space, the majority of FB in the ROI highly expressed the known CAF marker genes (e.g., VIM, FAP, POSTIN, and SPARC) (Fig. 6d), hinting at the transformation of FB to CAFs induced by the adjacent TSKs and conversely supporting the stemness of TSKs via direct cell- cell communication in space (Supplementary Fig. 6a). Notably, the TSKs appear to be extremely heterogeneous with respect to the broad range of EMT scores in the ROI; thus, TSKs were further classified into 268 EMT- like and 268 EMT- unlike populations (Supplementary Fig. 6b). By comparing their differentially expressed genes, EMT- like TSKs were dramatically enriched with ECM organization, proteoglycans in cancer, regulation of cell adhesion, and the VEGFA- VEGFR2 signaling pathway, representing more invasive properties compared with EMT- unlike TSKs (Fig. 6e). + +Additionally, EMT- like TSKs appear to be more communicative with surrounding CAFs in the TME in light of the greater number of cell- cell pairs over the LRIs that prominently mediate TSK- stroma communications in space, such as LAMB3- ITGB1, LAMA3- ITGB1, LAMC2- ITGB1, and MMP1- CD44 (Fig. 6f and Supplementary Fig. 6c, d, e). Metastasis- related laminins are essential for formation and function of the + +<--- Page Split ---> + +basement membrane, whereas MMPs are involved in ECM breakdown, both contributing to the aggravated malignancy of tumors. Moreover, CD44 expression on CAFs plays a supporting role in the induction of cellular stemness, wherein CAFs have a preference of cell- cell communications with EMT- like TSKs in space (Fig. 6g). Interestingly, EMT- unlike TSKs notably exhibited more communicative cell- cell pairs with Endo, whereas EMT- like TSKs exhibited significantly more communicative cell- cell pairs with CAFs over the matched LRIs (Fig. 6h), consistent with the observed contribution of CAFs to EMT in a broad range of tumors49, 50. + +## Discussion + +We have demonstrated the capabilities of SpaTalk to infer and visualize spatially resolved cell- cell communications mediated by significantly enriched LRIs under normal and disease states over existing representative datasets, including the single- cell ST data generated from STARmap, MERFISH, and seqFISH+, and the spot- based ST data obtained via Slide- seq and 10X Visium. + +There are two principles to decode the mechanisms of cell- cell communications: ligand- receptor proximity and LRT co- expression12. In a given tissue niche, cells are more likely to communicate with each other when they are spatially adjacent and activate downstream target genes in the receiving cell triggered by the LRI in proximal cells; thus, ST data are well suited to apply the two principles for inferring inter- cellular communications. Accordingly, our proposed SpaTalk realizes the integration of these two principles by incorporating the KNN and cell- cell graph network to filter spatially proximal cell pairs and corresponding LRIs, followed by utilizing the knowledge graph algorithm to model the LRT signal propagation process. Consequently, the + +<--- Page Split ---> + +performance of SpaTalk was superior to that of other methods over the benchmarked ST datasets with respect to several evaluation indices, demonstrating the reliability of the two principles in decoding cellular cross- talk, especially for juxtracrine and paracrine communication. + +Importantly, SpaTalk is applicable to either single- cell or spot- based ST datasets generated from mainstream ST technologies. For the former, SpaTalk assigns a label to each cell by selecting similar cell types with the top- ranked weight via NNLM for single- cell ST data, generating the ST atlas at single- cell resolution with known cell types for the subsequent inference of cell- cell communications. For spot- based ST data, SpaTalk selects and maps the optimal combination of cells in accordance with the decomposed optimal weight/percent of cell types via NNLM and the transcriptome profiles of spatial spots to reconstruct the ST atlas at single- cell resolution with known cell types. Notably, applications of SpaTalk to the mouse cortex and liver datasets sequenced by STARmap and Slide- seq, respectively, revealed the evidential LRIs in space that mediate the spatially resolved cell- cell communications contributing to normal physiological processes. Moreover, exploration of SpaTalk on the human skin SCC dataset obtained from 10X Visium identified the variable preference of communication among tumor subpopulations, CAF, and Endo. These cases convincingly demonstrate the universality of SpaTalk in decoding the mechanism of cell- cell communications in space underlying normal and disease tissues for single- cell and spot- based ST data. + +As unmatched scRNA- seq and ST data would directly influence the cell- type decomposition, an important feature of SpaTalk is the ability to assign a spot/cell into an unsure category considering the unseen cell types in the scRNA- seq reference. For + +<--- Page Split ---> + +example, the scRNA- seq reference for the Slide- seq mouse cortex ST data was obtained from another repository for the same tissue, resulting in numerous unsure cells by assuming one cell in each spot in terms of the high resolution (10 μm) of Slide- seq technology that almost approaches single- cell resolution. Additionally, the extremely low gene coverage of several spatial spots severely affects the regression model, which were regarded as the unsure type by SpaTalk. However, with respect to the human skin SCC datasets, SpaTalk removed the unsure type for the matched scRNA- seq and ST data. As the matched multi- modal datasets will undoubtedly become greater in number, application of SpaTalk and similar methods will be required for accurate inference of spatially resolved cell- cell communications. + +Additionally, SpaTalk characterizes the spatial distribution for each cell type within the reconstructed ST at single- cell resolution through the contour plot of cellular density in space, which enables analyzing the proximal relationship between paired cell types. Moreover, SpaTalk enables the statistical analyses and visualization of spatially proximal LRIs in space, forming a dynamic cell- cell communication network. Currently, it is hard to analyze and visualize the LRI at single- cell resolution for scRNA- seq data, wherein the common practice is to interpret the LRI for paired cell types. By incorporating spatial information, SpaTalk displays the enriched LRI at single- cell resolution via the spatially proximal co- expressed cell pairs, offering an informatively brand- new approach for the analysis and visualization of the LRI and its mediated cell- cell communication underlying the disease pathology from a novel perspective, as shown in the application of SpaTalk to the human skin SCC datasets. + +Adding spatial constraints in cell- cell communication inference is critical to the spatial analysis of juxtracrine and paracrine communications. However, this constraint + +<--- Page Split ---> + +inevitably causes the failure of inferring long- range communications such as endocrine and telecrine signaling. Classification of LRIs into short- range and long- range communications with prior knowledge might be helpful to infer the comprehensive communication categories computationally. Moreover, it is potentially beneficial to include other omics data with the increasing multi- modal datasets generated from state- of- the- art technologies such as 10X Multioome and Digital Spatial Profiling51 in studying spatially regulated cell- cell communications. Thus, more reliable computational models might be needed for more accurate integration of multi- modal data and inference. + +## Methods + +Datasets. For STARmap, the single- cell ST data of the mouse cortex (20180410- BY3_1kgenes) was obtained from a public data portal (https://www.dropbox.com/sh/f7ebheru1lbz91s/AABYSSjSTppBmVmVWl2H4sKa?dl=0). For MERFISH, the single- cell ST data of the naïve female mouse (Animal_ID: 1, Bregma: 0.26) hypothalamic preoptic region was downloaded from Dryad (https://datadryad.org/stash/dataset/doi:10.5061/dryad.8t8s248). For seqFISH+, the single- cell ST data of the mouse cortex and olfactory bulb were retrieved from the Github repository (https://github.com/CaiGroup/seqFISH-PLUS). For Slide-seq, the spot- based ST data of the mouse liver (Puck_180803_8) and somatosensory cortex (Puck_200306_03) were obtained from the Broad Institute Single Cell Portal (https://singlecell.broadinstitute.org/single_cell/study/SCP354/slide-seq-study) and (https://singlecell.broadinstitute.org/single_cell/study/SCP815/highly-sensitive- spatial-transcriptomics-at-near-cellular-resolution-with-slide-seqv2), respectively. For + +<--- Page Split ---> + +10X Visium, the spot- based ST data and scRNA- seq data of human SCC were downloaded from the Gene Expression Omnibus (GEO) repository (GSE144240). The mouse liver scRNA- seq data of non- parenchymal and parenchymal hepatic cells were refined from the MCA (https://figshare.com/articles/MCA_DGE_Data/5435866) and GSE125688, respectively. The mouse cortex scRNA- seq data were obtained from GSE71585. + +Data processing. For the liver scRNA- seq datasets, the non- parenchymal cells and parenchymal hepatic cells were collected from the MCA and GSE125688, respectively, wherein hepatocytes were classified into pericentral and periportal hepatocytes with principal component analyses and clustering analysis. For the MERFISH dataset, ependymal cells were excluded due to the limited cell number in the section (<2). For other datasets, all cells were included in the filtered matrices. Human and mouse gene symbols were revised in accordance with NCBI gene data (https://www.ncbi.nlm.nih.gov/gene/) updated on June 30, 2021, wherein unmatched genes and duplicated genes were removed. For all ST and scRNA- seq datasets, the raw counts were normalized via the global- scaling normalization method LogNormalize in preparation for running the subsequent scDeepSort pipeline. + +SpaTalk algorithm. The SpaTalk model consists of two components: cell- type decomposition and spatial LRI enrichment. The first component is to infer cell- type composition for single- cell or spot- based ST data, and the second component is to infer spatially proximal ligand- receptor interactions that mediate cell- cell communications in space. + +Cell- type decomposition. To dissect the cell- type composition for the ST data matrix \(T[n \times s](n\) genes and \(s\) spots/cells), NNLM was first applied to obtain the + +<--- Page Split ---> + +optimal proportion of cell types using the scRNA- seq data matrix \(S[n \times c]\) ( \(n\) genes and \(c\) cells) as the reference with \(k\) cell types. Let \(Y = \{y_1, y_2, \dots , y_n\}\) be the expression profile for each spot/cell to establish the following linear model: + +\[Y = X\beta +\epsilon\] + +where \(X = [n \times k]\) is the average expression profile generated from \(S\) and \(\epsilon\) represents random error. Mean relative entropy loss was then used to measure the difference between the predicted and observed values. Therefore, the objective function can be written as: + +\[argmin\{\beta \geq 0\} L(Y - X\beta) + \lambda_1R_1(\beta) + \lambda_\alpha R_\alpha (\beta) + \lambda_2R_2(\beta)\] + +where \(R_1\) , \(R_\alpha\) and \(R_2\) represent the L1, angle, and L2 regularization with non- negative \(\lambda_1\) , \(\lambda_\alpha\) and \(\lambda_2\) initialized by zero \(^{23, 24}\) . The model was trained with the above objective function using Lee's multiplicative iteration algorithm \(^{25}\) with default hyperparameters until convergence or after 10,000 iterations to generate the coefficient matrix \(C[k \times s]\) . + +For single- cell ST data, the cell type with the maximum coefficient was assigned to each cell. For spot- based ST data, let \(M\) be the maximum cell number for each spot, which was set to 30 for 10X Visium data and was set to 1 for Slide- seq data according to a recent review. In practice, the optimal cellular combination \(\omega\) for each spot was determined by the following function: + +\[\omega_{i}(i \in \{1,2,\dots,k\}) = \left\{ \begin{array}{ll} [M\beta_{i}] + 1 & ( \{M\beta_{i}\} \geq 0.5) \\ [M\beta_{i}] & ( \{M\beta_{i}\} < 0.5) \end{array} \right.\] + +wherein \([M\beta_{i}]\) and \(\{M\beta_{i}\}\) represent the integer and fractional parts of \(M\beta_{i}\) , respectively. For each spot, we randomly selected \(m\) \((m = \sum_{i = 1}^{k} \omega_{i})\) cells from \(S\) to compare their merged expression profile \(\epsilon\) with the ground truth according to the + +<--- Page Split ---> + +following function: + +\[argmin\{m\leq M\} \sum_{i = 1}^{n}(Y_{i} - \sum_{j = 1}^{m}\epsilon_{i}^{j})^{2}\] + +To assign a coordinate \((x, y)\) to each sampled cell, we applied stochastic \(\alpha \in [0, 1]\) and \(\theta \in [0, 360]\) in spot \((x_0, y_0)\) to locate the cell into the space detailed in the following function: + +\[x = x_0 + \alpha d_{min}\cos (\theta \pi /180)\] \[y = y_0 + \alpha d_{min}\sin (\theta \pi /180)\] + +where \(d_{min}\) represents the spatial distance of the closest neighbor spot. By integrating the optimal cellular combinations for all spots, ST data at single- cell resolution were reconstructed for the spot- based ST data. + +Spatial LRI enrichment. To generate the cell- cell distance matrix \(D\) , the Euclidean distance between cells was calculated using the single- cell spatial coordinates of ST data. The KNN algorithm was then applied to each cell to select the \(K\) nearest cells from \(D\) to construct the cell graph network. For ligand \(i\) of the sender (cell type A) and receptor \(j\) of the receiver (cell type B), the number of LRI pairs \((P_{Ai,Bj}^{0})\) was obtained from the graph network by counting the 1- hop neighbor nodes of receivers for each sender. The permutation test was then performed by randomly shuffling cell labels to recalculate the number of LRI pairs. By repeating this step Z times, a background distribution \(P = \{P_{Ai,Bj}^{1}, P_{Ai,Bj}^{2}, \dots , P_{Ai,Bj}^{Z}\}\) was obtained for comparison with the real interacting score, and the \(P\) value was calculated as follows: + +\[P_{Ai,Bj} = \text{card}\{x \in P \mid x \geq P_{Ai,Bj}^{0}\} /Z\] + +where \(P_{Ai,Bj}\) values less than 0.05 were filtered to calculate the intercellular score of LRI from senders to receivers \((S_{Ai,Bj}^{inter} = 1 - P_{Ai,Bj})\) . To further enrich the LRIs that activate downstream TFs, target genes, and the related pathways of receivers, the + +<--- Page Split ---> + +knowledge graph was introduced to model the intracellular signal propagation process. In practice, LRIs from CellTalkDB, pathways from KEGG and Reactome, and TFs from AnimalTFDB were integrated to construct the ligand- receptor- TF knowledge graph (LRT- KG), wherein the weight between entities represents the co- expressed coefficient. Taking the receptor as the query node, we incorporated the random- walk algorithm into the LRT- KG to filter and score the downstream activated \(t\) TFs with no more than 10 steps and Z iterations; thus, the probability \(p\) for each TF can be calculated with the ratio of successful hits from the query node to the target TF during the Z random walks. By integrating the co- expressed TFs and the corresponding target genes from the LRT- KG, the intracellular score of LRI from senders to receivers can be written as: + +\[S_{A i,B j}^{i n t r a} = \sum_{k = 1}^{t}\theta_{k}\times p_{k} / \eta_{k}\] + +where \(\theta\) represents the number of targeted genes, \(\eta\) represents the step from the receptor to the TF in the LRT- KG. By the sigmoid transformation for \(S_{A i,B j}^{i n t r a}\) , the final score of the LRI from cell type A to cell type B can be written as: + +\[S_{A i,B j} = \sqrt{S_{A i,B j}^{i n t e r}\times S_{A i,B j}^{i n t r a}}\] + +Comparison with other methods. STARmap, MERFISH, and seqFISH+ ST data were used to compare the performance of SpaTalk with other existing cell- type decomposition methods. For these single- cell ST data, all cells were split according to the fixed spatial distance and then merged into simulated spots as the benchmark datasets. RCTD, SPOTlight, Seurat, deconvSeq, and Stereoscope were benchmarked with the default parameters and evaluated with Pearson's correlation coefficient and RMSE over the predicted and real cell- type composition for each spot. + +Given the limited genes of MERFISH ST data, STARmap and seqFISH+ single- cell ST + +<--- Page Split ---> + +data (including 1020 and 10,000 genes, respectively) were used as the benchmark datasets to compare the performance of SpaTalk with other cell- cell communication inference methods (Giotto, SpaOTsc, NicheNet, CytoTalk, and CellCall). The one- sided Wilcoxon test was performed to evaluate the spatial proximity significance of the inferred LRIs by comparing the number of expressed LRIs between sender- receiver pairs and all cell- cell pairs, and the co- expressed percent of the LRI was calculated to evaluate the co- expression level by counting the number of expressed LRIs from senders to receivers from the cell- cell graph network. All methods were benchmarked with the default parameters and all inferred LRIs were unbiasedly evaluated with the above criteria except for the LRIs from SpaOTsc since the number of inferred LRIs was much larger than that of the other methods; thus, the top 1000 LRIs for each cell- cell communication were selected from SpaOTsc according to the final score. Given the significantly enriched biological processes or pathways in the receiver cell type, the Fisher exact test was adopted for pathway enrichment analysis with the KEGG and Reactome databases on the activated genes in receivers using the following function: + +\[P = \left(\frac{a + b}{a}\right)\left(\frac{c + d}{c}\right) / \left(\frac{n}{a + c}\right)\] + +where \(n = a + b + c + d\) ; \(a\) is the number of inferred target genes that match a given pathway, \(b\) is the number of given pathway genes that exclude \(a\) , \(c\) is the number of inferred target genes that unmatch a given pathway, and \(d\) is the number of all genes excluding \(a\) , \(b\) , and \(c\) . NicheNet, CytoTalk, and CellCall were benchmarked with the default parameters and the inferred target genes for each LRI were evaluated according to the significance of pathway enrichment. + +Pathway and biological process enrichment. The Metascape web tool + +<--- Page Split ---> + +(https://metascape.org/) was used to perform the enrichment analysis of pathways and biological processes, wherein the top 100 highly expressed genes were selected according to the fold change of the average gene expression. Gene Set Enrichment Analysis (GSEA) was performed using the ranked gene list with the clusterprofiler tool to enrich the significantly activated pathways and biological processes, whose signatures were obtained from the Molecular Signatures Database v7.4 (MSigDB, http://www.gsea- msigdb.org/gsea/msigdb), including the gene sets from Gene Ontology (GO) and the canonical pathway gene sets derived from the KEGG, Reactome, and WikiPathways pathway databases. + +Module scoring of hallmarks and signatures. Hallmark scoring of metabolism of xenobiotics by cytochrome P450, synthesis of bile acids and bile salts, TSK, and EMT was performed using the "AddModuleScore" function in Seurat with default parameters. Hallmark pathways and EMT were obtained from MSigDB, and the signature genes of the TSK were download from the original publication by Jin et al. Statistics. R (version 4.1.1) and GraphPad Prism 8 were used for all statistical analyses. + +## Data and code availability + +No new data was generated for this study. All data used in this study is publicly available as previously described. Source codes for the SpaTalk R package and the related scripts are available at github (https://github.com/ZJUFanLab/SpaTalk). + +## Acknowledgements + +This work was supported by the National Natural Science Foundation of China (81973701), the Key Program, National Natural Science Foundation of China + +<--- Page Split ---> + +(81930016), the Key Research and Development Program of China (2021YFA1100500), the Major Research Plan of the National Natural Science Foundation of China (No.92159202), the Natural Science Foundation of Zhejiang Province (LZ20H290002), the China Postdoctoral Science Foundation (2021M702828) and Westlake Laboratory (Westlake Laboratory of Life Sciences and Biomedicine). + +## Author contributions + +X.F., X.X., and H.C. conceived and designed the study. C.L., X.L., J.L., J.Q., and K.W. collected and analyzed the scRNA-seq and ST data. X.S., H.Y., J.C, and P.Y. implemented the algorithm of SpaTalk. X.S., C.L., and H.Y. developed the package of SpaTalk. All authors wrote the manuscript, read and approved the final manuscript. + +## Competing interests + +The authors declare no competing interests. + +## Reference + +1. Shao X, Lu X, Liao J, Chen H, Fan X. New avenues for systematically inferring cell-cell communication: through single-cell transcriptomics data. Protein Cell 11, 866-880 (2020). +2. Armingol E, Officer A, Harismendy O, Lewis NE. Deciphering cell-cell interactions and communication from gene expression. Nat Rev Genet 22, 71-88 (2021). +3. Shao X, et al. scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network. Nucleic Acids Res 49, e122 (2021). +4. Cabello-Aguilar S, Alame M, Kon-Sun-Tack F, Fau C, Lacroix M, Colinge J. SingleCellSignalR: inference of intercellular networks from single-cell transcriptomics. Nucleic Acids Res 48, e55 (2020). +5. Jung S, Singh K, Del Sol A. FunRes: resolving tissue-specific functional cell states based on a cell-cell communication network model. Brief Bioinform 22, (2021). +6. Browaeys R, Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods 17, 159-162 (2020). +7. Hu Y, Peng T, Gao L, Tan K. CytoTalk: De novo construction of signal transduction networks using single-cell transcriptomic data. Sci Adv 7, (2021). + +<--- Page Split ---> + +8. Zhang Y, et al. CellCall: integrating paired ligand-receptor and transcription factor activities for cell-cell communication. Nucleic Acids Res 49, 8520-8534 (2021).9. Zhang M, et al. Spatial molecular profiling: platforms, applications and analysis tools. Brief Bioinform 22, (2021).10. Lein E, Borm LE, Linnarsson S. The promise of spatial transcriptomics for neuroscience in the era of molecular cell typing. Science 358, 64-69 (2017).11. Liao J, Lu X, Shao X, Zhu L, Fan X. Uncovering an Organ's Molecular Architecture at Single-Cell Resolution by Spatially Resolved Transcriptomics. Trends Biotechnol 39, 43-58 (2021).12. Longo SK, Guo MG, Ji AL, Khavari PA. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat Rev Genet 22, 627-644 (2021).13. Moss A, et al. A single cell transcriptomics map of paracrine networks in the intrinsic cardiac nervous system. iScience 24, 102713 (2021).14. Stur E, et al. Spatially resolved transcriptomics of high-grade serous ovarian carcinoma. iScience 25, 103923 (2022).15. Almet AA, Cang Z, Jin S, Nie Q. The landscape of cell-cell communication through single-cell transcriptomics. Curr Opin Syst Biol 26, 12-23 (2021).16. Dries R, et al. Giotto: a toolbox for integrative analysis and visualization of spatial expression data. Genome Biol 22, 78 (2021).17. Cang Z, Nie Q. Inferring spatial and signaling relationships between cells from single cell transcriptomic data. Nat Commun 11, 2084 (2020).18. Mohamed SK, Nounu A, Novacek V. Biological applications of knowledge graph embedding models. Brief Bioinform 22, 1679-1693 (2021).19. Wang X, et al. Three-dimensional intact-tissue sequencing of single-cell transcriptional states. Science 361, (2018).20. Rodrigues SG, et al. Slide-seq: A scalable technology for measuring genome-wide expression at high spatial resolution. Science 363, 1463-1467 (2019).21. Stickels RR, et al. Highly sensitive spatial transcriptomics at near-cellular resolution with Slide-seqV2. Nat Biotechnol 39, 313-319 (2021).22. Ji AL, et al. Multimodal Analysis of Composition and Spatial Architecture in Human Squamous Cell Carcinoma. Cell 182, 497-514 e422 (2020).23. Pascual-Montano A, Carazo JM, Kochi K, Lehmann D, Pascual-Marqui RD. Nonsmooth nonnegative matrix factorization (nsNMF). IEEE Trans Pattern Anal Mach Intell 28, 403-415 (2006).24. Franc V, Hlaváč V, Navara M. Sequential Coordinate-Wise Algorithm for the Non-negative Least Squares Problem. In: Computer Analysis of Images and Patterns (eds Gagalowicz A, Philips W). Springer Berlin Heidelberg (2005).25. Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature 401, 788-791 (1999).26. Shao X, Liao J, Li C, Lu X, Cheng J, Fan X. CellTalkDB: a manually curated database of ligand-receptor interactions in humans and mice. Brief Bioinform 22, (2021).27. Hu H, Miao YR, Jia LH, Yu QY, Zhang Q, Guo AY. AnimalTFDB 3.0: a comprehensive resource for annotation and prediction of animal transcription factors. Nucleic Acids Res 47, D33-D38 (2019).28. Lee B, Zhang S, Poleksic A, Xie L. Heterogeneous Multi-Layered Network Model for Omics + +<--- Page Split ---> + +Data Integration and Analysis. Front Genet 10, 1381 (2019). + +Moffitt JR, et al. Molecular, spatial, and functional single- cell profiling of the hypothalamic preoptic region. Science 362, (2018). + +Eng CL, et al. Transcriptome- scale super- resolved imaging in tissues by RNA seqFISH. Nature 568, 235- 239 (2019). + +Cable DM, et al. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol, (2021). + +Stuart T, et al. Comprehensive Integration of Single- Cell Data. Cell 177, 1888- 1902 e1821 (2019). + +Elosua- Bayes M, Nieto P, Mereu E, Gut I, Heyn H. SPOTlight: seeded NMF regression to deconvolute spatial transcriptomics spots with single- cell transcriptomes. Nucleic Acids Res 49, e50 (2021). + +Du R, Carey V, Weiss ST. deconvSeq: deconvolution of cell mixture distribution in sequencing data. Bioinformatics 35, 5095- 5102 (2019). + +Andersson A, et al. Single- cell and spatial transcriptomics enables probabilistic inference of cell type topography. Commun Biol 3, 565 (2020). + +Tasic B, et al. Shared and distinct transcriptomic cell types across neocortical areas. Nature 563, 72- 78 (2018). + +Dippoppa M, Ranson A, Krumin M, Pachitariu M, Carandini M, Harris KD. Vision and Locomotion Shape the Interactions between Neuron Types in Mouse Visual Cortex. Neuron 98, 602- 615 e608 (2018). + +Krieglstein K, Zheng F, Unsicker K, Alzheimer C. More than being protective: functional roles for TGF- beta/activin signaling pathways at central synapses. Trends Neurosci 34, 421- 429 (2011). + +Allen NJ, Lyons DA. Glia as architects of central nervous system formation and function. Science 362, 181- 185 (2018). + +Sun M, Gewirtz JC, Bofenkamp L, Wickham RJ, Ge H, O'Connor MB. Canonical TGF- beta signaling is required for the balance of excitatory/inhibitory transmission within the hippocampus and prepulse inhibition of acoustic startle. J Neurosci 30, 6025- 6035 (2010). + +Nat Neurosci 19, 335- 346 (2016). + +Han X, et al. Mapping the Mouse Cell Atlas by Microwell- Seq. Cell 172, 1091- 1107 e1017 (2018). + +Pepe- Mooney BJ, et al. Single- Cell Analysis of the Liver Epithelium Reveals Dynamic Heterogeneity and an Essential Role for YAP in Homeostasis and Regeneration. Cell Stem Cell 25, 23- 38 e28 (2019). + +Shao X, Liao J, Lu X, Xue R, Ai N, Fan X. ssCATCH: Automatic Annotation on Cell Types of Clusters from Single- Cell RNA Sequencing Data. iScience 23, 100882 (2020). + +Yuan Z, et al. SEAM is a spatial single nuclear metabolomics method for dissecting tissue microenvironment. Nat Methods 18, 1223- 1232 (2021). + +Ben- Moshe S, Itzkovitz S. Spatial heterogeneity in the mammalian liver. Nat Rev Gastroenterol Hepatol 16, 395- 410 (2019). + +Margadant C, Sonnenberg A. Integrin- TGF- beta crosstalk in fibrosis, cancer and wound healing. EMBO Rep 11, 97- 105 (2010). + +<--- Page Split ---> + +805 48. Chen CH, Chen WY, Lin SF, Wong RJ. Epithelial-mesenchymal transition enhances response 806 to oncolytic herpesviral therapy through nectin-1. Hum Gene Ther 25, 539-551 (2014). 807 49. Wen S, et al. Cancer-associated fibroblast (CAF)-derived IL32 promotes breast cancer cell 808 invasion and metastasis via integrin beta3-p38 MAPK signalling. Cancer Lett 442, 320-332 809 (2019). 810 50. Asif PJ, Longobardi C, Hahne M, Medema JP. The Role of Cancer-Associated Fibroblasts in 811 Cancer Invasion and Metastasis. Cancers (Basel) 13, (2021). 812 51. Merritt CR, et al. Multiplex digital spatial profiling of proteins and RNA in fixed tissue. Nat 813 Biotechnol 38, 586-599 (2020). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryFigures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5_det.mmd b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..adfac82fd0a2f6f0db29c867f995bc2eae4f51d0 --- /dev/null +++ b/preprint/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5/preprint__0dc4159524178aa2b8818e1cd5a2fb769cdeba272c79bb5ec67f5f1a67e0d9b5_det.mmd @@ -0,0 +1,467 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 935, 210]]<|/det|> +# Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk + +<|ref|>text<|/ref|><|det|>[[42, 230, 583, 784]]<|/det|> +Xin Shao Zhejiang University https://orcid.org/0000- 0002- 1928- 3878 Chengyu Li Zhejiang University https://orcid.org/0000- 0003- 3144- 9460 Haihong Yang Zhejiang University Xiaoyan Lu Zhejiang University Jie Liao Zhejiang University https://orcid.org/0000- 0002- 6697- 8998 Jingyang Qian Zhejiang University Kai Wang Zhejiang University Junyun Cheng Zhejiang University Penghui Yang Zhejiang University Huajun Chen Zhejiang University Xiao Xu Zhejiang University Xiaohui Fan ( \(\boxed{\pmb{\times}}\) fanxh@zju.edu.cn ) Zhejiang University https://orcid.org/0000- 0002- 6336- 3007 + +<|ref|>sub_title<|/ref|><|det|>[[44, 825, 101, 842]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 862, 135, 880]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 898, 303, 917]]<|/det|> +Posted Date: April 22nd, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 936, 473, 955]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1576678/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 909, 87]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 123, 907, 167]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 30th, 2022. See the published version at https://doi.org/10.1038/s41467-022-32111-8. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[103, 84, 850, 142]]<|/det|> +# Knowledge-graph-based cell-cell communication inference for spatially resolved transcriptomic data with SpaTalk + +<|ref|>text<|/ref|><|det|>[[100, 150, 852, 800]]<|/det|> +Xin Shao1, 3, †, Chengyu Li3, †, Haihong Yang2, 4, †, Xiaoyan Lu3, Jie Liao3, Jingyang Qian3, Kai Wang1, Junyun Cheng3, Penghui Yang3, Huajun Chen2, 4, *, Xiao Xu1, 5, *, Xiaohui Fan1, 3, 5, *1 Key Laboratory of Integrated Oncology and Intelligent Medicine of Zhejiang Province, Department of Hepatobiliary and Pancreatic Surgery, Affiliated Hangzhou First People's Hospital, Zhejiang University School of Medicine, Hangzhou 310006, China. 2 Hangzhou Innovation Center, Zhejiang University, Hangzhou 310058, China 3 Pharmaceutical Informatics Institute, College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, China 4 College of Computer Science and Technology, Zhejiang University, Hangzhou 310027, China 5 Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou 310003, China. 1 These authors should be regarded as Joint First Authors 1 Corresponding author: Dr. Huajun Chen (huajunsir@zju.edu.cn), Dr. Xiao Xu (E-mail: zjxu@zju.edu.cn) and Dr. Xiaohui Fan (E-mail: fanxh@zju.edu.cn, Tel/Fax: 86-571- 88208596) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 87, 234, 104]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[144, 120, 853, 597]]<|/det|> +Spatially resolved transcriptomics (ST) provides genetic information in space toward elucidation of the spatial architecture in intact organs and the spatially resolved cell- cell communications mediating tissue homeostasis, development, and disease. To facilitate inference of spatially resolved cell- cell communications from ST data, we here present SpaTalk, which relies on a graph network and knowledge graph to model and score the ligand- receptor- target signaling network between spatially proximal cells, decomposed from ST data through a non- negative linear model and spatial mapping between single- cell RNA- sequencing and ST data. The performance of SpaTalk benchmarked on public single- cell ST datasets was superior to that of existing cell- cell communication inference methods. SpaTalk was then applied to STARmap, Slide- seq, and 10X Visium data, revealing the in- depth communicative mechanisms underlying normal and disease tissues with spatial structure. SpaTalk can uncover spatially resolved cell- cell communications for single- cell and spot- based ST data universally, providing new insights into spatial inter- cellular dynamics. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 86, 273, 104]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[145, 123, 852, 600]]<|/det|> +Cell- cell communications via secreting and receiving ligands frequently occur in multicellular organisms, which is a vital feature involving numerous biological processes1. Standard algorithms for inferring cell- cell communications mediated by ligand- receptor interactions (LRIs) primarily incorporate a database of known LRIs and single- cell transcriptomic data by delineating cell populations and their lineage relationships2, 3. One common strategy is to integrate the abundance of ligands and receptors for the inference of signals from senders to receivers based on the premise that highly co- expressed ligands and receptors are likely to mediate inter- cellular communications4, 5. Another strategy applies the downstream targets triggered by LRIs in receivers to enrich and score the ligand- receptor- target (LRT) signaling network6- 8. Although single- cell transcriptomic data can provide information on the genes contributing to cell- cell communications, the spatial information of cells is inevitably lost when dissociating tissues into single cells, thereby hindering the extension of current tools to investigate cell- cell communications in tissues with spatial structure9. + +<|ref|>text<|/ref|><|det|>[[145, 612, 852, 907]]<|/det|> +Recent technological advances in spatially resolved transcriptomics (ST) benefiting from spatial barcoding and imaging- based approaches have enabled the measurement of whole or mostly whole transcriptomes while retaining the spatial information10, 11, which have been increasingly adopted to generate new insights in the biological and biomedical domains, with dramatically improved accuracy and reliability in the inference of spatially proximal cell- cell communications12. Given the space- constrained nature of juxtacrine and paracrine signaling, such spatial gene expression information is vital to understand cell- cell communications mediating tissue homeostasis, development, and disease13, 14. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 521]]<|/det|> +Several methods have recently emerged to decode the mechanisms of cell- cell communications in space15. For example, Giotto utilizes preferential cell neighbors over single- cell ST datasets for each pair of cell types with an enrichment test to evaluate the likelihood of a given LRI based on proximal co- expressing cells and infer cell- cell communication in space16. SpaOTsc applies structured optimal transport mapping between scRNA- seq and ST data to assign a spatial position for each cell, resulting in a cell- cell distance as a transport cost to infer the ligand- receptor signaling network that mediates space- constrained cell- cell communication17. However, Giotto and SpaOTsc are limited to infer inter- cellular communications over single- cell ST data rather than the spot- based ST data and between paired cell types rather than paired cells. It still lacks of methods that can infer and visualize spatially resolved cell- cell communications at single- cell resolution over ST data to date, posing a great challenge for decoding spatial inter- cellular dynamics underlying disease pathology. + +<|ref|>text<|/ref|><|det|>[[144, 536, 852, 905]]<|/det|> +To address this challenge, we herein proposed SpaTalk, a spatially resolved cell- cell communication inference method by creatively integrating the principles of the ligand- receptor proximity and ligand- receptor- target (LRT) co- expression to model and score the LRT signaling network between spatially proximal cells relying on the graph network and knowledge graph approaches18. The performance of SpaTalk was evaluated on benchmarked datasets with remarkable superiority over other methods. By applying to STARmap19, Slide-seq20, 21, and 10X Visium22 datasets, SpaTalk revealed the in- depth communicative mechanisms underlying normal and disease tissues with spatial structure. Collectively, these results demonstrate SpaTalk as a useful and universal method that can help to uncover spatially resolved cell- cell communications for both single- cell and spot- based ST data, providing insights into the understanding + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 500, 101]]<|/det|> +of spatial inter- cellular dynamics in tissues. + +<|ref|>sub_title<|/ref|><|det|>[[147, 140, 222, 157]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[144, 175, 852, 685]]<|/det|> +Overview of the SpaTalk method. Fig. 1 provides an overview of the workflow for developing and testing SpaTalk, comprising two main components: (1) dissect the cell- type composition of ST data and (2) infer spatially resolved cell- cell communications over decomposed single- cell ST data (Fig. 1a). In the first component, the non- negative linear model (NNLM) \(^{23 - 25}\) was applied to decode the cell- type composition for a single- cell or spot- based ST data matrix using the scRNA- seq data matrix with k cell types as the underlying reference. By incorporating Lee's multiplicative iteration algorithm and relative entropy loss \(^{25}\) , the model was trained with default hyperparameters until convergence, producing a weight matrix representing the optimal proportion of cell types for each cell/spot. For single- cell ST data, the cell type with the maximum weight was assigned to label each cell. For spot- based ST data, the cell types with different weights were used as the reference to project the cells from scRNA- seq data onto the spatial spot (Fig. 1b). Through random sampling and deep iteration processes, the optimal cellular combination that most resembled the spatial spot was refined to reconstruct the single- cell ST data for spot- based ST data. + +<|ref|>text<|/ref|><|det|>[[144, 700, 852, 892]]<|/det|> +The second component of SpaTalk is to infer spatially resolved cell- cell communications and downstream signal pathways. To identify possible communications among cells mediated by LRIs, the principles of ligand- receptor proximity and ligand- receptor- target (LRT) co- expression were incorporated based on a recent review \(^{12}\) . In detail, the KNN algorithm is first applied to each cell in space to construct the cell graph network. For the ligand of the sender (cell type A) and the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 82, 852, 207]]<|/det|> +receptor of the receiver (cell type B), the number of LRI pairs is obtained from the graph network by counting the 1- hop neighbor nodes of receivers for each sender. A permutation test filters and scores the significantly enriched LRIs, generating the inter- cellular score (Fig. 1c). + +<|ref|>image<|/ref|><|det|>[[152, 230, 840, 750]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[149, 750, 850, 905]]<|/det|> +
Fig. 1 Workflow of the SpaTalk method and visualization. a Overview of SpaTalk, including the input, intermediate process of decoding spatially resolved cell–cell communications, and output. b Conceptual framework of cell-type decomposition with SpaTalk. Five different spatial technologies and datasets were selected and analyzed: spot-based ST data (Slide-seq and 10X Visium) and single-cell ST data (STARmap, MERFISH, and seqFISH+). NNLM was used to dissect the optimal proportion of cell types for the projection of cells from scRNA-seq reference data onto the spatial cells/spots, generating single-cell ST data with known cell types. c Schematic representation of SpaTalk to infer spatially resolved cell–cell communications
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 851, 207]]<|/det|> +mediated by LRIs. The inter- cellular and intra- cellular scores were obtained and combined from the cell- cell graph network and the LRT- knowledge graph (KG), respectively, by integrating the KNN, permutation test, and random walk algorithms. L, ligand; R, receptor; TF, transcription factor; T, target. d Visualization of spatially resolved cell- cell communications, including a heatmap, Sankey plot, and diagram of the LRI from senders to receivers in space, as well as ligand- receptor- target (LRT) signaling pathways over the reconstructed single- cell ST data. + +<|ref|>text<|/ref|><|det|>[[145, 240, 853, 540]]<|/det|> +The knowledge graph (KG) was then introduced to model the intracellular signal propagation process. In practice, LRIs from CellTalkDB26, pathways from Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome, and TFs from AnimalTFDB27 were integrated to construct the LRT- KG, wherein the weight between entities represents the co- expressed coefficient. Taking the receptor as the query node, we incorporated the random walk algorithm28 into the LRT- KG to filter and score the downstream activated TFs and calculate the intracellular score of the LRI from senders to receivers. Inter- cellular and intra- cellular scores are combined to rank the LRIs that mediate spatially resolved cell- cell communications. + +<|ref|>text<|/ref|><|det|>[[145, 554, 853, 781]]<|/det|> +SpaTalk also includes numerous visualization functions to characterize the cell- type composition and spatially resolved cell- cell communications, such as the diagram of the LRI from senders to receivers in space and LRT signaling pathways, over the reconstructed single- cell ST data (Fig. 1d). Five broad ranges of different spatial technologies and corresponding representative datasets were analyzed and visualized: spot- based ST data (Slide- seq20, 21 and 10X Visium22) and single- cell ST data (STARmap19, MERFISH29, and seqFISH+30). + +<|ref|>text<|/ref|><|det|>[[145, 797, 851, 885]]<|/det|> +Performance comparison of SpaTalk with other methods. The cell- type decomposition by SpaTalk is the foundation for subsequent analyses. To evaluate its performance, four single- cell ST datasets from the mouse cortex, hypothalamus, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 854, 558]]<|/det|> +olfactory bulb, and sub- ventricular zone were utilized (Supplementary Fig. 1a). All cells were split according to the fixed spatial distance and then merged into simulated spots as the benchmark datasets (Fig. 2a). The quality of predicted cell- type decompositions and expression profiles was evaluated by Pearson's correlation coefficient and the root mean square error (RMSE) based on the ground truth, wherein SpaTalk exhibited fantastic performance over the benchmark datasets (Supplementary Fig. 1b, c). Although the majority of existing cell- type deconvolution methods (RCTD31, Seurat32, SPOTlight33, deconvSeq34, and Stereoscope35) can achieve a decent correlation coefficient and low RMSE on spot deconvolution, SpaTalk outperformed these methods on most benchmark datasets with the top- ranked performance, except for the MERFISH dataset (Fig. 2b). The MERFISH dataset only includes 155 genes, whereas the STARmap and seqFISH+ datasets cover 1020 and 10,000 genes per cell, respectively, suggesting that SpaTalk is potentially more effective for spatial data with higher gene coverage. + +<|ref|>text<|/ref|><|det|>[[144, 571, 854, 904]]<|/det|> +We next compared the performance of SpaTalk with that of existing cell- cell communication inference methods (Supplementary Fig. 2a). Consequently, most methods exhibited a large fraction of overlapped predictions with the rest of the methods despite the different number of inferred cell- cell communications (Supplementary Fig. 2b, c), indicating the reproducible inference across these methods. Regarding the inferred LRIs (Supplementary Fig. 2d), we reasoned that the spatial distances of the inferred LRI between sender- receiver pairs will be shorter than those between all cell- cell pairs and thus the inferred LRI will be more co- expressed in local space as cells that are close are more likely to signal (Fig. 2c). The one- sided Wilcoxon test was performed to evaluate the spatial proximity significance of the inferred LRIs, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 312]]<|/det|> +and the co- expressed percentage of the LRI was calculated as the co- expression level using the cell- cell graph network. Although most LRIs inferred by other methods showed significantly closer spatial distances between sender- receiver pairs than that between all cell- cell pairs, superior performance of SpaTalk was observed, ranking first for both evaluation indices for STARmap datasets (Fig. 2d). Similarly, SpaTalk obtained a higher median \(- \log_{10}P\) value and co- expression percent on the seqFISH+ OB and SVZ datasets but not for SpaOTsc (Fig. 2e). + +<|ref|>image<|/ref|><|det|>[[147, 325, 844, 710]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 716, 850, 914]]<|/det|> +
Fig. 2 Superior performance of SpaTalk over existing methods. a Schematic diagram for generating simulated spot data. Cells were split according to the fixed spatial distance and then merged for the single-cell ST data with known cell types. b Performance comparison of SpaTalk with other existing cell-type deconvolution methods (RCTD, Seurat, SPOTlight, deconvSeq, Stereoscope). The asterisk represents the top-ranked method for each dataset. NA, not available. c Schematic illustration of the procedure and rationale for single-cell ST data to evaluate predicted LRIs that mediate spatially resolved cell–cell communications. d and e Performance comparison of SpaTalk with existing cell–cell communication inference methods (Giotto, SpaOTsc, NicheNet, CytoTalk, and CellCall) on the STARmap and seqFISH+ datasets. The \(P\) value
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 83, 851, 244]]<|/det|> +represents the difference of spatial distances between sender- receiver and all cell- cell pairs assessed with the Wilcoxon test. f Schematic illustration of the procedure and rationale for single- cell ST data to evaluate predicted downstream target and pathways underlying LRIs. g Performance comparison of SpaTalk on the inferred downstream targets with other methods (NicheNet, CytoTalk, and CellCall) over the STARmap and seqFISH+ datasets. The \(P\) value represents the significance of enriched pathways or biological processes from the KEGG and Reactome databases using inferred downstream targets with the Fisher exact test. + +<|ref|>text<|/ref|><|det|>[[145, 279, 852, 644]]<|/det|> +We compared the performance of SpaTalk for inference of intra- cellular signal pathways of the receiver cell type triggered by the LRI with those of NicheNet6, CytoTalk7, and CellCall8 that also infer the downstream targets of LRIs. We reasoned that a more accurate method would be more likely to enrich the receptor- related biological processes or pathways using the inferred downstream target genes in the receiver cell type (Fig. 2f); hence, the Fisher- exact test was adopted for pathway enrichment analysis with the KEGG and Reactome databases on target genes in receivers. The target genes inferred by all methods enriched the most intra- cellular pathways or biological processes triggered by the inter- cellular LRI (Fig. 2g). Nevertheless, SpaTalk exhibited the top- ranked performance over three benchmarked datasets, exceeding other existing methods in inference of the LRT signal network. + +<|ref|>text<|/ref|><|det|>[[145, 661, 852, 890]]<|/det|> +Identification of signal transmission among neurons and non- neuronal cells. SpaTalk was first applied to investigate and visualize the cell- cell communications over the STARmap ST dataset of the mouse visual cortex (Fig. 3a), including data of 1020 sequenced genes for 973 cells in space covering the spatial axis of excitatory neurons (eL2/3, eL4, eL5, eL6, annotated by anatomic cortical layers); Pvalb, Reln, Sst, and Vip- expressing neurons; and non- neuronal cells, including astrocytes (Astro), endothelial cells (Endo), microglia (Micro), oligodendrocytes (Oligo), and smooth muscle cells + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 522]]<|/det|> +(SMCs) from layers L1 though L6 to the corpus callosum and hippocampus. As shown in Fig. 3b, SpaTalk identified the spatial signal transmission among excitatory and inhibitory neurons and non- neuronal cells such as Sst- Sstr2, Inhba- Acvr1c, and Cort- Sstr2. For example, the direct cell- cell communication mediated by the Sst- Sstr2 interaction was observed between Sst- expressing neurons and Pvalb- expressing neurons (Supplementary Fig. 3a), which are known to regulate neuron activity via nonsynaptic, inter- neuronal communication36, 37. Concordantly, these identified LRIs among neurons and non- neuronal cells are associated with multiple biological processes and pathways that play vital roles in the regulation of physiological neural development and the balance of excitatory/inhibitory transmission in the central nervous system38, 39, including growth hormone synthesis, secretion, and action; neuroactive ligand- receptor interaction; signaling by activin; and the transforming growth factor (TGF)- beta signaling pathway (Fig. 3c). + +<|ref|>text<|/ref|><|det|>[[144, 536, 852, 904]]<|/det|> +Notably, Astro were relatively abundant across space of the L5 layer and colocalized with numerous excitatory neurons, exhibiting direct cell- cell communication with eL2/3, eL5, and eL6 (Supplementary Fig. 3b). Given the Cort- Sstr2 interaction between eL5 and Astro, eL5 highly expressed and secreted the CORT ligand to interact with the SSTR2 receptor on Astro with a highly overlapped distribution density of eL5 and Astro in the constrained space, wherein the spatial distances of Cort and Sstr2 in eL5- Astro pairs were significantly closer than those in all cell- cell pairs (Fig. 3d). Focusing on the intra- cellular signal pathway of the Cort- Sstr2 interaction reconstructed by SpaTalk, two downstream TFs were identified, Egr1 and Smad3, which are involved in canonical TGF- beta signaling, in line with previous findings40. Despite the relatively low- rise co- expression of target genes in receivers, the intra + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 242]]<|/det|> +cellular signal triggered by the Cort- Sstr2 interaction successfully propagated to the target genes, reaching a high percentage of cells expressing most target genes (Fig. 3e). The Fisher exact test showed significant enrichment with receptor- related pathways (Fig. 3f), indicating the reliability of eL5- Astro communications mediated by the Cort- Sstr2 interaction. + +<|ref|>text<|/ref|><|det|>[[144, 257, 853, 872]]<|/det|> +We then applied SpaTalk to another spot- based ST dataset of the mouse cortex (Slide- seq v2), including 17,545 unique genes among 42,550 spots in space, reaching up to 4000 expressed genes per spot. Leveraging previously published adult mouse cortical cell taxonomy by scRNA- seq data41 (Supplementary Fig. 3c), SpaTalk reconstructed the spatial transcriptomics atlas at the single- cell resolution for Slide- seq data, which showed consistent spatial localization of neurons and non- neuronal cells across different layers, including the major excitatory and inhibitory neurons and Oligo (Fig. 3g). Compared to STARmap data, oligodendrocyte progenitor cells, and Igtp-, Ndnf-, and Smad3- expressing neurons were also observed in the Slide- seq data. Concordantly, most cell- cell communications in STARmap data were also found in Slide- seq data except for those of distinct cell types in the two datasets (Supplementary Fig. 3d). For example, the eL5- Astro communications mediated by the Cort- Sstr2 interaction in space and the downstream targets such as SMAD3 were also observed in Slide- seq data (Fig. 3h), suggesting the universality of the spatially resolved cell- cell communications inferred by SpaTalk. In addition, the direct communications among Pvalb neurons, eL6, and Oligo were also significantly enriched, in accordance with the fact that neuron- oligo communication controls the oligodendrocyte function and myelin biogenesis (Supplementary Fig. 3e). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[149, 88, 846, 536]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 540, 850, 897]]<|/det|> +
Fig. 3 Identification of spatial inhibitory signal transmission among neurons and non-neuronal cells. a STARmap single-cell ST dataset of the mouse visual cortex involving 973 cells and 1020 genes. Astro, astrocytes; eL2/3, eL4, eL5, eL6, excitatory neuron subtypes; Endo, endothelial cells; HPC, hippocampus; Micro, microglia; Oligo, oligodendrocytes; SMC, smooth muscle cells; cc, corpus callosum. b Significantly enriched LRIs that mediate cell–cell communications among neurons and non-neuronal cells inferred by SpaTalk with \(P< 0.05\) . The \(P\) value represents the significance of spatial proximity of LRIs using the permutation test. c Sanky plot of the associations among ligands, receptors, and biological processes or pathways in the KEGG and Reactome databases that mediate cell–cell communications in the central nervous system. d Spatial distribution and intra-cellular signaling pathways of the Cort–Sstr2 pairs between the eL5 senders and Astro receivers. \(P\) values were calculated with the Wilcoxon test. e Co-expression of target genes in receivers and the percentage of expressed cells for target genes. f Significantly enriched biological processes and pathways with the ligand-receptor-target genes using the Fisher exact test. g Slide-seq spot-based ST dataset of the mouse visual cortex involving 42,550 spots and 22,542 genes. OPC, oligodendrocyte progenitor cell. h Communications of eL5–Astro mediated by the Cort–Sstr2 interaction in space and the intra-cellular signal pathway
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 470, 101]]<|/det|> +inferred by SpaTalk over Slide- seq data. + +<|ref|>text<|/ref|><|det|>[[144, 120, 852, 630]]<|/det|> +Metabolic modulation of periportal hepatocytes on pericentral hepatocytes. We applied SpaTalk to the Slide- seq ST dataset of the mouse liver covering 17,545 unique genes among 25,595 spots in space (Fig. 4a). To explore the cell- type composition of Slide- seq data, a mouse liver scRNA- seq reference integrating the non- parenchymal cells from the Mouse Cell Atlas (MCA) \(^{42}\) and the parenchymal hepatic cells from the GSE125688 dataset \(^{43}\) were utilized (Supplementary Fig. 4a), containing 6029 cells, including major immune cells such as macrophages (Macro), and the pericentral and periportal hepatocytes. The reconstructed single- cell ST atlas was perfectly accordant with the original outcome obtained by Slide- seq (Fig. 4b), wherein the expression of known marker genes \(^{44}\) and the percent for each cell type were highly correlated across spots (Supplementary Fig. 4b, c, d), such as the pericentrally and periportally zonated genes Cyp2e1 and Pck1 (Fig. 4c). Immune cells were hardly observed in each spot, with pericentral and periportal hepatocytes accounting for the major proportion across spots (Supplementary Fig. 4e); the same phenomenon was observed in recently published ST data of the healthy liver \(^{45}\) . + +<|ref|>text<|/ref|><|det|>[[144, 644, 852, 907]]<|/det|> +The cell- cell communications between pericentral and periportal hepatocytes were further explored by SpaTalk (Fig. 4d). Both hepatocyte types secrete and receive multiple ligands for their communication, forming spatially distributed metabolic cascades to cooperatively optimize the metabolic environment. For instance, with the gradient expression of enzymes in sequential lobule layers, pericentral hepatocytes perform the primary steroid, alcohol, and lipid metabolic processes, while periportal hepatocyte mainly carry out the small- molecule and monosaccharide biosynthetic processes, amino acid and triglyceride metabolic processes, and gluconeogenesis (Fig. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 451]]<|/det|> +4e, f), in line with the variable functions of zonated hepatocytes residing in the central and portal veins46. Notably, the periportal hepatocytes substantially expressed more ligands, including epidermal growth factor (Egf), transforming growth factor alpha (Tgfa), heparin-binding EGF-like growth factor (Hbegf), insulin-like growth factor 1 (Igf1), and vascular endothelial growth factor A (Vegfa), to promote the growth of pericentral hepatocytes. As the blood flows from the portal vein toward the central vein, this could reflect the fact that periportal hepatocytes respire most of the oxygen, leading to decreased oxygen concentrations along the lobule axis; thus, periportal hepatocytes secrete numerous growth factors via paracrine signaling to modulate the pericentral hepatocytes for the prevention of hypoxia and the maintenance and amelioration of pericentrally metabolic functions46. + +<|ref|>text<|/ref|><|det|>[[144, 466, 852, 904]]<|/det|> +Taking the LRI of Apob- Cd36 as an example, the spatially resolved cell- cell communications between periportal and pericentral hepatocytes mainly occurred across the mid-lobule layers (Fig. 4g). The gene product of Apob is an apolipoprotein of chylomicrons and low- density lipoproteins highly involved with the regulation of lipids and fatty acids metabolism through CD36, indicating the modulation of periportal hepatocytes on the metabolic microenvironment sensed by pericentral hepatocytes. From the reconstructed intra- cellular signal propagation network triggered by the Apob- Cd36 interaction (Fig. 4h), sequential target TFs were activated, including Ahr that regulates xenobiotic- metabolizing enzymes such as cytochrome P450, and Nr1h4 that regulates the expression of genes involved in bile acid synthesis and transport, in agreement with the corresponding module score of pericentral hepatocytes in space (Fig. 4i). The LRT network was also remarkably enriched in the AMPK and PPAR signaling pathways, which play crucial roles in the regulation of energy + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[89, 85, 851, 170]]<|/det|> +334 and metabolic homeostasis, suggesting the spatially fine- tuned cell- cell communications along the portal- central lobule axis for minimizing risks to pericentral hepatocytes. + +<|ref|>image<|/ref|><|det|>[[150, 191, 844, 636]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 644, 850, 901]]<|/det|> +
Fig. 4 Modulation of periportal hepatocytes on the metabolic microenvironment sensed by pericentral hepatocytes. a Slide-seq spot-based ST dataset of the mouse liver involving 25,595 spots and 17,545 genes. b Cell-type decomposition by SpaTalk. PC, pericentral; PP, periportal; Hep, hepatocytes. c Scaled Pearson's correlation coefficients between the expression of known marker genes and the percent for each cell type. DC, dendritic cell; Macro, macrophages. d Enriched LRIs that mediate cell–cell communications between pericentral and periportal hepatocytes. e Significant differentially expressed genes (DEGs) between periportal and pericentral hepatocytes assessed with the Wilcoxon test and the corresponding significantly enriched biological processes and pathways determined with the Metascape web tool. Representative DEGs are labeled beside the heatmap. f Significantly activated pathways in pericentral (up) and periportal (down) hepatocytes determined by Gene Set Enrichment Analysis (GSEA). g Communications from periportal hepatocytes to
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 85, 851, 163]]<|/det|> +pericentral hepatocytes mediated by the Apob- Cd36 interaction in space. h Intracellular signal pathway inferred by SpaTalk over Slide- seq data and the significantly enriched pathways over the ligand- receptor- target network. i Inferred module score of each hepatocyte type over the pathway signatures determined with Seurat. + +<|ref|>text<|/ref|><|det|>[[145, 183, 852, 622]]<|/det|> +Spatial characterization of cell types over 10X Visium data. Given the widely used 10X Visium tool in ST studies, we applied SpaTalk to a human skin squamous cell carcinoma (SCC) ST dataset published by Ji et al.22, who profiled SCC and matched normal tissues via 10X scRNA- seq and used Visium to identify a tumor- specific keratinocyte (TSK) in the tumor (Fig. 5a). Using the matched SCC scRNA- seq data of patient 2 as reference, the optimal cell- type composition for each spot was deconvoluted by SpaTalk, which exhibited a similar characterization with that histologically assessed from hematoxylin and eosin- stained frozen sections (Fig. 5b). The percent of TSK inferred by our method was compared to the TSK score based on markers (e.g., MMP10, PTHLH, LAMC2, and IL24) defined by Ji et al. across 646 spots (Fig. 5c). A high correlation between the TSK percent and score was observed (Fig. 5d). Moreover, the percent of inferred cell types was prominently associated with the expression of known marker genes, indicating the accuracy of SpaTalk for cell- type decomposition (Supplementary Fig. 5a, b). + +<|ref|>text<|/ref|><|det|>[[145, 636, 852, 897]]<|/det|> +Next, we reconstructed the single- cell ST profile by assuming a total of 30 cells in each spot according to a recent review12, which covered the main epithelial cells, including differentiating, cycling, and basal keratinocytes; melanocytes; fibroblasts (FB); Endo; natural killer (NK) cells; and T cells (Fig. 5e). Despite the asymmetrical distribution for most cell types, TSK, FB, and Endo showed specific patterns of locations in space, which were highly adjacent in some tumor areas (Fig. 5f), forming direct cell- cell communications in the tumor microenvironment (TME). By filtering cells from the TSK leading spots (score \(\geq 0.8\) ), we found that TSKs reside within a fibrovascular niche, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 383]]<|/det|> +resulting in high colocalization of TSKs, FB, and Endo at the TSK leading spots (Fig. 5g), in line with the previous findings22. Moreover, we used SpaTalk to investigate the cell- type composition over the ST data of another SCC patient. Despite a low percentage across 621 spatial spots, most TSKs centered on a handful of corner spots in space, exhibiting a highly consistent distribution with TSK scores (Fig. 5h and Supplementary Fig. 5c). Unsurprisingly, the fibrovascular niche was also observed in the TSK leading spots of patient 10 with clear spatial co- localization (Supplementary Fig. 5d), indicating the close cell- cell communication among TSKs, FB, and Endo in the TME underlying the occurrence and development of SCC. + +<|ref|>image<|/ref|><|det|>[[150, 400, 839, 875]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 880, 850, 898]]<|/det|> +
Fig. 5 Spatial characterization of tumor and stromal cells in human squamous cell
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 85, 852, 323]]<|/det|> +carcinoma Visium data with SpaTalk. a Visium spot- based ST dataset of human skin SCC in patient 2 with the matched scRNA- seq dataset involving the main keratinocytes (KC), stromal cells, and immune cells. b Cell- type decomposition by SpaTalk. Cyc, cycling; Diff, differentiating; NK, natural killer; FB, fibroblast. c TSK percent and TSK score across spatial spots. The expression of known TSK markers is plotted. d Pearson's correlation coefficient between the TSK percent and TSK score. f Contour plot of TSK, FB, and Endo based on the reconstructed single- cell ST atlas by SpaTalk. g TSK leading spots with a TSK score \(>0.8\) in space. The bar chart represents the number of different cell types and the line chart represents the number of neighbors adjacent to TSKs among the TSK leading spots. h Visium spot- based ST dataset of human skin SCC in patient 10 and the cell- type decomposition by SpaTalk showing the percent of TSKs across 621 spatial spots. + +<|ref|>text<|/ref|><|det|>[[145, 342, 852, 886]]<|/det|> +Reconstruction of TSK- stroma communications in space. To dissect the underlying LRI mediating the spatially resolved cell- cell communications between TSKs and stromal cells of the fibrovascular niche in the TME, we applied SpaTalk to infer the communications between TSK- FB and TSK- Endo pairs over the decomposed single- cell ST data of SCC in patient 2, including the top- ranked 20 LRIs based on the integrated inter- cellular and intra- cellular scores (Fig. 6a). Consistent with a TSK- fibrovascular niche, prominent TSK signaling to FB and Endo was mediated by several common ligand- receptor pairs, including VEGFA- NPR1, VEGFB- NPR1, PGF- SDC1, and CDH1- ITGAE, associated with tumor angiogenesis. Additionally, TSKs modulate FB through secreting matrix metallopeptidase (MMP)1 and MMP9, which are linked to tumor metastasis via cellular movement and extracellular matrix (ECM) disassembly. Conversely, FB and Endo prominently co- expressed numerous ligands such as MDK, HGF, HMGB1, and THBS1, matching TSK receptors that promote the proliferation and differentiation of TSKs (Fig. 6b). Further supporting TSKs as an epithelial mesenchymal transition (EMT)- like population, SpaTalk predicted that the widely expressed TGFB1 regulates TSKs. TSK receptors corresponding to additional ligands from FB and Endo + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 83, 850, 170]]<|/det|> +included several integrins (e.g., ITGA5 and ITGB6) and nectins (e.g., NECTIN1 and NECTIN2), highlighting other pathways associated with EMT and epithelial tumor invasion \(^{47,48}\) . + +<|ref|>image<|/ref|><|det|>[[155, 188, 840, 675]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 679, 850, 897]]<|/det|> +
Fig. 6 Reconstruction of cell–cell communications between TSK subpopulations and stromal cells in space with SpaTalk. a Top 20 inferred LRIs that mediate cell–cell communication from the TSK senders to the FB and Endo receivers. Colored blocks represent the known ligand–receptor pairs in CellTalkDB. The asterisk represents the significantly enriched LRIs determined by SpaTalk. b Top 20 inferred LRIs that mediate cell–cell communication from the FB and Endo senders to TSK receivers. c Region of interest (ROI) covering 42 spatial spots in space with a high total score of TSK, FB, and Endo according to their signature genes. The point plot shows the decomposed single-cell ST atlas determined by SpaTalk. d Expression of known cancer-associated fibroblast (CAF) markers across TSKs, FB, Endo, and other cells. e Differentially expressed genes (DEGs) between EMT-like and EMT-unlike TSKs and the corresponding
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 852, 202]]<|/det|> +enriched biological processes and pathways. Representative DEGs are labeled beside the point. f Number of cell- cell pairs over the LRIs from the EMT- like and EMT- unlike TSKs to CAFs. g Communications from EMT- like and EMT- unlike TSKs to CAFs mediated by the MMP1- CD44 interaction in space. h Comparison of communications among CAFs, Endo, EMT- like TSKs, and EMT- unlike TSKs with respect to the number of cell- cell pairs in space over the matched LRIs evaluated with a paired t test. + +<|ref|>text<|/ref|><|det|>[[144, 222, 852, 730]]<|/det|> +Next, we focused on a region of interest (ROI) covering 42 spatial spots in space for in- depth exploration of TSK- stroma communications, which exhibited a high total score of TSK, FB, and Endo according to their signature genes, occupying the major part in the ROI (Fig. 6c). By mapping cancer- associated fibroblast (CAF) markers to the cells in space, the majority of FB in the ROI highly expressed the known CAF marker genes (e.g., VIM, FAP, POSTIN, and SPARC) (Fig. 6d), hinting at the transformation of FB to CAFs induced by the adjacent TSKs and conversely supporting the stemness of TSKs via direct cell- cell communication in space (Supplementary Fig. 6a). Notably, the TSKs appear to be extremely heterogeneous with respect to the broad range of EMT scores in the ROI; thus, TSKs were further classified into 268 EMT- like and 268 EMT- unlike populations (Supplementary Fig. 6b). By comparing their differentially expressed genes, EMT- like TSKs were dramatically enriched with ECM organization, proteoglycans in cancer, regulation of cell adhesion, and the VEGFA- VEGFR2 signaling pathway, representing more invasive properties compared with EMT- unlike TSKs (Fig. 6e). + +<|ref|>text<|/ref|><|det|>[[145, 744, 852, 902]]<|/det|> +Additionally, EMT- like TSKs appear to be more communicative with surrounding CAFs in the TME in light of the greater number of cell- cell pairs over the LRIs that prominently mediate TSK- stroma communications in space, such as LAMB3- ITGB1, LAMA3- ITGB1, LAMC2- ITGB1, and MMP1- CD44 (Fig. 6f and Supplementary Fig. 6c, d, e). Metastasis- related laminins are essential for formation and function of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 852, 347]]<|/det|> +basement membrane, whereas MMPs are involved in ECM breakdown, both contributing to the aggravated malignancy of tumors. Moreover, CD44 expression on CAFs plays a supporting role in the induction of cellular stemness, wherein CAFs have a preference of cell- cell communications with EMT- like TSKs in space (Fig. 6g). Interestingly, EMT- unlike TSKs notably exhibited more communicative cell- cell pairs with Endo, whereas EMT- like TSKs exhibited significantly more communicative cell- cell pairs with CAFs over the matched LRIs (Fig. 6h), consistent with the observed contribution of CAFs to EMT in a broad range of tumors49, 50. + +<|ref|>sub_title<|/ref|><|det|>[[147, 383, 253, 401]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[147, 421, 852, 579]]<|/det|> +We have demonstrated the capabilities of SpaTalk to infer and visualize spatially resolved cell- cell communications mediated by significantly enriched LRIs under normal and disease states over existing representative datasets, including the single- cell ST data generated from STARmap, MERFISH, and seqFISH+, and the spot- based ST data obtained via Slide- seq and 10X Visium. + +<|ref|>text<|/ref|><|det|>[[147, 595, 852, 893]]<|/det|> +There are two principles to decode the mechanisms of cell- cell communications: ligand- receptor proximity and LRT co- expression12. In a given tissue niche, cells are more likely to communicate with each other when they are spatially adjacent and activate downstream target genes in the receiving cell triggered by the LRI in proximal cells; thus, ST data are well suited to apply the two principles for inferring inter- cellular communications. Accordingly, our proposed SpaTalk realizes the integration of these two principles by incorporating the KNN and cell- cell graph network to filter spatially proximal cell pairs and corresponding LRIs, followed by utilizing the knowledge graph algorithm to model the LRT signal propagation process. Consequently, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 206]]<|/det|> +performance of SpaTalk was superior to that of other methods over the benchmarked ST datasets with respect to several evaluation indices, demonstrating the reliability of the two principles in decoding cellular cross- talk, especially for juxtracrine and paracrine communication. + +<|ref|>text<|/ref|><|det|>[[144, 222, 852, 801]]<|/det|> +Importantly, SpaTalk is applicable to either single- cell or spot- based ST datasets generated from mainstream ST technologies. For the former, SpaTalk assigns a label to each cell by selecting similar cell types with the top- ranked weight via NNLM for single- cell ST data, generating the ST atlas at single- cell resolution with known cell types for the subsequent inference of cell- cell communications. For spot- based ST data, SpaTalk selects and maps the optimal combination of cells in accordance with the decomposed optimal weight/percent of cell types via NNLM and the transcriptome profiles of spatial spots to reconstruct the ST atlas at single- cell resolution with known cell types. Notably, applications of SpaTalk to the mouse cortex and liver datasets sequenced by STARmap and Slide- seq, respectively, revealed the evidential LRIs in space that mediate the spatially resolved cell- cell communications contributing to normal physiological processes. Moreover, exploration of SpaTalk on the human skin SCC dataset obtained from 10X Visium identified the variable preference of communication among tumor subpopulations, CAF, and Endo. These cases convincingly demonstrate the universality of SpaTalk in decoding the mechanism of cell- cell communications in space underlying normal and disease tissues for single- cell and spot- based ST data. + +<|ref|>text<|/ref|><|det|>[[145, 815, 851, 902]]<|/det|> +As unmatched scRNA- seq and ST data would directly influence the cell- type decomposition, an important feature of SpaTalk is the ability to assign a spot/cell into an unsure category considering the unseen cell types in the scRNA- seq reference. For + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 416]]<|/det|> +example, the scRNA- seq reference for the Slide- seq mouse cortex ST data was obtained from another repository for the same tissue, resulting in numerous unsure cells by assuming one cell in each spot in terms of the high resolution (10 μm) of Slide- seq technology that almost approaches single- cell resolution. Additionally, the extremely low gene coverage of several spatial spots severely affects the regression model, which were regarded as the unsure type by SpaTalk. However, with respect to the human skin SCC datasets, SpaTalk removed the unsure type for the matched scRNA- seq and ST data. As the matched multi- modal datasets will undoubtedly become greater in number, application of SpaTalk and similar methods will be required for accurate inference of spatially resolved cell- cell communications. + +<|ref|>text<|/ref|><|det|>[[144, 432, 852, 833]]<|/det|> +Additionally, SpaTalk characterizes the spatial distribution for each cell type within the reconstructed ST at single- cell resolution through the contour plot of cellular density in space, which enables analyzing the proximal relationship between paired cell types. Moreover, SpaTalk enables the statistical analyses and visualization of spatially proximal LRIs in space, forming a dynamic cell- cell communication network. Currently, it is hard to analyze and visualize the LRI at single- cell resolution for scRNA- seq data, wherein the common practice is to interpret the LRI for paired cell types. By incorporating spatial information, SpaTalk displays the enriched LRI at single- cell resolution via the spatially proximal co- expressed cell pairs, offering an informatively brand- new approach for the analysis and visualization of the LRI and its mediated cell- cell communication underlying the disease pathology from a novel perspective, as shown in the application of SpaTalk to the human skin SCC datasets. + +<|ref|>text<|/ref|><|det|>[[145, 850, 850, 903]]<|/det|> +Adding spatial constraints in cell- cell communication inference is critical to the spatial analysis of juxtracrine and paracrine communications. However, this constraint + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 383]]<|/det|> +inevitably causes the failure of inferring long- range communications such as endocrine and telecrine signaling. Classification of LRIs into short- range and long- range communications with prior knowledge might be helpful to infer the comprehensive communication categories computationally. Moreover, it is potentially beneficial to include other omics data with the increasing multi- modal datasets generated from state- of- the- art technologies such as 10X Multioome and Digital Spatial Profiling51 in studying spatially regulated cell- cell communications. Thus, more reliable computational models might be needed for more accurate integration of multi- modal data and inference. + +<|ref|>sub_title<|/ref|><|det|>[[148, 417, 240, 436]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[144, 455, 853, 896]]<|/det|> +Datasets. For STARmap, the single- cell ST data of the mouse cortex (20180410- BY3_1kgenes) was obtained from a public data portal (https://www.dropbox.com/sh/f7ebheru1lbz91s/AABYSSjSTppBmVmVWl2H4sKa?dl=0). For MERFISH, the single- cell ST data of the naïve female mouse (Animal_ID: 1, Bregma: 0.26) hypothalamic preoptic region was downloaded from Dryad (https://datadryad.org/stash/dataset/doi:10.5061/dryad.8t8s248). For seqFISH+, the single- cell ST data of the mouse cortex and olfactory bulb were retrieved from the Github repository (https://github.com/CaiGroup/seqFISH-PLUS). For Slide-seq, the spot- based ST data of the mouse liver (Puck_180803_8) and somatosensory cortex (Puck_200306_03) were obtained from the Broad Institute Single Cell Portal (https://singlecell.broadinstitute.org/single_cell/study/SCP354/slide-seq-study) and (https://singlecell.broadinstitute.org/single_cell/study/SCP815/highly-sensitive- spatial-transcriptomics-at-near-cellular-resolution-with-slide-seqv2), respectively. For + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 277]]<|/det|> +10X Visium, the spot- based ST data and scRNA- seq data of human SCC were downloaded from the Gene Expression Omnibus (GEO) repository (GSE144240). The mouse liver scRNA- seq data of non- parenchymal and parenchymal hepatic cells were refined from the MCA (https://figshare.com/articles/MCA_DGE_Data/5435866) and GSE125688, respectively. The mouse cortex scRNA- seq data were obtained from GSE71585. + +<|ref|>text<|/ref|><|det|>[[144, 291, 852, 660]]<|/det|> +Data processing. For the liver scRNA- seq datasets, the non- parenchymal cells and parenchymal hepatic cells were collected from the MCA and GSE125688, respectively, wherein hepatocytes were classified into pericentral and periportal hepatocytes with principal component analyses and clustering analysis. For the MERFISH dataset, ependymal cells were excluded due to the limited cell number in the section (<2). For other datasets, all cells were included in the filtered matrices. Human and mouse gene symbols were revised in accordance with NCBI gene data (https://www.ncbi.nlm.nih.gov/gene/) updated on June 30, 2021, wherein unmatched genes and duplicated genes were removed. For all ST and scRNA- seq datasets, the raw counts were normalized via the global- scaling normalization method LogNormalize in preparation for running the subsequent scDeepSort pipeline. + +<|ref|>text<|/ref|><|det|>[[144, 675, 852, 833]]<|/det|> +SpaTalk algorithm. The SpaTalk model consists of two components: cell- type decomposition and spatial LRI enrichment. The first component is to infer cell- type composition for single- cell or spot- based ST data, and the second component is to infer spatially proximal ligand- receptor interactions that mediate cell- cell communications in space. + +<|ref|>text<|/ref|><|det|>[[144, 850, 850, 902]]<|/det|> +Cell- type decomposition. To dissect the cell- type composition for the ST data matrix \(T[n \times s](n\) genes and \(s\) spots/cells), NNLM was first applied to obtain the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 850, 172]]<|/det|> +optimal proportion of cell types using the scRNA- seq data matrix \(S[n \times c]\) ( \(n\) genes and \(c\) cells) as the reference with \(k\) cell types. Let \(Y = \{y_1, y_2, \dots , y_n\}\) be the expression profile for each spot/cell to establish the following linear model: + +<|ref|>equation<|/ref|><|det|>[[444, 188, 550, 204]]<|/det|> +\[Y = X\beta +\epsilon\] + +<|ref|>text<|/ref|><|det|>[[144, 220, 851, 344]]<|/det|> +where \(X = [n \times k]\) is the average expression profile generated from \(S\) and \(\epsilon\) represents random error. Mean relative entropy loss was then used to measure the difference between the predicted and observed values. Therefore, the objective function can be written as: + +<|ref|>equation<|/ref|><|det|>[[234, 360, 761, 380]]<|/det|> +\[argmin\{\beta \geq 0\} L(Y - X\beta) + \lambda_1R_1(\beta) + \lambda_\alpha R_\alpha (\beta) + \lambda_2R_2(\beta)\] + +<|ref|>text<|/ref|><|det|>[[144, 394, 851, 554]]<|/det|> +where \(R_1\) , \(R_\alpha\) and \(R_2\) represent the L1, angle, and L2 regularization with non- negative \(\lambda_1\) , \(\lambda_\alpha\) and \(\lambda_2\) initialized by zero \(^{23, 24}\) . The model was trained with the above objective function using Lee's multiplicative iteration algorithm \(^{25}\) with default hyperparameters until convergence or after 10,000 iterations to generate the coefficient matrix \(C[k \times s]\) . + +<|ref|>text<|/ref|><|det|>[[144, 569, 853, 727]]<|/det|> +For single- cell ST data, the cell type with the maximum coefficient was assigned to each cell. For spot- based ST data, let \(M\) be the maximum cell number for each spot, which was set to 30 for 10X Visium data and was set to 1 for Slide- seq data according to a recent review. In practice, the optimal cellular combination \(\omega\) for each spot was determined by the following function: + +<|ref|>equation<|/ref|><|det|>[[277, 741, 716, 780]]<|/det|> +\[\omega_{i}(i \in \{1,2,\dots,k\}) = \left\{ \begin{array}{ll} [M\beta_{i}] + 1 & ( \{M\beta_{i}\} \geq 0.5) \\ [M\beta_{i}] & ( \{M\beta_{i}\} < 0.5) \end{array} \right.\] + +<|ref|>text<|/ref|><|det|>[[144, 794, 851, 887]]<|/det|> +wherein \([M\beta_{i}]\) and \(\{M\beta_{i}\}\) represent the integer and fractional parts of \(M\beta_{i}\) , respectively. For each spot, we randomly selected \(m\) \((m = \sum_{i = 1}^{k} \omega_{i})\) cells from \(S\) to compare their merged expression profile \(\epsilon\) with the ground truth according to the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 305, 101]]<|/det|> +following function: + +<|ref|>equation<|/ref|><|det|>[[321, 115, 675, 157]]<|/det|> +\[argmin\{m\leq M\} \sum_{i = 1}^{n}(Y_{i} - \sum_{j = 1}^{m}\epsilon_{i}^{j})^{2}\] + +<|ref|>text<|/ref|><|det|>[[147, 153, 850, 241]]<|/det|> +To assign a coordinate \((x, y)\) to each sampled cell, we applied stochastic \(\alpha \in [0, 1]\) and \(\theta \in [0, 360]\) in spot \((x_0, y_0)\) to locate the cell into the space detailed in the following function: + +<|ref|>equation<|/ref|><|det|>[[369, 255, 627, 295]]<|/det|> +\[x = x_0 + \alpha d_{min}\cos (\theta \pi /180)\] \[y = y_0 + \alpha d_{min}\sin (\theta \pi /180)\] + +<|ref|>text<|/ref|><|det|>[[147, 308, 850, 398]]<|/det|> +where \(d_{min}\) represents the spatial distance of the closest neighbor spot. By integrating the optimal cellular combinations for all spots, ST data at single- cell resolution were reconstructed for the spot- based ST data. + +<|ref|>text<|/ref|><|det|>[[147, 411, 850, 748]]<|/det|> +Spatial LRI enrichment. To generate the cell- cell distance matrix \(D\) , the Euclidean distance between cells was calculated using the single- cell spatial coordinates of ST data. The KNN algorithm was then applied to each cell to select the \(K\) nearest cells from \(D\) to construct the cell graph network. For ligand \(i\) of the sender (cell type A) and receptor \(j\) of the receiver (cell type B), the number of LRI pairs \((P_{Ai,Bj}^{0})\) was obtained from the graph network by counting the 1- hop neighbor nodes of receivers for each sender. The permutation test was then performed by randomly shuffling cell labels to recalculate the number of LRI pairs. By repeating this step Z times, a background distribution \(P = \{P_{Ai,Bj}^{1}, P_{Ai,Bj}^{2}, \dots , P_{Ai,Bj}^{Z}\}\) was obtained for comparison with the real interacting score, and the \(P\) value was calculated as follows: + +<|ref|>equation<|/ref|><|det|>[[345, 763, 652, 785]]<|/det|> +\[P_{Ai,Bj} = \text{card}\{x \in P \mid x \geq P_{Ai,Bj}^{0}\} /Z\] + +<|ref|>text<|/ref|><|det|>[[147, 800, 850, 893]]<|/det|> +where \(P_{Ai,Bj}\) values less than 0.05 were filtered to calculate the intercellular score of LRI from senders to receivers \((S_{Ai,Bj}^{inter} = 1 - P_{Ai,Bj})\) . To further enrich the LRIs that activate downstream TFs, target genes, and the related pathways of receivers, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 854, 410]]<|/det|> +knowledge graph was introduced to model the intracellular signal propagation process. In practice, LRIs from CellTalkDB, pathways from KEGG and Reactome, and TFs from AnimalTFDB were integrated to construct the ligand- receptor- TF knowledge graph (LRT- KG), wherein the weight between entities represents the co- expressed coefficient. Taking the receptor as the query node, we incorporated the random- walk algorithm into the LRT- KG to filter and score the downstream activated \(t\) TFs with no more than 10 steps and Z iterations; thus, the probability \(p\) for each TF can be calculated with the ratio of successful hits from the query node to the target TF during the Z random walks. By integrating the co- expressed TFs and the corresponding target genes from the LRT- KG, the intracellular score of LRI from senders to receivers can be written as: + +<|ref|>equation<|/ref|><|det|>[[381, 430, 613, 469]]<|/det|> +\[S_{A i,B j}^{i n t r a} = \sum_{k = 1}^{t}\theta_{k}\times p_{k} / \eta_{k}\] + +<|ref|>text<|/ref|><|det|>[[144, 483, 851, 575]]<|/det|> +where \(\theta\) represents the number of targeted genes, \(\eta\) represents the step from the receptor to the TF in the LRT- KG. By the sigmoid transformation for \(S_{A i,B j}^{i n t r a}\) , the final score of the LRI from cell type A to cell type B can be written as: + +<|ref|>equation<|/ref|><|det|>[[390, 593, 605, 628]]<|/det|> +\[S_{A i,B j} = \sqrt{S_{A i,B j}^{i n t e r}\times S_{A i,B j}^{i n t r a}}\] + +<|ref|>text<|/ref|><|det|>[[144, 642, 852, 867]]<|/det|> +Comparison with other methods. STARmap, MERFISH, and seqFISH+ ST data were used to compare the performance of SpaTalk with other existing cell- type decomposition methods. For these single- cell ST data, all cells were split according to the fixed spatial distance and then merged into simulated spots as the benchmark datasets. RCTD, SPOTlight, Seurat, deconvSeq, and Stereoscope were benchmarked with the default parameters and evaluated with Pearson's correlation coefficient and RMSE over the predicted and real cell- type composition for each spot. + +<|ref|>text<|/ref|><|det|>[[181, 884, 849, 904]]<|/det|> +Given the limited genes of MERFISH ST data, STARmap and seqFISH+ single- cell ST + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 590]]<|/det|> +data (including 1020 and 10,000 genes, respectively) were used as the benchmark datasets to compare the performance of SpaTalk with other cell- cell communication inference methods (Giotto, SpaOTsc, NicheNet, CytoTalk, and CellCall). The one- sided Wilcoxon test was performed to evaluate the spatial proximity significance of the inferred LRIs by comparing the number of expressed LRIs between sender- receiver pairs and all cell- cell pairs, and the co- expressed percent of the LRI was calculated to evaluate the co- expression level by counting the number of expressed LRIs from senders to receivers from the cell- cell graph network. All methods were benchmarked with the default parameters and all inferred LRIs were unbiasedly evaluated with the above criteria except for the LRIs from SpaOTsc since the number of inferred LRIs was much larger than that of the other methods; thus, the top 1000 LRIs for each cell- cell communication were selected from SpaOTsc according to the final score. Given the significantly enriched biological processes or pathways in the receiver cell type, the Fisher exact test was adopted for pathway enrichment analysis with the KEGG and Reactome databases on the activated genes in receivers using the following function: + +<|ref|>equation<|/ref|><|det|>[[364, 604, 633, 643]]<|/det|> +\[P = \left(\frac{a + b}{a}\right)\left(\frac{c + d}{c}\right) / \left(\frac{n}{a + c}\right)\] + +<|ref|>text<|/ref|><|det|>[[144, 654, 852, 848]]<|/det|> +where \(n = a + b + c + d\) ; \(a\) is the number of inferred target genes that match a given pathway, \(b\) is the number of given pathway genes that exclude \(a\) , \(c\) is the number of inferred target genes that unmatch a given pathway, and \(d\) is the number of all genes excluding \(a\) , \(b\) , and \(c\) . NicheNet, CytoTalk, and CellCall were benchmarked with the default parameters and the inferred target genes for each LRI were evaluated according to the significance of pathway enrichment. + +<|ref|>text<|/ref|><|det|>[[144, 864, 850, 883]]<|/det|> +Pathway and biological process enrichment. The Metascape web tool + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 853, 384]]<|/det|> +(https://metascape.org/) was used to perform the enrichment analysis of pathways and biological processes, wherein the top 100 highly expressed genes were selected according to the fold change of the average gene expression. Gene Set Enrichment Analysis (GSEA) was performed using the ranked gene list with the clusterprofiler tool to enrich the significantly activated pathways and biological processes, whose signatures were obtained from the Molecular Signatures Database v7.4 (MSigDB, http://www.gsea- msigdb.org/gsea/msigdb), including the gene sets from Gene Ontology (GO) and the canonical pathway gene sets derived from the KEGG, Reactome, and WikiPathways pathway databases. + +<|ref|>text<|/ref|><|det|>[[144, 397, 853, 590]]<|/det|> +Module scoring of hallmarks and signatures. Hallmark scoring of metabolism of xenobiotics by cytochrome P450, synthesis of bile acids and bile salts, TSK, and EMT was performed using the "AddModuleScore" function in Seurat with default parameters. Hallmark pathways and EMT were obtained from MSigDB, and the signature genes of the TSK were download from the original publication by Jin et al. Statistics. R (version 4.1.1) and GraphPad Prism 8 were used for all statistical analyses. + +<|ref|>sub_title<|/ref|><|det|>[[147, 626, 404, 645]]<|/det|> +## Data and code availability + +<|ref|>text<|/ref|><|det|>[[145, 664, 852, 754]]<|/det|> +No new data was generated for this study. All data used in this study is publicly available as previously described. Source codes for the SpaTalk R package and the related scripts are available at github (https://github.com/ZJUFanLab/SpaTalk). + +<|ref|>sub_title<|/ref|><|det|>[[147, 789, 344, 808]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[145, 828, 852, 881]]<|/det|> +This work was supported by the National Natural Science Foundation of China (81973701), the Key Program, National Natural Science Foundation of China + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 850, 241]]<|/det|> +(81930016), the Key Research and Development Program of China (2021YFA1100500), the Major Research Plan of the National Natural Science Foundation of China (No.92159202), the Natural Science Foundation of Zhejiang Province (LZ20H290002), the China Postdoctoral Science Foundation (2021M702828) and Westlake Laboratory (Westlake Laboratory of Life Sciences and Biomedicine). + +<|ref|>sub_title<|/ref|><|det|>[[148, 278, 357, 297]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[144, 317, 850, 440]]<|/det|> +X.F., X.X., and H.C. conceived and designed the study. C.L., X.L., J.L., J.Q., and K.W. collected and analyzed the scRNA-seq and ST data. X.S., H.Y., J.C, and P.Y. implemented the algorithm of SpaTalk. X.S., C.L., and H.Y. developed the package of SpaTalk. All authors wrote the manuscript, read and approved the final manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[148, 476, 347, 495]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[147, 516, 511, 533]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[147, 570, 249, 587]]<|/det|> +## Reference + +<|ref|>text<|/ref|><|det|>[[144, 608, 852, 905]]<|/det|> +1. Shao X, Lu X, Liao J, Chen H, Fan X. New avenues for systematically inferring cell-cell communication: through single-cell transcriptomics data. Protein Cell 11, 866-880 (2020). +2. Armingol E, Officer A, Harismendy O, Lewis NE. Deciphering cell-cell interactions and communication from gene expression. Nat Rev Genet 22, 71-88 (2021). +3. Shao X, et al. scDeepSort: a pre-trained cell-type annotation method for single-cell transcriptomics using deep learning with a weighted graph neural network. Nucleic Acids Res 49, e122 (2021). +4. Cabello-Aguilar S, Alame M, Kon-Sun-Tack F, Fau C, Lacroix M, Colinge J. SingleCellSignalR: inference of intercellular networks from single-cell transcriptomics. Nucleic Acids Res 48, e55 (2020). +5. Jung S, Singh K, Del Sol A. FunRes: resolving tissue-specific functional cell states based on a cell-cell communication network model. Brief Bioinform 22, (2021). +6. Browaeys R, Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods 17, 159-162 (2020). +7. Hu Y, Peng T, Gao L, Tan K. CytoTalk: De novo construction of signal transduction networks using single-cell transcriptomic data. Sci Adv 7, (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 87, 854, 900]]<|/det|> +8. Zhang Y, et al. CellCall: integrating paired ligand-receptor and transcription factor activities for cell-cell communication. Nucleic Acids Res 49, 8520-8534 (2021).9. Zhang M, et al. Spatial molecular profiling: platforms, applications and analysis tools. Brief Bioinform 22, (2021).10. Lein E, Borm LE, Linnarsson S. The promise of spatial transcriptomics for neuroscience in the era of molecular cell typing. Science 358, 64-69 (2017).11. Liao J, Lu X, Shao X, Zhu L, Fan X. Uncovering an Organ's Molecular Architecture at Single-Cell Resolution by Spatially Resolved Transcriptomics. Trends Biotechnol 39, 43-58 (2021).12. Longo SK, Guo MG, Ji AL, Khavari PA. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat Rev Genet 22, 627-644 (2021).13. Moss A, et al. A single cell transcriptomics map of paracrine networks in the intrinsic cardiac nervous system. iScience 24, 102713 (2021).14. Stur E, et al. Spatially resolved transcriptomics of high-grade serous ovarian carcinoma. iScience 25, 103923 (2022).15. Almet AA, Cang Z, Jin S, Nie Q. The landscape of cell-cell communication through single-cell transcriptomics. Curr Opin Syst Biol 26, 12-23 (2021).16. Dries R, et al. Giotto: a toolbox for integrative analysis and visualization of spatial expression data. Genome Biol 22, 78 (2021).17. Cang Z, Nie Q. Inferring spatial and signaling relationships between cells from single cell transcriptomic data. Nat Commun 11, 2084 (2020).18. Mohamed SK, Nounu A, Novacek V. Biological applications of knowledge graph embedding models. Brief Bioinform 22, 1679-1693 (2021).19. Wang X, et al. Three-dimensional intact-tissue sequencing of single-cell transcriptional states. Science 361, (2018).20. Rodrigues SG, et al. Slide-seq: A scalable technology for measuring genome-wide expression at high spatial resolution. Science 363, 1463-1467 (2019).21. Stickels RR, et al. Highly sensitive spatial transcriptomics at near-cellular resolution with Slide-seqV2. Nat Biotechnol 39, 313-319 (2021).22. Ji AL, et al. Multimodal Analysis of Composition and Spatial Architecture in Human Squamous Cell Carcinoma. Cell 182, 497-514 e422 (2020).23. Pascual-Montano A, Carazo JM, Kochi K, Lehmann D, Pascual-Marqui RD. Nonsmooth nonnegative matrix factorization (nsNMF). IEEE Trans Pattern Anal Mach Intell 28, 403-415 (2006).24. Franc V, Hlaváč V, Navara M. Sequential Coordinate-Wise Algorithm for the Non-negative Least Squares Problem. In: Computer Analysis of Images and Patterns (eds Gagalowicz A, Philips W). Springer Berlin Heidelberg (2005).25. Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization. Nature 401, 788-791 (1999).26. Shao X, Liao J, Li C, Lu X, Cheng J, Fan X. CellTalkDB: a manually curated database of ligand-receptor interactions in humans and mice. Brief Bioinform 22, (2021).27. Hu H, Miao YR, Jia LH, Yu QY, Zhang Q, Guo AY. AnimalTFDB 3.0: a comprehensive resource for annotation and prediction of animal transcription factors. Nucleic Acids Res 47, D33-D38 (2019).28. Lee B, Zhang S, Poleksic A, Xie L. Heterogeneous Multi-Layered Network Model for Omics + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[201, 85, 636, 101]]<|/det|> +Data Integration and Analysis. Front Genet 10, 1381 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 104, 850, 139]]<|/det|> +Moffitt JR, et al. Molecular, spatial, and functional single- cell profiling of the hypothalamic preoptic region. Science 362, (2018). + +<|ref|>text<|/ref|><|det|>[[201, 142, 848, 177]]<|/det|> +Eng CL, et al. Transcriptome- scale super- resolved imaging in tissues by RNA seqFISH. Nature 568, 235- 239 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 180, 850, 214]]<|/det|> +Cable DM, et al. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol, (2021). + +<|ref|>text<|/ref|><|det|>[[201, 216, 848, 250]]<|/det|> +Stuart T, et al. Comprehensive Integration of Single- Cell Data. Cell 177, 1888- 1902 e1821 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 253, 850, 306]]<|/det|> +Elosua- Bayes M, Nieto P, Mereu E, Gut I, Heyn H. SPOTlight: seeded NMF regression to deconvolute spatial transcriptomics spots with single- cell transcriptomes. Nucleic Acids Res 49, e50 (2021). + +<|ref|>text<|/ref|><|det|>[[201, 308, 850, 343]]<|/det|> +Du R, Carey V, Weiss ST. deconvSeq: deconvolution of cell mixture distribution in sequencing data. Bioinformatics 35, 5095- 5102 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 345, 850, 380]]<|/det|> +Andersson A, et al. Single- cell and spatial transcriptomics enables probabilistic inference of cell type topography. Commun Biol 3, 565 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 382, 850, 417]]<|/det|> +Tasic B, et al. Shared and distinct transcriptomic cell types across neocortical areas. Nature 563, 72- 78 (2018). + +<|ref|>text<|/ref|><|det|>[[201, 419, 850, 472]]<|/det|> +Dippoppa M, Ranson A, Krumin M, Pachitariu M, Carandini M, Harris KD. Vision and Locomotion Shape the Interactions between Neuron Types in Mouse Visual Cortex. Neuron 98, 602- 615 e608 (2018). + +<|ref|>text<|/ref|><|det|>[[201, 475, 850, 527]]<|/det|> +Krieglstein K, Zheng F, Unsicker K, Alzheimer C. More than being protective: functional roles for TGF- beta/activin signaling pathways at central synapses. Trends Neurosci 34, 421- 429 (2011). + +<|ref|>text<|/ref|><|det|>[[201, 530, 848, 565]]<|/det|> +Allen NJ, Lyons DA. Glia as architects of central nervous system formation and function. Science 362, 181- 185 (2018). + +<|ref|>text<|/ref|><|det|>[[201, 567, 850, 638]]<|/det|> +Sun M, Gewirtz JC, Bofenkamp L, Wickham RJ, Ge H, O'Connor MB. Canonical TGF- beta signaling is required for the balance of excitatory/inhibitory transmission within the hippocampus and prepulse inhibition of acoustic startle. J Neurosci 30, 6025- 6035 (2010). + +<|ref|>text<|/ref|><|det|>[[201, 640, 848, 655]]<|/det|> +Nat Neurosci 19, 335- 346 (2016). + +<|ref|>text<|/ref|><|det|>[[201, 658, 848, 692]]<|/det|> +Han X, et al. Mapping the Mouse Cell Atlas by Microwell- Seq. Cell 172, 1091- 1107 e1017 (2018). + +<|ref|>text<|/ref|><|det|>[[201, 695, 850, 748]]<|/det|> +Pepe- Mooney BJ, et al. Single- Cell Analysis of the Liver Epithelium Reveals Dynamic Heterogeneity and an Essential Role for YAP in Homeostasis and Regeneration. Cell Stem Cell 25, 23- 38 e28 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 751, 850, 786]]<|/det|> +Shao X, Liao J, Lu X, Xue R, Ai N, Fan X. ssCATCH: Automatic Annotation on Cell Types of Clusters from Single- Cell RNA Sequencing Data. iScience 23, 100882 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 788, 850, 822]]<|/det|> +Yuan Z, et al. SEAM is a spatial single nuclear metabolomics method for dissecting tissue microenvironment. Nat Methods 18, 1223- 1232 (2021). + +<|ref|>text<|/ref|><|det|>[[201, 825, 850, 859]]<|/det|> +Ben- Moshe S, Itzkovitz S. Spatial heterogeneity in the mammalian liver. Nat Rev Gastroenterol Hepatol 16, 395- 410 (2019). + +<|ref|>text<|/ref|><|det|>[[201, 862, 850, 897]]<|/det|> +Margadant C, Sonnenberg A. Integrin- TGF- beta crosstalk in fibrosis, cancer and wound healing. EMBO Rep 11, 97- 105 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 85, 852, 252]]<|/det|> +805 48. Chen CH, Chen WY, Lin SF, Wong RJ. Epithelial-mesenchymal transition enhances response 806 to oncolytic herpesviral therapy through nectin-1. Hum Gene Ther 25, 539-551 (2014). 807 49. Wen S, et al. Cancer-associated fibroblast (CAF)-derived IL32 promotes breast cancer cell 808 invasion and metastasis via integrin beta3-p38 MAPK signalling. Cancer Lett 442, 320-332 809 (2019). 810 50. Asif PJ, Longobardi C, Hahne M, Medema JP. The Role of Cancer-Associated Fibroblasts in 811 Cancer Invasion and Metastasis. Cancers (Basel) 13, (2021). 812 51. Merritt CR, et al. Multiplex digital spatial profiling of proteins and RNA in fixed tissue. Nat 813 Biotechnol 38, 586-599 (2020). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 130, 316, 150]]<|/det|> +SupplementaryFigures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/images_list.json b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..63f9e634f5c9a3b528afb5a33b0433901e745152 --- /dev/null +++ b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Schematic illustration of peptide nano-blanket impedes fibroblasts activation and", + "footnote": [], + "bbox": [ + [ + 45, + 93, + 950, + 480 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. a, TEM images of FR17 or sFD17 treated with MMP2 (scale bar \\(= 200 \\mathrm{nm}\\) ), FG8 and FFKY (scale bar \\(= 100 \\mathrm{nm}\\) ). b, Size distribution of FR17 and sFD17 after MMP2 cleavage. c, Molecular structure of FFKY. d, The FFKY assemblies generated at \\(\\mathrm{t} = 200 \\mathrm{ns}\\) of MD simulation of 16 FFKY molecules in water (containing NaCl for charge neutralization). e, Typical molecular cluster and the non-bonded interactions involved in FFKY assemblies in detail. f, Molecular structure of FG8. g, The FG8 assemblies generated at \\(\\mathrm{t} = 200 \\mathrm{ns}\\) of MD simulation of 16 FG8 molecules in water. h, Typical molecular cluster and the non-bonded interactions involved in FG8 assemblies in detail. The dashed purple lines denote the \\(\\pi -\\pi\\) stacking. The dashed blue lines denote the hydrogen bonds. The nitrogen atoms are labeled in blue. And the oxygen atoms are labeled in red.", + "footnote": [], + "bbox": [ + [ + 152, + 90, + 839, + 504 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. FR17 interrupted the activation of fibroblast induced by tumor derived factors. a,", + "footnote": [], + "bbox": [ + [ + 50, + 100, + 936, + 485 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. FR17 administration protected fibroblasts from activation to inhibit vascular", + "footnote": [], + "bbox": [ + [ + 60, + 95, + 933, + 453 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. FR17 administration prevented MDSC recruitment, pausing the development of", + "footnote": [], + "bbox": [ + [ + 168, + 95, + 825, + 590 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. FR17 administration inhibited lung metastasis in vivo by retarding PMN", + "footnote": [], + "bbox": [ + [ + 63, + 98, + 933, + 428 + ] + ], + "page_idx": 25 + } +] \ No newline at end of file diff --git a/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806.mmd b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1725f8ec6692ddf7ecf1503f8b9fa578bb38c814 --- /dev/null +++ b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806.mmd @@ -0,0 +1,459 @@ + +# Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche + +Yi Zhou Zhejiang University Peng Ke Shengli Clinical Medical College of Fujian Medical University Yiyi Xia Zhejiang University Honghui Wu Zhejiang University Zhentao Zhang Zhejiang University Haiqing Zhong Zhejiang University Qi Dai Zhejiang University Tiantian Wang Zhejiang University Mengting Lin Zhejiang University Yaosheng Li Zhejiang University Xinchi Jiang Zhejiang University Qiyao Yang Zhejiang University Yiying Lu Zhejiang University Xincheng Zhong Zhejiang University Min Han Zhejiang University https://orcid.org/0000- 0001- 9373- 8466 Jianqing Gao ( \(\square\) gaojianqing@zju.edu.cn) Zhejiang University + +<--- Page Split ---> + +## Article + +Keywords: pre- metastatic niche, self- assembled peptide, pre- metastasis associated fibroblasts, 2 vascular endothelial cells, myeloid- derived suppressor cells, tumor metastasis + +Posted Date: August 10th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 699014/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on May 25th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 30634- 8. + +<--- Page Split ---> + +# Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche + +3 + +4 Yi Zhou', Peng \(\mathrm{Ke}^{1}\) 4, Yiyi Xia', Honghui Wu', Zhentao Zhang', Haiqing Zhong', Qi Dai' 5, Tiantian Wang', Mengting Lin', Yaosheng Li', Xinchi Jiang', Qiyao Yang' 5, Yiying Lu', Xincheng Zhong', Min Han' 2 3\\*, Jianqing Gao' 2 3\\* 7 1 Institute of Pharmaceutics, Zhejiang Province Key Laboratory of Anti- Cancer Drug Research, 8 College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, P.R. China 9 2 Cancer Center of Zhejiang University, Zhejiang University, Hangzhou 310058, P.R. China. 10 3 Hangzhou Institute of Innovative Medicine, Zhejiang University, Hangzhou 310058, P.R. China. 11 4 Shengli Clinical Medical College of Fujian Medical University, Fuzhou, 350001, China. 12 5 Department of Radiation Oncology, Key Laboratory of Cancer Prevention and Intervention, The 13 Second Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou 310058, China. 14 \\* Corresponding author: hanmin@zju.edu.cn (M. H.); gaojianqing@zju.edu.cn (J. G.) + +<--- Page Split ---> + +1 **Keywords:** pre-metastatic niche, self-assembled peptide, pre-metastasis associated fibroblasts, vascular endothelial cells, myeloid-derived suppressor cells, tumor metastasis + +## Abstract + +In various types of malignant tumors, metastasis is responsible for most of the tumor- induced death. Though emerging technologies provided early detection of tumor metastasis or even warning of high metastatic risk before the actual occurrence of metastasis, clinical treatment on metastasis prevention lags far behind. Evidences have illustrated that primary tumor induced pre- metastatic niche (PMN) formation in distal organs by producing pro- metastasis factors, spreading the spark to ignite the distal microenvironment. Given the fundamental role of PMN in the development of metastases, interruption of PMN formation would be a promising strategy to take early actions against tumor metastasis. Here we report an enzyme- activatable assembled peptide FR17 that can serve as a “flame- retarding blanket” at PMN site specifically to extinguish the “fire” of tumor- supportive microenvironment adaption. Our experiment demonstrated that the assembled peptide successfully reversed extracellular matrix deposition, vascular leakage and angiogenesis through inhibition on fibroblasts activation in PMN, which suppressed the remodeling of metastasis- supportive host stromal, and further prevented the recruitment of myeloid cells to PMN and then recovered the immunosuppressive microenvironment. Cell transcriptomic analysis of the pulmonary recruited MDSC suggested that FR17 intervention could regulate immune response activation, immune cells chemotaxis and migration pathways. Consequently, FR17 administration effectively inhibited pulmonary PMN formation and postoperative metastasis of melanoma, with only 30% lung- metastasis occurrence was observed for FR17 treated group at the time point when 100% occurrence was observed for the control + +<--- Page Split ---> + +group and \(80\%\) occurrence for anti- PD1 treated group, offering a robust therapeutic strategy against PMN establishing to prevent metastasis. + +## Introduction + +Though therapeutic outcomes and survival rate for patients with various cancers have been greatly improved in last decades, effective treatments for patients with metastatic cancer are still limited. Though emerging technologies provided early detection of malignant transformation or even warning of high metastatic risk by biomarker screening before the actual occurrence of metastasis in clinic. clinical treatment on metastasis prevention lags far behind. The contemporary therapeutic strategies against metastasis in clinic, including systemic chemotherapy, radiotherapy and immunotherapy, mainly focus on the later time period of metastasis development, or at least after the arrival and colonization of disseminated tumor cells to the distal organs, which have gained unsatisfied clinical outcomes. Observations of the adjuvant chemotherapy resistance and treatment- related toxicities revealed the shortcoming of the currently available strategies. + +Growing evidence illustrated that an inflammatory, neovascularized, immunosuppressive, tumor supportive microenvironment has emerged before the arrival and colonization of disseminated tumor cells, which is termed as pre- metastatic niche (PMN) formation. Relevant studies revealed that complex interactions between multiple participants and alteration in regulative pathways energized the construction of PMN, such as primary tumor- derived cytokines and exosomes, myeloid- derived suppressor cells (MDSCs), and the tumor re- educated stromal environment including pre- metastasis associated fibroblasts, destabilized vasculature and extracellular matrix (ECM) (Fig. 1a). + +<--- Page Split ---> + +1 Since PMN was considered as the foundation laid for circulating tumor cells colonization and one 2 of the vital premises to develop metastasis in distant organs, we wonder if the early process of 3 tumor metastasis can be terminated or even totally prevented by interrupting the formation of 4 PMN. To view the entire process of PMN establishment and metastasis development as a progress 5 of the occurrence and spread of a huge forest fire, it would be more efficient to contain and beat 6 out the local flame by preventing the formation of PMN. Here we report an enzyme- activatable 7 assembled peptide FR17 that can serve as a "flame- retarding blanket" at PMN site specifically, 8 containing and suppressing the "fire blaze" of PMN formation to further develop into overt 9 metastasis (Fig. 1b). As a substrate peptide of matrix metalloproteinase 2 (MMP2), FR17 can 10 release self- assembly monomer FG8 to construct peptide nano- blanket in PMN stromal 11 microenvironment, impeding fibroblasts activation so as to prevent metastatic cascades. We 12 further explored the subsequent impact of inhibition on fibroblasts activation induced by FR17 13 intervene, revealing the underlining mechanism on cellular interactions among fibroblasts, 14 vascular endothelial cells and extracellular components, and intervention on PMN recruited 15 MDSCs via in vitro and in vivo experiments. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Schematic illustration of peptide nano-blanket impedes fibroblasts activation and
+ +1 Figure 1. Schematic illustration of peptide nano- blanket impedes fibroblasts activation and subsequent formation of pre- metastatic niche. a, Illustration of the pathological process of pre- metastatic niche (PMN) formation. The primary tumor produces pro- metastasis factors, such as tumor- derived secreted factors (TDSF), to induce fibroblast activation in metastatic destination organs. The tumor- educated activated fibroblasts serve to construct a metastasis- supportive host stromal, including extra cellular matrix (ECM) deposition and reconstruction, angiogenesis and vascular leakage, as well as myeloid- derived suppressor cells (MDSC) recruitment. b, After FR17 subcutaneous administration, the in- situ assembled peptide nano- blanket in PMN stromal microenvironment impedes fibroblasts activation so as to retard stromal and vessel pro- metastatic reconstruction, inhibiting MDSC recruitment and metastatic cascades. + +<--- Page Split ---> + +## 1 Results and Discussion + +## Peptide nano-blanket transformed from FR17. + +The matrix metalloproteinase 2 (MMP2)- activatable self- assembled branched peptide (FR17, FFK(GPLGLAGG- YVDKR)Y) consists of (1) the backbone of a self- assembly peptide domain Phe- Phe- Lys- Tyr (FFKY), which is derived from \(\beta\) - amyloid (Aβ) peptide14, 15; (2) thymopentin (TP5, Arg- Lys- Asp- Val- Tyr, RKDVY), the pentapeptide with perfect hydrophilic property and immune modulation effect, which is applied as the adjunctive therapeutic agent on cancer treatment in clinic to prevent postoperative infection and to activate immune response16, 17. TP5 was conjugated to the side- chain of Lys (K) of the main chain FFKY with the MMP2- cleavable peptide linker (Gly- Pro- Leu- Gly- Leu- Ala- Gly- Gly, GPLGLAGG)18, increasing the hydrophilic property of the entire peptide molecule. Furthermore, the sequence of TP5 in peptide FR17 was replaced by the scrambled pentapeptide of TP5 without immunomodulatory bioactivity, i.e. DVVKR, to form the scrambled group (sFD17, FFK(GPLGLAGG- RKVYD)Y) as peptide assembly control. + +Peptides were synthesized via Fmoc solid- phase peptide synthesis technology and the peptide sequences were verified by mass spectra (Supplementary Fig. 1- 2). When specifically cleaved by MMP2, both FR17 and sFD17 are able to release self- assembled monomer FG8 (FFK(GPLG)Y), constructing peptide self- assemblies, the peptide nano- blanket. The transmission electron microscopy (TEM) images (Fig. 2a) and the Cryo- TEM image (Supplementary Fig. 3a) showed the lamellar structure of the peptide self- assemblies formed by enzymatic degradation (Supplementary Fig. 3b) with an average diameter of \(\sim 500 \mathrm{nm}\) (Fig. 2b). In addition, circular + +<--- Page Split ---> + +1 dichroism (CD) analysis revealed alterations in the secondary structure of peptide by the presence 2 of enzyme (Supplementary Fig. 3c). The peptide nano-blanket is woven by the self-assembled 3 monomer FG8 (FFK(GPLG)Y), which would spontaneously fold into lamellar structure rather 4 than the fiber clusters aggregated by FFKY as we previously reported14. The peptide assembly 5 relies on the non-bonded interactions, typically hydrogen bonds and \(\pi - \pi\) stacking, between 6 adjacent peptide molecules. The self-assembling pattern of FG8 changed due to the branch 7 modification with GPLG side-chain. The all-atom molecular dynamics (MD) simulation of 8 FFKY-assemblies and FG8-assemblies gives a microscopic account of the impacts of branch 9 modification on peptide interactions. The MD simulation revealed that there are more 10 intermolecular hydrogen bonds involved in FG8 cluster, which are more ordered and mostly 11 formed between Phe of adjacent FG8 molecules (Supplementary Fig. 4a-b). While the 12 intermolecular hydrogen bonds involved in FFKY assembly are less-formed and disordered (Fig. 13 2c-h). Besides, the aggregation-induced luminescence effect (AIE) was also employed to monitor 14 the spontaneous aggregation of FG8, the self-assembled monomer, in aqueous system 15 (Supplementary Fig. 4c-e). Taken together, these results indicated the self-assembly property of 16 FG8 monomer, and MMP2-cleaved release of FG8 from FR17 or sFD17 to form the peptide 17 nano-blanket. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. a, TEM images of FR17 or sFD17 treated with MMP2 (scale bar \(= 200 \mathrm{nm}\) ), FG8 and FFKY (scale bar \(= 100 \mathrm{nm}\) ). b, Size distribution of FR17 and sFD17 after MMP2 cleavage. c, Molecular structure of FFKY. d, The FFKY assemblies generated at \(\mathrm{t} = 200 \mathrm{ns}\) of MD simulation of 16 FFKY molecules in water (containing NaCl for charge neutralization). e, Typical molecular cluster and the non-bonded interactions involved in FFKY assemblies in detail. f, Molecular structure of FG8. g, The FG8 assemblies generated at \(\mathrm{t} = 200 \mathrm{ns}\) of MD simulation of 16 FG8 molecules in water. h, Typical molecular cluster and the non-bonded interactions involved in FG8 assemblies in detail. The dashed purple lines denote the \(\pi -\pi\) stacking. The dashed blue lines denote the hydrogen bonds. The nitrogen atoms are labeled in blue. And the oxygen atoms are labeled in red.
+ +<--- Page Split ---> + +# The pathological process in the MCM-induced PMN model in vivo + +In order to study the effect of FR17 administration on the process of PMN development, an in vivo PMN model induced by melanoma- conditioned media (MCM) has been established according to the previous research on PMN19. Briefly, tumor- derived factors secreted from primary tumor were replaced by MCM intraperitoneal injection to the mice for 10 consecutive days from Day 1 to 10. On Day 7, when the tumor supportive microenvironment was successfully established in the lung, B16F10 melanoma cells were intravenously administrated as the simulation of circulative tumor cells (CTC) wandering through blood vessels and some would successfully colonize in the prepared "fire scene" in lung (Supplementary Fig. 5b). + +The pathological process in the lung was assessed from various aspects during MCM- induced PMN establishment from Day 3 to Day 13 (Supplementary Fig. 5- 6). Important cellular derived molecular components and cytokines were closely monitored during the process. The higher expression of the extracellular matrix (ECM) component fibronectin (FN) was generated by MCM inducement. Moreover, matrix metalloproteinase 9 (MMP9), vascular endothelial growth factor a (VEGFa) were up- regulated along with the pathological progress of lung PMN from Day 0 to Day 10 and reached a plateau on Day 10 (Supplementary Fig. 5a). An accordant trend was found on MMP2 level as well (Supplementary Fig. 5d). Meanwhile, the immune cell population analysis during the pathological process in the lung of MCM- induced PMN model was conducted by flow cytometry, which drew our attention to the crime culprit- cell induced PMN, MDSC, for the recruitment of MDSC increased through the timeline (Supplementary Fig. 6- 7). It's reported that MDSC contributes a lot in developing the immunosuppression microenvironment in PMN, preparing more suitable and fertile land for tumor cells to take root in13, 20. Expression level of the + +<--- Page Split ---> + +major inflammatory mediator TGF- \(\beta 1\) , possibly produced by MDSCs, increased gradually but sharply. Typical biomarkers produced by MDSC to exert immuno- suppressive effect, in other words, the incriminating tools for the microenvironment re- education, were also detected, i.e. reactive oxygen species (ROS), iNOS and arginase- 1 (Arg- 1) (Supplementary Fig. 5a, c) 21. Major characteristics of PMN also emerged in lung as time went by, such as activation of fibroblasts (Supplementary Fig. 5e- f), angiogenesis (Supplementary Fig. 5g) and increase of vascular permeability (Supplementary Fig. 5h) 12. All these data indicated that the in vivo PMN model had been well- established, reflecting the pathological process of PMN development. With the overall support provided by PMN, it would consequently accelerate and aggravate metastasis in vivo (Supplementary Fig. 8). + +# FR17 administration interrupts the activation of fibroblast induced by tumor derived factors + +According to previous researches on tumor metastasis22 and our assessment on MCM- induced PMN mice model, the activation of the resident fibroblasts in distal tissues could be regarded as the tipping point of the beginning of PMN establishing, raising the alarm about laying the foundations of the potential metastasis. It's reported that tumor- educated fibroblasts would serve to construct a tumor supportive host stromal via TGF- \(\beta\) signaling23, promoting ECM degradation and reconstruction, inducing angiogenic and pro- inflammatory response of endothelial cells, recruiting VLA- 4+ bone marrow- derived cells (BMDCs) by localized FN deposition for niche formation19. Indeed, lysyl oxidase (LOX) cross- linked collagen surrounding pre- metastasis associated fibroblasts attracts CD11b+ myeloid cells invasion in destination organs, which corresponds to metastatic efficiency24. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. FR17 interrupted the activation of fibroblast induced by tumor derived factors. a,
+ +Schematic drawing illustrates the in- situ assembled peptide nano- blanket interrupts the activation of fibroblast. When activated by tumor derived factors during PMN development, the expression of proangiogenic factors and ECM remodeling factors would be up- regulated in fibroblast, as well as the ECM components production. While the peptide nano- blanket could calm down fibroblast activation, down- regulating the above factors. b, Schematic illustration to show the protocol of MCM stimulation and peptide treatment on mice lung fibroblast (MLF) in vitro. c, Cell proliferation of MLF after MCM stimulation and peptide treatment (n = 4). d, Secretion of MMP9, VEGF and fibronectin in the culture media of MLF after stimulated by MCM, treated with or without peptide. n = 4 for treated groups and 3 for the control group. e, qPCR analysis of Acta2 (i.e., αSma), Mmp9, Vegfa, Fibronectin1 (Fn1) expression in MLF after MCM stimulation + +<--- Page Split ---> + +1 and peptide treatment. f, Migration assay and collagen gel contract assay to evaluate MLF cellular functions after MCM stimulation and peptide treatment (n = 3). Scale bar = 100 μm. g, Expression level of fibronectin in the lung harvested from the PMN model mice administrated with different peptides on Day 10. h, Representative images and semi-quantification of \(\alpha \mathrm{SMA}+\) fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Data is presented as mean \(\pm\) SD. One-way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. Scale bar = 50 μm. + +9 Here in a cell model simulating the impact of secreted factors derived from primary tumor on lung fibroblasts in vitro (Fig. 3b), FR17 cultivation along with MCM stimulation showed an interruption effect on tumor derived factors irritating lung fibroblasts (Fig. 3a). The treatment with FR17 or sFD17, which would form peptide nano-blanket assembled with the monomer FG8 that was released by MMP2-induced cleavage, successfully prevented the activation of fibroblasts by MCM. The inhibition on the expression of pathologic fibroblast biomarker alpha smooth muscle actin ( \(\alpha \mathrm{SMA}\) ) as well as the fibroblasts proliferation and cellular bio-functions (Fig. 3c-f) also substantiated the blocking of lung fibroblasts activity by FR17 or the scrambled control sFD17. The reverse of the increase in cytokines secretion, like MMP9, VEGF, FN, was determined by ELISA kit (Fig. 3d). The qPCR results further suggested that the peptide assemblies might exert biological regulatory effect on cells while not just acting like a physical shield to wrap on the surface of cells (Fig. 3e). Besides, when irritated by tumor derived factors to present the activated \(\alpha \mathrm{SMA}^{+}\) phenotype, the migration ability as well as the collagen gel + +<--- Page Split ---> + +contracting function of fibroblasts got promoted. By contrast, cultivation with FR17 minified these functional changes to a large extent (Fig. 3f). Administration of FR17 to the in vivo PMN model also gave great relief to the activation of the fibroblasts in lung (Fig. 3h). What's more, the above data also suggested that the "flame-retarding" effect on fibroblast activation was almost completely contributed by peptide nano-blanket on account of the similar outcomes of the scrambled control sFD17 to that of FR17 treatment. + +The stromal microenvironment, apart from fibroblasts, consisting of ECM and vasculature, could also be re- educated by tumor derived factors25. For ECM would have gone through remodeling to rebuild a supportive niche for tumor colonization in PMN model, Western blot analysis, Masson staining and Sirius Red staining revealed the down- regulation in the expression of FN (Fig. 3g) and collagen (Supplementary Fig. 9a- c) in the lungs of FR17 group and sFD17 group, which would have likely been over- expressed by the activated fibroblasts otherwise. Besides, versican expression in the lung was also lower in the FR17 or sFD17 intervened groups than that of PMN control group and TP5 treated group (Supplementary Fig. 9d), and the elevated expression of which has been reported to contribute to angiogenesis in tumor26. + +FR17 protects fibroblasts from activation to inhibit vascular leakage and angiogenesis. + +The activated fibroblasts in primary tumor, also known as cancer- associated fibroblasts (CAFs), have been reported to produce proangiogenic factors, such as VEGF, so as to promote tumor angiogenesis27. What's more, the secretion of MMPs by CAFs induced tumor vascular leakage, exacerbating MDSCs and tumor cells intravasion into the vascular system. Therefore, we + +<--- Page Split ---> + +assumed that, as an important participator and vanguard in the construction of PMN, the activated fibroblast may also contribute to angiogenesis and the increasing vascular permeability in PMN (Fig. 4a). Further exploration was carried out on mice vascular endothelial cells to find out whether fibroblasts promote angiogenesis and vascular permeability during the arounement induced by MCM (as shown in Supplementary Fig. 10a). The tube forming assay demonstrated the proangiogenic capability of fibroblasts activated by MCM (Fig. 4b). Meanwhile, the transwell permeability assay (as illustrated in Supplementary Fig. 10b) to mimic the inner layer of blood vessel in vitro verified that the fibroblast-conditioned medium (FCM) collected from activated mice lung fibroblasts (MLF) increased vascular permeability (Fig. 4c), indicated directly by the dismission of endothelial adherence junctions mediated by vascular endothelial cadherin, VE-cadherin (Fig. 4d). + +For all experiments conducted on endothelial cells, the different conditional FCM was obtained from fibroblasts pre- treated with FR17, sFD17 or TP5 on interrupting MCM stimulation separately. The conditional FCM pre- treated with FR17 or sFD17, rather than TP5, offset the increase in bEnd3 cell proliferation induced by the indirect stimulation of MCM (Supplementary Fig. 10c). The acceleration in neovascularization and the disruption on endothelial cell- cell connection to cause vascular leakage were made up by indirect FR17 or sFD17 treatment on MLF as indicated in Fig. 4b- c. + +To examine the influence of FR17 intervention on the expression level of the proangiogenic factors and vascular remodeling enzymes in pulmonary PMN in general, Western blot analysis was carried out. When compared to PMN control group, the increased expression of VEGFa, + +<--- Page Split ---> + +1 ANG2, MMP9, MMP2 was reversed by FR17 treatment almost back to normal levels (Fig. 4e, 2 S11). In addition, the scrambled control sFD17, which doesn't possess the drug-bioactivity of 3 TP5, exhibited similar inhibition result as FR17 while TP5 didn't, suggesting the protective 4 effect was contributed by the peptide assemblies formed by the enzyme-activatable self-assemble 5 monomer. Vascular leakage assay on PMN model in vivo verified that FR17 or sFD17 6 administration attenuated the enhancement of pulmonary vascular permeability (Fig. 4f-g). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. FR17 administration protected fibroblasts from activation to inhibit vascular
+ +leakage and angiogenesis. a, Illustration of the inhibition of vascular leakage and angiogenesis via the peptide nano- blanket protection on fibroblasts from tumor re- education. b, Tube forming assay was performed on bEnd3 cells which were treated with conditional FCM collected from MLF pre- stimulated with MCM and peptide. Scale bar \(= 100 \mu \mathrm{m}\) . c, Permeability of the endothelial cell layer in vitro when co- cultured with MLF pre- treated with MCM and peptide. Data is presented as mean \(\pm\) SD. n = 3. d, Integrality of the endothelial cell monolayer after cultivated with conditional FCM collected from MCM and peptide pre- stimulated MLF, indicated by VE- cadherin on the membrane (n = 3). The white box and the white arrows in the enlarged images indicate the tight junction between the endothelial cells. While the yellow box and yellow arrows indicate the disruption of cell- cell connection. Scale bar \(= 50 \mu \mathrm{m}\) in the upper panel. Scale bar \(= 30 \mu \mathrm{m}\) in the lower panel. e, Expression level of VEGFa, ANG2, MMP9 and + +<--- Page Split ---> + +MMP2 in the pulmonary PMN harvested from mice administrated with different peptides on Day 10. f & g, Vascular permeability of the pulmonary PMN after peptide administration. The semi-quantification was calculated from 5 random visual fields for each section taken from \(\mathrm{n} = 3\) biologically independent mice. Data is presented as mean \(\pm\) SD. Scale bar \(= 100 \mu \mathrm{m}\) . One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. + +Above data confirmed that FR17 can be employed as an in- situ spontaneously- assembled "flame- retarding blanket" to beat out the "flames" on fibroblasts caused by tumor derived factors in PMN, preventing pro- metastatic angiogenesis and vascular destabilization. + +# FR17 administration impedes MDSC recruitment to pulmonary PMN and modulates the + +# immuno-microenvironment. + +Given the fact that MDSC occupies a vital position in PMN construction and metastasis formation, MDSC has attracted wide concerns in recent years. It's firstly reported by David Lyden's team that VEGFR+ myeloid progenitor cells are recruited and formed cell clusters to the sites highly expressing FN, which is most likely produced by resident fibroblasts, in the distal tissues prior to the arrival of tumor cells19. The recruitment of MDSC was found to be related to the enhancement of ECM production and remodeling in PMN, including cross- linking collagen by LOX secreted by tumor cells24, FN deposition28, periostin (POSTN)- enrichment29. As MDSC plays a crucial role in developing immunosuppressive and inflammatory microenvironment, + +<--- Page Split ---> + +some pioneers have demonstrated that blocking the recruitment of MDSC could be a promising strategy to suppress early metastasis by preventing the development of breeding ground for tumor30- 33. + +In our PMN model, the accumulation of MDSC in lung was decreased in FR17 and sFD17 groups, revealed by immunofluorescent staining (Fig. 5a) and flow cytometry analysis on Day 10 (Fig. 5b). Moreover, focal enrichment areas of FN and the co-localization of LOX attracted an increasing number of CD11b+Gr+ MDSC in serial sections of the PMN lung comparing to healthy lung (Fig. 5a, Supplementary Fig. 12), consistent with the previous reports24. The immunofluorescent stains indicated that FR17 intervention successfully down-regulated the expression of LOX, FN and POSTN in the lung induced by tumor-derived factors to impede the construction of PMN, which probably resulted in the cut in MDSC recruitment (Supplementary Fig. 12). To evaluate prevention of peptide administration on developing PMN immunosuppressive environment, protein level of TGF- \(\beta\) in PMN (Fig. 5c), the well-known immuno-modulator produced by MDSC to regulate the establishment of immunosuppressive tumor supportive niche, as well as the pro-inflammatory cytokine IL-6 was detected (Fig. 5d). FR17 treatment successfully down-regulated TGF- \(\beta\) and IL-6 expression in lung, so as sFD17. In the meantime, TP5 administration exhibited partial abatement on the expression of these cytokines34, suggesting both the enzyme-responsive assembled peptide nano-blanket and the immunoregulatory agent TP5 facilitated the normalization of immunosuppressive and inflammatory environment of PMN. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. FR17 administration prevented MDSC recruitment, pausing the development of
+ +the immunosuppressive microenvironment in PMN. a, Serial sections of the lungs harvested from the model mice treated with different peptides. Serial sections show the distribution of lysyl oxidase (LOX) and Fibronectin (FN) and the co- location of CD11b+Gr1+ myeloid derived suppressor cells (MDSC). b, Recruitment of CD11b+Ly6g+ MDSC to the lungs of the model mice treated with different peptides on Day 10. Data is presented as mean ± SD. n = 4 biologically independent mice for treatment groups, n = 3 for the control group. c, Expression of TGF- \(\beta 1\) in the lung harvested from PMN model mice administrated with different peptides. d, + +<--- Page Split ---> + +1 IL- 6 expression in the lung tissue fluid collected from PMN model mice administrated with different peptides. Data is presented as mean \(\pm\) SD. \(\mathrm{n} = 3\) biologically independent mice. 3 One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. e, GO enrichment analysis of CD11b \(^+\) Ly6g \(^+\) MDSC sorted from different treatment groups. RNA preparations were extracted from CD11b \(^+\) Ly6g \(^+\) MDSCs sorted from lungs pooled from 10- 12 mice per sample. The size of the dots corresponds to the number of genes per pathway, and the color indicates \(p\) - value. Statistical significance was considered at least at \(p<\) 0.05. + +10 To find out whether the impact on MDSC- induced immunosuppressive microenvironment in PMN was mainly contributed by TP5 or by the in- situ assembled peptide nano- blanket, cell transcriptomic analysis of MDSCs recruited to the lung was carried out on different treatment groups. Given the fact that CD11b \(^+\) Ly6g \(^+\) MDSC takes the majority (over \(90\%\) ) of MDSC population in the lung of MCM- induced PMN model (Supplementary Fig. 6a- b), CD11b \(^+\) Ly6g \(^+\) MDSC were sorted on Day 10 from pulmonary PMN after different treatments as the representative subset for further transcription analysis (Supplementary Fig. 13). GO enrichment analysis indicated that the regulation effect of FR17 might relate to the activation of immune response pathway, cytokine production involved in immune response, regulation of leukocyte and lymphocyte activation, leukocyte chemotaxis and migration (Fig. 5e). Relative enriched pathways provided possible comments on the underlined mechanisms, including: regulation on cytokine biosynthetic process, adaptive immune response based on somatic recombination + +<--- Page Split ---> + +1 immune receptors built from immunoglobulin super-family domains, interferon- \(\gamma\) - mediated 2 signaling pathway and the impact on chemokine receptor binding as well as CXCR chemokine 3 receptor binding (Supplementary Fig. 14). From above, when compared with the differential 4 gene enrichment pathway of sFD17 and TP5, we found that the regulation on immune cells 5 chemotaxis and migration pathways was contributed by the in- situ peptide assemblies, for the 6 peptide nano- blanket would only form in FR17 or sFD17 treated lung while not in free TP5 7 treated group. And the Venn diagram and further enrichment analysis to put TP5 aside suggested 8 these would correspond to the regulation on cell surface receptor signaling pathway by peptide 9 assemblies (Supplementary Fig. 14b- c). In addition, the regulation on leukocyte differentiation 10 and T cell activation pathway was mainly contributed by TP5 (Supplementary Fig. 14d), which 11 has been commonly accepted as one of the mechanisms for TP5 to exert immune regulation 12 effect35-37. + +Moreover, the in- situ assembled peptide nano- blanket formed by FR17 inhibited tumor cells migration as well, with hardly any effect on cell viability (Supplementary Fig. 15). + +## FR17 administration inhibits melanoma lung metastasis in vivo. + +First, we evaluated the anti- metastasis efficacy of FR17 on MCM- induced lung metastasis model. C57/BL6 mice were pre- induced by MCM administration for 10 days to set up the tumor supportive pre- metastatic niche in lung. Tumor cells were then injected through the tail vein on Day 7 to simulate the circulative tumor cells (CTC) wandering through blood vessels and eventually settled down in pulmonary PMN. As illustrated in Fig. 6a, peptide subcutaneous + +<--- Page Split ---> + +1 administration started from Day 3 and ended at 2 weeks- post lung tumor inoculation. The 2 metastasis inhibition efficacy was evaluated in terms of the tumor module number on Day 28. 3 FR17 effectively suppressed tumorigenesis and the development of metastasis in lung, compared 4 to both the control and TP5 groups (Fig. 6b- d). Moreover, the good outcomes of sFD17 treatment 5 which was close to that of FR17 group emphasized the main role of nano- blanket assembled by 6 the enzyme- activated released monomer in intervening PMN construction. Therefore, the in- situ 7 assembled nano- blanket also played a part in inhibiting metastasis development and growth. 8 Preliminary safety evaluation on these model mice at the end of the experiment revealed no 9 severe safety concern of FR17 administration (Supplementary Fig. 16- 17). The systemic immune 10 modulation effect of TP5 was reflected on the recovery of thymus shrinking as well 11 (Supplementary Fig. 16d, 17). What's more, peptide administration prevented tumor cells 12 infiltration into the spleen (Supplementary Fig. 17). + +13 The anti- metastasis activity of FR17 was further verified on a post- surgery metastasis model30 14 (illustrated by Fig. 6e), which is closer to the clinical treatment of resectable melanoma 15 followed- up with adjuvant therapy for patients at high risk of recurrence. Here we compared 16 FR17 therapy with the rising- star of checkpoint inhibitors, PD1 antibody38- 39, which was 17 approved by FDA as adjuvant therapy on primary tumor excised with lymph node management 18 or metastatic melanoma40. As demonstrated in Fig. 6f, the recurrence of lung metastasis was put 19 off and the overall survival was greatly improved by FR17 therapy when compared with control 20 (Fig. 6g- i). The median overall survival prolonged for 23.3% (from 30 days to 37 days), catching 21 up with the outcomes of anti- PD1 treatment (from 30 days to 36 days). And the median + +<--- Page Split ---> + +1 lung-metastasis-free survival prolonged from less than 19 days of both control group and 2 anti-PD1 treated group to 23 days of FR17 treated group. At the first time- point for lung 3 metastasis monitoring via bioluminescence imaging, lung metastasis has been observed in all 4 of the 10 mice from the control group, 8 mice from anti-PD1 treated group while only 3 mice 5 from FR17 treated group on Day 19. The retard of the occurrence of lung metastasis after FR17 6 administration emphasized the importance of PMN intervention. What's more, the relapse rate 7 of the primary tumor post-resection was reduced from 10 / 10 in the control group to 4 / 10 in 8 FR17 treated group (Supplementary Fig. 18). + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. FR17 administration inhibited lung metastasis in vivo by retarding PMN
+ +formation. a, Schedule of MCM- induced lung metastasis model with peptide treatment. Peptide administration started from Day 3 for metastasis prevention. b, Number of the metastatic nodules in the lung of different treatment groups (n = 7). Data is presented as mean ± SD. One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. c, Images of the lung metastasis harvested on Day 28 from different treatment groups. d, Representative images of Hematoxylin & Eosin staining of lung sections from different treatment groups. The lung metastasis is circled by yellow dotted lines. Scale bar = 200 μm. e, Schedule of post- surgery metastasis model. Peptide administration started from Day 7 to 25. f, Representative in vivo bioluminescent images of mice without treatment or treated with FR17 or anti- PD1 (n = 10). The grey patches represent dead mice in the control group. g, Survival curves of the mice treated with FR17 or anti- PD1 or without treatment (n = 10). h, The cumulative + +<--- Page Split ---> + +incidence of new pulmonary metastases in the mice treated with FR17 or anti- PD1 or without treatment \((\mathrm{n} = 10)\) . i, Semi- quantification of the in vivo bioluminescent signals in the lungs of the mice treated with FR17 or anti- PD1 or without treatment \((\mathrm{n} = 5)\) . Data is presented as mean ± SD. + +## 6 Conclusions + +In summary, we have explored the magical retarding effect of the PMN microenvironment responsive- assembled peptide nano- blanket on fibroblasts activation, impeding PMN development. The enzyme- activatable assembled peptide FR17 can be enzyme- cleaved to release the self- assembly monomer FG8 to construct a lamellar structure, which is named as peptide nano- blanket. Experiments demonstrated that FR17 administration not only beat out the "flame" set up on resident fibroblasts induced by tumor derived factors, but also interrupted the subsequent PMN formation, including preventing pro- metastatic angiogenesis and vascular destabilization, and then intervening MDSCs' recruitment as well as their bio- functions. Astonishingly, when treated with sFD17, tumor- induced fibroblast activation was also impeded by peptide- assemblies without the assistance of TP5 so as to arrest the following pro- metastatic pathological process. This finding illustrated that the "flame- retarding" effect on fibroblast activation could be contributed by the drug- free peptide nano- blanket alone, presenting a broad application prospect of drug- free peptide assemblies to regulate PMN microenvironment and to prevent tumor distant metastasis. This work also elucidated the important role of pre- metastasis associated fibroblast in the complex interactions between major participators in PMN formation + +<--- Page Split ---> + +1 and metastasis development, suggesting that reprogramming or intervention on the key juncture could make a big difference on fighting against tumor metastasis. + +## 4 References + +5 1. Gdowski, A. S., Ranjan, A. & Vishwanatha, J. K. Current concepts in bone metastasis, contemporary therapeutic strategies and ongoing clinical trials. J. Exp. Clin. Cancer Res. 36, 108-121 (2017). 8 2. Pashayan, N. & Paul, P. D. P. The challenge of early detection in cancer. Science (New York, N.Y.) 368, 589-590 (2020). 9 3. Rodriguez-Ruiz, A. et al. Stand-alone artificial intelligence for breast cancer detection in mammography: Comparison with 101 radiologists. J. Natl. Cancer Inst. 111, 916-922 (2019). 4. Poudineh, M., Sargent, E. H., Pantel, K., Kelley, S. O. Profiling circulating tumour cells and other biomarkers of invasive cancers. Nat. Biomed. Eng. 2, 72-84 (2018). 5. Christensen, E. et al. Early detection of metastatic relapse and monitoring of therapeutic efficacy by ultra-deep sequencing of plasma cell-free DNA in patients with urothelial bladder carcinoma. J. Clin. Oncol. 37, 1547-1557 (2019). 6. Abbosh, C. et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. Nature 545, 446-451 (2017). 7. Curti, B. D. & Faries, M. B. Recent advances in the treatment of melanoma. N. Engl. J. Med. 384, 2229-2240 (2021). + +<--- Page Split ---> + +8. Oliver Sartor, M. D. & Johann S. de Bono, M. B. Metastatic prostate cancer. N. Engl. J. Med. 378, 645-657 (2018). +9. Zhang, C. et al. Radiotherapy and cytokine storm: Risk and mechanism. Front. Onco. 11, 670464 (2021). +10. Cardoso, F. et al. 70-Gene signature as an aid to treatment decisions in early-stage breast cancer. N. Engl. J. Med. 375, 717-29 (2016). +11. Peinado, H. et al. Pre-metastatic niches: organ-specific homes for metastases. Nat. Rev. Cancer 17, 302-317 (2017). +12. Liu, Y. & Cao, X. Characteristics and significance of the pre-metastatic niche. Cancer Cell 30, 668-681 (2016). +13. Zhou, Y., Han, M., & Gao, J. Q. Prognosis and targeting of pre-metastatic niche. J. Control. Release 325, 223-234 (2020). +14. Wang, T. T. et al. AIE/FRET-based versatile PEG-Pep-TPE/DOX nanoparticles for cancer therapy and real-time drug release monitoring. Biomater. Sci. 8, 118-124 (2020). +15. Guo, W. W. et al. Intracellular restructured reduced glutathione-responsive peptide nanofibers for synergetic tumor chemotherapy. Biomacromolecules 21, 444-453 (2020). +16. Zeng, F. L. et al. Clinical efficacy and safety of synthetic thymic peptides with chemotherapy for non-small cell lung cancer in China: A systematic review and meta-analysis of 27 randomized controlled trials following the PRISMA guidelines. Int. Immunopharmacol. 75, 105747 (2019). +17. Li, S. et al. Supramolecular nanofibrils formed by coassembly of clinically approved drugs for tumor photothermal immunotherapy. Adv. Mater. 33, e2100595 (2021). + +<--- Page Split ---> + +18. Qin, S. Y. et al. Theranostic GO-based nanohybrid for tumor induced imaging and potential combinational tumor therapy. Small 10, 599-608 (2014). +19. Kaplan, R. N. et al. VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche. Nature 438, 820-827 (2005). +20. Engblom, C., Pfirschke C. & Pittet M. J. The role of myeloid cells in cancer therapies. Nat. Rev. Cancer 16, 447-462 (2016). +21. Gabrilovich, D. I. Myeloid-derived suppressor cells. Cancer Immunol. Res. 5, 3-8 (2017). +22. Pein, M. et al. Metastasis-initiating cells induce and exploit a fibroblast niche to fuel malignant colonization of the lungs. Nat. Commun. 11, 1494 (2020). +23. Kong, J. et al. Extracellular vesicles of carcinoma-associated fibroblasts creates a pre-metastatic niche in the lung through activating fibroblasts. Molecular Cancer 18, 175 (2019). +24. Erler, J. T. et al. Hypoxia-induced lysyl oxidase is a critical mediator of bone marrow cell recruitment to form the premetastatic niche. Cancer Cell 15, 35-44 (2009). +25. Paolillo, M. & Schinelli, S. Extracellular matrix alterations in metastatic processes. Int. J. Mol. Sci. 20, 4947 (2019). +26. Asano, K. et al. Stromal versican regulates tumor growth by promoting angiogenesis. Sci. Rep. 7, 17225 (2017). +27. Zhou, X. et al. Melanoma cell-secreted exosomal miR-155-5p induce proangiogenic switch of cancer-associated fibroblasts via SOCS1/JAK2/STAT3 signaling pathway. J. Exp. Clin. Cancer Res. 37, 242 (2018). +28. Murgai, M. et al. KLF4-dependent perivascular cell plasticity mediates pre-metastatic niche + +<--- Page Split ---> + +1 formation and metastasis. Nat. Med. 23, 1176- 1190 (2017). 2 29. Wang, Z. et al. Periostin promotes immunosuppressive premetastatic niche formation to facilitate breast tumour metastasis. J. Pathol. 239, 484- 495 (2016). 4 30. Long, Y. et al. Self- delivery micellar nanoparticles prevent premetastatic niche formation by interfering with the early recruitment and vascular destruction of granulocytic myeloid- derived suppressor cells. Nano Lett. 20, 2219- 2229 (2020). 7 31. Jiang, T. et al. Metformin and Docosahexaenoic Acid Hybrid Micelles for Premetastatic Niche Modulation and Tumor Metastasis Suppression. Nano lett. 19, 3548- 3562 (2019). 9 32. Lu, Z et al. Epigenetic therapy inhibits metastases by disrupting premetastatic niches. Nature 579, 284- 290 (2020). 11 33. Kaczanowska, S. et al. Genetically engineered myeloid cells rebalance the core immune suppression program in metastasis. Cell 184, 2033- 2052 (2021). 13 34. Lunin, S. M. et al. Thymic peptides restrain the inflammatory response in mice with experimental autoimmune encephalomyelitis. Immunobiology 218, 402- 7 (2013). 15 35. Cascinelli, N. et al. Evaluation of clinical efficacy and tolerability of intravenous high dose thymopentin in advanced melanoma patients. Melanoma Res. 8, 83- 9 (1998). 17 36. Wang, Y. et al. The novel role of thymopentin in induction of maturation of bone marrow dendritic cells (BMDCs). Int. Immunopharmacol. 21, 255- 60 (2014). 19 37. Lunin, S. M. et al. Thymus peptides regulate activity of RAW 264.7 macrophage cells: inhibitory analysis and a role of signal cascades. Expert Opin. Ther. Targets 15, 1337- 46 (2011). 22 38. Rizvi, N. A. et al. Activity and safety of nivolumab, an anti- PD- 1 immune checkpoint + +<--- Page Split ---> + +1 inhibitor, for patients with advanced, refractory squamous non-small-cell lung cancer2 (CheckMate 063): a phase 2, single-arm trial. Lancet Oncol. 16, 257- 65 (2015).3 39. Gong, N. et al. Proton-driven transformable nanovaccine for cancer immunotherapy. Nat. Nanotechnol. 15, 1053- 1064 (2020).4 40. NIH. National Cancer Institute, Melanoma Treatment (PDQ#)- Health Professional Version.6 Available from: https://www.cancer.gov/types/skin/hp/melanoma-treatment-pdq#_402. + +<--- Page Split ---> + +## Acknowledgements + +This work was supported by the National Natural Science Foundation of China (Nos. 81673022). We thank Professor Yu Kang at College of Pharmaceutical Sciences, Zhejiang University for guidance on molecular dynamics simulation. We thank Qichun Wei's lab for providing the luciferase transfected B16F10. We thank Qin Han, Chenyu Yang, and Dandan Song at the Center of Cryo- Electron Microscopy (CCEM), Zhejiang University for their technical assistance on Confocal laser scanning microscopy and transmission electron microscopy. We thank Yanwei Li at the Core Facilities of Zhejiang University School of Medicine for technical assistance in flow cytometry analysis. + +## Author Contributions + +M.H. and Y.Z. conceived the project and designed the experiments. M.H. and J.G. supervised the project, discussed and commented on the manuscript. Y.Z. performed the majority of the experiments and data analysis. P.K. assisted with cell culture of B16F10, preparing MCM and animal experiments. Y.X. synthesized and characterized TPE- FR17 and TPE- FFKY. H.W. assisted with the establishment of animal models. Y.Z. performed the flow cytometry experiments with assistance from Z.Z., T.W., M.L., P.K., Y.S.L., Y.X., H.Z., X.Z. and Q.Y.; Y.Y.L. supported the operation of confocal microscope. Y.Z., Q.D., P.K., H.Z., Y.X., Y.S.L., X.J. and H.W. operated the tumor resection surgery. Y.Z. wrote the manuscript. All authors discussed the results and commented on the manuscript. + +<--- Page Split ---> + +1 Competing Interests statement + +2 All the authors declare no conflicting interests. + +<--- Page Split ---> + +## Methods + +Characterization of FR17 and sFD17. Peptide FR17 and sFD17 were synthesized via Fmoc solid- phase peptide synthesis technology by APeptide Shanghai. The molecule structure and amino acid sequence of peptide were confirmed by mass spectra. + +5 + +Enzyme induced assembly of FR17 and sFD17. Peptide FR17 or sFD17 was dissolved in HBS + +buffer (pH 7.4, containing \(50~\mathrm{mM}\) HEPES, \(150~\mathrm{mM}\) NaCl, \(10~\mathrm{mM}\) \(\mathrm{CaCl_2}\) ) at \(500~\mu \mathrm{M}\) . Activated + +hMMP2 was added to the peptide solution at the working concentration of \(1\mu \mathrm{g / mL}\) . The + +solutions were then incubated in \(37^{\circ}\mathrm{C}\) air- bath for \(24\mathrm{h}\) . The size of the enzyme treated sample, + +which has formed the peptide nano- blanket, was measured by Zetasizer Nano ZS (Malvern + +Instruments). The assembly morphology of the assembled nano- blanket was observed with + +transmission electron microscope (FEI Tecnai G2 spirit) for TEM image and \(200\mathrm{kv}\) transmission + +electron microscope (FEI Talos F200C) for Cryo- TEM image. + +14 + +All- atom molecular dynamics simulation. The molecular structures of peptide FFKY and FG8 + +monomer were first constructed via AMBERTOOL based on the AMBER14SB force field. + +Dynamics simulation runs were performed utilizing Gromacs 2018.4 package1. System + +configurations were visualized using VMD software2, and generated into images mainly + +employing GRACE software. The simulation was performed in water boxes containing 16 FFKY + +or FG8 molecules. FFKY system was simulated containing NaCl to neutralize electric charge of + +<--- Page Split ---> + +the amidogen on the side- chain if Lys. Energy was minimized according to the steepest- descent method. Bond lengths were constrained by the LINCS algorithms. The nonbonded LJ interactions were cut off at 1.2 nm. Electrostatics was treated utilizing the Particle Mesh Ewald (PME) scheme. All production runs were simulated in the NPT ensemble using V- rescale coupling scheme with the temperature maintained at 298.15 K and parrinello- rahman coupling scheme with pressure kept at 1.0 bar and isotropic coupling type. The time constants for the pressure and temperature couplings were respectively set to 2.0 and 0.2 ps. Besides, the compressibility value was \(4.5 \times 10^{- 5} \mathrm{bar}^{- 1}\) . Periodic boundary conditions with a time step of 0.002 ps were adopted. Simulations were carried out for 200 ns and the structural coordinate information was recorded per 50 ps. + +Cell lines and cell culture. B16F10 cells were purchased from Cell Bank of Chinese Academy of sciences (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium (DMEM, Cienry, China) containing \(10\%\) (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution ( \(100 \times\) , TBD, Tianjin, China) at \(1\%\) (v/v). And the luciferase transfected B16F10 (Luc- B16F10) was kindly gifted by Qichun Wei's lab, the Second Affiliated Hospital, Zhejiang University. The mouse lung fibroblasts (MLF) were purchased from iCell Bioscience Inc. (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium/Nutrient Mixture F- 12, 1:1 mixture (DMEM/F12 medium, Multicell, Wisent Int., Canada) containing \(10\%\) (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution ( \(100 \times\) , TBD, Tianjin, China) at \(1\%\) (v/v). The mouse endothelial cells bEnd3 was kindly gifted + +<--- Page Split ---> + +by Fuqiang Hu's lab, Institution of Pharmaceutics, College of Pharmaceutical Sciences, Zhejiang University. The bEnd3 cells were cultured in the same medium as B16F10. Cells were cultured in a cell incubator containing \(5\% \mathrm{CO}_2\) at \(37^{\circ}\mathrm{C}\) and passaged when reached \(80\% - 90\%\) confluence. The melanoma-conditioned medium (MCM) was obtained as follows: when B16F10 reached \(70 - 80\%\) confluence, washed with PBS and changed the medium to serum-free medium and incubated for \(24\mathrm{h}\) . The cell supernatants were collected, centrifuged at \(2000\mathrm{rpm}\) for \(10\mathrm{min}\) to discard the cell debris. The MLF-conditioned medium (FCM) was obtained as follows: MLFs were stimulated by MCM (supplemented with \(10\%\) (v/v) FBS) with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100\mu \mathrm{M}\) ) for \(48\mathrm{h}\) , then replaced with fresh complete medium and incubate for another \(24\mathrm{h}\) . The cell supernatants were collected, centrifuged at \(2000\mathrm{rpm}\) for \(10\mathrm{min}\) to discard the cell debris. + +12 + +Cell proliferation assays. MLFs in rapid proliferation were plated in 96- well plates at the density of \(4\times 10^{3}\) per well and cultured overnight. Former media were removed and the cells were cultivated with \(100~\mu \mathrm{L}\) MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100~\mu \mathrm{M}\) ). Cells cultivated with fresh complete medium was set up as control. After incubation for \(48\mathrm{h}\) , cell proliferation was measured using a Cell Counting Kit- 8 assay (Cat. 13E02A60, Boster Biotech., China) according to the manufacturer's instructions. The optical density at \(450\mathrm{nm}\) ( \(\mathrm{OD}_{450}\mathrm{nm}\) ) was measured using a multiwell plate reader (ELX800, BioTek, USA). Each group was repeated at least 4 times, and cell proliferation was presented as mean \(\pm\) SD. + +<--- Page Split ---> + +2 Cytokines secretion and gene expression of MLFs. An equal number of MLFs in rapid proliferation were seeded in a 24-well plate per well, cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in at least triplicate. Cells cultivated with fresh complete medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. MMP9 ELISA kit (EK0466, Boster Biotech., China) and VEGF ELISA kit (EK0541, Boster Biotech., China) were employed to measure the secretion level of MMP9 and VEGF in the cell supernatant. An equal number of MLFs in rapid proliferation were seeded in 24-well plate cultivated with (serum-free) MCM, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in quadruplicate. Cells cultivated with fresh DMEM/F12 medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. The secretion of FN was measured by FN ELISA kit (EK0351, Boster Biotech., China) according to the manufacturer's instructions. Cells with different treatments were harvested for RT-qPCR analysis to determine the gene expression level of M- Acta2 (Mouse αSMA), M- Mmp9, M- Vegfa and M- Fn1. The primer sequences are provided as follows: + +M- Gapdh- F: 5' GGTGTCTCCTGCGACTTCA 3' 58.2 M- Gapdh- R: 5' TGGTCCAGGGTTTCTTACTCC 3' 58.2 183bp M- Vegfa- F: 5' GCTACTGCCGTCCGATTGAG 3' 60.87 M- Vegfa- R: 5' ACTCCAGGGCTTCATCGTTACAG 3' 62.25 132bp M- Mmp9- F: 5' CACAGCCAACTATGACCAGGAT 3' 59.1 + +<--- Page Split ---> + +1 M-Mmp9-R: 5' CAGGAAGACGAAGGGGAAGA 3' 59.1 115bp + +2 M-Fn1-F: 5' CTATTTACCAACCGCAGACTCAC 3' 58.7 + +3 M-Fn1-R: 5' TGCTTGTTTCCTTGCGACTT 3' 58.5 115bp + +4 M-Acta2-F: 5' CAACTGGTATTGTGCTGGACTC 3' 57.3 + +5 M-Acta2-R: 5' ATCTCACGCTCGGGAGTAGT 3' 57.3 181bp + +6 + +7 Cell Migration assays. MLFs in rapid proliferation were plated in Culture- Inserts (2 Well, Ibidi, + +8 Germany) at the density of \(1 \times 10^{5}\) in \(70 \mu \mathrm{L}\) per well and grew to confluence overnight. + +9 Culture- Inserts as well as the former medium were gently removed. Then the MLFs were + +10 cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs + +11 (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) in triplicate. Cells cultivated with fresh + +12 complete medium was set up as control. Pictures of the cell scratches were taken under the + +13 microscope at \(0 \mathrm{h}\) . After incubated for \(24 \mathrm{h}\) , MLFs were fixed and stained with crystal violet. + +14 Pictures were taken and analyzed with ImageJ. + +15 + +16 Collagen gel contraction assay. An equal number of MLFs seeded in \(6 \mathrm{cm}\) dishes were + +17 cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs + +18 (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) for \(48 \mathrm{h}\) . Cells cultivated with fresh + +19 complete medium was set up as control. The pre- treated MLFs were digested and re- suspended + +20 at the density of \(2 \times 10^{6}\) per mL, kept on ice for later use. The neutral collagen solution was + +21 prepared as follows: \(224 \mu \mathrm{L}\) type I collagen gel ( \(3 \mathrm{mg / ml}\) , Cat. C8062, Solarbio, China) was + +<--- Page Split ---> + +quickly mixed with \(100~\mu \mathrm{L}\) conditional medium and \(8~\mu \mathrm{L}\) NaOH (0.1 N) on ice. Then \(300~\mu \mathrm{L}\) of the pre- treated MLFs suspension was added to the collagen solution on ice immediately. And the neutral cell- collagen mixture was added to 48- well plates \(200~\mu \mathrm{L}\) per well in triplicate and allowed to solidify for \(45\mathrm{min}\) at room temperature. After incubated at \(37^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) , the gels were photographed. ImageJ software was used to measure gel area and evaluate contraction. Gel contraction was assessed as the ratio of the gel area to the area of the well. + +Tube forming assay. Mouse endothelial bEnd3 cells seeded in \(6\mathrm{cm}\) dishes were cultivated with conditional FCM (from MLF previously stimulated by MCM with or without TP5, sFD17 or FR17 peptide as illustrated above) for \(24\mathrm{h}\) . The cells were harvested and seeded in 48- well plates pre- coated with Matrigel (Cat. 356230, BD, USA) in triplicate for each group. After \(6\mathrm{h}\) incubation, five visual fields were randomly chosen from each well and photographed by microscope (CKX53, OLYMPUS, Japan). The tube forming results were analyzed by ImageJ. + +Transwell permeability assay. An equal number of MLFs seeded in \(6\mathrm{cm}\) dishes were cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100~\mu \mathrm{M}\) ) for \(48\mathrm{h}\) . Cells cultivated with fresh complete medium was set up as control. The pre- treated MLFs were digested and seeded at the lower well of a 24- well transwell plate at \(2.5 \times 10^{5}\) cells per well. Each group was triplicate. Then, 7,000 Mouse endothelial bEnd3 cells were seeded on a \(0.4\mu \mathrm{m}\) Transwell insert (Cat. 3413, Corning Costar, USA) above the top of the well until grown to confluence. Rhodamine B- dextran (70 kDa, + +<--- Page Split ---> + +1 Cat. R9379, Sigma-Aldrich, USA) was added to the upper insert on the endothelial cell layer. 2 After 1 h incubation, the translocation of Rhodamine B-dextran from the insert to the lower well 3 passing through the endothelial cell layer was measured by a microplate reader (SPARK, 4 TECAN, Switzerland) at an excitation/emission wavelength of \(540 / 625 \mathrm{nm}\) . The relative 5 permeability of the cell layer was normalized by diving the fluorescence signals of the treatment 6 groups by the control group. + +8 Integrality of the endothelial cell monolayer. Mouse endothelial bEnd3 cells grown in 35- mm confocal dishes to \(100\%\) confluence were treated with conditional FCM, which was obtained from MLFs stimulated by MCM with or without the treatment of different peptides (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) as illustrated above. After incubated with conditional FCM for \(24 \mathrm{h}\) , the single endothelial cell layer was gently washed with PBS and fixed with \(4\%\) paraformaldehyde for \(15 \mathrm{min}\) , permeabilized with \(0.2\%\) Triton X- 100 for \(15 \mathrm{min}\) and blocked with \(2\%\) bovine serum albumin (BSA) for another \(15 \mathrm{min}\) . Cells were incubated with anti- VE- cadherin antibody (1:1000, Cat. Ab205336, Abcam, UK) containing \(0.2\%\) BSA and \(0.1\%\) Triton X- 100 in PBS at \(4^{\circ} \mathrm{C}\) overnight. Cells were washed with PBS for three times and incubated for \(1 \mathrm{h}\) with AF647 labeled goat anti- rabbit IgG (H+L) (1:100, Cat. 33113ES60, Yeasen, China). The nuclei were labeled by DAPI solution (ready- to- use) (Solarbio, China). Images were taken by confocal microscope (Leica, German). + +21 Mice and animal models. C57BL/6 mice (male, 5- week- old) purchased from Slaccas (Shanghai, + +<--- Page Split ---> + +1 China) were adaptive fed for more than one week for subsequent experiments. The animals were maintained under standard laboratory housing conditions where foods and water can be reached freely. All the animal experiments were conducted following the guidelines which have been approved by the Ethics Committee of Zhejiang University. + +5 For MCM- induced lung metastasis model, MCM (300 \(\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days from Day 1 to 10. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells ( \(1 \times 10^{5}\) per mice) was given to the mice. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS® Spectrum In Vivo Imaging System, PerkinElmer, USA) if viable. For bioluminescence imaging, mice were intraperitoneally injected with D- luciferin potassium salt (150 mg/kg, Gold Biotechnology, USA). The bioluminescence of pulmonary metastases was detected 10 min later. + +12 For post- surgery metastasis model, \(1 \times 10^{6}\) B16F10 cells were subcutaneously inoculated in the back of the C57BL/6 mice (male) aged 6 weeks above the right hindlimb on Day 1. On Day 12, tumor resection surgery was conducted on mice to remove the entire tumor tissues as well as the skin cover the tumor under anesthesia. On the next day, \(1 \times 10^{5}\) luciferase- expressing B16F10 cells were injected into mice through tail vein. The tumor recurrence and lung metastasis were monitored twice a week. The tumor volume and body weight were recorded twice a week. Tumor volume was calculated as (width \(^{2} \times\) length) / 2. + +19 Pre- metastatic niche study. MCM (300 \(\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells ( \(1 \times 10^{5}\) + +<--- Page Split ---> + +1 per mice) was given to the mice. On Day 3, 6, 10, 13, mice were euthanized for cardiac perfusion2 and the lung tissues were collected for further analysis.3 For Western-blot analysis, total proteins from the lung tissues were extracted using Tissue4 Protein Extraction Reagent (T- PER™, Cat. 78510, Thermo Pierce, Thermo Scientific, USA) and5 quantified with a Bradford Protein Assay Kit (Cat. P0010, Beyotime, Beijing, China). Samples6 (60 μg) were separated on 10% or 8% SDS- PAGE gels, then transferred to PVDF nitrocellulose7 membrane (Cat. IPVH00010, Merck Millipore). Membranes were incubated with the appropriate8 primary antibodies in 3% BSA, including Fibronectin (1:500, Cat. ab2413, Abcam, UK), MMP9 (1:1000, Cat. ab38898, Abcam, UK), VEGFa (1:500, Cat. ab119, Abcam, UK), TGF- β1 (1:1000,10 Cat. ab179695, Abcam, UK), iNOS (1:1000, Cat. ab204017, Abcam, UK), Arginase 1 (1:1000,11 Cat. ab124917, Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was12 used as control. After incubation with appropriate secondary antibody Goat anti- Mouse IgG13 (H+L) (1:5000, Cat. 31160, Thermo Pierce) or Goat anti- Rabbit IgG (H+L) (1:5000, Cat. 31210,14 Thermo Pierce), the intensity of the immunoreactive proteins was stabilized by SuperSignal®15 West Dura Extended Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X- ray16 film.17 For ELISA analysis, the lung tissues were ground and centrifuged to gain supernatant to measure18 the MMP2 (Cat. OM457413, Omnimabs, USA) and ROS (Cat. OM641674, Omnimabs, USA)19 level.20 For immunofluorescence staining, the left lung was fixed with 4% paraformaldehyde and 30%21 sucrose solution overnight, and embedded into paraffin and sliced into sections. The paraffin + +<--- Page Split ---> + +1 lung sections were deparaffinized and rehydrated, then stained with primary antibodies: αSMA2 (1:500, Cat. Ab7817, Abcam, UK), or CD34 (1:500, Cat. Ab81289, Abcam, UK). Secondary3 antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111-165-003, Jackson, USA) was4 utilized in 1:500 dilution and stained with DAPI before observation.5 For flow cytometry analysis, lung tissues harvested from mice were mechanically minced into6 1- 2 mm pieces using scissors and then dissociated into single cell suspension at 37 °C on a7 shaker for 30 min by enzymes. The digesting solution contains 2 mg/mL collagenase I (Cat.8 BS163, BioSharp, Germany), 2 mg/mL collagenase II (Cat. BS164, BioSharp, Germany) and9 DNase I (Cat. KGF008, KeyGEN BioTech., China). Digestion was stopped by adding 2 volumes10 PBS and filtered through a 70 μM cell strainer (Cat. CSS013070, Jet BIOFIL®, China). The cell11 suspension was centrifuged at 400 g for 5 min to discard the supernatant. Cell precipitations were12 then resuspended in 5 mL RBC lysis buffer (Cat. R1010, Solarbio, China) and centrifuged again13 to discard the supernatant. The single-cell-suspensions washed with PBS and resuspended were14 incubated with FITC-antimouse-CD45 (Cat. 553079, BD, USA), PE-antimouse-NK1.1 (Cat.15 108708, Biolegend, USA), PE-antimouse-CD3 (Cat. 100205, Biolegend, USA),16 PE-antimouse-TER119 (Cat. 116207, Biolegend, USA), PE-antimouse-CD19 (Cat. 152407,17 Biolegend, USA), APC-antimouse-CD11b (Cat. 101211, Biolegend, USA),18 BV605-antimouse-MHC II (Cat. 107639, Biolegend, USA), BB700-CD11c (Cat. 566505, BD,19 USA), BV421-antimouse-Ly6c (Cat. 562727, BD, USA) and PE/CF594-antimouse-Ly6g (Cat.20 562700, BD, USA), BV711-antimouse-F4/80 (Cat. 123147, Biolegend, USA),21 PE/Cy7-antimouse-CD103 (Cat. 121426, Biolegend, USA) antibodies in 100 μL 1% BSA + +<--- Page Split ---> + +containing \(50~\mu \mathrm{L}\) BD Brilliant Stain Buffer for \(30\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) in dark. After centrifuged and washed with PBS, cell pellets were fixed and membrane were perforated with Fix/Perm Buffer (Cat. 562574, BD, USA). The cell pellets were then stained with BV650-antimouse-CD206 (Cat. 141723, Biolegend, USA) for \(40\mathrm{min}\) in the dark at room temperature. After centrifuged and washed with PBS, the stained cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software. + +To investigate the impact of PMN formation induced by MCM injection, metastasis development and mice survival were monitored on MCM- induced PMN model and on the mice that didn't receive MCM injection but were inoculated directly with Luc- B16F10 cells \((1\times 10^{5}\mathrm{per}\) mice) on Day 7. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA). + +## Influence of peptide interference on mice PMN + +For mice PMN model, MCM \((300~\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells \((1\times 10^{5}\mathrm{per}\) mice) was given to the mice. Mice were randomly divided into 4 groups, namely control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at \(40~\mu \mathrm{M / kg / day}\) . On Day 10, mice were euthanized for cardiac perfusion and the lung tissues were collected for further analysis. + +The lung tissues harvested from different groups were fixed, embedded into paraffin and sliced + +<--- Page Split ---> + +into sections. To visualize the activation of lung fibroblasts and angiogenesis in pulmonary PMN, the paraffin lung sections after deparaffinization and rehydration, were stained with primary antibodies: αSMA (1:500, Cat. ab7817, Abcam, UK), or CD34 (1:500, Cat. ab81289, Abcam, UK). Secondary antibody Cy3 conjugated goat anti-rabbit IgG (Cat. 111- 165- 003, Jackson, USA) was utilized in 1:500 dilution and stained with DAPI before observation. To visualize the extracellular matrix environment alteration in pulmonary PMN, the lung sections were stained with appropriate primary antibodies: MMP2 (1:200, Cat. 10373- 2- ap, PTG, USA), or MMP9 (1:1000, Cat. ab228402, Abcam, UK), Secondary antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111- 165- 003, Jackson, USA) and DAPI. To visualize the collagen deposition in pulmonary PMN, Masson's trichrome staining of the lung sections were imaged and analyzed by ImageJ to calculate the collagen volume fraction (CVF) by dividing the blue collagen area by total tissue area. And the Sirius Red Staining was also carried out and the sections were visualized under the polarizing microscope (Nikon Eclipse Ci). To investigate the recruitment of MDSC in PMN, serial sections of lung tissues were stained with periostin (1:200, Cat. 19899- 1- AP, PTG, USA), or co- stained with LOX (1:200, Cat. ab174316, Abcam, UK) and Fibronectin (1:200, Cat. ab92572, Abcam, UK) antibody, or CD11b (1:2000, Cat. ab133357, Abcam, UK) and Gr- 1 (1:200, Cat. ab25377, Abcam, UK) antibody separately. Secondary antibody Cy3 conjugated goat anti- rabbit IgG (1:500, Cat. 111- 165- 003, Jackson, USA), goat anti- rabbit IgG conjugated to HRP (1:2000, Cat. ab6721, Abcam, UK) and fluorescent TSA- 488 (1:200, Wuhan Pinuofei, China) were applied according to Tyramide Signal Amplification technology. The stained sections were observed and imaged under the confocal microscope. Images were analyzed by ImageJ if necessary. + +<--- Page Split ---> + +1 To investigate the alteration of protein expression level in PMN, Western- blot or ELISA 2 experiments were carried out. For Western- blot assay, protein samples were extracted separately 3 from three independent mice from each group. Total proteins from the lung tissues were 4 extracted using Tissue Protein Extraction Reagent (T- PERTM, Cat. 78510, Thermo Pierce, 5 Thermo Scientific, USA) and quantified with a Bradford Protein Assay Kit (Cat. P0010, 6 Beyotime, Beijing, China). Samples (60 μg) were separated on 10% or 8% SDS- PAGE gels, then 7 transferred to PVDF nitrocellulose membrane (Cat. IPVH00010, Merck Millipore). Membranes 8 were incubated with the appropriate primary antibodies in 3% BSA, including Fibronectin (1:500, 9 Cat. ab2413, Abcam, UK), Versican (1:1000, Cat. ab270445, Abcam, UK), VEGFa (1:500, Cat. 10 ab119, Abcam, UK), ANG2 (1:500, Cat. ab155106, Abcam, UK), MMP9 (1:1000, Cat. ab38898, 11 Abcam, UK), MMP2 (1:500, Cat. ab97779, Abcam, UK), TGF- \(\beta 1\) (1:1000, Cat. ab179695, 12 Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was used as control. 13 After incubation with secondary antibody Goat anti- Mouse IgG (H+L) (1:5000, Cat. 31160, 14 Thermo Pierce) or Goat anti- Rabbit IgG (H+L) (1:5000, Cat. 31210, Thermo Pierce), the 15 intensity of the immunoreactive proteins was stabilized by SuperSignal® West Dura Extended 16 Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X- ray film. For ELISA assay, 17 lung tissues were ground and centrifuged to gain supernatant to measure the IL- 6 (Cat. EK0411, 18 Boster, China) level. + +20 In vivo vascular permeability assay. MCM (300 μL per mice) was intraperitoneally injected to 21 the mice for 7 consecutive days. And the mice were randomly divided into 4 groups, namely + +<--- Page Split ---> + +control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at \(40~\mu \mathrm{M / kg / day}\) . On day 7, \(100~\mathrm{mg / kg}\) Rhodamine B-dextran ( \(70\mathrm{kDa}\) , Cat. R9379, Sigma-Aldrich, USA) was intravenously injected to the mice. After \(3\mathrm{h}\) , mice were injected with FITC-lectin (Cat. L0770, Sigma-Aldrich, USA) at \(10\mathrm{mg / kg}\) through the tail- vein. Ten minutes later, each mouse was anesthetized and transcardiac perfused with \(20\mathrm{mL}\) saline to remove the excess dye and followed by \(5\mathrm{mL}\) of \(4\%\) formaldehyde. The lung tissues were formaldehyde- fixed and cryo- sectioned. Slices were observed and imaged by fluorescence microscopy for vascular leakage. There were 3 mice in each group and 5 visual fields were randomly chosen for each section. The relative vascular permeability was analyzed by dividing the dye leakage of each group by healthy control. + +Recruitment of MDSC to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and mechanically minced and digested to obtain the single- cell- suspensions as described above. The single- cell- suspensions washed with PBS and resuspended were incubated with APC- antimouse- CD11b (Cat. 101211, Biolegend, USA) and PE/Cy7- antimouse- Ly6g (Cat. 127617, Biolegend, USA) antibodies in \(100~\mu \mathrm{L}\) \(1\%\) BSA for 30 min at \(4^{\circ}\mathrm{C}\) in dark. After centrifuged and washed with PBS, cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software. + +mRNA sequencing of CD11b+Ly6g+ MDSC recruited to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and digested into single cells + +<--- Page Split ---> + +as introduced as above. The single-cell-suspensions after washing with PBS and re-suspension were incubated with APC-antimouse- CD11b(Cat. 101211, Biolegend, USA) and PE/Cy7-antimouse-Ly6g(Cat. 127617, Biolegend, USA) antibodies in \(100~\mu \mathrm{L}1\%\) BSA for 30 min at \(4^{\circ}\mathrm{C}\) in the dark. After centrifuged and washed with PBS, CD11b \(^+\) Ly6g \(^+\) MDSCs were sorted from the PMN lungs of 10- 12 individual mice from each group per sample by FACS (Beckman moflo Astrios EQ). Total RNA was extracted by TRIzol for cDNA preamplification using the NEBNext® Ultra™ Directional RNA Library Prep Kit for Illumina®, then analyzed using Qubit2.0 Fluorometer, Agilent 2100 bioanalyzer and qRT-PCR. Significantly enriched gene sets were defined as \(P\) values \(< 0.05\) comparing to the control group. + +Inhibition of tumor metastasis on MCM- induced PMN lung metastasis model. The PMN models were established as introduced above, mice were randomly divided into 4 groups (n = 7), namely control, TP5, sFD17 and FR17. From day 3 to 21, mice from different groups were subcutaneously administrated with saline, TP5 (40 \(\mu \mathrm{M / kg}\) per day), sFD17 (40 \(\mu \mathrm{M / kg}\) per day) and FR17 (40 \(\mu \mathrm{M / kg}\) per day) separately. Body weight was recorded every 3 days. On day 20, blood was collected from the submarginal ocular venous plexus under anesthesia for blood tests including complete blood count, alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN) and serum creatinine (CREA). On day 28, mice were euthanized and major organs, such as heart, liver, spleen, lung, kidney and thymus were collected after cardiac perfusion. Lung tumor nodules were then counted under the stereo microscope. The major organs were fixed, embedded into paraffin for Hematoxylin & Eosin staining. The thymus + +<--- Page Split ---> + +coefficient was calculated as the thymus weight divided by the body weight and the spleen coefficient was calculated as the spleen weight divided by the body weight. + +Inhibition of tumor metastasis post- surgery. The post- surgery metastasis model was established as illustrated above. Mice were randomly divided into 3 groups, namely control, anti- PD1 and FR17. For FR17 treatment, peptide was subcutaneously administrated to the mice from Day 7 to Day 25 at the dose of \(40~\mu \mathrm{M / kg}\) per day. For anti- PD1 treatment, \(100\mathrm{mg}\) anti- PD1 (Cat. BE0146, Bio X Cell, USA) was given by \(i.p\) . injection twice per week starting from day 3 post tumor resection and given two times per week from Day 15 to Day 25 for a total of 4 times. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA) until death. The recurrence of the excised subcutaneous tumor was closely monitored and tumor volume was measured and calculated following the ellipsoid volume formula: (width \(^2\times\) length) / 2. + +## Statistics + +Statistical analysis was performed using GraphPad Prism 8.0.1 (GraphPad Software, CA, USA). + +Data were presented as means \(\pm\) SD. Statistical evaluation of differences between experimental + +groups was performed by one- way ANOVA followed by Tukey's multiple comparisons test. + +Statistical significance was considered at least at \(p< 0.05\) . + +<--- Page Split ---> + +1 Abraham, M. J. et al. GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. \*SoftwareX\*, 1, 19- 25 (2015).2 Humphrey, W., Dalke A. & Schulten, K. VMD: visual molecular dynamics. \*J. molecular graphics\*, 14, 33- 38 (1996). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- SIV9.pdf- SIV9withoutauthorsinformation.pdf- SIV9withoutauthorsinformation.pdf- nrreportingsummary20210809T100118.224.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806_det.mmd b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..14a9edf689e0cea843e996e1766412c1ca82757f --- /dev/null +++ b/preprint/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806/preprint__0ddaa0694268812f17177a0729764fc95646c13558fc8d774118c31a5fc1c806_det.mmd @@ -0,0 +1,572 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 951, 175]]<|/det|> +# Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche + +<|ref|>text<|/ref|><|det|>[[42, 193, 582, 936]]<|/det|> +Yi Zhou Zhejiang University Peng Ke Shengli Clinical Medical College of Fujian Medical University Yiyi Xia Zhejiang University Honghui Wu Zhejiang University Zhentao Zhang Zhejiang University Haiqing Zhong Zhejiang University Qi Dai Zhejiang University Tiantian Wang Zhejiang University Mengting Lin Zhejiang University Yaosheng Li Zhejiang University Xinchi Jiang Zhejiang University Qiyao Yang Zhejiang University Yiying Lu Zhejiang University Xincheng Zhong Zhejiang University Min Han Zhejiang University https://orcid.org/0000- 0001- 9373- 8466 Jianqing Gao ( \(\square\) gaojianqing@zju.edu.cn) Zhejiang University + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 45, 101, 63]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 83, 955, 126]]<|/det|> +Keywords: pre- metastatic niche, self- assembled peptide, pre- metastasis associated fibroblasts, 2 vascular endothelial cells, myeloid- derived suppressor cells, tumor metastasis + +<|ref|>text<|/ref|><|det|>[[44, 144, 318, 163]]<|/det|> +Posted Date: August 10th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 182, 463, 201]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 699014/v1 + +<|ref|>text<|/ref|><|det|>[[42, 219, 910, 262]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 298, 909, 340]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 25th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 30634- 8. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[108, 92, 852, 151]]<|/det|> +# Peptide nano-blanket impedes fibroblasts activation and subsequent formation of pre-metastatic niche + +<|ref|>text<|/ref|><|det|>[[108, 170, 122, 184]]<|/det|> +3 + +<|ref|>text<|/ref|><|det|>[[103, 201, 855, 610]]<|/det|> +4 Yi Zhou', Peng \(\mathrm{Ke}^{1}\) 4, Yiyi Xia', Honghui Wu', Zhentao Zhang', Haiqing Zhong', Qi Dai' 5, Tiantian Wang', Mengting Lin', Yaosheng Li', Xinchi Jiang', Qiyao Yang' 5, Yiying Lu', Xincheng Zhong', Min Han' 2 3\\*, Jianqing Gao' 2 3\\* 7 1 Institute of Pharmaceutics, Zhejiang Province Key Laboratory of Anti- Cancer Drug Research, 8 College of Pharmaceutical Sciences, Zhejiang University, Hangzhou 310058, P.R. China 9 2 Cancer Center of Zhejiang University, Zhejiang University, Hangzhou 310058, P.R. China. 10 3 Hangzhou Institute of Innovative Medicine, Zhejiang University, Hangzhou 310058, P.R. China. 11 4 Shengli Clinical Medical College of Fujian Medical University, Fuzhou, 350001, China. 12 5 Department of Radiation Oncology, Key Laboratory of Cancer Prevention and Intervention, The 13 Second Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou 310058, China. 14 \\* Corresponding author: hanmin@zju.edu.cn (M. H.); gaojianqing@zju.edu.cn (J. G.) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 95, 854, 150]]<|/det|> +1 **Keywords:** pre-metastatic niche, self-assembled peptide, pre-metastasis associated fibroblasts, vascular endothelial cells, myeloid-derived suppressor cells, tumor metastasis + +<|ref|>sub_title<|/ref|><|det|>[[148, 179, 227, 195]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[140, 210, 855, 900]]<|/det|> +In various types of malignant tumors, metastasis is responsible for most of the tumor- induced death. Though emerging technologies provided early detection of tumor metastasis or even warning of high metastatic risk before the actual occurrence of metastasis, clinical treatment on metastasis prevention lags far behind. Evidences have illustrated that primary tumor induced pre- metastatic niche (PMN) formation in distal organs by producing pro- metastasis factors, spreading the spark to ignite the distal microenvironment. Given the fundamental role of PMN in the development of metastases, interruption of PMN formation would be a promising strategy to take early actions against tumor metastasis. Here we report an enzyme- activatable assembled peptide FR17 that can serve as a “flame- retarding blanket” at PMN site specifically to extinguish the “fire” of tumor- supportive microenvironment adaption. Our experiment demonstrated that the assembled peptide successfully reversed extracellular matrix deposition, vascular leakage and angiogenesis through inhibition on fibroblasts activation in PMN, which suppressed the remodeling of metastasis- supportive host stromal, and further prevented the recruitment of myeloid cells to PMN and then recovered the immunosuppressive microenvironment. Cell transcriptomic analysis of the pulmonary recruited MDSC suggested that FR17 intervention could regulate immune response activation, immune cells chemotaxis and migration pathways. Consequently, FR17 administration effectively inhibited pulmonary PMN formation and postoperative metastasis of melanoma, with only 30% lung- metastasis occurrence was observed for FR17 treated group at the time point when 100% occurrence was observed for the control + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 94, 852, 150]]<|/det|> +group and \(80\%\) occurrence for anti- PD1 treated group, offering a robust therapeutic strategy against PMN establishing to prevent metastasis. + +<|ref|>sub_title<|/ref|><|det|>[[147, 207, 260, 223]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[144, 250, 854, 600]]<|/det|> +Though therapeutic outcomes and survival rate for patients with various cancers have been greatly improved in last decades, effective treatments for patients with metastatic cancer are still limited. Though emerging technologies provided early detection of malignant transformation or even warning of high metastatic risk by biomarker screening before the actual occurrence of metastasis in clinic. clinical treatment on metastasis prevention lags far behind. The contemporary therapeutic strategies against metastasis in clinic, including systemic chemotherapy, radiotherapy and immunotherapy, mainly focus on the later time period of metastasis development, or at least after the arrival and colonization of disseminated tumor cells to the distal organs, which have gained unsatisfied clinical outcomes. Observations of the adjuvant chemotherapy resistance and treatment- related toxicities revealed the shortcoming of the currently available strategies. + +<|ref|>text<|/ref|><|det|>[[144, 630, 854, 910]]<|/det|> +Growing evidence illustrated that an inflammatory, neovascularized, immunosuppressive, tumor supportive microenvironment has emerged before the arrival and colonization of disseminated tumor cells, which is termed as pre- metastatic niche (PMN) formation. Relevant studies revealed that complex interactions between multiple participants and alteration in regulative pathways energized the construction of PMN, such as primary tumor- derived cytokines and exosomes, myeloid- derived suppressor cells (MDSCs), and the tumor re- educated stromal environment including pre- metastasis associated fibroblasts, destabilized vasculature and extracellular matrix (ECM) (Fig. 1a). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 92, 857, 636]]<|/det|> +1 Since PMN was considered as the foundation laid for circulating tumor cells colonization and one 2 of the vital premises to develop metastasis in distant organs, we wonder if the early process of 3 tumor metastasis can be terminated or even totally prevented by interrupting the formation of 4 PMN. To view the entire process of PMN establishment and metastasis development as a progress 5 of the occurrence and spread of a huge forest fire, it would be more efficient to contain and beat 6 out the local flame by preventing the formation of PMN. Here we report an enzyme- activatable 7 assembled peptide FR17 that can serve as a "flame- retarding blanket" at PMN site specifically, 8 containing and suppressing the "fire blaze" of PMN formation to further develop into overt 9 metastasis (Fig. 1b). As a substrate peptide of matrix metalloproteinase 2 (MMP2), FR17 can 10 release self- assembly monomer FG8 to construct peptide nano- blanket in PMN stromal 11 microenvironment, impeding fibroblasts activation so as to prevent metastatic cascades. We 12 further explored the subsequent impact of inhibition on fibroblasts activation induced by FR17 13 intervene, revealing the underlining mechanism on cellular interactions among fibroblasts, 14 vascular endothelial cells and extracellular components, and intervention on PMN recruited 15 MDSCs via in vitro and in vivo experiments. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[45, 93, 950, 480]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 498, 855, 515]]<|/det|> +
Figure 1. Schematic illustration of peptide nano-blanket impedes fibroblasts activation and
+ +<|ref|>text<|/ref|><|det|>[[103, 533, 857, 848]]<|/det|> +1 Figure 1. Schematic illustration of peptide nano- blanket impedes fibroblasts activation and subsequent formation of pre- metastatic niche. a, Illustration of the pathological process of pre- metastatic niche (PMN) formation. The primary tumor produces pro- metastasis factors, such as tumor- derived secreted factors (TDSF), to induce fibroblast activation in metastatic destination organs. The tumor- educated activated fibroblasts serve to construct a metastasis- supportive host stromal, including extra cellular matrix (ECM) deposition and reconstruction, angiogenesis and vascular leakage, as well as myeloid- derived suppressor cells (MDSC) recruitment. b, After FR17 subcutaneous administration, the in- situ assembled peptide nano- blanket in PMN stromal microenvironment impedes fibroblasts activation so as to retard stromal and vessel pro- metastatic reconstruction, inhibiting MDSC recruitment and metastatic cascades. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[103, 95, 348, 112]]<|/det|> +## 1 Results and Discussion + +<|ref|>sub_title<|/ref|><|det|>[[147, 141, 503, 158]]<|/det|> +## Peptide nano-blanket transformed from FR17. + +<|ref|>text<|/ref|><|det|>[[144, 186, 855, 611]]<|/det|> +The matrix metalloproteinase 2 (MMP2)- activatable self- assembled branched peptide (FR17, FFK(GPLGLAGG- YVDKR)Y) consists of (1) the backbone of a self- assembly peptide domain Phe- Phe- Lys- Tyr (FFKY), which is derived from \(\beta\) - amyloid (Aβ) peptide14, 15; (2) thymopentin (TP5, Arg- Lys- Asp- Val- Tyr, RKDVY), the pentapeptide with perfect hydrophilic property and immune modulation effect, which is applied as the adjunctive therapeutic agent on cancer treatment in clinic to prevent postoperative infection and to activate immune response16, 17. TP5 was conjugated to the side- chain of Lys (K) of the main chain FFKY with the MMP2- cleavable peptide linker (Gly- Pro- Leu- Gly- Leu- Ala- Gly- Gly, GPLGLAGG)18, increasing the hydrophilic property of the entire peptide molecule. Furthermore, the sequence of TP5 in peptide FR17 was replaced by the scrambled pentapeptide of TP5 without immunomodulatory bioactivity, i.e. DVVKR, to form the scrambled group (sFD17, FFK(GPLGLAGG- RKVYD)Y) as peptide assembly control. + +<|ref|>text<|/ref|><|det|>[[144, 640, 855, 881]]<|/det|> +Peptides were synthesized via Fmoc solid- phase peptide synthesis technology and the peptide sequences were verified by mass spectra (Supplementary Fig. 1- 2). When specifically cleaved by MMP2, both FR17 and sFD17 are able to release self- assembled monomer FG8 (FFK(GPLG)Y), constructing peptide self- assemblies, the peptide nano- blanket. The transmission electron microscopy (TEM) images (Fig. 2a) and the Cryo- TEM image (Supplementary Fig. 3a) showed the lamellar structure of the peptide self- assemblies formed by enzymatic degradation (Supplementary Fig. 3b) with an average diameter of \(\sim 500 \mathrm{nm}\) (Fig. 2b). In addition, circular + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 90, 856, 710]]<|/det|> +1 dichroism (CD) analysis revealed alterations in the secondary structure of peptide by the presence 2 of enzyme (Supplementary Fig. 3c). The peptide nano-blanket is woven by the self-assembled 3 monomer FG8 (FFK(GPLG)Y), which would spontaneously fold into lamellar structure rather 4 than the fiber clusters aggregated by FFKY as we previously reported14. The peptide assembly 5 relies on the non-bonded interactions, typically hydrogen bonds and \(\pi - \pi\) stacking, between 6 adjacent peptide molecules. The self-assembling pattern of FG8 changed due to the branch 7 modification with GPLG side-chain. The all-atom molecular dynamics (MD) simulation of 8 FFKY-assemblies and FG8-assemblies gives a microscopic account of the impacts of branch 9 modification on peptide interactions. The MD simulation revealed that there are more 10 intermolecular hydrogen bonds involved in FG8 cluster, which are more ordered and mostly 11 formed between Phe of adjacent FG8 molecules (Supplementary Fig. 4a-b). While the 12 intermolecular hydrogen bonds involved in FFKY assembly are less-formed and disordered (Fig. 13 2c-h). Besides, the aggregation-induced luminescence effect (AIE) was also employed to monitor 14 the spontaneous aggregation of FG8, the self-assembled monomer, in aqueous system 15 (Supplementary Fig. 4c-e). Taken together, these results indicated the self-assembly property of 16 FG8 monomer, and MMP2-cleaved release of FG8 from FR17 or sFD17 to form the peptide 17 nano-blanket. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[152, 90, 839, 504]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[110, 520, 852, 910]]<|/det|> +
Figure 2. Enzyme-activated self-assembly of FR17 and all-atom molecular dynamics (MD) simulation of the self-assembly of FG8. a, TEM images of FR17 or sFD17 treated with MMP2 (scale bar \(= 200 \mathrm{nm}\) ), FG8 and FFKY (scale bar \(= 100 \mathrm{nm}\) ). b, Size distribution of FR17 and sFD17 after MMP2 cleavage. c, Molecular structure of FFKY. d, The FFKY assemblies generated at \(\mathrm{t} = 200 \mathrm{ns}\) of MD simulation of 16 FFKY molecules in water (containing NaCl for charge neutralization). e, Typical molecular cluster and the non-bonded interactions involved in FFKY assemblies in detail. f, Molecular structure of FG8. g, The FG8 assemblies generated at \(\mathrm{t} = 200 \mathrm{ns}\) of MD simulation of 16 FG8 molecules in water. h, Typical molecular cluster and the non-bonded interactions involved in FG8 assemblies in detail. The dashed purple lines denote the \(\pi -\pi\) stacking. The dashed blue lines denote the hydrogen bonds. The nitrogen atoms are labeled in blue. And the oxygen atoms are labeled in red.
+ +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[147, 95, 649, 111]]<|/det|> +# The pathological process in the MCM-induced PMN model in vivo + +<|ref|>text<|/ref|><|det|>[[144, 140, 855, 420]]<|/det|> +In order to study the effect of FR17 administration on the process of PMN development, an in vivo PMN model induced by melanoma- conditioned media (MCM) has been established according to the previous research on PMN19. Briefly, tumor- derived factors secreted from primary tumor were replaced by MCM intraperitoneal injection to the mice for 10 consecutive days from Day 1 to 10. On Day 7, when the tumor supportive microenvironment was successfully established in the lung, B16F10 melanoma cells were intravenously administrated as the simulation of circulative tumor cells (CTC) wandering through blood vessels and some would successfully colonize in the prepared "fire scene" in lung (Supplementary Fig. 5b). + +<|ref|>text<|/ref|><|det|>[[144, 445, 855, 910]]<|/det|> +The pathological process in the lung was assessed from various aspects during MCM- induced PMN establishment from Day 3 to Day 13 (Supplementary Fig. 5- 6). Important cellular derived molecular components and cytokines were closely monitored during the process. The higher expression of the extracellular matrix (ECM) component fibronectin (FN) was generated by MCM inducement. Moreover, matrix metalloproteinase 9 (MMP9), vascular endothelial growth factor a (VEGFa) were up- regulated along with the pathological progress of lung PMN from Day 0 to Day 10 and reached a plateau on Day 10 (Supplementary Fig. 5a). An accordant trend was found on MMP2 level as well (Supplementary Fig. 5d). Meanwhile, the immune cell population analysis during the pathological process in the lung of MCM- induced PMN model was conducted by flow cytometry, which drew our attention to the crime culprit- cell induced PMN, MDSC, for the recruitment of MDSC increased through the timeline (Supplementary Fig. 6- 7). It's reported that MDSC contributes a lot in developing the immunosuppression microenvironment in PMN, preparing more suitable and fertile land for tumor cells to take root in13, 20. Expression level of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 95, 855, 448]]<|/det|> +major inflammatory mediator TGF- \(\beta 1\) , possibly produced by MDSCs, increased gradually but sharply. Typical biomarkers produced by MDSC to exert immuno- suppressive effect, in other words, the incriminating tools for the microenvironment re- education, were also detected, i.e. reactive oxygen species (ROS), iNOS and arginase- 1 (Arg- 1) (Supplementary Fig. 5a, c) 21. Major characteristics of PMN also emerged in lung as time went by, such as activation of fibroblasts (Supplementary Fig. 5e- f), angiogenesis (Supplementary Fig. 5g) and increase of vascular permeability (Supplementary Fig. 5h) 12. All these data indicated that the in vivo PMN model had been well- established, reflecting the pathological process of PMN development. With the overall support provided by PMN, it would consequently accelerate and aggravate metastasis in vivo (Supplementary Fig. 8). + +<|ref|>title<|/ref|><|det|>[[145, 475, 849, 493]]<|/det|> +# FR17 administration interrupts the activation of fibroblast induced by tumor derived factors + +<|ref|>text<|/ref|><|det|>[[144, 520, 855, 872]]<|/det|> +According to previous researches on tumor metastasis22 and our assessment on MCM- induced PMN mice model, the activation of the resident fibroblasts in distal tissues could be regarded as the tipping point of the beginning of PMN establishing, raising the alarm about laying the foundations of the potential metastasis. It's reported that tumor- educated fibroblasts would serve to construct a tumor supportive host stromal via TGF- \(\beta\) signaling23, promoting ECM degradation and reconstruction, inducing angiogenic and pro- inflammatory response of endothelial cells, recruiting VLA- 4+ bone marrow- derived cells (BMDCs) by localized FN deposition for niche formation19. Indeed, lysyl oxidase (LOX) cross- linked collagen surrounding pre- metastasis associated fibroblasts attracts CD11b+ myeloid cells invasion in destination organs, which corresponds to metastatic efficiency24. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 100, 936, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[140, 507, 844, 524]]<|/det|> +
Figure 3. FR17 interrupted the activation of fibroblast induced by tumor derived factors. a,
+ +<|ref|>text<|/ref|><|det|>[[105, 543, 846, 905]]<|/det|> +Schematic drawing illustrates the in- situ assembled peptide nano- blanket interrupts the activation of fibroblast. When activated by tumor derived factors during PMN development, the expression of proangiogenic factors and ECM remodeling factors would be up- regulated in fibroblast, as well as the ECM components production. While the peptide nano- blanket could calm down fibroblast activation, down- regulating the above factors. b, Schematic illustration to show the protocol of MCM stimulation and peptide treatment on mice lung fibroblast (MLF) in vitro. c, Cell proliferation of MLF after MCM stimulation and peptide treatment (n = 4). d, Secretion of MMP9, VEGF and fibronectin in the culture media of MLF after stimulated by MCM, treated with or without peptide. n = 4 for treated groups and 3 for the control group. e, qPCR analysis of Acta2 (i.e., αSma), Mmp9, Vegfa, Fibronectin1 (Fn1) expression in MLF after MCM stimulation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 93, 846, 339]]<|/det|> +1 and peptide treatment. f, Migration assay and collagen gel contract assay to evaluate MLF cellular functions after MCM stimulation and peptide treatment (n = 3). Scale bar = 100 μm. g, Expression level of fibronectin in the lung harvested from the PMN model mice administrated with different peptides on Day 10. h, Representative images and semi-quantification of \(\alpha \mathrm{SMA}+\) fibroblasts in the lung harvested from mice administrated with different peptides on Day 10. Data is presented as mean \(\pm\) SD. One-way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. Scale bar = 50 μm. + +<|ref|>text<|/ref|><|det|>[[101, 407, 846, 881]]<|/det|> +9 Here in a cell model simulating the impact of secreted factors derived from primary tumor on lung fibroblasts in vitro (Fig. 3b), FR17 cultivation along with MCM stimulation showed an interruption effect on tumor derived factors irritating lung fibroblasts (Fig. 3a). The treatment with FR17 or sFD17, which would form peptide nano-blanket assembled with the monomer FG8 that was released by MMP2-induced cleavage, successfully prevented the activation of fibroblasts by MCM. The inhibition on the expression of pathologic fibroblast biomarker alpha smooth muscle actin ( \(\alpha \mathrm{SMA}\) ) as well as the fibroblasts proliferation and cellular bio-functions (Fig. 3c-f) also substantiated the blocking of lung fibroblasts activity by FR17 or the scrambled control sFD17. The reverse of the increase in cytokines secretion, like MMP9, VEGF, FN, was determined by ELISA kit (Fig. 3d). The qPCR results further suggested that the peptide assemblies might exert biological regulatory effect on cells while not just acting like a physical shield to wrap on the surface of cells (Fig. 3e). Besides, when irritated by tumor derived factors to present the activated \(\alpha \mathrm{SMA}^{+}\) phenotype, the migration ability as well as the collagen gel + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 95, 845, 301]]<|/det|> +contracting function of fibroblasts got promoted. By contrast, cultivation with FR17 minified these functional changes to a large extent (Fig. 3f). Administration of FR17 to the in vivo PMN model also gave great relief to the activation of the fibroblasts in lung (Fig. 3h). What's more, the above data also suggested that the "flame-retarding" effect on fibroblast activation was almost completely contributed by peptide nano-blanket on account of the similar outcomes of the scrambled control sFD17 to that of FR17 treatment. + +<|ref|>text<|/ref|><|det|>[[144, 332, 845, 655]]<|/det|> +The stromal microenvironment, apart from fibroblasts, consisting of ECM and vasculature, could also be re- educated by tumor derived factors25. For ECM would have gone through remodeling to rebuild a supportive niche for tumor colonization in PMN model, Western blot analysis, Masson staining and Sirius Red staining revealed the down- regulation in the expression of FN (Fig. 3g) and collagen (Supplementary Fig. 9a- c) in the lungs of FR17 group and sFD17 group, which would have likely been over- expressed by the activated fibroblasts otherwise. Besides, versican expression in the lung was also lower in the FR17 or sFD17 intervened groups than that of PMN control group and TP5 treated group (Supplementary Fig. 9d), and the elevated expression of which has been reported to contribute to angiogenesis in tumor26. + +<|ref|>text<|/ref|><|det|>[[147, 730, 800, 748]]<|/det|> +FR17 protects fibroblasts from activation to inhibit vascular leakage and angiogenesis. + +<|ref|>text<|/ref|><|det|>[[144, 777, 845, 907]]<|/det|> +The activated fibroblasts in primary tumor, also known as cancer- associated fibroblasts (CAFs), have been reported to produce proangiogenic factors, such as VEGF, so as to promote tumor angiogenesis27. What's more, the secretion of MMPs by CAFs induced tumor vascular leakage, exacerbating MDSCs and tumor cells intravasion into the vascular system. Therefore, we + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 845, 490]]<|/det|> +assumed that, as an important participator and vanguard in the construction of PMN, the activated fibroblast may also contribute to angiogenesis and the increasing vascular permeability in PMN (Fig. 4a). Further exploration was carried out on mice vascular endothelial cells to find out whether fibroblasts promote angiogenesis and vascular permeability during the arounement induced by MCM (as shown in Supplementary Fig. 10a). The tube forming assay demonstrated the proangiogenic capability of fibroblasts activated by MCM (Fig. 4b). Meanwhile, the transwell permeability assay (as illustrated in Supplementary Fig. 10b) to mimic the inner layer of blood vessel in vitro verified that the fibroblast-conditioned medium (FCM) collected from activated mice lung fibroblasts (MLF) increased vascular permeability (Fig. 4c), indicated directly by the dismission of endothelial adherence junctions mediated by vascular endothelial cadherin, VE-cadherin (Fig. 4d). + +<|ref|>text<|/ref|><|det|>[[144, 520, 845, 765]]<|/det|> +For all experiments conducted on endothelial cells, the different conditional FCM was obtained from fibroblasts pre- treated with FR17, sFD17 or TP5 on interrupting MCM stimulation separately. The conditional FCM pre- treated with FR17 or sFD17, rather than TP5, offset the increase in bEnd3 cell proliferation induced by the indirect stimulation of MCM (Supplementary Fig. 10c). The acceleration in neovascularization and the disruption on endothelial cell- cell connection to cause vascular leakage were made up by indirect FR17 or sFD17 treatment on MLF as indicated in Fig. 4b- c. + +<|ref|>text<|/ref|><|det|>[[144, 796, 845, 888]]<|/det|> +To examine the influence of FR17 intervention on the expression level of the proangiogenic factors and vascular remodeling enzymes in pulmonary PMN in general, Western blot analysis was carried out. When compared to PMN control group, the increased expression of VEGFa, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 92, 847, 303]]<|/det|> +1 ANG2, MMP9, MMP2 was reversed by FR17 treatment almost back to normal levels (Fig. 4e, 2 S11). In addition, the scrambled control sFD17, which doesn't possess the drug-bioactivity of 3 TP5, exhibited similar inhibition result as FR17 while TP5 didn't, suggesting the protective 4 effect was contributed by the peptide assemblies formed by the enzyme-activatable self-assemble 5 monomer. Vascular leakage assay on PMN model in vivo verified that FR17 or sFD17 6 administration attenuated the enhancement of pulmonary vascular permeability (Fig. 4f-g). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 95, 933, 453]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 472, 843, 490]]<|/det|> +
Figure 4. FR17 administration protected fibroblasts from activation to inhibit vascular
+ +<|ref|>text<|/ref|><|det|>[[103, 508, 845, 910]]<|/det|> +leakage and angiogenesis. a, Illustration of the inhibition of vascular leakage and angiogenesis via the peptide nano- blanket protection on fibroblasts from tumor re- education. b, Tube forming assay was performed on bEnd3 cells which were treated with conditional FCM collected from MLF pre- stimulated with MCM and peptide. Scale bar \(= 100 \mu \mathrm{m}\) . c, Permeability of the endothelial cell layer in vitro when co- cultured with MLF pre- treated with MCM and peptide. Data is presented as mean \(\pm\) SD. n = 3. d, Integrality of the endothelial cell monolayer after cultivated with conditional FCM collected from MCM and peptide pre- stimulated MLF, indicated by VE- cadherin on the membrane (n = 3). The white box and the white arrows in the enlarged images indicate the tight junction between the endothelial cells. While the yellow box and yellow arrows indicate the disruption of cell- cell connection. Scale bar \(= 50 \mu \mathrm{m}\) in the upper panel. Scale bar \(= 30 \mu \mathrm{m}\) in the lower panel. e, Expression level of VEGFa, ANG2, MMP9 and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 93, 844, 264]]<|/det|> +MMP2 in the pulmonary PMN harvested from mice administrated with different peptides on Day 10. f & g, Vascular permeability of the pulmonary PMN after peptide administration. The semi-quantification was calculated from 5 random visual fields for each section taken from \(\mathrm{n} = 3\) biologically independent mice. Data is presented as mean \(\pm\) SD. Scale bar \(= 100 \mu \mathrm{m}\) . One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. + +<|ref|>text<|/ref|><|det|>[[110, 340, 844, 435]]<|/det|> +Above data confirmed that FR17 can be employed as an in- situ spontaneously- assembled "flame- retarding blanket" to beat out the "flames" on fibroblasts caused by tumor derived factors in PMN, preventing pro- metastatic angiogenesis and vascular destabilization. + +<|ref|>title<|/ref|><|det|>[[145, 512, 842, 528]]<|/det|> +# FR17 administration impedes MDSC recruitment to pulmonary PMN and modulates the + +<|ref|>title<|/ref|><|det|>[[147, 550, 364, 565]]<|/det|> +# immuno-microenvironment. + +<|ref|>text<|/ref|><|det|>[[144, 595, 844, 880]]<|/det|> +Given the fact that MDSC occupies a vital position in PMN construction and metastasis formation, MDSC has attracted wide concerns in recent years. It's firstly reported by David Lyden's team that VEGFR+ myeloid progenitor cells are recruited and formed cell clusters to the sites highly expressing FN, which is most likely produced by resident fibroblasts, in the distal tissues prior to the arrival of tumor cells19. The recruitment of MDSC was found to be related to the enhancement of ECM production and remodeling in PMN, including cross- linking collagen by LOX secreted by tumor cells24, FN deposition28, periostin (POSTN)- enrichment29. As MDSC plays a crucial role in developing immunosuppressive and inflammatory microenvironment, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 92, 847, 188]]<|/det|> +some pioneers have demonstrated that blocking the recruitment of MDSC could be a promising strategy to suppress early metastasis by preventing the development of breeding ground for tumor30- 33. + +<|ref|>text<|/ref|><|det|>[[144, 216, 845, 844]]<|/det|> +In our PMN model, the accumulation of MDSC in lung was decreased in FR17 and sFD17 groups, revealed by immunofluorescent staining (Fig. 5a) and flow cytometry analysis on Day 10 (Fig. 5b). Moreover, focal enrichment areas of FN and the co-localization of LOX attracted an increasing number of CD11b+Gr+ MDSC in serial sections of the PMN lung comparing to healthy lung (Fig. 5a, Supplementary Fig. 12), consistent with the previous reports24. The immunofluorescent stains indicated that FR17 intervention successfully down-regulated the expression of LOX, FN and POSTN in the lung induced by tumor-derived factors to impede the construction of PMN, which probably resulted in the cut in MDSC recruitment (Supplementary Fig. 12). To evaluate prevention of peptide administration on developing PMN immunosuppressive environment, protein level of TGF- \(\beta\) in PMN (Fig. 5c), the well-known immuno-modulator produced by MDSC to regulate the establishment of immunosuppressive tumor supportive niche, as well as the pro-inflammatory cytokine IL-6 was detected (Fig. 5d). FR17 treatment successfully down-regulated TGF- \(\beta\) and IL-6 expression in lung, so as sFD17. In the meantime, TP5 administration exhibited partial abatement on the expression of these cytokines34, suggesting both the enzyme-responsive assembled peptide nano-blanket and the immunoregulatory agent TP5 facilitated the normalization of immunosuppressive and inflammatory environment of PMN. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[168, 95, 825, 590]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[150, 607, 844, 625]]<|/det|> +
Figure 5. FR17 administration prevented MDSC recruitment, pausing the development of
+ +<|ref|>text<|/ref|><|det|>[[150, 644, 844, 888]]<|/det|> +the immunosuppressive microenvironment in PMN. a, Serial sections of the lungs harvested from the model mice treated with different peptides. Serial sections show the distribution of lysyl oxidase (LOX) and Fibronectin (FN) and the co- location of CD11b+Gr1+ myeloid derived suppressor cells (MDSC). b, Recruitment of CD11b+Ly6g+ MDSC to the lungs of the model mice treated with different peptides on Day 10. Data is presented as mean ± SD. n = 4 biologically independent mice for treatment groups, n = 3 for the control group. c, Expression of TGF- \(\beta 1\) in the lung harvested from PMN model mice administrated with different peptides. d, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 93, 848, 380]]<|/det|> +1 IL- 6 expression in the lung tissue fluid collected from PMN model mice administrated with different peptides. Data is presented as mean \(\pm\) SD. \(\mathrm{n} = 3\) biologically independent mice. 3 One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. e, GO enrichment analysis of CD11b \(^+\) Ly6g \(^+\) MDSC sorted from different treatment groups. RNA preparations were extracted from CD11b \(^+\) Ly6g \(^+\) MDSCs sorted from lungs pooled from 10- 12 mice per sample. The size of the dots corresponds to the number of genes per pathway, and the color indicates \(p\) - value. Statistical significance was considered at least at \(p<\) 0.05. + +<|ref|>text<|/ref|><|det|>[[100, 444, 847, 881]]<|/det|> +10 To find out whether the impact on MDSC- induced immunosuppressive microenvironment in PMN was mainly contributed by TP5 or by the in- situ assembled peptide nano- blanket, cell transcriptomic analysis of MDSCs recruited to the lung was carried out on different treatment groups. Given the fact that CD11b \(^+\) Ly6g \(^+\) MDSC takes the majority (over \(90\%\) ) of MDSC population in the lung of MCM- induced PMN model (Supplementary Fig. 6a- b), CD11b \(^+\) Ly6g \(^+\) MDSC were sorted on Day 10 from pulmonary PMN after different treatments as the representative subset for further transcription analysis (Supplementary Fig. 13). GO enrichment analysis indicated that the regulation effect of FR17 might relate to the activation of immune response pathway, cytokine production involved in immune response, regulation of leukocyte and lymphocyte activation, leukocyte chemotaxis and migration (Fig. 5e). Relative enriched pathways provided possible comments on the underlined mechanisms, including: regulation on cytokine biosynthetic process, adaptive immune response based on somatic recombination + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 92, 848, 530]]<|/det|> +1 immune receptors built from immunoglobulin super-family domains, interferon- \(\gamma\) - mediated 2 signaling pathway and the impact on chemokine receptor binding as well as CXCR chemokine 3 receptor binding (Supplementary Fig. 14). From above, when compared with the differential 4 gene enrichment pathway of sFD17 and TP5, we found that the regulation on immune cells 5 chemotaxis and migration pathways was contributed by the in- situ peptide assemblies, for the 6 peptide nano- blanket would only form in FR17 or sFD17 treated lung while not in free TP5 7 treated group. And the Venn diagram and further enrichment analysis to put TP5 aside suggested 8 these would correspond to the regulation on cell surface receptor signaling pathway by peptide 9 assemblies (Supplementary Fig. 14b- c). In addition, the regulation on leukocyte differentiation 10 and T cell activation pathway was mainly contributed by TP5 (Supplementary Fig. 14d), which 11 has been commonly accepted as one of the mechanisms for TP5 to exert immune regulation 12 effect35-37. + +<|ref|>text<|/ref|><|det|>[[145, 560, 842, 615]]<|/det|> +Moreover, the in- situ assembled peptide nano- blanket formed by FR17 inhibited tumor cells migration as well, with hardly any effect on cell viability (Supplementary Fig. 15). + +<|ref|>sub_title<|/ref|><|det|>[[147, 683, 629, 699]]<|/det|> +## FR17 administration inhibits melanoma lung metastasis in vivo. + +<|ref|>text<|/ref|><|det|>[[144, 728, 846, 899]]<|/det|> +First, we evaluated the anti- metastasis efficacy of FR17 on MCM- induced lung metastasis model. C57/BL6 mice were pre- induced by MCM administration for 10 days to set up the tumor supportive pre- metastatic niche in lung. Tumor cells were then injected through the tail vein on Day 7 to simulate the circulative tumor cells (CTC) wandering through blood vessels and eventually settled down in pulmonary PMN. As illustrated in Fig. 6a, peptide subcutaneous + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 845, 530]]<|/det|> +1 administration started from Day 3 and ended at 2 weeks- post lung tumor inoculation. The 2 metastasis inhibition efficacy was evaluated in terms of the tumor module number on Day 28. 3 FR17 effectively suppressed tumorigenesis and the development of metastasis in lung, compared 4 to both the control and TP5 groups (Fig. 6b- d). Moreover, the good outcomes of sFD17 treatment 5 which was close to that of FR17 group emphasized the main role of nano- blanket assembled by 6 the enzyme- activated released monomer in intervening PMN construction. Therefore, the in- situ 7 assembled nano- blanket also played a part in inhibiting metastasis development and growth. 8 Preliminary safety evaluation on these model mice at the end of the experiment revealed no 9 severe safety concern of FR17 administration (Supplementary Fig. 16- 17). The systemic immune 10 modulation effect of TP5 was reflected on the recovery of thymus shrinking as well 11 (Supplementary Fig. 16d, 17). What's more, peptide administration prevented tumor cells 12 infiltration into the spleen (Supplementary Fig. 17). + +<|ref|>text<|/ref|><|det|>[[144, 560, 845, 881]]<|/det|> +13 The anti- metastasis activity of FR17 was further verified on a post- surgery metastasis model30 14 (illustrated by Fig. 6e), which is closer to the clinical treatment of resectable melanoma 15 followed- up with adjuvant therapy for patients at high risk of recurrence. Here we compared 16 FR17 therapy with the rising- star of checkpoint inhibitors, PD1 antibody38- 39, which was 17 approved by FDA as adjuvant therapy on primary tumor excised with lymph node management 18 or metastatic melanoma40. As demonstrated in Fig. 6f, the recurrence of lung metastasis was put 19 off and the overall survival was greatly improved by FR17 therapy when compared with control 20 (Fig. 6g- i). The median overall survival prolonged for 23.3% (from 30 days to 37 days), catching 21 up with the outcomes of anti- PD1 treatment (from 30 days to 36 days). And the median + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 93, 847, 380]]<|/det|> +1 lung-metastasis-free survival prolonged from less than 19 days of both control group and 2 anti-PD1 treated group to 23 days of FR17 treated group. At the first time- point for lung 3 metastasis monitoring via bioluminescence imaging, lung metastasis has been observed in all 4 of the 10 mice from the control group, 8 mice from anti-PD1 treated group while only 3 mice 5 from FR17 treated group on Day 19. The retard of the occurrence of lung metastasis after FR17 6 administration emphasized the importance of PMN intervention. What's more, the relapse rate 7 of the primary tumor post-resection was reduced from 10 / 10 in the control group to 4 / 10 in 8 FR17 treated group (Supplementary Fig. 18). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[63, 98, 933, 428]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 446, 844, 464]]<|/det|> +
Figure 6. FR17 administration inhibited lung metastasis in vivo by retarding PMN
+ +<|ref|>text<|/ref|><|det|>[[100, 480, 848, 884]]<|/det|> +formation. a, Schedule of MCM- induced lung metastasis model with peptide treatment. Peptide administration started from Day 3 for metastasis prevention. b, Number of the metastatic nodules in the lung of different treatment groups (n = 7). Data is presented as mean ± SD. One- way ANOVA followed by Tukey's multiple comparisons test was employed for statistical evaluation. c, Images of the lung metastasis harvested on Day 28 from different treatment groups. d, Representative images of Hematoxylin & Eosin staining of lung sections from different treatment groups. The lung metastasis is circled by yellow dotted lines. Scale bar = 200 μm. e, Schedule of post- surgery metastasis model. Peptide administration started from Day 7 to 25. f, Representative in vivo bioluminescent images of mice without treatment or treated with FR17 or anti- PD1 (n = 10). The grey patches represent dead mice in the control group. g, Survival curves of the mice treated with FR17 or anti- PD1 or without treatment (n = 10). h, The cumulative + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 94, 845, 225]]<|/det|> +incidence of new pulmonary metastases in the mice treated with FR17 or anti- PD1 or without treatment \((\mathrm{n} = 10)\) . i, Semi- quantification of the in vivo bioluminescent signals in the lungs of the mice treated with FR17 or anti- PD1 or without treatment \((\mathrm{n} = 5)\) . Data is presented as mean ± SD. + +<|ref|>sub_title<|/ref|><|det|>[[110, 293, 255, 310]]<|/det|> +## 6 Conclusions + +<|ref|>text<|/ref|><|det|>[[140, 339, 845, 895]]<|/det|> +In summary, we have explored the magical retarding effect of the PMN microenvironment responsive- assembled peptide nano- blanket on fibroblasts activation, impeding PMN development. The enzyme- activatable assembled peptide FR17 can be enzyme- cleaved to release the self- assembly monomer FG8 to construct a lamellar structure, which is named as peptide nano- blanket. Experiments demonstrated that FR17 administration not only beat out the "flame" set up on resident fibroblasts induced by tumor derived factors, but also interrupted the subsequent PMN formation, including preventing pro- metastatic angiogenesis and vascular destabilization, and then intervening MDSCs' recruitment as well as their bio- functions. Astonishingly, when treated with sFD17, tumor- induced fibroblast activation was also impeded by peptide- assemblies without the assistance of TP5 so as to arrest the following pro- metastatic pathological process. This finding illustrated that the "flame- retarding" effect on fibroblast activation could be contributed by the drug- free peptide nano- blanket alone, presenting a broad application prospect of drug- free peptide assemblies to regulate PMN microenvironment and to prevent tumor distant metastasis. This work also elucidated the important role of pre- metastasis associated fibroblast in the complex interactions between major participators in PMN formation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 95, 844, 150]]<|/det|> +1 and metastasis development, suggesting that reprogramming or intervention on the key juncture could make a big difference on fighting against tumor metastasis. + +<|ref|>sub_title<|/ref|><|det|>[[102, 228, 245, 245]]<|/det|> +## 4 References + +<|ref|>text<|/ref|><|det|>[[101, 264, 846, 890]]<|/det|> +5 1. Gdowski, A. S., Ranjan, A. & Vishwanatha, J. K. Current concepts in bone metastasis, contemporary therapeutic strategies and ongoing clinical trials. J. Exp. Clin. Cancer Res. 36, 108-121 (2017). 8 2. Pashayan, N. & Paul, P. D. P. The challenge of early detection in cancer. Science (New York, N.Y.) 368, 589-590 (2020). 9 3. Rodriguez-Ruiz, A. et al. Stand-alone artificial intelligence for breast cancer detection in mammography: Comparison with 101 radiologists. J. Natl. Cancer Inst. 111, 916-922 (2019). 4. Poudineh, M., Sargent, E. H., Pantel, K., Kelley, S. O. Profiling circulating tumour cells and other biomarkers of invasive cancers. Nat. Biomed. Eng. 2, 72-84 (2018). 5. Christensen, E. et al. Early detection of metastatic relapse and monitoring of therapeutic efficacy by ultra-deep sequencing of plasma cell-free DNA in patients with urothelial bladder carcinoma. J. Clin. Oncol. 37, 1547-1557 (2019). 6. Abbosh, C. et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. Nature 545, 446-451 (2017). 7. Curti, B. D. & Faries, M. B. Recent advances in the treatment of melanoma. N. Engl. J. Med. 384, 2229-2240 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[100, 95, 848, 911]]<|/det|> +8. Oliver Sartor, M. D. & Johann S. de Bono, M. B. Metastatic prostate cancer. N. Engl. J. Med. 378, 645-657 (2018). +9. Zhang, C. et al. Radiotherapy and cytokine storm: Risk and mechanism. Front. Onco. 11, 670464 (2021). +10. Cardoso, F. et al. 70-Gene signature as an aid to treatment decisions in early-stage breast cancer. N. Engl. J. Med. 375, 717-29 (2016). +11. Peinado, H. et al. Pre-metastatic niches: organ-specific homes for metastases. Nat. Rev. Cancer 17, 302-317 (2017). +12. Liu, Y. & Cao, X. Characteristics and significance of the pre-metastatic niche. Cancer Cell 30, 668-681 (2016). +13. Zhou, Y., Han, M., & Gao, J. Q. Prognosis and targeting of pre-metastatic niche. J. Control. Release 325, 223-234 (2020). +14. Wang, T. T. et al. AIE/FRET-based versatile PEG-Pep-TPE/DOX nanoparticles for cancer therapy and real-time drug release monitoring. Biomater. Sci. 8, 118-124 (2020). +15. Guo, W. W. et al. Intracellular restructured reduced glutathione-responsive peptide nanofibers for synergetic tumor chemotherapy. Biomacromolecules 21, 444-453 (2020). +16. Zeng, F. L. et al. Clinical efficacy and safety of synthetic thymic peptides with chemotherapy for non-small cell lung cancer in China: A systematic review and meta-analysis of 27 randomized controlled trials following the PRISMA guidelines. Int. Immunopharmacol. 75, 105747 (2019). +17. Li, S. et al. Supramolecular nanofibrils formed by coassembly of clinically approved drugs for tumor photothermal immunotherapy. Adv. Mater. 33, e2100595 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 95, 845, 911]]<|/det|> +18. Qin, S. Y. et al. Theranostic GO-based nanohybrid for tumor induced imaging and potential combinational tumor therapy. Small 10, 599-608 (2014). +19. Kaplan, R. N. et al. VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche. Nature 438, 820-827 (2005). +20. Engblom, C., Pfirschke C. & Pittet M. J. The role of myeloid cells in cancer therapies. Nat. Rev. Cancer 16, 447-462 (2016). +21. Gabrilovich, D. I. Myeloid-derived suppressor cells. Cancer Immunol. Res. 5, 3-8 (2017). +22. Pein, M. et al. Metastasis-initiating cells induce and exploit a fibroblast niche to fuel malignant colonization of the lungs. Nat. Commun. 11, 1494 (2020). +23. Kong, J. et al. Extracellular vesicles of carcinoma-associated fibroblasts creates a pre-metastatic niche in the lung through activating fibroblasts. Molecular Cancer 18, 175 (2019). +24. Erler, J. T. et al. Hypoxia-induced lysyl oxidase is a critical mediator of bone marrow cell recruitment to form the premetastatic niche. Cancer Cell 15, 35-44 (2009). +25. Paolillo, M. & Schinelli, S. Extracellular matrix alterations in metastatic processes. Int. J. Mol. Sci. 20, 4947 (2019). +26. Asano, K. et al. Stromal versican regulates tumor growth by promoting angiogenesis. Sci. Rep. 7, 17225 (2017). +27. Zhou, X. et al. Melanoma cell-secreted exosomal miR-155-5p induce proangiogenic switch of cancer-associated fibroblasts via SOCS1/JAK2/STAT3 signaling pathway. J. Exp. Clin. Cancer Res. 37, 242 (2018). +28. Murgai, M. et al. KLF4-dependent perivascular cell plasticity mediates pre-metastatic niche + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 95, 848, 902]]<|/det|> +1 formation and metastasis. Nat. Med. 23, 1176- 1190 (2017). 2 29. Wang, Z. et al. Periostin promotes immunosuppressive premetastatic niche formation to facilitate breast tumour metastasis. J. Pathol. 239, 484- 495 (2016). 4 30. Long, Y. et al. Self- delivery micellar nanoparticles prevent premetastatic niche formation by interfering with the early recruitment and vascular destruction of granulocytic myeloid- derived suppressor cells. Nano Lett. 20, 2219- 2229 (2020). 7 31. Jiang, T. et al. Metformin and Docosahexaenoic Acid Hybrid Micelles for Premetastatic Niche Modulation and Tumor Metastasis Suppression. Nano lett. 19, 3548- 3562 (2019). 9 32. Lu, Z et al. Epigenetic therapy inhibits metastases by disrupting premetastatic niches. Nature 579, 284- 290 (2020). 11 33. Kaczanowska, S. et al. Genetically engineered myeloid cells rebalance the core immune suppression program in metastasis. Cell 184, 2033- 2052 (2021). 13 34. Lunin, S. M. et al. Thymic peptides restrain the inflammatory response in mice with experimental autoimmune encephalomyelitis. Immunobiology 218, 402- 7 (2013). 15 35. Cascinelli, N. et al. Evaluation of clinical efficacy and tolerability of intravenous high dose thymopentin in advanced melanoma patients. Melanoma Res. 8, 83- 9 (1998). 17 36. Wang, Y. et al. The novel role of thymopentin in induction of maturation of bone marrow dendritic cells (BMDCs). Int. Immunopharmacol. 21, 255- 60 (2014). 19 37. Lunin, S. M. et al. Thymus peptides regulate activity of RAW 264.7 macrophage cells: inhibitory analysis and a role of signal cascades. Expert Opin. Ther. Targets 15, 1337- 46 (2011). 22 38. Rizvi, N. A. et al. Activity and safety of nivolumab, an anti- PD- 1 immune checkpoint + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 95, 844, 315]]<|/det|> +1 inhibitor, for patients with advanced, refractory squamous non-small-cell lung cancer2 (CheckMate 063): a phase 2, single-arm trial. Lancet Oncol. 16, 257- 65 (2015).3 39. Gong, N. et al. Proton-driven transformable nanovaccine for cancer immunotherapy. Nat. Nanotechnol. 15, 1053- 1064 (2020).4 40. NIH. National Cancer Institute, Melanoma Treatment (PDQ#)- Health Professional Version.6 Available from: https://www.cancer.gov/types/skin/hp/melanoma-treatment-pdq#_402. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[150, 106, 297, 121]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[144, 150, 845, 435]]<|/det|> +This work was supported by the National Natural Science Foundation of China (Nos. 81673022). We thank Professor Yu Kang at College of Pharmaceutical Sciences, Zhejiang University for guidance on molecular dynamics simulation. We thank Qichun Wei's lab for providing the luciferase transfected B16F10. We thank Qin Han, Chenyu Yang, and Dandan Song at the Center of Cryo- Electron Microscopy (CCEM), Zhejiang University for their technical assistance on Confocal laser scanning microscopy and transmission electron microscopy. We thank Yanwei Li at the Core Facilities of Zhejiang University School of Medicine for technical assistance in flow cytometry analysis. + +<|ref|>sub_title<|/ref|><|det|>[[148, 494, 316, 508]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[144, 530, 847, 851]]<|/det|> +M.H. and Y.Z. conceived the project and designed the experiments. M.H. and J.G. supervised the project, discussed and commented on the manuscript. Y.Z. performed the majority of the experiments and data analysis. P.K. assisted with cell culture of B16F10, preparing MCM and animal experiments. Y.X. synthesized and characterized TPE- FR17 and TPE- FFKY. H.W. assisted with the establishment of animal models. Y.Z. performed the flow cytometry experiments with assistance from Z.Z., T.W., M.L., P.K., Y.S.L., Y.X., H.Z., X.Z. and Q.Y.; Y.Y.L. supported the operation of confocal microscope. Y.Z., Q.D., P.K., H.Z., Y.X., Y.S.L., X.J. and H.W. operated the tumor resection surgery. Y.Z. wrote the manuscript. All authors discussed the results and commented on the manuscript. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 96, 383, 111]]<|/det|> +1 Competing Interests statement + +<|ref|>text<|/ref|><|det|>[[110, 133, 480, 148]]<|/det|> +2 All the authors declare no conflicting interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[103, 103, 240, 123]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[144, 150, 844, 245]]<|/det|> +Characterization of FR17 and sFD17. Peptide FR17 and sFD17 were synthesized via Fmoc solid- phase peptide synthesis technology by APeptide Shanghai. The molecule structure and amino acid sequence of peptide were confirmed by mass spectra. + +<|ref|>text<|/ref|><|det|>[[103, 270, 120, 283]]<|/det|> +5 + +<|ref|>text<|/ref|><|det|>[[144, 322, 844, 339]]<|/det|> +Enzyme induced assembly of FR17 and sFD17. Peptide FR17 or sFD17 was dissolved in HBS + +<|ref|>text<|/ref|><|det|>[[144, 358, 844, 375]]<|/det|> +buffer (pH 7.4, containing \(50~\mathrm{mM}\) HEPES, \(150~\mathrm{mM}\) NaCl, \(10~\mathrm{mM}\) \(\mathrm{CaCl_2}\) ) at \(500~\mu \mathrm{M}\) . Activated + +<|ref|>text<|/ref|><|det|>[[144, 395, 844, 412]]<|/det|> +hMMP2 was added to the peptide solution at the working concentration of \(1\mu \mathrm{g / mL}\) . The + +<|ref|>text<|/ref|><|det|>[[144, 432, 844, 450]]<|/det|> +solutions were then incubated in \(37^{\circ}\mathrm{C}\) air- bath for \(24\mathrm{h}\) . The size of the enzyme treated sample, + +<|ref|>text<|/ref|><|det|>[[144, 470, 844, 487]]<|/det|> +which has formed the peptide nano- blanket, was measured by Zetasizer Nano ZS (Malvern + +<|ref|>text<|/ref|><|det|>[[144, 507, 844, 524]]<|/det|> +Instruments). The assembly morphology of the assembled nano- blanket was observed with + +<|ref|>text<|/ref|><|det|>[[144, 544, 844, 561]]<|/det|> +transmission electron microscope (FEI Tecnai G2 spirit) for TEM image and \(200\mathrm{kv}\) transmission + +<|ref|>text<|/ref|><|det|>[[144, 582, 590, 599]]<|/det|> +electron microscope (FEI Talos F200C) for Cryo- TEM image. + +<|ref|>text<|/ref|><|det|>[[144, 629, 120, 642]]<|/det|> +14 + +<|ref|>text<|/ref|><|det|>[[144, 680, 844, 699]]<|/det|> +All- atom molecular dynamics simulation. The molecular structures of peptide FFKY and FG8 + +<|ref|>text<|/ref|><|det|>[[144, 718, 844, 735]]<|/det|> +monomer were first constructed via AMBERTOOL based on the AMBER14SB force field. + +<|ref|>text<|/ref|><|det|>[[144, 755, 844, 773]]<|/det|> +Dynamics simulation runs were performed utilizing Gromacs 2018.4 package1. System + +<|ref|>text<|/ref|><|det|>[[144, 792, 844, 810]]<|/det|> +configurations were visualized using VMD software2, and generated into images mainly + +<|ref|>text<|/ref|><|det|>[[144, 830, 844, 847]]<|/det|> +employing GRACE software. The simulation was performed in water boxes containing 16 FFKY + +<|ref|>text<|/ref|><|det|>[[144, 867, 844, 884]]<|/det|> +or FG8 molecules. FFKY system was simulated containing NaCl to neutralize electric charge of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 94, 845, 455]]<|/det|> +the amidogen on the side- chain if Lys. Energy was minimized according to the steepest- descent method. Bond lengths were constrained by the LINCS algorithms. The nonbonded LJ interactions were cut off at 1.2 nm. Electrostatics was treated utilizing the Particle Mesh Ewald (PME) scheme. All production runs were simulated in the NPT ensemble using V- rescale coupling scheme with the temperature maintained at 298.15 K and parrinello- rahman coupling scheme with pressure kept at 1.0 bar and isotropic coupling type. The time constants for the pressure and temperature couplings were respectively set to 2.0 and 0.2 ps. Besides, the compressibility value was \(4.5 \times 10^{- 5} \mathrm{bar}^{- 1}\) . Periodic boundary conditions with a time step of 0.002 ps were adopted. Simulations were carried out for 200 ns and the structural coordinate information was recorded per 50 ps. + +<|ref|>text<|/ref|><|det|>[[148, 530, 844, 888]]<|/det|> +Cell lines and cell culture. B16F10 cells were purchased from Cell Bank of Chinese Academy of sciences (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium (DMEM, Cienry, China) containing \(10\%\) (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution ( \(100 \times\) , TBD, Tianjin, China) at \(1\%\) (v/v). And the luciferase transfected B16F10 (Luc- B16F10) was kindly gifted by Qichun Wei's lab, the Second Affiliated Hospital, Zhejiang University. The mouse lung fibroblasts (MLF) were purchased from iCell Bioscience Inc. (Shanghai, China) and cultured in Dulbecco's Modified Eagle Medium/Nutrient Mixture F- 12, 1:1 mixture (DMEM/F12 medium, Multicell, Wisent Int., Canada) containing \(10\%\) (v/v) fetal bovine serum (FBS, Gibco, Grand Island, USA) and penicillin-streptomycin Solution ( \(100 \times\) , TBD, Tianjin, China) at \(1\%\) (v/v). The mouse endothelial cells bEnd3 was kindly gifted + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 95, 847, 490]]<|/det|> +by Fuqiang Hu's lab, Institution of Pharmaceutics, College of Pharmaceutical Sciences, Zhejiang University. The bEnd3 cells were cultured in the same medium as B16F10. Cells were cultured in a cell incubator containing \(5\% \mathrm{CO}_2\) at \(37^{\circ}\mathrm{C}\) and passaged when reached \(80\% - 90\%\) confluence. The melanoma-conditioned medium (MCM) was obtained as follows: when B16F10 reached \(70 - 80\%\) confluence, washed with PBS and changed the medium to serum-free medium and incubated for \(24\mathrm{h}\) . The cell supernatants were collected, centrifuged at \(2000\mathrm{rpm}\) for \(10\mathrm{min}\) to discard the cell debris. The MLF-conditioned medium (FCM) was obtained as follows: MLFs were stimulated by MCM (supplemented with \(10\%\) (v/v) FBS) with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100\mu \mathrm{M}\) ) for \(48\mathrm{h}\) , then replaced with fresh complete medium and incubate for another \(24\mathrm{h}\) . The cell supernatants were collected, centrifuged at \(2000\mathrm{rpm}\) for \(10\mathrm{min}\) to discard the cell debris. + +<|ref|>text<|/ref|><|det|>[[101, 522, 123, 536]]<|/det|> +12 + +<|ref|>text<|/ref|><|det|>[[145, 568, 846, 888]]<|/det|> +Cell proliferation assays. MLFs in rapid proliferation were plated in 96- well plates at the density of \(4\times 10^{3}\) per well and cultured overnight. Former media were removed and the cells were cultivated with \(100~\mu \mathrm{L}\) MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100~\mu \mathrm{M}\) ). Cells cultivated with fresh complete medium was set up as control. After incubation for \(48\mathrm{h}\) , cell proliferation was measured using a Cell Counting Kit- 8 assay (Cat. 13E02A60, Boster Biotech., China) according to the manufacturer's instructions. The optical density at \(450\mathrm{nm}\) ( \(\mathrm{OD}_{450}\mathrm{nm}\) ) was measured using a multiwell plate reader (ELX800, BioTek, USA). Each group was repeated at least 4 times, and cell proliferation was presented as mean \(\pm\) SD. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 131, 846, 720]]<|/det|> +2 Cytokines secretion and gene expression of MLFs. An equal number of MLFs in rapid proliferation were seeded in a 24-well plate per well, cultivated with MCM supplemented with 10% (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in at least triplicate. Cells cultivated with fresh complete medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. MMP9 ELISA kit (EK0466, Boster Biotech., China) and VEGF ELISA kit (EK0541, Boster Biotech., China) were employed to measure the secretion level of MMP9 and VEGF in the cell supernatant. An equal number of MLFs in rapid proliferation were seeded in 24-well plate cultivated with (serum-free) MCM, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of 100 μM) for 48 h in quadruplicate. Cells cultivated with fresh DMEM/F12 medium was set up as control. The cell supernatants were collected, centrifuged at 2000 rpm for 5 min to discard cell debris. The secretion of FN was measured by FN ELISA kit (EK0351, Boster Biotech., China) according to the manufacturer's instructions. Cells with different treatments were harvested for RT-qPCR analysis to determine the gene expression level of M- Acta2 (Mouse αSMA), M- Mmp9, M- Vegfa and M- Fn1. The primer sequences are provided as follows: + +<|ref|>text<|/ref|><|det|>[[180, 737, 720, 910]]<|/det|> +M- Gapdh- F: 5' GGTGTCTCCTGCGACTTCA 3' 58.2 M- Gapdh- R: 5' TGGTCCAGGGTTTCTTACTCC 3' 58.2 183bp M- Vegfa- F: 5' GCTACTGCCGTCCGATTGAG 3' 60.87 M- Vegfa- R: 5' ACTCCAGGGCTTCATCGTTACAG 3' 62.25 132bp M- Mmp9- F: 5' CACAGCCAACTATGACCAGGAT 3' 59.1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 93, 720, 112]]<|/det|> +1 M-Mmp9-R: 5' CAGGAAGACGAAGGGGAAGA 3' 59.1 115bp + +<|ref|>text<|/ref|><|det|>[[110, 133, 648, 150]]<|/det|> +2 M-Fn1-F: 5' CTATTTACCAACCGCAGACTCAC 3' 58.7 + +<|ref|>text<|/ref|><|det|>[[110, 171, 704, 189]]<|/det|> +3 M-Fn1-R: 5' TGCTTGTTTCCTTGCGACTT 3' 58.5 115bp + +<|ref|>text<|/ref|><|det|>[[110, 210, 648, 226]]<|/det|> +4 M-Acta2-F: 5' CAACTGGTATTGTGCTGGACTC 3' 57.3 + +<|ref|>text<|/ref|><|det|>[[110, 248, 714, 265]]<|/det|> +5 M-Acta2-R: 5' ATCTCACGCTCGGGAGTAGT 3' 57.3 181bp + +<|ref|>text<|/ref|><|det|>[[110, 297, 123, 310]]<|/det|> +6 + +<|ref|>text<|/ref|><|det|>[[110, 340, 843, 359]]<|/det|> +7 Cell Migration assays. MLFs in rapid proliferation were plated in Culture- Inserts (2 Well, Ibidi, + +<|ref|>text<|/ref|><|det|>[[110, 379, 841, 397]]<|/det|> +8 Germany) at the density of \(1 \times 10^{5}\) in \(70 \mu \mathrm{L}\) per well and grew to confluence overnight. + +<|ref|>text<|/ref|><|det|>[[110, 417, 843, 435]]<|/det|> +9 Culture- Inserts as well as the former medium were gently removed. Then the MLFs were + +<|ref|>text<|/ref|><|det|>[[110, 455, 843, 473]]<|/det|> +10 cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs + +<|ref|>text<|/ref|><|det|>[[110, 493, 843, 511]]<|/det|> +11 (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) in triplicate. Cells cultivated with fresh + +<|ref|>text<|/ref|><|det|>[[110, 531, 843, 549]]<|/det|> +12 complete medium was set up as control. Pictures of the cell scratches were taken under the + +<|ref|>text<|/ref|><|det|>[[110, 569, 843, 587]]<|/det|> +13 microscope at \(0 \mathrm{h}\) . After incubated for \(24 \mathrm{h}\) , MLFs were fixed and stained with crystal violet. + +<|ref|>text<|/ref|><|det|>[[110, 608, 480, 624]]<|/det|> +14 Pictures were taken and analyzed with ImageJ. + +<|ref|>text<|/ref|><|det|>[[110, 656, 123, 669]]<|/det|> +15 + +<|ref|>text<|/ref|><|det|>[[110, 701, 843, 719]]<|/det|> +16 Collagen gel contraction assay. An equal number of MLFs seeded in \(6 \mathrm{cm}\) dishes were + +<|ref|>text<|/ref|><|det|>[[110, 739, 843, 757]]<|/det|> +17 cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs + +<|ref|>text<|/ref|><|det|>[[110, 777, 843, 795]]<|/det|> +18 (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) for \(48 \mathrm{h}\) . Cells cultivated with fresh + +<|ref|>text<|/ref|><|det|>[[110, 815, 843, 833]]<|/det|> +19 complete medium was set up as control. The pre- treated MLFs were digested and re- suspended + +<|ref|>text<|/ref|><|det|>[[110, 853, 843, 871]]<|/det|> +20 at the density of \(2 \times 10^{6}\) per mL, kept on ice for later use. The neutral collagen solution was + +<|ref|>text<|/ref|><|det|>[[110, 891, 843, 909]]<|/det|> +21 prepared as follows: \(224 \mu \mathrm{L}\) type I collagen gel ( \(3 \mathrm{mg / ml}\) , Cat. C8062, Solarbio, China) was + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 93, 846, 300]]<|/det|> +quickly mixed with \(100~\mu \mathrm{L}\) conditional medium and \(8~\mu \mathrm{L}\) NaOH (0.1 N) on ice. Then \(300~\mu \mathrm{L}\) of the pre- treated MLFs suspension was added to the collagen solution on ice immediately. And the neutral cell- collagen mixture was added to 48- well plates \(200~\mu \mathrm{L}\) per well in triplicate and allowed to solidify for \(45\mathrm{min}\) at room temperature. After incubated at \(37^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) , the gels were photographed. ImageJ software was used to measure gel area and evaluate contraction. Gel contraction was assessed as the ratio of the gel area to the area of the well. + +<|ref|>text<|/ref|><|det|>[[110, 370, 845, 585]]<|/det|> +Tube forming assay. Mouse endothelial bEnd3 cells seeded in \(6\mathrm{cm}\) dishes were cultivated with conditional FCM (from MLF previously stimulated by MCM with or without TP5, sFD17 or FR17 peptide as illustrated above) for \(24\mathrm{h}\) . The cells were harvested and seeded in 48- well plates pre- coated with Matrigel (Cat. 356230, BD, USA) in triplicate for each group. After \(6\mathrm{h}\) incubation, five visual fields were randomly chosen from each well and photographed by microscope (CKX53, OLYMPUS, Japan). The tube forming results were analyzed by ImageJ. + +<|ref|>text<|/ref|><|det|>[[101, 616, 845, 907]]<|/det|> +Transwell permeability assay. An equal number of MLFs seeded in \(6\mathrm{cm}\) dishes were cultivated with MCM supplemented with \(10\%\) (v/v) FBS, treated with or without peptide drugs (TP5, sFD17, FR17 at the concentration of \(100~\mu \mathrm{M}\) ) for \(48\mathrm{h}\) . Cells cultivated with fresh complete medium was set up as control. The pre- treated MLFs were digested and seeded at the lower well of a 24- well transwell plate at \(2.5 \times 10^{5}\) cells per well. Each group was triplicate. Then, 7,000 Mouse endothelial bEnd3 cells were seeded on a \(0.4\mu \mathrm{m}\) Transwell insert (Cat. 3413, Corning Costar, USA) above the top of the well until grown to confluence. Rhodamine B- dextran (70 kDa, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 92, 846, 304]]<|/det|> +1 Cat. R9379, Sigma-Aldrich, USA) was added to the upper insert on the endothelial cell layer. 2 After 1 h incubation, the translocation of Rhodamine B-dextran from the insert to the lower well 3 passing through the endothelial cell layer was measured by a microplate reader (SPARK, 4 TECAN, Switzerland) at an excitation/emission wavelength of \(540 / 625 \mathrm{nm}\) . The relative 5 permeability of the cell layer was normalized by diving the fluorescence signals of the treatment 6 groups by the control group. + +<|ref|>text<|/ref|><|det|>[[110, 360, 845, 814]]<|/det|> +8 Integrality of the endothelial cell monolayer. Mouse endothelial bEnd3 cells grown in 35- mm confocal dishes to \(100\%\) confluence were treated with conditional FCM, which was obtained from MLFs stimulated by MCM with or without the treatment of different peptides (TP5, sFD17, FR17 at the concentration of \(100 \mu \mathrm{M}\) ) as illustrated above. After incubated with conditional FCM for \(24 \mathrm{h}\) , the single endothelial cell layer was gently washed with PBS and fixed with \(4\%\) paraformaldehyde for \(15 \mathrm{min}\) , permeabilized with \(0.2\%\) Triton X- 100 for \(15 \mathrm{min}\) and blocked with \(2\%\) bovine serum albumin (BSA) for another \(15 \mathrm{min}\) . Cells were incubated with anti- VE- cadherin antibody (1:1000, Cat. Ab205336, Abcam, UK) containing \(0.2\%\) BSA and \(0.1\%\) Triton X- 100 in PBS at \(4^{\circ} \mathrm{C}\) overnight. Cells were washed with PBS for three times and incubated for \(1 \mathrm{h}\) with AF647 labeled goat anti- rabbit IgG (H+L) (1:100, Cat. 33113ES60, Yeasen, China). The nuclei were labeled by DAPI solution (ready- to- use) (Solarbio, China). Images were taken by confocal microscope (Leica, German). + +<|ref|>text<|/ref|><|det|>[[100, 890, 847, 908]]<|/det|> +21 Mice and animal models. C57BL/6 mice (male, 5- week- old) purchased from Slaccas (Shanghai, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 95, 845, 225]]<|/det|> +1 China) were adaptive fed for more than one week for subsequent experiments. The animals were maintained under standard laboratory housing conditions where foods and water can be reached freely. All the animal experiments were conducted following the guidelines which have been approved by the Ethics Committee of Zhejiang University. + +<|ref|>text<|/ref|><|det|>[[110, 255, 845, 500]]<|/det|> +5 For MCM- induced lung metastasis model, MCM (300 \(\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days from Day 1 to 10. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells ( \(1 \times 10^{5}\) per mice) was given to the mice. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS® Spectrum In Vivo Imaging System, PerkinElmer, USA) if viable. For bioluminescence imaging, mice were intraperitoneally injected with D- luciferin potassium salt (150 mg/kg, Gold Biotechnology, USA). The bioluminescence of pulmonary metastases was detected 10 min later. + +<|ref|>text<|/ref|><|det|>[[110, 530, 845, 776]]<|/det|> +12 For post- surgery metastasis model, \(1 \times 10^{6}\) B16F10 cells were subcutaneously inoculated in the back of the C57BL/6 mice (male) aged 6 weeks above the right hindlimb on Day 1. On Day 12, tumor resection surgery was conducted on mice to remove the entire tumor tissues as well as the skin cover the tumor under anesthesia. On the next day, \(1 \times 10^{5}\) luciferase- expressing B16F10 cells were injected into mice through tail vein. The tumor recurrence and lung metastasis were monitored twice a week. The tumor volume and body weight were recorded twice a week. Tumor volume was calculated as (width \(^{2} \times\) length) / 2. + +<|ref|>text<|/ref|><|det|>[[110, 805, 845, 910]]<|/det|> +19 Pre- metastatic niche study. MCM (300 \(\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells ( \(1 \times 10^{5}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 90, 848, 900]]<|/det|> +1 per mice) was given to the mice. On Day 3, 6, 10, 13, mice were euthanized for cardiac perfusion2 and the lung tissues were collected for further analysis.3 For Western-blot analysis, total proteins from the lung tissues were extracted using Tissue4 Protein Extraction Reagent (T- PER™, Cat. 78510, Thermo Pierce, Thermo Scientific, USA) and5 quantified with a Bradford Protein Assay Kit (Cat. P0010, Beyotime, Beijing, China). Samples6 (60 μg) were separated on 10% or 8% SDS- PAGE gels, then transferred to PVDF nitrocellulose7 membrane (Cat. IPVH00010, Merck Millipore). Membranes were incubated with the appropriate8 primary antibodies in 3% BSA, including Fibronectin (1:500, Cat. ab2413, Abcam, UK), MMP9 (1:1000, Cat. ab38898, Abcam, UK), VEGFa (1:500, Cat. ab119, Abcam, UK), TGF- β1 (1:1000,10 Cat. ab179695, Abcam, UK), iNOS (1:1000, Cat. ab204017, Abcam, UK), Arginase 1 (1:1000,11 Cat. ab124917, Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was12 used as control. After incubation with appropriate secondary antibody Goat anti- Mouse IgG13 (H+L) (1:5000, Cat. 31160, Thermo Pierce) or Goat anti- Rabbit IgG (H+L) (1:5000, Cat. 31210,14 Thermo Pierce), the intensity of the immunoreactive proteins was stabilized by SuperSignal®15 West Dura Extended Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X- ray16 film.17 For ELISA analysis, the lung tissues were ground and centrifuged to gain supernatant to measure18 the MMP2 (Cat. OM457413, Omnimabs, USA) and ROS (Cat. OM641674, Omnimabs, USA)19 level.20 For immunofluorescence staining, the left lung was fixed with 4% paraformaldehyde and 30%21 sucrose solution overnight, and embedded into paraffin and sliced into sections. The paraffin + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 92, 848, 922]]<|/det|> +1 lung sections were deparaffinized and rehydrated, then stained with primary antibodies: αSMA2 (1:500, Cat. Ab7817, Abcam, UK), or CD34 (1:500, Cat. Ab81289, Abcam, UK). Secondary3 antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111-165-003, Jackson, USA) was4 utilized in 1:500 dilution and stained with DAPI before observation.5 For flow cytometry analysis, lung tissues harvested from mice were mechanically minced into6 1- 2 mm pieces using scissors and then dissociated into single cell suspension at 37 °C on a7 shaker for 30 min by enzymes. The digesting solution contains 2 mg/mL collagenase I (Cat.8 BS163, BioSharp, Germany), 2 mg/mL collagenase II (Cat. BS164, BioSharp, Germany) and9 DNase I (Cat. KGF008, KeyGEN BioTech., China). Digestion was stopped by adding 2 volumes10 PBS and filtered through a 70 μM cell strainer (Cat. CSS013070, Jet BIOFIL®, China). The cell11 suspension was centrifuged at 400 g for 5 min to discard the supernatant. Cell precipitations were12 then resuspended in 5 mL RBC lysis buffer (Cat. R1010, Solarbio, China) and centrifuged again13 to discard the supernatant. The single-cell-suspensions washed with PBS and resuspended were14 incubated with FITC-antimouse-CD45 (Cat. 553079, BD, USA), PE-antimouse-NK1.1 (Cat.15 108708, Biolegend, USA), PE-antimouse-CD3 (Cat. 100205, Biolegend, USA),16 PE-antimouse-TER119 (Cat. 116207, Biolegend, USA), PE-antimouse-CD19 (Cat. 152407,17 Biolegend, USA), APC-antimouse-CD11b (Cat. 101211, Biolegend, USA),18 BV605-antimouse-MHC II (Cat. 107639, Biolegend, USA), BB700-CD11c (Cat. 566505, BD,19 USA), BV421-antimouse-Ly6c (Cat. 562727, BD, USA) and PE/CF594-antimouse-Ly6g (Cat.20 562700, BD, USA), BV711-antimouse-F4/80 (Cat. 123147, Biolegend, USA),21 PE/Cy7-antimouse-CD103 (Cat. 121426, Biolegend, USA) antibodies in 100 μL 1% BSA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 93, 847, 301]]<|/det|> +containing \(50~\mu \mathrm{L}\) BD Brilliant Stain Buffer for \(30\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) in dark. After centrifuged and washed with PBS, cell pellets were fixed and membrane were perforated with Fix/Perm Buffer (Cat. 562574, BD, USA). The cell pellets were then stained with BV650-antimouse-CD206 (Cat. 141723, Biolegend, USA) for \(40\mathrm{min}\) in the dark at room temperature. After centrifuged and washed with PBS, the stained cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software. + +<|ref|>text<|/ref|><|det|>[[108, 330, 845, 501]]<|/det|> +To investigate the impact of PMN formation induced by MCM injection, metastasis development and mice survival were monitored on MCM- induced PMN model and on the mice that didn't receive MCM injection but were inoculated directly with Luc- B16F10 cells \((1\times 10^{5}\mathrm{per}\) mice) on Day 7. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA). + +<|ref|>sub_title<|/ref|><|det|>[[145, 579, 504, 595]]<|/det|> +## Influence of peptide interference on mice PMN + +<|ref|>text<|/ref|><|det|>[[108, 624, 845, 833]]<|/det|> +For mice PMN model, MCM \((300~\mu \mathrm{L}\) per mice) was intraperitoneally injected to the mice for 10 consecutive days. On Day 7, a tail vein injection of B16F10 or Luc- B16F10 cells \((1\times 10^{5}\mathrm{per}\) mice) was given to the mice. Mice were randomly divided into 4 groups, namely control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at \(40~\mu \mathrm{M / kg / day}\) . On Day 10, mice were euthanized for cardiac perfusion and the lung tissues were collected for further analysis. + +<|ref|>text<|/ref|><|det|>[[108, 863, 844, 880]]<|/det|> +The lung tissues harvested from different groups were fixed, embedded into paraffin and sliced + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 847, 901]]<|/det|> +into sections. To visualize the activation of lung fibroblasts and angiogenesis in pulmonary PMN, the paraffin lung sections after deparaffinization and rehydration, were stained with primary antibodies: αSMA (1:500, Cat. ab7817, Abcam, UK), or CD34 (1:500, Cat. ab81289, Abcam, UK). Secondary antibody Cy3 conjugated goat anti-rabbit IgG (Cat. 111- 165- 003, Jackson, USA) was utilized in 1:500 dilution and stained with DAPI before observation. To visualize the extracellular matrix environment alteration in pulmonary PMN, the lung sections were stained with appropriate primary antibodies: MMP2 (1:200, Cat. 10373- 2- ap, PTG, USA), or MMP9 (1:1000, Cat. ab228402, Abcam, UK), Secondary antibody Cy3 conjugated goat anti-rabbit IgG (1:500, Cat. 111- 165- 003, Jackson, USA) and DAPI. To visualize the collagen deposition in pulmonary PMN, Masson's trichrome staining of the lung sections were imaged and analyzed by ImageJ to calculate the collagen volume fraction (CVF) by dividing the blue collagen area by total tissue area. And the Sirius Red Staining was also carried out and the sections were visualized under the polarizing microscope (Nikon Eclipse Ci). To investigate the recruitment of MDSC in PMN, serial sections of lung tissues were stained with periostin (1:200, Cat. 19899- 1- AP, PTG, USA), or co- stained with LOX (1:200, Cat. ab174316, Abcam, UK) and Fibronectin (1:200, Cat. ab92572, Abcam, UK) antibody, or CD11b (1:2000, Cat. ab133357, Abcam, UK) and Gr- 1 (1:200, Cat. ab25377, Abcam, UK) antibody separately. Secondary antibody Cy3 conjugated goat anti- rabbit IgG (1:500, Cat. 111- 165- 003, Jackson, USA), goat anti- rabbit IgG conjugated to HRP (1:2000, Cat. ab6721, Abcam, UK) and fluorescent TSA- 488 (1:200, Wuhan Pinuofei, China) were applied according to Tyramide Signal Amplification technology. The stained sections were observed and imaged under the confocal microscope. Images were analyzed by ImageJ if necessary. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 90, 848, 760]]<|/det|> +1 To investigate the alteration of protein expression level in PMN, Western- blot or ELISA 2 experiments were carried out. For Western- blot assay, protein samples were extracted separately 3 from three independent mice from each group. Total proteins from the lung tissues were 4 extracted using Tissue Protein Extraction Reagent (T- PERTM, Cat. 78510, Thermo Pierce, 5 Thermo Scientific, USA) and quantified with a Bradford Protein Assay Kit (Cat. P0010, 6 Beyotime, Beijing, China). Samples (60 μg) were separated on 10% or 8% SDS- PAGE gels, then 7 transferred to PVDF nitrocellulose membrane (Cat. IPVH00010, Merck Millipore). Membranes 8 were incubated with the appropriate primary antibodies in 3% BSA, including Fibronectin (1:500, 9 Cat. ab2413, Abcam, UK), Versican (1:1000, Cat. ab270445, Abcam, UK), VEGFa (1:500, Cat. 10 ab119, Abcam, UK), ANG2 (1:500, Cat. ab155106, Abcam, UK), MMP9 (1:1000, Cat. ab38898, 11 Abcam, UK), MMP2 (1:500, Cat. ab97779, Abcam, UK), TGF- \(\beta 1\) (1:1000, Cat. ab179695, 12 Abcam, UK). Antibody against GAPDH (1:10000, Cat. ab181602, Abcam) was used as control. 13 After incubation with secondary antibody Goat anti- Mouse IgG (H+L) (1:5000, Cat. 31160, 14 Thermo Pierce) or Goat anti- Rabbit IgG (H+L) (1:5000, Cat. 31210, Thermo Pierce), the 15 intensity of the immunoreactive proteins was stabilized by SuperSignal® West Dura Extended 16 Duration Substrate (Cat. 34075, Thermo Pierce) and visualized on X- ray film. For ELISA assay, 17 lung tissues were ground and centrifuged to gain supernatant to measure the IL- 6 (Cat. EK0411, 18 Boster, China) level. + +<|ref|>text<|/ref|><|det|>[[100, 828, 844, 888]]<|/det|> +20 In vivo vascular permeability assay. MCM (300 μL per mice) was intraperitoneally injected to 21 the mice for 7 consecutive days. And the mice were randomly divided into 4 groups, namely + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 94, 846, 456]]<|/det|> +control, TP5, sFD17 and FR17. Peptides, including TP5, sFD17 and FR17, were administrated separately to the mice subcutaneously from Day 3 at \(40~\mu \mathrm{M / kg / day}\) . On day 7, \(100~\mathrm{mg / kg}\) Rhodamine B-dextran ( \(70\mathrm{kDa}\) , Cat. R9379, Sigma-Aldrich, USA) was intravenously injected to the mice. After \(3\mathrm{h}\) , mice were injected with FITC-lectin (Cat. L0770, Sigma-Aldrich, USA) at \(10\mathrm{mg / kg}\) through the tail- vein. Ten minutes later, each mouse was anesthetized and transcardiac perfused with \(20\mathrm{mL}\) saline to remove the excess dye and followed by \(5\mathrm{mL}\) of \(4\%\) formaldehyde. The lung tissues were formaldehyde- fixed and cryo- sectioned. Slices were observed and imaged by fluorescence microscopy for vascular leakage. There were 3 mice in each group and 5 visual fields were randomly chosen for each section. The relative vascular permeability was analyzed by dividing the dye leakage of each group by healthy control. + +<|ref|>text<|/ref|><|det|>[[148, 530, 845, 775]]<|/det|> +Recruitment of MDSC to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and mechanically minced and digested to obtain the single- cell- suspensions as described above. The single- cell- suspensions washed with PBS and resuspended were incubated with APC- antimouse- CD11b (Cat. 101211, Biolegend, USA) and PE/Cy7- antimouse- Ly6g (Cat. 127617, Biolegend, USA) antibodies in \(100~\mu \mathrm{L}\) \(1\%\) BSA for 30 min at \(4^{\circ}\mathrm{C}\) in dark. After centrifuged and washed with PBS, cell pellets were analyzed by BD Fortessa flow cytometry. The data were analyzed using FlowJo software. + +<|ref|>text<|/ref|><|det|>[[148, 852, 845, 907]]<|/det|> +mRNA sequencing of CD11b+Ly6g+ MDSC recruited to PMN. On Day 10 of PMN mice model, lung tissues were harvested from different treatment groups and digested into single cells + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 95, 845, 416]]<|/det|> +as introduced as above. The single-cell-suspensions after washing with PBS and re-suspension were incubated with APC-antimouse- CD11b(Cat. 101211, Biolegend, USA) and PE/Cy7-antimouse-Ly6g(Cat. 127617, Biolegend, USA) antibodies in \(100~\mu \mathrm{L}1\%\) BSA for 30 min at \(4^{\circ}\mathrm{C}\) in the dark. After centrifuged and washed with PBS, CD11b \(^+\) Ly6g \(^+\) MDSCs were sorted from the PMN lungs of 10- 12 individual mice from each group per sample by FACS (Beckman moflo Astrios EQ). Total RNA was extracted by TRIzol for cDNA preamplification using the NEBNext® Ultra™ Directional RNA Library Prep Kit for Illumina®, then analyzed using Qubit2.0 Fluorometer, Agilent 2100 bioanalyzer and qRT-PCR. Significantly enriched gene sets were defined as \(P\) values \(< 0.05\) comparing to the control group. + +<|ref|>text<|/ref|><|det|>[[148, 494, 845, 890]]<|/det|> +Inhibition of tumor metastasis on MCM- induced PMN lung metastasis model. The PMN models were established as introduced above, mice were randomly divided into 4 groups (n = 7), namely control, TP5, sFD17 and FR17. From day 3 to 21, mice from different groups were subcutaneously administrated with saline, TP5 (40 \(\mu \mathrm{M / kg}\) per day), sFD17 (40 \(\mu \mathrm{M / kg}\) per day) and FR17 (40 \(\mu \mathrm{M / kg}\) per day) separately. Body weight was recorded every 3 days. On day 20, blood was collected from the submarginal ocular venous plexus under anesthesia for blood tests including complete blood count, alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN) and serum creatinine (CREA). On day 28, mice were euthanized and major organs, such as heart, liver, spleen, lung, kidney and thymus were collected after cardiac perfusion. Lung tumor nodules were then counted under the stereo microscope. The major organs were fixed, embedded into paraffin for Hematoxylin & Eosin staining. The thymus + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 94, 844, 152]]<|/det|> +coefficient was calculated as the thymus weight divided by the body weight and the spleen coefficient was calculated as the spleen weight divided by the body weight. + +<|ref|>text<|/ref|><|det|>[[103, 178, 850, 590]]<|/det|> +Inhibition of tumor metastasis post- surgery. The post- surgery metastasis model was established as illustrated above. Mice were randomly divided into 3 groups, namely control, anti- PD1 and FR17. For FR17 treatment, peptide was subcutaneously administrated to the mice from Day 7 to Day 25 at the dose of \(40~\mu \mathrm{M / kg}\) per day. For anti- PD1 treatment, \(100\mathrm{mg}\) anti- PD1 (Cat. BE0146, Bio X Cell, USA) was given by \(i.p\) . injection twice per week starting from day 3 post tumor resection and given two times per week from Day 15 to Day 25 for a total of 4 times. Lung metastasis was monitored twice a week by bioluminescence imaging (IVIS Spectrum, USA) until death. The recurrence of the excised subcutaneous tumor was closely monitored and tumor volume was measured and calculated following the ellipsoid volume formula: (width \(^2\times\) length) / 2. + +<|ref|>sub_title<|/ref|><|det|>[[148, 664, 219, 679]]<|/det|> +## Statistics + +<|ref|>text<|/ref|><|det|>[[144, 709, 843, 728]]<|/det|> +Statistical analysis was performed using GraphPad Prism 8.0.1 (GraphPad Software, CA, USA). + +<|ref|>text<|/ref|><|det|>[[144, 749, 843, 767]]<|/det|> +Data were presented as means \(\pm\) SD. Statistical evaluation of differences between experimental + +<|ref|>text<|/ref|><|det|>[[144, 788, 841, 805]]<|/det|> +groups was performed by one- way ANOVA followed by Tukey's multiple comparisons test. + +<|ref|>text<|/ref|><|det|>[[144, 826, 561, 842]]<|/det|> +Statistical significance was considered at least at \(p< 0.05\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 97, 848, 202]]<|/det|> +1 Abraham, M. J. et al. GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. \*SoftwareX\*, 1, 19- 25 (2015).2 Humphrey, W., Dalke A. & Schulten, K. VMD: visual molecular dynamics. \*J. molecular graphics\*, 14, 33- 38 (1996). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 43, 311, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 501, 230]]<|/det|> +- SIV9.pdf- SIV9withoutauthorsinformation.pdf- SIV9withoutauthorsinformation.pdf- nrreportingsummary20210809T100118.224.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/images_list.json b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..80651e1d1436a8dda1d76b9b3cf1d86994c7ed85 --- /dev/null +++ b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/images_list.json @@ -0,0 +1,78 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Genome wide scRNA-seq probe design and synthesis", + "footnote": [], + "bbox": [ + [ + 115, + 90, + 615, + 666 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Microfluidic probe-based scRNA-seq method and validation.", + "footnote": [], + "bbox": [ + [ + 131, + 90, + 608, + 722 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. scRNA-seq analysis reveals known and novel states in B subtilis.", + "footnote": [], + "bbox": [ + [ + 112, + 130, + 884, + 844 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2 Microfluidic probe- based scRNA- seq method and validation. a. Cells are fixed and permeabilized to allow penetration of thousands of unique, genome- specific oligonucleotide probes. Hybridized probes \"retrofit\" transcripts with a poly- A tail & UMI while unhybridized probes are washed away. b. Cells are then flowed through a commercial microfluidic device which encapsulates single cells into droplets (shown in c.) containing barcoded primers conjugated to a hydrogel microsphere and PCR reagents. d. Barcoded cDNA", + "footnote": [], + "bbox": [ + [ + 52, + 55, + 600, + 777 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [], + "page_idx": 33 + } +] \ No newline at end of file diff --git a/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd.mmd b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6c41d19e00ac998dad388fd613d1988ea469e7fd --- /dev/null +++ b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd.mmd @@ -0,0 +1,463 @@ + +# Droplet-based single cell RNA sequencing of bacteria identifies known and previously unseen cellular states + +Adam Rosenthal ( adam.rosenthal@iff.com) + +Dupont + +Ryan McNulty Dupont + +Duluxan Sritharan Harvard University + +Shichen Liu Dana- Farber Cancer Institute + +Sahand Hormoz + +Harvard Medical School / Dana- Farber Cancer Institute https://orcid.org/0000- 0002- 4384- 4428 + +## Biological Sciences - Article + +Keywords: Clonal Bacterial Populations, Specialized Cell States, Isogenic Bacterial Populations, Metabolic Pathway Segregation + +Posted Date: March 22nd, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 322406/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Droplet-based single cell RNA sequencing of bacteria identifies known and previously unseen cellular states + +Ryan McNulty \(^{1*}\) , Duluxan Sritharan \(^{2,3*}\) , Shichen Liu \(^{3}\) , Sahand Hormoz \(^{3,4,5\#}\) and Adam Z. Rosenthal \(^{1\#\wedge}\) + +## Affiliations + +\(^{1}\) DuPont Nutrition and Biosciences, Wilmington, DE 19803, USA \(^{2}\) Harvard Graduate Program in Biophysics, Harvard University, Cambridge, MA, USA \(^{3}\) Department of Data Science, Dana- Farber Cancer Institute, Boston, MA, USA \(^{4}\) Department of Systems Biology, Harvard Medical School, Boston, MA, USA \(^{5}\) Broad Institute of MIT and Harvard, Cambridge, MA, USA + +\*These authors contributed equally to this paper\#Correspondence: Sahand Hormoz@hms.harvard.edu and adam.rosenthal@iff.com\^current affiliation: IFF Health and Biosciences + +<--- Page Split ---> + +## Abstract + +Clonal bacterial populations rely on transcriptional variation to differentiate into specialized cell states that increase the community's fitness. Such heterogeneous gene expression is implicated in many fundamental microbial processes including sporulation, cell communication, detoxification, substrate utilization, competence, biofilm formation, motility, pathogenicity, and antibiotic resistance1. To identify these specialized cell states and determine the processes by which they develop, we need to study isogenic bacterial populations at the single cell level2,3. Here, we develop a method that uses DNA probes and leverages an existing commercial microfluidic platform (10X Chromium) to conduct bacterial single cell RNA sequencing. We sequenced the transcriptome of over 15,000 individual bacterial cells, detecting on average 365 transcripts mapping to 265 genes per cell in B. subtilis and 329 transcripts mapping to 149 genes per cell in E. coli. Our findings correctly identify known cell states and uncover previously unreported cell states. Interestingly, we find that some metabolic pathways segregate into distinct subpopulations across different bacteria and growth conditions, suggesting that some cellular processes may be more prone to differentiation than others. Our high throughput, highly resolved single cell transcriptomic platform can be broadly used for understanding heterogeneity in microbial populations. + +<--- Page Split ---> + +## Introduction + +1 In recent years, single cell RNA sequencing (scRNA- seq) has enabled the agnostic assessment of transcriptional heterogeneity in eukaryotic systems. Since the introduction of microfluidic techniques which allow encapsulation of cells and tagging of poly- adenylated transcripts in small droplets, scRNA- seq studies of mammalian populations have grown exponentially. However, tools to comprehensively study differential gene expression in bacterial populations remain severely limited due to several significant technical challenges. First, total mRNA abundance in bacteria is two orders of magnitude lower than that of eukaryotes, with a single bacterial cell containing approximately \(10^{3} - 10^{4}\) transcripts during exponential growth4. Second, transcriptional turnover is much faster in bacteria with mRNA half- life often on the scale of minutes, compared with hours for eukaryotic transcripts4. Third, bacterial transcripts do not intrinsically include a 3' poly- adenosine tail and, therefore, mRNA cannot easily be tagged and selectively enriched against rRNA, which makes up \(>95\%\) of the bacterial transcriptome4. Lastly, accessing mRNA requires cell permeabilization which is markedly more difficult in bacteria due to the diversity in membrane structure and variability of the peptidoglycan layer. We reasoned that a new method combining the advantages of microfluidic single cell barcoding in droplets with the ability to tag transcripts using in- situ hybridization of oligonucleotide probes could overcome these challenges. Here, we present a method for prokaryotic single cell RNA sequencing which utilizes a commercial, benchtop microfluidic device (Chromium Controller from 10X Genomics) and custom ssDNA probe libraries to resolve the mRNA profile of thousands of Gram positive (Bacillus subtilis 168) and Gram negative (Escherichia coli MG1655) bacterial cells. + +## Method Development + +## Probe Design & Library Generation + +To leverage existing microfluidic single cell sequencing platforms, we devised a method relying on tagging of individual transcripts with DNA probes. This approach requires the generation of a large oligonucleotide library that is complementary to all protein coding sequences within a genome (Figure 1). Multiple DNA regions of 50 bp were chosen from each ORF based on uniqueness as determined by UPS2 software5 or based on previously published oligonucleotide arrays6,7. These sequences then served as the hybridization regions of ssDNA probes which were designed to target mRNA by sequence complementarity. ssDNA probes also contained a 5' PCR handle for library generation, a Unique Molecular Identifier (UMI), and a 3' poly adenosine tail (A30) for retrofitting prokaryotic transcripts to the Chromium Single Cell 3' system (Figure 1a). Multiple probes (complementary to different regions) were designed for each gene to enhance transcript capture efficiency and decrease noise caused by poor hybridization and/or insufficient amplification of any given probe. Complete species libraries contained 29,765 probes for B subtilis and 21,527 probes for E. coli, and targeted 2,959 and 4,181 genes, respectively. Libraries were ordered at sub- femtomole quantities from Twist Biosciences (a one- time cost of 5¢ - 15¢ per probe) and amplified by rolling circle amplification8 to obtain a sufficient concentration (0.25 mg = 10.25 nM/library or approximately 0.35 pM of each probe) for scRNA- seq experiments (Figure1b, Methods, Supplementary Tables S1 and S2, Supplementary Figure S1). Probe libraries were completed by addition of randomized 12 bp UMI sequences and a poly- A tail and purified by PAGE. Completed libraries had a uniform coverage of probes (Figure 1e). The rolling circle amplification approach permits re- amplifying probe sets for unlimited subsequent experiments without the need to re- order probes, at an upfront cost of under 1¢ per cell (Supplementary Table S3). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Genome wide scRNA-seq probe design and synthesis
+ +a. A genome for the organism of interest is used to design a library of probes that are complementary to unique regions within each gene's coding sequence (CDS). Complementary ssDNA oligonucleotides are amended by addition of an extender sequence at the 5' end and 3' PCR handle. Sequences are ordered as an oligopool b. Probe sequences are inverted in-silico and restriction sites and handles are added for circle-to-circle amplification c. The large, inverse oligonucleotide library is commercially-synthesized and pooled. Ordered oligo-pools are amplified in a 3-step "rolling circle" reaction with Phi29 polymerase to produce the final, correctly oriented proto-probe library. d. A random 12mer sequence (UMI) and a 30-base polyA tail is added by primer extension with a blocked primer that is complementary to the 3' extender sequence. Single stranded DNA containing a universal PCR handle, UMI, complementary transcript region, and poly-A tail (listed 5' to 3') is purified. e. Finalized probe libraries are free of amplification primers and well distributed. + +<--- Page Split ---> + +Before microfluidic encapsulation, bacteria were fixed in \(1\%\) paraformaldehyde and permeabilized (Methods). Permeabilized bacteria were incubated with their corresponding DNA probe library. Non- hybridized probes were washed away (Figure 2a, Methods). Next, the bacteria were run through a 10X controller, where the DNA probes were captured and barcoded in a manner analogous to the barcoding of the transcriptome of mammalian cells (Figure 2 b- d). The resulting libraries were sequenced, preprocessed with custom scripts (https://gitlab.com/hormozlab/bacteria_scrnaSeq), and analyzed with the standard CellRanger pipeline and the Seurat analysis package \(^{9,10}\) (see Methods). + +## Microfluidic RNA-seq validation in bulk and single cell experiments + +To determine if our probe library can report on the transcriptional state of cells, we split a culture of \(B\) . subtilis cells in late exponential state into two aliquots of approximately \(10^{8}\) cells each. One aliquot was processed using a traditional RNA- seq protocol (Methods) while the second sample was fixed in formaldehyde for in situ hybridization with a probe set designed against the Bacillus subtilis genome. After incubation and washing, the probes that remained bound were amplified by PCR and processed into Illumina libraries. The output was compared to the traditional RNA- seq library. The hybridization and wash conditions at \(50^{\circ}\mathrm{C}\) gave results that are similar to traditional RNA- seq (Figure 2e; \(\mathrm{R}^2 = 0.55\) ). This value is comparable to the correlation observed when comparing RNA- seq to microarray experiments ( \(\mathrm{R}^2 = 0.57 - 0.65\) ) \(^{11,12}\) . + +Next, we tested if single cell transcriptomes can be captured by encapsulating individual bacteria using the 10X Chromium Controller after fixation and in situ probe hybridization. \(E\) . coli cells and \(B\) . subtilis cells were independently pre- treated with probes corresponding to their respective genomes. The prepared bacterial samples were subsequently mixed and loaded onto a microfluidic chip along with a custom PCR master mix containing a DNA primer designed to amplify the back- end PCR handle built into the probes (Figure 1, Methods). This custom mix replaced the standard cDNA reagents supplied by 10X Genomics as the aqueous phase within the droplets. Sequenced libraries from this experiment demonstrate that the microfluidic platform is able to successfully segregate and barcode individual microbial cells (Figure 2f). We observed a "doublet" rate of \(1.3\%\) for 3,373 captured cells, consistent with the expected rate of successful single cell encapsulation based on the specifications of the 10X microfluidic system (reported as an expected \(2.3\%\) multiplet events per 3,000 captured cells \(^{10}\) ). + +To assess whether the signal from individually encapsulated cells provides an unbiased readout of the transcriptomic state of the population, we compared the output produced by capturing probes from a bulk sample of cells \((\geq 10^{8}\) cells) to the signal obtained by summing the probe counts over thousands of individually tagged cells (aggregated UMI counts). The number of reads across all probes between the two samples were highly correlated ( \(\mathrm{R}^2 = 0.88\) , Figure 2g), confirming that the signal from single cells provides an unbiased representation of transcriptional states. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Microfluidic probe-based scRNA-seq method and validation.
+ +a. Cells are fixed and permeabilized to allow penetration of thousands of unique, genome-specific oligonucleotide probes. Hybridized probes "retrofit" transcripts with a poly-A tail & UMI while unhybridized probes are washed away. b. Cells are then flowed through a commercial microfluidic device which encapsulates single cells into droplets (shown in c.) containing barcoded primers conjugated to a hydrogel microsphere and PCR reagents. d. Barcoded cDNA is generated from the mRNA:probe hybridized complex via in-droplet PCR or RT reaction. Droplets are then broken, the pooled cDNA amplified further before sequencing. Single cell transcriptomes are resolved, clustered, and visualized. e. Transcriptomic estimation by probe hybridization in bulk samples is correlated to transcriptomic measurement of bulk samples by traditional RNA-seq f. Species mixture ("barnyard") plot demonstrates that single cells of different bacterial + +<--- Page Split ---> + +species can be resolved by barcode after microfluidic encapsulation. g. Aggregated probe- based signal from thousands of single cells correlates with the average probe- based signal of the bulk population. + +## Results + +## Bacterial scRNA-seq identifies known and previously hidden heterogeneity in B. subtilis + +We performed scRNA- seq on B. subtilis grown to late exponential phase in M9 minimal media supplemented with malate. In total, 2,784 cells were captured as determined by unique barcodes with \(550\pm 340\) (mean±SD, median=470) reads per cell. We detected a median of 325 mRNA transcripts per cell (approximately 10- 15% of the total mRNA pool13) corresponding to a median of 241 genes per cell. UMIs per probe ranged from 0- 45 for any given gene in any given cell. The most highly detected probes targeted genes encoding ribosomal proteins (rpsG, rpsH, rpsR, rpsM, rpsS, rplL, rplD, rplO, rplV), translation elongation machinery (fusA, tufA, map), and transcriptional machinery (rpoA). + +Next, we resolved distinct cellular states by dimensional reduction of single cell expression vectors followed by graph- based clustering and analysis of differential gene expression (DGE) using the Seurat package9. For B. subtilis in M9 minimal media in late log growth, we resolved four major transcriptomic signatures comprised of ten cell clusters that capture more subtle variations in gene expression (Figure 3 a- b, Supplementary Figure S2, Supplementary Table S6). Algorithmic grouping of cell expression profiles was robust to the clustering parameters (see Methods). As expected for minimal media cultures in late log, we observed a subpopulation of cells (clusters 6 and 8, approximately 10% of all cells) in the genetically competent state, the physiological state in which bacteria can uptake extracellular DNA and incorporate it into their genome. Cells in these two clusters differentially overexpress the competence master- regulator comK (logFC = 2.37, adjusted p- value = 4E- 179) as well as multiple genes within and downstream of competence operons (comC, comEA, comEC, comFA, comFB, comFC, comGA, comGC, comGD, comGE, comGG, recA, coiA, drpA, sbB, nucA, nin, clpC, recA). The fraction of competent cells in the population was measured by fluorescence microscopy using a fluorescent promoter- reporter of comG (Methods, Figure 3c left panel, Supplementary Figure S12a) and determined to be approximately 10.4% (IQR outlier detection - Methods) - similar to the fraction of the population comprising clusters 6 and 8 (9.3%, Supplementary Table S5). In total, we found that 45 of the 50 comK- regulon genes that were probed were differentially overexpressed in clusters 6 and 8 (adjusted p- values < 0.05; Figure 3d, Supplementary Table S6). Our results agree with numerous studies identifying the competence regulon in B. subtilis14,15, and the percentage of cells displaying natural competence falls within the range (3- 10%) of previous observations in similar media16,17. + +Unexpectedly for these growth conditions, we observed a small subpopulation of cells within the sample (cluster 9, 19 cells, 0.7% of the population) with a transcriptomic signature indicative of sporulation. In these cells, we observed significant upregulation of transcripts corresponding to sigma factors associated with sporulation, sigF (logFC = 1.2, adjusted p- value = 4E- 41) and sigG (logFC = 2.2, adjusted p- value = 2E- 63), as well as genes in the ger, cot, and spolVF operons within the cluster. Gene set enrichment analysis (GSEA) on all significantly upregulated genes in cluster 9 using ontological classes from the Gene Ontology Consortium18- 20 reveals a 4.3- fold + +<--- Page Split ---> + +enrichment of genes involved in sporulation (Fisher's exact test, \(\mathrm{FDR} = 1\mathrm{E} - 16\) ) as well as spore germination (fold enrichment \(= 8.1\) , \(\mathrm{FDR} = 1.7\mathrm{E} - 03\) ). This finding highlights the ability of probe- + +![](images/Figure_3.jpg) + +
Figure 3. scRNA-seq analysis reveals known and novel states in B subtilis.
+ +a. Heatmap of marker gene expression (z-score of log-transformed values) from 2,784 individual B. subtilis cells organized into 10 clusters. b. UMAP 2-dimensional representation of the 10 cell clusters reveals 4 + +<--- Page Split ---> + +highly distinct transcriptomic signatures c. Single cell expression of key marker genes for competence (comGE; clusters 6 and 8), sporulation (spollID, cluster 9) and arginine synthesis (argC, cluster 5) are highlighted on the UMAP (top panels, left to right). Volcano plots of genes expressed in at least \(25\%\) of cells in the corresponding clusters, with genes from the respective processes highlighted in green (middle panels). The lower panels show the presence of each heterogeneous marker in the population as confirmed by fluorescent promoter- reporter constructs \((P_{comG} - YFP, P_{cotY} - YFP, P_{argC} - YFP,\) respectively). A phase bright spore can be seen in the cotY- expressing cell. d. \(90\%\) of all comK regulon genes probed (45/50) were significantly differentially upregulated in cell clusters 6 and 8 (Bonferroni corrected P- value \(\leq 0.05\) for each gene) + +based scRNA- seq to detect rare and unexpected cell states in microbial cultures. To confirm this subpopulation, we created a promoter reporter for cotY, a spore coat gene expressed in the sporulating mother cell. Fluorescence observation identifies the presence of cotY expression and spores in a similar percentage of cells as recovered by scRNA- seq (0.5- 1%) (Figure 3c middle panel, Supplementary Figure S12d). + +The largest group of B. subtilis cells (comprising clusters 1, 2, 3, 5, and part of 7) accounted for \(\sim 45\%\) of the population and was characterized primarily by the upregulation of the dhb operon (logFCs \(>0.25\) , adjusted p- values \(< 2\mathrm{E - 05}\) ). Genes dhbA, dhbB, dhbC, dhbE, and dhbF encode the 5 enzymes implicated in the biosynthesis of bacillibactin - a catecholic siderophore which is produced and secreted in response to iron deprivation inside the cell. Relatedly, yusV, an ABC transporter of bacillibactin, was also found to be upregulated in these cells. Cells within cluster 5 are distinguished from the rest of the subpopulation by expression of genes related to arginine biosynthesis via ornithine. 10 of 14 genes associated with the cellular process were significantly upregulated in this cluster alone (gene set fold enrichment \(= 19.35\) , FDR \(= 3\mathrm{E - 08}\) ). A fluorescent promoter reporter for argC confirms heterogeneous expression of arginine genes with approximately \(2.1\%\) of cells in the high- argC expressing tail of the distribution - compared to \(7.2\%\) of cells in this state as determined by scRNA- seq (Figure 3c right panel, Supplementary Figure S12b, Methods, Supplementary Table S5). + +Finally, to see if microfluidic probe- based methods are robust with different DNA- labeling chemistries, we repeated scRNA- seq on B. subtilis, replacing in- droplet PCR with a probe barcoding step utilizing a reverse transcriptase enzyme in accordance with the 10X Chromium Single Cell 3' standard protocol for eukaryotic scRNA- seq. While resolution decreased (median of 89 transcripts and 86 unique genes per cell), we were still able to resolve distinct biological states including sporulating and competent cell clusters (Supplementary Figure S3, Supplementary Table S6). Taken together, our platform for single cell sequencing of individual bacterium correctly recapitulates the known cellular states of B. subtilis at the expected population fractions and identifies previously unknown cellular states. + +## Gene expression heterogeneity in clonal populations of E. coli + +With a validated method for identifying cellular states in Bacillus subtilis, we turned our attention to characterizing transcriptional heterogeneity in Escherichia coli. E. coli MG1655 was first grown in M9 minimal media, fixed at an \(\mathrm{OD}_{600} = 0.5\) (mid log), and subjected to probe- based scRNA- seq. We resolved 3,315 cells by unique single cell barcodes, detecting a median of 263 transcripts per cell representing a median of 165 distinct genes. Cells were partitioned into nine groups by graph + +<--- Page Split ---> + +based clustering of their gene expression profiles (Figure 4ac, Supplementary Figure S4, Supplementary Table S6). The lack of clearly distinguishable transcriptomic signatures after dimensionality reduction using UMAP suggests a cell population that exhibits more uniform gene expression profiles compared with Bacillus subtilis. However, as with B. subtilis, we observe that certain clusters can be assigned to specific biological processes by DGE and that cluster determination is robust to the choice of clustering parameters (Methods). Cells in cluster 8 (141 cells, \(4.3\%\) of the population) exhibit significant upregulation of fiml, fimC, and fimA ( \(\log \mathrm{FC} > 2.4\) , adjusted p- value \(< 1\mathrm{E - 166}\) ). The downstream genes of the fim operon encode for components of type 1 pili and are known to be expressed heterogeneously in cells upon inversion of the DNA element containing the fimA promoter ("phase on" fim switch) \(^{21,22}\) . Type 1 pili are associated with biofilm formation and pathogenicity due to the adhesive properties of the fimbriae \(^{23}\) . + +Relatedly, cells in cluster 1 (482 cells, \(14.5\%\) of population) uniquely upregulate genes implicated in cell motility including those encoding chemotaxis signaling proteins (cheA, cheW, tar; gene set fold enrichment \(>100\) , FDR \(= 3.1\mathrm{E - 02}\) ) and structural flagella components (fliC, flgL, flgG, flgD, flgE, flgC, flgF; gene set enrichment \(= 86.08\) , FDR \(= 2.6\mathrm{E - 06}\) ). This list includes genes regulated by class 1, 2, and 3 flagella promoters; three distinct regulons which control the sequential expression of flagella genes including the master regulator flhDC, components of the membrane- associated basal body and chemotaxis proteins, respectively \(^{44}\) . This model of assembly has been described as "just in time" gene expression that allows E. coli to discretize the energetically demanding biosynthesis process into checkpoint- controlled subprocesses. The fact that we observe upregulation of genes from multiple classes within the cluster, often within the same individual cells, indicates a more overlapping transcriptomic state, an observation that aligns with the fact that class 2 controlled fliA controls transcription of class 3 promoters upon completion of the basal body and export of class 2 flgM (fliA inhibitor) \(^{44}\) . To confirm this motile subpopulation of cells, we stained cells grown under the same conditions (Remel flagella stain) and visualized by phase microscopy. As shown in Figure 4d, we see assembled flagella emanating from a fraction of cells within the population. + +Cells within clusters 5, 6, and 7 demonstrate differential usage of carbamoyl phosphate within the culture. Clusters 5 and 7 upregulate carAB, which encode the subunits of carbamoyl phosphate synthetase, converting bicarbonate to carbamoyl phosphate as the first step in the de novo biosynthesis of both uridine- 5'- monophosphate (UMP) and arginine. Both clusters, arranged adjacently on the UMAP, also upregulate pyrI, pyrB & pyrC, three enzymes involved in the first two steps of UMP biosynthesis, implying a commitment of the cells towards pyrimidine biosynthesis. Alternatively, cells in cluster 6 shuttle carbamoyl phosphate into the arginine biosynthesis pathway as implied by the upregulation of argF, argI, argB, argG, argD ( \(\log \mathrm{FCS} > 0.55\) , adjusted p- values \(< 1\mathrm{E - 20}\) ). GSEA reinforces this observation, finding a significant enrichment of genes implicated in arginine biosynthesis via ornithine in these cells (gene set fold enrichment \(= 26.77\) , FDR \(= 2.02\mathrm{E - 03}\) ). + +Next, E. coli was grown in rich, LB media under anaerobic conditions. In the absence of exogenous electron acceptors permitting cellular respiration (e.g. O2, nitrite, nitrate, DMSO), E. coli performs "mixed acid" fermentation to sustain ATP generation by glycolysis, shuttling electrons into a variety of end- products (acetate, ethanol, succinate, formate) in order to replenish the pool of NAD+. The amount of each end- product produced by E. coli varies based on the redox state of the cell - an adaptability which allows E. coli to grow on a wide range of substrates under + +<--- Page Split ---> + +anoxic conditions and varying \(\mathsf{pH}^{24}\) . We performed scRNA- seq to elucidate whether mixed acid fermentation can be, in part, attributed to the separation of branched fermentative pathways into distinct subpopulations of cells. Graph- based clustering of gene expression vectors ( \(n = 3,763\) resolved cells) resulted in 8 cell clusters (Figure 4b,d, Supplementary Figure S5, Supplementary Table S6). Within the population, we observe the most highly abundant transcripts (top \(5\%\) by aggregated UMIs) corresponding to genes involved in anaerobic ATP generation including glycolysis (fbaB, mdh, aceE, fbaA, pgi, lpd, eno, gapA, pgk, gene set fold enrichment \(= 9.5\) compared with a null model where the top \(5\%\) most abundant genes were chosen randomly, FDR \(= 2.0\mathrm{E} - 4\) ), anaerobic respiration (dcuA, acnA, hyaC, glpB, frdC, mdh, frdB, hyaA, hyaB, narH, glpD, glpA, narG, hybO, glpC, fold enrichment \(= 4.9\) , FDR 1.22- 04), and mixed acid fermentation (hyaC, frdC, mdh, frdB, hyaA, hyaB, gldA, fold enrichment \(= 8.3\) , FDR \(= 3.7\mathrm{E} - 03\) ) in addition to genes related to ribosome biogenesis, transcription, and translation. Enrichment of these pathways within the most highly expressed genes at the population level is consistent with an anaerobic environment. + +We then analyzed the transcriptional states of cells in each cluster. Clusters 1 and 4 appear to represent a subpopulation of cells responding to over- acidification of the cytoplasm. Both groups significantly upregulate the gad operon (gadABC) as well as dps (logFCs \(> 0.34\) , adjusted p- values \(< 1\mathrm{E} - 32\) ). In addition, acid stress chaperone proteins hdeA and hdeB are overexpressed in cluster 4 (logFCs \(> 0.25\) , adjusted p- value \(< 1\mathrm{E} - 02\) ). Acid stress is an expected consequence of fermentation as acidic fermentation byproducts (e.g. acetate, lactate, formate) accumulate. Cells in cluster 5 (283 cells, \(7.5\%\) of total) had 157 genes significantly upregulated (adjusted p- values \(< 0.05\) ) compared with the rest of the population. Surprisingly, these genes belong to pathways involved in aerobic respiration (aceAB, sucABCD, nuoBMN, cyoABC, cydAB, sdhABC, fumA, gltA, acnB, icd, mdh; gene set fold enrichment \(= 11\) , FDR \(= 2.32\mathrm{E} - 13\) ), ATP synthesis coupled proton transport (atpABCDEHG; gene set fold enrichment \(= 24.47\) , FDR \(= 1.45\mathrm{E} - 05\) ), glycolysis (mdh, gpmA, aceE, fbaA, lpd, aceF, gapA; gene set fold enrichment \(= 10.24\) , FDR \(= 7.33\mathrm{E} - 04\) ), and gluconeogenesis (gpmA, fbaA, sdaB, pck, pps; gene set fold enrichment \(= 9.92\) , FDR \(= 1.03\mathrm{E} - 02\) ). The overexpression of these genes, many of which are involved in the TCA cycle (Figure 4g), may be an indication of oxygen contamination within the culture, resulting in a stark difference in transcriptional signatures and, consequently, cell behavior within the isogenic population. In addition, cells in cluster 5 also appear to be motile as genes pertaining to flagella structure (flgCDEFG; fold enrichment \(= 9.26\) , FDR \(= 1.28\mathrm{E} - 02\) ) and locomotion (fliCD, tar, mgIB, cheAW, flgK; fold enrichment \(= 5.16\) , FDR \(= 4.08\mathrm{E} - 04\) ) were upregulated & overrepresented in the subpopulation. Heterogeneous expression of flagella was confirmed microscopically using a remel stain (Figure 4f). Since cultures were maintained in capped tubes outside of an anaerobic chamber, is it possible air was introduced into the culture through an imperfect seal. Thus, we hypothesize that this induced cell population activated motility genes to gather at the liquid/air interface. Consistent with the idea of such an oxygen- gradient organization, a recent report used a high- throughput fluorescent in situ hybridization (FISH) analysis of approximately 100 genes to demonstrate the role of oxygen gradients in spatial partitioning within biofilms25. + +<--- Page Split ---> +![](images/Figure_1.jpg) + + +<--- Page Split ---> + +a. Heatmap of marker gene expression (z-score of log-transformed values) from 3,315 individual E. coli cells grown aerobically in minimal M9 media organized into 10 clusters. b. Heatmap of gene expression from 3,763 individual E. coli cells grown in anaerobic LB media organized into 9 clusters. c. UMAP 2-dimensional representation of the 10 cell clusters from aerobic M9 culture conditions. d. Flagellar components are heterogeneously expressed in aerobic M9 media - a key flagellar gene (flagellin - fliC) is preferentially expressed by cells in cluster 1. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. e. UMAP 2-dimensional representation of the 9 cell clusters from E coli fermenting LB media culture conditions. f. Flagellar components are heterogeneously expressed in fermentation conditions - a key flagellar gene (flagellar basal-body rod protein - flgC) is preferentially expressed by cells in cluster 5. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. g. Many aerobic respiration genes and other metabolic and flagellar genes are overexpressed in cluster 5. Volcano plots of genes over-expressed in at least 25% of cells in cluster 5 from these processes are highlighted in colors as labeled. Green labeled genes, involved in central-carbon metabolism, are also highlighted as green arrows in the metabolic map. + +Finally, cluster 8 (111 cells, \(3\%\) of total), which projects between the majority of cells and cluster 5 on the UMAP, is characterized by the strong upregulation of the tdc operon (tdcBCEG; logFC \(>3.0\) , adjusted p- values \(< 1\mathrm{E - 100}\) ). Under anaerobic conditions, if energy is low inside the cell, E. coli generates ATP by conversion of L- threonine to propionate. The tdc operon has been shown to be strongly induced by anaerobic conditions as well as catabolically repressed (CRP- controlled), suggesting these cells are sensing an anoxic environment while also deprived of glucose26. The tdc operon encodes proteins for the transport and degradation of L- threonine (and L- serine by analogous intermediates) for the purpose of ATP generation (1:1 conversion). While the operon is induced in the absence of oxygen, it has been shown to not be directly regulated by common oxygen- sensing systems inside the cell such as transcriptional regulators FNR and ArcA but rather by tcdA and tcdR (transcriptional activators)27. To the best of our knowledge, this is the first time threonine metabolism has been shown to be cellularly segregated in a microbial culture. Taken together, these findings demonstrate how single cell transcriptomics can inform on physiological and metabolic organization in complex microbial environments. + +## Discussion + +To agnostically characterize the collection of cell states within a microbial population, we developed an affordable (standard commercial 10X run costs plus \(\equiv 1c\) per cell in probes) bacterial scRNA- seq technique that demonstrates high throughput and sensitivity. Our method, which is based on combining microfluidic droplet partitioning of single cells with in situ hybridization of DNA probes, overcomes several difficulties associated with obtaining quality single cell mRNA signal from bacteria. Using bacterial scRNA- seq, we identify known cellular states including genetic competence and sporulation in B. subtilis and fimbriae production in E coli. In addition to these well characterized physiologies, we uncovered several previously unknown or sparsely studied heterogeneously expressed gene sets including metabolic pathways (amino- acid metabolism, carbon metabolism, siderophores) and physiological states (chemotaxis and motility). Interestingly, arginine biosynthesis genes appear to be heterogeneously expressed in both E coli and B subtilis in minimal media. Arginine synthesis is also isolated spatially (at the organ and cellular location level) in multicellular, eukaryotic organisms28. This conserved partitioning of a biochemical process suggests that some processes may be inherently more + +<--- Page Split ---> + +prone to segregate. We speculate that such segregation may be a result of incompatibilities between certain biochemical reactions and may have played a role in the initial development of multicellular structure. + +Along with the method we describe in this work, two other groups have separately introduced a high- throughput bacterial scRNA- seq approach that is based on combinatorial indexing29,30. A commonality between these studies and ours is the discovery of subpopulations of cells expressing genes for specialized function, including genetic competence. While the different techniques uncover common biology, each has advantages and disadvantages. Our approach requires an upfront investment in probe generation and prior understanding of the genome, and is therefore not intended to be used with novel or environmental organisms that are not well characterized. Conversely, once a probe set is synthesized, our tool is fast, cheap, and provides very deep coverage of transcripts as well as a more accurate method of quantification by UMI. In addition, the technique relies on commercial equipment that is now ubiquitous in laboratories and core facilities around the world. Critically, unlike the combinatorial indexing system, our in- situ hybridization technique ensures that sequencing is not wasted on ribosomal or other non- mRNA transcripts, which account for between 93- 97% of all transcripts that are sequenced by combinatorial indexing (ref). + +Our technique of combining in- situ hybridization with microfluidic single cell partitioning also invites several potential uses for eukaryotic samples which often rely on targeting the 3' poly- A tail; an approach that provides only a single site for transcript capture, excludes sequences lacking a poly- A, and provide limited information on splice variants. In addition, hybridization techniques are compatible with formalin- fixed paraffin- embedded tissues, the most common form of clinical sample, which are difficult to analyze using existing high- throughput single cell techniques. Finally, the extension of these techniques to mammalian and other eukaryotic cells could potentially be used to simultaneously identify the gene expression profile of an intracellular bacteria and its eukaryotic host cell. + +In the context of microbial analyses these new tools will enable investigation of several recently observed phenomena in bacterial cultures. In particular, we foresee their use in characterizing bacterial subpopulations that are differentially resistant to antibiotics (antibiotic persistence), and the physiological role of bacterial cells that differentially express toxin and virulence genes in bacterial infections. In eukaryotic systems, where mature scRNA- seq tools have existed for several years, computational methods extend beyond surveys of the different cell types to provide information about the dynamics of cellular populations such as pseudotime arrangements of cells31, lineage and causality maps and cell- type specific QTLs32. In bacterial communities we predict that these bioinformatic approaches will be used to understand dynamic processes like phage- bacteria interactions, the organization of biofilms, and lab- based evolution of traits. + +## Acknowledgements + +DS was funded in part by the Natural Sciences and Engineering Research Council of Canada (NSERC PGSD2- 517131- 2018). DS and SH acknowledge funding from NIH NIGMS R00GM118910, NIH NHLBI R01HL158269, U19 Systems Immunology Pilot Project Grant at Harvard University, and the Harvard University William F. Milton Fund. AZR and RAM were funded by a Dupont Science and Innovation award. Portions of this research were conducted on + +<--- Page Split ---> + +the O2 High Performance Compute Cluster, supported by the Research Computing Group, at Harvard Medical School. See http://rc.hms.harvard.edu for more information. + +## Methods + +## Strains and Growth Conditions + +Bacillus subtilis str. 168 was used for all single cell Bacillus experiments. For experiments on suspension cultures in glucose/malate, cells were grown at \(37^{\circ}C\) with vigorous shaking in M9 media supplemented with CaCl2 (0.1 mM), \(0.2\%\) glucose, tryptophan, and trace metal mix 8. Trace metal solution was made as a 100X concentrate using Na- EDTA 5.2 g, \(\mathrm{FeSO_4 - 7H_2O}\) 2100.0 mg, \(\mathrm{H_3BO_3}\) 330.0 mg, \(\mathrm{MnCl_2 - 4H_2O}\) 100.0 mg, \(\mathrm{CoCl_2 - 6H_2O}\) 190.0 mg, \(\mathrm{NiCl_2 - 6H_2O}\) 24.0 mg, \(\mathrm{CuCl_2 - 2H_2O}\) 2.0 mg, \(\mathrm{ZnSO_4 - 7H_2O}\) 144.0 mg, \(\mathrm{Na_2MoO_4 - 2H_2O}\) 1200.0 mg, DI Water, 999.0 ml. The solution pH was adjusted to 7.0. At OD of 0.4- 0.5 cultures were supplemented with \(50~\mathrm{mM}\) malate. + +Bacillus subtilis strain PY79 was used for all fluorescent promoter- reporter strains. comG reporter strain was previously described 8. Promoter reporters for cotY and argC were made as previously described 8. Briefly, YFP promoter reporters were cloned into the ECE174 backbone plasmid which uses sacA integration site and encodes chloramphenicol resistance (R. Middleton, obtained from the Bacillus Genetic Stock Center). Strains were made by genomic integration into the genome. Fluorescent reporters were integrated into the sacA site and checked by sequencing. Reporter strain detail is found in Supplementary Table 4. Reporter strains were grown in the same media and growth conditions used in single cell experiments. + +E. coli MG1655 was used in all E coli single cell experiments. For experiments in minimal media cells were grown overnight in M9 minimal media and incubated at \(37^{\circ}C\) with moderate shaking (200 rpm). Overnight cultures were diluted and subcultured and a fixed specimen was collected at mid exponential phase (OD 0.3-0.5). For experiments in anaerobic LB media cells were inoculated into degassed LB that was aliquoted inside a COY anaerobic chamber containing (95% N2 5% H2). Overnight culture was subcultured into a degassed media tube using a needle and syringe purged in N2 gas, and the samples were incubated anaerobically inside an anaerobic box with a gas-pack (manufacturer) at \(37^{\circ}C\) . Cells were fixed at OD 0.3-0.5 by addition of formaldehyde using a N2-purged syringe and needle directly into the tubes. + +## Probe set design + +Probes were designed with an mRNA complementary region approximately 50 bp in length flanked by a PCR handle (18- 23 bp) towards the 5' end and a 30 poly(dA) tract on the 3' end. On the ends of each probes we included a region allowing for circularization and cutting with hindIII as outlined by Schmidt et al 28. Probe sequences are included in Supplementary Tables 1 and 2. + +## Probe amplification + +Probes were amplified using a rolling circle method similar to the one used by Schmidt et al 28 with slight modifications. To amplify our probes, we did not use nicking enzymes but instead only + +<--- Page Split ---> + +used the HindIII digestion site in the rolling circle scheme \(^{28}\) . Our incubations were scaled by a factor of 5 from the detailed protocol provided by Schmidt et al. Otherwise, all other aspects of the protocol were as previously detailed. Amplified probes were eluted and purified by PAGE electrophoresis using \(15\%\) urea gels and ethanol precipitation as described in Sambrook et al \(^{59}\) . + +## UMI and Poly-A Addition + +Amplified "proto- probes" were extended to include a unique molecular index (UMI) and 3' poly- adenine tail by isothermal extension with a 3' blocked primer containing the reverse complementary sequences. 100 ul of probes and 1 mg blocked extension primer (Supplementary Table 3) were mixed with 100 units of klenow fragment and 1x NEB buffer 2 and incubated for 30 min at room temperature. The extended oligonucleotide library was selectively purified by PAGE electrophoresis (final probe length approximately 142 bp) using \(15\%\) urea gels followed by ethanol precipitation as described in Sambrook et al. \(^{59}\) + +## Fixation and in situ hybridization reactions + +Cells from one intact colony or 2 mL of cell culture were fixed using a 30 minute incubation in \(1\%\) paraformaldehyde (final concentration) at room temperature. Formaldehyde- fixed samples were washed with \(0.02\%\) SSC by gentle centrifugation ( \(6,000 \times \mathrm{G}\) ) for 2.5 minutes. After the wash, cell pellets were resuspended in 1 mL MAAM (4:1 V:V dilution of methanol to acetic acid). Samples were kept at \(- 20^{\circ}\mathrm{C}\) for up to 2 days before further processing. For in- situ probe hybridization, 150ul of fixed sample was centrifuged and washed once in 1x PBS to remove methanol and acetic acid. After the wash step cells were incubated in 200 ul PBS with 350u/ul of lysozyme solution (Epicentre ready- lyse, Epicentre Madison WI USA) for 30 minutes at room temperature. After 30 minutes cells were pelleted and washed with 500 ul PBS- tween (1 x PBS with 0.1 % Tween 20). Cells were then re- suspended with 100 ul of probe binding buffer consisting of 5x SSC, \(30\%\) formamide, 9mM citric acid (pH 6.0). \(0.1\%\) Tween 20, 50 ug.ml heparin, and \(10\%\) low MW dextran sulfate) \(^{60}\) . Cell suspensions were placed in a \(50^{\circ}\mathrm{C}\) shaker- incubator and allowed to pre- equilibrate for 1 hour. After an hour 50 ul of probes, (600 ng/ul) were added to each cell suspension, and samples were left to incubate overnight. + +The next day samples were washed 5 times in pre- warmed ( \(50^{\circ}\mathrm{C}\) ) probe- wash solution (5 x SSC, \(30\%\) formamide, 9 mM citric acid (pH 6.0), \(0.1\%\) Tween 20, and 50 ug/mL heparin). + +Before flowing on the 10X device cells were washed 3 times in 1x PBS and diluted as suggested in the 10X Chromium instruction manual. + +## Microfluidic encapsulation and droplet generation reaction + +Single cell partitioning, barcoding, and cDNA library generation was achieved using 10X Genomics's Chromium Controller with the Chromium Single Cell 3' Reagents Kit (v2 chemistry) as described by 10X Genomics (https://support.10xgenomics.com/single_cell- gene- expression/index/doc/user- guide- chromium- single_cell- 3- reagent- kits- user- guide- v2- chemistry). The protocol was modified to achieve bacterial scRNA- seq. During GEM generation, a master mix containing the following reagents (per rxn, not accounting for excess volume) was prepared: 33 uL of 4X ddPCR Multiplex Supermix (BioRad), 4 uL of custom primer (10 uM), 2.4 uL additive A (10X Genomics), 26.8 uL dH2O. All other reagents specified by 10X Genomics were omitted. + +<--- Page Split ---> + +Prepared cell samples were washed 3X in PBS and diluted to 1000 cells/uL before loading on the microfluidic chip ("Chip A Single Cell") with a targeted cell recovery of 10,000 cells. + +Library construction & sequencing + +After running the Chromium Controller, each sample (i.e. "reaction") was visually inspected to confirm successful GEM formation (should observe a well distributed emulsion with a volume of approximately 100 uL). Samples were then transferred to fresh PCR tubes and cycled at the following conditions (replacing the "GEM- RT Incubation Step" in 10X Genomics protocol): \(94^{\circ}C\) for 5 minutes, 6 cycles of (94°C for 30s followed by 50°C for 30s then 65°C for 30s), hold at 4°C. + +After PCR #1, the emulsion was broken and the pooled DNA purified using Dynabeads MyOne Silane as described in 10X Genomics' protocol. Purified DNA was amplified once more (replacing the "cDNA Amplification step" of 10X Genomics' protocol) using a master mix composed of 17 uL of purified DNA, 20 uL of Q5 Hot Start 2X MM, 1.5 uL of forward primer (10 uM) and 1.5 uL of reverse primer (10 uM). PCR conditions included a 30s incubation at 98C followed by 16 cycles of 10s at 98C, 20s at 62C, and 20s at 72C before a final extension at 72C for 2 min. After PCR #2, amplified DNA was purified using the NucleoSpin® Gel and PCR Clean- up kit (Macherey- Nagel) as per manufacturer's instructions. Purified DNA was run on Agilent TapeStation to confirm the presence of a band at 180 bp, indicating successful library generation. Libraries then prepared for sequencing by Illumina adapter addition via low cycle (n = 6) PCR with custom library preparation finishing primers (see Supplementary Table 3). + +Sequencing was done on an illumina nextSeq- 1000 instrument with 100 cycle reagents. Libraries were spiked with 30% phi- X and sequenced for 8bp in the i7 index direction and 119 bp for "Read 1". Addition of 30% phiX improved fastQ quality scores. + +Microscopy + +Cultures were visualized on a Leica DM3000 light microscope before single cell experiments to ensure samples were free of chains and clumps. Reporter strains were imaged using a 100X oil- immersion objective on a Zeiss Axio Observer inverted microscope equipped with a colibri- 7 LED fluorescent light source and an axiocam digital camera. Flagellar staining was done using the Remel flagellar stain (Thermo- Fisher Scientific) as per manufacturer instructions. + +Image analysis using cell segmentation and outlier identification + +Cells in microscopy images were segmented using the microbeJ program within FIJI (ImageJ). Outliers were identified by using the IQR outlier method. Specifically, quartiles and the interquartile range (IQR) were determined for each dataset. Outliers were determined by standard outlier detection parameters: cells that were more than 1.5 IQRs above Quartile 3 were designated as cells within an overexpressing population. Population histograms were plotted and outliers reported. + +Constructing a Reference Genome + +<--- Page Split ---> + +Here we describe how we constructed reference genomes tailored to the probesets used for each organism (see Methods Section Probeset Design). For a given organism, let \(P\) be the number of probes in the probeset, \(\tilde{s}_{dig} = AAGCTT\) be the \(L_{dig} = 6\) base pair (bp) nucleotide sequence of the HindIII digestion site (see Supplementary Figure S1a), \(\tilde{s}_{ext} = TGCTCGAAGAAACTATG\) be the \(L_{ext} = 17\) bp nucleotide sequence of the extender, \(\tilde{s}_{PCR} = TGTTAGGACGGGTCATC\) be the \(L_{PCR} = 18\) bp nucleotide sequence of the PCR handle, and \(\tilde{p}_{j}\) be the \(L_{j}\) bp nucleotide sequence of the \(j^{\text{th}}\) probe. Let \(\tilde{g}_{j} = [\tilde{s}_{dig}, \tilde{s}_{ext}, \tilde{p}_{j}, \tilde{s}_{PCR}]\) be the \(L_{dig} + L_{ext} + L_{j} + L_{PCR}\) bp nucleotide sequence formed by flanking the \(j^{\text{th}}\) probe by the HindIII digestion site, extender sequence, and PCR handle as shown in Figure 1a. We first constructed a FASTA file by concatenating all \(P\) of the \(\tilde{g}_{j}\) sequences. Next, we constructed a GTF file to index the FASTA. Typically, the GTF indexes genes in the FASTA by recording the location of the starting and ending bps of each gene. Since we have multiple probes per gene, and our FASTA is a concatenation of these probes, we therefore use the GTF to index probes instead. Specifically, we recorded that for the \(j^{\text{th}}\) probe, the starting and ending bps of \(\tilde{g}_{j}\) in the FASTA are \((\sum_{i = 1}^{j - 1} L_{dig} + L_{ext} + L_{i} + L_{PCR}) + 1\) and \((\sum_{i = 1}^{j} L_{dig} + L_{ext} + L_{i} + L_{PCR})\) respectively. We use \(\tilde{g}_{j}\) as the definition of a probe instead of \(\tilde{p}_{j}\) for the purpose of alignment because sequenced reads should contain the flanking sequences anyways so including them minimizes spurious alignment of off- target reads. Lastly, we constructed a reference genome by using the mkref command in CellRanger (v3.1.0), supplying the generated FASTA and GTF files to the --fasta and --genes arguments respectively. + +## Creating FASTQ Files for CellRanger + +Each read in the Illumina- sequenced FASTQ file contains the 10x CB, the 10x UMI, the Probe UMI as well as the mRNA sequence on a single line. This file therefore needs to be re- formatted to be compatible with CellRanger, so that the CB and UMI is stored in the R1 file and the mRNA sequence is stored in the R2 file. We also perform filtering at this stage so that only reads containing all the information needed by the CellRanger pipeline are retained. First, we discard reads whose length is less than or equal to 110 bp. Next, we aligned each read to \(\tilde{s}_{ext}\) using the nwalign(global=true) function in MATLAB (2019a), and discarded reads where the alignment score was \(< 17\) . We search for the extender sequence specifically, because for all probe designs the Extender sequence flanks the Probe UMI sequence (see Supplementary Figure S1c). + +We create two sets of FASTQ files using reads retained after these filtering steps. In the first set of FASTQ files, we record the 10x CB and Probe UMI in R1, with the Probe UMI obtained by extracting 12 bps in either the 5' or 3' direction from the Extender sequence according to the probe design (see Supplementary Figure S1c). In the second set of FASTQ files, we record the 10x CB and 10x UMI in R1. For both sets of FASTQs, we trimmed all nucleotides 5' (3') of the Extender sequence if the Extender is 5' (3') of the binding region, and record the result in R2. The two sets of FASTQs therefore have the same number of lines, identical R2 files and the same 10x CBs in the R1 files, and only differ in the UMI recorded in the R1 files. We obtained single cell count matrices from these reformatted FASTQs by using the count function in CellRanger (v3.1.0) and supplying the custom made genome (described in Methods Section Constructing a Reference Genome) as the --transcriptome argument. Since CellRanger discards reads if QC scores for UMIs are too low, the same set of reads may not necessarily end up being used in constituting single- cell count matrices from the two sets of FASTQ files. + +## Comparing Single cell Probe Expression Matrices Generated Using 10x vs Probe UMIs + +The output from CellRanger is a single cell matrix where each row, \(i\) , corresponds to one of \(N\) cells, and each column \(j\) corresponds to one of \(P\) probes. When the supplied FASTQ files have Probe (10x) UMIs recorded in R1, each entry in the single cell matrix, \(c_{i,j}^{P}\) ( \(c_{i,j}^{P}\) ) is the number of Probe (10x) UMIs in cell \(i\) corresponding to probe \(j\) . The Probe UMI remains fixed over multiple + +<--- Page Split ---> + +rounds of PCR and can therefore be treated as a classical UMI, while in our protocol, new 10x UMIs may be introduced on the same transcript at each PCR round. Therefore, \(c_{i,j}^{p'}\) may not accurately record the number of transcripts. We used \(c_{i,j}^{p'}\) for all results presented in the main text of the paper. Since it may be more convenient to use 10x UMIs directly, we also quantified the extent to which \(c_{i,j}^{p'}\) differs from \(c_{i,j}^{p'}\) by computing a joint distribution of count frequencies, \(h_{k,l} = \sum_{i}\sum_{j}I_{k}\left(c_{i,j}^{p'}\right)I_{l}\left(c_{i,j}^{p'}\right)\) , where \(I_{s}(t) = \delta_{s,t}\) (see histograms in Supplementary Figures S6- S8). + +On the one hand, when \(h_{k,l} > 0\) , for \(l > k\) , there is over- counting; the sequencing depth is large enough to detect spurious 10x UMIs introduced in latter PCR rounds. On the other hand, when \(h_{k,l} > 0\) , for \(l < k\) , there is under- counting; CellRanger detects that the same 10x UMI is coincidentally found on different probes, and discards all such reads. Since under- counting doesn't introduce spurious transcript counts and can be treated in the same vein as low sequencing depth or low capture efficiency, we do not consider it a confounder. The accuracy of a protocol relying only on 10x UMIs can therefore be quantified by the fraction of 10x UMIs corresponding to entries in \(h_{k,l}\) with \(l \leq k\) : + +\[Accuracy = \frac{\sum_{k}\sum_{l\leq k}l h_{k,l}}{\sum_{k}\sum_{l}l h_{k,l}}\] + +For the B. Subtilis sample, the accuracy of using 10x UMIs thus computed was only \(41\%\) . However, when we computed gene expression matrices using \(c_{i,j}^{p'}\) (see Methods Section Single cell Gene Expression Matrices from Single cell Probe Expression Matrices below) and performed differential gene expression (see Methods Section Transcriptomic Analysis and Visualization below), we were still able to detect clusters corresponding to competence and sporulation (see e.g. clusters #2 and #11 respectively in Supplementary Figure S6 and Supplementary Table 7). Likewise, for the E. coli sample grown in minimal media, the accuracy was only \(33\%\) , but we were still able to detect clusters corresponding to the fim operon, cell motility and differential usage of carbamoyl phosphate (see e.g. clusters #8, #2 and #7 respectively in Supplementary Figure S7 and Supplementary Table 7). Lastly, for the E. coli sample grown in LB media, the accuracy was \(88\%\) . Here too, we were able to recapitulate biological results obtained using Probe UMIs including clusters with different metabolic programming and upregulation of the tdc operon (see clusters #6 and #9 respectively in Supplementary Figure S8 and Supplementary Table 7). Taken together, this suggests that users of our technology may be able to use 10x UMIs directly i.e. \(c_{i,j}^{p'}\) . However, for the remainder of this study we restrict our attention to \(c_{i,j}^{p'}\) . + +## Single cell Gene Expression Matrices from Single cell Probe Expression Matrices + +In order to perform standard single cell analysis such as cell- similarity clustering and differential gene expression, we need to determine \(c_{i,k}^{g}\) , the number of transcripts of gene \(k\) in cell \(i\) , whereas the output from CellRanger is \(c_{i,j}^{p'}\) . Let \(G_{k}\) be the indices of probes in the probeset corresponding to gene \(k\) . Since multiple probes may bind to the same transcript, setting \(c_{i,k}^{g} = \sum_{j \in G_{k}} c_{i,j}^{p'}\) may result in an over- estimate of transcript counts. To be conservative while also retaining as many UMIs as possible, we therefore picked \(c_{i,k}^{g} = \max_{j \in G_{k}} c_{i,j}^{p'}\) which lower bounds gene expression. This choice resulted in \(66\%\) , \(79\%\) and \(81\%\) of the UMIs in \(c^{p}\) being retained in \(c^{g}\) , accounting for \(65\%\) , \(77\%\) and \(81\%\) of the total reads, for the B. Subtilis, E. Coli in minimal media and E. Coli in LB media samples respectively. + +That different probes may be selected to represent the same gene in different cells is not material, since probes are selected proportional to their binding affinity. However, this choice introduces a + +<--- Page Split ---> + +potential bias since the gene expression tabulated in \(c^g\) will tend to be higher for genes with more corresponding probes in the probeset, i.e. larger \(|G_k|\) . As a sanity check, we also tabulated single cell gene expression matrices using \(c_{i,k}^g = c_{i,k}^p, k^* = \underset {j\in G_k}{\operatorname{argmedian}}\sum_i c_{i,j}^p\) i.e. using a fixed probe across all cells for each gene corresponding to the probe with the median bulk expression. This choice only resulted in \(10\%\) , \(18\%\) and \(17\%\) of the UMLs in \(c^p\) being retained in \(c^g\) , for the \(B\) . Subtilis, \(E\) . Coli in minimal media and \(E\) . Coli in LB media samples respectively. Despite the decreased transcriptional resolution, we are able to still detect clusters corresponding to the results presented in the main text (see Supplementary Figures S9- S11 and Supplementary Table 8). We therefore defer investigation of more sophisticated methods of combining probe counts to form a single cell gene expression matrix to subsequent work. + +## Transcriptomic Analysis and Visualization + +Here we describe how we analyzed single cell gene expression matrices, \(c^g\) , using Seurat (v3.1.2 with default parameters except where indicated). We first log- transformed the data using the NormalizeData(normalization.method = "LogNormalize", scale_factor = 10000) function, and selected the 2000 most variable genes using FindVariableFeatures(selection.method = "vst", nfeatures = 2000). Then, we z- scored these highly- variable genes using ScaleData. Next, we performed linear dimensionality reduction using PCA down to 50 dimensions (RunPCA). Points in this embedding were used to construct UMAP plots (RunUMAP(dims=1:10)) and find neighbors for clustering (FindNeighbors). FindClusters(resolution=1.0) was used to run the Louvain clustering algorithm and generate clusters. We confirmed that the clusters highlighted in the main text appeared consistently for a range of resolutions from 0.5 to 1.5. Summary statistics of the number of genes/transcripts in a cell and cluster populations for the different samples can be found in Supplementary Table 5. Differential gene expression was performed using the FindMarkers(min.pct=0.25) function with a log- fold change cutoff of 0.25, whose output when applied to our data is shown in Supplementary Tables 6- 8. Heatmaps of differentially expressed genes were generated using the DoHeatMap function, and show the top over- expressed genes in each cluster with Bonferroni- corrected p- value \(< 0.05\) . + +## Data and Code Availability + +All data used in this paper will be made available on the NCBI GEO upon publication. Code to prepare custom references, reformat FASTQ files for the CellRanger pipeline and generate single- cell gene expression matrices from single- cell probe expression matrices is available at https://gitlab.com/hormozlab/bacteria_scrnaseq. All code and instructions to reproduce numerics and figures in the manuscript will be made available upon publication. + +<--- Page Split ---> + +## Supplementary Figures + +Figure S1: Detailed schematic of C2C- based proto- probe amplification and primer extension incorporation of probe- based UMI and polyA + +Figure S2: Heatmap of B subtilis cells from Figure 3 including all marker gene names + +Figure S3: scRNA- seq of B subtilis using in- droplet reverse transcription (RT) instead of in- droplet PCR identifies sporulation and competence populations + +Figure S4: Heatmap of aerobic M9 culture of E coli cells from Figure 4 including all marker gene names + +Figure S5: Heatmap of E coli fermenting in LB media from Figure 4 including all marker gene names + +Figure S6: scRNA- seq of B subtilis using 10x UMI instead of Probe UMI + +Figure S7: scRNA- seq of E coli cells in aerobic M9 culture using 10x UMI instead of Probe UMI + +Figure S8: scRNA- seq of E coli cells in LB media using 10x UMI instead of Probe UMI + +Figure S9: scRNA- seq of B subtilis using bulk median instead of per- cell maximum probe counts for gene expression + +Figure S10: scRNA- seq of E coli cells in aerobic M9 culture using bulk median instead of per- cell maximum probe counts for gene expression + +Figure S11: scRNA- seq of E coli cells in LB media using bulk median instead of per- cell maximum probe counts for gene expression + +Figure S12: Frequency of marker genes and spores in the population measured by microscopy + +## Supplementary Tables + +Table S1: B subtilis probes ordered from TWIST Bioscience + +Table S2: E coli probes ordered from TWIST Bioscience + +Table S3: Probe amplification reagents and costs + +Table S4: Strains used in this study + +Table S5: Sample summary + +Table S6: DGE analysis + +Table S7: DGE analysis using 10x UMI instead of Probe UMI + +Table S8: DGE analysis using bulk median instead of per- cell maximum probe counts + +Table S9: Genes in each section of the Venn diagram in Figure 3d + +Table S10: Genes used in volcano plots on Figure 3C and Figures S2, S4 and S5 + +## Supplementary Notes + +<--- Page Split ---> + +Supplemental note 1: Additional biology related to Figure 3 and Figure S2 Supplemental note 2: Additional info on Figure S3 and use of droplet RT instead of PCR Supplemental note 3: Additional biology related to Figure 4a and Figure S4 Supplemental note 4: Additional biology related to Figure 4b and Figure S5 Supplemental note 5: Calculation of probe concentrations + +## Author Contributions + +R.M., S.L, and A.Z.R. performed single cell and microbiology wet- lab experiments; R.M., D.S., S.H., and A.Z.R analyzed data. D.S. wrote custom analysis scripts. R.M., D.S., S.H. and A.Z.R. wrote and edited the manuscript. S.H. and A.Z.R. conceptualized the project; All authors designed experiments and reviewed the manuscript. + +## References + +1. Ackermann, M. A functional perspective on phenotypic heterogeneity in microorganisms. + +Nat. Rev. Microbiol. 13, 497- 508 (2015). + +2. Huang, K. C. Applications of imaging for bacterial systems biology. Curr. Opin. Microbiol. 27, 114-120 (2015). + +3. Locke, J. C. W. & Elowitz, M. B. Using movies to analyse gene circuit dynamics in single cells. Nat. Rev. Microbiol. 7, 383-392 (2009). + +4. Milo, R. & Phillips, R. Cell Biology by the Numbers. (Garland Science, 2015). + +5. Chen, S.-H. et al. UPS 2.0: unique probe selector for probe design and oligonucleotide microarrays at the pangenomic/genomic level. BMC Genomics 11 Suppl 4, S6 (2010). + +6. Gundlach, J., Rath, H., Herzberg, C., Mader, U. & Stülke, J. Second Messenger Signaling in Bacillus subtilis: Accumulation of Cyclic di-AMP Inhibits Biofilm Formation. Front. Microbiol. 7, 804 (2016). + +7. Wang, D. et al. The orphan gene ybjN conveys pleiotropic effects on multicellular behavior + +<--- Page Split ---> + +and survival of Escherichia coli. PLoS One 6, e25293 (2011).8. Schmidt, T. L. et al. Scalable amplification of strand subsets from chip-synthesized oligonucleotide libraries. Nat. Commun. 6, 1–7 (2015).9. Stuart, T. et al. Comprehensive Integration of Single cell Data. Cell 177, 1888–1902.e21 (2019).10. Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).11. Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L. & Wold, B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621–628 (2008).12. González-Ballester, D. et al. RNA-seq analysis of sulfur-deprived Chlamydomonas cells reveals aspects of acclimation critical for cell survival. Plant Cell 22, 2058–2084 (2010).13. Bartholomäus, A. et al. Bacteria differently regulate mRNA abundance to specifically respond to various stresses. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 374, 20150069 (2016).14. Ogura, M. et al. Whole-genome analysis of genes regulated by the Bacillus subtilis competence transcription factor ComK. J. Bacteriol. 184, 2344–2351 (2002).15. Berka, R. M. et al. Microarray analysis of the Bacillus subtilis K-state: genome-wide expression changes dependent on ComK. Mol. Microbiol. 43, 1331–1345 (2002).16. Dubnau, D. The regulation of genetic competence in Bacillus subtilis. Mol. Microbiol. 5, 11–18 (1991).17. Rosenthal, A. Z. et al. Metabolic interactions between dynamic bacterial subpopulations. Elife 7, (2018).18. Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29 (2000).19. Carbon, S. et al. AmiGO: online access to ontology and annotation data. Bioinformatics 25, 288–289 (2009). + +<--- Page Split ---> + +20. Mi, H., Muruganujan, A., Ebert, D., Huang, X. & Thomas, P. D. PANTHER version 14: more genomes, a new PANTHER GO-slim and improvements in enrichment analysis tools. Nucleic Acids Res. 47, D419–D426 (2019).21. Eisenstein, B. I. Phase variation of type 1 fimbriae in Escherichia coli is under transcriptional control. Science 214, 337–339 (1981).22. Adicipitaningrum, A. M., Blomfield, I. C. & Tans, S. J. Direct observation of type 1 fimbrial switching. EMBO Rep. 10, 527–532 (2009).23. Spaulding, C. N. et al. Functional role of the type 1 pilus rod structure in mediating host-pathogen interactions. Elife 7, e31662 (2018).24. Clark, D. P. The fermentation pathways of Escherichia coli. FEMS Microbiol. Rev. 5, 223–234 (1989).25. Dar, D., Dar, N., Cai, L. & Newman, D. K. In situ single cell activities of microbial populations revealed by spatial transcriptomics. 40 (2021).26. Simanshu, D. K., Chittori, S., Savithri, H. S. & Murthy, M. R. N. Structure and function of enzymes involved in the anaerobic degradation of L-threonine to propionate. J. Biosci. 32, 1195–1206 (2007).27. Sawers, G. A novel mechanism controls anaerobic and catabolite regulation of the Escherichia coli tdc operon: Regulation of the E. coli tdc operon. Mol. Microbiol. 39, 1285–1298 (2001).28. Wu, G. et al. Arginine metabolism and nutrition in growth, health and disease. Amino Acids 37, 153–168 (2009).29. Kuchina, A. et al. Microbial single cell RNA sequencing by split-pool barcoding. Science (2020) doi:10.1126/science.aba5257.30. Blattman, S. B., Jiang, W., Oikonomou, P. & Tavazoie, S. Prokaryotic single cell RNA sequencing by in situ combinatorial indexing. Nat Microbiol 5, 1192–1201 (2020).31. Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by + +<--- Page Split ---> + +752 pseudotemporal ordering of single cells. Nat. Biotechnol. 32, 381–386 (2014). + +753 32. van der Wijst, M. G. P. et al. Single cell RNA sequencing identifies celltype-specific cis-eQTLs and co-expression QTLs. Nat. Genet. 50, 493–497 (2018). + +<--- Page Split ---> + +For B. subtilis in minimal media, in addition to the competent and sporulating cell subpopulations discussed in the main text, we observe two other clearly distinguishable cell groups indicative of distinct biological states. The largest group of cells (comprising clusters 1, 2, 3, 5, and part of cluster 7) upregulate dhbA, dhbB, dhbC, dhbE, and dhbF (the 5 enzymes implicated in the biosynthesis of bacillibactin) as well as yusV, an ABC transporter of bacillibactin. In addition, at least 3 out of the 5 clusters within the subpopulation showed upregulation of genes involved in thiamine biosynthesis (tenA, tenI, thiD, thiG, thiF, thiS, thiO; gene set enrichment = 16.85, FDR = 2.79E- 05), amino acid biosynthesis (fold enrichment = 5.23, FDR = 1.48E- 07), amino acid activation (thrS, gatB, ileS, aspS; fold enrichment = 6.22, FDR = 3.46E- 02), surfactant biosynthesis (srfAA, srfAB, srfAC, srfAD), "de novo" IMP biosynthesis (gene set fold enrichment = 25.79, FDR = 9.10E- 09), purine biosynthesis (purB, purC, purD, purQ, purH, purF, purL, purK, purM, purS; gene set fold enrichment = 21.66, FDR = 1.83E- 02), pyrimidine biosynthesis (pyrAA, pyrK, pyrF, pyrH; gene set enrichment = 9.63, FDR = 2.59E- 02), chorismate biosynthesis (aroA, aroB, aroC, aroH, aroE; gene set fold enrichment = 18.05, FDR = 4.60E- 03), cell motility (hag, fliY, fliD, fliH, flhP; gene set enrichment = 6.19, FDR = 1.44E- 02), sulfate assimilation (cysl, cysH, cysC, cysJ; gene set enrichment = 18.05; FDR = 4.56E- 03), and biotin biosynthesis (bioA, bioB, bioD, bioF, bioW; gene set fold enrichment = 18.05, FDR = 4.65E- 03). Altogether, this subpopulation appears to represent the most metabolically active cell state within the total population, an observation further supported by the fact that genes encoding ribosomal components & translation machinery were also enriched (rpmI, rplU, rplR, rpsM, rpsE, rpsI, rpsN, rpsD, rpoB, rplV, rplX, rplB, rpmA, rpsS, rplS, rpsP, rplL, rpmD, rpmC; "translation" gene set fold enrichment = 7.81, FDR = 3.07E- 12). Cell cluster #2, part of the subpopulation, may represent the gateway into sporulation as early sporulation factors sigF, spooA, and spoIIAB are all uniquely upregulated and cells are projected in close proximity with cluster 9 after dimensional reduction. In addition, these cells downregulate genes implicated in motility (hag, fliD, fliG, fliY, fliK, fliM, flgK, flgL) in stark contrast to other cells within the subpopulation. Furthermore, cells within cluster #5 are distinguished from others in the "metabolically active" subpopulation by expression of genes related to arginine biosynthesis via ornithine. L- arginine is used by cells for protein synthesis as well as for production of polyamines such as putrescine and spermidine; compounds which have functions in nucleic acid binding, protein activation, and membrane stabilization due to their polycationic state. The 4th distinct cell state detected in B. subtilis, only slightly less in size compared to the "metabolically active" cell group (45% of total population), was characterized by lower expression of many of the genes upregulated in the "metabolically active" subpopulation. Uniquely, cells in cluster 0 (930 cells, 76% of the subpopulation) were found to overexpress sdpB (logFC = 0.68, adjusted p-value = 4.7E- 4). Under starvation conditions, B. subtilis produces an antimicrobial peptide through the sdpABC operon that is active against its own species, a tactic used by the cell to increase its own chance of survival and delay commitment to sporulation. While the sdpC gene encodes the 42AA toxic peptide, sdpA & sdpB are required for maturation of the peptide into its active form prior to secretion. sdpB is a multipass membrane protein which affects the toxicity of the final SDP peptide without being involved in signal peptide cleavage, disulfide bond formation, or secretion. The sdpABC operon was found expressed in cells within each of the four subpopulations, however, it is only within cluster #0 that sdpB was found to be significantly upregulated. This finding, along with the overall reduction in transcript load as compared to the "metabolically active" subpopulation, suggests a more dormant biological state in which cells appear to be facing a nutritional strain but have not yet committed to an alternative developmental program. This hypothesis is supported by the fact that cells in cluster 7 are split between the two largest subpopulations in the UMAP projection as well as by the clustering of a few cells associated with competence (part of cluster 8) near this dormant subpopulation – possibly suggesting a developmental transition between the two states. + +<--- Page Split ---> + +Supplemental note 2: Additional info on Figure S3 and use of droplet RT instead of PCR + +While the approach in Fig S3 (RT in drops instead of PCR) proved less sensitive in transcript capture compared to the final, optimized method, we still resolve distinct biological states upon dimensional reduction of expression vectors including cell subpopulations corresponding to competence and sporulation. Additionally, we observe a majority grouping of cells (clusters #0- 5) with significantly increased expression of ribosomal genes as well as genes involved in purine and amino acid biosynthesis - corresponding to the "metabolically active" subpopulation defined previously. scRNA- seq also revealed a subpopulation of cells (cluster #8, 1.9% of cells) within the sample exhibiting a state unforeseen in the previous run - characterized by the upregulation of genes encoding a manganese importer (mntABCD, logFCs > 1.99, adjusted p- values < 1E- 11), sigma factor sigW (logFC = 1.69, adjusted p- value = 5E- 16), and stress response proteins (yuaF, yceDEF, yqfA, yqeZ, logFCs > 0.99, adjusted p- values < 3.3E- 04) involved in maintaining cell membrane integrity. yceF, specifically, is a cell envelope stress protein that conveys resistance to manganese. katA, an iron- binding catalase which protects cells against toxification from hydrogen peroxide, was also upregulated. Altogether, this cluster of cells appears to be responding to stress induced by a low concentration of manganese within the environment by upregulating manganese uptake machinery. maeA was also upregulated in these cells (logFC = 1.9, adjusted p- value = 3.65E- 34). maeE encodes a malic enzyme necessary for catabolic repression of pyruvate import (pftAB) as well as for maintaining ATP levels during growth on the carbon source by conversion of malate to pyruvate and generation of NADH. + +## Supplemental note 3: Additional biology related to Figure 4a and Figure S4 + +Single cell transcriptomes of E. coli grown in M9 were clustered into ten groups. In the largest group of cells (cluster #0, 514 cells, 15.5% of total), ribosomal genes (rpl operon, rpsC, rpsD, rpmC, rpmD) as well as deaD and priB are amongst the most upregulated compared to the rest of the cell population. deaD (logFC = 0.7, adjusted p- value = 2.3E- 11) is a DEAD- box RNA helicase involved in ribosome biogenesis, translation initiation, and mRNA degradation at low temperatures - forming a cold- shock degradosome with RNase E. Kuchina et al. also identified a subpopulation of E. coli differentially expressing deaD, along with other cold shock proteins, in their scRNA- seq experiments and hypothesized the cluster was an artifact of sample preparation. Since our protocol fixes cells directly in culture, we find it unlikely that this state is a product of cold shock - a conclusion reinforced by the fact that we observe no other cold shock associated proteins within the set of upregulated genes. priB (logFC = 0.37, adjusted p- value = 6.4E- 12) encodes a protein within the primosome which catalyzes lag strand priming during DNA replication. Conversely, genes involved in pyrimidine biosynthesis (pyrl, pyrB, pyrC) were significantly downregulated within the cluster. Interspersed with cluster #0 on the dimensionally reduced plot, we observe a group of cells (cluster #8, 141 cells) highly upregulating genes from the fim operon, as discussed in the main text. Cells in cluster #2 appear less active overall, as all genes found to be significantly differentially expressed within the cluster are downregulated compared to the rest of the population. This may be an effect of poor probe penetration within the protocol or indicative of a more basal cell state. Cluster #3 reveals no significantly overrepresented gene sets by GSEA after accounting for the false discovery rate. Upregulated genes reported in the cluster include ompT and ompX as well as tolB and tolC (logFCs > 0.25, adjusted p- values < 1E- 14). In addition, scRNA- seq reveals a cluster of cells (cluster #4, 403 cells, 12% of total) that appear to be entering stationary phase; exhibiting a relative increase in + +<--- Page Split ---> + +expression of dps (logFC = 0.91, adjusted p- value = 4.17E- 26). dps encodes a protective protein which binds to the bacterial chromosome and sequesters intracellular Fe2+, forming a highly stable protein- DNA mineral complex which protects the DNA from oxidative damage. In E. coli, dps has been associated with the transition into stationary phase as cells quickly work to preserve genetic material from stressors including UV irradiation, metal toxicity, thermal fluctuations, and acid/base shock. Cluster #9 (117 cells, 3.5% of total) was also shown to upregulate dps (logFC = 1.05, adjusted p- value = 7.04E- 16) while highly expressing the gad (gadA, gadB, gadC) and hde (hdeA, hdeB, hdeD) operons. Both operons encode resistance to acid stress and fall under the gene set of "regulation of intracellular pH" (fold enrichment > 100, FDR = 6.72E- 03). gadA and gadB produce the subunits of glutamate carboxylase which allows cells to interconvert between L- glutamate and GABA by incorporation of an intracellular proton while gadC encodes the antiporter enabling L- glutamate uptake and GABA export. hdeA and hdeB encode acid stress chaperone proteins with differing pH optimums; both preventing the aggregation of periplasmic proteins denatured under acidic conditions. osmY, encoding a periplasmic chaperone protein, was also upregulated within this cluster and has been shown to be induced by entry into the stationary phase. Altogether, DGE analysis reveals a fraction of cells within the population that are experiencing significant stress. Cells in cluster #4 appear to be at the initiation of this state while cells in the cluster #9 are more clearly induced, expressing dps along with proteins dealing with acid, osmotic, and oxidative stresses. It is possible these clusters represent cells transitioning into stationary phase or, alternatively, a transient state experienced during active metabolism/exponential growth. Cells within clusters 5 and 7 upregulate genes involved in pyrimidine biosynthesis as discussed in the main text. Interestingly, cluster #5 differentially overexpresses argG (logFC = 0.89, adjusted p- value = 6.57E- 29), although no other genes specific to arginine biosynthesis were found to be upregulated. Both clusters upregulate ygjF, encoding a G/U mismatch-specific DNA glycosylase to correct mispairings in double stranded DNA that arise from alkylation or deamination of cytosine. ygjF is often active in stationary-phase cells 36. Cluster 6 is defined by the upregulation of genes involved in arginine biosynthesis as discussed in the main text. + +## Supplemental note 4: Additional biology related to Figure 4b and Figure S5 + +Upregulated genes within the largest cell cluster (cluster #0, 677 cells, 18% of total) included fliC (LogFC = 1.02, adjusted p- value = 3.81E- 54), uncharacterized genes ygeWXY (logFC > 0.7, adjusted p- values < 3.3E- 32), genes putatively involved in the conversion of adenine to guanine during nucleotide salvage (xdhA, xdhD, logFCs > 0.85, adjusted p- values < 1E- 34), ssnA (logFC = 0.80, adjusted p- value = 9.8E- 23), a putative, phase- dependent regulator of cell viability during later stage growth, and genes in the tna operon (tnaABC, logFCs > 0.45, adjusted p- values < 1E- 28). tnaA encodes tryptophanase, which converts L- tryptophan to indole and pyruvate as part of amino acid catabolism while tnaB is a low affinity permease and tnaC is a positively regulating leader peptide. Cell clusters #1 and #4, adjacent and densely packed groups on the UMAP, appear to represent a subpopulation of cells responding to over- acidification of the cytoplasm. Both clusters significantly upregulate the gad operon (gadABC) as well as dps. In addition, acid stress chaperone proteins hdeA and hdeB are overexpressed in cluster #4. No statistically significant gene sets were identified from upregulated genes in clusters #2- 3, however, cells in both clusters express genes related to anaerobic fermentation as expected under the culture conditions. Cells in cluster #5 (283 cells, 7.5% of total) are characterized by the upregulation of + +<--- Page Split ---> + +genes involved in aerobic respiration as discussed in the main text. We also observe the strong upregulation of genes involved in ribosome assembly (fold enrichment \(= 15.26\) , FDR \(= 6.16\mathrm{E - 21}\) ) and translation machinery (fold enrichment \(= 12.01\) , FDR \(= 6.26\mathrm{E - 33}\) ) in these cells, indicating a more active metabolic state which can most likely be attributed to the increases in ATP yield that is achieved by aerobic respiration. We observe similar enrichment of ribosome- related genes in cell cluster #6 (191 cells, \(5\%\) of total) along with the upregulation of a few genes in glycolysis (eno, lpd, aceEF, logFC \(>0.5\) , adjusted p- values \(< 2.2\mathrm{E - 06}\) ) and oxidative phosphorylation (atpD, cydAB, logFC \(>0.7\) , adjusted p- values \(< 1\mathrm{E - 06}\) ). These findings suggest that cluster #6 is in a differentiating state, transitioning between aerobic and anaerobic metabolism. Interestingly, cell cluster #7 (135 cells, \(3.6\%\) of total) appears to capture cells undergoing anaerobic metabolism while responding to oxidative stress. Gene sets describing anaerobic respiration (acnA, yjil, nrfA, frdC, frdA, ynfF, ynfE, nrfB, dmsA; fold enrichment \(= 8.00\) , FDR \(= 5.63\mathrm{E - 03}\) ) and oxidative stress (acnA, ahpF, osmC, qor, pgi, sufD, ydbK, yliT, yggE, clpA; fold enrichment \(=\) , FDR \(= 2.68\mathrm{E - 02}\) ) were both identified as overrepresented in the list of upregulated genes within the cluster. In addition, stress proteins such as hspQ (a heat shock response protein) and osmY (hyperosmotic response protein) were upregulated along with flu (logFC \(= 0.47\) , adjusted p- value \(= 2.88\mathrm{E - 14}\) ), a surface- display, self- recognizing adhesin protein encoding gene which enables auto aggregation and flocculation of cells in static liquid culture as well as biofilm formation by uropathogenic E coli in situ. Cluster #8 (111 cells, \(3\%\) of total) is characterized by the strong upregulation of the tdc operon (threonine metabolism under anaerobic conditions) as discussed in the main text. + +## Supplemental note 5: Calculation of probe concentrations + +## Determining Probe Concentration + +To determine the concentration of probes needed to saturate an approximate number of transcripts in solution, saturation was defined as total, nonspecific probe coverage for each transcript. In other words, a pool of transcripts is considered saturated when, for each molecule, there is an entire library of probes ( \(n = 1\) probe per unique design) exclusively dedicated. In reality, this system would be over- saturated as it contains enough probes to tag every transcript even if all transcripts have the same sequence. + +In our protocol, we prepare 150 uL of fixed cells from a 2X concentrated culture collected at mid- log phase. Assuming the culture contains 1E8 cells/mL prior to concentration and each bacterial cell contains 2,000 transcripts \(^{13}\) , the following estimate can be made using a probe set containing 20,000 unique ssDNA probes of length 138 bp ( \(\sim 42,500 \text{g/mol}\) ). + +<--- Page Split ---> + +\[\frac{2E8 \text{ cells}}{mL} * \left(\frac{2000 \text{ transcripts}}{cell}\right) = \frac{4E11 \text{ transcripts}}{mL} = 0.66 \text{ nM transcripts}\] + +\[\left(\frac{1 \text{ probeset}}{\text{transcript}}\right) \left(\frac{20,000 \text{ unique probes}}{1 \text{ probeset}}\right) = 20,000 \frac{\text{probes}}{\text{transcript}}\] + +\[0.66 \text{nM transcript} * \left(\frac{20,000 \text{ probes}}{\text{transcript}}\right) = 13,020 \text{nM probes} = 13.20 \text{uM probes}\] + +\[13.20 \text{uM probes} \left(\frac{1 \text{M}}{10^6 \text{uM}}\right) \left(\frac{42,500 \text{g}}{\text{mol}}\right) = \frac{0.561 \text{g}}{L} = \frac{560 \text{ng}}{\text{uL}}\] + +In our final experiments, we achieved a probe concentration of \(600 \text{ ng / uL}\) (0.6 mg/mL). To note, our calculations assume equal representation of all probe designs in the applied probe library which is unlikely due to biases introduced during probe synthesis and amplification. However, in all likelihood, this number represents a gross overestimate of probe requirements for a few reasons. Cell concentration and, therefore, transcript load was purposefully overestimated by assuming an exponentially growing culture contains 1E8 cells/mL; this number better represents the carrying capacity of B. subtilis in minimal media. In addition, our definition of "saturation" was overly strict, as discussed above. Lastly, probe libraries were designed redundantly to minimize noise in the downstream analysis by creating multiple probes to target different regions of the same gene. For the B. subtilis library, we designed, on average, 7 probes per gene - further assuring that our final probe concentration is sufficient to tag all available transcripts for scRNA-seq. + +<--- Page Split ---> + +## Figures + +![](images/Figure_2.jpg) + + + +
Figure 1
+ +Genome wide scRNA-seq probe design and synthesis a. A genome for the organism of interest is used to design a library of probes that are complementary to unique regions within each gene's coding sequence (CDS). Complementary ssDNA oligonucleotides are amended by addition of an extender sequence at the + +<--- Page Split ---> + +5' end and 3' PCR handle. Sequences are ordered as an oligopool b. Probe sequences are inverted in- silico and restriction sites and handles are added for circle- to- circle amplification c. The large, inverse oligonucleotide library is commercially- synthesized and pooled. Ordered oligo- pools are amplified in a 3- step "rolling circle" reaction with Phi29 polymerase to produce the final, correctly oriented proto- probe library. d. A random 12mer sequence (UMI) and a 30- base polyA tail is added by primer extension with a blocked primer that is complementary to the 3' extender sequence. Single stranded DNA containing a universal PCR handle, UMI, complementary transcript region, and poly- A tail (listed 5' to 3') is purified. e. Finalized probe libraries are free of amplification primers and well distributed. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 2
+ +Microfluidic probe- based scRNA- seq method and validation. a. Cells are fixed and permeabilized to allow penetration of thousands of unique, genome- specific oligonucleotide probes. Hybridized probes "retrofit" transcripts with a poly- A tail & UMI while unhybridized probes are washed away. b. Cells are then flowed through a commercial microfluidic device which encapsulates single cells into droplets (shown in c.) containing barcoded primers conjugated to a hydrogel microsphere and PCR reagents. d. Barcoded cDNA + +<--- Page Split ---> + +is generated from the mRNA:probe hybridized complex via in-droplet PCR or RT reaction. Droplets are then broken, the pooled cDNA amplified further before sequencing. Single cell transcriptomes are resolved, clustered, and visualized. e. Transcriptomic estimation by probe hybridization in bulk samples is correlated to transcriptomic measurement of bulk samples by traditional RNA-seq f. Species mixture ("barnyard") plot demonstrates that single cells of different bacterial species can be resolved by barcode after microfluidic encapsulation. g. Aggregated probe-based signal from thousands of single cells correlates with the average probe-based signal of the bulk population. + +![PLACEHOLDER_34_0] + + +<--- Page Split ---> + +## Figure 3 + +scRNA- seq analysis reveals known and novel states in B subtilis. a. Heatmap of marker gene expression (z- score of log- transformed values) from 2,784 individual B. subtilis cells organized into 10 clusters. b. UMAP 2- dimensional representation of the 10 cell clusters reveals 4 highly distinct transcriptomic signatures c. Single cell expression of key marker genes for competence (comGE; clusters 6 and 8), sporulation (spollID, cluster 9) and arginine synthesis (argC, cluster 5) are highlighted on the UMAP (top panels, left to right). Volcano plots of genes expressed in at least \(25\%\) of cells in the corresponding clusters, with genes from the respective processes highlighted in green (middle panels). The lower panels show the presence of each heterogeneous marker in the population as confirmed by fluorescent promoter- reporter constructs (PcomG- YFP, PcotY- YFP, PargC- YFP, respectively). A phase bright spore can be seen in the cotY- expressing cell. d. \(90\%\) of all comK regulon genes probed (45/50) were significantly differentially upregulated in cell clusters 6 and 8 (Bonferroni corrected P- value \(\leq 0.05\) for each gene) + +<--- Page Split ---> +![PLACEHOLDER_36_0] + +
Figure 4
+ +Heterogeneous gene expression in E coli grown in minimal medium and anaerobically. a. Heatmap of marker gene expression (z- score of log- transformed values) from 3,315 individual E. coli cells grown aerobically in minimal M9 media organized into 10 clusters. b. Heatmap of gene expression from 3,763 individual E. coli cells grown in anaerobic LB media organized into 9 clusters. c. UMAP 2- dimensional representation of the 10 cell clusters from aerobic M9 culture conditions. d. Flagellar components are + +<--- Page Split ---> + +heterogeneously expressed in aerobic M9 media - a key flagellar gene (flagellin - flic) is preferentially expressed by cells in cluster 1. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. e. UMAP 2- dimensional representation of the 9 cell clusters from E coli fermenting LB media culture conditions. f. Flagellar components are heterogeneously expressed in fermentation conditions - a key flagellar gene (flagellar basal- body rod protein - fIgC) is preferentially expressed by cells in cluster 5. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. g. Many aerobic respiration genes and other metabolic and flagellar genes are overexpressed in cluster 5. Volcano plots of genes over- expressed in at least \(25\%\) of cells in cluster 5 from these processes are highlighted in colors as labeled. Green labeled genes, involved in central- carbon metabolism, are also highlighted as green arrows in the metabolic map. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- SupptablesfinalizedforinitialsubmissionAutosaved.xlsx- suppfiguresfinalizedforinitialsubmission.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd_det.mmd b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7bb7a4996427cd4f3ea50495a8eaca43e441bba1 --- /dev/null +++ b/preprint/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd/preprint__0dfb2ae81f3db04badcf51c90bade8b085a06e2563ac059c26ab25153a277bfd_det.mmd @@ -0,0 +1,607 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 886, 208]]<|/det|> +# Droplet-based single cell RNA sequencing of bacteria identifies known and previously unseen cellular states + +<|ref|>text<|/ref|><|det|>[[44, 228, 460, 249]]<|/det|> +Adam Rosenthal ( adam.rosenthal@iff.com) + +<|ref|>text<|/ref|><|det|>[[52, 252, 125, 269]]<|/det|> +Dupont + +<|ref|>text<|/ref|><|det|>[[44, 276, 168, 315]]<|/det|> +Ryan McNulty Dupont + +<|ref|>text<|/ref|><|det|>[[44, 322, 210, 361]]<|/det|> +Duluxan Sritharan Harvard University + +<|ref|>text<|/ref|><|det|>[[44, 367, 305, 408]]<|/det|> +Shichen Liu Dana- Farber Cancer Institute + +<|ref|>text<|/ref|><|det|>[[44, 414, 189, 432]]<|/det|> +Sahand Hormoz + +<|ref|>text<|/ref|><|det|>[[50, 435, 888, 456]]<|/det|> +Harvard Medical School / Dana- Farber Cancer Institute https://orcid.org/0000- 0002- 4384- 4428 + +<|ref|>sub_title<|/ref|><|det|>[[44, 496, 284, 516]]<|/det|> +## Biological Sciences - Article + +<|ref|>text<|/ref|><|det|>[[42, 533, 956, 578]]<|/det|> +Keywords: Clonal Bacterial Populations, Specialized Cell States, Isogenic Bacterial Populations, Metabolic Pathway Segregation + +<|ref|>text<|/ref|><|det|>[[44, 595, 316, 615]]<|/det|> +Posted Date: March 22nd, 2021 + +<|ref|>text<|/ref|><|det|>[[42, 632, 463, 653]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 322406/v1 + +<|ref|>text<|/ref|><|det|>[[42, 669, 910, 713]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[115, 88, 882, 131]]<|/det|> +# Droplet-based single cell RNA sequencing of bacteria identifies known and previously unseen cellular states + +<|ref|>text<|/ref|><|det|>[[115, 168, 881, 206]]<|/det|> +Ryan McNulty \(^{1*}\) , Duluxan Sritharan \(^{2,3*}\) , Shichen Liu \(^{3}\) , Sahand Hormoz \(^{3,4,5\#}\) and Adam Z. Rosenthal \(^{1\#\wedge}\) + +<|ref|>sub_title<|/ref|><|det|>[[115, 243, 219, 261]]<|/det|> +## Affiliations + +<|ref|>text<|/ref|><|det|>[[113, 270, 857, 404]]<|/det|> +\(^{1}\) DuPont Nutrition and Biosciences, Wilmington, DE 19803, USA \(^{2}\) Harvard Graduate Program in Biophysics, Harvard University, Cambridge, MA, USA \(^{3}\) Department of Data Science, Dana- Farber Cancer Institute, Boston, MA, USA \(^{4}\) Department of Systems Biology, Harvard Medical School, Boston, MA, USA \(^{5}\) Broad Institute of MIT and Harvard, Cambridge, MA, USA + +<|ref|>text<|/ref|><|det|>[[113, 439, 855, 514]]<|/det|> +\*These authors contributed equally to this paper\#Correspondence: Sahand Hormoz@hms.harvard.edu and adam.rosenthal@iff.com\^current affiliation: IFF Health and Biosciences + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 90, 190, 106]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[114, 115, 882, 393]]<|/det|> +Clonal bacterial populations rely on transcriptional variation to differentiate into specialized cell states that increase the community's fitness. Such heterogeneous gene expression is implicated in many fundamental microbial processes including sporulation, cell communication, detoxification, substrate utilization, competence, biofilm formation, motility, pathogenicity, and antibiotic resistance1. To identify these specialized cell states and determine the processes by which they develop, we need to study isogenic bacterial populations at the single cell level2,3. Here, we develop a method that uses DNA probes and leverages an existing commercial microfluidic platform (10X Chromium) to conduct bacterial single cell RNA sequencing. We sequenced the transcriptome of over 15,000 individual bacterial cells, detecting on average 365 transcripts mapping to 265 genes per cell in B. subtilis and 329 transcripts mapping to 149 genes per cell in E. coli. Our findings correctly identify known cell states and uncover previously unreported cell states. Interestingly, we find that some metabolic pathways segregate into distinct subpopulations across different bacteria and growth conditions, suggesting that some cellular processes may be more prone to differentiation than others. Our high throughput, highly resolved single cell transcriptomic platform can be broadly used for understanding heterogeneity in microbial populations. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 222, 107]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[112, 116, 883, 462]]<|/det|> +1 In recent years, single cell RNA sequencing (scRNA- seq) has enabled the agnostic assessment of transcriptional heterogeneity in eukaryotic systems. Since the introduction of microfluidic techniques which allow encapsulation of cells and tagging of poly- adenylated transcripts in small droplets, scRNA- seq studies of mammalian populations have grown exponentially. However, tools to comprehensively study differential gene expression in bacterial populations remain severely limited due to several significant technical challenges. First, total mRNA abundance in bacteria is two orders of magnitude lower than that of eukaryotes, with a single bacterial cell containing approximately \(10^{3} - 10^{4}\) transcripts during exponential growth4. Second, transcriptional turnover is much faster in bacteria with mRNA half- life often on the scale of minutes, compared with hours for eukaryotic transcripts4. Third, bacterial transcripts do not intrinsically include a 3' poly- adenosine tail and, therefore, mRNA cannot easily be tagged and selectively enriched against rRNA, which makes up \(>95\%\) of the bacterial transcriptome4. Lastly, accessing mRNA requires cell permeabilization which is markedly more difficult in bacteria due to the diversity in membrane structure and variability of the peptidoglycan layer. We reasoned that a new method combining the advantages of microfluidic single cell barcoding in droplets with the ability to tag transcripts using in- situ hybridization of oligonucleotide probes could overcome these challenges. Here, we present a method for prokaryotic single cell RNA sequencing which utilizes a commercial, benchtop microfluidic device (Chromium Controller from 10X Genomics) and custom ssDNA probe libraries to resolve the mRNA profile of thousands of Gram positive (Bacillus subtilis 168) and Gram negative (Escherichia coli MG1655) bacterial cells. + +<|ref|>sub_title<|/ref|><|det|>[[115, 472, 299, 489]]<|/det|> +## Method Development + +<|ref|>sub_title<|/ref|><|det|>[[115, 498, 416, 515]]<|/det|> +## Probe Design & Library Generation + +<|ref|>text<|/ref|><|det|>[[111, 525, 883, 905]]<|/det|> +To leverage existing microfluidic single cell sequencing platforms, we devised a method relying on tagging of individual transcripts with DNA probes. This approach requires the generation of a large oligonucleotide library that is complementary to all protein coding sequences within a genome (Figure 1). Multiple DNA regions of 50 bp were chosen from each ORF based on uniqueness as determined by UPS2 software5 or based on previously published oligonucleotide arrays6,7. These sequences then served as the hybridization regions of ssDNA probes which were designed to target mRNA by sequence complementarity. ssDNA probes also contained a 5' PCR handle for library generation, a Unique Molecular Identifier (UMI), and a 3' poly adenosine tail (A30) for retrofitting prokaryotic transcripts to the Chromium Single Cell 3' system (Figure 1a). Multiple probes (complementary to different regions) were designed for each gene to enhance transcript capture efficiency and decrease noise caused by poor hybridization and/or insufficient amplification of any given probe. Complete species libraries contained 29,765 probes for B subtilis and 21,527 probes for E. coli, and targeted 2,959 and 4,181 genes, respectively. Libraries were ordered at sub- femtomole quantities from Twist Biosciences (a one- time cost of 5¢ - 15¢ per probe) and amplified by rolling circle amplification8 to obtain a sufficient concentration (0.25 mg = 10.25 nM/library or approximately 0.35 pM of each probe) for scRNA- seq experiments (Figure1b, Methods, Supplementary Tables S1 and S2, Supplementary Figure S1). Probe libraries were completed by addition of randomized 12 bp UMI sequences and a poly- A tail and purified by PAGE. Completed libraries had a uniform coverage of probes (Figure 1e). The rolling circle amplification approach permits re- amplifying probe sets for unlimited subsequent experiments without the need to re- order probes, at an upfront cost of under 1¢ per cell (Supplementary Table S3). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 90, 615, 666]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 672, 629, 690]]<|/det|> +
Figure 1. Genome wide scRNA-seq probe design and synthesis
+ +<|ref|>text<|/ref|><|det|>[[112, 697, 883, 870]]<|/det|> +a. A genome for the organism of interest is used to design a library of probes that are complementary to unique regions within each gene's coding sequence (CDS). Complementary ssDNA oligonucleotides are amended by addition of an extender sequence at the 5' end and 3' PCR handle. Sequences are ordered as an oligopool b. Probe sequences are inverted in-silico and restriction sites and handles are added for circle-to-circle amplification c. The large, inverse oligonucleotide library is commercially-synthesized and pooled. Ordered oligo-pools are amplified in a 3-step "rolling circle" reaction with Phi29 polymerase to produce the final, correctly oriented proto-probe library. d. A random 12mer sequence (UMI) and a 30-base polyA tail is added by primer extension with a blocked primer that is complementary to the 3' extender sequence. Single stranded DNA containing a universal PCR handle, UMI, complementary transcript region, and poly-A tail (listed 5' to 3') is purified. e. Finalized probe libraries are free of amplification primers and well distributed. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 228]]<|/det|> +Before microfluidic encapsulation, bacteria were fixed in \(1\%\) paraformaldehyde and permeabilized (Methods). Permeabilized bacteria were incubated with their corresponding DNA probe library. Non- hybridized probes were washed away (Figure 2a, Methods). Next, the bacteria were run through a 10X controller, where the DNA probes were captured and barcoded in a manner analogous to the barcoding of the transcriptome of mammalian cells (Figure 2 b- d). The resulting libraries were sequenced, preprocessed with custom scripts (https://gitlab.com/hormozlab/bacteria_scrnaSeq), and analyzed with the standard CellRanger pipeline and the Seurat analysis package \(^{9,10}\) (see Methods). + +<|ref|>sub_title<|/ref|><|det|>[[115, 236, 690, 255]]<|/det|> +## Microfluidic RNA-seq validation in bulk and single cell experiments + +<|ref|>text<|/ref|><|det|>[[113, 264, 883, 420]]<|/det|> +To determine if our probe library can report on the transcriptional state of cells, we split a culture of \(B\) . subtilis cells in late exponential state into two aliquots of approximately \(10^{8}\) cells each. One aliquot was processed using a traditional RNA- seq protocol (Methods) while the second sample was fixed in formaldehyde for in situ hybridization with a probe set designed against the Bacillus subtilis genome. After incubation and washing, the probes that remained bound were amplified by PCR and processed into Illumina libraries. The output was compared to the traditional RNA- seq library. The hybridization and wash conditions at \(50^{\circ}\mathrm{C}\) gave results that are similar to traditional RNA- seq (Figure 2e; \(\mathrm{R}^2 = 0.55\) ). This value is comparable to the correlation observed when comparing RNA- seq to microarray experiments ( \(\mathrm{R}^2 = 0.57 - 0.65\) ) \(^{11,12}\) . + +<|ref|>text<|/ref|><|det|>[[112, 429, 883, 638]]<|/det|> +Next, we tested if single cell transcriptomes can be captured by encapsulating individual bacteria using the 10X Chromium Controller after fixation and in situ probe hybridization. \(E\) . coli cells and \(B\) . subtilis cells were independently pre- treated with probes corresponding to their respective genomes. The prepared bacterial samples were subsequently mixed and loaded onto a microfluidic chip along with a custom PCR master mix containing a DNA primer designed to amplify the back- end PCR handle built into the probes (Figure 1, Methods). This custom mix replaced the standard cDNA reagents supplied by 10X Genomics as the aqueous phase within the droplets. Sequenced libraries from this experiment demonstrate that the microfluidic platform is able to successfully segregate and barcode individual microbial cells (Figure 2f). We observed a "doublet" rate of \(1.3\%\) for 3,373 captured cells, consistent with the expected rate of successful single cell encapsulation based on the specifications of the 10X microfluidic system (reported as an expected \(2.3\%\) multiplet events per 3,000 captured cells \(^{10}\) ). + +<|ref|>text<|/ref|><|det|>[[112, 646, 883, 768]]<|/det|> +To assess whether the signal from individually encapsulated cells provides an unbiased readout of the transcriptomic state of the population, we compared the output produced by capturing probes from a bulk sample of cells \((\geq 10^{8}\) cells) to the signal obtained by summing the probe counts over thousands of individually tagged cells (aggregated UMI counts). The number of reads across all probes between the two samples were highly correlated ( \(\mathrm{R}^2 = 0.88\) , Figure 2g), confirming that the signal from single cells provides an unbiased representation of transcriptional states. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[131, 90, 608, 722]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 728, 673, 745]]<|/det|> +
Figure 2. Microfluidic probe-based scRNA-seq method and validation.
+ +<|ref|>text<|/ref|><|det|>[[113, 755, 883, 897]]<|/det|> +a. Cells are fixed and permeabilized to allow penetration of thousands of unique, genome-specific oligonucleotide probes. Hybridized probes "retrofit" transcripts with a poly-A tail & UMI while unhybridized probes are washed away. b. Cells are then flowed through a commercial microfluidic device which encapsulates single cells into droplets (shown in c.) containing barcoded primers conjugated to a hydrogel microsphere and PCR reagents. d. Barcoded cDNA is generated from the mRNA:probe hybridized complex via in-droplet PCR or RT reaction. Droplets are then broken, the pooled cDNA amplified further before sequencing. Single cell transcriptomes are resolved, clustered, and visualized. e. Transcriptomic estimation by probe hybridization in bulk samples is correlated to transcriptomic measurement of bulk samples by traditional RNA-seq f. Species mixture ("barnyard") plot demonstrates that single cells of different bacterial + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 89, 882, 121]]<|/det|> +species can be resolved by barcode after microfluidic encapsulation. g. Aggregated probe- based signal from thousands of single cells correlates with the average probe- based signal of the bulk population. + +<|ref|>sub_title<|/ref|><|det|>[[115, 149, 182, 165]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 183, 870, 202]]<|/det|> +## Bacterial scRNA-seq identifies known and previously hidden heterogeneity in B. subtilis + +<|ref|>text<|/ref|><|det|>[[115, 216, 882, 357]]<|/det|> +We performed scRNA- seq on B. subtilis grown to late exponential phase in M9 minimal media supplemented with malate. In total, 2,784 cells were captured as determined by unique barcodes with \(550\pm 340\) (mean±SD, median=470) reads per cell. We detected a median of 325 mRNA transcripts per cell (approximately 10- 15% of the total mRNA pool13) corresponding to a median of 241 genes per cell. UMIs per probe ranged from 0- 45 for any given gene in any given cell. The most highly detected probes targeted genes encoding ribosomal proteins (rpsG, rpsH, rpsR, rpsM, rpsS, rplL, rplD, rplO, rplV), translation elongation machinery (fusA, tufA, map), and transcriptional machinery (rpoA). + +<|ref|>text<|/ref|><|det|>[[112, 372, 882, 752]]<|/det|> +Next, we resolved distinct cellular states by dimensional reduction of single cell expression vectors followed by graph- based clustering and analysis of differential gene expression (DGE) using the Seurat package9. For B. subtilis in M9 minimal media in late log growth, we resolved four major transcriptomic signatures comprised of ten cell clusters that capture more subtle variations in gene expression (Figure 3 a- b, Supplementary Figure S2, Supplementary Table S6). Algorithmic grouping of cell expression profiles was robust to the clustering parameters (see Methods). As expected for minimal media cultures in late log, we observed a subpopulation of cells (clusters 6 and 8, approximately 10% of all cells) in the genetically competent state, the physiological state in which bacteria can uptake extracellular DNA and incorporate it into their genome. Cells in these two clusters differentially overexpress the competence master- regulator comK (logFC = 2.37, adjusted p- value = 4E- 179) as well as multiple genes within and downstream of competence operons (comC, comEA, comEC, comFA, comFB, comFC, comGA, comGC, comGD, comGE, comGG, recA, coiA, drpA, sbB, nucA, nin, clpC, recA). The fraction of competent cells in the population was measured by fluorescence microscopy using a fluorescent promoter- reporter of comG (Methods, Figure 3c left panel, Supplementary Figure S12a) and determined to be approximately 10.4% (IQR outlier detection - Methods) - similar to the fraction of the population comprising clusters 6 and 8 (9.3%, Supplementary Table S5). In total, we found that 45 of the 50 comK- regulon genes that were probed were differentially overexpressed in clusters 6 and 8 (adjusted p- values < 0.05; Figure 3d, Supplementary Table S6). Our results agree with numerous studies identifying the competence regulon in B. subtilis14,15, and the percentage of cells displaying natural competence falls within the range (3- 10%) of previous observations in similar media16,17. + +<|ref|>text<|/ref|><|det|>[[115, 768, 882, 890]]<|/det|> +Unexpectedly for these growth conditions, we observed a small subpopulation of cells within the sample (cluster 9, 19 cells, 0.7% of the population) with a transcriptomic signature indicative of sporulation. In these cells, we observed significant upregulation of transcripts corresponding to sigma factors associated with sporulation, sigF (logFC = 1.2, adjusted p- value = 4E- 41) and sigG (logFC = 2.2, adjusted p- value = 2E- 63), as well as genes in the ger, cot, and spolVF operons within the cluster. Gene set enrichment analysis (GSEA) on all significantly upregulated genes in cluster 9 using ontological classes from the Gene Ontology Consortium18- 20 reveals a 4.3- fold + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 883, 125]]<|/det|> +enrichment of genes involved in sporulation (Fisher's exact test, \(\mathrm{FDR} = 1\mathrm{E} - 16\) ) as well as spore germination (fold enrichment \(= 8.1\) , \(\mathrm{FDR} = 1.7\mathrm{E} - 03\) ). This finding highlights the ability of probe- + +<|ref|>image<|/ref|><|det|>[[112, 130, 884, 844]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 848, 714, 865]]<|/det|> +
Figure 3. scRNA-seq analysis reveals known and novel states in B subtilis.
+ +<|ref|>text<|/ref|><|det|>[[112, 875, 884, 907]]<|/det|> +a. Heatmap of marker gene expression (z-score of log-transformed values) from 2,784 individual B. subtilis cells organized into 10 clusters. b. UMAP 2-dimensional representation of the 10 cell clusters reveals 4 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 232]]<|/det|> +highly distinct transcriptomic signatures c. Single cell expression of key marker genes for competence (comGE; clusters 6 and 8), sporulation (spollID, cluster 9) and arginine synthesis (argC, cluster 5) are highlighted on the UMAP (top panels, left to right). Volcano plots of genes expressed in at least \(25\%\) of cells in the corresponding clusters, with genes from the respective processes highlighted in green (middle panels). The lower panels show the presence of each heterogeneous marker in the population as confirmed by fluorescent promoter- reporter constructs \((P_{comG} - YFP, P_{cotY} - YFP, P_{argC} - YFP,\) respectively). A phase bright spore can be seen in the cotY- expressing cell. d. \(90\%\) of all comK regulon genes probed (45/50) were significantly differentially upregulated in cell clusters 6 and 8 (Bonferroni corrected P- value \(\leq 0.05\) for each gene) + +<|ref|>text<|/ref|><|det|>[[113, 258, 883, 344]]<|/det|> +based scRNA- seq to detect rare and unexpected cell states in microbial cultures. To confirm this subpopulation, we created a promoter reporter for cotY, a spore coat gene expressed in the sporulating mother cell. Fluorescence observation identifies the presence of cotY expression and spores in a similar percentage of cells as recovered by scRNA- seq (0.5- 1%) (Figure 3c middle panel, Supplementary Figure S12d). + +<|ref|>text<|/ref|><|det|>[[112, 360, 883, 585]]<|/det|> +The largest group of B. subtilis cells (comprising clusters 1, 2, 3, 5, and part of 7) accounted for \(\sim 45\%\) of the population and was characterized primarily by the upregulation of the dhb operon (logFCs \(>0.25\) , adjusted p- values \(< 2\mathrm{E - 05}\) ). Genes dhbA, dhbB, dhbC, dhbE, and dhbF encode the 5 enzymes implicated in the biosynthesis of bacillibactin - a catecholic siderophore which is produced and secreted in response to iron deprivation inside the cell. Relatedly, yusV, an ABC transporter of bacillibactin, was also found to be upregulated in these cells. Cells within cluster 5 are distinguished from the rest of the subpopulation by expression of genes related to arginine biosynthesis via ornithine. 10 of 14 genes associated with the cellular process were significantly upregulated in this cluster alone (gene set fold enrichment \(= 19.35\) , FDR \(= 3\mathrm{E - 08}\) ). A fluorescent promoter reporter for argC confirms heterogeneous expression of arginine genes with approximately \(2.1\%\) of cells in the high- argC expressing tail of the distribution - compared to \(7.2\%\) of cells in this state as determined by scRNA- seq (Figure 3c right panel, Supplementary Figure S12b, Methods, Supplementary Table S5). + +<|ref|>text<|/ref|><|det|>[[112, 600, 883, 756]]<|/det|> +Finally, to see if microfluidic probe- based methods are robust with different DNA- labeling chemistries, we repeated scRNA- seq on B. subtilis, replacing in- droplet PCR with a probe barcoding step utilizing a reverse transcriptase enzyme in accordance with the 10X Chromium Single Cell 3' standard protocol for eukaryotic scRNA- seq. While resolution decreased (median of 89 transcripts and 86 unique genes per cell), we were still able to resolve distinct biological states including sporulating and competent cell clusters (Supplementary Figure S3, Supplementary Table S6). Taken together, our platform for single cell sequencing of individual bacterium correctly recapitulates the known cellular states of B. subtilis at the expected population fractions and identifies previously unknown cellular states. + +<|ref|>sub_title<|/ref|><|det|>[[115, 774, 651, 792]]<|/det|> +## Gene expression heterogeneity in clonal populations of E. coli + +<|ref|>text<|/ref|><|det|>[[112, 808, 883, 894]]<|/det|> +With a validated method for identifying cellular states in Bacillus subtilis, we turned our attention to characterizing transcriptional heterogeneity in Escherichia coli. E. coli MG1655 was first grown in M9 minimal media, fixed at an \(\mathrm{OD}_{600} = 0.5\) (mid log), and subjected to probe- based scRNA- seq. We resolved 3,315 cells by unique single cell barcodes, detecting a median of 263 transcripts per cell representing a median of 165 distinct genes. Cells were partitioned into nine groups by graph + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 280]]<|/det|> +based clustering of their gene expression profiles (Figure 4ac, Supplementary Figure S4, Supplementary Table S6). The lack of clearly distinguishable transcriptomic signatures after dimensionality reduction using UMAP suggests a cell population that exhibits more uniform gene expression profiles compared with Bacillus subtilis. However, as with B. subtilis, we observe that certain clusters can be assigned to specific biological processes by DGE and that cluster determination is robust to the choice of clustering parameters (Methods). Cells in cluster 8 (141 cells, \(4.3\%\) of the population) exhibit significant upregulation of fiml, fimC, and fimA ( \(\log \mathrm{FC} > 2.4\) , adjusted p- value \(< 1\mathrm{E - 166}\) ). The downstream genes of the fim operon encode for components of type 1 pili and are known to be expressed heterogeneously in cells upon inversion of the DNA element containing the fimA promoter ("phase on" fim switch) \(^{21,22}\) . Type 1 pili are associated with biofilm formation and pathogenicity due to the adhesive properties of the fimbriae \(^{23}\) . + +<|ref|>text<|/ref|><|det|>[[112, 295, 883, 572]]<|/det|> +Relatedly, cells in cluster 1 (482 cells, \(14.5\%\) of population) uniquely upregulate genes implicated in cell motility including those encoding chemotaxis signaling proteins (cheA, cheW, tar; gene set fold enrichment \(>100\) , FDR \(= 3.1\mathrm{E - 02}\) ) and structural flagella components (fliC, flgL, flgG, flgD, flgE, flgC, flgF; gene set enrichment \(= 86.08\) , FDR \(= 2.6\mathrm{E - 06}\) ). This list includes genes regulated by class 1, 2, and 3 flagella promoters; three distinct regulons which control the sequential expression of flagella genes including the master regulator flhDC, components of the membrane- associated basal body and chemotaxis proteins, respectively \(^{44}\) . This model of assembly has been described as "just in time" gene expression that allows E. coli to discretize the energetically demanding biosynthesis process into checkpoint- controlled subprocesses. The fact that we observe upregulation of genes from multiple classes within the cluster, often within the same individual cells, indicates a more overlapping transcriptomic state, an observation that aligns with the fact that class 2 controlled fliA controls transcription of class 3 promoters upon completion of the basal body and export of class 2 flgM (fliA inhibitor) \(^{44}\) . To confirm this motile subpopulation of cells, we stained cells grown under the same conditions (Remel flagella stain) and visualized by phase microscopy. As shown in Figure 4d, we see assembled flagella emanating from a fraction of cells within the population. + +<|ref|>text<|/ref|><|det|>[[112, 587, 883, 779]]<|/det|> +Cells within clusters 5, 6, and 7 demonstrate differential usage of carbamoyl phosphate within the culture. Clusters 5 and 7 upregulate carAB, which encode the subunits of carbamoyl phosphate synthetase, converting bicarbonate to carbamoyl phosphate as the first step in the de novo biosynthesis of both uridine- 5'- monophosphate (UMP) and arginine. Both clusters, arranged adjacently on the UMAP, also upregulate pyrI, pyrB & pyrC, three enzymes involved in the first two steps of UMP biosynthesis, implying a commitment of the cells towards pyrimidine biosynthesis. Alternatively, cells in cluster 6 shuttle carbamoyl phosphate into the arginine biosynthesis pathway as implied by the upregulation of argF, argI, argB, argG, argD ( \(\log \mathrm{FCS} > 0.55\) , adjusted p- values \(< 1\mathrm{E - 20}\) ). GSEA reinforces this observation, finding a significant enrichment of genes implicated in arginine biosynthesis via ornithine in these cells (gene set fold enrichment \(= 26.77\) , FDR \(= 2.02\mathrm{E - 03}\) ). + +<|ref|>text<|/ref|><|det|>[[112, 794, 883, 900]]<|/det|> +Next, E. coli was grown in rich, LB media under anaerobic conditions. In the absence of exogenous electron acceptors permitting cellular respiration (e.g. O2, nitrite, nitrate, DMSO), E. coli performs "mixed acid" fermentation to sustain ATP generation by glycolysis, shuttling electrons into a variety of end- products (acetate, ethanol, succinate, formate) in order to replenish the pool of NAD+. The amount of each end- product produced by E. coli varies based on the redox state of the cell - an adaptability which allows E. coli to grow on a wide range of substrates under + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 331]]<|/det|> +anoxic conditions and varying \(\mathsf{pH}^{24}\) . We performed scRNA- seq to elucidate whether mixed acid fermentation can be, in part, attributed to the separation of branched fermentative pathways into distinct subpopulations of cells. Graph- based clustering of gene expression vectors ( \(n = 3,763\) resolved cells) resulted in 8 cell clusters (Figure 4b,d, Supplementary Figure S5, Supplementary Table S6). Within the population, we observe the most highly abundant transcripts (top \(5\%\) by aggregated UMIs) corresponding to genes involved in anaerobic ATP generation including glycolysis (fbaB, mdh, aceE, fbaA, pgi, lpd, eno, gapA, pgk, gene set fold enrichment \(= 9.5\) compared with a null model where the top \(5\%\) most abundant genes were chosen randomly, FDR \(= 2.0\mathrm{E} - 4\) ), anaerobic respiration (dcuA, acnA, hyaC, glpB, frdC, mdh, frdB, hyaA, hyaB, narH, glpD, glpA, narG, hybO, glpC, fold enrichment \(= 4.9\) , FDR 1.22- 04), and mixed acid fermentation (hyaC, frdC, mdh, frdB, hyaA, hyaB, gldA, fold enrichment \(= 8.3\) , FDR \(= 3.7\mathrm{E} - 03\) ) in addition to genes related to ribosome biogenesis, transcription, and translation. Enrichment of these pathways within the most highly expressed genes at the population level is consistent with an anaerobic environment. + +<|ref|>text<|/ref|><|det|>[[112, 345, 883, 797]]<|/det|> +We then analyzed the transcriptional states of cells in each cluster. Clusters 1 and 4 appear to represent a subpopulation of cells responding to over- acidification of the cytoplasm. Both groups significantly upregulate the gad operon (gadABC) as well as dps (logFCs \(> 0.34\) , adjusted p- values \(< 1\mathrm{E} - 32\) ). In addition, acid stress chaperone proteins hdeA and hdeB are overexpressed in cluster 4 (logFCs \(> 0.25\) , adjusted p- value \(< 1\mathrm{E} - 02\) ). Acid stress is an expected consequence of fermentation as acidic fermentation byproducts (e.g. acetate, lactate, formate) accumulate. Cells in cluster 5 (283 cells, \(7.5\%\) of total) had 157 genes significantly upregulated (adjusted p- values \(< 0.05\) ) compared with the rest of the population. Surprisingly, these genes belong to pathways involved in aerobic respiration (aceAB, sucABCD, nuoBMN, cyoABC, cydAB, sdhABC, fumA, gltA, acnB, icd, mdh; gene set fold enrichment \(= 11\) , FDR \(= 2.32\mathrm{E} - 13\) ), ATP synthesis coupled proton transport (atpABCDEHG; gene set fold enrichment \(= 24.47\) , FDR \(= 1.45\mathrm{E} - 05\) ), glycolysis (mdh, gpmA, aceE, fbaA, lpd, aceF, gapA; gene set fold enrichment \(= 10.24\) , FDR \(= 7.33\mathrm{E} - 04\) ), and gluconeogenesis (gpmA, fbaA, sdaB, pck, pps; gene set fold enrichment \(= 9.92\) , FDR \(= 1.03\mathrm{E} - 02\) ). The overexpression of these genes, many of which are involved in the TCA cycle (Figure 4g), may be an indication of oxygen contamination within the culture, resulting in a stark difference in transcriptional signatures and, consequently, cell behavior within the isogenic population. In addition, cells in cluster 5 also appear to be motile as genes pertaining to flagella structure (flgCDEFG; fold enrichment \(= 9.26\) , FDR \(= 1.28\mathrm{E} - 02\) ) and locomotion (fliCD, tar, mgIB, cheAW, flgK; fold enrichment \(= 5.16\) , FDR \(= 4.08\mathrm{E} - 04\) ) were upregulated & overrepresented in the subpopulation. Heterogeneous expression of flagella was confirmed microscopically using a remel stain (Figure 4f). Since cultures were maintained in capped tubes outside of an anaerobic chamber, is it possible air was introduced into the culture through an imperfect seal. Thus, we hypothesize that this induced cell population activated motility genes to gather at the liquid/air interface. Consistent with the idea of such an oxygen- gradient organization, a recent report used a high- throughput fluorescent in situ hybridization (FISH) analysis of approximately 100 genes to demonstrate the role of oxygen gradients in spatial partitioning within biofilms25. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 87, 794, 900]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 115, 883, 336]]<|/det|> +a. Heatmap of marker gene expression (z-score of log-transformed values) from 3,315 individual E. coli cells grown aerobically in minimal M9 media organized into 10 clusters. b. Heatmap of gene expression from 3,763 individual E. coli cells grown in anaerobic LB media organized into 9 clusters. c. UMAP 2-dimensional representation of the 10 cell clusters from aerobic M9 culture conditions. d. Flagellar components are heterogeneously expressed in aerobic M9 media - a key flagellar gene (flagellin - fliC) is preferentially expressed by cells in cluster 1. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. e. UMAP 2-dimensional representation of the 9 cell clusters from E coli fermenting LB media culture conditions. f. Flagellar components are heterogeneously expressed in fermentation conditions - a key flagellar gene (flagellar basal-body rod protein - flgC) is preferentially expressed by cells in cluster 5. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. g. Many aerobic respiration genes and other metabolic and flagellar genes are overexpressed in cluster 5. Volcano plots of genes over-expressed in at least 25% of cells in cluster 5 from these processes are highlighted in colors as labeled. Green labeled genes, involved in central-carbon metabolism, are also highlighted as green arrows in the metabolic map. + +<|ref|>text<|/ref|><|det|>[[112, 346, 883, 587]]<|/det|> +Finally, cluster 8 (111 cells, \(3\%\) of total), which projects between the majority of cells and cluster 5 on the UMAP, is characterized by the strong upregulation of the tdc operon (tdcBCEG; logFC \(>3.0\) , adjusted p- values \(< 1\mathrm{E - 100}\) ). Under anaerobic conditions, if energy is low inside the cell, E. coli generates ATP by conversion of L- threonine to propionate. The tdc operon has been shown to be strongly induced by anaerobic conditions as well as catabolically repressed (CRP- controlled), suggesting these cells are sensing an anoxic environment while also deprived of glucose26. The tdc operon encodes proteins for the transport and degradation of L- threonine (and L- serine by analogous intermediates) for the purpose of ATP generation (1:1 conversion). While the operon is induced in the absence of oxygen, it has been shown to not be directly regulated by common oxygen- sensing systems inside the cell such as transcriptional regulators FNR and ArcA but rather by tcdA and tcdR (transcriptional activators)27. To the best of our knowledge, this is the first time threonine metabolism has been shown to be cellularly segregated in a microbial culture. Taken together, these findings demonstrate how single cell transcriptomics can inform on physiological and metabolic organization in complex microbial environments. + +<|ref|>sub_title<|/ref|><|det|>[[115, 621, 213, 637]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 654, 883, 895]]<|/det|> +To agnostically characterize the collection of cell states within a microbial population, we developed an affordable (standard commercial 10X run costs plus \(\equiv 1c\) per cell in probes) bacterial scRNA- seq technique that demonstrates high throughput and sensitivity. Our method, which is based on combining microfluidic droplet partitioning of single cells with in situ hybridization of DNA probes, overcomes several difficulties associated with obtaining quality single cell mRNA signal from bacteria. Using bacterial scRNA- seq, we identify known cellular states including genetic competence and sporulation in B. subtilis and fimbriae production in E coli. In addition to these well characterized physiologies, we uncovered several previously unknown or sparsely studied heterogeneously expressed gene sets including metabolic pathways (amino- acid metabolism, carbon metabolism, siderophores) and physiological states (chemotaxis and motility). Interestingly, arginine biosynthesis genes appear to be heterogeneously expressed in both E coli and B subtilis in minimal media. Arginine synthesis is also isolated spatially (at the organ and cellular location level) in multicellular, eukaryotic organisms28. This conserved partitioning of a biochemical process suggests that some processes may be inherently more + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 882, 142]]<|/det|> +prone to segregate. We speculate that such segregation may be a result of incompatibilities between certain biochemical reactions and may have played a role in the initial development of multicellular structure. + +<|ref|>text<|/ref|><|det|>[[112, 156, 882, 398]]<|/det|> +Along with the method we describe in this work, two other groups have separately introduced a high- throughput bacterial scRNA- seq approach that is based on combinatorial indexing29,30. A commonality between these studies and ours is the discovery of subpopulations of cells expressing genes for specialized function, including genetic competence. While the different techniques uncover common biology, each has advantages and disadvantages. Our approach requires an upfront investment in probe generation and prior understanding of the genome, and is therefore not intended to be used with novel or environmental organisms that are not well characterized. Conversely, once a probe set is synthesized, our tool is fast, cheap, and provides very deep coverage of transcripts as well as a more accurate method of quantification by UMI. In addition, the technique relies on commercial equipment that is now ubiquitous in laboratories and core facilities around the world. Critically, unlike the combinatorial indexing system, our in- situ hybridization technique ensures that sequencing is not wasted on ribosomal or other non- mRNA transcripts, which account for between 93- 97% of all transcripts that are sequenced by combinatorial indexing (ref). + +<|ref|>text<|/ref|><|det|>[[112, 412, 882, 567]]<|/det|> +Our technique of combining in- situ hybridization with microfluidic single cell partitioning also invites several potential uses for eukaryotic samples which often rely on targeting the 3' poly- A tail; an approach that provides only a single site for transcript capture, excludes sequences lacking a poly- A, and provide limited information on splice variants. In addition, hybridization techniques are compatible with formalin- fixed paraffin- embedded tissues, the most common form of clinical sample, which are difficult to analyze using existing high- throughput single cell techniques. Finally, the extension of these techniques to mammalian and other eukaryotic cells could potentially be used to simultaneously identify the gene expression profile of an intracellular bacteria and its eukaryotic host cell. + +<|ref|>text<|/ref|><|det|>[[112, 582, 882, 754]]<|/det|> +In the context of microbial analyses these new tools will enable investigation of several recently observed phenomena in bacterial cultures. In particular, we foresee their use in characterizing bacterial subpopulations that are differentially resistant to antibiotics (antibiotic persistence), and the physiological role of bacterial cells that differentially express toxin and virulence genes in bacterial infections. In eukaryotic systems, where mature scRNA- seq tools have existed for several years, computational methods extend beyond surveys of the different cell types to provide information about the dynamics of cellular populations such as pseudotime arrangements of cells31, lineage and causality maps and cell- type specific QTLs32. In bacterial communities we predict that these bioinformatic approaches will be used to understand dynamic processes like phage- bacteria interactions, the organization of biofilms, and lab- based evolution of traits. + +<|ref|>sub_title<|/ref|><|det|>[[115, 771, 285, 787]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[112, 802, 882, 883]]<|/det|> +DS was funded in part by the Natural Sciences and Engineering Research Council of Canada (NSERC PGSD2- 517131- 2018). DS and SH acknowledge funding from NIH NIGMS R00GM118910, NIH NHLBI R01HL158269, U19 Systems Immunology Pilot Project Grant at Harvard University, and the Harvard University William F. Milton Fund. AZR and RAM were funded by a Dupont Science and Innovation award. Portions of this research were conducted on + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 883, 123]]<|/det|> +the O2 High Performance Compute Cluster, supported by the Research Computing Group, at Harvard Medical School. See http://rc.hms.harvard.edu for more information. + +<|ref|>sub_title<|/ref|><|det|>[[115, 234, 192, 250]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 267, 361, 283]]<|/det|> +## Strains and Growth Conditions + +<|ref|>text<|/ref|><|det|>[[115, 285, 883, 412]]<|/det|> +Bacillus subtilis str. 168 was used for all single cell Bacillus experiments. For experiments on suspension cultures in glucose/malate, cells were grown at \(37^{\circ}C\) with vigorous shaking in M9 media supplemented with CaCl2 (0.1 mM), \(0.2\%\) glucose, tryptophan, and trace metal mix 8. Trace metal solution was made as a 100X concentrate using Na- EDTA 5.2 g, \(\mathrm{FeSO_4 - 7H_2O}\) 2100.0 mg, \(\mathrm{H_3BO_3}\) 330.0 mg, \(\mathrm{MnCl_2 - 4H_2O}\) 100.0 mg, \(\mathrm{CoCl_2 - 6H_2O}\) 190.0 mg, \(\mathrm{NiCl_2 - 6H_2O}\) 24.0 mg, \(\mathrm{CuCl_2 - 2H_2O}\) 2.0 mg, \(\mathrm{ZnSO_4 - 7H_2O}\) 144.0 mg, \(\mathrm{Na_2MoO_4 - 2H_2O}\) 1200.0 mg, DI Water, 999.0 ml. The solution pH was adjusted to 7.0. At OD of 0.4- 0.5 cultures were supplemented with \(50~\mathrm{mM}\) malate. + +<|ref|>text<|/ref|><|det|>[[115, 428, 882, 558]]<|/det|> +Bacillus subtilis strain PY79 was used for all fluorescent promoter- reporter strains. comG reporter strain was previously described 8. Promoter reporters for cotY and argC were made as previously described 8. Briefly, YFP promoter reporters were cloned into the ECE174 backbone plasmid which uses sacA integration site and encodes chloramphenicol resistance (R. Middleton, obtained from the Bacillus Genetic Stock Center). Strains were made by genomic integration into the genome. Fluorescent reporters were integrated into the sacA site and checked by sequencing. Reporter strain detail is found in Supplementary Table 4. Reporter strains were grown in the same media and growth conditions used in single cell experiments. + +<|ref|>text<|/ref|><|det|>[[114, 571, 882, 717]]<|/det|> +E. coli MG1655 was used in all E coli single cell experiments. For experiments in minimal media cells were grown overnight in M9 minimal media and incubated at \(37^{\circ}C\) with moderate shaking (200 rpm). Overnight cultures were diluted and subcultured and a fixed specimen was collected at mid exponential phase (OD 0.3-0.5). For experiments in anaerobic LB media cells were inoculated into degassed LB that was aliquoted inside a COY anaerobic chamber containing (95% N2 5% H2). Overnight culture was subcultured into a degassed media tube using a needle and syringe purged in N2 gas, and the samples were incubated anaerobically inside an anaerobic box with a gas-pack (manufacturer) at \(37^{\circ}C\) . Cells were fixed at OD 0.3-0.5 by addition of formaldehyde using a N2-purged syringe and needle directly into the tubes. + +<|ref|>sub_title<|/ref|><|det|>[[115, 734, 253, 750]]<|/det|> +## Probe set design + +<|ref|>text<|/ref|><|det|>[[115, 752, 882, 822]]<|/det|> +Probes were designed with an mRNA complementary region approximately 50 bp in length flanked by a PCR handle (18- 23 bp) towards the 5' end and a 30 poly(dA) tract on the 3' end. On the ends of each probes we included a region allowing for circularization and cutting with hindIII as outlined by Schmidt et al 28. Probe sequences are included in Supplementary Tables 1 and 2. + +<|ref|>sub_title<|/ref|><|det|>[[115, 846, 270, 861]]<|/det|> +## Probe amplification + +<|ref|>text<|/ref|><|det|>[[115, 864, 882, 899]]<|/det|> +Probes were amplified using a rolling circle method similar to the one used by Schmidt et al 28 with slight modifications. To amplify our probes, we did not use nicking enzymes but instead only + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 163]]<|/det|> +used the HindIII digestion site in the rolling circle scheme \(^{28}\) . Our incubations were scaled by a factor of 5 from the detailed protocol provided by Schmidt et al. Otherwise, all other aspects of the protocol were as previously detailed. Amplified probes were eluted and purified by PAGE electrophoresis using \(15\%\) urea gels and ethanol precipitation as described in Sambrook et al \(^{59}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 181, 314, 198]]<|/det|> +## UMI and Poly-A Addition + +<|ref|>text<|/ref|><|det|>[[112, 199, 883, 328]]<|/det|> +Amplified "proto- probes" were extended to include a unique molecular index (UMI) and 3' poly- adenine tail by isothermal extension with a 3' blocked primer containing the reverse complementary sequences. 100 ul of probes and 1 mg blocked extension primer (Supplementary Table 3) were mixed with 100 units of klenow fragment and 1x NEB buffer 2 and incubated for 30 min at room temperature. The extended oligonucleotide library was selectively purified by PAGE electrophoresis (final probe length approximately 142 bp) using \(15\%\) urea gels followed by ethanol precipitation as described in Sambrook et al. \(^{59}\) + +<|ref|>sub_title<|/ref|><|det|>[[115, 346, 451, 363]]<|/det|> +## Fixation and in situ hybridization reactions + +<|ref|>text<|/ref|><|det|>[[112, 364, 883, 619]]<|/det|> +Cells from one intact colony or 2 mL of cell culture were fixed using a 30 minute incubation in \(1\%\) paraformaldehyde (final concentration) at room temperature. Formaldehyde- fixed samples were washed with \(0.02\%\) SSC by gentle centrifugation ( \(6,000 \times \mathrm{G}\) ) for 2.5 minutes. After the wash, cell pellets were resuspended in 1 mL MAAM (4:1 V:V dilution of methanol to acetic acid). Samples were kept at \(- 20^{\circ}\mathrm{C}\) for up to 2 days before further processing. For in- situ probe hybridization, 150ul of fixed sample was centrifuged and washed once in 1x PBS to remove methanol and acetic acid. After the wash step cells were incubated in 200 ul PBS with 350u/ul of lysozyme solution (Epicentre ready- lyse, Epicentre Madison WI USA) for 30 minutes at room temperature. After 30 minutes cells were pelleted and washed with 500 ul PBS- tween (1 x PBS with 0.1 % Tween 20). Cells were then re- suspended with 100 ul of probe binding buffer consisting of 5x SSC, \(30\%\) formamide, 9mM citric acid (pH 6.0). \(0.1\%\) Tween 20, 50 ug.ml heparin, and \(10\%\) low MW dextran sulfate) \(^{60}\) . Cell suspensions were placed in a \(50^{\circ}\mathrm{C}\) shaker- incubator and allowed to pre- equilibrate for 1 hour. After an hour 50 ul of probes, (600 ng/ul) were added to each cell suspension, and samples were left to incubate overnight. + +<|ref|>text<|/ref|><|det|>[[112, 620, 883, 655]]<|/det|> +The next day samples were washed 5 times in pre- warmed ( \(50^{\circ}\mathrm{C}\) ) probe- wash solution (5 x SSC, \(30\%\) formamide, 9 mM citric acid (pH 6.0), \(0.1\%\) Tween 20, and 50 ug/mL heparin). + +<|ref|>text<|/ref|><|det|>[[112, 656, 883, 691]]<|/det|> +Before flowing on the 10X device cells were washed 3 times in 1x PBS and diluted as suggested in the 10X Chromium instruction manual. + +<|ref|>sub_title<|/ref|><|det|>[[115, 708, 577, 725]]<|/det|> +## Microfluidic encapsulation and droplet generation reaction + +<|ref|>text<|/ref|><|det|>[[112, 726, 883, 875]]<|/det|> +Single cell partitioning, barcoding, and cDNA library generation was achieved using 10X Genomics's Chromium Controller with the Chromium Single Cell 3' Reagents Kit (v2 chemistry) as described by 10X Genomics (https://support.10xgenomics.com/single_cell- gene- expression/index/doc/user- guide- chromium- single_cell- 3- reagent- kits- user- guide- v2- chemistry). The protocol was modified to achieve bacterial scRNA- seq. During GEM generation, a master mix containing the following reagents (per rxn, not accounting for excess volume) was prepared: 33 uL of 4X ddPCR Multiplex Supermix (BioRad), 4 uL of custom primer (10 uM), 2.4 uL additive A (10X Genomics), 26.8 uL dH2O. All other reagents specified by 10X Genomics were omitted. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 127]]<|/det|> +Prepared cell samples were washed 3X in PBS and diluted to 1000 cells/uL before loading on the microfluidic chip ("Chip A Single Cell") with a targeted cell recovery of 10,000 cells. + +<|ref|>text<|/ref|><|det|>[[115, 144, 389, 161]]<|/det|> +Library construction & sequencing + +<|ref|>text<|/ref|><|det|>[[115, 177, 883, 269]]<|/det|> +After running the Chromium Controller, each sample (i.e. "reaction") was visually inspected to confirm successful GEM formation (should observe a well distributed emulsion with a volume of approximately 100 uL). Samples were then transferred to fresh PCR tubes and cycled at the following conditions (replacing the "GEM- RT Incubation Step" in 10X Genomics protocol): \(94^{\circ}C\) for 5 minutes, 6 cycles of (94°C for 30s followed by 50°C for 30s then 65°C for 30s), hold at 4°C. + +<|ref|>text<|/ref|><|det|>[[113, 284, 883, 486]]<|/det|> +After PCR #1, the emulsion was broken and the pooled DNA purified using Dynabeads MyOne Silane as described in 10X Genomics' protocol. Purified DNA was amplified once more (replacing the "cDNA Amplification step" of 10X Genomics' protocol) using a master mix composed of 17 uL of purified DNA, 20 uL of Q5 Hot Start 2X MM, 1.5 uL of forward primer (10 uM) and 1.5 uL of reverse primer (10 uM). PCR conditions included a 30s incubation at 98C followed by 16 cycles of 10s at 98C, 20s at 62C, and 20s at 72C before a final extension at 72C for 2 min. After PCR #2, amplified DNA was purified using the NucleoSpin® Gel and PCR Clean- up kit (Macherey- Nagel) as per manufacturer's instructions. Purified DNA was run on Agilent TapeStation to confirm the presence of a band at 180 bp, indicating successful library generation. Libraries then prepared for sequencing by Illumina adapter addition via low cycle (n = 6) PCR with custom library preparation finishing primers (see Supplementary Table 3). + +<|ref|>text<|/ref|><|det|>[[115, 503, 883, 559]]<|/det|> +Sequencing was done on an illumina nextSeq- 1000 instrument with 100 cycle reagents. Libraries were spiked with 30% phi- X and sequenced for 8bp in the i7 index direction and 119 bp for "Read 1". Addition of 30% phiX improved fastQ quality scores. + +<|ref|>text<|/ref|><|det|>[[115, 576, 209, 594]]<|/det|> +Microscopy + +<|ref|>text<|/ref|><|det|>[[115, 609, 883, 702]]<|/det|> +Cultures were visualized on a Leica DM3000 light microscope before single cell experiments to ensure samples were free of chains and clumps. Reporter strains were imaged using a 100X oil- immersion objective on a Zeiss Axio Observer inverted microscope equipped with a colibri- 7 LED fluorescent light source and an axiocam digital camera. Flagellar staining was done using the Remel flagellar stain (Thermo- Fisher Scientific) as per manufacturer instructions. + +<|ref|>text<|/ref|><|det|>[[115, 718, 625, 736]]<|/det|> +Image analysis using cell segmentation and outlier identification + +<|ref|>text<|/ref|><|det|>[[115, 752, 883, 860]]<|/det|> +Cells in microscopy images were segmented using the microbeJ program within FIJI (ImageJ). Outliers were identified by using the IQR outlier method. Specifically, quartiles and the interquartile range (IQR) were determined for each dataset. Outliers were determined by standard outlier detection parameters: cells that were more than 1.5 IQRs above Quartile 3 were designated as cells within an overexpressing population. Population histograms were plotted and outliers reported. + +<|ref|>text<|/ref|><|det|>[[115, 880, 395, 898]]<|/det|> +Constructing a Reference Genome + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 413]]<|/det|> +Here we describe how we constructed reference genomes tailored to the probesets used for each organism (see Methods Section Probeset Design). For a given organism, let \(P\) be the number of probes in the probeset, \(\tilde{s}_{dig} = AAGCTT\) be the \(L_{dig} = 6\) base pair (bp) nucleotide sequence of the HindIII digestion site (see Supplementary Figure S1a), \(\tilde{s}_{ext} = TGCTCGAAGAAACTATG\) be the \(L_{ext} = 17\) bp nucleotide sequence of the extender, \(\tilde{s}_{PCR} = TGTTAGGACGGGTCATC\) be the \(L_{PCR} = 18\) bp nucleotide sequence of the PCR handle, and \(\tilde{p}_{j}\) be the \(L_{j}\) bp nucleotide sequence of the \(j^{\text{th}}\) probe. Let \(\tilde{g}_{j} = [\tilde{s}_{dig}, \tilde{s}_{ext}, \tilde{p}_{j}, \tilde{s}_{PCR}]\) be the \(L_{dig} + L_{ext} + L_{j} + L_{PCR}\) bp nucleotide sequence formed by flanking the \(j^{\text{th}}\) probe by the HindIII digestion site, extender sequence, and PCR handle as shown in Figure 1a. We first constructed a FASTA file by concatenating all \(P\) of the \(\tilde{g}_{j}\) sequences. Next, we constructed a GTF file to index the FASTA. Typically, the GTF indexes genes in the FASTA by recording the location of the starting and ending bps of each gene. Since we have multiple probes per gene, and our FASTA is a concatenation of these probes, we therefore use the GTF to index probes instead. Specifically, we recorded that for the \(j^{\text{th}}\) probe, the starting and ending bps of \(\tilde{g}_{j}\) in the FASTA are \((\sum_{i = 1}^{j - 1} L_{dig} + L_{ext} + L_{i} + L_{PCR}) + 1\) and \((\sum_{i = 1}^{j} L_{dig} + L_{ext} + L_{i} + L_{PCR})\) respectively. We use \(\tilde{g}_{j}\) as the definition of a probe instead of \(\tilde{p}_{j}\) for the purpose of alignment because sequenced reads should contain the flanking sequences anyways so including them minimizes spurious alignment of off- target reads. Lastly, we constructed a reference genome by using the mkref command in CellRanger (v3.1.0), supplying the generated FASTA and GTF files to the --fasta and --genes arguments respectively. + +<|ref|>sub_title<|/ref|><|det|>[[115, 425, 415, 443]]<|/det|> +## Creating FASTQ Files for CellRanger + +<|ref|>text<|/ref|><|det|>[[112, 441, 883, 585]]<|/det|> +Each read in the Illumina- sequenced FASTQ file contains the 10x CB, the 10x UMI, the Probe UMI as well as the mRNA sequence on a single line. This file therefore needs to be re- formatted to be compatible with CellRanger, so that the CB and UMI is stored in the R1 file and the mRNA sequence is stored in the R2 file. We also perform filtering at this stage so that only reads containing all the information needed by the CellRanger pipeline are retained. First, we discard reads whose length is less than or equal to 110 bp. Next, we aligned each read to \(\tilde{s}_{ext}\) using the nwalign(global=true) function in MATLAB (2019a), and discarded reads where the alignment score was \(< 17\) . We search for the extender sequence specifically, because for all probe designs the Extender sequence flanks the Probe UMI sequence (see Supplementary Figure S1c). + +<|ref|>text<|/ref|><|det|>[[112, 584, 883, 794]]<|/det|> +We create two sets of FASTQ files using reads retained after these filtering steps. In the first set of FASTQ files, we record the 10x CB and Probe UMI in R1, with the Probe UMI obtained by extracting 12 bps in either the 5' or 3' direction from the Extender sequence according to the probe design (see Supplementary Figure S1c). In the second set of FASTQ files, we record the 10x CB and 10x UMI in R1. For both sets of FASTQs, we trimmed all nucleotides 5' (3') of the Extender sequence if the Extender is 5' (3') of the binding region, and record the result in R2. The two sets of FASTQs therefore have the same number of lines, identical R2 files and the same 10x CBs in the R1 files, and only differ in the UMI recorded in the R1 files. We obtained single cell count matrices from these reformatted FASTQs by using the count function in CellRanger (v3.1.0) and supplying the custom made genome (described in Methods Section Constructing a Reference Genome) as the --transcriptome argument. Since CellRanger discards reads if QC scores for UMIs are too low, the same set of reads may not necessarily end up being used in constituting single- cell count matrices from the two sets of FASTQ files. + +<|ref|>sub_title<|/ref|><|det|>[[115, 809, 810, 827]]<|/det|> +## Comparing Single cell Probe Expression Matrices Generated Using 10x vs Probe UMIs + +<|ref|>text<|/ref|><|det|>[[112, 826, 883, 896]]<|/det|> +The output from CellRanger is a single cell matrix where each row, \(i\) , corresponds to one of \(N\) cells, and each column \(j\) corresponds to one of \(P\) probes. When the supplied FASTQ files have Probe (10x) UMIs recorded in R1, each entry in the single cell matrix, \(c_{i,j}^{P}\) ( \(c_{i,j}^{P}\) ) is the number of Probe (10x) UMIs in cell \(i\) corresponding to probe \(j\) . The Probe UMI remains fixed over multiple + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 883, 206]]<|/det|> +rounds of PCR and can therefore be treated as a classical UMI, while in our protocol, new 10x UMIs may be introduced on the same transcript at each PCR round. Therefore, \(c_{i,j}^{p'}\) may not accurately record the number of transcripts. We used \(c_{i,j}^{p'}\) for all results presented in the main text of the paper. Since it may be more convenient to use 10x UMIs directly, we also quantified the extent to which \(c_{i,j}^{p'}\) differs from \(c_{i,j}^{p'}\) by computing a joint distribution of count frequencies, \(h_{k,l} = \sum_{i}\sum_{j}I_{k}\left(c_{i,j}^{p'}\right)I_{l}\left(c_{i,j}^{p'}\right)\) , where \(I_{s}(t) = \delta_{s,t}\) (see histograms in Supplementary Figures S6- S8). + +<|ref|>text<|/ref|><|det|>[[112, 206, 883, 343]]<|/det|> +On the one hand, when \(h_{k,l} > 0\) , for \(l > k\) , there is over- counting; the sequencing depth is large enough to detect spurious 10x UMIs introduced in latter PCR rounds. On the other hand, when \(h_{k,l} > 0\) , for \(l < k\) , there is under- counting; CellRanger detects that the same 10x UMI is coincidentally found on different probes, and discards all such reads. Since under- counting doesn't introduce spurious transcript counts and can be treated in the same vein as low sequencing depth or low capture efficiency, we do not consider it a confounder. The accuracy of a protocol relying only on 10x UMIs can therefore be quantified by the fraction of 10x UMIs corresponding to entries in \(h_{k,l}\) with \(l \leq k\) : + +<|ref|>equation<|/ref|><|det|>[[396, 355, 600, 395]]<|/det|> +\[Accuracy = \frac{\sum_{k}\sum_{l\leq k}l h_{k,l}}{\sum_{k}\sum_{l}l h_{k,l}}\] + +<|ref|>text<|/ref|><|det|>[[112, 405, 883, 666]]<|/det|> +For the B. Subtilis sample, the accuracy of using 10x UMIs thus computed was only \(41\%\) . However, when we computed gene expression matrices using \(c_{i,j}^{p'}\) (see Methods Section Single cell Gene Expression Matrices from Single cell Probe Expression Matrices below) and performed differential gene expression (see Methods Section Transcriptomic Analysis and Visualization below), we were still able to detect clusters corresponding to competence and sporulation (see e.g. clusters #2 and #11 respectively in Supplementary Figure S6 and Supplementary Table 7). Likewise, for the E. coli sample grown in minimal media, the accuracy was only \(33\%\) , but we were still able to detect clusters corresponding to the fim operon, cell motility and differential usage of carbamoyl phosphate (see e.g. clusters #8, #2 and #7 respectively in Supplementary Figure S7 and Supplementary Table 7). Lastly, for the E. coli sample grown in LB media, the accuracy was \(88\%\) . Here too, we were able to recapitulate biological results obtained using Probe UMIs including clusters with different metabolic programming and upregulation of the tdc operon (see clusters #6 and #9 respectively in Supplementary Figure S8 and Supplementary Table 7). Taken together, this suggests that users of our technology may be able to use 10x UMIs directly i.e. \(c_{i,j}^{p'}\) . However, for the remainder of this study we restrict our attention to \(c_{i,j}^{p'}\) . + +<|ref|>sub_title<|/ref|><|det|>[[112, 679, 764, 698]]<|/det|> +## Single cell Gene Expression Matrices from Single cell Probe Expression Matrices + +<|ref|>text<|/ref|><|det|>[[112, 696, 883, 864]]<|/det|> +In order to perform standard single cell analysis such as cell- similarity clustering and differential gene expression, we need to determine \(c_{i,k}^{g}\) , the number of transcripts of gene \(k\) in cell \(i\) , whereas the output from CellRanger is \(c_{i,j}^{p'}\) . Let \(G_{k}\) be the indices of probes in the probeset corresponding to gene \(k\) . Since multiple probes may bind to the same transcript, setting \(c_{i,k}^{g} = \sum_{j \in G_{k}} c_{i,j}^{p'}\) may result in an over- estimate of transcript counts. To be conservative while also retaining as many UMIs as possible, we therefore picked \(c_{i,k}^{g} = \max_{j \in G_{k}} c_{i,j}^{p'}\) which lower bounds gene expression. This choice resulted in \(66\%\) , \(79\%\) and \(81\%\) of the UMIs in \(c^{p}\) being retained in \(c^{g}\) , accounting for \(65\%\) , \(77\%\) and \(81\%\) of the total reads, for the B. Subtilis, E. Coli in minimal media and E. Coli in LB media samples respectively. + +<|ref|>text<|/ref|><|det|>[[112, 863, 883, 896]]<|/det|> +That different probes may be selected to represent the same gene in different cells is not material, since probes are selected proportional to their binding affinity. However, this choice introduces a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 265]]<|/det|> +potential bias since the gene expression tabulated in \(c^g\) will tend to be higher for genes with more corresponding probes in the probeset, i.e. larger \(|G_k|\) . As a sanity check, we also tabulated single cell gene expression matrices using \(c_{i,k}^g = c_{i,k}^p, k^* = \underset {j\in G_k}{\operatorname{argmedian}}\sum_i c_{i,j}^p\) i.e. using a fixed probe across all cells for each gene corresponding to the probe with the median bulk expression. This choice only resulted in \(10\%\) , \(18\%\) and \(17\%\) of the UMLs in \(c^p\) being retained in \(c^g\) , for the \(B\) . Subtilis, \(E\) . Coli in minimal media and \(E\) . Coli in LB media samples respectively. Despite the decreased transcriptional resolution, we are able to still detect clusters corresponding to the results presented in the main text (see Supplementary Figures S9- S11 and Supplementary Table 8). We therefore defer investigation of more sophisticated methods of combining probe counts to form a single cell gene expression matrix to subsequent work. + +<|ref|>sub_title<|/ref|><|det|>[[115, 277, 448, 294]]<|/det|> +## Transcriptomic Analysis and Visualization + +<|ref|>text<|/ref|><|det|>[[112, 294, 883, 550]]<|/det|> +Here we describe how we analyzed single cell gene expression matrices, \(c^g\) , using Seurat (v3.1.2 with default parameters except where indicated). We first log- transformed the data using the NormalizeData(normalization.method = "LogNormalize", scale_factor = 10000) function, and selected the 2000 most variable genes using FindVariableFeatures(selection.method = "vst", nfeatures = 2000). Then, we z- scored these highly- variable genes using ScaleData. Next, we performed linear dimensionality reduction using PCA down to 50 dimensions (RunPCA). Points in this embedding were used to construct UMAP plots (RunUMAP(dims=1:10)) and find neighbors for clustering (FindNeighbors). FindClusters(resolution=1.0) was used to run the Louvain clustering algorithm and generate clusters. We confirmed that the clusters highlighted in the main text appeared consistently for a range of resolutions from 0.5 to 1.5. Summary statistics of the number of genes/transcripts in a cell and cluster populations for the different samples can be found in Supplementary Table 5. Differential gene expression was performed using the FindMarkers(min.pct=0.25) function with a log- fold change cutoff of 0.25, whose output when applied to our data is shown in Supplementary Tables 6- 8. Heatmaps of differentially expressed genes were generated using the DoHeatMap function, and show the top over- expressed genes in each cluster with Bonferroni- corrected p- value \(< 0.05\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 564, 329, 581]]<|/det|> +## Data and Code Availability + +<|ref|>text<|/ref|><|det|>[[113, 590, 882, 678]]<|/det|> +All data used in this paper will be made available on the NCBI GEO upon publication. Code to prepare custom references, reformat FASTQ files for the CellRanger pipeline and generate single- cell gene expression matrices from single- cell probe expression matrices is available at https://gitlab.com/hormozlab/bacteria_scrnaseq. All code and instructions to reproduce numerics and figures in the manuscript will be made available upon publication. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 318, 108]]<|/det|> +## Supplementary Figures + +<|ref|>text<|/ref|><|det|>[[112, 116, 884, 153]]<|/det|> +Figure S1: Detailed schematic of C2C- based proto- probe amplification and primer extension incorporation of probe- based UMI and polyA + +<|ref|>text<|/ref|><|det|>[[112, 161, 800, 180]]<|/det|> +Figure S2: Heatmap of B subtilis cells from Figure 3 including all marker gene names + +<|ref|>text<|/ref|><|det|>[[112, 188, 884, 223]]<|/det|> +Figure S3: scRNA- seq of B subtilis using in- droplet reverse transcription (RT) instead of in- droplet PCR identifies sporulation and competence populations + +<|ref|>text<|/ref|><|det|>[[112, 232, 884, 267]]<|/det|> +Figure S4: Heatmap of aerobic M9 culture of E coli cells from Figure 4 including all marker gene names + +<|ref|>text<|/ref|><|det|>[[112, 276, 884, 311]]<|/det|> +Figure S5: Heatmap of E coli fermenting in LB media from Figure 4 including all marker gene names + +<|ref|>text<|/ref|><|det|>[[112, 320, 711, 339]]<|/det|> +Figure S6: scRNA- seq of B subtilis using 10x UMI instead of Probe UMI + +<|ref|>text<|/ref|><|det|>[[112, 347, 884, 367]]<|/det|> +Figure S7: scRNA- seq of E coli cells in aerobic M9 culture using 10x UMI instead of Probe UMI + +<|ref|>text<|/ref|><|det|>[[112, 375, 825, 394]]<|/det|> +Figure S8: scRNA- seq of E coli cells in LB media using 10x UMI instead of Probe UMI + +<|ref|>text<|/ref|><|det|>[[112, 402, 884, 438]]<|/det|> +Figure S9: scRNA- seq of B subtilis using bulk median instead of per- cell maximum probe counts for gene expression + +<|ref|>text<|/ref|><|det|>[[112, 446, 884, 482]]<|/det|> +Figure S10: scRNA- seq of E coli cells in aerobic M9 culture using bulk median instead of per- cell maximum probe counts for gene expression + +<|ref|>text<|/ref|><|det|>[[112, 490, 884, 526]]<|/det|> +Figure S11: scRNA- seq of E coli cells in LB media using bulk median instead of per- cell maximum probe counts for gene expression + +<|ref|>text<|/ref|><|det|>[[112, 543, 876, 562]]<|/det|> +Figure S12: Frequency of marker genes and spores in the population measured by microscopy + +<|ref|>sub_title<|/ref|><|det|>[[115, 575, 310, 593]]<|/det|> +## Supplementary Tables + +<|ref|>text<|/ref|><|det|>[[115, 601, 596, 620]]<|/det|> +Table S1: B subtilis probes ordered from TWIST Bioscience + +<|ref|>text<|/ref|><|det|>[[115, 629, 568, 647]]<|/det|> +Table S2: E coli probes ordered from TWIST Bioscience + +<|ref|>text<|/ref|><|det|>[[115, 656, 512, 675]]<|/det|> +Table S3: Probe amplification reagents and costs + +<|ref|>text<|/ref|><|det|>[[115, 684, 404, 702]]<|/det|> +Table S4: Strains used in this study + +<|ref|>text<|/ref|><|det|>[[115, 711, 343, 729]]<|/det|> +Table S5: Sample summary + +<|ref|>text<|/ref|><|det|>[[115, 738, 313, 756]]<|/det|> +Table S6: DGE analysis + +<|ref|>text<|/ref|><|det|>[[115, 765, 624, 784]]<|/det|> +Table S7: DGE analysis using 10x UMI instead of Probe UMI + +<|ref|>text<|/ref|><|det|>[[115, 792, 800, 811]]<|/det|> +Table S8: DGE analysis using bulk median instead of per- cell maximum probe counts + +<|ref|>text<|/ref|><|det|>[[115, 820, 648, 838]]<|/det|> +Table S9: Genes in each section of the Venn diagram in Figure 3d + +<|ref|>text<|/ref|><|det|>[[115, 847, 769, 866]]<|/det|> +Table S10: Genes used in volcano plots on Figure 3C and Figures S2, S4 and S5 + +<|ref|>sub_title<|/ref|><|det|>[[115, 875, 303, 893]]<|/det|> +## Supplementary Notes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 88, 833, 220]]<|/det|> +Supplemental note 1: Additional biology related to Figure 3 and Figure S2 Supplemental note 2: Additional info on Figure S3 and use of droplet RT instead of PCR Supplemental note 3: Additional biology related to Figure 4a and Figure S4 Supplemental note 4: Additional biology related to Figure 4b and Figure S5 Supplemental note 5: Calculation of probe concentrations + +<|ref|>sub_title<|/ref|><|det|>[[115, 281, 333, 301]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[115, 313, 884, 380]]<|/det|> +R.M., S.L, and A.Z.R. performed single cell and microbiology wet- lab experiments; R.M., D.S., S.H., and A.Z.R analyzed data. D.S. wrote custom analysis scripts. R.M., D.S., S.H. and A.Z.R. wrote and edited the manuscript. S.H. and A.Z.R. conceptualized the project; All authors designed experiments and reviewed the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[115, 444, 232, 464]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[110, 494, 884, 515]]<|/det|> +1. Ackermann, M. A functional perspective on phenotypic heterogeneity in microorganisms. + +<|ref|>text<|/ref|><|det|>[[150, 526, 477, 546]]<|/det|> +Nat. Rev. Microbiol. 13, 497- 508 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 558, 883, 608]]<|/det|> +2. Huang, K. C. Applications of imaging for bacterial systems biology. Curr. Opin. Microbiol. 27, 114-120 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 620, 883, 671]]<|/det|> +3. Locke, J. C. W. & Elowitz, M. B. Using movies to analyse gene circuit dynamics in single cells. Nat. Rev. Microbiol. 7, 383-392 (2009). + +<|ref|>text<|/ref|><|det|>[[110, 684, 763, 704]]<|/det|> +4. Milo, R. & Phillips, R. Cell Biology by the Numbers. (Garland Science, 2015). + +<|ref|>text<|/ref|><|det|>[[110, 716, 884, 767]]<|/det|> +5. Chen, S.-H. et al. UPS 2.0: unique probe selector for probe design and oligonucleotide microarrays at the pangenomic/genomic level. BMC Genomics 11 Suppl 4, S6 (2010). + +<|ref|>text<|/ref|><|det|>[[110, 780, 884, 864]]<|/det|> +6. Gundlach, J., Rath, H., Herzberg, C., Mader, U. & Stülke, J. Second Messenger Signaling in Bacillus subtilis: Accumulation of Cyclic di-AMP Inhibits Biofilm Formation. Front. Microbiol. 7, 804 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 877, 884, 897]]<|/det|> +7. Wang, D. et al. The orphan gene ybjN conveys pleiotropic effects on multicellular behavior + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 60, 888, 920]]<|/det|> +and survival of Escherichia coli. PLoS One 6, e25293 (2011).8. Schmidt, T. L. et al. Scalable amplification of strand subsets from chip-synthesized oligonucleotide libraries. Nat. Commun. 6, 1–7 (2015).9. Stuart, T. et al. Comprehensive Integration of Single cell Data. Cell 177, 1888–1902.e21 (2019).10. Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).11. Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L. & Wold, B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621–628 (2008).12. González-Ballester, D. et al. RNA-seq analysis of sulfur-deprived Chlamydomonas cells reveals aspects of acclimation critical for cell survival. Plant Cell 22, 2058–2084 (2010).13. Bartholomäus, A. et al. Bacteria differently regulate mRNA abundance to specifically respond to various stresses. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 374, 20150069 (2016).14. Ogura, M. et al. Whole-genome analysis of genes regulated by the Bacillus subtilis competence transcription factor ComK. J. Bacteriol. 184, 2344–2351 (2002).15. Berka, R. M. et al. Microarray analysis of the Bacillus subtilis K-state: genome-wide expression changes dependent on ComK. Mol. Microbiol. 43, 1331–1345 (2002).16. Dubnau, D. The regulation of genetic competence in Bacillus subtilis. Mol. Microbiol. 5, 11–18 (1991).17. Rosenthal, A. Z. et al. Metabolic interactions between dynamic bacterial subpopulations. Elife 7, (2018).18. Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. 25, 25–29 (2000).19. Carbon, S. et al. AmiGO: online access to ontology and annotation data. Bioinformatics 25, 288–289 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 66, 888, 936]]<|/det|> +20. Mi, H., Muruganujan, A., Ebert, D., Huang, X. & Thomas, P. D. PANTHER version 14: more genomes, a new PANTHER GO-slim and improvements in enrichment analysis tools. Nucleic Acids Res. 47, D419–D426 (2019).21. Eisenstein, B. I. Phase variation of type 1 fimbriae in Escherichia coli is under transcriptional control. Science 214, 337–339 (1981).22. Adicipitaningrum, A. M., Blomfield, I. C. & Tans, S. J. Direct observation of type 1 fimbrial switching. EMBO Rep. 10, 527–532 (2009).23. Spaulding, C. N. et al. Functional role of the type 1 pilus rod structure in mediating host-pathogen interactions. Elife 7, e31662 (2018).24. Clark, D. P. The fermentation pathways of Escherichia coli. FEMS Microbiol. Rev. 5, 223–234 (1989).25. Dar, D., Dar, N., Cai, L. & Newman, D. K. In situ single cell activities of microbial populations revealed by spatial transcriptomics. 40 (2021).26. Simanshu, D. K., Chittori, S., Savithri, H. S. & Murthy, M. R. N. Structure and function of enzymes involved in the anaerobic degradation of L-threonine to propionate. J. Biosci. 32, 1195–1206 (2007).27. Sawers, G. A novel mechanism controls anaerobic and catabolite regulation of the Escherichia coli tdc operon: Regulation of the E. coli tdc operon. Mol. Microbiol. 39, 1285–1298 (2001).28. Wu, G. et al. Arginine metabolism and nutrition in growth, health and disease. Amino Acids 37, 153–168 (2009).29. Kuchina, A. et al. Microbial single cell RNA sequencing by split-pool barcoding. Science (2020) doi:10.1126/science.aba5257.30. Blattman, S. B., Jiang, W., Oikonomou, P. & Tavazoie, S. Prokaryotic single cell RNA sequencing by in situ combinatorial indexing. Nat Microbiol 5, 1192–1201 (2020).31. Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 770, 108]]<|/det|> +752 pseudotemporal ordering of single cells. Nat. Biotechnol. 32, 381–386 (2014). + +<|ref|>text<|/ref|><|det|>[[55, 120, 884, 170]]<|/det|> +753 32. van der Wijst, M. G. P. et al. Single cell RNA sequencing identifies celltype-specific cis-eQTLs and co-expression QTLs. Nat. Genet. 50, 493–497 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 115, 884, 902]]<|/det|> +For B. subtilis in minimal media, in addition to the competent and sporulating cell subpopulations discussed in the main text, we observe two other clearly distinguishable cell groups indicative of distinct biological states. The largest group of cells (comprising clusters 1, 2, 3, 5, and part of cluster 7) upregulate dhbA, dhbB, dhbC, dhbE, and dhbF (the 5 enzymes implicated in the biosynthesis of bacillibactin) as well as yusV, an ABC transporter of bacillibactin. In addition, at least 3 out of the 5 clusters within the subpopulation showed upregulation of genes involved in thiamine biosynthesis (tenA, tenI, thiD, thiG, thiF, thiS, thiO; gene set enrichment = 16.85, FDR = 2.79E- 05), amino acid biosynthesis (fold enrichment = 5.23, FDR = 1.48E- 07), amino acid activation (thrS, gatB, ileS, aspS; fold enrichment = 6.22, FDR = 3.46E- 02), surfactant biosynthesis (srfAA, srfAB, srfAC, srfAD), "de novo" IMP biosynthesis (gene set fold enrichment = 25.79, FDR = 9.10E- 09), purine biosynthesis (purB, purC, purD, purQ, purH, purF, purL, purK, purM, purS; gene set fold enrichment = 21.66, FDR = 1.83E- 02), pyrimidine biosynthesis (pyrAA, pyrK, pyrF, pyrH; gene set enrichment = 9.63, FDR = 2.59E- 02), chorismate biosynthesis (aroA, aroB, aroC, aroH, aroE; gene set fold enrichment = 18.05, FDR = 4.60E- 03), cell motility (hag, fliY, fliD, fliH, flhP; gene set enrichment = 6.19, FDR = 1.44E- 02), sulfate assimilation (cysl, cysH, cysC, cysJ; gene set enrichment = 18.05; FDR = 4.56E- 03), and biotin biosynthesis (bioA, bioB, bioD, bioF, bioW; gene set fold enrichment = 18.05, FDR = 4.65E- 03). Altogether, this subpopulation appears to represent the most metabolically active cell state within the total population, an observation further supported by the fact that genes encoding ribosomal components & translation machinery were also enriched (rpmI, rplU, rplR, rpsM, rpsE, rpsI, rpsN, rpsD, rpoB, rplV, rplX, rplB, rpmA, rpsS, rplS, rpsP, rplL, rpmD, rpmC; "translation" gene set fold enrichment = 7.81, FDR = 3.07E- 12). Cell cluster #2, part of the subpopulation, may represent the gateway into sporulation as early sporulation factors sigF, spooA, and spoIIAB are all uniquely upregulated and cells are projected in close proximity with cluster 9 after dimensional reduction. In addition, these cells downregulate genes implicated in motility (hag, fliD, fliG, fliY, fliK, fliM, flgK, flgL) in stark contrast to other cells within the subpopulation. Furthermore, cells within cluster #5 are distinguished from others in the "metabolically active" subpopulation by expression of genes related to arginine biosynthesis via ornithine. L- arginine is used by cells for protein synthesis as well as for production of polyamines such as putrescine and spermidine; compounds which have functions in nucleic acid binding, protein activation, and membrane stabilization due to their polycationic state. The 4th distinct cell state detected in B. subtilis, only slightly less in size compared to the "metabolically active" cell group (45% of total population), was characterized by lower expression of many of the genes upregulated in the "metabolically active" subpopulation. Uniquely, cells in cluster 0 (930 cells, 76% of the subpopulation) were found to overexpress sdpB (logFC = 0.68, adjusted p-value = 4.7E- 4). Under starvation conditions, B. subtilis produces an antimicrobial peptide through the sdpABC operon that is active against its own species, a tactic used by the cell to increase its own chance of survival and delay commitment to sporulation. While the sdpC gene encodes the 42AA toxic peptide, sdpA & sdpB are required for maturation of the peptide into its active form prior to secretion. sdpB is a multipass membrane protein which affects the toxicity of the final SDP peptide without being involved in signal peptide cleavage, disulfide bond formation, or secretion. The sdpABC operon was found expressed in cells within each of the four subpopulations, however, it is only within cluster #0 that sdpB was found to be significantly upregulated. This finding, along with the overall reduction in transcript load as compared to the "metabolically active" subpopulation, suggests a more dormant biological state in which cells appear to be facing a nutritional strain but have not yet committed to an alternative developmental program. This hypothesis is supported by the fact that cells in cluster 7 are split between the two largest subpopulations in the UMAP projection as well as by the clustering of a few cells associated with competence (part of cluster 8) near this dormant subpopulation – possibly suggesting a developmental transition between the two states. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 135, 830, 153]]<|/det|> +Supplemental note 2: Additional info on Figure S3 and use of droplet RT instead of PCR + +<|ref|>text<|/ref|><|det|>[[112, 161, 883, 469]]<|/det|> +While the approach in Fig S3 (RT in drops instead of PCR) proved less sensitive in transcript capture compared to the final, optimized method, we still resolve distinct biological states upon dimensional reduction of expression vectors including cell subpopulations corresponding to competence and sporulation. Additionally, we observe a majority grouping of cells (clusters #0- 5) with significantly increased expression of ribosomal genes as well as genes involved in purine and amino acid biosynthesis - corresponding to the "metabolically active" subpopulation defined previously. scRNA- seq also revealed a subpopulation of cells (cluster #8, 1.9% of cells) within the sample exhibiting a state unforeseen in the previous run - characterized by the upregulation of genes encoding a manganese importer (mntABCD, logFCs > 1.99, adjusted p- values < 1E- 11), sigma factor sigW (logFC = 1.69, adjusted p- value = 5E- 16), and stress response proteins (yuaF, yceDEF, yqfA, yqeZ, logFCs > 0.99, adjusted p- values < 3.3E- 04) involved in maintaining cell membrane integrity. yceF, specifically, is a cell envelope stress protein that conveys resistance to manganese. katA, an iron- binding catalase which protects cells against toxification from hydrogen peroxide, was also upregulated. Altogether, this cluster of cells appears to be responding to stress induced by a low concentration of manganese within the environment by upregulating manganese uptake machinery. maeA was also upregulated in these cells (logFC = 1.9, adjusted p- value = 3.65E- 34). maeE encodes a malic enzyme necessary for catabolic repression of pyruvate import (pftAB) as well as for maintaining ATP levels during growth on the carbon source by conversion of malate to pyruvate and generation of NADH. + +<|ref|>sub_title<|/ref|><|det|>[[115, 510, 723, 528]]<|/det|> +## Supplemental note 3: Additional biology related to Figure 4a and Figure S4 + +<|ref|>text<|/ref|><|det|>[[112, 537, 883, 908]]<|/det|> +Single cell transcriptomes of E. coli grown in M9 were clustered into ten groups. In the largest group of cells (cluster #0, 514 cells, 15.5% of total), ribosomal genes (rpl operon, rpsC, rpsD, rpmC, rpmD) as well as deaD and priB are amongst the most upregulated compared to the rest of the cell population. deaD (logFC = 0.7, adjusted p- value = 2.3E- 11) is a DEAD- box RNA helicase involved in ribosome biogenesis, translation initiation, and mRNA degradation at low temperatures - forming a cold- shock degradosome with RNase E. Kuchina et al. also identified a subpopulation of E. coli differentially expressing deaD, along with other cold shock proteins, in their scRNA- seq experiments and hypothesized the cluster was an artifact of sample preparation. Since our protocol fixes cells directly in culture, we find it unlikely that this state is a product of cold shock - a conclusion reinforced by the fact that we observe no other cold shock associated proteins within the set of upregulated genes. priB (logFC = 0.37, adjusted p- value = 6.4E- 12) encodes a protein within the primosome which catalyzes lag strand priming during DNA replication. Conversely, genes involved in pyrimidine biosynthesis (pyrl, pyrB, pyrC) were significantly downregulated within the cluster. Interspersed with cluster #0 on the dimensionally reduced plot, we observe a group of cells (cluster #8, 141 cells) highly upregulating genes from the fim operon, as discussed in the main text. Cells in cluster #2 appear less active overall, as all genes found to be significantly differentially expressed within the cluster are downregulated compared to the rest of the population. This may be an effect of poor probe penetration within the protocol or indicative of a more basal cell state. Cluster #3 reveals no significantly overrepresented gene sets by GSEA after accounting for the false discovery rate. Upregulated genes reported in the cluster include ompT and ompX as well as tolB and tolC (logFCs > 0.25, adjusted p- values < 1E- 14). In addition, scRNA- seq reveals a cluster of cells (cluster #4, 403 cells, 12% of total) that appear to be entering stationary phase; exhibiting a relative increase in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 883, 537]]<|/det|> +expression of dps (logFC = 0.91, adjusted p- value = 4.17E- 26). dps encodes a protective protein which binds to the bacterial chromosome and sequesters intracellular Fe2+, forming a highly stable protein- DNA mineral complex which protects the DNA from oxidative damage. In E. coli, dps has been associated with the transition into stationary phase as cells quickly work to preserve genetic material from stressors including UV irradiation, metal toxicity, thermal fluctuations, and acid/base shock. Cluster #9 (117 cells, 3.5% of total) was also shown to upregulate dps (logFC = 1.05, adjusted p- value = 7.04E- 16) while highly expressing the gad (gadA, gadB, gadC) and hde (hdeA, hdeB, hdeD) operons. Both operons encode resistance to acid stress and fall under the gene set of "regulation of intracellular pH" (fold enrichment > 100, FDR = 6.72E- 03). gadA and gadB produce the subunits of glutamate carboxylase which allows cells to interconvert between L- glutamate and GABA by incorporation of an intracellular proton while gadC encodes the antiporter enabling L- glutamate uptake and GABA export. hdeA and hdeB encode acid stress chaperone proteins with differing pH optimums; both preventing the aggregation of periplasmic proteins denatured under acidic conditions. osmY, encoding a periplasmic chaperone protein, was also upregulated within this cluster and has been shown to be induced by entry into the stationary phase. Altogether, DGE analysis reveals a fraction of cells within the population that are experiencing significant stress. Cells in cluster #4 appear to be at the initiation of this state while cells in the cluster #9 are more clearly induced, expressing dps along with proteins dealing with acid, osmotic, and oxidative stresses. It is possible these clusters represent cells transitioning into stationary phase or, alternatively, a transient state experienced during active metabolism/exponential growth. Cells within clusters 5 and 7 upregulate genes involved in pyrimidine biosynthesis as discussed in the main text. Interestingly, cluster #5 differentially overexpresses argG (logFC = 0.89, adjusted p- value = 6.57E- 29), although no other genes specific to arginine biosynthesis were found to be upregulated. Both clusters upregulate ygjF, encoding a G/U mismatch-specific DNA glycosylase to correct mispairings in double stranded DNA that arise from alkylation or deamination of cytosine. ygjF is often active in stationary-phase cells 36. Cluster 6 is defined by the upregulation of genes involved in arginine biosynthesis as discussed in the main text. + +<|ref|>sub_title<|/ref|><|det|>[[115, 626, 723, 645]]<|/det|> +## Supplemental note 4: Additional biology related to Figure 4b and Figure S5 + +<|ref|>text<|/ref|><|det|>[[112, 653, 883, 895]]<|/det|> +Upregulated genes within the largest cell cluster (cluster #0, 677 cells, 18% of total) included fliC (LogFC = 1.02, adjusted p- value = 3.81E- 54), uncharacterized genes ygeWXY (logFC > 0.7, adjusted p- values < 3.3E- 32), genes putatively involved in the conversion of adenine to guanine during nucleotide salvage (xdhA, xdhD, logFCs > 0.85, adjusted p- values < 1E- 34), ssnA (logFC = 0.80, adjusted p- value = 9.8E- 23), a putative, phase- dependent regulator of cell viability during later stage growth, and genes in the tna operon (tnaABC, logFCs > 0.45, adjusted p- values < 1E- 28). tnaA encodes tryptophanase, which converts L- tryptophan to indole and pyruvate as part of amino acid catabolism while tnaB is a low affinity permease and tnaC is a positively regulating leader peptide. Cell clusters #1 and #4, adjacent and densely packed groups on the UMAP, appear to represent a subpopulation of cells responding to over- acidification of the cytoplasm. Both clusters significantly upregulate the gad operon (gadABC) as well as dps. In addition, acid stress chaperone proteins hdeA and hdeB are overexpressed in cluster #4. No statistically significant gene sets were identified from upregulated genes in clusters #2- 3, however, cells in both clusters express genes related to anaerobic fermentation as expected under the culture conditions. Cells in cluster #5 (283 cells, 7.5% of total) are characterized by the upregulation of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 883, 408]]<|/det|> +genes involved in aerobic respiration as discussed in the main text. We also observe the strong upregulation of genes involved in ribosome assembly (fold enrichment \(= 15.26\) , FDR \(= 6.16\mathrm{E - 21}\) ) and translation machinery (fold enrichment \(= 12.01\) , FDR \(= 6.26\mathrm{E - 33}\) ) in these cells, indicating a more active metabolic state which can most likely be attributed to the increases in ATP yield that is achieved by aerobic respiration. We observe similar enrichment of ribosome- related genes in cell cluster #6 (191 cells, \(5\%\) of total) along with the upregulation of a few genes in glycolysis (eno, lpd, aceEF, logFC \(>0.5\) , adjusted p- values \(< 2.2\mathrm{E - 06}\) ) and oxidative phosphorylation (atpD, cydAB, logFC \(>0.7\) , adjusted p- values \(< 1\mathrm{E - 06}\) ). These findings suggest that cluster #6 is in a differentiating state, transitioning between aerobic and anaerobic metabolism. Interestingly, cell cluster #7 (135 cells, \(3.6\%\) of total) appears to capture cells undergoing anaerobic metabolism while responding to oxidative stress. Gene sets describing anaerobic respiration (acnA, yjil, nrfA, frdC, frdA, ynfF, ynfE, nrfB, dmsA; fold enrichment \(= 8.00\) , FDR \(= 5.63\mathrm{E - 03}\) ) and oxidative stress (acnA, ahpF, osmC, qor, pgi, sufD, ydbK, yliT, yggE, clpA; fold enrichment \(=\) , FDR \(= 2.68\mathrm{E - 02}\) ) were both identified as overrepresented in the list of upregulated genes within the cluster. In addition, stress proteins such as hspQ (a heat shock response protein) and osmY (hyperosmotic response protein) were upregulated along with flu (logFC \(= 0.47\) , adjusted p- value \(= 2.88\mathrm{E - 14}\) ), a surface- display, self- recognizing adhesin protein encoding gene which enables auto aggregation and flocculation of cells in static liquid culture as well as biofilm formation by uropathogenic E coli in situ. Cluster #8 (111 cells, \(3\%\) of total) is characterized by the strong upregulation of the tdc operon (threonine metabolism under anaerobic conditions) as discussed in the main text. + +<|ref|>sub_title<|/ref|><|det|>[[115, 454, 585, 472]]<|/det|> +## Supplemental note 5: Calculation of probe concentrations + +<|ref|>sub_title<|/ref|><|det|>[[115, 482, 384, 499]]<|/det|> +## Determining Probe Concentration + +<|ref|>text<|/ref|><|det|>[[115, 509, 883, 611]]<|/det|> +To determine the concentration of probes needed to saturate an approximate number of transcripts in solution, saturation was defined as total, nonspecific probe coverage for each transcript. In other words, a pool of transcripts is considered saturated when, for each molecule, there is an entire library of probes ( \(n = 1\) probe per unique design) exclusively dedicated. In reality, this system would be over- saturated as it contains enough probes to tag every transcript even if all transcripts have the same sequence. + +<|ref|>text<|/ref|><|det|>[[115, 621, 882, 690]]<|/det|> +In our protocol, we prepare 150 uL of fixed cells from a 2X concentrated culture collected at mid- log phase. Assuming the culture contains 1E8 cells/mL prior to concentration and each bacterial cell contains 2,000 transcripts \(^{13}\) , the following estimate can be made using a probe set containing 20,000 unique ssDNA probes of length 138 bp ( \(\sim 42,500 \text{g/mol}\) ). + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[260, 103, 741, 140]]<|/det|> +\[\frac{2E8 \text{ cells}}{mL} * \left(\frac{2000 \text{ transcripts}}{cell}\right) = \frac{4E11 \text{ transcripts}}{mL} = 0.66 \text{ nM transcripts}\] + +<|ref|>equation<|/ref|><|det|>[[303, 160, 700, 197]]<|/det|> +\[\left(\frac{1 \text{ probeset}}{\text{transcript}}\right) \left(\frac{20,000 \text{ unique probes}}{1 \text{ probeset}}\right) = 20,000 \frac{\text{probes}}{\text{transcript}}\] + +<|ref|>equation<|/ref|><|det|>[[245, 219, 759, 255]]<|/det|> +\[0.66 \text{nM transcript} * \left(\frac{20,000 \text{ probes}}{\text{transcript}}\right) = 13,020 \text{nM probes} = 13.20 \text{uM probes}\] + +<|ref|>equation<|/ref|><|det|>[[308, 275, 700, 313]]<|/det|> +\[13.20 \text{uM probes} \left(\frac{1 \text{M}}{10^6 \text{uM}}\right) \left(\frac{42,500 \text{g}}{\text{mol}}\right) = \frac{0.561 \text{g}}{L} = \frac{560 \text{ng}}{\text{uL}}\] + +<|ref|>text<|/ref|><|det|>[[113, 333, 883, 541]]<|/det|> +In our final experiments, we achieved a probe concentration of \(600 \text{ ng / uL}\) (0.6 mg/mL). To note, our calculations assume equal representation of all probe designs in the applied probe library which is unlikely due to biases introduced during probe synthesis and amplification. However, in all likelihood, this number represents a gross overestimate of probe requirements for a few reasons. Cell concentration and, therefore, transcript load was purposefully overestimated by assuming an exponentially growing culture contains 1E8 cells/mL; this number better represents the carrying capacity of B. subtilis in minimal media. In addition, our definition of "saturation" was overly strict, as discussed above. Lastly, probe libraries were designed redundantly to minimize noise in the downstream analysis by creating multiple probes to target different regions of the same gene. For the B. subtilis library, we designed, on average, 7 probes per gene - further assuring that our final probe concentration is sufficient to tag all available transcripts for scRNA-seq. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[55, 101, 690, 830]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[43, 852, 120, 871]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[43, 894, 950, 958]]<|/det|> +Genome wide scRNA-seq probe design and synthesis a. A genome for the organism of interest is used to design a library of probes that are complementary to unique regions within each gene's coding sequence (CDS). Complementary ssDNA oligonucleotides are amended by addition of an extender sequence at the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 950, 223]]<|/det|> +5' end and 3' PCR handle. Sequences are ordered as an oligopool b. Probe sequences are inverted in- silico and restriction sites and handles are added for circle- to- circle amplification c. The large, inverse oligonucleotide library is commercially- synthesized and pooled. Ordered oligo- pools are amplified in a 3- step "rolling circle" reaction with Phi29 polymerase to produce the final, correctly oriented proto- probe library. d. A random 12mer sequence (UMI) and a 30- base polyA tail is added by primer extension with a blocked primer that is complementary to the 3' extender sequence. Single stranded DNA containing a universal PCR handle, UMI, complementary transcript region, and poly- A tail (listed 5' to 3') is purified. e. Finalized probe libraries are free of amplification primers and well distributed. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 55, 600, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 802, 116, 820]]<|/det|> +
Figure 2
+ +Microfluidic probe- based scRNA- seq method and validation. a. Cells are fixed and permeabilized to allow penetration of thousands of unique, genome- specific oligonucleotide probes. Hybridized probes "retrofit" transcripts with a poly- A tail & UMI while unhybridized probes are washed away. b. Cells are then flowed through a commercial microfluidic device which encapsulates single cells into droplets (shown in c.) containing barcoded primers conjugated to a hydrogel microsphere and PCR reagents. d. Barcoded cDNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 954, 202]]<|/det|> +is generated from the mRNA:probe hybridized complex via in-droplet PCR or RT reaction. Droplets are then broken, the pooled cDNA amplified further before sequencing. Single cell transcriptomes are resolved, clustered, and visualized. e. Transcriptomic estimation by probe hybridization in bulk samples is correlated to transcriptomic measurement of bulk samples by traditional RNA-seq f. Species mixture ("barnyard") plot demonstrates that single cells of different bacterial species can be resolved by barcode after microfluidic encapsulation. g. Aggregated probe-based signal from thousands of single cells correlates with the average probe-based signal of the bulk population. + +<|ref|>image<|/ref|><|det|>[[55, 210, 840, 930]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 116, 62]]<|/det|> +## Figure 3 + +<|ref|>text<|/ref|><|det|>[[41, 84, 950, 333]]<|/det|> +scRNA- seq analysis reveals known and novel states in B subtilis. a. Heatmap of marker gene expression (z- score of log- transformed values) from 2,784 individual B. subtilis cells organized into 10 clusters. b. UMAP 2- dimensional representation of the 10 cell clusters reveals 4 highly distinct transcriptomic signatures c. Single cell expression of key marker genes for competence (comGE; clusters 6 and 8), sporulation (spollID, cluster 9) and arginine synthesis (argC, cluster 5) are highlighted on the UMAP (top panels, left to right). Volcano plots of genes expressed in at least \(25\%\) of cells in the corresponding clusters, with genes from the respective processes highlighted in green (middle panels). The lower panels show the presence of each heterogeneous marker in the population as confirmed by fluorescent promoter- reporter constructs (PcomG- YFP, PcotY- YFP, PargC- YFP, respectively). A phase bright spore can be seen in the cotY- expressing cell. d. \(90\%\) of all comK regulon genes probed (45/50) were significantly differentially upregulated in cell clusters 6 and 8 (Bonferroni corrected P- value \(\leq 0.05\) for each gene) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 55, 650, 768]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 799, 118, 817]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 840, 936, 951]]<|/det|> +Heterogeneous gene expression in E coli grown in minimal medium and anaerobically. a. Heatmap of marker gene expression (z- score of log- transformed values) from 3,315 individual E. coli cells grown aerobically in minimal M9 media organized into 10 clusters. b. Heatmap of gene expression from 3,763 individual E. coli cells grown in anaerobic LB media organized into 9 clusters. c. UMAP 2- dimensional representation of the 10 cell clusters from aerobic M9 culture conditions. d. Flagellar components are + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 955, 271]]<|/det|> +heterogeneously expressed in aerobic M9 media - a key flagellar gene (flagellin - flic) is preferentially expressed by cells in cluster 1. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. e. UMAP 2- dimensional representation of the 9 cell clusters from E coli fermenting LB media culture conditions. f. Flagellar components are heterogeneously expressed in fermentation conditions - a key flagellar gene (flagellar basal- body rod protein - fIgC) is preferentially expressed by cells in cluster 5. The heterogeneous presence of flagella in the population is confirmed by flagellar staining. g. Many aerobic respiration genes and other metabolic and flagellar genes are overexpressed in cluster 5. Volcano plots of genes over- expressed in at least \(25\%\) of cells in cluster 5 from these processes are highlighted in colors as labeled. Green labeled genes, involved in central- carbon metabolism, are also highlighted as green arrows in the metabolic map. + +<|ref|>sub_title<|/ref|><|det|>[[44, 293, 311, 320]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 343, 765, 363]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 380, 562, 428]]<|/det|> +- SupptablesfinalizedforinitialsubmissionAutosaved.xlsx- suppfiguresfinalizedforinitialsubmission.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/images_list.json b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..191ef1b6b4eb731bbc7069e1b248751e1911845a --- /dev/null +++ b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/images_list.json @@ -0,0 +1,123 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1: Fly-by-feel-based flight control strategy.", + "footnote": [], + "bbox": [ + [ + 120, + 139, + 880, + 666 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: Aerodynamic feature extraction from wing strain data.", + "footnote": [], + "bbox": [ + [ + 123, + 130, + 870, + 585 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: Position control by sensing the wind direction and speed, which vary depending on the position.", + "footnote": [], + "bbox": [ + [ + 120, + 150, + 878, + 675 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4: Reinforcement learning in a 2 DOF environment.", + "footnote": [], + "bbox": [ + [ + 120, + 132, + 880, + 650 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5: Position control of a flapping-wing drone with the proposed fly-by-feel system.", + "footnote": [], + "bbox": [ + [ + 120, + 132, + 875, + 666 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6: Flight path control of a flapping-wing drone in windless environment.", + "footnote": [], + "bbox": [ + [ + 120, + 132, + 875, + 651 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Extended Data Figure 1. Correlation between the flapping-wing power and sensor signal from the wind. a, Schematic of the experimental setup. b, Real image showing the experimental setup. c, Change in the duty cycle as the motor power was increased from 15\\~100%. d, Variation in the thrust force of the drone as the motor power increased. The thrust force was recorded by nano 17 fixed under the flapping drone. e, Variation in the sensor signal as the motor power increased. f, In detail, as the motor power increased, the frequency of a single flapping cycle and signal amplitude both increased.", + "footnote": [], + "bbox": [ + [ + 170, + 100, + 825, + 576 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Extended Data Figure 3. Additional examples of the 1 DOF system. a, Snapshots of the episode when the target angle is \\(180^{\\circ}\\) . It moves quickly to reach target position and reduces its motor power when it passes over the target. b, The case when the target position angle is \\(270^{\\circ}\\) . When the target position is changed, the drone could recognize the position based on the sensor information. c, The case when the researcher disturbs the flight by pushing the drone backward and forward. Even if there is an intentional disturbance, a trained drone controls its motor power to maintains the target position. d, Motor power control and corresponding angle change for case 1. e, Motor power control and corresponding angle change for case 2. f, Motor power control and corresponding angle change for case 3.", + "footnote": [], + "bbox": [], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_10.jpg", + "caption": "Extended Data Figure 10. Methods of yielding odometry and results. a, Labelling procedure for comparing odometry and ground truth trajectories. b, Comparison of mean squared error between 5 folds of different training datasets. c, Comparison of mean squared error between 5 folds of different validation datasets. d, Representative result of comparison between odometry and ground truth trajectories from additional experiment.", + "footnote": [], + "bbox": [], + "page_idx": 37 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178.mmd b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178.mmd new file mode 100644 index 0000000000000000000000000000000000000000..967f0d5ca242edc3acb96f599de8b2ab407976dc --- /dev/null +++ b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178.mmd @@ -0,0 +1,509 @@ + +# Fly-by-Feel: Wing Strain-based Flight Control of Flapping-Wing Drones through Reinforcement Learning + +Daeshik Kang dskang@a.jou.ac.kr + +Ajou University https://orcid.org/0000- 0002- 7490- 9079 + +Seungyong Han Ajou University https://orcid.org/0000- 0003- 2870- 4088 + +Je- sung Koh Ajou University https://orcid.org/0000- 0002- 7739- 0882 + +Taewi Kim Ajou University https://orcid.org/0000- 0002- 8570- 348X + +Insic Hong Ajou University + +Sunghoon Im Ajou University https://orcid.org/0000- 0002- 2595- 157X + +Seungeun Rho Georgia Institute of Technology + +Minho Kim Ajou University + +Yeonwook Roh Ajou University https://orcid.org/0000- 0002- 2823- 2028 + +Changhwan Kim Ajou University https://orcid.org/0000- 0002- 0626- 730X + +Jieun Park Ajou University + +Daseul Lim Ajou University https://orcid.org/0000- 0002- 1990- 9092 + +Doohoe Lee Ajou University + +Seunggon Lee Ajou University https://orcid.org/0000- 0001- 6305- 985X + +Jingoo Lee + +<--- Page Split ---> + +Ajou University https://orcid.org/0000- 0002- 5783- 2583 + +Inryeol Back Ajou University https://orcid.org/0000- 0002- 0730- 6585 + +Joonho Lee ETH Zurich https://orcid.org/0000- 0002- 5072- 7385 + +Sungchul Seo Seogyeong University + +Uikyum Kim Ajou University + +Junggwang Cho Ajou University + +Myung Rae Hong Ajou University + +Sanghun Kang Ajou University + +Young-Man Choi Ajou University + +## Article + +Keywords: + +Posted Date: May 30th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4443963/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Machine Intelligence on September 20th, 2024. See the published version at https://doi.org/10.1038/s42256- 024- 00893- 9. + +<--- Page Split ---> + +# Fly-by-Feel: Wing Strain-based Flight Control of Flapping-Wing Drones through Reinforcement Learning + +Taewi \(\mathrm{Kim}^{1,\dagger}\) , Insic Hong \(^{1,\dagger}\) , Sunghoon \(\mathrm{Im}^{1,\dagger}\) , Seungeun \(\mathrm{Rho}^{2,\dagger}\) , Minho \(\mathrm{Kim}^{1}\) , Yeonwook \(\mathrm{Roh}^{1}\) , Changhwan \(\mathrm{Kim}^{1}\) , Jieun \(\mathrm{Park}^{1}\) , Daseul \(\mathrm{Lim}^{1}\) , Doohoe \(\mathrm{Lee}^{1}\) , Seunggon \(\mathrm{Lee}^{1}\) , Jingo \(\mathrm{Lee}^{1}\) , Inryeol \(\mathrm{Back}^{1}\) , Junggwang \(\mathrm{Cho}^{1}\) , Myung Rae \(\mathrm{Hong}^{1}\) , Sanghun \(\mathrm{Kang}^{1}\) , Joonho \(\mathrm{Lee}^{3}\) , Sungchul \(\mathrm{Seo}^{4}\) , Uikyum \(\mathrm{Kim}^{1}\) , Young- Man \(\mathrm{Choi}^{1}\) , Je- sung \(\mathrm{Koh}^{1,*}\) , Seungyong \(\mathrm{Han}^{1,*}\) , Daeshik \(\mathrm{Kang}^{1,*}\) + +\(^{1}\) Department of Mechanical Engineering, Ajou University, Suwon, Korea + +\(^{2}\) College of Computing, Georgia Institute of Technology, Georgia, USA + +\(^{3}\) Robotic Systems Lab, ETH- Zürich, Zürich, Switzerland + +\(^{4}\) Department of Nanochemical, Biological and Environmental Engineering, Seokyeong + +University, Korea + +14 + +\(^{+}\) These authors contributed equally to this work. + +16 + +\(^{*}\) Address correspondence to: + +Prof. Je- Sung Koh; jskoh@ajou.ac.kr + +Prof. Seungyong Han; sy84han@ajou.ac.kr + +Prof. Daeshik Kang; dskang@ajou.ac.kr + +<--- Page Split ---> + +## Abstract + +Although drone technology has progressed significantly, replicating the dynamic control and wind- sensing abilities of biological flights is still beyond our reach. Biological studies have revealed that insect wings are equipped with mechanoreceptors known as campaniform sensilla, which detect complex aerodynamic loads critical for flight agility. By leveraging robotic experiments designed to mimic these biological systems, we confirmed that wing strain provides crucial information about the drone's attitude, as well as the direction and velocity of the wind. We introduce a novel wing strain- based flight controller, termed 'fly- by- feel'. This methodology employs the aerodynamic forces exerted on a flapping drone's wings to deduce vital flight data, such as attitude and airflow without accelerometers and gyroscopic sensors. Our empirical approach spanned five key experiments: initially validating the wing strain sensor system for state information provision, followed by a single degree of freedom (1 DOF) control in changing winds, a two degrees of freedom (2 DOF) control for gravitational attitude adjustment, a test for position control in windy conditions, and finally, demonstrating precise flight path manipulation in a windless condition using only wing strain sensors. We have successfully demonstrated control of a flapping drone in a various environment using only wing strain sensors, with the aid of reinforcement learning- driven flight controller. The fly- by- feel system holds the potential to revolutionize autonomous drone operations, providing enhanced adaptability to environmental shifts. This will be beneficial across varied applications, from gust resistance to wind- assisted flight, paving the way toward the next generation of resilient and autonomous flying robots. + +<--- Page Split ---> + +## Introduction + +Understanding the aerodynamic mechanism \(^{1 - 3}\) of biological flights has led to remarkable progress in the system design of flapping- wing micro aerial vehicles (FWMAVs) such as RoboBee \(^{4}\) , DelFly \(^{5}\) , KUBeele \(^{6}\) , and Purdue Hummingbird \(^{7}\) . Although these FWMAVs offer high agility and energy efficiency at a small scale over quadcopters \(^{8}\) and fixed- wing drones \(^{9}\) , they encounter difficulties in terms of flight control and lightweight design \(^{10}\) . Specifically, controlling in windy conditions is challenging due to their complex dynamics and small size, which limit to generate sufficient thrust force and moments \(^{11}\) . To control FWMAVs, sensors and feedback system combining inertial measurement units (IMUs), vision cameras, and proportional- integral- derivative (PID) controllers are commonly utilized \(^{12,13}\) . However, they are not sufficient for controlling FWMAVs due to several reasons. Firstly, IMUs are typically located on the bodies, requiring body movement for disturbance detection, which can often cause the delayed stabilization of the robot \(^{14}\) . Second, existing sensors are inadequate to handle the flexible body of flapping wing drones due to their rigidity. Third, model- based controllers, with limited tuning parameters, struggle to manage the complex aerodynamics of flexible flapping wings \(^{15,16}\) . + +Flying insects (e.g. dragonflies, hawk moths, and hoverflies) possess a high level of flight ability, capable of maneuvers like hovering, turning, and gliding, even in unsteady aerodynamic environments \(^{17 - 22}\) . Notably, some species like the globe skimmer dragonfly traverse thousands of kilometers, from India to East Africa \(^{23}\) , by utilizing fast, high- altitude airstreams and navigating seasonally favorable directions \(^{24 - 28}\) . The superior sensing and control mechanism are attributed to active stabilization, supported by complex sensory systems including mechanosensory, vision, and olfaction \(^{29 - 31}\) . Among these sensory organs, the campaniform sensilla, which are strain- detecting mechanosensory distributed on the wings, enable the immediate, proprioceptive flight control \(^{32,33}\) . The wings of flying insects are subject to combined inertial- aerodynamic (aeroelastic) loads because of various aerodynamic phenomena, including airflow stagnation and separation, and nonlinear phenomena such as vortex growth and shedding \(^{34,35}\) . Moreover, recent researches claim that the campaniform sensilla detect the Coriolis force and function as inertial sensors based on electrophysiology data \(^{36 - 38}\) and numerical simulations \(^{39,40}\) . From an engineering perspective, the direct measurement of wing load is a significant advantage for flight control because the unsteadiness in aerodynamics makes it hard for conventional control systems to accurately + +<--- Page Split ---> + +estimate the wing load in real- time41. Therefore, inspired by the control systems of these flying insects, attempts have been made to develop controllers that can sensitively respond to changes in airflow by measuring the wing deformation of aircraft14. However, there have been no successful cases of flapping drones because the aerodynamics between easily deformable wings and airflow are extremely complex42,43, and developing sensing and control algorithms for stable flight and manipulation has been difficult, even with supercomputers44. + +In this article, we present a wing strain- based flight controller, which we term fly- by- feel. Given the computational and complexity challenges associated with modeling aerodynamics, we employed the powerful and cutting- edge method of reinforcement learning. Utilizing wing strain sensors, our goal is to assess the feasibility and reliability of such a system in autonomously determining flight paths. Our experimental approach is structured into five distinct phases. 1) Validation as a state informative sensor system: The initial phase of the research focuses on verifying the wing strain sensor's ability to provide useful and reliable state information for drone control. This foundational step is crucial in confirming the sensory system's detection capabilities. 2) Single degree of freedom (1 DOF) flight path control experiment: This stage evaluates how effectively the drone can adapt to changing wind speeds and directions, and how well it can manage one- dimensional flight paths using data from strain sensors. 3) Two degrees of freedom (2 DOF) attitude control experiment: This experiment tests the drone's ability to identify and maintain optimal pitch angles through gravity- based attitude adjustment, which is vital for precise posture control and flight path modification. 4) Position control in windy Environment: This phase tests the drone's capability to maintain its position in a wind- influenced environment solely using wing strain sensor data. Success in this experiment would underline the potential of strain sensors to manage complex aerodynamic conditions without additional sensory inputs. 5) Wing strain- based flight path control in clam environment: The final experimental step involves verifying the drone's ability to control its flight path in a windless environment using only the information provided by wing strain sensors, which would demonstrate the precision of location and attitude adjustments possible with this system. Each stage of this study meticulously verifies different aspects of the system, building upon each other to substantiate the utility and effectiveness of an autonomous flight control system based on strain sensor data. Through this series of empirical tests, we aim to implement the biological hypothesis that wing strain sensors are useful for immediate flight control in response to airflow changes + +<--- Page Split ---> + +into robotic experiments and explore the potential of wing strain- based control systems in real flight situations. + +## Main + +## Fly-by-feel control system incorporates RL based on wing strain + +Inspired by insects' agile flight abilities, the fly- by- feel control system shown in Fig. 1 enables rapid adaptation to complex airflow conditions and immediate response to sudden fluctuations. The campaniform sensilla are sensory organs distributed on the wing base that sense the strain on the wing45 (Fig. 1a middle, Supplementary Fig. 1a). The spike rate of neurons depends on the flapping motion of the wing, as shown in Fig. 1a left, and the campaniform sensilla can detect the rotational rate during flight38. To develop an efficient autonomous flight system by understanding the role of the Campaniform sensilla in flying insects, ultrasensitive crack- based strain sensors46- 48 were attached to the wing bases of commercial wing- flapping drones inspired by the sensory system of a flying insect (Fig. 1a right). + +When training a drone to fly, the flapping motion of the drone's wings continuously changes the airflow, causing a variety of complex aerodynamic phenomena. The crack- based strain sensors attached to the wing base collect aeroelastic information in various states (Fig. 1b). After analyzing the wing deformation signal with a temporal convolutional neural network (1D- CNN) model, we discovered that the signal contains possible clues for estimating the external environment's characteristics (wind direction and speed) and the internal state (attitude and flapping speed). To analyze how external flow information can specifically contribute to aircraft attitude control, we designed an experimental environment in complex airflow conditions that allows repetitive training to control the aircraft's rotational angles and degrees of freedom (Fig. 1c). The measured wing strain data is transmitted to a RL based controller in real time, and the RL agent is trained to find the optimal policy guaranteeing the highest reward via a RL algorithm. As a result, the RL- based flight controller can recognize different states and control the thrust and direction of the drone. + +The crack- based strain sensors, comprising metal layers deposited on a polyimide substrate, are very thin ( \(\sim 7.5 \mu \mathrm{m}\) ) and lightweight ( \(\sim 3 \mathrm{mg}\) ) and thus have little effect on the structural deformation of the wing (Fig. 1d right). The inset SEM image shows that channeled + +<--- Page Split ---> + +nanoscale cracks on the metal surface are regularly aligned at intervals of \(3 \mu \mathrm{m}\) , and when the sensor is mechanically deformed, the gap between cracks widens and causes a large change in electrical resistance. Due to nanoscale cracks on the metal layer surfaces, the crack- based sensor has a higher gauge factor ( \(\sim 30,000\) ) and signal- to- noise ratio than conventional strain gauges (Supplementary Figs. 2a- j). The flapping- wing drone uses two motors located on the body (motor 1) and tail (motor 2) to control its motion (Fig. 1d left). Motor 1 was replaced with a commercial DC motor (DCX06M, GPX06 A 15:1, Maxon, Co) known for its high durability and low driving current ( \(< 0.23 \mathrm{A}\) ) to overcome the increased weight and increased resistance of wires. Motor 1 is combined with a scotch yoke mechanism that converts the rotational motion into flapping motion to generate thrust force, and motor 2 controls the direction of the drone by steering the tail wing. When a positive voltage is applied to the motor 2, the tail wing is rotated to the clockwise, thus turning the drone counterclockwise. The attachment of crack- based strain sensors to a drone's flapping wings allows for the precise and stable measurement of trajectory changes and the aerodynamic forces exerted on the wing over a long time (Supplementary Fig. 3 and Supplementary Video 1). + +## Validation as state informative sensor system + +To analyze how accurately the wing deformation signal contains information about changes in wind direction and speed, we designed an experiment that the orientation of the drone was systematically controlled, allowing us to create scenarios with winds blowing from multiple directions and varying in speed. In the experiment, the relative wind direction was changed by controlling the drone's orientation. The drone rotated according to different combinations of yaw, pitch and roll angles, resulting in a total of 62 different wind vectors (Fig. 2a and Supplementary Video 2). Also, speed of wind varied in 3 steps, resulting in total 186 cases of blowing wind (Fig. 2b, Supplementary Table 1). The strain sensors attached to the wing base measured the strain, which was affected by the complex aerodynamic loads on the flapping wings. Not only the sensor signal changes with flapping- wing power (Extended Data Fig. 1), but it also varies with the direction and speed of the blowing wind (Extended Data Fig. 2). In this way, various information from environment is encoded into sensor signals. In order to decode the information, we used temporal convolutional neural network (1D CNN) and classified the direction and speed of the blowing wind. The raw sensor signals are given as input to the model, and the model predicts the wind direction and speed as + +<--- Page Split ---> + +outputs (Supplementary Fig. 4). + +After training, we verified the feasibility of state recognition using only strain information based on calculated error from trained regression (Fig. 2c) and confusion matrix from the trained classification model (Fig. 2d). Out of a total of 12,416 validation cases, the mean absolute value of theta error \((\theta_{error})\) was estimated to be 29 degrees. Specifically, \(76\%\) of the angle errors fell inside the range of 0 to 33 degrees. In addition, the mean value of speed error \((s_{error})\) was calculated as \(26\%\) , and \(63.5\%\) of the speed errors fell within the range of \(0 - 30\%\) of speed error. For classification model, the mean AUC was approximately 0.99 (Supplementary Fig. 4c), and the model classified the wind direction and speed with a mean accuracy of \(80\%\) (Fig. 2d), that means the model can predict the direction and speed of the blowing wind with about \(80\%\) of accuracy from 186 cases. In addition, the prediction accuracy is displayed as a color map in spherical coordinates (Fig. 2e). As shown in the color map, the prediction accuracy was highest in the front, lower left, and lower right regions, while the accuracy was the lowest in the upper rear region (Supplementary Fig. 5). This is because the blowing wind from front, left, and right regions made the greatest mechanical deformation at the wing base. These results confirm that the drone can determine the wind direction and speed for various airflows, especially when the front, left, and right sides are windward, by using only the strain information collected by the two sensors on the wing base. + +## 1 DOF control experiment + +Next, we designed an experimental setup that limited the drone's flight path to a single degree- of- freedom (DOF), to confirm its ability to utilize strain data for real- time adaptation to varying wind direction and speed, employing a RL controller (Soft- Actor- Critic algorithm; described in detail at Methods). The drone's body was anchored to a rotational encoder for simplified circular motion and rotated clockwise in x- y plane as the flapping power increased. In this setup, the direction and speed of the apparent wind acting on each wing changed according to the motor encoder's rotation angle \((\alpha)\) . At the initial position, the wind blew from the rear of the drone at a speed of \(3 \mathrm{m / s}\) , which is not enough to move the drone, but enough to assist it after it starts to fly. When thrusting force was generated by the flapping motion, a torsion spring dragged the drone back to the initial position. After the drone overcame the drag and passed over 90 degrees of rotational angle \((\alpha)\) , it had to + +<--- Page Split ---> + +overcome a faster wind speed (5 m/s), that makes the drone more difficult to reach the target position. The sensor signal at the wing base was used as state information in the RL algorithm. The motor encoder at the center measured the rotational angle change, which was used only to calculate the reward during training (Fig. 3a and Supplementary Fig. 6a). The angle of target position was \(180^{\circ}\) , and Fig. 3b shows images from one training episode (Supplementary Video 3). The power increased, then quickly decreased and finally stabilized as the angle fluctuated near \(180^{\circ}\) , as shown in Fig. 3c. Fig. 3d shows that the drone reacted sensitively to the state change. As the episode advanced, the drone chose another strategy and slowly increased and decreased the motor power to approach the target position more smoothly. + +We next compared the performance of the trained drone in these experiments with that of various untrained drones. There are 3 types of untrained drones: flapping in minimum maximum, and random motor power. Compared to the untrained cases, the RL- based drone adapted to the environment and outperformed on accumulated reward (score) as the episodes increase (Supplementary Figs. 6b and 6c). Furthermore, the drone maintained the target position in different angle ( \(270^{\circ}\) ), and even restored the target position when there was dragging- backward- disturbances (Extended Data Fig. 3). When the target position is changed to \(270^{\circ}\) , the only thing that has changed is the reward function. These results show that RL controlled drone can feel the direction of the wind by the strain information and control its flapping frequency in 1 DOF environment, to maintain target position where the highest reward is given. + +## 2 DOF control experiment + +The aim of the next experiment was to verify whether campaniform sensilla aid in controlling gravity- oriented attitudes (pitch angle) for insects, given the unclear nature of whether insects sense gravity during flight control49. Conversely, conventional drones use multi- axis inertial measurement unit (IMU) devices to control attitude. Thus, another experimental setup was devised to validate that the drone can maintain sophisticated pitch angle attitudes. In the experimental environment, drone had to find the optimal pitch angle and maintain balance to increase the rotational angle, which resulted in a higher reward. The body of the drone was anchored to a rotational encoder (rotation joint 1) by a triangle- shaped + +<--- Page Split ---> + +rod, and the center of the body frame was attached to a pin joint (rotation joint 2) so that the drone's pitch angle \((\beta)\) varied from \(- 30^{\circ}\) to \(60^{\circ}\) . Each joint moved separately as the drone's pitch angle changed, regardless of the rotation angle \((\alpha)\) (Fig. 4a and Supplementary Fig. 7a). An apparent wind caused a drag force, pushing the drone backward and creating counterclockwise momentum in rotation joint 2. Because the reward increased proportionately with the rotation angle, the drone received a higher reward when rotation joint 1 rotated counterclockwise. Therefore, the drone had to precisely control the horizontal pitch angle to obtain the highest score during the episode. The drone aimed to simply fly forward through the apparent wind; however, the drone had difficulty maintaining the optimal pitch angle as the rotational angle increased. + +In the experiment, the drone had to detect its own pitch angle based on the data collected by the wing strain sensor. As the drone is constrained to move clockwise or counterclockwise around the center of the motor encoder, only one crack- based sensor attached to the left side of the wing base was enough to acquire state information in the environment. The falling state is defined as the drone fails to maintain balance and the pitch angle decreases to less than \(0^{\circ}\) . In this state, because the direction of the blowing wind is perpendicular to the flapping direction, flapping motion does not assist to increase rotation angle \((\alpha)\) nor pitch angle \((\beta)\) . The only way to restore from this state is to stop flapping and wait for the apparent wind to recover the pitch angle horizontal with the direction of the wind. Fig. 4b shows this recovery process of the trained drone during a disturbance (Supplementary Video 4). As the drone balanced at the optimal pitch angle, the tail of the drone was lifted intentionally to induce the falling state. The drone recognized the falling state transition within 0.1 seconds and decreased its motor power to zero to recover the optimal pitch angle (Fig. 4c and Supplementary Fig. 7b). And accordingly, the amplitude of the sensor signals also decreased in response to the decreased motor power (Fig. 4d). Regardless of the flapping motion, the drone recognized the pitch angle with only one crack- based sensor and quickly recovered to reach a higher rotational angle. + +The score of the RL- agent- controlled drone increased as the number of training episodes increased, whereas the score of the untrained drones was unchanged (Supplementary Fig. 7c). In addition, the trends of the rotational angle and pitch angle of the trained and untrained drones over time clearly differed (Supplementary Figs. 7d and 7e). For the trained drone, the rotational angle converged to approximately \(10^{\circ}\) , where the highest reward is given + +<--- Page Split ---> + +in the designed environment, and the drone also maintained a pitch angle of approximately \(10^{\circ}\) that is optimal for flying in the higher rotational angle. In contrast, the rotational and pitch angles of the untrained model did not converge and instead fluctuated randomly. As the rotational angle \((\alpha)\) and the drone's pitch angle \((\beta)\) change in each time, direction of drag force by blowing wind changes very slightly and frequently, making the drone difficult to fly in the direction of increasing rotational angle. Despite this difficulty, the trained drone recognized the subtle change in state well and successfully flew forward through the apparent wind to get the highest reward during a given episode (Extended Data Figs. 4a- d and Supplementary Video 5). Moreover, even if disturbances were intentionally induced, the drone maintained its balance and controlled the motor power with agility (Extended Data Figs. 4e- h and Supplementary Video 6). After each disturbance, the drone rapidly reduced its motor power to maintain the pitch angle and keep the balance for a smooth flight. The response was quick enough for immediate recovery, which ensured that the drone did not remain in the falling state. These results show that the RL based drone can recognize forward range (90 degree) of the blowing wind and learn to get the highest reward during a given episode by maintaining optimal pitch angle in designed environment. + +## Position control in windy environment + +In general, flapping- wing drones require IMU sensor consisting of accelerometer, gyroscope, and magnetometer and optical flow sensors that can measure information in 3D space to achieve position and attitude control \(^{50,51}\) . In contrast, we indirectly measure the 3D space information by determining 4- dimensional information such as the x, y, z position coordinates and speed using only two strain sensors that measure 1- dimensional strain information. To determine the level of flight control that can be achieved by our proposed system, we designed the position and attitude control problem. In this problem, the commercial drone is trained to navigate to a target position in an unsteady airflow and receives a higher reward as the drone fly to the target position closely. Since the location- dependent airflow velocity is the only cue that the drone can use to identify the target position, the drone must decode the aeroelastic information from the wing deformation in a complex airflow. + +Inside the wind tunnel, we installed an experimental setup, consisting of motion + +<--- Page Split ---> + +capture cameras for measuring position information and a jig to secure the drone from damage due to collisions during repetitive training (Fig. 5a). Since the drone was passively aligned forward by the direction of the airflow, it could not hover without proper control. The drone's position information, measured by the motion capture cameras, was not used as a control input, but was used as a variable in the reward function to provide a reward for problem training (Supplementary Fig. 8). Whether during training or after training, the drone flies in a blind state without the aid of vision sensors. To generate an airflow to provide a positional cue, three small fans were installed at the entrance of the wind tunnel and one big fan was installed at the outlet. The airflow inside the wind tunnel was analyzed using an optical flow estimation method, and the complex asymmetric flow was visualized through color maps representing the wind direction and speed (Supplementary Fig. 9 and Supplementary Video 7). + +Fig. 5b shows a short period (between 7.7 and 9.4 seconds) during training as the drone flies to the middle target position (Supplementary Video 8). The motion of the drone can be explained by dividing it into four cases: clockwise rotation, counterclockwise rotation, acceleration, and deceleration. The drone's movement along the x- axis, signifying its acceleration and deceleration, is primarily influenced by the output changes of motor 1 that regulate the thrust, whereas its rotation along the y- axis is mainly controlled by the output changes of motor 2 that regulate the direction. Fig. 5c shows the trajectory of the drone over time in 3D space, facilitating a comparative analysis with the change in the output of the motors. The trained RL- based controller adjusts the output of the motors based on the wing deformation measured by the strain sensors to determine the optimal path with the highest reward (Fig. 5d). For example, during flight towards the target position, the drone increases the thrust duty cycle between intervals 1 and 2 to accelerate, increases the direction duty cycle between intervals 2 and 3 to rotate clockwise, and repetitively adjusts the trust and direction duty cycles between intervals 3 and 4 to maintain its position at the target position, demonstrating its ability to interpret information about the target position through the aerodynamic forces generated by the interaction of its two wings with the airflow. + +To validate the autonomous flight control results discussed above, Extended Data Fig. 5 displays further cases, such as training the drone to reach alternative target positions and comparing the outcomes with those of untrained drones. Target positions were tested on both the left and right sides of the drone to evaluate position control capability. While the drone + +<--- Page Split ---> + +was capable of detecting and approaching the target position in both instances, there was a significant discrepancy in control performance between the two (Supplementary Videos 9 and 10). In order to make a quantitative comparison of control performance, we used a 2D heat map of the drone's trajectory in the XY plane with a grid dimension of \(50\mathrm{mm}\) for both width and length. By comparing the areas of the heatmaps, it was discovered that when trained with the target on the left, the area was \(\sim 1.6\) times smaller than when trained with the target on the right (Extended Data Figs. 5a and 5c). Moreover, we observed that there were larger fluctuations in the output of both the trust and direction duty cycles when trained with the target on the right (Extended Data Figs. 5b and 5d), and in the 3D trajectory graphs, it was apparent that the scatter of the drone's trajectory was much larger when trained with the target on the right (Extended Data Figs. 5e and 5f). However, considering that both cases showed trust and direction duty cycles producing outputs close to their maximum, and the location of the black grids with the highest visitation frequency in the heat map was behind the target position, it can be inferred that the drone is unable to overcome the resistance caused by airflow due to hardware limitations and the velocity of the right side of the flow field is faster. + +To demonstrate the effectiveness of our proposed control strategy, we conducted a performance comparison between a trained drone and an untrained drone, with the target position set to the middle (Extended Data Fig. 6). The untrained drone showed a disorganized flight pattern, as shown in the heat map and 3D plot of its trajectory (Extended Data Figs. 6a and 6b, and Supplementary Video 11). In contrast, the trained drone repeatedly maintained its proximity to the target position (Extended Data Figs. 6d and 6e, and Supplementary Video 12). For obvious comparison, we displayed the frequency distribution of each drone within a radius of \(20 - 30\mathrm{cm}\) from the target position (Extended Data Figs. 6c and 6f). During a total episode that lasted 16 seconds, it was observed that the untrained drone stayed in the region for a duration of 1.08 seconds, which accounts for approximately \(6\%\) of the total steps taken. On the other hand, the trained drone was in the area for a duration of 5.22 seconds, indicating almost 29 percent of the total steps taken. Average position of the untrained and the trained is also compared (Extended Data Fig. 6g). We conducted a comparison of the mean value of position in the x- y plane. The average position of the untrained drone is located at coordinates \((20\mathrm{mm}, 70\mathrm{mm})\) exhibiting a standard deviation of \((315\mathrm{mm}, 280\mathrm{mm})\) . On the other hand, the average position of the trained drone is found at coordinates \((150\mathrm{mm}, 200\mathrm{mm})\) displaying a standard deviation of \((110\mathrm{mm}, 130\mathrm{mm})\) . Moreover, as the number of training epochs for the trained drone increased, the entropy and score decreased and increased, + +<--- Page Split ---> + +respectively, indicating a reduction in the uncertainty of the output value and successful training of the RL controller model (Extended Data Figs. 6e and 6f). In summary, the effectiveness of the proposed flapping-wing drone control system was confirmed through experimental testing in a wind tunnel, utilizing only two strain sensors to indirectly measure 3D space information for successful position and attitude control without the vision sensors and IMU. + +## Flight path control in a windless environment + +Based on experiments such as wind perception, 1DOF control, 2DOF control, and position control in wind tunnel, we systematically expanded the control utility and principles for implementing the fly- by- feel system. To demonstrate the stable controllability of the fly- by- feel system, we conducted a free flight demonstration with various motions only using the two strain sensors in a windless environment. In previous experiments, wind was an apparent clue to create large wing strain, but in a windless environment, only relatively small deformation due to attitude changes are the information for state estimation. + +We conducted experiments in an enough space (width: 6m, depth: 8m, height: 4m) for drone's natural flight motion (Fig. 6a left). For reliable training, we designed the drone launcher to ensure the same start conditions such as takeoff speed (Supplementary Fig. 10). In addition, very thin and lightweight wires (40 milligram per meter) were used to minimize the tensile force of the tether applied to the drone. By utilizing 12 sets of motion capture cameras, we were able to adjust the reward based on the drone's trajectory (Extended Data Fig. 7). Our first objective was to manage simple left and right motion during flight in order to verify the reliability of trajectory control (Extended Data Fig. 8). Throughout every training session, the RL agent gradually increases its scores and effectively adjusts the angle of the direction motor. Following this, in order to enhance the complexity of the problem, we attempted a combination of left and right motions to navigate in a zigzag pattern and aimed to achieve a circular flight trajectory. In contrast to previous experiments, we made use of the human demonstration method (Fig. 6a right), creating a Replay buffer with human- controlled episodes so that the RL agent had the opportunity to pick up on the desirable motion of each episode. With the strain sensors, the RL agent was able to control flying in both zigzag and circular motion after training (Fig 6b, d). Achieving precise control of the drone's motion + +<--- Page Split ---> + +proves to be challenging, with only a \(30\%\) success rate when researchers manually control its flight along the desired trajectory. However, the RL agent adeptly adjusted the power of the directional motor, resulting in an increase in score and a decrease in entropy as the episode progressed (Supplementary Fig. 11). + +In addition to controlling the directional motor, the thrust motor power was also trained for adjustment during free flight. Our objective was to train the RL agent to maintain altitude higher or lower than the takeoff altitude. Remarkably, the RL agent successfully orchestrated the thrust motor power and directional motor power to sustain appropriate altitude, achieving higher scores as the episodes progressed (Extended Data Fig. 9). + +Furthermore, we compared the odometry and ground truth of the trajectory (Fig. 6c, e) to confirm that the sensor could predict trajectory in the windless environment. In previous research, the accuracy of odometry has been improved by combining an air flow sensor that can detect the wind speed and direction with a IMU based attitude controllable multirotor52,53. Since wing deformation also carries information about wind speed and direction, we were able to reconstruct the drone's driving path. We used 0.5 seconds of sensor signal as input and extracted ground truth based on local coordinate of drone (Extended Data Fig. 10a left). We randomly divided the collection of episodes into a training set and a validation set, with a ratio of 9:1. Subsequently, we conducted training on a model that possesses the same architecture as the policy network of the RL agent but includes additional dropout layer to prevent overfitting (Extended Data Fig. 10a right). Validation episodes are not utilized during the training of the model. We conducted a total of 5 training sessions, each using different variations of training datasets. During the iteration of training epochs, we observed a decrease in the mean squared error (MSE) for both the training and validation datasets as the epochs progressed (Extended Data Fig. 10b, c). Furthermore, we calculated the root mean square error (RMSE) for the displayed episodes. For the zigzag motion, the RMSE was 587.7 mm in the x-axis, 165.5 mm in the y-axis, and 289.9 mm in the z-axis. On the other hand, for circular motion, the RMSE was 636.2 mm in the x-axis, 202.4 mm in the y-axis, 813.3 mm in the z-axis. Additionally, we performed supplementary experiments to ensure the repeatability, yielding different kind of odometry, such as the s- curve (Extended Data Fig. 10d). In this episode, the RMSE was 376.3 in the x-axis, 181 mm in the y-axis, and 404.5 mm in the z-axis. + +<--- Page Split ---> + +## Discussion + +Inspired by the sensory mechanisms of flying insects, particularly their ability to sense aeroelastic loads via strain- measuring organs known as campaniform sensilla, we developed a wing strain- based flight control system. The bioinspired design and engineering foundation of the proposed control system demonstrated that wing strain information effectively encodes the surrounding environments and the flying orientation during flight, even in noisy condition resulting from high vibration of flapping. We attached ultrasensitive, lightweight crack- based strain sensors to the wings of a drone and measured aeroelastic loads during flight. Although the sensor is lightweight (3 mg for a 1 mm width x 5 mm length) and flexible with a thickness of \(7.5 \mu \mathrm{m}\) and a Young's modulus of \(2.5 \mathrm{GPa}^{54}\) , further miniaturization is necessary for application in subgram drones with wing weights of 3 mg. In terms of miniaturization, crack sensors are advantageous due to their simple structure, which involves metal deposited on a thin plastic film, and can be efficiently fabricated using photolithography or laser ablation. We maintained solely deploying strain sensors within our system to evaluate the effectiveness of these sensors in controlling the flapping- wing drone. Our experimental approach was carefully organized into five distinct phases, aimed at rigorously testing our hypothesis and exploring the role of wing strain sensors in controlling a flapping- wing drone. Each phase provided results that reinforced the viability and ingenuity of utilizing wing strain sensors for autonomous drone flight control. These sensors offer strong real- time feedback for system modifications across different atmospheric conditions and enable advanced control strategies without relying on traditional sensors such as IMUs or vision systems. This highlights their potential to improve drone responsiveness and agility in challenging environments. + +Consequently, the use of just two strain sensors proved adequate for the drone to effectively track wind cues from multiple directions. The strain sensors at each wing's base comprehensively captured various features such as the wind speed and direction, drone's attitude, and flapping speed. In the robot experiments, following the only cue given from the wind, flight tasks such as positioning, balancing, and navigating were performed without employing other flight control sensors such as gyro sensors, accelerometers, and vision sensors, which were necessary in previous drones55. + +In our research, we demonstrated that the fly- by- feel control system could be + +<--- Page Split ---> + +implemented by employing bio- inspired strain sensors and end- to- end RL. The end- to- end RL minimizes the need for manual feature selection and intervention by directly mapping sensory input to control outputs, significantly streamlining the learning process and improving response accuracy in complex scenarios. The fly- by- feel flight control system provides a unique approach to navigation that is not solely reliant on GPS or visual cues. This method can adapt subtle shifts in complex aerodynamics, especially valuable in situations where GPS signals falter or are absent, and where visual systems are hindered by adverse environmental factors. + +By conducting robotic experiments, we confirmed that the wing strain of flapping wings contains information about the direction and speed of the wind. These findings support the hypothesis that flying insects can accurately sense wind direction and speed through the campaniform sensilla on their wings. Although antennae56, hair33,57, and bristles39 are known to be wind receptors, the aeroelastic forces that vary with wind can also serve as wind- sensing cues. However, these organs may have difficulty detecting all wind directions if they are localized to specific parts of the body or can be disturbed by flapping- induced flow vortices. During flight, the ambient wind, apparent wind and flapping motion- induced wind act simultaneously58. If a sensory organ is localized to a specific part of the body, such as antennae, it may have difficulty detecting all wind directions because it may be covered by the body or wings depending on the direction of the ambient wind. Hair and bristles distributed on the wings can be disturbed by the flow vortices induced by the flapping motion. In contrast, wings can obtain spatial- temporal flow information during flight. From this perspective, campaniform sensilla on wings are a strong candidate for organs enabling insects to choose the most favorable wind direction for long- distance migration and maneuver in complex and unpredictable environments. + +Developing a wing strain- based flight control model is very challenging. Natural flyers independently control their wing stroke amplitude, pitch angle amplitude, and phase lag to accomplish various maneuvers59. This kinematics generate and shed vortices into the flow and cause the wing shape to vary. Moreover, the large deformation of the wings results in complex fluid- structure interactions and highly coupled nonlinearities in fluid dynamics, aeroelasticity, flight dynamics, and control systems. Even simplified computational models require significant computational resources and often take several days for each simulation to converge60. While quadrotors, and Quadrupedal robots with rigid body components can + +<--- Page Split ---> + +trained in sim- to- real environments, thereby enhancing their control performance at a generalized level \(^{61 - 63}\) , the flapping wing drone systems are still at a fundamental stage in simulation owing to the deformability of body parts and corresponding complexity of aerodynamics. As we aimed to propose a novel control system for flapping drones inspired by nature, we utilized real- world reinforcement learning as an alternative to address these complexities. Since the proposed RL agent trains the neural network using only the strain signal of the wings recorded in a real- world environment, our scheme eliminates many of the calculations required for fluid- structure interactions in flight control systems. Real- world learning obviously has challenges such as limited sample data, delays in the system actuators or rewards, and reasoning system constraints \(^{64}\) . However, because of their offline domains, compensation for reward delays, and algorithms that can consider environmental constraints, RL schemes can be applied in systems that were previously considered too fragile or expensive for learning- based approaches \(^{65}\) . In our research, we used a low- cost toy drone, which has limited durability, but completed learning by giving a time term for each episode and setting the length of the episode to be as short as 20 seconds. However, with a more durable drone that has a higher degree of control freedom, we expect that more dynamic and high- maneuverability flight techniques can be learned. Future developments in this area could involve using more advanced drones to achieve higher performance or complex flight techniques. + +The campaniform sensilla are hypothesized to play a significant role in sensing gravity in flying insects. Although no dedicated gravity sensing organ has been observed in insects, it is possible that the deformation of the wings could vary based on their orientation relative to gravity. However, to accurately differentiate this deformation from that caused by wind and acceleration, multiple sensors may be required to detect complex spatiotemporal deformations of the wing, such as twisting or buckling. In future research, we intend to verify whether attaching multiple sensors to the drone wings enables gravity detection and advanced reactive autonomous flight, such as hovering and wind- assisted flight, without the use of accelerometers and gyro sensors. + +Overall, we expect the aerobatic fly- by- feel flight control system to be applied in many scenarios in the near future. + +<--- Page Split ---> + +## Methods + +Method overview. We aim to train RL agents that control a flapping drone. The drone must perform various tasks within an airflow, including maintaining a specific target position and executing different types of turns in desired directions. To train such agents, we employed the most straightforward method: directly collecting experience in real- world interactions and then training the agents with the data. + +In detail, we first embedded bioinspired crack- based sensors into a commercial flapping- wing robot. These sensors measure the deformation of the wing base for every time step, allowing agents to perceive the surrounding environment. This system can be viewed as a partially observable Markov decision process (POMDP). To address partial observability, we extracted the state of the robot using a 1- D convolutional neural network (1D CNN). The reward is calculated using a motion capture camera and motor encoder. Subsequently, any off- the- shelf RL algorithm can be applied to modify the policy of the agents to maximize the reward. In the following sections, we provide details about the chosen algorithm, the sensor fabrication process, and the experimental setup. + +## Reinforcement learning algorithm + +Drone control as a POMDP. We define the system as a partially observable Markov decision process (POMDP). A POMDP includes a set of states \(S\) , a set of actions \(A\) , a reward function \(r(s, a)\) , the probability distribution \(s' \sim P(s' | s, a)\) of the next state \(s'\) given current state and action, a set of observations \(\Omega\) , a set of conditional observation probabilities \(O\) , and a discount factor \(\gamma \in [0,1)\) . For every time step, the agent samples an action \(a \in A\) using policy \(a \sim \pi (a | s_{agent})\) . Here, since the agent cannot access the true state \(s\) of the environment, it usually construct \(s_{agent}\) , a belief state about the true state \(s\) using history of the observations. When the agent executes an action, the environment makes the state transition, and the agent receives the next observation \(o'\) . The goal of RL is to find a policy \(\pi (a | s_{agent})\) that maximizes the expected sum of the discounted reward for a given \(POMDP \equiv (S, A, r, P, \Omega , O, \gamma)\) . Since we parametrize \(\pi (a | s_{agent})\) through a neural network with parameters \(\theta\) , we denote the policy as \(\pi_{\theta}\) . Therefore, the goal of RL is to find \(\theta\) that satisfies the following. + +<--- Page Split ---> + +\[\theta = \underset {\theta}{\mathrm{argmax}}\mathbb{E}_{\pi_{\theta}}\left(\sum_{t}\gamma^{t}r_{t}\right)\] + +In our experimentation, we discretized the time into 0.05 second of intervals. At each step, the agent observes \(o_{t}\) from the signal received by the sensor attached to the drone's wing. Then, the agent constructs the agent state \(s_{t}\) based on the most recent 32 observations, which is equivalent to 1.6 seconds of time: + +\[s_{t} = (o_{t},o_{t - 1},\dots,o_{t - 31})\in R^{32}.\] + +The agent determines the action according to \(a_{t}\sim \pi_{\theta}(a_{t}|s_{t})\) . Here, note that \(a_{t}\) is the torque of the motor for sing so the action is continuous, such that \(a_{t}\in [0,1]\) . When \(a_{t}\) is equal to 1, the motor output is maximized, and when it is 0, the motor output is minimized. In all experiments, the discount factor \(\gamma\) is set to 0.98, empirically. Through the experiments, the drone executes an action and then the environment make the transition, so that we can gather the transition \(\tau_{t} = (o_{t},a_{t},r_{t + 1},o_{t + 1})\) for every time step. These experiences are then collected in the replay buffer, and the RL algorithm samples minibatches from the replay buffer to update the policy. Each minibatch of data is consists of 64 transitions. After certain number of updates, we obtain a new parameter \(\theta^{\prime}\) and send the parameter directly to the neural network to replace the previous parameter. This process is repeated until the algorithm converges. + +RL Algorithm. We use a soft actor- critic (SAC) \(^{66}\) algorithm as the off- the- shelf policy gradient algorithm. The SAC algorithm is particularly suitable for our problem because it is an off- policy RL algorithm, inherently sample- efficient, and well- suited for our setup. Widely used on- policy algorithms \(^{67 - 69}\) usually discard the data after performing single gradient updates, off- policy algorithms exhibit better sample efficiency as data can be reused multiple times. High sample efficiency is essential because we are gathering data from the real- world robot, and the amount of data is limited. + +Additionally, the SAC algorithm is known for its robustness since it is trained with a maximum entropy objective \(^{70}\) . \(\mathcal{H}\) of the policy \(\pi\) as following: + +\[\mathcal{H}\big(\pi (\cdot |s)\big) = \mathbb{E}_{a_{t}\sim \pi}\big(-log\pi (a_{t}|s_{t})\big)\] + +Since \(\mathcal{H}\) measures the randomness of the policy, maximizing \(\mathcal{H}\) indicates that the algorithm is in favor of stochastic policy. This ensures that the algorithm finds the policy as random as + +<--- Page Split ---> + +577 possible while achieving high rewards. More concretely, SAC aims to maximize the sum of both reward and the entropy of the policy for each time step. + +\[\theta_{s a c} = \underset {\theta}{\mathrm{argmax}}\mathbb{E}_{\pi_{\theta}}\left(\sum_{t}\gamma^{t}[r_{t} + \alpha \mathcal{H}(\pi (\cdot |s_{t}))]\right)\] + +Note that a temperature parameter \(\alpha\) is introduced to control the relative importance of the entropy term against the reward term. With the objective given, we update following parameters: \(\psi\) for the soft state value function \(V_{\psi}\) , \(\phi\) for the soft action value function \(Q_{\phi}\) , and \(\theta\) for the policy \(\pi_{\theta}\) . Here, \(V_{\psi}(s)\) is the expected sum of reward and entropy given state \(s\) , and \(Q_{\phi}(s, a)\) is the expected sum of reward and entropy given state \(s\) and action \(a\) . Both functions assume the future is sampled using policy \(\pi\) . We first update \(\psi\) to minimize the squared residual error: + +\[J_{V}(\psi) = \mathbb{E}_{s_{t}\sim D}\left[\frac{1}{2} (V_{\psi}(s_{t}) - \mathbb{E}_{a_{t}\sim \pi_{\theta}}[Q_{\phi}(s_{t},a_{t}) - log\pi_{\theta}(a_{t}|s_{t}))]^{2}\right]\] + +where \(D\) refers to the distribution of states and actions in the replay buffer. We then update \(\phi\) to minimize the soft Bellman residual \(J_{Q}(\phi)\) as follows: + +\[J_{Q}(\phi) = \mathbb{E}_{(s_{t},a_{t})\sim D}\left[\frac{1}{2} (Q_{\phi}(s_{t},a_{t}) - \hat{Q} (s_{t},a_{t}))^{2}\right],\] + +where \(\hat{Q} (s_{t},a_{t}) = r(s_{t},a_{t}) + \gamma \mathbb{E}_{s_{t + 1}}[V_{\bar{\psi}}(s_{t + 1})]\) . + +Here, \(\bar{\psi}\) denotes the exponential moving average of \(\psi\) , is introduced to stabilize the learning process. Finally, we update \(\theta\) to minimize the Kullback- Leibler (KL) divergence between current policy and the target density defined by Q- function as follows: + +\[\mathbb{E}_{s_{t}\sim D}\left[D_{K L}\left(\pi_{\theta}(s_{t})\right)\left|\frac{e x p\left(Q_{\phi}(s_{t},a_{t})\right)}{Z_{\phi}(s_{t})}\right)\right]\] + +\(Z_{\phi}(s_{t})\) is a partition function that normalizes the distribution \(exp(Q_{\phi}(s_{t},a_{t}))\) . More details of the algorithm can be found in the original paper66. + +Neural network architecture and training details. We use a modified version of SAC algorithm that is working only with \(\pi_{\theta}\) and \(Q_{\phi}\) 71. In this version of the algorithm, the value of \(\alpha\) can be trained concurrently by setting the target entropy explicitly. We use a 1D CNN architecture to properly encode residing patterns from sensor signals across time domain. In + +<--- Page Split ---> + +detail, \(s_t \in \mathbb{R}^{32}\) is fed into the neural network \(\pi_\theta\) , and then the two consecutive 1D CNN layers and a max pooling layer are applied in the time dimension. The 1D CNN layers employ 32 filters with width 5, and the pooling layer uses a window with width 2. The output tensor is flattened into a 1D vector and embedded into a 256- dimensional vector through a linear layer. Finally, the network outputs mean \(\mu\) and variance \(\sigma^2\) of the gaussian distribution. We sample final actions from the gaussian distribution. The action value function \(Q_\phi\) basically has the same structure but with an additional concatenation layer added right before the final output layer. We embed the action into a 64- dimensional vector and concatenate it with the flattened output of the CNN layers. \(\pi_\theta\) and \(Q_\phi\) have separate network parameters. A nonlinear ReLU function \(^{72}\) is applied to the outputs of all convolutional and linear layers. Other detailed hyperparameters are listed in Extended Data Table 2. + +## Equipping the sensors on the flapping-wing drone + +The crack- based sensor was fabricated in the same way as in previous studies \(^{46,73}\) . After covering the shadow mask on a \(7.5 \mu \mathrm{m}\) PI film substrate (3022- 5 Kapton thin film, Chemplex, Co), \(50 \mathrm{nm}\) chrome (Cr) and \(20 \mathrm{nm}\) gold (Au) thin films were deposited at a rate of \(0.5 \mathrm{\AA / s}\) at \(1.2 \times 10^{- 5} \mathrm{mTorr}\) using a thermal evaporator (DDHT- SB015, Dae Dong Hitech, Co). The deposited substrate was cut at \(5 \mathrm{mm}\) intervals and stretched \(2\%\) using a material tester (3342 UTM, Instron, Co) to generate nanoscale cracks on the surface, as this facilitated a high gauge factor of \(\sim 30,000\) at \(2\%\) strain. To measure the wind deformation, we equipped the fabricated crack- based sensors on the wings of a flapping drone (MetaFly, XTIM, Co). The measured weight and wingspan of the drone are \(10 \mathrm{g}\) and \(13.5 \mathrm{cm}\) , respectively. The PDMS dry transfer method \(^{74}\) was used to accurately attach the sensors to the thin wing base, which had a width of \(1 \mathrm{mm}\) , and a commercial strain gauge adhesive (CN adhesive, Tokyo Sokki Kenkyujo, Co) was used for strong and conformal bonding. Conductive epoxy (CW2400, Chemtronics, Co) was used for electrical connection between the sensor and wires, and commercial epoxy adhesive (EE- 05, Axia, Co) was applied to reduce the strain on the wiring interface. We used modified strip theory based on blade elemental analysis to calculate the lift and trust force using the wing strain (Supplementary Fig. 12). + +## Signal processing + +<--- Page Split ---> + +We acquired the strain sensor signal using DAQ: DEwESoft SIRIUS. We used a bandpass filter over the range 0.1\~55 Hz and normalized the signal to 0\~1 to increase the RL agent learning efficiency. All data from the DAQ system and motion capture camera module were transmitted to the desktop computer via serial communication and processed using a Python- language- based RL algorithm (PyTorch). + +## Wind prediction experiment (Fig. 2) + +Simulating 186 wind cases. The flapping- wing drone was anchored to a 3 DOF motor system with 3 motors bound along the x, y, and z axes. As shown in Fig. 4a, wind was generated by an electric fan, and the drone rotated with various combinations of yaw, pitch, and roll angles (Supplementary Figure 5- 10). A Teensy 4.0 microcontroller and L298N motor drivers were used. With this system, 62 wind directions were generated, as depicted in Fig. 4b. Three wind speeds were considered: 3 m/s, 5 m/s, and 7 m/s. The speed variation is expressed as colored arrows: 3 m/s in blue, 5 m/s in green, and 7 m/s in red. Three flapping powers were considered: 60%, 80% and 100% of the motor maximum. However, these powers did not affect the number of classification- labels. As shown in the sub yellow blocks in Fig. 4b, the signal varied depending on the wind direction and speed. + +Neural network. For 186 wind cases classification, we used neural network composed of two 1D convolutional layers followed by two fully connected layers. Each convolutional layer included 5 filters with size 50, yielding a total of 64 filters. The ReLU \(^{72}\) activation function and a max pooling layer were applied to the output of each convolutional layer (Supplementary Fig. 4b). For the fully connected layers, dropout was applied to prevent overfitting. The final output was transformed to the shape of a probability distribution with the softmax function, and cross entropy was used as the loss function. The model received 0.2 seconds of the sensor signal as input. For regression, there was difference in the number of nodes in fully connected layers and final output layers (Supplementary Fig. 4d). + +Calculating prediction errors. We used the inverse trigonometric function of cosine to determine the angle by dividing the dot product by the product of the magnitudes of each vector ( \(\vec{a}\) is prediction vector, and \(\vec{b}\) is ground vector): + +\[\theta_{error} = cos^{-1}\left(\frac{a\cdot b}{|a||b|}\right)\left(\vec{a} = (x_1i, y_1j, z_1k), \vec{b} = (x_2i, y_2j, z_2k)\right)\] + +<--- Page Split ---> + +On the other hand, we represented the speed error as a percentage, which is calculated by dividing the difference between the predicted vector magnitude and the ground vector magnitude by the ground vector magnitude: + +\[s_{error} = \left(\left|\frac{|a| - |b|}{|b|}\right|\right)\times 100\left(\vec{a} = (x_{1}i,y_{1}j,z_{1}k),\vec{b} = (x_{2}i,y_{2}j,z_{2}k)\right)\] + +## Experimental setup + +Wind tunnel. A canopy tent with a width, length, and height of \(1.5\mathrm{m}\) , \(3\mathrm{m}\) , and \(2.8\mathrm{m}\) , respectively, was installed as an indoor wind tunnel. The outlet and inlet were made by cutting the sidewall of the canopy tent. The gap between the ground and tent was sealed with tape to prevent air flow leakage. To generate air flow in the wind tunnel, a large air circulator (Shinil Electronics Co., Ltd) was placed at the outlet, and three small air circulators (Unimax Co, Ltd) were placed at the inlet. An asymmetric and complex air flow was formed by controlling the output of the air circulators. To analyze the airflow inside the wind tunnel, smoke was generated behind the air circulator using a smoke machine, and the turbulence flow was visualized with the smoke converted into linear motion by passing through a honeycomb lattice. + +1 DOF experiment (Fig. 3). The drone was fixed at a right angle to the end of a \(20\mathrm{cm}\) long stick, and the other end of the stick was fixed with a rotary joint (which was also an encoder). As a result, the drone moved in a circular motion when flapping its wings. The movement range of the drone was \(0^{\circ}\) to \(350^{\circ}\) . When the drone was not flapping its wings, a torsion spring allowed the drone to return to its original position. Constant wind in the - Y direction was generated by the two fans at speeds of \(5\mathrm{m / s}\) (upward) and \(3\mathrm{m / s}\) (downward). The drone was controlled by a DC motor that produced flapping motion, and the drone's tail wing was removed. Sensors were attached at the wing base where the stress was largest. The state provided to the RL controller was the final 0.5 seconds of the strain signals collected by the sensors on both wings and the power of the motor used by the drone to control itself. The controller's decision period was 0.05 that is an empirically determined value to handle rapid flow changes around the drone. So, each 10 second episode included 200 decisions. We used a Gaussian distribution function with 180 as the median for the reward function, and accordingly, the drone received the highest score when it maintained its position at \(180^{\circ}\) . + +<--- Page Split ---> + +2 DOF experiment (Fig. 4). The drone was anchored at a pin joint so that the drone's pitch angle varied from \(- 30^{\circ}\) to \(60^{\circ}\) . A triangle-shaped rod was attached to the upper encoder, and the reward was determined by the rotational angle of the rod. \(5\mathrm{m / s}\) of wind blew against the drone, and the drag force of the wind pushed the drone back. This resulted in the pitch angle turning counterclockwise, which allowed the drone to return to its original state. + +Position control in windy environment (Fig. 5). The drone was tethered with a nylon cord to prevent collisions and ensure that it returned to its initial position after each learning episode. The drone had 2 degrees of freedom (DOFs) of control with two motors and flew with 6 DOF movement. Due to the limited DOF control of the drone, the target position was given as \(x\) and \(y\) coordinates. We used a 2D multivariate probability distribution function as the reward function and input the target location as the mean. The SAC algorithm was used for RL, and motion capture cameras (PrimeX 13, Optitrack, Co) were used to provide location information as a reward. + +Flight path control experiments (Fig. 6). In a confined space measuring approximately 6 meters in width, 8 meters in depth, and 4 meters in height, we strategically positioned a total of 12 motion capture cameras and 5 passive markers reflecting infrared rays were attached to the drone. To facilitate the free flight of drones in this space, a dedicated drone launcher was developed (Supplementary Fig. 9). This launcher comprises rail frames, wheels, and a motor ensuring consistent start conditions such as velocity and initial angle of attack of the drone. At the beginning of each episode, the drone on the stage advances at a speed of \(1\mathrm{m / s}\) until it is halted by a mechanical brake. Leveraging the law of the inertia, the mounted drone takes off from the stage, initiating its flight by flapping its wings. During training, the drone was picked up and remounted to the launcher after each episode, and curtains and nets were hung around the walls to prevent collisions and potential damage to the drone. For actuation and data acquisition from the drone, a total of eight copper wires (American Wire Gauge 40, diameter: \(96.5\mu \mathrm{m}\) , length: \(8\mathrm{m}\) ) were utilized. Specifically, 2 wires were allocated for the thrusting motor, 2 for directional motor, and 4 for the two strain sensors attached on the drone's wings. As depicted in Figure 6, the actual dimensions of the drone's flying space are \(4\mathrm{m} \times 6\mathrm{m} \times 2\mathrm{m}\) (width \(x\) length \(x\) height). However, since the tethers are centrally attached to the floor, the operational space is effectively reduced to \(2\mathrm{m} \times 3\mathrm{m} \times 2\mathrm{m}\) (width \(x\) length \(x\) height), which is the area where the drone and tethers interact. This configuration minimizes the tether length, consequently reducing the tensile force exerted on the drone and preventing + +<--- Page Split ---> + +loss of thrust, thereby enhancing flight efficiency. With this arrangement, the maximum length of each tether is approximately \(4.12\mathrm{m}\) . Given the mass of the tether is \(40\mathrm{mg}\) per meter, and with eight wires connected, the total additional weight from the tethers amounts to \(1.32\mathrm{g}\) , which is only \(13.2\%\) of the drone's total weight. This confirms that the mass of the tethers is minimal and does not significantly impact the drone's performance. + +The reward function is designed based on each trajectory of the drone. For leftward flight, the reward is proportionally given according to the displacement in the negative direction of the x- axis and the positive direction of the y- axis. This means the drone receives a larger reward when moving to the left- front direction (Extended Data Fig. 9a). Conversely, in the case of rightward flight, the reward is proportional to the positive displacement along the x- axis and the positive direction of the y- axis (Extended Data Fig. 9b). For a mixed trajectory involving both leftward and rightward flight, such as the zigzag trajectory, the reward is adjusted based on the displacement along the y- axis (Extended Data Fig. 9c). Additionally, we increased the input size as the RL agent needed to consider more earlier standpoints. The reward is initially proportional to the displacement in the negative direction of the x- axis and changes in the opposite direction when the drone passes a certain point along the y- axis. We focused solely on the displacement in the x- axis to induce a dramatic directional change in the drone. Due to inertia, a substantial force is required to alter the drone's direction, making this strategy suitable for achieving our objective. + +Next, we trained the RL agent to control the drone to fly in a circular motion by adjusting the reward function (Extended Data Fig. 9d). The reward is determined proportionally by the dot product of two vectors extracted from sequential positions. Considering the displacement of the drone from its viewpoint, global vectors are rotated according to the local coordinate system of the drone itself. Since our experiment was conducted only in the real world and not in simulation, we fixed the thrust power of the drone to \(80\%\) of the duty cycle to reduce the time required for the RL agent to learn. In this environment, the RL agent is tasked with controlling the angle of the directional motor in the drone's tail. + +We also trained the RL agent to control the altitude of the drone, allowing it to fly at a higher or lower altitude than the starting point of the drone launcher. The reward function designates that the drone receives a higher reward when flying further away from the starting point while maintaining the target altitudes of 1.6 meters and 0.8 meters, respectively. The RL agent had to manage both thrust motor power and directional motor power to uphold desired + +<--- Page Split ---> + +altitude in this environment. + +## Acknowledgements + +D.K., S.H. and J.- S.K. acknowledge financial support from the Ajou University research fund. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (2021R1A6A3A01087289, 2022R1A2C2093100) and Korea Environment Industry & Technology Institute (KEITI) through Digital Infrastructure Building Project for Monitoring, Surveying and Evaluating the Environmental Health Program, funded by Korea Ministry of Environment (MOE) (2021003330009). + +## Author contributions + +T.K., I.H., S.I., S.R., J.- S.K., S.H., and D.K. led the developed of the concepts, designed the experiments, interpreted results, and wrote the paper. T.K., I.H., S.I. designed the experimental setups and S.R., T.K., and S.I. developed codes for reinforcement learning algorithm system. I.H and M.K. provided strain sensors on the flapping drone with support from Y.R., C.K., J.P., D.L., D.L., and S.L. as fabrication of the sensors. J.L. supported to design embedded circuit system and I.B. supported to build Motion Capture tracking system of the flapping drone. J.J. and Y.- M.C supported to build free flight experiments, and S.I. analyzed odometry of the drone in free flight with support from M.R.H. and S.K. In addition, J.L. checked for the reinforcement learning results, and S.S. contributed to progress the experiments. U.K. provided the kinematics ideas for implementing flapping motion of the drone. + +## Competing interests + +The authors declare no competing interests. + +## Data availability + +The data that support the findings of this study are available from the corresponding author upon reasonable request. + +<--- Page Split ---> + +## Code availability + +Codes for the reinforcement learning used in this study are available on following GitHub (https://github.com/TaewiKim/Fly- by- Feel). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1: Fly-by-feel-based flight control strategy.
+ +a, Mechanism for detecting wing deformation in death's head hawkmoth and flapping drone. Death's head hawkmoth : spike rate change via campaniform sensilla in response to wing movement. Flapping drone : resistance change of crack- based sensor according to wing flapping. b, The state information that can be derived from the wing deformation measured by the crack- based strain sensors in the fly- by- feel system. c, The reinforcement learning- based flight control system using wing strain information. d, Photograph of a flapping drone equipped with crack- based strain sensors on the wing bases and two motors on the body and tail for thrust and direction control, respectively. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: Aerodynamic feature extraction from wing strain data.
+ +a, Experimental design for changing the wind direction and speed. b, 186 sets of wing strain data were acquired by changing the wind direction and speed. c, Histogram according to the error range, Angle error and speed error. d, Confusion matrix from the test dataset. e, Visualization of the classification accuracy as a color map in spherical coordinates. Trained model well classified the red-colored range. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: Position control by sensing the wind direction and speed, which vary depending on the position.
+ +a, Schematic illustration of the environment and agent in the designed experiment with 1 DOF. 2 sensor signals and thrust motor power were used as input, while the rotation angle of the encoder was used as variable of reward function. b, Snapshots of an example episode in which a drone controls the motor power to maintain a target position. c, Strain sensor signal (top), motor power (middle) and drone angle (bottom). The range in the red box shows the signals in the snapshots in (b). After the drone passes over the target, it stops flapping to get target position. d, Magnified view of the black dotted box in (c). Trained drone reacts with agility as sensor signals change. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4: Reinforcement learning in a 2 DOF environment.
+ +a, Schematic illustration of the environment and agent in the designed experiment with 2 DOFs. 1 sensor signal was used as input, while the rotation angle of the encoder was used as variable of reward function. b, Snapshots of a representative episode. The drone controls the motor power despite the induced disturbance. c, Sensor signal (top), motor power (middle), and rotational and pitch angles of the drone (bottom). The range in the red box shows the signal in the snapshots in (b). As the trained drone recognizes the falling state, it stops flapping to recover proper pitch angle. d, Magnified view of the black dotted box in (c). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5: Position control of a flapping-wing drone with the proposed fly-by-feel system.
+ +a, Wind tunnel experiment setup to train a tethered flapping-wing drone and the training system configuration. b, Representative images of a flapping drone flying to the target position based on only information obtained by the strain sensors attached to the wings. c, Trajectory of the drone in x- y- z space. The black line is the flight trajectory before 7.5 seconds. d, Changes in wing strain and the output of motors of 7.5 to 9.5 seconds during the target positioning flight. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6: Flight path control of a flapping-wing drone in windless environment.
+ +a, Windless environmental setup designed for training a flapping-wing drone and the corresponding training system configuration. b, Sequential snapshots from a video illustrating the flight of a flapping-wing drone in a zigzag trajectory, relying solely on information acquired through strain sensors attached to its wings. c, Comparison between odometry and ground truth trajectory of the drone in zigzag motion. d, Sequential snapshots from a video showing a flapping drone executing a circular trajectory, guided exclusively by information obtained from strain sensors attached to its wings. e, Comparison between odometry and ground truth trajectory of the drone in circular motion. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Extended Data Figure 1. Correlation between the flapping-wing power and sensor signal from the wind. a, Schematic of the experimental setup. b, Real image showing the experimental setup. c, Change in the duty cycle as the motor power was increased from 15\~100%. d, Variation in the thrust force of the drone as the motor power increased. The thrust force was recorded by nano 17 fixed under the flapping drone. e, Variation in the sensor signal as the motor power increased. f, In detail, as the motor power increased, the frequency of a single flapping cycle and signal amplitude both increased.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + + +Extended Data Figure 2. Signal changes with changing wind direction. a, Five representative wind directions on the front side. b, Sensor signals according to each wind direction on the front side. In the head-blowing scenario, the heights of the signal peaks of the left and right sensors are similar. c, Five representative wind directions on the left side. d, Corresponding sensor signals of each direction on the left side. The biased wind direction causes more strain on the left wing. This is reflected in the sensor signal, as the height of the left wing signal is greater than that of the right wing signal. e, Five representative wind directions on the right side. f, Corresponding sensor signals of each direction on the right side. In contrast to the results of wind blowing on the left side, the height of the right wing signal is greater than that of the left wing signal. + +<--- Page Split ---> +![](images/Figure_10.jpg) + + +![PLACEHOLDER_37_1] + +
Extended Data Figure 3. Additional examples of the 1 DOF system. a, Snapshots of the episode when the target angle is \(180^{\circ}\) . It moves quickly to reach target position and reduces its motor power when it passes over the target. b, The case when the target position angle is \(270^{\circ}\) . When the target position is changed, the drone could recognize the position based on the sensor information. c, The case when the researcher disturbs the flight by pushing the drone backward and forward. Even if there is an intentional disturbance, a trained drone controls its motor power to maintains the target position. d, Motor power control and corresponding angle change for case 1. e, Motor power control and corresponding angle change for case 2. f, Motor power control and corresponding angle change for case 3.
+ +<--- Page Split ---> +![PLACEHOLDER_38_0] + + +Extended Data Figure 4. Extended episode for optimal flight learning. a, Snapshots of free flight in the 2 DOF system. The trained drone balanced to maintain proper pitch angle during flying. b, Signal amplitude of the crack- based sensor attached on the left wing's base. c, Change in the drone's motor power. The drone fine- tunes the power to sustain the optimal pitch angle \((\beta)\) . d, Changes in the rotational angle \((\alpha)\) and pitch angle \((\beta)\) . During flight, drone maintains optimal pitch angle and so that could get the highest rotation angle (approximately \(10^{\circ}\) ). e, Snapshots of a disturbed flight in the 2 DOF system. Even if there is an intentional disturbance, a trained drone reduces its motor power to avoid falling state. f, Variation in the signal amplitude, synchronized with (e). The perturbation events are marked with red circles. g, Power of the motor during the disturbed flight. After the perturbation events, the drone immediately reduces its power to maintain balance. h, Rotational angle \((\alpha)\) and pitch angle \((\beta)\) during the disturbed flight. The drone recovers the optimal angles quickly after the perturbation events. + +<--- Page Split ---> +![PLACEHOLDER_39_0] + + +Extended Data Figure 5. Position control of a flapping- wing drone when trained to reach different target positions. a and b, Trajectory and change in the wing deformation and the output of motors of a trained drone heading toward the left target position. c and d, Trajectory and change in the wing deformation and the output of motors of a trained drone headed toward the right target position 3 (fast airflow). e and f, 3D trajectories of the trained drone heading toward target position (e : left target position, f : right target position). + +<--- Page Split ---> +![PLACEHOLDER_40_0] + + +Extended Data Figure 6. Comparison of an untrained and trained drone. a and b, Plane and space trajectories of the untrained drone heading toward middle target position. c, Frequency distribution of untrained drone within a radius of 20\~30cm from the target point. d and e, Plane and space trajectories of the trained drone heading toward middle target position. f, Frequency distribution of trained drone within a radius of 20\~30cm from the target point. g, Comparison of mean value of the position from untrained and trained drone. h and i, Changes in the entropy and score of the artificial neural network with the number of training epochs. + +<--- Page Split ---> +![PLACEHOLDER_41_0] + + +Extended Data Figure 7. Reward functions for flight trajectory control experiments. a, Reward function designated for leftward flight. b, Reward function designated for rightward flight. c, Reward function designated for zigzag trajectory of flight. d, Reward function designated for circular motion flight. e, Reward function designated for higher altitude trajectory of flight. f, Reward function designated for lower altitude trajectory of flight. + +<--- Page Split ---> +![PLACEHOLDER_42_0] + + +Extended Data Figure 8. Results of left and rightward flight control. a, Sequential snapshots from a video depicting the drone's leftward flight. b, Sequential snapshots from a video displaying drone's rightward flight. c, Trajectory of the drone during the leftward flight. d, Scores observed as episode progressed during the training of leftward flight. e, Strain sensor signals during leftward flight and corresponding duty cycle of thrust motor power and directional motor power. f, Trajectory of the drone during rightward flight. g, Scores recorded as the episode progressed during the training of rightward flight. h, Strain sensor signals corresponding to the rightward flight, along with the associated duty cycle of thrust motor power and directional motor power. + +<--- Page Split ---> +![PLACEHOLDER_43_0] + + +Extended Data Figure 9. Results of higher and lower altitude flight control. a, Sequential snapshots from a video illustrating flight at higher altitudes. b, Sequential snapshots from a video displaying drone's flight at lower altitudes. c, Trajectory of the drone during the higher altitude flight. d, Scores recorded as episode progressed during the training of higher altitude flight. e, Strain sensor signals during the higher altitude flight, along with the corresponding duty cycle of thrust motor power and directional motor power. f, Trajectory of the drone during lower altitude flight. g, Scores recorded as the episodes progressed during the training of the lower height intended flight. h, Strain sensor signals corresponding to the lower height flight, along with the associated duty cycle of thrust motor power and directional motor power. + +<--- Page Split ---> +![PLACEHOLDER_44_0] + +
Extended Data Figure 10. Methods of yielding odometry and results. a, Labelling procedure for comparing odometry and ground truth trajectories. b, Comparison of mean squared error between 5 folds of different training datasets. c, Comparison of mean squared error between 5 folds of different validation datasets. d, Representative result of comparison between odometry and ground truth trajectories from additional experiment.
+ +<--- Page Split ---> + + +
HyperparameterValue
Learning rate of π0.0005
Learning rate of q0.001
Learning rate of α0.001
Initial α0.01
γ0.98
τ(for EMA update)0.01
Batch size32
Target entropy-1.0
Maximum buffer size3000
Initial training buffer size1000
Maximum episode length (number of steps)300
Decision period (s)0.05
+ +959 + +960 **Extended Data Table 1. Hyperparameter values.** + +961 + +<--- Page Split ---> + +962 References963 1 Ansari, S., Zbikowski, R. & Knowles, K. Aerodynamic modelling of insect- like flapping flight for micro air vehicles. Progress in aerospace sciences 42, 129- 172 (2006).965 2 Ma, K. Y., Chirarattananon, P., Fuller, S. B. & Wood, R. J. Controlled flight of a biologically inspired, insect- scale robot. Science 340, 603- 607 (2013).966 3 Chen, Y. et al. Controlled flight of a microrobot powered by soft artificial muscles. Nature 575, 324- 329 (2019).967 4 Jafferis, N. T., Helbling, E. F., Karpelson, M. & Wood, R. J. Untethered flight of an insect- sized flapping- wing microscale aerial vehicle. Nature 570, 491- 495 (2019).968 5 De Croon, G., De Clercq, K., Ruijsink, R., Remes, B. & De Wagter, C. Design, aerodynamics, and vision- based control of the DelFly. Int. J. Micro Air Veh. 1, 71- 97 (2009).969 6 Phan, H. V., Kang, T. & Park, H. C. Design and stable flight of a 21 g insect- like tailless flapping wing micro air vehicle with angular rates feedback control. Bioinspiration Biomimetics 12, 036006 (2017).970 7 Zhang, J., Fei, F., Tu, Z. & Deng, X. in 2017 IEEE International Conference on Robotics and Automation (ICRA). 5422- 5428 (IEEE).971 8 Mellinger, D. & Kumar, V. in 2011 IEEE international conference on robotics and automation. 2520- 2525 (IEEE).972 9 Elijah, T., Jamisola, R. S., Tjiparuro, Z. & Namoshe, M. A review on control and maneuvering of cooperative fixed- wing drones. International Journal of Dynamics and Control 9, 1332- 1349 (2021).973 10 de Croon, G. Flapping wing drones show off their skills. Sci. Robot. 5, eabd0233 (2020).974 11 Chirarattananon, P. et al. Dynamics and flight control of a flapping- wing robotic insect in the presence of wind gusts. Interface focus 7, 20160080 (2017).975 12 Baek, S. S., Bermudez, F. L. G. & Fearing, R. S. in 2011 IEEE/RSJ International Conference on Intelligent Robots and Systems. 2674- 2681 (IEEE).976 13 Nguyen, Q.- V. & Chan, W. L. Development and flight performance of a biologically- inspired tailless flapping- wing micro air vehicle with wing stroke plane modulation. Bioinspiration Biomimetics 14, 016015 (2018).977 14 Shin, H. S. et al. Bio- Inspired Large- Area Soft Sensing Skins to Measure UAV Wing Deformation in Flight. Adv. Funct. Mater. 31, 2100679 (2021).978 15 Tong, S., Weiping, Z., Jiawang, M. & Zihao, C. in 2019 IEEE International Conference on Unmanned Systems (ICUS). 842- 847 (IEEE).979 16 Woolf, P. J. Chemical process dynamics and controls. (2009).980 17 Walker, S. M., Thomas, A. L. & Taylor, G. K. Deformable wing kinematics in free- flying hoverflies. J. R. Soc. Interface 7, 131- 142 (2010).981 18 Somps, C. & Luttges, M. Dragonfly flight: novel uses of unsteady separated flows. Science 228, 1326- 1329 (1985).982 19 Wang, H., Zeng, L., Liu, H. & Yin, C. Measuring wing kinematics, flight trajectory and body + +<--- Page Split ---> + +1001 attitude during forward flight and turning maneuvers in dragonflies. J. Exp. Biol. 206, 745- 757 (2003). 1002 Springthorpe, D., Fernández, M. J. & Hedrick, T. L. Neuromuscular control of free-flight yaw turns in the hawkmoth Manduca sexta. J. Exp. Biol. 215, 1766- 1774 (2012). 1005 21 Salami, E., Ward, T. A., Montazer, E. & Ghazali, N. N. N. A review of aerodynamic studies on dragonfly flight. Proc. Inst. Mech. Eng., Part C 233, 6519- 6537 (2019). 1007 Mischiati, M. et al. Internal models direct dragonfly interception steering. Nature 517, 333- 338 (2015). 1009 23 Hedlund, J. S. et al. Unraveling the world's longest non- stop migration: the indian ocean crossing of the globe skimmer dragonfly. Front. Ecol. Evol. 9, 525 (2021). 1011 24 Gao, B. et al. Adaptive strategies of high- flying migratory hoverflies in response to wind currents. Proc. R. Soc. B 287, 20200406 (2020). 1013 25 Cardé, R. T. Animal migration: seasonal reversals of migrant moths. Curr. Biol. 18, R1007- R1009 (2008). 1015 26 Menz, M. H. et al. Individual tracking reveals long- distance flight- path control in a nocturnally migrating moth. Science 377, 764- 768 (2022). 1017 27 Hu, G. et al. Mass seasonal bioflows of high- flying insect migrants. Science 354, 1584- 1587 (2016). 1019 28 Hedlund, J. S. et al. Unraveling the World's longest non- stop migration: The Indian Ocean crossing of the Globe Skimmer Dragonfly. Front. Ecol. Evol., 525 (2021). 1021 29 Frye, M. A. & Dickinson, M. H. Motor output reflects the linear superposition of visual and olfactory inputs in Drosophila. J. Exp. Biol. 207, 123- 131 (2004). 1023 30 Sherman, A. & Dickinson, M. H. Summation of visual and mechanosensory feedback in Drosophila flight control. J. Exp. Biol. 207, 133- 142 (2004). 1025 31 Tuthill, J. C. & Wilson, R. I. Mechanosensation and adaptive motor control in insects. Curr. Biol. 26, R1022- R1038 (2016). 1027 32 Dickerson, B. H., Fox, J. L. & Sponberg, S. Functional diversity from generic encoding in insect campaniform sensilla. Current Opinion in Physiology 19, 194- 203 (2021). 1029 33 Taylor, G. K. & Krapp, H. G. Sensory systems and flight stability: what do insects measure and why? Advances in insect physiology 34, 231- 316 (2007). 1031 34 Combes, S. & Daniel, T. Flexural stiffness in insect wings II. Spatial distribution and dynamic wing bending. J. Exp. Biol. 206, 2989- 2997 (2003). 1033 35 Bomphrey, R. J. & Godoy- Diana, R. Insect and insect- inspired aerodynamics: unsteadiness, structural mechanics and flight control. Curr. Opin. Insect. Sci. 30, 26- 32 (2018). 1035 36 Dickerson, B. H., Aldworth, Z. N. & Daniel, T. L. Control of moth flight posture is mediated by wing mechanosensory feedback. Journal of Experimental Biology 217, 2301- 2308 (2014). 1037 37 Dickinson, M. H. Comparison of encoding properties of campaniform sensilla on the fly wing. Journal of experimental biology 151, 245- 261 (1990). + +<--- Page Split ---> + +1040 38 Pratt, B., Deora, T., Mohren, T. & Daniel, T. Neural evidence supports a dual sensory- motor role for insect wings. Proceedings of the Royal Society B: Biological Sciences 284, 20170969 (2017).1042 39 Fabian, J. et al. Systematic characterization of wing mechanosensors that monitor airflow and wing deformations. iScience 25, 104150 (2022).1045 40 Hinson, B. T. & Morgansen, K. A. Gyroscopic sensing in the wings of the hawkmoth Manduca sexta: the role of sensor location and directional sensitivity. Bioinspiration & biomimetics 10, 056013 (2015).1048 41 Taylor, G. Modelling the effects of unsteady flow phenomena on flapping flight dynamics- stability and control. Flow Phenomena in Nature: A Challenge to Engineering Design 1, 155- 166 (2007).1051 42 Sane, S. P. The aerodynamics of insect flight. Journal of experimental biology 206, 4191- 4208 (2003).1053 43 Chin, D. D. & Lentink, D. Flapping wing aerodynamics: from insects to vertebrates. Journal of Experimental Biology 219, 920- 932 (2016).1055 44 Breuer, K. (Nature Publishing Group, 2019).1056 45 Aiello, B. R. et al. Spatial distribution of campaniform sensilla mechanosensors on wings: form, function, and phylogeny. Current Opinion in Insect Science 48, 8- 17 (2021).1058 46 Kang, D. et al. Ultrasensitive mechanical crack- based sensor inspired by the spider sensory system. Nature 516, 222- 226 (2014).1060 47 Lee, E. et al. Effect of metal thickness on the sensitivity of crack- based sensors. Sensors 18, 2872 (2018).1062 48 Kim, T. et al. Polyimide encapsulation of spider- inspired crack- based sensors for durability improvement. Applied Sciences 8, 367 (2018).1064 49 Bender, J. A. & Frye, M. A. Invertebrate solutions for sensing gravity. Current biology 19, R186- R190 (2009).1066 50 Yang, R., Zhang, W., Mou, J., Zhang, B. & Zhang, Y. in Proceedings of 2022 International Conference on Autonomous Unmanned Systems (ICAUS 2022). 1432- 1443 (Springer).1068 51 Bloesch, M. et al. in 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems. 3102- 3107 (IEEE).1070 52 Tagliabue, A. & How, J. P. in 2021 IEEE International Conference on Robotics and Automation (ICRA). 5736- 5743 (IEEE).1072 53 Tagliabue, A. et al. in 2020 IEEE/RSJ international conference on intelligent robots and systems (IROS). 1645- 1652 (IEEE).1074 54 Kim, T. et al. Ultra- stable and tough bioinspired crack- based tactile sensor for small legged robots. npj Flexible Electronics 7, 22 (2023).1076 55 Pan, E., Liang, X. & Xu, W. Development of vision stabilizing system for a large- scale flapping- wing robotic bird. IEEE Sensors Journal 20, 8017- 8028 (2020).1078 56 Gewecke, M. Antennae: another wind- sensitive receptor in locusts. Nature 225, 1263- 1264 + +<--- Page Split ---> + +(1970). 57 Fratzl, P. & Barth, F. G. Biomaterial systems for mechanosensing and actuation. Nature 462, 442- 448 (2009). 58 van Breugel, F. in 2021 60th IEEE Conference on Decision and Control (CDC). 1399- 1406 (IEEE). 59 Shyy, W. et al. Recent progress in flapping wing aerodynamics and aeroelasticity. Progress in Aerospace Sciences 46, 284- 327 (2010). 60 Schwab, R., Reade, J. & Jankauski, M. Quasi three- dimensional deformable blade element and unsteady vortex lattice reduced- order modeling of fluid- structure interaction in flapping wings. Physics of Fluids 34, 121903 (2022). 61 Cheng, X., Shi, K., Agarwal, A. & Pathak, D. Extreme parkour with legged robots. arXiv preprint arXiv:2309.14341 (2023). 62 Huang, K., Rana, R., Spitzer, A., Shi, G. & Boots, B. Datt: Deep adaptive trajectory tracking for quadrotor control. arXiv preprint arXiv:2310.09053 (2023). 63 Zhuang, Z. et al. Robot parkour learning. arXiv preprint arXiv:2309.05665 (2023). 64 Lee, J., Hwangbo, J., Wellhausen, L., Koltun, V. & Hutter, M. Learning quadrupedal locomotion over challenging terrain. Science robotics 5, eabc5986 (2020). 65 Dulac- Arnold, G. et al. Challenges of real- world reinforcement learning: definitions, benchmarks and analysis. Machine Learning 110, 2419- 2468 (2021). 66 Haarnoja, T., Zhou, A., Abbeel, P. & Levine, S. in International conference on machine learning. 1861- 1870 (PMLR). 67 Mnih, V. et al. in International conference on machine learning. 1928- 1937 (PMLR). 68 Espeholt, L. et al. in International conference on machine learning. 1407- 1416 (PMLR). 69 Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms. arXiv:1707.06347 (2017). 70 Ziebart, B. D. Modeling purposeful adaptive behavior with the principle of maximum causal entropy. (Carnegie Mellon University, 2010). 71 Haarnoja, T. et al. Soft actor- critic algorithms and applications. arXiv preprint arXiv:1812.05905 (2018). 72 Agarap, A. F. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375 (2018). 73 Roh, Y. et al. Vital signal sensing and manipulation of a microscale organ with a multifunctional soft gripper. Science Robotics 6, eabi6774 (2021). 74 Sundar, V. C. et al. Elastomeric transistor stamps: reversible probing of charge transport in organic crystals. Science 303, 1644- 1646 (2004). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +ExtendedDataTable1.tif - SupplementaryVideo.zip - Supplementaryinformationclean.docx - Supplementaryinformationmarkedup.docx + +<--- Page Split ---> diff --git a/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178_det.mmd b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a581dc7bdc6cbf3b93e3939e2d42f902fafcc8d9 --- /dev/null +++ b/preprint/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178/preprint__0e2f50e176f45187ae703bea50f4a7a1087f322fd325fa4713273e388c1d0178_det.mmd @@ -0,0 +1,647 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 857, 210]]<|/det|> +# Fly-by-Feel: Wing Strain-based Flight Control of Flapping-Wing Drones through Reinforcement Learning + +<|ref|>text<|/ref|><|det|>[[44, 230, 245, 276]]<|/det|> +Daeshik Kang dskang@a.jou.ac.kr + +<|ref|>text<|/ref|><|det|>[[44, 301, 550, 330]]<|/det|> +Ajou University https://orcid.org/0000- 0002- 7490- 9079 + +<|ref|>text<|/ref|><|det|>[[44, 333, 550, 372]]<|/det|> +Seungyong Han Ajou University https://orcid.org/0000- 0003- 2870- 4088 + +<|ref|>text<|/ref|><|det|>[[44, 375, 550, 415]]<|/det|> +Je- sung Koh Ajou University https://orcid.org/0000- 0002- 7739- 0882 + +<|ref|>text<|/ref|><|det|>[[44, 419, 550, 459]]<|/det|> +Taewi Kim Ajou University https://orcid.org/0000- 0002- 8570- 348X + +<|ref|>text<|/ref|><|det|>[[44, 464, 188, 503]]<|/det|> +Insic Hong Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 509, 550, 549]]<|/det|> +Sunghoon Im Ajou University https://orcid.org/0000- 0002- 2595- 157X + +<|ref|>text<|/ref|><|det|>[[44, 554, 330, 595]]<|/det|> +Seungeun Rho Georgia Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 600, 188, 639]]<|/det|> +Minho Kim Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 644, 188, 683]]<|/det|> +Yeonwook Roh Ajou University https://orcid.org/0000- 0002- 2823- 2028 + +<|ref|>text<|/ref|><|det|>[[44, 688, 188, 727]]<|/det|> +Changhwan Kim Ajou University https://orcid.org/0000- 0002- 0626- 730X + +<|ref|>text<|/ref|><|det|>[[44, 732, 188, 771]]<|/det|> +Jieun Park Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 777, 188, 815]]<|/det|> +Daseul Lim Ajou University https://orcid.org/0000- 0002- 1990- 9092 + +<|ref|>text<|/ref|><|det|>[[44, 820, 188, 859]]<|/det|> +Doohoe Lee Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 864, 188, 903]]<|/det|> +Seunggon Lee Ajou University https://orcid.org/0000- 0001- 6305- 985X + +<|ref|>text<|/ref|><|det|>[[44, 908, 144, 927]]<|/det|> +Jingoo Lee + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 45, 546, 66]]<|/det|> +Ajou University https://orcid.org/0000- 0002- 5783- 2583 + +<|ref|>text<|/ref|><|det|>[[44, 71, 546, 112]]<|/det|> +Inryeol Back Ajou University https://orcid.org/0000- 0002- 0730- 6585 + +<|ref|>text<|/ref|><|det|>[[44, 117, 515, 159]]<|/det|> +Joonho Lee ETH Zurich https://orcid.org/0000- 0002- 5072- 7385 + +<|ref|>text<|/ref|><|det|>[[44, 163, 245, 204]]<|/det|> +Sungchul Seo Seogyeong University + +<|ref|>text<|/ref|><|det|>[[44, 210, 190, 250]]<|/det|> +Uikyum Kim Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 255, 190, 295]]<|/det|> +Junggwang Cho Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 301, 196, 342]]<|/det|> +Myung Rae Hong Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 347, 196, 387]]<|/det|> +Sanghun Kang Ajou University + +<|ref|>text<|/ref|><|det|>[[44, 392, 190, 433]]<|/det|> +Young-Man Choi Ajou University + +<|ref|>sub_title<|/ref|><|det|>[[44, 476, 104, 494]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 514, 137, 533]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 552, 297, 572]]<|/det|> +Posted Date: May 30th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 590, 475, 610]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4443963/v1 + +<|ref|>text<|/ref|><|det|>[[44, 627, 912, 671]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 688, 535, 709]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 743, 950, 788]]<|/det|> +Version of Record: A version of this preprint was published at Nature Machine Intelligence on September 20th, 2024. See the published version at https://doi.org/10.1038/s42256- 024- 00893- 9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[80, 98, 870, 164]]<|/det|> +# Fly-by-Feel: Wing Strain-based Flight Control of Flapping-Wing Drones through Reinforcement Learning + +<|ref|>text<|/ref|><|det|>[[80, 211, 885, 339]]<|/det|> +Taewi \(\mathrm{Kim}^{1,\dagger}\) , Insic Hong \(^{1,\dagger}\) , Sunghoon \(\mathrm{Im}^{1,\dagger}\) , Seungeun \(\mathrm{Rho}^{2,\dagger}\) , Minho \(\mathrm{Kim}^{1}\) , Yeonwook \(\mathrm{Roh}^{1}\) , Changhwan \(\mathrm{Kim}^{1}\) , Jieun \(\mathrm{Park}^{1}\) , Daseul \(\mathrm{Lim}^{1}\) , Doohoe \(\mathrm{Lee}^{1}\) , Seunggon \(\mathrm{Lee}^{1}\) , Jingo \(\mathrm{Lee}^{1}\) , Inryeol \(\mathrm{Back}^{1}\) , Junggwang \(\mathrm{Cho}^{1}\) , Myung Rae \(\mathrm{Hong}^{1}\) , Sanghun \(\mathrm{Kang}^{1}\) , Joonho \(\mathrm{Lee}^{3}\) , Sungchul \(\mathrm{Seo}^{4}\) , Uikyum \(\mathrm{Kim}^{1}\) , Young- Man \(\mathrm{Choi}^{1}\) , Je- sung \(\mathrm{Koh}^{1,*}\) , Seungyong \(\mathrm{Han}^{1,*}\) , Daeshik \(\mathrm{Kang}^{1,*}\) + +<|ref|>text<|/ref|><|det|>[[80, 397, 718, 417]]<|/det|> +\(^{1}\) Department of Mechanical Engineering, Ajou University, Suwon, Korea + +<|ref|>text<|/ref|><|det|>[[80, 437, 701, 457]]<|/det|> +\(^{2}\) College of Computing, Georgia Institute of Technology, Georgia, USA + +<|ref|>text<|/ref|><|det|>[[80, 477, 579, 497]]<|/det|> +\(^{3}\) Robotic Systems Lab, ETH- Zürich, Zürich, Switzerland + +<|ref|>text<|/ref|><|det|>[[80, 517, 825, 537]]<|/det|> +\(^{4}\) Department of Nanochemical, Biological and Environmental Engineering, Seokyeong + +<|ref|>text<|/ref|><|det|>[[80, 550, 266, 568]]<|/det|> +University, Korea + +<|ref|>text<|/ref|><|det|>[[80, 592, 90, 604]]<|/det|> +14 + +<|ref|>text<|/ref|><|det|>[[80, 631, 510, 650]]<|/det|> +\(^{+}\) These authors contributed equally to this work. + +<|ref|>text<|/ref|><|det|>[[80, 672, 90, 684]]<|/det|> +16 + +<|ref|>text<|/ref|><|det|>[[80, 704, 354, 722]]<|/det|> +\(^{*}\) Address correspondence to: + +<|ref|>text<|/ref|><|det|>[[80, 736, 424, 755]]<|/det|> +Prof. Je- Sung Koh; jskoh@ajou.ac.kr + +<|ref|>text<|/ref|><|det|>[[80, 777, 471, 796]]<|/det|> +Prof. Seungyong Han; sy84han@ajou.ac.kr + +<|ref|>text<|/ref|><|det|>[[80, 817, 447, 836]]<|/det|> +Prof. Daeshik Kang; dskang@ajou.ac.kr + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 101, 222, 122]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[115, 140, 881, 630]]<|/det|> +Although drone technology has progressed significantly, replicating the dynamic control and wind- sensing abilities of biological flights is still beyond our reach. Biological studies have revealed that insect wings are equipped with mechanoreceptors known as campaniform sensilla, which detect complex aerodynamic loads critical for flight agility. By leveraging robotic experiments designed to mimic these biological systems, we confirmed that wing strain provides crucial information about the drone's attitude, as well as the direction and velocity of the wind. We introduce a novel wing strain- based flight controller, termed 'fly- by- feel'. This methodology employs the aerodynamic forces exerted on a flapping drone's wings to deduce vital flight data, such as attitude and airflow without accelerometers and gyroscopic sensors. Our empirical approach spanned five key experiments: initially validating the wing strain sensor system for state information provision, followed by a single degree of freedom (1 DOF) control in changing winds, a two degrees of freedom (2 DOF) control for gravitational attitude adjustment, a test for position control in windy conditions, and finally, demonstrating precise flight path manipulation in a windless condition using only wing strain sensors. We have successfully demonstrated control of a flapping drone in a various environment using only wing strain sensors, with the aid of reinforcement learning- driven flight controller. The fly- by- feel system holds the potential to revolutionize autonomous drone operations, providing enhanced adaptability to environmental shifts. This will be beneficial across varied applications, from gust resistance to wind- assisted flight, paving the way toward the next generation of resilient and autonomous flying robots. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 101, 266, 122]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[115, 140, 882, 505]]<|/det|> +Understanding the aerodynamic mechanism \(^{1 - 3}\) of biological flights has led to remarkable progress in the system design of flapping- wing micro aerial vehicles (FWMAVs) such as RoboBee \(^{4}\) , DelFly \(^{5}\) , KUBeele \(^{6}\) , and Purdue Hummingbird \(^{7}\) . Although these FWMAVs offer high agility and energy efficiency at a small scale over quadcopters \(^{8}\) and fixed- wing drones \(^{9}\) , they encounter difficulties in terms of flight control and lightweight design \(^{10}\) . Specifically, controlling in windy conditions is challenging due to their complex dynamics and small size, which limit to generate sufficient thrust force and moments \(^{11}\) . To control FWMAVs, sensors and feedback system combining inertial measurement units (IMUs), vision cameras, and proportional- integral- derivative (PID) controllers are commonly utilized \(^{12,13}\) . However, they are not sufficient for controlling FWMAVs due to several reasons. Firstly, IMUs are typically located on the bodies, requiring body movement for disturbance detection, which can often cause the delayed stabilization of the robot \(^{14}\) . Second, existing sensors are inadequate to handle the flexible body of flapping wing drones due to their rigidity. Third, model- based controllers, with limited tuning parameters, struggle to manage the complex aerodynamics of flexible flapping wings \(^{15,16}\) . + +<|ref|>text<|/ref|><|det|>[[115, 519, 882, 908]]<|/det|> +Flying insects (e.g. dragonflies, hawk moths, and hoverflies) possess a high level of flight ability, capable of maneuvers like hovering, turning, and gliding, even in unsteady aerodynamic environments \(^{17 - 22}\) . Notably, some species like the globe skimmer dragonfly traverse thousands of kilometers, from India to East Africa \(^{23}\) , by utilizing fast, high- altitude airstreams and navigating seasonally favorable directions \(^{24 - 28}\) . The superior sensing and control mechanism are attributed to active stabilization, supported by complex sensory systems including mechanosensory, vision, and olfaction \(^{29 - 31}\) . Among these sensory organs, the campaniform sensilla, which are strain- detecting mechanosensory distributed on the wings, enable the immediate, proprioceptive flight control \(^{32,33}\) . The wings of flying insects are subject to combined inertial- aerodynamic (aeroelastic) loads because of various aerodynamic phenomena, including airflow stagnation and separation, and nonlinear phenomena such as vortex growth and shedding \(^{34,35}\) . Moreover, recent researches claim that the campaniform sensilla detect the Coriolis force and function as inertial sensors based on electrophysiology data \(^{36 - 38}\) and numerical simulations \(^{39,40}\) . From an engineering perspective, the direct measurement of wing load is a significant advantage for flight control because the unsteadiness in aerodynamics makes it hard for conventional control systems to accurately + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 98, 881, 241]]<|/det|> +estimate the wing load in real- time41. Therefore, inspired by the control systems of these flying insects, attempts have been made to develop controllers that can sensitively respond to changes in airflow by measuring the wing deformation of aircraft14. However, there have been no successful cases of flapping drones because the aerodynamics between easily deformable wings and airflow are extremely complex42,43, and developing sensing and control algorithms for stable flight and manipulation has been difficult, even with supercomputers44. + +<|ref|>text<|/ref|><|det|>[[113, 255, 881, 894]]<|/det|> +In this article, we present a wing strain- based flight controller, which we term fly- by- feel. Given the computational and complexity challenges associated with modeling aerodynamics, we employed the powerful and cutting- edge method of reinforcement learning. Utilizing wing strain sensors, our goal is to assess the feasibility and reliability of such a system in autonomously determining flight paths. Our experimental approach is structured into five distinct phases. 1) Validation as a state informative sensor system: The initial phase of the research focuses on verifying the wing strain sensor's ability to provide useful and reliable state information for drone control. This foundational step is crucial in confirming the sensory system's detection capabilities. 2) Single degree of freedom (1 DOF) flight path control experiment: This stage evaluates how effectively the drone can adapt to changing wind speeds and directions, and how well it can manage one- dimensional flight paths using data from strain sensors. 3) Two degrees of freedom (2 DOF) attitude control experiment: This experiment tests the drone's ability to identify and maintain optimal pitch angles through gravity- based attitude adjustment, which is vital for precise posture control and flight path modification. 4) Position control in windy Environment: This phase tests the drone's capability to maintain its position in a wind- influenced environment solely using wing strain sensor data. Success in this experiment would underline the potential of strain sensors to manage complex aerodynamic conditions without additional sensory inputs. 5) Wing strain- based flight path control in clam environment: The final experimental step involves verifying the drone's ability to control its flight path in a windless environment using only the information provided by wing strain sensors, which would demonstrate the precision of location and attitude adjustments possible with this system. Each stage of this study meticulously verifies different aspects of the system, building upon each other to substantiate the utility and effectiveness of an autonomous flight control system based on strain sensor data. Through this series of empirical tests, we aim to implement the biological hypothesis that wing strain sensors are useful for immediate flight control in response to airflow changes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 880, 142]]<|/det|> +into robotic experiments and explore the potential of wing strain- based control systems in real flight situations. + +<|ref|>sub_title<|/ref|><|det|>[[116, 195, 183, 214]]<|/det|> +## Main + +<|ref|>sub_title<|/ref|><|det|>[[115, 225, 671, 243]]<|/det|> +## Fly-by-feel control system incorporates RL based on wing strain + +<|ref|>text<|/ref|><|det|>[[115, 258, 882, 500]]<|/det|> +Inspired by insects' agile flight abilities, the fly- by- feel control system shown in Fig. 1 enables rapid adaptation to complex airflow conditions and immediate response to sudden fluctuations. The campaniform sensilla are sensory organs distributed on the wing base that sense the strain on the wing45 (Fig. 1a middle, Supplementary Fig. 1a). The spike rate of neurons depends on the flapping motion of the wing, as shown in Fig. 1a left, and the campaniform sensilla can detect the rotational rate during flight38. To develop an efficient autonomous flight system by understanding the role of the Campaniform sensilla in flying insects, ultrasensitive crack- based strain sensors46- 48 were attached to the wing bases of commercial wing- flapping drones inspired by the sensory system of a flying insect (Fig. 1a right). + +<|ref|>text<|/ref|><|det|>[[115, 514, 882, 828]]<|/det|> +When training a drone to fly, the flapping motion of the drone's wings continuously changes the airflow, causing a variety of complex aerodynamic phenomena. The crack- based strain sensors attached to the wing base collect aeroelastic information in various states (Fig. 1b). After analyzing the wing deformation signal with a temporal convolutional neural network (1D- CNN) model, we discovered that the signal contains possible clues for estimating the external environment's characteristics (wind direction and speed) and the internal state (attitude and flapping speed). To analyze how external flow information can specifically contribute to aircraft attitude control, we designed an experimental environment in complex airflow conditions that allows repetitive training to control the aircraft's rotational angles and degrees of freedom (Fig. 1c). The measured wing strain data is transmitted to a RL based controller in real time, and the RL agent is trained to find the optimal policy guaranteeing the highest reward via a RL algorithm. As a result, the RL- based flight controller can recognize different states and control the thrust and direction of the drone. + +<|ref|>text<|/ref|><|det|>[[115, 843, 880, 911]]<|/det|> +The crack- based strain sensors, comprising metal layers deposited on a polyimide substrate, are very thin ( \(\sim 7.5 \mu \mathrm{m}\) ) and lightweight ( \(\sim 3 \mathrm{mg}\) ) and thus have little effect on the structural deformation of the wing (Fig. 1d right). The inset SEM image shows that channeled + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 882, 465]]<|/det|> +nanoscale cracks on the metal surface are regularly aligned at intervals of \(3 \mu \mathrm{m}\) , and when the sensor is mechanically deformed, the gap between cracks widens and causes a large change in electrical resistance. Due to nanoscale cracks on the metal layer surfaces, the crack- based sensor has a higher gauge factor ( \(\sim 30,000\) ) and signal- to- noise ratio than conventional strain gauges (Supplementary Figs. 2a- j). The flapping- wing drone uses two motors located on the body (motor 1) and tail (motor 2) to control its motion (Fig. 1d left). Motor 1 was replaced with a commercial DC motor (DCX06M, GPX06 A 15:1, Maxon, Co) known for its high durability and low driving current ( \(< 0.23 \mathrm{A}\) ) to overcome the increased weight and increased resistance of wires. Motor 1 is combined with a scotch yoke mechanism that converts the rotational motion into flapping motion to generate thrust force, and motor 2 controls the direction of the drone by steering the tail wing. When a positive voltage is applied to the motor 2, the tail wing is rotated to the clockwise, thus turning the drone counterclockwise. The attachment of crack- based strain sensors to a drone's flapping wings allows for the precise and stable measurement of trajectory changes and the aerodynamic forces exerted on the wing over a long time (Supplementary Fig. 3 and Supplementary Video 1). + +<|ref|>sub_title<|/ref|><|det|>[[117, 512, 506, 530]]<|/det|> +## Validation as state informative sensor system + +<|ref|>text<|/ref|><|det|>[[115, 545, 882, 909]]<|/det|> +To analyze how accurately the wing deformation signal contains information about changes in wind direction and speed, we designed an experiment that the orientation of the drone was systematically controlled, allowing us to create scenarios with winds blowing from multiple directions and varying in speed. In the experiment, the relative wind direction was changed by controlling the drone's orientation. The drone rotated according to different combinations of yaw, pitch and roll angles, resulting in a total of 62 different wind vectors (Fig. 2a and Supplementary Video 2). Also, speed of wind varied in 3 steps, resulting in total 186 cases of blowing wind (Fig. 2b, Supplementary Table 1). The strain sensors attached to the wing base measured the strain, which was affected by the complex aerodynamic loads on the flapping wings. Not only the sensor signal changes with flapping- wing power (Extended Data Fig. 1), but it also varies with the direction and speed of the blowing wind (Extended Data Fig. 2). In this way, various information from environment is encoded into sensor signals. In order to decode the information, we used temporal convolutional neural network (1D CNN) and classified the direction and speed of the blowing wind. The raw sensor signals are given as input to the model, and the model predicts the wind direction and speed as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 100, 377, 118]]<|/det|> +outputs (Supplementary Fig. 4). + +<|ref|>text<|/ref|><|det|>[[115, 133, 883, 549]]<|/det|> +After training, we verified the feasibility of state recognition using only strain information based on calculated error from trained regression (Fig. 2c) and confusion matrix from the trained classification model (Fig. 2d). Out of a total of 12,416 validation cases, the mean absolute value of theta error \((\theta_{error})\) was estimated to be 29 degrees. Specifically, \(76\%\) of the angle errors fell inside the range of 0 to 33 degrees. In addition, the mean value of speed error \((s_{error})\) was calculated as \(26\%\) , and \(63.5\%\) of the speed errors fell within the range of \(0 - 30\%\) of speed error. For classification model, the mean AUC was approximately 0.99 (Supplementary Fig. 4c), and the model classified the wind direction and speed with a mean accuracy of \(80\%\) (Fig. 2d), that means the model can predict the direction and speed of the blowing wind with about \(80\%\) of accuracy from 186 cases. In addition, the prediction accuracy is displayed as a color map in spherical coordinates (Fig. 2e). As shown in the color map, the prediction accuracy was highest in the front, lower left, and lower right regions, while the accuracy was the lowest in the upper rear region (Supplementary Fig. 5). This is because the blowing wind from front, left, and right regions made the greatest mechanical deformation at the wing base. These results confirm that the drone can determine the wind direction and speed for various airflows, especially when the front, left, and right sides are windward, by using only the strain information collected by the two sensors on the wing base. + +<|ref|>sub_title<|/ref|><|det|>[[118, 596, 346, 613]]<|/det|> +## 1 DOF control experiment + +<|ref|>text<|/ref|><|det|>[[115, 629, 883, 895]]<|/det|> +Next, we designed an experimental setup that limited the drone's flight path to a single degree- of- freedom (DOF), to confirm its ability to utilize strain data for real- time adaptation to varying wind direction and speed, employing a RL controller (Soft- Actor- Critic algorithm; described in detail at Methods). The drone's body was anchored to a rotational encoder for simplified circular motion and rotated clockwise in x- y plane as the flapping power increased. In this setup, the direction and speed of the apparent wind acting on each wing changed according to the motor encoder's rotation angle \((\alpha)\) . At the initial position, the wind blew from the rear of the drone at a speed of \(3 \mathrm{m / s}\) , which is not enough to move the drone, but enough to assist it after it starts to fly. When thrusting force was generated by the flapping motion, a torsion spring dragged the drone back to the initial position. After the drone overcame the drag and passed over 90 degrees of rotational angle \((\alpha)\) , it had to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 883, 340]]<|/det|> +overcome a faster wind speed (5 m/s), that makes the drone more difficult to reach the target position. The sensor signal at the wing base was used as state information in the RL algorithm. The motor encoder at the center measured the rotational angle change, which was used only to calculate the reward during training (Fig. 3a and Supplementary Fig. 6a). The angle of target position was \(180^{\circ}\) , and Fig. 3b shows images from one training episode (Supplementary Video 3). The power increased, then quickly decreased and finally stabilized as the angle fluctuated near \(180^{\circ}\) , as shown in Fig. 3c. Fig. 3d shows that the drone reacted sensitively to the state change. As the episode advanced, the drone chose another strategy and slowly increased and decreased the motor power to approach the target position more smoothly. + +<|ref|>text<|/ref|><|det|>[[115, 354, 882, 619]]<|/det|> +We next compared the performance of the trained drone in these experiments with that of various untrained drones. There are 3 types of untrained drones: flapping in minimum maximum, and random motor power. Compared to the untrained cases, the RL- based drone adapted to the environment and outperformed on accumulated reward (score) as the episodes increase (Supplementary Figs. 6b and 6c). Furthermore, the drone maintained the target position in different angle ( \(270^{\circ}\) ), and even restored the target position when there was dragging- backward- disturbances (Extended Data Fig. 3). When the target position is changed to \(270^{\circ}\) , the only thing that has changed is the reward function. These results show that RL controlled drone can feel the direction of the wind by the strain information and control its flapping frequency in 1 DOF environment, to maintain target position where the highest reward is given. + +<|ref|>sub_title<|/ref|><|det|>[[118, 668, 346, 685]]<|/det|> +## 2 DOF control experiment + +<|ref|>text<|/ref|><|det|>[[115, 703, 882, 894]]<|/det|> +The aim of the next experiment was to verify whether campaniform sensilla aid in controlling gravity- oriented attitudes (pitch angle) for insects, given the unclear nature of whether insects sense gravity during flight control49. Conversely, conventional drones use multi- axis inertial measurement unit (IMU) devices to control attitude. Thus, another experimental setup was devised to validate that the drone can maintain sophisticated pitch angle attitudes. In the experimental environment, drone had to find the optimal pitch angle and maintain balance to increase the rotational angle, which resulted in a higher reward. The body of the drone was anchored to a rotational encoder (rotation joint 1) by a triangle- shaped + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 881, 340]]<|/det|> +rod, and the center of the body frame was attached to a pin joint (rotation joint 2) so that the drone's pitch angle \((\beta)\) varied from \(- 30^{\circ}\) to \(60^{\circ}\) . Each joint moved separately as the drone's pitch angle changed, regardless of the rotation angle \((\alpha)\) (Fig. 4a and Supplementary Fig. 7a). An apparent wind caused a drag force, pushing the drone backward and creating counterclockwise momentum in rotation joint 2. Because the reward increased proportionately with the rotation angle, the drone received a higher reward when rotation joint 1 rotated counterclockwise. Therefore, the drone had to precisely control the horizontal pitch angle to obtain the highest score during the episode. The drone aimed to simply fly forward through the apparent wind; however, the drone had difficulty maintaining the optimal pitch angle as the rotational angle increased. + +<|ref|>text<|/ref|><|det|>[[115, 355, 883, 767]]<|/det|> +In the experiment, the drone had to detect its own pitch angle based on the data collected by the wing strain sensor. As the drone is constrained to move clockwise or counterclockwise around the center of the motor encoder, only one crack- based sensor attached to the left side of the wing base was enough to acquire state information in the environment. The falling state is defined as the drone fails to maintain balance and the pitch angle decreases to less than \(0^{\circ}\) . In this state, because the direction of the blowing wind is perpendicular to the flapping direction, flapping motion does not assist to increase rotation angle \((\alpha)\) nor pitch angle \((\beta)\) . The only way to restore from this state is to stop flapping and wait for the apparent wind to recover the pitch angle horizontal with the direction of the wind. Fig. 4b shows this recovery process of the trained drone during a disturbance (Supplementary Video 4). As the drone balanced at the optimal pitch angle, the tail of the drone was lifted intentionally to induce the falling state. The drone recognized the falling state transition within 0.1 seconds and decreased its motor power to zero to recover the optimal pitch angle (Fig. 4c and Supplementary Fig. 7b). And accordingly, the amplitude of the sensor signals also decreased in response to the decreased motor power (Fig. 4d). Regardless of the flapping motion, the drone recognized the pitch angle with only one crack- based sensor and quickly recovered to reach a higher rotational angle. + +<|ref|>text<|/ref|><|det|>[[115, 782, 881, 900]]<|/det|> +The score of the RL- agent- controlled drone increased as the number of training episodes increased, whereas the score of the untrained drones was unchanged (Supplementary Fig. 7c). In addition, the trends of the rotational angle and pitch angle of the trained and untrained drones over time clearly differed (Supplementary Figs. 7d and 7e). For the trained drone, the rotational angle converged to approximately \(10^{\circ}\) , where the highest reward is given + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 882, 488]]<|/det|> +in the designed environment, and the drone also maintained a pitch angle of approximately \(10^{\circ}\) that is optimal for flying in the higher rotational angle. In contrast, the rotational and pitch angles of the untrained model did not converge and instead fluctuated randomly. As the rotational angle \((\alpha)\) and the drone's pitch angle \((\beta)\) change in each time, direction of drag force by blowing wind changes very slightly and frequently, making the drone difficult to fly in the direction of increasing rotational angle. Despite this difficulty, the trained drone recognized the subtle change in state well and successfully flew forward through the apparent wind to get the highest reward during a given episode (Extended Data Figs. 4a- d and Supplementary Video 5). Moreover, even if disturbances were intentionally induced, the drone maintained its balance and controlled the motor power with agility (Extended Data Figs. 4e- h and Supplementary Video 6). After each disturbance, the drone rapidly reduced its motor power to maintain the pitch angle and keep the balance for a smooth flight. The response was quick enough for immediate recovery, which ensured that the drone did not remain in the falling state. These results show that the RL based drone can recognize forward range (90 degree) of the blowing wind and learn to get the highest reward during a given episode by maintaining optimal pitch angle in designed environment. + +<|ref|>sub_title<|/ref|><|det|>[[118, 536, 450, 554]]<|/det|> +## Position control in windy environment + +<|ref|>text<|/ref|><|det|>[[115, 570, 882, 859]]<|/det|> +In general, flapping- wing drones require IMU sensor consisting of accelerometer, gyroscope, and magnetometer and optical flow sensors that can measure information in 3D space to achieve position and attitude control \(^{50,51}\) . In contrast, we indirectly measure the 3D space information by determining 4- dimensional information such as the x, y, z position coordinates and speed using only two strain sensors that measure 1- dimensional strain information. To determine the level of flight control that can be achieved by our proposed system, we designed the position and attitude control problem. In this problem, the commercial drone is trained to navigate to a target position in an unsteady airflow and receives a higher reward as the drone fly to the target position closely. Since the location- dependent airflow velocity is the only cue that the drone can use to identify the target position, the drone must decode the aeroelastic information from the wing deformation in a complex airflow. + +<|ref|>text<|/ref|><|det|>[[181, 875, 878, 893]]<|/det|> +Inside the wind tunnel, we installed an experimental setup, consisting of motion + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 99, 881, 390]]<|/det|> +capture cameras for measuring position information and a jig to secure the drone from damage due to collisions during repetitive training (Fig. 5a). Since the drone was passively aligned forward by the direction of the airflow, it could not hover without proper control. The drone's position information, measured by the motion capture cameras, was not used as a control input, but was used as a variable in the reward function to provide a reward for problem training (Supplementary Fig. 8). Whether during training or after training, the drone flies in a blind state without the aid of vision sensors. To generate an airflow to provide a positional cue, three small fans were installed at the entrance of the wind tunnel and one big fan was installed at the outlet. The airflow inside the wind tunnel was analyzed using an optical flow estimation method, and the complex asymmetric flow was visualized through color maps representing the wind direction and speed (Supplementary Fig. 9 and Supplementary Video 7). + +<|ref|>text<|/ref|><|det|>[[114, 404, 882, 791]]<|/det|> +Fig. 5b shows a short period (between 7.7 and 9.4 seconds) during training as the drone flies to the middle target position (Supplementary Video 8). The motion of the drone can be explained by dividing it into four cases: clockwise rotation, counterclockwise rotation, acceleration, and deceleration. The drone's movement along the x- axis, signifying its acceleration and deceleration, is primarily influenced by the output changes of motor 1 that regulate the thrust, whereas its rotation along the y- axis is mainly controlled by the output changes of motor 2 that regulate the direction. Fig. 5c shows the trajectory of the drone over time in 3D space, facilitating a comparative analysis with the change in the output of the motors. The trained RL- based controller adjusts the output of the motors based on the wing deformation measured by the strain sensors to determine the optimal path with the highest reward (Fig. 5d). For example, during flight towards the target position, the drone increases the thrust duty cycle between intervals 1 and 2 to accelerate, increases the direction duty cycle between intervals 2 and 3 to rotate clockwise, and repetitively adjusts the trust and direction duty cycles between intervals 3 and 4 to maintain its position at the target position, demonstrating its ability to interpret information about the target position through the aerodynamic forces generated by the interaction of its two wings with the airflow. + +<|ref|>text<|/ref|><|det|>[[115, 807, 882, 899]]<|/det|> +To validate the autonomous flight control results discussed above, Extended Data Fig. 5 displays further cases, such as training the drone to reach alternative target positions and comparing the outcomes with those of untrained drones. Target positions were tested on both the left and right sides of the drone to evaluate position control capability. While the drone + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 100, 882, 464]]<|/det|> +was capable of detecting and approaching the target position in both instances, there was a significant discrepancy in control performance between the two (Supplementary Videos 9 and 10). In order to make a quantitative comparison of control performance, we used a 2D heat map of the drone's trajectory in the XY plane with a grid dimension of \(50\mathrm{mm}\) for both width and length. By comparing the areas of the heatmaps, it was discovered that when trained with the target on the left, the area was \(\sim 1.6\) times smaller than when trained with the target on the right (Extended Data Figs. 5a and 5c). Moreover, we observed that there were larger fluctuations in the output of both the trust and direction duty cycles when trained with the target on the right (Extended Data Figs. 5b and 5d), and in the 3D trajectory graphs, it was apparent that the scatter of the drone's trajectory was much larger when trained with the target on the right (Extended Data Figs. 5e and 5f). However, considering that both cases showed trust and direction duty cycles producing outputs close to their maximum, and the location of the black grids with the highest visitation frequency in the heat map was behind the target position, it can be inferred that the drone is unable to overcome the resistance caused by airflow due to hardware limitations and the velocity of the right side of the flow field is faster. + +<|ref|>text<|/ref|><|det|>[[115, 479, 882, 914]]<|/det|> +To demonstrate the effectiveness of our proposed control strategy, we conducted a performance comparison between a trained drone and an untrained drone, with the target position set to the middle (Extended Data Fig. 6). The untrained drone showed a disorganized flight pattern, as shown in the heat map and 3D plot of its trajectory (Extended Data Figs. 6a and 6b, and Supplementary Video 11). In contrast, the trained drone repeatedly maintained its proximity to the target position (Extended Data Figs. 6d and 6e, and Supplementary Video 12). For obvious comparison, we displayed the frequency distribution of each drone within a radius of \(20 - 30\mathrm{cm}\) from the target position (Extended Data Figs. 6c and 6f). During a total episode that lasted 16 seconds, it was observed that the untrained drone stayed in the region for a duration of 1.08 seconds, which accounts for approximately \(6\%\) of the total steps taken. On the other hand, the trained drone was in the area for a duration of 5.22 seconds, indicating almost 29 percent of the total steps taken. Average position of the untrained and the trained is also compared (Extended Data Fig. 6g). We conducted a comparison of the mean value of position in the x- y plane. The average position of the untrained drone is located at coordinates \((20\mathrm{mm}, 70\mathrm{mm})\) exhibiting a standard deviation of \((315\mathrm{mm}, 280\mathrm{mm})\) . On the other hand, the average position of the trained drone is found at coordinates \((150\mathrm{mm}, 200\mathrm{mm})\) displaying a standard deviation of \((110\mathrm{mm}, 130\mathrm{mm})\) . Moreover, as the number of training epochs for the trained drone increased, the entropy and score decreased and increased, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 881, 240]]<|/det|> +respectively, indicating a reduction in the uncertainty of the output value and successful training of the RL controller model (Extended Data Figs. 6e and 6f). In summary, the effectiveness of the proposed flapping-wing drone control system was confirmed through experimental testing in a wind tunnel, utilizing only two strain sensors to indirectly measure 3D space information for successful position and attitude control without the vision sensors and IMU. + +<|ref|>sub_title<|/ref|><|det|>[[115, 291, 512, 309]]<|/det|> +## Flight path control in a windless environment + +<|ref|>text<|/ref|><|det|>[[115, 325, 881, 490]]<|/det|> +Based on experiments such as wind perception, 1DOF control, 2DOF control, and position control in wind tunnel, we systematically expanded the control utility and principles for implementing the fly- by- feel system. To demonstrate the stable controllability of the fly- by- feel system, we conducted a free flight demonstration with various motions only using the two strain sensors in a windless environment. In previous experiments, wind was an apparent clue to create large wing strain, but in a windless environment, only relatively small deformation due to attitude changes are the information for state estimation. + +<|ref|>text<|/ref|><|det|>[[115, 506, 881, 895]]<|/det|> +We conducted experiments in an enough space (width: 6m, depth: 8m, height: 4m) for drone's natural flight motion (Fig. 6a left). For reliable training, we designed the drone launcher to ensure the same start conditions such as takeoff speed (Supplementary Fig. 10). In addition, very thin and lightweight wires (40 milligram per meter) were used to minimize the tensile force of the tether applied to the drone. By utilizing 12 sets of motion capture cameras, we were able to adjust the reward based on the drone's trajectory (Extended Data Fig. 7). Our first objective was to manage simple left and right motion during flight in order to verify the reliability of trajectory control (Extended Data Fig. 8). Throughout every training session, the RL agent gradually increases its scores and effectively adjusts the angle of the direction motor. Following this, in order to enhance the complexity of the problem, we attempted a combination of left and right motions to navigate in a zigzag pattern and aimed to achieve a circular flight trajectory. In contrast to previous experiments, we made use of the human demonstration method (Fig. 6a right), creating a Replay buffer with human- controlled episodes so that the RL agent had the opportunity to pick up on the desirable motion of each episode. With the strain sensors, the RL agent was able to control flying in both zigzag and circular motion after training (Fig 6b, d). Achieving precise control of the drone's motion + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 880, 193]]<|/det|> +proves to be challenging, with only a \(30\%\) success rate when researchers manually control its flight along the desired trajectory. However, the RL agent adeptly adjusted the power of the directional motor, resulting in an increase in score and a decrease in entropy as the episode progressed (Supplementary Fig. 11). + +<|ref|>text<|/ref|><|det|>[[115, 207, 881, 325]]<|/det|> +In addition to controlling the directional motor, the thrust motor power was also trained for adjustment during free flight. Our objective was to train the RL agent to maintain altitude higher or lower than the takeoff altitude. Remarkably, the RL agent successfully orchestrated the thrust motor power and directional motor power to sustain appropriate altitude, achieving higher scores as the episodes progressed (Extended Data Fig. 9). + +<|ref|>text<|/ref|><|det|>[[113, 338, 882, 875]]<|/det|> +Furthermore, we compared the odometry and ground truth of the trajectory (Fig. 6c, e) to confirm that the sensor could predict trajectory in the windless environment. In previous research, the accuracy of odometry has been improved by combining an air flow sensor that can detect the wind speed and direction with a IMU based attitude controllable multirotor52,53. Since wing deformation also carries information about wind speed and direction, we were able to reconstruct the drone's driving path. We used 0.5 seconds of sensor signal as input and extracted ground truth based on local coordinate of drone (Extended Data Fig. 10a left). We randomly divided the collection of episodes into a training set and a validation set, with a ratio of 9:1. Subsequently, we conducted training on a model that possesses the same architecture as the policy network of the RL agent but includes additional dropout layer to prevent overfitting (Extended Data Fig. 10a right). Validation episodes are not utilized during the training of the model. We conducted a total of 5 training sessions, each using different variations of training datasets. During the iteration of training epochs, we observed a decrease in the mean squared error (MSE) for both the training and validation datasets as the epochs progressed (Extended Data Fig. 10b, c). Furthermore, we calculated the root mean square error (RMSE) for the displayed episodes. For the zigzag motion, the RMSE was 587.7 mm in the x-axis, 165.5 mm in the y-axis, and 289.9 mm in the z-axis. On the other hand, for circular motion, the RMSE was 636.2 mm in the x-axis, 202.4 mm in the y-axis, 813.3 mm in the z-axis. Additionally, we performed supplementary experiments to ensure the repeatability, yielding different kind of odometry, such as the s- curve (Extended Data Fig. 10d). In this episode, the RMSE was 376.3 in the x-axis, 181 mm in the y-axis, and 404.5 mm in the z-axis. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 100, 241, 121]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[115, 140, 881, 677]]<|/det|> +Inspired by the sensory mechanisms of flying insects, particularly their ability to sense aeroelastic loads via strain- measuring organs known as campaniform sensilla, we developed a wing strain- based flight control system. The bioinspired design and engineering foundation of the proposed control system demonstrated that wing strain information effectively encodes the surrounding environments and the flying orientation during flight, even in noisy condition resulting from high vibration of flapping. We attached ultrasensitive, lightweight crack- based strain sensors to the wings of a drone and measured aeroelastic loads during flight. Although the sensor is lightweight (3 mg for a 1 mm width x 5 mm length) and flexible with a thickness of \(7.5 \mu \mathrm{m}\) and a Young's modulus of \(2.5 \mathrm{GPa}^{54}\) , further miniaturization is necessary for application in subgram drones with wing weights of 3 mg. In terms of miniaturization, crack sensors are advantageous due to their simple structure, which involves metal deposited on a thin plastic film, and can be efficiently fabricated using photolithography or laser ablation. We maintained solely deploying strain sensors within our system to evaluate the effectiveness of these sensors in controlling the flapping- wing drone. Our experimental approach was carefully organized into five distinct phases, aimed at rigorously testing our hypothesis and exploring the role of wing strain sensors in controlling a flapping- wing drone. Each phase provided results that reinforced the viability and ingenuity of utilizing wing strain sensors for autonomous drone flight control. These sensors offer strong real- time feedback for system modifications across different atmospheric conditions and enable advanced control strategies without relying on traditional sensors such as IMUs or vision systems. This highlights their potential to improve drone responsiveness and agility in challenging environments. + +<|ref|>text<|/ref|><|det|>[[118, 690, 881, 857]]<|/det|> +Consequently, the use of just two strain sensors proved adequate for the drone to effectively track wind cues from multiple directions. The strain sensors at each wing's base comprehensively captured various features such as the wind speed and direction, drone's attitude, and flapping speed. In the robot experiments, following the only cue given from the wind, flight tasks such as positioning, balancing, and navigating were performed without employing other flight control sensors such as gyro sensors, accelerometers, and vision sensors, which were necessary in previous drones55. + +<|ref|>text<|/ref|><|det|>[[180, 872, 878, 891]]<|/det|> +In our research, we demonstrated that the fly- by- feel control system could be + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 881, 290]]<|/det|> +implemented by employing bio- inspired strain sensors and end- to- end RL. The end- to- end RL minimizes the need for manual feature selection and intervention by directly mapping sensory input to control outputs, significantly streamlining the learning process and improving response accuracy in complex scenarios. The fly- by- feel flight control system provides a unique approach to navigation that is not solely reliant on GPS or visual cues. This method can adapt subtle shifts in complex aerodynamics, especially valuable in situations where GPS signals falter or are absent, and where visual systems are hindered by adverse environmental factors. + +<|ref|>text<|/ref|><|det|>[[115, 305, 882, 693]]<|/det|> +By conducting robotic experiments, we confirmed that the wing strain of flapping wings contains information about the direction and speed of the wind. These findings support the hypothesis that flying insects can accurately sense wind direction and speed through the campaniform sensilla on their wings. Although antennae56, hair33,57, and bristles39 are known to be wind receptors, the aeroelastic forces that vary with wind can also serve as wind- sensing cues. However, these organs may have difficulty detecting all wind directions if they are localized to specific parts of the body or can be disturbed by flapping- induced flow vortices. During flight, the ambient wind, apparent wind and flapping motion- induced wind act simultaneously58. If a sensory organ is localized to a specific part of the body, such as antennae, it may have difficulty detecting all wind directions because it may be covered by the body or wings depending on the direction of the ambient wind. Hair and bristles distributed on the wings can be disturbed by the flow vortices induced by the flapping motion. In contrast, wings can obtain spatial- temporal flow information during flight. From this perspective, campaniform sensilla on wings are a strong candidate for organs enabling insects to choose the most favorable wind direction for long- distance migration and maneuver in complex and unpredictable environments. + +<|ref|>text<|/ref|><|det|>[[115, 708, 882, 899]]<|/det|> +Developing a wing strain- based flight control model is very challenging. Natural flyers independently control their wing stroke amplitude, pitch angle amplitude, and phase lag to accomplish various maneuvers59. This kinematics generate and shed vortices into the flow and cause the wing shape to vary. Moreover, the large deformation of the wings results in complex fluid- structure interactions and highly coupled nonlinearities in fluid dynamics, aeroelasticity, flight dynamics, and control systems. Even simplified computational models require significant computational resources and often take several days for each simulation to converge60. While quadrotors, and Quadrupedal robots with rigid body components can + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 99, 881, 563]]<|/det|> +trained in sim- to- real environments, thereby enhancing their control performance at a generalized level \(^{61 - 63}\) , the flapping wing drone systems are still at a fundamental stage in simulation owing to the deformability of body parts and corresponding complexity of aerodynamics. As we aimed to propose a novel control system for flapping drones inspired by nature, we utilized real- world reinforcement learning as an alternative to address these complexities. Since the proposed RL agent trains the neural network using only the strain signal of the wings recorded in a real- world environment, our scheme eliminates many of the calculations required for fluid- structure interactions in flight control systems. Real- world learning obviously has challenges such as limited sample data, delays in the system actuators or rewards, and reasoning system constraints \(^{64}\) . However, because of their offline domains, compensation for reward delays, and algorithms that can consider environmental constraints, RL schemes can be applied in systems that were previously considered too fragile or expensive for learning- based approaches \(^{65}\) . In our research, we used a low- cost toy drone, which has limited durability, but completed learning by giving a time term for each episode and setting the length of the episode to be as short as 20 seconds. However, with a more durable drone that has a higher degree of control freedom, we expect that more dynamic and high- maneuverability flight techniques can be learned. Future developments in this area could involve using more advanced drones to achieve higher performance or complex flight techniques. + +<|ref|>text<|/ref|><|det|>[[115, 576, 881, 790]]<|/det|> +The campaniform sensilla are hypothesized to play a significant role in sensing gravity in flying insects. Although no dedicated gravity sensing organ has been observed in insects, it is possible that the deformation of the wings could vary based on their orientation relative to gravity. However, to accurately differentiate this deformation from that caused by wind and acceleration, multiple sensors may be required to detect complex spatiotemporal deformations of the wing, such as twisting or buckling. In future research, we intend to verify whether attaching multiple sensors to the drone wings enables gravity detection and advanced reactive autonomous flight, such as hovering and wind- assisted flight, without the use of accelerometers and gyro sensors. + +<|ref|>text<|/ref|><|det|>[[115, 806, 880, 849]]<|/det|> +Overall, we expect the aerobatic fly- by- feel flight control system to be applied in many scenarios in the near future. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 100, 222, 121]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[117, 141, 881, 258]]<|/det|> +Method overview. We aim to train RL agents that control a flapping drone. The drone must perform various tasks within an airflow, including maintaining a specific target position and executing different types of turns in desired directions. To train such agents, we employed the most straightforward method: directly collecting experience in real- world interactions and then training the agents with the data. + +<|ref|>text<|/ref|><|det|>[[115, 273, 882, 490]]<|/det|> +In detail, we first embedded bioinspired crack- based sensors into a commercial flapping- wing robot. These sensors measure the deformation of the wing base for every time step, allowing agents to perceive the surrounding environment. This system can be viewed as a partially observable Markov decision process (POMDP). To address partial observability, we extracted the state of the robot using a 1- D convolutional neural network (1D CNN). The reward is calculated using a motion capture camera and motor encoder. Subsequently, any off- the- shelf RL algorithm can be applied to modify the policy of the agents to maximize the reward. In the following sections, we provide details about the chosen algorithm, the sensor fabrication process, and the experimental setup. + +<|ref|>sub_title<|/ref|><|det|>[[118, 540, 415, 557]]<|/det|> +## Reinforcement learning algorithm + +<|ref|>text<|/ref|><|det|>[[115, 572, 884, 899]]<|/det|> +Drone control as a POMDP. We define the system as a partially observable Markov decision process (POMDP). A POMDP includes a set of states \(S\) , a set of actions \(A\) , a reward function \(r(s, a)\) , the probability distribution \(s' \sim P(s' | s, a)\) of the next state \(s'\) given current state and action, a set of observations \(\Omega\) , a set of conditional observation probabilities \(O\) , and a discount factor \(\gamma \in [0,1)\) . For every time step, the agent samples an action \(a \in A\) using policy \(a \sim \pi (a | s_{agent})\) . Here, since the agent cannot access the true state \(s\) of the environment, it usually construct \(s_{agent}\) , a belief state about the true state \(s\) using history of the observations. When the agent executes an action, the environment makes the state transition, and the agent receives the next observation \(o'\) . The goal of RL is to find a policy \(\pi (a | s_{agent})\) that maximizes the expected sum of the discounted reward for a given \(POMDP \equiv (S, A, r, P, \Omega , O, \gamma)\) . Since we parametrize \(\pi (a | s_{agent})\) through a neural network with parameters \(\theta\) , we denote the policy as \(\pi_{\theta}\) . Therefore, the goal of RL is to find \(\theta\) that satisfies the following. + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[375, 99, 620, 148]]<|/det|> +\[\theta = \underset {\theta}{\mathrm{argmax}}\mathbb{E}_{\pi_{\theta}}\left(\sum_{t}\gamma^{t}r_{t}\right)\] + +<|ref|>text<|/ref|><|det|>[[115, 161, 881, 256]]<|/det|> +In our experimentation, we discretized the time into 0.05 second of intervals. At each step, the agent observes \(o_{t}\) from the signal received by the sensor attached to the drone's wing. Then, the agent constructs the agent state \(s_{t}\) based on the most recent 32 observations, which is equivalent to 1.6 seconds of time: + +<|ref|>equation<|/ref|><|det|>[[115, 270, 380, 291]]<|/det|> +\[s_{t} = (o_{t},o_{t - 1},\dots,o_{t - 31})\in R^{32}.\] + +<|ref|>text<|/ref|><|det|>[[115, 305, 882, 573]]<|/det|> +The agent determines the action according to \(a_{t}\sim \pi_{\theta}(a_{t}|s_{t})\) . Here, note that \(a_{t}\) is the torque of the motor for sing so the action is continuous, such that \(a_{t}\in [0,1]\) . When \(a_{t}\) is equal to 1, the motor output is maximized, and when it is 0, the motor output is minimized. In all experiments, the discount factor \(\gamma\) is set to 0.98, empirically. Through the experiments, the drone executes an action and then the environment make the transition, so that we can gather the transition \(\tau_{t} = (o_{t},a_{t},r_{t + 1},o_{t + 1})\) for every time step. These experiences are then collected in the replay buffer, and the RL algorithm samples minibatches from the replay buffer to update the policy. Each minibatch of data is consists of 64 transitions. After certain number of updates, we obtain a new parameter \(\theta^{\prime}\) and send the parameter directly to the neural network to replace the previous parameter. This process is repeated until the algorithm converges. + +<|ref|>text<|/ref|><|det|>[[115, 586, 882, 754]]<|/det|> +RL Algorithm. We use a soft actor- critic (SAC) \(^{66}\) algorithm as the off- the- shelf policy gradient algorithm. The SAC algorithm is particularly suitable for our problem because it is an off- policy RL algorithm, inherently sample- efficient, and well- suited for our setup. Widely used on- policy algorithms \(^{67 - 69}\) usually discard the data after performing single gradient updates, off- policy algorithms exhibit better sample efficiency as data can be reused multiple times. High sample efficiency is essential because we are gathering data from the real- world robot, and the amount of data is limited. + +<|ref|>text<|/ref|><|det|>[[115, 769, 881, 813]]<|/det|> +Additionally, the SAC algorithm is known for its robustness since it is trained with a maximum entropy objective \(^{70}\) . \(\mathcal{H}\) of the policy \(\pi\) as following: + +<|ref|>equation<|/ref|><|det|>[[344, 828, 651, 852]]<|/det|> +\[\mathcal{H}\big(\pi (\cdot |s)\big) = \mathbb{E}_{a_{t}\sim \pi}\big(-log\pi (a_{t}|s_{t})\big)\] + +<|ref|>text<|/ref|><|det|>[[115, 867, 881, 911]]<|/det|> +Since \(\mathcal{H}\) measures the randomness of the policy, maximizing \(\mathcal{H}\) indicates that the algorithm is in favor of stochastic policy. This ensures that the algorithm finds the policy as random as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 880, 143]]<|/det|> +577 possible while achieving high rewards. More concretely, SAC aims to maximize the sum of both reward and the entropy of the policy for each time step. + +<|ref|>equation<|/ref|><|det|>[[290, 156, 704, 206]]<|/det|> +\[\theta_{s a c} = \underset {\theta}{\mathrm{argmax}}\mathbb{E}_{\pi_{\theta}}\left(\sum_{t}\gamma^{t}[r_{t} + \alpha \mathcal{H}(\pi (\cdot |s_{t}))]\right)\] + +<|ref|>text<|/ref|><|det|>[[115, 220, 881, 395]]<|/det|> +Note that a temperature parameter \(\alpha\) is introduced to control the relative importance of the entropy term against the reward term. With the objective given, we update following parameters: \(\psi\) for the soft state value function \(V_{\psi}\) , \(\phi\) for the soft action value function \(Q_{\phi}\) , and \(\theta\) for the policy \(\pi_{\theta}\) . Here, \(V_{\psi}(s)\) is the expected sum of reward and entropy given state \(s\) , and \(Q_{\phi}(s, a)\) is the expected sum of reward and entropy given state \(s\) and action \(a\) . Both functions assume the future is sampled using policy \(\pi\) . We first update \(\psi\) to minimize the squared residual error: + +<|ref|>equation<|/ref|><|det|>[[217, 409, 777, 448]]<|/det|> +\[J_{V}(\psi) = \mathbb{E}_{s_{t}\sim D}\left[\frac{1}{2} (V_{\psi}(s_{t}) - \mathbb{E}_{a_{t}\sim \pi_{\theta}}[Q_{\phi}(s_{t},a_{t}) - log\pi_{\theta}(a_{t}|s_{t}))]^{2}\right]\] + +<|ref|>text<|/ref|><|det|>[[115, 461, 880, 506]]<|/det|> +where \(D\) refers to the distribution of states and actions in the replay buffer. We then update \(\phi\) to minimize the soft Bellman residual \(J_{Q}(\phi)\) as follows: + +<|ref|>equation<|/ref|><|det|>[[115, 520, 515, 552]]<|/det|> +\[J_{Q}(\phi) = \mathbb{E}_{(s_{t},a_{t})\sim D}\left[\frac{1}{2} (Q_{\phi}(s_{t},a_{t}) - \hat{Q} (s_{t},a_{t}))^{2}\right],\] + +<|ref|>text<|/ref|><|det|>[[115, 565, 500, 587]]<|/det|> +where \(\hat{Q} (s_{t},a_{t}) = r(s_{t},a_{t}) + \gamma \mathbb{E}_{s_{t + 1}}[V_{\bar{\psi}}(s_{t + 1})]\) . + +<|ref|>text<|/ref|><|det|>[[115, 602, 880, 673]]<|/det|> +Here, \(\bar{\psi}\) denotes the exponential moving average of \(\psi\) , is introduced to stabilize the learning process. Finally, we update \(\theta\) to minimize the Kullback- Leibler (KL) divergence between current policy and the target density defined by Q- function as follows: + +<|ref|>equation<|/ref|><|det|>[[328, 686, 668, 730]]<|/det|> +\[\mathbb{E}_{s_{t}\sim D}\left[D_{K L}\left(\pi_{\theta}(s_{t})\right)\left|\frac{e x p\left(Q_{\phi}(s_{t},a_{t})\right)}{Z_{\phi}(s_{t})}\right)\right]\] + +<|ref|>text<|/ref|><|det|>[[115, 744, 880, 791]]<|/det|> +\(Z_{\phi}(s_{t})\) is a partition function that normalizes the distribution \(exp(Q_{\phi}(s_{t},a_{t}))\) . More details of the algorithm can be found in the original paper66. + +<|ref|>text<|/ref|><|det|>[[115, 805, 880, 900]]<|/det|> +Neural network architecture and training details. We use a modified version of SAC algorithm that is working only with \(\pi_{\theta}\) and \(Q_{\phi}\) 71. In this version of the algorithm, the value of \(\alpha\) can be trained concurrently by setting the target entropy explicitly. We use a 1D CNN architecture to properly encode residing patterns from sensor signals across time domain. In + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 98, 881, 370]]<|/det|> +detail, \(s_t \in \mathbb{R}^{32}\) is fed into the neural network \(\pi_\theta\) , and then the two consecutive 1D CNN layers and a max pooling layer are applied in the time dimension. The 1D CNN layers employ 32 filters with width 5, and the pooling layer uses a window with width 2. The output tensor is flattened into a 1D vector and embedded into a 256- dimensional vector through a linear layer. Finally, the network outputs mean \(\mu\) and variance \(\sigma^2\) of the gaussian distribution. We sample final actions from the gaussian distribution. The action value function \(Q_\phi\) basically has the same structure but with an additional concatenation layer added right before the final output layer. We embed the action into a 64- dimensional vector and concatenate it with the flattened output of the CNN layers. \(\pi_\theta\) and \(Q_\phi\) have separate network parameters. A nonlinear ReLU function \(^{72}\) is applied to the outputs of all convolutional and linear layers. Other detailed hyperparameters are listed in Extended Data Table 2. + +<|ref|>sub_title<|/ref|><|det|>[[117, 419, 550, 439]]<|/det|> +## Equipping the sensors on the flapping-wing drone + +<|ref|>text<|/ref|><|det|>[[115, 453, 883, 842]]<|/det|> +The crack- based sensor was fabricated in the same way as in previous studies \(^{46,73}\) . After covering the shadow mask on a \(7.5 \mu \mathrm{m}\) PI film substrate (3022- 5 Kapton thin film, Chemplex, Co), \(50 \mathrm{nm}\) chrome (Cr) and \(20 \mathrm{nm}\) gold (Au) thin films were deposited at a rate of \(0.5 \mathrm{\AA / s}\) at \(1.2 \times 10^{- 5} \mathrm{mTorr}\) using a thermal evaporator (DDHT- SB015, Dae Dong Hitech, Co). The deposited substrate was cut at \(5 \mathrm{mm}\) intervals and stretched \(2\%\) using a material tester (3342 UTM, Instron, Co) to generate nanoscale cracks on the surface, as this facilitated a high gauge factor of \(\sim 30,000\) at \(2\%\) strain. To measure the wind deformation, we equipped the fabricated crack- based sensors on the wings of a flapping drone (MetaFly, XTIM, Co). The measured weight and wingspan of the drone are \(10 \mathrm{g}\) and \(13.5 \mathrm{cm}\) , respectively. The PDMS dry transfer method \(^{74}\) was used to accurately attach the sensors to the thin wing base, which had a width of \(1 \mathrm{mm}\) , and a commercial strain gauge adhesive (CN adhesive, Tokyo Sokki Kenkyujo, Co) was used for strong and conformal bonding. Conductive epoxy (CW2400, Chemtronics, Co) was used for electrical connection between the sensor and wires, and commercial epoxy adhesive (EE- 05, Axia, Co) was applied to reduce the strain on the wiring interface. We used modified strip theory based on blade elemental analysis to calculate the lift and trust force using the wing strain (Supplementary Fig. 12). + +<|ref|>sub_title<|/ref|><|det|>[[117, 890, 270, 907]]<|/det|> +## Signal processing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 98, 881, 217]]<|/det|> +We acquired the strain sensor signal using DAQ: DEwESoft SIRIUS. We used a bandpass filter over the range 0.1\~55 Hz and normalized the signal to 0\~1 to increase the RL agent learning efficiency. All data from the DAQ system and motion capture camera module were transmitted to the desktop computer via serial communication and processed using a Python- language- based RL algorithm (PyTorch). + +<|ref|>sub_title<|/ref|><|det|>[[117, 266, 432, 285]]<|/det|> +## Wind prediction experiment (Fig. 2) + +<|ref|>text<|/ref|><|det|>[[115, 299, 883, 540]]<|/det|> +Simulating 186 wind cases. The flapping- wing drone was anchored to a 3 DOF motor system with 3 motors bound along the x, y, and z axes. As shown in Fig. 4a, wind was generated by an electric fan, and the drone rotated with various combinations of yaw, pitch, and roll angles (Supplementary Figure 5- 10). A Teensy 4.0 microcontroller and L298N motor drivers were used. With this system, 62 wind directions were generated, as depicted in Fig. 4b. Three wind speeds were considered: 3 m/s, 5 m/s, and 7 m/s. The speed variation is expressed as colored arrows: 3 m/s in blue, 5 m/s in green, and 7 m/s in red. Three flapping powers were considered: 60%, 80% and 100% of the motor maximum. However, these powers did not affect the number of classification- labels. As shown in the sub yellow blocks in Fig. 4b, the signal varied depending on the wind direction and speed. + +<|ref|>text<|/ref|><|det|>[[115, 554, 883, 767]]<|/det|> +Neural network. For 186 wind cases classification, we used neural network composed of two 1D convolutional layers followed by two fully connected layers. Each convolutional layer included 5 filters with size 50, yielding a total of 64 filters. The ReLU \(^{72}\) activation function and a max pooling layer were applied to the output of each convolutional layer (Supplementary Fig. 4b). For the fully connected layers, dropout was applied to prevent overfitting. The final output was transformed to the shape of a probability distribution with the softmax function, and cross entropy was used as the loss function. The model received 0.2 seconds of the sensor signal as input. For regression, there was difference in the number of nodes in fully connected layers and final output layers (Supplementary Fig. 4d). + +<|ref|>text<|/ref|><|det|>[[115, 782, 881, 852]]<|/det|> +Calculating prediction errors. We used the inverse trigonometric function of cosine to determine the angle by dividing the dot product by the product of the magnitudes of each vector ( \(\vec{a}\) is prediction vector, and \(\vec{b}\) is ground vector): + +<|ref|>equation<|/ref|><|det|>[[223, 868, 774, 904]]<|/det|> +\[\theta_{error} = cos^{-1}\left(\frac{a\cdot b}{|a||b|}\right)\left(\vec{a} = (x_1i, y_1j, z_1k), \vec{b} = (x_2i, y_2j, z_2k)\right)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 99, 880, 167]]<|/det|> +On the other hand, we represented the speed error as a percentage, which is calculated by dividing the difference between the predicted vector magnitude and the ground vector magnitude by the ground vector magnitude: + +<|ref|>equation<|/ref|><|det|>[[201, 181, 795, 225]]<|/det|> +\[s_{error} = \left(\left|\frac{|a| - |b|}{|b|}\right|\right)\times 100\left(\vec{a} = (x_{1}i,y_{1}j,z_{1}k),\vec{b} = (x_{2}i,y_{2}j,z_{2}k)\right)\] + +<|ref|>sub_title<|/ref|><|det|>[[117, 273, 290, 291]]<|/det|> +## Experimental setup + +<|ref|>text<|/ref|><|det|>[[115, 306, 881, 547]]<|/det|> +Wind tunnel. A canopy tent with a width, length, and height of \(1.5\mathrm{m}\) , \(3\mathrm{m}\) , and \(2.8\mathrm{m}\) , respectively, was installed as an indoor wind tunnel. The outlet and inlet were made by cutting the sidewall of the canopy tent. The gap between the ground and tent was sealed with tape to prevent air flow leakage. To generate air flow in the wind tunnel, a large air circulator (Shinil Electronics Co., Ltd) was placed at the outlet, and three small air circulators (Unimax Co, Ltd) were placed at the inlet. An asymmetric and complex air flow was formed by controlling the output of the air circulators. To analyze the airflow inside the wind tunnel, smoke was generated behind the air circulator using a smoke machine, and the turbulence flow was visualized with the smoke converted into linear motion by passing through a honeycomb lattice. + +<|ref|>text<|/ref|><|det|>[[115, 561, 881, 902]]<|/det|> +1 DOF experiment (Fig. 3). The drone was fixed at a right angle to the end of a \(20\mathrm{cm}\) long stick, and the other end of the stick was fixed with a rotary joint (which was also an encoder). As a result, the drone moved in a circular motion when flapping its wings. The movement range of the drone was \(0^{\circ}\) to \(350^{\circ}\) . When the drone was not flapping its wings, a torsion spring allowed the drone to return to its original position. Constant wind in the - Y direction was generated by the two fans at speeds of \(5\mathrm{m / s}\) (upward) and \(3\mathrm{m / s}\) (downward). The drone was controlled by a DC motor that produced flapping motion, and the drone's tail wing was removed. Sensors were attached at the wing base where the stress was largest. The state provided to the RL controller was the final 0.5 seconds of the strain signals collected by the sensors on both wings and the power of the motor used by the drone to control itself. The controller's decision period was 0.05 that is an empirically determined value to handle rapid flow changes around the drone. So, each 10 second episode included 200 decisions. We used a Gaussian distribution function with 180 as the median for the reward function, and accordingly, the drone received the highest score when it maintained its position at \(180^{\circ}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 100, 881, 216]]<|/det|> +2 DOF experiment (Fig. 4). The drone was anchored at a pin joint so that the drone's pitch angle varied from \(- 30^{\circ}\) to \(60^{\circ}\) . A triangle-shaped rod was attached to the upper encoder, and the reward was determined by the rotational angle of the rod. \(5\mathrm{m / s}\) of wind blew against the drone, and the drag force of the wind pushed the drone back. This resulted in the pitch angle turning counterclockwise, which allowed the drone to return to its original state. + +<|ref|>text<|/ref|><|det|>[[115, 231, 881, 421]]<|/det|> +Position control in windy environment (Fig. 5). The drone was tethered with a nylon cord to prevent collisions and ensure that it returned to its initial position after each learning episode. The drone had 2 degrees of freedom (DOFs) of control with two motors and flew with 6 DOF movement. Due to the limited DOF control of the drone, the target position was given as \(x\) and \(y\) coordinates. We used a 2D multivariate probability distribution function as the reward function and input the target location as the mean. The SAC algorithm was used for RL, and motion capture cameras (PrimeX 13, Optitrack, Co) were used to provide location information as a reward. + +<|ref|>text<|/ref|><|det|>[[115, 437, 881, 900]]<|/det|> +Flight path control experiments (Fig. 6). In a confined space measuring approximately 6 meters in width, 8 meters in depth, and 4 meters in height, we strategically positioned a total of 12 motion capture cameras and 5 passive markers reflecting infrared rays were attached to the drone. To facilitate the free flight of drones in this space, a dedicated drone launcher was developed (Supplementary Fig. 9). This launcher comprises rail frames, wheels, and a motor ensuring consistent start conditions such as velocity and initial angle of attack of the drone. At the beginning of each episode, the drone on the stage advances at a speed of \(1\mathrm{m / s}\) until it is halted by a mechanical brake. Leveraging the law of the inertia, the mounted drone takes off from the stage, initiating its flight by flapping its wings. During training, the drone was picked up and remounted to the launcher after each episode, and curtains and nets were hung around the walls to prevent collisions and potential damage to the drone. For actuation and data acquisition from the drone, a total of eight copper wires (American Wire Gauge 40, diameter: \(96.5\mu \mathrm{m}\) , length: \(8\mathrm{m}\) ) were utilized. Specifically, 2 wires were allocated for the thrusting motor, 2 for directional motor, and 4 for the two strain sensors attached on the drone's wings. As depicted in Figure 6, the actual dimensions of the drone's flying space are \(4\mathrm{m} \times 6\mathrm{m} \times 2\mathrm{m}\) (width \(x\) length \(x\) height). However, since the tethers are centrally attached to the floor, the operational space is effectively reduced to \(2\mathrm{m} \times 3\mathrm{m} \times 2\mathrm{m}\) (width \(x\) length \(x\) height), which is the area where the drone and tethers interact. This configuration minimizes the tether length, consequently reducing the tensile force exerted on the drone and preventing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 99, 884, 217]]<|/det|> +loss of thrust, thereby enhancing flight efficiency. With this arrangement, the maximum length of each tether is approximately \(4.12\mathrm{m}\) . Given the mass of the tether is \(40\mathrm{mg}\) per meter, and with eight wires connected, the total additional weight from the tethers amounts to \(1.32\mathrm{g}\) , which is only \(13.2\%\) of the drone's total weight. This confirms that the mass of the tethers is minimal and does not significantly impact the drone's performance. + +<|ref|>text<|/ref|><|det|>[[115, 231, 882, 570]]<|/det|> +The reward function is designed based on each trajectory of the drone. For leftward flight, the reward is proportionally given according to the displacement in the negative direction of the x- axis and the positive direction of the y- axis. This means the drone receives a larger reward when moving to the left- front direction (Extended Data Fig. 9a). Conversely, in the case of rightward flight, the reward is proportional to the positive displacement along the x- axis and the positive direction of the y- axis (Extended Data Fig. 9b). For a mixed trajectory involving both leftward and rightward flight, such as the zigzag trajectory, the reward is adjusted based on the displacement along the y- axis (Extended Data Fig. 9c). Additionally, we increased the input size as the RL agent needed to consider more earlier standpoints. The reward is initially proportional to the displacement in the negative direction of the x- axis and changes in the opposite direction when the drone passes a certain point along the y- axis. We focused solely on the displacement in the x- axis to induce a dramatic directional change in the drone. Due to inertia, a substantial force is required to alter the drone's direction, making this strategy suitable for achieving our objective. + +<|ref|>text<|/ref|><|det|>[[115, 585, 882, 777]]<|/det|> +Next, we trained the RL agent to control the drone to fly in a circular motion by adjusting the reward function (Extended Data Fig. 9d). The reward is determined proportionally by the dot product of two vectors extracted from sequential positions. Considering the displacement of the drone from its viewpoint, global vectors are rotated according to the local coordinate system of the drone itself. Since our experiment was conducted only in the real world and not in simulation, we fixed the thrust power of the drone to \(80\%\) of the duty cycle to reduce the time required for the RL agent to learn. In this environment, the RL agent is tasked with controlling the angle of the directional motor in the drone's tail. + +<|ref|>text<|/ref|><|det|>[[115, 792, 882, 910]]<|/det|> +We also trained the RL agent to control the altitude of the drone, allowing it to fly at a higher or lower altitude than the starting point of the drone launcher. The reward function designates that the drone receives a higher reward when flying further away from the starting point while maintaining the target altitudes of 1.6 meters and 0.8 meters, respectively. The RL agent had to manage both thrust motor power and directional motor power to uphold desired + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 100, 348, 117]]<|/det|> +altitude in this environment. + +<|ref|>sub_title<|/ref|><|det|>[[118, 168, 288, 185]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[116, 201, 881, 344]]<|/det|> +D.K., S.H. and J.- S.K. acknowledge financial support from the Ajou University research fund. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (2021R1A6A3A01087289, 2022R1A2C2093100) and Korea Environment Industry & Technology Institute (KEITI) through Digital Infrastructure Building Project for Monitoring, Surveying and Evaluating the Environmental Health Program, funded by Korea Ministry of Environment (MOE) (2021003330009). + +<|ref|>sub_title<|/ref|><|det|>[[118, 381, 303, 399]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 414, 881, 679]]<|/det|> +T.K., I.H., S.I., S.R., J.- S.K., S.H., and D.K. led the developed of the concepts, designed the experiments, interpreted results, and wrote the paper. T.K., I.H., S.I. designed the experimental setups and S.R., T.K., and S.I. developed codes for reinforcement learning algorithm system. I.H and M.K. provided strain sensors on the flapping drone with support from Y.R., C.K., J.P., D.L., D.L., and S.L. as fabrication of the sensors. J.L. supported to design embedded circuit system and I.B. supported to build Motion Capture tracking system of the flapping drone. J.J. and Y.- M.C supported to build free flight experiments, and S.I. analyzed odometry of the drone in free flight with support from M.R.H. and S.K. In addition, J.L. checked for the reinforcement learning results, and S.S. contributed to progress the experiments. U.K. provided the kinematics ideas for implementing flapping motion of the drone. + +<|ref|>sub_title<|/ref|><|det|>[[118, 730, 293, 747]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[118, 763, 471, 781]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[118, 832, 262, 849]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[118, 865, 880, 908]]<|/det|> +The data that support the findings of this study are available from the corresponding author upon reasonable request. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 134, 266, 151]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[118, 167, 880, 213]]<|/det|> +Codes for the reinforcement learning used in this study are available on following GitHub (https://github.com/TaewiKim/Fly- by- Feel). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 139, 880, 666]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 100, 524, 118]]<|/det|> +
Fig. 1: Fly-by-feel-based flight control strategy.
+ +<|ref|>text<|/ref|><|det|>[[115, 668, 880, 801]]<|/det|> +a, Mechanism for detecting wing deformation in death's head hawkmoth and flapping drone. Death's head hawkmoth : spike rate change via campaniform sensilla in response to wing movement. Flapping drone : resistance change of crack- based sensor according to wing flapping. b, The state information that can be derived from the wing deformation measured by the crack- based strain sensors in the fly- by- feel system. c, The reinforcement learning- based flight control system using wing strain information. d, Photograph of a flapping drone equipped with crack- based strain sensors on the wing bases and two motors on the body and tail for thrust and direction control, respectively. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 130, 870, 585]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[66, 100, 650, 117]]<|/det|> +
Fig. 2: Aerodynamic feature extraction from wing strain data.
+ +<|ref|>text<|/ref|><|det|>[[117, 589, 880, 672]]<|/det|> +a, Experimental design for changing the wind direction and speed. b, 186 sets of wing strain data were acquired by changing the wind direction and speed. c, Histogram according to the error range, Angle error and speed error. d, Confusion matrix from the test dataset. e, Visualization of the classification accuracy as a color map in spherical coordinates. Trained model well classified the red-colored range. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 150, 878, 675]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 100, 878, 135]]<|/det|> +
Fig. 3: Position control by sensing the wind direction and speed, which vary depending on the position.
+ +<|ref|>text<|/ref|><|det|>[[115, 678, 880, 821]]<|/det|> +a, Schematic illustration of the environment and agent in the designed experiment with 1 DOF. 2 sensor signals and thrust motor power were used as input, while the rotation angle of the encoder was used as variable of reward function. b, Snapshots of an example episode in which a drone controls the motor power to maintain a target position. c, Strain sensor signal (top), motor power (middle) and drone angle (bottom). The range in the red box shows the signals in the snapshots in (b). After the drone passes over the target, it stops flapping to get target position. d, Magnified view of the black dotted box in (c). Trained drone reacts with agility as sensor signals change. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 132, 880, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 100, 601, 117]]<|/det|> +
Fig. 4: Reinforcement learning in a 2 DOF environment.
+ +<|ref|>text<|/ref|><|det|>[[115, 660, 880, 777]]<|/det|> +a, Schematic illustration of the environment and agent in the designed experiment with 2 DOFs. 1 sensor signal was used as input, while the rotation angle of the encoder was used as variable of reward function. b, Snapshots of a representative episode. The drone controls the motor power despite the induced disturbance. c, Sensor signal (top), motor power (middle), and rotational and pitch angles of the drone (bottom). The range in the red box shows the signal in the snapshots in (b). As the trained drone recognizes the falling state, it stops flapping to recover proper pitch angle. d, Magnified view of the black dotted box in (c). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 132, 875, 666]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 100, 852, 118]]<|/det|> +
Fig. 5: Position control of a flapping-wing drone with the proposed fly-by-feel system.
+ +<|ref|>text<|/ref|><|det|>[[115, 667, 880, 768]]<|/det|> +a, Wind tunnel experiment setup to train a tethered flapping-wing drone and the training system configuration. b, Representative images of a flapping drone flying to the target position based on only information obtained by the strain sensors attached to the wings. c, Trajectory of the drone in x- y- z space. The black line is the flight trajectory before 7.5 seconds. d, Changes in wing strain and the output of motors of 7.5 to 9.5 seconds during the target positioning flight. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 132, 875, 651]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[67, 100, 777, 118]]<|/det|> +
Fig. 6: Flight path control of a flapping-wing drone in windless environment.
+ +<|ref|>text<|/ref|><|det|>[[115, 664, 880, 806]]<|/det|> +a, Windless environmental setup designed for training a flapping-wing drone and the corresponding training system configuration. b, Sequential snapshots from a video illustrating the flight of a flapping-wing drone in a zigzag trajectory, relying solely on information acquired through strain sensors attached to its wings. c, Comparison between odometry and ground truth trajectory of the drone in zigzag motion. d, Sequential snapshots from a video showing a flapping drone executing a circular trajectory, guided exclusively by information obtained from strain sensors attached to its wings. e, Comparison between odometry and ground truth trajectory of the drone in circular motion. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[170, 100, 825, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 585, 880, 700]]<|/det|> +
Extended Data Figure 1. Correlation between the flapping-wing power and sensor signal from the wind. a, Schematic of the experimental setup. b, Real image showing the experimental setup. c, Change in the duty cycle as the motor power was increased from 15\~100%. d, Variation in the thrust force of the drone as the motor power increased. The thrust force was recorded by nano 17 fixed under the flapping drone. e, Variation in the sensor signal as the motor power increased. f, In detail, as the motor power increased, the frequency of a single flapping cycle and signal amplitude both increased.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 100, 874, 545]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 555, 881, 720]]<|/det|> +Extended Data Figure 2. Signal changes with changing wind direction. a, Five representative wind directions on the front side. b, Sensor signals according to each wind direction on the front side. In the head-blowing scenario, the heights of the signal peaks of the left and right sensors are similar. c, Five representative wind directions on the left side. d, Corresponding sensor signals of each direction on the left side. The biased wind direction causes more strain on the left wing. This is reflected in the sensor signal, as the height of the left wing signal is greater than that of the right wing signal. e, Five representative wind directions on the right side. f, Corresponding sensor signals of each direction on the right side. In contrast to the results of wind blowing on the left side, the height of the right wing signal is greater than that of the left wing signal. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 108, 876, 300]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[123, 300, 876, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 666, 880, 810]]<|/det|> +
Extended Data Figure 3. Additional examples of the 1 DOF system. a, Snapshots of the episode when the target angle is \(180^{\circ}\) . It moves quickly to reach target position and reduces its motor power when it passes over the target. b, The case when the target position angle is \(270^{\circ}\) . When the target position is changed, the drone could recognize the position based on the sensor information. c, The case when the researcher disturbs the flight by pushing the drone backward and forward. Even if there is an intentional disturbance, a trained drone controls its motor power to maintains the target position. d, Motor power control and corresponding angle change for case 1. e, Motor power control and corresponding angle change for case 2. f, Motor power control and corresponding angle change for case 3.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[152, 100, 850, 666]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 671, 883, 884]]<|/det|> +Extended Data Figure 4. Extended episode for optimal flight learning. a, Snapshots of free flight in the 2 DOF system. The trained drone balanced to maintain proper pitch angle during flying. b, Signal amplitude of the crack- based sensor attached on the left wing's base. c, Change in the drone's motor power. The drone fine- tunes the power to sustain the optimal pitch angle \((\beta)\) . d, Changes in the rotational angle \((\alpha)\) and pitch angle \((\beta)\) . During flight, drone maintains optimal pitch angle and so that could get the highest rotation angle (approximately \(10^{\circ}\) ). e, Snapshots of a disturbed flight in the 2 DOF system. Even if there is an intentional disturbance, a trained drone reduces its motor power to avoid falling state. f, Variation in the signal amplitude, synchronized with (e). The perturbation events are marked with red circles. g, Power of the motor during the disturbed flight. After the perturbation events, the drone immediately reduces its power to maintain balance. h, Rotational angle \((\alpha)\) and pitch angle \((\beta)\) during the disturbed flight. The drone recovers the optimal angles quickly after the perturbation events. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 100, 880, 630]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 645, 881, 745]]<|/det|> +Extended Data Figure 5. Position control of a flapping- wing drone when trained to reach different target positions. a and b, Trajectory and change in the wing deformation and the output of motors of a trained drone heading toward the left target position. c and d, Trajectory and change in the wing deformation and the output of motors of a trained drone headed toward the right target position 3 (fast airflow). e and f, 3D trajectories of the trained drone heading toward target position (e : left target position, f : right target position). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 100, 870, 570]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[101, 580, 881, 714]]<|/det|> +Extended Data Figure 6. Comparison of an untrained and trained drone. a and b, Plane and space trajectories of the untrained drone heading toward middle target position. c, Frequency distribution of untrained drone within a radius of 20\~30cm from the target point. d and e, Plane and space trajectories of the trained drone heading toward middle target position. f, Frequency distribution of trained drone within a radius of 20\~30cm from the target point. g, Comparison of mean value of the position from untrained and trained drone. h and i, Changes in the entropy and score of the artificial neural network with the number of training epochs. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 100, 870, 640]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 643, 880, 727]]<|/det|> +Extended Data Figure 7. Reward functions for flight trajectory control experiments. a, Reward function designated for leftward flight. b, Reward function designated for rightward flight. c, Reward function designated for zigzag trajectory of flight. d, Reward function designated for circular motion flight. e, Reward function designated for higher altitude trajectory of flight. f, Reward function designated for lower altitude trajectory of flight. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 100, 880, 585]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 592, 880, 741]]<|/det|> +Extended Data Figure 8. Results of left and rightward flight control. a, Sequential snapshots from a video depicting the drone's leftward flight. b, Sequential snapshots from a video displaying drone's rightward flight. c, Trajectory of the drone during the leftward flight. d, Scores observed as episode progressed during the training of leftward flight. e, Strain sensor signals during leftward flight and corresponding duty cycle of thrust motor power and directional motor power. f, Trajectory of the drone during rightward flight. g, Scores recorded as the episode progressed during the training of rightward flight. h, Strain sensor signals corresponding to the rightward flight, along with the associated duty cycle of thrust motor power and directional motor power. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 100, 878, 583]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 591, 880, 757]]<|/det|> +Extended Data Figure 9. Results of higher and lower altitude flight control. a, Sequential snapshots from a video illustrating flight at higher altitudes. b, Sequential snapshots from a video displaying drone's flight at lower altitudes. c, Trajectory of the drone during the higher altitude flight. d, Scores recorded as episode progressed during the training of higher altitude flight. e, Strain sensor signals during the higher altitude flight, along with the corresponding duty cycle of thrust motor power and directional motor power. f, Trajectory of the drone during lower altitude flight. g, Scores recorded as the episodes progressed during the training of the lower height intended flight. h, Strain sensor signals corresponding to the lower height flight, along with the associated duty cycle of thrust motor power and directional motor power. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 110, 870, 448]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 459, 880, 543]]<|/det|> +
Extended Data Figure 10. Methods of yielding odometry and results. a, Labelling procedure for comparing odometry and ground truth trajectories. b, Comparison of mean squared error between 5 folds of different training datasets. c, Comparison of mean squared error between 5 folds of different validation datasets. d, Representative result of comparison between odometry and ground truth trajectories from additional experiment.
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[272, 98, 722, 560]]<|/det|> + +
HyperparameterValue
Learning rate of π0.0005
Learning rate of q0.001
Learning rate of α0.001
Initial α0.01
γ0.98
τ(for EMA update)0.01
Batch size32
Target entropy-1.0
Maximum buffer size3000
Initial training buffer size1000
Maximum episode length (number of steps)300
Decision period (s)0.05
+ +<|ref|>text<|/ref|><|det|>[[60, 562, 95, 572]]<|/det|> +959 + +<|ref|>text<|/ref|><|det|>[[60, 589, 536, 600]]<|/det|> +960 **Extended Data Table 1. Hyperparameter values.** + +<|ref|>text<|/ref|><|det|>[[60, 615, 95, 625]]<|/det|> +961 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 90, 885, 910]]<|/det|> +962 References963 1 Ansari, S., Zbikowski, R. & Knowles, K. Aerodynamic modelling of insect- like flapping flight for micro air vehicles. Progress in aerospace sciences 42, 129- 172 (2006).965 2 Ma, K. Y., Chirarattananon, P., Fuller, S. B. & Wood, R. J. Controlled flight of a biologically inspired, insect- scale robot. Science 340, 603- 607 (2013).966 3 Chen, Y. et al. Controlled flight of a microrobot powered by soft artificial muscles. Nature 575, 324- 329 (2019).967 4 Jafferis, N. T., Helbling, E. F., Karpelson, M. & Wood, R. J. Untethered flight of an insect- sized flapping- wing microscale aerial vehicle. Nature 570, 491- 495 (2019).968 5 De Croon, G., De Clercq, K., Ruijsink, R., Remes, B. & De Wagter, C. Design, aerodynamics, and vision- based control of the DelFly. Int. J. Micro Air Veh. 1, 71- 97 (2009).969 6 Phan, H. V., Kang, T. & Park, H. C. Design and stable flight of a 21 g insect- like tailless flapping wing micro air vehicle with angular rates feedback control. Bioinspiration Biomimetics 12, 036006 (2017).970 7 Zhang, J., Fei, F., Tu, Z. & Deng, X. in 2017 IEEE International Conference on Robotics and Automation (ICRA). 5422- 5428 (IEEE).971 8 Mellinger, D. & Kumar, V. in 2011 IEEE international conference on robotics and automation. 2520- 2525 (IEEE).972 9 Elijah, T., Jamisola, R. S., Tjiparuro, Z. & Namoshe, M. A review on control and maneuvering of cooperative fixed- wing drones. International Journal of Dynamics and Control 9, 1332- 1349 (2021).973 10 de Croon, G. Flapping wing drones show off their skills. Sci. Robot. 5, eabd0233 (2020).974 11 Chirarattananon, P. et al. Dynamics and flight control of a flapping- wing robotic insect in the presence of wind gusts. Interface focus 7, 20160080 (2017).975 12 Baek, S. S., Bermudez, F. L. G. & Fearing, R. S. in 2011 IEEE/RSJ International Conference on Intelligent Robots and Systems. 2674- 2681 (IEEE).976 13 Nguyen, Q.- V. & Chan, W. L. Development and flight performance of a biologically- inspired tailless flapping- wing micro air vehicle with wing stroke plane modulation. Bioinspiration Biomimetics 14, 016015 (2018).977 14 Shin, H. S. et al. Bio- Inspired Large- Area Soft Sensing Skins to Measure UAV Wing Deformation in Flight. Adv. Funct. Mater. 31, 2100679 (2021).978 15 Tong, S., Weiping, Z., Jiawang, M. & Zihao, C. in 2019 IEEE International Conference on Unmanned Systems (ICUS). 842- 847 (IEEE).979 16 Woolf, P. J. Chemical process dynamics and controls. (2009).980 17 Walker, S. M., Thomas, A. L. & Taylor, G. K. Deformable wing kinematics in free- flying hoverflies. J. R. Soc. Interface 7, 131- 142 (2010).981 18 Somps, C. & Luttges, M. Dragonfly flight: novel uses of unsteady separated flows. Science 228, 1326- 1329 (1985).982 19 Wang, H., Zeng, L., Liu, H. & Yin, C. Measuring wing kinematics, flight trajectory and body + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 100, 884, 900]]<|/det|> +1001 attitude during forward flight and turning maneuvers in dragonflies. J. Exp. Biol. 206, 745- 757 (2003). 1002 Springthorpe, D., Fernández, M. J. & Hedrick, T. L. Neuromuscular control of free-flight yaw turns in the hawkmoth Manduca sexta. J. Exp. Biol. 215, 1766- 1774 (2012). 1005 21 Salami, E., Ward, T. A., Montazer, E. & Ghazali, N. N. N. A review of aerodynamic studies on dragonfly flight. Proc. Inst. Mech. Eng., Part C 233, 6519- 6537 (2019). 1007 Mischiati, M. et al. Internal models direct dragonfly interception steering. Nature 517, 333- 338 (2015). 1009 23 Hedlund, J. S. et al. Unraveling the world's longest non- stop migration: the indian ocean crossing of the globe skimmer dragonfly. Front. Ecol. Evol. 9, 525 (2021). 1011 24 Gao, B. et al. Adaptive strategies of high- flying migratory hoverflies in response to wind currents. Proc. R. Soc. B 287, 20200406 (2020). 1013 25 Cardé, R. T. Animal migration: seasonal reversals of migrant moths. Curr. Biol. 18, R1007- R1009 (2008). 1015 26 Menz, M. H. et al. Individual tracking reveals long- distance flight- path control in a nocturnally migrating moth. Science 377, 764- 768 (2022). 1017 27 Hu, G. et al. Mass seasonal bioflows of high- flying insect migrants. Science 354, 1584- 1587 (2016). 1019 28 Hedlund, J. S. et al. Unraveling the World's longest non- stop migration: The Indian Ocean crossing of the Globe Skimmer Dragonfly. Front. Ecol. Evol., 525 (2021). 1021 29 Frye, M. A. & Dickinson, M. H. Motor output reflects the linear superposition of visual and olfactory inputs in Drosophila. J. Exp. Biol. 207, 123- 131 (2004). 1023 30 Sherman, A. & Dickinson, M. H. Summation of visual and mechanosensory feedback in Drosophila flight control. J. Exp. Biol. 207, 133- 142 (2004). 1025 31 Tuthill, J. C. & Wilson, R. I. Mechanosensation and adaptive motor control in insects. Curr. Biol. 26, R1022- R1038 (2016). 1027 32 Dickerson, B. H., Fox, J. L. & Sponberg, S. Functional diversity from generic encoding in insect campaniform sensilla. Current Opinion in Physiology 19, 194- 203 (2021). 1029 33 Taylor, G. K. & Krapp, H. G. Sensory systems and flight stability: what do insects measure and why? Advances in insect physiology 34, 231- 316 (2007). 1031 34 Combes, S. & Daniel, T. Flexural stiffness in insect wings II. Spatial distribution and dynamic wing bending. J. Exp. Biol. 206, 2989- 2997 (2003). 1033 35 Bomphrey, R. J. & Godoy- Diana, R. Insect and insect- inspired aerodynamics: unsteadiness, structural mechanics and flight control. Curr. Opin. Insect. Sci. 30, 26- 32 (2018). 1035 36 Dickerson, B. H., Aldworth, Z. N. & Daniel, T. L. Control of moth flight posture is mediated by wing mechanosensory feedback. Journal of Experimental Biology 217, 2301- 2308 (2014). 1037 37 Dickinson, M. H. Comparison of encoding properties of campaniform sensilla on the fly wing. Journal of experimental biology 151, 245- 261 (1990). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 100, 886, 900]]<|/det|> +1040 38 Pratt, B., Deora, T., Mohren, T. & Daniel, T. Neural evidence supports a dual sensory- motor role for insect wings. Proceedings of the Royal Society B: Biological Sciences 284, 20170969 (2017).1042 39 Fabian, J. et al. Systematic characterization of wing mechanosensors that monitor airflow and wing deformations. iScience 25, 104150 (2022).1045 40 Hinson, B. T. & Morgansen, K. A. Gyroscopic sensing in the wings of the hawkmoth Manduca sexta: the role of sensor location and directional sensitivity. Bioinspiration & biomimetics 10, 056013 (2015).1048 41 Taylor, G. Modelling the effects of unsteady flow phenomena on flapping flight dynamics- stability and control. Flow Phenomena in Nature: A Challenge to Engineering Design 1, 155- 166 (2007).1051 42 Sane, S. P. The aerodynamics of insect flight. Journal of experimental biology 206, 4191- 4208 (2003).1053 43 Chin, D. D. & Lentink, D. Flapping wing aerodynamics: from insects to vertebrates. Journal of Experimental Biology 219, 920- 932 (2016).1055 44 Breuer, K. (Nature Publishing Group, 2019).1056 45 Aiello, B. R. et al. Spatial distribution of campaniform sensilla mechanosensors on wings: form, function, and phylogeny. Current Opinion in Insect Science 48, 8- 17 (2021).1058 46 Kang, D. et al. Ultrasensitive mechanical crack- based sensor inspired by the spider sensory system. Nature 516, 222- 226 (2014).1060 47 Lee, E. et al. Effect of metal thickness on the sensitivity of crack- based sensors. Sensors 18, 2872 (2018).1062 48 Kim, T. et al. Polyimide encapsulation of spider- inspired crack- based sensors for durability improvement. Applied Sciences 8, 367 (2018).1064 49 Bender, J. A. & Frye, M. A. Invertebrate solutions for sensing gravity. Current biology 19, R186- R190 (2009).1066 50 Yang, R., Zhang, W., Mou, J., Zhang, B. & Zhang, Y. in Proceedings of 2022 International Conference on Autonomous Unmanned Systems (ICAUS 2022). 1432- 1443 (Springer).1068 51 Bloesch, M. et al. in 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems. 3102- 3107 (IEEE).1070 52 Tagliabue, A. & How, J. P. in 2021 IEEE International Conference on Robotics and Automation (ICRA). 5736- 5743 (IEEE).1072 53 Tagliabue, A. et al. in 2020 IEEE/RSJ international conference on intelligent robots and systems (IROS). 1645- 1652 (IEEE).1074 54 Kim, T. et al. Ultra- stable and tough bioinspired crack- based tactile sensor for small legged robots. npj Flexible Electronics 7, 22 (2023).1076 55 Pan, E., Liang, X. & Xu, W. Development of vision stabilizing system for a large- scale flapping- wing robotic bird. IEEE Sensors Journal 20, 8017- 8028 (2020).1078 56 Gewecke, M. Antennae: another wind- sensitive receptor in locusts. Nature 225, 1263- 1264 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 100, 884, 825]]<|/det|> +(1970). 57 Fratzl, P. & Barth, F. G. Biomaterial systems for mechanosensing and actuation. Nature 462, 442- 448 (2009). 58 van Breugel, F. in 2021 60th IEEE Conference on Decision and Control (CDC). 1399- 1406 (IEEE). 59 Shyy, W. et al. Recent progress in flapping wing aerodynamics and aeroelasticity. Progress in Aerospace Sciences 46, 284- 327 (2010). 60 Schwab, R., Reade, J. & Jankauski, M. Quasi three- dimensional deformable blade element and unsteady vortex lattice reduced- order modeling of fluid- structure interaction in flapping wings. Physics of Fluids 34, 121903 (2022). 61 Cheng, X., Shi, K., Agarwal, A. & Pathak, D. Extreme parkour with legged robots. arXiv preprint arXiv:2309.14341 (2023). 62 Huang, K., Rana, R., Spitzer, A., Shi, G. & Boots, B. Datt: Deep adaptive trajectory tracking for quadrotor control. arXiv preprint arXiv:2310.09053 (2023). 63 Zhuang, Z. et al. Robot parkour learning. arXiv preprint arXiv:2309.05665 (2023). 64 Lee, J., Hwangbo, J., Wellhausen, L., Koltun, V. & Hutter, M. Learning quadrupedal locomotion over challenging terrain. Science robotics 5, eabc5986 (2020). 65 Dulac- Arnold, G. et al. Challenges of real- world reinforcement learning: definitions, benchmarks and analysis. Machine Learning 110, 2419- 2468 (2021). 66 Haarnoja, T., Zhou, A., Abbeel, P. & Levine, S. in International conference on machine learning. 1861- 1870 (PMLR). 67 Mnih, V. et al. in International conference on machine learning. 1928- 1937 (PMLR). 68 Espeholt, L. et al. in International conference on machine learning. 1407- 1416 (PMLR). 69 Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms. arXiv:1707.06347 (2017). 70 Ziebart, B. D. Modeling purposeful adaptive behavior with the principle of maximum causal entropy. (Carnegie Mellon University, 2010). 71 Haarnoja, T. et al. Soft actor- critic algorithms and applications. arXiv preprint arXiv:1812.05905 (2018). 72 Agarap, A. F. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375 (2018). 73 Roh, Y. et al. Vital signal sensing and manipulation of a microscale organ with a multifunctional soft gripper. Science Robotics 6, eabi6774 (2021). 74 Sundar, V. C. et al. Elastomeric transistor stamps: reversible probing of charge transport in organic crystals. Science 303, 1644- 1646 (2004). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 43, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 453, 230]]<|/det|> +ExtendedDataTable1.tif - SupplementaryVideo.zip - Supplementaryinformationclean.docx - Supplementaryinformationmarkedup.docx + +<--- Page Split ---> diff --git a/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/images_list.json b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ff490c4451588e67b4a0e8c3381ad21df2e2df70 --- /dev/null +++ b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/images_list.json @@ -0,0 +1,122 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Ga1-GDP thermostability and intrinsic tryptophan fluorescence are pH dependent. (A) Representative CD thermal melt (222 nm, 20–95 °C) of Ga1-GDP (15 μM) shows enhanced thermostability at higher pH (N = 3). (B) Representative CD spectral scans (200-250 nm) of Ga1-GDP collected at 45 and 55°C and pH 7.2 (N = 2). (C) Bar graph derived from temperature-dependent CD spectral scans highlighting the loss of alpha-helical but not beta-sheet secondary structure. (D) Representative intrinsic tryptophan fluorescence spectra of Ga1-GDP (2 μM, excitation = 280 nm, emission = 300 - 400 nm) shows enhanced fluorescence as the pH is increased from 5 to 7.2 (N = 2, performed in triplicate).", + "footnote": [], + "bbox": [ + [ + 164, + 88, + 827, + 450 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: Gai-GDP \\(^{1}\\mathrm{H}\\) - \\(^{15}\\mathrm{N}\\) 2D NMR spectral changes as a function of pH. (A) Representative 2D \\(^{1}\\mathrm{H}\\) - \\(^{15}\\mathrm{N}\\) TROSY-HSQCP spectral overlay of \\(^{2}\\mathrm{H}\\) , \\(^{13}\\mathrm{C}\\) , \\(^{15}\\mathrm{N}\\) -enriched Gai-GDP (230 \\(\\mu \\mathrm{M}\\) ) at pH 6.4, 6.8, and 7.2, highlighting peak shifts and line broadening. Spectra were acquired on a Bruker Avance III 850 MHz instrument at \\(25^{\\circ}\\mathrm{C}\\) ( \\(\\mathrm{N} = 2\\) ). (B) Chemical shift perturbation (CSP) and (C) peak intensity ratio \\((\\mathrm{Int}_{\\mathrm{pH}6.4} / \\mathrm{Int}_{\\mathrm{pH}7.2})\\) for pH 6.4 and 7.2. The striped horizontal line represents the mean value of \\(\\Delta \\delta\\) amide. Most of the residues that show significant CSP, and broadening lie within the Ras-like domain \\((\\alpha 1,\\alpha 5,\\beta 1,\\beta 2\\) and \\(\\beta 2 - \\beta 3\\) loop). (D) Spectral differences are highlighted on a ribbon diagram of GTP-bound Gai (PDB: 1CIP). NH residues with CSP greater than 0.03 ppm are represented in red. Residues with decreased peak intensity (line broadening) at pH 6.4 relative to pH 7.2 are shown in green. Residues missing or unassigned are shown in black color while unaffected residues are shown in blue color.", + "footnote": [], + "bbox": [ + [ + 112, + 95, + 876, + 610 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: Rates of Gai GDP nucleotide dissociation vary with pH. (A) Structure of Gai (PDB:1GP2) highlighting the position of tryptophan (W211) in Switch II and bound Mant-GDP. (B) Diagram displaying experimental setup of FRET-based determination of Gai nucleotide association and dissociation rates. (C) Rate of GDP dissociation from Gai as a function of pH by monitoring the time-dependent decrease in FRET emission of Gai-loaded Mant-GDP at 445 nm upon the addition of \\(7.5 \\mu \\mathrm{M}\\) GDP. (D) The rate of Mant-GDP dissociation decreases as the pH increases. Data are averages of two independent experiments performed in duplicate (±SE).", + "footnote": [], + "bbox": [ + [ + 125, + 92, + 872, + 560 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4: pH-dependent nucleotide exchange and stability of Gai variants from the GDP release network. (A) Ribbon diagram of GTP-bound Gai (PDB: 1CIP) highlighting electrostatic network near GDP release network. The helical domain, Ras-like domain and the Switch regions are shown in gray, blue, and green, respectively. Charged residues are shown as yellow sticks. (B) Rates of GDP dissociation is compared for WT Gai and Gai variants (H57T and K192Q) as a function of pH. Data are averages of two independent experiments performed in duplicate (±SE) (C) Representative CD melt profile of GDP-bound WT and variant Gai (N=2), Gai H57T (D) and K192Q (E) proteins. Gai release network variants retain pH-dependent thermal unfolding profiles similar to WT Gai.", + "footnote": [], + "bbox": [ + [ + 210, + 313, + 777, + 623 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. pH-dependent stability and tryptophan fluorescence of Gai Switch network variants. (A) The putative pH-dependent electrostatic network (red) within the Switch regions (green) is highlighted on the ribbon diagram of", + "footnote": [], + "bbox": [ + [ + 179, + 130, + 812, + 576 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6: MD simulations of WT Gai-GDP highlighting uncharged and charged states of E236, D237 and E245.", + "footnote": [], + "bbox": [ + [ + 120, + 88, + 880, + 390 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig. 7: BRET assay showing enhanced Gai-Gβγ association at lower intracellular pH for the Gai Switch variants. (A) Schematic diagram (BioRender) illustrating the BRET assay used to monitor receptor-mediated dissociation of Gai from Gβγ. For this assay, the GPCR neurotensin receptor, Ga Renilla luciferase 8 (Rluc8), Gβ3 and γ9-GFP fusion constructs were co-transfected in HEK29T cells and fluorescence (510 nm) monitored after addition of the substrate and agonist (neurotensin). Net BRET (dissociation) plotted as γ9-GFP (acceptor) over Ga-Rluc (donor) ratio as a function of log doses of neurotensin. Neurotensin receptor-mediated BRET shows that lowering pHi (from 7.2 to 6.7) either by altering extracellular pH (B) or by the addition of FCCP (C), decreases Gai dissociation from Gβγ in HEK29T cells. (D) WT Gai shows significantly higher dissociation from Gβγ relative to both double variant and triple variants (low pH mimetics of WT Gai). (E) The difference in net BRET signal as a function of FCCP concentration (0 μM in solid line and 2 μM in dotted line) for double variant and triple variants is reduced with respect to WT Gai, confirming key roles for E236, D237, and E245 in pH-mediated Gβγ dissociation for Gai-Gβγ complex. All BRET data are averages of two independent experiments performed in duplicate (±SE).", + "footnote": [], + "bbox": [ + [ + 225, + 90, + 765, + 586 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8: Proposed mechanism of Gai pH regulation. Activation of Gai by GPCRs leads to the release of Gai from \\(\\mathrm{G}\\beta \\gamma\\) and stimulation of signaling pathways. A distinct mechanism of signaling regulation may occur by intracellular pH regulation. Gai protein may act as intracellular pH sensors to regulate Gai- \\(\\mathrm{G}\\beta \\gamma\\) interactions with lower pH enhancing Gai- \\(\\mathrm{G}\\beta \\gamma\\) interactions to inhibit Gai signaling.", + "footnote": [], + "bbox": [ + [ + 117, + 288, + 875, + 592 + ] + ], + "page_idx": 26 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc.mmd b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc.mmd new file mode 100644 index 0000000000000000000000000000000000000000..e3f6a1d47d68caed5f7dd2367388ecaa51379a12 --- /dev/null +++ b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc.mmd @@ -0,0 +1,335 @@ + +# Molecular and Functional Profiling of Gai as an Intracellular pH Sensor + +Sharon Campbell campbes1@med.unc.edu + +University of North Carolina at Chapel Hill https://orcid.org/0000- 0003- 0311- 409X + +Ajit Prakash UNC Chapel Hill + +Zijian Li University of North Carolina https://orcid.org/0000- 0002- 6802- 9606 + +Venkat Chirasani University of North Carolina https://orcid.org/0000- 0003- 1416- 0438 + +Juhi Rasquinha University of North Carolina + +Natalie Valentin University of North Carolina + +Garrett Hubbard University of North Carolina + +Guowei Yin The Seventh Affiliated Hospital of Sun Yat- sen University + +Henrik Dohlman University of North Carolina at Chapel Hill + +## Article + +Keywords: + +Posted Date: April 30th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4203924/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on April 11th, 2025. See the published version at https://doi.org/10.1038/s41467-025-58323-2. + +<--- Page Split ---> + +## Molecular and Functional Profiling of Gαi as an Intracellular pH Sensor + +2 Ajit Prakash \(^{1}\) , Zijian Li \(^{1}\) , Venkata R. Chirasani \(^{1}\) , Juhi A. Rasquinha \(^{1}\) , Natalie H. Valentin \(^{2}\) , Garrett B. Hubbard \(^{1}\) , Guowei Yin \(^{3}\) , Henrik G. Dohlman \(^{2*}\) , Sharon L. Campbell \(^{1,4*}\) \(^{1}\) Department of Biochemistry & Biophysics, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA \(^{2}\) Department of Pharmacology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA \(^{3}\) The Seventh Affiliated Hospital of Sun Yat- sen University, Shenzhen, 518107, China \(^{4}\) Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA + +8 \*Address for correspondence: hdohlman@med.unc.edu and campbesl@med.unc.edu + +## Abstract: + +Heterotrimeric G proteins ( \(\mathrm{G}\alpha\) , \(\mathrm{G}\beta\) and \(\mathrm{G}\gamma\) ) act downstream of G- protein- coupled receptors (GPCRs) to mediate signaling pathways that regulate various physiological processes and human disease conditions. Previously, human \(\mathrm{G}\alpha \mathrm{i}\) and its yeast homolog Gpa1 have been reported to function as intracellular pH sensors, yet the pH sensing capabilities of \(\mathrm{G}\alpha \mathrm{i}\) and the underlying mechanism remain to be established. Herein, we identify a pH sensing network within \(\mathrm{G}\alpha \mathrm{i}\) , and evaluate the consequences of pH modulation on the structure and stability of the G- protein. We find that changes over the physiological pH range significantly alter the structure and stability of \(\mathrm{G}\alpha \mathrm{i}\) - GDP, with the protein undergoing a disorder- to- order transition as the pH is raised from 6.8 to 7.5. Further, we find that modulation of intracellular pH in HEK293 cells regulates \(\mathrm{G}\alpha \mathrm{i}\) - \(\mathrm{G}\beta \gamma\) release. Identification of key residues in the pH- sensing network allowed the generation of low pH mimetics that attenuate \(\mathrm{G}\alpha \mathrm{i}\) - \(\mathrm{G}\beta \gamma\) release. Our findings, taken together, indicate that pH- dependent structural changes in \(\mathrm{G}\alpha \mathrm{i}\) alter the agonist- mediated \(\mathrm{G}\beta \gamma\) dissociation necessary for proper signaling. + +<--- Page Split ---> + +Within the complex milieu of living cells, intracellular pH (pHi) is maintained within a narrow range, as even small changes in pH can affect a myriad of cellular processes including membrane potential, ion transport, cellular growth and metabolism1,2,3. Unsurprisingly, disruptions in pH regulation can contribute to the development of pathological conditions such as ischemic heart disease, cancer, and neurological disorders3,4. While various proton pumps and transporters play a role in regulating the flow of protons across the membrane to uphold pHi homeostasis, there are intracellular proteins (termed pH sensors) that sense and transmit pH signals, thus orchestrating the regulation of biochemical processes. Among these pH sensors are signal- transducing proteins. Notably, a study by Isom et. al. (2013) provided evidence that a subset of signal- transducing heterotrimeric G proteins may serve as intracellular pH sensors3. These membrane- associated proteins form a heterotrimeric complex (Gα, Gβ and Gγ) and act downstream of G- protein coupled receptors (GPCRs). GPCRs represent the most extensive group of membrane proteins that are targeted by approved drugs. They play a crucial role in orchestrating the majority of cellular responses to hormones and neurotransmitters through several signaling pathways mediated by different isoforms of G protein (Gαi, Gαs, Gα12/13, and Gαq) that receive and transduce signals through diverse pathways5,6. The Gβ and Gγ subunits associate to form a Gβγ heterodimer3 whereas the Gα subunit binds GDP or GTP and catalyzes GTP hydrolysis. The Gα subunit is comprised of two distinct domains: a helical domain and a Ras- like domain. Within the Ras- like domain, there are three key 'Switch' regions, namely SW- I, SW- II, and SW- III, which play an essential role in its activity (Fig. 2D). In the GDP- bound state, which is an inactive form of Gα protein, these Switch regions exhibit dynamic behavior. In contrast, in the GTP- bound state (the active form of Gα), they become more structured and less dynamic7. In the GDP- bound state, the Gα subunit is associated with the Gβγ complex. Upon GPCR stimulated Gα GTP- loading, the Gβγ subunits dissociate from Gα- GTP and along with Gα, promote activation of downstream signaling pathways. + +Isom et. al. (2013) developed a computer algorithm, pHinder, which predicted that both the mammalian Gαi isoform and yeast homolog Gpa1 contain a core of residues between the Ras- like and helical domains that may promote pH sensing properties8. In support of this prediction, both proteins showed pH- dependent changes in thermostability over a pH range from 5.5 to 8. + +<--- Page Split ---> + +They also found that Gpa1 undergoes phosphorylation under acidic conditions to attenuate pheromone- dependent stimulation of mitogen- activated protein kinases in the yeast4. While these findings, taken together suggest that mammalian Gai may function as a pH sensor, the molecular mechanism, and the biological consequence of pH sensing through Gai remain unknown. + +To expand on past observations, we characterized pH- dependent biochemical and structural/dynamic properties of GDP- bound Gai, identified the underlying pH- dependent electrostatic network, and determined functional consequences on cellular Gai activity. We find that the structure, stability, and dynamics of Gai in the GDP- bound state are highly dependent on pH over the physiological pH range due to a pH- responsive network within the Ras- like domain. These findings differ from a previous report where larger pH- dependent changes in thermostability were observed for the GTP- bound state with ionizable residues predicted to lie at the interface between the Ras- like and helical domains3. Our NMR, biophysical and computational analyses indicate that changes in the ionization state of residues within the pH sensing network promote a disorder- to- order transition in Gai over the physiological pH range, to populate a less ordered state than enhances Gai- Gβγ association in HEK293 cells at the lower end of the physiological pH range. Identification of the pH- sensing network allowed for the identification and generation of low pH mimetics that reduce pH- dependent Gβγ release from the Gai- Gβγ complex. Of note, the Gai Switch III region plays a key role in pH sensing, providing a new role of this understudied Switch region in agonist- mediated Gβγ release, a key important step for both Gai and Gβγ activation. Taken together, our studies indicate that Gai undergoes a pH- dependent order- to- disorder transition that modulates Gai- Gβγ interactions and Gai activation over the physiological pH range. + +## Materials and methods + +## Plasmid constructs + +A bacterial pET- SUMO vector containing the human Gai1 gene (GenBank accession no. BC026326) with the first 31 amino acids deleted, N- terminal 6xHis, and SUMO- tag were employed for in vitro analyses. Full- length human Ga containing Renilla luciferase (Ga- RLuc8), Gβ and Gγ containing green fluorescence protein 2 (Gγ- GFP2), and neurotensin receptor (NTR) constructs were employed for cell- based BRET assays4,8. + +<--- Page Split ---> + +## Site-directed mutagenesis + +Gai variants were generated using the Q5 Site- Directed Mutagenesis kit (NEB). Polymerase chain reaction (PCR) primers were designed using the NEBaseChanger (https://nebasechanger.neb.com), an online mutagenesis primer design tool powered by New England Biolabs from Eton Bioscience Inc. Mutagenesis was performed as described4. The variant constructs were sequenced by Sanger Sequencing (Genewiz) to confirm successful mutagenesis. Sequencing results for Gai variants were aligned with published sequences of wildtype (WT) Gai using Clustal Omega (European Molecular Biology Laboratory, Cambridgeshire, UK). + +## Expression and purification of WT Gai and its variants + +Gai proteins were overexpressed in E. coli Agilent BL21- codon plus (DE3)- RP- X competent cells. Cells were grown at \(37^{\circ}\mathrm{C}\) for 3- 4 hours to achieve an optical density at \(600\mathrm{nm}\) (OD 600) of 0.60 and then induced with \(500\mu \mathrm{M}\) isopropylthio- \(\beta\) - galactoside (IPTG). The temperature was then reduced to \(18^{\circ}\mathrm{C}\) and the cultures were left to grow overnight. After 20 hours, bacteria cells were harvested by centrifugation at 13,000 rpm and then resuspended in \(50\mathrm{mL}\) Dialysis buffer ( \(10\mathrm{mM}\) \(\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(1\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\beta\) - mercaptoethanol (BME), \(\mathrm{pH}7.0\) ). Resuspended cells were lysed using a sonicator (Fisher Scientific, #CL- 334) with an amplitude of 80 and 5 seconds on/off time ratio for 10 mins/L cells. Clarified lysate was passed through the Ni- NTA column equilibrated with 1 column volume (CV) (50 ml) of Dialysis buffer, washed with Wash buffer ( \(10\mathrm{mM}\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(1\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\mathrm{BME}\) , \(30\mathrm{mM}\) imidazole, \(\mathrm{pH}7.0\) ) and eluted using the Elution buffer ( \(10\mathrm{mM}\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(5\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\mathrm{BME}\) , \(250\mathrm{mM}\) imidazole, \(\mathrm{pH}7.0\) ). The elution fraction was dialyzed overnight in the presence of a ubiquitin- like- specific protease 1 (ULP1) to cleave the SUMO tag on Gai. After dialysis, the protein was again passed over a Ni- NTA column equilibrated with 3 CVs of water and 1 CV dialysis buffer. The protein flow- through was further purified by gel filtration chromatography using a Fast Protein Liquid Chromatography (FPLC) system (Akta Primeplus) and Superdex 75 column (Cytiva). The purity of the protein was verified by SDS PAGE gel electrophoresis, and the concentration of the purified protein was estimated by Thermo Scientific NanoDrop. + +## Circular Dichroism (CD) analyses + +<--- Page Split ---> + +Gai proteins were exchanged in CD buffer (10 mM \(\mathrm{K}_2\mathrm{HPO}_4\) , 500 \(\mu \mathrm{M}\) MgSO4, 500 \(\mu \mathrm{M}\) tris(2- carboxymethyl)phosphine (TCEP)) using an Amicon centrifugal filter unit (MilliporeSigma, #UFC901024) at \(3800\mathrm{xg}\) , diluted to \(15\mu \mathrm{M}\) (pH 6.0, 6.4, 6.8, 7.2, or 7.6) and centrifuged (14,000 rpm) for 10 mins at \(4^{\circ}\mathrm{C}\) . Temperature dependent CD experiments were performed on a Jasco J815CD spectrometer using \(15\mu \mathrm{M}\) WT or Gai variants protein in a \(1\mathrm{mm}\) path- length quartz cuvette (Hellma Analytics). Thermal melts were obtained at \(222\mathrm{nm}\) , over a temperature range of \(20–95^{\circ}\mathrm{C}\) , using a temperature increment of \(1\mathrm{C / min}\) . The CD spectral scans were collected for Gai proteins at different fixed temperatures (e.g. \(45^{\circ}\mathrm{C}\) , \(55^{\circ}\mathrm{C}\) and \(65^{\circ}\mathrm{C}\) ) by taking CD measurements every \(1\mathrm{nm}\) from \(200–250\mathrm{nm}\) . Secondary structure was evaluated using the online server BeStSel (Beta Structure Selection) \(^9\) . + +## Intrinsic tryptophan fluorescence + +Intrinsic tryptophan fluorescence assays were conducted using \(2\mu \mathrm{M}\) purified WT or variant GDP- loaded Gai proteins in \(200\mu \mathrm{l}\) of assay buffer (20 mM HEPES, \(50\mathrm{mMNaCl}\) , \(5\mathrm{mM}\mathrm{MgCl}_2\) , \(2\mathrm{mM}\) DTT) at different pH (pH 5, 6 and 7.2) in the 96 well plate. Gai proteins were excited at \(290\mathrm{nm}\) and intrinsic tryptophan fluorescence was measured from 300 to \(400\mathrm{nm}\) wavelength using a SpectraMax M4 Series Microplate Reader. + +## 2D NMR experiments + +Triple labeled \(^2\mathrm{H} / ^{13}\mathrm{C} / ^{15}\mathrm{N}\) Gai samples were generated by expressing Gai proteins in BL21(DE3)- RIPL E. coli cells and growing in minimal medium supplemented with \(1\mathrm{g / L}^{15}\mathrm{NH}_4\mathrm{Cl}\) , \(3\mathrm{g / L}^{13}\mathrm{C}\) - glucose, \(99\%\) D \(_2\mathrm{O}\) (Cambridge Isotope Labs) \(^7\) , and purified as described above. Triple labeled Gai was prepared in \(20\mathrm{mM}\) potassium phosphate ( \(\mathrm{pH} = 7.0\) ), \(50\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) DTT (dithiothreitol), \(5\mathrm{mM}\) magnesium chloride, and \(50\mu \mathrm{M}\) GDP containing \(5\%\) (v/v) D \(_2\mathrm{O}\) . For \(^{15}\mathrm{N}\) - enriched WT and Gai variants, bacterial expressed proteins were grown in minimal media containing \(1\mathrm{g / L}^{15}\mathrm{NH}_4\mathrm{Cl}\) as the sole nitrogen source and purified as described above. + +2D NMR \(^1\mathrm{H}–^{15}\mathrm{N}\) HSQC spectra of Gai were acquired on a Bruker Avance 850 MHz (14.1 T field strength) NMR spectrometer at \(25^{\circ}\mathrm{C}\) , with a cryogenic (TCI) \(5\mathrm{mm}\) triple resonance probe equipped with a z- axis gradient. The \(^1\mathrm{H}–^{15}\mathrm{N}\) HSQC spectra of Gai- GDP were assigned using a combination of triple resonance experiments, including 3D HNCA, HN(CO)CA, HN(CA)CO and HNCO \(^7\) . TROSY- based pulse sequences were used for sensitivity enhancement. Bruker- TopSpin + +<--- Page Split ---> + +was used to process the NMR data and NMRFAM- SPARKY was used to visualize and analyze the NMR spectra10. For assignments, BMRB 30078 was used as reference spectrum for Gaig- GDP11. pH titration studies were performed by calculating the chemical shift perturbation (CSP) of Gaig- GDP over a pH range from 6.4, 6.8, 7.0, 7.2, 7.4 and 7.6. Average 1H- 15N CSP were determined using the formula \(\Delta \delta = [(\Delta^{1}\mathrm{HN})^{2} + (\Delta^{15}\mathrm{N} / 5)^{2}]^{0.5}\) , as previously described12. PyMOL (https://pymol.org/2/) was used to generate all images of molecular structures. + +## Guanine nucleotide dissociation assay + +Gaig proteins were exchanged into nucleotide association buffer (20 mM HEPES, 50 mM NaCl, 5 mM MgCl2, 2 mM DTT) using an amino concentrator (3,800 g, 15 min, 3 rounds). In the cuvette, 0.75 μM (2'-(or- 3')- O-(N- Methylanthraniloyl) Guanosine 5'- Diphosphate, Disodium Salt (Mant- GDP, purchased from Thermofisher) was added to 1 ml of association buffer for the assay. The intrinsic tryptophan (W211) was excited at 280 nm and the Mant- GDP fluorescence intensity at 425 nm was measured as a function of time using a PerkinElmer LS55 luminescence spectrometer. After collecting data for a few seconds to obtain baseline fluorescence, purified 1 μM Gaig- GDP was added to the solution to initiate GDP association. Once the fluorescence reached saturation, 10 x (7.5 μM) GDP was added to the solution to initiate the GDP dissociation. + +## Molecular Dynamics (MD) simulations + +Gaig- GDP coordinates were extracted from the Gaig- Gβγ complex structure (PDB: 1GP2) to visualize Switch residues not observable in Gaig- GDP structures13. Modeller 9v21.2 was employed to relocate missing residues and generate charged and uncharged states for residues E236, D237 and E24514 to represent the charged state as higher pH (above 7.2) and the uncharged state as low pH (below 6.4). Following side- chain optimizations, minimum energy conformations of charged and uncharged GDP- bound Gaig states were identified. Subsequently, MD simulations were performed to investigate conformational and dynamic differences compared to WT GDP- bound Gaig. The CHARMM36 forcefield was used to parametrize the protein and GDP, and simulations were run using GROMACS- 2020.3.515. Gaig proteins were solvated in a water cubic box containing approximately 22,500 TIP3P water molecules and maintained a salt concentration of 150 mM through the addition of an appropriate number of Na+ and Cl- ions. The solvated system was energy minimized using the conjugate gradient algorithm and subsequently equilibrated using the V- rescale thermostat and the Parrinello- Rahman barostat16. Long- range electrostatic interactions + +<--- Page Split ---> + +were evaluated using the Particle- Mesh Ewald sum17, and all bonds involving hydrogen atoms were constrained using the LINCS algorithm18. Simulations were run for 250 ns and were repeated in triplicate to ensure reproducibility and maintain consistency. Structural figures were generated using PyMOL and graphical plots were created using Grace (http://plasma- gate.weizmann.ac.il/Grace/). + +## HADDOCK docking + +The HADDOCK online server was used for the computational docking of \(\mathrm{Ga}\) with \(\mathrm{G}\beta \gamma^{19}\) . The initial structure of \(\mathrm{Ga}\) with E236, D237, and E245 residue side chains in either the charged (higher \(\mathrm{pH}\) ) or uncharged state (lower \(\mathrm{pH}\) ) as described above, was obtained from the averaged structure of 250 ns MD simulations. The \(\mathrm{G}\beta \gamma\) structure was extracted from the \(\mathrm{Ga}\) - \(\mathrm{G}\beta \gamma\) complex crystal structure (PDB:1GP2). Contact residues between \(\mathrm{Ga}\) and \(\mathrm{G}\beta \gamma\) that form the trimeric complex structure were used as active residues and the passive residues were auto selected by the software. The docking model corresponding to the lowest energy was used for further analysis. + +## Cell culture + +HEK293T cells were maintained, passaged, and transfected in Dulbecco's Modified Eagle Medium (DMEM, Gibco) containing \(10\%\) sterilized Fetal Bovine Serum (FBS, Gibco), 100 Units/mL penicillin, and \(100 \mu \mathrm{g} / \mathrm{mL}\) streptomycin (Gibco- ThermoFisher, Waltham, MA) in a humidified atmosphere at \(37^{\circ}\mathrm{C}\) and \(5\% \mathrm{CO}_{2}\) . Transfection was carried out using a lipid- polymer- based transfection agent (Mirus Bio, MIR 5400, Madison, WI). After transfection, cells were plated in DMEM (Dulbecco's Modified Eagle Medium) containing \(10\%\) FBS (Fetal Bovine Serum), 100 Units/mL penicillin, and \(100 \mu \mathrm{g} / \mathrm{mL}\) streptomycin for BRET assays. + +## Intracellular pH modulation: + +The intracellular pH of HEK293 cells was altered by two different methods. In the first method, HEK293 cells were treated with \(1 - 5 \mu \mathrm{M}\) of trifluoromethoxycarbonylcyanide phenylhydrazone (FCCP) for 5 min. In the second method, intracellular pH was modulated by incubating HEK293 cells with Hanks' Balanced Salt Solution (HBSS) buffer at different pH for 15 min. Intracellular pH was quantified using the intracellular pH indicator, \(2',7'\) - bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein-acetoxymethyl ester (BCECF-AM) as described20. To convert BCECF-AM + +<--- Page Split ---> + +fluorescence to the intracellular pH values, HEK293 cells were treated with \(20 \mu \mathrm{M}\) nigericin and a calibration curve for \(\mathrm{pH}_i\) and BCECF- AM fluorescence was generated. + +## Bioluminescence Resonance Energy Transfer (BRET) assays + +BRET assays were carried out as described by Olsen et al. (2020) \(^{8}\) . Briefly, HEK293T cells were seeded in six- well dishes in DMEM containing \(10\%\) FBS, 100 Units/mL penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin at a density of 300,000 cells/well and allowed to grow to \(80 - 90\%\) confluency. Cells were then transfected using a 1:1:1:1 DNA ratio of NTR: \(\mathrm{Ga}\) - RLuc: \(\mathrm{G}\beta\) : \(\mathrm{G}\gamma\) - GFP2 (1000 ng/construct for six- well dishes). TransIT- 2020 (Mirus Biosciences, Madison, WI) was used to complex the DNA at a ratio of \(3 \mu \mathrm{L}\) Transit/ \(\mu \mathrm{g}\) DNA, in OptiMEM (Gibco- ThermoFisher, Waltham, MA) at a concentration of \(1 \mu \mathrm{g}\) DNA/ \(\mu \mathrm{L}\) OptiMEM. The next day, cells were harvested from the plate using trypsin (0.25% EDTA, Gibco, #25200056), and plated in poly- D- lysine- coated white, clear bottom 96- well assay plates (Greiner Bio- One, Monroe, NC) in DMEM containing \(1\%\) dialyzed FBS, 100 Units/mL penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin at a density of 50,000 cells/well. + +One day after seeding in 96- well assay plates, white backings (Perkin Elmer, Waltham, MA) were applied to the plate bottoms, and the growth medium was aspirated. Cells were washed three times with Assay buffer (1X HBSS, \(20 \mathrm{mM}\) HEPES, \(\mathrm{pH} 7.4\) ), then \(60 \mu \mathrm{L}\) of Assay buffer was added immediately, followed by a \(10 \mu \mathrm{L}\) addition of freshly prepared \(50 \mu \mathrm{M}\) coelenterazine 400a (1- bisdeoxycoelenterazine) (Nanolight Technologies, Pinetop, AZ) in each well. After a five- minute equilibration period, cells were treated with \(30 \mu \mathrm{L}\) of \(3 \mathrm{x}\) neurotransmitters ( \(3 \times 10^{- 5} \mathrm{M}\) ) for another 5 minutes. Plates were then read in a plate reader (Clariostar, BMG Labtech, Ortenberg, Germany) at \(395 \mathrm{nm}\) (Rluc8- coelenterazine 400a) and \(510 \mathrm{nm}\) (GFP2) with a slit width of \(8 \mathrm{nm}\) using 10 flashes per spiral well scan. Plates were read serially five times, and measurements from the third read were used in all analyses. BRET ratios were computed as the ratio of the GFP2 emission to Rluc8 emission. + +## Statistical Analysis + +CD melt- curves were fit to Boltzmann sigmoidal equation in Prism 9.3.1 (GraphPad Software, San Diego, CA). Concentration- response curves for BRET assays were fit to a three- parameter logistic equation. Raw BRET concentration- response curves were normalized to the best- fit maximum + +<--- Page Split ---> + +within a data set. BRET data were represented as mean \(\pm\) SEM. Data analysis was carried out using Prism 9.3.1 (GraphPad Software, San Diego, CA). + +## Results + +## pH modulation of GDP-bound Gai structure and dynamic properties + +Earlier investigations into mammalian Gai demonstrated changes in structure and stability in response to pH variations8. While these findings suggest a potential role for Gai as a pH sensor8, the molecular basis and functional relevance has yet to be established. Herein, we apply comprehensive and multidisciplinary approaches to evaluate how pH changes in the physiological range affect structure, stability and Gai1 activity in vitro and in cells. We first conducted CD experiments on GDP- bound Gai to monitor pH- dependent changes in thermal unfolding, stability and secondary structure. Further, to monitor thermal unfolding and stability, we collected CD spectra at \(222 \mathrm{nm}\) as a function of pH and temperature. As shown in Fig. 1A, a striking and gradual increase in thermal stability ( \(\mathrm{T_m} \sim 25^{\circ} \mathrm{C}\) ) of Gai- GDP was observed over the pH range from 5 to 7.3. Notably, the thermal unfolding transition for Gai- GDP appears cooperative at low pH but shifts to a multi- state unfolding transition over the physiological pH range (Fig. 1A). Gai contains two subdomains, a Ras- like and helical domain. To evaluate differential unfolding at higher pH (above pH 7.1), we conducted CD scans (200- 250 nm) for Gai- GDP as a function of temperature at pH 7.2. As shown in Fig. 1B and C, CD spectra revealed a significant reduction in alpha- helical propensity, but not the beta- sheet propensity as the temperature is raised from \(45 - 55^{\circ} \mathrm{C}\) . These findings indicate that the helical domain melts first, followed by the Ras- like domain, with the Ras- like domain showing the greatest change in thermal stability at higher pH. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Ga1-GDP thermostability and intrinsic tryptophan fluorescence are pH dependent. (A) Representative CD thermal melt (222 nm, 20–95 °C) of Ga1-GDP (15 μM) shows enhanced thermostability at higher pH (N = 3). (B) Representative CD spectral scans (200-250 nm) of Ga1-GDP collected at 45 and 55°C and pH 7.2 (N = 2). (C) Bar graph derived from temperature-dependent CD spectral scans highlighting the loss of alpha-helical but not beta-sheet secondary structure. (D) Representative intrinsic tryptophan fluorescence spectra of Ga1-GDP (2 μM, excitation = 280 nm, emission = 300 - 400 nm) shows enhanced fluorescence as the pH is increased from 5 to 7.2 (N = 2, performed in triplicate).
+ +Given the enhanced stability observed at higher pH, we employed intrinsic tryptophan fluorescence experiments using tryptophan (W211) in the SW-II region (Fig. 3A) as a fluorescence probe to monitor differences in solvent exposure as a function of pH which indirectly indicates pH- dependent switch conformational changes. It has previously been shown that the more dynamic and less ordered GDP- bound state of Ga1 promotes enhanced exposure of W211 resulting in a fluorescence decrease relative to that of the Ga1- GTP state7,21. As shown in Fig. 1D, intrinsic Ga1- GDP tryptophan fluorescence increases over the pH range from 5 to 7.2, suggesting that higher pH promotes decreased solvent accessibility possibly due to enhanced interactions and structural order, consistent with the greatly enhanced stability observed by CD. Taken together, these results support a disorder- to- order transition at higher physiological pH for Ga1 in its GDP- bound form. + +<--- Page Split ---> + +To further probe whether a pH- dependent disorder- to- order transition occurs in GDP- bound Gai, we conducted 2D NMR analyses. For these studies, we prepared \(^{15}\mathrm{N}\) enriched Gai- GDP and collected a 2D \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) Heteronuclear Single Quantum Coherence (HSQC) NMR spectra over a physiologically relevant pH range (6.4 - 7.6). Enrichment with \(^{15}\mathrm{N}\) allows the detection of backbone and sidechain NH resonances within Gai and provides a site- specific probe for every amino acid except proline. The 2D HSQC overlay of Gai- GDP at pH 6.4, 6.8 and 7.2 is shown in Fig. 2A, with an HSQC overlay comparing pH 6.4 versus pH 7.6 shown separately in Supplementary Fig. S1. Spectra acquired at lower pH show significant chemical shift changes, line broadening and loss of several peaks in comparison to those obtained at higher pH values, suggesting the protein is more dynamic at lower pH (Fig. 2A). This data correlates well with CD and tryptophan fluorescence data which suggests a less thermostable structure of Gai- GDP at lower pH. The pH- dependent HSQC changes observed are consistent with an earlier report which showed extensive broadening of NH peaks in the \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) HSQC spectrum of Gai- GDP at pH 6 relative to pH \(7^{22}\) . + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: Gai-GDP \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) 2D NMR spectral changes as a function of pH. (A) Representative 2D \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) TROSY-HSQCP spectral overlay of \(^{2}\mathrm{H}\) , \(^{13}\mathrm{C}\) , \(^{15}\mathrm{N}\) -enriched Gai-GDP (230 \(\mu \mathrm{M}\) ) at pH 6.4, 6.8, and 7.2, highlighting peak shifts and line broadening. Spectra were acquired on a Bruker Avance III 850 MHz instrument at \(25^{\circ}\mathrm{C}\) ( \(\mathrm{N} = 2\) ). (B) Chemical shift perturbation (CSP) and (C) peak intensity ratio \((\mathrm{Int}_{\mathrm{pH}6.4} / \mathrm{Int}_{\mathrm{pH}7.2})\) for pH 6.4 and 7.2. The striped horizontal line represents the mean value of \(\Delta \delta\) amide. Most of the residues that show significant CSP, and broadening lie within the Ras-like domain \((\alpha 1,\alpha 5,\beta 1,\beta 2\) and \(\beta 2 - \beta 3\) loop). (D) Spectral differences are highlighted on a ribbon diagram of GTP-bound Gai (PDB: 1CIP). NH residues with CSP greater than 0.03 ppm are represented in red. Residues with decreased peak intensity (line broadening) at pH 6.4 relative to pH 7.2 are shown in green. Residues missing or unassigned are shown in black color while unaffected residues are shown in blue color.
+ +Residue- specific chemical shift perturbation and peak intensity changes associated with changes in pH between 6.4 and 7.2 are plotted in Fig. 2B and C, respectively. Most pH- dependent spectral changes are localized to \(\alpha 1\) , \(\alpha 5\) , \(\beta 1\) , \(\beta 2\) and the \(\beta 2 - \beta 3\) loop within the Ras- like domain as mapped + +<--- Page Split ---> + +onto the Gai- GTP structure (PDB:1CIP) in Fig. 2D. Of note, several resonances within these key regions are undetectable in the GDP- bound state at all pH values (Fig. 2D, black) which somewhat limits NMR analyses. These findings are consistent with previous work showing that residues associated with the Switch regions in the GDP- bound state of Gai exhibit enhanced backbone dynamics and are broadened and undetectable compared to the resonances associated with the GTP- bound state7. Taken together, 2D NMR analyses, CD data and tryptophan fluorescence data indicate that the Switch regions of Gai- GDP are more dynamic and less structured at pH 6.4 - 6.8 compared to pH 7.2 - 7.5. + +## pH-dependence of GDP binding to Gai + +Gai when bound to GDP, adopts a conformational ensemble and dynamic properties distinct from that of the GTP- bound state7. This in turn drives recognition of regulatory factors and downstream targets. Given our findings that pH modulates Gai- GDP structure, stability and dynamics, we asked if pH could alter GDP binding to Gai protein. For that purpose, we performed Mant- GDP dissociation assays. For these assays, the rate of GDP dissociation was determined by adding excess GDP to Mant- GDP loaded Gai and monitoring Mant fluorescence changes (by FRET upon tryptophan excitation) as a function of time at different pH values (Fig. 3A- B). As shown in Fig. 3C- D, small differences in GDP dissociation rates were observed over the pH range (pH 6.8- 7.4), indicating that GDP binding is not significantly altered at physiological pH. + +## Identification and characterization of pH sensing residues of Gai-GDP + +To elucidate the molecular basis for pH- dependent stability and conformational dynamic changes in GDP- bound Gai, we sought to identify key residues involved in pH sensing. As extensive broadening of resonances in GDP- bound spectra prevented pKa determination by NMR, we examined and identified two networks of charged residues in regions, designated as the ‘GDP release network’ (Fig. 4A) and ‘Switch network’ (Fig. 5A), that showed pH- dependent changes in the NMR spectra (Fig. 2D). The GDP release network contains residues within \(\alpha 1\) , \(\alpha 5\) , \(\beta 2\) , and \(\beta 3\) , and was previously shown to be important for GDP release during the GPCR- mediated activation of the Gai23. We postulated that if this network plays a key role in pH sensing, mutation of charged residues within this network (e.g., H57, H188, K192, D193, H195 and D337) would reduce pH- dependent Gai thermostability, due to protonation and loss of electrostatic interactions that destabilize Gai tertiary structure. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: Rates of Gai GDP nucleotide dissociation vary with pH. (A) Structure of Gai (PDB:1GP2) highlighting the position of tryptophan (W211) in Switch II and bound Mant-GDP. (B) Diagram displaying experimental setup of FRET-based determination of Gai nucleotide association and dissociation rates. (C) Rate of GDP dissociation from Gai as a function of pH by monitoring the time-dependent decrease in FRET emission of Gai-loaded Mant-GDP at 445 nm upon the addition of \(7.5 \mu \mathrm{M}\) GDP. (D) The rate of Mant-GDP dissociation decreases as the pH increases. Data are averages of two independent experiments performed in duplicate (±SE).
+ +To examine whether this network modulates pH- dependent changes in stability and nucleotide binding, we generated several Gai variants and conducted pH- dependent CD thermal melt and nucleotide dissociation assays. To select neutral or uncharged amino acid substitutions that retain or have minimal effect on Gai structure, we employed the Rosetta modeling suite. Rosetta replaces a desired amino acid within the protein to all possible amino acid substitutions and predicts associated free energy changes24. Substitutions that minimally perturb free energy are predicted to retain protein structure. Using this strategy, we identified four variants (H57T, H188V, K192Q + +<--- Page Split ---> + +and H195N) predicted to retain Gai structure and stability. To evaluate pH- dependent thermostability associated with these Gai GDP release network variants, we acquired CD thermal melts at pH 6 and 7.2, as shown in Fig. 4. Of note, all four variants retained pH- dependent thermostability similar to WT Gai (data not shown for H188 and H195) (Fig. 4C- E). We also performed GDP dissociation assays to evaluate whether the variants alter nucleotide binding (Fig. 4B). All variants showed pH- dependent GDP dissociation rates similar to WT Gai. These findings indicate that the GDP release network does not significantly modulate pH- dependent stability or nucleotide binding (Fig. 4B- F). + +![](images/Figure_4.jpg) + +
Fig. 4: pH-dependent nucleotide exchange and stability of Gai variants from the GDP release network. (A) Ribbon diagram of GTP-bound Gai (PDB: 1CIP) highlighting electrostatic network near GDP release network. The helical domain, Ras-like domain and the Switch regions are shown in gray, blue, and green, respectively. Charged residues are shown as yellow sticks. (B) Rates of GDP dissociation is compared for WT Gai and Gai variants (H57T and K192Q) as a function of pH. Data are averages of two independent experiments performed in duplicate (±SE) (C) Representative CD melt profile of GDP-bound WT and variant Gai (N=2), Gai H57T (D) and K192Q (E) proteins. Gai release network variants retain pH-dependent thermal unfolding profiles similar to WT Gai.
+ +As a number of charged residues within the Switch regions form stabilizing electrostatic interactions in the active Gai- GTP bound state, we next postulated that the decreased stability of Gai- GDP at lower pH may result from protonation of pH- dependent Switch network (SW- I, SW- II, SW- III and \(\alpha 3\) ) residues. To test this hypothesis, we selected residues from the Switch network + +<--- Page Split ---> + +shown or predicted to be important for Switch stability in the GTP- bound form of Gai25. Then, based on Rosetta prediction, we mutated a subset of charged residues within this network (e.g. R205N, R208Q, E236L, D237G, R242Q and E245N) predicted to least perturb Gai structure (Fig. 5A). As shown in Fig. 5B- E, two of the variants located in SW- III (E236L and D237G) and \(\alpha 3\) (E245N), respectively, showed a reduction in pH- dependent thermostability between pH 6 and 7.2 relative to WT Gai. Moreover, a ‘double variant’ consisting of two substitutions (E236L and E237G) from SW- III showed further reduction in pH- induced stability but not complete abolishment of pH dependence (Fig. 5F). Yet, a ‘triple variant’ containing all three substitutions (Gai E236L, D237G and E245N) effectively eliminated pH- induced stability changes relative to WT Gai (Fig. 5G), suggesting a key role for these three Switch network residues (E236, D237 and E245) in stabilizing Gai at higher pH. Other variants from the Switch network at R205 and R208 did not show any change in pH- dependent thermostability in comparison to WT Gai protein (Supplementary Fig. S2). We also monitored tryptophan fluorescence of the GDP- bound Gai triple variant as a function of pH to examine changes in solvent accessibility of SW- II residue W211. Consistent with the loss in pH- dependent thermal stability, we found that the double variant and triple variant reduce and abolish (Supplementary Fig. S3) pH- dependent fluorescence intensity changes relative to WT Gai- GDP, respectively (Fig. 5H). As the CD thermal and fluorescence profiles associated with the double variant and triple variant mimic WT Gai at lower pH (pH 6), we refer to these variants as “low pH mimetics”. Further to confirm that the identified residues participate in pH sensing, we performed NMR analyses for the double variant at pH 6.4 and pH 7.2 in comparison to WT GDP- bound Gai. As shown in Supplementary Fig. S4, pH- dependent changes in peak intensity and chemical shift perturbations are significantly reduced in the double variant with respect to WT Gai. Taken together, our results point to key residues in the Switch network that regulate pH- dependent stability and conformational dynamic properties. MD simulations of pH- dependent Gai- GDP conformational dynamics We identified three residues in the GDP- bound Gai, including two SW- III residues (E236, D237) and one \(\alpha 3\) residue (E245), that appear to play a key role in pH- dependent stability and conformational dynamics. To evaluate how these residues form pH- dependent electrostatic interactions that stabilize the Switch regions at higher pH, we employed MD simulations. We generated Gai structures using the Gai- GDP crystal structure (PDB: 1GP2) as a starting point and then changed the protonation state of side chains associated with residues E236, D237 and + +<--- Page Split ---> + +E245 to simulate a low pH and high pH state, respectively. + +![](images/Figure_5.jpg) + +
Fig. 5. pH-dependent stability and tryptophan fluorescence of Gai Switch network variants. (A) The putative pH-dependent electrostatic network (red) within the Switch regions (green) is highlighted on the ribbon diagram of
+ +GTP- bound Gai (PDB: 1CIP). (B) Comparison of representative CD melt profiles obtained from two independent experiments at pH 6.0 and 7.2 for WT Gai- GDP, (C) Gai- GDP E245N, (D) Gai- GDP E236L, (E) Gai- GDP D237G (F) Gai- GDP double variant (E236L/D237G), and (G) a Gai- GDP triple variant (E236L/D237G/E245Q). The CD thermal profile for Gai single and double variants shows decreased pH- dependent thermal stability while the triple variant shows a complete loss of pH- dependent thermal stability compared to WT Gai. (H) Representative intrinsic tryptophan fluorescence spectra of Gai- GDP triple variant (2 \(\mu \mathrm{M}\) , excitation = 280 nm, emission = 300 - 400 nm) as a function of pH (N = 2, performed in triplicate). Consistent with the CD results, pH- dependent intrinsic tryptophan (W211) fluorescence observed for WT Gai- GDP is abolished for the Gai triple variant supporting a role for E236, D237 and E245 in pH- dependent stability and structural changes. + +thermal profile for Gai single and double variants shows decreased pH- dependent thermal stability while the triple variant shows a complete loss of pH- dependent thermal stability compared to WT Gai. (H) Representative intrinsic tryptophan fluorescence spectra of Gai- GDP triple variant (2 \(\mu \mathrm{M}\) , excitation = 280 nm, emission = 300 - 400 nm) as a function of pH (N = 2, performed in triplicate). Consistent with the CD results, pH- dependent intrinsic tryptophan (W211) fluorescence observed for WT Gai- GDP is abolished for the Gai triple variant supporting a role for E236, D237 and E245 in pH- dependent stability and structural changes. + +Analysis of MD simulation trajectories of GDP- bound Gai collected for 250 ns in charged versus uncharged states revealed higher root mean square deviation (RMSD) in the protonated or + +<--- Page Split ---> + +uncharged state. As shown in Fig. 6A, the RMSD of the uncharged state indicated higher dynamics, with RMSD values around \(0.4 \mathrm{nm}\) , while the charged state exhibited lower RMSD values of approximately \(0.25 \mathrm{nm}\) . Based on analyses of the MD trajectories, we attribute the RMSD reduction associated with the charged state to the formation of salt- bridge interactions within the Switch regions. To further probe the dynamic properties of the Gai- GDP Switch regions, we calculated residue- specific root mean square fluctuation (RMSF) in the uncharged state (Fig. 6B and C) and mapped RMSF changes onto the Gai- GTP crystal structure (PDB:1CIP). As shown in Fig. 6B and C, our simulations suggest increased dynamics of SW- I, SW- II, and SW- III in the uncharged state compared to the charged state, consistent with reduced thermal stability at lower pH. Overall, these findings suggest that at lower pH (pH 6.4- 7), there is a loss of electrostatic interactions within the Switch regions of GDP- Gai that promotes enhanced dynamics, consistent with observations from NMR and CD data. + +Representative snapshots extracted from the MD trajectories of Gai- GDP provide insights into the dynamic behavior of Gai as a function of pH. In the charged or higher pH state of Gai- GDP, we observe the transient formation of three critical salt- bridge interactions: E236 interacting with R205, E245 with R208, and E245 with K248 (Fig. 6D and F). Notably, these residues form electrostatic interactions as observed in crystal structures of active Gai- GTP state (PDB:1CIP) and play a pivotal role in the stabilization of the Switch regions. In particular, E236 and D237 side chains from SW- III interact with R205 and R208 in SW- II whereas \(\alpha 3\) residue E245 interacts with R208 \(^{25}\) . In the uncharged state of Gai- GDP (lower pH), conformations extracted from MD simulation trajectories exhibit a notable absence of these salt- bridge interactions (Fig. 6E), which we attribute to the destabilization of the SW- III and SW- II region (Fig. 6E). As a result, the protein displays increased dynamics, as evidenced by higher RMSD and RMSF values, suggesting greater structural fluctuations. The protonation of key residues interferes with the formation of electrostatic interactions leading to their destabilization. Consequently, the protein becomes more dynamic and less thermally stable under more acidic conditions. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6: MD simulations of WT Gai-GDP highlighting uncharged and charged states of E236, D237 and E245.
+ +To mimic the low pH state of Gai- GDP (uncharged state), E236, D237 and E245 side chains were protonated using the Gai- GDP crystal structure (PDB: 1GP2). MD simulations of Gai- GDP in both charged (deprotonated) and uncharged states (protonated) were performed for 250 ns. (A) Root means square deviation (RMSD) of Gai Cα atoms in charged (black) and uncharged (red) states, as estimated by superposing each frame of the trajectory to the corresponding starting structure. Gai- GDP shows higher RMSD in the uncharged state indicative of enhanced dynamics consistent with NMR at lower pH. (B) Residue- wise root mean square fluctuation (RMSF) derived from MD trajectory (100 ns) of charged (black) and uncharged (red) Gai- GDP, highlights enhanced Switch dynamics in the uncharged state. (C) Mapping of residues that display higher RMSF (orange) on the Gai- GTP crystal structure (1CIP) indicates that the SW- I, SW- II and SW- III become more dynamic in the uncharged state compared to the charged state, consistent with the reduced thermal stability and intrinsic tryptophan fluorescence at lower pH. (D) Time evolution of Gai - GDP salt- bridge formation associated with charged trajectory shows three salt- bridges in the charged state but not in the uncharged state of Gai. The cut- off value of salt- bridge distance = 3.2 Å is shown as a dotted line. (E) Representative conformation of Gai- GDP in the uncharged (protonated) state extracted from the MD trajectory shows a lack of salt- bridge interactions due to the destabilization of the SW- III region. (F) Zoomed section depicting salt- bridge interactions extracted from the corresponding MD trajectory (E236- R205, E245- R208, and E245- K248) for the average conformation of charged Gai- GDP. + +To predict how a pH- dependent disorder- to- order transition in Gai- GDP affects the interaction of Gai with binding partners, we analyzed interactions with GPCRs or \(\mathrm{GB}\gamma\) . Agonist- simulated GPCRs interact with Gai- GDP through \(\alpha 5\) , which is part of the GDP release network. As inspection of the crystal structure of \(\beta 1\) - adrenergic receptor with Gai and \(\mathrm{GB}\gamma\) (PDB: 7S0F) + +<--- Page Split ---> + +indicates GPCR engagement does not alter the structure of Switch II and III26, we predict that pH changes over the physiological pH range do not significantly modulate GPCR interactions with Gai. Conversely, \(\mathrm{G}\beta \gamma\) binds to Gai- GDP through the SW- II region which is more dynamic in the GDP- bound form than it is in the GTP- bound state. We propose, consistent with a recent study25, that residues from the Switch network (including the three identified pH sensing residues) provide stabilization of SW- II in the GTP- bound form of Gai which in turn, prevents interaction with \(\mathrm{G}\beta \gamma\) . Since our MD data show enhanced dynamics in the protonated or lower pH state of Gai- GDP, we predict that at the lower end of the physiological pH range, Gai engages \(\mathrm{G}\beta \gamma\) with higher affinity resulting in downregulation of Gai and \(\mathrm{G}\beta \gamma\) mediated downstream signaling (Fig. 8). To evaluate this hypothesis computationally, we performed HADDOCK docking of Gai- GDP with \(\mathrm{G}\beta \gamma\) in both deprotonated charged and protonated uncharged states. The Gai structures in the charged and uncharged states were obtained as described above, while the \(\mathrm{G}\beta \gamma\) structure was extracted from the Gai- \(\mathrm{G}\beta \gamma\) trimeric complex crystal structure (PDB: 1GP2). The HADDOCK docking score, which is a weighted sum of a variety of energy terms (including van der Waals, electrostatic, desolvation, and restraint violation), was calculated to be - 147 and - 203 for the charged and uncharged states, respectively. Evaluation of this docking score suggested higher affinity binding of Gai with \(\mathrm{G}\beta \gamma\) in the uncharged state due to additional polar contacts between Gai and \(\mathrm{G}\beta \gamma\) , implying that low pH enhances the binding of Gai to \(\mathrm{G}\beta \gamma\) . Taken together, our computational analyses suggest that Gai- \(\mathrm{G}\beta \gamma\) interactions are modulated by pH. + +## Analyses of pH-dependent Gai- \(\mathrm{G}\beta \gamma\) interactions + +\(\mathrm{G}\beta \gamma\) interacts with the dynamic SW- II region of Gai in the GDP- bound state. Upon upstream activation by GPCRs, Gai becomes activated through the exchange of GDP for GTP. In the GTP- bound state, the SW- II region becomes rigid and Gai is unable to adopt a conformation compatible with binding to \(\mathrm{G}\beta \gamma\) . Given our findings that Gai adopts a more dynamic state at the lower end of the physiological pH range (6.8- 7), we hypothesized that Gai- GDP engages \(\mathrm{G}\beta \gamma\) with a higher affinity at lower pH, whereas higher pH (7- 7.5) promotes \(\mathrm{G}\beta \gamma\) release from Gai- GDP. Consistent with this hypothesis, our MD analyses indicate that pH- mediated changes in the protonation state of key Switch residues alter electrostatic interactions and the dynamic properties of Switch regions of Gai in its GDP- bound form which may, in turn, modulate Gai- \(\mathrm{G}\beta \gamma\) interactions and thus the downstream signaling (Fig. 8). To experimentally evaluate the pH dependence of Gai- \(\mathrm{G}\beta \gamma\) + +<--- Page Split ---> + +interactions in a cellular context, we conducted bioluminescence resonance energy transfer (BRET) assays in human HEK293T cells. For these assays, WT Gai was tagged with Renilla luciferase (RLuc) and co- transfected with \(\mathrm{G}\gamma\) tagged with GFP, \(\mathrm{G}\beta\) and the neurotensin receptor. Upon stimulation with neurotensin (agonist), the \(\mathrm{G}\alpha\) - RLuc (energy donor) and \(\mathrm{G}\beta \gamma\) - GFP (energy acceptor) dissociate, with receptor- catalyzed dissociation of the \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) complex measured by comparing energy transfer from donor to acceptor (Fig. 7A). To directly evaluate whether lower intracellular \(\mathrm{pH}(\mathrm{pH}_i)\) reduces \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) dissociation, we employed two distinct approaches to alter and monitor \(\mathrm{pH}_i\) changes in HEK293 cells. Intracellular \(\mathrm{pH}\) was determined by treating HEK293 cells with BCECF dye and determining the ratio of emission intensity (detected at \(535~\mathrm{nm}\) ) when the dye is excited at \(\sim 490 \mathrm{nm}\) and \(\sim 440 \mathrm{nm}^{20}\) . To convert the fluorescence ratios to \(\mathrm{pH}_i\) , a calibration curve of HEK293 cells treated with \(5 \mu \mathrm{M}\) nigericin was generated (Supplementary Fig. S5A). Treatment of cells with nigericin makes the \(\mathrm{pH}_i\) equivalent to extracellular \(\mathrm{pH}(\mathrm{pH}_e)\) thus the calibration curve for \(\mathrm{pH}_e\) vs. fluorescence can be used to convert fluorescence into cellular \(\mathrm{pH}_i\) . To alter the \(\mathrm{pH}_i\) , we compared results using two different strategies. One common method to reduce \(\mathrm{pH}_i\) is the addition of Carbonyl cyanide 4- (trifluoromethoxy) phenylhydrazone (FCCP), an uncoupler of oxidative phosphorylation in the mitochondria \(^{27}\) . We determined that cells treated with \(1 \mu \mathrm{M}\) FCCP generated a \(\mathrm{pH}_i\) of \(\sim 7.04\) , while at \(5 \mu \mathrm{M}\) the \(\mathrm{pH}_i\) was further reduced to 6.85 (Supplementary Fig. S5B). As one drawback of FCCP is that it can alter mitochondrial energetics, we employed a second method to alter \(\mathrm{pH}_i\) by modulating extracellular \(\mathrm{pH}\) . As shown in Supplementary Fig. S5C, lowering extracellular \(\mathrm{pH}\) to 6 and 5, caused a reduction in \(\mathrm{pH}_i\) from 6.9 and 6.7, respectively which is consistent with a previous report \(^{28}\) . To evaluate the \(\mathrm{pH}\) dependence of \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) dissociation in cells, we conducted the BRET assay by expressing WT Gai in HEK293 cells and inducing acidosis in the cytoplasm by either altering extracellular \(\mathrm{pH}\) (Fig. 7B) or by the addition of varying concentrations of FCCP (Fig. 7C). Results obtained from the BRET assays show that lowering \(\mathrm{pH}_i\) ( \(\mathrm{pH}_i\) 7.2- 6.85 by FCCP or 7.2- 6.7 by changing \(\mathrm{pH}_e\) ) reduces agonist- mediated \(\mathrm{G}\beta \gamma\) release from \(\mathrm{G}\alpha \mathrm{i}\) , indicating that \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) interactions are highly dependent on \(\mathrm{pH}\) over the physiological \(\mathrm{pH}\) range. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Fig. 7: BRET assay showing enhanced Gai-Gβγ association at lower intracellular pH for the Gai Switch variants. (A) Schematic diagram (BioRender) illustrating the BRET assay used to monitor receptor-mediated dissociation of Gai from Gβγ. For this assay, the GPCR neurotensin receptor, Ga Renilla luciferase 8 (Rluc8), Gβ3 and γ9-GFP fusion constructs were co-transfected in HEK29T cells and fluorescence (510 nm) monitored after addition of the substrate and agonist (neurotensin). Net BRET (dissociation) plotted as γ9-GFP (acceptor) over Ga-Rluc (donor) ratio as a function of log doses of neurotensin. Neurotensin receptor-mediated BRET shows that lowering pHi (from 7.2 to 6.7) either by altering extracellular pH (B) or by the addition of FCCP (C), decreases Gai dissociation from Gβγ in HEK29T cells. (D) WT Gai shows significantly higher dissociation from Gβγ relative to both double variant and triple variants (low pH mimetics of WT Gai). (E) The difference in net BRET signal as a function of FCCP concentration (0 μM in solid line and 2 μM in dotted line) for double variant and triple variants is reduced with respect to WT Gai, confirming key roles for E236, D237, and E245 in pH-mediated Gβγ dissociation for Gai-Gβγ complex. All BRET data are averages of two independent experiments performed in duplicate (±SE).
+ +<--- Page Split ---> + +Further to confirm that the modulation of \(\mathrm{G}\beta \gamma\) release by pH occurs through the identified 3 residues of \(\mathrm{G}\alpha \mathrm{i}\) , we performed the BRET assay using the double or triple \(\mathrm{G}\alpha \mathrm{i}\) variants. As shown in Fig. 7D, agonist- stimulated \(\mathrm{G}\alpha \mathrm{i}\) dissociation from \(\mathrm{G}\beta \gamma\) is significantly reduced for the "double variant" and further reduced for the triple variant with respect to WT \(\mathrm{G}\alpha \mathrm{i}\) . These variants appear to serve as low pH mimetics, as they produce BRET similar to that obtained at lower \(\mathrm{pH_i}\) . This observation was further validated by BRET assays performed for double variant and triple variant at different \(\mathrm{pH_i}\) . Taken together, these findings suggest that enhanced \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) association at lower pH is due to the loss of electrostatic interactions within the Switch network needed for \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) formation. + +## Discussion: + +The recognition of pH sensors among biomolecules is crucial in understanding cellular processes, as they play a pivotal role in responding to changes in the proton concentration within the physiological range. While many biomolecules experience alterations in their protonation state, only a specific subset serves as pH sensors, exhibiting pH- dependent functional changes that influence cellular processes. + +One of the first pH- sensing proteins to be characterized is hemoglobin. In acidic environments, hemoglobin exhibits reduced affinity for oxygen, a phenomenon known as the Bohr effect. This effect is due to the protonation of specific amino acid residues, including histidine 146, located in the \(\alpha\) subunit of hemoglobin29. The protonation of H146 promotes the formation of a salt bridge with a nearby aspartate residue (D94) on the \(\beta\) subunit. This interaction stabilizes the deoxygenated or T- state conformation of hemoglobin, reducing its affinity for oxygen and facilitating the release of oxygen to tissues where it is needed30. After the identification of hemoglobin, proton- sensing ion channels and receptors governing cytosolic pH have been a focal point in research. + +More recent structural informatics calculations have shown that buried ionizable networks are a structural hallmark of pH sensitivity3, and the large pKa shifts exhibited by buried sidechains can be harnessed by proteins to drive pH- dependent changes in structure, stability, and function. Indeed, networks of buried ionizable amino acids are conserved in nearly all of the \(100+\) \(\mathrm{G}\alpha\) protein structures in the Protein Data Bank (PDB)3. On that basis, the cell signaling GTPases, \(\mathrm{G}\alpha \mathrm{i}\) and its yeast homolog Gpa1, were proposed to function as pH- sensing proteins. The study indicated that + +<--- Page Split ---> + +the stability of heterotrimeric \(\mathrm{G}\alpha\) proteins exhibits a significant dependence on pH levels across a broad range from 5 to 8.0. Furthermore, alterations in yeast intracellular \(\mathrm{pH} (\mathrm{pH}_i)\) were shown to promote Gpa1 phosphorylation and subsequently dampen the mitogen- activated protein (MAP) kinase signaling pathway4. In addition to our findings that Gαi serves as a pH sensor to regulate \(\mathrm{G}\alpha - \mathrm{G}\beta \gamma\) interactions, recent studies indicate that select GPCRs possess pH sensing properties that modulate extracellular signaling to \(\mathrm{G}\alpha\) proteins31,32,33. These and other GPCRs may work independently or synergistically with Gαi proteins to transduce extracellular pH- dependent signals to changes in intracellular \(\mathrm{pH}^{28,34,35}\) . + +Herein, we find that as pH is raised from 6.8 to 7.3, the thermostability of Gαi- GDP is greatly enhanced ( \(\Delta \mathrm{Tm} \sim 20^{\circ}\mathrm{C}\) ) due to the formation of an electrostatic network within the Ras- like domain. Since thermostability changes for GDP- bound Gαi were significantly larger ( \(\sim 20^{0}\) ), relative to the GTP- bound form ( \(\sim 8^{0}\) ), we focused on characterizing pH- dependent structural and dynamic changes associated with the GDP- bound form of Gαi in this study. Using pH- dependent NMR analyses and thermal stability profiling on WT and mutant proteins, we identified three key pH- sensing residues that drive pH- dependent thermostability changes. Two of the residues (E236 and D237) reside in SW- III of Gαi while the third residue (E245) lies in the neighboring α3 helix. Mutation of E236 and D237 in SW- III promotes Gβγ release in HEK293 cells, supporting the role of these residues in pH- dependent electrostatic interactions that allosterically regulate heterotrimer formation. While both Ras and heterotrimeric G- proteins contain SWI and SWII regions, Ras proteins lack the SW- III region found in Gα proteins. The presence of this unique SW- III region in heterotrimeric G- proteins may explain the unique pH sensing properties of Gαi and provide a novel role for the SW- III region in pH- dependent Gβγ release from the Gαi- Gβγ complex. Of note, the pH sensing network identified in this study is distinct from the network predicted by Isom et al3. This computationally predicted pH sensing network consists of residues at the Ras/helical domain interface (e.g. K46, D150, D200, D229, R242, K270, and K277). However, our NMR and CD data do not support this interaction network, as observable NMR resonances associated with K270 and K277 do not show pH- dependent perturbations and mutation of K46 and R242 did not alter pH- dependent thermostability changes measured by CD. + +Large changes in the stability of GTP- bound Gαi have previously been attributed to the formation of an interaction triad (termed as G- R- E Triad) involving residues G203 and R208 from + +<--- Page Split ---> + +SW- II and E245 from \(\alpha 3\) . This interaction triad is absent from the less stable and more dynamic GDP- bound state \(^{7,25}\) . Interestingly, E245 is also one of the residues identified in the pH sensing network. Moreover, E236 and D237 from SW- III interact with SW- II via residue R205 to stabilize both Switches in the GTP- bound forms of the protein. This nucleotide- dependent disorder- to- order transition is predicted to facilitate the release of \(\mathrm{G}\beta \gamma\) from Gai resulting in the activation of Gai. Consistent with these findings, SW- III residues E236 and D237 are involved in pH- dependent electrostatic interactions that appear to allosterically regulate \(\mathrm{G}\beta \gamma\) release. + +![](images/Figure_8.jpg) + +
Fig. 8: Proposed mechanism of Gai pH regulation. Activation of Gai by GPCRs leads to the release of Gai from \(\mathrm{G}\beta \gamma\) and stimulation of signaling pathways. A distinct mechanism of signaling regulation may occur by intracellular pH regulation. Gai protein may act as intracellular pH sensors to regulate Gai- \(\mathrm{G}\beta \gamma\) interactions with lower pH enhancing Gai- \(\mathrm{G}\beta \gamma\) interactions to inhibit Gai signaling.
+ +While several pH- sensing proteins contain titratable histidines \(^{36,37,38}\) , others such as EmrE contain aspartate and glutamate residues that titrate in the physiological range due to the formation of electrostatic networks \(^{32,39}\) . In these systems, even a small reduction in pH can alter the side chain protonation state leading to the neutralization of charge. In the case of Gai, we identified a network of three charged residues (E236, D237 and E245) that regulate pH- dependent thermostability changes in the Gai- GDP. To investigate how protonation/deprotonation of these residues might + +<--- Page Split ---> + +alter Gαi structure and dynamics in the GDP- bound state, we performed MD simulations. As starting points for the simulations, residues E236, D237 and E245 were generated in both charged and uncharged states, to represent higher (deprotonated) and lower pH (protonated) states, respectively. Results from these analyses suggest that the protonation of residues E236, D237 and E245 enhances Gαi- GDP dynamics due to the loss of electrostatic interactions between SW- II, SW- III, and α3. Moreover, we find that these residues make transient electrostatic interactions in Gαi- GDP at higher ( \(\sim\) pH 7.2), but not at lower pH. Since the binding of Gβγ to the Gαi depends on the dynamic properties of SW- II, we propose that lower physiological pH promotes Gβγ binding to Gαi resulting in attenuation of both Gαi and Gβγ mediated signaling pathways. Consistent with this premise, our findings indicate that higher intracellular pHi, promotes agonist- mediated Gβγ release from Gαi via the formation of pH- dependent electrostatic networks involving key residues within SW- III and α3. This in turn, will regulate Gαi and Gβγ downstream signaling. + +While the current investigation is centered on the Gαi isoform, sequence and structural alignment with other Gα isoforms (such as Gαs, Gαo, and Gαq) reveals conservation within the core pH sensing network (Supplementary Fig. S6)3. This analysis suggests the potential for pH- sensing properties in other Gα isoforms. The development of pH- insensitive forms of Gαi will facilitate investigations of hormone- and neurotransmitter signaling during physiological stresses, such as occur during glucose or oxygen deprivation, leading to changes in cellular pH40. + +## References: + +1. Madshus, I. H. Regulation of intracellular pH in eukaryotic cells. Biochem. J. 250, 1–8 (1988). +2. Damaghi, M., Wojtkowiak, J. W. & Gillies, R. J. pH sensing and regulation in cancer. Front. Physiol. 4 DEC, 1–10 (2013). +3. Isom, D. G. et al. Protons as second messenger regulators of G protein signaling. Mol. Cell 51, 531–538 (2013). +4. Hewitt, N. et al. Catalytic site mutations confer multiple states of G protein activation. Sci. Signal. 16, eabq7842 (2023). + +<--- Page Split ---> + +5. Murga, C. et al. G protein-coupled receptor kinase 2 (GRK2) as a potential therapeutic target in cardiovascular and metabolic diseases. Front. Pharmacol. 10, (2019). +6. Kimura, T., Pydi, S. P., Pham, J. & Tanaka, N. Metabolic Functions of G Protein-Coupled Receptors in Hepatocytes—Potential Applications for Diabetes and NAFLD. Biomolecules 10, (2020). +7. Goricanec, D. et al. Conformational dynamics of a G-protein \(\alpha\) subunit is tightly regulated by nucleotide binding. Proc. Natl. Acad. Sci. U. S. A. 113, E3629–E3638 (2016). +8. Olsen, R. H. J. et al. TRUPATH, an open-source biosensor platform for interrogating the GPCR transducermore. Nat. Chem. Biol. 16, 841–849 (2020). +9. Misconai, A. et al. BeSTSeI: webserver for secondary structure and fold prediction for protein CD spectroscopy. Nucleic Acids Res. 50, W90–W98 (2022). +10. Lee, W., Tonelli, M. & Markley, J. L. Structural bioinformatics NMRFAM-SPARKY: enhanced software for biomolecular NMR spectroscopy. 31, 1325–1327 (2015). +11. Goricanec, D. & Hagn, F. NMR backbone and methyl resonance assignments of an inhibitory G-alpha subunit in complex with GDP. Biomol. NMR Assign. 13, 131–137 (2019). +12. Prakash, A., Shin, J., Rajan, S. & Yoon, H. S. Structural basis of nucleic acid recognition by FK506-binding protein 25 (FKBP25), a nuclear immunophilin. 44, 2909–2925 (2016). +13. Wall, M. A. et al. The structure of the G protein heterotrimer Gα1β1γ2. Cell 83, 1047–1058 (1995). +14. Šali, A. & Blundell, T. L. Comparative Protein Modelling by Satisfaction of Spatial Restraints. J. Mol. Biol. 234, 779–815 (1993). +15. Huang, J. & Jr, A. D. M. CHARMM36 All-Atom Additive Protein Force Field : Validation Based on Comparison to NMR Data. 2135–2145 (2013) doi:10.1002/jcc.23354. +16. Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: A new molecular dynamics method. J. Appl. Phys. 52, 7182–7190 (1981). +17. Essmann, U. et al. A smooth particle mesh Ewald method. J. Chem. Phys. 103, 8577–8593 (1995). +18. Hess, B., Bekker, H., Berendsen, H. J. C. & Fraaije, J. G. E. M. LINCS: A Linear + +<--- Page Split ---> + +Constraint Solver for molecular simulations. J. Comput. Chem. 18, 1463- 1472 (1997). 19. van Zundert, G. C. P. et al. The HADDOCK2.2 Web Server: User- Friendly Integrative Modeling of Biomolecular Complexes. J. Mol. Biol. 428, 720- 725 (2016). 20. Behbahan, I. S., McBrian, M. A. & Kurdistani, S. K. A Protocol for Measurement of Intracellular pH. Bio- protocol 4, e1027 (2014). 21. Hamm, H. E., Meier, S. M., Liao, G. & Preininger, A. M. Trp fluorescence reveals an activation- dependent cation- π interaction in the switch II region of Gai proteins. Protein Sci. 18, 2326- 2335 (2009). 22. Shellhammer, J. P. et al. Amino acid metabolites that regulate G protein signaling during osmotic stress. PLOS Genet. 13, e1006829 (2017). 23. Ham, D. et al. Conformational switch that induces GDP release from Gi. J. Struct. Biol. 213, 107694 (2021). 24. Thieker, D. F. et al. Stabilizing proteins, simplified: A Rosetta- based webtool for predicting favorable mutations. Protein Sci. 31, e4428 (2022). 25. Knight, K. M. et al. A universal allosteric mechanism for G protein activation. Mol. Cell 81, 1384- 1396. e6 (2021). 26. Alegre, K. O. et al. Structural basis and mechanism of activation of two different families of G proteins by the same GPCR. Nat. Struct. Mol. Biol. 28, 936- 944 (2021). 27. Berezhnov, A. V. et al. Intracellular pH modulates autophagy and mitophagy. J. Biol. Chem. 291, 8701- 8708 (2016). 28. Kapolka, N. J. et al. Proton- gated coincidence detection is a common feature of GPCR signaling. Proc. Natl. Acad. Sci. U. S. A. 118, (2021). 29. Perutz, M. F., Gronenborn, A. M., Clore, G. M., Fogg, J. H. & Shih, D. T. - b. The pKa values of two histidine residues in human haemoglobin, the Bohr effect, and the dipole moments of α- helices. J. Mol. Biol. 183, 491- 498 (1985). 30. Dajnowicz, S. et al. Visualizing the Bohr effect in hemoglobin: neutron structure of equine cyanomethomoglobin in the R state and comparison with human deoxyhemoglobin in the T state. Acta Crystallogr. Sect. D, Struct. Biol. 72, 892- 903 (2016). 31. Levin, L. R. & Buck, J. Physiological roles of acid- base sensors. Annu. Rev. Physiol. 77, 347- 362 (2015). + +<--- Page Split ---> + +32. Li, J., Her, A. S. & Traaseth, N. J. Asymmetric protonation of glutamate residues drives a preferred transport pathway in EmrE. Proc. Natl. Acad. Sci. U. S. A. 118, (2021).33. Russell, J. L. et al. Regulated expression of pH sensing G protein-coupled receptor-68 identified through chemical biology defines a new drug target for ischemic heart disease. ACS Chem. Biol. 7, 1077–1083 (2012).34. Rowe, J. B., Kapolka, N. J., Taghon, G. J., Morgan, W. M. & Isom, D. G. The evolution and mechanism of GPCR proton sensing. J. Biol. Chem. 296, 100167 (2021).35. Isom, D. G. & Dohlman, H. G. Buried ionizable networks are an ancient hallmark of G protein-coupled receptor activation. Proc. Natl. Acad. Sci. U. S. A. 112, 5702–5707 (2015).36. Vercoulen, Y. et al. A histidine pH sensor regulates activation of the Ras-specific guanine nucleotide exchange factor RasGRP1. Elife 6, 1–26 (2017).37. Tomura, H., Mogi, C., Sato, K. & Okajima, F. Proton-sensing and lysolipid-sensitive G-protein-coupled receptors: A novel type of multi-functional receptors. Cell. Signal. 17, 1466–1476 (2005).38. Williamson, D. M., Elferich, J. & Shinde, U. Mechanism of fine-tuning pH sensors in proprotein convertases: Identification of a pH-sensing histidine pair in the propeptide of proprotein convertase 1/3. J. Biol. Chem. 290, 23214–23225 (2015).39. Schönichen, A., Webb, B. A., Jacobson, M. P. & Barber, D. L. Considering protonation as a posttranslational modification regulating protein structure and function. Annu. Rev. Biophys. 42, 289–314 (2013).40. Carr, R. et al. B-Arrestin-Biased Signaling Through the B2-Adrenergic Receptor Promotes Cardiomyocyte Contraction. Proc. Natl. Acad. Sci. U. S. A. 113, E4107–E4116 (2016). + +## Acknowledgments + +The authors thank Bryan Roth for providing BRET constructs and David Forest for the critical input in Rosetta predictions. Illustrative images in Figs. 3B, 7A and 8 were created with BioRender.com. Research reported in this publication was supported by National Institutes of Health (NIH) grants R35GM134962 (to S. L. Campbell) and R35GM118105 (to H. G Dohlman). + +## Author contributions + +<--- Page Split ---> + +729 Conceptualization: A.p., G.Y., H.G.D., S.L.C. Methodology: A.P., Z.L., J.R., N.H.V., G.B.H., G.Y., Experimentation: A.P., Z.L., J.R., N.H.V., G.B.H., G. Y., Simulation: V.R.C. Writing—original draft: A.P., Z.L., V.R.C., H.G.D., S.L.C. Writing—review & editing: G.Y., J.R., N.H.V. Analysis: A.P., Z.L., J.R., N.H.V., G.B.H. Supervision: H.G.D., S.L.C. + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementary figures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc_det.mmd b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b6eb1bcab572e46f922d29b7c0e6ede85d3ca1d3 --- /dev/null +++ b/preprint/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc/preprint__0e6062296811ed0247c4338e0381fd53a7dc3cfb8287474a87040085029684fc_det.mmd @@ -0,0 +1,420 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 860, 175]]<|/det|> +# Molecular and Functional Profiling of Gai as an Intracellular pH Sensor + +<|ref|>text<|/ref|><|det|>[[44, 195, 280, 240]]<|/det|> +Sharon Campbell campbes1@med.unc.edu + +<|ref|>text<|/ref|><|det|>[[50, 268, 778, 289]]<|/det|> +University of North Carolina at Chapel Hill https://orcid.org/0000- 0003- 0311- 409X + +<|ref|>text<|/ref|><|det|>[[44, 293, 198, 333]]<|/det|> +Ajit Prakash UNC Chapel Hill + +<|ref|>text<|/ref|><|det|>[[44, 340, 658, 381]]<|/det|> +Zijian Li University of North Carolina https://orcid.org/0000- 0002- 6802- 9606 + +<|ref|>text<|/ref|><|det|>[[44, 385, 658, 427]]<|/det|> +Venkat Chirasani University of North Carolina https://orcid.org/0000- 0003- 1416- 0438 + +<|ref|>text<|/ref|><|det|>[[44, 431, 298, 472]]<|/det|> +Juhi Rasquinha University of North Carolina + +<|ref|>text<|/ref|><|det|>[[44, 478, 298, 519]]<|/det|> +Natalie Valentin University of North Carolina + +<|ref|>text<|/ref|><|det|>[[44, 525, 298, 565]]<|/det|> +Garrett Hubbard University of North Carolina + +<|ref|>text<|/ref|><|det|>[[44, 571, 548, 612]]<|/det|> +Guowei Yin The Seventh Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 617, 420, 658]]<|/det|> +Henrik Dohlman University of North Carolina at Chapel Hill + +<|ref|>sub_title<|/ref|><|det|>[[44, 700, 103, 717]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 737, 137, 755]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 775, 300, 794]]<|/det|> +Posted Date: April 30th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 813, 475, 833]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4203924/v1 + +<|ref|>text<|/ref|><|det|>[[42, 850, 914, 893]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 910, 535, 931]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 914, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on April 11th, 2025. See the published version at https://doi.org/10.1038/s41467-025-58323-2. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[70, 88, 828, 112]]<|/det|> +## Molecular and Functional Profiling of Gαi as an Intracellular pH Sensor + +<|ref|>text<|/ref|><|det|>[[70, 118, 885, 240]]<|/det|> +2 Ajit Prakash \(^{1}\) , Zijian Li \(^{1}\) , Venkata R. Chirasani \(^{1}\) , Juhi A. Rasquinha \(^{1}\) , Natalie H. Valentin \(^{2}\) , Garrett B. Hubbard \(^{1}\) , Guowei Yin \(^{3}\) , Henrik G. Dohlman \(^{2*}\) , Sharon L. Campbell \(^{1,4*}\) \(^{1}\) Department of Biochemistry & Biophysics, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA \(^{2}\) Department of Pharmacology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA \(^{3}\) The Seventh Affiliated Hospital of Sun Yat- sen University, Shenzhen, 518107, China \(^{4}\) Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA + +<|ref|>text<|/ref|><|det|>[[70, 274, 787, 295]]<|/det|> +8 \*Address for correspondence: hdohlman@med.unc.edu and campbesl@med.unc.edu + +<|ref|>sub_title<|/ref|><|det|>[[115, 425, 196, 443]]<|/det|> +## Abstract: + +<|ref|>text<|/ref|><|det|>[[70, 465, 886, 802]]<|/det|> +Heterotrimeric G proteins ( \(\mathrm{G}\alpha\) , \(\mathrm{G}\beta\) and \(\mathrm{G}\gamma\) ) act downstream of G- protein- coupled receptors (GPCRs) to mediate signaling pathways that regulate various physiological processes and human disease conditions. Previously, human \(\mathrm{G}\alpha \mathrm{i}\) and its yeast homolog Gpa1 have been reported to function as intracellular pH sensors, yet the pH sensing capabilities of \(\mathrm{G}\alpha \mathrm{i}\) and the underlying mechanism remain to be established. Herein, we identify a pH sensing network within \(\mathrm{G}\alpha \mathrm{i}\) , and evaluate the consequences of pH modulation on the structure and stability of the G- protein. We find that changes over the physiological pH range significantly alter the structure and stability of \(\mathrm{G}\alpha \mathrm{i}\) - GDP, with the protein undergoing a disorder- to- order transition as the pH is raised from 6.8 to 7.5. Further, we find that modulation of intracellular pH in HEK293 cells regulates \(\mathrm{G}\alpha \mathrm{i}\) - \(\mathrm{G}\beta \gamma\) release. Identification of key residues in the pH- sensing network allowed the generation of low pH mimetics that attenuate \(\mathrm{G}\alpha \mathrm{i}\) - \(\mathrm{G}\beta \gamma\) release. Our findings, taken together, indicate that pH- dependent structural changes in \(\mathrm{G}\alpha \mathrm{i}\) alter the agonist- mediated \(\mathrm{G}\beta \gamma\) dissociation necessary for proper signaling. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 126, 886, 784]]<|/det|> +Within the complex milieu of living cells, intracellular pH (pHi) is maintained within a narrow range, as even small changes in pH can affect a myriad of cellular processes including membrane potential, ion transport, cellular growth and metabolism1,2,3. Unsurprisingly, disruptions in pH regulation can contribute to the development of pathological conditions such as ischemic heart disease, cancer, and neurological disorders3,4. While various proton pumps and transporters play a role in regulating the flow of protons across the membrane to uphold pHi homeostasis, there are intracellular proteins (termed pH sensors) that sense and transmit pH signals, thus orchestrating the regulation of biochemical processes. Among these pH sensors are signal- transducing proteins. Notably, a study by Isom et. al. (2013) provided evidence that a subset of signal- transducing heterotrimeric G proteins may serve as intracellular pH sensors3. These membrane- associated proteins form a heterotrimeric complex (Gα, Gβ and Gγ) and act downstream of G- protein coupled receptors (GPCRs). GPCRs represent the most extensive group of membrane proteins that are targeted by approved drugs. They play a crucial role in orchestrating the majority of cellular responses to hormones and neurotransmitters through several signaling pathways mediated by different isoforms of G protein (Gαi, Gαs, Gα12/13, and Gαq) that receive and transduce signals through diverse pathways5,6. The Gβ and Gγ subunits associate to form a Gβγ heterodimer3 whereas the Gα subunit binds GDP or GTP and catalyzes GTP hydrolysis. The Gα subunit is comprised of two distinct domains: a helical domain and a Ras- like domain. Within the Ras- like domain, there are three key 'Switch' regions, namely SW- I, SW- II, and SW- III, which play an essential role in its activity (Fig. 2D). In the GDP- bound state, which is an inactive form of Gα protein, these Switch regions exhibit dynamic behavior. In contrast, in the GTP- bound state (the active form of Gα), they become more structured and less dynamic7. In the GDP- bound state, the Gα subunit is associated with the Gβγ complex. Upon GPCR stimulated Gα GTP- loading, the Gβγ subunits dissociate from Gα- GTP and along with Gα, promote activation of downstream signaling pathways. + +<|ref|>text<|/ref|><|det|>[[113, 804, 884, 903]]<|/det|> +Isom et. al. (2013) developed a computer algorithm, pHinder, which predicted that both the mammalian Gαi isoform and yeast homolog Gpa1 contain a core of residues between the Ras- like and helical domains that may promote pH sensing properties8. In support of this prediction, both proteins showed pH- dependent changes in thermostability over a pH range from 5.5 to 8. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 188]]<|/det|> +They also found that Gpa1 undergoes phosphorylation under acidic conditions to attenuate pheromone- dependent stimulation of mitogen- activated protein kinases in the yeast4. While these findings, taken together suggest that mammalian Gai may function as a pH sensor, the molecular mechanism, and the biological consequence of pH sensing through Gai remain unknown. + +<|ref|>text<|/ref|><|det|>[[111, 210, 885, 655]]<|/det|> +To expand on past observations, we characterized pH- dependent biochemical and structural/dynamic properties of GDP- bound Gai, identified the underlying pH- dependent electrostatic network, and determined functional consequences on cellular Gai activity. We find that the structure, stability, and dynamics of Gai in the GDP- bound state are highly dependent on pH over the physiological pH range due to a pH- responsive network within the Ras- like domain. These findings differ from a previous report where larger pH- dependent changes in thermostability were observed for the GTP- bound state with ionizable residues predicted to lie at the interface between the Ras- like and helical domains3. Our NMR, biophysical and computational analyses indicate that changes in the ionization state of residues within the pH sensing network promote a disorder- to- order transition in Gai over the physiological pH range, to populate a less ordered state than enhances Gai- Gβγ association in HEK293 cells at the lower end of the physiological pH range. Identification of the pH- sensing network allowed for the identification and generation of low pH mimetics that reduce pH- dependent Gβγ release from the Gai- Gβγ complex. Of note, the Gai Switch III region plays a key role in pH sensing, providing a new role of this understudied Switch region in agonist- mediated Gβγ release, a key important step for both Gai and Gβγ activation. Taken together, our studies indicate that Gai undergoes a pH- dependent order- to- disorder transition that modulates Gai- Gβγ interactions and Gai activation over the physiological pH range. + +<|ref|>sub_title<|/ref|><|det|>[[115, 677, 311, 695]]<|/det|> +## Materials and methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 714, 277, 731]]<|/det|> +## Plasmid constructs + +<|ref|>text<|/ref|><|det|>[[112, 748, 884, 874]]<|/det|> +A bacterial pET- SUMO vector containing the human Gai1 gene (GenBank accession no. BC026326) with the first 31 amino acids deleted, N- terminal 6xHis, and SUMO- tag were employed for in vitro analyses. Full- length human Ga containing Renilla luciferase (Ga- RLuc8), Gβ and Gγ containing green fluorescence protein 2 (Gγ- GFP2), and neurotensin receptor (NTR) constructs were employed for cell- based BRET assays4,8. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 333, 108]]<|/det|> +## Site-directed mutagenesis + +<|ref|>text<|/ref|><|det|>[[111, 124, 886, 303]]<|/det|> +Gai variants were generated using the Q5 Site- Directed Mutagenesis kit (NEB). Polymerase chain reaction (PCR) primers were designed using the NEBaseChanger (https://nebasechanger.neb.com), an online mutagenesis primer design tool powered by New England Biolabs from Eton Bioscience Inc. Mutagenesis was performed as described4. The variant constructs were sequenced by Sanger Sequencing (Genewiz) to confirm successful mutagenesis. Sequencing results for Gai variants were aligned with published sequences of wildtype (WT) Gai using Clustal Omega (European Molecular Biology Laboratory, Cambridgeshire, UK). + +<|ref|>sub_title<|/ref|><|det|>[[113, 318, 581, 338]]<|/det|> +## Expression and purification of WT Gai and its variants + +<|ref|>text<|/ref|><|det|>[[110, 351, 886, 847]]<|/det|> +Gai proteins were overexpressed in E. coli Agilent BL21- codon plus (DE3)- RP- X competent cells. Cells were grown at \(37^{\circ}\mathrm{C}\) for 3- 4 hours to achieve an optical density at \(600\mathrm{nm}\) (OD 600) of 0.60 and then induced with \(500\mu \mathrm{M}\) isopropylthio- \(\beta\) - galactoside (IPTG). The temperature was then reduced to \(18^{\circ}\mathrm{C}\) and the cultures were left to grow overnight. After 20 hours, bacteria cells were harvested by centrifugation at 13,000 rpm and then resuspended in \(50\mathrm{mL}\) Dialysis buffer ( \(10\mathrm{mM}\) \(\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(1\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\beta\) - mercaptoethanol (BME), \(\mathrm{pH}7.0\) ). Resuspended cells were lysed using a sonicator (Fisher Scientific, #CL- 334) with an amplitude of 80 and 5 seconds on/off time ratio for 10 mins/L cells. Clarified lysate was passed through the Ni- NTA column equilibrated with 1 column volume (CV) (50 ml) of Dialysis buffer, washed with Wash buffer ( \(10\mathrm{mM}\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(1\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\mathrm{BME}\) , \(30\mathrm{mM}\) imidazole, \(\mathrm{pH}7.0\) ) and eluted using the Elution buffer ( \(10\mathrm{mM}\mathrm{KH_2PO_4}\) , \(20\mathrm{mM}\mathrm{K_2HPO_4}\) , \(0.15\mathrm{M}\mathrm{KCl}\) , \(5\mathrm{mM}\mathrm{MgCl_2}\) , \(10\mu \mathrm{M}\) GDP, \(1\mathrm{mM}\mathrm{BME}\) , \(250\mathrm{mM}\) imidazole, \(\mathrm{pH}7.0\) ). The elution fraction was dialyzed overnight in the presence of a ubiquitin- like- specific protease 1 (ULP1) to cleave the SUMO tag on Gai. After dialysis, the protein was again passed over a Ni- NTA column equilibrated with 3 CVs of water and 1 CV dialysis buffer. The protein flow- through was further purified by gel filtration chromatography using a Fast Protein Liquid Chromatography (FPLC) system (Akta Primeplus) and Superdex 75 column (Cytiva). The purity of the protein was verified by SDS PAGE gel electrophoresis, and the concentration of the purified protein was estimated by Thermo Scientific NanoDrop. + +<|ref|>sub_title<|/ref|><|det|>[[115, 862, 403, 881]]<|/det|> +## Circular Dichroism (CD) analyses + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 345]]<|/det|> +Gai proteins were exchanged in CD buffer (10 mM \(\mathrm{K}_2\mathrm{HPO}_4\) , 500 \(\mu \mathrm{M}\) MgSO4, 500 \(\mu \mathrm{M}\) tris(2- carboxymethyl)phosphine (TCEP)) using an Amicon centrifugal filter unit (MilliporeSigma, #UFC901024) at \(3800\mathrm{xg}\) , diluted to \(15\mu \mathrm{M}\) (pH 6.0, 6.4, 6.8, 7.2, or 7.6) and centrifuged (14,000 rpm) for 10 mins at \(4^{\circ}\mathrm{C}\) . Temperature dependent CD experiments were performed on a Jasco J815CD spectrometer using \(15\mu \mathrm{M}\) WT or Gai variants protein in a \(1\mathrm{mm}\) path- length quartz cuvette (Hellma Analytics). Thermal melts were obtained at \(222\mathrm{nm}\) , over a temperature range of \(20–95^{\circ}\mathrm{C}\) , using a temperature increment of \(1\mathrm{C / min}\) . The CD spectral scans were collected for Gai proteins at different fixed temperatures (e.g. \(45^{\circ}\mathrm{C}\) , \(55^{\circ}\mathrm{C}\) and \(65^{\circ}\mathrm{C}\) ) by taking CD measurements every \(1\mathrm{nm}\) from \(200–250\mathrm{nm}\) . Secondary structure was evaluated using the online server BeStSel (Beta Structure Selection) \(^9\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 362, 399, 381]]<|/det|> +## Intrinsic tryptophan fluorescence + +<|ref|>text<|/ref|><|det|>[[113, 399, 884, 529]]<|/det|> +Intrinsic tryptophan fluorescence assays were conducted using \(2\mu \mathrm{M}\) purified WT or variant GDP- loaded Gai proteins in \(200\mu \mathrm{l}\) of assay buffer (20 mM HEPES, \(50\mathrm{mMNaCl}\) , \(5\mathrm{mM}\mathrm{MgCl}_2\) , \(2\mathrm{mM}\) DTT) at different pH (pH 5, 6 and 7.2) in the 96 well plate. Gai proteins were excited at \(290\mathrm{nm}\) and intrinsic tryptophan fluorescence was measured from 300 to \(400\mathrm{nm}\) wavelength using a SpectraMax M4 Series Microplate Reader. + +<|ref|>sub_title<|/ref|><|det|>[[115, 545, 301, 564]]<|/det|> +## 2D NMR experiments + +<|ref|>text<|/ref|><|det|>[[112, 581, 884, 760]]<|/det|> +Triple labeled \(^2\mathrm{H} / ^{13}\mathrm{C} / ^{15}\mathrm{N}\) Gai samples were generated by expressing Gai proteins in BL21(DE3)- RIPL E. coli cells and growing in minimal medium supplemented with \(1\mathrm{g / L}^{15}\mathrm{NH}_4\mathrm{Cl}\) , \(3\mathrm{g / L}^{13}\mathrm{C}\) - glucose, \(99\%\) D \(_2\mathrm{O}\) (Cambridge Isotope Labs) \(^7\) , and purified as described above. Triple labeled Gai was prepared in \(20\mathrm{mM}\) potassium phosphate ( \(\mathrm{pH} = 7.0\) ), \(50\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) DTT (dithiothreitol), \(5\mathrm{mM}\) magnesium chloride, and \(50\mu \mathrm{M}\) GDP containing \(5\%\) (v/v) D \(_2\mathrm{O}\) . For \(^{15}\mathrm{N}\) - enriched WT and Gai variants, bacterial expressed proteins were grown in minimal media containing \(1\mathrm{g / L}^{15}\mathrm{NH}_4\mathrm{Cl}\) as the sole nitrogen source and purified as described above. + +<|ref|>text<|/ref|><|det|>[[112, 775, 884, 901]]<|/det|> +2D NMR \(^1\mathrm{H}–^{15}\mathrm{N}\) HSQC spectra of Gai were acquired on a Bruker Avance 850 MHz (14.1 T field strength) NMR spectrometer at \(25^{\circ}\mathrm{C}\) , with a cryogenic (TCI) \(5\mathrm{mm}\) triple resonance probe equipped with a z- axis gradient. The \(^1\mathrm{H}–^{15}\mathrm{N}\) HSQC spectra of Gai- GDP were assigned using a combination of triple resonance experiments, including 3D HNCA, HN(CO)CA, HN(CA)CO and HNCO \(^7\) . TROSY- based pulse sequences were used for sensitivity enhancement. Bruker- TopSpin + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 240]]<|/det|> +was used to process the NMR data and NMRFAM- SPARKY was used to visualize and analyze the NMR spectra10. For assignments, BMRB 30078 was used as reference spectrum for Gaig- GDP11. pH titration studies were performed by calculating the chemical shift perturbation (CSP) of Gaig- GDP over a pH range from 6.4, 6.8, 7.0, 7.2, 7.4 and 7.6. Average 1H- 15N CSP were determined using the formula \(\Delta \delta = [(\Delta^{1}\mathrm{HN})^{2} + (\Delta^{15}\mathrm{N} / 5)^{2}]^{0.5}\) , as previously described12. PyMOL (https://pymol.org/2/) was used to generate all images of molecular structures. + +<|ref|>sub_title<|/ref|><|det|>[[115, 256, 433, 275]]<|/det|> +## Guanine nucleotide dissociation assay + +<|ref|>text<|/ref|><|det|>[[111, 291, 884, 522]]<|/det|> +Gaig proteins were exchanged into nucleotide association buffer (20 mM HEPES, 50 mM NaCl, 5 mM MgCl2, 2 mM DTT) using an amino concentrator (3,800 g, 15 min, 3 rounds). In the cuvette, 0.75 μM (2'-(or- 3')- O-(N- Methylanthraniloyl) Guanosine 5'- Diphosphate, Disodium Salt (Mant- GDP, purchased from Thermofisher) was added to 1 ml of association buffer for the assay. The intrinsic tryptophan (W211) was excited at 280 nm and the Mant- GDP fluorescence intensity at 425 nm was measured as a function of time using a PerkinElmer LS55 luminescence spectrometer. After collecting data for a few seconds to obtain baseline fluorescence, purified 1 μM Gaig- GDP was added to the solution to initiate GDP association. Once the fluorescence reached saturation, 10 x (7.5 μM) GDP was added to the solution to initiate the GDP dissociation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 537, 444, 557]]<|/det|> +## Molecular Dynamics (MD) simulations + +<|ref|>text<|/ref|><|det|>[[111, 572, 884, 909]]<|/det|> +Gaig- GDP coordinates were extracted from the Gaig- Gβγ complex structure (PDB: 1GP2) to visualize Switch residues not observable in Gaig- GDP structures13. Modeller 9v21.2 was employed to relocate missing residues and generate charged and uncharged states for residues E236, D237 and E24514 to represent the charged state as higher pH (above 7.2) and the uncharged state as low pH (below 6.4). Following side- chain optimizations, minimum energy conformations of charged and uncharged GDP- bound Gaig states were identified. Subsequently, MD simulations were performed to investigate conformational and dynamic differences compared to WT GDP- bound Gaig. The CHARMM36 forcefield was used to parametrize the protein and GDP, and simulations were run using GROMACS- 2020.3.515. Gaig proteins were solvated in a water cubic box containing approximately 22,500 TIP3P water molecules and maintained a salt concentration of 150 mM through the addition of an appropriate number of Na+ and Cl- ions. The solvated system was energy minimized using the conjugate gradient algorithm and subsequently equilibrated using the V- rescale thermostat and the Parrinello- Rahman barostat16. Long- range electrostatic interactions + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 214]]<|/det|> +were evaluated using the Particle- Mesh Ewald sum17, and all bonds involving hydrogen atoms were constrained using the LINCS algorithm18. Simulations were run for 250 ns and were repeated in triplicate to ensure reproducibility and maintain consistency. Structural figures were generated using PyMOL and graphical plots were created using Grace (http://plasma- gate.weizmann.ac.il/Grace/). + +<|ref|>sub_title<|/ref|><|det|>[[115, 230, 293, 249]]<|/det|> +## HADDOCK docking + +<|ref|>text<|/ref|><|det|>[[112, 265, 884, 444]]<|/det|> +The HADDOCK online server was used for the computational docking of \(\mathrm{Ga}\) with \(\mathrm{G}\beta \gamma^{19}\) . The initial structure of \(\mathrm{Ga}\) with E236, D237, and E245 residue side chains in either the charged (higher \(\mathrm{pH}\) ) or uncharged state (lower \(\mathrm{pH}\) ) as described above, was obtained from the averaged structure of 250 ns MD simulations. The \(\mathrm{G}\beta \gamma\) structure was extracted from the \(\mathrm{Ga}\) - \(\mathrm{G}\beta \gamma\) complex crystal structure (PDB:1GP2). Contact residues between \(\mathrm{Ga}\) and \(\mathrm{G}\beta \gamma\) that form the trimeric complex structure were used as active residues and the passive residues were auto selected by the software. The docking model corresponding to the lowest energy was used for further analysis. + +<|ref|>sub_title<|/ref|><|det|>[[115, 460, 217, 479]]<|/det|> +## Cell culture + +<|ref|>text<|/ref|><|det|>[[112, 494, 884, 672]]<|/det|> +HEK293T cells were maintained, passaged, and transfected in Dulbecco's Modified Eagle Medium (DMEM, Gibco) containing \(10\%\) sterilized Fetal Bovine Serum (FBS, Gibco), 100 Units/mL penicillin, and \(100 \mu \mathrm{g} / \mathrm{mL}\) streptomycin (Gibco- ThermoFisher, Waltham, MA) in a humidified atmosphere at \(37^{\circ}\mathrm{C}\) and \(5\% \mathrm{CO}_{2}\) . Transfection was carried out using a lipid- polymer- based transfection agent (Mirus Bio, MIR 5400, Madison, WI). After transfection, cells were plated in DMEM (Dulbecco's Modified Eagle Medium) containing \(10\%\) FBS (Fetal Bovine Serum), 100 Units/mL penicillin, and \(100 \mu \mathrm{g} / \mathrm{mL}\) streptomycin for BRET assays. + +<|ref|>sub_title<|/ref|><|det|>[[115, 688, 362, 707]]<|/det|> +## Intracellular pH modulation: + +<|ref|>text<|/ref|><|det|>[[112, 723, 884, 875]]<|/det|> +The intracellular pH of HEK293 cells was altered by two different methods. In the first method, HEK293 cells were treated with \(1 - 5 \mu \mathrm{M}\) of trifluoromethoxycarbonylcyanide phenylhydrazone (FCCP) for 5 min. In the second method, intracellular pH was modulated by incubating HEK293 cells with Hanks' Balanced Salt Solution (HBSS) buffer at different pH for 15 min. Intracellular pH was quantified using the intracellular pH indicator, \(2',7'\) - bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein-acetoxymethyl ester (BCECF-AM) as described20. To convert BCECF-AM + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 883, 135]]<|/det|> +fluorescence to the intracellular pH values, HEK293 cells were treated with \(20 \mu \mathrm{M}\) nigericin and a calibration curve for \(\mathrm{pH}_i\) and BCECF- AM fluorescence was generated. + +<|ref|>sub_title<|/ref|><|det|>[[113, 151, 625, 171]]<|/det|> +## Bioluminescence Resonance Energy Transfer (BRET) assays + +<|ref|>text<|/ref|><|det|>[[112, 186, 884, 469]]<|/det|> +BRET assays were carried out as described by Olsen et al. (2020) \(^{8}\) . Briefly, HEK293T cells were seeded in six- well dishes in DMEM containing \(10\%\) FBS, 100 Units/mL penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin at a density of 300,000 cells/well and allowed to grow to \(80 - 90\%\) confluency. Cells were then transfected using a 1:1:1:1 DNA ratio of NTR: \(\mathrm{Ga}\) - RLuc: \(\mathrm{G}\beta\) : \(\mathrm{G}\gamma\) - GFP2 (1000 ng/construct for six- well dishes). TransIT- 2020 (Mirus Biosciences, Madison, WI) was used to complex the DNA at a ratio of \(3 \mu \mathrm{L}\) Transit/ \(\mu \mathrm{g}\) DNA, in OptiMEM (Gibco- ThermoFisher, Waltham, MA) at a concentration of \(1 \mu \mathrm{g}\) DNA/ \(\mu \mathrm{L}\) OptiMEM. The next day, cells were harvested from the plate using trypsin (0.25% EDTA, Gibco, #25200056), and plated in poly- D- lysine- coated white, clear bottom 96- well assay plates (Greiner Bio- One, Monroe, NC) in DMEM containing \(1\%\) dialyzed FBS, 100 Units/mL penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin at a density of 50,000 cells/well. + +<|ref|>text<|/ref|><|det|>[[112, 484, 884, 765]]<|/det|> +One day after seeding in 96- well assay plates, white backings (Perkin Elmer, Waltham, MA) were applied to the plate bottoms, and the growth medium was aspirated. Cells were washed three times with Assay buffer (1X HBSS, \(20 \mathrm{mM}\) HEPES, \(\mathrm{pH} 7.4\) ), then \(60 \mu \mathrm{L}\) of Assay buffer was added immediately, followed by a \(10 \mu \mathrm{L}\) addition of freshly prepared \(50 \mu \mathrm{M}\) coelenterazine 400a (1- bisdeoxycoelenterazine) (Nanolight Technologies, Pinetop, AZ) in each well. After a five- minute equilibration period, cells were treated with \(30 \mu \mathrm{L}\) of \(3 \mathrm{x}\) neurotransmitters ( \(3 \times 10^{- 5} \mathrm{M}\) ) for another 5 minutes. Plates were then read in a plate reader (Clariostar, BMG Labtech, Ortenberg, Germany) at \(395 \mathrm{nm}\) (Rluc8- coelenterazine 400a) and \(510 \mathrm{nm}\) (GFP2) with a slit width of \(8 \mathrm{nm}\) using 10 flashes per spiral well scan. Plates were read serially five times, and measurements from the third read were used in all analyses. BRET ratios were computed as the ratio of the GFP2 emission to Rluc8 emission. + +<|ref|>sub_title<|/ref|><|det|>[[113, 782, 276, 800]]<|/det|> +## Statistical Analysis + +<|ref|>text<|/ref|><|det|>[[112, 819, 883, 890]]<|/det|> +CD melt- curves were fit to Boltzmann sigmoidal equation in Prism 9.3.1 (GraphPad Software, San Diego, CA). Concentration- response curves for BRET assays were fit to a three- parameter logistic equation. Raw BRET concentration- response curves were normalized to the best- fit maximum + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 883, 135]]<|/det|> +within a data set. BRET data were represented as mean \(\pm\) SEM. Data analysis was carried out using Prism 9.3.1 (GraphPad Software, San Diego, CA). + +<|ref|>sub_title<|/ref|><|det|>[[115, 153, 180, 170]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 188, 695, 208]]<|/det|> +## pH modulation of GDP-bound Gai structure and dynamic properties + +<|ref|>text<|/ref|><|det|>[[112, 222, 884, 664]]<|/det|> +Earlier investigations into mammalian Gai demonstrated changes in structure and stability in response to pH variations8. While these findings suggest a potential role for Gai as a pH sensor8, the molecular basis and functional relevance has yet to be established. Herein, we apply comprehensive and multidisciplinary approaches to evaluate how pH changes in the physiological range affect structure, stability and Gai1 activity in vitro and in cells. We first conducted CD experiments on GDP- bound Gai to monitor pH- dependent changes in thermal unfolding, stability and secondary structure. Further, to monitor thermal unfolding and stability, we collected CD spectra at \(222 \mathrm{nm}\) as a function of pH and temperature. As shown in Fig. 1A, a striking and gradual increase in thermal stability ( \(\mathrm{T_m} \sim 25^{\circ} \mathrm{C}\) ) of Gai- GDP was observed over the pH range from 5 to 7.3. Notably, the thermal unfolding transition for Gai- GDP appears cooperative at low pH but shifts to a multi- state unfolding transition over the physiological pH range (Fig. 1A). Gai contains two subdomains, a Ras- like and helical domain. To evaluate differential unfolding at higher pH (above pH 7.1), we conducted CD scans (200- 250 nm) for Gai- GDP as a function of temperature at pH 7.2. As shown in Fig. 1B and C, CD spectra revealed a significant reduction in alpha- helical propensity, but not the beta- sheet propensity as the temperature is raised from \(45 - 55^{\circ} \mathrm{C}\) . These findings indicate that the helical domain melts first, followed by the Ras- like domain, with the Ras- like domain showing the greatest change in thermal stability at higher pH. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[164, 88, 827, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 469, 884, 616]]<|/det|> +
Fig. 1. Ga1-GDP thermostability and intrinsic tryptophan fluorescence are pH dependent. (A) Representative CD thermal melt (222 nm, 20–95 °C) of Ga1-GDP (15 μM) shows enhanced thermostability at higher pH (N = 3). (B) Representative CD spectral scans (200-250 nm) of Ga1-GDP collected at 45 and 55°C and pH 7.2 (N = 2). (C) Bar graph derived from temperature-dependent CD spectral scans highlighting the loss of alpha-helical but not beta-sheet secondary structure. (D) Representative intrinsic tryptophan fluorescence spectra of Ga1-GDP (2 μM, excitation = 280 nm, emission = 300 - 400 nm) shows enhanced fluorescence as the pH is increased from 5 to 7.2 (N = 2, performed in triplicate).
+ +<|ref|>text<|/ref|><|det|>[[112, 631, 884, 888]]<|/det|> +Given the enhanced stability observed at higher pH, we employed intrinsic tryptophan fluorescence experiments using tryptophan (W211) in the SW-II region (Fig. 3A) as a fluorescence probe to monitor differences in solvent exposure as a function of pH which indirectly indicates pH- dependent switch conformational changes. It has previously been shown that the more dynamic and less ordered GDP- bound state of Ga1 promotes enhanced exposure of W211 resulting in a fluorescence decrease relative to that of the Ga1- GTP state7,21. As shown in Fig. 1D, intrinsic Ga1- GDP tryptophan fluorescence increases over the pH range from 5 to 7.2, suggesting that higher pH promotes decreased solvent accessibility possibly due to enhanced interactions and structural order, consistent with the greatly enhanced stability observed by CD. Taken together, these results support a disorder- to- order transition at higher physiological pH for Ga1 in its GDP- bound form. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 885, 450]]<|/det|> +To further probe whether a pH- dependent disorder- to- order transition occurs in GDP- bound Gai, we conducted 2D NMR analyses. For these studies, we prepared \(^{15}\mathrm{N}\) enriched Gai- GDP and collected a 2D \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) Heteronuclear Single Quantum Coherence (HSQC) NMR spectra over a physiologically relevant pH range (6.4 - 7.6). Enrichment with \(^{15}\mathrm{N}\) allows the detection of backbone and sidechain NH resonances within Gai and provides a site- specific probe for every amino acid except proline. The 2D HSQC overlay of Gai- GDP at pH 6.4, 6.8 and 7.2 is shown in Fig. 2A, with an HSQC overlay comparing pH 6.4 versus pH 7.6 shown separately in Supplementary Fig. S1. Spectra acquired at lower pH show significant chemical shift changes, line broadening and loss of several peaks in comparison to those obtained at higher pH values, suggesting the protein is more dynamic at lower pH (Fig. 2A). This data correlates well with CD and tryptophan fluorescence data which suggests a less thermostable structure of Gai- GDP at lower pH. The pH- dependent HSQC changes observed are consistent with an earlier report which showed extensive broadening of NH peaks in the \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) HSQC spectrum of Gai- GDP at pH 6 relative to pH \(7^{22}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 95, 876, 610]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 626, 883, 818]]<|/det|> +
Fig. 2: Gai-GDP \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) 2D NMR spectral changes as a function of pH. (A) Representative 2D \(^{1}\mathrm{H}\) - \(^{15}\mathrm{N}\) TROSY-HSQCP spectral overlay of \(^{2}\mathrm{H}\) , \(^{13}\mathrm{C}\) , \(^{15}\mathrm{N}\) -enriched Gai-GDP (230 \(\mu \mathrm{M}\) ) at pH 6.4, 6.8, and 7.2, highlighting peak shifts and line broadening. Spectra were acquired on a Bruker Avance III 850 MHz instrument at \(25^{\circ}\mathrm{C}\) ( \(\mathrm{N} = 2\) ). (B) Chemical shift perturbation (CSP) and (C) peak intensity ratio \((\mathrm{Int}_{\mathrm{pH}6.4} / \mathrm{Int}_{\mathrm{pH}7.2})\) for pH 6.4 and 7.2. The striped horizontal line represents the mean value of \(\Delta \delta\) amide. Most of the residues that show significant CSP, and broadening lie within the Ras-like domain \((\alpha 1,\alpha 5,\beta 1,\beta 2\) and \(\beta 2 - \beta 3\) loop). (D) Spectral differences are highlighted on a ribbon diagram of GTP-bound Gai (PDB: 1CIP). NH residues with CSP greater than 0.03 ppm are represented in red. Residues with decreased peak intensity (line broadening) at pH 6.4 relative to pH 7.2 are shown in green. Residues missing or unassigned are shown in black color while unaffected residues are shown in blue color.
+ +<|ref|>text<|/ref|><|det|>[[112, 832, 883, 905]]<|/det|> +Residue- specific chemical shift perturbation and peak intensity changes associated with changes in pH between 6.4 and 7.2 are plotted in Fig. 2B and C, respectively. Most pH- dependent spectral changes are localized to \(\alpha 1\) , \(\alpha 5\) , \(\beta 1\) , \(\beta 2\) and the \(\beta 2 - \beta 3\) loop within the Ras- like domain as mapped + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 292]]<|/det|> +onto the Gai- GTP structure (PDB:1CIP) in Fig. 2D. Of note, several resonances within these key regions are undetectable in the GDP- bound state at all pH values (Fig. 2D, black) which somewhat limits NMR analyses. These findings are consistent with previous work showing that residues associated with the Switch regions in the GDP- bound state of Gai exhibit enhanced backbone dynamics and are broadened and undetectable compared to the resonances associated with the GTP- bound state7. Taken together, 2D NMR analyses, CD data and tryptophan fluorescence data indicate that the Switch regions of Gai- GDP are more dynamic and less structured at pH 6.4 - 6.8 compared to pH 7.2 - 7.5. + +<|ref|>sub_title<|/ref|><|det|>[[113, 309, 444, 328]]<|/det|> +## pH-dependence of GDP binding to Gai + +<|ref|>text<|/ref|><|det|>[[111, 344, 884, 574]]<|/det|> +Gai when bound to GDP, adopts a conformational ensemble and dynamic properties distinct from that of the GTP- bound state7. This in turn drives recognition of regulatory factors and downstream targets. Given our findings that pH modulates Gai- GDP structure, stability and dynamics, we asked if pH could alter GDP binding to Gai protein. For that purpose, we performed Mant- GDP dissociation assays. For these assays, the rate of GDP dissociation was determined by adding excess GDP to Mant- GDP loaded Gai and monitoring Mant fluorescence changes (by FRET upon tryptophan excitation) as a function of time at different pH values (Fig. 3A- B). As shown in Fig. 3C- D, small differences in GDP dissociation rates were observed over the pH range (pH 6.8- 7.4), indicating that GDP binding is not significantly altered at physiological pH. + +<|ref|>sub_title<|/ref|><|det|>[[113, 590, 707, 610]]<|/det|> +## Identification and characterization of pH sensing residues of Gai-GDP + +<|ref|>text<|/ref|><|det|>[[111, 625, 884, 908]]<|/det|> +To elucidate the molecular basis for pH- dependent stability and conformational dynamic changes in GDP- bound Gai, we sought to identify key residues involved in pH sensing. As extensive broadening of resonances in GDP- bound spectra prevented pKa determination by NMR, we examined and identified two networks of charged residues in regions, designated as the ‘GDP release network’ (Fig. 4A) and ‘Switch network’ (Fig. 5A), that showed pH- dependent changes in the NMR spectra (Fig. 2D). The GDP release network contains residues within \(\alpha 1\) , \(\alpha 5\) , \(\beta 2\) , and \(\beta 3\) , and was previously shown to be important for GDP release during the GPCR- mediated activation of the Gai23. We postulated that if this network plays a key role in pH sensing, mutation of charged residues within this network (e.g., H57, H188, K192, D193, H195 and D337) would reduce pH- dependent Gai thermostability, due to protonation and loss of electrostatic interactions that destabilize Gai tertiary structure. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 92, 872, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 578, 884, 703]]<|/det|> +
Fig. 3: Rates of Gai GDP nucleotide dissociation vary with pH. (A) Structure of Gai (PDB:1GP2) highlighting the position of tryptophan (W211) in Switch II and bound Mant-GDP. (B) Diagram displaying experimental setup of FRET-based determination of Gai nucleotide association and dissociation rates. (C) Rate of GDP dissociation from Gai as a function of pH by monitoring the time-dependent decrease in FRET emission of Gai-loaded Mant-GDP at 445 nm upon the addition of \(7.5 \mu \mathrm{M}\) GDP. (D) The rate of Mant-GDP dissociation decreases as the pH increases. Data are averages of two independent experiments performed in duplicate (±SE).
+ +<|ref|>text<|/ref|><|det|>[[112, 717, 884, 896]]<|/det|> +To examine whether this network modulates pH- dependent changes in stability and nucleotide binding, we generated several Gai variants and conducted pH- dependent CD thermal melt and nucleotide dissociation assays. To select neutral or uncharged amino acid substitutions that retain or have minimal effect on Gai structure, we employed the Rosetta modeling suite. Rosetta replaces a desired amino acid within the protein to all possible amino acid substitutions and predicts associated free energy changes24. Substitutions that minimally perturb free energy are predicted to retain protein structure. Using this strategy, we identified four variants (H57T, H188V, K192Q + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 884, 293]]<|/det|> +and H195N) predicted to retain Gai structure and stability. To evaluate pH- dependent thermostability associated with these Gai GDP release network variants, we acquired CD thermal melts at pH 6 and 7.2, as shown in Fig. 4. Of note, all four variants retained pH- dependent thermostability similar to WT Gai (data not shown for H188 and H195) (Fig. 4C- E). We also performed GDP dissociation assays to evaluate whether the variants alter nucleotide binding (Fig. 4B). All variants showed pH- dependent GDP dissociation rates similar to WT Gai. These findings indicate that the GDP release network does not significantly modulate pH- dependent stability or nucleotide binding (Fig. 4B- F). + +<|ref|>image<|/ref|><|det|>[[210, 313, 777, 623]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 644, 884, 790]]<|/det|> +
Fig. 4: pH-dependent nucleotide exchange and stability of Gai variants from the GDP release network. (A) Ribbon diagram of GTP-bound Gai (PDB: 1CIP) highlighting electrostatic network near GDP release network. The helical domain, Ras-like domain and the Switch regions are shown in gray, blue, and green, respectively. Charged residues are shown as yellow sticks. (B) Rates of GDP dissociation is compared for WT Gai and Gai variants (H57T and K192Q) as a function of pH. Data are averages of two independent experiments performed in duplicate (±SE) (C) Representative CD melt profile of GDP-bound WT and variant Gai (N=2), Gai H57T (D) and K192Q (E) proteins. Gai release network variants retain pH-dependent thermal unfolding profiles similar to WT Gai.
+ +<|ref|>text<|/ref|><|det|>[[113, 806, 876, 904]]<|/det|> +As a number of charged residues within the Switch regions form stabilizing electrostatic interactions in the active Gai- GTP bound state, we next postulated that the decreased stability of Gai- GDP at lower pH may result from protonation of pH- dependent Switch network (SW- I, SW- II, SW- III and \(\alpha 3\) ) residues. To test this hypothesis, we selected residues from the Switch network + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 876, 900]]<|/det|> +shown or predicted to be important for Switch stability in the GTP- bound form of Gai25. Then, based on Rosetta prediction, we mutated a subset of charged residues within this network (e.g. R205N, R208Q, E236L, D237G, R242Q and E245N) predicted to least perturb Gai structure (Fig. 5A). As shown in Fig. 5B- E, two of the variants located in SW- III (E236L and D237G) and \(\alpha 3\) (E245N), respectively, showed a reduction in pH- dependent thermostability between pH 6 and 7.2 relative to WT Gai. Moreover, a ‘double variant’ consisting of two substitutions (E236L and E237G) from SW- III showed further reduction in pH- induced stability but not complete abolishment of pH dependence (Fig. 5F). Yet, a ‘triple variant’ containing all three substitutions (Gai E236L, D237G and E245N) effectively eliminated pH- induced stability changes relative to WT Gai (Fig. 5G), suggesting a key role for these three Switch network residues (E236, D237 and E245) in stabilizing Gai at higher pH. Other variants from the Switch network at R205 and R208 did not show any change in pH- dependent thermostability in comparison to WT Gai protein (Supplementary Fig. S2). We also monitored tryptophan fluorescence of the GDP- bound Gai triple variant as a function of pH to examine changes in solvent accessibility of SW- II residue W211. Consistent with the loss in pH- dependent thermal stability, we found that the double variant and triple variant reduce and abolish (Supplementary Fig. S3) pH- dependent fluorescence intensity changes relative to WT Gai- GDP, respectively (Fig. 5H). As the CD thermal and fluorescence profiles associated with the double variant and triple variant mimic WT Gai at lower pH (pH 6), we refer to these variants as “low pH mimetics”. Further to confirm that the identified residues participate in pH sensing, we performed NMR analyses for the double variant at pH 6.4 and pH 7.2 in comparison to WT GDP- bound Gai. As shown in Supplementary Fig. S4, pH- dependent changes in peak intensity and chemical shift perturbations are significantly reduced in the double variant with respect to WT Gai. Taken together, our results point to key residues in the Switch network that regulate pH- dependent stability and conformational dynamic properties. MD simulations of pH- dependent Gai- GDP conformational dynamics We identified three residues in the GDP- bound Gai, including two SW- III residues (E236, D237) and one \(\alpha 3\) residue (E245), that appear to play a key role in pH- dependent stability and conformational dynamics. To evaluate how these residues form pH- dependent electrostatic interactions that stabilize the Switch regions at higher pH, we employed MD simulations. We generated Gai structures using the Gai- GDP crystal structure (PDB: 1GP2) as a starting point and then changed the protonation state of side chains associated with residues E236, D237 and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 90, 579, 108]]<|/det|> +E245 to simulate a low pH and high pH state, respectively. + +<|ref|>image<|/ref|><|det|>[[179, 130, 812, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 594, 884, 630]]<|/det|> +
Fig. 5. pH-dependent stability and tryptophan fluorescence of Gai Switch network variants. (A) The putative pH-dependent electrostatic network (red) within the Switch regions (green) is highlighted on the ribbon diagram of
+ +<|ref|>text<|/ref|><|det|>[[112, 634, 884, 700]]<|/det|> +GTP- bound Gai (PDB: 1CIP). (B) Comparison of representative CD melt profiles obtained from two independent experiments at pH 6.0 and 7.2 for WT Gai- GDP, (C) Gai- GDP E245N, (D) Gai- GDP E236L, (E) Gai- GDP D237G (F) Gai- GDP double variant (E236L/D237G), and (G) a Gai- GDP triple variant (E236L/D237G/E245Q). The CD thermal profile for Gai single and double variants shows decreased pH- dependent thermal stability while the triple variant shows a complete loss of pH- dependent thermal stability compared to WT Gai. (H) Representative intrinsic tryptophan fluorescence spectra of Gai- GDP triple variant (2 \(\mu \mathrm{M}\) , excitation = 280 nm, emission = 300 - 400 nm) as a function of pH (N = 2, performed in triplicate). Consistent with the CD results, pH- dependent intrinsic tryptophan (W211) fluorescence observed for WT Gai- GDP is abolished for the Gai triple variant supporting a role for E236, D237 and E245 in pH- dependent stability and structural changes. + +<|ref|>text<|/ref|><|det|>[[112, 704, 884, 828]]<|/det|> +thermal profile for Gai single and double variants shows decreased pH- dependent thermal stability while the triple variant shows a complete loss of pH- dependent thermal stability compared to WT Gai. (H) Representative intrinsic tryptophan fluorescence spectra of Gai- GDP triple variant (2 \(\mu \mathrm{M}\) , excitation = 280 nm, emission = 300 - 400 nm) as a function of pH (N = 2, performed in triplicate). Consistent with the CD results, pH- dependent intrinsic tryptophan (W211) fluorescence observed for WT Gai- GDP is abolished for the Gai triple variant supporting a role for E236, D237 and E245 in pH- dependent stability and structural changes. + +<|ref|>text<|/ref|><|det|>[[112, 844, 877, 891]]<|/det|> +Analysis of MD simulation trajectories of GDP- bound Gai collected for 250 ns in charged versus uncharged states revealed higher root mean square deviation (RMSD) in the protonated or + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 876, 397]]<|/det|> +uncharged state. As shown in Fig. 6A, the RMSD of the uncharged state indicated higher dynamics, with RMSD values around \(0.4 \mathrm{nm}\) , while the charged state exhibited lower RMSD values of approximately \(0.25 \mathrm{nm}\) . Based on analyses of the MD trajectories, we attribute the RMSD reduction associated with the charged state to the formation of salt- bridge interactions within the Switch regions. To further probe the dynamic properties of the Gai- GDP Switch regions, we calculated residue- specific root mean square fluctuation (RMSF) in the uncharged state (Fig. 6B and C) and mapped RMSF changes onto the Gai- GTP crystal structure (PDB:1CIP). As shown in Fig. 6B and C, our simulations suggest increased dynamics of SW- I, SW- II, and SW- III in the uncharged state compared to the charged state, consistent with reduced thermal stability at lower pH. Overall, these findings suggest that at lower pH (pH 6.4- 7), there is a loss of electrostatic interactions within the Switch regions of GDP- Gai that promotes enhanced dynamics, consistent with observations from NMR and CD data. + +<|ref|>text<|/ref|><|det|>[[111, 416, 877, 805]]<|/det|> +Representative snapshots extracted from the MD trajectories of Gai- GDP provide insights into the dynamic behavior of Gai as a function of pH. In the charged or higher pH state of Gai- GDP, we observe the transient formation of three critical salt- bridge interactions: E236 interacting with R205, E245 with R208, and E245 with K248 (Fig. 6D and F). Notably, these residues form electrostatic interactions as observed in crystal structures of active Gai- GTP state (PDB:1CIP) and play a pivotal role in the stabilization of the Switch regions. In particular, E236 and D237 side chains from SW- III interact with R205 and R208 in SW- II whereas \(\alpha 3\) residue E245 interacts with R208 \(^{25}\) . In the uncharged state of Gai- GDP (lower pH), conformations extracted from MD simulation trajectories exhibit a notable absence of these salt- bridge interactions (Fig. 6E), which we attribute to the destabilization of the SW- III and SW- II region (Fig. 6E). As a result, the protein displays increased dynamics, as evidenced by higher RMSD and RMSF values, suggesting greater structural fluctuations. The protonation of key residues interferes with the formation of electrostatic interactions leading to their destabilization. Consequently, the protein becomes more dynamic and less thermally stable under more acidic conditions. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 88, 880, 390]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 410, 883, 429]]<|/det|> +
Fig. 6: MD simulations of WT Gai-GDP highlighting uncharged and charged states of E236, D237 and E245.
+ +<|ref|>text<|/ref|><|det|>[[112, 432, 885, 778]]<|/det|> +To mimic the low pH state of Gai- GDP (uncharged state), E236, D237 and E245 side chains were protonated using the Gai- GDP crystal structure (PDB: 1GP2). MD simulations of Gai- GDP in both charged (deprotonated) and uncharged states (protonated) were performed for 250 ns. (A) Root means square deviation (RMSD) of Gai Cα atoms in charged (black) and uncharged (red) states, as estimated by superposing each frame of the trajectory to the corresponding starting structure. Gai- GDP shows higher RMSD in the uncharged state indicative of enhanced dynamics consistent with NMR at lower pH. (B) Residue- wise root mean square fluctuation (RMSF) derived from MD trajectory (100 ns) of charged (black) and uncharged (red) Gai- GDP, highlights enhanced Switch dynamics in the uncharged state. (C) Mapping of residues that display higher RMSF (orange) on the Gai- GTP crystal structure (1CIP) indicates that the SW- I, SW- II and SW- III become more dynamic in the uncharged state compared to the charged state, consistent with the reduced thermal stability and intrinsic tryptophan fluorescence at lower pH. (D) Time evolution of Gai - GDP salt- bridge formation associated with charged trajectory shows three salt- bridges in the charged state but not in the uncharged state of Gai. The cut- off value of salt- bridge distance = 3.2 Å is shown as a dotted line. (E) Representative conformation of Gai- GDP in the uncharged (protonated) state extracted from the MD trajectory shows a lack of salt- bridge interactions due to the destabilization of the SW- III region. (F) Zoomed section depicting salt- bridge interactions extracted from the corresponding MD trajectory (E236- R205, E245- R208, and E245- K248) for the average conformation of charged Gai- GDP. + +<|ref|>text<|/ref|><|det|>[[112, 791, 877, 890]]<|/det|> +To predict how a pH- dependent disorder- to- order transition in Gai- GDP affects the interaction of Gai with binding partners, we analyzed interactions with GPCRs or \(\mathrm{GB}\gamma\) . Agonist- simulated GPCRs interact with Gai- GDP through \(\alpha 5\) , which is part of the GDP release network. As inspection of the crystal structure of \(\beta 1\) - adrenergic receptor with Gai and \(\mathrm{GB}\gamma\) (PDB: 7S0F) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 877, 585]]<|/det|> +indicates GPCR engagement does not alter the structure of Switch II and III26, we predict that pH changes over the physiological pH range do not significantly modulate GPCR interactions with Gai. Conversely, \(\mathrm{G}\beta \gamma\) binds to Gai- GDP through the SW- II region which is more dynamic in the GDP- bound form than it is in the GTP- bound state. We propose, consistent with a recent study25, that residues from the Switch network (including the three identified pH sensing residues) provide stabilization of SW- II in the GTP- bound form of Gai which in turn, prevents interaction with \(\mathrm{G}\beta \gamma\) . Since our MD data show enhanced dynamics in the protonated or lower pH state of Gai- GDP, we predict that at the lower end of the physiological pH range, Gai engages \(\mathrm{G}\beta \gamma\) with higher affinity resulting in downregulation of Gai and \(\mathrm{G}\beta \gamma\) mediated downstream signaling (Fig. 8). To evaluate this hypothesis computationally, we performed HADDOCK docking of Gai- GDP with \(\mathrm{G}\beta \gamma\) in both deprotonated charged and protonated uncharged states. The Gai structures in the charged and uncharged states were obtained as described above, while the \(\mathrm{G}\beta \gamma\) structure was extracted from the Gai- \(\mathrm{G}\beta \gamma\) trimeric complex crystal structure (PDB: 1GP2). The HADDOCK docking score, which is a weighted sum of a variety of energy terms (including van der Waals, electrostatic, desolvation, and restraint violation), was calculated to be - 147 and - 203 for the charged and uncharged states, respectively. Evaluation of this docking score suggested higher affinity binding of Gai with \(\mathrm{G}\beta \gamma\) in the uncharged state due to additional polar contacts between Gai and \(\mathrm{G}\beta \gamma\) , implying that low pH enhances the binding of Gai to \(\mathrm{G}\beta \gamma\) . Taken together, our computational analyses suggest that Gai- \(\mathrm{G}\beta \gamma\) interactions are modulated by pH. + +<|ref|>sub_title<|/ref|><|det|>[[115, 604, 519, 624]]<|/det|> +## Analyses of pH-dependent Gai- \(\mathrm{G}\beta \gamma\) interactions + +<|ref|>text<|/ref|><|det|>[[112, 639, 886, 896]]<|/det|> +\(\mathrm{G}\beta \gamma\) interacts with the dynamic SW- II region of Gai in the GDP- bound state. Upon upstream activation by GPCRs, Gai becomes activated through the exchange of GDP for GTP. In the GTP- bound state, the SW- II region becomes rigid and Gai is unable to adopt a conformation compatible with binding to \(\mathrm{G}\beta \gamma\) . Given our findings that Gai adopts a more dynamic state at the lower end of the physiological pH range (6.8- 7), we hypothesized that Gai- GDP engages \(\mathrm{G}\beta \gamma\) with a higher affinity at lower pH, whereas higher pH (7- 7.5) promotes \(\mathrm{G}\beta \gamma\) release from Gai- GDP. Consistent with this hypothesis, our MD analyses indicate that pH- mediated changes in the protonation state of key Switch residues alter electrostatic interactions and the dynamic properties of Switch regions of Gai in its GDP- bound form which may, in turn, modulate Gai- \(\mathrm{G}\beta \gamma\) interactions and thus the downstream signaling (Fig. 8). To experimentally evaluate the pH dependence of Gai- \(\mathrm{G}\beta \gamma\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 886, 792]]<|/det|> +interactions in a cellular context, we conducted bioluminescence resonance energy transfer (BRET) assays in human HEK293T cells. For these assays, WT Gai was tagged with Renilla luciferase (RLuc) and co- transfected with \(\mathrm{G}\gamma\) tagged with GFP, \(\mathrm{G}\beta\) and the neurotensin receptor. Upon stimulation with neurotensin (agonist), the \(\mathrm{G}\alpha\) - RLuc (energy donor) and \(\mathrm{G}\beta \gamma\) - GFP (energy acceptor) dissociate, with receptor- catalyzed dissociation of the \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) complex measured by comparing energy transfer from donor to acceptor (Fig. 7A). To directly evaluate whether lower intracellular \(\mathrm{pH}(\mathrm{pH}_i)\) reduces \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) dissociation, we employed two distinct approaches to alter and monitor \(\mathrm{pH}_i\) changes in HEK293 cells. Intracellular \(\mathrm{pH}\) was determined by treating HEK293 cells with BCECF dye and determining the ratio of emission intensity (detected at \(535~\mathrm{nm}\) ) when the dye is excited at \(\sim 490 \mathrm{nm}\) and \(\sim 440 \mathrm{nm}^{20}\) . To convert the fluorescence ratios to \(\mathrm{pH}_i\) , a calibration curve of HEK293 cells treated with \(5 \mu \mathrm{M}\) nigericin was generated (Supplementary Fig. S5A). Treatment of cells with nigericin makes the \(\mathrm{pH}_i\) equivalent to extracellular \(\mathrm{pH}(\mathrm{pH}_e)\) thus the calibration curve for \(\mathrm{pH}_e\) vs. fluorescence can be used to convert fluorescence into cellular \(\mathrm{pH}_i\) . To alter the \(\mathrm{pH}_i\) , we compared results using two different strategies. One common method to reduce \(\mathrm{pH}_i\) is the addition of Carbonyl cyanide 4- (trifluoromethoxy) phenylhydrazone (FCCP), an uncoupler of oxidative phosphorylation in the mitochondria \(^{27}\) . We determined that cells treated with \(1 \mu \mathrm{M}\) FCCP generated a \(\mathrm{pH}_i\) of \(\sim 7.04\) , while at \(5 \mu \mathrm{M}\) the \(\mathrm{pH}_i\) was further reduced to 6.85 (Supplementary Fig. S5B). As one drawback of FCCP is that it can alter mitochondrial energetics, we employed a second method to alter \(\mathrm{pH}_i\) by modulating extracellular \(\mathrm{pH}\) . As shown in Supplementary Fig. S5C, lowering extracellular \(\mathrm{pH}\) to 6 and 5, caused a reduction in \(\mathrm{pH}_i\) from 6.9 and 6.7, respectively which is consistent with a previous report \(^{28}\) . To evaluate the \(\mathrm{pH}\) dependence of \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) dissociation in cells, we conducted the BRET assay by expressing WT Gai in HEK293 cells and inducing acidosis in the cytoplasm by either altering extracellular \(\mathrm{pH}\) (Fig. 7B) or by the addition of varying concentrations of FCCP (Fig. 7C). Results obtained from the BRET assays show that lowering \(\mathrm{pH}_i\) ( \(\mathrm{pH}_i\) 7.2- 6.85 by FCCP or 7.2- 6.7 by changing \(\mathrm{pH}_e\) ) reduces agonist- mediated \(\mathrm{G}\beta \gamma\) release from \(\mathrm{G}\alpha \mathrm{i}\) , indicating that \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) interactions are highly dependent on \(\mathrm{pH}\) over the physiological \(\mathrm{pH}\) range. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[225, 90, 765, 586]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 603, 884, 860]]<|/det|> +
Fig. 7: BRET assay showing enhanced Gai-Gβγ association at lower intracellular pH for the Gai Switch variants. (A) Schematic diagram (BioRender) illustrating the BRET assay used to monitor receptor-mediated dissociation of Gai from Gβγ. For this assay, the GPCR neurotensin receptor, Ga Renilla luciferase 8 (Rluc8), Gβ3 and γ9-GFP fusion constructs were co-transfected in HEK29T cells and fluorescence (510 nm) monitored after addition of the substrate and agonist (neurotensin). Net BRET (dissociation) plotted as γ9-GFP (acceptor) over Ga-Rluc (donor) ratio as a function of log doses of neurotensin. Neurotensin receptor-mediated BRET shows that lowering pHi (from 7.2 to 6.7) either by altering extracellular pH (B) or by the addition of FCCP (C), decreases Gai dissociation from Gβγ in HEK29T cells. (D) WT Gai shows significantly higher dissociation from Gβγ relative to both double variant and triple variants (low pH mimetics of WT Gai). (E) The difference in net BRET signal as a function of FCCP concentration (0 μM in solid line and 2 μM in dotted line) for double variant and triple variants is reduced with respect to WT Gai, confirming key roles for E236, D237, and E245 in pH-mediated Gβγ dissociation for Gai-Gβγ complex. All BRET data are averages of two independent experiments performed in duplicate (±SE).
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 293]]<|/det|> +Further to confirm that the modulation of \(\mathrm{G}\beta \gamma\) release by pH occurs through the identified 3 residues of \(\mathrm{G}\alpha \mathrm{i}\) , we performed the BRET assay using the double or triple \(\mathrm{G}\alpha \mathrm{i}\) variants. As shown in Fig. 7D, agonist- stimulated \(\mathrm{G}\alpha \mathrm{i}\) dissociation from \(\mathrm{G}\beta \gamma\) is significantly reduced for the "double variant" and further reduced for the triple variant with respect to WT \(\mathrm{G}\alpha \mathrm{i}\) . These variants appear to serve as low pH mimetics, as they produce BRET similar to that obtained at lower \(\mathrm{pH_i}\) . This observation was further validated by BRET assays performed for double variant and triple variant at different \(\mathrm{pH_i}\) . Taken together, these findings suggest that enhanced \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) association at lower pH is due to the loss of electrostatic interactions within the Switch network needed for \(\mathrm{G}\alpha \mathrm{i} - \mathrm{G}\beta \gamma\) formation. + +<|ref|>sub_title<|/ref|><|det|>[[114, 309, 211, 327]]<|/det|> +## Discussion: + +<|ref|>text<|/ref|><|det|>[[112, 351, 884, 477]]<|/det|> +The recognition of pH sensors among biomolecules is crucial in understanding cellular processes, as they play a pivotal role in responding to changes in the proton concentration within the physiological range. While many biomolecules experience alterations in their protonation state, only a specific subset serves as pH sensors, exhibiting pH- dependent functional changes that influence cellular processes. + +<|ref|>text<|/ref|><|det|>[[112, 500, 884, 729]]<|/det|> +One of the first pH- sensing proteins to be characterized is hemoglobin. In acidic environments, hemoglobin exhibits reduced affinity for oxygen, a phenomenon known as the Bohr effect. This effect is due to the protonation of specific amino acid residues, including histidine 146, located in the \(\alpha\) subunit of hemoglobin29. The protonation of H146 promotes the formation of a salt bridge with a nearby aspartate residue (D94) on the \(\beta\) subunit. This interaction stabilizes the deoxygenated or T- state conformation of hemoglobin, reducing its affinity for oxygen and facilitating the release of oxygen to tissues where it is needed30. After the identification of hemoglobin, proton- sensing ion channels and receptors governing cytosolic pH have been a focal point in research. + +<|ref|>text<|/ref|><|det|>[[112, 753, 884, 905]]<|/det|> +More recent structural informatics calculations have shown that buried ionizable networks are a structural hallmark of pH sensitivity3, and the large pKa shifts exhibited by buried sidechains can be harnessed by proteins to drive pH- dependent changes in structure, stability, and function. Indeed, networks of buried ionizable amino acids are conserved in nearly all of the \(100+\) \(\mathrm{G}\alpha\) protein structures in the Protein Data Bank (PDB)3. On that basis, the cell signaling GTPases, \(\mathrm{G}\alpha \mathrm{i}\) and its yeast homolog Gpa1, were proposed to function as pH- sensing proteins. The study indicated that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 291]]<|/det|> +the stability of heterotrimeric \(\mathrm{G}\alpha\) proteins exhibits a significant dependence on pH levels across a broad range from 5 to 8.0. Furthermore, alterations in yeast intracellular \(\mathrm{pH} (\mathrm{pH}_i)\) were shown to promote Gpa1 phosphorylation and subsequently dampen the mitogen- activated protein (MAP) kinase signaling pathway4. In addition to our findings that Gαi serves as a pH sensor to regulate \(\mathrm{G}\alpha - \mathrm{G}\beta \gamma\) interactions, recent studies indicate that select GPCRs possess pH sensing properties that modulate extracellular signaling to \(\mathrm{G}\alpha\) proteins31,32,33. These and other GPCRs may work independently or synergistically with Gαi proteins to transduce extracellular pH- dependent signals to changes in intracellular \(\mathrm{pH}^{28,34,35}\) . + +<|ref|>text<|/ref|><|det|>[[112, 312, 885, 835]]<|/det|> +Herein, we find that as pH is raised from 6.8 to 7.3, the thermostability of Gαi- GDP is greatly enhanced ( \(\Delta \mathrm{Tm} \sim 20^{\circ}\mathrm{C}\) ) due to the formation of an electrostatic network within the Ras- like domain. Since thermostability changes for GDP- bound Gαi were significantly larger ( \(\sim 20^{0}\) ), relative to the GTP- bound form ( \(\sim 8^{0}\) ), we focused on characterizing pH- dependent structural and dynamic changes associated with the GDP- bound form of Gαi in this study. Using pH- dependent NMR analyses and thermal stability profiling on WT and mutant proteins, we identified three key pH- sensing residues that drive pH- dependent thermostability changes. Two of the residues (E236 and D237) reside in SW- III of Gαi while the third residue (E245) lies in the neighboring α3 helix. Mutation of E236 and D237 in SW- III promotes Gβγ release in HEK293 cells, supporting the role of these residues in pH- dependent electrostatic interactions that allosterically regulate heterotrimer formation. While both Ras and heterotrimeric G- proteins contain SWI and SWII regions, Ras proteins lack the SW- III region found in Gα proteins. The presence of this unique SW- III region in heterotrimeric G- proteins may explain the unique pH sensing properties of Gαi and provide a novel role for the SW- III region in pH- dependent Gβγ release from the Gαi- Gβγ complex. Of note, the pH sensing network identified in this study is distinct from the network predicted by Isom et al3. This computationally predicted pH sensing network consists of residues at the Ras/helical domain interface (e.g. K46, D150, D200, D229, R242, K270, and K277). However, our NMR and CD data do not support this interaction network, as observable NMR resonances associated with K270 and K277 do not show pH- dependent perturbations and mutation of K46 and R242 did not alter pH- dependent thermostability changes measured by CD. + +<|ref|>text<|/ref|><|det|>[[113, 857, 884, 903]]<|/det|> +Large changes in the stability of GTP- bound Gαi have previously been attributed to the formation of an interaction triad (termed as G- R- E Triad) involving residues G203 and R208 from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 884, 267]]<|/det|> +SW- II and E245 from \(\alpha 3\) . This interaction triad is absent from the less stable and more dynamic GDP- bound state \(^{7,25}\) . Interestingly, E245 is also one of the residues identified in the pH sensing network. Moreover, E236 and D237 from SW- III interact with SW- II via residue R205 to stabilize both Switches in the GTP- bound forms of the protein. This nucleotide- dependent disorder- to- order transition is predicted to facilitate the release of \(\mathrm{G}\beta \gamma\) from Gai resulting in the activation of Gai. Consistent with these findings, SW- III residues E236 and D237 are involved in pH- dependent electrostatic interactions that appear to allosterically regulate \(\mathrm{G}\beta \gamma\) release. + +<|ref|>image<|/ref|><|det|>[[117, 288, 875, 592]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 647, 884, 730]]<|/det|> +
Fig. 8: Proposed mechanism of Gai pH regulation. Activation of Gai by GPCRs leads to the release of Gai from \(\mathrm{G}\beta \gamma\) and stimulation of signaling pathways. A distinct mechanism of signaling regulation may occur by intracellular pH regulation. Gai protein may act as intracellular pH sensors to regulate Gai- \(\mathrm{G}\beta \gamma\) interactions with lower pH enhancing Gai- \(\mathrm{G}\beta \gamma\) interactions to inhibit Gai signaling.
+ +<|ref|>text<|/ref|><|det|>[[113, 734, 884, 885]]<|/det|> +While several pH- sensing proteins contain titratable histidines \(^{36,37,38}\) , others such as EmrE contain aspartate and glutamate residues that titrate in the physiological range due to the formation of electrostatic networks \(^{32,39}\) . In these systems, even a small reduction in pH can alter the side chain protonation state leading to the neutralization of charge. In the case of Gai, we identified a network of three charged residues (E236, D237 and E245) that regulate pH- dependent thermostability changes in the Gai- GDP. To investigate how protonation/deprotonation of these residues might + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 885, 426]]<|/det|> +alter Gαi structure and dynamics in the GDP- bound state, we performed MD simulations. As starting points for the simulations, residues E236, D237 and E245 were generated in both charged and uncharged states, to represent higher (deprotonated) and lower pH (protonated) states, respectively. Results from these analyses suggest that the protonation of residues E236, D237 and E245 enhances Gαi- GDP dynamics due to the loss of electrostatic interactions between SW- II, SW- III, and α3. Moreover, we find that these residues make transient electrostatic interactions in Gαi- GDP at higher ( \(\sim\) pH 7.2), but not at lower pH. Since the binding of Gβγ to the Gαi depends on the dynamic properties of SW- II, we propose that lower physiological pH promotes Gβγ binding to Gαi resulting in attenuation of both Gαi and Gβγ mediated signaling pathways. Consistent with this premise, our findings indicate that higher intracellular pHi, promotes agonist- mediated Gβγ release from Gαi via the formation of pH- dependent electrostatic networks involving key residues within SW- III and α3. This in turn, will regulate Gαi and Gβγ downstream signaling. + +<|ref|>text<|/ref|><|det|>[[112, 444, 885, 596]]<|/det|> +While the current investigation is centered on the Gαi isoform, sequence and structural alignment with other Gα isoforms (such as Gαs, Gαo, and Gαq) reveals conservation within the core pH sensing network (Supplementary Fig. S6)3. This analysis suggests the potential for pH- sensing properties in other Gα isoforms. The development of pH- insensitive forms of Gαi will facilitate investigations of hormone- and neurotransmitter signaling during physiological stresses, such as occur during glucose or oxygen deprivation, leading to changes in cellular pH40. + +<|ref|>sub_title<|/ref|><|det|>[[115, 629, 215, 647]]<|/det|> +## References: + +<|ref|>text<|/ref|><|det|>[[110, 680, 852, 888]]<|/det|> +1. Madshus, I. H. Regulation of intracellular pH in eukaryotic cells. Biochem. J. 250, 1–8 (1988). +2. Damaghi, M., Wojtkowiak, J. W. & Gillies, R. J. pH sensing and regulation in cancer. Front. Physiol. 4 DEC, 1–10 (2013). +3. Isom, D. G. et al. Protons as second messenger regulators of G protein signaling. Mol. Cell 51, 531–538 (2013). +4. Hewitt, N. et al. Catalytic site mutations confer multiple states of G protein activation. Sci. Signal. 16, eabq7842 (2023). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 880, 898]]<|/det|> +5. Murga, C. et al. G protein-coupled receptor kinase 2 (GRK2) as a potential therapeutic target in cardiovascular and metabolic diseases. Front. Pharmacol. 10, (2019). +6. Kimura, T., Pydi, S. P., Pham, J. & Tanaka, N. Metabolic Functions of G Protein-Coupled Receptors in Hepatocytes—Potential Applications for Diabetes and NAFLD. Biomolecules 10, (2020). +7. Goricanec, D. et al. Conformational dynamics of a G-protein \(\alpha\) subunit is tightly regulated by nucleotide binding. Proc. Natl. Acad. Sci. U. S. A. 113, E3629–E3638 (2016). +8. Olsen, R. H. J. et al. TRUPATH, an open-source biosensor platform for interrogating the GPCR transducermore. Nat. Chem. Biol. 16, 841–849 (2020). +9. Misconai, A. et al. BeSTSeI: webserver for secondary structure and fold prediction for protein CD spectroscopy. Nucleic Acids Res. 50, W90–W98 (2022). +10. Lee, W., Tonelli, M. & Markley, J. L. Structural bioinformatics NMRFAM-SPARKY: enhanced software for biomolecular NMR spectroscopy. 31, 1325–1327 (2015). +11. Goricanec, D. & Hagn, F. NMR backbone and methyl resonance assignments of an inhibitory G-alpha subunit in complex with GDP. Biomol. NMR Assign. 13, 131–137 (2019). +12. Prakash, A., Shin, J., Rajan, S. & Yoon, H. S. Structural basis of nucleic acid recognition by FK506-binding protein 25 (FKBP25), a nuclear immunophilin. 44, 2909–2925 (2016). +13. Wall, M. A. et al. The structure of the G protein heterotrimer Gα1β1γ2. Cell 83, 1047–1058 (1995). +14. Šali, A. & Blundell, T. L. Comparative Protein Modelling by Satisfaction of Spatial Restraints. J. Mol. Biol. 234, 779–815 (1993). +15. Huang, J. & Jr, A. D. M. CHARMM36 All-Atom Additive Protein Force Field : Validation Based on Comparison to NMR Data. 2135–2145 (2013) doi:10.1002/jcc.23354. +16. Parrinello, M. & Rahman, A. Polymorphic transitions in single crystals: A new molecular dynamics method. J. Appl. Phys. 52, 7182–7190 (1981). +17. Essmann, U. et al. A smooth particle mesh Ewald method. J. Chem. Phys. 103, 8577–8593 (1995). +18. Hess, B., Bekker, H., Berendsen, H. J. C. & Fraaije, J. G. E. M. LINCS: A Linear + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 87, 875, 899]]<|/det|> +Constraint Solver for molecular simulations. J. Comput. Chem. 18, 1463- 1472 (1997). 19. van Zundert, G. C. P. et al. The HADDOCK2.2 Web Server: User- Friendly Integrative Modeling of Biomolecular Complexes. J. Mol. Biol. 428, 720- 725 (2016). 20. Behbahan, I. S., McBrian, M. A. & Kurdistani, S. K. A Protocol for Measurement of Intracellular pH. Bio- protocol 4, e1027 (2014). 21. Hamm, H. E., Meier, S. M., Liao, G. & Preininger, A. M. Trp fluorescence reveals an activation- dependent cation- π interaction in the switch II region of Gai proteins. Protein Sci. 18, 2326- 2335 (2009). 22. Shellhammer, J. P. et al. Amino acid metabolites that regulate G protein signaling during osmotic stress. PLOS Genet. 13, e1006829 (2017). 23. Ham, D. et al. Conformational switch that induces GDP release from Gi. J. Struct. Biol. 213, 107694 (2021). 24. Thieker, D. F. et al. Stabilizing proteins, simplified: A Rosetta- based webtool for predicting favorable mutations. Protein Sci. 31, e4428 (2022). 25. Knight, K. M. et al. A universal allosteric mechanism for G protein activation. Mol. Cell 81, 1384- 1396. e6 (2021). 26. Alegre, K. O. et al. Structural basis and mechanism of activation of two different families of G proteins by the same GPCR. Nat. Struct. Mol. Biol. 28, 936- 944 (2021). 27. Berezhnov, A. V. et al. Intracellular pH modulates autophagy and mitophagy. J. Biol. Chem. 291, 8701- 8708 (2016). 28. Kapolka, N. J. et al. Proton- gated coincidence detection is a common feature of GPCR signaling. Proc. Natl. Acad. Sci. U. S. A. 118, (2021). 29. Perutz, M. F., Gronenborn, A. M., Clore, G. M., Fogg, J. H. & Shih, D. T. - b. The pKa values of two histidine residues in human haemoglobin, the Bohr effect, and the dipole moments of α- helices. J. Mol. Biol. 183, 491- 498 (1985). 30. Dajnowicz, S. et al. Visualizing the Bohr effect in hemoglobin: neutron structure of equine cyanomethomoglobin in the R state and comparison with human deoxyhemoglobin in the T state. Acta Crystallogr. Sect. D, Struct. Biol. 72, 892- 903 (2016). 31. Levin, L. R. & Buck, J. Physiological roles of acid- base sensors. Annu. Rev. Physiol. 77, 347- 362 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 88, 881, 710]]<|/det|> +32. Li, J., Her, A. S. & Traaseth, N. J. Asymmetric protonation of glutamate residues drives a preferred transport pathway in EmrE. Proc. Natl. Acad. Sci. U. S. A. 118, (2021).33. Russell, J. L. et al. Regulated expression of pH sensing G protein-coupled receptor-68 identified through chemical biology defines a new drug target for ischemic heart disease. ACS Chem. Biol. 7, 1077–1083 (2012).34. Rowe, J. B., Kapolka, N. J., Taghon, G. J., Morgan, W. M. & Isom, D. G. The evolution and mechanism of GPCR proton sensing. J. Biol. Chem. 296, 100167 (2021).35. Isom, D. G. & Dohlman, H. G. Buried ionizable networks are an ancient hallmark of G protein-coupled receptor activation. Proc. Natl. Acad. Sci. U. S. A. 112, 5702–5707 (2015).36. Vercoulen, Y. et al. A histidine pH sensor regulates activation of the Ras-specific guanine nucleotide exchange factor RasGRP1. Elife 6, 1–26 (2017).37. Tomura, H., Mogi, C., Sato, K. & Okajima, F. Proton-sensing and lysolipid-sensitive G-protein-coupled receptors: A novel type of multi-functional receptors. Cell. Signal. 17, 1466–1476 (2005).38. Williamson, D. M., Elferich, J. & Shinde, U. Mechanism of fine-tuning pH sensors in proprotein convertases: Identification of a pH-sensing histidine pair in the propeptide of proprotein convertase 1/3. J. Biol. Chem. 290, 23214–23225 (2015).39. Schönichen, A., Webb, B. A., Jacobson, M. P. & Barber, D. L. Considering protonation as a posttranslational modification regulating protein structure and function. Annu. Rev. Biophys. 42, 289–314 (2013).40. Carr, R. et al. B-Arrestin-Biased Signaling Through the B2-Adrenergic Receptor Promotes Cardiomyocyte Contraction. Proc. Natl. Acad. Sci. U. S. A. 113, E4107–E4116 (2016). + +<|ref|>sub_title<|/ref|><|det|>[[115, 717, 272, 735]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[112, 752, 886, 850]]<|/det|> +The authors thank Bryan Roth for providing BRET constructs and David Forest for the critical input in Rosetta predictions. Illustrative images in Figs. 3B, 7A and 8 were created with BioRender.com. Research reported in this publication was supported by National Institutes of Health (NIH) grants R35GM134962 (to S. L. Campbell) and R35GM118105 (to H. G Dohlman). + +<|ref|>sub_title<|/ref|><|det|>[[115, 868, 296, 886]]<|/det|> +## Author contributions + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 88, 884, 188]]<|/det|> +729 Conceptualization: A.p., G.Y., H.G.D., S.L.C. Methodology: A.P., Z.L., J.R., N.H.V., G.B.H., G.Y., Experimentation: A.P., Z.L., J.R., N.H.V., G.B.H., G. Y., Simulation: V.R.C. Writing—original draft: A.P., Z.L., V.R.C., H.G.D., S.L.C. Writing—review & editing: G.Y., J.R., N.H.V. Analysis: A.P., Z.L., J.R., N.H.V., G.B.H. Supervision: H.G.D., S.L.C. + +<|ref|>sub_title<|/ref|><|det|>[[63, 205, 286, 224]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[63, 240, 460, 259]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 312, 150]]<|/det|> +Supplementary figures.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/images_list.json b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..71b78c93a7f124da3a19958cb5b19d929826862b --- /dev/null +++ b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/images_list.json @@ -0,0 +1,47 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Observed and simulated trends in mesoscale SST-LHF coupling. Global distribution of the decadal trends of mesoscale SST-LHF coupling strength as derived from ERA5 (a) and CESM-HR (b) during 1979-2022. The coupling strength is computed using the linear regression coefficient between high-pass filtered monthly SST and LHF (Methods) with trends significant at a 99% confidence level indicated by black dots. (c) Composites of SST (color shading) and LHF (contours) anomalies associated with mesoscale oceanic eddies during historical periods (1956-1975) in CESM-HR. Shown are winter season mean anomalies in warm minus cold eddy composites within four WBC regions (outlined by red boxes in a, b). The white circle represents one eddy radius and the white dot marks the eddy center. (d) Mesoscale SST-LHF coupling strength during historical (1956-1975) and future (2063-2082) periods in CESM-HR for warm (red) and cold (blue) eddies averaged across four WBC regions. The coupling strength is computed using the linear regression coefficient between SST and LHF anomalies within twice the radius of eddy composites. (e) Differences in mesoscale SST-LHF coupling strength between future and historical periods in CESM-HR for warm (red bars) and cold (blue bars) eddies in the KE, GS, ARC and BMC regions, with fractional differences labeled atop the respective bars.", + "footnote": [], + "bbox": [ + [ + 150, + 150, + 844, + 483 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Decomposition of mesoscale SST-LHF coupling response under greenhouse warming. (a) Differences in thermodynamic and dynamic adjustments between future and historical periods in CESM-HR in the KE region. From left to right, the terms plotted are thermodynamic adjustment \\(\\overline{U}_{10}(\\frac{aq_s'}{dsST'}, \\frac{aq_a'}{dsST'})\\) , dynamic adjustment \\(\\frac{au_{10}'}{dsST'}(\\overline{q}_s - \\overline{q}_a)\\) , large-scale surface wind \\((\\overline{U}_{10})\\) and moisture adjustment \\((\\frac{aq_s'}{dsST'}, \\frac{aq_a'}{dsST'})\\) . (b-d), as for a, but for the GS, ARC and BMC regions, respectively.", + "footnote": [], + "bbox": [ + [ + 125, + 120, + 864, + 450 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Decomposition of mesoscale moisture response under greenhouse warming. (a) Estimated changes in mesoscale SST-LHF coupling between future and historical periods in CESM-HR according to Eq. 2 in the KE GS, ARC and BMC regions. The large-scale SST averaged in the WBCs are displayed on the bottom axis for each respective region. Black lines denote the standard deviations of the corresponding variables. (b) Contributions of \\(\\overline{U}_{10}, \\frac{d^2 q_s(T)}{dT^2}\\) and \\(\\Delta SST\\) to the north-south hemispheric asymmetry in the KE GS, ARC and BMC regions. The total deviation from the hemispheric mean due to three terms in a specific region is approximated by \\((ABC)' \\approx A' \\bar{B} \\bar{C} + B' \\bar{A} \\bar{C} + C' \\bar{A} \\bar{B}\\) . Here, A, B, and C denotes the three terms; the overbar denotes a mean component (the average across the four WBCs); the prime denotes an anomaly component (the deviation from the hemispheric mean for each specific WBC region). (c) The historical large-scale surface wind, SST and the projected SST warming within the four WBC regions in CESM-HR. (d) as for c, but for observational data. Surface wind is derived from ERA5 and SST is derived from NOAA-OISST. \\(\\Delta SST\\) is computed based on the warming trend observed over the last 40 years.", + "footnote": [], + "bbox": [ + [ + 209, + 90, + 780, + 630 + ] + ], + "page_idx": 23 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d.mmd b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d.mmd new file mode 100644 index 0000000000000000000000000000000000000000..e037742a384ac1db834b469779a1bc7e40cea22d --- /dev/null +++ b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d.mmd @@ -0,0 +1,310 @@ + +# Midlatitude Mesoscale Thermal Air-sea Interaction Enhanced by Greenhouse Warming + +Xingzhi Zhang zhangxingzhi@ouc.edu.cn + +Ocean Univeristy of China Xiaohui Ma Ocean University of China https://orcid.org/0000- 0001- 9937- 3859 Lixin Wu Ocean University of China https://orcid.org/0000- 0002- 4694- 5531 Peiran Yang Laoshan Laboratory Fengfei Song Ocean University of China https://orcid.org/0000- 0002- 3004- 1749 Zhao Jing Ocean University of China https://orcid.org/0000- 0002- 8430- 9149 Hui Chen Ocean University of China Yushan Qu Ocean University of China Man Yuan Ocean University of China Zhaohui Chen Ocean University of China https://orcid.org/0000- 0002- 0830- 2332 Bolan Gan Ocean University of China https://orcid.org/0000- 0001- 7620- 485X + +## Article + +Keywords: + +Posted Date: February 19th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 3932615/v1 + +<--- Page Split ---> + +License: © ① This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on September 4th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52077- z. + +<--- Page Split ---> + +Midlatitude Mesoscale Thermal Air-sea Interaction Enhanced by Greenhouse Warming Xiaohui Ma1,2, Xingzhi Zhang1,*, Lixin Wu1,2, Peiran Yang2, Fengfei Song1,2, Zhao Jing1,2, Hui Chen1, Yushan Qu1, Man Yuan1, Zhaohui Chen1,2, Bolan Gan1,2 + +1. Frontiers Science Center for Deep Ocean Multispheres and Earth System and Key Laboratory of Physical Oceanography, Ocean University of China, Qingdao, China. + +2. Laoshan Laboratory, Qingdao, China. + +\*Corresponding author. E-mail: zhangxingzhi@ouc.edu.cn + +## Abstract + +The influence of greenhouse warming on mesoscale air- sea interactions, crucial for modulating ocean circulation and climate variability, remains largely unexplored due to the limited resolution of current climate models. This study addresses this gap by analyzing eddy- resolving high- resolution climate simulations and observations, focusing on the coupling between mesoscale sea surface temperature (SST) and latent heat flux (LHF) in winter. Our findings reveal a consistent increase in mesoscale SST- LHF coupling in the major western boundary current regions under warming, characterized by a heightened nonlinearity between warm and cold eddies and a more pronounced enhancement in the northern hemisphere. To understand the dynamics, we develop a theoretical framework that links mesoscale thermal coupling changes to large- scale factors, which indicates that the projected changes are collectively determined by historical background wind, SST, and the rate of SST warming. Among these factors, the large- scale SST and its warming rate are the primary drivers of hemispheric asymmetry in mesoscale coupling intensification. This study introduces a simplified approach for assessing the projected mesoscale thermal coupling changes and implies a growing significance of mesoscale air- sea interaction in shaping weather and climate patterns in a warming world. + +<--- Page Split ---> + +Mesoscale air- sea interactions, predominantly active in the Western Boundary Current (WBC) regions in the midlatitudes, play a critical role in modulating extratropical weather and climate systems (Chelton et al., 2004; Small et al., 2008; Bryan et al., 2010; Czaja et al., 2019; Seo et al., 2023). The interaction between mesoscale oceanic eddies and the atmosphere actively impacts precipitation, storms, large- scale atmospheric circulations, and provides feedback to the ocean, influencing oceanic circulations and climate variability (Xie, 2004; Skyllingstad et al., 2007; Chelton and Xie, 2010; Frenger et al., 2013; Souza et al., 2014; Ma et al., 2016, 2017; Foussard et al., 2019; Renault et al., 2019; Gan et al., 2023). A key aspect of this interaction is the coupling between sea surface temperature and turbulent heat flux (SST- THF), hereafter denoted as thermal coupling. In terms of the atmosphere, thermal coupling acts as an energy source, which is crucial for the genesis and development of weather systems and deep troposphere response (Parfitt et al 2016; Chen et al., 2017; Sheldon et al., 2017; Jiang et al., 2019; Zhang et al., 2019; Liu et al., 2021). In terms of the ocean, thermal coupling serves as an energy sink, which is the key to dissipating oceanic eddy energy and driving oceanic circulation response (Ma et al., 2016; Jing et al., 2020). + +How greenhouse warming will impact mesoscale oceanic eddies remains uncertain, let alone mesoscale air- sea coupling. Satellite observations reveal a rise in eddy activity in the WBC regions during recent decades, while climate models project heterogeneous eddy variations in different WBC regions under warming (Martinez- Moreno et al., 2021; Beech et al., 2022). Additionally, the theoretical framework for predicting the mesoscale thermal coupling change in response to greenhouse warming is currently absent. Thermodynamically, the SST warming and the associated water vapor increase governed by the Clausius- Clapeyron (C- C) relation, appear to strengthen the thermal air- sea coupling (primarily through the enhancement of latent heat flux) in a warming climate. Dynamically, the non- uniform warming between the upper and lower troposphere under anthropogenic forcing tends to increase + +<--- Page Split ---> + +atmospheric stability, inhibiting the vertical momentum transfer and thereby suppressing the surface wind and thermal air- sea coupling. This is further complicated by significant uncertainties in atmospheric circulation responses under climate change, introducing additional perturbations to SST- THF coupling. It is noteworthy that the aforementioned oceanic and atmospheric changes discussed within a conventional large- scale framework may not necessarily manifest at mesoscales. Physical processes governing the response of mesoscale thermal coupling to greenhouse warming are multifaceted, making it challenging to pinpoint the ultimate dominant factor. + +Utilizing an unprecedented set of high- resolution Community Earth System Model (referred to as CESM- HR, Methods) with \(\sim 0.25^{\circ}\) atmosphere and \(\sim 0.1^{\circ}\) ocean components that can explicitly resolve the mesoscale oceanic eddies and their coupling with the atmosphere, we investigated the impact of greenhouse warming on mesoscale thermal coupling in eddy- rich WBC regions. We then constructed a theoretical framework for estimating the mesoscale thermal coupling change in response to greenhouse warming through the decomposition of contributing factors. We further assessed the robustness of the findings by extending the analyses to High- Resolution Model Intercomparison Project (HighResMIP, Methods) models. + +## Mesoscale thermal coupling during the observational period + +We first evaluated the model's capability in representing the mesoscale thermal coupling in the CESM- HR simulations against observational data (Methods), in four major WBC regions, i.e., the Kuroshio Extension (KE), the Gulf Stream Extension (GS), the Agulhas Return Current (ARC) and the Brazil- Malvinas Confluence Region (BMC), by detecting eddies using sea surface height anomalies and constructing composite analyses with a reference frame centered on the eddy (see details in Methods). The model shows high fidelity in representing the statistical characteristics of eddies, such as the averaged number, amplitude and size, as detailed in Tab. S1. It also successfully reproduces the eddy- induced turbulent heat fluxes + +<--- Page Split ---> + +(including both sensible and latent components) along with their seasonality, for both anticyclonic warm and cyclonic cold eddies (Fig. S1). It is noted that the coupling strength peaks during the winter month. Furthermore, the eddy- induced sensible heat flux (SHF) is approximately half that of latent heat flux (LHF) and remains relatively stable between historical and future simulations (Fig. S2). Consequently, our subsequent analyses will focus on the mesoscale SST- LHF coupling in the hemispheric winter season — DJF for the Northern Hemisphere (NH) and JJA for the Southern Hemisphere (SH). + +To assess the potential impact of anthropogenic warming on mesoscale thermal coupling, we analyzed the decadal trend of mesoscale SST- LHF coupling in the observational periods based on high- pass spatial filtering fields (Methods). A significant intensification of mesoscale SST- LHF coupling at an approximate rate of \(1.5 \mathrm{W / m^2 / ^\circ C}\) per decade is detected in the WBCs over the past four decades, with the most pronounced increase observed in the GS and KE regions of the NH (Fig. 1a). The CESM- HR successfully simulates the enhanced mesoscale thermal coupling in the WBCs with a magnitude slightly higher than that recorded in the observations (Fig. 1b). The model also captures the north- south hemispheric asymmetry of the changes. The alignment between model simulated and observational trends in the past lends credence to the use of CESM- HR for investigating the mesoscale coupling response under climate change. + +## Mesoscale thermal coupling under greenhouse warming + +A linear regression between mesoscale SST and LHF based on eddy composites averaged across four WBC regions, yields an estimated coupling coefficient of approximately \(28 \mathrm{W / m^2 / ^\circ C}\) for both anticyclonic warm and cyclonic cold eddies in historical simulations (1956- 1975, Fig. 1c,d). The nonlinearity between warm and cold eddies appears to be minimal during the historical period. Under the high greenhouse forcing scenario of Representative Concentration Pathway 8.5, future simulations (2063- 2082) project an approximate \(13\%\) + +<--- Page Split ---> + +enhancement in mesoscale SST- LHF coupling to \(32\mathrm{W / m^2 / ^\circ C}\) (Fig. 1d). A detailed examination of the four WBCs reveals a consistent increase in coupling strength by \(6\%\) to \(20\%\) in future simulations compared to historical simulations (Fig. 1e). Importantly, future simulations project a greater increase of mesoscale SST- LHF coupling strength for warm eddies than for cold eddies, indicating heightened nonlinearity in a warming climate. The increase in nonlinearity is expected to amplify net heat flux from mesoscale oceanic eddies (Ma et al. 2015; Foussard et al., 2019), thereby impacting future weather and climate systems more significantly. Additionally, the projected increase in mesoscale SST- LHF coupling in the GS and KE is twice that of the ARC and BMC, pointing to a more substantial intensification in the NH than in the SH. Combined with the findings in the observational periods, the results indicate that the hemisphere discrepancy in mesoscale SST- LHF coupling is likely to not only persist but also to become more pronounced with ongoing climate warming. + +## Decomposition of contributing factors + +The LHF is proportional to both the surface wind and the humidity difference at the air- sea interface according to the Bulk formula (Large and Yeager 2004). To estimate the mesoscale SST- LHF coupling associated with oceanic eddies, we decompose the relevant fields into large- scale background and mesoscale components in line with previous studies (Yang et al., 2018; Yuan et al, 2023). The mesoscale SST- LHF coupling coefficient can be quantified using the following relationship (see detailed derivation in Methods): + +\[\frac{dQ_{L}^{\prime}}{d s S T^{\prime}} = \bar{\rho}_{a}\overline{A}_{V}\overline{C}_{e}[\overline{U}_{10}(\frac{d q_{s}^{\prime}}{d s S T^{\prime}} -\frac{d q_{a}^{\prime}}{d s S T^{\prime}}) + \frac{d U_{10}^{\prime}}{d s S T^{\prime}} (\overline{q}_{S} - \overline{q}_{a})] \quad (1)\] + +Here, the prime (') represents mesoscale anomalies defined by the high- pass spatial filtering, and the overbar (') denotes large- scale background excluding the mesoscale signal. + +The mesoscale SST- LHF coupling coefficient \((\frac{dQ_{L}^{\prime}}{d s S T^{\prime}})\) is determined by two components: the thermodynamic adjustment to mesoscale SST multiplied by the largescale wind + +<--- Page Split ---> + +\(\begin{array}{r l} & {(\overline{{U}}_{10}(\frac{d q_{s}^{\prime}}{d s S T^{\prime}} -\frac{d q_{a}^{\prime}}{d s S T^{\prime}}),} \end{array}\) , hereafter denoted as mesoscale thermodynamic adjustment term), and the dynamic adjustment to mesoscale SST multiplied by the large- scale humidity difference \(\begin{array}{r l} & {(\frac{d U_{10}^{\prime}}{d S S T^{\prime}} (\overline{{q}}_{S} - \overline{{q}}_{a}),} \end{array}\) , hereafter denoted as mesoscale dynamic adjustment term). Evaluation of historical and future simulations in the WBC regions reveals an increase in both terms due to greenhouse forcing (bars with black borders in Fig. 2a- d). Particularly, the enhancement of the thermodynamic adjustment term is about 3 to 5 times that of the dynamic adjustment term across all four WBCs (Fig. 2a- d), indicating the dominant contribution of the thermodynamic adjustment term to mesoscale SST- LHF coupling modulation. A further decomposition of the thermodynamic adjustment term reveals that the large- scale wind change between historical and future simulations is negligible (bars with magenta borders in Fig. 2a- d), while the predominant influence arises from the mesoscale moisture response to oceanic eddies, especially the specific humidity change at the ocean surface \((\frac{d q_{s}^{\prime}}{d S S T^{\prime}})\) is an order of magnitude greater than \(\frac{d q_{a}^{\prime}}{d S S T^{\prime}})\) . Collectively, the results suggest that the amplification in mesoscale moisture response is the principal driver for the strengthened mesoscale SST- LHF coupling under climate change. + +The above analyses indicate that changes in mesoscale SST- LHF coupling due to warming can be effectively estimated via the mesoscale moisture adjustment process. Nonetheless, this estimation still relies on the availability of both mesoscale and large- scale fields from historical and future simulations. To circumvent this, we apply a Taylor series expansion to the mesoscale moisture derivative (see detailed derivation in Methods). The resultant expression provides a simplified approach to assess mesoscale SST- LHF coupling change using large- scale fields: + +\[\left(\frac{d Q_{L}^{\prime}}{d S S T^{\prime}}\right)_{(F - P)} = (\overline{\rho}_{a}\overline{A}_{V}\overline{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}\left(\frac{d^{2}q_{s}(T)}{d T^{2}}\mid_{T = \overline{S S T}_{(P)}}\cdot \overline{\Delta S S T}\right) \quad (2)\] + +Here, ‘F’ represents future values and ‘P’ represents historical values in the past. It is evident + +<--- Page Split ---> + +that changes in mesoscale SST- LHF coupling are collectively affected by the large- scale wind and the curvature of the C- C scaling from the historical simulations, along with the projected warming of large- scale SST. Note that the curvature of the C- C scaling is inherently linked with the background SST, with higher temperature corresponding to a more pronounced moisture- temperature sensitivity. The relationship suggests that projections of future mesoscale SST- LHF coupling are significantly influenced by the historical large- scale oceanic and atmospheric conditions. Given that the large- scale wind and the curvature of the C- C scaling stay positive, it can be inferred that the direction of mesoscale SST- LHF coupling is exclusively determined by the sign of projected SST changes, leading to consistent intensification in line with the warming of underlying SST. + +We then assessed the regional variations in mesoscale SST- LHF coupling responses among the four WBC regions using Eq. (2), with emphasis on the north- south hemisphere asymmetry. The simplified relationship efficiently captures the more pronounced increase in coupling strength within the KE and GS regions compared to the ARC and BMC regions (Fig. 3a), consistent with the projected intensification of mesoscale SST- LHF coupling between future and historical simulations. Detailed examination of contributing factors in CESM- HR demonstrates consistently higher values for large- scale wind and SST in the KE and GS regions in the NH from historical simulations, alongside a more rapid SST warming rate (Fig. 3c), all of which jointly contribute to the enhanced augmentation in mesoscale coupling strength in the NH under climate change. The historical large- scale wind in the KE is approximately \(30\%\) stronger than that in the ARC and BMC, and the background SST in the GS is \(7^{\circ}\mathrm{C}\) higher than in the BMC, corroborated by observations (Fig. 3d). Additionally, the accelerated SST warming trend in the NH is also confirmed by the observational data (Fig. 3d). Further decomposition of the three factors indicates their respective contribution to the overall hemispheric discrepancy (Fig. 3b), with the large- scale SST warming rate being the dominant + +<--- Page Split ---> + +factor, followed by the background SST. + +## Validations in HighResMIP + +We extend the analyses to HighResMIP simulations (Methods) and to different warming periods (2030- 2050), aiming to verify the robustness of the findings. All models examined show an increase in mesoscale SST- LHF coupling in response to greenhouse warming across four WBCs by the year 2050 (Tab. 1). A more pronounced intensification of this coupling is projected in the NH compared to the SH, in line with CESM- HR results (Fig. 1e), albeit with a generally lower magnitude of change. The magnitude discrepancy is due to the HighResMIP projections here terminating in 2050, whereas CESM- HR projections shown in Fig 1 extend to a later period of the century (2080). + +The effectiveness of the proposed theoretical framework for estimating changes in mesoscale SST- LHF coupling due to greenhouse warming was also evaluated across different models. A comparison between the coupling strength changes estimated by the theoretical framework and those projected by CESM- HR reveals a significant linear relationship across four WBCs (Fig. S3). The estimated and actual projected changes in mesoscale SST- LHF coupling are highly correlated, with a correlation coefficient of around 0.7 in the KE and GS, and 0.5 in the ARC and BMC (Tab. 1). Similar linear correlations, ranging generally from 0.5 to 0.8 (Tab. 1), are found between estimated and projected mesoscale coupling strength changes in HighResMIP models, confirming the broad applicability of the simplified framework across diverse climate models. + +## Conclusions and Discussions + +How greenhouse warming will influence mesoscale air- sea interactions remains an open question. Utilizing eddy- resolving high- resolution CESM simulations, supported by observational and HighResMIP data, we found a ubiquitous intensification (approximately \(15\%\) ) of mesoscale SST- LHF coupling in WBC regions by the end of the \(21^{\text{st}}\) century under the + +<--- Page Split ---> + +RCP8.5 warming scenario. The intensification is characterized by an increased nonlinearity between warm and cold eddies and a more pronounced enhancement in the NH. Our findings suggest an increasingly important role of mesoscale oceanic processes in shaping weather and climate systems in a warming climate, especially in the NH. This underscores the critical need for climate models to incorporate mesoscale air- sea interaction for more reliable climate projections. + +We found that the mesoscale moisture response is the key factor driving the strengthened mesoscale SST- LHF coupling under climate change. To further understand the dynamics, we developed a theoretical framework to estimate the mesoscale moisture change. The framework builds a linkage between mesoscale coupling changes and large- scale fields, revealing that the projected mesoscale moisture changes can be estimated by the interplay among historical background wind, SST, and projected SST warming. The direction of mesoscale SST- LHF coupling changes is exclusively determined by the sign of projected SST changes. Given a scenario of SST warming, the mesoscale SST- LHF coupling will invariably exhibit intensification. The relationship highlights the importance of C- C scaling in determining changes in mesoscale SST- LHF coupling, suggesting that higher large- scale SST, determined either by historical baselines or future warming rates, are associated with greater rates of moisture increase and thereby enhanced augmentation of mesoscale SST- LHF coupling. + +The effectiveness of the proposed theoretical framework was evaluated across CESM- HR and HighResMIP models. The analysis reveals a robust linear correlation between estimated and projected changes in mesoscale SST- LHF, suggesting the broad applicability of the theoretical framework within major WBC regions. However, it is important to acknowledge that the prerequisite conditions for the theoretical framework to work are the mesoscale moisture adjustment significantly surpasses the mesoscale wind adjustment, the mesoscale moisture adjustment at the ocean surface significantly exceeds that at the atmospheric surface, + +<--- Page Split ---> + +and the large- scale wind changes due to warming is minimal. These conditions may not hold outside the WBCs, potentially undermining the framework's applicability. + +## Methods + +## CESM-HR, HighResMIP Simulations and Observations + +We utilized the high- resolution Community Earth System Model (CESM- HR) simulations with \(\sim 0.25^{\circ}\) atmosphere and \(\sim 0.1^{\circ}\) ocean components developed by the National Center for Atmosphere Research (NCAR) (Chang et al., 2020). The simulations include a 500- year preindustrial control simulation and a 250- year historical and future simulation from 1850- 2100. Historical radiative forcing is applied from 1850 to 2005 while the Representative Concentration Pathway 8.5 (RCP8.5, a high greenhouse gas emission) warming forcing is switched from 2006 onwards. Two 20- year periods with high- frequency (daily) output were chosen to assess the mesoscale thermal coupling response to greenhouse warming in CESM- HR: a historical period from 1956 to 1975 (referred to as HIS) and a future period from 2063 to 2082 (referred to as RCP). + +Five HighResMIP simulations with relatively high oceanic resolution were selected: CMCC- CM2- VHR4 ( \(0.25^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), HadGEM3- GC31- HH ( \(0.5^{\circ}\) atmosphere and \(0.08^{\circ}\) ocean), EC- Earth3P- HR ( \(0.5^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), CNRM- CM6- 1- HR ( \(1^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), MPI- ESM1- 2- XR ( \(0.5^{\circ}\) atmosphere and \(0.5^{\circ}\) ocean). We note that only one model (HadGEM3- GC31- HH) has comparable oceanic resolutions with CESM- HR, yet its atmospheric resolution is coarser. All these selected models include a 100- year historical and future (under RCP8.5 scenario) simulation from 1950- 2050. For a consistent comparison between CESM- HR and HighResMIP simulations, 1950- 1969 for historical and 2031- 2050 for future projections were selected for the corroborative analysis shown in Tab. 1. + +<--- Page Split ---> + +Daily sea surface height (SSH) derived from Copernicus Marine Environment Monitoring Service (CMEMS) during 2003- 2007 was used to identify eddies in observations (Tab. S1). Concurrently, SST obtained from the National Oceanic and Atmospheric Administration daily Optimum Interpolation SST (NOAA- OISST) and heat fluxes derived from Japanese Ocean Flux Data Sets with Use of Remote Sensing Observations version3 (J- OFURO3) during the same period were employed to construct observational eddy composites (Fig. S1). ERA5 reanalysis (the fifth generation European Centre for Medium- Range Weather Forecasts atmospheric reanalysis, Hersbach et al., 2020) with an extended temporal span from 1979 to 2022 was utilized to examine the decadal trend of mesoscale coupling strength (Fig. 1a). + +## Eddy detection, Eddy Composites and High-pass Filtering + +In both CMEMS observations and CESM- HR simulations, mesoscale eddies were detected using daily SSH anomalies derived by applying high- pass spatial filtering ( \(20^{\circ}\) longitude \(\times 10^{\circ}\) latitude) to remove the large- scale signal, following previous studies (Faghmous et al., 2013; Qu et al., 2022). Cyclonic (anticyclonic) eddies are classified by closed contours of SSH anomalies that include a single minimum (maximum), with an SSH anomaly increment (decrement) of \(0.05 \mathrm{cm}\) between successive contours. The edge of an eddy is delineated by the outmost closed contour of SSH anomalies. The radius of an eddy corresponds to the radius of a circle with an equivalent area to that enclosed by the outmost contour. The amplitude of an eddy is defined by the SSH anomaly difference between the eddy's peak and its defined edge. Only eddies with a radius ranging from 70 to \(200 \mathrm{km}\) and an amplitude exceeding \(3 \mathrm{cm}\) are included in the analysis. + +Eddy composites were constructed by aligning associated variables to the reference coordinate of the eddy core. The variables were normalized by the individual eddy radius and oriented to the prevailing direction of the large- scale background surface wind, in line with + +<--- Page Split ---> + +previous research (Frenger et al., 2013). Variables within twice the eddy radius were included for composite analysis. + +In addition to eddy composite analysis, we also applied a high- pass Loess Filter with a cutoff wavelength of \(30^{\circ}\) longitude \(\times 10^{\circ}\) latitude (similar to a \(5^{\circ}\times 5^{\circ}\) box car average, Bryan et al. 2010, Ma et al. 2016) in ERA5, CESM- HR and HighResMIP, to isolate mesoscale SST and LHF and examined the spatial distribution of their coupling strength (Fig. 1a,b and Fig S3). The coupling strength at each grid point was computed using the linear regression coefficient between high- pass filtered monthly SST and LHF over a spatial domain of \(4^{\circ}\times 4^{\circ}\) . It was noted that applying a high- pass filter directly to monthly data yields a coupling coefficient comparable to that obtained when the filter is first applied to daily data, which is then aggregated into a monthly mean before calculating the coefficient. The former method was selected for its computational efficiency. To highlight regions with pronounced mesoscale SST- LHF coupling, mesoscale signals where the SST anomaly fell below \(0.4^{\circ}\mathrm{C}\) in ERA5 and below \(0.6^{\circ}\mathrm{C}\) in CESM- HR were excluded when computing the decadal trend in coupling strength (Fig. 1a, b). + +## Decomposition of mesoscale SST-LHF coupling + +According to Bulk formula (Large and Yeager 2004), the latent heat flux \(Q_{L}\) is determined by the equation: + +\[Q_{L} = \rho_{a}\Lambda_{v}C_{e}U_{10}(q_{s} - q_{a}) \quad (1)\] + +Here, \(\rho_{a}\) is the surface air density, \(\Lambda_{v}\) is the latent heat of vaporization, and \(C_{e}\) is the transfer coefficients for evaporation. \(U_{10}\) is the 10m wind speed, \(q_{s}\) is the saturated specific humidity at the ocean surface, and \(q_{a}\) is air specific humidity at 2m. + +Following previous studies (Yang et al., 2018; Yuan et al, 2023), equation (1) can be decomposed into large- scale background and mesoscale components. The mesoscale component of latent heat flux is estimated as follows: + +<--- Page Split ---> + +\[Q_{L}^{t} = \bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}[\overline{U}_{10}(q_{s}^{t} - q_{a}^{t}) + U_{10}^{t}(\overline{q}_{s} - \overline{q}_{a})] \quad (2)\] + +Here, the prime (') represents mesoscale anomalies defined by the high- pass spatial filtering, and the overbar (') denotes large- scale background excluding the mesoscale signal. + +By differentiating with respect to SST, the mesoscale SST- LHF coupling is expressed as: + +\[\frac{dq_{L}^{t}}{dSST^{t}} = \bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}[\overline{U}_{10}(\frac{dq_{s}^{t}}{dSST^{t}} -\frac{dq_{a}^{t}}{dSST^{t}}) + \frac{dU_{10}^{t}}{dSST^{t}} (\overline{q}_{S} - \overline{q}_{a})] \quad (3)\] + +Based on calculations, the second term \(\frac{dU_{10}^{t}}{dSST^{t}} (\overline{q}_{S} - \overline{q}_{a})\) on the right- hand side of Eq. (3) is an order of magnitude smaller than the first term \(\overline{U}_{10}(\frac{dq_{s}^{t}}{dSST^{t}} -\frac{dq_{a}^{t}}{dSST^{t}})\) . Furthermore, \(\frac{dq_{a}^{t}}{dSST^{t}}\) is an order of magnitude smaller than \(\frac{dq_{s}^{t}}{dSST^{t}}\) . Disregarding the relatively smaller terms and assuming the changes in \(\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}\) is minimal under global warming, the mesoscale SST- LHF coupling changes is predominately influenced by \(\overline{U}_{10}\frac{dq_{s}^{t}}{dSST^{t}}\) , which can be represented as: + +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}[\overline{U}_{10(F)}\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \overline{U}_{10(P)}\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})}] \quad (4)\] + +Where 'F' represents future values and 'P' represents historical values in the past. As the large- scale wind, \(\overline{U}_{10(F)}\) and \(\overline{U}_{10(C)}\) , experience minimal changes in the WBCs, Eq. (4) can be further simplified as: + +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}[\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})}] \quad (5)\] + +Applying a Taylor expansion, the right- hand side term can be approximated as: + +\[\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})} = \frac{d^{2}q_{s}(T)}{dT^{2}} |_{T = \overline{SST}_{(P)}}\cdot \Delta SST \quad (6)\] + +Substituting (6) into (5) yields: + +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}\frac{d^{2}q_{s}(T)}{dT^{2}} |_{T = \overline{SST}_{(P)}}\cdot \Delta SST \quad (7)\] + +Where \(\frac{d^{2}q_{s}(T)}{dT^{2}}\) is determined by C- C scaling and exhibits an exponential increase with + +<--- Page Split ---> + +322 temperature. Note that the composite coefficient \((\bar{\rho}_{a}\bar{A}_{v}\bar{C}_{e})\) is retained to align the estimated 323 coupling strength changes with magnitude analogous to the actual projections. + +<--- Page Split ---> + +## Data Availability + +Data AvailabilityCMEMS data can be obtained through https://data.marine.copernicus.eu/products. NOAA- OISST can be accessed through https://www.ncei.noaa.gov/products/optimum- interpolation- sst. J- OFURO3 can be achieved through https://www.j- ofuro.com/en/. ERA5 reanalysis can be downloaded from https://doi.org/10.24381/cds.bd0915c6. The CESM simulations can be achieved through https://ihep.github.io/archive/products/ds_archive/Sunway_Runs.html. The HighResMIP data can be downloaded from https://pcmdi.llnl.gov/CMIP6/. + +## Code Availability + +Code AvailabilityMATLAB codes to reproduce the analyses are available upon request from the corresponding author or can be accessed through the link https://zenodo.org/records/10610386. + +## Acknowledgments + +AcknowledgmentsThis research is supported by the National Natural Science Foundation of China (42376025), Science and Technology Innovation Program of Laoshan Laboratory (LSKJ202300302, LSKJ202202503), Shandong Provincial Natural Science Foundation (ZR2022YQ29), Taishan Scholar Funds (tsqn202103028). We thank Sunway TaihuLight High- Performance Computer (Wuxi), Laoshan Laboratory in Qingdao and the National Supercomputing center in Jinan for providing the high resolution CESM simulations and high- performance computing resources that contributed to the research results reported in this paper. + +## Author Contributions + +Author ContributionsX. M. and X. Z conceived the study. X. M. instructed the investigation and wrote the manuscript. X. Z. performed the analyses and produced all figures. L. W. supervised the project. P. Y., F. S. and M. Y. contributed to the discussion of mesoscale thermal coupling decomposition. Y. Q and H. C. offered insights into eddy detection. Z. J., Z. C. and B. G. contributed to interpreting the results and improving the manuscript. + +## Competing Interests statement + +<--- Page Split ---> + +The authors declare no competing interests. + +## References + +Beech, Nathan, et al. Long- term evolution of ocean eddy activity in a warming world. Nature climate change 12.10 (2022): 910- 917 + +Bryan, Frank O., et al. Frontal scale air- sea interaction in high- resolution coupled climate models. Journal of Climate 23.23 (2010): 6277- 6291. + +Chang, Ping, et al. An unprecedented set of high- resolution earth system simulations for understanding multiscale interactions in climate variability and change. Journal of Advances in Modeling Earth Systems 12.12 (2020): e2020MS002298. + +Chelton, Dudley B., et al. Satellite measurements reveal persistent small- scale features in ocean winds. science 303.5660 (2004): 978- 983. + +Chen, Longjing, Yinglai Jia, and Qinyu Liu. Oceanic eddy- driven atmospheric secondary circulation in the winter Kuroshio Extension region. Journal of Oceanography 73 (2017): 295- 307. + +Czaja, Arnaud, et al. Simulating the midlatitude atmospheric circulation: what might we gain from high- resolution modeling of air- sea interactions?. Current Climate Change reports 5 (2019): 390- 406. + +Faghmous, James H., et al. A parameter- free spatio- temporal pattern mining model to catalog global ocean dynamics. 2013 IEEE 13th International conference on data mining. IEEE, 2013. + +Frenger, Ivy, et al. Imprint of Southern Ocean eddies on winds, clouds and rainfall. Nature geoscience 6.8 (2013): 608- 612. + +Foussard, A., G. Lapeyre, and R. Plougonven. Storm track response to oceanic eddies in idealized atmospheric simulations. Journal of Climate 32.2 (2019): 445- 463. + +<--- Page Split ---> + +Gan, Bolan, et al. North Atlantic subtropical mode water formation controlled by Gulf Stream fronts. National Science Review (2023): nwad133. + +Hersbach, Hans, et al. The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society 146.730 (2020): 1999- 2049. + +Jiang, Yuxi, et al. Effects of a cold ocean eddy on local atmospheric boundary layer near the Kuroshio Extension: In situ observations and model experiments. Journal of Geophysical Research: Atmospheres 124.11 (2019): 5779- 5790. + +Jing, Zhao, et al. Maintenance of mid- latitude oceanic fronts by mesoscale eddies. Science advances 6.31 (2020): eaba7880. + +Large, William George, and Stephen G. Yeager. Diurnal to decadal global forcing for ocean and sea- ice models: The data sets and flux climatologies. (2004). + +Liu, Xue, et al. Ocean fronts and eddies force atmospheric rivers and heavy precipitation in western North America. Nature communications 12.1 (2021): 1268. + +Ma, Xiaohui, et al. Distant Influence of Kuroshio Eddies on North Pacific Weather Patterns?. Sci Rep 5, 17785 (2015). https://doi.org/10.1038/srep17785 + +Ma, Xiaohui, et al. Western boundary currents regulated by interaction between ocean eddies and the atmosphere. Nature 535.7613 (2016): 533- 537. + +Ma, Xiaohui, et al. Importance of resolving Kuroshio front and eddy influence in simulating the North Pacific storm track. Journal of Climate 30.5 (2017): 1861- 1880. + +Martínez- Moreno, Josué, et al. Global changes in oceanic mesoscale currents over the satellite altimetry record. Nature Climate Change 11.5 (2021): 397- 403. + +<--- Page Split ---> + +Parfitt, R., Czaja, A., Minobe, S., & Kuwano- Yoshida, A. (2016). The atmospheric frontal response to SST perturbations in the Gulf Stream region. Geophysical Research Letters, 43(5), 2299- 2306. + +Qu, Yushan, et al. Spatial Structure of Vertical Motions and Associated Heat Flux Induced by Mesoscale Eddies in the Upper Kuroshio- Oyashio Extension. Journal of Geophysical Research: Oceans 127.10 (2022): e2022JC018781. + +Renault, Lionel, et al. Disentangling the mesoscale ocean- atmosphere interactions. Journal of Geophysical Research: Oceans 124.3 (2019): 2164- 2178. + +Seo, Hyodae, et al. "Ocean Mesoscale and Frontal- Scale Ocean- Atmosphere Interactions and Influence on Large- Scale Climate: A Review." Journal of Climate 36.7 (2023): 1981- 2013. + +Sheldon, L., Czaja, A., Vanniere, B., Morcrette, C., Sohet, B., Casado, M., & Smith, D. (2017). A 'warm path' for Gulf Stream- troposphere interactions. Tellus A: Dynamic Meteorology and Oceanography, 69(1), 1299397. + +Skyllingstad, Eric D., et al. Effects of mesoscale sea- surface temperature fronts on the marine atmospheric boundary layer. Boundary- layer meteorology 123 (2007): 219- 237. + +Small, RJ D., et al. Air- sea interaction over ocean fronts and eddies. Dynamics of Atmospheres and Oceans 45.3- 4 (2008): 274- 319. + +Souza, J. M. A. C., Bertrand Chapron, and Emmanuelle Autret. The surface thermal signature and air- sea coupling over the Agulhas rings propagating in the South Atlantic Ocean interior. Ocean Science 10.4 (2014): 633- 644. + +Xie, Shang- Ping. Satellite observations of cool ocean- atmosphere interaction. Bulletin of the American Meteorological Society 85.2 (2004): 195- 208. + +Yang, Peiran, Zhao Jing, and Lixin Wu. An assessment of representation of oceanic mesoscale eddy- atmosphere interaction in the current generation of general circulation models and reanalyses. Geophysical Research Letters 45.21 (2018): 11- 856. + +<--- Page Split ---> + +Yuan, Man, et al. Spatio- temporal variability of surface turbulent heat flux feedback for mesoscale sea surface temperature anomaly in the global ocean. Frontiers in Marine Science 9 (2022): 957796. Zhang, Xingzhi, Xiaohui Ma, and Lixin Wu. Effect of mesoscale oceanic eddies on extratropical cyclogenesis: A tracking approach. Journal of Geophysical Research: Atmospheres 124.12 (2019): 6411- 6422. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Observed and simulated trends in mesoscale SST-LHF coupling. Global distribution of the decadal trends of mesoscale SST-LHF coupling strength as derived from ERA5 (a) and CESM-HR (b) during 1979-2022. The coupling strength is computed using the linear regression coefficient between high-pass filtered monthly SST and LHF (Methods) with trends significant at a 99% confidence level indicated by black dots. (c) Composites of SST (color shading) and LHF (contours) anomalies associated with mesoscale oceanic eddies during historical periods (1956-1975) in CESM-HR. Shown are winter season mean anomalies in warm minus cold eddy composites within four WBC regions (outlined by red boxes in a, b). The white circle represents one eddy radius and the white dot marks the eddy center. (d) Mesoscale SST-LHF coupling strength during historical (1956-1975) and future (2063-2082) periods in CESM-HR for warm (red) and cold (blue) eddies averaged across four WBC regions. The coupling strength is computed using the linear regression coefficient between SST and LHF anomalies within twice the radius of eddy composites. (e) Differences in mesoscale SST-LHF coupling strength between future and historical periods in CESM-HR for warm (red bars) and cold (blue bars) eddies in the KE, GS, ARC and BMC regions, with fractional differences labeled atop the respective bars.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 Decomposition of mesoscale SST-LHF coupling response under greenhouse warming. (a) Differences in thermodynamic and dynamic adjustments between future and historical periods in CESM-HR in the KE region. From left to right, the terms plotted are thermodynamic adjustment \(\overline{U}_{10}(\frac{aq_s'}{dsST'}, \frac{aq_a'}{dsST'})\) , dynamic adjustment \(\frac{au_{10}'}{dsST'}(\overline{q}_s - \overline{q}_a)\) , large-scale surface wind \((\overline{U}_{10})\) and moisture adjustment \((\frac{aq_s'}{dsST'}, \frac{aq_a'}{dsST'})\) . (b-d), as for a, but for the GS, ARC and BMC regions, respectively.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Decomposition of mesoscale moisture response under greenhouse warming. (a) Estimated changes in mesoscale SST-LHF coupling between future and historical periods in CESM-HR according to Eq. 2 in the KE GS, ARC and BMC regions. The large-scale SST averaged in the WBCs are displayed on the bottom axis for each respective region. Black lines denote the standard deviations of the corresponding variables. (b) Contributions of \(\overline{U}_{10}, \frac{d^2 q_s(T)}{dT^2}\) and \(\Delta SST\) to the north-south hemispheric asymmetry in the KE GS, ARC and BMC regions. The total deviation from the hemispheric mean due to three terms in a specific region is approximated by \((ABC)' \approx A' \bar{B} \bar{C} + B' \bar{A} \bar{C} + C' \bar{A} \bar{B}\) . Here, A, B, and C denotes the three terms; the overbar denotes a mean component (the average across the four WBCs); the prime denotes an anomaly component (the deviation from the hemispheric mean for each specific WBC region). (c) The historical large-scale surface wind, SST and the projected SST warming within the four WBC regions in CESM-HR. (d) as for c, but for observational data. Surface wind is derived from ERA5 and SST is derived from NOAA-OISST. \(\Delta SST\) is computed based on the warming trend observed over the last 40 years.
+ +<--- Page Split ---> + + +Table. 1 Projected and estimated mesoscale SST-LHF coupling changes in CESM-HR and HighResMIP models. + +
Projected coupling changes (RCP-HIS) (W·m-2·°C-1)Correlation between projected and estimated changes
KEGSARCBMCKEGSARCBMC
CESM-HR3.2 (10.26%)3.5 (15.34%)1.9 (6.68%)1.5 (6.72%)0.710.690.450.49
CMCC-CM2-VHR44.1 (14.27%)2.5 (11.97%)1.5 (6.65%)0.07 (0.35%)0.760.650.600.54
HadGEM3-GC31-HH3.2 (11.95%)2.0 (8.76%)2.8 (11.34%)2.4 (11.88%)0.780.660.580.61
MPI-ESM1-2-XR2.9 (14.06%)2.0 (10.14%)1.6 (9%)0.9 (5.24%)0.410.690.720.78
EC-Earth3P-HR2.0 (8.37%)2.4 (11.24%)1.5 (6.41%)0.5 (2.72%)0.720.760.690.66
CNRM-CM6-1-HR2.2 (10.1%)2.4 (12.4%)1.2 (7.52%)0.9 (3.58%)0.690.70.670.58
+ +506 Note that mesoscale SST-LHF coupling strength changes analyzed here correspond to 507 differences between the historical period of 1950-1969 and the mid-21st century projections for 508 the period of 2031-2050. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d_det.mmd b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..ce286c1455befdae822469e037b6cb30eb578c08 --- /dev/null +++ b/preprint/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d/preprint__0e6380fd5cbe3c0c7e8a39f2b23083f37592c615a5adfd5421239921c07e585d_det.mmd @@ -0,0 +1,410 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 923, 177]]<|/det|> +# Midlatitude Mesoscale Thermal Air-sea Interaction Enhanced by Greenhouse Warming + +<|ref|>text<|/ref|><|det|>[[44, 196, 312, 242]]<|/det|> +Xingzhi Zhang zhangxingzhi@ouc.edu.cn + +<|ref|>text<|/ref|><|det|>[[44, 270, 641, 760]]<|/det|> +Ocean Univeristy of China Xiaohui Ma Ocean University of China https://orcid.org/0000- 0001- 9937- 3859 Lixin Wu Ocean University of China https://orcid.org/0000- 0002- 4694- 5531 Peiran Yang Laoshan Laboratory Fengfei Song Ocean University of China https://orcid.org/0000- 0002- 3004- 1749 Zhao Jing Ocean University of China https://orcid.org/0000- 0002- 8430- 9149 Hui Chen Ocean University of China Yushan Qu Ocean University of China Man Yuan Ocean University of China Zhaohui Chen Ocean University of China https://orcid.org/0000- 0002- 0830- 2332 Bolan Gan Ocean University of China https://orcid.org/0000- 0001- 7620- 485X + +<|ref|>sub_title<|/ref|><|det|>[[44, 794, 103, 811]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 831, 137, 850]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 870, 336, 888]]<|/det|> +Posted Date: February 19th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 907, 473, 925]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3932615/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 914, 88]]<|/det|> +License: © ① This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 106, 535, 126]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 161, 920, 204]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 4th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52077- z. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 881, 152]]<|/det|> +Midlatitude Mesoscale Thermal Air-sea Interaction Enhanced by Greenhouse Warming Xiaohui Ma1,2, Xingzhi Zhang1,*, Lixin Wu1,2, Peiran Yang2, Fengfei Song1,2, Zhao Jing1,2, Hui Chen1, Yushan Qu1, Man Yuan1, Zhaohui Chen1,2, Bolan Gan1,2 + +<|ref|>text<|/ref|><|det|>[[150, 195, 880, 240]]<|/det|> +1. Frontiers Science Center for Deep Ocean Multispheres and Earth System and Key Laboratory of Physical Oceanography, Ocean University of China, Qingdao, China. + +<|ref|>text<|/ref|><|det|>[[150, 251, 437, 266]]<|/det|> +2. Laoshan Laboratory, Qingdao, China. + +<|ref|>text<|/ref|><|det|>[[150, 279, 546, 294]]<|/det|> +\*Corresponding author. E-mail: zhangxingzhi@ouc.edu.cn + +<|ref|>sub_title<|/ref|><|det|>[[118, 332, 198, 348]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[113, 360, 884, 905]]<|/det|> +The influence of greenhouse warming on mesoscale air- sea interactions, crucial for modulating ocean circulation and climate variability, remains largely unexplored due to the limited resolution of current climate models. This study addresses this gap by analyzing eddy- resolving high- resolution climate simulations and observations, focusing on the coupling between mesoscale sea surface temperature (SST) and latent heat flux (LHF) in winter. Our findings reveal a consistent increase in mesoscale SST- LHF coupling in the major western boundary current regions under warming, characterized by a heightened nonlinearity between warm and cold eddies and a more pronounced enhancement in the northern hemisphere. To understand the dynamics, we develop a theoretical framework that links mesoscale thermal coupling changes to large- scale factors, which indicates that the projected changes are collectively determined by historical background wind, SST, and the rate of SST warming. Among these factors, the large- scale SST and its warming rate are the primary drivers of hemispheric asymmetry in mesoscale coupling intensification. This study introduces a simplified approach for assessing the projected mesoscale thermal coupling changes and implies a growing significance of mesoscale air- sea interaction in shaping weather and climate patterns in a warming world. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 85, 885, 565]]<|/det|> +Mesoscale air- sea interactions, predominantly active in the Western Boundary Current (WBC) regions in the midlatitudes, play a critical role in modulating extratropical weather and climate systems (Chelton et al., 2004; Small et al., 2008; Bryan et al., 2010; Czaja et al., 2019; Seo et al., 2023). The interaction between mesoscale oceanic eddies and the atmosphere actively impacts precipitation, storms, large- scale atmospheric circulations, and provides feedback to the ocean, influencing oceanic circulations and climate variability (Xie, 2004; Skyllingstad et al., 2007; Chelton and Xie, 2010; Frenger et al., 2013; Souza et al., 2014; Ma et al., 2016, 2017; Foussard et al., 2019; Renault et al., 2019; Gan et al., 2023). A key aspect of this interaction is the coupling between sea surface temperature and turbulent heat flux (SST- THF), hereafter denoted as thermal coupling. In terms of the atmosphere, thermal coupling acts as an energy source, which is crucial for the genesis and development of weather systems and deep troposphere response (Parfitt et al 2016; Chen et al., 2017; Sheldon et al., 2017; Jiang et al., 2019; Zhang et al., 2019; Liu et al., 2021). In terms of the ocean, thermal coupling serves as an energy sink, which is the key to dissipating oceanic eddy energy and driving oceanic circulation response (Ma et al., 2016; Jing et al., 2020). + +<|ref|>text<|/ref|><|det|>[[115, 577, 885, 891]]<|/det|> +How greenhouse warming will impact mesoscale oceanic eddies remains uncertain, let alone mesoscale air- sea coupling. Satellite observations reveal a rise in eddy activity in the WBC regions during recent decades, while climate models project heterogeneous eddy variations in different WBC regions under warming (Martinez- Moreno et al., 2021; Beech et al., 2022). Additionally, the theoretical framework for predicting the mesoscale thermal coupling change in response to greenhouse warming is currently absent. Thermodynamically, the SST warming and the associated water vapor increase governed by the Clausius- Clapeyron (C- C) relation, appear to strengthen the thermal air- sea coupling (primarily through the enhancement of latent heat flux) in a warming climate. Dynamically, the non- uniform warming between the upper and lower troposphere under anthropogenic forcing tends to increase + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 884, 333]]<|/det|> +atmospheric stability, inhibiting the vertical momentum transfer and thereby suppressing the surface wind and thermal air- sea coupling. This is further complicated by significant uncertainties in atmospheric circulation responses under climate change, introducing additional perturbations to SST- THF coupling. It is noteworthy that the aforementioned oceanic and atmospheric changes discussed within a conventional large- scale framework may not necessarily manifest at mesoscales. Physical processes governing the response of mesoscale thermal coupling to greenhouse warming are multifaceted, making it challenging to pinpoint the ultimate dominant factor. + +<|ref|>text<|/ref|><|det|>[[115, 347, 885, 592]]<|/det|> +Utilizing an unprecedented set of high- resolution Community Earth System Model (referred to as CESM- HR, Methods) with \(\sim 0.25^{\circ}\) atmosphere and \(\sim 0.1^{\circ}\) ocean components that can explicitly resolve the mesoscale oceanic eddies and their coupling with the atmosphere, we investigated the impact of greenhouse warming on mesoscale thermal coupling in eddy- rich WBC regions. We then constructed a theoretical framework for estimating the mesoscale thermal coupling change in response to greenhouse warming through the decomposition of contributing factors. We further assessed the robustness of the findings by extending the analyses to High- Resolution Model Intercomparison Project (HighResMIP, Methods) models. + +<|ref|>sub_title<|/ref|><|det|>[[120, 608, 640, 627]]<|/det|> +## Mesoscale thermal coupling during the observational period + +<|ref|>text<|/ref|><|det|>[[115, 641, 884, 890]]<|/det|> +We first evaluated the model's capability in representing the mesoscale thermal coupling in the CESM- HR simulations against observational data (Methods), in four major WBC regions, i.e., the Kuroshio Extension (KE), the Gulf Stream Extension (GS), the Agulhas Return Current (ARC) and the Brazil- Malvinas Confluence Region (BMC), by detecting eddies using sea surface height anomalies and constructing composite analyses with a reference frame centered on the eddy (see details in Methods). The model shows high fidelity in representing the statistical characteristics of eddies, such as the averaged number, amplitude and size, as detailed in Tab. S1. It also successfully reproduces the eddy- induced turbulent heat fluxes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 884, 300]]<|/det|> +(including both sensible and latent components) along with their seasonality, for both anticyclonic warm and cyclonic cold eddies (Fig. S1). It is noted that the coupling strength peaks during the winter month. Furthermore, the eddy- induced sensible heat flux (SHF) is approximately half that of latent heat flux (LHF) and remains relatively stable between historical and future simulations (Fig. S2). Consequently, our subsequent analyses will focus on the mesoscale SST- LHF coupling in the hemispheric winter season — DJF for the Northern Hemisphere (NH) and JJA for the Southern Hemisphere (SH). + +<|ref|>text<|/ref|><|det|>[[115, 314, 884, 660]]<|/det|> +To assess the potential impact of anthropogenic warming on mesoscale thermal coupling, we analyzed the decadal trend of mesoscale SST- LHF coupling in the observational periods based on high- pass spatial filtering fields (Methods). A significant intensification of mesoscale SST- LHF coupling at an approximate rate of \(1.5 \mathrm{W / m^2 / ^\circ C}\) per decade is detected in the WBCs over the past four decades, with the most pronounced increase observed in the GS and KE regions of the NH (Fig. 1a). The CESM- HR successfully simulates the enhanced mesoscale thermal coupling in the WBCs with a magnitude slightly higher than that recorded in the observations (Fig. 1b). The model also captures the north- south hemispheric asymmetry of the changes. The alignment between model simulated and observational trends in the past lends credence to the use of CESM- HR for investigating the mesoscale coupling response under climate change. + +<|ref|>sub_title<|/ref|><|det|>[[118, 675, 605, 694]]<|/det|> +## Mesoscale thermal coupling under greenhouse warming + +<|ref|>text<|/ref|><|det|>[[115, 707, 884, 890]]<|/det|> +A linear regression between mesoscale SST and LHF based on eddy composites averaged across four WBC regions, yields an estimated coupling coefficient of approximately \(28 \mathrm{W / m^2 / ^\circ C}\) for both anticyclonic warm and cyclonic cold eddies in historical simulations (1956- 1975, Fig. 1c,d). The nonlinearity between warm and cold eddies appears to be minimal during the historical period. Under the high greenhouse forcing scenario of Representative Concentration Pathway 8.5, future simulations (2063- 2082) project an approximate \(13\%\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 884, 465]]<|/det|> +enhancement in mesoscale SST- LHF coupling to \(32\mathrm{W / m^2 / ^\circ C}\) (Fig. 1d). A detailed examination of the four WBCs reveals a consistent increase in coupling strength by \(6\%\) to \(20\%\) in future simulations compared to historical simulations (Fig. 1e). Importantly, future simulations project a greater increase of mesoscale SST- LHF coupling strength for warm eddies than for cold eddies, indicating heightened nonlinearity in a warming climate. The increase in nonlinearity is expected to amplify net heat flux from mesoscale oceanic eddies (Ma et al. 2015; Foussard et al., 2019), thereby impacting future weather and climate systems more significantly. Additionally, the projected increase in mesoscale SST- LHF coupling in the GS and KE is twice that of the ARC and BMC, pointing to a more substantial intensification in the NH than in the SH. Combined with the findings in the observational periods, the results indicate that the hemisphere discrepancy in mesoscale SST- LHF coupling is likely to not only persist but also to become more pronounced with ongoing climate warming. + +<|ref|>sub_title<|/ref|><|det|>[[120, 478, 450, 496]]<|/det|> +## Decomposition of contributing factors + +<|ref|>text<|/ref|><|det|>[[115, 510, 884, 693]]<|/det|> +The LHF is proportional to both the surface wind and the humidity difference at the air- sea interface according to the Bulk formula (Large and Yeager 2004). To estimate the mesoscale SST- LHF coupling associated with oceanic eddies, we decompose the relevant fields into large- scale background and mesoscale components in line with previous studies (Yang et al., 2018; Yuan et al, 2023). The mesoscale SST- LHF coupling coefficient can be quantified using the following relationship (see detailed derivation in Methods): + +<|ref|>equation<|/ref|><|det|>[[244, 705, 813, 737]]<|/det|> +\[\frac{dQ_{L}^{\prime}}{d s S T^{\prime}} = \bar{\rho}_{a}\overline{A}_{V}\overline{C}_{e}[\overline{U}_{10}(\frac{d q_{s}^{\prime}}{d s S T^{\prime}} -\frac{d q_{a}^{\prime}}{d s S T^{\prime}}) + \frac{d U_{10}^{\prime}}{d s S T^{\prime}} (\overline{q}_{S} - \overline{q}_{a})] \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[115, 750, 881, 802]]<|/det|> +Here, the prime (') represents mesoscale anomalies defined by the high- pass spatial filtering, and the overbar (') denotes large- scale background excluding the mesoscale signal. + +<|ref|>text<|/ref|><|det|>[[115, 816, 881, 880]]<|/det|> +The mesoscale SST- LHF coupling coefficient \((\frac{dQ_{L}^{\prime}}{d s S T^{\prime}})\) is determined by two components: the thermodynamic adjustment to mesoscale SST multiplied by the largescale wind + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 85, 884, 606]]<|/det|> +\(\begin{array}{r l} & {(\overline{{U}}_{10}(\frac{d q_{s}^{\prime}}{d s S T^{\prime}} -\frac{d q_{a}^{\prime}}{d s S T^{\prime}}),} \end{array}\) , hereafter denoted as mesoscale thermodynamic adjustment term), and the dynamic adjustment to mesoscale SST multiplied by the large- scale humidity difference \(\begin{array}{r l} & {(\frac{d U_{10}^{\prime}}{d S S T^{\prime}} (\overline{{q}}_{S} - \overline{{q}}_{a}),} \end{array}\) , hereafter denoted as mesoscale dynamic adjustment term). Evaluation of historical and future simulations in the WBC regions reveals an increase in both terms due to greenhouse forcing (bars with black borders in Fig. 2a- d). Particularly, the enhancement of the thermodynamic adjustment term is about 3 to 5 times that of the dynamic adjustment term across all four WBCs (Fig. 2a- d), indicating the dominant contribution of the thermodynamic adjustment term to mesoscale SST- LHF coupling modulation. A further decomposition of the thermodynamic adjustment term reveals that the large- scale wind change between historical and future simulations is negligible (bars with magenta borders in Fig. 2a- d), while the predominant influence arises from the mesoscale moisture response to oceanic eddies, especially the specific humidity change at the ocean surface \((\frac{d q_{s}^{\prime}}{d S S T^{\prime}})\) is an order of magnitude greater than \(\frac{d q_{a}^{\prime}}{d S S T^{\prime}})\) . Collectively, the results suggest that the amplification in mesoscale moisture response is the principal driver for the strengthened mesoscale SST- LHF coupling under climate change. + +<|ref|>text<|/ref|><|det|>[[117, 617, 884, 833]]<|/det|> +The above analyses indicate that changes in mesoscale SST- LHF coupling due to warming can be effectively estimated via the mesoscale moisture adjustment process. Nonetheless, this estimation still relies on the availability of both mesoscale and large- scale fields from historical and future simulations. To circumvent this, we apply a Taylor series expansion to the mesoscale moisture derivative (see detailed derivation in Methods). The resultant expression provides a simplified approach to assess mesoscale SST- LHF coupling change using large- scale fields: + +<|ref|>equation<|/ref|><|det|>[[211, 844, 785, 881]]<|/det|> +\[\left(\frac{d Q_{L}^{\prime}}{d S S T^{\prime}}\right)_{(F - P)} = (\overline{\rho}_{a}\overline{A}_{V}\overline{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}\left(\frac{d^{2}q_{s}(T)}{d T^{2}}\mid_{T = \overline{S S T}_{(P)}}\cdot \overline{\Delta S S T}\right) \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[115, 888, 884, 907]]<|/det|> +Here, ‘F’ represents future values and ‘P’ represents historical values in the past. It is evident + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 884, 400]]<|/det|> +that changes in mesoscale SST- LHF coupling are collectively affected by the large- scale wind and the curvature of the C- C scaling from the historical simulations, along with the projected warming of large- scale SST. Note that the curvature of the C- C scaling is inherently linked with the background SST, with higher temperature corresponding to a more pronounced moisture- temperature sensitivity. The relationship suggests that projections of future mesoscale SST- LHF coupling are significantly influenced by the historical large- scale oceanic and atmospheric conditions. Given that the large- scale wind and the curvature of the C- C scaling stay positive, it can be inferred that the direction of mesoscale SST- LHF coupling is exclusively determined by the sign of projected SST changes, leading to consistent intensification in line with the warming of underlying SST. + +<|ref|>text<|/ref|><|det|>[[115, 412, 884, 892]]<|/det|> +We then assessed the regional variations in mesoscale SST- LHF coupling responses among the four WBC regions using Eq. (2), with emphasis on the north- south hemisphere asymmetry. The simplified relationship efficiently captures the more pronounced increase in coupling strength within the KE and GS regions compared to the ARC and BMC regions (Fig. 3a), consistent with the projected intensification of mesoscale SST- LHF coupling between future and historical simulations. Detailed examination of contributing factors in CESM- HR demonstrates consistently higher values for large- scale wind and SST in the KE and GS regions in the NH from historical simulations, alongside a more rapid SST warming rate (Fig. 3c), all of which jointly contribute to the enhanced augmentation in mesoscale coupling strength in the NH under climate change. The historical large- scale wind in the KE is approximately \(30\%\) stronger than that in the ARC and BMC, and the background SST in the GS is \(7^{\circ}\mathrm{C}\) higher than in the BMC, corroborated by observations (Fig. 3d). Additionally, the accelerated SST warming trend in the NH is also confirmed by the observational data (Fig. 3d). Further decomposition of the three factors indicates their respective contribution to the overall hemispheric discrepancy (Fig. 3b), with the large- scale SST warming rate being the dominant + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 85, 451, 103]]<|/det|> +factor, followed by the background SST. + +<|ref|>sub_title<|/ref|><|det|>[[120, 118, 362, 136]]<|/det|> +## Validations in HighResMIP + +<|ref|>text<|/ref|><|det|>[[117, 150, 884, 400]]<|/det|> +We extend the analyses to HighResMIP simulations (Methods) and to different warming periods (2030- 2050), aiming to verify the robustness of the findings. All models examined show an increase in mesoscale SST- LHF coupling in response to greenhouse warming across four WBCs by the year 2050 (Tab. 1). A more pronounced intensification of this coupling is projected in the NH compared to the SH, in line with CESM- HR results (Fig. 1e), albeit with a generally lower magnitude of change. The magnitude discrepancy is due to the HighResMIP projections here terminating in 2050, whereas CESM- HR projections shown in Fig 1 extend to a later period of the century (2080). + +<|ref|>text<|/ref|><|det|>[[117, 412, 884, 727]]<|/det|> +The effectiveness of the proposed theoretical framework for estimating changes in mesoscale SST- LHF coupling due to greenhouse warming was also evaluated across different models. A comparison between the coupling strength changes estimated by the theoretical framework and those projected by CESM- HR reveals a significant linear relationship across four WBCs (Fig. S3). The estimated and actual projected changes in mesoscale SST- LHF coupling are highly correlated, with a correlation coefficient of around 0.7 in the KE and GS, and 0.5 in the ARC and BMC (Tab. 1). Similar linear correlations, ranging generally from 0.5 to 0.8 (Tab. 1), are found between estimated and projected mesoscale coupling strength changes in HighResMIP models, confirming the broad applicability of the simplified framework across diverse climate models. + +<|ref|>sub_title<|/ref|><|det|>[[120, 740, 368, 757]]<|/det|> +## Conclusions and Discussions + +<|ref|>text<|/ref|><|det|>[[117, 772, 888, 889]]<|/det|> +How greenhouse warming will influence mesoscale air- sea interactions remains an open question. Utilizing eddy- resolving high- resolution CESM simulations, supported by observational and HighResMIP data, we found a ubiquitous intensification (approximately \(15\%\) ) of mesoscale SST- LHF coupling in WBC regions by the end of the \(21^{\text{st}}\) century under the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 883, 266]]<|/det|> +RCP8.5 warming scenario. The intensification is characterized by an increased nonlinearity between warm and cold eddies and a more pronounced enhancement in the NH. Our findings suggest an increasingly important role of mesoscale oceanic processes in shaping weather and climate systems in a warming climate, especially in the NH. This underscores the critical need for climate models to incorporate mesoscale air- sea interaction for more reliable climate projections. + +<|ref|>text<|/ref|><|det|>[[115, 281, 884, 660]]<|/det|> +We found that the mesoscale moisture response is the key factor driving the strengthened mesoscale SST- LHF coupling under climate change. To further understand the dynamics, we developed a theoretical framework to estimate the mesoscale moisture change. The framework builds a linkage between mesoscale coupling changes and large- scale fields, revealing that the projected mesoscale moisture changes can be estimated by the interplay among historical background wind, SST, and projected SST warming. The direction of mesoscale SST- LHF coupling changes is exclusively determined by the sign of projected SST changes. Given a scenario of SST warming, the mesoscale SST- LHF coupling will invariably exhibit intensification. The relationship highlights the importance of C- C scaling in determining changes in mesoscale SST- LHF coupling, suggesting that higher large- scale SST, determined either by historical baselines or future warming rates, are associated with greater rates of moisture increase and thereby enhanced augmentation of mesoscale SST- LHF coupling. + +<|ref|>text<|/ref|><|det|>[[115, 675, 884, 890]]<|/det|> +The effectiveness of the proposed theoretical framework was evaluated across CESM- HR and HighResMIP models. The analysis reveals a robust linear correlation between estimated and projected changes in mesoscale SST- LHF, suggesting the broad applicability of the theoretical framework within major WBC regions. However, it is important to acknowledge that the prerequisite conditions for the theoretical framework to work are the mesoscale moisture adjustment significantly surpasses the mesoscale wind adjustment, the mesoscale moisture adjustment at the ocean surface significantly exceeds that at the atmospheric surface, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 85, 881, 137]]<|/det|> +and the large- scale wind changes due to warming is minimal. These conditions may not hold outside the WBCs, potentially undermining the framework's applicability. + +<|ref|>sub_title<|/ref|><|det|>[[118, 184, 198, 201]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[118, 216, 605, 235]]<|/det|> +## CESM-HR, HighResMIP Simulations and Observations + +<|ref|>text<|/ref|><|det|>[[115, 248, 884, 563]]<|/det|> +We utilized the high- resolution Community Earth System Model (CESM- HR) simulations with \(\sim 0.25^{\circ}\) atmosphere and \(\sim 0.1^{\circ}\) ocean components developed by the National Center for Atmosphere Research (NCAR) (Chang et al., 2020). The simulations include a 500- year preindustrial control simulation and a 250- year historical and future simulation from 1850- 2100. Historical radiative forcing is applied from 1850 to 2005 while the Representative Concentration Pathway 8.5 (RCP8.5, a high greenhouse gas emission) warming forcing is switched from 2006 onwards. Two 20- year periods with high- frequency (daily) output were chosen to assess the mesoscale thermal coupling response to greenhouse warming in CESM- HR: a historical period from 1956 to 1975 (referred to as HIS) and a future period from 2063 to 2082 (referred to as RCP). + +<|ref|>text<|/ref|><|det|>[[115, 576, 884, 890]]<|/det|> +Five HighResMIP simulations with relatively high oceanic resolution were selected: CMCC- CM2- VHR4 ( \(0.25^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), HadGEM3- GC31- HH ( \(0.5^{\circ}\) atmosphere and \(0.08^{\circ}\) ocean), EC- Earth3P- HR ( \(0.5^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), CNRM- CM6- 1- HR ( \(1^{\circ}\) atmosphere and \(0.25^{\circ}\) ocean), MPI- ESM1- 2- XR ( \(0.5^{\circ}\) atmosphere and \(0.5^{\circ}\) ocean). We note that only one model (HadGEM3- GC31- HH) has comparable oceanic resolutions with CESM- HR, yet its atmospheric resolution is coarser. All these selected models include a 100- year historical and future (under RCP8.5 scenario) simulation from 1950- 2050. For a consistent comparison between CESM- HR and HighResMIP simulations, 1950- 1969 for historical and 2031- 2050 for future projections were selected for the corroborative analysis shown in Tab. 1. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 884, 400]]<|/det|> +Daily sea surface height (SSH) derived from Copernicus Marine Environment Monitoring Service (CMEMS) during 2003- 2007 was used to identify eddies in observations (Tab. S1). Concurrently, SST obtained from the National Oceanic and Atmospheric Administration daily Optimum Interpolation SST (NOAA- OISST) and heat fluxes derived from Japanese Ocean Flux Data Sets with Use of Remote Sensing Observations version3 (J- OFURO3) during the same period were employed to construct observational eddy composites (Fig. S1). ERA5 reanalysis (the fifth generation European Centre for Medium- Range Weather Forecasts atmospheric reanalysis, Hersbach et al., 2020) with an extended temporal span from 1979 to 2022 was utilized to examine the decadal trend of mesoscale coupling strength (Fig. 1a). + +<|ref|>sub_title<|/ref|><|det|>[[128, 412, 623, 432]]<|/det|> +## Eddy detection, Eddy Composites and High-pass Filtering + +<|ref|>text<|/ref|><|det|>[[115, 445, 884, 792]]<|/det|> +In both CMEMS observations and CESM- HR simulations, mesoscale eddies were detected using daily SSH anomalies derived by applying high- pass spatial filtering ( \(20^{\circ}\) longitude \(\times 10^{\circ}\) latitude) to remove the large- scale signal, following previous studies (Faghmous et al., 2013; Qu et al., 2022). Cyclonic (anticyclonic) eddies are classified by closed contours of SSH anomalies that include a single minimum (maximum), with an SSH anomaly increment (decrement) of \(0.05 \mathrm{cm}\) between successive contours. The edge of an eddy is delineated by the outmost closed contour of SSH anomalies. The radius of an eddy corresponds to the radius of a circle with an equivalent area to that enclosed by the outmost contour. The amplitude of an eddy is defined by the SSH anomaly difference between the eddy's peak and its defined edge. Only eddies with a radius ranging from 70 to \(200 \mathrm{km}\) and an amplitude exceeding \(3 \mathrm{cm}\) are included in the analysis. + +<|ref|>text<|/ref|><|det|>[[115, 805, 883, 891]]<|/det|> +Eddy composites were constructed by aligning associated variables to the reference coordinate of the eddy core. The variables were normalized by the individual eddy radius and oriented to the prevailing direction of the large- scale background surface wind, in line with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 881, 137]]<|/det|> +previous research (Frenger et al., 2013). Variables within twice the eddy radius were included for composite analysis. + +<|ref|>text<|/ref|><|det|>[[115, 149, 884, 563]]<|/det|> +In addition to eddy composite analysis, we also applied a high- pass Loess Filter with a cutoff wavelength of \(30^{\circ}\) longitude \(\times 10^{\circ}\) latitude (similar to a \(5^{\circ}\times 5^{\circ}\) box car average, Bryan et al. 2010, Ma et al. 2016) in ERA5, CESM- HR and HighResMIP, to isolate mesoscale SST and LHF and examined the spatial distribution of their coupling strength (Fig. 1a,b and Fig S3). The coupling strength at each grid point was computed using the linear regression coefficient between high- pass filtered monthly SST and LHF over a spatial domain of \(4^{\circ}\times 4^{\circ}\) . It was noted that applying a high- pass filter directly to monthly data yields a coupling coefficient comparable to that obtained when the filter is first applied to daily data, which is then aggregated into a monthly mean before calculating the coefficient. The former method was selected for its computational efficiency. To highlight regions with pronounced mesoscale SST- LHF coupling, mesoscale signals where the SST anomaly fell below \(0.4^{\circ}\mathrm{C}\) in ERA5 and below \(0.6^{\circ}\mathrm{C}\) in CESM- HR were excluded when computing the decadal trend in coupling strength (Fig. 1a, b). + +<|ref|>sub_title<|/ref|><|det|>[[117, 576, 531, 595]]<|/det|> +## Decomposition of mesoscale SST-LHF coupling + +<|ref|>text<|/ref|><|det|>[[117, 609, 881, 661]]<|/det|> +According to Bulk formula (Large and Yeager 2004), the latent heat flux \(Q_{L}\) is determined by the equation: + +<|ref|>equation<|/ref|><|det|>[[411, 675, 880, 696]]<|/det|> +\[Q_{L} = \rho_{a}\Lambda_{v}C_{e}U_{10}(q_{s} - q_{a}) \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[117, 708, 881, 800]]<|/det|> +Here, \(\rho_{a}\) is the surface air density, \(\Lambda_{v}\) is the latent heat of vaporization, and \(C_{e}\) is the transfer coefficients for evaporation. \(U_{10}\) is the 10m wind speed, \(q_{s}\) is the saturated specific humidity at the ocean surface, and \(q_{a}\) is air specific humidity at 2m. + +<|ref|>text<|/ref|><|det|>[[117, 814, 881, 900]]<|/det|> +Following previous studies (Yang et al., 2018; Yuan et al, 2023), equation (1) can be decomposed into large- scale background and mesoscale components. The mesoscale component of latent heat flux is estimated as follows: + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[364, 82, 879, 105]]<|/det|> +\[Q_{L}^{t} = \bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}[\overline{U}_{10}(q_{s}^{t} - q_{a}^{t}) + U_{10}^{t}(\overline{q}_{s} - \overline{q}_{a})] \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[67, 117, 881, 171]]<|/det|> +Here, the prime (') represents mesoscale anomalies defined by the high- pass spatial filtering, and the overbar (') denotes large- scale background excluding the mesoscale signal. + +<|ref|>text<|/ref|><|det|>[[67, 184, 881, 234]]<|/det|> +By differentiating with respect to SST, the mesoscale SST- LHF coupling is expressed as: + +<|ref|>equation<|/ref|><|det|>[[310, 247, 879, 280]]<|/det|> +\[\frac{dq_{L}^{t}}{dSST^{t}} = \bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}[\overline{U}_{10}(\frac{dq_{s}^{t}}{dSST^{t}} -\frac{dq_{a}^{t}}{dSST^{t}}) + \frac{dU_{10}^{t}}{dSST^{t}} (\overline{q}_{S} - \overline{q}_{a})] \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[67, 293, 881, 483]]<|/det|> +Based on calculations, the second term \(\frac{dU_{10}^{t}}{dSST^{t}} (\overline{q}_{S} - \overline{q}_{a})\) on the right- hand side of Eq. (3) is an order of magnitude smaller than the first term \(\overline{U}_{10}(\frac{dq_{s}^{t}}{dSST^{t}} -\frac{dq_{a}^{t}}{dSST^{t}})\) . Furthermore, \(\frac{dq_{a}^{t}}{dSST^{t}}\) is an order of magnitude smaller than \(\frac{dq_{s}^{t}}{dSST^{t}}\) . Disregarding the relatively smaller terms and assuming the changes in \(\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e}\) is minimal under global warming, the mesoscale SST- LHF coupling changes is predominately influenced by \(\overline{U}_{10}\frac{dq_{s}^{t}}{dSST^{t}}\) , which can be represented as: + +<|ref|>equation<|/ref|><|det|>[[227, 497, 879, 533]]<|/det|> +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}[\overline{U}_{10(F)}\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \overline{U}_{10(P)}\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})}] \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[67, 548, 881, 633]]<|/det|> +Where 'F' represents future values and 'P' represents historical values in the past. As the large- scale wind, \(\overline{U}_{10(F)}\) and \(\overline{U}_{10(C)}\) , experience minimal changes in the WBCs, Eq. (4) can be further simplified as: + +<|ref|>equation<|/ref|><|det|>[[247, 648, 879, 686]]<|/det|> +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}[\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})}] \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[178, 700, 821, 720]]<|/det|> +Applying a Taylor expansion, the right- hand side term can be approximated as: + +<|ref|>equation<|/ref|><|det|>[[295, 730, 879, 766]]<|/det|> +\[\left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(F)} - \left(\frac{dq_{s}^{t}}{dSST^{t}}\right)_{(\bar{P})} = \frac{d^{2}q_{s}(T)}{dT^{2}} |_{T = \overline{SST}_{(P)}}\cdot \Delta SST \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[179, 780, 433, 799]]<|/det|> +Substituting (6) into (5) yields: + +<|ref|>equation<|/ref|><|det|>[[255, 811, 879, 847]]<|/det|> +\[\left(\frac{dq_{L}^{t}}{dSST^{t}}\right)_{(F - P)} = (\bar{\rho}_{a}\overline{A}_{v}\bar{C}_{e})_{(P)}\cdot \overline{U}_{10(P)}\frac{d^{2}q_{s}(T)}{dT^{2}} |_{T = \overline{SST}_{(P)}}\cdot \Delta SST \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[179, 862, 881, 889]]<|/det|> +Where \(\frac{d^{2}q_{s}(T)}{dT^{2}}\) is determined by C- C scaling and exhibits an exponential increase with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 83, 884, 138]]<|/det|> +322 temperature. Note that the composite coefficient \((\bar{\rho}_{a}\bar{A}_{v}\bar{C}_{e})\) is retained to align the estimated 323 coupling strength changes with magnitude analogous to the actual projections. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 270, 103]]<|/det|> +## Data Availability + +<|ref|>text<|/ref|><|det|>[[117, 118, 883, 300]]<|/det|> +Data AvailabilityCMEMS data can be obtained through https://data.marine.copernicus.eu/products. NOAA- OISST can be accessed through https://www.ncei.noaa.gov/products/optimum- interpolation- sst. J- OFURO3 can be achieved through https://www.j- ofuro.com/en/. ERA5 reanalysis can be downloaded from https://doi.org/10.24381/cds.bd0915c6. The CESM simulations can be achieved through https://ihep.github.io/archive/products/ds_archive/Sunway_Runs.html. The HighResMIP data can be downloaded from https://pcmdi.llnl.gov/CMIP6/. + +<|ref|>sub_title<|/ref|><|det|>[[118, 315, 272, 332]]<|/det|> +## Code Availability + +<|ref|>text<|/ref|><|det|>[[118, 347, 882, 400]]<|/det|> +Code AvailabilityMATLAB codes to reproduce the analyses are available upon request from the corresponding author or can be accessed through the link https://zenodo.org/records/10610386. + +<|ref|>sub_title<|/ref|><|det|>[[118, 413, 281, 431]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[116, 445, 883, 660]]<|/det|> +AcknowledgmentsThis research is supported by the National Natural Science Foundation of China (42376025), Science and Technology Innovation Program of Laoshan Laboratory (LSKJ202300302, LSKJ202202503), Shandong Provincial Natural Science Foundation (ZR2022YQ29), Taishan Scholar Funds (tsqn202103028). We thank Sunway TaihuLight High- Performance Computer (Wuxi), Laoshan Laboratory in Qingdao and the National Supercomputing center in Jinan for providing the high resolution CESM simulations and high- performance computing resources that contributed to the research results reported in this paper. + +<|ref|>sub_title<|/ref|><|det|>[[118, 675, 311, 693]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[116, 707, 884, 857]]<|/det|> +Author ContributionsX. M. and X. Z conceived the study. X. M. instructed the investigation and wrote the manuscript. X. Z. performed the analyses and produced all figures. L. W. supervised the project. P. Y., F. S. and M. Y. contributed to the discussion of mesoscale thermal coupling decomposition. Y. Q and H. C. offered insights into eddy detection. Z. J., Z. C. and B. G. contributed to interpreting the results and improving the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[118, 872, 387, 889]]<|/det|> +## Competing Interests statement + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 85, 475, 103]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[118, 143, 217, 160]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[118, 191, 881, 235]]<|/det|> +Beech, Nathan, et al. Long- term evolution of ocean eddy activity in a warming world. Nature climate change 12.10 (2022): 910- 917 + +<|ref|>text<|/ref|><|det|>[[118, 265, 881, 309]]<|/det|> +Bryan, Frank O., et al. Frontal scale air- sea interaction in high- resolution coupled climate models. Journal of Climate 23.23 (2010): 6277- 6291. + +<|ref|>text<|/ref|><|det|>[[118, 339, 881, 406]]<|/det|> +Chang, Ping, et al. An unprecedented set of high- resolution earth system simulations for understanding multiscale interactions in climate variability and change. Journal of Advances in Modeling Earth Systems 12.12 (2020): e2020MS002298. + +<|ref|>text<|/ref|><|det|>[[118, 437, 881, 480]]<|/det|> +Chelton, Dudley B., et al. Satellite measurements reveal persistent small- scale features in ocean winds. science 303.5660 (2004): 978- 983. + +<|ref|>text<|/ref|><|det|>[[118, 510, 881, 578]]<|/det|> +Chen, Longjing, Yinglai Jia, and Qinyu Liu. Oceanic eddy- driven atmospheric secondary circulation in the winter Kuroshio Extension region. Journal of Oceanography 73 (2017): 295- 307. + +<|ref|>text<|/ref|><|det|>[[118, 608, 881, 676]]<|/det|> +Czaja, Arnaud, et al. Simulating the midlatitude atmospheric circulation: what might we gain from high- resolution modeling of air- sea interactions?. Current Climate Change reports 5 (2019): 390- 406. + +<|ref|>text<|/ref|><|det|>[[118, 707, 881, 751]]<|/det|> +Faghmous, James H., et al. A parameter- free spatio- temporal pattern mining model to catalog global ocean dynamics. 2013 IEEE 13th International conference on data mining. IEEE, 2013. + +<|ref|>text<|/ref|><|det|>[[118, 783, 881, 825]]<|/det|> +Frenger, Ivy, et al. Imprint of Southern Ocean eddies on winds, clouds and rainfall. Nature geoscience 6.8 (2013): 608- 612. + +<|ref|>text<|/ref|><|det|>[[118, 856, 881, 899]]<|/det|> +Foussard, A., G. Lapeyre, and R. Plougonven. Storm track response to oceanic eddies in idealized atmospheric simulations. Journal of Climate 32.2 (2019): 445- 463. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 109, 883, 153]]<|/det|> +Gan, Bolan, et al. North Atlantic subtropical mode water formation controlled by Gulf Stream fronts. National Science Review (2023): nwad133. + +<|ref|>text<|/ref|><|det|>[[115, 184, 883, 227]]<|/det|> +Hersbach, Hans, et al. The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society 146.730 (2020): 1999- 2049. + +<|ref|>text<|/ref|><|det|>[[115, 256, 883, 323]]<|/det|> +Jiang, Yuxi, et al. Effects of a cold ocean eddy on local atmospheric boundary layer near the Kuroshio Extension: In situ observations and model experiments. Journal of Geophysical Research: Atmospheres 124.11 (2019): 5779- 5790. + +<|ref|>text<|/ref|><|det|>[[115, 356, 883, 399]]<|/det|> +Jing, Zhao, et al. Maintenance of mid- latitude oceanic fronts by mesoscale eddies. Science advances 6.31 (2020): eaba7880. + +<|ref|>text<|/ref|><|det|>[[115, 429, 883, 472]]<|/det|> +Large, William George, and Stephen G. Yeager. Diurnal to decadal global forcing for ocean and sea- ice models: The data sets and flux climatologies. (2004). + +<|ref|>text<|/ref|><|det|>[[115, 502, 883, 545]]<|/det|> +Liu, Xue, et al. Ocean fronts and eddies force atmospheric rivers and heavy precipitation in western North America. Nature communications 12.1 (2021): 1268. + +<|ref|>text<|/ref|><|det|>[[115, 576, 881, 619]]<|/det|> +Ma, Xiaohui, et al. Distant Influence of Kuroshio Eddies on North Pacific Weather Patterns?. Sci Rep 5, 17785 (2015). https://doi.org/10.1038/srep17785 + +<|ref|>text<|/ref|><|det|>[[115, 650, 883, 693]]<|/det|> +Ma, Xiaohui, et al. Western boundary currents regulated by interaction between ocean eddies and the atmosphere. Nature 535.7613 (2016): 533- 537. + +<|ref|>text<|/ref|><|det|>[[115, 723, 883, 767]]<|/det|> +Ma, Xiaohui, et al. Importance of resolving Kuroshio front and eddy influence in simulating the North Pacific storm track. Journal of Climate 30.5 (2017): 1861- 1880. + +<|ref|>text<|/ref|><|det|>[[115, 797, 883, 840]]<|/det|> +Martínez- Moreno, Josué, et al. Global changes in oceanic mesoscale currents over the satellite altimetry record. Nature Climate Change 11.5 (2021): 397- 403. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 884, 152]]<|/det|> +Parfitt, R., Czaja, A., Minobe, S., & Kuwano- Yoshida, A. (2016). The atmospheric frontal response to SST perturbations in the Gulf Stream region. Geophysical Research Letters, 43(5), 2299- 2306. + +<|ref|>text<|/ref|><|det|>[[115, 183, 886, 250]]<|/det|> +Qu, Yushan, et al. Spatial Structure of Vertical Motions and Associated Heat Flux Induced by Mesoscale Eddies in the Upper Kuroshio- Oyashio Extension. Journal of Geophysical Research: Oceans 127.10 (2022): e2022JC018781. + +<|ref|>text<|/ref|><|det|>[[115, 281, 884, 324]]<|/det|> +Renault, Lionel, et al. Disentangling the mesoscale ocean- atmosphere interactions. Journal of Geophysical Research: Oceans 124.3 (2019): 2164- 2178. + +<|ref|>text<|/ref|><|det|>[[115, 355, 884, 398]]<|/det|> +Seo, Hyodae, et al. "Ocean Mesoscale and Frontal- Scale Ocean- Atmosphere Interactions and Influence on Large- Scale Climate: A Review." Journal of Climate 36.7 (2023): 1981- 2013. + +<|ref|>text<|/ref|><|det|>[[115, 428, 884, 496]]<|/det|> +Sheldon, L., Czaja, A., Vanniere, B., Morcrette, C., Sohet, B., Casado, M., & Smith, D. (2017). A 'warm path' for Gulf Stream- troposphere interactions. Tellus A: Dynamic Meteorology and Oceanography, 69(1), 1299397. + +<|ref|>text<|/ref|><|det|>[[115, 527, 884, 570]]<|/det|> +Skyllingstad, Eric D., et al. Effects of mesoscale sea- surface temperature fronts on the marine atmospheric boundary layer. Boundary- layer meteorology 123 (2007): 219- 237. + +<|ref|>text<|/ref|><|det|>[[115, 601, 883, 644]]<|/det|> +Small, RJ D., et al. Air- sea interaction over ocean fronts and eddies. Dynamics of Atmospheres and Oceans 45.3- 4 (2008): 274- 319. + +<|ref|>text<|/ref|><|det|>[[115, 675, 883, 741]]<|/det|> +Souza, J. M. A. C., Bertrand Chapron, and Emmanuelle Autret. The surface thermal signature and air- sea coupling over the Agulhas rings propagating in the South Atlantic Ocean interior. Ocean Science 10.4 (2014): 633- 644. + +<|ref|>text<|/ref|><|det|>[[115, 773, 883, 816]]<|/det|> +Xie, Shang- Ping. Satellite observations of cool ocean- atmosphere interaction. Bulletin of the American Meteorological Society 85.2 (2004): 195- 208. + +<|ref|>text<|/ref|><|det|>[[115, 846, 883, 915]]<|/det|> +Yang, Peiran, Zhao Jing, and Lixin Wu. An assessment of representation of oceanic mesoscale eddy- atmosphere interaction in the current generation of general circulation models and reanalyses. Geophysical Research Letters 45.21 (2018): 11- 856. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 108, 884, 280]]<|/det|> +Yuan, Man, et al. Spatio- temporal variability of surface turbulent heat flux feedback for mesoscale sea surface temperature anomaly in the global ocean. Frontiers in Marine Science 9 (2022): 957796. Zhang, Xingzhi, Xiaohui Ma, and Lixin Wu. Effect of mesoscale oceanic eddies on extratropical cyclogenesis: A tracking approach. Journal of Geophysical Research: Atmospheres 124.12 (2019): 6411- 6422. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 150, 844, 483]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 500, 881, 763]]<|/det|> +
Fig. 1 Observed and simulated trends in mesoscale SST-LHF coupling. Global distribution of the decadal trends of mesoscale SST-LHF coupling strength as derived from ERA5 (a) and CESM-HR (b) during 1979-2022. The coupling strength is computed using the linear regression coefficient between high-pass filtered monthly SST and LHF (Methods) with trends significant at a 99% confidence level indicated by black dots. (c) Composites of SST (color shading) and LHF (contours) anomalies associated with mesoscale oceanic eddies during historical periods (1956-1975) in CESM-HR. Shown are winter season mean anomalies in warm minus cold eddy composites within four WBC regions (outlined by red boxes in a, b). The white circle represents one eddy radius and the white dot marks the eddy center. (d) Mesoscale SST-LHF coupling strength during historical (1956-1975) and future (2063-2082) periods in CESM-HR for warm (red) and cold (blue) eddies averaged across four WBC regions. The coupling strength is computed using the linear regression coefficient between SST and LHF anomalies within twice the radius of eddy composites. (e) Differences in mesoscale SST-LHF coupling strength between future and historical periods in CESM-HR for warm (red bars) and cold (blue bars) eddies in the KE, GS, ARC and BMC regions, with fractional differences labeled atop the respective bars.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 120, 864, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 461, 881, 585]]<|/det|> +
Fig. 2 Decomposition of mesoscale SST-LHF coupling response under greenhouse warming. (a) Differences in thermodynamic and dynamic adjustments between future and historical periods in CESM-HR in the KE region. From left to right, the terms plotted are thermodynamic adjustment \(\overline{U}_{10}(\frac{aq_s'}{dsST'}, \frac{aq_a'}{dsST'})\) , dynamic adjustment \(\frac{au_{10}'}{dsST'}(\overline{q}_s - \overline{q}_a)\) , large-scale surface wind \((\overline{U}_{10})\) and moisture adjustment \((\frac{aq_s'}{dsST'}, \frac{aq_a'}{dsST'})\) . (b-d), as for a, but for the GS, ARC and BMC regions, respectively.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[209, 90, 780, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 640, 881, 888]]<|/det|> +
Fig. 3 Decomposition of mesoscale moisture response under greenhouse warming. (a) Estimated changes in mesoscale SST-LHF coupling between future and historical periods in CESM-HR according to Eq. 2 in the KE GS, ARC and BMC regions. The large-scale SST averaged in the WBCs are displayed on the bottom axis for each respective region. Black lines denote the standard deviations of the corresponding variables. (b) Contributions of \(\overline{U}_{10}, \frac{d^2 q_s(T)}{dT^2}\) and \(\Delta SST\) to the north-south hemispheric asymmetry in the KE GS, ARC and BMC regions. The total deviation from the hemispheric mean due to three terms in a specific region is approximated by \((ABC)' \approx A' \bar{B} \bar{C} + B' \bar{A} \bar{C} + C' \bar{A} \bar{B}\) . Here, A, B, and C denotes the three terms; the overbar denotes a mean component (the average across the four WBCs); the prime denotes an anomaly component (the deviation from the hemispheric mean for each specific WBC region). (c) The historical large-scale surface wind, SST and the projected SST warming within the four WBC regions in CESM-HR. (d) as for c, but for observational data. Surface wind is derived from ERA5 and SST is derived from NOAA-OISST. \(\Delta SST\) is computed based on the warming trend observed over the last 40 years.
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[157, 148, 841, 365]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[67, 86, 880, 120]]<|/det|> +Table. 1 Projected and estimated mesoscale SST-LHF coupling changes in CESM-HR and HighResMIP models. + +
Projected coupling changes (RCP-HIS) (W·m-2·°C-1)Correlation between projected and estimated changes
KEGSARCBMCKEGSARCBMC
CESM-HR3.2 (10.26%)3.5 (15.34%)1.9 (6.68%)1.5 (6.72%)0.710.690.450.49
CMCC-CM2-VHR44.1 (14.27%)2.5 (11.97%)1.5 (6.65%)0.07 (0.35%)0.760.650.600.54
HadGEM3-GC31-HH3.2 (11.95%)2.0 (8.76%)2.8 (11.34%)2.4 (11.88%)0.780.660.580.61
MPI-ESM1-2-XR2.9 (14.06%)2.0 (10.14%)1.6 (9%)0.9 (5.24%)0.410.690.720.78
EC-Earth3P-HR2.0 (8.37%)2.4 (11.24%)1.5 (6.41%)0.5 (2.72%)0.720.760.690.66
CNRM-CM6-1-HR2.2 (10.1%)2.4 (12.4%)1.2 (7.52%)0.9 (3.58%)0.690.70.670.58
+ +<|ref|>table_footnote<|/ref|><|det|>[[67, 366, 880, 415]]<|/det|> +506 Note that mesoscale SST-LHF coupling strength changes analyzed here correspond to 507 differences between the historical period of 1950-1969 and the mid-21st century projections for 508 the period of 2031-2050. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 92, 768, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 137, 149]]<|/det|> +- Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/images_list.json b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..31f490433def1ac08607812cc3ee8deb89419f72 --- /dev/null +++ b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/images_list.json @@ -0,0 +1,123 @@ +[ + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Vaccine acceptance by current cases and mortality", + "footnote": [], + "bbox": [], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Panel b. Vaccine acceptance and COVID-19 mortality", + "footnote": [], + "bbox": [ + [ + 125, + 420, + 733, + 664 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "Panel c. Vaccine acceptance and COVID-SCORE-10 (June 2020)", + "footnote": [], + "bbox": [ + [ + 123, + 110, + 763, + 365 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_2.jpg", + "caption": "Panel d. Vaccine acceptance and COVID-VAC", + "footnote": [], + "bbox": [ + [ + 120, + 442, + 748, + 667 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_3.jpg", + "caption": "Panel e. Vaccination rates and hesitancy", + "footnote": [], + "bbox": [ + [ + 128, + 110, + 755, + 345 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Vaccine acceptance for children among parents", + "footnote": [], + "bbox": [ + [ + 118, + 457, + 905, + 710 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Correlates of vaccine acceptance-sociodemographic factors and COVID-19 experience.", + "footnote": [], + "bbox": [], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Correlates of vaccine acceptance-COVID-VAC score, trust in government, anxiety and depression.", + "footnote": [], + "bbox": [ + [ + 115, + 120, + 940, + 856 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. Vaccine mandates acceptance.", + "footnote": [], + "bbox": [ + [ + 81, + 175, + 980, + 550 + ] + ], + "page_idx": 29 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e.mmd b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e.mmd new file mode 100644 index 0000000000000000000000000000000000000000..bb9afe0661f690d0be88672f5fd6173f2e862190 --- /dev/null +++ b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e.mmd @@ -0,0 +1,363 @@ + +# COVID-VAC: The second global study of COVID-19 vaccine acceptance + +Jeffrey Lazarus ( Jeffrey.Lazarus@lSGlobal.org ) Barcelona Institute for Global Health (ISGlobal) https://orcid.org/0000- 0001- 9618- 2299 + +Katarzyna Wyka Graduate School of Public Health & Health Policy, City University of New York (CUNY) + +Trenton White Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona https://orcid.org/0000- 0002- 0633- 4445 + +Camila Picchio Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona + +Kenneth Rabin City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +Scott Ratzan City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +Jeanna Parsons Leigh Dalhousie University + +Jia Hu University of Calgary + +Ayman El-Mohandes City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +## Article + +Keywords: COVID- 19, disease control, SARS- CoV- 2, vaccination + +Posted Date: August 14th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 780128/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 31441-x. + +<--- Page Split ---> + +# 1. 1. 1. 1. 1. 1. 1. 2. 2. 2. 2. 2. 2. 2. + +<--- Page Split ---> + +## COVID-VAC: The second global study of COVID-19 vaccine acceptance + +Jeffrey V Lazarus1,2\*, Katarzyna Wyka2, Trenton M White1, Camila A Picchio1, Kenneth Rabin2, Scott C Ratzan2, Jeanna Parsons Leigh3, Jia Hu4, Ayman El- Mohandes2 + +1. Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona, Spain + +2. Graduate School of Public Health & Health Policy, City University of New York (CUNY), New York, New York, United States + +3. School of Health Administration, Dalhousie University, Halifax, Canada + +4. University of Calgary, Calgary, Canada + +\* Corresponding Author: Jeffrey V Lazarus, jeffrey.lazarus@isglobal.org + +## Abstract (150/150 words) + +This one- year follow- up study investigated COVID- 19 vaccine acceptance in 23 countries. In June 2021, 75.2% of respondents reported to accept the vaccine, of which 49% had received at least one vaccine dose and 51% were willing to get it once available to them. Factors associated with COVID- 19 vaccine acceptance include perceived safety and trust in science, as measured by a novel COVID- VAC score, and personal experience with COVID- 19. Males, healthcare workers, well- to- do people, those with university degrees, and those whose physician recommended vaccination were more likely to accept it. Trust in governmental ability to address COVID- 19 was weakly associated with vaccine acceptance. Respondents strongly supported vaccination to travel internationally, but weakly supported vaccinating children for school attendance. Hesitancy rates remain high in LMICs. Health policymakers should aggressively communicate vaccination need and inform accurately about efficacy and safety compared to disease risk, and healthcare workers need a greater vaccine communications role. + +Keywords: COVID- 19, disease control, SARS- CoV- 2, vaccination + +<--- Page Split ---> + +## Introduction + +Vaccine hesitancy is a global challenge, declared by the World Health Organization in 2019 as one of the ten greatest global health threats1, and a vital concern since vaccination is the most effective intervention to control the ongoing COVID- 19 pandemic. In June 2020, as most countries were experiencing a second wave of SARS- CoV- 2 transmission, the authors of this study reported potentially low COVID- 19 vaccine acceptance among more than 13,000 respondents in 19 of the hardest- hit countries.2 As of 1 August 2021, 1.07 billion individuals globally had received one dose and another 1.14 billion received two doses of an available COVID- 19 vaccine; however, more than 70% of the world (90% in low- and middle- income countries (LMIC)) remains unvaccinated.3 + +Globally, rates of COVID- 19 vaccine acceptance vary. In a 2021 study of more than 30,000 respondents from eight countries, 27% of participants were hesitant, ranging from 9.6% in Brazil to 47.3% in France.4 Another survey of more than 18,000 respondents from 15 countries reported a range of 54% vaccine acceptance in France to 87% in India.5 A recent review of 15 peer- reviewed studies calculates a COVID- 19 vaccine acceptance rate of 77.6% in the general population and 86.1% among students.6 Another study reported vaccine acceptance ranges from 23.6% in Kuwait to 97% in Ecuador.7 The lowest rates of acceptance to date are reported in the Middle East, Africa, and several European countries (notably France, Italy, Poland, and Russia). + +The main causes of vaccine hesitancy in these studies included concerns about vaccine safety and efficacy, mistrust of governments and health and scientific institutions, individual religious and political beliefs, as well as misinformation and conspiracy theories.8- 11 Most studies reported men to be more likely to accept COVID- 19 vaccines than women.7 Other variables associated with lower vaccine acceptance include living in rural areas and lower income. Conversely, COVID- 19 vaccine acceptance correlated strongly among those who reported greater concern about the effects of + +<--- Page Split ---> + +COVID- 19 disease and those previously vaccinated against influenza.4 Possessing accurate knowledge about COVID- 19 disease transmissibility and prevention, fear of contracting COVID- 19, and higher income all correlated positively with acceptance, but having a chronic disease and being female did not.12 The association between age and vaccine acceptance varied geographically, with some studies showing greater acceptance among older (above 65 years)4 and others among younger (below 40 years) age groups.12 A survey conducted in nine LMICs in Asia and Africa showed that vaccine acceptance increased with a higher perception of effectiveness; 76.4% accepting a hypothetical 90% effective vaccine increased to 88.8% accepting if 95% effective.12 + +Following a year of substantial but very uneven global availability, administration and acceptance of COVID- 19 vaccines, further investigation is necessary to understand how knowledge, risk perception, trust, and other factors associated with acceptance continue to evolve. The present study was undertaken in June 2021 to assess vaccine acceptance and uptake among 23 countries that represent approximately 60% of the world's population. In addition, comparisons in acceptance rates among the 19 countries in our 2020 study were included. Novel investigations in this 2021 study include testing the predictive value of a new vaccine acceptance score (COVID- VAC), which represents perceptions of risk, trust, safety, equity, and efficacy; the impact of mental health on vaccine acceptance; and acceptance for a range of requirements for proof of vaccination, and vaccination for children. + +## Results + +## Sample characteristics + +23,000 participants from 23 countries responded to the survey. Approximately half were female (50.2%), ages 30- 59 (59.9%), and resided in LMICs (52.2%). One- fifth were university graduates (22.4%) (Table 1). Healthcare workers represented 10.8% of respondents. COVID- 19 illness (self or + +<--- Page Split ---> + +family) or loss of a family member to COVID- 19 were most common in Ecuador (48.3% and 19.2%), Brazil (38.3% and 18.3) and Peru (35.5% and 34.7%), and least common in China (1.5% and 2%), Singapore (3.3% and 5.3%), South Korea (3.8% and 2.1%), Ghana (5.8% and 1.5%), and Nigeria (6% and 2.3%). Loss of income was highest in lower- middle and upper- middle income countries (range 31.7% in Germany to 95.1% in Ecuador). Reported anxiety prevalence ranged from 9.2% in Ghana to 44.7% in Turkey, and depression was lowest in China (12.6%) and Ghana (12.9%) and highest in Turkey (38.7%). + +Vaccine acceptance, June 2021 + +In June 2021, global COVID- 19 vaccine acceptance was 75.2%, of whom 49% reported having received at least one vaccine dose and 51% were willing to get it once available to them (Figure 1). Reported vaccine acceptance was highest in China (97.6% (of whom 91.9% had received at least one dose)), Canada (79.2% (86.3%)), and the UK (81.2% (82.6%)) and lowest in Russia (51.6% (30.4%)), Nigeria (57% (3.5%)), and Poland (59.3% (74.2%) (Figure 1). The mean difference between these single- dose figures and official rates \(^{13}\) was 2.7%; reliable data for China are unavailable. + +Compared to June 2020, vaccine acceptance in the 19 countries previously surveyed increased from 71.5% to 75.2%, though an opposite trend was observed in South Africa (- 20.9%), the US (- 8.8%), Nigeria (- 8.2%), and Russia (- 3.3%). Vaccine acceptance when recommended by employer increased in all countries (global average increased from 48.1% in 2020 to 64.2% in 2021), except in Germany (- 3%) and South Korea (- 0.2%). Vaccine acceptance when recommended by one's doctor (range 56.4% in Russia to 95.1% in China), was higher in all countries compared to one's employer (range 44.4% in Russia to 96.7% in China). Notably, among respondents unsure/unwilling to vaccinate, potential vaccine acceptance was most likely if recommended by one's doctor, particularly in India (63.5%), Kenya (38%), Brazil (36.8%), Turkey (29.9%), South Korea (28.4%), and Russia (26.5%). Generally, + +<--- Page Split ---> + +those unsure/unwilling to take the vaccine were more likely to follow their doctors' recommendation \((21.9\% + / - 12.3\%)\) than employers' recommendation \((13.2\% + / - 7.7\%)\) ; \(p = 0.006\) (Figure 1). + +Vaccine acceptance correlated highly with perceptions of risk, trust, safety, equity, and efficacy, measured using COVID- VAC (June 2021: \(r = 0.85\) , \(p< 0.001\) ). In validation analyses, COVID- VAC was unidimensional with a Cronbach's alpha of 0.86. Vaccine acceptance weakly correlated with countries' current COVID- 19 case burden \((r = - 0.13\) , \(p > 0.05\) ), mortality \((r = - 0.25\) , \(p > 0.05\) ), and approval of government responses to the pandemic in June 2020, measured using COVID- SCORE- \(10^{11}\) \((r = 0.25\) , \(p > 0.05\) ) (Figure 2). + +Countries' vaccination rates \(^{13}\) and hesitancy showed moderate negative association \((r = 0.45\) , \(p = 0.034\) ) (Figure 2). Generally, countries in Africa had low vaccination uptake and high hesitancy; whereas India, South Korea, and countries in South America had higher uptake and lower hesitancy. Countries with \(40\% +\) uptake had similar hesitancy to the latter group, with the exception of Poland, France and the US, which had high hesitancy. + +Parents' willingness to vaccinate children was highest in China \((95\%)\) , Brazil \((91.3\%)\) , Ecuador \((85.9\%)\) , Peru \((85.1\%)\) and lowest in Russia \((35.5\%)\) , Poland \((46.3\%)\) and France \((48.5\%)\) . In all countries, willingness to vaccinate one's children was significantly greater among parents who accepted the vaccine \((p< 0.001)\) (Figure 3). + +Correlates of vaccine acceptance June 2021 + +In multivariable models, vaccine acceptance was associated with older respondents in Canada, France, Germany, South Korea, Sweden, UK, and US (aOR range, 1.02- 1.04) (Figure 4) and younger + +<--- Page Split ---> + +ones in Nigeria (aOR=0.89) and Ghana (aOR=0.93). Male gender was significantly positively correlated in Ghana, Nigeria, South Africa, and the US (aOR range 1.51- 8.84) and female gender in Kenya and Peru (aOR=0.39 and aOR=0.32). Having a university degree was significantly associated with vaccine acceptance in France, Mexico, South Africa, and the US (aOR range, 1.75- 5.55). Personal or family COVID- 19 illness or death of a family member due to COVID- 19 was strongly associated with vaccine acceptance, with statistically significant associations with personal or family illness in Germany, Peru, Poland, Turkey, Singapore, UK (aOR range 1.89- 5.72) and with loss of a family member in Brazil, France, Germany, Ghana, Mexico, Peru, Poland, Sweden, Turkey and the UK (aOR range 2.75- 9.25). Having low or no income was associated with less vaccine acceptance in Brazil, France, Germany, Nigeria, Peru, Russia, South Korea, Spain, Turkey, and the US (aOR range 0.08- 0.55). Moderate or severe income loss due to the pandemic was positively associated with vaccine acceptance in Ghana, Nigeria, Poland, and the US (aOR range 2.01- 6.27). + +Adjusted for socio- demographic factors and COVID- 19 experience, the COVID- VAC score was the most salient and consistent positive correlate of vaccine acceptance in all countries (aOR range, 1.41- 3.44) (Figure 5). The strongest correlation was with "I trust the science behind the COVID- 19 vaccines" and "safety of the vaccines." The lowest correlation was with "I trust my government is able to deliver the vaccine to everyone and everywhere in my country" (Supplemental Table 1). In univariable analyses, vaccine acceptance was significantly higher among respondents who reported that they trust their central and/or local government. This finding was not universal, however, and was not found in Brazil, China, Ecuador, and Peru (Supplemental Figure 1). After adjustment, trust in central government had a significantly positive association in Russia and Singapore (aOR=2.05 and aOR=5.28) and trust in local government had a negative association in South Africa (aOR=0.44). In Poland, vaccine acceptance was negatively correlated with trust in central government (aOR=.48) and positively correlated with trust in local government (aOR=2.9). Finally, vaccine acceptance positively correlated with anxiety in Canada (aOR=4.59) and Peru (aOR=3.64) and negatively + +<--- Page Split ---> + +correlated in South Korea (aOR=.43) and the US (aOR=.31), while depression was a negative factor in Canada (aOR=.21) and South Africa (aOR=.53) (Figure 5). + +Healthcare workers (HCW) + +Vaccine acceptance was significantly higher among HCWs globally compared to non- HCWs (91.8% vs. 83%, p<0.001, aOR=1.72, 95%CI [1.38, 2.13]) (Table 2). Vaccine hesitancy was significantly lower among physicians (3.1%) followed by nurses (6.5%), community health workers (7.8%), and other HCWs (14.0%). + +## Proof of vaccination requirements + +Support for requiring vaccines was highest in China, India, Peru and South Korea and lowest in Poland, Russia and Germany. Strong support was reported globally (74.4%) for proof of vaccination to travel internationally, particularly in China (93.3%), South Korea (87.8%), India (87.7%), Brazil (86.4%), and Ecuador (83%), with less support in Poland (52.7%) and Russia (52.5%). Overall support was 63.3% and 63.1% for vaccination to attend university and indoor activities, respectively. These ranged from lows in Russia of 33.1% for university and 37.6% for indoor activitisteos highs in China of 95.4% and 89.7%. Support for governments or employers to require vaccination followed similar trends. Agreement with mandates to vaccinate school children received the lowest support overall (58.2%), ranging from 24.4% in Russia to 86.6% in China (Figure 6). + +## Discussion + +Reported COVID- 19 vaccine acceptance, measured by intention to get vaccinated or having received at least one dose of an available vaccine, increased over the last year in 15 of the 19 countries studied both years and was 75.2% for the 23 countries studied in 2021, still below what is needed to + +<--- Page Split ---> + +achieve global herd immunity (80- 85%). Among the six- item COVID- VAC score, positive perceptions of safety and trust in the science behind vaccine development were the strongest predictors of acceptance, while perceptions of efficacy, benefit, and equity also held strong associations. Other determinants of vaccine acceptance varied by country; personal experience with COVID- 19 (e.g. sickness or loss of a family member) predominated, but other demographic characteristics (e.g. gender, education and income) showed varying degrees of association, and government trust was associated with vaccine acceptance in a few countries. + +In order to improve the global vaccination rate, some countries may require people to present proof of vaccination to attend work, school, or indoor activities and events. Our results found the strongest support for requirements targeting international travellers and support was weakest for requirements for school children. Importantly, vaccine acceptance improves among those indicating unwillingness to vaccinate if it is recommended by a doctor, and to a lesser extent by an employer. + +Misperceptions of vaccines as having high risks and low benefits drive vaccine hesitancy,14 which may also be influenced by lower trust in the science behind vaccine research and production.15- 19 In late 2019 and early 2020 the majority of people in most countries preceived the importance of science and trust in scientists positively.20 Our results corroborate these findings where perceptions of trust in science and vaccine safety were most predictive of vaccine acceptance. + +Healthcare workers are in a privileged position of trust and are therefore important sources of reliable and accurate vaccine information.21- 24 Among HCWs, COVID- 19 vaccine hesitancy ranged between 4.3% and 72%, with an average of 22.5%.25 Similar to the general population, the main reasons for vaccine hesitancy among HCWs were concerns about efficacy, safety, and potential side- effects. Demographic factors among HCWs such as male gender, older age and holding a doctoral degree were positively associated with acceptance of COVID- 19 vaccines.25 Our study aligns with + +<--- Page Split ---> + +similarly recorded trends of higher rates of vaccination acceptance among physicians compared to nurses. \(^{26 - 29}\) Among hesitant respondents in this survey, the advice of their physician was the factor most likely to change their minds, followed by the recommendation of their employer. This effect could be amplified if the physician had personally received the vaccine. + +In June 2020, trust in government to successfully address unexpected health threats, including the COVID- 19 pandemic, was one of the strongest factors associated with acceptance of an (at that time) unavailable vaccine. \(^{2}\) A 2020 study in Portugal, where 56% reported they would wait to take the COVID- 19 vaccine and 9% would refuse, showed that vaccine hesitancy was related to a poor perception of government and health service response as well as a lack of trust in the information provided, \(^{30}\) which is consistent with previous pandemic research. \(^{31}\) Yet, after accounting for sociodemographic, vaccine- specific, and COVID- 19 experience variables, our study found trust in central government to be significantly associated with vaccine acceptance only in Russia and Singapore and for local government in only Poland. It is notable also that Nigeria reported low vaccine acceptance and high distrust in governmental ability to respond to COVID- 19, \(^{32}\) but the association was not significant. This dissonance between trust in government and vaccine acceptance in our study could be related to the population's general dissatisfaction with government responses to the pandemic and its economic consequences, with vaccine acceptance being independent of such sentiments and more a reflection of personal experiences with COVID- 19 illness or loss of life and livelihood. Our results confirm that direct experience for self or family with the illness, and/or loss of a family member to the disease are independently associated with vaccine acceptance. + +As health systems in LMICs struggle to address COVID- 19, \(^{33}\) vaccine access has become cause for national frustrations. Most LMICs have been slow to receive and distribute vaccines, which are significantly more available in high- income countries, prompting critiques of global vaccine inequity. \(^{34}\) + +<--- Page Split ---> + +Most recent studies do not investigate income as a potential influencer on vaccine acceptance. A study conducted in Portugal found people who lost income during the pandemic were more hesitant. \(^{30}\) An Irish study showed that people with lower income were also more vaccine hesitant. \(^{35}\) In China, loss of income was greater among residents of areas with more severe COVID- 19 transmission and magnified existing social and economic disparities. \(^{36}\) In the US, having lower income was associated with higher risk of depression during the pandemic. \(^{37}\) Our results show a seemingly paradox where lower household income is associated with a greater level of hesitance, while loss of income due to the pandemic is associated with greater likelihood of vaccine acceptance. We suspect the perceived association between vaccination and return to normalcy is stronger among people who lost socioeconomic status due to pandemic- related job loss, while people whose income was low before the pandemic may not perceive a vaccine as making much of a difference in their lives. + +Globally, anxiety and depression increased during the pandemic while positive feelings such as happiness and life satisfaction decreased. \(^{38}\) A global review of COVID- 19 related mental health found anxiety levels of \(26.5\% - 44.6\%\) , depression rates of \(8.1\% - 25\%\) and insomnia levels of \(38\% .^{39}\) Stress levels were found to greatly differ ( \(3.8\%\) and \(68.3\%\) ). In several countries such as Bangladesh, India, and Pakistancases of suicide as a result of COVID- 19 fear have been reported. \(^{38}\) Across a range of timepoints and geographies, people who had COVID- 19 report more symptoms of anxiety, depression, and post- traumatic stress disorder than people without a COVID- 19 diagnosis. \(^{38}\) Similarly, people with pre- existing psychiatric conditions reported worsening of their psychiatric symptoms during the pandemic. \(^{40}\) + +Mental illness has been associated with a higher risk of COVID- 19 related mortality and morbidity, \(^{41}\) yet studies examining the impact of mental health on vaccine hesitancy are scarce. One German + +<--- Page Split ---> + +study did not show a relation between depression or anxiety and vaccine hesitancy. \(^{42}\) This is in line with a Danish study showing that, although people previously diagnosed with mental illnesses reported slightly lower vaccine acceptance compared to the general population (84.8% versus 89.5%), vaccine hesitancy among people with mental illnesses does not seem to be a deterrent to reaching herd immunity. \(^{43}\) In Ireland, by contrast, people who had received treatment for a mental health problem were more accepting of a vaccine, unlike UK respondents who showed no association. \(^{35}\) Our study results suggest that the effects of depression and anxiety are far from universal, with divergent associations between anxiety and depression and vaccine acceptance reported across the 23 countries. A deeper examination of cultural influences should be explored in future research. + +Increased vaccine scepticism may result from the dissemination of erroneous or inaccurate, and often politicized, \(^{17,44}\) information. Misinformation drives vaccine hesitancy, undermining confidence in vaccine safety and efficacy, as well as in equity of availability and access to COVID- 19 vaccines. \(^{45}\) For example, over two- thirds of vaccine videos on YouTube that were analyzed for content accuracy in May 2019 were found to have presented unreliable safety and efficacy information. \(^{46}\) However, in mid- February 2021, despite only 46% of Twitter poll respondents agreeing that all COVID- 19 vaccines are safe, 83% indicated they would accept a vaccine, while only 2% would only agree to accept one if it were mandatory to do so. \(^{47}\) Our results indicate that respondents are willing to trust their doctors' advice. HCWs should play a more direct role in disseminating clear and credible information whenever possible. + +To control the COVID- 19 pandemic, requirements for proof of vaccination, or vaccine mandates, are being considered, for example, to travel internationally or to attend work, school, or indoor events. A 2020 study surveying 1,200 Australians found that 73% of the respondents agreed with vaccine mandates for work, travel, and study. \(^{48}\) In line with our results, a US September 2020 study found + +<--- Page Split ---> + +that approximately half of the general population considered mandatory COVID- 19 vaccination for children attending school acceptable, a perspective that remains universally low in our sample. + +Mandates for adults by state governments were considered acceptable by \(40.9\%\) of the US population, whereas \(47.7\%\) accepted mandates by their employer to attend work. However, our 2020 global study showed that people were potentially more likely to accept voluntary over employer- mandated vaccination. "Choice architecture" that frames vaccination requirements as effective public health and disease prevention and control tools, which one chooses to accept in order to fully participate in society, as opposed to a violation of the individual's right to select medical treatment, may promote incremental vaccine uptake. As vaccines receive full approval from regulatory agencies, this may lead to improved perspectives on safety and efficacy, and vaccination campaigns based on such choice framing could convince more unvaccinated adults and young adults to accept vaccination and increase parental acceptance of vaccination for their children. + +One limitation of correlation analyses using actual vaccination rates is that countries with low vaccine access may produce unreliable results given this extrinsic factor. Additionally, our questionnaire asked about a general COVID- 19 vaccine, whereas several COVID- 19 vaccines, each with different efficacy results and targeted misinformation, are being distributed globally. This study is strengthened by maintaining a sampling methodology that ensures population representativeness between iterations. + +As COVID- 19 vaccination campaigns continue and coverage improves, further challenges will include increasing vaccination among those reporting lower vaccine confidence in addition to expanding equitable vaccine access to low- and middle- income countries. Validation of COVID- VAC confirms the importance of positive perception of vaccine safety, efficacy and equity for vaccine acceptance. Vaccine uptake may be improved through targeted requirements for vaccination in order to + +<--- Page Split ---> + +participate in certain activities. Misinformation may have greater opportunity to spread in settings where access to COVID- 19 vaccination is low, so accurate COVID- 19 vaccine communication should be delivered by trusted sources, like doctors, to promote vaccination and clearly explain its safety and benefits to individuals, families, and communities. + +## Methods + +Study participants + +Participants were recruited by Consensus Strategies using multiple international online panel providers to avoid coverage bias: Dynata provided 22,500 respondents across all 23 countries; and Consensus Strategies provided 500 respondents from Ghana. Respondents' identities were verified using IP addresses and mobile phone numbers. Participants were recruited for the panels via a variety of methods, including online, telephone, and direct mail solicitation and equitably compensated in compliance with ethical standards, varying by country and not exceeding USD 3 per completed survey. No personally identifiable information was collected or stored. + +## Sampling + +Strata included age (18- 29, 30- 39, 40- 49, 50- 59 and 60 years and older); gender (male, female, prefer not to say, and "other"); statistical regions (usually province or state, varies by country); and level of education (based on each country's educational system51), using global data from UNESCO the Organisation for Economic Co- operation and Development, and country data from Sweden, the United Kingdom, and the United States. Educational level was coded into two groups, those who had or had not completed a university degree. The number of participants who could enrol in each of + +<--- Page Split ---> + +these strata was calculated to reflect the distribution in the general population based on census/survey estimates provided by the World Bank and CIA World Factbook. Data were weighted by strata with each stratum requiring a minimum of 50 participants. Sampling was random and is described in detail elsewhere. \(^{51}\) + +## Data collection + +Survey data were collected between 25- 30 June 2021, from an online panel of 23,000 respondents aged \(\geq 18\) years from 23 countries (n=1,000 per country), comprised of those countries included in the 2020 study \(^{2}\) (n=19), augmented by four additional countries with high disease incidence \(^{52}\) and representing regions not represented in the previous study. The 23 countries are: Brazil, Canada, China, Ecuador, France, Germany, Ghana, India, Italy, Kenya, Mexico, Nigeria, Peru, Poland, Russia, Singapore, South Africa, South Korea, Spain, Sweden, Turkey, the United Kingdom (UK) and the United States (US). This study was approved and the survey administered by Emerson College, Boston, USA (institutional review board protocol no. 20- 023- F- E- 6/12- [R1] updated April 12, 2021). + +## Survey Instrument + +The instrument was developed by an expert panel following a comprehensive literature review of COVID- 19 vaccine acceptance studies and the authors' earlier studies of pandemic control measures \(^{51,53,54}\) and vaccination intent. \(^{14,55 - 60}\) The instrument included 1) a 6- item COVID- VAC score representing perceptions of risk, trust, safety, equity, and efficacy, identified via the literature review as important determinants of acceptance of a COVID- 19 vaccine and of routine immunization, 2) vaccination acceptance for self (whether respondent received at least one dose of a COVID- 19 vaccine and if not, whether he/she will take the COVID- 19 vaccine when it is available for them) and vaccination acceptance for their children, 3) vaccination acceptance if reccomended by + +<--- Page Split ---> + +one's employer or doctor, 4) COVID- 19 mandate acceptance (required by: (a) employers and (b) the government and for (c) university students, (d) school children, and (e) indoor activities like auditoriums, concerts, sports events, and (f) international travel), 5) trust in central and local government, 6) experience of anxiety and depression (moderate; 3- 4 days per week, or most or all of the time; 5- 7 days), 7) COVID- 19 experience (self or a family member became ill with COVID- 19, lost a family member to COVID- 19), and 8) demographic variables (age, gender, education, income, loss of income due to COVID- 19 (self- reported as "moderate" or "severe"), and health care worker (HCW) status (Supplemental Information 1). + +## Data analysis + +This study documents vaccine acceptance globally and by country at a point in time approximately six months after the first vaccine was authorized and made available for emergency use. First, vaccine acceptance was assessed using descriptive statistics. The association between a country's vaccine acceptance and current COVID- 19 cases and mortality (per million population), COVID- SCORE- 10 (June 2020),51 COVID- VAC score (June 2021), and vaccine hesitancy were each assessed using Pearson correlations. We also compared the acceptance of a hypothetical vaccine described in our June 2020 study and current COVID- 19 vaccine acceptance now that vaccines are available. Weighted multivariable logistic regressions were used to assess the relationship between vaccine acceptance and socio- demographic variables, COVID- 19 illness experience (personally or a family member), COVID- VAC score, trust in central and local government, and mental health. In addition to adult vaccine acceptance, we investigated attitudes regarding vaccination of children and requirements for proof of vaccination to travel internationally or to attend work, school, or indoor events using descriptive statistics. Finally, vaccine acceptance among HCWs was assessed across all countries combined. Statistical significance was set at alpha=0.05. Analyses were conducted in SAS 9.4. + +<--- Page Split ---> + +## Author Contributions + +JVL, SCR and AEM conceived the study. PFF collected the data. KW was responsible for coding and data analyses with input from TMW. JVL, TMW, CAP and AEM wrote the first draft of the paper. JVL, AEM, TMW, CAP, KW, KR, SCR, JPL, and JH edited subsequent revisions of the draft and approved the final manuscript. + +## Acknowledgements + +The authors would like to thank Patrick F Fox at Consensus Strategies for surveying the countries and reviewing the methodology. + +## Competing Interests Statement + +The authors declare no competing interests. + +## References + +1. World Health Organization. Ten threats to global health in 2019 [Internet]. 2019 [cited 2020 Oct 6]. Available from: https://www.who.int/news-room/spotlight/ten-threats-to-global-health-in-2019 +2. Lazarus JV, Ratzan SC, Palayew A, Gostin LO, Larson HJ, Rabin K, et al. A global survey of potential acceptance of a COVID-19 vaccine. Nat Med [Internet]. 2020; Available from: https://www.nature.com/articles/s41591-020-1124-9 +3. Mathieu E, Ritchie H, Ortiz-Ospina E, Roser M, Hasell J, Appel C, et al. A global database of COVID-19 vaccinations. Nat Hum Behav 2021 57 [Internet]. 2021 May 10 [cited 2021 Aug 3];5(7):947–53. Available from: https://www.nature.com/articles/s41562-021-01122-8 +4. Stojanovic J, Boucher VG, Gagne M, Gupta S, Joyal-Desmarais K, Paduano S, et al. Global Trends and Correlates of COVID-19 Vaccination Hesitancy: Findings from the iCARE Study. Vaccines 2021, Vol 9, Page 661 [Internet]. 2021 Jun 17 [cited 2021 Jul 9];9(6):661. Available from: https://www.mdpi.com/2076-393X/9/6/661/htm +5. Boyon N. COVID-19 vaccination intent is decreasing globally [Internet]. Ipsos and World Economic Forum. 2020 [cited 2021 Jul 29]. Available from: https://www.ipsos.com/en/global-attitudes-covid-19-vaccine-october-2020 +6. Troiano G, Nardi A. Vaccine hesitancy in the era of COVID-19. Public Health. 2021 May 1;194:245–51. +7. Sallam M. COVID-19 Vaccine Hesitancy Worldwide: A Concise Systematic Review of Vaccine Acceptance Rates. Vaccines 2021, Vol 9, Page 160 [Internet]. 2021 Feb 16 [cited 2021 Jul 9];9(2):160. Available from: https://www.mdpi.com/2076-393X/9/2/160/htm +8. Grossman VA. The COVID-19 Vaccine: Why the Hesitancy? J Radiol Nurs. 2021 Jun 1;40(2):116–9. +9. Sallam M, Dabbaseh D, Eid H, Hasan H, Taim D, Al-Mahzoum K, et al. Low COVID-19 Vaccine Acceptance Is Correlated with Conspiracy Beliefs among University Students in Jordan. Int J + +<--- Page Split ---> + +Environ Res Public Health [Internet]. 2021 Mar 1 [cited 2021 Apr 29];18(5):2407. Available from: https://www.mdpi.com/1660- 4601/18/5/2407 10. Loomba S, de Figueiredo A, Piatek SJ, de Graaf K, Larson HJ. Measuring the impact of COVID- 19 vaccine misinformation on vaccination intent in the UK and USA. Nat Hum Behav [Internet]. 2021 Mar 1 [cited 2021 Apr 28];5(3):337- 48. Available from: https://doi.org/10.1038/s41562- 021- 01056- 1 11. Chaccour C, Vilasanjuan R. Infodemic: How Has the Epidemic of Misinformation Affected the Response to COVID- 19? ISGlobal. 2020. 12. Suzanna Awang B, Edlaine F de MV, Ching Sin S, Won Sun C, Supa P, M Tasdik H, et al. Factors Affecting COVID- 19 Vaccine Acceptance: An International Survey among Low- and Middle- Income Countries. Vaccines [Internet]. 2021 May 1 [cited 2021 Jul 9];9(5). Available from: https://pubmed.ncbi.nlm.nih.gov/34067682/ 13. Our World in Data. Coronavirus (COVID- 19) Vaccinations [Internet]. Statistics and Research. 2021 [cited 2021 Apr 20]. Available from: https://ourworldindata.org/covid- vaccinations 14. Lane S, MacDonald NE, Marti M, Dumolard L. Vaccine hesitancy around the globe: Analysis of three years of WHO/UNICEF Joint Reporting Form data- 2015- 2017. Vaccine. 2018 Jun 18;36(26):3861- 7. 15. Cadeddu C, Daugbjerg S, Ricciardi W, Rosano A. Beliefs towards vaccination and trust in the scientific community in Italy. Vaccine [Internet]. 2020 Sep 29 [cited 2021 Jul 21];38(42):6609- 17. Available from: https://pubmed.ncbi.nlm.nih.gov/32788138/ 16. Palamenghi L, Barello S, Boccia S, Graffigna G. Mistrust in biomedical research and vaccine hesitancy: the forefront challenge in the battle against COVID- 19 in Italy. Eur J Epidemiol [Internet]. 2020 Aug 1 [cited 2021 Jul 27];35(8):785- 8. Available from: https://pubmed.ncbi.nlm.nih.gov/32808095/ 17. May T. Anti- Vaxxers, Politicization of Science, and the Need for Trust in Pandemic Response. J Health Commun [Internet]. 2020 [cited 2021 Jul 27];25(10):761- 3. Available from: https://pubmed.ncbi.nlm.nih.gov/33345732/ 18. Bicchieri C, Fatas E, Aldama A, Casas A, Deshpande I, Lauro M, et al. In science we (should) trust: Expectations and compliance across nine countries during the COVID- 19 pandemic. PLoS One [Internet]. 2021 Jun 1 [cited 2021 Jul 21];16(6). Available from: https://pubmed.ncbi.nlm.nih.gov/34086823/ 19. Veit W, Brown R, Earp B. In Science We Trust? Being Honest About the Limits of Medical Research During COVID- 19. Am J Bioeth [Internet]. 2021 [cited 2021 Jul 27];21(1):22- 4. Available from: https://pubmed.ncbi.nlm.nih.gov/33373581/ 20. Funk C, Tyson A, Kennedy B, Johnson C. Science and Scientists Held in High Esteem Across Global Publics [Internet]. 2020 Sep [cited 2021 Jul 20]. Available from: https://www.pewresearch.org/science/2020/09/29/science- and- scientists- held- in- high- esteem- across- global- publics/ 21. Arce JSS, Warren SS, Meriggi NF, Scacco A, McMurry N, Voors M, et al. COVID- 19 vaccine acceptance and hesitancy in low- and middle- income countries. Nat Med 2021 [Internet]. 2021 Jul 16 [cited 2021 Jul 27];1- 10. Available from: https://www.nature.com/articles/s41591- 021- 01454- y 22. Yaqub O, Castle- Clarke S, Sevdalis N, Chataway J. Attitudes to vaccination: A critical review. Soc Sci Med. 2014 Jul 1;112:1- 11. 23. Paterson P, Meurice F, Stanberry LR, Glismann S, Rosenthal SL, Larson HJ. Vaccine hesitancy and healthcare providers. Vaccine [Internet]. 2016 Dec 20 [cited 2021 Apr 28];34(52):6700- 6. Available from: https://pubmed.ncbi.nlm.nih.gov/27810314/ 24. Verger P, et al. Vaccine Hesitancy Among General Practitioners and Its Determinants During Controversies: A National Cross- sectional Survey in France. EBioMedicine [Internet]. 2015 Aug 1 [cited 2021 Jul 27];2(8):891- 7. Available from: https://pubmed.ncbi.nlm.nih.gov/26425696/ 25. Biswas N, Mustapha T, Khubchandani J, Price JH. The Nature and Extent of COVID- 19 + +<--- Page Split ---> + +Vaccination Hesitancy in Healthcare Workers. J Community Health [Internet]. 2021 [cited 2021 Jul 9];1. Available from: /pmc/articles/PMC8056370/26. Picchio CA, Carrasco MG, Sagué- Vilavella M, Rius C. Knowledge, attitudes and beliefs about vaccination in primary healthcare workers involved in the administration of systematic childhood vaccines, Barcelona, 2016/17. Eurosurveillance [Internet]. 2019 Feb 7 [cited 2021 May 20];24(6):1800117. Available from: https://www.eurosurveillance.org/content/10.2807/1560- 7917. ES.2019.24.6.180011727. Elizondo- Alzola U, Carrasco MG, Pinós L, Picchio CA, Rius C, Diez E. Vaccine hesitancy among paediatric nurses: Prevalence and associated factors. PLoS One [Internet]. 2021 May 1 [cited 2021 Jul 27];16(5):e0251735. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.025173528. Dzieciolowska S, Hamel D, Gadio S, Dionne M, Gagnon D, Robitaille L, et al. Covid- 19 vaccine acceptance, hesitancy, and refusal among Canadian healthcare workers: A multicenter survey. Am J Infect Control. 2021 Apr 28;29. Bauernfeind S, Hitzenbichler F, Huppertz G, Zeman F, Koller M, Schmidt B, et al. Brief report: attitudes towards Covid- 19 vaccination among hospital employees in a tertiary care university hospital in Germany in December 2020. Infect 2021 [Internet]. 2021 May 20 [cited 2021 Aug 3];1:1- 5. Available from: https://link.springer.com/article/10.1007/s15010- 021- 01622- 930. Soares P, Rocha JV, Moniz M, Gama A, Laires PA, Pedro AR, et al. Factors Associated with COVID- 19 Vaccine Hesitancy. Vaccines [Internet]. 2021 Mar 22 [cited 2021 Jul 9];9(3):300. Available from: https://www.mdpi.com/2076- 393X/9/3/300/html31. Siegrist M, Zingg A. The role of public trust during pandemics: Implications for crisis communication. Eur Psychol [Internet]. 2014 [cited 2020 Oct 6];19(1):23- 32. Available from: /record/2013- 34232- 00132. Ezeibe CC, Ilo C, Ezeibe EN, Oguonu CN, Nwankwo NA, Ajaero CK, et al. Political distrust and the spread of COVID- 19 in Nigeria. https://doi.org/101080/1744169220201828987 [Internet]. 2020 Dec 1 [cited 2021 Jul 21];15(12):1753- 66. Available from: https://www.tandfonline.com/doi/abs/10.1080/17441692.2020.182898733. Okereke M, Ukor NA, Adebisi YA, Ogunkola IO, Iyagbaye EF, Owhor GA, et al. Impact of COVID- 19 on access to healthcare in low- and middle- income countries: Current evidence and future recommendations. Int J Health Plan Manage [Internet]. 2021 Jan 1 [cited 2021 Jul 26];36(1):13- 7. Available from: https://onlinelibrary.wiley.com/doi/full/10.1002/hpm.306734. Acharya KP, Ghimire TR, Subramanya SH. Access to and equitable distribution of COVID- 19 vaccine in low- income countries. npj Vaccines [Internet]. 2021 Apr 14 [cited 2021 Jul 9];6(1):1- 3. Available from: https://www.nature.com/articles/s41541- 021- 00323- 635. Murphy J, Vallieres F, Bentall RP, Shevlin M, McBride O, Hartman TK, et al. Psychological characteristics associated with COVID- 19 vaccine hesitancy and resistance in Ireland and the United Kingdom. Nat Commun [Internet]. 2021 Dec 1 [cited 2021 May 5];12(1):1- 15. Available from: https://doi.org/10.1038/s41467- 020- 20226- 936. Qian Y, Fan W. Who loses income during the COVID- 19 outbreak? Evidence from China. Res Soc Stratif Mobil. 2020 Aug 1;68:100522.37. Ettman C, Abdalla S, Cohen G, Sampson L, Vivier P, Galea S. Prevalence of Depression Symptoms in US Adults Before and During the COVID- 19 Pandemic. JAMA Netw open [Internet]. 2020 Sep 3 [cited 2021 Jul 9];3(9). Available from: https://pubmed.ncbi.nlm.nih.gov/32876685/38. Hossain M, Tasnim S, Sultana A, Faizah F, Mazumder H, Zou L, et al. Epidemiology of mental health problems in COVID- 19: a review. F1000Research [Internet]. 2020 [cited 2021 Jul 9];9. Available from: https://pubmed.ncbi.nlm.nih.gov/33093946/39. Garcia- Iglesias J, Gomez- Salgado J, Martin- Pereira J, Fagundo- Rivera J, Ayuso- Murillo D, Martinez- Riera J, et al. Impacto del SARS- CoV- 2 (Covid- 19) en la salud mental de los + +<--- Page Split ---> + +profesionales sanitarios: una revisión sistemática. Rev Esp Salud Publica [Internet]. 2020 [cited 2021 Jul 9];94(1):e1- 20. Available from: https://medes.com/publication/152295 +40. Vindegaard N, Benros M. COVID- 19 pandemic and mental health consequences: Systematic review of the current evidence. Brain Behav Immun [Internet]. 2020 Oct 1 [cited 2021 Jul 9];89:531- 42. Available from: https://pubmed.ncbi.nlm.nih.gov/32485289/ +41. Mazereel V, Assche K Van, Detraux J, Hert M De. COVID- 19 vaccination for people with severe mental illness: why, what, and how? The Lancet Psychiatry [Internet]. 2021 May 1 [cited 2021 Jul 9];8(5):444- 50. Available from: http://www.thelancet.com/article/S2215036620305642/fulltext +42. Bendrau A, Plag J, Petzold MB, Strohle A. COVID- 19 vaccine hesitancy and related fears and anxiety. Int Immunopharmacol [Internet]. 2021 Aug 1 [cited 2021 Jul 9];97:107724. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8078903/ +43. Jefsen OH, Kølbæk P, Gil Y, Speed M, Dinesen PT, Sønderskov KM, et al. COVID- 19 vaccine willingness amongst patients with mental illness compared with the general population. Acta Neuropsychiatr [Internet]. 2021 [cited 2021 Jul 9];1- 4. Available from: https://www.cambridge.org/core/journals/acta-neuropsychiatrica/article/abs/covid19-vaccine-willingness-amongst-patients-with- mental- illness- compared- with- the- general- population/949CE2ADF019A3D78F64E704146EE348 +44. Lin C, Tu P, Beitsch LM. Confidence and receptivity for covid- 19 vaccines: A rapid systematic review. Vaccines [Internet]. 2021 Jan 1 [cited 2021 Apr 20];9(1):1- 32. Available from: https://pubmed.ncbi.nlm.nih.gov/33396832/ +45. Islam M, Kamal A, Kabir A, Southern D, Khan S, Hasan S, et al. COVID- 19 vaccine rumors and conspiracy theories: The need for cognitive inoculation against misinformation to improve vaccine adherence. PLoS One [Internet]. 2021 May 1 [cited 2021 Jul 19];16(5). Available from: https://pubmed.ncbi.nlm.nih.gov/33979412/ +46. Murphy M, Nanadiego FA, McCavera L, Nichols C, Kalekas P, Wachs D. Assessing the Validity and Accuracy of Online Videos on Vaccine Health Risks. Clin Pediatr (Phila) [Internet]. 2020 Feb 19 [cited 2021 Jul 9];59(4- 5):458- 66. Available from: https://journals.sagepub.com/doi/abs/10.1177/0009922820905866 +47. Eibensteiner F, Ritschl V, Nawaz F, Fazel S, Tsagkaris C, Kulnik S, et al. People's Willingness to Vaccinate Against COVID- 19 Despite Their Safety Concerns: Twitter Poll Analysis. J Med Internet Res [Internet]. 2021 Apr 1 [cited 2021 Jul 9];23(4). Available from: https://pubmed.ncbi.nlm.nih.gov/33872185/ +48. Smith DT, Attwell K, Evers U. Support for a COVID- 19 vaccine mandate in the face of safety concerns and political affiliations: An Australian study: Politics [Internet]. 2021 May 7 [cited 2021 Jul 9]; Available from: https://journals.sagepub.com/doi/10.1177/0263395711009066 +49. Largent EA, Persad G, Sangenito S, Glickman A, Boyle C, Emanuel EJ. US Public Attitudes Toward COVID- 19 Vaccine Mandates. JAMA Netw Open [Internet]. 2020 Dec 18 [cited 2021 Jul 9];3(12). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7749443/ +50. Dubov A, Phung C. Nudges or mandates? The ethics of mandatory flu vaccination. Vaccine [Internet]. 2015 May 21 [cited 2021 Jul 9];33(22):2530- 5. Available from: https://pubmed.ncbi.nlm.nih.gov/25869886/ +51. Lazarus JV, Ratzan S, Palayew A, Billari FC, Binagwaho A, Kimball S, et al. COVID- SCORE: A global survey to assess public perceptions of government responses to COVID- 19 (COVID- SCORE- 10). Hotchkiss D, editor. PLoS One. 2020 Oct;15(10):e0240011. +52. Worldometer. COVID- 19 data [Internet]. 2020 [cited 2020 Jun 30]. Available from: https://www.worldometers.info/coronavirus/about/#sources +53. Lazarus JV, Binagwaho A, El- Mohandes A, Fielding JE, Larson HJ, Plasència A, et al. Keeping governments accountable: the COVID- 19 Assessment Scorecard (COVID- SCORE). Nat Med [Internet]. 2020 Jun 11 [cited 2020 Jun 19];1- 4. Available from: https://doi.org/10.1038/s41591- 020- 0950- 0 + +<--- Page Split ---> + +54. White TM, Cash-Gibson L, Martin-Moreno JM, Matesanz R, Crespo J, Alfonso-Sanchez JL, et al. COVID-SCORE Spain: Public perceptions of key government COVID-19 control measures. Eur J Public Health [Internet]. 2021 Apr 19 [cited 2021 May 3]; Available from: https://academic.oup.com/eurpub/advance-article/doi/10.1093/eurpub/ckab066/6238158 +55. Larson HJ, Cooper LZ, Eskola J, Katz SL, Ratzan S. Addressing the vaccine confidence gap. Lancet. 2011;378(9790):526-35. +56. MacDonald NE, SAGE Working Group on Vaccine Hesitancy. Vaccine hesitancy: Definition, scope and determinants. Vaccine [Internet]. 2015 Aug 14 [cited 2020 Oct 6];33(34):4161-4. Available from: https://pubmed.ncbi.nlm.nih.gov/25896383/ +57. Larson HJ, Clarke RM, Jarrett C, Eckersberger E, Levine Z, Schulz WS, et al. Measuring trust in vaccination: A systematic review. Hum Vaccines Immunother [Internet]. 2018 Jul 3 [cited 2021 Apr 22];14(7):1599-609. Available from: /pmc/articles/PMC6067893/ +58. Quinn SC, Parmer J, Freimuth VS, Hilyard KM, Musa D, Kim KH. Exploring communication, trust in government, and vaccination intention later in the 2009 H1N1 pandemic: Results of a national survey. Biosecurity and Bioterrorism [Internet]. 2013 Jun 1 [cited 2020 Jul 9];11(2):96-106. Available from: http://www.knowledgenetworks.com +59. Karafillakis E, Larson HJ. The benefit of the doubt or doubts over benefits? A systematic literature review of perceived risks of vaccines in European populations. Vaccine [Internet]. 2017 Sep 5 [cited 2021 Apr 22];35(37):4840-50. Available from: https://pubmed.ncbi.nlm.nih.gov/28760616/ +60. Strategic Advisory Group of Experts on Immunization (SAGE). Report of the SAGE Working Group on Vaccine Hesitancy [Internet]. Geneva; 2014 Oct [cited 2021 Apr 28]. Available from: https://www.who.int/immunization/sage/meetings/2014/october/1_Report_WORKING_GR OUP_vaccine_hesitancy_final.pdf + +<--- Page Split ---> + +# Tables and figures + +Table 1. Sample characteristics + +
CountryBrazilCanadaChinaEcuadorFranceGermanyGhanaIndiaItalyKenyaMexicoNigeriaPeruPolandRussiaSouth KoreaSingaporeSpainSwedenTurkeyUnited KingdomUnited StatesGlobal average
%%%%%%%%%%%%%%%%%%
Age
Groups
18-291916.516.523.516.514.135.624.218.537.923.531.923.31615.82518.420.715.515.722.116.617.1
30-3920.317.722.422.115.215.364.421.22020.722.168.121.517.317.723.421.320.717.416.320.617.215.8
40-491916.523.522.116.516.521.22018.920.620.618.51920.319.92218.617.620.617.817.1
50-5924.116.516.514.717.717.616.716.913.816.216.517.318.717.218.417.117.415.717.616.618.4
\(60+\)17.732.921.217.634.236.516.724.68.817.61830.928.814.12219.531.134.719.131.831.624.3
Sex
Male49.149.351.549.749.248.950.650.448.749.349.850.549.448.146.248.949.849.348.949.648.749.148.949.3
Female50.950.148.549.750.750.549.2485149.950.149.249.951.553.250.649.750.250.749.649.950.45150.2
Prefer not
to		0.60.60.10.60.21.60.30.80.10.30.70.40.60.50.40.50.40.71.40.50.10.54
say/Other
Education (university degree)
No82.673.888888275859185.6978491.584754592.544.967.86775.77965.664.377.6
Yes17.426.21212182515914.43168.51625557.555.132.23324.32134.435.738.7
Income (country median)
More
than32.139.865.930.339.630.217.766.629.836.149.617.940.740.224.465.650.435.437.622.632.3434238.7
Median
+ +<--- Page Split ---> + +
Less than Median No income58.25329.336.753.261.142.819.457.828.242.44838.953.469.318.937.454.152.368.856.94946.146.7
Have you or anyone else in your household experienced a loss in income due to the COVID-19 pandemic?
Severe loss29.716.714.458.914.513.743.457.514.959.933.156.237.310.824.940.119.523.216.413.138.411.917.929.0
Moderate loss42.224.851.536.223.11825.927.241.229.345.220.65016.636.239.740.739.234.623.43724.718.232.4
No28.258.534.1562.568.230.715.243.910.721.723.212.672.538.920.239.937.649.163.524.663.463.838.6
COVID-19 experience
None43.384.996.532.468.885.792.770.373.27850.491.729.852.253.158.594.191.472.567.460.173.67169.2
Self/family member38.3111.548.325.49.45.88.720.115.228.2635.541.442.825.13.83.319.627.331.317.817.721.0
sick
Lost
family member18.34.1219.25.84.91.520.96.76.821.42.334.76.44.116.52.15.37.85.38.68.611.39.8
Health care worker
Yes5.59.897.710.81220.625.45.610.98.919.48.24.72.96.711.27.77.813.89.314.816.510.8
No94.590.29192.389.28879.474.694.489.191.180.691.895.397.193.388.892.392.286.290.785.283.589.2
Mental health
Anxiety25.124.912.321.728.720.69.226.726.124.429.413.339.320.636.63722.720.824.424.144.727.621.925.3
Depression25.323.112.623.123.119.212.91522.432.523.825.128.820.327.230.921.121.622.223.838.725.923.223.6
+ +Note: Anxiety and depression defined as symptoms moderate amount of time (3-4 days) or most or all of the time (5-7 days). + +<--- Page Split ---> +![](images/Figure_2.jpg) + + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Panel c. Vaccine acceptance if recommended by employer or one's doctor among 1) willing to get vaccinated, 2) unsure/unwilling to get vaccinated.
+ +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + + +<--- Page Split ---> +![](images/Figure_unknown_2.jpg) + +
Figure 2. Vaccine acceptance by current cases and mortality
+ +![](images/Figure_unknown_3.jpg) + +
Panel b. Vaccine acceptance and COVID-19 mortality
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Panel c. Vaccine acceptance and COVID-SCORE-10 (June 2020)
+ +![](images/Figure_4.jpg) + +
Panel d. Vaccine acceptance and COVID-VAC
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Panel e. Vaccination rates and hesitancy
+ +![](images/Figure_6.jpg) + +
Figure 3. Vaccine acceptance for children among parents
+ +
Panel b. By respondent's willingness to vaccinate self.
+ +<--- Page Split ---> +![PLACEHOLDER_30_0] + + +<--- Page Split ---> +![PLACEHOLDER_31_0] + +
Figure 4. Correlates of vaccine acceptance-sociodemographic factors and COVID-19 experience.
+ +Adjusted odds ratios (aOR) and 95%CI (log scale) from weighted multivariable logistic regression; reference categories: Female, No university degree, More than median income, No COVID-19 sickness/death, No loss of income. + +<--- Page Split ---> +![PLACEHOLDER_32_0] + +
Figure 5. Correlates of vaccine acceptance-COVID-VAC score, trust in government, anxiety and depression.
+ +Adjusted odds ratios (aOR) and 95%CI (log scale) from weighted multivariable logistic regression, adjusted for socio- demographic factors and COVID- 19 experience. + +<--- Page Split ---> + +Table 2. Vaccine acceptance by HCW status. + +
At least one dose receivedWill take when availableAt least one dose received or will receive when availableUnsure or not willing to take
n%%%%p-valueaOR(95%CI)
Not HCW1984052318317
All HCW329572.419.491.88.1<.0011.72(1.38,2.13)
Physician89185.6a11.2a96.9a3.1a<.001
Nurse61974.5b19.1b93.5b6.5b
Community Health Worker79069.6b22.5b92.2b7.8b
Other healthcare worker99561.6c24.5b86.1c13.9c
+ +p-value based on Chi-squared tests. Different subscripts denote statistically significant pairwise differences. aOR from multivariate logistic model after adjusting for demographic variables, COVID-19 experience and clustering of HWC in countries. + +<--- Page Split ---> +![PLACEHOLDER_34_0] + +
Figure 6. Vaccine mandates acceptance.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementalMaterialssubmit.docx + +<--- Page Split ---> diff --git a/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e_det.mmd b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b56de631805e8954fc6dcbb64dbaaecce17696be --- /dev/null +++ b/preprint/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e/preprint__0e703b6f5c04b85c2c6db02aba31424a86128e0d31c49b97e97c9e8a52fd580e_det.mmd @@ -0,0 +1,447 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 916, 175]]<|/det|> +# COVID-VAC: The second global study of COVID-19 vaccine acceptance + +<|ref|>text<|/ref|><|det|>[[44, 195, 824, 238]]<|/det|> +Jeffrey Lazarus ( Jeffrey.Lazarus@lSGlobal.org ) Barcelona Institute for Global Health (ISGlobal) https://orcid.org/0000- 0001- 9618- 2299 + +<|ref|>text<|/ref|><|det|>[[44, 243, 792, 285]]<|/det|> +Katarzyna Wyka Graduate School of Public Health & Health Policy, City University of New York (CUNY) + +<|ref|>text<|/ref|><|det|>[[44, 290, 904, 353]]<|/det|> +Trenton White Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona https://orcid.org/0000- 0002- 0633- 4445 + +<|ref|>text<|/ref|><|det|>[[44, 358, 904, 401]]<|/det|> +Camila Picchio Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona + +<|ref|>text<|/ref|><|det|>[[44, 405, 880, 448]]<|/det|> +Kenneth Rabin City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +<|ref|>text<|/ref|><|det|>[[44, 453, 880, 496]]<|/det|> +Scott Ratzan City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +<|ref|>text<|/ref|><|det|>[[44, 500, 240, 540]]<|/det|> +Jeanna Parsons Leigh Dalhousie University + +<|ref|>text<|/ref|><|det|>[[44, 546, 240, 586]]<|/det|> +Jia Hu University of Calgary + +<|ref|>text<|/ref|><|det|>[[44, 592, 880, 633]]<|/det|> +Ayman El-Mohandes City University of New York (CUNY) Graduate School of Public Health & Health Policy, New York + +<|ref|>sub_title<|/ref|><|det|>[[44, 674, 101, 691]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 710, 587, 730]]<|/det|> +Keywords: COVID- 19, disease control, SARS- CoV- 2, vaccination + +<|ref|>text<|/ref|><|det|>[[44, 749, 318, 768]]<|/det|> +Posted Date: August 14th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 787, 463, 806]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 780128/v1 + +<|ref|>text<|/ref|><|det|>[[44, 824, 909, 867]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 902, 945, 945]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 31441-x. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[0, 0, 997, 997]]<|/det|> +# 1. 1. 1. 1. 1. 1. 1. 2. 2. 2. 2. 2. 2. 2. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 652, 101]]<|/det|> +## COVID-VAC: The second global study of COVID-19 vaccine acceptance + +<|ref|>text<|/ref|><|det|>[[118, 118, 872, 153]]<|/det|> +Jeffrey V Lazarus1,2\*, Katarzyna Wyka2, Trenton M White1, Camila A Picchio1, Kenneth Rabin2, Scott C Ratzan2, Jeanna Parsons Leigh3, Jia Hu4, Ayman El- Mohandes2 + +<|ref|>text<|/ref|><|det|>[[174, 170, 856, 205]]<|/det|> +1. Barcelona Institute for Global Health (ISGlobal), Hospital Clínic, University of Barcelona, Barcelona, Spain + +<|ref|>text<|/ref|><|det|>[[174, 215, 842, 249]]<|/det|> +2. Graduate School of Public Health & Health Policy, City University of New York (CUNY), New York, New York, United States + +<|ref|>text<|/ref|><|det|>[[175, 259, 726, 275]]<|/det|> +3. School of Health Administration, Dalhousie University, Halifax, Canada + +<|ref|>text<|/ref|><|det|>[[176, 286, 483, 301]]<|/det|> +4. University of Calgary, Calgary, Canada + +<|ref|>text<|/ref|><|det|>[[118, 337, 655, 354]]<|/det|> +\* Corresponding Author: Jeffrey V Lazarus, jeffrey.lazarus@isglobal.org + +<|ref|>sub_title<|/ref|><|det|>[[118, 402, 315, 418]]<|/det|> +## Abstract (150/150 words) + +<|ref|>text<|/ref|><|det|>[[115, 432, 875, 803]]<|/det|> +This one- year follow- up study investigated COVID- 19 vaccine acceptance in 23 countries. In June 2021, 75.2% of respondents reported to accept the vaccine, of which 49% had received at least one vaccine dose and 51% were willing to get it once available to them. Factors associated with COVID- 19 vaccine acceptance include perceived safety and trust in science, as measured by a novel COVID- VAC score, and personal experience with COVID- 19. Males, healthcare workers, well- to- do people, those with university degrees, and those whose physician recommended vaccination were more likely to accept it. Trust in governmental ability to address COVID- 19 was weakly associated with vaccine acceptance. Respondents strongly supported vaccination to travel internationally, but weakly supported vaccinating children for school attendance. Hesitancy rates remain high in LMICs. Health policymakers should aggressively communicate vaccination need and inform accurately about efficacy and safety compared to disease risk, and healthcare workers need a greater vaccine communications role. + +<|ref|>text<|/ref|><|det|>[[118, 852, 593, 868]]<|/det|> +Keywords: COVID- 19, disease control, SARS- CoV- 2, vaccination + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 216, 100]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[115, 146, 875, 390]]<|/det|> +Vaccine hesitancy is a global challenge, declared by the World Health Organization in 2019 as one of the ten greatest global health threats1, and a vital concern since vaccination is the most effective intervention to control the ongoing COVID- 19 pandemic. In June 2020, as most countries were experiencing a second wave of SARS- CoV- 2 transmission, the authors of this study reported potentially low COVID- 19 vaccine acceptance among more than 13,000 respondents in 19 of the hardest- hit countries.2 As of 1 August 2021, 1.07 billion individuals globally had received one dose and another 1.14 billion received two doses of an available COVID- 19 vaccine; however, more than 70% of the world (90% in low- and middle- income countries (LMIC)) remains unvaccinated.3 + +<|ref|>text<|/ref|><|det|>[[115, 434, 880, 677]]<|/det|> +Globally, rates of COVID- 19 vaccine acceptance vary. In a 2021 study of more than 30,000 respondents from eight countries, 27% of participants were hesitant, ranging from 9.6% in Brazil to 47.3% in France.4 Another survey of more than 18,000 respondents from 15 countries reported a range of 54% vaccine acceptance in France to 87% in India.5 A recent review of 15 peer- reviewed studies calculates a COVID- 19 vaccine acceptance rate of 77.6% in the general population and 86.1% among students.6 Another study reported vaccine acceptance ranges from 23.6% in Kuwait to 97% in Ecuador.7 The lowest rates of acceptance to date are reported in the Middle East, Africa, and several European countries (notably France, Italy, Poland, and Russia). + +<|ref|>text<|/ref|><|det|>[[117, 721, 874, 899]]<|/det|> +The main causes of vaccine hesitancy in these studies included concerns about vaccine safety and efficacy, mistrust of governments and health and scientific institutions, individual religious and political beliefs, as well as misinformation and conspiracy theories.8- 11 Most studies reported men to be more likely to accept COVID- 19 vaccines than women.7 Other variables associated with lower vaccine acceptance include living in rural areas and lower income. Conversely, COVID- 19 vaccine acceptance correlated strongly among those who reported greater concern about the effects of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 879, 325]]<|/det|> +COVID- 19 disease and those previously vaccinated against influenza.4 Possessing accurate knowledge about COVID- 19 disease transmissibility and prevention, fear of contracting COVID- 19, and higher income all correlated positively with acceptance, but having a chronic disease and being female did not.12 The association between age and vaccine acceptance varied geographically, with some studies showing greater acceptance among older (above 65 years)4 and others among younger (below 40 years) age groups.12 A survey conducted in nine LMICs in Asia and Africa showed that vaccine acceptance increased with a higher perception of effectiveness; 76.4% accepting a hypothetical 90% effective vaccine increased to 88.8% accepting if 95% effective.12 + +<|ref|>text<|/ref|><|det|>[[115, 370, 879, 675]]<|/det|> +Following a year of substantial but very uneven global availability, administration and acceptance of COVID- 19 vaccines, further investigation is necessary to understand how knowledge, risk perception, trust, and other factors associated with acceptance continue to evolve. The present study was undertaken in June 2021 to assess vaccine acceptance and uptake among 23 countries that represent approximately 60% of the world's population. In addition, comparisons in acceptance rates among the 19 countries in our 2020 study were included. Novel investigations in this 2021 study include testing the predictive value of a new vaccine acceptance score (COVID- VAC), which represents perceptions of risk, trust, safety, equity, and efficacy; the impact of mental health on vaccine acceptance; and acceptance for a range of requirements for proof of vaccination, and vaccination for children. + +<|ref|>sub_title<|/ref|><|det|>[[118, 723, 176, 737]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[118, 786, 288, 801]]<|/det|> +## Sample characteristics + +<|ref|>text<|/ref|><|det|>[[117, 816, 860, 899]]<|/det|> +23,000 participants from 23 countries responded to the survey. Approximately half were female (50.2%), ages 30- 59 (59.9%), and resided in LMICs (52.2%). One- fifth were university graduates (22.4%) (Table 1). Healthcare workers represented 10.8% of respondents. COVID- 19 illness (self or + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 874, 294]]<|/det|> +family) or loss of a family member to COVID- 19 were most common in Ecuador (48.3% and 19.2%), Brazil (38.3% and 18.3) and Peru (35.5% and 34.7%), and least common in China (1.5% and 2%), Singapore (3.3% and 5.3%), South Korea (3.8% and 2.1%), Ghana (5.8% and 1.5%), and Nigeria (6% and 2.3%). Loss of income was highest in lower- middle and upper- middle income countries (range 31.7% in Germany to 95.1% in Ecuador). Reported anxiety prevalence ranged from 9.2% in Ghana to 44.7% in Turkey, and depression was lowest in China (12.6%) and Ghana (12.9%) and highest in Turkey (38.7%). + +<|ref|>text<|/ref|><|det|>[[118, 340, 352, 355]]<|/det|> +Vaccine acceptance, June 2021 + +<|ref|>text<|/ref|><|det|>[[115, 402, 876, 580]]<|/det|> +In June 2021, global COVID- 19 vaccine acceptance was 75.2%, of whom 49% reported having received at least one vaccine dose and 51% were willing to get it once available to them (Figure 1). Reported vaccine acceptance was highest in China (97.6% (of whom 91.9% had received at least one dose)), Canada (79.2% (86.3%)), and the UK (81.2% (82.6%)) and lowest in Russia (51.6% (30.4%)), Nigeria (57% (3.5%)), and Poland (59.3% (74.2%) (Figure 1). The mean difference between these single- dose figures and official rates \(^{13}\) was 2.7%; reliable data for China are unavailable. + +<|ref|>text<|/ref|><|det|>[[115, 624, 881, 900]]<|/det|> +Compared to June 2020, vaccine acceptance in the 19 countries previously surveyed increased from 71.5% to 75.2%, though an opposite trend was observed in South Africa (- 20.9%), the US (- 8.8%), Nigeria (- 8.2%), and Russia (- 3.3%). Vaccine acceptance when recommended by employer increased in all countries (global average increased from 48.1% in 2020 to 64.2% in 2021), except in Germany (- 3%) and South Korea (- 0.2%). Vaccine acceptance when recommended by one's doctor (range 56.4% in Russia to 95.1% in China), was higher in all countries compared to one's employer (range 44.4% in Russia to 96.7% in China). Notably, among respondents unsure/unwilling to vaccinate, potential vaccine acceptance was most likely if recommended by one's doctor, particularly in India (63.5%), Kenya (38%), Brazil (36.8%), Turkey (29.9%), South Korea (28.4%), and Russia (26.5%). Generally, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 856, 164]]<|/det|> +those unsure/unwilling to take the vaccine were more likely to follow their doctors' recommendation \((21.9\% + / - 12.3\%)\) than employers' recommendation \((13.2\% + / - 7.7\%)\) ; \(p = 0.006\) (Figure 1). + +<|ref|>text<|/ref|><|det|>[[115, 210, 877, 390]]<|/det|> +Vaccine acceptance correlated highly with perceptions of risk, trust, safety, equity, and efficacy, measured using COVID- VAC (June 2021: \(r = 0.85\) , \(p< 0.001\) ). In validation analyses, COVID- VAC was unidimensional with a Cronbach's alpha of 0.86. Vaccine acceptance weakly correlated with countries' current COVID- 19 case burden \((r = - 0.13\) , \(p > 0.05\) ), mortality \((r = - 0.25\) , \(p > 0.05\) ), and approval of government responses to the pandemic in June 2020, measured using COVID- SCORE- \(10^{11}\) \((r = 0.25\) , \(p > 0.05\) ) (Figure 2). + +<|ref|>text<|/ref|><|det|>[[115, 434, 875, 581]]<|/det|> +Countries' vaccination rates \(^{13}\) and hesitancy showed moderate negative association \((r = 0.45\) , \(p = 0.034\) ) (Figure 2). Generally, countries in Africa had low vaccination uptake and high hesitancy; whereas India, South Korea, and countries in South America had higher uptake and lower hesitancy. Countries with \(40\% +\) uptake had similar hesitancy to the latter group, with the exception of Poland, France and the US, which had high hesitancy. + +<|ref|>text<|/ref|><|det|>[[115, 625, 831, 740]]<|/det|> +Parents' willingness to vaccinate children was highest in China \((95\%)\) , Brazil \((91.3\%)\) , Ecuador \((85.9\%)\) , Peru \((85.1\%)\) and lowest in Russia \((35.5\%)\) , Poland \((46.3\%)\) and France \((48.5\%)\) . In all countries, willingness to vaccinate one's children was significantly greater among parents who accepted the vaccine \((p< 0.001)\) (Figure 3). + +<|ref|>text<|/ref|><|det|>[[118, 786, 445, 802]]<|/det|> +Correlates of vaccine acceptance June 2021 + +<|ref|>text<|/ref|><|det|>[[117, 848, 860, 899]]<|/det|> +In multivariable models, vaccine acceptance was associated with older respondents in Canada, France, Germany, South Korea, Sweden, UK, and US (aOR range, 1.02- 1.04) (Figure 4) and younger + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 883, 455]]<|/det|> +ones in Nigeria (aOR=0.89) and Ghana (aOR=0.93). Male gender was significantly positively correlated in Ghana, Nigeria, South Africa, and the US (aOR range 1.51- 8.84) and female gender in Kenya and Peru (aOR=0.39 and aOR=0.32). Having a university degree was significantly associated with vaccine acceptance in France, Mexico, South Africa, and the US (aOR range, 1.75- 5.55). Personal or family COVID- 19 illness or death of a family member due to COVID- 19 was strongly associated with vaccine acceptance, with statistically significant associations with personal or family illness in Germany, Peru, Poland, Turkey, Singapore, UK (aOR range 1.89- 5.72) and with loss of a family member in Brazil, France, Germany, Ghana, Mexico, Peru, Poland, Sweden, Turkey and the UK (aOR range 2.75- 9.25). Having low or no income was associated with less vaccine acceptance in Brazil, France, Germany, Nigeria, Peru, Russia, South Korea, Spain, Turkey, and the US (aOR range 0.08- 0.55). Moderate or severe income loss due to the pandemic was positively associated with vaccine acceptance in Ghana, Nigeria, Poland, and the US (aOR range 2.01- 6.27). + +<|ref|>text<|/ref|><|det|>[[115, 496, 876, 900]]<|/det|> +Adjusted for socio- demographic factors and COVID- 19 experience, the COVID- VAC score was the most salient and consistent positive correlate of vaccine acceptance in all countries (aOR range, 1.41- 3.44) (Figure 5). The strongest correlation was with "I trust the science behind the COVID- 19 vaccines" and "safety of the vaccines." The lowest correlation was with "I trust my government is able to deliver the vaccine to everyone and everywhere in my country" (Supplemental Table 1). In univariable analyses, vaccine acceptance was significantly higher among respondents who reported that they trust their central and/or local government. This finding was not universal, however, and was not found in Brazil, China, Ecuador, and Peru (Supplemental Figure 1). After adjustment, trust in central government had a significantly positive association in Russia and Singapore (aOR=2.05 and aOR=5.28) and trust in local government had a negative association in South Africa (aOR=0.44). In Poland, vaccine acceptance was negatively correlated with trust in central government (aOR=.48) and positively correlated with trust in local government (aOR=2.9). Finally, vaccine acceptance positively correlated with anxiety in Canada (aOR=4.59) and Peru (aOR=3.64) and negatively + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 876, 133]]<|/det|> +correlated in South Korea (aOR=.43) and the US (aOR=.31), while depression was a negative factor in Canada (aOR=.21) and South Africa (aOR=.53) (Figure 5). + +<|ref|>text<|/ref|><|det|>[[118, 180, 321, 196]]<|/det|> +Healthcare workers (HCW) + +<|ref|>text<|/ref|><|det|>[[116, 242, 870, 357]]<|/det|> +Vaccine acceptance was significantly higher among HCWs globally compared to non- HCWs (91.8% vs. 83%, p<0.001, aOR=1.72, 95%CI [1.38, 2.13]) (Table 2). Vaccine hesitancy was significantly lower among physicians (3.1%) followed by nurses (6.5%), community health workers (7.8%), and other HCWs (14.0%). + +<|ref|>sub_title<|/ref|><|det|>[[119, 403, 375, 419]]<|/det|> +## Proof of vaccination requirements + +<|ref|>text<|/ref|><|det|>[[115, 433, 876, 707]]<|/det|> +Support for requiring vaccines was highest in China, India, Peru and South Korea and lowest in Poland, Russia and Germany. Strong support was reported globally (74.4%) for proof of vaccination to travel internationally, particularly in China (93.3%), South Korea (87.8%), India (87.7%), Brazil (86.4%), and Ecuador (83%), with less support in Poland (52.7%) and Russia (52.5%). Overall support was 63.3% and 63.1% for vaccination to attend university and indoor activities, respectively. These ranged from lows in Russia of 33.1% for university and 37.6% for indoor activitisteos highs in China of 95.4% and 89.7%. Support for governments or employers to require vaccination followed similar trends. Agreement with mandates to vaccinate school children received the lowest support overall (58.2%), ranging from 24.4% in Russia to 86.6% in China (Figure 6). + +<|ref|>sub_title<|/ref|><|det|>[[118, 754, 201, 769]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[117, 817, 877, 899]]<|/det|> +Reported COVID- 19 vaccine acceptance, measured by intention to get vaccinated or having received at least one dose of an available vaccine, increased over the last year in 15 of the 19 countries studied both years and was 75.2% for the 23 countries studied in 2021, still below what is needed to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 83, 864, 293]]<|/det|> +achieve global herd immunity (80- 85%). Among the six- item COVID- VAC score, positive perceptions of safety and trust in the science behind vaccine development were the strongest predictors of acceptance, while perceptions of efficacy, benefit, and equity also held strong associations. Other determinants of vaccine acceptance varied by country; personal experience with COVID- 19 (e.g. sickness or loss of a family member) predominated, but other demographic characteristics (e.g. gender, education and income) showed varying degrees of association, and government trust was associated with vaccine acceptance in a few countries. + +<|ref|>text<|/ref|><|det|>[[116, 339, 877, 485]]<|/det|> +In order to improve the global vaccination rate, some countries may require people to present proof of vaccination to attend work, school, or indoor activities and events. Our results found the strongest support for requirements targeting international travellers and support was weakest for requirements for school children. Importantly, vaccine acceptance improves among those indicating unwillingness to vaccinate if it is recommended by a doctor, and to a lesser extent by an employer. + +<|ref|>text<|/ref|><|det|>[[116, 529, 872, 677]]<|/det|> +Misperceptions of vaccines as having high risks and low benefits drive vaccine hesitancy,14 which may also be influenced by lower trust in the science behind vaccine research and production.15- 19 In late 2019 and early 2020 the majority of people in most countries preceived the importance of science and trust in scientists positively.20 Our results corroborate these findings where perceptions of trust in science and vaccine safety were most predictive of vaccine acceptance. + +<|ref|>text<|/ref|><|det|>[[116, 721, 876, 900]]<|/det|> +Healthcare workers are in a privileged position of trust and are therefore important sources of reliable and accurate vaccine information.21- 24 Among HCWs, COVID- 19 vaccine hesitancy ranged between 4.3% and 72%, with an average of 22.5%.25 Similar to the general population, the main reasons for vaccine hesitancy among HCWs were concerns about efficacy, safety, and potential side- effects. Demographic factors among HCWs such as male gender, older age and holding a doctoral degree were positively associated with acceptance of COVID- 19 vaccines.25 Our study aligns with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 867, 198]]<|/det|> +similarly recorded trends of higher rates of vaccination acceptance among physicians compared to nurses. \(^{26 - 29}\) Among hesitant respondents in this survey, the advice of their physician was the factor most likely to change their minds, followed by the recommendation of their employer. This effect could be amplified if the physician had personally received the vaccine. + +<|ref|>text<|/ref|><|det|>[[115, 240, 875, 743]]<|/det|> +In June 2020, trust in government to successfully address unexpected health threats, including the COVID- 19 pandemic, was one of the strongest factors associated with acceptance of an (at that time) unavailable vaccine. \(^{2}\) A 2020 study in Portugal, where 56% reported they would wait to take the COVID- 19 vaccine and 9% would refuse, showed that vaccine hesitancy was related to a poor perception of government and health service response as well as a lack of trust in the information provided, \(^{30}\) which is consistent with previous pandemic research. \(^{31}\) Yet, after accounting for sociodemographic, vaccine- specific, and COVID- 19 experience variables, our study found trust in central government to be significantly associated with vaccine acceptance only in Russia and Singapore and for local government in only Poland. It is notable also that Nigeria reported low vaccine acceptance and high distrust in governmental ability to respond to COVID- 19, \(^{32}\) but the association was not significant. This dissonance between trust in government and vaccine acceptance in our study could be related to the population's general dissatisfaction with government responses to the pandemic and its economic consequences, with vaccine acceptance being independent of such sentiments and more a reflection of personal experiences with COVID- 19 illness or loss of life and livelihood. Our results confirm that direct experience for self or family with the illness, and/or loss of a family member to the disease are independently associated with vaccine acceptance. + +<|ref|>text<|/ref|><|det|>[[117, 785, 844, 900]]<|/det|> +As health systems in LMICs struggle to address COVID- 19, \(^{33}\) vaccine access has become cause for national frustrations. Most LMICs have been slow to receive and distribute vaccines, which are significantly more available in high- income countries, prompting critiques of global vaccine inequity. \(^{34}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 114, 876, 485]]<|/det|> +Most recent studies do not investigate income as a potential influencer on vaccine acceptance. A study conducted in Portugal found people who lost income during the pandemic were more hesitant. \(^{30}\) An Irish study showed that people with lower income were also more vaccine hesitant. \(^{35}\) In China, loss of income was greater among residents of areas with more severe COVID- 19 transmission and magnified existing social and economic disparities. \(^{36}\) In the US, having lower income was associated with higher risk of depression during the pandemic. \(^{37}\) Our results show a seemingly paradox where lower household income is associated with a greater level of hesitance, while loss of income due to the pandemic is associated with greater likelihood of vaccine acceptance. We suspect the perceived association between vaccination and return to normalcy is stronger among people who lost socioeconomic status due to pandemic- related job loss, while people whose income was low before the pandemic may not perceive a vaccine as making much of a difference in their lives. + +<|ref|>text<|/ref|><|det|>[[115, 529, 880, 804]]<|/det|> +Globally, anxiety and depression increased during the pandemic while positive feelings such as happiness and life satisfaction decreased. \(^{38}\) A global review of COVID- 19 related mental health found anxiety levels of \(26.5\% - 44.6\%\) , depression rates of \(8.1\% - 25\%\) and insomnia levels of \(38\% .^{39}\) Stress levels were found to greatly differ ( \(3.8\%\) and \(68.3\%\) ). In several countries such as Bangladesh, India, and Pakistancases of suicide as a result of COVID- 19 fear have been reported. \(^{38}\) Across a range of timepoints and geographies, people who had COVID- 19 report more symptoms of anxiety, depression, and post- traumatic stress disorder than people without a COVID- 19 diagnosis. \(^{38}\) Similarly, people with pre- existing psychiatric conditions reported worsening of their psychiatric symptoms during the pandemic. \(^{40}\) + +<|ref|>text<|/ref|><|det|>[[117, 848, 868, 899]]<|/det|> +Mental illness has been associated with a higher risk of COVID- 19 related mortality and morbidity, \(^{41}\) yet studies examining the impact of mental health on vaccine hesitancy are scarce. One German + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 868, 388]]<|/det|> +study did not show a relation between depression or anxiety and vaccine hesitancy. \(^{42}\) This is in line with a Danish study showing that, although people previously diagnosed with mental illnesses reported slightly lower vaccine acceptance compared to the general population (84.8% versus 89.5%), vaccine hesitancy among people with mental illnesses does not seem to be a deterrent to reaching herd immunity. \(^{43}\) In Ireland, by contrast, people who had received treatment for a mental health problem were more accepting of a vaccine, unlike UK respondents who showed no association. \(^{35}\) Our study results suggest that the effects of depression and anxiety are far from universal, with divergent associations between anxiety and depression and vaccine acceptance reported across the 23 countries. A deeper examination of cultural influences should be explored in future research. + +<|ref|>text<|/ref|><|det|>[[115, 434, 878, 740]]<|/det|> +Increased vaccine scepticism may result from the dissemination of erroneous or inaccurate, and often politicized, \(^{17,44}\) information. Misinformation drives vaccine hesitancy, undermining confidence in vaccine safety and efficacy, as well as in equity of availability and access to COVID- 19 vaccines. \(^{45}\) For example, over two- thirds of vaccine videos on YouTube that were analyzed for content accuracy in May 2019 were found to have presented unreliable safety and efficacy information. \(^{46}\) However, in mid- February 2021, despite only 46% of Twitter poll respondents agreeing that all COVID- 19 vaccines are safe, 83% indicated they would accept a vaccine, while only 2% would only agree to accept one if it were mandatory to do so. \(^{47}\) Our results indicate that respondents are willing to trust their doctors' advice. HCWs should play a more direct role in disseminating clear and credible information whenever possible. + +<|ref|>text<|/ref|><|det|>[[117, 785, 873, 899]]<|/det|> +To control the COVID- 19 pandemic, requirements for proof of vaccination, or vaccine mandates, are being considered, for example, to travel internationally or to attend work, school, or indoor events. A 2020 study surveying 1,200 Australians found that 73% of the respondents agreed with vaccine mandates for work, travel, and study. \(^{48}\) In line with our results, a US September 2020 study found + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 860, 134]]<|/det|> +that approximately half of the general population considered mandatory COVID- 19 vaccination for children attending school acceptable, a perspective that remains universally low in our sample. + +<|ref|>text<|/ref|><|det|>[[115, 178, 866, 516]]<|/det|> +Mandates for adults by state governments were considered acceptable by \(40.9\%\) of the US population, whereas \(47.7\%\) accepted mandates by their employer to attend work. However, our 2020 global study showed that people were potentially more likely to accept voluntary over employer- mandated vaccination. "Choice architecture" that frames vaccination requirements as effective public health and disease prevention and control tools, which one chooses to accept in order to fully participate in society, as opposed to a violation of the individual's right to select medical treatment, may promote incremental vaccine uptake. As vaccines receive full approval from regulatory agencies, this may lead to improved perspectives on safety and efficacy, and vaccination campaigns based on such choice framing could convince more unvaccinated adults and young adults to accept vaccination and increase parental acceptance of vaccination for their children. + +<|ref|>text<|/ref|><|det|>[[115, 529, 877, 707]]<|/det|> +One limitation of correlation analyses using actual vaccination rates is that countries with low vaccine access may produce unreliable results given this extrinsic factor. Additionally, our questionnaire asked about a general COVID- 19 vaccine, whereas several COVID- 19 vaccines, each with different efficacy results and targeted misinformation, are being distributed globally. This study is strengthened by maintaining a sampling methodology that ensures population representativeness between iterations. + +<|ref|>text<|/ref|><|det|>[[115, 752, 881, 899]]<|/det|> +As COVID- 19 vaccination campaigns continue and coverage improves, further challenges will include increasing vaccination among those reporting lower vaccine confidence in addition to expanding equitable vaccine access to low- and middle- income countries. Validation of COVID- VAC confirms the importance of positive perception of vaccine safety, efficacy and equity for vaccine acceptance. Vaccine uptake may be improved through targeted requirements for vaccination in order to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 866, 197]]<|/det|> +participate in certain activities. Misinformation may have greater opportunity to spread in settings where access to COVID- 19 vaccination is low, so accurate COVID- 19 vaccine communication should be delivered by trusted sources, like doctors, to promote vaccination and clearly explain its safety and benefits to individuals, families, and communities. + +<|ref|>sub_title<|/ref|><|det|>[[118, 276, 190, 291]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[118, 340, 257, 355]]<|/det|> +Study participants + +<|ref|>text<|/ref|><|det|>[[116, 401, 870, 612]]<|/det|> +Participants were recruited by Consensus Strategies using multiple international online panel providers to avoid coverage bias: Dynata provided 22,500 respondents across all 23 countries; and Consensus Strategies provided 500 respondents from Ghana. Respondents' identities were verified using IP addresses and mobile phone numbers. Participants were recruited for the panels via a variety of methods, including online, telephone, and direct mail solicitation and equitably compensated in compliance with ethical standards, varying by country and not exceeding USD 3 per completed survey. No personally identifiable information was collected or stored. + +<|ref|>sub_title<|/ref|><|det|>[[118, 659, 191, 674]]<|/det|> +## Sampling + +<|ref|>text<|/ref|><|det|>[[116, 721, 880, 899]]<|/det|> +Strata included age (18- 29, 30- 39, 40- 49, 50- 59 and 60 years and older); gender (male, female, prefer not to say, and "other"); statistical regions (usually province or state, varies by country); and level of education (based on each country's educational system51), using global data from UNESCO the Organisation for Economic Co- operation and Development, and country data from Sweden, the United Kingdom, and the United States. Educational level was coded into two groups, those who had or had not completed a university degree. The number of participants who could enrol in each of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 870, 197]]<|/det|> +these strata was calculated to reflect the distribution in the general population based on census/survey estimates provided by the World Bank and CIA World Factbook. Data were weighted by strata with each stratum requiring a minimum of 50 participants. Sampling was random and is described in detail elsewhere. \(^{51}\) + +<|ref|>sub_title<|/ref|><|det|>[[118, 244, 234, 259]]<|/det|> +## Data collection + +<|ref|>text<|/ref|><|det|>[[115, 305, 866, 550]]<|/det|> +Survey data were collected between 25- 30 June 2021, from an online panel of 23,000 respondents aged \(\geq 18\) years from 23 countries (n=1,000 per country), comprised of those countries included in the 2020 study \(^{2}\) (n=19), augmented by four additional countries with high disease incidence \(^{52}\) and representing regions not represented in the previous study. The 23 countries are: Brazil, Canada, China, Ecuador, France, Germany, Ghana, India, Italy, Kenya, Mexico, Nigeria, Peru, Poland, Russia, Singapore, South Africa, South Korea, Spain, Sweden, Turkey, the United Kingdom (UK) and the United States (US). This study was approved and the survey administered by Emerson College, Boston, USA (institutional review board protocol no. 20- 023- F- E- 6/12- [R1] updated April 12, 2021). + +<|ref|>sub_title<|/ref|><|det|>[[118, 604, 257, 619]]<|/det|> +## Survey Instrument + +<|ref|>text<|/ref|><|det|>[[115, 666, 876, 908]]<|/det|> +The instrument was developed by an expert panel following a comprehensive literature review of COVID- 19 vaccine acceptance studies and the authors' earlier studies of pandemic control measures \(^{51,53,54}\) and vaccination intent. \(^{14,55 - 60}\) The instrument included 1) a 6- item COVID- VAC score representing perceptions of risk, trust, safety, equity, and efficacy, identified via the literature review as important determinants of acceptance of a COVID- 19 vaccine and of routine immunization, 2) vaccination acceptance for self (whether respondent received at least one dose of a COVID- 19 vaccine and if not, whether he/she will take the COVID- 19 vaccine when it is available for them) and vaccination acceptance for their children, 3) vaccination acceptance if reccomended by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 325]]<|/det|> +one's employer or doctor, 4) COVID- 19 mandate acceptance (required by: (a) employers and (b) the government and for (c) university students, (d) school children, and (e) indoor activities like auditoriums, concerts, sports events, and (f) international travel), 5) trust in central and local government, 6) experience of anxiety and depression (moderate; 3- 4 days per week, or most or all of the time; 5- 7 days), 7) COVID- 19 experience (self or a family member became ill with COVID- 19, lost a family member to COVID- 19), and 8) demographic variables (age, gender, education, income, loss of income due to COVID- 19 (self- reported as "moderate" or "severe"), and health care worker (HCW) status (Supplemental Information 1). + +<|ref|>sub_title<|/ref|><|det|>[[118, 372, 221, 387]]<|/det|> +## Data analysis + +<|ref|>text<|/ref|><|det|>[[115, 432, 875, 897]]<|/det|> +This study documents vaccine acceptance globally and by country at a point in time approximately six months after the first vaccine was authorized and made available for emergency use. First, vaccine acceptance was assessed using descriptive statistics. The association between a country's vaccine acceptance and current COVID- 19 cases and mortality (per million population), COVID- SCORE- 10 (June 2020),51 COVID- VAC score (June 2021), and vaccine hesitancy were each assessed using Pearson correlations. We also compared the acceptance of a hypothetical vaccine described in our June 2020 study and current COVID- 19 vaccine acceptance now that vaccines are available. Weighted multivariable logistic regressions were used to assess the relationship between vaccine acceptance and socio- demographic variables, COVID- 19 illness experience (personally or a family member), COVID- VAC score, trust in central and local government, and mental health. In addition to adult vaccine acceptance, we investigated attitudes regarding vaccination of children and requirements for proof of vaccination to travel internationally or to attend work, school, or indoor events using descriptive statistics. Finally, vaccine acceptance among HCWs was assessed across all countries combined. Statistical significance was set at alpha=0.05. Analyses were conducted in SAS 9.4. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 111, 284, 126]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[117, 137, 880, 250]]<|/det|> +JVL, SCR and AEM conceived the study. PFF collected the data. KW was responsible for coding and data analyses with input from TMW. JVL, TMW, CAP and AEM wrote the first draft of the paper. JVL, AEM, TMW, CAP, KW, KR, SCR, JPL, and JH edited subsequent revisions of the draft and approved the final manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[118, 315, 271, 330]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[118, 342, 880, 375]]<|/det|> +The authors would like to thank Patrick F Fox at Consensus Strategies for surveying the countries and reviewing the methodology. + +<|ref|>sub_title<|/ref|><|det|>[[118, 413, 361, 428]]<|/det|> +## Competing Interests Statement + +<|ref|>text<|/ref|><|det|>[[118, 440, 451, 455]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[118, 494, 206, 508]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[115, 510, 876, 911]]<|/det|> +1. World Health Organization. Ten threats to global health in 2019 [Internet]. 2019 [cited 2020 Oct 6]. Available from: https://www.who.int/news-room/spotlight/ten-threats-to-global-health-in-2019 +2. Lazarus JV, Ratzan SC, Palayew A, Gostin LO, Larson HJ, Rabin K, et al. A global survey of potential acceptance of a COVID-19 vaccine. Nat Med [Internet]. 2020; Available from: https://www.nature.com/articles/s41591-020-1124-9 +3. Mathieu E, Ritchie H, Ortiz-Ospina E, Roser M, Hasell J, Appel C, et al. A global database of COVID-19 vaccinations. Nat Hum Behav 2021 57 [Internet]. 2021 May 10 [cited 2021 Aug 3];5(7):947–53. Available from: https://www.nature.com/articles/s41562-021-01122-8 +4. Stojanovic J, Boucher VG, Gagne M, Gupta S, Joyal-Desmarais K, Paduano S, et al. Global Trends and Correlates of COVID-19 Vaccination Hesitancy: Findings from the iCARE Study. Vaccines 2021, Vol 9, Page 661 [Internet]. 2021 Jun 17 [cited 2021 Jul 9];9(6):661. Available from: https://www.mdpi.com/2076-393X/9/6/661/htm +5. Boyon N. COVID-19 vaccination intent is decreasing globally [Internet]. Ipsos and World Economic Forum. 2020 [cited 2021 Jul 29]. Available from: https://www.ipsos.com/en/global-attitudes-covid-19-vaccine-october-2020 +6. Troiano G, Nardi A. Vaccine hesitancy in the era of COVID-19. Public Health. 2021 May 1;194:245–51. +7. Sallam M. COVID-19 Vaccine Hesitancy Worldwide: A Concise Systematic Review of Vaccine Acceptance Rates. Vaccines 2021, Vol 9, Page 160 [Internet]. 2021 Feb 16 [cited 2021 Jul 9];9(2):160. Available from: https://www.mdpi.com/2076-393X/9/2/160/htm +8. Grossman VA. The COVID-19 Vaccine: Why the Hesitancy? J Radiol Nurs. 2021 Jun 1;40(2):116–9. +9. Sallam M, Dabbaseh D, Eid H, Hasan H, Taim D, Al-Mahzoum K, et al. Low COVID-19 Vaccine Acceptance Is Correlated with Conspiracy Beliefs among University Students in Jordan. Int J + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 884, 901]]<|/det|> +Environ Res Public Health [Internet]. 2021 Mar 1 [cited 2021 Apr 29];18(5):2407. Available from: https://www.mdpi.com/1660- 4601/18/5/2407 10. Loomba S, de Figueiredo A, Piatek SJ, de Graaf K, Larson HJ. Measuring the impact of COVID- 19 vaccine misinformation on vaccination intent in the UK and USA. Nat Hum Behav [Internet]. 2021 Mar 1 [cited 2021 Apr 28];5(3):337- 48. Available from: https://doi.org/10.1038/s41562- 021- 01056- 1 11. Chaccour C, Vilasanjuan R. Infodemic: How Has the Epidemic of Misinformation Affected the Response to COVID- 19? ISGlobal. 2020. 12. Suzanna Awang B, Edlaine F de MV, Ching Sin S, Won Sun C, Supa P, M Tasdik H, et al. Factors Affecting COVID- 19 Vaccine Acceptance: An International Survey among Low- and Middle- Income Countries. Vaccines [Internet]. 2021 May 1 [cited 2021 Jul 9];9(5). Available from: https://pubmed.ncbi.nlm.nih.gov/34067682/ 13. Our World in Data. Coronavirus (COVID- 19) Vaccinations [Internet]. Statistics and Research. 2021 [cited 2021 Apr 20]. Available from: https://ourworldindata.org/covid- vaccinations 14. Lane S, MacDonald NE, Marti M, Dumolard L. Vaccine hesitancy around the globe: Analysis of three years of WHO/UNICEF Joint Reporting Form data- 2015- 2017. Vaccine. 2018 Jun 18;36(26):3861- 7. 15. Cadeddu C, Daugbjerg S, Ricciardi W, Rosano A. Beliefs towards vaccination and trust in the scientific community in Italy. Vaccine [Internet]. 2020 Sep 29 [cited 2021 Jul 21];38(42):6609- 17. Available from: https://pubmed.ncbi.nlm.nih.gov/32788138/ 16. Palamenghi L, Barello S, Boccia S, Graffigna G. Mistrust in biomedical research and vaccine hesitancy: the forefront challenge in the battle against COVID- 19 in Italy. Eur J Epidemiol [Internet]. 2020 Aug 1 [cited 2021 Jul 27];35(8):785- 8. Available from: https://pubmed.ncbi.nlm.nih.gov/32808095/ 17. May T. Anti- Vaxxers, Politicization of Science, and the Need for Trust in Pandemic Response. J Health Commun [Internet]. 2020 [cited 2021 Jul 27];25(10):761- 3. Available from: https://pubmed.ncbi.nlm.nih.gov/33345732/ 18. Bicchieri C, Fatas E, Aldama A, Casas A, Deshpande I, Lauro M, et al. In science we (should) trust: Expectations and compliance across nine countries during the COVID- 19 pandemic. PLoS One [Internet]. 2021 Jun 1 [cited 2021 Jul 21];16(6). Available from: https://pubmed.ncbi.nlm.nih.gov/34086823/ 19. Veit W, Brown R, Earp B. In Science We Trust? Being Honest About the Limits of Medical Research During COVID- 19. Am J Bioeth [Internet]. 2021 [cited 2021 Jul 27];21(1):22- 4. Available from: https://pubmed.ncbi.nlm.nih.gov/33373581/ 20. Funk C, Tyson A, Kennedy B, Johnson C. Science and Scientists Held in High Esteem Across Global Publics [Internet]. 2020 Sep [cited 2021 Jul 20]. Available from: https://www.pewresearch.org/science/2020/09/29/science- and- scientists- held- in- high- esteem- across- global- publics/ 21. Arce JSS, Warren SS, Meriggi NF, Scacco A, McMurry N, Voors M, et al. COVID- 19 vaccine acceptance and hesitancy in low- and middle- income countries. Nat Med 2021 [Internet]. 2021 Jul 16 [cited 2021 Jul 27];1- 10. Available from: https://www.nature.com/articles/s41591- 021- 01454- y 22. Yaqub O, Castle- Clarke S, Sevdalis N, Chataway J. Attitudes to vaccination: A critical review. Soc Sci Med. 2014 Jul 1;112:1- 11. 23. Paterson P, Meurice F, Stanberry LR, Glismann S, Rosenthal SL, Larson HJ. Vaccine hesitancy and healthcare providers. Vaccine [Internet]. 2016 Dec 20 [cited 2021 Apr 28];34(52):6700- 6. Available from: https://pubmed.ncbi.nlm.nih.gov/27810314/ 24. Verger P, et al. Vaccine Hesitancy Among General Practitioners and Its Determinants During Controversies: A National Cross- sectional Survey in France. EBioMedicine [Internet]. 2015 Aug 1 [cited 2021 Jul 27];2(8):891- 7. Available from: https://pubmed.ncbi.nlm.nih.gov/26425696/ 25. Biswas N, Mustapha T, Khubchandani J, Price JH. The Nature and Extent of COVID- 19 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 884, 902]]<|/det|> +Vaccination Hesitancy in Healthcare Workers. J Community Health [Internet]. 2021 [cited 2021 Jul 9];1. Available from: /pmc/articles/PMC8056370/26. Picchio CA, Carrasco MG, Sagué- Vilavella M, Rius C. Knowledge, attitudes and beliefs about vaccination in primary healthcare workers involved in the administration of systematic childhood vaccines, Barcelona, 2016/17. Eurosurveillance [Internet]. 2019 Feb 7 [cited 2021 May 20];24(6):1800117. Available from: https://www.eurosurveillance.org/content/10.2807/1560- 7917. ES.2019.24.6.180011727. Elizondo- Alzola U, Carrasco MG, Pinós L, Picchio CA, Rius C, Diez E. Vaccine hesitancy among paediatric nurses: Prevalence and associated factors. PLoS One [Internet]. 2021 May 1 [cited 2021 Jul 27];16(5):e0251735. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.025173528. Dzieciolowska S, Hamel D, Gadio S, Dionne M, Gagnon D, Robitaille L, et al. Covid- 19 vaccine acceptance, hesitancy, and refusal among Canadian healthcare workers: A multicenter survey. Am J Infect Control. 2021 Apr 28;29. Bauernfeind S, Hitzenbichler F, Huppertz G, Zeman F, Koller M, Schmidt B, et al. Brief report: attitudes towards Covid- 19 vaccination among hospital employees in a tertiary care university hospital in Germany in December 2020. Infect 2021 [Internet]. 2021 May 20 [cited 2021 Aug 3];1:1- 5. Available from: https://link.springer.com/article/10.1007/s15010- 021- 01622- 930. Soares P, Rocha JV, Moniz M, Gama A, Laires PA, Pedro AR, et al. Factors Associated with COVID- 19 Vaccine Hesitancy. Vaccines [Internet]. 2021 Mar 22 [cited 2021 Jul 9];9(3):300. Available from: https://www.mdpi.com/2076- 393X/9/3/300/html31. Siegrist M, Zingg A. The role of public trust during pandemics: Implications for crisis communication. Eur Psychol [Internet]. 2014 [cited 2020 Oct 6];19(1):23- 32. Available from: /record/2013- 34232- 00132. Ezeibe CC, Ilo C, Ezeibe EN, Oguonu CN, Nwankwo NA, Ajaero CK, et al. Political distrust and the spread of COVID- 19 in Nigeria. https://doi.org/101080/1744169220201828987 [Internet]. 2020 Dec 1 [cited 2021 Jul 21];15(12):1753- 66. Available from: https://www.tandfonline.com/doi/abs/10.1080/17441692.2020.182898733. Okereke M, Ukor NA, Adebisi YA, Ogunkola IO, Iyagbaye EF, Owhor GA, et al. Impact of COVID- 19 on access to healthcare in low- and middle- income countries: Current evidence and future recommendations. Int J Health Plan Manage [Internet]. 2021 Jan 1 [cited 2021 Jul 26];36(1):13- 7. Available from: https://onlinelibrary.wiley.com/doi/full/10.1002/hpm.306734. Acharya KP, Ghimire TR, Subramanya SH. Access to and equitable distribution of COVID- 19 vaccine in low- income countries. npj Vaccines [Internet]. 2021 Apr 14 [cited 2021 Jul 9];6(1):1- 3. Available from: https://www.nature.com/articles/s41541- 021- 00323- 635. Murphy J, Vallieres F, Bentall RP, Shevlin M, McBride O, Hartman TK, et al. Psychological characteristics associated with COVID- 19 vaccine hesitancy and resistance in Ireland and the United Kingdom. Nat Commun [Internet]. 2021 Dec 1 [cited 2021 May 5];12(1):1- 15. Available from: https://doi.org/10.1038/s41467- 020- 20226- 936. Qian Y, Fan W. Who loses income during the COVID- 19 outbreak? Evidence from China. Res Soc Stratif Mobil. 2020 Aug 1;68:100522.37. Ettman C, Abdalla S, Cohen G, Sampson L, Vivier P, Galea S. Prevalence of Depression Symptoms in US Adults Before and During the COVID- 19 Pandemic. JAMA Netw open [Internet]. 2020 Sep 3 [cited 2021 Jul 9];3(9). Available from: https://pubmed.ncbi.nlm.nih.gov/32876685/38. Hossain M, Tasnim S, Sultana A, Faizah F, Mazumder H, Zou L, et al. Epidemiology of mental health problems in COVID- 19: a review. F1000Research [Internet]. 2020 [cited 2021 Jul 9];9. Available from: https://pubmed.ncbi.nlm.nih.gov/33093946/39. Garcia- Iglesias J, Gomez- Salgado J, Martin- Pereira J, Fagundo- Rivera J, Ayuso- Murillo D, Martinez- Riera J, et al. Impacto del SARS- CoV- 2 (Covid- 19) en la salud mental de los + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 85, 880, 902]]<|/det|> +profesionales sanitarios: una revisión sistemática. Rev Esp Salud Publica [Internet]. 2020 [cited 2021 Jul 9];94(1):e1- 20. Available from: https://medes.com/publication/152295 +40. Vindegaard N, Benros M. COVID- 19 pandemic and mental health consequences: Systematic review of the current evidence. Brain Behav Immun [Internet]. 2020 Oct 1 [cited 2021 Jul 9];89:531- 42. Available from: https://pubmed.ncbi.nlm.nih.gov/32485289/ +41. Mazereel V, Assche K Van, Detraux J, Hert M De. COVID- 19 vaccination for people with severe mental illness: why, what, and how? The Lancet Psychiatry [Internet]. 2021 May 1 [cited 2021 Jul 9];8(5):444- 50. Available from: http://www.thelancet.com/article/S2215036620305642/fulltext +42. Bendrau A, Plag J, Petzold MB, Strohle A. COVID- 19 vaccine hesitancy and related fears and anxiety. Int Immunopharmacol [Internet]. 2021 Aug 1 [cited 2021 Jul 9];97:107724. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8078903/ +43. Jefsen OH, Kølbæk P, Gil Y, Speed M, Dinesen PT, Sønderskov KM, et al. COVID- 19 vaccine willingness amongst patients with mental illness compared with the general population. Acta Neuropsychiatr [Internet]. 2021 [cited 2021 Jul 9];1- 4. Available from: https://www.cambridge.org/core/journals/acta-neuropsychiatrica/article/abs/covid19-vaccine-willingness-amongst-patients-with- mental- illness- compared- with- the- general- population/949CE2ADF019A3D78F64E704146EE348 +44. Lin C, Tu P, Beitsch LM. Confidence and receptivity for covid- 19 vaccines: A rapid systematic review. Vaccines [Internet]. 2021 Jan 1 [cited 2021 Apr 20];9(1):1- 32. Available from: https://pubmed.ncbi.nlm.nih.gov/33396832/ +45. Islam M, Kamal A, Kabir A, Southern D, Khan S, Hasan S, et al. COVID- 19 vaccine rumors and conspiracy theories: The need for cognitive inoculation against misinformation to improve vaccine adherence. PLoS One [Internet]. 2021 May 1 [cited 2021 Jul 19];16(5). Available from: https://pubmed.ncbi.nlm.nih.gov/33979412/ +46. Murphy M, Nanadiego FA, McCavera L, Nichols C, Kalekas P, Wachs D. Assessing the Validity and Accuracy of Online Videos on Vaccine Health Risks. Clin Pediatr (Phila) [Internet]. 2020 Feb 19 [cited 2021 Jul 9];59(4- 5):458- 66. Available from: https://journals.sagepub.com/doi/abs/10.1177/0009922820905866 +47. Eibensteiner F, Ritschl V, Nawaz F, Fazel S, Tsagkaris C, Kulnik S, et al. People's Willingness to Vaccinate Against COVID- 19 Despite Their Safety Concerns: Twitter Poll Analysis. J Med Internet Res [Internet]. 2021 Apr 1 [cited 2021 Jul 9];23(4). Available from: https://pubmed.ncbi.nlm.nih.gov/33872185/ +48. Smith DT, Attwell K, Evers U. Support for a COVID- 19 vaccine mandate in the face of safety concerns and political affiliations: An Australian study: Politics [Internet]. 2021 May 7 [cited 2021 Jul 9]; Available from: https://journals.sagepub.com/doi/10.1177/0263395711009066 +49. Largent EA, Persad G, Sangenito S, Glickman A, Boyle C, Emanuel EJ. US Public Attitudes Toward COVID- 19 Vaccine Mandates. JAMA Netw Open [Internet]. 2020 Dec 18 [cited 2021 Jul 9];3(12). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7749443/ +50. Dubov A, Phung C. Nudges or mandates? The ethics of mandatory flu vaccination. Vaccine [Internet]. 2015 May 21 [cited 2021 Jul 9];33(22):2530- 5. Available from: https://pubmed.ncbi.nlm.nih.gov/25869886/ +51. Lazarus JV, Ratzan S, Palayew A, Billari FC, Binagwaho A, Kimball S, et al. COVID- SCORE: A global survey to assess public perceptions of government responses to COVID- 19 (COVID- SCORE- 10). Hotchkiss D, editor. PLoS One. 2020 Oct;15(10):e0240011. +52. Worldometer. COVID- 19 data [Internet]. 2020 [cited 2020 Jun 30]. Available from: https://www.worldometers.info/coronavirus/about/#sources +53. Lazarus JV, Binagwaho A, El- Mohandes A, Fielding JE, Larson HJ, Plasència A, et al. Keeping governments accountable: the COVID- 19 Assessment Scorecard (COVID- SCORE). Nat Med [Internet]. 2020 Jun 11 [cited 2020 Jun 19];1- 4. Available from: https://doi.org/10.1038/s41591- 020- 0950- 0 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 868, 468]]<|/det|> +54. White TM, Cash-Gibson L, Martin-Moreno JM, Matesanz R, Crespo J, Alfonso-Sanchez JL, et al. COVID-SCORE Spain: Public perceptions of key government COVID-19 control measures. Eur J Public Health [Internet]. 2021 Apr 19 [cited 2021 May 3]; Available from: https://academic.oup.com/eurpub/advance-article/doi/10.1093/eurpub/ckab066/6238158 +55. Larson HJ, Cooper LZ, Eskola J, Katz SL, Ratzan S. Addressing the vaccine confidence gap. Lancet. 2011;378(9790):526-35. +56. MacDonald NE, SAGE Working Group on Vaccine Hesitancy. Vaccine hesitancy: Definition, scope and determinants. Vaccine [Internet]. 2015 Aug 14 [cited 2020 Oct 6];33(34):4161-4. Available from: https://pubmed.ncbi.nlm.nih.gov/25896383/ +57. Larson HJ, Clarke RM, Jarrett C, Eckersberger E, Levine Z, Schulz WS, et al. Measuring trust in vaccination: A systematic review. Hum Vaccines Immunother [Internet]. 2018 Jul 3 [cited 2021 Apr 22];14(7):1599-609. Available from: /pmc/articles/PMC6067893/ +58. Quinn SC, Parmer J, Freimuth VS, Hilyard KM, Musa D, Kim KH. Exploring communication, trust in government, and vaccination intention later in the 2009 H1N1 pandemic: Results of a national survey. Biosecurity and Bioterrorism [Internet]. 2013 Jun 1 [cited 2020 Jul 9];11(2):96-106. Available from: http://www.knowledgenetworks.com +59. Karafillakis E, Larson HJ. The benefit of the doubt or doubts over benefits? A systematic literature review of perceived risks of vaccines in European populations. Vaccine [Internet]. 2017 Sep 5 [cited 2021 Apr 22];35(37):4840-50. Available from: https://pubmed.ncbi.nlm.nih.gov/28760616/ +60. Strategic Advisory Group of Experts on Immunization (SAGE). Report of the SAGE Working Group on Vaccine Hesitancy [Internet]. Geneva; 2014 Oct [cited 2021 Apr 28]. Available from: https://www.who.int/immunization/sage/meetings/2014/october/1_Report_WORKING_GR OUP_vaccine_hesitancy_final.pdf + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[83, 120, 185, 140]]<|/det|> +# Tables and figures + +<|ref|>table_caption<|/ref|><|det|>[[83, 172, 254, 187]]<|/det|> +Table 1. Sample characteristics + +<|ref|>table<|/ref|><|det|>[[83, 210, 951, 875]]<|/det|> + +
CountryBrazilCanadaChinaEcuadorFranceGermanyGhanaIndiaItalyKenyaMexicoNigeriaPeruPolandRussiaSouth KoreaSingaporeSpainSwedenTurkeyUnited KingdomUnited StatesGlobal average
%%%%%%%%%%%%%%%%%%
Age
Groups
18-291916.516.523.516.514.135.624.218.537.923.531.923.31615.82518.420.715.515.722.116.617.1
30-3920.317.722.422.115.215.364.421.22020.722.168.121.517.317.723.421.320.717.416.320.617.215.8
40-491916.523.522.116.516.521.22018.920.620.618.51920.319.92218.617.620.617.817.1
50-5924.116.516.514.717.717.616.716.913.816.216.517.318.717.218.417.117.415.717.616.618.4
\(60+\)17.732.921.217.634.236.516.724.68.817.61830.928.814.12219.531.134.719.131.831.624.3
Sex
Male49.149.351.549.749.248.950.650.448.749.349.850.549.448.146.248.949.849.348.949.648.749.148.949.3
Female50.950.148.549.750.750.549.2485149.950.149.249.951.553.250.649.750.250.749.649.950.45150.2
Prefer not
to		0.60.60.10.60.21.60.30.80.10.30.70.40.60.50.40.50.40.71.40.50.10.54
say/Other
Education (university degree)
No82.673.888888275859185.6978491.584754592.544.967.86775.77965.664.377.6
Yes17.426.21212182515914.43168.51625557.555.132.23324.32134.435.738.7
Income (country median)
More
than32.139.865.930.339.630.217.766.629.836.149.617.940.740.224.465.650.435.437.622.632.3434238.7
Median
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[83, 117, 952, 770]]<|/det|> +
Less than Median No income58.25329.336.753.261.142.819.457.828.242.44838.953.469.318.937.454.152.368.856.94946.146.7
Have you or anyone else in your household experienced a loss in income due to the COVID-19 pandemic?
Severe loss29.716.714.458.914.513.743.457.514.959.933.156.237.310.824.940.119.523.216.413.138.411.917.929.0
Moderate loss42.224.851.536.223.11825.927.241.229.345.220.65016.636.239.740.739.234.623.43724.718.232.4
No28.258.534.1562.568.230.715.243.910.721.723.212.672.538.920.239.937.649.163.524.663.463.838.6
COVID-19 experience
None43.384.996.532.468.885.792.770.373.27850.491.729.852.253.158.594.191.472.567.460.173.67169.2
Self/family member38.3111.548.325.49.45.88.720.115.228.2635.541.442.825.13.83.319.627.331.317.817.721.0
sick
Lost
family member18.34.1219.25.84.91.520.96.76.821.42.334.76.44.116.52.15.37.85.38.68.611.39.8
Health care worker
Yes5.59.897.710.81220.625.45.610.98.919.48.24.72.96.711.27.77.813.89.314.816.510.8
No94.590.29192.389.28879.474.694.489.191.180.691.895.397.193.388.892.392.286.290.785.283.589.2
Mental health
Anxiety25.124.912.321.728.720.69.226.726.124.429.413.339.320.636.63722.720.824.424.144.727.621.925.3
Depression25.323.112.623.123.119.212.91522.432.523.825.128.820.327.230.921.121.622.223.838.725.923.223.6
+ +<|ref|>text<|/ref|><|det|>[[85, 770, 748, 791]]<|/det|> +Note: Anxiety and depression defined as symptoms moderate amount of time (3-4 days) or most or all of the time (5-7 days). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[360, 95, 700, 905]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 115, 880, 500]]<|/det|> +<|ref|>image_footnote<|/ref|><|det|>[[85, 512, 866, 540]]<|/det|> +
Panel c. Vaccine acceptance if recommended by employer or one's doctor among 1) willing to get vaccinated, 2) unsure/unwilling to get vaccinated.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[380, 85, 880, 950]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 140, 692, 360]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 85, 579, 100]]<|/det|> +
Figure 2. Vaccine acceptance by current cases and mortality
+ +<|ref|>image<|/ref|><|det|>[[125, 420, 733, 664]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 394, 520, 410]]<|/det|> +
Panel b. Vaccine acceptance and COVID-19 mortality
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 110, 763, 365]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[119, 85, 586, 100]]<|/det|> +
Panel c. Vaccine acceptance and COVID-SCORE-10 (June 2020)
+ +<|ref|>image<|/ref|><|det|>[[120, 442, 748, 667]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[119, 414, 456, 429]]<|/det|> +
Panel d. Vaccine acceptance and COVID-VAC
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[128, 110, 755, 345]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 84, 422, 99]]<|/det|> +
Panel e. Vaccination rates and hesitancy
+ +<|ref|>image<|/ref|><|det|>[[118, 457, 905, 710]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 408, 553, 425]]<|/det|> +
Figure 3. Vaccine acceptance for children among parents
+<|ref|>image_caption<|/ref|><|det|>[[118, 750, 521, 766]]<|/det|> +
Panel b. By respondent's willingness to vaccinate self.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[600, 110, 860, 900]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 110, 868, 840]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 85, 852, 101]]<|/det|> +
Figure 4. Correlates of vaccine acceptance-sociodemographic factors and COVID-19 experience.
+ +<|ref|>text<|/ref|><|det|>[[118, 846, 848, 896]]<|/det|> +Adjusted odds ratios (aOR) and 95%CI (log scale) from weighted multivariable logistic regression; reference categories: Female, No university degree, More than median income, No COVID-19 sickness/death, No loss of income. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 120, 940, 856]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 85, 841, 115]]<|/det|> +
Figure 5. Correlates of vaccine acceptance-COVID-VAC score, trust in government, anxiety and depression.
+ +<|ref|>text<|/ref|><|det|>[[115, 860, 845, 894]]<|/det|> +Adjusted odds ratios (aOR) and 95%CI (log scale) from weighted multivariable logistic regression, adjusted for socio- demographic factors and COVID- 19 experience. + +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[118, 86, 456, 99]]<|/det|> +Table 2. Vaccine acceptance by HCW status. + +<|ref|>table<|/ref|><|det|>[[118, 110, 936, 300]]<|/det|> + +
At least one dose receivedWill take when availableAt least one dose received or will receive when availableUnsure or not willing to take
n%%%%p-valueaOR(95%CI)
Not HCW1984052318317
All HCW329572.419.491.88.1<.0011.72(1.38,2.13)
Physician89185.6a11.2a96.9a3.1a<.001
Nurse61974.5b19.1b93.5b6.5b
Community Health Worker79069.6b22.5b92.2b7.8b
Other healthcare worker99561.6c24.5b86.1c13.9c
+ +<|ref|>text<|/ref|><|det|>[[118, 362, 868, 408]]<|/det|> +p-value based on Chi-squared tests. Different subscripts denote statistically significant pairwise differences. aOR from multivariate logistic model after adjusting for demographic variables, COVID-19 experience and clustering of HWC in countries. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[81, 175, 980, 550]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[83, 120, 301, 141]]<|/det|> +
Figure 6. Vaccine mandates acceptance.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 384, 150]]<|/det|> +SupplementalMaterialssubmit.docx + +<--- Page Split ---> diff --git a/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/images_list.json b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..31d060f7e2445c415b9c11a17bd6b6fa77fa7c35 --- /dev/null +++ b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Schematic representation of the experimental design and methods (a) The 1 year long cheddar-making experiment. Microbial population dynamics were quantified during cheese ripening using a selective method of viable cell counting, which discriminates between thermophilic cocci, mesophilic cocci and non-starter lactic acid bacteria (NSLAB). Additionally, metabolic changes in the cheeses were measured using several targeted analytical chemistry approaches. (b) Schematic representation of the controlled milk experiment in the laboratory as well as of the methods (1-5) used in the controlled milk experiment. This second experiment involves the removal of additional strains, namely the individual exclusion of three major Lactococcus strains. The numbers indicate the different type of data and analysis as follows: 1. metatrancriptomics, 2. metabolomics, 3. genomics, 4. phylogenomics and 5. genomes-scale metabolic models (GEMs) and community simulations. Our integrative systems biology approach combined: i) the analysis of the SLAB community’s genomes, ii) the generation and simulation of their respective GEMs, iii) the analysis of the metatranscriptomes across the different strain removal conditions and iv) the quantification of key metabolites.", + "footnote": [], + "bbox": [ + [ + 92, + 225, + 904, + 551 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. S. thermophilus benefit the growth of lactococci community and influence the final flavor of cheese. (a-d) Microbial population dynamics during cheese ripening. Relationship between CFUs and time (months) for (a) Mesophilic, (c) Thermophilic cocci, and (d) non-starter lactic acid bacteria (NSLAB). (b) Presents the boxplot comparison of the different condition on Mesophilic cocci at 12 months. Note, significant reduction only on Mesophilic cocci is observed when S. thermophilus is absent. Condition All corresponds to whole culture which is composed from 24 number of strains. HP corresponds to an alternative method (hand packed) of the whole culture inoculation. -LB corresponds to the removal of a L. lactis blend population. -ST corresponds to the removal of the S. thermophilus strain. (e-h) Metabolome dynamics during cheese ripening. (e) 2-dimensional representation with the usage of UMAP of all samples based on the metabolomics measurements (Acids, Carbohydrates and Peptides). The colour indicates the time when the sample was taken and the shape indicates the four different conditions. Note, the stronger change is observed between 2 week and 3 months while the removal of ST (cross symbol) has a strong effect on the later time of cheese ripening. (f) Relative change of the metabolites on different time intervals including all the conditions, both colour and shape indicates the class of the metabolite. Note, the higher relative change of peptides followed by acids at the interval between 2 week and 3 months, later acids exhibit higher changes. The three measured sugars (Glucose, Lactose and Galactose) are excluded from this panel. (g) Selection of the six most discriminative metabolites out of 50 for the four conditions, colour indicate the condition while the right side shape the class of the metabolite. Note, the majority of peptides concentration are significant different when S. thermophilus is not present as well as with galactose, lactose and lactic acid. (h) Galactose concentration over time highlights the absence of galactose when S. thermophilus is not present. The galactose concentration remains steady from the start till the 9 month of cheese ripening and then shows signs of decline.", + "footnote": [], + "bbox": [ + [ + 91, + 157, + 905, + 484 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_5.jpg", + "caption": "Figure 3. The removal of different Lactococcus strains has a distinct effect on gene expression in S. thermophilus. (a-e) relationship of the Lactococcus phylogeny with the S. thermophilus transcription change on different dropout conditions. (a) Lactococcus phylogenetic tree separates the L. cremoris and L. lactis strains. The two clades are colored with transparent yellow and grey, respectively. The colors and shape of the tree tips indicate the strains presents in the SLAB culture. Tree empty tips correspond to the obtained complete NCBI genomes of Lactococcus. Additional information is presented in the outer layers of the tree. The first layer presents the unique k-mer content based on the SLAB culture's genomes (purple-scale color) and indicates higher number for the strains who have no close relative within the culture. The second layer presents the proxy of temperature stress tolerance based on Pearson Correlation Coefficient (PCC) between individual acidification curve of 30 and 40 degree \\(^\\circ \\mathrm{C}\\) (red-scale color). High values correspond to high temperature tolerance. The clade L. cremoris shows lower temperature tolerance, with few exceptions including the main LC strain (more elaborate data are presented in Supplementary Fig. 5). The third layer presents the percentile of transcribed sigletons (green-scale color) as well as the total number of the strains sigletons (outer bars). (b-e) Volcano plots presents the S. thermophilus transcription between the whole community (All) and the different Lactococcus dropout conditions, respectively. The colors corresponds to the 4 Lactococcus components. The S. thermophilus transcription observed to change more when LLm1 was left-out followed by LB, with 291 and 182 number of total significant up/down regulated genes, respectively. A smaller effect of 21 genes was observed when LC was left-out and only 1 gene when LLm2 was left out.", + "footnote": [], + "bbox": [ + [ + 90, + 197, + 905, + 479 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_6.jpg", + "caption": "Figure 4. S. thermophilus provides nitrogen in form of amino acids to L. lactis community, which is necessary for de novo nucleotide biosynthesis. (a) Summary of metabolic modeling analysis including individual model and community-based simulations. (b) Selected exchange fluxes predicted across individual flux balance analysis simulations carried in milk media variations for each set of metabolic models, in particular fermentation products, amino acids, and carbon sources highlighting differences in metabolic strategies across species. (c) Alluvial diagram showing predicted metabolic exchanges from community simulations in milk media variants, highlighting the fact the amino acid valine is strongly predicted across all simulation conditions where metabolic cross-talk is expected. Refer to Supplementary Figure 6 for more details on the metabolic models and simulation. (d-i) Transcriptomics profiles between All and -ST (S. thermophilus is left out) conditions of key functions for all the strains in SLAB culture. (d-f) Transcriptomics boxplots of branched chain amino acids (BCAA) aminotransferase, BCAA transport system 2 carrier protein and ammonium transporter (T) supports the predictions of metabolic exchanges. Note, (e) the higher transcription profile of S. thermophilus on BCAA transport system as well as the high transcription profile of S. thermophilus and the increase of LC transcription profile when S. thermophilus was left out. (g-j) Up-regulation of Lactococcus glutamine and nucleotide metabolism when S. thermophilus is left out. (g-i) Boxplots present the activities of nitrogen regulatory protein P-II, glutamine synthetase (glnA) and related transcriptional regulator (TR) (GlnR) for all community members. Note, the up-regulation of all the enzymes derived from Lactococcus strains when S. thermophilus is left out. (j) Metatranscriptomics represents the increased activity of Lactococcus reactions, which are incorporated into Escher maps. The pattern shows a coordinate up-regulation (with green) of the enzymes towards guanine biosynthesis (and uridine). (k) Schematic compilation of the compounds S. thermophilus may provide to Lactococcus community as well as selected transcriptional changes of the Lactococcus community when S. thermophilus was left out. The color and direction of the arrows represent the up- and down- regulation with green/up and red/down, respectively.", + "footnote": [], + "bbox": [ + [ + 86, + 103, + 910, + 504 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community The measured metabolome from the controlled milk experiment in response to stain removal conditions (a-d). (a) The PCA of key carbohydrates, acids and cheese flavor compounds. The strongest response observed from the removal of L. cremoris followed by coordinated response of LLm1, LB and S. thermophilus. Removal of LLm2 has no effect. The arrows represents the top 5 compounds (2,3-Pentanedione, Hexanal, Diacetyl, Acetoin and Ethyl acetate) responsible for the observed changes. Panels (b-c) represents the signal-to-noise ratio (S/N) of diacetyl and acetoin, respectively. Regarding those two compounds, absence of S. thermophilus from the community led to significant lower accumulation, while absence of L. cremoris leads to significant higher accumulation. The black horizontal line indicate the average value of the two compounds at milk prior to acidification. Paired t-test for each condition against the whole community (All) were performed with the symbols indicating the statistical significance results: ns: \\(\\mathrm{p} > 0.05\\) , \\*: \\(\\mathrm{p}< = 0.05\\) and \\*\\*: \\(\\mathrm{p}< = 0.01\\) . (d) Schematic compilation of the compounds that changed between the whole SLAB culture and when L. cremoris was left out (for detailed boxplots see supplementary). The color represent the removal conditions; yellow indicates the removal of L. cremoris while red represents the whole SLAB culture. The upwards arrow indicates the increase of the compounds relative to the compounds concentration in milk. The X symbol indicates the lack of production of the compounds in the respective condition. (e) Simplified metabolic escher maps from citrate to \\(\\alpha\\) -ketoglutarate as well as to diacetyl through pyruvate and (S)-2-Acetolactate. Metabolites and reactions are indicated with black and grey color in the graph, respectively. Full colored circles and open colored circles indicate the high and low transcription per strain, respectively. Lack of circle visualization indicates the lack of the corresponding orthologous group genes. Note, the transcriptional fluxes from citrate-sodium symporter towards diacetyl, which are present to all members in Lactococcus community. In addition, LC and LLm1 manifest higher transcription among the Lactococcus community on the genes related to acetolactate decarboxylase. Only LC strain manifests transcriptional flux through diacetyl reductase and butanediol dehydrogenase as well as in reactions towards \\(\\alpha\\) -ketoglutarate. (f) Predicted fluxes based on metabolic modeling across the different simulated media.", + "footnote": [], + "bbox": [ + [ + 171, + 75, + 820, + 535 + ] + ], + "page_idx": 34 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323.mmd b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323.mmd new file mode 100644 index 0000000000000000000000000000000000000000..52b627f6285056d9d9dfb2ccec9c306b8324b3c7 --- /dev/null +++ b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323.mmd @@ -0,0 +1,390 @@ + +# Inter-species metabolic interactions in cheese flavour formation + +Chrats Melkonian ( \(\boxed{\infty}\) chrats.melkonian@gmail.com) Utrecht University & Wageningen University & Research https://orcid.org/0000- 0002- 1448- 5542 + +Francisco Zorrilla Medical Research Council Toxicology Unit, University of Cambridge + +Inge Kjærbølling Novozymes + +Sonja Blasche University of Cambridge + +Daniel Machado Norwegian University of Science and Technology + +Mette Junge Chr. Hansen + +Kim Soerensen Chr. Hansen + +Lene Andersen Chr. Hansen + +Kiran Patil University of Cambridge https://orcid.org/0000- 0002- 6166- 8640 + +Ahmad Zeidan Chr. Hansen + +## Article + +Keywords: Cheddar cheese, Microbial interactions, multi- omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris + +Posted Date: March 16th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2574132/v1 + +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on December 21st, 2023. See the published version at https://doi.org/10.1038/s41467-023-41059-2. + +<--- Page Split ---> + +# Inter-species metabolic interactions in cheese + +## flavour formation + +3. Chrats Melkonian1,6,7,*, Francisco Zorrilla2, Inge Kjærbølling1, Sonja Blasche2, Daniel Machado3, Mette Junge4, Kim Ib Soerensen4, Lene Tranberg Andersen5, Kiran R Patil2, and Ahmad A. Zeidan1,* + +6. Systems Biology, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark +7. Medical Research Council Toxicology Unit, University of Cambridge, Cambridge, UK +8. Department of Biotechnology and Food Science, Norwegian University of Science and Technology, Trondheim, 7491, Norway +9. Global Application, Food Cultures & Enzymes, Chr. Hansen A/S, 2970 Hørsholm, Denmark +10. Strain Development, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark +11. Theoretical Biology and Bioinformatics, Science for Life, Utrecht University, Utrecht, the Netherlands +12. Bioinformatics Group, Wageningen University and Research, Wageningen, Netherlands +13. Authors to whom correspondence should be addressed: chrats.melkonian@gmail.com, dkahze@chr-hansen.com + +## ABSTRACT + +<--- Page Split ---> + +Cheese fermentation and flavour formation are governed by complex biochemical reactions driven by polymicrobial activity. While the compositional dynamics of cheese microbiomes is relatively well mapped, the mechanistic role of microbial interactions in flavour formation is yet unknown. We microbially and metabolically characterised a year- long Cheddar cheese process using a commonly used starter culture containing Streptococcus thermophilus and Lactococcus strains. By using an experimental strategy whereby certain strains were left out from the starting mixture, we identified the critical role of S. thermophilus in boosting Lactococcus growth and in shaping flavour compound profile. Controlled milk fermentations with systematic exclusion of single Lactococcus strains, combined with genomics, genome- scale metabolic modelling, and metatranscriptomics, indicated that proteolytic activity of S. thermophilus relieves nitrogen limitation for Lactococcus and boosts de novo nucleotide biosynthesis. While S. thermophilus had large contribution to the flavour profile, L. cremoris also played a role by limiting diacetyl and acetoin formation which leads to off- flavour when in excess. This off- flavour control could be attributed to different metabolic re- routing of citrate between L. cremoris and other L. lactis strains. Further, closely related L. lactis strains exhibited different interaction patterns with S. thermophilus highlighting the importance of strain- specificity in cheese- making. Overall, our results bring forward the critical role of competitive and cooperative microbial interactions shaping cheese flavour profile. + +Keywords: Cheddar cheese, Microbial interactions, multi- omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris + +<--- Page Split ---> + +## Introduction + +Fermented foods based on microbial consortia (e.g., cheese, kefir and kombucha) constitute a large part of the modern diet with many reported health benefits [1, 2]. In cheese, well defined Starter Lactic Acid Bacteria (SLAB) cultures are used to ensure consistent and flavorful products. During cheese making, microbes encounter dynamic conditions characterized by different nutrient availability and different stresses, starting with an initial feast during milk fermentation, followed by a long period of famine during cheese ripening. Cheese- making thus provides a controlled system with well characterized dynamics to understand the role of microbial interactions in shaping the characteristics and quality of fermented foods [3]. + +In industrial cheese making, SLAB cultures are responsible for milk acidification via lactose fermentation. They are predominantly composed of mesophilic strains of Lactococcus, including the species L. lactis and L. cremoris as well as thermophilic strains of Streptococcus thermophilus. A rich literature exists on the physiology of both species in mono- cultures [4, 5, 6], and genome- scale metabolic models have been used to model their metabolism [7, 8]. In contrast, the interactions between the two taxa is still largely unknown from a mechanistic perspective despite indications of strong inter- dependencies [9]. Further, the current knowledge on cheese microbial interactions is limited to the scope of simplified artificial milk media, concerns pairwise associations, or focus only on the short time interval during the milk acidification. Less is known about the interactions involving more than two SLAB strains and how interaction networks evolve during the cheese ripening steps. Metabar coding and metagenomics approaches have also been used to study the cheese microbiome [10, 11, 12]. While these studies provide a good overview on the community dynamics during cheese- making using higher- order taxonomy, they are limited in uncovering inter- species interactions and strain- level diversity. + +To uncover the role of microbial interactions and interacting agents, we used a strain dropout strategy whereby one single strain or a strain group was left out from the starter culture, and the cheese- making process was characterized in its entirety. We used an industrially relevant SLAB culture, containing: one Streptococcus thermophilus (ST), two major L. lactis (LLm1 & LLm2), one major L. cremoris (LC), + +<--- Page Split ---> + +44 and a mixture of 21 L. cremoris and L. lactis strains in smaller fractions (hereafter, Lactococcus blend (LB)). We investigated the effect of leaving out S. thermophilus and Lactococcus blend in a year- long cheddar- making experiment (Fig. 1a); followed by controlled milk experiments investigating hypotheses generated from the year- long experiments. For the latter, we used an integrative systems biology approach that combined different layers of biological information (Fig. 1b). + +<--- Page Split ---> + +## Results + +## 1 S. thermophilus supports lactococci growth and shapes the metabolic profile during cheese ripening + +To study the interactions between the members of the SLAB culture, we started by quantifying the population dynamics during a year- long ripening cheese making experiment (see Methods). We used for variations of starter cultures: i) containing all member species of an industrial starter culture, viz., S. thermophilus (ST), L. cremoris (LC), L. lactis (LLm1 and LLm2), and Lactococcus blend (LB); ii) the same as (i) but prepared independently (HP); iii) excluding LB; and iv) excluding ST (Fig. 1a). In the three conditions that included S. thermophilus, we observed a trend of slow decline in the population of lactococci and S. thermophilus, starting with 9.25 and ending at 8 log10 CFU/g (Fig. 2a- c). In the single condition that excluded S. thermophilus, we observed a much steeper decline in lactococci population, ending at 6.5 log10 CFU/g (Fig. 2a- b). Also, the declining lactococci population exhibited different trajectories between the two batches, albeit both converging at 9 and 12 months (Fig. 2a- b). The non- SLAB count increased from 0 to 7.5 log10 CFU/g, in the highest case (Fig. 2d). Based on 95% confidence interval estimates, we observed high variability among the non- SLAB populations, but with no appreciable overall difference between the conditions (Fig. 2d). Leaving out Lactococcus blend or changing the way of packing the different strains in the culture did not result in any visible response to the community dynamics. Thus, inter- species interactions, especially involving S. thermophilus, appear to be driving the overall population dynamics. + +To measure the effect of the microbial interactions in the resulting phenotype of the cheese, we used targeted metabolomics analyses. Across conditions, the biggest change in the metabolic profiles of cheese was observed between the two- week mark and three months (Fig. 2e). This can be attributed to changes in peptide composition followed by the general increase in amino acids concentrations (Fig. 2f). Furthermore, cheese metabolomes continue to change at a slower pace between the three- to twelve- month marks, resulting in different end profiles. A notable time- depended accumulation of compounds after the third month was observed on acetic acid, tyramine, g- amino butyric acid, putrescine and cadaverine + +<--- Page Split ---> + +(Fig. 2f & Supplementary Fig. 1). The cheeses produced without S. thermophilus formed a separate cluster relative to the rest (Fig. 2e). This pattern can be ascribed to compounds with significantly different concentrations across strain removal conditions, especially when S. thermophilus was absent. The majority of these compounds are peptides that were either not accumulated, accumulated in lower amounts or accumulated in higher amounts without the addition of S. thermophilus. Lactose, galactose and lactic acid concentrations were also found to be significantly different when S. thermophilus was absent (Fig. 2g, ANOVA \(F(3,36) = 208 \pm 334, p < 0.001, \eta^2 = 0.86 \pm 0.08\) & Supplementary Fig. 2). This difference was already notable in the samples from the second week of cheese ripening (Fig. 2h & Supplementary Fig. 3a,c&e). These results points towards the significant influence of S. thermophilus growth during milk fermentation and its non- growing yet active cells during the long term cheese ripening. + +The presence of S. thermophilus was found to benefit both the growth of the Lactococcus community and the final metabolic profile of the cheddar cheese. Notably, the growth benefit is not clearly visible before two weeks to three months, a time- frame that may be considered prohibitively long for experiments in the laboratory. Instead, these effects become more evident as the experimental time- frame expands to one year. The cheese metabolic profiles indicated differences in peptide composition between the conditions, a result likely stemming from the non- growing metabolic activity of S. thermophilus. In accordance with the literature, the presence of S. thermophilus reveals a few key metabolic changes from the milk fermentation; the complete consumption of lactose and the production of galactose [13, 14]. Together, the compositional and metabolic profiles led us to hypothesize that both early and late S. thermophilus activity is critical for the long term effect on Lactococcus population. + +## 2 S. thermophilus has strain-specific interactions with the members of Lactococcus lactis community + +To assess the role of S. thermophilus, we performed controlled experiments wherein additional strains were removed. Samples were taken when milk fermentation reached the transition from exponential to stationary phase (at pH 5) and gene expression was monitored using metatranscriptomics. To further investigate the microbial interactions within the SLAB culture, we also analyzed the genomes of the individual strains + +<--- Page Split ---> + +in the culture and their phylogenetic relationship (Fig. 3a). The phylogenomic analysis separated the species L. lactis and L. cremoris into two distinct groups. Two of the main Lactococcus strains, LC and LLm2, have no close relatives within the culture community. For the third main Lactococcus strain, LLm1, we identified two close relatives (LL- LB01 & LL- LB02), belonging to the Lactococcus blend (Fig. 3 & Supplementary Table 1). The rest of the strains in the Lactococcus blend formed distinct sub- clusters within the cremoris clade (Fig. 3). By employing pan- genome analysis on Lactococcus strains, we identified singletons (i.e. a gene found only in one genome) for each member of the culture. The highest number of singletons was harbored by the main LLm2 followed by LC, and LLm1 where its number of singletons was in the same range as the Lactococcus blend strains, with numbers 288, 134, 79 and average of \(53.6 \pm 18.6\) , respectively. In parallel, we evaluated both the presence and activity level (using metatranscriptomics) of the Lactococcus strains by calculating percentile of the transcribed singletons per genome. We found a high number of transcribed singletons on all the main Lactococcus strains, while this number varied considerably among the strains of the Lactococcus blend, raising questions regarding whether or not these strains were active (Fig. 3a & Supplementary Table 4). We hypothesized that the temperature dynamics during the experiment could influence the strain activity. To test this, by carrying out milk fermentation using each of the individual Lactococcus strains at different temperatures, we calculated a proxy for their temperature stress tolerance (Pearson correlation between curves of 30 against 37, 40 and 43 degrees). The results showed that the L. lactis strains have higher temperature stress tolerance in the tested ranges, including Lactococcus blend strains LL- LB01 and LL- LB02 (PCC \(0.96 \pm 0.05\) ). The L. cremoris strains exhibited low temperature stress tolerance (PCC \(0.7 \pm 0.24\) ), with the main LC being one of the most stress tolerant strains within this group (PCC range 0.84- 0.99) (Supplementary Fig. 5a&b, Fig. 3a). + +We next explored the extent to which we could discriminate between the different strains. The number of unique k- mers in each genome ranged from 368 to 622455 and they were found to be dependent on the presence of close phylogenetic relative strains (Fig. 3a). The k- mer transcription abundance indicated the feasibility of discriminating between the three main Lactococcus strains (LC, LLm1 and LLm2) and the one S. thermophilus strain, but not between the strains in the Lactococcus blend (Supplementary Fig. 4). Using differential expression analysis, we investigated the interaction effect on S. thermophilus + +<--- Page Split ---> + +transcriptome by leaving out each Lactococcus strain/blend from the culture. We found the strongest response when LLm1 was left out followed by Lactococcus blend, LC and LLm2 with 291, 182, 21 and 1 gene(s) significantly differential expressed, respectively (Fig. 3b- e, Supplementary Table 6 & sections below). Although both LLm1 and LLm2 belong to L. lactis clade, the latter strain stands out by having the highest number of singletons as well as having the second highest number of unique k- mers (Fig. 3). By removing strains with a different degree of relatedness we identified that different Lactococcus strains have distinct effects on gene expression of S. thermophilus. This finding can be attributed to the different genetic background and the diverse phylogenetic relationships within the Lactococcus community, highlighting the diversity of potential microbial interactions between S. thermophilus and the Lactococcus community. + +## 3 S. thermophilus proteolytic activity may benefit Lactococcus community by providing nitrogen sources + +To investigate mechanisms of inter- species interactions, we generated genome- scale metabolic models for the individual strains in the culture and performed flux balance analysis - based simulations. To explore the effect of changes in growth medium composition, we generated models gap- filled on relevant variations of milk- mimicking media. Supplementary Fig. 6a shows overall model statistics for the different species, while Supplementary Fig. 6b shows the specific reactions added through gap- filling on different media across the models (see Supplementary Section 0.1). Flux balance simulations with individual models were carried out using the different model sets in aerobic and anaerobic variants of fermented and non- fermented milk (Fig. 3a,b, Supplementary Fig. 6c,d). All species consistently consumed lactose in simulations where growth was feasible. Several differences in amino acid uptake/secretion were observed among the strains. The branched chain amino acid valine was exported by S. thermophilus in all but one simulation condition, whereas LC, LLm1, and LLm2 consumed this amino acid in four, six, and three simulations conditions, respectively. Additional amino acid uptake/secretion predictions are reported at Supplementary Section 0.1 + +To assess potential cross- feeding, we performed community simulations using the SMETANA framework (see Supplementary Section 0.1). We used the calculated SMETANA scores, ranging from 0 to + +<--- Page Split ---> + +1, as a measure of predicted interaction confidence (0 being lowest confidence and 1 being the highest) [15]. These simulation indicate that S. thermophilus provides valine to Lactococcus strains in all three community simulation conditions (gap- filled on rich aerobic milk & simulated on minimal aerobic milk, gap- filled on rich anaerobic milk & simulated on minimal aerobic milk, and gap- filled on rich anaerobic milk & simulated on minimal anaerobic milk) where exchanges were predicted between community members (Figure 3c). Not only was this interaction consistent, it also occurred with a SMETANA score of 1 in every case, suggesting that this is a key ecological interaction for maintaining the composition and function of the community. Serine exchanges from S. thermophilus to LLm1 were predicted in 2 out of 3 community simulation conditions, with an average SMETANA score of \(0.38 \pm 0.01\) . Exchanges from S. thermophilus to LC involving glycine and ammonium were predicted in 1 out of the 3 simulation conditions, with SMETANA scores of 0.3 and 0.29, respectively (Figure 3c). Finally, an exchange from S. thermophilus to LLm2 involving alanine was predicted under only 1 simulation condition with a SMETANA score of 0.42. Overall, the metabolic simulations strongly indicate cross- feeding between the community members, with S. thermophilus standing out as a key donor (Figure 3c). + +To complement the genomic and metabolic modeling analysis, we investigated the changes in S. thermophilus and Lactococcus community transcriptome. The principal component analysis showed pattern consistency with the differential transcriptome analysis (Fig. 3c- f, Supplementary Section 0.2). Regarding changes in S. thermophilus's transcriptome, we found up- regulated genes annotated as oligopeptide- binding protein (AmiA) and oligopeptide transport system permease protein (OppB, OppC, OppD & OppF) (see Supplementary Section 0.2). Seven down- regulated genes are annotated as transporters, such as glutamine ABC transporter permease protein (GlnP), cadmium, cobalt and zinc/H(+)- K(+) antiporter (Supplementary Section 0.2). Further, transcriptional repressors such as the catabolite control protein A (CcpA) and nitrogen- metabolism- regulating- proteins such as GTP- sensing transcriptional pleiotropic repressor (CodY) were found to be transcriptionally active. In line with the putative interactions predicted using metabolic modelling, we found a gene annotated as branched- chain amino acids (BCAA) transaminase expressed in S. thermophilus, suggesting that valine biosynthesis is active. Moreover, we found a gene annotated as BCAA transport system 2 carrier protein (BrnQ) as well expressed, suggesting that S. thermophilus is secreting valine. Complementary differential pathway enrichment analysis supported + +<--- Page Split ---> + +further that valine biosynthesis was enriched in S. thermophilus metatranscriptome (Fig. 4d- e). Overall, the metabolic modeling and metatranscriptome analysis together suggests that S. thermophilus can act as a branched chain amino acid donor to Lactococcus. It is well known that Lactococcus harbor a number of amino acid auxotrophies including valine [16]. In addition, the SMETANA simulations provide insights into the metabolic strategies that may be employed by community members under challenging media conditions. + +We next investigated transcriptional change in the Lactococcus community in the presence and absence of S. thermophilus. As it is not possible to distinguish between the transcriptome profiles of all the different Lactococcus strains, we merged these into one group for pan- genome analysis and pan- metabolic modelling. Strain- to- strain transcriptional changes were investigated when possible (see Supplementary Section 0.2). The most up- regulated OGs in Lactococcus when S. thermophilus is left out are involved in nitrogen assimilation, namely nitrogen regulatory protein P- II, related transcriptional regulator (GlnR), ammonium transporter (amtB), glutamine synthetase (glnA) and glutamine transport ATP- binding protein (glnQ). Additionally, we confirmed that GlnR in Lactococcus regulates the three operons, namely amtB, glnA, and glnQ (Fig. 4f- i, Supplementary Table 2 & Supplementary Section 0.2). The purine and pyrimidine biosynthesis pathways are deferentially expressed in a coordinated manner. Several genes in the pathways were up- regulated, in particular the genes associated with the reactions leading to uridine and guanine synthesis (Fig. 4j & supplementary). Both pathways are connected via L- glutamine synthetase. Expression patterns in amino acid metabolism show a mixed pattern of both up- and down- regulation (see Supplementary Section 0.2). Overall, the metatranscriptomics analysis further supports that S. thermophilus cross- feed Lactococcus community. One hypothesis that could explain this is that S. thermophilus has a higher proteolytic activity, thereby providing more peptides and amino acids (available nitrogen source) in the culture which Lactococcus takes advantage of. When S. thermophilus is absent, there is a lower degree of nitrogen availability causing upregulation of the nitrogen assimilation and nucleotide syntheses pathway in Lactococcus. + +<--- Page Split ---> + +## 4 Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community + +To elucidate the role of interactions between the SLAB strains in the development of cheese flavor, we performed a targeted metabolomics analysis. Leaving out \(L\) . cremoris (LC) led to the strongest metabolic response followed by a weaker and different response when leaving out LLm1, S. thermophilus and the Lactococcus blend (see Supplementary Section 0.3). Surprisingly, leaving out LLm2 did not lead to a marked change in the metabolic profile of the acidified milk (Fig. 5a). Six flavor compounds, namely heptanal, hexanal, 2- ethyl- furan, 2,3- pentanedione, diacetyl and acetoin, were either detected at significantly higher concentrations or only produced when this main \(L\) . cremoris strain was removed (Fig. 5b- c & Supplementary Fig. 10). A graphical summary of the metabolome differences between the whole culture and the culture lacking LC is shown in Fig. 5d. Many of the flavor compounds found were significantly altered by the strain removal strategy, with the majority corresponding to byproducts of secondary metabolism with unknown microbial biosynthesis pathways or gene clusters. Yet, investigation of their patterns highlights the numerous ways these compounds may cross- feed within the SLAB culture. For example, 2,3- Pentanedione could be donated by S. thermophilus to LC while ethyl hexanoate, ethyl acetate and 2- Methyl- 3- thiolanone likely produced by LC. + +C4 aroma compounds such as diacetyl and acetoin are known derivatives of citrate metabolism with characteristic contribution to buttery- like aroma. Although desirable in small quantities, e.g less than 0.05 rag. per 100 g of diacetyl content, higher amount could lead to off- flavors [17, 18]. Therefore, we investigated whether the increase in diacetyl and acetoin concentrations triggered by the removal of LC could be explained by the metatranscriptome profiles. Indeed, we found differential expression patterns in genes related to diacetyl and acetoin metabolism. Therefore, we examined the gene expression levels in the metabolic route from citrate towards diacetyl and acetoin through pyruvate as well as an alternative route towards \(\alpha\) - ketoglutarate. The citrate- sodium symporter that is present only on Lactococcus strains was highly expressed implying active uptake of the citrate available in milk by \(L\) . lactis. From citrate, Lactococcus strains showed expression of the genes associated with the pyruvate + +<--- Page Split ---> + +carboxylase and acetolactate synthase reactions, whereas only LC and LLm1 had the genes related to acetolactate decarboxylase expressed at high levels. On the other route towards \(\alpha\) - ketoglutarate and further, genes associated with aconitate hydratase and isocitrate dehydrogenase are only expressed in LC and S. thermophilus (Fig. 5e & Supplementary Fig. 11). Metabolic modeling simulations across the different medium conditions are largely corroborated by the metatranscriptomics results with the exception of three reactions involving acetoin. These are acetolactate decarboxylase (ACLDC), which does not carry a flux in the LC model, and 2,3- butanediol dehydrogenase (BTDD_RR) and diacetyl reductase (ACTD), which carry fluxes in LLm1 model but not in LC model (Fig. 5e & f). Taken together, the presence of the main L. cremoris strain in the cultures results in the prevention of an undesirable increase in the levels of some key flavor compounds, including diacetyl and acetoin. This can be attributed to the degradation of diacetyl, e.g., through the action of some alcohol dehydrogenases from LC. Alternatively, LC might be competing with the other strains in the culture on citrate and subsequently converting it to other products than diacetyl and acetoin, such as ethyl- hexanoate, ethyl- acetate and 2- methyl- 3- thiolanone, which normally happens in mixed culture fermentation following the rapid drop of redox potential in the beginning of fermentation. This would result in the reduction of the overall amount of citrate available for diacetyl and acetoin formation. + +## Discussion + +Interactions in microbial communities are reported in a wide range of ecosystems [19, 20, 21, 22, 23]. While such interactions are widely reported also in food microbial communities [24, 3, 25, 26, 27, 28], very few studies have provided insights into the molecular agents that mediate the interactions. Uncovering interactions in situ is particularly challenging due to complexity of the communities involved as well as that of the medium such as milk. Here, we combined genomics, metatranscriptomics, metabolomics, and metabolic modelling to uncover key microbial interactions in Cheddar cheese- making. Notably, we used industrial strains and a full cycle of year- long cheese ripening. We report competitive and cross- feeding metabolic interactions between the lactic acid bacteria used in Cheddar cheese production. Firstly, L. cremoris strain competes with L. lactis for the milk's available citrate, resulting in accumulation of key + +<--- Page Split ---> + +metabolites such as diacetyl and acetoin in the final product. Such interactions take place within the first 5 hours of milk fermentation and strongly influences the final cheese flavor. Secondly, S. thermophilus provides the necessary nitrogen source to the Lactococcus community, explaining the significant one- year long growth benefit of the Lactococcus strain population and the different final cheese metabolome profile. In addition, Lactococcus strains affect the activity of S. thermophilus differently. While the presence of one L. lactis strain hardly influences the activity of S. thermophilus and the development of the final cheese flavor profile, the presence of a different L. lactis strain influences both significantly. Often in literature the attempts to study microbial interactions disregard the strain diversity. Yet, recent studies highlight the importance of strain interactions, claiming the major role of those in predicting eco- evolutionary dynamics [29]. Our results show how strain- specific metabolic interactions between microbes shape the biochemical profile of cheese, and provide targets towards the rational design and assembly of microbial communities with the aim of fine- tuning cheese flavor. More broadly, the study provides a blue- print to uncovering in situ interactions in complex food microbial ecosystems. + +## Methods + +## Starter lactic acid bacteria culture + +A cheese making experiment was designed to investigate the effect of removing selected strains from a defined- strain SLAB culture. Wild- type strains were obtained from the Chr. Hansen Culture Collection and were originally isolated from dairy products or cultures. By applying this strain removal strategy, it becomes possible to gain insights into the individual role of the strain but also its interactions within the microbial community. The original defined- strain SLAB culture consisted of 5 different bulks: 1 bulk with single strain S. thermophils, 1 bulk with a multi strain mixture of 21 Lactococcus strains belonging to both species of L. lactic and L. cremoris [30], which were inoculated as a single component (hereafter, Lactococcus blend), 2 bulks with single strain L. lactis, and 1 bulk with single strain L. cremoris. The original defined- strain SLAB culture was composed from of 25 strains and all bacterial and all strains applied are listed in Supplementary Table 1. + +To apply the strain removal strategy 4 defined- strain SLAB cultures were designed. Condition ALL, + +<--- Page Split ---> + +corresponded to the original defined- strain SLAB composed from of 25 strains packed under industrial conditions. Condition - HP, corresponded to the original defined- strain SLAB composed from of 25 strains packed under non- industrial condition by hand. Condition - LB, corresponded to the removal of the Lactococcus blend. Condition - ST, corresponded to the removal of the S. thermophils strain. The amount added varied between the conditions to target the same acidification profile in the cheese vat while the cheeses making process was kept constant. The target was to obtain a pH of 5.35 at milling and a final moisture in non- fat solids (MNFS) of the cheeses at \(54\% \pm 0.5\%\) measured after 2 weeks of ripening. The designed defined- strain SLAB for conditions - ALL, - HP, and - LB were all dosed using \(8.1\mathrm{g}\) of culture/100L, whereas the designed defined- strain SLAB for condition - ST was dosed at a concentration of \(17.3\mathrm{g / 100L}\) to compensated for missing contribution to acidification from S. thermophils. + +## Cheese making + +Cheeses were produced at the Application and Technology Center of Chr. Hansen A/S (Hørsholm, Denmark). The composition of the pasteurized cheese milk (organic milk, Naturmælk, Denmark) was measured using a Milkscan™ (FOSS, Hillerød, Denmark), and the fat level was adjusted with pasteurized cream (organic cream \(38\%\) fat, Naturmælk, Denmark) to obtain a protein to fat ratio of 0.90. The cheese milk was heated to \(32^{\circ}\mathrm{C}\) and each cheese vat was filled with \(150\mathrm{kg}\) . The milk was ripened for 40 min with the starter culture before the coagulant (CHY- MAX® Plus, Chr. Hansen A/S) was added. After 30 min of coagulation, the gel was cut into cubes \(10\mathrm{x}10\mathrm{mm}\) . The whey and the cheese grains were stirred for 10 min before heating to \(38^{\circ}\mathrm{C}\) over 40 min and the final scalding and stirring period was 45 min before whey drainage at pH 6.4- 6.5. After 20 min, the cheese curd was cut into 12 blocks and rearranged in blocks 3 times during the next 75 min until pH 5.35 was obtained. The blocks where then milled, and the chips were dry salted (1.7- 2.0% salt in dry matter) by manually adding and mixing the salt 2 times during 15 min. The salted chips were mixed every 5 min over a period of 30 min. Following molding, the chips were pressed at 2 bars for 15 min followed by 5 bars for 17 hours. The cheeses were vacuum- packed and stored at \(9^{\circ}\mathrm{C}\) until sampling after 0.5, 3, 6, 9 and 12 months. The cheese gross composition (moisture, fat, protein and total solids) was estimated after 2 weeks of ripening by FoodScan™ (FOSS, Hillerød, Denmark). The NaCl content was estimated by analyzing the chloride concentration by automated potentiometric + +<--- Page Split ---> + +endpoint titration (DL50, Mettler- Toledo A/S, Glostrup, Denmark). pH was measured potentiometrically (PHC2002- 8, Radiometer Analytical SAS, Lyon, France) in a paste prepared by mixing \(10\mathrm{g}\) of grated cheese with \(10\mathrm{ml}\) of deionized water with a wooden spatula. All analysis were carried out in duplicate. + +In summary, for the one year- long ripening cheese making experiment we inoculated three variations of the SLAB cultures into four milk tanks. Two milk tanks were inoculated with the complete set of strains, one used as the control and the other to test a different approach of culture inoculations (All and All_HP, respectively). The two other variations of the culture were prepared using the 'strain removal strategy' with the aim of testing the effect of \(L\) lactis blend (All- LB) and of the single \(S\) thermophilus (All- ST) during cheese- making. Following the acidification phase, we sampled the resulting cheddar cheeses in 5 time points over the one year period. We followed two batch experiments started on two consecutive days. + +## Microbial community dynamics + +Grated cheese (5.0 g) was mixed with \(45.0\mathrm{g}\) of autoclaved \(2\%\) sodium citrate buffer, \(\mathrm{pH}7.5,46^{\circ}\mathrm{C}\) . The mixture was homogenized in a Stomacher® blender for 4 minutes at medium setting to dissolve the cheese and suspend the bacteria present. Sequential of 10- fold dilutions were prepared as required in \(0.1\%\) peptone/0.15M NaCl in water, \(\mathrm{pH}7.0\) . The total population of Lactococcus spp. was determined on pour plated M17 Agar (Difco, USA) after aerobic incubation for 5 days at \(30^{\circ}\mathrm{C}\) . The total population of thermophilic cocci (ST) was determined on pour plated M17 Agar after aerobic incubation for 3 days at \(37^{\circ}\mathrm{C}\) . The total population of non- starter lactic acid bacteria (NS- LAB) was determined on overlaid Rogosa Agar (Sigma Aldrich, USA) after incubation for 7 days at \(37^{\circ}\mathrm{C}\) . The microbial analysis was carried out in duplicate at each sampling point. + +## Carbohydrates and organic acids in cheese + +The content of lactose, glucose, galactose, lactate, citrate and acetate were quantified using a Dionex ICS- 3000 RFIC- EGTM dual system equipped with an amperometric detector (Dionex, Sunnyvale, CA, USA). The separation was performed using an anion- exchange column (CarboPac® PA20, \(3\mathrm{x}150\mathrm{mm}\) , \(6.5\mu \mathrm{m}\) ) and an ion- exclusion column (IonPac® ICE- AS6, \(9\mathrm{x}250\mathrm{mm}\) , \(8\mu \mathrm{m}\) ). Grated cheese (3.0 g) was mixed with \(15\mathrm{ml}83\mathrm{mM}\) PCA containing \(2\mathrm{mM}\) Na- EDTA, \(33\mathrm{mM}\) arabinose and \(48\mathrm{mM}\) 2- hydroxyisobutyric acid + +<--- Page Split ---> + +(internal standards) and rotated for \(30\mathrm{min}\) at room temperature. This extract was subsequent centrifuged \((5000\mathrm{ng},30\mathrm{min},4^{\circ}\mathrm{C})\) and the supernatant was filtered \((0.45\mu \mathrm{m})\) . The supernatant was diluted (600- fold) before analysis. For analysis of carbohydrates \(25\mu \mathrm{L}\) was injected and the separation was performed at \(26^{\circ}\mathrm{C}\) with a flow rate of \(1.3\mathrm{mL / min}\) increasing the KOH concentration as follow: \(1\mathrm{mM}\) KOH for \(5\mathrm{min}\) from \(1\mathrm{mM}\) to \(20\mathrm{mM}\) KOH over \(0.5\mathrm{min}\) , then kept at \(20\mathrm{mM}\) KOH for \(6.6\mathrm{min}\) min before re- equilibration to \(1\mathrm{mM}\) KOH over \(5\mathrm{min}\) . For analysis of organic acids \(50\mu \mathrm{L}\) was injected and the separation was performed at \(30^{\circ}\mathrm{C}\) with a flow rate of \(1\mathrm{mL / min}\) using \(0.4\mathrm{mM}\) heptfluorobutyric acid in \(5\%\) acetonitrile as eluent. The analysis was performed in duplicate at each sampling point. + +## Proteolysis in cheese + +Grated cheese \((2.0\mathrm{g})\) was mixed with \(18\mathrm{ml}\) distilled water and blended for \(2\mathrm{min}\) at \(25,000\mathrm{rpm}\) (Ultra- Turrax model T25). This suspension was used for determination of total nitrogen (TN), Non- Casein nitrogen (NCN), and non- protein nitrogen (NPN) using the fractionation procedure described in standard NF ISO 27871 (ISO, 2011). The nitrogen content for each fraction was determined using the Kjeldahl method as described in standard NF EN ISO 8968- 1 (ISO, 2014). The level of primary proteolysis (NCN) and secondary proteolysis (NPN) were expressed as percentage of TN found in the cheese. The analysis was performed in duplicate at each sampling point. Finally, the metabolic profile of the cheeses was quantified by measuring 30, 3, 28 and 248 features belonging to the classes of acids, sugars, flavor- related organic compounds and peptides, respectively. + +## Genome sequencing and assembly + +Genomic DNA for de novo short read whole- genome shotgun sequencing (WGS) was extracted from \(1\mathrm{mL}\) of overnight culture (M17 Broth (Difco) supplemented with glucose) of each of the strains (at \(OD600 = 1\) ) with DNeasy Blood and Tissue kit on QiaCube system (Qiagen, Germany) following the manufacturer's protocol. Prior to extraction, cell pellets were washed twice in TES buffer ( \(50\mathrm{mM}\) TRIS pH 8.0, \(1\mathrm{mM}\) EDTA pH 8.5, \(20\%\) sucrose) and then resuspended in \(180\mathrm{uL}\) of pre- lysis TET buffer (20 mM TRIS- Cl pH 8.0, \(2\mathrm{mM}\) EDTA pH 8.5, \(1.2\%\) Triton X- 100, \(20\mathrm{mg / mL}\) lysozyme, \(2\mu \mathrm{L}\) of \(25\mathrm{U / \mu L}\) mutantsolysin, \(4\mu \mathrm{l}\) of \(100\mathrm{mg / mL}\) RNase A). Genomic libraries were prepared using KAPA LTP Library + +<--- Page Split ---> + +Preparation Kit KK8230 (Roche, Switzerland) on a Biomek 4000 Liquid Handler (Beckman Coulter, USA). A portion of \(500\mathrm{ng}\) of genomic DNA diluted in EB buffer (Tris- Cl, pH 8.0) was mechanically fragmented on Bioruptor® Standard (Diagenode, USA) with 12 sonication cycles (30 sec ON/OFF) to obtain an average fragment size of \(300\mathrm{bp}\) (CV%: 1.5). Fragmented DNA was processed following the KK8230 kit manufacture's protocol. Following the adapter ligation step, adapter- modified DNA fragments were enriched by 8- cycle PCR. AMPure XP (Beckman Coulter) paramagnetic beads were used for clean- ups to purify fragments at average size between 450 to 550 bp. Concentration of gDNA and double stranded DNA libraries were measured by Qubit® Fluorimeter using Qubit dsDNA Broad range and Qubit 1x dsDNA HS assays (Thermo Fisher Scientific, USA), respectively. Average dsDNA library size distribution was determined using an Agilent HS NGS Fragment (1- 6000 bp) kit on an Agilent Fragment Analyzer (Agilent Technologies, USA). Libraries were normalized and pooled in NPB solution (10 mM Tris- Cl, pH 8.0, \(0.05\%\) Tween 20) to the final concentration of \(10\mathrm{nM}\) . Following denaturation in \(0.2\mathrm{NNaOH}\) , \(10\mathrm{pM}\) of pooled libraries in \(600\mu \mathrm{L}\) ice- cold HT1 buffer were loaded onto the flow cell provided in the MiSeq Reagent kit v3 (600 cycles) and sequenced on a MiSeq platform (Illumina Inc., San Diego, USA) with a paired- end protocol and read lengths of 301 nucleotides. + +All processing of the short reads was done in either CLC Genomics Workbench versions 9.5.3, 9.5.4 or 10.1.1. The short reads were mapped with default parameters to the reference sequence of the phage Phi X 174 using the tool "Map reads to reference". Unmapped reads from the mapping were trimmed for quality using the PHRED score 23 as the threshold and with the non- default parameter of discarding reads that were less than 50 base pairs long using the tool "Trim Sequences". The trimmed reads were de novo assembled with default parameters except for the minimum contig length was set to 600 base pairs using the tool "De Novo Assembly". Afterwards, a decontamination step was performed where contigs with low depth of coverage were removed using a custom plugin written by Qiagen. The decontamination step first removes all contigs where the depth of coverage is below 15X and afterwards removes all contigs where the depth of coverage is below \(25\%\) of the median depth of coverage for the entire genome assembly. Gene calling of the filtered contigs was done with Prodigal version 2.6.2 using the default parameters. Finally, the genome assemblies with annotated genes were functionally annotated with BLAST against a local annotation database using a custom plugin written by Qiagen. + +<--- Page Split ---> + +## Comparative genomics and Phylogenomics + +A L. lactis and L. cremoris pangenome was created without strain LC- LB16 which found to have 6 times more singletons than all the other strains. The number of core, accessory, and unique (singletons) genes in the pangenome is 1247, 3308, and 2323, respectively. The OGs from this pangenome were used in the comparative genomics and differential metatranscriptome analyses as well as in the creation of a pan- metabolic network of L. lactis and L. cremoris for microbial community modeling. We identified all k- mers of length up to 31 bps (k=31) in the genomes of all the strains and unique k- mers in each strain were filtered. For the set of unique k- mers we mapped the metatrascriptomic reads to get their abundance and kept only those above 7 counts. Normalization of the number of counts per strain performed based on the number of unique k- mers identified in each of the strains. From the total pool of 54667 genes \(15.8\%\) was removed as the cluster above 96 cd- hit identity was mixed among the three taxa. still \(84.2\%\) consider significant number to look on L. cremoris and L. lactis differences. For the phylogenetic analysis 22 genomes from the SLAB culture were used alongside with 107 complete genomes of L. lactis and L. cremoris, which were retrieved from NCBI results in total to 129 genomes. CD- HIT on \(80\%\) amino acid identity thresholds along with a custom- made R script were used to identify a set of 464 monocore marker gene sets [31, 32]. The trees were constructed using PhyloPhlAn v3.0 [33] using the PhyloPhlAn parameters "- accurate" and "- diversity low", which translate to the usage of a pfasum60 substitution matrix. In addition, multiple sequencing alignments (msa) were performed with muscle and trimming was performed by removing columns with at least one nucleotide appearing above a threshold of 0.99. The final maximum likelihood tree was constructed on the concatenated DNA msa using raxmlHPC, 100 bootstrap and GTRGAMMAI model [34, 35]. + +## Generation and simulation of genome-scale metabolic models + +Genome- scale metabolic models were generated for S. thermophilus, Lactococcus LLm1, LLm2, and LC using CarveMe v1.4.1 [36] based on the protein sequences of assembled genomes. First, Prodigal v2.6.3 [37] was used to generate open- reading- frame (ORF) annotated protein sequence files from the corresponding DNA fasta files. Additionally, milk media from a recent kefir publication was used for + +<--- Page Split ---> + +gap- filling during model generation [28]. More specifically, we formulated four biologically relevant variations of the milk composition, including aerobic rich milk, anaerobic rich milk, aerobic depleted milk, and anaerobic depleted milk. The former two media represent the initial composition of milk, while the latter two media represent milk after it has been depleted by fermentation. A version of each of the four species was generated by gap- filling on each of the four milk media variations, as well as without gap- filling, resulting in a total of 20 models. All models were generated using the default CarveMe universal bacterial model template. Note that although gram positive bacteria are generally not modeled with a periplasmic compartment due to its smaller size relative to gram positive bacteria, all models generated with the automated CarveMe tool contain such a compartment. Individual model simulations were carried out for all models on each of the four media variations using the reframed v1.2.1 and cobrapy v.0.20.0 metabolic modeling packages [38]. Community simulation of genome- scale metabolic models was also carried out in the different media variations, using SMETANA v1.2.0 [15]. All associated data, including code used for generating and plotting results, is available on GitHub. + +## Inoculations in milk + +All milk fermentations were carried out in organic low fat (1.5%) milk from the organic dairy NATUR- M/ELK, Tinglev, Denmark. which consists of skim milk powder at a level of dry matter of 9.5% (w/v) reconstituted in distilled water and pasteurized at \(99^{\circ}\mathrm{C}\) for 30 min, followed by cooling to \(30^{\circ}\mathrm{C}\) (Sørensen et al., 2016). Starting material for all inoculations of were concentrated F- DVS® (Direct Vat Set) strains and cultures for direct inoculation into. All cultures were inoculated \(0.02\%\) with F- DVS material by firstly transfer \(2.0\mathrm{gF - DVS}\) material to \(200\mathrm{ml}\) cold milk. After mixing, \(4\mathrm{ml}\) were transferred to \(200\mathrm{ml}\) pre- warmed milk ( \(32^{\circ}\mathrm{C}\) ) for the final fermentation. The addition percentages of the different strains and cultures for the 6 different culture blends is given in Supplementary Table 15. + +## Milk acidifications and sampling + +After inoculation into bottles with milk placed at \(32^{\circ}\mathrm{C}\) in a waterbath, pH measurement was started and acidification was followed using a CINAC system (Corrieu et al., 1988). After 59 min, the temperature was raised to \(38^{\circ}\mathrm{C}\) and at 183 min, the temperature was reduced to the final temperature, \(35^{\circ}\mathrm{C}\) , for 20 + +<--- Page Split ---> + +hours. The 6 different culture blends were each run in triplicate. After around 5 hours of acidification during the exponential growth phase when reaching pH 5, 1 g fermented milk from each fermentation and 1 g unfermented milk were sampled and added \(200\mu 14\mathrm{NH}2\mathrm{SO}4\) to prepare for analysis for organic acids, Carbohydrates and Volatiles. In addition 1 g was samples for mRNA extraction. + +## Carbohydrates, acids and volatiles in milk + +For the preparation of fermented milk samples used for carbohydrate and small organic acid analysis, 1 g or 1 mL of sample was quenched with \(200\mu \mathrm{L}4\mathrm{N}\) sulfuric acid (H2SO4) in a \(7\mathrm{mL}\) glass tube, mixed and stored at \(- 20^{\circ}\mathrm{C}\) until analysis. The analytes for carbohydrate analysis were extracted from the sample and proteins get deproteinated and precipitated by treatment with an aqueous perchloric acid (PCA) solution. The samples get further diluted to fit into the dynamic range of the quantification. Arabinose is added as an internal standard. The diluted samples are analyzed on a Dionex ICS- 3000 system (Thermo Fischer Scientific, Waltham (MA), USA) using an analytical anion- exchange column and a pulsed amperometric detector (PAD). For quantification a one- point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (arabinose). The analytes for analysis of small organic acids were extracted from the sample and proteins get deproteinated and precipitated adding an aqueous PCA solution. The samples get further diluted to fit into the dynamic range of the quantification. Adipic acid is added as an internal standard. The diluted samples are analyzed on a Dionex ICS- 3000 or ICS- 5000 system (Thermo Fischer Scientific, Waltham, MA, USA) using an analytical ion exclusion column and a suppressed conductivity detector (SCD). For quantification, an 8- point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (adipic acid). + +For the preparation of the volatile organic compounds (VOC's) samples, 1 g or 1 mL of each sample was transferred to a headspace vial (20 mL) with \(200\mu \mathrm{L}\) of 4N H2SO4 and sealed with teflon- lined aluminium caps. The samples were then identified using a static head space sampler connected to a Gas Chromatograph with Flame Ionization Detector (GC- FID) (Perkin Elmer, MA, USA) and equipped with a HP- FFAP column. The identification of VOC's was based on retention time in comparison with that of standards. The injector and detector were maintained at \(180^{\circ}\mathrm{C}\) and \(220^{\circ}\mathrm{C}\) , respectively. The oven + +<--- Page Split ---> + +was initially heated to \(60^{\circ}\mathrm{C}\) and held for \(2\mathrm{min}\) , then increased to \(230^{\circ}\mathrm{C}\) and held for \(0.5\mathrm{min}\) . The calculation of the concentration of each compound in each sample was based on the peak height divided by the response factor. The response factor is established suing standard solutions by the quotient of the peak height divided by the known sample concentration. All of the chemicals and analytes used for the standard solutions were provided by Sigma- Aldrich, Munich, Germany. Overall with these approaches, we measured a total of 43 metabolites comprising five carbohydrates, eight organic acids and 30 important flavor compounds. + +## RNA extraction, sequencing and analysis + +Total RNA was extracted using the RNeasy Mini kit (Qiagen 74104) and rRNA was depleted with the NEBNext rRNA Depletion Kit (Bacteria). The sequencing library was then prepared using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina and the libraries were sequenced by EMBL Genomic Core Facility using the Illumina NextSeq 500 system, read length 150bp. The RNA- reads are preprocessed and filtered based on minimum quality and length using the NGLess pipeline and substrim. Given a read it finds the longest substring, such that all bases are of at least the given quality. The parameters minimum quality and minimum length were set to 25 and 100, respectively. The filtered reads are mapped to the genomes of all strains present in the Culture using bbmap – if one read has multiple hits random distribution is used (selecting one top- scoring site randomly). The total number of reads ranges from 4,855,868 to 18,812,096 in samples 4B and 1B, respectively, and the percentage of mapped reads is in the range of 93- 95% (Supplementary Table 5). htseq- count was used to count the number of reads per gene. The resulting counts grouped per orthology group (custom python script) based on a previously created pangenome created for the pan- metabolic network. Normalization was performed and differential gene expression with DESeq2 and SARTools in R [39, 40]. Further differentially expressed genes, Lactococcus pan- metabolic network information and KEGG pathways were based on custom analysis [41, 21]. The data was normalized using DESeq2 to correct systematic technical biases and make it possible to compare read counts across samples. The median scaling factor for each sample was used. This is done for the data containing all genes. RegPrecise database was use for validation of key functional regulation [42]. + +<--- Page Split ---> + +## Acknowledgments + +AcknowledgmentsThe authors wish to thank Jannik Vindeloev and Ana Rute Neves for helpful discussions. Karsten Hellmuth, Eric Johansen and Mads Bennedsen for their help with project management and funding acquisition. Also, Kosai Al- Nakeeb, Anna Koza, Jacob Bælum, and Martin Abel- Kistrup for their significant contribution at genome sequencing and assembly. Lisandra Zepeda and Ida Bærholm Schnell for their help during cheese sampling. Finally, Gunnar Øregaard for providing acidification data for individual strains. The research of CM was supported by a Grand Solution grant from Innovation Fund Denmark (grant no. 6150- 00033B), The FoodTranscriptomics project and by the Dutch Research Council, as part of the MiCRop Consortium (NWO/OCW grant no. 024.004.014) + +## Data Availability Statement + +Data Availability StatementProcessed data, including metabolomics, metatranscriptomics count tables as well as the code to reproduce the results are available in https://github.com/Chrats- Melkonian/mi_cheese. Genomes and raw metatranscriptomics data will become available before publication. + +## Author contributions statement + +Author contributions statementCM, organizing ideas, writing, comparative genomics, phylogenomics, transcriptomics and bioinformatics analyses. FZ, genome- scale metabolic modeling, community simulations and writing. IK, initial transcriptomics, bioinformatics analyses and genome- scale metabolic modeling. MJ and KSr, carried out the experiments and drafted the manuscript part relating to sections “Inoculations in milk” and “Milk acidifications and sampling”. SB, mRNA extraction of the milk fermentation samples for mRNA sequencing. DM, analysed initial metatranscriptomics data, performed model reconstruction and community simulations. LTA, participated in conceptualization of the study, design and supervise the year- long cheddar- making experiment. KrP, participated in conceptualization of the study, funding acquisition, supervision of modeling and critically revising the manuscript. AZ, participated in conceptualization of the study, funding acquisition, supervision of experimental design, data analysis and bioinformatics, and + +<--- Page Split ---> + +critically revising the manuscript. + +## Declaration of competing interest + +CM, IK, MJ, KsR, LtA and AAZ are present or previous employees at Chr. Hansen A/S, a global supplier microbial cultures for food fermentation. The authors' views presented in this study, however, are solely based on scientific grounds and do not reflect the commercial interest of their employer. + +## References + +1. Smith J, Yeluripati J, Smith P, Nayak DR. Potential yield challenges to scale-up of zero budget natural farming. Nature Sustainability. 2020 Mar;3(3):247-252. Available from: https://doi.org/10.1038/s41893-019-0469-x. + +2. Wastyk HC, Fragiadakis GK, Perelman D, Dahan D, Merrill BD, Yu FB, et al. Gut-microbiota-targeted diets modulate human immune status. Cell. 2021;184(16):4137-4153. 4137. Available from: https://doi.org/10.1016/j.cell.2021.06.019. + +3. Blaya J, Barzideh Z, LaPointe G. Symposium review: Interaction of starter cultures and nonstarter lactic acid bacteria in the cheese environment. Journal of Dairy Science. 2018 Apr;101(4):3611-3629. Available from: https://doi.org/10.3168/jds.2017-13345. + +4. Song AA, In LLA, Lim SHE, Rahim RA. Erratum to: A review on Lactococcus lactis: from food to factory. Microbial Cell Factories. 2017 Aug;16(1). Available from: https://doi.org/10.1186/s12934-017-0754-1. + +5. Markakiou S, Gaspar P, Johansen E, Zeidan AA, Neves AR. Harnessing the metabolic potential of Streptococcus thermophilus for new biotechnological applications. Current Opinion in Biotechnology. 2020 Feb;61:142-152. Available from: https://doi.org/10.1016/j.copbio.2019.12.019. + +6. Rau MH, Gaspar P, Jensen ML, Geppel A, Neves AR, Zeidan AA. Genome-Scale Metabolic Modeling Combined with Transcriptome Profiling Provides Mechanistic Understanding of Streptococcus ther + +<--- Page Split ---> + +547 mophilus CH8 Metabolism. Applied and Environmental Microbiology. 2022 Aug;88(16). Available from: https://doi.org/10.1128/aem.00780- 22. + +7. Pastink MI, Teusink B, Hols P, Visser S, de Vos WM, Hugenholtz J. Genome-Scale Model of Streptococcus thermophilus LMG18311 for Metabolic Comparison of Lactic Acid Bacteria. Applied and Environmental Microbiology. 2009 Jun;75(11):3627-3633. Available from: https://doi.org/10.1128/aem.00138-09. + +8. Oliveira AP, Nielsen J, Förster J. Modeling Lactococcus lactis using a genome-scale flux model. BMC Microbiology. 2005 Jun;5(1). Available from: https://doi.org/10.1186/1471-2180-5-39. + +9. Champagne CP, Gagnon D, St-Gelais D, Vuillemand JC. Interactions between Lactococcus lactis and Streptococcus thermophilus strains in Cheddar cheese processing conditions. International Dairy Journal. 2009 Nov;19(11):669-674. Available from: https://doi.org/10.1016/j.idairyj.2009.06.002. + +10. Duru IC, Laine P, Andreevskaya M, Paulin L, Kananen S, Tynkkynen S, et al. Metagenomic and metatranscriptomic analysis of the microbial community in Swiss-type Maasdam cheese during ripening. International Journal of Food Microbiology. 2018 Sep;281:10-22. Available from: https://doi.org/10.1016/j.ijfoodmicro.2018.05.017. + +11. Salazar JK, Carstens CK, Ramachandran P, Shazer AG, Narula SS, Reed E, et al. Metagenomics of pasteurized and unpasteurized gouda cheese using targeted 16S rDNA sequencing. BMC Microbiology. 2018 Nov;18(1). Available from: https://doi.org/10.1186/s12866-018-1323-4. + +12. Walsh AM, Macori G, Kilcawley KN, Cotter PD. Meta-analysis of cheese microbiomes highlights contributions to multiple aspects of quality. Nature Food. 2020 Aug;1(8):500-510. Available from: https://doi.org/10.1038/s43016-020-0129-3. + +13. Hutkins R, Halambeck SM, Morris HA. Use of Galactose-Fermenting Streptococcus thermophilus in the Manufacture of Swiss, Mozzarella, and Short-Method Cheddar Cheese. Journal of Dairy Science. 1986 Jan;69(1):1-8. Available from: https://doi.org/10.3168/jds.s0022-0302(86)80361-7. + +<--- Page Split ---> + +14. Proust L, Haudebourg E, Sourabié A, Pedersen M, Besançon I, Monnet V, et al. Multi-omics Approach Reveals How Yeast Extract Peptides Shape Streptococcus thermophilus Metabolism. Applied and Environmental Microbiology. 2020 oct;86(22). Available from: https://doi.org/10.1128% 2Faem.01446-20. + +15. Zeleznik A, Andrejev S, Ponomarova O, Mende DR, Bork P, Patil KR. Metabolic dependencies drive species co-occurrence in diverse microbial communities. Proceedings of the National Academy of Sciences. 2015 May;112(20):6449-6454. Available from: https://doi.org/10.1073/pnas.1421834112. + +16. Nugroho ADW, Kleerebezem M, Bachmann H. Growth, dormancy and lysis: the complex relation of starter culture physiology and cheese flavour formation. Current Opinion in Food Science. 2021;39:22-30. + +17. Smid EJ, Kleerebezem M. Production of Aroma Compounds in Lactic Fermentations. Annual Review of Food Science and Technology. 2014 feb;5(1):313-326. Available from: https://doi.org/10.1146%2Fannurev-food-030713-092339. + +18. Calbert HE, Price WV. A Study of the Diacetyl in Cheese. I. Diacetyl Content and Flavor of Cheddar Cheese. Journal of Dairy Science. 1949 jun;32(6):515-520. Available from: https://doi.org/10.3168%2Fjds.s0022-0302%2849%2992073-1. + +19. Baumler AJ, Sperandio V. Interactions between the microbiota and pathogenic bacteria in the gut. Nature. 2016 Jul;535(7610):85-93. Available from: https://doi.org/10.1038/nature18849. + +20. Lara EG, van der Windt I, Molenaar D, de Vos MGJ, Melkonian C. Using Functional Annotations to Study Pairwise Interactions in Urinary Tract Infection Communities. Genes. 2021 aug;12(8):1221. Available from: https://doi.org/10.3390%2Fgenes12081221. + +21. Melkonian C, Fillinger L, Atashgahi S, da Rocha UN, Kuiper E, Olivier B, et al. High biodiversity in a benzene-degrading nitrate-reducing culture is sustained by a few primary consumers. Communications Biology. 2021 may;4(1). Available from: https://doi.org/10.1038%2Fs42003-021-01948-y. + +<--- Page Split ---> + +22. Melkonian C, Seidl MF, van der Hooft JJJ, de Vos MGJ. Metabolic interactions shape a community's phenotype. Trends in Microbiology. 2022 jul;30(7):609-611. Available from: https://doi.org/10.1016%2Fj.tim.2022.05.001. + +23. Machado D, Maistrenko OM, Andrejev S, Kim Y, Bork P, Patil KR, et al. Polarization of microbial communities between competitive and cooperative metabolism. Nature Ecology & Evolution. 2021 Jan;5(2):195-203. Available from: https://doi.org/10.1038/s41559-020-01353-4. + +24. Irlinger F, Mounier J. Microbial interactions in cheese: implications for cheese quality and safety. Current Opinion in Biotechnology. 2009 Apr;20(2):142-148. Available from: https://doi.org/10.1016/j.copbio.2009.02.016. + +25. Jiang N, Wu R, Wu C, Wang R, Wu J, Shi H. Multi-omics approaches to elucidate the role of interactions between microbial communities in cheese flavor and quality. Food Reviews International. 2022 Apr;1-13. Available from: https://doi.org/10.1080/87559129.2022.2070199. + +26. Parente E, Zotta T, Ricciardi A. A review of methods for the inference and experimental confirmation of microbial association networks in cheese. International Journal of Food Microbiology. 2022 May;368:109618. Available from: https://doi.org/10.1016/j.ijfoodmicro.2022.109618. + +27. Ponomarova O, Gabrielli N, Sévin DC, Mülleder M, Zirngibl K, Bulyha K, et al. Yeast Creates a Niche for Symbiotic Lactic Acid Bacteria through Nitrogen Overflow. Cell Systems. 2017 Oct;5(4):345-357.e6. Available from: https://doi.org/10.1016/j.cels.2017.09.002. + +28. Blasche S, Kim Y, Mars RA, Machado D, Maansson M, Kafka E, et al. Metabolic cooperation and spatiotemporal niche partitioning in a kefir microbial community. Nature microbiology. 2021;6(2):196-208. + +29. Goyal A, Bittleston LS, Leventhal GE, Lu L, Cordero OX. Interactions between strains govern the eco-evolutionary dynamics of microbial communities. eLife. 2022 feb;11. Available from: https://doi.org/10.7554%2Felife.74987. + +<--- Page Split ---> + +30. Oren A, Garrity GM. Notification that new names of prokaryotes, new combinations and new taxonomic opinions have appeared in volume 71, part 3 of the IJSEM. International Journal of Systematic and Evolutionary Microbiology. 2021 Jun;71(6). Available from: https://doi.org/10.1099/ijsem.0.004812. + +31. Fu L, Niu B, Zhu Z, Wu S, Li W. CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics. 2012 oct;28(23):3150-3152. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbts565. + +32. R Core Team. R: A language and environment for statistical computing. Vienna, Austria; 2020. Available from: https://www.R-project.org/. + +33. Asnicar F, Thomas AM, Beghini F, Mengoni C, Manara S, Manghi P, et al. Precise phylogenetic analysis of microbial isolates and genomes from metagenomes using PhyloPhlAn 3.0. Nature Communications. 2020 may;11(1). Available from: https://doi.org/10.1038%2Fs41467-020-16366-7. + +34. Edgar RC. MUSCLE: multiple sequence alignment with high accuracy and high throughput. Nucleic Acids Research. 2004 mar;32(5):1792-1797. Available from: https://doi.org/10.1093%2Fnar%2Fgkh340. + +35. Stamatakis A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics. 2014 jan;30(9):1312-1313. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbtu033. + +36. Machado D, Andrejev S, Tramontano M, Patil KR. Fast automated reconstruction of genome-scale metabolic models for microbial species and communities. Nucleic Acids Research. 2018 06;46(15):7542-7553. Available from: https://doi.org/10.1093/nar/gky537. + +37. Hyatt D, Chen GL, LoCascio PF, Land ML, Larimer FW, Hauser LJ. Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC bioinformatics. 2010;11(1):1-11. + +38. Ebrahim A, Lerman JA, Palsson BO, Hyduke DR. COBRApy: constraints-based reconstruction and analysis for python. BMC systems biology. 2013;7(1):1-6. + +<--- Page Split ---> + +652 39. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology. 2014 dec;15(12). Available from: https://doi.org/10.1186%2Fs13059- 014- 0550- 8. + +655 40. Varet H, Brillet-Guéguen L, Coppée JY, Dillies MA. SARTools: A DESeq2- and EdgeR- Based R Pipeline for Comprehensive Differential Analysis of RNA-Seq Data. PLOS ONE. 2016 jun;11(6):e0157022. Available from: https://doi.org/10.1371%2Fjournal.pone.0157022. + +659 41. Melkonian C, Gottstein W, Blasche S, Kim Y, Abel-Kistrup M, Swiegers H, et al. Finding Functional Differences Between Species in a Microbial Community: Case Studies in Wine Fermentation and Kefir Culture. Frontiers in Microbiology. 2019 jun;10. Available from: https://doi.org/10.3389%2Ffmicb.2019.01347. + +663 42. Novichkov PS, Kazakov AE, Ravcheev DA, Leyn SA, Kovaleva GY, Sutormin RA, et al. Reg- Precise 3.0 - A resource for genome-scale exploration of transcriptional regulation in bacteria. BMC Genomics. 2013;14(1):745. Available from: https://doi.org/10.1186/1471- 2164- 14- 745. + +## figures and tables + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Schematic representation of the experimental design and methods (a) The 1 year long cheddar-making experiment. Microbial population dynamics were quantified during cheese ripening using a selective method of viable cell counting, which discriminates between thermophilic cocci, mesophilic cocci and non-starter lactic acid bacteria (NSLAB). Additionally, metabolic changes in the cheeses were measured using several targeted analytical chemistry approaches. (b) Schematic representation of the controlled milk experiment in the laboratory as well as of the methods (1-5) used in the controlled milk experiment. This second experiment involves the removal of additional strains, namely the individual exclusion of three major Lactococcus strains. The numbers indicate the different type of data and analysis as follows: 1. metatrancriptomics, 2. metabolomics, 3. genomics, 4. phylogenomics and 5. genomes-scale metabolic models (GEMs) and community simulations. Our integrative systems biology approach combined: i) the analysis of the SLAB community’s genomes, ii) the generation and simulation of their respective GEMs, iii) the analysis of the metatranscriptomes across the different strain removal conditions and iv) the quantification of key metabolites.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. S. thermophilus benefit the growth of lactococci community and influence the final flavor of cheese. (a-d) Microbial population dynamics during cheese ripening. Relationship between CFUs and time (months) for (a) Mesophilic, (c) Thermophilic cocci, and (d) non-starter lactic acid bacteria (NSLAB). (b) Presents the boxplot comparison of the different condition on Mesophilic cocci at 12 months. Note, significant reduction only on Mesophilic cocci is observed when S. thermophilus is absent. Condition All corresponds to whole culture which is composed from 24 number of strains. HP corresponds to an alternative method (hand packed) of the whole culture inoculation. -LB corresponds to the removal of a L. lactis blend population. -ST corresponds to the removal of the S. thermophilus strain. (e-h) Metabolome dynamics during cheese ripening. (e) 2-dimensional representation with the usage of UMAP of all samples based on the metabolomics measurements (Acids, Carbohydrates and Peptides). The colour indicates the time when the sample was taken and the shape indicates the four different conditions. Note, the stronger change is observed between 2 week and 3 months while the removal of ST (cross symbol) has a strong effect on the later time of cheese ripening. (f) Relative change of the metabolites on different time intervals including all the conditions, both colour and shape indicates the class of the metabolite. Note, the higher relative change of peptides followed by acids at the interval between 2 week and 3 months, later acids exhibit higher changes. The three measured sugars (Glucose, Lactose and Galactose) are excluded from this panel. (g) Selection of the six most discriminative metabolites out of 50 for the four conditions, colour indicate the condition while the right side shape the class of the metabolite. Note, the majority of peptides concentration are significant different when S. thermophilus is not present as well as with galactose, lactose and lactic acid. (h) Galactose concentration over time highlights the absence of galactose when S. thermophilus is not present. The galactose concentration remains steady from the start till the 9 month of cheese ripening and then shows signs of decline.
+ +<--- Page Split ---> +![](images/Supplementary_Figure_5.jpg) + +
Figure 3. The removal of different Lactococcus strains has a distinct effect on gene expression in S. thermophilus. (a-e) relationship of the Lactococcus phylogeny with the S. thermophilus transcription change on different dropout conditions. (a) Lactococcus phylogenetic tree separates the L. cremoris and L. lactis strains. The two clades are colored with transparent yellow and grey, respectively. The colors and shape of the tree tips indicate the strains presents in the SLAB culture. Tree empty tips correspond to the obtained complete NCBI genomes of Lactococcus. Additional information is presented in the outer layers of the tree. The first layer presents the unique k-mer content based on the SLAB culture's genomes (purple-scale color) and indicates higher number for the strains who have no close relative within the culture. The second layer presents the proxy of temperature stress tolerance based on Pearson Correlation Coefficient (PCC) between individual acidification curve of 30 and 40 degree \(^\circ \mathrm{C}\) (red-scale color). High values correspond to high temperature tolerance. The clade L. cremoris shows lower temperature tolerance, with few exceptions including the main LC strain (more elaborate data are presented in Supplementary Fig. 5). The third layer presents the percentile of transcribed sigletons (green-scale color) as well as the total number of the strains sigletons (outer bars). (b-e) Volcano plots presents the S. thermophilus transcription between the whole community (All) and the different Lactococcus dropout conditions, respectively. The colors corresponds to the 4 Lactococcus components. The S. thermophilus transcription observed to change more when LLm1 was left-out followed by LB, with 291 and 182 number of total significant up/down regulated genes, respectively. A smaller effect of 21 genes was observed when LC was left-out and only 1 gene when LLm2 was left out.
+ +<--- Page Split ---> +![](images/Supplementary_Figure_6.jpg) + +
Figure 4. S. thermophilus provides nitrogen in form of amino acids to L. lactis community, which is necessary for de novo nucleotide biosynthesis. (a) Summary of metabolic modeling analysis including individual model and community-based simulations. (b) Selected exchange fluxes predicted across individual flux balance analysis simulations carried in milk media variations for each set of metabolic models, in particular fermentation products, amino acids, and carbon sources highlighting differences in metabolic strategies across species. (c) Alluvial diagram showing predicted metabolic exchanges from community simulations in milk media variants, highlighting the fact the amino acid valine is strongly predicted across all simulation conditions where metabolic cross-talk is expected. Refer to Supplementary Figure 6 for more details on the metabolic models and simulation. (d-i) Transcriptomics profiles between All and -ST (S. thermophilus is left out) conditions of key functions for all the strains in SLAB culture. (d-f) Transcriptomics boxplots of branched chain amino acids (BCAA) aminotransferase, BCAA transport system 2 carrier protein and ammonium transporter (T) supports the predictions of metabolic exchanges. Note, (e) the higher transcription profile of S. thermophilus on BCAA transport system as well as the high transcription profile of S. thermophilus and the increase of LC transcription profile when S. thermophilus was left out. (g-j) Up-regulation of Lactococcus glutamine and nucleotide metabolism when S. thermophilus is left out. (g-i) Boxplots present the activities of nitrogen regulatory protein P-II, glutamine synthetase (glnA) and related transcriptional regulator (TR) (GlnR) for all community members. Note, the up-regulation of all the enzymes derived from Lactococcus strains when S. thermophilus is left out. (j) Metatranscriptomics represents the increased activity of Lactococcus reactions, which are incorporated into Escher maps. The pattern shows a coordinate up-regulation (with green) of the enzymes towards guanine biosynthesis (and uridine). (k) Schematic compilation of the compounds S. thermophilus may provide to Lactococcus community as well as selected transcriptional changes of the Lactococcus community when S. thermophilus was left out. The color and direction of the arrows represent the up- and down- regulation with green/up and red/down, respectively.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community The measured metabolome from the controlled milk experiment in response to stain removal conditions (a-d). (a) The PCA of key carbohydrates, acids and cheese flavor compounds. The strongest response observed from the removal of L. cremoris followed by coordinated response of LLm1, LB and S. thermophilus. Removal of LLm2 has no effect. The arrows represents the top 5 compounds (2,3-Pentanedione, Hexanal, Diacetyl, Acetoin and Ethyl acetate) responsible for the observed changes. Panels (b-c) represents the signal-to-noise ratio (S/N) of diacetyl and acetoin, respectively. Regarding those two compounds, absence of S. thermophilus from the community led to significant lower accumulation, while absence of L. cremoris leads to significant higher accumulation. The black horizontal line indicate the average value of the two compounds at milk prior to acidification. Paired t-test for each condition against the whole community (All) were performed with the symbols indicating the statistical significance results: ns: \(\mathrm{p} > 0.05\) , \*: \(\mathrm{p}< = 0.05\) and \*\*: \(\mathrm{p}< = 0.01\) . (d) Schematic compilation of the compounds that changed between the whole SLAB culture and when L. cremoris was left out (for detailed boxplots see supplementary). The color represent the removal conditions; yellow indicates the removal of L. cremoris while red represents the whole SLAB culture. The upwards arrow indicates the increase of the compounds relative to the compounds concentration in milk. The X symbol indicates the lack of production of the compounds in the respective condition. (e) Simplified metabolic escher maps from citrate to \(\alpha\) -ketoglutarate as well as to diacetyl through pyruvate and (S)-2-Acetolactate. Metabolites and reactions are indicated with black and grey color in the graph, respectively. Full colored circles and open colored circles indicate the high and low transcription per strain, respectively. Lack of circle visualization indicates the lack of the corresponding orthologous group genes. Note, the transcriptional fluxes from citrate-sodium symporter towards diacetyl, which are present to all members in Lactococcus community. In addition, LC and LLm1 manifest higher transcription among the Lactococcus community on the genes related to acetolactate decarboxylase. Only LC strain manifests transcriptional flux through diacetyl reductase and butanediol dehydrogenase as well as in reactions towards \(\alpha\) -ketoglutarate. (f) Predicted fluxes based on metabolic modeling across the different simulated media.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- MICheeseSUPP.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323_det.mmd b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9ab52664fc355fa70de9071eca32baf5f3f331c1 --- /dev/null +++ b/preprint/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323/preprint__0e85f9efde7d6b5ff7517772dbef871099da6254a7b2b3ce7363284a60076323_det.mmd @@ -0,0 +1,507 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 852, 175]]<|/det|> +# Inter-species metabolic interactions in cheese flavour formation + +<|ref|>text<|/ref|><|det|>[[44, 195, 890, 240]]<|/det|> +Chrats Melkonian ( \(\boxed{\infty}\) chrats.melkonian@gmail.com) Utrecht University & Wageningen University & Research https://orcid.org/0000- 0002- 1448- 5542 + +<|ref|>text<|/ref|><|det|>[[44, 244, 639, 286]]<|/det|> +Francisco Zorrilla Medical Research Council Toxicology Unit, University of Cambridge + +<|ref|>text<|/ref|><|det|>[[44, 291, 230, 333]]<|/det|> +Inge Kjærbølling Novozymes + +<|ref|>text<|/ref|><|det|>[[44, 338, 268, 377]]<|/det|> +Sonja Blasche University of Cambridge + +<|ref|>text<|/ref|><|det|>[[44, 383, 480, 424]]<|/det|> +Daniel Machado Norwegian University of Science and Technology + +<|ref|>text<|/ref|><|det|>[[44, 429, 163, 468]]<|/det|> +Mette Junge Chr. Hansen + +<|ref|>text<|/ref|><|det|>[[44, 474, 175, 513]]<|/det|> +Kim Soerensen Chr. Hansen + +<|ref|>text<|/ref|><|det|>[[44, 520, 175, 559]]<|/det|> +Lene Andersen Chr. Hansen + +<|ref|>text<|/ref|><|det|>[[44, 565, 625, 606]]<|/det|> +Kiran Patil University of Cambridge https://orcid.org/0000- 0002- 6166- 8640 + +<|ref|>text<|/ref|><|det|>[[44, 612, 175, 651]]<|/det|> +Ahmad Zeidan Chr. Hansen + +<|ref|>sub_title<|/ref|><|det|>[[44, 697, 101, 714]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 733, 899, 776]]<|/det|> +Keywords: Cheddar cheese, Microbial interactions, multi- omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris + +<|ref|>text<|/ref|><|det|>[[44, 794, 314, 813]]<|/det|> +Posted Date: March 16th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 832, 473, 851]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2574132/v1 + +<|ref|>text<|/ref|><|det|>[[44, 870, 905, 911]]<|/det|> +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 930, 531, 950]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 68, 955, 111]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on December 21st, 2023. See the published version at https://doi.org/10.1038/s41467-023-41059-2. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[66, 73, 805, 103]]<|/det|> +# Inter-species metabolic interactions in cheese + +<|ref|>sub_title<|/ref|><|det|>[[66, 127, 358, 155]]<|/det|> +## flavour formation + +<|ref|>text<|/ref|><|det|>[[66, 177, 891, 250]]<|/det|> +3. Chrats Melkonian1,6,7,*, Francisco Zorrilla2, Inge Kjærbølling1, Sonja Blasche2, Daniel Machado3, Mette Junge4, Kim Ib Soerensen4, Lene Tranberg Andersen5, Kiran R Patil2, and Ahmad A. Zeidan1,* + +<|ref|>text<|/ref|><|det|>[[66, 277, 895, 496]]<|/det|> +6. Systems Biology, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark +7. Medical Research Council Toxicology Unit, University of Cambridge, Cambridge, UK +8. Department of Biotechnology and Food Science, Norwegian University of Science and Technology, Trondheim, 7491, Norway +9. Global Application, Food Cultures & Enzymes, Chr. Hansen A/S, 2970 Hørsholm, Denmark +10. Strain Development, R&D Discovery, Chr. Hansen A/S, 2970 Hørsholm, Denmark +11. Theoretical Biology and Bioinformatics, Science for Life, Utrecht University, Utrecht, the Netherlands +12. Bioinformatics Group, Wageningen University and Research, Wageningen, Netherlands +13. Authors to whom correspondence should be addressed: chrats.melkonian@gmail.com, dkahze@chr-hansen.com + +<|ref|>sub_title<|/ref|><|det|>[[68, 537, 220, 557]]<|/det|> +## ABSTRACT + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 123, 910, 572]]<|/det|> +Cheese fermentation and flavour formation are governed by complex biochemical reactions driven by polymicrobial activity. While the compositional dynamics of cheese microbiomes is relatively well mapped, the mechanistic role of microbial interactions in flavour formation is yet unknown. We microbially and metabolically characterised a year- long Cheddar cheese process using a commonly used starter culture containing Streptococcus thermophilus and Lactococcus strains. By using an experimental strategy whereby certain strains were left out from the starting mixture, we identified the critical role of S. thermophilus in boosting Lactococcus growth and in shaping flavour compound profile. Controlled milk fermentations with systematic exclusion of single Lactococcus strains, combined with genomics, genome- scale metabolic modelling, and metatranscriptomics, indicated that proteolytic activity of S. thermophilus relieves nitrogen limitation for Lactococcus and boosts de novo nucleotide biosynthesis. While S. thermophilus had large contribution to the flavour profile, L. cremoris also played a role by limiting diacetyl and acetoin formation which leads to off- flavour when in excess. This off- flavour control could be attributed to different metabolic re- routing of citrate between L. cremoris and other L. lactis strains. Further, closely related L. lactis strains exhibited different interaction patterns with S. thermophilus highlighting the importance of strain- specificity in cheese- making. Overall, our results bring forward the critical role of competitive and cooperative microbial interactions shaping cheese flavour profile. + +<|ref|>text<|/ref|><|det|>[[70, 597, 900, 650]]<|/det|> +Keywords: Cheddar cheese, Microbial interactions, multi- omics, metabolic modeling, Streptococcus thermophilus, Lactococcus lactis, Lactococcus cremoris + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[85, 88, 229, 110]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[80, 133, 912, 368]]<|/det|> +Fermented foods based on microbial consortia (e.g., cheese, kefir and kombucha) constitute a large part of the modern diet with many reported health benefits [1, 2]. In cheese, well defined Starter Lactic Acid Bacteria (SLAB) cultures are used to ensure consistent and flavorful products. During cheese making, microbes encounter dynamic conditions characterized by different nutrient availability and different stresses, starting with an initial feast during milk fermentation, followed by a long period of famine during cheese ripening. Cheese- making thus provides a controlled system with well characterized dynamics to understand the role of microbial interactions in shaping the characteristics and quality of fermented foods [3]. + +<|ref|>text<|/ref|><|det|>[[78, 383, 912, 767]]<|/det|> +In industrial cheese making, SLAB cultures are responsible for milk acidification via lactose fermentation. They are predominantly composed of mesophilic strains of Lactococcus, including the species L. lactis and L. cremoris as well as thermophilic strains of Streptococcus thermophilus. A rich literature exists on the physiology of both species in mono- cultures [4, 5, 6], and genome- scale metabolic models have been used to model their metabolism [7, 8]. In contrast, the interactions between the two taxa is still largely unknown from a mechanistic perspective despite indications of strong inter- dependencies [9]. Further, the current knowledge on cheese microbial interactions is limited to the scope of simplified artificial milk media, concerns pairwise associations, or focus only on the short time interval during the milk acidification. Less is known about the interactions involving more than two SLAB strains and how interaction networks evolve during the cheese ripening steps. Metabar coding and metagenomics approaches have also been used to study the cheese microbiome [10, 11, 12]. While these studies provide a good overview on the community dynamics during cheese- making using higher- order taxonomy, they are limited in uncovering inter- species interactions and strain- level diversity. + +<|ref|>text<|/ref|><|det|>[[78, 784, 911, 894]]<|/det|> +To uncover the role of microbial interactions and interacting agents, we used a strain dropout strategy whereby one single strain or a strain group was left out from the starter culture, and the cheese- making process was characterized in its entirety. We used an industrially relevant SLAB culture, containing: one Streptococcus thermophilus (ST), two major L. lactis (LLm1 & LLm2), one major L. cremoris (LC), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 78, 911, 220]]<|/det|> +44 and a mixture of 21 L. cremoris and L. lactis strains in smaller fractions (hereafter, Lactococcus blend (LB)). We investigated the effect of leaving out S. thermophilus and Lactococcus blend in a year- long cheddar- making experiment (Fig. 1a); followed by controlled milk experiments investigating hypotheses generated from the year- long experiments. For the latter, we used an integrative systems biology approach that combined different layers of biological information (Fig. 1b). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 90, 177, 111]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[87, 139, 907, 201]]<|/det|> +## 1 S. thermophilus supports lactococci growth and shapes the metabolic profile during cheese ripening + +<|ref|>text<|/ref|><|det|>[[78, 220, 912, 700]]<|/det|> +To study the interactions between the members of the SLAB culture, we started by quantifying the population dynamics during a year- long ripening cheese making experiment (see Methods). We used for variations of starter cultures: i) containing all member species of an industrial starter culture, viz., S. thermophilus (ST), L. cremoris (LC), L. lactis (LLm1 and LLm2), and Lactococcus blend (LB); ii) the same as (i) but prepared independently (HP); iii) excluding LB; and iv) excluding ST (Fig. 1a). In the three conditions that included S. thermophilus, we observed a trend of slow decline in the population of lactococci and S. thermophilus, starting with 9.25 and ending at 8 log10 CFU/g (Fig. 2a- c). In the single condition that excluded S. thermophilus, we observed a much steeper decline in lactococci population, ending at 6.5 log10 CFU/g (Fig. 2a- b). Also, the declining lactococci population exhibited different trajectories between the two batches, albeit both converging at 9 and 12 months (Fig. 2a- b). The non- SLAB count increased from 0 to 7.5 log10 CFU/g, in the highest case (Fig. 2d). Based on 95% confidence interval estimates, we observed high variability among the non- SLAB populations, but with no appreciable overall difference between the conditions (Fig. 2d). Leaving out Lactococcus blend or changing the way of packing the different strains in the culture did not result in any visible response to the community dynamics. Thus, inter- species interactions, especially involving S. thermophilus, appear to be driving the overall population dynamics. + +<|ref|>text<|/ref|><|det|>[[77, 714, 912, 915]]<|/det|> +To measure the effect of the microbial interactions in the resulting phenotype of the cheese, we used targeted metabolomics analyses. Across conditions, the biggest change in the metabolic profiles of cheese was observed between the two- week mark and three months (Fig. 2e). This can be attributed to changes in peptide composition followed by the general increase in amino acids concentrations (Fig. 2f). Furthermore, cheese metabolomes continue to change at a slower pace between the three- to twelve- month marks, resulting in different end profiles. A notable time- depended accumulation of compounds after the third month was observed on acetic acid, tyramine, g- amino butyric acid, putrescine and cadaverine + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 77, 912, 373]]<|/det|> +(Fig. 2f & Supplementary Fig. 1). The cheeses produced without S. thermophilus formed a separate cluster relative to the rest (Fig. 2e). This pattern can be ascribed to compounds with significantly different concentrations across strain removal conditions, especially when S. thermophilus was absent. The majority of these compounds are peptides that were either not accumulated, accumulated in lower amounts or accumulated in higher amounts without the addition of S. thermophilus. Lactose, galactose and lactic acid concentrations were also found to be significantly different when S. thermophilus was absent (Fig. 2g, ANOVA \(F(3,36) = 208 \pm 334, p < 0.001, \eta^2 = 0.86 \pm 0.08\) & Supplementary Fig. 2). This difference was already notable in the samples from the second week of cheese ripening (Fig. 2h & Supplementary Fig. 3a,c&e). These results points towards the significant influence of S. thermophilus growth during milk fermentation and its non- growing yet active cells during the long term cheese ripening. + +<|ref|>text<|/ref|><|det|>[[60, 388, 912, 680]]<|/det|> +The presence of S. thermophilus was found to benefit both the growth of the Lactococcus community and the final metabolic profile of the cheddar cheese. Notably, the growth benefit is not clearly visible before two weeks to three months, a time- frame that may be considered prohibitively long for experiments in the laboratory. Instead, these effects become more evident as the experimental time- frame expands to one year. The cheese metabolic profiles indicated differences in peptide composition between the conditions, a result likely stemming from the non- growing metabolic activity of S. thermophilus. In accordance with the literature, the presence of S. thermophilus reveals a few key metabolic changes from the milk fermentation; the complete consumption of lactose and the production of galactose [13, 14]. Together, the compositional and metabolic profiles led us to hypothesize that both early and late S. thermophilus activity is critical for the long term effect on Lactococcus population. + +<|ref|>sub_title<|/ref|><|det|>[[60, 727, 909, 788]]<|/det|> +## 2 S. thermophilus has strain-specific interactions with the members of Lactococcus lactis community + +<|ref|>text<|/ref|><|det|>[[60, 811, 911, 923]]<|/det|> +To assess the role of S. thermophilus, we performed controlled experiments wherein additional strains were removed. Samples were taken when milk fermentation reached the transition from exponential to stationary phase (at pH 5) and gene expression was monitored using metatranscriptomics. To further investigate the microbial interactions within the SLAB culture, we also analyzed the genomes of the individual strains + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 75, 912, 740]]<|/det|> +in the culture and their phylogenetic relationship (Fig. 3a). The phylogenomic analysis separated the species L. lactis and L. cremoris into two distinct groups. Two of the main Lactococcus strains, LC and LLm2, have no close relatives within the culture community. For the third main Lactococcus strain, LLm1, we identified two close relatives (LL- LB01 & LL- LB02), belonging to the Lactococcus blend (Fig. 3 & Supplementary Table 1). The rest of the strains in the Lactococcus blend formed distinct sub- clusters within the cremoris clade (Fig. 3). By employing pan- genome analysis on Lactococcus strains, we identified singletons (i.e. a gene found only in one genome) for each member of the culture. The highest number of singletons was harbored by the main LLm2 followed by LC, and LLm1 where its number of singletons was in the same range as the Lactococcus blend strains, with numbers 288, 134, 79 and average of \(53.6 \pm 18.6\) , respectively. In parallel, we evaluated both the presence and activity level (using metatranscriptomics) of the Lactococcus strains by calculating percentile of the transcribed singletons per genome. We found a high number of transcribed singletons on all the main Lactococcus strains, while this number varied considerably among the strains of the Lactococcus blend, raising questions regarding whether or not these strains were active (Fig. 3a & Supplementary Table 4). We hypothesized that the temperature dynamics during the experiment could influence the strain activity. To test this, by carrying out milk fermentation using each of the individual Lactococcus strains at different temperatures, we calculated a proxy for their temperature stress tolerance (Pearson correlation between curves of 30 against 37, 40 and 43 degrees). The results showed that the L. lactis strains have higher temperature stress tolerance in the tested ranges, including Lactococcus blend strains LL- LB01 and LL- LB02 (PCC \(0.96 \pm 0.05\) ). The L. cremoris strains exhibited low temperature stress tolerance (PCC \(0.7 \pm 0.24\) ), with the main LC being one of the most stress tolerant strains within this group (PCC range 0.84- 0.99) (Supplementary Fig. 5a&b, Fig. 3a). + +<|ref|>text<|/ref|><|det|>[[88, 750, 911, 923]]<|/det|> +We next explored the extent to which we could discriminate between the different strains. The number of unique k- mers in each genome ranged from 368 to 622455 and they were found to be dependent on the presence of close phylogenetic relative strains (Fig. 3a). The k- mer transcription abundance indicated the feasibility of discriminating between the three main Lactococcus strains (LC, LLm1 and LLm2) and the one S. thermophilus strain, but not between the strains in the Lactococcus blend (Supplementary Fig. 4). Using differential expression analysis, we investigated the interaction effect on S. thermophilus + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 77, 912, 344]]<|/det|> +transcriptome by leaving out each Lactococcus strain/blend from the culture. We found the strongest response when LLm1 was left out followed by Lactococcus blend, LC and LLm2 with 291, 182, 21 and 1 gene(s) significantly differential expressed, respectively (Fig. 3b- e, Supplementary Table 6 & sections below). Although both LLm1 and LLm2 belong to L. lactis clade, the latter strain stands out by having the highest number of singletons as well as having the second highest number of unique k- mers (Fig. 3). By removing strains with a different degree of relatedness we identified that different Lactococcus strains have distinct effects on gene expression of S. thermophilus. This finding can be attributed to the different genetic background and the diverse phylogenetic relationships within the Lactococcus community, highlighting the diversity of potential microbial interactions between S. thermophilus and the Lactococcus community. + +<|ref|>sub_title<|/ref|><|det|>[[57, 388, 911, 450]]<|/det|> +## 3 S. thermophilus proteolytic activity may benefit Lactococcus community by providing nitrogen sources + +<|ref|>text<|/ref|><|det|>[[55, 472, 914, 856]]<|/det|> +To investigate mechanisms of inter- species interactions, we generated genome- scale metabolic models for the individual strains in the culture and performed flux balance analysis - based simulations. To explore the effect of changes in growth medium composition, we generated models gap- filled on relevant variations of milk- mimicking media. Supplementary Fig. 6a shows overall model statistics for the different species, while Supplementary Fig. 6b shows the specific reactions added through gap- filling on different media across the models (see Supplementary Section 0.1). Flux balance simulations with individual models were carried out using the different model sets in aerobic and anaerobic variants of fermented and non- fermented milk (Fig. 3a,b, Supplementary Fig. 6c,d). All species consistently consumed lactose in simulations where growth was feasible. Several differences in amino acid uptake/secretion were observed among the strains. The branched chain amino acid valine was exported by S. thermophilus in all but one simulation condition, whereas LC, LLm1, and LLm2 consumed this amino acid in four, six, and three simulations conditions, respectively. Additional amino acid uptake/secretion predictions are reported at Supplementary Section 0.1 + +<|ref|>text<|/ref|><|det|>[[57, 872, 911, 922]]<|/det|> +To assess potential cross- feeding, we performed community simulations using the SMETANA framework (see Supplementary Section 0.1). We used the calculated SMETANA scores, ranging from 0 to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[79, 78, 912, 494]]<|/det|> +1, as a measure of predicted interaction confidence (0 being lowest confidence and 1 being the highest) [15]. These simulation indicate that S. thermophilus provides valine to Lactococcus strains in all three community simulation conditions (gap- filled on rich aerobic milk & simulated on minimal aerobic milk, gap- filled on rich anaerobic milk & simulated on minimal aerobic milk, and gap- filled on rich anaerobic milk & simulated on minimal anaerobic milk) where exchanges were predicted between community members (Figure 3c). Not only was this interaction consistent, it also occurred with a SMETANA score of 1 in every case, suggesting that this is a key ecological interaction for maintaining the composition and function of the community. Serine exchanges from S. thermophilus to LLm1 were predicted in 2 out of 3 community simulation conditions, with an average SMETANA score of \(0.38 \pm 0.01\) . Exchanges from S. thermophilus to LC involving glycine and ammonium were predicted in 1 out of the 3 simulation conditions, with SMETANA scores of 0.3 and 0.29, respectively (Figure 3c). Finally, an exchange from S. thermophilus to LLm2 involving alanine was predicted under only 1 simulation condition with a SMETANA score of 0.42. Overall, the metabolic simulations strongly indicate cross- feeding between the community members, with S. thermophilus standing out as a key donor (Figure 3c). + +<|ref|>text<|/ref|><|det|>[[78, 510, 912, 923]]<|/det|> +To complement the genomic and metabolic modeling analysis, we investigated the changes in S. thermophilus and Lactococcus community transcriptome. The principal component analysis showed pattern consistency with the differential transcriptome analysis (Fig. 3c- f, Supplementary Section 0.2). Regarding changes in S. thermophilus's transcriptome, we found up- regulated genes annotated as oligopeptide- binding protein (AmiA) and oligopeptide transport system permease protein (OppB, OppC, OppD & OppF) (see Supplementary Section 0.2). Seven down- regulated genes are annotated as transporters, such as glutamine ABC transporter permease protein (GlnP), cadmium, cobalt and zinc/H(+)- K(+) antiporter (Supplementary Section 0.2). Further, transcriptional repressors such as the catabolite control protein A (CcpA) and nitrogen- metabolism- regulating- proteins such as GTP- sensing transcriptional pleiotropic repressor (CodY) were found to be transcriptionally active. In line with the putative interactions predicted using metabolic modelling, we found a gene annotated as branched- chain amino acids (BCAA) transaminase expressed in S. thermophilus, suggesting that valine biosynthesis is active. Moreover, we found a gene annotated as BCAA transport system 2 carrier protein (BrnQ) as well expressed, suggesting that S. thermophilus is secreting valine. Complementary differential pathway enrichment analysis supported + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 77, 912, 250]]<|/det|> +further that valine biosynthesis was enriched in S. thermophilus metatranscriptome (Fig. 4d- e). Overall, the metabolic modeling and metatranscriptome analysis together suggests that S. thermophilus can act as a branched chain amino acid donor to Lactococcus. It is well known that Lactococcus harbor a number of amino acid auxotrophies including valine [16]. In addition, the SMETANA simulations provide insights into the metabolic strategies that may be employed by community members under challenging media conditions. + +<|ref|>text<|/ref|><|det|>[[55, 268, 912, 835]]<|/det|> +We next investigated transcriptional change in the Lactococcus community in the presence and absence of S. thermophilus. As it is not possible to distinguish between the transcriptome profiles of all the different Lactococcus strains, we merged these into one group for pan- genome analysis and pan- metabolic modelling. Strain- to- strain transcriptional changes were investigated when possible (see Supplementary Section 0.2). The most up- regulated OGs in Lactococcus when S. thermophilus is left out are involved in nitrogen assimilation, namely nitrogen regulatory protein P- II, related transcriptional regulator (GlnR), ammonium transporter (amtB), glutamine synthetase (glnA) and glutamine transport ATP- binding protein (glnQ). Additionally, we confirmed that GlnR in Lactococcus regulates the three operons, namely amtB, glnA, and glnQ (Fig. 4f- i, Supplementary Table 2 & Supplementary Section 0.2). The purine and pyrimidine biosynthesis pathways are deferentially expressed in a coordinated manner. Several genes in the pathways were up- regulated, in particular the genes associated with the reactions leading to uridine and guanine synthesis (Fig. 4j & supplementary). Both pathways are connected via L- glutamine synthetase. Expression patterns in amino acid metabolism show a mixed pattern of both up- and down- regulation (see Supplementary Section 0.2). Overall, the metatranscriptomics analysis further supports that S. thermophilus cross- feed Lactococcus community. One hypothesis that could explain this is that S. thermophilus has a higher proteolytic activity, thereby providing more peptides and amino acids (available nitrogen source) in the culture which Lactococcus takes advantage of. When S. thermophilus is absent, there is a lower degree of nitrogen availability causing upregulation of the nitrogen assimilation and nucleotide syntheses pathway in Lactococcus. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[55, 87, 908, 150]]<|/det|> +## 4 Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community + +<|ref|>text<|/ref|><|det|>[[55, 170, 912, 586]]<|/det|> +To elucidate the role of interactions between the SLAB strains in the development of cheese flavor, we performed a targeted metabolomics analysis. Leaving out \(L\) . cremoris (LC) led to the strongest metabolic response followed by a weaker and different response when leaving out LLm1, S. thermophilus and the Lactococcus blend (see Supplementary Section 0.3). Surprisingly, leaving out LLm2 did not lead to a marked change in the metabolic profile of the acidified milk (Fig. 5a). Six flavor compounds, namely heptanal, hexanal, 2- ethyl- furan, 2,3- pentanedione, diacetyl and acetoin, were either detected at significantly higher concentrations or only produced when this main \(L\) . cremoris strain was removed (Fig. 5b- c & Supplementary Fig. 10). A graphical summary of the metabolome differences between the whole culture and the culture lacking LC is shown in Fig. 5d. Many of the flavor compounds found were significantly altered by the strain removal strategy, with the majority corresponding to byproducts of secondary metabolism with unknown microbial biosynthesis pathways or gene clusters. Yet, investigation of their patterns highlights the numerous ways these compounds may cross- feed within the SLAB culture. For example, 2,3- Pentanedione could be donated by S. thermophilus to LC while ethyl hexanoate, ethyl acetate and 2- Methyl- 3- thiolanone likely produced by LC. + +<|ref|>text<|/ref|><|det|>[[55, 601, 912, 895]]<|/det|> +C4 aroma compounds such as diacetyl and acetoin are known derivatives of citrate metabolism with characteristic contribution to buttery- like aroma. Although desirable in small quantities, e.g less than 0.05 rag. per 100 g of diacetyl content, higher amount could lead to off- flavors [17, 18]. Therefore, we investigated whether the increase in diacetyl and acetoin concentrations triggered by the removal of LC could be explained by the metatranscriptome profiles. Indeed, we found differential expression patterns in genes related to diacetyl and acetoin metabolism. Therefore, we examined the gene expression levels in the metabolic route from citrate towards diacetyl and acetoin through pyruvate as well as an alternative route towards \(\alpha\) - ketoglutarate. The citrate- sodium symporter that is present only on Lactococcus strains was highly expressed implying active uptake of the citrate available in milk by \(L\) . lactis. From citrate, Lactococcus strains showed expression of the genes associated with the pyruvate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 77, 912, 555]]<|/det|> +carboxylase and acetolactate synthase reactions, whereas only LC and LLm1 had the genes related to acetolactate decarboxylase expressed at high levels. On the other route towards \(\alpha\) - ketoglutarate and further, genes associated with aconitate hydratase and isocitrate dehydrogenase are only expressed in LC and S. thermophilus (Fig. 5e & Supplementary Fig. 11). Metabolic modeling simulations across the different medium conditions are largely corroborated by the metatranscriptomics results with the exception of three reactions involving acetoin. These are acetolactate decarboxylase (ACLDC), which does not carry a flux in the LC model, and 2,3- butanediol dehydrogenase (BTDD_RR) and diacetyl reductase (ACTD), which carry fluxes in LLm1 model but not in LC model (Fig. 5e & f). Taken together, the presence of the main L. cremoris strain in the cultures results in the prevention of an undesirable increase in the levels of some key flavor compounds, including diacetyl and acetoin. This can be attributed to the degradation of diacetyl, e.g., through the action of some alcohol dehydrogenases from LC. Alternatively, LC might be competing with the other strains in the culture on citrate and subsequently converting it to other products than diacetyl and acetoin, such as ethyl- hexanoate, ethyl- acetate and 2- methyl- 3- thiolanone, which normally happens in mixed culture fermentation following the rapid drop of redox potential in the beginning of fermentation. This would result in the reduction of the overall amount of citrate available for diacetyl and acetoin formation. + +<|ref|>sub_title<|/ref|><|det|>[[90, 614, 217, 636]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[50, 660, 912, 923]]<|/det|> +Interactions in microbial communities are reported in a wide range of ecosystems [19, 20, 21, 22, 23]. While such interactions are widely reported also in food microbial communities [24, 3, 25, 26, 27, 28], very few studies have provided insights into the molecular agents that mediate the interactions. Uncovering interactions in situ is particularly challenging due to complexity of the communities involved as well as that of the medium such as milk. Here, we combined genomics, metatranscriptomics, metabolomics, and metabolic modelling to uncover key microbial interactions in Cheddar cheese- making. Notably, we used industrial strains and a full cycle of year- long cheese ripening. We report competitive and cross- feeding metabolic interactions between the lactic acid bacteria used in Cheddar cheese production. Firstly, L. cremoris strain competes with L. lactis for the milk's available citrate, resulting in accumulation of key + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 78, 912, 464]]<|/det|> +metabolites such as diacetyl and acetoin in the final product. Such interactions take place within the first 5 hours of milk fermentation and strongly influences the final cheese flavor. Secondly, S. thermophilus provides the necessary nitrogen source to the Lactococcus community, explaining the significant one- year long growth benefit of the Lactococcus strain population and the different final cheese metabolome profile. In addition, Lactococcus strains affect the activity of S. thermophilus differently. While the presence of one L. lactis strain hardly influences the activity of S. thermophilus and the development of the final cheese flavor profile, the presence of a different L. lactis strain influences both significantly. Often in literature the attempts to study microbial interactions disregard the strain diversity. Yet, recent studies highlight the importance of strain interactions, claiming the major role of those in predicting eco- evolutionary dynamics [29]. Our results show how strain- specific metabolic interactions between microbes shape the biochemical profile of cheese, and provide targets towards the rational design and assembly of microbial communities with the aim of fine- tuning cheese flavor. More broadly, the study provides a blue- print to uncovering in situ interactions in complex food microbial ecosystems. + +<|ref|>sub_title<|/ref|><|det|>[[90, 507, 190, 528]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[90, 556, 408, 576]]<|/det|> +## Starter lactic acid bacteria culture + +<|ref|>text<|/ref|><|det|>[[80, 593, 912, 888]]<|/det|> +A cheese making experiment was designed to investigate the effect of removing selected strains from a defined- strain SLAB culture. Wild- type strains were obtained from the Chr. Hansen Culture Collection and were originally isolated from dairy products or cultures. By applying this strain removal strategy, it becomes possible to gain insights into the individual role of the strain but also its interactions within the microbial community. The original defined- strain SLAB culture consisted of 5 different bulks: 1 bulk with single strain S. thermophils, 1 bulk with a multi strain mixture of 21 Lactococcus strains belonging to both species of L. lactic and L. cremoris [30], which were inoculated as a single component (hereafter, Lactococcus blend), 2 bulks with single strain L. lactis, and 1 bulk with single strain L. cremoris. The original defined- strain SLAB culture was composed from of 25 strains and all bacterial and all strains applied are listed in Supplementary Table 1. + +<|ref|>text<|/ref|><|det|>[[125, 903, 909, 924]]<|/det|> +To apply the strain removal strategy 4 defined- strain SLAB cultures were designed. Condition ALL, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 78, 912, 372]]<|/det|> +corresponded to the original defined- strain SLAB composed from of 25 strains packed under industrial conditions. Condition - HP, corresponded to the original defined- strain SLAB composed from of 25 strains packed under non- industrial condition by hand. Condition - LB, corresponded to the removal of the Lactococcus blend. Condition - ST, corresponded to the removal of the S. thermophils strain. The amount added varied between the conditions to target the same acidification profile in the cheese vat while the cheeses making process was kept constant. The target was to obtain a pH of 5.35 at milling and a final moisture in non- fat solids (MNFS) of the cheeses at \(54\% \pm 0.5\%\) measured after 2 weeks of ripening. The designed defined- strain SLAB for conditions - ALL, - HP, and - LB were all dosed using \(8.1\mathrm{g}\) of culture/100L, whereas the designed defined- strain SLAB for condition - ST was dosed at a concentration of \(17.3\mathrm{g / 100L}\) to compensated for missing contribution to acidification from S. thermophils. + +<|ref|>sub_title<|/ref|><|det|>[[90, 412, 236, 432]]<|/det|> +## Cheese making + +<|ref|>text<|/ref|><|det|>[[78, 448, 912, 925]]<|/det|> +Cheeses were produced at the Application and Technology Center of Chr. Hansen A/S (Hørsholm, Denmark). The composition of the pasteurized cheese milk (organic milk, Naturmælk, Denmark) was measured using a Milkscan™ (FOSS, Hillerød, Denmark), and the fat level was adjusted with pasteurized cream (organic cream \(38\%\) fat, Naturmælk, Denmark) to obtain a protein to fat ratio of 0.90. The cheese milk was heated to \(32^{\circ}\mathrm{C}\) and each cheese vat was filled with \(150\mathrm{kg}\) . The milk was ripened for 40 min with the starter culture before the coagulant (CHY- MAX® Plus, Chr. Hansen A/S) was added. After 30 min of coagulation, the gel was cut into cubes \(10\mathrm{x}10\mathrm{mm}\) . The whey and the cheese grains were stirred for 10 min before heating to \(38^{\circ}\mathrm{C}\) over 40 min and the final scalding and stirring period was 45 min before whey drainage at pH 6.4- 6.5. After 20 min, the cheese curd was cut into 12 blocks and rearranged in blocks 3 times during the next 75 min until pH 5.35 was obtained. The blocks where then milled, and the chips were dry salted (1.7- 2.0% salt in dry matter) by manually adding and mixing the salt 2 times during 15 min. The salted chips were mixed every 5 min over a period of 30 min. Following molding, the chips were pressed at 2 bars for 15 min followed by 5 bars for 17 hours. The cheeses were vacuum- packed and stored at \(9^{\circ}\mathrm{C}\) until sampling after 0.5, 3, 6, 9 and 12 months. The cheese gross composition (moisture, fat, protein and total solids) was estimated after 2 weeks of ripening by FoodScan™ (FOSS, Hillerød, Denmark). The NaCl content was estimated by analyzing the chloride concentration by automated potentiometric + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 78, 910, 157]]<|/det|> +endpoint titration (DL50, Mettler- Toledo A/S, Glostrup, Denmark). pH was measured potentiometrically (PHC2002- 8, Radiometer Analytical SAS, Lyon, France) in a paste prepared by mixing \(10\mathrm{g}\) of grated cheese with \(10\mathrm{ml}\) of deionized water with a wooden spatula. All analysis were carried out in duplicate. + +<|ref|>text<|/ref|><|det|>[[90, 177, 912, 380]]<|/det|> +In summary, for the one year- long ripening cheese making experiment we inoculated three variations of the SLAB cultures into four milk tanks. Two milk tanks were inoculated with the complete set of strains, one used as the control and the other to test a different approach of culture inoculations (All and All_HP, respectively). The two other variations of the culture were prepared using the 'strain removal strategy' with the aim of testing the effect of \(L\) lactis blend (All- LB) and of the single \(S\) thermophilus (All- ST) during cheese- making. Following the acidification phase, we sampled the resulting cheddar cheeses in 5 time points over the one year period. We followed two batch experiments started on two consecutive days. + +<|ref|>sub_title<|/ref|><|det|>[[90, 411, 383, 431]]<|/det|> +## Microbial community dynamics + +<|ref|>text<|/ref|><|det|>[[90, 448, 912, 712]]<|/det|> +Grated cheese (5.0 g) was mixed with \(45.0\mathrm{g}\) of autoclaved \(2\%\) sodium citrate buffer, \(\mathrm{pH}7.5,46^{\circ}\mathrm{C}\) . The mixture was homogenized in a Stomacher® blender for 4 minutes at medium setting to dissolve the cheese and suspend the bacteria present. Sequential of 10- fold dilutions were prepared as required in \(0.1\%\) peptone/0.15M NaCl in water, \(\mathrm{pH}7.0\) . The total population of Lactococcus spp. was determined on pour plated M17 Agar (Difco, USA) after aerobic incubation for 5 days at \(30^{\circ}\mathrm{C}\) . The total population of thermophilic cocci (ST) was determined on pour plated M17 Agar after aerobic incubation for 3 days at \(37^{\circ}\mathrm{C}\) . The total population of non- starter lactic acid bacteria (NS- LAB) was determined on overlaid Rogosa Agar (Sigma Aldrich, USA) after incubation for 7 days at \(37^{\circ}\mathrm{C}\) . The microbial analysis was carried out in duplicate at each sampling point. + +<|ref|>sub_title<|/ref|><|det|>[[90, 744, 496, 764]]<|/det|> +## Carbohydrates and organic acids in cheese + +<|ref|>text<|/ref|><|det|>[[90, 782, 912, 923]]<|/det|> +The content of lactose, glucose, galactose, lactate, citrate and acetate were quantified using a Dionex ICS- 3000 RFIC- EGTM dual system equipped with an amperometric detector (Dionex, Sunnyvale, CA, USA). The separation was performed using an anion- exchange column (CarboPac® PA20, \(3\mathrm{x}150\mathrm{mm}\) , \(6.5\mu \mathrm{m}\) ) and an ion- exclusion column (IonPac® ICE- AS6, \(9\mathrm{x}250\mathrm{mm}\) , \(8\mu \mathrm{m}\) ). Grated cheese (3.0 g) was mixed with \(15\mathrm{ml}83\mathrm{mM}\) PCA containing \(2\mathrm{mM}\) Na- EDTA, \(33\mathrm{mM}\) arabinose and \(48\mathrm{mM}\) 2- hydroxyisobutyric acid + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 912, 312]]<|/det|> +(internal standards) and rotated for \(30\mathrm{min}\) at room temperature. This extract was subsequent centrifuged \((5000\mathrm{ng},30\mathrm{min},4^{\circ}\mathrm{C})\) and the supernatant was filtered \((0.45\mu \mathrm{m})\) . The supernatant was diluted (600- fold) before analysis. For analysis of carbohydrates \(25\mu \mathrm{L}\) was injected and the separation was performed at \(26^{\circ}\mathrm{C}\) with a flow rate of \(1.3\mathrm{mL / min}\) increasing the KOH concentration as follow: \(1\mathrm{mM}\) KOH for \(5\mathrm{min}\) from \(1\mathrm{mM}\) to \(20\mathrm{mM}\) KOH over \(0.5\mathrm{min}\) , then kept at \(20\mathrm{mM}\) KOH for \(6.6\mathrm{min}\) min before re- equilibration to \(1\mathrm{mM}\) KOH over \(5\mathrm{min}\) . For analysis of organic acids \(50\mu \mathrm{L}\) was injected and the separation was performed at \(30^{\circ}\mathrm{C}\) with a flow rate of \(1\mathrm{mL / min}\) using \(0.4\mathrm{mM}\) heptfluorobutyric acid in \(5\%\) acetonitrile as eluent. The analysis was performed in duplicate at each sampling point. + +<|ref|>sub_title<|/ref|><|det|>[[90, 348, 291, 368]]<|/det|> +## Proteolysis in cheese + +<|ref|>text<|/ref|><|det|>[[55, 384, 912, 648]]<|/det|> +Grated cheese \((2.0\mathrm{g})\) was mixed with \(18\mathrm{ml}\) distilled water and blended for \(2\mathrm{min}\) at \(25,000\mathrm{rpm}\) (Ultra- Turrax model T25). This suspension was used for determination of total nitrogen (TN), Non- Casein nitrogen (NCN), and non- protein nitrogen (NPN) using the fractionation procedure described in standard NF ISO 27871 (ISO, 2011). The nitrogen content for each fraction was determined using the Kjeldahl method as described in standard NF EN ISO 8968- 1 (ISO, 2014). The level of primary proteolysis (NCN) and secondary proteolysis (NPN) were expressed as percentage of TN found in the cheese. The analysis was performed in duplicate at each sampling point. Finally, the metabolic profile of the cheeses was quantified by measuring 30, 3, 28 and 248 features belonging to the classes of acids, sugars, flavor- related organic compounds and peptides, respectively. + +<|ref|>sub_title<|/ref|><|det|>[[89, 684, 420, 705]]<|/det|> +## Genome sequencing and assembly + +<|ref|>text<|/ref|><|det|>[[55, 720, 912, 923]]<|/det|> +Genomic DNA for de novo short read whole- genome shotgun sequencing (WGS) was extracted from \(1\mathrm{mL}\) of overnight culture (M17 Broth (Difco) supplemented with glucose) of each of the strains (at \(OD600 = 1\) ) with DNeasy Blood and Tissue kit on QiaCube system (Qiagen, Germany) following the manufacturer's protocol. Prior to extraction, cell pellets were washed twice in TES buffer ( \(50\mathrm{mM}\) TRIS pH 8.0, \(1\mathrm{mM}\) EDTA pH 8.5, \(20\%\) sucrose) and then resuspended in \(180\mathrm{uL}\) of pre- lysis TET buffer (20 mM TRIS- Cl pH 8.0, \(2\mathrm{mM}\) EDTA pH 8.5, \(1.2\%\) Triton X- 100, \(20\mathrm{mg / mL}\) lysozyme, \(2\mu \mathrm{L}\) of \(25\mathrm{U / \mu L}\) mutantsolysin, \(4\mu \mathrm{l}\) of \(100\mathrm{mg / mL}\) RNase A). Genomic libraries were prepared using KAPA LTP Library + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[79, 77, 912, 526]]<|/det|> +Preparation Kit KK8230 (Roche, Switzerland) on a Biomek 4000 Liquid Handler (Beckman Coulter, USA). A portion of \(500\mathrm{ng}\) of genomic DNA diluted in EB buffer (Tris- Cl, pH 8.0) was mechanically fragmented on Bioruptor® Standard (Diagenode, USA) with 12 sonication cycles (30 sec ON/OFF) to obtain an average fragment size of \(300\mathrm{bp}\) (CV%: 1.5). Fragmented DNA was processed following the KK8230 kit manufacture's protocol. Following the adapter ligation step, adapter- modified DNA fragments were enriched by 8- cycle PCR. AMPure XP (Beckman Coulter) paramagnetic beads were used for clean- ups to purify fragments at average size between 450 to 550 bp. Concentration of gDNA and double stranded DNA libraries were measured by Qubit® Fluorimeter using Qubit dsDNA Broad range and Qubit 1x dsDNA HS assays (Thermo Fisher Scientific, USA), respectively. Average dsDNA library size distribution was determined using an Agilent HS NGS Fragment (1- 6000 bp) kit on an Agilent Fragment Analyzer (Agilent Technologies, USA). Libraries were normalized and pooled in NPB solution (10 mM Tris- Cl, pH 8.0, \(0.05\%\) Tween 20) to the final concentration of \(10\mathrm{nM}\) . Following denaturation in \(0.2\mathrm{NNaOH}\) , \(10\mathrm{pM}\) of pooled libraries in \(600\mu \mathrm{L}\) ice- cold HT1 buffer were loaded onto the flow cell provided in the MiSeq Reagent kit v3 (600 cycles) and sequenced on a MiSeq platform (Illumina Inc., San Diego, USA) with a paired- end protocol and read lengths of 301 nucleotides. + +<|ref|>text<|/ref|><|det|>[[55, 540, 912, 923]]<|/det|> +All processing of the short reads was done in either CLC Genomics Workbench versions 9.5.3, 9.5.4 or 10.1.1. The short reads were mapped with default parameters to the reference sequence of the phage Phi X 174 using the tool "Map reads to reference". Unmapped reads from the mapping were trimmed for quality using the PHRED score 23 as the threshold and with the non- default parameter of discarding reads that were less than 50 base pairs long using the tool "Trim Sequences". The trimmed reads were de novo assembled with default parameters except for the minimum contig length was set to 600 base pairs using the tool "De Novo Assembly". Afterwards, a decontamination step was performed where contigs with low depth of coverage were removed using a custom plugin written by Qiagen. The decontamination step first removes all contigs where the depth of coverage is below 15X and afterwards removes all contigs where the depth of coverage is below \(25\%\) of the median depth of coverage for the entire genome assembly. Gene calling of the filtered contigs was done with Prodigal version 2.6.2 using the default parameters. Finally, the genome assemblies with annotated genes were functionally annotated with BLAST against a local annotation database using a custom plugin written by Qiagen. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 85, 498, 105]]<|/det|> +## Comparative genomics and Phylogenomics + +<|ref|>text<|/ref|><|det|>[[78, 120, 912, 722]]<|/det|> +A L. lactis and L. cremoris pangenome was created without strain LC- LB16 which found to have 6 times more singletons than all the other strains. The number of core, accessory, and unique (singletons) genes in the pangenome is 1247, 3308, and 2323, respectively. The OGs from this pangenome were used in the comparative genomics and differential metatranscriptome analyses as well as in the creation of a pan- metabolic network of L. lactis and L. cremoris for microbial community modeling. We identified all k- mers of length up to 31 bps (k=31) in the genomes of all the strains and unique k- mers in each strain were filtered. For the set of unique k- mers we mapped the metatrascriptomic reads to get their abundance and kept only those above 7 counts. Normalization of the number of counts per strain performed based on the number of unique k- mers identified in each of the strains. From the total pool of 54667 genes \(15.8\%\) was removed as the cluster above 96 cd- hit identity was mixed among the three taxa. still \(84.2\%\) consider significant number to look on L. cremoris and L. lactis differences. For the phylogenetic analysis 22 genomes from the SLAB culture were used alongside with 107 complete genomes of L. lactis and L. cremoris, which were retrieved from NCBI results in total to 129 genomes. CD- HIT on \(80\%\) amino acid identity thresholds along with a custom- made R script were used to identify a set of 464 monocore marker gene sets [31, 32]. The trees were constructed using PhyloPhlAn v3.0 [33] using the PhyloPhlAn parameters "- accurate" and "- diversity low", which translate to the usage of a pfasum60 substitution matrix. In addition, multiple sequencing alignments (msa) were performed with muscle and trimming was performed by removing columns with at least one nucleotide appearing above a threshold of 0.99. The final maximum likelihood tree was constructed on the concatenated DNA msa using raxmlHPC, 100 bootstrap and GTRGAMMAI model [34, 35]. + +<|ref|>sub_title<|/ref|><|det|>[[90, 774, 669, 794]]<|/det|> +## Generation and simulation of genome-scale metabolic models + +<|ref|>text<|/ref|><|det|>[[90, 812, 911, 923]]<|/det|> +Genome- scale metabolic models were generated for S. thermophilus, Lactococcus LLm1, LLm2, and LC using CarveMe v1.4.1 [36] based on the protein sequences of assembled genomes. First, Prodigal v2.6.3 [37] was used to generate open- reading- frame (ORF) annotated protein sequence files from the corresponding DNA fasta files. Additionally, milk media from a recent kefir publication was used for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 77, 912, 463]]<|/det|> +gap- filling during model generation [28]. More specifically, we formulated four biologically relevant variations of the milk composition, including aerobic rich milk, anaerobic rich milk, aerobic depleted milk, and anaerobic depleted milk. The former two media represent the initial composition of milk, while the latter two media represent milk after it has been depleted by fermentation. A version of each of the four species was generated by gap- filling on each of the four milk media variations, as well as without gap- filling, resulting in a total of 20 models. All models were generated using the default CarveMe universal bacterial model template. Note that although gram positive bacteria are generally not modeled with a periplasmic compartment due to its smaller size relative to gram positive bacteria, all models generated with the automated CarveMe tool contain such a compartment. Individual model simulations were carried out for all models on each of the four media variations using the reframed v1.2.1 and cobrapy v.0.20.0 metabolic modeling packages [38]. Community simulation of genome- scale metabolic models was also carried out in the different media variations, using SMETANA v1.2.0 [15]. All associated data, including code used for generating and plotting results, is available on GitHub. + +<|ref|>sub_title<|/ref|><|det|>[[72, 499, 275, 518]]<|/det|> +## Inoculations in milk + +<|ref|>text<|/ref|><|det|>[[70, 536, 912, 767]]<|/det|> +All milk fermentations were carried out in organic low fat (1.5%) milk from the organic dairy NATUR- M/ELK, Tinglev, Denmark. which consists of skim milk powder at a level of dry matter of 9.5% (w/v) reconstituted in distilled water and pasteurized at \(99^{\circ}\mathrm{C}\) for 30 min, followed by cooling to \(30^{\circ}\mathrm{C}\) (Sørensen et al., 2016). Starting material for all inoculations of were concentrated F- DVS® (Direct Vat Set) strains and cultures for direct inoculation into. All cultures were inoculated \(0.02\%\) with F- DVS material by firstly transfer \(2.0\mathrm{gF - DVS}\) material to \(200\mathrm{ml}\) cold milk. After mixing, \(4\mathrm{ml}\) were transferred to \(200\mathrm{ml}\) pre- warmed milk ( \(32^{\circ}\mathrm{C}\) ) for the final fermentation. The addition percentages of the different strains and cultures for the 6 different culture blends is given in Supplementary Table 15. + +<|ref|>sub_title<|/ref|><|det|>[[72, 805, 393, 825]]<|/det|> +## Milk acidifications and sampling + +<|ref|>text<|/ref|><|det|>[[70, 842, 911, 922]]<|/det|> +After inoculation into bottles with milk placed at \(32^{\circ}\mathrm{C}\) in a waterbath, pH measurement was started and acidification was followed using a CINAC system (Corrieu et al., 1988). After 59 min, the temperature was raised to \(38^{\circ}\mathrm{C}\) and at 183 min, the temperature was reduced to the final temperature, \(35^{\circ}\mathrm{C}\) , for 20 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 78, 911, 191]]<|/det|> +hours. The 6 different culture blends were each run in triplicate. After around 5 hours of acidification during the exponential growth phase when reaching pH 5, 1 g fermented milk from each fermentation and 1 g unfermented milk were sampled and added \(200\mu 14\mathrm{NH}2\mathrm{SO}4\) to prepare for analysis for organic acids, Carbohydrates and Volatiles. In addition 1 g was samples for mRNA extraction. + +<|ref|>sub_title<|/ref|><|det|>[[88, 224, 483, 245]]<|/det|> +## Carbohydrates, acids and volatiles in milk + +<|ref|>text<|/ref|><|det|>[[88, 260, 912, 737]]<|/det|> +For the preparation of fermented milk samples used for carbohydrate and small organic acid analysis, 1 g or 1 mL of sample was quenched with \(200\mu \mathrm{L}4\mathrm{N}\) sulfuric acid (H2SO4) in a \(7\mathrm{mL}\) glass tube, mixed and stored at \(- 20^{\circ}\mathrm{C}\) until analysis. The analytes for carbohydrate analysis were extracted from the sample and proteins get deproteinated and precipitated by treatment with an aqueous perchloric acid (PCA) solution. The samples get further diluted to fit into the dynamic range of the quantification. Arabinose is added as an internal standard. The diluted samples are analyzed on a Dionex ICS- 3000 system (Thermo Fischer Scientific, Waltham (MA), USA) using an analytical anion- exchange column and a pulsed amperometric detector (PAD). For quantification a one- point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (arabinose). The analytes for analysis of small organic acids were extracted from the sample and proteins get deproteinated and precipitated adding an aqueous PCA solution. The samples get further diluted to fit into the dynamic range of the quantification. Adipic acid is added as an internal standard. The diluted samples are analyzed on a Dionex ICS- 3000 or ICS- 5000 system (Thermo Fischer Scientific, Waltham, MA, USA) using an analytical ion exclusion column and a suppressed conductivity detector (SCD). For quantification, an 8- point calibration curve is used. Concentrations are calculated based on the chromatographic peak heights after normalizing to the internal standard (adipic acid). + +<|ref|>text<|/ref|><|det|>[[88, 752, 911, 923]]<|/det|> +For the preparation of the volatile organic compounds (VOC's) samples, 1 g or 1 mL of each sample was transferred to a headspace vial (20 mL) with \(200\mu \mathrm{L}\) of 4N H2SO4 and sealed with teflon- lined aluminium caps. The samples were then identified using a static head space sampler connected to a Gas Chromatograph with Flame Ionization Detector (GC- FID) (Perkin Elmer, MA, USA) and equipped with a HP- FFAP column. The identification of VOC's was based on retention time in comparison with that of standards. The injector and detector were maintained at \(180^{\circ}\mathrm{C}\) and \(220^{\circ}\mathrm{C}\) , respectively. The oven + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 77, 912, 280]]<|/det|> +was initially heated to \(60^{\circ}\mathrm{C}\) and held for \(2\mathrm{min}\) , then increased to \(230^{\circ}\mathrm{C}\) and held for \(0.5\mathrm{min}\) . The calculation of the concentration of each compound in each sample was based on the peak height divided by the response factor. The response factor is established suing standard solutions by the quotient of the peak height divided by the known sample concentration. All of the chemicals and analytes used for the standard solutions were provided by Sigma- Aldrich, Munich, Germany. Overall with these approaches, we measured a total of 43 metabolites comprising five carbohydrates, eight organic acids and 30 important flavor compounds. + +<|ref|>sub_title<|/ref|><|det|>[[88, 322, 476, 343]]<|/det|> +## RNA extraction, sequencing and analysis + +<|ref|>text<|/ref|><|det|>[[85, 355, 912, 928]]<|/det|> +Total RNA was extracted using the RNeasy Mini kit (Qiagen 74104) and rRNA was depleted with the NEBNext rRNA Depletion Kit (Bacteria). The sequencing library was then prepared using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina and the libraries were sequenced by EMBL Genomic Core Facility using the Illumina NextSeq 500 system, read length 150bp. The RNA- reads are preprocessed and filtered based on minimum quality and length using the NGLess pipeline and substrim. Given a read it finds the longest substring, such that all bases are of at least the given quality. The parameters minimum quality and minimum length were set to 25 and 100, respectively. The filtered reads are mapped to the genomes of all strains present in the Culture using bbmap – if one read has multiple hits random distribution is used (selecting one top- scoring site randomly). The total number of reads ranges from 4,855,868 to 18,812,096 in samples 4B and 1B, respectively, and the percentage of mapped reads is in the range of 93- 95% (Supplementary Table 5). htseq- count was used to count the number of reads per gene. The resulting counts grouped per orthology group (custom python script) based on a previously created pangenome created for the pan- metabolic network. Normalization was performed and differential gene expression with DESeq2 and SARTools in R [39, 40]. Further differentially expressed genes, Lactococcus pan- metabolic network information and KEGG pathways were based on custom analysis [41, 21]. The data was normalized using DESeq2 to correct systematic technical biases and make it possible to compare read counts across samples. The median scaling factor for each sample was used. This is done for the data containing all genes. RegPrecise database was use for validation of key functional regulation [42]. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 88, 300, 111]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[88, 133, 912, 368]]<|/det|> +AcknowledgmentsThe authors wish to thank Jannik Vindeloev and Ana Rute Neves for helpful discussions. Karsten Hellmuth, Eric Johansen and Mads Bennedsen for their help with project management and funding acquisition. Also, Kosai Al- Nakeeb, Anna Koza, Jacob Bælum, and Martin Abel- Kistrup for their significant contribution at genome sequencing and assembly. Lisandra Zepeda and Ida Bærholm Schnell for their help during cheese sampling. Finally, Gunnar Øregaard for providing acidification data for individual strains. The research of CM was supported by a Grand Solution grant from Innovation Fund Denmark (grant no. 6150- 00033B), The FoodTranscriptomics project and by the Dutch Research Council, as part of the MiCRop Consortium (NWO/OCW grant no. 024.004.014) + +<|ref|>sub_title<|/ref|><|det|>[[90, 411, 393, 434]]<|/det|> +## Data Availability Statement + +<|ref|>text<|/ref|><|det|>[[90, 457, 910, 540]]<|/det|> +Data Availability StatementProcessed data, including metabolomics, metatranscriptomics count tables as well as the code to reproduce the results are available in https://github.com/Chrats- Melkonian/mi_cheese. Genomes and raw metatranscriptomics data will become available before publication. + +<|ref|>sub_title<|/ref|><|det|>[[90, 585, 443, 606]]<|/det|> +## Author contributions statement + +<|ref|>text<|/ref|><|det|>[[88, 630, 912, 925]]<|/det|> +Author contributions statementCM, organizing ideas, writing, comparative genomics, phylogenomics, transcriptomics and bioinformatics analyses. FZ, genome- scale metabolic modeling, community simulations and writing. IK, initial transcriptomics, bioinformatics analyses and genome- scale metabolic modeling. MJ and KSr, carried out the experiments and drafted the manuscript part relating to sections “Inoculations in milk” and “Milk acidifications and sampling”. SB, mRNA extraction of the milk fermentation samples for mRNA sequencing. DM, analysed initial metatranscriptomics data, performed model reconstruction and community simulations. LTA, participated in conceptualization of the study, design and supervise the year- long cheddar- making experiment. KrP, participated in conceptualization of the study, funding acquisition, supervision of modeling and critically revising the manuscript. AZ, participated in conceptualization of the study, funding acquisition, supervision of experimental design, data analysis and bioinformatics, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 80, 355, 99]]<|/det|> +critically revising the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[56, 142, 466, 165]]<|/det|> +## Declaration of competing interest + +<|ref|>text<|/ref|><|det|>[[56, 188, 912, 270]]<|/det|> +CM, IK, MJ, KsR, LtA and AAZ are present or previous employees at Chr. Hansen A/S, a global supplier microbial cultures for food fermentation. The authors' views presented in this study, however, are solely based on scientific grounds and do not reflect the commercial interest of their employer. + +<|ref|>sub_title<|/ref|><|det|>[[56, 313, 220, 333]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[95, 358, 912, 437]]<|/det|> +1. Smith J, Yeluripati J, Smith P, Nayak DR. Potential yield challenges to scale-up of zero budget natural farming. Nature Sustainability. 2020 Mar;3(3):247-252. Available from: https://doi.org/10.1038/s41893-019-0469-x. + +<|ref|>text<|/ref|><|det|>[[95, 454, 912, 536]]<|/det|> +2. Wastyk HC, Fragiadakis GK, Perelman D, Dahan D, Merrill BD, Yu FB, et al. Gut-microbiota-targeted diets modulate human immune status. Cell. 2021;184(16):4137-4153. 4137. Available from: https://doi.org/10.1016/j.cell.2021.06.019. + +<|ref|>text<|/ref|><|det|>[[95, 552, 912, 634]]<|/det|> +3. Blaya J, Barzideh Z, LaPointe G. Symposium review: Interaction of starter cultures and nonstarter lactic acid bacteria in the cheese environment. Journal of Dairy Science. 2018 Apr;101(4):3611-3629. Available from: https://doi.org/10.3168/jds.2017-13345. + +<|ref|>text<|/ref|><|det|>[[95, 650, 912, 730]]<|/det|> +4. Song AA, In LLA, Lim SHE, Rahim RA. Erratum to: A review on Lactococcus lactis: from food to factory. Microbial Cell Factories. 2017 Aug;16(1). Available from: https://doi.org/10.1186/s12934-017-0754-1. + +<|ref|>text<|/ref|><|det|>[[95, 745, 912, 856]]<|/det|> +5. Markakiou S, Gaspar P, Johansen E, Zeidan AA, Neves AR. Harnessing the metabolic potential of Streptococcus thermophilus for new biotechnological applications. Current Opinion in Biotechnology. 2020 Feb;61:142-152. Available from: https://doi.org/10.1016/j.copbio.2019.12.019. + +<|ref|>text<|/ref|><|det|>[[95, 872, 912, 924]]<|/det|> +6. Rau MH, Gaspar P, Jensen ML, Geppel A, Neves AR, Zeidan AA. Genome-Scale Metabolic Modeling Combined with Transcriptome Profiling Provides Mechanistic Understanding of Streptococcus ther + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 78, 912, 128]]<|/det|> +547 mophilus CH8 Metabolism. Applied and Environmental Microbiology. 2022 Aug;88(16). Available from: https://doi.org/10.1128/aem.00780- 22. + +<|ref|>text<|/ref|><|det|>[[97, 145, 912, 255]]<|/det|> +7. Pastink MI, Teusink B, Hols P, Visser S, de Vos WM, Hugenholtz J. Genome-Scale Model of Streptococcus thermophilus LMG18311 for Metabolic Comparison of Lactic Acid Bacteria. Applied and Environmental Microbiology. 2009 Jun;75(11):3627-3633. Available from: https://doi.org/10.1128/aem.00138-09. + +<|ref|>text<|/ref|><|det|>[[97, 272, 912, 351]]<|/det|> +8. Oliveira AP, Nielsen J, Förster J. Modeling Lactococcus lactis using a genome-scale flux model. BMC Microbiology. 2005 Jun;5(1). Available from: https://doi.org/10.1186/1471-2180-5-39. + +<|ref|>text<|/ref|><|det|>[[97, 368, 912, 477]]<|/det|> +9. Champagne CP, Gagnon D, St-Gelais D, Vuillemand JC. Interactions between Lactococcus lactis and Streptococcus thermophilus strains in Cheddar cheese processing conditions. International Dairy Journal. 2009 Nov;19(11):669-674. Available from: https://doi.org/10.1016/j.idairyj.2009.06.002. + +<|ref|>text<|/ref|><|det|>[[97, 493, 912, 604]]<|/det|> +10. Duru IC, Laine P, Andreevskaya M, Paulin L, Kananen S, Tynkkynen S, et al. Metagenomic and metatranscriptomic analysis of the microbial community in Swiss-type Maasdam cheese during ripening. International Journal of Food Microbiology. 2018 Sep;281:10-22. Available from: https://doi.org/10.1016/j.ijfoodmicro.2018.05.017. + +<|ref|>text<|/ref|><|det|>[[97, 620, 912, 700]]<|/det|> +11. Salazar JK, Carstens CK, Ramachandran P, Shazer AG, Narula SS, Reed E, et al. Metagenomics of pasteurized and unpasteurized gouda cheese using targeted 16S rDNA sequencing. BMC Microbiology. 2018 Nov;18(1). Available from: https://doi.org/10.1186/s12866-018-1323-4. + +<|ref|>text<|/ref|><|det|>[[97, 716, 912, 796]]<|/det|> +12. Walsh AM, Macori G, Kilcawley KN, Cotter PD. Meta-analysis of cheese microbiomes highlights contributions to multiple aspects of quality. Nature Food. 2020 Aug;1(8):500-510. Available from: https://doi.org/10.1038/s43016-020-0129-3. + +<|ref|>text<|/ref|><|det|>[[97, 812, 912, 920]]<|/det|> +13. Hutkins R, Halambeck SM, Morris HA. Use of Galactose-Fermenting Streptococcus thermophilus in the Manufacture of Swiss, Mozzarella, and Short-Method Cheddar Cheese. Journal of Dairy Science. 1986 Jan;69(1):1-8. Available from: https://doi.org/10.3168/jds.s0022-0302(86)80361-7. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 912, 191]]<|/det|> +14. Proust L, Haudebourg E, Sourabié A, Pedersen M, Besançon I, Monnet V, et al. Multi-omics Approach Reveals How Yeast Extract Peptides Shape Streptococcus thermophilus Metabolism. Applied and Environmental Microbiology. 2020 oct;86(22). Available from: https://doi.org/10.1128% 2Faem.01446-20. + +<|ref|>text<|/ref|><|det|>[[56, 205, 912, 315]]<|/det|> +15. Zeleznik A, Andrejev S, Ponomarova O, Mende DR, Bork P, Patil KR. Metabolic dependencies drive species co-occurrence in diverse microbial communities. Proceedings of the National Academy of Sciences. 2015 May;112(20):6449-6454. Available from: https://doi.org/10.1073/pnas.1421834112. + +<|ref|>text<|/ref|><|det|>[[56, 332, 912, 410]]<|/det|> +16. Nugroho ADW, Kleerebezem M, Bachmann H. Growth, dormancy and lysis: the complex relation of starter culture physiology and cheese flavour formation. Current Opinion in Food Science. 2021;39:22-30. + +<|ref|>text<|/ref|><|det|>[[56, 427, 911, 508]]<|/det|> +17. Smid EJ, Kleerebezem M. Production of Aroma Compounds in Lactic Fermentations. Annual Review of Food Science and Technology. 2014 feb;5(1):313-326. Available from: https://doi.org/10.1146%2Fannurev-food-030713-092339. + +<|ref|>text<|/ref|><|det|>[[56, 523, 912, 605]]<|/det|> +18. Calbert HE, Price WV. A Study of the Diacetyl in Cheese. I. Diacetyl Content and Flavor of Cheddar Cheese. Journal of Dairy Science. 1949 jun;32(6):515-520. Available from: https://doi.org/10.3168%2Fjds.s0022-0302%2849%2992073-1. + +<|ref|>text<|/ref|><|det|>[[56, 620, 911, 700]]<|/det|> +19. Baumler AJ, Sperandio V. Interactions between the microbiota and pathogenic bacteria in the gut. Nature. 2016 Jul;535(7610):85-93. Available from: https://doi.org/10.1038/nature18849. + +<|ref|>text<|/ref|><|det|>[[56, 716, 911, 796]]<|/det|> +20. Lara EG, van der Windt I, Molenaar D, de Vos MGJ, Melkonian C. Using Functional Annotations to Study Pairwise Interactions in Urinary Tract Infection Communities. Genes. 2021 aug;12(8):1221. Available from: https://doi.org/10.3390%2Fgenes12081221. + +<|ref|>text<|/ref|><|det|>[[56, 812, 912, 924]]<|/det|> +21. Melkonian C, Fillinger L, Atashgahi S, da Rocha UN, Kuiper E, Olivier B, et al. High biodiversity in a benzene-degrading nitrate-reducing culture is sustained by a few primary consumers. Communications Biology. 2021 may;4(1). Available from: https://doi.org/10.1038%2Fs42003-021-01948-y. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 912, 160]]<|/det|> +22. Melkonian C, Seidl MF, van der Hooft JJJ, de Vos MGJ. Metabolic interactions shape a community's phenotype. Trends in Microbiology. 2022 jul;30(7):609-611. Available from: https://doi.org/10.1016%2Fj.tim.2022.05.001. + +<|ref|>text<|/ref|><|det|>[[55, 183, 911, 266]]<|/det|> +23. Machado D, Maistrenko OM, Andrejev S, Kim Y, Bork P, Patil KR, et al. Polarization of microbial communities between competitive and cooperative metabolism. Nature Ecology & Evolution. 2021 Jan;5(2):195-203. Available from: https://doi.org/10.1038/s41559-020-01353-4. + +<|ref|>text<|/ref|><|det|>[[55, 288, 911, 369]]<|/det|> +24. Irlinger F, Mounier J. Microbial interactions in cheese: implications for cheese quality and safety. Current Opinion in Biotechnology. 2009 Apr;20(2):142-148. Available from: https://doi.org/10.1016/j.copbio.2009.02.016. + +<|ref|>text<|/ref|><|det|>[[55, 392, 912, 475]]<|/det|> +25. Jiang N, Wu R, Wu C, Wang R, Wu J, Shi H. Multi-omics approaches to elucidate the role of interactions between microbial communities in cheese flavor and quality. Food Reviews International. 2022 Apr;1-13. Available from: https://doi.org/10.1080/87559129.2022.2070199. + +<|ref|>text<|/ref|><|det|>[[55, 497, 912, 610]]<|/det|> +26. Parente E, Zotta T, Ricciardi A. A review of methods for the inference and experimental confirmation of microbial association networks in cheese. International Journal of Food Microbiology. 2022 May;368:109618. Available from: https://doi.org/10.1016/j.ijfoodmicro.2022.109618. + +<|ref|>text<|/ref|><|det|>[[55, 633, 912, 715]]<|/det|> +27. Ponomarova O, Gabrielli N, Sévin DC, Mülleder M, Zirngibl K, Bulyha K, et al. Yeast Creates a Niche for Symbiotic Lactic Acid Bacteria through Nitrogen Overflow. Cell Systems. 2017 Oct;5(4):345-357.e6. Available from: https://doi.org/10.1016/j.cels.2017.09.002. + +<|ref|>text<|/ref|><|det|>[[55, 737, 912, 820]]<|/det|> +28. Blasche S, Kim Y, Mars RA, Machado D, Maansson M, Kafka E, et al. Metabolic cooperation and spatiotemporal niche partitioning in a kefir microbial community. Nature microbiology. 2021;6(2):196-208. + +<|ref|>text<|/ref|><|det|>[[55, 842, 912, 925]]<|/det|> +29. Goyal A, Bittleston LS, Leventhal GE, Lu L, Cordero OX. Interactions between strains govern the eco-evolutionary dynamics of microbial communities. eLife. 2022 feb;11. Available from: https://doi.org/10.7554%2Felife.74987. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 77, 912, 191]]<|/det|> +30. Oren A, Garrity GM. Notification that new names of prokaryotes, new combinations and new taxonomic opinions have appeared in volume 71, part 3 of the IJSEM. International Journal of Systematic and Evolutionary Microbiology. 2021 Jun;71(6). Available from: https://doi.org/10.1099/ijsem.0.004812. + +<|ref|>text<|/ref|><|det|>[[50, 207, 912, 291]]<|/det|> +31. Fu L, Niu B, Zhu Z, Wu S, Li W. CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics. 2012 oct;28(23):3150-3152. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbts565. + +<|ref|>text<|/ref|><|det|>[[50, 308, 911, 360]]<|/det|> +32. R Core Team. R: A language and environment for statistical computing. Vienna, Austria; 2020. Available from: https://www.R-project.org/. + +<|ref|>text<|/ref|><|det|>[[50, 375, 912, 488]]<|/det|> +33. Asnicar F, Thomas AM, Beghini F, Mengoni C, Manara S, Manghi P, et al. Precise phylogenetic analysis of microbial isolates and genomes from metagenomes using PhyloPhlAn 3.0. Nature Communications. 2020 may;11(1). Available from: https://doi.org/10.1038%2Fs41467-020-16366-7. + +<|ref|>text<|/ref|><|det|>[[50, 504, 911, 586]]<|/det|> +34. Edgar RC. MUSCLE: multiple sequence alignment with high accuracy and high throughput. Nucleic Acids Research. 2004 mar;32(5):1792-1797. Available from: https://doi.org/10.1093%2Fnar%2Fgkh340. + +<|ref|>text<|/ref|><|det|>[[50, 603, 911, 687]]<|/det|> +35. Stamatakis A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics. 2014 jan;30(9):1312-1313. Available from: https://doi.org/10.1093%2Fbioinformatics%2Fbtu033. + +<|ref|>text<|/ref|><|det|>[[50, 703, 911, 787]]<|/det|> +36. Machado D, Andrejev S, Tramontano M, Patil KR. Fast automated reconstruction of genome-scale metabolic models for microbial species and communities. Nucleic Acids Research. 2018 06;46(15):7542-7553. Available from: https://doi.org/10.1093/nar/gky537. + +<|ref|>text<|/ref|><|det|>[[50, 803, 911, 855]]<|/det|> +37. Hyatt D, Chen GL, LoCascio PF, Land ML, Larimer FW, Hauser LJ. Prodigal: prokaryotic gene recognition and translation initiation site identification. BMC bioinformatics. 2010;11(1):1-11. + +<|ref|>text<|/ref|><|det|>[[50, 871, 911, 924]]<|/det|> +38. Ebrahim A, Lerman JA, Palsson BO, Hyduke DR. COBRApy: constraints-based reconstruction and analysis for python. BMC systems biology. 2013;7(1):1-6. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 912, 157]]<|/det|> +652 39. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology. 2014 dec;15(12). Available from: https://doi.org/10.1186%2Fs13059- 014- 0550- 8. + +<|ref|>text<|/ref|><|det|>[[55, 175, 912, 286]]<|/det|> +655 40. Varet H, Brillet-Guéguen L, Coppée JY, Dillies MA. SARTools: A DESeq2- and EdgeR- Based R Pipeline for Comprehensive Differential Analysis of RNA-Seq Data. PLOS ONE. 2016 jun;11(6):e0157022. Available from: https://doi.org/10.1371%2Fjournal.pone.0157022. + +<|ref|>text<|/ref|><|det|>[[55, 303, 912, 412]]<|/det|> +659 41. Melkonian C, Gottstein W, Blasche S, Kim Y, Abel-Kistrup M, Swiegers H, et al. Finding Functional Differences Between Species in a Microbial Community: Case Studies in Wine Fermentation and Kefir Culture. Frontiers in Microbiology. 2019 jun;10. Available from: https://doi.org/10.3389%2Ffmicb.2019.01347. + +<|ref|>text<|/ref|><|det|>[[55, 429, 912, 540]]<|/det|> +663 42. Novichkov PS, Kazakov AE, Ravcheev DA, Leyn SA, Kovaleva GY, Sutormin RA, et al. Reg- Precise 3.0 - A resource for genome-scale exploration of transcriptional regulation in bacteria. BMC Genomics. 2013;14(1):745. Available from: https://doi.org/10.1186/1471- 2164- 14- 745. + +<|ref|>sub_title<|/ref|><|det|>[[58, 583, 293, 604]]<|/det|> +## figures and tables + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[92, 225, 904, 551]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 564, 910, 770]]<|/det|> +
Figure 1. Schematic representation of the experimental design and methods (a) The 1 year long cheddar-making experiment. Microbial population dynamics were quantified during cheese ripening using a selective method of viable cell counting, which discriminates between thermophilic cocci, mesophilic cocci and non-starter lactic acid bacteria (NSLAB). Additionally, metabolic changes in the cheeses were measured using several targeted analytical chemistry approaches. (b) Schematic representation of the controlled milk experiment in the laboratory as well as of the methods (1-5) used in the controlled milk experiment. This second experiment involves the removal of additional strains, namely the individual exclusion of three major Lactococcus strains. The numbers indicate the different type of data and analysis as follows: 1. metatrancriptomics, 2. metabolomics, 3. genomics, 4. phylogenomics and 5. genomes-scale metabolic models (GEMs) and community simulations. Our integrative systems biology approach combined: i) the analysis of the SLAB community’s genomes, ii) the generation and simulation of their respective GEMs, iii) the analysis of the metatranscriptomes across the different strain removal conditions and iv) the quantification of key metabolites.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[91, 157, 905, 484]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 494, 908, 837]]<|/det|> +
Figure 2. S. thermophilus benefit the growth of lactococci community and influence the final flavor of cheese. (a-d) Microbial population dynamics during cheese ripening. Relationship between CFUs and time (months) for (a) Mesophilic, (c) Thermophilic cocci, and (d) non-starter lactic acid bacteria (NSLAB). (b) Presents the boxplot comparison of the different condition on Mesophilic cocci at 12 months. Note, significant reduction only on Mesophilic cocci is observed when S. thermophilus is absent. Condition All corresponds to whole culture which is composed from 24 number of strains. HP corresponds to an alternative method (hand packed) of the whole culture inoculation. -LB corresponds to the removal of a L. lactis blend population. -ST corresponds to the removal of the S. thermophilus strain. (e-h) Metabolome dynamics during cheese ripening. (e) 2-dimensional representation with the usage of UMAP of all samples based on the metabolomics measurements (Acids, Carbohydrates and Peptides). The colour indicates the time when the sample was taken and the shape indicates the four different conditions. Note, the stronger change is observed between 2 week and 3 months while the removal of ST (cross symbol) has a strong effect on the later time of cheese ripening. (f) Relative change of the metabolites on different time intervals including all the conditions, both colour and shape indicates the class of the metabolite. Note, the higher relative change of peptides followed by acids at the interval between 2 week and 3 months, later acids exhibit higher changes. The three measured sugars (Glucose, Lactose and Galactose) are excluded from this panel. (g) Selection of the six most discriminative metabolites out of 50 for the four conditions, colour indicate the condition while the right side shape the class of the metabolite. Note, the majority of peptides concentration are significant different when S. thermophilus is not present as well as with galactose, lactose and lactic acid. (h) Galactose concentration over time highlights the absence of galactose when S. thermophilus is not present. The galactose concentration remains steady from the start till the 9 month of cheese ripening and then shows signs of decline.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 197, 905, 479]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 495, 910, 803]]<|/det|> +
Figure 3. The removal of different Lactococcus strains has a distinct effect on gene expression in S. thermophilus. (a-e) relationship of the Lactococcus phylogeny with the S. thermophilus transcription change on different dropout conditions. (a) Lactococcus phylogenetic tree separates the L. cremoris and L. lactis strains. The two clades are colored with transparent yellow and grey, respectively. The colors and shape of the tree tips indicate the strains presents in the SLAB culture. Tree empty tips correspond to the obtained complete NCBI genomes of Lactococcus. Additional information is presented in the outer layers of the tree. The first layer presents the unique k-mer content based on the SLAB culture's genomes (purple-scale color) and indicates higher number for the strains who have no close relative within the culture. The second layer presents the proxy of temperature stress tolerance based on Pearson Correlation Coefficient (PCC) between individual acidification curve of 30 and 40 degree \(^\circ \mathrm{C}\) (red-scale color). High values correspond to high temperature tolerance. The clade L. cremoris shows lower temperature tolerance, with few exceptions including the main LC strain (more elaborate data are presented in Supplementary Fig. 5). The third layer presents the percentile of transcribed sigletons (green-scale color) as well as the total number of the strains sigletons (outer bars). (b-e) Volcano plots presents the S. thermophilus transcription between the whole community (All) and the different Lactococcus dropout conditions, respectively. The colors corresponds to the 4 Lactococcus components. The S. thermophilus transcription observed to change more when LLm1 was left-out followed by LB, with 291 and 182 number of total significant up/down regulated genes, respectively. A smaller effect of 21 genes was observed when LC was left-out and only 1 gene when LLm2 was left out.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[86, 103, 910, 504]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 513, 910, 892]]<|/det|> +
Figure 4. S. thermophilus provides nitrogen in form of amino acids to L. lactis community, which is necessary for de novo nucleotide biosynthesis. (a) Summary of metabolic modeling analysis including individual model and community-based simulations. (b) Selected exchange fluxes predicted across individual flux balance analysis simulations carried in milk media variations for each set of metabolic models, in particular fermentation products, amino acids, and carbon sources highlighting differences in metabolic strategies across species. (c) Alluvial diagram showing predicted metabolic exchanges from community simulations in milk media variants, highlighting the fact the amino acid valine is strongly predicted across all simulation conditions where metabolic cross-talk is expected. Refer to Supplementary Figure 6 for more details on the metabolic models and simulation. (d-i) Transcriptomics profiles between All and -ST (S. thermophilus is left out) conditions of key functions for all the strains in SLAB culture. (d-f) Transcriptomics boxplots of branched chain amino acids (BCAA) aminotransferase, BCAA transport system 2 carrier protein and ammonium transporter (T) supports the predictions of metabolic exchanges. Note, (e) the higher transcription profile of S. thermophilus on BCAA transport system as well as the high transcription profile of S. thermophilus and the increase of LC transcription profile when S. thermophilus was left out. (g-j) Up-regulation of Lactococcus glutamine and nucleotide metabolism when S. thermophilus is left out. (g-i) Boxplots present the activities of nitrogen regulatory protein P-II, glutamine synthetase (glnA) and related transcriptional regulator (TR) (GlnR) for all community members. Note, the up-regulation of all the enzymes derived from Lactococcus strains when S. thermophilus is left out. (j) Metatranscriptomics represents the increased activity of Lactococcus reactions, which are incorporated into Escher maps. The pattern shows a coordinate up-regulation (with green) of the enzymes towards guanine biosynthesis (and uridine). (k) Schematic compilation of the compounds S. thermophilus may provide to Lactococcus community as well as selected transcriptional changes of the Lactococcus community when S. thermophilus was left out. The color and direction of the arrows represent the up- and down- regulation with green/up and red/down, respectively.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[171, 75, 820, 535]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 546, 911, 958]]<|/det|> +
Figure 5. Cheese flavor compounds are strongly influenced by the interactions within Lactococcus community The measured metabolome from the controlled milk experiment in response to stain removal conditions (a-d). (a) The PCA of key carbohydrates, acids and cheese flavor compounds. The strongest response observed from the removal of L. cremoris followed by coordinated response of LLm1, LB and S. thermophilus. Removal of LLm2 has no effect. The arrows represents the top 5 compounds (2,3-Pentanedione, Hexanal, Diacetyl, Acetoin and Ethyl acetate) responsible for the observed changes. Panels (b-c) represents the signal-to-noise ratio (S/N) of diacetyl and acetoin, respectively. Regarding those two compounds, absence of S. thermophilus from the community led to significant lower accumulation, while absence of L. cremoris leads to significant higher accumulation. The black horizontal line indicate the average value of the two compounds at milk prior to acidification. Paired t-test for each condition against the whole community (All) were performed with the symbols indicating the statistical significance results: ns: \(\mathrm{p} > 0.05\) , \*: \(\mathrm{p}< = 0.05\) and \*\*: \(\mathrm{p}< = 0.01\) . (d) Schematic compilation of the compounds that changed between the whole SLAB culture and when L. cremoris was left out (for detailed boxplots see supplementary). The color represent the removal conditions; yellow indicates the removal of L. cremoris while red represents the whole SLAB culture. The upwards arrow indicates the increase of the compounds relative to the compounds concentration in milk. The X symbol indicates the lack of production of the compounds in the respective condition. (e) Simplified metabolic escher maps from citrate to \(\alpha\) -ketoglutarate as well as to diacetyl through pyruvate and (S)-2-Acetolactate. Metabolites and reactions are indicated with black and grey color in the graph, respectively. Full colored circles and open colored circles indicate the high and low transcription per strain, respectively. Lack of circle visualization indicates the lack of the corresponding orthologous group genes. Note, the transcriptional fluxes from citrate-sodium symporter towards diacetyl, which are present to all members in Lactococcus community. In addition, LC and LLm1 manifest higher transcription among the Lactococcus community on the genes related to acetolactate decarboxylase. Only LC strain manifests transcriptional flux through diacetyl reductase and butanediol dehydrogenase as well as in reactions towards \(\alpha\) -ketoglutarate. (f) Predicted fluxes based on metabolic modeling across the different simulated media.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 130, 250, 150]]<|/det|> +- MICheeseSUPP.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/images_list.json b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..f11ca1b5c427b380d966343899a517b2b481e1c3 --- /dev/null +++ b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/images_list.json @@ -0,0 +1,55 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Crystal structure and sodium-ion conduction characteristic of crystalline \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . a Rietveld refinement based on the powder XRD. b Nyquist plots of \\(\\mathrm{Na_5SmSi_4O_{12}}\\) from 25 to \\(175^{\\circ}\\mathrm{C}\\) . c Arrhenius plot of the conductivity values for \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . d Top-down and e perspective view of crystal structures of crystalline \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . f-g Molecular dynamics simulation trajectories of \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . h Minimum potential energy path along \\(\\mathrm{Na^{+}}\\) diffusion route in crystalline \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . i Solid-state \\(^{23}\\mathrm{Na}\\) NMR spectrum and its simulation for the crystalline \\(\\mathrm{Na_5SmSi_4O_{12}}\\) . The gray line is experimental data and the green-dashed line is the sum of simulation. j Saturation recovery fitting curve for the data obtained at room temperature. k Temperature dependence of \\(^{23}\\mathrm{Na}\\) NMR relaxation rate as a function of temperature in \\(\\mathrm{K}^{-1}\\) . The solid line is the fit according to Eq. (1). The derivation of the data is not used for the fit.", + "footnote": [], + "bbox": [ + [ + 156, + 92, + 840, + 571 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Characterization of interfacial evolution and \\(\\mathrm{Na_5SmSi_4O_{12}}\\) after cycling in \\(\\mathrm{Na}||\\mathrm{Na_5SmSi_4O_{12}}||\\mathrm{Na}\\) symmetric cells. a Cycling performance of \\(\\mathrm{Na}||\\mathrm{Na_5SmSi_4O_{12}}||\\mathrm{Na}\\) at room temperature. b Nyquist plots of the SE-based symmetric cell after cycling for different times. Cross-sectional SEM images of the Na metal/ \\(\\mathrm{Na_5SmSi_4O_{12}}\\) interfaces: c pristine and d 100 h cycling. e XRD profiles of \\(\\mathrm{Na_5SmSi_4O_{12}}\\) after cycling for different times. f HRTEM and g SAED patterns of \\(\\mathrm{Na_5SmSi_4O_{12}}\\) after cycling.", + "footnote": [], + "bbox": [ + [ + 196, + 210, + 789, + 580 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. The superiority of amorphous bulk materials and interface. a Solid-state \\(^{23}\\mathrm{Na}\\) NMR of the pristine crystalline and the cycled amorphous \\(\\mathrm{Na_{5}SmSi_{4}O_{12}}\\) . b \\(^{23}\\mathrm{Na}\\) NMR relaxation rate of Na cycled \\(\\mathrm{Na_{5}SmSi_{4}O_{12}}\\) as a function of temperature in \\(\\mathrm{K}^{-1}\\) . c Nanoindentation load-displacement curves of crystalline \\(\\mathrm{Na_{5}SmSi_{4}O_{12}}\\) and amorphous \\(\\mathrm{Na_{5}SmSi_{4}O_{12}}\\) . d Schematic of interface morphology evolution during sodium plating/stripping.", + "footnote": [], + "bbox": [], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. CTA mechanism. a Structure of the \\(\\mathrm{Na_5SmSi_4O_{12}||Li_5SmSi_4O_{12}}\\) interface model b molecular dynamics trajectories of the interface model. Molecular dynamics simulation trajectories of crystalline c \\(\\mathrm{Li_5SmSi_4O_{12}dNa_5SmSi_4O_{12}}\\) and e \\(\\mathrm{Li_5Na_2SmSi_4O_{12}}\\) . f Minimum potential energy path along Li diffusion route in crystalline \\(\\mathrm{Na_2Li_5SmSi_4O_{12}}\\) . g XRD profiles of \\(\\mathrm{Na_5 - xLi_5SmSi_4O_{12}}\\) after cycling for different times. Solid-state h \\(^{23}\\mathrm{Na}\\) and i \\(^{7}\\mathrm{Li}\\) NMR of the \\(\\mathrm{Na_5SmSi_4O_{12}}\\) with different cycling times.", + "footnote": [], + "bbox": [ + [ + 163, + 90, + 825, + 455 + ] + ], + "page_idx": 19 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5.mmd b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5.mmd new file mode 100644 index 0000000000000000000000000000000000000000..21ecdcdd3d0b9dc36782eb8573c7114c400f1253 --- /dev/null +++ b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5.mmd @@ -0,0 +1,310 @@ + +# Electrochemically induced crystalline-to-amorphization transformation in sodium samarium silicate solid electrolyte for long-lasting all-solid-state sodium metal batteries + +Ge Sun Key Laboratory of Physics and Technology for Advanced Batteries (Ministry of Education), State Key Laboratory of Superhard Materials, College of Physics, Jilin University + +Chenjie Lou Center for High Pressure Science and Technology Advanced Research + +Zhixuan Wei Jilin University + +Shiyu Yao Jilin University + +Ziheng Lu University of Cambridge + +Gang Chen Jilin University + +Ze Shen Nanyang Technological University https://orcid.org/0000- 0001- 7432- 7936 + +Mingxue Tang Center for High Pressure Science & Technology Advanced Research + +Fei Du ( \(\circleddash\) dafei@jlu.edu.cn) Jilin University https://orcid.org/0000- 0001- 6413- 0689 + +Article + +Keywords: + +Posted Date: March 7th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2623650/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on October 16th, 2023. See the published version at https://doi.org/10.1038/s41467-023-42308-0. + +<--- Page Split ---> + +Electrochemically induced crystalline-to-amorphization transformation in sodium samarium silicate solid electrolyte for long-lasting all-solid-state sodium metal batteriesGe Sun1,4, Chenjie Lou2,4, Zhixuan Wei1, Shiyu Yao1, Ziheng Lu3\*, Gang Chen1, Zexiang Shen1, Mingxue Tang2\*, Fei Du1\*1Key Laboratory of Physics and Technology for Advanced Batteries (Ministry of Education), State Key Laboratory of Superhard Materials, College of Physics, Jilin University, Changchun, 130012, China.2Center for High Pressure Science and Technology Advanced Research (HPSTAR), Beijing 100193, China.3Department of Materials Science & Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge CB3 0FS, United Kingdom4These authors contributed equally: Ge Sun, Chenjie Lou.5Email: zluag@connect.ust.hk (Z. L.); mingxue.tang@hpostar.ac.cn (M. T.); dufei@jlu.edu.cn (F. D.) + +<--- Page Split ---> + +## Abstract + +Exploiting solid electrolyte (SE) materials with high ionic conductivity, good interfacial compatibility, and ultraconformal contact with electrode are essential for solid- state sodium metal batteries (SSBs). Here we report a crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) SE which features high room- temperature ionic conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) and a low activation energy of \(0.15 \mathrm{eV}\) . All- solid- state symmetric cell with \(\mathrm{Na_5SmSi_4O_{12}}\) delivers excellent cycling life over \(800 \mathrm{~h}\) at \(0.15 \mathrm{mA} \mathrm{h} \mathrm{cm}^{- 2}\) and high critical current density of \(1.4 \mathrm{mA} \mathrm{cm}^{- 2}\) . Such excellent electrochemical performance is attributed to an electrochemically induced in- situ crystalline- to- amorphous (CTA) transformation propagating from the interface to the bulk during repeated deposition and stripping of sodium, which lead to faster ionic transport and superior interfacial properties. Impressively, the \(\mathrm{Na_3V_2(PO_4)_3} \| \mathrm{Na_5SmSi_4O_{12}} \| \mathrm{Na} \mathrm{SSBs}\) achieves a remarkable cycling performance over 4000 cycles (6 months) with no capacity loss. These results not only identify \(\mathrm{Na_5SmSi_4O_{12}}\) as a promising SE, but also emphasize the potential of the CTA transition as a promising mechanism towards long- lasting SSBs. + +## Introduction + +All- solid- state batteries (ASSBs) are expected to provide key improvements over today's rechargeable batteries owing to the inherent merits of solid electrolytes (SEs) such as high safety, long- lasting life, and high energy density1, 2. Among them, sodium- based ASSBs are drawing ever- increasing interest because of the abundant resource in nature, beneficial to cost saving and sustainability3, 4. Despite the progress of sodium- + +<--- Page Split ---> + +based ASSBs in the past few years, it is still a significant challenge to exploit low- cost and facile synthesized SEs with high ionic conductivity, excellent mechanical and chemical stability. Moreover, the poor wetting of the solid- solid interface with sluggish interfacial kinetics is a big hurdle to future Na- ASSBs development5, 6. + +Generally, Na- based inorganic SEs can be divided into two categories, e.g., sulfides like \(\mathrm{Na_3SbS_4}\) , \(\mathrm{Na_{11}Sn_2PS_{12}}\) , \(\mathrm{Na_7P_3S_{11}}\) and oxides including \(\mathrm{Na - \beta^{\prime\prime} - Al_2O_3}\) , NASICON- type7. Sulfide electrolytes feature higher ionic conductivity and better ductility than the oxides \(\mathrm{SEs^{8, 9}}\) . However, their chemical instability against air and narrow electrochemical window are likely to induce complex side reaction, leading to shortened cycling lives10, 11. In contrast, NASICON- type oxides could deliver high thermal and chemical stability, and low thermal expansion12, 13. Nevertheless, large grain boundary resistance and harsh synthesis conditions are critically challenging for their application14, 15. Recently, our group reported a novel structure of SE, \(\mathrm{Na_5YSi_4O_{12}}\) with a high room- temperature ionic conductivity of \(1.59 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) , comparable with the sulfide \(\mathrm{SEs^{16}}\) . Furthermore, in comparison with the NASICON- type materials, \(\mathrm{Na_5YSi_4O_{12}}\) can be synthesized at a lower temperature which is beneficial to cost- and energy- saving. More impressively, the stable structure provides sufficient freedom of materials optimization and design by substituting the Y sites by for instance, In, Sc, and the rare earth Lu- Sm, to further enhance the ionic conductivity and understand the ion- conducting behavior of this class of materials17, 18. + +Besides the difficulty in the SEs exploration, another challenge of developing Na- ASSBs lies in the electrode- electrolyte interfaces, in the mechanics throughout the cell, + +<--- Page Split ---> + +and in processing at scale19. To begin with, the poor ductility and crystallographic orientation- dependent ionic transport properties of oxide SEs likely induce a large interfacial resistance and sluggish kinetics20, 21. In addition, during the continuous dissolution of sodium, the formation of pores at the sodium metal anode interface will further worsen the interfacial physical contact and generate unavoidable surface defects that could disturb sodium- ion flux and work as the nucleation center, leading to rapid dendrite nucleation and growth22, 23. Moreover, most SEs are intrinsically thermodynamically unstable against sodium metal, which induces degradation and forms mixed conducting interphases, further accelerating the dendrite propagation24, 25. To address these issues, many approaches are employed to fabricate an artificial layer on the surface of sodium metal to improve sodium wetting, chemical stability, and thus reduce interfacial impedance, such as \(\mathrm{TiO_2^{26}}\) , \(\mathrm{SnS_2^{27}}\) , etc28, 29. However, they face practical limitations, such as complex synthesis procedures and difficulty in controlling the thickness and achieving acceptable adhesion. Even worse, the artificial interface layer is likely to introduce additional interfacial issues with bulk SEs, such as unexpected phase transition, uneven ion flux distribution, and electrostatic potential drop and formation of “space- charge layer”, seriously limiting the ion transport and reducing the cycle life of ASSBs30. + +Herein, we report a new member of the \(\mathrm{Na_5MSi_4O_{12}}\) family with \(\mathrm{M} = \mathrm{Sm}\) . It has the highest room- temperature ionic conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) among the \(\mathrm{Na_5MSi_4O_{12}}\) family that have been reported. Interestingly, we observe an electrochemically induced crystalline- to- amorphous (CTA) transformation of + +<--- Page Split ---> + +\(\mathrm{Na_5SmSi_4O_{12}}\) SE during repeated deposition and stripping of Na. This CTA transition is attributed to the lattice stress generated upon \(\mathrm{Na^{+}}\) transportation rather than phase transformation due to chemical instability. When applied in a Li symmetric cell with the same cell configuration (Li|Na5SmSi4O12||Li), this CTA process is speeded up because of the mismatch between \(\mathrm{Li^{+}}\) and \(\mathrm{Na^{+}}\) ionic radius, which further improves the selectivity of the Li ASSBs. Beneficial from the enhanced mechanical properties, decreased ion mobility activation energy, and lower interfacial energy of amorphous material and interface than crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) , symmetric Na cells deliver a low overpotential of \(\sim 26 \mathrm{mV}\) and ultra- stable cycling performance over \(800 \mathrm{h}\) at \(0.15 \mathrm{mA} \mathrm{cm}^{- 2}\) . Moreover, the amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) facilitates intimate contact of SE with Na metal and brings essentially improved critical current density (CCD) of \(1.4 \mathrm{mA} \mathrm{cm}^{- 2}\) in comparison with the initial crystalline stage \((0.6 \mathrm{mA} \mathrm{cm}^{- 2})\) . By virtue of the decreased resistance of sodium metal anode, the assembled quasi- solid- state \(\mathrm{Na_3V_2(PO_4)_3||Na_5SmSi_4O_{12}||Na}\) cell demonstrates an ultra- long cycle lives over 4000 cycles with \(\sim 100\%\) Coulombic efficiency and capacity retention, indicative of the promising application of \(\mathrm{Na_5SmSi_4O_{12}}\) SE in future large- scale energy storage. + +<--- Page Split ---> + +## Results + +## Synthesis and characterization of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) + +Hexagonal- prismatic crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) was successfully synthesized via two- step solid- state reaction, first at \(800^{\circ}\mathrm{C}\) for \(8\mathrm{h}\) and then \(950^{\circ}\mathrm{C}\) for \(20\mathrm{h}\) , according to the stoichiometric mixtures of \(\mathrm{Na_2CO_3}\) , \(\mathrm{Sm_2O_3}\) and \(\mathrm{SiO_2}\) . As compared in the XRD patterns before and after the second sintering (Supplementary Fig. 1), pure \(\mathrm{Na_5SmSi_4O_{12}}\) can form after the first sintering. And the second sintering helps to achieve a dense ceramic pellet with few interfacial and bulk pores (Supplementary Fig. 2), beneficial to lower the grain boundary resistance and increase the ionic conductivity. Note that the sintering temperature of \(\mathrm{Na_5SmSi_4O_{12}}\) is lower than other oxide SEs (Supplementary Table 1), such as \(\mathrm{Na_5YSi_4O_{12}}\) , \(\mathrm{Na - \beta^{\prime\prime} - Al_2O_3}\) , \(\mathrm{Na_3Zr_2Si_2PO_{12}}\) and its derivatives, etc., good for the cost- and energy- saving. As shown in Fig. 1a, X- ray diffraction (XRD) was taken to identify the structural property of \(\mathrm{Na_5SmSi_4O_{12}}\) , and the Rietveld refined parameters are listed in Supplementary Table 2. All the diffraction peaks can be indexed into a hexagonal system with space group \(R - 3c\) , and the lattice parameters are calculated as \(a = b = 22.14609\mathrm{\AA}\) and \(c = 12.68858\mathrm{\AA}\) . Energy dispersive X- ray spectroscopy (Supplementary Fig. 3) suggests all the elements of \(\mathrm{Na}\) , \(\mathrm{Sm}\) , \(\mathrm{Si}\) and \(\mathrm{O}\) are uniformly dispersed in the as- prepared \(\mathrm{Na_5SmSi_4O_{12}}\) . + +The ionic conductivity of \(\mathrm{Na_5SmSi_4O_{12}}\) was then studied by the alternating current (AC) impedance spectra. The Nyquist plot at room temperature is, as presented in Fig. 1b, composed with a semicircle at high frequency and a tail at low frequency. Via fitting the bulk and grain boundary resistance, the total ionic conductivity of \(\mathrm{Na_5SmSi_4O_{12}}\) is + +<--- Page Split ---> + +calculated as \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) , among the highest values in the existing ceramic electrolytes (Supplementary Table 1). Activation energy \((E_{\mathrm{a}})\) of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) was further evaluated via temperature- dependent AC impedance spectra, as displayed in Fig. 1b. With increasing temperature, the high frequency semicircle gradually vanishes, indicative of enhanced \(\mathrm{Na}^{+}\) transport across the grain boundaries. By fitting the Arrhenius plot (Fig. 1c), \(E_{a}\) is calculated as small as \(0.15 \mathrm{eV}\) , indicative of a rapid ion hoping. In addition, the electronic conductivity of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) was measured as about \(5.8 \times 10^{- 10} \mathrm{~S} \mathrm{~cm}^{- 1}\) via a direct current (DC) polarization measurement (Supplementary Fig. 4). The intrinsic electronic insulation can effectively reduce the self- discharge of batteries and suppress dendrite growth, enabling \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) as a good candidate for sodium based \(\mathrm{ASSB}_{5}^{31}\) . Besides the merits of high ionic and low electronic conductivities, \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) demonstrates superior moisture stability, whose XRD pattern shows no change after soaking in deionized water for 48 h (Supplementary Fig. 5) or exposing to air for 45 days (Supplementary Fig. 6), showing great potential in future industrial application. + +To further reveal the ion conduction mechanism, molecular dynamics simulation was carried out. As shown in Fig. 1d- g, Na ions in \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) can be characterized by their mobility, i.e., the non- mobile ones \(\mathrm{Na}_{1}\) , \(\mathrm{Na}_{2}\) , \(\mathrm{Na}_{3}\) and the mobile ones \(\mathrm{Na}_{4}\) , \(\mathrm{Na}_{5}\) , \(\mathrm{Na}_{6}\) . The conductivity is contributed by a percolating conduction pathway in the mobile region, whereas the rest of Na ions serve as a pillar to hold the structure together. Such a behavior was also proposed in the same family of materials from our previous work16. Interestingly, along the ion conduction pathway, three distinctive sites are + +<--- Page Split ---> + +revealed where Na ions can stably sit in, denoted as sites A, B, and C. These sites are connected to each other via the zigzag like channel as observed from the molecular dynamics simulation trajectory. Furthermore, the diffusion barrier of Na ions between these sites are found to be \(\sim 0.3 \mathrm{eV}\) , which is relatively low in comparison with other reported solid ion conductors32, 33. However, such a value is larger than the activation energy from experiment. We assign such a deviation to the concerted Na hopping behavior. The \(\sim 0.3 \mathrm{eV}\) barrier was calculated by assuming a vacancy-mediated uncorrelated conduction mechanism, while the Na concentration is relatively high and concerted motion is favored. The case of concerted motion is further estimated by assuming a two-ion correlated hopping mechanism, as shown in Fig. 1h. According to such a mechanism, the diffusion barrier along the same path dropped to \(\sim 0.19 \mathrm{eV}\) , close to the experimental value. + +Furthermore, solid- state 23Na nuclear magnetic resonance (NMR) measurements were carried out to reveal the atomic local structure and dynamics mobility34, 35, 36. As exhibited in Fig. 1i, the pristine \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) conductor shows multiple 23Na resonances, which can be further deconvoluted into six types of signals. Via the aid of crystal structure and the proportion of sodium, the peak at 8.8 ppm is assigned to the mobile sodium at Na5 site, the resonance at 4.5 ppm and 1.6 ppm are attributed to Na1 and Na3, the signal at - 16.8 ppm is from Na4, and the rest peaks at - 23.4 ppm and - 28.6 ppm are assigned to sodium at Na2 and Na6 sites, respectively. The simulated details are listed in Supplementary Table 3, from which the proportion of each Na atom agrees with the theoretical results. To obtain more information about \(\mathrm{Na}^{+}\) dynamics properties, + +<--- Page Split ---> + +NMR spectra at different temperatures were performed. As displayed in Supplementary Fig. 7, the \(\mathrm{Na_5SmSi_4O_{12}}\) show similar spectra upon increasing temperature from \(213\mathrm{K}\) to \(367\mathrm{K}\) , except decrease in signal due to Boltzmann distribution. No obvious change is observed for spectral configuration, possibly attributed to the stable structure caused by the non- mobile Na ions at \(\mathrm{Na1}\) , \(\mathrm{Na2}\) , and \(\mathrm{Na3}\) sites. Therefore, we turn to a more temperature sensitive parameter, spin- lattice relaxation (SLR) time \((T_{1})\) . Since the strong quadrupole interaction is observed for \(^{23}\mathrm{Na}\) NMR spectrum in our system, saturation recovery technique is employed to determine \(T_{1}\) values \(^{37}\) . Via fitting the spectral intensity vs. saturation recovery period measured at room- temperature (Fig. 1j), two relaxation time values of 0.5 and \(20\mathrm{ms}\) are obtained for the non- mobile and mobile Na ions, respectively \(^{38}\) . Here, the data analysis is simplified to non- mobile and mobile for convenience. According to Eq. (1), the fitting of the relaxation times as a function of temperature yields an activation energy \(E_{\mathrm{a}} \approx 0.13\mathrm{eV}\) for the mobile \(\mathrm{Na^{+}}\) , which is in good agreement with the results from AC impedance (0.15 eV). + +\[R_{1} = 1 / T_{1}\propto \omega_{0}^{\beta}\mathrm{exp}[-E_{a} / (kT)] \quad (1)\] + +Where \(R_{1}\) is NMR spin- lattice relaxation rate, the reciprocal of \(T_{1}\) , \(\omega_{0}\) is resonance frequency, \(\beta\) is modified exponent, \(E_{\mathrm{a}}\) is activation energy, \(k\) is Boltzmann constant, \(T\) is the absolute temperature in K. Note that the derivate data points marked by hollow circle, as displayed in Fig. 1k, were excluded from the fits since the \(R_{1}\) rates \((1 / T_{1})\) recorded at low temperature range are mainly governed by non- diffusive background effects, such as lattice vibrations or coupling by paramagnetic impurities \(^{39,40,41}\) . + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Crystal structure and sodium-ion conduction characteristic of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . a Rietveld refinement based on the powder XRD. b Nyquist plots of \(\mathrm{Na_5SmSi_4O_{12}}\) from 25 to \(175^{\circ}\mathrm{C}\) . c Arrhenius plot of the conductivity values for \(\mathrm{Na_5SmSi_4O_{12}}\) . d Top-down and e perspective view of crystal structures of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . f-g Molecular dynamics simulation trajectories of \(\mathrm{Na_5SmSi_4O_{12}}\) . h Minimum potential energy path along \(\mathrm{Na^{+}}\) diffusion route in crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . i Solid-state \(^{23}\mathrm{Na}\) NMR spectrum and its simulation for the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . The gray line is experimental data and the green-dashed line is the sum of simulation. j Saturation recovery fitting curve for the data obtained at room temperature. k Temperature dependence of \(^{23}\mathrm{Na}\) NMR relaxation rate as a function of temperature in \(\mathrm{K}^{-1}\) . The solid line is the fit according to Eq. (1). The derivation of the data is not used for the fit.
+ +Electrochemical performance and crystalline-to-amorphous (CTA) transition of \(\mathrm{Na_5SmSi_4O_{12}}\) solid electrolyte. + +<--- Page Split ---> + +To further evaluate the chemical and electrochemical stability of SE against Na metal, symmetric all- solid- state Na cell using \(\mathrm{Na_5SmSi_4O_{12}}\) as the electrolyte was fabricated and tested by the repeatedly galvanostatic stripping and plating at different current densities. The charge/discharge profiles of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) without any interfacial modification were recorded at a current density ranging from 0.05 to \(0.15\mathrm{mAcm}^{- 2}\) with \(120\mathrm{min}\) per cycle (Fig. 2a). The cell displays stable and long cyclic performance which maintains an overvoltage of \(\sim 26\mathrm{mV}\) at \(0.15\mathrm{mAcm}^{- 2}\) with negligible fluctuations over \(800\mathrm{h}\) , indicating sodium dendrite- free plating/stripping and excellent kinetic stability of the \(\mathrm{Na_5SmSi_4O_{12}}\) against Na metal. Afterwards, electrochemical impedance was collected from the symmetric cell after cycling 150, 200 and \(300\mathrm{h}\) to reveal the changes in the internal resistance, as illustrated Fig. 2b and Supplementary Table 4. The resistance of SE (both \(\mathrm{R_b}\) and \(\mathrm{R_{GB}}\) ) remains nearly unchanged as the sodium plating/stripping proceeding. In contrast, the interfacial resistance \(\mathrm{(R_{int})}\) between sodium metal and SE decreases at initial few cycles and then stabilizes at a relatively low value, demonstrating excellent compatibility between \(\mathrm{Na_5SmSi_4O_{12}}\) and sodium metal. This phenomenon is quite different from most of reported SEs without surface modification, whose \(\mathrm{R_{int}}\) increases gradually with increasing cycling times with the result of the excessive internal resistance and failure of the cells due to interfacial side reactions or the formation of holes \(^{25,28,42}\) . + +Scanning electron microscopy (SEM) images of the electrodes at different plating/stripping stages are shown in Fig. 2c and 2d, which demonstrates an interesting tendency of gradually vanishing gap at the interface between Na and \(\mathrm{Na_5SmSi_4O_{12}}\) after + +<--- Page Split ---> + +cycling and an ultraconformal interfacial contact was created. XRD pattern after cycling 200 h suggests a transformation from the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) SE into the amorphous state (Fig. 2e). Such CTA transition starts from the surface and then gradually propagates into the bulk of the SE pellets since there is no clear reflection peaks from the depth- profiled XRD pattern (Supplementary Fig. 8). To further confirm the CTA transition, HRTEM and SAED measurements were undertaken before (Supplementary Fig. 9) and after cycling (Fig. 2f- g). There are no observed lattice fringe and diffraction spot for the cycled SE sample, confirming the electrochemistry- induced CTA transition. Because of the amorphous state, it is difficult to determine the possible local coordination based on the XRD measurement. Therefore, Raman spectra were then used to examine the short- range vibration changes. As displayed in Supplementary Fig. 10, crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) exhibits seven vibration peaks43: the bands in the 900- 1100 cm- 1 region are assigned to Si- O- stretching vibrations, the symmetrical band at \(\sim 624 \mathrm{cm^{- 1}}\) is corresponding to O- Si- O bending mode. The low- frequency bands (<550 cm- 1) are attributed to Sm- O and Na- O bond vibrations in their polyhedral. Though the cycled \(\mathrm{Na_5SmSi_4O_{12}}\) SE loses its long- term ordering (Fig. 2e- g), it maintains nearly all the vibrational peaks that confirms no chemical reaction between the interface of SE and Na metal. While the disappearance of the peak at 1041 cm- 1 might be related to the damage of the fracture of Si- O bond by the amorphous transition. Furthermore, X- ray photoelectron spectroscopy (XPS) suggests that there are no changes in the Sm 3d, Na 1s and Si 2p XPS spectra after cycling, indicative of no redox reaction occurred between sodium metal and \(\mathrm{Na_5SmSi_4O_{12}}\) (Supplementary + +<--- Page Split ---> + +1 Fig. 11). Thus, it can be concluded that \(\mathrm{Na_5SmSi_4O_{12}}\) SE demonstrates an interesting 2 CTA transition with an excellent electrochemical stability that enables it as the ideal 3 SE for sodium based ASSBs. + +![](images/Figure_2.jpg) + +
Fig. 2. Characterization of interfacial evolution and \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling in \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) symmetric cells. a Cycling performance of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) at room temperature. b Nyquist plots of the SE-based symmetric cell after cycling for different times. Cross-sectional SEM images of the Na metal/ \(\mathrm{Na_5SmSi_4O_{12}}\) interfaces: c pristine and d 100 h cycling. e XRD profiles of \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling for different times. f HRTEM and g SAED patterns of \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling.
+ +## Electrochemical performances of solid-state-batteries. + +Critical current density (CCD) and long- term cycling performance of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) with a long single deposition time were further measured to evaluate the capability of amorphous materials and interface in suppressing dendrite growth. As displayed in Fig. 3a, CCD for \(\mathrm{Na}||\mathrm{amorphous Na_5SmSi_4O_{12}}||\mathrm{Na}\) cell is + +<--- Page Split ---> + +achieved as \(1.4\mathrm{mAcm}^{- 2}\) at \(1.4\mathrm{mA hcm}^{- 2}\) , higher than the CCD for \(\mathrm{Na}\|\) crystalline \(\mathrm{Na_5SmSi_4O_{12}||Na}\) (0.4 mA h cm \(^{- 2}\) , Supplementary Fig. 12). This behavior is mainly because the uneven metal deposition leads to rapid growth of sodium dendrites along the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) grain boundary with a large deposition current and hence results in the short circuit rapidly. In addition, as shown in Supplementary Fig. 13a, under a large area capacity, the overpotential increases during the cycling and drops suddenly less than \(25\mathrm{h}\) , which suggests that crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) can be easily penetrated by sodium dendrites. By contrast, if a low area capacity of \(0.05\mathrm{mA hcm}^{- 2}\) is applied in advance to make the electrolyte transition to an amorphous state, the \(\mathrm{Na}\|\) amorphous \(\mathrm{Na_5SmSi_4O_{12}||Na}\) cell displays a stable and long cycling performance which maintains the overvoltage at around \(20\mathrm{mV}\) over \(500\mathrm{h}\) (Supplementary Fig. 13b). All these results indicate the strong capability of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) in suppressing Na dendrite formation. + +To further emphasize the superiority of amorphous interface and bulk materials, a \(\mathrm{Na_3V_2(PO_4)_3}\) (NVP) \(\| \mathrm{Na_5SmSi_4O_{12}||Na}\) SSB was constructed and evaluated at room temperature, as illustrated in the schematic figure (Fig. 3b). The cyclic voltammetry (CV) curve of stainless steel (SS) \(\| \mathrm{Na_5SmSi_4O_{12}||Na}\) cell in Fig. 3c shows that \(\mathrm{Na_5SmSi_4O_{12}}\) possesses a wide electrochemical stability window more than \(5\mathrm{V}\) , which is high enough to ensure that the electrolyte does not undergo phase transition within the working voltage range. As shown in Supplementary Fig. 14a, quite flat charge- discharge voltage profiles at \(2.3 - 3.9\mathrm{V}\) with an initial discharge capacity of \(112\mathrm{mA h}\) \(\mathrm{g}^{- 1}\) were observed, which matches well with the characteristic NVP redox plateaus in + +<--- Page Split ---> + +1 liquid electrolyte (Supplementary Fig. 14b). In addition, the high initial Coulombic 2 efficiency of \(99\%\) indicates that there is no irreversible side reaction between 3 \(\mathrm{Na_5SmSi_4O_{12}}\) and Na anode or NVP cathode. Meanwhile, an excellent cycling 4 performance is achieved with a high- capacity retention of \(95\%\) after 100 cycles (Fig. 5d). Furthermore, specific capacities of \(102\mathrm{mA h g^{- 1}}\) , \(98\mathrm{mA h g^{- 1}}\) and \(93\mathrm{mA h g^{- 1}}\) can 6 be obtained at 0.5, 0.75 and 1 C- rates, respectively, suggestive of a superior rate 7 capability (Fig. 3e). Finally, the long- term cycling stability was estimated at a current 8 rate of 2 C (Fig. 3f). Impressively, there is no obvious capacity loss during the repeated 9 4000 cycles (6 months). The capacity remained to be \(\sim 95\mathrm{mA h g^{- 1}}\) . All these features 10 verify the unique interface properties between \(\mathrm{Na_5SmSi_4O_{12}}\) and sodium, which can not 11 only enable sufficient contact with sodium metal, but also inhibit the side reaction and 12 the dendrite growth during cycle process. + +![](images/Figure_4.jpg) + + +<--- Page Split ---> + +Fig. 3. The electrochemical performance of solid-state batteries. a Potential response of \(\mathrm{Na}||\mathrm{Na}5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) cell during the CCD measurement with a low area capacity of \(0.05\mathrm{mA h cm^{- 2}}\) is applied for \(150\mathrm{h}\) in advance. b Schematic illustration of \(\mathrm{Na}_3\mathrm{V}_2(\mathrm{PO}_4)_3||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) . c CV curve of \(\mathrm{SS}||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) cell at a scanning speed of \(5\mathrm{mV s^{- 1}}\) . d Cycling performance at \(0.5\mathrm{C}\) - rate (1 C corresponds to \(118\mathrm{mA g^{- 1}}\) ). e Rate capability at 0.2, 0.5, 0.75 and 1 C-rate. f Long-term cycle life at \(2\mathrm{C}\) - rate. + +## Discussion + +## Cell stabilization by CTA transition. + +According to the above- mentioned results, the electrochemical- induced CTA transition plays a key role in stabilizing the interfacial properties, suppressing the dendrite formation, and thus increasing the long- term stability and high- rate capability. This superiority of the amorphous stage from the interface to the bulk SE material can be understood in terms of the following two aspects. Firstly, the ionic conductivity of amorphous bulk material was improved. The pathway and the ion transport properties of amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) were investigated in detail by solid- state NMR. Figure 4a shows the \(^{23}\mathrm{Na}\) NMR spectra before and after metallic Na cycling by using crystalline \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) as electrolyte. The \(^{23}\mathrm{Na}\) NMR spectrum did not change significantly before and after cycling, except changes of the mobile ions (Na4, Na5, and Na6), indicative of their slight redistribution upon cycling. Nevertheless, NMR results demonstrate the non- mobile and mobile segments for Na migration within stable structure. In addition, spin- lattice relaxation time \(T_1\) of \(^{23}\mathrm{Na}\) were measured at different temperatures for amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) , as shown in Fig. 4b and Supplementary Fig. 15. The activation energy of the mobile \(\mathrm{Na}^+\) of amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) is calculated as \(0.07\mathrm{eV}\) lower than that of the pristine crystalline state (0.13 eV in Fig. 1k). The decrease in the active energy strongly suggests an enhanced \(\mathrm{Na}^+\) hopping ability for the + +<--- Page Split ---> + +amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) with higher ionic conductivity. Secondly, the interfacial issues, such as high interfacial resistance and metal dendrite growth, are strongly alleviated. The schematic illustration of sodium deposition is provided in Fig. 4d. A locally solid- solid contact induces a heterogeneous sodium- ion flux, whereas a tight interfacial contact of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) lead to uniform deposition of sodium. As mentioned above in Fig. 2b, there is obvious decrease in the \(\mathrm{R_{int}}\) upon cycling at different plating/stripping stages. The interesting phenomenon can be attributed to ultraconformal interfacial contact between Na metal and electrolyte due to the increased contact area (Fig. 2c- d). Furthermore, the isotropic surface of amorphous material could enhance the wettability with significantly reduced interface resistance and eliminates the influence of crystallographic orientation- dependent ionic transport since the interfacial energy of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) is calculated as \(0.33\mathrm{Jm}^{- 2}\) with sodium, lower than the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) (0.56 J \(\mathrm{m}^{- 2}\) ) (Supplementary Fig. 16). In addition, the amorphous- \(\mathrm{Na_5SmSi_4O_{12}}\) deliver the high mechanical strength, beneficial to inhibiting the dendrite growth. As shown in Fig. 4c, nanoindentation technique was employed to evaluate the Young's modulus \(E\) and hardness \(H^{44,45}\) . As for the amorphous sample, the \(E\) and \(H\) are calculated to be \(\sim 79.9\) GPa and \(\sim 3.8\) GPa, respectively, higher than the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) ( \(\sim 72.6\) GPa and \(\sim 2.8\) GPa). In summary, the intrinsic microstructural and compositional homogeneity, as well as low electronic conductivity, alleviate the potential fluctuations at local positions in the SE and suppress sodium propagation and penetration into amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) . + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 4. The superiority of amorphous bulk materials and interface. a Solid-state \(^{23}\mathrm{Na}\) NMR of the pristine crystalline and the cycled amorphous \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . b \(^{23}\mathrm{Na}\) NMR relaxation rate of Na cycled \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) as a function of temperature in \(\mathrm{K}^{-1}\) . c Nanoindentation load-displacement curves of crystalline \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) and amorphous \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . d Schematic of interface morphology evolution during sodium plating/stripping.
+ +## Mechanism of CTA transition. + +Electrochemically- induced solid- state amorphization (SSA) transformations are popular in the alloying type anode, like the transition from nano Si into amorphous Li- Si phases after electrochemical lithiation46, 47. Nevertheless, this kind of SSA process usually undergoes a chemical phase transition, which is disastrous for SEs. To the best of our knowledge, it is the first observation about the electrochemically induced CTA transformation in oxide SEs during alkali metal plating/stripping without chemical phase transition. Generally, the SSA transformation depends on the thermodynamic driving force (pressure, defects, internal stress, etc.), and the existence of kinetic constraints (mainly because of the low experimental temperature) that prevent the + +<--- Page Split ---> + +formation of full equilibrium crystalline phases48, 49, 50. For example, when ion conduction is non- homogeneous, the local variance may lead to large local stress, which could in turn lead to the collapse of the structure. Through cycling, SSA transformation may occur. In this case, we calculate the formation energies of the crystalline and the amorphous phases sampled from quenched melts. As shown in Supplementary Fig. 17, the crystalline phases are thermodynamically favored which an energy \(\sim 60 \mathrm{meV}\) atom1 lower than the amorphous structures. This indicates that kinetics may play a critical role in the SSA. One way to magnify such kinetic driving force is to apply large lattice strain on the system. In this context, we computationally substitute the Na ions to Li ions in the structures and further compute the energy difference between the crystalline and the amorphous phases. Interestingly, when the Li concentration reaches \(\sim 2 / 5\) , the energy difference significantly drops to \(\sim 10 \mathrm{meV}\) atom- 1. This further supports our assumption that the driving force led by the large lattice strain could be the origin of the SSA. + +Following the above results, a Li- Na exchange process is captured by atomistic simulations, since the ionic radius of \(\mathrm{Li^{+}}\) (76 pm) is smaller than \(\mathrm{Na^{+}}\) (102 pm) that could induce much larger lattice strain. Large- scale molecular dynamics is run on the hypothetical \(\mathrm{Na_5SmSi_4O_{12} / Li_5SmSi_4O_{12}}\) using a machine- learned forcefield, as shown in Fig. 5a. Interestingly, the Li ions first exchanges with the mobile Na ions at the reaction front followed by mixing with the non- mobile ones, see Fig. 5b- e. When the pillar Na ions are replaced by smaller Li ions, the structure start to collapse. During the initial exchange process, crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) and \(\mathrm{Na_5 - xLi_xSmSi_4O_{12}}\) could co- exist. + +<--- Page Split ---> + +We computationally evaluated how Li substitution affects the ion diffusion. As shown in Fig. 5f, when all mobile Na ions are replaced by Li, Li diffusion barrier along the original zig-zag route dropped from \(\sim 0.3 \mathrm{eV}\) to \(\sim 0.25 \mathrm{eV}\) , indicating a faster diffusion kinetics, which may in turn, result in faster SSA during Li exchange. In conclusion, lattice strain induced by ion intercalation is the main driving force of amorphous transformation. Once the thermodynamic driving force is present, the kinetic hindrance at room- temperature prevents the transition back to amorphous \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) . + +Guided by the computational results, a hybrid symmetric cell of \(\mathrm{Li} \| \mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12} \| \mathrm{Li}\) was fabricated to accelerate this SSA transition. As shown in Supplementary Fig. 18, the cell reveals uniform plating and stripping overpotential profiles with an increased current density. This phenomenon suggests that hybrid movement of \(\mathrm{Li}^{+} / \mathrm{Na}^{+}\) within the bulk of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) and an effective plating/stripping of \(\mathrm{Na}^{+}\) at Li anode. As expected, \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) exhibits a much shorter amorphous time within \(100 \mathrm{~h}\) (Fig. 5g and Supplementary Fig. 19), confirming that the CTA transition of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) is mainly triggered by the lattice strain and speeded up because of the mismatch between \(\mathrm{Li}^{+}\) and \(\mathrm{Na}^{+}\) ionic radius. Furthermore, an obvious reflections shift and weakening can be observed in the XRD patterns after experiencing different cycling time (Supplementary Fig. 20). This shift indicates the existence of microscopic strain in the lattice, and the gradual accumulation of stress leads to the break of more bonds. Thus, with the increase of cycling time, the crystallinity of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) weakens, and the intensity of XRD peaks decreases gradually. To assess the cationic electrochemical exchange mechanism, XPS analysis of SE operating in the hybrid cell + +<--- Page Split ---> + +was carried out after cycling. As shown in Supplementary Fig. 21, the decrease in Na 1s peak and appearance of Li 1s peak prove that \(\mathrm{Li^{+}}\) ions can successfully replace part of \(\mathrm{Na^{+}}\) ions and the strong \(\mathrm{Li^{+}}\) mobility within hexagonal-prismatic \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . In addition, there is no observed new Raman peaks after cycling in the hybrid symmetric cell (Supplementary Fig. 22), which indicates that \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) SE demonstrates an excellent thermodynamic stability with Li metal. + +Figure 5h shows the \(^{23}\mathrm{Na}\) spectra of \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) cycled with Li metal under different times, from which the stripped Li will exchange with Na on its way when across the electrolyte. The signals at - 16.8 ppm - 23.4 ppm, being assigned to sodium at \(\mathrm{Na4}\) and \(\mathrm{Na6}\) sites, significantly weakened, reflecting the transport activity of \(\mathrm{Na4}\) and \(\mathrm{Na6}\) sodium sites during polarization. In addition, the signal at 8.8 ppm, which is assigned to sodium at \(\mathrm{Na5}\) site, is weakened but fluctuates in intensity at different cycle times, indicating it likely serves as 'bridge' for transporting \(\mathrm{Na / Li}\) ions during polarization. These results further conclude that there is a possible 3D pathway of \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) between \(\mathrm{Na4 - Na5 - Na6}\) . Figure 5i and Supplementary Fig. 23 display the \(^{7}\mathrm{Li}\) NMR spectrum of \(\mathrm{Na_{5 - x}Li_{x}SmSi_{4}O_{12}}\) after Li cycled different times (100 and 200 h). The \(^{7}\mathrm{Li}\) NMR spectrum of the electrolyte cycled for 200 h is broader than that for 100 h, indicating the growth of amorphous phase. \(^{6}\mathrm{Li}\) NMR spectrum after Li cycling is shown in Supplementary Fig. 24. Both \(^{7}\mathrm{Li}\) and \(^{6}\mathrm{Li}\) NMR spectra present two components, corresponding to the different mobile Na sites, such as \(\mathrm{Na4}\) and \(\mathrm{Na6}\) . + +<--- Page Split ---> +![PLACEHOLDER_23_0] + +
Fig. 5. CTA mechanism. a Structure of the \(\mathrm{Na_5SmSi_4O_{12}||Li_5SmSi_4O_{12}}\) interface model b molecular dynamics trajectories of the interface model. Molecular dynamics simulation trajectories of crystalline c \(\mathrm{Li_5SmSi_4O_{12}dNa_5SmSi_4O_{12}}\) and e \(\mathrm{Li_5Na_2SmSi_4O_{12}}\) . f Minimum potential energy path along Li diffusion route in crystalline \(\mathrm{Na_2Li_5SmSi_4O_{12}}\) . g XRD profiles of \(\mathrm{Na_5 - xLi_5SmSi_4O_{12}}\) after cycling for different times. Solid-state h \(^{23}\mathrm{Na}\) and i \(^{7}\mathrm{Li}\) NMR of the \(\mathrm{Na_5SmSi_4O_{12}}\) with different cycling times.
+ +In summary, dense crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) was prepared and exhibits a high room- temperature conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) . Driven by microscopic strain in the lattice, the \(\mathrm{Na_5SmSi_4O_{12}}\) undergoes an amorphous transformation during the cycling in both symmetric \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) and \(\mathrm{Li}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Ci}\) cells. On the one hand, the increased contact area greatly reduces the interfacial resistance between sodium metal and electrolyte and promotes the homogeneous deposition of sodium. On the other, the isotropic ionic transport feature of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) eliminates the ion- blocking crystallographic orientation, uniform distribution of the current and homogeneous metal nucleation at the anode interface will be promoted. Thus, the + +<--- Page Split ---> + +1 sodium symmetrical cells manifest stable cycling performance for \(800\mathrm{h}\) at \(0.15\mathrm{mA}\) 2 \(\mathrm{cm}^{- 2}@1\mathrm{h}\) and \(500\mathrm{h}\) at \(0.05\mathrm{mA}\mathrm{cm}^{- 2}@5\mathrm{h}\) ( \(25^{\circ}\mathrm{C}\) ). Furthermore, the successful 3 operation of \(\mathrm{Na_3V_2(PO_4)_3\|Na_5SmSi_4O_{12}\|Na}\) solid- state sodium batteries with excellent 4 electrochemical performance further implies the superiority of \(\mathrm{Na_5SmSi_4O_{12}}\) electrolyte. 5 + +## Methods + +## Materials synthesis: + +The \(\mathrm{Na_5SmSi_4O_{12}}\) pellets were synthesized by a solid- state sintering method using \(\mathrm{Na_2CO_3}\) , \(\mathrm{Sm_2O_3}\) and \(\mathrm{SiO_2}\) as the starting materials. First, the raw materials of analytical grade were mixed by ball- milling at a milling speed constant of \(600\mathrm{rpm}\) for \(15\mathrm{h}\) . The mixture was dried at \(80^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) and calcined at \(800^{\circ}\mathrm{C}\) for \(8\mathrm{h}\) . Then the powder was put into a cylindrical pressing mold with diameter of \(15\mathrm{mm}\) and pressed under a pressure of \(300\mathrm{MPa}\) . The pressed pellets were then sintered at \(950^{\circ}\mathrm{C}\) for \(20\mathrm{h}\) . Finally, buff pellets were obtained after sintering. + +The \(\mathrm{Na_3V_2(PO_4)_3}\) (NVP) cathode material was prepared by the sol- gel method according to our previous work (Supplementary Fig. 25)51. Firstly, the stoichiometric amount of \(\mathrm{Na_2CO_3}\) , \(\mathrm{NH_4VO_3}\) and \(\mathrm{NH_4H_2PO_4}\) with a molar ratio of \(3:4:6\) was dissolved in deionized water. Secondly, \(0.02\mathrm{M}\) aqueous citric acid [HOC(COOH)(CH2COOH)2] solution was added dropwise into the solution until the ratio of vanadium: citric acid equals to 2: 1. Then the gel could be acquired by drying the precursor in an oven at \(120^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) . Finally, the NVP/C powder can be acquired after heat treatments in two steps, first at \(350^{\circ}\mathrm{C}\) for \(5\mathrm{h}\) and then at \(750^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) under a nitrogen atmosphere. + +<--- Page Split ---> + +## Characterization method: + +X- ray diffraction (XRD) patterns were recorded by a Bruker D8 Advance diffractometer with Cu Kα radiation and RigaKu D/max- 2550 diffractometer (1.6 kW, Cu Kα radiation, \(\lambda = 1.5406 \AA\) ), followed by Rietveld refinement using Fullprof software for the crystal structure analysis. The microscopy characteristics of the samples were investigated by Hitachi Regulus8100 FESEM, high resolution transmission electron microscope (HRTEM, talos F200X) and selected area electron diffraction (SAED). The elemental mapping was used to analyze the element distribution of the samples. X- ray photoemission spectrum (XPS) was carried out on a thermo scientific NEXSA spectrometer. Raman spectra were examined using a Renishaw Raman microscope (model 2000) with Ar- ion laser excitation. Prior to analysis the interface properties between Na/Li metal and Na5SmSi4O12 electrolyte after cycling, emery paper was employed to remove any residual metal on the surface. All \(^{6}\mathrm{Li}\) , \(^{7}\mathrm{Li}\) and \(^{23}\mathrm{Na}\) magic angle spinning (MAS) NMR experiments were acquired on Bruker 400 MHz (9.4 T) magnets with AVANCE NEO consoles using Bruker 3.2 mm HXY MAS probe. The samples were filled into rotors inside Argon glove box. The Larmor frequencies for \(^{6}\mathrm{Li}\) , \(^{7}\mathrm{Li}\) and \(^{23}\mathrm{Na}\) were 58.89, 155.53 and 105.86 MHz, respectively. All spectra were acquired by using one- pulse program and were referenced to 1 M LiCl ( \(^{6}\mathrm{Li}\) and \(^{7}\mathrm{Li}\) ) and 1 M NaCl ( \(^{23}\mathrm{Na}\) ) solutions with chemical shifts at 0 ppm. The spinning rate \(\nu_{\mathrm{rot}}\) was set to 14 kHz. \(^{23}\mathrm{Na}\) spin- lattice relaxation times (\(T_{1}\)) were recorded by using the saturation recovery pulse sequence. The varying temperature experiments were protected by \(\mathrm{N}_{2}\) atmosphere. Nanoindentation + +<--- Page Split ---> + +measurement was taken on a nanoindentation tester (Agilent Nano Indenter G200) equipped with a three- sided pyramidal Berkovich diamond indenter. The applied standard loading, holding, and unloading times were 10, 5, and 10 s, respectively. During the testing, the load- displacement curves up to pellet cracking were recorded and utilized to calculate the Young's modulus \(E\) and hardness \(H\) using the Oliver- Pharr method. Indentations with maximum indentation load of 1 mN are conducted on the surface of SE pellets. The reduced modulus \(E_{\mathrm{r}}\) was determined by the unloading stiffness and projected contact area. By assuming a Poisson's ratio of 0.3 for samples and 0.07 for single crystalline diamond, their Young's moduli were estimated. + +## Impedance spectrum test: + +For the conductivity measurement, silver was spread on both sides of the ceramic pellets as blocking electrodes. AC impedance spectra were recorded using a Solartron 1260 impedance analyzer over a frequency range of 5 MHz to 1 Hz, with an applied root mean square AC voltage of 30 mV. The temperature dependence of the conductivity was measured in the same way at several specific temperatures ranging from 25 to 175 °C. For conductivity test at each temperature, the samples were allowed to equilibrate for 2 h prior to measurements. The resistances of the \(\mathrm{Na}||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) symmetric cells were tested under the same conditions. + +## Electrochemical measurement: + +To obtain NVP cathode, NVP active material (70 wt.%), Super P conductive additive (20 wt.%) and carboxymethyl cellulose (CMC) binder (10 wt.%) were dissolved in water to form a homogeneous slurry, and then uniformly coated onto an aluminum foil + +<--- Page Split ---> + +current collector. After drying for \(12\mathrm{h}\) at \(80^{\circ}\mathrm{C}\) , the electrode was punched into \(1\mathrm{cm}\) diameter wafers for use with the loading mass of \(1.0\mathrm{- }1.5\mathrm{mgcm}^{- 2}\) . Sodium foil was employed as anode. The \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) pellet was used as both separator and electrolyte. \(20\mu \mathrm{L}1\mathrm{MNaClO_4}\) in ethylene carbonate (EC) and propylene carbonate (PC) (1:1 v/v) with the addition of 5 vol.% fluoroethylene carbonate (FEC) was added as the interfacial wetting agent at the cathode side. The 2032- type coin cells were assembled in an argon- filled glovebox. Galvanostatic charge- discharge tests were performed in a cutoff potential window of 2.3- 3.9 V by using Land- 2100 automatic battery tester. All- solid- state \(\mathrm{Na\|Na_{5}SmSi_{4}O_{12}\|Na}\) and \(\mathrm{Li\|Na_{5}SmSi_{4}O_{12}\|Li}\) symmetric cells were assembled to test the sodium/lithium metal stripping/plating at \(25^{\circ}\mathrm{C}\) and \(50^{\circ}\mathrm{C}\) , respectively. In addition, electrochemical stable window of electrolytes was examined by cyclic voltammetry (CV) measurement with the stainless steel \(\mathrm{||Na_{5}SmSi_{4}O_{12}||Na}\) cell in the voltage range of - 1 to \(8\mathrm{V}\) at a scanning rate of \(5\mathrm{mV}\mathrm{s}^{- 1}\) . The direct current (DC) polarization measurement was performed on \(\mathrm{Ag\|Na_{5}SmSi_{4}O_{12}\|Ag}\) cell with a \(300\mathrm{mV}\) potential and the current response was measured for \(100\mathrm{min}\) at ambient temperature. The CV measurement and the DC measurement were performed using a Bio- Logic electrochemical workstation. + +## Computational methods: + +Density functional theory calculations were carried out using the VASP6.3 \(^{52,53}\) package following the setup we used in previous work \(^{54,55,56}\) . Briefly, the PBE exchange- correlation functional was adopted with a planewave basis and a cutoff energy of 520 eV. The reciprocal space was sampled using Monkhorst- Pack grids with a spacing of + +<--- Page Split ---> + +0.04 Å⁻¹. The convergence for electron self- consistent computations and structural optimizations are set to 10⁻⁴ eV atom⁻¹ and 10⁻³ eV atom⁻¹, respectively. Due to the need of large- scale molecular dynamics simulations, we trained a machine learning forcefield based on ab initio molecular dynamics simulations (AIMD)⁵⁴,⁵⁵. Trajectories from smaller systems with 5 compositions sampled evenly between Na₅SmSi₄O₁₂ and Li₅SmSi₄O₁₂ were collected at high temperatures together with their relaxation trajectories. The energy and forces were used as the label to train the model. The production MD simulations were carried out using such a forcefield. The systems were equilibrated at 600 K for 100 ps and gradually dropped to 300 K with a period of another 100 ps under an NPT thermostat at ambient pressure. Then an NVT production run was carried out at 300 K for 200 ps. The time step was 1 fs. For the interface model, we adopted a universal machine learning model which can capture the ground state structures and their energy. Two interface models were built to mimic the interaction between the Li electrode and the crystalline and amorphous SE. The models were first equilibrated at 500 K for 1000 fs followed by energy minimization. The interfacial energy was calculated by subtracting the energy of the interfaces from the sum of energies of independent bulk phases. + +## References + +1. Deysher G, et al. Transport and mechanical aspects of all-solid-state lithium batteries. Mater. Today Phys. 24, 100679 (2022). +2. Wu Y, Wang S, Li H, Chen L, Wu F. Progress in thermal stability of all-solid-state-Li-ion-batteries. InfoMat 3, 827-853 (2021). + +<--- Page Split ---> + +3. Feng X, et al. Review of modification strategies in emerging inorganic solidstate electrolytes for lithium, sodium, and potassium batteries. Joule 6, 543-587 (2022). +4. Tian Y, et al. Promises and Challenges of Next-Generation "Beyond Li-ion" Batteries for Electric Vehicles and Grid Decarbonization. Chem. Rev. 121, 1623-1669 (2021). +5. Oh JAS, He L, Chua B, Zeng K, Lu L. Inorganic sodium solid-state electrolyte and interface with sodium metal for room-temperature metal solid-state batteries. Energy Storage Mater. 34, 28-44 (2021). +6. Tang B, Jaschin PW, Li X, Bo S-H, Zhou Z. Critical interface between inorganic solid-state electrolyte and sodium metal. Mater. Today 41, 200-218 (2020). +7. Wu JF, et al. Inorganic Solid Electrolytes for All-Solid-State Sodium Batteries: Fundamentals and Strategies for Battery Optimization. Adv. Funct. Mater. 31, 2008165 (2020). +8. Feng X, et al. Heavily Tungsten-Doped Sodium Thioantimonate Solid-State Electrolytes with Exceptionally Low Activation Energy for Ionic Diffusion. Angew. Chem. Int. Ed. Engl. 60, 26158-26166 (2021). +9. Hayashi A, et al. A sodium-ion sulfide solid electrolyte with unprecedented conductivity at room temperature. Nat. Commun. 10, 5266 (2019). +10. Wu EA, et al. A stable cathode-solid electrolyte composite for high-voltage, long-cycle-life solid-state sodium-ion batteries. Nat. Commun. 12, 1256 (2021). +11. Zhu Y, Mo Y. Materials Design Principles for Air-Stable Lithium/Sodium Solid + +<--- Page Split ---> + +Electrolytes. Angew. Chem. Int. Ed. Engl. 59, 17472- 17476 (2020). 12. Sun F, et al. Origin of High Ionic Conductivity of Sc- Doped Sodium- Rich NASICON Solid- State Electrolytes. Adv. Funct. Mater. 31, 2102129 (2021). 13. Yang J, et al. Ultrastable All- Solid- State Sodium Rechargeable Batteries. ACS Energy Lett. 5, 2835- 2841 (2020). 14. Zou Z, et al. Relationships Between \(\mathrm{Na + }\) Distribution, Concerted Migration, and Diffusion Properties in Rhombohedral NASICON. Adv. Energy Mater. 10, 2001486 (2020). 15. Lu Y, Li L, Zhang Q, Niu Z, Chen J. Electrolyte and Interface Engineering for Solid- State Sodium Batteries. Joule 2, 1747- 1770 (2018). 16. Sun G, et al. \(\mathrm{Na_5YSi_4O_{12}}\) : A sodium superionic conductor for ultrastable quasi- solid- state sodium- ion batteries. Energy Storage Mater. 41, 196- 202 (2021). 17. Schilm J, et al. Influence of \(\mathrm{R = Y}\) , Gd, Sm on Crystallization and Sodium Ion Conductivity of \(\mathrm{Na_5RSi_4O_{12}}\) Phase. Mater. (Basel) 15, (2022). 18. Shannon RD, Taylor BE, Gier TE, Chen HY, Berzins T. Ionic conductivity in sodium yttrium silicon oxide \(\mathrm{(Na_5YSi_4O_{12})}\) - type silicates. Inorg. Chem. 17, 958- 964 (2002). 19. Zhou W, Li Y, Xin S, Goodenough JB. Rechargeable Sodium All- Solid- State Battery. ACS Cent. Sci. 3, 52- 57 (2017). 20. Zahiri B, et al. Revealing the role of the cathode- electrolyte interface on solid- state batteries. Nat. Mater. 20, 1392- 1400 (2021). 21. Wang X, Chen J, Wang D, Mao Z. Improving the alkali metal + +<--- Page Split ---> + +electrode/inorganic solid electrolyte contact via room- temperature ultrasound solid welding. Nat. Commun. 12, 7109 (2021). + +Oh JAS, Sun J, Goh M, Chua B, Zeng K, Lu L. A Robust Solid- Solid Interface Using Sodium- Tin Alloy Modified Metallic Sodium Anode Paving Way for All- Solid- State Battery. Adv. Energy Mater. 11, 2101228 (2021). + +Kasemchainan J, et al. Critical stripping current leads to dendrite formation on plating in lithium anode solid electrolyte cells. Nat. Mater. 18, 1105- 1111 (2019). + +Ji X, et al. Solid- State Electrolyte Design for Lithium Dendrite Suppression. Adv. Mater. 32, e2002741 (2020). + +Wang S, Xu H, Li W, Dolocan A, Manthiram A. Interfacial Chemistry in Solid- State Batteries: Formation of Interphase and Its Consequences. J. Am. Chem. Soc. 140, 250- 257 (2018). + +Gao Z, et al. TiO2 as Second Phase in \(\mathrm{Na_3Zr_2Si_2PO_{12}}\) to Suppress Dendrite Growth in Sodium Metal Solid- State Batteries. Adv. Energy Mater. 12, 2103607 (2022). + +Wang X, Chen J, Mao Z, Wang D. In situ construction of a stable interface induced by the \(\mathrm{SnS_2}\) ultra- thin layer for dendrite restriction in a solid- state sodium metal battery. J. Mater. Chem. A 9, 16039- 16045 (2021). + +Wang C, et al. Grain Boundary Design of Solid Electrolyte Actualizing Stable All- Solid- State Sodium Batteries. Small 17, e2103819 (2021). + +Yu X, Manthiram A. Sodium- Sulfur Batteries with a Polymer- Coated NASICON- type Sodium- Ion Solid Electrolyte. Matter 1, 439- 451 (2019). + +<--- Page Split ---> + +30. Swift MW, Qi Y. First-Principles Prediction of Potentials and Space-Charge Layers in All-Solid-State Batteries. Phys. Rev. Lett. 122, 167701 (2019). +31. Han F, et al. High electronic conductivity as the origin of lithium dendrite formation within solid electrolytes. Nat. Energy 4, 187-196 (2019). +32. He X, Bai Q, Liu Y, Nolan AM, Ling C, Mo Y. Crystal Structural Framework of Lithium Super-Ionic Conductors. Adv. Energy Mater. 9, 1902078 (2019). +33. Zhang Z, et al. Correlated Migration Invokes Higher \(\mathrm{Na^{+}}\) -Ion Conductivity in NaSICON-Type Solid Electrolytes. Adv. Energy Mater. 9, 1902373 (2019). +34. Zheng J, Tang M, Hu YY. Lithium Ion Pathway within \(\mathrm{Li_7La_3Zr_2O_{12}}\) -Polyethylene Oxide Composite Electrolytes. Angew. Chem. Int. Ed. Engl. 55, 12538-12542 (2016). +35. Hui K, et al. Yolk-shell nanoarchitecture for stabilizing a \(\mathrm{Ce_2S_3}\) anode. Carbon Energy 3, 709-720 (2021). +36. Liu J, et al. Regulating Hybrid Anodes for Efficient \(\mathrm{Li^{+} / Na^{+}}\) Storage. ACS Mater. Lett. 4, 1411-1421 (2022). +37. Zhou Q, et al. NASICON \(\mathrm{Li_{1.2}Mg_{0.1}Zr_{1.9}(PO_4)_3}\) Solid Electrolyte for an All-Solid-State Li-Metal Battery. Small Methods 4, 2000764 (2020). +38. Wang D, et al. Toward Understanding the Lithium Transport Mechanism in Garnet-type Solid Electrolytes: \(\mathrm{Li^{+}}\) Ion Exchanges and Their Mobility at Octahedral/Tetrahedral Sites. Chem. Mater. 27, 6650-6659 (2015). +39. Buschmann H, et al. Structure and dynamics of the fast lithium ion conductor " \(\mathrm{Li_7La_3Zr_2O_{12}}\) ". Phys. Chem. Chem. Phys. 13, 19378-19392 (2011). + +<--- Page Split ---> + +40. Kuhn A, Narayanan S, Spencer L, Goward G, Thangadurai V, Wilkening M. Li self-diffusion in garnet-type \(\mathrm{Li_7La_3Zr_2O_{12}}\) as probed directly by diffusion-induced Li7 spin-lattice relaxation NMR spectroscopy. Phys. Rev. B 83, (2011). +41. Düvel A, Kuhn A, Robben L, Wilkening M, Heitjans P. Mechanosynthesis of Solid Electrolytes: Preparation, Characterization, and Li Ion Transport Properties of Garnet-Type Al-Doped \(\mathrm{Li_7La_3Zr_2O_{12}}\) Crystallizing with Cubic Symmetry. J. Phys. Chem. C 116, 15192-15202 (2012). +42. Chung H, Kang B. Mechanical and Thermal Failure Induced by Contact between a \(\mathrm{Li_{1.5}Al_{0.5}Ge_{1.5}(PO_4)_3}\) Solid Electrolyte and Li Metal in an All Solid-State Li Cell. Chem. Mater. 29, 8611-8619 (2017). +43. Ahmadzadeh M, Olds TA, Scrimshire A, Bingham PA, McCloy JS. Structure and properties of \(\mathrm{Na_5FeSi_4O_{12}}\) crystallized from \(\mathrm{5Na_2O-Fe_2O_3-8SiO_2}\) glass. Acta Crystallogr., Sect. C: Struct. Chem. 74, 1595-1602 (2018). +44. Xu R, Sun H, de Vasconcelos LS, Zhao K. Mechanical and Structural Degradation of \(\mathrm{LiNi_5Mn_7Co_2O_2}\) Cathode in Li-Ion Batteries: An Experimental Study. J. Electrochem. Soc. 164, A3333 (2017). +45. Chi X, et al. An electrochemically stable homogeneous glassy electrolyte formed at room temperature for all-solid-state sodium batteries. Nat. Commun. 13, 2854 (2022). +46. Limthongkul P, Jang Y-I, Dudney NJ, Chiang Y-M. Electrochemically-driven solid-state amorphization in lithium-metal anodes. J. Power Sources 119-121, + +<--- Page Split ---> + +1 604- 609 (2003). 2 47. Tang Y, Gao Y, Liu L, Zhang Y, Xie J, Zeng X. Electrochemically driven 3 amorphization of (Li)- Ti- P- O nanoparticles embedded in porous CNTs for 4 superior lithium storage performance. Chem. Eng. J. 409, 127373 (2021). 5 48. Johnson WL. Crystal- to- glass transformation in metallic materials. Mater. Sci. 6 Eng. 97, 1- 13 (1988). 7 49. Fecht HJ. Defect- induced melting and solid- state amorphization. Nature 356, 8 133- 135 (1992). 9 50. Lyapin AG, Brazhkin VV. Pressure- induced lattice instability and solid- state 10 amorphization. Phys. Rev. B 54, 12036- 12048 (1996). 11 51. Wang D, et al. \(\mathrm{Na_3V_2(PO_4)_3 / C}\) composite as the intercalation- type anode 12 material for sodium- ion batteries with superior rate capability and long- cycle 13 life. J. Mater. Chem. A 3, 8636- 8642 (2015). 14 52. Kresse G, Hafner J. Ab initio molecular dynamics for liquid metals. Phy. Rev. 15 B 47, 558- 561 (1993). 16 53. Kresse G, Hafner J. Ab initio molecular- dynamics simulation of the liquid- 17 metal- - amorphous- semiconductor transition in germanium. Phy. Rev. B 49, 18 14251- 14269 (1994). 19 54. Hou Z, Zhou R, Min Z, Lu Z, Zhang B. Realizing Wide- Temperature Reversible 20 Ca Metal Anodes through a \(\mathrm{Ca^{2 + }}\) - Conducting Artificial Layer. ACS Energy Lett. 21 274- 279 (2022). 22 55. Hou Z, et al. Correlation between Electrolyte Chemistry and Solid Electrolyte + +<--- Page Split ---> + +1 Interphase for Reversible Ca Metal Anodes. Angew. Chem. Int. Ed. 61, 2 e202214796 (2022). 3 56. Lu Z, Ciucci F. Metal Borohydrides as Electrolytes for Solid-State Li, Na, Mg, 4 and Ca Batteries: A First-Principles Study. Chem. Mater. 29, 9308- 9319 (2017). 5 Acknowledgments 6 This work was supported by the National Natural Science Foundation of China with 7 Grant No. 12274176, 51972142, and 21974007. We also would like to thank the support 8 from the Department of Science and Technology of Jilin Province with Grant No. 9 20220201118GX and 20210301021GX. 10 Author contributions 11 G. S. and C. L. contributed equally to this work. F. D., M. T. and Z. L. designed and 12 supervised the project. G. S. performed materials synthesis, electrochemical tests, and 13 wrote the manuscript. C. L. and M. T. performed NMR experiments and analyses. Z. L. 14 performed and discussed DFT calculation results. Z. W., S. Y., G. C., and Z. S. revised 15 the manuscript. All the authors participated in the discussion and provided constructive 16 advice for the experimental design. 17 Competing interests 18 All other authors declare they have no competing interests. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- supplementarymaterials.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5_det.mmd b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f49537919ebd269c415fdfbeb381ef03d5ff2090 --- /dev/null +++ b/preprint/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5/preprint__0e87b1e2631222ab8fd748792a389f1e1b25ccf5d46ae6423d387890919518d5_det.mmd @@ -0,0 +1,373 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 936, 241]]<|/det|> +# Electrochemically induced crystalline-to-amorphization transformation in sodium samarium silicate solid electrolyte for long-lasting all-solid-state sodium metal batteries + +<|ref|>text<|/ref|><|det|>[[44, 263, 925, 325]]<|/det|> +Ge Sun Key Laboratory of Physics and Technology for Advanced Batteries (Ministry of Education), State Key Laboratory of Superhard Materials, College of Physics, Jilin University + +<|ref|>text<|/ref|><|det|>[[44, 330, 660, 372]]<|/det|> +Chenjie Lou Center for High Pressure Science and Technology Advanced Research + +<|ref|>text<|/ref|><|det|>[[44, 377, 180, 416]]<|/det|> +Zhixuan Wei Jilin University + +<|ref|>text<|/ref|><|det|>[[44, 422, 184, 462]]<|/det|> +Shiyu Yao Jilin University + +<|ref|>text<|/ref|><|det|>[[44, 468, 268, 510]]<|/det|> +Ziheng Lu University of Cambridge + +<|ref|>text<|/ref|><|det|>[[44, 515, 184, 555]]<|/det|> +Gang Chen Jilin University + +<|ref|>text<|/ref|><|det|>[[44, 561, 710, 603]]<|/det|> +Ze Shen Nanyang Technological University https://orcid.org/0000- 0001- 7432- 7936 + +<|ref|>text<|/ref|><|det|>[[44, 608, 640, 650]]<|/det|> +Mingxue Tang Center for High Pressure Science & Technology Advanced Research + +<|ref|>text<|/ref|><|det|>[[44, 655, 544, 697]]<|/det|> +Fei Du ( \(\circleddash\) dafei@jlu.edu.cn) Jilin University https://orcid.org/0000- 0001- 6413- 0689 + +<|ref|>text<|/ref|><|det|>[[44, 740, 101, 757]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 777, 135, 795]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 815, 303, 834]]<|/det|> +Posted Date: March 7th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 853, 474, 872]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2623650/v1 + +<|ref|>text<|/ref|><|det|>[[44, 890, 910, 933]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 100, 936, 142]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on October 16th, 2023. See the published version at https://doi.org/10.1038/s41467-023-42308-0. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 95, 855, 699]]<|/det|> +Electrochemically induced crystalline-to-amorphization transformation in sodium samarium silicate solid electrolyte for long-lasting all-solid-state sodium metal batteriesGe Sun1,4, Chenjie Lou2,4, Zhixuan Wei1, Shiyu Yao1, Ziheng Lu3\*, Gang Chen1, Zexiang Shen1, Mingxue Tang2\*, Fei Du1\*1Key Laboratory of Physics and Technology for Advanced Batteries (Ministry of Education), State Key Laboratory of Superhard Materials, College of Physics, Jilin University, Changchun, 130012, China.2Center for High Pressure Science and Technology Advanced Research (HPSTAR), Beijing 100193, China.3Department of Materials Science & Metallurgy, University of Cambridge, 27 Charles Babbage Road, Cambridge CB3 0FS, United Kingdom4These authors contributed equally: Ge Sun, Chenjie Lou.5Email: zluag@connect.ust.hk (Z. L.); mingxue.tang@hpostar.ac.cn (M. T.); dufei@jlu.edu.cn (F. D.) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 96, 227, 111]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[144, 128, 853, 636]]<|/det|> +Exploiting solid electrolyte (SE) materials with high ionic conductivity, good interfacial compatibility, and ultraconformal contact with electrode are essential for solid- state sodium metal batteries (SSBs). Here we report a crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) SE which features high room- temperature ionic conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) and a low activation energy of \(0.15 \mathrm{eV}\) . All- solid- state symmetric cell with \(\mathrm{Na_5SmSi_4O_{12}}\) delivers excellent cycling life over \(800 \mathrm{~h}\) at \(0.15 \mathrm{mA} \mathrm{h} \mathrm{cm}^{- 2}\) and high critical current density of \(1.4 \mathrm{mA} \mathrm{cm}^{- 2}\) . Such excellent electrochemical performance is attributed to an electrochemically induced in- situ crystalline- to- amorphous (CTA) transformation propagating from the interface to the bulk during repeated deposition and stripping of sodium, which lead to faster ionic transport and superior interfacial properties. Impressively, the \(\mathrm{Na_3V_2(PO_4)_3} \| \mathrm{Na_5SmSi_4O_{12}} \| \mathrm{Na} \mathrm{SSBs}\) achieves a remarkable cycling performance over 4000 cycles (6 months) with no capacity loss. These results not only identify \(\mathrm{Na_5SmSi_4O_{12}}\) as a promising SE, but also emphasize the potential of the CTA transition as a promising mechanism towards long- lasting SSBs. + +<|ref|>sub_title<|/ref|><|det|>[[148, 688, 260, 704]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[144, 722, 851, 891]]<|/det|> +All- solid- state batteries (ASSBs) are expected to provide key improvements over today's rechargeable batteries owing to the inherent merits of solid electrolytes (SEs) such as high safety, long- lasting life, and high energy density1, 2. Among them, sodium- based ASSBs are drawing ever- increasing interest because of the abundant resource in nature, beneficial to cost saving and sustainability3, 4. Despite the progress of sodium- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 92, 852, 225]]<|/det|> +based ASSBs in the past few years, it is still a significant challenge to exploit low- cost and facile synthesized SEs with high ionic conductivity, excellent mechanical and chemical stability. Moreover, the poor wetting of the solid- solid interface with sluggish interfacial kinetics is a big hurdle to future Na- ASSBs development5, 6. + +<|ref|>text<|/ref|><|det|>[[144, 240, 852, 821]]<|/det|> +Generally, Na- based inorganic SEs can be divided into two categories, e.g., sulfides like \(\mathrm{Na_3SbS_4}\) , \(\mathrm{Na_{11}Sn_2PS_{12}}\) , \(\mathrm{Na_7P_3S_{11}}\) and oxides including \(\mathrm{Na - \beta^{\prime\prime} - Al_2O_3}\) , NASICON- type7. Sulfide electrolytes feature higher ionic conductivity and better ductility than the oxides \(\mathrm{SEs^{8, 9}}\) . However, their chemical instability against air and narrow electrochemical window are likely to induce complex side reaction, leading to shortened cycling lives10, 11. In contrast, NASICON- type oxides could deliver high thermal and chemical stability, and low thermal expansion12, 13. Nevertheless, large grain boundary resistance and harsh synthesis conditions are critically challenging for their application14, 15. Recently, our group reported a novel structure of SE, \(\mathrm{Na_5YSi_4O_{12}}\) with a high room- temperature ionic conductivity of \(1.59 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) , comparable with the sulfide \(\mathrm{SEs^{16}}\) . Furthermore, in comparison with the NASICON- type materials, \(\mathrm{Na_5YSi_4O_{12}}\) can be synthesized at a lower temperature which is beneficial to cost- and energy- saving. More impressively, the stable structure provides sufficient freedom of materials optimization and design by substituting the Y sites by for instance, In, Sc, and the rare earth Lu- Sm, to further enhance the ionic conductivity and understand the ion- conducting behavior of this class of materials17, 18. + +<|ref|>text<|/ref|><|det|>[[144, 835, 849, 891]]<|/det|> +Besides the difficulty in the SEs exploration, another challenge of developing Na- ASSBs lies in the electrode- electrolyte interfaces, in the mechanics throughout the cell, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 852, 748]]<|/det|> +and in processing at scale19. To begin with, the poor ductility and crystallographic orientation- dependent ionic transport properties of oxide SEs likely induce a large interfacial resistance and sluggish kinetics20, 21. In addition, during the continuous dissolution of sodium, the formation of pores at the sodium metal anode interface will further worsen the interfacial physical contact and generate unavoidable surface defects that could disturb sodium- ion flux and work as the nucleation center, leading to rapid dendrite nucleation and growth22, 23. Moreover, most SEs are intrinsically thermodynamically unstable against sodium metal, which induces degradation and forms mixed conducting interphases, further accelerating the dendrite propagation24, 25. To address these issues, many approaches are employed to fabricate an artificial layer on the surface of sodium metal to improve sodium wetting, chemical stability, and thus reduce interfacial impedance, such as \(\mathrm{TiO_2^{26}}\) , \(\mathrm{SnS_2^{27}}\) , etc28, 29. However, they face practical limitations, such as complex synthesis procedures and difficulty in controlling the thickness and achieving acceptable adhesion. Even worse, the artificial interface layer is likely to introduce additional interfacial issues with bulk SEs, such as unexpected phase transition, uneven ion flux distribution, and electrostatic potential drop and formation of “space- charge layer”, seriously limiting the ion transport and reducing the cycle life of ASSBs30. + +<|ref|>text<|/ref|><|det|>[[147, 760, 852, 891]]<|/det|> +Herein, we report a new member of the \(\mathrm{Na_5MSi_4O_{12}}\) family with \(\mathrm{M} = \mathrm{Sm}\) . It has the highest room- temperature ionic conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) among the \(\mathrm{Na_5MSi_4O_{12}}\) family that have been reported. Interestingly, we observe an electrochemically induced crystalline- to- amorphous (CTA) transformation of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 853, 675]]<|/det|> +\(\mathrm{Na_5SmSi_4O_{12}}\) SE during repeated deposition and stripping of Na. This CTA transition is attributed to the lattice stress generated upon \(\mathrm{Na^{+}}\) transportation rather than phase transformation due to chemical instability. When applied in a Li symmetric cell with the same cell configuration (Li|Na5SmSi4O12||Li), this CTA process is speeded up because of the mismatch between \(\mathrm{Li^{+}}\) and \(\mathrm{Na^{+}}\) ionic radius, which further improves the selectivity of the Li ASSBs. Beneficial from the enhanced mechanical properties, decreased ion mobility activation energy, and lower interfacial energy of amorphous material and interface than crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) , symmetric Na cells deliver a low overpotential of \(\sim 26 \mathrm{mV}\) and ultra- stable cycling performance over \(800 \mathrm{h}\) at \(0.15 \mathrm{mA} \mathrm{cm}^{- 2}\) . Moreover, the amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) facilitates intimate contact of SE with Na metal and brings essentially improved critical current density (CCD) of \(1.4 \mathrm{mA} \mathrm{cm}^{- 2}\) in comparison with the initial crystalline stage \((0.6 \mathrm{mA} \mathrm{cm}^{- 2})\) . By virtue of the decreased resistance of sodium metal anode, the assembled quasi- solid- state \(\mathrm{Na_3V_2(PO_4)_3||Na_5SmSi_4O_{12}||Na}\) cell demonstrates an ultra- long cycle lives over 4000 cycles with \(\sim 100\%\) Coulombic efficiency and capacity retention, indicative of the promising application of \(\mathrm{Na_5SmSi_4O_{12}}\) SE in future large- scale energy storage. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 95, 215, 111]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[147, 131, 654, 149]]<|/det|> +## Synthesis and characterization of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) + +<|ref|>text<|/ref|><|det|>[[144, 164, 852, 751]]<|/det|> +Hexagonal- prismatic crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) was successfully synthesized via two- step solid- state reaction, first at \(800^{\circ}\mathrm{C}\) for \(8\mathrm{h}\) and then \(950^{\circ}\mathrm{C}\) for \(20\mathrm{h}\) , according to the stoichiometric mixtures of \(\mathrm{Na_2CO_3}\) , \(\mathrm{Sm_2O_3}\) and \(\mathrm{SiO_2}\) . As compared in the XRD patterns before and after the second sintering (Supplementary Fig. 1), pure \(\mathrm{Na_5SmSi_4O_{12}}\) can form after the first sintering. And the second sintering helps to achieve a dense ceramic pellet with few interfacial and bulk pores (Supplementary Fig. 2), beneficial to lower the grain boundary resistance and increase the ionic conductivity. Note that the sintering temperature of \(\mathrm{Na_5SmSi_4O_{12}}\) is lower than other oxide SEs (Supplementary Table 1), such as \(\mathrm{Na_5YSi_4O_{12}}\) , \(\mathrm{Na - \beta^{\prime\prime} - Al_2O_3}\) , \(\mathrm{Na_3Zr_2Si_2PO_{12}}\) and its derivatives, etc., good for the cost- and energy- saving. As shown in Fig. 1a, X- ray diffraction (XRD) was taken to identify the structural property of \(\mathrm{Na_5SmSi_4O_{12}}\) , and the Rietveld refined parameters are listed in Supplementary Table 2. All the diffraction peaks can be indexed into a hexagonal system with space group \(R - 3c\) , and the lattice parameters are calculated as \(a = b = 22.14609\mathrm{\AA}\) and \(c = 12.68858\mathrm{\AA}\) . Energy dispersive X- ray spectroscopy (Supplementary Fig. 3) suggests all the elements of \(\mathrm{Na}\) , \(\mathrm{Sm}\) , \(\mathrm{Si}\) and \(\mathrm{O}\) are uniformly dispersed in the as- prepared \(\mathrm{Na_5SmSi_4O_{12}}\) . + +<|ref|>text<|/ref|><|det|>[[145, 760, 850, 891]]<|/det|> +The ionic conductivity of \(\mathrm{Na_5SmSi_4O_{12}}\) was then studied by the alternating current (AC) impedance spectra. The Nyquist plot at room temperature is, as presented in Fig. 1b, composed with a semicircle at high frequency and a tail at low frequency. Via fitting the bulk and grain boundary resistance, the total ionic conductivity of \(\mathrm{Na_5SmSi_4O_{12}}\) is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 92, 854, 636]]<|/det|> +calculated as \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) , among the highest values in the existing ceramic electrolytes (Supplementary Table 1). Activation energy \((E_{\mathrm{a}})\) of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) was further evaluated via temperature- dependent AC impedance spectra, as displayed in Fig. 1b. With increasing temperature, the high frequency semicircle gradually vanishes, indicative of enhanced \(\mathrm{Na}^{+}\) transport across the grain boundaries. By fitting the Arrhenius plot (Fig. 1c), \(E_{a}\) is calculated as small as \(0.15 \mathrm{eV}\) , indicative of a rapid ion hoping. In addition, the electronic conductivity of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) was measured as about \(5.8 \times 10^{- 10} \mathrm{~S} \mathrm{~cm}^{- 1}\) via a direct current (DC) polarization measurement (Supplementary Fig. 4). The intrinsic electronic insulation can effectively reduce the self- discharge of batteries and suppress dendrite growth, enabling \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) as a good candidate for sodium based \(\mathrm{ASSB}_{5}^{31}\) . Besides the merits of high ionic and low electronic conductivities, \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) demonstrates superior moisture stability, whose XRD pattern shows no change after soaking in deionized water for 48 h (Supplementary Fig. 5) or exposing to air for 45 days (Supplementary Fig. 6), showing great potential in future industrial application. + +<|ref|>text<|/ref|><|det|>[[145, 650, 852, 891]]<|/det|> +To further reveal the ion conduction mechanism, molecular dynamics simulation was carried out. As shown in Fig. 1d- g, Na ions in \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) can be characterized by their mobility, i.e., the non- mobile ones \(\mathrm{Na}_{1}\) , \(\mathrm{Na}_{2}\) , \(\mathrm{Na}_{3}\) and the mobile ones \(\mathrm{Na}_{4}\) , \(\mathrm{Na}_{5}\) , \(\mathrm{Na}_{6}\) . The conductivity is contributed by a percolating conduction pathway in the mobile region, whereas the rest of Na ions serve as a pillar to hold the structure together. Such a behavior was also proposed in the same family of materials from our previous work16. Interestingly, along the ion conduction pathway, three distinctive sites are + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 93, 852, 520]]<|/det|> +revealed where Na ions can stably sit in, denoted as sites A, B, and C. These sites are connected to each other via the zigzag like channel as observed from the molecular dynamics simulation trajectory. Furthermore, the diffusion barrier of Na ions between these sites are found to be \(\sim 0.3 \mathrm{eV}\) , which is relatively low in comparison with other reported solid ion conductors32, 33. However, such a value is larger than the activation energy from experiment. We assign such a deviation to the concerted Na hopping behavior. The \(\sim 0.3 \mathrm{eV}\) barrier was calculated by assuming a vacancy-mediated uncorrelated conduction mechanism, while the Na concentration is relatively high and concerted motion is favored. The case of concerted motion is further estimated by assuming a two-ion correlated hopping mechanism, as shown in Fig. 1h. According to such a mechanism, the diffusion barrier along the same path dropped to \(\sim 0.19 \mathrm{eV}\) , close to the experimental value. + +<|ref|>text<|/ref|><|det|>[[144, 538, 852, 891]]<|/det|> +Furthermore, solid- state 23Na nuclear magnetic resonance (NMR) measurements were carried out to reveal the atomic local structure and dynamics mobility34, 35, 36. As exhibited in Fig. 1i, the pristine \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) conductor shows multiple 23Na resonances, which can be further deconvoluted into six types of signals. Via the aid of crystal structure and the proportion of sodium, the peak at 8.8 ppm is assigned to the mobile sodium at Na5 site, the resonance at 4.5 ppm and 1.6 ppm are attributed to Na1 and Na3, the signal at - 16.8 ppm is from Na4, and the rest peaks at - 23.4 ppm and - 28.6 ppm are assigned to sodium at Na2 and Na6 sites, respectively. The simulated details are listed in Supplementary Table 3, from which the proportion of each Na atom agrees with the theoretical results. To obtain more information about \(\mathrm{Na}^{+}\) dynamics properties, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 92, 852, 597]]<|/det|> +NMR spectra at different temperatures were performed. As displayed in Supplementary Fig. 7, the \(\mathrm{Na_5SmSi_4O_{12}}\) show similar spectra upon increasing temperature from \(213\mathrm{K}\) to \(367\mathrm{K}\) , except decrease in signal due to Boltzmann distribution. No obvious change is observed for spectral configuration, possibly attributed to the stable structure caused by the non- mobile Na ions at \(\mathrm{Na1}\) , \(\mathrm{Na2}\) , and \(\mathrm{Na3}\) sites. Therefore, we turn to a more temperature sensitive parameter, spin- lattice relaxation (SLR) time \((T_{1})\) . Since the strong quadrupole interaction is observed for \(^{23}\mathrm{Na}\) NMR spectrum in our system, saturation recovery technique is employed to determine \(T_{1}\) values \(^{37}\) . Via fitting the spectral intensity vs. saturation recovery period measured at room- temperature (Fig. 1j), two relaxation time values of 0.5 and \(20\mathrm{ms}\) are obtained for the non- mobile and mobile Na ions, respectively \(^{38}\) . Here, the data analysis is simplified to non- mobile and mobile for convenience. According to Eq. (1), the fitting of the relaxation times as a function of temperature yields an activation energy \(E_{\mathrm{a}} \approx 0.13\mathrm{eV}\) for the mobile \(\mathrm{Na^{+}}\) , which is in good agreement with the results from AC impedance (0.15 eV). + +<|ref|>equation<|/ref|><|det|>[[399, 610, 848, 632]]<|/det|> +\[R_{1} = 1 / T_{1}\propto \omega_{0}^{\beta}\mathrm{exp}[-E_{a} / (kT)] \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[144, 648, 852, 853]]<|/det|> +Where \(R_{1}\) is NMR spin- lattice relaxation rate, the reciprocal of \(T_{1}\) , \(\omega_{0}\) is resonance frequency, \(\beta\) is modified exponent, \(E_{\mathrm{a}}\) is activation energy, \(k\) is Boltzmann constant, \(T\) is the absolute temperature in K. Note that the derivate data points marked by hollow circle, as displayed in Fig. 1k, were excluded from the fits since the \(R_{1}\) rates \((1 / T_{1})\) recorded at low temperature range are mainly governed by non- diffusive background effects, such as lattice vibrations or coupling by paramagnetic impurities \(^{39,40,41}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[156, 92, 840, 571]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 604, 852, 789]]<|/det|> +
Fig. 1. Crystal structure and sodium-ion conduction characteristic of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . a Rietveld refinement based on the powder XRD. b Nyquist plots of \(\mathrm{Na_5SmSi_4O_{12}}\) from 25 to \(175^{\circ}\mathrm{C}\) . c Arrhenius plot of the conductivity values for \(\mathrm{Na_5SmSi_4O_{12}}\) . d Top-down and e perspective view of crystal structures of crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . f-g Molecular dynamics simulation trajectories of \(\mathrm{Na_5SmSi_4O_{12}}\) . h Minimum potential energy path along \(\mathrm{Na^{+}}\) diffusion route in crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . i Solid-state \(^{23}\mathrm{Na}\) NMR spectrum and its simulation for the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) . The gray line is experimental data and the green-dashed line is the sum of simulation. j Saturation recovery fitting curve for the data obtained at room temperature. k Temperature dependence of \(^{23}\mathrm{Na}\) NMR relaxation rate as a function of temperature in \(\mathrm{K}^{-1}\) . The solid line is the fit according to Eq. (1). The derivation of the data is not used for the fit.
+ +<|ref|>text<|/ref|><|det|>[[100, 806, 850, 864]]<|/det|> +Electrochemical performance and crystalline-to-amorphous (CTA) transition of \(\mathrm{Na_5SmSi_4O_{12}}\) solid electrolyte. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 853, 789]]<|/det|> +To further evaluate the chemical and electrochemical stability of SE against Na metal, symmetric all- solid- state Na cell using \(\mathrm{Na_5SmSi_4O_{12}}\) as the electrolyte was fabricated and tested by the repeatedly galvanostatic stripping and plating at different current densities. The charge/discharge profiles of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) without any interfacial modification were recorded at a current density ranging from 0.05 to \(0.15\mathrm{mAcm}^{- 2}\) with \(120\mathrm{min}\) per cycle (Fig. 2a). The cell displays stable and long cyclic performance which maintains an overvoltage of \(\sim 26\mathrm{mV}\) at \(0.15\mathrm{mAcm}^{- 2}\) with negligible fluctuations over \(800\mathrm{h}\) , indicating sodium dendrite- free plating/stripping and excellent kinetic stability of the \(\mathrm{Na_5SmSi_4O_{12}}\) against Na metal. Afterwards, electrochemical impedance was collected from the symmetric cell after cycling 150, 200 and \(300\mathrm{h}\) to reveal the changes in the internal resistance, as illustrated Fig. 2b and Supplementary Table 4. The resistance of SE (both \(\mathrm{R_b}\) and \(\mathrm{R_{GB}}\) ) remains nearly unchanged as the sodium plating/stripping proceeding. In contrast, the interfacial resistance \(\mathrm{(R_{int})}\) between sodium metal and SE decreases at initial few cycles and then stabilizes at a relatively low value, demonstrating excellent compatibility between \(\mathrm{Na_5SmSi_4O_{12}}\) and sodium metal. This phenomenon is quite different from most of reported SEs without surface modification, whose \(\mathrm{R_{int}}\) increases gradually with increasing cycling times with the result of the excessive internal resistance and failure of the cells due to interfacial side reactions or the formation of holes \(^{25,28,42}\) . + +<|ref|>text<|/ref|><|det|>[[147, 799, 851, 891]]<|/det|> +Scanning electron microscopy (SEM) images of the electrodes at different plating/stripping stages are shown in Fig. 2c and 2d, which demonstrates an interesting tendency of gradually vanishing gap at the interface between Na and \(\mathrm{Na_5SmSi_4O_{12}}\) after + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 852, 895]]<|/det|> +cycling and an ultraconformal interfacial contact was created. XRD pattern after cycling 200 h suggests a transformation from the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) SE into the amorphous state (Fig. 2e). Such CTA transition starts from the surface and then gradually propagates into the bulk of the SE pellets since there is no clear reflection peaks from the depth- profiled XRD pattern (Supplementary Fig. 8). To further confirm the CTA transition, HRTEM and SAED measurements were undertaken before (Supplementary Fig. 9) and after cycling (Fig. 2f- g). There are no observed lattice fringe and diffraction spot for the cycled SE sample, confirming the electrochemistry- induced CTA transition. Because of the amorphous state, it is difficult to determine the possible local coordination based on the XRD measurement. Therefore, Raman spectra were then used to examine the short- range vibration changes. As displayed in Supplementary Fig. 10, crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) exhibits seven vibration peaks43: the bands in the 900- 1100 cm- 1 region are assigned to Si- O- stretching vibrations, the symmetrical band at \(\sim 624 \mathrm{cm^{- 1}}\) is corresponding to O- Si- O bending mode. The low- frequency bands (<550 cm- 1) are attributed to Sm- O and Na- O bond vibrations in their polyhedral. Though the cycled \(\mathrm{Na_5SmSi_4O_{12}}\) SE loses its long- term ordering (Fig. 2e- g), it maintains nearly all the vibrational peaks that confirms no chemical reaction between the interface of SE and Na metal. While the disappearance of the peak at 1041 cm- 1 might be related to the damage of the fracture of Si- O bond by the amorphous transition. Furthermore, X- ray photoelectron spectroscopy (XPS) suggests that there are no changes in the Sm 3d, Na 1s and Si 2p XPS spectra after cycling, indicative of no redox reaction occurred between sodium metal and \(\mathrm{Na_5SmSi_4O_{12}}\) (Supplementary + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 93, 850, 188]]<|/det|> +1 Fig. 11). Thus, it can be concluded that \(\mathrm{Na_5SmSi_4O_{12}}\) SE demonstrates an interesting 2 CTA transition with an excellent electrochemical stability that enables it as the ideal 3 SE for sodium based ASSBs. + +<|ref|>image<|/ref|><|det|>[[196, 210, 789, 580]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 586, 850, 696]]<|/det|> +
Fig. 2. Characterization of interfacial evolution and \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling in \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) symmetric cells. a Cycling performance of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) at room temperature. b Nyquist plots of the SE-based symmetric cell after cycling for different times. Cross-sectional SEM images of the Na metal/ \(\mathrm{Na_5SmSi_4O_{12}}\) interfaces: c pristine and d 100 h cycling. e XRD profiles of \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling for different times. f HRTEM and g SAED patterns of \(\mathrm{Na_5SmSi_4O_{12}}\) after cycling.
+ +<|ref|>sub_title<|/ref|><|det|>[[145, 715, 614, 732]]<|/det|> +## Electrochemical performances of solid-state-batteries. + +<|ref|>text<|/ref|><|det|>[[144, 752, 850, 882]]<|/det|> +Critical current density (CCD) and long- term cycling performance of \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) with a long single deposition time were further measured to evaluate the capability of amorphous materials and interface in suppressing dendrite growth. As displayed in Fig. 3a, CCD for \(\mathrm{Na}||\mathrm{amorphous Na_5SmSi_4O_{12}}||\mathrm{Na}\) cell is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 92, 852, 560]]<|/det|> +achieved as \(1.4\mathrm{mAcm}^{- 2}\) at \(1.4\mathrm{mA hcm}^{- 2}\) , higher than the CCD for \(\mathrm{Na}\|\) crystalline \(\mathrm{Na_5SmSi_4O_{12}||Na}\) (0.4 mA h cm \(^{- 2}\) , Supplementary Fig. 12). This behavior is mainly because the uneven metal deposition leads to rapid growth of sodium dendrites along the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) grain boundary with a large deposition current and hence results in the short circuit rapidly. In addition, as shown in Supplementary Fig. 13a, under a large area capacity, the overpotential increases during the cycling and drops suddenly less than \(25\mathrm{h}\) , which suggests that crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) can be easily penetrated by sodium dendrites. By contrast, if a low area capacity of \(0.05\mathrm{mA hcm}^{- 2}\) is applied in advance to make the electrolyte transition to an amorphous state, the \(\mathrm{Na}\|\) amorphous \(\mathrm{Na_5SmSi_4O_{12}||Na}\) cell displays a stable and long cycling performance which maintains the overvoltage at around \(20\mathrm{mV}\) over \(500\mathrm{h}\) (Supplementary Fig. 13b). All these results indicate the strong capability of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) in suppressing Na dendrite formation. + +<|ref|>text<|/ref|><|det|>[[147, 575, 852, 892]]<|/det|> +To further emphasize the superiority of amorphous interface and bulk materials, a \(\mathrm{Na_3V_2(PO_4)_3}\) (NVP) \(\| \mathrm{Na_5SmSi_4O_{12}||Na}\) SSB was constructed and evaluated at room temperature, as illustrated in the schematic figure (Fig. 3b). The cyclic voltammetry (CV) curve of stainless steel (SS) \(\| \mathrm{Na_5SmSi_4O_{12}||Na}\) cell in Fig. 3c shows that \(\mathrm{Na_5SmSi_4O_{12}}\) possesses a wide electrochemical stability window more than \(5\mathrm{V}\) , which is high enough to ensure that the electrolyte does not undergo phase transition within the working voltage range. As shown in Supplementary Fig. 14a, quite flat charge- discharge voltage profiles at \(2.3 - 3.9\mathrm{V}\) with an initial discharge capacity of \(112\mathrm{mA h}\) \(\mathrm{g}^{- 1}\) were observed, which matches well with the characteristic NVP redox plateaus in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 92, 854, 520]]<|/det|> +1 liquid electrolyte (Supplementary Fig. 14b). In addition, the high initial Coulombic 2 efficiency of \(99\%\) indicates that there is no irreversible side reaction between 3 \(\mathrm{Na_5SmSi_4O_{12}}\) and Na anode or NVP cathode. Meanwhile, an excellent cycling 4 performance is achieved with a high- capacity retention of \(95\%\) after 100 cycles (Fig. 5d). Furthermore, specific capacities of \(102\mathrm{mA h g^{- 1}}\) , \(98\mathrm{mA h g^{- 1}}\) and \(93\mathrm{mA h g^{- 1}}\) can 6 be obtained at 0.5, 0.75 and 1 C- rates, respectively, suggestive of a superior rate 7 capability (Fig. 3e). Finally, the long- term cycling stability was estimated at a current 8 rate of 2 C (Fig. 3f). Impressively, there is no obvious capacity loss during the repeated 9 4000 cycles (6 months). The capacity remained to be \(\sim 95\mathrm{mA h g^{- 1}}\) . All these features 10 verify the unique interface properties between \(\mathrm{Na_5SmSi_4O_{12}}\) and sodium, which can not 11 only enable sufficient contact with sodium metal, but also inhibit the side reaction and 12 the dendrite growth during cycle process. + +<|ref|>image<|/ref|><|det|>[[163, 536, 831, 893]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 85, 852, 195]]<|/det|> +Fig. 3. The electrochemical performance of solid-state batteries. a Potential response of \(\mathrm{Na}||\mathrm{Na}5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) cell during the CCD measurement with a low area capacity of \(0.05\mathrm{mA h cm^{- 2}}\) is applied for \(150\mathrm{h}\) in advance. b Schematic illustration of \(\mathrm{Na}_3\mathrm{V}_2(\mathrm{PO}_4)_3||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) . c CV curve of \(\mathrm{SS}||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) cell at a scanning speed of \(5\mathrm{mV s^{- 1}}\) . d Cycling performance at \(0.5\mathrm{C}\) - rate (1 C corresponds to \(118\mathrm{mA g^{- 1}}\) ). e Rate capability at 0.2, 0.5, 0.75 and 1 C-rate. f Long-term cycle life at \(2\mathrm{C}\) - rate. + +<|ref|>sub_title<|/ref|><|det|>[[148, 216, 243, 232]]<|/det|> +## Discussion + +<|ref|>sub_title<|/ref|><|det|>[[148, 252, 464, 270]]<|/det|> +## Cell stabilization by CTA transition. + +<|ref|>text<|/ref|><|det|>[[144, 285, 852, 904]]<|/det|> +According to the above- mentioned results, the electrochemical- induced CTA transition plays a key role in stabilizing the interfacial properties, suppressing the dendrite formation, and thus increasing the long- term stability and high- rate capability. This superiority of the amorphous stage from the interface to the bulk SE material can be understood in terms of the following two aspects. Firstly, the ionic conductivity of amorphous bulk material was improved. The pathway and the ion transport properties of amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) were investigated in detail by solid- state NMR. Figure 4a shows the \(^{23}\mathrm{Na}\) NMR spectra before and after metallic Na cycling by using crystalline \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) as electrolyte. The \(^{23}\mathrm{Na}\) NMR spectrum did not change significantly before and after cycling, except changes of the mobile ions (Na4, Na5, and Na6), indicative of their slight redistribution upon cycling. Nevertheless, NMR results demonstrate the non- mobile and mobile segments for Na migration within stable structure. In addition, spin- lattice relaxation time \(T_1\) of \(^{23}\mathrm{Na}\) were measured at different temperatures for amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) , as shown in Fig. 4b and Supplementary Fig. 15. The activation energy of the mobile \(\mathrm{Na}^+\) of amorphous \(\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}\) is calculated as \(0.07\mathrm{eV}\) lower than that of the pristine crystalline state (0.13 eV in Fig. 1k). The decrease in the active energy strongly suggests an enhanced \(\mathrm{Na}^+\) hopping ability for the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 852, 860]]<|/det|> +amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) with higher ionic conductivity. Secondly, the interfacial issues, such as high interfacial resistance and metal dendrite growth, are strongly alleviated. The schematic illustration of sodium deposition is provided in Fig. 4d. A locally solid- solid contact induces a heterogeneous sodium- ion flux, whereas a tight interfacial contact of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) lead to uniform deposition of sodium. As mentioned above in Fig. 2b, there is obvious decrease in the \(\mathrm{R_{int}}\) upon cycling at different plating/stripping stages. The interesting phenomenon can be attributed to ultraconformal interfacial contact between Na metal and electrolyte due to the increased contact area (Fig. 2c- d). Furthermore, the isotropic surface of amorphous material could enhance the wettability with significantly reduced interface resistance and eliminates the influence of crystallographic orientation- dependent ionic transport since the interfacial energy of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) is calculated as \(0.33\mathrm{Jm}^{- 2}\) with sodium, lower than the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) (0.56 J \(\mathrm{m}^{- 2}\) ) (Supplementary Fig. 16). In addition, the amorphous- \(\mathrm{Na_5SmSi_4O_{12}}\) deliver the high mechanical strength, beneficial to inhibiting the dendrite growth. As shown in Fig. 4c, nanoindentation technique was employed to evaluate the Young's modulus \(E\) and hardness \(H^{44,45}\) . As for the amorphous sample, the \(E\) and \(H\) are calculated to be \(\sim 79.9\) GPa and \(\sim 3.8\) GPa, respectively, higher than the crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) ( \(\sim 72.6\) GPa and \(\sim 2.8\) GPa). In summary, the intrinsic microstructural and compositional homogeneity, as well as low electronic conductivity, alleviate the potential fluctuations at local positions in the SE and suppress sodium propagation and penetration into amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[175, 100, 812, 430]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 438, 852, 530]]<|/det|> +
Fig. 4. The superiority of amorphous bulk materials and interface. a Solid-state \(^{23}\mathrm{Na}\) NMR of the pristine crystalline and the cycled amorphous \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . b \(^{23}\mathrm{Na}\) NMR relaxation rate of Na cycled \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) as a function of temperature in \(\mathrm{K}^{-1}\) . c Nanoindentation load-displacement curves of crystalline \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) and amorphous \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . d Schematic of interface morphology evolution during sodium plating/stripping.
+ +<|ref|>sub_title<|/ref|><|det|>[[150, 549, 414, 565]]<|/det|> +## Mechanism of CTA transition. + +<|ref|>text<|/ref|><|det|>[[144, 585, 852, 900]]<|/det|> +Electrochemically- induced solid- state amorphization (SSA) transformations are popular in the alloying type anode, like the transition from nano Si into amorphous Li- Si phases after electrochemical lithiation46, 47. Nevertheless, this kind of SSA process usually undergoes a chemical phase transition, which is disastrous for SEs. To the best of our knowledge, it is the first observation about the electrochemically induced CTA transformation in oxide SEs during alkali metal plating/stripping without chemical phase transition. Generally, the SSA transformation depends on the thermodynamic driving force (pressure, defects, internal stress, etc.), and the existence of kinetic constraints (mainly because of the low experimental temperature) that prevent the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 92, 853, 595]]<|/det|> +formation of full equilibrium crystalline phases48, 49, 50. For example, when ion conduction is non- homogeneous, the local variance may lead to large local stress, which could in turn lead to the collapse of the structure. Through cycling, SSA transformation may occur. In this case, we calculate the formation energies of the crystalline and the amorphous phases sampled from quenched melts. As shown in Supplementary Fig. 17, the crystalline phases are thermodynamically favored which an energy \(\sim 60 \mathrm{meV}\) atom1 lower than the amorphous structures. This indicates that kinetics may play a critical role in the SSA. One way to magnify such kinetic driving force is to apply large lattice strain on the system. In this context, we computationally substitute the Na ions to Li ions in the structures and further compute the energy difference between the crystalline and the amorphous phases. Interestingly, when the Li concentration reaches \(\sim 2 / 5\) , the energy difference significantly drops to \(\sim 10 \mathrm{meV}\) atom- 1. This further supports our assumption that the driving force led by the large lattice strain could be the origin of the SSA. + +<|ref|>text<|/ref|><|det|>[[144, 612, 853, 891]]<|/det|> +Following the above results, a Li- Na exchange process is captured by atomistic simulations, since the ionic radius of \(\mathrm{Li^{+}}\) (76 pm) is smaller than \(\mathrm{Na^{+}}\) (102 pm) that could induce much larger lattice strain. Large- scale molecular dynamics is run on the hypothetical \(\mathrm{Na_5SmSi_4O_{12} / Li_5SmSi_4O_{12}}\) using a machine- learned forcefield, as shown in Fig. 5a. Interestingly, the Li ions first exchanges with the mobile Na ions at the reaction front followed by mixing with the non- mobile ones, see Fig. 5b- e. When the pillar Na ions are replaced by smaller Li ions, the structure start to collapse. During the initial exchange process, crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) and \(\mathrm{Na_5 - xLi_xSmSi_4O_{12}}\) could co- exist. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 852, 335]]<|/det|> +We computationally evaluated how Li substitution affects the ion diffusion. As shown in Fig. 5f, when all mobile Na ions are replaced by Li, Li diffusion barrier along the original zig-zag route dropped from \(\sim 0.3 \mathrm{eV}\) to \(\sim 0.25 \mathrm{eV}\) , indicating a faster diffusion kinetics, which may in turn, result in faster SSA during Li exchange. In conclusion, lattice strain induced by ion intercalation is the main driving force of amorphous transformation. Once the thermodynamic driving force is present, the kinetic hindrance at room- temperature prevents the transition back to amorphous \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) . + +<|ref|>text<|/ref|><|det|>[[144, 352, 852, 895]]<|/det|> +Guided by the computational results, a hybrid symmetric cell of \(\mathrm{Li} \| \mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12} \| \mathrm{Li}\) was fabricated to accelerate this SSA transition. As shown in Supplementary Fig. 18, the cell reveals uniform plating and stripping overpotential profiles with an increased current density. This phenomenon suggests that hybrid movement of \(\mathrm{Li}^{+} / \mathrm{Na}^{+}\) within the bulk of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) and an effective plating/stripping of \(\mathrm{Na}^{+}\) at Li anode. As expected, \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) exhibits a much shorter amorphous time within \(100 \mathrm{~h}\) (Fig. 5g and Supplementary Fig. 19), confirming that the CTA transition of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) is mainly triggered by the lattice strain and speeded up because of the mismatch between \(\mathrm{Li}^{+}\) and \(\mathrm{Na}^{+}\) ionic radius. Furthermore, an obvious reflections shift and weakening can be observed in the XRD patterns after experiencing different cycling time (Supplementary Fig. 20). This shift indicates the existence of microscopic strain in the lattice, and the gradual accumulation of stress leads to the break of more bonds. Thus, with the increase of cycling time, the crystallinity of \(\mathrm{Na}_{5} \mathrm{Sm} \mathrm{Si}_{4} \mathrm{O}_{12}\) weakens, and the intensity of XRD peaks decreases gradually. To assess the cationic electrochemical exchange mechanism, XPS analysis of SE operating in the hybrid cell + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 852, 300]]<|/det|> +was carried out after cycling. As shown in Supplementary Fig. 21, the decrease in Na 1s peak and appearance of Li 1s peak prove that \(\mathrm{Li^{+}}\) ions can successfully replace part of \(\mathrm{Na^{+}}\) ions and the strong \(\mathrm{Li^{+}}\) mobility within hexagonal-prismatic \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) . In addition, there is no observed new Raman peaks after cycling in the hybrid symmetric cell (Supplementary Fig. 22), which indicates that \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) SE demonstrates an excellent thermodynamic stability with Li metal. + +<|ref|>text<|/ref|><|det|>[[144, 315, 853, 820]]<|/det|> +Figure 5h shows the \(^{23}\mathrm{Na}\) spectra of \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) cycled with Li metal under different times, from which the stripped Li will exchange with Na on its way when across the electrolyte. The signals at - 16.8 ppm - 23.4 ppm, being assigned to sodium at \(\mathrm{Na4}\) and \(\mathrm{Na6}\) sites, significantly weakened, reflecting the transport activity of \(\mathrm{Na4}\) and \(\mathrm{Na6}\) sodium sites during polarization. In addition, the signal at 8.8 ppm, which is assigned to sodium at \(\mathrm{Na5}\) site, is weakened but fluctuates in intensity at different cycle times, indicating it likely serves as 'bridge' for transporting \(\mathrm{Na / Li}\) ions during polarization. These results further conclude that there is a possible 3D pathway of \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) between \(\mathrm{Na4 - Na5 - Na6}\) . Figure 5i and Supplementary Fig. 23 display the \(^{7}\mathrm{Li}\) NMR spectrum of \(\mathrm{Na_{5 - x}Li_{x}SmSi_{4}O_{12}}\) after Li cycled different times (100 and 200 h). The \(^{7}\mathrm{Li}\) NMR spectrum of the electrolyte cycled for 200 h is broader than that for 100 h, indicating the growth of amorphous phase. \(^{6}\mathrm{Li}\) NMR spectrum after Li cycling is shown in Supplementary Fig. 24. Both \(^{7}\mathrm{Li}\) and \(^{6}\mathrm{Li}\) NMR spectra present two components, corresponding to the different mobile Na sites, such as \(\mathrm{Na4}\) and \(\mathrm{Na6}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[163, 90, 825, 455]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 457, 852, 567]]<|/det|> +
Fig. 5. CTA mechanism. a Structure of the \(\mathrm{Na_5SmSi_4O_{12}||Li_5SmSi_4O_{12}}\) interface model b molecular dynamics trajectories of the interface model. Molecular dynamics simulation trajectories of crystalline c \(\mathrm{Li_5SmSi_4O_{12}dNa_5SmSi_4O_{12}}\) and e \(\mathrm{Li_5Na_2SmSi_4O_{12}}\) . f Minimum potential energy path along Li diffusion route in crystalline \(\mathrm{Na_2Li_5SmSi_4O_{12}}\) . g XRD profiles of \(\mathrm{Na_5 - xLi_5SmSi_4O_{12}}\) after cycling for different times. Solid-state h \(^{23}\mathrm{Na}\) and i \(^{7}\mathrm{Li}\) NMR of the \(\mathrm{Na_5SmSi_4O_{12}}\) with different cycling times.
+ +<|ref|>text<|/ref|><|det|>[[144, 585, 851, 900]]<|/det|> +In summary, dense crystalline \(\mathrm{Na_5SmSi_4O_{12}}\) was prepared and exhibits a high room- temperature conductivity of \(2.90 \times 10^{- 3} \mathrm{~S} \mathrm{~cm}^{- 1}\) . Driven by microscopic strain in the lattice, the \(\mathrm{Na_5SmSi_4O_{12}}\) undergoes an amorphous transformation during the cycling in both symmetric \(\mathrm{Na}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Na}\) and \(\mathrm{Li}||\mathrm{Na_5SmSi_4O_{12}}||\mathrm{Ci}\) cells. On the one hand, the increased contact area greatly reduces the interfacial resistance between sodium metal and electrolyte and promotes the homogeneous deposition of sodium. On the other, the isotropic ionic transport feature of amorphous \(\mathrm{Na_5SmSi_4O_{12}}\) eliminates the ion- blocking crystallographic orientation, uniform distribution of the current and homogeneous metal nucleation at the anode interface will be promoted. Thus, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[100, 92, 855, 250]]<|/det|> +1 sodium symmetrical cells manifest stable cycling performance for \(800\mathrm{h}\) at \(0.15\mathrm{mA}\) 2 \(\mathrm{cm}^{- 2}@1\mathrm{h}\) and \(500\mathrm{h}\) at \(0.05\mathrm{mA}\mathrm{cm}^{- 2}@5\mathrm{h}\) ( \(25^{\circ}\mathrm{C}\) ). Furthermore, the successful 3 operation of \(\mathrm{Na_3V_2(PO_4)_3\|Na_5SmSi_4O_{12}\|Na}\) solid- state sodium batteries with excellent 4 electrochemical performance further implies the superiority of \(\mathrm{Na_5SmSi_4O_{12}}\) electrolyte. 5 + +<|ref|>sub_title<|/ref|><|det|>[[147, 280, 227, 297]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[148, 317, 322, 334]]<|/det|> +## Materials synthesis: + +<|ref|>text<|/ref|><|det|>[[144, 352, 852, 595]]<|/det|> +The \(\mathrm{Na_5SmSi_4O_{12}}\) pellets were synthesized by a solid- state sintering method using \(\mathrm{Na_2CO_3}\) , \(\mathrm{Sm_2O_3}\) and \(\mathrm{SiO_2}\) as the starting materials. First, the raw materials of analytical grade were mixed by ball- milling at a milling speed constant of \(600\mathrm{rpm}\) for \(15\mathrm{h}\) . The mixture was dried at \(80^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) and calcined at \(800^{\circ}\mathrm{C}\) for \(8\mathrm{h}\) . Then the powder was put into a cylindrical pressing mold with diameter of \(15\mathrm{mm}\) and pressed under a pressure of \(300\mathrm{MPa}\) . The pressed pellets were then sintered at \(950^{\circ}\mathrm{C}\) for \(20\mathrm{h}\) . Finally, buff pellets were obtained after sintering. + +<|ref|>text<|/ref|><|det|>[[144, 611, 852, 892]]<|/det|> +The \(\mathrm{Na_3V_2(PO_4)_3}\) (NVP) cathode material was prepared by the sol- gel method according to our previous work (Supplementary Fig. 25)51. Firstly, the stoichiometric amount of \(\mathrm{Na_2CO_3}\) , \(\mathrm{NH_4VO_3}\) and \(\mathrm{NH_4H_2PO_4}\) with a molar ratio of \(3:4:6\) was dissolved in deionized water. Secondly, \(0.02\mathrm{M}\) aqueous citric acid [HOC(COOH)(CH2COOH)2] solution was added dropwise into the solution until the ratio of vanadium: citric acid equals to 2: 1. Then the gel could be acquired by drying the precursor in an oven at \(120^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) . Finally, the NVP/C powder can be acquired after heat treatments in two steps, first at \(350^{\circ}\mathrm{C}\) for \(5\mathrm{h}\) and then at \(750^{\circ}\mathrm{C}\) for \(12\mathrm{h}\) under a nitrogen atmosphere. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 95, 374, 112]]<|/det|> +## Characterization method: + +<|ref|>text<|/ref|><|det|>[[144, 130, 852, 900]]<|/det|> +X- ray diffraction (XRD) patterns were recorded by a Bruker D8 Advance diffractometer with Cu Kα radiation and RigaKu D/max- 2550 diffractometer (1.6 kW, Cu Kα radiation, \(\lambda = 1.5406 \AA\) ), followed by Rietveld refinement using Fullprof software for the crystal structure analysis. The microscopy characteristics of the samples were investigated by Hitachi Regulus8100 FESEM, high resolution transmission electron microscope (HRTEM, talos F200X) and selected area electron diffraction (SAED). The elemental mapping was used to analyze the element distribution of the samples. X- ray photoemission spectrum (XPS) was carried out on a thermo scientific NEXSA spectrometer. Raman spectra were examined using a Renishaw Raman microscope (model 2000) with Ar- ion laser excitation. Prior to analysis the interface properties between Na/Li metal and Na5SmSi4O12 electrolyte after cycling, emery paper was employed to remove any residual metal on the surface. All \(^{6}\mathrm{Li}\) , \(^{7}\mathrm{Li}\) and \(^{23}\mathrm{Na}\) magic angle spinning (MAS) NMR experiments were acquired on Bruker 400 MHz (9.4 T) magnets with AVANCE NEO consoles using Bruker 3.2 mm HXY MAS probe. The samples were filled into rotors inside Argon glove box. The Larmor frequencies for \(^{6}\mathrm{Li}\) , \(^{7}\mathrm{Li}\) and \(^{23}\mathrm{Na}\) were 58.89, 155.53 and 105.86 MHz, respectively. All spectra were acquired by using one- pulse program and were referenced to 1 M LiCl ( \(^{6}\mathrm{Li}\) and \(^{7}\mathrm{Li}\) ) and 1 M NaCl ( \(^{23}\mathrm{Na}\) ) solutions with chemical shifts at 0 ppm. The spinning rate \(\nu_{\mathrm{rot}}\) was set to 14 kHz. \(^{23}\mathrm{Na}\) spin- lattice relaxation times (\(T_{1}\)) were recorded by using the saturation recovery pulse sequence. The varying temperature experiments were protected by \(\mathrm{N}_{2}\) atmosphere. Nanoindentation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 94, 852, 410]]<|/det|> +measurement was taken on a nanoindentation tester (Agilent Nano Indenter G200) equipped with a three- sided pyramidal Berkovich diamond indenter. The applied standard loading, holding, and unloading times were 10, 5, and 10 s, respectively. During the testing, the load- displacement curves up to pellet cracking were recorded and utilized to calculate the Young's modulus \(E\) and hardness \(H\) using the Oliver- Pharr method. Indentations with maximum indentation load of 1 mN are conducted on the surface of SE pellets. The reduced modulus \(E_{\mathrm{r}}\) was determined by the unloading stiffness and projected contact area. By assuming a Poisson's ratio of 0.3 for samples and 0.07 for single crystalline diamond, their Young's moduli were estimated. + +<|ref|>sub_title<|/ref|><|det|>[[149, 437, 373, 454]]<|/det|> +## Impedance spectrum test: + +<|ref|>text<|/ref|><|det|>[[144, 471, 852, 750]]<|/det|> +For the conductivity measurement, silver was spread on both sides of the ceramic pellets as blocking electrodes. AC impedance spectra were recorded using a Solartron 1260 impedance analyzer over a frequency range of 5 MHz to 1 Hz, with an applied root mean square AC voltage of 30 mV. The temperature dependence of the conductivity was measured in the same way at several specific temperatures ranging from 25 to 175 °C. For conductivity test at each temperature, the samples were allowed to equilibrate for 2 h prior to measurements. The resistances of the \(\mathrm{Na}||\mathrm{Na}_5\mathrm{SmSi}_4\mathrm{O}_{12}||\mathrm{Na}\) symmetric cells were tested under the same conditions. + +<|ref|>sub_title<|/ref|><|det|>[[149, 776, 417, 793]]<|/det|> +## Electrochemical measurement: + +<|ref|>text<|/ref|><|det|>[[144, 812, 852, 905]]<|/det|> +To obtain NVP cathode, NVP active material (70 wt.%), Super P conductive additive (20 wt.%) and carboxymethyl cellulose (CMC) binder (10 wt.%) were dissolved in water to form a homogeneous slurry, and then uniformly coated onto an aluminum foil + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 852, 708]]<|/det|> +current collector. After drying for \(12\mathrm{h}\) at \(80^{\circ}\mathrm{C}\) , the electrode was punched into \(1\mathrm{cm}\) diameter wafers for use with the loading mass of \(1.0\mathrm{- }1.5\mathrm{mgcm}^{- 2}\) . Sodium foil was employed as anode. The \(\mathrm{Na_{5}SmSi_{4}O_{12}}\) pellet was used as both separator and electrolyte. \(20\mu \mathrm{L}1\mathrm{MNaClO_4}\) in ethylene carbonate (EC) and propylene carbonate (PC) (1:1 v/v) with the addition of 5 vol.% fluoroethylene carbonate (FEC) was added as the interfacial wetting agent at the cathode side. The 2032- type coin cells were assembled in an argon- filled glovebox. Galvanostatic charge- discharge tests were performed in a cutoff potential window of 2.3- 3.9 V by using Land- 2100 automatic battery tester. All- solid- state \(\mathrm{Na\|Na_{5}SmSi_{4}O_{12}\|Na}\) and \(\mathrm{Li\|Na_{5}SmSi_{4}O_{12}\|Li}\) symmetric cells were assembled to test the sodium/lithium metal stripping/plating at \(25^{\circ}\mathrm{C}\) and \(50^{\circ}\mathrm{C}\) , respectively. In addition, electrochemical stable window of electrolytes was examined by cyclic voltammetry (CV) measurement with the stainless steel \(\mathrm{||Na_{5}SmSi_{4}O_{12}||Na}\) cell in the voltage range of - 1 to \(8\mathrm{V}\) at a scanning rate of \(5\mathrm{mV}\mathrm{s}^{- 1}\) . The direct current (DC) polarization measurement was performed on \(\mathrm{Ag\|Na_{5}SmSi_{4}O_{12}\|Ag}\) cell with a \(300\mathrm{mV}\) potential and the current response was measured for \(100\mathrm{min}\) at ambient temperature. The CV measurement and the DC measurement were performed using a Bio- Logic electrochemical workstation. + +<|ref|>sub_title<|/ref|><|det|>[[148, 732, 365, 749]]<|/det|> +## Computational methods: + +<|ref|>text<|/ref|><|det|>[[147, 767, 852, 898]]<|/det|> +Density functional theory calculations were carried out using the VASP6.3 \(^{52,53}\) package following the setup we used in previous work \(^{54,55,56}\) . Briefly, the PBE exchange- correlation functional was adopted with a planewave basis and a cutoff energy of 520 eV. The reciprocal space was sampled using Monkhorst- Pack grids with a spacing of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 853, 708]]<|/det|> +0.04 Å⁻¹. The convergence for electron self- consistent computations and structural optimizations are set to 10⁻⁴ eV atom⁻¹ and 10⁻³ eV atom⁻¹, respectively. Due to the need of large- scale molecular dynamics simulations, we trained a machine learning forcefield based on ab initio molecular dynamics simulations (AIMD)⁵⁴,⁵⁵. Trajectories from smaller systems with 5 compositions sampled evenly between Na₅SmSi₄O₁₂ and Li₅SmSi₄O₁₂ were collected at high temperatures together with their relaxation trajectories. The energy and forces were used as the label to train the model. The production MD simulations were carried out using such a forcefield. The systems were equilibrated at 600 K for 100 ps and gradually dropped to 300 K with a period of another 100 ps under an NPT thermostat at ambient pressure. Then an NVT production run was carried out at 300 K for 200 ps. The time step was 1 fs. For the interface model, we adopted a universal machine learning model which can capture the ground state structures and their energy. Two interface models were built to mimic the interaction between the Li electrode and the crystalline and amorphous SE. The models were first equilibrated at 500 K for 1000 fs followed by energy minimization. The interfacial energy was calculated by subtracting the energy of the interfaces from the sum of energies of independent bulk phases. + +<|ref|>sub_title<|/ref|><|det|>[[148, 725, 245, 741]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[144, 760, 851, 890]]<|/det|> +1. Deysher G, et al. Transport and mechanical aspects of all-solid-state lithium batteries. Mater. Today Phys. 24, 100679 (2022). +2. Wu Y, Wang S, Li H, Chen L, Wu F. Progress in thermal stability of all-solid-state-Li-ion-batteries. InfoMat 3, 827-853 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 92, 853, 900]]<|/det|> +3. Feng X, et al. Review of modification strategies in emerging inorganic solidstate electrolytes for lithium, sodium, and potassium batteries. Joule 6, 543-587 (2022). +4. Tian Y, et al. Promises and Challenges of Next-Generation "Beyond Li-ion" Batteries for Electric Vehicles and Grid Decarbonization. Chem. Rev. 121, 1623-1669 (2021). +5. Oh JAS, He L, Chua B, Zeng K, Lu L. Inorganic sodium solid-state electrolyte and interface with sodium metal for room-temperature metal solid-state batteries. Energy Storage Mater. 34, 28-44 (2021). +6. Tang B, Jaschin PW, Li X, Bo S-H, Zhou Z. Critical interface between inorganic solid-state electrolyte and sodium metal. Mater. Today 41, 200-218 (2020). +7. Wu JF, et al. Inorganic Solid Electrolytes for All-Solid-State Sodium Batteries: Fundamentals and Strategies for Battery Optimization. Adv. Funct. Mater. 31, 2008165 (2020). +8. Feng X, et al. Heavily Tungsten-Doped Sodium Thioantimonate Solid-State Electrolytes with Exceptionally Low Activation Energy for Ionic Diffusion. Angew. Chem. Int. Ed. Engl. 60, 26158-26166 (2021). +9. Hayashi A, et al. A sodium-ion sulfide solid electrolyte with unprecedented conductivity at room temperature. Nat. Commun. 10, 5266 (2019). +10. Wu EA, et al. A stable cathode-solid electrolyte composite for high-voltage, long-cycle-life solid-state sodium-ion batteries. Nat. Commun. 12, 1256 (2021). +11. Zhu Y, Mo Y. Materials Design Principles for Air-Stable Lithium/Sodium Solid + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[100, 90, 854, 895]]<|/det|> +Electrolytes. Angew. Chem. Int. Ed. Engl. 59, 17472- 17476 (2020). 12. Sun F, et al. Origin of High Ionic Conductivity of Sc- Doped Sodium- Rich NASICON Solid- State Electrolytes. Adv. Funct. Mater. 31, 2102129 (2021). 13. Yang J, et al. Ultrastable All- Solid- State Sodium Rechargeable Batteries. ACS Energy Lett. 5, 2835- 2841 (2020). 14. Zou Z, et al. Relationships Between \(\mathrm{Na + }\) Distribution, Concerted Migration, and Diffusion Properties in Rhombohedral NASICON. Adv. Energy Mater. 10, 2001486 (2020). 15. Lu Y, Li L, Zhang Q, Niu Z, Chen J. Electrolyte and Interface Engineering for Solid- State Sodium Batteries. Joule 2, 1747- 1770 (2018). 16. Sun G, et al. \(\mathrm{Na_5YSi_4O_{12}}\) : A sodium superionic conductor for ultrastable quasi- solid- state sodium- ion batteries. Energy Storage Mater. 41, 196- 202 (2021). 17. Schilm J, et al. Influence of \(\mathrm{R = Y}\) , Gd, Sm on Crystallization and Sodium Ion Conductivity of \(\mathrm{Na_5RSi_4O_{12}}\) Phase. Mater. (Basel) 15, (2022). 18. Shannon RD, Taylor BE, Gier TE, Chen HY, Berzins T. Ionic conductivity in sodium yttrium silicon oxide \(\mathrm{(Na_5YSi_4O_{12})}\) - type silicates. Inorg. Chem. 17, 958- 964 (2002). 19. Zhou W, Li Y, Xin S, Goodenough JB. Rechargeable Sodium All- Solid- State Battery. ACS Cent. Sci. 3, 52- 57 (2017). 20. Zahiri B, et al. Revealing the role of the cathode- electrolyte interface on solid- state batteries. Nat. Mater. 20, 1392- 1400 (2021). 21. Wang X, Chen J, Wang D, Mao Z. Improving the alkali metal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[205, 95, 850, 150]]<|/det|> +electrode/inorganic solid electrolyte contact via room- temperature ultrasound solid welding. Nat. Commun. 12, 7109 (2021). + +<|ref|>text<|/ref|><|det|>[[205, 168, 850, 260]]<|/det|> +Oh JAS, Sun J, Goh M, Chua B, Zeng K, Lu L. A Robust Solid- Solid Interface Using Sodium- Tin Alloy Modified Metallic Sodium Anode Paving Way for All- Solid- State Battery. Adv. Energy Mater. 11, 2101228 (2021). + +<|ref|>text<|/ref|><|det|>[[205, 279, 852, 334]]<|/det|> +Kasemchainan J, et al. Critical stripping current leads to dendrite formation on plating in lithium anode solid electrolyte cells. Nat. Mater. 18, 1105- 1111 (2019). + +<|ref|>text<|/ref|><|det|>[[205, 352, 849, 408]]<|/det|> +Ji X, et al. Solid- State Electrolyte Design for Lithium Dendrite Suppression. Adv. Mater. 32, e2002741 (2020). + +<|ref|>text<|/ref|><|det|>[[205, 426, 850, 520]]<|/det|> +Wang S, Xu H, Li W, Dolocan A, Manthiram A. Interfacial Chemistry in Solid- State Batteries: Formation of Interphase and Its Consequences. J. Am. Chem. Soc. 140, 250- 257 (2018). + +<|ref|>text<|/ref|><|det|>[[205, 538, 850, 630]]<|/det|> +Gao Z, et al. TiO2 as Second Phase in \(\mathrm{Na_3Zr_2Si_2PO_{12}}\) to Suppress Dendrite Growth in Sodium Metal Solid- State Batteries. Adv. Energy Mater. 12, 2103607 (2022). + +<|ref|>text<|/ref|><|det|>[[205, 648, 850, 742]]<|/det|> +Wang X, Chen J, Mao Z, Wang D. In situ construction of a stable interface induced by the \(\mathrm{SnS_2}\) ultra- thin layer for dendrite restriction in a solid- state sodium metal battery. J. Mater. Chem. A 9, 16039- 16045 (2021). + +<|ref|>text<|/ref|><|det|>[[205, 760, 850, 815]]<|/det|> +Wang C, et al. Grain Boundary Design of Solid Electrolyte Actualizing Stable All- Solid- State Sodium Batteries. Small 17, e2103819 (2021). + +<|ref|>text<|/ref|><|det|>[[205, 833, 850, 889]]<|/det|> +Yu X, Manthiram A. Sodium- Sulfur Batteries with a Polymer- Coated NASICON- type Sodium- Ion Solid Electrolyte. Matter 1, 439- 451 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 92, 855, 896]]<|/det|> +30. Swift MW, Qi Y. First-Principles Prediction of Potentials and Space-Charge Layers in All-Solid-State Batteries. Phys. Rev. Lett. 122, 167701 (2019). +31. Han F, et al. High electronic conductivity as the origin of lithium dendrite formation within solid electrolytes. Nat. Energy 4, 187-196 (2019). +32. He X, Bai Q, Liu Y, Nolan AM, Ling C, Mo Y. Crystal Structural Framework of Lithium Super-Ionic Conductors. Adv. Energy Mater. 9, 1902078 (2019). +33. Zhang Z, et al. Correlated Migration Invokes Higher \(\mathrm{Na^{+}}\) -Ion Conductivity in NaSICON-Type Solid Electrolytes. Adv. Energy Mater. 9, 1902373 (2019). +34. Zheng J, Tang M, Hu YY. Lithium Ion Pathway within \(\mathrm{Li_7La_3Zr_2O_{12}}\) -Polyethylene Oxide Composite Electrolytes. Angew. Chem. Int. Ed. Engl. 55, 12538-12542 (2016). +35. Hui K, et al. Yolk-shell nanoarchitecture for stabilizing a \(\mathrm{Ce_2S_3}\) anode. Carbon Energy 3, 709-720 (2021). +36. Liu J, et al. Regulating Hybrid Anodes for Efficient \(\mathrm{Li^{+} / Na^{+}}\) Storage. ACS Mater. Lett. 4, 1411-1421 (2022). +37. Zhou Q, et al. NASICON \(\mathrm{Li_{1.2}Mg_{0.1}Zr_{1.9}(PO_4)_3}\) Solid Electrolyte for an All-Solid-State Li-Metal Battery. Small Methods 4, 2000764 (2020). +38. Wang D, et al. Toward Understanding the Lithium Transport Mechanism in Garnet-type Solid Electrolytes: \(\mathrm{Li^{+}}\) Ion Exchanges and Their Mobility at Octahedral/Tetrahedral Sites. Chem. Mater. 27, 6650-6659 (2015). +39. Buschmann H, et al. Structure and dynamics of the fast lithium ion conductor " \(\mathrm{Li_7La_3Zr_2O_{12}}\) ". Phys. Chem. Chem. Phys. 13, 19378-19392 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 90, 854, 898]]<|/det|> +40. Kuhn A, Narayanan S, Spencer L, Goward G, Thangadurai V, Wilkening M. Li self-diffusion in garnet-type \(\mathrm{Li_7La_3Zr_2O_{12}}\) as probed directly by diffusion-induced Li7 spin-lattice relaxation NMR spectroscopy. Phys. Rev. B 83, (2011). +41. Düvel A, Kuhn A, Robben L, Wilkening M, Heitjans P. Mechanosynthesis of Solid Electrolytes: Preparation, Characterization, and Li Ion Transport Properties of Garnet-Type Al-Doped \(\mathrm{Li_7La_3Zr_2O_{12}}\) Crystallizing with Cubic Symmetry. J. Phys. Chem. C 116, 15192-15202 (2012). +42. Chung H, Kang B. Mechanical and Thermal Failure Induced by Contact between a \(\mathrm{Li_{1.5}Al_{0.5}Ge_{1.5}(PO_4)_3}\) Solid Electrolyte and Li Metal in an All Solid-State Li Cell. Chem. Mater. 29, 8611-8619 (2017). +43. Ahmadzadeh M, Olds TA, Scrimshire A, Bingham PA, McCloy JS. Structure and properties of \(\mathrm{Na_5FeSi_4O_{12}}\) crystallized from \(\mathrm{5Na_2O-Fe_2O_3-8SiO_2}\) glass. Acta Crystallogr., Sect. C: Struct. Chem. 74, 1595-1602 (2018). +44. Xu R, Sun H, de Vasconcelos LS, Zhao K. Mechanical and Structural Degradation of \(\mathrm{LiNi_5Mn_7Co_2O_2}\) Cathode in Li-Ion Batteries: An Experimental Study. J. Electrochem. Soc. 164, A3333 (2017). +45. Chi X, et al. An electrochemically stable homogeneous glassy electrolyte formed at room temperature for all-solid-state sodium batteries. Nat. Commun. 13, 2854 (2022). +46. Limthongkul P, Jang Y-I, Dudney NJ, Chiang Y-M. Electrochemically-driven solid-state amorphization in lithium-metal anodes. J. Power Sources 119-121, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[101, 92, 852, 895]]<|/det|> +1 604- 609 (2003). 2 47. Tang Y, Gao Y, Liu L, Zhang Y, Xie J, Zeng X. Electrochemically driven 3 amorphization of (Li)- Ti- P- O nanoparticles embedded in porous CNTs for 4 superior lithium storage performance. Chem. Eng. J. 409, 127373 (2021). 5 48. Johnson WL. Crystal- to- glass transformation in metallic materials. Mater. Sci. 6 Eng. 97, 1- 13 (1988). 7 49. Fecht HJ. Defect- induced melting and solid- state amorphization. Nature 356, 8 133- 135 (1992). 9 50. Lyapin AG, Brazhkin VV. Pressure- induced lattice instability and solid- state 10 amorphization. Phys. Rev. B 54, 12036- 12048 (1996). 11 51. Wang D, et al. \(\mathrm{Na_3V_2(PO_4)_3 / C}\) composite as the intercalation- type anode 12 material for sodium- ion batteries with superior rate capability and long- cycle 13 life. J. Mater. Chem. A 3, 8636- 8642 (2015). 14 52. Kresse G, Hafner J. Ab initio molecular dynamics for liquid metals. Phy. Rev. 15 B 47, 558- 561 (1993). 16 53. Kresse G, Hafner J. Ab initio molecular- dynamics simulation of the liquid- 17 metal- - amorphous- semiconductor transition in germanium. Phy. Rev. B 49, 18 14251- 14269 (1994). 19 54. Hou Z, Zhou R, Min Z, Lu Z, Zhang B. Realizing Wide- Temperature Reversible 20 Ca Metal Anodes through a \(\mathrm{Ca^{2 + }}\) - Conducting Artificial Layer. ACS Energy Lett. 21 274- 279 (2022). 22 55. Hou Z, et al. Correlation between Electrolyte Chemistry and Solid Electrolyte + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 92, 852, 768]]<|/det|> +1 Interphase for Reversible Ca Metal Anodes. Angew. Chem. Int. Ed. 61, 2 e202214796 (2022). 3 56. Lu Z, Ciucci F. Metal Borohydrides as Electrolytes for Solid-State Li, Na, Mg, 4 and Ca Batteries: A First-Principles Study. Chem. Mater. 29, 9308- 9319 (2017). 5 Acknowledgments 6 This work was supported by the National Natural Science Foundation of China with 7 Grant No. 12274176, 51972142, and 21974007. We also would like to thank the support 8 from the Department of Science and Technology of Jilin Province with Grant No. 9 20220201118GX and 20210301021GX. 10 Author contributions 11 G. S. and C. L. contributed equally to this work. F. D., M. T. and Z. L. designed and 12 supervised the project. G. S. performed materials synthesis, electrochemical tests, and 13 wrote the manuscript. C. L. and M. T. performed NMR experiments and analyses. Z. L. 14 performed and discussed DFT calculation results. Z. W., S. Y., G. C., and Z. S. revised 15 the manuscript. All the authors participated in the discussion and provided constructive 16 advice for the experimental design. 17 Competing interests 18 All other authors declare they have no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 330, 150]]<|/det|> +- supplementarymaterials.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/images_list.json b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..171408091de70d5aa20505ce62676f870f27d524 --- /dev/null +++ b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Structural models for polydopamine. In the centre: the first step in the polydopamine formation process consisting of (1) auto-oxidation of dopamine, (2) intramolecular cyclization, (3) oxidation and (4) isomerization. For the structures (i) – (x) see text.", + "footnote": [], + "bbox": [ + [ + 250, + 81, + 748, + 425 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 (a) Particle size distributions – as deduced from dynamic light scattering of a dopamine solution after different polymerization times with and without halloysite nanotubes (HNTs); (b) time evolution of the particle size during polydopamine growth in the presence and absence of HNTs; (c) BJH pore size distribution; (d) TGA graphs for HNTs after different exposure times to the dopamine solution and (e) time evolution of the polydopamine mass deposited on the HNT surface during polymerization. EDS elemental mapping images for HNT after (f) 1 h, (g) 4 h and (h) 24 h of polydopamine polymerization.", + "footnote": [], + "bbox": [ + [ + 122, + 85, + 884, + 775 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Solid-state NMR of PDA formed in absence and presence of HNTs. (a) Carbon-13 cross-polarization (CP) magic angle spinning NMR spectra for polydopamine formed in the absence of HNTs (PDA-A) after 24 h of polymerization time. Spectra are shown for short (80 \\(\\mu \\mathrm{s}\\) , black) and long (2ms; red) CP contact times. The former experiment shows only protonated carbons, where the latter also includes non-protonated carbons. (b) Analogous data for polydopamine deposited on the HNT surface (HNT-PDA), after 24h. The inset highlights the lack of the peak at 105 ppm in the case of the HNT-PDA sample. These spectra were recorded at 18 kHz MAS on a 600 MHz NMR instrument.", + "footnote": [], + "bbox": [ + [ + 125, + 509, + 870, + 692 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Chemical structures of the expected building blocks of polydopamine. DA and DAQ are shown in their positively charged protonated states, the expected major species in the employed solution conditions.", + "footnote": [], + "bbox": [ + [ + 295, + 81, + 690, + 392 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 High-resolution XPS spectra of the C1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.", + "footnote": [], + "bbox": [ + [ + 133, + 108, + 881, + 562 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 High-resolution XPS spectra of the N1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.", + "footnote": [], + "bbox": [ + [ + 132, + 90, + 875, + 545 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig. 7 Time dependent evolution of (a) C1s and (b) N1s components for HNT-PDA during dopamine polymerization.", + "footnote": [], + "bbox": [ + [ + 125, + 90, + 856, + 337 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8 Time evolution the concentration of the main building blocks during polydopamine film formation on halloysite nanotubes as deduced from the numerical model. For details, see text and Supplementary.", + "footnote": [], + "bbox": [ + [ + 130, + 256, + 864, + 626 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Fig. 9 Proposed model for polydopamine film formation.", + "footnote": [], + "bbox": [ + [ + 113, + 80, + 844, + 335 + ] + ], + "page_idx": 17 + } +] \ No newline at end of file diff --git a/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785.mmd b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f3042d77d9ffb6457b9d52fbf32f9933f77e7c5d --- /dev/null +++ b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785.mmd @@ -0,0 +1,274 @@ + +# New insights in polydopamine formation via surface adsorption + +Hamoon HemmatpourZernike Institute for Advanced MaterialsOreste De Luca ( oreste.deluca89@gmail.com)Zernike Institute for Advanced Materials https://orcid.org/0000- 0002- 4428- 0863Dominic CrestaniZernike Institute for Advanced Materials https://orcid.org/0000- 0002 - 2584- 503XAlessia LasorsaZernike Institute for Advanced MaterialsPatrick van der WelUniversity of Groningen https://orcid.org/0000- 0002- 5390- 3321Katja LoosUniversity of Groningen https://orcid.org/0000- 0002- 4613- 1159Theodosis GiousisUniversity of IoanninaVahid Haddadi AslAmirkabir University of TechnologyPetra RudolfZernike Institute for Advanced Materials, University of Groningen https://orcid.org/0000- 0002- 4418- 1769 + +Article + +Keywords: + +Posted Date: January 5th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1195259/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# New insights in polydopamine formation via surface adsorption + +Hamoon Hemmatpour1,2, Oreste De Luca1\*, Dominic Crestani1, Alessia Lasorsa1, Patrick C.A. van der Wel1, Katja Loos1, Theodosis Giousis1,3, Vahid Haddadi Asl2 and Petra Rudolf1\* + +1 Zernike Institute for Advanced Materials, University of Groningen, Nijenborgh 4, 9747AG Groningen, the Netherlands + +2 Department of Polymer Engineering and Color Technology, Amirkabir University of Technology, P.O. Box 1587- 4413, Tehran, Iran + +3 Department of Materials Science & Engineering, University of Ioannina, 45110, Ioannina, Greece. + +\*Corresponding authors. Email: oreste.deluca89@gmail.com, p.rudolf@rug.nl + +## Abstract + +Polydopamine is a biomimetic self- adherent polymer, which can be easily deposited on a wide variety of materials. Despite the rapidly increasing interest in polydopamine- based coatings, the polymerization mechanism and the key intermediate species formed during the deposition process are still controversial. Herein, we report a systematic investigation of polydopamine formation on halloysite nanotubes; the negative charge and high surface area of halloysite nanotubes favour the capture of intermediates that are involved in polydopamine formation and decelerate the kinetics of the process, to unravel the various polymerization steps. Data from X- ray photoelectron and solid- state nuclear magnetic resonance spectroscopies demonstrate that in the initial stage of polydopamine deposition, oxidative coupling reaction of the dopaminechrome molecules is the main reaction pathway that leads to formation of polycatecholamine oligomers as an intermediate and the post cyclization of the linear oligomers occurs subsequently. Furthermore, Tris molecules are incorporated into the initially formed oligomers. + +<--- Page Split ---> + +Natural phenomena have often served as an inspiration for designing new synthetic materials. In recent years, the extraordinary ability of the marine mussels to attach to virtually all types of inorganic and organic surfaces has attracted particular attention \(^{1}\) . Clues for such universal adhesion ability lie in the high content of catechol (L- 3,4- dihydroxyphenylalanine, L- Dopa) as well as primary and secondary amines (lysine and histidine) in the mussel's adhesive proteins, secreted at the mussel- substrate interface \(^{2}\) . Inspired by this, Messersmith et al. demonstrated that dopamine, an analogue of L- Dopa, is prone to undergo self- polymerization to form a thin, surface- adherent polydopamine film (PDA) on a vast variety of materials \(^{3}\) . The simplest protocol for coating an object with PDA involves immersing it in an alkaline solution of dopamine and waiting for a PDA layer to form spontaneously on the surface and reach a thickness of typically \(1 - 100 \text{nm}^4\) . PDA coatings therefore overcome the limitations of previous surface modification methods that require specific substrates or harsh chemical conditions \(^{5}\) . These unique features of a PDA coating in combination with its high biocompatibility and post- functionalization possibilities have triggered an exponentially growing interest for a broad range of applications, including energy storage, environmental remediation cell encapsulation and drug delivery \(^{6,7}\) . + +Understanding how the PDA coating is formed and what is ultimately its structure, is essential for optimizing its performance for tailored applications. However, its insolubility in nearly all aqueous and organic solvents complicates the structural evaluation of PDA \(^{8}\) . In addition, in most cases, the formation of a PDA film on a surface is associated with the formation of PDA particles in solution and it has been shown that the structure of the film formed on the surface agrees with a process in which particles form in the solution and then adsorb on the surface \(^{9}\) . These mechanistic complexities as well as the observation that the structural characteristics of PDA films depend on the initial dopamine concentration as well as the type of the oxidizing agent used, have led to reports of diverse structures for PDA in literature \(^{10}\) . The first step of the PDA formation pathway, which comprises oxidation, intermolecular cyclization and isomerization reactions, has been clearly described by a variety of studies (central circle in Fig. 1) \(^{11}\) . However, the following step/s leading to PDA formation are still controversial. The proposed structural models for PDA can be divided in three main categories, namely polymeric, physical and trimer- based models, as sketched in Fig. 1. Considering the polymeric models, two primary opposite models were proposed: (a) the “Eumelanin model”, which predicts that PDA formation mimics the synthetic pathway of melanin pigments in living organisms, and where the 5,6- dihydroxyindole (DHI) unit is considered the main building block of PDA (Fig. 1 (i)) \(^{3}\) , and (b) the “open- chain poly(catechol/quionine) model” that views PDA as a linear sequence of catecholamine units bonded through biphenyl- type bonds (Fig. 1(ii)) \(^{12}\) . In parallel to these two models, Della Vecchia et al. \(^{13}\) have proposed a three- component structure of PDA, which comprises uncycled (catecholamine) and cyclized (indole) units, as well as pyrrolecarboxylic acid (PDCA) moieties with covalent incorporation of trisaminoethane (Tris) (Fig. 1 (iii)). Along similar lines, a eumelanin- like polymer chain consisting of DHI units with different degrees of saturation and open- chain dopamine units has also been suggested + +<--- Page Split ---> + +(Fig. 1(iv))14. Moreover, Delparastan et al.8, provided evidence for the high-molecular weight polymeric nature of PDA films wherein the subunits are covalently connected. In contrast to the above studies supporting the existence of covalent bonds between PDA constituents, several researchers have proposed a supramolecular structure for PDA, assembled by physical interactions rather than covalent bonds. For example, using HPLC and NMR analysis Hong et al.15, have identified a physical trimer of (dopamine)2/DHI formed via a self-assembly and stabilized by van der Waals interactions and hydrogen bonds (Fig. 1(v)). Using solid-state spectroscopic techniques, Dreyer et al.16 have argued that PDA is a supramolecular aggregate consisting of 5,6-dihydroxyindoline and its quinone derivate, held together by hydrogen-bonding, π- π interaction and charge transfer interactions (Fig. 1(vi)). This view has been supported by Chan17, who investigated the PDA formation process by mass spectrometry coupled with isotope-labelling techniques, and provided evidence for non-covalently bound supramolecular aggregates of dopamine and a cyclized intermediate, dopaminechrome (DAC) being the major components of PDA (Fig. 1(vii)). However, most of the studies published in recent years argued that PDA is mainly composed of oligomers, mainly trimers, formed by covalent coupling oxidation, which bind via non-covalent interactions. Reported observations to support this view include the study performed by Ding et al.18, who, based on high-resolution mass spectroscopy, has proposed a covalently-bonded trimer of (DHI)2/PDCA, which links up through non-covalent interactions to build the supramolecular structure of PDA (Fig. 1(viii)). Alfieri et al.9 have suggested a polycyclic complex as the main constituent involved in PDA formation rather than DHI-based oligomers (Fig. 1(ix)) and Lyu et al.19, also based on a mass spectroscopy study of PDA formation, have proposed a molecular structure resulting from a complex interplay between dopaminechrome and dopamine units, as the major component of the PDA structure (Fig. 1(x)). + +As a consequence of all these discussions, a unified and unambiguous picture of PDA formation has still not emerged in the literature. In addition, most of the previously proposed models for PDA are based on experimental data from mass- spectroscopy methods that are inherently inadequate to discern the chemical structures with similar molecular weight but different in the functional groups. Accordingly, an investigation of which functional groups are formed and how they change during dopamine polymerization is essential to gain a better understanding of the PDA formation mechanism. Using X- ray photoelectron spectroscopy, Houang et al.20, have shown that the functional groups in the PDA film deposited on mica change during the first 5 min of deposition, but remain constant for deposition times \(>5\) min. A similarly fast kinetic behaviour was also observed by Zangmeister et al.21, and Rella et al.22, when they evaluated the surface chemical composition of a PDA film deposited on gold as a function of deposition time. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Structural models for polydopamine. In the centre: the first step in the polydopamine formation process consisting of (1) auto-oxidation of dopamine, (2) intramolecular cyclization, (3) oxidation and (4) isomerization. For the structures (i) – (x) see text.
+ +The aim of this study is to evaluate the kinetics of PDA formation on the surface of a nanometer- sized object because by mixing nano- objects with a large amount of adsorption sites into the alkaline solution of dopamine, the adsorption of intermediates species can be enhanced allowing for a better identification. We investigated PDA formation on halloysite nanotubes (HNTs), a low- cost aluminosilicate mineral with chemical structure of \(\mathrm{Si}_2\mathrm{Al}_2\mathrm{O}_5(\mathrm{OH})_4\cdot \mathrm{nH}_2\mathrm{O}\) and a special multi- walled structure where the outer surface is composed of negatively charged tetrahedral \(\mathrm{SiO}_2\) , while the octahedral \(\mathrm{Al}_2\mathrm{O}_3\) at the inner surface is positively charged in aqueous solution23. The evolution of the PDA layer deposited on the HNTs was monitored by X- ray photoelectron (XPS) and solid state nuclear magnetic resonance (ssNMR) spectroscopies, nitrogen adsorption/desorption porosimetry and energy dispersive X- ray microanalysis (EDS). Our results show that the adsorption of the intermediates of dopamine polymerization on HNT results in a deceleration of the kinetics of PDA formation, which enables the characterization of the intermediates of the reaction. From these data a mechanism for PDA formation is deduced in which polycatecholamine is an intermediate species. + +## Results and discussion + +Time evolution of polydopamine formation on halloysite nanotubes. The exact properties of halloysite nanotubes depend on the specific mineral deposit from which they were mined23. + +<--- Page Split ---> + +A detailed characterization of the pristine HNTs used in this study can be found in the Supplementary Fig. 1. To gain insight on the dopamine adsorption on HNTs, an experiment was performed using dynamic light scattering (DLS) measurments where the particle size distribution of HNTs suspension was examined before and after adding dopamine. Since trisaminomethane (Tris, see Fig. 4) was not present in this experiment, no dopamine polymerization was expected. The particle size distributions obtained are shown in Supplementary Fig. 2. HNTs alone show a relatively broad size distribution centered at \(320~\mathrm{nm}\) ; after adding dopamine to the solution, the particle size distribution shifts to larger values centred at \(1200~\mathrm{nm}\) . This increase reflects dopamine adsorption on the HNT surface as confirmed by the presence of nitrogen from XPS analysis (see Supplementary Fig. 3). Indeed, the negative surface charge of HNTs, their relatively high surface area and large pore volume, favour adsorption of different compounds ranging from drugs to dye molecules, especially positively charged ones24. Considering the \(\mathsf{pK}_{\mathsf{a}}\) values of dopamine \((\mathsf{pK}_{\mathsf{a}1} = 8.9\pm 0.1\) and \(\mathsf{pK}_{\mathsf{a}2} = 10.5\pm 0.1)^{25}\) and the pH of the solution ( \(\mathsf{pH} = 8.5\) ), we can deduce that part of the dopamine molecules are protonated and positively charged. This suggests that the electrostatic interactions between protonated dopamine and the negatively charged HNT surface have a leading role in the adsorption. Dopamine adsorption results in the compensation of the negative surface charge of the HNTs. Consequently, HNTs form larger aggregates because the repulsive interaction between the nanotubes is reduced. + +In the next step of our investigation, the impacts of the HNTs on the kinetics of polydopamine formation was studied by performing time dependent DLS measurements. In the first set of experiments, the auto- oxidation of a dopamine solution ( \(10~\mathrm{mM}\) dopamine, \(10~\mathrm{mM}\) Tris at pH 8.5) in the presence of HNTs at a concentration of \(0.2~\mathrm{mg / ml}\) was examined. Then the experiment was repeated without adding HNTs in order to establish if the presence of HNTs interferes with the polydopamine particle growth. In both cases a gradual darkening of the solution during the reaction was observed, followed by the precipitation of a black insoluble material. However, the darkening proceeded faster in presence of HNTs, suggesting a higher nucleation rate for polydopamine growth. Fig. 2a shows the particle size distribution in the dopamine solution after different reaction times for both experiments. At all reaction times, including the initial phase, a monomodal distribution was observed, which gradually shifts towards larger sizes as time increases. The absence of a bimodal distribution implies that the polydopamine particles grow uniformly all the time and do not coexist with oligomers formed in the solution at the initial stages of the reaction. This monomodal distribution is particularly remarkable in the case where HNTs are present because it also indicates that only one type of particles is present. In other words, we do not have a combination of HNT particles (coated with dopamine) and separate dopamine- derived particles free in solution. Instead the data suggest primarily growth of a polydopamine layer on the nanotubes. The time evolution of the particle size shows that in the absence of HNTs, the dimension of the polydopamine particles is about \(180~\mathrm{nm}\) after 10 minutes of reaction, and gradually increases to \(1500~\mathrm{nm}\) after \(3\mathrm{~h}\) . When HNTs are present, the particle size is \(1500~\mathrm{nm}\) after 10 minutes of reaction and reaches \(3000~\mathrm{nm}\) after \(3\mathrm{~h}\) (see Fig. 2b). The small discrepancy observed between Fig.2a + +<--- Page Split ---> + +and Fig.2b is due to the fact that the data shown in Fig.2b represent the average particle size obtained from triplicated measurements, while data in Fig.2a is based on a single measurement. These findings suggest that the formation of free- floating oligomers, which has been observed in the initial phase of dopamine polymerization, can be suppressed in the presence of HNTs. From these results we infer that the oligomers and intermediate species, which are involved in the polydopamine formation, are adsorbed by HNTs. Accordingly, the increase of the average particle size observed during the dopamine polymerization experiment in the presence of HNTs can be the result of nanotube aggregation as a consequence of the intermediate adsorption. + +The impact of dopamine adsorption and polymerization on the porosity of HNTs was evaluated by \(\mathsf{N}_2\) adsorption/desorption analysis for different reaction times; the isotherms are shown in Supplementary Fig. 4, while the Barrett, Joyner and Halenda (BJH) pore size distributions are depicted in Fig. 2c. The pore size distribution of HNTs changes significantly as the dopamine polymerization proceeds. The feature located at 10- 20 nm, attributed to the inner pores of the HNTs, gradually decreases and shifts towards smaller widths (see the arrow in Fig. 2c). This result indicates that the oligomeric and/or polymeric compounds are filling the inner pores of the HNTs as the reaction proceeds. During dopamine polymerization, the pore volume of the macro pores (>50 nm) increases, suggesting that polydopamine deposition on the HNT surface induces the formation of larger aggregates, in agreement with the DLS data discussed above. + +To quantify the amounts of organics deposited on the HNTs at different polymerization times, thermogravimetric analysis (TGA) was performed (see Fig. 2d- e). For the derivative graphs, see Supplementary Fig. 5. The gradual decrease in the residual weight after heating up to \(700^{\circ}\) C is observed for all samples in Fig. 2d. A non- linear growth of the polydopamine mass deposited on HNTs can be observed in Fig. 2e, pointing to a higher deposition rate in the initial phase of the polymerization, i.e. up to two hours of reaction, followed by slower deposition. The steady increase in polydopamine mass during polymerization suggests that longer polymerization times result in larger amounts of polydopamine deposited on the HNT surface, as reported by Mondin et al. \(^{26}\) for aluminum oxide particles. Energy dispersive X- ray spectroscopy elemental mapping analysis was employed to investigate the distribution of the deposited organic species on the individual nanotube. Fig. 2f- h shows the EDS maps acquired on a single nanotube after 3 different polymerization times, i.e. 1, 4 and 24 h. Carbon is detected all over the nanotube in all three samples and the signal becomes more intense as the polymerization time increases, suggesting that the nanotubes are more and more covered by carbon species during dopamine polymerization. + +All these observations encouraged us to investigate the chemical structure of the adsorbed species on the HNTs in more detail to better understand the mechanism of the polydopamine formation. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 (a) Particle size distributions – as deduced from dynamic light scattering of a dopamine solution after different polymerization times with and without halloysite nanotubes (HNTs); (b) time evolution of the particle size during polydopamine growth in the presence and absence of HNTs; (c) BJH pore size distribution; (d) TGA graphs for HNTs after different exposure times to the dopamine solution and (e) time evolution of the polydopamine mass deposited on the HNT surface during polymerization. EDS elemental mapping images for HNT after (f) 1 h, (g) 4 h and (h) 24 h of polydopamine polymerization.
+ +<--- Page Split ---> + +The carbon signal in the left bottom corner of (f) is due to the carbon grid that was used as support for the nanotubes. + +Solid State \(^{13}\mathrm{C}\) nuclear magnetic resonance spectroscopy. A first indication that the chemical structure of polydopamine is different when polydopamine aggregates form in solution and when the polymer is deposited on HNTs comes from carbon- 13 cross- polarization magic angle spinning nuclear magnetic resonance spectroscopy (CP/MAS \(^{13}\mathrm{C}\) - NMR). Representative spectra of polydopamine- covered HNTs (HNT- PDA) and polydopamine aggregates (PDA- A) after 24 h of polymerization are shown in Fig. 3. For the analysis of the spectra, we considered the chemical species sketched in Fig. 4 as the expected building blocks for polydopamine, based on the previously accepted proposal that dopamine forms indole/indoline- like structures upon oxidation \(^{11}\) . In Fig. 3a, the two well- resolved signals around 35 and 45 ppm in the aliphatic region are assigned to the sp \(^3\) carbons in the uncycled units \(^{27}\) , i.e. dopamine (DA) and dopamine quinone (DAQ), as well as to carbons at 2 and 3 positions of the indoline ring in dopaminechrome (DAC), its tautomeric structure (TS) and pyrrole dicarboxylic acid (PDCA), see Fig. 4. In addition, a signal at 60 ppm is clearly observed in the spectrum of PDA- A, while the same feature is less pronounced in the spectrum of HNT- PDA (see Fig. 3b). Della Vecchia et al. \(^{13}\) observed the same peak and assigned it to the aliphatic carbon of trisaminomethane, which is used to adjust the pH of the reaction. They concluded that polydopamine particles prepared in the presence of Tris might contain covalently bonded Tris molecules, especially when the initial dopamine concentration is relatively low. + +![](images/Figure_3.jpg) + +
Fig. 3 Solid-state NMR of PDA formed in absence and presence of HNTs. (a) Carbon-13 cross-polarization (CP) magic angle spinning NMR spectra for polydopamine formed in the absence of HNTs (PDA-A) after 24 h of polymerization time. Spectra are shown for short (80 \(\mu \mathrm{s}\) , black) and long (2ms; red) CP contact times. The former experiment shows only protonated carbons, where the latter also includes non-protonated carbons. (b) Analogous data for polydopamine deposited on the HNT surface (HNT-PDA), after 24h. The inset highlights the lack of the peak at 105 ppm in the case of the HNT-PDA sample. These spectra were recorded at 18 kHz MAS on a 600 MHz NMR instrument.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 Chemical structures of the expected building blocks of polydopamine. DA and DAQ are shown in their positively charged protonated states, the expected major species in the employed solution conditions.
+ +Small shoulders at \(\sim 30\) and 48 ppm are identifiable in both spectra of HNT- PDA and PDA- A, confirming that aliphatic carbons become disordered upon dopamine polymerization27. This agrees with the aliphatic carbons being present in different chemical structures like DA, DAC, TS and PDCA in the polydopamine structure. A signal at \(\sim 170\) ppm, which is assigned to the resonance of the carbon atoms in the carbonyl or carboxylic acid functional groups related to the DAC, TS, DAQ and PDCA structures13,14, would be expected and linked to the oxidative formation of the quinone structures. However, in our ssNMR spectra there is no clear signal at this chemical shift even at the longer CP contact time (2 ms) where non- protonated carbons should be detected. The absence of this peak suggests at best a low amount of this functionality in the sample. A potential confounding factor could be a high degree of mobility of the specific groups, since high motion make CP transfers less effective28. However, we deem this unlikely given that the carbonyl groups are integrated into the compounds' ring structure (Fig. 4) combined with the overall rigidity of the material as evidenced by the contact time dependence of the CP ssNMR in Fig. 3. + +The signals in the range of \(105 - 150\) ppm are assigned to aromatic species. To simplify the interpretation of these resonances, 13C CP MAS spectra of the representative samples were acquired with shorter CP contact times, i.e. 80 μs (see Fig. 3, red lines), following the study of Leibscher et al.14, and allowing the observation of only protonated carbons. As shown in Fig. 3a, four main signals at 105, 119, 130 and 146 ppm are detected in the aromatic region of the polydopamine spectrum acquired with 2 ms contact time, while only two of the four, at about 105 and 119 ppm, are observed with 80 μs contact time. This implies that the latter two signals + +<--- Page Split ---> + +belong to protonated carbons while the other two belong to quaternary or otherwise nonprotonated carbons. The downfield resonance at \(\sim 146\) ppm can be ascribed to the carbon atoms bonded to oxygen in phenyl form (see the structures of DA, TS, DHI and Tris in Fig. 4). In addition, the bridged carbon atom located at the 7a position in DAC, DHI and TS might contribute to this resonance since it is adjacent to nitrogen and a high downfield shift is commonly observed for such deshielded moieties14. Another downfield resonance observed at \(\sim 130\) ppm is attributed to the quaternary bridged carbon atom in the 3a position of the cyclized structures (DAC, DHI, TS) and to the carbon at the 3 position of the uncyclized structures (DA and DAQ)27. A shoulder at \(\sim 105\) ppm can be attributed to aromatic CH in 2 and 3 positions in DHI14,27. The observation of such a resonance confirms the presence of DHI in the polydopamine aggregates structure, implying that dehydrogenation occurs upon cyclization of the dopamine units. Finally, the signal at 119 ppm can reasonably be assigned to the carbon atoms not discussed so far, i.e. carbon located at 4 and 7 positions of the benzene ring in all the considered structures. + +All the main signals of the polydopamine spectrum are also observed in the spectrum of HNT- PDA (see Fig. 3b). However, the peak (shoulder) representative of DHI at \(\sim 105\) ppm is much less pronounced here, suggesting that less DHI is present in the polydopamine deposited on HNTs. Previous studies reported contradictory findings about the role of DHI in the polydopamine film formation. Ding et al.18 investigated the structure of polydopamine deposited on \(\mathrm{TiO_2}\) using TOF- SIMS analysis, finding two main peaks at \(\mathrm{m} / \mathrm{z} = 149\) and 402; the authors assigned the former to DHI, and the latter to the trimer complex (DHI)2/PDCA as the main component of polydopamine. However, this proposal is not supported by our findings since a) two well- resolved signals in the aliphatic region are strong evidence for \(\mathrm{sp^3}\) carbons in HNT- PDA but not present in the structure proposed by Ding et al.18; b) if the (DHI)2/PDCA trimer was the main component of HNT- PDA, we would expect an intense DHI peak at \(\sim 105\) ppm in the ssNMR spectrum. This discrepancy can be rationalized considering that mass spectroscopy cannot distinguish between two chemical structures with the same molecular weight as DAC and DHI, and hence the peak at \(\mathrm{m} / \mathrm{z} = 149\) in TOF- SIMS spectrum of polydopamine could be also ascribed to DAC. In his study by HPLC coupled with isotope- labelling techniques17, Chan provided strong evidence supporting uncycled dopamine and DAC as the main components for polydopamine, in reasonably good agreement with our observations. In addition, Alfieri et al.9 found that in contrast to dopamine, DHI cannot form a coating on quartz under the commonly used conditions for polymerization of dopamine (Tris solution at pH 8.5) and dismissed the decisive role of DHI in the polydopamine film formation. However, the structure containing pyran rings and one tertiary nitrogen proposed by these authors is contradicted by our XPS data of polydopamine deposited on HNTs discussed below, where a noticeable amount of primary amine groups is detected. This raises the question of why the DHI peak is less pronounced in the spectrum of HNT- PDA compared with that of PDA- A. The answer comes from the extremely weak basic strength of DHI molecule (pKa for the strongest base = - 6.3)29. Such a low basicity ensures that DHI molecules are neutral in the polymerization medium (pH = 8.5), and as a consequence, they cannot associate with the + +<--- Page Split ---> + +negatively charged HNTs surfaces. This can explain the low intensity of the DHI peak for HNT- PDA. Based on this interpretation, the presence of DHI molecules in HNT- PDA has been neglected in our further analysis. + +XPS analysis. In the next step of our investigation, XPS was used to study the PDA film formation kinetics by evaluating the chemical composition of the PDA deposited on the HNT surfaces as a function of the reaction time. In addition, to have a reference, a detailed XPS characterization of the PDA- A samples isolated from an aqueous alkaline dopamine solution after \(24 \text{h}\) can be found in the Supplementary Fig. 6. In parallel, to investigate the effects of the substrate dimensions on the kinetics of PDA film formation, a substrate with macroscopic dimensions \((1.5 \times 1.5 \text{cm}^2)\) and a chemical nature similar to the external surface of HNTs, namely a Si wafer covered by an oxide layer, was exposed to the dopamine polymerization medium for different times, and the obtained PDA films were also analysed by XPS. The survey spectra of HNT- PDA and \(\text{SiO}_2\text{- PDA}\) as a function of deposition time are shown in Supplementary Fig. 7. All samples exhibit the carbon and nitrogen signatures, indicating the formation of a polydopamine layer on the surface. The stoichiometric analysis was performed by collecting the detailed core level spectra of the constituent elements of HNTs, \(\text{SiO}_2\) and polydopamine, i.e. \(\text{Al}_2\text{p}\) , \(\text{Si}_2\text{p}\) , \(\text{O}_1\text{s}\) , \(\text{C}_1\text{s}\) and \(\text{N}_1\text{s}\) , and deducing the corresponding atomic percentages in the probed volume in order to evaluate the evolution of the elemental composition as a function of dopamine polymerization time (Supplementary Table 1). For all samples, the atomic percentage of Al and/or Si decreases continuously with increasing deposition time, while the atomic percentages of nitrogen and carbon rise, indicating PDA film growth on the surface. In addition, the \(\text{N/C}\) atomic ratio \((0.10 \pm 0.02)\) in all samples is in agreement with the value of polydopamine \((\text{N/C} = 0.10)\) , confirming the presence of PDA on the surface. + +Typical detailed XPS spectra of the C1s core level region for HNT- PDA samples are shown in Fig. 5. The same chemical species as observed in the C1s spectrum of PDA- A (see Supplementary Fig. 6b) can be identified in the C1s spectra of HNT- PDA samples, namely C- C/C=C (red, observed at a binding energy (BE) of \(284.8 \text{eV}\) ), C- O/C- N (blue, \(\text{BE} = 286.1 \text{eV}\) ), C=O/C=N (green, \(\text{BE} = 287.5 \text{eV}\) ), O=C- O (pink, \(\text{BE} = 289.2 \text{eV}\) ) bonds and shake- up component at \(\text{BE}\) of \(291.3 \text{eV}^{18,30,31}\) . + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 High-resolution XPS spectra of the C1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.
+ +The spectrum of the N1s core level region changes significantly as the polymerization proceeds (see Fig. 6). In the initial phase, the N1s peak can be fitted with two components, attributed to \(\mathrm{R - NH_3^+}\) (marked in orange in Fig. 6, \(\mathrm{BE} = 402.5 \mathrm{eV}\) ) and \(\mathrm{R - NH - R}\) (purple, \(\mathrm{BE} = 400.1 \mathrm{eV}\) ) moieties \(^{18,32}\) . Considering the building blocks shown in Fig. 4, a primary amine component points out the presence of both DA and DAQ, while secondary amines feature in DAC and PDCA. However, as the reaction proceeds, a third component appears at \(\mathrm{BE} = 398.8 \mathrm{eV}\) , which is the characteristic BE of an imine structure (green) \(^{30}\) . + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6 High-resolution XPS spectra of the N1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.
+ +Fig. 7a- b shows the evolution of the relative percentages of carbon and nitrogen species during the polymerization as deduced from their photoemission intensities. It is worth noting that the ratio between C- C/C=C and C- O/C- N components is below 1 at the beginning of the reaction, i.e. after 5 min and 15 min of reaction time, while it stabilizes at 1 as the polymerization proceeds (see Fig. 7a). This finding is somewhat surprising; indeed, considering only the structures originating from dopamine, i.e. DAQ, DAC, TS and PDCA, this ratio should not be less than 1. Its lower value is therefore a robust indication for the presence of another molecular structure with alcohol or amine functional groups in the initial phases of the polymerization, namely the Tris molecule. Della Vecchia et al. reported similar observations13,33, identifying incorporation of Tris into the polydopamine structure obtained from an aged alkaline solution of dopamine after 24 h of reaction. The most striking result that emerges from the curve fitting of the N1s spectra is that in the initial phase of polymerization, primary amine is the main component, while the secondary amine contribution becomes more pronounced with increasing reaction time (see Fig. 7b). + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Fig. 7 Time dependent evolution of (a) C1s and (b) N1s components for HNT-PDA during dopamine polymerization.
+ +The prominent primary amine component confirms the presence of intact Tris units at the beginning of the reaction, which corroborates the interpretation of the C1s spectra. In addition, as the secondary amine component becomes more important, the imine feature starts to be noticeable in the N1s spectra and its intensity continuously increases as the polymerization proceeds. These findings led us to conclude that the imine component is attributed to TS, which is a tautomer of DAC. It should be noted that the spectra of the C1s and N1s core level regions for HNT- PDA samples after 240 min of reaction are identical to those of PDA- A samples, suggesting that the surface charge of HNTs does not affect the distribution of functional groups in the PDA film deposited on the surface. Based on the above results, we can infer that the main building blocks of the polydopamine film are DA, DAQ, DAC, PDCA, TS and Tris. Surprisingly, in contrast to HNTs, after deconvolution of the N1s and C1s spectra for \(\mathrm{SiO_2}\) - PDA samples, no significant changes in the relative amounts of nitrogen and carbon species were observed during the reaction; all samples show relative percentages comparable to PDA aggregates (Supplementary Fig. 8 and Table 2). As already mentioned in the introduction, Huang et al. \(^{20}\) systematically monitored the growth of a PDA film on a mica surface and observed an evolution of the N1s and C1s spectra as a function of reaction time close to ours for HNT- PDA; however, in their case 60- 300 s were sufficient for the formation of a complete PDA layer with thickness ranging between 0.5 and 1.1 nm. In the case of our oxidized Si substrate, we do not observe this evolution. We suggest that this is due to the much lower number of adsorption sites on the small \(\mathrm{SiO_2}\) surface \((2.25 \times 10^{- 4} \mathrm{~m}^2)\) as compared to the much larger surface \((18.6 \mathrm{~m}^2\) for 0.4 g HNT) present in the solution when HNTs are added: very few of the intermediates and oligomers associated with the PDA aggregation can be adsorbed on the oxidized Si surface, while the majority of them form PDA aggregates in solution, which consequently deposit on the substrate. However, when HNTs are present, the + +<--- Page Split ---> + +amount of the available adsorption sites is so large that intermediates and oligomers adsorb on the nanotubes to such an extent that we can no longer detect PDA aggregates in solution. On the HNT surface we can therefore follow the PDA polymerization and unravel the mechanism. + +Towards a rational model for the formation of polydopamine. To determine which intermediates are formed in PDA polymerization, and therefore understand how the polymerization proceeds, an empirical model was developed based on the analysis of the XPS data. The relative percentage of each building block was calculated by using a numerical method. Further details on the model development and the calculations are presented in the Supplementary information. + +Fig. 8 shows the relative percentage of each building block during the formation of the polydopamine film. In Fig. 8b, one sees a high amount of dopamine quinone present after 5 min of polymerization; this is expected according to the commonly accepted mechanism for polydopamine formation, where the first step is the auto- oxidation of dopamine (DA) to dopamine quinone (DAQ) in an alkaline solution11. However, as the reaction proceeds, the amount of DAQ decreases, suggesting that this molecule is consumed during polymerization. In addition, as illustrated in Fig. 8a, the amount of dopamine (DA) in the deposited film increases in the initial phases of the reaction, but starts to decrease after 60 min of polymerization. This mirrors the characteristic result of kinetics of intermediate species in consecutive reactions, where a certain molecule formed in the first reaction is consumed in the second one. Accordingly, we propose that the reaction responsible for increasing the dopamine concentration in the film structure is the crosslinking of quinone molecules (DAQ) resulting in the formation of polycatecholamine34. The second reaction consists then in the oxidation and cyclization of polycatecholamine leading to the creation of dopaminechrome (DAC) units, as confirmed by the increase of that building block seen in Fig. 8c. However, Fig. 8d shows an increase as a function of polymerization time also for pyrrole dicarboxylic acid (PDCA), suggesting that PDCA also originates from DAC. This is consistent with the study of Napolitano et al.35, which revealed the possibility of oxidative breakage of o- quinone rings resulting in the formation of pyrrole dicarboxylic acid species. However, the rate at which PDCA increases is less than that of DAC, indicating that only a small portion of dopaminechrome units undergoes the oxidative ring breakage. In addition, if one considers the Fig. 8e, one sees that the increase in TS during polymerization identifies this building block as a product of DAC tautomerization. Finally, Fig. 8f shows that Tris is more prominently present at the beginning of the reaction, suggesting the adsorption of the Tris units by hydrogen bonds and/or electrostatic interaction on the HNT surface. The growth of the polydopamine layer on the HNTs leads to a decrease of Tris, as the reaction proceeds. + +Overall, all these results allow us to draw a complete picture for the formation of the polydopamine film as illustrated in Fig. 9. As a first step, dopamine molecules (DA) undergo a pH- induced auto- oxidation reaction resulting in quinone molecules (DAQ) formation. Then crosslinking of DAQ molecules occurs through the formation of biphenyl bond, which leads to + +<--- Page Split ---> + +polycatecholamine as an intermediate for the film growth process. These results corroborate the findings of Burzio et al. \(^{36}\) who reported that the \(o\) - quinone groups can react through reverse dismutation with catechol groups to produce two semiquinone radicals that couple to form di- catechol crosslinks. Then uncycled units in the polycatecholamine undergo simultaneously oxidation and inter- molecular cyclization to generate dopaminechrome. In the next step, the tautomerization of DAC molecules causes the appearance of a small quantity of TS molecules in the polydopamine film. Lastly, the oxidative ring breakage reaction of DAC generates PDCA units, which can be detected in low amounts in the polydopamine film. + +![](images/Figure_8.jpg) + +
Fig. 8 Time evolution the concentration of the main building blocks during polydopamine film formation on halloysite nanotubes as deduced from the numerical model. For details, see text and Supplementary.
+ +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Fig. 9 Proposed model for polydopamine film formation.
+ +In conclusion, this study presents a systematic investigation of polydopamine film formation on the nanometer- sized surface of halloysite nanotubes. The role of 5,6- dihydroxyindole molecules in PDA film formation process was ruled out by a comparative inspection of the ssNMR spectra of PDA film formed on the HNTS and that of PDA aggregates. Interestingly, monitoring the process kinetics showed that the presence of HNTS in the dopamine polymerization medium resulted in a deceleration of the PDA film formation process because of massive adsorption of intermediates onto HNT surfaces. As a consequence of such a decelerated process, the intermediates involved in the formation of the PDA film could be identified and a complete picture for the mechanism of the process could be drawn up. Our results show that oxidative coupling of the quinone units leading to formation the polycatecholamine species is the most dominant reaction in the initial phase of the film formation process. As the reaction time proceeds, intermolecular cyclization gradually occurs in the dopamine units that are present in the polycatecholamine species. In addition, we could provide experimental evidence for the inclusion of Tris in the layer during the initial phase of the film formation process, while the presence of this molecule become less important as the reaction time proceeds. The results presented here give new insights in the PDA film formation, which can be further exploited for tuning the properties and functions of the PDA coating. + +## Methods + +Materials. Halloysite nanotubes (HNT, \(\mathrm{Al}_2\mathrm{Si}_2\mathrm{O}_5(\mathrm{OH})_4.2\mathrm{H}_2\mathrm{O}\) ), dopamine hydrochloride (DA, \(\mathrm{C}_8\mathrm{H}_{11}\mathrm{O}_2\mathrm{N}\cdot \mathrm{HCl}\) , \(\geq 98\%\) , AR) and Tris (hydroxymethyl) aminoethane (Tris, \(\mathrm{C}_4\mathrm{H}_{11}\mathrm{O}_3\mathrm{N}\) , \(\geq 99.9\%\) , AR), were purchased from Sigma Aldrich and used as received. All the solutions were prepared using Milli- Q water. + +Preparation of PDA aggregates. PDA aggregates were synthesized by adding 0.4 g of DA to \(200~\mathrm{ml}\) of \(10~\mathrm{mM}\) Tris solution at \(\mathrm{pH} = 8.5\) . The colour of the solution gradually changed from colourless to black, + +<--- Page Split ---> + +indicating the formation of PDA. After \(24h\) of stirring at ambient temperature, the black precipitate was separated by centrifugation and dried in a vacuum jar. This product is labelled as PDA- A. + +PDA coating procedure. \(0.8g\) of HNTs were added to \(200ml\) of \(10mM\) Tris aqueous solution at \(\mathsf{pH} = 8.5\) The mixture was sonicated for 30 min to assure good dispersion, then \(0.4g\) of DA was added to the suspension, followed by stirring at room temperature for different time intervals, namely 5, 15, 30, 60, 120 and \(240mm\) . The product was separated by centrifugation and washed several times with Milli- Q water to remove the unreacted reagents. Then, the obtained pellet was dried under vacuum overnight; it is denoted as HNT- PDA. In a complementary experiment, in order to investigate the effects of substrate dimensions on the PDA formation process, Si wafers covered by a native oxide layer (Siegert Wafer, thickness \(525\pm 20\mu m\) ) with dimensions of \(1.5\times 1.5cm^2\) were immersed in a \(50ml\) of a freshly prepared solution of DA \((2mg / ml)\) and Tris \((10mM)\) . The reaction mixture was then shaken at \(150rpm\) and ambient temperature using an orbital shaker incubator (VWR, Incubating Mini Shaker). After certain time intervals, the substrate was removed from the solution, rinsed several times with Milli- Q water and gently dried with an argon flow. The coated substrate is referred to as \(\mathrm{SiO_2}\) - PDA. + +Characterization. Dynamic light scattering (DLS) measurements were conducted using a Malvern Zetasizer Ultra. Stock solutions of DA \((20mg / ml)\) and Tris \((1M)\) were prepared and filtered through a \(0.2\mu m\) pore size filter to remove the non- dissolved species. \(0.1ml\) of the DA stock solution was diluted to \(1ml\) by Milli- Q water or a HNT suspension at a concentration of \(0.2mg / ml\) . Then, \(10\mu l\) of the Tris stock solution was added to start the dopamine polymerization. After a certain time of stirring, the reaction was stopped by acidifying the solution to \(\mathsf{pH} = 2\) using a 4M HCl solution. Acidified solutions were then transferred to the dust- free disposable polystyrene cuvettes for DLS measurements. It is worth note that for non- spherical particles, the particle size obtained from DLS analysis represents the size of spherical aggregates31. The mass percentage of the PDA layer deposited on the HNTs was studied through thermogravimetric analysis (TGA) by heating samples approximately \(5mg\) at a rate of \(5^{\circ}C / min\) from 25 to \(700^{\circ}C\) under \(\mathsf{N}_2\) atmosphere, using a PL Thermogravimetric analyser (Polymer laboratories, TGA 1000, UK). Porosimetry studies was carried out using a Micrometrics ASAP 2420 V2.05 instrument and the \(\mathsf{N}_2\) adsorption/desorption isotherms were acquired at - \(196^{\circ}C\) . The Barrett, Joyner and Halenda (BJH) model was applied to the \(\mathsf{N}_2\) desorption data to obtain the pore size distribution. A JEOL 2010F transmission electron microscope (TEM), operating at \(200keV\) , was utilized to investigate the structure of the nanotubes. For this HNT powder was sonicated in ethanol and the suspension dropcast on a carbon- coated gold grid. Energy dispersive X- ray spectroscopy (EDS) was performed using an EDAX Octane silicon drift detector with an accelerating voltage of \(20kV\) , in conjugation with a FEI- Philips FEG- XL30s scanning electron microscope (SEM). Solid state- nuclear magnetic resonance (ssNMR) was preformed on a Bruker AVANCE NEO 600 MHz (14.1 T) spectrometer, using a \(3.2 - mm\) EFree HCN MAS Probe from Bruker Biospin, at \(18kHz\) MAS and \(283K\) temperature.13 C ssNMR spectra were acquired using the cross- polarization (CP) pulse sequence, with \(5\mu s\) \(90^{\circ}\) carbon pulse and \(2.5\mu s\) \(90^{\circ}\) proton pulse. Two distinct experimental conditions were used for the two investigated samples: short and long CP contact time, corresponding to \(80\mu s\) and \(2ms\) . The free induction decays were recorded under high- power proton decoupling ( \(100kHz\) ) using the two- pulse phase modulation scheme (TPPM)37, with a recycle delay of \(2.5s\) and by averaging \(1k\) and \(20k\) transients for PDA- A and HNT- PDA respectively. NMR spectra were processed with NMRPipe38. Chemical shifts were referenced to \(4,4\) - dimethyl- 4- silapentane- 1- 1 sulfonic acid using the adamantane13 C chemical shifts as an external reference, as previously described39,40. A Surface Science SSX- 100 ESCA instrument with a monochromatic Al Kα X- ray source (hv = \(1486.6eV\) ) operating in a vacuum of + +<--- Page Split ---> + +\(2\times 10^{- 9}\) mbar was used to perform the X- ray photoelectron spectroscopy (XPS) analysis. Samples were prepared ex situ by pressing dried powders of PDA- A or HNT- PDA onto silver substrates (previously prepared by flattening Ag pearls (Goodfellow, silver lump AG006100, purity \(99.999\%\) , size: \(3\mathrm{mm}\) ) in a press (RHC, 30 ton pillar press). Data acquisition was performed on an area of \(1000\mu \mathrm{m}^2\) and the electron take- off angle with respect to surface normal was \(37^{\circ}\) . The energy resolution was set to 1.3 eV for both the survey spectra and the detailed spectra of the C1s and N1s core level regions; a charge neutralizer system in optimized conditions was used during the XPS measurements to compensate for charging effects. Binding energies are reported \(\pm 0.1\) eV and referenced to the \(\mathrm{Si2p}\) photoemission peak centered at a binding energy of \(102.7\mathrm{eV}^{41}\) . The detailed spectra were analysed with the help of least- squares curve- fitting program Winspec (University of Namur, Belgium). Deconvolution of the spectra included a Shirley baseline subtraction and fitting with peak profiles taken as a convolution of Gaussian and Lorentzian functions. The uncertainty in the peak intensity determination is within \(2\%\) for all core levels reported. All measurements were carried out on freshly prepared samples and on three spots in order to check for homogeneity. + +## Data availability + +The authors declare that all the relevant data supporting the findings of this study are available within the article and its Supplementary information files. Source data are provided with this paper. + +## References + +1. Ma, Y. et al. Mussel-Inspired Underwater Adhesives-from Adhesion Mechanisms to Engineering Applications: A Critical Review. Rev. Adhes. Adhes. 9, 167-188 (2020). +2. Li, X. et al. Protein-mediated bioadhesion in marine organisms: a review. Mar. Environ. Res. 170, 105409 (2021). +3. Lee, H.; Dellatore, S. M.; Miller, W. M.; Messersmith, S. B. Mussel-inspired surface chemistry for multifunctional coatings. Science 318, 426-430 (2007). +4. Ryu, J. H.; Messersmith, P. B.; Lee, H. Polydopamine surface chemistry: a decade of discovery. ACS Appl. Mater. Interfaces 10, 7523-7540 (2018). +5. Barclay, T. G.; Hegab, H. M.; Clarke, S. R.; Ginic-Markovic, M. Versatile surface modification using polydopamine and related polycatecholamines: Chemistry, structure, and applications. Adv. Mater. Interfaces 4, 1601192 (2017). +6. Lee, H. A.; Park, E.; Lee, H. Polydopamine and its derivative surface chemistry in material science: a focused review for studies at KAIST. Adv. Mater. 32, 1907505 (2020). +7. Wang, L. H. et al. A bioinspired scaffold for rapid oxygenation of cell encapsulation systems. Nat. Commun. 12, 5846 (2021). +8. Delparastan, P.; Malollari, K. G.; Lee, H.; Messersmith, P. B. Direct evidence for the polymeric nature of polydopamine. Angew. Chem. 131, 1089-1094 (2019). +9. Alfieri, M. L. et al. Structural basis of polydopamine film formation: probing 5, 6-dihydroxyindole-based eumelanin type units and the porphyrin issue. ACS Appl. Mater. Interfaces 10, 7670-7680 (2017). +10. Cheng, W. et al. Versatile polydopamine platforms: synthesis and promising applications for surface modification and advanced nanomedicine. ACS Nano 13, 8537-8565 (2019). +11. Liebscher, J. Chemistry of polydopamine–scope, variation, and limitation. Eur. J. Org. Chem. 2019, 4976-4994 (2019). + +<--- Page Split ---> + +12. Jiang, J.-H. et al. Surface modification of PE porous membranes based on the strong adhesion of polydopamine and covalent immobilization of heparin. J. Membr. Sci. 364, 194-202 (2010). + +13. Della Vecchia, N. F. et al. Building-block diversity in polydopamine underpins a multifunctional eumelanin-type platform tunable through a quinone control point. Adv. Funct. Mater. 23, 1331-1340 (2013). + +14. Liebscher, J. R. et al. Structure of polydopamine: a never-ending story? Langmuir 29, 10539-10548 (2013). + +15. Hong, S. et al. Non-covalent self-assembly and covalent polymerization co-contribute to polydopamine formation. Adv. Funct. Mater. 22, 4711-4717 (2012). + +16. Dreyer, D. R. et al. Elucidating the structure of poly (dopamine). Langmuir 28, 6428-6435 (2012). + +17. Chan, W. Investigation of the chemical structure and formation mechanism of polydopamine from self-assembly of dopamine by liquid chromatography/mass spectrometry coupled with isotope-labelling techniques. Rapid Commun. Mass Spectrom. 33, 429-436 (2019). + +18. Ding, Y. et al. Insights into the aggregation/deposition and structure of a polydopamine film. Langmuir 30, 12258-12269 (2014). + +19. Lyu, Q.; Hsueh, N.; Chai, C. L. Unravelling the polydopamine mystery: is the end in sight? Polym. Chem. 10, 5771-5777 (2019). + +20. Huang, Z.-H. et al. Polydopamine ultrathin film growth on mica via in-situ polymerization of dopamine with applications for silver-based antimicrobial coatings. Materials 14, 671 (2021). + +21. Zangmeister, R. A.; Morris, T. A.; Tarlov, M. J. Characterization of polydopamine thin films deposited at short times by autoxidation of dopamine. Langmuir 29, 8619-8628 (2013). + +22. Rella, S. et al. Investigation of polydopamine coatings by X-ray Photoelectron Spectroscopy as an effective tool for improving biomolecule conjugation. Appl. Surf. Sci. 447, 31-39 (2018). + +23. Danyluk, N.; Tomaszewska, J.; Tatarchuk, T. Halloysite nanotubes and halloysite-based composites for environmental and biomedical applications. J. Mol. Liq. 309, 113077 (2020). + +24. Anastopoulos, I. et al. A review on halloysite-based adsorbents to remove pollutants in water and wastewater. J. Mol. Liq. 269, 855-868 (2018). + +25. Sanchez-Rivera, A. et al. Spectrophotometric study on the stability of dopamine and the determination of its acidity constants. Spectrochim. Acta A 59, 3193-3203 (2003). + +26. Mondin, G. et al. Investigations of mussel-inspired polydopamine deposition on WC and Al2O3 particles: The influence of particle size and material. Mater. Chem. Phys. 148, 624-630 (2014). + +27. Circu, M.; Filip, C. Closer to the polydopamine structure: new insights from a combined 13 C/1 H/2 H solid-state NMR study on deuterated samples. Polym. Chem. 9, 3379-3387 (2018). + +28. Matlahov, I.; van der Wel, P. C. Hidden motions and motion-induced invisibility: Dynamics-based spectral editing in solid-state NMR. Methods 148, 123-135 (2018). + +29. Wishart, D.S. et al. HMDB 4.0 — The Human Metabolome Database for 2018. Nucleic Acids Res. 46, D608-17 (2018). + +30. Qin, J. et al. Achieving stable Na metal cycling via polydopamine/multilayer graphene coating of a polypropylene separator. Nat. Commun. 12, 5786 (2021). + +31. Lee, K. et al. Laser-induced graphitization of polydopamine leads to enhanced mechanical performance while preserving multifunctionality. Nat. Commun. 11, 4848 (2020). + +32. Wang, W. et al. Self-healing performance and corrosion resistance of graphene oxide–mesoporous silicon layer–nanosphere structure coating under marine alternating hydrostatic pressure. Chem. Eng. J. 361, 792-804 (2019). + +<--- Page Split ---> + +33. Della Vecchia, N. F. et al. Tris buffer modulates polydopamine growth, aggregation, and paramagnetic properties. Langmuir 30, 9811-9818 (2014). + +34. Yang, J.; Stuart, M. A. C.; Kamperman, M. Jack of all trades: versatile catechol crosslinking mechanisms. Chem. Soc. Rev. 43, 8271-8298 (2014). +35. Napolitano, A.; Pezzella, A.; d'Ischia, M.; Prota, G. New pyrrole acids by oxidative degradation of eumelanins with hydrogen peroxide. Further hints to the mechanism of pigment breakdown. Tetrahedron 52, 8775-8780 (1996). +36. Burzio, L. A.; Waite, J. H. Cross-linking in adhesive quinoproteins: studies with model decapeptides. Biochemistry 39, 11147-11153 (2000). +37. Bennett, A. et al. Heteronuclear decoupling in rotating solids. J. Chem. Phys. 103(16), 6951-6958 (1995). +38. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277-293 (1995). +39. Morcombe, C. R.; Zilm, K. W. Chemical shift referencing in MAS solid state NMR. J. Magn. Reson. 162, 479-486 (2003). +40. Harris, R. K. et al. Further conventions for NMR shielding and chemical shift (IUPAC Recommendation 2008). Magn. Reson. Chem. 46, 582-598 (2008). +41. Liu, M. et al. Properties of halloysite nanotube-epoxy resin hybrids and the interfacial reactions in the systems. Nanotechnology 18, 455703 (2007). + +## Aknowledgements + +The authors thank Jur van Dijken for assistance with the TGA measurements. H.H. acknowledges the Ubbo Emmius Programme for financially supporting his PhD project. This work benefitted from financial support by the Advanced Materials research programme of the Zernike National Research Centre under the Bonus Incentive Scheme of the Dutch Ministry for Education, Culture and Science. + +## Author contributions + +H. H. proposed, planned and executed all the experiments, analysed the data and wrote the main paper; O.D.L. contributed to the interpretation of XPS data and revised the paper; D.C. performed the main experiments with H.H.; A. L. and P. v. d. W. performed the ssNMR experiments and data analysis; T. G. contributed to the experiments; K. L., V. H. A. revised the paper; P.R. designed and supervised the experiments, and revised the paper. All authors critically reviewed and approved the paper. + +## Competing interests + +The authors declare no competing interests. + +## Additional information + +Supplementary information. The online version contains supplementary material available at: Correspondence and request for materials should be addressed to O. D. L. and P. R. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementaryinformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785_det.mmd b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..91a5d0c826df8606fb1f3bf2ec33346a509890ea --- /dev/null +++ b/preprint/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785/preprint__0e98f07b05f42867a0cab0b5a75d6066fa6bbf6e25596f6f2dfc461cea42a785_det.mmd @@ -0,0 +1,349 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 955, 175]]<|/det|> +# New insights in polydopamine formation via surface adsorption + +<|ref|>text<|/ref|><|det|>[[44, 195, 763, 636]]<|/det|> +Hamoon HemmatpourZernike Institute for Advanced MaterialsOreste De Luca ( oreste.deluca89@gmail.com)Zernike Institute for Advanced Materials https://orcid.org/0000- 0002- 4428- 0863Dominic CrestaniZernike Institute for Advanced Materials https://orcid.org/0000- 0002 - 2584- 503XAlessia LasorsaZernike Institute for Advanced MaterialsPatrick van der WelUniversity of Groningen https://orcid.org/0000- 0002- 5390- 3321Katja LoosUniversity of Groningen https://orcid.org/0000- 0002- 4613- 1159Theodosis GiousisUniversity of IoanninaVahid Haddadi AslAmirkabir University of TechnologyPetra RudolfZernike Institute for Advanced Materials, University of Groningen https://orcid.org/0000- 0002- 4418- 1769 + +<|ref|>text<|/ref|><|det|>[[44, 671, 101, 689]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 710, 135, 728]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 747, 320, 767]]<|/det|> +Posted Date: January 5th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 786, 474, 805]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1195259/v1 + +<|ref|>text<|/ref|><|det|>[[44, 823, 910, 866]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 85, 881, 147]]<|/det|> +# New insights in polydopamine formation via surface adsorption + +<|ref|>text<|/ref|><|det|>[[115, 159, 864, 197]]<|/det|> +Hamoon Hemmatpour1,2, Oreste De Luca1\*, Dominic Crestani1, Alessia Lasorsa1, Patrick C.A. van der Wel1, Katja Loos1, Theodosis Giousis1,3, Vahid Haddadi Asl2 and Petra Rudolf1\* + +<|ref|>text<|/ref|><|det|>[[115, 208, 864, 245]]<|/det|> +1 Zernike Institute for Advanced Materials, University of Groningen, Nijenborgh 4, 9747AG Groningen, the Netherlands + +<|ref|>text<|/ref|><|det|>[[115, 256, 812, 292]]<|/det|> +2 Department of Polymer Engineering and Color Technology, Amirkabir University of Technology, P.O. Box 1587- 4413, Tehran, Iran + +<|ref|>text<|/ref|><|det|>[[115, 303, 863, 338]]<|/det|> +3 Department of Materials Science & Engineering, University of Ioannina, 45110, Ioannina, Greece. + +<|ref|>text<|/ref|><|det|>[[117, 378, 757, 395]]<|/det|> +\*Corresponding authors. Email: oreste.deluca89@gmail.com, p.rudolf@rug.nl + +<|ref|>sub_title<|/ref|><|det|>[[115, 435, 191, 450]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[115, 454, 884, 712]]<|/det|> +Polydopamine is a biomimetic self- adherent polymer, which can be easily deposited on a wide variety of materials. Despite the rapidly increasing interest in polydopamine- based coatings, the polymerization mechanism and the key intermediate species formed during the deposition process are still controversial. Herein, we report a systematic investigation of polydopamine formation on halloysite nanotubes; the negative charge and high surface area of halloysite nanotubes favour the capture of intermediates that are involved in polydopamine formation and decelerate the kinetics of the process, to unravel the various polymerization steps. Data from X- ray photoelectron and solid- state nuclear magnetic resonance spectroscopies demonstrate that in the initial stage of polydopamine deposition, oxidative coupling reaction of the dopaminechrome molecules is the main reaction pathway that leads to formation of polycatecholamine oligomers as an intermediate and the post cyclization of the linear oligomers occurs subsequently. Furthermore, Tris molecules are incorporated into the initially formed oligomers. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 110, 884, 429]]<|/det|> +Natural phenomena have often served as an inspiration for designing new synthetic materials. In recent years, the extraordinary ability of the marine mussels to attach to virtually all types of inorganic and organic surfaces has attracted particular attention \(^{1}\) . Clues for such universal adhesion ability lie in the high content of catechol (L- 3,4- dihydroxyphenylalanine, L- Dopa) as well as primary and secondary amines (lysine and histidine) in the mussel's adhesive proteins, secreted at the mussel- substrate interface \(^{2}\) . Inspired by this, Messersmith et al. demonstrated that dopamine, an analogue of L- Dopa, is prone to undergo self- polymerization to form a thin, surface- adherent polydopamine film (PDA) on a vast variety of materials \(^{3}\) . The simplest protocol for coating an object with PDA involves immersing it in an alkaline solution of dopamine and waiting for a PDA layer to form spontaneously on the surface and reach a thickness of typically \(1 - 100 \text{nm}^4\) . PDA coatings therefore overcome the limitations of previous surface modification methods that require specific substrates or harsh chemical conditions \(^{5}\) . These unique features of a PDA coating in combination with its high biocompatibility and post- functionalization possibilities have triggered an exponentially growing interest for a broad range of applications, including energy storage, environmental remediation cell encapsulation and drug delivery \(^{6,7}\) . + +<|ref|>text<|/ref|><|det|>[[115, 430, 884, 909]]<|/det|> +Understanding how the PDA coating is formed and what is ultimately its structure, is essential for optimizing its performance for tailored applications. However, its insolubility in nearly all aqueous and organic solvents complicates the structural evaluation of PDA \(^{8}\) . In addition, in most cases, the formation of a PDA film on a surface is associated with the formation of PDA particles in solution and it has been shown that the structure of the film formed on the surface agrees with a process in which particles form in the solution and then adsorb on the surface \(^{9}\) . These mechanistic complexities as well as the observation that the structural characteristics of PDA films depend on the initial dopamine concentration as well as the type of the oxidizing agent used, have led to reports of diverse structures for PDA in literature \(^{10}\) . The first step of the PDA formation pathway, which comprises oxidation, intermolecular cyclization and isomerization reactions, has been clearly described by a variety of studies (central circle in Fig. 1) \(^{11}\) . However, the following step/s leading to PDA formation are still controversial. The proposed structural models for PDA can be divided in three main categories, namely polymeric, physical and trimer- based models, as sketched in Fig. 1. Considering the polymeric models, two primary opposite models were proposed: (a) the “Eumelanin model”, which predicts that PDA formation mimics the synthetic pathway of melanin pigments in living organisms, and where the 5,6- dihydroxyindole (DHI) unit is considered the main building block of PDA (Fig. 1 (i)) \(^{3}\) , and (b) the “open- chain poly(catechol/quionine) model” that views PDA as a linear sequence of catecholamine units bonded through biphenyl- type bonds (Fig. 1(ii)) \(^{12}\) . In parallel to these two models, Della Vecchia et al. \(^{13}\) have proposed a three- component structure of PDA, which comprises uncycled (catecholamine) and cyclized (indole) units, as well as pyrrolecarboxylic acid (PDCA) moieties with covalent incorporation of trisaminoethane (Tris) (Fig. 1 (iii)). Along similar lines, a eumelanin- like polymer chain consisting of DHI units with different degrees of saturation and open- chain dopamine units has also been suggested + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 80, 884, 561]]<|/det|> +(Fig. 1(iv))14. Moreover, Delparastan et al.8, provided evidence for the high-molecular weight polymeric nature of PDA films wherein the subunits are covalently connected. In contrast to the above studies supporting the existence of covalent bonds between PDA constituents, several researchers have proposed a supramolecular structure for PDA, assembled by physical interactions rather than covalent bonds. For example, using HPLC and NMR analysis Hong et al.15, have identified a physical trimer of (dopamine)2/DHI formed via a self-assembly and stabilized by van der Waals interactions and hydrogen bonds (Fig. 1(v)). Using solid-state spectroscopic techniques, Dreyer et al.16 have argued that PDA is a supramolecular aggregate consisting of 5,6-dihydroxyindoline and its quinone derivate, held together by hydrogen-bonding, π- π interaction and charge transfer interactions (Fig. 1(vi)). This view has been supported by Chan17, who investigated the PDA formation process by mass spectrometry coupled with isotope-labelling techniques, and provided evidence for non-covalently bound supramolecular aggregates of dopamine and a cyclized intermediate, dopaminechrome (DAC) being the major components of PDA (Fig. 1(vii)). However, most of the studies published in recent years argued that PDA is mainly composed of oligomers, mainly trimers, formed by covalent coupling oxidation, which bind via non-covalent interactions. Reported observations to support this view include the study performed by Ding et al.18, who, based on high-resolution mass spectroscopy, has proposed a covalently-bonded trimer of (DHI)2/PDCA, which links up through non-covalent interactions to build the supramolecular structure of PDA (Fig. 1(viii)). Alfieri et al.9 have suggested a polycyclic complex as the main constituent involved in PDA formation rather than DHI-based oligomers (Fig. 1(ix)) and Lyu et al.19, also based on a mass spectroscopy study of PDA formation, have proposed a molecular structure resulting from a complex interplay between dopaminechrome and dopamine units, as the major component of the PDA structure (Fig. 1(x)). + +<|ref|>text<|/ref|><|det|>[[115, 562, 884, 800]]<|/det|> +As a consequence of all these discussions, a unified and unambiguous picture of PDA formation has still not emerged in the literature. In addition, most of the previously proposed models for PDA are based on experimental data from mass- spectroscopy methods that are inherently inadequate to discern the chemical structures with similar molecular weight but different in the functional groups. Accordingly, an investigation of which functional groups are formed and how they change during dopamine polymerization is essential to gain a better understanding of the PDA formation mechanism. Using X- ray photoelectron spectroscopy, Houang et al.20, have shown that the functional groups in the PDA film deposited on mica change during the first 5 min of deposition, but remain constant for deposition times \(>5\) min. A similarly fast kinetic behaviour was also observed by Zangmeister et al.21, and Rella et al.22, when they evaluated the surface chemical composition of a PDA film deposited on gold as a function of deposition time. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[250, 81, 748, 425]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 439, 883, 492]]<|/det|> +
Fig. 1 Structural models for polydopamine. In the centre: the first step in the polydopamine formation process consisting of (1) auto-oxidation of dopamine, (2) intramolecular cyclization, (3) oxidation and (4) isomerization. For the structures (i) – (x) see text.
+ +<|ref|>text<|/ref|><|det|>[[115, 515, 884, 813]]<|/det|> +The aim of this study is to evaluate the kinetics of PDA formation on the surface of a nanometer- sized object because by mixing nano- objects with a large amount of adsorption sites into the alkaline solution of dopamine, the adsorption of intermediates species can be enhanced allowing for a better identification. We investigated PDA formation on halloysite nanotubes (HNTs), a low- cost aluminosilicate mineral with chemical structure of \(\mathrm{Si}_2\mathrm{Al}_2\mathrm{O}_5(\mathrm{OH})_4\cdot \mathrm{nH}_2\mathrm{O}\) and a special multi- walled structure where the outer surface is composed of negatively charged tetrahedral \(\mathrm{SiO}_2\) , while the octahedral \(\mathrm{Al}_2\mathrm{O}_3\) at the inner surface is positively charged in aqueous solution23. The evolution of the PDA layer deposited on the HNTs was monitored by X- ray photoelectron (XPS) and solid state nuclear magnetic resonance (ssNMR) spectroscopies, nitrogen adsorption/desorption porosimetry and energy dispersive X- ray microanalysis (EDS). Our results show that the adsorption of the intermediates of dopamine polymerization on HNT results in a deceleration of the kinetics of PDA formation, which enables the characterization of the intermediates of the reaction. From these data a mechanism for PDA formation is deduced in which polycatecholamine is an intermediate species. + +<|ref|>sub_title<|/ref|><|det|>[[117, 838, 306, 853]]<|/det|> +## Results and discussion + +<|ref|>text<|/ref|><|det|>[[117, 857, 884, 894]]<|/det|> +Time evolution of polydopamine formation on halloysite nanotubes. The exact properties of halloysite nanotubes depend on the specific mineral deposit from which they were mined23. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 884, 461]]<|/det|> +A detailed characterization of the pristine HNTs used in this study can be found in the Supplementary Fig. 1. To gain insight on the dopamine adsorption on HNTs, an experiment was performed using dynamic light scattering (DLS) measurments where the particle size distribution of HNTs suspension was examined before and after adding dopamine. Since trisaminomethane (Tris, see Fig. 4) was not present in this experiment, no dopamine polymerization was expected. The particle size distributions obtained are shown in Supplementary Fig. 2. HNTs alone show a relatively broad size distribution centered at \(320~\mathrm{nm}\) ; after adding dopamine to the solution, the particle size distribution shifts to larger values centred at \(1200~\mathrm{nm}\) . This increase reflects dopamine adsorption on the HNT surface as confirmed by the presence of nitrogen from XPS analysis (see Supplementary Fig. 3). Indeed, the negative surface charge of HNTs, their relatively high surface area and large pore volume, favour adsorption of different compounds ranging from drugs to dye molecules, especially positively charged ones24. Considering the \(\mathsf{pK}_{\mathsf{a}}\) values of dopamine \((\mathsf{pK}_{\mathsf{a}1} = 8.9\pm 0.1\) and \(\mathsf{pK}_{\mathsf{a}2} = 10.5\pm 0.1)^{25}\) and the pH of the solution ( \(\mathsf{pH} = 8.5\) ), we can deduce that part of the dopamine molecules are protonated and positively charged. This suggests that the electrostatic interactions between protonated dopamine and the negatively charged HNT surface have a leading role in the adsorption. Dopamine adsorption results in the compensation of the negative surface charge of the HNTs. Consequently, HNTs form larger aggregates because the repulsive interaction between the nanotubes is reduced. + +<|ref|>text<|/ref|><|det|>[[115, 464, 884, 902]]<|/det|> +In the next step of our investigation, the impacts of the HNTs on the kinetics of polydopamine formation was studied by performing time dependent DLS measurements. In the first set of experiments, the auto- oxidation of a dopamine solution ( \(10~\mathrm{mM}\) dopamine, \(10~\mathrm{mM}\) Tris at pH 8.5) in the presence of HNTs at a concentration of \(0.2~\mathrm{mg / ml}\) was examined. Then the experiment was repeated without adding HNTs in order to establish if the presence of HNTs interferes with the polydopamine particle growth. In both cases a gradual darkening of the solution during the reaction was observed, followed by the precipitation of a black insoluble material. However, the darkening proceeded faster in presence of HNTs, suggesting a higher nucleation rate for polydopamine growth. Fig. 2a shows the particle size distribution in the dopamine solution after different reaction times for both experiments. At all reaction times, including the initial phase, a monomodal distribution was observed, which gradually shifts towards larger sizes as time increases. The absence of a bimodal distribution implies that the polydopamine particles grow uniformly all the time and do not coexist with oligomers formed in the solution at the initial stages of the reaction. This monomodal distribution is particularly remarkable in the case where HNTs are present because it also indicates that only one type of particles is present. In other words, we do not have a combination of HNT particles (coated with dopamine) and separate dopamine- derived particles free in solution. Instead the data suggest primarily growth of a polydopamine layer on the nanotubes. The time evolution of the particle size shows that in the absence of HNTs, the dimension of the polydopamine particles is about \(180~\mathrm{nm}\) after 10 minutes of reaction, and gradually increases to \(1500~\mathrm{nm}\) after \(3\mathrm{~h}\) . When HNTs are present, the particle size is \(1500~\mathrm{nm}\) after 10 minutes of reaction and reaches \(3000~\mathrm{nm}\) after \(3\mathrm{~h}\) (see Fig. 2b). The small discrepancy observed between Fig.2a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 884, 261]]<|/det|> +and Fig.2b is due to the fact that the data shown in Fig.2b represent the average particle size obtained from triplicated measurements, while data in Fig.2a is based on a single measurement. These findings suggest that the formation of free- floating oligomers, which has been observed in the initial phase of dopamine polymerization, can be suppressed in the presence of HNTs. From these results we infer that the oligomers and intermediate species, which are involved in the polydopamine formation, are adsorbed by HNTs. Accordingly, the increase of the average particle size observed during the dopamine polymerization experiment in the presence of HNTs can be the result of nanotube aggregation as a consequence of the intermediate adsorption. + +<|ref|>text<|/ref|><|det|>[[115, 263, 884, 480]]<|/det|> +The impact of dopamine adsorption and polymerization on the porosity of HNTs was evaluated by \(\mathsf{N}_2\) adsorption/desorption analysis for different reaction times; the isotherms are shown in Supplementary Fig. 4, while the Barrett, Joyner and Halenda (BJH) pore size distributions are depicted in Fig. 2c. The pore size distribution of HNTs changes significantly as the dopamine polymerization proceeds. The feature located at 10- 20 nm, attributed to the inner pores of the HNTs, gradually decreases and shifts towards smaller widths (see the arrow in Fig. 2c). This result indicates that the oligomeric and/or polymeric compounds are filling the inner pores of the HNTs as the reaction proceeds. During dopamine polymerization, the pore volume of the macro pores (>50 nm) increases, suggesting that polydopamine deposition on the HNT surface induces the formation of larger aggregates, in agreement with the DLS data discussed above. + +<|ref|>text<|/ref|><|det|>[[115, 483, 884, 780]]<|/det|> +To quantify the amounts of organics deposited on the HNTs at different polymerization times, thermogravimetric analysis (TGA) was performed (see Fig. 2d- e). For the derivative graphs, see Supplementary Fig. 5. The gradual decrease in the residual weight after heating up to \(700^{\circ}\) C is observed for all samples in Fig. 2d. A non- linear growth of the polydopamine mass deposited on HNTs can be observed in Fig. 2e, pointing to a higher deposition rate in the initial phase of the polymerization, i.e. up to two hours of reaction, followed by slower deposition. The steady increase in polydopamine mass during polymerization suggests that longer polymerization times result in larger amounts of polydopamine deposited on the HNT surface, as reported by Mondin et al. \(^{26}\) for aluminum oxide particles. Energy dispersive X- ray spectroscopy elemental mapping analysis was employed to investigate the distribution of the deposited organic species on the individual nanotube. Fig. 2f- h shows the EDS maps acquired on a single nanotube after 3 different polymerization times, i.e. 1, 4 and 24 h. Carbon is detected all over the nanotube in all three samples and the signal becomes more intense as the polymerization time increases, suggesting that the nanotubes are more and more covered by carbon species during dopamine polymerization. + +<|ref|>text<|/ref|><|det|>[[116, 783, 882, 840]]<|/det|> +All these observations encouraged us to investigate the chemical structure of the adsorbed species on the HNTs in more detail to better understand the mechanism of the polydopamine formation. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[122, 85, 884, 775]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 779, 884, 888]]<|/det|> +
Fig. 2 (a) Particle size distributions – as deduced from dynamic light scattering of a dopamine solution after different polymerization times with and without halloysite nanotubes (HNTs); (b) time evolution of the particle size during polydopamine growth in the presence and absence of HNTs; (c) BJH pore size distribution; (d) TGA graphs for HNTs after different exposure times to the dopamine solution and (e) time evolution of the polydopamine mass deposited on the HNT surface during polymerization. EDS elemental mapping images for HNT after (f) 1 h, (g) 4 h and (h) 24 h of polydopamine polymerization.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 117]]<|/det|> +The carbon signal in the left bottom corner of (f) is due to the carbon grid that was used as support for the nanotubes. + +<|ref|>text<|/ref|><|det|>[[114, 128, 884, 488]]<|/det|> +Solid State \(^{13}\mathrm{C}\) nuclear magnetic resonance spectroscopy. A first indication that the chemical structure of polydopamine is different when polydopamine aggregates form in solution and when the polymer is deposited on HNTs comes from carbon- 13 cross- polarization magic angle spinning nuclear magnetic resonance spectroscopy (CP/MAS \(^{13}\mathrm{C}\) - NMR). Representative spectra of polydopamine- covered HNTs (HNT- PDA) and polydopamine aggregates (PDA- A) after 24 h of polymerization are shown in Fig. 3. For the analysis of the spectra, we considered the chemical species sketched in Fig. 4 as the expected building blocks for polydopamine, based on the previously accepted proposal that dopamine forms indole/indoline- like structures upon oxidation \(^{11}\) . In Fig. 3a, the two well- resolved signals around 35 and 45 ppm in the aliphatic region are assigned to the sp \(^3\) carbons in the uncycled units \(^{27}\) , i.e. dopamine (DA) and dopamine quinone (DAQ), as well as to carbons at 2 and 3 positions of the indoline ring in dopaminechrome (DAC), its tautomeric structure (TS) and pyrrole dicarboxylic acid (PDCA), see Fig. 4. In addition, a signal at 60 ppm is clearly observed in the spectrum of PDA- A, while the same feature is less pronounced in the spectrum of HNT- PDA (see Fig. 3b). Della Vecchia et al. \(^{13}\) observed the same peak and assigned it to the aliphatic carbon of trisaminomethane, which is used to adjust the pH of the reaction. They concluded that polydopamine particles prepared in the presence of Tris might contain covalently bonded Tris molecules, especially when the initial dopamine concentration is relatively low. + +<|ref|>image<|/ref|><|det|>[[125, 509, 870, 692]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 708, 884, 835]]<|/det|> +
Fig. 3 Solid-state NMR of PDA formed in absence and presence of HNTs. (a) Carbon-13 cross-polarization (CP) magic angle spinning NMR spectra for polydopamine formed in the absence of HNTs (PDA-A) after 24 h of polymerization time. Spectra are shown for short (80 \(\mu \mathrm{s}\) , black) and long (2ms; red) CP contact times. The former experiment shows only protonated carbons, where the latter also includes non-protonated carbons. (b) Analogous data for polydopamine deposited on the HNT surface (HNT-PDA), after 24h. The inset highlights the lack of the peak at 105 ppm in the case of the HNT-PDA sample. These spectra were recorded at 18 kHz MAS on a 600 MHz NMR instrument.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[295, 81, 690, 392]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 409, 883, 463]]<|/det|> +
Fig. 4 Chemical structures of the expected building blocks of polydopamine. DA and DAQ are shown in their positively charged protonated states, the expected major species in the employed solution conditions.
+ +<|ref|>text<|/ref|><|det|>[[115, 487, 884, 765]]<|/det|> +Small shoulders at \(\sim 30\) and 48 ppm are identifiable in both spectra of HNT- PDA and PDA- A, confirming that aliphatic carbons become disordered upon dopamine polymerization27. This agrees with the aliphatic carbons being present in different chemical structures like DA, DAC, TS and PDCA in the polydopamine structure. A signal at \(\sim 170\) ppm, which is assigned to the resonance of the carbon atoms in the carbonyl or carboxylic acid functional groups related to the DAC, TS, DAQ and PDCA structures13,14, would be expected and linked to the oxidative formation of the quinone structures. However, in our ssNMR spectra there is no clear signal at this chemical shift even at the longer CP contact time (2 ms) where non- protonated carbons should be detected. The absence of this peak suggests at best a low amount of this functionality in the sample. A potential confounding factor could be a high degree of mobility of the specific groups, since high motion make CP transfers less effective28. However, we deem this unlikely given that the carbonyl groups are integrated into the compounds' ring structure (Fig. 4) combined with the overall rigidity of the material as evidenced by the contact time dependence of the CP ssNMR in Fig. 3. + +<|ref|>text<|/ref|><|det|>[[115, 767, 884, 905]]<|/det|> +The signals in the range of \(105 - 150\) ppm are assigned to aromatic species. To simplify the interpretation of these resonances, 13C CP MAS spectra of the representative samples were acquired with shorter CP contact times, i.e. 80 μs (see Fig. 3, red lines), following the study of Leibscher et al.14, and allowing the observation of only protonated carbons. As shown in Fig. 3a, four main signals at 105, 119, 130 and 146 ppm are detected in the aromatic region of the polydopamine spectrum acquired with 2 ms contact time, while only two of the four, at about 105 and 119 ppm, are observed with 80 μs contact time. This implies that the latter two signals + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 883, 360]]<|/det|> +belong to protonated carbons while the other two belong to quaternary or otherwise nonprotonated carbons. The downfield resonance at \(\sim 146\) ppm can be ascribed to the carbon atoms bonded to oxygen in phenyl form (see the structures of DA, TS, DHI and Tris in Fig. 4). In addition, the bridged carbon atom located at the 7a position in DAC, DHI and TS might contribute to this resonance since it is adjacent to nitrogen and a high downfield shift is commonly observed for such deshielded moieties14. Another downfield resonance observed at \(\sim 130\) ppm is attributed to the quaternary bridged carbon atom in the 3a position of the cyclized structures (DAC, DHI, TS) and to the carbon at the 3 position of the uncyclized structures (DA and DAQ)27. A shoulder at \(\sim 105\) ppm can be attributed to aromatic CH in 2 and 3 positions in DHI14,27. The observation of such a resonance confirms the presence of DHI in the polydopamine aggregates structure, implying that dehydrogenation occurs upon cyclization of the dopamine units. Finally, the signal at 119 ppm can reasonably be assigned to the carbon atoms not discussed so far, i.e. carbon located at 4 and 7 positions of the benzene ring in all the considered structures. + +<|ref|>text<|/ref|><|det|>[[115, 363, 883, 902]]<|/det|> +All the main signals of the polydopamine spectrum are also observed in the spectrum of HNT- PDA (see Fig. 3b). However, the peak (shoulder) representative of DHI at \(\sim 105\) ppm is much less pronounced here, suggesting that less DHI is present in the polydopamine deposited on HNTs. Previous studies reported contradictory findings about the role of DHI in the polydopamine film formation. Ding et al.18 investigated the structure of polydopamine deposited on \(\mathrm{TiO_2}\) using TOF- SIMS analysis, finding two main peaks at \(\mathrm{m} / \mathrm{z} = 149\) and 402; the authors assigned the former to DHI, and the latter to the trimer complex (DHI)2/PDCA as the main component of polydopamine. However, this proposal is not supported by our findings since a) two well- resolved signals in the aliphatic region are strong evidence for \(\mathrm{sp^3}\) carbons in HNT- PDA but not present in the structure proposed by Ding et al.18; b) if the (DHI)2/PDCA trimer was the main component of HNT- PDA, we would expect an intense DHI peak at \(\sim 105\) ppm in the ssNMR spectrum. This discrepancy can be rationalized considering that mass spectroscopy cannot distinguish between two chemical structures with the same molecular weight as DAC and DHI, and hence the peak at \(\mathrm{m} / \mathrm{z} = 149\) in TOF- SIMS spectrum of polydopamine could be also ascribed to DAC. In his study by HPLC coupled with isotope- labelling techniques17, Chan provided strong evidence supporting uncycled dopamine and DAC as the main components for polydopamine, in reasonably good agreement with our observations. In addition, Alfieri et al.9 found that in contrast to dopamine, DHI cannot form a coating on quartz under the commonly used conditions for polymerization of dopamine (Tris solution at pH 8.5) and dismissed the decisive role of DHI in the polydopamine film formation. However, the structure containing pyran rings and one tertiary nitrogen proposed by these authors is contradicted by our XPS data of polydopamine deposited on HNTs discussed below, where a noticeable amount of primary amine groups is detected. This raises the question of why the DHI peak is less pronounced in the spectrum of HNT- PDA compared with that of PDA- A. The answer comes from the extremely weak basic strength of DHI molecule (pKa for the strongest base = - 6.3)29. Such a low basicity ensures that DHI molecules are neutral in the polymerization medium (pH = 8.5), and as a consequence, they cannot associate with the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 82, 882, 141]]<|/det|> +negatively charged HNTs surfaces. This can explain the low intensity of the DHI peak for HNT- PDA. Based on this interpretation, the presence of DHI molecules in HNT- PDA has been neglected in our further analysis. + +<|ref|>text<|/ref|><|det|>[[115, 164, 884, 581]]<|/det|> +XPS analysis. In the next step of our investigation, XPS was used to study the PDA film formation kinetics by evaluating the chemical composition of the PDA deposited on the HNT surfaces as a function of the reaction time. In addition, to have a reference, a detailed XPS characterization of the PDA- A samples isolated from an aqueous alkaline dopamine solution after \(24 \text{h}\) can be found in the Supplementary Fig. 6. In parallel, to investigate the effects of the substrate dimensions on the kinetics of PDA film formation, a substrate with macroscopic dimensions \((1.5 \times 1.5 \text{cm}^2)\) and a chemical nature similar to the external surface of HNTs, namely a Si wafer covered by an oxide layer, was exposed to the dopamine polymerization medium for different times, and the obtained PDA films were also analysed by XPS. The survey spectra of HNT- PDA and \(\text{SiO}_2\text{- PDA}\) as a function of deposition time are shown in Supplementary Fig. 7. All samples exhibit the carbon and nitrogen signatures, indicating the formation of a polydopamine layer on the surface. The stoichiometric analysis was performed by collecting the detailed core level spectra of the constituent elements of HNTs, \(\text{SiO}_2\) and polydopamine, i.e. \(\text{Al}_2\text{p}\) , \(\text{Si}_2\text{p}\) , \(\text{O}_1\text{s}\) , \(\text{C}_1\text{s}\) and \(\text{N}_1\text{s}\) , and deducing the corresponding atomic percentages in the probed volume in order to evaluate the evolution of the elemental composition as a function of dopamine polymerization time (Supplementary Table 1). For all samples, the atomic percentage of Al and/or Si decreases continuously with increasing deposition time, while the atomic percentages of nitrogen and carbon rise, indicating PDA film growth on the surface. In addition, the \(\text{N/C}\) atomic ratio \((0.10 \pm 0.02)\) in all samples is in agreement with the value of polydopamine \((\text{N/C} = 0.10)\) , confirming the presence of PDA on the surface. + +<|ref|>text<|/ref|><|det|>[[115, 584, 884, 700]]<|/det|> +Typical detailed XPS spectra of the C1s core level region for HNT- PDA samples are shown in Fig. 5. The same chemical species as observed in the C1s spectrum of PDA- A (see Supplementary Fig. 6b) can be identified in the C1s spectra of HNT- PDA samples, namely C- C/C=C (red, observed at a binding energy (BE) of \(284.8 \text{eV}\) ), C- O/C- N (blue, \(\text{BE} = 286.1 \text{eV}\) ), C=O/C=N (green, \(\text{BE} = 287.5 \text{eV}\) ), O=C- O (pink, \(\text{BE} = 289.2 \text{eV}\) ) bonds and shake- up component at \(\text{BE}\) of \(291.3 \text{eV}^{18,30,31}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[133, 108, 881, 562]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 576, 880, 611]]<|/det|> +
Fig. 5 High-resolution XPS spectra of the C1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.
+ +<|ref|>text<|/ref|><|det|>[[115, 634, 884, 771]]<|/det|> +The spectrum of the N1s core level region changes significantly as the polymerization proceeds (see Fig. 6). In the initial phase, the N1s peak can be fitted with two components, attributed to \(\mathrm{R - NH_3^+}\) (marked in orange in Fig. 6, \(\mathrm{BE} = 402.5 \mathrm{eV}\) ) and \(\mathrm{R - NH - R}\) (purple, \(\mathrm{BE} = 400.1 \mathrm{eV}\) ) moieties \(^{18,32}\) . Considering the building blocks shown in Fig. 4, a primary amine component points out the presence of both DA and DAQ, while secondary amines feature in DAC and PDCA. However, as the reaction proceeds, a third component appears at \(\mathrm{BE} = 398.8 \mathrm{eV}\) , which is the characteristic BE of an imine structure (green) \(^{30}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[132, 90, 875, 545]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 567, 881, 602]]<|/det|> +
Fig. 6 High-resolution XPS spectra of the N1s core level region for HNT-PDA after (a) 5 min, (b) 15 min, (c) 30 min, (d) 60 min, (e) 120 min and (f) 240 min of polydopamine polymerization.
+ +<|ref|>text<|/ref|><|det|>[[115, 624, 884, 904]]<|/det|> +Fig. 7a- b shows the evolution of the relative percentages of carbon and nitrogen species during the polymerization as deduced from their photoemission intensities. It is worth noting that the ratio between C- C/C=C and C- O/C- N components is below 1 at the beginning of the reaction, i.e. after 5 min and 15 min of reaction time, while it stabilizes at 1 as the polymerization proceeds (see Fig. 7a). This finding is somewhat surprising; indeed, considering only the structures originating from dopamine, i.e. DAQ, DAC, TS and PDCA, this ratio should not be less than 1. Its lower value is therefore a robust indication for the presence of another molecular structure with alcohol or amine functional groups in the initial phases of the polymerization, namely the Tris molecule. Della Vecchia et al. reported similar observations13,33, identifying incorporation of Tris into the polydopamine structure obtained from an aged alkaline solution of dopamine after 24 h of reaction. The most striking result that emerges from the curve fitting of the N1s spectra is that in the initial phase of polymerization, primary amine is the main component, while the secondary amine contribution becomes more pronounced with increasing reaction time (see Fig. 7b). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 90, 856, 337]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 358, 881, 396]]<|/det|> +
Fig. 7 Time dependent evolution of (a) C1s and (b) N1s components for HNT-PDA during dopamine polymerization.
+ +<|ref|>text<|/ref|><|det|>[[115, 418, 883, 899]]<|/det|> +The prominent primary amine component confirms the presence of intact Tris units at the beginning of the reaction, which corroborates the interpretation of the C1s spectra. In addition, as the secondary amine component becomes more important, the imine feature starts to be noticeable in the N1s spectra and its intensity continuously increases as the polymerization proceeds. These findings led us to conclude that the imine component is attributed to TS, which is a tautomer of DAC. It should be noted that the spectra of the C1s and N1s core level regions for HNT- PDA samples after 240 min of reaction are identical to those of PDA- A samples, suggesting that the surface charge of HNTs does not affect the distribution of functional groups in the PDA film deposited on the surface. Based on the above results, we can infer that the main building blocks of the polydopamine film are DA, DAQ, DAC, PDCA, TS and Tris. Surprisingly, in contrast to HNTs, after deconvolution of the N1s and C1s spectra for \(\mathrm{SiO_2}\) - PDA samples, no significant changes in the relative amounts of nitrogen and carbon species were observed during the reaction; all samples show relative percentages comparable to PDA aggregates (Supplementary Fig. 8 and Table 2). As already mentioned in the introduction, Huang et al. \(^{20}\) systematically monitored the growth of a PDA film on a mica surface and observed an evolution of the N1s and C1s spectra as a function of reaction time close to ours for HNT- PDA; however, in their case 60- 300 s were sufficient for the formation of a complete PDA layer with thickness ranging between 0.5 and 1.1 nm. In the case of our oxidized Si substrate, we do not observe this evolution. We suggest that this is due to the much lower number of adsorption sites on the small \(\mathrm{SiO_2}\) surface \((2.25 \times 10^{- 4} \mathrm{~m}^2)\) as compared to the much larger surface \((18.6 \mathrm{~m}^2\) for 0.4 g HNT) present in the solution when HNTs are added: very few of the intermediates and oligomers associated with the PDA aggregation can be adsorbed on the oxidized Si surface, while the majority of them form PDA aggregates in solution, which consequently deposit on the substrate. However, when HNTs are present, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 83, 883, 161]]<|/det|> +amount of the available adsorption sites is so large that intermediates and oligomers adsorb on the nanotubes to such an extent that we can no longer detect PDA aggregates in solution. On the HNT surface we can therefore follow the PDA polymerization and unravel the mechanism. + +<|ref|>text<|/ref|><|det|>[[115, 185, 883, 302]]<|/det|> +Towards a rational model for the formation of polydopamine. To determine which intermediates are formed in PDA polymerization, and therefore understand how the polymerization proceeds, an empirical model was developed based on the analysis of the XPS data. The relative percentage of each building block was calculated by using a numerical method. Further details on the model development and the calculations are presented in the Supplementary information. + +<|ref|>text<|/ref|><|det|>[[114, 303, 884, 824]]<|/det|> +Fig. 8 shows the relative percentage of each building block during the formation of the polydopamine film. In Fig. 8b, one sees a high amount of dopamine quinone present after 5 min of polymerization; this is expected according to the commonly accepted mechanism for polydopamine formation, where the first step is the auto- oxidation of dopamine (DA) to dopamine quinone (DAQ) in an alkaline solution11. However, as the reaction proceeds, the amount of DAQ decreases, suggesting that this molecule is consumed during polymerization. In addition, as illustrated in Fig. 8a, the amount of dopamine (DA) in the deposited film increases in the initial phases of the reaction, but starts to decrease after 60 min of polymerization. This mirrors the characteristic result of kinetics of intermediate species in consecutive reactions, where a certain molecule formed in the first reaction is consumed in the second one. Accordingly, we propose that the reaction responsible for increasing the dopamine concentration in the film structure is the crosslinking of quinone molecules (DAQ) resulting in the formation of polycatecholamine34. The second reaction consists then in the oxidation and cyclization of polycatecholamine leading to the creation of dopaminechrome (DAC) units, as confirmed by the increase of that building block seen in Fig. 8c. However, Fig. 8d shows an increase as a function of polymerization time also for pyrrole dicarboxylic acid (PDCA), suggesting that PDCA also originates from DAC. This is consistent with the study of Napolitano et al.35, which revealed the possibility of oxidative breakage of o- quinone rings resulting in the formation of pyrrole dicarboxylic acid species. However, the rate at which PDCA increases is less than that of DAC, indicating that only a small portion of dopaminechrome units undergoes the oxidative ring breakage. In addition, if one considers the Fig. 8e, one sees that the increase in TS during polymerization identifies this building block as a product of DAC tautomerization. Finally, Fig. 8f shows that Tris is more prominently present at the beginning of the reaction, suggesting the adsorption of the Tris units by hydrogen bonds and/or electrostatic interaction on the HNT surface. The growth of the polydopamine layer on the HNTs leads to a decrease of Tris, as the reaction proceeds. + +<|ref|>text<|/ref|><|det|>[[116, 825, 883, 902]]<|/det|> +Overall, all these results allow us to draw a complete picture for the formation of the polydopamine film as illustrated in Fig. 9. As a first step, dopamine molecules (DA) undergo a pH- induced auto- oxidation reaction resulting in quinone molecules (DAQ) formation. Then crosslinking of DAQ molecules occurs through the formation of biphenyl bond, which leads to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 884, 240]]<|/det|> +polycatecholamine as an intermediate for the film growth process. These results corroborate the findings of Burzio et al. \(^{36}\) who reported that the \(o\) - quinone groups can react through reverse dismutation with catechol groups to produce two semiquinone radicals that couple to form di- catechol crosslinks. Then uncycled units in the polycatecholamine undergo simultaneously oxidation and inter- molecular cyclization to generate dopaminechrome. In the next step, the tautomerization of DAC molecules causes the appearance of a small quantity of TS molecules in the polydopamine film. Lastly, the oxidative ring breakage reaction of DAC generates PDCA units, which can be detected in low amounts in the polydopamine film. + +<|ref|>image<|/ref|><|det|>[[130, 256, 864, 626]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 636, 883, 689]]<|/det|> +
Fig. 8 Time evolution the concentration of the main building blocks during polydopamine film formation on halloysite nanotubes as deduced from the numerical model. For details, see text and Supplementary.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 80, 844, 335]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 345, 541, 361]]<|/det|> +
Fig. 9 Proposed model for polydopamine film formation.
+ +<|ref|>text<|/ref|><|det|>[[115, 384, 884, 744]]<|/det|> +In conclusion, this study presents a systematic investigation of polydopamine film formation on the nanometer- sized surface of halloysite nanotubes. The role of 5,6- dihydroxyindole molecules in PDA film formation process was ruled out by a comparative inspection of the ssNMR spectra of PDA film formed on the HNTS and that of PDA aggregates. Interestingly, monitoring the process kinetics showed that the presence of HNTS in the dopamine polymerization medium resulted in a deceleration of the PDA film formation process because of massive adsorption of intermediates onto HNT surfaces. As a consequence of such a decelerated process, the intermediates involved in the formation of the PDA film could be identified and a complete picture for the mechanism of the process could be drawn up. Our results show that oxidative coupling of the quinone units leading to formation the polycatecholamine species is the most dominant reaction in the initial phase of the film formation process. As the reaction time proceeds, intermolecular cyclization gradually occurs in the dopamine units that are present in the polycatecholamine species. In addition, we could provide experimental evidence for the inclusion of Tris in the layer during the initial phase of the film formation process, while the presence of this molecule become less important as the reaction time proceeds. The results presented here give new insights in the PDA film formation, which can be further exploited for tuning the properties and functions of the PDA coating. + +<|ref|>sub_title<|/ref|><|det|>[[116, 768, 195, 783]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[116, 786, 883, 857]]<|/det|> +Materials. Halloysite nanotubes (HNT, \(\mathrm{Al}_2\mathrm{Si}_2\mathrm{O}_5(\mathrm{OH})_4.2\mathrm{H}_2\mathrm{O}\) ), dopamine hydrochloride (DA, \(\mathrm{C}_8\mathrm{H}_{11}\mathrm{O}_2\mathrm{N}\cdot \mathrm{HCl}\) , \(\geq 98\%\) , AR) and Tris (hydroxymethyl) aminoethane (Tris, \(\mathrm{C}_4\mathrm{H}_{11}\mathrm{O}_3\mathrm{N}\) , \(\geq 99.9\%\) , AR), were purchased from Sigma Aldrich and used as received. All the solutions were prepared using Milli- Q water. + +<|ref|>text<|/ref|><|det|>[[116, 867, 883, 902]]<|/det|> +Preparation of PDA aggregates. PDA aggregates were synthesized by adding 0.4 g of DA to \(200~\mathrm{ml}\) of \(10~\mathrm{mM}\) Tris solution at \(\mathrm{pH} = 8.5\) . The colour of the solution gradually changed from colourless to black, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 82, 881, 118]]<|/det|> +indicating the formation of PDA. After \(24h\) of stirring at ambient temperature, the black precipitate was separated by centrifugation and dried in a vacuum jar. This product is labelled as PDA- A. + +<|ref|>text<|/ref|><|det|>[[115, 128, 883, 347]]<|/det|> +PDA coating procedure. \(0.8g\) of HNTs were added to \(200ml\) of \(10mM\) Tris aqueous solution at \(\mathsf{pH} = 8.5\) The mixture was sonicated for 30 min to assure good dispersion, then \(0.4g\) of DA was added to the suspension, followed by stirring at room temperature for different time intervals, namely 5, 15, 30, 60, 120 and \(240mm\) . The product was separated by centrifugation and washed several times with Milli- Q water to remove the unreacted reagents. Then, the obtained pellet was dried under vacuum overnight; it is denoted as HNT- PDA. In a complementary experiment, in order to investigate the effects of substrate dimensions on the PDA formation process, Si wafers covered by a native oxide layer (Siegert Wafer, thickness \(525\pm 20\mu m\) ) with dimensions of \(1.5\times 1.5cm^2\) were immersed in a \(50ml\) of a freshly prepared solution of DA \((2mg / ml)\) and Tris \((10mM)\) . The reaction mixture was then shaken at \(150rpm\) and ambient temperature using an orbital shaker incubator (VWR, Incubating Mini Shaker). After certain time intervals, the substrate was removed from the solution, rinsed several times with Milli- Q water and gently dried with an argon flow. The coated substrate is referred to as \(\mathrm{SiO_2}\) - PDA. + +<|ref|>text<|/ref|><|det|>[[115, 352, 883, 904]]<|/det|> +Characterization. Dynamic light scattering (DLS) measurements were conducted using a Malvern Zetasizer Ultra. Stock solutions of DA \((20mg / ml)\) and Tris \((1M)\) were prepared and filtered through a \(0.2\mu m\) pore size filter to remove the non- dissolved species. \(0.1ml\) of the DA stock solution was diluted to \(1ml\) by Milli- Q water or a HNT suspension at a concentration of \(0.2mg / ml\) . Then, \(10\mu l\) of the Tris stock solution was added to start the dopamine polymerization. After a certain time of stirring, the reaction was stopped by acidifying the solution to \(\mathsf{pH} = 2\) using a 4M HCl solution. Acidified solutions were then transferred to the dust- free disposable polystyrene cuvettes for DLS measurements. It is worth note that for non- spherical particles, the particle size obtained from DLS analysis represents the size of spherical aggregates31. The mass percentage of the PDA layer deposited on the HNTs was studied through thermogravimetric analysis (TGA) by heating samples approximately \(5mg\) at a rate of \(5^{\circ}C / min\) from 25 to \(700^{\circ}C\) under \(\mathsf{N}_2\) atmosphere, using a PL Thermogravimetric analyser (Polymer laboratories, TGA 1000, UK). Porosimetry studies was carried out using a Micrometrics ASAP 2420 V2.05 instrument and the \(\mathsf{N}_2\) adsorption/desorption isotherms were acquired at - \(196^{\circ}C\) . The Barrett, Joyner and Halenda (BJH) model was applied to the \(\mathsf{N}_2\) desorption data to obtain the pore size distribution. A JEOL 2010F transmission electron microscope (TEM), operating at \(200keV\) , was utilized to investigate the structure of the nanotubes. For this HNT powder was sonicated in ethanol and the suspension dropcast on a carbon- coated gold grid. Energy dispersive X- ray spectroscopy (EDS) was performed using an EDAX Octane silicon drift detector with an accelerating voltage of \(20kV\) , in conjugation with a FEI- Philips FEG- XL30s scanning electron microscope (SEM). Solid state- nuclear magnetic resonance (ssNMR) was preformed on a Bruker AVANCE NEO 600 MHz (14.1 T) spectrometer, using a \(3.2 - mm\) EFree HCN MAS Probe from Bruker Biospin, at \(18kHz\) MAS and \(283K\) temperature.13 C ssNMR spectra were acquired using the cross- polarization (CP) pulse sequence, with \(5\mu s\) \(90^{\circ}\) carbon pulse and \(2.5\mu s\) \(90^{\circ}\) proton pulse. Two distinct experimental conditions were used for the two investigated samples: short and long CP contact time, corresponding to \(80\mu s\) and \(2ms\) . The free induction decays were recorded under high- power proton decoupling ( \(100kHz\) ) using the two- pulse phase modulation scheme (TPPM)37, with a recycle delay of \(2.5s\) and by averaging \(1k\) and \(20k\) transients for PDA- A and HNT- PDA respectively. NMR spectra were processed with NMRPipe38. Chemical shifts were referenced to \(4,4\) - dimethyl- 4- silapentane- 1- 1 sulfonic acid using the adamantane13 C chemical shifts as an external reference, as previously described39,40. A Surface Science SSX- 100 ESCA instrument with a monochromatic Al Kα X- ray source (hv = \(1486.6eV\) ) operating in a vacuum of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 883, 338]]<|/det|> +\(2\times 10^{- 9}\) mbar was used to perform the X- ray photoelectron spectroscopy (XPS) analysis. Samples were prepared ex situ by pressing dried powders of PDA- A or HNT- PDA onto silver substrates (previously prepared by flattening Ag pearls (Goodfellow, silver lump AG006100, purity \(99.999\%\) , size: \(3\mathrm{mm}\) ) in a press (RHC, 30 ton pillar press). Data acquisition was performed on an area of \(1000\mu \mathrm{m}^2\) and the electron take- off angle with respect to surface normal was \(37^{\circ}\) . The energy resolution was set to 1.3 eV for both the survey spectra and the detailed spectra of the C1s and N1s core level regions; a charge neutralizer system in optimized conditions was used during the XPS measurements to compensate for charging effects. Binding energies are reported \(\pm 0.1\) eV and referenced to the \(\mathrm{Si2p}\) photoemission peak centered at a binding energy of \(102.7\mathrm{eV}^{41}\) . The detailed spectra were analysed with the help of least- squares curve- fitting program Winspec (University of Namur, Belgium). Deconvolution of the spectra included a Shirley baseline subtraction and fitting with peak profiles taken as a convolution of Gaussian and Lorentzian functions. The uncertainty in the peak intensity determination is within \(2\%\) for all core levels reported. All measurements were carried out on freshly prepared samples and on three spots in order to check for homogeneity. + +<|ref|>sub_title<|/ref|><|det|>[[116, 362, 255, 377]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[116, 380, 883, 415]]<|/det|> +The authors declare that all the relevant data supporting the findings of this study are available within the article and its Supplementary information files. Source data are provided with this paper. + +<|ref|>sub_title<|/ref|><|det|>[[116, 439, 212, 454]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[113, 456, 885, 897]]<|/det|> +1. Ma, Y. et al. Mussel-Inspired Underwater Adhesives-from Adhesion Mechanisms to Engineering Applications: A Critical Review. Rev. Adhes. Adhes. 9, 167-188 (2020). +2. Li, X. et al. Protein-mediated bioadhesion in marine organisms: a review. Mar. Environ. Res. 170, 105409 (2021). +3. Lee, H.; Dellatore, S. M.; Miller, W. M.; Messersmith, S. B. Mussel-inspired surface chemistry for multifunctional coatings. Science 318, 426-430 (2007). +4. Ryu, J. H.; Messersmith, P. B.; Lee, H. Polydopamine surface chemistry: a decade of discovery. ACS Appl. Mater. Interfaces 10, 7523-7540 (2018). +5. Barclay, T. G.; Hegab, H. M.; Clarke, S. R.; Ginic-Markovic, M. Versatile surface modification using polydopamine and related polycatecholamines: Chemistry, structure, and applications. Adv. Mater. Interfaces 4, 1601192 (2017). +6. Lee, H. A.; Park, E.; Lee, H. Polydopamine and its derivative surface chemistry in material science: a focused review for studies at KAIST. Adv. Mater. 32, 1907505 (2020). +7. Wang, L. H. et al. A bioinspired scaffold for rapid oxygenation of cell encapsulation systems. Nat. Commun. 12, 5846 (2021). +8. Delparastan, P.; Malollari, K. G.; Lee, H.; Messersmith, P. B. Direct evidence for the polymeric nature of polydopamine. Angew. Chem. 131, 1089-1094 (2019). +9. Alfieri, M. L. et al. Structural basis of polydopamine film formation: probing 5, 6-dihydroxyindole-based eumelanin type units and the porphyrin issue. ACS Appl. Mater. Interfaces 10, 7670-7680 (2017). +10. Cheng, W. et al. Versatile polydopamine platforms: synthesis and promising applications for surface modification and advanced nanomedicine. ACS Nano 13, 8537-8565 (2019). +11. Liebscher, J. Chemistry of polydopamine–scope, variation, and limitation. Eur. J. Org. Chem. 2019, 4976-4994 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 85, 885, 120]]<|/det|> +12. Jiang, J.-H. et al. Surface modification of PE porous membranes based on the strong adhesion of polydopamine and covalent immobilization of heparin. J. Membr. Sci. 364, 194-202 (2010). + +<|ref|>text<|/ref|><|det|>[[115, 120, 884, 172]]<|/det|> +13. Della Vecchia, N. F. et al. Building-block diversity in polydopamine underpins a multifunctional eumelanin-type platform tunable through a quinone control point. Adv. Funct. Mater. 23, 1331-1340 (2013). + +<|ref|>text<|/ref|><|det|>[[115, 175, 884, 209]]<|/det|> +14. Liebscher, J. R. et al. Structure of polydopamine: a never-ending story? Langmuir 29, 10539-10548 (2013). + +<|ref|>text<|/ref|><|det|>[[115, 212, 884, 247]]<|/det|> +15. Hong, S. et al. Non-covalent self-assembly and covalent polymerization co-contribute to polydopamine formation. Adv. Funct. Mater. 22, 4711-4717 (2012). + +<|ref|>text<|/ref|><|det|>[[115, 249, 875, 267]]<|/det|> +16. Dreyer, D. R. et al. Elucidating the structure of poly (dopamine). Langmuir 28, 6428-6435 (2012). + +<|ref|>text<|/ref|><|det|>[[115, 269, 883, 321]]<|/det|> +17. Chan, W. Investigation of the chemical structure and formation mechanism of polydopamine from self-assembly of dopamine by liquid chromatography/mass spectrometry coupled with isotope-labelling techniques. Rapid Commun. Mass Spectrom. 33, 429-436 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 324, 881, 358]]<|/det|> +18. Ding, Y. et al. Insights into the aggregation/deposition and structure of a polydopamine film. Langmuir 30, 12258-12269 (2014). + +<|ref|>text<|/ref|><|det|>[[115, 360, 880, 395]]<|/det|> +19. Lyu, Q.; Hsueh, N.; Chai, C. L. Unravelling the polydopamine mystery: is the end in sight? Polym. Chem. 10, 5771-5777 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 397, 883, 432]]<|/det|> +20. Huang, Z.-H. et al. Polydopamine ultrathin film growth on mica via in-situ polymerization of dopamine with applications for silver-based antimicrobial coatings. Materials 14, 671 (2021). + +<|ref|>text<|/ref|><|det|>[[115, 434, 881, 469]]<|/det|> +21. Zangmeister, R. A.; Morris, T. A.; Tarlov, M. J. Characterization of polydopamine thin films deposited at short times by autoxidation of dopamine. Langmuir 29, 8619-8628 (2013). + +<|ref|>text<|/ref|><|det|>[[115, 471, 881, 507]]<|/det|> +22. Rella, S. et al. Investigation of polydopamine coatings by X-ray Photoelectron Spectroscopy as an effective tool for improving biomolecule conjugation. Appl. Surf. Sci. 447, 31-39 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 509, 881, 544]]<|/det|> +23. Danyluk, N.; Tomaszewska, J.; Tatarchuk, T. Halloysite nanotubes and halloysite-based composites for environmental and biomedical applications. J. Mol. Liq. 309, 113077 (2020). + +<|ref|>text<|/ref|><|det|>[[115, 547, 881, 581]]<|/det|> +24. Anastopoulos, I. et al. A review on halloysite-based adsorbents to remove pollutants in water and wastewater. J. Mol. Liq. 269, 855-868 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 584, 881, 619]]<|/det|> +25. Sanchez-Rivera, A. et al. Spectrophotometric study on the stability of dopamine and the determination of its acidity constants. Spectrochim. Acta A 59, 3193-3203 (2003). + +<|ref|>text<|/ref|><|det|>[[115, 621, 881, 656]]<|/det|> +26. Mondin, G. et al. Investigations of mussel-inspired polydopamine deposition on WC and Al2O3 particles: The influence of particle size and material. Mater. Chem. Phys. 148, 624-630 (2014). + +<|ref|>text<|/ref|><|det|>[[115, 658, 881, 693]]<|/det|> +27. Circu, M.; Filip, C. Closer to the polydopamine structure: new insights from a combined 13 C/1 H/2 H solid-state NMR study on deuterated samples. Polym. Chem. 9, 3379-3387 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 695, 881, 730]]<|/det|> +28. Matlahov, I.; van der Wel, P. C. Hidden motions and motion-induced invisibility: Dynamics-based spectral editing in solid-state NMR. Methods 148, 123-135 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 732, 881, 767]]<|/det|> +29. Wishart, D.S. et al. HMDB 4.0 — The Human Metabolome Database for 2018. Nucleic Acids Res. 46, D608-17 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 768, 881, 803]]<|/det|> +30. Qin, J. et al. Achieving stable Na metal cycling via polydopamine/multilayer graphene coating of a polypropylene separator. Nat. Commun. 12, 5786 (2021). + +<|ref|>text<|/ref|><|det|>[[115, 804, 881, 839]]<|/det|> +31. Lee, K. et al. Laser-induced graphitization of polydopamine leads to enhanced mechanical performance while preserving multifunctionality. Nat. Commun. 11, 4848 (2020). + +<|ref|>text<|/ref|><|det|>[[115, 840, 881, 890]]<|/det|> +32. Wang, W. et al. Self-healing performance and corrosion resistance of graphene oxide–mesoporous silicon layer–nanosphere structure coating under marine alternating hydrostatic pressure. Chem. Eng. J. 361, 792-804 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 884, 130]]<|/det|> +33. Della Vecchia, N. F. et al. Tris buffer modulates polydopamine growth, aggregation, and paramagnetic properties. Langmuir 30, 9811-9818 (2014). + +<|ref|>text<|/ref|><|det|>[[115, 118, 884, 432]]<|/det|> +34. Yang, J.; Stuart, M. A. C.; Kamperman, M. Jack of all trades: versatile catechol crosslinking mechanisms. Chem. Soc. Rev. 43, 8271-8298 (2014). +35. Napolitano, A.; Pezzella, A.; d'Ischia, M.; Prota, G. New pyrrole acids by oxidative degradation of eumelanins with hydrogen peroxide. Further hints to the mechanism of pigment breakdown. Tetrahedron 52, 8775-8780 (1996). +36. Burzio, L. A.; Waite, J. H. Cross-linking in adhesive quinoproteins: studies with model decapeptides. Biochemistry 39, 11147-11153 (2000). +37. Bennett, A. et al. Heteronuclear decoupling in rotating solids. J. Chem. Phys. 103(16), 6951-6958 (1995). +38. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277-293 (1995). +39. Morcombe, C. R.; Zilm, K. W. Chemical shift referencing in MAS solid state NMR. J. Magn. Reson. 162, 479-486 (2003). +40. Harris, R. K. et al. Further conventions for NMR shielding and chemical shift (IUPAC Recommendation 2008). Magn. Reson. Chem. 46, 582-598 (2008). +41. Liu, M. et al. Properties of halloysite nanotube-epoxy resin hybrids and the interfacial reactions in the systems. Nanotechnology 18, 455703 (2007). + +<|ref|>sub_title<|/ref|><|det|>[[117, 461, 276, 477]]<|/det|> +## Aknowledgements + +<|ref|>text<|/ref|><|det|>[[115, 483, 884, 580]]<|/det|> +The authors thank Jur van Dijken for assistance with the TGA measurements. H.H. acknowledges the Ubbo Emmius Programme for financially supporting his PhD project. This work benefitted from financial support by the Advanced Materials research programme of the Zernike National Research Centre under the Bonus Incentive Scheme of the Dutch Ministry for Education, Culture and Science. + +<|ref|>sub_title<|/ref|><|det|>[[117, 604, 295, 620]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 629, 884, 743]]<|/det|> +H. H. proposed, planned and executed all the experiments, analysed the data and wrote the main paper; O.D.L. contributed to the interpretation of XPS data and revised the paper; D.C. performed the main experiments with H.H.; A. L. and P. v. d. W. performed the ssNMR experiments and data analysis; T. G. contributed to the experiments; K. L., V. H. A. revised the paper; P.R. designed and supervised the experiments, and revised the paper. All authors critically reviewed and approved the paper. + +<|ref|>sub_title<|/ref|><|det|>[[117, 772, 288, 788]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[117, 797, 450, 813]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[117, 842, 311, 858]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[115, 868, 883, 907]]<|/det|> +Supplementary information. The online version contains supplementary material available at: Correspondence and request for materials should be addressed to O. D. L. and P. R. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 354, 150]]<|/det|> +Supplementaryinformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/images_list.json b/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..9f05e8d29d2a77b4d8c60b41656eac49d581e984 --- /dev/null +++ b/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/images_list.json @@ -0,0 +1,47 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Overview of sample collection, preeclampsia diagnosis, and delivery time points across patient and control groups. Bar graph illustrating the number of samples collected at each gestational week for the EOPE (a), LOPE (c) and control (e) groups. Color represents the time point of sample collection: T1 (9-14 gestational weeks); T2 (18-28 gestational weeks); T3 (at the time of preeclampsia diagnosis or >28 gestational weeks). Density plot showing the relative frequency of preeclampsia diagnosis and delivery across gestational weeks for the EOPE (b), LOPE (d) and control (f) groups.", + "footnote": [], + "bbox": [ + [ + 148, + 130, + 660, + 752 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. CfRNA abundance by organ/tissue origin in EOPE, LOPE patients and controls. (a) Total number of transcripts detected in different organs/tissues of interest. The percentage indicates the proportion of cfRNA transcripts identified relative to the total transcripts catalogued for each organ/tissue in the Human Protein Atlas database. (b) Box plots show cfRNA abundance scores by tissue of origin at each time point, calculated as the sum of log-transformed CPM-TMM normalised counts. The color represents the group. The horizontal line represents the median, with the lower and upper edges marking the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range. Sample sizes for each time point and group are as follows: T1 (EOPE, n=41; LOPE, n=43; control, n=75); T2 (EOPE, n=40; LOPE, n=41; control, n=73); T3 (EOPE, n=18 vs. control, n=36; LOPE, n=23 vs. control, n=39). P-values were determined by Wilcoxon rank-sum test with two tails. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.", + "footnote": [], + "bbox": [ + [ + 170, + 85, + 722, + 628 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. Performance evaluation and feature importance analysis of the first trimester (T1) predictive models for EOPE and LOPE. Receiver operating characteristic curve for the predictive models of EOPE (a) and LOPE (d). The X-axis represents the False Negative Rate, while the Y-axis represents the True Positive Rate. Violin plots displaying the distribution of correctly and misclassified patients and controls based on the classifier score obtained from the predictive model for EOPE (b) and LOPE (e). The X-axis shows the real obstetric outcome, while the Y-axis shows the predicted outcome. Participants with a classifier score above the threshold were classified as preeclampsia patients, and those below the", + "footnote": [], + "bbox": [ + [ + 166, + 80, + 800, + 730 + ] + ], + "page_idx": 22 + } +] \ No newline at end of file diff --git a/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a.mmd b/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a.mmd new file mode 100644 index 0000000000000000000000000000000000000000..03497838406b6fc57e3de5d6eac71e4694e0eca6 --- /dev/null +++ b/preprint/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a/preprint__0ea74b5c8de9b0e46fcdecaea8a66b6a4b59669369e79a3310d00aabe8f4f20a.mmd @@ -0,0 +1,455 @@ + +# Maternal Plasma Cell-Free RNA as a Predictor of Early and Late-Onset Preeclampsia Throughout Pregnancy + +Tamara Garrido- Gómez tgarrido@fundacioncarlossimon.com + +Carlos Simon Foundation - INCLIVA Health Research Institute https://orcid.org/0000- 0002- 6584- 4832 Nerea Castillo- Marco + +Carlos Simon Foundation, iPremom Pregnancy healthcare Diagnostics https://orcid.org/0000- 0002- 4817- 4777 + +Teresa Cordero Carlos Simon Foundation, iPremom Pregnancy healthcare Diagnostics + +Marina Igual iPremom Pregnancy healthcare Diagnostics + +Irene Muñoz- Blat Carlos Simon Foundation - INCLIVA Health Research Institute https://orcid.org/0000- 0003- 0543- 1064 + +Carla Gómez- Álvarez iPremom Pregnancy healthcare Diagnostics + +Neus Bernat- González iPremom Pregnancy healthcare Diagnostics + +Ángela Gaspar- Doménech iPremom Pregnancy healthcare Diagnostics + +Érika Ortiz- Domingo iPremom Pregnancy healthcare Diagnostics + +Alba Vives iPremom Pregnancy healthcare Diagnostics + +Sheila Ortega- Sanchís Carlos Simon Foundation - INCLIVA Health Research Institute https://orcid.org/0000- 0001- 6956- 4910 + +Rogelio Monfort- Ortiz University Hospital La Fe + +Petr Volkov Carlos Simon Foundation + +Juan Luis Delgado Fetal Medicine Unit Murcia- IMIB Arrixaca https://orcid.org/0000- 0003- 1687- 0222 + +<--- Page Split ---> + +Laura Hernandez-Hernandez Fetal Medicine Unit Murcia-IMIB Arrixaca Esther Canovas Fetal Medicine Unit Murcia-IMIB Arrixaca https://orcid.org/0000-0003-4595-2219 + +Maria M Gil Hospital Torrejon https://orcid.org/0000- 0002- 9993- 5249 + +Belén Santacruz Hospital Universitario de Torrejon, Universidad Francisco de Vitoria + +Nieves Luisa Gonzalez- Gonzalez University of La Laguna + +Walter Plasencia Hospital Universitario de Canarias + +Alfredo Perales- Marín University Hospital La Fe, University of Valencia + +Beatriz Marcos- Puig University Hospital La Fe + +Ana Palacios Hospital Alicante + +Iñigo Melchor Corcóstegui BioCruces Health Research Institute https://orcid.org/0000- 0002- 4190- 3180 + +Alicia Martin- Martinez Complejo Hospitalario Universitario Insular + +Taysa Benitez- Delgado Complejo Hospitalario Universitario Insular + +Carlos Simon Carlos Simon Foundation- INCLIVA Health Research Institute https://orcid.org/0000- 0003- 0902- 9531 + +Article + +Keywords: + +Posted Date: January 17th, 2025 + +DOI: https://doi.org/10.21203/rs.3.rs- 5684050/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +## Additional Declarations: + +Table 1 is available in the Supplementary Files section. + +<--- Page Split ---> + +Yes there is potential Competing Interest. N.C- M., M.I., T.G- G., C.S. are inventors on a patent application (EP24383276.3) covering methods for determining the risk of preeclampsia. N.C- M., T.C., M.I., C.G- A., N.B- G., A.G- D., E.O- D., A.V., T.G- G. are employees of iPremom Pregnancy Healthcare Diagnostics. C.S. is a founder of iPremom Pregnancy Healthcare Diagnostics + +Version of Record: A version of this preprint was published at Nature Communications on October 20th, 2025. See the published version at https://doi.org/10.1038/s41467-025-64215-2. + +<--- Page Split ---> + +# 1 Maternal Plasma Cell-Free RNA as a Predictor of Early and Late-Onset + +## 2 Preeclampsia Throughout Pregnancy + +3 Nerea Castillo-Marco¹,², Teresa Cordero¹,², Marina Igual², Irene Muñoz-Blat¹, Carla Gómez-Álvarez², 4 Neus Bernat-González², Ángela Gaspar-Doménech², Érika Ortíz-Domingo², Alba Vives², Sheila Ortega- 5 Sanchís¹, Rogelio Monfort-Ortiz³, Petr Volkov¹, Juan Luis Delgado⁴, Laura Hernandez-Hernandez⁴, 6 Esther Canovas⁴, Maria del Mar Gil⁵,⁶,⁷, Belén Santacruz⁵,⁷, Nieves Luisa Gonzalez-Gonzalez⁸, Walter 7 Plasencia⁹, Alfredo Perales-Marín³,¹⁰, Beatriz Marcos-Puig³, Ana María Palacios-Marqués¹¹,¹²,¹³, Íñigo 8 Melchor¹⁴, Alicia Martín-Martínez¹⁵, Taysa Benitez-Delgado¹⁵, Carlos Simón¹,²,¹⁰,¹⁶, Tamara Garrido- 9 Gómez¹,², on behalf of the PREMOM Consortium¹⁷. + +## 10 PREMOM Consortium: + +11 Aiartzaguena A, Alonso-Menéndez S, Amezcua A, Andrada-Ripolles C, Arias-Valdés EM, Arnal-Burró +12 AM, Barbazán MJ, Batalla-Urrea R, Baviera-Royo P, Bondía M, Burgos J, Cabrera-Leon CR, Campillos- +13 Maza JM, Casanova MC, Cuenca-Gómez D, De Bonrostro-Torralba C, De Leon-Socorro S, Del Campo A, +14 Delgado-Gonzalez JL, Diaz-Lozano P, Fabre M, Ferrer E, Florez-Herrero S, Francés-Ferré J, García- +15 Izquierdo O, García-Sousa V, Gibbone E, Goiti H, Gomez A, Gonzalez E, Gregorio-González SE, +16 Hernández-Suárez M, Herrero-Serrano R, Jimenez-Mendez A, Jodar-Perez MA, Larrea-Ortíz Quintana +17 MM, Lasierra-Beamonte A, Lobo-Valentín RM, López-Soto A, Lozano-Moreno A, Macías-Alonso MJ, +18 Martín-Arias A, Martín-Medrano EM, Martínez-Cendán JP, Martínez-Rivero I, Meabe A, Melchor JC, +19 Mico Y, Montesinos-Albert M, Moreno-Reviriego A, Nieto-Tous M, Orive-Boluda A, Oros D, Padrón- +20 Pérez E, Paules C, Peña-Lobo SM, Pérez-Pascual E, Recio V, Reula-Blasco C, Rodríguez L, Romero MI, +21 Rubert L, Ruiz-Martínez S, Ruiz-Peña AC, Salinas A, Sanchez-Martínez E, Satorres-Pérez E, Villar- +22 Graullera E. + +## 23 Affiliations: + +24 ¹ Carlos Simon Foundation, Valencia, Spain +25 ² R&D Department, iPremom Pregnancy healthcare Diagnostics, Valencia, Spain +26 ³ Department of Obstetrics, University Hospital La Fe, Valencia, Spain. +27 ⁴ Fetal Medicine Unit Murcia-IMIB Arrixaca +28 ⁵ Obstetrics and Gynecology Department, Hospital Universitario de Torrejón, Madrid, Spain. +29 ⁶ Obstetrics and Gynecology Department, Hospital Universitario La Paz, Madrid, Spain. +30 ⁷ School of Medicine, Universidad Francisco de Vitoria, Madrid, Spain. +31 ⁸ Department of Obstetrics and Gynecology, University of La Laguna, Tenerife. Spain. +32 ⁹ Department of Obstetrics and Gynecology, Hospital Universitario de Canarias, Tenerife. Spain. +33 ¹⁰ Department of Pediatrics, Obstetrics, and Gynecology, University of Valencia, Valencia, Spain. +34 ¹¹ Obstetrics and Gynecology Department, Dr. Balmis General University Hospital, Alicante, Spain. +35 ¹² Alicante Institute for Health and Biomedical Research (ISABIAL), Alicante, Spain. +36 ¹³ Department of Gynecology, Miguel Hernández University, Alicante, Spain. +37 ¹⁴ Obstetrics and Gynecology Service, BioCruces Health Research Institute, Hospital Universitario Cruces (Basque Country University), Biscay, Spain. +38 ¹⁵ Department of Obstetrics and Gynecology, Complejo Hospitalario Universitario Insular, Materno Infantil, Las Palmas, Spain. + +<--- Page Split ---> + +16 Department of Obstetrics and Gynecology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA 17 A complete list of contributors of the PREMOM Consortium is provided in the Supplementary Information. + +## Abstract + +Early- onset (EOPE) and late- onset preeclampsia (LOPE) pose significant challenges to maternal and child health, highlighting the need for early, non- invasive risk identification. In this prospective longitudinal study, we followed 9,586 pregnant women, collecting blood samples each trimester: 9- 14 weeks (T1), 18- 28 weeks (T2), and after 28 weeks or at preeclampsia diagnosis (T3). Plasma cell- free RNA (cfRNA) signatures were analyzed in women who developed EOPE (n=42) or LOPE (n=43) and compared to matched normotensive controls (n=75). Mapping cfRNA origins and performing differential abundance analysis provided insights into multi- organ impacts, revealing distinct transcriptional features of EOPE and LOPE. We developed a first- trimester EOPE predictive model using 36 transcripts, achieving 83% sensitivity, 88% specificity, and an AUC of 0.85, detecting risk 18.0 weeks before onset. A second- trimester model based on 87 cfRNA transcripts, predicted EOPE 8.5 weeks prior to onset with 87% sensitivity, 84% specificity, and an AUC of 0.85. For LOPE model, detecting risk 14.9 weeks before onset, used 92 cfRNAs, with 86% sensitivity, 89% specificity, and an AUC of 0.88. EOPE models were enriched for decidua- associated transcripts, highlighting the maternal involvement in this subtype, while LOPE models showed diverse tissue responses, paving the way for improved subtype differentiation and tailored interventions to mitigate preeclampsia risks. + +<--- Page Split ---> + +## 73 Introduction + +Maternal and infant mortality during pregnancy and labor are critical indicators of community and national health1,2. Most pregnancy complications arise from disorders that develop during the periconeptional phase, particularly during embryonic implantation and early placentation3. + +Preeclampsia — a life- threatening obstetric syndrome— is characterized by new- onset of hypertension after 20 weeks of gestation, accompanied by signs of kidney, liver, or brain damage4. Each year, preeclampsia contributes to 14% of maternal deaths worldwide, leaving a lasting impact on survivors' health5. It also constitutes a significant public health burden, incurring \$1.03 billion in maternal healthcare costs and an additional \$1.15 billion for neonatal care in infants born to mothers affected by preeclampsia within the first year after birth in the United States6. + +The heterogeneity of preeclampsia is notable, differentiated by the timing of onset and severity of symptoms. Early- onset preeclampsia (EOPE) arises before 34 weeks of gestation, necessitating emergency delivery to mitigate risks to maternal and fetal health7,8. In contrast, late- onset preeclampsia (LOPE) manifests after 34 weeks and can lead to severe maternal organ damage such as kidney, liver, or brain damage8- 11. Therefore, there is an urgent need for straightforward, non- invasive methods for early diagnosis of preeclampsia in the first trimester to implement preventive strategies effectively12- 14. + +Since the maternal decidua regulates the initial steps of maternal- embryo communication, decidualization resistance (DR) — characterized by defective endometrial cell differentiation— results in abnormal placentation, which has been associated with the etiology of major obstetric syndromes, including preeclampsia15- 19, even though symptoms may manifest later in gestation15- 19. Recently, we provided an in- depth multi- omics characterization of DR in former EOPE patients, further underscoring the uterine contribution to this pathological condition20. + +Analyzing plasma cell- free RNA (cfRNA) through liquid biopsy (i.e., from a blood sample) has emerged as a promising non- invasive tool for molecular monitoring in pregnancy, offering insights into physiological and pathological events21- 23. In this study, we prospectively analyzed the cfRNA profiles in 9,586 pregnant women across the three trimesters of pregnancy, comparing EOPE and LOPE with normotensive controls. This approach facilitated the characterization of the circulating transcriptome by mapping the tissue origins and transcriptional changes associated with EOPE and LOPE, revealing that both subtypes display distinct transcriptional differences compared to controls. Our research identified cfRNA profiles that exhibited robust predictive performance for EOPE in both the first (averaging 18.0 weeks before diagnosis) and second trimesters (averaging 8.5 weeks prior to clinical + +<--- Page Split ---> + +onset), as well as for LOPE in the second trimester (14.9 weeks prior to clinical onset). Monitoring cfRNA profiles not only aids in predicting the risk of developing preeclampsia but also allows the differentiation of both subtypes of preeclampsia and the evaluation of different organ damage in affected patients, providing insights into their prognosis. + +## Results + +## Clinical study design and participants baseline characteristics + +A total of 9,586 pregnant women with singleton pregnancies were enrolled in this prospective and longitudinal study in fourteen tertiary hospitals in Spain (ClinicalTrials.gov Identifier: NCT04990141). Uncomplicated pregnancies that progressed to term (>37 weeks) were classified as normotensive controls, while those diagnosed with EOPE or LOPE, were categorized according to current established ACOG4 and FIGO24 clinical guidelines. This comprehensive approach enabled us to characterize cfRNA profiles throughout the progression of pregnancy in both controls and preeclampsia patients (EOPE and LOPE), as we collected samples during each trimester and at the time of preeclampsia diagnosis (Extended Data Fig. 1a). + +In the study cohort, we included 85 patients who developed EOPE or LOPE, enabling a comparable modeling of both preeclampsia subtypes. Overall, this subset included EOPE (n=42), LOPE (n=43) patients and normotensive controls (n=75). We selected participants in the control group matched by gestational age at sample collection, maternal age, and parity to mitigate confounding effects due to biological variation (Extended Data Fig. 1b,c). Maternal characteristics, clinical symptoms, and birth outcomes are summarized in Table 1. There were no significant differences in maternal age, parity or smoking status between patients and controls. However, body mass index (BMI), a known preeclampsia risk factor, was notably higher in EOPE patients than in the control group (p = 0.0005); at the same time, LOPE patients didn't have a statistically higher BMI index (p = 0.1142). Additionally, ethnicity varied significantly between EOPE and controls (p = 0.097), but not between LOPE and controls (p = 0.1132). Natural conception rate was lower in EOPE patients compared to controls (p = 0.0426) but did not differ significantly in LOPE (p = 1). EOPE was diagnosed at 30.0 ± 3.4 weeks, with severe symptoms in 76.2% of patients; LOPE was diagnosed at 36.5 ± 1.8 weeks, with severe symptoms in 41.9% of patients. Birth outcomes for EOPE and LOPE included higher rates of small for gestational age, preterm birth, cesarean delivery, and lower fetal weight (p < 0.0001). Specifically, preterm deliveries occurred in 87.8% of EOPE patients and in 41.9% of LOPE patients, with cesarean sections required in 69.0% and 44.2% of patients, respectively. In contrast, all deliveries in the control group + +<--- Page Split ---> + +occurred at term, and only \(16\%\) involved cesarean sections. Fetal sex did not differ between groups. EOPE patients had significantly higher rates of stillbirth \((11.9\%)\) and post- delivery complications \((p< 0.001)\) , with \(35.2\%\) of mothers and \(50.0\%\) of neonates requiring intensive care. In comparison, among patients with LOPE, \(18.6\%\) of mothers and \(16.3\%\) of newborns required intensive care, whereas no mothers or newborns in the control group needed intensive care. + +For each participant, we analyzed total cfRNA from three peripheral blood samples collected between 9- 14 weeks (T1), 18- 28 weeks (T2), and at the time of diagnosis of EOPE and LOPE or after 28 weeks (T3) (Fig. 1). Data on the gestational weeks of blood sample collection are summarized in Supplementary Table 1. Due to clinical emergencies necessitating the termination of pregnancy, the third sample could not be collected from fourteen EOPE patients and seven LOPE patients. For the development of predictive models, the cohort was randomly stratified into discovery (70% of patients) and validation (30% of patients) sets, with the discovery set used to create the predictive model and the validation set to assess its accuracy (Extended Data Fig. 1a). + +## Profiling the tissue origin and dynamics of cfRNA in EOPE and LOPE through pregnancy + +We analyzed a total of 29,871 cfRNA transcripts after applying quality filtering and normalization processes. To determine the tissue origins of the identified transcripts, we compared our cfRNA dataset to the Human Protein Atlas database24, focusing on transcripts classified as "enriched" or "enhanced" in specific tissues or organs. Our experimental protocol detected over \(90\%\) of these classified transcripts for each targeted organ or tissue of interest (Fig. 2a), indicating a robust coverage of tissue- specific cfRNA signatures in our dataset. + +We then calculated the organ/tissue- specific signature score for patients and controls at three time points during pregnancy (T1, T2 and T3) (Fig. 2b). In EOPE patients, a significant increase in cfRNA transcripts from the liver, kidney, and decidua was identified at T2 \((p< 0.05)\) , indicating tissue specific damage approximately eight weeks before diagnosis (Fig. 2b). At T3, when clinical symptoms appear, EOPE patients displayed a significantly higher signature score \((p< 0.0001)\) for additional organs including brain, lungs, placenta, and lymphoid tissues, signaling widespread organ injury (Fig. 2b). In contrast in LOPE patients, tissue- specific transcripts suggesting organ damage was only observed at T3 \((p< 0.001)\) , with lower levels of significance than those in EOPE (Fig. 2b). + +To decode cfRNA dynamics throughout pregnancy, we performed a differential abundance analysis at each time point, elucidating molecular changes in the circulating transcriptome associated with disease progression and offering insights into underlying mechanisms. At the time of diagnosis (T3), + +<--- Page Split ---> + +we identified 24,336 transcripts with significantly altered expression in EOPE patients compared to controls (FDR < 0.05) (Extended Data Fig. 3a and Supplementary Table 2). In contrast, LOPE patients exhibited 11,859 differentially abundant transcripts (FDR < 0.05) (Extended Data Fig. 3b and Supplementary Table 3). Notably, only 251 cfRNAs showed differential abundance in T2 for EOPE patients (FDR < 0.05), whereas no differentially abundant cfRNAs were detected in T1 for either EOPE or LOPE patients, nor in T2 for LOPE. These findings suggest that transcriptomic alterations emerge as EOPE progresses, while LOPE remains largely unchanged. + +Gene ontology overrepresentation analysis within the differentially abundant cfRNAs at diagnosis revealed biological processes indicative of fetal and maternal organ- specific damage (FDR < 0.05) (Extended Data Fig. 3c and Supplementary Table 4). Both, EOPE and LOPE patients displayed significant enrichment in key biological processes, including transport across the blood- brain barrier, renal water homeostasis, regulation of blood pressure and cognition, which are hallmark processes of the pathology. Importantly, signatures of fetal tissue damage were identified in both EOPE and LOPE, with a notably greater impact in EOPE patients. Distinct biological processes were associated with either EOPE or LOPE. In EOPE, overrepresentation analysis revealed significantly enriched pathways related to neuronal death, renal filtration, and immune dysfunction – including interleukin- 8 production, response to interleukin- 4, neutrophil- mediated immunity, and antimicrobial humoral immune response. In contrast, LOPE cfRNA profile showed signatures linked to heart and brain function (p < 0.05), suggesting significant damage to these organs. + +Thus, cfRNA profile analysis at diagnosis (T3) underscores more extensive transcriptomic alterations in EOPE compared to LOPE, highlighting an exacerbated proinflammatory state as a defining feature. These findings underscore the impacts of the disease on multiple organ systems and suggest that cfRNA profiling can serve as a powerful tool for characterization of preeclampsia subtypes. Additionally, the identification of distinct biological processes linked to each preeclampsia subtype emphasizes the need for tailored therapeutic approaches targeting specific dysfunctions observed in EOPE and LOPE. + +## Early prediction of EOPE and LOPE in the first trimester of pregnancy + +Given the evidence that cfRNA profiles reflect molecular changes throughout pregnancy, disruptions in these pathways may help identify pregnancies at risk for EOPE or LOPE. Here, we developed a model for EOPE risk assessment based on plasma cfRNA profiles in the first trimester (T1), approximately 18.0 weeks before clinical onset. + +<--- Page Split ---> + +Our optimal predictive model for EOPE utilized 36 cfRNA transcripts (Supplementary Table 5) and was evaluated with a leave- one- out cross- validation approach to estimate disease risk. In a hold- out validation set, the model achieved a sensitivity of \(83\%\) and specificity of \(88\%\) , with an area under the receiver operator characteristic curve (AUC) of 0.85 (Fig. 3a and Supplementary Table 6). Nearly all samples were correctly classified, with minimal misclassifications observed reinforcing the model's robustness and indicating no evidence of overfitting (Fig. 3b). Relative contribution of individual cfRNA transcripts to the model's performance are detailed in Fig. 3c + +Further analysis of these 36 transcripts revealed that 17 (47.2%) were identified as markers of DR in women with a history of severe preeclampsia, including CBR3, MMP7, MDK, TRIB1, PAEP20. The model also incorporates cfRNA transcripts known to be disrupted in preeclamptic placentas, such as RFLBN25, and CD7426, as well as others associated with intrauterine growth restriction, such as CCL4L227 and MYL628. + +Using the same computational approach, we developed a predictive model for LOPE in the first trimester (T1), with predictions averaging 24.9 weeks before clinical onset. However, the model's performance in the validation set was limited, achieving a sensitivity of \(69\%\) , specificity of \(67\%\) , and an AUC of 0.68 (Fig. 3d and Supplementary Table 6), underscoring the challenges in early LOPE prediction. Misclassified samples are shown in Fig. 3e, and the relative contribution of individual cfRNAs to predictive accuracy detailed in Fig. 3f. While predictive capability was limited, analysis of the selected cfRNAs offers insights into LOPE mechanisms. + +Further exploration revealed that several of these cfRNAs map to protein- coding genes with known roles in cardiovascular, hepatic, and immune functions, including PRR23D1, SnoRD126, CD52, TRDV3. Unlike EOPE, no cfRNA transcripts in this model were associated with decidua, underscoring distinct pathophysiological pathways for EOPE and LOPE. + +In conclusion, our findings demonstrate the effectiveness of cfRNA signatures in predicting EOPE during the first trimester, while LOPE prediction remains challenging, likely reflecting fundamental differences in pathophysiology between EOPE and LOPE. + +## Early prediction for EOPE and LOPE in the second trimester of pregnancy + +We next investigated the potential for early detection of EOPE and LOPE in the second trimester (T2). The most effective predictive model for EOPE was based on 87 cfRNA transcripts (Supplementary Table 5), achieving a sensitivity of \(87\%\) and specificity of \(84\%\) , with an AUC of 0.85 in the validation set (Fig. 4a and Supplementary Table 6). Misclassified samples are shown in Fig. 4b, and importance scores for + +<--- Page Split ---> + +each transcript are illustrated in Fig. 4c. This model reliably identifies patients at risk for EOPE between 18 and 28 weeks of gestation, approximately 8.5 weeks before clinical onset. + +Further investigation into the tissue- specific origin of these transcripts revealed that 32 (36.8%) are associated with DR signature previously described in endometrial tissue from women with a history of severe preeclampsia, including CCL20, CXCR4, IGF1, RBP4, SQSTM1, WNT5A20. The persistence of decidual contributions as EOPE approaches underscores the maternal decidua's role in its pathophysiology. The model also includes inflammatory mediators such as SQSTM1, IL1B, CCL20, FASLG and TREM1, as well as transcripts encoding T cell receptors (e.g. TRAV21, TRBV27, TRBV5- 7). Additionally, it incorporates anti- inflammatory mediators like ALOX5AP, an immunosuppressive gene linked to recurrent miscarriage29, and IL19. Transcripts such as RBP4, which directly influences blood pressure regulation30, NRBF2 involved in autophagy and liver protection31, and WNT5A, and WNT5A, a key regulator of placental growth32, further support the model's clinical relevance. + +The top- performing predictive model for LOPE at T2 included 92 cfRNAs (Supplementary Table 5), achieving a sensitivity of 86%, specificity of 89%, and an AUC of 0.88 in the validation cohort (Fig. 4d and Supplementary Table 6). An analysis of misclassified samples is shown in Fig. 4e, with the contributions of individual cfRNAs to predictive accuracy detailed in Fig. 4f. + +Pathway enrichment analysis revealed that this model included cfRNAs related to immune function, such as CFHR1 and CFHR3, involved in complement activation, and immunoglobulin transcripts (e.g., IGKV3D- 20, IGKV3D- 11, IGHV5- 10- 1, IGHV3- 69- 1), and CXCR5, linked to B- cell migration33. Additionally, the model incorporates a cfRNA corresponding to HISLA, highly expressed in the liver34, and LINC0141935. Notably, most predictive cfRNAs were classified as non- coding RNAs or pseudogenes with no annotated function. In contrast to the EOPE model, this LOPE model includes only two cfRNAs related to DR, HES4 and SPEF1. + +## Discussion + +Previous efforts to develop screening tests for preeclampsia have primarily focused on circulating biomarkers related to placental dysfunction, such as sFLT1 and PlGF36. These tests have been validated for use starting at 23 weeks of gestation, with their strongest predictive accuracy typically observed within two weeks of symptom onset. Consequently, they are recommended for patients with suspected preeclampsia37,38. While these tests are particularly useful for short- term prediction, placental dysfunction- based tests are also utilized as early as the first trimester. They are often combined with maternal epidemiological factors and ultrasound or Doppler parameters. However, + +<--- Page Split ---> + +they face significant limitations in their effectiveness and application39- 42. In settings where guidelines from the National Institute for Health and Care Excellence (NICE) and the ACOG are applied, screening primarily relies on pregnancy- related factors and maternal characteristics. While this approach minimizes additional costs, it has low sensitivity (<41%)43,44. + +Predictive models based on cfRNAs from liquid biopsy, grounded in biological plausibility and applicable early in pregnancy, offering potential improvements for the clinical management of preeclampsia21,23,45, yet they have not been clinically applied. Building on this foundation, we prospectively collected blood samples from 9,586 pregnant women across three gestational trimesters (9–42 weeks), generating a comprehensive longitudinal dataset of cfRNA profiles related to EOPE or LOPE progression. The performance metrics demonstrate substantial advancements in leveraging cfRNA signatures for early detection of EOPE in both the first and second trimesters, as well as LOPE in the second trimester. + +Our study represents an advancement over previously published research in this field. By leveraging a study design that provides a dynamic and comprehensive view of cfRNA changes across all three trimesters of pregnancy, we capture critical information spanning pre- manifestation stages to disease onset. This approach enables earlier detection and offers deeper insights into disease progression compared to prior studies. Additionally, our clinically well- defined cohort allows for precise differentiation between EOPE and LOPE. The transcriptional insights we have uncovered highlight distinct pathophysiological trajectories for EOPE and LOPE, paving the way for more targeted and patient- specific interventions. Moreover, the cfRNA- based algorithms developed in our study demonstrate improved predictive performance, offering clinically relevant tools for distinguishing between EOPE and LOPE. These algorithms outperform current state- of- the- art methods, providing a robust framework for earlier and more accurate diagnosis, ultimately enhancing patient care and outcomes. + +Our analysis highlighted the tissue- specific origins of the detected cfRNAs, offering further insight into the pathophysiology of both subtypes of preeclampsia. For EOPE patients, early signs of tissue distress were observed in the liver, kidney, and decidua at T2, suggesting that these organs may be affected up to eight weeks before clinical diagnosis. By the time of the clinical onset (T3), cfRNA levels associated with critical organs such as the placenta, brain and lungs showed marked elevation in EOPE, indicating widespread organ involvement likely due to apoptotic processes releasing cfRNA into circulation. In LOPE patients, although cfRNA levels also increased by T3, levels were lower compared to EOPE. Furthermore, differential abundance analysis at diagnosis revealed distinct transcriptomic profile in + +<--- Page Split ---> + +EOPE, with more pronounced cfRNA changes than in LOPE reflecting potential differences in severity and inflammatory response between both preeclampsia subtypes. + +These distinctions between the subtypes extended to the biological roles of cfRNAs included in the predictive models. In models predicting EOPE, a substantial proportion of cfRNA transcripts were associated with genes involved in decidualization and DR, along with some placental- related transcripts. In contrast, cfRNA transcripts associated with LOPE prediction reflecting broader systemic contributions including placental malfunction. + +A limitation of our study is that the control group consisted of women without obstetric complications. Although this design allows a clear distinction between patients with preeclampsia and those with uncomplicated pregnancies, it may not fully reflect the diverse clinical backgrounds encountered in real- world settings, where women who develop other obstetric conditions later in pregnancy may also be screened. In addition, external validation is required to further validate of the diagnostic performance, which is currently underway (ClinicalTrials.gov Identifier: NCT06716242), to ensure the generalizability of the predictive models before their clinical application. + +Overall, by analyzing circulating RNAs from a single blood sample at T1 or T2, our approach provides a reliable, standardized diagnostic measure that minimizes subjective interpretation and reduces variability in clinical decision- making. This streamlined strategy simplifies risk stratification, improving both the accuracy and efficiency of preeclampsia screening and facilitating personalized patient monitoring. + +## MATERIAL AND METHODS + +## Study design + +This prospective, multicenter case- control study involved fourteen hospitals across Spain (ClinicalTrials.gov Identifier: NCT04990141). Given the incidence rate of preeclampsia, the cohort size was designed to capture a minimum of 30 patients of EOPE over the course of the study. Approval was obtained from the relevant Clinical Research Ethics Committees at each site, and written informed consent was collected from all participants prior to blood collection and sample anonymization. A total of 9,586 pregnant women were enrolled based on the following criteria: signed informed consent, age over 18, singleton pregnancy, and first blood sample collection within 9–14 gestational weeks. Each participant provided 20 mL of peripheral blood in the three trimesters of pregnancy, coinciding with routine clinical follow- up: (T1) 9–14 weeks, (T2) 18–28 weeks, and (T3) >28 weeks or at the time of preeclampsia diagnosis. Gestational age was confirmed via ultrasound during the first trimester. + +<--- Page Split ---> + +Clinical data for each participant were recorded in an electronic data capture system. All blood samples were processed to isolate plasma and stored at \(- 80^{\circ}C\) until pregnancy outcomes were available. Preeclampsia patients were diagnosed following ACOG4 and FIGO46 guidelines, as per the clinical protocol of each hospital involved. EOPE patients (n=42) and a subset of LOPE patients (n=43) were selected and matched with normotensive pregnant women with uncomplicated pregnancies as controls (n=75). Participants in the control group were selected based on matching gestational age at the time of blood collection, maternal age, and parity, utilizing Euclidean distance for optimal pairing. Patients and controls were randomly stratified following a 70:30 proportion into two cohorts: discovery and validation sets. Sample sizes for the EOPE, LOPE, and control groups in each cohort are detailed in Supplementary Table 6. The discovery set was used for feature selection, model training, and optimization, with model performance assessed by leave-one-out cross-validation. The optimal model from this process was then applied to the validation set to assess the predictive performance, yielding metrics based on an unexposed sample set. The bioinformatic workflow is detailed in Extended Data Fig. 4. + +## Blood sample processing and storage + +Peripheral blood samples (20 mL) were collected in Streck Cell- Free DNA BCT tubes (Illumina, San Diego, CA), stored, shipped at room temperature, and processed within seven days to obtain the plasma fraction. All blood samples were centrifuged for 15 min at 1,600 x g and \(4^{\circ}C\) . Plasma was transferred to a new collection tube and stored at \(- 80^{\circ}C\) until use. + +## cfrNA isolation, library preparation, and sequencing + +Plasma supernatant samples (n=457) from the study patients (n=160) were centrifuged for 10 min at 13,000 x g. Following the manufacturer's protocol, cfRNA from 2 mL of plasma was isolated using MiRNeasy Serum/Plasma Advanced Kit (Qiagen, Hilden, Germany). According to the manufacturer's protocol, cDNA libraries from total cfRNA samples were prepared using Illumina RNA Prep with Enrichment (L) Tagmentation (Illumina, San Diego, CA). cDNA libraries were quantified using an Agilent D1000 ScreenTape in a 4200 TapeStation system (Agilent Technologies Inc). Libraries were normalized to 10 nM and pooled in equal volumes. The pool concentration was quantified by qPCR using the KAPA Library Quantification Kit (Kapa Biosystems Inc) and an Agilent D1000 ScreenTape in a 4200 TapeStation system (Agilent Technologies Inc). The mean value was used to establish pool concentration, which was then sequenced in a NextSeq 500/550 High Output kit with 2.5 cartridges of 150 cycles (Illumina, San Diego, CA). + +## Sequencing data processing + +<--- Page Split ---> + +Raw reads were aligned to the human reference genome (GRCh38 Gencode v38 Primary Assembly) using STAR (2.7.10a). The SAM/BAM files were further processed using SAMtools (v.1.6). Only reads with mapping quality more significant than \(90\%\) were maintained (MAPQ score obtained from the alignment). The duplicated reads were removed with Picard MarkDuplicates (v.2.27.4). The mapping and the quantification of the reads were done using featureCounts (v.2.0.1). Read statistics were estimated using FastQC (v.0.11.9) and ResqQC (v.5.0.1) and summarized using MultiQC (v.1.13). + +## Sample quality filtering + +Three key quality parameters related to the sequencing process were estimated for each analyzed sample: RNA degradation, DNA contamination, and rRNA fraction as previously defined22,47. Samples were retained for further analysis if they met the established cut- off values for each parameter: RNA degradation (cut- off: \(40\%\) ), DNA contamination (cut- off ratio: 3), and rRNA fraction (cut- off: \(15\%\) ). Principal Component Analysis (PCA) was used as an additional quality control measure. Samples deviating by more than 3 standard deviations from the mean of the first and second components for each dataset were excluded from the analysis. + +## Differential abundance analysis + +CfRNAs differentially abundant between EOPE or LOPE patients and controls at each time point (T1, T2, T3) were identified using the limma- Voom method from the Bioconductor package limma (v3.60.5). For the T3 samples, comparisons only included patients whose samples were collected at the time of EOPE or LOPE diagnosis and gestationally matched control samples collected during routine medical appointments. Genes with False Discovery Rate (FDR) less or equal to 0.05 were considered statistically significant. + +## Enrichment analysis + +Gene Ontology (GO) analyses were performed to identify biological processes using the enrichGO function from the clusterProfiler R package (v4.2.2). The input consisted of cfRNAs that were differentially abundant between EOPE and controls, as well as LOPE and controls (FDR \(< 0.05\) ). The p- value adjustment method used was FDR, with a significance threshold set at 0.05 (FDR \(< 0.05\) ). + +## Estimating signature scores for each tissue + +Gene sets for each tissue of interest were derived from the Human Protein Atlas database24, which includes gene expression data across tissues, focusing specifically on transcripts classified as either "enriched" or "enhanced" within those tissues. The signature score in our dataset was calculated by summing the log- transformed, normalized counts of each gene in the set. For the T3 samples, + +<--- Page Split ---> + +comparisons included only those patients whose samples were collected at the time of EOPE or LOPE diagnosis and gestationally matched control samples collected during routine medical appointments, to avoid potential bias. Differences between groups were assessed using the Wilcoxon rank- sum test. + +## Data splitting + +Our study cohort was divided into discovery and validation sets to develop and evaluate the predictive models, following best practices to prevent overfitting in artificial intelligence. Using stratified sampling based on obstetric outcomes (patient/control groups) and the scikit- learn library (v1.5.1) in Python, \(70\%\) of participants were allocated to the discovery set. + +## CfRNA count normalization + +CfRNAs were filtered based on their detection value, and only cfRNAs with levels over more than 0.5 counts per million reads (CPMs) in \(\geq 70\%\) of discovery samples after removing outlier samples were kept. Discovery set CPMs were normalized using the "deseq median ratio normalization" with pydeseq2 (v0.4.1). The validation set were then normalized with the same algorithm using size factors, from discovery set as described in MLSeq package48,49. Batch effect and other possible confounding factors were assessed using PCA, hierarchical clustering with Spearman correlation as a distance metric, and variance component analysis. A differential abundance analysis was performed on the processed counts of the discovery set, comparing the patients and controls with the shrunk log2 fold changes from pyDeseq2 (v0.4.1) using the default parameters and options. The ranking of genes, based on their adjusted and non- adjusted p- value was obtained to see the differential abundant cfRNA of controls versus patients. Finally, the normalized counts of each sample of discovery and validation sets were re- scaled to 0- 1 range with a min- max scaling process. + +## Feature selection + +The Lasso regression model was used to select the more relevant cfRNAs to discriminate between patients and controls. The discovery dataset was used with a lasso regression algorithm (v1.5.2, sklearn.linear_model.Lasso) with a penalty term (alpha) of 0.5 and the case condition as a dependent part and the cfRNA abundance levels as the independent components, resulting in a regression formula that assigns a coefficient to each cfRNA variable, indicating the correlation between the condition and each variable. The number of cfRNAs selected was determined by a minimum coefficient threshold, which determined whether a cfRNA was relevant or not. Different minimum coefficient thresholds, ranging from 0 to the maximum coefficient in increments of 0.05, were tested to determine the optimal set of cfRNAs. The F1- score was calculated for each set of cfRNAs using the strategy of leave- one- out cross- validation, and the set that yield the highest F1- score metric was selected. Lasso + +<--- Page Split ---> + +regression was chosen over other feature selection methods due to the relatively small sample size, which can lead to model overfitting. The penalty term of the model helps to counteract overfitting by shrinking and selecting features with less importance50. + +## Algorithm selection and optimization + +For the development of the optimal predictor, the discovery set was used, with cfRNA selection performed using the lasso regression method as previously described. . Six different algorithms were tested with python (v3.10.6): support vector machine (v1.5.2, sklearn.svm.SVC), Elastic Net Linear Regression (v1.5.2, sklearn.linear_model.EastNet), Lasso Linear Regression (v1.5.2, sklearn.linear_model.Lasso), Random Forest (v1.5.2, sklearn.ensemble.RandomForestClassifier), XGBoost (v1.7.6 xgboost. XGBClassifier) and TabPFN (v0.1.10 tabpfn.TabPFNClassifier). Each algorithm was trained with the best parameters calculated with a grid search applied with a cross- fold strategy. The evaluation of the predictive capacity of each model was done with a leave- one- out cross- validation with the discovery samples. The algorithm providing the best F1- score was selected for each group of samples: EOPE in the first trimester (EOPE T1), in the second trimester (EOPE T2), LOPE in the first trimester (LOPE T1) and LOPE in the second trimester (LOPE T2). The resulting chosen algorithms were TabPFN for T1 EOPE and T2 LOPE and support vector machine for T2 EOPE and T1 LOPE. + +## Predictive model training + +For each model (EOPE T1, LOPE T1, EOPE T2, LOPE T2), the ML algorithm showing the highest F1- score and its best parameters was trained with the discovery dataset. To evaluate the predictive capacity with the discovery data, a strategy of leave- one- out was performed. The selected algorithm was trained N number of times. In each iteration, one sample was isolated, and the rest were used to fit the model. The fitted model was used to predict the label of the isolated sample, and the result of the prediction was added to a pool of predicted labels that were used to calculate the discovery leave- one- out metrics. Finally, the algorithm was fitted with all the discovery samples, and the obtained trained model was used to predict the labels in the validation dataset and evaluate the performance with never seen samples. + +## Model evaluation + +We evaluated the predictive performance of each model (EOPE T1, LOPE T1, EOPE T2, LOPE T2) using two approaches: (1) leave- one- out cross- validation in the discovery dataset, and (2) predictions on the validation dataset using the final model. Model performance was assessed with key metrics, including accuracy, sensitivity, specificity, AUC, and F1- score. + +## Data availability + +<--- Page Split ---> + +The cell- free RNA- sequencing data generated for this manuscript has been uploaded to Gene Expression Omnibus database under the accession number (REDACTED 11 AUGUST 2025). The uploaded data include Raw count matrices obtained with FeatureCount. The raw sequences are not publicly available due to privacy concerns. However, they are available from the corresponding authors (C.S, carlos.simon@uv.es; T.G, tgarrido@fundacioncarlossimon.com) upon reasonable request and with permission of the Institutional Review Board of the Spanish hospitals involved. + +## REFERENCES + +1. Organization WH. 2018 Global Reference List of 100 Core Health Indicators (plus health-related SDGs). Geneva; 2018. +2. Say L, Chou D, Gemmill A, et al. Global causes of maternal death: a WHO systematic analysis. Lancet Glob Health. Jun 2014;2(6):e323-33. doi:10.1016/S2214- 109X(14)70227-X +3. Moreno I, Capalbo A, Mas A, et al. The human periconceptional maternal-embryonic space in health and disease. Physiol Rev. Jul 01 2023;103(3):1965-2038. doi:10.1152/physrev.00050.2021 +4. ACOG Practice Bulletin No. 202 Summary: Gestational Hypertension and Preeclampsia. Obstet Gynecol. Jan 2019;133(1):1. doi:10.1097/AOG.0000000000003019 +5. Hinkle SN, Schisterman EF, Liu D, et al. Pregnancy Complications and Long-Term Mortality in a Diverse Cohort. Circulation. Mar 28 2023;147(13):1014-1025. doi:10.1161/CIRCULATIONAHA.122.062177 +6. Stevens W, Shih T, Incerti D, et al. Short-term costs of preeclampsia to the United States health care system. Am J Obstet Gynecol. Sep 2017;217(3):237-248.e16. doi:10.1016/j.ajog.2017.04.032 +7. Lisonkova S, Joseph KS. Incidence of preeclampsia: risk factors and outcomes associated with early- versus late-onset disease. Am J Obstet Gynecol. Dec 2013;209(6):544.e1-544.e12. doi:10.1016/j.ajog.2013.08.019 +8. Lisonkova S, Sabr Y, Mayer C, Young C, Skoll A, Joseph KS. Maternal morbidity associated with early-onset and late-onset preeclampsia. Obstet Gynecol. Oct 2014;124(4):771-781. doi:10.1097/AOG.0000000000000472 +9. de Havenon A, Delic A, Stulberg E, et al. Association of Preeclampsia With Incident Stroke in Later Life Among Women in the Framingham Heart Study. JAMA Netw Open. Apr 01 2021;4(4):e215077. doi:10.1001/jamanetworkopen.2021.5077 +10. Conti-Ramsden FI, Nathan HL, De Greeff A, et al. Pregnancy-Related Acute Kidney Injury in Preeclampsia: Risk Factors and Renal Outcomes. Hypertension. Nov 2019;74(5):1144-1151. doi:10.1161/HYPERTENSIONAHA.119.13089 +11. Lisonkova S, Bone JN, Muraca GM, et al. Incidence and risk factors for severe preeclampsia, hemolysis, elevated liver enzymes, and low platelet count syndrome, and eclampsia at preterm and term gestation: a population-based study. Am J Obstet Gynecol. Nov 2021;225(5):538.e1-538.e19. doi:10.1016/j.ajog.2021.04.261 +12. Rolnik DL, Wright D, Poon LC, et al. Aspirin versus Placebo in Pregnancies at High Risk for Preterm Preeclampsia. N Engl J Med. Aug 17 2017;377(7):613-622. doi:10.1056/NEJMoa1704559 +13. Davidson KW, Barry MJ, Mangione CM, et al. Aspirin Use to Prevent Preeclampsia and Related Morbidity and Mortality: US Preventive Services Task Force Recommendation Statement. JAMA. Sep 28 2021;326(12):1186-1191. doi:10.1001/jama.2021.14781 +14. Bibbins-Domingo K, Grossman DC, Curry SJ, et al. Screening for Preeclampsia: US Preventive Services Task Force Recommendation Statement. JAMA. Apr 25 2017;317(16):1661-1667. doi:10.1001/jama.2017.3439 + +<--- Page Split ---> + +15. Garrido-Gomez T, Dominguez F, Quiñonero A, et al. Defective decidualization during and after severe preeclampsia reveals a possible maternal contribution to the etiology. Proc Natl Acad Sci U S A. 10 03 2017;114(40):E8468-E8477. doi:10.1073/pnas.1706546114 +16. Garrido-Gomez T, Castillo-Marco N, Clemente-Ciscar M, et al. Disrupted PGR-B and ESR1 signaling underlies defective decidualization linked to severe preeclampsia. Elife. 10 28 2021;10doi:10.7554/eLife.70753 +17. Garrido-Gomez T, Castillo-Marco N, Cordero T, Simón C. Decidualization resistance in the origin of preeclampsia. Am J Obstet Gynecol. 02 2022;226(2S):S886-S894. doi:10.1016/j.ajog.2020.09.039 +18. Ng SW, Norwitz GA, Pavlicev M, Tilburgs T, Simón C, Norwitz ER. Endometrial Decidualization: The Primary Driver of Pregnancy Health. Int J Mol Sci. Jun 08 2020;21(11)doi:10.3390/ijms21114092 +19. Woods L, Perez-Garcia V, Kieckbusch J, et al. Decidualisation and placentation defects are a major cause of age-related reproductive decline. Nat Commun. Sep 05 2017;8(1):352. doi:10.1038/s41467-017-00308-x +20. Muñoz-Blatt I, Pérez-Moraga R, Castillo-Marco N, et al. Multi-Omics based mapping of Decidualization Resistance in patients with a history of Severe Preeclampsia. Nat Med. Jan 07 2025. doi:10.1038/s41591-024-03407-7. +21. Munchel S, Rohrback S, Randise-Hinchliff C, et al. Circulating transcripts in maternal blood reflect a molecular signature of early-onset preeclampsia. Sci Transl Med. Jul 01 2020;12(550)doi:10.1126/scitranslmed.aaz0131 +22. Moufarrej MN, Wong RJ, Shaw GM, Stevenson DK, Quake SR. Investigating Pregnancy and Its Complications Using Circulating Cell-Free RNA in Women's Blood During Gestation. Front Pediatr. 2020;8:605219. doi:10.3389/fped.2020.605219 +23. Rasmussen M, Reddy M, Nolan R, et al. RNA profiles reveal signatures of future health and disease in pregnancy. Nature. 01 2022;601(7893):422-427. doi:10.1038/s41586-021-04249-w +24. Uhlén M, Fagerberg L, Hallström BM, et al. Proteomics. Tissue-based map of the human proteome. Science. Jan 23 2015;347(6220):1260419. doi:10.1126/science.1260419 +25. Nevalainen J, Skarp S, Savolainen ER, Ryynänen M, Järvenpää J. Intrauterine growth restriction and placental gene expression in severe preeclampsia, comparing early-onset and late-onset forms. J Perinat Med. Oct 26 2017;45(7):869-877. doi:10.1515/jpm-2016-0406 +26. Przybyl L, Haase N, Golic M, et al. CD74-Downregulation of Placental Macrophage-Trophoblastic Interactions in Preeclampsia. Circ Res. Jun 24 2016;119(1):55-68. doi:10.1161/CIRCRESAHA.116.308304 +27. Medina-Bastidas D, Guzmán-Huerta M, Borboa-Olivares H, et al. Placental Microarray Profiling Reveals Common mRNA and lncRNA Expression Patterns in Preeclampsia and Intrauterine Growth Restriction. Int J Mol Sci. May 20 2020;21(10)doi:10.3390/ijms21103597 +28. Timofeeva AV, Fedorov IS, Brzhozovskiy AG, et al. miRNAs and Their Gene Targets-A Clue to Differentiate Pregnancies with Small for Gestational Age Newborns, Intrauterine Growth Restriction, and Preeclampsia. Diagnostics (Basel). Apr 20 2021;11(4)doi:10.3390/diagnostics11040729 +29. Wang F, Jia W, Fan M, et al. Single-cell Immune Landscape of Human Recurrent Miscarriage. Genomics Proteomics Bioinformatics. Apr 2021;19(2):208-222. doi:10.1016/j.gpb.2020.11.002 +30. Kraus BJ, Sartoretto JL, Polak P, et al. Novel role for retinol-binding protein 4 in the regulation of blood pressure. FASEB J. Aug 2015;29(8):3133-40. doi:10.1096/fj.14-266064 +31. Lu J, He L, Behrends C, et al. NRBF2 regulates autophagy and prevents liver injury by modulating Atg14L-linked phosphatidylinositol-3 kinase III activity. Nat Commun. May 22 2014;5:3920. doi:10.1038/ncomms4920 +32. Lu Z, Hardt J, Kim JJ. Global analysis of genes regulated by HOXA10 in decidualization reveals a role in cell proliferation. Mol Hum Reprod. Jun 2008;14(6):357-66. doi:10.1093/molehr/gan023 +33. Stelzer G, Rosen N, Plaschkes I, et al. The GeneCards Suite: From Gene Data Mining to Disease Genome Sequence Analyses. Curr Protoc Bioinformatics. Jun 20 2016;54:1.30.1-1.30.33. doi:10.1002/cpbi.5 + +<--- Page Split ---> + +34. Fagerberg L, Hallström BM, Oksvold P, et al. Analysis of the human tissue-specific expression by genome-wide integration of transcriptomics and antibody-based proteomics. Mol Cell Proteomics. Feb 2014;13(2):397-406. doi:10.1074/mcp.M113.035600 +35. Dang H, Chen L, Tang P, et al. LINC01419 promotes cell proliferation and metastasis in hepatocellular carcinoma by enhancing NDRG1 promoter activity. Cell Oncol (Dordr). Oct 2020;43(5):931-947. doi:10.1007/s13402-020-00540-6 +36. Zeisler H, Llurba E, Chantraine F, et al. Predictive Value of the sFlt-1:PlGF Ratio in Women with Suspected Preeclampsia. N Engl J Med. Jan 07 2016;374(1):13-22. doi:10.1056/NEJMoa1414838 +37. Hund M, Allegranza D, Schoedl M, Dilba P, Verhagen-Kamerbeek W, Stepan H. Multicenter prospective clinical study to evaluate the prediction of short-term outcome in pregnant women with suspected preeclampsia (PROGNOSIS): study protocol. BMC Pregnancy Childbirth. Sep 18 2014;14:324. doi:10.1186/1471-2393-14-324 +38. Thadhani R, Lemoine E, Rana S, et al. Circulating Angiogenic Factor Levels in Hypertensive Disorders of Pregnancy. NEJM Evid. Dec 2022;1(12):EVIDoa2200161. doi:10.1056/EVIDoa2200161 +39. Chaeemsaithong P, Sahota DS, Poon LC. First trimester preeclampsia screening and prediction. Am J Obstet Gynecol. Feb 2022;226(2S):S1071-S1097.e2. doi:10.1016/j.ajog.2020.07.020 +40. Rana S, Lemoine E, Granger JP, Karumanchi SA. Preeclampsia: Pathophysiology, Challenges, and Perspectives. Circ Res. Mar 29 2019;124(7):1094-1112. doi:10.1161/CIRCRESAHA.118.313276 +41. Agrawal S, Cerdeira AS, Redman C, Vatish M. Meta-Analysis and Systematic Review to Assess the Role of Soluble FMS-Like Tyrosine Kinase-1 and Placenta Growth Factor Ratio in Prediction of Preeclampsia: The SaPPPhirE Study. Hypertension. Feb 2018;71(2):306-316. doi:10.1161/HYPERTENSIONAHA.117.10182 +42. MacDonald TM, Walker SP, Hannan NJ, Tong S, Kaitu'u-Lino TJ. Clinical tools and biomarkers to predict preeclampsia. EBioMedicine. Jan 2022;75:103780. doi:10.1016/j.ebiom.2021.103780 +43. ACOG Committee Opinion No. 743: Low-Dose Aspirin Use During Pregnancy. Obstet Gynecol. Jul 2018;132(1):e44-e52. doi:10.1097/AOG.0000000000002708 +44. Visintin C, Mugglestone MA, Almerie MQ, et al. Management of hypertensive disorders during pregnancy: summary of NICE guidance. BMJ. Aug 25 2010;341:c2207. doi:10.1136/bmj.c2207 +45. Moufarrej MN, Vorperian SK, Wong RJ, et al. Early prediction of preeclampsia in pregnancy with cell-free RNA. Nature. Feb 2022;602(7898):689-694. doi:10.1038/s41586-022-04410-z +46. Poon LC, Shennan A, Hyett JA, et al. The International Federation of Gynecology and Obstetrics (FIGO) initiative on pre-eclampsia: A pragmatic guide for first-trimester screening and prevention. Int J Gynaecol Obstet. May 2019;145 Suppl 1(Suppl 1):1-33. doi:10.1002/ijgo.12802 +47. Tong YK, Lo YM. Diagnostic developments involving cell-free (circulating) nucleic acids. Clin Chim Acta. Jan 2006;363(1-2):187-96. doi:10.1016/j.cccn.2005.05.048 +48. Goksuluk D, Zarrasiz G, Korkmaz S, et al. MLSeq: Machine learning interface for RNA-sequencing data. Comput Methods Programs Biomed. Jul 2019;175:223-231. doi:10.1016/j.cmpb.2019.04.007 +49. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550. doi:10.1186/s13059-014-0550-8 +50. Ji Hyung Lee ZS, Zhan Gao. On LASSO for predictive regression. Journal of Econometrics. 2022;229(2):322-349. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Overview of sample collection, preeclampsia diagnosis, and delivery time points across patient and control groups. Bar graph illustrating the number of samples collected at each gestational week for the EOPE (a), LOPE (c) and control (e) groups. Color represents the time point of sample collection: T1 (9-14 gestational weeks); T2 (18-28 gestational weeks); T3 (at the time of preeclampsia diagnosis or >28 gestational weeks). Density plot showing the relative frequency of preeclampsia diagnosis and delivery across gestational weeks for the EOPE (b), LOPE (d) and control (f) groups.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. CfRNA abundance by organ/tissue origin in EOPE, LOPE patients and controls. (a) Total number of transcripts detected in different organs/tissues of interest. The percentage indicates the proportion of cfRNA transcripts identified relative to the total transcripts catalogued for each organ/tissue in the Human Protein Atlas database. (b) Box plots show cfRNA abundance scores by tissue of origin at each time point, calculated as the sum of log-transformed CPM-TMM normalised counts. The color represents the group. The horizontal line represents the median, with the lower and upper edges marking the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range. Sample sizes for each time point and group are as follows: T1 (EOPE, n=41; LOPE, n=43; control, n=75); T2 (EOPE, n=40; LOPE, n=41; control, n=73); T3 (EOPE, n=18 vs. control, n=36; LOPE, n=23 vs. control, n=39). P-values were determined by Wilcoxon rank-sum test with two tails. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3. Performance evaluation and feature importance analysis of the first trimester (T1) predictive models for EOPE and LOPE. Receiver operating characteristic curve for the predictive models of EOPE (a) and LOPE (d). The X-axis represents the False Negative Rate, while the Y-axis represents the True Positive Rate. Violin plots displaying the distribution of correctly and misclassified patients and controls based on the classifier score obtained from the predictive model for EOPE (b) and LOPE (e). The X-axis shows the real obstetric outcome, while the Y-axis shows the predicted outcome. Participants with a classifier score above the threshold were classified as preeclampsia patients, and those below the
+ +<--- Page Split ---> + +threshold were predicted as controls. Bar plot illustrating each cfRNAs contribution to the performance of the predictive models of EOPE (c) and LOPE (f). The X-axis shows the feature importance scores, which quantify the relative contribution of each cfRNA to the model's predictions, with higher scores indicating features that play a more significant role in discriminating between outcomes. CfRNAs associated with DR are marked with an asterisk. AUC, area under the curve. + +<--- Page Split ---> +![PLACEHOLDER_24_0] + + +<--- Page Split ---> + +Fig. 4. Performance evaluation and feature importance analysis of the second trimester (T2) predictive models for EOPE and LOPE. Receiver operating characteristic curve for the predictive models of EOPE (a) and LOPE (d). The X-axis represents the False Negative Rate, while the Y-axis represents the True Positive Rate. Violin plots displaying the distribution of correctly and misclassified patients and controls based on the classifier score obtained from the predictive model for EOPE (b) and LOPE (e). The X-axis shows the obstetric outcome, while the Y-axis shows the predicted outcome. Participants with a predicted probability above the threshold were classified as preeclampsia patients, and those below the threshold were predicted as controls. Bar plot illustrating each cfRNAs contribution to the performance of the predictive models of EOPE (c) and LOPE (f). The X-axis shows the feature importance scores, which quantify the relative contribution of each cfRNA to the model's predictions, with higher scores indicating features that play a more significant role in discriminating between outcomes. cfRNAs associated with DR are marked with an asterisk. AUC, area under the curve. + +## Extended Data Figures + +![PLACEHOLDER_25_0] + + +Extended Data Fig. 1. Study Design and clinical/epidemiological variables matched for the selection of patients and controls. (a) Workflow from sample collection to patients and controls stratification into discovery and validation sets. (b) Box and bar plots showing that EOPE patients (n=42) and controls + +<--- Page Split ---> + +(n=75) groups are matched for gestational age at sample collection (T1: 9- 14 gestational weeks; T2: 18- 28 gestational weeks; T3: at the time of preeclampsia diagnosis or >28 gestational weeks), maternal age and primiparity. (c) Box and bar plots showing that LOPE patients (n=43) and controls (n=75) are matched for gestational age at sample collection, maternal age, and primiparity. In box plots, the horizontal line represents the median, with the lower and upper edges marking the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range. Color represents the EOPE, LOPE or control group. P-values were determined using the two- tailed Wilcoxon rank- sum test, with Bonferroni adjustment for p- values. \*\*\*\*P < 0.0001. + +<--- Page Split ---> +![PLACEHOLDER_27_0] + + +Extended Data Fig. 2. Sequencing quality parameters applied to identify outlier samples. a, RNA degradation. b, rRNA fraction. c, DNA contamination. The dashed line represents the quality threshold; samples exceeding this threshold were identified as outliers. In box plots, the horizontal line represents the median and the lower and upper edges mark the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range. Color represents the group. d, e. Principal component analysis at the three different time points (T1: 9-14 gestational weeks; T2: 18-28 gestational weeks; T3: + +<--- Page Split ---> + +at the time of diagnosis in disease or \(>28\) gestational weeks) of sample collection for EOPE compared to controls (d) and LOPE compared to controls (e). Color represents the group. Arrow points out the samples deviating more than 3 standard deviations from the mean of the first and second components, which were considered outliers. + +![PLACEHOLDER_28_0] + + +Extended Data Fig. 3. Differentially abundant cfRNAs in early-onset and late-onset preeclampsia at diagnosis and enriched biological processes. a,b, Volcano plots showing differentially abundant cfRNAs in EOPE \((n = 20)\) (a) and LOPE \((n = 34)\) (b) compared to controls \((n = 36)\) . Significant changes in cfRNA abundance is represented by color. c, Dot plot illustrating significantly enriched biological process using as input cfRNAs differentially increased in patients compared to controls. The plot shows common pathways in EOPE and LOPE as well as pathways unique to each preeclampsia subtype. Shape represents the group. The gradient color represents the ratio \((\%)\) calculated by dividing the number of transcripts altered by the total number of transcripts described in the pathway. + +<--- Page Split ---> +![PLACEHOLDER_29_0] + + +Extended Data Fig. 4. Bioinformatic workflow to develop the predictive models. Participants were divided into two groups: a discovery set (70% of samples) for model training and optimization, and a validation set (30% of samples) for independent performance assessment. In the discovery phase, cfRNA counts were normalized, and features (cfRNAs with predictive potential) were selected using a Lasso regression method. Several cfRNA sets were generated by adjusting Lasso coefficient thresholds, and each set was evaluated for model performance using the F1 score, with the set achieving the highest F1 score (80% in this example) chosen. The selected cfRNA set was then used to train and evaluate multiple models via a leave-one-out cross-validation approach, where each sample was excluded sequentially, the model trained on the remaining samples, and then tested on the excluded sample. This process ensured each sample was used for both training and testing, yielding robust performance metrics. The model with optimal performance was selected and applied to the validation set. In the validation phase, the cfRNA matrix was filtered to include only model-selected features, which served as input for the predictive model. Based on correct and incorrect classifications, final performance metrics for the model were calculated. + +## TABLES + +Table 1. Maternal characteristics and pregnancy outcomes for the selected subset of participants. + +## SUPPLEMENTARY TABLES + +Supplementary Table 1. Gestational age at blood sample for patients and controls in the selected subset of participants. + +<--- Page Split ---> + +701 Supplementary Table 2. Differentially abundant cfRNA transcripts at diagnosis in early- onset preeclampsia patients compared to normotensive controls. + +703 Supplementary Table 3. Differentially abundant cfRNA transcripts at diagnosis in late-onset preeclampsia patients compared to normotensive controls. + +705 Supplementary Table 4. Gene ontology analysis for increased abundant cfRNA in early-onset preeclampsia and late-onset preeclampsia. + +707 Supplementary Table 5. cfRNAs composing the predictive models for early-onset preeclampsia and late-onset preeclampsia at the first and second trimester of pregnancy. + +709 Supplementary Table 6. Summary of predictive model performance metrics for early-onset preeclampsia and late-onset preeclampsia during the first and second trimesters of pregnancy. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Table1.xlsx- Supp.Table1.xlsx- Supp.Table2.xlsx- Supp.Table3.xlsx- Supp.Table4.xlsx- Supp.Table5.xlsx- Supp.Table6.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/images_list.json b/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ee932a39a15f8e69a69fc0ec6ecaa053131c8752 --- /dev/null +++ b/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/images_list.json @@ -0,0 +1,122 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Characterisation of Pt/SiO₂ catalyst and TAP CO oxidation experiments. a,b, Representative TEM images with particle size distribution (inset) of (a) fresh and (b) spent Pt/SiO₂ catalyst after all of the TAP experiments. c, Argon normalised exit flux curves of m/z = 28 (CO), 32 (O₂), and 44 (CO₂) at (i) 25 and (ii) 350 °C for a pulse set of 6.6% CO 13.4% O₂ gas mixture in an inert Ar tracer over CO*-covered Pt/SiO₂ catalyst. d, Temperature-dependent integrated exit flux of m/z = 44 (CO₂) normalised via Ar on (i) pristine and (ii) CO-covered Pt/SiO₂ catalyst for CO oxidation while heating from 25–350–25 °C at a heating rate of 8 °C/min.", + "footnote": [], + "bbox": [ + [ + 217, + 90, + 770, + 512 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Isotopic labelling TAP experiment. Temperature-dependent integrated exit flux of \\(\\mathrm{m} / \\mathrm{z} = 44\\) \\((\\mathrm{CO}_2)\\) and \\(\\mathrm{m} / \\mathrm{z} = 45\\) \\((^{13}\\mathrm{CO}_2)\\) normalised via Ar from the TAP experiment where an oxygen rich \\(\\mathrm{CO} / \\mathrm{O}_2\\) mixture (1:2 molar ratio, \\(6.6\\%\\) \\(\\mathrm{CO}13.4\\%\\) \\(\\mathrm{O}_2\\) ) was pulsed over a \\(^{13}\\mathrm{CO}^*\\) -covered \\(\\mathrm{Pt / SiO}_2\\) catalyst while heating from \\(25–350^{\\circ}\\mathrm{C}\\) at a heating rate of \\(8^{\\circ}\\mathrm{C} / \\mathrm{min}\\) .", + "footnote": [], + "bbox": [ + [ + 300, + 456, + 758, + 720 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. Isothermal \\(\\mathbf{O}_2\\) titration experiments at 100 and \\(200^{\\circ}\\mathrm{C}\\) over the \\(\\mathbf{CO}^{*}\\) -covered Pt/SiO2 catalyst. a,b, Integrated Ar normalised exit flux of \\(\\mathrm{O_2}\\) and \\(\\mathrm{CO_2}\\) during the \\(\\mathrm{O_2}\\) titration experiments over the \\(\\mathrm{CO}^{*}\\) -covered \\(\\mathrm{Pt / SiO_2}\\) catalyst (left), and the apparent rate constant \\((k^{\\prime}_{app})\\) as a function of cumulative \\(\\mathrm{CO_2}\\) produced (right) at (a) 100 and (b) \\(200^{\\circ}\\mathrm{C}\\) . Yellow lines are provided to guide the eye to the two kinetic regimes.", + "footnote": [], + "bbox": [ + [ + 260, + 92, + 704, + 460 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Model fitting of isothermal \\(\\mathrm{O_2}\\) titration experiment using MZTRT. a, Schematic of the three-pathway model. b, Experimentally measured and model exit flux curves for pulse set 40 of \\(\\mathrm{O_2 / Ar}\\) during \\(\\mathrm{O_2}\\) titration experiment at \\(100^{\\circ}\\mathrm{C}\\) . c, Experimentally measured and model fitted integrated exit flux curves for the whole \\(\\mathrm{O_2}\\) titration experiment at \\(100^{\\circ}\\mathrm{C}\\) (corresponding to Fig. 3a). d, Rate constants calculated from model fitting with \\(95\\%\\) confidence intervals included. Inset shows small but non-zero value for \\(k_{r,1}\\) .", + "footnote": [], + "bbox": [ + [ + 115, + 174, + 870, + 575 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Characterisation of \\(\\mathbf{CO}^*\\) covered catalyst via DRIFTS, TAP, and kinetic modelling. a, DRIFT spectra for CO adsorption on the \\(\\mathbf{CO}^*\\) -covered \\(2\\mathrm{nm}\\mathrm{Pt} / \\mathrm{SiO}_2\\) catalyst where \\(\\mathbf{CO}^*\\) was preadsorbed at 35, 100, 200, and \\(350^{\\circ}\\mathrm{C}\\) . The spectra were obtained after cooling the catalyst to \\(35^{\\circ}\\mathrm{C}\\) . All DRIFTS measurements were acquired at \\(35^{\\circ}\\mathrm{C}\\) which avoids the influence of temperature on vibrational features \\(^{45}\\) . b, Illustration for adsorbed \\(\\mathbf{CO}^*\\) sites (well-coordinated and under-coordinated) on the surface of a model Pt nanoparticle (regular, truncated octahedron; 586 atoms; \\(\\sim 2.17 \\mathrm{nm}\\) -size). c, Integrated Ar normalised exit flux of \\(\\mathrm{m} / \\mathrm{z} = 44\\) ( \\(\\mathbf{CO}_2\\) ) during TPO experiments on the \\(\\mathbf{CO}^*\\) -covered \\(\\mathrm{Pt} / \\mathrm{SiO}_2\\) catalyst where \\(\\mathbf{CO}^*\\) was preadsorbed at 25, 100, 200, and \\(350^{\\circ}\\mathrm{C}\\) . For TPO experiments, the \\(\\mathbf{CO}^*\\) was preadsorbed at \\(25 - 350^{\\circ}\\mathrm{C}\\) on the catalyst and the catalyst was cooled to \\(25^{\\circ}\\mathrm{C}\\) . Then \\(\\mathrm{O}_2\\) was repeatedly pulsed over the catalyst while being linearly heated to \\(350^{\\circ}\\mathrm{C}\\) at \\(8^{\\circ}\\mathrm{C} / \\mathrm{min}\\) . d–g, Deconvoluted \\(\\mathbf{CO}_2\\) production pathways and \\(\\mathbf{CO}_2\\) production pathway ratios calculated using the regressed kinetic model (MZTRT) for the TPO experiments where \\(\\mathbf{CO}^*\\) was preadsorbed at (d) 25, (e) 100, (f) 200, (g) \\(350^{\\circ}\\mathrm{C}\\) .", + "footnote": [], + "bbox": [ + [ + 120, + 92, + 872, + 494 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6. Comparison between pathway ratio (TPO experiment) and area ratio of \\(\\mathbf{CO}^*\\) sites (DRIFTS investigation). A direct correlation between the total amount of pathway 1 and pathway 2 calculated from the kinetic model, and the total amount of UC and WC sites from the DRIFTS investigation is found.", + "footnote": [], + "bbox": [ + [ + 264, + 90, + 732, + 373 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_12.jpg", + "caption": "Fig. 7. Calculated rate constants \\((k_{r,1}\\) and \\(k_{r,2}\\) ) for the surface reaction between adsorbed oxygen and CO from the three-pathway kinetic model with \\(95\\%\\) confidence intervals overlaid. a,b, Isothermal titration experiments where CO was adsorbed at 100 and \\(200^{\\circ}\\mathrm{C}\\) . Increasing total \\(\\mathrm{CO}_2\\) produced is correlated with decreasing CO coverage. c,d, TPO experiments where CO was adsorbed at temperature ranging from \\(25 - 200^{\\circ}\\mathrm{C}\\) . When the production of \\(\\mathrm{CO}_2\\) is sufficiently low in the TPO experiment \\((>200^{\\circ}\\mathrm{C})\\) the signal/noise ratio of the \\(\\mathrm{CO}_2\\) exit flux curves significantly decreases, which in turn decreases the confidence in the model fitting, particularly for pathway 2, as shown in Supplementary Fig. 12.", + "footnote": [], + "bbox": [ + [ + 196, + 316, + 808, + 727 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8. Summary of the three-pathway model for the reaction of \\(\\mathbf{O}_2\\) with preabsorbed \\(\\mathbf{CO}^*\\) over \\(\\mathbf{SiO}_2\\) -supporrted 2 nm Pt nanoparticles.", + "footnote": [], + "bbox": [ + [ + 115, + 130, + 880, + 375 + ] + ], + "page_idx": 15 + } +] \ No newline at end of file diff --git a/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7_det.mmd b/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..cf545eaae2c7296ff3604f4d5f6231f5c746dafe --- /dev/null +++ b/preprint/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7/preprint__0eccaef55dcb59ca1b81c7f04d09c150ba54c7571b656e22e7118c670e4141a7_det.mmd @@ -0,0 +1,543 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 886, 176]]<|/det|> +# Interrogating Site Dependent Kinetics over SiO2- Supported Pt Nanoparticles + +<|ref|>text<|/ref|><|det|>[[44, 196, 182, 214]]<|/det|> +Christian Reece + +<|ref|>text<|/ref|><|det|>[[55, 223, 390, 240]]<|/det|> +christianrecee@fas.harvard.edu + +<|ref|>text<|/ref|><|det|>[[44, 269, 576, 382]]<|/det|> +Harvard University https://orcid.org/0000- 0002- 3626- 7546 Taek-Seung Kim Harvard University https://orcid.org/0000- 0001- 8137- 0326 Christopher O'Connor Harvard University https://orcid.org/0000- 0002- 9224- 9342 + +<|ref|>sub_title<|/ref|><|det|>[[44, 423, 102, 440]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 460, 135, 478]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 498, 341, 517]]<|/det|> +Posted Date: September 6th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 536, 473, 555]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3235489/v1 + +<|ref|>text<|/ref|><|det|>[[42, 574, 909, 615]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 635, 530, 654]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 690, 914, 733]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on March 7th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 46496- 1. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[149, 90, 848, 135]]<|/det|> +# Interrogating Site Dependent Kinetics over \(\mathrm{SiO_2}\) -Supported Pt Nanoparticles + +<|ref|>text<|/ref|><|det|>[[240, 135, 755, 154]]<|/det|> +Taek- Seung Kim, Christopher R. O'Connor and Christian Reece\* + +<|ref|>text<|/ref|><|det|>[[115, 171, 700, 208]]<|/det|> +Rowland Institute at Harvard, Harvard University, Cambridge, MA 02142 \*Corresponding author: christianreece@fas.harvard.edu + +<|ref|>sub_title<|/ref|><|det|>[[115, 243, 191, 259]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[112, 274, 884, 486]]<|/det|> +A detailed knowledge of reaction kinetics is key to the development of new more efficient heterogeneous catalytic processes. However, the ability to resolve site dependent kinetics has been largely limited to surface science experiments on model systems. Herein, we can bypass the pressure, materials, and temperature gaps, resolving and quantifying two distinct pathways for CO oxidation over \(\mathrm{SiO_2}\) - supported \(2\mathrm{nmPt}\) nanoparticles under operando conditions. We find that the pathway distribution directly correlates with the distribution of well- coordinated (e.g., terrace) and under- coordinated (e.g., edge, vertex) CO adsorption sites on the \(2\mathrm{nmPt}\) nanoparticles as measured by in situ DRIFTS. We conclude that well- coordinated sites follow classic Langmuir- Hinshelwood kinetics, but under- coordinated sites follow non- standard kinetics with CO oxidation being barrierless but conversely also slow. This fundamental method of kinetic site deconvolution is broadly applicable to other catalytic systems, affording bridging of the complexity gap in heterogeneous catalysis. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 133, 882, 326]]<|/det|> +Heterogeneous catalytic processes are the foundation of the chemical industry. However, our ability to rationalise and predict the behaviour of these complex industrial processes has been largely limited due to the significant complexity gap1 that exists between our fundamental understanding of catalytic systems and their application. Surface science experiments over planar model catalysts have been able to precisely resolve intrinsic catalytic kinetics and dynamic catalytic behaviour2- 6; however, their application to “real-world” catalytic systems is often limited due to perceived pressure, material and temperature gaps7. These gaps are even more apparent in small (≤ 5 nm) supported nanoparticle systems where the metals no longer retain their bulk-like properties and notable support- metal interactions can exist8,9. Therefore, a method of directly measuring intrinsic kinetics over complex multi- faceted supported nanoparticle catalysts is highly desirable. + +<|ref|>text<|/ref|><|det|>[[112, 342, 882, 707]]<|/det|> +As the rate of a reaction is defined by the specific geometric and electronic characteristics of an “active site”, we propose that measurement of an intrinsic rate constant (directly related to the free energy of the reaction) would be specific to a given active site. Therefore, a precisely resolved kinetic coefficient (or set of kinetic coefficients) can act as a parameterised representation of a given active site. CO oxidation has been utilised as a fundamental model reaction for understanding surface catalytic processes for decades4,10- 13. In particular, CO oxidation over Pt catalysts has attracted significant attention due to its complex and dynamic surface chemistry, with the reaction generally understood to proceed via a Langmuir- Hinshelwood mechanism14,15. More recently, CO oxidation was used as a probe to study atom utilisation from nanoparticles to the single- site limit13,16. Due to the variations in physical structure with changing nanoparticle size17 and reaction environment18,19 understanding the catalytic role of geometric surface sites (i.e., structural- sensitivity) has been20,21, and still is22 of great interest. For example, the activity of Pt catalysts has been considered size- dependent for C–H bond activation in the methane reforming reaction23,24, but conversely their activity for CO oxidation has typically been regarded as size- independent15,25. However, the possibility of site- dependence for CO oxidation on non- reducible oxide- supported Pt catalysts was recently reported26,27 where through a combination of operando DRIFTS and steady- state kinetic measurements it was thought that different reaction kinetics for under- coordinated (UC) and well- coordinated (WC) sites exists on the metallic Pt surface. However, as the study was performed under steady state conditions their analysis was limited to simple reaction orders and apparent activation energies. + +<|ref|>text<|/ref|><|det|>[[113, 724, 882, 881]]<|/det|> +Non- steady state techniques such as Temporal Analysis of Products (TAP, see section: The Temporal Analysis of Products Experiment) are an effective way of measuring intrinsic kinetics as they can provide information about each sequential elementary step for the overall reaction. The TAP experiment serves to bridge the perceived pressure, material, and temperature gaps with the peak pressure during a pulse over the catalyst being on the order of 1 mbar28 (similar to other operando methods) and by using a packed bed microreactor (allowing powdered samples) which can be heated to the reaction temperature. As the pulse of reactant gas contains significantly fewer molecules than the number of reactive sites on the catalyst, a single pulse is not considered to change the surface significantly. However, by repeatedly pulsing it becomes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 881, 126]]<|/det|> +possible to dynamically evolve the catalyst state using a technique known as chemical calculus29 making titration- like experiments extremely powerful30- 34. + +<|ref|>text<|/ref|><|det|>[[113, 142, 883, 405]]<|/det|> +In this work we demonstrate that with the combination of TAP experiments, kinetic modelling, and DRIFTS measurements it is possible to precisely resolve site- dependent kinetics on working catalysts under operando conditions. We provide a detailed insight into the site- dependent intrinsic kinetics for CO oxidation over a well- defined 2 nm- sized Pt/SiO2 catalyst. Using isotopic labelling, we deconvolute the production of CO2 that arises from the reaction with preadsorbed 13CO* and from the adsorption/reaction of reactant 12CO in the gas phase under CO oxidation conditions. We identify two distinct kinetic features in the adsorbed 13CO*- driven CO2 production, which when combined with isothermal titration experiments are identified as two distinct pathways for CO oxidation. Regression of a kinetic model to the TAP exit flux curves was used to quantify the distribution of each pathway and calculate the intrinsic kinetics for the surface reaction of adsorbed oxygen with preadsorbed CO* as a function of temperature and coverage. Finally, by combining this data with comparable DRIFTS measurements, we are able to deduce site- specific kinetics as a direct relationship between the distribution of kinetic pathways and the distribution of well- coordinated (e.g., terrace like) and under- coordinated (e.g., edge, vertex) sites. + +<|ref|>sub_title<|/ref|><|det|>[[115, 421, 179, 437]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[112, 454, 883, 720]]<|/det|> +Initial characterisation of pristine and CO*- covered 2 nm- sized Pt/SiO2 catalysts. Uniform and well- dispersed Pt nanoparticles were synthesized via a conventional polyol method (Supplementary Fig. 1) and were deposited on a SiO2 support with a concentration of 0.72 wt%. The average size and distribution of the nanoparticles was measured to be \(1.82 \pm 0.51\) nm (Fig. 1a) on the fresh catalyst, and \(1.94 \pm 0.37\) nm (Fig. 1b) over the spent catalyst. As no sintering of the nanoparticles occurred throughout the duration of the TAP experiments, we can directly correlate the catalytic behaviour to this narrow distribution of particles approximately 2 nm in diameter. The catalytic activity of the 2 nm Pt/SiO2 catalyst was tested under steady- state CO oxidation conditions (2.5% CO, 5% O2) with an apparent activation energy of \(89 \pm 3\) kJ/mol measured between 130 and 160 °C, and reaction orders of \(1.02 \pm 0.06\) in CO and \(1.22 \pm 0.1\) in O2 (Supplementary Fig. 2). These values match well with previous results over Pt catalysts with similar average particle sizes15,26,35- 37, demonstrating that the Pt/SiO2 catalyst used in this work is comparable with ones reported previously, albeit with a narrower size distribution. Finally, we see excellent reproducibility for all TAP experiments performed in this work (Supplementary Fig. 3). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[217, 90, 770, 512]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 519, 884, 641]]<|/det|> +
Fig. 1. Characterisation of Pt/SiO₂ catalyst and TAP CO oxidation experiments. a,b, Representative TEM images with particle size distribution (inset) of (a) fresh and (b) spent Pt/SiO₂ catalyst after all of the TAP experiments. c, Argon normalised exit flux curves of m/z = 28 (CO), 32 (O₂), and 44 (CO₂) at (i) 25 and (ii) 350 °C for a pulse set of 6.6% CO 13.4% O₂ gas mixture in an inert Ar tracer over CO*-covered Pt/SiO₂ catalyst. d, Temperature-dependent integrated exit flux of m/z = 44 (CO₂) normalised via Ar on (i) pristine and (ii) CO-covered Pt/SiO₂ catalyst for CO oxidation while heating from 25–350–25 °C at a heating rate of 8 °C/min.
+ +<|ref|>text<|/ref|><|det|>[[113, 657, 883, 903]]<|/det|> +The fresh 2 nm Pt/SiO₂ catalyst was loaded into a home- built TAP reactor³⁸ and the catalytic activity for CO oxidation was measured by pulsing an oxygen rich CO/O₂ mixture (1:2 molar ratio, 6.6% CO 13.4% O₂) over a pristine and CO*- covered catalyst while heating from 25 to 350 °C at a heating rate of 8 °C/min. At 25 °C no conversion of CO or O₂ and no production of CO₂ was measured (Fig. 1c-(i)), whereas at 350 °C near 100% conversion of the CO to CO₂ is recorded (Fig. 1c-(ii)). To simplify the comparison between the experiments, the exit flux curves for every pulse set in the experiment were integrated and normalised to the inert Ar tracer (Fig. 1d). The pristine Pt/SiO₂ catalyst (Fig. 1d-(i)) shows a gradual increase in activity from 0% CO conversion at room temperature to almost 100% CO conversion above 100 °C. Similar reactivity for CO₂ production between the heating and cooling steps was found indicating the catalyst state remains consistent during the experiment. However, the CO*- covered catalyst (Fig. 1d-(ii)) shows increased production of CO₂ compared with the pristine catalyst. The excess O₂ in the reactant gas mixture is able to react with the preadsorbed CO* and act as a titrant, sequentially removing the preadsorbed reactive sites with increasing temperatures. By 350 °C it is assumed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 882, 179]]<|/det|> +that almost all of the preadsorbed \(\mathrm{CO}^*\) has either been reacted or desorbed off the catalyst surface as the activity during cooling was the same as that measured over the pristine catalyst. Interestingly, during the heating ramp the \(\mathrm{CO}^*\) - covered \(2\mathrm{nm}\mathrm{Pt / SiO}_2\) catalyst showed two catalytic features, with peaks in \(\mathrm{CO}_2\) production around 100 and \(200^{\circ}\mathrm{C}\) . However, further experiments were required to precisely understand the cause for these two kinetic features. + +<|ref|>text<|/ref|><|det|>[[112, 194, 886, 440]]<|/det|> +Identifying pathways for oxidation of preadsorbed \(\mathrm{CO}^*\) . In an attempt to further rationalise the two kinetic features of observed in Fig. 1d-(ii), isotopically labelled \(^{13}\mathrm{CO}\) was used to prepare a \(^{13}\mathrm{CO}^*\) - covered catalyst. This allows a precise deconvolution of the \(\mathrm{CO}_2\) produced from the reaction of the \(6.6\%\) \(\mathrm{CO}13.4\%\) \(\mathrm{O}_2\) gas mixture over the catalyst \((\mathrm{m} / \mathrm{z} = 44)\) and the \(\mathrm{CO}_2\) produced from the reaction of \(\mathrm{O}_2\) with preadsorbed \(^{13}\mathrm{CO}^*\) \((\mathrm{m} / \mathrm{z} = 45)\) . The heating rate experiment was repeated on the \(^{13}\mathrm{CO}\) - covered \(\mathrm{Pt / SiO}_2\) catalyst under the exact same conditions as used in Fig. 1d-(ii). Interestingly, the reaction of gas phase \(\mathrm{CO}\) and \(\mathrm{O}_2\) with the catalyst (Fig. 2, red triangles) was similar to that measured over the pristine catalyst whereas the preadsorbed \(^{13}\mathrm{CO}^*\) - driven \(\mathrm{CO}_2\) production (Fig. 2, orange circles) shows two well- defined kinetic features around 100 and \(200^{\circ}\mathrm{C}\) . The total \(\mathrm{CO}_2\) production \((^{12}\mathrm{CO}_2 + ^{13}\mathrm{CO}_2\) , Fig. 2, black circles) also matched well with the \(\mathrm{CO}_2\) measured in Fig. 1d-(ii). As in other temperature programmed techniques such as Temporal Programmed Oxidation (TPO), the two peaks in the measured signal would indicate two different kinetic pathways (often prescribed to different sites) for the oxidation of preadsorbed \(\mathrm{CO}\) with gas phase \(\mathrm{O}_2\) . + +<|ref|>image<|/ref|><|det|>[[300, 456, 758, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 722, 883, 794]]<|/det|> +
Fig. 2. Isotopic labelling TAP experiment. Temperature-dependent integrated exit flux of \(\mathrm{m} / \mathrm{z} = 44\) \((\mathrm{CO}_2)\) and \(\mathrm{m} / \mathrm{z} = 45\) \((^{13}\mathrm{CO}_2)\) normalised via Ar from the TAP experiment where an oxygen rich \(\mathrm{CO} / \mathrm{O}_2\) mixture (1:2 molar ratio, \(6.6\%\) \(\mathrm{CO}13.4\%\) \(\mathrm{O}_2\) ) was pulsed over a \(^{13}\mathrm{CO}^*\) -covered \(\mathrm{Pt / SiO}_2\) catalyst while heating from \(25–350^{\circ}\mathrm{C}\) at a heating rate of \(8^{\circ}\mathrm{C} / \mathrm{min}\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[260, 92, 704, 460]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 465, 883, 555]]<|/det|> +
Fig. 3. Isothermal \(\mathbf{O}_2\) titration experiments at 100 and \(200^{\circ}\mathrm{C}\) over the \(\mathbf{CO}^{*}\) -covered Pt/SiO2 catalyst. a,b, Integrated Ar normalised exit flux of \(\mathrm{O_2}\) and \(\mathrm{CO_2}\) during the \(\mathrm{O_2}\) titration experiments over the \(\mathrm{CO}^{*}\) -covered \(\mathrm{Pt / SiO_2}\) catalyst (left), and the apparent rate constant \((k^{\prime}_{app})\) as a function of cumulative \(\mathrm{CO_2}\) produced (right) at (a) 100 and (b) \(200^{\circ}\mathrm{C}\) . Yellow lines are provided to guide the eye to the two kinetic regimes.
+ +<|ref|>text<|/ref|><|det|>[[113, 569, 883, 764]]<|/det|> +To directly probe the kinetic features at 100 and \(200^{\circ}\mathrm{C}\) , isothermal \(\mathrm{O_2}\) titration experiments were carried out. First, CO was pulsed over the catalyst at the reaction temperature until it was saturated with adsorbed \(\mathrm{CO}^*\) . Then, the preadsorbed \(\mathrm{CO}^*\) was titrated off sequentially with a series of \(\mathrm{O_2}\) pulses. This affords calculation of the apparent rate constant for the oxidation of \(\mathrm{CO}^*\) as a function of CO coverage \(^{31,34}\) . In the first few pulses \(\sim 100\%\) conversion of \(\mathrm{O_2}\) is observed at both temperatures (Fig. 3- left). Then, as \(\mathrm{O_2}\) is repeatedly pulsed, the \(\mathrm{CO_2}\) production decreases, and the exit flux of \(\mathrm{O_2}\) increases up to the saturation state on the Pt surface, confirming that complete removal of the reactive \(\mathrm{CO}^*\) occurs. To estimate the kinetics of the preadsorbed \(\mathrm{CO}^*\) consumption, we plot the temperature corrected apparent rate constant as a function of cumulative \(\mathrm{CO_2}\) produced (indicative of \(\mathrm{CO}^*\) coverage) in Fig. 3a,b- right, which is calculated using \(^{31}\) : + +<|ref|>equation<|/ref|><|det|>[[328, 787, 832, 824]]<|/det|> +\[k_{app}^{'} = \frac{X}{1 - X}\sqrt{T}\approx k_{app}\approx k_{a}\theta_{CO^{*}} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[113, 847, 883, 905]]<|/det|> +Where \(k_{app}^{\prime}\) is the temperature corrected apparent rate constant, \(X\) is the fractional \(\mathrm{O_2}\) conversion, \(T\) is the temperature (K), \(k_{app}\) is the apparent rate constant (s \(^{- 1}\) ), \(k_{a}\) is the intrinsic rate constant for the reaction of \(\mathrm{O_2}\) with preadsorbed \(\mathrm{CO}^*\) , and \(\theta_{CO^*}\) is the coverage of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 882, 181]]<|/det|> +preadsorbed \(\mathrm{CO^{*}}\) . On account of the usage of the same catalyst bed throughout all the pulse experiments, variations in \(k_{app}\) show a direct correlation to the intrinsic rate constant for the \(\mathrm{O_2}\) titration experiment \(^{39,40}\) . Specifically, a linearly decrease in \(k_{app}\) with \(\theta_{CO^{*}}\) (or \(\mathrm{CO_2}\) produced) indicates the presence of a unique intrinsic rate constants \(\left(\mathrm{k_a}\right)\) for the reaction of \(\mathrm{O_2}\) with \(\mathrm{CO^*}\) through the following relationship \(^{31,41}\) : + +<|ref|>equation<|/ref|><|det|>[[375, 200, 832, 240]]<|/det|> +\[\frac{\Delta k_{app}}{\Delta\theta_{CO^{*}}} = \frac{(1 - \epsilon)}{\epsilon V} k_a \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[115, 263, 882, 423]]<|/det|> +Where \(\epsilon\) is the void fraction of the packed bed, and V is volume of the catalyst zone. We identify two linear regimes for the reaction of \(\mathrm{O_2}\) with preadsorbed \(\mathrm{CO^*}\) in both isothermal titration experiments (Fig. 3, yellow line), indicating that two intrinsic rate constants, and as such two pathways exist simultaneously at both temperatures. We identify a slow and fast intrinsic reaction rate for \(\mathrm{CO_2}\) production, as shown by the gradients in Fig. 3- right. At high relative coverage of \(\mathrm{CO^*}\) the "fast" pathway dominates, at lower \(\mathrm{CO^*}\) coverages the "slow" pathway dominates during the titration experiment. Coupling this insight with the two kinetic regimes seen in the temperature programmed experiments, we feel confident in claiming that at least two different pathways for the oxidation of \(\mathrm{CO^*}\) by \(\mathrm{O_2}\) exist on the \(2\mathrm{nm}\mathrm{Pt / SiO_2}\) catalyst. + +<|ref|>text<|/ref|><|det|>[[113, 438, 882, 899]]<|/det|> +Quantifying active species using MZTRT and DRIFTS. Due to its well- defined nature, it is possible to precisely resolve the intrinsic kinetics of catalytic processes using the TAP experiment \(^{40,42}\) . For linear (first order or pseudo first order) reactions Multi- Zone TAP Reactor Theory \(^{43,44}\) (MZTRT) is a powerful and efficient tool for simulating TAP exit flux responses. The model of the experiment was built using the generalised form of MZTRT (see Supplementary I) with the Symmetric Thin- Zone assumption applied to the catalyst zone \(^{40}\) . For the reaction of oxygen with preadsorbed \(\mathrm{CO^*}\) , a three- pathway model was identified as the most likely candidate (see Supplementary II; Fig. 4a and Supplementary Fig. 4). The experimental results for two catalytic features ( \(\mathrm{CO_2}\) production around 100 and \(200^{\circ}\mathrm{C}\) in Fig. 2) and two kinetic regimes (two pathways for the reaction of \(\mathrm{O_2}\) in Fig. 3) would indicate two separate pathways for oxidation of \(\mathrm{CO^*}\) . As the oxygen balance (i.e., oxygen released as \(\mathrm{CO_2}\) / oxygen consumed) is not always 1 throughout the experiment a third pathway of irreversible oxygen adsorption is necessary in the model. It is very important to note that the model regression is performed on each set of exit flux curves individually (Fig. 4b) with the model fit to both the shape and magnitude of Ar, \(\mathrm{O_2}\) , and \(\mathrm{CO_2}\) exit flux curves. The kinetic model can be broken down into two parts. First, the apparent adsorption rate constants \(k_{a,1}'\) and \(k_{a,2}'\) represent the irreversible adsorption and subsequent reaction of gaseous \(\mathrm{O_2}\) with preabsorbed \(\mathrm{CO^*}\) , and the adsorption rate constant \(k_{a,3}'\) represents the irreversible adsorption of \(\mathrm{O_2}\) to sites where no further reaction with \(\mathrm{CO^*}\) takes place. Second, the intrinsic rate constants \(k_{r,1}\) and \(k_{r,2}\) represent the rate at which the surface reaction between adsorbed oxygen and \(\mathrm{CO^*}\) occurs for pathways 1 and 2, respectively. In short, the adsorption constants \(k_{a}'\) control the magnitude of the \(\mathrm{O_2}\) and \(\mathrm{CO_2}\) exit flux curves, whereas the surface reaction constants \(k_{r}\) control the shape. As each pulse set is regressed individually during the isothermal \(\mathrm{O_2}\) titration experiment (Fig. 4c), a set of rate constants for the reaction of oxygen with preabsorbed \(\mathrm{CO^*}\) can be calculated at each point (Fig. 4d). Models of varying complexity were tested, but only the three- pathway model was able to precisely recreate the experimental data without being overparameterised (see Supplementary II; + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 882, 143]]<|/det|> +Supplementary Fig. 5). The model fitting exhibits a high degree of confidence and consistency with all TAP experiments in this work, although decreased confidence in the intrinsic surface reaction constants is observed at the limit of very low \(\mathrm{CO_2}\) production. + +<|ref|>image<|/ref|><|det|>[[115, 174, 870, 575]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 580, 884, 687]]<|/det|> +
Fig. 4. Model fitting of isothermal \(\mathrm{O_2}\) titration experiment using MZTRT. a, Schematic of the three-pathway model. b, Experimentally measured and model exit flux curves for pulse set 40 of \(\mathrm{O_2 / Ar}\) during \(\mathrm{O_2}\) titration experiment at \(100^{\circ}\mathrm{C}\) . c, Experimentally measured and model fitted integrated exit flux curves for the whole \(\mathrm{O_2}\) titration experiment at \(100^{\circ}\mathrm{C}\) (corresponding to Fig. 3a). d, Rate constants calculated from model fitting with \(95\%\) confidence intervals included. Inset shows small but non-zero value for \(k_{r,1}\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 92, 872, 494]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 498, 883, 725]]<|/det|> +
Fig. 5. Characterisation of \(\mathbf{CO}^*\) covered catalyst via DRIFTS, TAP, and kinetic modelling. a, DRIFT spectra for CO adsorption on the \(\mathbf{CO}^*\) -covered \(2\mathrm{nm}\mathrm{Pt} / \mathrm{SiO}_2\) catalyst where \(\mathbf{CO}^*\) was preadsorbed at 35, 100, 200, and \(350^{\circ}\mathrm{C}\) . The spectra were obtained after cooling the catalyst to \(35^{\circ}\mathrm{C}\) . All DRIFTS measurements were acquired at \(35^{\circ}\mathrm{C}\) which avoids the influence of temperature on vibrational features \(^{45}\) . b, Illustration for adsorbed \(\mathbf{CO}^*\) sites (well-coordinated and under-coordinated) on the surface of a model Pt nanoparticle (regular, truncated octahedron; 586 atoms; \(\sim 2.17 \mathrm{nm}\) -size). c, Integrated Ar normalised exit flux of \(\mathrm{m} / \mathrm{z} = 44\) ( \(\mathbf{CO}_2\) ) during TPO experiments on the \(\mathbf{CO}^*\) -covered \(\mathrm{Pt} / \mathrm{SiO}_2\) catalyst where \(\mathbf{CO}^*\) was preadsorbed at 25, 100, 200, and \(350^{\circ}\mathrm{C}\) . For TPO experiments, the \(\mathbf{CO}^*\) was preadsorbed at \(25 - 350^{\circ}\mathrm{C}\) on the catalyst and the catalyst was cooled to \(25^{\circ}\mathrm{C}\) . Then \(\mathrm{O}_2\) was repeatedly pulsed over the catalyst while being linearly heated to \(350^{\circ}\mathrm{C}\) at \(8^{\circ}\mathrm{C} / \mathrm{min}\) . d–g, Deconvoluted \(\mathbf{CO}_2\) production pathways and \(\mathbf{CO}_2\) production pathway ratios calculated using the regressed kinetic model (MZTRT) for the TPO experiments where \(\mathbf{CO}^*\) was preadsorbed at (d) 25, (e) 100, (f) 200, (g) \(350^{\circ}\mathrm{C}\) .
+ +<|ref|>text<|/ref|><|det|>[[113, 740, 882, 881]]<|/det|> +To rationalise if the two pathways for oxidation of the preadsorbed \(\mathbf{CO}^*\) were correlated to the geometric structure of metallic Pt sites (e.g., terrace, edge, vertex sites) on the \(2\mathrm{nm}\) nanoparticles, a series of DRIFTS and TPO experiments were performed where CO was preadsorbed at \(25 - 350^{\circ}\mathrm{C}\) (Fig. 5). In general, it is known that the binding energy for adsorbed CO on the under- coordinated (UC) sites (e.g., edge, vertex sites) is higher than that on the well- coordinated (WC) sites (e.g., terrace sites). This implies that when CO is adsorbed at higher temperatures, an increase in populated UC sites relative to the populated WC sites would be expected \(^{46}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 883, 528]]<|/det|> +From the DRIFTS investigation in Fig. 5a, it is shown that the pre- adsorption of CO at 35, 100, 200, and \(350^{\circ}\mathrm{C}\) (35CO, 100CO, 200CO, 350CO, respectively) populates linear bound CO to WC sites, UC sites, and bridge bound CO as evidenced by three \(\mathrm{v(C - O)}\) bands (Fig. 5a,b features 1- 3). The 35CO yields \(\mathrm{v(C - O)}\) bands centred at \(2075\mathrm{cm}^{- 1}\) with a small high frequency shoulder (feature 1), a small band near \(2042\mathrm{cm}^{- 1}\) (feature 2), and a broad band at \(1805\mathrm{cm}^{- 1}\) (feature 3). Increasing the adsorption temperature to \(100^{\circ}\mathrm{C}\) causes a slight decrease in the intensity of feature 1 and a shift in features 2 and 3 to lower frequency. For the 200CO and 350CO, there is a dramatic decrease in the intensity and slight shift to lower frequency of feature 1. The decrease in the intensity is attributed to the reduction of the total number of adsorbed \(\mathrm{CO}^{*}\) with increasing temperature. In turn the surface is predominantly relatively strongly bound CO on UC sites. A larger shift to lower frequency of features 2 and 3 is observed, which clarifies feature 2 as a distinct peak from feature 1. The position of the \(\mathrm{v(C - O)}\) frequencies as a function of CO adsorption temperature are reported in Supplementary Table 1. Feature 1 with a peak maximum from \(2075\mathrm{cm}^{- 1}\) is assigned to the collective oscillation of linear bound CO to WC sites (linear WC) where the high frequency shoulder is attributed to a dense CO phase for high CO coverages. Feature 2 with a peak maximum from \(2042\mathrm{cm}^{- 1}\) is assigned to the collective oscillation of linear CO to UC sites (linear UC). The frequency of the \(\mathrm{v(C - O)}\) for linear CO on the \(2\mathrm{nm}\mathrm{Pt}\) nanoparticles catalyst is in agreement with previous investigations when considering the frequency is dependent on the CO coverage \(^{26,47 - 57}\) , Pt coordination environment \(^{26,35,51 - 63}\) , and nanoparticle size \(^{26,57 - 61}\) . Feature 3 with a peak maximum from \(1805\mathrm{cm}^{- 1}\) is assigned to the collective oscillation of bridge bound CO to Pt sites and is also in reasonable agreement with previous work \(^{49 - 54,64 - 67}\) . In contrast to the linear CO vibrational features, the identification of unique bridge CO features for different Pt coordination environments was not possible. The \(\mathrm{CO}^{*}\) adsorption site information from the DRIFTS analysis is graphically summarised in Fig. 5b. + +<|ref|>text<|/ref|><|det|>[[113, 542, 882, 890]]<|/det|> +Alongside the DRIFTS investigation, comparable TPO experiments were performed where \(\mathrm{CO}^{*}\) was preadsorbed at temperatures from \(25 - 350^{\circ}\mathrm{C}\) and was titrated using \(\mathrm{O_2}\) pulses while heating from \(25 - 350^{\circ}\mathrm{C}\) at \(8^{\circ}\mathrm{C / min}\) (Fig. 5c). Interestingly, the 25CO TPO shows an onset temperature for \(\mathrm{CO_2}\) production at around \(40^{\circ}\mathrm{C}\) , whereas the TPO results of both 100CO and 200CO show 10- 20 times higher \(\mathrm{CO_2}\) production at \(40^{\circ}\mathrm{C}\) . It is believed that when CO is preadsorbed at \(25^{\circ}\mathrm{C}\) the surface is fully covered in \(\mathrm{CO}^{*}\) meaning there are no vacant sites for adsorption/dissociation of \(\mathrm{O_2}\) (i.e., CO poisoning). We prescribe the increased activity for \(\mathrm{CO_2}\) production in the 100CO and 200CO TPO experiments to the increased number of vacant sites on the partially \(\mathrm{CO}^{*}\) - covered surface. This is supported by an additional TPO experiment where CO is pulsed to saturation over the 100CO catalyst at \(25^{\circ}\mathrm{C}\) . The \(100\mathrm{CO} + 25\mathrm{CO}\) TPO experiment shows similar tendency to that of 25CO (Supplementary Fig. 6), which implies that the low temperature \(\mathrm{CO_2}\) production at \(25^{\circ}\mathrm{C}\) is vacancy driven. Further, isothermal titration experiments performed at \(25^{\circ}\mathrm{C}\) show no \(\mathrm{CO_2}\) production (Supplementary Fig. 7). When combined, this would indicate that \(\mathrm{CO_2}\) production by oxidation of preadsorbed \(\mathrm{CO}^{*}\) is driven by a Langmuir- Hinshelwood type reaction (as is expected for CO oxidation over Pt) \(^{14,15}\) . However, we cannot rule out other methods such as \(\mathrm{CO}^{*}\) assisted \(\mathrm{O_2}\) dissociation \(^{15}\) . Interestingly, the low- temperature oxidation of preadsorbed \(\mathrm{CO}^{*}\) would indicate that the Langmuir- Hinshelwood surface reaction between oxygen and \(\mathrm{CO}^{*}\) over \(2\mathrm{nm}\mathrm{Pt}\) nanoparticles has a very low activation barrier, which is counter to previous work on single crystals and supported Pt catalysts with activation energies ranging from 37 to \(85\mathrm{kJ / mol}^{15,36,68,69}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 106, 882, 300]]<|/det|> +Similar to the isotopically labelled experiments, two kinetic features are observed: a spike around \(80^{\circ}\mathrm{C}\) (first feature) and a long shoulder from \(100–350^{\circ}\mathrm{C}\) (second feature) appear in the 25CO TPO. The first kinetic feature appears at low temperatures \((< 50^{\circ}\mathrm{C})\) in both the 100CO and 200CO which we ascribe to the low- temperature \(\mathrm{CO}^{*}\) conversion mentioned above. Due to decreasing \(\mathrm{CO}^{*}\) coverage throughout the titration experiment, the \(\mathrm{CO}_{2}\) production would be expected to decrease. However, in the 100CO experiment a distinct second catalytic feature appears around \(125^{\circ}\mathrm{C}\) , which suggests the presence of temperature- dependent kinetics. Further, the second catalytic feature disappears as the adsorption temperature increases from 100 to \(200^{\circ}\mathrm{C}\) , which corresponds to the large decrease in WC sites from the DRIFTS results in Fig. 5a. This strongly suggests that site dependent kinetics does exist for the oxidation of \(\mathrm{CO}^{*}\) over \(\mathrm{SiO}_{2}\) - supported \(2\mathrm{nm}\mathrm{Pt}\) nanoparticles. + +<|ref|>text<|/ref|><|det|>[[113, 316, 883, 531]]<|/det|> +To calculate the intrinsic kinetics at each point in the TPO experiments, the rate constants for the model in Fig. 4a were calculated by regressing the MZTRT Symmetric Thin- Zone model to every pulse set in the experiments. All rate constants for the 25CO, 100CO, 200CO, and 350CO TPO experiments are shown in Supplementary Fig. 8. Also, the MATLAB script used to process the 25CO TPO experiment is included alongside the Supplementary Information. As during the TAP experiment the pulse size is sufficiently small that coverage of \(\mathrm{CO}^{*}\) species is not changed by any appreciable amount during a pulse, the first order irreversible adsorption rate constants \(k_{a,1}^{\prime}\) and \(k_{a,2}^{\prime}\) are pseudo first order and are proportional to the concentration of the adsorbed \(\mathrm{CO}^{*}\) species involved in each pathway \((k_{a,3}^{\prime}\) would be proportional to the number of empty sites). This means that it becomes possible to deconvolute the amount of \(\mathrm{CO}_{2}\) produced from each pathway as shown in in Fig. 5d- g. From the MZTRT Symmetric Thin- Zone model, the conversion of \(\mathrm{O}_{2}\) in each pulse set can be calculated using \(^{40,70}\) : + +<|ref|>equation<|/ref|><|det|>[[303, 555, 828, 600]]<|/det|> +\[X_{O_{2}} = \frac{\left(k_{a,1}^{\prime} + k_{a,2}^{\prime} + k_{a,3}^{\prime}\right)\left(L / 2D_{e}\right)}{1 + \left(k_{a,1}^{\prime} + k_{a,2}^{\prime} + k_{a,3}^{\prime}\right)\left(L / 2D_{e}\right)} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[113, 626, 882, 739]]<|/det|> +Where \(X_{O_{2}}\) conversion of the reactant, \(D_{e}^{R}\) is the diffusivity of the reactant \((\mathrm{cm}^{2} / \mathrm{cm}^{3})\) , \(L\) is the length of the catalyst bed (cm), and \(k_{a,n}^{\prime}\) represents the apparent pseudo- first order adsorption/reaction constant for each individual site included in the model (cm \(\mathrm{s}^{- 1}\) ) and is the parameter that is calculated during the model fitting. The apparent pseudo- first order adsorption/reaction constant is linearly related to the intrinsic adsorption/reaction rate constant through the following relationship \(^{41,43}\) : + +<|ref|>equation<|/ref|><|det|>[[340, 764, 828, 803]]<|/det|> +\[k_{a,n}^{\prime} = k_{a,n}\theta_{n}L_{cat}\frac{S_{v}(1 - \epsilon_{b})}{\epsilon_{b}} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[113, 834, 882, 908]]<|/det|> +Where \(k_{a,n}\) is the intrinsic adsorption/reaction constant \((\mathrm{cm}^{3} \mathrm{mol} \mathrm{s}^{- 1} \mathrm{s}^{- 1})\) and \(\theta_{n}\) is the concentration of \(\mathrm{CO}^{*}\) in the case of sites 1 and 2, and the coverage of empty irreversible adsorption sites in the case of site 3 \((\mathrm{mol} \mathrm{cm}^{- 3})\) , \(L_{cat}\) is the length of the catalyst zone \((\mathrm{cm}) S_{v}\) is the surface area of the catalyst per volume of catalyst \((\mathrm{cm}^{2} / \mathrm{cm}^{3})\) . This is a slight modification to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 164]]<|/det|> +forms of the equation previously published40, as the MZTRT Symmetric Thin- Zone model does not explicitly include a value for \(L_{cat}\) and so it is lumped into the apparent pseudo- first order adsorption/reaction constant \(k_{a,n}^{\prime}\) . For each pathway included in the model, the conversion of oxygen specific to each pathway can be deconvoluted using the following: + +<|ref|>equation<|/ref|><|det|>[[297, 180, 829, 225]]<|/det|> +\[X_{O_{2},n} = \frac{k_{a,n}^{\prime}(L / 2D_{e})}{1 + \left(k_{a,1}^{\prime} + k_{a,2}^{\prime} + k_{a,3}^{\prime}\right)(L / 2D_{e})} \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[113, 245, 883, 283]]<|/det|> +The amount of \(\mathrm{CO}_{2}\) produced through can be calculated from the conversion of oxygen specific to pathways 1 and 2 by considering the reaction stoichiometry using the following: + +<|ref|>equation<|/ref|><|det|>[[383, 310, 829, 335]]<|/det|> +\[M_{0,norm}^{CO_{2},n} = 2X_{O_{2},n} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[113, 365, 883, 507]]<|/det|> +From the pathway deconvolution in Fig. 5d- g, we find that pathway 1 is dominant for \(\mathrm{CO}_{2}\) production at low- temperatures, whereas at elevated temperatures pathway 2 becomes the dominant pathway. The \(\mathrm{CO}_{2}\) yield from each pathway can be easily calculated by summing the deconvoluted \(\mathrm{CO}_{2}\) production throughout the TPO experiment (Supplementary Fig. 9). We find that the \(\mathrm{CO}_{2}\) yield of pathway 1 stays relatively constant up to a CO adsorption temperature of \(200^{\circ}\mathrm{C}\) , whereas the \(\mathrm{CO}_{2}\) yield of pathway 2 rapidly decreases. Further, when \(\mathrm{CO}\) is preadsorbed at \(350^{\circ}\mathrm{C}\) , the amount of \(\mathrm{CO}\) species related to pathways 1 and 2 becomes approximately equal on the surface. + +<|ref|>text<|/ref|><|det|>[[113, 522, 883, 767]]<|/det|> +To identify if any correlation between the two pathways and geometric surface sites exists, we summarised the ratio of the \(\mathrm{CO}_{2}\) yield of pathway 1 to pathway 2 calculated from the kinetic model with the quantitative DRIFTS analysis for the population of linear UC and WC \(\mathrm{CO}^{*}\) sites in Fig. 6. The fraction of pathway 1 from the kinetic model (Fig. 6, black) increases with increasing CO adsorption temperature. From the quantitative deconvolution of the DRIFTS spectra for the population of UC and WC \(\mathrm{CO}^{*}\) sites (see Supplementary III for details on quantification), the fraction of populated linear CO at UC sites is also increasing with CO adsorption temperature. Deconvolution of the bridge CO to WC and UC sites is not possible, but there is a shift to lower frequency of the bridge peak maximum with temperature that is consistent with increasing the relative population of UC Pt sites. Most notably, we find a straightforward relationship between the amount of each pathway as calculated from the kinetic model, and the total amount of UC and WC sites calculated using DRIFTS. Due to this direct relationship, we claim that pathway 1 mainly occurs on UC sites, and pathway 2 occurs mainly at WC sites. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[264, 90, 732, 373]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 377, 883, 450]]<|/det|> +
Fig. 6. Comparison between pathway ratio (TPO experiment) and area ratio of \(\mathbf{CO}^*\) sites (DRIFTS investigation). A direct correlation between the total amount of pathway 1 and pathway 2 calculated from the kinetic model, and the total amount of UC and WC sites from the DRIFTS investigation is found.
+ +<|ref|>text<|/ref|><|det|>[[113, 465, 883, 872]]<|/det|> +Measuring the Intrinsic Kinetics of CO oxidation on UC and WC sites. Along with the reactive site information, the intrinsic surface reaction constants \(k_{r,1}\) and \(k_{r,2}\) (related to the reaction of adsorbed oxygen with adsorbed CO for pathways 1 and 2) from the isothermal titration and TPO experiments were calculated and are shown in Fig. 7. As mentioned previously, the reaction over UC sites is ascribed to pathway 1, whereas the reaction over WC sites is ascribed to pathway 2. It can be seen that the rate constant for pathway 1 \((k_{r,1})\) is significantly lower than pathway 2 \((k_{r,2})\) under all conditions probed, with pathway 1 being a slow reaction between adsorbed oxygen and CO and pathway 2 being fast. Interestingly, pathway 1 shows no dependence on temperature or the coverage of CO in both the isothermal O2 titration (Fig. 7a,b) and the TPO (Fig. 7c,d) experiments but is unintuitively a slow reaction. One such rationalisation for this behaviour based on transition state theory would be that the reaction results in a significant loss in entropy in the transition state \(^{71}\) . As adsorbed oxygen is considered to be immobile under these conditions, it could be hypothesized that a CO molecule strongly bound to a vertex of a nanoparticle has a large number of degrees of freedom, but during the transition state to make \(\mathrm{CO_2}\) the species would become immobile due to the strongly bound oxygen, losing a large number of degrees of freedom. However, as CO adsorption on Pt is not well described by DFT \(^{72,73}\) , this would be difficult to confirm, so this idea remains conceptual. Experiments using molecular beam scattering \(^{14,68}\) or velocity resolved kinetics \(^{74}\) do not observe this barrierless reaction, but it should be noted that these were performed on Pt single crystals, whereas this work is performed on 2 nm Pt supported nanoparticles, and as such the nanoparticle size effect must also be considered. Further, due to the slow nature of the reaction, this pathway would be difficult to isolate as it would certainly be limited by CO desorption under steady-state conditions at any appreciable CO pressure. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 882, 300]]<|/det|> +We find that pathway 2 shows more classic kinetic behaviour. Under isothermal conditions we find a linear dependence on the coverage of \(\mathrm{CO}^*\) (Fig. 7b) indicating lateral interactions between adsorbed \(\mathrm{CO}^*\) molecules and/or adsorbed \(\mathrm{O}^*\) , which is expected based on previous work36. Under non-isothermal conditions an exponential increase in rate with increasing temperature (Fig. 7d) is observed above \(80^{\circ}\mathrm{C}\) . It should be noted that as the surface coverage is also changing during the TPO experiment, a classic Arrhenius style analysis to calculate activation energies and pre-exponential factors is non-trivial and will be attempted in future publications. Further, we observed complex kinetic behaviour in pathway 2 below \(80^{\circ}\mathrm{C}\) (which was not recreated in the isothermal \(\mathrm{O}_2\) titration experiments) which cannot be described by classical Arrhenius kinetics. It could be hypothesized that some restructuring of the Pt nanoparticles is occurring around \(100^{\circ}\mathrm{C}\) , but this will be evaluated in future work combining these methods with techniques such as X- ray Absorption Spectroscopy. + +<|ref|>image<|/ref|><|det|>[[196, 316, 808, 727]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 736, 882, 880]]<|/det|> +
Fig. 7. Calculated rate constants \((k_{r,1}\) and \(k_{r,2}\) ) for the surface reaction between adsorbed oxygen and CO from the three-pathway kinetic model with \(95\%\) confidence intervals overlaid. a,b, Isothermal titration experiments where CO was adsorbed at 100 and \(200^{\circ}\mathrm{C}\) . Increasing total \(\mathrm{CO}_2\) produced is correlated with decreasing CO coverage. c,d, TPO experiments where CO was adsorbed at temperature ranging from \(25 - 200^{\circ}\mathrm{C}\) . When the production of \(\mathrm{CO}_2\) is sufficiently low in the TPO experiment \((>200^{\circ}\mathrm{C})\) the signal/noise ratio of the \(\mathrm{CO}_2\) exit flux curves significantly decreases, which in turn decreases the confidence in the model fitting, particularly for pathway 2, as shown in Supplementary Fig. 12.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 130, 880, 375]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 401, 881, 437]]<|/det|> +
Fig. 8. Summary of the three-pathway model for the reaction of \(\mathbf{O}_2\) with preabsorbed \(\mathbf{CO}^*\) over \(\mathbf{SiO}_2\) -supporrted 2 nm Pt nanoparticles.
+ +<|ref|>text<|/ref|><|det|>[[113, 452, 882, 732]]<|/det|> +The combination of TAP, kinetic modelling, and DRIFTS measurements provides unparalleled levels of kinetic insight into the dynamic site dependent activity for CO oxidation over 2 nm- sized \(\mathrm{Pt / SiO_2}\) catalysts. The precisely defined nature of the TAP experiment means that fine kinetic features are resolvable by modelling the exit flux curves using efficient analytical functions. By coupling this insight with DRIFTS measurements, we have been able to identify that site- specific kinetics exists for CO oxidation over 2 nm Pt nanoparticles with the three pathways for the interaction of oxygen with \(\mathrm{CO}^*\) summarised in Fig. 8. We find two pathways for the oxidation of \(\mathrm{CO}^*\) and one pathway for the irreversible adsorption of oxygen (no reaction). Pathway 1 mainly occurs at the UC sites and has slow kinetics, is coverage independent, and is temperature independent. On the contrary, pathway 2 mainly occurs at WC sites with fast kinetics, is highly dependent on the \(\mathrm{CO}^*\) or \(\mathrm{O}^*\) coverage and shows an exponential increase with temperature. These results serve as a significant insight into understanding the kinetics of various reactive sites in heterogeneous catalysis. Typically, it has been widely accepted that increasing the number of UC sites is advantageous for CO conversion22,27. However, even though the reaction at the UC sites is found to be barrierless, it has slow kinetics which may not be optimal for CO conversion. + +<|ref|>text<|/ref|><|det|>[[114, 748, 882, 870]]<|/det|> +Once the reaction pathways have been assigned to specific sites, this method of kinetic site deconvolution can be applied generally, allowing features such as restructuring, sintering, and even potentially metal- support interactions to be understood. The quantitative nature of the TAP experiment means simultaneous calculation of the number of active sites, their distribution on the surface, and the intrinsic kinetics of the surface reactions is now possible. We believe that our approach is general enough to apply to other catalytic systems and can serve as a new toolkit in the characterisation of heterogeneous catalysts. + +<|ref|>sub_title<|/ref|><|det|>[[115, 870, 190, 885]]<|/det|> +## Methods + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 883, 417]]<|/det|> +Preparation of Pt/SiO₂ catalyst. In a typical synthesis of 2 nm- sized Pt nanoparticles, \(^{75}\) 100 mg of chloroplatinic acid hydrate \((\mathrm{H}_{2}\mathrm{PtCl}_{6} \cdot \mathrm{xH}_{2}\mathrm{O}, 99.9\%)\) , Sigma- Aldrich), 20 mg of poly(vinylpyrrolidone) (PVP, \(\mathrm{M}_{\mathrm{w}} = 40000\) , Sigma- Aldrich), 2.5 mL of sodium hydroxide (NaOH, 1 N) were dissolved in 10 mL of ethylene glycol (Sigma- Aldrich) in a 50 mL three- neck roundbottom flask. The flask was heated to 80 °C and evacuated for 30 min with vigorous stirring, then heated to 200 °C. The solution was kept at 200 °C for 2 hr with Ar gas purging. After the reaction, the solution cooled down to RT. The colloidal suspension was diluted to 50 mL of acetone and centrifuged at 6500 rpm for 10 min twice, repeatedly. Then, the as- synthesized Pt nanoparticles were re- dispersed in 40 mL of ethanol. Then, 10 mL of the Pt- dispersed solution was dropped onto the 0.3 g of \(\mathrm{SiO}_{2}\) powder (pretreated at 700 °C for 1 hr; 10–20 nm, Sigma- Aldrich) with vigorous stirring. The suspension was sonicated for 20 min, subsequently evaporating the solvent in a vacuum at 50 °C overnight. The microstructure of the Pt/SiO₂ catalyst was investigated via XRD measurement (Supplementary Fig. 1b) and shows diffraction peaks at 39.5 and 46°, corresponding to the (111) and (200) planes of reference Pt (JCPDS #04- 0802). For characterisation and TAP experiments, the Pt/SiO₂ catalyst was calcined at 500 °C for 1 hr in air condition to remove majority of carbonaceous capping agent with a ramp up rate of 1 °C/min. To prepare the metallic Pt nanoparticles, the Pt/SiO₂ catalyst was reduced at 300 °C for 1 hr with a ramp up rate of 1 °C/min under 10% \(\mathrm{H}_{2}\) in Ar flow. + +<|ref|>text<|/ref|><|det|>[[114, 433, 882, 558]]<|/det|> +Characterisations. The morphology and size distribution of Pt/SiO₂ catalyst were investigated by transmission electron microscopy (TEM; ARM 200F, JEOL) at 200 kV. The concentration of catalyst was determined using inductively coupled plasma optical emission spectroscopy (ICP- OES; 5110 ICP- OES; Agilent). The catalyst microstructure was measured by X- ray diffractometer (XRD; D2 Phaser, Bruker). The steady- state flow catalytic activity was measured in a home- built ambient pressure flow reactor that has been previously described \(^{76}\) with all gas mixtures reported balanced in Ar. + +<|ref|>text<|/ref|><|det|>[[114, 574, 882, 749]]<|/det|> +Temporal Analysis of Products experiments. The TAP technique has been described extensively in the literature \(^{41,42,77}\) , but it is summarised here. During the TAP experiment a nanomole pulse of gas \((\sim 10^{15}\) molecules, 108 \(\mu\) s pulse width) is sent into a packed bed microreactor that is held at ultra- high vacuum \((\sim 10^{- 9}\) torr). During the experiment, the pulsed gas diffuses through the packed bed via Knudsen Diffusion where it can interact with the catalyst surface. Eventually the gas diffuses out the exit of the microreactor and the exit flux of is measured via mass spectrometry. Due to the precisely defined nature of Knudsen Diffusion, the shape (and magnitude) of the exit flux curves provides highly resolved kinetic insight, in particular when coupled with kinetic modelling of the exit flux curves (see section: Modelling of Temporal Analysis of Products Pulse Responses). + +<|ref|>text<|/ref|><|det|>[[114, 765, 882, 906]]<|/det|> +For the TAP experiments, the Pt/SiO₂ catalyst is first calcined at 350 °C by injecting 1000 pulse sets of large O₂ pulses (160 \(\mu\) s pulse width) until the CO₂ signal (m/z = 44) becomes near- zero to minimise the decomposition of the remaining carbonaceous capping. Before all pulse/transient response experiments, the Pt/SiO₂ catalyst is reduced at 350 °C by injecting 600 pulse sets of large H₂ pulses (160 \(\mu\) s pulse width) to achieve a metallic Pt surface. We utilise a home- built TAP reactor \(^{38}\) where the microreactor contains a layer of commercial sand (29.7 mm; 50–70 mesh SiO₂; Sigma- Aldrich) followed by a layer of Pt/SiO₂ catalyst (5.4 mg) followed by a final layer of commercial sand (34.4 mm) for a total reactor length of 64.1 mm. The exit flux of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 882, 179]]<|/det|> +CO, \(\mathrm{O_2}\) , Ar, and \(\mathrm{CO_2}\) is monitored via mass spectrometry. The integrated exit flux of the Ar tracer is used for normalisation of all pulse experiments (see Supplementary IV). As the mass spectrometer can only investigate one \(\mathrm{m / z}\) value per pulse, multiple pulses are used to scan the whole range of \(\mathrm{m / z}\) values and are combined to one pulse set38. All gas mixtures reported are balanced in Ar. + +<|ref|>text<|/ref|><|det|>[[112, 195, 882, 579]]<|/det|> +Diffused Reflectance Infrared Fourier Transform Spectroscopy experiments. DRIFTS experiments were carried out in a low- temperature reaction chamber (Harrick Scientific) equipped with ZnSe windows, mounted inside the sample compartment of a Bruker Invenio FT- IR spectrometer using a Praying Mantis diffuse reflectance accessory (Harrick Scientific). The catalyst sample was prepared by pressing approximately \(2\mathrm{mg}\) of \(2\mathrm{nm}\) Pt/SiO2 onto a 304 stainless- steel mesh ( \(150 \times 150\) mesh). The DRIFTS reactor was loaded by placing the catalyst- containing mesh on top of approximately \(110\mathrm{mg}\) of \(120\mathrm{gr}\) SiC, an inert support with high thermal conductivity. There can exist large temperature gradients between the thermocouple contact in a DRIFTS reactor cell and the catalyst surface temperature exposed to the infrared beam.78 Therefore, a thermocouple was mounted in physical contact with the bottom of the stainless- steel mesh and the temperature gradient to the catalyst surface at \(350^{\circ}\mathrm{C}\) was less than \(20^{\circ}\mathrm{C}\) as calibrated by an optical pyrometer. All DRIFTS experiments used a total volumetric flow rate of \(100\mathrm{scm}\) . Each absorbance spectrum was obtained by averaging 200 background and sample scans at a resolution of \(4\mathrm{cm}^{- 1}\) using a liquid- nitrogen- cooled \(\mathrm{HgCdTe}\) (MCT) detector, while the Praying Mantis diffuse reflectance accessory and FT- IR spectrometer was purged with dry \(\mathrm{N_2}\) . The background measurement was acquired after the catalyst sample was annealed at \(350^{\circ}\mathrm{C}\) for \(30\mathrm{min}\) in \(5\%\) \(\mathrm{H_2}\) in Ar and cooled to \(35^{\circ}\mathrm{C}\) in Ar. The sample measurements were acquired after the catalyst sample was annealed at \(30^{\circ}\mathrm{C / min}\) in Ar to the CO adsorption temperature, the temperature was maintained for \(10\mathrm{min}\) in \(0.1\%\) CO in Ar until saturation was achieved, and the sample was cooled to \(35^{\circ}\mathrm{C}\) in Ar. A description of the quantitative analysis of the DRIFTS spectra is provided in the Supplementary (see Supplementary III). + +<|ref|>text<|/ref|><|det|>[[112, 594, 882, 897]]<|/det|> +Modelling of Temporal Analysis of Products Pulse Responses. To simulate the TAP exit flux response curves, Multi- Zone TAP Reactor Theory43,44 (MZTRT) was utilised with the catalyst zone being approximated as a Thin Zone40 in the centre of the microreactor between two layers of inert packing. To perform the curve fitting first, the experimentally measured signals were normalised to the inert Ar tracer and corrected using their corresponding calibration factors (see Supplementary IV). Then, the curves were further normalised to the concentration of reactant gas in the pulsed mixture such that the integrated area under the reactant curve is 0 at \(100\%\) conversion and 1 at \(0\%\) conversion. Next, the diffusivity of Argon in the packed bed reactor was calculated by fitting a one- zone TAP model to the Ar exit flux curve. As the Knudsen diffusivity is proportional to \(\sqrt{1 / M}\) where \(M\) is the molecular weight of the gas, it becomes possible to calculate the diffusivity of the reactant ( \(\mathrm{O_2}\) , \(M = 32\) ) and product ( \(\mathrm{CO_2}\) , \(M = 44\) ) gases by scaling the diffusivity relative to the inert gas (Ar, \(M = 40\) ). When performing the fitting of the reactant and product curves, the diffusivities of the gases, the reactor length, and the void fraction of the reactor were all fixed, with the only variables being the rate constants for the corresponding model. It is very important to note the regression is performed on each set of exit flux response curves (i.e., exit flux plotted as a function of time) individually. Therefore, the rate constants are calculated separately at each pulse set (and catalyst state) during the experiment. All curve fitting + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 883, 179]]<|/det|> +was performed in the MATLAB environment using the lsqcurvefit function, with the \(95\%\) confidence intervals for the fitted variables evaluated using the nlparci function. A full description of the MZTRT model used in this work and how the model fitting is performed is available in Supplementary Information I, II, IV, and an example of the MATLAB script used to simulate the TAP experiments is included alongside this paper. + +<|ref|>sub_title<|/ref|><|det|>[[115, 195, 255, 212]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 230, 882, 300]]<|/det|> +The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files. An example MATLAB script for the modelling of the TAP response curves is also included alongside the data for the 25TPO experiment. All other relevant data is available from the authors upon reasonable request. + +<|ref|>sub_title<|/ref|><|det|>[[115, 317, 280, 334]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[115, 350, 880, 370]]<|/det|> +C.R. gratefully acknowledges the Rowland Fellowship through the Rowland Institute at Harvard. + +<|ref|>sub_title<|/ref|><|det|>[[115, 386, 295, 404]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 420, 880, 494]]<|/det|> +Taek- Seung Kim: Conceptualization, Investigation, Formal analysis, Writing - original draft Christopher R. O'Connor: Investigation, Writing - original draft Christian Reece: Conceptualization, Formal analysis, Supervision, Writing - original draft, Writing - review & editing + +<|ref|>text<|/ref|><|det|>[[115, 508, 882, 561]]<|/det|> +T.S.K performed the synthesis, characterisation and TAP experiments. C.R.O. performed the DRIFTS experiments. C.R. performed the TAP modelling and supervised the project. All authors in frequent discussions and contributed significantly to writing the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[115, 578, 312, 595]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[115, 611, 882, 648]]<|/det|> +Supplementary information accompanies this paper. The MATLAB script accompanying this paper requires the Curve Fitting Toolbox and the Statistics and Machine Learning Toolbox. + +<|ref|>sub_title<|/ref|><|det|>[[115, 664, 363, 683]]<|/det|> +## Competing financial interests + +<|ref|>text<|/ref|><|det|>[[115, 699, 532, 717]]<|/det|> +The authors declare no competing financial interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 734, 208, 751]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[115, 768, 857, 821]]<|/det|> +1. Mou, T. et al. Bridging the complexity gap in computational heterogeneous catalysis with machine learning. Nat. Catal. 6, 122-136 (2023). + +<|ref|>text<|/ref|><|det|>[[115, 837, 870, 891]]<|/det|> +2. Zhdanov, V. P. Impact of surface science on the understanding of kinetics of heterogeneous catalytic reactions. Surf. Sci. 500, 966-985 (2002). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 860, 144]]<|/det|> +3. Libuda, J. & Freund, H. J. Molecular beam experiments on model catalysts. Surf. Sci. Rep. 57, 157-298 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 157, 857, 248]]<|/det|> +4. Van Spronsen, M. A., Frenken, J. W. M. & Groot, I. M. N. Surface science under reaction conditions: CO oxidation on Pt and Pd model catalysts. Chem. Soc. Rev. 46, 4347-4374 (2017). + +<|ref|>text<|/ref|><|det|>[[110, 262, 840, 355]]<|/det|> +5. Nguyen, L., Tao, F. F., Tang, Y., Dou, J. & Bao, X.-J. Understanding catalyst surfaces during catalysis through near ambient pressure X-ray photoelectron spectroscopy. Chem. Rev. 119, 6822-6905 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 367, 882, 423]]<|/det|> +6. Lee, J. D. et al. Dilute alloys based on Au, Ag, or Cu for efficient catalysis: from synthesis to active sites. Chem. Rev. 122, 8758-8808 (2022). + +<|ref|>text<|/ref|><|det|>[[110, 435, 875, 527]]<|/det|> +7. Reece, C. & Madix, R. J. Moving from fundamental knowledge of kinetics and mechanisms on surfaces to prediction of catalyst performance in reactors. ACS Catal. 11, 3048-3066 (2021). + +<|ref|>text<|/ref|><|det|>[[110, 540, 860, 596]]<|/det|> +8. Haruta, M. et al. Low-temperature oxidation of CO over fold supported on TiO₂, α-Fe₂O₃, and Co₃O₄. J. Catal. 144, 175-192 (1993). + +<|ref|>text<|/ref|><|det|>[[110, 609, 879, 700]]<|/det|> +9. Reske, R., Mistry, H., Behafarid, F., Roldan Cuenya, B. & Strasser, P. Particle size effects in the catalytic electroreduction of CO₂ on Cu nanoparticles. J. Am. Chem. Soc. 136, 6978-6986 (2014). + +<|ref|>text<|/ref|><|det|>[[110, 714, 870, 805]]<|/det|> +10. Freund, H.-J., Meijer, G., Scheffler, M., Schlögl, R. & Wolf, M. CO oxidation as a prototypical reaction for heterogeneous processes. Angew. Chem. Int. Ed. 50, 10064-10094 (2011). + +<|ref|>text<|/ref|><|det|>[[110, 819, 787, 875]]<|/det|> +11. Kim, T.-S. et al. Catalytic synergy on PtNi bimetal catalysts driven by interfacial intermediate structures. ACS Catal. 10, 10459-10467 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 852, 145]]<|/det|> +12. Kim, T.-S. et al. Catalytic boosting on AuCu bimetallic nanoparticles by oxygen-induced atomic restructuring. Appl. Catal. B: Environ. 331, 122704 (2023). + +<|ref|>text<|/ref|><|det|>[[110, 157, 850, 213]]<|/det|> +13. Maurer, F. et al. Tracking the formation, fate and consequence for catalytic activity of Pt single sites on \(\mathrm{CeO_2}\) . Nat. Catal. 3, 824–833 (2020). + +<|ref|>text<|/ref|><|det|>[[110, 227, 865, 283]]<|/det|> +14. Campbell, C. T., Ertl, G., Kuipers, H. & Segner, J. A molecular beam study of the catalytic oxidation of CO on a Pt(111) surface. J. Chem. Phys. 73, 5862–5873 (1980). + +<|ref|>text<|/ref|><|det|>[[110, 296, 840, 353]]<|/det|> +15. Allian, A. D. et al. Chemisorption of CO and mechanism of CO oxidation on supported platinum nanoclusters. J. Am. Chem. Soc. 133, 4498–4517 (2011). + +<|ref|>text<|/ref|><|det|>[[110, 366, 836, 422]]<|/det|> +16. Yang, X.-F. et al. Single-atom catalysts: a new frontier in heterogeneous catalysis. Acc. Chem. Res. 46, 1740–1748 (2013). + +<|ref|>text<|/ref|><|det|>[[110, 435, 875, 491]]<|/det|> +17. Ding, R. et al. Size-dependent shape distributions of platinum nanoparticles. Nanoscale Adv. 4, 3978–3986 (2022). + +<|ref|>text<|/ref|><|det|>[[110, 505, 864, 561]]<|/det|> +18. Vogt, C. et al. Dynamic restructuring of supported metal nanoparticles and its implications for structure insensitive catalysis. Nat. Commun. 12, 7096 (2021). + +<|ref|>text<|/ref|><|det|>[[110, 575, 863, 666]]<|/det|> +19. Chee, S. W., Arce-Ramos, J. M., Li, W., Genest, A. & Mirsaidov, U. Structural changes in noble metal nanoparticles during CO oxidation and their impact on catalyst activity. Nat. Commun. 11, 2133 (2020). + +<|ref|>text<|/ref|><|det|>[[110, 680, 860, 736]]<|/det|> +20. Somorjai, G. A. & Carrazza, J. Structure sensitivity of catalytic reactions. Ind. Eng. Chem. Fund. 25, 63–69 (1986). + +<|ref|>text<|/ref|><|det|>[[110, 750, 874, 806]]<|/det|> +21. Beniya, A. et al. CO oxidation activity of non-reducible oxide-supported mass-selected few-atom Pt single-clusters. Nat. Commun. 11, 1888 (2020). + +<|ref|>text<|/ref|><|det|>[[110, 820, 857, 876]]<|/det|> +22. Vogt, C. & Weckhuysen, B. M. The concept of active site in heterogeneous catalysis. Nat. Rev. Chem. 6, 89–111 (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 872, 178]]<|/det|> +23. Wei, J. & Iglesia, E. Mechanism and site requirements for activation and chemical conversion of methane on supported Pt clusters and turnover rate comparisons among noble metals. J. Phys. Chem. B 108, 4094–4103 (2004). + +<|ref|>text<|/ref|><|det|>[[110, 193, 840, 283]]<|/det|> +24. Ramallo-López, J. M. et al. Complementary methods for cluster size distribution measurements: supported platinum nanoclusters in methane reforming catalysts. J. Mol. Catal. A-Chem. 228, 299–307 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 298, 866, 353]]<|/det|> +25. Engel, T. & Ertl, G. A molecular beam investigation of the catalytic oxidation of CO on Pd (111). J. Chem. Phys. 69, 1267–1281 (1978). + +<|ref|>text<|/ref|><|det|>[[110, 367, 845, 457]]<|/det|> +26. Kale, M. J. & Christopher, P. Utilizing quantitative in situ FTIR spectroscopy to identify well-coordinated Pt atoms as the active site for CO oxidation on \(\mathrm{Al}_2\mathrm{O}_3\) -supported Pt catalysts. ACS Catal. 6, 5599–5609 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 472, 833, 527]]<|/det|> +27. Neumann, S. et al. Insights into the reaction mechanism and particle size effects of CO oxidation over supported Pt nanoparticle catalysts. J. Catal. 377, 662–672 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 541, 826, 596]]<|/det|> +28. Darshini Srinivasan, P. In situ and operando tools and methods for characterization of heterogeneous catalysts. (University of Kansas, 2019). + +<|ref|>text<|/ref|><|det|>[[110, 611, 879, 665]]<|/det|> +29. Yablonsky, G. S. Decoding complexity of chemical reactions. Theor. Found. Chem. Eng. 48, 608–613 (2014). + +<|ref|>text<|/ref|><|det|>[[110, 680, 858, 771]]<|/det|> +30. Reece, C., Redekop, E. A., Karakalos, S., Friend, C. M. & Madix, R. J. Crossing the great divide between single-crystal reactivity and actual catalyst selectivity with pressure transients. Nat. Catal. 1, 852–859 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 786, 874, 875]]<|/det|> +31. Karatok, M., Madix, R. J., Van Der Hoeven, J. E. S., Aizenberg, J. & Reece, C. Quantifying oxygen induced surface enrichment of a dilute PdAu alloy catalyst. Catal. Sci. Technol. 11, 7530–7534 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 857, 178]]<|/det|> +32. Widmann, D. & Behm, R. J. Activation of molecular oxygen and the nature of the active oxygen species for CO oxidation on oxide supported au catalysts. Accounts Chem. Res. 47, 740-749 (2014). + +<|ref|>text<|/ref|><|det|>[[110, 192, 875, 248]]<|/det|> +33. Shekhtman, S. O., Yablonsky, G. S., Glevaes, J. T. & Fushimi, R. R. Thin-zone TAP reactor as a basis of 'state-by-state transient screening'. Chem. Eng. Sci. 59, 5493-5500 (2004). + +<|ref|>text<|/ref|><|det|>[[110, 261, 863, 355]]<|/det|> +34. Shekhtman, S. O., Goguet, A., Burch, R., Hardacre, C. & Maguire, N. CO multipulse TAP studies of 2% Pt/CeO₂ catalyst: Influence of catalyst pretreatment and temperature on the number of active sites observed. J. Catal. 253, 303-311 (2008). + +<|ref|>text<|/ref|><|det|>[[110, 368, 812, 424]]<|/det|> +35. Ding, K. et al. Identification of active sites in CO oxidation and water-gas shift over supported Pt catalysts. Science 350, 189-192 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 436, 830, 529]]<|/det|> +36. Bourane, A. & Bianchi, D. Oxidation of CO on a Pt/Al₂O₃ catalyst: from the surface elementary steps to lighting-off tests: II. kinetic study of the oxidation of adsorbed CO species using mass spectroscopy. J. Catal. 209, 114-125 (2002). + +<|ref|>text<|/ref|><|det|>[[110, 542, 833, 598]]<|/det|> +37. Engel, T. & Ertl, G. Elementary steps in the catalytic oxidation of carbon monoxide on platinum metals. in Advances in Catalysis vol. 28 1-78 (Elsevier, 1979). + +<|ref|>text<|/ref|><|det|>[[110, 610, 877, 666]]<|/det|> +38. Brandão, L., High, E., Kim, T.-S. & Reece, C. Simplifying the temporal analysis of products reactor. Preprint at https://doi.org/10.26434/chemrxiv-2023-d1wgj (2023). + +<|ref|>text<|/ref|><|det|>[[110, 679, 875, 771]]<|/det|> +39. Shekhtman, S. O., Maguire, N., Goguet, A., Burch, R. & Hardacre, C. Three primary kinetic characteristics observed in a pulse-response TAP experiment. Catal. Today 121, 255-260 (2007). + +<|ref|>text<|/ref|><|det|>[[110, 785, 841, 841]]<|/det|> +40. Shekhtman, S. O., Yablonsky, G. S., Chen, S. & Glevaes, J. T. Thin-zone TAP-reactor - theory and application. Chem. Eng. Sci. 54, 4371-4378 (1999). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 877, 144]]<|/det|> +41. Gleaves, J. T., Yablonskii, G. S., Phanawadee, P. & Schuurman, Y. TAP-2: An interrogative kinetics approach. Appl. Catal. A-Gen. 160, 55-88 (1997). + +<|ref|>text<|/ref|><|det|>[[110, 157, 825, 213]]<|/det|> +42. Fushimi, R. Temporal analysis of product (TAP). in Springer Handbook of Advanced Catalyst Characterization 899-934 (Springer, 2023). + +<|ref|>text<|/ref|><|det|>[[110, 227, 870, 319]]<|/det|> +43. Constales, D., Yablonsky, G. S., Marin, G. B. & Gleaves, J. T. Multi-zone tap-reactors theory and application: I. The global transfer matrix equation. Chem. Eng. Sci. 56, 133-149 (2001). + +<|ref|>text<|/ref|><|det|>[[110, 333, 854, 424]]<|/det|> +44. Constales, D., Yablonsky, G. S., Marin, G. B. & Gleaves, J. T. Multi-zone TAP-reactors theory and application. III Multi-response theory and criteria of instantaneousness. Chem. Eng. Sci. 59, 3725-3736 (2004). + +<|ref|>text<|/ref|><|det|>[[110, 437, 863, 493]]<|/det|> +45. Geissler, P. L. Temperature dependence of inhomogeneous broadening: on the meaning of isosbestic points. J. Am. Chem. Soc. 127, 14930-14935 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 506, 818, 562]]<|/det|> +46. Hammer, B., Nielsen, O. H. & Nørskov, J. K. Structure sensitivity in adsorption: CO interaction with stepped and reconstructed Pt surfaces. Catal. Lett. 46, 31-35 (1997). + +<|ref|>text<|/ref|><|det|>[[110, 576, 839, 631]]<|/det|> +47. Meunier, F. C., Elgayyar, T., Dembélé, K. & Kaper, H. Stability of Pt-Adsorbed CO on Catalysts for Room Temperature-Oxidation of CO. Catalysts 12, 532 (2022). + +<|ref|>text<|/ref|><|det|>[[110, 645, 848, 701]]<|/det|> +48. Lundwall, M. J., McClure, S. M. & Goodman, D. W. Probing terrace and step sites on Pt nanoparticles using CO and ethylene. J. Phys. Chem. C 114, 7904-7912 (2010). + +<|ref|>text<|/ref|><|det|>[[110, 714, 840, 770]]<|/det|> +49. Hayden, B. E. & Bradshaw, A. M. The adsorption of CO on Pt(111) studied by infrared reflection-absorption spectroscopy. Surf. Sci. 125, 787-802 (1983). + +<|ref|>text<|/ref|><|det|>[[110, 784, 848, 840]]<|/det|> +50. Carrasco, E. et al. Infrared spectra of high coverage CO adsorption structures on Pt(111). Surf. Sci. 604, 1320-1325 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 875, 177]]<|/det|> +51. Hopster, H. & Ibach, H. Adsorption of CO on Pt(111) and Pt 6(111) × (111) studied by high resolution electron energy loss spectroscopy and thermal desorption spectroscopy. Surf. Sci. 77, 109-117 (1978). + +<|ref|>text<|/ref|><|det|>[[110, 193, 830, 248]]<|/det|> +52. McClellan, M. R., Gland, J. L. & McFeeley, F. R. Carbon monoxide adsorption on the kinked Pt(321) surface. Surf. Sci. 112, 63-77 (1981). + +<|ref|>text<|/ref|><|det|>[[110, 262, 881, 317]]<|/det|> +53. Yoshinobu, J., Tsukahara, N., Yasui, F., Mukai, K. & Yamashita, Y. Lateral displacement by transient mobility in chemisorption of CO on Pt(997). Phys. Rev. Lett. 90, 248301 (2003). + +<|ref|>text<|/ref|><|det|>[[110, 331, 881, 422]]<|/det|> +54. Luo, J. S., Tobin, R. G., Lambert, D. K., Fisher, G. B. & DiMaggio, C. L. CO adsorption site occupation on Pt(335): a quantitative investigation using TPD and EELS. Surf. Sci. 274, 53-62 (1992). + +<|ref|>text<|/ref|><|det|>[[110, 437, 881, 492]]<|/det|> +55. Hayden, B. E., Kretzschmar, K., Bradshaw, A. M. & Greenler, R. G. An infrared study of the adsorption of CO on a stepped platinum surface. Surf. Sci. 149, 394-406 (1985). + +<|ref|>text<|/ref|><|det|>[[110, 507, 870, 562]]<|/det|> +56. Banholzer, W. F. & Masel, R. I. A reflection-absorption infrared study of carbon monoxide and nitric oxide adsorption on platinum(100). Surf. Sci. 137, 339-360 (1984). + +<|ref|>text<|/ref|><|det|>[[110, 576, 825, 670]]<|/det|> +57. Fanson, P. T., Delgass, W. N. & Lauterbach, J. Island formation during kinetic rate oscillations in the oxidation of CO over Pt/SiO₂: a transient Fourier transform infrared spectrometry study. J. Catal. 204, 35-52 (2001). + +<|ref|>text<|/ref|><|det|>[[110, 683, 866, 738]]<|/det|> +58. Brandt, K. & Hughes, M. R. The interpretation of CO adsorbed on Pt/SiO₂ of two different particle-size distributions. Surf. Sci. 286, 15-25 (1993). + +<|ref|>text<|/ref|><|det|>[[110, 752, 876, 843]]<|/det|> +59. Avanesian, T. et al. Quantitative and atomic-scale view of CO-induced Pt nanoparticle surface reconstruction at saturation coverage via DFT calculations coupled with in situ TEM and IR. J. Am. Chem. Soc. 139, 4551-4558 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 840, 178]]<|/det|> +60. Kappers, M. J. & Van Der Maas, J. H. Correlation between CO frequency and Pt coordination number. A DRIFT study on supported Pt catalysts. Catal. Lett. 10, 365–373 (1991). + +<|ref|>text<|/ref|><|det|>[[110, 193, 877, 250]]<|/det|> +61. Hossain, M. J., Rahman, M. M. & Jafar Sharif, Md. Preference for low-coordination sites by adsorbed CO on small platinum nanoparticles. Nanoscale Adv. 2, 1245–1252 (2020). + +<|ref|>text<|/ref|><|det|>[[110, 264, 833, 355]]<|/det|> +62. Liu, A. et al. Getting insights into the temperature-specific active sites on platinum nanoparticles for CO oxidation: A combined in situ spectroscopic and ab Initio density functional theory study. ACS Catal. 9, 7759–7768 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 368, 855, 425]]<|/det|> +63. Kustov, L. M., Ostgard, D. & Sachtler, W. M. H. IR spectroscopic study of Pt/Kl zeolites using adsorption of CO as a molecular probe. Catal. Lett. 9, 121–126 (1991). + +<|ref|>text<|/ref|><|det|>[[110, 438, 860, 495]]<|/det|> +64. Di, M. et al. Chasing \(\mathrm{PtO_x}\) species in ceria supported platinum during CO oxidation extinction with correlative operando spectroscopic techniques. J. Catal. 409, 1–11 (2022). + +<|ref|>text<|/ref|><|det|>[[110, 508, 879, 600]]<|/det|> +65. Klimavicius, V., Neumann, S., Kunz, S., Gutmann, T. & Buntkowsky, G. Room temperature CO oxidation catalysed by supported Pt nanoparticles revealed by solid-state NMR and DNP spectroscopy. Catal. Sci. Technol. 9, 3743–3752 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 612, 856, 668]]<|/det|> +66. Persson, B. N. J., Tüshaus, M. & Bradshaw, A. M. On the nature of dense CO adlayers. J. Chem. Phys. 92, 5034–5046 (1990). + +<|ref|>text<|/ref|><|det|>[[110, 682, 875, 738]]<|/det|> +67. Baró, A. M. & Ibach, H. New study of CO adsorption at low temperature (90 K) on Pt (111) by EELS. J. Chem. Phys. 71, 4812 (1979). + +<|ref|>text<|/ref|><|det|>[[110, 752, 879, 808]]<|/det|> +68. Zaera, F., Liu, J. & Xu, M. Isothermal study of the kinetics of carbon monoxide oxidation on Pt(111): Rate dependence on surface coverages. J. Chem. Phys. 106, 4204–4215 (1997). + +<|ref|>text<|/ref|><|det|>[[110, 821, 836, 877]]<|/det|> +69. Mhadeshwar, A. B. & Vlachos, D. G. A thermodynamically consistent surface reaction mechanism for CO oxidation on Pt. Combust. Flame 142, 289–298 (2005). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 830, 108]]<|/det|> +70. Shekhtman, S. O. & Yablonsky, G. S. Thin-zone TAP reactor versus differential PFR: + +<|ref|>text<|/ref|><|det|>[[144, 123, 844, 143]]<|/det|> +analysis of concentration nonuniformity for gas–solid systems. Ind. Eng. Chem. Res. 44, + +<|ref|>text<|/ref|><|det|>[[144, 158, 300, 177]]<|/det|> +6518–6522 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 193, 797, 212]]<|/det|> +71. Dumesic, J. A., Rudd, D. F., Aparicio, L. M., Rekoske, J. E. & Treviño, A. A. The + +<|ref|>text<|/ref|><|det|>[[144, 228, 765, 247]]<|/det|> +microkinetics of heterogeneous catalysis. (American Chemical Society, 1993). + +<|ref|>text<|/ref|><|det|>[[110, 262, 812, 281]]<|/det|> +72. Janthon, P., Viñes, F., Sirijaraensre, J., Limtrakul, J. & Illas, F. Adding pieces to the + +<|ref|>text<|/ref|><|det|>[[144, 297, 812, 317]]<|/det|> +CO/Pt(111) puzzle: the role of dispersion. J. Phys. Chem. C 121, 3970–3977 (2017). + +<|ref|>text<|/ref|><|det|>[[110, 332, 844, 352]]<|/det|> +73. Feibelman, P. J. et al. The CO/Pt(111) puzzle. J. Phys. Chem. B 105, 4018–4025 (2001). + +<|ref|>text<|/ref|><|det|>[[110, 367, 875, 422]]<|/det|> +74. Neugebohren, J. et al. Velocity-resolved kinetics of site-specific carbon monoxide oxidation on platinum surfaces. Nature 558, 280–283 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 437, 872, 457]]<|/det|> +75. Song, H. C. et al. The effect of the oxidation states of supported oxides on catalytic activity: + +<|ref|>text<|/ref|><|det|>[[144, 472, 788, 492]]<|/det|> +CO oxidation studies on Pt/cobalt oxide. Chem. Commun. 55, 9503–9506 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 507, 803, 527]]<|/det|> +76. High, E. A., Lee, E. & Reece, C. A transient flow reactor for rapid gas switching at + +<|ref|>text<|/ref|><|det|>[[144, 543, 732, 562]]<|/det|> +atmospheric pressure. Review of Scientific Instruments 94, 054101 (2023). + +<|ref|>text<|/ref|><|det|>[[110, 578, 875, 629]]<|/det|> +77. Morgan, K. et al. Forty years of temporal analysis of products. Catal. Sci. Technol. 7, 2416–2439 (2017). + +<|ref|>text<|/ref|><|det|>[[110, 644, 875, 700]]<|/det|> +78. Meunier, F. C. Pitfalls and benefits of in situ and operando diffuse reflectance FT-IR spectroscopy (DRIFTS) applied to catalytic reactions. React. Chem. Eng. 1, 134–141 (2016). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 434, 178]]<|/det|> +SiteDependentKineticsSI.pdf 23InterrogatingSites3PathwayModel.zip + +<--- Page Split ---> diff --git a/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353.mmd b/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9399e9d4c7787b0eb9db525144f98ea6c1470713 --- /dev/null +++ b/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353.mmd @@ -0,0 +1,361 @@ + +# DTPA-Receptor – A novel reporter gene system for the specific and sensitive PET imaging of CAR-T cells and AAV transduced cells + +Wolfgang Weber w.weber@tum.de + +Technical University of Munich Volker Morath Technical University of Munich https://orcid.org/0000- 0001- 7636- 0537 + +Katja Fritschle Technical University of Munich + +Linda Warmuth Technical University of Munich + +Markus Anneser Technical University of Munich + +Sarah Dötsch Technical University of Munich + +Milica Zivanic Technical University of Munich + +Luisa Krumwiede Helmholtz Munich + +Philipp BösI Technical University of Munich + +Tarik Bozoglu Medical Department I – Cardiology, Angiology, Pneumology, Klinikum rechts der Isar – Technical University of Munich, Ismaninger Strasse 22, 81675 Munich + +Stephanie Robu Technical University of Munich + +Silvana Libertini Novartis + +Susanne Kossatz Technical University of Munich + +Christian Kupatt + +<--- Page Split ---> + +Large Animal Models in Cardiovascular Research, Clinic for Cardiology, TU Munich, Germany https://orcid.org/0000- 0002- 3611- 1218 + +Markus Schwaiger Katja Steiger Institute of Pathology, School of Medicine, Technical University of Munich https://orcid.org/0000- 0002- 7269- 5433 + +Dirk Busch Technical University of Munich https://orcid.org/0000- 0001- 8713- 093X + +Arne Skerra Technical University Munich https://orcid.org/0000- 0002- 5717- 498X + +## Article + +Keywords: reporter gene imaging, ATMP, synthetic biomarker, positron emission tomography (PET), cell tracking + +Posted Date: August 23rd, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3200226/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: Yes there is potential Competing Interest. V.M., K.F., A.Sk., and W.W. have applied for intellectual property rights on Anticains- based reporter genes (WO 20022/101492 A1). A.Sk. is a co- founder and shareholder of Pieris Pharmaceuticals. + +Version of Record: A version of this preprint was published at Nature Biomedical Engineering on June 13th, 2025. See the published version at https://doi.org/10.1038/s41551- 025- 01415- 7. + +<--- Page Split ---> + +## Abstract + +Advanced Therapy Medicinal Products (ATMPs), such as cell and gene therapies, necessitate a reliable diagnostic method for quantitative monitoring. We developed a novel reporter gene system for PET imaging consisting of a membrane- anchored Anticalin protein (DTPA- R) that acts as a high- affinity receptor for the radioligand \([^{18}\mathrm{F}]\) F- DTPA- Tb. The reporter protein shows high cell surface expression of up to \(\sim 1\times 10^{6}\) receptors per cell. After systemic administration, the pharmacologically inert radioligand rapidly clears via the renal route and, at \(\mathrm{t} = 90\) min, generates a strong signal of \(22.1\%\) dD/g for DTPA- R- expressing PC3 cells compared to \(0.2\%\) dD/g for DTPA- R- negative controls (ratio: 125). The detection limit for JurkatDTPA- R cells was 500 cells in a PET phantom ex vivo and 8,000 if subcutaneously injected. In vivo expansion and migration of CD19- CAR- TDPA- R cells was successfully monitored over four weeks with a linear relationship between PET signal and CAR- T cell number. Furthermore, our reporter system allowed quantitative and longitudinal imaging of AAV9 viral vectors with a linear dose- to- signal relation. In summary, DTPA- R shows high potential for in vivo monitoring of ATMP- based therapies. + +## Introduction + +Recent regulatory approvals of cell and gene therapies \(^{1,2}\) have shown that such advanced therapy medicinal products (ATMPs) can be used in clinical practice and have the potential to cure life- threatening diseases. Despite these exciting breakthroughs, the development of new cell and virus based ATMPs remains hampered by limited knowledge about the biodistribution and persistence of these live drugs \(^{3}\) . Technologies to monitor the distribution of Chimeric Antigen Receptor (CAR)- T cells in vivo non- invasively and quantitatively could greatly improve our understanding of their trafficking, therapeutic efficacy, and off- target toxicity \(^{4}\) . Likewise, Adeno- associated virus (AAV) based gene therapies would benefit from quantifying the location, magnitude, and duration of transgene expression. + +Direct ex vivo labeling of cells (e.g., with iron oxide particles) is a sensitive approach for studying the distribution of transplanted cells early after injection. Although sensitive, this approach is limited at later time points because the label is diluted by each cell division, which prevents the assessment of cell proliferation. Furthermore, the label persists even after the cells have died or were phagocytosed. This limits the utility of direct labeling for CAR- T cell tracking, where a small fraction of the graft massively expands following activation. \(^{5}\) + +As an alternative, cell graft proliferation and viability can be imaged by an indirect labeling strategy. This approach relies either on endogenous biomarkers or reporter genes (synthetic biomarkers) that can be detected by imaging \(^{6}\) . Various luciferase enzymes and fluorescent proteins have been used as reporter genes to study transplanted cells in mice by means of optical imaging techniques \(^{7}\) . However, light is strongly attenuated and scattered by biological tissues, which limits imaging depth and quantification in optical imaging \(^{8}\) , particularly in larger animals or humans \(^{7}\) . + +<--- Page Split ---> + +A viable modality for reporter gene imaging should allow for whole body- imaging, be highly sensitive and provide a quantitative, depth- independent signal6. Furthermore, imaging should be cross- sectional to allow for precise anatomical localization. PET (positron emission tomography) fulfills all of these requirements, thus enabling the spatiotemporal imaging of transgenes on a whole- body scale while being widely available for preclinical and clinical imaging4. The sensitivity of current clinical scanners lies in the picomolar range5 and has been further increased by a factor of 40 with the recent introduction of total body PET scanners9. + +The ideal reporter gene/protein should be small, bio- orthogonal – i.e., not expressed endogenously and not affect cell function – and non- immunogenic. For PET imaging, a straightforward radiolabeling procedure with a commonly available radioisotope such as fluorine- 18 would be desirable. The small molecule probe should show low non- specific binding to cells, undergo rapid kidney clearance, but remain in the circulation sufficiently long to ensure quantitative labeling of cells expressing the reporter gene. Notably, despite numerous efforts over three decades10, no reporter gene system fulfills these requirements in a clinical setting. Herpes simplex thymidine kinase (HSV- tk)11 has been extensively studied as a reporter gene, but this viral protein is highly immunogenic with a high seroprevalence. On the other hand, systems like the sodium/iodide symporter (NIS)12, somatostatin receptor 2 (SSTR2)13, and truncated prostate- specific membrane antigen (tPSMA)14, 15 are not immunogenic but are expressed endogenously by various tissues which hampers their application for the specific whole- body imaging of grafted cells. + +To address this unmet medical need, we developed and characterized a new reporter gene and radioligand system that combines high sensitivity with exquisite specific. Our approach is based on an Anticalin16, 17 that is expressed on the cell surface and binds a bio- orthogonal 18F- labeled probe. Anticalins are engineered binding proteins based on the lipocalin protein family18, endogenous plasma proteins comprising a single polypeptide with a robust \(\beta\) - barrel fold19. We have extensively studied this new reporter system in vivo and demonstrated its suitability to monitor CAR- T cells and AAV- mediated gene transfer in mice. + +## Results + +## Design of the reporter protein and radioligand + +The PET reporter protein has been designed as an artificial cell surface protein exhibiting a highly specific binding site for a small molecule ligand (Fig. 1A). The protein construct comprises the lipocalin- 2 signal peptide, an engineered lipocalin binding protein (Anticalin18), a peptide linker containing the V5- tag (a multifunctional molecular handle for staining of ATMPs)20 and the \(\alpha\) - helical human CD4 transmembrane domain. This fusion protein contains only 257 amino acid residues and is encoded in a 774 base pairs (bp) open reading frame (ORF; for sequences see the Supplementary Information). Finally, for multimodal detection, an optional fluorescent protein was added to the C- terminal + +<--- Page Split ---> + +cytoplasmic end, such as mRuby \(^{321}\) or miRFP720 \(^{22}\) (Fig. 1B). Two Anticilins were chosen to construct different reporter proteins with two ligand specificities: (i) the Anticalin CL31d (DTPAR), which binds CHX- A"- DTPA- metal complexes with a \(K_{\mathrm{D}}\) of \(\sim 500 \mathrm{pM}^{16,23}\) and the Anticalin D6.4(Q77E) (ColchiR) which binds colchicine with a \(K_{\mathrm{D}}\) of \(\sim 20 \mathrm{pM}^{24}\) . The cognate reporter probe features CHX- A"- DTPA- metal (Fig. 1C) or colchicine (Fig. 1D), respectively, a PEG \(_{4}\) - linker to facilitate access to the Anticalin binding pocket without steric hindrance, optionally hydrophilic groups to improve pharmacokinetics and a labeling group for efficient incorporation of the \(^{18}\mathrm{F}\) isotope to enable PET detection (Fig. 1C). + +## Analysis of reporter protein variants + +To compare the two Anticalin- ligand pairs, we generated cell lines expressing the respective reporter genes by retroviral transduction of the human T cell line Jurkat, human embryonic kidney cells (HEK293T), and the prostate carcinoma cell line PC3. Initially, cell surface expression of the DTPA- R or ColchiR reporter proteins on PC3 cells was confirmed by fluorescence microscopy using an anti- V5- tag antibody (Fig. 1E). Membrane localization of mRuby3 in DTPARmRuby3 expressing HEK293T cells indicated efficient transport of the reporter protein through the secretory pathway (Fig. S1). Absolute reporter protein numbers per cell were measured by flow cytometry using an anti- V5- tag antibody and MESF calibration beads. Expression levels on Jurkat cells ranged from \(\sim 16,000\) for DTPA- R- miRFP720 to around 1 million receptors for DTPA- R (Fig. 1F). Omitting the optional fluorescent protein increased expression, e.g., DTPA- R showed 8.3- fold ( \(p < 0.0001\) ) higher expression than DTPA- R- mRuby3 (Fig. 1F). Next, we studied the influence of T cell activation on reporter gene expression by comparing PMA/lonomycin- activated to untreated Jurkat cells, which did not result in significant differences (Fig. 1F & S3A). Furthermore, the type of Anticalin influenced the expression levels: a 4.9- fold ( \(p < 0.0001\) ) higher expression of DTPA- R was measured compared to Colchi- R (Fig. 1F, S2 & S3A). We also compared the expression level of DTPA- R with a previously described reporter gene that uses the murine single chain variable fragment (scFv) C825 \(^{25}\) , or its humanized version huC825 \(^{26}\) , to bind DOTA- metal complexes (Fig. S3A). Using the same expression cassette and membrane anchor domain for these scFvs, DTPAR showed a 5.8 (muC825) or 8.9 (huC825) fold higher median V5- AF488 signal after retroviral transduction. The number of genomically inserted expression cassettes per cell was quantified by ddPCR and showed similar levels for the different constructs (Fig. S3B). However, transgene mRNA levels varied between constructs (Fig. S3C). While protein levels of Anticalin- based reporters correlated with mRNA levels ( \(R^{2} = 0.98\) ), scFv- based reporter genes produced high mRNA levels, which did not translate into elevated protein levels, indicating superior protein folding, stability, and cell- surface presentation for Anticalin- based reporter proteins (Fig. S3D). + +A potential metabolic burden of the reporter gene expression was assessed by measuring doubling times of non- activated or PMA/lonomycin- activated Jurkat cell lines using a carboxyfluorescein succinimidyl ester (CFSE) based proliferation assay (Fig. 1G). The doubling time of wild- type Jurkat cells (non- activated: \(18.3 \pm 4.3 \mathrm{h}\) / activated: \(22.2 \pm 4.7 \mathrm{h}\) ) was not substantially changed by reporter gene expression (DTPA- R: \(17.0 \pm 4.1 \mathrm{h}\) / \(20.6 \pm 4.5 \mathrm{h}\) ). + +<--- Page Split ---> + +Further applications of the V5- tag as part of the DTPA- R include the isolation of reporter gene expressing cells using magnetic- activated cell sorting. Employing the V5- tag, JurkatDTPA- R cells could be isolated by magnetic- activated cell sorting (MACS)27 from a 5:95 cell mixture with high efficacy and purity (> 98%) (Fig. 1H). Finally, the V5- tag also enabled the detection of the reporter protein at the cellular level via immunohistochemistry (IHC) (Fig. S4). + +## Synthesis & characterization of [18F]-colchicine and [18F]-DTPA + +First, we investigated the binding of \(\mathsf{NH}_2\) - CHX- A"- DTPA in complex with \(^{90}\mathrm{Ylll}\) to JurkatDTPA- R cells using a competitive assay, resulting in a half- maximal inhibitory concentration (IC50) of 1.4 nM (Fig. 2A). However, fluorine- 18 is the preferred PET radioisotope as it is widely available and almost every decay leads to the emission of a positron (ratio 96.9%) with desirably low energy (Emean=250 keV, average positron range of 0.6 mm28). On the other hand, positron- emitting radiometals have inferior physical characteristics, and their \(\mathsf{NH}_2\) - CHX- A"- DTPA complexes often bind with lower affinity to DTPA- R than \(\mathsf{Ylll}\) or \(\mathsf{Tb}^{III}\) complexes16, 23. For this reason, we designed a DTPA- based radioligand that can be charged with radioactive or non- radioactive metal ions and features a second group that can be easily radiolabeled with fluorine- 18 (Fig. 2B & S6). This radiolabeling approach uses a trimethylamine leaving group29, similar to the approved radioligand [18F]PSMA- 100730. For comparison, we synthesized analogous radioligand precursors for Colchi- R employing different numbers of strongly polar Glu residues and a precursor with two Glu residues for DTPA- R (Fig. S5). After quality control (Fig. S6E), both compounds were radio- fluorinated in a single step and purified by reversed phase HPLC to obtain highly pure radioligands (Fig. 2C & S7A). For [18F]-F- Nic- D- Glu2- PEG4- CHX- A"- DTPA- \(\mathsf{Tb}^{III}\) (dubbed [18F]-F- DTPA) a radiochemical yield of \(\sim 20\%\) and a radiochemical purity of \(>98\%\) were achieved. Subsequently, the 18F- labeled radioligands were tested in ligand binding assays with DTPA- R and Colchi- R expressing PC3 cells (Fig. 2DE & S7B). The binding of both radioligands was highly specific to cells expressing the corresponding receptor ( \(\sim 1,000\) - fold for DTPA- R and \(\sim 500\) - fold for Colchi- R) and could be efficiently blocked by competition with non- radioactive ligand ( \(\sim 100\) - fold for DTPA- R and \(\sim 400\) - fold for Colchi- R). + +## Dynamic PET imaging of mice bearing xenograft tumors + +As a proof- of- principle for reporter gene imaging, the respective radioligands were injected into CD1- nude mice carrying subcutaneous PC3DTPA- R and PC3Colchi- R xenograft tumors above the right and left shoulder, respectively (Fig. 3AB & S7C). Dynamic [18F]- F- DTPA PET imaging showed fast, almost exclusively renal clearance, accompanied by increasing signals in the kidneys, ureter, and urinary bladder, with no retention in other tissues (Fig. 3A). The radioligand showed rapid accumulation in the PC3DTPA- R xenograft tumor, resulting in a stable activity concentration of 22.8±0.6% Injected Dose (ID)max/g between 30 and 90 min post injection (p.i.). While both xenograft tumors were clearly visible on MR images (Fig. 3B), only the PC3DTPAR tumor could be delineated in PET. The maximum activity + +<--- Page Split ---> + +concentration in the \(\mathrm{PC3^{Colchir}R}\) xenograft \((0.18\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g})\) was 125- fold lower than for the cognate \(\mathrm{PC3^{DTPA - R}}\) xenograft. \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in the blood pool was quickly cleared with a plasma half- life of 0.5- 1.5 min, which resulted in a tumor- to- blood ratio of 261 after 90 min. Furthermore, in a second static PET scan 6 h p.i., still \(48\%\) of the initial \(\mathrm{PC3^{DTPA - R}}\) signal was retained (Fig. 3CD). Overall, these results indicate an exquisite specificity of the DTPA- R reporter gene system in vivo. + +The biodistribution of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA at 90 min p.i. in healthy female and male mice was analyzed ex vivo to study sex- specific differences \((n = 5\% +5\%)\) . Very low retention of below \(0.2\% \mathrm{ID} / \mathrm{g}\) was measured for most tissues (Fig. 3E), except for the kidneys \((0.9\% \mathrm{ID} / \mathrm{g})\) and the hepato- biliary excretion route (small intestine: \(0.2\% \mathrm{ID} / \mathrm{g}\) ; large intestine: \(0.1\% \mathrm{ID} / \mathrm{g}\) ). No significant sex- specific differences in biodistribution were observed. (Fig. 3E). + +In vivo stability of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA was analyzed after intravenous (i.v.) injection using size- exclusion chromatography (SEC) to separate intact radioligand from fragments with lower molecular weight. \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA within urine samples collected from two mice \(\sim 3\) h p.i. was \(>97\%\) intact, indicating serum stability (Fig. 3F). + +For comparison, a dynamic PET study was conducted for the radioligand \([^{18}\mathrm{F}]\mathrm{F}\) - Nic- Glu \(_2\) - PEG \(_4\) - colchicine ([18F]F- colchicine). This radioligand showed not only renal (19.8%ID) but also significant hepato- biliary excretion (69.7%ID), which is undesirable due to the strong and variable background PET signals in the gastrointestinal tract (Fig. S7D). The elevated hepato- biliary excretion of \([^{18}\mathrm{F}]\mathrm{F}\) - colchicine is likely explained by its lower hydrophilicity, as reflected by the \(\mathrm{LogD}_{7,4}\) of - 2.97 vs. - 3.58 for \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA. Taken together, these pharmacokinetic data, as well as the higher expression levels of the DTPA- R on transfected cells and the much higher radioligand uptake in the xenograft tumor, prompted us to focus on the further development of the DTPA- R reporter system. + +## Design and evaluation of DTPA-R labeled CAR-T cells + +To investigate the application of DTPA- R in the context of human CART cell therapy, we assessed in an animal study if the migration, proliferation, and tissue infiltration of \(\mathrm{CART^{DTPA - R}}\) cells can be imaged. To this end, we employed a well- characterized CD19- targeting 2nd- generation CAR based on the scFv FMC63 (Fig. S8A) \(^{31}\) . The corresponding expression cassette comprised the CAR followed by a 2A self- cleaving sequence separating a second membrane protein, the truncated Epidermal Growth Factor Receptor (EGFRt) \(^{32}\) . EGFRt lacks normal EGFR function but is still recognized by cetuximab (Erbitux) for in vivo cell ablation \(^{33}\) . To generate \(\mathrm{CART^{DTPAR}}\) cells, we replaced the EGFRt in the retroviral expression cassette with DTPA- R, which enables PET imaging, apart from optionally serving for cell ablation (Fig. S8B). Both CAR- T vectors were used to transduce human peripheral blood mononuclear cells (PBMCs), resulting in a surface expression of approximately 100,000 DTPA- R molecules per CAR- T cell (Fig. S8C). + +<--- Page Split ---> + +There was no substantial difference in CD3, CD4, and CXCR3 expression levels or \(\mathsf{CD8^{+} / CD4^{+}}\) ratios between CAR- TEGFRt and CAR- TDPTA- R cells. (Fig. S8CD). CAR- T cell function was assessed in a real- time assay of the CAR- mediated killing of HEKCD19 target cells (xCELLigence; Fig. S8E) as well as a 51Cr- release assay on Raji cells (Fig. S8F). Specific lysis was comparable between the aCD19CARTEGFRt and aCD19CARTDPTAR cell treated groups in both assays. + +## PET imaging of CAR-T cell therapy + +Both CAR- T cells were used to treat NSG mice engrafted with a systemic Raji tumor, which is known to primarily home to the bone marrow34. After 7 days of Raji engraftment, \(2\times 10^{6}\) aCD19- CARTDPTAR or reference aCD19- CARTEGFRt cells were infused. This experimental setting ensured constant CART expansion due to non- complete tumor eradication, which was imaged with [18F]FDPTA on days 4, 8, 15, 22, and 30 (Fig. 4A). PET signals indicated the presence of CARTDPTAR cells at day 4 in the spleen and from day 8 onwards, with increasing intensity, in the bone marrow of the extremities as well as spine, scull, and axillary lymph nodes (Fig. 4B or Fig. S9 for all animals). In contrast, PET signals were absent in mice injected with aCD19- CARTEGFRt cells, demonstrating the specificity of [18F]F- DTPA (Fig. 4C). + +Quantification of the cumulated specific PET signals allowed the assessment of CAR- T cell expansion over time (Fig. 4D). Furthermore, we used flow cytometry at the end of the study, or when the humane endpoint was reached, to quantify the numbers of CAR- T (Fig. 4E) and Raji cells (Fig. 4F). This ex vivo analysis showed differences in CAR- T cell numbers, confirming the high variability between animals and the need for imaging of these therapies. In addition, 3D cross- sectional images of the [18F]F- DTPA PET/MR allowed the monitoring of local CAR- T infiltration (Fig. 4G- I). Longitudinal PET analysis revealed different trajectories of CAR- T infiltrated lesions (Fig. 4HI). Ex vivo analysis by V5- IHC of sagittal spine sections allowed CAR- TDPTA- R identification on the cellular level, which confirmed CAR- T infiltration into individual vertebra bodies ( \(\sim 1.5\times 0.5 \text{mm}\) ) and, occasionally, invasion through the vertebral bone into the spinal canal (Fig. 4J). Importantly, the pattern of CAR- T infiltration determined by IHC was very well matched by the signals in PET/MR (Fig. 4K). Furthermore, the co- localization of CAR- TDPTAR with Raji cells was confirmed by IHC co- staining of CD19 and the V5- tag (Fig. 4L). Interestingly, lesions with a CAR- T cell plateau in longitudinal PET imaging (blue and orange, Fig. 4I) showed lower tumor cell density, which indicated better tumor clearance by CART cells compared to neoplasms with constantly increasing PET signal (green, Fig. 4I & S10A). Other organs like the spleen, kidneys, liver, and lung only showed few CAR- TDPTA- R cells in V5- IHC (Fig. S10B). + +Correlation between the actual number of CAR- TDPTA- R cells detected by flow cytometry after extraction from a hollow bone and the cumulated PET signal obtained for the same extremity (Fig. 5A & S11AB) was performed for mice sacrificed on day 8 or 15 (Fig. 5B, gating see S11C) and on day 16 or 31 (Fig. S11D). A linear relationship ( \(R^2 = 0.92\) ) indicated that PET imaging of DTPA- R can quantitatively monitor tissue- specific CAR- T cell infiltration. Utilizing the resulting regression line, we calculated a detection limit of around 1,200 CARTDPTAR cells in delineated lesions in the bone (Fig. 5C). To further characterize + +<--- Page Split ---> + +the relationship between PET signal and the number of DTPA- R expressing cells, we measured the signal of defined numbers of JurkatDTPAR cells in a PET phantom in vitro. For cells labeled with [18F]F- DTPA, we observed a linear correlation (R2 = 0.999) with a detection limit of 500 cells (Fig. 5D). In the next step, we determined the detection limit in tissue with low perfusion. Therefore, we injected 4,000 to 64,000 JurkatDTPA- R cells, or CARTDTPAR cells, in the dorsal subcutis, immediately followed by i.v. injection of [18F]F- DTPA (Fig. 5EF). Here, 8,000 cells were clearly detectable by PET. Again, a linear correlation between cell number and PET signal was observed (R2 = 0.96 and R2 = 0.99, respectively; see Fig. 5EF). + +## PET imaging of gene transfer mediated by AAV9 viral vectors + +In vivo gene therapy represents another relevant application scenario for reporter gene imaging, where PET imaging can help assess the roles of administration routes, dosing regimens, pharmacokinetics of capsid- modified vectors, or endurance of gene expression. As a common example for gene therapy, we have selected vectors based on the Adeno- associated virus (AAV)35. In particular, we have focused on the serotype AAV9, which sparked great interest for transducing muscle and neuronal tissue36 and provided the basis for the FDA- approved drug Zolgensma35. We constructed a transfer plasmid flanked by AAV2 inverted terminal repeats (ITR) and used it to produce AAV2/9 viral vectors encoding the DTPA- R reporter gene under the control of the CMV promoter (Fig. 6A). First, we investigated the reporter gene expression in mice after systemic injection of AAV9DTPA - R. Immunocompetent C57BL/6 (Fig. 6B) and immunocompromised CD1- nude (Fig. 6C) mice (n = 3 per group) were dosed with \(2.5 \times 10^{12}\) viral genomes (vg) per mouse and, on days 14, 21 and 28 [18F]F- DTPA PET scans were recorded (Fig. 6BC). Although cohorts were treated in parallel using the same virus batch, there were marked differences between the transduction patterns obtained in C57BL/6 and CD1- nude mice. Isocontour segmentation of the heart yielded a value of \(15.3 \pm 3.8\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) for C57BL/6 and \(28.7 \pm 5.4\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) for CD1- nude mice. Within each cohort, animals showed comparable transduction patterns (Fig. S12). Moreover, we observed consistent trajectories of the repeated PET measurements, indicating stability of the vector expression over time (Fig. 6D). + +We then i.v. injected four different titers, ranging from \(1 \times 10^{11}\) to \(2.5 \times 10^{12}\) vg/mouse, into C57BL/6 mice (n = 3 per group). After 7 days, [18F]F- DTPA PET/MR scans were recorded for all cohorts and an untreated control cohort (Fig. 6E or Fig. S13 for all animals). The transduction patterns were similar to the previous experiment; for example, mice with the highest applied titer showed clear PET signals in dorsal brown adipose tissue (BAT; \(\% \mathrm{ID}_{\mathrm{max}}: 7.7 \pm 1.5\) ), heart muscle ( \(\% \mathrm{ID}_{\mathrm{max}}: 6.1 \pm 3.6\) ), liver ( \(\% \mathrm{ID}_{\mathrm{max}}: 11.3 \pm 1.4\) ) and adrenals ( \(\% \mathrm{ID}_{\mathrm{max}}: 29.5 \pm 7.0\) ), whereas the background uptake in the upper body of the control mice was \(\sim 0.5 \pm 0.06\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) . Of note, the high resolution in axial planes provided by the combination of PET with an isomorph MRT allowed the precise analysis of transduction events (Fig. 6F) far beyond previous results obtained by bioluminescence imaging36. + +<--- Page Split ---> + +Lower titers resulted in gradually reduced PET signals within the respective tissues. Importantly, PET signals ( \(\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) ) showed a linear correlation with the administered dose in BAT ( \(R^2 = 0.73\) ), heart muscle ( \(R^2 = 0.60\) ), liver ( \(R^2 = 0.92\) ), and adrenals ( \(R^2 = 0.89\) ), which underlines the quantitative character of DTPA- R PET imaging (Fig. 6G). Well- perfused organs showed a steeper increase of the PET signal with increasing AAV titer. This observation could either be due to a higher transduction rate in these organs or a better local delivery of [18F]F- DTPA. To test this hypothesis, we compared PET data for different tissues with perfusion values from the literature \(^{37,38}\) . Indeed, a linear correlation between the perfusion index and the slope of PET signal increase was seen (Fig. 6H). + +Next, we stained tissues by V5- IHC and observed transduction of individual cells, for example in the cortex of adrenal glands (Fig. 6l). Quantification using automated positive cell detection showed \(2\%\) positive cells for animals transduced with the lowest, \(33\%\) for the low, and \(72.2\%\) for the high dose, which yielded a perfectly linear relation with the injected vector titer ( \(R^2 = 1\) ; Fig. 6K). Also, the relationship between the measured PET signal and the injected AAV titer (Fig. 6J), or the transduction levels from histology (Fig. 6L), indicated a linear correlation ( \(R^2 = 0.98\) or 0.96, respectively). Transduction levels were further quantified by ddPCR, revealing a linear correlation between the number of viral genomes and the PET signals in the liver ( \(R^2 = 0.97\) ; Fig. 6M). A multivariate analysis correlating the measured PET signals with injected AAV9 doses, mRNA levels, and viral genomes in the liver revealed a linear relationship ( \(R^2 = 1\) , Fig. 6N). + +## Discussion + +We have developed a novel reporter gene system based on a membrane- anchored Anticalin that specifically binds a small- molecule radioligand, enabling quantitative and longitudinal PET imaging of ATMPs with remarkable specificity and sensitivity. Based on PET with [18F]F- DTPA, we have demonstrated the suitability of DTPAR in relevant use cases such as CAR- T cell therapy of CD19 lymphoma \(^{39}\) and AAV9 gene therapy \(^{35}\) . CAR- T cell movement and tumor homing in a mouse model of systemic lymphoma could be visualized and quantitatively tracked in vivo over 30 days. Furthermore, the transduction of AAV9 vectors in distinct tissues could be quantitatively analyzed by molecular imaging. This theranostic approach, combining novel cell and gene therapies with a quantitative imaging modality using a universal reporter system that potentially can bridge preclinical development and clinical evaluation, should facilitate the clinical translation of ATMPs \(^{40}\) . To this end, our PET reporter system offers promising functional features, which are described below. + +High expression level of the reporter protein determines strength of the signal, which has been demonstrated for Jurkat \(^{\mathrm{DTPA - R}}\) cells displaying \(\sim 1 \times 10^6\) receptors per cell without measurable negative effects on cellular fitness or T cell function. This number is approximately 10- fold higher compared to well- known lymphocyte receptors, such as the B cell receptor ( \(1.2 \times 10^5\) copies \(^{41}\) ) or CD4 ( \(\sim 1 \times 10^5\) copies \(^{42}\) ). Direct comparison of the Anticalin CL31d specific for [18F]FDTPA with the scFv huC825, which + +<--- Page Split ---> + +binds DOTA- metal, resulted in an 8.9- fold higher expression of DTPA- R (Fig. S2). Low surface densities for huC825- based reporter proteins are also reflected by transfected HEK293T cells expressing only \(\sim\) \(1.5\times 10^{4}\) receptors per cell \(^{26}\) . In contrast to scFv antibody fragments, known for their oligomerization and aggregation tendencies \(^{43}\) , Anticalins possess a robust fold, are composed of a single polypeptide chain, and can be expressed as recombinant proteins at high levels \(^{18}\) . + +Minimal gene size is of importance due to the limited packaging capacities of viral and non- viral gene shuttles and met by DTPA- R, which is encoded by just 774 bp ( \(\sim 26.4\) kDa for the mature fusion protein), in contrast to most other PET reporter proteins such as HSV1- tk (1,131 bp) \(^{11}\) , NIS (1,932 bp) \(^{12,44}\) , tPSMA \(^{(\mathrm{N9del})}\) (2,226 bp) \(^{15}\) , SSTR2 (1,110 bp) \(^{13}\) , DAbR1- 2A- GFP (2,280 bp) \(^{25}\) , SNAPtag (846 bp) \(^{45}\) , eDHFR (480 bp) \(^{46}\) or the EGFRt (1,074 bp) \(^{32}\) . + +Functional inertness, including lack of biological activity, interference, and toxicity of both the reporter gene and probe, are crucial factors as the capability of in vivo imaging is second to the therapeutic function of an ATMP. We have investigated the proliferation kinetics, activation status, receptor expression, and cellular toxicity of CAR- T cells and found no difference due to the DTPA- R expression compared to EGFRt. Given the binding activity of DTPA- R for an exogenous hapten, interferences with cellular processes are much less likely than for reporter proteins that lead to ion transport over the cell membrane (NIS), represent tumor- associated surface antigens (tPSMA, SSTR2) or possess catalytic activities (HSV- tk, PSMA, SNAPtag, eDHFR). + +Stability of the PET signal over time is important to allow reproducible imaging, which mainly relies on the receptor- ligand affinity. While the soluble recombinant CL31d protein shows a \(K_{\mathrm{D}}\) value of 543 pM for the complex with CHX- A"DTPA- Y \(^{17}\) , the half- maximal inhibitory concentration (IC \(_{50}\) ) of \(\mathrm{NH}_{2}\) - CHX- A"- DTPA- \(^{90}\mathrm{Y}\) for DTPA- R expressed on the cell surface was slightly higher, with 1.4 nM. Nevertheless, the high stability of the [ \(^{18}\mathrm{F}\) ]F - DTPA signal within the tumor was demonstrated in dynamic PET scans. This is not the case for the NIS reporter gene, for example, due to the naturally occurring efflux of the radioactive iodide \(^{44}\) . Potential internalization of DTPA- R is very low due to the absence of the cytosolic CD4 domain, which triggers internalization of CD \(^{47}\) . In addition, DTPA- R expression levels are independent of T cell activation, which is commonly known to heavily influence expression profiles in T cells \(^{48}\) . + +Linear relation between PET signal and the number of DTPA- R- labelled cells is mandatory to move beyond qualitative imaging. The DTPA- R system allows a strong correlation of the number of CART cells in the bone marrow or the applied AAV9 virus titer with the corresponding PET signal, thus enabling non- invasive quantification. The amount of bound [ \(^{18}\mathrm{F}\) ]F- DTPA ligand is only governed by the local concentration of DTPA- R and the Law of Mass Action, contrasting to the much more complex and unpredictable relationships for transporter- or enzyme- type reporter proteins. + +<--- Page Split ---> + +Criteria for a corresponding radioligand \(^{49}\) include its chemical composition as well as the choice of the radioisotope for PET imaging. The choice of \(^{18}\mathrm{F}\) , with its ideal physical half- life, high positron yield, and low positron energy, allows detection with high sensitivity and resolution, especially if compared to isotopes such as \(^{111}\mathrm{In}\) or \(^{86}\mathrm{Y}\) , used for DAbR1 / C825 reporter probes \(^{25,26}\) . Other advantages of \(^{18}\mathrm{F}\) are the relatively short half- life of 109 min, which enables repeated serial imaging in a daily interval, while the half- live is long enough to allow multiple PET scans from a single batch of radioligand. Furthermore, the low radiation exposure does not hamper the function of the ATMP nor lead to a significant absorbed biological dose. The stability of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in blood was demonstrated, together with a stable PC3 \(^{DTPA^{- }}\) \(R\) signal in the time- activity curve. Nevertheless, achieving high specific activity remains a challenge for \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA synthesis, as well as for other \(^{18}\mathrm{F}\) - based radioligands, and constitutes a current area of improvement. + +The absence of endogenous binding activity was impressively demonstrated by the very low background accumulation of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in various organs. This is in marked contrast to all fully human reporter proteins described so far, including NIS with its thyroidal, gastric, mucosal, salivary, and lactating mammary gland expression \(^{50}\) . Furthermore, the biodistribution analysis of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in female and male mice showed no sex- specific differences. Excretion of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA was mainly seen via the renal route, which led to defined signals in the kidneys and bladder. In contrast, the radioligand for the C825- scFv- baed reporter gene system caused a diffuse signal in the abdomen at \(t = 30\) min p.i., and only \(\sim 80\%\) of the \([^{86}\mathrm{Y}]\mathrm{Y}\) - DOTA- Bn was cleared via the renal route \(^{26}\) . + +Furthermore, the ease of probe preparation and the price per imaging day are of importance both in biomedical research and in a clinical setting. In this regard, the DTPA- R system is attractive as the chemical synthesis of the ligand precursor is straight- forward. Furthermore, radiolabeling involves only few steps and the supply of \([^{18}\mathrm{F}]\mathrm{F}^{- }\) is neither limiting nor expensive. + +Finally, the sensitivity of signal detection is a crucial aspect, which depends on the number of binding sites, radioligand affinity, physical decay characteristics of the isotope and the specific activity of the radioligand. Using a PET phantom, we determined a detection limit of 500 cells while a standard \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA imaging protocol allowed the clear detection of as few as \(8 \times 10^{3}\) Jurkat \(^{DTPA - R}\) or αCD19- CAR- \(^{TDPA - R}\) cells in vivo. Reported sensitivity for DTPA- R is in line or even higher when compared to other reporter systems \(^{50}\) . In a comparative study with primary T cells transfected with human reporter genes \(^{51}\) , only the norepinephrine transporter (NET) with \([^{18}\mathrm{F}]\mathrm{F}\) - MFBG was able to detect \(3 - 4 \times 10^{4}\) cells, while for HSV- TK/ \([^{18}\mathrm{F}]\mathrm{F}\) - FEAU and NIS/ \([^{124}\mathrm{I}]\) - iodide a sensitivity of \(\sim 3 \times 10^{5}\) and \(\sim 1 \times 10^{6}\) was reported, respectively. In direct comparison with those reporter systems DTPA- R/ \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA not only achieved a higher sensitivity, but also shows the superior excretion profile which leads to lower background signals \(^{51}\) . + +<--- Page Split ---> + +In summary, the DTPA- R & [18F]F- DTPA system meets all relevant design and functional requirements for a universal reporter gene system, which may boost future ATMPs by making them PET- traceable. + +## Declarations + +Acknowledgments + +The authors wish to thank Markus Mittelhauser, Hannes Rolbieski, Sybille Reder, and Natalie Röder for PET/MR acquisition. Prof. Dr. Franz Schilling and Dr. Geoffrey Topping for help with MR sequence optimization, Olga Seelbach and Marion Mielke for histology, Anja Wolf for production of AAV9 vectors, Dr. Ritu Mishra for FACS sorting, Dr. Andreas Eichinger for help with the generation of the structural model, Prof. Bernhard Küster and Prof. Gil Westmeyer for cell line (all from TU Munich) and Lara Bontadelli (Novartis) for qPCR and ddPCR analysis. + +This study was funded by the National Centre for the 3Rs via CRACK IT Challenge 32 (NC/C01905/1 & NC/C019202/1) and SFB824 of the German research foundation (projects A8, Z1 & Z2). Anticalin® is a registered trademark of Pieris Pharmaceuticals GmbH. + +Author Contributions + +GlaxoSmithKline and Novartis supported the project with in- kind contributions in the frame of the CRACK IT program of the NC3Rs. + +Competing interests + +V.M., K.F., A.Sk., and W.W. have applied for intellectual property rights on Anticalins- based reporter genes (WO 20022/101492 A1). A.Sk. is a co- founder and shareholder of Pieris Pharmaceuticals. + +## References + +1. Kingwell, K. CAR T therapies drive into new terrain. Nat Rev Drug Discov 16, 301-304 (2017). +2. Mendell, J.R. et al. Current Clinical Applications of In Vivo Gene Therapy with AAVs. Mol Ther 29, 464-488 (2021). +3. Minn, I., Rowe, S.P. & Pomper, M.G. Enhancing CAR T-cell therapy through cellular imaging and radiotherapy. Lancet Oncol 20, e443-e451 (2019). +4. Krebs, S., Dacek, M.M., Carter, L.M., Scheinberg, D.A. & Larson, S.M. CAR Chase: Where Do Engineered Cells Go in Humans? Front Oncol 10, 577773 (2020). +5. Iafrate, M. & Fruhwirth, G.O. How Non-invasive in vivo Cell Tracking Supports the Development and Translation of Cancer Immunotherapies. Front Physiol 11, 154 (2020). +6. Ashmore-Harris, C., Iafrate, M., Saleem, A. & Fruhwirth, G.O. Non-invasive Reporter Gene Imaging of Cell Therapies, including T Cells and Stem Cells. Mol Ther 28, 1392-1416 (2020). + +<--- Page Split ---> + +7. Serganova, I. & Blasberg, R.G. Molecular Imaging with Reporter Genes: Has Its Promise Been Delivered? J Nucl Med 60, 1665-1681 (2019). +8. Volpe, A., Kurtys, E. & Fruhwirth, G.O. Cousins at work: How combining medical with optical imaging enhances in vivo cell tracking. Int J Biochem Cell Biol 102, 40-50 (2018). +9. Cherry, S.R. et al. Total-Body PET: Maximizing Sensitivity to Create New Opportunities for Clinical Research and Patient Care. J Nucl Med 59, 3-12 (2018). +10. Tjuvajev, J.G. et al. Imaging the expression of transfected genes in vivo. Cancer Res 55, 6126-6132 (1995). +11. Yaghoubi, S.S. et al. Noninvasive detection of therapeutic cytolytic T cells with \(^{18}\mathrm{F}\)-FHBG PET in a patient with glioma. Nat Clin Pract Oncol 6, 53-58 (2009). +12. Merron, A. et al. SPECT/CT imaging of oncolytic adenovirus propagation in tumours in vivo using the Na/I symporter as a reporter gene. Gene Ther 14, 1731-1738 (2007). +13. Vedvyas, Y. et al. Longitudinal PET imaging demonstrates biphasic CAR T cell responses in survivors. JCI Insight 1, e90064 (2016). +14. Castanares, M.A. et al. Evaluation of prostate-specific membrane antigen as an imaging reporter. J Nucl Med 55, 805-811 (2014). +15. Minn, I. et al. Imaging CAR T cell therapy with PSMA-targeted positron emission tomography. Sci Adv 5, eaaw5096 (2019). +16. Kim, H.J., Eichinger, A. & Skerra, A. High-affinity recognition of lanthanide(III) chelate complexes by a reprogrammed human lipocalin 2. J Am Chem Soc 131, 3565-3576 (2009). +17. Eggenstein, E., Eichinger, A., Kim, H.J. & Skerra, A. Structure-guided engineering of Anticilins with improved binding behavior and biochemical characteristics for application in radio-immuno imaging and/or therapy. J Struct Biol (2013). +18. Deuschle, F.C., Ilyukhina, E. & Skerra, A. Anticalin(R) proteins: from bench to bedside. Expert Opin Biol Ther 21, 509-518 (2021). +19. Schiefner, A. & Skerra, A. The menagerie of human lipocalins: a natural protein scaffold for molecular recognition of physiological compounds. Acc Chem Res 48, 976-985 (2015). +20. Dunn, C., O'Dowd, A. & Randall, R.E. Fine mapping of the binding sites of monoclonal antibodies raised against the Pk tag. J Immunol Methods 224, 141-150 (1999). +21. Bajar, B.T. et al. Improving brightness and photostability of green and red fluorescent proteins for live cell imaging and FRET reporting. Sci Rep 6, 20889 (2016). +22. Shcherbakova, D.M., Cox Cammer, N., Huisman, T.M., Verkhusha, V.V. & Hodgson, L. Direct multiplex imaging and optogenetics of Rho GTPases enabled by near-infrared FRET. Nat Chem Biol 14, 591-600 (2018). +23. Eggenstein, E., Eichinger, A., Kim, H.J. & Skerra, A. Structure-guided engineering of Anticilins with improved binding behavior and biochemical characteristics for application in radio-immuno imaging and/or therapy. J Struct Biol 185, 203-214 (2013). + +<--- Page Split ---> + +24. Barkovskiy, M., Ilyukhina, E., Dauner, M., Eichinger, A. & Skerra, A. An engineered lipocalin that tightly complexes the plant poison colchicine for use as antidote and in bioanalytical applications. Biol Chem 400, 351–366 (2019). + +25. Krebs, S. et al. Antibody with infinite affinity for in vivo tracking of genetically engineered lymphocytes. J Nucl Med 59, 1894–1900 (2018). + +26. Dacek, M.M. et al. Engineered Cells as a Test Platform for Radiohaptens in Pretargeted Imaging and Radioimmunotherapy Applications. Bioconjug Chem 32, 649–654 (2021). + +27. Sutermaster, B.A. & Darling, E.M. Considerations for high-yield, high-throughput cell enrichment: fluorescence versus magnetic sorting. Sci Rep 9, 227 (2019). + +28. Sanchez-Crespo, A. Comparison of Gallium-68 and Fluorine-18 imaging characteristics in positron emission tomography. Appl Radiat Isot 76, 55–62 (2013). + +29. Richarz, R. et al. Neither azeotropic drying, nor base nor other additives: a minimalist approach to \(^{18}\mathrm{F}\) labeling. Org Biomol Chem 12, 8094–8099 (2014). + +30. Cardinale, J. et al. Development of PSMA-1007-Related Series of \(^{18}\mathrm{F}\) -Labeled Glu-Ureido-Type PSMA Inhibitors. J Med Chem 63, 10897–10907 (2020). + +31. Sommermeyer, D. et al. Fully human CD19-specific chimeric antigen receptors for T-cell therapy. Leukemia 31, 2191–2199 (2017). + +32. Wang, X. et al. A transgene-encoded cell surface polypeptide for selection, in vivo tracking, and ablation of engineered cells. Blood 118, 1255–1263 (2011). + +33. Paszkiewicz, P.J. et al. Targeted antibody-mediated depletion of murine CD19 CAR T cells permanently reverses B cell aplasia. J Clin Invest 126, 4262–4272 (2016). + +34. Lee, J.C. et al. In vivo inhibition of human CD19-targeted effector T cells by natural T regulatory cells in a xenotransplant murine model of B cell malignancy. Cancer Res 71, 2871–2881 (2011). + +35. Wang, D., Tai, P.W.L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat Rev Drug Discov 18, 358–378 (2019). + +36. Zincarelli, C., Soltys, S., Rengo, G. & Rabinowitz, J.E. Analysis of AAV serotypes 1–9 mediated gene expression and tropism in mice after systemic injection. Mol Ther 16, 1073–1080 (2008). + +37. Brown, L.V., Gaffney, E.A., Ager, A., Wagg, J. & Coles, M.C. Quantifying the limits of CAR T-cell delivery in mice and men. J R Soc Interface 18, 20201013 (2021). + +38. Gjedde, S.B. & Gjedde, A. Organ Blood-Flow Rates and Cardiac-Output of the Balb-C Mouse. Comp Biochem Phys A 67, 671–674 (1980). + +39. Amini, L. et al. Preparing for CAR T cell therapy: patient selection, bridging therapies and lymphodepletion. Nat Rev Clin Oncol (2022). + +40. Weber, W.A., Varasteh, Z., Fritschle, K. & Morath, V. A Theranostic Approach for CAR-T Cell Therapy. Clin Cancer Res 28, 5241–5243 (2022). + +41. Yang, J. & Reth, M. The dissociation activation model of B cell antigen receptor triggering. FEBS Lett 584, 4872–4877 (2010). + +<--- Page Split ---> + +42. Wang, M. et al. Quantifying CD4 receptor protein in two human CD4 + lymphocyte preparations for quantitative flow cytometry. Clin Proteomics 11, 43 (2014). +43. Lee, C.C., Perchiacca, J.M. & Tessier, P.M. Toward aggregation-resistant antibodies by design. Trends Biotechnol 31, 612–620 (2013). +44. Ravera, S., Reyna-Neyra, A., Ferrandino, G., Amzel, L.M. & Carrasco, N. The Sodium/Iodide Symporter (NIS): Molecular Physiology and Preclinical and Clinical Applications. Annu Rev Physiol 79, 261–289 (2017). +45. Stotz, S., Bowden, G.D., Cotton, J.M., Pichler, B.J. & Maurer, A. Covalent \(^{18}\mathrm{F}\) -Radiotracers for SNAPtag: A New Toolbox for Reporter Gene Imaging. Pharmaceuticals (Basel) 14 (2021). +46. Sellmyer, M.A. et al. Imaging CAR T Cell Trafficking with eDHFR as a PET Reporter Gene. Mol Ther 28, 42–51 (2020). +47. Bojkowska, K. et al. Measuring in vivo protein half-life. Chem Biol 18, 805–815 (2011). +48. van der Windt, G.J. & Pearce, E.L. Metabolic switching and fuel choice during T-cell differentiation and memory development. Immunol Rev 249, 27–42 (2012). +49. Volpe, A., Pillarsetty, N.V.K., Lewis, J.S. & Ponomarev, V. Applications of nuclear-based imaging in gene and cell therapy: probe considerations. Mol Ther Oncolytics 20, 447–458 (2021). +50. Shao, F. et al. Radionuclide-based molecular imaging allows CAR-T cellular visualization and therapeutic monitoring. Theranostics 11, 6800–6817 (2021). +51. Moroz, M.A. et al. Comparative Analysis of T Cell Imaging with Human Nuclear Reporter Genes. J Nucl Med 56, 1055–1060 (2015). + +## Figures + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +## Design and characterization of DTPA-R reporter protein + +Schematic representation of the DTPA- R(ceptor) reporter gene system composed of the reporter protein DTPA- R and the cognate PET reporter probe [18F]F- DTPA- metal. (A) Molecular model (PyMol) based on the crystal structure of the Anticalin/CHX- A"- DTPA- Y complex (PDB ID: 4IAX) and the NMR + +<--- Page Split ---> + +structure of the CD3 transmembrane domain (PDB ID: 2KLU). (B) Design of the coding region for the Anticalin- based reporter protein with a promoter, the Lcn2 signal peptide (sp), the mature Anticalin, the V5- tag, the CD4 transmembrane domain and, optionally, a fluorescent protein. (C) Chemical structure of the PET radioligand \([^{18}\mathrm{F}]\mathrm{F}\) - Nic- Glu2- PEG4- CHX- A"- DTPA- metal, dubbed \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA and (D) the ligand moiety of \([^{18}\mathrm{F}]\mathrm{F}\) - colchicine. (E) Fluorescence microscopy of PC3DTPA- R and PC3Colchi- R cells stained with Hoechst 33342 (cell nucleus) and an AlexaFluor488- conjugated anti- V5- tag antibody (reporter protein). (F) Quantification of reporter protein surface densities by flow cytometry with MESF beads (four individual experiments, t- test). (G) Analysis of transduced Jurkat cell lines for their proliferation kinetics using the CFSE assay, error bars indicate distribution in one experiment. (H) Separation of transgenic JurkatDTPA- R cells from not transduced Jurkat cells using the anti- V5- tag antibody for magnetic- activated cell sorting (MACS). + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +![](images/Figure_3.jpg) + + +## Preparation and characterization of the \([^{18}\mathrm{F}]\) F-DTPA radioligand + +(A) Binding of \(\mathrm{NH_2 - CHX - A^{\prime\prime} - DTPA^{\prime\prime}90Y}\) competed by \(\mathrm{NH_2 - CHX - A^{\prime\prime} - DTPA - 89Y}\) to Jurkat cells expressing DTPA-R-mRuby3. (B) Radiosynthesis procedure including radiofluorination, preparative HPLC, and charging of the CHX-A"-DTPA chelator with \(\mathrm{Na^{+}Tb^{III}}\) . (C) Quality control by analytical HPLC of the radiosynthesis product before and after preparative HPLC purification. (D) Binding assay of \([^{18}\mathrm{F}]\) F-DTPA to PC3DTPA-R or PC3Colchi-R cells (statistical analysis: t-test). (E) Analysis of the internalization of \([^{18}\mathrm{F}]\) F-DTPA upon binding to PC3 cells by analyzing the Accutase-cleavable and the non-cleavable + +<--- Page Split ---> + +fraction. Accutase efficiently cleaves the DTPA- R ectodomain and, thus, allows the identification of bound radioligand that is accessible on the cell surface (statistical analysis: t- test). + +![](images/Figure_4.jpg) + +
Figure 3
+ +Pharmacokinetic and stability of [18F]- DTPA in vivo + +(A- D) Dynamic PET scan of CD1- nude mice carrying subcutaneous PC3DTPA- R (right shoulder) and PC3Colchi- R (left shoulder) xenograft tumors. (A) Maximum Intensity Projections (MIPs) of a dynamic PET + +<--- Page Split ---> + +scan between 1 and 90 min p.i. of 11.4 MBq [18F]F- DTPA and (B) an axial PET plane through both tumors at \(0 - 25\%\) ID/g and \(0 - 2.5\%\) ID/g, showing selective accumulation in the PC3DTPA- R tumor but no elevated signal in the PC3Colchi- R tumor. (C) Linear and (D) logarithmic representation of the time- activity curve derived from the dynamic PET scan. VOI quantification using 10 pixel spheres revealed increasing signal- to- background ratios over time. (E) Ex vivo biodistribution analysis at \(t = 90\) min after [18F]F- DTPA i.v. injection into male and female C57BL/6 mice. Please note, mice were awake between injection and \(t = 90\) min, and not under anesthesia as for the dynamic PET scan, causing different pharmacokinetic profiles. No significant differences were found (t- test). (F) Separation of the intact [18F]F- DTPA radioligand, its hydrolysis fragments, and urine samples collected from mice injected beforehand with [18F]F- DTPA by size- exclusion chromatography. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 4
+ +## Longitudinal PET imaging study of DTPA-R expressing CAR-T cells + +(A) The design of the longitudinal CAR-T cell study involved NSG mice injected i.v. with Raji tumor cells \((0.5 \times 10^6)\) and, after 7 days, i.v. injection of \(2 \times 10^6\) DTPA-R or EGFRt expressing anti-CD19 CAR-T cells. PET scans were recorded during treatment on days 4, 8, 15, 22 and 30. MIP of an exemplary (B) αCD19-CAR-TDTPA-R and (C) αCD19-CAR-TEGFRt mouse each is depicted for the respective days. (D) The + +<--- Page Split ---> + +cumulative specific signal from every \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA scan was quantified for the upper body (above the gall bladder) and the hind legs. This quantification allowed monitoring of global CAR- T cell expansion and/or decline over time. FACS analysis of total cell numbers of (E) Raji tumor cells and (F) CAR- T cells in different extremities, spleen, and blood. (G- J) Quantification of local CAR- T infiltrations over the course of the treatment. (G) A sagittal PET/MR scan on day 22 revealed different bone marrow infiltrations. (H) Axial PET/MR enabled the quantitative assessment of local CAR- T cell infiltration over time. (I) Longitudinal VOI quantification allowed the differentiation between tumor lesions experiencing an intermediate CAR- T expansion and subsequent decline (grey, blue, and orange arrowheads in G/H) and lesions with constant CAR- T infiltration and/or expansion (green arrowhead in G/H). (J) Immunohistochemistry of sagittal spine sections containing (green, blue & orange) lesions with infiltrating CAR- T cells (V5- tag, brown). (K) Overlay of PET signals on the V5- IHC. (L) Double staining of a consecutive tissue section for CAR- T cells (V5- tag, red) and Raji lymphoma cells (CD19, brown). + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 5
+ +## Quantification and detection limit for DTPA-R labeled lymphocytes + +(A- B) Flow cytometry quantification of CAR- \(\mathsf{TDPA - R}\) cells in the bone marrow of mice after PET imaging. (A) Cells within hollow bones were harvested for ex vivo analysis. (B) Correlation of CAR- T cell numbers found by flow cytometry with the PET signal obtained from hollow bones (with exemplary PET/MR cross- section images). (C) Individual lesions in the bone marrow were quantified, and CAR- \(\mathsf{TDPA - R}\) cell numbers were calculated by interpolation using the regression line from (B). (D) Maximum signal capacity was assessed by staining Jurkat \(\mathsf{DTPA - R}\) cells in vitro with \([^{18}\mathsf{F}]\mathsf{F}\) - DTPA. PET phantom study: Jurkat \(\mathsf{DTPA - R}\) or Jurkat \(\mathsf{Colchi - R}\) cells were incubated with \([^{18}\mathsf{F}]\mathsf{F}\) - DTPA, washed three times, counted, and + +<--- Page Split ---> + +10 μl of the suspension was transferred into 200 μl PCR tubes. PET signals were recorded for 60 min. (E, F) Spot assay: JurkatDTPA-R cells (E) or aCD19-CAR-DTPA-R cells (F) were injected subcutaneously into the back of mice. Five samples of a 1:2 dilution series of the DTPA-R expressing cells and of a control cell line were used. 90 min after i.v. [18F]F-DTPA injection, PET scans were recorded for 20 min. %ID of the spots was quantified by V0ls and plotted against the number of injected cells. + +![PLACEHOLDER_24_0] + +
Figure 6
+ +<--- Page Split ---> + +## PET imaging of AAV9 gene therapy via DTPA-R + +PET imaging of AAV9 gene therapy via DTPA- R C57BL/6 or CD1- nude mice were injected i.v. with (A) AAV9/2 viral vectors encoding the DTPA- R. MIPs are shown for a longitudinal study of (B) immunocompetent C57BL/6 mice and (C) immunocompromised CD1- nude mice at t=14, 21, and 28 days after injection of \(2.5 \times 10^{12}\) vg per mouse. (D) Quantification of PET signals by VOI spheres in selected organs throughout the study (statistical analysis: t- test). (E) MIP of C57BL/6 mice transduced with different titers \((1 \times 10^{11}, 5 \times 10^{11}, 1 \times 10^{12}\) and \(2.5 \times 10^{12}\) vg/mouse) as well as untreated control mice imaged after 7 days. (F) MIP and axial PET/MR overlays allow the identification of anatomical structures accumulating \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA. (G) Correlation of injected AAV9 titer and quantified PET signals for brown adipose tissue (BAT), liver, heart, and adrenal glands (n=3 per titer). (H) The slope of the increase in PET signals was correlated with the perfusion rate (obtained from literature) of the respective organ. (I) Sagittal PET and V5- IHC of the adrenal gland with quantification of positive cells in the zona fasciculata of the adrenal gland cortex. Correlations of (J) positive cell count and (K) PET signal with the injected viral vector dose, as well as (L) with each other. Correlation of detected vector genomes and PET signal (M). (N) Multivariate analysis correlating the PET signal (actual Y) with AAV9 dose, mRNA and viral genomes (predicted Y) in the liver. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementary DTPARPETreportergeneimagingNatBiotechnol.docx + +<--- Page Split ---> diff --git a/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353_det.mmd b/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d36554dbe6284bb5115a886bf16045ce87d9839c --- /dev/null +++ b/preprint/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353/preprint__11246a6b552172842bbbfed45762db4c17f3d4b47c6a40e36318ef72fec51353_det.mmd @@ -0,0 +1,466 @@ +<|ref|>title<|/ref|><|det|>[[43, 108, 923, 207]]<|/det|> +# DTPA-Receptor – A novel reporter gene system for the specific and sensitive PET imaging of CAR-T cells and AAV transduced cells + +<|ref|>text<|/ref|><|det|>[[43, 230, 211, 275]]<|/det|> +Wolfgang Weber w.weber@tum.de + +<|ref|>text<|/ref|><|det|>[[43, 303, 685, 370]]<|/det|> +Technical University of Munich Volker Morath Technical University of Munich https://orcid.org/0000- 0001- 7636- 0537 + +<|ref|>text<|/ref|><|det|>[[43, 375, 325, 416]]<|/det|> +Katja Fritschle Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 421, 325, 462]]<|/det|> +Linda Warmuth Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 468, 325, 508]]<|/det|> +Markus Anneser Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 514, 325, 554]]<|/det|> +Sarah Dötsch Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 560, 325, 600]]<|/det|> +Milica Zivanic Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 606, 217, 645]]<|/det|> +Luisa Krumwiede Helmholtz Munich + +<|ref|>text<|/ref|><|det|>[[43, 652, 325, 692]]<|/det|> +Philipp BösI Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 698, 891, 762]]<|/det|> +Tarik Bozoglu Medical Department I – Cardiology, Angiology, Pneumology, Klinikum rechts der Isar – Technical University of Munich, Ismaninger Strasse 22, 81675 Munich + +<|ref|>text<|/ref|><|det|>[[43, 768, 325, 808]]<|/det|> +Stephanie Robu Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 814, 185, 853]]<|/det|> +Silvana Libertini Novartis + +<|ref|>text<|/ref|><|det|>[[43, 860, 325, 900]]<|/det|> +Susanne Kossatz Technical University of Munich + +<|ref|>text<|/ref|><|det|>[[43, 906, 187, 925]]<|/det|> +Christian Kupatt + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 45, 858, 88]]<|/det|> +Large Animal Models in Cardiovascular Research, Clinic for Cardiology, TU Munich, Germany https://orcid.org/0000- 0002- 3611- 1218 + +<|ref|>text<|/ref|><|det|>[[43, 92, 911, 160]]<|/det|> +Markus Schwaiger Katja Steiger Institute of Pathology, School of Medicine, Technical University of Munich https://orcid.org/0000- 0002- 7269- 5433 + +<|ref|>text<|/ref|><|det|>[[43, 186, 696, 228]]<|/det|> +Dirk Busch Technical University of Munich https://orcid.org/0000- 0001- 8713- 093X + +<|ref|>text<|/ref|><|det|>[[43, 231, 664, 273]]<|/det|> +Arne Skerra Technical University Munich https://orcid.org/0000- 0002- 5717- 498X + +<|ref|>sub_title<|/ref|><|det|>[[43, 313, 104, 331]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 350, 940, 394]]<|/det|> +Keywords: reporter gene imaging, ATMP, synthetic biomarker, positron emission tomography (PET), cell tracking + +<|ref|>text<|/ref|><|det|>[[43, 410, 321, 430]]<|/det|> +Posted Date: August 23rd, 2023 + +<|ref|>text<|/ref|><|det|>[[42, 448, 476, 468]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3200226/v1 + +<|ref|>text<|/ref|><|det|>[[42, 485, 916, 529]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 546, 945, 611]]<|/det|> +Additional Declarations: Yes there is potential Competing Interest. V.M., K.F., A.Sk., and W.W. have applied for intellectual property rights on Anticains- based reporter genes (WO 20022/101492 A1). A.Sk. is a co- founder and shareholder of Pieris Pharmaceuticals. + +<|ref|>text<|/ref|><|det|>[[42, 647, 925, 691]]<|/det|> +Version of Record: A version of this preprint was published at Nature Biomedical Engineering on June 13th, 2025. See the published version at https://doi.org/10.1038/s41551- 025- 01415- 7. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 158, 68]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[40, 82, 945, 363]]<|/det|> +Advanced Therapy Medicinal Products (ATMPs), such as cell and gene therapies, necessitate a reliable diagnostic method for quantitative monitoring. We developed a novel reporter gene system for PET imaging consisting of a membrane- anchored Anticalin protein (DTPA- R) that acts as a high- affinity receptor for the radioligand \([^{18}\mathrm{F}]\) F- DTPA- Tb. The reporter protein shows high cell surface expression of up to \(\sim 1\times 10^{6}\) receptors per cell. After systemic administration, the pharmacologically inert radioligand rapidly clears via the renal route and, at \(\mathrm{t} = 90\) min, generates a strong signal of \(22.1\%\) dD/g for DTPA- R- expressing PC3 cells compared to \(0.2\%\) dD/g for DTPA- R- negative controls (ratio: 125). The detection limit for JurkatDTPA- R cells was 500 cells in a PET phantom ex vivo and 8,000 if subcutaneously injected. In vivo expansion and migration of CD19- CAR- TDPA- R cells was successfully monitored over four weeks with a linear relationship between PET signal and CAR- T cell number. Furthermore, our reporter system allowed quantitative and longitudinal imaging of AAV9 viral vectors with a linear dose- to- signal relation. In summary, DTPA- R shows high potential for in vivo monitoring of ATMP- based therapies. + +<|ref|>sub_title<|/ref|><|det|>[[44, 385, 204, 411]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[41, 426, 950, 612]]<|/det|> +Recent regulatory approvals of cell and gene therapies \(^{1,2}\) have shown that such advanced therapy medicinal products (ATMPs) can be used in clinical practice and have the potential to cure life- threatening diseases. Despite these exciting breakthroughs, the development of new cell and virus based ATMPs remains hampered by limited knowledge about the biodistribution and persistence of these live drugs \(^{3}\) . Technologies to monitor the distribution of Chimeric Antigen Receptor (CAR)- T cells in vivo non- invasively and quantitatively could greatly improve our understanding of their trafficking, therapeutic efficacy, and off- target toxicity \(^{4}\) . Likewise, Adeno- associated virus (AAV) based gene therapies would benefit from quantifying the location, magnitude, and duration of transgene expression. + +<|ref|>text<|/ref|><|det|>[[41, 628, 947, 764]]<|/det|> +Direct ex vivo labeling of cells (e.g., with iron oxide particles) is a sensitive approach for studying the distribution of transplanted cells early after injection. Although sensitive, this approach is limited at later time points because the label is diluted by each cell division, which prevents the assessment of cell proliferation. Furthermore, the label persists even after the cells have died or were phagocytosed. This limits the utility of direct labeling for CAR- T cell tracking, where a small fraction of the graft massively expands following activation. \(^{5}\) + +<|ref|>text<|/ref|><|det|>[[41, 781, 947, 924]]<|/det|> +As an alternative, cell graft proliferation and viability can be imaged by an indirect labeling strategy. This approach relies either on endogenous biomarkers or reporter genes (synthetic biomarkers) that can be detected by imaging \(^{6}\) . Various luciferase enzymes and fluorescent proteins have been used as reporter genes to study transplanted cells in mice by means of optical imaging techniques \(^{7}\) . However, light is strongly attenuated and scattered by biological tissues, which limits imaging depth and quantification in optical imaging \(^{8}\) , particularly in larger animals or humans \(^{7}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 950, 210]]<|/det|> +A viable modality for reporter gene imaging should allow for whole body- imaging, be highly sensitive and provide a quantitative, depth- independent signal6. Furthermore, imaging should be cross- sectional to allow for precise anatomical localization. PET (positron emission tomography) fulfills all of these requirements, thus enabling the spatiotemporal imaging of transgenes on a whole- body scale while being widely available for preclinical and clinical imaging4. The sensitivity of current clinical scanners lies in the picomolar range5 and has been further increased by a factor of 40 with the recent introduction of total body PET scanners9. + +<|ref|>text<|/ref|><|det|>[[39, 227, 950, 507]]<|/det|> +The ideal reporter gene/protein should be small, bio- orthogonal – i.e., not expressed endogenously and not affect cell function – and non- immunogenic. For PET imaging, a straightforward radiolabeling procedure with a commonly available radioisotope such as fluorine- 18 would be desirable. The small molecule probe should show low non- specific binding to cells, undergo rapid kidney clearance, but remain in the circulation sufficiently long to ensure quantitative labeling of cells expressing the reporter gene. Notably, despite numerous efforts over three decades10, no reporter gene system fulfills these requirements in a clinical setting. Herpes simplex thymidine kinase (HSV- tk)11 has been extensively studied as a reporter gene, but this viral protein is highly immunogenic with a high seroprevalence. On the other hand, systems like the sodium/iodide symporter (NIS)12, somatostatin receptor 2 (SSTR2)13, and truncated prostate- specific membrane antigen (tPSMA)14, 15 are not immunogenic but are expressed endogenously by various tissues which hampers their application for the specific whole- body imaging of grafted cells. + +<|ref|>text<|/ref|><|det|>[[41, 524, 940, 688]]<|/det|> +To address this unmet medical need, we developed and characterized a new reporter gene and radioligand system that combines high sensitivity with exquisite specific. Our approach is based on an Anticalin16, 17 that is expressed on the cell surface and binds a bio- orthogonal 18F- labeled probe. Anticalins are engineered binding proteins based on the lipocalin protein family18, endogenous plasma proteins comprising a single polypeptide with a robust \(\beta\) - barrel fold19. We have extensively studied this new reporter system in vivo and demonstrated its suitability to monitor CAR- T cells and AAV- mediated gene transfer in mice. + +<|ref|>sub_title<|/ref|><|det|>[[44, 712, 144, 737]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[44, 749, 738, 781]]<|/det|> +## Design of the reporter protein and radioligand + +<|ref|>text<|/ref|><|det|>[[41, 797, 947, 958]]<|/det|> +The PET reporter protein has been designed as an artificial cell surface protein exhibiting a highly specific binding site for a small molecule ligand (Fig. 1A). The protein construct comprises the lipocalin- 2 signal peptide, an engineered lipocalin binding protein (Anticalin18), a peptide linker containing the V5- tag (a multifunctional molecular handle for staining of ATMPs)20 and the \(\alpha\) - helical human CD4 transmembrane domain. This fusion protein contains only 257 amino acid residues and is encoded in a 774 base pairs (bp) open reading frame (ORF; for sequences see the Supplementary Information). Finally, for multimodal detection, an optional fluorescent protein was added to the C- terminal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 950, 216]]<|/det|> +cytoplasmic end, such as mRuby \(^{321}\) or miRFP720 \(^{22}\) (Fig. 1B). Two Anticilins were chosen to construct different reporter proteins with two ligand specificities: (i) the Anticalin CL31d (DTPAR), which binds CHX- A"- DTPA- metal complexes with a \(K_{\mathrm{D}}\) of \(\sim 500 \mathrm{pM}^{16,23}\) and the Anticalin D6.4(Q77E) (ColchiR) which binds colchicine with a \(K_{\mathrm{D}}\) of \(\sim 20 \mathrm{pM}^{24}\) . The cognate reporter probe features CHX- A"- DTPA- metal (Fig. 1C) or colchicine (Fig. 1D), respectively, a PEG \(_{4}\) - linker to facilitate access to the Anticalin binding pocket without steric hindrance, optionally hydrophilic groups to improve pharmacokinetics and a labeling group for efficient incorporation of the \(^{18}\mathrm{F}\) isotope to enable PET detection (Fig. 1C). + +<|ref|>sub_title<|/ref|><|det|>[[42, 217, 591, 248]]<|/det|> +## Analysis of reporter protein variants + +<|ref|>text<|/ref|><|det|>[[38, 260, 955, 814]]<|/det|> +To compare the two Anticalin- ligand pairs, we generated cell lines expressing the respective reporter genes by retroviral transduction of the human T cell line Jurkat, human embryonic kidney cells (HEK293T), and the prostate carcinoma cell line PC3. Initially, cell surface expression of the DTPA- R or ColchiR reporter proteins on PC3 cells was confirmed by fluorescence microscopy using an anti- V5- tag antibody (Fig. 1E). Membrane localization of mRuby3 in DTPARmRuby3 expressing HEK293T cells indicated efficient transport of the reporter protein through the secretory pathway (Fig. S1). Absolute reporter protein numbers per cell were measured by flow cytometry using an anti- V5- tag antibody and MESF calibration beads. Expression levels on Jurkat cells ranged from \(\sim 16,000\) for DTPA- R- miRFP720 to around 1 million receptors for DTPA- R (Fig. 1F). Omitting the optional fluorescent protein increased expression, e.g., DTPA- R showed 8.3- fold ( \(p < 0.0001\) ) higher expression than DTPA- R- mRuby3 (Fig. 1F). Next, we studied the influence of T cell activation on reporter gene expression by comparing PMA/lonomycin- activated to untreated Jurkat cells, which did not result in significant differences (Fig. 1F & S3A). Furthermore, the type of Anticalin influenced the expression levels: a 4.9- fold ( \(p < 0.0001\) ) higher expression of DTPA- R was measured compared to Colchi- R (Fig. 1F, S2 & S3A). We also compared the expression level of DTPA- R with a previously described reporter gene that uses the murine single chain variable fragment (scFv) C825 \(^{25}\) , or its humanized version huC825 \(^{26}\) , to bind DOTA- metal complexes (Fig. S3A). Using the same expression cassette and membrane anchor domain for these scFvs, DTPAR showed a 5.8 (muC825) or 8.9 (huC825) fold higher median V5- AF488 signal after retroviral transduction. The number of genomically inserted expression cassettes per cell was quantified by ddPCR and showed similar levels for the different constructs (Fig. S3B). However, transgene mRNA levels varied between constructs (Fig. S3C). While protein levels of Anticalin- based reporters correlated with mRNA levels ( \(R^{2} = 0.98\) ), scFv- based reporter genes produced high mRNA levels, which did not translate into elevated protein levels, indicating superior protein folding, stability, and cell- surface presentation for Anticalin- based reporter proteins (Fig. S3D). + +<|ref|>text<|/ref|><|det|>[[41, 827, 950, 940]]<|/det|> +A potential metabolic burden of the reporter gene expression was assessed by measuring doubling times of non- activated or PMA/lonomycin- activated Jurkat cell lines using a carboxyfluorescein succinimidyl ester (CFSE) based proliferation assay (Fig. 1G). The doubling time of wild- type Jurkat cells (non- activated: \(18.3 \pm 4.3 \mathrm{h}\) / activated: \(22.2 \pm 4.7 \mathrm{h}\) ) was not substantially changed by reporter gene expression (DTPA- R: \(17.0 \pm 4.1 \mathrm{h}\) / \(20.6 \pm 4.5 \mathrm{h}\) ). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 951, 163]]<|/det|> +Further applications of the V5- tag as part of the DTPA- R include the isolation of reporter gene expressing cells using magnetic- activated cell sorting. Employing the V5- tag, JurkatDTPA- R cells could be isolated by magnetic- activated cell sorting (MACS)27 from a 5:95 cell mixture with high efficacy and purity (> 98%) (Fig. 1H). Finally, the V5- tag also enabled the detection of the reporter protein at the cellular level via immunohistochemistry (IHC) (Fig. S4). + +<|ref|>sub_title<|/ref|><|det|>[[42, 163, 945, 224]]<|/det|> +## Synthesis & characterization of [18F]-colchicine and [18F]-DTPA + +<|ref|>text<|/ref|><|det|>[[38, 238, 951, 702]]<|/det|> +First, we investigated the binding of \(\mathsf{NH}_2\) - CHX- A"- DTPA in complex with \(^{90}\mathrm{Ylll}\) to JurkatDTPA- R cells using a competitive assay, resulting in a half- maximal inhibitory concentration (IC50) of 1.4 nM (Fig. 2A). However, fluorine- 18 is the preferred PET radioisotope as it is widely available and almost every decay leads to the emission of a positron (ratio 96.9%) with desirably low energy (Emean=250 keV, average positron range of 0.6 mm28). On the other hand, positron- emitting radiometals have inferior physical characteristics, and their \(\mathsf{NH}_2\) - CHX- A"- DTPA complexes often bind with lower affinity to DTPA- R than \(\mathsf{Ylll}\) or \(\mathsf{Tb}^{III}\) complexes16, 23. For this reason, we designed a DTPA- based radioligand that can be charged with radioactive or non- radioactive metal ions and features a second group that can be easily radiolabeled with fluorine- 18 (Fig. 2B & S6). This radiolabeling approach uses a trimethylamine leaving group29, similar to the approved radioligand [18F]PSMA- 100730. For comparison, we synthesized analogous radioligand precursors for Colchi- R employing different numbers of strongly polar Glu residues and a precursor with two Glu residues for DTPA- R (Fig. S5). After quality control (Fig. S6E), both compounds were radio- fluorinated in a single step and purified by reversed phase HPLC to obtain highly pure radioligands (Fig. 2C & S7A). For [18F]-F- Nic- D- Glu2- PEG4- CHX- A"- DTPA- \(\mathsf{Tb}^{III}\) (dubbed [18F]-F- DTPA) a radiochemical yield of \(\sim 20\%\) and a radiochemical purity of \(>98\%\) were achieved. Subsequently, the 18F- labeled radioligands were tested in ligand binding assays with DTPA- R and Colchi- R expressing PC3 cells (Fig. 2DE & S7B). The binding of both radioligands was highly specific to cells expressing the corresponding receptor ( \(\sim 1,000\) - fold for DTPA- R and \(\sim 500\) - fold for Colchi- R) and could be efficiently blocked by competition with non- radioactive ligand ( \(\sim 100\) - fold for DTPA- R and \(\sim 400\) - fold for Colchi- R). + +<|ref|>sub_title<|/ref|><|det|>[[44, 700, 904, 732]]<|/det|> +## Dynamic PET imaging of mice bearing xenograft tumors + +<|ref|>text<|/ref|><|det|>[[40, 744, 955, 936]]<|/det|> +As a proof- of- principle for reporter gene imaging, the respective radioligands were injected into CD1- nude mice carrying subcutaneous PC3DTPA- R and PC3Colchi- R xenograft tumors above the right and left shoulder, respectively (Fig. 3AB & S7C). Dynamic [18F]- F- DTPA PET imaging showed fast, almost exclusively renal clearance, accompanied by increasing signals in the kidneys, ureter, and urinary bladder, with no retention in other tissues (Fig. 3A). The radioligand showed rapid accumulation in the PC3DTPA- R xenograft tumor, resulting in a stable activity concentration of 22.8±0.6% Injected Dose (ID)max/g between 30 and 90 min post injection (p.i.). While both xenograft tumors were clearly visible on MR images (Fig. 3B), only the PC3DTPAR tumor could be delineated in PET. The maximum activity + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 951, 165]]<|/det|> +concentration in the \(\mathrm{PC3^{Colchir}R}\) xenograft \((0.18\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g})\) was 125- fold lower than for the cognate \(\mathrm{PC3^{DTPA - R}}\) xenograft. \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in the blood pool was quickly cleared with a plasma half- life of 0.5- 1.5 min, which resulted in a tumor- to- blood ratio of 261 after 90 min. Furthermore, in a second static PET scan 6 h p.i., still \(48\%\) of the initial \(\mathrm{PC3^{DTPA - R}}\) signal was retained (Fig. 3CD). Overall, these results indicate an exquisite specificity of the DTPA- R reporter gene system in vivo. + +<|ref|>text<|/ref|><|det|>[[41, 183, 955, 297]]<|/det|> +The biodistribution of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA at 90 min p.i. in healthy female and male mice was analyzed ex vivo to study sex- specific differences \((n = 5\% +5\%)\) . Very low retention of below \(0.2\% \mathrm{ID} / \mathrm{g}\) was measured for most tissues (Fig. 3E), except for the kidneys \((0.9\% \mathrm{ID} / \mathrm{g})\) and the hepato- biliary excretion route (small intestine: \(0.2\% \mathrm{ID} / \mathrm{g}\) ; large intestine: \(0.1\% \mathrm{ID} / \mathrm{g}\) ). No significant sex- specific differences in biodistribution were observed. (Fig. 3E). + +<|ref|>text<|/ref|><|det|>[[41, 315, 936, 406]]<|/det|> +In vivo stability of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA was analyzed after intravenous (i.v.) injection using size- exclusion chromatography (SEC) to separate intact radioligand from fragments with lower molecular weight. \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA within urine samples collected from two mice \(\sim 3\) h p.i. was \(>97\%\) intact, indicating serum stability (Fig. 3F). + +<|ref|>text<|/ref|><|det|>[[40, 424, 945, 615]]<|/det|> +For comparison, a dynamic PET study was conducted for the radioligand \([^{18}\mathrm{F}]\mathrm{F}\) - Nic- Glu \(_2\) - PEG \(_4\) - colchicine ([18F]F- colchicine). This radioligand showed not only renal (19.8%ID) but also significant hepato- biliary excretion (69.7%ID), which is undesirable due to the strong and variable background PET signals in the gastrointestinal tract (Fig. S7D). The elevated hepato- biliary excretion of \([^{18}\mathrm{F}]\mathrm{F}\) - colchicine is likely explained by its lower hydrophilicity, as reflected by the \(\mathrm{LogD}_{7,4}\) of - 2.97 vs. - 3.58 for \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA. Taken together, these pharmacokinetic data, as well as the higher expression levels of the DTPA- R on transfected cells and the much higher radioligand uptake in the xenograft tumor, prompted us to focus on the further development of the DTPA- R reporter system. + +<|ref|>sub_title<|/ref|><|det|>[[45, 616, 848, 647]]<|/det|> +## Design and evaluation of DTPA-R labeled CAR-T cells + +<|ref|>text<|/ref|><|det|>[[40, 661, 940, 898]]<|/det|> +To investigate the application of DTPA- R in the context of human CART cell therapy, we assessed in an animal study if the migration, proliferation, and tissue infiltration of \(\mathrm{CART^{DTPA - R}}\) cells can be imaged. To this end, we employed a well- characterized CD19- targeting 2nd- generation CAR based on the scFv FMC63 (Fig. S8A) \(^{31}\) . The corresponding expression cassette comprised the CAR followed by a 2A self- cleaving sequence separating a second membrane protein, the truncated Epidermal Growth Factor Receptor (EGFRt) \(^{32}\) . EGFRt lacks normal EGFR function but is still recognized by cetuximab (Erbitux) for in vivo cell ablation \(^{33}\) . To generate \(\mathrm{CART^{DTPAR}}\) cells, we replaced the EGFRt in the retroviral expression cassette with DTPA- R, which enables PET imaging, apart from optionally serving for cell ablation (Fig. S8B). Both CAR- T vectors were used to transduce human peripheral blood mononuclear cells (PBMCs), resulting in a surface expression of approximately 100,000 DTPA- R molecules per CAR- T cell (Fig. S8C). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 46, 945, 167]]<|/det|> +There was no substantial difference in CD3, CD4, and CXCR3 expression levels or \(\mathsf{CD8^{+} / CD4^{+}}\) ratios between CAR- TEGFRt and CAR- TDPTA- R cells. (Fig. S8CD). CAR- T cell function was assessed in a real- time assay of the CAR- mediated killing of HEKCD19 target cells (xCELLigence; Fig. S8E) as well as a 51Cr- release assay on Raji cells (Fig. S8F). Specific lysis was comparable between the aCD19CARTEGFRt and aCD19CARTDPTAR cell treated groups in both assays. + +<|ref|>sub_title<|/ref|><|det|>[[44, 168, 567, 201]]<|/det|> +## PET imaging of CAR-T cell therapy + +<|ref|>text<|/ref|><|det|>[[41, 214, 944, 405]]<|/det|> +Both CAR- T cells were used to treat NSG mice engrafted with a systemic Raji tumor, which is known to primarily home to the bone marrow34. After 7 days of Raji engraftment, \(2\times 10^{6}\) aCD19- CARTDPTAR or reference aCD19- CARTEGFRt cells were infused. This experimental setting ensured constant CART expansion due to non- complete tumor eradication, which was imaged with [18F]FDPTA on days 4, 8, 15, 22, and 30 (Fig. 4A). PET signals indicated the presence of CARTDPTAR cells at day 4 in the spleen and from day 8 onwards, with increasing intensity, in the bone marrow of the extremities as well as spine, scull, and axillary lymph nodes (Fig. 4B or Fig. S9 for all animals). In contrast, PET signals were absent in mice injected with aCD19- CARTEGFRt cells, demonstrating the specificity of [18F]F- DTPA (Fig. 4C). + +<|ref|>text<|/ref|><|det|>[[40, 420, 951, 794]]<|/det|> +Quantification of the cumulated specific PET signals allowed the assessment of CAR- T cell expansion over time (Fig. 4D). Furthermore, we used flow cytometry at the end of the study, or when the humane endpoint was reached, to quantify the numbers of CAR- T (Fig. 4E) and Raji cells (Fig. 4F). This ex vivo analysis showed differences in CAR- T cell numbers, confirming the high variability between animals and the need for imaging of these therapies. In addition, 3D cross- sectional images of the [18F]F- DTPA PET/MR allowed the monitoring of local CAR- T infiltration (Fig. 4G- I). Longitudinal PET analysis revealed different trajectories of CAR- T infiltrated lesions (Fig. 4HI). Ex vivo analysis by V5- IHC of sagittal spine sections allowed CAR- TDPTA- R identification on the cellular level, which confirmed CAR- T infiltration into individual vertebra bodies ( \(\sim 1.5\times 0.5 \text{mm}\) ) and, occasionally, invasion through the vertebral bone into the spinal canal (Fig. 4J). Importantly, the pattern of CAR- T infiltration determined by IHC was very well matched by the signals in PET/MR (Fig. 4K). Furthermore, the co- localization of CAR- TDPTAR with Raji cells was confirmed by IHC co- staining of CD19 and the V5- tag (Fig. 4L). Interestingly, lesions with a CAR- T cell plateau in longitudinal PET imaging (blue and orange, Fig. 4I) showed lower tumor cell density, which indicated better tumor clearance by CART cells compared to neoplasms with constantly increasing PET signal (green, Fig. 4I & S10A). Other organs like the spleen, kidneys, liver, and lung only showed few CAR- TDPTA- R cells in V5- IHC (Fig. S10B). + +<|ref|>text<|/ref|><|det|>[[41, 809, 944, 950]]<|/det|> +Correlation between the actual number of CAR- TDPTA- R cells detected by flow cytometry after extraction from a hollow bone and the cumulated PET signal obtained for the same extremity (Fig. 5A & S11AB) was performed for mice sacrificed on day 8 or 15 (Fig. 5B, gating see S11C) and on day 16 or 31 (Fig. S11D). A linear relationship ( \(R^2 = 0.92\) ) indicated that PET imaging of DTPA- R can quantitatively monitor tissue- specific CAR- T cell infiltration. Utilizing the resulting regression line, we calculated a detection limit of around 1,200 CARTDPTAR cells in delineated lesions in the bone (Fig. 5C). To further characterize + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 43, 950, 240]]<|/det|> +the relationship between PET signal and the number of DTPA- R expressing cells, we measured the signal of defined numbers of JurkatDTPAR cells in a PET phantom in vitro. For cells labeled with [18F]F- DTPA, we observed a linear correlation (R2 = 0.999) with a detection limit of 500 cells (Fig. 5D). In the next step, we determined the detection limit in tissue with low perfusion. Therefore, we injected 4,000 to 64,000 JurkatDTPA- R cells, or CARTDTPAR cells, in the dorsal subcutis, immediately followed by i.v. injection of [18F]F- DTPA (Fig. 5EF). Here, 8,000 cells were clearly detectable by PET. Again, a linear correlation between cell number and PET signal was observed (R2 = 0.96 and R2 = 0.99, respectively; see Fig. 5EF). + +<|ref|>sub_title<|/ref|><|det|>[[42, 213, 852, 273]]<|/det|> +## PET imaging of gene transfer mediated by AAV9 viral vectors + +<|ref|>text<|/ref|><|det|>[[39, 288, 951, 690]]<|/det|> +In vivo gene therapy represents another relevant application scenario for reporter gene imaging, where PET imaging can help assess the roles of administration routes, dosing regimens, pharmacokinetics of capsid- modified vectors, or endurance of gene expression. As a common example for gene therapy, we have selected vectors based on the Adeno- associated virus (AAV)35. In particular, we have focused on the serotype AAV9, which sparked great interest for transducing muscle and neuronal tissue36 and provided the basis for the FDA- approved drug Zolgensma35. We constructed a transfer plasmid flanked by AAV2 inverted terminal repeats (ITR) and used it to produce AAV2/9 viral vectors encoding the DTPA- R reporter gene under the control of the CMV promoter (Fig. 6A). First, we investigated the reporter gene expression in mice after systemic injection of AAV9DTPA - R. Immunocompetent C57BL/6 (Fig. 6B) and immunocompromised CD1- nude (Fig. 6C) mice (n = 3 per group) were dosed with \(2.5 \times 10^{12}\) viral genomes (vg) per mouse and, on days 14, 21 and 28 [18F]F- DTPA PET scans were recorded (Fig. 6BC). Although cohorts were treated in parallel using the same virus batch, there were marked differences between the transduction patterns obtained in C57BL/6 and CD1- nude mice. Isocontour segmentation of the heart yielded a value of \(15.3 \pm 3.8\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) for C57BL/6 and \(28.7 \pm 5.4\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) for CD1- nude mice. Within each cohort, animals showed comparable transduction patterns (Fig. S12). Moreover, we observed consistent trajectories of the repeated PET measurements, indicating stability of the vector expression over time (Fig. 6D). + +<|ref|>text<|/ref|><|det|>[[39, 707, 950, 920]]<|/det|> +We then i.v. injected four different titers, ranging from \(1 \times 10^{11}\) to \(2.5 \times 10^{12}\) vg/mouse, into C57BL/6 mice (n = 3 per group). After 7 days, [18F]F- DTPA PET/MR scans were recorded for all cohorts and an untreated control cohort (Fig. 6E or Fig. S13 for all animals). The transduction patterns were similar to the previous experiment; for example, mice with the highest applied titer showed clear PET signals in dorsal brown adipose tissue (BAT; \(\% \mathrm{ID}_{\mathrm{max}}: 7.7 \pm 1.5\) ), heart muscle ( \(\% \mathrm{ID}_{\mathrm{max}}: 6.1 \pm 3.6\) ), liver ( \(\% \mathrm{ID}_{\mathrm{max}}: 11.3 \pm 1.4\) ) and adrenals ( \(\% \mathrm{ID}_{\mathrm{max}}: 29.5 \pm 7.0\) ), whereas the background uptake in the upper body of the control mice was \(\sim 0.5 \pm 0.06\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) . Of note, the high resolution in axial planes provided by the combination of PET with an isomorph MRT allowed the precise analysis of transduction events (Fig. 6F) far beyond previous results obtained by bioluminescence imaging36. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 945, 236]]<|/det|> +Lower titers resulted in gradually reduced PET signals within the respective tissues. Importantly, PET signals ( \(\% \mathrm{ID}_{\mathrm{max}} / \mathrm{g}\) ) showed a linear correlation with the administered dose in BAT ( \(R^2 = 0.73\) ), heart muscle ( \(R^2 = 0.60\) ), liver ( \(R^2 = 0.92\) ), and adrenals ( \(R^2 = 0.89\) ), which underlines the quantitative character of DTPA- R PET imaging (Fig. 6G). Well- perfused organs showed a steeper increase of the PET signal with increasing AAV titer. This observation could either be due to a higher transduction rate in these organs or a better local delivery of [18F]F- DTPA. To test this hypothesis, we compared PET data for different tissues with perfusion values from the literature \(^{37,38}\) . Indeed, a linear correlation between the perfusion index and the slope of PET signal increase was seen (Fig. 6H). + +<|ref|>text<|/ref|><|det|>[[40, 252, 952, 488]]<|/det|> +Next, we stained tissues by V5- IHC and observed transduction of individual cells, for example in the cortex of adrenal glands (Fig. 6l). Quantification using automated positive cell detection showed \(2\%\) positive cells for animals transduced with the lowest, \(33\%\) for the low, and \(72.2\%\) for the high dose, which yielded a perfectly linear relation with the injected vector titer ( \(R^2 = 1\) ; Fig. 6K). Also, the relationship between the measured PET signal and the injected AAV titer (Fig. 6J), or the transduction levels from histology (Fig. 6L), indicated a linear correlation ( \(R^2 = 0.98\) or 0.96, respectively). Transduction levels were further quantified by ddPCR, revealing a linear correlation between the number of viral genomes and the PET signals in the liver ( \(R^2 = 0.97\) ; Fig. 6M). A multivariate analysis correlating the measured PET signals with injected AAV9 doses, mRNA levels, and viral genomes in the liver revealed a linear relationship ( \(R^2 = 1\) , Fig. 6N). + +<|ref|>sub_title<|/ref|><|det|>[[44, 510, 191, 535]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[40, 550, 951, 806]]<|/det|> +We have developed a novel reporter gene system based on a membrane- anchored Anticalin that specifically binds a small- molecule radioligand, enabling quantitative and longitudinal PET imaging of ATMPs with remarkable specificity and sensitivity. Based on PET with [18F]F- DTPA, we have demonstrated the suitability of DTPAR in relevant use cases such as CAR- T cell therapy of CD19 lymphoma \(^{39}\) and AAV9 gene therapy \(^{35}\) . CAR- T cell movement and tumor homing in a mouse model of systemic lymphoma could be visualized and quantitatively tracked in vivo over 30 days. Furthermore, the transduction of AAV9 vectors in distinct tissues could be quantitatively analyzed by molecular imaging. This theranostic approach, combining novel cell and gene therapies with a quantitative imaging modality using a universal reporter system that potentially can bridge preclinical development and clinical evaluation, should facilitate the clinical translation of ATMPs \(^{40}\) . To this end, our PET reporter system offers promising functional features, which are described below. + +<|ref|>text<|/ref|><|det|>[[41, 821, 950, 940]]<|/det|> +High expression level of the reporter protein determines strength of the signal, which has been demonstrated for Jurkat \(^{\mathrm{DTPA - R}}\) cells displaying \(\sim 1 \times 10^6\) receptors per cell without measurable negative effects on cellular fitness or T cell function. This number is approximately 10- fold higher compared to well- known lymphocyte receptors, such as the B cell receptor ( \(1.2 \times 10^5\) copies \(^{41}\) ) or CD4 ( \(\sim 1 \times 10^5\) copies \(^{42}\) ). Direct comparison of the Anticalin CL31d specific for [18F]FDTPA with the scFv huC825, which + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 44, 950, 163]]<|/det|> +binds DOTA- metal, resulted in an 8.9- fold higher expression of DTPA- R (Fig. S2). Low surface densities for huC825- based reporter proteins are also reflected by transfected HEK293T cells expressing only \(\sim\) \(1.5\times 10^{4}\) receptors per cell \(^{26}\) . In contrast to scFv antibody fragments, known for their oligomerization and aggregation tendencies \(^{43}\) , Anticalins possess a robust fold, are composed of a single polypeptide chain, and can be expressed as recombinant proteins at high levels \(^{18}\) . + +<|ref|>text<|/ref|><|det|>[[41, 180, 944, 298]]<|/det|> +Minimal gene size is of importance due to the limited packaging capacities of viral and non- viral gene shuttles and met by DTPA- R, which is encoded by just 774 bp ( \(\sim 26.4\) kDa for the mature fusion protein), in contrast to most other PET reporter proteins such as HSV1- tk (1,131 bp) \(^{11}\) , NIS (1,932 bp) \(^{12,44}\) , tPSMA \(^{(\mathrm{N9del})}\) (2,226 bp) \(^{15}\) , SSTR2 (1,110 bp) \(^{13}\) , DAbR1- 2A- GFP (2,280 bp) \(^{25}\) , SNAPtag (846 bp) \(^{45}\) , eDHFR (480 bp) \(^{46}\) or the EGFRt (1,074 bp) \(^{32}\) . + +<|ref|>text<|/ref|><|det|>[[41, 314, 933, 495]]<|/det|> +Functional inertness, including lack of biological activity, interference, and toxicity of both the reporter gene and probe, are crucial factors as the capability of in vivo imaging is second to the therapeutic function of an ATMP. We have investigated the proliferation kinetics, activation status, receptor expression, and cellular toxicity of CAR- T cells and found no difference due to the DTPA- R expression compared to EGFRt. Given the binding activity of DTPA- R for an exogenous hapten, interferences with cellular processes are much less likely than for reporter proteins that lead to ion transport over the cell membrane (NIS), represent tumor- associated surface antigens (tPSMA, SSTR2) or possess catalytic activities (HSV- tk, PSMA, SNAPtag, eDHFR). + +<|ref|>text<|/ref|><|det|>[[40, 511, 945, 753]]<|/det|> +Stability of the PET signal over time is important to allow reproducible imaging, which mainly relies on the receptor- ligand affinity. While the soluble recombinant CL31d protein shows a \(K_{\mathrm{D}}\) value of 543 pM for the complex with CHX- A"DTPA- Y \(^{17}\) , the half- maximal inhibitory concentration (IC \(_{50}\) ) of \(\mathrm{NH}_{2}\) - CHX- A"- DTPA- \(^{90}\mathrm{Y}\) for DTPA- R expressed on the cell surface was slightly higher, with 1.4 nM. Nevertheless, the high stability of the [ \(^{18}\mathrm{F}\) ]F - DTPA signal within the tumor was demonstrated in dynamic PET scans. This is not the case for the NIS reporter gene, for example, due to the naturally occurring efflux of the radioactive iodide \(^{44}\) . Potential internalization of DTPA- R is very low due to the absence of the cytosolic CD4 domain, which triggers internalization of CD \(^{47}\) . In addition, DTPA- R expression levels are independent of T cell activation, which is commonly known to heavily influence expression profiles in T cells \(^{48}\) . + +<|ref|>text<|/ref|><|det|>[[41, 770, 949, 909]]<|/det|> +Linear relation between PET signal and the number of DTPA- R- labelled cells is mandatory to move beyond qualitative imaging. The DTPA- R system allows a strong correlation of the number of CART cells in the bone marrow or the applied AAV9 virus titer with the corresponding PET signal, thus enabling non- invasive quantification. The amount of bound [ \(^{18}\mathrm{F}\) ]F- DTPA ligand is only governed by the local concentration of DTPA- R and the Law of Mass Action, contrasting to the much more complex and unpredictable relationships for transporter- or enzyme- type reporter proteins. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 953, 305]]<|/det|> +Criteria for a corresponding radioligand \(^{49}\) include its chemical composition as well as the choice of the radioisotope for PET imaging. The choice of \(^{18}\mathrm{F}\) , with its ideal physical half- life, high positron yield, and low positron energy, allows detection with high sensitivity and resolution, especially if compared to isotopes such as \(^{111}\mathrm{In}\) or \(^{86}\mathrm{Y}\) , used for DAbR1 / C825 reporter probes \(^{25,26}\) . Other advantages of \(^{18}\mathrm{F}\) are the relatively short half- life of 109 min, which enables repeated serial imaging in a daily interval, while the half- live is long enough to allow multiple PET scans from a single batch of radioligand. Furthermore, the low radiation exposure does not hamper the function of the ATMP nor lead to a significant absorbed biological dose. The stability of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in blood was demonstrated, together with a stable PC3 \(^{DTPA^{- }}\) \(R\) signal in the time- activity curve. Nevertheless, achieving high specific activity remains a challenge for \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA synthesis, as well as for other \(^{18}\mathrm{F}\) - based radioligands, and constitutes a current area of improvement. + +<|ref|>text<|/ref|><|det|>[[39, 321, 952, 512]]<|/det|> +The absence of endogenous binding activity was impressively demonstrated by the very low background accumulation of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in various organs. This is in marked contrast to all fully human reporter proteins described so far, including NIS with its thyroidal, gastric, mucosal, salivary, and lactating mammary gland expression \(^{50}\) . Furthermore, the biodistribution analysis of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA in female and male mice showed no sex- specific differences. Excretion of \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA was mainly seen via the renal route, which led to defined signals in the kidneys and bladder. In contrast, the radioligand for the C825- scFv- baed reporter gene system caused a diffuse signal in the abdomen at \(t = 30\) min p.i., and only \(\sim 80\%\) of the \([^{86}\mathrm{Y}]\mathrm{Y}\) - DOTA- Bn was cleared via the renal route \(^{26}\) . + +<|ref|>text<|/ref|><|det|>[[41, 527, 936, 620]]<|/det|> +Furthermore, the ease of probe preparation and the price per imaging day are of importance both in biomedical research and in a clinical setting. In this regard, the DTPA- R system is attractive as the chemical synthesis of the ligand precursor is straight- forward. Furthermore, radiolabeling involves only few steps and the supply of \([^{18}\mathrm{F}]\mathrm{F}^{- }\) is neither limiting nor expensive. + +<|ref|>text<|/ref|><|det|>[[39, 636, 940, 902]]<|/det|> +Finally, the sensitivity of signal detection is a crucial aspect, which depends on the number of binding sites, radioligand affinity, physical decay characteristics of the isotope and the specific activity of the radioligand. Using a PET phantom, we determined a detection limit of 500 cells while a standard \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA imaging protocol allowed the clear detection of as few as \(8 \times 10^{3}\) Jurkat \(^{DTPA - R}\) or αCD19- CAR- \(^{TDPA - R}\) cells in vivo. Reported sensitivity for DTPA- R is in line or even higher when compared to other reporter systems \(^{50}\) . In a comparative study with primary T cells transfected with human reporter genes \(^{51}\) , only the norepinephrine transporter (NET) with \([^{18}\mathrm{F}]\mathrm{F}\) - MFBG was able to detect \(3 - 4 \times 10^{4}\) cells, while for HSV- TK/ \([^{18}\mathrm{F}]\mathrm{F}\) - FEAU and NIS/ \([^{124}\mathrm{I}]\) - iodide a sensitivity of \(\sim 3 \times 10^{5}\) and \(\sim 1 \times 10^{6}\) was reported, respectively. In direct comparison with those reporter systems DTPA- R/ \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA not only achieved a higher sensitivity, but also shows the superior excretion profile which leads to lower background signals \(^{51}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 949, 90]]<|/det|> +In summary, the DTPA- R & [18F]F- DTPA system meets all relevant design and functional requirements for a universal reporter gene system, which may boost future ATMPs by making them PET- traceable. + +<|ref|>sub_title<|/ref|><|det|>[[44, 112, 210, 138]]<|/det|> +## Declarations + +<|ref|>text<|/ref|><|det|>[[44, 155, 208, 174]]<|/det|> +Acknowledgments + +<|ref|>text<|/ref|><|det|>[[42, 191, 939, 326]]<|/det|> +The authors wish to thank Markus Mittelhauser, Hannes Rolbieski, Sybille Reder, and Natalie Röder for PET/MR acquisition. Prof. Dr. Franz Schilling and Dr. Geoffrey Topping for help with MR sequence optimization, Olga Seelbach and Marion Mielke for histology, Anja Wolf for production of AAV9 vectors, Dr. Ritu Mishra for FACS sorting, Dr. Andreas Eichinger for help with the generation of the structural model, Prof. Bernhard Küster and Prof. Gil Westmeyer for cell line (all from TU Munich) and Lara Bontadelli (Novartis) for qPCR and ddPCR analysis. + +<|ref|>text<|/ref|><|det|>[[42, 342, 933, 410]]<|/det|> +This study was funded by the National Centre for the 3Rs via CRACK IT Challenge 32 (NC/C01905/1 & NC/C019202/1) and SFB824 of the German research foundation (projects A8, Z1 & Z2). Anticalin® is a registered trademark of Pieris Pharmaceuticals GmbH. + +<|ref|>text<|/ref|><|det|>[[44, 428, 227, 448]]<|/det|> +Author Contributions + +<|ref|>text<|/ref|><|det|>[[42, 465, 952, 510]]<|/det|> +GlaxoSmithKline and Novartis supported the project with in- kind contributions in the frame of the CRACK IT program of the NC3Rs. + +<|ref|>text<|/ref|><|det|>[[44, 527, 223, 547]]<|/det|> +Competing interests + +<|ref|>text<|/ref|><|det|>[[42, 563, 950, 608]]<|/det|> +V.M., K.F., A.Sk., and W.W. have applied for intellectual property rights on Anticalins- based reporter genes (WO 20022/101492 A1). A.Sk. is a co- founder and shareholder of Pieris Pharmaceuticals. + +<|ref|>sub_title<|/ref|><|det|>[[44, 631, 193, 656]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[55, 672, 940, 940]]<|/det|> +1. Kingwell, K. CAR T therapies drive into new terrain. Nat Rev Drug Discov 16, 301-304 (2017). +2. Mendell, J.R. et al. Current Clinical Applications of In Vivo Gene Therapy with AAVs. Mol Ther 29, 464-488 (2021). +3. Minn, I., Rowe, S.P. & Pomper, M.G. Enhancing CAR T-cell therapy through cellular imaging and radiotherapy. Lancet Oncol 20, e443-e451 (2019). +4. Krebs, S., Dacek, M.M., Carter, L.M., Scheinberg, D.A. & Larson, S.M. CAR Chase: Where Do Engineered Cells Go in Humans? Front Oncol 10, 577773 (2020). +5. Iafrate, M. & Fruhwirth, G.O. How Non-invasive in vivo Cell Tracking Supports the Development and Translation of Cancer Immunotherapies. Front Physiol 11, 154 (2020). +6. Ashmore-Harris, C., Iafrate, M., Saleem, A. & Fruhwirth, G.O. Non-invasive Reporter Gene Imaging of Cell Therapies, including T Cells and Stem Cells. Mol Ther 28, 1392-1416 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 44, 945, 952]]<|/det|> +7. Serganova, I. & Blasberg, R.G. Molecular Imaging with Reporter Genes: Has Its Promise Been Delivered? J Nucl Med 60, 1665-1681 (2019). +8. Volpe, A., Kurtys, E. & Fruhwirth, G.O. Cousins at work: How combining medical with optical imaging enhances in vivo cell tracking. Int J Biochem Cell Biol 102, 40-50 (2018). +9. Cherry, S.R. et al. Total-Body PET: Maximizing Sensitivity to Create New Opportunities for Clinical Research and Patient Care. J Nucl Med 59, 3-12 (2018). +10. Tjuvajev, J.G. et al. Imaging the expression of transfected genes in vivo. Cancer Res 55, 6126-6132 (1995). +11. Yaghoubi, S.S. et al. Noninvasive detection of therapeutic cytolytic T cells with \(^{18}\mathrm{F}\)-FHBG PET in a patient with glioma. Nat Clin Pract Oncol 6, 53-58 (2009). +12. Merron, A. et al. SPECT/CT imaging of oncolytic adenovirus propagation in tumours in vivo using the Na/I symporter as a reporter gene. Gene Ther 14, 1731-1738 (2007). +13. Vedvyas, Y. et al. Longitudinal PET imaging demonstrates biphasic CAR T cell responses in survivors. JCI Insight 1, e90064 (2016). +14. Castanares, M.A. et al. Evaluation of prostate-specific membrane antigen as an imaging reporter. J Nucl Med 55, 805-811 (2014). +15. Minn, I. et al. Imaging CAR T cell therapy with PSMA-targeted positron emission tomography. Sci Adv 5, eaaw5096 (2019). +16. Kim, H.J., Eichinger, A. & Skerra, A. High-affinity recognition of lanthanide(III) chelate complexes by a reprogrammed human lipocalin 2. J Am Chem Soc 131, 3565-3576 (2009). +17. Eggenstein, E., Eichinger, A., Kim, H.J. & Skerra, A. Structure-guided engineering of Anticilins with improved binding behavior and biochemical characteristics for application in radio-immuno imaging and/or therapy. J Struct Biol (2013). +18. Deuschle, F.C., Ilyukhina, E. & Skerra, A. Anticalin(R) proteins: from bench to bedside. Expert Opin Biol Ther 21, 509-518 (2021). +19. Schiefner, A. & Skerra, A. The menagerie of human lipocalins: a natural protein scaffold for molecular recognition of physiological compounds. Acc Chem Res 48, 976-985 (2015). +20. Dunn, C., O'Dowd, A. & Randall, R.E. Fine mapping of the binding sites of monoclonal antibodies raised against the Pk tag. J Immunol Methods 224, 141-150 (1999). +21. Bajar, B.T. et al. Improving brightness and photostability of green and red fluorescent proteins for live cell imaging and FRET reporting. Sci Rep 6, 20889 (2016). +22. Shcherbakova, D.M., Cox Cammer, N., Huisman, T.M., Verkhusha, V.V. & Hodgson, L. Direct multiplex imaging and optogenetics of Rho GTPases enabled by near-infrared FRET. Nat Chem Biol 14, 591-600 (2018). +23. Eggenstein, E., Eichinger, A., Kim, H.J. & Skerra, A. Structure-guided engineering of Anticilins with improved binding behavior and biochemical characteristics for application in radio-immuno imaging and/or therapy. J Struct Biol 185, 203-214 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 42, 950, 95]]<|/det|> +24. Barkovskiy, M., Ilyukhina, E., Dauner, M., Eichinger, A. & Skerra, A. An engineered lipocalin that tightly complexes the plant poison colchicine for use as antidote and in bioanalytical applications. Biol Chem 400, 351–366 (2019). + +<|ref|>text<|/ref|><|det|>[[46, 118, 850, 161]]<|/det|> +25. Krebs, S. et al. Antibody with infinite affinity for in vivo tracking of genetically engineered lymphocytes. J Nucl Med 59, 1894–1900 (2018). + +<|ref|>text<|/ref|><|det|>[[46, 166, 950, 211]]<|/det|> +26. Dacek, M.M. et al. Engineered Cells as a Test Platform for Radiohaptens in Pretargeted Imaging and Radioimmunotherapy Applications. Bioconjug Chem 32, 649–654 (2021). + +<|ref|>text<|/ref|><|det|>[[46, 216, 915, 260]]<|/det|> +27. Sutermaster, B.A. & Darling, E.M. Considerations for high-yield, high-throughput cell enrichment: fluorescence versus magnetic sorting. Sci Rep 9, 227 (2019). + +<|ref|>text<|/ref|><|det|>[[46, 264, 935, 309]]<|/det|> +28. Sanchez-Crespo, A. Comparison of Gallium-68 and Fluorine-18 imaging characteristics in positron emission tomography. Appl Radiat Isot 76, 55–62 (2013). + +<|ref|>text<|/ref|><|det|>[[46, 313, 930, 358]]<|/det|> +29. Richarz, R. et al. Neither azeotropic drying, nor base nor other additives: a minimalist approach to \(^{18}\mathrm{F}\) labeling. Org Biomol Chem 12, 8094–8099 (2014). + +<|ref|>text<|/ref|><|det|>[[46, 365, 951, 410]]<|/det|> +30. Cardinale, J. et al. Development of PSMA-1007-Related Series of \(^{18}\mathrm{F}\) -Labeled Glu-Ureido-Type PSMA Inhibitors. J Med Chem 63, 10897–10907 (2020). + +<|ref|>text<|/ref|><|det|>[[46, 415, 914, 460]]<|/det|> +31. Sommermeyer, D. et al. Fully human CD19-specific chimeric antigen receptors for T-cell therapy. Leukemia 31, 2191–2199 (2017). + +<|ref|>text<|/ref|><|det|>[[46, 465, 909, 510]]<|/det|> +32. Wang, X. et al. A transgene-encoded cell surface polypeptide for selection, in vivo tracking, and ablation of engineered cells. Blood 118, 1255–1263 (2011). + +<|ref|>text<|/ref|><|det|>[[46, 515, 861, 560]]<|/det|> +33. Paszkiewicz, P.J. et al. Targeted antibody-mediated depletion of murine CD19 CAR T cells permanently reverses B cell aplasia. J Clin Invest 126, 4262–4272 (2016). + +<|ref|>text<|/ref|><|det|>[[46, 564, 950, 610]]<|/det|> +34. Lee, J.C. et al. In vivo inhibition of human CD19-targeted effector T cells by natural T regulatory cells in a xenotransplant murine model of B cell malignancy. Cancer Res 71, 2871–2881 (2011). + +<|ref|>text<|/ref|><|det|>[[46, 614, 950, 659]]<|/det|> +35. Wang, D., Tai, P.W.L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat Rev Drug Discov 18, 358–378 (2019). + +<|ref|>text<|/ref|><|det|>[[46, 664, 940, 709]]<|/det|> +36. Zincarelli, C., Soltys, S., Rengo, G. & Rabinowitz, J.E. Analysis of AAV serotypes 1–9 mediated gene expression and tropism in mice after systemic injection. Mol Ther 16, 1073–1080 (2008). + +<|ref|>text<|/ref|><|det|>[[46, 713, 951, 758]]<|/det|> +37. Brown, L.V., Gaffney, E.A., Ager, A., Wagg, J. & Coles, M.C. Quantifying the limits of CAR T-cell delivery in mice and men. J R Soc Interface 18, 20201013 (2021). + +<|ref|>text<|/ref|><|det|>[[46, 762, 933, 807]]<|/det|> +38. Gjedde, S.B. & Gjedde, A. Organ Blood-Flow Rates and Cardiac-Output of the Balb-C Mouse. Comp Biochem Phys A 67, 671–674 (1980). + +<|ref|>text<|/ref|><|det|>[[46, 811, 849, 856]]<|/det|> +39. Amini, L. et al. Preparing for CAR T cell therapy: patient selection, bridging therapies and lymphodepletion. Nat Rev Clin Oncol (2022). + +<|ref|>text<|/ref|><|det|>[[46, 860, 936, 905]]<|/det|> +40. Weber, W.A., Varasteh, Z., Fritschle, K. & Morath, V. A Theranostic Approach for CAR-T Cell Therapy. Clin Cancer Res 28, 5241–5243 (2022). + +<|ref|>text<|/ref|><|det|>[[46, 909, 950, 954]]<|/det|> +41. Yang, J. & Reth, M. The dissociation activation model of B cell antigen receptor triggering. FEBS Lett 584, 4872–4877 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 44, 940, 536]]<|/det|> +42. Wang, M. et al. Quantifying CD4 receptor protein in two human CD4 + lymphocyte preparations for quantitative flow cytometry. Clin Proteomics 11, 43 (2014). +43. Lee, C.C., Perchiacca, J.M. & Tessier, P.M. Toward aggregation-resistant antibodies by design. Trends Biotechnol 31, 612–620 (2013). +44. Ravera, S., Reyna-Neyra, A., Ferrandino, G., Amzel, L.M. & Carrasco, N. The Sodium/Iodide Symporter (NIS): Molecular Physiology and Preclinical and Clinical Applications. Annu Rev Physiol 79, 261–289 (2017). +45. Stotz, S., Bowden, G.D., Cotton, J.M., Pichler, B.J. & Maurer, A. Covalent \(^{18}\mathrm{F}\) -Radiotracers for SNAPtag: A New Toolbox for Reporter Gene Imaging. Pharmaceuticals (Basel) 14 (2021). +46. Sellmyer, M.A. et al. Imaging CAR T Cell Trafficking with eDHFR as a PET Reporter Gene. Mol Ther 28, 42–51 (2020). +47. Bojkowska, K. et al. Measuring in vivo protein half-life. Chem Biol 18, 805–815 (2011). +48. van der Windt, G.J. & Pearce, E.L. Metabolic switching and fuel choice during T-cell differentiation and memory development. Immunol Rev 249, 27–42 (2012). +49. Volpe, A., Pillarsetty, N.V.K., Lewis, J.S. & Ponomarev, V. Applications of nuclear-based imaging in gene and cell therapy: probe considerations. Mol Ther Oncolytics 20, 447–458 (2021). +50. Shao, F. et al. Radionuclide-based molecular imaging allows CAR-T cellular visualization and therapeutic monitoring. Theranostics 11, 6800–6817 (2021). +51. Moroz, M.A. et al. Comparative Analysis of T Cell Imaging with Human Nuclear Reporter Genes. J Nucl Med 56, 1055–1060 (2015). + +<|ref|>sub_title<|/ref|><|det|>[[44, 556, 143, 581]]<|/det|> +## Figures + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[40, 45, 888, 792]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[43, 802, 115, 820]]<|/det|> +
Figure 1
+ +<|ref|>sub_title<|/ref|><|det|>[[43, 843, 523, 863]]<|/det|> +## Design and characterization of DTPA-R reporter protein + +<|ref|>text<|/ref|><|det|>[[42, 882, 916, 949]]<|/det|> +Schematic representation of the DTPA- R(ceptor) reporter gene system composed of the reporter protein DTPA- R and the cognate PET reporter probe [18F]F- DTPA- metal. (A) Molecular model (PyMol) based on the crystal structure of the Anticalin/CHX- A"- DTPA- Y complex (PDB ID: 4IAX) and the NMR + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 44, 950, 303]]<|/det|> +structure of the CD3 transmembrane domain (PDB ID: 2KLU). (B) Design of the coding region for the Anticalin- based reporter protein with a promoter, the Lcn2 signal peptide (sp), the mature Anticalin, the V5- tag, the CD4 transmembrane domain and, optionally, a fluorescent protein. (C) Chemical structure of the PET radioligand \([^{18}\mathrm{F}]\mathrm{F}\) - Nic- Glu2- PEG4- CHX- A"- DTPA- metal, dubbed \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA and (D) the ligand moiety of \([^{18}\mathrm{F}]\mathrm{F}\) - colchicine. (E) Fluorescence microscopy of PC3DTPA- R and PC3Colchi- R cells stained with Hoechst 33342 (cell nucleus) and an AlexaFluor488- conjugated anti- V5- tag antibody (reporter protein). (F) Quantification of reporter protein surface densities by flow cytometry with MESF beads (four individual experiments, t- test). (G) Analysis of transduced Jurkat cell lines for their proliferation kinetics using the CFSE assay, error bars indicate distribution in one experiment. (H) Separation of transgenic JurkatDTPA- R cells from not transduced Jurkat cells using the anti- V5- tag antibody for magnetic- activated cell sorting (MACS). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 52, 480, 270]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[45, 732, 117, 752]]<|/det|> +
Figure 2
+ +<|ref|>image<|/ref|><|det|>[[515, 55, 933, 707]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[44, 772, 595, 794]]<|/det|> +## Preparation and characterization of the \([^{18}\mathrm{F}]\) F-DTPA radioligand + +<|ref|>text<|/ref|><|det|>[[41, 811, 944, 960]]<|/det|> +(A) Binding of \(\mathrm{NH_2 - CHX - A^{\prime\prime} - DTPA^{\prime\prime}90Y}\) competed by \(\mathrm{NH_2 - CHX - A^{\prime\prime} - DTPA - 89Y}\) to Jurkat cells expressing DTPA-R-mRuby3. (B) Radiosynthesis procedure including radiofluorination, preparative HPLC, and charging of the CHX-A"-DTPA chelator with \(\mathrm{Na^{+}Tb^{III}}\) . (C) Quality control by analytical HPLC of the radiosynthesis product before and after preparative HPLC purification. (D) Binding assay of \([^{18}\mathrm{F}]\) F-DTPA to PC3DTPA-R or PC3Colchi-R cells (statistical analysis: t-test). (E) Analysis of the internalization of \([^{18}\mathrm{F}]\) F-DTPA upon binding to PC3 cells by analyzing the Accutase-cleavable and the non-cleavable + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 900, 87]]<|/det|> +fraction. Accutase efficiently cleaves the DTPA- R ectodomain and, thus, allows the identification of bound radioligand that is accessible on the cell surface (statistical analysis: t- test). + +<|ref|>image<|/ref|><|det|>[[60, 92, 930, 790]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 814, 117, 833]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 856, 485, 878]]<|/det|> +Pharmacokinetic and stability of [18F]- DTPA in vivo + +<|ref|>text<|/ref|><|det|>[[42, 898, 950, 942]]<|/det|> +(A- D) Dynamic PET scan of CD1- nude mice carrying subcutaneous PC3DTPA- R (right shoulder) and PC3Colchi- R (left shoulder) xenograft tumors. (A) Maximum Intensity Projections (MIPs) of a dynamic PET + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 955, 283]]<|/det|> +scan between 1 and 90 min p.i. of 11.4 MBq [18F]F- DTPA and (B) an axial PET plane through both tumors at \(0 - 25\%\) ID/g and \(0 - 2.5\%\) ID/g, showing selective accumulation in the PC3DTPA- R tumor but no elevated signal in the PC3Colchi- R tumor. (C) Linear and (D) logarithmic representation of the time- activity curve derived from the dynamic PET scan. VOI quantification using 10 pixel spheres revealed increasing signal- to- background ratios over time. (E) Ex vivo biodistribution analysis at \(t = 90\) min after [18F]F- DTPA i.v. injection into male and female C57BL/6 mice. Please note, mice were awake between injection and \(t = 90\) min, and not under anesthesia as for the dynamic PET scan, causing different pharmacokinetic profiles. No significant differences were found (t- test). (F) Separation of the intact [18F]F- DTPA radioligand, its hydrolysis fragments, and urine samples collected from mice injected beforehand with [18F]F- DTPA by size- exclusion chromatography. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[40, 40, 952, 770]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[43, 787, 117, 805]]<|/det|> +
Figure 4
+ +<|ref|>sub_title<|/ref|><|det|>[[43, 827, 618, 848]]<|/det|> +## Longitudinal PET imaging study of DTPA-R expressing CAR-T cells + +<|ref|>text<|/ref|><|det|>[[42, 864, 940, 957]]<|/det|> +(A) The design of the longitudinal CAR-T cell study involved NSG mice injected i.v. with Raji tumor cells \((0.5 \times 10^6)\) and, after 7 days, i.v. injection of \(2 \times 10^6\) DTPA-R or EGFRt expressing anti-CD19 CAR-T cells. PET scans were recorded during treatment on days 4, 8, 15, 22 and 30. MIP of an exemplary (B) αCD19-CAR-TDTPA-R and (C) αCD19-CAR-TEGFRt mouse each is depicted for the respective days. (D) The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[36, 45, 951, 319]]<|/det|> +cumulative specific signal from every \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA scan was quantified for the upper body (above the gall bladder) and the hind legs. This quantification allowed monitoring of global CAR- T cell expansion and/or decline over time. FACS analysis of total cell numbers of (E) Raji tumor cells and (F) CAR- T cells in different extremities, spleen, and blood. (G- J) Quantification of local CAR- T infiltrations over the course of the treatment. (G) A sagittal PET/MR scan on day 22 revealed different bone marrow infiltrations. (H) Axial PET/MR enabled the quantitative assessment of local CAR- T cell infiltration over time. (I) Longitudinal VOI quantification allowed the differentiation between tumor lesions experiencing an intermediate CAR- T expansion and subsequent decline (grey, blue, and orange arrowheads in G/H) and lesions with constant CAR- T infiltration and/or expansion (green arrowhead in G/H). (J) Immunohistochemistry of sagittal spine sections containing (green, blue & orange) lesions with infiltrating CAR- T cells (V5- tag, brown). (K) Overlay of PET signals on the V5- IHC. (L) Double staining of a consecutive tissue section for CAR- T cells (V5- tag, red) and Raji lymphoma cells (CD19, brown). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 41, 940, 690]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 712, 118, 731]]<|/det|> +
Figure 5
+ +<|ref|>sub_title<|/ref|><|det|>[[44, 752, 618, 773]]<|/det|> +## Quantification and detection limit for DTPA-R labeled lymphocytes + +<|ref|>text<|/ref|><|det|>[[41, 790, 951, 956]]<|/det|> +(A- B) Flow cytometry quantification of CAR- \(\mathsf{TDPA - R}\) cells in the bone marrow of mice after PET imaging. (A) Cells within hollow bones were harvested for ex vivo analysis. (B) Correlation of CAR- T cell numbers found by flow cytometry with the PET signal obtained from hollow bones (with exemplary PET/MR cross- section images). (C) Individual lesions in the bone marrow were quantified, and CAR- \(\mathsf{TDPA - R}\) cell numbers were calculated by interpolation using the regression line from (B). (D) Maximum signal capacity was assessed by staining Jurkat \(\mathsf{DTPA - R}\) cells in vitro with \([^{18}\mathsf{F}]\mathsf{F}\) - DTPA. PET phantom study: Jurkat \(\mathsf{DTPA - R}\) or Jurkat \(\mathsf{Colchi - R}\) cells were incubated with \([^{18}\mathsf{F}]\mathsf{F}\) - DTPA, washed three times, counted, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 940, 163]]<|/det|> +10 μl of the suspension was transferred into 200 μl PCR tubes. PET signals were recorded for 60 min. (E, F) Spot assay: JurkatDTPA-R cells (E) or aCD19-CAR-DTPA-R cells (F) were injected subcutaneously into the back of mice. Five samples of a 1:2 dilution series of the DTPA-R expressing cells and of a control cell line were used. 90 min after i.v. [18F]F-DTPA injection, PET scans were recorded for 20 min. %ID of the spots was quantified by V0ls and plotted against the number of injected cells. + +<|ref|>image<|/ref|><|det|>[[40, 160, 875, 911]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 920, 117, 938]]<|/det|> +
Figure 6
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 45, 450, 65]]<|/det|> +## PET imaging of AAV9 gene therapy via DTPA-R + +<|ref|>text<|/ref|><|det|>[[40, 82, 951, 408]]<|/det|> +PET imaging of AAV9 gene therapy via DTPA- R C57BL/6 or CD1- nude mice were injected i.v. with (A) AAV9/2 viral vectors encoding the DTPA- R. MIPs are shown for a longitudinal study of (B) immunocompetent C57BL/6 mice and (C) immunocompromised CD1- nude mice at t=14, 21, and 28 days after injection of \(2.5 \times 10^{12}\) vg per mouse. (D) Quantification of PET signals by VOI spheres in selected organs throughout the study (statistical analysis: t- test). (E) MIP of C57BL/6 mice transduced with different titers \((1 \times 10^{11}, 5 \times 10^{11}, 1 \times 10^{12}\) and \(2.5 \times 10^{12}\) vg/mouse) as well as untreated control mice imaged after 7 days. (F) MIP and axial PET/MR overlays allow the identification of anatomical structures accumulating \([^{18}\mathrm{F}]\mathrm{F}\) - DTPA. (G) Correlation of injected AAV9 titer and quantified PET signals for brown adipose tissue (BAT), liver, heart, and adrenal glands (n=3 per titer). (H) The slope of the increase in PET signals was correlated with the perfusion rate (obtained from literature) of the respective organ. (I) Sagittal PET and V5- IHC of the adrenal gland with quantification of positive cells in the zona fasciculata of the adrenal gland cortex. Correlations of (J) positive cell count and (K) PET signal with the injected viral vector dose, as well as (L) with each other. Correlation of detected vector genomes and PET signal (M). (N) Multivariate analysis correlating the PET signal (actual Y) with AAV9 dose, mRNA and viral genomes (predicted Y) in the liver. + +<|ref|>sub_title<|/ref|><|det|>[[44, 429, 312, 456]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 480, 767, 500]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 517, 666, 538]]<|/det|> +Supplementary DTPARPETreportergeneimagingNatBiotechnol.docx + +<--- Page Split ---> diff --git a/preprint/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7.mmd b/preprint/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7.mmd new file mode 100644 index 0000000000000000000000000000000000000000..3c445041d84d6095c2340f0296537e4617a10b57 --- /dev/null +++ b/preprint/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7/preprint__114b2c7af2a7edcc27507d19510f53d2b72a74abd673eef2f33846e234ceb8b7.mmd @@ -0,0 +1,178 @@ + +# Microscopic mechanism for fluctuating pair density wave + +Chandan Setty ( settychandan@gmail.com ) + +Rice University + +Laura Fanfarillo + +SISSA International School for Advanced Studies https://orcid.org/0000- 0002- 6452- 8520 + +Peter Hirschfeld + +University of Florida + +## Article + +Keywords: superconductors, fluctuating pair density wave, pairing phases + +Posted Date: November 18th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 1021881/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Microscopic mechanism for fluctuating pair density wave + +Chandan Setty \(^{\oplus ,\dagger ,1,2}\) Laura Fanfarillo \(^{\oplus ,\diamond ,1,3}\) and P. J. Hirschfeld \(^{\otimes 1}\) \(^{1}\) Department of Physics, University of Florida, Gainesville, Florida, USA \(^{2}\) Department of Physics and Astronomy, Rice Center for Quantum Materials, Rice University, Houston, Texas 77005, USA \(^{3}\) Scuola Internazionale Superiore di Studi Avanzati (SISSA), Via Bonomea 265, 34136 Trieste, Italy + +In weakly coupled BCS superconductors, only electrons within a tiny energy window around the Fermi energy, \(E_{F}\) , form Cooper pairs. This may not be the case in strong coupling superconductors such as cuprates, FeSe, SrTiO \(_3\) or cold atom condensates where the pairing scale, \(E_{B}\) , becomes comparable or even larger than \(E_{F}\) . In cuprates, for example, a plausible candidate for the pseudogap state at low doping is a fluctuating pair density wave, but no microscopic model has yet been found which supports such a state. In this work, we write an analytically solvable model to examine pairing phases in the strongly coupled regime and in the presence of anisotropic interactions. Already for moderate coupling we find an unusual finite temperature phase, below an instability temperature \(T_{i}\) , where local pair correlations have non- zero center- of- mass momentum but lack long- range order. At low temperature, this fluctuating pair density wave can condense either to a uniform \(d\) - wave superconductor or the widely postulated pair- density wave phase depending on the interaction strength. Our minimal model offers a unified microscopic framework to understand the emergence of both fluctuating and long range pair density waves in realistic systems. + +Spatially uniform superconducting (SC) order formed from Cooper pairs with zero center- of- mass momentum is the energetically favored ground state in the conventional theory of Bardeen, Cooper and Schrieffer (BCS) [1]. Equivalently, the SC instability is signaled by a divergence in the static pair- fluctuation propagator, \(L(\mathbf{q}, \Omega = 0)\) , at \(\mathbf{q} = 0\) once the pair instability temperature, \(T_{i}\) , is achieved [2]. On the other hand, a non- uniform order with non- zero center- of- mass momentum Cooper pair can occur when the divergence of the pair fluctuation propagator is shifted to non- zero \(\mathbf{q}\) . First proposed by Fulde and Farrell (FF) [3] and independently by Larkin and Ovchinnikov (LO) [4], these solutions are stabilized in the presence of explicit time- reversal symmetry breaking from an external magnetic field. A modulated order parameter can also be realized in the presence of time- reversal symmetry where the spatial average of the gap vanishes. Termed pair- density waves (PDWs), these states are posited to exist in a variety of systems, including high- temperature cuprate superconductors (for a review, see Ref. [5] and references therein). + +While PDWs have been subject to much theoretical [7- 20] and numerical [21- 27] interest, a clear- cut analytically solvable model describing their origin from microscopic ingredients is lacking. From the experimental point of view, the interest for modulated pairing phases has been triggered by increasing experimental evidence for short- ranged PDW order in the underdoped region of the phase diagram of cuprates [27- 38]. In particular, [33] reported the first clear observation via scanning tunneling spectroscopy of a vortex- induced PDW in \(\mathrm{Bi}_{2}\mathrm{Sr}_{2}\mathrm{CaCu}_{2}\mathrm{O}_{8}\) at low temperature. More recent STM experiments provide further evidence in favor of a short- range PDW coexisting with the \(d\) - wave superconductivity in the SC phase and evolving into a PDW state in the pseudogap region [27, 38]. This phase is characterized by a gap at finite temperatures but lacks long- range order, and can be characterized as a "fluctuating pair density wave", locally pinned by disorder. Such a state also provides an explanation for many other experimental signatures of the cuprates, including the existence of vestigial charge density wave order arising from partial melting of a PDW [5, 16, 39]. However, there is currently no microscopic model supporting this picture. Hence it is urgent to seek a unified framework that subsumes both fluctuating and long- range ordered PDW phases under a single paradigm by providing a concrete description of their origin. + +ized by a gap at finite temperatures but lacks long- range order, and can be characterized as a "fluctuating pair density wave", locally pinned by disorder. Such a state also provides an explanation for many other experimental signatures of the cuprates, including the existence of vestigial charge density wave order arising from partial melting of a PDW [5, 16, 39]. However, there is currently no microscopic model supporting this picture. Hence it is urgent to seek a unified framework that subsumes both fluctuating and long- range ordered PDW phases under a single paradigm by providing a concrete description of their origin. + +In this Article, we show that a Fermi liquid subjected to a finite anisotropic interaction is unstable toward a modulated SC phase in the strong coupling limit. Whether this phase is a "fluctuating" PDW (FPDW) or long- range order PDW is determined by temperature as well as the coupling strength defined by the ratio \(\alpha = E_{B} / E_{F}\) , with \(E_{F}\) the Fermi energy and \(E_{B}\) the bound state energy for pair formation. + +Our strategy is to solve the self- consistent gap equation for the homogeneous \(d\) - wave superconductor and analyze the momentum dependence of the SC fluctuations. The expansion of the static pair propagator \(L_{\mathbf{q}}\) in powers of momentum transfer \(\mathbf{q}\) , can reveal, in fact, critical fluctuations of Cooper pairs with finite center- of- mass momentum, that makes the homogeneous solution unstable towards a modulated SC phase. This is indeed what we find already at intermediate coupling \(\alpha \sim 0.7\) . The emergence of such a state is linked to the existence of fluctuating terms that lower the momentum rigidity of the Cooper pairs. These terms directly follow from the anisotropy of the pairing interaction that affects the momentum dependence of the pairing susceptibility already in the normal phase. + +Our results are summarized in the phase diagram, + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
FIG. 1: Phase diagram for \(\alpha = E_{B} / E_{F}\) vs \(T\) . The instability temperature for the \(d\) -wave superconductor, \(T_{i}\) , defines the transition from a Fermi liquid (light blue) to the SC state. At weak coupling the pairing state is a homogeneous \(d\) -wave superconductor (gold). Increasing \(\alpha\) the system develops critical fluctuations at finite momentum and the \(d\) -wave SC state becomes unstable toward a non-homogeneous SC state (pink and purple regions). \(T^{*}\) is the temperature at which the momentum rigidity parameter \(c_{2}\) vanishes. The fluctuating PDW (pink) condenses below a coherence temperature \(T_{c}\) into a long-range ordered state that can be an homogeneous \(d\) -wave SC state (gold) or a PDW (purple) depending on the coupling strength, schematically represented by a solid line. \(T_{c}\) coincides with \(T_{i}\) at weak coupling while at strong coupling it is expected that \(T_{c} < T_{i}\) [6]. Note that the actual instability temperature of the FPDW, \(\bar{T}_{i}\) , is somewhat higher than \(T_{i}\) (see Supplementary Material). Temperatures are renormalized by the energy range of the pairing, \(\Lambda\) that is the largest energy scale of our model.
+ +Fig. 1. \(T_{i}\) is the instability temperature of the homogeneous \(d\) - wave state. The analysis of fluctuations allows us to define two different regimes. At weak coupling, \(\alpha \ll 1\) , the uniform \(d\) - wave paired state is the ground state; at larger \(\alpha\) (strong coupling), SC fluctuations at finite momentum lead to two modulated pairing phases – the \(T = 0\) PDW ground state and a higher temperature FPDW phase that condenses into a PDW ordered phase below a coherence temperature \((T_{c})\) . As expected in the BCS limit, the instability temperature \(T_{i}\) and the coherence temperature \(T_{c}\) coincide at weak coupling, while they decouple in the strong coupling regime where we find well formed pairs with no coherence. We do not perform here any calculation of the coherence temperature inside the modulated phase, however in analogy with results obtained for homogeneous \(s\) - wave superconductors in the strong coupling limit [6] we anticipate \(T_{c} < T_{i}\) for \(\alpha > 1\) . The FPDW is found for temperature \(T_{c} < T < T_{i}\) and it is characterized by pairs with finite momentum with no coherence. At \(T = 0\) , the ground state can be either the uniform \(d\) - wave solution or the long- range PDW depending on the value of \(\alpha\) . Hence our model captures two + +key experimentally postulated modulated Cooper phases – a FPDW and a long- range PDW – in a single unified scheme. + +The mechanism we present in this paper predicts spatially modulated pairing phases for \(\alpha = E_{B} / E_{F} > 1\) , i.e. in strongly coupled electronic systems, with anisotropic interactions. Examples of low- density electronic materials include the Fe- based superconductor FeSe where quantum oscillations [40] as well as transport and scanning tunneling spectroscopy [41] show that both the electron and hole pockets are tiny with Fermi energies comparable or even smaller than the SC gap and for which we find several proposals of BCS- BEC cross- over physics in the literature [42- 44]. Other "mixed- band" superconductors such as O vacancy- or Nb- doped SrTiO₃ have one partially filled band with a large Fermi surface while the Fermi level intersects the other at or close to the band bottom [45]. Even if these materials typically have more than one band close to or crossing the Fermi level, the results from our minimal model may eventually provide a suitable starting- point for the analysis of possible instabilities towards modulated pairing states in dilute multiorbital superconductors. Our results may also be relevant to the recent observation of superconductivity in twisted- bilayer graphene [46] where interactions can be large compared to the bandwidth leading to large inter- particle distances [47] and hence possible strongly- coupled Cooper pairing. + +The modulated phases we propose in this work, that include both the long- range ordered PDW as well as the FPDW at finite temperature, are distinct from earlier proposals in literature. Loder and coworkers [11], considered similar models characterized by nearest neighbor attractive interaction with \(d\) - wave symmetry and found Cooper pairing with finite center- of- mass momentum above a critical interaction strength. In Refs. [19, 20], a modulated superconducting state is found in models which have correlated pair- hopping interactions. Other models that admit modulated SC ground states were proposed in the context of cold atoms [10] where local interactions were considered in systems with multiple bands. Those references focused on the analysis of the long- range ordered state (mainly at zero temperature) without exploring the FPDW phase. The key contribution of our work is it provides an analytically tractable model where both fluctuating and long- range ordered PDWs can be explained under a single unified framework. + +## MODEL + +Let's consider a single band SC system. The kinetic part of the Hamiltonian reads \(H_{0} = \sum_{\mathbf{k}\sigma} \xi_{\mathbf{k}} c_{\mathbf{k}\sigma} c_{\mathbf{k}\sigma}\) , where \(\xi_{\mathbf{k}} = \epsilon_{\mathbf{k}} - \mu\) , \(\mu\) is the chemical potential, \(\epsilon_{\mathbf{k}} = \mathbf{k}^{2} / 2m\) the parabolic dispersion and we further assume + +<--- Page Split ---> + +\(2m = 1\) . The pairing interaction is given by + +\[H_{I} = -g\sum_{\mathbf{q}}\theta_{\mathbf{q}}^{\dagger}\theta_{\mathbf{q}}, \quad (1)\] + +\(g\) is the constant SC coupling and \(\theta_{\mathbf{q}}\) is defined as + +\[\theta_{\mathbf{q}} = \sum_{\mathbf{k}}f_{\mathbf{k},\mathbf{q}}c_{-\mathbf{k} + \frac{\mathbf{q}}{2},\downarrow}c_{\mathbf{k} + \frac{\mathbf{q}}{2},\uparrow}. \quad (2)\] + +where \(f_{\mathbf{k},\mathbf{q}} = (h_{\mathbf{k} - \mathbf{q} / 2} + h_{\mathbf{k} + \mathbf{q} / 2}) / 2\) is a form factor. In this work \(h_{\mathbf{k}}\) can be any anisotropic form factor; we consider, e.g. \(h_{\mathbf{k}} = \left(k_{x}^{2} - k_{y}^{2}\right) / \Lambda\) with \(d\) - wave form. Our results do not depend qualitatively on the exact form of the anisotropy, provided it is strong enough, but they are distinct from the conventional \(s\) - wave case \(f_{\mathbf{k},\mathbf{q}} = 1\) . Here \(\Lambda\) is the pairing energy scale. + +We use the standard Hubbard- Stratonovich transformation to decouple the interaction term, Eq.1 and to derive the effective action in term of the bosonic pairing field \(\Delta\) (for a detailed derivation see Supplemental Material). + +In standard BCS superconductors, the mean- field value of the pairing field is defined by minimizing the action with respect to the homogeneous \(\mathbf{q} = 0\) value of \(\Delta\) and then solving this equation together with the one for the chemical potential. To study fluctuations of the pairing field around the mean- field value, we instead analyze the gaussian action obtained by retaining up to the second order in the fluctuating field with arbitrary momentum \(\mathbf{q}\) given by + +\[S_{G}[\Delta_{\mathbf{q}}] = \sum_{\mathbf{q}}L_{\mathbf{q}}^{-1}\Delta_{\mathbf{q}}^{2}. \quad (3)\] + +The static pairing susceptibility is explicitly given by \(L_{\mathbf{q}}^{- 1} = g^{- 1} + \Pi_{\mathbf{q}}\) , where the particle- particle propagator reads + +\[\Pi_{\mathbf{q}} = \frac{T}{V}\sum_{\mathbf{k}n}\frac{(i\omega_{n} + \xi_{\mathbf{k} + \mathbf{q}})(i\omega_{n} - \xi_{\mathbf{k}}) - f_{\mathbf{k},0}f_{\mathbf{k} + \mathbf{q},0}\Delta^{2}}{(\omega_{n}^{2} + E_{\mathbf{k}}^{2})(\omega_{n}^{2} + E_{\mathbf{k} + \mathbf{q}}^{2})} f_{\mathbf{k},\mathbf{q}}^{2}, \quad (4)\] + +with \(E_{\mathbf{k}}^{2} = \xi_{\mathbf{k}}^{2} + f_{\mathbf{k},0}^{2}\Delta^{2}\) . Here \(T\) is the temperature and \(V\) the volume. Note that since \(2m = 1\) , energies have dimensions of 2- D \(V^{- 1}\) , and \(L_{\mathbf{q}}^{- 1}\) is therefore dimensionless. + +The static susceptibility can be expanded in the hydrodynamic limit as + +\[L_{\mathbf{q}}^{-1} = c_{0} + c_{2}q^{2}. \quad (5)\] + +The instability temperature is defined as the highest temperature at which the susceptibility diverges, i.e. \(c_{0} = g^{- 1} + \Pi_{0}|_{T = T_{i}} = 0\) , as we assume that the minimum of the action, Eq. 3, is associated with the homogeneous order parameter. The coefficient \(c_{2} = (\partial^{2}L_{\mathbf{q}}^{- 1} / \partial q^{2}|_{q = 0}) / 2\) provides instead information about the momentum rigidity of the fluctuating Cooper pairs i.e. the energy needed to + +move the center- of- mass momentum of the Cooper pairs from zero to a finite value. A negative momentum rigidity, \(c_{2}< 0\) , implies that finite momentum fluctuations can lower the energy of the system making the homogeneous SC solution unstable. This means that the highest temperature at which the pairing susceptibility, Eq. 5, diverges is actually associated to a critical mode with finite momentum. + +In what follows we analyze the momentum- dependence of the static susceptibility, Eq. 5, looking for a sign change of the momentum rigidity parameter \(c_{2}\) and using it as a proxy to identify possible spatially modulated SC regions in the phase diagram. It is worth noticing that \(c_{2}\) is directly affected by the momentum properties of the pairing susceptibility i.e. the pairing symmetry. From Eq. 4, it is easy to verify that the anisotropy of the interactions affects the momentum dependence of the propagator not only in the SC phase via the symmetry of the SC order parameter, but also above the instability temperature \(T_{i}\) where \(\Delta = 0\) due to the overall form factor \(f_{\mathbf{k},\mathbf{q}}^{2}\) at the numerator. This reflects in a strong momentum dependence of the contributions to the rigidity parameter depending on the symmetry of the pairing interaction. We discuss below how this affects the development of critical finite- momentum fluctuations. + +## RESULTS AND DISCUSSION + +The mean- field analysis for the homogeneous \(d\) - wave superconductor is shown in Fig.2. In panels (a)- (b) we report the self- consistent numerical mean- field solutions for the pairing function \(\Delta\) and the chemical potential \(\mu\) as a function of temperature \(T\) for three representative cases of the pairing strength \(\alpha = E_{B} / E_{F} = 0.5,1.0,2.0\) , where for simplicity the weak- coupling expression \(E_{B} = \Lambda e^{- 2 / g}\) is used at all \(\alpha\) . In panels (c)- (d) we show the same mean- field results at \(T = T_{i}\) and \(T = 0\) as a function of \(\alpha\) . The change of sign of the chemical potential with increasing coupling strength is well- known from the BCS- BEC crossover problem [48- 52]. In the weak- coupling regime, the pairs are loosely bound and we recover the BCS expression \(\mu \sim E_{F}\) . As the interaction increases, all fermions strongly bind in pairs and \(\mu\) becomes negative and proportional to \(- E_{B}\) . In both the weak and strong coupling limits, the curves are similar to those derived for \(s\) - wave superconductors in [52], showing that the \(d\) - wave symmetry of the pairing interaction does not affect the mean- field results qualitatively. + +We first study the SC fluctuations above the instability temperature by analyze the static pairing susceptibility in the hydrodynamic limit, Eq. 5. The mass term \(c_{0}\) is positive and vanishes as the temperature approaches the instability temperature as expected from a Ginzburg- Landau description of the transition. + +The analysis of the momentum rigidity of the fluctuat + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
FIG. 2: Mean-field results for the spatially homogeneous \(d\) -wave superconductor. (a,b) Self-consistent solutions of the pairing order parameter \(\Delta (T)\) and the chemical potential \(\mu (T)\) for three representative values of \(\alpha\) . Temperatures are normalized to the instability temperature \(T_{i}\) defined as the temperature at which the static pairing susceptibility \(L_{q = 0}\) diverges, while \(\Delta\) and \(\mu\) are scaled with \(\Lambda\) . (c) Instability temperature \(T_{i}\) and chemical potential \(\mu_{i} \equiv \mu (T = T_{i})\) as a function of \(\alpha\) . (d) \(T = 0\) solutions: \(\Delta_{0} \equiv \Delta (T = 0)\) and chemical potential \(\mu_{0} \equiv \mu (T = 0)\) as a function of \(\alpha\) . For comparison we show also the results of the isotropic s-wave case in dashed lines. Computations are performed using \(\Lambda = 11\) , \(E_{F} = 2.2\) in units of \(2m = 1\) .
+ +ing pairs above \(T_{i}\) is shown in Fig. 3. The weak coupling region is characterized by a standard regime of fluctuations with \(c_{2} > 0\) . Here Cooper pairs with zero center- of- mass momentum are stable. Increasing \(\alpha\) , the momentum rigidity for the \(d\) - wave pairing interaction (continuous line) monotonically decreases and becomes negative at intermediate coupling, \(\alpha > 0.7\) , as shown in Fig. 3(a). This means that finite momentum critical fluctuations grow, increasing the coupling strength up to a critical value of the interaction for which the homogeneous SC solution can become unstable toward a modulated phase. Notice that \(c_{2}\) becomes very small and eventually changes sign in the crossover between weak and strong coupling where also the chemical potential changes sign from positive to negative, see inset Fig. 3(a). The result changes qualitatively for the isotropic \(s\) - wave interaction (dashed line) where the rigidity parameter decreases but remains positive even at strong coupling for the set of model parameter of our study [53]. + +To characterize the modulated SC state and check its stability, we expand the static susceptibility to higher order in momentum + +![](images/Figure_3.jpg) + +
FIG. 3: Coefficients of the momentum-expansion of the static susceptibility as a function of the coupling strength \(\alpha = E_{B} / E_{F}\) at \(T = T_{i}\) . (a) The momentum rigidity \(c_{2}(\alpha)\) for \(d\) -wave (solid line) and \(s\) -wave pairing interaction (dashed line). In the anisotropic \(d\) -wave case \(c_{2}\) becomes negative at intermediate coupling, \(\alpha \sim 0.7\) indicating that the homogeneous \(d\) -wave SC is unstable. Inset: \(c_{2}(\mu_{i})\) , the sign change of the momentum rigidity occurs around the same range in which \(\mu_{i}\) turns from positive to negative values. The momentum rigidity for the isotropic \(s\) -wave case remains positive regardless the coupling strength. (b) \(c_{n}(\alpha)\) coefficients, \(n = 2,4,6\) , for the \(d\) -wave pairing. The positive value of \(c_{6}\) allows to recover the stability of the action. The computation of the higher order coefficients allows to define the finite momentum of the critical mode (see Fig. 4) and the relative instability temperature. We use here the same set of parameters of Fig. 2 and plot the results in dimensionless units i.e. \(c_{n} \equiv c_{n} \Lambda^{n / 2}\) .
+ +\[L_{\mathbf{q}}^{-1} = \sum_{n}c_{n}\mathbf{q}^{n},\quad \mathrm{with}\quad c_{n} = \frac{1}{n}\frac{\partial^{n}L_{\mathbf{q}}^{-1}}{\partial\mathbf{q}^{n}}\bigg|_{{\mathbf{q}} = 0} \quad (6)\] + +We report the coefficients of the momentum expansion at \(T_{i}\) in Fig. 3(b). Results are shown as a function of \(\alpha\) for the coupling regime in which \(c_{2} \lesssim 0\) . We need to expand the susceptibility up \(n = 6\) to find \(c_{6} > 0\) , since for our set of model parameters \(c_{4} < 0\) as in the conventional BCS case. + +We analyze the momentum dependence of the static susceptibility at \(T_{i}\) in Fig. 4, where we show the expansion of Eq. 6 up to sixth order for different values of \(\alpha\) . At the instability temperature, \(c_{0} = 0\) by definition and the minimum of the function is determined by the higher order coefficients. At weak coupling, where \(c_{2}\) is large and positive, the minimum of \(L_{\mathbf{q}}^{- 1}\) is located at zero momentum. As the pairing interaction increases \(c_{2}\) becomes small and eventually changes sign at \(\alpha \sim 0.7\) . Here, since \(c_{4} < 0\) , the minimum shifts discontinuously to a finite momentum \(\bar{Q}\) , i.e. by increasing the interactions the modulated phase emerges at \(T_{i}\) via a first order transition from the homogeneous \(d\) - wave SC solution, in analogy with the results found at \(T = 0\) in [11, 19]. The non- zero value of \(\bar{Q}\) at \(\alpha \sim 0.7\) signals the formation of the FPDW state with finite momentum pairing but no long range coherent order. Note that the finite order parameter jump \(\bar{Q}\) is a non- universal quantity and depends on microscopic details of the chosen model, as is a + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
FIG. 4: Momentum dependence of the sixth order expansion of \(L^{-1}\) at \(T_{i}\) (dimensionless units). At weak coupling, \(\alpha = 0.22\) , we find the homogeneous \(d\) -wave SC. The momentum rigidity \(c_{2}\) is large and positive, and the minimum of the inverse of the susceptibility is at \(q = 0\) . At intermediate coupling, \(\alpha \sim 0.7\) , \(c_{2}\) vanishes and the minimum of \(L^{-1}\) appears at a finite \(\bar{Q}\) of order 1. Same set of parameters of Fig. 2
+ +feature of any generic first order transition. The momentum characterizing the modulated phase shifts toward larger values increasing the coupling parameters. In the strong coupling regime, \(\alpha \gg 1\) , the minimum occurs at \(q / \sqrt{\Lambda} \gg 1\) , (not shown), for this range of the interaction the analysis of the momentum characterizing the modulated phase requires the implementation of a non perturbative approach. + +The sign change of the momentum rigidity parameter discussed at \(T = T_{i}\) can be traced down in temperature (dashed line in Fig. 1). At \(T = 0\) the homogeneous \(d\) - wave state becomes unstable, now toward a PDW, for a slightly higher value of the coupling where the chemical potential \(\mu\) also changes sign (see Fig.2d). The stability of the PDW phase requires expanding up to the sixth- order, \(c_{4} < 0\) , \(c_{6} > 0\) as we show in the Supplementary Material. + +The results of our numerical study are summarized in the phase diagram of Fig. 1. We characterized the SC region below \(T_{i}\) by the sign of the momentum rigidity parameter (dashed line). The sign change of the \(c_{2}\) coefficient at strong coupling signals the presence of critical SC fluctuations at finite momentum that make the \(d\) - wave homogeneous state unstable toward either an FPDW or PDW. The pink and purple regions indicate the FPDW and the long- range ordered PDW state at high and low temperatures respectively. We leave for future work the explicit calculation of the coherence temperature below which the FPDW condenses. The color gradient indicates approximately the expected \(T_{c}(\alpha)\) behaviour based on previous analysis of the coherence energy scale for the homogeneous \(s\) - wave SC state [6]. + +Analytical calculations of the momentum rigidity can + +be easily performed within a simplified model in which the chemical potential is used as parameter. Both at \(T_{i}\) and \(T = 0\) , we find qualitatively the same results discussed within the numerical study. In particular, within the analytical calculations sketched in the Supplementary Material, the momentum rigidity parameter follows the chemical potential behaviour, i.e. \(c_{2}(\mu) < 0\) for \(\mu < 0\) . This relation is qualitatively in agreement with the numerical study performed computing self- consistently \(\mu (\alpha)\) , as one can see from the inset of Fig. 3(a). + +The strategy implemented here to investigate how finite momentum fluctuations become critical at strong coupling is based on the analysis of the momentum rigidity parameter. This method presents two main advantages with respect to other theoretical approaches. On the hand, as already discussed, it allows us to explore the finite temperature regime and analyze the FPDW state. On the other hand, it provides a physical understanding of the importance of the anisotropy of the pairing interactions in the development of the modulated phase. As one can see in Eq. 4, the symmetry of the pairing interactions dramatically affects the momentum dependence of the propagator not only in the SC phase, but also in the normal one when \(\Delta = 0\) due to the overall form factor \(f_{\mathbf{k},\mathbf{q}}^{2}\) . This is reflected in a strong momentum dependence of the contribution to the momentum rigidity parameter. In fact, after performing analytically the Matsubara summation, the computation of the \(c_{2}\) coefficient reduces to an integral over the Brillouin zone \(c_{2} = \frac{1}{V} \sum_{\mathbf{k}} I_{2}(\mathbf{k})\) . The expression for \(I_{2}\) is given in the Supplemental Material, but here we show here in Fig. 5 2D maps of \(I_{2}(\mathbf{k})\) for both \(s\) - wave and \(d\) - wave at \(T = 0\) and \(T = T_{i}\) . In the isotropic \(s\) - wave case, the contributions to the momentum rigidity coming from different momenta, \(I_{2}(\mathbf{k})\) , are positive at any \((k_{x}, k_{y})\) . On the other hand, in the \(d\) - wave case the contributions to the momentum rigidity coming from the nodal regions are negative and dominate the overall sign of the \(c_{2}\) coefficient. + +## CONCLUSIONS + +A consistent explanation for the occurrence of both static and fluctuating Cooper pairs with finite momentum in the phase diagram of materials such as cuprates has been a long- standing problem. This is primarily because an identification of the microscopic ingredients driving such exotic pairing has been elusive. The results in this paper point toward a simple and unified framework that naturally promotes both fluctuating and static pair- density wave (FPDW and PDW) phases over their zero momentum counterparts. Fig. 1 summarizes the main conclusions of our work, supported not only by numerical evaluations but also transparent analytical estimates (see Supplemental Material). The two key ingredients resulting in a high temperature FPDW and low tem + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
FIG. 5: \(I_{2}(k_{x},k_{y})\) color maps at \(T = 0\) and \(T = T_{i}\) and \(\alpha = 2.0\) for the \(s\) -wave and \(d\) -wave case. The anisotropy of the interactions affects the momentum dependence of the propagator both in the SC and normal phase. This reflects in a strong momentum dependence of the contribution to the momentum rigidity parameter. For the \(d\) -wave case negative contributions to the rigidity are found both at \(T = 0\) and \(T = T_{i}\) from \(k\) -points close to the nodal region. We use the same set of parameters of Fig. 2 and plot the result in dimensionless units for momenta \(|k_{i}| / \sqrt{\Lambda} < \pi\) , with \(i = x, y\) .
+ +perature PDW phases are a) anisotropic (e.g. \(d\) - wave) pair interactions and b) intermediate to strong coupling ratio of \(\alpha = \frac{E_{B}}{E_{F}}\) , where \(E_{B}\) is the pair binding energy for two electrons on the Fermi surface in the presence of an attractive interaction, and \(E_{F}\) is the Fermi energy. Well below a critical value of \(\alpha \sim 0.7\) , only uniform zero momentum \(d\) - wave pairing is favored. In the approximate range of \(0.7 \lesssim \alpha \lesssim 1.5\) , the FPDW phase, characterized by a negative momentum rigidity \(c_{2}\) and positive \(c_{6}\) (see Fig. 3), is stable over a range of temperatures below the instability temperature \(T_{i}\) . However, in this range of \(\alpha\) a uniform \(d\) - wave pair is still favored at zero temperature. For \(\alpha \gtrsim 1.5\) , the PDW phase is more stable than a uniform solution at \(T = 0\) and a finite momentum pair exists for all temperatures below \(T_{i}\) . The modulation wave vector \(\mathbf{Q}\) of the paired phases is determined by the ratio \(\alpha\) and acquires a jump with increasing \(\alpha\) as in a first order transition (see Fig. 4). + +The FPDW and PDW phases are stabilized by contributions to the fluctuation free energy arising from momenta close to the nodal regions in the Brillouin zone. These contributions, which also should drive strong anisotropy in the phase stiffness near \(T_{i}\) , are suppressed (enhanced) at weak (strong) coupling thus leading to a modulated phase above a critical pairing strength. This + +simplified picture is confirmed from our numerical calculations (Fig. 5). Finally, while our work primarily focuses on the instability temperature \(T_{i}\) in the strong coupling limit, the behavior of the condensation temperature \(T_{c}\) and the fluctuations around the PDW ground state in this setting are open problems that will require further investigations. Our work does not consider the competing effects of a nematic superconducting phase that has been phenomenologically found to suppress the PDW at \(T > 0\) in 2D [8, 12]. In addition, even if allowed by our model, we have not addressed the possible coexistence at low \(T\) of a PDW and a homogeneous \(d\) - wave superconductor, as suggested by cuprate experiments [5, 27]. Our results as such set the stage for future microscopic descriptions of modulated superconductivity in strongly coupled materials. + +## ACKNOWLEDGEMENTS + +We thank P. Abbamonte, B.M. Andersen, S. Caprara, A. Chubukov, E. Fradkin, S. A. Kivelson, M. Granath and A. Toschi for useful discussions and suggestions. L. F. acknowledges support by the European Union's Horizon 2020 research and innovation programme through the Marie Sklodowska- Curie grant SuperCoop (Grant No 838526). This work is supported by the DOE grant number DE- FG02- 05ER46236. + +\(\oplus\) These authors contributed equally to this work. \(\dagger\) csctty@rice.edu \(\diamond\) laura.fanfariello@ufl.edu \(\otimes\) pjh@phys.ufl.edu + +[1] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957). [2] A. Larkin and A. Varlamov, Theory of fluctuations in superconductors (Clarendon Press, 2005). [3] P. Fulde and R. A. Ferrell, Physical Review 135, A550 (1964). [4] A. Larkin and Y. N. Ovchinnikov, JETP 20, 762 (1965). [5] D. F. Agterberg, J. Davis, S. D. Edkins, E. Fradkin, D. J. Van Harlingen, P. A. Lee, L. Radzihovsky, J. M. Tranquada, Y. Wang, et al., arXiv preprint arXiv:1904.09687 (2019). [6] A. Toschi, M. Capone, and C. Castellani, Phys. Rev. B 72, 235118 (2005). [7] E. Berg, E. Fradkin, E.- A. Kim, S. A. Kivelson, V. Oganesyan, J. M. Tranquada, and S. Zhang, Physical review letters 99, 127003 (2007). [8] E. Berg, E. Fradkin, and S. A. Kivelson, Nature Physics 5, 830 (2009). [9] E. Berg, E. Fradkin, S. A. Kivelson, and J. M. Tranquada, New Journal of Physics 11, 115004 (2009). [10] P. Nikolić, A. A. Burkov, and A. Paramekanti, Physical Review B 81, 012504 (2010). + +<--- Page Split ---> + +[11] F. Loder, A. P. Kampf, and T. Kopp, Physical Review B 81, 020511 (2010).[12] D. G. Barci and E. Fradkin, Physical Review B 83, 100509 (2011).[13] F. Loder, S. Graser, A. P. Kampf, and T. Kopp, Physical review letters 107, 187001 (2011).[14] P. Nikolic and Z. Tesanovic, Physical Review B 83, 064501 (2011).[15] A. Jaefari and E. Fradkin, Physical Review B 85, 035104 (2012).[16] P. A. Lee, Physical Review X 4, 031017 (2014).[17] R. Soto-Garrido and E. Fradkin, Physical Review B 89, 165126 (2014).[18] R. Soto-Garrido, G. Y. Cho, and E. Fradkin, Physical Review B 91, 195102 (2015).[19] J. Wardi and M. Granath, Physical Review B 96, 224503 (2017).[20] J. Wardi, B. M. Andersen, and M. Granath, Physical Review B 98, 224501 (2018).[21] A. Himeda, T. Kato, and M. Ogata, Physical review letters 88, 117001 (2002).[22] M. Raczkowski, M. Capello, D. Poilblanc, R. Fresard, and A. M. Oles, Physical Review B 76, 140505 (2007).[23] M. Capello, M. Raczkowski, and D. Poilblanc, Physical Review B 77, 224502 (2008).[24] K.- Y. Yang, W. Q. Chen, T. M. Rice, M. Sigrist, and F.- C. Zhang, New Journal of Physics 11, 055053 (2009).[25] E. Berg, E. Fradkin, and S. A. Kivelson, Physical review letters 105, 146403 (2010).[26] P. Corboz, T. M. Rice, and M. Troyer, Physical review letters 113, 046402 (2014).[27] P. Choubey, S. H. Joo, K. Fujita, Z. Du, S. D. Edkins, M. H. Hamidian, H. Eisaki, S. Uchida, A. P. Mackenzie, J. Lee, J. C. S. Davis, and P. J. Hirschfeld, Proceedings of the National Academy of Sciences 117, 14805 (2020), https://www.pnas.org/content/117/26/14805.full.pdf .[28] Q. Li, M. Hucker, G. Gu, A. Tsvelik, and J. Tranquada, Physical review letters 99, 067001 (2007).[29] J. Tranquada, G. Gu, M. Hucker, Q. Jie, H.- J. Kang, R. Klingeler, Q. Li, N. Tristan, J. Wen, G. Xu, et al., Physical Review B 78, 174529 (2008).[30] A. Schafgans, A. LaForge, S. Dordevic, M. Qazilbash, W. Padilla, K. Burch, Z. Li, S. Komiya, Y. Ando, and D. Basov, Physical review letters 104, 157002 (2010).[31] A. Schafgans, C. Homes, G. Gu, S. Komiya, Y. Ando, and D. Basov, Physical Review B 82, 100505 (2010).[32] H. Jacobsen, I. Zaliznyak, A. Savici, B. Winn, S. Chang, M. Hucker, G. Gu, and J. Tranquada, Physical Review B 92, 174525 (2015).[33] M. Hamidian, S. Edkins, S. H. Joo, A. Kostin, H. Eisaki, S. Uchida, M. Lawler, E.- A. Kim, A. Mackenzie, K. Fujita, et al., Nature 532, 343 (2016).[34] R. Zhong, J. Schneeloch, H. Chi, Q. Li, G. Gu, and J. Tranquada, Physical Review B 97, 134520 (2018).[35] S. Rajasekaran, J.- i. Okamoto, L. Mathey, M. Fechner, V. Thampy, G. D. Gu, and A. Cavalleri, Science 359, 575 (2018).[36] S. D. Edkins, A. Kostin, K. Fujita, A. P. Mackenzie, H. Eisaki, S. Uchida, S. Sachdev, M. J. Lawler, E.- A. Kim, J. Seamus Davis, and M. H. Hamidian, Science 364, 976 (2019).[37] Z. Du, H. Li, S. H. Joo, E. P. Donoway, J. Lee, J. S. Davis, G. Gu, P. D. Johnson, and K. Fujita, Nature 580, 65 (2020). + +[38] S. Wang, P. Choubey, Y. X. Chong, W. Chen, W. Ren, H. Eisaki, S. Uchida, P. J. Hirschfeld, and J. Seamus Davis, arXiv preprint arXiv:2105.06518 (2021).[39] S. A. Kivelson, I. P. Bindloss, E. Fradkin, V. Oganesyan, J. Tranquada, A. Kapitulnik, and C. Howald, Reviews of Modern Physics 75, 1201 (2003).[40] T. Terashima, N. Kikugawa, A. Kiswandhi, E.- S. Choi, J. S. Brooks, S. Kasahara, T. Watashige, H. Ikeda, T. Shibauchi, Y. Matsuda, et al., Physical Review B 90, 144517 (2014).[41] S. Kasahara, T. Watashige, T. Hanaguri, Y. Kohsaka, T. Yamashita, Y. Shimoyama, Y. Mizukami, R. Endo, H. Ikeda, K. Aoyama, et al., Proceedings of the National Academy of Sciences 111, 16309 (2014).[42] K. Okazaki, Y. Ito, Y. Ota, Y. Kotani, T. Shimojima, T. Kiss, S. Watanabe, C.- T. Chen, S. Niitaka, T. Hanaguri, et al., Scientific reports 4, 1 (2014).[43] S. Kasahara, T. Watashige, T. Hanaguri, Y. Kohsaka, T. Yamashita, Y. Shimoyama, Y. Mizukami, R. Endo, H. Ikeda, K. Aoyama, et al., Proceedings of the National Academy of Sciences 111, 16309 (2014).[44] T. Watashige, S. Arsenijevic, T. Yamashita, D. Terazawa, T. Onishi, L. Opherden, S. Kasahara, Y. Tokiwa, Y. Kasahara, T. Shibauchi, et al., Journal of the Physical Society of Japan 86, 014707 (2017).[45] C. Collignon, X. Lin, C. W. Rischau, B. FauquA, and K. Behnia, Annual Review of Condensed Matter Physics 10, 25 (2019), https://doi.org/10.1146/annurev- conmatphys- 031218- 013144 .[46] Y. Cao, V. Fatemi, S. Fang, K. Watanabe, T. Taniguchi, E. Kaxiras, and P. Jarillo- Herrero, Nature 556, 43 (2018).[47] B. Padhi, C. Setty, and P. W. Phillips, Nano letters 18, 6175 (2018).[48] P. Nozieres and S. Schmitt- Rink, Journal of Low Temperature Physics 59, 195 (1985).[49] M. Randeria and E. Taylor, Annual Review of Condensed Matter Physics 5, 209 (2014), https://doi.org/10.1146/annurev- conmatphys- 031113- 133829 .[50] R. D. Duncan and C. A. R. Sá de Melo, Phys. Rev. B 62, 9675 (2000).[51] L. Benfatto, A. Toschi, S. Caprara, and C. Castellani, Phys. Rev. B 66, 054515 (2002).[52] A. V. Chubukov, I. Eremin, and D. V. Efremov, Phys. Rev. B 93, 174516 (2016).[53] This result differs from the analysis of [51], in which a sign change of the rigidity parameter is found for a SC system with \(s\) - wave pairing symmetry at strong coupling. In this case, however, the authors first perform a strong coupling expansion of the pairing susceptibility and only subsequently expand the approximated result in powers of \(q\) . + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SMPDW.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/images_list.json b/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..81e581051a8b6053b92148fe778842e540f1f59a --- /dev/null +++ b/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Dynamical Ollivier-Ricci curvature capturing the spreading of diffusion processes. a Snapshot at time \\(\\log \\tau = 0.15\\) of a pair of diffusion measures \\(\\mathbf{p}_{i}(\\tau)\\) and \\(\\mathbf{p}_{j}(\\tau)\\) started at nodes \\(i\\) , \\(j\\) of a stochastic block model network ( \\(n = 120\\) with four equal clusters with edge probabilities 0.8 within clusters and 0.1 or 0.02 between clusters). When \\(i\\) and \\(j\\) are in the same cluster, the measures overlap significantly. The size of half-circles is proportional to the amount of mass on the respective nodes. b For \\(i^{\\prime}\\) , \\(j^{\\prime}\\) in different clusters the measures remain largely disjoint. c Optimal transport plan \\(\\zeta (\\tau)\\) superimposed with \\(\\mathbf{p}_{i}(\\tau)\\) , \\(\\mathbf{p}_{j}(\\tau)\\) . When \\(i\\) , \\(j\\) (colored dashed lines) lie in the same cluster only diagonal elements \\(\\zeta_{uu}\\) are positive, meaning only geodesics within a cluster transport significant mass. The white dashed lines correspond to the four clusters. d Same as c, but with diffusions started at nodes \\(i^{\\prime}\\) and \\(j^{\\prime}\\) in different clusters. Only entries \\(\\zeta_{uv}\\) with \\(u\\) and \\(v\\) in different clusters have significant nonzero weight. e Geodesic distance matrix showing the block structure of the network. f The evolution of the edge curvatures (Eq. (2)) against time, with the highlighted lines corresponding to edges in a, b. Here \\(\\kappa_{ij}(\\tau) \\simeq 0.75\\) indicates scales when local mixing occurs between diffusion pairs. The dashed vertical lines show two such scales ( \\(\\log \\tau = 0.15\\) , 0.43). g, h Graph edges coloured by the curvature reveals the clusters at the two scales.", + "footnote": [], + "bbox": [ + [ + 84, + 66, + 912, + 400 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Edge curvature gap indicates the presence of clusters where spectral clustering fails. Two-partition symmetric SBM graph in the a dense regime ( \\(p_{\\mathrm{in}} = 0.5\\) , \\(p_{\\mathrm{out}} = 0.1\\) ) and b sparse regime ( \\(p_{\\mathrm{in}} = 8 / n\\) , \\(p_{\\mathrm{out}} = 0.5 / n\\) ). Edges are coloured by the curvature ( \\(n = 100\\) , \\(\\log \\tau = 0.83\\) ). c, d The histogram of eigenvalues obtained from five SBM realisations in dense and sparse regime, respectively. In the dense regime, the eigenvalue \\(\\lambda_{c}\\) corresponding to the community structure is well separated from the bulk eigenvalues, but overlaps in the sparse regime. e, f The evolution of edge curvatures driven against diffusion time. A gap between the curvatures of within-edges and between-edges is associated with the presence of clusters. When \\(\\kappa_{ij} > 0.75\\) (horizontal dashed line) the diffusions are well mixed across the respective edges. The curvature gap is maximal at \\(\\tau_{\\kappa} \\approx \\lambda_{c}^{-1}\\) (orange and black vertical lines). g There is no curvature gap in the limiting ER graph (inset, \\(p_{\\mathrm{in}} = p_{\\mathrm{out}} = (8 + 0.5) / (2n)\\) ). h Maximal curvature gap averaged over 20 SBM realisations for each fixed \\(\\bar{k}\\) with \\(10^{4}\\) nodes, against edge density ratio. The horizontal line marks the estimated background noise level. The intersection of this line with the mean curvature gap defines \\(r_{k}^{*}\\) , the largest possible edge density ratio to detect clusters. i Phase diagram of critical edge density ratio against average degree. The numerically obtained critical edge density ratios computed from the curvature gap are superimposed with the theoretical Kesten-Stigum detection limit (dashed line) and show excellent agreement. Gray shaded area denotes the regime where detection is possible.", + "footnote": [], + "bbox": [ + [ + 84, + 60, + 914, + 468 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Detecting communities using pairs of diffusions near the weak recovery limit. a Difference in eigenvectors \\(\\Delta \\phi_{s}\\) (Eq. (8)) between diffusion processes started at adjacent nodes for a single sparse SBM network \\((p_{\\mathrm{in}} = 3 / n\\) , \\(p_{\\mathrm{out}} = 0.5 / n\\) , \\(n = 10^{4}\\) ). Each dot marks \\((\\lambda_{s},\\Delta \\phi_{s})\\) for the 50 smallest eigenvectors, colored by the correlation of the corresponding eigenvector with the ground truth, shown in the inset. b The eigenvector with the highest \\(\\Delta \\phi_{s}\\) encodes the cluster structure (solid line), whereas the second eigenvector \\(\\phi_{2}\\) , used by spectral clustering methods, are driven by high random fluctuations. c Correlation of eigenvectors with ground truth against distance to KS limit ( \\(n = 10^{5}\\) , \\(\\bar{k} = 3\\) ). The eigenvector identified by the highest \\(\\Delta \\phi_{s}\\) approaches the correlation with the ground truth of the best eigenvector in the spectrum. All eigenvectors become uncorrelated with the ground truth at the KS limit.", + "footnote": [], + "bbox": [ + [ + 125, + 63, + 861, + 222 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Clustering networks based on multiscale geometric modularity a Clustering statistics computed based on \\(10^{2}\\) Louvain realisations for the multiscale stochastic block model graph. Vertical dashed lines show scales at which stable clusters are detected based on low variation of information at a given scale and persistent low variation of information between Louvain realisations across scales. The communities obtained at these scales are shown on b for \\(\\log \\tau = -2.3\\) and c for \\(\\log \\tau = -1\\) . Edges are coloured by the curvatures at the respective scales. d Clustering statistics for the European power grid. Two representative stable scales are shown e for \\(\\log \\tau = -0.95\\) and f for \\(\\log \\tau = -0.5\\) . g Clustering statistics for and the network of C. elegans single-neuron homeobox gene expressions show a plateau of stable scales with very similar partitions. h Clustering statistics obtained with Markov stability shows stable scales only at small times with single-node communities, indicating overfitting, and many non-robust partitions at larger scales with high variation of information. i Distance from ground truth based on structural neuronal types or the predicted clusters. Geometric modularity obtained significantly better performance than Markov stability. j Clustering of the C. elegans homeobox gene expression data obtained from geometric modularity optimisation superimposed with the ground truth.", + "footnote": [], + "bbox": [ + [ + 108, + 60, + 890, + 744 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_1.jpg", + "caption": "Supplementary Figure 1: Ricci and dynamical Ollivier Ricci curvature on canonical surfaces and graph structures. a Ricci curvature on a manifold. The geodesic distance of close points \\(x\\) and \\(y\\) on average changes when translated by parallel vectors \\(\\mathbf{v}\\) and \\(\\mathbf{v}^{\\prime}\\) on the unit circle in the tangent planes at \\(x\\) and \\(y\\) . On planes the points remain equidistant, on spheres the points contract and on hyperbolic surface they expand. b Canonical graphs with edges coloured by the dynamical OR curvature (Eq. (2)) for \\(\\tau = 1\\) show that positively and negatively curved graphs have qualitatively different topologies. Away from the boundaries, tree-like topologies are negatively curved, grid-like topologies are flat (zero curvature), whereas clique-like topologies attain positive curvature. Nodes are coloured by the average edge curvature across the neighbours. c Analogously, the differential geometric notion of Ricci curvature is negative on hyperbolic surfaces, zero on planes and positive on spherical surfaces.", + "footnote": [], + "bbox": [ + [ + 88, + 111, + 910, + 470 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 72, + 105, + 927, + 456 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 70, + 56, + 936, + 490 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 80, + 625, + 914, + 812 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 78, + 52, + 905, + 777 + ] + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f_det.mmd b/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1c331b299fa0de3a601dcddcd58195220c1a3ebb --- /dev/null +++ b/preprint/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f/preprint__118e48f8906d9884fe96b830f39986096147254471f7b700649dc45674684d7f_det.mmd @@ -0,0 +1,676 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 940, 175]]<|/det|> +# Unfolding the multiscale structure of networks with dynamical Ollivier-Ricci curvature + +<|ref|>text<|/ref|><|det|>[[44, 194, 787, 239]]<|/det|> +Adam Gosztolai ( \(\boxed{ \begin{array}{r l} \end{array} }\) adam.gosztolai@epfl.ch) École Polytechnique Fédérale de Lausanne https://orcid.org/0000- 0002- 0699- 5825 + +<|ref|>text<|/ref|><|det|>[[44, 243, 270, 283]]<|/det|> +Alexis Arnaudon Imperial College London + +<|ref|>sub_title<|/ref|><|det|>[[44, 325, 102, 343]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 361, 836, 382]]<|/det|> +Keywords: network structure, learning algorithms, network processes, Ollivier- Ricci curvature + +<|ref|>text<|/ref|><|det|>[[44, 400, 333, 419]]<|/det|> +Posted Date: February 16th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 438, 463, 458]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 222407/v1 + +<|ref|>text<|/ref|><|det|>[[44, 476, 910, 519]]<|/det|> +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 554, 907, 597]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 27th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 24884- 1. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[144, 115, 852, 168]]<|/det|> +# Unfolding the multiscale structure of networks with dynamical Ollivier-Ricci curvature + +<|ref|>text<|/ref|><|det|>[[318, 186, 678, 205]]<|/det|> +Adam Gosztolai\* \(^{1}\) and Alexis Arnaudon \(^{1,2}\) + +<|ref|>text<|/ref|><|det|>[[102, 219, 896, 270]]<|/det|> +\(^{1}\) Neuroengineering Laboratory, Brain Mind Institute & Interfaculty Institute of Bioengineering, EPFL, Lausanne, Switzerland \(^{2}\) Department of Mathematics, Imperial College London, London, United Kingdom + +<|ref|>sub_title<|/ref|><|det|>[[462, 283, 535, 299]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[129, 316, 868, 499]]<|/det|> +Defining the geometry of networks is typically associated with embedding in low- dimensional spaces such as manifolds. This approach has helped design efficient learning algorithms, unveil network symmetries and study dynamical network processes. However, the choice of embedding space is network- specific, and incompatible spaces can result in information loss. Here, we define a dynamic edge curvature for the study of arbitrary networks measuring the similarity between pairs of dynamical network processes seeded at nearby nodes. We show that the evolution of the curvature distribution exhibits gaps at characteristic timescales indicating bottleneck- edges that limit information spreading. Importantly, curvature gaps robustly encode communities until the phase transition of detectability, where spectral clustering methods fail. We use this insight to derive geometric modularity optimisation and demonstrate it on the European power grid and the C. elegans homeobox gene regulatory network finding previously unidentified communities on multiple scales. Our work suggests using network geometry for studying and controlling the structure of and information spreading on networks. + +<|ref|>text<|/ref|><|det|>[[67, 520, 912, 760]]<|/det|> +Real- world networks are rarely embedded in physical or Euclidean spaces, which complicates their analysis. However, to correctly represent node similarities, it is typical to assume that the network's nodes lie in a low- dimensional subspace, such as a manifold or linear subspace \(^{1}\) . Having this geometric backbone permits the efficient functioning of standard clustering methods, including ones based on Euclidean geometric features such as k- means or expectation maximisation \(^{2}\) . A related means of geometrising networks is possible by embedding nodes into a continuous space. For example, the hyperbolic space of constant negative curvature provides a natural parametrisation of complex networks to unveil their self- similar clusters across scales \(^{3,4}\) . Likewise, embedding networks into a geometric space based on a suitable distance metric between dynamical network processes has helped reveal their functional organisation \(^{5,6}\) . However, in general, there is no guarantee that a network is compatible with a given metric space without suffering significant distortion \(^{7}\) . At the same time, a network may have several, not necessarily self- similar, geometric representations arising, for example, from clusters at multiple resolutions \(^{8}\) . Thus, there is a need for a geometric notion that does not require embedding, yet allows studying the multiscale structure of a general class of networks. + +<|ref|>text<|/ref|><|det|>[[68, 760, 912, 881]]<|/det|> +A promising candidate is the Ollivier- Ricci (OR) curvature \(^{9}\) , which measures the change of local connectivity from one node to another, given by the cost of transporting a unit of mass between their respective neighbourhoods. Instead of imposing a geometry on the network through embedding, the OR curvature induces an effective geometry that has precise interpretation in limiting cases. In fact, it is the only one among a number of discrete curvature notions \(^{10,11}\) known to converge rigorously to the traditional Ricci curvature of a Riemannian manifold \(^{12}\) . The OR curvature is also related to graph theoretical objects, including the local clustering coefficient and bounds on the spectrum of the graph Laplacian \(^{13,14}\) , and has + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 66, 912, 102]]<|/det|> +lead to advances in applications such as studying the robustness of economic networks \(^{15}\) , characterising the human brain structural connectivity \(^{16}\) and designing clustering heuristics \(^{17,18}\) . + +<|ref|>text<|/ref|><|det|>[[57, 101, 913, 291]]<|/det|> +However, several properties of the OR curvature hinder its widespread applicability to study network clusters. Since it depends on structural neighbourhoods, it lacks a resolution parameter to define a geometry on different resolutions, that is necessary to study the multiscale structure in real- world networks. Further, the OR curvature of an edge is a local quantity, in a sense that it is controlled by the degree of its endpoints \(^{13}\) . Thus, it may provide a suboptimal geometric representation of sparse networks - including many real- world networks where each node connects only to a few others - in which node degrees vary widely. This lack of robustness of the OR curvature for sparse networks also precludes its use for studying from a geometric perspective the phase transition occurring as the community structure gets weaker and become abruptly undetectable \(^{19 - 21}\) . In other words, there is a need for a geometric notion that does not rely on embeddings, robustly captures multiscale clusters in real networks, and captures the phase transition at the limit of cluster detection. + +<|ref|>sub_title<|/ref|><|det|>[[57, 312, 172, 331]]<|/det|> +## 54 Results + +<|ref|>sub_title<|/ref|><|det|>[[57, 347, 627, 367]]<|/det|> +## 55 Dynamical Ollivier-Ricci curvature from graph diffusion + +<|ref|>text<|/ref|><|det|>[[57, 374, 913, 550]]<|/det|> +We address this need by combining two distinct frameworks - network- driven dynamical processes and geometry with OR curvature. The spreading of network- driven dynamical processes is shaped by the heterogeneity of the network. In turn, one may infer the network structure by observing properties of their evolution. We focus on Markov diffusion processes \(^{8,22 - 24}\) , a class of linear dynamical systems which is rich enough to capture several properties of nonlinear processes on networks \(^{25,26}\) . On a connected network \(G\) weighted by pairwise distances \(w_{ij}\) , the continuous time diffusion is constructed by the standard procedure \(^{27}\) of defining the normalised graph Laplacian matrix \(\mathbf{L} \coloneqq \mathbf{K}^{- 1}(\mathbf{K} - \mathbf{A})\) , where \(\mathbf{K}\) is the diagonal matrix of node degrees with \(K_{ii} = \sum_{j} A_{ij}\) and \(\mathbf{A}\) is the weighted adjacency matrix encoding similarities between nodes. For example, one may simply take \(A_{ij} = \max_{ij} w_{ij} - w_{ij}\) , or \(A_{ij} = e^{- w_{ij}}\) . Then, the probability measure of the diffusion started from the unit mass \(\delta_{i}\) on node \(i\) (Fig. 1a, b) evolves according to + +<|ref|>equation<|/ref|><|det|>[[435, 557, 910, 578]]<|/det|> +\[\mathbf{p}_{i}(\tau) = \delta_{i}e^{-\tau \mathbf{L}}. \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[57, 586, 913, 692]]<|/det|> +In analogy to the Ricci curvature on a manifold, the classical OR curvature \(^{9,28}\) measures the change of one- step neighbourhoods between nodes (see Methods and Supplementary Fig. 1a for background). Here, instead of structural neighbourhoods we consider distributions generated by diffusion processes across scales \(\tau\) . Specifically, we start a diffusion process at each node \(i = 1, \ldots , n\) to obtain a set of measures \(\mathbf{p}_{i}(\tau)\) . We then define the dynamic Ollivier- Ricci curvature of an edge as the distance of adjacent pairs of measures relative to the distance of their starting points + +<|ref|>equation<|/ref|><|det|>[[372, 699, 910, 735]]<|/det|> +\[\kappa_{ij}(\tau) \coloneqq 1 - \frac{\mathcal{W}_{1}(\mathbf{p}_{i}(\tau), \mathbf{p}_{j}(\tau))}{w_{ij}}, \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[57, 744, 913, 849]]<|/det|> +whenever \(ij\) is an edge and 0 otherwise. Intuitively, Eq. (2) measures how much 'closer' diffusions get over time when started \(w_{ij}\) distance apart, measured by \(\mathcal{W}_{1}\) , the optimal transport distance \(^{29}\) . It is obtained as a solution to a minimisation problem (Eq. (13) in Methods) and encodes the least cost of transporting the measure \(\mathbf{p}_{i}(\tau)\) to \(\mathbf{p}_{j}(\tau)\) via the edges on the graph. The minimiser of this problem is the optimal transport plan represented as a matrix \(\zeta (\tau)\) . The entries of this matrix shown on Fig. 1c, d quantify how much mass is moved between each pair of nodes \(u\) and \(v\) along their connecting geodesic of length \(d_{uv}\) (Fig. 1e). + +<|ref|>text<|/ref|><|det|>[[57, 848, 913, 935]]<|/det|> +As expected, our definition recovers the classical OR curvature \(^{9}\) as a first- order approximation for small times \(\tau \ll 1\) . Indeed, \(\mathbf{p}_{i}(\tau) \simeq \delta_{i}\mathbf{K}^{- 1}\mathbf{A}\) is the one- step measure encoding the local connectivity. Further, the dynamical OR curvature inherits the geometric intuition of the classical definition. Notably, on canonical trees- like and cliques- like networks the \(\kappa_{ij}(\tau)\) is negative and positive, respectively, for all finite scales \(\tau\) analogously to the Ricci curvature on hyperboloids and spheres (Supplementary Fig. 1b, c). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 66, 912, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 408, 914, 607]]<|/det|> +
Figure 1: Dynamical Ollivier-Ricci curvature capturing the spreading of diffusion processes. a Snapshot at time \(\log \tau = 0.15\) of a pair of diffusion measures \(\mathbf{p}_{i}(\tau)\) and \(\mathbf{p}_{j}(\tau)\) started at nodes \(i\) , \(j\) of a stochastic block model network ( \(n = 120\) with four equal clusters with edge probabilities 0.8 within clusters and 0.1 or 0.02 between clusters). When \(i\) and \(j\) are in the same cluster, the measures overlap significantly. The size of half-circles is proportional to the amount of mass on the respective nodes. b For \(i^{\prime}\) , \(j^{\prime}\) in different clusters the measures remain largely disjoint. c Optimal transport plan \(\zeta (\tau)\) superimposed with \(\mathbf{p}_{i}(\tau)\) , \(\mathbf{p}_{j}(\tau)\) . When \(i\) , \(j\) (colored dashed lines) lie in the same cluster only diagonal elements \(\zeta_{uu}\) are positive, meaning only geodesics within a cluster transport significant mass. The white dashed lines correspond to the four clusters. d Same as c, but with diffusions started at nodes \(i^{\prime}\) and \(j^{\prime}\) in different clusters. Only entries \(\zeta_{uv}\) with \(u\) and \(v\) in different clusters have significant nonzero weight. e Geodesic distance matrix showing the block structure of the network. f The evolution of the edge curvatures (Eq. (2)) against time, with the highlighted lines corresponding to edges in a, b. Here \(\kappa_{ij}(\tau) \simeq 0.75\) indicates scales when local mixing occurs between diffusion pairs. The dashed vertical lines show two such scales ( \(\log \tau = 0.15\) , 0.43). g, h Graph edges coloured by the curvature reveals the clusters at the two scales.
+ +<|ref|>text<|/ref|><|det|>[[58, 633, 910, 668]]<|/det|> +In the following we are interested in studying the curvature distribution across edges when the network structure deviates from these canonical topologies. + +<|ref|>sub_title<|/ref|><|det|>[[80, 685, 700, 705]]<|/det|> +## Edge curvature gap differences in rate of information spreading + +<|ref|>text<|/ref|><|det|>[[80, 712, 912, 799]]<|/det|> +Most real- world networks exhibit organisation on several scales. As an illustration, the unweighted stochastic block model (SBM) network \(^{30}\) of four equal- size clusters contains two nontrivial scales if the edges are drawn independently with probability 0.8 within clusters and 0.1 or 0.02 between clusters (Fig. 1a, b). We show that multiscale structure can be revealed by scanning through a finite range of scales \(\tau\) and studying snapshots of curvature distribution across edges. + +<|ref|>text<|/ref|><|det|>[[80, 799, 912, 920]]<|/det|> +The characteristic scales of a network are related to the overlap between pairs of diffusion measures \(\mathbf{p}_{i}(\tau)\) , \(\mathbf{p}_{j}(\tau)\) . This overlap depends on the starting points \(i\) , \(j\) and on network clusters which can confine diffusions on well- connected regions for long times before reaching the stationary state \(\pi^{8,22,23,31}\) , with \(\pi_{i} = \mathbf{K}_{ii} / \sum_{i} \mathbf{K}_{ii}\) . This transient phenomenon is reflected by the structure of the optimal transport matrix \(\zeta (\tau)\) . If \(i\) , \(j\) lie within the same subnetwork, the measures quickly overlap (Fig. 1a) and only diagonal entries of \(\zeta (\tau)\) are positive (Fig. 1c), weighing only short, within- cluster geodesics. By contrast, started at different subnetworks, the measures remain almost disjoint (Fig. 1b) and \(\zeta (\tau)\) is forced to select longer + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 67, 686, 85]]<|/det|> +geodesics (Fig. 1d, e), reflected by the large entries in the off- diagonal block. + +<|ref|>text<|/ref|><|det|>[[55, 85, 914, 241]]<|/det|> +The evolution of the edge curvature \(\kappa_{ij}(\tau)\) (Fig. 1f) aggregates the information in \(\zeta (\tau)\) into a single number that is related to the rate of mass exchange between subnetworks at a given scale. We see in Fig. 1f that, initially, when all nodes support disjoint point masses and the diffusions have not yet mixed, \(\begin{array}{r}{\lim_{\tau \to 0}\kappa_{ij}(\tau)\to 1 - \mathcal{W}_{1}(\delta_{i},\delta_{j}) / d_{ij} = 0} \end{array}\) . At the other extreme, as the diffusions reach stationary state, \(\begin{array}{r}{\lim_{\tau \to \infty}\kappa_{ij}(\tau)\to 1 - \mathcal{W}_{1}(\pi ,\pi) / d_{ij} = 1} \end{array}\) . At intermediate scales, the curvature can take values between 1 and some finite negative number depending on the graph. We find that, as the curvature of an edge evolves, the scale at which it approaches unity indicates how easy it is to propagate information between the subnetworks. More precisely, in the Methods, we prove that this scale gives an upper bound on the mixing time \(\tau_{ij}^{\mathrm{mix}}\) of the diffusion pair, namely, + +<|ref|>equation<|/ref|><|det|>[[376, 253, 910, 319]]<|/det|> +\[\begin{array}{l}{\tau_{ij}^{\mathrm{mix}}\coloneqq \frac{1}{2}\sum_{uv}|\zeta_{uv}(\tau) - \zeta_{uv}(\infty)|}\\ {\leq \min \{\tau :\kappa_{ij}(\tau)\geq 0.75\} ,} \end{array} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[55, 326, 912, 378]]<|/det|> +where \(\zeta (\tau)\) is the optimal transport plan with marginals \(\mathbf{p}_{i}(\tau)\) and \(\mathbf{p}_{j}(\tau)\) . Note that \(\kappa_{ij}(\tau)\geq 0.75\) does not imply that the corresponding diffusion processes have approached stationary state independently, but only that they exchange negligible mass at that or larger scales. + +<|ref|>text<|/ref|><|det|>[[55, 377, 913, 565]]<|/det|> +Importantly, a gap in the distribution of curvatures appears when the curvature exceeds 0.75 for some edges while remaining less than 0.75 for others indicating a network bottleneck that limits mass flow. To illustrate this, Fig. 1f shows three groups of edges, those with most positive curvature are found within clusters, while the other two groups of edges are found between pairs of clusters. Figs. 1g, h correspond to two scales on Fig. 1f ( \(\log \tau = 0.15\) , 0.43) where the curvature has exceeded 0.75 for some groups of edges, indicating the diffusions are well mixed within these groups, but not across other edges for which the curvature is less than 0.75. The latter mark bottleneck edges which lie between the expected partitions with 4 and 2 clusters, respectively. This simple example shows the importance of the scale parameter \(\tau\) in our curvature definition to capture the network structure at multiple scales. Before applying this to real networks, we take a closer look at the curvature gap in the theoretical context of the stochastic block model. + +<|ref|>sub_title<|/ref|><|det|>[[60, 583, 794, 602]]<|/det|> +## Curvature gap is a robust indicator of clusters in stochastic block models + +<|ref|>text<|/ref|><|det|>[[55, 610, 913, 714]]<|/det|> +Since in our example any pair of diffusions are supported by one (Fig. 1a) or two (Fig. 1b) clusters, we focus on studying the subgraph \(G\) induced by two clusters (Fig. 2a). The subgraph \(G\) is a realisation of \(\mathcal{G} = \mathrm{SSBM}(n / 2, p_{\mathrm{in}}, p_{\mathrm{out}})\) , the symmetric SBM composed of two planted partitions of equal size. Edges are generated independently with probability \(p_{\mathrm{in}}\) within- clusters and probability \(p_{\mathrm{out}}\) between- clusters. We will denote the ground truth with \(C_{i}\in \{1, - 1\}\) for each node \(i\) and define \(\bar{k} = n(p_{\mathrm{in}} + p_{\mathrm{out}}) / 2\) as the average degree. + +<|ref|>text<|/ref|><|det|>[[55, 712, 913, 919]]<|/det|> +Classical spectral clustering methods \(^{27}\) perform well for dense graphs (Fig. 2a), where \(\bar{k}\) is an increasing function of \(n\) . This suppresses fluctuations for large \(n\) causing a spectral gap to appear when the eigenvalue \(\lambda_{c}\) of the Laplacian matrix \(\mathbf{L}\) of \(G\) separates from bulk eigenvalues arising from randomness \(^{27}\) (Fig. 2c). In this dense regime, \(\lambda_{c}\) is well approximated by \(\langle \lambda_{c}\rangle_{\mathcal{G}} = 2p_{\mathrm{out}} / (p_{\mathrm{in}} + p_{\mathrm{out}})\) , the second eigenvalue of the ensemble averaged Laplacian \(\langle \mathbf{L}\rangle_{\mathcal{G}}\) (see Supplementary Note 1). Since \(\lambda_{c}\) can be identified due to the spectral gap, clustering involves simply labelling nodes by the sign of the entries of the corresponding eigenvector, which also approximates the ensemble average eigenvector \(\phi_{c}(u) = 1 / \sqrt{n}\) when \(C_{u} = 1\) and \(- 1 / \sqrt{n}\) when \(C_{u} = - 1\) . However, for sparse graphs (Fig. 2b), where \(\bar{k}\) is constant (independent of \(n\) ), the spectral gap ceases to exist \(^{32}\) (Fig. 2d). Thus, spectral algorithms relying on identifying \(\lambda_{c}\) perform no better than chance. To perform clustering in this regime, one needs to go beyond spectral clustering using, for example, the belief propagation method in statistical physics or the related non- backtracking operator whose spectrum is better behaved \(^{19,21}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 60, 914, 468]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[83, 476, 914, 705]]<|/det|> +
Figure 2: Edge curvature gap indicates the presence of clusters where spectral clustering fails. Two-partition symmetric SBM graph in the a dense regime ( \(p_{\mathrm{in}} = 0.5\) , \(p_{\mathrm{out}} = 0.1\) ) and b sparse regime ( \(p_{\mathrm{in}} = 8 / n\) , \(p_{\mathrm{out}} = 0.5 / n\) ). Edges are coloured by the curvature ( \(n = 100\) , \(\log \tau = 0.83\) ). c, d The histogram of eigenvalues obtained from five SBM realisations in dense and sparse regime, respectively. In the dense regime, the eigenvalue \(\lambda_{c}\) corresponding to the community structure is well separated from the bulk eigenvalues, but overlaps in the sparse regime. e, f The evolution of edge curvatures driven against diffusion time. A gap between the curvatures of within-edges and between-edges is associated with the presence of clusters. When \(\kappa_{ij} > 0.75\) (horizontal dashed line) the diffusions are well mixed across the respective edges. The curvature gap is maximal at \(\tau_{\kappa} \approx \lambda_{c}^{-1}\) (orange and black vertical lines). g There is no curvature gap in the limiting ER graph (inset, \(p_{\mathrm{in}} = p_{\mathrm{out}} = (8 + 0.5) / (2n)\) ). h Maximal curvature gap averaged over 20 SBM realisations for each fixed \(\bar{k}\) with \(10^{4}\) nodes, against edge density ratio. The horizontal line marks the estimated background noise level. The intersection of this line with the mean curvature gap defines \(r_{k}^{*}\) , the largest possible edge density ratio to detect clusters. i Phase diagram of critical edge density ratio against average degree. The numerically obtained critical edge density ratios computed from the curvature gap are superimposed with the theoretical Kesten-Stigum detection limit (dashed line) and show excellent agreement. Gray shaded area denotes the regime where detection is possible.
+ +<|ref|>text<|/ref|><|det|>[[84, 730, 912, 783]]<|/det|> +To see how robustly the dynamical OR curvature indicates the presence of clusters in the symmetric SBM, let us construct a measure on the curvature evolution. To this end, we define the curvature gap as the difference between the mean curvatures of within- and between- edges + +<|ref|>equation<|/ref|><|det|>[[332, 792, 910, 826]]<|/det|> +\[\delta \kappa (\tau)\coloneqq \frac{1}{\sigma}\left|\langle \kappa_{ij}(\tau)\rangle_{C_i = C_j} - \langle \kappa_{ij}(\tau)\rangle_{C_i\neq C_j}\right| \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[84, 837, 912, 932]]<|/det|> +where the averages are on within and between- edges, normalised by \(\sigma = \sqrt{\frac{1}{2}\left(\sigma_{\mathrm{within}}^{2} + \sigma_{\mathrm{between}}^{2}\right)}\) in terms of the standard deviations of both sets of curvatures. This measure is adapted from the sensitivity index in signal detection theory, known to be, asymptotically, the most powerful statistical test for discriminating two distributions. Large curvature gap \(\delta \kappa (\tau)\) indicates that the within and between edges have curvatures different enough for the clusters to be recovered (Fig. 2e, f). Correspondingly, in the limits \(\tau \to 0, \infty\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 63, 861, 222]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 230, 913, 368]]<|/det|> +
Figure 3: Detecting communities using pairs of diffusions near the weak recovery limit. a Difference in eigenvectors \(\Delta \phi_{s}\) (Eq. (8)) between diffusion processes started at adjacent nodes for a single sparse SBM network \((p_{\mathrm{in}} = 3 / n\) , \(p_{\mathrm{out}} = 0.5 / n\) , \(n = 10^{4}\) ). Each dot marks \((\lambda_{s},\Delta \phi_{s})\) for the 50 smallest eigenvectors, colored by the correlation of the corresponding eigenvector with the ground truth, shown in the inset. b The eigenvector with the highest \(\Delta \phi_{s}\) encodes the cluster structure (solid line), whereas the second eigenvector \(\phi_{2}\) , used by spectral clustering methods, are driven by high random fluctuations. c Correlation of eigenvectors with ground truth against distance to KS limit ( \(n = 10^{5}\) , \(\bar{k} = 3\) ). The eigenvector identified by the highest \(\Delta \phi_{s}\) approaches the correlation with the ground truth of the best eigenvector in the spectrum. All eigenvectors become uncorrelated with the ground truth at the KS limit.
+ +<|ref|>text<|/ref|><|det|>[[84, 393, 912, 512]]<|/det|> +where the curvatures are uniform across the graph \(\delta \kappa (\tau)\) vanishes and, likewise, in the absence of structure \((p_{\mathrm{in}}\approx p_{\mathrm{out}}\) in the Erdos- Renyi (ER) limit) we have \(\delta \kappa (\tau) = 0\) for all \(\tau\) (Fig. 2g). At intermediate scales, we find that the scale of maximal curvature gap occurs at \(\tau_{\kappa}\) at which point the curvatures of within- edges is \(\kappa_{ij}(\tau_{\kappa})\approx 0.75\) . In agreement with Eq. (3), this indicates well- mixed diffusions across these edges relative to low- curvature bottleneck edges between clusters, which indicate incomplete mixing. We also find that \(\tau_{\kappa}\approx \lambda_{c}^{- 1}\) (Fig. 2e, f). These results show that positive curvature gap is associated with the presence of clusters. + +<|ref|>text<|/ref|><|det|>[[84, 512, 912, 600]]<|/det|> +What is the minimum curvature gap needed to detect clusters? Previous works on the limits of cluster detection has shown that if the clusters are too weak (high \(r\coloneqq p_{\mathrm{out}} / p_{\mathrm{in}}\) ) or the graph too sparse (low \(\bar{k}\) ), no algorithm can assign the vertices to communities better than chance, or distinguish \(G\) from an Erdos- Renyi graph ( \(r = 1\) ). This is known as the limit of weak- recovery or detection and is characterised by the Kesten- Stigum (KS) threshold \(r = r_{\mathrm{KS}} = (\bar{k} - \sqrt{\bar{k}}) / (\bar{k} + \sqrt{\bar{k}})^{19,20,34}\) . + +<|ref|>text<|/ref|><|det|>[[84, 600, 912, 737]]<|/det|> +To study this limit, we sampled 20 networks from \(\mathcal{G}\) for a range of \(\bar{k}\) and \(r\) . For each sample, we computed the maximal curvature gap \(\delta \kappa^{*}\coloneqq \max_{\tau}\delta \kappa (\tau)\) and formed the ensemble average quantity \(\langle \delta \kappa^{*}\rangle_{\mathcal{G}}\) . As \(r\) increases for a given \(\bar{k}\) we observe that \(\langle \delta \kappa^{*}\rangle_{\mathcal{G}}\) decreases exponentially until a certain noise level (Fig. 2h). The critical edge density ratio \(r_{\bar{k}}^{*}\) can be estimated as the smallest \(r\) where \(\langle \delta \kappa^{*}\rangle_{\mathcal{G}}\) dropped below a threshold background noise level, estimated here at 0.035 (black horizontal line). This choice of threshold is not absolute, as it is affected by the finite- size effect of the SBM graphs. An analytical derivation of this threshold is out of scope of this work, but our numerical experiment clearly shows that the curvature gap detects a signal from the planted partitions up to the KS limit (Fig. 2i). + +<|ref|>sub_title<|/ref|><|det|>[[85, 755, 558, 773]]<|/det|> +## Geometric cluster detection in the sparse regime + +<|ref|>text<|/ref|><|det|>[[85, 781, 912, 850]]<|/det|> +Given that the curvature gap (Eq. (4)) indicates the presence of clusters until the fundamental KS limit we asked if this information could be used to recover the ground truth partition. The definition of curvature gap (Eq. (4)) suggests looking for equilibrium configurations of the unit- temperature Boltzmann distribution over the cluster assignments \(C\) , + +<|ref|>equation<|/ref|><|det|>[[387, 860, 910, 880]]<|/det|> +\[\mathbb{P}(C|\kappa)\propto e^{\sum_{ij}\kappa_{ij}(\tau)\delta (C_i,C_j)}, \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[84, 896, 912, 930]]<|/det|> +where \(\kappa\) is a matrix with entries \(\kappa_{ij}\) and the sum is over all edges \(ij\) . The distribution involves only within- edges because finding those is equivalent to finding between- edges, up to a normalisation factor. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 66, 912, 137]]<|/det|> +The distribution \(\mathbb{P}(C|\kappa)\) is important because all of its equilibrium states are equivalent and correlate with the ground truth partition of the symmetric SBM \(\mathcal{G}\) . To see this, we connect \(\mathbb{P}(C|\kappa)\) to the posterior distribution \(\mathbb{P}(C|G)\) of the cluster assignments obtained given the graph drawn from \(\mathcal{G}\) . In the sparse regime, the likelihood of observing \(G\) with a given cluster assignment \(C\) is + +<|ref|>equation<|/ref|><|det|>[[341, 145, 910, 191]]<|/det|> +\[\mathbb{P}(G|C)\propto \prod_{ij}\left(\frac{p_{in}}{p_{\mathrm{out}}}\right)^{\delta (C_i,C_j)}\propto \mathbb{P}(C|G) \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[55, 202, 912, 257]]<|/det|> +(see Eq. (17) in Methods). The second part of Eq. (6) results from Bayes' theorem using a uniform prior on \(C\) , since a priori all configurations are equally likely. It has been previously shown \(^{19}\) that \(P(C|G)\) is equivalent to the Boltzmann distribution of an Ising model with constant interaction strength + +<|ref|>equation<|/ref|><|det|>[[403, 266, 910, 288]]<|/det|> +\[\mathbb{P}(C|G)\propto e^{\beta \sum_{ij}\delta (C_i,C_j)} \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[55, 300, 912, 404]]<|/det|> +with inverse temperature \(\beta = \log (p_{\mathrm{in}} / p_{\mathrm{out}})\approx p_{\mathrm{in}} - p_{\mathrm{out}}\) . Note that one of the equilibrium states is trivial assigning all nodes to one cluster. However, asymptotically ( \(n\to \infty\) ) the probability of this state vanishes and the Boltzmann distribution is uniform over all other configurations with group sizes \(n / 2\) and \(p_{\mathrm{out}}n / 2\) between- edges \(^{19}\) . The fact that one of these states is the ground truth partition, and all equilibrium states of Eq. (7) are equivalent up to a permutation of nodes within clusters means they are indistinguishable from the ground truth partition. + +<|ref|>text<|/ref|><|det|>[[55, 403, 912, 550]]<|/det|> +Due to the equivalence between Eq. (6) and (7), to prove the equivalence between Eq. (5) and (6) we show that Eq. (5) can also be reduced to Eq. (7). The main insight is that the dynamical OR curvature (Eq. (2)) is constructed using pairs of diffusions, as opposed to single diffusions. Thus, eigenmodes arising from random fluctuations are reflected equally in the spectrum of both diffusions and cancel out upon taking differences over all adjacent node pairs. This allows recovering the community eigenvector \(\phi_{c}\) even in the sparse regime where when there is no spectral gap and \(\lambda_{c}\) is no longer identifiable from the spectrum (Fig. 2d). Specifically, using pairs of diffusions, we use the spectral expansion to write \(\sum_{ij}\left(p_{i}^{u}(\tau) - p_{j}^{u}(\tau)\right) = \sum_{s}e^{\lambda_{s}\tau}\phi_{s}\Delta \phi_{s}\) where + +<|ref|>equation<|/ref|><|det|>[[389, 560, 910, 599]]<|/det|> +\[\Delta \phi_{s}\coloneqq \sum_{ij}\left(\phi_{s}(i) - \phi_{s}(j)\right). \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[55, 608, 912, 782]]<|/det|> +We find that, on a single SBM realisation, \(\Delta \phi_{s}\) is large for only a few eigenvectors \(\phi_{s}\) and diminishing for others Fig. 3a). Importantly only those eigenvectors with large \(\Delta \phi_{s}\) correlate strongly with the ground truth (Fig. 3a inset). As seen in Fig. 3b, the best eigenvector is not \(\phi_{2}\) , i.e., the one whose eigenvalue is second in the spectrum and is used by spectral clustering methods, but the one whose eigenvalue is inside the bulk in Fig. 2d and thus cannot be identified by looking at the spectrum alone. The correlation with the ground truth for \(\phi_{c}\) with the highest \(\Delta \phi_{s}\) averaged over 50 SBM realisations remains close to the highest achievable among all eigenvectors as the KS bound is approached. Meanwhile, \(\phi_{2}\) is suboptimal (Fig. 3c). We also found that, close to the KS bound, often a few other eigenvectors with similarly high \(\Delta \phi_{s}\) appear, suggesting an improved clustering method combining several top eigenvectors, but this is out of scope here. + +<|ref|>text<|/ref|><|det|>[[55, 781, 912, 891]]<|/det|> +To express the curvature in the exponent of Eq. (5) we use the dual formulation of the optimal transport distance (Eq. (14) in Methods). The fact that \(\Delta \phi_{c}\) dominates the contribution from other eigenvectors, allows us to approximate \(\sum_{ij}\left(p_{i}^{u}(\tau) - p_{j}^{u}(\tau)\right) = e^{\lambda_{c}\tau}\phi_{c}\Delta \phi_{c} + \epsilon_{\phi}\propto e^{\lambda_{c}\tau}\phi_{c} + \epsilon_{\phi}\) , where \(\epsilon_{\phi}\) is an asymptotically small term. We use this expression, together with the duality formula (Eq. (14)) to express Eq. (5). Finally, in the sparse regime, we may make a tree- like approximation of the neighbourhoods of \(i\) and \(j\) to find that Eq. (5) reduces to + +<|ref|>equation<|/ref|><|det|>[[377, 900, 910, 921]]<|/det|> +\[\mathbb{P}(C|\kappa)\propto e^{|p_{\mathrm{in}} - p_{\mathrm{out}}|\sum_{ij}\delta (C_i,C_j)}. \quad (9)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 66, 912, 120]]<|/det|> +We refer the reader to the Methods for details. Eq. (9) is the same as Eq. (7) when the communities are assortative \((p_{\mathrm{in}} > p_{\mathrm{out}})\) . We then conclude that the curvatures encode the communities of the symmetric SBM and allow it to be recovered until close to the Kesten- Stigum bound. + +<|ref|>text<|/ref|><|det|>[[70, 120, 912, 154]]<|/det|> +In the next section, we present a clustering algorithm based on this insight that can find multiscale clusters in real- world networks. + +<|ref|>sub_title<|/ref|><|det|>[[70, 171, 696, 191]]<|/det|> +## Geometric modularity for the multiscale clustering of networks + +<|ref|>text<|/ref|><|det|>[[70, 198, 912, 251]]<|/det|> +To exploit the property of the dynamical OR curvature to give multiple geometric representations, we develop a multiscale graph clustering algorithm for real- world networks. Using Eq. (5), we introduce the geometric modularity function + +<|ref|>equation<|/ref|><|det|>[[327, 259, 910, 302]]<|/det|> +\[Q_{\kappa}(C,\tau) = \frac{1}{2m_{\kappa}}\sum_{ij}\left(\kappa_{ij}(\tau) - \kappa_{0}\right)\delta (C_{i},C_{j}), \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[70, 312, 912, 451]]<|/det|> +where \(2m_{\kappa} = \sum_{ij}|\kappa_{ij}|\) is a normalisation factor and \(\kappa_{0} = \max_{ij}\kappa_{ij}(\tau_{\mathrm{min}})\) is a constant ensuring that all edges have small non- positive curvature at the smallest computed scale \(\tau_{\mathrm{min}}\) . Hence optimising Eq. (10) at small times yields separate communities for each node whereas at large times, when \(\kappa_{ij}(\tau)\to 1\) for all \(ij\) , all nodes are merged to a single community. At intermediate scales, the curvatures will have negative and positive values on different edges, making the detection of non- trivial clusters possible without a statistical null- model. This is in contrast to classical modularity \(^{35}\) , which minimises the expected number of edges between clusters, and requires a statistical null- model (typically the configuration model), which can hinder identifying functional communities based on dynamics \(^{5}\) . + +<|ref|>text<|/ref|><|det|>[[70, 451, 912, 604]]<|/det|> +To detect robust partitions at several scales, we sample the cluster landscape \(Q_{\kappa}(C,\tau)\) at a sequence of scales \(\tau\) spanning the entire dynamical range of the curvature and, at each \(\tau\) , optimising Eq. (10) using the Louvain algorithm \(^{36,37}\) with 200 random initialisations. At a given \(\tau\) , we take the cluster with the highest geometric modularity and deem it robust if it has a low variation of information \(\mathrm{VI}_{\tau}\) against 50 other randomly chosen clusters at this scale, as well as low variation of information \(\mathrm{VI}_{\tau \tau^{\prime}}\) against the best cluster assignments at nearby scales \(\tau^{\prime}\) . As an example, we show in Fig. 4a the result of this computation on our four- partition SBM graph with two hard- coded scales. We clearly see two large plateaus with low \(\mathrm{VI}_{\tau}\) and \(\mathrm{VI}_{\tau \tau^{\prime}}\) , corresponding to robust clusters, shown in Fig. 4b,c. At the smallest scales we find no robust communities shown by the sharp increase in the number of communities and the large \(\mathrm{VI}_{\tau}\) . + +<|ref|>text<|/ref|><|det|>[[70, 604, 912, 809]]<|/det|> +Due to the link between high edge curvature and well- mixed state (Eq. (3)), we expected that at robust scales the clusters will correspond to those regions which have a high amount of redundant information, and thus can be disconnected without affecting the dynamics within them. To see this, we applied this clustering algorithm to the European power grid graph in Fig. 4d,e,f, an unweighted network of major electrical lines, which has been previously analysed for robustness \(^{38}\) , multiscale communities \(^{39}\) and centrality \(^{24}\) . The multiscale community structure can be clearly seen with the many minima of the \(\mathrm{VI}_{\tau}\) function in Fig. 4d. We displayed two scales in Fig. 4e,f which unfold parts of the power grid which have been historically independently developed. The smaller scale (at around \(\log \tau = - 0.95\) ) marks countries or economical and historical alliances (Skandinavia, Benelux, Czechoslovakia, Balkans, etc.). Likewise, the larger scale (at around \(\log \tau = - 0.5\) ) divides historical Eastern- Western Europe. Interestingly with the boundary in Germany runs along the iron curtain, which also demarcates the regions between major electricity companies. + +<|ref|>text<|/ref|><|det|>[[70, 810, 912, 897]]<|/det|> +Finally, we analysed a recent dataset of homeobox gene expression in single neurons of \(C\) . elegans in Fig. 4g, h, i and \(\mathrm{j}^{40}\) . This work found based on a multivariate linear regression that the homeobox gene expression profile in a given anatomical neuron class can explain on average 74% of the expression level of the remaining genes in that neuron class. We therefore asked whether the homeobox gene expression profile has sufficient information to cluster neurons into their known anatomical classes. + +<|ref|>text<|/ref|><|det|>[[70, 897, 912, 931]]<|/det|> +The data contains a binary feature vector for each of the 301 neurons, indicating the presence of a protein expressed by any of the 105 homeobox genes in the given neuron. To convert this data into a + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[108, 60, 890, 744]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 768, 913, 965]]<|/det|> +
Figure 4: Clustering networks based on multiscale geometric modularity a Clustering statistics computed based on \(10^{2}\) Louvain realisations for the multiscale stochastic block model graph. Vertical dashed lines show scales at which stable clusters are detected based on low variation of information at a given scale and persistent low variation of information between Louvain realisations across scales. The communities obtained at these scales are shown on b for \(\log \tau = -2.3\) and c for \(\log \tau = -1\) . Edges are coloured by the curvatures at the respective scales. d Clustering statistics for the European power grid. Two representative stable scales are shown e for \(\log \tau = -0.95\) and f for \(\log \tau = -0.5\) . g Clustering statistics for and the network of C. elegans single-neuron homeobox gene expressions show a plateau of stable scales with very similar partitions. h Clustering statistics obtained with Markov stability shows stable scales only at small times with single-node communities, indicating overfitting, and many non-robust partitions at larger scales with high variation of information. i Distance from ground truth based on structural neuronal types or the predicted clusters. Geometric modularity obtained significantly better performance than Markov stability. j Clustering of the C. elegans homeobox gene expression data obtained from geometric modularity optimisation superimposed with the ground truth.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 67, 912, 188]]<|/det|> +graph with nodes being neurons, we first eliminated all homeobox genes co- expressed in none or more than \(90\%\) of the neurons to retain 67 homeobox genes. We then constructed an all- to- all graph adjacency matrix weighted by the Jaccard similarity index between expression profiles of neurons. To increase the number of edges with negative curvature, thus improve the detection at the smallest scales, we sparsified this network using a geometric sparsification method \(^{41}\) with parameter \(\gamma = 0.01\) . This method retains at most a fraction \(\gamma\) edges of the original graph as minimum spanning tree augmented by edges relevant for preserving local or global geometry of the graph. + +<|ref|>text<|/ref|><|det|>[[57, 188, 912, 426]]<|/det|> +The results of our clustering algorithm on this graph is shown on Fig. 4g and compared with the result of Markov stability \(^{8}\) (Fig. 4h), a multiscale method based on persistence of diffusions. Geometric modularity obtains a large range of robust scales with highly similar clusters - as shown by the low \(\mathrm{VI}_{\tau}\) and \(\mathrm{VI}_{\tau \tau^{\prime}}\) . These scales correlate closely with the known ground truth of 117 anatomical neuron classes (Fig. 4i). In contrast, for Markov stability \(^{8}\) , the scales with low \(VI_{\tau}\) overfit the graph finding too many clusters (Fig. 4h) which correlate less with the ground truth (Fig. 4i). Likewise, hierarchical clustering fails to identify the ground truth communities \(^{40}\) . On Fig. 4j we superimpose the best clustering from geometric modularity against the ground- truth. We observe little differences, apart from VA and AS nodes as well as VD and DD often clustered together. Careful look reveals close biological relationship between these classes; all four classes correspond to motor neurons, with pairs expressing the same neurotransmitters – VA, AS expressing acetylcholine and VD, DD expressing gamma- aminobutyric acid (GABA). These novel results give direct quantitative support to the claim that homeobox gene expression patterns encode structural neuron types. We also observe other stable partitions at larger scale, but they did not correlate the ground- truth. + +<|ref|>text<|/ref|><|det|>[[58, 427, 911, 461]]<|/det|> +Overall, these results give a strong demonstration that our method is able to find stable clusters in sparse graphs, and provide meaningful insights into distinct types of real- world networks. + +<|ref|>sub_title<|/ref|><|det|>[[58, 480, 188, 496]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[58, 506, 912, 592]]<|/det|> +We introduced the concept of dynamical Ollivier- Ricci (OR) curvature which defines an effective geometry from pairs of diffusion processes on the network. Instead of imposing the requirement of a manifold approximation or embedding, used by previous geometric approaches \(^{4 - 6,23}\) , our approach constructs a geometric object - the weighted and signed edge curvature matrix - capturing progressively coarser features as the diffusion processes evolve. + +<|ref|>text<|/ref|><|det|>[[58, 592, 912, 745]]<|/det|> +Real- world networks often exhibit community structure on multiple scales, based on difference between the rates of information propagation in regions the network on various timescales. We showed that the edge curvature matrix carries a precise meaning in this context and bounds the rate of information flow across edges. Consequentially, curvature gaps, differences between edge curvatures within and between regions, indicate network bottlenecks. This result does not rely on the dynamics being linear diffusions, making it suitable to study the interaction of arbitrary dynamical processes. We expect that, in the future, this approach can be used to tune the geometry of the graph to control the flux or interaction of network- driven dynamical processes, for example, leading to better insights to synchronisation problems or metapopulation models \(^{42}\) . + +<|ref|>text<|/ref|><|det|>[[58, 746, 912, 934]]<|/det|> +Although diffusion processes constructed from the graph Laplacian have been explored for network clustering \(^{8,23}\) , our work differs in the use of diffusion pairs to construct the curvature. Two diffusions pick up random variations in the graph independently, which can be exploited to average out non- informative fluctuations. On stochastic block models, this feature allows the curvature gap to robustly indicate clusters in the sparse regime down to the fundamental limit, where clustering methods relying on the spectral gap in the Laplacian fail \(^{32}\) . We also found a new measure of eigenvalue quality, able to select the best eigenvector to be used in spectral methods. Interestingly, the edge curvatures are defined on the set of shortest paths which cannot contain the same edge twice, a subset of the set of non- backtracking walks. Our results are therefore consistent with previous works on the limits of cluster detection using statistical physics objects including the spectrum of non- backtracking operator \(^{21}\) or related message passing approaches \(^{19}\) . We expect this insight to provide a new avenue to study the fundamental limits of efficient clustering from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 70, 275, 84]]<|/det|> +a geometric perspective. + +<|ref|>text<|/ref|><|det|>[[55, 85, 912, 153]]<|/det|> +Finally, using geometric modularity we built an easy- to- use algorithm to study the multiscale community structure of real networks. We demonstrated on the European power grid network and a recent dataset of \(C\) . elegans single- neuron homeobox gene expressions that our method can find robust and interpretable communities on multiple scales on diverse datasets without the tendency of overfitting. + +<|ref|>text<|/ref|><|det|>[[55, 153, 912, 188]]<|/det|> +Overall, we expect our insights connecting dynamical processes, geometry and network clustering to open new avenues to studying and controlling the structural and dynamical properties of networks. + +<|ref|>sub_title<|/ref|><|det|>[[56, 207, 270, 223]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[55, 232, 912, 300]]<|/det|> +AG acknowledges support from an HFSP Cross- disciplinary Postdoctoral Fellowship (LT000669/2020- C). We thank Mauricio Barahona for insightful discussions on this topic, Jonas Braun and István Tomon for their helpful comments on the manuscript and Daniel Morales for inspiring us to analyse the \(C\) . elegans dataset. + +<|ref|>sub_title<|/ref|><|det|>[[56, 321, 293, 338]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[56, 348, 465, 365]]<|/det|> +A.G. and A.A. contributed equally to this work. + +<|ref|>sub_title<|/ref|><|det|>[[56, 383, 248, 401]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[56, 410, 912, 445]]<|/det|> +The code to reproduce the results in our paper and to perform geometric modularity optimisation is available at https://github.com/agosztolai/geometric_clustering. + +<|ref|>sub_title<|/ref|><|det|>[[56, 463, 245, 480]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[56, 490, 710, 507]]<|/det|> +The raw data supporting the results is available from the authors upon request. + +<|ref|>sub_title<|/ref|><|det|>[[56, 530, 189, 550]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[56, 566, 456, 584]]<|/det|> +## Classical Ollivier-Ricci edge curvature + +<|ref|>text<|/ref|><|det|>[[56, 592, 912, 679]]<|/det|> +To contrast the dynamical Ollivier Ricci curvature in Eq. (2), we recap the definition of the classical formulation \(^{9}\) , which is a generalisation of the Ricci curvature of manifolds in differential geometry. Briefly, consider two close points \(x\) and \(y\) on a manifold as well as a vector \(\mathbf{v}\) on the tangent plane at \(x\) and another tangent vector \(\mathbf{v}^{\prime}\) in the tangent plane at \(y\) that is parallel to \(\mathbf{v}\) , i.e., obtained by parallel transport along the geodesic connecting \(x, y\) (Supplementary Figure 1a). These vectors shift \(x\) and \(y\) to nearby points \(x^{\prime}\) and \(y^{\prime}\) , which will be at a distance \(d_{x^{\prime}y^{\prime}} \approx d_{xy}(1 - ||\mathbf{v}||^{2}K_{w} / 2)\) , where \(K_{w}\) is the sectional curvature. The Ricci curvature \(Ric_{xy}\) between points \(x, y\) is then defined as the average sectional curvature and is proportional to + +<|ref|>equation<|/ref|><|det|>[[431, 686, 910, 717]]<|/det|> +\[R i c_{x y}\propto 1 - \frac{\langle d_{x^{\prime}y^{\prime}}\rangle}{d_{x y}}, \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[56, 725, 912, 782]]<|/det|> +where \(\langle \cdot \rangle\) denotes the average over all vectors \(\mathbf{w}, \mathbf{w}^{\prime}\) running over the unit sphere in the tangent planes at \(x\) and \(y\) . In other words, it measures how much geodesics expand or contract on average around points \(x, y\) . On flat planes the geodesics stay equally separated hence \(Ric_{xy} = 0\) , on spheres the geodesics contract hence \(Ric_{xy} > 0\) , whereas in hyperbolic spaces they expand, hence \(Ric_{xy} < 0\) (Supplementary Figure 1a). + +<|ref|>text<|/ref|><|det|>[[56, 781, 912, 820]]<|/det|> +The classical Ollivier- Ricci curvature \(^{9}\) is defined by direct analogy to this. Consider two adjacent nodes \(i\) and \(j\) and place weights on their immediate neighbours in proportion to the edge weights, namely, \(\mathbf{p}_{i} = \delta_{i}\mathbf{K}^{- 1}\mathbf{A}\) . Then the Ollivier Ricci curvature becomes + +<|ref|>equation<|/ref|><|det|>[[425, 816, 910, 846]]<|/det|> +\[\kappa_{i j} = 1 - \frac{\mathcal{W}_{1}(\mathbf{p}_{i},\mathbf{p}_{j})}{d_{i j}}. \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[56, 850, 912, 880]]<|/det|> +Contrast this expression to the dynamical Ollivier Ricci curvature in Eq. (2), which considers diffusion measures which weight progressively larger neighbourhoods as \(\tau\) increases. + +<|ref|>text<|/ref|><|det|>[[56, 879, 912, 935]]<|/det|> +We remark that Eq. (12) and the dynamical OR curvature in (2) are valid for any two nodes \(i, j\) if the denominator is replaced by the weighted geodesic distance between \(i\) and \(j\) , but for this work, it suffices to consider adjacent nodes. Indeed, for any non- adjacent nodes \(uv\) , \(\kappa_{uv} \geq \kappa_{u^{\prime}v^{\prime}}\) , where \(u^{\prime}v^{\prime}\) is an adjacent pair lying on the geodesic connecting \(u, v\) (Proposition 19 in Ref. \(^{9}\) ), meaning that local curvatures control global curvatures. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[45, 66, 288, 83]]<|/det|> +347 Wasserstein distance + +<|ref|>text<|/ref|><|det|>[[50, 93, 912, 123]]<|/det|> +348 To measure the distance between a pair of measures \(\mathbf{p}_{i}(\tau)\) and \(\mathbf{p}_{j}(\tau)\) we use the optimal transport distance \(^{29}\) (also known as 349 1- Wasserstein or earth- mover distance), defined as + +<|ref|>equation<|/ref|><|det|>[[344, 130, 910, 193]]<|/det|> +\[\begin{array}{l}{\mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau)) = \min_{\zeta}\sum_{u v}d_{u v}\zeta_{u v},}\\ {\mathrm{subject~to}\sum_{v}\zeta_{u v} = p_{i}^{u}(\tau),\quad \sum_{u}\zeta_{u v} = p_{j}^{v}(\tau).} \end{array} \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[50, 200, 912, 230]]<|/det|> +350 The constraints in Eq. (13) ensure that the optimal transport plan \(\zeta (\tau) \in \mathbb{R}^{n \times n}\) is a coupling of the measures \(\mathbf{p}_{i}(\tau)\) , \(\mathbf{p}_{j}(\tau)\) , i.e., \(\zeta (\tau)\) is a joint distribution that admits \(\mathbf{p}_{i}(\tau)\) and \(\mathbf{p}_{j}(\tau)\) as marginals. + +<|ref|>text<|/ref|><|det|>[[50, 229, 876, 244]]<|/det|> +352 An equivalent formulation of this distance can be constructed from the Kantorovich- Rubinstein duality \(^{29}\) , given by + +<|ref|>equation<|/ref|><|det|>[[341, 250, 910, 280]]<|/det|> +\[\mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau)) = \sup_{f}\sum_{u}f(u)[p_{i}^{u}(\tau) - p_{j}^{u}(\tau)] \quad (14)\] + +<|ref|>text<|/ref|><|det|>[[50, 287, 637, 303]]<|/det|> +353 where the supremum is taken over all 1- Lipschitz functions \(f\) on the graph, that is, + +<|ref|>equation<|/ref|><|det|>[[432, 311, 910, 328]]<|/det|> +\[|f(u) - f(v)|\leq d_{u v} \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[50, 336, 236, 350]]<|/det|> +354 for any node pair \(u\) , \(v\) . + +<|ref|>sub_title<|/ref|><|det|>[[50, 370, 343, 388]]<|/det|> +355 Computational complexity + +<|ref|>text<|/ref|><|det|>[[50, 395, 912, 424]]<|/det|> +356 The computational complexity of our clustering method is determined by three components: the computation of the diffusion measures (Eq. (1)), the computation of the optimal transport distance (Eq. (13)) and the computation of the clustering. + +<|ref|>text<|/ref|><|det|>[[50, 423, 912, 550]]<|/det|> +357 We compute each diffusion measure in Eq. (1) by the scaling and squaring algorithm of Ref. \(^{43}\) . To our knowledge, the complexity of this algorithm is not known, but we found it to be better than computing the matrix exponential which runs in time \(\mathcal{O}(n^{3})\) and then multiplying by the initial condition. The exact computation of Eq. (13) is performed by interior point methods which have a complexity \(\mathcal{O}(n^{3} \log n)\) . However, note that in our work the explicit computation of \(\zeta (\tau)\) is not required and, moreover, there is typically a significant overlap between the measures \(\mathbf{p}_{i}\) , \(\mathbf{p}_{j}\) whenever \(i\) , \(j\) lie in the same highly connected region. Based on these observations we use recent approximate algorithms to compute the transport cost permitting near \(\mathcal{O}(n)\) - time computation of the optimal transport distance \(^{44,45}\) . This yields a computational complexity of \(\mathcal{O}(nm)\) for sparse graphs with worst case \(\mathcal{O}(n^{3})\) . These methods also allow GPU parallelisation, which we recommend using for large \((n \gg 10^{3})\) and dense graphs. + +<|ref|>text<|/ref|><|det|>[[80, 548, 830, 564]]<|/det|> +The third component is the complexity of the Louvain algorithm which is linear \(\mathcal{O}(n)\) for sparse networks \(^{36}\) . + +<|ref|>sub_title<|/ref|><|det|>[[50, 582, 620, 600]]<|/det|> +368 Upper bound on the mixing time in terms of curvature + +<|ref|>text<|/ref|><|det|>[[50, 606, 912, 650]]<|/det|> +369 Here we prove inequality (3), which gives an upper bound on the mixing time of the coupled diffusions with measures \(\mathbf{p}_{i}(\tau)\) , \(\mathbf{p}_{j}(\tau)\) in terms the dynamical OR curvature. The \(\epsilon\) - mixing time is defined as the smallest \(\tau\) where the law of the coupled process, the optimal transport plan \(\zeta (\tau)\) , is within an \(\epsilon\) radius of the stationary distribution + +<|ref|>equation<|/ref|><|det|>[[365, 657, 910, 674]]<|/det|> +\[\tau_{ij}(\epsilon)\coloneqq \min \{\tau :||\zeta (\tau) - \zeta (\infty)||_{\mathrm{TV}}\leq \epsilon \} , \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[50, 682, 912, 712]]<|/det|> +372 where the notion of "close to stationarity" is quantified by the total variation distance \(\begin{array}{r}{||\zeta (\tau) - \zeta (\infty)||_{\mathrm{TV}}\coloneqq \frac{1}{2}\sum_{u v}|\zeta_{u v}(\tau) - } \end{array}\) \(\zeta_{u v}(\infty)]\) . Since \(\mathbf{p}_{i}(\tau)\) and \(\mathbf{p}_{j}(\tau)\) are marginals of \(\zeta (\tau)\) we have that + +<|ref|>equation<|/ref|><|det|>[[315, 718, 680, 754]]<|/det|> +\[\tau_{ij}(\epsilon) = \min \{\tau :||\mathbf{p}_{i}(\tau) - \pi ||_{\mathrm{TV}} + ||\mathbf{p}_{j}(\tau) - \pi ||_{\mathrm{TV}}\leq \epsilon \}\] \[\qquad = \min \{\tau :||\mathbf{p}_{i}(\tau) - \mathbf{p}_{j}(\tau)||_{\mathrm{TV}}\leq \epsilon \} ,\] + +<|ref|>text<|/ref|><|det|>[[50, 760, 912, 790]]<|/det|> +374 where we used the independence of the diffusion processes. From here, we may follow Ref. \(^{46}\) and use the Csiszar- KullbackPinsker inequality for the optimal transport distance + +<|ref|>equation<|/ref|><|det|>[[339, 799, 656, 814]]<|/det|> +\[||\mathbf{p}_{j}(\tau) - \mathbf{p}_{j}(\infty)||_{\mathrm{TV}}\leq (1 / d_{0})\mathcal{W}_{1}(\mathbf{p}_{j}(\tau),\mathbf{p}_{j}(\tau)),\] + +<|ref|>text<|/ref|><|det|>[[50, 822, 850, 838]]<|/det|> +376 where \(d_{0} = \min_{ij}d_{ij}\) is a global graph constant, which can therefore be absorbed into \(\epsilon\) . This gives an upper bound + +<|ref|>equation<|/ref|><|det|>[[364, 845, 632, 880]]<|/det|> +\[\tau_{ij}(\epsilon^{\prime})\leq \min \{\tau :\mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau))\leq \epsilon^{\prime}\}\] \[\qquad = \min \{\tau :\kappa_{ij}(\tau)\geq 1 - \epsilon^{\prime}\} ,\] + +<|ref|>text<|/ref|><|det|>[[50, 888, 912, 918]]<|/det|> +377 with \(\epsilon^{\prime} = d_{0}\epsilon\) which is what we set out to show. Note that choosing any \(\epsilon^{\prime}\in (0,1 / 2)\) ensures exponential convergence rate to the stationary measure \(^{47}\) and by convention, we take the middle of this range and define \(\tau_{ij}^{\mathrm{mix}}\coloneqq \tau_{ij}^{\mathrm{mix}}(1 / 4)\) to obtain Eq. (3). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[80, 65, 910, 85]]<|/det|> +## Connection between geometric modularity and the symmetric stochastic block model + +<|ref|>text<|/ref|><|det|>[[55, 91, 912, 150]]<|/det|> +In this section, we prove that the Boltzmann distribution of cluster assignments given the edge curvatures \(\mathbb{P}(C|\kappa)\) (Eq. (5)) has equilibrium states which are indistinguishable from the ground truth partition of the SBM. We show this by reducing \(\mathbb{P}(C|\kappa)\) as well as the posterior distribution \(\mathbb{P}(C|G)\) , to the same constant interaction Ising model (Eq. (7)). In the remainder of this section we work in the sparse regime, where \(p_{\mathrm{in}}\) , \(p_{\mathrm{out}} = O(1 / n)\) . + +<|ref|>text<|/ref|><|det|>[[55, 148, 912, 178]]<|/det|> +First, we recap the well- known equivalence of the SBM and the Ising model \(^{19}\) . Let \(E\) denote the set of edges. The probability distribution of the symmetric SBM for two clusters can be written as \(^{30}\) + +<|ref|>equation<|/ref|><|det|>[[319, 183, 910, 283]]<|/det|> +\[\begin{array}{r l} & {\mathbb{P}(G|C) = p_{\mathrm{out}}^{\mathrm{e}}(1 - p_{\mathrm{out}})^{\binom{n}{2} -e}\times}\\ & {\qquad \times \prod_{i j\in E}\left(\frac{p_{\mathrm{in}}}{p_{\mathrm{out}}}\right)^{\delta (C_{i},C_{j})}\prod_{i j\notin E}\left(\frac{1 - p_{\mathrm{in}}}{1 - p_{\mathrm{out}}}\right)^{\delta (C_{i},C_{j})}}\\ & {\qquad \propto \prod_{i j\in E}\left(\frac{p_{\mathrm{in}}}{p_{\mathrm{out}}}\right)^{\delta (C_{i},C_{j})}} \end{array} \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[55, 288, 912, 346]]<|/det|> +where \(e\) is the total number of edges and in the last line we used that the effect of non- edges is weak in the sparse regime. Therefore, by Bayes' theorem with uniform prior one obtains the posterior distribution \(\mathbb{P}(C|G) \propto \mathbb{P}(G|C)\) . As a result, the probability of clusters generated by the SBM is equivalent to the Ising model with uniform interaction with Boltzmann distribution given by Eq. (7) \(^{19}\) . + +<|ref|>text<|/ref|><|det|>[[55, 344, 912, 374]]<|/det|> +Second, we reduce the Boltzmann distribution of clusters given the edge curvature to same Ising model in Eq. (7). From Eq. (5) we have + +<|ref|>equation<|/ref|><|det|>[[358, 378, 910, 418]]<|/det|> +\[\begin{array}{r l} & {\mathbb{P}(C|\kappa)\propto e^{\sum_{i j}\kappa_{i j}(\tau)\delta (C_{i},C_{j})}}\\ & {\qquad \propto e^{\sum_{i j}[1 - \mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau))]\delta (C_{i},C_{j})},} \end{array} \quad (18)\] + +<|ref|>text<|/ref|><|det|>[[55, 424, 912, 483]]<|/det|> +where in the last line we used the definition of the curvature in Eq. (2). Comparing Eq. (18) with Eq. (7) note that \(1 - \mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau))\) is non- constant and has a non- linear dependence on the scale \(\tau\) . However, it is possible to express it in terms of \(p_{\mathrm{in}}\) , \(p_{\mathrm{in}}\) to make the connection to the Ising model. Let us write the diffusion measures in Eq. (1) in terms of the spectral decomposition of \(\mathbf{L}\) as + +<|ref|>equation<|/ref|><|det|>[[319, 487, 910, 526]]<|/det|> +\[p_{i}^{k}(\tau) = \delta_{i}\sum_{s = 1}^{n}e^{-\lambda_{s}\tau}\phi_{s}(k)\phi_{s}(l) = \sum_{s = 1}^{n}e^{-\lambda_{s}\tau}\phi_{s}(\boldsymbol {k})\phi_{s}(\boldsymbol {i}). \quad (19)\] + +<|ref|>text<|/ref|><|det|>[[55, 531, 912, 616]]<|/det|> +At this point let us remark that in the dense regime where \(p_{\mathrm{in}}\) , \(p_{\mathrm{out}} = O(1)\) , the first two eigenmodes \((\lambda_{1},\phi_{1})\) and \((\lambda_{c},\phi_{c})\) dominate and the second eigenmode contains the anti- symmetric eigenvector \(\phi_{c}(u) = 1 / \sqrt{n}\) when \(C_{u} = 1\) and \(- 1 / \sqrt{n}\) when \(C_{u} = 2\) that is associated with the community structure (Fig. 2c). Thus, one can follow spectral clustering methods \(^{27}\) to find the sparsest cut between clusters using \(\phi_{c}\) . In contrast, in the sparse regime, the dominant eigenmodes will be driven by random fluctuations in the node degrees across the graph \(^{48}\) , thus spectral clustering algorithms based on \(\mathbf{L}\) are suboptimal (Fig. 2d). + +<|ref|>text<|/ref|><|det|>[[55, 614, 912, 643]]<|/det|> +However, the coupled diffusion pair allows for cancelling out random fluctuations in their spectrum. To see this, consider for a between- edge \(ij\) the difference + +<|ref|>equation<|/ref|><|det|>[[319, 648, 910, 728]]<|/det|> +\[\begin{array}{r l r} & {} & {\sum_{i j\in E}p_{i}^{k}(\tau) - p_{j}^{k}(\tau) = \sum_{i j\in E}\sum_{s = 1}^{n}e^{-\lambda_{s}\tau}\phi_{s}(k)[\phi_{s}(i) - \phi_{s}(j)]}\\ & {} & {= \sum_{s = 1}^{n}e^{-\lambda_{s}\tau}\phi_{s}(k)\Delta \phi_{s},} \end{array} \quad (20)\] + +<|ref|>text<|/ref|><|det|>[[55, 732, 912, 775]]<|/det|> +where \(\Delta \phi_{s}\) is defined in Eq. (8). The first term involves the constant eigenvector \(\phi_{1}\) corresponding to the stationary state. Therefore, \(\phi_{1}(i) = \phi_{1}(j)\) for all \(ij\) and thus its contributions cancels out when taking differences. Further, for eigenvectors \(\phi_{s}\) with \(s \neq 1\) , \(c\) we have asymptotically \((n \to \infty)\) that + +<|ref|>equation<|/ref|><|det|>[[466, 775, 530, 789]]<|/det|> +\[\Delta \phi_{s}\to 0\] + +<|ref|>text<|/ref|><|det|>[[55, 792, 912, 821]]<|/det|> +(Fig. 3). As a result, the only contribution we are left with is coming from the anti- symmetric eigenmode \((\lambda_{c},\phi_{c})\) . Thus we have that + +<|ref|>equation<|/ref|><|det|>[[310, 816, 910, 855]]<|/det|> +\[\sum_{i j\in E}(p_{i}^{u}(\tau) - p_{j}^{u}(\tau)) = \left\{ \begin{array}{l l}{\epsilon_{\phi},} & {\mathrm{if~}C_{i} = C_{j},}\\ {e^{-\lambda_{s}\tau}\phi_{c}\Delta \phi_{c} + \epsilon_{\phi},} & {\mathrm{if~}C_{i}\neq C_{j},} \end{array} \right. \quad (21)\] + +<|ref|>text<|/ref|><|det|>[[55, 857, 790, 872]]<|/det|> +where \(\epsilon_{\phi}\) represents the contribution from the random eigenvectors which is negligible in the limit \(n \to \infty\) . + +<|ref|>text<|/ref|><|det|>[[55, 870, 912, 900]]<|/det|> +To compute \(\mathcal{W}_{1}\) in the exponent of Eq. (18), we use Kantorovich- Rubinstein duality (Eq. (14)). Using Eq. (21) in Eq. (14) and ignoring asymptotically small terms, we consider the quantity + +<|ref|>equation<|/ref|><|det|>[[325, 905, 515, 938]]<|/det|> +\[\sum_{i j\in E}\sum_{u}f(u)\left[p_{i}^{u}(\tau) - p_{j}^{u}(\tau)\right]\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[325, 63, 910, 210]]<|/det|> +\[\begin{array}{l}{= e^{-\lambda_{c}\tau}\sum_{u}f(u)\phi_{c}(u)}\\ {= \frac{e^{-\lambda_{c}\tau}}{n}\bigg[\sum_{u:C_{u} = 1}f(u) - \sum_{u:C_{u} = 2}f(u)\bigg]}\\ {= \frac{e^{-\lambda_{c}\tau}}{n}\bigg[\sum_{u:C_{u} = 1}(f(u) - f(i)) - \sum_{u:C_{u} = 2}(f(u) - f(j))}\\ {\qquad +\sum_{u:C_{u} = 1}f(i) - \sum_{u:C_{u} = 2}f(j)\bigg].} \end{array} \quad (22)\] + +<|ref|>text<|/ref|><|det|>[[52, 216, 912, 260]]<|/det|> +412 In the sparse regime, we may make a tree- like approximation in the neighbourhood of \(i\) . This means that the number of \(i\) neighbours of \(i\) at distance \(q\) inside the cluster is \(p_{n}^{q}(n / 2)^{q}\) , ignoring terms of order \(O(1 / n)\) and beyond. Considering only \(i\) nodes at unit distance ( \(q = 1\) ), we approximate Eq. (22) as + +<|ref|>equation<|/ref|><|det|>[[295, 264, 702, 512]]<|/det|> +\[\begin{array}{r l} & {\frac{e^{-\lambda_{c}\tau}}{n}\bigg[\sum_{\substack{u:C_{u}=1}}\left(f(u)-f(i)\right)-\sum_{\substack{u:C_{u}=2}}\left(f(u)-f(i)\right)}\\ & {\qquad+\sum_{\substack{u:C_{u}=1}}\left(f(u)-f(j)\right)-\sum_{\substack{u:C_{u}=2}}\left(f(u)-f(j)\right)}\\ & {\qquad+\sum_{\substack{u:C_{u}=1}}\left(f(i)-\sum_{\substack{u:C_{u}=2}}\left(f(i)+\sum_{\substack{u:C_{u}=1}}\left(f(j)-\sum_{\substack{u:C_{u}=2}}\left(f(j)\right)\right)\right)\right)}\\ & {\qquad=\frac{e^{-\lambda_{c}\tau}}{n}\bigg[\sum_{\substack{u:C_{u}=1}}\left(f(u)-f(i)\right)-\sum_{\substack{u:C_{u}=2}}\left(f(u)-f(i)\right)}\\ & {\qquad+\sum_{\substack{u:C_{u}=1}}\left(f(u)-f(j)\right)-\sum_{\substack{u:C_{u}=2}}\left(f(u)-f(j)\right)}\\ & {\qquad+\sum_{\substack{u:C_{u}=1}}\left(f(u)-f(j)\right)-\sum_{\substack{u:C_{u}=2}}\left(f(u)-f(j)\right)}\\ & {\qquad+\frac{n}{2}p_{n}(f(i)-f(j))-\frac{n}{2}p_{n}(f(i)-f(j))\bigg].} \end{array} \quad (22)\] + +<|ref|>text<|/ref|><|det|>[[52, 516, 555, 531]]<|/det|> +415 Then, taking the supremum over all 1- Lipschitz functions \(f\) , we obtain + +<|ref|>equation<|/ref|><|det|>[[364, 538, 910, 602]]<|/det|> +\[\begin{array}{l}{\sum_{i j\in E}\mathcal{W}_{1}(\mathbf{p}_{i}(\tau),\mathbf{p}_{j}(\tau))(1 - \delta (C_{i},C_{j}))}\\ {\approx e^{-\lambda_{c}\tau}(p_{\mathrm{in}} + p_{\mathrm{out}})\left(1 + \frac{|p_{\mathrm{in}} - p_{\mathrm{out}}|}{2(p_{\mathrm{in}} + p_{\mathrm{out}})}\right)} \end{array} \quad (23)\] + +<|ref|>text<|/ref|><|det|>[[52, 607, 686, 623]]<|/det|> +416 Substituting this into Eq. (18) and noting that \(p_{\mathrm{in}} + p_{\mathrm{out}}\) is constant we obtain at a fixed \(\tau\) + +<|ref|>equation<|/ref|><|det|>[[333, 630, 655, 671]]<|/det|> +\[\mathbb{P}(C|\kappa)\propto \exp \left[\left(\frac{|p_{\mathrm{in}} - p_{\mathrm{out}}|}{2(p_{\mathrm{in}} + p_{\mathrm{out}})}\right)\sum_{i\in E}\delta (C_{i},C_{j})\right],\] + +<|ref|>text<|/ref|><|det|>[[52, 678, 580, 693]]<|/det|> +417 which up to a constant of proportionality equals the expression in Eq. (9). + +<|ref|>sub_title<|/ref|><|det|>[[52, 716, 211, 735]]<|/det|> +## 418 References + +<|ref|>text<|/ref|><|det|>[[52, 749, 910, 778]]<|/det|> +419 [1] Tenenbaum, J. B., de Silva, V. & Langford, J. C. A global geometric framework for nonlinear dimensionality reduction. Science 290, 2319- 2323 (2000). + +<|ref|>text<|/ref|><|det|>[[52, 787, 910, 817]]<|/det|> +421 [2] Ding, C., Xiaofeng He, Hongyuan Zha & Simon, H. D. Adaptive dimension reduction for clustering high dimensional data. In 2002 IEEE International Conference on Data Mining, 2002. Proceedings., 147- 154 (2002). + +<|ref|>text<|/ref|><|det|>[[52, 826, 910, 855]]<|/det|> +423 [3] Serrano, M. A., Krioukov, D. & Boguña, M. Self- similarity of complex networks and hidden metric spaces. Phys. Rev. Lett. 100, 078701 (2008). + +<|ref|>text<|/ref|><|det|>[[52, 865, 910, 894]]<|/det|> +425 [4] García- Pérez, G., Boguña, M. & Serrano, M. Á. Multiscale unfolding of real networks by geometric renormalization. Nat. Phys. 1- 9 (2018). + +<|ref|>text<|/ref|><|det|>[[52, 904, 910, 933]]<|/det|> +427 [5] De Domenico, M. Diffusion geometry unravels the emergence of functional clusters in collective phenomena. Phys. Rev. Lett. 118, 168301 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 70, 912, 101]]<|/det|> +[6] Brockmann, D. & Helbing, D. The hidden geometry of complex, network- driven contagion phenomena. Science 342, 1337- 1342 (2013). + +<|ref|>text<|/ref|><|det|>[[50, 108, 653, 125]]<|/det|> +[7] Matousek, J. Lectures on discrete geometry, vol. 212 (Springer New York, 2013). + +<|ref|>text<|/ref|><|det|>[[50, 133, 912, 163]]<|/det|> +[8] Delvenne, J.- C., Yaliraki, S. N. & Barahona, M. Stability of graph communities across time scales. Proc. Natl. Acad. Sci. U.S.A. 107, 12755- 12760 (2010). + +<|ref|>text<|/ref|><|det|>[[50, 172, 787, 189]]<|/det|> +[9] Ollivier, Y. Ricci curvature of Markov chains on metric spaces. J. Funct. Anal. 256, 810- 864 (2009). + +<|ref|>text<|/ref|><|det|>[[50, 197, 715, 214]]<|/det|> +[10] Sturm, K.- T. On the geometry of metric measure spaces. Acta Math. 196, 65- 131 (2006). + +<|ref|>text<|/ref|><|det|>[[50, 222, 912, 238]]<|/det|> +[11] Lott, J. & Villani, C. Ricci curvature for metric- measure spaces via optimal transport. Ann. Math. 169, 903- 991 (2009). + +<|ref|>text<|/ref|><|det|>[[50, 246, 912, 277]]<|/det|> +[12] van der Hoorn, P., Cunningham, W. J., Lippner, G., Trugenberger, C. & Krioukov, D. Ollivier- Ricci curvature convergence in random geometric graphs. Preprint at https://arxiv.org/abs/2008.01209 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 285, 912, 315]]<|/det|> +[13] Jost, J. & Liu, S. Ollivier's ricci curvature, local clustering and curvature- dimension inequalities on graphs. Discrete & Comput. Geom. 51, 300- 322 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 323, 912, 354]]<|/det|> +[14] Bauer, F., Jost, J. & Liu, S. Ollivier- Ricci curvature and the spectrum of the normalized graph Laplace operator. Math. Res. Lett. 19, 1185- 1205 (2012). + +<|ref|>text<|/ref|><|det|>[[50, 362, 912, 393]]<|/det|> +[15] Sandhu, R. S., Georgiou, T. T. & Tannenbaum, A. R. Ricci curvature: An economic indicator for market fragility and systemic risk. Sci. Adv. 2 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 401, 912, 432]]<|/det|> +[16] Farooq, H., Chen, Y., Georgiou, T. T., Tannenbaum, A. & Lenglet, C. Network curvature as a hallmark of brain structural connectivity. Nat. Commun. 10, 4937 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 440, 912, 471]]<|/det|> +[17] Sia, J., Jonckheere, E. & Bogdan, P. Ollivier- Ricci curvature- based method to community detection in complex networks. Sci. Rep. 9, 9800 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 479, 900, 496]]<|/det|> +[18] Ni, C.- C., Lin, Y.- Y., Luo, F. & Gao, J. Community detection on networks with Ricci flow. Sci. Rep. 9, 9984 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 504, 912, 535]]<|/det|> +[19] Decelle, A., Krzakala, F., Moore, C. & Zdeborová, L. Asymptotic analysis of the stochastic block model for modular networks and its algorithmic applications. Phys. Rev. E 84, 066106 (2011). + +<|ref|>text<|/ref|><|det|>[[50, 543, 912, 574]]<|/det|> +[20] Abbe, E. & Sandon, C. Community detection in general stochastic block models: Fundamental limits and efficient algorithms for recovery. In 2015 IEEE 56th Annual Symposium on Foundations of Computer Science, 670- 688 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 582, 912, 628]]<|/det|> +[21] Massoulie, L. Community detection thresholds and the weak Ramanujan property. In Proceedings of the Forty- Sixth Annual ACM Symposium on Theory of Computing, STOC '14, 694- 703 (Association for Computing Machinery, New York, NY, USA, 2014). + +<|ref|>text<|/ref|><|det|>[[50, 636, 863, 653]]<|/det|> +[22] Gfeller, D. & De Los Rios, P. Spectral coarse graining of complex networks. Phys. Rev. Lett. 99, 038701 (2007). + +<|ref|>text<|/ref|><|det|>[[50, 661, 912, 692]]<|/det|> +[23] Coifman, R. R. et al. Geometric diffusions as a tool for harmonic analysis and structure definition of data: Diffusion maps. Proc. Natl. Acad. Sci. U.S.A. 102, 7426- 7431 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 700, 912, 731]]<|/det|> +[24] Arnaudon, A., Peach, R. L. & Barahona, M. Scale- dependent measure of network centrality from diffusion dynamics. Phys. Rev. Res. 2, 033104 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 739, 912, 770]]<|/det|> +[25] Schaub, M. T., Billeh, Y. N., Anastassiou, C. A., Koch, C. & Barahona, M. Emergence of slow- switching assemblies in structured neuronal networks. PLoS Comp Biol 11, e1004196- 28 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 778, 912, 809]]<|/det|> +[26] Young, H. P. Innovation diffusion in heterogeneous populations: Contagion, social influence, and social learning. Am. Econ. Rev. 99, 1899- 1924 (2009). + +<|ref|>text<|/ref|><|det|>[[50, 817, 686, 834]]<|/det|> +[27] Chung, F. R. K. Spectral Graph Theory, vol. 92 (American Mathematical Soc., 1997). + +<|ref|>text<|/ref|><|det|>[[50, 842, 912, 874]]<|/det|> +[28] Veysseire, L. Coarse Ricci curvature for continuous- time Markov processes. Preprint at https://arxiv.org/abs/1202.0420 (2012). + +<|ref|>text<|/ref|><|det|>[[50, 882, 644, 899]]<|/det|> +[29] Villani, C. Optimal transport: old and new (Springer, Berlin Heidelberg, 2009). + +<|ref|>text<|/ref|><|det|>[[50, 907, 870, 924]]<|/det|> +[30] Holland, P. W., Laskey, K. B. & Leinhardt, S. Stochastic blockmodels: First steps. Soc. Netw. 5, 109- 137 (1983). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 68, 914, 101]]<|/det|> +[31] Gosztolai, A., Carrillo, J. A. & Barahona, M. Collective search with finite perception: Transient dynamics and search efficiency. Front. Phys. 6, 153 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 108, 914, 139]]<|/det|> +[32] Kawamoto, T. & Kabashima, Y. Limitations in the spectral method for graph partitioning: Detectability threshold and localization of eigenvectors. Phys. Rev. E 91, 062803 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 146, 914, 177]]<|/det|> +[33] Kay, S. M. Fundamentals of statistical signal processing. Prentice Hall signal processing series (Prentice Hall PTR, Upper Saddle River, NJ, 1993). + +<|ref|>text<|/ref|><|det|>[[50, 185, 911, 202]]<|/det|> +[34] Mossel, E., Neeman, J. & Sly, A. A proof of the block model threshold conjecture. Combinatorica 38, 665- 708 (2018). + +<|ref|>text<|/ref|><|det|>[[50, 210, 914, 241]]<|/det|> +[35] Newman, M. E. J. Modularity and community structure in networks. Proc. Natl. Acad. Sci. U.S.A. 103, 8577- 8582 (2006). + +<|ref|>text<|/ref|><|det|>[[50, 249, 914, 280]]<|/det|> +[36] Blondel, V. D., Guillaume, J.- L., Lambiotte, R. & Lefebvre, E. Fast unfolding of communities in large networks. J. Stat. Mech. Theory Exp. 2008, P10008 (2008). + +<|ref|>text<|/ref|><|det|>[[50, 288, 857, 305]]<|/det|> +[37] PyGenStability: unsupervised clustering with generalised Louvain and Markov Stability. In preparation (2021). + +<|ref|>text<|/ref|><|det|>[[50, 313, 914, 343]]<|/det|> +[38] Rosas- Casals, M., Valverde, S. & Solé, R. V. Topological vulnerability of the European power grid under errors and attacks. Int. J. Bifurcat. Chaos 17, 2465- 2475 (2007). + +<|ref|>text<|/ref|><|det|>[[50, 352, 914, 383]]<|/det|> +[39] Schaub, M. T., Delvenne, J.- C., Yaliraki, S. N. & Barahona, M. Markov dynamics as a zooming lens for multiscale community detection: Non clique- like communities and the field- of- view limit. PLOS One 7, 1- 11 (2012). + +<|ref|>text<|/ref|><|det|>[[50, 391, 914, 421]]<|/det|> +[40] Reilly, M. B., Cros, C., Varol, E., Yemini, E. & Hobert, O. Unique homeobox codes delineate all the neuron classes of C. elegans. Nature 584, 595- 601 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 430, 914, 475]]<|/det|> +[41] Beguerisse- Diaz, M., Vangelov, B. & Barahona, M. Finding role communities in directed networks using role- based similarity, markov stability and the relaxed minimum spanning tree. 2013 IEEE Global Conference on Signal and Information Processing (2013). + +<|ref|>text<|/ref|><|det|>[[50, 483, 914, 513]]<|/det|> +[42] Davis, J. T., Perra, N., Zhang, Q., Moreno, Y. & Vespignani, A. Phase transitions in information spreading on structured populations. Nat. Phys. 1- 8 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 521, 914, 552]]<|/det|> +[43] Al- Mohy, A. H. & Higham, N. J. A new scaling and squaring algorithm for the matrix exponential. SIAM J. Matrix Anal. Appl. 31, 970- 989 (2010). + +<|ref|>text<|/ref|><|det|>[[50, 560, 914, 605]]<|/det|> +[44] Cuturi, M. Sinkhorn distances: Lightspeed computation of optimal transport. In Burges, C. J. C., Bottou, L., Welling, M., Ghahramani, Z. & Weinberger, K. Q. (eds.) Adv. Neural. Inf. Process. Syst. 26, 2292- 2300 (Curran Associates, Inc., 2013). + +<|ref|>text<|/ref|><|det|>[[50, 613, 914, 658]]<|/det|> +[45] Atasu, K. & Mittelholzer, T. Linear- complexity data- parallel earth mover's distance approximations. In Chaudhuri, K. & Salakhutdinov, R. (eds.) Proceedings of the 36th International Conference on Machine Learning, vol. 97 of Proceedings of Machine Learning Research, 364- 373 (PMLR, Long Beach, California, USA, 2019). + +<|ref|>text<|/ref|><|det|>[[50, 666, 755, 682]]<|/det|> +[46] Paulin, D. Mixing and concentration by Ricci curvature. J. Func. Anal. 270, 1623- 1662 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 690, 875, 707]]<|/det|> +[47] Levin, D. A., Peres, Y. & Wilmer, E. L. Markov chains and mixing times (American Mathematical Society, 2006). + +<|ref|>text<|/ref|><|det|>[[50, 715, 914, 746]]<|/det|> +[48] Krivelevich, M. & Sudakov, B. The largest eigenvalue of sparse random graphs. Preprint at https://arxiv.org/math/0106066v1 (2001). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[88, 111, 910, 470]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 477, 913, 616]]<|/det|> +
Supplementary Figure 1: Ricci and dynamical Ollivier Ricci curvature on canonical surfaces and graph structures. a Ricci curvature on a manifold. The geodesic distance of close points \(x\) and \(y\) on average changes when translated by parallel vectors \(\mathbf{v}\) and \(\mathbf{v}^{\prime}\) on the unit circle in the tangent planes at \(x\) and \(y\) . On planes the points remain equidistant, on spheres the points contract and on hyperbolic surface they expand. b Canonical graphs with edges coloured by the dynamical OR curvature (Eq. (2)) for \(\tau = 1\) show that positively and negatively curved graphs have qualitatively different topologies. Away from the boundaries, tree-like topologies are negatively curved, grid-like topologies are flat (zero curvature), whereas clique-like topologies attain positive curvature. Nodes are coloured by the average edge curvature across the neighbours. c Analogously, the differential geometric notion of Ricci curvature is negative on hyperbolic surfaces, zero on planes and positive on spherical surfaces.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[60, 108, 346, 130]]<|/det|> +## Supplementary Note 1 + +<|ref|>text<|/ref|><|det|>[[55, 141, 912, 210]]<|/det|> +In this section, we compute the spectrum of the expected normalised Laplacian matrix of the symmetric SBM \(\mathcal{G}(n,k_{\mathrm{in}} / n,k_{\mathrm{out}} / n)\) . Here \(k_{\mathrm{in}}\) and \(k_{\mathrm{out}}\) are constants representing the expected number of edges within and across clusters, respectively. The expected adjacency matrix of the symmetric stochastic block model is: + +<|ref|>equation<|/ref|><|det|>[[346, 205, 910, 247]]<|/det|> +\[\langle \mathbf{A}\rangle_{\mathcal{G}} = \left(\begin{array}{c c}{\frac{k_{\mathrm{in}}}{n}\mathbf{1}_{n / 2\times n / 2}} & {\frac{k_{\mathrm{out}}}{n}\mathbf{1}_{n / 2\times n / 2}}\\ {\frac{k_{\mathrm{out}}}{n}\mathbf{1}_{n / 2\times \mathrm{n / 2}}} & {\frac{k_{\mathrm{in}}}{n}\mathbf{1}_{n / 2\times n / 2}}\end{array}\right). \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[55, 250, 559, 269]]<|/det|> +Then, the expected normalised Laplacian matrix is given by + +<|ref|>equation<|/ref|><|det|>[[290, 276, 910, 328]]<|/det|> +\[\langle \mathbf{L}\rangle_{\mathcal{G}} = \mathbf{I} - \left(\frac{\frac{2k_{\mathrm{in}}}{n(k_{\mathrm{in}} + k_{\mathrm{out}})}\mathbf{1}_{n / 2\times n / 2}}{\frac{2k_{\mathrm{out}}}{n(k_{\mathrm{in}} + k_{\mathrm{out}})}\mathbf{1}_{n / 2\times n / 2}}\frac{\frac{2k_{\mathrm{out}}}{n(k_{\mathrm{in}} + k_{\mathrm{out}})}\mathbf{1}_{n / 2\times n / 2}}{\frac{2k_{\mathrm{in}}}{n(k_{\mathrm{in}} + k_{\mathrm{out}})}\mathbf{1}_{n / 2\times n / 2}}\right) \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[55, 334, 912, 389]]<|/det|> +The first eigenvector is \(\phi_{1} = \mathbf{1}_{n} / \sqrt{n}\in \mathbb{R}^{n}\) , with the corresponding eigenvalue being \(\lambda_{1} = 0\) . The second eigenvector has two values, one on each cluster of the SBM. Taking \(\phi_{c}(u) = 1 / \sqrt{n}\) for \(1\leq u\leq n / 2\) and \(- 1 / \sqrt{n}\) for \(n / 2< u\leq n\) one has asymptotically + +<|ref|>equation<|/ref|><|det|>[[146, 396, 910, 435]]<|/det|> +\[\langle \mathbf{L}\rangle_{\mathcal{G}}\phi_{c} = \left[1 - \frac{k_{\mathrm{in}}}{k_{\mathrm{in}} + k_{\mathrm{out}}}\frac{2}{n} -\frac{k_{\mathrm{in}}}{k_{\mathrm{in}} + k_{\mathrm{out}}} \frac{2}{n} \left(\frac{n}{2} -1\right) + \frac{k_{\mathrm{out}}}{k_{\mathrm{in}} + k_{\mathrm{out}}} \frac{2}{n} \frac{n}{2}\right]\phi_{c} \xrightarrow{n\to\infty} \frac{2k_{\mathrm{out}}}{k_{\mathrm{in}} + k_{\mathrm{out}}}\phi_{c}. \quad (3)\] + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[72, 105, 927, 456]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[43, 480, 115, 499]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 522, 499, 541]]<|/det|> +Please see the manuscript file to view figure caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 56, 936, 490]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 508, 118, 527]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[44, 551, 499, 570]]<|/det|> +Please see the manuscript file to view figure caption. + +<|ref|>image<|/ref|><|det|>[[80, 625, 914, 812]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 832, 118, 851]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[44, 875, 499, 894]]<|/det|> +Please see the manuscript file to view figure caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[78, 52, 905, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 116, 819]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[44, 842, 499, 861]]<|/det|> +Please see the manuscript file to view figure caption. + +<--- Page Split ---> diff --git a/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/images_list.json b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c051b40792924cda594159a80de6cd51124d51f5 --- /dev/null +++ b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Bulk mechanical behaviour of Clashach sandstone. (a) Differential stress and AE event rate evolution with time. The transition from constant strain rate loading ( \\(10^{-5} / \\mathrm{s}\\) ) to constant AE event rate loading (1 AE /s) occurred early in stage (ii). During stage (iv), shear zone propagation occurs first across the sample, and then down the sample. (b) Photograph of the failed sample showing the localised shear damage zone. (b) Differential stress plotted against axial strain. Letters a–g refer to the image labels in Figs. 2 and 3. Young’s modulus, \\(E = 19.369 \\pm 0.028 \\mathrm{GPa}\\) , was calculated over the range shown. AE activity began at 40% of peak stress, \\(\\sigma_{\\mathrm{P}}\\) , with sample yield at \\(0.85\\sigma_{\\mathrm{P}}\\) . The AE feedback control (1 AE/s) modulated the strain rate from \\(0.93\\sigma_{\\mathrm{P}}\\) .", + "footnote": [], + "bbox": [ + [ + 91, + 198, + 840, + 384 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Micro-scale damage evolution close to and following peak stress. (a-i) Reconstructed 2D \\(\\mu \\mathrm{CT}\\) slice (x,y-oriented, where x and y are perpendicular to each other and to the direction of loading, and x is across-strike and y is along-strike of the shear zone) showing", + "footnote": [], + "bbox": [ + [ + 92, + 123, + 905, + 725 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. 3D incremental strain fields from the onset of strain localisation (marked in Fig. 1c). Incremental dilation, \\(\\Delta \\epsilon_{\\mathrm{d}}\\) (blue-pink) and shear strain, \\(\\Delta \\epsilon_{\\mathrm{s}}\\) (yellow-green) were calculated from digital volume correlation between successive pairs of \\(\\mu \\mathrm{CT}\\) volumes and are shown (a) parallel to strike (y,z orientation) and (b) perpendicular to strike (x,z orientation). The lower threshold of 0.0017 was set at four standard deviations from the mean error distribution of \\(\\Delta \\epsilon_{\\mathrm{s}}\\) (Supplementary Figure S6) and the upper threshold shows regions with strain \\(>0.01\\) (maximum", + "footnote": [], + "bbox": [ + [ + 115, + 99, + 876, + 655 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Relationship between AE amplitude (in log-scale) and local incremental strain at the AE location. (a) Incremental shear strain, \\(\\Delta \\epsilon_{\\mathrm{s}}\\) , and (b) incremental dilation, \\(\\Delta \\epsilon_{\\mathrm{d}}\\) , each normalized by their respective maximum. The solid black ellipses highlight the lack of large AE events at locations of large strain; evidence for predominantly aseismic deformation. Conversely, the dashed black ellipses highlight locations of large strain which are connected to mostly smaller AE events. Larger AE events are mostly associated with smaller strains.", + "footnote": [], + "bbox": [ + [ + 230, + 109, + 740, + 420 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Shear fracture energy estimated from bulk strain and from local incremental strain in the shear zone. (a) Procedure for estimating \\(\\mathrm{G_c}\\) from shear stress and inelastic axial displacement resolved on the fault (reproduced from Fig. 1 in 24) showing (top) schematic of sample with boundary stresses \\(\\sigma_{1}\\) (axial) and \\(\\sigma_{3}\\) (radial), the angle \\(\\theta\\) that fault F makes with \\(\\sigma_{1}\\) , and the shear stress \\(\\tau\\) acting on the fault; (bottom left) obtaining inelastic axial displacement \\(\\Delta \\mathrm{l}\\) from the differential stress vs axial displacement curve and then resolving it onto F to obtain the relative slip \\(\\Delta \\mathrm{u}\\) ; and (bottom right) obtaining \\(\\mathrm{G_c}\\) by integration under \\(\\tau\\) - \\(\\Delta \\mathrm{u}\\) curve from peak shear stress \\(\\tau_{\\mathrm{P}}\\) to residual frictional strength \\(\\tau_{\\mathrm{F}}\\) . (b) Relationship between \\(\\tau\\) and \\(\\Delta \\mathrm{u}\\) for \\(\\theta = 30.3^{\\circ}\\) (Supplementary Table 1) for (i) yellow; uniform slip from average bulk axial stress and strain data, (ii) orange; uniform slip from bulk axial stress and strain data at each \\(\\mu \\mathrm{CT}\\) scan time, and", + "footnote": [], + "bbox": [ + [ + 120, + 90, + 868, + 515 + ] + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01.mmd b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01.mmd new file mode 100644 index 0000000000000000000000000000000000000000..baeeb88cfba73827c3b6116b604fe359a094c2ae --- /dev/null +++ b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01.mmd @@ -0,0 +1,480 @@ + +# Seismic events miss important grain-scale mechanisms governed by kinematics during shear failure of porous rock + +Alexis Cartwright- Taylor ( \(\boxed{\bullet}\) alexis.cartwright- taylor@ed.ac.uk) University of Edinburgh https://orcid.org/0000- 0001- 7805- 4576 + +Maria- Daphne Mangriotis University of Edinburgh + +Ian Main University of Edinburgh + +Ian B. Butler University of Edinburgh + +Florian Fusseis University of Edinburgh + +Martin Ling Edinburgh Hacklab + +Edward Ando Univ. Grenoble Alpes, CNRS, Grenoble INP, 3SR, Grenoble https://orcid.org/0000- 0001- 5509- 5287 + +Andrew Curtis University of Edinburgh + +Andrew Bell University of Edinburgh https://orcid.org/0000- 0002- 5633- 6289 + +Alyssa Crippen University of Edinburgh + +Roberto Rizzo University of Edinburgh + +Sina Marti University of Edinburgh + +Derek Leung University of Edinburgh + +Oxana Magdysyuk Diamond Light Source https://orcid.org/0000- 0003- 3842- 3239 + +Article + +<--- Page Split ---> + +Keywords: micro- seismic events, kinematics, catastrophic failure + +Posted Date: November 4th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 1034813/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on October 18th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33855- z. + +<--- Page Split ---> + +# Seismic events miss important grain-scale mechanisms governed by kinematics during shear failure of porous rock + +A. Cartwright-Taylor1\\*, +M.-D. Mangriotis1, +I. +G. Main1, +I. +B. Butler1, +F. Fusseis1, +M. Ling2, +E. Ando3, +A. Curtis1, +A. +F. Bell1, +A. Crippen1, +R. +E. Rizzo1, +S. Marti1, +D. +D. Leung1, +O. +V. Magdysyuk4 + +## Affiliations + +1School of GeoSciences, University of Edinburgh, Edinburgh, UK. + +2Independent Electronics Developer, Edinburgh Hacklab, Edinburgh, UK. + +3Laboratoire 3SR, Université Grenoble Alpes, Grenoble, France. + +4Beamline I12- JEEP, Diamond Light Source, Didcot, UK. + +\*Correspondence to: alexis.cartwright- taylor@ed.ac.uk + +## Abstract + +Catastrophic failure in brittle, porous materials initiates when smaller- scale fractures localize along an emergent failure plane or 'fault' in a transition from stable crack growth to dynamic rupture. Due to the extremely rapid nature of this critical transition, the precise micro- mechanisms involved are poorly understood and difficult to capture. Here, we observe these micro- mechanisms directly by controlling the microcracking rate to slow down the transition in a unique rock deformation experiment that combines acoustic monitoring with contemporaneous in- situ x- ray imaging of the microstructure. The results provide the first integrated picture of how damage and associated micro- seismic events emerge and evolve together during localisation and failure and allow us to ground truth some previous inferences from mechanical and seismic monitoring alone. They also highlight where such inferences + +<--- Page Split ---> + +miss important kinematically- governed grain- scale mechanisms prior to and during shear failure. Contrary to common assumption, we find seismic amplitude is not correlated with local imaged strain; large local strain often occurs with small acoustic emissions, and vice versa. Local strain is predominantly aseismic, explained in part by grain/crack rotation along an emergent shear zone, and the shear fracture energy calculated from local dilation and shear strain on the fault is half of that inferred from the bulk deformation. This improvement in process- based understanding holds out the prospect of reducing systematic errors in forecasting system- sized catastrophic failure in a variety of applications. + +## Main Text + +Catastrophic failure of porous materials is important for wide range of applications and on a variety of scales, from natural and induced earthquakes to the failure of synthetic materials and engineered structures. Understanding the micro- structural processes that operate during weakening and failure of these materials, and the associated strain partition between seismic and aseismic components, is a critical barrier to reducing the significant uncertainties involved in inferring local deformation and strain rates from remotely accessed field- scale seismic or geodetic data. Without this ability, we are unable to improve methods for forecasting and mitigating the risks associated with catastrophic failure. + +The key driving mechanism of catastrophic failure under compression is the concentration of precursory damage along localised zones of deformation, eventually resulting in system- sized failure along a distinct and emergent sub- planar discontinuity (1). Mean field models explain some aspects of the observed behaviour, but rely on average properties that cannot account for localisation or the detailed micro- mechanics (2). They are often derived solely from the properties of recorded acoustic emissions (AE), which are assumed to be representative of the local strain, and cannot account for local aseismic mechanisms that typically constitute \(>99\%\) (3) of the total accumulated strain energy. The + +<--- Page Split ---> + +partition between seismic and aseismic strain directly constrains the rheology and the detectability of seismic precursors to failure, and is currently the largest source of uncertainty in operational forecasting of seismic risk from subsurface engineering projects (4). However, we have limited constraints on its evolution at a large scale, and none at the microscopic scale. + +The size, location and fracture mode of individual micro- cracking events are commonly inferred from AE waveforms, sometimes calibrated against thin sections that reveal the microscopic processes destructively after the test. These processes include elastic compaction, pore collapse, and the development of small tensile micro- cracks oriented parallel to the maximum principal stress (1, 5, 6). Cracking is a form of local failure either due to locally weaker material or local stress concentrations resulting from changes in geometry (7- 11). Under a constant deformation rate, the peak stress often coincides with the onset of an extremely rapid, non- linear acceleration of AE event rate (5, 12), resulting in violent, abrupt failure and rapid stress drop. The non- linear rheology and short time scales ( \(< 40\) s in Clashach sandstone; Supplementary Figure 3) make the micro- mechanisms involved in the transition from stable crack nucleation along a localised shear zone to system- sized rupture extremely difficult to capture and characterize in real time. However, failure can be extended in duration by controlling the loading rate to maintain a constant AE event rate (1, 13, 14). This procedure prevents the acceleration of crack damage and can extend to minutes or even hours the microscopic processes that usually occur over a few seconds, enabling the post- peak- stress region to be studied under quasi- static conditions. Mapping AE source locations during such an experiment provided the first in- situ view of crack localisation and shear zone growth, and an estimate of the associated shear fracture energy, a key parameter in the mechanics of earthquakes and faulting (1, 13). + +More recently, in- situ high- energy x- ray micro- tomography (μCT) time- lapse imaging of rock deformation experiments has enabled the non- destructive characterization of microstructural damage and local strains (15- 17). Such studies have so far been limited to constant strain rate experiments + +<--- Page Split ---> + +where changes after peak stress occur too rapidly to be captured, even with the high- speed imaging capabilities of a synchrotron. Here, we address this research gap by presenting detailed in- situ images of rock failure obtained during experiments in a novel x- ray transparent triaxial deformation cell (see Methods), which for the first time integrates acoustic monitoring with fast time- lapse synchrotron x- ray imaging, on the beamline I12- JEEP (18) at the Diamond Light Source, UK. We slowed down failure using feedback from the AE event rate (see (1) and Supplementary Note), extended the failure time from \(\sim 1\) minute to \(\sim 50\) minutes (Supplementary Figure 3), and captured sample weakening and failure in a series of eighteen 3D x- ray \(\mu \mathrm{CT}\) volumes (Fig. 1a), along with the locations and amplitudes of the AE sources. This allowed us to unlock the relationship between the seismic sources, the local strain and the associated underlying micro- mechanisms. We find that AE miss important developments on the grain scale; early strain localisation does not necessarily lock the system in; and the micro- mechanics on the grain scale are governed by kinematics, in turn governed by the rock configuration. + +## Micro-mechanics of strain localisation and failure from in-situ x-ray images + +Under AE- controlled quasi- static loading, the sample eventually experienced system- sized brittle failure along a localised shear zone (Fig. 1), with x- ray \(\mu \mathrm{CT}\) volumes of the precursory microscale processes obtained at the times indicated. The details of damage localisation and shear zone development are shown in post- yield 2D \(\mu \mathrm{CT}\) slices (Fig. 2) and incremental 3D strain fields obtained by Digital Volume Correlation (Fig. 3). + +AE activity preceded initial strain localisation during early loading [stage (i) in Fig. 1; Fig. 3 first panel], consistent with previous AE studies on porous rocks using much larger samples (5, 6, 19) or in- situ \(\mu \mathrm{CT}\) imaging of smaller samples (20, 21). Between yield and shortly after peak stress [stage (ii); Fig. 3 panels a- b], the spatial distribution of shear strain closely followed that of dilation, and competing strain clusters localised along three distinct conjugate planes of similar amplitude and dip + +<--- Page Split ---> + +(30° to maximum principal stress; typical of optimally-oriented faults in nature) but variable strike, indicating self-organised exploration of candidate shear zones. These direct observations highlight the exploratory nature of emergent localisation in a complex system. + +In stage (iii), dilation and shear strain concentrated along the critically- oriented shear zone soon after peak stress (Fig. 3 panel c), initially with a brief hiatus in the damage rate (Supplementary Figure 4). The shear zone emerged spontaneously from the self- organised localisation of numerous, narrow en- echelon tensile microcracks, which nucleated simultaneously along the whole length of the shear zone (Fig. 2b and c; slice c) as the upper part of the sample moved as one continuum relative to the lower part. These cracks, which mainly crossed a single whole grain, nucleated at pore edges, at grain contacts where Hertzian stresses accumulated within force chains, and as wing cracks at the tips of sliding grain boundary cracks (consistent with 7- 11). As damage mechanisms localised increasingly on the shear zone, initial diffuse compaction in the bulk was swamped by localised dilation and shear strain (Supplementary Figure 6; Supplementary Movies 3- 8) as a fat tail emerged in their respective frequency- amplitude distributions, which eventually became bimodal (Supplementary Figure 5). + +In stage (iv), dilation and shear strain were highly correlated in the shear zone (Fig. 3), consistent with several mechanisms accommodating bulk shear motion (Fig. 2c; slices c- f). These included en- echelon, tensile cracks nucleating and widening to produce extension and volumetric strain; aseismic rotation of tensile cracks associated with neighboring grain rotation, which prevented tensile crack propagation and supported the walls of the shear zone to maintain finite grain- size thickness throughout failure; and sliding on favourably- and unfavourably- oriented cracks, including some resembling Riedel shear zones. Grain fragmentation generated a proto- cataclasite within the shear zone as whole grains disintegrated, partial grains fractured off the shear zone walls, and fractured grains filled cavities (Fig. 2c; slices d- g). These observations highlight the significant contribution of + +<--- Page Split ---> + +aseismic mechanisms (i.e., rotation) to the overall failure process, relative to seismic mechanisms (i.e., cracking, stick-slip sliding; 6). + +Eventually coherent slip occurred on a contiguous fault 'plane' that reached the sample boundary on all sides (stage v; Fig. 3 panels f- g), with corresponding reduction of local strain amplitude and local strain rate as the main failure mechanism transferred from local breakdown and rotation to coherent slip (Supplementary Figures 4 and 5). All the micro- mechanisms just described remained active in accommodating coherent slip, explaining the remaining patches of dilation (Fig. 3 panels f- g), including rotation of fragments still attached to the upper part of the sample forming local asperities. As asperities broke away from the upper part of the sample, synthetic shear cracks/voids developed along the walls of the shear zone, facilitating further sliding. + +In summary, we observe a breakdown sequence from failure of individual grains due to point loading, through the formation of a proto- cataclastic due to grain rotation and fragmentation, to an ultra- cataclastic due to further grain fragmentation during coherent slip. + +## AE location and the seismic strain partition factor + +We recorded \(\sim 3600\) AE events above ambient noise, some \(5\%\) of which were selected using objective criteria for location analysis (see Methods). The most likely location for each event was inferred by constraining it to the maximum local strain within a kinematically- derived hyperboloid of possible locations (see Methods and Supplementary Figure 11). The same unique location was found for \(85\%\) of the located AE events in both the dilation and shear strain fields due to the high correlation between the two strain fields. Even with the maximum strain constraint, the largest AE events did not occur at locations of high local strain (Fig. 4 solid black ellipses), implying the seismic strain partition coefficient is highly variable in space. Many small events occurred in regions of high local strain (Fig. + +<--- Page Split ---> + +4 dashed black ellipses), consistent with deformation being primarily aseismic in the shear zone, while many large events occurred in regions of low local strain in the bulk (Fig. 4 and Supplementary Figure 12). Dilation and shear strain are positively correlated over the whole experiment, but this positive correlation is much stronger at the AE locations (Supplementary Figure 13), consistent with the large mixed mode component of seismic moment tensors observed in (6). + +We inferred a seismic strain partition coefficient for bulk deformation of \(0.2\%\) from summing the inferred scalar seismic moments (see Methods), verifying that deformation is primarily aseismic. This is much lower than the \(1\%\) inferred by (3), most likely due to using a less brittle material (sandstone) under quasi- static rather than constant strain rate loading. Both estimates are a lower bound due to the finite signal to noise ratio. + +## Rupture energy + +The shear fracture energy or energy release rate, \(\mathrm{G}_{\mathrm{c}}\) , is a crucial parameter for modelling shear fracture propagation (23). It is the energy required per unit area for breakdown processes, such as tensile fracturing, to create the new fault surface. It characterises the strain- softening region of the stress- strain curve and is conventionally estimated from bulk axial stress and strain data (Fig. 5a). Here we estimate \(\mathrm{G}_{\mathrm{c}}\) both from bulk and from knowledge of the local strain in the shear zone (see Methods), finding that bulk estimates are biased to large values. + +We computed \(\mathrm{G}_{\mathrm{c}}\) (Fig. 5c) by integrating the shear stress versus fault slip record (Fig. 5b) for (i) uniform slip on a sample- sized fault from bulk axial stress and strain (as in 24), (ii) uniform slip from bulk axial stress and strain at each \(\mu \mathrm{CT}\) scan time, and (iii) total observed local slip measured directly from local dilation and shear strains within the shear zone. We observed a smaller critical slip distance, \(\Delta \mathrm{u}^{*}\) (Fig. 5a) for local slip (0.14 mm) than for uniform slip (0.2 mm), while the integrated curves + +<--- Page Split ---> + +yielded \(\mathrm{G_{c - i}} = 5.06 \mathrm{kJ / m^2}\) , \(\mathrm{G_{c - ii}} = 3.72 \mathrm{kJ / m^2}\) and \(\mathrm{G_{c - iii}} = 2.53 \mathrm{kJ / m^2}\) , where dilation and shear strains contributed \(36\%\) and \(64\%\) of \(\mathrm{G_{c - iii}}\) respectively. The discrepancy between \(\mathrm{G_{c - i}}\) and \(\mathrm{G_{c - ii}}\) arises from the feedback control: many of the \(\mu \mathrm{CT}\) scan times after peak stress coincided with a reduction in stress and strain as the ram backed off to maintain the AE event rate (Fig. 1a). However, local shear fracture energy, \(\mathrm{G_{c - iii}}\) , is only \(0.68\mathrm{G_{c - ii}}\) and \(0.5\mathrm{G_{c - i}}\) . Hence, bulk estimates of \(\mathrm{G_{c}}\) for uniform slip across a sample- sized fault can significantly exceed those determined directly from local slip measurements in the developing shear zone. + +## Comparison with laboratory observations of shear failure + +The evolving shear zone imaged in the 3D x- ray volumes and strain fields validates many inferences from classic AE experiments, such as the nucleation and growth of a shear zone containing the eventual fault plane due to the spontaneous localisation of en- echelon tensile microcracks (1, 13). These data also provide new insight into the micro- mechanics of strain localisation and shear failure, notably: the grain size control on failure; the self- organised exploration of candidate shear zones close to peak stress; the continued correlation between dilation and shear strain throughout sample weakening; the relative importance of aseismic mechanisms such as crack rotation in accommodating bulk shear deformation; the lack of correlation between locations of large seismic events and regions of high local strain; the very low and locally highly variable seismic strain partition coefficient; and the relatively low local shear fracture energy compared to bulk estimates. + +The strong correlation between local dilation and shear strain is consistent with in- situ \(\mu \mathrm{CT}\) observations up to peak stress (16). Here, we show that this correlation continues throughout quasi- static failure, confirming, at higher resolution, inferences from seismic velocity tomography (25). If volumetric strain is a proxy for tensile cracking, the correlation confirms the existence of a cohesive zone, but with crack damage distributed throughout the shear zone rather than concentrated solely in a + +<--- Page Split ---> + +breakdown zone at the propagating front of a pre- existing discontinuity, as proposed by Barenblatt (26). This observation is consistent with the fact that no contiguous fault exists until very late in the process, and even then cracking continues to occur at the grain scale by fracturing individual grains and breaking remaining grain- scale asperities distributed across the whole shear zone. + +The Gutenberg- Richter \(b\) - value estimate from the AE was \(1.94 \pm 0.04\) (Supplementary Figure 10), slightly higher than \(b \sim 1.7\) estimated for granite under similar loading conditions (1). Both are likely high due to the forced quasi- static failure protocol. When \(b > 1.5\) , the moment release is dominated by smaller events, so our very low estimate for the seismic strain partition factor (0.2%) may be due in part to the lack of detection of smaller events above the relatively high ambient noise present in the synchrotron beamline. However, this does not explain the relative absence of large AE events in high strain regions. The distribution of AE locations reflects the high \(b\) - value, with many moderate/small events occurring in regions of large directly measured strain. One explanation could be that the reduction in local stiffness with increasing damage leads to the preference for smaller events along the shear zone, with fewer but larger AE occurring at locally stiff regions off- fault where strain energy accumulates with relatively little deformation. These results show that the seismic strain partition coefficient is highly variable locally and therefore the AE source amplitude is not necessarily representative of the local strain. This is consistent with similar counterintuitive scaling shown between avalanche size (related to strain) and average energy (related to AE amplitude) in a local load sharing fibre bundle model (27). These findings may help to explain similar spatial variability inferred at field scale and constrain model uncertainty when forecasting seismic risk (4). + +Our estimates of \(\mathrm{G}_{\mathrm{c}}\) are close to that estimated for Berea sandstone under the same assumptions (13), and an order of magnitude smaller than reported for granite (1, 24, 25). Since \(\mathrm{G}_{\mathrm{c - iii}}\) (for local slip in a propagating shear zone) is only 50- 68% of \(\mathrm{G}_{\mathrm{c}}\) for uniform slip on a sample- sized fault, significant additional work ( \(\sim 32 - 50\%\) ) must be done in the rest of the sample. However, the average off- fault to + +<--- Page Split ---> + +on- fault incremental shear and volumetric strain ratios during failure are only \(9\%\) and \(3\%\) respectively (Supplementary Table 1), consistent with ratios of off- fault dissipated energy to on- fault shear fracture energy in granite (25). The remaining \(20 - 38\%\) of additional work must be accommodated by aseismic mechanisms within the shear zone (i.e., crack and grain rotation and silent grain rearrangement). Slip due to rotation for individual cracks can be as large as \(77 \pm 29\%\) of the local relative slip (Supplementary Table 2), and hence likely accounts for a significant proportion of the shear fracture energy. + +## Relating micro-mechanics with natural slip systems and tectonic-scale kinematics + +Our results highlight the complexity of damage processes occurring within shear zones during localisation and through weakening and failure. Local crack rotation with antithetic slip (Fig 2c- iii) offers an additional mechanism for local stress rotation and slip on unfavourably- orientated faults without the need for high pore pressure (28). It may also help to explain observations of interlaced orthogonal faults and accompanying geometric complexities (such as en- echelon faulting, aftershock migration and event triggering from one orientation to another) in the 2019 Ridgecrest earthquake sequence and other examples (29), by providing alternative pathways along which to accommodate strain during a cascading rupture process. + +Our observations of crack rotation provide experimental evidence to support models of tectonic kinematics (e.g., 30- 33), which postulate that rotation in rifting margins and strike- slip settings can emerge as a result of local strength heterogeneity in the crust. In these cases, a strong ‘fixed’ zone (in our case, the rock matrix between en- echelon tensile microcracks) transmits the bulk drag, preventing pure shear and enabling extension and rotation, with rotation facilitated by the surrounding mechanically weak regions (in our case, the tensile cracks themselves) that accommodate the rotation- induced deformation. The rotation in our experiments is driven by shearing oblique to the en- echelon + +<--- Page Split ---> + +tensile cracks, whereby the strong grain segments transmit the shearing motion and prevent tensile crack propagation beyond a single grain, thereby limiting the width of the shear zone to the grain scale. Crack coalescence eventually occurs by grain fragmentation within the shear zone and by the breaking of asperities, forming strands of connected crack porosity aligned along the shear zone in patterns consistent with dynamic shear failure (34), coalescence of en- echelon rift segments (35) and kink band development in granodiorite (36). The potential for slip along either and/or both surfaces of the shear zone, as well as planes within the shear zone, is also found on a larger scale in nature (37). + +This discussion highlights the potential for re- examination of the microstructures and inferred mechanisms associated with larger- scale seismic and aseismic processes in light of the results presented here, in particular the conclusion that seismic events miss important grain- scale mechanisms governed by kinematics before and during shear failure. This improvement in process- based understanding holds out the prospect of reducing systematic errors in forecasting system- sized catastrophic failure in a variety of applications. + +## Methods + +For a description of how we implemented the AE feedback control of deformation, see the Supplementary Note. + +## Experimental material + +A Permian aeolian sandstone (Clashach) was chosen as the experimental material. Clashach is a quartz- rich arenite composed of \(>92\%\) quartz grains, \(< 8\%\) K- feldspar and subordinate lithics. It is well- sorted with fine to medium- sized grains \(0.25 - 0.4 \mathrm{mm}\) in diameter (38). A highly cemented Clashach sample was used (17% porosity), which behaved in a mechanically brittle manner when loaded to failure at shallow crustal pressures, and emitted sufficient acoustic emissions for feedback control. Cylindrical + +<--- Page Split ---> + +cores were obtained using a diamond core drill and the ends of the cores were ground flat and parallel on a lathe. + +## Experimental equipment + +The experiment used our lightweight (3.5 kg) x- ray transparent triaxial deformation apparatus, Stör Mjölnir (Supplementary Figure 1), developed at the University of Edinburgh. Named after Thor's hammer in Norse mythology, it is an up- scaled version of our miniature triaxial cell, Mjölnir (39), with the addition of a linear variable displacement transducer (LVDT) to measure axial displacement, and two piezoelectric 'Glaser- type' transducers, positioned axially (Supplementary Figure 1), to passively detect acoustic emissions (AE) and actively monitor ultrasonic velocities. These sensors (signal in Volts proportional to the true normal displacement, \(u\) , of the received elastic wave) were connected to an Applied Seismology Consulting Ltd (ASC) micro- seismic monitoring system by means of two ASC pre- amplifiers. This allowed us to capture AE waveforms and conduct ultrasonic velocity surveys throughout the experiment. Stör Mjölnir is constructed of grade 5 titanium alloy, with an x- ray "transparent" pressure vessel made of 7068- T6 aluminium alloy. It accommodates cylindrical samples of 10 mm diameter and 25 mm length and can attain confining pressures up to 50 MPa and apply axial stresses up to 500 MPa. It was installed on the x- ray microtomography rotation stage in Experimental Hutch 1 (EH1) of beamline I12- JEEP at the Diamond Light Source. + +## Experimental protocol + +The experiment was conducted at ambient temperature. The Clashach core was jacketed in silicone tubing and sealed between the pistons within the pressure vessel (Supplementary Figure 1). An effective pressure ( \(\mathrm{P}_{\mathrm{eff}}\) ) was applied and maintained at 20 MPa throughout the test (confining pressure \(\mathrm{P}_{\mathrm{c}} = 25 \mathrm{MPa}\) and pore fluid pressure \(\mathrm{P}_{\mathrm{p}} = 5 \mathrm{MPa}\) , where \(\mathrm{P}_{\mathrm{eff}} = \mathrm{P}_{\mathrm{c}} - \mathrm{P}_{\mathrm{p}}\) . A hydrostatic starting pressure condition (zero differential stress) was achieved by simultaneously increasing the axial pressure to match the confining pressure. Tomographic reference scans were acquired before pressurisation and + +<--- Page Split ---> + +again prior to loading (at zero differential stress) to obtain the initial state of the sample. Two scans were acquired at zero differential stress to characterize the error in the digital volume correlation by correlating two volumes in which the state of the sample was identical (Supplementary Figure 7). + +After the initial scans, the sample was loaded continuously at a constant deformation rate of \(10^{- 5}\) \(\mathrm{s}^{- 1}\) until the desired AE event rate (1 event/s) was reached, at which point loading at this constant AE rate took over. This protocol extended the failure time from \(\sim 1\) minute to 50 minutes, equivalent to an average bulk strain rate of \(10^{- 7} \mathrm{s}^{- 1}\) , sufficient to capture 18 high- quality \(\mu \mathrm{CT}\) volumes after peak stress. The sample underwent triaxial deformation to failure evident from a rollover at peak differential stress followed by a gentle decrease in stress ending in an almost constant stress on completion of loading (Fig. 1a and Supplementary Figure 3). + +Imaging the deformation in- situ on beamline I12- JEEP (18) was achieved with a 53 keV monochromatic beam detected by a PCO. edge light sensor with I12 in- house optical module of 7.91 x \(7.91 \mu \mathrm{m}\) /pixel resolution and \(20 \mathrm{mm} \times 12 \mathrm{mm}\) field of view. Tomographic volumes of the whole sample, comprising two discrete overlapping scans of the top and bottom of the sample, with vertical translation in between, were acquired every 85 s throughout the experiment. Individual scans were acquired in approx. 40 seconds and consisted of 1800 projections with a 0.0035s exposure time. At each end of the sample, 0.2 mm (next to the piston) was not captured due to limits on the x- ray field of view, which had a vertical dimension of \(12 \mathrm{mm}\) . + +Ultrasonic velocity surveys were conducted every 5 minutes and acoustic emissions monitored continuously throughout the experiment, along with actuator pressure and axial displacement, confining pressure, and pore fluid pressure and volume. + +Tomographic reconstruction + +<--- Page Split ---> + +The tomographic dataset was fully reconstructed on the Diamond cluster using the framework SAVU (40) with automatic centre detection (41), ring removal (42) and distortion correction (43), a Paganin phase filter with delta/beta ratio of 0.6 (44) and a 2D GPU- accelerated filtered backprojection ('FBP_CUDA') reconstruction algorithm using the integrated ASTRA toolbox (45- 48). This yielded 16- bit image volumes of \(2560 \times 2560 \times 1800\) voxels, with a voxel size of \(7.91 \times 7.91 \times 7.91 \mu \mathrm{m}^3\) . Each pair of reconstructed volumes comprising both ends of the sample, acquired with an overlapping region of 520 vertical pixels (4 mm), was merged together using digital volume correlation (DVC) to match the overlapping ends and generate a time- series of tomographic volumes of the full sample (orange dots in Fig. 1 and blue dots in Supplementary Figure 3). + +## Estimation of bulk stress and strain + +Axial displacement was converted to axial strain, \(\epsilon\) , which was then corrected for rig stiffness, \(\mathrm{k}_{\mathrm{rig}}\) to obtain the axial sample strain, \(\epsilon\) , as follows: \(\epsilon = \epsilon /\sigma - \sigma (1 / \mathrm{k}_{\mathrm{rig}})\) . Differential stress, \(\sigma\) , on the sample was obtained from \(\sigma = \sigma_{1} - \sigma_{3}\) , where \(\sigma_{1}\) is the axial stress, equal to the load, F, on the sample (obtained from a calibration of the measured ram pressure with a load cell) divided by the surface area, A, of the sample end \((\sigma_{1} = \mathrm{F / A})\) , and \(\sigma_{3} = \mathrm{P}_{\mathrm{eff}}\) . + +## Dynamic wave velocity estimation + +Changes in ultrasonic compressional- wave velocity were measured by means of the pulse transmission technique (49) between the pair of AE transducers, with source- receiver geometries at opposite ends of the sample (Supplementary Figures 1 and 11). Velocity surveys were conducted by the ASC Pulser Interface Unit at constant time intervals (5 minutes) during the experiment. A pulse of acoustic energy was sent through the sample from each transducer in turn and its arrival at the other sensor was recorded. To understand the system delay (including travel- time through pistons) a seismic test using the pistons without any sample in between them was recorded, and the difference in time between pulse onset at the origin and pulse onset retrieval at the other piston was calculated as the time- correction. + +<--- Page Split ---> + +Detected AE signals were amplified and sent to the two- channel monitoring system, which recorded the full event waveforms, including arrival time, amplitude and first- motion information. The pre- amplification and trigger threshold (based on an instantaneous amplitude) were set at 70 dB and 280 mV respectively, with the aim of controlling the AE event rate effectively. The gain was determined from a benchmark pencil lead- break test in the laboratory prior to visiting the synchrotron, and the trigger threshold was determined by the ambient noise in the experimental hutch on the I12- JEEP beamline. AE event rates were calculated from the number of events recorded in each 10 s time window. AE amplitudes as a function of time, AE event rate and frequency- magnitude distributions are shown in Supplementary Figure 10. + +We interrogated the AE events to obtain a sub- set of the most reliable events for location analysis, ensuring that: (i) individual events were represented in both sensors' recordings, (ii) the AE waveforms satisfied a minimum Signal to Noise Ratio (SNR), (iii) the relative time- delay for a specific AE recorded at the two sensors was robust against noise and yielded a depth uncertainty less than 1 pixel, (iv) the maximum relative time- delay for a specific AE recorded at the two sensors obeyed physical limits given sample dimensions/velocity. The relative time- delays between top and bottom sensor were estimated by cross- correlation of the top and bottom waveforms. Given the limitation of only two sensors at the opposite ends of the rock sample, the experimental geometry led to inherent ambiguity in AE location. However, the specific relative time- delay of an AE event, and knowledge of the dynamic velocity allowed us to define a circular hyperboloid, which marks a 3D surface of constant relative time- delay (Supplementary Figure 11) and therefore a kinematic constraint for AE location. + +Using the AE events that passed the selection criteria ( \(\sim 5\%\) of 3600 events) we defined each hyperboloid as the locus of potential elementary volumes (with thickness \(= 1\) pixel) for the AE location. Time- binning of AE relative to the occurrence of the two scans used to obtain a particular + +<--- Page Split ---> + +strain increment from the DVC permitted comparison of AE with the local incremental strains. Assuming that each AE occurred at the largest incremental local strain within the kinematically- defined hyperboloid, a unique AE location was established (location constrained by local strain). Given the large correlation between locations of large volumetric strain vs. deviatoric strain, AE locations were defined within the same pixel whether constrained by volumetric or deviatoric strain for \(85\%\) of the located AE. Where two or more AE had the same time- bin and the same kinematic constraint, we assumed that the largest AE would be correlated with the largest strain, effectively assigning AE locations sequentially from largest to lowest AE within a time- bin. Whilst the assumption of linking AE to the largest local strain should bias AE locations to be linked to larger local strains, the results showed that, effectively via proof by contradiction, the largest AEs do not occur at locations of large local strain, with the locations of large local strains being linked to small/moderate AE (Fig. 4). This observation, in turn, is evidence that deformation is primarily aseismic. + +Local incremental 3D strain fields from digital volume correlation between neighboring scans + +In addition to observing quasi- static fault development directly in the reconstructed \(\mu \mathrm{CT}\) volumes, we quantified the full, incremental 3D strain tensor evolution using DVC between adjacent pairs of tomograms using the open- source software SPAM (50). The procedure involved computing linear displacement vectors between nodes (with 40 pixel spacing, equivalent to \(316.4 \mu \mathrm{m}\) ) identified in the reference image and the same nodes identified in the deformed image by tracking a representative volume (window size 40 pixels) around the nodes. The transformation gradient tensor (local derivatives of displacement), \(\mathbf{F}\) , was then computed using a Q8 shape function linking 8 neighbouring nodes, with \(\mathbf{F}\) computed in the centre of the element. \(\mathbf{F}\) was decomposed into the symmetric stretch tensor, \(\mathbf{U}\) , which was further decomposed into an isotropic and a deviatoric part. The volumetric strain (first invariant of the strain tensor showing dilation/compaction) was defined as the determinant of the transformation gradient tensor, \(\left| \mathbf{F} \right| - 1\) . Here, we followed soil mechanics convention and defined + +<--- Page Split ---> + +dilation as positive volumetric strain and compaction as negative volumetric strain. The deviatoric strain (second invariant of the strain tensor showing shear deformation and referred to throughout as the shear strain) was defined as the Euclidean norm of the deviatoric part \(\| \mathbf{U}_{dev}\|\) . + +Estimation of the overall seismic strain partition factor + +To estimate the seismic strain partition factor, we followed the approach of (3), who estimated the seismic moment of the largest observed AE event based on the largest observed fault, in order to scale the distribution of seismic moments from all their events. We used (51) to obtain the scalar seismic moments \(M_{o} = CV\Delta \sigma\) , with \(C = 16 / 7\) , \(V = r^{3}\) the crack volume and \(\Delta \sigma\) the stress drop. The estimated maximum for the scalar seismic moment was 1.5e- 3 Nm, using the Madariaga source model (52) with a radius, \(r\) , equal to one- half the maximum observed crack size ( \(2r \sim 800 \mu \mathrm{m}\) ) and a stress drop of 1 MPa. The latter stress drop is an intermediate stress drop encompassing 23 different studies of natural, mining induced, and fracking induced seismicity and laboratory AE events (Supplementary Figure 13; 53, 54). The maxima of the AE envelopes were scaled to the maximum scalar seismic moment to obtain the scalar seismic moment distribution. The Kostrov (55) strain was estimated from \(\Delta \epsilon_{ij} = \frac{1}{2\mu\Delta V}\sum_{i = 1}^{N}M_{ij}\) , with \(M_{ij} = \sqrt{2M_{o}U_{ij}}\) (56), where \(U_{ij}\) is the unit displacement tensor, \(\mu\) the shear modulus and \(\Delta V\) the total representative volume. Finally, the strain partition factor was estimated from \(\chi = \frac{\gamma_{AE}}{\gamma_{F}}\) (57, 58, 3), with \(\gamma_{AE}\) representing the sum of scalar seismic moments multiplied by \(\frac{\sqrt{2}}{2\mu\Delta V}\) and \(\gamma_{F}\) the average volumetric and deviatoric strains accumulated throughout the deformation experiment. + +Estimation of bulk and local shear fracture energy + +<--- Page Split ---> + +The breakdown zone (slip- weakening) model (Fig. 5a; 23, 24) considers a fault as a shear crack with peak shear strength, \(\tau_{\mathrm{P}}\) . Relative slip on the fault, \(\Delta \mathrm{u}\) , initiates at \(\tau_{\mathrm{P}}\) and then strength degrades from \(\tau_{\mathrm{P}}\) to a constant residual frictional strength, \(\tau_{\mathrm{F}}\) , as \(\Delta \mathrm{u}\) increases to a value \(\Delta \mathrm{u}^{*}\) , the critical slip- weakening distance, where tensile fracturing (mode I; dilation) transitions to frictional sliding (mode II; shear). This leads to a breakdown zone of dimension \(\omega_{0}\) at the shear crack tip. \(\mathrm{G}_{\mathrm{c}}\) for breakdown processes at the crack tip can be evaluated by integrating under the \(\tau\) - \(\Delta \mathrm{u}\) curve (23; Fig. 5a). \(\Delta \mathrm{u}\) is resolved from the bulk inelastic axial displacement, \(\Delta \mathrm{l}\) , and depends on the angle, \(\theta\) , which the fault makes with the direction of loading. Shear stress depends on \(\theta\) and the boundary stress conditions. Three main assumptions are involved: all axial strain beyond peak stress is accommodated by shear slip on the fault, the fault is sample- sized throughout, and \(\omega_{0}\) is small compared with the size of the fault. + +To obtain local geometric and strain information for the shear zone itself defined by the local strain fields that were output from the DVC (Fig. 3 and Supplementary Figure 4), we separated the shear zone core from other correlated strain clusters by thresholding above 0.004 strain. We then used a connected component object analysis to obtain the best- fitting ellipsoid for the shear zone at each strain increment, and hence obtained the angle \(\theta\) of the shear zone with respect to the direction of loading (Supplementary Table 1) from the ellipsoid eigenvectors. The evolving shear zone object is shown in Supplementary Figure 9. + +Following (23, 24), we then estimated \(\mathrm{G}_{\mathrm{c}} = \int_{0}^{\Delta \mathrm{u}^{*}}[\tau (\Delta \mathrm{u}) - \tau_{\mathrm{F}}]\mathrm{d}(\Delta \mathrm{u})\) for: + +(i) Uniform slip on a sample-sized fault (as in 24) by first resolving relative slip, \(\Delta \mathrm{u}\) , from the bulk inelastic axial displacement, \(\Delta \mathrm{l}\) , and \(\theta\) as follows: \(\Delta \mathrm{u} = \Delta \mathrm{l} / \cos \theta\) , and then calculating shear stress, \(\tau = [(\sigma_{1} - \sigma_{3}) / 2] * \sin (2\theta)\) , from \(\theta\) and the boundary stress conditions. A cubic fit was used to obtain an average bulk shear stress vs relative slip curve and we assumed that all axial shortening was caused by slip along the fault from \(\tau_{\mathrm{P}}\) . Axial shortening was corrected for elastic strain using the intact Young's modulus for the sample (Figs. 1c and 5a). + +<--- Page Split ---> + +(ii) Uniform slip at each \(\mu \mathrm{CT}\) scan time, using the bulk shear strain and relative slip measurements at those times (rather than an average), in order to compare the uniform slip model directly with the observed local slip on the shear zone itself (estimated from the 3D strain fields). We assumed that all axial shortening was caused by slip along the fault from the point of final localization (scan c). + +(iii) Total observed local slip on the shear zone itself from local dilation and shear strain. We assumed that slip started from the point of final localization (scan c) and used the bulk shear strain calculated in (ii). For the shear contribution to slip, we extracted the mean local incremental shear strain in the shear zone object, \(\Delta \epsilon_{\mathrm{shear}}\) , from the connected component object analysis described above. We corrected \(\sum \overline{\Delta\epsilon_{\mathrm{shear}}}\) for elastic strain using the intact shear modulus of the sample, obtained from the shear stress vs \(\sum \overline{\Delta\epsilon_{\mathrm{shear}}}\) curve (Supplementary Figure 8). Finally, we obtained local shear slip, \(\Delta u_{\mathrm{shear}}\) , from an engineering strains approximation: \(\Delta u_{\mathrm{shear}} = 2 * \sum \overline{\Delta\epsilon_{\mathrm{shear}}} * \mathrm{DVC}\) node spacing (59). For the radial dilation contribution to slip, we extracted the mean local incremental radial dilation in the shear zone object, \(\overline{\Delta\epsilon_{\mathrm{dilaton}}}\) , and corrected \(\sum \overline{\Delta\epsilon_{\mathrm{dilaton}}}\) for elastic radial strain using the elastic axial strain calculated above and an estimate of Poisson's ratio for Clashach sandstone (60). We then resolved the inelastic dilation onto the shear zone orientation, obtaining local dilational slip from an engineering strains approximation: \(\Delta u_{\mathrm{dilaton}} = \sum (\overline{\Delta\epsilon_{\mathrm{dilaton}}} /\sin \theta) * \mathrm{DVC}\) node spacing. + +## Acknowledgments + +We would like to thank the University of Edinburgh Geosciences Workshop for their support in developing the experimental apparatus, and Jonathan Singh for useful discussions about data file + +<--- Page Split ---> + +formats and estimation of acoustic waveform arrival times. We also acknowledge Diamond Light Source for time on beamline I12- JEEP under proposal MG22517. This work is supported by the UK's Natural Environment Research Council (NERC) through the CATFAIL project NE/R001693/1 Catastrophic failure: what controls precursory localisation in rocks? + +## Author contributions + +Conceptualization (A. Bell, I. Butler, A. Cartwright- Taylor, A. Curtis, F. Fusseis, I. Main), data curation (A. Cartwright- Taylor, M. Mangriotis), formal analysis (A. Cartwright- Taylor, M. Mangriotis, E. Andò, A. Bell, A. Crippen), funding acquisition (A. Bell, I. Butler, A. Cartwright- Taylor, A. Curtis, F. Fusseis, I. Main), investigation (I. Butler, A. Cartwright- Taylor, F. Fusseis, D. Leung, O. V. Magdysyuk, S. Marti, R. Rizzo), methodology (E. Andò, M. Ling), project administration (A. Cartwright- Taylor, I. Main), resources (I. Butler, E. Andò, M. Ling, O. V. Magdysyuk), software (E. Andò, M. Ling), supervision (I. Main, A. Curtis, A. Bell, I. Butler, F. Fusseis), validation (A. Cartwright- Taylor, M. Mangriotis, I. Main, A. Curtis.), writing – original draft (A. Cartwright- Taylor, M. Mangriotis, I. Main, M. Ling), writing – review and editing (A. Bell, I. Butler, A. Curtis, F. Fusseis, O. V. Magdysyuk, R. Rizzo). + +## Statement of competing interests + +The authors declare no competing interests. + +## Data and materials availability + +The data supporting our conclusions can be found in main text, in the supplementary information, and in the \(\mu \mathrm{CT}\) data sets held at the NERC repository (link tbc). When using the \(\mu \mathrm{CT}\) data sets, please cite (dataset publication tbc) and see (link tbc) for the metadata. + +<--- Page Split ---> + +## References + +1. D. Lockner, J. D. Byerlee, V. Kuksenko, A. Ponomarev, A. Sidorin, Quasi-static fault growth and shear fracture energy in granite. Nature 350 39-42 (1991). + +2. I. Main, A damage mechanics model for power-law creep and earthquake aftershock and foreshock sequences. Geophys. J. Int. 142, 151-161 (2000). + +3. G. Dresen, G. Kwiatek, T. Goebel, Y. Ben-Zion, Seismic and aseismic preparatory processes before large stick-slip failure. Pure Appl. Geophys. 177, 5741-5760 (2020). + +4. S. Bourne, S. J. Oates, J. van Elk, D. Doornhof, A seismological model for earthquakes induced by fluid extraction from a subsurface reservoir. J. Geophys. Res. 119, 8991-9015 (2014). + +5. P. R. Sammonds, P. G. Meredith, I. G. Main, Role of pore fluids in the generation of seismic precursors to shear fracture. Nature 359, 228-230 (1992). + +6. C. C. Graham, S. Stanchits, I. G. Main, G. Dresen, Comparison of polarity and moment tensor inversion methods for source analysis of acoustic emission data. Int. J. Rock Mech. Min. Sci. 47, 161-169 (2010). + +7. N. Cho, C. D. Martin, D. S. Sego, A clumped particle model for rock. Int. J. Rock Mech. Min. Sci. 44(7), 997-1010 (2007). + +8. D. O. Potyondy, P. A. Cundall, A bonded particle model for rock. Int. J. Rock Mech. Min. Sci. 41(8), 1329-1364 (2004). + +9. C. G. Sammis, M. F. Ashby, The failure of brittle porous solids under compressive stress states. Acta Metall. 34, 511-526 (1986). + +10. M. F. Ashby, S. D. Hallam, The failure of brittle solids containing small cracks under compressive stress states. Acta Metall. 34, 497-510 (1986). + +<--- Page Split ---> + +11. H. Hori, S. Nemat-Nasser, Compression-induced microcrack growth in brittle solids: axial splitting and shear failure. J. Geophys. Res. 90(B4), 3105-3125 (1985). + +12. J. Vasseur, F. B. Wadsworth, Y. Lavallée, A. Bell, I. G. Main, D. B. Dingwell, Heterogeneity: The key to failure forecasting. Sci. Rep. 5, 13259 (2015). + +13. D. Lockner, J. D. Byerlee, V. Kuksenko, A. Ponomarev, A., Sidorin, "Observations of quasi-static fault growth from acoustic emissions" in Fault Mechanics and Transport Properties of Rocks, B. Evans, T.-F. Wong, Eds. (Academic Press, 1992), vol. 51, chap. 1, pp. 3-31. + +14. M. Terada, T. Yanagidani, S. Ehara, "AE rate controlled compression test of rocks" in 3rd Conf. on Acoustic Emission / Microseismic Activity in Geologic Structures and Materials, H. R. Hardy, F. W. Leighton, Eds. (Trans Tech, 1984), pp. 159-171. + +15. L. Huang, P. Baud, B. Cordonnier, F. Renard, L. Liu, T.-F. Wong, Synchrotron X-ray imaging in 4D: multiscale failure and compaction localization in triaxially compressed porous limestone. Earth Planet. Sci. Lett. 528, 115831 (2019). + +16. J. McBeck, Y. Ben-Zion, F. Renard, The mixology of precursory strain partitioning approaching brittle failure in rocks. Geophys. J. Int. 221, 1856-1872 (2020). + +17. A. Cartwright-Taylor, I. G. Main, I. B. Butler, F. Fusseis, M. Flynn, A. King, Catastrophic failure: how and when? Insights from 4-D in situ x-ray microtomography. J. Geophys. Res. Solid Earth 125, e2020JB019642 (2020). + +18. M. Drakopoulos, T. Connolley, C. Reinhard, R. Atwood, O. Magdysyuk, N. Vo, M. Hart, L. Connor, B. Humphreys, G. Howell, S. Davies, T. Hill, G. Wilkin, U. Pedersen, A. Foster, N. De Maio, M. Basham, F. Yuan, K. Wanelik, I12: the Joint Engineering, Environment and Processing (JEEP) beamline at Diamond Light Source. J. Synchrotron Rad. 22, 828-838 (2015). + +<--- Page Split ---> + +19. I. O. Ojala, I. G. Main, B. T. Ngwenya, Strain rate and temperature dependence of Omori law scaling constants of AE data: implications for earthquake foreshock-aftershock sequences. Geophys. Res. Lett. 31, L24617 (2004). + +20. F. Renard, J. McBeck, B. Cordonnier, X. Zheng, N. Kandula, J. R. Sanchez, M. Kobchenko, C. Noiriel, W. Zhu, P. Meakin, F. Fussies, D. K. Dysthe, Dynamic in situ three-dimensional imaging and digital volume correlation analysis to quantify strain localization and fracture coalescence in sandstone. Pure Appl. Geophys. 176, 1083-1115 (2018). + +21. F. Renard, J. McBeck, N. Kandula, B. Cordonnier, P. Meakin, Y. Ben-Zion, Volumetric and shear processes in crystalline rock approaching faulting. Proc. Natl. Acad. Sci. 116, 16234-16239 (2019). + +22. S. Lennartz-Sassinek, I.G. Main, M. Zaiser, C. Graham, Acceleration and localization of subcritical crack growth in a natural composite material. Phys. Rev. E 90, 052401 (2014). + +23. J. R. Rice, "The mechanics of earthquake rupture" in Physics of the Earth's Interior, A. M. Dziewonski, E. Boschi, Eds. (Italian Phys. Soc. 1980), pp. 555-649. + +24. T.-F. Wong, Shear fracture energy of Westerly granite from post-failure behaviour. J. Geophys. Res. 87, 990-1000 (1982). + +25. F. M. Aben, N. Brantut, T. M. Mitchell, E. C. David, Rupture energetics in crustal rock from laboratory-scale seismic tomography. Geophys. Res. Lett. 46, 7337-7344. + +26. G. I. Barenblatt, The mathematical theory of equilibrium cracks in brittle fracture. Adv. Appl. Mech. 7, 55-129 (1962). + +27. S. Roy, S. Biswas, Size distribution of emitted energies in local load sharing fiber bundles. Front. Phys. 9, 643602 (2021). + +28. D. R. Faulkner, T. M. Mitchell, D. Healy, M. J. Heap, Slip on 'weak' faults by the rotation of regional stress in the fracture damage zone. Nature 444, 922-925 (2006). + +<--- Page Split ---> + +29. Z. E. Ross, B. Idini, Z. Jia, O. L. Stephenson, M. Zhong, X. Wang, Z. Zhan, M. Simons, E. J. Fielding, S.-H. Yun, E. Hauksson, A. W. Moore, Z. Liu, J. Jung, Hierarchical interlocked orthogonal faulting in the 2019 Ridgecrest earthquake sequence. Science 366, 346-351 (2019). + +30. T. Taymaz, J. Jackson and D. McKenzie, Active tectonics of the north and central Aegean Sea. Geophys. J. Int. 106, 433-490 (1991). + +31. H. Schouten, K. D. Klitgord, D. G. Gallo, Edge-driven microplate kinematics. J. Geophys. Res. 98, 6689-6701 (1993). + +32. A. Glerum, S. Brune, D. S. Stamps, M. R. Strecker, Victoria continental microplate dynamics controlled by the lithospheric strength distribution of the East African Rift. Nat. Comm. 11, 2881 (2020). + +33. G. Duclaux, R. S. Huismans, D. A. May, Rotation, narrowing, and preferential reactivation of brittle structures during oblique rifting. Earth Planet. Sci. Lett. 531, 115952 (2020). + +34. R. E. Rizzo, D. Healy, M. J. Heap, N. J. Farrell, Detecting the onset of strain localization using two-dimensional wavelet analysis on sandstone deformed at different effective pressures. J. Geophys. Res. Solid Earth 123, 10460-10478 (2018). + +35. G. Corti, R. Cioni, Z. Franceschini, F. Sani, S. Scaillet, P. Molin, I. Isola, F. Mazzarini, S. Brune, D. Keir, A. Erbello, A. Muluneh, F. Illsley-Kemp, A. Glerum. Aborted propagation of the Ethiopian rift caused by linkage with the Kenyan rift. Nat. Comm. 10, 1309 (2019). + +36. R. K. Davies, D. D. Pollard, Relations between left-lateral strike-slip faults and right-lateral monoclonal kink bands in granodiorite, Mt. Abbot Quadrangle, Sierra Nevada, California. Pure Appl. Geophys. 124, 177-201 (1986). + +37. T. M. Mitchell, D. R. Faulkner, Towards quantifying the matrix permeability of fault damage zones in low porosity rocks. Earth Planet. Sci. Lett. 339-340, 24-31 (2012). + +<--- Page Split ---> + +38. N. J. C. Farrell, D. Healy, Anisotropic pore fabrics in faulted porous sandstones. J. Struct. Geol. 104, 125-141 (2017). + +39. I. B. Butler, F. Fusseis, A. Cartwright-Taylor, M. Flynn, Mjölnir: a miniature triaxial rock deformation apparatus for 4D synchrotron x-ray micro-tomography. J. Synchrotron Rad. 27, 1681-1687 (2020). + +40. N. Wadeson, M. Basham, Savu: a Python-based, MPI framework for simultaneous processing of multiple, N-dimensional, large tomography datasets. arXiv:1610.08015 (2016). + +41. N. T. Vo, M. Drakopoulos, R. C. Atwood, C. Reinhard, Reliable method for calculating the center of rotation in parallel-beam tomography. Opt. Express 22, 19078-19086 (2014). + +42. N. T. Vo, M. Drakopoulos, R. C. Atwood, Superior techniques for eliminating ring artifacts in X-ray micro-tomography. Opt. Express 26, 28396-28412 (2018). + +43. N. T. Vo, R. C. Atwood, M. Drakopoulos, Radial lens distortion correction with sub-pixel accuracy for X-ray micro-tomography. Opt. Express 23, 32859-32868 (2015). + +44. D. Paganin, S. C. Mayo, T. E. Gureyev, P. R. Miller, S. W. Wilkins, Simultaneous phase and amplitude extraction from a single defocused image of a homogeneous object. J. Microsc. 206, 33-40 (2002). + +45. D. Pelt, D. Gürsoy, W. J. Palenstijn, J. Sijbers, F. De Carlo, K. J. Batenburg, Integration of tomopy and the astra toolbox for advanced processing and reconstruction of tomographic synchrotron data. J. Synchrotron Rad. 23, 842-849 (2016). + +46. W. van Aarle, W. J. Palenstijn, J. De Beenhouwer, T. Altantzis, S. Bals, K. J. Batenburg, J. Sijbers, The ASTRA toolbox: a platform for advanced algorithm development in electron tomography. + +Ultramicroscopy 157, 35-47 (2015). + +<--- Page Split ---> + +47. W. van Aarle, W. J. Palenstijn, J. Cant, E. Janssens, F. Bleichrodt, A. Dabravolski, J. De Beenhouwer, K. J. Batenburg, J. Sijbers, Fast and flexible x-ray tomography using the ASTRA toolbox. Opt. Express 24, 25129-25147 (2016). + +48. W. J. Palenstijn, K. J. Batenburg, J. Sijbers, Performance improvements for iterative electron tomography reconstruction using graphics processing units (GPUs). J. Struct. Biol. 176, 250-253 (2011). + +49. F. Birch, The velocity of compressional waves in rocks to 10 kilobars, part 1. J. Geophys. Res. 65, 1083-1102 (1960). + +50. O. Stamati, E. Andò, E. Roubin, R. Cailletaud, M. Wiebicke, G. Pinzon, C. Couture, R. C. Hurley, R. Caulk, D. Caillerie, T. Matsushima, P. Bésuelle, F. Bertoni, T. Arnaud, A. O. Laborin, R. Rorato, Y. Sun, A. Tengattini, O. Okubadejo, J.-B. Colliat, M. Saadatfar, F. E. Garcia, C. Papazoglou, I. Vego, S. Brisard, J. Dijkstra, G. Birmpilis, SPAM: software for practical analysis of materials. J. Open Source Softw. 5(51), 2286 (2020). https://ttk.gricad-pages.univ-grenoble-alpes.fr/spam/index.html. + +51. H. Kanamori, D. L. Anderson, Theoretical basis of some empirical relations in seismology. Bull. Seismol. Soc. Am. 65, 1073-1095 (1975). + +52. R. Madariaga, Dynamics of an expanding circular fault, Bull. Seismol. Soc. Am. 66(3), 639-666 (1976). + +53. A. B. Blanke, G. Kwiatek, T. H. W. Goebel, M. Bohnhoff, G. Dresen, Stress drop-magnitude dependence of acoustic emissions during laboratory stick-slip. Geophys. J. Int. 224, 1371-1380 (2021). + +<--- Page Split ---> + +54. G. Kwiatek, K. Plenkers, G. Dresen, JAGUARS Research Group, Source parameters of pico-seismicity recorded at Mponeng Deep Gold Mine, South Africa: implications for scaling relations, Bull. Seismol. Soc. Am. 101(6), 2592–2608 (2011). + +55. V. Kostrov, Seismic moment and energy of earthquakes and seismic flow of rock. Earth Phys. 1, 23-40 (1974). + +56. P. G. Silver, T. H. Jordan, Optimal estimation of scalar seismic moment. Geophys. J. Int. 70(3), 755-787 (1982). + +57. C. H. Scholz, J. Campos, The seismic coupling of subduction zones revisited. J. Geophys. Res. 117, B05310 (2012). + +58. G. C. McLaskey, D. A. Lockner, Preslip and cascade processes initiating laboratory stick slip. J. Geophys. Res. 119, 6323-6336 (2014). + +59. https://www.continuummechanics.org/strain.html accessed 22 March 2021 + +60. M. H. Lamorde, "Development and application of a novel approach to sand production prediction", thesis, Heriot-Watt University (2015). + +## Supplementary Information + +Supplementary Note: Feedback control of deformation + +Supplementary Figures 1- 14 + +Supplementary Tables 1- 2 + +Supplementary Movies 1- 8 + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Bulk mechanical behaviour of Clashach sandstone. (a) Differential stress and AE event rate evolution with time. The transition from constant strain rate loading ( \(10^{-5} / \mathrm{s}\) ) to constant AE event rate loading (1 AE /s) occurred early in stage (ii). During stage (iv), shear zone propagation occurs first across the sample, and then down the sample. (b) Photograph of the failed sample showing the localised shear damage zone. (b) Differential stress plotted against axial strain. Letters a–g refer to the image labels in Figs. 2 and 3. Young’s modulus, \(E = 19.369 \pm 0.028 \mathrm{GPa}\) , was calculated over the range shown. AE activity began at 40% of peak stress, \(\sigma_{\mathrm{P}}\) , with sample yield at \(0.85\sigma_{\mathrm{P}}\) . The AE feedback control (1 AE/s) modulated the strain rate from \(0.93\sigma_{\mathrm{P}}\) .
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. Micro-scale damage evolution close to and following peak stress. (a-i) Reconstructed 2D \(\mu \mathrm{CT}\) slice (x,y-oriented, where x and y are perpendicular to each other and to the direction of loading, and x is across-strike and y is along-strike of the shear zone) showing
+ +<--- Page Split ---> + +the plane (orange line) of a re- slice of the original \(\mu \mathrm{CT}\) volume (x,z- oriented, where z is the direction of loading) where the shear zone initially localised. The corresponding across- strike (x,z- oriented) re- slice is shown between the grey arrows with the shear zone highlighted by dash- dot lines. (a- ii) histogram of greyscale x- ray intensity values (a proxy for density) showing evolution within the shear zone region (white line in inset image) and an increasing number of partial volume voxels representing increasing damage in the shear zone. (a- iii) Zoomed- in view (region between the grey arrows) of the x,z- oriented slice in (A- i) highlighting the narrow shear zone (dash- dot line) formed after peak stress, and the region of damage that formed after yield but before peak stress (dotted circle). (b) Further zoomed- in view of the across- strike slices (x,z- oriented) showing shear zone emergence and development in region shown by dashed box in (a- iii) for scans a- g (see also Figs. 1c and 3). (c) Even further zoomed- in slices (x,z- oriented) highlighting the variety of micro- mechanisms involved in shear zone formation: numbers (i)- (iii) correspond to the dashed boxes in (b; slice a). When the whole time- series is viewed as an animation (Supplementary Movies 1 and 2), the micro- mechanisms illustrated by the annotations are apparent. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3. 3D incremental strain fields from the onset of strain localisation (marked in Fig. 1c). Incremental dilation, \(\Delta \epsilon_{\mathrm{d}}\) (blue-pink) and shear strain, \(\Delta \epsilon_{\mathrm{s}}\) (yellow-green) were calculated from digital volume correlation between successive pairs of \(\mu \mathrm{CT}\) volumes and are shown (a) parallel to strike (y,z orientation) and (b) perpendicular to strike (x,z orientation). The lower threshold of 0.0017 was set at four standard deviations from the mean error distribution of \(\Delta \epsilon_{\mathrm{s}}\) (Supplementary Figure S6) and the upper threshold shows regions with strain \(>0.01\) (maximum
+ +<--- Page Split ---> + +\(\Delta \epsilon_{\mathrm{s}}\) and \(\Delta \epsilon_{\mathrm{d}}\) were \(\sim 0.04\) ; Supplementary Figures S4 and S5). The thresholds were chosen to visually highlight regions of localised strain. Letters a- g correspond to those in Figs. 1c and 2, with the strain increment following the scan. Strain localisation began at \(0.7\sigma_{\mathrm{P}}\) . + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4. Relationship between AE amplitude (in log-scale) and local incremental strain at the AE location. (a) Incremental shear strain, \(\Delta \epsilon_{\mathrm{s}}\) , and (b) incremental dilation, \(\Delta \epsilon_{\mathrm{d}}\) , each normalized by their respective maximum. The solid black ellipses highlight the lack of large AE events at locations of large strain; evidence for predominantly aseismic deformation. Conversely, the dashed black ellipses highlight locations of large strain which are connected to mostly smaller AE events. Larger AE events are mostly associated with smaller strains.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5. Shear fracture energy estimated from bulk strain and from local incremental strain in the shear zone. (a) Procedure for estimating \(\mathrm{G_c}\) from shear stress and inelastic axial displacement resolved on the fault (reproduced from Fig. 1 in 24) showing (top) schematic of sample with boundary stresses \(\sigma_{1}\) (axial) and \(\sigma_{3}\) (radial), the angle \(\theta\) that fault F makes with \(\sigma_{1}\) , and the shear stress \(\tau\) acting on the fault; (bottom left) obtaining inelastic axial displacement \(\Delta \mathrm{l}\) from the differential stress vs axial displacement curve and then resolving it onto F to obtain the relative slip \(\Delta \mathrm{u}\) ; and (bottom right) obtaining \(\mathrm{G_c}\) by integration under \(\tau\) - \(\Delta \mathrm{u}\) curve from peak shear stress \(\tau_{\mathrm{P}}\) to residual frictional strength \(\tau_{\mathrm{F}}\) . (b) Relationship between \(\tau\) and \(\Delta \mathrm{u}\) for \(\theta = 30.3^{\circ}\) (Supplementary Table 1) for (i) yellow; uniform slip from average bulk axial stress and strain data, (ii) orange; uniform slip from bulk axial stress and strain data at each \(\mu \mathrm{CT}\) scan time, and
+ +<--- Page Split ---> + +(iii) green; total observed local slip in the developing shear zone, obtained from direct dilation and shear strain measurements. We obtained \(G_{\mathrm{c}}\) by integration under the \(\tau\) - \(\Delta u\) curve from peak \(\tau\) to \(\tau_{\mathrm{F}}\) (dotted black line). The contributions to (iii) from local shear strain (blue) and local dilation (purple) are also shown. (C) Evolution of \(G_{\mathrm{c}}\) calculated from (B) as \(\Delta u\) increases. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- supplementarymovie1.mp4- supplementarymovie2.mp4- supplementarymovie3.mp4- supplementarymovie4.mp4- supplementarymovie5.mp4- supplementarymovie6.mp4- supplementarymovie7.mp4- cartwrighttayloretalsupplementaryinformation.pdf- supplementarymovie8.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01_det.mmd b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6472bbac3ad50a9d4853a8ee74f2885f69060704 --- /dev/null +++ b/preprint/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01/preprint__11ac4d843ca54591d0dfde44ab37622789c29dd798127cd4877d6a51cfe2cf01_det.mmd @@ -0,0 +1,632 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 928, 208]]<|/det|> +# Seismic events miss important grain-scale mechanisms governed by kinematics during shear failure of porous rock + +<|ref|>text<|/ref|><|det|>[[44, 230, 617, 270]]<|/det|> +Alexis Cartwright- Taylor ( \(\boxed{\bullet}\) alexis.cartwright- taylor@ed.ac.uk) University of Edinburgh https://orcid.org/0000- 0001- 7805- 4576 + +<|ref|>text<|/ref|><|det|>[[44, 277, 266, 316]]<|/det|> +Maria- Daphne Mangriotis University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 324, 260, 384]]<|/det|> +Ian Main University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 390, 260, 430]]<|/det|> +Ian B. Butler University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 416, 260, 456]]<|/det|> +Florian Fusseis University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 463, 223, 502]]<|/det|> +Martin Ling Edinburgh Hacklab + +<|ref|>text<|/ref|><|det|>[[44, 508, 900, 550]]<|/det|> +Edward Ando Univ. Grenoble Alpes, CNRS, Grenoble INP, 3SR, Grenoble https://orcid.org/0000- 0001- 5509- 5287 + +<|ref|>text<|/ref|><|det|>[[44, 556, 260, 595]]<|/det|> +Andrew Curtis University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 601, 617, 642]]<|/det|> +Andrew Bell University of Edinburgh https://orcid.org/0000- 0002- 5633- 6289 + +<|ref|>text<|/ref|><|det|>[[44, 648, 260, 687]]<|/det|> +Alyssa Crippen University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 694, 260, 733]]<|/det|> +Roberto Rizzo University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 740, 260, 779]]<|/det|> +Sina Marti University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 787, 260, 826]]<|/det|> +Derek Leung University of Edinburgh + +<|ref|>text<|/ref|><|det|>[[44, 833, 610, 873]]<|/det|> +Oxana Magdysyuk Diamond Light Source https://orcid.org/0000- 0003- 3842- 3239 + +<|ref|>text<|/ref|><|det|>[[44, 914, 101, 931]]<|/det|> +Article + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 45, 602, 65]]<|/det|> +Keywords: micro- seismic events, kinematics, catastrophic failure + +<|ref|>text<|/ref|><|det|>[[44, 83, 335, 102]]<|/det|> +Posted Date: November 4th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 121, 475, 141]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1034813/v1 + +<|ref|>text<|/ref|><|det|>[[44, 158, 911, 202]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 236, 936, 280]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on October 18th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33855- z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[111, 78, 886, 140]]<|/det|> +# Seismic events miss important grain-scale mechanisms governed by kinematics during shear failure of porous rock + +<|ref|>text<|/ref|><|det|>[[92, 164, 902, 220]]<|/det|> +A. Cartwright-Taylor1\\*, +M.-D. Mangriotis1, +I. +G. Main1, +I. +B. Butler1, +F. Fusseis1, +M. Ling2, +E. Ando3, +A. Curtis1, +A. +F. Bell1, +A. Crippen1, +R. +E. Rizzo1, +S. Marti1, +D. +D. Leung1, +O. +V. Magdysyuk4 + +<|ref|>sub_title<|/ref|><|det|>[[90, 260, 187, 277]]<|/det|> +## Affiliations + +<|ref|>text<|/ref|><|det|>[[88, 300, 672, 320]]<|/det|> +1School of GeoSciences, University of Edinburgh, Edinburgh, UK. + +<|ref|>text<|/ref|><|det|>[[88, 343, 673, 363]]<|/det|> +2Independent Electronics Developer, Edinburgh Hacklab, Edinburgh, UK. + +<|ref|>text<|/ref|><|det|>[[88, 386, 600, 405]]<|/det|> +3Laboratoire 3SR, Université Grenoble Alpes, Grenoble, France. + +<|ref|>text<|/ref|><|det|>[[88, 428, 550, 448]]<|/det|> +4Beamline I12- JEEP, Diamond Light Source, Didcot, UK. + +<|ref|>text<|/ref|><|det|>[[88, 471, 528, 490]]<|/det|> +\*Correspondence to: alexis.cartwright- taylor@ed.ac.uk + +<|ref|>sub_title<|/ref|><|det|>[[90, 557, 166, 574]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[86, 597, 900, 897]]<|/det|> +Catastrophic failure in brittle, porous materials initiates when smaller- scale fractures localize along an emergent failure plane or 'fault' in a transition from stable crack growth to dynamic rupture. Due to the extremely rapid nature of this critical transition, the precise micro- mechanisms involved are poorly understood and difficult to capture. Here, we observe these micro- mechanisms directly by controlling the microcracking rate to slow down the transition in a unique rock deformation experiment that combines acoustic monitoring with contemporaneous in- situ x- ray imaging of the microstructure. The results provide the first integrated picture of how damage and associated micro- seismic events emerge and evolve together during localisation and failure and allow us to ground truth some previous inferences from mechanical and seismic monitoring alone. They also highlight where such inferences + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 896, 333]]<|/det|> +miss important kinematically- governed grain- scale mechanisms prior to and during shear failure. Contrary to common assumption, we find seismic amplitude is not correlated with local imaged strain; large local strain often occurs with small acoustic emissions, and vice versa. Local strain is predominantly aseismic, explained in part by grain/crack rotation along an emergent shear zone, and the shear fracture energy calculated from local dilation and shear strain on the fault is half of that inferred from the bulk deformation. This improvement in process- based understanding holds out the prospect of reducing systematic errors in forecasting system- sized catastrophic failure in a variety of applications. + +<|ref|>sub_title<|/ref|><|det|>[[90, 386, 181, 404]]<|/det|> +## Main Text + +<|ref|>text<|/ref|><|det|>[[87, 425, 900, 656]]<|/det|> +Catastrophic failure of porous materials is important for wide range of applications and on a variety of scales, from natural and induced earthquakes to the failure of synthetic materials and engineered structures. Understanding the micro- structural processes that operate during weakening and failure of these materials, and the associated strain partition between seismic and aseismic components, is a critical barrier to reducing the significant uncertainties involved in inferring local deformation and strain rates from remotely accessed field- scale seismic or geodetic data. Without this ability, we are unable to improve methods for forecasting and mitigating the risks associated with catastrophic failure. + +<|ref|>text<|/ref|><|det|>[[87, 678, 907, 908]]<|/det|> +The key driving mechanism of catastrophic failure under compression is the concentration of precursory damage along localised zones of deformation, eventually resulting in system- sized failure along a distinct and emergent sub- planar discontinuity (1). Mean field models explain some aspects of the observed behaviour, but rely on average properties that cannot account for localisation or the detailed micro- mechanics (2). They are often derived solely from the properties of recorded acoustic emissions (AE), which are assumed to be representative of the local strain, and cannot account for local aseismic mechanisms that typically constitute \(>99\%\) (3) of the total accumulated strain energy. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 888, 194]]<|/det|> +partition between seismic and aseismic strain directly constrains the rheology and the detectability of seismic precursors to failure, and is currently the largest source of uncertainty in operational forecasting of seismic risk from subsurface engineering projects (4). However, we have limited constraints on its evolution at a large scale, and none at the microscopic scale. + +<|ref|>text<|/ref|><|det|>[[86, 214, 904, 799]]<|/det|> +The size, location and fracture mode of individual micro- cracking events are commonly inferred from AE waveforms, sometimes calibrated against thin sections that reveal the microscopic processes destructively after the test. These processes include elastic compaction, pore collapse, and the development of small tensile micro- cracks oriented parallel to the maximum principal stress (1, 5, 6). Cracking is a form of local failure either due to locally weaker material or local stress concentrations resulting from changes in geometry (7- 11). Under a constant deformation rate, the peak stress often coincides with the onset of an extremely rapid, non- linear acceleration of AE event rate (5, 12), resulting in violent, abrupt failure and rapid stress drop. The non- linear rheology and short time scales ( \(< 40\) s in Clashach sandstone; Supplementary Figure 3) make the micro- mechanisms involved in the transition from stable crack nucleation along a localised shear zone to system- sized rupture extremely difficult to capture and characterize in real time. However, failure can be extended in duration by controlling the loading rate to maintain a constant AE event rate (1, 13, 14). This procedure prevents the acceleration of crack damage and can extend to minutes or even hours the microscopic processes that usually occur over a few seconds, enabling the post- peak- stress region to be studied under quasi- static conditions. Mapping AE source locations during such an experiment provided the first in- situ view of crack localisation and shear zone growth, and an estimate of the associated shear fracture energy, a key parameter in the mechanics of earthquakes and faulting (1, 13). + +<|ref|>text<|/ref|><|det|>[[88, 816, 890, 905]]<|/det|> +More recently, in- situ high- energy x- ray micro- tomography (μCT) time- lapse imaging of rock deformation experiments has enabled the non- destructive characterization of microstructural damage and local strains (15- 17). Such studies have so far been limited to constant strain rate experiments + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 910, 475]]<|/det|> +where changes after peak stress occur too rapidly to be captured, even with the high- speed imaging capabilities of a synchrotron. Here, we address this research gap by presenting detailed in- situ images of rock failure obtained during experiments in a novel x- ray transparent triaxial deformation cell (see Methods), which for the first time integrates acoustic monitoring with fast time- lapse synchrotron x- ray imaging, on the beamline I12- JEEP (18) at the Diamond Light Source, UK. We slowed down failure using feedback from the AE event rate (see (1) and Supplementary Note), extended the failure time from \(\sim 1\) minute to \(\sim 50\) minutes (Supplementary Figure 3), and captured sample weakening and failure in a series of eighteen 3D x- ray \(\mu \mathrm{CT}\) volumes (Fig. 1a), along with the locations and amplitudes of the AE sources. This allowed us to unlock the relationship between the seismic sources, the local strain and the associated underlying micro- mechanisms. We find that AE miss important developments on the grain scale; early strain localisation does not necessarily lock the system in; and the micro- mechanics on the grain scale are governed by kinematics, in turn governed by the rock configuration. + +<|ref|>sub_title<|/ref|><|det|>[[89, 538, 725, 559]]<|/det|> +## Micro-mechanics of strain localisation and failure from in-situ x-ray images + +<|ref|>text<|/ref|><|det|>[[88, 579, 899, 740]]<|/det|> +Under AE- controlled quasi- static loading, the sample eventually experienced system- sized brittle failure along a localised shear zone (Fig. 1), with x- ray \(\mu \mathrm{CT}\) volumes of the precursory microscale processes obtained at the times indicated. The details of damage localisation and shear zone development are shown in post- yield 2D \(\mu \mathrm{CT}\) slices (Fig. 2) and incremental 3D strain fields obtained by Digital Volume Correlation (Fig. 3). + +<|ref|>text<|/ref|><|det|>[[88, 760, 895, 923]]<|/det|> +AE activity preceded initial strain localisation during early loading [stage (i) in Fig. 1; Fig. 3 first panel], consistent with previous AE studies on porous rocks using much larger samples (5, 6, 19) or in- situ \(\mu \mathrm{CT}\) imaging of smaller samples (20, 21). Between yield and shortly after peak stress [stage (ii); Fig. 3 panels a- b], the spatial distribution of shear strain closely followed that of dilation, and competing strain clusters localised along three distinct conjugate planes of similar amplitude and dip + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 70, 896, 159]]<|/det|> +(30° to maximum principal stress; typical of optimally-oriented faults in nature) but variable strike, indicating self-organised exploration of candidate shear zones. These direct observations highlight the exploratory nature of emergent localisation in a complex system. + +<|ref|>text<|/ref|><|det|>[[87, 181, 901, 551]]<|/det|> +In stage (iii), dilation and shear strain concentrated along the critically- oriented shear zone soon after peak stress (Fig. 3 panel c), initially with a brief hiatus in the damage rate (Supplementary Figure 4). The shear zone emerged spontaneously from the self- organised localisation of numerous, narrow en- echelon tensile microcracks, which nucleated simultaneously along the whole length of the shear zone (Fig. 2b and c; slice c) as the upper part of the sample moved as one continuum relative to the lower part. These cracks, which mainly crossed a single whole grain, nucleated at pore edges, at grain contacts where Hertzian stresses accumulated within force chains, and as wing cracks at the tips of sliding grain boundary cracks (consistent with 7- 11). As damage mechanisms localised increasingly on the shear zone, initial diffuse compaction in the bulk was swamped by localised dilation and shear strain (Supplementary Figure 6; Supplementary Movies 3- 8) as a fat tail emerged in their respective frequency- amplitude distributions, which eventually became bimodal (Supplementary Figure 5). + +<|ref|>text<|/ref|><|det|>[[87, 572, 900, 872]]<|/det|> +In stage (iv), dilation and shear strain were highly correlated in the shear zone (Fig. 3), consistent with several mechanisms accommodating bulk shear motion (Fig. 2c; slices c- f). These included en- echelon, tensile cracks nucleating and widening to produce extension and volumetric strain; aseismic rotation of tensile cracks associated with neighboring grain rotation, which prevented tensile crack propagation and supported the walls of the shear zone to maintain finite grain- size thickness throughout failure; and sliding on favourably- and unfavourably- oriented cracks, including some resembling Riedel shear zones. Grain fragmentation generated a proto- cataclasite within the shear zone as whole grains disintegrated, partial grains fractured off the shear zone walls, and fractured grains filled cavities (Fig. 2c; slices d- g). These observations highlight the significant contribution of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 898, 125]]<|/det|> +aseismic mechanisms (i.e., rotation) to the overall failure process, relative to seismic mechanisms (i.e., cracking, stick-slip sliding; 6). + +<|ref|>text<|/ref|><|det|>[[88, 145, 888, 412]]<|/det|> +Eventually coherent slip occurred on a contiguous fault 'plane' that reached the sample boundary on all sides (stage v; Fig. 3 panels f- g), with corresponding reduction of local strain amplitude and local strain rate as the main failure mechanism transferred from local breakdown and rotation to coherent slip (Supplementary Figures 4 and 5). All the micro- mechanisms just described remained active in accommodating coherent slip, explaining the remaining patches of dilation (Fig. 3 panels f- g), including rotation of fragments still attached to the upper part of the sample forming local asperities. As asperities broke away from the upper part of the sample, synthetic shear cracks/voids developed along the walls of the shear zone, facilitating further sliding. + +<|ref|>text<|/ref|><|det|>[[88, 432, 885, 524]]<|/det|> +In summary, we observe a breakdown sequence from failure of individual grains due to point loading, through the formation of a proto- cataclastic due to grain rotation and fragmentation, to an ultra- cataclastic due to further grain fragmentation during coherent slip. + +<|ref|>sub_title<|/ref|><|det|>[[89, 587, 512, 607]]<|/det|> +## AE location and the seismic strain partition factor + +<|ref|>text<|/ref|><|det|>[[87, 628, 907, 896]]<|/det|> +We recorded \(\sim 3600\) AE events above ambient noise, some \(5\%\) of which were selected using objective criteria for location analysis (see Methods). The most likely location for each event was inferred by constraining it to the maximum local strain within a kinematically- derived hyperboloid of possible locations (see Methods and Supplementary Figure 11). The same unique location was found for \(85\%\) of the located AE events in both the dilation and shear strain fields due to the high correlation between the two strain fields. Even with the maximum strain constraint, the largest AE events did not occur at locations of high local strain (Fig. 4 solid black ellipses), implying the seismic strain partition coefficient is highly variable in space. Many small events occurred in regions of high local strain (Fig. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 900, 230]]<|/det|> +4 dashed black ellipses), consistent with deformation being primarily aseismic in the shear zone, while many large events occurred in regions of low local strain in the bulk (Fig. 4 and Supplementary Figure 12). Dilation and shear strain are positively correlated over the whole experiment, but this positive correlation is much stronger at the AE locations (Supplementary Figure 13), consistent with the large mixed mode component of seismic moment tensors observed in (6). + +<|ref|>text<|/ref|><|det|>[[88, 250, 907, 410]]<|/det|> +We inferred a seismic strain partition coefficient for bulk deformation of \(0.2\%\) from summing the inferred scalar seismic moments (see Methods), verifying that deformation is primarily aseismic. This is much lower than the \(1\%\) inferred by (3), most likely due to using a less brittle material (sandstone) under quasi- static rather than constant strain rate loading. Both estimates are a lower bound due to the finite signal to noise ratio. + +<|ref|>sub_title<|/ref|><|det|>[[90, 477, 224, 495]]<|/det|> +## Rupture energy + +<|ref|>text<|/ref|><|det|>[[88, 516, 907, 711]]<|/det|> +The shear fracture energy or energy release rate, \(\mathrm{G}_{\mathrm{c}}\) , is a crucial parameter for modelling shear fracture propagation (23). It is the energy required per unit area for breakdown processes, such as tensile fracturing, to create the new fault surface. It characterises the strain- softening region of the stress- strain curve and is conventionally estimated from bulk axial stress and strain data (Fig. 5a). Here we estimate \(\mathrm{G}_{\mathrm{c}}\) both from bulk and from knowledge of the local strain in the shear zone (see Methods), finding that bulk estimates are biased to large values. + +<|ref|>text<|/ref|><|det|>[[88, 733, 907, 894]]<|/det|> +We computed \(\mathrm{G}_{\mathrm{c}}\) (Fig. 5c) by integrating the shear stress versus fault slip record (Fig. 5b) for (i) uniform slip on a sample- sized fault from bulk axial stress and strain (as in 24), (ii) uniform slip from bulk axial stress and strain at each \(\mu \mathrm{CT}\) scan time, and (iii) total observed local slip measured directly from local dilation and shear strains within the shear zone. We observed a smaller critical slip distance, \(\Delta \mathrm{u}^{*}\) (Fig. 5a) for local slip (0.14 mm) than for uniform slip (0.2 mm), while the integrated curves + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 907, 300]]<|/det|> +yielded \(\mathrm{G_{c - i}} = 5.06 \mathrm{kJ / m^2}\) , \(\mathrm{G_{c - ii}} = 3.72 \mathrm{kJ / m^2}\) and \(\mathrm{G_{c - iii}} = 2.53 \mathrm{kJ / m^2}\) , where dilation and shear strains contributed \(36\%\) and \(64\%\) of \(\mathrm{G_{c - iii}}\) respectively. The discrepancy between \(\mathrm{G_{c - i}}\) and \(\mathrm{G_{c - ii}}\) arises from the feedback control: many of the \(\mu \mathrm{CT}\) scan times after peak stress coincided with a reduction in stress and strain as the ram backed off to maintain the AE event rate (Fig. 1a). However, local shear fracture energy, \(\mathrm{G_{c - iii}}\) , is only \(0.68\mathrm{G_{c - ii}}\) and \(0.5\mathrm{G_{c - i}}\) . Hence, bulk estimates of \(\mathrm{G_{c}}\) for uniform slip across a sample- sized fault can significantly exceed those determined directly from local slip measurements in the developing shear zone. + +<|ref|>sub_title<|/ref|><|det|>[[89, 364, 576, 384]]<|/det|> +## Comparison with laboratory observations of shear failure + +<|ref|>text<|/ref|><|det|>[[86, 405, 907, 741]]<|/det|> +The evolving shear zone imaged in the 3D x- ray volumes and strain fields validates many inferences from classic AE experiments, such as the nucleation and growth of a shear zone containing the eventual fault plane due to the spontaneous localisation of en- echelon tensile microcracks (1, 13). These data also provide new insight into the micro- mechanics of strain localisation and shear failure, notably: the grain size control on failure; the self- organised exploration of candidate shear zones close to peak stress; the continued correlation between dilation and shear strain throughout sample weakening; the relative importance of aseismic mechanisms such as crack rotation in accommodating bulk shear deformation; the lack of correlation between locations of large seismic events and regions of high local strain; the very low and locally highly variable seismic strain partition coefficient; and the relatively low local shear fracture energy compared to bulk estimates. + +<|ref|>text<|/ref|><|det|>[[87, 761, 880, 923]]<|/det|> +The strong correlation between local dilation and shear strain is consistent with in- situ \(\mu \mathrm{CT}\) observations up to peak stress (16). Here, we show that this correlation continues throughout quasi- static failure, confirming, at higher resolution, inferences from seismic velocity tomography (25). If volumetric strain is a proxy for tensile cracking, the correlation confirms the existence of a cohesive zone, but with crack damage distributed throughout the shear zone rather than concentrated solely in a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 909, 194]]<|/det|> +breakdown zone at the propagating front of a pre- existing discontinuity, as proposed by Barenblatt (26). This observation is consistent with the fact that no contiguous fault exists until very late in the process, and even then cracking continues to occur at the grain scale by fracturing individual grains and breaking remaining grain- scale asperities distributed across the whole shear zone. + +<|ref|>text<|/ref|><|det|>[[87, 214, 900, 761]]<|/det|> +The Gutenberg- Richter \(b\) - value estimate from the AE was \(1.94 \pm 0.04\) (Supplementary Figure 10), slightly higher than \(b \sim 1.7\) estimated for granite under similar loading conditions (1). Both are likely high due to the forced quasi- static failure protocol. When \(b > 1.5\) , the moment release is dominated by smaller events, so our very low estimate for the seismic strain partition factor (0.2%) may be due in part to the lack of detection of smaller events above the relatively high ambient noise present in the synchrotron beamline. However, this does not explain the relative absence of large AE events in high strain regions. The distribution of AE locations reflects the high \(b\) - value, with many moderate/small events occurring in regions of large directly measured strain. One explanation could be that the reduction in local stiffness with increasing damage leads to the preference for smaller events along the shear zone, with fewer but larger AE occurring at locally stiff regions off- fault where strain energy accumulates with relatively little deformation. These results show that the seismic strain partition coefficient is highly variable locally and therefore the AE source amplitude is not necessarily representative of the local strain. This is consistent with similar counterintuitive scaling shown between avalanche size (related to strain) and average energy (related to AE amplitude) in a local load sharing fibre bundle model (27). These findings may help to explain similar spatial variability inferred at field scale and constrain model uncertainty when forecasting seismic risk (4). + +<|ref|>text<|/ref|><|det|>[[88, 781, 901, 906]]<|/det|> +Our estimates of \(\mathrm{G}_{\mathrm{c}}\) are close to that estimated for Berea sandstone under the same assumptions (13), and an order of magnitude smaller than reported for granite (1, 24, 25). Since \(\mathrm{G}_{\mathrm{c - iii}}\) (for local slip in a propagating shear zone) is only 50- 68% of \(\mathrm{G}_{\mathrm{c}}\) for uniform slip on a sample- sized fault, significant additional work ( \(\sim 32 - 50\%\) ) must be done in the rest of the sample. However, the average off- fault to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 899, 300]]<|/det|> +on- fault incremental shear and volumetric strain ratios during failure are only \(9\%\) and \(3\%\) respectively (Supplementary Table 1), consistent with ratios of off- fault dissipated energy to on- fault shear fracture energy in granite (25). The remaining \(20 - 38\%\) of additional work must be accommodated by aseismic mechanisms within the shear zone (i.e., crack and grain rotation and silent grain rearrangement). Slip due to rotation for individual cracks can be as large as \(77 \pm 29\%\) of the local relative slip (Supplementary Table 2), and hence likely accounts for a significant proportion of the shear fracture energy. + +<|ref|>sub_title<|/ref|><|det|>[[90, 364, 772, 385]]<|/det|> +## Relating micro-mechanics with natural slip systems and tectonic-scale kinematics + +<|ref|>text<|/ref|><|det|>[[86, 404, 884, 670]]<|/det|> +Our results highlight the complexity of damage processes occurring within shear zones during localisation and through weakening and failure. Local crack rotation with antithetic slip (Fig 2c- iii) offers an additional mechanism for local stress rotation and slip on unfavourably- orientated faults without the need for high pore pressure (28). It may also help to explain observations of interlaced orthogonal faults and accompanying geometric complexities (such as en- echelon faulting, aftershock migration and event triggering from one orientation to another) in the 2019 Ridgecrest earthquake sequence and other examples (29), by providing alternative pathways along which to accommodate strain during a cascading rupture process. + +<|ref|>text<|/ref|><|det|>[[86, 691, 904, 920]]<|/det|> +Our observations of crack rotation provide experimental evidence to support models of tectonic kinematics (e.g., 30- 33), which postulate that rotation in rifting margins and strike- slip settings can emerge as a result of local strength heterogeneity in the crust. In these cases, a strong ‘fixed’ zone (in our case, the rock matrix between en- echelon tensile microcracks) transmits the bulk drag, preventing pure shear and enabling extension and rotation, with rotation facilitated by the surrounding mechanically weak regions (in our case, the tensile cracks themselves) that accommodate the rotation- induced deformation. The rotation in our experiments is driven by shearing oblique to the en- echelon + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 904, 300]]<|/det|> +tensile cracks, whereby the strong grain segments transmit the shearing motion and prevent tensile crack propagation beyond a single grain, thereby limiting the width of the shear zone to the grain scale. Crack coalescence eventually occurs by grain fragmentation within the shear zone and by the breaking of asperities, forming strands of connected crack porosity aligned along the shear zone in patterns consistent with dynamic shear failure (34), coalescence of en- echelon rift segments (35) and kink band development in granodiorite (36). The potential for slip along either and/or both surfaces of the shear zone, as well as planes within the shear zone, is also found on a larger scale in nature (37). + +<|ref|>text<|/ref|><|det|>[[86, 320, 901, 515]]<|/det|> +This discussion highlights the potential for re- examination of the microstructures and inferred mechanisms associated with larger- scale seismic and aseismic processes in light of the results presented here, in particular the conclusion that seismic events miss important grain- scale mechanisms governed by kinematics before and during shear failure. This improvement in process- based understanding holds out the prospect of reducing systematic errors in forecasting system- sized catastrophic failure in a variety of applications. + +<|ref|>sub_title<|/ref|><|det|>[[90, 574, 165, 591]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[90, 614, 803, 670]]<|/det|> +For a description of how we implemented the AE feedback control of deformation, see the Supplementary Note. + +<|ref|>sub_title<|/ref|><|det|>[[90, 692, 268, 711]]<|/det|> +## Experimental material + +<|ref|>text<|/ref|><|det|>[[88, 726, 907, 886]]<|/det|> +A Permian aeolian sandstone (Clashach) was chosen as the experimental material. Clashach is a quartz- rich arenite composed of \(>92\%\) quartz grains, \(< 8\%\) K- feldspar and subordinate lithics. It is well- sorted with fine to medium- sized grains \(0.25 - 0.4 \mathrm{mm}\) in diameter (38). A highly cemented Clashach sample was used (17% porosity), which behaved in a mechanically brittle manner when loaded to failure at shallow crustal pressures, and emitted sufficient acoustic emissions for feedback control. Cylindrical + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 70, 892, 123]]<|/det|> +cores were obtained using a diamond core drill and the ends of the cores were ground flat and parallel on a lathe. + +<|ref|>sub_title<|/ref|><|det|>[[90, 148, 287, 166]]<|/det|> +## Experimental equipment + +<|ref|>text<|/ref|><|det|>[[87, 180, 904, 658]]<|/det|> +The experiment used our lightweight (3.5 kg) x- ray transparent triaxial deformation apparatus, Stör Mjölnir (Supplementary Figure 1), developed at the University of Edinburgh. Named after Thor's hammer in Norse mythology, it is an up- scaled version of our miniature triaxial cell, Mjölnir (39), with the addition of a linear variable displacement transducer (LVDT) to measure axial displacement, and two piezoelectric 'Glaser- type' transducers, positioned axially (Supplementary Figure 1), to passively detect acoustic emissions (AE) and actively monitor ultrasonic velocities. These sensors (signal in Volts proportional to the true normal displacement, \(u\) , of the received elastic wave) were connected to an Applied Seismology Consulting Ltd (ASC) micro- seismic monitoring system by means of two ASC pre- amplifiers. This allowed us to capture AE waveforms and conduct ultrasonic velocity surveys throughout the experiment. Stör Mjölnir is constructed of grade 5 titanium alloy, with an x- ray "transparent" pressure vessel made of 7068- T6 aluminium alloy. It accommodates cylindrical samples of 10 mm diameter and 25 mm length and can attain confining pressures up to 50 MPa and apply axial stresses up to 500 MPa. It was installed on the x- ray microtomography rotation stage in Experimental Hutch 1 (EH1) of beamline I12- JEEP at the Diamond Light Source. + +<|ref|>sub_title<|/ref|><|det|>[[90, 679, 269, 698]]<|/det|> +## Experimental protocol + +<|ref|>text<|/ref|><|det|>[[87, 720, 895, 915]]<|/det|> +The experiment was conducted at ambient temperature. The Clashach core was jacketed in silicone tubing and sealed between the pistons within the pressure vessel (Supplementary Figure 1). An effective pressure ( \(\mathrm{P}_{\mathrm{eff}}\) ) was applied and maintained at 20 MPa throughout the test (confining pressure \(\mathrm{P}_{\mathrm{c}} = 25 \mathrm{MPa}\) and pore fluid pressure \(\mathrm{P}_{\mathrm{p}} = 5 \mathrm{MPa}\) , where \(\mathrm{P}_{\mathrm{eff}} = \mathrm{P}_{\mathrm{c}} - \mathrm{P}_{\mathrm{p}}\) . A hydrostatic starting pressure condition (zero differential stress) was achieved by simultaneously increasing the axial pressure to match the confining pressure. Tomographic reference scans were acquired before pressurisation and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 70, 877, 160]]<|/det|> +again prior to loading (at zero differential stress) to obtain the initial state of the sample. Two scans were acquired at zero differential stress to characterize the error in the digital volume correlation by correlating two volumes in which the state of the sample was identical (Supplementary Figure 7). + +<|ref|>text<|/ref|><|det|>[[88, 181, 901, 411]]<|/det|> +After the initial scans, the sample was loaded continuously at a constant deformation rate of \(10^{- 5}\) \(\mathrm{s}^{- 1}\) until the desired AE event rate (1 event/s) was reached, at which point loading at this constant AE rate took over. This protocol extended the failure time from \(\sim 1\) minute to 50 minutes, equivalent to an average bulk strain rate of \(10^{- 7} \mathrm{s}^{- 1}\) , sufficient to capture 18 high- quality \(\mu \mathrm{CT}\) volumes after peak stress. The sample underwent triaxial deformation to failure evident from a rollover at peak differential stress followed by a gentle decrease in stress ending in an almost constant stress on completion of loading (Fig. 1a and Supplementary Figure 3). + +<|ref|>text<|/ref|><|det|>[[88, 432, 896, 696]]<|/det|> +Imaging the deformation in- situ on beamline I12- JEEP (18) was achieved with a 53 keV monochromatic beam detected by a PCO. edge light sensor with I12 in- house optical module of 7.91 x \(7.91 \mu \mathrm{m}\) /pixel resolution and \(20 \mathrm{mm} \times 12 \mathrm{mm}\) field of view. Tomographic volumes of the whole sample, comprising two discrete overlapping scans of the top and bottom of the sample, with vertical translation in between, were acquired every 85 s throughout the experiment. Individual scans were acquired in approx. 40 seconds and consisted of 1800 projections with a 0.0035s exposure time. At each end of the sample, 0.2 mm (next to the piston) was not captured due to limits on the x- ray field of view, which had a vertical dimension of \(12 \mathrm{mm}\) . + +<|ref|>text<|/ref|><|det|>[[88, 719, 896, 809]]<|/det|> +Ultrasonic velocity surveys were conducted every 5 minutes and acoustic emissions monitored continuously throughout the experiment, along with actuator pressure and axial displacement, confining pressure, and pore fluid pressure and volume. + +<|ref|>text<|/ref|><|det|>[[90, 832, 316, 850]]<|/det|> +Tomographic reconstruction + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 905, 370]]<|/det|> +The tomographic dataset was fully reconstructed on the Diamond cluster using the framework SAVU (40) with automatic centre detection (41), ring removal (42) and distortion correction (43), a Paganin phase filter with delta/beta ratio of 0.6 (44) and a 2D GPU- accelerated filtered backprojection ('FBP_CUDA') reconstruction algorithm using the integrated ASTRA toolbox (45- 48). This yielded 16- bit image volumes of \(2560 \times 2560 \times 1800\) voxels, with a voxel size of \(7.91 \times 7.91 \times 7.91 \mu \mathrm{m}^3\) . Each pair of reconstructed volumes comprising both ends of the sample, acquired with an overlapping region of 520 vertical pixels (4 mm), was merged together using digital volume correlation (DVC) to match the overlapping ends and generate a time- series of tomographic volumes of the full sample (orange dots in Fig. 1 and blue dots in Supplementary Figure 3). + +<|ref|>sub_title<|/ref|><|det|>[[90, 392, 370, 411]]<|/det|> +## Estimation of bulk stress and strain + +<|ref|>text<|/ref|><|det|>[[88, 432, 907, 594]]<|/det|> +Axial displacement was converted to axial strain, \(\epsilon\) , which was then corrected for rig stiffness, \(\mathrm{k}_{\mathrm{rig}}\) to obtain the axial sample strain, \(\epsilon\) , as follows: \(\epsilon = \epsilon /\sigma - \sigma (1 / \mathrm{k}_{\mathrm{rig}})\) . Differential stress, \(\sigma\) , on the sample was obtained from \(\sigma = \sigma_{1} - \sigma_{3}\) , where \(\sigma_{1}\) is the axial stress, equal to the load, F, on the sample (obtained from a calibration of the measured ram pressure with a load cell) divided by the surface area, A, of the sample end \((\sigma_{1} = \mathrm{F / A})\) , and \(\sigma_{3} = \mathrm{P}_{\mathrm{eff}}\) . + +<|ref|>sub_title<|/ref|><|det|>[[90, 615, 366, 635]]<|/det|> +## Dynamic wave velocity estimation + +<|ref|>text<|/ref|><|det|>[[87, 655, 907, 921]]<|/det|> +Changes in ultrasonic compressional- wave velocity were measured by means of the pulse transmission technique (49) between the pair of AE transducers, with source- receiver geometries at opposite ends of the sample (Supplementary Figures 1 and 11). Velocity surveys were conducted by the ASC Pulser Interface Unit at constant time intervals (5 minutes) during the experiment. A pulse of acoustic energy was sent through the sample from each transducer in turn and its arrival at the other sensor was recorded. To understand the system delay (including travel- time through pistons) a seismic test using the pistons without any sample in between them was recorded, and the difference in time between pulse onset at the origin and pulse onset retrieval at the other piston was calculated as the time- correction. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 110, 886, 411]]<|/det|> +Detected AE signals were amplified and sent to the two- channel monitoring system, which recorded the full event waveforms, including arrival time, amplitude and first- motion information. The pre- amplification and trigger threshold (based on an instantaneous amplitude) were set at 70 dB and 280 mV respectively, with the aim of controlling the AE event rate effectively. The gain was determined from a benchmark pencil lead- break test in the laboratory prior to visiting the synchrotron, and the trigger threshold was determined by the ambient noise in the experimental hutch on the I12- JEEP beamline. AE event rates were calculated from the number of events recorded in each 10 s time window. AE amplitudes as a function of time, AE event rate and frequency- magnitude distributions are shown in Supplementary Figure 10. + +<|ref|>text<|/ref|><|det|>[[86, 432, 905, 808]]<|/det|> +We interrogated the AE events to obtain a sub- set of the most reliable events for location analysis, ensuring that: (i) individual events were represented in both sensors' recordings, (ii) the AE waveforms satisfied a minimum Signal to Noise Ratio (SNR), (iii) the relative time- delay for a specific AE recorded at the two sensors was robust against noise and yielded a depth uncertainty less than 1 pixel, (iv) the maximum relative time- delay for a specific AE recorded at the two sensors obeyed physical limits given sample dimensions/velocity. The relative time- delays between top and bottom sensor were estimated by cross- correlation of the top and bottom waveforms. Given the limitation of only two sensors at the opposite ends of the rock sample, the experimental geometry led to inherent ambiguity in AE location. However, the specific relative time- delay of an AE event, and knowledge of the dynamic velocity allowed us to define a circular hyperboloid, which marks a 3D surface of constant relative time- delay (Supplementary Figure 11) and therefore a kinematic constraint for AE location. + +<|ref|>text<|/ref|><|det|>[[87, 824, 877, 914]]<|/det|> +Using the AE events that passed the selection criteria ( \(\sim 5\%\) of 3600 events) we defined each hyperboloid as the locus of potential elementary volumes (with thickness \(= 1\) pixel) for the AE location. Time- binning of AE relative to the occurrence of the two scans used to obtain a particular + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 901, 475]]<|/det|> +strain increment from the DVC permitted comparison of AE with the local incremental strains. Assuming that each AE occurred at the largest incremental local strain within the kinematically- defined hyperboloid, a unique AE location was established (location constrained by local strain). Given the large correlation between locations of large volumetric strain vs. deviatoric strain, AE locations were defined within the same pixel whether constrained by volumetric or deviatoric strain for \(85\%\) of the located AE. Where two or more AE had the same time- bin and the same kinematic constraint, we assumed that the largest AE would be correlated with the largest strain, effectively assigning AE locations sequentially from largest to lowest AE within a time- bin. Whilst the assumption of linking AE to the largest local strain should bias AE locations to be linked to larger local strains, the results showed that, effectively via proof by contradiction, the largest AEs do not occur at locations of large local strain, with the locations of large local strains being linked to small/moderate AE (Fig. 4). This observation, in turn, is evidence that deformation is primarily aseismic. + +<|ref|>text<|/ref|><|det|>[[90, 495, 830, 515]]<|/det|> +Local incremental 3D strain fields from digital volume correlation between neighboring scans + +<|ref|>text<|/ref|><|det|>[[86, 536, 904, 908]]<|/det|> +In addition to observing quasi- static fault development directly in the reconstructed \(\mu \mathrm{CT}\) volumes, we quantified the full, incremental 3D strain tensor evolution using DVC between adjacent pairs of tomograms using the open- source software SPAM (50). The procedure involved computing linear displacement vectors between nodes (with 40 pixel spacing, equivalent to \(316.4 \mu \mathrm{m}\) ) identified in the reference image and the same nodes identified in the deformed image by tracking a representative volume (window size 40 pixels) around the nodes. The transformation gradient tensor (local derivatives of displacement), \(\mathbf{F}\) , was then computed using a Q8 shape function linking 8 neighbouring nodes, with \(\mathbf{F}\) computed in the centre of the element. \(\mathbf{F}\) was decomposed into the symmetric stretch tensor, \(\mathbf{U}\) , which was further decomposed into an isotropic and a deviatoric part. The volumetric strain (first invariant of the strain tensor showing dilation/compaction) was defined as the determinant of the transformation gradient tensor, \(\left| \mathbf{F} \right| - 1\) . Here, we followed soil mechanics convention and defined + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 881, 161]]<|/det|> +dilation as positive volumetric strain and compaction as negative volumetric strain. The deviatoric strain (second invariant of the strain tensor showing shear deformation and referred to throughout as the shear strain) was defined as the Euclidean norm of the deviatoric part \(\| \mathbf{U}_{dev}\|\) . + +<|ref|>text<|/ref|><|det|>[[90, 182, 523, 202]]<|/det|> +Estimation of the overall seismic strain partition factor + +<|ref|>text<|/ref|><|det|>[[88, 222, 910, 800]]<|/det|> +To estimate the seismic strain partition factor, we followed the approach of (3), who estimated the seismic moment of the largest observed AE event based on the largest observed fault, in order to scale the distribution of seismic moments from all their events. We used (51) to obtain the scalar seismic moments \(M_{o} = CV\Delta \sigma\) , with \(C = 16 / 7\) , \(V = r^{3}\) the crack volume and \(\Delta \sigma\) the stress drop. The estimated maximum for the scalar seismic moment was 1.5e- 3 Nm, using the Madariaga source model (52) with a radius, \(r\) , equal to one- half the maximum observed crack size ( \(2r \sim 800 \mu \mathrm{m}\) ) and a stress drop of 1 MPa. The latter stress drop is an intermediate stress drop encompassing 23 different studies of natural, mining induced, and fracking induced seismicity and laboratory AE events (Supplementary Figure 13; 53, 54). The maxima of the AE envelopes were scaled to the maximum scalar seismic moment to obtain the scalar seismic moment distribution. The Kostrov (55) strain was estimated from \(\Delta \epsilon_{ij} = \frac{1}{2\mu\Delta V}\sum_{i = 1}^{N}M_{ij}\) , with \(M_{ij} = \sqrt{2M_{o}U_{ij}}\) (56), where \(U_{ij}\) is the unit displacement tensor, \(\mu\) the shear modulus and \(\Delta V\) the total representative volume. Finally, the strain partition factor was estimated from \(\chi = \frac{\gamma_{AE}}{\gamma_{F}}\) (57, 58, 3), with \(\gamma_{AE}\) representing the sum of scalar seismic moments multiplied by \(\frac{\sqrt{2}}{2\mu\Delta V}\) and \(\gamma_{F}\) the average volumetric and deviatoric strains accumulated throughout the deformation experiment. + +<|ref|>text<|/ref|><|det|>[[90, 836, 486, 856]]<|/det|> +Estimation of bulk and local shear fracture energy + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 68, 901, 405]]<|/det|> +The breakdown zone (slip- weakening) model (Fig. 5a; 23, 24) considers a fault as a shear crack with peak shear strength, \(\tau_{\mathrm{P}}\) . Relative slip on the fault, \(\Delta \mathrm{u}\) , initiates at \(\tau_{\mathrm{P}}\) and then strength degrades from \(\tau_{\mathrm{P}}\) to a constant residual frictional strength, \(\tau_{\mathrm{F}}\) , as \(\Delta \mathrm{u}\) increases to a value \(\Delta \mathrm{u}^{*}\) , the critical slip- weakening distance, where tensile fracturing (mode I; dilation) transitions to frictional sliding (mode II; shear). This leads to a breakdown zone of dimension \(\omega_{0}\) at the shear crack tip. \(\mathrm{G}_{\mathrm{c}}\) for breakdown processes at the crack tip can be evaluated by integrating under the \(\tau\) - \(\Delta \mathrm{u}\) curve (23; Fig. 5a). \(\Delta \mathrm{u}\) is resolved from the bulk inelastic axial displacement, \(\Delta \mathrm{l}\) , and depends on the angle, \(\theta\) , which the fault makes with the direction of loading. Shear stress depends on \(\theta\) and the boundary stress conditions. Three main assumptions are involved: all axial strain beyond peak stress is accommodated by shear slip on the fault, the fault is sample- sized throughout, and \(\omega_{0}\) is small compared with the size of the fault. + +<|ref|>text<|/ref|><|det|>[[87, 424, 910, 654]]<|/det|> +To obtain local geometric and strain information for the shear zone itself defined by the local strain fields that were output from the DVC (Fig. 3 and Supplementary Figure 4), we separated the shear zone core from other correlated strain clusters by thresholding above 0.004 strain. We then used a connected component object analysis to obtain the best- fitting ellipsoid for the shear zone at each strain increment, and hence obtained the angle \(\theta\) of the shear zone with respect to the direction of loading (Supplementary Table 1) from the ellipsoid eigenvectors. The evolving shear zone object is shown in Supplementary Figure 9. + +<|ref|>text<|/ref|><|det|>[[147, 676, 725, 705]]<|/det|> +Following (23, 24), we then estimated \(\mathrm{G}_{\mathrm{c}} = \int_{0}^{\Delta \mathrm{u}^{*}}[\tau (\Delta \mathrm{u}) - \tau_{\mathrm{F}}]\mathrm{d}(\Delta \mathrm{u})\) for: + +<|ref|>text<|/ref|><|det|>[[116, 725, 907, 922]]<|/det|> +(i) Uniform slip on a sample-sized fault (as in 24) by first resolving relative slip, \(\Delta \mathrm{u}\) , from the bulk inelastic axial displacement, \(\Delta \mathrm{l}\) , and \(\theta\) as follows: \(\Delta \mathrm{u} = \Delta \mathrm{l} / \cos \theta\) , and then calculating shear stress, \(\tau = [(\sigma_{1} - \sigma_{3}) / 2] * \sin (2\theta)\) , from \(\theta\) and the boundary stress conditions. A cubic fit was used to obtain an average bulk shear stress vs relative slip curve and we assumed that all axial shortening was caused by slip along the fault from \(\tau_{\mathrm{P}}\) . Axial shortening was corrected for elastic strain using the intact Young's modulus for the sample (Figs. 1c and 5a). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 68, 904, 230]]<|/det|> +(ii) Uniform slip at each \(\mu \mathrm{CT}\) scan time, using the bulk shear strain and relative slip measurements at those times (rather than an average), in order to compare the uniform slip model directly with the observed local slip on the shear zone itself (estimated from the 3D strain fields). We assumed that all axial shortening was caused by slip along the fault from the point of final localization (scan c). + +<|ref|>text<|/ref|><|det|>[[115, 250, 907, 734]]<|/det|> +(iii) Total observed local slip on the shear zone itself from local dilation and shear strain. We assumed that slip started from the point of final localization (scan c) and used the bulk shear strain calculated in (ii). For the shear contribution to slip, we extracted the mean local incremental shear strain in the shear zone object, \(\Delta \epsilon_{\mathrm{shear}}\) , from the connected component object analysis described above. We corrected \(\sum \overline{\Delta\epsilon_{\mathrm{shear}}}\) for elastic strain using the intact shear modulus of the sample, obtained from the shear stress vs \(\sum \overline{\Delta\epsilon_{\mathrm{shear}}}\) curve (Supplementary Figure 8). Finally, we obtained local shear slip, \(\Delta u_{\mathrm{shear}}\) , from an engineering strains approximation: \(\Delta u_{\mathrm{shear}} = 2 * \sum \overline{\Delta\epsilon_{\mathrm{shear}}} * \mathrm{DVC}\) node spacing (59). For the radial dilation contribution to slip, we extracted the mean local incremental radial dilation in the shear zone object, \(\overline{\Delta\epsilon_{\mathrm{dilaton}}}\) , and corrected \(\sum \overline{\Delta\epsilon_{\mathrm{dilaton}}}\) for elastic radial strain using the elastic axial strain calculated above and an estimate of Poisson's ratio for Clashach sandstone (60). We then resolved the inelastic dilation onto the shear zone orientation, obtaining local dilational slip from an engineering strains approximation: \(\Delta u_{\mathrm{dilaton}} = \sum (\overline{\Delta\epsilon_{\mathrm{dilaton}}} /\sin \theta) * \mathrm{DVC}\) node spacing. + +<|ref|>sub_title<|/ref|><|det|>[[90, 807, 247, 825]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[88, 846, 850, 902]]<|/det|> +We would like to thank the University of Edinburgh Geosciences Workshop for their support in developing the experimental apparatus, and Jonathan Singh for useful discussions about data file + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 896, 196]]<|/det|> +formats and estimation of acoustic waveform arrival times. We also acknowledge Diamond Light Source for time on beamline I12- JEEP under proposal MG22517. This work is supported by the UK's Natural Environment Research Council (NERC) through the CATFAIL project NE/R001693/1 Catastrophic failure: what controls precursory localisation in rocks? + +<|ref|>sub_title<|/ref|><|det|>[[90, 225, 271, 244]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[86, 264, 904, 600]]<|/det|> +Conceptualization (A. Bell, I. Butler, A. Cartwright- Taylor, A. Curtis, F. Fusseis, I. Main), data curation (A. Cartwright- Taylor, M. Mangriotis), formal analysis (A. Cartwright- Taylor, M. Mangriotis, E. Andò, A. Bell, A. Crippen), funding acquisition (A. Bell, I. Butler, A. Cartwright- Taylor, A. Curtis, F. Fusseis, I. Main), investigation (I. Butler, A. Cartwright- Taylor, F. Fusseis, D. Leung, O. V. Magdysyuk, S. Marti, R. Rizzo), methodology (E. Andò, M. Ling), project administration (A. Cartwright- Taylor, I. Main), resources (I. Butler, E. Andò, M. Ling, O. V. Magdysyuk), software (E. Andò, M. Ling), supervision (I. Main, A. Curtis, A. Bell, I. Butler, F. Fusseis), validation (A. Cartwright- Taylor, M. Mangriotis, I. Main, A. Curtis.), writing – original draft (A. Cartwright- Taylor, M. Mangriotis, I. Main, M. Ling), writing – review and editing (A. Bell, I. Butler, A. Curtis, F. Fusseis, O. V. Magdysyuk, R. Rizzo). + +<|ref|>sub_title<|/ref|><|det|>[[90, 630, 366, 649]]<|/det|> +## Statement of competing interests + +<|ref|>text<|/ref|><|det|>[[90, 672, 434, 691]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[90, 722, 351, 741]]<|/det|> +## Data and materials availability + +<|ref|>text<|/ref|><|det|>[[88, 763, 896, 854]]<|/det|> +The data supporting our conclusions can be found in main text, in the supplementary information, and in the \(\mu \mathrm{CT}\) data sets held at the NERC repository (link tbc). When using the \(\mu \mathrm{CT}\) data sets, please cite (dataset publication tbc) and see (link tbc) for the metadata. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 185, 88]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[88, 110, 895, 168]]<|/det|> +1. D. Lockner, J. D. Byerlee, V. Kuksenko, A. Ponomarev, A. Sidorin, Quasi-static fault growth and shear fracture energy in granite. Nature 350 39-42 (1991). + +<|ref|>text<|/ref|><|det|>[[88, 189, 904, 245]]<|/det|> +2. I. Main, A damage mechanics model for power-law creep and earthquake aftershock and foreshock sequences. Geophys. J. Int. 142, 151-161 (2000). + +<|ref|>text<|/ref|><|det|>[[88, 266, 907, 323]]<|/det|> +3. G. Dresen, G. Kwiatek, T. Goebel, Y. Ben-Zion, Seismic and aseismic preparatory processes before large stick-slip failure. Pure Appl. Geophys. 177, 5741-5760 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 343, 902, 400]]<|/det|> +4. S. Bourne, S. J. Oates, J. van Elk, D. Doornhof, A seismological model for earthquakes induced by fluid extraction from a subsurface reservoir. J. Geophys. Res. 119, 8991-9015 (2014). + +<|ref|>text<|/ref|><|det|>[[88, 420, 853, 477]]<|/det|> +5. P. R. Sammonds, P. G. Meredith, I. G. Main, Role of pore fluids in the generation of seismic precursors to shear fracture. Nature 359, 228-230 (1992). + +<|ref|>text<|/ref|><|det|>[[88, 497, 880, 590]]<|/det|> +6. C. C. Graham, S. Stanchits, I. G. Main, G. Dresen, Comparison of polarity and moment tensor inversion methods for source analysis of acoustic emission data. Int. J. Rock Mech. Min. Sci. 47, 161-169 (2010). + +<|ref|>text<|/ref|><|det|>[[88, 610, 890, 666]]<|/det|> +7. N. Cho, C. D. Martin, D. S. Sego, A clumped particle model for rock. Int. J. Rock Mech. Min. Sci. 44(7), 997-1010 (2007). + +<|ref|>text<|/ref|><|det|>[[88, 687, 861, 743]]<|/det|> +8. D. O. Potyondy, P. A. Cundall, A bonded particle model for rock. Int. J. Rock Mech. Min. Sci. 41(8), 1329-1364 (2004). + +<|ref|>text<|/ref|><|det|>[[88, 764, 874, 820]]<|/det|> +9. C. G. Sammis, M. F. Ashby, The failure of brittle porous solids under compressive stress states. Acta Metall. 34, 511-526 (1986). + +<|ref|>text<|/ref|><|det|>[[88, 841, 897, 897]]<|/det|> +10. M. F. Ashby, S. D. Hallam, The failure of brittle solids containing small cracks under compressive stress states. Acta Metall. 34, 497-510 (1986). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 902, 125]]<|/det|> +11. H. Hori, S. Nemat-Nasser, Compression-induced microcrack growth in brittle solids: axial splitting and shear failure. J. Geophys. Res. 90(B4), 3105-3125 (1985). + +<|ref|>text<|/ref|><|det|>[[88, 145, 900, 202]]<|/det|> +12. J. Vasseur, F. B. Wadsworth, Y. Lavallée, A. Bell, I. G. Main, D. B. Dingwell, Heterogeneity: The key to failure forecasting. Sci. Rep. 5, 13259 (2015). + +<|ref|>text<|/ref|><|det|>[[88, 222, 896, 316]]<|/det|> +13. D. Lockner, J. D. Byerlee, V. Kuksenko, A. Ponomarev, A., Sidorin, "Observations of quasi-static fault growth from acoustic emissions" in Fault Mechanics and Transport Properties of Rocks, B. Evans, T.-F. Wong, Eds. (Academic Press, 1992), vol. 51, chap. 1, pp. 3-31. + +<|ref|>text<|/ref|><|det|>[[88, 335, 904, 428]]<|/det|> +14. M. Terada, T. Yanagidani, S. Ehara, "AE rate controlled compression test of rocks" in 3rd Conf. on Acoustic Emission / Microseismic Activity in Geologic Structures and Materials, H. R. Hardy, F. W. Leighton, Eds. (Trans Tech, 1984), pp. 159-171. + +<|ref|>text<|/ref|><|det|>[[88, 447, 907, 540]]<|/det|> +15. L. Huang, P. Baud, B. Cordonnier, F. Renard, L. Liu, T.-F. Wong, Synchrotron X-ray imaging in 4D: multiscale failure and compaction localization in triaxially compressed porous limestone. Earth Planet. Sci. Lett. 528, 115831 (2019). + +<|ref|>text<|/ref|><|det|>[[88, 560, 882, 616]]<|/det|> +16. J. McBeck, Y. Ben-Zion, F. Renard, The mixology of precursory strain partitioning approaching brittle failure in rocks. Geophys. J. Int. 221, 1856-1872 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 636, 894, 728]]<|/det|> +17. A. Cartwright-Taylor, I. G. Main, I. B. Butler, F. Fusseis, M. Flynn, A. King, Catastrophic failure: how and when? Insights from 4-D in situ x-ray microtomography. J. Geophys. Res. Solid Earth 125, e2020JB019642 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 748, 888, 874]]<|/det|> +18. M. Drakopoulos, T. Connolley, C. Reinhard, R. Atwood, O. Magdysyuk, N. Vo, M. Hart, L. Connor, B. Humphreys, G. Howell, S. Davies, T. Hill, G. Wilkin, U. Pedersen, A. Foster, N. De Maio, M. Basham, F. Yuan, K. Wanelik, I12: the Joint Engineering, Environment and Processing (JEEP) beamline at Diamond Light Source. J. Synchrotron Rad. 22, 828-838 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 860, 160]]<|/det|> +19. I. O. Ojala, I. G. Main, B. T. Ngwenya, Strain rate and temperature dependence of Omori law scaling constants of AE data: implications for earthquake foreshock-aftershock sequences. Geophys. Res. Lett. 31, L24617 (2004). + +<|ref|>text<|/ref|><|det|>[[88, 180, 890, 307]]<|/det|> +20. F. Renard, J. McBeck, B. Cordonnier, X. Zheng, N. Kandula, J. R. Sanchez, M. Kobchenko, C. Noiriel, W. Zhu, P. Meakin, F. Fussies, D. K. Dysthe, Dynamic in situ three-dimensional imaging and digital volume correlation analysis to quantify strain localization and fracture coalescence in sandstone. Pure Appl. Geophys. 176, 1083-1115 (2018). + +<|ref|>text<|/ref|><|det|>[[88, 327, 904, 384]]<|/det|> +21. F. Renard, J. McBeck, N. Kandula, B. Cordonnier, P. Meakin, Y. Ben-Zion, Volumetric and shear processes in crystalline rock approaching faulting. Proc. Natl. Acad. Sci. 116, 16234-16239 (2019). + +<|ref|>text<|/ref|><|det|>[[88, 404, 904, 461]]<|/det|> +22. S. Lennartz-Sassinek, I.G. Main, M. Zaiser, C. Graham, Acceleration and localization of subcritical crack growth in a natural composite material. Phys. Rev. E 90, 052401 (2014). + +<|ref|>text<|/ref|><|det|>[[88, 481, 839, 538]]<|/det|> +23. J. R. Rice, "The mechanics of earthquake rupture" in Physics of the Earth's Interior, A. M. Dziewonski, E. Boschi, Eds. (Italian Phys. Soc. 1980), pp. 555-649. + +<|ref|>text<|/ref|><|det|>[[88, 558, 872, 615]]<|/det|> +24. T.-F. Wong, Shear fracture energy of Westerly granite from post-failure behaviour. J. Geophys. Res. 87, 990-1000 (1982). + +<|ref|>text<|/ref|><|det|>[[88, 636, 852, 693]]<|/det|> +25. F. M. Aben, N. Brantut, T. M. Mitchell, E. C. David, Rupture energetics in crustal rock from laboratory-scale seismic tomography. Geophys. Res. Lett. 46, 7337-7344. + +<|ref|>text<|/ref|><|det|>[[88, 713, 904, 770]]<|/det|> +26. G. I. Barenblatt, The mathematical theory of equilibrium cracks in brittle fracture. Adv. Appl. Mech. 7, 55-129 (1962). + +<|ref|>text<|/ref|><|det|>[[88, 791, 888, 848]]<|/det|> +27. S. Roy, S. Biswas, Size distribution of emitted energies in local load sharing fiber bundles. Front. Phys. 9, 643602 (2021). + +<|ref|>text<|/ref|><|det|>[[88, 868, 857, 925]]<|/det|> +28. D. R. Faulkner, T. M. Mitchell, D. Healy, M. J. Heap, Slip on 'weak' faults by the rotation of regional stress in the fracture damage zone. Nature 444, 922-925 (2006). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 857, 160]]<|/det|> +29. Z. E. Ross, B. Idini, Z. Jia, O. L. Stephenson, M. Zhong, X. Wang, Z. Zhan, M. Simons, E. J. Fielding, S.-H. Yun, E. Hauksson, A. W. Moore, Z. Liu, J. Jung, Hierarchical interlocked orthogonal faulting in the 2019 Ridgecrest earthquake sequence. Science 366, 346-351 (2019). + +<|ref|>text<|/ref|><|det|>[[87, 181, 866, 237]]<|/det|> +30. T. Taymaz, J. Jackson and D. McKenzie, Active tectonics of the north and central Aegean Sea. Geophys. J. Int. 106, 433-490 (1991). + +<|ref|>text<|/ref|><|det|>[[87, 258, 900, 313]]<|/det|> +31. H. Schouten, K. D. Klitgord, D. G. Gallo, Edge-driven microplate kinematics. J. Geophys. Res. 98, 6689-6701 (1993). + +<|ref|>text<|/ref|><|det|>[[87, 334, 881, 427]]<|/det|> +32. A. Glerum, S. Brune, D. S. Stamps, M. R. Strecker, Victoria continental microplate dynamics controlled by the lithospheric strength distribution of the East African Rift. Nat. Comm. 11, 2881 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 448, 864, 504]]<|/det|> +33. G. Duclaux, R. S. Huismans, D. A. May, Rotation, narrowing, and preferential reactivation of brittle structures during oblique rifting. Earth Planet. Sci. Lett. 531, 115952 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 524, 870, 617]]<|/det|> +34. R. E. Rizzo, D. Healy, M. J. Heap, N. J. Farrell, Detecting the onset of strain localization using two-dimensional wavelet analysis on sandstone deformed at different effective pressures. J. Geophys. Res. Solid Earth 123, 10460-10478 (2018). + +<|ref|>text<|/ref|><|det|>[[88, 637, 886, 728]]<|/det|> +35. G. Corti, R. Cioni, Z. Franceschini, F. Sani, S. Scaillet, P. Molin, I. Isola, F. Mazzarini, S. Brune, D. Keir, A. Erbello, A. Muluneh, F. Illsley-Kemp, A. Glerum. Aborted propagation of the Ethiopian rift caused by linkage with the Kenyan rift. Nat. Comm. 10, 1309 (2019). + +<|ref|>text<|/ref|><|det|>[[88, 749, 870, 840]]<|/det|> +36. R. K. Davies, D. D. Pollard, Relations between left-lateral strike-slip faults and right-lateral monoclonal kink bands in granodiorite, Mt. Abbot Quadrangle, Sierra Nevada, California. Pure Appl. Geophys. 124, 177-201 (1986). + +<|ref|>text<|/ref|><|det|>[[88, 860, 904, 916]]<|/det|> +37. T. M. Mitchell, D. R. Faulkner, Towards quantifying the matrix permeability of fault damage zones in low porosity rocks. Earth Planet. Sci. Lett. 339-340, 24-31 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 69, 875, 125]]<|/det|> +38. N. J. C. Farrell, D. Healy, Anisotropic pore fabrics in faulted porous sandstones. J. Struct. Geol. 104, 125-141 (2017). + +<|ref|>text<|/ref|><|det|>[[88, 146, 899, 237]]<|/det|> +39. I. B. Butler, F. Fusseis, A. Cartwright-Taylor, M. Flynn, Mjölnir: a miniature triaxial rock deformation apparatus for 4D synchrotron x-ray micro-tomography. J. Synchrotron Rad. 27, 1681-1687 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 258, 886, 314]]<|/det|> +40. N. Wadeson, M. Basham, Savu: a Python-based, MPI framework for simultaneous processing of multiple, N-dimensional, large tomography datasets. arXiv:1610.08015 (2016). + +<|ref|>text<|/ref|><|det|>[[88, 335, 894, 391]]<|/det|> +41. N. T. Vo, M. Drakopoulos, R. C. Atwood, C. Reinhard, Reliable method for calculating the center of rotation in parallel-beam tomography. Opt. Express 22, 19078-19086 (2014). + +<|ref|>text<|/ref|><|det|>[[88, 412, 890, 468]]<|/det|> +42. N. T. Vo, M. Drakopoulos, R. C. Atwood, Superior techniques for eliminating ring artifacts in X-ray micro-tomography. Opt. Express 26, 28396-28412 (2018). + +<|ref|>text<|/ref|><|det|>[[88, 489, 904, 545]]<|/det|> +43. N. T. Vo, R. C. Atwood, M. Drakopoulos, Radial lens distortion correction with sub-pixel accuracy for X-ray micro-tomography. Opt. Express 23, 32859-32868 (2015). + +<|ref|>text<|/ref|><|det|>[[88, 566, 900, 658]]<|/det|> +44. D. Paganin, S. C. Mayo, T. E. Gureyev, P. R. Miller, S. W. Wilkins, Simultaneous phase and amplitude extraction from a single defocused image of a homogeneous object. J. Microsc. 206, 33-40 (2002). + +<|ref|>text<|/ref|><|det|>[[88, 679, 901, 771]]<|/det|> +45. D. Pelt, D. Gürsoy, W. J. Palenstijn, J. Sijbers, F. De Carlo, K. J. Batenburg, Integration of tomopy and the astra toolbox for advanced processing and reconstruction of tomographic synchrotron data. J. Synchrotron Rad. 23, 842-849 (2016). + +<|ref|>text<|/ref|><|det|>[[88, 792, 902, 849]]<|/det|> +46. W. van Aarle, W. J. Palenstijn, J. De Beenhouwer, T. Altantzis, S. Bals, K. J. Batenburg, J. Sijbers, The ASTRA toolbox: a platform for advanced algorithm development in electron tomography. + +<|ref|>text<|/ref|><|det|>[[118, 861, 407, 881]]<|/det|> +Ultramicroscopy 157, 35-47 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 870, 159]]<|/det|> +47. W. van Aarle, W. J. Palenstijn, J. Cant, E. Janssens, F. Bleichrodt, A. Dabravolski, J. De Beenhouwer, K. J. Batenburg, J. Sijbers, Fast and flexible x-ray tomography using the ASTRA toolbox. Opt. Express 24, 25129-25147 (2016). + +<|ref|>text<|/ref|><|det|>[[88, 180, 884, 272]]<|/det|> +48. W. J. Palenstijn, K. J. Batenburg, J. Sijbers, Performance improvements for iterative electron tomography reconstruction using graphics processing units (GPUs). J. Struct. Biol. 176, 250-253 (2011). + +<|ref|>text<|/ref|><|det|>[[88, 292, 896, 348]]<|/det|> +49. F. Birch, The velocity of compressional waves in rocks to 10 kilobars, part 1. J. Geophys. Res. 65, 1083-1102 (1960). + +<|ref|>text<|/ref|><|det|>[[88, 368, 901, 565]]<|/det|> +50. O. Stamati, E. Andò, E. Roubin, R. Cailletaud, M. Wiebicke, G. Pinzon, C. Couture, R. C. Hurley, R. Caulk, D. Caillerie, T. Matsushima, P. Bésuelle, F. Bertoni, T. Arnaud, A. O. Laborin, R. Rorato, Y. Sun, A. Tengattini, O. Okubadejo, J.-B. Colliat, M. Saadatfar, F. E. Garcia, C. Papazoglou, I. Vego, S. Brisard, J. Dijkstra, G. Birmpilis, SPAM: software for practical analysis of materials. J. Open Source Softw. 5(51), 2286 (2020). https://ttk.gricad-pages.univ-grenoble-alpes.fr/spam/index.html. + +<|ref|>text<|/ref|><|det|>[[88, 586, 880, 642]]<|/det|> +51. H. Kanamori, D. L. Anderson, Theoretical basis of some empirical relations in seismology. Bull. Seismol. Soc. Am. 65, 1073-1095 (1975). + +<|ref|>text<|/ref|><|det|>[[88, 662, 875, 718]]<|/det|> +52. R. Madariaga, Dynamics of an expanding circular fault, Bull. Seismol. Soc. Am. 66(3), 639-666 (1976). + +<|ref|>text<|/ref|><|det|>[[88, 740, 870, 832]]<|/det|> +53. A. B. Blanke, G. Kwiatek, T. H. W. Goebel, M. Bohnhoff, G. Dresen, Stress drop-magnitude dependence of acoustic emissions during laboratory stick-slip. Geophys. J. Int. 224, 1371-1380 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 896, 160]]<|/det|> +54. G. Kwiatek, K. Plenkers, G. Dresen, JAGUARS Research Group, Source parameters of pico-seismicity recorded at Mponeng Deep Gold Mine, South Africa: implications for scaling relations, Bull. Seismol. Soc. Am. 101(6), 2592–2608 (2011). + +<|ref|>text<|/ref|><|det|>[[88, 180, 884, 238]]<|/det|> +55. V. Kostrov, Seismic moment and energy of earthquakes and seismic flow of rock. Earth Phys. 1, 23-40 (1974). + +<|ref|>text<|/ref|><|det|>[[88, 258, 875, 314]]<|/det|> +56. P. G. Silver, T. H. Jordan, Optimal estimation of scalar seismic moment. Geophys. J. Int. 70(3), 755-787 (1982). + +<|ref|>text<|/ref|><|det|>[[88, 335, 905, 392]]<|/det|> +57. C. H. Scholz, J. Campos, The seismic coupling of subduction zones revisited. J. Geophys. Res. 117, B05310 (2012). + +<|ref|>text<|/ref|><|det|>[[88, 412, 879, 469]]<|/det|> +58. G. C. McLaskey, D. A. Lockner, Preslip and cascade processes initiating laboratory stick slip. J. Geophys. Res. 119, 6323-6336 (2014). + +<|ref|>text<|/ref|><|det|>[[88, 490, 708, 510]]<|/det|> +59. https://www.continuummechanics.org/strain.html accessed 22 March 2021 + +<|ref|>text<|/ref|><|det|>[[88, 531, 902, 589]]<|/det|> +60. M. H. Lamorde, "Development and application of a novel approach to sand production prediction", thesis, Heriot-Watt University (2015). + +<|ref|>sub_title<|/ref|><|det|>[[89, 661, 329, 681]]<|/det|> +## Supplementary Information + +<|ref|>text<|/ref|><|det|>[[88, 702, 523, 722]]<|/det|> +Supplementary Note: Feedback control of deformation + +<|ref|>text<|/ref|><|det|>[[89, 744, 316, 764]]<|/det|> +Supplementary Figures 1- 14 + +<|ref|>text<|/ref|><|det|>[[89, 787, 300, 806]]<|/det|> +Supplementary Tables 1- 2 + +<|ref|>text<|/ref|><|det|>[[89, 829, 306, 848]]<|/det|> +Supplementary Movies 1- 8 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[91, 198, 840, 384]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 406, 899, 671]]<|/det|> +
Fig. 1. Bulk mechanical behaviour of Clashach sandstone. (a) Differential stress and AE event rate evolution with time. The transition from constant strain rate loading ( \(10^{-5} / \mathrm{s}\) ) to constant AE event rate loading (1 AE /s) occurred early in stage (ii). During stage (iv), shear zone propagation occurs first across the sample, and then down the sample. (b) Photograph of the failed sample showing the localised shear damage zone. (b) Differential stress plotted against axial strain. Letters a–g refer to the image labels in Figs. 2 and 3. Young’s modulus, \(E = 19.369 \pm 0.028 \mathrm{GPa}\) , was calculated over the range shown. AE activity began at 40% of peak stress, \(\sigma_{\mathrm{P}}\) , with sample yield at \(0.85\sigma_{\mathrm{P}}\) . The AE feedback control (1 AE/s) modulated the strain rate from \(0.93\sigma_{\mathrm{P}}\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[92, 123, 905, 725]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 757, 907, 831]]<|/det|> +
Fig. 2. Micro-scale damage evolution close to and following peak stress. (a-i) Reconstructed 2D \(\mu \mathrm{CT}\) slice (x,y-oriented, where x and y are perpendicular to each other and to the direction of loading, and x is across-strike and y is along-strike of the shear zone) showing
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[81, 110, 912, 595]]<|/det|> +the plane (orange line) of a re- slice of the original \(\mu \mathrm{CT}\) volume (x,z- oriented, where z is the direction of loading) where the shear zone initially localised. The corresponding across- strike (x,z- oriented) re- slice is shown between the grey arrows with the shear zone highlighted by dash- dot lines. (a- ii) histogram of greyscale x- ray intensity values (a proxy for density) showing evolution within the shear zone region (white line in inset image) and an increasing number of partial volume voxels representing increasing damage in the shear zone. (a- iii) Zoomed- in view (region between the grey arrows) of the x,z- oriented slice in (A- i) highlighting the narrow shear zone (dash- dot line) formed after peak stress, and the region of damage that formed after yield but before peak stress (dotted circle). (b) Further zoomed- in view of the across- strike slices (x,z- oriented) showing shear zone emergence and development in region shown by dashed box in (a- iii) for scans a- g (see also Figs. 1c and 3). (c) Even further zoomed- in slices (x,z- oriented) highlighting the variety of micro- mechanisms involved in shear zone formation: numbers (i)- (iii) correspond to the dashed boxes in (b; slice a). When the whole time- series is viewed as an animation (Supplementary Movies 1 and 2), the micro- mechanisms illustrated by the annotations are apparent. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 99, 876, 655]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 670, 880, 866]]<|/det|> +
Fig. 3. 3D incremental strain fields from the onset of strain localisation (marked in Fig. 1c). Incremental dilation, \(\Delta \epsilon_{\mathrm{d}}\) (blue-pink) and shear strain, \(\Delta \epsilon_{\mathrm{s}}\) (yellow-green) were calculated from digital volume correlation between successive pairs of \(\mu \mathrm{CT}\) volumes and are shown (a) parallel to strike (y,z orientation) and (b) perpendicular to strike (x,z orientation). The lower threshold of 0.0017 was set at four standard deviations from the mean error distribution of \(\Delta \epsilon_{\mathrm{s}}\) (Supplementary Figure S6) and the upper threshold shows regions with strain \(>0.01\) (maximum
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 857, 178]]<|/det|> +\(\Delta \epsilon_{\mathrm{s}}\) and \(\Delta \epsilon_{\mathrm{d}}\) were \(\sim 0.04\) ; Supplementary Figures S4 and S5). The thresholds were chosen to visually highlight regions of localised strain. Letters a- g correspond to those in Figs. 1c and 2, with the strain increment following the scan. Strain localisation began at \(0.7\sigma_{\mathrm{P}}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[230, 109, 740, 420]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 455, 872, 650]]<|/det|> +
Fig. 4. Relationship between AE amplitude (in log-scale) and local incremental strain at the AE location. (a) Incremental shear strain, \(\Delta \epsilon_{\mathrm{s}}\) , and (b) incremental dilation, \(\Delta \epsilon_{\mathrm{d}}\) , each normalized by their respective maximum. The solid black ellipses highlight the lack of large AE events at locations of large strain; evidence for predominantly aseismic deformation. Conversely, the dashed black ellipses highlight locations of large strain which are connected to mostly smaller AE events. Larger AE events are mostly associated with smaller strains.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 90, 868, 515]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 538, 871, 870]]<|/det|> +
Fig. 5. Shear fracture energy estimated from bulk strain and from local incremental strain in the shear zone. (a) Procedure for estimating \(\mathrm{G_c}\) from shear stress and inelastic axial displacement resolved on the fault (reproduced from Fig. 1 in 24) showing (top) schematic of sample with boundary stresses \(\sigma_{1}\) (axial) and \(\sigma_{3}\) (radial), the angle \(\theta\) that fault F makes with \(\sigma_{1}\) , and the shear stress \(\tau\) acting on the fault; (bottom left) obtaining inelastic axial displacement \(\Delta \mathrm{l}\) from the differential stress vs axial displacement curve and then resolving it onto F to obtain the relative slip \(\Delta \mathrm{u}\) ; and (bottom right) obtaining \(\mathrm{G_c}\) by integration under \(\tau\) - \(\Delta \mathrm{u}\) curve from peak shear stress \(\tau_{\mathrm{P}}\) to residual frictional strength \(\tau_{\mathrm{F}}\) . (b) Relationship between \(\tau\) and \(\Delta \mathrm{u}\) for \(\theta = 30.3^{\circ}\) (Supplementary Table 1) for (i) yellow; uniform slip from average bulk axial stress and strain data, (ii) orange; uniform slip from bulk axial stress and strain data at each \(\mu \mathrm{CT}\) scan time, and
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 880, 214]]<|/det|> +(iii) green; total observed local slip in the developing shear zone, obtained from direct dilation and shear strain measurements. We obtained \(G_{\mathrm{c}}\) by integration under the \(\tau\) - \(\Delta u\) curve from peak \(\tau\) to \(\tau_{\mathrm{F}}\) (dotted black line). The contributions to (iii) from local shear strain (blue) and local dilation (purple) are also shown. (C) Evolution of \(G_{\mathrm{c}}\) calculated from (B) as \(\Delta u\) increases. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[58, 130, 520, 365]]<|/det|> +- supplementarymovie1.mp4- supplementarymovie2.mp4- supplementarymovie3.mp4- supplementarymovie4.mp4- supplementarymovie5.mp4- supplementarymovie6.mp4- supplementarymovie7.mp4- cartwrighttayloretalsupplementaryinformation.pdf- supplementarymovie8.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/images_list.json b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..5aeaffc77b085439d0be290d180b8cf3729ea4fb --- /dev/null +++ b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/images_list.json @@ -0,0 +1,122 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Overview of molecular silicon clusters. Selected examples of saturated (I - III), siliconoxide (IV - VI) and Zintl-type (VII, VIII) clusters. Silicon is depicted as blue, \\(\\mathrm{SiMe_2}\\) units as yellow and Si-SiCl3 units as red spheres. TMS = Trimethylsilyl; \\(\\mathrm{Dis} = \\mathrm{TMS}_2\\mathrm{HC}\\) ; Mes = 2,4,6-Trimethylphenyl; Tip = 2,4,6- Tri-iso-propylphenyl; Dipp = 2,6-Di-iso-propylphenyl; \\(\\mathrm{MeHyp} = \\mathrm{TMS}_3\\mathrm{Si}\\) ; \\(\\mathrm{Cy} = \\mathrm{cyclohexyl}\\) .", + "footnote": [], + "bbox": [ + [ + 140, + 375, + 876, + 750 + ] + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Scheme 1: General synthetic procedure. Synthesis of trisilylated silicon clusters 2a-d via separation of four and nine atomic clusters in liquid ammonia and subsequent silylation.", + "footnote": [], + "bbox": [ + [ + 115, + 231, + 880, + 400 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Pictures of synthetic sequence. Ammonia solution of \\(\\mathsf{K}_{12}\\mathsf{Si}_{17}\\) and 2.2.2-Cryptand after storage at \\(-40^{\\circ}C\\) for twelve hours (left) and dried filtrate \\(\\mathsf{K}[\\mathsf{K}(2.2.2\\text{-} \\mathsf{cypt})]_2[\\mathsf{Si}_9\\mathsf{H}]\\) (orange powder) and dried filtration residue (grey powder) (right).", + "footnote": [], + "bbox": [ + [ + 115, + 555, + 639, + 721 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Stacked RAMAN spectra. a) \\(\\mathsf{K}_{12}\\mathsf{Si}_{17}\\) , b) \\(\\mathsf{K}_4\\mathsf{Si}_4\\) , c) filtration residue, and d) filtrate. Vibrations are given in \\(\\mathrm{cm^{-1}}\\) .", + "footnote": [], + "bbox": [ + [ + 128, + 189, + 867, + 628 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Crystal structure of [Si9H]3- unit in [K(2.2.2-crypt)]4[Si9H]-8.5NH3. left: front view; right: top view. Ellipsoids are shown at \\(50\\%\\) probability level. Selected bond length (A): Si1A-H: 1.55(4); Si1A-Si2B: 2.345(3); Si1A-Si3B: 3.125(4); Si1A-Si4: 2.323(2); Si1A-Si5: 2.4238(12); Si1A-Si6: 2.4338(12); Si2B-Si3B: 2.578(3); Si2B-Si4: 3.751(3); Si2B-Si6: 2.520(2); Si2B-Si7: 2.4637(14); Si3B-Si4: 2.532(2); Si3B-Si7: 2.4555(12); Si3B-Si8: 2.4322(11); Si4-Si5: 2.4912(12); Si4-Si8: 2.4258(10); Si5-Si9: 2.4466(11); Si6-Si9: 2.4312(10); Si7-Si9: 2.4356(13); Si8-Si9: 2.4545(11); Si2B-Si4/Si1A-Si3B: 1.20.", + "footnote": [], + "bbox": [ + [ + 128, + 177, + 874, + 375 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: \\(^{1}\\mathrm{H}\\) NMR of the dried filtrate. (400 MHz, DMF-\\(d_{7}\\) , 300 K).", + "footnote": [], + "bbox": [ + [ + 120, + 275, + 870, + 680 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: ESI(-) MS spectrum of the dried filtrate in MeCN. a) \\(\\{[K(2.2.2\\text{-} \\text{crypt})][\\text{Si9}] + 2\\text{H}\\}^{-}\\) (m/z = 670.37); b) \\(\\{[K(2.2.2\\text{-} \\text{crypt})][\\text{Si9}] + 2\\text{H} + \\text{meen}\\}^{-}\\) (m/z = 711.42).", + "footnote": [], + "bbox": [ + [ + 120, + 81, + 880, + 430 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7: Crystal structure of the anionic unit in [K(2.2.2-crypt)][\\(^{Me}HypsSi_3\\)]-thf (2a). Si is depicted as blue ellipsoids at 50% probability level and carbon grey as wires. Hydrogen atoms, co-crystallised [K(2.2.2-crypt)]+ and thf are omitted for clarity.", + "footnote": [], + "bbox": [ + [ + 118, + 81, + 496, + 317 + ] + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811.mmd b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811.mmd new file mode 100644 index 0000000000000000000000000000000000000000..430328fdfca621a35b5cc8bec158171ff725927f --- /dev/null +++ b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811.mmd @@ -0,0 +1,203 @@ + +# An Efficient Multi-Gram Access in a Two-step Synthesis to Soluble Nine-atomic Silylated Silicon Clusters + +Thomas Fässler + +Thomas.Faessler@lrz.tum.de + +Technical University of Munich https://orcid.org/0000- 0001- 9460- 8882 Kevin Frankiewicz Technical University of Munich https://orcid.org/0000- 0002- 2752- 3632 Nicole Willeit Technical University of Munich Viktor Hlukhyy Technical University of Munich + +## Article + +Keywords: + +Posted Date: March 29th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4073358/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on December 23rd, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 55211- z. + +<--- Page Split ---> + +# An Efficient Multi-Gram Access in a Two-step Synthesis to Soluble Nine-atomic Silylated Silicon Clusters + +Kevin M. Frankiewicz \(^{1,2}\) , Nicole S. Willeit \(^{1,2}\) , Viktor Hlukhyy \(^{1}\) & Thomas F. Fässler \(^{1,2*}\) + +Corresponding author: Thomas F. Fässler + +Tel.: +49 (0)89 289 13131 + +E- mail address: thomas.faessler@lrz.tum.de (Thomas. F. Fässler) + +Technical University of Munich, Chair of Inorganic Chemistry with Focus on Novel Materials, Lichtenbergstraße 4, D- 85748 Garching, Germany + +WACKER Institute of Silicon Chemistry, Technische Universität München, Lichtenbergstraße 4, D- 85748 Garching, Germany + +## Abstract + +Silicon is by far the most important semiconducting material. However, solution- based synthetic approaches for unsaturated silicon- rich molecules require less efficient multi- step syntheses. We report on a straightforward access to soluble, polyhedral \(\mathrm{Si}_9\) clusters from the binary phase \(\mathrm{K}_{12}\mathrm{Si}_{17}\) , which contains both \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) clusters. \([\mathrm{Si}_4]^{4 - }\) ions, characterised by a high charge per atom ratio, behave as strong reducing agents, preventing \([\mathrm{Si}_9]^{4 - }\) from directed reactions. By the here reported separation of \([\mathrm{Si}_4]^{4 - }\) , \(\mathrm{Si}_9\) clusters are isolated as monoprotonated \([\mathrm{Si}_9\mathrm{H}]^{3 - }\) ions on a multi- gram scale and further crystallised as their 2.2.2- Cryptate salt. 20 grams of the product can be obtained through this two- step procedure - a new starting point for silicon \(Zintl\) chemistry such as the first isolation and structural characterisation of a trisilylated \([\mathrm{MeHyp}_3\mathrm{Si}_9]^{- }\) cluster. + +<--- Page Split ---> + +## 26 Introduction + +27 With the progressing technologisation of our society and the accompanying miniaturisation of 28 electronic devices, physicists and chemists face new challenges. Silicon stands out as the most 29 important semiconducting material by far. However, traditional manufacturing methods, such 30 as lithography and etching of crystalline silicon (top- down) or Chemical Vapor Deposition 31 (CVD) of volatile silanes for producing nanostructured components, are reaching their limits. 32 Nevertheless, not only manufacturing these materials requires different approaches. Bulk 33 materials and semiconducting materials in the nanometre range significantly differ in their 34 optical and electronic properties (quantum confinement). Molecular precursors could provide 35 an answer to new manufacturing methods and the necessity for targeted investigations of 36 quantum confinement effects in low- dimensional materials (quantum dots, wires, and wells). 37 Alongside the targeted synthesis of silicon nanoparticles, defined molecular (silicon) clusters 38 are also considered model systems for studying physical and chemical processes in 39 nanomaterials. + +![](images/Figure_1.jpg) + +
Figure 1: Overview of molecular silicon clusters. Selected examples of saturated (I - III), siliconoxide (IV - VI) and Zintl-type (VII, VIII) clusters. Silicon is depicted as blue, \(\mathrm{SiMe_2}\) units as yellow and Si-SiCl3 units as red spheres. TMS = Trimethylsilyl; \(\mathrm{Dis} = \mathrm{TMS}_2\mathrm{HC}\) ; Mes = 2,4,6-Trimethylphenyl; Tip = 2,4,6- Tri-iso-propylphenyl; Dipp = 2,6-Di-iso-propylphenyl; \(\mathrm{MeHyp} = \mathrm{TMS}_3\mathrm{Si}\) ; \(\mathrm{Cy} = \mathrm{cyclohexyl}\) .
+ +45 The targeted synthesis of saturated cage oligosilanes \(^{1,2}\) , unsaturated siliconoids \(^{3 - 5}\) and Zintl 46 clusters \(^{6 - 8}\) has been intensively studied in past decades. In 1970, West and co- workers 47 achieved the preparation of a cage oligosilane under reductive conditions starting from 48 chlorosilane precursors for the first time. \(^{9}\) Such Wurtz- type couplings or metathesis reactions + +<--- Page Split ---> + +also provide access to paradigmatic clusters like silaprismanes \(^{10 - 12}\) , - cubanes \(^{13 - 16}\) , and - tetrahedranes \(^{17,18}\) (II), primarily impressive by their structural beauty. Following West's initial hints towards the synthesis of a permethylated sila- adamantane \(^{19}\) , Marschner et al. established the synthetic pathway to the sila- adamantane derivative I, representing a molecular fragment of the diamond structure of elemental silicon. Via a simple and elegant cascade of silyl abstraction and silylation steps, coupled with subsequent Lewis acid mediated isomerisation, this molecule was obtained in a stepwise synthesis from readily available TMS \(_4\) Si. \(^{20}\) Apart from the synthesis of perchlorocyclohexasilane, the disproportionation of the versatile precursor Si \(_2\) Cl \(_6\) also enables the synthesis of an endohedral, chloride- decorated silafullerane (III). \(^{21 - 23}\) In addition to the described saturated silicon clusters, Breher, Scheschkewitz, and Lips report on unsaturated so- called siliconoid clusters. Breher obtained the silapropellane IV via the co- reduction of Mes \(_2\) SiCl \(_2\) and Si \(_2\) Cl \(_6\) , exhibiting a biradicaloid character of the transannular interaction between both bridgeheads. \(^{24}\) Scheschkewitz's bridged silapropellane V is structurally related \(^{25}\) , which can be derived from aromatic hexasilabenzenene in a thermal or photochemical rearrangement and serves as a starting point for a rich cluster functionalisation and expansion chemistry. \(^{26 - 29}\) Lips et al. demonstrated the suitability of amido ligands \(^{30,31}\) in stabilising six- atomic silicon clusters (VI), accessible via both the reductive coupling of corresponding bromosilane precursors and the thermal transformation of a zwitterionic tetrasilane \(^{32}\) . + +While all discussed routes require multistep syntheses to build up molecules with multiple Si- Si bonds, an alternative method involves the direct formation of bare silicon clusters in a one- step synthesis by the solid- state reaction of silicon with alkali metals. This approach has yielded considerable success in the case of heavier Ge \(_9\) clusters. For instance, the solid- state phase M \(_4\) Ge \(_9\) (M = Na- Cs) \(^{33}\) can easily be obtained and readily dissolve in various polar solvents. In recent years, a rich chemistry has been established around the nine- atomic [Ge \(_9\) ] \(^{4 - }\) cluster anion such as (organo)functionalisation, \(^{34 - 49}\) metalation \(^{42,50 - 57}\) and cluster growth \(^{58 - 64}\) . However, unlike in the case of heavier tetrel elements (Ge- Pb), [Si \(_9\) ] \(^{4 - }\) cannot be selectively obtained in a solid- state reaction. Instead, only such phases with the composition M \(_{12}\) Si \(_{17}\) (equivalent to \(\{(M^{+})_{12}([Si_4]^{4 - })_2([Si_9]^{4 - }); M = Na - Cs\}\) are accessible, which contain also four- atomic [Si \(_4\) ] \(^{4 - }\) clusters in addition to the desired nine- atomic clusters. \(^{65}\) Due to the highly reductive character of these tetrahedral ions, direct reactions of K \(_{12}\) Si \(_{17}\) with electrophilic reagents like chlorosilanes are not feasible. As a result, silicon- based Zintl cluster chemistry lagged behind the heavier homologues up to now and is limited to reactions and studies in liquid ammonia. Apart from characterising bare anions [Si \(_9\) ] \(^{x - }\) (x = 2- 4) \(^{66 - 68}\) monoprotonated ions like [Si \(_4\) H] \(^{3 - 69}\) and [Si \(_9\) H] \(^{3 - 70}\) by Korber et al., there are only six examples of metalated silicon- based Zintl ions, such as [PhZnSi \(_9\) ] \(^{3 - }\) , \(^{71}\) [(Ni(CO) \(_2\) ] \(_2\) (μ- Si \(_9\) ) \(_2\) ] \(^{8 - }\) , \(^{72}\) [NHC \(^{Dipp}\) Cu(η \(^4\) - Si \(_9\) )H] \(^{2 - }\) , \(^{73}\) and [(NHC \(^{tBu}\) Au) \(_6\) (η \(^2\) - Si \(_4\) )] \(^{2 - }\) , \(^{74}\) as well as [(MesCu) \(_2\) Si \(_4\) ] \(^{4 - }\) \(^{75}\) and [NHC \(^{Dipp}\) Cu(η \(^4\) - Si \(_9\) )] \(^{3 - }\) \(^{76}\) . Both the removal of the highly reactive four- atomic silicon clusters and the avoidance of liquid ammonia as a reaction medium are crucial to further develop this kind of chemistry and overcome synthetic limitations. + +<--- Page Split ---> + +89 In a first step, we recently made the nine- atomic [Si9]4- clusters available for reactions in 90 organic solvents by dissolving K12Si17 in ammonia with 2.2.2- Cryptand and subsequent solvent 91 removal. Starting from this so- called activated precursor phase, the extraction of bis- 92 protonated [Si9H2]2- clusters in pyridine and the transformation into disubstituted dianions of 93 the form [R2Si9]2- (R = MeHyp, 'Bu2HSi, Cy3Sn) in THF was achieved.77- 79 However, the four- 94 atomic clusters continue interfering with the respective electrophilic reagents, causing limited 95 functional group tolerance, poor product purity and low yields. 96 In this work, we report now on a wet chemical access to a synthetic K4Si9 analogue via 97 separation of four- and nine- atomic clusters in ammonia solution. Obtaining such a precursor 98 compound represents a key step in the still largely unexplored chemistry of nine- atomic silicon 99 clusters and allows for the selective synthesis of trisilylated [K(2.2.2- crypt)][(R3Si)3Si9] cluster 100 salts. + +<--- Page Split ---> + +## Results + +## Separation of \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) in liquid ammonia. + +Our previous work shows that the solid- state phase \(\mathsf{K}_{12}\mathsf{Si}_{17}\) can be converted into an activated form, accessible for follow- up reactions like silylation. This conversion is done by dissolving \(\mathsf{K}_{12}\mathsf{Si}_{17}\) in liquid ammonia with 2.2.2- Cryptand as sequestering agent and subsequent solvent removal.78,79 Nevertheless, interfering four- atomic clusters might be still present in this activated phase. + +![](images/Figure_unknown_0.jpg) + +
Scheme 1: General synthetic procedure. Synthesis of trisilylated silicon clusters 2a-d via separation of four and nine atomic clusters in liquid ammonia and subsequent silylation.
+ +In order to separate \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) we exploit their different solubilities in liquid ammonia. Keeping the ammonia extract of \(\mathsf{K}_{12}\mathsf{Si}_{17}\) for approximately twelve hours at \(- 40^{\circ}C\) , we observe the formation of a bright red solid under a reddish- brown solution. After filtration of the mixture and removal of the solvent from the filtrate, an orange, coarse solid was isolated, while the former red filtration residue changed into a grey, fine- powdery solid (Figure 2). + +![](images/Figure_2.jpg) + +
Figure 2: Pictures of synthetic sequence. Ammonia solution of \(\mathsf{K}_{12}\mathsf{Si}_{17}\) and 2.2.2-Cryptand after storage at \(-40^{\circ}C\) for twelve hours (left) and dried filtrate \(\mathsf{K}[\mathsf{K}(2.2.2\text{-} \mathsf{cypt})]_2[\mathsf{Si}_9\mathsf{H}]\) (orange powder) and dried filtration residue (grey powder) (right).
+ +Surprisingly, the Raman measurements of the residue and filtrate after solvent removal (Fig. 3d) and filtration residue (Fig. 3c) show a clear separation of the four- and nine- atomic cluster species. For the residue, the most intense resonances at \(477~\mathrm{cm^{- 1}}\) , \(287~\mathrm{cm^{- 1}}\) and \(272~\mathrm{cm^{- 1}}\) can be assigned by comparison with the solid- state phase \(\mathsf{K}_4\mathsf{Si}_4\) (Fig. 3b), which contains exclusively four- atomic \([\mathrm{Si}_4]^{4 - }\) clusters. The slight shift in the resonances is due to the non- identical chemical environment within the crystalline solid and the amorphous filtration residue. In contrast, the Raman spectrum of the filtrate does not show any \(\mathrm{Si}_4\) band, as can be + +<--- Page Split ---> + +detected in \(\mathsf{K}_4\mathsf{Si}_4\) and \(\mathsf{K}_{12}\mathsf{Si}_{17}\) . The resonances at \(294~\mathrm{cm^{- 1}}\) and \(384~\mathrm{cm^{- 1}}\) agree with the Raman data already described by Schering for \(\mathsf{Cs}_4\mathsf{Si}_9\) which was obtained by the thermal decomposition of \(\mathsf{Cs}_4\mathsf{Si}_4\) .80 The third resonance at \(248~\mathrm{cm^{- 1}}\) indicates a further mode due to the different symmetry with respect to \(C_4v\) symmetric [Si9]4- ion. + +![](images/Figure_3.jpg) + +
Figure 3: Stacked RAMAN spectra. a) \(\mathsf{K}_{12}\mathsf{Si}_{17}\) , b) \(\mathsf{K}_4\mathsf{Si}_4\) , c) filtration residue, and d) filtrate. Vibrations are given in \(\mathrm{cm^{-1}}\) .
+ +An elemental analysis of the dried filtrate is in accordance with an approximate composition of \(\mathrm{K}[\mathrm{K}(2.2.2\text{- } \mathrm{crypt})]_2[\mathrm{Si}_9]\) , indicating a diminished cluster charge of 3-. Clusters with this charge have been observed for both paramagnetic [Si9]3- 67 and protonated [Si9H]3- 70 in ammonia. + +Single crystals as orange blocks suitable for SC- XRD were obtained by vapour diffusion of \(\mathrm{Et}_2\mathrm{O}\) into an ammonia solution of the filtrate with an additional equivalent of 2.2.2- Cryptand. The structure determination results in the composition of \([\mathrm{K}(2.2.2\text{- } \mathrm{crypt})]_3[\mathrm{Si}_9\mathrm{H}]\cdot 8.5\mathrm{NH}_3\) (1). The crystal structure analysis unambiguously shows the formation of a three- fold negatively charged cluster. The exceptional good data quality allows for further structure refinement and the localisation of a hydrogen atom at the cluster from the difference Fourier map and a free refinement of the position of the H atom. + +<--- Page Split ---> + +145 The asymmetric unit contains a monoprotonated, threefold negatively charged [Si9H]3- 146 cluster, three [K(2.2.2- crypt)]+ counter ions and 8.5 equivalents of co- crystallised ammonia. 147 The co- crystallised ammonia primarily occupies the voids between the [K(2.2.2- crypt)]+ units 148 and the cluster. This results in a particular thermal and mechanical sensitivity of the crystals. + +![](images/Figure_4.jpg) + +
Figure 4: Crystal structure of [Si9H]3- unit in [K(2.2.2-crypt)]4[Si9H]-8.5NH3. left: front view; right: top view. Ellipsoids are shown at \(50\%\) probability level. Selected bond length (A): Si1A-H: 1.55(4); Si1A-Si2B: 2.345(3); Si1A-Si3B: 3.125(4); Si1A-Si4: 2.323(2); Si1A-Si5: 2.4238(12); Si1A-Si6: 2.4338(12); Si2B-Si3B: 2.578(3); Si2B-Si4: 3.751(3); Si2B-Si6: 2.520(2); Si2B-Si7: 2.4637(14); Si3B-Si4: 2.532(2); Si3B-Si7: 2.4555(12); Si3B-Si8: 2.4322(11); Si4-Si5: 2.4912(12); Si4-Si8: 2.4258(10); Si5-Si9: 2.4466(11); Si6-Si9: 2.4312(10); Si7-Si9: 2.4356(13); Si8-Si9: 2.4545(11); Si2B-Si4/Si1A-Si3B: 1.20.
+ +The refinement shows the presence of an orientationally disordered \(C_s\) symmetric [Si9H]3- 156 cluster (for details, see the Supplementary Discussion). For the main orientation \(\alpha\) (Fig. 4) the 157 refinement allows for the localisation of the proton at the Si1A position with a bond length of 158 1.55(4) Å, which is in the range of typical Si- H bond distances. The cluster framework shows 159 the expected involvement of the substituted Si1A position of the open square plane. Thus, the 160 Si1A- Si2B (2.345(3) Å) and Si1A- Si4 (2.323(2) Å) distances are significantly shortened 161 compared to the Si2B- Si3B (2.578(3) Å) and Si3B- Si4 (2.532(2) Å) distances of the open square 162 plane. The ratio of the square diagonals (Si2B- Si4/Si1A- Si3B) of 1.20 clearly shows the 163 deviation from the ideal \(C_{4v}\) symmetry of the parent [Si9]4- ion.68 The structural characteristic 164 of shorter Si- Si bonds at cluster atoms with ligands supports the existence of a H atoms at Si1A 165 and agrees with findings for the solvate [K(DB- 18- crown- 6)][K(2.2.2- crypt)]2[Si9H]·NH3.70 166 The crystallographic data are further supported by mass spectra of acetonitrile (MeCN) 167 solutions of the filtration residue showing protonated Si9 species (Figure 6). Additionally, 1H 168 NMR (Figure 5) of K[K(2.2.2- crypt)]2[Si9H] in DMF- d7 verifies the existence of monoprotonated 169 cluster species. The prominent signal in the high- field region at \(- 1.80\) ppm aligns with the 170 spectral range reported for [Si9H2]2- .77 The characteristic satellite pattern emerges from scalar 171 coupling to all nine cluster atoms. This pattern displays the superposition of all possible 172 isotopologues caused by the low natural abundance of NMR active 29Si (Natural 173 abundancy \(= 4.7\%\) ) in the cluster framework. Consequently, the intense main singlet results 174 from all non- NMR active isotopologues ([28/30Si9H]3- ). While the first set of satellites is due to 175 a doublet splitting of the isotopologues [28/30Si9H]3- , the second is due to a triplet splitting + +<--- Page Split ---> + +176 of the isotopologues \([^{28 / 30}\mathrm{Si}_{7}^{29}\mathrm{Si}_{2}\mathrm{H}]^{3 - }\) . The satellite signals of higher isotopologues are not 177 detectable due to the low natural abundance of \(^{29}\mathrm{Si}\) . A full overview of the statistical intensity 178 distribution for the superposition of all isotopologues is given in the supporting information 179 (Supplementary Fig. 6 and Table 6) and in accordance with previous work. \(^{77}\) The exceptionally 180 small coupling constant of \(J(^{1}\mathrm{H},^{29}\mathrm{Si}) = 19.5 \mathrm{Hz}\) and the interaction of the proton with all nine 181 silicon atoms of the cluster framework paints the picture of a highly dynamic system at room 182 temperature. Those migration tendencies have already been described by Korber et al. for 183 \([\mathrm{Si}_{9}\mathrm{H}]^{3 - 70}\) in liquid ammonia at low temperature and by Eichhorn et al. for \([\mathrm{Sn}_{9}\mathrm{R}]^{3 - }\) ( \(\mathrm{R} = \mathrm{H}^{81}\) , 184 \(^{1}\mathrm{Pr}\) , \(\mathrm{SnCy}_{3}^{82}\) ) at room temperature. + +![](images/Figure_5.jpg) + +
Figure 5: \(^{1}\mathrm{H}\) NMR of the dried filtrate. (400 MHz, DMF-\(d_{7}\) , 300 K).
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6: ESI(-) MS spectrum of the dried filtrate in MeCN. a) \(\{[K(2.2.2\text{-} \text{crypt})][\text{Si9}] + 2\text{H}\}^{-}\) (m/z = 670.37); b) \(\{[K(2.2.2\text{-} \text{crypt})][\text{Si9}] + 2\text{H} + \text{meen}\}^{-}\) (m/z = 711.42).
+ +## Reactivity Study of [Si9H]3-. + +The access to isolated \(\text{Si}_9\) clusters on a multi- gram scale, in the absence of highly reductive \(\text{[Si}_4]^{4- }\) clusters, provides promising and well- defined conditions for follow- up reactions of nina- atomic silicon clusters. Silylation reactions of \(\text{K[K(2.2.2- crypt)]}_2[\text{Si}_9\text{H}]\) in tetrahydrofuran (THF) lead to the isolation and the fist structural characterisation of trisilylated cluster salts [K(2.2.2- crypt)][(R3Si)3Si9] (2a- 2d) in good yields as orange- brown solids. All compounds were characterised by NMR and ESI- MS analyses. Yellow block- shaped single crystals of \(\text{[K(2.2.2- crypt)]}[\text{MeHyp}_3\text{Si}_9]\) - thf were grown from a thf solution at \(- 32^{\circ}\text{C}\) over two weeks. The crystals structure of the molecular anion in 2a is depicted in Fig. 7. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7: Crystal structure of the anionic unit in [K(2.2.2-crypt)][\(^{Me}HypsSi_3\)]-thf (2a). Si is depicted as blue ellipsoids at 50% probability level and carbon grey as wires. Hydrogen atoms, co-crystallised [K(2.2.2-crypt)]+ and thf are omitted for clarity.
+ +2a crystallises in the monoclinic space group \(P2_{1} / n\) (14) (a = 15.0913(4) Å, b = 24.7859(6) Å, c = 23.9571(6) Å, \(\alpha = 90^{\circ}\) , \(\beta = 90.959(2)^{\circ}\) , \(\gamma = 90^{\circ}\) , \(V = 8959.9(4) \text{Å}^3\) ) with one trisilylated \([^{Me}HypsSi_3]^{- }\) cluster anion, one [K(2.2.2- crypt)]+ unit and one disordered thf molecule in the asymmetric unit. Analogously to the homologous germanium cluster, the present silicon cluster can also be described as a \(D_{3h}\) symmetric threefold capped trigonal prism. The attachment of a further hypersilyl group to the \(C_{2v}\) symmetric dianion \([^{Me}HypsSi_3]^{2 - 79}\) leads to a closure of the planar square plane Si1- Si4- Si8- Si5 by shortening the Si1- Si8 bond from 3.770(7) Å to 3.2569(14) Å. At the same time, the remaining prism edges (Si2- Si9 and Si3- Si7) are shortened by 0.526 Å and 0.430 Å, respectively. The attachment of the third silyl substituent at the Si\(_9\) cluster induces the same geometric changes that have been described for the homologous \([^{Me}Hyps,Ge_9]^{(4-n)- }\) (Table 1)\(^{40,83}\) and \([^{Me}Hyps,Sn_9]^{(4-n)- }\) clusters (n = 2, 3)\(^{84,85}\). As expected, the cluster framework undergoes a significant contraction from tin and germanium to silicon. + +<--- Page Split ---> + + +Table 1: Selected bond length (A) in 2a and related cluster species. + +
Bond[MeHyp2Sig]2-79[MeHyp3Sig]- (2a)[MeHyp2Ges]2-40[MeHyp3Ges]-83 a)
exo MeHyp-E42.357(5)2.3326(12)2.4071(9)2.3646(14)
exo MeHyp-E52.339(5)2.3264(12)2.4098(9)2.3741(15)
exo MeHyp-E6-2.3258(12)-2.3688(14)
E1-E42.427(5)2.4115(12)2.5528(5)2.5321(8)
E1-E52.398(5)2.4075(13)2.5469(5)2.5213(8)
E4-E82.396(5)2.4050(12)2.5415(5)2.5366(8)
E5-E82.395(4)2.4071(12)2.5486(5)2.5314(8)
E1-E22.488(5)2.4786(13)2.6439(5)2.7081(7)
E1-E32.485(6)2.4981(13)2.6399(5)2.6526(8)
E2-E32.585(6)2.5010(13)2.7046(5)2.6921(8)
E1-E83.770(7)3.2569(14)3.9867(9)3.4966(9)
E2-E92.738(5)3.2639(14)2.8837(5)3.2196(9)
E3-E72.772(6)3.2020(14)2.6005(5)3.4904(9)
E7-E92.569(6)2.5098(12)2.7076(5)2.6640(8)
E2-E62.475(7)2.4100(13)2.5824(5)2.5234(8)
E3-E62.468(6)2.4130(13)2.5872(5)2.5222(7)
E6-E72.470(6)2.4057(12)2.5751(8)2.5238(8)
E6-E92.427(6)2.3997(12)2.5819(5)2.5343(8)
+ +a) Two independent clusters in unit cell. + +As expected from the crystal structure and analogously to [MeHyp3Ges] - 83, 2a behaves also \(D_{3h}\) symmetric on NMR time scale in thf- \(d_8\) at room temperature. Hence, the three hypersily groups collapse to two resonances at \(- 130.03\) ppm and \(- 8.71\) ppm. These signals can be attributed to the exo- bonded silicon atoms (STMS3) and the TMS groups, respectively. Further, in the high field at \(- 171.3\) ppm, the three capping positions show one signal, while + +<--- Page Split ---> + +the six equivalent prism positions exhibit a strongly shielded signal at \(- 352.8 \text{ppm}\) . The NMR data are consistent with the data described in our previous studies. \(^{79}\) + +Table 2: Selected spectroscopic data of 2a-2d. \(^{29}\mathrm{Si}\) NMR shifts of \(\mathrm{Si}_{\mathrm{cap}}\) (red) and \(\mathrm{Si}_{\mathrm{prism}}\) (blue) and homonuclear \(J\) coupling (298 K, 79.5 MHz, thf-\(d_8\) ). + +
R3Siδ(SiCap) / ppmδ(SiPrism) / ppm1J(29SiCap, 29SiPrism) / Hz[K(2.2.2-crypt)]+
MeHyp (2a)-171.3-352.840.1Si3Si
EtHyp (2b)-171.3-352.842.7
MeHypMe2Si (2c)-146.3-347.424.4
tBu2FSi (2d) a)-177.7-357.3723.2
+ +a) \(^{1}J(^{19}\mathrm{F},^{29}\mathrm{Si}) = 340.9 \mathrm{Hz};^{2}J(^{19}\mathrm{F},^{29}\mathrm{Si}) = 19.4 \mathrm{Hz}.\) + +The remaining derivatives 2b- d exhibit similar behaviour, with the cap and prism signals falling within the characteristic range of \(- 160 \text{ppm}\) and \(- 350 \text{ppm}\) . The high quality of the data enables the determination of the \(^{1}J(^{29}\mathrm{Si},^{29}\mathrm{Si})\) homonuclear couplings between the cap and prism atoms (Table 2). The coupling constants of the sterically demanding hypersilyl groups (40.1 Hz for 2a and 42.7 Hz 2b) are significantly higher than those of the sterically less demanding silyl groups in 2c (24.4 Hz) and 2d (23.2 Hz). These couplings differ from localised Si- Si bonds as in cyclic oligosilanes ( \(^{1}J(^{29}\mathrm{Si},^{29}\mathrm{Si}) \approx 50 - 70 \text{Hz})^{86}\) and may indicate possible dynamic processes in the cluster framework, which have not yet been described in homologous silylated clusters. The access to NMR active cluster frameworks could reveal processes that have remained hidden to us so far. + +<--- Page Split ---> + +## Conclusion + +ConclusionThe presence of strongly reductive \([Si_4]^{4- }\) clusters in the solid- state phase \(\mathsf{K}_{12}\mathsf{Si}_{17}\) limits the directed conversion of nine- atomic \([Si_9]^{4- }\) silicon clusters. However, in this work we have shown that the separation of both cluster species is easily possible on a multi- gram scale in liquid ammonia and provides a valuable synthetic access to \([Si_9H]^{3- }\) ions for the first time. \([Si_9H]^{3- }\) shows a pronounced tautomerisation tendency, in which the proton rapidly migrates over the entire nine- atomic cluster framework. In addition, those monoprotonated silicon clusters in the form of the salt \(\mathsf{K}[\mathsf{K}(2.2.2\text{- } \mathsf{crypt})_2[Si_9H]\) represent a synthetic equivalent to \([Si_9]^{4- }\) ions that are still not accessible in an isolated form via a solid- state approach. Thus, the trisilylated cluster salts \([\mathsf{K}(2.2.2\text{- } \mathsf{crypt})][(\mathsf{R}_3\mathsf{Si})_3\mathsf{Si}_9]\) (2a- d) are obtained in good yields and high purity by direct silylation of \(\mathsf{K}[\mathsf{K}(2.2.2\text{- } \mathsf{crypt})_2[Si_9H]\) with the corresponding chlorosilanes. The spectroscopic behaviour as well as the crystallographic characterisation of 2a proofs the strong similarities between silicon and germanium based Zintl clusters. With the present work, we were able to close a significant gap in the chemistry of group 14 Zintl ions. Studies on the further reactivity of the obtained trisilylated monoanions are underway. + +<--- Page Split ---> + +262 References263 1 Sekiguchi, A. & SAKURA, H. Cage and Cluster Compounds of. Adv. Organomet. Chem., 264 1, (1995).265 2 Lee, V. Y. Organosilicon Compounds: Theory and experiment (synthesis). (Academic 266 Press, 2017).267 3 Heider, Y. & Scheschkewitz, D. Molecular Silicon Clusters. Chem. Rev. 121, 9674- 9718, 268 (2021).269 4 Heider, Y. & Scheschkewitz, D. Stable unsaturated silicon clusters (siliconoids). Dalton 270 Trans. 47, 7104- 7112, (2018).271 5 Schnepf, A. Metalloid group 14 cluster compounds: An introduction and perspectives 272 to this novel group of cluster compounds. Chem. Soc. Rev. 36, 745- 758, (2007).273 6 Scharfe, S., Kraus, F., Stegmaier, S., Schier, A. & Fässler, T. F. Zintl Ions, Cage 274 Compounds, and Intermetalloid Clusters of Group 14 and Group 15 Elements. Angew. 275 Chem. Int. Ed. 50, 3630- 3670, (2011).276 7 Liu, C. & Sun, Z.- M. Recent advances in structural chemistry of Group 14 Zintl ions. 277 Coord. Chem. Rev. 382, 32- 56, (2019).278 8 Wilson, R. J., Weinert, B. & Dehnen, S. Recent developments in Zintl cluster chemistry. 279 Dalton Trans. 47, 14861- 14869, (2018).280 9 West, R. & Indriksons, A. Cyclic polysilanes. V. Tetradecylmethylbicyclo [2.2. 2] 281 octasilane, a cage polysilane. J. Am. Chem. Soc. 92, 6704- 6705, (1970).282 10 Sekiguchi, A., Yatabe, T., Kabuto, C. & Sakurai, H. Chemistry of organosilicon 283 compounds. 303. The missing hexasilaprismane: synthesis, x- ray analysis and 284 photochemical reactions. J. Am. Chem. Soc. 115, 5853- 5854, (1993).285 11 Abersfelder, K. et al. Contraction and Expansion of the Silicon Scaffold of Stable Si6R6 286 Isomers. J. Am. Chem. Soc. 134, 16008- 16016, (2012).287 12 Li, Y., Li, J., Zhang, J., Song, H. & Cui, C. Isolation of R6Si6 Dianion: A Bridged Tricyclic 288 Isomer of Dianionic Hexasilabenzen. J. Am. Chem. Soc. 140, 1219- 1222, (2018).289 13 Matsumoto, H. et al. The first octasilacubane system: synthesis of octakis- (t- 290 butyldimethylsilyl)pentacyclo[4.2.0.02,5.03,8.04,7]octasilane. Chem. Commun., 1083- 1084, (1988).291 14 Matsumoto, H., Higuchi, K., Kyushin, S. & Goto, M. Octakis(1,1,2- 293 trimethylpropyl)octasilacubane: Synthesis, Molecular Structure, and Unusual 294 Properties. Angew. Chem. Int. Ed. 31, 1354- 1356, (1992).295 15 Sekiguchi, A., Yatabe, T., Kamatani, H., Kabuto, C. & Sakurai, H. Chemistry of 296 organosilicon compounds. 293. Preparation, characterization, and crystal structures of 297 octasilacubanes and octagermacubanes. J. Am. Chem. Soc. 114, 6260- 6262, (1992).298 16 Furukawa, K., Fujino, M. & Matsumoto, N. Cubic silicon cluster. Appl. Phys. Lett. 60, 299 2744- 2745, (1992).300 17 Wiberg, N., Finger, C. M. M. & Polborn, K. Tetrakis(tri- tert- butylsilyl)- tetrahedro- 301 tetrasilane (tBu3Si)4Si4: The First Molecular Silicon Compound with a Si4 Tetrahedron. 302 Angew. Chem. Int. Ed. 32, 1054- 1056, (1993).303 18 Ichinohe, M., Toyoshima, M., Kinjo, R. & Sekiguchi, A. Tetrasilatetrahedranide: A 304 Silicon Cage Anion. J. Am. Chem. Soc. 125, 13328- 13329, (2003).305 19 West, R. & Indriksons, A. Cyclic polysilanes. VI. Bicyclic and cage 306 permethylcyclopolysilanes. J. Am. Chem. Soc. 94, 6110- 6115, (1972).307 20 Fischer, J., Baumgartner, J. & Marschner, C. Synthesis and Structure of Sila- 308 Adamantane. Science 310, 825- 825, (2005). + +<--- Page Split ---> + +309 21 Tillmann, J. et al. One-Step Synthesis of a [20]Silafullerane with an Endohedral Chloride 310 lon. Angew. Chem. Int. Ed. 54, 5429- 5433, (2015). 311 22 Bamberg, M. et al. [Cl@Si2OH20]–: Parent Siladodecahedrane with Endohedral 312 Chloride lon. J. Am. Chem. Soc. 143, 10865- 10871, (2021). 313 23 Bamberg, M. et al. Regioselective Derivatization of Silylated [20]Silafulleranes. J. Am. 314 Chem. Soc. 145, 11440- 11448, (2023). 315 24 Nied, D., Köppe, R., Klopper, W., Schnöckel, H. & Breher, F. Synthesis of a 316 Pentasilapropellane. Exploring the Nature of a Stretched Silicon- Silicon Bond in a 317 Nonclassical Molecule. J. Am. Chem. Soc. 132, 10264- 10265, (2010). 318 25 Abersfelder, K., White, A. J. P., Berger, R. J. F., Rzepa, H. S. & Scheschkewitz, D. A Stable 319 Derivative of the Global Minimum on the Si6H6 Potential Energy Surface. Angew. 320 Chem. Int. Ed. 50, 7936- 7939, (2011). 321 26 Leszczyńska, K. I. et al. Atomically Precise Expansion of Unsaturated Silicon Clusters. 322 Angew. Chem. Int. Ed. 58, 5124- 5128, (2019). 323 27 Willmes, P. et al. Isolation and Versatile Derivatization of an Unsaturated Anionic 324 Silicon Cluster (Siliconoid). Angew. Chem. Int. Ed. 55, 2907- 2910, (2016). 325 28 Heider, Y., Willmes, P., Huch, V., Zimmer, M. & Scheschkewitz, D. Boron and 326 Phosphorus Containing Heterosiliconoids: Stable p- and n- Doped Unsaturated Silicon 327 Clusters. J. Am. Chem. Soc. 141, 19498- 19504, (2019). 328 29 Poitiers, N. E., Giarrana, L., Huch, V., Zimmer, M. & Scheschkewitz, D. Exohedral 329 functionalization vs. core expansion of siliconoids with Group 9 metals: catalytic 330 activity in alkene isomerization. Chem. Sci. 11, 7782- 7788, (2020). 331 30 Helmer, J., Hepp, A., Berger, R. J. F. & Lips, F. Synthesis and functionalization of the six- 332 vertex anionic amido- substituted silicon cluster [Si6{N(SiMe3)Ph}5]–. Dalton Trans. 52, 1339- 14945, (2023). 333 31 Keuter, J., Hepp, A., Daniliuc, C. G., Feldt, M. & Lips, F. Cycloadditions with a Stable 334 Charge- Separated Cyclobutadiene- Type Amido- Substituted Silicon Ring Compound. 335 Angew. Chem. Int. Ed. 60, 21761- 21766, (2021). 336 32 Keuter, J. et al. A highly unsaturated six- vertex amido- substituted silicon cluster. Chem. 337 Sci. 11, 5895- 5901, (2020). 338 33 Von Schnering, H. G. et al. Binary alkali metal compounds with the zintl anions [Ge9]4- 340 and [Sn9]4–. Zeitschrift für anorganische und allgemeine Chemie 623, 1037- 1039, 341 (1997). 342 34 Benda, C. B., Wang, J.- Q., Wahl, B. & Fässler, T. F. Syntheses and 1H NMR Spectra of 343 Substituted Zintl Ions [Ge9Rn](4–n)–: Crystal Structures of [Ge9R]3– (R = 2,4,6- 344 Me3C6H2, CHCH2) and Indication of Tris- Vinylated Clusters. Eur. J. Inorg. Chem. 2011, 345 4262- 4269, (2011). 346 35 Bentlohner, M. M., Klein, W., Fard, Z. H., Jantke, L.- A. & Fässler, T. F. Linking 347 Deltahedral Zintl Clusters with Conjugated Organic Building Blocks: Synthesis and 348 Characterization of the Zintl Triad [R- Ge9- CH@CH@CH@CH- Ge9- R]4–. Angew. Chem. Int. 349 Ed. 54, 3748- 3753, (2015). 350 36 Hull, M. W. & Sevov, S. C. Functionalization of Nine- Atom Deltahedral Zintl Ions with 351 Organic Substituents: Detailed Studies of the Reactions. J. Am. Chem. Soc. 131, 9026- 352 9037, (2009). 353 37 Frischhut, S., Bentlohner, M. M., Klein, W. & Fässler, T. F. Synthesis of Zintl Triads 354 Comprising Extended Conjugated π-Electronic Systems: [RGe9- CH=CH- CH=CH- 355 Ge9R]4– (R = - CH=CH2, - C(CH3)=CH- CH=N(CH2)2NH2). Inorg. Chem. 56, 10691- 356 10698, (2017). + +<--- Page Split ---> + +357 38 Benda, C. B., He, H., Klein, W., Somer, M. & Fässler, T. F. Bisvinylated [R- Ge9- Ge9- 358 R]4- Cluster Dimers. Z. Anorg. Allg. Chem. 641, 1080- 1086, (2015). 359 39 Hull, M. W. & Sevov, S. C. Addition of Alkenes to Deltahedral Zintl Clusters by Reaction 360 with Alkynes: Synthesis and Structure of [FcCHCHCHGe9CHCHCHFc]2-, an Organo- 361 Zintl- Organometallic Anion. Angew. Chem. Int. Ed. 46, 6695- 6698, (2007). 362 40 Kysliak, O. & Schnepf, A. {Ge9[Si(SiMe3)3]2}2- : a starting point for mixed substituted 363 metalloid germanium clusters. Dalton Trans. 45, 2404- 2408, (2016). 364 41 Mayer, K., Schiegerl, L. J., Kratky, T., Günther, S. & Fässler, T. F. Targeted attachment 365 of functional groups at Ge9 clusters via silylation reactions. Chem. Commun. 53, 11798- 366 11801, (2017). 367 42 Schiegerl, L. J., Geitner, F. S., Fischer, C., Klein, W. & Fässler, T. F. Functionalization of 368 [Ge9] with Small Silanes:{Ge9(SiR3)3}- (R = iBu, iPr, Et) and the Structures of 369 (CuNHCdipp)[Ge9(Si(iBu)3)3], (K- 18c6)Au[Ge9(Si(iBu)3)3]2, and (K- 370 18c6)2[Ge9(Si(iBu)3)2]. Z. Anorg. Allg. Chem. 642, 1419- 1426, (2016). 371 43 Kysliak, O., Kunz, T. & Schnepf, A. Metalloid Ge9R3- Clusters with Various Silyl 372 Substituents: From Shielded to Open Cluster Cores. Eur. J. Inorg. Chem. 2017, 805- 810, 373 (2017). 374 44 Kysliak, O., Schrenk, C. & Schnepf, A. {Ge9[Si(SiMe3)2(SiPh3)]3}-: Ligand Modification 375 in Metalloid Germanium Cluster Chemistry. Inorg. Chem. 54, 7083- 7088, (2015). 376 45 Kysliak, O., Schrenk, C. & Schnepf, A. {[Si(SiMe3)3]2Ge9- SiMe2- (C6H4)- SiMe2- 377 Ge9[Si(SiMe3)3]2K}-: The Connection of Metalloid Clusters via an Organic Linker. Inorg. Chem. 56, 9693- 9697, (2017). 378 Frischhut, S., Machado de Carvalho, J. G., Karttunen, A. J. & Fässler, T. F. On the Affinity 379 between Fullerenes and Deltahedral Zintl Ions: A UV/Vis Spectroscopic Investigation. 380 Z. Anorg. Allg. Chem. 644, 1337- 1343, (2018). 381 Li, F. & Sevov, S. C. Synthesis, Structures, and Solution Dynamics of Tetrasubstituted 382 Nine- Atom Germanium Deltahedral Clusters. J. Am. Chem. Soc. 136, 12056- 12063, 383 (2014). 384 Frischhut, S. & Fässler, T. F. Synthesis of low- oxidation- state germanium clusters 385 comprising a functional anchor group - synthesis and characterization of [(Ge0)5(Ge- 386 R)3(Ge- (CH2)n- CH \(\checkmark\) CH2)] with R = Si(SiMe3)3. Dalton Trans. 47, 3223- 3226, (2018). 387 Frischhut, S., Klein, W., Drees, M. & Fässler, T. F. Acylation of Homoatomic Ge9 Cages 388 and Subsequent Decarbonylation. Chem. Eur. J. 24, 9009- 9014, (2018). 389 Frischhut, S. et al. Capping nido- Nonagermanide Clusters with M- PPh3 and Dynamics 390 in Solution: Synthesis and Structure of closo- [(Me3Si)3Si]3Et[Ge9M](PPh3) (M = Ni, Pt). 391 Organometallics 37, 4560- 4567, (2018). 392 Henke, F., Schenk, C. & Schnepf, A. {Si(SiMe3)3]3Ge9M(CO)3- (M = Cr, Mo, W): 393 Coordination Chemistry with metalloid Clusters. Dalton Trans. 40, 6704- 6710, (2011). 394 Schenk, C. & Schnepf, A. {Ge9R3Cr(CO)5}- and {Ge9R3Cr(CO)3}-: a metalloid cluster 395 (Ge9R3- ) as a flexible ligand in coordination chemistry [R = Si(SiMe3)3]. Chem. 396 Commun., 3208- 3210, (2009). 397 Li, F., Muñoz- Castro, A. & Sevov, S. C. [(Me3Si)Si]3EtGe9Pd(PPh3), a 398 Pentafunctionalized Deltahedral Zintl Cluster: Synthesis, Structure, and Solution 399 Dynamics. Angew. Chem. Int. Ed. 55, 8630- 8633, (2016). 400 Willeit, N. S., Klein, W., Coburger, P., Fritz- Langhals, E. & Fässler, T. F. Functionalised 401 [Ge9Ni] Clusters as Homogeneous Single- Site Catalysts for Olefin Isomerisation 402 Reactions. ChemCatChem n/a, e202301200. + +<--- Page Split ---> + +404 55 Schiegerl, L. J. et al. Silylated Ge(9) Clusters as New Ligands for Cyclic (Alkyl)amino and 405 Mesoionic Carbene Copper Complexes. Inorg. Chem. 58, 3256- 3264, (2019). 406 56 Gienger, C., Schynowski, L., Schaefer, J., Schrenk, C. & Schnepf, A. New Intermetalloid 407 Ge9- clusters with Copper and Gold: Filling Vacancies in the Cluster Chemistry of 408 [Ge9(Hyp)3]– (Hyp=Si(SiMe3)3). Eur. J. Inorg. Chem. 26, e202200738, (2023). 409 57 Mayer, K., Schiegerl, L. J. & Fässler, T. F. On the Reactivity of Silylated Ge9 Clusters: 410 Synthesis and Characterization of [ZnCp*(Ge9(Si(SiMe3)3)3)], 411 [CuPiPr3(Ge9(Si(SiMe3)3)3)], and [(CuPiPr3)4(Ge9(SiPh3)2)2]. Chem. Eur. J. 22, 18794- 412 18800, (2016). 413 58 Ugrinov, A. & Sevov, S. C. [Ge9Ge9Ge9]6- : A Linear Trimer of 27 Germanium Atoms. J. 414 Am. Chem. Soc. 124, 10990- 10991, (2002). 415 59 Nienhaus, A., Hoffmann, S. D. & Fässler, T. F. First Synthesis of Group- 14 Polyanions by 416 Extraction of a Binary Alloy with dmf and a Novel Conformation of the (Ge9- Ge9)6- 417 Dimer: Crystal Structures of [K6(Ge9- Ge9)](dmf)12, [Rb6(Ge9- Ge9)](dmf)12 and 418 [K2.5Cs3.5(Ge9- Ge9)](dmf)12. Z. Anorg. Allg. Chem. 632, 1752- 1758, (2006). 419 60 Hauptmann, R. & Fässler, T. F. Low Dimensional Arrangements of the Zintl Ion [Ge9- 420 Ge9]6— and Chemical Bonding in [Ge6]2—, [Ge9- Ge9]6—, and {[Ge9]}]2—. Z. Anorg. 421 Allg. Chem. 629, 2266- 2273, (2003). 422 61 Yong, L., Hoffmann, S. D. & Fässler, T. F. The Controlled Oxidative Coupling of Ge94- 423 Zintl Anions to a Linear Trimer [Ge9=Ge9=Ge9]6- . Z. Anorg. Allg. Chem. 631, 1149- 424 1153, (2005). 425 62 Xu, L. & Sevov, S. C. Oxidative coupling of deltahedral [Ge9] 4- Zintl ions. J. Am. Chem. 426 Soc. 121, 9245- 9246, (1999). 427 63 Scharfe, S. & Fässler, Thomas F. Synthesis of (Ge9- Ge9)6- Dimeric Zintl Ions in Liquid 428 Ammonia Solutions of K4Ge9: Low- Dimensional Coordination Networks in the Crystal 429 Structure of the Ammoniates Kn[K([2.2.2]crypt)]6- n[Ge9- Ge9](NH3)m (n = 2, 3, and 430 4). Z. Anorg. Allg. Chem. 637, 901- 906, (2011). 431 64 Ugrinov, A. & Sevov, S. C. Synthesis of a chain of nine- atom germanium clusters 432 accompanied with dimerization of the sequestering agent. C. R. Chim. 8, 1878- 1882, 433 (2005). 434 65 Hoch, C., Wendorff, M. & Röhr, C. Synthesis and crystal structure of the tetrelides 435 A12M17 (A=Na, K, Rb, Cs; M=Si, Ge, Sn) and A4Pb9 (A=K, Rb). J. Alloys Compd. 361, 436 206- 221, (2003). 437 66 Goicoechea, J. M. & Sevov, S. C. Ligand- Free Deltahedral Clusters of Silicon in Solution: 438 Synthesis, Structure, and Electrochemistry of Si92. Inorganic Chemistry 44, 2654- 2658, 439 (2005). 440 67 Goicoechea, J. M. & Sevov, S. C. Naked Deltahedral Silicon Clusters in Solution: 441 Synthesis and Characterization of Si93- and Si52. J. Am. Chem. Soc. 126, 6860- 6861, 442 (2004). 443 68 Benda, C. B., Henneberger, T., Klein, W. & Fässler, T. F. [Si4]4- and [Si9]4- Clusters 444 Crystallized from Liquid Ammonia Solution – Synthesis and Characterization of 445 K8[Si4][Si9]·(NH3)14.6. Zeitschrift für anorganische und allgemeine Chemie 643, 146- 446 148, (2017). 447 69 Hastreiter, F. et al. Elusive Zintl Ions [μ-HSi4]3- and [Si5]2- in Liquid Ammonia: 448 Protonation States, Sites, and Bonding Situation Evaluated by NMR and Theory. 449 Angew. Chem. Int. Ed. 58, 3133- 3137, (2019). 450 70 Lorenz, C. et al. The Structure of [HSi9]3- in the Solid State and Its Unexpected Highly 451 Dynamic Behavior in Solution. Angew. Chem. Int. Ed. 57, 12956- 12960, (2018). + +<--- Page Split ---> + +452 71 Goicoechea, J. M. & Sevov, S. C. Organozinc Derivatives of Deltahedral Zintl Ions: Synthesis and Characterization of closo- [E9Zn(C6H5)]3- (E = Si, Ge, Sn, Pb). Organometallics 25, 4530- 4536, (2006). 453 72 Joseph, S. et al. Chemistry with Bare Silicon Clusters in Solution: A Transition- Metal Complex of a Polysilicide Anion. Angew. Chem. Int. Ed. 48, 8770- 8772, (2009). 454 73 Streitferdt, V. et al. NMR- Spectroscopic Detection of an Elusive Protonated and Coinage Metalated Silicide [NHCDippCu(η4- Si9)H]2- in Solution. Eur. J. Chem. 2021, 458 3648- 3690, (2021). 455 74 Tiefenthaler, S. M. et al. On the Reactivity of NHctBuAuCl towards Rb6Cs6Si17: The First Gold- Silicon Cluster [(NHctBuAu)6(η2- Si4)]Cl2- 7NH3 and an Imide Capped Gold Triangle (NHctBuAu)3NHCl. Z. Anorg. Allg. Chem. 646, 1595- 1602, (2020). 456 75 Waibel, M., Kraus, F., Scharfe, S., Wahl, B. & Fässler, T. F. [(MesCu)2(η3- Si4)]4- : A Mesitylcopper- Stabilized Tetrasilicide Tetraanion. Angew. Chem. Int. Ed. 49, 6611- 6615, (2010). 457 76 Geitner, F. S. & Fässler, T. F. Low oxidation state silicon clusters - synthesis and structure of [NHCDippCu(η4- Si9)]3- . Chem. Commun. 53, 12974- 12977, (2017). 458 77 Schiegerl, L. J. et al. Charged Si9 Clusters in Neat Solids and the Detection of [H2Si9]2- in Solution: A Combined NMR, Raman, Mass Spectrometric, and Quantum Chemical Investigation. Angew. Chem. Int. Ed. 57, 12950- 12955, (2018). 459 78 Schiegerl, L. J., Karttunen, A. J., Klein, W. & Fässler, T. F. Anionic Siliconoids from Zintl Phases: R3Si9- with Six and R2Si92- with Seven Unsubstituted Exposed Silicon Cluster Atoms (R=Si(tBu)2H). Chem. Eur. J. 24, 19171- 19174, (2018). 460 79 Schiegerl, L. J., Karttunen, A. J., Klein, W. & Fässler, T. F. Silicon clusters with six and seven unsubstituted vertices via a two- step reaction from elemental silicon. Chem. Sci. 10, 9130- 9139, (2019). 461 80 von Schnering, H. G. et al. Das Clusteranion Si94- . Angew. Chem. 110, 2507- 2509, (1998). 462 81 Kocak, F. S. et al. Surprising Acid/Base and Ion- Sequestration Chemistry of Sn94- : HSn93- , Ni@HSn93- , and the Sn93- Ion Revisited. J. Am. Chem. Soc. 134, 9733- 9740, (2012). 463 82 Kocak, F. S., Zavalij, P. Y., Lam, Y.- F. & Eichhorn, B. W. Substituent- dependent exchange mechanisms in highly fluxional RSn93- anions. Chem. Commun., 4197- 4199, (2009). 464 83 Li, F. & Sevov, S. C. Rational Synthesis of [Ge9[Si(SiMe3)3]3]- from Its Parent Zintl Ion Ge94- . Inorg. Chem. 51, 2706- 2708, (2012). 465 84 Schrenk, C., Winter, F., Pöttgen, R. & Schnepf, A. {Sn9[Si(SiMe3)3]2}- : A Metalloid Tin Cluster Compound With a Sn9 Core of Oxidation State Zero. Inorg. Chem. 51, 8583- 8588, (2012). 466 85 Schrenk, C., Neumaier, M. & Schnepf, A. {Sn9[Si(SiMe3)3]3}- and {Sn8Si[Si(SiMe3)3]3}- : Variations of the E9 Cage of Metalloid Group 14 Clusters. Inorganic Chemistry 51, 3989- 3995, (2012). 467 86 Jenkner, P. K., Spielberger, A., Eibl, M. & Hengge, E. 29Si NMR Spektroskopie von Cyclopentasilanen. Spectrochim. Acta A 49, 161- 165, (1993). + +<--- Page Split ---> + +## Acknowledgement + +The authors are thankful for the financial support by WACKER Chemie AG. The authors thank B.Sc. Vivienne Wolde and B.Sc. Thanh N. Trân for their assistance in the synthesis of 2.2.2- Cryptand. + +## Author contributions + +K.M.F. conceived and performed the syntheses of cluster compounds and collected the single- crystal X- ray data of 1 and 2a, solved and refined the structure of 1, performed the electrospray- ionisation mass spectrometry and prepared samples for further analyses. V. H. solved and refined the structure of 2a. T.F.F. supervised the work. K.M.F wrote the paper. K.M.F and N.S.W conceived and performed the synthesis of 2.2.2- Cryptand. + +## Competing Interests + +The authors declare no competing interests. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- NCOMMS2414778SupportingInformation.pdf- NCOMMS24147781.cif- NCOMMS24147782a.cif + +<--- Page Split ---> diff --git a/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811_det.mmd b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..00a8a293d7a43a524db05c5853f2b2903c81b47d --- /dev/null +++ b/preprint/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811/preprint__11b9e66bad4ba115da6cbbbd336e319b922ab42a87300eaec1f724b205793811_det.mmd @@ -0,0 +1,248 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 903, 207]]<|/det|> +# An Efficient Multi-Gram Access in a Two-step Synthesis to Soluble Nine-atomic Silylated Silicon Clusters + +<|ref|>text<|/ref|><|det|>[[44, 230, 191, 248]]<|/det|> +Thomas Fässler + +<|ref|>text<|/ref|><|det|>[[55, 258, 344, 274]]<|/det|> +Thomas.Faessler@lrz.tum.de + +<|ref|>text<|/ref|><|det|>[[44, 303, 685, 460]]<|/det|> +Technical University of Munich https://orcid.org/0000- 0001- 9460- 8882 Kevin Frankiewicz Technical University of Munich https://orcid.org/0000- 0002- 2752- 3632 Nicole Willeit Technical University of Munich Viktor Hlukhyy Technical University of Munich + +<|ref|>sub_title<|/ref|><|det|>[[44, 503, 104, 520]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 541, 137, 559]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 578, 315, 597]]<|/det|> +Posted Date: March 29th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 616, 475, 635]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4073358/v1 + +<|ref|>text<|/ref|><|det|>[[44, 654, 914, 696]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 715, 535, 734]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 770, 914, 813]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on December 23rd, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 55211- z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[92, 97, 855, 149]]<|/det|> +# An Efficient Multi-Gram Access in a Two-step Synthesis to Soluble Nine-atomic Silylated Silicon Clusters + +<|ref|>text<|/ref|><|det|>[[92, 189, 790, 206]]<|/det|> +Kevin M. Frankiewicz \(^{1,2}\) , Nicole S. Willeit \(^{1,2}\) , Viktor Hlukhyy \(^{1}\) & Thomas F. Fässler \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[174, 231, 515, 248]]<|/det|> +Corresponding author: Thomas F. Fässler + +<|ref|>text<|/ref|><|det|>[[175, 252, 386, 268]]<|/det|> +Tel.: +49 (0)89 289 13131 + +<|ref|>text<|/ref|><|det|>[[175, 273, 707, 290]]<|/det|> +E- mail address: thomas.faessler@lrz.tum.de (Thomas. F. Fässler) + +<|ref|>text<|/ref|><|det|>[[175, 293, 881, 331]]<|/det|> +Technical University of Munich, Chair of Inorganic Chemistry with Focus on Novel Materials, Lichtenbergstraße 4, D- 85748 Garching, Germany + +<|ref|>text<|/ref|><|det|>[[175, 335, 880, 374]]<|/det|> +WACKER Institute of Silicon Chemistry, Technische Universität München, Lichtenbergstraße 4, D- 85748 Garching, Germany + +<|ref|>sub_title<|/ref|><|det|>[[93, 399, 191, 414]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[90, 418, 884, 630]]<|/det|> +Silicon is by far the most important semiconducting material. However, solution- based synthetic approaches for unsaturated silicon- rich molecules require less efficient multi- step syntheses. We report on a straightforward access to soluble, polyhedral \(\mathrm{Si}_9\) clusters from the binary phase \(\mathrm{K}_{12}\mathrm{Si}_{17}\) , which contains both \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) clusters. \([\mathrm{Si}_4]^{4 - }\) ions, characterised by a high charge per atom ratio, behave as strong reducing agents, preventing \([\mathrm{Si}_9]^{4 - }\) from directed reactions. By the here reported separation of \([\mathrm{Si}_4]^{4 - }\) , \(\mathrm{Si}_9\) clusters are isolated as monoprotonated \([\mathrm{Si}_9\mathrm{H}]^{3 - }\) ions on a multi- gram scale and further crystallised as their 2.2.2- Cryptate salt. 20 grams of the product can be obtained through this two- step procedure - a new starting point for silicon \(Zintl\) chemistry such as the first isolation and structural characterisation of a trisilylated \([\mathrm{MeHyp}_3\mathrm{Si}_9]^{- }\) cluster. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[66, 83, 224, 99]]<|/det|> +## 26 Introduction + +<|ref|>text<|/ref|><|det|>[[66, 103, 884, 373]]<|/det|> +27 With the progressing technologisation of our society and the accompanying miniaturisation of 28 electronic devices, physicists and chemists face new challenges. Silicon stands out as the most 29 important semiconducting material by far. However, traditional manufacturing methods, such 30 as lithography and etching of crystalline silicon (top- down) or Chemical Vapor Deposition 31 (CVD) of volatile silanes for producing nanostructured components, are reaching their limits. 32 Nevertheless, not only manufacturing these materials requires different approaches. Bulk 33 materials and semiconducting materials in the nanometre range significantly differ in their 34 optical and electronic properties (quantum confinement). Molecular precursors could provide 35 an answer to new manufacturing methods and the necessity for targeted investigations of 36 quantum confinement effects in low- dimensional materials (quantum dots, wires, and wells). 37 Alongside the targeted synthesis of silicon nanoparticles, defined molecular (silicon) clusters 38 are also considered model systems for studying physical and chemical processes in 39 nanomaterials. + +<|ref|>image<|/ref|><|det|>[[140, 375, 876, 750]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[66, 760, 883, 817]]<|/det|> +
Figure 1: Overview of molecular silicon clusters. Selected examples of saturated (I - III), siliconoxide (IV - VI) and Zintl-type (VII, VIII) clusters. Silicon is depicted as blue, \(\mathrm{SiMe_2}\) units as yellow and Si-SiCl3 units as red spheres. TMS = Trimethylsilyl; \(\mathrm{Dis} = \mathrm{TMS}_2\mathrm{HC}\) ; Mes = 2,4,6-Trimethylphenyl; Tip = 2,4,6- Tri-iso-propylphenyl; Dipp = 2,6-Di-iso-propylphenyl; \(\mathrm{MeHyp} = \mathrm{TMS}_3\mathrm{Si}\) ; \(\mathrm{Cy} = \mathrm{cyclohexyl}\) .
+ +<|ref|>text<|/ref|><|det|>[[66, 826, 883, 907]]<|/det|> +45 The targeted synthesis of saturated cage oligosilanes \(^{1,2}\) , unsaturated siliconoids \(^{3 - 5}\) and Zintl 46 clusters \(^{6 - 8}\) has been intensively studied in past decades. In 1970, West and co- workers 47 achieved the preparation of a cage oligosilane under reductive conditions starting from 48 chlorosilane precursors for the first time. \(^{9}\) Such Wurtz- type couplings or metathesis reactions + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 80, 884, 477]]<|/det|> +also provide access to paradigmatic clusters like silaprismanes \(^{10 - 12}\) , - cubanes \(^{13 - 16}\) , and - tetrahedranes \(^{17,18}\) (II), primarily impressive by their structural beauty. Following West's initial hints towards the synthesis of a permethylated sila- adamantane \(^{19}\) , Marschner et al. established the synthetic pathway to the sila- adamantane derivative I, representing a molecular fragment of the diamond structure of elemental silicon. Via a simple and elegant cascade of silyl abstraction and silylation steps, coupled with subsequent Lewis acid mediated isomerisation, this molecule was obtained in a stepwise synthesis from readily available TMS \(_4\) Si. \(^{20}\) Apart from the synthesis of perchlorocyclohexasilane, the disproportionation of the versatile precursor Si \(_2\) Cl \(_6\) also enables the synthesis of an endohedral, chloride- decorated silafullerane (III). \(^{21 - 23}\) In addition to the described saturated silicon clusters, Breher, Scheschkewitz, and Lips report on unsaturated so- called siliconoid clusters. Breher obtained the silapropellane IV via the co- reduction of Mes \(_2\) SiCl \(_2\) and Si \(_2\) Cl \(_6\) , exhibiting a biradicaloid character of the transannular interaction between both bridgeheads. \(^{24}\) Scheschkewitz's bridged silapropellane V is structurally related \(^{25}\) , which can be derived from aromatic hexasilabenzenene in a thermal or photochemical rearrangement and serves as a starting point for a rich cluster functionalisation and expansion chemistry. \(^{26 - 29}\) Lips et al. demonstrated the suitability of amido ligands \(^{30,31}\) in stabilising six- atomic silicon clusters (VI), accessible via both the reductive coupling of corresponding bromosilane precursors and the thermal transformation of a zwitterionic tetrasilane \(^{32}\) . + +<|ref|>text<|/ref|><|det|>[[110, 479, 884, 920]]<|/det|> +While all discussed routes require multistep syntheses to build up molecules with multiple Si- Si bonds, an alternative method involves the direct formation of bare silicon clusters in a one- step synthesis by the solid- state reaction of silicon with alkali metals. This approach has yielded considerable success in the case of heavier Ge \(_9\) clusters. For instance, the solid- state phase M \(_4\) Ge \(_9\) (M = Na- Cs) \(^{33}\) can easily be obtained and readily dissolve in various polar solvents. In recent years, a rich chemistry has been established around the nine- atomic [Ge \(_9\) ] \(^{4 - }\) cluster anion such as (organo)functionalisation, \(^{34 - 49}\) metalation \(^{42,50 - 57}\) and cluster growth \(^{58 - 64}\) . However, unlike in the case of heavier tetrel elements (Ge- Pb), [Si \(_9\) ] \(^{4 - }\) cannot be selectively obtained in a solid- state reaction. Instead, only such phases with the composition M \(_{12}\) Si \(_{17}\) (equivalent to \(\{(M^{+})_{12}([Si_4]^{4 - })_2([Si_9]^{4 - }); M = Na - Cs\}\) are accessible, which contain also four- atomic [Si \(_4\) ] \(^{4 - }\) clusters in addition to the desired nine- atomic clusters. \(^{65}\) Due to the highly reductive character of these tetrahedral ions, direct reactions of K \(_{12}\) Si \(_{17}\) with electrophilic reagents like chlorosilanes are not feasible. As a result, silicon- based Zintl cluster chemistry lagged behind the heavier homologues up to now and is limited to reactions and studies in liquid ammonia. Apart from characterising bare anions [Si \(_9\) ] \(^{x - }\) (x = 2- 4) \(^{66 - 68}\) monoprotonated ions like [Si \(_4\) H] \(^{3 - 69}\) and [Si \(_9\) H] \(^{3 - 70}\) by Korber et al., there are only six examples of metalated silicon- based Zintl ions, such as [PhZnSi \(_9\) ] \(^{3 - }\) , \(^{71}\) [(Ni(CO) \(_2\) ] \(_2\) (μ- Si \(_9\) ) \(_2\) ] \(^{8 - }\) , \(^{72}\) [NHC \(^{Dipp}\) Cu(η \(^4\) - Si \(_9\) )H] \(^{2 - }\) , \(^{73}\) and [(NHC \(^{tBu}\) Au) \(_6\) (η \(^2\) - Si \(_4\) )] \(^{2 - }\) , \(^{74}\) as well as [(MesCu) \(_2\) Si \(_4\) ] \(^{4 - }\) \(^{75}\) and [NHC \(^{Dipp}\) Cu(η \(^4\) - Si \(_9\) )] \(^{3 - }\) \(^{76}\) . Both the removal of the highly reactive four- atomic silicon clusters and the avoidance of liquid ammonia as a reaction medium are crucial to further develop this kind of chemistry and overcome synthetic limitations. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 82, 884, 336]]<|/det|> +89 In a first step, we recently made the nine- atomic [Si9]4- clusters available for reactions in 90 organic solvents by dissolving K12Si17 in ammonia with 2.2.2- Cryptand and subsequent solvent 91 removal. Starting from this so- called activated precursor phase, the extraction of bis- 92 protonated [Si9H2]2- clusters in pyridine and the transformation into disubstituted dianions of 93 the form [R2Si9]2- (R = MeHyp, 'Bu2HSi, Cy3Sn) in THF was achieved.77- 79 However, the four- 94 atomic clusters continue interfering with the respective electrophilic reagents, causing limited 95 functional group tolerance, poor product purity and low yields. 96 In this work, we report now on a wet chemical access to a synthetic K4Si9 analogue via 97 separation of four- and nine- atomic clusters in ammonia solution. Obtaining such a precursor 98 compound represents a key step in the still largely unexplored chemistry of nine- atomic silicon 99 clusters and allows for the selective synthesis of trisilylated [K(2.2.2- crypt)][(R3Si)3Si9] cluster 100 salts. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 84, 181, 99]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 105, 535, 123]]<|/det|> +## Separation of \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) in liquid ammonia. + +<|ref|>text<|/ref|><|det|>[[115, 126, 883, 227]]<|/det|> +Our previous work shows that the solid- state phase \(\mathsf{K}_{12}\mathsf{Si}_{17}\) can be converted into an activated form, accessible for follow- up reactions like silylation. This conversion is done by dissolving \(\mathsf{K}_{12}\mathsf{Si}_{17}\) in liquid ammonia with 2.2.2- Cryptand as sequestering agent and subsequent solvent removal.78,79 Nevertheless, interfering four- atomic clusters might be still present in this activated phase. + +<|ref|>image<|/ref|><|det|>[[115, 231, 880, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 411, 880, 440]]<|/det|> +
Scheme 1: General synthetic procedure. Synthesis of trisilylated silicon clusters 2a-d via separation of four and nine atomic clusters in liquid ammonia and subsequent silylation.
+ +<|ref|>text<|/ref|><|det|>[[115, 451, 883, 553]]<|/det|> +In order to separate \([\mathrm{Si}_4]^{4 - }\) and \([\mathrm{Si}_9]^{4 - }\) we exploit their different solubilities in liquid ammonia. Keeping the ammonia extract of \(\mathsf{K}_{12}\mathsf{Si}_{17}\) for approximately twelve hours at \(- 40^{\circ}C\) , we observe the formation of a bright red solid under a reddish- brown solution. After filtration of the mixture and removal of the solvent from the filtrate, an orange, coarse solid was isolated, while the former red filtration residue changed into a grey, fine- powdery solid (Figure 2). + +<|ref|>image<|/ref|><|det|>[[115, 555, 639, 721]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 733, 880, 761]]<|/det|> +
Figure 2: Pictures of synthetic sequence. Ammonia solution of \(\mathsf{K}_{12}\mathsf{Si}_{17}\) and 2.2.2-Cryptand after storage at \(-40^{\circ}C\) for twelve hours (left) and dried filtrate \(\mathsf{K}[\mathsf{K}(2.2.2\text{-} \mathsf{cypt})]_2[\mathsf{Si}_9\mathsf{H}]\) (orange powder) and dried filtration residue (grey powder) (right).
+ +<|ref|>text<|/ref|><|det|>[[115, 771, 883, 916]]<|/det|> +Surprisingly, the Raman measurements of the residue and filtrate after solvent removal (Fig. 3d) and filtration residue (Fig. 3c) show a clear separation of the four- and nine- atomic cluster species. For the residue, the most intense resonances at \(477~\mathrm{cm^{- 1}}\) , \(287~\mathrm{cm^{- 1}}\) and \(272~\mathrm{cm^{- 1}}\) can be assigned by comparison with the solid- state phase \(\mathsf{K}_4\mathsf{Si}_4\) (Fig. 3b), which contains exclusively four- atomic \([\mathrm{Si}_4]^{4 - }\) clusters. The slight shift in the resonances is due to the non- identical chemical environment within the crystalline solid and the amorphous filtration residue. In contrast, the Raman spectrum of the filtrate does not show any \(\mathrm{Si}_4\) band, as can be + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 883, 167]]<|/det|> +detected in \(\mathsf{K}_4\mathsf{Si}_4\) and \(\mathsf{K}_{12}\mathsf{Si}_{17}\) . The resonances at \(294~\mathrm{cm^{- 1}}\) and \(384~\mathrm{cm^{- 1}}\) agree with the Raman data already described by Schering for \(\mathsf{Cs}_4\mathsf{Si}_9\) which was obtained by the thermal decomposition of \(\mathsf{Cs}_4\mathsf{Si}_4\) .80 The third resonance at \(248~\mathrm{cm^{- 1}}\) indicates a further mode due to the different symmetry with respect to \(C_4v\) symmetric [Si9]4- ion. + +<|ref|>image<|/ref|><|det|>[[128, 189, 867, 628]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 645, 833, 660]]<|/det|> +
Figure 3: Stacked RAMAN spectra. a) \(\mathsf{K}_{12}\mathsf{Si}_{17}\) , b) \(\mathsf{K}_4\mathsf{Si}_4\) , c) filtration residue, and d) filtrate. Vibrations are given in \(\mathrm{cm^{-1}}\) .
+ +<|ref|>text<|/ref|><|det|>[[113, 670, 883, 733]]<|/det|> +An elemental analysis of the dried filtrate is in accordance with an approximate composition of \(\mathrm{K}[\mathrm{K}(2.2.2\text{- } \mathrm{crypt})]_2[\mathrm{Si}_9]\) , indicating a diminished cluster charge of 3-. Clusters with this charge have been observed for both paramagnetic [Si9]3- 67 and protonated [Si9H]3- 70 in ammonia. + +<|ref|>text<|/ref|><|det|>[[113, 753, 883, 899]]<|/det|> +Single crystals as orange blocks suitable for SC- XRD were obtained by vapour diffusion of \(\mathrm{Et}_2\mathrm{O}\) into an ammonia solution of the filtrate with an additional equivalent of 2.2.2- Cryptand. The structure determination results in the composition of \([\mathrm{K}(2.2.2\text{- } \mathrm{crypt})]_3[\mathrm{Si}_9\mathrm{H}]\cdot 8.5\mathrm{NH}_3\) (1). The crystal structure analysis unambiguously shows the formation of a three- fold negatively charged cluster. The exceptional good data quality allows for further structure refinement and the localisation of a hydrogen atom at the cluster from the difference Fourier map and a free refinement of the position of the H atom. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 83, 883, 165]]<|/det|> +145 The asymmetric unit contains a monoprotonated, threefold negatively charged [Si9H]3- 146 cluster, three [K(2.2.2- crypt)]+ counter ions and 8.5 equivalents of co- crystallised ammonia. 147 The co- crystallised ammonia primarily occupies the voids between the [K(2.2.2- crypt)]+ units 148 and the cluster. This results in a particular thermal and mechanical sensitivity of the crystals. + +<|ref|>image<|/ref|><|det|>[[128, 177, 874, 375]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 390, 881, 457]]<|/det|> +
Figure 4: Crystal structure of [Si9H]3- unit in [K(2.2.2-crypt)]4[Si9H]-8.5NH3. left: front view; right: top view. Ellipsoids are shown at \(50\%\) probability level. Selected bond length (A): Si1A-H: 1.55(4); Si1A-Si2B: 2.345(3); Si1A-Si3B: 3.125(4); Si1A-Si4: 2.323(2); Si1A-Si5: 2.4238(12); Si1A-Si6: 2.4338(12); Si2B-Si3B: 2.578(3); Si2B-Si4: 3.751(3); Si2B-Si6: 2.520(2); Si2B-Si7: 2.4637(14); Si3B-Si4: 2.532(2); Si3B-Si7: 2.4555(12); Si3B-Si8: 2.4322(11); Si4-Si5: 2.4912(12); Si4-Si8: 2.4258(10); Si5-Si9: 2.4466(11); Si6-Si9: 2.4312(10); Si7-Si9: 2.4356(13); Si8-Si9: 2.4545(11); Si2B-Si4/Si1A-Si3B: 1.20.
+ +<|ref|>text<|/ref|><|det|>[[113, 470, 883, 912]]<|/det|> +The refinement shows the presence of an orientationally disordered \(C_s\) symmetric [Si9H]3- 156 cluster (for details, see the Supplementary Discussion). For the main orientation \(\alpha\) (Fig. 4) the 157 refinement allows for the localisation of the proton at the Si1A position with a bond length of 158 1.55(4) Å, which is in the range of typical Si- H bond distances. The cluster framework shows 159 the expected involvement of the substituted Si1A position of the open square plane. Thus, the 160 Si1A- Si2B (2.345(3) Å) and Si1A- Si4 (2.323(2) Å) distances are significantly shortened 161 compared to the Si2B- Si3B (2.578(3) Å) and Si3B- Si4 (2.532(2) Å) distances of the open square 162 plane. The ratio of the square diagonals (Si2B- Si4/Si1A- Si3B) of 1.20 clearly shows the 163 deviation from the ideal \(C_{4v}\) symmetry of the parent [Si9]4- ion.68 The structural characteristic 164 of shorter Si- Si bonds at cluster atoms with ligands supports the existence of a H atoms at Si1A 165 and agrees with findings for the solvate [K(DB- 18- crown- 6)][K(2.2.2- crypt)]2[Si9H]·NH3.70 166 The crystallographic data are further supported by mass spectra of acetonitrile (MeCN) 167 solutions of the filtration residue showing protonated Si9 species (Figure 6). Additionally, 1H 168 NMR (Figure 5) of K[K(2.2.2- crypt)]2[Si9H] in DMF- d7 verifies the existence of monoprotonated 169 cluster species. The prominent signal in the high- field region at \(- 1.80\) ppm aligns with the 170 spectral range reported for [Si9H2]2- .77 The characteristic satellite pattern emerges from scalar 171 coupling to all nine cluster atoms. This pattern displays the superposition of all possible 172 isotopologues caused by the low natural abundance of NMR active 29Si (Natural 173 abundancy \(= 4.7\%\) ) in the cluster framework. Consequently, the intense main singlet results 174 from all non- NMR active isotopologues ([28/30Si9H]3- ). While the first set of satellites is due to 175 a doublet splitting of the isotopologues [28/30Si9H]3- , the second is due to a triplet splitting + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 82, 884, 272]]<|/det|> +176 of the isotopologues \([^{28 / 30}\mathrm{Si}_{7}^{29}\mathrm{Si}_{2}\mathrm{H}]^{3 - }\) . The satellite signals of higher isotopologues are not 177 detectable due to the low natural abundance of \(^{29}\mathrm{Si}\) . A full overview of the statistical intensity 178 distribution for the superposition of all isotopologues is given in the supporting information 179 (Supplementary Fig. 6 and Table 6) and in accordance with previous work. \(^{77}\) The exceptionally 180 small coupling constant of \(J(^{1}\mathrm{H},^{29}\mathrm{Si}) = 19.5 \mathrm{Hz}\) and the interaction of the proton with all nine 181 silicon atoms of the cluster framework paints the picture of a highly dynamic system at room 182 temperature. Those migration tendencies have already been described by Korber et al. for 183 \([\mathrm{Si}_{9}\mathrm{H}]^{3 - 70}\) in liquid ammonia at low temperature and by Eichhorn et al. for \([\mathrm{Sn}_{9}\mathrm{R}]^{3 - }\) ( \(\mathrm{R} = \mathrm{H}^{81}\) , 184 \(^{1}\mathrm{Pr}\) , \(\mathrm{SnCy}_{3}^{82}\) ) at room temperature. + +<|ref|>image<|/ref|><|det|>[[120, 275, 870, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 692, 520, 707]]<|/det|> +
Figure 5: \(^{1}\mathrm{H}\) NMR of the dried filtrate. (400 MHz, DMF-\(d_{7}\) , 300 K).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 81, 880, 430]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 444, 880, 474]]<|/det|> +
Figure 6: ESI(-) MS spectrum of the dried filtrate in MeCN. a) \(\{[K(2.2.2\text{-} \text{crypt})][\text{Si9}] + 2\text{H}\}^{-}\) (m/z = 670.37); b) \(\{[K(2.2.2\text{-} \text{crypt})][\text{Si9}] + 2\text{H} + \text{meen}\}^{-}\) (m/z = 711.42).
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 515, 346, 532]]<|/det|> +## Reactivity Study of [Si9H]3-. + +<|ref|>text<|/ref|><|det|>[[113, 535, 883, 704]]<|/det|> +The access to isolated \(\text{Si}_9\) clusters on a multi- gram scale, in the absence of highly reductive \(\text{[Si}_4]^{4- }\) clusters, provides promising and well- defined conditions for follow- up reactions of nina- atomic silicon clusters. Silylation reactions of \(\text{K[K(2.2.2- crypt)]}_2[\text{Si}_9\text{H}]\) in tetrahydrofuran (THF) lead to the isolation and the fist structural characterisation of trisilylated cluster salts [K(2.2.2- crypt)][(R3Si)3Si9] (2a- 2d) in good yields as orange- brown solids. All compounds were characterised by NMR and ESI- MS analyses. Yellow block- shaped single crystals of \(\text{[K(2.2.2- crypt)]}[\text{MeHyp}_3\text{Si}_9]\) - thf were grown from a thf solution at \(- 32^{\circ}\text{C}\) over two weeks. The crystals structure of the molecular anion in 2a is depicted in Fig. 7. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 81, 496, 317]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[110, 332, 881, 361]]<|/det|> +
Figure 7: Crystal structure of the anionic unit in [K(2.2.2-crypt)][\(^{Me}HypsSi_3\)]-thf (2a). Si is depicted as blue ellipsoids at 50% probability level and carbon grey as wires. Hydrogen atoms, co-crystallised [K(2.2.2-crypt)]+ and thf are omitted for clarity.
+ +<|ref|>text<|/ref|><|det|>[[112, 370, 883, 642]]<|/det|> +2a crystallises in the monoclinic space group \(P2_{1} / n\) (14) (a = 15.0913(4) Å, b = 24.7859(6) Å, c = 23.9571(6) Å, \(\alpha = 90^{\circ}\) , \(\beta = 90.959(2)^{\circ}\) , \(\gamma = 90^{\circ}\) , \(V = 8959.9(4) \text{Å}^3\) ) with one trisilylated \([^{Me}HypsSi_3]^{- }\) cluster anion, one [K(2.2.2- crypt)]+ unit and one disordered thf molecule in the asymmetric unit. Analogously to the homologous germanium cluster, the present silicon cluster can also be described as a \(D_{3h}\) symmetric threefold capped trigonal prism. The attachment of a further hypersilyl group to the \(C_{2v}\) symmetric dianion \([^{Me}HypsSi_3]^{2 - 79}\) leads to a closure of the planar square plane Si1- Si4- Si8- Si5 by shortening the Si1- Si8 bond from 3.770(7) Å to 3.2569(14) Å. At the same time, the remaining prism edges (Si2- Si9 and Si3- Si7) are shortened by 0.526 Å and 0.430 Å, respectively. The attachment of the third silyl substituent at the Si\(_9\) cluster induces the same geometric changes that have been described for the homologous \([^{Me}Hyps,Ge_9]^{(4-n)- }\) (Table 1)\(^{40,83}\) and \([^{Me}Hyps,Sn_9]^{(4-n)- }\) clusters (n = 2, 3)\(^{84,85}\). As expected, the cluster framework undergoes a significant contraction from tin and germanium to silicon. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[113, 103, 881, 760]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[115, 84, 534, 98]]<|/det|> +Table 1: Selected bond length (A) in 2a and related cluster species. + +
Bond[MeHyp2Sig]2-79[MeHyp3Sig]- (2a)[MeHyp2Ges]2-40[MeHyp3Ges]-83 a)
exo MeHyp-E42.357(5)2.3326(12)2.4071(9)2.3646(14)
exo MeHyp-E52.339(5)2.3264(12)2.4098(9)2.3741(15)
exo MeHyp-E6-2.3258(12)-2.3688(14)
E1-E42.427(5)2.4115(12)2.5528(5)2.5321(8)
E1-E52.398(5)2.4075(13)2.5469(5)2.5213(8)
E4-E82.396(5)2.4050(12)2.5415(5)2.5366(8)
E5-E82.395(4)2.4071(12)2.5486(5)2.5314(8)
E1-E22.488(5)2.4786(13)2.6439(5)2.7081(7)
E1-E32.485(6)2.4981(13)2.6399(5)2.6526(8)
E2-E32.585(6)2.5010(13)2.7046(5)2.6921(8)
E1-E83.770(7)3.2569(14)3.9867(9)3.4966(9)
E2-E92.738(5)3.2639(14)2.8837(5)3.2196(9)
E3-E72.772(6)3.2020(14)2.6005(5)3.4904(9)
E7-E92.569(6)2.5098(12)2.7076(5)2.6640(8)
E2-E62.475(7)2.4100(13)2.5824(5)2.5234(8)
E3-E62.468(6)2.4130(13)2.5872(5)2.5222(7)
E6-E72.470(6)2.4057(12)2.5751(8)2.5238(8)
E6-E92.427(6)2.3997(12)2.5819(5)2.5343(8)
+ +<|ref|>table_footnote<|/ref|><|det|>[[115, 760, 358, 774]]<|/det|> +a) Two independent clusters in unit cell. + +<|ref|>text<|/ref|><|det|>[[115, 799, 883, 901]]<|/det|> +As expected from the crystal structure and analogously to [MeHyp3Ges] - 83, 2a behaves also \(D_{3h}\) symmetric on NMR time scale in thf- \(d_8\) at room temperature. Hence, the three hypersily groups collapse to two resonances at \(- 130.03\) ppm and \(- 8.71\) ppm. These signals can be attributed to the exo- bonded silicon atoms (STMS3) and the TMS groups, respectively. Further, in the high field at \(- 171.3\) ppm, the three capping positions show one signal, while + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 123]]<|/det|> +the six equivalent prism positions exhibit a strongly shielded signal at \(- 352.8 \text{ppm}\) . The NMR data are consistent with the data described in our previous studies. \(^{79}\) + +<|ref|>table<|/ref|><|det|>[[115, 166, 880, 355]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[115, 135, 608, 163]]<|/det|> +Table 2: Selected spectroscopic data of 2a-2d. \(^{29}\mathrm{Si}\) NMR shifts of \(\mathrm{Si}_{\mathrm{cap}}\) (red) and \(\mathrm{Si}_{\mathrm{prism}}\) (blue) and homonuclear \(J\) coupling (298 K, 79.5 MHz, thf-\(d_8\) ). + +
R3Siδ(SiCap) / ppmδ(SiPrism) / ppm1J(29SiCap, 29SiPrism) / Hz[K(2.2.2-crypt)]+
MeHyp (2a)-171.3-352.840.1Si3Si
EtHyp (2b)-171.3-352.842.7
MeHypMe2Si (2c)-146.3-347.424.4
tBu2FSi (2d) a)-177.7-357.3723.2
+ +<|ref|>table_footnote<|/ref|><|det|>[[115, 359, 396, 374]]<|/det|> +a) \(^{1}J(^{19}\mathrm{F},^{29}\mathrm{Si}) = 340.9 \mathrm{Hz};^{2}J(^{19}\mathrm{F},^{29}\mathrm{Si}) = 19.4 \mathrm{Hz}.\) + +<|ref|>text<|/ref|><|det|>[[114, 390, 883, 597]]<|/det|> +The remaining derivatives 2b- d exhibit similar behaviour, with the cap and prism signals falling within the characteristic range of \(- 160 \text{ppm}\) and \(- 350 \text{ppm}\) . The high quality of the data enables the determination of the \(^{1}J(^{29}\mathrm{Si},^{29}\mathrm{Si})\) homonuclear couplings between the cap and prism atoms (Table 2). The coupling constants of the sterically demanding hypersilyl groups (40.1 Hz for 2a and 42.7 Hz 2b) are significantly higher than those of the sterically less demanding silyl groups in 2c (24.4 Hz) and 2d (23.2 Hz). These couplings differ from localised Si- Si bonds as in cyclic oligosilanes ( \(^{1}J(^{29}\mathrm{Si},^{29}\mathrm{Si}) \approx 50 - 70 \text{Hz})^{86}\) and may indicate possible dynamic processes in the cluster framework, which have not yet been described in homologous silylated clusters. The access to NMR active cluster frameworks could reveal processes that have remained hidden to us so far. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 84, 210, 99]]<|/det|> +## Conclusion + +<|ref|>text<|/ref|><|det|>[[112, 105, 884, 397]]<|/det|> +ConclusionThe presence of strongly reductive \([Si_4]^{4- }\) clusters in the solid- state phase \(\mathsf{K}_{12}\mathsf{Si}_{17}\) limits the directed conversion of nine- atomic \([Si_9]^{4- }\) silicon clusters. However, in this work we have shown that the separation of both cluster species is easily possible on a multi- gram scale in liquid ammonia and provides a valuable synthetic access to \([Si_9H]^{3- }\) ions for the first time. \([Si_9H]^{3- }\) shows a pronounced tautomerisation tendency, in which the proton rapidly migrates over the entire nine- atomic cluster framework. In addition, those monoprotonated silicon clusters in the form of the salt \(\mathsf{K}[\mathsf{K}(2.2.2\text{- } \mathsf{crypt})_2[Si_9H]\) represent a synthetic equivalent to \([Si_9]^{4- }\) ions that are still not accessible in an isolated form via a solid- state approach. Thus, the trisilylated cluster salts \([\mathsf{K}(2.2.2\text{- } \mathsf{crypt})][(\mathsf{R}_3\mathsf{Si})_3\mathsf{Si}_9]\) (2a- d) are obtained in good yields and high purity by direct silylation of \(\mathsf{K}[\mathsf{K}(2.2.2\text{- } \mathsf{crypt})_2[Si_9H]\) with the corresponding chlorosilanes. The spectroscopic behaviour as well as the crystallographic characterisation of 2a proofs the strong similarities between silicon and germanium based Zintl clusters. With the present work, we were able to close a significant gap in the chemistry of group 14 Zintl ions. Studies on the further reactivity of the obtained trisilylated monoanions are underway. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 70, 884, 905]]<|/det|> +262 References263 1 Sekiguchi, A. & SAKURA, H. Cage and Cluster Compounds of. Adv. Organomet. Chem., 264 1, (1995).265 2 Lee, V. Y. Organosilicon Compounds: Theory and experiment (synthesis). (Academic 266 Press, 2017).267 3 Heider, Y. & Scheschkewitz, D. Molecular Silicon Clusters. Chem. Rev. 121, 9674- 9718, 268 (2021).269 4 Heider, Y. & Scheschkewitz, D. Stable unsaturated silicon clusters (siliconoids). Dalton 270 Trans. 47, 7104- 7112, (2018).271 5 Schnepf, A. Metalloid group 14 cluster compounds: An introduction and perspectives 272 to this novel group of cluster compounds. Chem. Soc. Rev. 36, 745- 758, (2007).273 6 Scharfe, S., Kraus, F., Stegmaier, S., Schier, A. & Fässler, T. F. Zintl Ions, Cage 274 Compounds, and Intermetalloid Clusters of Group 14 and Group 15 Elements. Angew. 275 Chem. Int. Ed. 50, 3630- 3670, (2011).276 7 Liu, C. & Sun, Z.- M. Recent advances in structural chemistry of Group 14 Zintl ions. 277 Coord. Chem. Rev. 382, 32- 56, (2019).278 8 Wilson, R. J., Weinert, B. & Dehnen, S. Recent developments in Zintl cluster chemistry. 279 Dalton Trans. 47, 14861- 14869, (2018).280 9 West, R. & Indriksons, A. Cyclic polysilanes. V. Tetradecylmethylbicyclo [2.2. 2] 281 octasilane, a cage polysilane. J. Am. Chem. Soc. 92, 6704- 6705, (1970).282 10 Sekiguchi, A., Yatabe, T., Kabuto, C. & Sakurai, H. Chemistry of organosilicon 283 compounds. 303. The missing hexasilaprismane: synthesis, x- ray analysis and 284 photochemical reactions. J. Am. Chem. Soc. 115, 5853- 5854, (1993).285 11 Abersfelder, K. et al. Contraction and Expansion of the Silicon Scaffold of Stable Si6R6 286 Isomers. J. Am. Chem. Soc. 134, 16008- 16016, (2012).287 12 Li, Y., Li, J., Zhang, J., Song, H. & Cui, C. Isolation of R6Si6 Dianion: A Bridged Tricyclic 288 Isomer of Dianionic Hexasilabenzen. J. Am. Chem. Soc. 140, 1219- 1222, (2018).289 13 Matsumoto, H. et al. The first octasilacubane system: synthesis of octakis- (t- 290 butyldimethylsilyl)pentacyclo[4.2.0.02,5.03,8.04,7]octasilane. Chem. Commun., 1083- 1084, (1988).291 14 Matsumoto, H., Higuchi, K., Kyushin, S. & Goto, M. Octakis(1,1,2- 293 trimethylpropyl)octasilacubane: Synthesis, Molecular Structure, and Unusual 294 Properties. Angew. Chem. Int. Ed. 31, 1354- 1356, (1992).295 15 Sekiguchi, A., Yatabe, T., Kamatani, H., Kabuto, C. & Sakurai, H. Chemistry of 296 organosilicon compounds. 293. Preparation, characterization, and crystal structures of 297 octasilacubanes and octagermacubanes. J. Am. Chem. Soc. 114, 6260- 6262, (1992).298 16 Furukawa, K., Fujino, M. & Matsumoto, N. Cubic silicon cluster. Appl. Phys. Lett. 60, 299 2744- 2745, (1992).300 17 Wiberg, N., Finger, C. M. M. & Polborn, K. Tetrakis(tri- tert- butylsilyl)- tetrahedro- 301 tetrasilane (tBu3Si)4Si4: The First Molecular Silicon Compound with a Si4 Tetrahedron. 302 Angew. Chem. Int. Ed. 32, 1054- 1056, (1993).303 18 Ichinohe, M., Toyoshima, M., Kinjo, R. & Sekiguchi, A. Tetrasilatetrahedranide: A 304 Silicon Cage Anion. J. Am. Chem. Soc. 125, 13328- 13329, (2003).305 19 West, R. & Indriksons, A. Cyclic polysilanes. VI. Bicyclic and cage 306 permethylcyclopolysilanes. J. Am. Chem. Soc. 94, 6110- 6115, (1972).307 20 Fischer, J., Baumgartner, J. & Marschner, C. Synthesis and Structure of Sila- 308 Adamantane. Science 310, 825- 825, (2005). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 78, 884, 920]]<|/det|> +309 21 Tillmann, J. et al. One-Step Synthesis of a [20]Silafullerane with an Endohedral Chloride 310 lon. Angew. Chem. Int. Ed. 54, 5429- 5433, (2015). 311 22 Bamberg, M. et al. [Cl@Si2OH20]–: Parent Siladodecahedrane with Endohedral 312 Chloride lon. J. Am. Chem. Soc. 143, 10865- 10871, (2021). 313 23 Bamberg, M. et al. Regioselective Derivatization of Silylated [20]Silafulleranes. J. Am. 314 Chem. Soc. 145, 11440- 11448, (2023). 315 24 Nied, D., Köppe, R., Klopper, W., Schnöckel, H. & Breher, F. Synthesis of a 316 Pentasilapropellane. Exploring the Nature of a Stretched Silicon- Silicon Bond in a 317 Nonclassical Molecule. J. Am. Chem. Soc. 132, 10264- 10265, (2010). 318 25 Abersfelder, K., White, A. J. P., Berger, R. J. F., Rzepa, H. S. & Scheschkewitz, D. A Stable 319 Derivative of the Global Minimum on the Si6H6 Potential Energy Surface. Angew. 320 Chem. Int. Ed. 50, 7936- 7939, (2011). 321 26 Leszczyńska, K. I. et al. Atomically Precise Expansion of Unsaturated Silicon Clusters. 322 Angew. Chem. Int. Ed. 58, 5124- 5128, (2019). 323 27 Willmes, P. et al. Isolation and Versatile Derivatization of an Unsaturated Anionic 324 Silicon Cluster (Siliconoid). Angew. Chem. Int. Ed. 55, 2907- 2910, (2016). 325 28 Heider, Y., Willmes, P., Huch, V., Zimmer, M. & Scheschkewitz, D. Boron and 326 Phosphorus Containing Heterosiliconoids: Stable p- and n- Doped Unsaturated Silicon 327 Clusters. J. Am. Chem. Soc. 141, 19498- 19504, (2019). 328 29 Poitiers, N. E., Giarrana, L., Huch, V., Zimmer, M. & Scheschkewitz, D. Exohedral 329 functionalization vs. core expansion of siliconoids with Group 9 metals: catalytic 330 activity in alkene isomerization. Chem. Sci. 11, 7782- 7788, (2020). 331 30 Helmer, J., Hepp, A., Berger, R. J. F. & Lips, F. Synthesis and functionalization of the six- 332 vertex anionic amido- substituted silicon cluster [Si6{N(SiMe3)Ph}5]–. Dalton Trans. 52, 1339- 14945, (2023). 333 31 Keuter, J., Hepp, A., Daniliuc, C. G., Feldt, M. & Lips, F. Cycloadditions with a Stable 334 Charge- Separated Cyclobutadiene- Type Amido- Substituted Silicon Ring Compound. 335 Angew. Chem. Int. Ed. 60, 21761- 21766, (2021). 336 32 Keuter, J. et al. A highly unsaturated six- vertex amido- substituted silicon cluster. Chem. 337 Sci. 11, 5895- 5901, (2020). 338 33 Von Schnering, H. G. et al. Binary alkali metal compounds with the zintl anions [Ge9]4- 340 and [Sn9]4–. Zeitschrift für anorganische und allgemeine Chemie 623, 1037- 1039, 341 (1997). 342 34 Benda, C. B., Wang, J.- Q., Wahl, B. & Fässler, T. F. Syntheses and 1H NMR Spectra of 343 Substituted Zintl Ions [Ge9Rn](4–n)–: Crystal Structures of [Ge9R]3– (R = 2,4,6- 344 Me3C6H2, CHCH2) and Indication of Tris- Vinylated Clusters. Eur. J. Inorg. Chem. 2011, 345 4262- 4269, (2011). 346 35 Bentlohner, M. M., Klein, W., Fard, Z. H., Jantke, L.- A. & Fässler, T. F. Linking 347 Deltahedral Zintl Clusters with Conjugated Organic Building Blocks: Synthesis and 348 Characterization of the Zintl Triad [R- Ge9- CH@CH@CH@CH- Ge9- R]4–. Angew. Chem. Int. 349 Ed. 54, 3748- 3753, (2015). 350 36 Hull, M. W. & Sevov, S. C. Functionalization of Nine- Atom Deltahedral Zintl Ions with 351 Organic Substituents: Detailed Studies of the Reactions. J. Am. Chem. Soc. 131, 9026- 352 9037, (2009). 353 37 Frischhut, S., Bentlohner, M. M., Klein, W. & Fässler, T. F. Synthesis of Zintl Triads 354 Comprising Extended Conjugated π-Electronic Systems: [RGe9- CH=CH- CH=CH- 355 Ge9R]4– (R = - CH=CH2, - C(CH3)=CH- CH=N(CH2)2NH2). Inorg. Chem. 56, 10691- 356 10698, (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 78, 885, 904]]<|/det|> +357 38 Benda, C. B., He, H., Klein, W., Somer, M. & Fässler, T. F. Bisvinylated [R- Ge9- Ge9- 358 R]4- Cluster Dimers. Z. Anorg. Allg. Chem. 641, 1080- 1086, (2015). 359 39 Hull, M. W. & Sevov, S. C. Addition of Alkenes to Deltahedral Zintl Clusters by Reaction 360 with Alkynes: Synthesis and Structure of [FcCHCHCHGe9CHCHCHFc]2-, an Organo- 361 Zintl- Organometallic Anion. Angew. Chem. Int. Ed. 46, 6695- 6698, (2007). 362 40 Kysliak, O. & Schnepf, A. {Ge9[Si(SiMe3)3]2}2- : a starting point for mixed substituted 363 metalloid germanium clusters. Dalton Trans. 45, 2404- 2408, (2016). 364 41 Mayer, K., Schiegerl, L. J., Kratky, T., Günther, S. & Fässler, T. F. Targeted attachment 365 of functional groups at Ge9 clusters via silylation reactions. Chem. Commun. 53, 11798- 366 11801, (2017). 367 42 Schiegerl, L. J., Geitner, F. S., Fischer, C., Klein, W. & Fässler, T. F. Functionalization of 368 [Ge9] with Small Silanes:{Ge9(SiR3)3}- (R = iBu, iPr, Et) and the Structures of 369 (CuNHCdipp)[Ge9(Si(iBu)3)3], (K- 18c6)Au[Ge9(Si(iBu)3)3]2, and (K- 370 18c6)2[Ge9(Si(iBu)3)2]. Z. Anorg. Allg. Chem. 642, 1419- 1426, (2016). 371 43 Kysliak, O., Kunz, T. & Schnepf, A. Metalloid Ge9R3- Clusters with Various Silyl 372 Substituents: From Shielded to Open Cluster Cores. Eur. J. Inorg. Chem. 2017, 805- 810, 373 (2017). 374 44 Kysliak, O., Schrenk, C. & Schnepf, A. {Ge9[Si(SiMe3)2(SiPh3)]3}-: Ligand Modification 375 in Metalloid Germanium Cluster Chemistry. Inorg. Chem. 54, 7083- 7088, (2015). 376 45 Kysliak, O., Schrenk, C. & Schnepf, A. {[Si(SiMe3)3]2Ge9- SiMe2- (C6H4)- SiMe2- 377 Ge9[Si(SiMe3)3]2K}-: The Connection of Metalloid Clusters via an Organic Linker. Inorg. Chem. 56, 9693- 9697, (2017). 378 Frischhut, S., Machado de Carvalho, J. G., Karttunen, A. J. & Fässler, T. F. On the Affinity 379 between Fullerenes and Deltahedral Zintl Ions: A UV/Vis Spectroscopic Investigation. 380 Z. Anorg. Allg. Chem. 644, 1337- 1343, (2018). 381 Li, F. & Sevov, S. C. Synthesis, Structures, and Solution Dynamics of Tetrasubstituted 382 Nine- Atom Germanium Deltahedral Clusters. J. Am. Chem. Soc. 136, 12056- 12063, 383 (2014). 384 Frischhut, S. & Fässler, T. F. Synthesis of low- oxidation- state germanium clusters 385 comprising a functional anchor group - synthesis and characterization of [(Ge0)5(Ge- 386 R)3(Ge- (CH2)n- CH \(\checkmark\) CH2)] with R = Si(SiMe3)3. Dalton Trans. 47, 3223- 3226, (2018). 387 Frischhut, S., Klein, W., Drees, M. & Fässler, T. F. Acylation of Homoatomic Ge9 Cages 388 and Subsequent Decarbonylation. Chem. Eur. J. 24, 9009- 9014, (2018). 389 Frischhut, S. et al. Capping nido- Nonagermanide Clusters with M- PPh3 and Dynamics 390 in Solution: Synthesis and Structure of closo- [(Me3Si)3Si]3Et[Ge9M](PPh3) (M = Ni, Pt). 391 Organometallics 37, 4560- 4567, (2018). 392 Henke, F., Schenk, C. & Schnepf, A. {Si(SiMe3)3]3Ge9M(CO)3- (M = Cr, Mo, W): 393 Coordination Chemistry with metalloid Clusters. Dalton Trans. 40, 6704- 6710, (2011). 394 Schenk, C. & Schnepf, A. {Ge9R3Cr(CO)5}- and {Ge9R3Cr(CO)3}-: a metalloid cluster 395 (Ge9R3- ) as a flexible ligand in coordination chemistry [R = Si(SiMe3)3]. Chem. 396 Commun., 3208- 3210, (2009). 397 Li, F., Muñoz- Castro, A. & Sevov, S. C. [(Me3Si)Si]3EtGe9Pd(PPh3), a 398 Pentafunctionalized Deltahedral Zintl Cluster: Synthesis, Structure, and Solution 399 Dynamics. Angew. Chem. Int. Ed. 55, 8630- 8633, (2016). 400 Willeit, N. S., Klein, W., Coburger, P., Fritz- Langhals, E. & Fässler, T. F. Functionalised 401 [Ge9Ni] Clusters as Homogeneous Single- Site Catalysts for Olefin Isomerisation 402 Reactions. ChemCatChem n/a, e202301200. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 80, 884, 920]]<|/det|> +404 55 Schiegerl, L. J. et al. Silylated Ge(9) Clusters as New Ligands for Cyclic (Alkyl)amino and 405 Mesoionic Carbene Copper Complexes. Inorg. Chem. 58, 3256- 3264, (2019). 406 56 Gienger, C., Schynowski, L., Schaefer, J., Schrenk, C. & Schnepf, A. New Intermetalloid 407 Ge9- clusters with Copper and Gold: Filling Vacancies in the Cluster Chemistry of 408 [Ge9(Hyp)3]– (Hyp=Si(SiMe3)3). Eur. J. Inorg. Chem. 26, e202200738, (2023). 409 57 Mayer, K., Schiegerl, L. J. & Fässler, T. F. On the Reactivity of Silylated Ge9 Clusters: 410 Synthesis and Characterization of [ZnCp*(Ge9(Si(SiMe3)3)3)], 411 [CuPiPr3(Ge9(Si(SiMe3)3)3)], and [(CuPiPr3)4(Ge9(SiPh3)2)2]. Chem. Eur. J. 22, 18794- 412 18800, (2016). 413 58 Ugrinov, A. & Sevov, S. C. [Ge9Ge9Ge9]6- : A Linear Trimer of 27 Germanium Atoms. J. 414 Am. Chem. Soc. 124, 10990- 10991, (2002). 415 59 Nienhaus, A., Hoffmann, S. D. & Fässler, T. F. First Synthesis of Group- 14 Polyanions by 416 Extraction of a Binary Alloy with dmf and a Novel Conformation of the (Ge9- Ge9)6- 417 Dimer: Crystal Structures of [K6(Ge9- Ge9)](dmf)12, [Rb6(Ge9- Ge9)](dmf)12 and 418 [K2.5Cs3.5(Ge9- Ge9)](dmf)12. Z. Anorg. Allg. Chem. 632, 1752- 1758, (2006). 419 60 Hauptmann, R. & Fässler, T. F. Low Dimensional Arrangements of the Zintl Ion [Ge9- 420 Ge9]6— and Chemical Bonding in [Ge6]2—, [Ge9- Ge9]6—, and {[Ge9]}]2—. Z. Anorg. 421 Allg. Chem. 629, 2266- 2273, (2003). 422 61 Yong, L., Hoffmann, S. D. & Fässler, T. F. The Controlled Oxidative Coupling of Ge94- 423 Zintl Anions to a Linear Trimer [Ge9=Ge9=Ge9]6- . Z. Anorg. Allg. Chem. 631, 1149- 424 1153, (2005). 425 62 Xu, L. & Sevov, S. C. Oxidative coupling of deltahedral [Ge9] 4- Zintl ions. J. Am. Chem. 426 Soc. 121, 9245- 9246, (1999). 427 63 Scharfe, S. & Fässler, Thomas F. Synthesis of (Ge9- Ge9)6- Dimeric Zintl Ions in Liquid 428 Ammonia Solutions of K4Ge9: Low- Dimensional Coordination Networks in the Crystal 429 Structure of the Ammoniates Kn[K([2.2.2]crypt)]6- n[Ge9- Ge9](NH3)m (n = 2, 3, and 430 4). Z. Anorg. Allg. Chem. 637, 901- 906, (2011). 431 64 Ugrinov, A. & Sevov, S. C. Synthesis of a chain of nine- atom germanium clusters 432 accompanied with dimerization of the sequestering agent. C. R. Chim. 8, 1878- 1882, 433 (2005). 434 65 Hoch, C., Wendorff, M. & Röhr, C. Synthesis and crystal structure of the tetrelides 435 A12M17 (A=Na, K, Rb, Cs; M=Si, Ge, Sn) and A4Pb9 (A=K, Rb). J. Alloys Compd. 361, 436 206- 221, (2003). 437 66 Goicoechea, J. M. & Sevov, S. C. Ligand- Free Deltahedral Clusters of Silicon in Solution: 438 Synthesis, Structure, and Electrochemistry of Si92. Inorganic Chemistry 44, 2654- 2658, 439 (2005). 440 67 Goicoechea, J. M. & Sevov, S. C. Naked Deltahedral Silicon Clusters in Solution: 441 Synthesis and Characterization of Si93- and Si52. J. Am. Chem. Soc. 126, 6860- 6861, 442 (2004). 443 68 Benda, C. B., Henneberger, T., Klein, W. & Fässler, T. F. [Si4]4- and [Si9]4- Clusters 444 Crystallized from Liquid Ammonia Solution – Synthesis and Characterization of 445 K8[Si4][Si9]·(NH3)14.6. Zeitschrift für anorganische und allgemeine Chemie 643, 146- 446 148, (2017). 447 69 Hastreiter, F. et al. Elusive Zintl Ions [μ-HSi4]3- and [Si5]2- in Liquid Ammonia: 448 Protonation States, Sites, and Bonding Situation Evaluated by NMR and Theory. 449 Angew. Chem. Int. Ed. 58, 3133- 3137, (2019). 450 70 Lorenz, C. et al. The Structure of [HSi9]3- in the Solid State and Its Unexpected Highly 451 Dynamic Behavior in Solution. Angew. Chem. Int. Ed. 57, 12956- 12960, (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 80, 886, 821]]<|/det|> +452 71 Goicoechea, J. M. & Sevov, S. C. Organozinc Derivatives of Deltahedral Zintl Ions: Synthesis and Characterization of closo- [E9Zn(C6H5)]3- (E = Si, Ge, Sn, Pb). Organometallics 25, 4530- 4536, (2006). 453 72 Joseph, S. et al. Chemistry with Bare Silicon Clusters in Solution: A Transition- Metal Complex of a Polysilicide Anion. Angew. Chem. Int. Ed. 48, 8770- 8772, (2009). 454 73 Streitferdt, V. et al. NMR- Spectroscopic Detection of an Elusive Protonated and Coinage Metalated Silicide [NHCDippCu(η4- Si9)H]2- in Solution. Eur. J. Chem. 2021, 458 3648- 3690, (2021). 455 74 Tiefenthaler, S. M. et al. On the Reactivity of NHctBuAuCl towards Rb6Cs6Si17: The First Gold- Silicon Cluster [(NHctBuAu)6(η2- Si4)]Cl2- 7NH3 and an Imide Capped Gold Triangle (NHctBuAu)3NHCl. Z. Anorg. Allg. Chem. 646, 1595- 1602, (2020). 456 75 Waibel, M., Kraus, F., Scharfe, S., Wahl, B. & Fässler, T. F. [(MesCu)2(η3- Si4)]4- : A Mesitylcopper- Stabilized Tetrasilicide Tetraanion. Angew. Chem. Int. Ed. 49, 6611- 6615, (2010). 457 76 Geitner, F. S. & Fässler, T. F. Low oxidation state silicon clusters - synthesis and structure of [NHCDippCu(η4- Si9)]3- . Chem. Commun. 53, 12974- 12977, (2017). 458 77 Schiegerl, L. J. et al. Charged Si9 Clusters in Neat Solids and the Detection of [H2Si9]2- in Solution: A Combined NMR, Raman, Mass Spectrometric, and Quantum Chemical Investigation. Angew. Chem. Int. Ed. 57, 12950- 12955, (2018). 459 78 Schiegerl, L. J., Karttunen, A. J., Klein, W. & Fässler, T. F. Anionic Siliconoids from Zintl Phases: R3Si9- with Six and R2Si92- with Seven Unsubstituted Exposed Silicon Cluster Atoms (R=Si(tBu)2H). Chem. Eur. J. 24, 19171- 19174, (2018). 460 79 Schiegerl, L. J., Karttunen, A. J., Klein, W. & Fässler, T. F. Silicon clusters with six and seven unsubstituted vertices via a two- step reaction from elemental silicon. Chem. Sci. 10, 9130- 9139, (2019). 461 80 von Schnering, H. G. et al. Das Clusteranion Si94- . Angew. Chem. 110, 2507- 2509, (1998). 462 81 Kocak, F. S. et al. Surprising Acid/Base and Ion- Sequestration Chemistry of Sn94- : HSn93- , Ni@HSn93- , and the Sn93- Ion Revisited. J. Am. Chem. Soc. 134, 9733- 9740, (2012). 463 82 Kocak, F. S., Zavalij, P. Y., Lam, Y.- F. & Eichhorn, B. W. Substituent- dependent exchange mechanisms in highly fluxional RSn93- anions. Chem. Commun., 4197- 4199, (2009). 464 83 Li, F. & Sevov, S. C. Rational Synthesis of [Ge9[Si(SiMe3)3]3]- from Its Parent Zintl Ion Ge94- . Inorg. Chem. 51, 2706- 2708, (2012). 465 84 Schrenk, C., Winter, F., Pöttgen, R. & Schnepf, A. {Sn9[Si(SiMe3)3]2}- : A Metalloid Tin Cluster Compound With a Sn9 Core of Oxidation State Zero. Inorg. Chem. 51, 8583- 8588, (2012). 466 85 Schrenk, C., Neumaier, M. & Schnepf, A. {Sn9[Si(SiMe3)3]3}- and {Sn8Si[Si(SiMe3)3]3}- : Variations of the E9 Cage of Metalloid Group 14 Clusters. Inorganic Chemistry 51, 3989- 3995, (2012). 467 86 Jenkner, P. K., Spielberger, A., Eibl, M. & Hengge, E. 29Si NMR Spektroskopie von Cyclopentasilanen. Spectrochim. Acta A 49, 161- 165, (1993). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 84, 276, 100]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[115, 105, 883, 166]]<|/det|> +The authors are thankful for the financial support by WACKER Chemie AG. The authors thank B.Sc. Vivienne Wolde and B.Sc. Thanh N. Trân for their assistance in the synthesis of 2.2.2- Cryptand. + +<|ref|>sub_title<|/ref|><|det|>[[117, 191, 295, 206]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 211, 883, 312]]<|/det|> +K.M.F. conceived and performed the syntheses of cluster compounds and collected the single- crystal X- ray data of 1 and 2a, solved and refined the structure of 1, performed the electrospray- ionisation mass spectrometry and prepared samples for further analyses. V. H. solved and refined the structure of 2a. T.F.F. supervised the work. K.M.F wrote the paper. K.M.F and N.S.W conceived and performed the synthesis of 2.2.2- Cryptand. + +<|ref|>sub_title<|/ref|><|det|>[[117, 337, 288, 353]]<|/det|> +## Competing Interests + +<|ref|>text<|/ref|><|det|>[[117, 358, 480, 374]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 480, 202]]<|/det|> +- NCOMMS2414778SupportingInformation.pdf- NCOMMS24147781.cif- NCOMMS24147782a.cif + +<--- Page Split ---> diff --git a/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/images_list.json b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..5399563b489effb3196c70b335dd66ba133d5bc5 --- /dev/null +++ b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/images_list.json @@ -0,0 +1,107 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: A topographic recurrent network model encodes temporal information of video frames via internal wave activity. (a) Schematic of the network model. Neurons (circles) are arranged on a two-dimensional grid and are", + "footnote": [], + "bbox": [ + [ + 160, + 400, + 828, + 855 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: The network can forecast a simple video input many frames into the future. (a) As in the classification example (Figure 1), a video frame projects into the network in a spatially local manner and a recurrent network interaction occurs, generating internal wave activity on top of the projection. The network outputs an image from its network state via a matrix of trainable weights. Training entails one-shot linear regression between a set of network states and the corresponding desired output frames (the one-step-ahead next frames). Shown: a schematic representation of the one-shot linear regression for one time step. (b) Once training of the readout weights is complete, closed-loop forecasting begins. To properly test how well the network model learned the underlying spatiotemporal process from the training data, it is deprived of ground-truth data of any kind during this step. Instead,", + "footnote": [], + "bbox": [ + [ + 133, + 220, + 861, + 765 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Moving bump forecast performance depends on specific properties of the recurrent connections. (a) Structural similarity between a forecast frame and the ground truth as a function of closed-loop forecast video", + "footnote": [], + "bbox": [ + [ + 114, + 694, + 888, + 869 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: The recurrent network performs next-frame forecasting of a natural video input. (a) Training follows as in the moving bump example (Fig. 2a). (b) Next-frame closed-loop forecasting follows as in the moving bump example (Fig. 2b). (c) Video frames of the data: a person walking. (d) Corresponding closed-loop forecasts generated by the network model in the case of optimal recurrence. (e) Corresponding network states for the optimal-recurrence case (panel d). Cosine of phase shown. (f) Same as d, but in the absence of recurrence.", + "footnote": [], + "bbox": [ + [ + 143, + 90, + 857, + 732 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Natural movie forecast performance depends on specific properties of the recurrent connections. (a) Several examples of closed-loop forecast performance. Structural similarity between a forecast frame and the ground truth as a function of video frame during closed-loop forecasting. Each curve corresponds to a different ratio of recurrent strength to input strength. Curves have been smoothed by a moving-average filter (filter width of 30 time steps). Shaded error is the absolute difference between filtered and unfiltered. (b) Total structural similarity, in which a single SSIM is computed for the whole movie, as a function of the recurrence-to-input ratio. In the parameter space, each point differs only in recurrent strength. Smoothing and error shading is the same as in a.", + "footnote": [], + "bbox": [ + [ + 152, + 560, + 845, + 780 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: Randomly shuffling recurrent connections eliminates nTWs and ability to forecast. (a) Left: the topographic network model used throughout this study, featuring feedforward projections of the image input (red lines) and local distance-dependent horizontal connectivity (blue lines). There are also synaptic time delays proportional to a neuron pair's separation distance within the horizontal recurrent circuitry. Right: by randomizing the horizontal connection weights and time delays assigned to the network neurons, the topography in the network is", + "footnote": [], + "bbox": [ + [ + 191, + 100, + 808, + 816 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7: The network is capable of forecasting multiple movies without being retrained. (a) The recurrent network model was adapted to contain a higher-level competitive-learning process. Left: Readout matrices were learned separately for separate examples. Right: Storing the learned readout matrices in a higher-level matrix, the", + "footnote": [], + "bbox": [ + [ + 137, + 440, + 857, + 836 + ] + ], + "page_idx": 12 + } +] \ No newline at end of file diff --git a/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545.mmd b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545.mmd new file mode 100644 index 0000000000000000000000000000000000000000..898ea440b3c92d09bb2edb88bc87d71163a4c6a3 --- /dev/null +++ b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545.mmd @@ -0,0 +1,397 @@ + +# What computations can be done with traveling waves in visual cortex? + +Gabriel Benigno Western University + +Roberto Budzinski Western University + +Zachary Davis Salk Institute for Biological Studies https://orcid.org/0000- 0003- 4440- 9011 + +John Reynolds The Salk Institute for Biological Studies https://orcid.org/0000- 0001- 6988- 4607 + +Lyle Muller ( ☑ Imuller2@uwo.ca ) Western University https://orcid.org/0000- 0001- 5165- 9890 + +Article + +Keywords: + +Posted Date: August 29th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1903144/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# What computations can be done with traveling waves in visual cortex? + +Gabriel B. Benigno \(^{1,2,3}\) , Roberto C. Budzinski \(^{1,2,3}\) , Zachary Davis \(^{4}\) , John Reynolds \(^{4}\) , Lyle Muller \(^{1,2,3,*}\) + +Affiliations: \(^{1}\) Department of Mathematics, Western University, London, ON, Canada, \(^{2}\) Brain and Mind Institute, Western University, London, ON, Canada, \(^{3}\) Western Academy for Advanced Research, Western University, London, ON, Canada, \(^{4}\) The Salk Institute for Biological Studies, La Jolla, CA, USA, \(^{*}\) Corresponding author: Imuller2@uwo.ca. + +Recent analyses have found waves of neural activity traveling across entire visual cortical areas in awake animals. These traveling waves modulate excitability of local networks and perceptual sensitivity. The general computational role for these spatiotemporal patterns in the visual system, however, remains unclear. Here, we hypothesize that traveling waves endow the brain with the capacity to predict complex and naturalistic visual inputs. We present a new network model whose connections can be rapidly and efficiently trained to predict natural movies. After training, a few input frames from a movie trigger complex wave patterns that drive accurate predictions many frames into the future, solely from the network's connections. When the recurrent connections that drive waves are randomly shuffled, both traveling waves and the ability to predict are eliminated. These results show traveling waves could play an essential computational role in the visual system by embedding continuous spatiotemporal structures over spatial maps. + +## Introduction + +Five percent of synapses received by a neuron in visual cortex arrive through the feedforward (FF) pathway that conveys sensory input from the eyes \(^{1}\) . While these FF synapses are strong \(^{2}\) , "horizontal" recurrent connections coming from within the cortical region make up about 80% of total synaptic inputs, with 95% of these connections arising from a very local patch (2 mm) around the cell \(^{1}\) . The anatomy of the visual system thus indicates that cortical neurons interact with other neurons across the retinotopically organized maps \(^{3}\) that assign nearby points in visual space to nearby points in a cortical region, via these horizontal connections. Models of the visual system predominantly focus only on FF \(^{4,5}\) and feedback (FB) \(^{6}\) connections. One result of this focus is that neurons in visual cortex are often conceived to be non- interacting "feature + +<--- Page Split ---> + +detectors", with fixed selectivity to features in visual input (driven by FF connections) that can be modulated by expectations generated in higher visual areas (driven by FB connections). Neuroscientists have long been interested in how horizontal connections shape neuronal selectivity \(^{7,8}\) and "non- classical" receptive fields \(^{9 - 13}\) . More recently, neuroscientists have also been interested in adding these connections to deep learning models to understand neuronal selectivity in visual cortex \(^{14,15}\) . It remains unclear, however, how horizontal connections shape the moment- by- moment dynamics and computations in cortex while processing visual input. + +Recent analyses of large- scale recordings have revealed that horizontal connections profoundly shape spatiotemporal dynamics in cortex. Traveling waves driven by horizontal connections have been observed in visual cortex of anesthetized animals \(^{16 - 21}\) . The relevance of traveling waves had previously been called into question, as they were thought to disappear in the awake state \(^{22}\) or to be suppressed by high- contrast visual stimuli \(^{19,23}\) . Recent analyses of neural activity at the single- trial level, however, have revealed spontaneous \(^{24}\) and stimulus- evoked \(^{25}\) activity patterns that travel smoothly across entire cortical regions in awake, behaving primates during normal vision. These neural traveling waves (nTWs) shift the balance of excitation and inhibition as they propagate across cortex, sparsely modulating spiking activity as they pass \(^{26}\) . Because they drive fluctuations in neural excitability \(^{24,27}\) , nTWs show that neurons at one point in a visual area (representing a small section of visual space) can strongly interact with neurons across the entire cortical region. These results thus indicate that cortical neurons may share information about visual scenes broadly across the retinotopic map, through nTWs generated by horizontal connections. + +What computations, then, can be done with waves of neural activity traveling across a map of visual space? To address this question, we studied a complex- valued neural network (cv- NN) processing visual inputs ranging from simple stimuli to natural movies. cv- NNs exhibit similar or superior performance to standard, real- valued neural networks in many supervised learning tasks \(^{28}\) , and have been used effectively in explaining biological neural dynamics \(^{29}\) . Here, we modified the standard FF architecture used in deep learning and computer vision to include horizontal recurrent connections, where neurons in a single processing layer form a web of interconnections similar to the horizontal connections in visual cortex. Horizontal recurrent connections are thought to provide advantages \(^{14}\) over the standard FF architecture used in computer vision tasks \(^{5,30}\) ; however, current methods for training recurrent models severely limit both the time window over which recurrent activity can be considered and the ease with which + +<--- Page Split ---> + +the networks can be trained31,32. In recent work, we have introduced a mathematical approach to understand the recurrent dynamics in a specific complex- valued model33. Here, we leverage this understanding to train recurrent complex- valued networks to process visual inputs, ranging from simple stimuli to naturalistic movie scenes. The training process for the cv- NN is rapid and efficient, requiring only minutes of desktop computer time. The resulting networks can predict learned movies many frames into the future, entirely from their internal dynamics alone, without external input. During prediction, the recurrent network exhibits prominent nTWs, ranging from simple waves propagating out from a small local input25 to complex traveling wave patterns34, raising the possibility that nTWs enable processing spatiotemporally complex, natural, and dynamic visual scenes. + +## Results + +The cv- NN consists of an input layer sending movie frames to a recurrently connected network of model neurons. An individual movie frame, serving as input to the network, is represented by a two- dimensional grid of pixels (input frame, Fig. 1a), and each pixel projects to the recurrently connected layer through FF connections (red lines, Fig. 1a). The recurrently connected layer is arranged on a two- dimensional grid, analogous to the retinotopic arrangement of neurons in visual regions. Horizontal interconnections within the cv- NN then drive recurrent interactions in the network (blue lines, Fig. 1a). Both FF and horizontal recurrent projections in the cv- NN are matched to the approximate scale of connectivity in visual cortex35,36, so that a single pixel in an input movie drives a local patch of neurons, with overlapping horizontal connections, in the cv- NN. Lastly, neurons in the recurrent layer communicate with time delays approximating axonal conduction speeds along horizontal fibers37, which have recently been shown to shape spiking neural activity into nTWs26. The combination of FF input and dense interconnections generates complex patterns of activity in the recurrent layer (Fig. 1b). Here, we focus on these recurrent activity patterns to understand their computational role for movie inputs ranging from simple to complex. + +nTWs can simultaneously encode stimulus position and time of onset over spatial maps + +To understand how nTWs propagating over sensory maps could facilitate visual computation, we first studied the dynamics generated in response to a single point stimulus. Without recurrent connections, a short point stimulus generates a small bump of activity that remains centered on + +<--- Page Split ---> + +the point of input ("without recurrence", Fig. 1c). With recurrent connections, however, the point stimulus generates a wave that propagates out from the point of input ("with recurrence", Fig. 1c). We then studied these stimulus- evoked waves, which are qualitatively similar to those previously observed in visual cortex of awake primate25, in a simple decoding task. Specifically, we let the point stimulus appear at a random time in a series of input frames, and then trained a linear classifier to decode the time of stimulus onset from the network activity at the final frame. As expected, without recurrent connections, the classifier performed at chance- level accuracy in this task (Fig. 1c, right; Methods - Time- of- stimulus prediction task). With recurrence, however, the classifier selects the correct time of stimulus appearance from the final network state with \(100\%\) accuracy. This initial example shows that traveling waves of neural activity, when propagating on an orderly retinotopic map, can allow decoding of stimulus onset time, in addition to stimulus location, even after the stimulus is no longer present. + +![](images/Figure_1.jpg) + +
Figure 1: A topographic recurrent network model encodes temporal information of video frames via internal wave activity. (a) Schematic of the network model. Neurons (circles) are arranged on a two-dimensional grid and are
+ +<--- Page Split ---> + +recurrently connected (blue) locally in space like the cortical sheet. A natural image input projects locally into the network via feedforward connections (red), mimicking retinotopy. (b) Example dynamic of the network model. Due to the spatially local projection of the input image, an imprint of the image is visible in the grid of network activity. Due to the local recurrent connectivity, intrinsic wave activity is generated alongside the input projection. (c) In a sequence of six frames, exactly one of the first five contains a point stimulus, and the other frames do not. These frames are sequentially input to the network. Top row: When the network has no recurrence, the stimulus projection remains stationary. Bottom row: With recurrence, from the time of stimulus, the network activity contains a projection of the stimulus and a wave radiating outward. Right: A linear classifier that received the final network state in the no- recurrence case could not predict the time of stimulus beyond chance- level accuracy. In contrast, using the classifier with the sixth with- recurrence network state allowed \(100\%\) accuracy since the feedforward projection of the point stimulus triggered a radiating wave that encoded the time of stimulus in the subsequent network states. + +nTWs aid forecasting movie inputs from simple to complex + +Can nTWs enable the processing of the complex, dynamic, and non- stationary visual scenes that we encounter in our natural experience? We approached this question in several steps. We first asked whether, given an input frame from a movie, the cv- NN could be trained to accurately predict the following frame. To perform this more complicated task, we introduced a learning rule that requires training only a linear readout of the recurrent layer (Fig. 2a). This procedure is analogous to a complex- valued implementation of the reservoir computing paradigm38 that has recently proven very powerful for learning the dynamics of chaotic systems in physics39. This training process, however, has never before been applied to naturalistic movie scenes. We find the cv- NN can be rapidly, reliably, and efficiently trained to predict the next frame in a movie input (Supplementary Materials - Section III, Table S2, Moving Bump Input). Surprisingly, with a cv- NN trained on a movie input, the predicted next frame generated by the network can then be provided as input, in place of the original movie (Fig. 2b). We call this process closed- loop forecasting of entire visual scenes. + +The visual cortex readily processes and operates on dynamic visual inputs on timescales of milliseconds to seconds. We then asked whether closed- loop forecasting in this system could work on the scale of tens to hundreds of frames in an input movie. Starting with the first half of a movie containing a simple moving bump stimulus tracing out a trajectory in two- dimensional space (Fig. 2c), we find that the trained cv- NN can produce the entire second half of the movie as output from its trained synaptic weights alone (Fig. 2d and Movie S1). As in the previous example, activity in the recurrent layer exhibits a dynamic spatiotemporal pattern extending beyond the immediate FF imprint of the stimulus and structured by the recurrent connections in the network (Fig. 2e and Movie S1). These results demonstrate that recurrent cv- NNs can produce simple video inputs from their recurrent connections through this rapid and efficient + +<--- Page Split ---> + +training process. Finally, when we remove the recurrent connections, the cv- NN produces an activity pattern that represents only the average of FF stimulus imprints, without having learned the underlying spatiotemporal process40. In this case, the cv- NN no longer produces an accurate closed- loop forecast (Fig. 2f). These results demonstrate the importance of both the spatiotemporal patterns in the reservoir and the horizontal recurrent dynamics generating them. + +![](images/Figure_2.jpg) + +
Figure 2: The network can forecast a simple video input many frames into the future. (a) As in the classification example (Figure 1), a video frame projects into the network in a spatially local manner and a recurrent network interaction occurs, generating internal wave activity on top of the projection. The network outputs an image from its network state via a matrix of trainable weights. Training entails one-shot linear regression between a set of network states and the corresponding desired output frames (the one-step-ahead next frames). Shown: a schematic representation of the one-shot linear regression for one time step. (b) Once training of the readout weights is complete, closed-loop forecasting begins. To properly test how well the network model learned the underlying spatiotemporal process from the training data, it is deprived of ground-truth data of any kind during this step. Instead,
+ +<--- Page Split ---> + +the forecast next frame at one time step serves as the input frame for the following time step. (c) Video frames of the data: a bump tracing an orbit. (d) Corresponding closed- loop forecasts generated by the network model with optimal recurrence. (e) Network activity for the optimal- recurrence case. Cosine of phase of activation shown. (f) Closed- loop forecast in the case without recurrence. + +We find that closed- loop forecast performance in this system depends on two key factors: (1) the ratio of horizontal recurrent strength to feedforward input strength and (2) the spatial extent of the recurrence. To study the first factor in detail, we measured closed- loop forecast performance using an index of structural similarity (SSIM) \(^{41}\) , which quantifies the perceptual match between two images. We studied SSIM between movie frames produced by the closed- loop forecast process and the ground truth at different ratios of recurrence to input (Fig. 3a; see also Fig. S1 and Methods - Network connectivity and Network dynamics). Once the stimulus is removed and the closed- loop forecast begins (video frame 1, Fig. 3a), forecast performance in cv- NNs with low recurrent strength quickly drops close to zero (light blue line, Fig. 3a). By contrast, cv- NNs at optimal recurrent strength sustain closed- loop forecasts for long timescales (gray line, Fig. 3a), extending beyond 100 video frames into the future. Importantly, networks where recurrence is too strong also perform poorly, with SSIM dropping near zero within a short timeframe (copper line, Fig. 3a). Systematic quantification of SSIM across ratios of recurrent strength to input strength reveals that performance is best when the recurrence and input are approximately balanced (Fig. 3b), highlighting the importance of the interplay between these two fundamental circuit patterns in visual cortex. We next studied performance as a function of the spatial extent of recurrent connectivity. The best performance occurs for recurrent lengths on approximately the same spatial scale as the moving bump stimulus (Fig. 3c), with performance dropping for recurrent lengths outside this range. This result demonstrates that recurrent connections, which span from local to long- range in visual cortex \(^{6,42}\) , utilize features in the closed- loop forecasting task best when matched to the spatial scale of the input. + +![](images/Figure_3.jpg) + +
Figure 3: Moving bump forecast performance depends on specific properties of the recurrent connections. (a) Structural similarity between a forecast frame and the ground truth as a function of closed-loop forecast video
+ +<--- Page Split ---> + +frame. Each curve corresponds to a different network parameter implementation. Curves have been smoothed by a moving- average filter (filter width of 30 time steps). Shaded error is the absolute difference between filtered and unfiltered. (b) Total structural similarity, in which a single SSIM is calculated for the whole movie, as a function of the recurrence- to- input ratio. In the parameter space, each point differs only in recurrent strength. Smoothing and error shading is the same as in a. (c) Total structural similarity as a function of recurrent length. In the three- dimensional parameter space comprising the recurrent strength (rs), recurrent length (rl), and input strength (is), averages across rs- is planes at fixed rl were computed (gray curve). The peak coincides with the standard- deviation width of the Gaussian bump stimulus (dashed vertical line). Shaded area: variance. Solid black curve: maximum structural similarity at each recurrent length. + +The visual system readily processes richly textured and naturalistic visual scenes. To examine this type of stimulus in the cv- NN, we considered naturalistic video inputs for next- frame prediction and closed- loop forecasting. To do this, we used videos from the Weizmann Human Action Dataset \(^{43}\) . As above, we trained linear readout weights of the cv- NN on these individual naturalistic movie inputs (Fig. 4a) and then tested whether, given the first half of the input movie, the network could produce the second half in a closed- loop forecast (Fig. 4b). Even with a much more sophisticated input than the previous examples, the cv- NN can be trained rapidly and efficiently on the natural movie inputs (Supplementary Materials - Section III, Table S2, Walking Person Input). As in previous examples, at optimal values of the network parameters (Methods - Parameter optimization), the cv- NN accurately produces the natural movie using only its connection weights (Fig. 4c,d and Movie S2). In this case, the recurrent connections in the cv- NN create complex wave patterns (Fig. 4e and Movie S2). The recurrent connections and their resulting complex activity patterns are important for success in this task, as networks without recurrence do not produce accurate closed- loop forecasts (Fig. 4f). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: The recurrent network performs next-frame forecasting of a natural video input. (a) Training follows as in the moving bump example (Fig. 2a). (b) Next-frame closed-loop forecasting follows as in the moving bump example (Fig. 2b). (c) Video frames of the data: a person walking. (d) Corresponding closed-loop forecasts generated by the network model in the case of optimal recurrence. (e) Corresponding network states for the optimal-recurrence case (panel d). Cosine of phase shown. (f) Same as d, but in the absence of recurrence.
+ +We then studied what specific features of the recurrent connections enable predicting naturalistic movie inputs. As in the moving bump example, networks perform best when recurrence and input are approximately balanced, and the performance quickly decays when + +<--- Page Split ---> + +the recurrence is too weak or too strong (Figs. 5a,b). This result shows that, as in the simple case of the moving bump, the complex spatiotemporal predictions generated by the network depend on a sophisticated interplay between input and recurrent connections. We next studied the role of connection topography and distance- dependent time delays. To do this, we started with networks that achieve accurate predictions and randomly shuffled both the connections and time delays (Fig. 6a). We then compared the closed- loop forecast performance and network activity in the topographic and shuffled cases. In the topographic case, the cv- NN produces accurate predictions and complex traveling wave patterns, as before (Fig. 6b,c). The shuffled versions of the cv- NN, however, produce spatiotemporally unstructured activity in the recurrent layer (Fig. 6d) and do not achieve accurate closed- loop forecasts, even after retraining (Fig. 6e; see also Supplementary Materials - Table S3 and Movie S3). Finally, the specific spatiotemporal structure of the input movie is also important: a cv- NN at the optimal hyperparameters for a natural movie cannot be retrained to do closed- loop forecasting on a randomized (phase- shuffled) version of the same movie (Supplementary Materials - Table S1), demonstrating that the cv- NN utilizes the specific spatiotemporal correlations in the movie to generate its forecast. Taken together, these results demonstrate that the complex spatiotemporal patterns generated by horizontal recurrent connections in the cv- NN enable performance on next- frame prediction and closed- loop forecasting tasks for sophisticated natural movie inputs. + +![](images/Figure_5.jpg) + +
Figure 5: Natural movie forecast performance depends on specific properties of the recurrent connections. (a) Several examples of closed-loop forecast performance. Structural similarity between a forecast frame and the ground truth as a function of video frame during closed-loop forecasting. Each curve corresponds to a different ratio of recurrent strength to input strength. Curves have been smoothed by a moving-average filter (filter width of 30 time steps). Shaded error is the absolute difference between filtered and unfiltered. (b) Total structural similarity, in which a single SSIM is computed for the whole movie, as a function of the recurrence-to-input ratio. In the parameter space, each point differs only in recurrent strength. Smoothing and error shading is the same as in a.
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6: Randomly shuffling recurrent connections eliminates nTWs and ability to forecast. (a) Left: the topographic network model used throughout this study, featuring feedforward projections of the image input (red lines) and local distance-dependent horizontal connectivity (blue lines). There are also synaptic time delays proportional to a neuron pair's separation distance within the horizontal recurrent circuitry. Right: by randomizing the horizontal connection weights and time delays assigned to the network neurons, the topography in the network is
+ +<--- Page Split ---> + +removed. (b) The network activity of the topographic network in response to frames of a natural movie input. (c) Closed- loop forecasts generated by the topographic network. Forecast frames correspond to network states in a. (d) Network activity of the shuffled network. (e) Closed- loop forecasts generated by the shuffled network. + +The nTW network model is capable of forecasting multiple movies without retraining + +We lastly sought to understand whether the cv- NN could perform closed- loop forecasts on multiple movies it had previously learned, and switch flexibly with changing inputs. To do this, we implemented a simple competitive process (Methods - Movie switching), so that the network could adapt its output based on the similarity of its prediction to its input (Fig. 7a). When performing a closed- loop forecast, this extended network model can receive a new input from its learned set, and then rapidly switch to closed- loop forecasting this new movie input within a few frames (Fig. 7b and Movie S4). This result demonstrates that the process of closed- loop forecasting, mediated by horizontal recurrent fibers in the network, can generalize to realistic visual conditions with multiple, changing input streams. + +![](images/Figure_7.jpg) + +
Figure 7: The network is capable of forecasting multiple movies without being retrained. (a) The recurrent network model was adapted to contain a higher-level competitive-learning process. Left: Readout matrices were learned separately for separate examples. Right: Storing the learned readout matrices in a higher-level matrix, the
+ +<--- Page Split ---> + +present network state drove the aggregate matrix toward either of the learned matrices via an unsupervised competitive learning rule. (b) Beginning with feeding frames from movie 1, the network takes some time to recall the learned matrix that results in an accurate closed- loop forecast. Quickly switching to a different movie, the network once again takes some time to adjust its output weights before converging to the correct ones for an accurate closed- loop forecast. + +## Discussion + +In this work, we have introduced a model to understand whether traveling waves generated by horizontal connections in visual cortex may play a computational role in processing natural visual inputs. By adapting a recurrent neural network model using a new dynamical update rule and a new learning rule, this model can efficiently learn video inputs ranging from simple visual stimuli to complex natural scenes. Further, this network model is broadly consistent with spatiotemporal dynamics recently observed in the visual system of the alert primate. In the case of a single point stimulus (Fig. 1c), the network produces a traveling wave radiating out from the point of input, similar to the responses observed in primary and secondary visual cortex of the awake monkey in response to a brief visual input25. In the case of a moving bump stimulus (Fig. 2), the network produces a bump of activity, reflecting the movie input but embedded within a larger spatiotemporal pattern, consistent with anticipatory responses observed in retinal populations44 and in primary visual cortex45. Finally, in the case of naturalistic movie inputs (Fig. 4), the network produces complex spatiotemporal patterns, which can be mathematically described in this model as the summation of multiple traveling waves33,34. The responses of this network to natural movie inputs ranging from simple to complex are thus qualitatively consistent with observations of neuronal dynamics in vivo. + +These results provide fundamental insight into the function of horizontal recurrent connections, whose effect on the moment- by- moment dynamics in the visual system has remained unexplained. While there has been much interest in the function of recurrent horizontal fibers in visual cortex, for example in explaining direction and orientation selectivity in \(\mathrm{V}^{17,8}\) , or in center- surround models of the receptive field11,13,46, general computational roles for traveling waves generated by the massive recurrent circuitry in single cortical areas on the single- trial level remain unknown. Successful models of the visual system, including feature- based models and deep convolutional neural networks, have provided insight into how neural systems could process single image inputs, but explain only a fraction of the variance in neural responses to natural sensory stimuli15,47,48. The cv- NN may provide new opportunities for understanding how + +<--- Page Split ---> + +the visual system processes continuously updated, movie- like visual inputs, where information is extracted from the visual environment moment- by- moment as it comes from the eye. The sophisticated closed- loop movie forecasts produced by this network, and the fact that this closed- loop forecast process can generalize to multiple movie inputs, represent an important step in explaining the computational role of recurrent connections and traveling waves in visual cortex. + +## Methods + +## Network connectivity + +The recurrent network is arranged on a square grid of \(N\) nodes. The network grid is treated as a discretized Euclidean plane such that the side lengths span distances of unity. Boundaries are not periodic. The recurrent weight \(w_{ij}\) from node \(j\) to node \(i\) is inversely proportional to their Euclidean distance \(d_{ij}\) so as to give local connectivity like that of the neocortical sheet. Specifically, \(w_{ij}\) is Gaussian as a function of \(d_{ij}\) : + +\[w_{ij} = ae^{-d_{ij}^2 / (2b^2)}.\] + +The coefficient \(a\) is called the recurrent strength and the standard deviation \(b\) is called the recurrent length. Both are free parameters. The maximum possible value of \(d_{ij}\) is \(\sqrt{2}\) (corner to corner), and, for example, \(b = 1\) means that the recurrent length equals the network side length. Further, all \(\mathbb{N}^2\) such weights are strictly positive, and the N- by- N matrix of such weights is symmetric ( \(w_{ij} = w_{ji}\) ). Diagonal weights ( \(w_{ii}\) ) are not set to zero. + +## Network dynamics + +Network dynamics are given by a complex- valued equation. A complex number \(z\) is of the form \(z = x + \mathrm{i}y\) , where \(x\) is the real part, \(y\) is the imaginary part, and \(i\) is the imaginary constant defined as \(\mathrm{i}^2 = - 1\) . Equivalently, \(z = me^{\mathrm{i}\phi}\) , where \(m\) is the modulus and \(\phi\) is the argument. A complex number is intuitively visualized as a two- dimensional vector, where \((x,y)\) is its Cartesian representation and \((m,\phi)\) is its polar representation. What distinguishes a complex number from a standard two- dimensional vector is the multiplication rule: multiplication of two + +<--- Page Split ---> + +complex numbers corresponds to both a scaling and a rotation in the so- called complex plane. This property makes complex- valued representations of observable quantities more concise than real- valued representations, and thus, complex numbers are a central tool in physics and engineering. From the perspective of biological vision, a complex- valued representation is useful. Since phase information is important for representing visual inputs, complex- valued models, which efficiently represent phase in the argument \(\phi\) , are ideal. Indeed, complex- valued models of vision are widely explored49. Given the practical utility of artificial neural networks and deep learning (including for modeling biological neural networks), complex- valued neural networks, in which the neural activations are complex- valued, are of great interest. However, they are notoriously difficult to train, especially in a recurrent architecture50. We make an advance here on this front by choosing a unique dynamical equation and by exploiting the advantages of reservoir computing. + +The discrete- time dynamical equation for each node \(i\) is + +\[\begin{array}{l}{a_{i}[t] = a_{i}[t - 1] + \left(x_{i}[t] - \mathrm{i}\sum_{j = 1}^{N}w_{ij}\mathrm{e}^{\mathrm{i}\left(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1]\right)}\right),}\\ {a_{i}[t]:= a_{i}[t] / \left|a_{i}[t]\right|.} \end{array} \quad (1b)\] + +Here, \(a_{i}[t]\) is the complex- valued activation, \(x_{i}[t]\) is the feedforward input of the image stimulus to node \(i\) , and \(w_{ij}\) is the recurrent weight from node \(j\) to node \(j\) (Methods - Network connectivity). Further, \(\tau_{ij}\) is the discrete time delay between nodes \(i\) and \(j\) , given by \(\tau_{ij} = \mathrm{round}[d_{ij} / v]\) in which the Euclidean distance \(d_{ij}\) between nodes \(i\) and \(j\) (Methods - Network connectivity) is scaled by the parameter \(v\) , which represents the speed of activation transmission across the network, and \(\mathrm{round}[d_{ij} / v]\) rounds \(d_{ij} / v\) to the nearest integer in accord with the discrete- time dynamics. The value of \(v\) is randomly sampled between 0 and 0.1 \((v \in (0, 0.1))\) , meaning the activation travels a distance of up to one- tenth the network side length per time step. Lastly, the modulus of \(a_{i}[t]\) (i.e., \(|a_{i}[t]|\) ) is normalized (Equation 1b), which confines \(a_{i}[t]\) on the complex unit circle, and thus, the phase of \(a_{i}[t]\) contains the dynamics. We note that modulus normalization is a common operation used in complex- valued neural networks50. + +<--- Page Split ---> + +The specific form of Equation 1a is unique compared to other complex- valued neural- network equations because it involves a pairwise node attraction \(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1]\) . Another system with pairwise attraction is the Kuramoto model, a popular model for studying synchronization in nonlinear systems \(^{51 - 53}\) . Our presented system has a correspondence with the Kuramoto model \(^{54}\) , and allows the description of the dynamics for individual realization in terms of the eigenvalues and eigenvectors of the network \(^{33}\) . Along with the choice of local network connectivity and distance- dependent delays, the presented system gives rise to meaningful spatiotemporal self- organization dynamics, and for this reason, the recurrent weights \(\{w_{ij}\}\) need not be trained. + +The initial network state is \(a_{i}[0] = 0 + 0\mathrm{i}\) for all nodes, and the first several time steps contain transient activity associated with the input disrupting the initial steady state of the system. For the time- of- stimulus prediction task, this transient activity is important to the model and was used, while for the next- frame forecasting task, it is distracting to the model and was discarded. + +## Image read-in + +Each discrete time step, a digital grayscale image is read into the network. Prior to read- in, the image is mean- subtracted and divided by its standard deviation across all its pixels (i.e., z- scored). Image read- in is accomplished with a local feedforward projection, which mimics retinotopy and preserves the spatial correlations in the image. Technically, this is a two- dimensional interpolation using the bilinear kernel common in image processing, which takes a weighted average in the nearest 2- by- 2 pixel neighbourhood. The projected image has \(\sqrt{N}\) rows and \(\sqrt{N}\) columns like the network grid, and each pixel intensity of the projected image is given by \(x_{i}[t]\) (Equation 1a). Lastly, \(x_{i}[t]\) is scaled according to \(x_{i}[t] := \epsilon x_{i}[t]\) , where \(\epsilon\) is called the input strength. In our model, \(\epsilon\) is the third and final free parameter after the recurrent strength and recurrent length. + +## Time-of-stimulus prediction task + +Classification was performed using the basic perceptron. For an input vector \(\vec{v} = (1, v_{1}, \dots , v_{N})^{T}\) , where \(v_{1}, \dots , v_{N}\) are features, and a label \(l \in \{0, 1\}\) , the goal is to find a hyperplane \(\vec{u}^{T} \vec{v} = b + u_{1} v_{1} + \dots + u_{N} v_{N} = 0\) , where \(\vec{u} = (b, u_{1}, \dots , u_{N})^{T}\) is a + +<--- Page Split ---> + +vector containing the bias \(b\) and weights \(u_{1},\ldots ,u_{N}\) , that separates the data in the N- dimensional feature space according to their binary class (0 or 1). During training, with a sub- optimal \(\vec{u}\) vector and one example \(\vec{v}\) vector, the output classification \(l = H(\vec{u}^{T}\vec{v})\) is computed, where \(H(\cdot)\) is the Heaviside step function defined as unity for positive argument and zero otherwise. For the desired classification \(d\) (either 0 or 1), the signed distance \(\Delta = d - l\) is computed, where \(\Delta \in \{- 1,0,1\}\) . With each new example \(\vec{v}\) , the \(\vec{u}\) vector is updated using the delta rule \(\vec{u}:= \vec{u} +\lambda \vec{v}\Delta\) , where \(\lambda\) is the learning rate. To use the perceptron in multiclass classification, the one- versus- rest scheme is used. That is, for the set of classes \(C = \{c_{1},c_{2},\ldots ,c_{i},\ldots ,c_{M}\}\) , binary classification is performed separately \(M\) times. Each time \(i\) , the two classes are defined such that \(c_{i} = 1\) and \(C\setminus c_{i} = 0\) , where " \(\cdot\) " denotes the set difference. Then, there are M weight vectors \(\vec{u}_{1},\ldots ,\vec{u}_{i},\ldots ,\vec{u}_{M}\) , and M inner products \(f_{1} = \vec{u}_{1}^{T}\vec{v}\) , ..., \(f_{i} = \vec{u}_{i}^{T}\vec{v}\) , ..., \(f_{M} = u_{M}^{T}\vec{v}\) for a given data vector \(\vec{v}\) . The multiclass classification is \(\mathrm{argmax}_{c_{i}}\{f_{1},\ldots ,f_{i},\ldots ,f_{M}\}\) . + +In the time- of- stimulus classification task (Fig. 1c), input frames were 50 by 50 pixels, and the network was 50 by 50 nodes. There were six frames. One of the first five frames was randomly chosen to contain the point stimulus, and the remaining frames were entirely zero intensity. The point stimulus was an isotropic two- dimensional Gaussian of standard deviation 0.05, and the input frames are defined on the Cartesian grid \([- 2,2]\times [- 2,2]\) . The stimulus was centered in the frame. The sequence of frames was sequentially input to the network. There are exactly five classes: each of the first five frames in which the point stimulus could occur. The column vector of activations corresponding to the final (sixth) frame was used as predictors for all trials. The task was repeated 100 000 times, with the time of stimulus (1 or 2 or 3 or 4 or 5) randomly rechosen each time. + +## Next-frame forecasting + +The network outputs an image of \(M_{r}\) rows and \(M_{c}\) columns of pixels- the same size as the input image- at each time step. In both examples (moving bump and natural movie), the network was 50 by 50 nodes. Recalling that \(a_{i}[t]\) is the complex- valued activation of node \(i\) at discrete time t (Equations 1a and 1b), the output transformation is linear: + +<--- Page Split ---> + +\[y_{i}[t] = \sum_{j = 1}^{N}v_{ij}a_{j}[t]^{\prime}.\] + +Here, \(y_{i}[t]\) is the \(i^{\mathrm{th}}\) pixel intensity of the output image, and \(v_{ij}\) is the \((i,j)^{\mathrm{th}}\) readout weight of the M- by- N matrix \(V\) , where \(M = M_{r}M_{c}\) . The prime notation (') indicates that \(a[t] = \left(a_{1}[t], a_{2}[t], \dots , a_{N}[t]\right)^{T}\) was mean- subtracted, which was done to avoid an intercept term during training. + +The readout weights \(\{v_{ij}\}\) of \(V\) are the only weights trained in our model, making our network a reservoir computer. Reservoir computers are recurrent neural networks that avoid the issues associated with training recurrent weights, and have been shown to perform well in time series forecasting \(^{38}\) . Suppose training begins at time step 1, after discarding the initial transient, and ends at time step T. Defining \(a[t]^{\prime} = \left(a_{1}[t]^{\prime}, a_{2}[t]^{\prime}, \dots , a_{N}[t]^{\prime}\right)^{T}\) , the matrix of regressors is then + +\[A = \left[a[1]^{\prime} a[2]^{\prime} \dots a[T]^{\prime}\right],\] + +and the matrix of regressands (desired outputs) is + +\[D = \left[f[2] f[3] \dots f[T + 1]\right].\] + +Hence, the desired outputs are simply the set of one- step- ahead frames. Here, \(f[t]\) is the column vectorization of the \(t^{\mathrm{th}}\) input image frame (before read- in), and is also mean- subtracted. Training entails ordinary least- squares linear regression between \(A\) and \(D\) . Because \(D\) is highly underdetermined (containing far fewer frames than pixels per frame), the matrix 2- norm of \(V\) was simultaneously minimized during regression to reduce model bias. + +Following training is closed- loop forecasting. At this point, the network activation has been primed by being driven with the training frames, and the readout matrix \(V\) has been trained. In the first time step of closed- loop forecasting, we input the corresponding video frame. + +<--- Page Split ---> + +Subsequently for steps \(\{t\}\) , the predicted output at time step \(t\) serves as the input for time step \(t + 1\) . + +In the moving bump example (Fig. 2), the frames are 30 by 30 pixels and defined on a \([- 2,2] \times [- 2,2]\) Cartesian grid. A two- dimensional isotropic Gaussian of standard deviation 0.2 traced a Lissajous curve given by the parametric equations \(x_{c}(t) = \sin (t / 3)\) and \(y_{c}(t) = \cos (t)\) , where \((x_{c}, y_{c})\) is the center of the Gaussian in space and \(t\) is a continuously valued time variable \(^{55}\) . The Lissajous trajectory was discretized to have 100 frames per cycle. The first cycle was discarded to omit the initial transient network activity, the network was trained on the subsequent 3 cycles, and closed- loop forecasting was performed on the 2 cycles subsequent to that. + +In the natural video example (Fig. 4), a walking video from the Weizmann Human Action Dataset \(^{56}\) was used, in which a person walks across the scene. We present several key examples here, but note that the model successfully performs closed- loop forecasting for all movies in this dataset. Segmentation masks of the people in the videos are included with this dataset (https://www.wisdom.weizmann.ac.il/vision/SpaceTimeActions.html). Using these masks, we cropped the frames so that the person was centered throughout the entire walk, giving frames of approximately 80 by 50 pixels. Without performing this step, our network model would fail: the training data would be independent from the closed- loop forecast data since they would occupy exclusive regions of the pixel space, and the model would not generalize to the prediction data. Such nonstationary data have been successfully taught to networks with approximate translation invariance, and translation invariance is likely used in the brain to learn such processes \(^{57}\) . However, translation invariance is beyond the scope of our study. The frames were then resized to be exactly 80 by 50 pixels. Finally, each video was around 70 frames long. To get more frames without interpolation, we "bookended" each video by concatenating it with its temporal reverse sequence, where one cycle consists of the original frames followed by the bookended frames. The result is a longer video with the same spatiotemporal statistics. The first cycle was discarded to omit the initial transient network activity, the network was trained on the subsequent 3 cycles, and closed- loop prediction was performed on the 2 cycles subsequent to that. + +<--- Page Split ---> + +To measure the balance between feedforward input and recurrent interaction, we devised the recurrence- to- input ratio. Per Equation 1a, the input and recurrence terms are the column vectors + +\[x[t] = \{x_{i}[t]\}_{i = 1}^{N}\] + +and + +\[r[t] = \left\{-\mathrm{i}\sum_{j = 1}^{N}w_{ij}\mathrm{e}^{\mathrm{i}(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1])}\right\}_{i = 1}^{N},\] + +respectively. Further, let the matrices + +\[R = \left[r[t] r[t + 1] \dots r[t']\right]\] + +and + +\[X = \left[x[t] x[t + 1] \dots x[t']\right]\] + +be the horizontal concatenations of \(r[t]\) and \(x[t]\) , respectively, over closed- loop forecast times \(\{t, t + 1, \ldots , t'\}\) . The ratio is defined as + +\[\frac{\|R\|_F}{\|X\|_F},\] + +where \(\| G\| _F\) denotes the Frobenius matrix norm of a matrix G, which is equivalent to the Euclidean vector norm of the vectorization of G. + +Movie switching + +<--- Page Split ---> + +The network was trained on two movie inputs: one of a walking person (movie 1) and one of a jumping person (movie 2), both from the Weizmann dataset. The same recurrent matrix was used in each case- only the learned matrices \((V_{1}\) and \(V_{2}\) , respectively) differed. Let \(\mathcal{V} = c V_{1} + (1 - c) V_{2}\) , where \(c \in [0,1]\) . \(\mathcal{V}\) stores both learned matrices, and the present input modulates the relative contribution of \(V_{1}\) and \(V_{2}\) using an update rule for \(c\) . The structural similarity between the input and output were computed at each time step t ( \(\mathrm{SSIM}[t]\) ), and the change thereof was computed at each time step as \(\Delta \mathrm{SSIM} = \mathrm{SSIM}[t] - \mathrm{SSIM}[t - 1]\) . The update rule is \(c := c + \Delta c\) , where \(\Delta c = - \eta \operatorname {sgn}[\Delta \mathrm{SSIM}]\) and \(\eta\) is the learning rate, set to 0.1. Depending on which movie (movie 1 or movie 2) drives the network, \(c\) tends toward 1 or 0, respectively. Once this happens, this driving input is removed and closed- loop forecasting commences as described. Switching entails instantaneously transitioning from closed- loop forecasting of one movie to driving the network with the frames of another movie. \(c\) then updates as described and is followed by closed- loop forecasting again. + +## Parameter optimization + +The random- search algorithm was used to optimize parameters for closed- loop forecasting. Within specified bounds, each parameter was randomly sampled, giving a point in the parameter space. The parameter space was randomly sampled in this way many times, and each time, the structural similarity index was computed as the performance index. The bounds within which the parameters were sampled are given in Table I. + +
parametersampled interval
recurrent strength(0, 0.2)
recurrent length(0, 0.2)
input strength(0, 0.2)
v(0, 0.1)
+ +Table I: Intervals over which model parameters were randomly searched during optimization. + +Acknowledgements: This work was supported by the Canadian Institute for Health Research and NSF (NeuroNex Grant No. 2015276), BrainsCAN at Western University through the Canada First Research Excellence Fund (CFREF), Gatsby Charitable Foundation, the Fiona and Sanjay Jha Chair in Neuroscience, the Swartz Foundation, Compute Ontario (computeontario.ca), + +<--- Page Split ---> + +Compute Canada (computecanada.ca), SPIRITS 2020 of Kyoto University, and the Western Academy for Advanced Research. R.C.B. gratefully acknowledges the Western Institute for Neuroscience Clinical Research Postdoctoral Fellowship. G.B.B. gratefully acknowledges the Canadian Open Neuroscience Platform (Graduate Scholarship), the Vector Institute (Postgraduate Affiliate), and the National Sciences and Engineering Research Council of Canada (Canada Graduate Scholarship - Doctoral). The authors thank Alex Busch for her help with illustrations. + +Data and code availability: All data and code associated with this work are available at https://github.com/mullerlab. + +Contributions: Conceptualization: G.B.B., L.M.; data curation: G.B.B.; formal analysis: G.B.B., L.M.; funding acquisition: J.R., L.M.; investigation: G.B.B., L.M.; methodology: G.B.B., R.C.B., Z.D., J.R., L.M.; supervision: J.R., L.M.; visualization: G.B.B.; writing- original draft: G.B.B., L.M.; writing- review and editing: G.B.B., R.C.B., Z.D., J.R., L.M. + +## References + +1. Markov, N. T. et al. Weight consistency specifies regularities of macaque cortical networks. Cereb. Cortex 21, 1254–1272 (2011). +2. Bruno, R. M. & Sakmann, B. Cortex is driven by weak but synchronously active thalamocortical synapses. Science 312, 1622–1627 (2006). +3. Swindale, N. V. Visual map. Scholarpedia J. 3, 4607 (2008). +4. Hubel, D. H. & Wiesel, T. N. Receptive fields of single neurones in the cat's striate cortex. J. Physiol. 148, 574–591 (1959). +5. Riesenhuber, M. & Poggio, T. Hierarchical models of object recognition in cortex. Nat. Neurosci. 2, 1019–1025 (1999). +6. Angelucci, A. et al. Circuits for local and global signal integration in primary visual cortex. J. Neurosci. 22, 8633–8646 (2002). +7. Douglas, R. J., Koch, C., Mahowald, M., Martin, K. A. & Suarez, H. H. Recurrent excitation in neocortical circuits. Science 269, 981–985 (1995). +8. Sompolinsky, H. & Shapley, R. New perspectives on the mechanisms for orientation selectivity. Curr. + +<--- Page Split ---> + +Opin. Neurobiol. 7, 514–522 (1997). + +9. Blakemore, C. & Tobin, E. A. Lateral inhibition between orientation detectors in the cat's visual cortex. Exp. Brain Res. 15, 439–440 (1972). + +10. Allman, J., Miezin, F. & McGuinness, E. Stimulus specific responses from beyond the classical receptive field: neurophysiological mechanisms for local-global comparisons in visual neurons. Annu. Rev. Neurosci. 8, 407–430 (1985). + +11. Field, D. J., Hayes, A. & Hess, R. F. Contour integration by the human visual system: Evidence for a local 'association field'. Vision Res. 33, 173–193 (1993). + +12. Gilbert, C. D. Adult cortical dynamics. Physiol. Rev. 78, 467–485 (1998). + +13. Albright, T. D. & Stoner, G. R. Contextual influences on visual processing. Annu. Rev. Neurosci. 25, 339–379 (2002). + +14. Kietzmann, T. C. et al. Recurrence is required to capture the representational dynamics of the human visual system. Proc. Natl. Acad. Sci. U. S. A. 116, 21854–21863 (2019). + +15. Kar, K., Kubilius, J., Schmidt, K., Issa, E. B. & DiCarlo, J. J. Evidence that recurrent circuits are critical to the ventral stream's execution of core object recognition behavior. Nat. Neurosci. 22, 974–983 (2019). + +16. Bringuier, V., Chavane, F., Glaeser, L. & Frégnac, Y. Horizontal propagation of visual activity in the synaptic integration field of area 17 neurons. Science 283, 695–699 (1999). + +17. Roland, P. E. et al. Cortical feedback depolarization waves: a mechanism of top-down influence on early visual areas. Proc. Natl. Acad. Sci. U. S. A. 103, 12586–12591 (2006). + +18. Xu, W., Huang, X., Takagaki, K. & Wu, J.-Y. Compression and reflection of visually evoked cortical waves. Neuron 55, 119–129 (2007). + +19. Nauhaus, I., Busse, L., Carandini, M. & Ringach, D. L. Stimulus contrast modulates functional connectivity in visual cortex. Nat. Neurosci. 12, 70–76 (2009). + +20. Reimer, A., Hubka, P., Engel, A. K. & Kral, A. Fast propagating waves within the rodent auditory cortex. Cereb. Cortex 21, 166–177 (2011). + +21. Townsend, R. G. et al. Emergence of complex wave patterns in primate cerebral cortex. J. Neurosci. 35, 4657–4662 (2015). + +<--- Page Split ---> + +22. Slovin, H., Arieli, A., Hildesheim, R. & Grinvald, A. Long-term voltage-sensitive dye imaging reveals cortical dynamics in behaving monkeys. J. Neurophysiol. 88, 3421–3438 (2002). + +23. Sato, T. K., Nauhaus, I. & Carandini, M. Traveling waves in visual cortex. Neuron 75, 218–229 (2012). + +24. Davis, Z. W., Muller, L., Martinez-Trujillo, J., Sejnowski, T. & Reynolds, J. H. Spontaneous travelling cortical waves gate perception in behaving primates. Nature 587, 432–436 (2020). + +25. Muller, L., Reynaud, A., Chavane, F. & Destexhe, A. The stimulus-evoked population response in visual cortex of awake monkey is a propagating wave. Nat. Commun. 5, 3675 (2014). + +26. Davis, Z. et al. Spontaneous traveling waves naturally emerge from horizontal fiber time delays and travel through locally asynchronous-irregular states. Nature Communications (2021). + +27. Takahashi, K. et al. Large-scale spatiotemporal spike patterning consistent with wave propagation in motor cortex. Nat. Commun. 6, 7169 (2015). + +28. Trabelsi, C. et al. Deep Complex Networks. arXiv [cs.NE] (2017). + +29. Heeger, D. J. & Mackey, W. E. Oscillatory recurrent gated neural integrator circuits (ORGaNICs), a unifying theoretical framework for neural dynamics. Proc. Natl. Acad. Sci. U. S. A. 116, 22783–22794 (2019). + +30. Krizhevsky, A., Sutskever, I. & Hinton, G. E. Imagenet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 25, (2012). + +31. Graves, A. Supervised Sequence Labelling with Recurrent Neural Networks. Studies in Computational Intelligence (2012) doi:10.1007/978-3-642-24797-2. + +32. Hwang, K. & Sung, W. Online Sequence Training of Recurrent Neural Networks with Connectionist Temporal Classification. arXiv [cs.LG] (2015). + +33. Budzinski, R. C. et al. Geometry unites synchrony, chimeras, and waves in nonlinear oscillator networks. Chaos 32, 031104 (2022). + +34. Muller, L., Chavane, F., Reynolds, J. & Sejnowski, T. J. Cortical travelling waves: mechanisms and computational principles. Nat. Rev. Neurosci. 19, 255–268 (2018). + +35. Hellwig, B. A quantitative analysis of the local connectivity between pyramidal neurons in layers 2/3 of the rat visual cortex. Biol. Cybern. 82, 111–121 (2000). + +<--- Page Split ---> + +36. Binzegger, T., Douglas, R. J. & Martin, K. A. C. A quantitative map of the circuit of cat primary visual cortex. J. Neurosci. 24, 8441–8453 (2004). + +37. Girard, P., Hupé, J. M. & Bullier, J. Feedforward and feedback connections between areas V1 and V2 of the monkey have similar rapid conduction velocities. J. Neurophysiol. 85, 1328–1331 (2001). + +38. Jaeger, H. & Haas, H. Harnessing Nonlinearity: Predicting Chaotic Systemsand Saving Energy in Wireless Communication. Science 304, (2004). + +39. Pathak, J., Hunt, B., Girvan, M., Lu, Z. & Ott, E. Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach. Phys. Rev. Lett. 120, 024102 (2018). + +40. Mathieu, M., Couprie, C. & LeCun, Y. Deep multi-scale video prediction beyond mean square error. arXiv [cs.LG] (2015). + +41. Wang, Z., Bovik, A. C., Sheikh, H. R. & Simoncelli, E. P. Image quality assessment: from error visibility to structural similarity. IEEE Trans. Image Process. 13, 600–612 (2004). + +42. Stettler, D. D., Das, A., Bennett, J. & Gilbert, C. D. Lateral connectivity and contextual interactions in macaque primary visual cortex. Neuron 36, 739–750 (2002). + +43. Moshe, B., Lena, G., Eli, S., Michal, I. & Ronen, B. Actions as space-time shapes. in Proceedings of the Tenth IEEE International Conference on Computer Vision, Beijing, China 17–20 (2005). + +44. Berry, M. J., Brivanlou, I. H., Jordan, T. A. & Meister, M. Anticipation of moving stimuli by the retina. Nature vol. 398 334–338 (1999). + +45. Benvenuti, G. et al. Anticipatory responses along motion trajectories in awake monkey area V1. 2020.03.26.010017 (2020) doi:10.1101/2020.03.26.010017. + +46. Hess, R. F. & Dakin, S. C. Contour integration in the peripheral field. Vision Res. 39, 947–959 (1999). + +47. Olshausen, B. A. How Close Are We to Understanding V1? Neural Comput. 17, 1665–1699 (2005). + +48. Schrimpf, M. et al. Brain-Score: Which Artificial Neural Network for Object Recognition is most Brain-Like? bioRxiv 407007 (2020) doi:10.1101/407007. + +49. Cadieu, C. F. & Olshausen, B. A. Learning intermediate-level representations of form and motion from natural movies. Neural Comput. 24, 827–866 (2012). + +50. Bassey, J., Qian, L. & Li, X. A Survey of Complex-Valued Neural Networks. arXiv [stat.ML] (2021). + +51. Acebrón, J. A., Bonilla, L. L., Pérez Vicente, C. J., Ritort, F. & Spigler, R. The Kuramoto model: A + +<--- Page Split ---> + +simple paradigm for synchronization phenomena. Rev. Mod. Phys. 77, 137–185 (2005). + +52. Rodrigues, F. A., Peron, T. K. D. M., Ji, P. & Kurths, J. The Kuramoto model in complex networks. Phys. Rep. 610, 1–98 (2016). + +53. Breakspear, M., Heitmann, S. & Daffertshofer, A. Generative models of cortical oscillations: neurobiological implications of the kuramoto model. Front. Hum. Neurosci. 4, 190 (2010). + +54. Muller, L., Mináč, J. & Nguyen, T. T. Algebraic approach to the Kuramoto model. Phys Rev E 104, L022201 (2021). + +55. Heim, N. & Avery, J. E. Adaptive Anomaly Detection in Chaotic Time Series with a Spatially Aware Echo State Network. arXiv [cs.NE] (2019). + +56. Gorelick, L., Blank, M., Shechtman, E., Irani, M. & Basri, R. Actions as space-time shapes. IEEE Trans. Pattern Anal. Mach. Intell. 29, 2247–2253 (2007). + +57. Anselmi, F. et al. Unsupervised learning of invariant representations. Theor. Comput. Sci. 633, 112–121 (2016). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- wc1 supplement.pdf- movies1.mp4- movies2.mp4- movies3.mp4- movies4.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545_det.mmd b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..ed6438661c23d1dc192664d3a093e9cdd1507b65 --- /dev/null +++ b/preprint/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545/preprint__11be1947534bff4b18a5d6916674863dc9009e00419509497a68870547ccb545_det.mmd @@ -0,0 +1,526 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 855, 175]]<|/det|> +# What computations can be done with traveling waves in visual cortex? + +<|ref|>text<|/ref|><|det|>[[44, 196, 218, 237]]<|/det|> +Gabriel Benigno Western University + +<|ref|>text<|/ref|><|det|>[[44, 243, 218, 283]]<|/det|> +Roberto Budzinski Western University + +<|ref|>text<|/ref|><|det|>[[44, 290, 720, 330]]<|/det|> +Zachary Davis Salk Institute for Biological Studies https://orcid.org/0000- 0003- 4440- 9011 + +<|ref|>text<|/ref|><|det|>[[44, 336, 757, 377]]<|/det|> +John Reynolds The Salk Institute for Biological Studies https://orcid.org/0000- 0001- 6988- 4607 + +<|ref|>text<|/ref|><|det|>[[44, 380, 576, 422]]<|/det|> +Lyle Muller ( ☑ Imuller2@uwo.ca ) Western University https://orcid.org/0000- 0001- 5165- 9890 + +<|ref|>text<|/ref|><|det|>[[44, 464, 102, 481]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 502, 137, 520]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 539, 321, 559]]<|/det|> +Posted Date: August 29th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 577, 474, 597]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1903144/v1 + +<|ref|>text<|/ref|><|det|>[[44, 614, 910, 657]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[165, 88, 828, 109]]<|/det|> +# What computations can be done with traveling waves in visual cortex? + +<|ref|>text<|/ref|><|det|>[[155, 125, 844, 164]]<|/det|> +Gabriel B. Benigno \(^{1,2,3}\) , Roberto C. Budzinski \(^{1,2,3}\) , Zachary Davis \(^{4}\) , John Reynolds \(^{4}\) , Lyle Muller \(^{1,2,3,*}\) + +<|ref|>text<|/ref|><|det|>[[114, 181, 877, 273]]<|/det|> +Affiliations: \(^{1}\) Department of Mathematics, Western University, London, ON, Canada, \(^{2}\) Brain and Mind Institute, Western University, London, ON, Canada, \(^{3}\) Western Academy for Advanced Research, Western University, London, ON, Canada, \(^{4}\) The Salk Institute for Biological Studies, La Jolla, CA, USA, \(^{*}\) Corresponding author: Imuller2@uwo.ca. + +<|ref|>text<|/ref|><|det|>[[113, 300, 884, 608]]<|/det|> +Recent analyses have found waves of neural activity traveling across entire visual cortical areas in awake animals. These traveling waves modulate excitability of local networks and perceptual sensitivity. The general computational role for these spatiotemporal patterns in the visual system, however, remains unclear. Here, we hypothesize that traveling waves endow the brain with the capacity to predict complex and naturalistic visual inputs. We present a new network model whose connections can be rapidly and efficiently trained to predict natural movies. After training, a few input frames from a movie trigger complex wave patterns that drive accurate predictions many frames into the future, solely from the network's connections. When the recurrent connections that drive waves are randomly shuffled, both traveling waves and the ability to predict are eliminated. These results show traveling waves could play an essential computational role in the visual system by embedding continuous spatiotemporal structures over spatial maps. + +<|ref|>sub_title<|/ref|><|det|>[[115, 637, 222, 654]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[114, 684, 884, 895]]<|/det|> +Five percent of synapses received by a neuron in visual cortex arrive through the feedforward (FF) pathway that conveys sensory input from the eyes \(^{1}\) . While these FF synapses are strong \(^{2}\) , "horizontal" recurrent connections coming from within the cortical region make up about 80% of total synaptic inputs, with 95% of these connections arising from a very local patch (2 mm) around the cell \(^{1}\) . The anatomy of the visual system thus indicates that cortical neurons interact with other neurons across the retinotopically organized maps \(^{3}\) that assign nearby points in visual space to nearby points in a cortical region, via these horizontal connections. Models of the visual system predominantly focus only on FF \(^{4,5}\) and feedback (FB) \(^{6}\) connections. One result of this focus is that neurons in visual cortex are often conceived to be non- interacting "feature + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 87, 884, 252]]<|/det|> +detectors", with fixed selectivity to features in visual input (driven by FF connections) that can be modulated by expectations generated in higher visual areas (driven by FB connections). Neuroscientists have long been interested in how horizontal connections shape neuronal selectivity \(^{7,8}\) and "non- classical" receptive fields \(^{9 - 13}\) . More recently, neuroscientists have also been interested in adding these connections to deep learning models to understand neuronal selectivity in visual cortex \(^{14,15}\) . It remains unclear, however, how horizontal connections shape the moment- by- moment dynamics and computations in cortex while processing visual input. + +<|ref|>text<|/ref|><|det|>[[113, 279, 884, 610]]<|/det|> +Recent analyses of large- scale recordings have revealed that horizontal connections profoundly shape spatiotemporal dynamics in cortex. Traveling waves driven by horizontal connections have been observed in visual cortex of anesthetized animals \(^{16 - 21}\) . The relevance of traveling waves had previously been called into question, as they were thought to disappear in the awake state \(^{22}\) or to be suppressed by high- contrast visual stimuli \(^{19,23}\) . Recent analyses of neural activity at the single- trial level, however, have revealed spontaneous \(^{24}\) and stimulus- evoked \(^{25}\) activity patterns that travel smoothly across entire cortical regions in awake, behaving primates during normal vision. These neural traveling waves (nTWs) shift the balance of excitation and inhibition as they propagate across cortex, sparsely modulating spiking activity as they pass \(^{26}\) . Because they drive fluctuations in neural excitability \(^{24,27}\) , nTWs show that neurons at one point in a visual area (representing a small section of visual space) can strongly interact with neurons across the entire cortical region. These results thus indicate that cortical neurons may share information about visual scenes broadly across the retinotopic map, through nTWs generated by horizontal connections. + +<|ref|>text<|/ref|><|det|>[[113, 637, 884, 899]]<|/det|> +What computations, then, can be done with waves of neural activity traveling across a map of visual space? To address this question, we studied a complex- valued neural network (cv- NN) processing visual inputs ranging from simple stimuli to natural movies. cv- NNs exhibit similar or superior performance to standard, real- valued neural networks in many supervised learning tasks \(^{28}\) , and have been used effectively in explaining biological neural dynamics \(^{29}\) . Here, we modified the standard FF architecture used in deep learning and computer vision to include horizontal recurrent connections, where neurons in a single processing layer form a web of interconnections similar to the horizontal connections in visual cortex. Horizontal recurrent connections are thought to provide advantages \(^{14}\) over the standard FF architecture used in computer vision tasks \(^{5,30}\) ; however, current methods for training recurrent models severely limit both the time window over which recurrent activity can be considered and the ease with which + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 323]]<|/det|> +the networks can be trained31,32. In recent work, we have introduced a mathematical approach to understand the recurrent dynamics in a specific complex- valued model33. Here, we leverage this understanding to train recurrent complex- valued networks to process visual inputs, ranging from simple stimuli to naturalistic movie scenes. The training process for the cv- NN is rapid and efficient, requiring only minutes of desktop computer time. The resulting networks can predict learned movies many frames into the future, entirely from their internal dynamics alone, without external input. During prediction, the recurrent network exhibits prominent nTWs, ranging from simple waves propagating out from a small local input25 to complex traveling wave patterns34, raising the possibility that nTWs enable processing spatiotemporally complex, natural, and dynamic visual scenes. + +<|ref|>sub_title<|/ref|><|det|>[[115, 353, 182, 369]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[113, 399, 884, 755]]<|/det|> +The cv- NN consists of an input layer sending movie frames to a recurrently connected network of model neurons. An individual movie frame, serving as input to the network, is represented by a two- dimensional grid of pixels (input frame, Fig. 1a), and each pixel projects to the recurrently connected layer through FF connections (red lines, Fig. 1a). The recurrently connected layer is arranged on a two- dimensional grid, analogous to the retinotopic arrangement of neurons in visual regions. Horizontal interconnections within the cv- NN then drive recurrent interactions in the network (blue lines, Fig. 1a). Both FF and horizontal recurrent projections in the cv- NN are matched to the approximate scale of connectivity in visual cortex35,36, so that a single pixel in an input movie drives a local patch of neurons, with overlapping horizontal connections, in the cv- NN. Lastly, neurons in the recurrent layer communicate with time delays approximating axonal conduction speeds along horizontal fibers37, which have recently been shown to shape spiking neural activity into nTWs26. The combination of FF input and dense interconnections generates complex patterns of activity in the recurrent layer (Fig. 1b). Here, we focus on these recurrent activity patterns to understand their computational role for movie inputs ranging from simple to complex. + +<|ref|>text<|/ref|><|det|>[[115, 783, 812, 801]]<|/det|> +nTWs can simultaneously encode stimulus position and time of onset over spatial maps + +<|ref|>text<|/ref|><|det|>[[115, 831, 883, 898]]<|/det|> +To understand how nTWs propagating over sensory maps could facilitate visual computation, we first studied the dynamics generated in response to a single point stimulus. Without recurrent connections, a short point stimulus generates a small bump of activity that remains centered on + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 371]]<|/det|> +the point of input ("without recurrence", Fig. 1c). With recurrent connections, however, the point stimulus generates a wave that propagates out from the point of input ("with recurrence", Fig. 1c). We then studied these stimulus- evoked waves, which are qualitatively similar to those previously observed in visual cortex of awake primate25, in a simple decoding task. Specifically, we let the point stimulus appear at a random time in a series of input frames, and then trained a linear classifier to decode the time of stimulus onset from the network activity at the final frame. As expected, without recurrent connections, the classifier performed at chance- level accuracy in this task (Fig. 1c, right; Methods - Time- of- stimulus prediction task). With recurrence, however, the classifier selects the correct time of stimulus appearance from the final network state with \(100\%\) accuracy. This initial example shows that traveling waves of neural activity, when propagating on an orderly retinotopic map, can allow decoding of stimulus onset time, in addition to stimulus location, even after the stimulus is no longer present. + +<|ref|>image<|/ref|><|det|>[[160, 400, 828, 855]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 865, 883, 897]]<|/det|> +
Figure 1: A topographic recurrent network model encodes temporal information of video frames via internal wave activity. (a) Schematic of the network model. Neurons (circles) are arranged on a two-dimensional grid and are
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 255]]<|/det|> +recurrently connected (blue) locally in space like the cortical sheet. A natural image input projects locally into the network via feedforward connections (red), mimicking retinotopy. (b) Example dynamic of the network model. Due to the spatially local projection of the input image, an imprint of the image is visible in the grid of network activity. Due to the local recurrent connectivity, intrinsic wave activity is generated alongside the input projection. (c) In a sequence of six frames, exactly one of the first five contains a point stimulus, and the other frames do not. These frames are sequentially input to the network. Top row: When the network has no recurrence, the stimulus projection remains stationary. Bottom row: With recurrence, from the time of stimulus, the network activity contains a projection of the stimulus and a wave radiating outward. Right: A linear classifier that received the final network state in the no- recurrence case could not predict the time of stimulus beyond chance- level accuracy. In contrast, using the classifier with the sixth with- recurrence network state allowed \(100\%\) accuracy since the feedforward projection of the point stimulus triggered a radiating wave that encoded the time of stimulus in the subsequent network states. + +<|ref|>text<|/ref|><|det|>[[115, 277, 582, 295]]<|/det|> +nTWs aid forecasting movie inputs from simple to complex + +<|ref|>text<|/ref|><|det|>[[113, 324, 884, 633]]<|/det|> +Can nTWs enable the processing of the complex, dynamic, and non- stationary visual scenes that we encounter in our natural experience? We approached this question in several steps. We first asked whether, given an input frame from a movie, the cv- NN could be trained to accurately predict the following frame. To perform this more complicated task, we introduced a learning rule that requires training only a linear readout of the recurrent layer (Fig. 2a). This procedure is analogous to a complex- valued implementation of the reservoir computing paradigm38 that has recently proven very powerful for learning the dynamics of chaotic systems in physics39. This training process, however, has never before been applied to naturalistic movie scenes. We find the cv- NN can be rapidly, reliably, and efficiently trained to predict the next frame in a movie input (Supplementary Materials - Section III, Table S2, Moving Bump Input). Surprisingly, with a cv- NN trained on a movie input, the predicted next frame generated by the network can then be provided as input, in place of the original movie (Fig. 2b). We call this process closed- loop forecasting of entire visual scenes. + +<|ref|>text<|/ref|><|det|>[[114, 661, 884, 896]]<|/det|> +The visual cortex readily processes and operates on dynamic visual inputs on timescales of milliseconds to seconds. We then asked whether closed- loop forecasting in this system could work on the scale of tens to hundreds of frames in an input movie. Starting with the first half of a movie containing a simple moving bump stimulus tracing out a trajectory in two- dimensional space (Fig. 2c), we find that the trained cv- NN can produce the entire second half of the movie as output from its trained synaptic weights alone (Fig. 2d and Movie S1). As in the previous example, activity in the recurrent layer exhibits a dynamic spatiotemporal pattern extending beyond the immediate FF imprint of the stimulus and structured by the recurrent connections in the network (Fig. 2e and Movie S1). These results demonstrate that recurrent cv- NNs can produce simple video inputs from their recurrent connections through this rapid and efficient + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 202]]<|/det|> +training process. Finally, when we remove the recurrent connections, the cv- NN produces an activity pattern that represents only the average of FF stimulus imprints, without having learned the underlying spatiotemporal process40. In this case, the cv- NN no longer produces an accurate closed- loop forecast (Fig. 2f). These results demonstrate the importance of both the spatiotemporal patterns in the reservoir and the horizontal recurrent dynamics generating them. + +<|ref|>image<|/ref|><|det|>[[133, 220, 861, 765]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 784, 883, 905]]<|/det|> +
Figure 2: The network can forecast a simple video input many frames into the future. (a) As in the classification example (Figure 1), a video frame projects into the network in a spatially local manner and a recurrent network interaction occurs, generating internal wave activity on top of the projection. The network outputs an image from its network state via a matrix of trainable weights. Training entails one-shot linear regression between a set of network states and the corresponding desired output frames (the one-step-ahead next frames). Shown: a schematic representation of the one-shot linear regression for one time step. (b) Once training of the readout weights is complete, closed-loop forecasting begins. To properly test how well the network model learned the underlying spatiotemporal process from the training data, it is deprived of ground-truth data of any kind during this step. Instead,
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 882, 150]]<|/det|> +the forecast next frame at one time step serves as the input frame for the following time step. (c) Video frames of the data: a bump tracing an orbit. (d) Corresponding closed- loop forecasts generated by the network model with optimal recurrence. (e) Network activity for the optimal- recurrence case. Cosine of phase of activation shown. (f) Closed- loop forecast in the case without recurrence. + +<|ref|>text<|/ref|><|det|>[[112, 177, 884, 677]]<|/det|> +We find that closed- loop forecast performance in this system depends on two key factors: (1) the ratio of horizontal recurrent strength to feedforward input strength and (2) the spatial extent of the recurrence. To study the first factor in detail, we measured closed- loop forecast performance using an index of structural similarity (SSIM) \(^{41}\) , which quantifies the perceptual match between two images. We studied SSIM between movie frames produced by the closed- loop forecast process and the ground truth at different ratios of recurrence to input (Fig. 3a; see also Fig. S1 and Methods - Network connectivity and Network dynamics). Once the stimulus is removed and the closed- loop forecast begins (video frame 1, Fig. 3a), forecast performance in cv- NNs with low recurrent strength quickly drops close to zero (light blue line, Fig. 3a). By contrast, cv- NNs at optimal recurrent strength sustain closed- loop forecasts for long timescales (gray line, Fig. 3a), extending beyond 100 video frames into the future. Importantly, networks where recurrence is too strong also perform poorly, with SSIM dropping near zero within a short timeframe (copper line, Fig. 3a). Systematic quantification of SSIM across ratios of recurrent strength to input strength reveals that performance is best when the recurrence and input are approximately balanced (Fig. 3b), highlighting the importance of the interplay between these two fundamental circuit patterns in visual cortex. We next studied performance as a function of the spatial extent of recurrent connectivity. The best performance occurs for recurrent lengths on approximately the same spatial scale as the moving bump stimulus (Fig. 3c), with performance dropping for recurrent lengths outside this range. This result demonstrates that recurrent connections, which span from local to long- range in visual cortex \(^{6,42}\) , utilize features in the closed- loop forecasting task best when matched to the spatial scale of the input. + +<|ref|>image<|/ref|><|det|>[[114, 694, 888, 869]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 872, 884, 904]]<|/det|> +
Figure 3: Moving bump forecast performance depends on specific properties of the recurrent connections. (a) Structural similarity between a forecast frame and the ground truth as a function of closed-loop forecast video
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 224]]<|/det|> +frame. Each curve corresponds to a different network parameter implementation. Curves have been smoothed by a moving- average filter (filter width of 30 time steps). Shaded error is the absolute difference between filtered and unfiltered. (b) Total structural similarity, in which a single SSIM is calculated for the whole movie, as a function of the recurrence- to- input ratio. In the parameter space, each point differs only in recurrent strength. Smoothing and error shading is the same as in a. (c) Total structural similarity as a function of recurrent length. In the three- dimensional parameter space comprising the recurrent strength (rs), recurrent length (rl), and input strength (is), averages across rs- is planes at fixed rl were computed (gray curve). The peak coincides with the standard- deviation width of the Gaussian bump stimulus (dashed vertical line). Shaded area: variance. Solid black curve: maximum structural similarity at each recurrent length. + +<|ref|>text<|/ref|><|det|>[[113, 247, 884, 579]]<|/det|> +The visual system readily processes richly textured and naturalistic visual scenes. To examine this type of stimulus in the cv- NN, we considered naturalistic video inputs for next- frame prediction and closed- loop forecasting. To do this, we used videos from the Weizmann Human Action Dataset \(^{43}\) . As above, we trained linear readout weights of the cv- NN on these individual naturalistic movie inputs (Fig. 4a) and then tested whether, given the first half of the input movie, the network could produce the second half in a closed- loop forecast (Fig. 4b). Even with a much more sophisticated input than the previous examples, the cv- NN can be trained rapidly and efficiently on the natural movie inputs (Supplementary Materials - Section III, Table S2, Walking Person Input). As in previous examples, at optimal values of the network parameters (Methods - Parameter optimization), the cv- NN accurately produces the natural movie using only its connection weights (Fig. 4c,d and Movie S2). In this case, the recurrent connections in the cv- NN create complex wave patterns (Fig. 4e and Movie S2). The recurrent connections and their resulting complex activity patterns are important for success in this task, as networks without recurrence do not produce accurate closed- loop forecasts (Fig. 4f). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[143, 90, 857, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 744, 883, 821]]<|/det|> +
Figure 4: The recurrent network performs next-frame forecasting of a natural video input. (a) Training follows as in the moving bump example (Fig. 2a). (b) Next-frame closed-loop forecasting follows as in the moving bump example (Fig. 2b). (c) Video frames of the data: a person walking. (d) Corresponding closed-loop forecasts generated by the network model in the case of optimal recurrence. (e) Corresponding network states for the optimal-recurrence case (panel d). Cosine of phase shown. (f) Same as d, but in the absence of recurrence.
+ +<|ref|>text<|/ref|><|det|>[[114, 835, 882, 902]]<|/det|> +We then studied what specific features of the recurrent connections enable predicting naturalistic movie inputs. As in the moving bump example, networks perform best when recurrence and input are approximately balanced, and the performance quickly decays when + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 540]]<|/det|> +the recurrence is too weak or too strong (Figs. 5a,b). This result shows that, as in the simple case of the moving bump, the complex spatiotemporal predictions generated by the network depend on a sophisticated interplay between input and recurrent connections. We next studied the role of connection topography and distance- dependent time delays. To do this, we started with networks that achieve accurate predictions and randomly shuffled both the connections and time delays (Fig. 6a). We then compared the closed- loop forecast performance and network activity in the topographic and shuffled cases. In the topographic case, the cv- NN produces accurate predictions and complex traveling wave patterns, as before (Fig. 6b,c). The shuffled versions of the cv- NN, however, produce spatiotemporally unstructured activity in the recurrent layer (Fig. 6d) and do not achieve accurate closed- loop forecasts, even after retraining (Fig. 6e; see also Supplementary Materials - Table S3 and Movie S3). Finally, the specific spatiotemporal structure of the input movie is also important: a cv- NN at the optimal hyperparameters for a natural movie cannot be retrained to do closed- loop forecasting on a randomized (phase- shuffled) version of the same movie (Supplementary Materials - Table S1), demonstrating that the cv- NN utilizes the specific spatiotemporal correlations in the movie to generate its forecast. Taken together, these results demonstrate that the complex spatiotemporal patterns generated by horizontal recurrent connections in the cv- NN enable performance on next- frame prediction and closed- loop forecasting tasks for sophisticated natural movie inputs. + +<|ref|>image<|/ref|><|det|>[[152, 560, 845, 780]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 783, 883, 888]]<|/det|> +
Figure 5: Natural movie forecast performance depends on specific properties of the recurrent connections. (a) Several examples of closed-loop forecast performance. Structural similarity between a forecast frame and the ground truth as a function of video frame during closed-loop forecasting. Each curve corresponds to a different ratio of recurrent strength to input strength. Curves have been smoothed by a moving-average filter (filter width of 30 time steps). Shaded error is the absolute difference between filtered and unfiltered. (b) Total structural similarity, in which a single SSIM is computed for the whole movie, as a function of the recurrence-to-input ratio. In the parameter space, each point differs only in recurrent strength. Smoothing and error shading is the same as in a.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[191, 100, 808, 816]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 820, 883, 897]]<|/det|> +
Figure 6: Randomly shuffling recurrent connections eliminates nTWs and ability to forecast. (a) Left: the topographic network model used throughout this study, featuring feedforward projections of the image input (red lines) and local distance-dependent horizontal connectivity (blue lines). There are also synaptic time delays proportional to a neuron pair's separation distance within the horizontal recurrent circuitry. Right: by randomizing the horizontal connection weights and time delays assigned to the network neurons, the topography in the network is
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 882, 134]]<|/det|> +removed. (b) The network activity of the topographic network in response to frames of a natural movie input. (c) Closed- loop forecasts generated by the topographic network. Forecast frames correspond to network states in a. (d) Network activity of the shuffled network. (e) Closed- loop forecasts generated by the shuffled network. + +<|ref|>text<|/ref|><|det|>[[117, 157, 783, 175]]<|/det|> +The nTW network model is capable of forecasting multiple movies without retraining + +<|ref|>text<|/ref|><|det|>[[113, 204, 884, 415]]<|/det|> +We lastly sought to understand whether the cv- NN could perform closed- loop forecasts on multiple movies it had previously learned, and switch flexibly with changing inputs. To do this, we implemented a simple competitive process (Methods - Movie switching), so that the network could adapt its output based on the similarity of its prediction to its input (Fig. 7a). When performing a closed- loop forecast, this extended network model can receive a new input from its learned set, and then rapidly switch to closed- loop forecasting this new movie input within a few frames (Fig. 7b and Movie S4). This result demonstrates that the process of closed- loop forecasting, mediated by horizontal recurrent fibers in the network, can generalize to realistic visual conditions with multiple, changing input streams. + +<|ref|>image<|/ref|><|det|>[[137, 440, 857, 836]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 848, 883, 895]]<|/det|> +
Figure 7: The network is capable of forecasting multiple movies without being retrained. (a) The recurrent network model was adapted to contain a higher-level competitive-learning process. Left: Readout matrices were learned separately for separate examples. Right: Storing the learned readout matrices in a higher-level matrix, the
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 882, 166]]<|/det|> +present network state drove the aggregate matrix toward either of the learned matrices via an unsupervised competitive learning rule. (b) Beginning with feeding frames from movie 1, the network takes some time to recall the learned matrix that results in an accurate closed- loop forecast. Quickly switching to a different movie, the network once again takes some time to adjust its output weights before converging to the correct ones for an accurate closed- loop forecast. + +<|ref|>sub_title<|/ref|><|det|>[[115, 213, 212, 230]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[113, 258, 884, 639]]<|/det|> +In this work, we have introduced a model to understand whether traveling waves generated by horizontal connections in visual cortex may play a computational role in processing natural visual inputs. By adapting a recurrent neural network model using a new dynamical update rule and a new learning rule, this model can efficiently learn video inputs ranging from simple visual stimuli to complex natural scenes. Further, this network model is broadly consistent with spatiotemporal dynamics recently observed in the visual system of the alert primate. In the case of a single point stimulus (Fig. 1c), the network produces a traveling wave radiating out from the point of input, similar to the responses observed in primary and secondary visual cortex of the awake monkey in response to a brief visual input25. In the case of a moving bump stimulus (Fig. 2), the network produces a bump of activity, reflecting the movie input but embedded within a larger spatiotemporal pattern, consistent with anticipatory responses observed in retinal populations44 and in primary visual cortex45. Finally, in the case of naturalistic movie inputs (Fig. 4), the network produces complex spatiotemporal patterns, which can be mathematically described in this model as the summation of multiple traveling waves33,34. The responses of this network to natural movie inputs ranging from simple to complex are thus qualitatively consistent with observations of neuronal dynamics in vivo. + +<|ref|>text<|/ref|><|det|>[[113, 666, 884, 902]]<|/det|> +These results provide fundamental insight into the function of horizontal recurrent connections, whose effect on the moment- by- moment dynamics in the visual system has remained unexplained. While there has been much interest in the function of recurrent horizontal fibers in visual cortex, for example in explaining direction and orientation selectivity in \(\mathrm{V}^{17,8}\) , or in center- surround models of the receptive field11,13,46, general computational roles for traveling waves generated by the massive recurrent circuitry in single cortical areas on the single- trial level remain unknown. Successful models of the visual system, including feature- based models and deep convolutional neural networks, have provided insight into how neural systems could process single image inputs, but explain only a fraction of the variance in neural responses to natural sensory stimuli15,47,48. The cv- NN may provide new opportunities for understanding how + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 227]]<|/det|> +the visual system processes continuously updated, movie- like visual inputs, where information is extracted from the visual environment moment- by- moment as it comes from the eye. The sophisticated closed- loop movie forecasts produced by this network, and the fact that this closed- loop forecast process can generalize to multiple movie inputs, represent an important step in explaining the computational role of recurrent connections and traveling waves in visual cortex. + +<|ref|>sub_title<|/ref|><|det|>[[115, 256, 191, 273]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 304, 283, 321]]<|/det|> +## Network connectivity + +<|ref|>text<|/ref|><|det|>[[113, 350, 884, 477]]<|/det|> +The recurrent network is arranged on a square grid of \(N\) nodes. The network grid is treated as a discretized Euclidean plane such that the side lengths span distances of unity. Boundaries are not periodic. The recurrent weight \(w_{ij}\) from node \(j\) to node \(i\) is inversely proportional to their Euclidean distance \(d_{ij}\) so as to give local connectivity like that of the neocortical sheet. Specifically, \(w_{ij}\) is Gaussian as a function of \(d_{ij}\) : + +<|ref|>equation<|/ref|><|det|>[[423, 504, 574, 527]]<|/det|> +\[w_{ij} = ae^{-d_{ij}^2 / (2b^2)}.\] + +<|ref|>text<|/ref|><|det|>[[113, 558, 884, 681]]<|/det|> +The coefficient \(a\) is called the recurrent strength and the standard deviation \(b\) is called the recurrent length. Both are free parameters. The maximum possible value of \(d_{ij}\) is \(\sqrt{2}\) (corner to corner), and, for example, \(b = 1\) means that the recurrent length equals the network side length. Further, all \(\mathbb{N}^2\) such weights are strictly positive, and the N- by- N matrix of such weights is symmetric ( \(w_{ij} = w_{ji}\) ). Diagonal weights ( \(w_{ii}\) ) are not set to zero. + +<|ref|>sub_title<|/ref|><|det|>[[115, 710, 265, 726]]<|/det|> +## Network dynamics + +<|ref|>text<|/ref|><|det|>[[113, 755, 884, 908]]<|/det|> +Network dynamics are given by a complex- valued equation. A complex number \(z\) is of the form \(z = x + \mathrm{i}y\) , where \(x\) is the real part, \(y\) is the imaginary part, and \(i\) is the imaginary constant defined as \(\mathrm{i}^2 = - 1\) . Equivalently, \(z = me^{\mathrm{i}\phi}\) , where \(m\) is the modulus and \(\phi\) is the argument. A complex number is intuitively visualized as a two- dimensional vector, where \((x,y)\) is its Cartesian representation and \((m,\phi)\) is its polar representation. What distinguishes a complex number from a standard two- dimensional vector is the multiplication rule: multiplication of two + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 375]]<|/det|> +complex numbers corresponds to both a scaling and a rotation in the so- called complex plane. This property makes complex- valued representations of observable quantities more concise than real- valued representations, and thus, complex numbers are a central tool in physics and engineering. From the perspective of biological vision, a complex- valued representation is useful. Since phase information is important for representing visual inputs, complex- valued models, which efficiently represent phase in the argument \(\phi\) , are ideal. Indeed, complex- valued models of vision are widely explored49. Given the practical utility of artificial neural networks and deep learning (including for modeling biological neural networks), complex- valued neural networks, in which the neural activations are complex- valued, are of great interest. However, they are notoriously difficult to train, especially in a recurrent architecture50. We make an advance here on this front by choosing a unique dynamical equation and by exploiting the advantages of reservoir computing. + +<|ref|>text<|/ref|><|det|>[[114, 402, 560, 420]]<|/det|> +The discrete- time dynamical equation for each node \(i\) is + +<|ref|>equation<|/ref|><|det|>[[229, 447, 767, 536]]<|/det|> +\[\begin{array}{l}{a_{i}[t] = a_{i}[t - 1] + \left(x_{i}[t] - \mathrm{i}\sum_{j = 1}^{N}w_{ij}\mathrm{e}^{\mathrm{i}\left(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1]\right)}\right),}\\ {a_{i}[t]\coloneqq a_{i}[t] / \left|a_{i}[t]\right|.} \end{array} \quad (1b)\] + +<|ref|>text<|/ref|><|det|>[[112, 565, 884, 874]]<|/det|> +Here, \(a_{i}[t]\) is the complex- valued activation, \(x_{i}[t]\) is the feedforward input of the image stimulus to node \(i\) , and \(w_{ij}\) is the recurrent weight from node \(j\) to node \(j\) (Methods - Network connectivity). Further, \(\tau_{ij}\) is the discrete time delay between nodes \(i\) and \(j\) , given by \(\tau_{ij} = \mathrm{round}[d_{ij} / v]\) in which the Euclidean distance \(d_{ij}\) between nodes \(i\) and \(j\) (Methods - Network connectivity) is scaled by the parameter \(v\) , which represents the speed of activation transmission across the network, and \(\mathrm{round}[d_{ij} / v]\) rounds \(d_{ij} / v\) to the nearest integer in accord with the discrete- time dynamics. The value of \(v\) is randomly sampled between 0 and 0.1 \((v \in (0, 0.1))\) , meaning the activation travels a distance of up to one- tenth the network side length per time step. Lastly, the modulus of \(a_{i}[t]\) (i.e., \(|a_{i}[t]|\) ) is normalized (Equation 1b), which confines \(a_{i}[t]\) on the complex unit circle, and thus, the phase of \(a_{i}[t]\) contains the dynamics. We note that modulus normalization is a common operation used in complex- valued neural networks50. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 306]]<|/det|> +The specific form of Equation 1a is unique compared to other complex- valued neural- network equations because it involves a pairwise node attraction \(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1]\) . Another system with pairwise attraction is the Kuramoto model, a popular model for studying synchronization in nonlinear systems \(^{51 - 53}\) . Our presented system has a correspondence with the Kuramoto model \(^{54}\) , and allows the description of the dynamics for individual realization in terms of the eigenvalues and eigenvectors of the network \(^{33}\) . Along with the choice of local network connectivity and distance- dependent delays, the presented system gives rise to meaningful spatiotemporal self- organization dynamics, and for this reason, the recurrent weights \(\{w_{ij}\}\) need not be trained. + +<|ref|>text<|/ref|><|det|>[[114, 335, 884, 430]]<|/det|> +The initial network state is \(a_{i}[0] = 0 + 0\mathrm{i}\) for all nodes, and the first several time steps contain transient activity associated with the input disrupting the initial steady state of the system. For the time- of- stimulus prediction task, this transient activity is important to the model and was used, while for the next- frame forecasting task, it is distracting to the model and was discarded. + +<|ref|>sub_title<|/ref|><|det|>[[115, 460, 228, 477]]<|/det|> +## Image read-in + +<|ref|>text<|/ref|><|det|>[[113, 505, 884, 750]]<|/det|> +Each discrete time step, a digital grayscale image is read into the network. Prior to read- in, the image is mean- subtracted and divided by its standard deviation across all its pixels (i.e., z- scored). Image read- in is accomplished with a local feedforward projection, which mimics retinotopy and preserves the spatial correlations in the image. Technically, this is a two- dimensional interpolation using the bilinear kernel common in image processing, which takes a weighted average in the nearest 2- by- 2 pixel neighbourhood. The projected image has \(\sqrt{N}\) rows and \(\sqrt{N}\) columns like the network grid, and each pixel intensity of the projected image is given by \(x_{i}[t]\) (Equation 1a). Lastly, \(x_{i}[t]\) is scaled according to \(x_{i}[t] \coloneqq \epsilon x_{i}[t]\) , where \(\epsilon\) is called the input strength. In our model, \(\epsilon\) is the third and final free parameter after the recurrent strength and recurrent length. + +<|ref|>sub_title<|/ref|><|det|>[[117, 779, 372, 796]]<|/det|> +## Time-of-stimulus prediction task + +<|ref|>text<|/ref|><|det|>[[114, 826, 884, 904]]<|/det|> +Classification was performed using the basic perceptron. For an input vector \(\vec{v} = (1, v_{1}, \dots , v_{N})^{T}\) , where \(v_{1}, \dots , v_{N}\) are features, and a label \(l \in \{0, 1\}\) , the goal is to find a hyperplane \(\vec{u}^{T} \vec{v} = b + u_{1} v_{1} + \dots + u_{N} v_{N} = 0\) , where \(\vec{u} = (b, u_{1}, \dots , u_{N})^{T}\) is a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 85, 884, 432]]<|/det|> +vector containing the bias \(b\) and weights \(u_{1},\ldots ,u_{N}\) , that separates the data in the N- dimensional feature space according to their binary class (0 or 1). During training, with a sub- optimal \(\vec{u}\) vector and one example \(\vec{v}\) vector, the output classification \(l = H(\vec{u}^{T}\vec{v})\) is computed, where \(H(\cdot)\) is the Heaviside step function defined as unity for positive argument and zero otherwise. For the desired classification \(d\) (either 0 or 1), the signed distance \(\Delta = d - l\) is computed, where \(\Delta \in \{- 1,0,1\}\) . With each new example \(\vec{v}\) , the \(\vec{u}\) vector is updated using the delta rule \(\vec{u}\coloneqq \vec{u} +\lambda \vec{v}\Delta\) , where \(\lambda\) is the learning rate. To use the perceptron in multiclass classification, the one- versus- rest scheme is used. That is, for the set of classes \(C = \{c_{1},c_{2},\ldots ,c_{i},\ldots ,c_{M}\}\) , binary classification is performed separately \(M\) times. Each time \(i\) , the two classes are defined such that \(c_{i} = 1\) and \(C\setminus c_{i} = 0\) , where " \(\cdot\) " denotes the set difference. Then, there are M weight vectors \(\vec{u}_{1},\ldots ,\vec{u}_{i},\ldots ,\vec{u}_{M}\) , and M inner products \(f_{1} = \vec{u}_{1}^{T}\vec{v}\) , ..., \(f_{i} = \vec{u}_{i}^{T}\vec{v}\) , ..., \(f_{M} = u_{M}^{T}\vec{v}\) for a given data vector \(\vec{v}\) . The multiclass classification is \(\mathrm{argmax}_{c_{i}}\{f_{1},\ldots ,f_{i},\ldots ,f_{M}\}\) . + +<|ref|>text<|/ref|><|det|>[[113, 458, 884, 697]]<|/det|> +In the time- of- stimulus classification task (Fig. 1c), input frames were 50 by 50 pixels, and the network was 50 by 50 nodes. There were six frames. One of the first five frames was randomly chosen to contain the point stimulus, and the remaining frames were entirely zero intensity. The point stimulus was an isotropic two- dimensional Gaussian of standard deviation 0.05, and the input frames are defined on the Cartesian grid \([- 2,2]\times [- 2,2]\) . The stimulus was centered in the frame. The sequence of frames was sequentially input to the network. There are exactly five classes: each of the first five frames in which the point stimulus could occur. The column vector of activations corresponding to the final (sixth) frame was used as predictors for all trials. The task was repeated 100 000 times, with the time of stimulus (1 or 2 or 3 or 4 or 5) randomly rechosen each time. + +<|ref|>sub_title<|/ref|><|det|>[[115, 727, 300, 744]]<|/det|> +## Next-frame forecasting + +<|ref|>text<|/ref|><|det|>[[113, 774, 883, 871]]<|/det|> +The network outputs an image of \(M_{r}\) rows and \(M_{c}\) columns of pixels- the same size as the input image- at each time step. In both examples (moving bump and natural movie), the network was 50 by 50 nodes. Recalling that \(a_{i}[t]\) is the complex- valued activation of node \(i\) at discrete time t (Equations 1a and 1b), the output transformation is linear: + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[416, 88, 580, 139]]<|/det|> +\[y_{i}[t] = \sum_{j = 1}^{N}v_{ij}a_{j}[t]^{\prime}.\] + +<|ref|>text<|/ref|><|det|>[[113, 170, 884, 279]]<|/det|> +Here, \(y_{i}[t]\) is the \(i^{\mathrm{th}}\) pixel intensity of the output image, and \(v_{ij}\) is the \((i,j)^{\mathrm{th}}\) readout weight of the M- by- N matrix \(V\) , where \(M = M_{r}M_{c}\) . The prime notation (') indicates that \(a[t] = \left(a_{1}[t], a_{2}[t], \dots , a_{N}[t]\right)^{T}\) was mean- subtracted, which was done to avoid an intercept term during training. + +<|ref|>text<|/ref|><|det|>[[113, 308, 884, 460]]<|/det|> +The readout weights \(\{v_{ij}\}\) of \(V\) are the only weights trained in our model, making our network a reservoir computer. Reservoir computers are recurrent neural networks that avoid the issues associated with training recurrent weights, and have been shown to perform well in time series forecasting \(^{38}\) . Suppose training begins at time step 1, after discarding the initial transient, and ends at time step T. Defining \(a[t]^{\prime} = \left(a_{1}[t]^{\prime}, a_{2}[t]^{\prime}, \dots , a_{N}[t]^{\prime}\right)^{T}\) , the matrix of regressors is then + +<|ref|>equation<|/ref|><|det|>[[381, 488, 614, 531]]<|/det|> +\[A = \left[a[1]^{\prime} a[2]^{\prime} \dots a[T]^{\prime}\right],\] + +<|ref|>text<|/ref|><|det|>[[113, 562, 514, 581]]<|/det|> +and the matrix of regressands (desired outputs) is + +<|ref|>equation<|/ref|><|det|>[[368, 610, 630, 653]]<|/det|> +\[D = \left[f[2] f[3] \dots f[T + 1]\right].\] + +<|ref|>text<|/ref|><|det|>[[113, 685, 884, 808]]<|/det|> +Hence, the desired outputs are simply the set of one- step- ahead frames. Here, \(f[t]\) is the column vectorization of the \(t^{\mathrm{th}}\) input image frame (before read- in), and is also mean- subtracted. Training entails ordinary least- squares linear regression between \(A\) and \(D\) . Because \(D\) is highly underdetermined (containing far fewer frames than pixels per frame), the matrix 2- norm of \(V\) was simultaneously minimized during regression to reduce model bias. + +<|ref|>text<|/ref|><|det|>[[113, 835, 883, 904]]<|/det|> +Following training is closed- loop forecasting. At this point, the network activation has been primed by being driven with the training frames, and the readout matrix \(V\) has been trained. In the first time step of closed- loop forecasting, we input the corresponding video frame. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 135]]<|/det|> +Subsequently for steps \(\{t\}\) , the predicted output at time step \(t\) serves as the input for time step \(t + 1\) . + +<|ref|>text<|/ref|><|det|>[[113, 162, 884, 364]]<|/det|> +In the moving bump example (Fig. 2), the frames are 30 by 30 pixels and defined on a \([- 2,2] \times [- 2,2]\) Cartesian grid. A two- dimensional isotropic Gaussian of standard deviation 0.2 traced a Lissajous curve given by the parametric equations \(x_{c}(t) = \sin (t / 3)\) and \(y_{c}(t) = \cos (t)\) , where \((x_{c}, y_{c})\) is the center of the Gaussian in space and \(t\) is a continuously valued time variable \(^{55}\) . The Lissajous trajectory was discretized to have 100 frames per cycle. The first cycle was discarded to omit the initial transient network activity, the network was trained on the subsequent 3 cycles, and closed- loop forecasting was performed on the 2 cycles subsequent to that. + +<|ref|>text<|/ref|><|det|>[[112, 388, 884, 842]]<|/det|> +In the natural video example (Fig. 4), a walking video from the Weizmann Human Action Dataset \(^{56}\) was used, in which a person walks across the scene. We present several key examples here, but note that the model successfully performs closed- loop forecasting for all movies in this dataset. Segmentation masks of the people in the videos are included with this dataset (https://www.wisdom.weizmann.ac.il/vision/SpaceTimeActions.html). Using these masks, we cropped the frames so that the person was centered throughout the entire walk, giving frames of approximately 80 by 50 pixels. Without performing this step, our network model would fail: the training data would be independent from the closed- loop forecast data since they would occupy exclusive regions of the pixel space, and the model would not generalize to the prediction data. Such nonstationary data have been successfully taught to networks with approximate translation invariance, and translation invariance is likely used in the brain to learn such processes \(^{57}\) . However, translation invariance is beyond the scope of our study. The frames were then resized to be exactly 80 by 50 pixels. Finally, each video was around 70 frames long. To get more frames without interpolation, we "bookended" each video by concatenating it with its temporal reverse sequence, where one cycle consists of the original frames followed by the bookended frames. The result is a longer video with the same spatiotemporal statistics. The first cycle was discarded to omit the initial transient network activity, the network was trained on the subsequent 3 cycles, and closed- loop prediction was performed on the 2 cycles subsequent to that. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 155]]<|/det|> +To measure the balance between feedforward input and recurrent interaction, we devised the recurrence- to- input ratio. Per Equation 1a, the input and recurrence terms are the column vectors + +<|ref|>equation<|/ref|><|det|>[[429, 183, 567, 206]]<|/det|> +\[x[t] = \{x_{i}[t]\}_{i = 1}^{N}\] + +<|ref|>text<|/ref|><|det|>[[113, 238, 149, 254]]<|/det|> +and + +<|ref|>equation<|/ref|><|det|>[[315, 283, 680, 338]]<|/det|> +\[r[t] = \left\{-\mathrm{i}\sum_{j = 1}^{N}w_{ij}\mathrm{e}^{\mathrm{i}(a_{j}[t - 1 - \tau_{ij}] - a_{i}[t - 1])}\right\}_{i = 1}^{N},\] + +<|ref|>text<|/ref|><|det|>[[113, 367, 411, 385]]<|/det|> +respectively. Further, let the matrices + +<|ref|>equation<|/ref|><|det|>[[381, 415, 617, 456]]<|/det|> +\[R = \left[r[t] r[t + 1] \dots r[t']\right]\] + +<|ref|>text<|/ref|><|det|>[[113, 489, 149, 505]]<|/det|> +and + +<|ref|>equation<|/ref|><|det|>[[377, 536, 621, 578]]<|/det|> +\[X = \left[x[t] x[t + 1] \dots x[t']\right]\] + +<|ref|>text<|/ref|><|det|>[[113, 609, 883, 660]]<|/det|> +be the horizontal concatenations of \(r[t]\) and \(x[t]\) , respectively, over closed- loop forecast times \(\{t, t + 1, \ldots , t'\}\) . The ratio is defined as + +<|ref|>equation<|/ref|><|det|>[[465, 690, 530, 730]]<|/det|> +\[\frac{\|R\|_F}{\|X\|_F},\] + +<|ref|>text<|/ref|><|det|>[[113, 764, 883, 810]]<|/det|> +where \(\| G\| _F\) denotes the Frobenius matrix norm of a matrix G, which is equivalent to the Euclidean vector norm of the vectorization of G. + +<|ref|>text<|/ref|><|det|>[[114, 840, 245, 857]]<|/det|> +Movie switching + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 884, 413]]<|/det|> +The network was trained on two movie inputs: one of a walking person (movie 1) and one of a jumping person (movie 2), both from the Weizmann dataset. The same recurrent matrix was used in each case- only the learned matrices \((V_{1}\) and \(V_{2}\) , respectively) differed. Let \(\mathcal{V} = c V_{1} + (1 - c) V_{2}\) , where \(c \in [0,1]\) . \(\mathcal{V}\) stores both learned matrices, and the present input modulates the relative contribution of \(V_{1}\) and \(V_{2}\) using an update rule for \(c\) . The structural similarity between the input and output were computed at each time step t ( \(\mathrm{SSIM}[t]\) ), and the change thereof was computed at each time step as \(\Delta \mathrm{SSIM} = \mathrm{SSIM}[t] - \mathrm{SSIM}[t - 1]\) . The update rule is \(c \coloneqq c + \Delta c\) , where \(\Delta c = - \eta \operatorname {sgn}[\Delta \mathrm{SSIM}]\) and \(\eta\) is the learning rate, set to 0.1. Depending on which movie (movie 1 or movie 2) drives the network, \(c\) tends toward 1 or 0, respectively. Once this happens, this driving input is removed and closed- loop forecasting commences as described. Switching entails instantaneously transitioning from closed- loop forecasting of one movie to driving the network with the frames of another movie. \(c\) then updates as described and is followed by closed- loop forecasting again. + +<|ref|>sub_title<|/ref|><|det|>[[115, 442, 302, 459]]<|/det|> +## Parameter optimization + +<|ref|>text<|/ref|><|det|>[[113, 488, 884, 605]]<|/det|> +The random- search algorithm was used to optimize parameters for closed- loop forecasting. Within specified bounds, each parameter was randomly sampled, giving a point in the parameter space. The parameter space was randomly sampled in this way many times, and each time, the structural similarity index was computed as the performance index. The bounds within which the parameters were sampled are given in Table I. + +<|ref|>table<|/ref|><|det|>[[115, 631, 882, 742]]<|/det|> + +
parametersampled interval
recurrent strength(0, 0.2)
recurrent length(0, 0.2)
input strength(0, 0.2)
v(0, 0.1)
+ +<|ref|>table_footnote<|/ref|><|det|>[[115, 743, 850, 761]]<|/det|> +Table I: Intervals over which model parameters were randomly searched during optimization. + +<|ref|>text<|/ref|><|det|>[[113, 814, 884, 905]]<|/det|> +Acknowledgements: This work was supported by the Canadian Institute for Health Research and NSF (NeuroNex Grant No. 2015276), BrainsCAN at Western University through the Canada First Research Excellence Fund (CFREF), Gatsby Charitable Foundation, the Fiona and Sanjay Jha Chair in Neuroscience, the Swartz Foundation, Compute Ontario (computeontario.ca), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 252]]<|/det|> +Compute Canada (computecanada.ca), SPIRITS 2020 of Kyoto University, and the Western Academy for Advanced Research. R.C.B. gratefully acknowledges the Western Institute for Neuroscience Clinical Research Postdoctoral Fellowship. G.B.B. gratefully acknowledges the Canadian Open Neuroscience Platform (Graduate Scholarship), the Vector Institute (Postgraduate Affiliate), and the National Sciences and Engineering Research Council of Canada (Canada Graduate Scholarship - Doctoral). The authors thank Alex Busch for her help with illustrations. + +<|ref|>text<|/ref|><|det|>[[113, 280, 883, 323]]<|/det|> +Data and code availability: All data and code associated with this work are available at https://github.com/mullerlab. + +<|ref|>text<|/ref|><|det|>[[113, 351, 883, 443]]<|/det|> +Contributions: Conceptualization: G.B.B., L.M.; data curation: G.B.B.; formal analysis: G.B.B., L.M.; funding acquisition: J.R., L.M.; investigation: G.B.B., L.M.; methodology: G.B.B., R.C.B., Z.D., J.R., L.M.; supervision: J.R., L.M.; visualization: G.B.B.; writing- original draft: G.B.B., L.M.; writing- review and editing: G.B.B., R.C.B., Z.D., J.R., L.M. + +<|ref|>sub_title<|/ref|><|det|>[[114, 472, 214, 490]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[112, 506, 876, 905]]<|/det|> +1. Markov, N. T. et al. Weight consistency specifies regularities of macaque cortical networks. Cereb. Cortex 21, 1254–1272 (2011). +2. Bruno, R. M. & Sakmann, B. Cortex is driven by weak but synchronously active thalamocortical synapses. Science 312, 1622–1627 (2006). +3. Swindale, N. V. Visual map. Scholarpedia J. 3, 4607 (2008). +4. Hubel, D. H. & Wiesel, T. N. Receptive fields of single neurones in the cat's striate cortex. J. Physiol. 148, 574–591 (1959). +5. Riesenhuber, M. & Poggio, T. Hierarchical models of object recognition in cortex. Nat. Neurosci. 2, 1019–1025 (1999). +6. Angelucci, A. et al. Circuits for local and global signal integration in primary visual cortex. J. Neurosci. 22, 8633–8646 (2002). +7. Douglas, R. J., Koch, C., Mahowald, M., Martin, K. A. & Suarez, H. H. Recurrent excitation in neocortical circuits. Science 269, 981–985 (1995). +8. Sompolinsky, H. & Shapley, R. New perspectives on the mechanisms for orientation selectivity. Curr. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 88, 412, 105]]<|/det|> +Opin. Neurobiol. 7, 514–522 (1997). + +<|ref|>text<|/ref|><|det|>[[113, 116, 883, 164]]<|/det|> +9. Blakemore, C. & Tobin, E. A. Lateral inhibition between orientation detectors in the cat's visual cortex. Exp. Brain Res. 15, 439–440 (1972). + +<|ref|>text<|/ref|><|det|>[[115, 174, 880, 252]]<|/det|> +10. Allman, J., Miezin, F. & McGuinness, E. Stimulus specific responses from beyond the classical receptive field: neurophysiological mechanisms for local-global comparisons in visual neurons. Annu. Rev. Neurosci. 8, 407–430 (1985). + +<|ref|>text<|/ref|><|det|>[[115, 262, 875, 310]]<|/det|> +11. Field, D. J., Hayes, A. & Hess, R. F. Contour integration by the human visual system: Evidence for a local 'association field'. Vision Res. 33, 173–193 (1993). + +<|ref|>text<|/ref|><|det|>[[115, 320, 670, 339]]<|/det|> +12. Gilbert, C. D. Adult cortical dynamics. Physiol. Rev. 78, 467–485 (1998). + +<|ref|>text<|/ref|><|det|>[[115, 349, 868, 397]]<|/det|> +13. Albright, T. D. & Stoner, G. R. Contextual influences on visual processing. Annu. Rev. Neurosci. 25, 339–379 (2002). + +<|ref|>text<|/ref|><|det|>[[115, 407, 880, 455]]<|/det|> +14. Kietzmann, T. C. et al. Recurrence is required to capture the representational dynamics of the human visual system. Proc. Natl. Acad. Sci. U. S. A. 116, 21854–21863 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 465, 838, 541]]<|/det|> +15. Kar, K., Kubilius, J., Schmidt, K., Issa, E. B. & DiCarlo, J. J. Evidence that recurrent circuits are critical to the ventral stream's execution of core object recognition behavior. Nat. Neurosci. 22, 974–983 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 552, 860, 600]]<|/det|> +16. Bringuier, V., Chavane, F., Glaeser, L. & Frégnac, Y. Horizontal propagation of visual activity in the synaptic integration field of area 17 neurons. Science 283, 695–699 (1999). + +<|ref|>text<|/ref|><|det|>[[115, 610, 864, 658]]<|/det|> +17. Roland, P. E. et al. Cortical feedback depolarization waves: a mechanism of top-down influence on early visual areas. Proc. Natl. Acad. Sci. U. S. A. 103, 12586–12591 (2006). + +<|ref|>text<|/ref|><|det|>[[115, 668, 860, 716]]<|/det|> +18. Xu, W., Huang, X., Takagaki, K. & Wu, J.-Y. Compression and reflection of visually evoked cortical waves. Neuron 55, 119–129 (2007). + +<|ref|>text<|/ref|><|det|>[[115, 726, 828, 774]]<|/det|> +19. Nauhaus, I., Busse, L., Carandini, M. & Ringach, D. L. Stimulus contrast modulates functional connectivity in visual cortex. Nat. Neurosci. 12, 70–76 (2009). + +<|ref|>text<|/ref|><|det|>[[115, 784, 841, 832]]<|/det|> +20. Reimer, A., Hubka, P., Engel, A. K. & Kral, A. Fast propagating waves within the rodent auditory cortex. Cereb. Cortex 21, 166–177 (2011). + +<|ref|>text<|/ref|><|det|>[[115, 842, 872, 890]]<|/det|> +21. Townsend, R. G. et al. Emergence of complex wave patterns in primate cerebral cortex. J. Neurosci. 35, 4657–4662 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 866, 135]]<|/det|> +22. Slovin, H., Arieli, A., Hildesheim, R. & Grinvald, A. Long-term voltage-sensitive dye imaging reveals cortical dynamics in behaving monkeys. J. Neurophysiol. 88, 3421–3438 (2002). + +<|ref|>text<|/ref|><|det|>[[113, 146, 832, 193]]<|/det|> +23. Sato, T. K., Nauhaus, I. & Carandini, M. Traveling waves in visual cortex. Neuron 75, 218–229 (2012). + +<|ref|>text<|/ref|><|det|>[[113, 204, 867, 251]]<|/det|> +24. Davis, Z. W., Muller, L., Martinez-Trujillo, J., Sejnowski, T. & Reynolds, J. H. Spontaneous travelling cortical waves gate perception in behaving primates. Nature 587, 432–436 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 262, 853, 309]]<|/det|> +25. Muller, L., Reynaud, A., Chavane, F. & Destexhe, A. The stimulus-evoked population response in visual cortex of awake monkey is a propagating wave. Nat. Commun. 5, 3675 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 320, 870, 368]]<|/det|> +26. Davis, Z. et al. Spontaneous traveling waves naturally emerge from horizontal fiber time delays and travel through locally asynchronous-irregular states. Nature Communications (2021). + +<|ref|>text<|/ref|><|det|>[[113, 379, 874, 426]]<|/det|> +27. Takahashi, K. et al. Large-scale spatiotemporal spike patterning consistent with wave propagation in motor cortex. Nat. Commun. 6, 7169 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 437, 618, 455]]<|/det|> +28. Trabelsi, C. et al. Deep Complex Networks. arXiv [cs.NE] (2017). + +<|ref|>text<|/ref|><|det|>[[113, 466, 879, 541]]<|/det|> +29. Heeger, D. J. & Mackey, W. E. Oscillatory recurrent gated neural integrator circuits (ORGaNICs), a unifying theoretical framework for neural dynamics. Proc. Natl. Acad. Sci. U. S. A. 116, 22783–22794 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 553, 861, 600]]<|/det|> +30. Krizhevsky, A., Sutskever, I. & Hinton, G. E. Imagenet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 25, (2012). + +<|ref|>text<|/ref|><|det|>[[113, 611, 780, 657]]<|/det|> +31. Graves, A. Supervised Sequence Labelling with Recurrent Neural Networks. Studies in Computational Intelligence (2012) doi:10.1007/978-3-642-24797-2. + +<|ref|>text<|/ref|><|det|>[[113, 669, 864, 715]]<|/det|> +32. Hwang, K. & Sung, W. Online Sequence Training of Recurrent Neural Networks with Connectionist Temporal Classification. arXiv [cs.LG] (2015). + +<|ref|>text<|/ref|><|det|>[[113, 727, 832, 773]]<|/det|> +33. Budzinski, R. C. et al. Geometry unites synchrony, chimeras, and waves in nonlinear oscillator networks. Chaos 32, 031104 (2022). + +<|ref|>text<|/ref|><|det|>[[113, 785, 861, 831]]<|/det|> +34. Muller, L., Chavane, F., Reynolds, J. & Sejnowski, T. J. Cortical travelling waves: mechanisms and computational principles. Nat. Rev. Neurosci. 19, 255–268 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 843, 866, 890]]<|/det|> +35. Hellwig, B. A quantitative analysis of the local connectivity between pyramidal neurons in layers 2/3 of the rat visual cortex. Biol. Cybern. 82, 111–121 (2000). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 870, 135]]<|/det|> +36. Binzegger, T., Douglas, R. J. & Martin, K. A. C. A quantitative map of the circuit of cat primary visual cortex. J. Neurosci. 24, 8441–8453 (2004). + +<|ref|>text<|/ref|><|det|>[[112, 145, 881, 193]]<|/det|> +37. Girard, P., Hupé, J. M. & Bullier, J. Feedforward and feedback connections between areas V1 and V2 of the monkey have similar rapid conduction velocities. J. Neurophysiol. 85, 1328–1331 (2001). + +<|ref|>text<|/ref|><|det|>[[112, 203, 850, 251]]<|/det|> +38. Jaeger, H. & Haas, H. Harnessing Nonlinearity: Predicting Chaotic Systemsand Saving Energy in Wireless Communication. Science 304, (2004). + +<|ref|>text<|/ref|><|det|>[[112, 261, 876, 309]]<|/det|> +39. Pathak, J., Hunt, B., Girvan, M., Lu, Z. & Ott, E. Model-Free Prediction of Large Spatiotemporally Chaotic Systems from Data: A Reservoir Computing Approach. Phys. Rev. Lett. 120, 024102 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 319, 870, 367]]<|/det|> +40. Mathieu, M., Couprie, C. & LeCun, Y. Deep multi-scale video prediction beyond mean square error. arXiv [cs.LG] (2015). + +<|ref|>text<|/ref|><|det|>[[112, 377, 830, 425]]<|/det|> +41. Wang, Z., Bovik, A. C., Sheikh, H. R. & Simoncelli, E. P. Image quality assessment: from error visibility to structural similarity. IEEE Trans. Image Process. 13, 600–612 (2004). + +<|ref|>text<|/ref|><|det|>[[112, 435, 875, 483]]<|/det|> +42. Stettler, D. D., Das, A., Bennett, J. & Gilbert, C. D. Lateral connectivity and contextual interactions in macaque primary visual cortex. Neuron 36, 739–750 (2002). + +<|ref|>text<|/ref|><|det|>[[112, 493, 875, 541]]<|/det|> +43. Moshe, B., Lena, G., Eli, S., Michal, I. & Ronen, B. Actions as space-time shapes. in Proceedings of the Tenth IEEE International Conference on Computer Vision, Beijing, China 17–20 (2005). + +<|ref|>text<|/ref|><|det|>[[112, 551, 866, 600]]<|/det|> +44. Berry, M. J., Brivanlou, I. H., Jordan, T. A. & Meister, M. Anticipation of moving stimuli by the retina. Nature vol. 398 334–338 (1999). + +<|ref|>text<|/ref|><|det|>[[112, 610, 840, 657]]<|/det|> +45. Benvenuti, G. et al. Anticipatory responses along motion trajectories in awake monkey area V1. 2020.03.26.010017 (2020) doi:10.1101/2020.03.26.010017. + +<|ref|>text<|/ref|><|det|>[[112, 668, 879, 687]]<|/det|> +46. Hess, R. F. & Dakin, S. C. Contour integration in the peripheral field. Vision Res. 39, 947–959 (1999). + +<|ref|>text<|/ref|><|det|>[[112, 697, 866, 715]]<|/det|> +47. Olshausen, B. A. How Close Are We to Understanding V1? Neural Comput. 17, 1665–1699 (2005). + +<|ref|>text<|/ref|><|det|>[[112, 726, 833, 774]]<|/det|> +48. Schrimpf, M. et al. Brain-Score: Which Artificial Neural Network for Object Recognition is most Brain-Like? bioRxiv 407007 (2020) doi:10.1101/407007. + +<|ref|>text<|/ref|><|det|>[[112, 784, 850, 832]]<|/det|> +49. Cadieu, C. F. & Olshausen, B. A. Learning intermediate-level representations of form and motion from natural movies. Neural Comput. 24, 827–866 (2012). + +<|ref|>text<|/ref|><|det|>[[112, 842, 856, 861]]<|/det|> +50. Bassey, J., Qian, L. & Li, X. A Survey of Complex-Valued Neural Networks. arXiv [stat.ML] (2021). + +<|ref|>text<|/ref|><|det|>[[112, 871, 848, 890]]<|/det|> +51. Acebrón, J. A., Bonilla, L. L., Pérez Vicente, C. J., Ritort, F. & Spigler, R. The Kuramoto model: A + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 783, 107]]<|/det|> +simple paradigm for synchronization phenomena. Rev. Mod. Phys. 77, 137–185 (2005). + +<|ref|>text<|/ref|><|det|>[[113, 117, 852, 165]]<|/det|> +52. Rodrigues, F. A., Peron, T. K. D. M., Ji, P. & Kurths, J. The Kuramoto model in complex networks. Phys. Rep. 610, 1–98 (2016). + +<|ref|>text<|/ref|><|det|>[[113, 175, 808, 223]]<|/det|> +53. Breakspear, M., Heitmann, S. & Daffertshofer, A. Generative models of cortical oscillations: neurobiological implications of the kuramoto model. Front. Hum. Neurosci. 4, 190 (2010). + +<|ref|>text<|/ref|><|det|>[[113, 233, 855, 281]]<|/det|> +54. Muller, L., Mináč, J. & Nguyen, T. T. Algebraic approach to the Kuramoto model. Phys Rev E 104, L022201 (2021). + +<|ref|>text<|/ref|><|det|>[[113, 291, 857, 339]]<|/det|> +55. Heim, N. & Avery, J. E. Adaptive Anomaly Detection in Chaotic Time Series with a Spatially Aware Echo State Network. arXiv [cs.NE] (2019). + +<|ref|>text<|/ref|><|det|>[[113, 349, 848, 397]]<|/det|> +56. Gorelick, L., Blank, M., Shechtman, E., Irani, M. & Basri, R. Actions as space-time shapes. IEEE Trans. Pattern Anal. Mach. Intell. 29, 2247–2253 (2007). + +<|ref|>text<|/ref|><|det|>[[113, 407, 830, 455]]<|/det|> +57. Anselmi, F. et al. Unsupervised learning of invariant representations. Theor. Comput. Sci. 633, 112–121 (2016). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 131, 258, 258]]<|/det|> +- wc1 supplement.pdf- movies1.mp4- movies2.mp4- movies3.mp4- movies4.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/images_list.json b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d90e65999bd81867824d12211ceac7ccc74232cd --- /dev/null +++ b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1: Map of the proxy network. Black Xs indicate the centroid of binned drought atlas sites. Grey markers indicate PAGES2k sites. The size of the PAGES2k markers correspond to the length of each record. Filled color contours show the field correlation between the SAM index and DJF sea level pressure from 20CR [88, 89] over the period 1958-2000 CE.", + "footnote": [], + "bbox": [ + [ + 245, + 250, + 748, + 635 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: Evolution of the reconstruction over time. Top: Comparison of the annual reconstruction (blue) with the Marshall index (red) over the instrumental era. Shading indicates the 5-95 percentiles of the reconstruction. Middle: Evolution of the annual reconstruction (blue) and 31-year lowpass filtered (black) over the Common Era. Shading indicates the 5-95 percentiles of the lowpass filtered series. Bottom: Composition of the proxy network over time. Colors for proxy types are as follows: Dark blue (coral), pink (ANZDA), grey (SADA), dark red (trees), light blue (glacier ice), black (lake sediment).", + "footnote": [], + "bbox": [ + [ + 188, + 189, + 808, + 640 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: Optimal Sensor Analysis. Top: Maps of the potential ability for drought atlas (left) and PAGES2k (right) sites to constrain reconstruction posterior variance. Middle: Evolution of reconstruction posterior variance over time. Yellow bars indicate the addition of the indicated proxy to the network. Bottom: Ranked histograms of the seven sites with greatest potential influence in the early reconstruction (left), immediately before the addition of drought atlases (middle) and for the full network (right). Potential influence is determined as the uncertainty constrained by a single-proxy network.", + "footnote": [], + "bbox": [ + [ + 191, + 145, + 805, + 680 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4: SAM climate responses. Top left: Wavelet coherence of the reconstructed SAM index with the solar forcing reconstruction [112]. Bottom left: Composite mean response to major volcanic eruptions. Shading indicates 5-95 percentiles. Blue line is the ensemble mean. Right: Instrumental SAM trends for the Marshall Index (top right) and reconstruction (bottom right). Colored points indicate trends calculated from a sliding window centered on the given year. Solid (dotted) contours surround statistically significant trends at the \\(90\\%\\) confidence interval relative to the reconstruction over the period 1500-1900 CE (1-1900 CE).", + "footnote": [], + "bbox": [ + [ + 120, + 222, + 876, + 612 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5: Comparison of SAM reconstructions over the last millennium. All reconstructions are smoothed via a 30-year Gaussian filter and normalized to the period 1400-1850 CE.", + "footnote": [], + "bbox": [ + [ + 188, + 358, + 808, + 558 + ] + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192.mmd b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192.mmd new file mode 100644 index 0000000000000000000000000000000000000000..64eb51211c382f31c4186d64377768a8e6aa9ce5 --- /dev/null +++ b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192.mmd @@ -0,0 +1,569 @@ + +# Trends and variability in the Southern Annular Mode over the Common Era + +Jonathan King ( jonking93@email.arizona.edu ) + +University of Arizona + +Kevin Anchukaitis The University of Arizona + +Kathryn Allen University of Tasmania + +Tessa Vance University of Tasmania + +Amy Hessl West Virginia University https://orcid.org/0000- 0002- 2279- 6260 + +Article + +Keywords: + +Posted Date: August 17th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1958191/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Trends and variability in the Southern Annular Mode over the Common Era + +Jonathan King \(^{1,2*}\) , Kevin J. Anchukaitis \(^{1,2,3}\) , Kathryn Allen \(^{4,5,6}\) , Tessa Vance \(^{7}\) and Amy Hessl \(^{8}\) + +\(^{1}\) Department of Geosciences, University of Arizona, Tucson, AZ 85721 USA. \(^{2}\) Laboratory of Tree- Ring Research, University of Arizona, Tucson, AZ 85721 USA. \(^{3}\) School of Geography, Development, and Environment, University of Arizona, Tucson, AZ 85721 USA. \(^{4}\) School of Geography, Planning and Spatial Sciences, University of Tasmania, Hobart 7001. \(^{5}\) School of Ecosystem and Forest Sciences, University of Melbourne, Richmond, VIC Australia 3121. \(^{6}\) Centre of Excellence for Australian Biodiversity and Heritage, University of New South Wales, Australia. \(^{7}\) Australian Antarctic Program Partnership, Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, Australia. \(^{8}\) Department of Geology and Geography, West Virginia University, Morgantown, WV USA. + +## Abstract + +The Southern Annular Mode (SAM) is the leading mode of atmospheric variability in the extratropical Southern Hemisphere and has wide ranging effects on ecosystems and societies. Despite the SAM's importance, paleoclimate reconstructions disagree on its variability and trends. Here, we use data assimilation to reconstruct the SAM over the last 2000 years using temperature and drought- sensitive climate proxies. Our method does not assume a stationary relationship between proxy records and the SAM over an instrumental calibration period, so our reconstruction is less sensitive to the teleconnection variability that has hindered previous reconstructions. Our approach also allows us to identify critical paleoclimate records and quantify reconstruction + +<--- Page Split ---> + +uncertainty through time. We find no evidence for a forced response in SAM variability prior to the 20th century. We also find the modern positive trend is outside the range of the prior 2000 years, but only on multidecadal time scales. + +## Introduction + +The Southern Annular Mode (SAM) is the leading mode of atmospheric variability in the extratropical Southern Hemisphere and is characterized by a mostly zonally- symmetric mass oscillation with anti- correlated pressure anomalies over the mid- latitudes and Antarctica [1- 4, and see Figure 1]. The SAM's phases capture the strength and position of the mid- latitude westerly winds and the subtropical jet, such that positive phases promote a poleward shift of storm tracks and intensification of the circumpolar westerly belt, while negative phases promote an equator- ward shift of storm tracks and weakening of the westerly winds. Variability in the SAM therefore has wide ranging effects across the Southern Hemisphere. Positive phases of the SAM are linked to cooling over Australia and central Antarctica, as well as warming over the Antarctic Peninsula and southern South America [5- 11]. Hydroclimate effects of the positive phase include drying over southern South America, western South Africa, southern Australia, and New Zealand, as well as increased precipitation over central and eastern Australia and southeastern South America [7, 10, 12- 16]. SAM variations have also been linked to wildfire activity in South America and south- east Australia [17- 21], changes in sea ice distribution [8, 22- 25], and ocean- atmosphere carbon exchange [26- 28]. Understanding SAM variability is therefore important for both societies and ecosystems throughout the Southern Hemisphere, particularly in sub- tropical- temperate regions projected to experience a future drying climate. + +Since the 1950s, the SAM has exhibited a trend toward a more positive state [4, 29, 30], which is attributed to stratospheric ozone depletion and rising concentrations of atmospheric \(\mathrm{CO_2}\) [31- 35]. This positive trend has potentially contributed to severe droughts, including the Day Zero Cape Town drought [36] and Millennium Drought in Australia [37, 38], as well as increased fire activity [17- 19]. Given these impacts, it is important to place the SAM's recent behavior in a long- term perspective and assess the relative influence of anthropogenic forcing and natural climate variability. In the context of multi- decadal trends, reconstructions spanning multiple centuries are necessary to resolve forced responses from the SAM's internal variability. Instrumental records of the SAM only extend through the mid- 1900s and longer reanalysis- derived indices show low correlations with one another and differences in variability prior to the 1950s [30, 39], so characterizing the SAM's long- term behavior requires paleoclimate reconstructions derived from natural climate archives. + +There are several existing multi- century SAM reconstructions [40, 41, 42] (henceforth, V12, A14, and D18), but they show limited agreement prior to the 1850s [42, 43]. Most indicate a negative phase in the SAM during the late 1400s, but both trends and decadal- scale variability show large discrepancies aside from this feature. + +<--- Page Split ---> + +There are several potential reasons for these differences. Firstly, all three reconstructions rely on the calibration of proxy records directly with an instrumental SAM index. This implicitly makes two important assumptions for each reconstruction: first, that the relationship of proxy records to local climate variables is stationary over time; and second, that the SAM's teleconnections with local climate variables are stationary and well- represented by the instrumental record. While the first is reasonable and a necessary assumption of most paleoclimate analyses, multiple studies cast doubt on this second point, and regional complexity in the climate response to specific SAM phases further decreases the likelihood of this assumption holding. For instance, even over the instrumental period, SAM exhibits non- stationary connections with precipitation and temperature anomalies in southern South America, Australasia, and the Antarctic Peninsula [13, 44], and many of the proxy records in existing SAM reconstructions come from these areas [43]. Evolving concentrations of greenhouse gases, stratospheric ozone, connections with ENSO, and stochastic climate variability can also affect the SAM's influence on regional climates over multi- decadal time scales [29, 32, 45, 46]. Pseudo- proxy experiments have also shown that non- stationary teleconnections cause reconstruction skill to vary widely with the selection of different calibration windows [47]. This effect is particularly pronounced for proxy networks with fewer than 20 sites, which is common in the early portions of SAM reconstructions. To mitigate such effects, D18 explicitly screened for stationarity in their reconstruction, although this required calibration with a longer and therefore less reliable observational record [48]. + +Differences between SAM reconstructions may also result from the selection of different reconstruction targets and proxy networks. For example, A14 targets an annual SAM index, whereas V12 and D18 target an austral summer (DJF) SAM index. D18 found that annual reconstructions were much more sensitive to the selection of proxy sites and calibration windows and they conclude that annual products may exhibit increased sensitivity to non- stationary teleconnections, which may partly explain the differences between the reconstructions. Additionally, each index has been reconstructed using a different proxy network with a different geographic extent. A14 targets the Drake Passage sector, using a mix of terrestrial proxy types from southern South America as well as Antarctic ice cores. In comparison, V12 targets the Pacific sector, using a network of tree- ring chronologies from South America and New Zealand. D18 uses the most spatially extensive network, including tree- ring records [40], Antarctic ice cores, PAGES2k South American proxies [49], and coral records from the tropical Pacific [50]. Furthermore, A14 utilizes a temperature- sensitive proxy network, while V12 and D18 leverage both temperature and hydroclimate- sensitive proxies. Given the variability of the SAM's teleconnections on regional scales [13, 44], and the climate sensitivities of different proxy types [48], these variations in proxy- network design may further help explain reconstruction differences. It is often difficult to assess the influence and contribution of individual proxy records in multiproxy reconstructions, so the cause of any reconstructed index's behavior are often unclear. This is particularly relevant in the period prior to 1400 CE, when the sparsity of proxy networks leaves the reconstructions vulnerable to the dominant influence of just a few records. Ultimately, as a consequence + +<--- Page Split ---> + +of these uncertainties and the differences in existing reconstructions, the evolution of the SAM over the Common Era and its response to external forcing remains poorly constrained [43, 51]. + +To address these uncertainties, here we reconstruct the austral summer (DJF) SAM index over the Common Era at annual resolution using offline paleoclimate data assimilation (DA). DA is a recently developed reconstruction technique that integrates climate proxy records with the dynamical behavior captured by climate models [52, 53]. In brief, DA uses forward or proxy- system models [54, 55] to translate climate model states into the same dimensions or 'space' as a collection of climate proxy records. This allows direct comparison of the model output with the proxy records. The climate model states are then updated to more closely match the proxy records, and a model- derived estimate of climate system covariance is used to propagate the update to reconstruction targets, such as the SAM. DA has recently been used to reconstruct surface air temperature anomalies [56- 59], geopotential height fields [57], the response to volcanic eruptions [60, 61], sea ice extent [62], sea surface temperatures [63], and hydroclimate variables [64]. In this study, we assimilate the PAGES2k temperature- sensitive proxy network [49], the South American Drought Atlas [SADA; 65], and the Australia- New Zealand Drought Atlas [ANZDA; 66] (Figure 1) using a suite of last millennium general- circulation climate models (see Methods and Supplemental Table 1) to reconstruct the austral summer SAM index over the last 2,000 years. Here, we follow Gong et al. [2] and define the SAM index using the difference of zonal- mean pressure anomalies (see Methods, Equation 1). + +In the context of SAM reconstructions, DA offers several additional advantages relative to traditional methods. Firstly, our method does not calibrate proxy records against an instrumental SAM index directly; instead, we calibrate proxy forward models using local climate variables, like temperature and precipitation, near the proxy sites. Consequently, our calibration does not assume stationary SAM teleconnections and only requires the stability of proxy relationships to their local climate. Additionally, we estimate covariance between proxies and the SAM using thousands of years of climate model output. As a result of this, our proxy- SAM relationships are not sensitive to potentially anomalous decadal- or centennial- scale variations in the SAM's behavior. Furthermore, DA is amenable to the use of a range of proxy types as well as gridded climate records with spatial autocorrelation, and we leverage this to incorporate the two existing tree- ring based drought atlases into our reconstruction. Previous work indicates that SAM reconstructions using hydroclimate- sensitive sites are more skillful than those using strictly temperature- sensitive proxy networks [47]. Each drought atlas provides extensive coverage for at least the last five centuries and each incorporates over 150 tree- ring records. They therefore represent a significant source of hydroclimate information available for our reconstruction. + +Finally, our DA method allows us to incorporate an optimal sensor analysis [67] as part of the final reconstruction. Traditionally, optimal sensor analyses have been used to identify ideal regions for future proxy development [67- 70]; however, they can also be applied within a DA framework to quantitatively assess the power of different proxy sites as the overall network evolves through time. We use this to identify the proxy sites that are most likely to drive the reconstruction in each time + +<--- Page Split ---> + +step, which helps characterize the reconstruction's overall behavior. This information is particularly useful in the early part of this Common Era reconstruction, when the sparse network size can give high weights to a limited number of records. + +## Results + +We assess the skill of our SAM reconstruction relative to the Marshall [4] and Fogt indices [71, 72], two commonly used instrumental SAM indices (see Methods for further details). Before comparing time series, we first normalize the Fogt index and our reconstruction to the Marshall index, such that the mean and variance of the detrended normalized time series match those of the detrended Marshall index over the period 1958- 2000 CE. This places all series in the same unit space while preserving differences in the instrumental trend. Examining skill values (Table 1; Figure 2, upper panel), we find that the reconstruction's correlation with the Marshall index (1958- 2000 CE) is \(r = 0.72(p \ll 0.001)\) , which is comparable to that reported for A14 \((r = 0.75, p \ll 0.001))\) . With respect to the 20th century Fogt index, our reconstruction correlates at \(r = 0.65, p \ll 0.001)\) , somewhat higher than A14 \((r = 0.51, p \ll 0.001))\) . Our RMSE values with the Marshall index (1.45) are similar to, albeit slightly higher than, those reported by D18 (1.32). We emphasize that our reconstruction is not calibrated directly to the SAM index, so the agreement with the Marshall and Fogt indices is not built- in to our reconstruction method and thus represents a more independent skill metric. + +We next characterize the reconstruction's behavior over the last two millennia (Figure 2, center). The reconstruction exhibits minimal evidence for trends over most of the first millennium of the Common Era, although the third and seventh centuries are both marked by increased multidecadal variability as the SAM alternates between negative and positive phases. A more strongly negative anomaly in the early 1000s is followed by a notable 100- year positive trend that concludes with the most positive anomalies outside of the instrumental era. The SAM persists in a positive state until the late 1400s, when it abruptly decreases to strongly negative values. After this event, the index returns to near- zero mean anomalies. It has a peak in the mid- 1700s and begins exhibiting a positive trend in the early 1800s. This trend intensifies in the later half of the 20th century, and the reconstruction ends with the most positive SAM anomalies observed during the Common Era. + +Reconstruction uncertainty ranges from \(\pm 4.5\) anomaly units in the early reconstruction to less than 2.3 after 1500 CE (Figure 2b). We note that, because we use a stationary prior, the reconstruction years are treated as fully independent of one another. While this is common in many reconstruction techniques, it does not represent the reality of the SAM, which exhibits persistence on interannual time scales due to potential connections with the stratosphere [73, 74], tropical variability [75, 76], and external forcing [35, 77]. One consequence of this is that the uncertainty estimates shown here likely overestimate the true reconstruction uncertainty. Overall, uncertainty decreases as the reconstruction approaches the present day, a result of the increasing size of the proxy network (Figure 2, bottom panel). + +<--- Page Split ---> + +We use our optimal sensor framework to identify which proxies are most responsible for reducing reconstruction uncertainty over time (Figure 3). A proxy's ability to reduce uncertainty corresponds to its influence on the reconstruction, so this analysis also allows us to identify which proxies most strongly influence the reconstruction at a given point in time. The first 900 years of the reconstruction are most strongly affected by the Mt. Read (Tasmania) tree- ring record with additional support from the Plateau Remote, WDC06A, and WDC05A ice cores. At 900 CE, the Oroko (New Zealand) tree ring chronology joins the network and supplants Mt. Read as the most influential record. Two large decreases in reconstruction uncertainty occur in 1400 and 1500 CE, which correspond to the addition of the SADA and ANZDA, respectively. + +Examining the reconstruction's response to external forcing, we find no coherence with the solar forcing series and no significant common response to major volcanic eruptions (Figure 4). By contrast, the reconstruction exhibits significant positive trends in the latter half of the twentieth century. However, these modern trends are only significant on time- scales greater than approximately 40 years; trends over shorter time scales fall within the reconstructed range of trends from natural variability. Here, we define natural variability using the distribution of trends in periods of the pre- industrial reconstruction. Examining the Marshall Index, we similarly find that trends shorter than about 35 years are within the reconstructed range of natural variability, but that trends longer than about 35 years fall outside this range. The Marshall Index exhibits its most positive, significant trends for intervals centered on the early 1980s. Although this period is near the end of our reconstruction and less well resolved than preceding decades, we note that the reconstruction similarly exhibits strongly- positive, significant trends centered on the early 1980s. Here we have quantified natural variability using the distribution of reconstructed trends over the period 1500- 1900 CE, the years including both drought atlases. If we instead use the period of the full reconstruction (1- 1900 CE), the tests become more stringent. Significant trend in the reconstruction is limited to the last 55- 80 year interval, and Marshall Index trends are only significant when containing the interval 1964- 2000 CE. We also experiment with using the early portion of the reconstruction (1- 899 CE) to quantify natural variability and find these results are similar to those using the full reconstruction period (Supplemental Figure 1). + +## Discussion + +Our reconstruction suggests that the SAM is dominated by internal variability at least throughout the pre- industrial Common Era. This finding is in agreement with D18, who likewise found minimal influence of solar and volcanic forcing on their reconstruction. Volcanic signals have likewise been a challenge to detect in Southern Hemisphere temperature reconstructions [78]. Some studies have proposed that that an observed relationship between SAM and ENSO [41, 76, 79- 81] could provide a pathway for solar forcing [82] to influence the SAM [43, 83]; however, our results do not support this mechanism during the the Common Era. In a set of model simulations, Wright et al. [83] found that increasing the amplitude of the prescribed + +<--- Page Split ---> + +solar variability lead to a significant relationship between solar forcing and the simulated SAM. These authors suggest that using high amplitude solar forcing could help reconcile SAM reconstructions with climate simulations; however, the lack of solar signals in our reconstruction differs notably from their findings and instead further supports the realism of low- amplitude solar forcing scenarios [84–86]. + +By contrast with solar and volcanic forcings, our analysis indicates that the most recent multi- decadal trend is outside the range of natural variability and reflects the SAM's response to anthropogenic forcing. We emphasize that this modern trend is only significant for intervals longer than about 40 years when assessed against the 1500- 1900 CE period, or intervals of about 55 years when considering the full Common Era. Shorter trend periods remain within the range of natural variability, even for the most recent intervals. The significance of the modern positive trend therefore reflects its anomalous persistence, rather than the amplitude of its decadal- scale variation alone. The significance of these longer trends emphasizes the importance of the paleoclimate record, particularly given the uncertainties in instrumental SAM records prior to the late twentieth century [39, 87]. We also note that the modern positive trend is only outside of the range of natural variability for trends spanning the years from about 1940- 2000 CE. Trends are generally not significant during the early 1900s, and are even negative for the 50 year period centered on the 1930s. These results help establish the onset of the modern positive trend at around 1940 CE. This timing coincides with increasing emissions of ozone- depleting substances and greenhouse gasses, and is consistent with literature attributing the modern trend to stratospheric ozone depletion and rising levels of atmospheric \(\mathrm{CO_2}\) [31–35]. + +We next compare our reconstruction with the V12, A14, and D18 products (Figure 5). We normalize the mean and variance of each index over the period 1400- 1850 CE to allow comparison of the series in the same unit space. We select the year 1400 CE because it is the first year with values for all four reconstructions and we end the normalization in 1850 CE to limit the sensitivity of our comparison to differing representations of the post- industrial trend. All four indices agree on the existence of a strong positive trend during the late twentieth century; however, all show limited coherence with one another prior to about 1850 CE, as noted in previous studies [42, 43]. The limited agreement of these reconstructions reduces confidence in the significance of modern trends [51], and the causes of these discrepancies include differing seasonal expressions, different proxy networks, and the relative weights of proxies within those networks. Additionally, V12, A14, and D18 all rely on calibration with the instrumental SAM index, which can cause uncertainty when there is non- stationarity in the teleconnection of local climate with the SAM. Ultimately, our reconstruction does not solve the problem of differing reconstructions and similarly shows limited agreement with all of V12, A14, and D18. However, our assimilation does not rely on calibration with the SAM index, and offers a potential improvement by reducing uncertainty from non- stationary teleconnections. + +An additional advantage of our reconstruction is the transparency provided by the optimal sensor's assessment of the relative weights and influence of proxy records in our network. In general, we find that our reconstruction is most strongly influenced by the two drought atlases, followed by the Mt. Read (Tasmania), Oroko (New + +<--- Page Split ---> + +Zealand), and Pink Pine (New Zealand) tree ring chronologies, and also the Plateau Remote, Siple Station, WDC06A, and WDC06B ice cores. We note here that a minor change in reconstruction uncertainty does not imply that a proxy has a weak effect on the reconstruction, because highly influential proxies from the same location may present redundant climate signals. For example, the Pink Pine chronology is the third most potentially influential PAGES2k record (Figure 3, lower right), but has a relatively small effect on reconstruction uncertainty when added to the network in 1457 CE (Figure 3, center). This is because much of the Pink Pine climate signal is already represented by the nearby Oroko site. However, such redundant sites are valuable because they make the reconstruction less sensitive to non- climatic noise from a single highly- influential proxy record. In the case of Pink Pine and Oroko, spreading the southern New Zealand climate signal over two influential records allows either site to partially correct for non- climatic noise in the other. A proxy's potential influence reflects both its covariance with the SAM and the ability of our proxy estimates to accurately estimate the record. Ultimately, assuming our estimates of climate covariance are accurate, the influential sites are those most likely to contribute skill to the reconstruction. + +Overall, we find that tree- ring chronologies from Tasmania and New Zealand, the West Antarctic ice cores, and the drought atlas locations in Tasmania, southern New Zealand, the eastern edge of Australia, and southeast South America all have the greatest potential for reconstructing SAM (Figure 3, upper panels). This suggests that additional proxy development in these regions, or extensions of shorter existing records such as the Oroko and Pink Pine tree- ring chronologies or the Siple Station ice core, would be valuable for improving the skill of future SAM reconstructions. However, we caution that location alone is not sufficient for proxy utility and that future proxy development must demonstrate a robust sensitivity to local climate that connects them to the SAM. We also note that, in our optimal sensor framework, a proxy's potential influence is a function of (1) the accuracy of our forward (proxy system) models, and (2) the covariance of the resulting proxy estimates with the SAM in the climate models. As a result, our analysis may currently undervalue proxies from regions with limited climate model agreement, and future improvements in both climate and proxy system models may allow paleoclimate data from other regions to contribute to skillful reconstructions of the SAM. + +## Conclusions + +Our study provides the first reconstruction of the Southern Annular Mode at annual resolution over the entire Common Era. We use a data assimilation method that does not calibrate the proxies directly against the instrumental SAM index, so the reconstruction is not sensitive to observed SAM non- stationarity in the modern era. Our reconstruction leverages both the SADA and ANZDA in addition to the PAGES2k proxy network and represents a significant increase in paleoclimate information available to reconstruct the SAM. Optimal sensor analysis indicates that the first 1400 years of the reconstruction are strongly influenced by the Oroko and Mt. Read tree- ring chronologies, with additional support from the Plateau Remote, WDC06A, and + +<--- Page Split ---> + +WDC05A ice cores. As the SADA and ANZDA are added to the proxy network (1400 CE and 1500 CE, respectively), the drought atlases become strong drivers of the reconstruction's behavior. + +Our reconstruction provides a foundation with which to assess the drivers of the SAM's behavior over the Common Era; such assessments are critical given the SAM's importance to societies and effects on climate variability throughout the Southern Hemisphere. Although our index and existing SAM reconstructions show limited agreement with one another, all products exhibit the most strongly positive and persistent SAM trend during the last several decades. We find that the modern positive trend in the SAM is outside the range of natural variability over the previous millennium, further confirming a response to anthropogenic forcing. Prior to the most recent decades, we find no relationship between SAM variability and external climate forcing, suggesting that its behavior is dominated by internal variability over the pre- industrial Common Era. + +<--- Page Split ---> + +## Methods and Materials + +## Southern Annular Mode Index + +In this study, we use the Gong et al. [2] definition of the SAM index: + +\[\mathrm{SAM} = P_{40^{\circ}\mathrm{S}}^{*} - P_{65^{\circ}\mathrm{S}}^{*} \quad (1)\] + +where \(P_{X}^{*}\) indicates the normalized zonal- mean sea level pressure (SLP) at a particular latitude. The latitudes \(40^{\circ}\mathrm{S}\) and \(65^{\circ}\mathrm{S}\) were selected as the zonal- means with the most strongly anti- correlated SLP anomalies across the mid- and high- latitude Southern Hemisphere. We use this definition, as opposed to an index derived from a principal component analysis because the latitudes of the most strongly anticorrelated SLP anomalies are robust across the climate models considered in our assimilation (Supplemental Tables 1, 2). We target the austral summer (DJF; December - February) SAM because this corresponds to the seasonality of the climate response of the majority of our proxy network. D18 also suggests that summer SAM reconstructions are more robust to proxy network design than annual reconstructions, which further supports this choice. When calculating the SAM index, we normalize seasonal mean values, rather than individual months. Austral summers span months from two calendar years, and this can introduce date ambiguities for annual records, particularly tree- ring chronologies. Throughout this paper, we use the convention that the year of an austral summer value matches the calendar year of the associated January. + +## Data Products + +## Reanalysis and Instrumental Indices + +We use monthly precipitation and air- temperature fields from the Twentieth Century Reanalysis V3 [20CR; 88, 89] to calibrate our DA method. The 20CR is based on an 80- member ensemble Kalman Filter, and extends from 1850 CE to present at 2 degree resolution. Because of its role in our assimilation method, this effectively sets an upper bound on the resolution of any gridded spatial product used in this reconstruction. We also use the austral summer Marshall Index[4] and Fogt Index [71, 72] to assess the skill of our reconstruction in the modern era. The Marshall Index estimates the Gong et al. [2] definition of the SAM (Equation 1), and is based on data from 12 weather stations (6 near \(40^{\circ}\mathrm{S}\) , and 6 near \(65^{\circ}\mathrm{S}\) ). Because it uses station data, the Marshall index is not subject to the spurious trends observed in high- latitude Southern Hemisphere reanalysis pressure fields [4]. The Fogt index is constructed using a principal component regression of station pressure data and calibrated to the Marshall index. These indices are commonly used as a comparison point for SAM reconstructions[40- 42], + +## Climate Proxies + +In this reconstruction, we assimilate the PAGES2k temperature- sensitive proxy network [49], the South American Drought Atlas (SADA) [65], and the Australia- New + +<--- Page Split ---> + +Zealand Drought Atlas (ANZDA) [66]. We limit all three datasets to those sites or locations south of \(25^{\circ}\mathrm{S}\) . Pseudo- proxy tests of other latitude bounds suggests that reconstruction skill is minimally affected by the use of more northward proxy sites and agreement with the instrumental record exhibits a slight maximum for a bound at \(25^{\circ}\mathrm{S}\) (Supplemental Figure 2). Overall, this domain maximizes the number of SAM- sensitive proxy sites in our network, while minimizing the effects of distal proxies that primarily reflect other climate signals. + +From the PAGES2k dataset, we include all sites from the PAGES2k global temperature reconstruction that have annual or sub- annual temporal resolution. To maintain a common timescale, we bin all sub- annual sites to annual resolution. Our PAGES2k network therefore consists of 40 proxy records: 12 tree- ring chronologies, 3 lake sediment cores, 5 corals, 19 ice- cores, and 1 borehole- derived temperature reconstruction (Supplemental Table 3). The tree- ring records are from Tasmania, New Zealand, and the central Andes. The longest two chronologies are from Mt. Read, Tasmania and Oroko, New Zealand, which begin in 494 BCE and 900 CE, respectively; the remaining tree chronologies mostly begin between 1450 CE and 1550 CE. The three lake sediment proxies are derived from the central and southern Andes. The longest record (Laguna Chemical) spans the complete Common Era, while Lagunas Escondida and Aculeo begin in 400 CE and 816 CE, respectively. The five coral records are from the Houtman Abrolhos Islands off the west coast of Australia and begin between 1795 CE and 1900 CE. The Antarctica ice core records have varying temporal coverage. Four sites cover the full Common Era (Plateau Remote, WDC06A, James Ross Island, WAIS- Divide), six more extend at least one millennium, and the remaining nine begin between 1140 CE and 1703 CE. The borehole reconstruction is from WAIS- Divide and begins in 8 CE. For the 40 proxy set, full coverage extends from 1903 CE to 1983 CE with 20 sites remaining by 2000 CE. + +The SADA and ANZDA are gridded tree- ring reconstructions of the self- calibrated Palmer Drought Severity Index (PDSI) during austral summer at annual resolution [65, 66]. The SADA is derived from 286 temperature and precipitation- sensitive tree- ring chronologies and begins in 1400 CE. The atlas covers all of South America south of \(12.25^{\circ}\mathrm{S}\) at \(0.5^{\circ}\) resolution. Similarly, ANZDA is derived from 176 tree- ring chronologies, as well as one coral record, and begins in 1500 CE. The ANZDA covers Australia east of \(136.25^{\circ}\mathrm{E}\) , and New Zealand, also at \(0.5^{\circ}\) resolution. The SAM is strongly associated with droughts and pluvials in the domains of both atlases [65], supporting their inclusion in our network. Both atlases have significantly higher spatial resolution than the reanalysis data and climate model output used for our reconstruction method. To permit calculations that require the same spatial resolution, we bin both atlases to the lowest resolution spatial grid relevant to a given experiment. For the main reconstruction, after applying latitude screening, our SADA and ANZDA networks consist of 104 and 71 binned records, each on a \(2^{\circ} \times 2.5^{\circ}\) grid. It is worth noting that several of the PAGES2k tree ring records used in our reconstruction were also used to construct the drought atlases, and these repeat records might initially appear to duplicate information in the reconstruction. However, our Kalman filter method explicitly accounts for covariance between proxy records, and + +<--- Page Split ---> + +down- weights proxies with repeated information accordingly. Additional details for this process can be found in the following section. + +## Reconstruction Method + +## Kalman Filter + +Our reconstruction uses an ensemble Kalman Filter approach [EnKF; 90], which follows the update equation: + +\[\mathbf{X}_{\mathbf{a}} = \mathbf{X}_{\mathbf{p}} - \mathbf{K}(\mathbf{Y} - \hat{\mathbf{Y}}) \quad (2)\] + +in each reconstructed time step. Here, the \(\mathbf{X}_{\mathbf{p}}\) and \(\mathbf{X}_{\mathbf{a}}\) matrices are the initial (prior) and updated (analysis) ensembles of climate model states. Each row holds a target climate variable, and each column a different selection of climate model output (ensemble member). \(\mathbf{Y}\) is a matrix of proxy values for the time step; the columns of \(\mathbf{Y}\) are constant, and each row holds the value from a particular proxy record repeated once for each ensemble member. \(\hat{\mathbf{Y}}\) holds the model estimates of the proxy values; each row has the estimates for a particular proxy site, and each column has the estimates from a particular ensemble member. \(\mathbf{K}\) is the Kalman gain: + +\[\mathbf{K} = \mathrm{cov}(\mathbf{X}_{\mathbf{p}},\hat{\mathbf{Y}})[\mathrm{cov}(\hat{\mathbf{Y}}) + \mathbf{R}]^{-1} \quad (3)\] + +where \(\mathbf{R}\) is the matrix of proxy error- covariances. As previously mentioned, the Kalman filter accounts for duplication of information across repeated proxy records. This occurs via the \(\mathrm{cov}(\hat{\mathbf{Y}})\) term in Equation 3, which reduces proxy weights in the Kalman gain as a function of shared proxy covariance. Note that any shared covariance derived from proxies' relationships with the SAM is balanced by the \(\mathrm{cov}(\mathbf{X}_{\mathbf{p}},\hat{\mathbf{Y}})\) term in Equation 3. We use a square- root variant of EnKF [91, 92]. This modifies equations 2 and 3 to update the ensemble mean and deviations separately, and precludes the need for perturbed observations [93]. The Kalman filter can be expressed as a recursive Bayesian filter [94, 95], so we will often refer to \(\mathbf{X}_{\mathbf{p}}\) and \(\mathbf{X}_{\mathbf{a}}\) as the prior and posterior in this paper. + +## Prior + +We construct the prior using output from climate models with paleoclimate simulations of the last millennium (Supplemental Table 1). We use a multi- model ensemble (MME), which has been found to reduce error relative to single model assimilations [59, 96]. Our MME consists of CCSM4, CESM- LME, MPI, and MRI, which represent the set of last millennium simulations with spatial resolutions greater than or at the resolution of the 20CR reanalysis. As such, this selection does not require us to bin the drought atlases to lower resolutions than 20CR, which allows us to extract maximum information from SADA and ANZDA. We also tested a larger MME consisting of 10 models with last millennium simulations regardless of resolution. Our tests show that the high- resolution MME maximizes reconstruction skill (Supplemental Figure 3). For CCSM4, MPI, and MRI, we use output from the + +<--- Page Split ---> + +PMIP3 last1000 (850- 1850 CE) and historical (1851- 2005 CE) experiments, specifically ensemble member rli1p1. For CESM- LME, we use output from full- forcing run 2 (850- 2005 CE). While the PAGES2k proxy network does include stable oxygen isotope proxies, there are too few high- resolution last millennium isotope- enabled paleoclimate model simulations available to construct a multi- model prior[96]. + +We use an offline, stationary prior for our assimilation. Offline approaches [97, 98] differ from classical Kalman Filters in that updates are not used to inform model simulation. Instead, offline methods use pre- existing model output to build the prior in each time step. The offline approach has been shown to compare favorably with classical (online) methods in paleoclimate contexts but at a fraction of the computational cost [99, 100]. The stationary prior indicates that we use the same ensemble as the prior for each reconstructed time step. This is common in paleoclimate DA applications [52, 57, 101] and is justified by the limited forecast skill of climate models beyond the annual reconstruction time scale [102]. However, stationary priors have been observed to artificially reduce the variability of reconstructions as proxy networks become more sparse [59]. Consequently, our use of stationary priors necessitates a correction for the reconstruction's variability, which is detailed in the methods below. + +To build each prior, we first calculate the DJF SAM time- series for each model, normalizing zonal SLP means to the pre- industrial period (850- 1849 CE). We then concatenate the SAM index time- series from each model in every year of model output. The final prior has a total of 4624 ensemble members from 4 high- resolution models. + +## Proxy Forward Models and Error Covariances + +The proxy modeling process begins by designing a forward model for each assimilated proxy record. For the PAGES2k records, we follow previous studies [53, 59] and use simple univariate linear models: + +\[\hat{\mathbf{Y}} = a\mathbf{T} + b \quad (4)\] + +where \(\hat{\mathbf{Y}}\) is a vector of proxy estimates, and \(\mathbf{T}\) is a vector of seasonal temperature means. Here, the seasonal means used for each site is taken from the seasonal sensitivity reported in the PAGES2k metadata [49]. We determine the coefficients \(a\) and \(b\) by calibrating each proxy PAGES2k record to the corresponding climate data from 20CR. For each proxy site, we first determine the seasonal sensitivity and then linearly regress the proxy record against the seasonal- mean temperature vector from the closest 20CR grid point in all overlapping years from 1950 - 2000 CE. The regression slope and intercept are then used as coefficients a and b. For the drought atlases, we estimate proxies by calculating PDSI [103] using the Thornthwaite estimation of potential evapotranspiration [104]. This uses monthly mean temperature and precipitation from a drought atlas grid cell to compute monthly PDSI values for each year. We then use the austral summer means of these monthly values as the proxy estimates. Effectively: + +\[\mathbf{Y} = \mathrm{mean}\big[\mathrm{PDSI}_{\mathrm{Thornthwaite}}(\mathbf{T},\mathbf{P})\big]_{\mathrm{DJF}} \quad (5)\] + +<--- Page Split ---> + +where T and P are monthly temperature and precipitation, and Y is the drought atlas estimate. We estimate proxy values for the model priors by applying Equations 4 and 5 to climate model output and matching each year's estimates to the associated ensemble member in the prior. + +Although the PDSI calculation in Equation 5 uses the Thornthwaite approximation, both drought atlases target an observational dataset based on the Pennam-Monteith method [65, 66, 105]. However, both the Thornthwaite and Pennam-Monteith equations have been shown to perform similarly when applied to preindustrial simulations, and this agreement occurs because the simplifying assumptions of the Thornthwaite method remain valid over the relatively confined range of last millennium temperatures [106]. For the purposes of this study, the Thornthwaite method provides two further advantages: First, the Thornthwaite equation is more computationally tractable, which allows us to apply it to the large spatial regions and the multiple millennium- length climate model simulations used for priors in our assimilation method. Second, because the Thornthwaite calculation requires fewer climate model data fields to estimate the PDSI [104, 105], opportunities for climate model biases to degrade the reconstruction are reduced. + +We next estimate the proxy error covariances. These error covariances describe the uncertainty in the comparison of observed records to the proxy estimates \((\mathbf{Y} - \mathbf{Y}\mathbf{e})\) . In a classical Kalman Filter, the estimates \((\hat{\mathbf{Y}})\) are known perfectly and this uncertainty is derived from the observations \((\mathbf{Y})\) , so \(\mathbf{R}\) is often referred to as observation uncertainty. In paleoclimate contexts, this situation is inverted: proxy measurements are typically precise and uncertainty derives from the simplifications and parameterizations inherent in the estimation equations. Hence, we quantify \(\mathbf{R}\) by running Equations 4 and 5 on the 20CR dataset (from 1950- 2000 CE) and comparing the estimated proxy values to the real records. The differences between the two sets of values are used to estimate the errors inherent in using simple models and relatively coarse climate data to estimate the temporal behavior of the proxy records. Most EnKF paleoclimate efforts assume that proxy errors are independent, such that \(\mathbf{R}\) is a diagonal matrix [52, 53, 57, 64]. This is justified for datasets like PAGES2k, for which proxy uncertainties are dominated by local biological, physical, and mechanistic effects [48]. However, the drought atlas grid points are strongly spatially correlated, so this assumption is not appropriate in this study. Instead, we calculate independent error- variances for the proxies in the PAGES2k network, and full errorcovariances for both SADA and ANZDA. Hence, \(\mathbf{R}\) is block- diagonal, rather than strictly diagonal. We estimate uncertainty in the final reconstruction from the spread of the assimilation posterior. + +## Variance Correction + +The use of stationary priors creates artifacts in the variability of raw reconstruction. As the proxy network becomes sparse, less information is incorporated in the Kalman Filter, and the updated state is less able to move off the prior mean. This causes reconstruction variability to increase with the size of the proxy network and independently of the climatological record. We apply a variance adjustment scheme + +<--- Page Split ---> + +to correct for this effect. Variance adjustments are common in paleoclimate reconstructions [107–110] and are inherent to simpler methods like Composite Plus Scale [111]. + +Here, we use a series of frozen- network assimilations to adjust temporal variance. There are five sites in our proxy network with observations in every year of the reconstruction. We first assimilate this five- site network over the full interval 1- 2000 CE to derive a baseline time- series that is not affected by changes to the proxy network. We next determine each unique set of proxy sites used to update one or more time steps in the reconstruction. We then assimilate each set of proxies over the time steps for which all of the proxies have recorded values, and determine the ratio of this assimilation's standard deviation to that of the baseline time series over all overlapping years: + +\[P(\mathrm{set}) = \sigma_{\mathrm{set}} / \sigma_{\mathrm{Baseline}} \quad (6)\] + +We then calculate a scaling factor for each time step using the normalized ratio for the associated proxy set: + +\[w(t) = P(\mathrm{set}(t)) / \mathrm{max}(P) \quad (7)\] + +A comparison of the raw and variance- adjusted reconstructions is provided in Supplemental Figure 4. + +## Optimal Sensor Analysis + +We follow a previously established framework for optimal sensor analyses [67]. In brief, the method quantifies the ability of proxy sites to reduce the variance of a metric in a posterior ensemble. Here, we use the SAM as our metric, so the optimal sensor analysis here assesses the ability of sites to reduce uncertainty in the index across the reconstruction posterior. We first compute the total reduction in SAM posterior variance using the complete set of proxies with observations in each time step. We also quantify each site's ability to reduce reconstruction uncertainty when no other sites are in the proxy network. We refer to this quantity as 'potential percent constrained variance'. + +## External Forcing Analyses + +We begin our external forcing analysis by investigating the SAM's response to natural climate forcings. We first use a wavelet coherence analysis to examine the relationship between our SAM reconstruction and a time series of reconstructed solar forcing [110, 112]. We next use a superposed epoch analysis [113] to determine the reconstruction's composite mean response to major volcanic eruptions. We used the eVolv2k V3 volcanic forcing dataset [114, 115] to select events with a total forcing magnitude greater than or equal to that of Krakatoa. This yielded 28 eruption years: 87, 169, 266, 433, 536, 540, 574, 626, 682, 817, 939, 1108, 1171, 1182, 1230, 1257, 1276, 1286, 1345, 1458, 1600, 1640, 1695, 1783, 1809, 1815, 1831, and 1883. For the SEA, we normalized each event to the mean of the preceding 5 years and examined the composite mean response over the 10 years following volcanic events. We tested + +<--- Page Split ---> + +the significance of the observed response by bootstrapping 5,000 SEA time series via random draws of 28 event years from the remaining years in the reconstruction. + +We next consider the SAM's response to anthropogenic forcings using both our reconstruction and the Marshall index. Before quantifying trends, we first normalize our reconstruction to the Marshall index, such that the mean and variance of the detrended normalized reconstruction matches those of the detrended Marshall index over the years of common overlap (1958- 2000 CE). This places the series in the same unit space while preserving differences in the instrumental trend. We then calculate moving trends for the reconstruction over the years 1900- 2000 CE using trend window lengths from 31 to 101 years. Similarly, we calculate moving trends for the Marshall Index over the years 1958- 2020 CE using trend window lengths from 31 to 63 years. We then use the reconstruction to assess the significance of these trends. For each trend window length, we calculate the distribution of trends with the given window length from the reconstruction over the years 1500- 1900 CE, and we define this distribution as the natural variability for that trend length. We then use the 90% confidence intervals of each distribution to determine a significance threshold for trends of the associated length. We also repeat this process using trend distributions from the intervals 1- 1900 CE and 1- 899 CE to examine the sensitivity of this analysis to different portions of the reconstruction. + +## Caveats and Limitations + +Our DA method does not require a calibration with the instrumental SAM, which limits sensitivity to non- stationarity in the SAM during the instrumental era. However, the trade- off is the influence of proxy forward model and climate model biases on the reconstruction. In the case of proxy models, any biases typically reduce the weight of the proxy in the assimilation, thereby limiting its effect on the reconstruction. We note that improving the accuracy or sophistication of the proxy forward models could increase the influence of many records; for example, transitioning the statistical forward models used here for the PAGES2k sites to more mechanistically accurate proxy system models [54] could potentially improve the reconstruction [101]. However, efforts to develop more complex proxy system models must also exercise caution, as excessive complexity and poorly constrained parameters may lead to overfitting and artificially high skill in the instrumental era at the expense of accuracy during the earlier reconstruction. In this study, we retain the simpler statistical forward models because (1) the PAGES2k proxies are reported to be temperature sensitive [49], (2) statistical proxy models remain the most common and tractable approach for paleoclimate data assimilation to date [53, 57, 59], and (3) the simple statistical model eliminates errors caused by the interaction of climate model biases with forward models that rely on absolute units. + +With respect to climate models, biases in the mean state can affect proxy estimates that include parametrizations or thresholds based on absolute units. However, covariance biases are a greater concern, as they introduce errors in the propagation of information from the proxy records to the reconstruction target. For example, some of the climate models considered in this study simulate a SAM pattern that is too zonally symmetric and that overestimates the SAM's influence on overall Southern + +<--- Page Split ---> + +Hemisphere circulation [116]. Such teleconnection biases can cause the assimilation to overestimate the covariance between various proxies and the SAM, thereby increasing reconstruction error. In this study, we use a multi- model ensemble (MME) to reduce the effects of covariance bias from any one model [59, 96]. We note that we weight each model equally, which effectively treats each model as independent. In reality, many models share common features or code, so this equal weighting may bias an ensemble towards the most similar models [117, 118]. For example, the CCSM4 and CESM- LME output used in our MME are both from models developed by the US National Center for Atmospheric Research (NCAR) and may more closely resemble one another than the MPI or MRI models. Future efforts may wish to test different model composition and weights when constructing a MME prior. + +Finally, our use of a stationary offline prior implies a stationary estimate of climate system covariance when considered over the full reconstruction period. Although we use a long- term estimate of the SAM's climate covariance, the true covariance may vary on multi- decadal scales [13, 44], and these variations will not be captured in our approach. While the assumption of a reasonably stationary covariance is implicitly common to most spatial reconstruction methods [119, 120], the application of transient offline priors [102, 121, 122] or online assimilation techniques [123] may enhance future data assimilation reconstruction, although these approaches must balance the utility of evolving covariance estimates with reduced ensemble sizes. + +## Acknowledgments + +We thank David Meko for providing the original PDSI estimation code on which ours is based. This research is supported by a grant from the US National Science Foundation's Paleo Perspectives on Climatic Change program (P2C2) AGS- 1803946 to K.J.A. and A.H. T.V. is supported by an Australian Research Council Special Research Initiative for Antarctic Gateway Partnership (SR140300001) and the Australian Antarctic Program Partnership (ASCI000002). K.A. is supported by a grant from the Australian Research Council (FT200100102). We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. + +## Author contributions + +K.J.A., A.H., and J.K. designed the research. J.K. wrote all the software and code, improved and enhanced the assimilation method, conducted all calculations, and produced all the figures. A.H., T.V., and K.A. provided guidance, data, and knowledge for individual proxies and Southern Hemisphere climate variability. All authors (J.K., K.J.A., A.H., T.V. and K.A.) interpreted results and wrote the paper. + +## Competing interests + +The authors declare that they have no conflicts of interest. + +<--- Page Split ---> + +## Data and code availability + +The data and code used to produce this analysis, as well as the final reconstruction, will be made publicly available in a Zenodo repository with a permanent DOI pending review. + +## Supplementary materials + +Supplemental Figure 1: Trend analysis using the early reconstruction (1- 899 CE) to quantify natural variability and trend significance. + +Supplemental Figure 2: Correlation of reconstructions with the Marshall Index as a function of latitude cutoff thresholds. + +Supplemental Figure 3: Comparison of reconstruction skill using different multimodel priors. + +Supplemental Figure 4: Raw and variance- adjusted reconstruction time series. + +Supplemental Table 1: Climate models tested in assimilation priors. + +Supplemental Table 2: Latitudes of maximum negative SLP correlation in climate models and 20CR. + +Supplemental Table 3: PAGES2k sites used in the final reconstruction. + +<--- Page Split ---> + +## References + +[1] Rogers, J. C. & Van Loon, H. Spatial variability of sea level pressure and 500 mb height anomalies over the southern hemisphere. Monthly Weather Review 110 (10), 1375- 1392 (1982). + +[2] Gong, D. & Wang, S. Definition of antarctic oscillation index. Geophysical Research Letters 26 (4), 459- 462 (1999). + +[3] Thompson, D. W. & Wallace, J. M. Annular modes in the extratropical circulation. part i: Month- to- month variability. Journal of Climate 13 (5), 1000- 1016 (2000). + +[4] Marshall, G. J. Trends in the southern annular mode from observations and reanalyses. Journal of Climate 16 (24), 4134- 4143 (2003). + +[5] Thompson, D. W. & Solomon, S. Interpretation of recent southern hemisphere climate change. Science 296 (5569), 895- 899 (2002). + +[6] Kwok, R. & Comiso, J. C. Spatial patterns of variability in antarctic surface temperature: Connections to the southern hemisphere annular mode and the southern oscillation. Geophysical Research Letters 29 (14), 50- 1 (2002). + +[7] Gillett, N. P., Kell, T. D. & Jones, P. Regional climate impacts of the southern annular mode. Geophysical Research Letters 33 (23) (2006). + +[8] Gupta, A. S. & England, M. H. Coupled ocean- atmosphere- ice response to variations in the southern annular mode. Journal of Climate 19 (18), 4457- 4486 (2006). + +[9] Marshall, G. J., Orr, A., Van Lipzig, N. P. & King, J. C. The impact of a changing southern hemisphere annular mode on antarctic peninsula summer temperatures. Journal of Climate 19 (20), 5388- 5404 (2006). + +[10] Hendon, H. H., Thompson, D. W. & Wheeler, M. C. Australian rainfall and surface temperature variations associated with the southern hemisphere annular mode. Journal of Climate 20 (11), 2452- 2467 (2007). + +[11] Van Lipzig, N. P., Marshall, G. J., Orr, A. & King, J. C. The relationship between the southern hemisphere annular mode and antarctic peninsula summer temperatures: Analysis of a high- resolution model climatology. Journal of Climate 21 (8), 1649- 1668 (2008). + +[12] Silvestri, G. E. & Vera, C. S. Antarctic oscillation signal on precipitation anomalies over southeastern south america. Geophysical Research Letters 30 (21) (2003). + +<--- Page Split ---> + +[13] Silvestri, G. & Vera, C. Nonstationary impacts of the southern annular mode on southern hemisphere climate. Journal of Climate 22 (22), 6142- 6148 (2009). + +[14] Cai, W. & Cowan, T. Sam and regional rainfall in ipcc ar4 models: Can anthropogenic forcing account for southwest western australian winter rainfall reduction? Geophysical Research Letters 33 (24) (2006). + +[15] Reason, C. & Rouault, M. Links between the antarctic oscillation and winter rainfall over western south africa. Geophysical Research Letters 32 (7) (2005). + +[16] Kidston, J., Renwick, J. & McGregor, J. Hemispheric-scale seasonality of the southern annular mode and impacts on the climate of new zealand. Journal of Climate 22 (18), 4759- 4770 (2009). + +[17] Holz, A. & Veblen, T. T. Variability in the southern annular mode determines wildfire activity in patagonia. Geophysical Research Letters 38 (14) (2011). + +[18] Mariani, M. & Fletcher, M.- S. The southern annular mode determines interannual and centennial- scale fire activity in temperate southwest tasmania, australia. Geophysical Research Letters 43 (4), 1702- 1709 (2016). + +[19] Holz, A. et al. Southern annular mode drives multicentury wildfire activity in southern south america. Proceedings of the National Academy of Sciences 114 (36), 9552- 9557 (2017). + +[20] Abram, N. J. et al. Connections of climate change and variability to large and extreme forest fires in southeast australia. Communications Earth & Environment 2 (1), 1- 17 (2021). + +[21] Lim, E.- P. et al. The 2019 southern hemisphere stratospheric polar vortex weakening and its impacts. Bulletin of the American Meteorological Society 102 (6), E1150- E1171 (2021). + +[22] Hall, A. & Visbeck, M. Synchronous variability in the southern hemisphere atmosphere, sea ice, and ocean resulting from the annular mode. Journal of Climate 15 (21), 3043- 3057 (2002). + +[23] Stammerjohn, S. E., Martinson, D., Smith, R., Yuan, X. & Rind, D. Trends in antarctic annual sea ice retreat and advance and their relation to el niño- southern oscillation and southern annular mode variability. Journal of Geophysical Research: Oceans 113 (C3) (2008). + +[24] Simpkins, G. R., Ciasto, L. M., Thompson, D. W. & England, M. H. Seasonal relationships between large- scale climate variability and antarctic sea ice concentration. Journal of Climate 25 (16), 5451- 5469 (2012). + +<--- Page Split ---> + +[25] Kohyama, T. & Hartmann, D. L. Antarctic sea ice response to weather and climate modes of variability. Journal of Climate 29 (2), 721-741 (2016). + +[26] Lovenduski, N. S., Gruber, N., Doney, S. C. & Lima, I. D. Enhanced co2 outgassing in the southern ocean from a positive phase of the southern annular mode. Global Biogeochemical Cycles 21 (2) (2007). + +[27] Lovenduski, N. S., Fay, A. R. & McKinley, G. A. Observing multidecadal trends in southern ocean co2 uptake: What can we learn from an ocean model? Global Biogeochemical Cycles 29 (4), 416-426 (2015). + +[28] Sallée, J.- B., Speer, K. & Rintoul, S. Zonally asymmetric response of the southern ocean mixed-layer depth to the southern annular mode. Nature Geoscience 3 (4), 273-279 (2010). + +[29] Thompson, D. W., Wallace, J. M. & Hegerl, G. C. Annular modes in the extratropical circulation. part ii: Trends. Journal of Climate 13 (5), 1018-1036 (2000). + +[30] Fogt, R. L. & Marshall, G. J. The southern annular mode: variability, trends, and climate impacts across the southern hemisphere. Wiley Interdisciplinary Reviews: Climate Change 11 (4), e652 (2020). + +[31] Arblaster, J. M. & Meehl, G. A. Contributions of external forcings to southern annular mode trends. Journal of Climate 19 (12), 2896-2905 (2006). + +[32] Thompson, D. W. et al. Signatures of the antarctic ozone hole in southern hemisphere surface climate change. Nature Geoscience 4 (11), 741-749 (2011). + +[33] Polvani, L. M., Waugh, D. W., Correa, G. J. & Son, S.- W. Stratospheric ozone depletion: The main driver of twentieth-century atmospheric circulation changes in the southern hemisphere. Journal of Climate 24 (3), 795-812 (2011). + +[34] England, M. R., Polvani, L. M., Smith, K. L., Landrum, L. & Holland, M. M. Robust response of the amundsen sea low to stratospheric ozone depletion. Geophysical Research Letters 43 (15), 8207-8213 (2016). + +[35] Jones, J. M. et al. Assessing recent trends in high-latitude southern hemisphere surface climate. Nature Climate Change 6 (10), 917-926 (2016). + +[36] Sousa, P. M., Blamey, R. C., Reason, C. J., Ramos, A. M. & Trigo, R. M. The 'day zero' cape town drought and the poleward migration of moisture corridors. Environmental Research Letters 13 (12), 124025 (2018). + +<--- Page Split ---> + +[37] Verdon- Kidd, D. C. & Kiem, A. S. Nature and causes of protracted droughts in southeast australia: Comparison between the federation, wwii, and big dry droughts. Geophysical Research Letters 36 (22) (2009). [38] Cai, W., Van Rensch, P., Borlace, S. & Cowan, T. Does the southern annular mode contribute to the persistence of the multidecade- long drought over southwest western australia? Geophysical Research Letters 38 (14) (2011). [39] Barrucand, M. G., Zitto, M. E., Piotrkowski, R., Canziani, P. & O'Neill, A. Historical sam index time series: linear and nonlinear analysis. International Journal of Climatology 38, e1091- e1106 (2018). [40] Villalba, R. et al. Unusual southern hemisphere tree growth patterns induced by changes in the southern annular mode. Nature Geoscience 5 (11), 793- 798 (2012). [41] Abram, N. J. et al. Evolution of the southern annular mode during the past millennium. Nature Climate Change 4 (7), 564- 569 (2014). [42] Dätwyler, C. et al. Teleconnection stationarity, variability and trends of the southern annular mode (sam) during the last millennium. Climate Dynamics 51 (5), 2321- 2339 (2018). [43] Hessl, A., Allen, K. J., Vance, T., Abram, N. J. & Saunders, K. M. Reconstructions of the southern annular mode (sam) during the last millennium. Progress in Physical Geography 41 (6), 834- 849 (2017). [44] Gallant, A. J., Phipps, S. J., Karoly, D. J., Mullan, A. B. & Lorrey, A. M. Nonstationary australasian teleconnections and implications for paleoclimate reconstructions. Journal of Climate 26 (22), 8827- 8849 (2013). [45] Brönnimann, S. et al. Tropical circulation and precipitation response to ozone depletion and recovery. Environmental Research Letters 12 (6), 064011 (2017). [46] Yun, K.- S. & Timmermann, A. Decadal monsoon- enso relationships reexamined. Geophysical Research Letters 45 (4), 2014- 2021 (2018). [47] Huiskamp, W. & McGregor, S. Quantifying southern annular mode paleoreconstruction skill in a model framework. Climate of the Past 17 (5), 1819- 1839 (2021). [48] Jones, P. D. et al. High- resolution palaeoclimatology of the last millennium: a review of current status and future prospects. The Holocene 19 (1), 3- 49 (2009). + +<--- Page Split ---> + +Trends and variability in the Southern Annular Mode over the Common Era 23 + +[49] PAGES2k Consortium. A global multiproxy database for temperature reconstructions of the common era. Scientific Data 4 (2017). + +[50] Tierney, J. E. et al. Tropical sea surface temperatures for the past four centuries reconstructed from coral archives. Paleoceanography 30 (3), 226- 252 (2015). + +[51] Gulev, S. K. et al. in Changing State of the Climate System (eds Masson- Delmotte, V. et al.) Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change Ch. 2 (Cambridge University Press, 2021). + +[52] Steiger, N. J., Hakim, G. J., Steig, E. J., Battisti, D. S. & Roe, G. H. Assimilation of time- averaged pseudoproxies for climate reconstruction. Journal of Climate 27 (1), 426- 441 (2014). + +[53] Hakim, G. J. et al. The last millennium climate reanalysis project: Framework and first results. Journal of Geophysical Research: Atmospheres 121 (12), 6745- 6764 (2016). + +[54] Evans, M. N., Tolwinski- Ward, S. E., Thompson, D. M. & Anchukaitis, K. J. Applications of proxy system modeling in high resolution paleoclimatology. Quaternary Science Reviews 76, 16- 28 (2013). + +[55] Dee, S. et al. Prysm: An open- source framework for proxy system modeling, with applications to oxygen- isotope systems. Journal of Advances in Modeling Earth Systems 7 (3), 1220- 1247 (2015). + +[56] Goosse, H. et al. The role of forcing and internal dynamics in explaining the "medieval climate anomaly". Climate Dynamics 39 (12), 2847- 2866 (2012). + +[57] Tardif, R. et al. Last millennium reanalysis with an expanded proxy database and seasonal proxy modeling. Climate of the Past 15 (4), 1251- 1273 (2019). + +[58] Neukom, R., Steiger, N., Gómez- Navarro, J. J., Wang, J. & Werner, J. P. No evidence for globally coherent warm and cold periods over the preindustrial common era. Nature 571 (7766), 550- 554 (2019). + +[59] King, J. M. et al. A data assimilation approach to last millennium temperature field reconstruction using a limited high- sensitivity proxy network. Journal of Climate 1- 64 (2021). + +[60] Zhu, F., Emile- Geay, J., Hakim, G. J., King, J. & Anchukaitis, K. J. Resolving the differences in the simulated and reconstructed temperature response to volcanism. Geophysical Research Letters 47 (8), e2019GL086908 (2020). + +<--- Page Split ---> + +[61] Zhu, F. et al. A re- appraisal of the ENSO response to volcanism with paleoclimate data assimilation. Nature Communications 13 (1) (2022). https://doi.org/10.1038/s41467-022-28210-1. + +[62] Dalaiden, Q., Goosse, H., Rezsöházy, J. & Thomas, E. R. Reconstructing atmospheric circulation and sea- ice extent in the west antarctic over the past 200 years using data assimilation. Climate Dynamics 1- 25 (2021). + +[63] Tierney, J. E. et al. Glacial cooling and climate sensitivity revisited. Nature 584 (7822), 569- 573 (2020). + +[64] Steiger, N. J., Smerdon, J. E., Cook, E. R. & Cook, B. I. A reconstruction of global hydroclimate and dynamical variables over the common era. Scientific Data 5 (1), 1- 15 (2018). + +[65] Morales, M. S. et al. Six hundred years of south american tree rings reveal an increase in severe hydroclimatic events since mid- 20th century. Proceedings of the National Academy of Sciences 117 (29), 16816- 16823 (2020). + +[66] Palmer, J. G. et al. Drought variability in the eastern australia and new zealand summer drought atlas (anzda, ce 1500- 2012) modulated by the interdecadal pacific oscillation. Environmental Research Letters 10 (12), 124002 (2015). + +[67] Comboul, M., Emile- Geay, J., Hakim, G. J. & Evans, M. N. Paleoclimate sampling as a sensor placement problem. Journal of Climate 28 (19), 7717- 7740 (2015). + +[68] Bradley, R. S. Jones, P. D., Bradley, R. S. & Jouzel, J. (eds) Are there optimum sites for global paleotemperature reconstruction? (eds Jones, P. D., Bradley, R. S. & Jouzel, J.) Climatic Variations and Forcing Mechanisms of the Last 2000 Years, 603- 624 (Springer Berlin Heidelberg, Berlin, Heidelberg, 1996). + +[69] Evans, M. N., Kaplan, A. & Cane, M. A. Optimal sites for coral- based reconstruction of global sea surface temperature. Paleoceanography 13 (5), 502- 516 (1998). + +[70] Mauger, G., Bumbaco, K., Hakim, G. & Mote, P. Optimal design of a climatological network: Beyond practical considerations. Geoscientific Instrumentation, Methods and Data Systems 2 (2), 199- 212 (2013). + +[71] Fogt, R. L. et al. Historical sam variability. part ii: Twentieth- century variability and trends from reconstructions, observations, and the ipcc ar4 models. Journal of Climate 22 (20), 5346- 5365 (2009). + +[72] Jones, J. M. et al. Historical sam variability. part i: Century- length seasonal reconstructions. Journal of Climate 22 (20), 5319- 5345 (2009). + +<--- Page Split ---> + +[73] Lim, E.- P., Hendon, H. & Thompson, D. Seasonal evolution of stratosphere- troposphere coupling in the southern hemisphere and implications for the predictability of surface climate. Journal of Geophysical Research: Atmospheres 123 (21), 12–002 (2018). + +[74] Byrne, N. J., Shepherd, T. G. & Polichtchouk, I. Subseasonal- to- seasonal predictability of the southern hemisphere eddy- driven jet during austral spring and early summer. Journal of Geophysical Research: Atmospheres 124 (13), 6841–6855 (2019). + +[75] L'Heureux, M. L. & Thompson, D. W. Observed relationships between the el niño–southern oscillation and the extratropical zonal- mean circulation. Journal of Climate 19 (2), 276–287 (2006). + +[76] Wilson, A. B., Bromwich, D. H. & Hines, K. M. Simulating the mutual forcing of anomalous high southern latitude atmospheric circulation by el niño flavors and the southern annular mode. Journal of Climate 29 (6), 2291–2309 (2016). + +[77] Gillett, N. P., Fyfe, J. C. & Parker, D. E. Attribution of observed sea level pressure trends to greenhouse gas, aerosol, and ozone changes. Geophysical Research Letters 40 (10), 2302–2306 (2013). + +[78] PAGES 2k-PMIP3 group. Continental- scale temperature variability in PMIP3 simulations and PAGES 2k regional temperature reconstructions over the past millennium. Climate of the Past 11 (12), 1673–1699 (2015). https://doi.org/10.5194/cp-11-1673-2015. + +[79] Ding, Q., Steig, E. J., Battisti, D. S. & Wallace, J. M. Influence of the tropics on the southern annular mode. Journal of Climate 25 (18), 6330–6348 (2012). + +[80] Fogt, R. L., Bromwich, D. H. & Hines, K. M. Understanding the sam influence on the south pacific enso teleconnection. Climate Dynamics 36 (7), 1555–1576 (2011). + +[81] Fogt, R. L. & Bromwich, D. H. Decadal variability of the enso teleconnection to the high- latitude south pacific governed by coupling with the southern annular mode. Journal of Climate 19 (6), 979–997 (2006). + +[82] Emile- Geay, J., Cobb, K. M., Mann, M. E. & Wittenberg, A. T. Estimating central equatorial pacific sst variability over the past millennium. part ii: Reconstructions and implications. Journal of Climate 26 (7), 2329–2352 (2013). + +[83] Wright, N. M., Krause, C. E., Phipps, S. J., Boschat, G. & Abram, N. J. Influence of long- term changes in solar irradiance forcing on the southern annular mode. Climate of the Past 18 (6), 1509–1528 (2022). + +<--- Page Split ---> + +[84] Ammann, C. M., Joos, F., Schimel, D. S., Otto- Bliesner, B. L. & Tomas, R. A. Solar influence on climate during the past millennium: Results from transient simulations with the NCAR Climate System Model. Proceedings of the National Academy of Sciences 104 (10), 3713- 3718 (2007). + +[85] Hind, A. & Moberg, A. Past millennial solar forcing magnitude. Climate dynamics 41 (9), 2527- 2537 (2013). + +[86] Schurer, A. P., Tett, S. F. & Hegerl, G. C. Small influence of solar variability on climate over the past millennium. Nature Geoscience 7 (2), 104- 108 (2014). + +[87] Ho, M., Kiem, A. & Verdon- Kidd, D. The southern annular mode: a comparison of indices. Hydrology and Earth System Sciences 16 (3), 967- 982 (2012). + +[88] Compo, G. P. et al. The twentieth century reanalysis project. Quarterly Journal of the Royal Meteorological Society 137 (654), 1- 28 (2011). + +[89] Slivinski, L. C. et al. Towards a more reliable historical reanalysis: Improvements for version 3 of the twentieth century reanalysis system. Quarterly Journal of the Royal Meteorological Society 145 (724), 2876- 2908 (2019). + +[90] Evensen, G. Sequential data assimilation with a nonlinear quasi- geostrophic model using monte carlo methods to forecast error statistics. Journal of Geophysical Research: Oceans 99 (C5), 10143- 10162 (1994). + +[91] Andrews, A. A square root formulation of the kalman covariance equations. AIAA Journal 6 (6), 1165- 1166 (1968). + +[92] Tippett, M. K., Anderson, J. L., Bishop, C. H., Hamill, T. M. & Whitaker, J. S. Ensemble square root filters. Monthly Weather Review 131 (7), 1485- 1490 (2003). + +[93] Whitaker, J. S. & Hamill, T. M. Ensemble data assimilation without perturbed observations. Monthly Weather Review 130 (7), 1913- 1924 (2002). + +[94] Chen, Z. et al. Bayesian filtering: From kalman filters to particle filters, and beyond. Statistics 182 (1), 1- 69 (2003). + +[95] Wikle, C. K. & Berliner, L. M. A bayesian tutorial for data assimilation. Physica D: Nonlinear Phenomena 230 (1- 2), 1- 16 (2007). + +[96] Parsons, L. A. et al. Do multi- model ensembles improve reconstruction skill in paleoclimate data assimilation? Earth and Space Science 8 (4), e2020EA001467 (2021). + +[97] Oke, P. R., Allen, J. S., Miller, R. N., Egbert, G. D. & Kosro, P. M. Assimilation of surface velocity data into a primitive equation coastal ocean model. + +<--- Page Split ---> + +Trends and variability in the Southern Annular Mode over the Common Era 27 + +Journal of Geophysical Research: Oceans 107 (C9), 5- 1 (2002). + +[98] Evensen, G. The ensemble kalman filter: Theoretical formulation and practical implementation. Ocean Dynamics 53 (4), 343- 367 (2003). + +[99] Matsikaris, A., Widmann, M. & Jungclaus, J. On- line and off- line data assimilation in palaeoclimatology: a case study. Climate of the Past 11 (1), 81- 93 (2015). + +[100] Acevedo, W., Fallah, B., Reich, S. & Cubasch, U. Assimilation of pseudotree- ring- width observations into an atmospheric general circulation model. Climate of the Past 13 (5), 545- 557 (2017). + +[101] Dee, S. G., Steiger, N. J., Emile- Geay, J. & Hakim, G. J. On the utility of proxy system models for estimating climate states over the common era. Journal of Advances in Modeling Earth Systems 8 (3), 1164- 1179 (2016). + +[102] Bhend, J., Franke, J., Folini, D., Wild, M. & Brönnimann, S. An ensemble- based approach to climate reconstructions. Climate of the Past 8 (3), 963- 976 (2012). + +[103] Palmer, W. C. Meteorological drought Vol. 30 (US Department of Commerce, Weather Bureau, 1965). + +[104] Thornthwaite, C. W. An approach toward a rational classification of climate. Geographical Review 38 (1), 55- 94 (1948). + +[105] Allen, R., Smith, M., Perrier, A. & Pereira, L. S. An update for the definition of reference evapotranspiration. ICID bulletin 43 (2), 1- 34 (1994). + +[106] Smerdon, J. E., Cook, B. I., Cook, E. R. & Seager, R. Bridging past and future climate across paleoclimatic reconstructions, observations, and models: A hydroclimate case study. Journal of Climate 28 (8), 3212- 3231 (2015). + +[107] Cook, E. R., Meko, D. M., Stahle, D. W. & Cleaveland, M. K. Drought reconstructions for the continental united states. Journal of Climate 12 (4), 1145- 1162 (1999). + +[108] Esper, J., Frank, D. C., Wilson, R. J. & Briffa, K. R. Effect of scaling and regression on reconstructed temperature amplitude for the past millennium. Geophysical Research Letters 32 (7) (2005). + +[109] Frank, D., Esper, J. & Cook, E. R. Adjustment for proxy number and coherence in a large- scale temperature reconstruction. Geophysical Research Letters 34 (16) (2007). + +[110] Anchukaitis, K. J. et al. Last millennium northern hemisphere summer temperatures from tree rings: Part ii, spatially resolved reconstructions. Quaternary + +<--- Page Split ---> + +1006 Science Reviews 163, 1- 22 (2017). + +1007 [111] Mann, M. E. & Jones, P. D. Global surface temperatures over the past two millennia. Geophysical Research Letters 30 (15) (2003). + +1009 [112] Schmidt, G. et al. Climate forcing reconstructions for use in pmp simulations of the last millennium (v1.1). Geoscientific Model Development 5 (1), 185- 191 (2012). + +1012 [113] Haurwitz, M. W. & Brier, G. W. A critique of the superposed epoch analysis method: its application to solar- weather relations. Monthly Weather Review 109 (10), 2074- 2079 (1981). + +1015 [114] Sigl, M. et al. Timing and climate forcing of volcanic eruptions for the past 2,500 years. Nature 523 (7562), 543- 549 (2015). + +1017 [115] Toohey, M. & Sigl, M. Volcanic stratospheric sulfur injections and aerosol optical depth from 500 bce to 1900 ce. Earth System Science Data 9 (2), 809- 831 (2017). + +1020 [116] Zheng, F., Li, J., Clark, R. T. & Nnamchi, H. C. Simulation and projection of the southern hemisphere annular mode in cmip5 models. Journal of Climate 26 (24), 9860- 9879 (2013). + +1023 [117] Knutti, R., Masson, D. & Gettelman, A. Climate model genealogy: Generation cmip5 and how we got there. Geophysical Research Letters 40 (6), 1194- 1199 (2013). + +1026 [118] Sanderson, B. M., Knutti, R. & Caldwell, P. A representative democracy to reduce interdependency in a multimodel ensemble. Journal of Climate 28 (13), 5171- 5194 (2015). + +1029 [119] Tingley, M. P. et al. Piecing together the past: statistical insights into paleoclimatic reconstructions. Quaternary Science Reviews 35, 1- 22 (2012). + +1032 [120] Amrhein, D. E., Hakim, G. J. & Parsons, L. A. Quantifying structural uncertainty in paleoclimate data assimilation with an application to the last millennium. Geophysical Research Letters 47 (22), e2020GL090485 (2020). + +1035 [121] Franke, J., Valler, V., Brönnimann, S., Neukom, R. & Jaume- Santero, F. The importance of input data quality and quantity in climate field reconstructions- results from the assimilation of various tree- ring collections. Climate of the Past 16 (3), 1061- 1074 (2020). + +1039 [122] Osman, M. B. et al. Globally resolved surface temperatures singe the last glacial maximum. Nature 599, 239- 244 (2021). + +<--- Page Split ---> + +Trends and variability in the Southern Annular Mode over the Common Era 29 + +[123] Perkins, W. A. & Hakim, G. J. Reconstructing paleoclimate fields using online data assimilation with a linear inverse model. Climate of the Past 13 (5), 421–436 (2017). + +<--- Page Split ---> + +Springer Nature 2021 LATE[X template] + +Table 1: Reconstruction Skill metrics for the Southern Annular Mode calculated against instrumental indices over the given time periods + +
MetricMarshall Index (1958-2000)Fogt Index (1958-2000)Fogt Index (1866-2000)
Correlation (p ≪ 0.001)0.720.670.56
RMSE1.451.561.80
σ Ratio0.971.031.15
Mean Bias-0.260.45-0.29
+ +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1: Map of the proxy network. Black Xs indicate the centroid of binned drought atlas sites. Grey markers indicate PAGES2k sites. The size of the PAGES2k markers correspond to the length of each record. Filled color contours show the field correlation between the SAM index and DJF sea level pressure from 20CR [88, 89] over the period 1958-2000 CE.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: Evolution of the reconstruction over time. Top: Comparison of the annual reconstruction (blue) with the Marshall index (red) over the instrumental era. Shading indicates the 5-95 percentiles of the reconstruction. Middle: Evolution of the annual reconstruction (blue) and 31-year lowpass filtered (black) over the Common Era. Shading indicates the 5-95 percentiles of the lowpass filtered series. Bottom: Composition of the proxy network over time. Colors for proxy types are as follows: Dark blue (coral), pink (ANZDA), grey (SADA), dark red (trees), light blue (glacier ice), black (lake sediment).
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: Optimal Sensor Analysis. Top: Maps of the potential ability for drought atlas (left) and PAGES2k (right) sites to constrain reconstruction posterior variance. Middle: Evolution of reconstruction posterior variance over time. Yellow bars indicate the addition of the indicated proxy to the network. Bottom: Ranked histograms of the seven sites with greatest potential influence in the early reconstruction (left), immediately before the addition of drought atlases (middle) and for the full network (right). Potential influence is determined as the uncertainty constrained by a single-proxy network.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4: SAM climate responses. Top left: Wavelet coherence of the reconstructed SAM index with the solar forcing reconstruction [112]. Bottom left: Composite mean response to major volcanic eruptions. Shading indicates 5-95 percentiles. Blue line is the ensemble mean. Right: Instrumental SAM trends for the Marshall Index (top right) and reconstruction (bottom right). Colored points indicate trends calculated from a sliding window centered on the given year. Solid (dotted) contours surround statistically significant trends at the \(90\%\) confidence interval relative to the reconstruction over the period 1500-1900 CE (1-1900 CE).
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5: Comparison of SAM reconstructions over the last millennium. All reconstructions are smoothed via a 30-year Gaussian filter and normalized to the period 1400-1850 CE.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192_det.mmd b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..782325ba07f9597c706bb04586c7eefa58fcc0be --- /dev/null +++ b/preprint/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192/preprint__11d679e272b4101343c10f45f7fc62ffe52a6fab3a156e0d5858dbdbac38c192_det.mmd @@ -0,0 +1,777 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 950, 174]]<|/det|> +# Trends and variability in the Southern Annular Mode over the Common Era + +<|ref|>text<|/ref|><|det|>[[44, 195, 490, 216]]<|/det|> +Jonathan King ( jonking93@email.arizona.edu ) + +<|ref|>text<|/ref|><|det|>[[52, 219, 238, 237]]<|/det|> +University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 244, 275, 282]]<|/det|> +Kevin Anchukaitis The University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 289, 258, 328]]<|/det|> +Kathryn Allen University of Tasmania + +<|ref|>text<|/ref|><|det|>[[44, 336, 258, 374]]<|/det|> +Tessa Vance University of Tasmania + +<|ref|>text<|/ref|><|det|>[[44, 381, 620, 422]]<|/det|> +Amy Hessl West Virginia University https://orcid.org/0000- 0002- 2279- 6260 + +<|ref|>text<|/ref|><|det|>[[44, 464, 102, 481]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 502, 137, 520]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 539, 320, 558]]<|/det|> +Posted Date: August 17th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 577, 474, 596]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1958191/v1 + +<|ref|>text<|/ref|><|det|>[[42, 614, 910, 657]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[204, 170, 792, 233]]<|/det|> +# Trends and variability in the Southern Annular Mode over the Common Era + +<|ref|>text<|/ref|><|det|>[[205, 260, 789, 305]]<|/det|> +Jonathan King \(^{1,2*}\) , Kevin J. Anchukaitis \(^{1,2,3}\) , Kathryn Allen \(^{4,5,6}\) , Tessa Vance \(^{7}\) and Amy Hessl \(^{8}\) + +<|ref|>text<|/ref|><|det|>[[140, 315, 857, 641]]<|/det|> +\(^{1}\) Department of Geosciences, University of Arizona, Tucson, AZ 85721 USA. \(^{2}\) Laboratory of Tree- Ring Research, University of Arizona, Tucson, AZ 85721 USA. \(^{3}\) School of Geography, Development, and Environment, University of Arizona, Tucson, AZ 85721 USA. \(^{4}\) School of Geography, Planning and Spatial Sciences, University of Tasmania, Hobart 7001. \(^{5}\) School of Ecosystem and Forest Sciences, University of Melbourne, Richmond, VIC Australia 3121. \(^{6}\) Centre of Excellence for Australian Biodiversity and Heritage, University of New South Wales, Australia. \(^{7}\) Australian Antarctic Program Partnership, Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, Australia. \(^{8}\) Department of Geology and Geography, West Virginia University, Morgantown, WV USA. + +<|ref|>sub_title<|/ref|><|det|>[[459, 711, 537, 726]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[167, 732, 830, 897]]<|/det|> +The Southern Annular Mode (SAM) is the leading mode of atmospheric variability in the extratropical Southern Hemisphere and has wide ranging effects on ecosystems and societies. Despite the SAM's importance, paleoclimate reconstructions disagree on its variability and trends. Here, we use data assimilation to reconstruct the SAM over the last 2000 years using temperature and drought- sensitive climate proxies. Our method does not assume a stationary relationship between proxy records and the SAM over an instrumental calibration period, so our reconstruction is less sensitive to the teleconnection variability that has hindered previous reconstructions. Our approach also allows us to identify critical paleoclimate records and quantify reconstruction + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[166, 82, 830, 133]]<|/det|> +uncertainty through time. We find no evidence for a forced response in SAM variability prior to the 20th century. We also find the modern positive trend is outside the range of the prior 2000 years, but only on multidecadal time scales. + +<|ref|>sub_title<|/ref|><|det|>[[112, 189, 288, 211]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[111, 224, 884, 585]]<|/det|> +The Southern Annular Mode (SAM) is the leading mode of atmospheric variability in the extratropical Southern Hemisphere and is characterized by a mostly zonally- symmetric mass oscillation with anti- correlated pressure anomalies over the mid- latitudes and Antarctica [1- 4, and see Figure 1]. The SAM's phases capture the strength and position of the mid- latitude westerly winds and the subtropical jet, such that positive phases promote a poleward shift of storm tracks and intensification of the circumpolar westerly belt, while negative phases promote an equator- ward shift of storm tracks and weakening of the westerly winds. Variability in the SAM therefore has wide ranging effects across the Southern Hemisphere. Positive phases of the SAM are linked to cooling over Australia and central Antarctica, as well as warming over the Antarctic Peninsula and southern South America [5- 11]. Hydroclimate effects of the positive phase include drying over southern South America, western South Africa, southern Australia, and New Zealand, as well as increased precipitation over central and eastern Australia and southeastern South America [7, 10, 12- 16]. SAM variations have also been linked to wildfire activity in South America and south- east Australia [17- 21], changes in sea ice distribution [8, 22- 25], and ocean- atmosphere carbon exchange [26- 28]. Understanding SAM variability is therefore important for both societies and ecosystems throughout the Southern Hemisphere, particularly in sub- tropical- temperate regions projected to experience a future drying climate. + +<|ref|>text<|/ref|><|det|>[[111, 585, 884, 819]]<|/det|> +Since the 1950s, the SAM has exhibited a trend toward a more positive state [4, 29, 30], which is attributed to stratospheric ozone depletion and rising concentrations of atmospheric \(\mathrm{CO_2}\) [31- 35]. This positive trend has potentially contributed to severe droughts, including the Day Zero Cape Town drought [36] and Millennium Drought in Australia [37, 38], as well as increased fire activity [17- 19]. Given these impacts, it is important to place the SAM's recent behavior in a long- term perspective and assess the relative influence of anthropogenic forcing and natural climate variability. In the context of multi- decadal trends, reconstructions spanning multiple centuries are necessary to resolve forced responses from the SAM's internal variability. Instrumental records of the SAM only extend through the mid- 1900s and longer reanalysis- derived indices show low correlations with one another and differences in variability prior to the 1950s [30, 39], so characterizing the SAM's long- term behavior requires paleoclimate reconstructions derived from natural climate archives. + +<|ref|>text<|/ref|><|det|>[[112, 820, 884, 890]]<|/det|> +There are several existing multi- century SAM reconstructions [40, 41, 42] (henceforth, V12, A14, and D18), but they show limited agreement prior to the 1850s [42, 43]. Most indicate a negative phase in the SAM during the late 1400s, but both trends and decadal- scale variability show large discrepancies aside from this feature. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 82, 884, 479]]<|/det|> +There are several potential reasons for these differences. Firstly, all three reconstructions rely on the calibration of proxy records directly with an instrumental SAM index. This implicitly makes two important assumptions for each reconstruction: first, that the relationship of proxy records to local climate variables is stationary over time; and second, that the SAM's teleconnections with local climate variables are stationary and well- represented by the instrumental record. While the first is reasonable and a necessary assumption of most paleoclimate analyses, multiple studies cast doubt on this second point, and regional complexity in the climate response to specific SAM phases further decreases the likelihood of this assumption holding. For instance, even over the instrumental period, SAM exhibits non- stationary connections with precipitation and temperature anomalies in southern South America, Australasia, and the Antarctic Peninsula [13, 44], and many of the proxy records in existing SAM reconstructions come from these areas [43]. Evolving concentrations of greenhouse gases, stratospheric ozone, connections with ENSO, and stochastic climate variability can also affect the SAM's influence on regional climates over multi- decadal time scales [29, 32, 45, 46]. Pseudo- proxy experiments have also shown that non- stationary teleconnections cause reconstruction skill to vary widely with the selection of different calibration windows [47]. This effect is particularly pronounced for proxy networks with fewer than 20 sites, which is common in the early portions of SAM reconstructions. To mitigate such effects, D18 explicitly screened for stationarity in their reconstruction, although this required calibration with a longer and therefore less reliable observational record [48]. + +<|ref|>text<|/ref|><|det|>[[110, 479, 884, 893]]<|/det|> +Differences between SAM reconstructions may also result from the selection of different reconstruction targets and proxy networks. For example, A14 targets an annual SAM index, whereas V12 and D18 target an austral summer (DJF) SAM index. D18 found that annual reconstructions were much more sensitive to the selection of proxy sites and calibration windows and they conclude that annual products may exhibit increased sensitivity to non- stationary teleconnections, which may partly explain the differences between the reconstructions. Additionally, each index has been reconstructed using a different proxy network with a different geographic extent. A14 targets the Drake Passage sector, using a mix of terrestrial proxy types from southern South America as well as Antarctic ice cores. In comparison, V12 targets the Pacific sector, using a network of tree- ring chronologies from South America and New Zealand. D18 uses the most spatially extensive network, including tree- ring records [40], Antarctic ice cores, PAGES2k South American proxies [49], and coral records from the tropical Pacific [50]. Furthermore, A14 utilizes a temperature- sensitive proxy network, while V12 and D18 leverage both temperature and hydroclimate- sensitive proxies. Given the variability of the SAM's teleconnections on regional scales [13, 44], and the climate sensitivities of different proxy types [48], these variations in proxy- network design may further help explain reconstruction differences. It is often difficult to assess the influence and contribution of individual proxy records in multiproxy reconstructions, so the cause of any reconstructed index's behavior are often unclear. This is particularly relevant in the period prior to 1400 CE, when the sparsity of proxy networks leaves the reconstructions vulnerable to the dominant influence of just a few records. Ultimately, as a consequence + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 82, 884, 138]]<|/det|> +of these uncertainties and the differences in existing reconstructions, the evolution of the SAM over the Common Era and its response to external forcing remains poorly constrained [43, 51]. + +<|ref|>text<|/ref|><|det|>[[110, 138, 884, 479]]<|/det|> +To address these uncertainties, here we reconstruct the austral summer (DJF) SAM index over the Common Era at annual resolution using offline paleoclimate data assimilation (DA). DA is a recently developed reconstruction technique that integrates climate proxy records with the dynamical behavior captured by climate models [52, 53]. In brief, DA uses forward or proxy- system models [54, 55] to translate climate model states into the same dimensions or 'space' as a collection of climate proxy records. This allows direct comparison of the model output with the proxy records. The climate model states are then updated to more closely match the proxy records, and a model- derived estimate of climate system covariance is used to propagate the update to reconstruction targets, such as the SAM. DA has recently been used to reconstruct surface air temperature anomalies [56- 59], geopotential height fields [57], the response to volcanic eruptions [60, 61], sea ice extent [62], sea surface temperatures [63], and hydroclimate variables [64]. In this study, we assimilate the PAGES2k temperature- sensitive proxy network [49], the South American Drought Atlas [SADA; 65], and the Australia- New Zealand Drought Atlas [ANZDA; 66] (Figure 1) using a suite of last millennium general- circulation climate models (see Methods and Supplemental Table 1) to reconstruct the austral summer SAM index over the last 2,000 years. Here, we follow Gong et al. [2] and define the SAM index using the difference of zonal- mean pressure anomalies (see Methods, Equation 1). + +<|ref|>text<|/ref|><|det|>[[110, 479, 884, 784]]<|/det|> +In the context of SAM reconstructions, DA offers several additional advantages relative to traditional methods. Firstly, our method does not calibrate proxy records against an instrumental SAM index directly; instead, we calibrate proxy forward models using local climate variables, like temperature and precipitation, near the proxy sites. Consequently, our calibration does not assume stationary SAM teleconnections and only requires the stability of proxy relationships to their local climate. Additionally, we estimate covariance between proxies and the SAM using thousands of years of climate model output. As a result of this, our proxy- SAM relationships are not sensitive to potentially anomalous decadal- or centennial- scale variations in the SAM's behavior. Furthermore, DA is amenable to the use of a range of proxy types as well as gridded climate records with spatial autocorrelation, and we leverage this to incorporate the two existing tree- ring based drought atlases into our reconstruction. Previous work indicates that SAM reconstructions using hydroclimate- sensitive sites are more skillful than those using strictly temperature- sensitive proxy networks [47]. Each drought atlas provides extensive coverage for at least the last five centuries and each incorporates over 150 tree- ring records. They therefore represent a significant source of hydroclimate information available for our reconstruction. + +<|ref|>text<|/ref|><|det|>[[110, 785, 884, 893]]<|/det|> +Finally, our DA method allows us to incorporate an optimal sensor analysis [67] as part of the final reconstruction. Traditionally, optimal sensor analyses have been used to identify ideal regions for future proxy development [67- 70]; however, they can also be applied within a DA framework to quantitatively assess the power of different proxy sites as the overall network evolves through time. We use this to identify the proxy sites that are most likely to drive the reconstruction in each time + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 884, 137]]<|/det|> +step, which helps characterize the reconstruction's overall behavior. This information is particularly useful in the early part of this Common Era reconstruction, when the sparse network size can give high weights to a limited number of records. + +<|ref|>sub_title<|/ref|><|det|>[[113, 189, 215, 210]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[111, 223, 884, 512]]<|/det|> +We assess the skill of our SAM reconstruction relative to the Marshall [4] and Fogt indices [71, 72], two commonly used instrumental SAM indices (see Methods for further details). Before comparing time series, we first normalize the Fogt index and our reconstruction to the Marshall index, such that the mean and variance of the detrended normalized time series match those of the detrended Marshall index over the period 1958- 2000 CE. This places all series in the same unit space while preserving differences in the instrumental trend. Examining skill values (Table 1; Figure 2, upper panel), we find that the reconstruction's correlation with the Marshall index (1958- 2000 CE) is \(r = 0.72(p \ll 0.001)\) , which is comparable to that reported for A14 \((r = 0.75, p \ll 0.001))\) . With respect to the 20th century Fogt index, our reconstruction correlates at \(r = 0.65, p \ll 0.001)\) , somewhat higher than A14 \((r = 0.51, p \ll 0.001))\) . Our RMSE values with the Marshall index (1.45) are similar to, albeit slightly higher than, those reported by D18 (1.32). We emphasize that our reconstruction is not calibrated directly to the SAM index, so the agreement with the Marshall and Fogt indices is not built- in to our reconstruction method and thus represents a more independent skill metric. + +<|ref|>text<|/ref|><|det|>[[111, 512, 884, 727]]<|/det|> +We next characterize the reconstruction's behavior over the last two millennia (Figure 2, center). The reconstruction exhibits minimal evidence for trends over most of the first millennium of the Common Era, although the third and seventh centuries are both marked by increased multidecadal variability as the SAM alternates between negative and positive phases. A more strongly negative anomaly in the early 1000s is followed by a notable 100- year positive trend that concludes with the most positive anomalies outside of the instrumental era. The SAM persists in a positive state until the late 1400s, when it abruptly decreases to strongly negative values. After this event, the index returns to near- zero mean anomalies. It has a peak in the mid- 1700s and begins exhibiting a positive trend in the early 1800s. This trend intensifies in the later half of the 20th century, and the reconstruction ends with the most positive SAM anomalies observed during the Common Era. + +<|ref|>text<|/ref|><|det|>[[111, 728, 884, 909]]<|/det|> +Reconstruction uncertainty ranges from \(\pm 4.5\) anomaly units in the early reconstruction to less than 2.3 after 1500 CE (Figure 2b). We note that, because we use a stationary prior, the reconstruction years are treated as fully independent of one another. While this is common in many reconstruction techniques, it does not represent the reality of the SAM, which exhibits persistence on interannual time scales due to potential connections with the stratosphere [73, 74], tropical variability [75, 76], and external forcing [35, 77]. One consequence of this is that the uncertainty estimates shown here likely overestimate the true reconstruction uncertainty. Overall, uncertainty decreases as the reconstruction approaches the present day, a result of the increasing size of the proxy network (Figure 2, bottom panel). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 82, 884, 280]]<|/det|> +We use our optimal sensor framework to identify which proxies are most responsible for reducing reconstruction uncertainty over time (Figure 3). A proxy's ability to reduce uncertainty corresponds to its influence on the reconstruction, so this analysis also allows us to identify which proxies most strongly influence the reconstruction at a given point in time. The first 900 years of the reconstruction are most strongly affected by the Mt. Read (Tasmania) tree- ring record with additional support from the Plateau Remote, WDC06A, and WDC05A ice cores. At 900 CE, the Oroko (New Zealand) tree ring chronology joins the network and supplants Mt. Read as the most influential record. Two large decreases in reconstruction uncertainty occur in 1400 and 1500 CE, which correspond to the addition of the SADA and ANZDA, respectively. + +<|ref|>text<|/ref|><|det|>[[110, 281, 884, 677]]<|/det|> +Examining the reconstruction's response to external forcing, we find no coherence with the solar forcing series and no significant common response to major volcanic eruptions (Figure 4). By contrast, the reconstruction exhibits significant positive trends in the latter half of the twentieth century. However, these modern trends are only significant on time- scales greater than approximately 40 years; trends over shorter time scales fall within the reconstructed range of trends from natural variability. Here, we define natural variability using the distribution of trends in periods of the pre- industrial reconstruction. Examining the Marshall Index, we similarly find that trends shorter than about 35 years are within the reconstructed range of natural variability, but that trends longer than about 35 years fall outside this range. The Marshall Index exhibits its most positive, significant trends for intervals centered on the early 1980s. Although this period is near the end of our reconstruction and less well resolved than preceding decades, we note that the reconstruction similarly exhibits strongly- positive, significant trends centered on the early 1980s. Here we have quantified natural variability using the distribution of reconstructed trends over the period 1500- 1900 CE, the years including both drought atlases. If we instead use the period of the full reconstruction (1- 1900 CE), the tests become more stringent. Significant trend in the reconstruction is limited to the last 55- 80 year interval, and Marshall Index trends are only significant when containing the interval 1964- 2000 CE. We also experiment with using the early portion of the reconstruction (1- 899 CE) to quantify natural variability and find these results are similar to those using the full reconstruction period (Supplemental Figure 1). + +<|ref|>sub_title<|/ref|><|det|>[[111, 711, 258, 733]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[110, 746, 884, 908]]<|/det|> +Our reconstruction suggests that the SAM is dominated by internal variability at least throughout the pre- industrial Common Era. This finding is in agreement with D18, who likewise found minimal influence of solar and volcanic forcing on their reconstruction. Volcanic signals have likewise been a challenge to detect in Southern Hemisphere temperature reconstructions [78]. Some studies have proposed that that an observed relationship between SAM and ENSO [41, 76, 79- 81] could provide a pathway for solar forcing [82] to influence the SAM [43, 83]; however, our results do not support this mechanism during the the Common Era. In a set of model simulations, Wright et al. [83] found that increasing the amplitude of the prescribed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 884, 173]]<|/det|> +solar variability lead to a significant relationship between solar forcing and the simulated SAM. These authors suggest that using high amplitude solar forcing could help reconcile SAM reconstructions with climate simulations; however, the lack of solar signals in our reconstruction differs notably from their findings and instead further supports the realism of low- amplitude solar forcing scenarios [84–86]. + +<|ref|>text<|/ref|><|det|>[[112, 173, 884, 496]]<|/det|> +By contrast with solar and volcanic forcings, our analysis indicates that the most recent multi- decadal trend is outside the range of natural variability and reflects the SAM's response to anthropogenic forcing. We emphasize that this modern trend is only significant for intervals longer than about 40 years when assessed against the 1500- 1900 CE period, or intervals of about 55 years when considering the full Common Era. Shorter trend periods remain within the range of natural variability, even for the most recent intervals. The significance of the modern positive trend therefore reflects its anomalous persistence, rather than the amplitude of its decadal- scale variation alone. The significance of these longer trends emphasizes the importance of the paleoclimate record, particularly given the uncertainties in instrumental SAM records prior to the late twentieth century [39, 87]. We also note that the modern positive trend is only outside of the range of natural variability for trends spanning the years from about 1940- 2000 CE. Trends are generally not significant during the early 1900s, and are even negative for the 50 year period centered on the 1930s. These results help establish the onset of the modern positive trend at around 1940 CE. This timing coincides with increasing emissions of ozone- depleting substances and greenhouse gasses, and is consistent with literature attributing the modern trend to stratospheric ozone depletion and rising levels of atmospheric \(\mathrm{CO_2}\) [31–35]. + +<|ref|>text<|/ref|><|det|>[[111, 496, 884, 821]]<|/det|> +We next compare our reconstruction with the V12, A14, and D18 products (Figure 5). We normalize the mean and variance of each index over the period 1400- 1850 CE to allow comparison of the series in the same unit space. We select the year 1400 CE because it is the first year with values for all four reconstructions and we end the normalization in 1850 CE to limit the sensitivity of our comparison to differing representations of the post- industrial trend. All four indices agree on the existence of a strong positive trend during the late twentieth century; however, all show limited coherence with one another prior to about 1850 CE, as noted in previous studies [42, 43]. The limited agreement of these reconstructions reduces confidence in the significance of modern trends [51], and the causes of these discrepancies include differing seasonal expressions, different proxy networks, and the relative weights of proxies within those networks. Additionally, V12, A14, and D18 all rely on calibration with the instrumental SAM index, which can cause uncertainty when there is non- stationarity in the teleconnection of local climate with the SAM. Ultimately, our reconstruction does not solve the problem of differing reconstructions and similarly shows limited agreement with all of V12, A14, and D18. However, our assimilation does not rely on calibration with the SAM index, and offers a potential improvement by reducing uncertainty from non- stationary teleconnections. + +<|ref|>text<|/ref|><|det|>[[112, 821, 882, 893]]<|/det|> +An additional advantage of our reconstruction is the transparency provided by the optimal sensor's assessment of the relative weights and influence of proxy records in our network. In general, we find that our reconstruction is most strongly influenced by the two drought atlases, followed by the Mt. Read (Tasmania), Oroko (New + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 81, 884, 388]]<|/det|> +Zealand), and Pink Pine (New Zealand) tree ring chronologies, and also the Plateau Remote, Siple Station, WDC06A, and WDC06B ice cores. We note here that a minor change in reconstruction uncertainty does not imply that a proxy has a weak effect on the reconstruction, because highly influential proxies from the same location may present redundant climate signals. For example, the Pink Pine chronology is the third most potentially influential PAGES2k record (Figure 3, lower right), but has a relatively small effect on reconstruction uncertainty when added to the network in 1457 CE (Figure 3, center). This is because much of the Pink Pine climate signal is already represented by the nearby Oroko site. However, such redundant sites are valuable because they make the reconstruction less sensitive to non- climatic noise from a single highly- influential proxy record. In the case of Pink Pine and Oroko, spreading the southern New Zealand climate signal over two influential records allows either site to partially correct for non- climatic noise in the other. A proxy's potential influence reflects both its covariance with the SAM and the ability of our proxy estimates to accurately estimate the record. Ultimately, assuming our estimates of climate covariance are accurate, the influential sites are those most likely to contribute skill to the reconstruction. + +<|ref|>text<|/ref|><|det|>[[111, 388, 884, 677]]<|/det|> +Overall, we find that tree- ring chronologies from Tasmania and New Zealand, the West Antarctic ice cores, and the drought atlas locations in Tasmania, southern New Zealand, the eastern edge of Australia, and southeast South America all have the greatest potential for reconstructing SAM (Figure 3, upper panels). This suggests that additional proxy development in these regions, or extensions of shorter existing records such as the Oroko and Pink Pine tree- ring chronologies or the Siple Station ice core, would be valuable for improving the skill of future SAM reconstructions. However, we caution that location alone is not sufficient for proxy utility and that future proxy development must demonstrate a robust sensitivity to local climate that connects them to the SAM. We also note that, in our optimal sensor framework, a proxy's potential influence is a function of (1) the accuracy of our forward (proxy system) models, and (2) the covariance of the resulting proxy estimates with the SAM in the climate models. As a result, our analysis may currently undervalue proxies from regions with limited climate model agreement, and future improvements in both climate and proxy system models may allow paleoclimate data from other regions to contribute to skillful reconstructions of the SAM. + +<|ref|>sub_title<|/ref|><|det|>[[113, 711, 280, 733]]<|/det|> +## Conclusions + +<|ref|>text<|/ref|><|det|>[[112, 746, 884, 908]]<|/det|> +Our study provides the first reconstruction of the Southern Annular Mode at annual resolution over the entire Common Era. We use a data assimilation method that does not calibrate the proxies directly against the instrumental SAM index, so the reconstruction is not sensitive to observed SAM non- stationarity in the modern era. Our reconstruction leverages both the SADA and ANZDA in addition to the PAGES2k proxy network and represents a significant increase in paleoclimate information available to reconstruct the SAM. Optimal sensor analysis indicates that the first 1400 years of the reconstruction are strongly influenced by the Oroko and Mt. Read tree- ring chronologies, with additional support from the Plateau Remote, WDC06A, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 82, 884, 137]]<|/det|> +WDC05A ice cores. As the SADA and ANZDA are added to the proxy network (1400 CE and 1500 CE, respectively), the drought atlases become strong drivers of the reconstruction's behavior. + +<|ref|>text<|/ref|><|det|>[[110, 137, 884, 334]]<|/det|> +Our reconstruction provides a foundation with which to assess the drivers of the SAM's behavior over the Common Era; such assessments are critical given the SAM's importance to societies and effects on climate variability throughout the Southern Hemisphere. Although our index and existing SAM reconstructions show limited agreement with one another, all products exhibit the most strongly positive and persistent SAM trend during the last several decades. We find that the modern positive trend in the SAM is outside the range of natural variability over the previous millennium, further confirming a response to anthropogenic forcing. Prior to the most recent decades, we find no relationship between SAM variability and external climate forcing, suggesting that its behavior is dominated by internal variability over the pre- industrial Common Era. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[111, 78, 433, 101]]<|/det|> +## Methods and Materials + +<|ref|>sub_title<|/ref|><|det|>[[111, 114, 476, 135]]<|/det|> +## Southern Annular Mode Index + +<|ref|>text<|/ref|><|det|>[[111, 143, 746, 162]]<|/det|> +In this study, we use the Gong et al. [2] definition of the SAM index: + +<|ref|>equation<|/ref|><|det|>[[388, 180, 881, 202]]<|/det|> +\[\mathrm{SAM} = P_{40^{\circ}\mathrm{S}}^{*} - P_{65^{\circ}\mathrm{S}}^{*} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[111, 206, 884, 494]]<|/det|> +where \(P_{X}^{*}\) indicates the normalized zonal- mean sea level pressure (SLP) at a particular latitude. The latitudes \(40^{\circ}\mathrm{S}\) and \(65^{\circ}\mathrm{S}\) were selected as the zonal- means with the most strongly anti- correlated SLP anomalies across the mid- and high- latitude Southern Hemisphere. We use this definition, as opposed to an index derived from a principal component analysis because the latitudes of the most strongly anticorrelated SLP anomalies are robust across the climate models considered in our assimilation (Supplemental Tables 1, 2). We target the austral summer (DJF; December - February) SAM because this corresponds to the seasonality of the climate response of the majority of our proxy network. D18 also suggests that summer SAM reconstructions are more robust to proxy network design than annual reconstructions, which further supports this choice. When calculating the SAM index, we normalize seasonal mean values, rather than individual months. Austral summers span months from two calendar years, and this can introduce date ambiguities for annual records, particularly tree- ring chronologies. Throughout this paper, we use the convention that the year of an austral summer value matches the calendar year of the associated January. + +<|ref|>sub_title<|/ref|><|det|>[[111, 515, 282, 534]]<|/det|> +## Data Products + +<|ref|>sub_title<|/ref|><|det|>[[111, 544, 505, 563]]<|/det|> +## Reanalysis and Instrumental Indices + +<|ref|>text<|/ref|><|det|>[[111, 572, 884, 824]]<|/det|> +We use monthly precipitation and air- temperature fields from the Twentieth Century Reanalysis V3 [20CR; 88, 89] to calibrate our DA method. The 20CR is based on an 80- member ensemble Kalman Filter, and extends from 1850 CE to present at 2 degree resolution. Because of its role in our assimilation method, this effectively sets an upper bound on the resolution of any gridded spatial product used in this reconstruction. We also use the austral summer Marshall Index[4] and Fogt Index [71, 72] to assess the skill of our reconstruction in the modern era. The Marshall Index estimates the Gong et al. [2] definition of the SAM (Equation 1), and is based on data from 12 weather stations (6 near \(40^{\circ}\mathrm{S}\) , and 6 near \(65^{\circ}\mathrm{S}\) ). Because it uses station data, the Marshall index is not subject to the spurious trends observed in high- latitude Southern Hemisphere reanalysis pressure fields [4]. The Fogt index is constructed using a principal component regression of station pressure data and calibrated to the Marshall index. These indices are commonly used as a comparison point for SAM reconstructions[40- 42], + +<|ref|>sub_title<|/ref|><|det|>[[111, 844, 285, 863]]<|/det|> +## Climate Proxies + +<|ref|>text<|/ref|><|det|>[[111, 872, 884, 908]]<|/det|> +In this reconstruction, we assimilate the PAGES2k temperature- sensitive proxy network [49], the South American Drought Atlas (SADA) [65], and the Australia- New + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 81, 884, 208]]<|/det|> +Zealand Drought Atlas (ANZDA) [66]. We limit all three datasets to those sites or locations south of \(25^{\circ}\mathrm{S}\) . Pseudo- proxy tests of other latitude bounds suggests that reconstruction skill is minimally affected by the use of more northward proxy sites and agreement with the instrumental record exhibits a slight maximum for a bound at \(25^{\circ}\mathrm{S}\) (Supplemental Figure 2). Overall, this domain maximizes the number of SAM- sensitive proxy sites in our network, while minimizing the effects of distal proxies that primarily reflect other climate signals. + +<|ref|>text<|/ref|><|det|>[[111, 208, 884, 550]]<|/det|> +From the PAGES2k dataset, we include all sites from the PAGES2k global temperature reconstruction that have annual or sub- annual temporal resolution. To maintain a common timescale, we bin all sub- annual sites to annual resolution. Our PAGES2k network therefore consists of 40 proxy records: 12 tree- ring chronologies, 3 lake sediment cores, 5 corals, 19 ice- cores, and 1 borehole- derived temperature reconstruction (Supplemental Table 3). The tree- ring records are from Tasmania, New Zealand, and the central Andes. The longest two chronologies are from Mt. Read, Tasmania and Oroko, New Zealand, which begin in 494 BCE and 900 CE, respectively; the remaining tree chronologies mostly begin between 1450 CE and 1550 CE. The three lake sediment proxies are derived from the central and southern Andes. The longest record (Laguna Chemical) spans the complete Common Era, while Lagunas Escondida and Aculeo begin in 400 CE and 816 CE, respectively. The five coral records are from the Houtman Abrolhos Islands off the west coast of Australia and begin between 1795 CE and 1900 CE. The Antarctica ice core records have varying temporal coverage. Four sites cover the full Common Era (Plateau Remote, WDC06A, James Ross Island, WAIS- Divide), six more extend at least one millennium, and the remaining nine begin between 1140 CE and 1703 CE. The borehole reconstruction is from WAIS- Divide and begins in 8 CE. For the 40 proxy set, full coverage extends from 1903 CE to 1983 CE with 20 sites remaining by 2000 CE. + +<|ref|>text<|/ref|><|det|>[[111, 550, 884, 874]]<|/det|> +The SADA and ANZDA are gridded tree- ring reconstructions of the self- calibrated Palmer Drought Severity Index (PDSI) during austral summer at annual resolution [65, 66]. The SADA is derived from 286 temperature and precipitation- sensitive tree- ring chronologies and begins in 1400 CE. The atlas covers all of South America south of \(12.25^{\circ}\mathrm{S}\) at \(0.5^{\circ}\) resolution. Similarly, ANZDA is derived from 176 tree- ring chronologies, as well as one coral record, and begins in 1500 CE. The ANZDA covers Australia east of \(136.25^{\circ}\mathrm{E}\) , and New Zealand, also at \(0.5^{\circ}\) resolution. The SAM is strongly associated with droughts and pluvials in the domains of both atlases [65], supporting their inclusion in our network. Both atlases have significantly higher spatial resolution than the reanalysis data and climate model output used for our reconstruction method. To permit calculations that require the same spatial resolution, we bin both atlases to the lowest resolution spatial grid relevant to a given experiment. For the main reconstruction, after applying latitude screening, our SADA and ANZDA networks consist of 104 and 71 binned records, each on a \(2^{\circ} \times 2.5^{\circ}\) grid. It is worth noting that several of the PAGES2k tree ring records used in our reconstruction were also used to construct the drought atlases, and these repeat records might initially appear to duplicate information in the reconstruction. However, our Kalman filter method explicitly accounts for covariance between proxy records, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 118]]<|/det|> +down- weights proxies with repeated information accordingly. Additional details for this process can be found in the following section. + +<|ref|>sub_title<|/ref|><|det|>[[112, 141, 391, 161]]<|/det|> +## Reconstruction Method + +<|ref|>sub_title<|/ref|><|det|>[[112, 170, 268, 189]]<|/det|> +## Kalman Filter + +<|ref|>text<|/ref|><|det|>[[112, 198, 884, 234]]<|/det|> +Our reconstruction uses an ensemble Kalman Filter approach [EnKF; 90], which follows the update equation: + +<|ref|>equation<|/ref|><|det|>[[380, 250, 881, 273]]<|/det|> +\[\mathbf{X}_{\mathbf{a}} = \mathbf{X}_{\mathbf{p}} - \mathbf{K}(\mathbf{Y} - \hat{\mathbf{Y}}) \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[112, 291, 884, 436]]<|/det|> +in each reconstructed time step. Here, the \(\mathbf{X}_{\mathbf{p}}\) and \(\mathbf{X}_{\mathbf{a}}\) matrices are the initial (prior) and updated (analysis) ensembles of climate model states. Each row holds a target climate variable, and each column a different selection of climate model output (ensemble member). \(\mathbf{Y}\) is a matrix of proxy values for the time step; the columns of \(\mathbf{Y}\) are constant, and each row holds the value from a particular proxy record repeated once for each ensemble member. \(\hat{\mathbf{Y}}\) holds the model estimates of the proxy values; each row has the estimates for a particular proxy site, and each column has the estimates from a particular ensemble member. \(\mathbf{K}\) is the Kalman gain: + +<|ref|>equation<|/ref|><|det|>[[334, 451, 881, 475]]<|/det|> +\[\mathbf{K} = \mathrm{cov}(\mathbf{X}_{\mathbf{p}},\hat{\mathbf{Y}})[\mathrm{cov}(\hat{\mathbf{Y}}) + \mathbf{R}]^{-1} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[111, 480, 884, 660]]<|/det|> +where \(\mathbf{R}\) is the matrix of proxy error- covariances. As previously mentioned, the Kalman filter accounts for duplication of information across repeated proxy records. This occurs via the \(\mathrm{cov}(\hat{\mathbf{Y}})\) term in Equation 3, which reduces proxy weights in the Kalman gain as a function of shared proxy covariance. Note that any shared covariance derived from proxies' relationships with the SAM is balanced by the \(\mathrm{cov}(\mathbf{X}_{\mathbf{p}},\hat{\mathbf{Y}})\) term in Equation 3. We use a square- root variant of EnKF [91, 92]. This modifies equations 2 and 3 to update the ensemble mean and deviations separately, and precludes the need for perturbed observations [93]. The Kalman filter can be expressed as a recursive Bayesian filter [94, 95], so we will often refer to \(\mathbf{X}_{\mathbf{p}}\) and \(\mathbf{X}_{\mathbf{a}}\) as the prior and posterior in this paper. + +<|ref|>sub_title<|/ref|><|det|>[[111, 683, 172, 700]]<|/det|> +## Prior + +<|ref|>text<|/ref|><|det|>[[111, 710, 884, 907]]<|/det|> +We construct the prior using output from climate models with paleoclimate simulations of the last millennium (Supplemental Table 1). We use a multi- model ensemble (MME), which has been found to reduce error relative to single model assimilations [59, 96]. Our MME consists of CCSM4, CESM- LME, MPI, and MRI, which represent the set of last millennium simulations with spatial resolutions greater than or at the resolution of the 20CR reanalysis. As such, this selection does not require us to bin the drought atlases to lower resolutions than 20CR, which allows us to extract maximum information from SADA and ANZDA. We also tested a larger MME consisting of 10 models with last millennium simulations regardless of resolution. Our tests show that the high- resolution MME maximizes reconstruction skill (Supplemental Figure 3). For CCSM4, MPI, and MRI, we use output from the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 172]]<|/det|> +PMIP3 last1000 (850- 1850 CE) and historical (1851- 2005 CE) experiments, specifically ensemble member rli1p1. For CESM- LME, we use output from full- forcing run 2 (850- 2005 CE). While the PAGES2k proxy network does include stable oxygen isotope proxies, there are too few high- resolution last millennium isotope- enabled paleoclimate model simulations available to construct a multi- model prior[96]. + +<|ref|>text<|/ref|><|det|>[[111, 172, 884, 405]]<|/det|> +We use an offline, stationary prior for our assimilation. Offline approaches [97, 98] differ from classical Kalman Filters in that updates are not used to inform model simulation. Instead, offline methods use pre- existing model output to build the prior in each time step. The offline approach has been shown to compare favorably with classical (online) methods in paleoclimate contexts but at a fraction of the computational cost [99, 100]. The stationary prior indicates that we use the same ensemble as the prior for each reconstructed time step. This is common in paleoclimate DA applications [52, 57, 101] and is justified by the limited forecast skill of climate models beyond the annual reconstruction time scale [102]. However, stationary priors have been observed to artificially reduce the variability of reconstructions as proxy networks become more sparse [59]. Consequently, our use of stationary priors necessitates a correction for the reconstruction's variability, which is detailed in the methods below. + +<|ref|>text<|/ref|><|det|>[[111, 406, 884, 495]]<|/det|> +To build each prior, we first calculate the DJF SAM time- series for each model, normalizing zonal SLP means to the pre- industrial period (850- 1849 CE). We then concatenate the SAM index time- series from each model in every year of model output. The final prior has a total of 4624 ensemble members from 4 high- resolution models. + +<|ref|>sub_title<|/ref|><|det|>[[111, 515, 616, 535]]<|/det|> +## Proxy Forward Models and Error Covariances + +<|ref|>text<|/ref|><|det|>[[111, 542, 884, 596]]<|/det|> +The proxy modeling process begins by designing a forward model for each assimilated proxy record. For the PAGES2k records, we follow previous studies [53, 59] and use simple univariate linear models: + +<|ref|>equation<|/ref|><|det|>[[435, 611, 880, 633]]<|/det|> +\[\hat{\mathbf{Y}} = a\mathbf{T} + b \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[111, 637, 884, 872]]<|/det|> +where \(\hat{\mathbf{Y}}\) is a vector of proxy estimates, and \(\mathbf{T}\) is a vector of seasonal temperature means. Here, the seasonal means used for each site is taken from the seasonal sensitivity reported in the PAGES2k metadata [49]. We determine the coefficients \(a\) and \(b\) by calibrating each proxy PAGES2k record to the corresponding climate data from 20CR. For each proxy site, we first determine the seasonal sensitivity and then linearly regress the proxy record against the seasonal- mean temperature vector from the closest 20CR grid point in all overlapping years from 1950 - 2000 CE. The regression slope and intercept are then used as coefficients a and b. For the drought atlases, we estimate proxies by calculating PDSI [103] using the Thornthwaite estimation of potential evapotranspiration [104]. This uses monthly mean temperature and precipitation from a drought atlas grid cell to compute monthly PDSI values for each year. We then use the austral summer means of these monthly values as the proxy estimates. Effectively: + +<|ref|>equation<|/ref|><|det|>[[292, 887, 880, 910]]<|/det|> +\[\mathbf{Y} = \mathrm{mean}\big[\mathrm{PDSI}_{\mathrm{Thornthwaite}}(\mathbf{T},\mathbf{P})\big]_{\mathrm{DJF}} \quad (5)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 155]]<|/det|> +where T and P are monthly temperature and precipitation, and Y is the drought atlas estimate. We estimate proxy values for the model priors by applying Equations 4 and 5 to climate model output and matching each year's estimates to the associated ensemble member in the prior. + +<|ref|>text<|/ref|><|det|>[[111, 154, 884, 389]]<|/det|> +Although the PDSI calculation in Equation 5 uses the Thornthwaite approximation, both drought atlases target an observational dataset based on the Pennam-Monteith method [65, 66, 105]. However, both the Thornthwaite and Pennam-Monteith equations have been shown to perform similarly when applied to preindustrial simulations, and this agreement occurs because the simplifying assumptions of the Thornthwaite method remain valid over the relatively confined range of last millennium temperatures [106]. For the purposes of this study, the Thornthwaite method provides two further advantages: First, the Thornthwaite equation is more computationally tractable, which allows us to apply it to the large spatial regions and the multiple millennium- length climate model simulations used for priors in our assimilation method. Second, because the Thornthwaite calculation requires fewer climate model data fields to estimate the PDSI [104, 105], opportunities for climate model biases to degrade the reconstruction are reduced. + +<|ref|>text<|/ref|><|det|>[[110, 388, 884, 750]]<|/det|> +We next estimate the proxy error covariances. These error covariances describe the uncertainty in the comparison of observed records to the proxy estimates \((\mathbf{Y} - \mathbf{Y}\mathbf{e})\) . In a classical Kalman Filter, the estimates \((\hat{\mathbf{Y}})\) are known perfectly and this uncertainty is derived from the observations \((\mathbf{Y})\) , so \(\mathbf{R}\) is often referred to as observation uncertainty. In paleoclimate contexts, this situation is inverted: proxy measurements are typically precise and uncertainty derives from the simplifications and parameterizations inherent in the estimation equations. Hence, we quantify \(\mathbf{R}\) by running Equations 4 and 5 on the 20CR dataset (from 1950- 2000 CE) and comparing the estimated proxy values to the real records. The differences between the two sets of values are used to estimate the errors inherent in using simple models and relatively coarse climate data to estimate the temporal behavior of the proxy records. Most EnKF paleoclimate efforts assume that proxy errors are independent, such that \(\mathbf{R}\) is a diagonal matrix [52, 53, 57, 64]. This is justified for datasets like PAGES2k, for which proxy uncertainties are dominated by local biological, physical, and mechanistic effects [48]. However, the drought atlas grid points are strongly spatially correlated, so this assumption is not appropriate in this study. Instead, we calculate independent error- variances for the proxies in the PAGES2k network, and full errorcovariances for both SADA and ANZDA. Hence, \(\mathbf{R}\) is block- diagonal, rather than strictly diagonal. We estimate uncertainty in the final reconstruction from the spread of the assimilation posterior. + +<|ref|>sub_title<|/ref|><|det|>[[111, 790, 333, 809]]<|/det|> +## Variance Correction + +<|ref|>text<|/ref|><|det|>[[110, 817, 883, 910]]<|/det|> +The use of stationary priors creates artifacts in the variability of raw reconstruction. As the proxy network becomes sparse, less information is incorporated in the Kalman Filter, and the updated state is less able to move off the prior mean. This causes reconstruction variability to increase with the size of the proxy network and independently of the climatological record. We apply a variance adjustment scheme + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 884, 138]]<|/det|> +to correct for this effect. Variance adjustments are common in paleoclimate reconstructions [107–110] and are inherent to simpler methods like Composite Plus Scale [111]. + +<|ref|>text<|/ref|><|det|>[[112, 138, 884, 300]]<|/det|> +Here, we use a series of frozen- network assimilations to adjust temporal variance. There are five sites in our proxy network with observations in every year of the reconstruction. We first assimilate this five- site network over the full interval 1- 2000 CE to derive a baseline time- series that is not affected by changes to the proxy network. We next determine each unique set of proxy sites used to update one or more time steps in the reconstruction. We then assimilate each set of proxies over the time steps for which all of the proxies have recorded values, and determine the ratio of this assimilation's standard deviation to that of the baseline time series over all overlapping years: + +<|ref|>equation<|/ref|><|det|>[[384, 317, 881, 337]]<|/det|> +\[P(\mathrm{set}) = \sigma_{\mathrm{set}} / \sigma_{\mathrm{Baseline}} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[112, 340, 884, 376]]<|/det|> +We then calculate a scaling factor for each time step using the normalized ratio for the associated proxy set: + +<|ref|>equation<|/ref|><|det|>[[369, 391, 881, 412]]<|/det|> +\[w(t) = P(\mathrm{set}(t)) / \mathrm{max}(P) \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[111, 415, 884, 452]]<|/det|> +A comparison of the raw and variance- adjusted reconstructions is provided in Supplemental Figure 4. + +<|ref|>sub_title<|/ref|><|det|>[[112, 470, 404, 491]]<|/det|> +## Optimal Sensor Analysis + +<|ref|>text<|/ref|><|det|>[[111, 499, 884, 662]]<|/det|> +We follow a previously established framework for optimal sensor analyses [67]. In brief, the method quantifies the ability of proxy sites to reduce the variance of a metric in a posterior ensemble. Here, we use the SAM as our metric, so the optimal sensor analysis here assesses the ability of sites to reduce uncertainty in the index across the reconstruction posterior. We first compute the total reduction in SAM posterior variance using the complete set of proxies with observations in each time step. We also quantify each site's ability to reduce reconstruction uncertainty when no other sites are in the proxy network. We refer to this quantity as 'potential percent constrained variance'. + +<|ref|>sub_title<|/ref|><|det|>[[111, 681, 425, 702]]<|/det|> +## External Forcing Analyses + +<|ref|>text<|/ref|><|det|>[[110, 710, 884, 908]]<|/det|> +We begin our external forcing analysis by investigating the SAM's response to natural climate forcings. We first use a wavelet coherence analysis to examine the relationship between our SAM reconstruction and a time series of reconstructed solar forcing [110, 112]. We next use a superposed epoch analysis [113] to determine the reconstruction's composite mean response to major volcanic eruptions. We used the eVolv2k V3 volcanic forcing dataset [114, 115] to select events with a total forcing magnitude greater than or equal to that of Krakatoa. This yielded 28 eruption years: 87, 169, 266, 433, 536, 540, 574, 626, 682, 817, 939, 1108, 1171, 1182, 1230, 1257, 1276, 1286, 1345, 1458, 1600, 1640, 1695, 1783, 1809, 1815, 1831, and 1883. For the SEA, we normalized each event to the mean of the preceding 5 years and examined the composite mean response over the 10 years following volcanic events. We tested + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 884, 118]]<|/det|> +the significance of the observed response by bootstrapping 5,000 SEA time series via random draws of 28 event years from the remaining years in the reconstruction. + +<|ref|>text<|/ref|><|det|>[[111, 119, 884, 424]]<|/det|> +We next consider the SAM's response to anthropogenic forcings using both our reconstruction and the Marshall index. Before quantifying trends, we first normalize our reconstruction to the Marshall index, such that the mean and variance of the detrended normalized reconstruction matches those of the detrended Marshall index over the years of common overlap (1958- 2000 CE). This places the series in the same unit space while preserving differences in the instrumental trend. We then calculate moving trends for the reconstruction over the years 1900- 2000 CE using trend window lengths from 31 to 101 years. Similarly, we calculate moving trends for the Marshall Index over the years 1958- 2020 CE using trend window lengths from 31 to 63 years. We then use the reconstruction to assess the significance of these trends. For each trend window length, we calculate the distribution of trends with the given window length from the reconstruction over the years 1500- 1900 CE, and we define this distribution as the natural variability for that trend length. We then use the 90% confidence intervals of each distribution to determine a significance threshold for trends of the associated length. We also repeat this process using trend distributions from the intervals 1- 1900 CE and 1- 899 CE to examine the sensitivity of this analysis to different portions of the reconstruction. + +<|ref|>sub_title<|/ref|><|det|>[[113, 447, 400, 467]]<|/det|> +## Caveats and Limitations + +<|ref|>text<|/ref|><|det|>[[110, 475, 884, 800]]<|/det|> +Our DA method does not require a calibration with the instrumental SAM, which limits sensitivity to non- stationarity in the SAM during the instrumental era. However, the trade- off is the influence of proxy forward model and climate model biases on the reconstruction. In the case of proxy models, any biases typically reduce the weight of the proxy in the assimilation, thereby limiting its effect on the reconstruction. We note that improving the accuracy or sophistication of the proxy forward models could increase the influence of many records; for example, transitioning the statistical forward models used here for the PAGES2k sites to more mechanistically accurate proxy system models [54] could potentially improve the reconstruction [101]. However, efforts to develop more complex proxy system models must also exercise caution, as excessive complexity and poorly constrained parameters may lead to overfitting and artificially high skill in the instrumental era at the expense of accuracy during the earlier reconstruction. In this study, we retain the simpler statistical forward models because (1) the PAGES2k proxies are reported to be temperature sensitive [49], (2) statistical proxy models remain the most common and tractable approach for paleoclimate data assimilation to date [53, 57, 59], and (3) the simple statistical model eliminates errors caused by the interaction of climate model biases with forward models that rely on absolute units. + +<|ref|>text<|/ref|><|det|>[[110, 800, 884, 908]]<|/det|> +With respect to climate models, biases in the mean state can affect proxy estimates that include parametrizations or thresholds based on absolute units. However, covariance biases are a greater concern, as they introduce errors in the propagation of information from the proxy records to the reconstruction target. For example, some of the climate models considered in this study simulate a SAM pattern that is too zonally symmetric and that overestimates the SAM's influence on overall Southern + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 82, 884, 280]]<|/det|> +Hemisphere circulation [116]. Such teleconnection biases can cause the assimilation to overestimate the covariance between various proxies and the SAM, thereby increasing reconstruction error. In this study, we use a multi- model ensemble (MME) to reduce the effects of covariance bias from any one model [59, 96]. We note that we weight each model equally, which effectively treats each model as independent. In reality, many models share common features or code, so this equal weighting may bias an ensemble towards the most similar models [117, 118]. For example, the CCSM4 and CESM- LME output used in our MME are both from models developed by the US National Center for Atmospheric Research (NCAR) and may more closely resemble one another than the MPI or MRI models. Future efforts may wish to test different model composition and weights when constructing a MME prior. + +<|ref|>text<|/ref|><|det|>[[111, 281, 884, 440]]<|/det|> +Finally, our use of a stationary offline prior implies a stationary estimate of climate system covariance when considered over the full reconstruction period. Although we use a long- term estimate of the SAM's climate covariance, the true covariance may vary on multi- decadal scales [13, 44], and these variations will not be captured in our approach. While the assumption of a reasonably stationary covariance is implicitly common to most spatial reconstruction methods [119, 120], the application of transient offline priors [102, 121, 122] or online assimilation techniques [123] may enhance future data assimilation reconstruction, although these approaches must balance the utility of evolving covariance estimates with reduced ensemble sizes. + +<|ref|>sub_title<|/ref|><|det|>[[113, 466, 364, 489]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[111, 500, 884, 680]]<|/det|> +We thank David Meko for providing the original PDSI estimation code on which ours is based. This research is supported by a grant from the US National Science Foundation's Paleo Perspectives on Climatic Change program (P2C2) AGS- 1803946 to K.J.A. and A.H. T.V. is supported by an Australian Research Council Special Research Initiative for Antarctic Gateway Partnership (SR140300001) and the Australian Antarctic Program Partnership (ASCI000002). K.A. is supported by a grant from the Australian Research Council (FT200100102). We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output. + +<|ref|>sub_title<|/ref|><|det|>[[111, 706, 402, 727]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[111, 741, 884, 830]]<|/det|> +K.J.A., A.H., and J.K. designed the research. J.K. wrote all the software and code, improved and enhanced the assimilation method, conducted all calculations, and produced all the figures. A.H., T.V., and K.A. provided guidance, data, and knowledge for individual proxies and Southern Hemisphere climate variability. All authors (J.K., K.J.A., A.H., T.V. and K.A.) interpreted results and wrote the paper. + +<|ref|>sub_title<|/ref|><|det|>[[111, 855, 388, 878]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[111, 891, 640, 908]]<|/det|> +The authors declare that they have no conflicts of interest. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[111, 78, 470, 101]]<|/det|> +## Data and code availability + +<|ref|>text<|/ref|><|det|>[[111, 113, 884, 168]]<|/det|> +The data and code used to produce this analysis, as well as the final reconstruction, will be made publicly available in a Zenodo repository with a permanent DOI pending review. + +<|ref|>sub_title<|/ref|><|det|>[[111, 188, 460, 211]]<|/det|> +## Supplementary materials + +<|ref|>text<|/ref|><|det|>[[111, 223, 884, 260]]<|/det|> +Supplemental Figure 1: Trend analysis using the early reconstruction (1- 899 CE) to quantify natural variability and trend significance. + +<|ref|>text<|/ref|><|det|>[[111, 277, 884, 313]]<|/det|> +Supplemental Figure 2: Correlation of reconstructions with the Marshall Index as a function of latitude cutoff thresholds. + +<|ref|>text<|/ref|><|det|>[[111, 330, 884, 366]]<|/det|> +Supplemental Figure 3: Comparison of reconstruction skill using different multimodel priors. + +<|ref|>text<|/ref|><|det|>[[144, 384, 860, 404]]<|/det|> +Supplemental Figure 4: Raw and variance- adjusted reconstruction time series. + +<|ref|>text<|/ref|><|det|>[[144, 420, 765, 439]]<|/det|> +Supplemental Table 1: Climate models tested in assimilation priors. + +<|ref|>text<|/ref|><|det|>[[111, 456, 884, 492]]<|/det|> +Supplemental Table 2: Latitudes of maximum negative SLP correlation in climate models and 20CR. + +<|ref|>text<|/ref|><|det|>[[144, 510, 789, 529]]<|/det|> +Supplemental Table 3: PAGES2k sites used in the final reconstruction. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[112, 77, 263, 100]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[131, 110, 885, 168]]<|/det|> +[1] Rogers, J. C. & Van Loon, H. Spatial variability of sea level pressure and 500 mb height anomalies over the southern hemisphere. Monthly Weather Review 110 (10), 1375- 1392 (1982). + +<|ref|>text<|/ref|><|det|>[[131, 181, 885, 219]]<|/det|> +[2] Gong, D. & Wang, S. Definition of antarctic oscillation index. Geophysical Research Letters 26 (4), 459- 462 (1999). + +<|ref|>text<|/ref|><|det|>[[131, 232, 885, 288]]<|/det|> +[3] Thompson, D. W. & Wallace, J. M. Annular modes in the extratropical circulation. part i: Month- to- month variability. Journal of Climate 13 (5), 1000- 1016 (2000). + +<|ref|>text<|/ref|><|det|>[[131, 301, 885, 339]]<|/det|> +[4] Marshall, G. J. Trends in the southern annular mode from observations and reanalyses. Journal of Climate 16 (24), 4134- 4143 (2003). + +<|ref|>text<|/ref|><|det|>[[131, 352, 885, 390]]<|/det|> +[5] Thompson, D. W. & Solomon, S. Interpretation of recent southern hemisphere climate change. Science 296 (5569), 895- 899 (2002). + +<|ref|>text<|/ref|><|det|>[[131, 402, 885, 459]]<|/det|> +[6] Kwok, R. & Comiso, J. C. Spatial patterns of variability in antarctic surface temperature: Connections to the southern hemisphere annular mode and the southern oscillation. Geophysical Research Letters 29 (14), 50- 1 (2002). + +<|ref|>text<|/ref|><|det|>[[131, 472, 885, 510]]<|/det|> +[7] Gillett, N. P., Kell, T. D. & Jones, P. Regional climate impacts of the southern annular mode. Geophysical Research Letters 33 (23) (2006). + +<|ref|>text<|/ref|><|det|>[[131, 523, 885, 580]]<|/det|> +[8] Gupta, A. S. & England, M. H. Coupled ocean- atmosphere- ice response to variations in the southern annular mode. Journal of Climate 19 (18), 4457- 4486 (2006). + +<|ref|>text<|/ref|><|det|>[[131, 592, 885, 649]]<|/det|> +[9] Marshall, G. J., Orr, A., Van Lipzig, N. P. & King, J. C. The impact of a changing southern hemisphere annular mode on antarctic peninsula summer temperatures. Journal of Climate 19 (20), 5388- 5404 (2006). + +<|ref|>text<|/ref|><|det|>[[131, 661, 885, 717]]<|/det|> +[10] Hendon, H. H., Thompson, D. W. & Wheeler, M. C. Australian rainfall and surface temperature variations associated with the southern hemisphere annular mode. Journal of Climate 20 (11), 2452- 2467 (2007). + +<|ref|>text<|/ref|><|det|>[[131, 730, 885, 804]]<|/det|> +[11] Van Lipzig, N. P., Marshall, G. J., Orr, A. & King, J. C. The relationship between the southern hemisphere annular mode and antarctic peninsula summer temperatures: Analysis of a high- resolution model climatology. Journal of Climate 21 (8), 1649- 1668 (2008). + +<|ref|>text<|/ref|><|det|>[[131, 817, 885, 873]]<|/det|> +[12] Silvestri, G. E. & Vera, C. S. Antarctic oscillation signal on precipitation anomalies over southeastern south america. Geophysical Research Letters 30 (21) (2003). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 80, 885, 137]]<|/det|> +[13] Silvestri, G. & Vera, C. Nonstationary impacts of the southern annular mode on southern hemisphere climate. Journal of Climate 22 (22), 6142- 6148 (2009). + +<|ref|>text<|/ref|><|det|>[[67, 150, 885, 206]]<|/det|> +[14] Cai, W. & Cowan, T. Sam and regional rainfall in ipcc ar4 models: Can anthropogenic forcing account for southwest western australian winter rainfall reduction? Geophysical Research Letters 33 (24) (2006). + +<|ref|>text<|/ref|><|det|>[[67, 220, 885, 275]]<|/det|> +[15] Reason, C. & Rouault, M. Links between the antarctic oscillation and winter rainfall over western south africa. Geophysical Research Letters 32 (7) (2005). + +<|ref|>text<|/ref|><|det|>[[67, 289, 885, 345]]<|/det|> +[16] Kidston, J., Renwick, J. & McGregor, J. Hemispheric-scale seasonality of the southern annular mode and impacts on the climate of new zealand. Journal of Climate 22 (18), 4759- 4770 (2009). + +<|ref|>text<|/ref|><|det|>[[67, 358, 885, 396]]<|/det|> +[17] Holz, A. & Veblen, T. T. Variability in the southern annular mode determines wildfire activity in patagonia. Geophysical Research Letters 38 (14) (2011). + +<|ref|>text<|/ref|><|det|>[[67, 409, 885, 465]]<|/det|> +[18] Mariani, M. & Fletcher, M.- S. The southern annular mode determines interannual and centennial- scale fire activity in temperate southwest tasmania, australia. Geophysical Research Letters 43 (4), 1702- 1709 (2016). + +<|ref|>text<|/ref|><|det|>[[67, 478, 885, 534]]<|/det|> +[19] Holz, A. et al. Southern annular mode drives multicentury wildfire activity in southern south america. Proceedings of the National Academy of Sciences 114 (36), 9552- 9557 (2017). + +<|ref|>text<|/ref|><|det|>[[67, 547, 885, 603]]<|/det|> +[20] Abram, N. J. et al. Connections of climate change and variability to large and extreme forest fires in southeast australia. Communications Earth & Environment 2 (1), 1- 17 (2021). + +<|ref|>text<|/ref|><|det|>[[67, 616, 885, 672]]<|/det|> +[21] Lim, E.- P. et al. The 2019 southern hemisphere stratospheric polar vortex weakening and its impacts. Bulletin of the American Meteorological Society 102 (6), E1150- E1171 (2021). + +<|ref|>text<|/ref|><|det|>[[67, 685, 885, 741]]<|/det|> +[22] Hall, A. & Visbeck, M. Synchronous variability in the southern hemisphere atmosphere, sea ice, and ocean resulting from the annular mode. Journal of Climate 15 (21), 3043- 3057 (2002). + +<|ref|>text<|/ref|><|det|>[[67, 754, 885, 828]]<|/det|> +[23] Stammerjohn, S. E., Martinson, D., Smith, R., Yuan, X. & Rind, D. Trends in antarctic annual sea ice retreat and advance and their relation to el niño- southern oscillation and southern annular mode variability. Journal of Geophysical Research: Oceans 113 (C3) (2008). + +<|ref|>text<|/ref|><|det|>[[67, 841, 885, 897]]<|/det|> +[24] Simpkins, G. R., Ciasto, L. M., Thompson, D. W. & England, M. H. Seasonal relationships between large- scale climate variability and antarctic sea ice concentration. Journal of Climate 25 (16), 5451- 5469 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 65, 884, 110]]<|/det|> +[25] Kohyama, T. & Hartmann, D. L. Antarctic sea ice response to weather and climate modes of variability. Journal of Climate 29 (2), 721-741 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 130, 884, 186]]<|/det|> +[26] Lovenduski, N. S., Gruber, N., Doney, S. C. & Lima, I. D. Enhanced co2 outgassing in the southern ocean from a positive phase of the southern annular mode. Global Biogeochemical Cycles 21 (2) (2007). + +<|ref|>text<|/ref|><|det|>[[120, 201, 884, 256]]<|/det|> +[27] Lovenduski, N. S., Fay, A. R. & McKinley, G. A. Observing multidecadal trends in southern ocean co2 uptake: What can we learn from an ocean model? Global Biogeochemical Cycles 29 (4), 416-426 (2015). + +<|ref|>text<|/ref|><|det|>[[120, 270, 884, 325]]<|/det|> +[28] Sallée, J.- B., Speer, K. & Rintoul, S. Zonally asymmetric response of the southern ocean mixed-layer depth to the southern annular mode. Nature Geoscience 3 (4), 273-279 (2010). + +<|ref|>text<|/ref|><|det|>[[120, 340, 884, 394]]<|/det|> +[29] Thompson, D. W., Wallace, J. M. & Hegerl, G. C. Annular modes in the extratropical circulation. part ii: Trends. Journal of Climate 13 (5), 1018-1036 (2000). + +<|ref|>text<|/ref|><|det|>[[120, 409, 884, 464]]<|/det|> +[30] Fogt, R. L. & Marshall, G. J. The southern annular mode: variability, trends, and climate impacts across the southern hemisphere. Wiley Interdisciplinary Reviews: Climate Change 11 (4), e652 (2020). + +<|ref|>text<|/ref|><|det|>[[120, 479, 884, 515]]<|/det|> +[31] Arblaster, J. M. & Meehl, G. A. Contributions of external forcings to southern annular mode trends. Journal of Climate 19 (12), 2896-2905 (2006). + +<|ref|>text<|/ref|><|det|>[[120, 529, 884, 584]]<|/det|> +[32] Thompson, D. W. et al. Signatures of the antarctic ozone hole in southern hemisphere surface climate change. Nature Geoscience 4 (11), 741-749 (2011). + +<|ref|>text<|/ref|><|det|>[[120, 599, 884, 671]]<|/det|> +[33] Polvani, L. M., Waugh, D. W., Correa, G. J. & Son, S.- W. Stratospheric ozone depletion: The main driver of twentieth-century atmospheric circulation changes in the southern hemisphere. Journal of Climate 24 (3), 795-812 (2011). + +<|ref|>text<|/ref|><|det|>[[120, 686, 884, 740]]<|/det|> +[34] England, M. R., Polvani, L. M., Smith, K. L., Landrum, L. & Holland, M. M. Robust response of the amundsen sea low to stratospheric ozone depletion. Geophysical Research Letters 43 (15), 8207-8213 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 755, 884, 791]]<|/det|> +[35] Jones, J. M. et al. Assessing recent trends in high-latitude southern hemisphere surface climate. Nature Climate Change 6 (10), 917-926 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 805, 884, 860]]<|/det|> +[36] Sousa, P. M., Blamey, R. C., Reason, C. J., Ramos, A. M. & Trigo, R. M. The 'day zero' cape town drought and the poleward migration of moisture corridors. Environmental Research Letters 13 (12), 124025 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 79, 884, 915]]<|/det|> +[37] Verdon- Kidd, D. C. & Kiem, A. S. Nature and causes of protracted droughts in southeast australia: Comparison between the federation, wwii, and big dry droughts. Geophysical Research Letters 36 (22) (2009). [38] Cai, W., Van Rensch, P., Borlace, S. & Cowan, T. Does the southern annular mode contribute to the persistence of the multidecade- long drought over southwest western australia? Geophysical Research Letters 38 (14) (2011). [39] Barrucand, M. G., Zitto, M. E., Piotrkowski, R., Canziani, P. & O'Neill, A. Historical sam index time series: linear and nonlinear analysis. International Journal of Climatology 38, e1091- e1106 (2018). [40] Villalba, R. et al. Unusual southern hemisphere tree growth patterns induced by changes in the southern annular mode. Nature Geoscience 5 (11), 793- 798 (2012). [41] Abram, N. J. et al. Evolution of the southern annular mode during the past millennium. Nature Climate Change 4 (7), 564- 569 (2014). [42] Dätwyler, C. et al. Teleconnection stationarity, variability and trends of the southern annular mode (sam) during the last millennium. Climate Dynamics 51 (5), 2321- 2339 (2018). [43] Hessl, A., Allen, K. J., Vance, T., Abram, N. J. & Saunders, K. M. Reconstructions of the southern annular mode (sam) during the last millennium. Progress in Physical Geography 41 (6), 834- 849 (2017). [44] Gallant, A. J., Phipps, S. J., Karoly, D. J., Mullan, A. B. & Lorrey, A. M. Nonstationary australasian teleconnections and implications for paleoclimate reconstructions. Journal of Climate 26 (22), 8827- 8849 (2013). [45] Brönnimann, S. et al. Tropical circulation and precipitation response to ozone depletion and recovery. Environmental Research Letters 12 (6), 064011 (2017). [46] Yun, K.- S. & Timmermann, A. Decadal monsoon- enso relationships reexamined. Geophysical Research Letters 45 (4), 2014- 2021 (2018). [47] Huiskamp, W. & McGregor, S. Quantifying southern annular mode paleoreconstruction skill in a model framework. Climate of the Past 17 (5), 1819- 1839 (2021). [48] Jones, P. D. et al. High- resolution palaeoclimatology of the last millennium: a review of current status and future prospects. The Holocene 19 (1), 3- 49 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 55, 884, 75]]<|/det|> +Trends and variability in the Southern Annular Mode over the Common Era 23 + +<|ref|>text<|/ref|><|det|>[[120, 84, 884, 121]]<|/det|> +[49] PAGES2k Consortium. A global multiproxy database for temperature reconstructions of the common era. Scientific Data 4 (2017). + +<|ref|>text<|/ref|><|det|>[[120, 132, 884, 170]]<|/det|> +[50] Tierney, J. E. et al. Tropical sea surface temperatures for the past four centuries reconstructed from coral archives. Paleoceanography 30 (3), 226- 252 (2015). + +<|ref|>text<|/ref|><|det|>[[120, 183, 884, 274]]<|/det|> +[51] Gulev, S. K. et al. in Changing State of the Climate System (eds Masson- Delmotte, V. et al.) Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change Ch. 2 (Cambridge University Press, 2021). + +<|ref|>text<|/ref|><|det|>[[120, 287, 886, 344]]<|/det|> +[52] Steiger, N. J., Hakim, G. J., Steig, E. J., Battisti, D. S. & Roe, G. H. Assimilation of time- averaged pseudoproxies for climate reconstruction. Journal of Climate 27 (1), 426- 441 (2014). + +<|ref|>text<|/ref|><|det|>[[120, 356, 884, 412]]<|/det|> +[53] Hakim, G. J. et al. The last millennium climate reanalysis project: Framework and first results. Journal of Geophysical Research: Atmospheres 121 (12), 6745- 6764 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 425, 884, 481]]<|/det|> +[54] Evans, M. N., Tolwinski- Ward, S. E., Thompson, D. M. & Anchukaitis, K. J. Applications of proxy system modeling in high resolution paleoclimatology. Quaternary Science Reviews 76, 16- 28 (2013). + +<|ref|>text<|/ref|><|det|>[[120, 494, 884, 550]]<|/det|> +[55] Dee, S. et al. Prysm: An open- source framework for proxy system modeling, with applications to oxygen- isotope systems. Journal of Advances in Modeling Earth Systems 7 (3), 1220- 1247 (2015). + +<|ref|>text<|/ref|><|det|>[[120, 563, 884, 601]]<|/det|> +[56] Goosse, H. et al. The role of forcing and internal dynamics in explaining the "medieval climate anomaly". Climate Dynamics 39 (12), 2847- 2866 (2012). + +<|ref|>text<|/ref|><|det|>[[120, 614, 884, 652]]<|/det|> +[57] Tardif, R. et al. Last millennium reanalysis with an expanded proxy database and seasonal proxy modeling. Climate of the Past 15 (4), 1251- 1273 (2019). + +<|ref|>text<|/ref|><|det|>[[120, 665, 884, 721]]<|/det|> +[58] Neukom, R., Steiger, N., Gómez- Navarro, J. J., Wang, J. & Werner, J. P. No evidence for globally coherent warm and cold periods over the preindustrial common era. Nature 571 (7766), 550- 554 (2019). + +<|ref|>text<|/ref|><|det|>[[120, 734, 884, 790]]<|/det|> +[59] King, J. M. et al. A data assimilation approach to last millennium temperature field reconstruction using a limited high- sensitivity proxy network. Journal of Climate 1- 64 (2021). + +<|ref|>text<|/ref|><|det|>[[120, 803, 884, 859]]<|/det|> +[60] Zhu, F., Emile- Geay, J., Hakim, G. J., King, J. & Anchukaitis, K. J. Resolving the differences in the simulated and reconstructed temperature response to volcanism. Geophysical Research Letters 47 (8), e2019GL086908 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 80, 886, 138]]<|/det|> +[61] Zhu, F. et al. A re- appraisal of the ENSO response to volcanism with paleoclimate data assimilation. Nature Communications 13 (1) (2022). https://doi.org/10.1038/s41467-022-28210-1. + +<|ref|>text<|/ref|><|det|>[[67, 150, 886, 207]]<|/det|> +[62] Dalaiden, Q., Goosse, H., Rezsöházy, J. & Thomas, E. R. Reconstructing atmospheric circulation and sea- ice extent in the west antarctic over the past 200 years using data assimilation. Climate Dynamics 1- 25 (2021). + +<|ref|>text<|/ref|><|det|>[[67, 220, 886, 258]]<|/det|> +[63] Tierney, J. E. et al. Glacial cooling and climate sensitivity revisited. Nature 584 (7822), 569- 573 (2020). + +<|ref|>text<|/ref|><|det|>[[67, 270, 886, 327]]<|/det|> +[64] Steiger, N. J., Smerdon, J. E., Cook, E. R. & Cook, B. I. A reconstruction of global hydroclimate and dynamical variables over the common era. Scientific Data 5 (1), 1- 15 (2018). + +<|ref|>text<|/ref|><|det|>[[67, 339, 886, 397]]<|/det|> +[65] Morales, M. S. et al. Six hundred years of south american tree rings reveal an increase in severe hydroclimatic events since mid- 20th century. Proceedings of the National Academy of Sciences 117 (29), 16816- 16823 (2020). + +<|ref|>text<|/ref|><|det|>[[67, 408, 886, 466]]<|/det|> +[66] Palmer, J. G. et al. Drought variability in the eastern australia and new zealand summer drought atlas (anzda, ce 1500- 2012) modulated by the interdecadal pacific oscillation. Environmental Research Letters 10 (12), 124002 (2015). + +<|ref|>text<|/ref|><|det|>[[67, 477, 886, 534]]<|/det|> +[67] Comboul, M., Emile- Geay, J., Hakim, G. J. & Evans, M. N. Paleoclimate sampling as a sensor placement problem. Journal of Climate 28 (19), 7717- 7740 (2015). + +<|ref|>text<|/ref|><|det|>[[67, 546, 886, 622]]<|/det|> +[68] Bradley, R. S. Jones, P. D., Bradley, R. S. & Jouzel, J. (eds) Are there optimum sites for global paleotemperature reconstruction? (eds Jones, P. D., Bradley, R. S. & Jouzel, J.) Climatic Variations and Forcing Mechanisms of the Last 2000 Years, 603- 624 (Springer Berlin Heidelberg, Berlin, Heidelberg, 1996). + +<|ref|>text<|/ref|><|det|>[[67, 633, 886, 690]]<|/det|> +[69] Evans, M. N., Kaplan, A. & Cane, M. A. Optimal sites for coral- based reconstruction of global sea surface temperature. Paleoceanography 13 (5), 502- 516 (1998). + +<|ref|>text<|/ref|><|det|>[[67, 702, 886, 759]]<|/det|> +[70] Mauger, G., Bumbaco, K., Hakim, G. & Mote, P. Optimal design of a climatological network: Beyond practical considerations. Geoscientific Instrumentation, Methods and Data Systems 2 (2), 199- 212 (2013). + +<|ref|>text<|/ref|><|det|>[[67, 771, 886, 829]]<|/det|> +[71] Fogt, R. L. et al. Historical sam variability. part ii: Twentieth- century variability and trends from reconstructions, observations, and the ipcc ar4 models. Journal of Climate 22 (20), 5346- 5365 (2009). + +<|ref|>text<|/ref|><|det|>[[67, 841, 886, 878]]<|/det|> +[72] Jones, J. M. et al. Historical sam variability. part i: Century- length seasonal reconstructions. Journal of Climate 22 (20), 5319- 5345 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 80, 884, 155]]<|/det|> +[73] Lim, E.- P., Hendon, H. & Thompson, D. Seasonal evolution of stratosphere- troposphere coupling in the southern hemisphere and implications for the predictability of surface climate. Journal of Geophysical Research: Atmospheres 123 (21), 12–002 (2018). + +<|ref|>text<|/ref|><|det|>[[120, 168, 884, 242]]<|/det|> +[74] Byrne, N. J., Shepherd, T. G. & Polichtchouk, I. Subseasonal- to- seasonal predictability of the southern hemisphere eddy- driven jet during austral spring and early summer. Journal of Geophysical Research: Atmospheres 124 (13), 6841–6855 (2019). + +<|ref|>text<|/ref|><|det|>[[120, 256, 884, 310]]<|/det|> +[75] L'Heureux, M. L. & Thompson, D. W. Observed relationships between the el niño–southern oscillation and the extratropical zonal- mean circulation. Journal of Climate 19 (2), 276–287 (2006). + +<|ref|>text<|/ref|><|det|>[[120, 324, 884, 378]]<|/det|> +[76] Wilson, A. B., Bromwich, D. H. & Hines, K. M. Simulating the mutual forcing of anomalous high southern latitude atmospheric circulation by el niño flavors and the southern annular mode. Journal of Climate 29 (6), 2291–2309 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 392, 884, 447]]<|/det|> +[77] Gillett, N. P., Fyfe, J. C. & Parker, D. E. Attribution of observed sea level pressure trends to greenhouse gas, aerosol, and ozone changes. Geophysical Research Letters 40 (10), 2302–2306 (2013). + +<|ref|>text<|/ref|><|det|>[[120, 461, 884, 534]]<|/det|> +[78] PAGES 2k-PMIP3 group. Continental- scale temperature variability in PMIP3 simulations and PAGES 2k regional temperature reconstructions over the past millennium. Climate of the Past 11 (12), 1673–1699 (2015). https://doi.org/10.5194/cp-11-1673-2015. + +<|ref|>text<|/ref|><|det|>[[120, 548, 884, 585]]<|/det|> +[79] Ding, Q., Steig, E. J., Battisti, D. S. & Wallace, J. M. Influence of the tropics on the southern annular mode. Journal of Climate 25 (18), 6330–6348 (2012). + +<|ref|>text<|/ref|><|det|>[[120, 600, 884, 654]]<|/det|> +[80] Fogt, R. L., Bromwich, D. H. & Hines, K. M. Understanding the sam influence on the south pacific enso teleconnection. Climate Dynamics 36 (7), 1555–1576 (2011). + +<|ref|>text<|/ref|><|det|>[[120, 669, 884, 723]]<|/det|> +[81] Fogt, R. L. & Bromwich, D. H. Decadal variability of the enso teleconnection to the high- latitude south pacific governed by coupling with the southern annular mode. Journal of Climate 19 (6), 979–997 (2006). + +<|ref|>text<|/ref|><|det|>[[120, 737, 884, 810]]<|/det|> +[82] Emile- Geay, J., Cobb, K. M., Mann, M. E. & Wittenberg, A. T. Estimating central equatorial pacific sst variability over the past millennium. part ii: Reconstructions and implications. Journal of Climate 26 (7), 2329–2352 (2013). + +<|ref|>text<|/ref|><|det|>[[120, 825, 884, 879]]<|/det|> +[83] Wright, N. M., Krause, C. E., Phipps, S. J., Boschat, G. & Abram, N. J. Influence of long- term changes in solar irradiance forcing on the southern annular mode. Climate of the Past 18 (6), 1509–1528 (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 80, 886, 157]]<|/det|> +[84] Ammann, C. M., Joos, F., Schimel, D. S., Otto- Bliesner, B. L. & Tomas, R. A. Solar influence on climate during the past millennium: Results from transient simulations with the NCAR Climate System Model. Proceedings of the National Academy of Sciences 104 (10), 3713- 3718 (2007). + +<|ref|>text<|/ref|><|det|>[[123, 168, 884, 206]]<|/det|> +[85] Hind, A. & Moberg, A. Past millennial solar forcing magnitude. Climate dynamics 41 (9), 2527- 2537 (2013). + +<|ref|>text<|/ref|><|det|>[[123, 219, 884, 257]]<|/det|> +[86] Schurer, A. P., Tett, S. F. & Hegerl, G. C. Small influence of solar variability on climate over the past millennium. Nature Geoscience 7 (2), 104- 108 (2014). + +<|ref|>text<|/ref|><|det|>[[123, 270, 884, 325]]<|/det|> +[87] Ho, M., Kiem, A. & Verdon- Kidd, D. The southern annular mode: a comparison of indices. Hydrology and Earth System Sciences 16 (3), 967- 982 (2012). + +<|ref|>text<|/ref|><|det|>[[123, 340, 884, 378]]<|/det|> +[88] Compo, G. P. et al. The twentieth century reanalysis project. Quarterly Journal of the Royal Meteorological Society 137 (654), 1- 28 (2011). + +<|ref|>text<|/ref|><|det|>[[123, 390, 884, 447]]<|/det|> +[89] Slivinski, L. C. et al. Towards a more reliable historical reanalysis: Improvements for version 3 of the twentieth century reanalysis system. Quarterly Journal of the Royal Meteorological Society 145 (724), 2876- 2908 (2019). + +<|ref|>text<|/ref|><|det|>[[123, 459, 886, 516]]<|/det|> +[90] Evensen, G. Sequential data assimilation with a nonlinear quasi- geostrophic model using monte carlo methods to forecast error statistics. Journal of Geophysical Research: Oceans 99 (C5), 10143- 10162 (1994). + +<|ref|>text<|/ref|><|det|>[[123, 528, 884, 567]]<|/det|> +[91] Andrews, A. A square root formulation of the kalman covariance equations. AIAA Journal 6 (6), 1165- 1166 (1968). + +<|ref|>text<|/ref|><|det|>[[123, 579, 884, 636]]<|/det|> +[92] Tippett, M. K., Anderson, J. L., Bishop, C. H., Hamill, T. M. & Whitaker, J. S. Ensemble square root filters. Monthly Weather Review 131 (7), 1485- 1490 (2003). + +<|ref|>text<|/ref|><|det|>[[123, 648, 884, 686]]<|/det|> +[93] Whitaker, J. S. & Hamill, T. M. Ensemble data assimilation without perturbed observations. Monthly Weather Review 130 (7), 1913- 1924 (2002). + +<|ref|>text<|/ref|><|det|>[[123, 699, 884, 737]]<|/det|> +[94] Chen, Z. et al. Bayesian filtering: From kalman filters to particle filters, and beyond. Statistics 182 (1), 1- 69 (2003). + +<|ref|>text<|/ref|><|det|>[[123, 750, 884, 788]]<|/det|> +[95] Wikle, C. K. & Berliner, L. M. A bayesian tutorial for data assimilation. Physica D: Nonlinear Phenomena 230 (1- 2), 1- 16 (2007). + +<|ref|>text<|/ref|><|det|>[[123, 801, 884, 857]]<|/det|> +[96] Parsons, L. A. et al. Do multi- model ensembles improve reconstruction skill in paleoclimate data assimilation? Earth and Space Science 8 (4), e2020EA001467 (2021). + +<|ref|>text<|/ref|><|det|>[[123, 871, 884, 909]]<|/det|> +[97] Oke, P. R., Allen, J. S., Miller, R. N., Egbert, G. D. & Kosro, P. M. Assimilation of surface velocity data into a primitive equation coastal ocean model. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 48, 880, 66]]<|/det|> +Trends and variability in the Southern Annular Mode over the Common Era 27 + +<|ref|>text<|/ref|><|det|>[[120, 82, 770, 101]]<|/det|> +Journal of Geophysical Research: Oceans 107 (C9), 5- 1 (2002). + +<|ref|>text<|/ref|><|det|>[[120, 115, 884, 152]]<|/det|> +[98] Evensen, G. The ensemble kalman filter: Theoretical formulation and practical implementation. Ocean Dynamics 53 (4), 343- 367 (2003). + +<|ref|>text<|/ref|><|det|>[[120, 166, 884, 220]]<|/det|> +[99] Matsikaris, A., Widmann, M. & Jungclaus, J. On- line and off- line data assimilation in palaeoclimatology: a case study. Climate of the Past 11 (1), 81- 93 (2015). + +<|ref|>text<|/ref|><|det|>[[120, 235, 884, 289]]<|/det|> +[100] Acevedo, W., Fallah, B., Reich, S. & Cubasch, U. Assimilation of pseudotree- ring- width observations into an atmospheric general circulation model. Climate of the Past 13 (5), 545- 557 (2017). + +<|ref|>text<|/ref|><|det|>[[120, 304, 884, 358]]<|/det|> +[101] Dee, S. G., Steiger, N. J., Emile- Geay, J. & Hakim, G. J. On the utility of proxy system models for estimating climate states over the common era. Journal of Advances in Modeling Earth Systems 8 (3), 1164- 1179 (2016). + +<|ref|>text<|/ref|><|det|>[[120, 372, 884, 427]]<|/det|> +[102] Bhend, J., Franke, J., Folini, D., Wild, M. & Brönnimann, S. An ensemble- based approach to climate reconstructions. Climate of the Past 8 (3), 963- 976 (2012). + +<|ref|>text<|/ref|><|det|>[[120, 441, 884, 477]]<|/det|> +[103] Palmer, W. C. Meteorological drought Vol. 30 (US Department of Commerce, Weather Bureau, 1965). + +<|ref|>text<|/ref|><|det|>[[120, 492, 884, 528]]<|/det|> +[104] Thornthwaite, C. W. An approach toward a rational classification of climate. Geographical Review 38 (1), 55- 94 (1948). + +<|ref|>text<|/ref|><|det|>[[120, 542, 884, 579]]<|/det|> +[105] Allen, R., Smith, M., Perrier, A. & Pereira, L. S. An update for the definition of reference evapotranspiration. ICID bulletin 43 (2), 1- 34 (1994). + +<|ref|>text<|/ref|><|det|>[[120, 593, 884, 648]]<|/det|> +[106] Smerdon, J. E., Cook, B. I., Cook, E. R. & Seager, R. Bridging past and future climate across paleoclimatic reconstructions, observations, and models: A hydroclimate case study. Journal of Climate 28 (8), 3212- 3231 (2015). + +<|ref|>text<|/ref|><|det|>[[120, 662, 884, 716]]<|/det|> +[107] Cook, E. R., Meko, D. M., Stahle, D. W. & Cleaveland, M. K. Drought reconstructions for the continental united states. Journal of Climate 12 (4), 1145- 1162 (1999). + +<|ref|>text<|/ref|><|det|>[[120, 731, 884, 785]]<|/det|> +[108] Esper, J., Frank, D. C., Wilson, R. J. & Briffa, K. R. Effect of scaling and regression on reconstructed temperature amplitude for the past millennium. Geophysical Research Letters 32 (7) (2005). + +<|ref|>text<|/ref|><|det|>[[120, 801, 884, 855]]<|/det|> +[109] Frank, D., Esper, J. & Cook, E. R. Adjustment for proxy number and coherence in a large- scale temperature reconstruction. Geophysical Research Letters 34 (16) (2007). + +<|ref|>text<|/ref|><|det|>[[120, 871, 884, 907]]<|/det|> +[110] Anchukaitis, K. J. et al. Last millennium northern hemisphere summer temperatures from tree rings: Part ii, spatially resolved reconstructions. Quaternary + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 85, 500, 103]]<|/det|> +1006 Science Reviews 163, 1- 22 (2017). + +<|ref|>text<|/ref|><|det|>[[60, 115, 884, 153]]<|/det|> +1007 [111] Mann, M. E. & Jones, P. D. Global surface temperatures over the past two millennia. Geophysical Research Letters 30 (15) (2003). + +<|ref|>text<|/ref|><|det|>[[60, 165, 884, 220]]<|/det|> +1009 [112] Schmidt, G. et al. Climate forcing reconstructions for use in pmp simulations of the last millennium (v1.1). Geoscientific Model Development 5 (1), 185- 191 (2012). + +<|ref|>text<|/ref|><|det|>[[60, 234, 884, 289]]<|/det|> +1012 [113] Haurwitz, M. W. & Brier, G. W. A critique of the superposed epoch analysis method: its application to solar- weather relations. Monthly Weather Review 109 (10), 2074- 2079 (1981). + +<|ref|>text<|/ref|><|det|>[[60, 303, 884, 340]]<|/det|> +1015 [114] Sigl, M. et al. Timing and climate forcing of volcanic eruptions for the past 2,500 years. Nature 523 (7562), 543- 549 (2015). + +<|ref|>text<|/ref|><|det|>[[60, 353, 884, 408]]<|/det|> +1017 [115] Toohey, M. & Sigl, M. Volcanic stratospheric sulfur injections and aerosol optical depth from 500 bce to 1900 ce. Earth System Science Data 9 (2), 809- 831 (2017). + +<|ref|>text<|/ref|><|det|>[[60, 421, 884, 478]]<|/det|> +1020 [116] Zheng, F., Li, J., Clark, R. T. & Nnamchi, H. C. Simulation and projection of the southern hemisphere annular mode in cmip5 models. Journal of Climate 26 (24), 9860- 9879 (2013). + +<|ref|>text<|/ref|><|det|>[[60, 491, 884, 546]]<|/det|> +1023 [117] Knutti, R., Masson, D. & Gettelman, A. Climate model genealogy: Generation cmip5 and how we got there. Geophysical Research Letters 40 (6), 1194- 1199 (2013). + +<|ref|>text<|/ref|><|det|>[[60, 560, 884, 616]]<|/det|> +1026 [118] Sanderson, B. M., Knutti, R. & Caldwell, P. A representative democracy to reduce interdependency in a multimodel ensemble. Journal of Climate 28 (13), 5171- 5194 (2015). + +<|ref|>text<|/ref|><|det|>[[60, 629, 884, 685]]<|/det|> +1029 [119] Tingley, M. P. et al. Piecing together the past: statistical insights into paleoclimatic reconstructions. Quaternary Science Reviews 35, 1- 22 (2012). + +<|ref|>text<|/ref|><|det|>[[60, 699, 884, 754]]<|/det|> +1032 [120] Amrhein, D. E., Hakim, G. J. & Parsons, L. A. Quantifying structural uncertainty in paleoclimate data assimilation with an application to the last millennium. Geophysical Research Letters 47 (22), e2020GL090485 (2020). + +<|ref|>text<|/ref|><|det|>[[60, 768, 884, 841]]<|/det|> +1035 [121] Franke, J., Valler, V., Brönnimann, S., Neukom, R. & Jaume- Santero, F. The importance of input data quality and quantity in climate field reconstructions- results from the assimilation of various tree- ring collections. Climate of the Past 16 (3), 1061- 1074 (2020). + +<|ref|>text<|/ref|><|det|>[[60, 856, 884, 892]]<|/det|> +1039 [122] Osman, M. B. et al. Globally resolved surface temperatures singe the last glacial maximum. Nature 599, 239- 244 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 45, 816, 64]]<|/det|> +Trends and variability in the Southern Annular Mode over the Common Era 29 + +<|ref|>text<|/ref|><|det|>[[65, 81, 885, 136]]<|/det|> +[123] Perkins, W. A. & Hakim, G. J. Reconstructing paleoclimate fields using online data assimilation with a linear inverse model. Climate of the Past 13 (5), 421–436 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[328, 0, 670, 20]]<|/det|> +Springer Nature 2021 LATE[X template] + +<|ref|>table<|/ref|><|det|>[[201, 419, 792, 508]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[111, 520, 885, 558]]<|/det|> +Table 1: Reconstruction Skill metrics for the Southern Annular Mode calculated against instrumental indices over the given time periods + +
MetricMarshall Index (1958-2000)Fogt Index (1958-2000)Fogt Index (1866-2000)
Correlation (p ≪ 0.001)0.720.670.56
RMSE1.451.561.80
σ Ratio0.971.031.15
Mean Bias-0.260.45-0.29
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[245, 250, 748, 635]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 636, 885, 727]]<|/det|> +
Fig. 1: Map of the proxy network. Black Xs indicate the centroid of binned drought atlas sites. Grey markers indicate PAGES2k sites. The size of the PAGES2k markers correspond to the length of each record. Filled color contours show the field correlation between the SAM index and DJF sea level pressure from 20CR [88, 89] over the period 1958-2000 CE.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[188, 189, 808, 640]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 643, 885, 789]]<|/det|> +
Fig. 2: Evolution of the reconstruction over time. Top: Comparison of the annual reconstruction (blue) with the Marshall index (red) over the instrumental era. Shading indicates the 5-95 percentiles of the reconstruction. Middle: Evolution of the annual reconstruction (blue) and 31-year lowpass filtered (black) over the Common Era. Shading indicates the 5-95 percentiles of the lowpass filtered series. Bottom: Composition of the proxy network over time. Colors for proxy types are as follows: Dark blue (coral), pink (ANZDA), grey (SADA), dark red (trees), light blue (glacier ice), black (lake sediment).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[191, 145, 805, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 684, 885, 829]]<|/det|> +
Fig. 3: Optimal Sensor Analysis. Top: Maps of the potential ability for drought atlas (left) and PAGES2k (right) sites to constrain reconstruction posterior variance. Middle: Evolution of reconstruction posterior variance over time. Yellow bars indicate the addition of the indicated proxy to the network. Bottom: Ranked histograms of the seven sites with greatest potential influence in the early reconstruction (left), immediately before the addition of drought atlases (middle) and for the full network (right). Potential influence is determined as the uncertainty constrained by a single-proxy network.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 222, 876, 612]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 622, 884, 768]]<|/det|> +
Fig. 4: SAM climate responses. Top left: Wavelet coherence of the reconstructed SAM index with the solar forcing reconstruction [112]. Bottom left: Composite mean response to major volcanic eruptions. Shading indicates 5-95 percentiles. Blue line is the ensemble mean. Right: Instrumental SAM trends for the Marshall Index (top right) and reconstruction (bottom right). Colored points indicate trends calculated from a sliding window centered on the given year. Solid (dotted) contours surround statistically significant trends at the \(90\%\) confidence interval relative to the reconstruction over the period 1500-1900 CE (1-1900 CE).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[188, 358, 808, 558]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 562, 884, 617]]<|/det|> +
Fig. 5: Comparison of SAM reconstructions over the last millennium. All reconstructions are smoothed via a 30-year Gaussian filter and normalized to the period 1400-1850 CE.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 130, 225, 150]]<|/det|> +Supplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/images_list.json b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..0f9c4f915528849cf4cf0b39907488f2e0827eba --- /dev/null +++ b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/images_list.json @@ -0,0 +1,40 @@ +[ + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "a", + "footnote": [], + "bbox": [ + [ + 140, + 60, + 833, + 687 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Jussila et al.", + "footnote": [], + "bbox": [], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Jussila et al.", + "footnote": [], + "bbox": [ + [ + 115, + 50, + 500, + 680 + ] + ], + "page_idx": 28 + } +] \ No newline at end of file diff --git a/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae.mmd b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae.mmd new file mode 100644 index 0000000000000000000000000000000000000000..3a120b2da0134aab729131817ffba59a119b8459 --- /dev/null +++ b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae.mmd @@ -0,0 +1,437 @@ + +# Live imaging and conditional disruption of native vertebrate planar cell polarity + +Maria Jussila The Hospital for Sick Children Curtis Boswell Yale University School of Medicine Brian Ciruna ( \(\square\) ciruna@sickkids.ca) The Hospital for Sick Children + +## Article + +Keywords: + +Posted Date: August 18th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 764931/v1 + +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on September 23rd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33322- 9. + +<--- Page Split ---> + +# Live imaging and conditional disruption of native vertebrate planar cell polarity + +2 + +Maria Jussila \(^{1}\) , Curtis W. Boswell \(^{1,2,3}\) and Brian Ciruna \(^{1,2,*}\) + +4 + +1. Program in Developmental & Stem Cell Biology, The Hospital for Sick Children, 686 Bay + +Street, Toronto, Ontario, M5G 0A4, Canada. + +2. Department of Molecular Genetics, The University of Toronto, Toronto, Ontario, M5S 1A8, + +8 Canada. + +3. Current address: Department of Genetics, Yale University School of Medicine, New Haven, + +10 CT 06510, USA. + +11 \* Corresponding author. Email: ciruna@sickkids.ca + +<--- Page Split ---> + +## Abstract + +Tissue- wide coordination of polarized cytoskeletal organization and cell behaviour, critical for organ morphogenesis and tumour progression, is controlled by asymmetric membrane localization of non- canonical Wnt/planar cell polarity (PCP) signalling components. Understanding the dynamic regulation of PCP thus requires visualization of these core proteins. In vertebrates, immunohistochemical studies have provided snapshots of PCP within tissues while exogenous, fluorescently- tagged reporters have been introduced for live imaging of cell polarity. However, the functionality and spatiotemporal relevance of static and exogenous PCP readouts remain uncertain. Here we fluorescently tag an endogenous core PCP component, Vangl2, in zebrafish. We report on the authentic regulation of vertebrate PCP, through live imaging of this native sfGFP- Vangl2 protein during embryogenesis. We couple sfGFP- Vangl2 with conditional zGrad GFP- nanobody degradation methodologies for tissue- specific interrogation of PCP function. Together, our studies provide crucial insights into the establishment and maintenance of vertebrate PCP and create a powerful experimental paradigm for future investigations. + +## Introduction + +The non- canonical Wnt/planar cell polarity (PCP) signalling pathway controls polarized, uniform orientation of cells within a plane of a tissue. PCP pathway has been shown to regulate a number of developmental and homeostatic processes by directing collective asymmetric modification of the cell cytoskeleton, leading groups of cells to either move or divide in a shared direction, or position their organelles such as the basal body asymmetrically \(^{1 - 5}\) . Molecular genetic studies in Drosophila have identified essential roles for asymmetric membrane localization of opposing PCP signalling proteins in both the establishment and intercellular propagation of polarity \(^{6}\) . + +<--- Page Split ---> + +Despite identification of a multitude of PCP- regulated morphogenetic processes, characterization of vertebrate PCP at the cellular and molecular level is still lacking. This is challenging, because detailed observation requires live imaging capabilities and the ability to differentiate asymmetric protein localization across membranes of adjacent cells. In vertebrates, immunohistochemical studies have provided snapshots of PCP within tissues7,8 while exogenous, fluorescently- tagged reporter proteins have been introduced for mosaic imaging of PCP. Although great complexities in the dynamic regulation of PCP have been reported across tissues and polarized cell behaviours,9- 12 the functionality and spatiotemporal accuracy of these static and exogenous readouts remain uncertain. The biological relevance of studies utilizing exogenous, fluorescently- tagged PCP molecules remains to be determined, as overexpression of PCP proteins can disrupt cell polarity11,13 and functionality of published PCP fusion proteins has not been validated via rescue of orthologous mutant phenotypes.12,14- 16 Furthermore, non- physiological expression of exogenous proteins can mask subtle localization patterns or signalling dynamics. This is highlighted by observations that in many occasions, co- expression of another PCP component is required for asymmetric membrane localization of the fluorescent fusion protein of interest.17- 20 + +To understand the dynamic regulation of native PCP, we have aimed to overcome limitations imposed by exogenous cell polarity reporters by imaging and manipulating the endogenous zebrafish Vangl2 protein, a core PCP component and essential regulator of vertebrate planar polarity.21,22 Using CRISPR/Cas9 gene editing, we introduced a superfolder GFP (sfGFP) tag onto the N- terminus of endogenous Vangl2 to generate a functional and native PCP reporter protein. By performing detailed embryonic cell transplantation and single- cell level analyses, we documented the dynamic sub- cellular localization of Vangl2 over the course of neural tube + +<--- Page Split ---> + +formation, revealing novel associations of sfGFP- Vang12 positive cell membranes with polarized basal bodies in floorplate cells. Furthermore, we exploited zGrad \(^{23}\) GFP- specific protein degradation methodologies to conditionally interrogate sfGFP- Vang12 function in floor plate cells, demonstrating an essential role for Vang12 in both the establishment and maintenance of polarized basal body positioning as well as an inherent and surprising directionality to the intercellular propagation of PCP. Our work highlights the importance and utility of studying endogenous proteins in the context of PCP, and opens the door to conditional analysis of PCP function beyond embryonic development. + +## Results and Discussion + +To visualize the dynamic regulation of endogenous PCP signalling in zebrafish, we used CRISPR/Cas9 genome editing to target a superfolder green fluorescent protein (sfGFP) onto the N- terminus of Vang12, an essential and membrane- localized PCP signalling protein (Fig. 1a; targeting details in Methods section). Embryos carrying the knock- in allele showed prominent sfGFP signal in the brain and spinal cord at 22 hours post fertilization (hpf) (Fig. 1b). In striking contrast to the \(vangl2^{m209 / 209}\) loss- of- function mutant phenotype, which is embryonic lethal \(^{21}\) , fish homozygous for the \(vangl2^{sfGFP}\) allele or trans- heterozygote for \(vangl2^{sfGFP}\) and \(vangl2^{m209}\) alleles are viable, fertile and appear morphologically normal (Fig. 1c). Closer inspection revealed that \(vangl2^{sfGFP / sfGFP}\) embryos are slightly but significantly shorter than wildtype siblings at 24 hpf (Fig. 1d). Interestingly, heterozygote \(vangl2^{m209 / + }\) embryos are similarly shorter than wildtype, which has not been previously reported. \(Vangl2^{sfGFP / m209}\) transheterozygote embryos demonstrate a further shortening of the body axis but are significantly less affected than the \(vangl2^{m209 / m209}\) mutants (Fig. 1d). Altogether these data indicate that, although mildly hypomorphic, the sfGFP + +<--- Page Split ---> + +Vangl2 fusion protein is functional and can therefore be used to interrogate endogenous PCP signalling dynamics. + +To determine the onset of Vangl2 expression, we imaged vangl2sfGFP/sfGFP embryos (hereafter referred to as vangl2sfGFP) through blastula and gastrula stages. Weak Vangl2 membrane localization could be observed in a subset of cells as early as high stage (3.3 hpf), which became more ubiquitous through early epiboly stages (4.6 hpf; Fig. 1e). Interestingly, weak Vangl2 membrane- localization was also observed within enveloping layer (EVL) cells at these stages (Supplementary Fig. 1a). These observations are consistent with a role for Vangl2 prior to gastrulation, perhaps in promoting tissue cohesion in margin cells during early blastoderm morphogenesis.24 Vangl2 expression levels increased, and membrane localization became more prominent with the onset of gastrulation at shield stage, persisting through \(75\%\) epiboly and bud stages (Fig. 1e, Supplementary Fig. 1b- e). This corresponds to the known role for Vangl2 in the regulation of convergence and extension movements.21 Although immunohistochemical analyses of Vangl2 expression have yielded similar observations14 the superior consistency, sensitivity and live- imaging capabilities of sfGFP- Vangl2 expression highlight its utility for experimental studies. + +In addition to membrane- localized Vangl2, we also observed punctate intracellular accumulations, which became prominent at the onset of gastrulation when they concentrated into large cytosolic aggregates (Fig. 1e). To investigate the identity of these compartments, we analysed Vangl2 localization in relation to known markers of intracellular sorting organelles (Fig. 1f- i and Supplementary Fig. 1b- e). In published studies, early and late endosome markers Rab5 and Rab7 colocalize with PCP components Flamingo/Celsr and Frizzled- 3a in Drosophila wing, mouse keratinocytes and zebrafish neural tube.10,15,25- 27 The recycling endosome marker + +<--- Page Split ---> + +Rab11 also colocalizes with Flamingo/Celsr in mouse skin and in Drosophila wing, and with Vangl2 in cell culture. \(^{15,28,29}\) However, vertebrate studies have used exogenous PCP reporters, which might not recapitulate the endogenous subcellular protein localization and may induce trafficking artifacts. While we observed that Rab5c colocalized with some of the smaller Vangl2 puncta at both shield and \(90\%\) epiboly stages (Fig. 1f and Supplementary Fig. 1b), and that Rab11a colocalized with Vangl2 at the cell membrane but not within the cytoplasm (Fig. 1h and Supplementary Fig. 1d), the majority of Vangl2 puncta colocalized with mCherry- Rab7 (Fig. 1g and Supplementary Fig. 1c) suggesting late endosomal transport to lysosomes or the trans- Golgi network (TGN). Prior to gastrulation the zebrafish Golgi is fragmented, \(^{30}\) and whereas we could see occasional sfGFP- Vangl2 signal in the proximity of the TGN reporter dsRed- GalT, Vangl2 was predominantly localized elsewhere in the cell (Fig. 1i and Supplementary Fig. 1e). Taken together these results suggest that, at early gastrulation stages, Vangl2 is targeted to late endosomes for degradation. + +Following gastrulation, strong Vangl2 expression was observed in the central nervous system, whereas only weak sfGFP- Vangl2 signal could be detected in surrounding axial and paraxial mesodermal lineages (Figs. 1b, 2a). Previous work has demonstrated that exogenous GFPTagged Vangl2 or its cytoplasmic binding partner Prickle polarize anteriorly in neuroepithelial cells in zebrafish and Xenopus. \(^{9,12,14,16,26,31}\) To characterize the asymmetric distribution of endogenous Vangl2, we performed cell transplantation at blastoderm stages to generate clones of \(vangl2^{yGGFP}\) cells within wildtype host embryos and imaged Vangl2 localization in individual neuroepithelial cells at multiple time points during neural tube formation (Fig. 2b). At all stages, weak sfGFP- Vangl2 signal could be observed across the entire neuroepithelial cell membrane, with bright membrane sfGFP- Vangl2 domains enriched in puncta at anterior cell surfaces (Fig. + +<--- Page Split ---> + +2c, Supplementary Fig.2a, Supplementary Movie 1). Vangl2 localization varied considerably between cells, including apical and basal distribution patterns, as well as localization into cell protrusions (Fig. 2c and Supplementary Fig.2a). Previously, exogenous Vangl2 reporter localization has been quantified along single cellular planes. To gain a global view of Vangl2 distribution, we performed 3- dimensional quantification of sfGFP- Vangl2 membrane- enrichment in comparison to an RFP membrane reporter (mRFP) using the spot function in Imaris (detailed description in Methods). We plotted the subset of membrane spots containing the highest sfGFP- Vangl2 and mRFP fluorescent intensities across anterior- posterior and apico- basal cell axes (Supplementary Fig.3A) and found that, despite the variation between cells, Vangl2 was significantly enriched at the anterior membrane at all stages analysed (Fig. 2d). Vangl2 localization was less uniformly polarized along the apico- basal axis, with Vangl2 showing strongest apical enrichment at the 13 somite- stage concomitant to neural tube midline formation \(^{16}\) (Fig. 2e). + +To examine the dynamics of Vangl2 protein localization, we performed time- lapse imaging of \(vangl2^{yGFP}\) chimeric embryos between 10- 13 somite stages. Persistent sfGFP- Vangl2 enrichment was observed along anterior membrane domains (Supplementary Movie 1), supporting evidence from multiple Drosophila, Xenopus and mouse studies (predominantly using exogenous reporters) that PCP components concentrate into potential "signalosomes" on the cell membrane. \(^{9,16,20,28,32 - 35}\) In Xenopus, exogenous Vangl2 and Prickle reporters accumulate with actomyosin at shrinking neuroepithelial cell junctions \(^{9}\) and, in mesodermal cells undergoing convergence and extension movements, Prickle2 colocalizes with the LifeAct filamentous actin reporter. \(^{35,36}\) These observations suggest that Vangl2/PCP directly influences cytoskeletal reorganization. To investigate the relationship between anterior Vangl2 enrichment and actin + +<--- Page Split ---> + +organization, we co- expressed a mCherry- LifeAct reporter in vangl2sfGFP cells. Live imaging revealed that the majority of neuroepithelial cell protrusive activity and LifeAct localization was restricted to apical and basal surfaces, and not to anterior membranes where sfGFP- Vangl2 was enriched (Supplementary Movies 2- 3). However, sfGFP- Vangl2 could be observed accumulating within active cellular protrusions at the anterior membrane (Supplementary Fig.2a, Supplementary Movie 4), consistent with known roles for PCP and Vangl2 in regulating filopodia and cell protrusive activity during gastrulation and facial branchiomotor neuron migration.11,12,37 Together, although the presence of bright sfGFP- Vangl2 puncta supports the notion that PCP signalling functions in specific subdomains on the anterior membrane, the presence of Vangl2 membrane enrichment was not predictive of a particular membrane behaviour. + +To further understand how neighbouring cells affect asymmetric Vangl2 localization into different membrane- domains, we performed cell transplantation experiments to generate clones of vangl2sfGFP cells within a vangl2tk50f loss- of- function host embryo (Fig. 2f). In this PCP- deficient environment, vangl2sfGFP cells clustered together, suggesting that Vangl2 expressing cells are able to recognize and preferentially interact with one other. While Vangl2 remained localized to the cell membrane, planar polarized membrane enrichment was lost and Vangl2 appeared to concentrate at cell protrusions (Fig. 2i and Supplementary Fig.2b). We picked a subset of the most mediolaterally aligned cells to quantify asymmetric Vangl2 localization and found no difference in the distribution of the brightest Vangl2 and mRFP spots along the anterior- posterior or apico- basal axes (Fig. 2g- h; Supplementary Fig. 3d). Loss of polarity in vangl2sfGFP positive clones supports a cell non- autonomous regulation of vertebrate PCP. The + +<--- Page Split ---> + +persistence of Vangl2 in cell protrusions suggests that it might function in mediating local signalling that is independent of its role in global anterior- posterior neural tube planar polarity. + +The re- organization and re- establishment of cell polarity over the course of cell division remains a poorly understood feature of vertebrate PCP. In zebrafish, asymmetric localization of + +exogenous Prickle- GFP reporter molecules is lost upon neuroepithelial cell division, but is re- established as daughter cells reintegrate into the neuroepithelium. \(^{16}\) In mouse epidermis, different + +studies have shown PCP components to be internalized during mitosis or to persist on cell membranes in a polarized manner. \(^{15,38}\) Exogenous Vangl2, specifically, has been shown to + +persist on cell membranes and to get trans- internalized from neighbouring cell membranes into + +the dividing cell. \(^{10}\) To determine the fate of endogenous Vangl2 during mitosis, we imaged + +vangl2 \(^{9}\) GFP neuroepithelial cells over the course of cell division. Notably, asymmetric sfGFP- + +Vangl2 distribution was lost upon cell rounding, but Vangl2 remained at the cell membrane + +(Supplementary Movie 3- 4). We observed sfGFP- Vangl2 enriched at the poles of dividing cells, + +as well as to basolateral membrane projections that are maintained during division + +(Supplementary Movies 3- 4). Although some intracellular sfGFP- Vangl2 signal was observed in + +dividing cells, we did not observe significant endocytosis of membrane Vangl2 or trans- + +endocytosis between neighbouring cells (Supplementary Movies 3- 4). Reacquisition of Vangl2 + +polarization in daughter cells was not obvious immediately upon division (Supplementary Movie + +2). Of interest, sfGFP- Vangl2 positive membrane protrusions from distant neighbours could be + +observed interacting with dividing cells (Supplementary Movie 4). Out of 24 cells that had an + +anterior Vangl2 positive protrusion, 11 were touching a dividing cell, and the same was observed + +for 15/32 apical protrusions (total n=112 cells). Our results show that Vangl2 remains on the + +<--- Page Split ---> + +membrane but loses its asymmetry during mitosis, and that re- establishment of cell polarity may be regulated by contacts from neighbouring cells. + +In addition to cytoskeletal organization, polarized basal body (BB) positioning has been identified as a conserved readout of PCP from insects to vertebrates. \(^{39}\) In zebrafish floorplate cells, BBs are localized to the posterior apical membrane by 24 hpf, and the establishment of BB polarity requires Vangl2. \(^{13,26,40}\) To examine the relationship between Vangl2 localization and BB positioning, we imaged the floorplate of \(vangl2^{sfGFP}\) embryos at 22 and 32hpf. In the sagittal plane, sfGFP- Vangl2 was enriched on the anterior apical cell membrane (Fig. 3a- b), in accordance with observations using an exogenous GFP- Vangl2 transgenic reporter. \(^{12,26}\) However, in some floorplate cells apically enriched Vangl2 appeared to extend anteriorly towards neighbouring cells (Fig. 2a, Fig. 3a- b). This bright apical Vangl2 signal was closely associated with BBs labelled with Centrin- mCherry, and with Arl13b- positive cilia (Fig. 3a- b). We imaged individual transplanted \(vangl2^{sfGFP}\) floorplate cells and found that Vangl2 was indeed localized asymmetrically on anterior cell membranes, and that in some cells it further localized to an anterior apical projection (Fig. 3c). Labelling BBs in transplanted \(vangl2^{sfGFP}\) cells with Centrin- mCherry confirmed the proximity of anterior Vangl2 with posteriorly polarized BBs (Fig. 3d). Furthermore, labelling of host cell BBs confirmed that Vangl2- positive apical membrane closely associated with the posterior BB of neighbouring cells (Fig. 3e, 11/17 cells). Of interest exogenous Frizzled3a, a putative intracellular binding partner of Vangl2, has been shown to be enriched posteriorly into puncta at the base of cilia in zebrafish. \(^{26}\) This localization pattern suggests that PCP activity could act to physically deform the floorplate cell membranes at the site of BB docking, perhaps influencing the posterior tilting of the floorplate motile cilia. \(^{13}\) + +<--- Page Split ---> + +To determine whether Vang12 also controls the maintenance of BB polarity, we took advantage of zGrad GFP- nanobody protein degradation methodologies to conditionally manipulate endogenous sfGFP- Vang12 protein levels. \(^{23}\) We first confirmed that zGrad is non- toxic to embryos, and that zGrad- injected vang12sfGFP/sfGFP embryos phenocopy vang12 loss- of- function mutants (Supplementary Fig. 4a- b). Furthermore, because vang12 mutants are subject to maternal effects \(^{16}\) , we investigated the consequences of zGrad expression across a full genotypic spectrum of maternal and zygotic vang12sfGFP alleles. Strikingly, sequential zGrad- mediated degradation of maternal and zygotic vang12sfGFP gene products yielded progressively more severe phenotypes as measured by embryo length and axial length- width ratios (Fig. 4a, Supplementary Fig. 4c), recapitulating maternal and zygotic vang12 mutant phenotypes. This highlights the sensitivity and functionality of sfGFP- Vang12 zGrad protein degradation methods. + +To conditionally manipulate sfGFP- Vang12 protein levels within the floorplate, we generated a ubiquitously expressed Tg(β- actin2::loxP- mCherry- STOP- loxP- zGrad) transgene and crossed it to the Tg(foxj1a::iCre) line \(^{41}\) , which expresses Cre recombinase (activating zGrad) specifically within motile- ciliated cell lineages. At 32 hpf, when BB polarity has been established, sfGFP- Vang12 was still present at the membrane of experimental zGrad; foxj1a::iCre; vang12sfGFP floorplate cells (data not shown). However, mosaic loss of sfGFP- Vang12 could be observed by 48 hpf (Fig. 4b), allowing us to analyse both cell- autonomous and non- autonomous roles for Vang12 in the maintenance of BB polarity. Images were scored for the loss of Vang12 in individual cells and their immediate anterior and posterior neighbours. Irrespective of their Vang12 expression status, cells surrounded by sfGFP- negative neighbours demonstrated a loss of BB polarity when compared to control populations. Strikingly, the presence of sfGFP- Vang12 in a posterior neighbouring cell was associated with restored BB polarity in Vang12- negative cells, + +<--- Page Split ---> + +even if the posterior neighbour demonstrated abnormal BB positioning (Fig. 4c- d). In contrast, the presence of sfGFP- Vangl2 in anterior neighbours did not restore polarity of Vangl2 negative cells. These results demonstrate that Vangl2 is required for the maintenance of BB polarity, and that Vangl2 functions non- cell autonomously to propagate PCP in a posterior to anterior direction. + +Taken together, our work provides essential new insights into the establishment and maintenance of vertebrate PCP. Live imaging of endogenous Vangl2 has authenticated broad principles predicted using exogenously introduced reporter molecules, including organization of PCP across the embryonic anterior- posterior axis and concentration of core PCP proteins into 'signalosomes' at the cell membrane. It has also yielded novel findings that 1) demonstrate dynamic redistribution of Vangl2 within discrete subcellular membrane domains over the course of gastrulation and neurulation, 2) suggest tissue- specific contexts for published tenets regarding membrane recycling of PCP proteins during mitosis or their association with the actin cytoskeleton, and 3) indicate an inherent and surprising directionality to the intercellular propagation of PCP. Outside of C. elegans,42 functional fluorescently- tagged PCP knock- in alleles have not been widely reported. Our vangl2sfGFP allele thus represents an invaluable resource that, when combined with embryonic cell- transplantation, live- imaging and conditional zGrad degradation methodologies, establishes a powerful experimental paradigm for future investigations into vertebrate cell polarity. + +<--- Page Split ---> + +## Methods + +## Animal husbandry + +Zebrafish husbandry and experimental protocols were approved by the Hospital for Sick Children's Animal Care Committee, and all protocols were performed in accordance with Canadian Council on Animal Care guidelines. Wild- type zebrafish from TU strains were used. The \(v a n g l2\) mutant allele \(t r i^{k50f}\) was used for cell transplantation experiments \(^{21}\) , and the \(t r i^{m209}\) allele was used in to validate the \(v a n g l2^{y G F P}\) allele functionality because of the ability to use a published PCR mediated genotyping strategy \(^{21,43}\) . Tg(foxj1a::iCre) line has been previously described \(^{41}\) . Embryos from natural matings were grown at \(28^{\circ}\mathrm{C}\) unless otherwise indicated. + +## sfGFP-Vangl2 targeting strategy + +## Target Selection + +A previously validated guide RNA (gRNA) was used to target the sequence surrounding the start codon of \(v a n g l2\) (targeting sequence GGATAACGAGTCGCAGTACTCGG) \(^{44}\) . As a proof of principle, a solution containing \(300~\mathrm{mM}\) KCl, \(100~\mathrm{ng}\) \(v a n g l2\) gRNA, \(200~\mathrm{ng}\) recombinant Cas9 (PNA Bio cat. CP01) and \(10\%\) phenol red was incubated at \(37^{\circ}\mathrm{C}\) for 5 minutes to form ribonucleoprotein complexes (RNPs), and injected into wild- type embryos at the 1- cell stage. Embryos were (i) scored for \(v a n g l2\) /tribolite phenotypes at 28 hpf and (ii) assessed for loss of a ScaI restriction site by PCR to gauge mutagenesis efficiency. + +<--- Page Split ---> + +Prior to vector construction, male and female Tu wild- type animals were screened by PCR for polymorphisms surrounding vang12 exon 2. Left and right arms of homology of \(\sim 1 - 1.2\) kb in length were amplified from fin clips and subjected for Sanger sequencing. Fish containing highly similar amplicons were selected for use in targeting experiments (4 males and 4 females). A \(\sim 2.2\) kb amplicon surrounding vang12 exon 2 was amplified from these wildtype fish and cloned into pDONR221. A superfolder GFP (sfGFP) was amplified from sfGFP- N1 (a kind gift from Michael Davidson and Geoffrey Waldo, Addgene #54737) \(^{45}\) and cloned into the vang12 targeting intermediate with a 3x(GGGGS) flexible linker by megaprimer PCR \(^{46}\) and verified for in- frame inclusion. Site- directed mutagenesis was performed on this intermediate to abolish the vang12 gRNA targeting site to prevent vector from being cleaved in vivo. This final targeting intermediate was amplified and cloned into XhoI and EcoRI sites of pKHR5 (a kind gift from D. Grunwald, Addgene #74593). All targeting intermediates and final clones were fully verified by Sanger sequencing. + +## Knock-In Injections + +A \(2.5 \mu \mathrm{l}\) mix containing \(20 \mathrm{ng}\) of vang12 targeting vector, \(1.25 \mathrm{U}\) of I- SceI (NEB cat. R0694S), \(0.5 \mathrm{x}\) Cutsmart buffer and \(10\%\) phenol red was prepared. A second \(2.5 \mu \mathrm{l}\) mix containing \(300 \mathrm{mM}\) KCl, \(100 \mathrm{ng}\) vang12 gRNA and \(200 \mathrm{ng}\) recombinant Cas9 (PNA Bio cat. CP01) was incubated at \(37^{\circ} \mathrm{C}\) for 5 minutes. The two mixes were combined together and one nanolitre was injected into incrosses between pre- designated Tu wild- type animals. Mosaic GFP fluorescence was observed in several F0 embryos which were separated and grown to adulthood. Adult F0 animals were outcrossed to wildtype Tu fish and screened for GFP+ fluorescence. A founder transmitting \(36\%\) + +<--- Page Split ---> + +of GFP+ embryos was identified, and F1 embryos were screened by PCR using the following primers, F: CCGCGCTCTCCAGTCCGTCA, R: CGAGAGCTGCGTGAGTGTGAA, to ensure precise editing of the vangl2 locus. + +Line Establishment and Maintenance + +The F1 fish were incrossed to produce homozygous F2 fish. Embryos were screened either visually by GFP fluorescence or genotyped using the primers listed above. + +## Transgenesis + +Transgenesis was performed using standard Gateway- compatible procedures47. The zGrad open reading frame was amplified from pCS2(+)- zGrad (a kind gift from H. Knaut; Addgene #119716) and cloned into pDONR- P2rP3 to generate p3E- zGrad. To generate Tg( \(\beta\) actin2::loxP mCherry STOP loxP zGrad) zebrafish, p5E- \(\beta\) actin247, pME- loxP mCherry STOP loxP (a gift from K. Kwan), and p3E- zGrad were recombined into pDEST Tol2 pA2 transgenesis vector. Embryos were injected at the one- cell stage with 25 pg of assembled transgene and 25 pg of Tol2 mRNA. Embryos were sorted at 48 hpf for mCherry reporter expression and were subsequently grown to adulthood. Individuals were bred to TU wild- type zebrafish to generate stable F1 lines. Clones were fully verified by Sanger sequencing. + +## Plasmids and embryo microinjections + +Rab5c, 7 and 11a plasmids were a kind gift from B. Link, mCherry- LifeAct- 7 was a kind gift from M. Davidson (Addgene #54491), pCS2(+)- zGrad was a gift from H. Knaut (Addgene #119716)23, pME- mTagBFP- CAAX was a gift from N. Cole (Addgene # 75149)48, zebrafish + +<--- Page Split ---> + +codon- optimized mScarlet was a kind gift from T. Thiele, centrin- mCherry was a kind gift from + +A. Salic and GalT-RFP was a kind gift from J. Wallingford. N-terminal mCherry-linker was + +cloned into the Rab plasmids and mCherry-LifeAct and mTagBFP-CAAX were cloned into + +pCS2+ using Gateway technology (Thermo Fisher Scientific). The plasmids were linearized and + +sense-strand-capped mRNA was synthesized with the mMESSAGE mMACHINE system + +(Ambion). 15 pg membrane-localized monomeric RFP (mRFP), 20 pg mTagBFP-CAAX, 30 pg + +mCherry-LifeAct, 5- 25 pg mCherry- Centrin, 2.5 pg Arl13b- mScarlet, 10 pg mCherry- Rab5c, 10 + +pg mCherry- Rab7, 15 pg mCherry- Rab11a, 10 pg GalT- RFP and 50- 100 pg zGrad were injected + +at one- cell stage. + +## Cell transplantations + +Cell transplantations were performed at mid- blastula stages, as described previously49. For the + +early somite stage imaging, embryos were grown at \(25^{\circ}\mathrm{C}\) after transplantations. + +## Imaging + +Brightfield embryonic imaging was performed on Leica M80 microscope with a MC170 HD + +camera (Leica). Wide- field fluorescence imaging was performed on a Zeiss AXIO Zoom V16 + +microscope (EMS3/SyCoP3, Zeiss) and ORCA-Flash 4.0 C1140- 22C camera (Hamamatsu). + +Confocal imaging was performed on Leica TCS SP8 Lightning confocal with Leica DMI8 + +microscope, 40x/1.3 water immersion objective and HyD detectors using the 405, 488 and 552 + +laser lines (Leica). The images were deconvolved using the Lightning module. All time- lapse + +imaging was performed using a resonant scanner with 16 times averaging. Maximum intensity + +projections and time lapse images were recorded using Imaris (Bitplane). + +<--- Page Split ---> + +## Measurements of embryonic lengths and L/W ratio + +Embryo lengths were measured by manually drawing a line from the front of the head to the tip of the tail in ImageJ. Embryonic length/width ratios were calculated by measuring the widest and longest points of krox20 and myoD expression. Statistical testing was performed using GraphPad Prism (GraphPad Software). + +## Quantification of asymmetric Vang12 membrane-localization + +Comparison of Vang12 and mRFP membrane enrichment was performed using the spot function in Imaris. In short, membrane- signals for mRFP or sfGFP were decorated with spots, and the spots for the two channels were merged. A reference point was placed in the centre of each cell to determine the anterior- posterior (X) and apical- basal (Y) axis. The position of each spot was determined as their distance to the reference point. Mean fluorescence intensity for mRFP and sfGFP within each spot was calculated, and the \(10\%\) of brightest spots for each fluorophore were plotted against their normalized X and Y positions using GraphPad Prism (GraphPad Software). Statistical testing was performed using GraphPad Prism (GraphPad Software). + +## BB positioning quantification + +Basal body position was defined as a ratio between the distance of the basal body from the posterior membrane and the anterior- posterior floor plate cell length normalized to 1. Measurements were done in Imaris and statistical testing was performed using GraphPad Prism (GraphPad Software). The values were divided into three categories: posterior (0.66- 1), medial (0.33- 0.66) and anterior (0- 0.33). + +<--- Page Split ---> + +## Whole-mount in situ hybridization + +Antisense RNA probes were prepared using DIG RNA labelling kit (Roche) from linearized + +DNA. Whole mount RNA in situ hybridizations were performed according to Rose et al. \(^{50}\) . + +Samples were genotyped after imaging. + +## References + +1. Devenport, D. Tissue morphodynamics: Translating planar polarity cues into polarized cell behaviors. Semin. Cell Dev. Biol. doi:10.1016/j.semcdb.2016.03.012. (2016) + +2. Davey, C. F. & Moens, C. B. Planar cell polarity in moving cells: think globally, act locally. Development 144, 187–200 (2017). + +3. Humphries, A. C. & Mlodzik, M. From instruction to output: Wnt/PCP signaling in development and cancer. Curr. Opin. Cell Biol. 51, 110–116 (2018). + +4. Jussila, M. & Ciruna, B. Zebrafish models of non-canonical Wnt/planar cell polarity signalling: fishing for valuable insight into vertebrate polarized cell behavior. Wiley Interdiscip. Rev. Dev. Biol. 6. (2017). + +5. Butler, M. T. & Wallingford, J. B. Planar cell polarity in development and disease. Nat. Rev. Mol. Cell Biol. 18, 375–388 (2017). + +6. Carvajal-Gonzalez, J. M. & Mlodzik, M. Mechanisms of planar cell polarity establishment in Drosophila. F1000Prime Rep. 6, 98 (2014). + +7. Devenport, D. & Fuchs, E. Planar polarization in embryonic epidermis orchestrates global + +<--- Page Split ---> + +asymmetric morphogenesis of hair follicles. Nat. Cell Biol. 10, 1257–1268 (2008).8. Montcouquiol, M. et al. Asymmetric localization of Vangl2 and Fz3 indicate novel mechanisms for planar cell polarity in mammals. J. Neurosci. 26, 5265–5275 (2006).9. Butler, M. T. & Wallingford, J. B. Spatial and temporal analysis of PCP protein dynamics during neural tube closure. eLife 7, e36456 (2018).10. Heck, B. W. & Devenport, D. Trans-endocytosis of planar cell polarity complexes during cell division. Curr. Biol. 27, 3725-3733. e4 (2017).11. Love, A. M., Prince, D. J. & Jessen, J. R. Vangl2-dependent regulation of membrane protrusions and directed migration requires a fibronectin extracellular matrix. Development. 145, dev165472 (2018).12. Davey, C. F., Mathewson, A. W. & Moens, C. B. PCP Signaling between migrating neurons and their planar-polarized neuroepithelial environment controls filopodial dynamics and directional migration. PLOS Genet. 12, e1005934 (2016).13. Borovina, A., Superina, S., Voskas, D. & Ciruna, B. Vangl2 directs the posterior tilting and asymmetric localization of motile primary cilia. Nat. Cell Biol. 12, 407–412 (2010).14. Roszko, I., S. Sepich, D., Jessen, J. R., Chandrasekhar, A. & Solnica-Krezel, L. A dynamic intracellular distribution of Vangl2 accompanies cell polarization during zebrafish gastrulation. Development 142, 2508–2520 (2015).15. Devenport, D., Oristian, D., Heller, E. & Fuchs, E. Mitotic internalization of planar cell polarity proteins preserves tissue polarity. Nat. Cell Biol. 13, 893–902 (2011).16. Ciruna, B., Jenny, A., Lee, D., Mlodzik, M. & Schier, A. F. Planar cell polarity signalling + +<--- Page Split ---> + +couples cell division and morphogenesis during neurulation. Nature 439, 220–224 (2006). + +Panousopoulou, E., Tyson, R. A., Bretschneider, T. & Green, J. B. A. The distribution of Dishevelled in convergently extending mesoderm. Dev. Biol. 382, 496–503 (2013). + +Chu, C.-W. & Sokol, S. Y. Wnt proteins can direct planar cell polarity in vertebrate ectoderm. eLife 5, e16463 (2016). + +Witzel, S., Zimyanin, V., Carreira-Barbosa, F., Tada, M. & Heisenberg, C. P. Wnt11 controls cell contact persistence by local accumulation of Frizzled 7 at the plasma membrane. J. Cell Biol. 175, 791–802 (2006). + +Butler, M. T. & Wallingford, J. B. Control of vertebrate core planar cell polarity protein localization and dynamics by Prickle 2. Development 142, 3429–39 (2015). + +Jessen, J. R. et al. Zebrafish trilobite identifies new roles for Strabismus in gastrulation and neuronal movements. Nat. Cell Biol. 4, 610–615 (2002). + +Park, M. & Moon, R. T. The planar cell-polarity gene stbm regulates cell behaviour and cell fate in vertebrate embryos. Nat. Cell Biol. 4, 20–25 (2002). + +Yamaguchi, N., Colak-Champollion, T. & Knaut, H. ZGrad is a nanobody-based degron system that inactivates proteins in zebrafish. eLife 8, e43125 (2019). + +Petridou, N. I., Grigolon, S., Salbreux, G., Hannezo, E. & Heisenberg, C. P. Fluidization-mediated tissue spreading by mitotic cell rounding and non-canonical Wnt signalling. Nat. Cell Biol. 21, 169–178 (2019). + +Mottola, G., Classen, A.-K., González-Gaitán, M., Eaton, S. & Zerial, M. A novel function for the Rab5 effector Rabenosyn-5 in planar cell polarity. Development 137, + +<--- Page Split ---> + +26. Mathewson, A. W., Berman, D. G. & Moens, C. B. Microtubules are required for the maintenance of planar cell polarity in monociliated floorplate cells. Dev. Biol. 452, 21–33 (2019).27. Strutt, H. & Strutt, D. Differential stability of flamingo protein complexes underlies the establishment of planar polarity. Curr. Biol. 18, 1555–1564 (2008).28. Strutt, H., Warrington, S. J. & Strutt, D. Dynamics of core planar polarity protein turnover and stable assembly into discrete membrane subdomains. Dev. Cell 20, 511–525 (2011).29. Williams, B. B. et al. VANGL2 regulates membrane trafficking of MMP14 to control cell polarity and migration. J. Cell Sci. 125, 2141–2147 (2012).30. Sepich, D. S. & Solnica-Krezel, L. Intracellular Golgi Complex organization reveals tissue specific polarity during zebrafish embryogenesis. Dev. Dyn. 245, 678–691 (2016).31. Ossipova, O., Kim, K. & Sokol, S. Y. Planar polarization of Vangl2 in the vertebrate neural plate is controlled by Wnt and Myosin II signaling. Biol. Open 4, 722–730 (2015).32. Strutt, H., Gamage, J. & Strutt, D. Robust asymmetric localization of planar polarity proteins is associated with organization into signalosome-like domains of variable stoichiometry. Cell Rep. 17, 2660–2671 (2016).33. Stahley, S. N., Basta, L. P., Sharan, R. & Devenport, D. Celsr1 adhesive interactions mediate the asymmetric organization of planar polarity complexes. eLife 10, e62097 (2021).34. Shi, D. et al. Dynamics of planar cell polarity protein Vangl2 in the mouse oviduct + +<--- Page Split ---> + +epithelium. Mech. Dev. 141, 78- 89 (2016). + +Shindo, A., Inoue, Y., Kinoshita, M. & Wallingford, J. B. PCP- dependent transcellular regulation of actomyosin oscillation facilitates convergent extension of vertebrate tissue. Dev. Biol. 446, 159- 167 (2019). + +Riedl, J. et al. Lifeact: a versatile marker to visualize F- actin. Nat. Methods 5, 605- 607 (2008). + +Brunt, L. et al. Vangl2 promotes the formation of long cytonemes to enable distant Wnt/β- catenin signaling. Nat. Commun. 12, 2058 (2021). + +Oozeer, F., Yates, L. L., Dean, C. & Formstone, C. J. A role for core planar polarity proteins in cell contact- mediated orientation of planar cell division across the mammalian embryonic skin. Sci. Rep. 7, 1880 (2017). + +Carvajal- Gonzalez, J. M., Roman, A.- C. & Mlodzik, M. Positioning of centrioles is a conserved readout of Frizzled planar cell polarity signalling. Nat. Commun. 7, 11135 (2016). + +Donati, A., Anselme, I., Schneider- Maunoury, S. & Vesque, C. Planar polarization of cilia in the zebrafish floor- plate involves Par3- mediated posterior localization of highly motile basal bodies. Development 148, dev196386 (2021). + +Van Gennip, J. L. M., Boswell, C. W. & Ciruna, B. Neuroinflammatory signals drive spinal curve formation in zebrafish models of idiopathic scoliosis. Sci. Adv. 4, 1- 12 (2018). + +Heppert, J. K., Pani, A. M., Roberts, A. M., Dickinson, D. J. & Goldstein, B. A CRISPR + +<--- Page Split ---> + +tagging- based screen reveals localized players in Wnt- directed asymmetric cell division. Genetics 208, 1147–1164 (2018). + +Gao, B. et al. Wnt signaling gradients establish planar cell polarity by inducing Vangl2 phosphorylation through Ror2. Dev. Cell 20, 163–176 (2011). + +Shah, A. N., Davey, C. F., Whitebirch, A. C., Miller, A. C. & Moens, C. B. Rapid reverse genetic screening using CRISPR in zebrafish. Zebrafish 13, 152–153 (2016). + +Pédelacq, J.- D. et al. Engineering and characterization of a superfolder green fluorescent protein. Nat. Biotechnol. 24, 79–88 (2006). + +Geiser, M., Cèbe, R., Drewello, D. & Schmitz, R. Integration of PCR fragments at any specific site within cloning vectors without the use of restriction enzymes and DNA ligase. BioTechniques 31, 88–92 (2001). + +Kwan, K. M. et al. The Tol2kit: A multisite gateway- based construction Kit for Tol2 transposon transgenesis constructs. Dev. Dyn. 236, 3088–3099 (2007). + +Don, E. K. et al. A Tol2 Gateway- Compatible Toolbox for the study of the nervous system and neurodegenerative disease. Zebrafish 14, 69–72 (2017). + +Ho, R. K. & Kane, D. A. Cell- autonomous action of zebrafish spt- 1 mutation in specific mesodermal precursors. Nature 348, 728–730 (1990). + +Rose, C. D. et al. SCO- Spondin defects and neuroinflammation are conserved mechanisms driving spinal deformity across genetic models of idiopathic scoliosis. Curr. Biol. 30, 2363- 2373. e6 (2020). + +<--- Page Split ---> + +## Acknowledgements + +We gratefully acknowledge the SickKids' Imaging Facility for assistance with confocal microscopy and the SickKids' Zebrafish Facility technicians for excellent zebrafish care. This work was supported, in part, by funding from Canadian Institutes of Health Research (FDN- 167285) and the Canada Research Chair program to B.C., and the Sigrid Juselius Foundation to M.J. + +## Author Contributions + +M.J. and C.W.B. performed Vangl2 targeting, M.J. performed all Vangl2 knock- in line validation and Vangl2 imaging experiments and analysis, C.W.B. generated the zGrad transgene, B.C. supervised research studies, M.J. and B.C. wrote the manuscript with input from C.W.B. + +## Declaration of Interests + +The authors declare no competing interests. + +## Supplementary Movie Titles + +Supplementary Movie 1. Vangl2 enrichment on anterior membranes is dynamic over time Supplementary Movie 2. Vangl2 enrichment on the anterior membrane does not colocalize with filamentous actin Supplementary Movie 3. Vangl2 localizes to anterior apical membranes and to basal membrane extensions in dividing cells Supplementary Movie 4. Vangl2 anterior polarization is lost during cell division + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
a
+ +Figure 1. sfGFP- Vang12 is functional and membrane localized during early zebrafish development + +(a) A schematic illustration of the sfGFP-Vang12 knock-in targeting strategy. + +(b) Wide-field fluorescence images demonstrating sfGFP-Vang12 expression in \(vang12^{sfGFP + }\) and \(vang12^{sfGFP / sfGFP}\) embryos at 28 hours post fertilization (hpf). + +(c) Lateral view of wild-type, \(vang12^{sfGFP / sfGFP}\) , \(vang12^{sfGFP / m209}\) and \(vang12^{m209 / m209}\) embryos at 24 hpf. + +(d) Embryo length quantification at 24 hpf. Data shows representative experiments combined from independent crosses that were repeated 2-4 times. Statistical analysis was performed using one-way ANOVA with Tukey's multiple comparisons test (****P<0.0001). Wild-type \(n = 14\) , \(vang12^{sfGFP + }\) \(n = 22\) , \(vang12^{sfGFP / sfGFP}\) \(n = 20\) , \(vang12^{m209 / + }\) \(n = 82\) , \(vang12^{sfGFP / m209}\) \(n = 72\) , \(vang12^{m209}\) \(n = 63\) . + +(e) Live confocal images of \(vang12^{sfGFP / sfGFP}\) (hereafter referred as \(vang12^{sfGFP}\) ) embryos at blastula through early gastrula stages, as indicated. Ectodermal cells were imaged in gastrulating embryos. Embryos were injected with mRNA coding for a membrane-localized monomeric RFP (mRFP) reporter. All sfGFP images were acquired using identical settings. + +(f-i) Live confocal images of shield staged \(vang12^{sfGFP}\) embryos injected with mRNA coding for mCherry-Rab5c (f), mCherry-Rab7 (g), mCherry-Rab11a (h) or GalT-RFP (i) reporter constructs. Arrows point at cytoplasmic Vang12 puncta in close proximity to respective reporters, and arrowheads point at isolated Vang12 puncta. + +<--- Page Split ---> + +
Figure 2. Jussila et al.
+ +![](images/Figure_3.jpg) + + +<--- Page Split ---> + +Figure 2. Planar polarized enrichment of Vang12 on anterior neuroepithelial membranes is controlled by cell non- autonomous PCP signalling + +(a) Confocal images comparing sfGFP-Vang12 and mRFP localization at 22 hours post fertilization (hpf) in dorsal (top) and lateral (bottom) orientations. Anterior is to the left in all images. Strong Vang12 expression is observed on cell membranes in the neural tube (NT) including the floor plate (FP), while only weak expression is observed in somites (S) and notochord (NC). + +(b) A schematic illustrating transplantation of mRFP-labelled vang12sfGFP donor cells into WT (wild-type) host embryos, at sphere stage. Chimeric embryos were imaged dorsally, at positions between the \(2^{\mathrm{nd}}\) and \(8^{\mathrm{th}}\) somite. +(c) Confocal images of mRFP and sfGFP-Vang12 localization in the developing spinal cord of chimeric embryos at 4-somite, 10-somite and 13-somite stages of development. Maximum intensity projections are shown. Arrows point to Vang12 enrichment on anterior neuroepithelial cell membranes, and arrowheads to anterior apical membranes at the neural midline. + +(d-e) Distribution of the brightest sfGFP-Vang12 and membrane-RFP spots along anterior-posterior (d) and apical-basal (e) axes of vang12sfGFP neuroepithelial cells, quantified at four consecutive stages of neural tube morphogenesis. Data from multiple transplanted vang12sfGFP cells was pooled (3-5 somites, \(\mathrm{n} = 7\) ; 7 somites, \(\mathrm{n} = 7\) ; 8-10 somites, \(\mathrm{n} = 6\) ; 13 somites, \(\mathrm{n} = 5\) ) and axial lengths normalized to 1. Details of the quantification can be found in the Methods section. Statistical analysis was performed using a two-tailed Mann-Whitney test. + +(f) A schematic illustrating transplantation of mRFP-labelled vang12sfGFP donor cells into vang12k50f loss-of-function host embryos, at sphere stage. + +(g-h) Distribution of the brightest sfGFP-Vang12 and membrane-RFP spots along anterior-posterior (g) and apical-basal (h) axes of vang12sfGFP neuroepithelial cells transplanted into a vang12k50f mutant host, quantified at two consecutive stages of neural tube morphogenesis. Data from multiple transplanted vang12sfGFP cells was pooled (10 somites, \(\mathrm{n} = 7\) ; 12-13 somites, \(\mathrm{n} = 9\) ). + +(i) Confocal images of mRFP and sfGFP-Vang12 localization in vang12sfGFP neuroepithelial cells within vang12k50f mutant hosts, at 10-somite and 13-somite stages of development. Maximum intensity projections are shown. Arrows point at Vang12 enrichment on cell protrusions. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 3. Jussila et al.
+ +(a) Confocal images of sfGFP-Vang12 localization in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters within floorplate cells of a vang12sfGFP embryo at 22hpf. Lateral view, anterior to the left. +(b) Confocal images of sfGFP-Vang12 localization in relation to cilia axoneme (Arl13b-mScarlet) and membrane (mTagBFP) reporters within floorplate cells of a vang12sfGFP embryo at 32hpf. Lateral view, anterior to the left. +(c) Confocal image comparing sfGFP-Vang12 and mRFP localization in floorplate cells of a 23-somite staged chimeric embryo (mRFP mRNA-injected vang12sfGFP cells transplanted into WT hosts). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 enrichment on anterior cell membranes. Arrowhead points at Vang12 enriched membrane protruding from the anterior apical surface. + +(d) Confocal image comparing sfGFP-Vang12 and Centrin-mCherry localization in floorplate cells of a 15-somite staged chimeric embryo (Centrin-mCherry mRNA-injected vang12sfGFP cells transplanted into WT hosts). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 on the anterior membrane that is closely associated with a basal body docked on the posterior membrane. Asterisk indicates anterior unlabelled host cell that is not Centrin-mCherry positive. + +(e) Confocal image comparing sfGFP-Vang12 localization in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters in neighbouring floorplate cells of a 15-somite staged chimeric embryo (vang12sfGFP cells transplanted into WT hosts injected with Centrin-mCherry and mTagBFP mRNA). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 on the anterior membrane of a donor cell that is closely associated with a basal body docked on the posterior membrane of a host cell. + +<--- Page Split ---> +![PLACEHOLDER_29_0] + +
Figure 4. Jussila et al.
+ +Figure 4. Conditional degradation of sfGFP-Vang12 reveals a cell non-autonomous requirement for Vang12 in maintaining basal body polarity + +(a) Quantification of embryo lengths for zGrad mRNA-injected and control embryos at 24 hpf. Maternal and zygotic genotypes are as indicated. Data is pooled from three independent experiments. Control n from left to right: n=53, n=62, n=61, n=36. zGrad n from left to right: n=55, n=51, n=53, n=54. Statistical analysis was performed using Kruskal-Wallis test with Dunn's multiple comparisons test. + +(b) Confocal images of sfGFP-Vang12 expression in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters within floorplate cells of control vang12sfGFP; Tg(loxP-mCherry-STOP-loxP-zGrad) embryos (left) and experimental vang12sfGFP; Tg(loxP-mCherry-STOP-loxP- zGrad); Tg(foxj1a-iCre) embryos (right) at 48 hpf. Dorsal view, anterior to the left. Arrows indicate residual sfGFP-Vang12 expression in a subset of floorplate cells following zGrad expression. Asterisks indicate mispolarized basal bodies following zGrad expression. + +(c) Quantification of basal body localization within the posterior, medial or anterior thirds of floor plate cells for control and vang12sfGFP; Tg(loxP-mCherry-STOP-loxP-zGrad); Tg(foxj1a-cre) embryos described in (b). Cells were categorized as A-D based on the presence or absence of sfGFP-Vang12 signal within both the cell and its immediate neighbours, as illustrated in (d). Control n=164, A n=256, B n=20, C n=29, D n=19. + +(d) A schematic illustrating cell categories quantified in (c). Grey colour indicates cells that have lost sfGFP-Vang12 signal; green colour indicates residual sfGFP-Vang12 expression. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- JussilaetalMovie3NatComm.mp4- JussilaetalMovie2NatComm.mp4- JussilaetalSupplementaryFiguresNatComm.pdf- JussilaetalMovie1NatComm.mp4- JussilaetalMovie4NatComm.mp4- JussilaetalSupplementaryFiguresNatComm.pdf- nrreportingsummaryNCOMMS2129897.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae_det.mmd b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..018a4d898dba47a4aa30d209643ef011d6c29b27 --- /dev/null +++ b/preprint/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae/preprint__11db3cfd5c744972ee0a8b55468e38de752c594b0359a992b41825002038edae_det.mmd @@ -0,0 +1,588 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 896, 177]]<|/det|> +# Live imaging and conditional disruption of native vertebrate planar cell polarity + +<|ref|>text<|/ref|><|det|>[[44, 196, 377, 333]]<|/det|> +Maria Jussila The Hospital for Sick Children Curtis Boswell Yale University School of Medicine Brian Ciruna ( \(\square\) ciruna@sickkids.ca) The Hospital for Sick Children + +<|ref|>sub_title<|/ref|><|det|>[[44, 372, 102, 389]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 409, 137, 427]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 446, 318, 466]]<|/det|> +Posted Date: August 18th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 485, 463, 504]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 764931/v1 + +<|ref|>text<|/ref|><|det|>[[44, 522, 910, 565]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 601, 914, 644]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 23rd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33322- 9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[70, 88, 789, 110]]<|/det|> +# Live imaging and conditional disruption of native vertebrate planar cell polarity + +<|ref|>text<|/ref|><|det|>[[70, 140, 88, 154]]<|/det|> +2 + +<|ref|>text<|/ref|><|det|>[[70, 179, 575, 200]]<|/det|> +Maria Jussila \(^{1}\) , Curtis W. Boswell \(^{1,2,3}\) and Brian Ciruna \(^{1,2,*}\) + +<|ref|>text<|/ref|><|det|>[[70, 228, 88, 242]]<|/det|> +4 + +<|ref|>text<|/ref|><|det|>[[70, 268, 852, 290]]<|/det|> +1. Program in Developmental & Stem Cell Biology, The Hospital for Sick Children, 686 Bay + +<|ref|>text<|/ref|><|det|>[[70, 304, 468, 323]]<|/det|> +Street, Toronto, Ontario, M5G 0A4, Canada. + +<|ref|>text<|/ref|><|det|>[[70, 348, 861, 370]]<|/det|> +2. Department of Molecular Genetics, The University of Toronto, Toronto, Ontario, M5S 1A8, + +<|ref|>text<|/ref|><|det|>[[70, 385, 180, 403]]<|/det|> +8 Canada. + +<|ref|>text<|/ref|><|det|>[[70, 428, 861, 449]]<|/det|> +3. Current address: Department of Genetics, Yale University School of Medicine, New Haven, + +<|ref|>text<|/ref|><|det|>[[70, 464, 250, 482]]<|/det|> +10 CT 06510, USA. + +<|ref|>text<|/ref|><|det|>[[70, 508, 525, 528]]<|/det|> +11 \* Corresponding author. Email: ciruna@sickkids.ca + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 91, 191, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 131, 868, 608]]<|/det|> +Tissue- wide coordination of polarized cytoskeletal organization and cell behaviour, critical for organ morphogenesis and tumour progression, is controlled by asymmetric membrane localization of non- canonical Wnt/planar cell polarity (PCP) signalling components. Understanding the dynamic regulation of PCP thus requires visualization of these core proteins. In vertebrates, immunohistochemical studies have provided snapshots of PCP within tissues while exogenous, fluorescently- tagged reporters have been introduced for live imaging of cell polarity. However, the functionality and spatiotemporal relevance of static and exogenous PCP readouts remain uncertain. Here we fluorescently tag an endogenous core PCP component, Vangl2, in zebrafish. We report on the authentic regulation of vertebrate PCP, through live imaging of this native sfGFP- Vangl2 protein during embryogenesis. We couple sfGFP- Vangl2 with conditional zGrad GFP- nanobody degradation methodologies for tissue- specific interrogation of PCP function. Together, our studies provide crucial insights into the establishment and maintenance of vertebrate PCP and create a powerful experimental paradigm for future investigations. + +<|ref|>sub_title<|/ref|><|det|>[[115, 633, 225, 650]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[111, 675, 884, 905]]<|/det|> +The non- canonical Wnt/planar cell polarity (PCP) signalling pathway controls polarized, uniform orientation of cells within a plane of a tissue. PCP pathway has been shown to regulate a number of developmental and homeostatic processes by directing collective asymmetric modification of the cell cytoskeleton, leading groups of cells to either move or divide in a shared direction, or position their organelles such as the basal body asymmetrically \(^{1 - 5}\) . Molecular genetic studies in Drosophila have identified essential roles for asymmetric membrane localization of opposing PCP signalling proteins in both the establishment and intercellular propagation of polarity \(^{6}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 876, 633]]<|/det|> +Despite identification of a multitude of PCP- regulated morphogenetic processes, characterization of vertebrate PCP at the cellular and molecular level is still lacking. This is challenging, because detailed observation requires live imaging capabilities and the ability to differentiate asymmetric protein localization across membranes of adjacent cells. In vertebrates, immunohistochemical studies have provided snapshots of PCP within tissues7,8 while exogenous, fluorescently- tagged reporter proteins have been introduced for mosaic imaging of PCP. Although great complexities in the dynamic regulation of PCP have been reported across tissues and polarized cell behaviours,9- 12 the functionality and spatiotemporal accuracy of these static and exogenous readouts remain uncertain. The biological relevance of studies utilizing exogenous, fluorescently- tagged PCP molecules remains to be determined, as overexpression of PCP proteins can disrupt cell polarity11,13 and functionality of published PCP fusion proteins has not been validated via rescue of orthologous mutant phenotypes.12,14- 16 Furthermore, non- physiological expression of exogenous proteins can mask subtle localization patterns or signalling dynamics. This is highlighted by observations that in many occasions, co- expression of another PCP component is required for asymmetric membrane localization of the fluorescent fusion protein of interest.17- 20 + +<|ref|>text<|/ref|><|det|>[[110, 655, 876, 888]]<|/det|> +To understand the dynamic regulation of native PCP, we have aimed to overcome limitations imposed by exogenous cell polarity reporters by imaging and manipulating the endogenous zebrafish Vangl2 protein, a core PCP component and essential regulator of vertebrate planar polarity.21,22 Using CRISPR/Cas9 gene editing, we introduced a superfolder GFP (sfGFP) tag onto the N- terminus of endogenous Vangl2 to generate a functional and native PCP reporter protein. By performing detailed embryonic cell transplantation and single- cell level analyses, we documented the dynamic sub- cellular localization of Vangl2 over the course of neural tube + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 875, 355]]<|/det|> +formation, revealing novel associations of sfGFP- Vang12 positive cell membranes with polarized basal bodies in floorplate cells. Furthermore, we exploited zGrad \(^{23}\) GFP- specific protein degradation methodologies to conditionally interrogate sfGFP- Vang12 function in floor plate cells, demonstrating an essential role for Vang12 in both the establishment and maintenance of polarized basal body positioning as well as an inherent and surprising directionality to the intercellular propagation of PCP. Our work highlights the importance and utility of studying endogenous proteins in the context of PCP, and opens the door to conditional analysis of PCP function beyond embryonic development. + +<|ref|>sub_title<|/ref|><|det|>[[115, 424, 308, 442]]<|/det|> +## Results and Discussion + +<|ref|>text<|/ref|><|det|>[[111, 466, 880, 909]]<|/det|> +To visualize the dynamic regulation of endogenous PCP signalling in zebrafish, we used CRISPR/Cas9 genome editing to target a superfolder green fluorescent protein (sfGFP) onto the N- terminus of Vang12, an essential and membrane- localized PCP signalling protein (Fig. 1a; targeting details in Methods section). Embryos carrying the knock- in allele showed prominent sfGFP signal in the brain and spinal cord at 22 hours post fertilization (hpf) (Fig. 1b). In striking contrast to the \(vangl2^{m209 / 209}\) loss- of- function mutant phenotype, which is embryonic lethal \(^{21}\) , fish homozygous for the \(vangl2^{sfGFP}\) allele or trans- heterozygote for \(vangl2^{sfGFP}\) and \(vangl2^{m209}\) alleles are viable, fertile and appear morphologically normal (Fig. 1c). Closer inspection revealed that \(vangl2^{sfGFP / sfGFP}\) embryos are slightly but significantly shorter than wildtype siblings at 24 hpf (Fig. 1d). Interestingly, heterozygote \(vangl2^{m209 / + }\) embryos are similarly shorter than wildtype, which has not been previously reported. \(Vangl2^{sfGFP / m209}\) transheterozygote embryos demonstrate a further shortening of the body axis but are significantly less affected than the \(vangl2^{m209 / m209}\) mutants (Fig. 1d). Altogether these data indicate that, although mildly hypomorphic, the sfGFP + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 848, 144]]<|/det|> +Vangl2 fusion protein is functional and can therefore be used to interrogate endogenous PCP signalling dynamics. + +<|ref|>text<|/ref|><|det|>[[110, 165, 877, 644]]<|/det|> +To determine the onset of Vangl2 expression, we imaged vangl2sfGFP/sfGFP embryos (hereafter referred to as vangl2sfGFP) through blastula and gastrula stages. Weak Vangl2 membrane localization could be observed in a subset of cells as early as high stage (3.3 hpf), which became more ubiquitous through early epiboly stages (4.6 hpf; Fig. 1e). Interestingly, weak Vangl2 membrane- localization was also observed within enveloping layer (EVL) cells at these stages (Supplementary Fig. 1a). These observations are consistent with a role for Vangl2 prior to gastrulation, perhaps in promoting tissue cohesion in margin cells during early blastoderm morphogenesis.24 Vangl2 expression levels increased, and membrane localization became more prominent with the onset of gastrulation at shield stage, persisting through \(75\%\) epiboly and bud stages (Fig. 1e, Supplementary Fig. 1b- e). This corresponds to the known role for Vangl2 in the regulation of convergence and extension movements.21 Although immunohistochemical analyses of Vangl2 expression have yielded similar observations14 the superior consistency, sensitivity and live- imaging capabilities of sfGFP- Vangl2 expression highlight its utility for experimental studies. + +<|ref|>text<|/ref|><|det|>[[110, 666, 880, 896]]<|/det|> +In addition to membrane- localized Vangl2, we also observed punctate intracellular accumulations, which became prominent at the onset of gastrulation when they concentrated into large cytosolic aggregates (Fig. 1e). To investigate the identity of these compartments, we analysed Vangl2 localization in relation to known markers of intracellular sorting organelles (Fig. 1f- i and Supplementary Fig. 1b- e). In published studies, early and late endosome markers Rab5 and Rab7 colocalize with PCP components Flamingo/Celsr and Frizzled- 3a in Drosophila wing, mouse keratinocytes and zebrafish neural tube.10,15,25- 27 The recycling endosome marker + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 876, 530]]<|/det|> +Rab11 also colocalizes with Flamingo/Celsr in mouse skin and in Drosophila wing, and with Vangl2 in cell culture. \(^{15,28,29}\) However, vertebrate studies have used exogenous PCP reporters, which might not recapitulate the endogenous subcellular protein localization and may induce trafficking artifacts. While we observed that Rab5c colocalized with some of the smaller Vangl2 puncta at both shield and \(90\%\) epiboly stages (Fig. 1f and Supplementary Fig. 1b), and that Rab11a colocalized with Vangl2 at the cell membrane but not within the cytoplasm (Fig. 1h and Supplementary Fig. 1d), the majority of Vangl2 puncta colocalized with mCherry- Rab7 (Fig. 1g and Supplementary Fig. 1c) suggesting late endosomal transport to lysosomes or the trans- Golgi network (TGN). Prior to gastrulation the zebrafish Golgi is fragmented, \(^{30}\) and whereas we could see occasional sfGFP- Vangl2 signal in the proximity of the TGN reporter dsRed- GalT, Vangl2 was predominantly localized elsewhere in the cell (Fig. 1i and Supplementary Fig. 1e). Taken together these results suggest that, at early gastrulation stages, Vangl2 is targeted to late endosomes for degradation. + +<|ref|>text<|/ref|><|det|>[[110, 551, 883, 888]]<|/det|> +Following gastrulation, strong Vangl2 expression was observed in the central nervous system, whereas only weak sfGFP- Vangl2 signal could be detected in surrounding axial and paraxial mesodermal lineages (Figs. 1b, 2a). Previous work has demonstrated that exogenous GFPTagged Vangl2 or its cytoplasmic binding partner Prickle polarize anteriorly in neuroepithelial cells in zebrafish and Xenopus. \(^{9,12,14,16,26,31}\) To characterize the asymmetric distribution of endogenous Vangl2, we performed cell transplantation at blastoderm stages to generate clones of \(vangl2^{yGGFP}\) cells within wildtype host embryos and imaged Vangl2 localization in individual neuroepithelial cells at multiple time points during neural tube formation (Fig. 2b). At all stages, weak sfGFP- Vangl2 signal could be observed across the entire neuroepithelial cell membrane, with bright membrane sfGFP- Vangl2 domains enriched in puncta at anterior cell surfaces (Fig. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 530]]<|/det|> +2c, Supplementary Fig.2a, Supplementary Movie 1). Vangl2 localization varied considerably between cells, including apical and basal distribution patterns, as well as localization into cell protrusions (Fig. 2c and Supplementary Fig.2a). Previously, exogenous Vangl2 reporter localization has been quantified along single cellular planes. To gain a global view of Vangl2 distribution, we performed 3- dimensional quantification of sfGFP- Vangl2 membrane- enrichment in comparison to an RFP membrane reporter (mRFP) using the spot function in Imaris (detailed description in Methods). We plotted the subset of membrane spots containing the highest sfGFP- Vangl2 and mRFP fluorescent intensities across anterior- posterior and apico- basal cell axes (Supplementary Fig.3A) and found that, despite the variation between cells, Vangl2 was significantly enriched at the anterior membrane at all stages analysed (Fig. 2d). Vangl2 localization was less uniformly polarized along the apico- basal axis, with Vangl2 showing strongest apical enrichment at the 13 somite- stage concomitant to neural tube midline formation \(^{16}\) (Fig. 2e). + +<|ref|>text<|/ref|><|det|>[[111, 551, 883, 886]]<|/det|> +To examine the dynamics of Vangl2 protein localization, we performed time- lapse imaging of \(vangl2^{yGFP}\) chimeric embryos between 10- 13 somite stages. Persistent sfGFP- Vangl2 enrichment was observed along anterior membrane domains (Supplementary Movie 1), supporting evidence from multiple Drosophila, Xenopus and mouse studies (predominantly using exogenous reporters) that PCP components concentrate into potential "signalosomes" on the cell membrane. \(^{9,16,20,28,32 - 35}\) In Xenopus, exogenous Vangl2 and Prickle reporters accumulate with actomyosin at shrinking neuroepithelial cell junctions \(^{9}\) and, in mesodermal cells undergoing convergence and extension movements, Prickle2 colocalizes with the LifeAct filamentous actin reporter. \(^{35,36}\) These observations suggest that Vangl2/PCP directly influences cytoskeletal reorganization. To investigate the relationship between anterior Vangl2 enrichment and actin + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 881, 458]]<|/det|> +organization, we co- expressed a mCherry- LifeAct reporter in vangl2sfGFP cells. Live imaging revealed that the majority of neuroepithelial cell protrusive activity and LifeAct localization was restricted to apical and basal surfaces, and not to anterior membranes where sfGFP- Vangl2 was enriched (Supplementary Movies 2- 3). However, sfGFP- Vangl2 could be observed accumulating within active cellular protrusions at the anterior membrane (Supplementary Fig.2a, Supplementary Movie 4), consistent with known roles for PCP and Vangl2 in regulating filopodia and cell protrusive activity during gastrulation and facial branchiomotor neuron migration.11,12,37 Together, although the presence of bright sfGFP- Vangl2 puncta supports the notion that PCP signalling functions in specific subdomains on the anterior membrane, the presence of Vangl2 membrane enrichment was not predictive of a particular membrane behaviour. + +<|ref|>text<|/ref|><|det|>[[110, 481, 870, 852]]<|/det|> +To further understand how neighbouring cells affect asymmetric Vangl2 localization into different membrane- domains, we performed cell transplantation experiments to generate clones of vangl2sfGFP cells within a vangl2tk50f loss- of- function host embryo (Fig. 2f). In this PCP- deficient environment, vangl2sfGFP cells clustered together, suggesting that Vangl2 expressing cells are able to recognize and preferentially interact with one other. While Vangl2 remained localized to the cell membrane, planar polarized membrane enrichment was lost and Vangl2 appeared to concentrate at cell protrusions (Fig. 2i and Supplementary Fig.2b). We picked a subset of the most mediolaterally aligned cells to quantify asymmetric Vangl2 localization and found no difference in the distribution of the brightest Vangl2 and mRFP spots along the anterior- posterior or apico- basal axes (Fig. 2g- h; Supplementary Fig. 3d). Loss of polarity in vangl2sfGFP positive clones supports a cell non- autonomous regulation of vertebrate PCP. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 875, 140]]<|/det|> +persistence of Vangl2 in cell protrusions suggests that it might function in mediating local signalling that is independent of its role in global anterior- posterior neural tube planar polarity. + +<|ref|>text<|/ref|><|det|>[[110, 165, 881, 219]]<|/det|> +The re- organization and re- establishment of cell polarity over the course of cell division remains a poorly understood feature of vertebrate PCP. In zebrafish, asymmetric localization of + +<|ref|>text<|/ref|><|det|>[[110, 231, 881, 280]]<|/det|> +exogenous Prickle- GFP reporter molecules is lost upon neuroepithelial cell division, but is re- established as daughter cells reintegrate into the neuroepithelium. \(^{16}\) In mouse epidermis, different + +<|ref|>text<|/ref|><|det|>[[110, 293, 880, 340]]<|/det|> +studies have shown PCP components to be internalized during mitosis or to persist on cell membranes in a polarized manner. \(^{15,38}\) Exogenous Vangl2, specifically, has been shown to + +<|ref|>text<|/ref|><|det|>[[110, 352, 870, 373]]<|/det|> +persist on cell membranes and to get trans- internalized from neighbouring cell membranes into + +<|ref|>text<|/ref|><|det|>[[110, 385, 835, 408]]<|/det|> +the dividing cell. \(^{10}\) To determine the fate of endogenous Vangl2 during mitosis, we imaged + +<|ref|>text<|/ref|><|det|>[[110, 420, 856, 442]]<|/det|> +vangl2 \(^{9}\) GFP neuroepithelial cells over the course of cell division. Notably, asymmetric sfGFP- + +<|ref|>text<|/ref|><|det|>[[110, 454, 840, 475]]<|/det|> +Vangl2 distribution was lost upon cell rounding, but Vangl2 remained at the cell membrane + +<|ref|>text<|/ref|><|det|>[[110, 486, 875, 508]]<|/det|> +(Supplementary Movie 3- 4). We observed sfGFP- Vangl2 enriched at the poles of dividing cells, + +<|ref|>text<|/ref|><|det|>[[110, 520, 765, 540]]<|/det|> +as well as to basolateral membrane projections that are maintained during division + +<|ref|>text<|/ref|><|det|>[[110, 552, 877, 574]]<|/det|> +(Supplementary Movies 3- 4). Although some intracellular sfGFP- Vangl2 signal was observed in + +<|ref|>text<|/ref|><|det|>[[110, 585, 810, 607]]<|/det|> +dividing cells, we did not observe significant endocytosis of membrane Vangl2 or trans- + +<|ref|>text<|/ref|><|det|>[[110, 618, 870, 640]]<|/det|> +endocytosis between neighbouring cells (Supplementary Movies 3- 4). Reacquisition of Vangl2 + +<|ref|>text<|/ref|><|det|>[[110, 652, 880, 673]]<|/det|> +polarization in daughter cells was not obvious immediately upon division (Supplementary Movie + +<|ref|>text<|/ref|><|det|>[[110, 685, 875, 707]]<|/det|> +2). Of interest, sfGFP- Vangl2 positive membrane protrusions from distant neighbours could be + +<|ref|>text<|/ref|><|det|>[[110, 718, 875, 740]]<|/det|> +observed interacting with dividing cells (Supplementary Movie 4). Out of 24 cells that had an + +<|ref|>text<|/ref|><|det|>[[110, 752, 880, 773]]<|/det|> +anterior Vangl2 positive protrusion, 11 were touching a dividing cell, and the same was observed + +<|ref|>text<|/ref|><|det|>[[110, 785, 880, 806]]<|/det|> +for 15/32 apical protrusions (total n=112 cells). Our results show that Vangl2 remains on the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 875, 145]]<|/det|> +membrane but loses its asymmetry during mitosis, and that re- establishment of cell polarity may be regulated by contacts from neighbouring cells. + +<|ref|>text<|/ref|><|det|>[[111, 163, 881, 860]]<|/det|> +In addition to cytoskeletal organization, polarized basal body (BB) positioning has been identified as a conserved readout of PCP from insects to vertebrates. \(^{39}\) In zebrafish floorplate cells, BBs are localized to the posterior apical membrane by 24 hpf, and the establishment of BB polarity requires Vangl2. \(^{13,26,40}\) To examine the relationship between Vangl2 localization and BB positioning, we imaged the floorplate of \(vangl2^{sfGFP}\) embryos at 22 and 32hpf. In the sagittal plane, sfGFP- Vangl2 was enriched on the anterior apical cell membrane (Fig. 3a- b), in accordance with observations using an exogenous GFP- Vangl2 transgenic reporter. \(^{12,26}\) However, in some floorplate cells apically enriched Vangl2 appeared to extend anteriorly towards neighbouring cells (Fig. 2a, Fig. 3a- b). This bright apical Vangl2 signal was closely associated with BBs labelled with Centrin- mCherry, and with Arl13b- positive cilia (Fig. 3a- b). We imaged individual transplanted \(vangl2^{sfGFP}\) floorplate cells and found that Vangl2 was indeed localized asymmetrically on anterior cell membranes, and that in some cells it further localized to an anterior apical projection (Fig. 3c). Labelling BBs in transplanted \(vangl2^{sfGFP}\) cells with Centrin- mCherry confirmed the proximity of anterior Vangl2 with posteriorly polarized BBs (Fig. 3d). Furthermore, labelling of host cell BBs confirmed that Vangl2- positive apical membrane closely associated with the posterior BB of neighbouring cells (Fig. 3e, 11/17 cells). Of interest exogenous Frizzled3a, a putative intracellular binding partner of Vangl2, has been shown to be enriched posteriorly into puncta at the base of cilia in zebrafish. \(^{26}\) This localization pattern suggests that PCP activity could act to physically deform the floorplate cell membranes at the site of BB docking, perhaps influencing the posterior tilting of the floorplate motile cilia. \(^{13}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 881, 460]]<|/det|> +To determine whether Vang12 also controls the maintenance of BB polarity, we took advantage of zGrad GFP- nanobody protein degradation methodologies to conditionally manipulate endogenous sfGFP- Vang12 protein levels. \(^{23}\) We first confirmed that zGrad is non- toxic to embryos, and that zGrad- injected vang12sfGFP/sfGFP embryos phenocopy vang12 loss- of- function mutants (Supplementary Fig. 4a- b). Furthermore, because vang12 mutants are subject to maternal effects \(^{16}\) , we investigated the consequences of zGrad expression across a full genotypic spectrum of maternal and zygotic vang12sfGFP alleles. Strikingly, sequential zGrad- mediated degradation of maternal and zygotic vang12sfGFP gene products yielded progressively more severe phenotypes as measured by embryo length and axial length- width ratios (Fig. 4a, Supplementary Fig. 4c), recapitulating maternal and zygotic vang12 mutant phenotypes. This highlights the sensitivity and functionality of sfGFP- Vang12 zGrad protein degradation methods. + +<|ref|>text<|/ref|><|det|>[[110, 479, 875, 888]]<|/det|> +To conditionally manipulate sfGFP- Vang12 protein levels within the floorplate, we generated a ubiquitously expressed Tg(β- actin2::loxP- mCherry- STOP- loxP- zGrad) transgene and crossed it to the Tg(foxj1a::iCre) line \(^{41}\) , which expresses Cre recombinase (activating zGrad) specifically within motile- ciliated cell lineages. At 32 hpf, when BB polarity has been established, sfGFP- Vang12 was still present at the membrane of experimental zGrad; foxj1a::iCre; vang12sfGFP floorplate cells (data not shown). However, mosaic loss of sfGFP- Vang12 could be observed by 48 hpf (Fig. 4b), allowing us to analyse both cell- autonomous and non- autonomous roles for Vang12 in the maintenance of BB polarity. Images were scored for the loss of Vang12 in individual cells and their immediate anterior and posterior neighbours. Irrespective of their Vang12 expression status, cells surrounded by sfGFP- negative neighbours demonstrated a loss of BB polarity when compared to control populations. Strikingly, the presence of sfGFP- Vang12 in a posterior neighbouring cell was associated with restored BB polarity in Vang12- negative cells, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 877, 248]]<|/det|> +even if the posterior neighbour demonstrated abnormal BB positioning (Fig. 4c- d). In contrast, the presence of sfGFP- Vangl2 in anterior neighbours did not restore polarity of Vangl2 negative cells. These results demonstrate that Vangl2 is required for the maintenance of BB polarity, and that Vangl2 functions non- cell autonomously to propagate PCP in a posterior to anterior direction. + +<|ref|>text<|/ref|><|det|>[[110, 272, 880, 746]]<|/det|> +Taken together, our work provides essential new insights into the establishment and maintenance of vertebrate PCP. Live imaging of endogenous Vangl2 has authenticated broad principles predicted using exogenously introduced reporter molecules, including organization of PCP across the embryonic anterior- posterior axis and concentration of core PCP proteins into 'signalosomes' at the cell membrane. It has also yielded novel findings that 1) demonstrate dynamic redistribution of Vangl2 within discrete subcellular membrane domains over the course of gastrulation and neurulation, 2) suggest tissue- specific contexts for published tenets regarding membrane recycling of PCP proteins during mitosis or their association with the actin cytoskeleton, and 3) indicate an inherent and surprising directionality to the intercellular propagation of PCP. Outside of C. elegans,42 functional fluorescently- tagged PCP knock- in alleles have not been widely reported. Our vangl2sfGFP allele thus represents an invaluable resource that, when combined with embryonic cell- transplantation, live- imaging and conditional zGrad degradation methodologies, establishes a powerful experimental paradigm for future investigations into vertebrate cell polarity. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 191, 108]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[113, 160, 275, 179]]<|/det|> +## Animal husbandry + +<|ref|>text<|/ref|><|det|>[[111, 192, 860, 420]]<|/det|> +Zebrafish husbandry and experimental protocols were approved by the Hospital for Sick Children's Animal Care Committee, and all protocols were performed in accordance with Canadian Council on Animal Care guidelines. Wild- type zebrafish from TU strains were used. The \(v a n g l2\) mutant allele \(t r i^{k50f}\) was used for cell transplantation experiments \(^{21}\) , and the \(t r i^{m209}\) allele was used in to validate the \(v a n g l2^{y G F P}\) allele functionality because of the ability to use a published PCR mediated genotyping strategy \(^{21,43}\) . Tg(foxj1a::iCre) line has been previously described \(^{41}\) . Embryos from natural matings were grown at \(28^{\circ}\mathrm{C}\) unless otherwise indicated. + +<|ref|>sub_title<|/ref|><|det|>[[115, 473, 390, 492]]<|/det|> +## sfGFP-Vangl2 targeting strategy + +<|ref|>sub_title<|/ref|><|det|>[[115, 508, 248, 526]]<|/det|> +## Target Selection + +<|ref|>text<|/ref|><|det|>[[111, 540, 880, 771]]<|/det|> +A previously validated guide RNA (gRNA) was used to target the sequence surrounding the start codon of \(v a n g l2\) (targeting sequence GGATAACGAGTCGCAGTACTCGG) \(^{44}\) . As a proof of principle, a solution containing \(300~\mathrm{mM}\) KCl, \(100~\mathrm{ng}\) \(v a n g l2\) gRNA, \(200~\mathrm{ng}\) recombinant Cas9 (PNA Bio cat. CP01) and \(10\%\) phenol red was incubated at \(37^{\circ}\mathrm{C}\) for 5 minutes to form ribonucleoprotein complexes (RNPs), and injected into wild- type embryos at the 1- cell stage. Embryos were (i) scored for \(v a n g l2\) /tribolite phenotypes at 28 hpf and (ii) assessed for loss of a ScaI restriction site by PCR to gauge mutagenesis efficiency. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 123, 881, 562]]<|/det|> +Prior to vector construction, male and female Tu wild- type animals were screened by PCR for polymorphisms surrounding vang12 exon 2. Left and right arms of homology of \(\sim 1 - 1.2\) kb in length were amplified from fin clips and subjected for Sanger sequencing. Fish containing highly similar amplicons were selected for use in targeting experiments (4 males and 4 females). A \(\sim 2.2\) kb amplicon surrounding vang12 exon 2 was amplified from these wildtype fish and cloned into pDONR221. A superfolder GFP (sfGFP) was amplified from sfGFP- N1 (a kind gift from Michael Davidson and Geoffrey Waldo, Addgene #54737) \(^{45}\) and cloned into the vang12 targeting intermediate with a 3x(GGGGS) flexible linker by megaprimer PCR \(^{46}\) and verified for in- frame inclusion. Site- directed mutagenesis was performed on this intermediate to abolish the vang12 gRNA targeting site to prevent vector from being cleaved in vivo. This final targeting intermediate was amplified and cloned into XhoI and EcoRI sites of pKHR5 (a kind gift from D. Grunwald, Addgene #74593). All targeting intermediates and final clones were fully verified by Sanger sequencing. + +<|ref|>sub_title<|/ref|><|det|>[[115, 611, 273, 629]]<|/det|> +## Knock-In Injections + +<|ref|>text<|/ref|><|det|>[[111, 644, 884, 876]]<|/det|> +A \(2.5 \mu \mathrm{l}\) mix containing \(20 \mathrm{ng}\) of vang12 targeting vector, \(1.25 \mathrm{U}\) of I- SceI (NEB cat. R0694S), \(0.5 \mathrm{x}\) Cutsmart buffer and \(10\%\) phenol red was prepared. A second \(2.5 \mu \mathrm{l}\) mix containing \(300 \mathrm{mM}\) KCl, \(100 \mathrm{ng}\) vang12 gRNA and \(200 \mathrm{ng}\) recombinant Cas9 (PNA Bio cat. CP01) was incubated at \(37^{\circ} \mathrm{C}\) for 5 minutes. The two mixes were combined together and one nanolitre was injected into incrosses between pre- designated Tu wild- type animals. Mosaic GFP fluorescence was observed in several F0 embryos which were separated and grown to adulthood. Adult F0 animals were outcrossed to wildtype Tu fish and screened for GFP+ fluorescence. A founder transmitting \(36\%\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 872, 179]]<|/det|> +of GFP+ embryos was identified, and F1 embryos were screened by PCR using the following primers, F: CCGCGCTCTCCAGTCCGTCA, R: CGAGAGCTGCGTGAGTGTGAA, to ensure precise editing of the vangl2 locus. + +<|ref|>text<|/ref|><|det|>[[112, 228, 411, 247]]<|/det|> +Line Establishment and Maintenance + +<|ref|>text<|/ref|><|det|>[[112, 262, 833, 316]]<|/det|> +The F1 fish were incrossed to produce homozygous F2 fish. Embryos were screened either visually by GFP fluorescence or genotyped using the primers listed above. + +<|ref|>sub_title<|/ref|><|det|>[[113, 368, 227, 386]]<|/det|> +## Transgenesis + +<|ref|>text<|/ref|><|det|>[[110, 400, 884, 702]]<|/det|> +Transgenesis was performed using standard Gateway- compatible procedures47. The zGrad open reading frame was amplified from pCS2(+)- zGrad (a kind gift from H. Knaut; Addgene #119716) and cloned into pDONR- P2rP3 to generate p3E- zGrad. To generate Tg( \(\beta\) actin2::loxP mCherry STOP loxP zGrad) zebrafish, p5E- \(\beta\) actin247, pME- loxP mCherry STOP loxP (a gift from K. Kwan), and p3E- zGrad were recombined into pDEST Tol2 pA2 transgenesis vector. Embryos were injected at the one- cell stage with 25 pg of assembled transgene and 25 pg of Tol2 mRNA. Embryos were sorted at 48 hpf for mCherry reporter expression and were subsequently grown to adulthood. Individuals were bred to TU wild- type zebrafish to generate stable F1 lines. Clones were fully verified by Sanger sequencing. + +<|ref|>sub_title<|/ref|><|det|>[[113, 752, 434, 771]]<|/det|> +## Plasmids and embryo microinjections + +<|ref|>text<|/ref|><|det|>[[112, 785, 850, 875]]<|/det|> +Rab5c, 7 and 11a plasmids were a kind gift from B. Link, mCherry- LifeAct- 7 was a kind gift from M. Davidson (Addgene #54491), pCS2(+)- zGrad was a gift from H. Knaut (Addgene #119716)23, pME- mTagBFP- CAAX was a gift from N. Cole (Addgene # 75149)48, zebrafish + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 875, 109]]<|/det|> +codon- optimized mScarlet was a kind gift from T. Thiele, centrin- mCherry was a kind gift from + +<|ref|>text<|/ref|><|det|>[[110, 123, 844, 144]]<|/det|> +A. Salic and GalT-RFP was a kind gift from J. Wallingford. N-terminal mCherry-linker was + +<|ref|>text<|/ref|><|det|>[[110, 157, 840, 178]]<|/det|> +cloned into the Rab plasmids and mCherry-LifeAct and mTagBFP-CAAX were cloned into + +<|ref|>text<|/ref|><|det|>[[110, 192, 875, 213]]<|/det|> +pCS2+ using Gateway technology (Thermo Fisher Scientific). The plasmids were linearized and + +<|ref|>text<|/ref|><|det|>[[110, 227, 825, 248]]<|/det|> +sense-strand-capped mRNA was synthesized with the mMESSAGE mMACHINE system + +<|ref|>text<|/ref|><|det|>[[110, 262, 875, 283]]<|/det|> +(Ambion). 15 pg membrane-localized monomeric RFP (mRFP), 20 pg mTagBFP-CAAX, 30 pg + +<|ref|>text<|/ref|><|det|>[[110, 297, 875, 318]]<|/det|> +mCherry-LifeAct, 5- 25 pg mCherry- Centrin, 2.5 pg Arl13b- mScarlet, 10 pg mCherry- Rab5c, 10 + +<|ref|>text<|/ref|><|det|>[[110, 332, 875, 352]]<|/det|> +pg mCherry- Rab7, 15 pg mCherry- Rab11a, 10 pg GalT- RFP and 50- 100 pg zGrad were injected + +<|ref|>text<|/ref|><|det|>[[110, 368, 250, 386]]<|/det|> +at one- cell stage. + +<|ref|>sub_title<|/ref|><|det|>[[113, 438, 294, 457]]<|/det|> +## Cell transplantations + +<|ref|>text<|/ref|><|det|>[[110, 471, 858, 492]]<|/det|> +Cell transplantations were performed at mid- blastula stages, as described previously49. For the + +<|ref|>text<|/ref|><|det|>[[110, 507, 743, 527]]<|/det|> +early somite stage imaging, embryos were grown at \(25^{\circ}\mathrm{C}\) after transplantations. + +<|ref|>sub_title<|/ref|><|det|>[[113, 578, 189, 596]]<|/det|> +## Imaging + +<|ref|>text<|/ref|><|det|>[[110, 610, 848, 631]]<|/det|> +Brightfield embryonic imaging was performed on Leica M80 microscope with a MC170 HD + +<|ref|>text<|/ref|><|det|>[[110, 645, 857, 666]]<|/det|> +camera (Leica). Wide- field fluorescence imaging was performed on a Zeiss AXIO Zoom V16 + +<|ref|>text<|/ref|><|det|>[[110, 680, 842, 700]]<|/det|> +microscope (EMS3/SyCoP3, Zeiss) and ORCA-Flash 4.0 C1140- 22C camera (Hamamatsu). + +<|ref|>text<|/ref|><|det|>[[110, 714, 828, 735]]<|/det|> +Confocal imaging was performed on Leica TCS SP8 Lightning confocal with Leica DMI8 + +<|ref|>text<|/ref|><|det|>[[110, 749, 860, 770]]<|/det|> +microscope, 40x/1.3 water immersion objective and HyD detectors using the 405, 488 and 552 + +<|ref|>text<|/ref|><|det|>[[110, 784, 856, 805]]<|/det|> +laser lines (Leica). The images were deconvolved using the Lightning module. All time- lapse + +<|ref|>text<|/ref|><|det|>[[110, 819, 864, 840]]<|/det|> +imaging was performed using a resonant scanner with 16 times averaging. Maximum intensity + +<|ref|>text<|/ref|><|det|>[[110, 854, 691, 874]]<|/det|> +projections and time lapse images were recorded using Imaris (Bitplane). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 546, 109]]<|/det|> +## Measurements of embryonic lengths and L/W ratio + +<|ref|>text<|/ref|><|det|>[[112, 123, 881, 250]]<|/det|> +Embryo lengths were measured by manually drawing a line from the front of the head to the tip of the tail in ImageJ. Embryonic length/width ratios were calculated by measuring the widest and longest points of krox20 and myoD expression. Statistical testing was performed using GraphPad Prism (GraphPad Software). + +<|ref|>sub_title<|/ref|><|det|>[[113, 299, 626, 319]]<|/det|> +## Quantification of asymmetric Vang12 membrane-localization + +<|ref|>text<|/ref|><|det|>[[112, 331, 877, 597]]<|/det|> +Comparison of Vang12 and mRFP membrane enrichment was performed using the spot function in Imaris. In short, membrane- signals for mRFP or sfGFP were decorated with spots, and the spots for the two channels were merged. A reference point was placed in the centre of each cell to determine the anterior- posterior (X) and apical- basal (Y) axis. The position of each spot was determined as their distance to the reference point. Mean fluorescence intensity for mRFP and sfGFP within each spot was calculated, and the \(10\%\) of brightest spots for each fluorophore were plotted against their normalized X and Y positions using GraphPad Prism (GraphPad Software). Statistical testing was performed using GraphPad Prism (GraphPad Software). + +<|ref|>sub_title<|/ref|><|det|>[[115, 648, 364, 667]]<|/det|> +## BB positioning quantification + +<|ref|>text<|/ref|><|det|>[[112, 680, 870, 842]]<|/det|> +Basal body position was defined as a ratio between the distance of the basal body from the posterior membrane and the anterior- posterior floor plate cell length normalized to 1. Measurements were done in Imaris and statistical testing was performed using GraphPad Prism (GraphPad Software). The values were divided into three categories: posterior (0.66- 1), medial (0.33- 0.66) and anterior (0- 0.33). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 89, 410, 108]]<|/det|> +## Whole-mount in situ hybridization + +<|ref|>text<|/ref|><|det|>[[112, 123, 842, 144]]<|/det|> +Antisense RNA probes were prepared using DIG RNA labelling kit (Roche) from linearized + +<|ref|>text<|/ref|><|det|>[[112, 158, 833, 179]]<|/det|> +DNA. Whole mount RNA in situ hybridizations were performed according to Rose et al. \(^{50}\) . + +<|ref|>text<|/ref|><|det|>[[112, 194, 428, 214]]<|/det|> +Samples were genotyped after imaging. + +<|ref|>sub_title<|/ref|><|det|>[[113, 274, 209, 292]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[112, 343, 852, 400]]<|/det|> +1. Devenport, D. Tissue morphodynamics: Translating planar polarity cues into polarized cell behaviors. Semin. Cell Dev. Biol. doi:10.1016/j.semcdb.2016.03.012. (2016) + +<|ref|>text<|/ref|><|det|>[[112, 422, 833, 478]]<|/det|> +2. Davey, C. F. & Moens, C. B. Planar cell polarity in moving cells: think globally, act locally. Development 144, 187–200 (2017). + +<|ref|>text<|/ref|><|det|>[[112, 501, 828, 558]]<|/det|> +3. Humphries, A. C. & Mlodzik, M. From instruction to output: Wnt/PCP signaling in development and cancer. Curr. Opin. Cell Biol. 51, 110–116 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 581, 833, 670]]<|/det|> +4. Jussila, M. & Ciruna, B. Zebrafish models of non-canonical Wnt/planar cell polarity signalling: fishing for valuable insight into vertebrate polarized cell behavior. Wiley Interdiscip. Rev. Dev. Biol. 6. (2017). + +<|ref|>text<|/ref|><|det|>[[112, 694, 857, 750]]<|/det|> +5. Butler, M. T. & Wallingford, J. B. Planar cell polarity in development and disease. Nat. Rev. Mol. Cell Biol. 18, 375–388 (2017). + +<|ref|>text<|/ref|><|det|>[[112, 774, 881, 830]]<|/det|> +6. Carvajal-Gonzalez, J. M. & Mlodzik, M. Mechanisms of planar cell polarity establishment in Drosophila. F1000Prime Rep. 6, 98 (2014). + +<|ref|>text<|/ref|><|det|>[[112, 854, 878, 875]]<|/det|> +7. Devenport, D. & Fuchs, E. Planar polarization in embryonic epidermis orchestrates global + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 88, 870, 911]]<|/det|> +asymmetric morphogenesis of hair follicles. Nat. Cell Biol. 10, 1257–1268 (2008).8. Montcouquiol, M. et al. Asymmetric localization of Vangl2 and Fz3 indicate novel mechanisms for planar cell polarity in mammals. J. Neurosci. 26, 5265–5275 (2006).9. Butler, M. T. & Wallingford, J. B. Spatial and temporal analysis of PCP protein dynamics during neural tube closure. eLife 7, e36456 (2018).10. Heck, B. W. & Devenport, D. Trans-endocytosis of planar cell polarity complexes during cell division. Curr. Biol. 27, 3725-3733. e4 (2017).11. Love, A. M., Prince, D. J. & Jessen, J. R. Vangl2-dependent regulation of membrane protrusions and directed migration requires a fibronectin extracellular matrix. Development. 145, dev165472 (2018).12. Davey, C. F., Mathewson, A. W. & Moens, C. B. PCP Signaling between migrating neurons and their planar-polarized neuroepithelial environment controls filopodial dynamics and directional migration. PLOS Genet. 12, e1005934 (2016).13. Borovina, A., Superina, S., Voskas, D. & Ciruna, B. Vangl2 directs the posterior tilting and asymmetric localization of motile primary cilia. Nat. Cell Biol. 12, 407–412 (2010).14. Roszko, I., S. Sepich, D., Jessen, J. R., Chandrasekhar, A. & Solnica-Krezel, L. A dynamic intracellular distribution of Vangl2 accompanies cell polarization during zebrafish gastrulation. Development 142, 2508–2520 (2015).15. Devenport, D., Oristian, D., Heller, E. & Fuchs, E. Mitotic internalization of planar cell polarity proteins preserves tissue polarity. Nat. Cell Biol. 13, 893–902 (2011).16. Ciruna, B., Jenny, A., Lee, D., Mlodzik, M. & Schier, A. F. Planar cell polarity signalling + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[162, 88, 876, 110]]<|/det|> +couples cell division and morphogenesis during neurulation. Nature 439, 220–224 (2006). + +<|ref|>text<|/ref|><|det|>[[162, 135, 872, 190]]<|/det|> +Panousopoulou, E., Tyson, R. A., Bretschneider, T. & Green, J. B. A. The distribution of Dishevelled in convergently extending mesoderm. Dev. Biol. 382, 496–503 (2013). + +<|ref|>text<|/ref|><|det|>[[162, 214, 828, 268]]<|/det|> +Chu, C.-W. & Sokol, S. Y. Wnt proteins can direct planar cell polarity in vertebrate ectoderm. eLife 5, e16463 (2016). + +<|ref|>text<|/ref|><|det|>[[162, 293, 835, 383]]<|/det|> +Witzel, S., Zimyanin, V., Carreira-Barbosa, F., Tada, M. & Heisenberg, C. P. Wnt11 controls cell contact persistence by local accumulation of Frizzled 7 at the plasma membrane. J. Cell Biol. 175, 791–802 (2006). + +<|ref|>text<|/ref|><|det|>[[162, 408, 861, 463]]<|/det|> +Butler, M. T. & Wallingford, J. B. Control of vertebrate core planar cell polarity protein localization and dynamics by Prickle 2. Development 142, 3429–39 (2015). + +<|ref|>text<|/ref|><|det|>[[162, 488, 853, 543]]<|/det|> +Jessen, J. R. et al. Zebrafish trilobite identifies new roles for Strabismus in gastrulation and neuronal movements. Nat. Cell Biol. 4, 610–615 (2002). + +<|ref|>text<|/ref|><|det|>[[162, 568, 861, 622]]<|/det|> +Park, M. & Moon, R. T. The planar cell-polarity gene stbm regulates cell behaviour and cell fate in vertebrate embryos. Nat. Cell Biol. 4, 20–25 (2002). + +<|ref|>text<|/ref|><|det|>[[162, 647, 866, 702]]<|/det|> +Yamaguchi, N., Colak-Champollion, T. & Knaut, H. ZGrad is a nanobody-based degron system that inactivates proteins in zebrafish. eLife 8, e43125 (2019). + +<|ref|>text<|/ref|><|det|>[[162, 727, 877, 817]]<|/det|> +Petridou, N. I., Grigolon, S., Salbreux, G., Hannezo, E. & Heisenberg, C. P. Fluidization-mediated tissue spreading by mitotic cell rounding and non-canonical Wnt signalling. Nat. Cell Biol. 21, 169–178 (2019). + +<|ref|>text<|/ref|><|det|>[[162, 843, 840, 897]]<|/det|> +Mottola, G., Classen, A.-K., González-Gaitán, M., Eaton, S. & Zerial, M. A novel function for the Rab5 effector Rabenosyn-5 in planar cell polarity. Development 137, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 90, 875, 911]]<|/det|> +26. Mathewson, A. W., Berman, D. G. & Moens, C. B. Microtubules are required for the maintenance of planar cell polarity in monociliated floorplate cells. Dev. Biol. 452, 21–33 (2019).27. Strutt, H. & Strutt, D. Differential stability of flamingo protein complexes underlies the establishment of planar polarity. Curr. Biol. 18, 1555–1564 (2008).28. Strutt, H., Warrington, S. J. & Strutt, D. Dynamics of core planar polarity protein turnover and stable assembly into discrete membrane subdomains. Dev. Cell 20, 511–525 (2011).29. Williams, B. B. et al. VANGL2 regulates membrane trafficking of MMP14 to control cell polarity and migration. J. Cell Sci. 125, 2141–2147 (2012).30. Sepich, D. S. & Solnica-Krezel, L. Intracellular Golgi Complex organization reveals tissue specific polarity during zebrafish embryogenesis. Dev. Dyn. 245, 678–691 (2016).31. Ossipova, O., Kim, K. & Sokol, S. Y. Planar polarization of Vangl2 in the vertebrate neural plate is controlled by Wnt and Myosin II signaling. Biol. Open 4, 722–730 (2015).32. Strutt, H., Gamage, J. & Strutt, D. Robust asymmetric localization of planar polarity proteins is associated with organization into signalosome-like domains of variable stoichiometry. Cell Rep. 17, 2660–2671 (2016).33. Stahley, S. N., Basta, L. P., Sharan, R. & Devenport, D. Celsr1 adhesive interactions mediate the asymmetric organization of planar polarity complexes. eLife 10, e62097 (2021).34. Shi, D. et al. Dynamics of planar cell polarity protein Vangl2 in the mouse oviduct + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 87, 513, 108]]<|/det|> +epithelium. Mech. Dev. 141, 78- 89 (2016). + +<|ref|>text<|/ref|><|det|>[[57, 131, 864, 225]]<|/det|> +Shindo, A., Inoue, Y., Kinoshita, M. & Wallingford, J. B. PCP- dependent transcellular regulation of actomyosin oscillation facilitates convergent extension of vertebrate tissue. Dev. Biol. 446, 159- 167 (2019). + +<|ref|>text<|/ref|><|det|>[[57, 247, 857, 304]]<|/det|> +Riedl, J. et al. Lifeact: a versatile marker to visualize F- actin. Nat. Methods 5, 605- 607 (2008). + +<|ref|>text<|/ref|><|det|>[[57, 327, 880, 384]]<|/det|> +Brunt, L. et al. Vangl2 promotes the formation of long cytonemes to enable distant Wnt/β- catenin signaling. Nat. Commun. 12, 2058 (2021). + +<|ref|>text<|/ref|><|det|>[[57, 405, 877, 500]]<|/det|> +Oozeer, F., Yates, L. L., Dean, C. & Formstone, C. J. A role for core planar polarity proteins in cell contact- mediated orientation of planar cell division across the mammalian embryonic skin. Sci. Rep. 7, 1880 (2017). + +<|ref|>text<|/ref|><|det|>[[57, 521, 840, 615]]<|/det|> +Carvajal- Gonzalez, J. M., Roman, A.- C. & Mlodzik, M. Positioning of centrioles is a conserved readout of Frizzled planar cell polarity signalling. Nat. Commun. 7, 11135 (2016). + +<|ref|>text<|/ref|><|det|>[[57, 636, 880, 729]]<|/det|> +Donati, A., Anselme, I., Schneider- Maunoury, S. & Vesque, C. Planar polarization of cilia in the zebrafish floor- plate involves Par3- mediated posterior localization of highly motile basal bodies. Development 148, dev196386 (2021). + +<|ref|>text<|/ref|><|det|>[[57, 750, 841, 844]]<|/det|> +Van Gennip, J. L. M., Boswell, C. W. & Ciruna, B. Neuroinflammatory signals drive spinal curve formation in zebrafish models of idiopathic scoliosis. Sci. Adv. 4, 1- 12 (2018). + +<|ref|>text<|/ref|><|det|>[[57, 866, 866, 888]]<|/det|> +Heppert, J. K., Pani, A. M., Roberts, A. M., Dickinson, D. J. & Goldstein, B. A CRISPR + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 88, 864, 144]]<|/det|> +tagging- based screen reveals localized players in Wnt- directed asymmetric cell division. Genetics 208, 1147–1164 (2018). + +<|ref|>text<|/ref|><|det|>[[163, 168, 857, 224]]<|/det|> +Gao, B. et al. Wnt signaling gradients establish planar cell polarity by inducing Vangl2 phosphorylation through Ror2. Dev. Cell 20, 163–176 (2011). + +<|ref|>text<|/ref|><|det|>[[163, 247, 875, 303]]<|/det|> +Shah, A. N., Davey, C. F., Whitebirch, A. C., Miller, A. C. & Moens, C. B. Rapid reverse genetic screening using CRISPR in zebrafish. Zebrafish 13, 152–153 (2016). + +<|ref|>text<|/ref|><|det|>[[163, 327, 869, 383]]<|/det|> +Pédelacq, J.- D. et al. Engineering and characterization of a superfolder green fluorescent protein. Nat. Biotechnol. 24, 79–88 (2006). + +<|ref|>text<|/ref|><|det|>[[163, 407, 850, 500]]<|/det|> +Geiser, M., Cèbe, R., Drewello, D. & Schmitz, R. Integration of PCR fragments at any specific site within cloning vectors without the use of restriction enzymes and DNA ligase. BioTechniques 31, 88–92 (2001). + +<|ref|>text<|/ref|><|det|>[[163, 523, 836, 579]]<|/det|> +Kwan, K. M. et al. The Tol2kit: A multisite gateway- based construction Kit for Tol2 transposon transgenesis constructs. Dev. Dyn. 236, 3088–3099 (2007). + +<|ref|>text<|/ref|><|det|>[[163, 602, 825, 658]]<|/det|> +Don, E. K. et al. A Tol2 Gateway- Compatible Toolbox for the study of the nervous system and neurodegenerative disease. Zebrafish 14, 69–72 (2017). + +<|ref|>text<|/ref|><|det|>[[163, 681, 863, 737]]<|/det|> +Ho, R. K. & Kane, D. A. Cell- autonomous action of zebrafish spt- 1 mutation in specific mesodermal precursors. Nature 348, 728–730 (1990). + +<|ref|>text<|/ref|><|det|>[[163, 761, 870, 852]]<|/det|> +Rose, C. D. et al. SCO- Spondin defects and neuroinflammation are conserved mechanisms driving spinal deformity across genetic models of idiopathic scoliosis. Curr. Biol. 30, 2363- 2373. e6 (2020). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 281, 108]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[111, 123, 870, 280]]<|/det|> +We gratefully acknowledge the SickKids' Imaging Facility for assistance with confocal microscopy and the SickKids' Zebrafish Facility technicians for excellent zebrafish care. This work was supported, in part, by funding from Canadian Institutes of Health Research (FDN- 167285) and the Canada Research Chair program to B.C., and the Sigrid Juselius Foundation to M.J. + +<|ref|>sub_title<|/ref|><|det|>[[115, 334, 301, 352]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[111, 367, 881, 456]]<|/det|> +M.J. and C.W.B. performed Vangl2 targeting, M.J. performed all Vangl2 knock- in line validation and Vangl2 imaging experiments and analysis, C.W.B. generated the zGrad transgene, B.C. supervised research studies, M.J. and B.C. wrote the manuscript with input from C.W.B. + +<|ref|>sub_title<|/ref|><|det|>[[115, 508, 315, 526]]<|/det|> +## Declaration of Interests + +<|ref|>text<|/ref|><|det|>[[115, 542, 460, 561]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 611, 356, 630]]<|/det|> +## Supplementary Movie Titles + +<|ref|>text<|/ref|><|det|>[[112, 645, 875, 857]]<|/det|> +Supplementary Movie 1. Vangl2 enrichment on anterior membranes is dynamic over time Supplementary Movie 2. Vangl2 enrichment on the anterior membrane does not colocalize with filamentous actin Supplementary Movie 3. Vangl2 localizes to anterior apical membranes and to basal membrane extensions in dividing cells Supplementary Movie 4. Vangl2 anterior polarization is lost during cell division + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 60, 833, 687]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[140, 70, 188, 81]]<|/det|> +
a
+ +<|ref|>text<|/ref|><|det|>[[113, 708, 802, 725]]<|/det|> +Figure 1. sfGFP- Vang12 is functional and membrane localized during early zebrafish development + +<|ref|>text<|/ref|><|det|>[[115, 724, 620, 740]]<|/det|> +(a) A schematic illustration of the sfGFP-Vang12 knock-in targeting strategy. + +<|ref|>text<|/ref|><|det|>[[113, 737, 860, 770]]<|/det|> +(b) Wide-field fluorescence images demonstrating sfGFP-Vang12 expression in \(vang12^{sfGFP + }\) and \(vang12^{sfGFP / sfGFP}\) embryos at 28 hours post fertilization (hpf). + +<|ref|>text<|/ref|><|det|>[[115, 768, 780, 787]]<|/det|> +(c) Lateral view of wild-type, \(vang12^{sfGFP / sfGFP}\) , \(vang12^{sfGFP / m209}\) and \(vang12^{m209 / m209}\) embryos at 24 hpf. + +<|ref|>text<|/ref|><|det|>[[115, 785, 872, 849]]<|/det|> +(d) Embryo length quantification at 24 hpf. Data shows representative experiments combined from independent crosses that were repeated 2-4 times. Statistical analysis was performed using one-way ANOVA with Tukey's multiple comparisons test (****P<0.0001). Wild-type \(n = 14\) , \(vang12^{sfGFP + }\) \(n = 22\) , \(vang12^{sfGFP / sfGFP}\) \(n = 20\) , \(vang12^{m209 / + }\) \(n = 82\) , \(vang12^{sfGFP / m209}\) \(n = 72\) , \(vang12^{m209}\) \(n = 63\) . + +<|ref|>text<|/ref|><|det|>[[115, 847, 872, 912]]<|/det|> +(e) Live confocal images of \(vang12^{sfGFP / sfGFP}\) (hereafter referred as \(vang12^{sfGFP}\) ) embryos at blastula through early gastrula stages, as indicated. Ectodermal cells were imaged in gastrulating embryos. Embryos were injected with mRNA coding for a membrane-localized monomeric RFP (mRFP) reporter. All sfGFP images were acquired using identical settings. + +<|ref|>text<|/ref|><|det|>[[115, 910, 877, 960]]<|/det|> +(f-i) Live confocal images of shield staged \(vang12^{sfGFP}\) embryos injected with mRNA coding for mCherry-Rab5c (f), mCherry-Rab7 (g), mCherry-Rab11a (h) or GalT-RFP (i) reporter constructs. Arrows point at cytoplasmic Vang12 puncta in close proximity to respective reporters, and arrowheads point at isolated Vang12 puncta. + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[321, 40, 515, 55]]<|/det|> +
Figure 2. Jussila et al.
+ +<|ref|>image<|/ref|><|det|>[[321, 66, 660, 870]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 120, 872, 152]]<|/det|> +Figure 2. Planar polarized enrichment of Vang12 on anterior neuroepithelial membranes is controlled by cell non- autonomous PCP signalling + +<|ref|>text<|/ref|><|det|>[[113, 152, 875, 215]]<|/det|> +(a) Confocal images comparing sfGFP-Vang12 and mRFP localization at 22 hours post fertilization (hpf) in dorsal (top) and lateral (bottom) orientations. Anterior is to the left in all images. Strong Vang12 expression is observed on cell membranes in the neural tube (NT) including the floor plate (FP), while only weak expression is observed in somites (S) and notochord (NC). + +<|ref|>text<|/ref|><|det|>[[113, 215, 875, 310]]<|/det|> +(b) A schematic illustrating transplantation of mRFP-labelled vang12sfGFP donor cells into WT (wild-type) host embryos, at sphere stage. Chimeric embryos were imaged dorsally, at positions between the \(2^{\mathrm{nd}}\) and \(8^{\mathrm{th}}\) somite. +(c) Confocal images of mRFP and sfGFP-Vang12 localization in the developing spinal cord of chimeric embryos at 4-somite, 10-somite and 13-somite stages of development. Maximum intensity projections are shown. Arrows point to Vang12 enrichment on anterior neuroepithelial cell membranes, and arrowheads to anterior apical membranes at the neural midline. + +<|ref|>text<|/ref|><|det|>[[113, 309, 872, 390]]<|/det|> +(d-e) Distribution of the brightest sfGFP-Vang12 and membrane-RFP spots along anterior-posterior (d) and apical-basal (e) axes of vang12sfGFP neuroepithelial cells, quantified at four consecutive stages of neural tube morphogenesis. Data from multiple transplanted vang12sfGFP cells was pooled (3-5 somites, \(\mathrm{n} = 7\) ; 7 somites, \(\mathrm{n} = 7\) ; 8-10 somites, \(\mathrm{n} = 6\) ; 13 somites, \(\mathrm{n} = 5\) ) and axial lengths normalized to 1. Details of the quantification can be found in the Methods section. Statistical analysis was performed using a two-tailed Mann-Whitney test. + +<|ref|>text<|/ref|><|det|>[[113, 388, 875, 420]]<|/det|> +(f) A schematic illustrating transplantation of mRFP-labelled vang12sfGFP donor cells into vang12k50f loss-of-function host embryos, at sphere stage. + +<|ref|>text<|/ref|><|det|>[[113, 419, 872, 481]]<|/det|> +(g-h) Distribution of the brightest sfGFP-Vang12 and membrane-RFP spots along anterior-posterior (g) and apical-basal (h) axes of vang12sfGFP neuroepithelial cells transplanted into a vang12k50f mutant host, quantified at two consecutive stages of neural tube morphogenesis. Data from multiple transplanted vang12sfGFP cells was pooled (10 somites, \(\mathrm{n} = 7\) ; 12-13 somites, \(\mathrm{n} = 9\) ). + +<|ref|>text<|/ref|><|det|>[[113, 480, 877, 528]]<|/det|> +(i) Confocal images of mRFP and sfGFP-Vang12 localization in vang12sfGFP neuroepithelial cells within vang12k50f mutant hosts, at 10-somite and 13-somite stages of development. Maximum intensity projections are shown. Arrows point at Vang12 enrichment on cell protrusions. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 60, 512, 520]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 35, 306, 52]]<|/det|> +
Figure 3. Jussila et al.
+ +<|ref|>text<|/ref|><|det|>[[113, 589, 861, 732]]<|/det|> +(a) Confocal images of sfGFP-Vang12 localization in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters within floorplate cells of a vang12sfGFP embryo at 22hpf. Lateral view, anterior to the left. +(b) Confocal images of sfGFP-Vang12 localization in relation to cilia axoneme (Arl13b-mScarlet) and membrane (mTagBFP) reporters within floorplate cells of a vang12sfGFP embryo at 32hpf. Lateral view, anterior to the left. +(c) Confocal image comparing sfGFP-Vang12 and mRFP localization in floorplate cells of a 23-somite staged chimeric embryo (mRFP mRNA-injected vang12sfGFP cells transplanted into WT hosts). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 enrichment on anterior cell membranes. Arrowhead points at Vang12 enriched membrane protruding from the anterior apical surface. + +<|ref|>text<|/ref|><|det|>[[113, 732, 881, 810]]<|/det|> +(d) Confocal image comparing sfGFP-Vang12 and Centrin-mCherry localization in floorplate cells of a 15-somite staged chimeric embryo (Centrin-mCherry mRNA-injected vang12sfGFP cells transplanted into WT hosts). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 on the anterior membrane that is closely associated with a basal body docked on the posterior membrane. Asterisk indicates anterior unlabelled host cell that is not Centrin-mCherry positive. + +<|ref|>text<|/ref|><|det|>[[113, 810, 882, 904]]<|/det|> +(e) Confocal image comparing sfGFP-Vang12 localization in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters in neighbouring floorplate cells of a 15-somite staged chimeric embryo (vang12sfGFP cells transplanted into WT hosts injected with Centrin-mCherry and mTagBFP mRNA). Maximum intensity projections, dorsal (top) and lateral (bottom) views are shown, anterior to the left. Arrows point at Vang12 on the anterior membrane of a donor cell that is closely associated with a basal body docked on the posterior membrane of a host cell. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 50, 500, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 28, 308, 45]]<|/det|> +
Figure 4. Jussila et al.
+ +<|ref|>text<|/ref|><|det|>[[115, 700, 883, 732]]<|/det|> +Figure 4. Conditional degradation of sfGFP-Vang12 reveals a cell non-autonomous requirement for Vang12 in maintaining basal body polarity + +<|ref|>text<|/ref|><|det|>[[115, 732, 883, 795]]<|/det|> +(a) Quantification of embryo lengths for zGrad mRNA-injected and control embryos at 24 hpf. Maternal and zygotic genotypes are as indicated. Data is pooled from three independent experiments. Control n from left to right: n=53, n=62, n=61, n=36. zGrad n from left to right: n=55, n=51, n=53, n=54. Statistical analysis was performed using Kruskal-Wallis test with Dunn's multiple comparisons test. + +<|ref|>text<|/ref|><|det|>[[115, 795, 880, 874]]<|/det|> +(b) Confocal images of sfGFP-Vang12 expression in relation to basal body (Centrin-mCherry) and membrane (mTagBFP) reporters within floorplate cells of control vang12sfGFP; Tg(loxP-mCherry-STOP-loxP-zGrad) embryos (left) and experimental vang12sfGFP; Tg(loxP-mCherry-STOP-loxP- zGrad); Tg(foxj1a-iCre) embryos (right) at 48 hpf. Dorsal view, anterior to the left. Arrows indicate residual sfGFP-Vang12 expression in a subset of floorplate cells following zGrad expression. Asterisks indicate mispolarized basal bodies following zGrad expression. + +<|ref|>text<|/ref|><|det|>[[115, 874, 880, 937]]<|/det|> +(c) Quantification of basal body localization within the posterior, medial or anterior thirds of floor plate cells for control and vang12sfGFP; Tg(loxP-mCherry-STOP-loxP-zGrad); Tg(foxj1a-cre) embryos described in (b). Cells were categorized as A-D based on the presence or absence of sfGFP-Vang12 signal within both the cell and its immediate neighbours, as illustrated in (d). Control n=164, A n=256, B n=20, C n=29, D n=19. + +<|ref|>text<|/ref|><|det|>[[112, 936, 880, 969]]<|/det|> +(d) A schematic illustrating cell categories quantified in (c). Grey colour indicates cells that have lost sfGFP-Vang12 signal; green colour indicates residual sfGFP-Vang12 expression. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 500, 310]]<|/det|> +- JussilaetalMovie3NatComm.mp4- JussilaetalMovie2NatComm.mp4- JussilaetalSupplementaryFiguresNatComm.pdf- JussilaetalMovie1NatComm.mp4- JussilaetalMovie4NatComm.mp4- JussilaetalSupplementaryFiguresNatComm.pdf- nrreportingsummaryNCOMMS2129897.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/images_list.json b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..0637a088a01e8ddab3bf3fa98dbe804cbde1a0dc --- /dev/null +++ b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/images_list.json @@ -0,0 +1 @@ +[] \ No newline at end of file diff --git a/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46.mmd b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46.mmd new file mode 100644 index 0000000000000000000000000000000000000000..082eb55f13cc897bac96f8aee1444be4d9916aed --- /dev/null +++ b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46.mmd @@ -0,0 +1,620 @@ + +# L3MBTL3 and active STAT3 collaboratively up-regulate SNAIL transcription to promote metastasis in breast cancer + +Jianpeng Xiao The Third School of Clinical Medicine, Southern Medical University + +Jie Wang The Third School of Clinical Medicine, Southern Medical University + +Jialun Li East China Normal University + +Jie Xiao School of Laboratory Medicine and Biotechnology, Southern Medical University + +Cuicui Liu Southern Medical University Affiliated Fengxian Hospital + +Libi Tan Southern Medical University Affiliated Fengxian Hospital + +Yanhong Tu The Second Affiliated Hospital, The Chinese University of Hong Kong + +Ruifang Yang Anhui University of Science and Technology Affiliated Fengxian Hospital + +Yujie Pei Anhui University of Science and Technology Affiliated Fengxian Hospital + +Minghua Wang The Second Affiliated Hospital, The Chinese University of Hong Kong + +Jiemin Wong East China Normal University https://orcid.org/0000- 0002- 5311- 842X + +Binhua Zhou University of Kentucky College of Medicine https://orcid.org/0000- 0002- 0983- 2060 + +Jing Li Southern Medical University Affiliated Fengxian Hospital + +Jing Feng + +feng.jing71921@163. com + +The Third School of Clinical Medicine, Southern Medical University + +<--- Page Split ---> + +## Article + +## Keywords: + +Posted Date: March 15th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2601300/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on January 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55617- 9. + +<--- Page Split ---> + +L3MBTL3 and active STAT3 collaboratively up-regulate SNAIL transcription to promote metastasis in breast cancer + +## Authors + +Jianpeng Xiao \(^{1,2,3\#}\) , Jie Wang \(^{1,2\#}\) , Jialun Li \(^{4}\) , Jie Xiao \(^{5}\) , CuiCui Liu \(^{2}\) , Libi Tan \(^{2,5}\) , Yanhong Tu \(^{3}\) , Ruifang Yang \(^{6}\) , Yujie Pei \(^{6}\) , Minghua Wang \(^{3}\) , Jiemin Wong \(^{4}\) , Binhua P Zhou \(^{7}\) , Jing Li \(^{2,5*}\) , Jing Feng \(^{1,2,3,5*}\) + +\(^{1}\) The Third School of Clinical Medicine, Southern Medical University, Guangzhou, 510630, China. + +\(^{2}\) Department of Laboratory Medicine & Central Laboratory, Southern Medical University Affiliated Fengxian Hospital, Shanghai, 201499, China. + +\(^{3}\) The Second Affiliated Hospital, The Chinese University of Hong Kong, Shenzhen, 518172, China. + +\(^{4}\) Shanghai Key Laboratory of Regulatory Biology, Institute of Biomedical Sciences and School of Life Sciences, East China Normal University, Shanghai, 200241 China. + +\(^{5}\) School of Laboratory Medicine and Biotechnology, Southern Medical University, Guangzhou, 510515, China. + +\(^{6}\) Anhui University of Science and Technology Affiliated Fengxian Hospital, Shanghai, 201499, China. + +\(^{7}\) Department of Molecular and Cellular Biochemistry, Markey Cancer Center, University of Kentucky College of Medicine, Lexington, 40536, USA. + +#These authors contributed equally: Jianpeng Xiao, Jie Wang + +These authors jointly supervised this work: Jing Feng, Jing Li. Email: fengjing71921@163. com; lijing228nanfang@163. com. + +<--- Page Split ---> + +## Abstract + +Active STAT3 pathway promotes epithelial mesenchymal transition, migration, invasion and metastasis in cancer. STAT3 upregulates the transcription of epithelial mesenchymal transition key transcription factor SNAIL through DNA- binding independent way. However, the mechanism that how is STAT3 recruited to SNAIL promoter to upregulate transcription in DNA independent way is still unclear. In our study, we found a lysine methylated binding protein L3MBTL3 was positively associated with metastatic and poor prognosis in breast cancer. L3MBTL3 also promoted epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. Mechanism analysis found L3MBTL3 interacted with active STAT3 and recruited active STAT3 on SNAIL promoter to up- regulated SNAIL transcription. The interaction between L3MBTL3 and active STAT3 was required for SNAIL transcription up- regulation, migration and invasion in breast cancer, while the methylated lysine binding activity of L3MBTL3 is not required for these functions. In conclusion, L3MBTL3 and active STAT3 collaboratively up- regulate SNAIL transcription to promote breast cancer metastasis. + +<--- Page Split ---> + +## Introduction + +Breast cancer, developing breast lobules, tubes or connective tissue it the most common malignancy in women'. Metastasis is the main cause of breast cancer high mortality'. It is a convoluted and multi- step process: the tumor cell disseminates from primary tumor sites, migrating to the target organs, then develops into a distal secondary tumor'. Epithelial- mesenchymal transition, a biological process switches immobile epithelial cells into mobile mesenchymal cells, is one of the crucial steps in initiating metastasis'. + +There are many key transcription factors involved in the epithelial- mesenchymal transition process. SNAIL (Snail family transcriptional repressor 1), as a direct transcription repressor of epithelial morphology promoter \(E\) - cad (cadherin1), drives epithelial- mesenchymal transition process and induces migratory and invasive behavior in cancer cells and non- cancer cells3- 5. Enhancing SNAIL gene expression in primary tumors promotes cellular motility and the consequent acquisition of metastatic properties. In contrast, targeted SNAIL or down- regulated the transcription of SNAIL gene can reverse epithelial- mesenchymal transition process6- 8. The transcription of SNAIL is regulated by many signaling pathway, including \(\mathrm{FGF}^9\) , \(\mathrm{Wnt^{10}}\) , \(\mathrm{TGF}\beta^{11}\) and STAT312 signaling pathway, which collectively form the microenvironment of epithelial- mesenchymal transition and metastasis13. + +STAT3 (Signal transducer and activator of transcription 3), as a member of STAT family, is involved in various basic cell functions14,15. Aberrant accumulation of phosphorylated STAT3 is observed in nearly \(70\%\) of cancers including breast cancer, and phosphorylated STAT3 is involved in metastasis leading the poor clinical outcome in patients16. It is reported that STAT3 plays crucial roles in epithelial- mesenchymal transition, migration, invasion and metastasis. In keratinocytes and ovarian adenocarcinoma, STAT3 is required for cell motility, silencing STAT3 can + +<--- Page Split ---> + +reverse the phenotype of cell motility17,18. Persistent activation STAT3 enhances cell migration of prostate epithelial cells19. High phosphorylated STAT3 level correlates with increased invasion and metastasis in cutaneous squamous cell carcinoma20. In contrast, targeting STAT3 blocks invasion and metastasis formation in variety of cell lines both in vitro and in vivo21,22. IL6 (Cytokine interleukin 6) activates STAT3 signal pathway to induce epithelial-mesenchymal transition and promote metastasis23- 27. In IL6- STAT3 signal pathway, IL6 triggers STAT3 phosphorylation at its Y705 site. The phosphorylated STAT3 forms dimer and localizes into the nucleus, then binds to the promoter to up- regulate the transcription of target genes28,29. STAT3 activation mediated by IL- 6 can upregulate epithelial- mesenchymal transition- inducing transcription factors such as SNAIL, ZEB1 (zinc finger E- box binding homebox 1), TWIST1 (twist family bHLH transcription factor 1) and JUNB (JunB proto- oncogene), and then induces epithelial- mesenchymal transition to enhance metastasis23- 26. STAT3 also can further enhanced SNAIL transcription under TGF- \(\beta\) induce condition12. Knocking down STAT3 or inhibiting STAT3 activity significantly reduces the induction of SNAIL by TGF- \(\beta^{12}\) . STAT3 often recognizes DNA motifs of target gene promoter by its DNA binding domain and up- regulates target gene transcription28,29. However, on SNAIL promoter, neither the DNA- binding activity of STAT3 nor canonical STAT3 DNA binding motif of promoter is required for SNAIL transcriptional upregulation12, suggesting that there is still DNA- binding independent transcriptional regulation mechanism. + +L3MBTL3 (L3MBTL histone methyl- lysine binding protein 3) is a member of the malignant brain tumor (MBT) family of chromatin- interacting transcriptional repressors, which binds histone lysine methylation H4K20me230. As a methyl- binding protein, L3MBTL3 can also bind the lysine methylation of non- histone protein through + +<--- Page Split ---> + +its MBT domain \(^{31,32}\) and recruit ubiquitin E3 ligase CRL4 \(^{DCAF5}\) to mediate the methylation- dependent degradation of substrate proteins (such as SOX2 (SRY- box transcription factor 2), E2F1 (E2F transcription factor 1), DNMT1 (DNA methyltransferase1), SMARCC1 and SMARCC2 (SWI/SNF related matrix associated actin dependent regulator of chromatin subfamily c member 1 and 2)) \(^{31 - 33}\) . In addition, L3MBTL3 interacts with transcription factor RBPJ (recombination signal binding protein for immunoglobulin kappa J region) to repress target gene transcription of Notch signaling \(^{34,35}\) . Knocking out L3MBTL3 in mice leads to immature myeloid progenitors accumulation, and hence decreases mature myeloid blood cells, causing embryonic lethal at the late stage \(^{36}\) . The GWAS (post- genome- wide association study) analysis shows L3MBTL3 is associated with multiple sclerosis \(^{37,38}\) . In cancer, there are a cancer risk loci of L3MBTL3 (rs9375701), which significantly associated with low L3MBTL3 expression in breast and prostate cancer \(^{39}\) . However, the functions of L3MBTL3 in breast cancer are still unclear. + +In our study, we found L3MBTL3 was positively correlated with metastatic and poor prognosis in breast cancer and promoted epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. Protein interaction analysis revealed L3MBTL3 interacted with active STAT3. L3MBTL3 could recruit active STAT3 on SNAIL promoter to up- regulated SNAIL transcription. The interaction between L3MBTL3 and active STAT3 was required for SNAIL transcription up- regulation, migration and invasion in breast cancer, while the methylated lysine binding activity of L3MBTL3 is not required for these functions. Overall, these findings indicate that L3MBTL3 interacting with active STAT3 and up- regulates SNAIL transcription to promote breast cancer metastasis. + +<--- Page Split ---> + +## Results + +High expression of L3MBTL3 is associated with metastatic and poor prognosis in breast cancer. + +In order to explore the function of L3MBTL3 in breast cancer, we first examined the expression of L3MBTL3 in 43 primary breast tumor samples (Tumor) and paired lymph node metastasis samples (Metastasis) and tumor- adjacent tissue samples (AdjN) by immunohistochemistry (IHC). L3MBTL3 protein levels significantly increased in lymph node metastasis of breast tumor comparing to primary breast tumor (Fig. 1a and Fig. 1b). We also noticed that L3MBTL3 highly expressed and localized in nuclear in mammary gland (Fig. 1a and Fig. 1b), while in lymph node metastasis samples L3MBTL3 highly expressed and localized in the nuclear and cytoplasm (Fig. 1a). We then analyzed datasets of Breast Cancer Integrative Platform (http://www.omicsnet.org/bcancer/database), which integrated gene expression and clinical data, and found that high L3MBTL3 expression group (n=103) showed lower distant metastasis free survival comparing to low L3MBTL3 expression group (n=99) (GSE12276 (n=202)) (Fig. 1c). Therefore, these data suggest that high L3MBTL3 expression is associated with breast cancer metastasis and poor prognosis. + +L3MBTL3 promotes epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. + +The above observations prompted us to investigate the function of L3MBTL3 in breast cancer metastasis. We examined the expression levels of L3MBTL3 in different breast cancer cell lines and found that L3MBTL3 expression positively correlated with key epithelial- mesenchymal transition transcription factor SNAIL, and negatively correlated with epithelial cell marker E- cad which is transcriptional down- regulated by + +<--- Page Split ---> + +SNAIL (Fig. 2a). We established L3MBTL3 over-expression stable cell line in MCF7 cells, L3MBTL3 knockout and rescue stable cell line in MDA- MB- 231 and SUM159PT cells (Fig. 2b), then observed the L3MBTL3 function in epithelial- mesenchymal transition, migration and invasion of breast cancer cells. We found over- expressing L3MBTL3 resulted in a spindle like mesenchymal phenotype and loss of cell contact in MCF7 cell (Fig. 2c). Conversely, knocking out L3MBTL3 in MDA- MB- 231 and SUM159PT cell showed epithelial like morphology (Fig. 2d). And restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 and SUM159PT cell rescued the mesenchymal morphology (Fig. 2e). We also observed that over- expression L3MBTL3 greatly promoted the migration of MCF7 cell (Supplementary Fig. 1a and Fig. 2f). In contrast, knocking out L3MBTL3 significantly inhibited migration and invasion in MDA- MB- 231 and SUM159PT cells (Supplementary Fig. 1b and Fig. 2g). And restoring L3MBTL3 expression significantly rescued the migration and invasion ability of MDA- MB- 231 and SUM159PT cells (Supplementary Fig. 1c and Fig. 2h). In addition, we established L3MBTL3 knockdown stable cell line in mouse breast cancer cell 4T1 (Fig. 2i) to verify the metastasis- promoting effect of L3MBTL3 in breast cancer using breast cancer lung metastasis mouse model. We found silencing L3MBTL3 significantly inhibited tumor lung metastasis in vivo, either (Fig. 2j). Thus, these results suggest that L3MBTL3 promotes epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. + +## L3MBTL3 interacts with STAT3 and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. + +To further explore the mechanism of L3MBTL3, we performed RNA- seq in L3MBTL3 knockout SUM159PT cells. GSEA analysis of hallmark pathway gene signatures + +<--- Page Split ---> + +showed knocking out L3MBTL3 significantly down- regulated epithelial- mesenchymal transition pathway (Fig. 3a). We also performed immunoprecipitation in L3MBTL3 rescue MDA- MB- 231 cells and identified the interaction protein of L3MBTL3 through Mass Spectrum (Fig. 3b). According to previous study, L3MBTL3 binds histone lysine methylation30, and interacts with transcription factor RBPJ to repress target gene transcription of Notch signaling34,35. + +The immunofluorescence data showed L3MBTL3 mainly localized in the nucleus (Fig. 3d), suggesting L3MBTL3 functions primarily in the nucleus. So we selected the L3MBTL3 interaction protein, which was involving in epithelial- mesenchymal transition process and localized in nuclear. We found 71 L3MBTL3 interaction protein, including transcription factor STAT3, TP53 and STAT1 (Fig. 3c and Supplementary Table 3). In addition, the GSEA analysis of hallmark pathway gene signatures showed knocking out L3MBTL3 significantly down- regulated IL6- JAK- STAT3 signaling pathway (Fig. 3a), which also promoted epithelial- mesenchymal transition and metastasis in cancer. The immunofluorescence data also showed L3MBTL3 and STAT3 co- localized in nuclear (Fig. 3d). These data suggest that L3MBTL3 may interact with STAT3 to regulate the transcription of target genes. Further immunoprecipitation showed L3MBTL3 endogenously interacted with STAT3 in MDA- MB- 231 and SUM159PT cells (Fig. 3e), and exogenously interacted with STAT3 in Flag- L3MBTL3 and HA- STAT3 over- expression 293T cells and L3MBTL3 rescue MDA- MB- 231 cells (Fig. 3f and Fig. 3g). Interestingly, we found L3MBTL3 showed much stronger binding ability with p- STAT3Y705 (active form of STAT3) than STAT3 both in endogenous (Fig. 3e) and exogenous (Fig. 3g). In order to further confirm the active STAT3 binding specificity of L3MBTL3, we expressed and purified GST- STAT3, GST- STAT3Y705F (STAT3 inactive form) and GST- STAT3 C (STAT3 + +<--- Page Split ---> + +constantly active form) by E. coli, and performed GST-pulldown assay. We observed that the interaction between L3MBTL3 and STAT3 was significantly increased after STAT3 was constantly activated (STAT3 C) and decreased after STAT3 was inactivated (STAT3Y705F) (Fig. 3h). In addition, we over-expressed L3MBTL3, HA- STAT3, HA- STAT3Y705F (STAT3 inactive form) and HA- STAT3Y705E (mimic p- STAT3Y705, STAT3 active form) in 293T cells and performed immunoprecipitation. We found the interaction between L3MBTL3 and STAT3 was significantly decreased after STAT3 was inactivated (STAT3Y705F) but greatly increased after STAT3 was activated (STAT3Y705E) (Fig. 3i). As we know, IL6 induces STAT3 phosphorylation and then activates STAT3 signaling pathway40. We treated L3MBTL3 rescue MDA- MB- 231 cells with IL6, then examined the interaction between L3MBTL3 and STAT3. The immunoprecipitation showed both the STAT3 and p- STAT3Y705 binding ability of L3MBTL3 increased after IL6 treatment (Fig. 3j), suggesting the IL6 activated STAT3 to enhance the interaction between L3MBTL3 and STAT3. Together, these results indicates that L3MBTL3 interacts with STAT3, and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. + +## L3MBTL3 up-regulates epithelial-mesenchymal transition and IL6-JAK-STAT3 pathway in breast cancer. + +To further confirm the regulation function of L3MBTL3 in epithelial- mesenchymal transition and IL6- JAK- STAT3 signaling pathway, we also performed RNA- seq in L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells. GSEA analysis of hallmark pathway gene signatures showed knocking out L3MBTL3 revealed robust disruptions of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes in SUM159PT cells (Fig. 4a and Fig. 4b), but over- expressing + +<--- Page Split ---> + +L3MBTL3 in MCF7 cells or restoring L3MBTL3 in L3MBTL3 knockout SUM159PT cells significantly up- regulated the transcription of epithelial- mesenchymal transition signature target genes (Fig. 4a). We further used the TCGA datasets (TCGA BRCA 2000 and TCGA BRCA 2017) and GEO datasets (GSE202203) to perform the GSVA analysis according to L3MBTL3 expression level. Consist with the GSEA analysis results in breast cancer cells, the expression of epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway genes were elevated more in L3MBTL3 high tumors than in L3MBTL3 low tumors (Fig. 4c). In support of the notion that L3MBTL3 is a candidate driver of epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in breast cancer, we further analyzed the relationship between L3MBTL3 expression and epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in TCGA breast cancer datasets (TCGA BRCA 2000 and TCGA BRCA 2017) and GEO datasets (GSE202203). The analysis results showed that L3MBTL3 expression positively association with epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway (Fig. 4d). Thus, these data suggest L3MBTL3 up- regulates epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in breast cancer. + +## L3MBTL3 promotes transcription of SNAIL in breast cancer. + +We performed ChIP- seq to further analyze the transcription regulation mechanism of L3MBTL3 in SUM159PT cells. ChIP- seq analysis showed L3MBTL3 mainly bound on the promoter (77.4% of L3MBTL3 binding sites), specially near the transcription start site (promoter (<=1kb), 67.32% of L3MBTL3 binding sites) (Supplementary Fig. 2a). The L3MBTL3 binding peaks showed similar pattern with STAT341 in genome- wide (Fig. 5a). Further analysis showed that there were 4878 genes both occupied by STAT3 and L3MBTL3 on promoter in SUM159PT cells (Fig. 5b). We then overlap + +<--- Page Split ---> + +these genes with 932 different expression genes in L3MBTL3 rescue SUM159PT cells, and found 267 different expression genes (Fig. 5b), which suggested L3MBTL3 and STAT3 could collaboratively regulate transcription of target genes. After analyzing the 267 different expression genes, we found the transcription of epithelial-mesenchymal transition up-regulated genes were decreased after knocking out L3MBTL3 in SUM159PT cells, but increased after restoring L3MBTL3 expression in L3MBTL3 knockout SUM159PT cells (Fig. 5c). In these genes, we identified the key epithelial-mesenchymal transition transcription factor- SNAIL, which often inhibited its target gene transcription in epithelial-mesenchymal transition process \(^{3 - 5}\) . There were a significant increase trend of SNAIL transcription and a decrease trend of SNAIL target genes E- cad, OCLN and DSP transcription after L3MBTL3 over-expressing in MCF7 cells (Fig. 5d). Western blot showed that the protein level of SNAIL greatly increased and E- cad greatly decreased in L3MBTL3 over-expression MCF7 cells (Fig. 5e). Immunofluorescence staining also showed that over-expressing L3MBTL3 significantly decreased E- cad expression in MCF7 cells (Fig. 5f). We also observed that there were a significant decrease trend of SNAIL transcription and a increase trend of SNAIL target genes OCLN, DSP and CLDN9 transcription after knocking out L3MBTL3 in MDA- MB- 231 and SUM159PT cells (Fig. 5g). Western blot also showed that knocking out L3MBTL3 significantly decreased SNAIL protein level in MDA- MB- 231 and SUM159PT cells (Fig. 5i). Restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 and SUM159PT cells recused SNAIL transcription (Fig. 5h) and protein level (Fig. 5j), and inhibited the transcription of SNAIL target genes OCLN, DSP and CLDN9 (Fig. 5h). + +We further analyzed the expression association between L3MBTL3 and SNAIL in TCGA datasets (TCGA BRCA 2000 (n=1097) and TCGA BRCA 2017 (n=1218)) and + +<--- Page Split ---> + +GEO datasets (GSE202203 (n=3207)). The L3MBTL3 expression was positive association with the SNAIL expression in breast cancer tumors (Fig. 6a). We also analyzed the relationship between the co-expression level of L3MBTL3 and SNAIL and the distant metastasis free survival using the GEO dataset (GSE12276, n=202) and found that breast cancer patients with high L3MBTL3 and high SNAIL expression group (red line, n=53) showed lower distant metastasis free survival comparing with low L3MBTL3 and low SNAIL expression group (blue line, n=49) (Fig. 6b). The immunohistochemistry analysis also showed that L3MBTL3 expression positively correlated with SNAIL expression in primary tumor (Tumor) and lymph node metastasis (Metastasis) of breast cancer samples (Fig. 6c). All these data indicated that L3MBTL3 up-regulated SNAIL transcription in breast cancer. + +## L3MBTL3 recruits active STAT3 on SNAIL promoter to up-regulate SNAIL transcription. + +To further study the L3MBTL3 regulation mechanism on SNAIL gene in breast cancer, we analyzed L3MBTL3 and STAT341 ChIP-seq results and found that L3MBTL3 and STAT3 co- localized on SNAIL promoter (Fig. 7a). The ChIP assay showed region P3 (- 563, - 474) was the key binding site of L3MBTL3 on SNAIL promoter (Fig. 7b). The further ChIP assay showed L3MBTL3 and active STAT3 (p- STAT3Y705) co- localized on the region P3 of SNAIL promoter in MDA- MB- 231 (Fig. 7d) and SUM159PT (Fig. 7e) cells. Knocking out L3MBTL3 significantly reduced the binding ability of p- STAT3Y705 on region P3 of SNAIL promoter in MDA- MB- 231 (Fig. 7d) and SUM159PT (Fig. 7e) cells. And restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 cells significantly increased the binding ability of p- STAT3Y705 on region P3 of SNAIL promoter (Fig. 7d). We also cloning the SNAIL + +<--- Page Split ---> + +promoter region (- 600, +80) (including region P3) in pGL3 plasmid and performed Luciferase assay. The results showed that over- expressing L3MBTL3 significantly promoted the SNAIL promoter region (- 600, +80) triggered Luciferase reporter gene transcription (Fig. 7e and Fig. 7f). However, STAT3 inhibitor NIF could totally depleted the transcription up- regulation mediated by L3MBTL3 (Fig. 7f). Thus, all these results indicated L3MBTL3 recruited active STAT3 on SNAIL promoter to up- regulated SNAIL transcription in breast cancer. + +Considering STAT3 often binds target gene through the STAT3 DNA binding motif on target gene promoter \(^{28,29}\) , we performed motif analysis of L3MBTL3 binding sites and found that there were some sites enriched for STAT3 DNA binding motif (Supplementary Fig. 3a). However, on SNAIL promoter, we observed that the L3MBTL3- STAT3 co- localization sites are including neither the classical STAT3 DNA binding motif (TTC(N3)GAA) nor other reported STAT3 DNA binding motif (such as AGG(N3)AGG) \(^{42}\) . These results are consistent with the report that the DNA- binding activity of STAT3 and canonical STAT3 DNA binding motif on the SNAIL promoter are not required for SNAIL transcription up- regulation \(^{12}\) . Our results can help to explain how STAT3 up- regulated SNAIL transcription through a DNA- binding independent way. + +## L3MBTL3 up-regulates SNAIL expression and promotes migration and invasion dependent on STAT3 in breast cancer. + +To further evaluate the co- transcriptional regulation function of L3MBTL3 and STAT3 in breast cancer, we used STAT3 siRNA and inhibitor in L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells. Knocking down STAT3 significantly reduced SNAIL expression (Fig. 8a and Fig. 8b). The migration assay + +<--- Page Split ---> + +showed that knocking down STAT3 significantly reduced migration ability of L3MBTL3 over- expression MCF7 cells (Supplementary Fig. 4a and Fig. 8c) and L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4b and Fig. 8d). The invasion assay also showed that knocking down STAT3 greatly reduced invasion ability of L3MBTL3 rescue SUM159PT cell (Supplementary Fig. 4c and Fig. 8e). In L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells, inhibiting STAT3 activity by STAT3 inhibitor NIF also significantly reduced SNAIL expression (Fig. 8f and Fig. 8g). The migration assay showed that inhibiting STAT3 activity by NIF significantly reduced migration ability of L3MBTL3 over- expression MCF7 cells (Supplementary Fig. 4d and Fig. 8h) and L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4e and Fig. 8i). The invasion assay also showed that inhibiting STAT3 activity by NIF significantly reduced invasion ability of L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4f and Fig. 8j). Thus, all these results indicate L3MBTL3 up- regulates SNAIL expression and promotes migration and invasion dependent on STAT3 in breast cancer cell. + +## The interaction between L3MBTL3 and activated STAT3 is required for SNAIL transcription up-regulation, migration and invasion in breast cancer. + +In order to study the characterization of the molecular interplay between STAT3 and L3MBTL3, a series of L3MBTL3 deletion mutants were employed to identify p- STAT3Y705 interacting domain of L3MBTL3. In GST- pulldown assay, we observed that L3- 1 (1- 204aa) was required for the L3MBTL3/p- STAT3Y705 interaction (Fig. 9a). Immunoprecipitation assay also showed that the L3MBTL3 wildtype (Flag- L3) and L3MBTL3 deletion mutants containing L3- 1 (Flag- L3- 5 and Flag- L3- 6) showed the p- STAT3Y705 binding ability, while the L3MBTL3 deletion mutants without L3- 1 (Flag + +<--- Page Split ---> + +L3- 2 and Flag- L3- 4) lost p- STAT3Y705 binding ability (Fig. 9b). As we narrowed down our analysis to even shorter region, we identified 176- 193aa region was the key p- STAT3Y705 binding site of L3MBTL3 (Fig. 9c and Fig. 9d). Immunoprecipitation showed that deleting 176- 193aa region of L3MBTL3 significantly reduced the interaction between L3MBTL3 and STAT3 or p- STAT3Y705 in 293T (Fig. 10a and Fig. 10b). In order to study the function of L3MBTL3 176- 193aa region, we expressed L3MBTL3 wild- type and L3MBTL3 mutation L3- \(\Delta 176 - 193\) in MCF7 or L3MBTL3 knockout SUM159PT cells. The Q- PCR and WB showed that deleting 176- 193aa region of L3MBTL3 impeded the increase of SNAIL expression induced by L3MBTL3 (Fig. 10c and Fig. 10d). Comparing to L3MBTL3 wild- type, the migration and invasion ability was remarkably weakened after deleting 176- 193aa region of L3MBTL3 (Supplementary Fig. 5a, Supplementary Fig. 5b and Fig. 10e). + +It was reported that L3MBTL3 as a lysine methylated reader protein, bound methylated lysine through its key methylated lysine binding site D381 to degrade substrate protein31,32. To examine whether L3MBTL3 promoted epithelial- mesenchymal transition, migration and invasion dependent on its lysine methylated reader ability, we expressed L3MBTL3 wild- type and L3MBTL3 methylated lysine binding mutation L3MBTL3- D381N in MCF7 or L3MBTL3 knockout SUM159PT cells. Immunoprecipitation showed that L3MBTL3 D381N mutation did not significantly change the interaction between L3MBTL3 and STAT3 or p- STAT3Y705 (Supplementary Fig. 6a and Supplementary Fig. 6b). Comparing to L3MBTL3 wild- type, L3MBTL3 D381N mutation did not significantly change the SNAIL expression (Supplementary Fig. 6c and Supplementary Fig. 6d) and migration and invasion ability (Supplementary Fig. 6e). + +Thus, all these data indicated the interaction between L3MBTL3 and active + +<--- Page Split ---> + +STAT3 is required for epithelial- mesenchymal transition, migration and invasion in breast cancer while the methylated lysine binding ability of L3MBTL3 is not required for these functions described above. + +In conclusion, L3MBTL3 and active STAT3 collaboratively up- regulate SNAIL transcription to promote breast cancer metastasis (Fig. 10f). + +## Discussion + +As previously reported, there are a cancer risk loci of L3MBTL3 (rs9375701), which significantly associated with low L3MBTL3 expression in breast cancer39, suggesting L3MBTL3 may negatively associate with breast cancer. Consist with this report, we also observed the low expression of L3MBTL3 in primary breast tumor. However, our study showed that L3MBTL3 was positively correlated with metastatic and poor prognosis in breast cancer, suggesting an important role of L3MBTL3 in breast cancer metastasis. This speculation was confirmed by our observation that L3MBTL3 play a crucial role in promoting epithelial- mesenchymal transition, migration, invasion and breast tumor metastasis. Our results reveal that L3MBTL3 interacts with active STAT3 at shared binding sites on SNAIL promoter to up- regulate SNAIL transcription and promote epithelial- mesenchymal transition, migration and invasion. Therefore, we disclose a metastasis enhancer function of L3MBTL3 in breast cancer in this work. + +L3MBTL3 has been known as a chromatin- interacting transcriptional repressor, interacting with RBPJ to repress target gene transcription of Notch signaling34. However, we described a transcriptional activation role for L3MBTL3 in breast cancer. Knocking out L3MBTL3 revealed robust disruptions of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes in breast cancer cells. In contrast, over- expressing L3MBTL3 or restoring L3MBTL3 in L3MBTL3 knockout breast + +<--- Page Split ---> + +cancer cell significantly up- regulated the transcription of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes. In addition, clinical data also suggest that L3MBTL3 expression positively correlated with epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target gene transcription in breast cancer. Our study also showed L3MBTL3 up- regulated the transcription of SNAIL by recruiting active STAT3 on SNAIL promoter. These results indicate the transcription activator function of L3MBTL3. + +It is reported that L3MBTL3 as a lysine methylation reader, binds the lysine methylation of non- histone protein by its MBT domain \(^{31,32}\) and recruits CRL4 \(^{DCAF5}\) ubiquitin E3 ligase to mediate the methylation- dependent degradation \(^{31 - 33}\) . In our study, we used L3MBTL3 lysine methylation binding activity mutant D381N to explore epithelial- mesenchymal transition and metastasis function of L3MBTL3. We found D381N mutant did not significantly change the expression of SNAIL, cell migration and invasion, indicating that the regulation mechanism of L3MBTL3 on epithelial- mesenchymal transition, migration and invasion are lysine methylation binding activity independent. + +STAT3 as a transcription factor often up- regulates the transcription of target gene through its DNA binding domain. However, previous study is reported that STAT3 bound on SNAIL promoter in DNA binding independent way \(^{12}\) . In our study, L3MBTL3 interacted with active STAT3 and recruited active STAT3 on SNAIL promoter, then up- regulated SNAIL transcription to promote epithelial- mesenchymal transition and metastasis in breast cancer. The interaction between L3MBTL3 and active STAT3 is required for SNAIL transcription upregulation and breast cancer cell invasion and migration. Sequence analysis showed that the L3MBTL3- STAT3 colocalization sites on SNAIL promoter is including neither the classical nor other known STAT3 DNA + +<--- Page Split ---> + +binding motif. These findings support the notion that STAT3 binds on SNAIL promoter and up- regulated transcription independent on DNA binding of STAT312. Our results further explain how STAT3 is recruited on SNAIL promoter and up- regulates SNAIL transcription in DNA binding independent way. + +STAT3 is phosphorylated at its Y705 site to form dimer and localizes into the nucleus, then binds the promoter of target gene to up- regulate transcription28,29. We reveal that L3MBTL3 prefers to interact with active STAT3, implying that L3MBTL3 may play a role in STAT3 activation. We observed that the phosphorylation level of STAT3 was not changed in L3MBTL3 over- expression or knockout breast cancer cells, indicated that L3MBTL3 did not regulated STAT3 phosphorylation. However, after STAT3 been activated, the increased STAT3 binding ability of L3MBTL3 suggests L3MBTL3 prefers to bind Y705 phosphorylated STAT3 or STAT3 dimer. In addition, phosphorylated STAT3 often forms an anti- parallel dimer, so it is possible that two molecule L3MBTL3 binds STAT3 anti- parallel dimer to form L3MBTL3- STAT3 tetramer. We also mapped 176- 193aa region of L3MBTL3 as the key STAT3 binding region. The identification of key STAT3 binding region of L3MBTL3 can provide important information for designing compounds or peptides to disrupt L3MBTL3- STAT3 interaction against metastasis. The interaction working model between L3MBTL3 and STAT3 is still waiting for discovering. Further biochemical characterization combined with structural analysis should help to resolve these questions and treat breast cancer metastasis. + +## Methods + +## Clinical specimens + +<--- Page Split ---> + +43 pairs of HCC (female patients, the age range is between 25 and 79) samples were obtained from the Second Affiliated Hospital of Chinese University of Hong Kong (Shenzhen, China). All patients were diagnosed with breast cancer by postoperative pathology and were free of other cancers and chronic diseases. All samples were collected with the informed consent of the patients and the experiments were approved by research ethics committee at the Second Affiliated Hospital of Chinese University of Hong Kong. + +## Plasmids, sgRNA, siRNA and shRNA + +The sequence of L3MBTL3 ORF used in the plasmids was the CDS sequence of NM_001007102 in GenBank™. The lentivirus plasmid Plvx- IRES- Neo- FLAG- L3MBTL3, the eukaryotic expression plasmids pcDNA3- Flag- L3MBTL3, pcDNA3- Flag- L3- 2, pcDNA3- Flag- L3- 4, pcDNA3- Flag- L3- 5, pcDNA3- Flag- L3- 6, and the prokaryotic expression plasmids pGEX4T- GST- L3- 1, pGEX4T- GST- L3- 2, and pGEX4T- GST- L3- 3 were constructed using lentivirus plasmid GFP- L3MBTL3. The eukaryotic expression plasmids pcDNA3- FLAG- L3- Δ176- 193 and pcDNA3- FLAG- L3- D381N were constructed using pcDNA3- FLAG- L3MBTL3 as PCR template follow the Site- Directed Mutagenesis protocol. The prokaryotic expression plasmids pGEX4T- GST- L3- 1- Δ2- 64, pGEX4T- GST- L3- 1- Δ65- 104, pGEX4T- GST- L3- 1- Δ105- 204, pGEX4T- GST- L3- 1- Δ105- 137, pGEX4T- GST- L3- 1- Δ138- 162, pGEX4T- GST- L3- 1- Δ163- 175, pGEX4T- GST- L3- 1- Δ176- 193 and pGEX4T- GST- L3- 1- Δ194- 204 were constructed using pGEX4T- GST- L3- 1 as PCR template follow the Site- Directed Mutagenesis. The eukaryotic expression plasmids pcDNA3- HA- STAT3Y705F were constructed using the lentivirus plasmid cFUGW- Flag- STAT3Y705F. The eukaryotic expression plasmids pcDNA3- HA- STAT3Y705E were constructed using + +<--- Page Split ---> + +pcDNA3- HA- STAT3 as PCR template follow the Site- Directed Mutagenesis protocol. + +The prokaryotic expression plasmid pGEX4T- GST- STAT3, pGEX4T- GST- + +STAT3Y705F and pGEX4T- GST- STAT3 C were constructed using pcDNA3- HA- + +STAT3, cFUGW- Flag- STAT3Y705F and cFUGW- Flag- STAT3 C, respectively. pGL3- + +SNAIL (- 600, +80) was constructed using genomic DNA of 293T cell. + +The small interfering RNA of negative control (siNC) and STAT3 siRNA (siSTAT3- 1 and siSTAT3- 2) were purchased from GenePharma (Shanghai, China). The sgRNAs of L3MBTL3 (L3- KO4 and L3- KO5) and the shRNAs of NC and mouse L3MBTL3 shRNA (L3- KD1 and L3- KD2) were purchased from TranSheepBio (Suzhou, China). The sequence of sgRNA, siRNA and shRNA were shown in + +(Suzhou, China). The sequence of sgRNA, siRNA and shRNA were shown in + +Supplementary Table 1. + +## Generation of stable cell lines + +Lentiviral plasmid GFP- L3MBTL3 was used to generate L3MBTL3 over- expression MCF7 stable cell line and the positive cells were screened by Flow cytometer. The L3MBTL3 knockout MDA- MB- 231 or SUM159PT stable cell lines or the L3MBTL3 knockdown 4T1 stable cell line were established by sgRNA or shRNA, respectively. The positive cells were screened by growth with puromycin (#A11138- 03, Gibco, USA) for two weeks. The L3MBTL3 rescue MDA- MB- 231 or SUM159PT stable cell lines were established by using lentivirus plasmid Plvx- IRES- Neo- FLAG- L3MBTL3 and L3- KO5 MDA- MB- 231 or SUM159PT stable cell lines, and the positive cells were screened by growth with G418 (#A1720- 5G, Merck, USA) for two weeks. The expression level of L3MBTL3 was analyzed by Q- PCR or Western blot. + +## Antibodies, inhibitor and cytokine + +<--- Page Split ---> + +Antibodies against GAPDH (#AC002, ABclonal Technology, China), \(\beta\) - ACTIN (#20536- 1- AP, Proteintech, China), E- cad (#3195S, Cell Signaling Technology, USA), SNAIL (#3879S, Cell Signaling Technology, USA), STAT3 (#60199- 1- Ig Proteintech, USA), p- STAT3Y705 (#9145S, Cell Signaling Technology, USA), HA (#3724S, Cell Signaling Technology, USA), FLAG (#8146S, Cell Signaling Technology, USA) and GST (#AF2299, Beyotime, China) were used for Western blot. L3MBTL3 (#ab99928, Abcam, UK) and L3MBTL3 (#A7289, ABclonal Technology, China) were used for human and mouse L3MBTL3 Western blot, respectively. + +Antibodies against L3MBTL3 (#ab99928, Abcam, UK), STAT3 (#60199- 1- Ig Proteintech, USA) and p- STAT3Y705 (#9145S, Cell Signaling Technology, USA) were used for and immunoprecipitation. + +Antibodies against L3MBTL3 (#a302- 854a, Bethyl, USA) and p- STAT3Y705 (#9145S, Cell Signaling Technology, USA) were used for chromatin immunoprecipitation. + +Antibodies against L3MBTL3 (#ab99928, Abcam, UK) and SNAIL (#A5544, ABclonal Technology, China) were used for Immunohistochemistry. + +Antibody against E- cad (#610181, BD Biosciences, USA), STAT3 (#60199- 1- Ig Proteintech, USA) and FLAG (#8146S, Cell Signaling Technology, USA) were used for immunofluorescence. + +To inhibit STAT3 pathway, the cells were treated with growth medium containing \(100 \mu \mathrm{M}\) STAT3 inhibitor NIF (HY- B1436, MedChemExpress, USA) for 16 hours. To activate STAT3 pathway, the cells were treated with growth medium containing \(10 \mathrm{ng / ml}\) Cytokine IL- 6 (#AF- 200- 06, Peprotech, USA) for 30 min. + +# Immunohistochemistry + +<--- Page Split ---> + +Immunohistochemistry was performed according to standard protocol. In brief, slides were incubated at \(60^{\circ}\mathrm{C}\) for \(1\mathrm{h}\) , followed by \(10\mathrm{min}\) incubations in xylene and rehydration in graded alcohol to water. Sodium citrate buffer (pH 6.0) was used for antigen retrieval and slides were incubated with primary antibody overnight at \(4^{\circ}\mathrm{C}\) and stained with VECTASTAIN Elite ABC kit (PK- 4001, Vector labs, USA). Bound antibody was detected with DAB (3,3'- diaminobenzidine) peroxidase kit (SK4100, Vector labs, USA). + +Staining intensity and extent of staining were graded as follows: negative (score 0), weak (score 1), moderate (score 2), and strong (score 3). Extent of staining was also grouped into quantiles according to the percentage of high- staining cells per field: negative (score 0), \(25\%\) (score 1), \(26\% - 50\%\) (score 2), \(51\% - 75\%\) (score 3), and \(76\% - 100\%\) (score 4). All immunohistochemical staining was evaluated and scored by at least two independent pathologists blinded to the experimental groups. + +## Migration and invasion assay + +The migration and invasion assay were performed as previously described \(^{43}\) . \(8\mu \mathrm{m}\) pore size Boyden chambers (353097, Corning, USA) were used to migration assay. \(8\mu \mathrm{m}\) pore size Byden chambers (353097, Corning, USA) precoated with Matrigel (356231, Coring, USA) were used to invasion assay. + +## Metastasis assay + +We used a protocol approved by the Ethics Committee of East China Normal University. 6- week- old female BALB/c mice were used to established spontaneous lung metastasis model. The NC and L3MBTL3 knockdown (L3- KD1 and L3- KD2) 4T1 cells were trypsinized and suspended in a 1:2 (vol/vol) mix of Matrigel (Corning) and PBS. Then + +<--- Page Split ---> + +the 4T1 cells (5×105 cells per gland, mice were randomly divided into 3 groups, 9 mice per group) were injected into the fourth mammary fat pads of BALB/c mice. All the mice were sacrificed at the end of the experiment and the lung were collected for metastasis analysis. + +## RNA-seq and Q-PCR + +The total mRNAs were extracted from cells using TRIzol reagent (15596018, Ambion, USA). For RNA- seq, the RNA- seq libraries from 200ng of total RNA were prepared by using Hieff NGS Ultima Dual- mode mRNA Library Prep Kit for Illumina (yeasen, Cat12301), according to the manufacturer's instructions. Libraries were validated with an QIAxcel Advanced system (QIAGEN, Germany). Sequencing was performed on Illumina HiSeq 2500 sequencer or Illumina NovaSeq 6000. + +For Q- PCR, reverse transcription was performed using PrimeScriptTM RT reagent kit (#RR047A, TaKaRa, Japan). The Q- PCR was performed using Power SYBR Green PCR Master Mix (#4367659, Applied Biosystems, USA) on ABI QuantStudio™ 6 Flex System. Primers of Q- PCR were shown in Supplementary Table 2. + +## GSEA and GSVA analysis + +The 50 hallmark gene sets in the MSigDB databases (https://www.gsea- msigdb.org/gsea/msigdb) were used for the Gene Set Enrichment Analysis (GSEA)44 or Gene Set Variation Analysis (GSVA)45 in RNA- seq of breast cancer cells, TCGA Breast Cancer datasets (TCGA BRCA 2017: https://xenabrowser.net/datapages/?dataset=TCGA.BRCA.sampleMap%2FHiSeqV2 &host=https%3A%2F%2Ftcga.xenahubs.net&removeHub=https%3A%2F%2Fxena.tr eehouse.gi.ucsc.edu%3A443 and TCGA BRCA 2000: + +<--- Page Split ---> + +https://portal.gdc.cancer.gov/repository?facetTab=cases&filters=%7B%22content%2 + +2%3A%5B%7B%22content%22%3A%7B%22field%22%3A%22cases.project.projec + +t_id%22%2C%22value%22%3A%5B%22TCGA- + +BRCA%22%5D%7D%2C%22op%22%3A%22in%22%7D%2C%7B%22content%22 + +%3A%7B%22field%22%3A%22files.data_category%22%2C%22value%22%3A%5 + +B%22Transcriptome%20Profiling%22%5D%7D%2C%22op%22%3A%22in%22%7 + +D%5D%2C%22op%22%3A%22and%22%7D&searchTableTab=cases) and GEO + +datasets (GSE202203) to assess the relationship between L3MBTL3 expression level + +and epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway. + +## Immunofluorescence + +Cells were fixed with 4% paraformaldehyde for 20 min followed by a 15-min cell permeabilization with 1% (vol/vol) Triton X-100 in PBS. Then the cells were blocked in PBS with 2% (wt/vol) BSA for 1h at 37°C. Antibody were diluted in PBS with 2% (wt/vol) BSA and incubated with the cells overnight at 4°C. After rinsing, the cells were incubated with secondary antibodies conjugated to Alexa Fluor 568 (#A-11036, Invitrogen, USA) or Alexa Fluor 488 (#A11001, Invitrogen, USA) at 37°C for 30 min. After stained with DAPI (1 ng/ml)(#C1006, Beyotime, China), the cells were quickly rinsed in PBS and then imaged. + +## Immunoprecipitation, mass spectrum analysis and GST-pulldown + +For immunoprecipitation, cells were collected by centrifugation and lysed in lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.1% NP40, 8% glycerol, protease inhibitor cocktail (#P1010, Beyotime, China)) on ice for 30 min. Then the cell lysate supernatant was incubated with antibodies and protein A+G-agarose beads + +<--- Page Split ---> + +(#P2055, Beyotime, China), M2-conjugate agarose beads (#A2220, Merck, USA) or anti- HA nanobody conjugated agarose beads (#HNA- 25- 500, PromiOne, China) by rotation overnight at \(4^{\circ}\mathrm{C}\) . Then the beads were analyzed by Western blot after washed for 5 times. + +For mass spectrum analysis, the protein was immunoprecipitated using M2- conjugate agarose beads (#A2220, Merck, USA). After washing, the beads were eluted by \(3 \times \mathrm{FLAG}\) peptide (#F4799, Merck, USA). The elution protein was verified using silver staining, then the target lane was excised and subjected to analyze by an EASY- nLC 1200 UHPLC system (ThermoFisher Scientific, USA) coupled to a Q Exactive HF- X mass spectrometer (ThermoFisher Scientific, USA). + +For GST- pulldown assay, GST- tag protein was expressed in E. coli and incubated with the BeyoGold™ GST- tag Purification Resin (#P2251, Beyotime, China) beads. After washing, the beads were incubated with 293T cell lysate supernatant by rotation overnight at \(4^{\circ}\mathrm{C}\) . Then the beads were analyzed by Western blot after washed 5 times. + +## ChIP sequence and ChIP assay + +ChIP sequence and ChIP assayChIP were performed according to the protocol as previously described46. Briefly, cells were cross- linked using 1% formaldehyde and sonicated to generate DNA fragments of about 500 bp. For immunoprecipitation, Dynabeads™ Protein A or G magnetic beads (Invitrogen) were incubated with antibodies and chromatin fraction of cell lysate overnight at \(4^{\circ}\mathrm{C}\) with rotation. After washing, the protein- DNA complexes were eluted. Then the DNA was purified and used for Q- PCR. Primers were shown in + +For ChIP- seq, the sequencing libraries were generated using NEBNext Ultra DNA Library Prep Kit from Illumina (NEB, USA) by following manufacturer's + +<--- Page Split ---> + +recommendations and index codes were added to attribute sequences to each sample. The clustering of the index- coded samples was performed on a cBot Cluster Generation System using HiSeq Rapid Duo cBot Sample Loading Kit (Illumia) according to the manufacturer's instructions. The libraries were sequenced on the Illumina NovaSeq 6000. + +## ChIP-seq analysis and motif analysis + +After filtering and deduping, uniquely mapped tags of L3MBTL3 and STAT3 (GSE85579 and GSE152203)41 ChIP- seq were mapped to the Homo sapiens reference genome (hg19) by Bowtie2. Normalized genome- wide signal- coverage tracks from raw- read alignment files were built by Samtools, deeptools, and MACS2. Visualization of the ChIP- seq signal at enriched genomic regions (Profile and Heatmap) was achieved by using deepTools. Peak- associated genes and motif were identified by using the annotate Peaks function of HOMER. Each of the STAT3 or L3MBTL3- binding sites were assigned to the nearest gene. Further annotation information includes whether a peak is in the Promoter, 5' UTR, 3' UTR, Exon, Intron, Downstream or Distal Intergenic. + +For motif analysis, the DNA segments from L3MBTL3 or STAT3 binding sites were analyzed by homer script findMotifGenome.pl with argument "hg19" to detect enrichment of de novo and known TF motif. Homer results were further manually analyzed. Those motifs with high background enrichment and possible false positives were removed. + +## Luciferase assay + +Plasmids derived from pGL3 vectors were co- transfected with Renilla luciferase vectors (phRL- TK) and the luciferase assay were performed using the Dual- Luciferase + +<--- Page Split ---> + +Reporter Assay System (#E1910, Promega, USA) according to the manufacturer's instructions. + +## Statistical analysis + +SPSS version 17.0 software was used for statistical analysis. Results are expressed as mean \(\pm\) standard deviation (SD) and Student's t- test (two- tailed) was used to indicate significant differences between groups. In all experiments, \(\mathrm{p}< 0.05\) was considered to indicate statistical significance. + +## Data Availability + +The RNA- seq and ChIP- seq data generated in this study have been deposited in Sequence Read Archive (SRA) repository under SRA accession number PRJNA934109 and PRJNA934165, respectively. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE47 partner repository with the dataset identifier PXD040150. Source data are provided in this paper. + +## References + +1. Park, M. et al. Breast Cancer Metastasis: Mechanisms and Therapeutic Implications. Int J Mol Sci 23, 6806 (2022). +2. Buyuk, B., Jin, S. & Ye, K. Epithelial-to-Mesenchymal Transition Signaling Pathways Responsible for Breast Cancer Metastasis. Cell Mol Bioeng 15, 1–13 (2022). +3. Zhou, B. P. et al. Dual regulation of Snail by GSK-3beta-mediated phosphorylation in control of epithelial-mesenchymal transition. Nat Cell Biol 6, 931–940 (2004). + +<--- Page Split ---> + +4. Barralo-Gimeno, A. & Nieto, M. A. The Snail genes as inducers of cell movement and survival: implications in development and cancer. Development 132, 3151–3161 (2005). + +5. Lamouille, S., Xu, J. & Derynck, R. Molecular mechanisms of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol 15, 178–196 (2014). + +6. Olmeda, D., Jordá, M., Peinado, H., Fabra, A. & Cano, A. Snail silencing effectively suppresses tumour growth and invasiveness. Oncogene 26, 1862–1874 (2007). + +7. Olmeda, D. et al. Snail and Sna12 collaborate on tumor growth and metastasis properties of mouse skin carcinoma cell lines. Oncogene 27, 4690–4701 (2008). + +8. Azmi, A. S. et al. Systems analysis reveals a transcriptional reversal of the mesenchymal phenotype induced by SNAIL-inhibitor GN-25. BMC Syst Biol 7, 85 (2013). + +9. Ciruna, B. & Rossant, J. FGF signaling regulates mesoderm cell fate specification and morphogenetic movement at the primitive streak. Dev Cell 1, 37–49 (2001). + +10. Yook, J. I., Li, X.-Y., Ota, I., Fearon, E. R. & Weiss, S. J. Wnt-dependent regulation of the E-cadherin repressor snail. J Biol Chem 280, 11740–11748 (2005). + +11. Cho, H. J., Baek, K. E., Saika, S., Jeong, M.-J. & Yoo, J. Snail is required for transforming growth factor-beta-induced epithelial-mesenchymal transition by activating PI3 kinase/Akt signal pathway. Biochem Biophys Res Commun 353, 337–343 (2007). + +12. Saitoh, M. et al. STAT3 integrates cooperative Ras and TGF-β signals that induce Snail expression. Oncogene 35, 1049–1057 (2016). + +13. Talbot, L. J., Bhattacharya, S. D. & Kuo, P. C. Epithelial-mesenchymal transition, the tumor microenvironment, and metastatic behavior of epithelial malignancies. Int J Biochem Mol Biol 3, 117–136 (2012). + +<--- Page Split ---> + +14. Wang, H.-Q. et al. STAT3 pathway in cancers: Past, present, and future. MedComm (2020) 3, e124 (2022). + +15. Hirano, T., Ishihara, K. & Hibi, M. Roles of STAT3 in mediating the cell growth, differentiation and survival signals relayed through the IL-6 family of cytokine receptors. Oncogene 19, 2548–2556 (2000). + +16. Kamran, M. Z., Patil, P. & Gude, R. P. Role of STAT3 in cancer metastasis and translational advances. Biomed Res Int 2013, 421821 (2013). + +17. Silver, D. L., Naora, H., Liu, J., Cheng, W. & Montell, D. J. Activated signal transducer and activator of transcription (STAT) 3: localization in focal adhesions and function in ovarian cancer cell motility. Cancer Res 64, 3550–3558 (2004). + +18. Sano, S. et al. Keratinocyte-specific ablation of Stat3 exhibits impaired skin remodeling, but does not affect skin morphogenesis. EMBO J 18, 4657–4668 (1999). + +19. Azare, J. et al. Constitutively activated Stat3 induces tumorigenesis and enhances cell motility of prostate epithelial cells through integrin beta 6. Mol Cell Biol 27, 4444–4453 (2007). + +20. Suiqing, C., Min, Z. & Lirong, C. Overexpression of phosphorylated-STAT3 correlated with the invasion and metastasis of cutaneous squamous cell carcinoma. J Dermatol 32, 354–360 (2005). + +21. Qiu, Z. et al. RNA interference-mediated signal transducers and activators of transcription 3 gene silencing inhibits invasion and metastasis of human pancreatic cancer cells. Cancer Sci 98, 1099–1106 (2007). + +22. Li, H. dong et al. STAT3 knockdown reduces pancreatic cancer cell invasiveness and matrix metalloproteinase-7 expression in nude mice. PLoS One 6, e25941 (2011). + +23. Sullivan, N. J. et al. Interleukin-6 induces an epithelial-mesenchymal transition phenotype in + +<--- Page Split ---> + +human breast cancer cells. Oncogene 28, 2940- 2947 (2009). + +24. Yadav, A., Kumar, B., Datta, J., Teknos, T. N. & Kumar, P. IL-6 promotes head and neck tumor metastasis by inducing epithelial-mesenchymal transition via the JAK-STAT3-SNAIL signaling pathway. Mol Cancer Res 9, 1658-1667 (2011). + +25. Xiong, H. et al. Roles of STAT3 and ZEB1 proteins in E-cadherin down-regulation and human colorectal cancer epithelial-mesenchymal transition. J Biol Chem 287, 5819-5832 (2012). + +26. Gong, C. et al. Abnormally expressed JunB transactivated by IL-6/STAT3 signaling promotes uveal melanoma aggressiveness via epithelial-mesenchymal transition. Biosci Rep 38, BSR20180532 (2018). + +27. Teng, Y., Ross, J. L. & Cowell, J. K. The involvement of JAK-STAT3 in cell motility, invasion, and metastasis. JAKSTAT 3, e28086 (2014). + +28. Bromberg, J. & Darnell, J. E. The role of STATs in transcriptional control and their impact on cellular function. Oncogene 19, 2468-2473 (2000). + +29. Horvath, C. M., Wen, Z. & Darnell, J. E. A STAT protein domain that determines DNA sequence recognition suggests a novel DNA-binding domain. Genes Dev 9, 984-994 (1995). + +30. James, L. I. et al. Discovery of a chemical probe for the L3MBTL3 methyllysine reader domain. Nat Chem Biol 9, 184-191 (2013). + +31. Zhang, C. et al. Proteolysis of methylated SOX2 protein is regulated by L3MBTL3 and CRL4DCAF5 ubiquitin ligase. J Biol Chem 294, 476-489 (2019). + +32. Leng, F. et al. Methylated DNMT1 and E2F1 are targeted for proteolysis by L3MBTL3 and CRL4DCAF5 ubiquitin ligase. Nat Commun 9, 1641 (2018). + +33. Guo, P. et al. The assembly of mammalian SWI/SNF chromatin remodeling complexes is + +<--- Page Split ---> + +regulated by lysine-methylation dependent proteolysis. Nat Commun 13, 6696 (2022). + +34. Xu, T. et al. RBPJ/CBF1 interacts with L3MBTL3/MBT1 to promote repression of Notch signaling via histone demethylase KDM1A/LSD1. EMBO J 36, 3232-3249 (2017). + +35. Hall, D. et al. The structure, binding and function of a Notch transcription complex involving RBPJ and the epigenetic reader protein L3MBTL3. Nucleic Acids Res 50, 13083-13099 (2022). + +36. Arai, S. & Miyazaki, T. Impaired maturation of myeloid progenitors in mice lacking novel Polycomb group protein MBT-1. EMBO J 24, 1863-1873 (2005). + +37. Alcina, A. et al. Identification of the genetic mechanism that associates L3MBTL3 to multiple sclerosis. Hum Mol Genet 31, 2155-2163 (2022). + +38. Nazari Mehrabani, S. Z. et al. Association of SHMT1, MAZ, ERG, and L3MBTL3 Gene Polymorphisms with Susceptibility to Multiple Sclerosis. Biochem Genet 57, 355-370 (2019). + +39. Kar, S. P. et al. Genome-Wide Meta-Analyses of Breast, Ovarian, and Prostate Cancer Association Studies Identify Multiple New Susceptibility Loci Shared by at Least Two Cancer Types. Cancer Discov 6, 1052-1067 (2016). + +40. Kolosenko, I., Grander, D. & Tamm, K. P. IL-6 activated JAK/STAT3 pathway and sensitivity to Hsp90 inhibitors in multiple myeloma. Curr Med Chem 21, 3042-3047 (2014). + +41. Conway, M. E. et al. STAT3 and GR Cooperate to Drive Gene Expression and Growth of Basal-Like Triple-Negative Breast Cancer. Cancer Res 80, 4355-4370 (2020). + +42. Xu, L. et al. The STAT3 HIES mutation is a gain-of-function mutation that activates genes via AGG-element carrying promoters. Nucleic Acids Res 43, 8898-8912 (2015). + +43. Zhang, H. et al. Histone demethylase KDM2A suppresses EGF-TSPAN8 pathway to inhibit breast cancer cell migration and invasion in vitro. Biochem Biophys Res Commun 628, 104-109 + +<--- Page Split ---> + +754 (2022). 755 44. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for 756 interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545- 15550 (2005). 757 45. Liberzon, A. et al. Molecular signatures database (MSigDB) 3.0. Bioinformatics 27, 1739- 758 1740 (2011). 759 46. Schmidt, D. et al. ChIP-seq: using high-throughput sequencing to discover protein-DNA 760 interactions. Methods 48, 240- 248 (2009). 761 47. Perez- Riverol, Y. et al. The PRIDE database resources in 2022: a hub for mass spectrometry- 762 based proteomics evidences. Nucleic Acids Res 50, D543- D552 (2022). + +<--- Page Split ---> + +## Acknowledgements + +We thank Prof. Mingyao Liu and Prof. Zhengfang Yi from East China Normal University for providing cFUGW- Flag- STAT3Y705F and cFUGW- Flag- STAT3 C plasmids. We thank Prof. Jian Luo from Tongji University for providing pcDNA3- HA- STAT3 plasmid. We thank Prof. Arrowsmith from University of North Carolina at Chapel Hill for providing L3MBTL3 full- length plasmid GFP- L3MBTL3. This study was supported by grants from the National Natural Science Foundation of China (82173044 and 81772817), Shanghai Medical Leadership Program (2019LJ25), Medical and Health Technology Public Relations Project of Shenzhen Longgang District Science and Technology Innovation Special Fund (LGKCYLWS2022001), Special Program for Collaborative Innovation in Shanghai University of Medicine & Health Sciences (SPCI- 17- 15- 001), and Key Subject Construction Project of Shanghai Municipal Commission of Health and Family Planning, China (ZK2019B29). + +## Author Contributions + +Jing Feng and Jing Li: study design; Jianpeng Xiao, Jie Wang, Jialun Li, Jie Xiao, CuiCui Liu, Libi Tan, Yanhong Tu, Ruifang Yang, Yujie Pei, Minghua Wang, Jiemin Wong, Binhua P Zhou: study conduct, data analysis, and its interpretation; Jing Li and Jianpeng Xiao: manuscript drafting. + +## Competing Interests + +The authors declare no competing interests. + +## ORCID + +<--- Page Split ---> + +788 Jing Feng: http://orcid.org/0000- 0003- 1084- 4920 + +789 Jing Li: https://orcid.org/0000- 0003- 4852- 5781 + +<--- Page Split ---> + +Fig. 1 + +![PLACEHOLDER_36_0] + + + +![PLACEHOLDER_36_1] + + +<--- Page Split ---> + +Fig. 1 High L3MBLT3 expression is associated with metastatic and poor clinical outcomes of breast cancer. a Human breast cancer tissues consisting of primary breast cancer samples (Tumor) and paired tumor- adjacent tissue samples (AdjN) and breast cancer lymph node metastasis samples (Metastasis) were subjected to Immunohistochemistry staining for L3MBLT3 (scale bar, \(25 \mu \mathrm{m}\) ). b H- scores of L3MBTL3 protein in 43 breast cancer patients of primary breast cancer samples (Tumor) and paired tumor- adjacent tissue samples (AdjN) and breast cancer lymph node metastasis samples (Metastasis). The data were analyzed using Student's t test. \(\mathrm{n} = 43\) . \(^{*}\mathrm{p}< 0.05\) , \(***\mathrm{p}< 0.001\) . c Distant metastasis survival of breast cancer patients with L3MBLT3 high and low expression (GSE12276, \(\mathrm{n} = 202\) ). L3MBLT3 high, \(\mathrm{n} = 103\) ; L3MBLT3 low, \(\mathrm{n} = 99\) ; \(\mathrm{p} = 0.00153024\) . + +<--- Page Split ---> + +
Fig. 2
+ +![PLACEHOLDER_38_0] + + +<--- Page Split ---> + +Fig. 2 L3MBTL3 promotes epithelial-mesenchymal transition, migration and invasion and metastasis in breast cancer. a Western blot showed L3MBTL3 highly expressed in basal-like breast cancer cells, which was positively correlated with epithelial-mesenchymal transition factor SNAIL and negatively correlated with epithelial cell marker E-cad. Antibody \(\beta\) - ACTIN was used as a loading control. b Western blot showed the L3MBTL3 expression level in L3MBTL3 over-expression MCF7 cells, L3MBTL3 knockout and rescue MDA-MB-231 and SUM159PT cells. Antibody GAPDH and \(\beta\) - ACTIN were used as loading controls. c Over-expressing L3MBTL3 converted MCF7 cell into spindle-shaped mesenchymal-like cells (Scale bar, \(100 \mu \mathrm{m}\) ). d Knocking out L3MBTL3 converted MDA-MB-231 and SUM159PT cell into epithelial like cell (Scale bar, \(100 \mu \mathrm{m}\) ). e Restoring L3MBTL3 rescued the mesenchymal morphology of L3MBTL3 knockout MDA-MB-231 and SUM159PT cells (Scale bar, \(100 \mu \mathrm{m}\) ). f Over-expressing L3MBTL3 increased the migration ability of MCF7 cells. g Knocking out L3MBTL3 decreased the migration and invasion ability of MDA-MB-231 and SUM159PT cells. h Restoring L3MBTL3 rescued the migration and invasion ability of L3MBTL3 knockout MDA-MB-231 and SUM159PT cells. i Western blot showed the knocking down efficiency of L3MBTL3 in 4T1 mouse breast cancer cell. Antibody GAPDH was used as a loading control. j The L3MBTL3 knockdown 4T1 cells were injection to mice to induce lung metastasis. Photographs showed the lung metastasis focus (Scale bar, \(5 \mathrm{mm}\) ). The numbers of metastatic nodules in the lungs per mouse were quantified (mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 9\) ). For f, g and h panels, bars indicate the mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(* \mathrm{p} < 0.05\) , \(** \mathrm{p} < 0.01\) , \(*** \mathrm{p} < 0.005\) , \(**** \mathrm{p} < 0.001\) . + +<--- Page Split ---> + +
Fig. 3
+ +![PLACEHOLDER_40_0] + + +<--- Page Split ---> + +Fig. 3 L3MBTL3 interacts with STAT3 and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. a Differences in pathway activities scored by GSEA between control (C) and L3MBTL3 knockout (L3- KO4 and L3- KO5) group in SUM159PT cells. Normalized enrichment score (NES)>1.4, p<0.05. b In L3MBTL3 rescued MDA- MB- 231 cells, immunoprecipitation of Flag- L3MBTL3 interacting protein was visualized by silver gel staining. c Network analysis (Cytoscape v 3.9.1) of the significant nuclear proteins involving in epithelial- mesenchymal transition process (score>500) bound to L3MBTL3 identified through immunoprecipitation and mass spectrometry proteomics. L3MBTL3 was highlighted by red. The transcription factors were highlighted by orange. The other nuclear proteins were highlighted by blue. d Immunofluorescence showed the L3MBTL3 and STAT3 co- localized in nucleus in L3MBTL3 rescue MDA- MB- 231 cells. Nuclei were stained with DAPI (Scale bar, 20 μm). e Co- immunoprecipitation showed that endogenous L3MBTL3 interacted with endogenous STAT3 and p- STAT3Y705 in MDA- MB- 231 and SUM159PT cells. f Co- immunoprecipitation showed that exogenous Flag- L3MBTL3 interacted with exogenous HA- STAT3 in 293T cells. g Co- immunoprecipitation showed that exogenous Flag- L3MBTL3 interacted with endogenous STAT3 and p- STAT3Y705 in L3MBTL3 rescue MAD- MB- 231 cells. Antibody β- ACTIN was used as loading control. The asterisk showed the unspecific band. h GST pull- down showed that, comparing to wildtype STAT3, STAT3 C (constantly active form) had higher L3MBTL3 binding ability, but STAT3Y705F (deactive form) had lower L3MBTL3 binding ability. The asterisk showed the unspecific band. i Co- immunoprecipitation showed that, comparing to wildtype STAT3, exogenous Flag- L3MBTL3 had higher binding ability on HA- STAT3Y705E (mimic p- STAT3 Y705, active form), but had lower binding ability on HA- STAT3Y705F (deactive form). j Co- immunoprecipitation showed + +<--- Page Split ---> + +858 that the binding ability of L3MBTL3 to STAT3 or p- STAT3Y705 increased after L3MBTL3 rescue MAD- MB- 231 cells treated with IL6. Antibody \(\beta\) - ACTIN was used as loading control. + +<--- Page Split ---> +![PLACEHOLDER_43_0] + + +<--- Page Split ---> + +Fig. 4 L3MBTL3 up-regulated epithelial-mesenchymal transition and IL6-JAK- STAT3 pathway. a GSEA plots depicting the enrichment of epithelial-mesenchymal transition genes down-regulated in L3MBTL3 knockout SUM159PT cells and up-regulated in L3MBTL3 over-expression MCF7 cells and L3MBTL3 rescued SUM159PT cells. p<0.05. b GSEA plots depicting the enrichment of IL6-JAK-STAT3 pathway genes down-regulated in L3MBTL3 knockout SUM159PT cells. p<0.05. c Heatmap of GSVA enrichment scores of epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway in L3MBTL3 high and low expression tumors. The gene expression datasets were form TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203, n=3207). d Scatter plots showing the positive correlation of transcript expression between L3MBTL3 and epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway in TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203, n=3207.). p<0.0001. + +<--- Page Split ---> + +
Fig. 5
+ +![PLACEHOLDER_45_0] + + +<--- Page Split ---> + +Fig. 5 L3MBTL3 up-regulated transcription of SNAIL in breast cancer. a ChIP-seq profiles (up) and heat maps of ChIP-Seq signal intensity (down) of STAT3 and L3MBTL3 form - 2.5 kb before transcription start site (TSS) to 2.5 kb after transcription end site (TES) in SUM159PT cells. b Venn diagram of the overlapped L3MBTL3 and STAT3 occupied genes detected by ChIP-seq in SUM159PT cell and different expression genes detected by RNA-seq in L3MBTL3 rescue SUM159PT cells. c Heatmaps of transcription changes of epithelial-mesenchymal transition up-regulated genes in control (C+V), L3MBTL3 knockout (L3-KO5+V) and rescue (L3-KO5+L3) SUM159PT cells. d Q-PCR analyzed showed that over-expressing L3MBTL3 in MCF7 cells up-regulated the transcription of SNAIL and decreased transcription of SNAIL target genes E-cad, OCLN, and DSP. e WB analyzed showed that over-expressing L3MBTL3 in MCF7 cells increased SNAIL expression and decreased SNAIL target genes E-cad expression. Antibody GAPDH was used as a loading control. f Immunofluorescence showed the E-cad expression decreased in L3MBTL3 overexpression MCF7 cells. Nuclei were stained with DAPI (Scale bar, 20 μm). g, h Q-PCR analyzed showed that (g) knocking out L3MBTL3 in MDA-MB-231 and SUM159PT cells down-regulated the transcription of SNAIL and up-regulated the transcription of SNAIL target genes OCLN, DSP and CLDN9, and (h) restoring L3MBTL3 in L3MBTL3 knockout MDA-MB-231 and SUM159PT cells up-regulated the transcription of SNAIL and down-regulated the transcription of SNAIL target genes OCLN, DSP and CLDN9. i, j WB analyzed showed that (i) knocking out L3MBTL3 in MDA-MB-231 and SUM159PT cells decreased SNAIL expression and (j) restoring L3MBTL3 in L3MBTL3 knockout MDA-MB-231 and SUM159PT cells increased SNAIL expression. Antibody \(\beta\) - ACTIN was used as a loading control. For (d), (g) and (h) panels, bars indicate the mean \(\pm \mathrm{SD}\) , n = 3. All the data were analyzed using Student’s + +<--- Page Split ---> + +906 t test. *p < 0.05, **p < 0.01, ***p < 0.005, ****p < 0.001. + +<--- Page Split ---> +![PLACEHOLDER_48_0] + + +<--- Page Split ---> + +Fig. 6 L3MBTL3 and SNAIL expression are positive correlation and are associated with low distant metastasis survival. a Scatter plots showing the positive correlation between L3MBTL3 and SNAIL transcription level in TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203 (n=3207)). p<0.05. b Distant metastasis survival in breast cancer dataset (GSE12276, n=202) was analyzed according to L3MBLT3 and SNAIL transcription level (L3MBTL3 low SNAIL low, n=49; L3MBTL3 low SNAIL high, n=51; L3MBTL3 high SNAIL low, n=51; L3MBTL3 high SNAIL high, n=53) The Distant metastasis survival of L3MBTL3 high SNAIL high group (n=53) was much lower than L3MBTL3 low SNAIL low group (n=49). p = 0.0035. c Human breast cancer tissue consisting of primary breast cancer samples (Tumor) and paired breast cancer lymph node metastasis samples (Metastasis) were subjected to Immunohistochemistry staining for L3MBLT3 and SNAIL. Immunohistochemistry showed that L3MBTL3 and SNAIL expression were positively correlation in breast cancer tumors (scale bar, 50 μm). + +<--- Page Split ---> + +
Fig. 7
+ +![PLACEHOLDER_50_0] + + +<--- Page Split ---> + +Fig. 7 L3MBTL3 recruits active STAT3 on SNAIL to promote transcription. a Schematic presentation L3MBTL3 and STAT3 binding sites on SNAIL gene promoter. Region P1 (-3704, -3518), P2 (-2185, -2114), P3 (-563, -474), P4 (+2543, +2679) and P5 (+3093, +3240) showed the location of the specific amplicons analyzed by the ChIP assay. b ChIP analyses using a L3MBTL3 antibody or a control IgG were performed in SUM159PT cells. Primers specific for region P1, P2, P3, P4 and P5 on SNAIL gene promoter were used. Data derived from Q-PCR analysis showed region P3 was the key L3MBTL3 binding sites on SNAIL promoter. c ChIP assay showed that both L3MBTL3 and STAT3 binding on region P3 of SNAIL promoter. Knocking out L3MBTL3 decreased the recruitment of p-STAT3Y705 on region P3 of SNAIL promoter in SUM159PT cells. d ChIP assay showed that both L3MBTL3 and STAT3 binding on region P3 of SNAIL promoter. Knocking out L3MBTL3 decreased the recruitment of p-STAT3Y705 on region P3 of SNAIL promoter, and restoring L3MBTL3 expression increased recruitment of p-STAT3Y705 on region P3 of SNAIL promoter in MDA-MB-231 cells. e 293T cells were co-transfected with pGL3-SNAIL (-600, +80) reporter plasmid, STAT3 and L3MBTL3 expression plasmids. Western blot showed L3MBTL3 expression. Antibody \(\beta\) -ACTIN was used as a loading control. Luciferase reporter gene assay showed L3MBTL3 over-expression increased the transcription activity of SNAIL promoter. f 293T cells were cotransfected with pGL3-SNAIL (-600, +80) reporter, STAT3 and L3MBTL3 expression plasmid with treat or untreat STAT3 inhibitor NIF. Western blot showed the L3MBTL3, STAT3 and p-STAT3Y705 level. Antibody \(\beta\) -ACTIN was used as a loading control. Luciferase reporter gene assay showed NIF significantly inhibited the transcription activity increase of SNAIL promoter induced by L3MBTL3. For b-f panels, bars indicate the mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(^{*}\mathrm{p}< 0.05\) , \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.001\) . + +<--- Page Split ---> + +
Fig. 8
+ +![PLACEHOLDER_52_0] + + +<--- Page Split ---> + +Fig. 8. L3MBTL3 promotes epithelial-mesenchymal transition, migration and invasion dependent on STAT3 in breast cancer. a, b Western blot showed knocking down STAT3 in (a) L3MBTL3 over-expressing MCF7 cells and (b) L3MBTL3 rescue SUM159PT cells decreased SNAIL expression. c, d Knocking down STAT3 in (c) L3MBTL3 over-expressing MCF7 cells and (d) L3MBTL3 rescue SUM159PT cells inhibited cell migration. e Knocking down STAT3 in L3MBTL3 rescue SUM159PT cells inhibited cell invasion. f, g Western blot showed inhibiting STAT3 activity in (f) L3MBTL3 over-expressing MCF7 cells and (g) L3MBTL3 rescue SUM159PT cells by NIF decreased SNAIL expression. h, i Inhibiting STAT3 activity in (h) L3MBTL3 over-expressing MCF7 cells and (i) L3MBTL3 rescue SUM159PT cells by NIF suppressed cell migration. j Inhibiting STAT3 activity in L3MBTL3 rescue SUM159PT cells by NIF suppressed cell invasion. For c, d, e, h, i and j panels, bars indicate the mean±SD, \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.001\) , ns: no significance. + +<--- Page Split ---> + +
Fig. 9
+ +![PLACEHOLDER_54_0] + + +<--- Page Split ---> + +Fig. 9. Region 176- 193aa is the key site of L3MBTL3 interacting with active STAT3. a GST pull- down assay showed the recombinant L3MBTL3 truncated protein GST- L3- 1 (1- 204aa) directly interacted with p- STAT3Y705. b Co- immunoprecipitation showed that L3MBTL3 interacted with p- STAT3Y705 required L3MBTL3 fragment (1- 204aa). c GST pull- down assay showed that the region 105- 204aa of L3MBTL3 directly interacted with p- STAT3Y705. d GST pull- down assay showed that the region 176- 193aa was the key site of L3MBTL3 interacting with p- STAT3Y705. + +<--- Page Split ---> + +
Fig. 10
+ +![PLACEHOLDER_56_0] + + + +![PLACEHOLDER_56_1] + + +<--- Page Split ---> + +Fig. 10. The interaction between L3MBTL3 and active STAT3 is required for SNAIL expression, cell migration and invasion in breast cancer. A Co- immunoprecipitation showed that deleting region 176- 193aa of L3MBTL3 decreased the interaction between exogenous Flag- L3MBTL3 and exogenous HA- STAT3 in 293T cells. b Co- immunoprecipitation showed that deleting region 176- 193aa of L3MBTL3 decreased the interaction between exogenous Flag- L3MBTL3 and endogenous p- STAT3Y705 in 293T cells. c, d (c) Q- PCR and (d) Western blot showed the transcription and protein level of L3MBTL3 and SNAIL in MCF7 cells and L3MBTL3 knockout SUM159PT cells transfected with vector (V), L3MBTL3 (L3) and L3MBTL3 deletion (L3- Δ176- 193) plasmids. SNAIL expression decreased after deleting region 176- 193aa of L3MBTL3. e Comparing to L3MBTL wild- type, deleting region 176- 193aa of L3MBTL3 decreased the ability of L3MBTL3 promoting cell migration in MCF7 cells and L3MBTL3 knockout SUM159PT cells, and also decreased the ability of L3MBTL3 promoting cell invasion in L3MBTL3 knockout SUM159PT cells. f A schematic of how L3MBTL3 and STAT3 collaboratively up- regulate SNAIL transcription to promote metastasis in breast cancer. For c and e panels, bars indicate the mean±SD, n = 3. All the data were analyzed using Student's t test. \*p < 0.05, \*\*\*p < 0.005, \*\*\*\*p < 0.001. + +<--- Page Split ---> + +1000 SUPPLEMENTARY DATA + +1001 Supplementary Data are available at NC Online. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- supplementaryinformation217.pdf- nrreportingsummary32flattened.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46_det.mmd b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..bb47e7cbf11bf7dae40393cdb68ee77549d3f45f --- /dev/null +++ b/preprint/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46/preprint__11fa165dcda69a9e111310361e67fb0c209ae0e0a7d2707e1af781fffb988a46_det.mmd @@ -0,0 +1,781 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 941, 207]]<|/det|> +# L3MBTL3 and active STAT3 collaboratively up-regulate SNAIL transcription to promote metastasis in breast cancer + +<|ref|>text<|/ref|><|det|>[[44, 230, 634, 273]]<|/det|> +Jianpeng Xiao The Third School of Clinical Medicine, Southern Medical University + +<|ref|>text<|/ref|><|det|>[[44, 278, 633, 320]]<|/det|> +Jie Wang The Third School of Clinical Medicine, Southern Medical University + +<|ref|>text<|/ref|><|det|>[[44, 325, 312, 366]]<|/det|> +Jialun Li East China Normal University + +<|ref|>text<|/ref|><|det|>[[44, 371, 741, 413]]<|/det|> +Jie Xiao School of Laboratory Medicine and Biotechnology, Southern Medical University + +<|ref|>text<|/ref|><|det|>[[44, 418, 545, 460]]<|/det|> +Cuicui Liu Southern Medical University Affiliated Fengxian Hospital + +<|ref|>text<|/ref|><|det|>[[44, 465, 545, 506]]<|/det|> +Libi Tan Southern Medical University Affiliated Fengxian Hospital + +<|ref|>text<|/ref|><|det|>[[44, 511, 653, 552]]<|/det|> +Yanhong Tu The Second Affiliated Hospital, The Chinese University of Hong Kong + +<|ref|>text<|/ref|><|det|>[[44, 557, 680, 599]]<|/det|> +Ruifang Yang Anhui University of Science and Technology Affiliated Fengxian Hospital + +<|ref|>text<|/ref|><|det|>[[44, 604, 680, 645]]<|/det|> +Yujie Pei Anhui University of Science and Technology Affiliated Fengxian Hospital + +<|ref|>text<|/ref|><|det|>[[44, 650, 653, 691]]<|/det|> +Minghua Wang The Second Affiliated Hospital, The Chinese University of Hong Kong + +<|ref|>text<|/ref|><|det|>[[44, 696, 670, 737]]<|/det|> +Jiemin Wong East China Normal University https://orcid.org/0000- 0002- 5311- 842X + +<|ref|>text<|/ref|><|det|>[[44, 742, 787, 783]]<|/det|> +Binhua Zhou University of Kentucky College of Medicine https://orcid.org/0000- 0002- 0983- 2060 + +<|ref|>text<|/ref|><|det|>[[44, 788, 545, 829]]<|/det|> +Jing Li Southern Medical University Affiliated Fengxian Hospital + +<|ref|>text<|/ref|><|det|>[[44, 834, 133, 853]]<|/det|> +Jing Feng + +<|ref|>text<|/ref|><|det|>[[54, 860, 290, 877]]<|/det|> +feng.jing71921@163. com + +<|ref|>text<|/ref|><|det|>[[52, 905, 631, 924]]<|/det|> +The Third School of Clinical Medicine, Southern Medical University + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 46, 103, 63]]<|/det|> +## Article + +<|ref|>sub_title<|/ref|><|det|>[[44, 84, 135, 102]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 121, 315, 140]]<|/det|> +Posted Date: March 15th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 159, 474, 179]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2601300/v1 + +<|ref|>text<|/ref|><|det|>[[42, 197, 916, 240]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 257, 534, 277]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 312, 936, 355]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55617- 9. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[100, 92, 850, 154]]<|/det|> +L3MBTL3 and active STAT3 collaboratively up-regulate SNAIL transcription to promote metastasis in breast cancer + +<|ref|>sub_title<|/ref|><|det|>[[147, 189, 234, 206]]<|/det|> +## Authors + +<|ref|>text<|/ref|><|det|>[[144, 213, 850, 300]]<|/det|> +Jianpeng Xiao \(^{1,2,3\#}\) , Jie Wang \(^{1,2\#}\) , Jialun Li \(^{4}\) , Jie Xiao \(^{5}\) , CuiCui Liu \(^{2}\) , Libi Tan \(^{2,5}\) , Yanhong Tu \(^{3}\) , Ruifang Yang \(^{6}\) , Yujie Pei \(^{6}\) , Minghua Wang \(^{3}\) , Jiemin Wong \(^{4}\) , Binhua P Zhou \(^{7}\) , Jing Li \(^{2,5*}\) , Jing Feng \(^{1,2,3,5*}\) + +<|ref|>text<|/ref|><|det|>[[144, 340, 850, 390]]<|/det|> +\(^{1}\) The Third School of Clinical Medicine, Southern Medical University, Guangzhou, 510630, China. + +<|ref|>text<|/ref|><|det|>[[144, 405, 850, 456]]<|/det|> +\(^{2}\) Department of Laboratory Medicine & Central Laboratory, Southern Medical University Affiliated Fengxian Hospital, Shanghai, 201499, China. + +<|ref|>text<|/ref|><|det|>[[144, 470, 850, 521]]<|/det|> +\(^{3}\) The Second Affiliated Hospital, The Chinese University of Hong Kong, Shenzhen, 518172, China. + +<|ref|>text<|/ref|><|det|>[[144, 535, 850, 585]]<|/det|> +\(^{4}\) Shanghai Key Laboratory of Regulatory Biology, Institute of Biomedical Sciences and School of Life Sciences, East China Normal University, Shanghai, 200241 China. + +<|ref|>text<|/ref|><|det|>[[144, 600, 850, 650]]<|/det|> +\(^{5}\) School of Laboratory Medicine and Biotechnology, Southern Medical University, Guangzhou, 510515, China. + +<|ref|>text<|/ref|><|det|>[[144, 665, 850, 715]]<|/det|> +\(^{6}\) Anhui University of Science and Technology Affiliated Fengxian Hospital, Shanghai, 201499, China. + +<|ref|>text<|/ref|><|det|>[[144, 730, 850, 781]]<|/det|> +\(^{7}\) Department of Molecular and Cellular Biochemistry, Markey Cancer Center, University of Kentucky College of Medicine, Lexington, 40536, USA. + +<|ref|>text<|/ref|><|det|>[[144, 796, 650, 815]]<|/det|> +#These authors contributed equally: Jianpeng Xiao, Jie Wang + +<|ref|>text<|/ref|><|det|>[[144, 830, 850, 880]]<|/det|> +These authors jointly supervised this work: Jing Feng, Jing Li. Email: fengjing71921@163. com; lijing228nanfang@163. com. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 86, 240, 104]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[144, 120, 852, 600]]<|/det|> +Active STAT3 pathway promotes epithelial mesenchymal transition, migration, invasion and metastasis in cancer. STAT3 upregulates the transcription of epithelial mesenchymal transition key transcription factor SNAIL through DNA- binding independent way. However, the mechanism that how is STAT3 recruited to SNAIL promoter to upregulate transcription in DNA independent way is still unclear. In our study, we found a lysine methylated binding protein L3MBTL3 was positively associated with metastatic and poor prognosis in breast cancer. L3MBTL3 also promoted epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. Mechanism analysis found L3MBTL3 interacted with active STAT3 and recruited active STAT3 on SNAIL promoter to up- regulated SNAIL transcription. The interaction between L3MBTL3 and active STAT3 was required for SNAIL transcription up- regulation, migration and invasion in breast cancer, while the methylated lysine binding activity of L3MBTL3 is not required for these functions. In conclusion, L3MBTL3 and active STAT3 collaboratively up- regulate SNAIL transcription to promote breast cancer metastasis. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 85, 280, 104]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[145, 121, 852, 300]]<|/det|> +Breast cancer, developing breast lobules, tubes or connective tissue it the most common malignancy in women'. Metastasis is the main cause of breast cancer high mortality'. It is a convoluted and multi- step process: the tumor cell disseminates from primary tumor sites, migrating to the target organs, then develops into a distal secondary tumor'. Epithelial- mesenchymal transition, a biological process switches immobile epithelial cells into mobile mesenchymal cells, is one of the crucial steps in initiating metastasis'. + +<|ref|>text<|/ref|><|det|>[[144, 317, 852, 664]]<|/det|> +There are many key transcription factors involved in the epithelial- mesenchymal transition process. SNAIL (Snail family transcriptional repressor 1), as a direct transcription repressor of epithelial morphology promoter \(E\) - cad (cadherin1), drives epithelial- mesenchymal transition process and induces migratory and invasive behavior in cancer cells and non- cancer cells3- 5. Enhancing SNAIL gene expression in primary tumors promotes cellular motility and the consequent acquisition of metastatic properties. In contrast, targeted SNAIL or down- regulated the transcription of SNAIL gene can reverse epithelial- mesenchymal transition process6- 8. The transcription of SNAIL is regulated by many signaling pathway, including \(\mathrm{FGF}^9\) , \(\mathrm{Wnt^{10}}\) , \(\mathrm{TGF}\beta^{11}\) and STAT312 signaling pathway, which collectively form the microenvironment of epithelial- mesenchymal transition and metastasis13. + +<|ref|>text<|/ref|><|det|>[[144, 679, 852, 894]]<|/det|> +STAT3 (Signal transducer and activator of transcription 3), as a member of STAT family, is involved in various basic cell functions14,15. Aberrant accumulation of phosphorylated STAT3 is observed in nearly \(70\%\) of cancers including breast cancer, and phosphorylated STAT3 is involved in metastasis leading the poor clinical outcome in patients16. It is reported that STAT3 plays crucial roles in epithelial- mesenchymal transition, migration, invasion and metastasis. In keratinocytes and ovarian adenocarcinoma, STAT3 is required for cell motility, silencing STAT3 can + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 78, 852, 760]]<|/det|> +reverse the phenotype of cell motility17,18. Persistent activation STAT3 enhances cell migration of prostate epithelial cells19. High phosphorylated STAT3 level correlates with increased invasion and metastasis in cutaneous squamous cell carcinoma20. In contrast, targeting STAT3 blocks invasion and metastasis formation in variety of cell lines both in vitro and in vivo21,22. IL6 (Cytokine interleukin 6) activates STAT3 signal pathway to induce epithelial-mesenchymal transition and promote metastasis23- 27. In IL6- STAT3 signal pathway, IL6 triggers STAT3 phosphorylation at its Y705 site. The phosphorylated STAT3 forms dimer and localizes into the nucleus, then binds to the promoter to up- regulate the transcription of target genes28,29. STAT3 activation mediated by IL- 6 can upregulate epithelial- mesenchymal transition- inducing transcription factors such as SNAIL, ZEB1 (zinc finger E- box binding homebox 1), TWIST1 (twist family bHLH transcription factor 1) and JUNB (JunB proto- oncogene), and then induces epithelial- mesenchymal transition to enhance metastasis23- 26. STAT3 also can further enhanced SNAIL transcription under TGF- \(\beta\) induce condition12. Knocking down STAT3 or inhibiting STAT3 activity significantly reduces the induction of SNAIL by TGF- \(\beta^{12}\) . STAT3 often recognizes DNA motifs of target gene promoter by its DNA binding domain and up- regulates target gene transcription28,29. However, on SNAIL promoter, neither the DNA- binding activity of STAT3 nor canonical STAT3 DNA binding motif of promoter is required for SNAIL transcriptional upregulation12, suggesting that there is still DNA- binding independent transcriptional regulation mechanism. + +<|ref|>text<|/ref|><|det|>[[145, 772, 851, 890]]<|/det|> +L3MBTL3 (L3MBTL histone methyl- lysine binding protein 3) is a member of the malignant brain tumor (MBT) family of chromatin- interacting transcriptional repressors, which binds histone lysine methylation H4K20me230. As a methyl- binding protein, L3MBTL3 can also bind the lysine methylation of non- histone protein through + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 81, 852, 530]]<|/det|> +its MBT domain \(^{31,32}\) and recruit ubiquitin E3 ligase CRL4 \(^{DCAF5}\) to mediate the methylation- dependent degradation of substrate proteins (such as SOX2 (SRY- box transcription factor 2), E2F1 (E2F transcription factor 1), DNMT1 (DNA methyltransferase1), SMARCC1 and SMARCC2 (SWI/SNF related matrix associated actin dependent regulator of chromatin subfamily c member 1 and 2)) \(^{31 - 33}\) . In addition, L3MBTL3 interacts with transcription factor RBPJ (recombination signal binding protein for immunoglobulin kappa J region) to repress target gene transcription of Notch signaling \(^{34,35}\) . Knocking out L3MBTL3 in mice leads to immature myeloid progenitors accumulation, and hence decreases mature myeloid blood cells, causing embryonic lethal at the late stage \(^{36}\) . The GWAS (post- genome- wide association study) analysis shows L3MBTL3 is associated with multiple sclerosis \(^{37,38}\) . In cancer, there are a cancer risk loci of L3MBTL3 (rs9375701), which significantly associated with low L3MBTL3 expression in breast and prostate cancer \(^{39}\) . However, the functions of L3MBTL3 in breast cancer are still unclear. + +<|ref|>text<|/ref|><|det|>[[144, 541, 852, 856]]<|/det|> +In our study, we found L3MBTL3 was positively correlated with metastatic and poor prognosis in breast cancer and promoted epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. Protein interaction analysis revealed L3MBTL3 interacted with active STAT3. L3MBTL3 could recruit active STAT3 on SNAIL promoter to up- regulated SNAIL transcription. The interaction between L3MBTL3 and active STAT3 was required for SNAIL transcription up- regulation, migration and invasion in breast cancer, while the methylated lysine binding activity of L3MBTL3 is not required for these functions. Overall, these findings indicate that L3MBTL3 interacting with active STAT3 and up- regulates SNAIL transcription to promote breast cancer metastasis. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 85, 225, 103]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[147, 120, 850, 172]]<|/det|> +High expression of L3MBTL3 is associated with metastatic and poor prognosis in breast cancer. + +<|ref|>text<|/ref|><|det|>[[147, 185, 852, 633]]<|/det|> +In order to explore the function of L3MBTL3 in breast cancer, we first examined the expression of L3MBTL3 in 43 primary breast tumor samples (Tumor) and paired lymph node metastasis samples (Metastasis) and tumor- adjacent tissue samples (AdjN) by immunohistochemistry (IHC). L3MBTL3 protein levels significantly increased in lymph node metastasis of breast tumor comparing to primary breast tumor (Fig. 1a and Fig. 1b). We also noticed that L3MBTL3 highly expressed and localized in nuclear in mammary gland (Fig. 1a and Fig. 1b), while in lymph node metastasis samples L3MBTL3 highly expressed and localized in the nuclear and cytoplasm (Fig. 1a). We then analyzed datasets of Breast Cancer Integrative Platform (http://www.omicsnet.org/bcancer/database), which integrated gene expression and clinical data, and found that high L3MBTL3 expression group (n=103) showed lower distant metastasis free survival comparing to low L3MBTL3 expression group (n=99) (GSE12276 (n=202)) (Fig. 1c). Therefore, these data suggest that high L3MBTL3 expression is associated with breast cancer metastasis and poor prognosis. + +<|ref|>text<|/ref|><|det|>[[147, 680, 850, 730]]<|/det|> +L3MBTL3 promotes epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. + +<|ref|>text<|/ref|><|det|>[[147, 743, 852, 896]]<|/det|> +The above observations prompted us to investigate the function of L3MBTL3 in breast cancer metastasis. We examined the expression levels of L3MBTL3 in different breast cancer cell lines and found that L3MBTL3 expression positively correlated with key epithelial- mesenchymal transition transcription factor SNAIL, and negatively correlated with epithelial cell marker E- cad which is transcriptional down- regulated by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 80, 853, 730]]<|/det|> +SNAIL (Fig. 2a). We established L3MBTL3 over-expression stable cell line in MCF7 cells, L3MBTL3 knockout and rescue stable cell line in MDA- MB- 231 and SUM159PT cells (Fig. 2b), then observed the L3MBTL3 function in epithelial- mesenchymal transition, migration and invasion of breast cancer cells. We found over- expressing L3MBTL3 resulted in a spindle like mesenchymal phenotype and loss of cell contact in MCF7 cell (Fig. 2c). Conversely, knocking out L3MBTL3 in MDA- MB- 231 and SUM159PT cell showed epithelial like morphology (Fig. 2d). And restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 and SUM159PT cell rescued the mesenchymal morphology (Fig. 2e). We also observed that over- expression L3MBTL3 greatly promoted the migration of MCF7 cell (Supplementary Fig. 1a and Fig. 2f). In contrast, knocking out L3MBTL3 significantly inhibited migration and invasion in MDA- MB- 231 and SUM159PT cells (Supplementary Fig. 1b and Fig. 2g). And restoring L3MBTL3 expression significantly rescued the migration and invasion ability of MDA- MB- 231 and SUM159PT cells (Supplementary Fig. 1c and Fig. 2h). In addition, we established L3MBTL3 knockdown stable cell line in mouse breast cancer cell 4T1 (Fig. 2i) to verify the metastasis- promoting effect of L3MBTL3 in breast cancer using breast cancer lung metastasis mouse model. We found silencing L3MBTL3 significantly inhibited tumor lung metastasis in vivo, either (Fig. 2j). Thus, these results suggest that L3MBTL3 promotes epithelial- mesenchymal transition, migration, invasion and metastasis in breast cancer. + +<|ref|>sub_title<|/ref|><|det|>[[147, 772, 848, 821]]<|/det|> +## L3MBTL3 interacts with STAT3 and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. + +<|ref|>text<|/ref|><|det|>[[147, 836, 850, 888]]<|/det|> +To further explore the mechanism of L3MBTL3, we performed RNA- seq in L3MBTL3 knockout SUM159PT cells. GSEA analysis of hallmark pathway gene signatures + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 266]]<|/det|> +showed knocking out L3MBTL3 significantly down- regulated epithelial- mesenchymal transition pathway (Fig. 3a). We also performed immunoprecipitation in L3MBTL3 rescue MDA- MB- 231 cells and identified the interaction protein of L3MBTL3 through Mass Spectrum (Fig. 3b). According to previous study, L3MBTL3 binds histone lysine methylation30, and interacts with transcription factor RBPJ to repress target gene transcription of Notch signaling34,35. + +<|ref|>text<|/ref|><|det|>[[144, 280, 852, 898]]<|/det|> +The immunofluorescence data showed L3MBTL3 mainly localized in the nucleus (Fig. 3d), suggesting L3MBTL3 functions primarily in the nucleus. So we selected the L3MBTL3 interaction protein, which was involving in epithelial- mesenchymal transition process and localized in nuclear. We found 71 L3MBTL3 interaction protein, including transcription factor STAT3, TP53 and STAT1 (Fig. 3c and Supplementary Table 3). In addition, the GSEA analysis of hallmark pathway gene signatures showed knocking out L3MBTL3 significantly down- regulated IL6- JAK- STAT3 signaling pathway (Fig. 3a), which also promoted epithelial- mesenchymal transition and metastasis in cancer. The immunofluorescence data also showed L3MBTL3 and STAT3 co- localized in nuclear (Fig. 3d). These data suggest that L3MBTL3 may interact with STAT3 to regulate the transcription of target genes. Further immunoprecipitation showed L3MBTL3 endogenously interacted with STAT3 in MDA- MB- 231 and SUM159PT cells (Fig. 3e), and exogenously interacted with STAT3 in Flag- L3MBTL3 and HA- STAT3 over- expression 293T cells and L3MBTL3 rescue MDA- MB- 231 cells (Fig. 3f and Fig. 3g). Interestingly, we found L3MBTL3 showed much stronger binding ability with p- STAT3Y705 (active form of STAT3) than STAT3 both in endogenous (Fig. 3e) and exogenous (Fig. 3g). In order to further confirm the active STAT3 binding specificity of L3MBTL3, we expressed and purified GST- STAT3, GST- STAT3Y705F (STAT3 inactive form) and GST- STAT3 C (STAT3 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 595]]<|/det|> +constantly active form) by E. coli, and performed GST-pulldown assay. We observed that the interaction between L3MBTL3 and STAT3 was significantly increased after STAT3 was constantly activated (STAT3 C) and decreased after STAT3 was inactivated (STAT3Y705F) (Fig. 3h). In addition, we over-expressed L3MBTL3, HA- STAT3, HA- STAT3Y705F (STAT3 inactive form) and HA- STAT3Y705E (mimic p- STAT3Y705, STAT3 active form) in 293T cells and performed immunoprecipitation. We found the interaction between L3MBTL3 and STAT3 was significantly decreased after STAT3 was inactivated (STAT3Y705F) but greatly increased after STAT3 was activated (STAT3Y705E) (Fig. 3i). As we know, IL6 induces STAT3 phosphorylation and then activates STAT3 signaling pathway40. We treated L3MBTL3 rescue MDA- MB- 231 cells with IL6, then examined the interaction between L3MBTL3 and STAT3. The immunoprecipitation showed both the STAT3 and p- STAT3Y705 binding ability of L3MBTL3 increased after IL6 treatment (Fig. 3j), suggesting the IL6 activated STAT3 to enhance the interaction between L3MBTL3 and STAT3. Together, these results indicates that L3MBTL3 interacts with STAT3, and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. + +<|ref|>sub_title<|/ref|><|det|>[[147, 641, 850, 691]]<|/det|> +## L3MBTL3 up-regulates epithelial-mesenchymal transition and IL6-JAK-STAT3 pathway in breast cancer. + +<|ref|>text<|/ref|><|det|>[[145, 704, 852, 890]]<|/det|> +To further confirm the regulation function of L3MBTL3 in epithelial- mesenchymal transition and IL6- JAK- STAT3 signaling pathway, we also performed RNA- seq in L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells. GSEA analysis of hallmark pathway gene signatures showed knocking out L3MBTL3 revealed robust disruptions of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes in SUM159PT cells (Fig. 4a and Fig. 4b), but over- expressing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 853, 595]]<|/det|> +L3MBTL3 in MCF7 cells or restoring L3MBTL3 in L3MBTL3 knockout SUM159PT cells significantly up- regulated the transcription of epithelial- mesenchymal transition signature target genes (Fig. 4a). We further used the TCGA datasets (TCGA BRCA 2000 and TCGA BRCA 2017) and GEO datasets (GSE202203) to perform the GSVA analysis according to L3MBTL3 expression level. Consist with the GSEA analysis results in breast cancer cells, the expression of epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway genes were elevated more in L3MBTL3 high tumors than in L3MBTL3 low tumors (Fig. 4c). In support of the notion that L3MBTL3 is a candidate driver of epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in breast cancer, we further analyzed the relationship between L3MBTL3 expression and epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in TCGA breast cancer datasets (TCGA BRCA 2000 and TCGA BRCA 2017) and GEO datasets (GSE202203). The analysis results showed that L3MBTL3 expression positively association with epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway (Fig. 4d). Thus, these data suggest L3MBTL3 up- regulates epithelial- mesenchymal transition and IL6- JAK- STAT3 pathway in breast cancer. + +<|ref|>sub_title<|/ref|><|det|>[[147, 641, 682, 660]]<|/det|> +## L3MBTL3 promotes transcription of SNAIL in breast cancer. + +<|ref|>text<|/ref|><|det|>[[144, 671, 853, 890]]<|/det|> +We performed ChIP- seq to further analyze the transcription regulation mechanism of L3MBTL3 in SUM159PT cells. ChIP- seq analysis showed L3MBTL3 mainly bound on the promoter (77.4% of L3MBTL3 binding sites), specially near the transcription start site (promoter (<=1kb), 67.32% of L3MBTL3 binding sites) (Supplementary Fig. 2a). The L3MBTL3 binding peaks showed similar pattern with STAT341 in genome- wide (Fig. 5a). Further analysis showed that there were 4878 genes both occupied by STAT3 and L3MBTL3 on promoter in SUM159PT cells (Fig. 5b). We then overlap + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 78, 852, 831]]<|/det|> +these genes with 932 different expression genes in L3MBTL3 rescue SUM159PT cells, and found 267 different expression genes (Fig. 5b), which suggested L3MBTL3 and STAT3 could collaboratively regulate transcription of target genes. After analyzing the 267 different expression genes, we found the transcription of epithelial-mesenchymal transition up-regulated genes were decreased after knocking out L3MBTL3 in SUM159PT cells, but increased after restoring L3MBTL3 expression in L3MBTL3 knockout SUM159PT cells (Fig. 5c). In these genes, we identified the key epithelial-mesenchymal transition transcription factor- SNAIL, which often inhibited its target gene transcription in epithelial-mesenchymal transition process \(^{3 - 5}\) . There were a significant increase trend of SNAIL transcription and a decrease trend of SNAIL target genes E- cad, OCLN and DSP transcription after L3MBTL3 over-expressing in MCF7 cells (Fig. 5d). Western blot showed that the protein level of SNAIL greatly increased and E- cad greatly decreased in L3MBTL3 over-expression MCF7 cells (Fig. 5e). Immunofluorescence staining also showed that over-expressing L3MBTL3 significantly decreased E- cad expression in MCF7 cells (Fig. 5f). We also observed that there were a significant decrease trend of SNAIL transcription and a increase trend of SNAIL target genes OCLN, DSP and CLDN9 transcription after knocking out L3MBTL3 in MDA- MB- 231 and SUM159PT cells (Fig. 5g). Western blot also showed that knocking out L3MBTL3 significantly decreased SNAIL protein level in MDA- MB- 231 and SUM159PT cells (Fig. 5i). Restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 and SUM159PT cells recused SNAIL transcription (Fig. 5h) and protein level (Fig. 5j), and inhibited the transcription of SNAIL target genes OCLN, DSP and CLDN9 (Fig. 5h). + +<|ref|>text<|/ref|><|det|>[[147, 837, 850, 889]]<|/det|> +We further analyzed the expression association between L3MBTL3 and SNAIL in TCGA datasets (TCGA BRCA 2000 (n=1097) and TCGA BRCA 2017 (n=1218)) and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 430]]<|/det|> +GEO datasets (GSE202203 (n=3207)). The L3MBTL3 expression was positive association with the SNAIL expression in breast cancer tumors (Fig. 6a). We also analyzed the relationship between the co-expression level of L3MBTL3 and SNAIL and the distant metastasis free survival using the GEO dataset (GSE12276, n=202) and found that breast cancer patients with high L3MBTL3 and high SNAIL expression group (red line, n=53) showed lower distant metastasis free survival comparing with low L3MBTL3 and low SNAIL expression group (blue line, n=49) (Fig. 6b). The immunohistochemistry analysis also showed that L3MBTL3 expression positively correlated with SNAIL expression in primary tumor (Tumor) and lymph node metastasis (Metastasis) of breast cancer samples (Fig. 6c). All these data indicated that L3MBTL3 up-regulated SNAIL transcription in breast cancer. + +<|ref|>sub_title<|/ref|><|det|>[[147, 477, 849, 529]]<|/det|> +## L3MBTL3 recruits active STAT3 on SNAIL promoter to up-regulate SNAIL transcription. + +<|ref|>text<|/ref|><|det|>[[144, 540, 852, 890]]<|/det|> +To further study the L3MBTL3 regulation mechanism on SNAIL gene in breast cancer, we analyzed L3MBTL3 and STAT341 ChIP-seq results and found that L3MBTL3 and STAT3 co- localized on SNAIL promoter (Fig. 7a). The ChIP assay showed region P3 (- 563, - 474) was the key binding site of L3MBTL3 on SNAIL promoter (Fig. 7b). The further ChIP assay showed L3MBTL3 and active STAT3 (p- STAT3Y705) co- localized on the region P3 of SNAIL promoter in MDA- MB- 231 (Fig. 7d) and SUM159PT (Fig. 7e) cells. Knocking out L3MBTL3 significantly reduced the binding ability of p- STAT3Y705 on region P3 of SNAIL promoter in MDA- MB- 231 (Fig. 7d) and SUM159PT (Fig. 7e) cells. And restoring L3MBTL3 expression in L3MBTL3 knockout MDA- MB- 231 cells significantly increased the binding ability of p- STAT3Y705 on region P3 of SNAIL promoter (Fig. 7d). We also cloning the SNAIL + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 300]]<|/det|> +promoter region (- 600, +80) (including region P3) in pGL3 plasmid and performed Luciferase assay. The results showed that over- expressing L3MBTL3 significantly promoted the SNAIL promoter region (- 600, +80) triggered Luciferase reporter gene transcription (Fig. 7e and Fig. 7f). However, STAT3 inhibitor NIF could totally depleted the transcription up- regulation mediated by L3MBTL3 (Fig. 7f). Thus, all these results indicated L3MBTL3 recruited active STAT3 on SNAIL promoter to up- regulated SNAIL transcription in breast cancer. + +<|ref|>text<|/ref|><|det|>[[144, 313, 852, 660]]<|/det|> +Considering STAT3 often binds target gene through the STAT3 DNA binding motif on target gene promoter \(^{28,29}\) , we performed motif analysis of L3MBTL3 binding sites and found that there were some sites enriched for STAT3 DNA binding motif (Supplementary Fig. 3a). However, on SNAIL promoter, we observed that the L3MBTL3- STAT3 co- localization sites are including neither the classical STAT3 DNA binding motif (TTC(N3)GAA) nor other reported STAT3 DNA binding motif (such as AGG(N3)AGG) \(^{42}\) . These results are consistent with the report that the DNA- binding activity of STAT3 and canonical STAT3 DNA binding motif on the SNAIL promoter are not required for SNAIL transcription up- regulation \(^{12}\) . Our results can help to explain how STAT3 up- regulated SNAIL transcription through a DNA- binding independent way. + +<|ref|>sub_title<|/ref|><|det|>[[147, 707, 849, 756]]<|/det|> +## L3MBTL3 up-regulates SNAIL expression and promotes migration and invasion dependent on STAT3 in breast cancer. + +<|ref|>text<|/ref|><|det|>[[145, 771, 852, 890]]<|/det|> +To further evaluate the co- transcriptional regulation function of L3MBTL3 and STAT3 in breast cancer, we used STAT3 siRNA and inhibitor in L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells. Knocking down STAT3 significantly reduced SNAIL expression (Fig. 8a and Fig. 8b). The migration assay + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 563]]<|/det|> +showed that knocking down STAT3 significantly reduced migration ability of L3MBTL3 over- expression MCF7 cells (Supplementary Fig. 4a and Fig. 8c) and L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4b and Fig. 8d). The invasion assay also showed that knocking down STAT3 greatly reduced invasion ability of L3MBTL3 rescue SUM159PT cell (Supplementary Fig. 4c and Fig. 8e). In L3MBTL3 over- expression MCF7 cells and L3MBTL3 rescue SUM159PT cells, inhibiting STAT3 activity by STAT3 inhibitor NIF also significantly reduced SNAIL expression (Fig. 8f and Fig. 8g). The migration assay showed that inhibiting STAT3 activity by NIF significantly reduced migration ability of L3MBTL3 over- expression MCF7 cells (Supplementary Fig. 4d and Fig. 8h) and L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4e and Fig. 8i). The invasion assay also showed that inhibiting STAT3 activity by NIF significantly reduced invasion ability of L3MBTL3 rescue SUM159PT cells (Supplementary Fig. 4f and Fig. 8j). Thus, all these results indicate L3MBTL3 up- regulates SNAIL expression and promotes migration and invasion dependent on STAT3 in breast cancer cell. + +<|ref|>sub_title<|/ref|><|det|>[[147, 608, 848, 658]]<|/det|> +## The interaction between L3MBTL3 and activated STAT3 is required for SNAIL transcription up-regulation, migration and invasion in breast cancer. + +<|ref|>text<|/ref|><|det|>[[147, 671, 852, 891]]<|/det|> +In order to study the characterization of the molecular interplay between STAT3 and L3MBTL3, a series of L3MBTL3 deletion mutants were employed to identify p- STAT3Y705 interacting domain of L3MBTL3. In GST- pulldown assay, we observed that L3- 1 (1- 204aa) was required for the L3MBTL3/p- STAT3Y705 interaction (Fig. 9a). Immunoprecipitation assay also showed that the L3MBTL3 wildtype (Flag- L3) and L3MBTL3 deletion mutants containing L3- 1 (Flag- L3- 5 and Flag- L3- 6) showed the p- STAT3Y705 binding ability, while the L3MBTL3 deletion mutants without L3- 1 (Flag + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 465]]<|/det|> +L3- 2 and Flag- L3- 4) lost p- STAT3Y705 binding ability (Fig. 9b). As we narrowed down our analysis to even shorter region, we identified 176- 193aa region was the key p- STAT3Y705 binding site of L3MBTL3 (Fig. 9c and Fig. 9d). Immunoprecipitation showed that deleting 176- 193aa region of L3MBTL3 significantly reduced the interaction between L3MBTL3 and STAT3 or p- STAT3Y705 in 293T (Fig. 10a and Fig. 10b). In order to study the function of L3MBTL3 176- 193aa region, we expressed L3MBTL3 wild- type and L3MBTL3 mutation L3- \(\Delta 176 - 193\) in MCF7 or L3MBTL3 knockout SUM159PT cells. The Q- PCR and WB showed that deleting 176- 193aa region of L3MBTL3 impeded the increase of SNAIL expression induced by L3MBTL3 (Fig. 10c and Fig. 10d). Comparing to L3MBTL3 wild- type, the migration and invasion ability was remarkably weakened after deleting 176- 193aa region of L3MBTL3 (Supplementary Fig. 5a, Supplementary Fig. 5b and Fig. 10e). + +<|ref|>text<|/ref|><|det|>[[145, 477, 852, 858]]<|/det|> +It was reported that L3MBTL3 as a lysine methylated reader protein, bound methylated lysine through its key methylated lysine binding site D381 to degrade substrate protein31,32. To examine whether L3MBTL3 promoted epithelial- mesenchymal transition, migration and invasion dependent on its lysine methylated reader ability, we expressed L3MBTL3 wild- type and L3MBTL3 methylated lysine binding mutation L3MBTL3- D381N in MCF7 or L3MBTL3 knockout SUM159PT cells. Immunoprecipitation showed that L3MBTL3 D381N mutation did not significantly change the interaction between L3MBTL3 and STAT3 or p- STAT3Y705 (Supplementary Fig. 6a and Supplementary Fig. 6b). Comparing to L3MBTL3 wild- type, L3MBTL3 D381N mutation did not significantly change the SNAIL expression (Supplementary Fig. 6c and Supplementary Fig. 6d) and migration and invasion ability (Supplementary Fig. 6e). + +<|ref|>text<|/ref|><|det|>[[186, 870, 850, 890]]<|/det|> +Thus, all these data indicated the interaction between L3MBTL3 and active + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 83, 850, 168]]<|/det|> +STAT3 is required for epithelial- mesenchymal transition, migration and invasion in breast cancer while the methylated lysine binding ability of L3MBTL3 is not required for these functions described above. + +<|ref|>text<|/ref|><|det|>[[147, 181, 849, 235]]<|/det|> +In conclusion, L3MBTL3 and active STAT3 collaboratively up- regulate SNAIL transcription to promote breast cancer metastasis (Fig. 10f). + +<|ref|>sub_title<|/ref|><|det|>[[148, 282, 257, 301]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[147, 317, 852, 700]]<|/det|> +As previously reported, there are a cancer risk loci of L3MBTL3 (rs9375701), which significantly associated with low L3MBTL3 expression in breast cancer39, suggesting L3MBTL3 may negatively associate with breast cancer. Consist with this report, we also observed the low expression of L3MBTL3 in primary breast tumor. However, our study showed that L3MBTL3 was positively correlated with metastatic and poor prognosis in breast cancer, suggesting an important role of L3MBTL3 in breast cancer metastasis. This speculation was confirmed by our observation that L3MBTL3 play a crucial role in promoting epithelial- mesenchymal transition, migration, invasion and breast tumor metastasis. Our results reveal that L3MBTL3 interacts with active STAT3 at shared binding sites on SNAIL promoter to up- regulate SNAIL transcription and promote epithelial- mesenchymal transition, migration and invasion. Therefore, we disclose a metastasis enhancer function of L3MBTL3 in breast cancer in this work. + +<|ref|>text<|/ref|><|det|>[[147, 712, 850, 894]]<|/det|> +L3MBTL3 has been known as a chromatin- interacting transcriptional repressor, interacting with RBPJ to repress target gene transcription of Notch signaling34. However, we described a transcriptional activation role for L3MBTL3 in breast cancer. Knocking out L3MBTL3 revealed robust disruptions of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes in breast cancer cells. In contrast, over- expressing L3MBTL3 or restoring L3MBTL3 in L3MBTL3 knockout breast + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 83, 852, 299]]<|/det|> +cancer cell significantly up- regulated the transcription of epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target genes. In addition, clinical data also suggest that L3MBTL3 expression positively correlated with epithelial- mesenchymal transition and IL6- JAK- STAT3 signature target gene transcription in breast cancer. Our study also showed L3MBTL3 up- regulated the transcription of SNAIL by recruiting active STAT3 on SNAIL promoter. These results indicate the transcription activator function of L3MBTL3. + +<|ref|>text<|/ref|><|det|>[[145, 313, 852, 593]]<|/det|> +It is reported that L3MBTL3 as a lysine methylation reader, binds the lysine methylation of non- histone protein by its MBT domain \(^{31,32}\) and recruits CRL4 \(^{DCAF5}\) ubiquitin E3 ligase to mediate the methylation- dependent degradation \(^{31 - 33}\) . In our study, we used L3MBTL3 lysine methylation binding activity mutant D381N to explore epithelial- mesenchymal transition and metastasis function of L3MBTL3. We found D381N mutant did not significantly change the expression of SNAIL, cell migration and invasion, indicating that the regulation mechanism of L3MBTL3 on epithelial- mesenchymal transition, migration and invasion are lysine methylation binding activity independent. + +<|ref|>text<|/ref|><|det|>[[145, 607, 852, 890]]<|/det|> +STAT3 as a transcription factor often up- regulates the transcription of target gene through its DNA binding domain. However, previous study is reported that STAT3 bound on SNAIL promoter in DNA binding independent way \(^{12}\) . In our study, L3MBTL3 interacted with active STAT3 and recruited active STAT3 on SNAIL promoter, then up- regulated SNAIL transcription to promote epithelial- mesenchymal transition and metastasis in breast cancer. The interaction between L3MBTL3 and active STAT3 is required for SNAIL transcription upregulation and breast cancer cell invasion and migration. Sequence analysis showed that the L3MBTL3- STAT3 colocalization sites on SNAIL promoter is including neither the classical nor other known STAT3 DNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 851, 201]]<|/det|> +binding motif. These findings support the notion that STAT3 binds on SNAIL promoter and up- regulated transcription independent on DNA binding of STAT312. Our results further explain how STAT3 is recruited on SNAIL promoter and up- regulates SNAIL transcription in DNA binding independent way. + +<|ref|>text<|/ref|><|det|>[[144, 214, 852, 759]]<|/det|> +STAT3 is phosphorylated at its Y705 site to form dimer and localizes into the nucleus, then binds the promoter of target gene to up- regulate transcription28,29. We reveal that L3MBTL3 prefers to interact with active STAT3, implying that L3MBTL3 may play a role in STAT3 activation. We observed that the phosphorylation level of STAT3 was not changed in L3MBTL3 over- expression or knockout breast cancer cells, indicated that L3MBTL3 did not regulated STAT3 phosphorylation. However, after STAT3 been activated, the increased STAT3 binding ability of L3MBTL3 suggests L3MBTL3 prefers to bind Y705 phosphorylated STAT3 or STAT3 dimer. In addition, phosphorylated STAT3 often forms an anti- parallel dimer, so it is possible that two molecule L3MBTL3 binds STAT3 anti- parallel dimer to form L3MBTL3- STAT3 tetramer. We also mapped 176- 193aa region of L3MBTL3 as the key STAT3 binding region. The identification of key STAT3 binding region of L3MBTL3 can provide important information for designing compounds or peptides to disrupt L3MBTL3- STAT3 interaction against metastasis. The interaction working model between L3MBTL3 and STAT3 is still waiting for discovering. Further biochemical characterization combined with structural analysis should help to resolve these questions and treat breast cancer metastasis. + +<|ref|>sub_title<|/ref|><|det|>[[148, 807, 239, 825]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[148, 844, 311, 861]]<|/det|> +## Clinical specimens + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 300]]<|/det|> +43 pairs of HCC (female patients, the age range is between 25 and 79) samples were obtained from the Second Affiliated Hospital of Chinese University of Hong Kong (Shenzhen, China). All patients were diagnosed with breast cancer by postoperative pathology and were free of other cancers and chronic diseases. All samples were collected with the informed consent of the patients and the experiments were approved by research ethics committee at the Second Affiliated Hospital of Chinese University of Hong Kong. + +<|ref|>sub_title<|/ref|><|det|>[[147, 346, 473, 364]]<|/det|> +## Plasmids, sgRNA, siRNA and shRNA + +<|ref|>text<|/ref|><|det|>[[144, 375, 852, 894]]<|/det|> +The sequence of L3MBTL3 ORF used in the plasmids was the CDS sequence of NM_001007102 in GenBank™. The lentivirus plasmid Plvx- IRES- Neo- FLAG- L3MBTL3, the eukaryotic expression plasmids pcDNA3- Flag- L3MBTL3, pcDNA3- Flag- L3- 2, pcDNA3- Flag- L3- 4, pcDNA3- Flag- L3- 5, pcDNA3- Flag- L3- 6, and the prokaryotic expression plasmids pGEX4T- GST- L3- 1, pGEX4T- GST- L3- 2, and pGEX4T- GST- L3- 3 were constructed using lentivirus plasmid GFP- L3MBTL3. The eukaryotic expression plasmids pcDNA3- FLAG- L3- Δ176- 193 and pcDNA3- FLAG- L3- D381N were constructed using pcDNA3- FLAG- L3MBTL3 as PCR template follow the Site- Directed Mutagenesis protocol. The prokaryotic expression plasmids pGEX4T- GST- L3- 1- Δ2- 64, pGEX4T- GST- L3- 1- Δ65- 104, pGEX4T- GST- L3- 1- Δ105- 204, pGEX4T- GST- L3- 1- Δ105- 137, pGEX4T- GST- L3- 1- Δ138- 162, pGEX4T- GST- L3- 1- Δ163- 175, pGEX4T- GST- L3- 1- Δ176- 193 and pGEX4T- GST- L3- 1- Δ194- 204 were constructed using pGEX4T- GST- L3- 1 as PCR template follow the Site- Directed Mutagenesis. The eukaryotic expression plasmids pcDNA3- HA- STAT3Y705F were constructed using the lentivirus plasmid cFUGW- Flag- STAT3Y705F. The eukaryotic expression plasmids pcDNA3- HA- STAT3Y705E were constructed using + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 850, 103]]<|/det|> +pcDNA3- HA- STAT3 as PCR template follow the Site- Directed Mutagenesis protocol. + +<|ref|>text<|/ref|><|det|>[[145, 117, 850, 135]]<|/det|> +The prokaryotic expression plasmid pGEX4T- GST- STAT3, pGEX4T- GST- + +<|ref|>text<|/ref|><|det|>[[145, 149, 850, 168]]<|/det|> +STAT3Y705F and pGEX4T- GST- STAT3 C were constructed using pcDNA3- HA- + +<|ref|>text<|/ref|><|det|>[[145, 182, 850, 201]]<|/det|> +STAT3, cFUGW- Flag- STAT3Y705F and cFUGW- Flag- STAT3 C, respectively. pGL3- + +<|ref|>text<|/ref|><|det|>[[145, 215, 722, 233]]<|/det|> +SNAIL (- 600, +80) was constructed using genomic DNA of 293T cell. + +<|ref|>text<|/ref|><|det|>[[145, 247, 851, 366]]<|/det|> +The small interfering RNA of negative control (siNC) and STAT3 siRNA (siSTAT3- 1 and siSTAT3- 2) were purchased from GenePharma (Shanghai, China). The sgRNAs of L3MBTL3 (L3- KO4 and L3- KO5) and the shRNAs of NC and mouse L3MBTL3 shRNA (L3- KD1 and L3- KD2) were purchased from TranSheepBio (Suzhou, China). The sequence of sgRNA, siRNA and shRNA were shown in + +<|ref|>text<|/ref|><|det|>[[145, 380, 851, 399]]<|/det|> +(Suzhou, China). The sequence of sgRNA, siRNA and shRNA were shown in + +<|ref|>text<|/ref|><|det|>[[147, 413, 355, 431]]<|/det|> +Supplementary Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[147, 473, 405, 490]]<|/det|> +## Generation of stable cell lines + +<|ref|>text<|/ref|><|det|>[[145, 505, 853, 820]]<|/det|> +Lentiviral plasmid GFP- L3MBTL3 was used to generate L3MBTL3 over- expression MCF7 stable cell line and the positive cells were screened by Flow cytometer. The L3MBTL3 knockout MDA- MB- 231 or SUM159PT stable cell lines or the L3MBTL3 knockdown 4T1 stable cell line were established by sgRNA or shRNA, respectively. The positive cells were screened by growth with puromycin (#A11138- 03, Gibco, USA) for two weeks. The L3MBTL3 rescue MDA- MB- 231 or SUM159PT stable cell lines were established by using lentivirus plasmid Plvx- IRES- Neo- FLAG- L3MBTL3 and L3- KO5 MDA- MB- 231 or SUM159PT stable cell lines, and the positive cells were screened by growth with G418 (#A1720- 5G, Merck, USA) for two weeks. The expression level of L3MBTL3 was analyzed by Q- PCR or Western blot. + +<|ref|>sub_title<|/ref|><|det|>[[147, 867, 446, 884]]<|/det|> +## Antibodies, inhibitor and cytokine + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 333]]<|/det|> +Antibodies against GAPDH (#AC002, ABclonal Technology, China), \(\beta\) - ACTIN (#20536- 1- AP, Proteintech, China), E- cad (#3195S, Cell Signaling Technology, USA), SNAIL (#3879S, Cell Signaling Technology, USA), STAT3 (#60199- 1- Ig Proteintech, USA), p- STAT3Y705 (#9145S, Cell Signaling Technology, USA), HA (#3724S, Cell Signaling Technology, USA), FLAG (#8146S, Cell Signaling Technology, USA) and GST (#AF2299, Beyotime, China) were used for Western blot. L3MBTL3 (#ab99928, Abcam, UK) and L3MBTL3 (#A7289, ABclonal Technology, China) were used for human and mouse L3MBTL3 Western blot, respectively. + +<|ref|>text<|/ref|><|det|>[[147, 345, 851, 429]]<|/det|> +Antibodies against L3MBTL3 (#ab99928, Abcam, UK), STAT3 (#60199- 1- Ig Proteintech, USA) and p- STAT3Y705 (#9145S, Cell Signaling Technology, USA) were used for and immunoprecipitation. + +<|ref|>text<|/ref|><|det|>[[147, 442, 851, 529]]<|/det|> +Antibodies against L3MBTL3 (#a302- 854a, Bethyl, USA) and p- STAT3Y705 (#9145S, Cell Signaling Technology, USA) were used for chromatin immunoprecipitation. + +<|ref|>text<|/ref|><|det|>[[147, 541, 851, 594]]<|/det|> +Antibodies against L3MBTL3 (#ab99928, Abcam, UK) and SNAIL (#A5544, ABclonal Technology, China) were used for Immunohistochemistry. + +<|ref|>text<|/ref|><|det|>[[147, 607, 851, 690]]<|/det|> +Antibody against E- cad (#610181, BD Biosciences, USA), STAT3 (#60199- 1- Ig Proteintech, USA) and FLAG (#8146S, Cell Signaling Technology, USA) were used for immunofluorescence. + +<|ref|>text<|/ref|><|det|>[[147, 704, 851, 825]]<|/det|> +To inhibit STAT3 pathway, the cells were treated with growth medium containing \(100 \mu \mathrm{M}\) STAT3 inhibitor NIF (HY- B1436, MedChemExpress, USA) for 16 hours. To activate STAT3 pathway, the cells were treated with growth medium containing \(10 \mathrm{ng / ml}\) Cytokine IL- 6 (#AF- 200- 06, Peprotech, USA) for 30 min. + +<|ref|>title<|/ref|><|det|>[[148, 870, 352, 887]]<|/det|> +# Immunohistochemistry + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 310]]<|/det|> +Immunohistochemistry was performed according to standard protocol. In brief, slides were incubated at \(60^{\circ}\mathrm{C}\) for \(1\mathrm{h}\) , followed by \(10\mathrm{min}\) incubations in xylene and rehydration in graded alcohol to water. Sodium citrate buffer (pH 6.0) was used for antigen retrieval and slides were incubated with primary antibody overnight at \(4^{\circ}\mathrm{C}\) and stained with VECTASTAIN Elite ABC kit (PK- 4001, Vector labs, USA). Bound antibody was detected with DAB (3,3'- diaminobenzidine) peroxidase kit (SK4100, Vector labs, USA). + +<|ref|>text<|/ref|><|det|>[[144, 325, 852, 507]]<|/det|> +Staining intensity and extent of staining were graded as follows: negative (score 0), weak (score 1), moderate (score 2), and strong (score 3). Extent of staining was also grouped into quantiles according to the percentage of high- staining cells per field: negative (score 0), \(25\%\) (score 1), \(26\% - 50\%\) (score 2), \(51\% - 75\%\) (score 3), and \(76\% - 100\%\) (score 4). All immunohistochemical staining was evaluated and scored by at least two independent pathologists blinded to the experimental groups. + +<|ref|>sub_title<|/ref|><|det|>[[149, 555, 403, 572]]<|/det|> +## Migration and invasion assay + +<|ref|>text<|/ref|><|det|>[[144, 587, 852, 715]]<|/det|> +The migration and invasion assay were performed as previously described \(^{43}\) . \(8\mu \mathrm{m}\) pore size Boyden chambers (353097, Corning, USA) were used to migration assay. \(8\mu \mathrm{m}\) pore size Byden chambers (353097, Corning, USA) precoated with Matrigel (356231, Coring, USA) were used to invasion assay. + +<|ref|>sub_title<|/ref|><|det|>[[149, 765, 292, 781]]<|/det|> +## Metastasis assay + +<|ref|>text<|/ref|><|det|>[[144, 795, 853, 912]]<|/det|> +We used a protocol approved by the Ethics Committee of East China Normal University. 6- week- old female BALB/c mice were used to established spontaneous lung metastasis model. The NC and L3MBTL3 knockdown (L3- KD1 and L3- KD2) 4T1 cells were trypsinized and suspended in a 1:2 (vol/vol) mix of Matrigel (Corning) and PBS. Then + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 851, 201]]<|/det|> +the 4T1 cells (5×105 cells per gland, mice were randomly divided into 3 groups, 9 mice per group) were injected into the fourth mammary fat pads of BALB/c mice. All the mice were sacrificed at the end of the experiment and the lung were collected for metastasis analysis. + +<|ref|>sub_title<|/ref|><|det|>[[148, 248, 336, 266]]<|/det|> +## RNA-seq and Q-PCR + +<|ref|>text<|/ref|><|det|>[[144, 280, 852, 469]]<|/det|> +The total mRNAs were extracted from cells using TRIzol reagent (15596018, Ambion, USA). For RNA- seq, the RNA- seq libraries from 200ng of total RNA were prepared by using Hieff NGS Ultima Dual- mode mRNA Library Prep Kit for Illumina (yeasen, Cat12301), according to the manufacturer's instructions. Libraries were validated with an QIAxcel Advanced system (QIAGEN, Germany). Sequencing was performed on Illumina HiSeq 2500 sequencer or Illumina NovaSeq 6000. + +<|ref|>text<|/ref|><|det|>[[144, 483, 851, 600]]<|/det|> +For Q- PCR, reverse transcription was performed using PrimeScriptTM RT reagent kit (#RR047A, TaKaRa, Japan). The Q- PCR was performed using Power SYBR Green PCR Master Mix (#4367659, Applied Biosystems, USA) on ABI QuantStudio™ 6 Flex System. Primers of Q- PCR were shown in Supplementary Table 2. + +<|ref|>sub_title<|/ref|><|det|>[[148, 648, 379, 666]]<|/det|> +## GSEA and GSVA analysis + +<|ref|>text<|/ref|><|det|>[[144, 680, 852, 896]]<|/det|> +The 50 hallmark gene sets in the MSigDB databases (https://www.gsea- msigdb.org/gsea/msigdb) were used for the Gene Set Enrichment Analysis (GSEA)44 or Gene Set Variation Analysis (GSVA)45 in RNA- seq of breast cancer cells, TCGA Breast Cancer datasets (TCGA BRCA 2017: https://xenabrowser.net/datapages/?dataset=TCGA.BRCA.sampleMap%2FHiSeqV2 &host=https%3A%2F%2Ftcga.xenahubs.net&removeHub=https%3A%2F%2Fxena.tr eehouse.gi.ucsc.edu%3A443 and TCGA BRCA 2000: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 848, 101]]<|/det|> +https://portal.gdc.cancer.gov/repository?facetTab=cases&filters=%7B%22content%2 + +<|ref|>text<|/ref|><|det|>[[144, 116, 848, 133]]<|/det|> +2%3A%5B%7B%22content%22%3A%7B%22field%22%3A%22cases.project.projec + +<|ref|>text<|/ref|><|det|>[[144, 149, 558, 165]]<|/det|> +t_id%22%2C%22value%22%3A%5B%22TCGA- + +<|ref|>text<|/ref|><|det|>[[144, 181, 848, 198]]<|/det|> +BRCA%22%5D%7D%2C%22op%22%3A%22in%22%7D%2C%7B%22content%22 + +<|ref|>text<|/ref|><|det|>[[144, 214, 840, 231]]<|/det|> +%3A%7B%22field%22%3A%22files.data_category%22%2C%22value%22%3A%5 + +<|ref|>text<|/ref|><|det|>[[144, 247, 840, 264]]<|/det|> +B%22Transcriptome%20Profiling%22%5D%7D%2C%22op%22%3A%22in%22%7 + +<|ref|>text<|/ref|><|det|>[[144, 280, 850, 297]]<|/det|> +D%5D%2C%22op%22%3A%22and%22%7D&searchTableTab=cases) and GEO + +<|ref|>text<|/ref|><|det|>[[144, 313, 850, 330]]<|/det|> +datasets (GSE202203) to assess the relationship between L3MBTL3 expression level + +<|ref|>text<|/ref|><|det|>[[144, 346, 704, 363]]<|/det|> +and epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway. + +<|ref|>sub_title<|/ref|><|det|>[[144, 412, 333, 428]]<|/det|> +## Immunofluorescence + +<|ref|>text<|/ref|><|det|>[[144, 445, 850, 696]]<|/det|> +Cells were fixed with 4% paraformaldehyde for 20 min followed by a 15-min cell permeabilization with 1% (vol/vol) Triton X-100 in PBS. Then the cells were blocked in PBS with 2% (wt/vol) BSA for 1h at 37°C. Antibody were diluted in PBS with 2% (wt/vol) BSA and incubated with the cells overnight at 4°C. After rinsing, the cells were incubated with secondary antibodies conjugated to Alexa Fluor 568 (#A-11036, Invitrogen, USA) or Alexa Fluor 488 (#A11001, Invitrogen, USA) at 37°C for 30 min. After stained with DAPI (1 ng/ml)(#C1006, Beyotime, China), the cells were quickly rinsed in PBS and then imaged. + +<|ref|>sub_title<|/ref|><|det|>[[144, 753, 718, 770]]<|/det|> +## Immunoprecipitation, mass spectrum analysis and GST-pulldown + +<|ref|>text<|/ref|><|det|>[[144, 786, 852, 901]]<|/det|> +For immunoprecipitation, cells were collected by centrifugation and lysed in lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.1% NP40, 8% glycerol, protease inhibitor cocktail (#P1010, Beyotime, China)) on ice for 30 min. Then the cell lysate supernatant was incubated with antibodies and protein A+G-agarose beads + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 83, 851, 200]]<|/det|> +(#P2055, Beyotime, China), M2-conjugate agarose beads (#A2220, Merck, USA) or anti- HA nanobody conjugated agarose beads (#HNA- 25- 500, PromiOne, China) by rotation overnight at \(4^{\circ}\mathrm{C}\) . Then the beads were analyzed by Western blot after washed for 5 times. + +<|ref|>text<|/ref|><|det|>[[145, 215, 852, 398]]<|/det|> +For mass spectrum analysis, the protein was immunoprecipitated using M2- conjugate agarose beads (#A2220, Merck, USA). After washing, the beads were eluted by \(3 \times \mathrm{FLAG}\) peptide (#F4799, Merck, USA). The elution protein was verified using silver staining, then the target lane was excised and subjected to analyze by an EASY- nLC 1200 UHPLC system (ThermoFisher Scientific, USA) coupled to a Q Exactive HF- X mass spectrometer (ThermoFisher Scientific, USA). + +<|ref|>text<|/ref|><|det|>[[145, 411, 851, 530]]<|/det|> +For GST- pulldown assay, GST- tag protein was expressed in E. coli and incubated with the BeyoGold™ GST- tag Purification Resin (#P2251, Beyotime, China) beads. After washing, the beads were incubated with 293T cell lysate supernatant by rotation overnight at \(4^{\circ}\mathrm{C}\) . Then the beads were analyzed by Western blot after washed 5 times. + +<|ref|>sub_title<|/ref|><|det|>[[149, 575, 419, 593]]<|/det|> +## ChIP sequence and ChIP assay + +<|ref|>text<|/ref|><|det|>[[145, 606, 852, 818]]<|/det|> +ChIP sequence and ChIP assayChIP were performed according to the protocol as previously described46. Briefly, cells were cross- linked using 1% formaldehyde and sonicated to generate DNA fragments of about 500 bp. For immunoprecipitation, Dynabeads™ Protein A or G magnetic beads (Invitrogen) were incubated with antibodies and chromatin fraction of cell lysate overnight at \(4^{\circ}\mathrm{C}\) with rotation. After washing, the protein- DNA complexes were eluted. Then the DNA was purified and used for Q- PCR. Primers were shown in + +<|ref|>text<|/ref|><|det|>[[147, 836, 850, 889]]<|/det|> +For ChIP- seq, the sequencing libraries were generated using NEBNext Ultra DNA Library Prep Kit from Illumina (NEB, USA) by following manufacturer's + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 83, 852, 233]]<|/det|> +recommendations and index codes were added to attribute sequences to each sample. The clustering of the index- coded samples was performed on a cBot Cluster Generation System using HiSeq Rapid Duo cBot Sample Loading Kit (Illumia) according to the manufacturer's instructions. The libraries were sequenced on the Illumina NovaSeq 6000. + +<|ref|>sub_title<|/ref|><|det|>[[147, 281, 468, 299]]<|/det|> +## ChIP-seq analysis and motif analysis + +<|ref|>text<|/ref|><|det|>[[144, 312, 853, 590]]<|/det|> +After filtering and deduping, uniquely mapped tags of L3MBTL3 and STAT3 (GSE85579 and GSE152203)41 ChIP- seq were mapped to the Homo sapiens reference genome (hg19) by Bowtie2. Normalized genome- wide signal- coverage tracks from raw- read alignment files were built by Samtools, deeptools, and MACS2. Visualization of the ChIP- seq signal at enriched genomic regions (Profile and Heatmap) was achieved by using deepTools. Peak- associated genes and motif were identified by using the annotate Peaks function of HOMER. Each of the STAT3 or L3MBTL3- binding sites were assigned to the nearest gene. Further annotation information includes whether a peak is in the Promoter, 5' UTR, 3' UTR, Exon, Intron, Downstream or Distal Intergenic. + +<|ref|>text<|/ref|><|det|>[[145, 608, 852, 757]]<|/det|> +For motif analysis, the DNA segments from L3MBTL3 or STAT3 binding sites were analyzed by homer script findMotifGenome.pl with argument "hg19" to detect enrichment of de novo and known TF motif. Homer results were further manually analyzed. Those motifs with high background enrichment and possible false positives were removed. + +<|ref|>sub_title<|/ref|><|det|>[[147, 806, 292, 823]]<|/det|> +## Luciferase assay + +<|ref|>text<|/ref|><|det|>[[145, 836, 851, 888]]<|/det|> +Plasmids derived from pGL3 vectors were co- transfected with Renilla luciferase vectors (phRL- TK) and the luciferase assay were performed using the Dual- Luciferase + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 850, 135]]<|/det|> +Reporter Assay System (#E1910, Promega, USA) according to the manufacturer's instructions. + +<|ref|>sub_title<|/ref|><|det|>[[147, 182, 310, 199]]<|/det|> +## Statistical analysis + +<|ref|>text<|/ref|><|det|>[[144, 214, 851, 338]]<|/det|> +SPSS version 17.0 software was used for statistical analysis. Results are expressed as mean \(\pm\) standard deviation (SD) and Student's t- test (two- tailed) was used to indicate significant differences between groups. In all experiments, \(\mathrm{p}< 0.05\) was considered to indicate statistical significance. + +<|ref|>sub_title<|/ref|><|det|>[[147, 386, 320, 405]]<|/det|> +## Data Availability + +<|ref|>text<|/ref|><|det|>[[144, 421, 852, 605]]<|/det|> +The RNA- seq and ChIP- seq data generated in this study have been deposited in Sequence Read Archive (SRA) repository under SRA accession number PRJNA934109 and PRJNA934165, respectively. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE47 partner repository with the dataset identifier PXD040150. Source data are provided in this paper. + +<|ref|>sub_title<|/ref|><|det|>[[147, 653, 261, 672]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[144, 700, 852, 904]]<|/det|> +1. Park, M. et al. Breast Cancer Metastasis: Mechanisms and Therapeutic Implications. Int J Mol Sci 23, 6806 (2022). +2. Buyuk, B., Jin, S. & Ye, K. Epithelial-to-Mesenchymal Transition Signaling Pathways Responsible for Breast Cancer Metastasis. Cell Mol Bioeng 15, 1–13 (2022). +3. Zhou, B. P. et al. Dual regulation of Snail by GSK-3beta-mediated phosphorylation in control of epithelial-mesenchymal transition. Nat Cell Biol 6, 931–940 (2004). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 850, 150]]<|/det|> +4. Barralo-Gimeno, A. & Nieto, M. A. The Snail genes as inducers of cell movement and survival: implications in development and cancer. Development 132, 3151–3161 (2005). + +<|ref|>text<|/ref|><|det|>[[144, 168, 850, 223]]<|/det|> +5. Lamouille, S., Xu, J. & Derynck, R. Molecular mechanisms of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol 15, 178–196 (2014). + +<|ref|>text<|/ref|><|det|>[[144, 242, 850, 297]]<|/det|> +6. Olmeda, D., Jordá, M., Peinado, H., Fabra, A. & Cano, A. Snail silencing effectively suppresses tumour growth and invasiveness. Oncogene 26, 1862–1874 (2007). + +<|ref|>text<|/ref|><|det|>[[144, 315, 850, 370]]<|/det|> +7. Olmeda, D. et al. Snail and Sna12 collaborate on tumor growth and metastasis properties of mouse skin carcinoma cell lines. Oncogene 27, 4690–4701 (2008). + +<|ref|>text<|/ref|><|det|>[[144, 389, 850, 444]]<|/det|> +8. Azmi, A. S. et al. Systems analysis reveals a transcriptional reversal of the mesenchymal phenotype induced by SNAIL-inhibitor GN-25. BMC Syst Biol 7, 85 (2013). + +<|ref|>text<|/ref|><|det|>[[144, 463, 850, 518]]<|/det|> +9. Ciruna, B. & Rossant, J. FGF signaling regulates mesoderm cell fate specification and morphogenetic movement at the primitive streak. Dev Cell 1, 37–49 (2001). + +<|ref|>text<|/ref|><|det|>[[144, 537, 850, 592]]<|/det|> +10. Yook, J. I., Li, X.-Y., Ota, I., Fearon, E. R. & Weiss, S. J. Wnt-dependent regulation of the E-cadherin repressor snail. J Biol Chem 280, 11740–11748 (2005). + +<|ref|>text<|/ref|><|det|>[[144, 611, 850, 704]]<|/det|> +11. Cho, H. J., Baek, K. E., Saika, S., Jeong, M.-J. & Yoo, J. Snail is required for transforming growth factor-beta-induced epithelial-mesenchymal transition by activating PI3 kinase/Akt signal pathway. Biochem Biophys Res Commun 353, 337–343 (2007). + +<|ref|>text<|/ref|><|det|>[[144, 723, 850, 778]]<|/det|> +12. Saitoh, M. et al. STAT3 integrates cooperative Ras and TGF-β signals that induce Snail expression. Oncogene 35, 1049–1057 (2016). + +<|ref|>text<|/ref|><|det|>[[144, 797, 850, 889]]<|/det|> +13. Talbot, L. J., Bhattacharya, S. D. & Kuo, P. C. Epithelial-mesenchymal transition, the tumor microenvironment, and metastatic behavior of epithelial malignancies. Int J Biochem Mol Biol 3, 117–136 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 850, 150]]<|/det|> +14. Wang, H.-Q. et al. STAT3 pathway in cancers: Past, present, and future. MedComm (2020) 3, e124 (2022). + +<|ref|>text<|/ref|><|det|>[[144, 169, 850, 260]]<|/det|> +15. Hirano, T., Ishihara, K. & Hibi, M. Roles of STAT3 in mediating the cell growth, differentiation and survival signals relayed through the IL-6 family of cytokine receptors. Oncogene 19, 2548–2556 (2000). + +<|ref|>text<|/ref|><|det|>[[144, 280, 850, 335]]<|/det|> +16. Kamran, M. Z., Patil, P. & Gude, R. P. Role of STAT3 in cancer metastasis and translational advances. Biomed Res Int 2013, 421821 (2013). + +<|ref|>text<|/ref|><|det|>[[144, 354, 850, 446]]<|/det|> +17. Silver, D. L., Naora, H., Liu, J., Cheng, W. & Montell, D. J. Activated signal transducer and activator of transcription (STAT) 3: localization in focal adhesions and function in ovarian cancer cell motility. Cancer Res 64, 3550–3558 (2004). + +<|ref|>text<|/ref|><|det|>[[144, 465, 850, 520]]<|/det|> +18. Sano, S. et al. Keratinocyte-specific ablation of Stat3 exhibits impaired skin remodeling, but does not affect skin morphogenesis. EMBO J 18, 4657–4668 (1999). + +<|ref|>text<|/ref|><|det|>[[144, 539, 850, 594]]<|/det|> +19. Azare, J. et al. Constitutively activated Stat3 induces tumorigenesis and enhances cell motility of prostate epithelial cells through integrin beta 6. Mol Cell Biol 27, 4444–4453 (2007). + +<|ref|>text<|/ref|><|det|>[[144, 613, 850, 668]]<|/det|> +20. Suiqing, C., Min, Z. & Lirong, C. Overexpression of phosphorylated-STAT3 correlated with the invasion and metastasis of cutaneous squamous cell carcinoma. J Dermatol 32, 354–360 (2005). + +<|ref|>text<|/ref|><|det|>[[144, 687, 850, 778]]<|/det|> +21. Qiu, Z. et al. RNA interference-mediated signal transducers and activators of transcription 3 gene silencing inhibits invasion and metastasis of human pancreatic cancer cells. Cancer Sci 98, 1099–1106 (2007). + +<|ref|>text<|/ref|><|det|>[[144, 799, 850, 853]]<|/det|> +22. Li, H. dong et al. STAT3 knockdown reduces pancreatic cancer cell invasiveness and matrix metalloproteinase-7 expression in nude mice. PLoS One 6, e25941 (2011). + +<|ref|>text<|/ref|><|det|>[[144, 873, 850, 890]]<|/det|> +23. Sullivan, N. J. et al. Interleukin-6 induces an epithelial-mesenchymal transition phenotype in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 580, 111]]<|/det|> +human breast cancer cells. Oncogene 28, 2940- 2947 (2009). + +<|ref|>text<|/ref|><|det|>[[144, 131, 850, 222]]<|/det|> +24. Yadav, A., Kumar, B., Datta, J., Teknos, T. N. & Kumar, P. IL-6 promotes head and neck tumor metastasis by inducing epithelial-mesenchymal transition via the JAK-STAT3-SNAIL signaling pathway. Mol Cancer Res 9, 1658-1667 (2011). + +<|ref|>text<|/ref|><|det|>[[144, 242, 850, 297]]<|/det|> +25. Xiong, H. et al. Roles of STAT3 and ZEB1 proteins in E-cadherin down-regulation and human colorectal cancer epithelial-mesenchymal transition. J Biol Chem 287, 5819-5832 (2012). + +<|ref|>text<|/ref|><|det|>[[144, 316, 850, 409]]<|/det|> +26. Gong, C. et al. Abnormally expressed JunB transactivated by IL-6/STAT3 signaling promotes uveal melanoma aggressiveness via epithelial-mesenchymal transition. Biosci Rep 38, BSR20180532 (2018). + +<|ref|>text<|/ref|><|det|>[[144, 428, 850, 483]]<|/det|> +27. Teng, Y., Ross, J. L. & Cowell, J. K. The involvement of JAK-STAT3 in cell motility, invasion, and metastasis. JAKSTAT 3, e28086 (2014). + +<|ref|>text<|/ref|><|det|>[[144, 502, 850, 557]]<|/det|> +28. Bromberg, J. & Darnell, J. E. The role of STATs in transcriptional control and their impact on cellular function. Oncogene 19, 2468-2473 (2000). + +<|ref|>text<|/ref|><|det|>[[144, 576, 850, 631]]<|/det|> +29. Horvath, C. M., Wen, Z. & Darnell, J. E. A STAT protein domain that determines DNA sequence recognition suggests a novel DNA-binding domain. Genes Dev 9, 984-994 (1995). + +<|ref|>text<|/ref|><|det|>[[144, 650, 850, 704]]<|/det|> +30. James, L. I. et al. Discovery of a chemical probe for the L3MBTL3 methyllysine reader domain. Nat Chem Biol 9, 184-191 (2013). + +<|ref|>text<|/ref|><|det|>[[144, 724, 850, 778]]<|/det|> +31. Zhang, C. et al. Proteolysis of methylated SOX2 protein is regulated by L3MBTL3 and CRL4DCAF5 ubiquitin ligase. J Biol Chem 294, 476-489 (2019). + +<|ref|>text<|/ref|><|det|>[[144, 799, 850, 853]]<|/det|> +32. Leng, F. et al. Methylated DNMT1 and E2F1 are targeted for proteolysis by L3MBTL3 and CRL4DCAF5 ubiquitin ligase. Nat Commun 9, 1641 (2018). + +<|ref|>text<|/ref|><|det|>[[144, 873, 850, 890]]<|/det|> +33. Guo, P. et al. The assembly of mammalian SWI/SNF chromatin remodeling complexes is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 95, 759, 112]]<|/det|> +regulated by lysine-methylation dependent proteolysis. Nat Commun 13, 6696 (2022). + +<|ref|>text<|/ref|><|det|>[[144, 131, 850, 187]]<|/det|> +34. Xu, T. et al. RBPJ/CBF1 interacts with L3MBTL3/MBT1 to promote repression of Notch signaling via histone demethylase KDM1A/LSD1. EMBO J 36, 3232-3249 (2017). + +<|ref|>text<|/ref|><|det|>[[144, 206, 850, 231]]<|/det|> +35. Hall, D. et al. The structure, binding and function of a Notch transcription complex involving RBPJ and the epigenetic reader protein L3MBTL3. Nucleic Acids Res 50, 13083-13099 (2022). + +<|ref|>text<|/ref|><|det|>[[144, 245, 830, 262]]<|/det|> +36. Arai, S. & Miyazaki, T. Impaired maturation of myeloid progenitors in mice lacking novel Polycomb group protein MBT-1. EMBO J 24, 1863-1873 (2005). + +<|ref|>text<|/ref|><|det|>[[144, 282, 850, 337]]<|/det|> +37. Alcina, A. et al. Identification of the genetic mechanism that associates L3MBTL3 to multiple sclerosis. Hum Mol Genet 31, 2155-2163 (2022). + +<|ref|>text<|/ref|><|det|>[[144, 360, 850, 406]]<|/det|> +38. Nazari Mehrabani, S. Z. et al. Association of SHMT1, MAZ, ERG, and L3MBTL3 Gene Polymorphisms with Susceptibility to Multiple Sclerosis. Biochem Genet 57, 355-370 (2019). + +<|ref|>text<|/ref|><|det|>[[144, 428, 850, 483]]<|/det|> +39. Kar, S. P. et al. Genome-Wide Meta-Analyses of Breast, Ovarian, and Prostate Cancer Association Studies Identify Multiple New Susceptibility Loci Shared by at Least Two Cancer Types. Cancer Discov 6, 1052-1067 (2016). + +<|ref|>text<|/ref|><|det|>[[144, 504, 852, 590]]<|/det|> +40. Kolosenko, I., Grander, D. & Tamm, K. P. IL-6 activated JAK/STAT3 pathway and sensitivity to Hsp90 inhibitors in multiple myeloma. Curr Med Chem 21, 3042-3047 (2014). + +<|ref|>text<|/ref|><|det|>[[144, 612, 850, 667]]<|/det|> +41. Conway, M. E. et al. STAT3 and GR Cooperate to Drive Gene Expression and Growth of Basal-Like Triple-Negative Breast Cancer. Cancer Res 80, 4355-4370 (2020). + +<|ref|>text<|/ref|><|det|>[[144, 688, 850, 743]]<|/det|> +42. Xu, L. et al. The STAT3 HIES mutation is a gain-of-function mutation that activates genes via AGG-element carrying promoters. Nucleic Acids Res 43, 8898-8912 (2015). + +<|ref|>text<|/ref|><|det|>[[144, 763, 850, 818]]<|/det|> +43. Zhang, H. et al. Histone demethylase KDM2A suppresses EGF-TSPAN8 pathway to inhibit breast cancer cell migration and invasion in vitro. Biochem Biophys Res Commun 628, 104-109 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 95, 850, 412]]<|/det|> +754 (2022). 755 44. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for 756 interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545- 15550 (2005). 757 45. Liberzon, A. et al. Molecular signatures database (MSigDB) 3.0. Bioinformatics 27, 1739- 758 1740 (2011). 759 46. Schmidt, D. et al. ChIP-seq: using high-throughput sequencing to discover protein-DNA 760 interactions. Methods 48, 240- 248 (2009). 761 47. Perez- Riverol, Y. et al. The PRIDE database resources in 2022: a hub for mass spectrometry- 762 based proteomics evidences. Nucleic Acids Res 50, D543- D552 (2022). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 85, 346, 105]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[144, 121, 852, 504]]<|/det|> +We thank Prof. Mingyao Liu and Prof. Zhengfang Yi from East China Normal University for providing cFUGW- Flag- STAT3Y705F and cFUGW- Flag- STAT3 C plasmids. We thank Prof. Jian Luo from Tongji University for providing pcDNA3- HA- STAT3 plasmid. We thank Prof. Arrowsmith from University of North Carolina at Chapel Hill for providing L3MBTL3 full- length plasmid GFP- L3MBTL3. This study was supported by grants from the National Natural Science Foundation of China (82173044 and 81772817), Shanghai Medical Leadership Program (2019LJ25), Medical and Health Technology Public Relations Project of Shenzhen Longgang District Science and Technology Innovation Special Fund (LGKCYLWS2022001), Special Program for Collaborative Innovation in Shanghai University of Medicine & Health Sciences (SPCI- 17- 15- 001), and Key Subject Construction Project of Shanghai Municipal Commission of Health and Family Planning, China (ZK2019B29). + +<|ref|>sub_title<|/ref|><|det|>[[148, 555, 371, 574]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[144, 590, 852, 710]]<|/det|> +Jing Feng and Jing Li: study design; Jianpeng Xiao, Jie Wang, Jialun Li, Jie Xiao, CuiCui Liu, Libi Tan, Yanhong Tu, Ruifang Yang, Yujie Pei, Minghua Wang, Jiemin Wong, Binhua P Zhou: study conduct, data analysis, and its interpretation; Jing Li and Jianpeng Xiao: manuscript drafting. + +<|ref|>sub_title<|/ref|><|det|>[[148, 763, 355, 783]]<|/det|> +## Competing Interests + +<|ref|>text<|/ref|><|det|>[[148, 800, 503, 817]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[148, 872, 230, 890]]<|/det|> +## ORCID + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 84, 550, 101]]<|/det|> +788 Jing Feng: http://orcid.org/0000- 0003- 1084- 4920 + +<|ref|>text<|/ref|><|det|>[[90, 117, 534, 135]]<|/det|> +789 Jing Li: https://orcid.org/0000- 0003- 4852- 5781 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[168, 140, 220, 156]]<|/det|> +Fig. 1 + +<|ref|>image<|/ref|><|det|>[[220, 189, 512, 515]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[540, 190, 785, 515]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 852, 430]]<|/det|> +Fig. 1 High L3MBLT3 expression is associated with metastatic and poor clinical outcomes of breast cancer. a Human breast cancer tissues consisting of primary breast cancer samples (Tumor) and paired tumor- adjacent tissue samples (AdjN) and breast cancer lymph node metastasis samples (Metastasis) were subjected to Immunohistochemistry staining for L3MBLT3 (scale bar, \(25 \mu \mathrm{m}\) ). b H- scores of L3MBTL3 protein in 43 breast cancer patients of primary breast cancer samples (Tumor) and paired tumor- adjacent tissue samples (AdjN) and breast cancer lymph node metastasis samples (Metastasis). The data were analyzed using Student's t test. \(\mathrm{n} = 43\) . \(^{*}\mathrm{p}< 0.05\) , \(***\mathrm{p}< 0.001\) . c Distant metastasis survival of breast cancer patients with L3MBLT3 high and low expression (GSE12276, \(\mathrm{n} = 202\) ). L3MBLT3 high, \(\mathrm{n} = 103\) ; L3MBLT3 low, \(\mathrm{n} = 99\) ; \(\mathrm{p} = 0.00153024\) . + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[166, 103, 225, 120]]<|/det|> +
Fig. 2
+ +<|ref|>image<|/ref|><|det|>[[152, 131, 833, 610]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 80, 854, 841]]<|/det|> +Fig. 2 L3MBTL3 promotes epithelial-mesenchymal transition, migration and invasion and metastasis in breast cancer. a Western blot showed L3MBTL3 highly expressed in basal-like breast cancer cells, which was positively correlated with epithelial-mesenchymal transition factor SNAIL and negatively correlated with epithelial cell marker E-cad. Antibody \(\beta\) - ACTIN was used as a loading control. b Western blot showed the L3MBTL3 expression level in L3MBTL3 over-expression MCF7 cells, L3MBTL3 knockout and rescue MDA-MB-231 and SUM159PT cells. Antibody GAPDH and \(\beta\) - ACTIN were used as loading controls. c Over-expressing L3MBTL3 converted MCF7 cell into spindle-shaped mesenchymal-like cells (Scale bar, \(100 \mu \mathrm{m}\) ). d Knocking out L3MBTL3 converted MDA-MB-231 and SUM159PT cell into epithelial like cell (Scale bar, \(100 \mu \mathrm{m}\) ). e Restoring L3MBTL3 rescued the mesenchymal morphology of L3MBTL3 knockout MDA-MB-231 and SUM159PT cells (Scale bar, \(100 \mu \mathrm{m}\) ). f Over-expressing L3MBTL3 increased the migration ability of MCF7 cells. g Knocking out L3MBTL3 decreased the migration and invasion ability of MDA-MB-231 and SUM159PT cells. h Restoring L3MBTL3 rescued the migration and invasion ability of L3MBTL3 knockout MDA-MB-231 and SUM159PT cells. i Western blot showed the knocking down efficiency of L3MBTL3 in 4T1 mouse breast cancer cell. Antibody GAPDH was used as a loading control. j The L3MBTL3 knockdown 4T1 cells were injection to mice to induce lung metastasis. Photographs showed the lung metastasis focus (Scale bar, \(5 \mathrm{mm}\) ). The numbers of metastatic nodules in the lungs per mouse were quantified (mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 9\) ). For f, g and h panels, bars indicate the mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(* \mathrm{p} < 0.05\) , \(** \mathrm{p} < 0.01\) , \(*** \mathrm{p} < 0.005\) , \(**** \mathrm{p} < 0.001\) . + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[164, 103, 220, 120]]<|/det|> +
Fig. 3
+ +<|ref|>image<|/ref|><|det|>[[150, 120, 825, 789]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 80, 852, 895]]<|/det|> +Fig. 3 L3MBTL3 interacts with STAT3 and activating STAT3 enhances the interaction between L3MBTL3 and STAT3. a Differences in pathway activities scored by GSEA between control (C) and L3MBTL3 knockout (L3- KO4 and L3- KO5) group in SUM159PT cells. Normalized enrichment score (NES)>1.4, p<0.05. b In L3MBTL3 rescued MDA- MB- 231 cells, immunoprecipitation of Flag- L3MBTL3 interacting protein was visualized by silver gel staining. c Network analysis (Cytoscape v 3.9.1) of the significant nuclear proteins involving in epithelial- mesenchymal transition process (score>500) bound to L3MBTL3 identified through immunoprecipitation and mass spectrometry proteomics. L3MBTL3 was highlighted by red. The transcription factors were highlighted by orange. The other nuclear proteins were highlighted by blue. d Immunofluorescence showed the L3MBTL3 and STAT3 co- localized in nucleus in L3MBTL3 rescue MDA- MB- 231 cells. Nuclei were stained with DAPI (Scale bar, 20 μm). e Co- immunoprecipitation showed that endogenous L3MBTL3 interacted with endogenous STAT3 and p- STAT3Y705 in MDA- MB- 231 and SUM159PT cells. f Co- immunoprecipitation showed that exogenous Flag- L3MBTL3 interacted with exogenous HA- STAT3 in 293T cells. g Co- immunoprecipitation showed that exogenous Flag- L3MBTL3 interacted with endogenous STAT3 and p- STAT3Y705 in L3MBTL3 rescue MAD- MB- 231 cells. Antibody β- ACTIN was used as loading control. The asterisk showed the unspecific band. h GST pull- down showed that, comparing to wildtype STAT3, STAT3 C (constantly active form) had higher L3MBTL3 binding ability, but STAT3Y705F (deactive form) had lower L3MBTL3 binding ability. The asterisk showed the unspecific band. i Co- immunoprecipitation showed that, comparing to wildtype STAT3, exogenous Flag- L3MBTL3 had higher binding ability on HA- STAT3Y705E (mimic p- STAT3 Y705, active form), but had lower binding ability on HA- STAT3Y705F (deactive form). j Co- immunoprecipitation showed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[91, 82, 852, 170]]<|/det|> +858 that the binding ability of L3MBTL3 to STAT3 or p- STAT3Y705 increased after L3MBTL3 rescue MAD- MB- 231 cells treated with IL6. Antibody \(\beta\) - ACTIN was used as loading control. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[178, 135, 833, 590]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 83, 852, 530]]<|/det|> +Fig. 4 L3MBTL3 up-regulated epithelial-mesenchymal transition and IL6-JAK- STAT3 pathway. a GSEA plots depicting the enrichment of epithelial-mesenchymal transition genes down-regulated in L3MBTL3 knockout SUM159PT cells and up-regulated in L3MBTL3 over-expression MCF7 cells and L3MBTL3 rescued SUM159PT cells. p<0.05. b GSEA plots depicting the enrichment of IL6-JAK-STAT3 pathway genes down-regulated in L3MBTL3 knockout SUM159PT cells. p<0.05. c Heatmap of GSVA enrichment scores of epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway in L3MBTL3 high and low expression tumors. The gene expression datasets were form TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203, n=3207). d Scatter plots showing the positive correlation of transcript expression between L3MBTL3 and epithelial-mesenchymal transition or IL6-JAK-STAT3 pathway in TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203, n=3207.). p<0.0001. + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[165, 103, 222, 119]]<|/det|> +
Fig. 5
+ +<|ref|>image<|/ref|><|det|>[[155, 127, 835, 760]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 87, 852, 895]]<|/det|> +Fig. 5 L3MBTL3 up-regulated transcription of SNAIL in breast cancer. a ChIP-seq profiles (up) and heat maps of ChIP-Seq signal intensity (down) of STAT3 and L3MBTL3 form - 2.5 kb before transcription start site (TSS) to 2.5 kb after transcription end site (TES) in SUM159PT cells. b Venn diagram of the overlapped L3MBTL3 and STAT3 occupied genes detected by ChIP-seq in SUM159PT cell and different expression genes detected by RNA-seq in L3MBTL3 rescue SUM159PT cells. c Heatmaps of transcription changes of epithelial-mesenchymal transition up-regulated genes in control (C+V), L3MBTL3 knockout (L3-KO5+V) and rescue (L3-KO5+L3) SUM159PT cells. d Q-PCR analyzed showed that over-expressing L3MBTL3 in MCF7 cells up-regulated the transcription of SNAIL and decreased transcription of SNAIL target genes E-cad, OCLN, and DSP. e WB analyzed showed that over-expressing L3MBTL3 in MCF7 cells increased SNAIL expression and decreased SNAIL target genes E-cad expression. Antibody GAPDH was used as a loading control. f Immunofluorescence showed the E-cad expression decreased in L3MBTL3 overexpression MCF7 cells. Nuclei were stained with DAPI (Scale bar, 20 μm). g, h Q-PCR analyzed showed that (g) knocking out L3MBTL3 in MDA-MB-231 and SUM159PT cells down-regulated the transcription of SNAIL and up-regulated the transcription of SNAIL target genes OCLN, DSP and CLDN9, and (h) restoring L3MBTL3 in L3MBTL3 knockout MDA-MB-231 and SUM159PT cells up-regulated the transcription of SNAIL and down-regulated the transcription of SNAIL target genes OCLN, DSP and CLDN9. i, j WB analyzed showed that (i) knocking out L3MBTL3 in MDA-MB-231 and SUM159PT cells decreased SNAIL expression and (j) restoring L3MBTL3 in L3MBTL3 knockout MDA-MB-231 and SUM159PT cells increased SNAIL expression. Antibody \(\beta\) - ACTIN was used as a loading control. For (d), (g) and (h) panels, bars indicate the mean \(\pm \mathrm{SD}\) , n = 3. All the data were analyzed using Student’s + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 82, 625, 101]]<|/det|> +906 t test. *p < 0.05, **p < 0.01, ***p < 0.005, ****p < 0.001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[170, 135, 850, 405]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 530]]<|/det|> +Fig. 6 L3MBTL3 and SNAIL expression are positive correlation and are associated with low distant metastasis survival. a Scatter plots showing the positive correlation between L3MBTL3 and SNAIL transcription level in TCGA breast cancer datasets (TCGA BRCA 2000, n=1097; and TCGA BRCA 2017, n=1218.) and GEO datasets (GSE202203 (n=3207)). p<0.05. b Distant metastasis survival in breast cancer dataset (GSE12276, n=202) was analyzed according to L3MBLT3 and SNAIL transcription level (L3MBTL3 low SNAIL low, n=49; L3MBTL3 low SNAIL high, n=51; L3MBTL3 high SNAIL low, n=51; L3MBTL3 high SNAIL high, n=53) The Distant metastasis survival of L3MBTL3 high SNAIL high group (n=53) was much lower than L3MBTL3 low SNAIL low group (n=49). p = 0.0035. c Human breast cancer tissue consisting of primary breast cancer samples (Tumor) and paired breast cancer lymph node metastasis samples (Metastasis) were subjected to Immunohistochemistry staining for L3MBLT3 and SNAIL. Immunohistochemistry showed that L3MBTL3 and SNAIL expression were positively correlation in breast cancer tumors (scale bar, 50 μm). + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[165, 102, 222, 118]]<|/det|> +
Fig. 7
+ +<|ref|>image<|/ref|><|det|>[[170, 133, 820, 448]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 852, 899]]<|/det|> +Fig. 7 L3MBTL3 recruits active STAT3 on SNAIL to promote transcription. a Schematic presentation L3MBTL3 and STAT3 binding sites on SNAIL gene promoter. Region P1 (-3704, -3518), P2 (-2185, -2114), P3 (-563, -474), P4 (+2543, +2679) and P5 (+3093, +3240) showed the location of the specific amplicons analyzed by the ChIP assay. b ChIP analyses using a L3MBTL3 antibody or a control IgG were performed in SUM159PT cells. Primers specific for region P1, P2, P3, P4 and P5 on SNAIL gene promoter were used. Data derived from Q-PCR analysis showed region P3 was the key L3MBTL3 binding sites on SNAIL promoter. c ChIP assay showed that both L3MBTL3 and STAT3 binding on region P3 of SNAIL promoter. Knocking out L3MBTL3 decreased the recruitment of p-STAT3Y705 on region P3 of SNAIL promoter in SUM159PT cells. d ChIP assay showed that both L3MBTL3 and STAT3 binding on region P3 of SNAIL promoter. Knocking out L3MBTL3 decreased the recruitment of p-STAT3Y705 on region P3 of SNAIL promoter, and restoring L3MBTL3 expression increased recruitment of p-STAT3Y705 on region P3 of SNAIL promoter in MDA-MB-231 cells. e 293T cells were co-transfected with pGL3-SNAIL (-600, +80) reporter plasmid, STAT3 and L3MBTL3 expression plasmids. Western blot showed L3MBTL3 expression. Antibody \(\beta\) -ACTIN was used as a loading control. Luciferase reporter gene assay showed L3MBTL3 over-expression increased the transcription activity of SNAIL promoter. f 293T cells were cotransfected with pGL3-SNAIL (-600, +80) reporter, STAT3 and L3MBTL3 expression plasmid with treat or untreat STAT3 inhibitor NIF. Western blot showed the L3MBTL3, STAT3 and p-STAT3Y705 level. Antibody \(\beta\) -ACTIN was used as a loading control. Luciferase reporter gene assay showed NIF significantly inhibited the transcription activity increase of SNAIL promoter induced by L3MBTL3. For b-f panels, bars indicate the mean \(\pm \mathrm{SD}\) , \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(^{*}\mathrm{p}< 0.05\) , \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.001\) . + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[166, 102, 223, 119]]<|/det|> +
Fig. 8
+ +<|ref|>image<|/ref|><|det|>[[170, 123, 810, 682]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 536]]<|/det|> +Fig. 8. L3MBTL3 promotes epithelial-mesenchymal transition, migration and invasion dependent on STAT3 in breast cancer. a, b Western blot showed knocking down STAT3 in (a) L3MBTL3 over-expressing MCF7 cells and (b) L3MBTL3 rescue SUM159PT cells decreased SNAIL expression. c, d Knocking down STAT3 in (c) L3MBTL3 over-expressing MCF7 cells and (d) L3MBTL3 rescue SUM159PT cells inhibited cell migration. e Knocking down STAT3 in L3MBTL3 rescue SUM159PT cells inhibited cell invasion. f, g Western blot showed inhibiting STAT3 activity in (f) L3MBTL3 over-expressing MCF7 cells and (g) L3MBTL3 rescue SUM159PT cells by NIF decreased SNAIL expression. h, i Inhibiting STAT3 activity in (h) L3MBTL3 over-expressing MCF7 cells and (i) L3MBTL3 rescue SUM159PT cells by NIF suppressed cell migration. j Inhibiting STAT3 activity in L3MBTL3 rescue SUM159PT cells by NIF suppressed cell invasion. For c, d, e, h, i and j panels, bars indicate the mean±SD, \(\mathrm{n} = 3\) . All the data were analyzed using Student's t test. \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.001\) , ns: no significance. + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[165, 103, 223, 120]]<|/det|> +
Fig. 9
+ +<|ref|>image<|/ref|><|det|>[[212, 120, 800, 789]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 83, 852, 300]]<|/det|> +Fig. 9. Region 176- 193aa is the key site of L3MBTL3 interacting with active STAT3. a GST pull- down assay showed the recombinant L3MBTL3 truncated protein GST- L3- 1 (1- 204aa) directly interacted with p- STAT3Y705. b Co- immunoprecipitation showed that L3MBTL3 interacted with p- STAT3Y705 required L3MBTL3 fragment (1- 204aa). c GST pull- down assay showed that the region 105- 204aa of L3MBTL3 directly interacted with p- STAT3Y705. d GST pull- down assay showed that the region 176- 193aa was the key site of L3MBTL3 interacting with p- STAT3Y705. + +<--- Page Split ---> +<|ref|>image_caption<|/ref|><|det|>[[166, 101, 234, 118]]<|/det|> +
Fig. 10
+ +<|ref|>image<|/ref|><|det|>[[166, 123, 821, 400]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[166, 421, 686, 750]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 82, 852, 640]]<|/det|> +Fig. 10. The interaction between L3MBTL3 and active STAT3 is required for SNAIL expression, cell migration and invasion in breast cancer. A Co- immunoprecipitation showed that deleting region 176- 193aa of L3MBTL3 decreased the interaction between exogenous Flag- L3MBTL3 and exogenous HA- STAT3 in 293T cells. b Co- immunoprecipitation showed that deleting region 176- 193aa of L3MBTL3 decreased the interaction between exogenous Flag- L3MBTL3 and endogenous p- STAT3Y705 in 293T cells. c, d (c) Q- PCR and (d) Western blot showed the transcription and protein level of L3MBTL3 and SNAIL in MCF7 cells and L3MBTL3 knockout SUM159PT cells transfected with vector (V), L3MBTL3 (L3) and L3MBTL3 deletion (L3- Δ176- 193) plasmids. SNAIL expression decreased after deleting region 176- 193aa of L3MBTL3. e Comparing to L3MBTL wild- type, deleting region 176- 193aa of L3MBTL3 decreased the ability of L3MBTL3 promoting cell migration in MCF7 cells and L3MBTL3 knockout SUM159PT cells, and also decreased the ability of L3MBTL3 promoting cell invasion in L3MBTL3 knockout SUM159PT cells. f A schematic of how L3MBTL3 and STAT3 collaboratively up- regulate SNAIL transcription to promote metastasis in breast cancer. For c and e panels, bars indicate the mean±SD, n = 3. All the data were analyzed using Student's t test. \*p < 0.05, \*\*\*p < 0.005, \*\*\*\*p < 0.001. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[85, 85, 435, 103]]<|/det|> +1000 SUPPLEMENTARY DATA + +<|ref|>text<|/ref|><|det|>[[80, 122, 540, 140]]<|/det|> +1001 Supplementary Data are available at NC Online. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 398, 177]]<|/det|> +- supplementaryinformation217.pdf- nrreportingsummary32flattened.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/images_list.json b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d6c4535ff9c608ff392aabb0843d1c508a13b02e --- /dev/null +++ b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/images_list.json @@ -0,0 +1,212 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "FIG. 1: Left: a set of atoms is interpreted as an atomic graph with local neighborhoods. Middle: every atom carries a set of scalar and vector features with it. Right: atoms exchange information via filters, that are again scalars and vectors. The interactions of features and filters define five interactions.", + "footnote": [], + "bbox": [ + [ + 100, + 70, + 916, + 220 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "FIG. 2: The NequIP network architecture. Left: atomic numbers are embedded into \\(l = 0\\) features, which are refined through a series of interaction blocks, creating \\(l = 0\\) and \\(l = 1\\) features. An output block generates atomic energies, which are pooled to give the total predicted energy. Middle: the interaction block consists of a series of convolutions, interweaved with self-interaction layers, equivariant nonlinearities and concatenation. Right: the convolution combines the radial function \\(R(r)\\) which operates only on interatomic distances with the spherical harmonics based on unit vector \\(\\hat{r}\\) via a tensor product.", + "footnote": [], + "bbox": [ + [ + 130, + 74, + 907, + 370 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "FIG. 3: Perspective view of atomic configurations of (a) bidentate HCOO (b) monodentate HCOO and (c) \\(\\mathrm{CO_2}\\) and a hydrogen adatom on a \\(\\mathrm{Cu(110)}\\) surface. The blue, red, black, and white spheres represent Cu, O, C, and H atoms, respectively. The subset shown in each subplot is the corresponding top view along the \\(< 110>\\) orientation.", + "footnote": [], + "bbox": [ + [ + 85, + 670, + 500, + 772 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "FIG. 4: Quenched glass structure of \\(\\mathrm{Li_4P_2O_7}\\) . The insets show the P-O-O tetrahedral bond angle (bottom left) as well as the O-P-P bridging angle between corner-sharing phosphate tetrahedra (top right).", + "footnote": [], + "bbox": [ + [ + 565, + 268, + 858, + 510 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "FIG. 5: Left: Radial Distribution Function, middle: Angular Distribution Function, bridging oxygen, right: Angular Distribution Function, tetrahedral bond angle. All are defined as probability density functions.", + "footnote": [], + "bbox": [ + [ + 86, + 66, + 914, + 185 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "FIG. 6: Comparison of Lithium mean square displacement of AIMD and NequIP trajectories.", + "footnote": [], + "bbox": [ + [ + 100, + 266, + 470, + 450 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "FIG. 7: Log-log plot of the predictive error in forces of NeuIP with \\(l = 0\\) vs. \\(l = 0 / l = 1\\) interactions as a function of data set size, measured via the force MAE.", + "footnote": [], + "bbox": [ + [ + 84, + 528, + 500, + 689 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 45, + 95, + 945, + 260 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 55, + 409, + 945, + 750 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 44, + 52, + 955, + 270 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 58, + 55, + 905, + 767 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 55, + 50, + 940, + 180 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 50, + 294, + 943, + 733 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7", + "footnote": [], + "bbox": [ + [ + 44, + 48, + 950, + 415 + ] + ], + "page_idx": 26 + } +] \ No newline at end of file diff --git a/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb.mmd b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb.mmd new file mode 100644 index 0000000000000000000000000000000000000000..07997b5fe06e38473b9161005ee2317c74818177 --- /dev/null +++ b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb.mmd @@ -0,0 +1,514 @@ + +# SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials + +Simon Batzner Harvard University + +Tess Smidt Lawrence Berkeley National Laboratory https://orcid.org/0000- 0001- 5581- 5344 + +Lixin Sun Massachusetts Institute of Technology + +Jonathan Mailoa Bosch Research and Technology Center + +Mordechai Kombluth Robert Bosch Research and Technology Center https://orcid.org/0000- 0001- 6705- 8133 + +Nicola Molinari Harvard University + +Boris Kozinsky ( bkoz@seas.harvard.edu) Harvard University https://orcid.org/0000- 0002- 0638- 539X + +## Article + +Keywords: Molecular Dynamics Simulations, Symmetry- aware Models, Equivariant Convolutions, Geometric Tensors + +Posted Date: March 1st, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 244137/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on May 4th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29939- 5. + +<--- Page Split ---> + +# SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials + +Simon Batzner, \(^{1}\) Tess E. Smidt, \(^{2}\) Lixin Sun, \(^{1}\) Jonathan P. Mailoa, \(^{3}\) + +Mordechai Kornbluth, \(^{3}\) Nicola Molinari, \(^{1}\) and Boris Kozinsky \(^{1,3}\) + +\(^{1}\) John A. Paulson School of Engineering and Applied Sciences, + +Harvard University, Cambridge, MA 02138, USA + +\(^{2}\) Computational Research Division and Center for Advanced Mathematics for Energy Research Applications, + +Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA + +\(^{3}\) Robert Bosch Research and Technology Center, Cambridge, MA 02139, USA + +This work presents Neural Equivariant Interatomic Potentials (NequlP), a SE(3)- equivariant neural network approach for learning interatomic potentials from ab- initio calculations for molecular dynamics simulations. While most contemporary symmetry- aware models use invariant convolutions and only act on scalars, NequlP employs SE(3)- equivariant convolutions for interactions of geometric tensors, resulting in a more information- rich and faithful representation of atomic environments. The method achieves state- of- the- art accuracy on a challenging set of diverse molecules and materials while exhibiting remarkable data efficiency. NequlP outperforms existing models with up to three orders of magnitude fewer training data, challenging the widely held belief that deep neural networks require massive training sets. The high data efficiency of the method allows for the construction of accurate potentials using high- order quantum chemical level of theory as reference and enables high- fidelity molecular dynamics simulations over long time scales. + +## INTRODUCTION + +Molecular dynamics (MD) simulations are an indispensable tool for computational discovery in fields \(^{37}\) as diverse as energy storage, catalysis, and biological processes [1- 3]. While the atomic forces required to integrate Newton's equations of motion can in principle be obtained with high fidelity from quantum- mechanical calculations such as density functional theory (DFT), in practice the unfavorable computational scaling of first- principles methods limits simulations to short time scales and small numbers of atoms. This prohibits the study of many interesting physical phenomena beyond the time and length scales that are currently + +accessible, even on the largest supercomputers. Owing to their simple functional form, classical models for the atomic potential energy can typically be evaluated orders of magnitude faster than using first- principles methods, thereby enabling the study of large numbers of atoms over long time scales. However, due to their limited mathematical form, classical interatomic potentials, or force fields, are inherently limited in their predictive accuracy which has historically led to a fundamental trade- off between obtaining high computational efficiency while also predicting faithful dynamics of the system under study. The construction of flexible models of the interatomic potential energy based on Machine Learning (ML- IP), and in particular + +<--- Page Split ---> + +48 Neural Networks (NN- IP), has shown great promise in a providing a way to move past this dilemma, promising a to learn high- fidelity potentials from ab- initio reference calculations while retaining favorable computational efficiency [4- 13]. One of the limiting factors of NN- IPs is that they typically require collection of large training sets of ab- initio calculations, often including thousands or even millions of reference structures [4, 9, 10, 14- 16]. This computationally expensive process of training data collection has severely limited the adoption of NN- IPs as it quickly becomes a bottleneck in the development of force- fields for new systems. Kernel- based approaches, such as e.g. Gaussian Processes (GP) [5, 8] or Kernel Ridge Regression (KRR) [17], are a way to remedy this problem as they often generalize better from limited sample sizes. However, such methods generally tend to exhibit poor computational scaling with the number of reference configurations, in both training (cubic in training set size) and prediction (linear in training set size). This limits both the amount of training data they can be trained on as well as the length and size of simulations that can be simulated with them. + +accurate interatomic potentials from limited data sets of fewer than 1,000 or even as little as 100 reference ab- initio calculations, where other methods require orders of magnitude more. It is worth noting that on small molecular data sets, NequIP outperforms not only other neural networks, but is also competitive with kernel- based approaches, which typically obtain better predictive accuracy than NN- IPs on small data sets (although at significant additional cost scaling in training and prediction). We further demonstrate high data efficiency and accuracy with state- of- the- art results on a training set of molecular data obtained at the quantum chemical coupled- cluster level of theory. Finally, we validate the method through a series of simulations and demonstrate that we can reproduce with high fidelity structural and kinetic properties computed from NequIP simulations in comparison to ab- initio molecular dynamics simulations (AIMD). We directly verify that the performance gains are connected with the unique SE(3)- equivariant convolution architecture of the new NequIP model. + +70 + +In this work, we present the Neural Equivariant Interatomic Potential (NequIP), a highly data- efficient + +deep learning approach for learning interatomic potentials from reference first- principles calculations. We show that the proposed method obtains high- accuracy compared to existing ML- IP methods across a wide variety of systems, including small molecules, water in different phases, an amorphous solid, a reaction at a solid/gas interface, and a Lithium superionic conductor. Furthermore, we find that NequIP exhibits exceptional data efficiency, enabling the construction of + +## Related Work + +First applications of machine learning for the development of interatomic potentials were built on descriptor- based approaches combined with shallow neural networks or Gaussian Processes [4, 5], designed to exhibit invariance with respect to translation, permutation of atoms of the same chemical species, and rotation. Recently, rotationally invariant graph neural networks (GNN- IPs) have emerged as a powerful architecture for deep learning of interatomic potentials + +<--- Page Split ---> + +that eliminates the need for hand- crafted descriptors and allows to instead learn representations on graphs of atoms from invariant features of geometric data (e.g. radial distances or angles) [9- 11, 13]. In GNN- IPs, atomic structures are represented by collections of nodes and edges, where nodes in the graph correspond to individual atoms and edges are typically defined by simply connecting every atom to all other atoms that are closer than some cutoff distance \(r_c\) . Node/atom \(i\) is associated with a feature \(\mathbf{h}_i \in \mathbb{R}^h\) , consisting of scalar values, which is iteratively refined via a series of convolutions over neighboring atoms based on both the distance to neighboring atoms and their features \(\mathbf{h}_j\) . This iterative process allows information to be propagated along the atomic graph through a series of convolutional layers and can be viewed as a message- passing scheme [18]. Operating only on interatomic distances allows GNN- IPs to rotation- and translation- invariant, making both output as well as features internal to the network invariant to rotations. In contrast, the method outlined in this work uses relative position vectors rather than simply distances (scalars), which makes internal features instead equivariant to rotation and allows for angular information to be used by rotationally equivariant filters. Similar to other methods, we can restrict convolutions to only a local subset of all other atoms that lie closer to the central atom than a chosen cutoff distance \(r_c\) , see Figure 1, left. + +A series of related methods have recently been proposed: DimeNet [11] expands on using pairwise interactions in a single convolution to include angular, three- body terms, but individual features are still + +comprised of scalars (distances and three- body angles are invariant to rotation), as opposed to vectors used in this work. Another central difference to NequIP is that DimeNet explicitly enumerates angles between pairs of atoms and operates on a basis embedding of distances and angles, whereas NequIP operates on relative position vectors and a basis embedding of distances, and thus never explicitly computes three- body angles. Cormorant [19] uses an equivariant neural network for property prediction on small molecules. This method is demonstrated on potential energies of small molecules but not on atomic forces or systems with periodic boundary conditions. Townshend et al. [20] use the framework of Tensor- Field Networks [21] to directly predict atomic force vectors. The predicted forces are not guaranteed by construction to conserve energy since they are not obtained as gradients of the total potential energy. This may lead to problems in simulations of molecular dynamics over long times. None of these three works [11, 19, 20] demonstrates capability to perform molecular dynamics simulations. + +In this work we present a deep learning energy- conserving interatomic potential for both molecules and materials built on SE(3)- equivariant convolutions over geometric tensors that yields state- of- the- art accuracy, outstanding data- efficiency, and can with high fidelity reproduce structural and kinetic properties from molecular dynamics simulations. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
FIG. 1: Left: a set of atoms is interpreted as an atomic graph with local neighborhoods. Middle: every atom carries a set of scalar and vector features with it. Right: atoms exchange information via filters, that are again scalars and vectors. The interactions of features and filters define five interactions.
+ +## RESULTS + +Equivariance + +The concept of equivariance arises naturally in machine learning of atomistic systems (see e.g. [22]): physical properties have well- defined transformation properties under translation and rotation of a set of atoms. As a simple example, if a molecule is rotated in space, the vectors of its atomic dipoles or forces also rotate accordingly, via equivariant transformation. Equivariant neural networks are able to more generally represent tensor properties and tensor operations of physical systems (e.g. vector addition, dot products, and cross products). Equivariant neural networks are guaranteed to preserve the known transformation properties of physical systems under a change of coordinates because they are explicitly constructed from equivariant operations. Formally, a function \(f: X \to Y\) is equivariant with respect to a group \(G\) that acts on \(X\) and \(Y\) if: + +where \(D_{X}[g]\) and \(D_{Y}[g]\) are the representations of the group element \(g\) in the vector spaces \(X\) and \(Y\) , respectively. In this work, we focus on equivariance with respect to \(\mathrm{SE}(3)\) , i.e. the group of rotations and translations in 3D space. + +## Neural Equivariant Interatomic Potentials + +Given a set of atoms (a molecule or a material), we aim to find a mapping from atomic positions \(\vec{r}_{i}\) and chemical species (identified by atomic numbers \(Z_{i}\) ) to the total potential energy and the forces acting on the atoms: + +\[f:\{\vec{r}_{i},Z_{i}\} \to E_{pot} \quad (2)\] + +Forces are obtained as gradients of the predicted potential energy with respect to the atomic positions, which guarantees energy conservation: + +\[\vec{F}_{i} = -\nabla_{i}E_{pot} \quad (3)\] + +Gradients can be obtained with relatively low computational overhead in modern auto- differentiation frameworks such as TensorFlow or PyTorch [23, 24]. Following previous work [4], we further define the total + +<--- Page Split ---> + +potential energy of the system as a sum of atomic potential energies: + +\[E_{p o t} = \sum_{i\in N_{a t o m s}}E_{i,a t o m i c} \quad (4)\] + +These atomic local energies \(E_{i,atomic}\) are the scalar node attributes predicted by the graph neural network. Even though the output of NequIP is the predicted potential energy \(E_{pot}\) , which is invariant under translations and rotations, the network contains internal features that are tensors which are equivariant to rotation. This constitutes the core difference between NequIP and existing scalar- valued invariant GNN- IPs. The remainder of this section will discuss the design of the network in further detail. + +A series of methods has been introduced to realize + +A series of methods has been introduced to realize rotationally equivariant neural networks [13, 21, 25, 26]. Here, we build on the layers introduced in Tensor- Field Networks (TFN) [21], which enable the construction of neural networks that exhibit equivariance to translation, permutation, and rotation. Every atom in NequIP is associated with a feature comprised of tensors of different order: scalars, vectors, and higher- order tensors. Formally, these features are geometric objects that comprise a direct sum of irreducible representations of the SO(3) symmetry group. Second, the convolutions that operate on these geometric objects are equivariant functions instead of invariant ones, i.e. if a feature layer \(k\) is rotated, then the output of the convolution from layer \(k \to k + 1\) rotates accordingly. In practice, the features are implemented as a dictionary \(V_{acm}^{(l)}\) with keys \(l\) , where \(l = 0, 1, 2, \ldots\) is a non- negative integer and is called the "rotation order", labeling the irreducible + +representations. The indices \(a\) , \(c\) , \(m\) , correspond to the atoms, the channels (elements of the feature), and the representation index which takes values \(m \in [- l, l]\) , respectively. + +Convolution operations are naturally translation invariant, since their filters act on relative interatomic distance vectors. Moreover, they are permutation invariant since all convolution contributions are summed. Note that while atomic features are equivariant to permutation of atom indices, globally, the total potential energy of the system is invariant to permutation. To achieve rotation equivariance, the convolution filters are constrained to be products of learnable radial functions and spherical harmonics, which are equivariant under SO(3) [21]: + +\[F(\vec{r}_{ij}) = R(r_{ij})Y_{m}^{(l)}(\hat{r}_{ij}) \quad (5)\] + +where if \(\vec{r}_{ij}\) denotes the relative position from central atom \(i\) to neighboring atom \(j\) , \(\hat{r}_{ij}\) and \(r_{ij}\) are the associated unit vector and interatomic distance, respectively. It should be noted that all learnable weights in the filter lie in the rotationally invariant radial function \(R(r_{ij})\) . This radial function is implemented as a small neural network with one hidden layer and a shifted softplus activation function [9], operating on interatomic distances expressed in a basis of choice, \(R(r_{ij}) : \mathbb{R}^{N_{b}} \to \mathbb{R}^{h}\) , where \(N_{b}\) is the number of basis functions and \(h\) is the feature dimension: + +\[R(r_{ij}) = W_{2}\ln (0.5\exp (W_{1}B(r_{ij})) + 0.5) \quad (6)\] + +<--- Page Split ---> + +where \(W_{1} \in \mathbb{R}^{N_{hidden} \times N_{b}}\) and \(W_{2} \in \mathbb{R}^{h \times N_{hidden}}\) are 296 weight matrices, \(h\) is the dimension of the feature and 298 \(N_{hidden}\) is the dimension of the hidden layer in the 300 feed- forward neural network (in our experiments, we use 301 \(N_{hidden} = N_{b}\) , resulting in comparatively small neural 302 networks for the radial function). Radial Bessel functions 303 and a polynomial envelope function \(f_{env}\) discussed in 304 recent work [11] are used to expand the interatomic 305 distances: + +where \(r_{c}\) is a local cutoff radius, restricting interactions 310 to atoms closer than some cutoff distance and \(f_{env}\) is the polynomial defined in [11] with \(p = 6\) operating on the interatomic distances normalized by the cutoff radius \(r_{c}^{i,j}\) . The use of cutoffs/local atomic environments allows the computational cost of evaluation to scale linearly with the number of atoms. Similar to [11], we initialize the Bessel functions with \(n = [1,2,\dots,N_{b}]\) and subsequently optimize \(n\pi\) via backpropagation rather than keeping it constant. For systems with periodic boundary conditions, we use the neighbor list functionality as implemented in the ASE code [27] to identify appropriate atomic neighbors and then convolve over them. + +Finally, in the convolution, the input feature tensor 294 and the filter have to again be combined in an 295 equivariant manner, which is achieved via a geometric 296 + +tensor product, yielding an output feature that again is rotationally equivariant. A tensor product of two geometric tensors is computed via Clebsch- Gordan coefficients, as outlined in [21]. Since NequIP deals with force vectors, the network design is simplified by only using scalar (l=0) and vector (l=1) representations. Thus, we can enumerate five distinct products or "interactions" between \(l = 0\) and \(l = 1\) filters and \(l = 0\) and \(l = 1\) features that correspond to simple operations between scalars and vectors: + +\(\bullet 0\otimes 0\to 0\) (product of two scalars) \(\bullet 0\otimes 1\to 1\) (scalar multiplication of a vector) \(\bullet 1\otimes 0\to 1\) (scalar multiplication of a vector) \(\bullet 1\otimes 1\to 0\) (dot product of two vectors) \(\bullet 1\otimes 1\to 1\) (cross product of two vectors) + +It is trivial to include higher- order interactions, and previous works have increased the rotation order beyond \(l = 1\) [20, 28]. However, it should be noted that every interaction comes with additional trainable radial functions and hence additional weights, which thus adds to the model capacity, increasing the number of model weights and the memory footprint of the model. Omitting all higher- order interactions that go beyond the \(0\otimes 0\to 0\) interaction will result in a conventional GNN- IP with invariant convolutions over scalar features, similar to e.g. SchNet [9]. Finally, as outlined in [21], a full convolutional layer \(\mathcal{L}\) implementing an interaction with filter \(f\) acting on an input \(i\) producing output \(o\) : \(l_{f}\otimes l_{i}\to l_{o}\) is given by: + +<--- Page Split ---> + +\[\mathcal{L}_{acm_o}^{(l_o)}(\vec{r}_a,V_{acm_i}^{(l_i)}) = \sum_{m_f,m_i}C_{(l_f,m_f)(l_i,m_i)}^{(l_o,m_o)}\sum_{b\in S}R_c^{(l_f,l_i)}(r_{ab})Y_{m_f}^{(l_f)}(\hat{r}_{ab})V_{bcm_i}^{(l_i)} \quad (8)\] + +where \(a\) and \(b\) index the central atom of the convolution and the neighboring atom \(b \in S\) , respectively, and \(C\) indicates the Clebsch- Gordan coefficients. As an example of this, we write out a full \(1 \otimes 1 \to 1\) operation (corresponding to a cross- product) in the Methods section. After every convolution, output tensors of \(a_{362}\) rotation order \(l\) stemming from different tensor products are concatenated on a per- atom basis. To update atomic features, the model also leverages self- interaction layers similar to SchNet [9], corresponding to dense layers that are applied in an atom- wise fashion with weights shared across atoms. While different weights are used for different rotation orders, the same set of weights is applied for all representation indices \(m\) of a given rotation order \(l\) . Shifted softplus nonlinearities [9] are used as rotation- equivariant nonlinearities introduced in [21], which are applied to the Euclidean norm of the input feature, the output of which is in turn combined with the input tensor, thus preserving overall equivariance. + +The NequIP network architecture, shown in Figure 2, is built on an atomic embedding, followed by a series of interaction blocks, and finally an output block: + +Embedding: following SchNet, the initial feature is generated using a trainable embedding, that operates on the atomic number \(Z_{i}\) (represented via a one- hot encoding) alone, implemented via a trainable self- interaction layer. + +Interaction Block: interaction blocks encodes interactions between neighboring atoms: the core of this block is the convolution function, outlined in equation 8. For every output rotation order \(l_{o}\) , the features from different tensor product interactions are concatenated to give a new feature, which is in return refined with atom- wise self- interaction layers and equivariant non- linearities. We equip interactions blocks with a ResNet- style update [29] where the input feature \(\mathbf{x}\) is updated atom- wise via the output of an interaction block \(f(\mathbf{x})\) that gives the final feature \(r(\mathbf{x}) = f(\mathbf{x}) + \mathbf{x}\) (features are added element- wise in the \(m\) - dimension). Note that this operation is equivariant since the addition of an equivariant feature \(\mathbf{x}\) and an equivariant function \(f(\mathbf{x})\) preserves equivariance. While later interaction blocks include all five interactions outlined above, the first interaction block operates on the \(l = 0\) embedding with a \(0 \otimes 0 \to 0\) and a \(0 \otimes 1 \to 1\) only. + +interactions between neighboring atoms: the core of this block is the convolution function, outlined in equation 8. For every output rotation order \(l_{o}\) , the features from different tensor product interactions are concatenated to give a new feature, which is in return refined with atom- wise self- interaction layers and equivariant non- linearities. We equip interactions blocks with a ResNet- style update [29] where the input feature \(\mathbf{x}\) is updated atom- wise via the output of an interaction block \(f(\mathbf{x})\) that gives the final feature \(r(\mathbf{x}) = f(\mathbf{x}) + \mathbf{x}\) (features are added element- wise in the \(m\) - dimension). Note that this operation is equivariant since the addition of an equivariant feature \(\mathbf{x}\) and an equivariant function \(f(\mathbf{x})\) preserves equivariance. While later interaction blocks include all five interactions outlined above, the first interaction block operates on the \(l = 0\) embedding with a \(0 \otimes 0 \to 0\) and a \(0 \otimes 1 \to 1\) only. + +Output Block: the \(l = 0\) feature of the final convolution is passed to an output block, which consists of another atom- wise self- interaction layer, an equivariant non- linearity, and a final atom- wise self- interaction layer. + +The scalar atomic outputs of the final layer can be interpreted as atomic potential energies which are summed to give the total predicted potential energy of the system (Equation 4). Forces are subsequently obtained as the negative gradient of the predicted total potential energy, thereby ensuring both energy + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
FIG. 2: The NequIP network architecture. Left: atomic numbers are embedded into \(l = 0\) features, which are refined through a series of interaction blocks, creating \(l = 0\) and \(l = 1\) features. An output block generates atomic energies, which are pooled to give the total predicted energy. Middle: the interaction block consists of a series of convolutions, interweaved with self-interaction layers, equivariant nonlinearities and concatenation. Right: the convolution combines the radial function \(R(r)\) which operates only on interatomic distances with the spherical harmonics based on unit vector \(\hat{r}\) via a tensor product.
+ +387 conservation as well as rotation- equivariant forces (see equation 3). + +## Experiments + +390 We validate the proposed method on a series of diverse 391 and challenging data sets: first we demonstrate that we 392 improve upon state- of- the- art accuracy on MD- 17, a data 393 set of small, organic molecules that is widely used for 394 benchmarking ML- IPs [9, 11, 17, 30, 31]. Next, we show 395 that NequIP can also accurately learn forces obtained on 396 small molecules at the quantum chemical CCSD(T) level 397 [31], opening the door to scalable and efficient molecular 398 dynamics simulations with beyond- DFT accuracy. To 399 broaden the applicability of the method beyond small 400 isolated molecules, we explore a series of extended 401 + +systems with periodic boundary conditions, consisting of both surfaces and bulk materials: water in different phases [15, 32], a chemical reaction at a solid/gas interface, an amorphous Lithium Phosphate [12], and a Li superionic conductor [13]. Details of the training procedure are provided in the Methods section. + +## MD-17 small molecule dynamics + +We first evaluate NequIP on MD- 17 [17, 30, 31], a data set of eight small organic molecules in which reference values of energy and forces are generated by ab- initio MD simulations with DFT. For training we use N=1,000 structure configurations for each molecule, sampled uniformly from the full data set, the same number of configurations for validation, and evaluate + +<--- Page Split ---> + +the test error on all remaining configurations in the data set. The mean absolute error in the force components is shown in Table I in units of [meV/A]. We compare results using NequIP with those from published leading ML- IP models that were also trained on 1,000 structures: in particular SchNet [9], DimeNet [11] (both graph neural networks), sGDML [31], and FCHL19/GPR (kernel- based methods) [33]. We find that NequIP outperforms SchNet and sGDML on all molecules in the data set, DimeNet on 7 out of 8 molecules (on par on the remaining one), and performs on par with FCHL/19GPR. The consistent improvement in accuracy upon sGDML and the comparable performance to FCHL19/GPR are particularly surprising, as these are based on kernel methods, that typically tend to be more sample efficient. It should be noted, however, that the evaluation cost of kernel methods scales linearly with the number of training configurations. Note also that some molecules, NequIP trained on 1,000 configurations even performs as well as SchNet trained on 50,000 structures [9]: on aspirin and naphthalene, for example, the NequIP network trained on 1,000 structures produces mean absolute errors in the forces of 15.1 meV/A and 4.2 meV/A, respectively, compared to 14.3 meV/A and 4.8 meV/A of SchNet trained on 50x more molecules, hinting that NequIP exhibits exceptional data efficiency. On other molecules such as ethanol, however, SchNet trained with 50,000 molecules still clearly outperforms NequIP trained with 1,000 molecules (2.2 meV/A for SchNet for N=50,000 vs 9.0 meV/A for NequIP for N=1,000). + +Ability to achieve high accuracy on a comparatively small data set opens the door to training models on expensive high- order ab- initio quantum chemical methods. It has been shown that DFT can fail to capture important subtleties in the potential energy surface, potentially even identifying the wrong ground states [31]. This problem can be remedied through the use of more accurate reference calculations, such as coupled cluster methods CCSD(T), typically regarded as the gold standard of quantum chemistry. However, the high computational cost of CCSD(T) has thus far hindered the use of reference data structures at this level of theory, prohibited by the need for large data sets that are required by available NN- IPs. Leveraging the high data efficiency of NequIP, we evaluate it on a set of molecules computed at quantum chemical accuracy (aspirin at CCSD, all others at CCSD(T)) [31] and compare the results to those reported for sGDML [31]. The training/validation set consists of a total of 1,000 molecular structures which we split into 950 for training and 50 for validation (sampled uniformly), and we test the accuracy on all remaining structures (we use the train/test split provided with the data set, but further split the training set into training and validation sets). We find that NequIP achieves lower errors on four out of five molecules, performing on par with sGDML on the fifth molecule, as shown in Table II. + +To demonstrate the applicability of NequIP beyond small molecules, we evaluate the method on a series of + +<--- Page Split ---> + + +
MoleculeNequIPSchNetsGDMLDimeNetFCHL19/GPR
Aspirin15.158.529.521.620.7
Benzene [17]8.113.4n/a8.1n/a
Benzene [31]2.3n/a2.6n/an/a
Ethanol9.016.914.310.05.9
Malonaldehyde14.628.617.816.610.6
Naphthalene4.225.24.89.36.5
Salicylic Acid10.336.912.116.29.6
Toluene4.424.76.19.48.8
Uracil7.524.310.413.14.6
+ +TABLE I: MAE of force components on the MD-17 data set, trained on 1,000 configurations, forces in units of [meV/A]. For the benzene molecule, two different data set exists from [17], [31] with different levels of accuracy in the DFT reference data. + +
MoleculeNequIPsGDML
Aspirin14.733.0
Benzene0.81.7
Ethanol9.415.2
Malonaldehyde16.016.0
Toluene4.49.1
+ +TABLE II: Force MAE for molecules at CCSD/CCSD(T) accuracy, reported in units of [meV/A], with 1,000 reference configurations). + +476 extended systems with periodic boundary conditions. As + +477 a first example we use a joint data set consisting of liquid + +478 water and three ice structures [15,32], computed at the + +479 PBE0-TS level of theory. This data set contains [15]: + +480 a) liquid water, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=300\mathrm {K},\) computed via path- + +481 integral AIMD, b) ice Ih, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=273\mathrm {K},\) computed + +482 via path-integral AIMD c) ice Ih, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=330\mathrm {K},\) computed via classical AIMD d) ice Ih, \(\mathrm {P}=2.13\mathrm {kbar},\) + +484 \(\mathrm {T}=238\mathrm {K},\) computed via classical AIMD. The liquid water + +485 system consists of 64 \(\mathrm {H}_{2}\mathrm {O}\) molecules (192 atoms), while + +486 the ice structures consist of 96 \(\mathrm {H}_{2}\mathrm {O}\) molecules (288 + +487 atoms). A DeepMD NN-IP model was previously trained + +488 [15] for water and ice using a joint training set containing + +489 133,500 reference calculations of these four systems. Tos + +490 assess data efficiency of the NequIP architecture, wes + +491 similarly train a model jointly on all four parts of the data + +492 set, but using only 133 structures for training, i.e. 1000x + +498 fewer data. The 133 structures were sampled uniformly from the full data set available online, consisting of water and ice structures, made up of a total of 140,000 frames, coming from the same MD trajectories that were used in the earlier work [15]. We also use a validation set of 50 frames and report the test accuracy on all remaining structures in the data set. Table III shows the comparison of the predictive force accuracy of NequIP trained on the 133 structures vs DeepMD trained on 133,500 structures. We find that with 1000x fewer training data, NequIP significantly outperforms DeepMD on all four parts of the data set. + +Heterogeneous catalysis of formate dehydrogenation + +Next, we demonstrate application of NequIP to a catalytic surface reaction. In particular, we investigate the dynamics of formate undergoing dehydrogenation decomposition \(\left(\mathrm {HCOO}^{*}\rightarrow \mathrm {H}^{*}+\mathrm {CO}_{2}\right)\) on a Cu \(<110>\) surface (see Figure 3). This system is highly heterogeneous, with both metallic and covalent types of bonding as well as charge transfer occurring between the metal and the molecule, making this a particularly challenging test system. Different states of the molecule + +<--- Page Split ---> + + +
SystemNequIP, 133 data pointsDeepMD, 133,500 data points
Liquid Water35.940.4
Ice Ih (b)25.943.3
Ice Ih (c)16.626.8
Ice Ih (d)13.525.4
+ +TABLE III: Root mean square error (RMSE) of force components on liquid water and the three ices in units of [meV/A]. Note that the NequIP model was trained on \(< 0.1\%\) of the training data of DeepMD. + +
ElementMAE
C55.8
O86.7
H42.0
Cu54.5
Total structure55.6
+ +TABLE IV: MAE of force components for Formate on Cu system, per- element basis. The training set consists of 2,500 structures, force units are [meV/A] + +also lead to dissimilar C- O bond lengths [34, 35]. Training structures consist of 48 Cu atoms and 4 atoms of the molecule (HCOO\\* or \(\mathrm{CO_2 + H^*}\) ). The MAE of the predicted forces using a NequIP model trained on 2,500 structures is shown in Table IV, demonstrating that NequIP is able to accurately model the interatomic forces for this complex reactive system. A more detailed analysis of the resulting dynamics will be subject of a separate study. + +![](images/Figure_3.jpg) + +
FIG. 3: Perspective view of atomic configurations of (a) bidentate HCOO (b) monodentate HCOO and (c) \(\mathrm{CO_2}\) and a hydrogen adatom on a \(\mathrm{Cu(110)}\) surface. The blue, red, black, and white spheres represent Cu, O, C, and H atoms, respectively. The subset shown in each subplot is the corresponding top view along the \(< 110>\) orientation.
+ +Lithium Phosphate Amorphous Glass Formation + +![](images/Figure_4.jpg) + +
FIG. 4: Quenched glass structure of \(\mathrm{Li_4P_2O_7}\) . The insets show the P-O-O tetrahedral bond angle (bottom left) as well as the O-P-P bridging angle between corner-sharing phosphate tetrahedra (top right).
+ +To examine the ability of the model to capture dynamical properties, we demonstrate that NequIP can describe structural dynamics in amorphous lithium phosphate with composition \(\mathrm{Li_4P_2O_7}\) . This material is a member of the promising family of solid electrolytes for Li- metal batteries [12, 36, 37], with non- trivial Li- ion transport and phase transformation behaviors. The training data set consists of two 50ps- long AIMD simulations, one of the molten structure at \(\mathrm{T} = 3000\) K, followed by another of a quenched glass structure at \(\mathrm{T} = 600\) K. We train NequIP on a subset of 1,000 structures of the molten trajectory, each consisting of + +<--- Page Split ---> + +208 atoms, and sampled uniformly from the full data set of 25,000 AIMD frames. We use a validation set of 100 structures, and evaluate the model on all remaining structures. Table V shows the test set error in the force components on both the test set from the AIMD molten trajectory and the full AIMD quenched glass trajectory. To then evaluate the physical fidelity of the trained model, we use it to run a MD simulation of length ps at T=600 K in the NVT ensemble and compare the total radial distribution function (RDF) without element distinction as well as the angular distribution functions (ADF) of the P-O-O (P central atom) and O-P-P (central atom) angles to the ab-into trajectory at the same temperature. The P-O-O angle corresponds to the tetrahedral bond angle, while the O-P-P corresponds to a bridging angle between corner-sharing phosphates tetrahedra (Figure 4). Figure 5 shows that NequIP can accurately reproduce the RDF and the two ADFs, in comparison with AIMD, after training on only 1,000 structures. This demonstrates that the model generates the glass state and recovers its dynamics and structures almost perfectly, having seen only the high-temperature molten training data. + +data set of 30,874 structures to study Li diffusion in \(\mathrm{Li}_3\mathrm{PO}_4\) . Here we demonstrate that not only does NequIP obtain small errors in the force components, but it also accurately predicts the diffusivity after training on a data set obtained from an AIMD simulation. Again, we find that very small training sets lead to highly accurate models, as shown in Table V for training set sizes of 10, 100, 1,000 and 2,500 structures. We run a series of MD simulations with the NequIP potential trained on 2,500 structures in the NVT ensemble at the same temperature as the AIMD simulation for a total simulation time of 50 ps and a time step of 0.25 fs, which we found advantageous for reliability and stability of long simulations. We measure the Li diffusivity in ten Nequip-driven MD simulations (computed via the slope of the mean square displacement), all of length 50 ps and started from different initial velocities, randomly sampled from a Maxwell-Boltzmann distribution. We find a mean diffusivity of \(1.42 \times 10^{-5} \mathrm{~cm}^2 / \mathrm{s}\) , in excellent agreement with the diffusivity of \(1.38 \times 10^{-5} \mathrm{~cm}^2 / \mathrm{s}\) computed from AIMD, thus achieving a relative error of as little as 3%. Figure 6 shows the mean square displacements of Li for an example run. + +Lithium Thiophosphate Superionic Transport + +
SystemData Set SizeMAE
LiPS10157.1
LiPS10050.0
LiPS1,00025.1
LiPS2,50024.1
Li4P2O7, melt1,00063.2
Li4P2O7, quench1,00036.9
+ +To show that NequIP can model kinetic transport properties from small training sets at high accuracy, we study Li- ion diffusivity in LiPS ( \(\mathrm{Li}_{6.75}\mathrm{P}_3\mathrm{S}_{11}\) ) a crystalline superionic Li conductor, consisting of a simulation cell of 83 atoms [13]. MD is widely used to study diffusion; however, training a ML- IP to the accuracy required to accurately predict kinetic properties has in the past required large training set sizes ([38] e.g. uses a + +TABLE V: Force MAE for LiPS and \(\mathrm{Li}_4\mathrm{P}_2\mathrm{O}_7\) for different data set sizes in units of [meV/A]. The model for \(\mathrm{Li}_4\mathrm{P}_2\mathrm{O}_7\) was trained exclusively on structures from the melted trajectory, the reported test errors show the MAE on both the test set of the melted trajectory as well as the full quench trajectory. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
FIG. 5: Left: Radial Distribution Function, middle: Angular Distribution Function, bridging oxygen, right: Angular Distribution Function, tetrahedral bond angle. All are defined as probability density functions.
+ +![](images/Figure_6.jpg) + +
FIG. 6: Comparison of Lithium mean square displacement of AIMD and NequIP trajectories.
+ +## Data Efficiency + +In the above experiments, NequIP exhibits exceptionally high data efficiency, i.e. it can be trained successfully to state- of- the- art accuracy from \(\mathrm{62}\) unexpectedly small training sets. It is interesting to consider the reasons for such high performance and verify that it is connected to the equivariant nature of the \(\mathrm{62}\) model. First, it is important to note that each training configuration contains multiple labels, thus increasing the total number of labels available beyond just the potential energy label associated with each structure. In particular, for a training set of \(M\) first- principles calculations with structures consisting of \(N\) atoms, the total number of labels available is \(M(3N + 1)\) since every + +force component on every atom constitutes a label and so does the total energy of the reference calculation (we only train to atomic forces and not energies, thus using \(3MN\) force components as labels). + +In order to gain insight into the reasons behind increased accuracy and data efficiency, we perform a series of experiments with the goal of isolating the effect of using equivariant convolutions of geometric tensors compared to invariant convolutions over scalars. In particular, we run a set of experiments for a system with a fixed number of training configurations in which we explicitly turn on or off interactions of higher order than \(l = 0\) . This defines two settings: first, we train the network with both \(l = 0\) and \(l = 1\) features and all five interactions as previously outlined in this work. Second, when all interactions involving \(l = 1\) are turned off, this turns the network into a conventional invariant GNN- IP, involving only invariant convolutions over scalar features in a SchNet- style fashion. As a test system we chose bulk water: in particular we use the data set introduced in [39], consisting of 1,593 bulk liquid water structures with 64 water molecules each. We train a series of networks with identical hyperparameters, but vary the training set sizes between 10 and 1,000 structures, sampled uniformly from the full data set, as well as a validation set consisting of 100 structures. We then evaluate the error on all + +<--- Page Split ---> + +remaining structures for a given training set size. As shown in Figure 7, we find that the equivariant setting (using \(l = 0\) and \(l = 1\) ) significantly outperforms the invariant setting (using only \(l = 0\) ) for all data set sizes as measured by the MAE of force components. This suggests that it is indeed the use of tensor (in our specific case vector) features and equivariant convolutions that enables the high data efficiency of NeuIP. We further note, that in [39], a Behler-Parrinello Neural Network (BPNN) was trained on 1303 structures, yielding RMSE of \(\approx 120 \mathrm{meV / \AA}\) in forces when evaluated the remaining 290 structures. We find that NeuIP models trained with as little as 50 and 100 data points obtain RMSEs of \(122.9 \mathrm{meV / \AA}\) and \(93.3 \mathrm{meV / \AA}\) on their respective test sets (note that Figure 7 shows the MAE). This provides further evidence that NeuIP exhibits significantly improved data efficiency in comparison with existing methods. + +![](images/Figure_7.jpg) + +
FIG. 7: Log-log plot of the predictive error in forces of NeuIP with \(l = 0\) vs. \(l = 0 / l = 1\) interactions as a function of data set size, measured via the force MAE.
+ +## Computational Efficiency + +Finally, we report the computational efficiency of NeuIP and compare it to that of the ab- into methods on two examples shown in this work: for a molecular + +Finally, we report the computational efficiency of NeuIP and compare it to that of the ab- into methods on two examples shown in this work: for a molecular system, we choose the Toluenne molecule, computed at the CCSD(T)- level of theory [31]; for a material with periodic boundary conditions, we choose the Formate on Cu system, in which reference data were obtained with DFT. For both systems, we report the time required for a single force call on a CPU node with 32 cores. The results are shown in Table VI. In both cases, NeuIP gives a large speed- up over the ab- initio methods. In the case of the Toluenne system, this means that 58.4 minutes of a NeuIP simulation can obtain the simulation time equaling one century of a CCSD(T) simulation. + +## DISCUSSION + +We demonstrate that the Neural Equivariant Interatomic Potential (NeuIP), a new type of graph neural network built on SE(3)- equivariant convolutions exhibits state- of- the- art accuracy and exceptional data efficiency on data sets of small molecules and periodic materials. Furthermore, we find that we can reproduce structural and kinetic properties from molecular dynamics simulations with very high fidelity in comparison to ab- initio simulations. The ability to both learn from small numbers of reference samples, while retaining high computational efficiency opens the door to performing simulations of large systems over long time- scales at quantum mechanical accuracy, using DFT or higher order methods such as coupled- cluster or quantum Monte Carlo data as reference. We expect the new method will enable researchers in computational chemistry, physics, biology, and materials science to conduct molecular dynamics simulations of complex reactions and phase transformations at increased accuracy and efficiency. + +<--- Page Split ---> + + +
SystemNumber of atomsNequlPAb-initioSpeed-up
Toluene1516 ms4 hours*900,000
Formate on Cu5258 ms1045.6 s18,028
+ +TABLE VI: Time required for a single force call for NequIP in comparison to CCSD(T) for Toluene and DFT for Formate on Cu; * personal communication with Stefan Chmiela and Alexandre Tkatchenko. + +687 **METHODS** 715 contains a total of 140,000 structures, of which 100,000 are liquid water and 20,000 are Ice Ih b),10,000 are Ice 688 **Reference Data Sets** 717 Ih c), and another 10,000 are Ice Ih d). + +689 718 + +690 MD-17: MD-17 [17, 30, 31] is a data set of \(T_{175}\) **Formate decomposition on Cu:** The decomposition + +691 eight small organic molecules, obtained from MD720 process of formate on Cu involves configurations + +692 simulations at T=500K and computed at the \(T_{721}\) corresponding to the cleavage of the C-H bond, initial + +693 PBE+vdW-TS level of electronic structure theory,722 and intermediate states (monodentate, bidentate formate + +694 resulting in data set sizes between 133,770 and \(T_{723}\) on Cu \(<110>\) ) and final states (H ad-atom with a + +695 993,237 structures. The data set was obtained from \(T_{724}\) desorbed CO2 in the gas phase). Nudged elastic band + +696 http://quantum-machine.org/gdm1/#datasets. 725 (NEB) method was first used to generate an initial + +697 726 reaction path of the C-H bond breaking. 12 short + +698 Molecules@CCSD/CCSD(T): The data set of \(T_{727}\) ab initio molecular dynamics, starting from different + +699 small molecules at CCSD and CCSD(T) accuracy728 NEB images, were run to collect a total of 6855 DFT + +700 [31] contains positions, energies, and forces for five727 structures. The CP2K [40] code was employed for the + +701 different small molecules: Asprin (CCSD), Benzene,730 AIMD simulations. Each trajectory was generated with + +702 Malondaldehyde, Toluene, Ethanol (all CCSD(T)).731 a time step of 0.5 fs and 500 total steps. We train + +703 Each data set consists of 1,500 structures with the \(T_{732}\) NequlP on 2,500 reference structures sampled uniformly + +704 exception of Ethanol, for which 2,000 structure are733 from the full data set of 6,855 structures, use a validation + +705 available. For more detailed information, we direct734 set of 250 structures and evaluate the mean absolute + +706 the reader to [31]. The data set was obtained from735 error on all remaining structures. Due to the unbalanced + +707 http://quantum-machine.org/gdm1/#datasets. 736 nature of the data set (more atoms of Cu than in the + +708 molecule), we use a per-element weighed loss function in + +709 **Liquid Water and Ice:** The data set of liquid waters738 which atoms C, O1, O2, and H and the sum of all Cu + +710 and ice structures [15, 32] was generated from classical739 atoms all receive equal weights. + +711 AIMD and path-integral AIMD simulations at different740 + +712 temperatures and pressures, computed with a PBE0-TS741 \(Li_{4}P_{2}O_{7}\) glass: The Li4P2O7 ab-initio data were + +713 functional [15]. The data set, obtained from http:742 generated using an ab-initio melt-quench MD simulation, + +714 //www.deepmd.org/database/deeppot-se-data/, 743 starting with a stoichiometric crystal of 208 atoms (space + +<--- Page Split ---> + +group P21/c) in a periodic box of \(10.4 \times 14.0 \times 16.0 \mathrm{cm}\) Å. The dynamics used the Vienna Ab- Initio Simulation Package (VASP) [41- 43], with a generalized gradient PBE functional [44], projector augmented wave (PAW) pseudopotentials [45], a NVT ensemble and a Nose- Hover thermostat, a time step of 2 fs, a plane- wave cutoff of 400 eV, and a \(\Gamma\) - point reciprocal- space mesh. The crystal was melted at 3000 K for 50 ps, then immediately quenched to 600 K and run for another 50 ps. The resulting structure was confirmed to be amorphous by plotting the radial distribution function of P- P distances. The training was performed only on the molten portion, and the MD simulations for quenched simulation. + +LiPS: Lithium phosphorus sulfide (LiPS) based materials are known to exhibit high lithium conductivity, making them attractive as solid- state electrolytes for lithium- ion batteries. Other examples of known materials in this family of superionic conductors are LiGePS and LiCuPS- based compounds. The training data set is taken from a previous study on graph neural network force field [13], where the LiPS training data were generated using ab- initio MD of an LiPS structure with Li- vacancy ( \(\mathrm{Li}_{6.75}\mathrm{P}_{3}\mathrm{S}_{11}\) ) consisting of 27 Li, 12 P and 44 S atoms respectively. The structure was first equilibrated and then run at 520 K using the NVT ensemble for 50 ps with a 2.0 fs time step. The full data set contains 25,001 MD frames. We set aside 10,000 frames as a fixed test set. From the remaining frames, we choose training set sizes of 10, 100, 1,000, and 2,500 frames with a fixed validation set size of 100. In order to generate a diverse training set, we sample both the training and validation sets in a way such that 30% of + +both of them are comprised of the structures with the shortest interatomic distances out of all frames not in the test set and the remaining 70% of the training and validation set are uniformly sampled. + +Liquid Water, Cheng et al.: The training set used in the data efficiency experiments on water consists of 1,593 reference calculations of bulk liquid water at the revPBE0- D3 level of accuracy, with each structure containing 192 atoms, as given in [39]. Further information can be found in [39]. The data set was obtained from https://github.com/BingqingCheng/ab- initio- thermodynamics- of- water. + +Molecular Dynamics Simulations. To run MD simulations, NequIP force outputs were integrated with the Atomic Simulation Environment (ASE) [27] in which we implement a custom version of the Nose- Hoover thermostat. We use this in- house implementation for the both the \(\mathrm{Li}_{4}\mathrm{P}_{2}\mathrm{O}_{7}\) as well as the LiPS MD simulations. The thermostat parameter was chosen to match the temperature fluctuations observed in the AIMD run. The RDF and ADFs for \(\mathrm{Li}_{4}\mathrm{P}_{2}\mathrm{O}_{7}\) were computed with a maximum distance of 10 Å (RDF) and 2.5 Å (both ADFs). + +Training. Networks are trained using a loss function based on atomic forces: + +\[\mathcal{L} = \frac{1}{3N}\sum_{i = 1}^{N}\sum_{\alpha = 1}^{3}|| - \frac{\partial\hat{E}}{\partial r_{i,\alpha}} -F_{i,\alpha}||^{2} \quad (9)\] + +where N is the number of atoms in the system and \(\hat{E}\) is the predicted potential energy. Note that we do + +<--- Page Split ---> + +not train on energies since atomic forces are the only quantities required to integrate Newton's equations of motion. Since the predicted forces are computed as the gradient of a scalar potential, they are still conservative. If energies are of interest, however, one can add them to the loss function and determine the relative weightings via a trade- off parameter as done in previous works [9, 11]. In a similar fashion, it is trivial to add other quantities of interest to the loss function (e.g predicting atomic charges or multipole tensors can be of interest for modeling long- range interactions), where they may be scalar fields, vector fields, or higher- order tensor fields. + +Hyperparameters. Training of models was performed on NVIDIA Tesla V100 GPUs. Throughout all experiments shown in this work, we use a feature dimension of \(h = 64\) , 6 interaction blocks, \(N_{b} = 8\) Bessel basis functions and radial neural networks with one hidden layer, also of hidden dimension \(N_{hidden} = 8\) , giving light- weight radial functions with a comparatively small number of parameters. The final interaction block is followed by the output block, which first reduces the feature dimension to 16 through a self- interaction layer. An equivariant non- linearity is applied and finally through another self- interaction layer the feature dimension is reduced to a single scalar output value associated with each atom that is then summed over to give the total potential energy. Weights were initialized with the uniform Xavier initialization in the radial networks and orthogonal initialization in the self- interaction layers, biases were initialized with constant value of 0. In all experiments, we use the Adam optimizer [46] with the TensorFlow 1.14 default settings of \(\beta_{1} = 0.9\) , \(\beta_{2} = 0.999\) , and \(\epsilon = 10^{- 8}\) . We decrease the initial learning rate of 0.001 by a decay factor of 0.8 whenever the validation RMSE in the forces has not seen an improvement for a given number of epochs: for the small molecule tasks, we set this learning rate patience to 1,000, for all other tasks we use 100. We continuously save the model with the best validation RMSE and use the model with the overall best RMSE for evaluation on the test set and MD simulations. We stop the training if either a maximum number of 50,000 epochs (one epochs equals a full pass over the training set) has been reached, or the validation force RMSE has not improved for 2,500 epochs, or the maximum training time has been exceeded, whichever occurs first. All systems were trained for a maximum of 8 days (consisting of four runs of 48- hour time- limited compute jobs, which are restarted from the best saved model, i.e. potentially including repeats in the training) with the exception of the \(\mathrm{Li_4P_2O_7}\) , which was trained for 12 days (six 48- hour compute jobs) and the LiPS systems, which were trained for 4 days (two 48- hour compute jobs). We use a batch size of 5 structures for all small molecule tasks, and a batch size of 1 structure for all other tasks. We found small batch sizes to be important for obtaining high predictive accuracy. We also found it important to choose the radial cutoff distance \(r_{c}\) appropriately. A list of the cutoff radii in units of [Å] that were used for the different systems is given in Table VII. + +the initial learning rate of 0.001 by a decay factor of 0.8 whenever the validation RMSE in the forces has not seen an improvement for a given number of epochs: for the small molecule tasks, we set this learning rate patience to 1,000, for all other tasks we use 100. We continuously save the model with the best validation RMSE and use the model with the overall best RMSE for evaluation on the test set and MD simulations. We stop the training if either a maximum number of 50,000 epochs (one epochs equals a full pass over the training set) has been reached, or the validation force RMSE has not improved for 2,500 epochs, or the maximum training time has been exceeded, whichever occurs first. All systems were trained for a maximum of 8 days (consisting of four runs of 48-hour time-limited compute jobs, which are restarted from the best saved model, i.e. potentially including repeats in the training) with the exception of the \(\mathrm{Li_4P_2O_7}\) , which was trained for 12 days (six 48-hour compute jobs) and the LiPS systems, which were trained for 4 days (two 48-hour compute jobs). We use a batch size of 5 structures for all small molecule tasks, and a batch size of 1 structure for all other tasks. We found small batch sizes to be important for obtaining high predictive accuracy. We also found it important to choose the radial cutoff distance \(r_{c}\) appropriately. A list of the cutoff radii in units of [Å] that were used for the different systems is given in Table VII. + +Example of a tensor product interaction. To illustrate that the interactions outlined in this work reduce to a set of five simple operations, we write out the example of a full \(1 \otimes 1 \to 1\) interaction, i.e. a convolution that uses a \(l = 1\) filter to operate on a \(l = 1\) feature, yielding again a \(l = 1\) output. This corresponds to + +<--- Page Split ---> + +\(l_{i} = l_{f} = l_{o} = 1\) , facilitating a cross-product interaction + +between two \(l = 1\) tensors. In this case, the Clebsch + +\[C_{(l_{f} = 1,m_{f}),(l_{i} = 1,m_{i})}^{(l_{0} = 1,m_{o})}\propto \epsilon_{o f i} = \left\{ \begin{array}{l l}{1} & {(o,f,i)\in \{(1,2,3),(2,3,1),(3,1,2)\}}\\ {-1} & {(o,f,i)\in \{(1,3,2),(2,1,3),(3,2,1)\}}\\ {0} & {\mathrm{else}} \end{array} \right. \quad (10)\] + +Evaluating equation 8 as well as \(\epsilon_{o f i}\) and using thes relationship \(Y^{(1)}(\hat{r})\propto \hat{r}\) , we recognize the output as + +the vector cross product, taken here between the relative positions and the input feature element \(V_{bc}^{(l = 1)}\) + +\[\mathcal{L}_{ac}^{(l_{o} = 1)}(\vec{r}_{a},V_{ac}^{(l_{i} = 1)}) = \left\{ \begin{array}{l l}{\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{2}V_{bc3}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{3}V_{bc2}^{(l = 1)}}\\ {\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{3}V_{\mathrm{bc}1}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{1}V_{\mathrm{bc}3}^{(l = 1)}}\\ {\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{1}V_{\mathrm{bc}2}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{2}V_{\mathrm{bc}1}^{(l = 1)}} \end{array} \right\} \quad (11)\] + +
Data SetCutoff
MD-17 [17, 30, 31]4.0
Molecules, CCSD/CCSD(T) [31]4.0
Water+Ices, DeepMD [15, 32]6.0
Formate on Cu5.0
Li4P2O7 [12]5.0
LiPS [13]5.0
Water, data efficiency tests [39]4.5
+ +TABLE VII: Radial cutoff in units of [A]. + +## DATA AVAILABILITY + +The code and data sets will be made available upon publication. + +[1] Richards, W. D. et al. Design and synthesis of the superionic conductor na 10 snp 2 s 12. Nature + +communications 7, 1- 8 (2016). [2] Boero, M., Parrinello, M. & Terakura, K. First principles molecular dynamics study of ziegler- natta heterogeneous catalysis. Journal of the American Chemical Society 120, 2746- 2752 (1998). [3] Lindorff- Larsen, K., Piana, S., Dror, R. O. & Shaw, D. E. How fast- folding proteins fold. Science 334, 517- 520 (2011). [4] Behler, J. & Parrinello, M. Generalized neural- network representation of high- dimensional potential- energy surfaces. Physical review letters 98, 146401 (2007). [5] Bartók, A. P., Payne, M. C., Kondor, R. & Csányi, G. Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons. Physical review letters 104, 136403 (2010). [6] Shapeev, A. V. Moment tensor potentials: A class of systematically improvable interatomic potentials. + +<--- Page Split ---> + +[7] Thompson, A. P., Swiler, L. P., Trott, C. R., Foiles, S. M. 944 & Tucker, G. J. Spectral neighbor analysis method for automated generation of quantum- accurate interatomic potentials. Journal of Computational Physics 285, 316- 947 330 (2015). [8] Vandermause, J. et al. On- the- fly active learning of interpretable bayesian force fields for atomistic rare- events. npj Computational Materials 6, 1- 11 (2020). [9] Schütt, K. et al. Schnet: A continuous- filters convolutional neural network for modeling quantum- interactions. In Advances in neural information processing systems, 991- 1001 (2017). [10] Unke, O. T. & Meuwly, M. Physnet: A neural network for predicting energies, forces, dipole moments, and partial- charges. Journal of chemical theory and computation 15, 968 3678- 3693 (2019). [11] Klicpera, J., GroB, J. & Gunnemann, S. Directional message passing for molecular graphs. arXiv preprint arXiv:2003.03123 (2020). [12] Mailoa, J. P. et al. A fast neural network approach for direct covariant forces prediction in complex multi- element extended systems. Nature machine intelligence 1, 471- 479 (2019). [13] Park, C. W. et al. Accurate and scalable multi- element graph neural network force field and molecular dynamics with direct force architecture. arXiv preprint arXiv:2007.14444 (2020). [14] Artrit, N. & Kolpak, A. M. Understanding the composition and activity of electrocatalytic nanoalloys in aqueous solvents: A combination of dft and accurate neural network potentials. Nano letters 14, 2670- 2676 (2014). [15] Zhang, L., Han, J., Wang, H., Car, R. & Weinan, E. Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics. Physical review letters 120, 143001 (2018). + +[16] Smith, J. S., Isayev, O. & Roitberg, A. E. Ani- 1: an extensible neural network potential with dft accuracy at force field computational cost. Chemical science 8, 3192- 3203 (2017). [17] Chmiela, S. et al. Machine learning of accurate energy- conserving molecular force fields. Science advances 3, e1603015 (2017). [18] Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. arXiv preprint arXiv:1704.01212 (2017). [19] Anderson, B., Hy, T. S. & Kondor, R. Cormorant: Covariant molecular neural networks. In Advances in Neural Information Processing Systems, 14537- 14546 (2019). [20] Townshend, R. J., Townshend, B., Eismann, S. & Dror, R. O. Geometric prediction: Moving beyond scalars. arXiv preprint arXiv:2006.14163 (2020). [21] Thomas, N. et al. Tensor field networks: Rotation- and translation- equivariant neural networks for 3d point clouds. arXiv preprint arXiv:1802.08219 (2018). [22] Grisafi, A., Wilkins, D. M., Willatt, M. J. & Ceriotti, M. Atomic- scale representation and statistical learning of tensorial properties. In Machine Learning in Chemistry: Data- Driven Algorithms, Learning Systems, and Predictions, 1- 21 (ACS Publications, 2019). [23] Abadi, M. et al. Tensorflow: Large- scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467 (2016). [24] Paszke, A. et al. Pytorch: An imperative style, high- performance deep learning library. In Advances in neural information processing systems, 8026- 8037 (2019). [25] Weiler, M., Geiger, M., Welling, M., Boomsma, W. & Cohen, T. S. 3d steerable cnns: Learning rotationally equivariant features in volumetric data. In Advances in Neural Information Processing Systems, 10381- 10392 (2018). [26] Kondor, R., Lin, Z. & Trivedi, S. Clebsch- gordan nets: a fully fourier space spherical convolutional neural + +<--- Page Split ---> + +network. In Advances in Neural Information Processing Systems, 10117- 10126 (2018). [27] Larsen, A. H. et al. The atomic simulation environment—a python library for working with atoms202 Journal of Physics: Condensed Matter 29, 273002a3 (2017). URL https://doi.org/10.1088%2F1361- 648x%2F2Faa680e. [28] Smidt, T. E., Geiger, M. & Miller, B. K. Findings symmetry breaking order parameters with euclidean202 neural networks. arXiv preprint arXiv:2007.02005a3 (2020). [29] He, K., Zhang, X., Ren, S. & Sun, J. Deep residual202 learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770- 778 (2016). [30] Schütt, K. T., Arbabzadah, F., Chmiela, S., Müller202 K. R. & Tkatchenko, A. Quantum- chemical insights from202 deep tensor neural networks. Nature Communications 8, 13890 (2017). URL https://doi.org/10.1038/ncomms13890. [31] Chmiela, S., Sauceda, H. E., Müller, K.- R. &202 Tkatchenko, A. Towards exact molecular dynamics202 simulations with machine- learned force fields. Nature Communications 9, 3887 (2018). [32] Ko, H.- Y. et al. Isotope effects in liquid water via deep202 potential molecular dynamics. Molecular Physics 117, 1047 3269- 3281 (2019). [33] Christensen, A. S., Bratholm, L. A., Faber, F. A. &202 Anatole von Lilienfeld, O. Fchl revisited: Faster and202 more accurate quantum machine learning. The Journal of Chemical Physics 152, 044107 (2020). [34] Sim, W. S., Gardner, P. & King, D. A. Multiplex202 bonding configurations of adsorbed formate on ag111o1 The Journal of Physical Chemistry 100, 12509- 12516a2 (1996). 00000. [35] Wang, G., Morikawa, Y., Matsumoto, T. & Nakamura202 J. Why is formate synthesis insensitive to copper surfaces202 structures? The Journal of Physical Chemistry B 110, 1056 + +9- 11 (2006). 00050. [36] Yu, X., Bates, J. B., Jellison, G. E. & Hart, F. X. A stable thin- film lithium electrolyte: Lithium phosphorus oxynitride. Journal of The Electrochemical Society 144, 524- 532 (1997). URL https://doi.org/10.1149/1.1837443. [37] Westover, A. S. et al. Plasma synthesis of spherical crystalline and amorphous electrolyte nanopowders for solid- state batteries. ACS Applied Materials & Interfaces 12, 11570- 11578 (2020). URL https://doi.org/10.1021/acsami.9b20812. [38] Li, W., Ando, Y., Minamitani, E. & Watanabe, S. Study of li atom diffusion in amorphous li3po4 with neural network potential. The Journal of chemical physics 147, 214106 (2017). [39] Cheng, B., Engel, E. A., Behler, J., Dellago, C. & Ceriotti, M. Ab initio thermodynamics of liquid and solid water. Proceedings of the National Academy of Sciences 116, 1110- 1115 (2019). [40] Hutter, J., Iannuzzi, M., Schiffmann, F. & VandeVondele, J. cp2k: atomistic simulations of condensed matter systems. WIREs Computational Molecular Science 4, 15- 25 (2014). 00000. [41] Kresse, G. & Hafner, J. Ab initiomolecular dynamics for liquid metals. Physical Review B 47, 558- 561 (1993). URL https://doi.org/10.1103/physrevb.47.558. [42] Kresse, G. & Furthmüller, J. Efficiency of ab- initio total energy calculations for metals and semiconductors using a plane- wave basis set. Computational Materials Science 6, 15- 50 (1996). URL https://doi.org/10.1016/0927- 0256(96)00008- 0. [43] Kresse, G. & Furthmüller, J. Efficient iterative schemes forab initiototal- energy calculations using a plane- wave basis set. Physical Review B 54, 11169- 11186 (1996). URL https://doi.org/10.1103/physrevb.54.11169. [44] Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Physical Review Letters 77, 3865- 3868 (1996). URL https://doi.org/ + +<--- Page Split ---> + +1057 10.1103/physrevlett.77.3865. 1058 [45] Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented- wave method. Physical Review B 59, 1758- 1775 (1999). URL https://doi.org/ 1061 10.1103/physrevb.59.1758. 1062 [46] Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014). 1063 + +## ACKNOWLEDGEMENTS + +1065 We thank Jonathan Vandermause, Cheol Woo Park, 1066 David Clark, Kostiantyn Lapchevskyi, Mario Geiger 1067 Joshua Rackers, and Benjamin Kurt Miller for helpful 1068 discussions. 1069 We thank Stefan Chmiela and Alexandre Tkatchenko 1070 for providing the timing data for the Toluene system. 1071 Work at Harvard was supported by Bosch Research 1072 and the Integrated Mesoscale Architectures for 1073 Sustainable Catalysis (IMASC), an Energy Frontier 1074 Research Center funded by the US Department of 1075 Energy (DOE), Office of Science, Office of Basic Energy 1076 Sciences under Award No. DE- SC0012573. N.M. 1077 acknowledges support from the Department of the Navy; 1078 Office of Naval Research. + +1079 Work at Bosch Research was partially supported by 1080 ARPA- E Award No. DE- AR0000775 and used resources 1081 of the Oak Ridge Leadership Computing Facility at Oak 1082 Ridge National Laboratory, which is supported by the 1083 Office of Science of the Department of Energy under 1084 Contract DE- AC05- 00OR22725. 1085 T.E.S. was supported by the Laboratory Directed 1086 Research and Development Program of Lawrence 1087 Berkeley National Laboratory and the Center 1088 for Advanced Mathematics for Energy Research + +Applications, both under U.S. Department of Energy Contract No. DE- AC02- 05CH11231. + +The authors acknowledge computing resources provided by the Harvard University FAS Division of Science Research Computing Group and by the Texas Advanced Computing Center (TACC) at The University of Texas at Austin under allocations DMR20009 and DMR20013. + +## AUTHOR CONTRIBUTIONS + +S.B. initiated the project, conceived the NequIP model, implemented the software and conducted all software experiments under the guidance of B.K. T.E.S. contributed to the conception of the model, guidance of computational experiments, and the software implementation. L.S. created the data set for formate/Cu, guided work on training and MD simulations on this system, and contributed to development of the software implementation. J.P.M. guided the work on the LiPS conductor and implemented the thermostat for MD simulations together with S.B.. M.K. created the AIMD data set of \(\mathrm{Li_4P_2O_7}\) , wrote software for the analysis of MD results and guided the benchmarking on this system. N.M. wrote software for the computation of the diffusion results and guided discussions on the interpretation of results. B.K. supervised the project from conception to design of experiments, implementation, theory, as well as analysis of data. All authors contributed to the manuscript and the discussion of results. + +## COMPETING INTERESTS + +The authors declare no competing interests. + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +Left: a set of atoms is interpreted as an atomic graph with local neighborhoods. Middle: every atom carries a set of scalar and vector features with it. Right: atoms exchange information via filters, that are again scalars and vectors. The interactions of features and filters define five interactions. + +![](images/Figure_2.jpg) + +
Figure 2
+ +The NequIP network architecture. Left: atomic numbers are embedded into \(I = 0\) features, which are refined through a series of interaction blocks, creating \(I = 0\) and \(I = 1\) features. An output block generates atomic energies, which are pooled to give the total predicted energy. Middle: the interaction block consists of a series of convolutions, interweaved with self- interaction layers, equivariant nonlinearities and concatenation. Right: the convolution combines the radial function \(R(r)\) which operates only on interatomic distances with the spherical harmonics based on unit vector \(r^*\) via a tensor product. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Perspective view of atomic configurations of (a) bidentate HCOO (b) monodentate HCOO and (c) CO2 and a hydrogen adatom on a Cu(110) surface. The blue, red, black, and white spheres represent Cu, O, C, and H atoms, respectively. The subset shown in each subplot is the corresponding top view along the < \(110>\) orientation. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Quenched glass structure of Li4P2O7. The insets show the P- O- O tetrahedral bond angle (bottom left) as well as the O- P- P bridging angle between corner- sharing phosphate tetrahedra (top right). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5
+ +Left: Radial Distribution Function, middle: Angular Distribution Function, bridging oxygen, right: Angular Distribution Function, tetrahedral bond angle. All are defined as probability density functions. + +![](images/Figure_6.jpg) + +
Figure 6
+ +Comparison of Lithium mean square displacement of AIMD and NequIP trajectories. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7
+ +Log- log plot of the predictive error in forces of NequIP with \(l = 0\) vs. \(l = 0 / l = 1\) interactions as a function of data set size, measured via the force MAE. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- codesubmission.zip + +<--- Page Split ---> diff --git a/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb_det.mmd b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9e40024a4a2c9c060e2b16a17691467b7fbf29a1 --- /dev/null +++ b/preprint/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb/preprint__11fe674e450c321782c05dca0f68a54e2fd8b7f5e7c3af3b147e8fb38ede0fdb_det.mmd @@ -0,0 +1,700 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 923, 175]]<|/det|> +# SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials + +<|ref|>text<|/ref|><|det|>[[44, 196, 216, 236]]<|/det|> +Simon Batzner Harvard University + +<|ref|>text<|/ref|><|det|>[[44, 242, 756, 285]]<|/det|> +Tess Smidt Lawrence Berkeley National Laboratory https://orcid.org/0000- 0001- 5581- 5344 + +<|ref|>text<|/ref|><|det|>[[44, 290, 395, 331]]<|/det|> +Lixin Sun Massachusetts Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 336, 401, 377]]<|/det|> +Jonathan Mailoa Bosch Research and Technology Center + +<|ref|>text<|/ref|><|det|>[[44, 382, 820, 424]]<|/det|> +Mordechai Kombluth Robert Bosch Research and Technology Center https://orcid.org/0000- 0001- 6705- 8133 + +<|ref|>text<|/ref|><|det|>[[44, 428, 202, 468]]<|/det|> +Nicola Molinari Harvard University + +<|ref|>text<|/ref|><|det|>[[44, 473, 576, 515]]<|/det|> +Boris Kozinsky ( bkoz@seas.harvard.edu) Harvard University https://orcid.org/0000- 0002- 0638- 539X + +<|ref|>sub_title<|/ref|><|det|>[[44, 556, 102, 574]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 593, 883, 636]]<|/det|> +Keywords: Molecular Dynamics Simulations, Symmetry- aware Models, Equivariant Convolutions, Geometric Tensors + +<|ref|>text<|/ref|><|det|>[[44, 654, 300, 674]]<|/det|> +Posted Date: March 1st, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 692, 463, 712]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 244137/v1 + +<|ref|>text<|/ref|><|det|>[[44, 729, 910, 772]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 808, 949, 851]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 4th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29939- 5. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[137, 62, 866, 110]]<|/det|> +# SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials + +<|ref|>text<|/ref|><|det|>[[251, 133, 750, 150]]<|/det|> +Simon Batzner, \(^{1}\) Tess E. Smidt, \(^{2}\) Lixin Sun, \(^{1}\) Jonathan P. Mailoa, \(^{3}\) + +<|ref|>text<|/ref|><|det|>[[260, 160, 737, 177]]<|/det|> +Mordechai Kornbluth, \(^{3}\) Nicola Molinari, \(^{1}\) and Boris Kozinsky \(^{1,3}\) + +<|ref|>text<|/ref|><|det|>[[285, 188, 717, 203]]<|/det|> +\(^{1}\) John A. Paulson School of Engineering and Applied Sciences, + +<|ref|>text<|/ref|><|det|>[[330, 211, 672, 225]]<|/det|> +Harvard University, Cambridge, MA 02138, USA + +<|ref|>text<|/ref|><|det|>[[140, 232, 863, 248]]<|/det|> +\(^{2}\) Computational Research Division and Center for Advanced Mathematics for Energy Research Applications, + +<|ref|>text<|/ref|><|det|>[[272, 255, 732, 270]]<|/det|> +Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA + +<|ref|>text<|/ref|><|det|>[[237, 277, 766, 292]]<|/det|> +\(^{3}\) Robert Bosch Research and Technology Center, Cambridge, MA 02139, USA + +<|ref|>text<|/ref|><|det|>[[173, 309, 832, 544]]<|/det|> +This work presents Neural Equivariant Interatomic Potentials (NequlP), a SE(3)- equivariant neural network approach for learning interatomic potentials from ab- initio calculations for molecular dynamics simulations. While most contemporary symmetry- aware models use invariant convolutions and only act on scalars, NequlP employs SE(3)- equivariant convolutions for interactions of geometric tensors, resulting in a more information- rich and faithful representation of atomic environments. The method achieves state- of- the- art accuracy on a challenging set of diverse molecules and materials while exhibiting remarkable data efficiency. NequlP outperforms existing models with up to three orders of magnitude fewer training data, challenging the widely held belief that deep neural networks require massive training sets. The high data efficiency of the method allows for the construction of accurate potentials using high- order quantum chemical level of theory as reference and enables high- fidelity molecular dynamics simulations over long time scales. + +<|ref|>sub_title<|/ref|><|det|>[[214, 578, 359, 593]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[57, 630, 496, 927]]<|/det|> +Molecular dynamics (MD) simulations are an indispensable tool for computational discovery in fields \(^{37}\) as diverse as energy storage, catalysis, and biological processes [1- 3]. While the atomic forces required to integrate Newton's equations of motion can in principle be obtained with high fidelity from quantum- mechanical calculations such as density functional theory (DFT), in practice the unfavorable computational scaling of first- principles methods limits simulations to short time scales and small numbers of atoms. This prohibits the study of many interesting physical phenomena beyond the time and length scales that are currently + +<|ref|>text<|/ref|><|det|>[[512, 577, 920, 923]]<|/det|> +accessible, even on the largest supercomputers. Owing to their simple functional form, classical models for the atomic potential energy can typically be evaluated orders of magnitude faster than using first- principles methods, thereby enabling the study of large numbers of atoms over long time scales. However, due to their limited mathematical form, classical interatomic potentials, or force fields, are inherently limited in their predictive accuracy which has historically led to a fundamental trade- off between obtaining high computational efficiency while also predicting faithful dynamics of the system under study. The construction of flexible models of the interatomic potential energy based on Machine Learning (ML- IP), and in particular + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 60, 494, 610]]<|/det|> +48 Neural Networks (NN- IP), has shown great promise in a providing a way to move past this dilemma, promising a to learn high- fidelity potentials from ab- initio reference calculations while retaining favorable computational efficiency [4- 13]. One of the limiting factors of NN- IPs is that they typically require collection of large training sets of ab- initio calculations, often including thousands or even millions of reference structures [4, 9, 10, 14- 16]. This computationally expensive process of training data collection has severely limited the adoption of NN- IPs as it quickly becomes a bottleneck in the development of force- fields for new systems. Kernel- based approaches, such as e.g. Gaussian Processes (GP) [5, 8] or Kernel Ridge Regression (KRR) [17], are a way to remedy this problem as they often generalize better from limited sample sizes. However, such methods generally tend to exhibit poor computational scaling with the number of reference configurations, in both training (cubic in training set size) and prediction (linear in training set size). This limits both the amount of training data they can be trained on as well as the length and size of simulations that can be simulated with them. + +<|ref|>text<|/ref|><|det|>[[512, 65, 920, 585]]<|/det|> +accurate interatomic potentials from limited data sets of fewer than 1,000 or even as little as 100 reference ab- initio calculations, where other methods require orders of magnitude more. It is worth noting that on small molecular data sets, NequIP outperforms not only other neural networks, but is also competitive with kernel- based approaches, which typically obtain better predictive accuracy than NN- IPs on small data sets (although at significant additional cost scaling in training and prediction). We further demonstrate high data efficiency and accuracy with state- of- the- art results on a training set of molecular data obtained at the quantum chemical coupled- cluster level of theory. Finally, we validate the method through a series of simulations and demonstrate that we can reproduce with high fidelity structural and kinetic properties computed from NequIP simulations in comparison to ab- initio molecular dynamics simulations (AIMD). We directly verify that the performance gains are connected with the unique SE(3)- equivariant convolution architecture of the new NequIP model. + +<|ref|>text<|/ref|><|det|>[[58, 630, 74, 640]]<|/det|> +70 + +<|ref|>text<|/ref|><|det|>[[58, 648, 494, 691]]<|/det|> +In this work, we present the Neural Equivariant Interatomic Potential (NequIP), a highly data- efficient + +<|ref|>text<|/ref|><|det|>[[58, 700, 495, 914]]<|/det|> +deep learning approach for learning interatomic potentials from reference first- principles calculations. We show that the proposed method obtains high- accuracy compared to existing ML- IP methods across a wide variety of systems, including small molecules, water in different phases, an amorphous solid, a reaction at a solid/gas interface, and a Lithium superionic conductor. Furthermore, we find that NequIP exhibits exceptional data efficiency, enabling the construction of + +<|ref|>sub_title<|/ref|><|det|>[[671, 652, 763, 666]]<|/det|> +## Related Work + +<|ref|>text<|/ref|><|det|>[[512, 696, 920, 914]]<|/det|> +First applications of machine learning for the development of interatomic potentials were built on descriptor- based approaches combined with shallow neural networks or Gaussian Processes [4, 5], designed to exhibit invariance with respect to translation, permutation of atoms of the same chemical species, and rotation. Recently, rotationally invariant graph neural networks (GNN- IPs) have emerged as a powerful architecture for deep learning of interatomic potentials + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 60, 496, 792]]<|/det|> +that eliminates the need for hand- crafted descriptors and allows to instead learn representations on graphs of atoms from invariant features of geometric data (e.g. radial distances or angles) [9- 11, 13]. In GNN- IPs, atomic structures are represented by collections of nodes and edges, where nodes in the graph correspond to individual atoms and edges are typically defined by simply connecting every atom to all other atoms that are closer than some cutoff distance \(r_c\) . Node/atom \(i\) is associated with a feature \(\mathbf{h}_i \in \mathbb{R}^h\) , consisting of scalar values, which is iteratively refined via a series of convolutions over neighboring atoms based on both the distance to neighboring atoms and their features \(\mathbf{h}_j\) . This iterative process allows information to be propagated along the atomic graph through a series of convolutional layers and can be viewed as a message- passing scheme [18]. Operating only on interatomic distances allows GNN- IPs to rotation- and translation- invariant, making both output as well as features internal to the network invariant to rotations. In contrast, the method outlined in this work uses relative position vectors rather than simply distances (scalars), which makes internal features instead equivariant to rotation and allows for angular information to be used by rotationally equivariant filters. Similar to other methods, we can restrict convolutions to only a local subset of all other atoms that lie closer to the central atom than a chosen cutoff distance \(r_c\) , see Figure 1, left. + +<|ref|>text<|/ref|><|det|>[[57, 824, 496, 915]]<|/det|> +A series of related methods have recently been proposed: DimeNet [11] expands on using pairwise interactions in a single convolution to include angular, three- body terms, but individual features are still + +<|ref|>text<|/ref|><|det|>[[512, 64, 920, 585]]<|/det|> +comprised of scalars (distances and three- body angles are invariant to rotation), as opposed to vectors used in this work. Another central difference to NequIP is that DimeNet explicitly enumerates angles between pairs of atoms and operates on a basis embedding of distances and angles, whereas NequIP operates on relative position vectors and a basis embedding of distances, and thus never explicitly computes three- body angles. Cormorant [19] uses an equivariant neural network for property prediction on small molecules. This method is demonstrated on potential energies of small molecules but not on atomic forces or systems with periodic boundary conditions. Townshend et al. [20] use the framework of Tensor- Field Networks [21] to directly predict atomic force vectors. The predicted forces are not guaranteed by construction to conserve energy since they are not obtained as gradients of the total potential energy. This may lead to problems in simulations of molecular dynamics over long times. None of these three works [11, 19, 20] demonstrates capability to perform molecular dynamics simulations. + +<|ref|>text<|/ref|><|det|>[[512, 747, 919, 914]]<|/det|> +In this work we present a deep learning energy- conserving interatomic potential for both molecules and materials built on SE(3)- equivariant convolutions over geometric tensors that yields state- of- the- art accuracy, outstanding data- efficiency, and can with high fidelity reproduce structural and kinetic properties from molecular dynamics simulations. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 70, 916, 220]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 236, 920, 283]]<|/det|> +
FIG. 1: Left: a set of atoms is interpreted as an atomic graph with local neighborhoods. Middle: every atom carries a set of scalar and vector features with it. Right: atoms exchange information via filters, that are again scalars and vectors. The interactions of features and filters define five interactions.
+ +<|ref|>sub_title<|/ref|><|det|>[[244, 312, 327, 327]]<|/det|> +## RESULTS + +<|ref|>text<|/ref|><|det|>[[244, 362, 331, 376]]<|/det|> +Equivariance + +<|ref|>text<|/ref|><|det|>[[55, 412, 496, 830]]<|/det|> +The concept of equivariance arises naturally in machine learning of atomistic systems (see e.g. [22]): physical properties have well- defined transformation properties under translation and rotation of a set of atoms. As a simple example, if a molecule is rotated in space, the vectors of its atomic dipoles or forces also rotate accordingly, via equivariant transformation. Equivariant neural networks are able to more generally represent tensor properties and tensor operations of physical systems (e.g. vector addition, dot products, and cross products). Equivariant neural networks are guaranteed to preserve the known transformation properties of physical systems under a change of coordinates because they are explicitly constructed from equivariant operations. Formally, a function \(f: X \to Y\) is equivariant with respect to a group \(G\) that acts on \(X\) and \(Y\) if: + +<|ref|>text<|/ref|><|det|>[[485, 311, 920, 429]]<|/det|> +where \(D_{X}[g]\) and \(D_{Y}[g]\) are the representations of the group element \(g\) in the vector spaces \(X\) and \(Y\) , respectively. In this work, we focus on equivariance with respect to \(\mathrm{SE}(3)\) , i.e. the group of rotations and translations in 3D space. + +<|ref|>sub_title<|/ref|><|det|>[[577, 467, 857, 482]]<|/det|> +## Neural Equivariant Interatomic Potentials + +<|ref|>text<|/ref|><|det|>[[485, 510, 919, 602]]<|/det|> +Given a set of atoms (a molecule or a material), we aim to find a mapping from atomic positions \(\vec{r}_{i}\) and chemical species (identified by atomic numbers \(Z_{i}\) ) to the total potential energy and the forces acting on the atoms: + +<|ref|>equation<|/ref|><|det|>[[647, 630, 916, 649]]<|/det|> +\[f:\{\vec{r}_{i},Z_{i}\} \to E_{pot} \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[485, 678, 919, 745]]<|/det|> +Forces are obtained as gradients of the predicted potential energy with respect to the atomic positions, which guarantees energy conservation: + +<|ref|>equation<|/ref|><|det|>[[664, 772, 916, 792]]<|/det|> +\[\vec{F}_{i} = -\nabla_{i}E_{pot} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[485, 821, 919, 912]]<|/det|> +Gradients can be obtained with relatively low computational overhead in modern auto- differentiation frameworks such as TensorFlow or PyTorch [23, 24]. Following previous work [4], we further define the total + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 65, 490, 105]]<|/det|> +potential energy of the system as a sum of atomic potential energies: + +<|ref|>equation<|/ref|><|det|>[[197, 130, 490, 166]]<|/det|> +\[E_{p o t} = \sum_{i\in N_{a t o m s}}E_{i,a t o m i c} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[56, 185, 490, 426]]<|/det|> +These atomic local energies \(E_{i,atomic}\) are the scalar node attributes predicted by the graph neural network. Even though the output of NequIP is the predicted potential energy \(E_{pot}\) , which is invariant under translations and rotations, the network contains internal features that are tensors which are equivariant to rotation. This constitutes the core difference between NequIP and existing scalar- valued invariant GNN- IPs. The remainder of this section will discuss the design of the network in further detail. + +<|ref|>text<|/ref|><|det|>[[56, 444, 490, 460]]<|/det|> +A series of methods has been introduced to realize + +<|ref|>text<|/ref|><|det|>[[56, 468, 490, 910]]<|/det|> +A series of methods has been introduced to realize rotationally equivariant neural networks [13, 21, 25, 26]. Here, we build on the layers introduced in Tensor- Field Networks (TFN) [21], which enable the construction of neural networks that exhibit equivariance to translation, permutation, and rotation. Every atom in NequIP is associated with a feature comprised of tensors of different order: scalars, vectors, and higher- order tensors. Formally, these features are geometric objects that comprise a direct sum of irreducible representations of the SO(3) symmetry group. Second, the convolutions that operate on these geometric objects are equivariant functions instead of invariant ones, i.e. if a feature layer \(k\) is rotated, then the output of the convolution from layer \(k \to k + 1\) rotates accordingly. In practice, the features are implemented as a dictionary \(V_{acm}^{(l)}\) with keys \(l\) , where \(l = 0, 1, 2, \ldots\) is a non- negative integer and is called the "rotation order", labeling the irreducible + +<|ref|>text<|/ref|><|det|>[[511, 65, 919, 156]]<|/det|> +representations. The indices \(a\) , \(c\) , \(m\) , correspond to the atoms, the channels (elements of the feature), and the representation index which takes values \(m \in [- l, l]\) , respectively. + +<|ref|>text<|/ref|><|det|>[[511, 195, 919, 465]]<|/det|> +Convolution operations are naturally translation invariant, since their filters act on relative interatomic distance vectors. Moreover, they are permutation invariant since all convolution contributions are summed. Note that while atomic features are equivariant to permutation of atom indices, globally, the total potential energy of the system is invariant to permutation. To achieve rotation equivariance, the convolution filters are constrained to be products of learnable radial functions and spherical harmonics, which are equivariant under SO(3) [21]: + +<|ref|>equation<|/ref|><|det|>[[628, 521, 916, 541]]<|/det|> +\[F(\vec{r}_{ij}) = R(r_{ij})Y_{m}^{(l)}(\hat{r}_{ij}) \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[511, 568, 919, 836]]<|/det|> +where if \(\vec{r}_{ij}\) denotes the relative position from central atom \(i\) to neighboring atom \(j\) , \(\hat{r}_{ij}\) and \(r_{ij}\) are the associated unit vector and interatomic distance, respectively. It should be noted that all learnable weights in the filter lie in the rotationally invariant radial function \(R(r_{ij})\) . This radial function is implemented as a small neural network with one hidden layer and a shifted softplus activation function [9], operating on interatomic distances expressed in a basis of choice, \(R(r_{ij}) : \mathbb{R}^{N_{b}} \to \mathbb{R}^{h}\) , where \(N_{b}\) is the number of basis functions and \(h\) is the feature dimension: + +<|ref|>equation<|/ref|><|det|>[[571, 898, 916, 915]]<|/det|> +\[R(r_{ij}) = W_{2}\ln (0.5\exp (W_{1}B(r_{ij})) + 0.5) \quad (6)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 62, 504, 283]]<|/det|> +where \(W_{1} \in \mathbb{R}^{N_{hidden} \times N_{b}}\) and \(W_{2} \in \mathbb{R}^{h \times N_{hidden}}\) are 296 weight matrices, \(h\) is the dimension of the feature and 298 \(N_{hidden}\) is the dimension of the hidden layer in the 300 feed- forward neural network (in our experiments, we use 301 \(N_{hidden} = N_{b}\) , resulting in comparatively small neural 302 networks for the radial function). Radial Bessel functions 303 and a polynomial envelope function \(f_{env}\) discussed in 304 recent work [11] are used to expand the interatomic 305 distances: + +<|ref|>text<|/ref|><|det|>[[55, 395, 504, 737]]<|/det|> +where \(r_{c}\) is a local cutoff radius, restricting interactions 310 to atoms closer than some cutoff distance and \(f_{env}\) is the polynomial defined in [11] with \(p = 6\) operating on the interatomic distances normalized by the cutoff radius \(r_{c}^{i,j}\) . The use of cutoffs/local atomic environments allows the computational cost of evaluation to scale linearly with the number of atoms. Similar to [11], we initialize the Bessel functions with \(n = [1,2,\dots,N_{b}]\) and subsequently optimize \(n\pi\) via backpropagation rather than keeping it constant. For systems with periodic boundary conditions, we use the neighbor list functionality as implemented in the ASE code [27] to identify appropriate atomic neighbors and then convolve over them. + +<|ref|>text<|/ref|><|det|>[[55, 785, 504, 851]]<|/det|> +Finally, in the convolution, the input feature tensor 294 and the filter have to again be combined in an 295 equivariant manner, which is achieved via a geometric 296 + +<|ref|>text<|/ref|><|det|>[[514, 65, 920, 308]]<|/det|> +tensor product, yielding an output feature that again is rotationally equivariant. A tensor product of two geometric tensors is computed via Clebsch- Gordan coefficients, as outlined in [21]. Since NequIP deals with force vectors, the network design is simplified by only using scalar (l=0) and vector (l=1) representations. Thus, we can enumerate five distinct products or "interactions" between \(l = 0\) and \(l = 1\) filters and \(l = 0\) and \(l = 1\) features that correspond to simple operations between scalars and vectors: + +<|ref|>text<|/ref|><|det|>[[537, 330, 875, 490]]<|/det|> +\(\bullet 0\otimes 0\to 0\) (product of two scalars) \(\bullet 0\otimes 1\to 1\) (scalar multiplication of a vector) \(\bullet 1\otimes 0\to 1\) (scalar multiplication of a vector) \(\bullet 1\otimes 1\to 0\) (dot product of two vectors) \(\bullet 1\otimes 1\to 1\) (cross product of two vectors) + +<|ref|>text<|/ref|><|det|>[[514, 506, 920, 852]]<|/det|> +It is trivial to include higher- order interactions, and previous works have increased the rotation order beyond \(l = 1\) [20, 28]. However, it should be noted that every interaction comes with additional trainable radial functions and hence additional weights, which thus adds to the model capacity, increasing the number of model weights and the memory footprint of the model. Omitting all higher- order interactions that go beyond the \(0\otimes 0\to 0\) interaction will result in a conventional GNN- IP with invariant convolutions over scalar features, similar to e.g. SchNet [9]. Finally, as outlined in [21], a full convolutional layer \(\mathcal{L}\) implementing an interaction with filter \(f\) acting on an input \(i\) producing output \(o\) : \(l_{f}\otimes l_{i}\to l_{o}\) is given by: + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[250, 60, 916, 98]]<|/det|> +\[\mathcal{L}_{acm_o}^{(l_o)}(\vec{r}_a,V_{acm_i}^{(l_i)}) = \sum_{m_f,m_i}C_{(l_f,m_f)(l_i,m_i)}^{(l_o,m_o)}\sum_{b\in S}R_c^{(l_f,l_i)}(r_{ab})Y_{m_f}^{(l_f)}(\hat{r}_{ab})V_{bcm_i}^{(l_i)} \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[55, 155, 504, 650]]<|/det|> +where \(a\) and \(b\) index the central atom of the convolution and the neighboring atom \(b \in S\) , respectively, and \(C\) indicates the Clebsch- Gordan coefficients. As an example of this, we write out a full \(1 \otimes 1 \to 1\) operation (corresponding to a cross- product) in the Methods section. After every convolution, output tensors of \(a_{362}\) rotation order \(l\) stemming from different tensor products are concatenated on a per- atom basis. To update atomic features, the model also leverages self- interaction layers similar to SchNet [9], corresponding to dense layers that are applied in an atom- wise fashion with weights shared across atoms. While different weights are used for different rotation orders, the same set of weights is applied for all representation indices \(m\) of a given rotation order \(l\) . Shifted softplus nonlinearities [9] are used as rotation- equivariant nonlinearities introduced in [21], which are applied to the Euclidean norm of the input feature, the output of which is in turn combined with the input tensor, thus preserving overall equivariance. + +<|ref|>text<|/ref|><|det|>[[55, 684, 503, 750]]<|/det|> +The NequIP network architecture, shown in Figure 2, is built on an atomic embedding, followed by a series of interaction blocks, and finally an output block: + +<|ref|>text<|/ref|><|det|>[[110, 774, 503, 890]]<|/det|> +Embedding: following SchNet, the initial feature is generated using a trainable embedding, that operates on the atomic number \(Z_{i}\) (represented via a one- hot encoding) alone, implemented via a trainable self- interaction layer. + +<|ref|>text<|/ref|><|det|>[[110, 911, 503, 927]]<|/det|> +Interaction Block: interaction blocks encodes interactions between neighboring atoms: the core of this block is the convolution function, outlined in equation 8. For every output rotation order \(l_{o}\) , the features from different tensor product interactions are concatenated to give a new feature, which is in return refined with atom- wise self- interaction layers and equivariant non- linearities. We equip interactions blocks with a ResNet- style update [29] where the input feature \(\mathbf{x}\) is updated atom- wise via the output of an interaction block \(f(\mathbf{x})\) that gives the final feature \(r(\mathbf{x}) = f(\mathbf{x}) + \mathbf{x}\) (features are added element- wise in the \(m\) - dimension). Note that this operation is equivariant since the addition of an equivariant feature \(\mathbf{x}\) and an equivariant function \(f(\mathbf{x})\) preserves equivariance. While later interaction blocks include all five interactions outlined above, the first interaction block operates on the \(l = 0\) embedding with a \(0 \otimes 0 \to 0\) and a \(0 \otimes 1 \to 1\) only. + +<|ref|>text<|/ref|><|det|>[[553, 155, 920, 627]]<|/det|> +interactions between neighboring atoms: the core of this block is the convolution function, outlined in equation 8. For every output rotation order \(l_{o}\) , the features from different tensor product interactions are concatenated to give a new feature, which is in return refined with atom- wise self- interaction layers and equivariant non- linearities. We equip interactions blocks with a ResNet- style update [29] where the input feature \(\mathbf{x}\) is updated atom- wise via the output of an interaction block \(f(\mathbf{x})\) that gives the final feature \(r(\mathbf{x}) = f(\mathbf{x}) + \mathbf{x}\) (features are added element- wise in the \(m\) - dimension). Note that this operation is equivariant since the addition of an equivariant feature \(\mathbf{x}\) and an equivariant function \(f(\mathbf{x})\) preserves equivariance. While later interaction blocks include all five interactions outlined above, the first interaction block operates on the \(l = 0\) embedding with a \(0 \otimes 0 \to 0\) and a \(0 \otimes 1 \to 1\) only. + +<|ref|>text<|/ref|><|det|>[[543, 645, 919, 761]]<|/det|> +Output Block: the \(l = 0\) feature of the final convolution is passed to an output block, which consists of another atom- wise self- interaction layer, an equivariant non- linearity, and a final atom- wise self- interaction layer. + +<|ref|>text<|/ref|><|det|>[[513, 785, 919, 927]]<|/det|> +The scalar atomic outputs of the final layer can be interpreted as atomic potential energies which are summed to give the total predicted potential energy of the system (Equation 4). Forces are subsequently obtained as the negative gradient of the predicted total potential energy, thereby ensuring both energy + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[130, 74, 907, 370]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 389, 911, 485]]<|/det|> +
FIG. 2: The NequIP network architecture. Left: atomic numbers are embedded into \(l = 0\) features, which are refined through a series of interaction blocks, creating \(l = 0\) and \(l = 1\) features. An output block generates atomic energies, which are pooled to give the total predicted energy. Middle: the interaction block consists of a series of convolutions, interweaved with self-interaction layers, equivariant nonlinearities and concatenation. Right: the convolution combines the radial function \(R(r)\) which operates only on interatomic distances with the spherical harmonics based on unit vector \(\hat{r}\) via a tensor product.
+ +<|ref|>text<|/ref|><|det|>[[55, 509, 492, 553]]<|/det|> +387 conservation as well as rotation- equivariant forces (see equation 3). + +<|ref|>sub_title<|/ref|><|det|>[[238, 600, 336, 615]]<|/det|> +## Experiments + +<|ref|>text<|/ref|><|det|>[[55, 644, 492, 912]]<|/det|> +390 We validate the proposed method on a series of diverse 391 and challenging data sets: first we demonstrate that we 392 improve upon state- of- the- art accuracy on MD- 17, a data 393 set of small, organic molecules that is widely used for 394 benchmarking ML- IPs [9, 11, 17, 30, 31]. Next, we show 395 that NequIP can also accurately learn forces obtained on 396 small molecules at the quantum chemical CCSD(T) level 397 [31], opening the door to scalable and efficient molecular 398 dynamics simulations with beyond- DFT accuracy. To 399 broaden the applicability of the method beyond small 400 isolated molecules, we explore a series of extended 401 + +<|ref|>text<|/ref|><|det|>[[504, 509, 920, 653]]<|/det|> +systems with periodic boundary conditions, consisting of both surfaces and bulk materials: water in different phases [15, 32], a chemical reaction at a solid/gas interface, an amorphous Lithium Phosphate [12], and a Li superionic conductor [13]. Details of the training procedure are provided in the Methods section. + +<|ref|>sub_title<|/ref|><|det|>[[610, 700, 825, 715]]<|/det|> +## MD-17 small molecule dynamics + +<|ref|>text<|/ref|><|det|>[[504, 744, 920, 912]]<|/det|> +We first evaluate NequIP on MD- 17 [17, 30, 31], a data set of eight small organic molecules in which reference values of energy and forces are generated by ab- initio MD simulations with DFT. For training we use N=1,000 structure configurations for each molecule, sampled uniformly from the full data set, the same number of configurations for validation, and evaluate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 60, 503, 816]]<|/det|> +the test error on all remaining configurations in the data set. The mean absolute error in the force components is shown in Table I in units of [meV/A]. We compare results using NequIP with those from published leading ML- IP models that were also trained on 1,000 structures: in particular SchNet [9], DimeNet [11] (both graph neural networks), sGDML [31], and FCHL19/GPR (kernel- based methods) [33]. We find that NequIP outperforms SchNet and sGDML on all molecules in the data set, DimeNet on 7 out of 8 molecules (on par on the remaining one), and performs on par with FCHL/19GPR. The consistent improvement in accuracy upon sGDML and the comparable performance to FCHL19/GPR are particularly surprising, as these are based on kernel methods, that typically tend to be more sample efficient. It should be noted, however, that the evaluation cost of kernel methods scales linearly with the number of training configurations. Note also that some molecules, NequIP trained on 1,000 configurations even performs as well as SchNet trained on 50,000 structures [9]: on aspirin and naphthalene, for example, the NequIP network trained on 1,000 structures produces mean absolute errors in the forces of 15.1 meV/A and 4.2 meV/A, respectively, compared to 14.3 meV/A and 4.8 meV/A of SchNet trained on 50x more molecules, hinting that NequIP exhibits exceptional data efficiency. On other molecules such as ethanol, however, SchNet trained with 50,000 molecules still clearly outperforms NequIP trained with 1,000 molecules (2.2 meV/A for SchNet for N=50,000 vs 9.0 meV/A for NequIP for N=1,000). + +<|ref|>text<|/ref|><|det|>[[508, 108, 920, 784]]<|/det|> +Ability to achieve high accuracy on a comparatively small data set opens the door to training models on expensive high- order ab- initio quantum chemical methods. It has been shown that DFT can fail to capture important subtleties in the potential energy surface, potentially even identifying the wrong ground states [31]. This problem can be remedied through the use of more accurate reference calculations, such as coupled cluster methods CCSD(T), typically regarded as the gold standard of quantum chemistry. However, the high computational cost of CCSD(T) has thus far hindered the use of reference data structures at this level of theory, prohibited by the need for large data sets that are required by available NN- IPs. Leveraging the high data efficiency of NequIP, we evaluate it on a set of molecules computed at quantum chemical accuracy (aspirin at CCSD, all others at CCSD(T)) [31] and compare the results to those reported for sGDML [31]. The training/validation set consists of a total of 1,000 molecular structures which we split into 950 for training and 50 for validation (sampled uniformly), and we test the accuracy on all remaining structures (we use the train/test split provided with the data set, but further split the training set into training and validation sets). We find that NequIP achieves lower errors on four out of five molecules, performing on par with sGDML on the fifth molecule, as shown in Table II. + +<|ref|>text<|/ref|><|det|>[[510, 870, 919, 912]]<|/det|> +To demonstrate the applicability of NequIP beyond small molecules, we evaluate the method on a series of + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[105, 60, 898, 223]]<|/det|> + +
MoleculeNequIPSchNetsGDMLDimeNetFCHL19/GPR
Aspirin15.158.529.521.620.7
Benzene [17]8.113.4n/a8.1n/a
Benzene [31]2.3n/a2.6n/an/a
Ethanol9.016.914.310.05.9
Malonaldehyde14.628.617.816.610.6
Naphthalene4.225.24.89.36.5
Salicylic Acid10.336.912.116.29.6
Toluene4.424.76.19.48.8
Uracil7.524.310.413.14.6
+ +<|ref|>table_caption<|/ref|><|det|>[[95, 237, 912, 281]]<|/det|> +TABLE I: MAE of force components on the MD-17 data set, trained on 1,000 configurations, forces in units of [meV/A]. For the benzene molecule, two different data set exists from [17], [31] with different levels of accuracy in the DFT reference data. + +<|ref|>table<|/ref|><|det|>[[90, 305, 499, 405]]<|/det|> + +
MoleculeNequIPsGDML
Aspirin14.733.0
Benzene0.81.7
Ethanol9.415.2
Malonaldehyde16.016.0
Toluene4.49.1
+ +<|ref|>table_caption<|/ref|><|det|>[[117, 419, 457, 464]]<|/det|> +TABLE II: Force MAE for molecules at CCSD/CCSD(T) accuracy, reported in units of [meV/A], with 1,000 reference configurations). + +<|ref|>text<|/ref|><|det|>[[56, 495, 496, 508]]<|/det|> +476 extended systems with periodic boundary conditions. As + +<|ref|>text<|/ref|><|det|>[[56, 516, 496, 530]]<|/det|> +477 a first example we use a joint data set consisting of liquid + +<|ref|>text<|/ref|><|det|>[[56, 540, 496, 553]]<|/det|> +478 water and three ice structures [15,32], computed at the + +<|ref|>text<|/ref|><|det|>[[56, 565, 496, 579]]<|/det|> +479 PBE0-TS level of theory. This data set contains [15]: + +<|ref|>text<|/ref|><|det|>[[56, 590, 496, 604]]<|/det|> +480 a) liquid water, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=300\mathrm {K},\) computed via path- + +<|ref|>text<|/ref|><|det|>[[56, 616, 490, 630]]<|/det|> +481 integral AIMD, b) ice Ih, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=273\mathrm {K},\) computed + +<|ref|>text<|/ref|><|det|>[[56, 641, 496, 680]]<|/det|> +482 via path-integral AIMD c) ice Ih, \(\mathrm {P}=1\mathrm {bar},\mathrm {T}=330\mathrm {K},\) computed via classical AIMD d) ice Ih, \(\mathrm {P}=2.13\mathrm {kbar},\) + +<|ref|>text<|/ref|><|det|>[[56, 694, 496, 708]]<|/det|> +484 \(\mathrm {T}=238\mathrm {K},\) computed via classical AIMD. The liquid water + +<|ref|>text<|/ref|><|det|>[[56, 720, 496, 734]]<|/det|> +485 system consists of 64 \(\mathrm {H}_{2}\mathrm {O}\) molecules (192 atoms), while + +<|ref|>text<|/ref|><|det|>[[56, 745, 496, 759]]<|/det|> +486 the ice structures consist of 96 \(\mathrm {H}_{2}\mathrm {O}\) molecules (288 + +<|ref|>text<|/ref|><|det|>[[56, 770, 496, 784]]<|/det|> +487 atoms). A DeepMD NN-IP model was previously trained + +<|ref|>text<|/ref|><|det|>[[56, 795, 496, 810]]<|/det|> +488 [15] for water and ice using a joint training set containing + +<|ref|>text<|/ref|><|det|>[[56, 821, 496, 835]]<|/det|> +489 133,500 reference calculations of these four systems. Tos + +<|ref|>text<|/ref|><|det|>[[56, 846, 496, 860]]<|/det|> +490 assess data efficiency of the NequIP architecture, wes + +<|ref|>text<|/ref|><|det|>[[56, 872, 496, 885]]<|/det|> +491 similarly train a model jointly on all four parts of the data + +<|ref|>text<|/ref|><|det|>[[56, 897, 496, 911]]<|/det|> +492 set, but using only 133 structures for training, i.e. 1000x + +<|ref|>text<|/ref|><|det|>[[504, 308, 920, 600]]<|/det|> +498 fewer data. The 133 structures were sampled uniformly from the full data set available online, consisting of water and ice structures, made up of a total of 140,000 frames, coming from the same MD trajectories that were used in the earlier work [15]. We also use a validation set of 50 frames and report the test accuracy on all remaining structures in the data set. Table III shows the comparison of the predictive force accuracy of NequIP trained on the 133 structures vs DeepMD trained on 133,500 structures. We find that with 1000x fewer training data, NequIP significantly outperforms DeepMD on all four parts of the data set. + +<|ref|>text<|/ref|><|det|>[[546, 652, 888, 664]]<|/det|> +Heterogeneous catalysis of formate dehydrogenation + +<|ref|>text<|/ref|><|det|>[[504, 696, 920, 911]]<|/det|> +Next, we demonstrate application of NequIP to a catalytic surface reaction. In particular, we investigate the dynamics of formate undergoing dehydrogenation decomposition \(\left(\mathrm {HCOO}^{*}\rightarrow \mathrm {H}^{*}+\mathrm {CO}_{2}\right)\) on a Cu \(<110>\) surface (see Figure 3). This system is highly heterogeneous, with both metallic and covalent types of bonding as well as charge transfer occurring between the metal and the molecule, making this a particularly challenging test system. Different states of the molecule + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[170, 61, 832, 147]]<|/det|> + +
SystemNequIP, 133 data pointsDeepMD, 133,500 data points
Liquid Water35.940.4
Ice Ih (b)25.943.3
Ice Ih (c)16.626.8
Ice Ih (d)13.525.4
+ +<|ref|>table<|/ref|><|det|>[[153, 219, 420, 314]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[102, 158, 902, 191]]<|/det|> +TABLE III: Root mean square error (RMSE) of force components on liquid water and the three ices in units of [meV/A]. Note that the NequIP model was trained on \(< 0.1\%\) of the training data of DeepMD. + +
ElementMAE
C55.8
O86.7
H42.0
Cu54.5
Total structure55.6
+ +<|ref|>text<|/ref|><|det|>[[88, 325, 485, 372]]<|/det|> +TABLE IV: MAE of force components for Formate on Cu system, per- element basis. The training set consists of 2,500 structures, force units are [meV/A] + +<|ref|>text<|/ref|><|det|>[[55, 411, 490, 630]]<|/det|> +also lead to dissimilar C- O bond lengths [34, 35]. Training structures consist of 48 Cu atoms and 4 atoms of the molecule (HCOO\\* or \(\mathrm{CO_2 + H^*}\) ). The MAE of the predicted forces using a NequIP model trained on 2,500 structures is shown in Table IV, demonstrating that NequIP is able to accurately model the interatomic forces for this complex reactive system. A more detailed analysis of the resulting dynamics will be subject of a separate study. + +<|ref|>image<|/ref|><|det|>[[85, 670, 500, 772]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[87, 779, 491, 884]]<|/det|> +
FIG. 3: Perspective view of atomic configurations of (a) bidentate HCOO (b) monodentate HCOO and (c) \(\mathrm{CO_2}\) and a hydrogen adatom on a \(\mathrm{Cu(110)}\) surface. The blue, red, black, and white spheres represent Cu, O, C, and H atoms, respectively. The subset shown in each subplot is the corresponding top view along the \(< 110>\) orientation.
+ +<|ref|>text<|/ref|><|det|>[[520, 220, 878, 234]]<|/det|> +Lithium Phosphate Amorphous Glass Formation + +<|ref|>image<|/ref|><|det|>[[565, 268, 858, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[519, 530, 914, 592]]<|/det|> +
FIG. 4: Quenched glass structure of \(\mathrm{Li_4P_2O_7}\) . The insets show the P-O-O tetrahedral bond angle (bottom left) as well as the O-P-P bridging angle between corner-sharing phosphate tetrahedra (top right).
+ +<|ref|>text<|/ref|><|det|>[[504, 620, 920, 912]]<|/det|> +To examine the ability of the model to capture dynamical properties, we demonstrate that NequIP can describe structural dynamics in amorphous lithium phosphate with composition \(\mathrm{Li_4P_2O_7}\) . This material is a member of the promising family of solid electrolytes for Li- metal batteries [12, 36, 37], with non- trivial Li- ion transport and phase transformation behaviors. The training data set consists of two 50ps- long AIMD simulations, one of the molten structure at \(\mathrm{T} = 3000\) K, followed by another of a quenched glass structure at \(\mathrm{T} = 600\) K. We train NequIP on a subset of 1,000 structures of the molten trajectory, each consisting of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 66, 494, 636]]<|/det|> +208 atoms, and sampled uniformly from the full data set of 25,000 AIMD frames. We use a validation set of 100 structures, and evaluate the model on all remaining structures. Table V shows the test set error in the force components on both the test set from the AIMD molten trajectory and the full AIMD quenched glass trajectory. To then evaluate the physical fidelity of the trained model, we use it to run a MD simulation of length ps at T=600 K in the NVT ensemble and compare the total radial distribution function (RDF) without element distinction as well as the angular distribution functions (ADF) of the P-O-O (P central atom) and O-P-P (central atom) angles to the ab-into trajectory at the same temperature. The P-O-O angle corresponds to the tetrahedral bond angle, while the O-P-P corresponds to a bridging angle between corner-sharing phosphates tetrahedra (Figure 4). Figure 5 shows that NequIP can accurately reproduce the RDF and the two ADFs, in comparison with AIMD, after training on only 1,000 structures. This demonstrates that the model generates the glass state and recovers its dynamics and structures almost perfectly, having seen only the high-temperature molten training data. + +<|ref|>text<|/ref|><|det|>[[513, 66, 920, 633]]<|/det|> +data set of 30,874 structures to study Li diffusion in \(\mathrm{Li}_3\mathrm{PO}_4\) . Here we demonstrate that not only does NequIP obtain small errors in the force components, but it also accurately predicts the diffusivity after training on a data set obtained from an AIMD simulation. Again, we find that very small training sets lead to highly accurate models, as shown in Table V for training set sizes of 10, 100, 1,000 and 2,500 structures. We run a series of MD simulations with the NequIP potential trained on 2,500 structures in the NVT ensemble at the same temperature as the AIMD simulation for a total simulation time of 50 ps and a time step of 0.25 fs, which we found advantageous for reliability and stability of long simulations. We measure the Li diffusivity in ten Nequip-driven MD simulations (computed via the slope of the mean square displacement), all of length 50 ps and started from different initial velocities, randomly sampled from a Maxwell-Boltzmann distribution. We find a mean diffusivity of \(1.42 \times 10^{-5} \mathrm{~cm}^2 / \mathrm{s}\) , in excellent agreement with the diffusivity of \(1.38 \times 10^{-5} \mathrm{~cm}^2 / \mathrm{s}\) computed from AIMD, thus achieving a relative error of as little as 3%. Figure 6 shows the mean square displacements of Li for an example run. + +<|ref|>table<|/ref|><|det|>[[55, 664, 949, 780]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[55, 680, 437, 695]]<|/det|> +Lithium Thiophosphate Superionic Transport + +
SystemData Set SizeMAE
LiPS10157.1
LiPS10050.0
LiPS1,00025.1
LiPS2,50024.1
Li4P2O7, melt1,00063.2
Li4P2O7, quench1,00036.9
+ +<|ref|>text<|/ref|><|det|>[[55, 720, 490, 912]]<|/det|> +To show that NequIP can model kinetic transport properties from small training sets at high accuracy, we study Li- ion diffusivity in LiPS ( \(\mathrm{Li}_{6.75}\mathrm{P}_3\mathrm{S}_{11}\) ) a crystalline superionic Li conductor, consisting of a simulation cell of 83 atoms [13]. MD is widely used to study diffusion; however, training a ML- IP to the accuracy required to accurately predict kinetic properties has in the past required large training set sizes ([38] e.g. uses a + +<|ref|>text<|/ref|><|det|>[[520, 792, 915, 880]]<|/det|> +TABLE V: Force MAE for LiPS and \(\mathrm{Li}_4\mathrm{P}_2\mathrm{O}_7\) for different data set sizes in units of [meV/A]. The model for \(\mathrm{Li}_4\mathrm{P}_2\mathrm{O}_7\) was trained exclusively on structures from the melted trajectory, the reported test errors show the MAE on both the test set of the melted trajectory as well as the full quench trajectory. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[86, 66, 914, 185]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 194, 919, 227]]<|/det|> +
FIG. 5: Left: Radial Distribution Function, middle: Angular Distribution Function, bridging oxygen, right: Angular Distribution Function, tetrahedral bond angle. All are defined as probability density functions.
+ +<|ref|>image<|/ref|><|det|>[[100, 266, 470, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 472, 460, 503]]<|/det|> +
FIG. 6: Comparison of Lithium mean square displacement of AIMD and NequIP trajectories.
+ +<|ref|>sub_title<|/ref|><|det|>[[228, 547, 345, 562]]<|/det|> +## Data Efficiency + +<|ref|>text<|/ref|><|det|>[[78, 592, 502, 915]]<|/det|> +In the above experiments, NequIP exhibits exceptionally high data efficiency, i.e. it can be trained successfully to state- of- the- art accuracy from \(\mathrm{62}\) unexpectedly small training sets. It is interesting to consider the reasons for such high performance and verify that it is connected to the equivariant nature of the \(\mathrm{62}\) model. First, it is important to note that each training configuration contains multiple labels, thus increasing the total number of labels available beyond just the potential energy label associated with each structure. In particular, for a training set of \(M\) first- principles calculations with structures consisting of \(N\) atoms, the total number of labels available is \(M(3N + 1)\) since every + +<|ref|>text<|/ref|><|det|>[[512, 255, 919, 348]]<|/det|> +force component on every atom constitutes a label and so does the total energy of the reference calculation (we only train to atomic forces and not energies, thus using \(3MN\) force components as labels). + +<|ref|>text<|/ref|><|det|>[[512, 368, 920, 915]]<|/det|> +In order to gain insight into the reasons behind increased accuracy and data efficiency, we perform a series of experiments with the goal of isolating the effect of using equivariant convolutions of geometric tensors compared to invariant convolutions over scalars. In particular, we run a set of experiments for a system with a fixed number of training configurations in which we explicitly turn on or off interactions of higher order than \(l = 0\) . This defines two settings: first, we train the network with both \(l = 0\) and \(l = 1\) features and all five interactions as previously outlined in this work. Second, when all interactions involving \(l = 1\) are turned off, this turns the network into a conventional invariant GNN- IP, involving only invariant convolutions over scalar features in a SchNet- style fashion. As a test system we chose bulk water: in particular we use the data set introduced in [39], consisting of 1,593 bulk liquid water structures with 64 water molecules each. We train a series of networks with identical hyperparameters, but vary the training set sizes between 10 and 1,000 structures, sampled uniformly from the full data set, as well as a validation set consisting of 100 structures. We then evaluate the error on all + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 65, 500, 510]]<|/det|> +remaining structures for a given training set size. As shown in Figure 7, we find that the equivariant setting (using \(l = 0\) and \(l = 1\) ) significantly outperforms the invariant setting (using only \(l = 0\) ) for all data set sizes as measured by the MAE of force components. This suggests that it is indeed the use of tensor (in our specific case vector) features and equivariant convolutions that enables the high data efficiency of NeuIP. We further note, that in [39], a Behler-Parrinello Neural Network (BPNN) was trained on 1303 structures, yielding RMSE of \(\approx 120 \mathrm{meV / \AA}\) in forces when evaluated the remaining 290 structures. We find that NeuIP models trained with as little as 50 and 100 data points obtain RMSEs of \(122.9 \mathrm{meV / \AA}\) and \(93.3 \mathrm{meV / \AA}\) on their respective test sets (note that Figure 7 shows the MAE). This provides further evidence that NeuIP exhibits significantly improved data efficiency in comparison with existing methods. + +<|ref|>image<|/ref|><|det|>[[84, 528, 500, 689]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[90, 700, 500, 747]]<|/det|> +
FIG. 7: Log-log plot of the predictive error in forces of NeuIP with \(l = 0\) vs. \(l = 0 / l = 1\) interactions as a function of data set size, measured via the force MAE.
+ +<|ref|>sub_title<|/ref|><|det|>[[191, 804, 384, 820]]<|/det|> +## Computational Efficiency + +<|ref|>text<|/ref|><|det|>[[52, 847, 499, 912]]<|/det|> +Finally, we report the computational efficiency of NeuIP and compare it to that of the ab- into methods on two examples shown in this work: for a molecular + +<|ref|>text<|/ref|><|det|>[[513, 65, 920, 335]]<|/det|> +Finally, we report the computational efficiency of NeuIP and compare it to that of the ab- into methods on two examples shown in this work: for a molecular system, we choose the Toluenne molecule, computed at the CCSD(T)- level of theory [31]; for a material with periodic boundary conditions, we choose the Formate on Cu system, in which reference data were obtained with DFT. For both systems, we report the time required for a single force call on a CPU node with 32 cores. The results are shown in Table VI. In both cases, NeuIP gives a large speed- up over the ab- initio methods. In the case of the Toluenne system, this means that 58.4 minutes of a NeuIP simulation can obtain the simulation time equaling one century of a CCSD(T) simulation. + +<|ref|>sub_title<|/ref|><|det|>[[660, 375, 771, 391]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[513, 419, 920, 912]]<|/det|> +We demonstrate that the Neural Equivariant Interatomic Potential (NeuIP), a new type of graph neural network built on SE(3)- equivariant convolutions exhibits state- of- the- art accuracy and exceptional data efficiency on data sets of small molecules and periodic materials. Furthermore, we find that we can reproduce structural and kinetic properties from molecular dynamics simulations with very high fidelity in comparison to ab- initio simulations. The ability to both learn from small numbers of reference samples, while retaining high computational efficiency opens the door to performing simulations of large systems over long time- scales at quantum mechanical accuracy, using DFT or higher order methods such as coupled- cluster or quantum Monte Carlo data as reference. We expect the new method will enable researchers in computational chemistry, physics, biology, and materials science to conduct molecular dynamics simulations of complex reactions and phase transformations at increased accuracy and efficiency. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[145, 61, 858, 117]]<|/det|> + +
SystemNumber of atomsNequlPAb-initioSpeed-up
Toluene1516 ms4 hours*900,000
Formate on Cu5258 ms1045.6 s18,028
+ +<|ref|>table_caption<|/ref|><|det|>[[95, 130, 910, 158]]<|/det|> +TABLE VI: Time required for a single force call for NequIP in comparison to CCSD(T) for Toluene and DFT for Formate on Cu; * personal communication with Stefan Chmiela and Alexandre Tkatchenko. + +<|ref|>text<|/ref|><|det|>[[60, 193, 920, 255]]<|/det|> +687 **METHODS** 715 contains a total of 140,000 structures, of which 100,000 are liquid water and 20,000 are Ice Ih b),10,000 are Ice 688 **Reference Data Sets** 717 Ih c), and another 10,000 are Ice Ih d). + +<|ref|>text<|/ref|><|det|>[[60, 264, 500, 275]]<|/det|> +689 718 + +<|ref|>text<|/ref|><|det|>[[60, 290, 920, 304]]<|/det|> +690 MD-17: MD-17 [17, 30, 31] is a data set of \(T_{175}\) **Formate decomposition on Cu:** The decomposition + +<|ref|>text<|/ref|><|det|>[[60, 315, 920, 328]]<|/det|> +691 eight small organic molecules, obtained from MD720 process of formate on Cu involves configurations + +<|ref|>text<|/ref|><|det|>[[60, 339, 920, 352]]<|/det|> +692 simulations at T=500K and computed at the \(T_{721}\) corresponding to the cleavage of the C-H bond, initial + +<|ref|>text<|/ref|><|det|>[[60, 362, 920, 375]]<|/det|> +693 PBE+vdW-TS level of electronic structure theory,722 and intermediate states (monodentate, bidentate formate + +<|ref|>text<|/ref|><|det|>[[60, 386, 920, 399]]<|/det|> +694 resulting in data set sizes between 133,770 and \(T_{723}\) on Cu \(<110>\) ) and final states (H ad-atom with a + +<|ref|>text<|/ref|><|det|>[[60, 410, 920, 423]]<|/det|> +695 993,237 structures. The data set was obtained from \(T_{724}\) desorbed CO2 in the gas phase). Nudged elastic band + +<|ref|>text<|/ref|><|det|>[[60, 435, 920, 448]]<|/det|> +696 http://quantum-machine.org/gdm1/#datasets. 725 (NEB) method was first used to generate an initial + +<|ref|>text<|/ref|><|det|>[[60, 460, 920, 472]]<|/det|> +697 726 reaction path of the C-H bond breaking. 12 short + +<|ref|>text<|/ref|><|det|>[[60, 492, 920, 505]]<|/det|> +698 Molecules@CCSD/CCSD(T): The data set of \(T_{727}\) ab initio molecular dynamics, starting from different + +<|ref|>text<|/ref|><|det|>[[60, 516, 920, 529]]<|/det|> +699 small molecules at CCSD and CCSD(T) accuracy728 NEB images, were run to collect a total of 6855 DFT + +<|ref|>text<|/ref|><|det|>[[60, 540, 920, 553]]<|/det|> +700 [31] contains positions, energies, and forces for five727 structures. The CP2K [40] code was employed for the + +<|ref|>text<|/ref|><|det|>[[60, 564, 920, 577]]<|/det|> +701 different small molecules: Asprin (CCSD), Benzene,730 AIMD simulations. Each trajectory was generated with + +<|ref|>text<|/ref|><|det|>[[60, 588, 920, 601]]<|/det|> +702 Malondaldehyde, Toluene, Ethanol (all CCSD(T)).731 a time step of 0.5 fs and 500 total steps. We train + +<|ref|>text<|/ref|><|det|>[[60, 612, 920, 625]]<|/det|> +703 Each data set consists of 1,500 structures with the \(T_{732}\) NequlP on 2,500 reference structures sampled uniformly + +<|ref|>text<|/ref|><|det|>[[60, 636, 920, 650]]<|/det|> +704 exception of Ethanol, for which 2,000 structure are733 from the full data set of 6,855 structures, use a validation + +<|ref|>text<|/ref|><|det|>[[60, 661, 920, 674]]<|/det|> +705 available. For more detailed information, we direct734 set of 250 structures and evaluate the mean absolute + +<|ref|>text<|/ref|><|det|>[[60, 685, 920, 698]]<|/det|> +706 the reader to [31]. The data set was obtained from735 error on all remaining structures. Due to the unbalanced + +<|ref|>text<|/ref|><|det|>[[60, 709, 920, 722]]<|/det|> +707 http://quantum-machine.org/gdm1/#datasets. 736 nature of the data set (more atoms of Cu than in the + +<|ref|>text<|/ref|><|det|>[[60, 733, 920, 746]]<|/det|> +708 molecule), we use a per-element weighed loss function in + +<|ref|>text<|/ref|><|det|>[[60, 767, 920, 780]]<|/det|> +709 **Liquid Water and Ice:** The data set of liquid waters738 which atoms C, O1, O2, and H and the sum of all Cu + +<|ref|>text<|/ref|><|det|>[[60, 792, 740, 805]]<|/det|> +710 and ice structures [15, 32] was generated from classical739 atoms all receive equal weights. + +<|ref|>text<|/ref|><|det|>[[60, 816, 500, 828]]<|/det|> +711 AIMD and path-integral AIMD simulations at different740 + +<|ref|>text<|/ref|><|det|>[[60, 840, 920, 853]]<|/det|> +712 temperatures and pressures, computed with a PBE0-TS741 \(Li_{4}P_{2}O_{7}\) glass: The Li4P2O7 ab-initio data were + +<|ref|>text<|/ref|><|det|>[[60, 864, 920, 877]]<|/det|> +713 functional [15]. The data set, obtained from http:742 generated using an ab-initio melt-quench MD simulation, + +<|ref|>text<|/ref|><|det|>[[60, 888, 920, 901]]<|/det|> +714 //www.deepmd.org/database/deeppot-se-data/, 743 starting with a stoichiometric crystal of 208 atoms (space + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 65, 496, 405]]<|/det|> +group P21/c) in a periodic box of \(10.4 \times 14.0 \times 16.0 \mathrm{cm}\) Å. The dynamics used the Vienna Ab- Initio Simulation Package (VASP) [41- 43], with a generalized gradient PBE functional [44], projector augmented wave (PAW) pseudopotentials [45], a NVT ensemble and a Nose- Hover thermostat, a time step of 2 fs, a plane- wave cutoff of 400 eV, and a \(\Gamma\) - point reciprocal- space mesh. The crystal was melted at 3000 K for 50 ps, then immediately quenched to 600 K and run for another 50 ps. The resulting structure was confirmed to be amorphous by plotting the radial distribution function of P- P distances. The training was performed only on the molten portion, and the MD simulations for quenched simulation. + +<|ref|>text<|/ref|><|det|>[[55, 444, 496, 914]]<|/det|> +LiPS: Lithium phosphorus sulfide (LiPS) based materials are known to exhibit high lithium conductivity, making them attractive as solid- state electrolytes for lithium- ion batteries. Other examples of known materials in this family of superionic conductors are LiGePS and LiCuPS- based compounds. The training data set is taken from a previous study on graph neural network force field [13], where the LiPS training data were generated using ab- initio MD of an LiPS structure with Li- vacancy ( \(\mathrm{Li}_{6.75}\mathrm{P}_{3}\mathrm{S}_{11}\) ) consisting of 27 Li, 12 P and 44 S atoms respectively. The structure was first equilibrated and then run at 520 K using the NVT ensemble for 50 ps with a 2.0 fs time step. The full data set contains 25,001 MD frames. We set aside 10,000 frames as a fixed test set. From the remaining frames, we choose training set sizes of 10, 100, 1,000, and 2,500 frames with a fixed validation set size of 100. In order to generate a diverse training set, we sample both the training and validation sets in a way such that 30% of + +<|ref|>text<|/ref|><|det|>[[511, 65, 919, 157]]<|/det|> +both of them are comprised of the structures with the shortest interatomic distances out of all frames not in the test set and the remaining 70% of the training and validation set are uniformly sampled. + +<|ref|>text<|/ref|><|det|>[[511, 193, 920, 384]]<|/det|> +Liquid Water, Cheng et al.: The training set used in the data efficiency experiments on water consists of 1,593 reference calculations of bulk liquid water at the revPBE0- D3 level of accuracy, with each structure containing 192 atoms, as given in [39]. Further information can be found in [39]. The data set was obtained from https://github.com/BingqingCheng/ab- initio- thermodynamics- of- water. + +<|ref|>text<|/ref|><|det|>[[511, 420, 920, 688]]<|/det|> +Molecular Dynamics Simulations. To run MD simulations, NequIP force outputs were integrated with the Atomic Simulation Environment (ASE) [27] in which we implement a custom version of the Nose- Hoover thermostat. We use this in- house implementation for the both the \(\mathrm{Li}_{4}\mathrm{P}_{2}\mathrm{O}_{7}\) as well as the LiPS MD simulations. The thermostat parameter was chosen to match the temperature fluctuations observed in the AIMD run. The RDF and ADFs for \(\mathrm{Li}_{4}\mathrm{P}_{2}\mathrm{O}_{7}\) were computed with a maximum distance of 10 Å (RDF) and 2.5 Å (both ADFs). + +<|ref|>text<|/ref|><|det|>[[511, 725, 918, 765]]<|/det|> +Training. Networks are trained using a loss function based on atomic forces: + +<|ref|>equation<|/ref|><|det|>[[589, 808, 916, 852]]<|/det|> +\[\mathcal{L} = \frac{1}{3N}\sum_{i = 1}^{N}\sum_{\alpha = 1}^{3}|| - \frac{\partial\hat{E}}{\partial r_{i,\alpha}} -F_{i,\alpha}||^{2} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[511, 871, 918, 913]]<|/det|> +where N is the number of atoms in the system and \(\hat{E}\) is the predicted potential energy. Note that we do + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 66, 504, 380]]<|/det|> +not train on energies since atomic forces are the only quantities required to integrate Newton's equations of motion. Since the predicted forces are computed as the gradient of a scalar potential, they are still conservative. If energies are of interest, however, one can add them to the loss function and determine the relative weightings via a trade- off parameter as done in previous works [9, 11]. In a similar fashion, it is trivial to add other quantities of interest to the loss function (e.g predicting atomic charges or multipole tensors can be of interest for modeling long- range interactions), where they may be scalar fields, vector fields, or higher- order tensor fields. + +<|ref|>text<|/ref|><|det|>[[54, 393, 504, 916]]<|/det|> +Hyperparameters. Training of models was performed on NVIDIA Tesla V100 GPUs. Throughout all experiments shown in this work, we use a feature dimension of \(h = 64\) , 6 interaction blocks, \(N_{b} = 8\) Bessel basis functions and radial neural networks with one hidden layer, also of hidden dimension \(N_{hidden} = 8\) , giving light- weight radial functions with a comparatively small number of parameters. The final interaction block is followed by the output block, which first reduces the feature dimension to 16 through a self- interaction layer. An equivariant non- linearity is applied and finally through another self- interaction layer the feature dimension is reduced to a single scalar output value associated with each atom that is then summed over to give the total potential energy. Weights were initialized with the uniform Xavier initialization in the radial networks and orthogonal initialization in the self- interaction layers, biases were initialized with constant value of 0. In all experiments, we use the Adam optimizer [46] with the TensorFlow 1.14 default settings of \(\beta_{1} = 0.9\) , \(\beta_{2} = 0.999\) , and \(\epsilon = 10^{- 8}\) . We decrease the initial learning rate of 0.001 by a decay factor of 0.8 whenever the validation RMSE in the forces has not seen an improvement for a given number of epochs: for the small molecule tasks, we set this learning rate patience to 1,000, for all other tasks we use 100. We continuously save the model with the best validation RMSE and use the model with the overall best RMSE for evaluation on the test set and MD simulations. We stop the training if either a maximum number of 50,000 epochs (one epochs equals a full pass over the training set) has been reached, or the validation force RMSE has not improved for 2,500 epochs, or the maximum training time has been exceeded, whichever occurs first. All systems were trained for a maximum of 8 days (consisting of four runs of 48- hour time- limited compute jobs, which are restarted from the best saved model, i.e. potentially including repeats in the training) with the exception of the \(\mathrm{Li_4P_2O_7}\) , which was trained for 12 days (six 48- hour compute jobs) and the LiPS systems, which were trained for 4 days (two 48- hour compute jobs). We use a batch size of 5 structures for all small molecule tasks, and a batch size of 1 structure for all other tasks. We found small batch sizes to be important for obtaining high predictive accuracy. We also found it important to choose the radial cutoff distance \(r_{c}\) appropriately. A list of the cutoff radii in units of [Å] that were used for the different systems is given in Table VII. + +<|ref|>text<|/ref|><|det|>[[512, 66, 920, 740]]<|/det|> +the initial learning rate of 0.001 by a decay factor of 0.8 whenever the validation RMSE in the forces has not seen an improvement for a given number of epochs: for the small molecule tasks, we set this learning rate patience to 1,000, for all other tasks we use 100. We continuously save the model with the best validation RMSE and use the model with the overall best RMSE for evaluation on the test set and MD simulations. We stop the training if either a maximum number of 50,000 epochs (one epochs equals a full pass over the training set) has been reached, or the validation force RMSE has not improved for 2,500 epochs, or the maximum training time has been exceeded, whichever occurs first. All systems were trained for a maximum of 8 days (consisting of four runs of 48-hour time-limited compute jobs, which are restarted from the best saved model, i.e. potentially including repeats in the training) with the exception of the \(\mathrm{Li_4P_2O_7}\) , which was trained for 12 days (six 48-hour compute jobs) and the LiPS systems, which were trained for 4 days (two 48-hour compute jobs). We use a batch size of 5 structures for all small molecule tasks, and a batch size of 1 structure for all other tasks. We found small batch sizes to be important for obtaining high predictive accuracy. We also found it important to choose the radial cutoff distance \(r_{c}\) appropriately. A list of the cutoff radii in units of [Å] that were used for the different systems is given in Table VII. + +<|ref|>text<|/ref|><|det|>[[514, 772, 920, 916]]<|/det|> +Example of a tensor product interaction. To illustrate that the interactions outlined in this work reduce to a set of five simple operations, we write out the example of a full \(1 \otimes 1 \to 1\) interaction, i.e. a convolution that uses a \(l = 1\) filter to operate on a \(l = 1\) feature, yielding again a \(l = 1\) output. This corresponds to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 65, 488, 83]]<|/det|> +\(l_{i} = l_{f} = l_{o} = 1\) , facilitating a cross-product interaction + +<|ref|>text<|/ref|><|det|>[[56, 91, 488, 108]]<|/det|> +between two \(l = 1\) tensors. In this case, the Clebsch + +<|ref|>equation<|/ref|><|det|>[[243, 149, 916, 240]]<|/det|> +\[C_{(l_{f} = 1,m_{f}),(l_{i} = 1,m_{i})}^{(l_{0} = 1,m_{o})}\propto \epsilon_{o f i} = \left\{ \begin{array}{l l}{1} & {(o,f,i)\in \{(1,2,3),(2,3,1),(3,1,2)\}}\\ {-1} & {(o,f,i)\in \{(1,3,2),(2,1,3),(3,2,1)\}}\\ {0} & {\mathrm{else}} \end{array} \right. \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[56, 302, 488, 345]]<|/det|> +Evaluating equation 8 as well as \(\epsilon_{o f i}\) and using thes relationship \(Y^{(1)}(\hat{r})\propto \hat{r}\) , we recognize the output as + +<|ref|>text<|/ref|><|det|>[[510, 301, 919, 345]]<|/det|> +the vector cross product, taken here between the relative positions and the input feature element \(V_{bc}^{(l = 1)}\) + +<|ref|>equation<|/ref|><|det|>[[240, 384, 916, 466]]<|/det|> +\[\mathcal{L}_{ac}^{(l_{o} = 1)}(\vec{r}_{a},V_{ac}^{(l_{i} = 1)}) = \left\{ \begin{array}{l l}{\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{2}V_{bc3}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{3}V_{bc2}^{(l = 1)}}\\ {\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{3}V_{\mathrm{bc}1}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{1}V_{\mathrm{bc}3}^{(l = 1)}}\\ {\sum_{b\in B}R_{c}(r_{ab})\hat{r}_{1}V_{\mathrm{bc}2}^{(l = 1)} - \sum_{b\in B}R_{c}(r_{ab})\hat{r}_{2}V_{\mathrm{bc}1}^{(l = 1)}} \end{array} \right\} \quad (11)\] + +<|ref|>table<|/ref|><|det|>[[86, 523, 492, 650]]<|/det|> + +
Data SetCutoff
MD-17 [17, 30, 31]4.0
Molecules, CCSD/CCSD(T) [31]4.0
Water+Ices, DeepMD [15, 32]6.0
Formate on Cu5.0
Li4P2O7 [12]5.0
LiPS [13]5.0
Water, data efficiency tests [39]4.5
+ +<|ref|>text<|/ref|><|det|>[[137, 664, 433, 681]]<|/det|> +TABLE VII: Radial cutoff in units of [A]. + +<|ref|>sub_title<|/ref|><|det|>[[195, 740, 377, 755]]<|/det|> +## DATA AVAILABILITY + +<|ref|>text<|/ref|><|det|>[[56, 785, 490, 825]]<|/det|> +The code and data sets will be made available upon publication. + +<|ref|>text<|/ref|><|det|>[[56, 888, 490, 925]]<|/det|> +[1] Richards, W. D. et al. Design and synthesis of the superionic conductor na 10 snp 2 s 12. Nature + +<|ref|>text<|/ref|><|det|>[[510, 525, 919, 920]]<|/det|> +communications 7, 1- 8 (2016). [2] Boero, M., Parrinello, M. & Terakura, K. First principles molecular dynamics study of ziegler- natta heterogeneous catalysis. Journal of the American Chemical Society 120, 2746- 2752 (1998). [3] Lindorff- Larsen, K., Piana, S., Dror, R. O. & Shaw, D. E. How fast- folding proteins fold. Science 334, 517- 520 (2011). [4] Behler, J. & Parrinello, M. Generalized neural- network representation of high- dimensional potential- energy surfaces. Physical review letters 98, 146401 (2007). [5] Bartók, A. P., Payne, M. C., Kondor, R. & Csányi, G. Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons. Physical review letters 104, 136403 (2010). [6] Shapeev, A. V. Moment tensor potentials: A class of systematically improvable interatomic potentials. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 66, 490, 880]]<|/det|> +[7] Thompson, A. P., Swiler, L. P., Trott, C. R., Foiles, S. M. 944 & Tucker, G. J. Spectral neighbor analysis method for automated generation of quantum- accurate interatomic potentials. Journal of Computational Physics 285, 316- 947 330 (2015). [8] Vandermause, J. et al. On- the- fly active learning of interpretable bayesian force fields for atomistic rare- events. npj Computational Materials 6, 1- 11 (2020). [9] Schütt, K. et al. Schnet: A continuous- filters convolutional neural network for modeling quantum- interactions. In Advances in neural information processing systems, 991- 1001 (2017). [10] Unke, O. T. & Meuwly, M. Physnet: A neural network for predicting energies, forces, dipole moments, and partial- charges. Journal of chemical theory and computation 15, 968 3678- 3693 (2019). [11] Klicpera, J., GroB, J. & Gunnemann, S. Directional message passing for molecular graphs. arXiv preprint arXiv:2003.03123 (2020). [12] Mailoa, J. P. et al. A fast neural network approach for direct covariant forces prediction in complex multi- element extended systems. Nature machine intelligence 1, 471- 479 (2019). [13] Park, C. W. et al. Accurate and scalable multi- element graph neural network force field and molecular dynamics with direct force architecture. arXiv preprint arXiv:2007.14444 (2020). [14] Artrit, N. & Kolpak, A. M. Understanding the composition and activity of electrocatalytic nanoalloys in aqueous solvents: A combination of dft and accurate neural network potentials. Nano letters 14, 2670- 2676 (2014). [15] Zhang, L., Han, J., Wang, H., Car, R. & Weinan, E. Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics. Physical review letters 120, 143001 (2018). + +<|ref|>text<|/ref|><|det|>[[512, 68, 921, 895]]<|/det|> +[16] Smith, J. S., Isayev, O. & Roitberg, A. E. Ani- 1: an extensible neural network potential with dft accuracy at force field computational cost. Chemical science 8, 3192- 3203 (2017). [17] Chmiela, S. et al. Machine learning of accurate energy- conserving molecular force fields. Science advances 3, e1603015 (2017). [18] Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry. arXiv preprint arXiv:1704.01212 (2017). [19] Anderson, B., Hy, T. S. & Kondor, R. Cormorant: Covariant molecular neural networks. In Advances in Neural Information Processing Systems, 14537- 14546 (2019). [20] Townshend, R. J., Townshend, B., Eismann, S. & Dror, R. O. Geometric prediction: Moving beyond scalars. arXiv preprint arXiv:2006.14163 (2020). [21] Thomas, N. et al. Tensor field networks: Rotation- and translation- equivariant neural networks for 3d point clouds. arXiv preprint arXiv:1802.08219 (2018). [22] Grisafi, A., Wilkins, D. M., Willatt, M. J. & Ceriotti, M. Atomic- scale representation and statistical learning of tensorial properties. In Machine Learning in Chemistry: Data- Driven Algorithms, Learning Systems, and Predictions, 1- 21 (ACS Publications, 2019). [23] Abadi, M. et al. Tensorflow: Large- scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467 (2016). [24] Paszke, A. et al. Pytorch: An imperative style, high- performance deep learning library. In Advances in neural information processing systems, 8026- 8037 (2019). [25] Weiler, M., Geiger, M., Welling, M., Boomsma, W. & Cohen, T. S. 3d steerable cnns: Learning rotationally equivariant features in volumetric data. In Advances in Neural Information Processing Systems, 10381- 10392 (2018). [26] Kondor, R., Lin, Z. & Trivedi, S. Clebsch- gordan nets: a fully fourier space spherical convolutional neural + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 60, 504, 900]]<|/det|> +network. In Advances in Neural Information Processing Systems, 10117- 10126 (2018). [27] Larsen, A. H. et al. The atomic simulation environment—a python library for working with atoms202 Journal of Physics: Condensed Matter 29, 273002a3 (2017). URL https://doi.org/10.1088%2F1361- 648x%2F2Faa680e. [28] Smidt, T. E., Geiger, M. & Miller, B. K. Findings symmetry breaking order parameters with euclidean202 neural networks. arXiv preprint arXiv:2007.02005a3 (2020). [29] He, K., Zhang, X., Ren, S. & Sun, J. Deep residual202 learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770- 778 (2016). [30] Schütt, K. T., Arbabzadah, F., Chmiela, S., Müller202 K. R. & Tkatchenko, A. Quantum- chemical insights from202 deep tensor neural networks. Nature Communications 8, 13890 (2017). URL https://doi.org/10.1038/ncomms13890. [31] Chmiela, S., Sauceda, H. E., Müller, K.- R. &202 Tkatchenko, A. Towards exact molecular dynamics202 simulations with machine- learned force fields. Nature Communications 9, 3887 (2018). [32] Ko, H.- Y. et al. Isotope effects in liquid water via deep202 potential molecular dynamics. Molecular Physics 117, 1047 3269- 3281 (2019). [33] Christensen, A. S., Bratholm, L. A., Faber, F. A. &202 Anatole von Lilienfeld, O. Fchl revisited: Faster and202 more accurate quantum machine learning. The Journal of Chemical Physics 152, 044107 (2020). [34] Sim, W. S., Gardner, P. & King, D. A. Multiplex202 bonding configurations of adsorbed formate on ag111o1 The Journal of Physical Chemistry 100, 12509- 12516a2 (1996). 00000. [35] Wang, G., Morikawa, Y., Matsumoto, T. & Nakamura202 J. Why is formate synthesis insensitive to copper surfaces202 structures? The Journal of Physical Chemistry B 110, 1056 + +<|ref|>text<|/ref|><|det|>[[504, 60, 923, 899]]<|/det|> +9- 11 (2006). 00050. [36] Yu, X., Bates, J. B., Jellison, G. E. & Hart, F. X. A stable thin- film lithium electrolyte: Lithium phosphorus oxynitride. Journal of The Electrochemical Society 144, 524- 532 (1997). URL https://doi.org/10.1149/1.1837443. [37] Westover, A. S. et al. Plasma synthesis of spherical crystalline and amorphous electrolyte nanopowders for solid- state batteries. ACS Applied Materials & Interfaces 12, 11570- 11578 (2020). URL https://doi.org/10.1021/acsami.9b20812. [38] Li, W., Ando, Y., Minamitani, E. & Watanabe, S. Study of li atom diffusion in amorphous li3po4 with neural network potential. The Journal of chemical physics 147, 214106 (2017). [39] Cheng, B., Engel, E. A., Behler, J., Dellago, C. & Ceriotti, M. Ab initio thermodynamics of liquid and solid water. Proceedings of the National Academy of Sciences 116, 1110- 1115 (2019). [40] Hutter, J., Iannuzzi, M., Schiffmann, F. & VandeVondele, J. cp2k: atomistic simulations of condensed matter systems. WIREs Computational Molecular Science 4, 15- 25 (2014). 00000. [41] Kresse, G. & Hafner, J. Ab initiomolecular dynamics for liquid metals. Physical Review B 47, 558- 561 (1993). URL https://doi.org/10.1103/physrevb.47.558. [42] Kresse, G. & Furthmüller, J. Efficiency of ab- initio total energy calculations for metals and semiconductors using a plane- wave basis set. Computational Materials Science 6, 15- 50 (1996). URL https://doi.org/10.1016/0927- 0256(96)00008- 0. [43] Kresse, G. & Furthmüller, J. Efficient iterative schemes forab initiototal- energy calculations using a plane- wave basis set. Physical Review B 54, 11169- 11186 (1996). URL https://doi.org/10.1103/physrevb.54.11169. [44] Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Physical Review Letters 77, 3865- 3868 (1996). URL https://doi.org/ + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 66, 494, 252]]<|/det|> +1057 10.1103/physrevlett.77.3865. 1058 [45] Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented- wave method. Physical Review B 59, 1758- 1775 (1999). URL https://doi.org/ 1061 10.1103/physrevb.59.1758. 1062 [46] Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014). 1063 + +<|ref|>sub_title<|/ref|><|det|>[[185, 265, 391, 280]]<|/det|> +## ACKNOWLEDGEMENTS + +<|ref|>text<|/ref|><|det|>[[50, 300, 496, 660]]<|/det|> +1065 We thank Jonathan Vandermause, Cheol Woo Park, 1066 David Clark, Kostiantyn Lapchevskyi, Mario Geiger 1067 Joshua Rackers, and Benjamin Kurt Miller for helpful 1068 discussions. 1069 We thank Stefan Chmiela and Alexandre Tkatchenko 1070 for providing the timing data for the Toluene system. 1071 Work at Harvard was supported by Bosch Research 1072 and the Integrated Mesoscale Architectures for 1073 Sustainable Catalysis (IMASC), an Energy Frontier 1074 Research Center funded by the US Department of 1075 Energy (DOE), Office of Science, Office of Basic Energy 1076 Sciences under Award No. DE- SC0012573. N.M. 1077 acknowledges support from the Department of the Navy; 1078 Office of Naval Research. + +<|ref|>text<|/ref|><|det|>[[50, 666, 496, 900]]<|/det|> +1079 Work at Bosch Research was partially supported by 1080 ARPA- E Award No. DE- AR0000775 and used resources 1081 of the Oak Ridge Leadership Computing Facility at Oak 1082 Ridge National Laboratory, which is supported by the 1083 Office of Science of the Department of Energy under 1084 Contract DE- AC05- 00OR22725. 1085 T.E.S. was supported by the Laboratory Directed 1086 Research and Development Program of Lawrence 1087 Berkeley National Laboratory and the Center 1088 for Advanced Mathematics for Energy Research + +<|ref|>text<|/ref|><|det|>[[514, 65, 919, 108]]<|/det|> +Applications, both under U.S. Department of Energy Contract No. DE- AC02- 05CH11231. + +<|ref|>text<|/ref|><|det|>[[514, 116, 920, 258]]<|/det|> +The authors acknowledge computing resources provided by the Harvard University FAS Division of Science Research Computing Group and by the Texas Advanced Computing Center (TACC) at The University of Texas at Austin under allocations DMR20009 and DMR20013. + +<|ref|>sub_title<|/ref|><|det|>[[599, 291, 835, 307]]<|/det|> +## AUTHOR CONTRIBUTIONS + +<|ref|>text<|/ref|><|det|>[[514, 333, 920, 828]]<|/det|> +S.B. initiated the project, conceived the NequIP model, implemented the software and conducted all software experiments under the guidance of B.K. T.E.S. contributed to the conception of the model, guidance of computational experiments, and the software implementation. L.S. created the data set for formate/Cu, guided work on training and MD simulations on this system, and contributed to development of the software implementation. J.P.M. guided the work on the LiPS conductor and implemented the thermostat for MD simulations together with S.B.. M.K. created the AIMD data set of \(\mathrm{Li_4P_2O_7}\) , wrote software for the analysis of MD results and guided the benchmarking on this system. N.M. wrote software for the computation of the diffusion results and guided discussions on the interpretation of results. B.K. supervised the project from conception to design of experiments, implementation, theory, as well as analysis of data. All authors contributed to the manuscript and the discussion of results. + +<|ref|>sub_title<|/ref|><|det|>[[608, 861, 825, 876]]<|/det|> +## COMPETING INTERESTS + +<|ref|>text<|/ref|><|det|>[[531, 904, 847, 920]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 144, 69]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[45, 95, 945, 260]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 293, 115, 313]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 335, 936, 401]]<|/det|> +Left: a set of atoms is interpreted as an atomic graph with local neighborhoods. Middle: every atom carries a set of scalar and vector features with it. Right: atoms exchange information via filters, that are again scalars and vectors. The interactions of features and filters define five interactions. + +<|ref|>image<|/ref|><|det|>[[55, 409, 945, 750]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 772, 118, 792]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[41, 813, 956, 949]]<|/det|> +The NequIP network architecture. Left: atomic numbers are embedded into \(I = 0\) features, which are refined through a series of interaction blocks, creating \(I = 0\) and \(I = 1\) features. An output block generates atomic energies, which are pooled to give the total predicted energy. Middle: the interaction block consists of a series of convolutions, interweaved with self- interaction layers, equivariant nonlinearities and concatenation. Right: the convolution combines the radial function \(R(r)\) which operates only on interatomic distances with the spherical harmonics based on unit vector \(r^*\) via a tensor product. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 52, 955, 270]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 295, 118, 315]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 337, 941, 427]]<|/det|> +Perspective view of atomic configurations of (a) bidentate HCOO (b) monodentate HCOO and (c) CO2 and a hydrogen adatom on a Cu(110) surface. The blue, red, black, and white spheres represent Cu, O, C, and H atoms, respectively. The subset shown in each subplot is the corresponding top view along the < \(110>\) orientation. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 55, 905, 767]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 800, 117, 820]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 945, 887]]<|/det|> +Quenched glass structure of Li4P2O7. The insets show the P- O- O tetrahedral bond angle (bottom left) as well as the O- P- P bridging angle between corner- sharing phosphate tetrahedra (top right). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 50, 940, 180]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 202, 119, 222]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 243, 940, 286]]<|/det|> +Left: Radial Distribution Function, middle: Angular Distribution Function, bridging oxygen, right: Angular Distribution Function, tetrahedral bond angle. All are defined as probability density functions. + +<|ref|>image<|/ref|><|det|>[[50, 294, 943, 733]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 760, 118, 780]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[42, 803, 770, 823]]<|/det|> +Comparison of Lithium mean square displacement of AIMD and NequIP trajectories. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 48, 950, 415]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 437, 117, 457]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[42, 480, 955, 523]]<|/det|> +Log- log plot of the predictive error in forces of NequIP with \(l = 0\) vs. \(l = 0 / l = 1\) interactions as a function of data set size, measured via the force MAE. + +<|ref|>sub_title<|/ref|><|det|>[[44, 546, 311, 573]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 596, 765, 617]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 635, 259, 654]]<|/det|> +- codesubmission.zip + +<--- Page Split ---> diff --git a/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/images_list.json b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d2119704a0349beda1788c6a26f75436f6174501 --- /dev/null +++ b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Experimental setup. (A) Schematic of the micro- and nanofabricated reactor chip. It is comprised of a U-shaped microfluidic in- and a simple straight outlet system that connects to the model catalyst bed (the nanofluidic reactor with the reaction zone) on either side, as well as to the high-pressure gas handling system and the QMS, respectively. The cross-sectional dimensions of the microfluidic channels are also shown. (B) Schematic depiction of the \\(30\\mu \\mathrm{m}\\) long well-mixed reaction zone (the model catalyst bed that contains the nanoparticle(s)) with cross sectional dimensions \\(10\\mu \\mathrm{m}\\times 200\\mathrm{nm}\\) , as well as its connections to the microfluidic in- and outlet system via a smaller microfluidic inlet channel (also with cross sectional dimensions \\(10\\mu \\mathrm{m}\\times 200\\mathrm{nm}\\) ) and via a nanofluidic outlet \"capillary\" with cross sectional dimensions \\(500\\mathrm{nm}\\times 200\\mathrm{nm}\\) . Note that the schematics are not drawn to scale. (C) Dark-field scattering microscopy image of a nanofluidic catalyst bed containing 1000 nanoparticles arranged in a regular array in the reaction zone. The inset depicts a side-view scanning electron microscopy (SEM) image of the used \\(\\mathrm{Au / SiO_2 / Pd}\\) hybrid nanostructures. Scale bar 100 nm. The catalytically inert Au element enables single-particle dark-field scattering (DF) imaging to confirm the presence of the correct number of catalyst particles inside the sealed nanofluidic reactor, without itself participating in the reaction, as we have shown earlier [41]. (D) Schematic depiction and zoomed-in DF image of the reaction zone of the nanofluidic catalyst bed used in our experiments, with a regular array of \\(n = 1000\\) nanoparticles that become distinctly visible in the DF image. Also shown are two top-view SEM images of a similar array of nanoparticles, as well as a side-view of a single nanoparticle, prepared on an open surface to facilitate SEM imaging. Scale bar 100 nm. (E) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing \\(n = 10\\) Pd nanoparticles used in our experiments. (F) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing a single \\((n = 1)\\) Pd nanoparticle used in our experiments. (G) Schematic depiction and zoomed-in DF image of the reaction zone of the empty catalyst bed without Pd nanoparticles \\((n = 0)\\) , used in our experiments as a negative control.", + "footnote": [], + "bbox": [ + [ + 202, + 92, + 816, + 450 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 CO oxidation experiments at different temperatures on 1000 and 10 Pd nanoparticles. CO and \\(\\mathrm{O_2}\\) pulse sequence applied at a reactor temperature of (A) \\(T = 450^{\\circ}\\mathrm{C}\\) , (B) \\(T = 340^{\\circ}\\mathrm{C}\\) , (C) \\(T = 280^{\\circ}\\mathrm{C}\\) . Note that we start all experiments with a pure \\(6\\%\\) CO pulse in Ar carrier gas and subsequently decrease the relative CO concentration and increase the relative \\(\\mathrm{O_2}\\) concentration, such that \\(0 \\leq \\alpha_{CO} = P_{CO} / (P_{O_2} + P_{CO}) \\leq 1\\) , while keeping the total reactant concentration constant at \\(6\\%\\) . Corresponding baseline-adjusted (BA - see Methods for explanation of the BA-procedure) \\(\\mathrm{CO_2}\\) counts measured by the QMS for 1000 Pd nanoparticles at (D) \\(T = 450^{\\circ}\\mathrm{C}\\) , (E) \\(T = 340^{\\circ}\\mathrm{C}\\) and (F) \\(T = 280^{\\circ}\\mathrm{C}\\) . We note distinct responses at all three temperatures with the expected decrease in reaction rate for lower \\(T\\) . The \"transient\" overshoots observed for small \\(\\alpha_{CO}\\) values are the consequence of mass flow controller \"dial-in\" to the correct flow rate. (G-I) Corresponding \\(\\mathrm{CO_2}\\) counts measured by the QMS for the 10 Pd nanoparticle sample. While at \\(T = 450^{\\circ}\\mathrm{C}\\) a relatively clear \\(\\mathrm{CO_2}\\) signal is still obtained, it becomes increasingly weaker at the lower temperatures. (J) Mean value of the measured \\(\\mathrm{CO_2}\\) QMS counts for each reactant pulse plotted vs. the corresponding \\(\\alpha_{CO}\\) value for both the \\(n = 1000\\) and the \\(n = 10\\) Pd nanoparticle samples and obtained at \\(T = 450^{\\circ}\\mathrm{C}\\) . Shaded area depicts the standard deviation of measured BA-counts across each respective pulse. We note a maximum in the \\(\\mathrm{CO_2}\\) formation rate at \\(\\alpha_{CO}^{max} = 0.65\\) for both \\(n = 1000\\) and \\(n = 10\\) , as well as a significantly larger deviation of the derived counts for \\(n = 10\\) . (K) Same as (J) but for \\(T = 340^{\\circ}\\mathrm{C}\\) . We note a shift of \\(\\alpha_{CO}^{max}\\) to lower values for both samples due to increasing CO poisoning and a broadening/flattening of the overall trend for \\(n = 10\\) , which is the consequence of the system approaching the detection limit of the QMS and the consequent significant uncertainty in the derived counts. (L) Same as (J) and (K) but for \\(T = 280^{\\circ}\\mathrm{C}\\) . We note an even further reduction of \\(\\alpha_{CO}^{max}\\) and further uncertainty increase for \\(n = 10\\) .", + "footnote": [], + "bbox": [ + [ + 208, + 87, + 833, + 362 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 The constrained denoising auto-encoder. The underlying signal that we want to access in our experiments, i.e., the \\(\\mathrm{CO_2}\\) production rate step functions as a function of time, is corrupted by both correlated and uncorrelated noise in the experimentally measured readout from the QMS. Therefore, the experimental QMS readout is input to a denoising autoencoder, which transforms the input into a minimally dimensioned representation of the underlying signal and then reconstructs it sans corruption, which means, it reconstructs the original signal, rather then the QMS readout that also contains different types of noise. The DAE is trained with a standard L2 norm between reconstructed and underlying signal, and the latent space is constrained by an L1 norm and a consistency loss which forces the latent space to take on the form of the underlying signal.", + "footnote": [], + "bbox": [ + [ + 217, + 87, + 829, + 303 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Direct comparison of standard and DAE-enhanced QMS readout for 1000 and 10 Pd nanoparticles. (A) Baseline-adjusted (BA - see Methods for explanation) raw QMS counts for \\(\\mathrm{CO_2}\\) (grey line) together with the mean \\(\\mathrm{CO_2}\\) signal (blue line - obtained as the average measured BA-count across each pulse) for \\(n = 1000\\) Pd nanoparticles across the entire \\(\\alpha_{CO}\\) range, \\(\\alpha_{CO}\\in (0,1)\\) and at \\(450^{\\circ}\\mathrm{C}\\) . (B) BA-QMS counts for \\(\\mathrm{CO_2}\\) (grey line) together with the mean \\(\\mathrm{CO_2}\\) signal (green line) for \\(n = 10\\) Pd nanoparticles across the entire \\(\\alpha_{CO}\\) range and at \\(450^{\\circ}\\mathrm{C}\\) . (C) BA-QMS counts for \\(\\mathrm{CO_2}\\) (grey line) together with the \\(\\mathrm{CO_2}\\) signal denoised by the DAE (orange line) for \\(n = 10\\) across the entire \\(\\alpha_{CO}\\) range, \\(\\alpha_{CO}\\in (0,1)\\) , and at \\(450^{\\circ}\\mathrm{C}\\) . (D-F) Same as (A-C) but at \\(280^{\\circ}\\mathrm{C}\\) . (G) BA-mean \\(\\mathrm{CO_2}\\) counts for \\(\\alpha_{CO}\\) sweeps at reactor temperatures ranging from \\(280^{\\circ}\\mathrm{C}\\) to \\(450^{\\circ}\\mathrm{C}\\) in \\(20^{\\circ}\\mathrm{C}\\) steps for \\(n = 1000\\) . Red dots indicate the \\(\\alpha_{CO}\\) value corresponding to the highest reaction rate, \\(\\alpha_{CO}^{max}\\) . (H) Same as (G) but for \\(n = 10\\) , using \\(10^{\\circ}\\mathrm{C}\\) temperature steps. (I) Same as (H) but with the QMS signal denoised by the DAE. (J) \\(\\alpha_{CO}^{max}\\) values extracted from (G) for all temperatures for \\(n = 1000\\) , plotted as function of \\(\\alpha_{CO}\\) . We note the constant stoichiometric \\(\\alpha_{CO}^{max}\\) value of 0.65 down to \\(T = 360^{\\circ}\\mathrm{C}\\) . Below this temperature, we find a systematic shift to lower \\(\\alpha_{CO}^{max}\\) values, as a consequence of increasing CO poisoning. (K) Same as (J) but for \\(n = 10\\) . We note the qualitatively similar trend compared to \\(n = 1000\\) but also the significantly higher spread in the data points. (L) Same as (K) but based on the DAE-denoised QMS signal in (I). Clearly, the uncertainty in \\(\\alpha_{CO}^{max}\\) is significantly reduced and the T-dependent trend of \\(\\alpha_{CO}^{max}\\) for both \\(n = 1000\\) and \\(n = 10\\) is now very similar. The small discrepancies are discussed in the text.", + "footnote": [], + "bbox": [ + [ + 181, + 100, + 734, + 362 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Online mass spectrometry from a single Pd nanoparticle. (A) Baseline-adjusted (BA) raw QMS counts for \\(\\mathrm{CO_2}\\) (grey line) together with the mean \\(\\mathrm{CO_2}\\) signal (green line - obtained as average measured BA-count across each pulse) for \\(n = 1\\) Pd nanoparticle, measured across the entire \\(\\alpha_{CO}\\) range, \\(\\alpha_{CO}\\in (0,1)\\) , and at \\(450^{\\circ}\\mathrm{C}\\) . (B) Same as (A) but for an empty nanochannel, i.e., \\(n = 0\\) . Note the very similar and stochastic appearance of apparent \\(\\mathrm{CO_2}\\) pulses both for \\(n = 1\\) and \\(n = 0\\) when analyzed in the standard way. (C) Mean BA-CO2 counts extracted for each \\(\\alpha_{CO}\\) pulse based on the standard analysis across the entire \\(\\alpha_{CO}\\) range, \\(\\alpha_{CO}\\in (0,1)\\) , and at \\(450^{\\circ}\\mathrm{C}\\) , for both \\(n = 0\\) (black lines) and \\(n = 1\\) (green lines). For \\(n = 1\\) we executed 7 consecutive \\(\\alpha_{CO}\\) sweeps and for \\(n = 0\\) we executed 5 consecutive \\(\\alpha_{CO}\\) sweeps. As key observation, we note high reproducibility of the \\(\\alpha_{CO}\\) sweeps for both \\(n = 0\\) and \\(n = 1\\) , as well as that no clear activity trend as function of \\(\\alpha_{CO}\\) is resolved for the \\(n = 1\\) sample, despite a generally slightly higher number of counts compared the \\(n = 0\\) reference. (D) Same as (A) but for DAE-deniosed BA-QMS \\(\\mathrm{CO_2}\\) counts (purple line). Note the significantly smaller but at the same time non-stochastic \\(\\mathrm{CO_2}\\) pulses resolved by the DAE compared to the standard analysis. (E) Same as (D) but for an empty nanochannel, i.e., \\(n = 0\\) . Note that using the DAE, a flat baseline at zero BA-counts is obtained for the empty nanochannel. (F) Same as (C) but for DAE-deniosed data, where distinct reaction rate maxima are resolved for \\(n = 1\\) (purple curves), with \\(\\alpha_{CO}^{max}\\) ranging between 0.65 and 0.6 for temperatures between \\(450^{\\circ}\\mathrm{C}\\) and \\(410^{\\circ}\\mathrm{C}\\) (inset), in good agreement with the earlier results for \\(n = 10\\) and \\(n = 1000\\) . For lower temperatures, as well as for \\(n = 0\\) (black lines representing multiple sweeps at \\(450^{\\circ}\\mathrm{C}\\) ), the DAE outputs counts \\(< 1\\) , which is physically unreasonable and thus defined as the limit of detection. (G) \\(\\alpha_{CO}^{max}\\) values for \\(n = 1000\\) (blue), \\(n = 10\\) DAE-deniosed (orange) and \\(n = 1\\) DAE-deniosed (purple) for \\(400 \\leq T \\leq 450^{\\circ}\\mathrm{C}\\) , i.e., the range for which the \\(n = 1\\) DAE-signal is above the limit of detection. (H) \\(\\mathrm{CO_2}\\) counts normalized by the number of particles on the respective sample, obtained at \\(\\alpha_{CO}^{max}\\) for \\(n = 1000\\) (blue), \\(n = 10\\) DAE-deniosed (orange) and \\(n = 1\\) DAE-deniosed (purple) for \\(400 \\leq T \\leq 450^{\\circ}\\mathrm{C}\\) .", + "footnote": [], + "bbox": [ + [ + 225, + 144, + 772, + 400 + ] + ], + "page_idx": 12 + } +] \ No newline at end of file diff --git a/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d.mmd b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d.mmd new file mode 100644 index 0000000000000000000000000000000000000000..37940c7c42492face81bdc0b1a6d074be42ad8ba --- /dev/null +++ b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d.mmd @@ -0,0 +1,363 @@ + +# Deep-learning enabled online mass spectrometry of the reaction product from a single catalyst nanoparticle + +Christoph Langhammer clangham@chalmers.se + +Chalmers University of Technology https://orcid.org/0000- 0003- 2180- 1379 Henrik Klein Moberg Chalmers University of Technology Ievgen Nedrygailov Chalmers University of Technology David Albinsson Chalmers University of Technology Joachim Fritzsche Chalmers Univerity of Technology https://orcid.org/0000- 0001- 8660- 2624 + +## Article + +Keywords: Single Particle Catalysis, Nanofluidic Reactors, CO Oxidation, Palladium Nanoparticles, Deep Learning, Denoising Auto- Encoder, Spectrometry, Heterogeneous Catalysis. + +Posted Date: March 29th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 3797128/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on August 5th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 62602- 3. + +<--- Page Split ---> + +# Deep-learning enabled online mass spectrometry of the reaction product from a single catalyst nanoparticle + +Henrik Klein Moberg, Ievgen Nedrygailov, David Albinson, Joachim Fritzsche, Christoph Langhammer\* + +Department of Physics, Chalmers University of Technology, Göteborg, SE- 41296 Sweden. + +\*Corresponding author(s). E- mail(s): clangham@chalmers.se; + +Contributing authors: henrik.kleinmoberg@chalmers.se; + +nedrygailov@gmail.com; david@envue- technologies.com; + +joafri@chalmers.se; + +## Abstract + +Extracting tiny signals from noise is a generic challenge in experimental science. In catalysis, this challenge manifests itself as the need to quantify chemical reactions on nanoscopic surface areas, such as single nanoparticles or even single atoms. Here, we address this challenge by combining the ability of nanofluidic reactors to focus reaction product from tiny catalyst surfaces towards online mass spectrometric analysis with the unrivalled ability of a constrained denoising autoencoder to discern tiny signals from noise. Using CO oxidation on Pd as model reaction, we demonstrate that the catalyst surface area required for online mass spectrometry can be reduced by \(\approx 3\) orders of magnitude compared to state of the art, down to a single nanoparticle with \(0.0072 \pm 0.00086 \mu \mathrm{m}^2\) surface area. These results constitute a new paradigm for online reaction analysis in single particle catalysis and advocate deep learning to improve resolution in mass spectrometry in general. + +Keywords: Single Particle Catalysis, Nanofluidic Reactors, CO Oxidation, Palladium Nanoparticles, Deep Learning, Denoising Auto- Encoder, Spectrometry, Heterogeneous Catalysis. + +<--- Page Split ---> + +## 1 Introduction + +The ability to detect and analyze weak signals amidst high levels of noise has led to groundbreaking discoveries across a myriad of scientific disciplines using widely different experimental methodologies [1- 10]. Mass spectrometry constitutes such an experimental method since it enables the detection of very small amounts of particles, and has become a workhorse from (astro)physics to biology to chemistry [11, 12]. Accordingly, the method comes in different variants tailored for the species subject to characterization and the operational conditions. Mechanistically, it counts ionized particles, molecules or atoms in vacuum by separating them according to their mass/charge ratio, and thereby determines their (molecular) weight. + +In catalysis and surface science, mass spectrometry is widely used to quantitatively analyze the composition of molecules desorbing from surfaces, often from a surfacechemical reaction [13, 14]. In this application, usually a quadrupole is used for mass selection due to its high resolution and sensitivity [15, 16]. Such quadrupole mass spectrometers (QMS) have been instrumental in the development of our understanding of nanocatalytic processes. Nevertheless, the technical advance of QMS systems used for so- called residual gas analysis has been limited during the last decades and, therefore, no significant progress has been made in terms of improving their limit of detection and their ability to discern very weak signals from noise [24, 25]. Concurrently, this state- of- the- art (SotA) is becoming an increasingly limiting factor since accurately measuring catalytic activity and selectivity from small amounts of precious new catalyst materials, such as size- or shape- selected model catalysts [26] or single atom catalysts [27- 29] that can only be obtained in small quantities, is crucial for the development of new catalyst materials and for advancing our understanding of catalytic processes. To this end, in the most extreme implementation, the aim of so- called single particle catalysis is to study catalytic reactions on individual nanoparticles to overcome ensemble averaging effects [30]. + +The first strategy to enable such experiments employs detection of photon or electron signals that report on either the product molecules formed, on reactant molecules consumed, on the catalyst particle itself, or on temperature changes generated by the reaction [30- 35]. While both elegant and effective for the specific reactions chosen, these approaches lack the wide and generic applicability of mass spectrometry. Furthermore, they often cannot deliver information about both amounts and molecular weight of the species involved in the reaction. + +To address this limitation, miniaturization of the reactor to the level of microreactors [36- 38] and even nanoreactors [39] to reduce the catalyst surface area necessary to obtain measurable (QMS) signals has been implemented. Very high sensitivity can be obtained in such systems by dramatically reducing reactor volume and thereby directing the entire gas flow from the catalyst bed to a QMS to maximize the fraction of reaction product available for analysis. In this way, as we recently have demonstrated using a single nanofluidic channel as the catalyst bed, the minimal required active catalyst surface area for online QMS measurements was reduced by ca. 1.5 orders + +<--- Page Split ---> + +compared to the lower limit proposed for microreactors [36], i.e., down to ca. \(10\mu \mathrm{m}^{2}\) active surface area [40, 41]. + +While this indeed was a significant advance, it is still 2- 3 orders away from the ultimate dream of online QMS measurements from single nanoparticles. To overcome this gap we employ here a new type of nanofluidic reactor and combine it with a constrained convolutional denoising auto- encoder. This harnesses the reaction product focusing capability of nanofluidic reactors with the versatility, high- resolution and sensitivity of the QMS, with the power of deep learning to detect and analyze very weak signals hidden in noise. As we demonstrate on the example of CO oxidation, i.e., \(CO + \frac{1}{2} O_{2} \rightarrow CO_{2}\) , over a Pd model catalyst, this approach leapfrogs the limit of detection of a SotA QMS system by \(\approx 3\) orders, and enables online mass spectrometric analysis of reaction product from a single Pd nanoparticle with active surface area \(A \approx 0.0072 \pm 0.00086\mu \mathrm{m}^{2}\) \((7200 \pm 860\mathrm{nm}^{2})\) . + +Finally, we note that the application of machine learning to different modalities of mass spectrometry has a short but lively history [42],[43] [44], [45], [46- 49] but the application of machine learning to discern small QMS signals from noise, as we do here, is hitherto completely unexplored. + +## Results + +The reaction of CO with \(\mathrm{O}_{2}\) over Pd catalysts forming \(\mathrm{CO}_{2}\) is one of the most studied reactions in catalysis both due to its practical relevance and due to its role in studies of structure- function correlations and (surface) oxide formation [50, 51]. Therefore, we chose it as our model reaction with the aim to enable QMS measurements of the \(\mathrm{CO}_{2}\) formed on a single Pd nanoparticle. We have developed a nanofluidic reactor connected to U- shape microfluidic in- and a straight outlet channel(s), which are connected to a macroscopic reactant inlet system operated by mass flow controllers at 3 bar, and to a QMS mounted on a UHV chamber with \(1 \cdot 10^{- 10}\) mbar base pressure (Figure 1A and Supplementary Figure 1). This in- and outlet system is connected to a nanofluidic catalyst bed with \(30 \mu \mathrm{m} \times 10 \mu \mathrm{m} \times 200 \mathrm{nm}(L \times W \times H)\) dimensions via a nanofluidic inlet channel ( \(505 \mu \mathrm{m} \times 10 \mu \mathrm{m} \times 200 \mathrm{nm}\) ) and an outlet channel with dimensions \(100 \mu \mathrm{m} \times 500 \mathrm{nm} \times 200 \mathrm{nm}\) (Figure 1B, C). This design ensures that the catalyst is operated in the well- mixed regime, where concentration gradients due to reactant conversion are effectively eliminated [40]. Using a resistive heater, the nanoreactor zone can be heated up to \(450^{\circ} \mathrm{C}\) [41]. For this study, we have fabricated four nanoreactor chips with identical gas in- and outlet systems, as well as catalyst beds. They were decorated with (regular arrays of) \(n = 1000, 10, 1\) or \(0 \mathrm{Pd}\) particles with diameter \(d = 59.4 \pm 3.7 \mathrm{nm}\) and height \(h = 23.5 \pm 1.6 \mathrm{nm}\) , as SEM image analysis reveals, grown onto a \(8 \mathrm{nm}\) thick \(\mathrm{SiO}_{2}\) support layer that separates them from a chemically inert plasmonic Au nanoparticles with \(d = 97.6 \pm 6.6 \mathrm{nm}\) and \(h = 35.3 \pm 2.3 \mathrm{nm}\) (Figure 1C- G, Supplementary Figure 2 and Methods), as we have used earlier ([41, 52]). This corresponds to an active Pd surface area \(A \approx 0.0072 \pm 0.00086 \mu \mathrm{m}^{2}\) \((7200 \pm 860 \mathrm{nm}^{2})\) per particle. We chose this particular hybrid nanoparticle design, where the Au element serves as plasmonic light scatterer (Pd is optically dark [53]), to enable the verification + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Experimental setup. (A) Schematic of the micro- and nanofabricated reactor chip. It is comprised of a U-shaped microfluidic in- and a simple straight outlet system that connects to the model catalyst bed (the nanofluidic reactor with the reaction zone) on either side, as well as to the high-pressure gas handling system and the QMS, respectively. The cross-sectional dimensions of the microfluidic channels are also shown. (B) Schematic depiction of the \(30\mu \mathrm{m}\) long well-mixed reaction zone (the model catalyst bed that contains the nanoparticle(s)) with cross sectional dimensions \(10\mu \mathrm{m}\times 200\mathrm{nm}\) , as well as its connections to the microfluidic in- and outlet system via a smaller microfluidic inlet channel (also with cross sectional dimensions \(10\mu \mathrm{m}\times 200\mathrm{nm}\) ) and via a nanofluidic outlet "capillary" with cross sectional dimensions \(500\mathrm{nm}\times 200\mathrm{nm}\) . Note that the schematics are not drawn to scale. (C) Dark-field scattering microscopy image of a nanofluidic catalyst bed containing 1000 nanoparticles arranged in a regular array in the reaction zone. The inset depicts a side-view scanning electron microscopy (SEM) image of the used \(\mathrm{Au / SiO_2 / Pd}\) hybrid nanostructures. Scale bar 100 nm. The catalytically inert Au element enables single-particle dark-field scattering (DF) imaging to confirm the presence of the correct number of catalyst particles inside the sealed nanofluidic reactor, without itself participating in the reaction, as we have shown earlier [41]. (D) Schematic depiction and zoomed-in DF image of the reaction zone of the nanofluidic catalyst bed used in our experiments, with a regular array of \(n = 1000\) nanoparticles that become distinctly visible in the DF image. Also shown are two top-view SEM images of a similar array of nanoparticles, as well as a side-view of a single nanoparticle, prepared on an open surface to facilitate SEM imaging. Scale bar 100 nm. (E) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing \(n = 10\) Pd nanoparticles used in our experiments. (F) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing a single \((n = 1)\) Pd nanoparticle used in our experiments. (G) Schematic depiction and zoomed-in DF image of the reaction zone of the empty catalyst bed without Pd nanoparticles \((n = 0)\) , used in our experiments as a negative control.
+ +<--- Page Split ---> + +of the presence of the anticipated number of particles in the enclosed nanoreactors using dark field scattering microscopy (Figure 1D- G). + +## CO Oxidation on Pd model Catalysts in Nanofluidic Reactors + +To prepare these systems for experiments in reaction environment, we first exposed them to a conditioning sequence, followed by a full CO oxidation sequence at \(280^{\circ}\mathrm{C}\) (see Methods and Supplementary Figures 3- 5). Subsequently, we initiated the experiment sequence consisting of 15 min \(\mathrm{CO / O_2}\) mixture pulses at a constant \(6\%\) reactant concentration and starting with \(6\%\) percent CO, separated by 15 min in Ar. Such sequences were executed from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) in \(20^{\circ}\mathrm{C}\) steps for the \(\mathrm{n = 1000}\) sample and \(10^{\circ}\mathrm{C}\) steps for the \(\mathrm{n = 10}\) and \(\mathrm{n = 1}\) samples, to sweep the entire \(0\leq \alpha_{CO} =\) \(P_{CO} / (P_{O_2} + P_{CO})\leq 1\) range in steps of 0.05, where \(\alpha_{CO}\) is the relative CO concentration in the gas mixture. (Figure 2A- C). At each new temperature a single 30 min pulse of \(4\%\) \(\mathrm{O_2}\) and \(2\%\) CO in Ar carrier gas, followed by a 15 min pure Ar pulse, was applied to reset the state of the catalyst. + +Focusing first on the highest reaction temperature of \(450^{\circ}\mathrm{C}\) and \(n = 1000\) (total active surface area of \(A\approx 7200\pm 860\mathrm{nm}^2\cdot 1000 = 7.2\pm 0.86\mu \mathrm{m}^2)\) , we observe a QMS response distinctly above the noise floor for all \(\alpha_{CO}\in (0,1)\) values (Figure 2D). Furthermore, starting from \(\alpha_{CO} = 1\) , we observe the characteristic reaction rate increase as more \(\mathrm{O_2}\) is added until the maximum rate is reached at \(\alpha_{CO}^{max} = 0.65\) . This is very close to the stoichiometric \(\alpha_{CO} = 0.66\) (the small difference is the consequence of the 0.05 \(\alpha_{CO}\) steps), which indicates that CO poisoning is very mild at this temperature and that mass transport gradients are negligible, as expected for a well- mixed catalyst bed. Upon decreasing \(\alpha_{CO}\) beyond \(\alpha_{CO}^{max}\) , the reaction rate decreases in an essentially linear fashion. + +Decreasing temperature to \(340^{\circ}\mathrm{C}\) induces both a global decrease in reaction rate and a shift of \(\alpha_{CO}^{max}\) to a lower value of 0.5, due to stronger CO poisoning (Figure 2E). This trend continues upon temperature reduction to \(280^{\circ}\mathrm{C}\) (Figure 2F) with a QMS signal still distinctly above the noise floor and \(\alpha_{CO}^{max}\) reduced to 0.3. These measurements thus corroborate our experimental setup and approach since they deliver results in agreement with the well- established understanding of the CO oxidation over Pd catalysts [51, 54]. Hence, they constitute a relevant baseline as we further reduce the catalyst surface area towards a single Pd nanoparticle. + +As the first step, we repeated the experiments for an identical nanoreactor but with \(n = 10\) , which corresponds to \(A\approx 7200\pm 860\mathrm{nm}^2\cdot 10 = 0.072\pm 0.0086\mu \mathrm{m}^2\) (Figure 2G- I). We find that at \(450^{\circ}\mathrm{C}\) , \(\mathrm{CO_2}\) pulses still are resolved and that the counts is approximately two orders smaller compared to \(n = 1000\) . Importantly, however, we also see that the \(\mathrm{CO_2}\) signal induced by the reactant pulses approaches the noise floor. Accordingly, reducing temperature makes discerning \(\mathrm{CO_2}\) produced increasingly difficult. This becomes clear when extracting the \(\mathrm{CO_2}\) counts for each \(\alpha_{CO}\) pulse and directly comparing the \(\mathrm{CO_2}\) counts vs. the \(\alpha_{CO}\) trend for the \(n = 1000\) and \(n = 10\) samples. At \(T = 450^{\circ}\mathrm{C}\) , the trends obtained are qualitatively very similar and thus corroborate that the catalyst is operated at identical conditions (Figure 2J). It is + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 CO oxidation experiments at different temperatures on 1000 and 10 Pd nanoparticles. CO and \(\mathrm{O_2}\) pulse sequence applied at a reactor temperature of (A) \(T = 450^{\circ}\mathrm{C}\) , (B) \(T = 340^{\circ}\mathrm{C}\) , (C) \(T = 280^{\circ}\mathrm{C}\) . Note that we start all experiments with a pure \(6\%\) CO pulse in Ar carrier gas and subsequently decrease the relative CO concentration and increase the relative \(\mathrm{O_2}\) concentration, such that \(0 \leq \alpha_{CO} = P_{CO} / (P_{O_2} + P_{CO}) \leq 1\) , while keeping the total reactant concentration constant at \(6\%\) . Corresponding baseline-adjusted (BA - see Methods for explanation of the BA-procedure) \(\mathrm{CO_2}\) counts measured by the QMS for 1000 Pd nanoparticles at (D) \(T = 450^{\circ}\mathrm{C}\) , (E) \(T = 340^{\circ}\mathrm{C}\) and (F) \(T = 280^{\circ}\mathrm{C}\) . We note distinct responses at all three temperatures with the expected decrease in reaction rate for lower \(T\) . The "transient" overshoots observed for small \(\alpha_{CO}\) values are the consequence of mass flow controller "dial-in" to the correct flow rate. (G-I) Corresponding \(\mathrm{CO_2}\) counts measured by the QMS for the 10 Pd nanoparticle sample. While at \(T = 450^{\circ}\mathrm{C}\) a relatively clear \(\mathrm{CO_2}\) signal is still obtained, it becomes increasingly weaker at the lower temperatures. (J) Mean value of the measured \(\mathrm{CO_2}\) QMS counts for each reactant pulse plotted vs. the corresponding \(\alpha_{CO}\) value for both the \(n = 1000\) and the \(n = 10\) Pd nanoparticle samples and obtained at \(T = 450^{\circ}\mathrm{C}\) . Shaded area depicts the standard deviation of measured BA-counts across each respective pulse. We note a maximum in the \(\mathrm{CO_2}\) formation rate at \(\alpha_{CO}^{max} = 0.65\) for both \(n = 1000\) and \(n = 10\) , as well as a significantly larger deviation of the derived counts for \(n = 10\) . (K) Same as (J) but for \(T = 340^{\circ}\mathrm{C}\) . We note a shift of \(\alpha_{CO}^{max}\) to lower values for both samples due to increasing CO poisoning and a broadening/flattening of the overall trend for \(n = 10\) , which is the consequence of the system approaching the detection limit of the QMS and the consequent significant uncertainty in the derived counts. (L) Same as (J) and (K) but for \(T = 280^{\circ}\mathrm{C}\) . We note an even further reduction of \(\alpha_{CO}^{max}\) and further uncertainty increase for \(n = 10\) .
+ +also visible that the counts vs. \(\alpha_{CO}\) curve is slightly wider and flatter, and exhibits a significantly larger standard deviation (shaded area) for \(n = 10\) , in particular in the low reaction rate regimes at small and high \(\alpha_{CO}\) values. Reducing temperature further amplifies this effect as the signal approaches the noise floor for \(n = 10\) , with counts in the 10 - 40 range at \(T = 280^{\circ} \mathrm{C}\) (Figure 2K,L). Consequently, it becomes increasingly difficult to extract the \(\mathrm{CO_2}\) signal and we are approaching the limit of + +<--- Page Split ---> + +detection from a conventional data analysis perspective, where simply the number of counts is extracted from the QMS signal. + +In principle, various strategies to address this issue exist, e.g., digital filtering [55], Fourier transforms [56, 57], wavelet transforms [58], etc., or statistical methods, like principal component analysis (PCA) [59], among others. When applied correctly, they enable the extraction of the desired signal from the accompanying noise and enhance the visibility of the underlying patterns in the data. However, these well- established denoising techniques have inherent limitations in the context of denoising QMS signals. Specifically, digital filtering can inadvertently remove crucial signals if they overlap with noise frequencies [60], Fourier and wavelet transforms may falter if the signal doesn't meet their assumptions about periodicity or localization [61] and PCA, which assumes the primary variance in data is due to the signal, might not be effective if noise has high variance or the signal is confined to a tight range [62]. Given these challenges, a Denoising Auto- Encoder (DAE), i.e., an artificial neural network, offers a promising solution since it can be trained to discern the structure of complex noisy data and reconstruct the underlying signal [63] (see Supplementary Section 3 and Supplementary Table 1). In practice, this is achieved by introducing examples of simulated signal with experimentally measured noise and training the network to reconstruct the (original) simulated signal. By also constraining the autoencoder's latent space to have a step function distribution, we introduce a prior, which enables denoising well below Signal- to- Noise Ratio (SNR) of 1 and ensures robustness in the final representations [64]. + +## A Constrained Denoising Auto-Encoder to improve the QMS Limit of Detection + +We applied the DAE to our QMS experiments by using experimentally measured QMS readout consisting of intrinsic measurement noise and noise induced by the nanoreactor setup (e.g. fluctuations in \(\mathrm{CO}_{2}\) concentration stemming from impurities in the used gases and/or tiny leaks), combined with an underlying true \(\mathrm{CO}_{2}\) signal stemming from the catalytic reaction on the Pd nanoparticle(s), as input (Figure 3). This combined signal is then compressed through encoder \(E_{\theta}\) to form a latent space inside the bottleneck, which is constrained through a consistency loss to form a step function distribution. This compressed representation is then upsampled in decoder \(D_{\theta}\) to form the reconstructed true signal, i.e., the QMS readout with a deconvolved underlying true \(\mathrm{CO}_{2}\) signal. In this design, the DAE can learn complex non- linear relationships between the noise and the true signal, enabling it to handle the kind of noise- signal interactions that might confound the other noise reducing methods mentioned above. For a comparison between the DAE and other denoising techniques applied to our QMS data, see Supplementary Section 4 and Supplementary Figures 6- 8. + +To evaluate the performance of the DAE, we compare the measured baseline- adjusted (BA) QMS \(\mathrm{CO}_{2}\) counts (see Methods for explanation of the BA- procedure) as function of \(\alpha_{CO}\) for \(n = 1000\) and \(n = 10\) , for temperatures between \(280^{\circ}\mathrm{C}\) and \(450^{\circ}\mathrm{C}\) , as well as the DAE- denoised \(\mathrm{CO}_{2}\) counts for \(n = 10\) (Figure 4). Focusing first on \(n = 1000\) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 The constrained denoising auto-encoder. The underlying signal that we want to access in our experiments, i.e., the \(\mathrm{CO_2}\) production rate step functions as a function of time, is corrupted by both correlated and uncorrelated noise in the experimentally measured readout from the QMS. Therefore, the experimental QMS readout is input to a denoising autoencoder, which transforms the input into a minimally dimensioned representation of the underlying signal and then reconstructs it sans corruption, which means, it reconstructs the original signal, rather then the QMS readout that also contains different types of noise. The DAE is trained with a standard L2 norm between reconstructed and underlying signal, and the latent space is constrained by an L1 norm and a consistency loss which forces the latent space to take on the form of the underlying signal.
+ +at \(450^{\circ}\mathrm{C}\) , the catalyst exhibits the highest reaction rate at \(\alpha_{CO}^{max} = 0.65\) and all \(\mathrm{CO_2}\) pulses are clearly discernible (Figure 4A). Executing identical analysis for \(n = 10\) at \(450^{\circ}\mathrm{C}\) reveals individual \(\mathrm{CO_2}\) pulses and \(\alpha_{CO}^{max}\approx 0.6\) (Figure 4B). Passing the \(n = 10\) QMS data obtained at \(450^{\circ}\mathrm{C}\) through the DAE reveals that it delivers more distinct and systematically in/decreasing \(\mathrm{CO_2}\) pulses, with a maximum rate at \(\alpha_{CO}^{max} = 0.65\) , in excellent agreement with \(n = 1000\) (Figure 4C). Repeating the same analysis for \(280^{\circ}\) C reveals the anticipated reduction of the overall rate for both samples and a distinct shift of \(\alpha_{CO}^{max}\) to \(\alpha_{CO}^{max}\approx 0.2\) , due to severe CO poisoning (Figure 4D- F). Importantly, for \(n = 10\) , it is now clear that the standard analysis is approaching the limit of detection where the \(\mathrm{CO_2}\) pulses are stochastic, whereas the DAE- denoised data still exhibits the same clear trend as seen for \(n = 1000\) with the standard analysis. + +As the next step, we extract the mean \(\mathrm{CO_2}\) counts for each \(\alpha_{CO}\) pulse along the \(\alpha_{CO}\) sweep for \(n = 1000\) , obtained for temperature steps of \(20^{\circ}\mathrm{C}\) from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) (Figure 4G). This reveals a constant and stoichiometric \(\alpha_{CO}^{max} = 0.65\) down to \(380^{\circ}\mathrm{C}\) , below which \(\alpha_{CO}^{max}\) systematically shifts to smaller values due to increasing CO poisoning, and reaches \(\alpha_{CO}^{max} = 0.25\) at \(280^{\circ}\mathrm{C}\) . Furthermore, it reveals a decrease in reaction rate from \(5240\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(450^{\circ}\mathrm{C}\) , to \(1700\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(280^{\circ}\mathrm{C}\) . Plotting the same data for \(n = 10\) obtained for \(10^{\circ}\mathrm{C}\) temperature steps, and comparing the standard analysis (Figure 4H) with the DAE- based analysis (Figure 4I) reveals qualitatively very similar behaviour. However, crucially, SNR is very poor for + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 Direct comparison of standard and DAE-enhanced QMS readout for 1000 and 10 Pd nanoparticles. (A) Baseline-adjusted (BA - see Methods for explanation) raw QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (blue line - obtained as the average measured BA-count across each pulse) for \(n = 1000\) Pd nanoparticles across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) and at \(450^{\circ}\mathrm{C}\) . (B) BA-QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (green line) for \(n = 10\) Pd nanoparticles across the entire \(\alpha_{CO}\) range and at \(450^{\circ}\mathrm{C}\) . (C) BA-QMS counts for \(\mathrm{CO_2}\) (grey line) together with the \(\mathrm{CO_2}\) signal denoised by the DAE (orange line) for \(n = 10\) across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) . (D-F) Same as (A-C) but at \(280^{\circ}\mathrm{C}\) . (G) BA-mean \(\mathrm{CO_2}\) counts for \(\alpha_{CO}\) sweeps at reactor temperatures ranging from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) in \(20^{\circ}\mathrm{C}\) steps for \(n = 1000\) . Red dots indicate the \(\alpha_{CO}\) value corresponding to the highest reaction rate, \(\alpha_{CO}^{max}\) . (H) Same as (G) but for \(n = 10\) , using \(10^{\circ}\mathrm{C}\) temperature steps. (I) Same as (H) but with the QMS signal denoised by the DAE. (J) \(\alpha_{CO}^{max}\) values extracted from (G) for all temperatures for \(n = 1000\) , plotted as function of \(\alpha_{CO}\) . We note the constant stoichiometric \(\alpha_{CO}^{max}\) value of 0.65 down to \(T = 360^{\circ}\mathrm{C}\) . Below this temperature, we find a systematic shift to lower \(\alpha_{CO}^{max}\) values, as a consequence of increasing CO poisoning. (K) Same as (J) but for \(n = 10\) . We note the qualitatively similar trend compared to \(n = 1000\) but also the significantly higher spread in the data points. (L) Same as (K) but based on the DAE-denoised QMS signal in (I). Clearly, the uncertainty in \(\alpha_{CO}^{max}\) is significantly reduced and the T-dependent trend of \(\alpha_{CO}^{max}\) for both \(n = 1000\) and \(n = 10\) is now very similar. The small discrepancies are discussed in the text.
+ +the standard analysis to the point where the underlying chemical dynamics become obscured by stochastic shifts in, e.g. \(\alpha_{CO}^{max}\) . This is different for the DAE- based data, which reproduces the T- dependent \(\alpha_{CO}^{max}\) trend found for the \(n = 1000\) sample very clearly (Figure 4J- L). The situation is similar when comparing the absolute BA - \(\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(450^{\circ}\mathrm{C}\) . For \(n = 1000\) , we count 5240 \(\mathrm{CO_2}\) molecules, whereas we count 90 and 65 for \(n = 10\) for the standard and DAE- based analysis, respectively. The DAE- based value of 65 is indeed proportional to a \(100\times\) decrease in catalytic surface area (compared to \(n = 1000\) ), within a relative error of roughly \(10\%\) . The results + +<--- Page Split ---> + +demonstrate that the DAE effectively reconstructs the catalysts' \(\mathrm{CO}_{2}\) production rate and, e.g., the temperature- dependent poisoning state of the system, even when dealing with considerably noisy data and the QMS operated very close to its limit of detection. + +## Resolving \(CO_{2}\) Reaction Product formed on a single Pd Nanoparticle using the DAE + +We put the DAE to the test by further reducing the amount of catalyst by a factor 10, down to a single Pd nanoparticle with a surface area of approx. \(7200 \mathrm{nm}^{2}\) . We used a chip with \(n = 1\) (cf. Figure 1E) and an empty chip with \(n = 0\) as the control (cf. Figure 1F). Clearly, even at the highest temperature of \(450^{\circ}\mathrm{C}\) , the \(\alpha_{CO}\) sweeps for \(n = 0\) and \(n = 1\) look very similar using the standard analysis (Figure 5A- B). They are characterized by small stochastic \(\mathrm{CO}_{2}\) pulses, indicating a certain level of background \(\mathrm{CO}_{2}\) in the reactant gas mixture, as well as the signal stemming from the \(\mathrm{CO}_{2}\) produced by the catalyst particle drowning in background noise. This is further corroborated by extracting the BA- \(\mathrm{CO}_{2}\) counts and plotting them vs. \(\alpha_{CO}\) for both \(n = 1\) (7 identical sweeps) and \(n = 0\) (5 identical sweeps), which reveals that no clear activity trend as function of \(\alpha_{CO}\) is resolved for \(n = 1\) , despite a generally slightly higher number of counts compared to \(n = 0\) (Figure 5C and Supplementary Figures 5 and 9). + +Applying the DAE first to the \(\alpha_{CO}\) sweep at \(450^{\circ}\mathrm{C}\) for \(n = 1\) reveals a different picture (Figure 5D). Rather than stochastic \(\mathrm{CO}_{2}\) pulses, a clear trend of in- and decreasing \(\mathrm{CO}_{2}\) counts along the \(\alpha_{CO}\) sweep is recovered, which again exhibits a maximum at the stoichiometric \(\alpha_{CO}^{max} = 0.65\) . This is in very good agreement with the corresponding experiments for \(n = 10\) and \(n = 1000\) . Applying the DAE to the \(\alpha_{CO}\) sweep at \(450^{\circ}\mathrm{C}\) for the \(n = 0\) control outputs a close to completely flat baseline at zero BA- counts, corroborating that the DAE is able to eliminate the noise contaminating the raw QMS signal (Figure 5E). + +Encouraged, we apply the DAE to \(\alpha_{CO}\) sweeps measured at temperatures \(330^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) for \(n = 1\) and plot the extracted \(\mathrm{CO}_{2}\) counts vs. \(\alpha_{CO}\) (Figure 5F). Remarkably, we clearly resolve the anticipated increase of the reaction rate to an \(\alpha_{CO}^{max} = 0.65\) for decreasing \(\alpha_{CO}\) , and the subsequent decrease of the reaction rate upon decreasing \(\alpha_{CO}\) . Furthermore, the extracted number of \(\mathrm{CO}_{2}\) counts \(\approx 6\) is approximately 10 times lower than for \(n = 10\) at \(450^{\circ}\mathrm{C}\) , which is in excellent agreement with a reduction of the active surface area by a factor of 10 from \(n = 10\) to \(n = 1\) . Finally, upon reduction of temperature, we see an indication of a shift of \(\alpha_{CO}^{max}\) to lower values at \(410^{\circ}\mathrm{C}\) , before the DAE- signal also drops below the limit of detection of a single count. Similar analysis of the \(n = 0\) control at \(450^{\circ}\mathrm{C}\) consistently delivers \(\mathrm{CO}_{2}\) counts very close to or significantly below the limit of detection of a single count and thus confirms the ability of the DAE to resolve \(\mathrm{CO}_{2}\) produced by the single Pd nanoparticle (Figure 5F). + +We put these results into perspective by comparing them to the data obtained for \(n = 10\) and \(n = 1000\) . Specifically, we plot the \(\alpha_{CO}^{max}\) values for \(n = 1000\) , \(n = 10\) DAE- denoised and \(n = 1\) DAE- denoised for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) , i.e., the range for which the \(n = 1\) DAE- signal is above the limit of detection (Figure 5G). These values are + +<--- Page Split ---> + +very similar for all three samples and very close to the stoichiometric value of \(\alpha_{CO}^{max} = 0.66\) . This is important because it confirms that the catalyst experiences identical reaction conditions in all three implementations in terms of number of particles. We also compare the \(\mathrm{CO_2}\) counts obtained at \(\alpha_{CO}^{max}\) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) normalized by the number of particles on the respective sample, i.e., \(n = 1000\) , \(n = 10\) and \(n = 1\) (Figure 5H). This reveals particle- number- normalized \(\mathrm{CO_2}\) count values between 4 and 6 and thus corroborates direct scaling of extracted QMS counts with catalyst surface area, within the expected level of uncertainty that is imposed, e.g., by slightly different particle dimensions on each of the three chips. + +## Conclusions + +We have demonstrated how the combination of nanofluidic reactors and DAE- based deep learning enables the reduction of catalyst surface area required for online QMS analysis of reaction product in the gas phase by \(\approx 3\) orders from the current SotA, down to the level of single nanoparticles. This breakthrough was enabled by the ability of nanofluidic reactors to focus reaction product from tiny catalyst surfaces towards analysis, and the unrivalled capacity of autoencoders to discern tiny signals from noise. We have illustrated this on the example of the CO oxidation reaction on Pd nanoparticle model catalysts \(A \approx 0.0072 \pm 0.00086 \mu \mathrm{m}^2\) ( \(7200 \pm 860 \mathrm{~nm}^2\) ) per particle, localized inside well- mixed nanofluidic catalyst beds. We analyzed nanoreactors with \(n = 1000\) , 10, 1 and 0 Pd particles across a temperature range from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) and found the characteristic dependence of the reaction rate on \(\alpha_{CO} = P_{CO} / (P_{O_2} + P_{CO})\) , as well as a stoichiometric highest reaction rate at \(\alpha_{CO}^{max} = 0.66\) for the highest temperatures, with a distinct shift to lower \(\alpha_{CO}\) at lower temperatures, due to CO poisoning. While standard data analysis enabled meaningful results for \(n = 1000\) and \(n = 10\) (at the highest temperatures) only, DAE denoising produced reliable QMS data even for \(n = 1\) , and facilitated the analysis of the reaction dynamics in terms of \(\alpha_{CO}\) - dependent surface poisoning also for a single Pd nanoparticle. + +In a wider perspective, we highlight that our approach is not specific to CO oxidation, and can be adapted for other catalytic reactions and in principle any measurement problem, in which the structure of the underlying true signal is known. Our results thus constitute a new paradigm to significantly improve the resolution in online reaction analysis in (single particle) catalysis and advocate deep learning to extract tiny QMS signals from noise. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 Online mass spectrometry from a single Pd nanoparticle. (A) Baseline-adjusted (BA) raw QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (green line - obtained as average measured BA-count across each pulse) for \(n = 1\) Pd nanoparticle, measured across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) . (B) Same as (A) but for an empty nanochannel, i.e., \(n = 0\) . Note the very similar and stochastic appearance of apparent \(\mathrm{CO_2}\) pulses both for \(n = 1\) and \(n = 0\) when analyzed in the standard way. (C) Mean BA-CO2 counts extracted for each \(\alpha_{CO}\) pulse based on the standard analysis across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) , for both \(n = 0\) (black lines) and \(n = 1\) (green lines). For \(n = 1\) we executed 7 consecutive \(\alpha_{CO}\) sweeps and for \(n = 0\) we executed 5 consecutive \(\alpha_{CO}\) sweeps. As key observation, we note high reproducibility of the \(\alpha_{CO}\) sweeps for both \(n = 0\) and \(n = 1\) , as well as that no clear activity trend as function of \(\alpha_{CO}\) is resolved for the \(n = 1\) sample, despite a generally slightly higher number of counts compared the \(n = 0\) reference. (D) Same as (A) but for DAE-deniosed BA-QMS \(\mathrm{CO_2}\) counts (purple line). Note the significantly smaller but at the same time non-stochastic \(\mathrm{CO_2}\) pulses resolved by the DAE compared to the standard analysis. (E) Same as (D) but for an empty nanochannel, i.e., \(n = 0\) . Note that using the DAE, a flat baseline at zero BA-counts is obtained for the empty nanochannel. (F) Same as (C) but for DAE-deniosed data, where distinct reaction rate maxima are resolved for \(n = 1\) (purple curves), with \(\alpha_{CO}^{max}\) ranging between 0.65 and 0.6 for temperatures between \(450^{\circ}\mathrm{C}\) and \(410^{\circ}\mathrm{C}\) (inset), in good agreement with the earlier results for \(n = 10\) and \(n = 1000\) . For lower temperatures, as well as for \(n = 0\) (black lines representing multiple sweeps at \(450^{\circ}\mathrm{C}\) ), the DAE outputs counts \(< 1\) , which is physically unreasonable and thus defined as the limit of detection. (G) \(\alpha_{CO}^{max}\) values for \(n = 1000\) (blue), \(n = 10\) DAE-deniosed (orange) and \(n = 1\) DAE-deniosed (purple) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) , i.e., the range for which the \(n = 1\) DAE-signal is above the limit of detection. (H) \(\mathrm{CO_2}\) counts normalized by the number of particles on the respective sample, obtained at \(\alpha_{CO}^{max}\) for \(n = 1000\) (blue), \(n = 10\) DAE-deniosed (orange) and \(n = 1\) DAE-deniosed (purple) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) .
+ +<--- Page Split ---> + +## Methods + +## Deep Learning Architecture + +We employed a constrained denoising autoencoder to denoise the experimental QMS data. This architecture was designed to accept an input shape of 6400 timesteps, equalling the full size of a single \(\alpha_{CO}\) sweep. The encoder segment consists of seven convolutional layers, each with 32 neurons and a kernel size of 9. These layers are systematically interspersed with max- pooling operations to achieve optimal data compression and utilized the Leaky Rectified Linear Unit (LeakyReLU) as its activation function due to its proven efficacy in handling complex non- linearities whilst avoiding mode collapse, vanishing and exploding gradients. Following this sequence, the encoded data is transformed into its latent space through a dense layer, reducing its dimensionality to equal the number of steps of the \(\alpha_{CO}\) sweep. This compressed representation is then reshaped and combined with the previously encoded data to retain both quantitative and structural information of the underlying signal to the decoder. + +In the symmetrical decoder segment, the model reconstructs the data through another series of seven convolutional layers, each followed by an upsampling operation, ensuring accurate data restoration. The final layer then produces a reconstruction of the initial data. + +To enhance the model's accuracy and ensure it only outputs functions representative of the underlying signal, we incorporated a consistency loss function which compares mean values across specific data segments of the reconstructed input with their corresponding latent space representations. This function ensures that the autoencoder's reconstructed output aligns structurally with the original data. The network is trained with an MSE function between target and predicted data segments, as elaborated upon in Supplementary Table 1 and corresponding discussion. Overall, this architecture effectively balances noise reduction, feature retention, and data consistency. + +The architecture was implemented through TensorFlow [65] and the code can be found in the corresponding GitHub page. + +## Synthetic Data Generation + +The system utilizes two generator functions: a generator for the underlying signal and a generator for the noise. The noise is generated first, and is defined by its standard deviation and shape of the gas pulses. It is generated in two parts; firstly from typical stochastic fluctuations in the QMS readout itself such as thermal noise, environmental fluctuations, residual particles (flicker noise), shot noise and ion feedback noise. We model these by a gaussian distribution with a high range of variability, to encapsulate the (white noise) effects of all these sources in the training set. Secondly, there is contamination noise, resultant of reagent reactions which occur outside the catalytically active surface area of the nanoparticles. We model these by an added term for each gas pulse, proportional to the concentration of input reagents to the system, also with a high variability. + +<--- Page Split ---> + +The signal generator creates a step function with a desired signal- to- noise ratio \(SNR = \frac{\mu_{signal}}{\sigma_{noise}}\) , which is chosen during training to match the SNR of 1- 10 Pd NP reactor readouts based on the given noise. The steps are always placed the same (i.e. in an on- off pulsing form), but each with a random signal value. Thus, there is no inherent inductive bias regarding the relation of one pulse to another within the same overall step function. + +## Deep Learning Training + +The training consists of two steps. Firstly, a curriculum learning scheme is employed on artificially generated signals on gaussian- distributed noise with a pre- defined SNR. The SNR is drawn from a uniform distribution in the range \(SNR \in (0.95, 1)\) , quantitatively corresponding roughly to the highest amount of experimentally measured catalytic activity of the \(n = 10\) Pd sample at \(T = 450^{\circ}\mathrm{C}\) . The lower end of this range decreases by a factor 0.05 to a minimum of 0 each time the network is trained to convergence, as defined by an early stopping mechanism with a patience of 32 epochs. This scheme helps prevent problems of vanishing gradient and thus mode collapse resultant of directly training on extremely- low SNR examples. The artificial data generation allows the model to train on a broad range of noises and signal distributions, increasing its robustness by forcing it to learn generalizable functions for denoising. The loss curve for this training can be found in Supplementary Figure 13. + +The second step consists of training on artificial signal distributions in the range \(SNR \in (0, 1)\) , corresponding to the full range of SNRs considered in this work, generated on examples of experimentally measured noise. This fine- tuning step ensures that the final functions learnt by the network work appropriately for the particular noise distributions relevant to the experimental conditions at inference time. The noise is here defined as the output of a CO oxidation sequence of a \(n = 0\) Pd sample chip at \(T = 40^{\circ}\mathrm{C}\) . + +## Preprocessing + +QMS data was preprocessed using a custom Python script (available online, see Supplementary Materials) developed for this study. To adjust for baseline shifts of the QMS signal over time, which can occur if the pressure inside the UHV chamber on which the QMS is mounted has not yet entirely stabilized, the following procedure is applied. First, a separate linear fit is made to the QMS signal for each pulse in pure Ar. Secondly, this fit is subtracted from said QMS signal and from the QMS signal of the next corresponding \(\alpha_{CO}\) pulse in time. This shifts the baseline to a consistent level for each individual pulse within a sweep. This procedure is referred to baseline adjustment (BA) in the main text. Outliers, defined as individual QMS data points deviating more than three standard deviations from the mean, were removed. The mean QMS counts value for each period was thereafter calculated and thus constituted the standard analysis for this study. Finally, the data are scaled down by a factor 100 to bring it in range with the order of magnitude of a standard gaussian, which is how the neural network's weights are initialized and therefore stabilizes training. + +<--- Page Split ---> + +## Experimental Setup + +Experimental SetupExperiments were performed on a setup reported in detail earlier [40, 41] (see Supplementary Figure 1). In brief, it is comprised of an elevated/atmospheric pressure gas mixing/handling unit connected to the inlet side of a nanofluidic chip holder that, on the outlet side, is connected to an ultrahigh vacuum system (base pressure of \(10^{- 10}\) mbar) equipped with a triple filter QMS equipped with a pulsed ion counting detector (Hiden HAL/3F PIC) and a gold- plated ion source. The QMS was operated in SEM mode and at 12 s sampling rate. The gas mixing/handling unit was built from quarter- inch stainless steel tubing with Vacuum Coupling Radiation (VCR) (Swagelok) fittings as connectors. The nanofluidic chip was hosted in a stainless steel connection block with welded VCR fittings, which provided gas and QMS connection. The connection block featured interior gas lines connected to exterior flexible stainless steel tubing. The gas inlet and outlet of the chip were sealed with perfluoroelastomer (FPM) O- rings to the connection block. To avoid unwanted background signals, a constant flow of Ar (18 ml/min) was flushed around the O- rings, limiting any diffusion to Ar. Connectors on the connector block allowed for integration with four Bronkhorst Low \(\Delta \mathrm{P}\) mass flow controllers and a pressure controller (max pressure of 10 bar). This system facilitate the creation of a gas mixture with up to four different gases and a defined inlet pressure up to 10 bar (here we used 4 bar). All gas flows and pressures were controlled and monitored through a customized LabVIEW program. The system can be heated up to \(450^{\circ}\mathrm{C}\) K using a microfabricated resistive heater on the backside of the chip. Temperature readout was performed through a four- wire resistance temperature probe microfabricated on the backside of the chip at the position of the nanofluidic system. The chip was electrically connected for heating and temperature readout through six gold- plated electronic spring spins embedded in a machined ceramic block. The pins are connected to a temperature controller (Lakeshore 335) operated via a LabVIEW program. The chips are calibrated by placing one in an oil bath with a thermocouple readout to monitor the temperature before any measurements are performed, see Supplementary Figure 14. The chip holder is equipped with a water- cooled copper block to maintain a constant temperature, helping to reduce mechanical movements due to heat transfer into the holder assembly. + +## Sample Mounting and Pre-treatment + +Sample Mounting and Pre- treatmentPrior to mounting of a nanofluidic chip, the pipe connecting the outlet of the nanofluidic chip to the inlet of the QMS vacuum chamber are heated to \(353\mathrm{K}\) to minimize water adsorption. Once the chip was mounted, the system was pumped for up to 72 hours, until the base pressure of the QMS chamber reached the desired of \(10^{- 10}\) mbar. After mounting and pumping/baking, the sample was heated to \(280^{\circ}\mathrm{C}\) and subsequently exposed to 20 cycles of alternating pulses of \(10\%\) CO in Ar carrier gas and \(15\%\) \(\mathrm{O_2}\) in Ar carrier gas, each 15 min long, followed by a complete \(\alpha_{CO}\) sweep, in order to activate the catalyst and reach a stable QMS signal baseline. + +<--- Page Split ---> + +## CO Oxidation Experiments + +For the CO oxidation \(\alpha_{CO}\) sweep experiments, CO (10% in Ar) and \(\mathrm{O_2}\) (15% in Ar) were used with Ar carrier gas (99.99999% purity). The inlet pressure was set to 4 bar, and a total flow of \(10\mathrm{ml / min}\) through the microchannels was applied. The experiment sequence consists of 15 min pulses of \(\mathrm{CO / O_2}\) mixtures at a constant 6% percent reactant concentration, separated by 15 min in pure Ar (Figure 2A- C) at constant temperatures ranging from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) . In each pulse, the concentration of \(\mathrm{CO / O_2}\) is varied such that \(\alpha_{CO}\) is systematically varied from 1 to 0 in steps of 0.05 for each subsequent pulse. The only exception is the first (and last) two pulses, which vary in steps of 0.11 (0.02) and 0.02 (0.11), respectively. This is due to instrumental limitations of the MFC controllers at very low values of absolute flow, where they are unable to control the flow accurately. Each experiment is also started with a single 30 min pulse of \(4\%\) \(\mathrm{O_2}\) and \(2\%\) CO, followed by a 15 min Ar pulse, to reset the state of the catalyst before each sequence. + +## Nanofabrication of the Nanoreactor Chip + +Fabrication of the nanofluidic systems was carried out in cleanroom facilities of Fed. Std.209 E Class 10 - 100, using electron- beam lithography (JBX- 9300FS / JEOL Ltd), direct- laser lithography (Heidelberg Instruments DWL 2000), photolithography (MA 6 / Suss MicroTec), reactive- ion etching (Plasmalab 100 ICP180 / Oxford Plasma Technology and STS ICP), electron- beam evaporation (PVD 225 / Lesker), magnetron sputtering (MS150 / FHR), deep reactive- ion etching (STS ICP / STS) and wet oxidation (wet oxidation / Centrotherm), fusion bonding (AWF 12/65 / Lenton), and dicing (DAD3350 / Disco). In particular, the fabrication steps comprised the following processing steps of a \(4^{\circ}\) - silicon (p- type) wafer: + +Fabrication of alignment marks: Spin coating HMDS adhesion promoter (MicroChem) at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 120 sec. Spin coating UV5 (MicroChem) at \(2000\mathrm{rpm}\) for 60 sec and soft baking at \(130^{\circ}\mathrm{C}\) for 120 sec. Electron- beam exposure of alignment marks for both optical and electron- beam lithography at \(10\mathrm{nA}\) with a shot pitch of \(20\mathrm{nm}\) and \(34\mu \mathrm{C / cm^2}\) exposure dose. Postexposure bake at \(130^{\circ}\mathrm{C}\) for 90 sec. Development in MF- 24A (Microposit) for 90 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Reactive- ion etching (RIE) for 15 sec at \(60\mathrm{mTorr}\) chamber pressure, \(60\mathrm{W}\) RF- power, \(60\mathrm{scm}\) \(\mathrm{O_2}\) flow (descum). RIE for 15 min at \(40\mathrm{mTorr}\) chamber pressure, \(50\mathrm{W}\) RF- power, \(100\mathrm{W}\) ICP- power, \(50\mathrm{scm}\) \(\mathrm{Cl}_2\) flow (600 nm etch depth in silicon). + +Thermal oxidation: Cleaning in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. Wet oxidation in water atmosphere for 45 min at \(950^{\circ}\mathrm{C}\) (200 nm thermal oxide). + +Fabrication of nanochannels: Spin coating HMDS adhesion promoter (MicroChem) at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 120 sec. Spin coating UV5 (MicroChem) at \(2000\mathrm{rpm}\) for 60 sec and soft baking at \(130^{\circ}\mathrm{C}\) for 120 sec. Electron- beam exposure of nanochannels at \(1\mathrm{nA}\) with a shot pitch of \(10\mathrm{nm}\) and \(34\mu \mathrm{C / cm^2}\) exposure dose. Post- exposure bake at \(130^{\circ}\mathrm{C}\) for 90 sec. Development in + +<--- Page Split ---> + +MF- 24A (Microposit) for 90 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. RIE for 10 sec at \(40\mathrm{mTorr}\) chamber pressure, \(40\mathrm{W}\) RF- power, \(40\mathrm{scm}\) \(\mathrm{O}_2\) flow (descum). RIE for 100 sec at \(8\mathrm{mTorr}\) chamber pressure, \(50\mathrm{W}\) RF- power, \(50\mathrm{scm}\) \(\mathrm{NF}_3\) flow (70 nm etch depth in thermal oxide). Cleaning in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. + +Fabrication of microchannels: Spin coating HMDS at 3000 rpm for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Spin coating S1813 (Shipley) at 3000 rpm for 30 sec and soft baking at \(115^{\circ}\mathrm{C}\) for 2 min. Expose microchannels for 10 sec in contact aligner at \(6\mathrm{mW / cm^2}\) intensity. Development in MF- 319 (Microposit) for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Buffered oxide- etch for 3 min to remove thermal oxide, rinsing in water and drying under \(\mathrm{N}_2\) stream. Deep reactive- ion etching for 100 cycles of 7 sec at \(6\mathrm{mTorr}\) chamber pressure, \(800\mathrm{W}\) RF- power, \(8\mathrm{W}\) platen power, \(130\mathrm{scm}\) \(\mathrm{SF}_6\) flow (Si- etch), and of 5 sec at \(6\mathrm{mTorr}\) chamber pressure, \(800\mathrm{W}\) RF- power, \(8\mathrm{W}\) platen power, \(85\mathrm{scm}\) \(\mathrm{C}_4\mathrm{F}_8\) flow (passivation) at a rate of \(600\mathrm{nm / cycle}\) . Removal of resist in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. The resulting channels have a depth of \(60\mu \mathrm{m}\) measured using a Dektak 150 surface profiler. + +Fabrication of inlets (from backside): Magnetron- sputtering of \(200\mathrm{nm}\) Al (hard mask). Spin coating S1813 at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Expose inlets for 10 sec in contact aligner at \(6\mathrm{mW / cm^2}\) intensity. Development in MF- 319 for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Aluminum wet etch \(\mathrm{(H_3PO_4\cdot CH_3COOH\cdot HNO_3\cdot H_2O}\) (4:4:1:1)) for 10 min to clear the hard mask at inlet positions. Deep reactive- ion etching for 300 cycles of 12 sec at \(5\mathrm{mTorr}\) chamber pressure, \(600\mathrm{W}\) RF- power, \(10\mathrm{W}\) platen power, \(130\mathrm{scm}\) \(\mathrm{SF}_6\) flow (Si- etch), and of 7 sec at \(5\mathrm{mTorr}\) chamber pressure, \(600\mathrm{W}\) RF- power, \(10\mathrm{W}\) platen power, \(85\mathrm{scm}\) \(\mathrm{C}_4\mathrm{F}_8\) flow (passivation) at a rate of \(2\mu \mathrm{m / cycle}\) . Removal of Al- hard mask in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. + +Fabrication of heater elements on the backside: Spin coating HMDS at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Spin coating LOR3A (MicroChemicals) at \(3000\mathrm{rpm}\) for 30 sec and soft baking at \(180^{\circ}\mathrm{C}\) for 5 min. Spin coating S1813 (Shipley) at \(3000\mathrm{rpm}\) for 30 sec and soft baking at \(115^{\circ}\mathrm{C}\) for 2 min. Expose heater elements with direct- laser lithography at \(10\mathrm{mW / cm^2}\) intensity. Development in MF- 319 (Microposit) for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Electron- beam evaporation of \(10\mathrm{nm}\mathrm{Cr} / 100\mathrm{nm}\mathrm{Pt}\) . Lift- off in remover Rem1165 (MicroChemicals), rinsing in isopropanol, and drying under \(\mathrm{N}_2\) stream. + +Fabrication of nanoparticles inside nanochannels: Spin coating Copolymer MMA(8.5)MMA (MicroChem Corporation, \(10\mathrm{wt}\%\) diluted in anisole) at \(6000\mathrm{rpm}\) for 60 sec and soft baking on a hotplate at \(180^{\circ}\mathrm{C}\) for 5 min. Spin coating ZEP520A: anisole (1:2) at \(3000\mathrm{rpm}\) for 60 sec and soft baking at \(180^{\circ}\mathrm{C}\) for 5 min. Electronbeam exposure at \(1\mathrm{nA}\) with a shot pitch of \(2\mathrm{nm}\) and \(280\mu \mathrm{C / cm^2}\) exposure dose. Development in n- amyl acetate for 60 sec, rinsing in isopropanol and drying under \(\mathrm{N}_2\) stream. Development in methyl isobutyl ketone:isopropanol (1:1) for 60 sec, rinsing in + +<--- Page Split ---> + +isopropanol and drying under \(\mathrm{N}_2\) stream. Electron- beam evaporation of \(\mathrm{Au / SiO_2 / Pd}\) triple layer. Lift- off in acetone, rinsing in isopropanol, and drying under \(\mathrm{N}_2\) stream. + +Fusion bonding: (a) Cleaning of the substrate together with a lid (175 \(\mu \mathrm{m}\) thick \(4^{,}\) pyrex, UniversityWafers) in \(\mathrm{H_2O:H_2O_2:NH_3OH}\) (5:1:1) for \(10\mathrm{min}\) at \(80^{\circ}\mathrm{C}\) . (b) Prebonding the lid to the substrate by bringing surfaces together and manually applying pressure. (c) Fusion bonding of the lid to the substrate for \(5\mathrm{h}\) in \(\mathrm{N}_2\) atmosphere at \(550^{\circ}\mathrm{C}\) ( \(^{\circ}\mathrm{C / min}\) ramp rate). + +Dicing of bonded wafers: Cutting nanofluidic chips from the bonded wafer using a resin bonded diamond blade of \(250\mu \mathrm{m}\) thickness (Dicing Blade Technology) at 35 krpm and \(1\mathrm{mm / s}\) feed rate. + +## Acknowledgments + +This research has received funding from the European Research Council (ERC) under the European Union's Horizon Europe research and innovation program (101043480/NACAREI) and under the European Union's Horizon 2020 research and innovation programme (678941/SINCAT), and from the Knut and Alice Wallenberg Foundation projects 2016.0210 and 2015.0055. Part of this work was carried out at the Chalmers MC2 cleanroom facility and at the Chalmers Materials Analysis Laboratory (CMAL). + +## References + +[1] Abbott, B. et al. Observation of gravitational waves from a binary black hole merger. Physical Review Letters 116, 061102 (2016). + +[2] Driggers, J. et al. Increasing the astrophysical reach of the advanced virgo detector via the application of squeezed vacuum states of light. Physical Review Letters 122, 231102 (2019). + +[3] Littenberg, T. B. & Cornish, N. J. Bayesian inference for spectral estimation of gravitational wave detector noise. Physical Review D 91, 084034 (2015). + +[4] Chen, S. et al. Common- red- signal analysis with 24- yr high- precision timing of the european pulsar timing array: Inferences in the stochastic gravitational- wave background search. Monthly Notices of the Royal Astronomical Society 508, 4970- 4993 (2021). + +[5] Lai, Z. & Leng, Y. Generalized parameter- adjusted stochastic resonance of duffing oscillator and its application to weak- signal detection. Sensors 15, 21327- 21345 (2015). + +[6] Lopez, C., Zhong, W., Lu, S., Cong, F. & Cortese, I. Stochastic resonance in an underdamped system with fitzhug- nagumo potential for weak signal detection. Journal of Sound and Vibration 408, 16- 28 (2017). + +<--- Page Split ---> + +[7] Jin, Y. & Moura, J. M. F. Time- reversal detection using antenna arrays. IEEE Transactions on Signal Processing 57, 1396- 1414 (2009). + +[8] Nelson, D. Special purpose correlation functions for improved signal detection and parameter estimation, Vol. 3, 237- 240 (IEEE, 1993). + +[9] Werley, C., Teo, S. M. & Nelson, K. Pulsed laser noise analysis and pump- probe signal detection with a data acquisition card. Review of Scientific Instruments 82, 123108 (2011). + +[10] Algburi, R. N. A. & Gao, H. Health assessment and fault detection system for an industrial robot using the rotary encoder signal. Energies 12, 2816 (2019). + +[11] Griffiths, W. J. & Wang, Y. Mass spectrometry: from proteomics to metabolomics and lipidomics. Chemical Society Reviews 37, 1432- 1448 (2008). + +[12] Grayson, S. M. & Frechet, J. M. Applications of mass spectrometry to polymers. Mass spectrometry reviews 21, 223- 248 (2002). + +[13] Mrksich, M. & Whitesides, G. M. Mass spectrometry of self- assembled monolayers: a new tool for molecular surface science. Science 269, 664- 665 (1995). + +[14] Shappell, N. W. Mass spectrometry in the analysis of perfluorinated pollutants. Environmental science & technology 42, 4869- 4874 (2008). + +[15] Brennwald, M. S., Schmidt, M., Oser, J. & Kipfer, R. A portable and autonomous mass spectrometric system for on- site environmental gas analysis. Environmental science & technology 50, 13455- 13463 (2016). + +[16] Tsugawa, H., Kanazawa, M., Ogiwara, A. & Arita, M. Understanding and practical use of in- source fragmentation in liquid chromatography- mass spectrometry. Journal of mass spectrometry 48, 730- 740 (2013). + +[17] Martin, L. B. et al. Mass spectrometry imaging for plant biology: a review. Phytochemistry reviews 13, 445- 488 (2014). + +[18] Buchberger, A., DeLaney, K., Johnson, J. & Li, L. Mass spectrometry imaging: A review of emerging advancements and future insights. Analytical Chemistry 90 (2017). + +[19] Kersten, R. D. et al. Mass spectrometry imaging of complex microbial communities. Accounts of chemical research 45, 455- 464 (2012). + +[20] Bush, M. F. et al. Mass spectrometry imaging of biological tissue: an approach for multicenter studies. Analytical chemistry 89, 2558- 2563 (2017). + +[21] Kasemo, B. Quartz tube orifice leaks for local, fast- response gas sampling to mass spectrometers. Review of Scientific Instruments 50, 1602- 1604 (1979). + +<--- Page Split ---> + +[22] Veisi, P., Dong, X., Dong, J., Fenselau, C. & Zhang, H. Mass spectrometry imaging for visualizing in situ metabolism of endogenous metabolites and dietary phytochemicals. \*Metabolites\* 4, 131 (2014). + +[23] Ligterink, W., Vincent, S. R., Masood, M. M., Choi, Y. H. & Verpoorte, R. Mass spectrometry imaging and gc- ms profiling of the mammalian peripheral sensory- motor circuit. \*Journal of visualized experiments: JoVE\* (2014). + +[24] Dahl, J. M., Maiolica, A., Gstaiger, M. & Aebersold, R. Mass spectrometry for the study of protein- protein interactions. \*Methods\* 24, 230- 239 (2001). + +[25] Abushareeda, W., Lyris, E., Kraiem, S., Thomas, A. & Thevis, M. Mass spectrometry in doping control: Structure characterization and analytical assays. \*Mass spectrometry reviews\* 36, 196- 217 (2017). + +[26] Li, Y., Lei, Y. & Huang, W. Size- selected clusters: preparation, structure, and catalytic properties. \*Advances in Physics: X\* 3, 1437698 (2018). + +[27] Qiao, B. et al. Single- atom catalysis of CO oxidation using Pt1/FeOx. \*Nature chemistry\* 3, 634- 641 (2011). + +[28] Zhang, X. et al. Single- atom catalysts: synthetic strategies and electrochemical applications. \*Joule\* 2, 1242- 1264 (2018). + +[29] Liu, L. & Corma, A. Single- atom catalysts: a new frontier in heterogeneous catalysis. \*Accounts of chemical research\* 51, 789- 799 (2018). + +[30] Sambur, J. B. et al. Sub- particle reaction and photocurrent mapping to optimize catalyst- modified photoanodes. \*Nature\* 530, 77- 80 (2016). + +[31] Wei, H. & Xu, H. Shell- isolated nanoparticle- enhanced raman spectroscopy: principle and applications. \*Chemical Society Reviews\* 40, 2494- 2507 (2011). + +[32] Xu, W., Zhang, Y. & Chen, T. \*Single Particle Nanocatalysis: Fundamentals and Applications\* (2019). + +[33] Hartman, T., Geitenbeek, R. G., Wondergem, C. S., van der Stam, W. & Weckhuysen, B. M. Operando nanoscale sensors in catalysis: All eyes on catalyst particles. \*ACS Nano\* 14, 3725- 3735 (2020). + +[34] Hartman, T., Wondergem, C. S., Kumar, N., van den Berg, A. & Weckhuysen, B. M. Surface- and tip- enhanced raman spectroscopy in catalysis. \*The Journal of Physical Chemistry Letters\* 7, 1570- 1584 (2016). + +[35] Alekseeva, S., Nedrygailov, I. I. & Langhammer, C. Single particle plasmonics for materials science and single particle catalysis. \*ACS Photonics\* 6, 1319- 1330 (2019). + +<--- Page Split ---> + +[36] Henriksen, G., Hoff, P., Larsen, R. & Alstad, J. Microreactor and method for preparing a radiolabeled complex or a biomolecule conjugate. US Patent (2009). + +[37] Schouten, J. C., Nijhuis, T. A., van den Oetelaar, L. C. & van den Bleek, C. M. Microreactors for kinetic studies of heterogeneous catalytic reactions. Applied Catalysis A: General 221, 253- 264 (2001). + +[38] Vesborg, P. C., Olsen, J. L., Henriksen, T. R., Chorkendorff, I. & Hansen, O. Gas- phase photocatalysis in micro- reactors. Chemical Engineering Journal 160, 738- 741 (2010). + +[39] Schouten, K. J. P., Pérez- Gallent, E. & Koper, M. T. Nanoreactors: engineering catalyst particles with enzyme- like behavior. Chemical Science 8, 5159- 5166 (2017). + +[40] Albinsson, D. et al. Copper catalysis at operando conditions—bridging the gap between single nanoparticle probing and catalyst- bed- averaging. Nature Communications 11, 4832 (2020). + +[41] Albinsson, D. et al. Operando detection of single nanoparticle activity dynamics inside a model pore catalyst material. Science Advances 6, eaba7678 (2020). + +[42] Boiko, D. A., Kozlov, K. S., Burykina, J. V., Ilyushenkova, V. V. & Ananikov, V. P. Fully automated unconstrained analysis of high- resolution mass spectrometry data with machine learning. Journal of the American Chemical Society 144, 14590- 14606 (2022). URL https://doi.org/10.1021/jacs.2c03631. + +[43] Abdelmoula, W. M. et al. Peak learning of mass spectrometry imaging data using artificial neural networks. Nature Communications 12, 5544 (2021). URL https://doi.org/10.1038/s41467- 021- 25744- 8. + +[44] Liebal, U. W., Phan, A. N. T., Sudhakar, M., Raman, K. & Blank, L. M. Machine learning applications for mass Spectrometry- Based metabolomics. Metabolites 10 (2020). + +[45] Ji, H., Deng, H., Lu, H. & Zhang, Z. Predicting a molecular fingerprint from an electron ionization mass spectrum with deep neural networks. Analytical Chemistry 92, 8649- 8653 (2020). URL https://doi.org/10.1021/acs.analchem.0c01450. + +[46] Jetybayeva, A. et al. A review on recent machine learning applications for imaging mass spectrometry studies. Journal of Applied Physics 133, 020702 (2023). URL https://doi.org/10.1063/5.0100948. + +[47] Zhou, Z. & Zare, R. N. Personal information from latent fingerprints using desorption electrospray ionization mass spectrometry and machine learning. Analytical Chemistry 89, 1369- 1372 (2017). URL https://doi.org/10.1021/acs.analchem. + +<--- Page Split ---> + +[48] Gessulat, S. et al. Prosit: proteome- wide prediction of peptide tandem mass spectra by deep learning. Nature Methods 16, 509- 518 (2019). URL https://doi.org/10.1038/s41592- 019- 0426- 7. [49] Tran, N. H. et al. Deep learning enables de novo peptide sequencing from data- independent- acquisition mass spectrometry. Nature Methods 16, 63- 66 (2019). URL https://doi.org/10.1038/s41592- 018- 0260- 3. [50] Zhou, Y., Wang, Z. & Liu, C. Perspective on CO oxidation over Pd- based catalysts. Catal. Sci. Technol. 5, 69- 81 (2015).[51] van Spronsen, M. A., Frenken, J. W. M. & Groot, I. M. N. Surface science under reaction conditions: CO oxidation on Pt and Pd model catalysts. Chem. Soc. Rev. 46, 4347- 4374 (2017).[52] Albinsson, D. et al. Shedding light on CO oxidation surface chemistry on single Pt catalyst nanoparticles inside a nanofluidic model pore. ACS Catalysis 11, 2021- 2033 (2021).[53] Langhammer, C., Kasemo, B. & Zoric, I. Absorption and scattering of light by Pt, Pd, Ag, and Au nanodisks: Absolute cross sections and branching ratios. The Journal of Chemical Physics 126, 194702 (2007).[54] Zhou, Y., Wang, Z. & Liu, C. Perspective on CO oxidation over Pd- based catalysts. Catal. Sci. Technol. 5, 69- 81 (2015).[55] Iwai, T. et al. A transient signal acquisition and processing method for microdroplet injection system inductively coupled plasma mass spectrometry (m- disc- ip- ms). J. Anal. At. Spectrom. 30, 1617- 1622 (2015).[56] Didriche, K., Lauzin, C., Foldes, T., de Ghellinck D'Elseghem Vaerenwijck, X. & Herman, M. The fantasio+ set- up to investigate jet- cooled molecules: focus on overtone bands of the acetylene dimer. Molecular Physics 108, 2155- 2163 (2010).[57] Fornelli, L. et al. Advancing top- down analysis of the human proteome using a benchtop quadrupole- orbitrap mass spectrometer. Journal of Proteome Research 16, 609- 618 (2017).[58] Chou, S.- W., Shiu, G.- R., Chang, H.- C. & Peng, W.- P. Wavelet- based method for time- domain noise analysis and reduction in a frequency- scan ion trap mass spectrometer. Journal of the American Society for Mass Spectrometry 23, 1855- 1864 (2012).[59] Soltysik, D. A., Thomasson, D., Rajan, S. & Biassou, N. Improving the use of principal component analysis to reduce physiological noise and motion artifacts + +<--- Page Split ---> + +to increase the sensitivity of task- based fMRI. Journal of Neuroscience Methods 241, 18- 29 (2015). + +[60] Cattai, T., Delfino, A., Scarano, G. & Colonnese, S. Vipda: A visually driven point cloud denoising algorithm based on anisotropic point cloud filtering. Frontiers in Signal Processing (2022). + +[61] Komorowski, D. & Pietraszek, S. The use of continuous wavelet transform based on the fast fourier transform in the analysis of multi- channel electrogastrography recordings. Journal of Medical Systems 40, 10 (2015). + +[62] Patil, R. & Bhosale, S. Medical image denoising techniques: A review. IJONEST (2022). + +[63] Vincent, P., Larochelle, H., Bengio, Y. & Manzagol, P.- A. Extracting and composing robust features with denoising autoencoders, ICML '08, 1096- 1103 (Association for Computing Machinery, New York, NY, USA, 2008). + +[64] Schmidhuber, J. Deep learning in neural networks: An overview. Neural Networks 61, 85- 117 (2015). + +[65] Abadi, M. et al. TensorFlow: Large- scale machine learning on heterogeneous systems (2015). Software available from tensorflow.org. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +KleinMobergetalSl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d_det.mmd b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9ab7b54bb5ac5b668ed8d5a5b2e38a5deffc823f --- /dev/null +++ b/preprint/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d/preprint__120d1ba0c5257251bfb472817798c67457e686f3580cb7e2c0ef83341eef279d_det.mmd @@ -0,0 +1,492 @@ +<|ref|>title<|/ref|><|det|>[[43, 107, 945, 208]]<|/det|> +# Deep-learning enabled online mass spectrometry of the reaction product from a single catalyst nanoparticle + +<|ref|>text<|/ref|><|det|>[[44, 229, 279, 275]]<|/det|> +Christoph Langhammer clangham@chalmers.se + +<|ref|>text<|/ref|><|det|>[[44, 301, 720, 510]]<|/det|> +Chalmers University of Technology https://orcid.org/0000- 0003- 2180- 1379 Henrik Klein Moberg Chalmers University of Technology Ievgen Nedrygailov Chalmers University of Technology David Albinsson Chalmers University of Technology Joachim Fritzsche Chalmers Univerity of Technology https://orcid.org/0000- 0001- 8660- 2624 + +<|ref|>sub_title<|/ref|><|det|>[[44, 550, 103, 567]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 587, 940, 630]]<|/det|> +Keywords: Single Particle Catalysis, Nanofluidic Reactors, CO Oxidation, Palladium Nanoparticles, Deep Learning, Denoising Auto- Encoder, Spectrometry, Heterogeneous Catalysis. + +<|ref|>text<|/ref|><|det|>[[44, 648, 315, 666]]<|/det|> +Posted Date: March 29th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 686, 475, 704]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3797128/v1 + +<|ref|>text<|/ref|><|det|>[[44, 723, 916, 765]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 784, 535, 803]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 839, 925, 881]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on August 5th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 62602- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[197, 163, 758, 240]]<|/det|> +# Deep-learning enabled online mass spectrometry of the reaction product from a single catalyst nanoparticle + +<|ref|>text<|/ref|><|det|>[[210, 269, 744, 304]]<|/det|> +Henrik Klein Moberg, Ievgen Nedrygailov, David Albinson, Joachim Fritzsche, Christoph Langhammer\* + +<|ref|>text<|/ref|><|det|>[[195, 319, 759, 352]]<|/det|> +Department of Physics, Chalmers University of Technology, Göteborg, SE- 41296 Sweden. + +<|ref|>text<|/ref|><|det|>[[230, 387, 723, 404]]<|/det|> +\*Corresponding author(s). E- mail(s): clangham@chalmers.se; + +<|ref|>text<|/ref|><|det|>[[250, 411, 702, 426]]<|/det|> +Contributing authors: henrik.kleinmoberg@chalmers.se; + +<|ref|>text<|/ref|><|det|>[[250, 428, 702, 444]]<|/det|> +nedrygailov@gmail.com; david@envue- technologies.com; + +<|ref|>text<|/ref|><|det|>[[399, 445, 553, 459]]<|/det|> +joafri@chalmers.se; + +<|ref|>sub_title<|/ref|><|det|>[[443, 493, 512, 505]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[206, 509, 750, 680]]<|/det|> +Extracting tiny signals from noise is a generic challenge in experimental science. In catalysis, this challenge manifests itself as the need to quantify chemical reactions on nanoscopic surface areas, such as single nanoparticles or even single atoms. Here, we address this challenge by combining the ability of nanofluidic reactors to focus reaction product from tiny catalyst surfaces towards online mass spectrometric analysis with the unrivalled ability of a constrained denoising autoencoder to discern tiny signals from noise. Using CO oxidation on Pd as model reaction, we demonstrate that the catalyst surface area required for online mass spectrometry can be reduced by \(\approx 3\) orders of magnitude compared to state of the art, down to a single nanoparticle with \(0.0072 \pm 0.00086 \mu \mathrm{m}^2\) surface area. These results constitute a new paradigm for online reaction analysis in single particle catalysis and advocate deep learning to improve resolution in mass spectrometry in general. + +<|ref|>text<|/ref|><|det|>[[205, 697, 745, 733]]<|/det|> +Keywords: Single Particle Catalysis, Nanofluidic Reactors, CO Oxidation, Palladium Nanoparticles, Deep Learning, Denoising Auto- Encoder, Spectrometry, Heterogeneous Catalysis. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[208, 83, 384, 101]]<|/det|> +## 1 Introduction + +<|ref|>text<|/ref|><|det|>[[207, 111, 831, 240]]<|/det|> +The ability to detect and analyze weak signals amidst high levels of noise has led to groundbreaking discoveries across a myriad of scientific disciplines using widely different experimental methodologies [1- 10]. Mass spectrometry constitutes such an experimental method since it enables the detection of very small amounts of particles, and has become a workhorse from (astro)physics to biology to chemistry [11, 12]. Accordingly, the method comes in different variants tailored for the species subject to characterization and the operational conditions. Mechanistically, it counts ionized particles, molecules or atoms in vacuum by separating them according to their mass/charge ratio, and thereby determines their (molecular) weight. + +<|ref|>text<|/ref|><|det|>[[207, 247, 831, 490]]<|/det|> +In catalysis and surface science, mass spectrometry is widely used to quantitatively analyze the composition of molecules desorbing from surfaces, often from a surfacechemical reaction [13, 14]. In this application, usually a quadrupole is used for mass selection due to its high resolution and sensitivity [15, 16]. Such quadrupole mass spectrometers (QMS) have been instrumental in the development of our understanding of nanocatalytic processes. Nevertheless, the technical advance of QMS systems used for so- called residual gas analysis has been limited during the last decades and, therefore, no significant progress has been made in terms of improving their limit of detection and their ability to discern very weak signals from noise [24, 25]. Concurrently, this state- of- the- art (SotA) is becoming an increasingly limiting factor since accurately measuring catalytic activity and selectivity from small amounts of precious new catalyst materials, such as size- or shape- selected model catalysts [26] or single atom catalysts [27- 29] that can only be obtained in small quantities, is crucial for the development of new catalyst materials and for advancing our understanding of catalytic processes. To this end, in the most extreme implementation, the aim of so- called single particle catalysis is to study catalytic reactions on individual nanoparticles to overcome ensemble averaging effects [30]. + +<|ref|>text<|/ref|><|det|>[[207, 496, 831, 596]]<|/det|> +The first strategy to enable such experiments employs detection of photon or electron signals that report on either the product molecules formed, on reactant molecules consumed, on the catalyst particle itself, or on temperature changes generated by the reaction [30- 35]. While both elegant and effective for the specific reactions chosen, these approaches lack the wide and generic applicability of mass spectrometry. Furthermore, they often cannot deliver information about both amounts and molecular weight of the species involved in the reaction. + +<|ref|>text<|/ref|><|det|>[[207, 603, 831, 718]]<|/det|> +To address this limitation, miniaturization of the reactor to the level of microreactors [36- 38] and even nanoreactors [39] to reduce the catalyst surface area necessary to obtain measurable (QMS) signals has been implemented. Very high sensitivity can be obtained in such systems by dramatically reducing reactor volume and thereby directing the entire gas flow from the catalyst bed to a QMS to maximize the fraction of reaction product available for analysis. In this way, as we recently have demonstrated using a single nanofluidic channel as the catalyst bed, the minimal required active catalyst surface area for online QMS measurements was reduced by ca. 1.5 orders + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 789, 115]]<|/det|> +compared to the lower limit proposed for microreactors [36], i.e., down to ca. \(10\mu \mathrm{m}^{2}\) active surface area [40, 41]. + +<|ref|>text<|/ref|><|det|>[[165, 121, 790, 278]]<|/det|> +While this indeed was a significant advance, it is still 2- 3 orders away from the ultimate dream of online QMS measurements from single nanoparticles. To overcome this gap we employ here a new type of nanofluidic reactor and combine it with a constrained convolutional denoising auto- encoder. This harnesses the reaction product focusing capability of nanofluidic reactors with the versatility, high- resolution and sensitivity of the QMS, with the power of deep learning to detect and analyze very weak signals hidden in noise. As we demonstrate on the example of CO oxidation, i.e., \(CO + \frac{1}{2} O_{2} \rightarrow CO_{2}\) , over a Pd model catalyst, this approach leapfrogs the limit of detection of a SotA QMS system by \(\approx 3\) orders, and enables online mass spectrometric analysis of reaction product from a single Pd nanoparticle with active surface area \(A \approx 0.0072 \pm 0.00086\mu \mathrm{m}^{2}\) \((7200 \pm 860\mathrm{nm}^{2})\) . + +<|ref|>text<|/ref|><|det|>[[165, 284, 790, 342]]<|/det|> +Finally, we note that the application of machine learning to different modalities of mass spectrometry has a short but lively history [42],[43] [44], [45], [46- 49] but the application of machine learning to discern small QMS signals from noise, as we do here, is hitherto completely unexplored. + +<|ref|>sub_title<|/ref|><|det|>[[165, 358, 253, 375]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[165, 385, 790, 728]]<|/det|> +The reaction of CO with \(\mathrm{O}_{2}\) over Pd catalysts forming \(\mathrm{CO}_{2}\) is one of the most studied reactions in catalysis both due to its practical relevance and due to its role in studies of structure- function correlations and (surface) oxide formation [50, 51]. Therefore, we chose it as our model reaction with the aim to enable QMS measurements of the \(\mathrm{CO}_{2}\) formed on a single Pd nanoparticle. We have developed a nanofluidic reactor connected to U- shape microfluidic in- and a straight outlet channel(s), which are connected to a macroscopic reactant inlet system operated by mass flow controllers at 3 bar, and to a QMS mounted on a UHV chamber with \(1 \cdot 10^{- 10}\) mbar base pressure (Figure 1A and Supplementary Figure 1). This in- and outlet system is connected to a nanofluidic catalyst bed with \(30 \mu \mathrm{m} \times 10 \mu \mathrm{m} \times 200 \mathrm{nm}(L \times W \times H)\) dimensions via a nanofluidic inlet channel ( \(505 \mu \mathrm{m} \times 10 \mu \mathrm{m} \times 200 \mathrm{nm}\) ) and an outlet channel with dimensions \(100 \mu \mathrm{m} \times 500 \mathrm{nm} \times 200 \mathrm{nm}\) (Figure 1B, C). This design ensures that the catalyst is operated in the well- mixed regime, where concentration gradients due to reactant conversion are effectively eliminated [40]. Using a resistive heater, the nanoreactor zone can be heated up to \(450^{\circ} \mathrm{C}\) [41]. For this study, we have fabricated four nanoreactor chips with identical gas in- and outlet systems, as well as catalyst beds. They were decorated with (regular arrays of) \(n = 1000, 10, 1\) or \(0 \mathrm{Pd}\) particles with diameter \(d = 59.4 \pm 3.7 \mathrm{nm}\) and height \(h = 23.5 \pm 1.6 \mathrm{nm}\) , as SEM image analysis reveals, grown onto a \(8 \mathrm{nm}\) thick \(\mathrm{SiO}_{2}\) support layer that separates them from a chemically inert plasmonic Au nanoparticles with \(d = 97.6 \pm 6.6 \mathrm{nm}\) and \(h = 35.3 \pm 2.3 \mathrm{nm}\) (Figure 1C- G, Supplementary Figure 2 and Methods), as we have used earlier ([41, 52]). This corresponds to an active Pd surface area \(A \approx 0.0072 \pm 0.00086 \mu \mathrm{m}^{2}\) \((7200 \pm 860 \mathrm{nm}^{2})\) per particle. We chose this particular hybrid nanoparticle design, where the Au element serves as plasmonic light scatterer (Pd is optically dark [53]), to enable the verification + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[202, 92, 816, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[205, 463, 832, 733]]<|/det|> +
Fig. 1 Experimental setup. (A) Schematic of the micro- and nanofabricated reactor chip. It is comprised of a U-shaped microfluidic in- and a simple straight outlet system that connects to the model catalyst bed (the nanofluidic reactor with the reaction zone) on either side, as well as to the high-pressure gas handling system and the QMS, respectively. The cross-sectional dimensions of the microfluidic channels are also shown. (B) Schematic depiction of the \(30\mu \mathrm{m}\) long well-mixed reaction zone (the model catalyst bed that contains the nanoparticle(s)) with cross sectional dimensions \(10\mu \mathrm{m}\times 200\mathrm{nm}\) , as well as its connections to the microfluidic in- and outlet system via a smaller microfluidic inlet channel (also with cross sectional dimensions \(10\mu \mathrm{m}\times 200\mathrm{nm}\) ) and via a nanofluidic outlet "capillary" with cross sectional dimensions \(500\mathrm{nm}\times 200\mathrm{nm}\) . Note that the schematics are not drawn to scale. (C) Dark-field scattering microscopy image of a nanofluidic catalyst bed containing 1000 nanoparticles arranged in a regular array in the reaction zone. The inset depicts a side-view scanning electron microscopy (SEM) image of the used \(\mathrm{Au / SiO_2 / Pd}\) hybrid nanostructures. Scale bar 100 nm. The catalytically inert Au element enables single-particle dark-field scattering (DF) imaging to confirm the presence of the correct number of catalyst particles inside the sealed nanofluidic reactor, without itself participating in the reaction, as we have shown earlier [41]. (D) Schematic depiction and zoomed-in DF image of the reaction zone of the nanofluidic catalyst bed used in our experiments, with a regular array of \(n = 1000\) nanoparticles that become distinctly visible in the DF image. Also shown are two top-view SEM images of a similar array of nanoparticles, as well as a side-view of a single nanoparticle, prepared on an open surface to facilitate SEM imaging. Scale bar 100 nm. (E) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing \(n = 10\) Pd nanoparticles used in our experiments. (F) Schematic depiction and zoomed-in DF image of the reaction zone of the catalyst bed containing a single \((n = 1)\) Pd nanoparticle used in our experiments. (G) Schematic depiction and zoomed-in DF image of the reaction zone of the empty catalyst bed without Pd nanoparticles \((n = 0)\) , used in our experiments as a negative control.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 87, 790, 116]]<|/det|> +of the presence of the anticipated number of particles in the enclosed nanoreactors using dark field scattering microscopy (Figure 1D- G). + +<|ref|>sub_title<|/ref|><|det|>[[165, 130, 780, 147]]<|/det|> +## CO Oxidation on Pd model Catalysts in Nanofluidic Reactors + +<|ref|>text<|/ref|><|det|>[[165, 153, 790, 310]]<|/det|> +To prepare these systems for experiments in reaction environment, we first exposed them to a conditioning sequence, followed by a full CO oxidation sequence at \(280^{\circ}\mathrm{C}\) (see Methods and Supplementary Figures 3- 5). Subsequently, we initiated the experiment sequence consisting of 15 min \(\mathrm{CO / O_2}\) mixture pulses at a constant \(6\%\) reactant concentration and starting with \(6\%\) percent CO, separated by 15 min in Ar. Such sequences were executed from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) in \(20^{\circ}\mathrm{C}\) steps for the \(\mathrm{n = 1000}\) sample and \(10^{\circ}\mathrm{C}\) steps for the \(\mathrm{n = 10}\) and \(\mathrm{n = 1}\) samples, to sweep the entire \(0\leq \alpha_{CO} =\) \(P_{CO} / (P_{O_2} + P_{CO})\leq 1\) range in steps of 0.05, where \(\alpha_{CO}\) is the relative CO concentration in the gas mixture. (Figure 2A- C). At each new temperature a single 30 min pulse of \(4\%\) \(\mathrm{O_2}\) and \(2\%\) CO in Ar carrier gas, followed by a 15 min pure Ar pulse, was applied to reset the state of the catalyst. + +<|ref|>text<|/ref|><|det|>[[165, 317, 790, 459]]<|/det|> +Focusing first on the highest reaction temperature of \(450^{\circ}\mathrm{C}\) and \(n = 1000\) (total active surface area of \(A\approx 7200\pm 860\mathrm{nm}^2\cdot 1000 = 7.2\pm 0.86\mu \mathrm{m}^2)\) , we observe a QMS response distinctly above the noise floor for all \(\alpha_{CO}\in (0,1)\) values (Figure 2D). Furthermore, starting from \(\alpha_{CO} = 1\) , we observe the characteristic reaction rate increase as more \(\mathrm{O_2}\) is added until the maximum rate is reached at \(\alpha_{CO}^{max} = 0.65\) . This is very close to the stoichiometric \(\alpha_{CO} = 0.66\) (the small difference is the consequence of the 0.05 \(\alpha_{CO}\) steps), which indicates that CO poisoning is very mild at this temperature and that mass transport gradients are negligible, as expected for a well- mixed catalyst bed. Upon decreasing \(\alpha_{CO}\) beyond \(\alpha_{CO}^{max}\) , the reaction rate decreases in an essentially linear fashion. + +<|ref|>text<|/ref|><|det|>[[165, 466, 790, 580]]<|/det|> +Decreasing temperature to \(340^{\circ}\mathrm{C}\) induces both a global decrease in reaction rate and a shift of \(\alpha_{CO}^{max}\) to a lower value of 0.5, due to stronger CO poisoning (Figure 2E). This trend continues upon temperature reduction to \(280^{\circ}\mathrm{C}\) (Figure 2F) with a QMS signal still distinctly above the noise floor and \(\alpha_{CO}^{max}\) reduced to 0.3. These measurements thus corroborate our experimental setup and approach since they deliver results in agreement with the well- established understanding of the CO oxidation over Pd catalysts [51, 54]. Hence, they constitute a relevant baseline as we further reduce the catalyst surface area towards a single Pd nanoparticle. + +<|ref|>text<|/ref|><|det|>[[165, 587, 790, 730]]<|/det|> +As the first step, we repeated the experiments for an identical nanoreactor but with \(n = 10\) , which corresponds to \(A\approx 7200\pm 860\mathrm{nm}^2\cdot 10 = 0.072\pm 0.0086\mu \mathrm{m}^2\) (Figure 2G- I). We find that at \(450^{\circ}\mathrm{C}\) , \(\mathrm{CO_2}\) pulses still are resolved and that the counts is approximately two orders smaller compared to \(n = 1000\) . Importantly, however, we also see that the \(\mathrm{CO_2}\) signal induced by the reactant pulses approaches the noise floor. Accordingly, reducing temperature makes discerning \(\mathrm{CO_2}\) produced increasingly difficult. This becomes clear when extracting the \(\mathrm{CO_2}\) counts for each \(\alpha_{CO}\) pulse and directly comparing the \(\mathrm{CO_2}\) counts vs. the \(\alpha_{CO}\) trend for the \(n = 1000\) and \(n = 10\) samples. At \(T = 450^{\circ}\mathrm{C}\) , the trends obtained are qualitatively very similar and thus corroborate that the catalyst is operated at identical conditions (Figure 2J). It is + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[208, 87, 833, 362]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[205, 387, 832, 636]]<|/det|> +
Fig. 2 CO oxidation experiments at different temperatures on 1000 and 10 Pd nanoparticles. CO and \(\mathrm{O_2}\) pulse sequence applied at a reactor temperature of (A) \(T = 450^{\circ}\mathrm{C}\) , (B) \(T = 340^{\circ}\mathrm{C}\) , (C) \(T = 280^{\circ}\mathrm{C}\) . Note that we start all experiments with a pure \(6\%\) CO pulse in Ar carrier gas and subsequently decrease the relative CO concentration and increase the relative \(\mathrm{O_2}\) concentration, such that \(0 \leq \alpha_{CO} = P_{CO} / (P_{O_2} + P_{CO}) \leq 1\) , while keeping the total reactant concentration constant at \(6\%\) . Corresponding baseline-adjusted (BA - see Methods for explanation of the BA-procedure) \(\mathrm{CO_2}\) counts measured by the QMS for 1000 Pd nanoparticles at (D) \(T = 450^{\circ}\mathrm{C}\) , (E) \(T = 340^{\circ}\mathrm{C}\) and (F) \(T = 280^{\circ}\mathrm{C}\) . We note distinct responses at all three temperatures with the expected decrease in reaction rate for lower \(T\) . The "transient" overshoots observed for small \(\alpha_{CO}\) values are the consequence of mass flow controller "dial-in" to the correct flow rate. (G-I) Corresponding \(\mathrm{CO_2}\) counts measured by the QMS for the 10 Pd nanoparticle sample. While at \(T = 450^{\circ}\mathrm{C}\) a relatively clear \(\mathrm{CO_2}\) signal is still obtained, it becomes increasingly weaker at the lower temperatures. (J) Mean value of the measured \(\mathrm{CO_2}\) QMS counts for each reactant pulse plotted vs. the corresponding \(\alpha_{CO}\) value for both the \(n = 1000\) and the \(n = 10\) Pd nanoparticle samples and obtained at \(T = 450^{\circ}\mathrm{C}\) . Shaded area depicts the standard deviation of measured BA-counts across each respective pulse. We note a maximum in the \(\mathrm{CO_2}\) formation rate at \(\alpha_{CO}^{max} = 0.65\) for both \(n = 1000\) and \(n = 10\) , as well as a significantly larger deviation of the derived counts for \(n = 10\) . (K) Same as (J) but for \(T = 340^{\circ}\mathrm{C}\) . We note a shift of \(\alpha_{CO}^{max}\) to lower values for both samples due to increasing CO poisoning and a broadening/flattening of the overall trend for \(n = 10\) , which is the consequence of the system approaching the detection limit of the QMS and the consequent significant uncertainty in the derived counts. (L) Same as (J) and (K) but for \(T = 280^{\circ}\mathrm{C}\) . We note an even further reduction of \(\alpha_{CO}^{max}\) and further uncertainty increase for \(n = 10\) .
+ +<|ref|>text<|/ref|><|det|>[[207, 653, 832, 739]]<|/det|> +also visible that the counts vs. \(\alpha_{CO}\) curve is slightly wider and flatter, and exhibits a significantly larger standard deviation (shaded area) for \(n = 10\) , in particular in the low reaction rate regimes at small and high \(\alpha_{CO}\) values. Reducing temperature further amplifies this effect as the signal approaches the noise floor for \(n = 10\) , with counts in the 10 - 40 range at \(T = 280^{\circ} \mathrm{C}\) (Figure 2K,L). Consequently, it becomes increasingly difficult to extract the \(\mathrm{CO_2}\) signal and we are approaching the limit of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 87, 790, 115]]<|/det|> +detection from a conventional data analysis perspective, where simply the number of counts is extracted from the QMS signal. + +<|ref|>text<|/ref|><|det|>[[165, 122, 790, 407]]<|/det|> +In principle, various strategies to address this issue exist, e.g., digital filtering [55], Fourier transforms [56, 57], wavelet transforms [58], etc., or statistical methods, like principal component analysis (PCA) [59], among others. When applied correctly, they enable the extraction of the desired signal from the accompanying noise and enhance the visibility of the underlying patterns in the data. However, these well- established denoising techniques have inherent limitations in the context of denoising QMS signals. Specifically, digital filtering can inadvertently remove crucial signals if they overlap with noise frequencies [60], Fourier and wavelet transforms may falter if the signal doesn't meet their assumptions about periodicity or localization [61] and PCA, which assumes the primary variance in data is due to the signal, might not be effective if noise has high variance or the signal is confined to a tight range [62]. Given these challenges, a Denoising Auto- Encoder (DAE), i.e., an artificial neural network, offers a promising solution since it can be trained to discern the structure of complex noisy data and reconstruct the underlying signal [63] (see Supplementary Section 3 and Supplementary Table 1). In practice, this is achieved by introducing examples of simulated signal with experimentally measured noise and training the network to reconstruct the (original) simulated signal. By also constraining the autoencoder's latent space to have a step function distribution, we introduce a prior, which enables denoising well below Signal- to- Noise Ratio (SNR) of 1 and ensures robustness in the final representations [64]. + +<|ref|>sub_title<|/ref|><|det|>[[165, 420, 775, 453]]<|/det|> +## A Constrained Denoising Auto-Encoder to improve the QMS Limit of Detection + +<|ref|>text<|/ref|><|det|>[[165, 460, 790, 674]]<|/det|> +We applied the DAE to our QMS experiments by using experimentally measured QMS readout consisting of intrinsic measurement noise and noise induced by the nanoreactor setup (e.g. fluctuations in \(\mathrm{CO}_{2}\) concentration stemming from impurities in the used gases and/or tiny leaks), combined with an underlying true \(\mathrm{CO}_{2}\) signal stemming from the catalytic reaction on the Pd nanoparticle(s), as input (Figure 3). This combined signal is then compressed through encoder \(E_{\theta}\) to form a latent space inside the bottleneck, which is constrained through a consistency loss to form a step function distribution. This compressed representation is then upsampled in decoder \(D_{\theta}\) to form the reconstructed true signal, i.e., the QMS readout with a deconvolved underlying true \(\mathrm{CO}_{2}\) signal. In this design, the DAE can learn complex non- linear relationships between the noise and the true signal, enabling it to handle the kind of noise- signal interactions that might confound the other noise reducing methods mentioned above. For a comparison between the DAE and other denoising techniques applied to our QMS data, see Supplementary Section 4 and Supplementary Figures 6- 8. + +<|ref|>text<|/ref|><|det|>[[165, 682, 790, 739]]<|/det|> +To evaluate the performance of the DAE, we compare the measured baseline- adjusted (BA) QMS \(\mathrm{CO}_{2}\) counts (see Methods for explanation of the BA- procedure) as function of \(\alpha_{CO}\) for \(n = 1000\) and \(n = 10\) , for temperatures between \(280^{\circ}\mathrm{C}\) and \(450^{\circ}\mathrm{C}\) , as well as the DAE- denoised \(\mathrm{CO}_{2}\) counts for \(n = 10\) (Figure 4). Focusing first on \(n = 1000\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[217, 87, 829, 303]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 320, 832, 424]]<|/det|> +
Fig. 3 The constrained denoising auto-encoder. The underlying signal that we want to access in our experiments, i.e., the \(\mathrm{CO_2}\) production rate step functions as a function of time, is corrupted by both correlated and uncorrelated noise in the experimentally measured readout from the QMS. Therefore, the experimental QMS readout is input to a denoising autoencoder, which transforms the input into a minimally dimensioned representation of the underlying signal and then reconstructs it sans corruption, which means, it reconstructs the original signal, rather then the QMS readout that also contains different types of noise. The DAE is trained with a standard L2 norm between reconstructed and underlying signal, and the latent space is constrained by an L1 norm and a consistency loss which forces the latent space to take on the form of the underlying signal.
+ +<|ref|>text<|/ref|><|det|>[[206, 439, 832, 596]]<|/det|> +at \(450^{\circ}\mathrm{C}\) , the catalyst exhibits the highest reaction rate at \(\alpha_{CO}^{max} = 0.65\) and all \(\mathrm{CO_2}\) pulses are clearly discernible (Figure 4A). Executing identical analysis for \(n = 10\) at \(450^{\circ}\mathrm{C}\) reveals individual \(\mathrm{CO_2}\) pulses and \(\alpha_{CO}^{max}\approx 0.6\) (Figure 4B). Passing the \(n = 10\) QMS data obtained at \(450^{\circ}\mathrm{C}\) through the DAE reveals that it delivers more distinct and systematically in/decreasing \(\mathrm{CO_2}\) pulses, with a maximum rate at \(\alpha_{CO}^{max} = 0.65\) , in excellent agreement with \(n = 1000\) (Figure 4C). Repeating the same analysis for \(280^{\circ}\) C reveals the anticipated reduction of the overall rate for both samples and a distinct shift of \(\alpha_{CO}^{max}\) to \(\alpha_{CO}^{max}\approx 0.2\) , due to severe CO poisoning (Figure 4D- F). Importantly, for \(n = 10\) , it is now clear that the standard analysis is approaching the limit of detection where the \(\mathrm{CO_2}\) pulses are stochastic, whereas the DAE- denoised data still exhibits the same clear trend as seen for \(n = 1000\) with the standard analysis. + +<|ref|>text<|/ref|><|det|>[[206, 603, 832, 732]]<|/det|> +As the next step, we extract the mean \(\mathrm{CO_2}\) counts for each \(\alpha_{CO}\) pulse along the \(\alpha_{CO}\) sweep for \(n = 1000\) , obtained for temperature steps of \(20^{\circ}\mathrm{C}\) from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) (Figure 4G). This reveals a constant and stoichiometric \(\alpha_{CO}^{max} = 0.65\) down to \(380^{\circ}\mathrm{C}\) , below which \(\alpha_{CO}^{max}\) systematically shifts to smaller values due to increasing CO poisoning, and reaches \(\alpha_{CO}^{max} = 0.25\) at \(280^{\circ}\mathrm{C}\) . Furthermore, it reveals a decrease in reaction rate from \(5240\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(450^{\circ}\mathrm{C}\) , to \(1700\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(280^{\circ}\mathrm{C}\) . Plotting the same data for \(n = 10\) obtained for \(10^{\circ}\mathrm{C}\) temperature steps, and comparing the standard analysis (Figure 4H) with the DAE- based analysis (Figure 4I) reveals qualitatively very similar behaviour. However, crucially, SNR is very poor for + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[181, 100, 734, 362]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 387, 791, 601]]<|/det|> +
Fig. 4 Direct comparison of standard and DAE-enhanced QMS readout for 1000 and 10 Pd nanoparticles. (A) Baseline-adjusted (BA - see Methods for explanation) raw QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (blue line - obtained as the average measured BA-count across each pulse) for \(n = 1000\) Pd nanoparticles across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) and at \(450^{\circ}\mathrm{C}\) . (B) BA-QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (green line) for \(n = 10\) Pd nanoparticles across the entire \(\alpha_{CO}\) range and at \(450^{\circ}\mathrm{C}\) . (C) BA-QMS counts for \(\mathrm{CO_2}\) (grey line) together with the \(\mathrm{CO_2}\) signal denoised by the DAE (orange line) for \(n = 10\) across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) . (D-F) Same as (A-C) but at \(280^{\circ}\mathrm{C}\) . (G) BA-mean \(\mathrm{CO_2}\) counts for \(\alpha_{CO}\) sweeps at reactor temperatures ranging from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) in \(20^{\circ}\mathrm{C}\) steps for \(n = 1000\) . Red dots indicate the \(\alpha_{CO}\) value corresponding to the highest reaction rate, \(\alpha_{CO}^{max}\) . (H) Same as (G) but for \(n = 10\) , using \(10^{\circ}\mathrm{C}\) temperature steps. (I) Same as (H) but with the QMS signal denoised by the DAE. (J) \(\alpha_{CO}^{max}\) values extracted from (G) for all temperatures for \(n = 1000\) , plotted as function of \(\alpha_{CO}\) . We note the constant stoichiometric \(\alpha_{CO}^{max}\) value of 0.65 down to \(T = 360^{\circ}\mathrm{C}\) . Below this temperature, we find a systematic shift to lower \(\alpha_{CO}^{max}\) values, as a consequence of increasing CO poisoning. (K) Same as (J) but for \(n = 10\) . We note the qualitatively similar trend compared to \(n = 1000\) but also the significantly higher spread in the data points. (L) Same as (K) but based on the DAE-denoised QMS signal in (I). Clearly, the uncertainty in \(\alpha_{CO}^{max}\) is significantly reduced and the T-dependent trend of \(\alpha_{CO}^{max}\) for both \(n = 1000\) and \(n = 10\) is now very similar. The small discrepancies are discussed in the text.
+ +<|ref|>text<|/ref|><|det|>[[165, 618, 791, 735]]<|/det|> +the standard analysis to the point where the underlying chemical dynamics become obscured by stochastic shifts in, e.g. \(\alpha_{CO}^{max}\) . This is different for the DAE- based data, which reproduces the T- dependent \(\alpha_{CO}^{max}\) trend found for the \(n = 1000\) sample very clearly (Figure 4J- L). The situation is similar when comparing the absolute BA - \(\mathrm{CO_2}\) counts at \(\alpha_{CO}^{max}\) at \(450^{\circ}\mathrm{C}\) . For \(n = 1000\) , we count 5240 \(\mathrm{CO_2}\) molecules, whereas we count 90 and 65 for \(n = 10\) for the standard and DAE- based analysis, respectively. The DAE- based value of 65 is indeed proportional to a \(100\times\) decrease in catalytic surface area (compared to \(n = 1000\) ), within a relative error of roughly \(10\%\) . The results + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 831, 130]]<|/det|> +demonstrate that the DAE effectively reconstructs the catalysts' \(\mathrm{CO}_{2}\) production rate and, e.g., the temperature- dependent poisoning state of the system, even when dealing with considerably noisy data and the QMS operated very close to its limit of detection. + +<|ref|>sub_title<|/ref|><|det|>[[207, 144, 763, 177]]<|/det|> +## Resolving \(CO_{2}\) Reaction Product formed on a single Pd Nanoparticle using the DAE + +<|ref|>text<|/ref|><|det|>[[207, 183, 832, 368]]<|/det|> +We put the DAE to the test by further reducing the amount of catalyst by a factor 10, down to a single Pd nanoparticle with a surface area of approx. \(7200 \mathrm{nm}^{2}\) . We used a chip with \(n = 1\) (cf. Figure 1E) and an empty chip with \(n = 0\) as the control (cf. Figure 1F). Clearly, even at the highest temperature of \(450^{\circ}\mathrm{C}\) , the \(\alpha_{CO}\) sweeps for \(n = 0\) and \(n = 1\) look very similar using the standard analysis (Figure 5A- B). They are characterized by small stochastic \(\mathrm{CO}_{2}\) pulses, indicating a certain level of background \(\mathrm{CO}_{2}\) in the reactant gas mixture, as well as the signal stemming from the \(\mathrm{CO}_{2}\) produced by the catalyst particle drowning in background noise. This is further corroborated by extracting the BA- \(\mathrm{CO}_{2}\) counts and plotting them vs. \(\alpha_{CO}\) for both \(n = 1\) (7 identical sweeps) and \(n = 0\) (5 identical sweeps), which reveals that no clear activity trend as function of \(\alpha_{CO}\) is resolved for \(n = 1\) , despite a generally slightly higher number of counts compared to \(n = 0\) (Figure 5C and Supplementary Figures 5 and 9). + +<|ref|>text<|/ref|><|det|>[[207, 375, 832, 490]]<|/det|> +Applying the DAE first to the \(\alpha_{CO}\) sweep at \(450^{\circ}\mathrm{C}\) for \(n = 1\) reveals a different picture (Figure 5D). Rather than stochastic \(\mathrm{CO}_{2}\) pulses, a clear trend of in- and decreasing \(\mathrm{CO}_{2}\) counts along the \(\alpha_{CO}\) sweep is recovered, which again exhibits a maximum at the stoichiometric \(\alpha_{CO}^{max} = 0.65\) . This is in very good agreement with the corresponding experiments for \(n = 10\) and \(n = 1000\) . Applying the DAE to the \(\alpha_{CO}\) sweep at \(450^{\circ}\mathrm{C}\) for the \(n = 0\) control outputs a close to completely flat baseline at zero BA- counts, corroborating that the DAE is able to eliminate the noise contaminating the raw QMS signal (Figure 5E). + +<|ref|>text<|/ref|><|det|>[[207, 496, 832, 667]]<|/det|> +Encouraged, we apply the DAE to \(\alpha_{CO}\) sweeps measured at temperatures \(330^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) for \(n = 1\) and plot the extracted \(\mathrm{CO}_{2}\) counts vs. \(\alpha_{CO}\) (Figure 5F). Remarkably, we clearly resolve the anticipated increase of the reaction rate to an \(\alpha_{CO}^{max} = 0.65\) for decreasing \(\alpha_{CO}\) , and the subsequent decrease of the reaction rate upon decreasing \(\alpha_{CO}\) . Furthermore, the extracted number of \(\mathrm{CO}_{2}\) counts \(\approx 6\) is approximately 10 times lower than for \(n = 10\) at \(450^{\circ}\mathrm{C}\) , which is in excellent agreement with a reduction of the active surface area by a factor of 10 from \(n = 10\) to \(n = 1\) . Finally, upon reduction of temperature, we see an indication of a shift of \(\alpha_{CO}^{max}\) to lower values at \(410^{\circ}\mathrm{C}\) , before the DAE- signal also drops below the limit of detection of a single count. Similar analysis of the \(n = 0\) control at \(450^{\circ}\mathrm{C}\) consistently delivers \(\mathrm{CO}_{2}\) counts very close to or significantly below the limit of detection of a single count and thus confirms the ability of the DAE to resolve \(\mathrm{CO}_{2}\) produced by the single Pd nanoparticle (Figure 5F). + +<|ref|>text<|/ref|><|det|>[[207, 674, 832, 731]]<|/det|> +We put these results into perspective by comparing them to the data obtained for \(n = 10\) and \(n = 1000\) . Specifically, we plot the \(\alpha_{CO}^{max}\) values for \(n = 1000\) , \(n = 10\) DAE- denoised and \(n = 1\) DAE- denoised for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) , i.e., the range for which the \(n = 1\) DAE- signal is above the limit of detection (Figure 5G). These values are + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 215]]<|/det|> +very similar for all three samples and very close to the stoichiometric value of \(\alpha_{CO}^{max} = 0.66\) . This is important because it confirms that the catalyst experiences identical reaction conditions in all three implementations in terms of number of particles. We also compare the \(\mathrm{CO_2}\) counts obtained at \(\alpha_{CO}^{max}\) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) normalized by the number of particles on the respective sample, i.e., \(n = 1000\) , \(n = 10\) and \(n = 1\) (Figure 5H). This reveals particle- number- normalized \(\mathrm{CO_2}\) count values between 4 and 6 and thus corroborates direct scaling of extracted QMS counts with catalyst surface area, within the expected level of uncertainty that is imposed, e.g., by slightly different particle dimensions on each of the three chips. + +<|ref|>sub_title<|/ref|><|det|>[[165, 230, 307, 249]]<|/det|> +## Conclusions + +<|ref|>text<|/ref|><|det|>[[165, 258, 790, 500]]<|/det|> +We have demonstrated how the combination of nanofluidic reactors and DAE- based deep learning enables the reduction of catalyst surface area required for online QMS analysis of reaction product in the gas phase by \(\approx 3\) orders from the current SotA, down to the level of single nanoparticles. This breakthrough was enabled by the ability of nanofluidic reactors to focus reaction product from tiny catalyst surfaces towards analysis, and the unrivalled capacity of autoencoders to discern tiny signals from noise. We have illustrated this on the example of the CO oxidation reaction on Pd nanoparticle model catalysts \(A \approx 0.0072 \pm 0.00086 \mu \mathrm{m}^2\) ( \(7200 \pm 860 \mathrm{~nm}^2\) ) per particle, localized inside well- mixed nanofluidic catalyst beds. We analyzed nanoreactors with \(n = 1000\) , 10, 1 and 0 Pd particles across a temperature range from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) and found the characteristic dependence of the reaction rate on \(\alpha_{CO} = P_{CO} / (P_{O_2} + P_{CO})\) , as well as a stoichiometric highest reaction rate at \(\alpha_{CO}^{max} = 0.66\) for the highest temperatures, with a distinct shift to lower \(\alpha_{CO}\) at lower temperatures, due to CO poisoning. While standard data analysis enabled meaningful results for \(n = 1000\) and \(n = 10\) (at the highest temperatures) only, DAE denoising produced reliable QMS data even for \(n = 1\) , and facilitated the analysis of the reaction dynamics in terms of \(\alpha_{CO}\) - dependent surface poisoning also for a single Pd nanoparticle. + +<|ref|>text<|/ref|><|det|>[[165, 507, 790, 593]]<|/det|> +In a wider perspective, we highlight that our approach is not specific to CO oxidation, and can be adapted for other catalytic reactions and in principle any measurement problem, in which the structure of the underlying true signal is known. Our results thus constitute a new paradigm to significantly improve the resolution in online reaction analysis in (single particle) catalysis and advocate deep learning to extract tiny QMS signals from noise. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[225, 144, 772, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 429, 832, 700]]<|/det|> +
Fig. 5 Online mass spectrometry from a single Pd nanoparticle. (A) Baseline-adjusted (BA) raw QMS counts for \(\mathrm{CO_2}\) (grey line) together with the mean \(\mathrm{CO_2}\) signal (green line - obtained as average measured BA-count across each pulse) for \(n = 1\) Pd nanoparticle, measured across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) . (B) Same as (A) but for an empty nanochannel, i.e., \(n = 0\) . Note the very similar and stochastic appearance of apparent \(\mathrm{CO_2}\) pulses both for \(n = 1\) and \(n = 0\) when analyzed in the standard way. (C) Mean BA-CO2 counts extracted for each \(\alpha_{CO}\) pulse based on the standard analysis across the entire \(\alpha_{CO}\) range, \(\alpha_{CO}\in (0,1)\) , and at \(450^{\circ}\mathrm{C}\) , for both \(n = 0\) (black lines) and \(n = 1\) (green lines). For \(n = 1\) we executed 7 consecutive \(\alpha_{CO}\) sweeps and for \(n = 0\) we executed 5 consecutive \(\alpha_{CO}\) sweeps. As key observation, we note high reproducibility of the \(\alpha_{CO}\) sweeps for both \(n = 0\) and \(n = 1\) , as well as that no clear activity trend as function of \(\alpha_{CO}\) is resolved for the \(n = 1\) sample, despite a generally slightly higher number of counts compared the \(n = 0\) reference. (D) Same as (A) but for DAE-deniosed BA-QMS \(\mathrm{CO_2}\) counts (purple line). Note the significantly smaller but at the same time non-stochastic \(\mathrm{CO_2}\) pulses resolved by the DAE compared to the standard analysis. (E) Same as (D) but for an empty nanochannel, i.e., \(n = 0\) . Note that using the DAE, a flat baseline at zero BA-counts is obtained for the empty nanochannel. (F) Same as (C) but for DAE-deniosed data, where distinct reaction rate maxima are resolved for \(n = 1\) (purple curves), with \(\alpha_{CO}^{max}\) ranging between 0.65 and 0.6 for temperatures between \(450^{\circ}\mathrm{C}\) and \(410^{\circ}\mathrm{C}\) (inset), in good agreement with the earlier results for \(n = 10\) and \(n = 1000\) . For lower temperatures, as well as for \(n = 0\) (black lines representing multiple sweeps at \(450^{\circ}\mathrm{C}\) ), the DAE outputs counts \(< 1\) , which is physically unreasonable and thus defined as the limit of detection. (G) \(\alpha_{CO}^{max}\) values for \(n = 1000\) (blue), \(n = 10\) DAE-deniosed (orange) and \(n = 1\) DAE-deniosed (purple) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) , i.e., the range for which the \(n = 1\) DAE-signal is above the limit of detection. (H) \(\mathrm{CO_2}\) counts normalized by the number of particles on the respective sample, obtained at \(\alpha_{CO}^{max}\) for \(n = 1000\) (blue), \(n = 10\) DAE-deniosed (orange) and \(n = 1\) DAE-deniosed (purple) for \(400 \leq T \leq 450^{\circ}\mathrm{C}\) .
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[165, 81, 270, 100]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[165, 111, 444, 129]]<|/det|> +## Deep Learning Architecture + +<|ref|>text<|/ref|><|det|>[[165, 135, 790, 306]]<|/det|> +We employed a constrained denoising autoencoder to denoise the experimental QMS data. This architecture was designed to accept an input shape of 6400 timesteps, equalling the full size of a single \(\alpha_{CO}\) sweep. The encoder segment consists of seven convolutional layers, each with 32 neurons and a kernel size of 9. These layers are systematically interspersed with max- pooling operations to achieve optimal data compression and utilized the Leaky Rectified Linear Unit (LeakyReLU) as its activation function due to its proven efficacy in handling complex non- linearities whilst avoiding mode collapse, vanishing and exploding gradients. Following this sequence, the encoded data is transformed into its latent space through a dense layer, reducing its dimensionality to equal the number of steps of the \(\alpha_{CO}\) sweep. This compressed representation is then reshaped and combined with the previously encoded data to retain both quantitative and structural information of the underlying signal to the decoder. + +<|ref|>text<|/ref|><|det|>[[165, 313, 790, 370]]<|/det|> +In the symmetrical decoder segment, the model reconstructs the data through another series of seven convolutional layers, each followed by an upsampling operation, ensuring accurate data restoration. The final layer then produces a reconstruction of the initial data. + +<|ref|>text<|/ref|><|det|>[[165, 377, 790, 491]]<|/det|> +To enhance the model's accuracy and ensure it only outputs functions representative of the underlying signal, we incorporated a consistency loss function which compares mean values across specific data segments of the reconstructed input with their corresponding latent space representations. This function ensures that the autoencoder's reconstructed output aligns structurally with the original data. The network is trained with an MSE function between target and predicted data segments, as elaborated upon in Supplementary Table 1 and corresponding discussion. Overall, this architecture effectively balances noise reduction, feature retention, and data consistency. + +<|ref|>text<|/ref|><|det|>[[165, 498, 789, 528]]<|/det|> +The architecture was implemented through TensorFlow [65] and the code can be found in the corresponding GitHub page. + +<|ref|>sub_title<|/ref|><|det|>[[165, 541, 433, 558]]<|/det|> +## Synthetic Data Generation + +<|ref|>text<|/ref|><|det|>[[165, 564, 790, 722]]<|/det|> +The system utilizes two generator functions: a generator for the underlying signal and a generator for the noise. The noise is generated first, and is defined by its standard deviation and shape of the gas pulses. It is generated in two parts; firstly from typical stochastic fluctuations in the QMS readout itself such as thermal noise, environmental fluctuations, residual particles (flicker noise), shot noise and ion feedback noise. We model these by a gaussian distribution with a high range of variability, to encapsulate the (white noise) effects of all these sources in the training set. Secondly, there is contamination noise, resultant of reagent reactions which occur outside the catalytically active surface area of the nanoparticles. We model these by an added term for each gas pulse, proportional to the concentration of input reagents to the system, also with a high variability. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 831, 172]]<|/det|> +The signal generator creates a step function with a desired signal- to- noise ratio \(SNR = \frac{\mu_{signal}}{\sigma_{noise}}\) , which is chosen during training to match the SNR of 1- 10 Pd NP reactor readouts based on the given noise. The steps are always placed the same (i.e. in an on- off pulsing form), but each with a random signal value. Thus, there is no inherent inductive bias regarding the relation of one pulse to another within the same overall step function. + +<|ref|>sub_title<|/ref|><|det|>[[207, 186, 444, 203]]<|/det|> +## Deep Learning Training + +<|ref|>text<|/ref|><|det|>[[207, 209, 832, 380]]<|/det|> +The training consists of two steps. Firstly, a curriculum learning scheme is employed on artificially generated signals on gaussian- distributed noise with a pre- defined SNR. The SNR is drawn from a uniform distribution in the range \(SNR \in (0.95, 1)\) , quantitatively corresponding roughly to the highest amount of experimentally measured catalytic activity of the \(n = 10\) Pd sample at \(T = 450^{\circ}\mathrm{C}\) . The lower end of this range decreases by a factor 0.05 to a minimum of 0 each time the network is trained to convergence, as defined by an early stopping mechanism with a patience of 32 epochs. This scheme helps prevent problems of vanishing gradient and thus mode collapse resultant of directly training on extremely- low SNR examples. The artificial data generation allows the model to train on a broad range of noises and signal distributions, increasing its robustness by forcing it to learn generalizable functions for denoising. The loss curve for this training can be found in Supplementary Figure 13. + +<|ref|>text<|/ref|><|det|>[[207, 387, 832, 486]]<|/det|> +The second step consists of training on artificial signal distributions in the range \(SNR \in (0, 1)\) , corresponding to the full range of SNRs considered in this work, generated on examples of experimentally measured noise. This fine- tuning step ensures that the final functions learnt by the network work appropriately for the particular noise distributions relevant to the experimental conditions at inference time. The noise is here defined as the output of a CO oxidation sequence of a \(n = 0\) Pd sample chip at \(T = 40^{\circ}\mathrm{C}\) . + +<|ref|>sub_title<|/ref|><|det|>[[207, 502, 345, 519]]<|/det|> +## Preprocessing + +<|ref|>text<|/ref|><|det|>[[207, 525, 832, 725]]<|/det|> +QMS data was preprocessed using a custom Python script (available online, see Supplementary Materials) developed for this study. To adjust for baseline shifts of the QMS signal over time, which can occur if the pressure inside the UHV chamber on which the QMS is mounted has not yet entirely stabilized, the following procedure is applied. First, a separate linear fit is made to the QMS signal for each pulse in pure Ar. Secondly, this fit is subtracted from said QMS signal and from the QMS signal of the next corresponding \(\alpha_{CO}\) pulse in time. This shifts the baseline to a consistent level for each individual pulse within a sweep. This procedure is referred to baseline adjustment (BA) in the main text. Outliers, defined as individual QMS data points deviating more than three standard deviations from the mean, were removed. The mean QMS counts value for each period was thereafter calculated and thus constituted the standard analysis for this study. Finally, the data are scaled down by a factor 100 to bring it in range with the order of magnitude of a standard gaussian, which is how the neural network's weights are initialized and therefore stabilizes training. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[166, 85, 364, 101]]<|/det|> +## Experimental Setup + +<|ref|>text<|/ref|><|det|>[[165, 108, 790, 536]]<|/det|> +Experimental SetupExperiments were performed on a setup reported in detail earlier [40, 41] (see Supplementary Figure 1). In brief, it is comprised of an elevated/atmospheric pressure gas mixing/handling unit connected to the inlet side of a nanofluidic chip holder that, on the outlet side, is connected to an ultrahigh vacuum system (base pressure of \(10^{- 10}\) mbar) equipped with a triple filter QMS equipped with a pulsed ion counting detector (Hiden HAL/3F PIC) and a gold- plated ion source. The QMS was operated in SEM mode and at 12 s sampling rate. The gas mixing/handling unit was built from quarter- inch stainless steel tubing with Vacuum Coupling Radiation (VCR) (Swagelok) fittings as connectors. The nanofluidic chip was hosted in a stainless steel connection block with welded VCR fittings, which provided gas and QMS connection. The connection block featured interior gas lines connected to exterior flexible stainless steel tubing. The gas inlet and outlet of the chip were sealed with perfluoroelastomer (FPM) O- rings to the connection block. To avoid unwanted background signals, a constant flow of Ar (18 ml/min) was flushed around the O- rings, limiting any diffusion to Ar. Connectors on the connector block allowed for integration with four Bronkhorst Low \(\Delta \mathrm{P}\) mass flow controllers and a pressure controller (max pressure of 10 bar). This system facilitate the creation of a gas mixture with up to four different gases and a defined inlet pressure up to 10 bar (here we used 4 bar). All gas flows and pressures were controlled and monitored through a customized LabVIEW program. The system can be heated up to \(450^{\circ}\mathrm{C}\) K using a microfabricated resistive heater on the backside of the chip. Temperature readout was performed through a four- wire resistance temperature probe microfabricated on the backside of the chip at the position of the nanofluidic system. The chip was electrically connected for heating and temperature readout through six gold- plated electronic spring spins embedded in a machined ceramic block. The pins are connected to a temperature controller (Lakeshore 335) operated via a LabVIEW program. The chips are calibrated by placing one in an oil bath with a thermocouple readout to monitor the temperature before any measurements are performed, see Supplementary Figure 14. The chip holder is equipped with a water- cooled copper block to maintain a constant temperature, helping to reduce mechanical movements due to heat transfer into the holder assembly. + +<|ref|>sub_title<|/ref|><|det|>[[165, 550, 533, 566]]<|/det|> +## Sample Mounting and Pre-treatment + +<|ref|>text<|/ref|><|det|>[[165, 573, 790, 688]]<|/det|> +Sample Mounting and Pre- treatmentPrior to mounting of a nanofluidic chip, the pipe connecting the outlet of the nanofluidic chip to the inlet of the QMS vacuum chamber are heated to \(353\mathrm{K}\) to minimize water adsorption. Once the chip was mounted, the system was pumped for up to 72 hours, until the base pressure of the QMS chamber reached the desired of \(10^{- 10}\) mbar. After mounting and pumping/baking, the sample was heated to \(280^{\circ}\mathrm{C}\) and subsequently exposed to 20 cycles of alternating pulses of \(10\%\) CO in Ar carrier gas and \(15\%\) \(\mathrm{O_2}\) in Ar carrier gas, each 15 min long, followed by a complete \(\alpha_{CO}\) sweep, in order to activate the catalyst and reach a stable QMS signal baseline. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[207, 84, 478, 101]]<|/det|> +## CO Oxidation Experiments + +<|ref|>text<|/ref|><|det|>[[207, 107, 832, 293]]<|/det|> +For the CO oxidation \(\alpha_{CO}\) sweep experiments, CO (10% in Ar) and \(\mathrm{O_2}\) (15% in Ar) were used with Ar carrier gas (99.99999% purity). The inlet pressure was set to 4 bar, and a total flow of \(10\mathrm{ml / min}\) through the microchannels was applied. The experiment sequence consists of 15 min pulses of \(\mathrm{CO / O_2}\) mixtures at a constant 6% percent reactant concentration, separated by 15 min in pure Ar (Figure 2A- C) at constant temperatures ranging from \(280^{\circ}\mathrm{C}\) to \(450^{\circ}\mathrm{C}\) . In each pulse, the concentration of \(\mathrm{CO / O_2}\) is varied such that \(\alpha_{CO}\) is systematically varied from 1 to 0 in steps of 0.05 for each subsequent pulse. The only exception is the first (and last) two pulses, which vary in steps of 0.11 (0.02) and 0.02 (0.11), respectively. This is due to instrumental limitations of the MFC controllers at very low values of absolute flow, where they are unable to control the flow accurately. Each experiment is also started with a single 30 min pulse of \(4\%\) \(\mathrm{O_2}\) and \(2\%\) CO, followed by a 15 min Ar pulse, to reset the state of the catalyst before each sequence. + +<|ref|>sub_title<|/ref|><|det|>[[207, 307, 616, 324]]<|/det|> +## Nanofabrication of the Nanoreactor Chip + +<|ref|>text<|/ref|><|det|>[[207, 330, 832, 460]]<|/det|> +Fabrication of the nanofluidic systems was carried out in cleanroom facilities of Fed. Std.209 E Class 10 - 100, using electron- beam lithography (JBX- 9300FS / JEOL Ltd), direct- laser lithography (Heidelberg Instruments DWL 2000), photolithography (MA 6 / Suss MicroTec), reactive- ion etching (Plasmalab 100 ICP180 / Oxford Plasma Technology and STS ICP), electron- beam evaporation (PVD 225 / Lesker), magnetron sputtering (MS150 / FHR), deep reactive- ion etching (STS ICP / STS) and wet oxidation (wet oxidation / Centrotherm), fusion bonding (AWF 12/65 / Lenton), and dicing (DAD3350 / Disco). In particular, the fabrication steps comprised the following processing steps of a \(4^{\circ}\) - silicon (p- type) wafer: + +<|ref|>text<|/ref|><|det|>[[207, 466, 832, 608]]<|/det|> +Fabrication of alignment marks: Spin coating HMDS adhesion promoter (MicroChem) at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 120 sec. Spin coating UV5 (MicroChem) at \(2000\mathrm{rpm}\) for 60 sec and soft baking at \(130^{\circ}\mathrm{C}\) for 120 sec. Electron- beam exposure of alignment marks for both optical and electron- beam lithography at \(10\mathrm{nA}\) with a shot pitch of \(20\mathrm{nm}\) and \(34\mu \mathrm{C / cm^2}\) exposure dose. Postexposure bake at \(130^{\circ}\mathrm{C}\) for 90 sec. Development in MF- 24A (Microposit) for 90 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Reactive- ion etching (RIE) for 15 sec at \(60\mathrm{mTorr}\) chamber pressure, \(60\mathrm{W}\) RF- power, \(60\mathrm{scm}\) \(\mathrm{O_2}\) flow (descum). RIE for 15 min at \(40\mathrm{mTorr}\) chamber pressure, \(50\mathrm{W}\) RF- power, \(100\mathrm{W}\) ICP- power, \(50\mathrm{scm}\) \(\mathrm{Cl}_2\) flow (600 nm etch depth in silicon). + +<|ref|>text<|/ref|><|det|>[[207, 614, 831, 659]]<|/det|> +Thermal oxidation: Cleaning in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. Wet oxidation in water atmosphere for 45 min at \(950^{\circ}\mathrm{C}\) (200 nm thermal oxide). + +<|ref|>text<|/ref|><|det|>[[207, 666, 832, 739]]<|/det|> +Fabrication of nanochannels: Spin coating HMDS adhesion promoter (MicroChem) at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 120 sec. Spin coating UV5 (MicroChem) at \(2000\mathrm{rpm}\) for 60 sec and soft baking at \(130^{\circ}\mathrm{C}\) for 120 sec. Electron- beam exposure of nanochannels at \(1\mathrm{nA}\) with a shot pitch of \(10\mathrm{nm}\) and \(34\mu \mathrm{C / cm^2}\) exposure dose. Post- exposure bake at \(130^{\circ}\mathrm{C}\) for 90 sec. Development in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 85, 790, 158]]<|/det|> +MF- 24A (Microposit) for 90 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. RIE for 10 sec at \(40\mathrm{mTorr}\) chamber pressure, \(40\mathrm{W}\) RF- power, \(40\mathrm{scm}\) \(\mathrm{O}_2\) flow (descum). RIE for 100 sec at \(8\mathrm{mTorr}\) chamber pressure, \(50\mathrm{W}\) RF- power, \(50\mathrm{scm}\) \(\mathrm{NF}_3\) flow (70 nm etch depth in thermal oxide). Cleaning in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. + +<|ref|>text<|/ref|><|det|>[[165, 165, 790, 336]]<|/det|> +Fabrication of microchannels: Spin coating HMDS at 3000 rpm for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Spin coating S1813 (Shipley) at 3000 rpm for 30 sec and soft baking at \(115^{\circ}\mathrm{C}\) for 2 min. Expose microchannels for 10 sec in contact aligner at \(6\mathrm{mW / cm^2}\) intensity. Development in MF- 319 (Microposit) for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Buffered oxide- etch for 3 min to remove thermal oxide, rinsing in water and drying under \(\mathrm{N}_2\) stream. Deep reactive- ion etching for 100 cycles of 7 sec at \(6\mathrm{mTorr}\) chamber pressure, \(800\mathrm{W}\) RF- power, \(8\mathrm{W}\) platen power, \(130\mathrm{scm}\) \(\mathrm{SF}_6\) flow (Si- etch), and of 5 sec at \(6\mathrm{mTorr}\) chamber pressure, \(800\mathrm{W}\) RF- power, \(8\mathrm{W}\) platen power, \(85\mathrm{scm}\) \(\mathrm{C}_4\mathrm{F}_8\) flow (passivation) at a rate of \(600\mathrm{nm / cycle}\) . Removal of resist in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. The resulting channels have a depth of \(60\mu \mathrm{m}\) measured using a Dektak 150 surface profiler. + +<|ref|>text<|/ref|><|det|>[[165, 342, 790, 486]]<|/det|> +Fabrication of inlets (from backside): Magnetron- sputtering of \(200\mathrm{nm}\) Al (hard mask). Spin coating S1813 at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Expose inlets for 10 sec in contact aligner at \(6\mathrm{mW / cm^2}\) intensity. Development in MF- 319 for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Aluminum wet etch \(\mathrm{(H_3PO_4\cdot CH_3COOH\cdot HNO_3\cdot H_2O}\) (4:4:1:1)) for 10 min to clear the hard mask at inlet positions. Deep reactive- ion etching for 300 cycles of 12 sec at \(5\mathrm{mTorr}\) chamber pressure, \(600\mathrm{W}\) RF- power, \(10\mathrm{W}\) platen power, \(130\mathrm{scm}\) \(\mathrm{SF}_6\) flow (Si- etch), and of 7 sec at \(5\mathrm{mTorr}\) chamber pressure, \(600\mathrm{W}\) RF- power, \(10\mathrm{W}\) platen power, \(85\mathrm{scm}\) \(\mathrm{C}_4\mathrm{F}_8\) flow (passivation) at a rate of \(2\mu \mathrm{m / cycle}\) . Removal of Al- hard mask in \(50\mathrm{mL}\mathrm{H}_2\mathrm{O}_2 + 100\mathrm{mL}\mathrm{H}_2\mathrm{SO}_4\) at \(130^{\circ}\mathrm{C}\) for 10 min, rinsing in water and drying under \(\mathrm{N}_2\) stream. + +<|ref|>text<|/ref|><|det|>[[165, 491, 790, 607]]<|/det|> +Fabrication of heater elements on the backside: Spin coating HMDS at \(3000\mathrm{rpm}\) for 30 sec and soft baking on a hotplate at \(115^{\circ}\mathrm{C}\) for 2 min. Spin coating LOR3A (MicroChemicals) at \(3000\mathrm{rpm}\) for 30 sec and soft baking at \(180^{\circ}\mathrm{C}\) for 5 min. Spin coating S1813 (Shipley) at \(3000\mathrm{rpm}\) for 30 sec and soft baking at \(115^{\circ}\mathrm{C}\) for 2 min. Expose heater elements with direct- laser lithography at \(10\mathrm{mW / cm^2}\) intensity. Development in MF- 319 (Microposit) for 60 sec, rinsing in water and drying under \(\mathrm{N}_2\) stream. Electron- beam evaporation of \(10\mathrm{nm}\mathrm{Cr} / 100\mathrm{nm}\mathrm{Pt}\) . Lift- off in remover Rem1165 (MicroChemicals), rinsing in isopropanol, and drying under \(\mathrm{N}_2\) stream. + +<|ref|>text<|/ref|><|det|>[[165, 613, 790, 715]]<|/det|> +Fabrication of nanoparticles inside nanochannels: Spin coating Copolymer MMA(8.5)MMA (MicroChem Corporation, \(10\mathrm{wt}\%\) diluted in anisole) at \(6000\mathrm{rpm}\) for 60 sec and soft baking on a hotplate at \(180^{\circ}\mathrm{C}\) for 5 min. Spin coating ZEP520A: anisole (1:2) at \(3000\mathrm{rpm}\) for 60 sec and soft baking at \(180^{\circ}\mathrm{C}\) for 5 min. Electronbeam exposure at \(1\mathrm{nA}\) with a shot pitch of \(2\mathrm{nm}\) and \(280\mu \mathrm{C / cm^2}\) exposure dose. Development in n- amyl acetate for 60 sec, rinsing in isopropanol and drying under \(\mathrm{N}_2\) stream. Development in methyl isobutyl ketone:isopropanol (1:1) for 60 sec, rinsing in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 831, 116]]<|/det|> +isopropanol and drying under \(\mathrm{N}_2\) stream. Electron- beam evaporation of \(\mathrm{Au / SiO_2 / Pd}\) triple layer. Lift- off in acetone, rinsing in isopropanol, and drying under \(\mathrm{N}_2\) stream. + +<|ref|>text<|/ref|><|det|>[[207, 122, 832, 195]]<|/det|> +Fusion bonding: (a) Cleaning of the substrate together with a lid (175 \(\mu \mathrm{m}\) thick \(4^{,}\) pyrex, UniversityWafers) in \(\mathrm{H_2O:H_2O_2:NH_3OH}\) (5:1:1) for \(10\mathrm{min}\) at \(80^{\circ}\mathrm{C}\) . (b) Prebonding the lid to the substrate by bringing surfaces together and manually applying pressure. (c) Fusion bonding of the lid to the substrate for \(5\mathrm{h}\) in \(\mathrm{N}_2\) atmosphere at \(550^{\circ}\mathrm{C}\) ( \(^{\circ}\mathrm{C / min}\) ramp rate). + +<|ref|>text<|/ref|><|det|>[[207, 200, 832, 244]]<|/det|> +Dicing of bonded wafers: Cutting nanofluidic chips from the bonded wafer using a resin bonded diamond blade of \(250\mu \mathrm{m}\) thickness (Dicing Blade Technology) at 35 krpm and \(1\mathrm{mm / s}\) feed rate. + +<|ref|>sub_title<|/ref|><|det|>[[208, 259, 421, 278]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[207, 287, 832, 388]]<|/det|> +This research has received funding from the European Research Council (ERC) under the European Union's Horizon Europe research and innovation program (101043480/NACAREI) and under the European Union's Horizon 2020 research and innovation programme (678941/SINCAT), and from the Knut and Alice Wallenberg Foundation projects 2016.0210 and 2015.0055. Part of this work was carried out at the Chalmers MC2 cleanroom facility and at the Chalmers Materials Analysis Laboratory (CMAL). + +<|ref|>sub_title<|/ref|><|det|>[[207, 402, 335, 421]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[212, 427, 832, 450]]<|/det|> +[1] Abbott, B. et al. Observation of gravitational waves from a binary black hole merger. Physical Review Letters 116, 061102 (2016). + +<|ref|>text<|/ref|><|det|>[[214, 465, 832, 510]]<|/det|> +[2] Driggers, J. et al. Increasing the astrophysical reach of the advanced virgo detector via the application of squeezed vacuum states of light. Physical Review Letters 122, 231102 (2019). + +<|ref|>text<|/ref|><|det|>[[214, 521, 832, 551]]<|/det|> +[3] Littenberg, T. B. & Cornish, N. J. Bayesian inference for spectral estimation of gravitational wave detector noise. Physical Review D 91, 084034 (2015). + +<|ref|>text<|/ref|><|det|>[[214, 562, 832, 620]]<|/det|> +[4] Chen, S. et al. Common- red- signal analysis with 24- yr high- precision timing of the european pulsar timing array: Inferences in the stochastic gravitational- wave background search. Monthly Notices of the Royal Astronomical Society 508, 4970- 4993 (2021). + +<|ref|>text<|/ref|><|det|>[[214, 631, 832, 675]]<|/det|> +[5] Lai, Z. & Leng, Y. Generalized parameter- adjusted stochastic resonance of duffing oscillator and its application to weak- signal detection. Sensors 15, 21327- 21345 (2015). + +<|ref|>text<|/ref|><|det|>[[214, 686, 832, 730]]<|/det|> +[6] Lopez, C., Zhong, W., Lu, S., Cong, F. & Cortese, I. Stochastic resonance in an underdamped system with fitzhug- nagumo potential for weak signal detection. Journal of Sound and Vibration 408, 16- 28 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[168, 87, 792, 116]]<|/det|> +[7] Jin, Y. & Moura, J. M. F. Time- reversal detection using antenna arrays. IEEE Transactions on Signal Processing 57, 1396- 1414 (2009). + +<|ref|>text<|/ref|><|det|>[[170, 126, 792, 156]]<|/det|> +[8] Nelson, D. Special purpose correlation functions for improved signal detection and parameter estimation, Vol. 3, 237- 240 (IEEE, 1993). + +<|ref|>text<|/ref|><|det|>[[170, 167, 792, 211]]<|/det|> +[9] Werley, C., Teo, S. M. & Nelson, K. Pulsed laser noise analysis and pump- probe signal detection with a data acquisition card. Review of Scientific Instruments 82, 123108 (2011). + +<|ref|>text<|/ref|><|det|>[[165, 222, 790, 251]]<|/det|> +[10] Algburi, R. N. A. & Gao, H. Health assessment and fault detection system for an industrial robot using the rotary encoder signal. Energies 12, 2816 (2019). + +<|ref|>text<|/ref|><|det|>[[165, 262, 790, 291]]<|/det|> +[11] Griffiths, W. J. & Wang, Y. Mass spectrometry: from proteomics to metabolomics and lipidomics. Chemical Society Reviews 37, 1432- 1448 (2008). + +<|ref|>text<|/ref|><|det|>[[165, 302, 790, 332]]<|/det|> +[12] Grayson, S. M. & Frechet, J. M. Applications of mass spectrometry to polymers. Mass spectrometry reviews 21, 223- 248 (2002). + +<|ref|>text<|/ref|><|det|>[[165, 343, 790, 372]]<|/det|> +[13] Mrksich, M. & Whitesides, G. M. Mass spectrometry of self- assembled monolayers: a new tool for molecular surface science. Science 269, 664- 665 (1995). + +<|ref|>text<|/ref|><|det|>[[165, 383, 790, 412]]<|/det|> +[14] Shappell, N. W. Mass spectrometry in the analysis of perfluorinated pollutants. Environmental science & technology 42, 4869- 4874 (2008). + +<|ref|>text<|/ref|><|det|>[[165, 423, 792, 467]]<|/det|> +[15] Brennwald, M. S., Schmidt, M., Oser, J. & Kipfer, R. A portable and autonomous mass spectrometric system for on- site environmental gas analysis. Environmental science & technology 50, 13455- 13463 (2016). + +<|ref|>text<|/ref|><|det|>[[165, 478, 790, 522]]<|/det|> +[16] Tsugawa, H., Kanazawa, M., Ogiwara, A. & Arita, M. Understanding and practical use of in- source fragmentation in liquid chromatography- mass spectrometry. Journal of mass spectrometry 48, 730- 740 (2013). + +<|ref|>text<|/ref|><|det|>[[165, 533, 790, 562]]<|/det|> +[17] Martin, L. B. et al. Mass spectrometry imaging for plant biology: a review. Phytochemistry reviews 13, 445- 488 (2014). + +<|ref|>text<|/ref|><|det|>[[165, 573, 790, 616]]<|/det|> +[18] Buchberger, A., DeLaney, K., Johnson, J. & Li, L. Mass spectrometry imaging: A review of emerging advancements and future insights. Analytical Chemistry 90 (2017). + +<|ref|>text<|/ref|><|det|>[[165, 627, 790, 656]]<|/det|> +[19] Kersten, R. D. et al. Mass spectrometry imaging of complex microbial communities. Accounts of chemical research 45, 455- 464 (2012). + +<|ref|>text<|/ref|><|det|>[[165, 667, 790, 697]]<|/det|> +[20] Bush, M. F. et al. Mass spectrometry imaging of biological tissue: an approach for multicenter studies. Analytical chemistry 89, 2558- 2563 (2017). + +<|ref|>text<|/ref|><|det|>[[165, 708, 790, 737]]<|/det|> +[21] Kasemo, B. Quartz tube orifice leaks for local, fast- response gas sampling to mass spectrometers. Review of Scientific Instruments 50, 1602- 1604 (1979). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[204, 87, 833, 130]]<|/det|> +[22] Veisi, P., Dong, X., Dong, J., Fenselau, C. & Zhang, H. Mass spectrometry imaging for visualizing in situ metabolism of endogenous metabolites and dietary phytochemicals. \*Metabolites\* 4, 131 (2014). + +<|ref|>text<|/ref|><|det|>[[206, 140, 832, 184]]<|/det|> +[23] Ligterink, W., Vincent, S. R., Masood, M. M., Choi, Y. H. & Verpoorte, R. Mass spectrometry imaging and gc- ms profiling of the mammalian peripheral sensory- motor circuit. \*Journal of visualized experiments: JoVE\* (2014). + +<|ref|>text<|/ref|><|det|>[[206, 195, 832, 224]]<|/det|> +[24] Dahl, J. M., Maiolica, A., Gstaiger, M. & Aebersold, R. Mass spectrometry for the study of protein- protein interactions. \*Methods\* 24, 230- 239 (2001). + +<|ref|>text<|/ref|><|det|>[[206, 236, 832, 279]]<|/det|> +[25] Abushareeda, W., Lyris, E., Kraiem, S., Thomas, A. & Thevis, M. Mass spectrometry in doping control: Structure characterization and analytical assays. \*Mass spectrometry reviews\* 36, 196- 217 (2017). + +<|ref|>text<|/ref|><|det|>[[206, 290, 832, 319]]<|/det|> +[26] Li, Y., Lei, Y. & Huang, W. Size- selected clusters: preparation, structure, and catalytic properties. \*Advances in Physics: X\* 3, 1437698 (2018). + +<|ref|>text<|/ref|><|det|>[[206, 330, 832, 360]]<|/det|> +[27] Qiao, B. et al. Single- atom catalysis of CO oxidation using Pt1/FeOx. \*Nature chemistry\* 3, 634- 641 (2011). + +<|ref|>text<|/ref|><|det|>[[206, 371, 832, 400]]<|/det|> +[28] Zhang, X. et al. Single- atom catalysts: synthetic strategies and electrochemical applications. \*Joule\* 2, 1242- 1264 (2018). + +<|ref|>text<|/ref|><|det|>[[206, 411, 832, 440]]<|/det|> +[29] Liu, L. & Corma, A. Single- atom catalysts: a new frontier in heterogeneous catalysis. \*Accounts of chemical research\* 51, 789- 799 (2018). + +<|ref|>text<|/ref|><|det|>[[206, 451, 832, 480]]<|/det|> +[30] Sambur, J. B. et al. Sub- particle reaction and photocurrent mapping to optimize catalyst- modified photoanodes. \*Nature\* 530, 77- 80 (2016). + +<|ref|>text<|/ref|><|det|>[[206, 492, 832, 520]]<|/det|> +[31] Wei, H. & Xu, H. Shell- isolated nanoparticle- enhanced raman spectroscopy: principle and applications. \*Chemical Society Reviews\* 40, 2494- 2507 (2011). + +<|ref|>text<|/ref|><|det|>[[206, 531, 832, 560]]<|/det|> +[32] Xu, W., Zhang, Y. & Chen, T. \*Single Particle Nanocatalysis: Fundamentals and Applications\* (2019). + +<|ref|>text<|/ref|><|det|>[[206, 572, 832, 615]]<|/det|> +[33] Hartman, T., Geitenbeek, R. G., Wondergem, C. S., van der Stam, W. & Weckhuysen, B. M. Operando nanoscale sensors in catalysis: All eyes on catalyst particles. \*ACS Nano\* 14, 3725- 3735 (2020). + +<|ref|>text<|/ref|><|det|>[[206, 626, 832, 670]]<|/det|> +[34] Hartman, T., Wondergem, C. S., Kumar, N., van den Berg, A. & Weckhuysen, B. M. Surface- and tip- enhanced raman spectroscopy in catalysis. \*The Journal of Physical Chemistry Letters\* 7, 1570- 1584 (2016). + +<|ref|>text<|/ref|><|det|>[[206, 681, 832, 724]]<|/det|> +[35] Alekseeva, S., Nedrygailov, I. I. & Langhammer, C. Single particle plasmonics for materials science and single particle catalysis. \*ACS Photonics\* 6, 1319- 1330 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 87, 792, 117]]<|/det|> +[36] Henriksen, G., Hoff, P., Larsen, R. & Alstad, J. Microreactor and method for preparing a radiolabeled complex or a biomolecule conjugate. US Patent (2009). + +<|ref|>text<|/ref|><|det|>[[165, 127, 792, 170]]<|/det|> +[37] Schouten, J. C., Nijhuis, T. A., van den Oetelaar, L. C. & van den Bleek, C. M. Microreactors for kinetic studies of heterogeneous catalytic reactions. Applied Catalysis A: General 221, 253- 264 (2001). + +<|ref|>text<|/ref|><|det|>[[165, 181, 790, 225]]<|/det|> +[38] Vesborg, P. C., Olsen, J. L., Henriksen, T. R., Chorkendorff, I. & Hansen, O. Gas- phase photocatalysis in micro- reactors. Chemical Engineering Journal 160, 738- 741 (2010). + +<|ref|>text<|/ref|><|det|>[[165, 236, 790, 279]]<|/det|> +[39] Schouten, K. J. P., Pérez- Gallent, E. & Koper, M. T. Nanoreactors: engineering catalyst particles with enzyme- like behavior. Chemical Science 8, 5159- 5166 (2017). + +<|ref|>text<|/ref|><|det|>[[165, 291, 790, 333]]<|/det|> +[40] Albinsson, D. et al. Copper catalysis at operando conditions—bridging the gap between single nanoparticle probing and catalyst- bed- averaging. Nature Communications 11, 4832 (2020). + +<|ref|>text<|/ref|><|det|>[[165, 345, 790, 373]]<|/det|> +[41] Albinsson, D. et al. Operando detection of single nanoparticle activity dynamics inside a model pore catalyst material. Science Advances 6, eaba7678 (2020). + +<|ref|>text<|/ref|><|det|>[[165, 385, 790, 442]]<|/det|> +[42] Boiko, D. A., Kozlov, K. S., Burykina, J. V., Ilyushenkova, V. V. & Ananikov, V. P. Fully automated unconstrained analysis of high- resolution mass spectrometry data with machine learning. Journal of the American Chemical Society 144, 14590- 14606 (2022). URL https://doi.org/10.1021/jacs.2c03631. + +<|ref|>text<|/ref|><|det|>[[165, 454, 790, 496]]<|/det|> +[43] Abdelmoula, W. M. et al. Peak learning of mass spectrometry imaging data using artificial neural networks. Nature Communications 12, 5544 (2021). URL https://doi.org/10.1038/s41467- 021- 25744- 8. + +<|ref|>text<|/ref|><|det|>[[165, 508, 790, 550]]<|/det|> +[44] Liebal, U. W., Phan, A. N. T., Sudhakar, M., Raman, K. & Blank, L. M. Machine learning applications for mass Spectrometry- Based metabolomics. Metabolites 10 (2020). + +<|ref|>text<|/ref|><|det|>[[165, 562, 790, 618]]<|/det|> +[45] Ji, H., Deng, H., Lu, H. & Zhang, Z. Predicting a molecular fingerprint from an electron ionization mass spectrum with deep neural networks. Analytical Chemistry 92, 8649- 8653 (2020). URL https://doi.org/10.1021/acs.analchem.0c01450. + +<|ref|>text<|/ref|><|det|>[[165, 631, 790, 673]]<|/det|> +[46] Jetybayeva, A. et al. A review on recent machine learning applications for imaging mass spectrometry studies. Journal of Applied Physics 133, 020702 (2023). URL https://doi.org/10.1063/5.0100948. + +<|ref|>text<|/ref|><|det|>[[165, 686, 790, 728]]<|/det|> +[47] Zhou, Z. & Zare, R. N. Personal information from latent fingerprints using desorption electrospray ionization mass spectrometry and machine learning. Analytical Chemistry 89, 1369- 1372 (2017). URL https://doi.org/10.1021/acs.analchem. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[206, 100, 832, 732]]<|/det|> +[48] Gessulat, S. et al. Prosit: proteome- wide prediction of peptide tandem mass spectra by deep learning. Nature Methods 16, 509- 518 (2019). URL https://doi.org/10.1038/s41592- 019- 0426- 7. [49] Tran, N. H. et al. Deep learning enables de novo peptide sequencing from data- independent- acquisition mass spectrometry. Nature Methods 16, 63- 66 (2019). URL https://doi.org/10.1038/s41592- 018- 0260- 3. [50] Zhou, Y., Wang, Z. & Liu, C. Perspective on CO oxidation over Pd- based catalysts. Catal. Sci. Technol. 5, 69- 81 (2015).[51] van Spronsen, M. A., Frenken, J. W. M. & Groot, I. M. N. Surface science under reaction conditions: CO oxidation on Pt and Pd model catalysts. Chem. Soc. Rev. 46, 4347- 4374 (2017).[52] Albinsson, D. et al. Shedding light on CO oxidation surface chemistry on single Pt catalyst nanoparticles inside a nanofluidic model pore. ACS Catalysis 11, 2021- 2033 (2021).[53] Langhammer, C., Kasemo, B. & Zoric, I. Absorption and scattering of light by Pt, Pd, Ag, and Au nanodisks: Absolute cross sections and branching ratios. The Journal of Chemical Physics 126, 194702 (2007).[54] Zhou, Y., Wang, Z. & Liu, C. Perspective on CO oxidation over Pd- based catalysts. Catal. Sci. Technol. 5, 69- 81 (2015).[55] Iwai, T. et al. A transient signal acquisition and processing method for microdroplet injection system inductively coupled plasma mass spectrometry (m- disc- ip- ms). J. Anal. At. Spectrom. 30, 1617- 1622 (2015).[56] Didriche, K., Lauzin, C., Foldes, T., de Ghellinck D'Elseghem Vaerenwijck, X. & Herman, M. The fantasio+ set- up to investigate jet- cooled molecules: focus on overtone bands of the acetylene dimer. Molecular Physics 108, 2155- 2163 (2010).[57] Fornelli, L. et al. Advancing top- down analysis of the human proteome using a benchtop quadrupole- orbitrap mass spectrometer. Journal of Proteome Research 16, 609- 618 (2017).[58] Chou, S.- W., Shiu, G.- R., Chang, H.- C. & Peng, W.- P. Wavelet- based method for time- domain noise analysis and reduction in a frequency- scan ion trap mass spectrometer. Journal of the American Society for Mass Spectrometry 23, 1855- 1864 (2012).[59] Soltysik, D. A., Thomasson, D., Rajan, S. & Biassou, N. Improving the use of principal component analysis to reduce physiological noise and motion artifacts + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[198, 86, 790, 115]]<|/det|> +to increase the sensitivity of task- based fMRI. Journal of Neuroscience Methods 241, 18- 29 (2015). + +<|ref|>text<|/ref|><|det|>[[165, 126, 790, 170]]<|/det|> +[60] Cattai, T., Delfino, A., Scarano, G. & Colonnese, S. Vipda: A visually driven point cloud denoising algorithm based on anisotropic point cloud filtering. Frontiers in Signal Processing (2022). + +<|ref|>text<|/ref|><|det|>[[165, 181, 790, 225]]<|/det|> +[61] Komorowski, D. & Pietraszek, S. The use of continuous wavelet transform based on the fast fourier transform in the analysis of multi- channel electrogastrography recordings. Journal of Medical Systems 40, 10 (2015). + +<|ref|>text<|/ref|><|det|>[[165, 236, 790, 265]]<|/det|> +[62] Patil, R. & Bhosale, S. Medical image denoising techniques: A review. IJONEST (2022). + +<|ref|>text<|/ref|><|det|>[[165, 276, 790, 320]]<|/det|> +[63] Vincent, P., Larochelle, H., Bengio, Y. & Manzagol, P.- A. Extracting and composing robust features with denoising autoencoders, ICML '08, 1096- 1103 (Association for Computing Machinery, New York, NY, USA, 2008). + +<|ref|>text<|/ref|><|det|>[[165, 331, 790, 360]]<|/det|> +[64] Schmidhuber, J. Deep learning in neural networks: An overview. Neural Networks 61, 85- 117 (2015). + +<|ref|>text<|/ref|><|det|>[[165, 372, 790, 400]]<|/det|> +[65] Abadi, M. et al. TensorFlow: Large- scale machine learning on heterogeneous systems (2015). Software available from tensorflow.org. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 768, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 280, 149]]<|/det|> +KleinMobergetalSl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/images_list.json b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d3f47c16bf9858009e3a0710fcef03dd8534d956 --- /dev/null +++ b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. The model pipeline. We (A) examine chest radiographs across several datasets with diverse populations, (B) learn a DenseNet from these data (training across all patients simultaneously), predicting the presence of the \"No Finding\" label – indicates the algorithm detect no disease for image, and (C) compare false positive rate (FPR) of this model on different subpopulations (including sex, race, age, and insurance type) to examine the algorithm's underdiagnosis rate. If the model has a higher underdiagnosis rates on certain subpopulations, such as female or Medicaid patients, this would induce significant health disparities that lead to higher rate of no clinical treatment for certain subpopulations, were the model deployed.", + "footnote": [], + "bbox": [ + [ + 130, + 92, + 884, + 277 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Analyzing underdiagnoses over subgroups of sex, age, race & insurance type within the MIMIC-CXR (CXR) dataset. The results are averaged over 5 run with different random seed ± 95% confidence interval. A. The underdiagnosis rate (measured by “No Finding” FPR). Female, 0-20, Black and/or low-income patients under Medicaid insurance have the largest underdiagnosis", + "footnote": [], + "bbox": [ + [ + 115, + 87, + 870, + 830 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 66, + 102, + 930, + 310 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 50, + 37, + 796, + 789 + ] + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4.mmd b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4.mmd new file mode 100644 index 0000000000000000000000000000000000000000..48f29679e71e2cfc1a77cb80c7b943a1d30df48f --- /dev/null +++ b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4.mmd @@ -0,0 +1,185 @@ + +# Medical imaging algorithms exacerbate biases in underdiagnosis + +Laleh Seyyed- Kalantari ( \(\boxed{ \begin{array}{r l} \end{array} }\) laleh@cs.toronto.edu ) University of Toronto, Vector Institute https://orcid.org/0000- 0002- 1059- 7125 + +Guanxiong Liu University of Toronto, Vector Institute + +Matthew McDermott Massachusetts Institute of Technology + +Irene Chen Massachusetts Institute of Technology + +Marzyeh Ghassemi University of Toronto https://orcid.org/0000- 0001- 6349- 7251 + +## Letter + +Keywords: artificial intelligence (AI), machine learning, underdiagnosis + +Posted Date: February 4th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 151985/v1 + +License: \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Medicine on December 10th, 2021. See the published version at https://doi.org/10.1038/s41591- 021- 01595- 0. + +<--- Page Split ---> + +# Medical imaging algorithms exacerbate biases in underdiagnosis + +Laleh Seyyed- Kalantari \(^{1,2*}\) , Guanxiong Liu \(^{1,2}\) , Matthew McDermott \(^{3}\) , Irene Y. Chen \(^{3}\) , Marzyeh Ghassemi \(^{1,2}\) + +\(^{1}\) University of Toronto. + +\(^{2}\) Vector Institute. + +\(^{3}\) Massachusetts Institute of Technology. + +\*Correspondence to: laleh@cs.toronto.edu. + +Artificial intelligence (AI) systems have increasingly achieved expert- level performance, particularly in medical imaging (1). However, there is growing concern that AI systems will reflect and amplify human bias against under- served subpopulations (2- 7). Such biases are especially troubling in the context of underdiagnosis: if AI systems falsely predict that patients are healthy, patients would be denied care when they need it most. This use case is particularly relevant in the context of existing health disparities where high underdiagnosis rates for under- served subgroups are well documented (8- 11). Although bias in underdiagnosis can potentially delay access to medical treatment unequally, underdiagnosis due of AI has been relatively unexplored. In this work we examine algorithmic underdiagnosis in chest X- ray pathology classifiers and find that classifiers consistently and selectively underdiagnose under- served patients, actively amplifying the existing biases in clinical care. These effects are worse on intersectional subpopulations, e.g., Black females, and persist across three large and a multi- source chest X- ray dataset. Our work demonstrates that deploying AI systems risks exacerbating biases present in current care practices. Developers, clinical staff, and regulators must address the serious ethical concerns of -- and barriers to -- effective deployment of these models in the clinic. + +## Introduction + +As artificial intelligence (AI) algorithms increasingly affect decision making in society (12), researchers have raised concerns about algorithms creating or amplifying biases in calculators documented (2- 8). While AI algorithms in specific circumstances can potentially reduce bias (13), direct application of AI has also been shown to systemize bias in a range of settings (2- 7, 15). This tension is particularly pressing in healthcare, where AI systems could improve patient health (4), but can also exhibit biases (2- 7). Motivated by demonstrations that AI algorithms can match specialist performance in particularly in medical imaging (1) and the global radiologist shortage (16), AI- based diagnostic tools looks a clear case for deployment. + +While there is much work in algorithmic bias (14) and bias in health (2- 7, 8- 11), the topic of AI- driven underdiagnosis has been relatively unexplored. Crucially, underdiagnosis -- defined as falsely claiming the patient is healthy -- leads to no clinical treatment when a patient needs it most. The existing clinical landscape demonstrates biases in underdiagnosis against under- served subpopulations. For example, Black patients compared to non- Hispanic White with chronic obstructive pulmonary disease are more under- diagnosed (9), in clinical applications the risk score threshold is adjusted racially, which is potentially harmful for Black patients (8), or the quality of care delivered to the patients within the same hospital varies by the patient insurance type (11). Such biases can manifest algorithmically, e.g. females receiving a longer time to diagnosis than males with the same medical conditions (10). In medical imaging specifically, algorithmic bias + +<--- Page Split ---> + + +
SubgroupAttributeCXRCXPNIHALL
#images371,858223,648112,120707626
SexMale52.17%59.36%56.49%55.13%
Female47.83%40.64%43.51%44.87%
Age0-202.20%0.87%6.09%2.40%
20-4019.51%13.18%25.96%18.53%
40-6037.20%31.00%43.83%36.29%
60-8034.12%38.94%23.11%33.90%
80+6.96%16.01%1.01%8.88%
RaceAsian3.24%------
Black18.59%------
Hispanic6.41%------
Native0.29%------
White67.64%------
Other3.83%------
InsuranceMedicare46.07%------
Medicaid8.98%------
Other44.95%------
AUC0.834 ± 0.0010.805±0.0010.835 ± 0.0020.859±0.001
+ +Table 1. The summary statistics across datasets. The description of MIMIC- CXR (CXR) (17), CheXpert (CXP) (18), and Chest- Xray8 (NIH) (19) dataset as well as the multi- source dataset (ALL) composed of the aggregation of CXR, CXP, and NIH on shared labels. The reported AUCs are the averages on 14, 14, 15, and 8 labels of the CXR, CXP, NIH, and ALL dataset. + +from chest X- rays has demonstrated asymmetric diagnosis for a range of diseases (6) in males patients and female, but underdiagnosis has not been investigated. + +In this work, we perform a systematic study of underdiagnosis bias in a chest X- ray prediction model across three large public radiology datasets, MIMIC- CXR (CXR) (17), CheXpert (CXP) (18), and Chest- Xray8 (NIH) (19), as well as a multi- source dataset combining all three on shared diseases. Motivated by known differences in disease manifestation in patients by sex (6), age (20), and race (8), and the effect of insurance type in quality of received care (11), we report results across all of these factors. Also we use insurance type as an imperfect proxy of socioeconomic status - e.g. patients with Medicaid insurance are often low income. + +We find that algorithms trained on all settings exhibit systemic underdiagnosis biases in under- served subpopulations, including females, Black and Hispanic, younger patients, and patients of lower socioeconomic status (with Medicaid insurance). Further, we show that our observations are not consistent with an increase in overall noise on these subgroups, but instead are reflective of a specific increase in underdiagnosis alone. We find these effects persist for intersectional subgroups (e.g., Black female), and are not consistently worse in the smallest intersectional groups. + +## Methodology + +We train distinct chest X- ray diagnosis models in four settings: the MIMIC- CXR dataset (CXR, 371,858 images from 65,079 patients) (17), CheXpert (CXP, 223,648 images from 64,740 + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. The model pipeline. We (A) examine chest radiographs across several datasets with diverse populations, (B) learn a DenseNet from these data (training across all patients simultaneously), predicting the presence of the "No Finding" label – indicates the algorithm detect no disease for image, and (C) compare false positive rate (FPR) of this model on different subpopulations (including sex, race, age, and insurance type) to examine the algorithm's underdiagnosis rate. If the model has a higher underdiagnosis rates on certain subpopulations, such as female or Medicaid patients, this would induce significant health disparities that lead to higher rate of no clinical treatment for certain subpopulations, were the model deployed.
+ +patients) (18), Chest- Xray8 (NIH, 112,120 images from 30,805 patients) (19), and a multi- source combination of all three (ALL, 707626 images from 129,819 patients). The CXR, CXP, and NIH datasets have relatively equal rates of male and female patients, and most patients are between 40 and 80 years old. Note that the CXP and NIH datasets only report patient sex1 and age, whereas the CXR additionally reports patient race and insurance type for a large subset of images. The race and insurance type attributes are highly skewed in the dataset, where White are the majority and patients with Medicaid insurance are the minorities within the dataset. The NIH dataset has only frontal view images where other datasets have also lateral view images. For more detailed summary statistics across datasets, see Table 1. For our medical imaging predictive model and training scheme, we follow best practices (7) and train a 121- layer DenseNet (21), with weights initialized using ImageNet (22). Additional details of model training and construction can be found in Appendix Section Model Training Details. + +To assess and compare underdiagnosis rates across subpopulations, we compute the false positive rate (FPR) of model prediction for the "No Finding" label, which indicates no disease diagnosed, though patient suffer from at least a diagnostic disease condition. We then compare these FPRs across subpopulations including age and sex on all four datasets, as well as race and insurance type on the CXR dataset specifically. For a visual illustration of our model pipeline, see Fig. 1. + +Additionally, we measure the False Negative Rate (FNR) for the "No Finding" label across all subgroups. This measure is useful to help us differentiate between overall model noise (e.g., predictions are flipped at random in either direction), which would result in roughly correlated FPR and FNR rates across subgroups, and selective model noise (e.g., predictions are selectively biased towards a prediction of "No Finding"), which would result in un- or anti- correlated FPR + +<--- Page Split ---> + +and FNR rates. While both kinds of noise are problematic, the latter is a form of technical bias amplification, as it would show the known bias of clinical underdiagnosis is being selectively amplified by the algorithm- - i.e., the model is not only failing to diagnose those patients clinicians, but is also failing to diagnose other patients as well. + +## Results + +Undergadnosis occurs in under- served patient subpopulations. We find the underdiagnosis rate for all datasets differs dramatically based on all measured subpopulations. In Fig. 2A we show the subgroup- specific underdiagnosis for CXR dataset on race, sex, age, and insurance type. We observed female, patients under 20 years old, Black, Hispanic, and patients with Medicaid insurance receive higher rates of algorithmic underdiagnosis than other groups. In other words, these groups are at higher risk of being falsely flagged as "healthy", and receiving no clinical treatment. Results on other datasets follow a similar trend and they are shown in the Appendix. + +Undergadnosis occurs in intersectional groups. We investigate intersectional groups - - here defined as patients who belong to two subpopulations, e.g., Black female. Similar to prior work in face detection (15), we find that intersectional subgroups (see Fig. 2B) often have compounded biases in algorithmic underdiagnosis. For instance, in CXR dataset Hispanic females have a higher "No Finding" FPR than White females (see Fig. 2B- 1). Also, patients less than 20 years, female, Black, or patients with Medicaid insurance who are often low income has the largest underdiagnosis rates (see Fig. 2B- 2). The underdiagnosis rate for the intersection of Black patients with another subgroup of age, sex, and insurance type (see Fig. 2B- 3) and patients with Medicaid insurance with another subgroup of sex, age, and race (see Fig. 2B- 4) are also depicted in Fig. 2B. It is observable that the patients who are the member of two under- served subgroups are experiencing larger underdiagnosis rate. In another word, though female in subpopulation study of underdiagnosis rate (Fig. 2A) have shown to have a larger underdiagnosis rate, not all females are misdiagnosed at the same rate (see Fig. 2B- 1). The intersection underdiagnosis for other datasets is shown in the Appendix where they also follow a similar pattern. + +Undergadnosis is not a result of subgroup- specific overall noise. As illustrated in Fig. 2C (FNR for 'No Finding'), FPR and FNR show inverse relationships across different under- served subgroups on the CXR dataset (though this finding is consistent across all datasets, see Appendix), for both overall and intersectional subgroups (see Fig. 2D). In other word. For emphasis, we restate that this finding is not consistent with a simple increase in overall noise for specific subgroups, but instead indicates that under- served subpopulations are being aggressively flagged erroneously as healthy by the algorithm, without a corresponding increase in false negatives. + +## Conclusion + +We demonstrate evidence of AI- based underdiagnosis against under- served subpopulations in diagnostic algorithms trained on chest X- rays. Clinically, underdiagnosis is of key importance because undiagnosed patients incorrectly receive no treatment. We observe, across three large- scale datasets and a combined multi- source dataset, under- served subpopulations are consistently at significant risk of algorithmic underdiagnosis. Additionally, patients in intersectional subgroups (e.g., Black female) are particularly susceptible to algorithmic underdiagnosis. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. Analyzing underdiagnoses over subgroups of sex, age, race & insurance type within the MIMIC-CXR (CXR) dataset. The results are averaged over 5 run with different random seed ± 95% confidence interval. A. The underdiagnosis rate (measured by “No Finding” FPR). Female, 0-20, Black and/or low-income patients under Medicaid insurance have the largest underdiagnosis
+ +<--- Page Split ---> + +rate, indicating the greatest disparity. B. The intersectional underdiagnosis rates within only female (B1), ages 0- 20 (B2), Black (B3), or Medicaid (B4) patients. In these plots, we see that intersectional identities are often underdiagnosed even more heavily than the group in aggregate (e.g., Medicaid female patients are underdiagnosed more than Medicare female patients). C, D We compute the same analyses on the overall subgroups (C), and the intersectional subgroups (D1- 4) but now examining the "No Finding" FNR. If we observed a commensurate increase in FNR alongside the increase in FPR observed in A, B, this would indicate these results are tracking an increase in overall noise. Instead, we typically observe an inverse correlation between FPR and FNR, indicating the model is selectively underdiagnosing these vulnerable subpopulations. Throughout, subgroups labeled in gray text, with results omitted, indicate the subgroup has too few members ( \(\leq = 15\) ) to be used reliably. + +This discrepancy is especially interesting in the context of known biases in clinical care itself, in which under- served subpopulations are often underdiagnosed by doctors without a simultaneous increase in privileged group overdiagnosis (9). Our prediction labels are provided by these same doctors, and are therefore not an unbiased ground truth - - in other words, our labels should already suffer from this same bias that our model is then additionally exhibiting. This is a form of bias amplification, when a model's predicted outputs amplify a known source of error in the data generative process (23) or data distribution (24). This is an especially dangerous outcome for machine learning models in healthcare, as it indicates that the existing biases in health practice risk being magnified, rather than ameliorated, by algorithmic decisions based on large (707,626 images), multi- source datasets. While this evaluation is for chest x- ray diagnostic imaging, this issue is likely widespread across data sources, and prediction tasks. + +Our findings demonstrate a concrete way that algorithms escalate existing systemic health inequities. As algorithms move from the lab to the real world, we must consider the ethical concerns about the accessibility of medical treatment for under- served subpopulations and effective and ethical deployment of these models. + +## References: + +1. Pranav, R., et al., CheXNet: Radiologist-level pneumonia detection on chest x-rays with deep learning, in (CVPR, 2017). +2. Wiens, J., et al. Do no harm: a roadmap for responsible machine learning for health care. Nat Med 25, 1337–1340 (2019). +3. Char, D. S., Eisenstein, L. G. & Jones D. S., Implementing Machine Learning in Health Care — Addressing Ethical Challenges, N Engl J Med. 378(11), 981–983 (2018). +4. Chen, I. Y., Joshi, S. & Ghassemi, M., Treating health disparities with artificial intelligence, Nat Med. 26(1), 16–17 (2020). +5. Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S., Dissecting racial bias in an algorithm used to manage the health of populations, Science 366, 447–453 (2019). +6. Larrazabal, A. J., et al., Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. PNAS 117(23), 12592–12594 (2020). + +<--- Page Split ---> + +7. Seyyed-Kalantari, L., et al., CheXclusion: Fairness gaps in deep chest X-ray classifiers, in (PSB, 2021), p. 232-243. + +8. Vyas, D. A., Eisenstein, L. G., & Jones, D. S., Hidden in Plain Sight—Reconsidering the Use of Race Correction in Clinical Algorithms, N Engl J Med. 383(9), 874-882 (2020). + +9. Mamary, A. J., et al., Race and gender disparities are evident in COPD underdiagnoses across all severities of measured airflow obstruction, Chronic Obstructive Pulmonary Diseases: Journal of the COPD Foundation 5(3), 177-184 (2018). + +10. Sun, T. Y., et al., Exploring gender disparities in time to diagnosis; TY Sun, OJ Walk IV, JL Chen, HR Nieva, N Elhadad - in (ML4H at NeurIPS, 2020). + +11. Spencer, C. S., Gaskin, D. J. & Roberts, E. T., The quality of care delivered to the patients within the same hospital varies by insurance type, Health Aff (Milwood), 32(10), 1731-9 (2013). + +12. Raghavan, M., Barocas S. & Kleinberg J., Mitigating bias in algorithmic hiring: Evaluating claims and practices (FAT\*, 2020), p. 469-481. + +13. Cowgill, B., "Bias and productivity in humans and machines, Upjohn Institute Working Paper", (W. E. Upjohn Institute for Employment Research, Kalamazoo, MI, 2018). + +14. Dwork, C., et al., Fairness through awareness, in (ITCS, 2012), p. 214-226. + +15. Buolamwini, J. & Gebru, T., Gender shades: Intersectional accuracy disparities in commercial gender classification, PMLR 81, 77-91 (2018). + +16. Rimmer, A., Radiologist shortage leaves patient care at risk, BMJ: 359 (2017). + +17. Johnson, A. E. W., et al., MIMIC-CXR: A large publicly available database of labeled chest radiographs, Science Data 6, 317 (2019). + +18. Irvin, J., et al., CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison, in (CVPR, 2019). + +19. Wang, X., et al., ChestX-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases, (CVPR, 2017), p. 2097-2106. + +20. Bhatt, M. L. B., Kant, S. & Bhaskar R., Pulmonary tuberculosis as differential diagnosis of lung cancer, South Asian J. of cancer 1(1), 36-42 (2012). + +21. Iandola, F., et al. Densenet: Implementing efficient ConvNet descriptor pyramids, in (CVPR, 2014). + +22. Russakovsky, O., et al., Imagenet large scale visual recognition challenge, IJCV, 115(3) 211-252, (2015). + +23. Oakden, L., Dunnmon, J., Carneiro, G. & Re, C., Hidden stratification causes clinically meaningful failures in machine learning for medical imaging, in (CHIL, 2020), p. 151-159. + +24. Zhao, J., et al., Men also like shopping: Reducing gender bias amplification using corpus-level constraints, (EMNLP, 2017), p. 2979-2989. + +<--- Page Split ---> + +Acknowledgments: we acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) - funding number PDF- 516984, Microsoft Research, Canadian Institute for Advanced Research (CIFAR,) NSERC Discovery Grant; Author contributions: Each named author has substantially contributed to conducting the underlying research and drafting the manuscript.; Competing interests: Authors declare no competing interests.; and Data and materials availability: All 3 datasets that we have used for this work are public under data use agreements. MIMIC- CXR dataset available at: https://physionet.org/content/mimic- cxr/2.0.0/ CheXpert dataset is available at: https://stanfordmlgroup.github.io/competitions/chexpert/ ChestX- ray8 dataset is available at: https://www.nih.gov/news- events/news- releases/nih- clinical- center- provides- one- largest- publicly- available- chest- x- ray- datasets- scientific- community. Code availability All code used in the analysis will be available in a public repository for purposes of reproducing or extending the analysis. The link to the code will be added to the text of the paper for the camera ready version. + +Supplementary Materials: Model Training Details Additional Results Figs. S1 to S3 + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +The model pipeline. We (A) examine chest radiographs across several datasets with diverse populations, (B) learn a DenseNet from these data (training across all patients simultaneously), predicting the presence of the "No Finding" label – indicates the algorithm detect no disease for image, and (C) compare false positive rate (FPR) of this model on different subpopulations (including sex, race, age, and insurance type) to examine the algorithm's underdiagnosis rate. If the model has a higher underdiagnosis rates on certain subpopulations, such as female or Medicaid patients, this would induce significant health disparities that lead to higher rate of no clinical treatment for certain subpopulations, were the model deployed. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Analyzing underdiagnoses over subgroups of sex, age, race & insurance type within the MIMIC- CXR (CXR) dataset. The results are averaged over run with different random seed \(\pm 95\%\) confidence interval. A. The underdiagnosis rate (measured by "No Finding" FPR). Female, 0- 20, Black and/or low- income patients under Medicaid insurance have the largest underdiagnosis rate, indicating the greatest disparity. B. The intersectional underdiagnosis rates within only female (B1), ages 0- 20 (B2), Black (B3), or Medicaid (B4) + +<--- Page Split ---> + +patients. In these plots, we see that intersectional identities are often underdiagnosed even more heavily than the group in aggregate (e.g., Medicaid female patients are underdiagnosed more than Medicare female patients). C, D We compute the same analyses on the overall subgroups (C), and the intersectional subgroups (D1- 4) 5 but now examining the "No Finding" FNR. If we observed a commensurate increase in FNR alongside the increase in FPR observed in A, B, this would indicate these results are tracking an increase in overall noise. Instead, we typically observe an inverse correlation between FPR and FNR, indicating the model is selectively underdiagnosing these vulnerable subpopulations. Throughout, subgroups labeled in gray text, with results omitted, indicate the subgroup has too few members (<= 15) to be used reliably. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- UndeSup.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4_det.mmd b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9c4ec60a224f93936e6d723218ba3f90074c8d0c --- /dev/null +++ b/preprint/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4/preprint__12124df0a77be69e70f09415b91e907640e4c22018e976b4ef8ec3ee133430e4_det.mmd @@ -0,0 +1,247 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 902, 175]]<|/det|> +# Medical imaging algorithms exacerbate biases in underdiagnosis + +<|ref|>text<|/ref|><|det|>[[44, 195, 736, 238]]<|/det|> +Laleh Seyyed- Kalantari ( \(\boxed{ \begin{array}{r l} \end{array} }\) laleh@cs.toronto.edu ) University of Toronto, Vector Institute https://orcid.org/0000- 0002- 1059- 7125 + +<|ref|>text<|/ref|><|det|>[[44, 243, 380, 283]]<|/det|> +Guanxiong Liu University of Toronto, Vector Institute + +<|ref|>text<|/ref|><|det|>[[44, 290, 395, 330]]<|/det|> +Matthew McDermott Massachusetts Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 336, 395, 376]]<|/det|> +Irene Chen Massachusetts Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 382, 594, 423]]<|/det|> +Marzyeh Ghassemi University of Toronto https://orcid.org/0000- 0001- 6349- 7251 + +<|ref|>sub_title<|/ref|><|det|>[[44, 464, 96, 481]]<|/det|> +## Letter + +<|ref|>text<|/ref|><|det|>[[44, 501, 650, 521]]<|/det|> +Keywords: artificial intelligence (AI), machine learning, underdiagnosis + +<|ref|>text<|/ref|><|det|>[[44, 540, 322, 559]]<|/det|> +Posted Date: February 4th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 578, 463, 597]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 151985/v1 + +<|ref|>text<|/ref|><|det|>[[42, 615, 910, 657]]<|/det|> +License: \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 694, 936, 737]]<|/det|> +Version of Record: A version of this preprint was published at Nature Medicine on December 10th, 2021. See the published version at https://doi.org/10.1038/s41591- 021- 01595- 0. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[180, 98, 813, 119]]<|/det|> +# Medical imaging algorithms exacerbate biases in underdiagnosis + +<|ref|>text<|/ref|><|det|>[[123, 124, 872, 160]]<|/det|> +Laleh Seyyed- Kalantari \(^{1,2*}\) , Guanxiong Liu \(^{1,2}\) , Matthew McDermott \(^{3}\) , Irene Y. Chen \(^{3}\) , Marzyeh Ghassemi \(^{1,2}\) + +<|ref|>text<|/ref|><|det|>[[115, 181, 300, 199]]<|/det|> +\(^{1}\) University of Toronto. + +<|ref|>text<|/ref|><|det|>[[115, 207, 250, 224]]<|/det|> +\(^{2}\) Vector Institute. + +<|ref|>text<|/ref|><|det|>[[115, 232, 430, 250]]<|/det|> +\(^{3}\) Massachusetts Institute of Technology. + +<|ref|>text<|/ref|><|det|>[[115, 258, 456, 274]]<|/det|> +\*Correspondence to: laleh@cs.toronto.edu. + +<|ref|>text<|/ref|><|det|>[[113, 275, 882, 536]]<|/det|> +Artificial intelligence (AI) systems have increasingly achieved expert- level performance, particularly in medical imaging (1). However, there is growing concern that AI systems will reflect and amplify human bias against under- served subpopulations (2- 7). Such biases are especially troubling in the context of underdiagnosis: if AI systems falsely predict that patients are healthy, patients would be denied care when they need it most. This use case is particularly relevant in the context of existing health disparities where high underdiagnosis rates for under- served subgroups are well documented (8- 11). Although bias in underdiagnosis can potentially delay access to medical treatment unequally, underdiagnosis due of AI has been relatively unexplored. In this work we examine algorithmic underdiagnosis in chest X- ray pathology classifiers and find that classifiers consistently and selectively underdiagnose under- served patients, actively amplifying the existing biases in clinical care. These effects are worse on intersectional subpopulations, e.g., Black females, and persist across three large and a multi- source chest X- ray dataset. Our work demonstrates that deploying AI systems risks exacerbating biases present in current care practices. Developers, clinical staff, and regulators must address the serious ethical concerns of -- and barriers to -- effective deployment of these models in the clinic. + +<|ref|>sub_title<|/ref|><|det|>[[115, 555, 224, 570]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[115, 578, 882, 718]]<|/det|> +As artificial intelligence (AI) algorithms increasingly affect decision making in society (12), researchers have raised concerns about algorithms creating or amplifying biases in calculators documented (2- 8). While AI algorithms in specific circumstances can potentially reduce bias (13), direct application of AI has also been shown to systemize bias in a range of settings (2- 7, 15). This tension is particularly pressing in healthcare, where AI systems could improve patient health (4), but can also exhibit biases (2- 7). Motivated by demonstrations that AI algorithms can match specialist performance in particularly in medical imaging (1) and the global radiologist shortage (16), AI- based diagnostic tools looks a clear case for deployment. + +<|ref|>text<|/ref|><|det|>[[113, 725, 882, 902]]<|/det|> +While there is much work in algorithmic bias (14) and bias in health (2- 7, 8- 11), the topic of AI- driven underdiagnosis has been relatively unexplored. Crucially, underdiagnosis -- defined as falsely claiming the patient is healthy -- leads to no clinical treatment when a patient needs it most. The existing clinical landscape demonstrates biases in underdiagnosis against under- served subpopulations. For example, Black patients compared to non- Hispanic White with chronic obstructive pulmonary disease are more under- diagnosed (9), in clinical applications the risk score threshold is adjusted racially, which is potentially harmful for Black patients (8), or the quality of care delivered to the patients within the same hospital varies by the patient insurance type (11). Such biases can manifest algorithmically, e.g. females receiving a longer time to diagnosis than males with the same medical conditions (10). In medical imaging specifically, algorithmic bias + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[113, 99, 881, 411]]<|/det|> + +
SubgroupAttributeCXRCXPNIHALL
#images371,858223,648112,120707626
SexMale52.17%59.36%56.49%55.13%
Female47.83%40.64%43.51%44.87%
Age0-202.20%0.87%6.09%2.40%
20-4019.51%13.18%25.96%18.53%
40-6037.20%31.00%43.83%36.29%
60-8034.12%38.94%23.11%33.90%
80+6.96%16.01%1.01%8.88%
RaceAsian3.24%------
Black18.59%------
Hispanic6.41%------
Native0.29%------
White67.64%------
Other3.83%------
InsuranceMedicare46.07%------
Medicaid8.98%------
Other44.95%------
AUC0.834 ± 0.0010.805±0.0010.835 ± 0.0020.859±0.001
+ +<|ref|>text<|/ref|><|det|>[[115, 425, 882, 496]]<|/det|> +Table 1. The summary statistics across datasets. The description of MIMIC- CXR (CXR) (17), CheXpert (CXP) (18), and Chest- Xray8 (NIH) (19) dataset as well as the multi- source dataset (ALL) composed of the aggregation of CXR, CXP, and NIH on shared labels. The reported AUCs are the averages on 14, 14, 15, and 8 labels of the CXR, CXP, NIH, and ALL dataset. + +<|ref|>text<|/ref|><|det|>[[115, 509, 882, 545]]<|/det|> +from chest X- rays has demonstrated asymmetric diagnosis for a range of diseases (6) in males patients and female, but underdiagnosis has not been investigated. + +<|ref|>text<|/ref|><|det|>[[115, 551, 882, 675]]<|/det|> +In this work, we perform a systematic study of underdiagnosis bias in a chest X- ray prediction model across three large public radiology datasets, MIMIC- CXR (CXR) (17), CheXpert (CXP) (18), and Chest- Xray8 (NIH) (19), as well as a multi- source dataset combining all three on shared diseases. Motivated by known differences in disease manifestation in patients by sex (6), age (20), and race (8), and the effect of insurance type in quality of received care (11), we report results across all of these factors. Also we use insurance type as an imperfect proxy of socioeconomic status - e.g. patients with Medicaid insurance are often low income. + +<|ref|>text<|/ref|><|det|>[[115, 682, 882, 805]]<|/det|> +We find that algorithms trained on all settings exhibit systemic underdiagnosis biases in under- served subpopulations, including females, Black and Hispanic, younger patients, and patients of lower socioeconomic status (with Medicaid insurance). Further, we show that our observations are not consistent with an increase in overall noise on these subgroups, but instead are reflective of a specific increase in underdiagnosis alone. We find these effects persist for intersectional subgroups (e.g., Black female), and are not consistently worse in the smallest intersectional groups. + +<|ref|>sub_title<|/ref|><|det|>[[115, 837, 226, 853]]<|/det|> +## Methodology + +<|ref|>text<|/ref|><|det|>[[115, 853, 881, 889]]<|/det|> +We train distinct chest X- ray diagnosis models in four settings: the MIMIC- CXR dataset (CXR, 371,858 images from 65,079 patients) (17), CheXpert (CXP, 223,648 images from 64,740 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[130, 92, 884, 277]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 286, 883, 427]]<|/det|> +
Fig. 1. The model pipeline. We (A) examine chest radiographs across several datasets with diverse populations, (B) learn a DenseNet from these data (training across all patients simultaneously), predicting the presence of the "No Finding" label – indicates the algorithm detect no disease for image, and (C) compare false positive rate (FPR) of this model on different subpopulations (including sex, race, age, and insurance type) to examine the algorithm's underdiagnosis rate. If the model has a higher underdiagnosis rates on certain subpopulations, such as female or Medicaid patients, this would induce significant health disparities that lead to higher rate of no clinical treatment for certain subpopulations, were the model deployed.
+ +<|ref|>text<|/ref|><|det|>[[113, 440, 883, 650]]<|/det|> +patients) (18), Chest- Xray8 (NIH, 112,120 images from 30,805 patients) (19), and a multi- source combination of all three (ALL, 707626 images from 129,819 patients). The CXR, CXP, and NIH datasets have relatively equal rates of male and female patients, and most patients are between 40 and 80 years old. Note that the CXP and NIH datasets only report patient sex1 and age, whereas the CXR additionally reports patient race and insurance type for a large subset of images. The race and insurance type attributes are highly skewed in the dataset, where White are the majority and patients with Medicaid insurance are the minorities within the dataset. The NIH dataset has only frontal view images where other datasets have also lateral view images. For more detailed summary statistics across datasets, see Table 1. For our medical imaging predictive model and training scheme, we follow best practices (7) and train a 121- layer DenseNet (21), with weights initialized using ImageNet (22). Additional details of model training and construction can be found in Appendix Section Model Training Details. + +<|ref|>text<|/ref|><|det|>[[113, 650, 882, 756]]<|/det|> +To assess and compare underdiagnosis rates across subpopulations, we compute the false positive rate (FPR) of model prediction for the "No Finding" label, which indicates no disease diagnosed, though patient suffer from at least a diagnostic disease condition. We then compare these FPRs across subpopulations including age and sex on all four datasets, as well as race and insurance type on the CXR dataset specifically. For a visual illustration of our model pipeline, see Fig. 1. + +<|ref|>text<|/ref|><|det|>[[113, 761, 882, 850]]<|/det|> +Additionally, we measure the False Negative Rate (FNR) for the "No Finding" label across all subgroups. This measure is useful to help us differentiate between overall model noise (e.g., predictions are flipped at random in either direction), which would result in roughly correlated FPR and FNR rates across subgroups, and selective model noise (e.g., predictions are selectively biased towards a prediction of "No Finding"), which would result in un- or anti- correlated FPR + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 882, 159]]<|/det|> +and FNR rates. While both kinds of noise are problematic, the latter is a form of technical bias amplification, as it would show the known bias of clinical underdiagnosis is being selectively amplified by the algorithm- - i.e., the model is not only failing to diagnose those patients clinicians, but is also failing to diagnose other patients as well. + +<|ref|>sub_title<|/ref|><|det|>[[115, 192, 178, 208]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[115, 216, 882, 340]]<|/det|> +Undergadnosis occurs in under- served patient subpopulations. We find the underdiagnosis rate for all datasets differs dramatically based on all measured subpopulations. In Fig. 2A we show the subgroup- specific underdiagnosis for CXR dataset on race, sex, age, and insurance type. We observed female, patients under 20 years old, Black, Hispanic, and patients with Medicaid insurance receive higher rates of algorithmic underdiagnosis than other groups. In other words, these groups are at higher risk of being falsely flagged as "healthy", and receiving no clinical treatment. Results on other datasets follow a similar trend and they are shown in the Appendix. + +<|ref|>text<|/ref|><|det|>[[115, 346, 882, 590]]<|/det|> +Undergadnosis occurs in intersectional groups. We investigate intersectional groups - - here defined as patients who belong to two subpopulations, e.g., Black female. Similar to prior work in face detection (15), we find that intersectional subgroups (see Fig. 2B) often have compounded biases in algorithmic underdiagnosis. For instance, in CXR dataset Hispanic females have a higher "No Finding" FPR than White females (see Fig. 2B- 1). Also, patients less than 20 years, female, Black, or patients with Medicaid insurance who are often low income has the largest underdiagnosis rates (see Fig. 2B- 2). The underdiagnosis rate for the intersection of Black patients with another subgroup of age, sex, and insurance type (see Fig. 2B- 3) and patients with Medicaid insurance with another subgroup of sex, age, and race (see Fig. 2B- 4) are also depicted in Fig. 2B. It is observable that the patients who are the member of two under- served subgroups are experiencing larger underdiagnosis rate. In another word, though female in subpopulation study of underdiagnosis rate (Fig. 2A) have shown to have a larger underdiagnosis rate, not all females are misdiagnosed at the same rate (see Fig. 2B- 1). The intersection underdiagnosis for other datasets is shown in the Appendix where they also follow a similar pattern. + +<|ref|>text<|/ref|><|det|>[[115, 597, 882, 720]]<|/det|> +Undergadnosis is not a result of subgroup- specific overall noise. As illustrated in Fig. 2C (FNR for 'No Finding'), FPR and FNR show inverse relationships across different under- served subgroups on the CXR dataset (though this finding is consistent across all datasets, see Appendix), for both overall and intersectional subgroups (see Fig. 2D). In other word. For emphasis, we restate that this finding is not consistent with a simple increase in overall noise for specific subgroups, but instead indicates that under- served subpopulations are being aggressively flagged erroneously as healthy by the algorithm, without a corresponding increase in false negatives. + +<|ref|>sub_title<|/ref|><|det|>[[115, 752, 210, 768]]<|/det|> +## Conclusion + +<|ref|>text<|/ref|><|det|>[[115, 776, 882, 881]]<|/det|> +We demonstrate evidence of AI- based underdiagnosis against under- served subpopulations in diagnostic algorithms trained on chest X- rays. Clinically, underdiagnosis is of key importance because undiagnosed patients incorrectly receive no treatment. We observe, across three large- scale datasets and a combined multi- source dataset, under- served subpopulations are consistently at significant risk of algorithmic underdiagnosis. Additionally, patients in intersectional subgroups (e.g., Black female) are particularly susceptible to algorithmic underdiagnosis. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 87, 870, 830]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 835, 883, 907]]<|/det|> +
Fig. 2. Analyzing underdiagnoses over subgroups of sex, age, race & insurance type within the MIMIC-CXR (CXR) dataset. The results are averaged over 5 run with different random seed ± 95% confidence interval. A. The underdiagnosis rate (measured by “No Finding” FPR). Female, 0-20, Black and/or low-income patients under Medicaid insurance have the largest underdiagnosis
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 882, 281]]<|/det|> +rate, indicating the greatest disparity. B. The intersectional underdiagnosis rates within only female (B1), ages 0- 20 (B2), Black (B3), or Medicaid (B4) patients. In these plots, we see that intersectional identities are often underdiagnosed even more heavily than the group in aggregate (e.g., Medicaid female patients are underdiagnosed more than Medicare female patients). C, D We compute the same analyses on the overall subgroups (C), and the intersectional subgroups (D1- 4) but now examining the "No Finding" FNR. If we observed a commensurate increase in FNR alongside the increase in FPR observed in A, B, this would indicate these results are tracking an increase in overall noise. Instead, we typically observe an inverse correlation between FPR and FNR, indicating the model is selectively underdiagnosing these vulnerable subpopulations. Throughout, subgroups labeled in gray text, with results omitted, indicate the subgroup has too few members ( \(\leq = 15\) ) to be used reliably. + +<|ref|>text<|/ref|><|det|>[[113, 313, 882, 505]]<|/det|> +This discrepancy is especially interesting in the context of known biases in clinical care itself, in which under- served subpopulations are often underdiagnosed by doctors without a simultaneous increase in privileged group overdiagnosis (9). Our prediction labels are provided by these same doctors, and are therefore not an unbiased ground truth - - in other words, our labels should already suffer from this same bias that our model is then additionally exhibiting. This is a form of bias amplification, when a model's predicted outputs amplify a known source of error in the data generative process (23) or data distribution (24). This is an especially dangerous outcome for machine learning models in healthcare, as it indicates that the existing biases in health practice risk being magnified, rather than ameliorated, by algorithmic decisions based on large (707,626 images), multi- source datasets. While this evaluation is for chest x- ray diagnostic imaging, this issue is likely widespread across data sources, and prediction tasks. + +<|ref|>text<|/ref|><|det|>[[114, 513, 882, 582]]<|/det|> +Our findings demonstrate a concrete way that algorithms escalate existing systemic health inequities. As algorithms move from the lab to the real world, we must consider the ethical concerns about the accessibility of medical treatment for under- served subpopulations and effective and ethical deployment of these models. + +<|ref|>sub_title<|/ref|><|det|>[[114, 615, 214, 631]]<|/det|> +## References: + +<|ref|>text<|/ref|><|det|>[[111, 639, 884, 888]]<|/det|> +1. Pranav, R., et al., CheXNet: Radiologist-level pneumonia detection on chest x-rays with deep learning, in (CVPR, 2017). +2. Wiens, J., et al. Do no harm: a roadmap for responsible machine learning for health care. Nat Med 25, 1337–1340 (2019). +3. Char, D. S., Eisenstein, L. G. & Jones D. S., Implementing Machine Learning in Health Care — Addressing Ethical Challenges, N Engl J Med. 378(11), 981–983 (2018). +4. Chen, I. Y., Joshi, S. & Ghassemi, M., Treating health disparities with artificial intelligence, Nat Med. 26(1), 16–17 (2020). +5. Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S., Dissecting racial bias in an algorithm used to manage the health of populations, Science 366, 447–453 (2019). +6. Larrazabal, A. J., et al., Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. PNAS 117(23), 12592–12594 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 884, 127]]<|/det|> +7. Seyyed-Kalantari, L., et al., CheXclusion: Fairness gaps in deep chest X-ray classifiers, in (PSB, 2021), p. 232-243. + +<|ref|>text<|/ref|><|det|>[[111, 132, 884, 169]]<|/det|> +8. Vyas, D. A., Eisenstein, L. G., & Jones, D. S., Hidden in Plain Sight—Reconsidering the Use of Race Correction in Clinical Algorithms, N Engl J Med. 383(9), 874-882 (2020). + +<|ref|>text<|/ref|><|det|>[[111, 175, 883, 228]]<|/det|> +9. Mamary, A. J., et al., Race and gender disparities are evident in COPD underdiagnoses across all severities of measured airflow obstruction, Chronic Obstructive Pulmonary Diseases: Journal of the COPD Foundation 5(3), 177-184 (2018). + +<|ref|>text<|/ref|><|det|>[[111, 233, 883, 270]]<|/det|> +10. Sun, T. Y., et al., Exploring gender disparities in time to diagnosis; TY Sun, OJ Walk IV, JL Chen, HR Nieva, N Elhadad - in (ML4H at NeurIPS, 2020). + +<|ref|>text<|/ref|><|det|>[[111, 276, 883, 329]]<|/det|> +11. Spencer, C. S., Gaskin, D. J. & Roberts, E. T., The quality of care delivered to the patients within the same hospital varies by insurance type, Health Aff (Milwood), 32(10), 1731-9 (2013). + +<|ref|>text<|/ref|><|det|>[[111, 335, 883, 372]]<|/det|> +12. Raghavan, M., Barocas S. & Kleinberg J., Mitigating bias in algorithmic hiring: Evaluating claims and practices (FAT\*, 2020), p. 469-481. + +<|ref|>text<|/ref|><|det|>[[111, 377, 883, 415]]<|/det|> +13. Cowgill, B., "Bias and productivity in humans and machines, Upjohn Institute Working Paper", (W. E. Upjohn Institute for Employment Research, Kalamazoo, MI, 2018). + +<|ref|>text<|/ref|><|det|>[[111, 420, 741, 439]]<|/det|> +14. Dwork, C., et al., Fairness through awareness, in (ITCS, 2012), p. 214-226. + +<|ref|>text<|/ref|><|det|>[[111, 445, 883, 482]]<|/det|> +15. Buolamwini, J. & Gebru, T., Gender shades: Intersectional accuracy disparities in commercial gender classification, PMLR 81, 77-91 (2018). + +<|ref|>text<|/ref|><|det|>[[111, 488, 768, 507]]<|/det|> +16. Rimmer, A., Radiologist shortage leaves patient care at risk, BMJ: 359 (2017). + +<|ref|>text<|/ref|><|det|>[[111, 512, 883, 549]]<|/det|> +17. Johnson, A. E. W., et al., MIMIC-CXR: A large publicly available database of labeled chest radiographs, Science Data 6, 317 (2019). + +<|ref|>text<|/ref|><|det|>[[111, 555, 883, 591]]<|/det|> +18. Irvin, J., et al., CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison, in (CVPR, 2019). + +<|ref|>text<|/ref|><|det|>[[111, 597, 883, 650]]<|/det|> +19. Wang, X., et al., ChestX-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases, (CVPR, 2017), p. 2097-2106. + +<|ref|>text<|/ref|><|det|>[[111, 656, 883, 692]]<|/det|> +20. Bhatt, M. L. B., Kant, S. & Bhaskar R., Pulmonary tuberculosis as differential diagnosis of lung cancer, South Asian J. of cancer 1(1), 36-42 (2012). + +<|ref|>text<|/ref|><|det|>[[111, 698, 883, 735]]<|/det|> +21. Iandola, F., et al. Densenet: Implementing efficient ConvNet descriptor pyramids, in (CVPR, 2014). + +<|ref|>text<|/ref|><|det|>[[111, 741, 883, 778]]<|/det|> +22. Russakovsky, O., et al., Imagenet large scale visual recognition challenge, IJCV, 115(3) 211-252, (2015). + +<|ref|>text<|/ref|><|det|>[[111, 784, 883, 820]]<|/det|> +23. Oakden, L., Dunnmon, J., Carneiro, G. & Re, C., Hidden stratification causes clinically meaningful failures in machine learning for medical imaging, in (CHIL, 2020), p. 151-159. + +<|ref|>text<|/ref|><|det|>[[111, 826, 883, 862]]<|/det|> +24. Zhao, J., et al., Men also like shopping: Reducing gender bias amplification using corpus-level constraints, (EMNLP, 2017), p. 2979-2989. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 883, 316]]<|/det|> +Acknowledgments: we acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) - funding number PDF- 516984, Microsoft Research, Canadian Institute for Advanced Research (CIFAR,) NSERC Discovery Grant; Author contributions: Each named author has substantially contributed to conducting the underlying research and drafting the manuscript.; Competing interests: Authors declare no competing interests.; and Data and materials availability: All 3 datasets that we have used for this work are public under data use agreements. MIMIC- CXR dataset available at: https://physionet.org/content/mimic- cxr/2.0.0/ CheXpert dataset is available at: https://stanfordmlgroup.github.io/competitions/chexpert/ ChestX- ray8 dataset is available at: https://www.nih.gov/news- events/news- releases/nih- clinical- center- provides- one- largest- publicly- available- chest- x- ray- datasets- scientific- community. Code availability All code used in the analysis will be available in a public repository for purposes of reproducing or extending the analysis. The link to the code will be added to the text of the paper for the camera ready version. + +<|ref|>text<|/ref|><|det|>[[113, 331, 338, 400]]<|/det|> +Supplementary Materials: Model Training Details Additional Results Figs. S1 to S3 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[66, 102, 930, 310]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 339, 115, 358]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[40, 380, 955, 562]]<|/det|> +The model pipeline. We (A) examine chest radiographs across several datasets with diverse populations, (B) learn a DenseNet from these data (training across all patients simultaneously), predicting the presence of the "No Finding" label – indicates the algorithm detect no disease for image, and (C) compare false positive rate (FPR) of this model on different subpopulations (including sex, race, age, and insurance type) to examine the algorithm's underdiagnosis rate. If the model has a higher underdiagnosis rates on certain subpopulations, such as female or Medicaid patients, this would induce significant health disparities that lead to higher rate of no clinical treatment for certain subpopulations, were the model deployed. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 37, 796, 789]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 802, 117, 821]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[40, 843, 950, 956]]<|/det|> +Analyzing underdiagnoses over subgroups of sex, age, race & insurance type within the MIMIC- CXR (CXR) dataset. The results are averaged over run with different random seed \(\pm 95\%\) confidence interval. A. The underdiagnosis rate (measured by "No Finding" FPR). Female, 0- 20, Black and/or low- income patients under Medicaid insurance have the largest underdiagnosis rate, indicating the greatest disparity. B. The intersectional underdiagnosis rates within only female (B1), ages 0- 20 (B2), Black (B3), or Medicaid (B4) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 958, 248]]<|/det|> +patients. In these plots, we see that intersectional identities are often underdiagnosed even more heavily than the group in aggregate (e.g., Medicaid female patients are underdiagnosed more than Medicare female patients). C, D We compute the same analyses on the overall subgroups (C), and the intersectional subgroups (D1- 4) 5 but now examining the "No Finding" FNR. If we observed a commensurate increase in FNR alongside the increase in FPR observed in A, B, this would indicate these results are tracking an increase in overall noise. Instead, we typically observe an inverse correlation between FPR and FNR, indicating the model is selectively underdiagnosing these vulnerable subpopulations. Throughout, subgroups labeled in gray text, with results omitted, indicate the subgroup has too few members (<= 15) to be used reliably. + +<|ref|>sub_title<|/ref|><|det|>[[44, 270, 311, 297]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 320, 765, 340]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 358, 199, 377]]<|/det|> +- UndeSup.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/images_list.json b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..1de8b520ce74f14e7eeb9d3c89de71788eec5ee3 --- /dev/null +++ b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: The neural network architecture of PepNet for de novo peptide sequencing.", + "footnote": [], + "bbox": [ + [ + 140, + 90, + 863, + 345 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Comparison of the performance of PepNet, PointNovo and DeepNovo on the charge \\(2+\\) (a) and charge \\(3+\\) (b) spectra in the human proteomics dataset. Here, the Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses.", + "footnote": [], + "bbox": [ + [ + 156, + 90, + 842, + 220 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: The precision-recall (PR) curves of the peptides sequenced by PepNet, PointNovo, and DeepNovo on the spectra of charge \\(2+\\) (a) and charge \\(3+\\) (b) in the human proteomics dataset. The two dotted lines represent the precision of 0.99 and 0.95, respectively.", + "footnote": [], + "bbox": [ + [ + 156, + 310, + 842, + 455 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: The performance of PepNet, PointNovo, and DeepNovo on the peptides with different length for the spectra of charge 2+ (top) and charge 3+ (bottom) in the human proteomics dataset. Here, filtered peptide accuracy indicates the peptide-level accuracies of the sequenced peptides after removing those with un-matched precursor masses.", + "footnote": [], + "bbox": [ + [ + 172, + 91, + 822, + 308 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Performance of PepNet, PointNovo and DeepNovo on the spectra of charge 2+ (a) and charge 3+ (b) in the proteomics datasets acquired from different species.", + "footnote": [], + "bbox": [ + [ + 135, + 88, + 861, + 630 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: Comparison of de novo sequencing results of PepNet (the pulled out parts) with database searching (identified by MaxQuant) on spectra of (a) charge 2+, and (b) charge 3+. Here we only retain de novo sequencing results that with matched precursor mass and confidence score \\(>0.5\\) , which further searched against the Uniprot human database.", + "footnote": [], + "bbox": [ + [ + 118, + 198, + 880, + 352 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7: The sequencing accuracy (left) and the PR curve (right) of PepNet, PointNovo, and DeepNovo-DIA on a dataset of DIA-derived MS/MS spectra. The Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses. The two dotted lines in the PR curve represent the precision of 0.99 and 0.95, respectively.", + "footnote": [], + "bbox": [ + [ + 139, + 90, + 861, + 243 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8: The positional and peptide-level accuracy of PepNet on the input \\(2+\\) (a) and \\(3+\\) (a) spectra in the testing dataset, on which different number of most intensive peaks are retained. The error bar represent the \\(99\\%\\) confidence interval.", + "footnote": [], + "bbox": [ + [ + 137, + 583, + 860, + 732 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9: The positional accuracy at each index of the peptides, for sequenced peptide length from 6 to 20. (a) for \\(2+\\) spectra. (b) for \\(3+\\) spectra.", + "footnote": [], + "bbox": [ + [ + 135, + 198, + 861, + 345 + ] + ], + "page_idx": 10 + } +] \ No newline at end of file diff --git a/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef.mmd b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9a837ebd7f7c00caec9f1021d5fbbcb8a1d92f81 --- /dev/null +++ b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef.mmd @@ -0,0 +1,327 @@ + +# PepNet: A Fully Convolutional Neural Network for De novo Peptide Sequencing + +Kaiyuan Liu Indiana University Bloomington https://orcid.org/0000- 0002- 3404- 2802 + +Yuzhen Ye School of Informatics and Computing, Indiana University Bloomington Haixu Tang ( \(\circleddash\) hatang@indiana.edu) Indiana University Bloomington + +## Methods Article + +Keywords: + +Posted Date: February 17th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1341615/v1 + +License: © \(\circleddash\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# PepNet: A Fully Convolutional Neural Network for De novo Peptide Sequencing + +Kaiyuan Liu, Yuzhen Ye, and Haixu Tang\* + +Luddy School of Informatics, Computing, and Engineering, Indiana University, Bloomington, Indiana 47405, United States + +E- mail: hatang@indiana.edu + +## Abstract + +The de novo peptide sequencing, which does not rely on a comprehensive target sequence database, provided us a way to identify novel peptides from tandem mass (MS/MS) spectra. However, current de novo sequencing algorithms suffer from lower accuracy and coverage, which hinders their applications in proteomics. In this paper, we present PepNet, a fully convolutional neural network (CNN) for high accuracy de novo peptide sequencing. It takes an MS/MS spectrum (represented as a high dimensional vector) as input, and outputs the optimal peptide sequence along with its confidence score. Our model was trained using a total of 30 million high- energy collisional dissociation (HCD) MS/MS spectra from multiple human peptide spectral libraries. The evaluation results show that PepNet significantly outperformed currently best- performing de novo sequencing algorithms (e.g. PointNovo and DeepNovo) at both peptide level accuracy and positional level accuracy. In addition, PepNet can sequence a large fraction of spectra that were not identified by database search engines, and thus could be used as a complementary tool of database search engines for peptide identification in proteomics. + +## 15 Introduction + +The past decade has witnessed the great advances of mass spectrometry techniques, in particular liquid chromatography coupled tandem mass spectrometry (LC- MS/MS). With enhanced throughput and sensitivity, LC- MS/MS has become one of the most commonly used approaches to functional studies of proteins at the whole proteome scale across various physiological (e.g., diseases) conditions in higher organisms including human. + +In a typical proteomics experiment, after MS/MS spectra are acquired, the first and arguably most important step is to identify the peptides from these spectra. Numerous algorithms have been developed to + +<--- Page Split ---> + +address this problem, which mostly falls into three categories: protein database searching, spectral library searching, and de novo sequencing. Pioneered the peptide sequence tag method \(^{1}\) and the Sequest algorithm, \(^{2}\) protein database searching is the predominant approach used for peptide identification. Till now many peptide searching engines have been developed, including commonly used tools such as Mascot, \(^{3}\) X!Tandem \(^{4}\) OMSSA, \(^{5}\) MyriMatch, \(^{6}\) Protein Prospector \(^{7,8}\) and MSGF \(^{+}\) . \(^{9}\) Those methods compare the experimental spectra with the theoretical spectra generated from the peptides in a protein database and report those likely true peptide- spectrum matches (PSMs). + +In contrast, the spectral library search approach compares newly acquired MS/MS spectra against a library containing previously characterized experimental spectra that were used in early computational analysis. \(^{10,11}\) Thanks to the improved repeatability and reproducibility of MS/MS data as well as the increasing availability of massive experimental spectra (e.g., from the proteomics data repository \(^{12}\) and large- scale synthetic peptides projects \(^{13}\) ), the spectral library search approach has become more increasingly adopted and is implemented in software tools such as X!hunter, \(^{14}\) SpectraST \(^{15}\) and msSLASH. \(^{16}\) + +Finally, de novo sequencing algorithms attempt to derive a peptide sequence directly from its MS/MS spectrum without using references such as a spectral library or a protein sequence database. \(^{17}\) Many de novo sequencing algorithms adopted a graph theoretical formulation to compute the longest path in the spectrum graph by employing a dynamic programming algorithm \(^{18,19}\) and adaptive scoring schemes. \(^{20- 22}\) With the advancement of high- resolution MS instruments, the performance of de novo sequencing algorithms improves significantly, \(^{23,24}\) in particular with more sophisticated scoring schemes. More recently, DeepNovo \(^{25- 27}\) and its successor PointNovo \(^{28}\) was developed using deep learning algorithms, which automatically learn the fragment ion patterns relevant to peptide sequences from massive MS/MS spectra of peptides and reported improved performance. + +## 2 Results + +## 2.1 Accurate HCD-MS/MS spectra de novo sequencing by deep learning + +We present a deep learning algorithm, PepNet, that directly outputs the peptide sequence from a given HCD- MS/MS spectra with high accuracy. As depicted in Figure 1, the input for our model is a 20,000 dimensional vector representation of the spectrum (for details see the Method section). The input vector will go through six continuing temporal convolutional network (TCN) blocks \(^{29}\) and down- sampling layers to capture the relationships between observed peaks, as depicted in the TCN branch of Fig. 1. Apparently, those TCN blocks working on different resolution levels. Thus we add a merging branch (top- down branch + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: The neural network architecture of PepNet for de novo peptide sequencing.
+ +1 in Fig. 1) that fuse the global and local information into a single feature tensor. Finally, the feature tensor is then converted by a softmax decoding layer into a probabilities matrix of size \(30 \times 23\) , in which each column contains the softmax probabilities of the target amino acid at the corresponding position in the peptide. From the final probabilities matrix, we derive the optimal peptide sequence by choosing the amino acid with the highest probability at each position. Additionally, the feature tensor is also sent to several relatively easy auxiliary tasks (auxiliary tasks branch in Fig. 1) to guide and regulate for the target de novo sequencing task. + +It's worth noting that, we did not force the model to output a peptide that has the matched precursor mass; instead, we design the loss function to encourage the model to output the peptide sequence containing as many correct amino acids as possible. This approach can prevent the model from outputting a peptide with the desirable amino acid composition, but neglect the sequence information in the input MS/MS spectrum, in particular during the early stage of the training process when the model has not learned sufficient rules and patterns. + +We collected the HCD spectra from multiple peptide spectral libraries (see details in the Method section) by retaining spectra with the charges of \(1+\) to \(4+\) and from peptides of length no longer than 30, which result in 3,041,570 HCD- MS/MS spectra from 1,066,296 distinct peptides. We implemented using Tensorflow30 and train the model using the Adam optimizer31 for 50 epochs (using learning rate of 0.001). The entire PepNet model contains about 10 million parameters. Notably, we did not distinguish spectra that were acquired by different types of instruments during training and testing, as we observed that the HCD spectra acquired using different instruments (e.g., Orbitrap, Fusion, or Q- Exactive) could all be sequenced at high accuracy + +<--- Page Split ---> + +1 by a single model. PepNet is released as open- source software for de novo peptide sequencing at GitHub (https://github.com/lkytal/PepNet). We also provided an online tool (https://denovo.predfull.com/) for submitting MS/MS spectra (in MGF format) to de novo peptide sequencing. + +## 4 Evaluation Criteria + +5 We evaluated the performance of PepNet over several proteomics datasets. For these datasets, we used the database searching results from MaxQuant \(^{32}\) with scores above the threshold of 80 and the precursor mass differences smaller than 100 ppm as the ground truth to compute the accuracy of the de novo sequencing results. Then we computed two measurements to evaluate the accuracy of the de novo peptide sequencing: the positional accuracy, defined as the fraction of correct amino acid residues reported by PepNet, and the peptide-level accuracy, defined as the fraction of spectra that the sequenced peptides are completely correct. Note that in both measurements, we do not distinguish the amino acids of Leucine and Isoleucine. The two accuracy measurements, as well as the Precision-Recall (PR) curves of PepNet, were compared with the current state- of- the- art de novo peptide sequencing algorithm PointNovo \(^{26}\) and its predecessor DeepNovo. \(^{25}\) Finally, we present here only the de novo sequencing of 2+ and 3+ HCD spectra of unmodified peptides as they are most common in shotgun proteomics data, although this model can be easily extended to sequencing spectra of other charges or sequencing those from the peptides containing common Posttranslational modifications (PTMs). + +## 5 Performance evaluation on a large-scale human proteomics data set + +We first evaluated the performance of PepNet using the HCD spectra from a large- scale human proteomics project (ProteomExchange ID: PXD019483) in comparison with PointNovo and DeepNovo. Here, we report the de novo sequencing results on the subset of identified spectra from the original study. \(^{33}\) in which 468,266 charge 2+ and 87,977 charge 3+ spectra were identified by MaxQuant (i.e., receiving the score 80 or above and mass difference smaller than 100 ppm, as described above), respectively. We will report the de novo sequencing results on the remaining unidentified spectra in a separate section below. + +As shown in Fig. 2, PepNet reported completely correct peptides on 79.1% 2+ and 51.4% 3+ identified spectra, on which the positional accuracy reaches 92.4% and 77.7%, respectively. In comparison, among the peptides reported by PointNovo, 70.4% and 44.2% are completely correct for 2+ and 3+ spectra, respectively, on which the positional accuracies are 72.6% and 55.3%, respectively; among the peptides reported by DeepNovo, only 53.2% and 24.5% are completely correct for 2+ and 3+ spectra, respectively, on which the positional accuracies are 72.5% and 48.5%, respectively. Also, the precision- recall (PR) curves (Fig. 3) + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Comparison of the performance of PepNet, PointNovo and DeepNovo on the charge \(2+\) (a) and charge \(3+\) (b) spectra in the human proteomics dataset. Here, the Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses.
+ +![](images/Figure_3.jpg) + +
Figure 3: The precision-recall (PR) curves of the peptides sequenced by PepNet, PointNovo, and DeepNovo on the spectra of charge \(2+\) (a) and charge \(3+\) (b) in the human proteomics dataset. The two dotted lines represent the precision of 0.99 and 0.95, respectively.
+ +1 showed that PepNet can sequence most peptides with higher accuracy; in particular, it can sequence the peptides from a majority of spectra (i.e., high recall) even when a high threshold of precision (95% or 99%) is used. + +2 peptides from a majority of spectra (i.e., high recall) even when a high threshold of precision (95% or 99%) is used. + +4 We further investigated the performance of PepNet, PointNovo, and DeepNovo on spectra from the + +5 peptides of different lengths, as depicted in Fig. 4. Not surprisingly, the performance of PepNet (as well + +6 as PointNovo and DeepNovo) is reduced with the increasing lengths of peptides, perhaps because 1) longer + +7 peptides are more challenging to be de novo sequenced (with less peptide sequence information in their + +8 MS/MS spectra), and 2) the training dataset contains relatively fewer training samples of longer peptides. + +9 However, the performance of PepNet is persistently better than PointNovo and DeepNovo, especially for + +10 longer peptides. + +## 11 Performance evaluation on proteomics data from non-human organisms + +12 Next, we evaluated the performance of PepNet on the proteomics data from a variety of non- human organ + +13 isms. We used the HCD spectra from a large- scale proteomics project (ProteomExchange ID: PXD014877) + +14 aiming to elucidate the evolutionary landscape of the proteomes in 57 organisms including bacteria, fungi, + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: The performance of PepNet, PointNovo, and DeepNovo on the peptides with different length for the spectra of charge 2+ (top) and charge 3+ (bottom) in the human proteomics dataset. Here, filtered peptide accuracy indicates the peptide-level accuracies of the sequenced peptides after removing those with un-matched precursor masses.
+ +1 plants, and animals. Similar to the results shown above, here we report the de novo sequencing results on + +2 the subset of identified spectra from the original study.33 + +3 As illustrated in Fig. 5, the performance of PepNet is consistent over various organisms. The positional accuracy on the peptides reported by PepNet is \(0.87 \pm 0.02\) on the sequenced peptides from \(2+\) spectra, and \(0.79 \pm 0.07\) on the sequenced peptides from \(3+\) spectra, while the peptide-level accuracy are \(0.76 \pm 0.03\) and \(0.49 \pm 0.09\) on \(2+\) and \(3+\) spectra, respectively. The performance of PointNovo is consistently lower than PepNet by about 0.1 for the positional accuracy and by about 0.15 for the peptide-level accuracy, while the performance of DeepNovo is even lower. Notably, the performance of PepNet on the proteomics data from these varieties of organisms is largely comparable with the performance on the human proteome data (see above), indicating that even though PepNet was trained using MS/MS spectra mostly from human peptides, the model is generalized well for the de novo sequencing of non- human peptides. + +## De novo sequencing of spectra not identified by database searching engines + +12 De novo sequencing of spectra not identified by database searching enginesHere, we demonstrate that PepNet can identify a large fraction of MS/MS spectra that cannot be confidently identified by the database searching engines. We applied PepNet to the subset of MS/MS spectra that were not unidentified by MaxQuant in the human proteomics dataset (ProteomExchange ID: PXD019483, as described above). We then computed a confidence score for each sequenced peptide as the product of the probabilities of the amino acid at all positions in the peptide. After removing sequenced peptides with unmatched precursor masses or with a confidence score lower than 0.5, PepNet reported 5,772,599 unique peptides from 8,856,090 MS/MS spectra (out of the whole set of 19,360,564 MS/MS spectra). Note that + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Performance of PepNet, PointNovo and DeepNovo on the spectra of charge 2+ (a) and charge 3+ (b) in the proteomics datasets acquired from different species.
+ +1 here the confidence score reflects the probability of the sequenced peptide to be correctly considered by the 2 model; hence, we choose the threshold of 0.5, when the probability of the peptide to be correct is higher 3 than that to be incorrect. 4 We then searched these sequenced peptides against the human proteins in the UniProt34 database by 5 using RAPSearch235 (without distinguishing amino acids of Leu and Ile). We observed that 375,043 (4.2%; 6 including 54,644 unique peptides) of these sequenced peptides matched perfectly with human proteins, while 7 another 179,368 (2.0%; including 35,083 unique peptides) contain only one substitution comparing with hu- 8 man proteins (Fig. 6). These results suggest that the de novo sequencing by PepNet is complementary to the + +<--- Page Split ---> + +1 database searching engines (such as MaxQuant), as further analyses of the de novo sequenced peptides that were not identified by database searching engines may lead to the discovery of additional peptides/proteins expressed in the proteome samples, including those not present in the target protein database (e.g., the peptides containing substitutions). + +![](images/Figure_6.jpg) + +
Figure 6: Comparison of de novo sequencing results of PepNet (the pulled out parts) with database searching (identified by MaxQuant) on spectra of (a) charge 2+, and (b) charge 3+. Here we only retain de novo sequencing results that with matched precursor mass and confidence score \(>0.5\) , which further searched against the Uniprot human database.
+ +## 5 Performance of PepNet on proteomics data from Data-Independent Acquisition (DIA) + +7 We further demonstrate that PepNet is also capable of the de novo sequencing of the MS/MS spectra derived from the data acquired by using Data Independent Acquisition (DIA). We compared the performance of PepNet against PointNovo and DeepNovo-DIA on a dataset acquired from a human plasma sample (used for the evaluation of DeepNovo- DIA \(^{36}\) ), which contains a total of 57,909 MS/MS spectra derived from DIA data. Notably, DeepNovo-DIA is a refined DeepNovo model using DIA-derived MS/MS spectra as training data; in contrast, we directly applied the PepNet model trained on the HCD-MS/MS spectra from Data Dependent Acquisition (DDA) as described above without any further refinement. + +It's not surprising that the performance of PepNet is significantly better than PointNovo, as PointNovo was not trained on the DIA data. Interestingly, PepNet also outperforms DeepNovo-DIA which is especially fine- tuned for the DIA- derived spectra. As shown in Figure 7, PepNet achieved comparable performance on the DIA- derived spectra as on the DDA- acquired spectra, with the positional accuracy of 0.69 and peptide- level accuracy of 0.49 on the combined set of \(2+\) and \(03+\) spectra. These results indicate the PepNet model is robust for the de novo sequencing of not only the DDA- acquired MS/MS spectra but also the MS/MS spectra derived from DIA data. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7: The sequencing accuracy (left) and the PR curve (right) of PepNet, PointNovo, and DeepNovo-DIA on a dataset of DIA-derived MS/MS spectra. The Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses. The two dotted lines in the PR curve represent the precision of 0.99 and 0.95, respectively.
+ +## 1 Observed factors related to performance + +2 We found multiple factors influencing the performance of PepNet that are worth discussing. We observed that the number of most intensive peaks retained in the input MS/MS spectrum to PepNet has a significant impact on its performance, as shown in Fig. 8; the performance of PepNet significantly increases when more peaks are retained in the spectra given as input the PepNet. This result indicates that the peaks of low intensities, including some non- backbone fragment ions that were considered as "noise" peaks and thus ignored by conventional de novo sequencing algorithms, also provide useful information for the PepNet model, especially helpful for determining the residues at some positions where the supportive backbone ions are missing. + +![](images/Figure_8.jpg) + +
Figure 8: The positional and peptide-level accuracy of PepNet on the input \(2+\) (a) and \(3+\) (a) spectra in the testing dataset, on which different number of most intensive peaks are retained. The error bar represent the \(99\%\) confidence interval.
+ +Besides, we observed that the positional accuracy by PepNet shared a similar trend for the sequenced peptides of different lengths. As shown in Fig. 9, for the charge \(2+\) spectra, the positional accuracy is low for the first few amino acid residues at the N- terminus, and becomes very high before decreasing gradually until the last residue at the C- terminus. It's not surprising that the C- terminal residue is determined accurately + +<--- Page Split ---> + +1 because most tested spectra are from tryptic peptides. This trend of sequencing error distribution is largely 2 due to the coverage of the observed fragment ions: the first few residues are easier to determine because the 3 \(b_{1}\) and \(b_{2}\) (and even some \(b_{3}\) ) ions are often missing in HCD spectra, while y- ions are weaker than b- ions, 4 causing the C- terminal residues are harder to be determined. + +![](images/Figure_9.jpg) + +
Figure 9: The positional accuracy at each index of the peptides, for sequenced peptide length from 6 to 20. (a) for \(2+\) spectra. (b) for \(3+\) spectra.
+ +## 5 Discussion + +6 In this paper, we present PepNet, a novel deep learning model for high accuracy de novo peptide sequencing 7 from HCD- MS/MS spectra. We first show that PepNet is capable of sequencing human MS/MS spectra with 8 high accuracy (with \(92.4\%\) and \(77.7\%\) positional accuracy on \(2+\) and \(3+\) human spectra from PXD019483, 9 respectively), then we show that the PepNet can also perform constantly well across the proteomes of many 10 non- human organisms (by testing on dataset PXD014877). Furthermore, the de novo results on unidentified 11 spectra show that PepNet could discover a comparable number of new identifications as the protein database 12 search engines, and thus can be used as a powerful tool for proteomic data analyses when a comprehensive 13 target protein sequence database is not available (e.g., in metaproteomics 37). + +14 We believe that the ability to sequence peptides with high accuracy will enable the increasing applications 15 of de novo peptide sequencing in proteomics data analyses. In addition to the peptide sequencing for HCD 16 spectra as presented in this paper, PepNet can be extended to the MS/MS spectra acquired by using other 17 fragmentation methods, such as the electron transfer dissociation (ETD), Electron- Transfer/Higher- Energy 18 Collision Dissociation (EThcD), photodissociation (PD) and the infrared multiphoton dissociation (IRMPD). 19 These methods were often considered to result in complex MS/MS spectra, in which those rich information 20 embedded in the complex MS/MS spectra may hopefully improve the accuracy of de novo peptide sequencing. 21 Furthermore, the low positional sequencing error rate of PepNet will enable the de novo sequencing of whole 22 proteins (e.g., antibodies) based on overlapping peptides (e.g., generated from multiple protease cleavages). 23 Therefore, we anticipate that PepNet will enhance the efficiency in proteomics data analyses and will benefit + +<--- Page Split ---> + +1 the community of life science research. + +## 2 Methods + +## 3 Representation of the MS/MS spectra + +4 We represent the input MS/MS spectrum as a one- dimensional (1- D) vector by binning the spectrum with a given bin width. We considered only the peaks within the range of mass- to- charge ratio (m/z) between 0 and 2000 as most experimental spectra do not contain peaks with m/z above 2,000. By default, we use a bin width of 0.1, which yields the vector representation of 20,000 dimensions. Based on our experiment, using an even smaller bin size (i.e., higher mass resolution) did not improve the performance of de novo sequencing but required longer running times. Finally, we removed the precursor peak in each spectrum, and normalized each spectrum by dividing each peak over the intensity of the maximum peak in that spectrum, similar to our previous work for MS/MS spectrum prediction. \(^{38}\) + +## 12 Deep Neural Network Architecture + +13 As depicted in the Result section (Figure 1), the input for our model is a 20,000 dimensional vector representation of the spectrum as described in the previous section. As depicted in the TCN branch of Fig. 1, six continuing temporal convolutional network (TCN) blocks \(^{29}\) and down- sampling layers are designed to capture the relationships between observed peaks. + +Although those TCN blocks are capable of extracting most information from the input MS/MS spectrum, they work at different levels of resolution, and thus retrieve complementary information: the topmost TCN block emphasizes the detailed local structures of the input spectrum whereas the bottom- most block considers mostly the global features of the spectrum. We follow the idea of the Feature Pyramid Networks \(^{39}\) by adding a top- down branch that merges output from all TCN blocks into a single feature tensor (thus fusing the global and local information together), as depicted in Fig. 1. + +The above feature tensor will be concatenated with meta- information (e.g. charge of the spectra, M/z of the precursor, normalized collision energy) to yield the final feature tensor, which is then converted into a final probabilities matrix of the size \(30 \times 23\) by a softmax decoding layer. As each column in the matrix represents the probabilities of each amino acid at the corresponding position, we can derive the optimal peptide sequence by choosing the amino acid with the highest probability column by column, until we meet the position at which the ending character is of the highest probability. As a post- correction to this strategy, if the theoretical mass of the inferred peptide differs from the experimental precursor mass more than 100 + +<--- Page Split ---> + +1 ppm, we attempt to check if any sub- optimal peptide has matched precursor mass: we substitute the amino acid at each position with the one with the second- highest probability; if the resulting peptide has the matched precursor mass, it will be output as the final de novo sequencing result; otherwise, the original optimal peptide sequence will still be reported. + +5 In addition to the de novo sequencing task, several relatively easy tasks (auxiliary task branch in Fig. 6. 1) are trained simultaneously for archiving better performance, including 1) whether the target peptide is a tryptic peptide; 2) the length of the target peptide; 3) the amino acid composition of the target peptide; 4) the composition of adjacent amino acid pairs in the peptide. These auxiliary tasks serve as guidance and regulations for the de novo sequencing task. + +10 Because the PepNet model is fully convolutional, it decodes the entire peptide sequence simultaneously rather than one amino acid at a time as adopted by the Recurrent Neural Network (RNN) in DeepNovo 12 .25,26 + +## 13 Training Datasets and Process + +14 To compile the training data set, we collected HCD spectra from multiple peptide spectral libraries including the NIST HCD library, \(^{40}\) the NIST Synthetic HCD library, \(^{40}\) the Human HCD library from MassIVE, \(^{41}\) and the synthetic HCD library from ProteomeTools. \(^{13}\) The number of spectra in these libraries is summarized in Supplementary Table 1. In total, by retaining spectra with the charges of \(1+\) to \(4+\) and from peptides of length no longer than 30, we collected 2,962,341 HCD-MS/MS spectra from 1,048,993 distinct peptides. + +19 The whole data set was randomly split into the two subsets, one subset containing 2,908,323 (95%) spectra as the training data, and the other subset containing the remaining 133,247 (5%) spectra serves as the cross- validation set. We ensured that the training and testing data shared no spectra from the same peptide in order to avoid information leakage. The complete training and testing data sets are available in the supplementary files. + +24 We use the Adam optimizer \(^{31}\) with the learning rate of 0.001 to train the model for 50 epochs (batch size of 64 spectra per GPU). We warm- up the learning rate linearly from 0.0001 to 0.001 within the first 10 epochs for training stability. The complete training process takes around 20 hours using 8 cards of NVIDIA A6000 GPU. We pick the model weight from the epoch that performs best on the cross- validation set after the training finished, which achieved over 0.9 positional accuracy on charge \(2+\) spectra and over 0.78 positional accuracy on charge \(3+\) spectra. More details can be found in the Training performance section of the Supplementary Materials. + +<--- Page Split ---> + +## 1 Configuration of PointNovo and DeepNovo + +1 Configuration of PointNovo and DeepNovo2 To build the PointNovo and DeepNovo for compassion, we directly used the source code provided by their original publication without modification. For DeepNovo (and also DeepNovo- DIA), we directly used the provided pre- trained weights for all testing tasks. While for PointNovo, as the authors did not provide pre- trained weights, we retrain the model using the intact training dataset provided in the paper. Both PointNovo and DeepNovo were executed by using their default configurations without modifications. + +## 7 Data availability + +7 Data availability8 The proteomic data of used for this study were taken from previous datasets with the identifier PXD019483 and PXD014877. The MaxQuant searching results provided in these studies were directly reused. The trained models of PepNet were deposited at https://zenodo.org/record/5807120 while the experiment results were deposited at https://zenodo.org/record/6003282. + +## 12 Code availability + +12 Code availability13 Source code and scripts are available on GitHub at https://github.com/lkytal/pepnet. + +## 14 Acknowledgement + +14 Acknowledgement15 This work was supported by National Science Foundation (DBI- 2011271), National Institutes of Health (1R01AI143254) and Indiana University Precision Health Initiative (PHI). + +## 17 Competing interests + +18 The authors declare that they have no conflict of interest. + +## 19 Author contributions + +19 Author contributions20 K.L. and H.T. conceived the project. K.L. designed and implemented the deep learning model. K.L., Y.Y., and H.T. analyzed the data. K.L., Y.Y., and H.T. wrote the manuscript. All authors read and revised the manuscript. + +<--- Page Split ---> + +## References + +(1) Mann, M.; Wilm, M. Error-tolerant identification of peptides in sequence databases by peptide sequence tags. Analytical chemistry 1994, 66, 4390-4399. +(2) Eng, J. K.; McCormack, A. L.; Yates, J. R. An approach to correlate tandem mass spectral data of peptides with amino acid sequences in a protein database. Journal of the american society for mass spectrometry 1994, 5, 976-989. +(3) Hirosawa, M.; Hoshida, M.; Ishikawa, M.; Toya, T. MASCOT: multiple alignment system for protein sequences based on three-way dynamic programming. Bioinformatics 1993, 9, 161-167. +(4) Craig, R.; Beavis, R. C. TANDEM: matching proteins with tandem mass spectra. Bioinformatics 2004, 20, 1466-1467. +(5) Geer, L. Y.; Markey, S. P.; Kowalak, J. A.; Wagner, L.; Xu, M.; Maynard, D. M.; Yang, X.; Shi, W.; Bryant, S. H. Open mass spectrometry search algorithm. Journal of proteome research 2004, 3, 958-964. +(6) Tabb, D. L.; Fernando, C. G.; Chambers, M. C. MyriMatch: highly accurate tandem mass spectral peptide identification by multivariate hypergeometric analysis. Journal of proteome research 2007, 6, 654-661. +(7) Clauser, K. R.; Baker, P.; Burlingame, A. L. Role of accurate mass measurement (±10 ppm) in protein identification strategies employing MS or MS/MS and database searching. Analytical chemistry 1999, 71, 2871-2882. +(8) Chalkley, R. J.; Baker, P. R.; Medzihradszky, K. F.; Lynn, A. J.; Burlingame, A. In-depth analysis of tandem mass spectrometry data from disparate instrument types. Molecular & Cellular Proteomics 2008, 7, 2386-2398. +(9) Kim, S.; Pevzner, P. A. MS-GF+ makes progress towards a universal database search tool for proteomics. Nature communications 2014, 5, 1-10. +(10) Ausloos, P.; Clifton, C.; Lias, S.; Mikaya, A.; Stein, S. E.; Tchekhovskoi, D. V.; Sparkman, O.; Zaikin, V.; Zhu, D. The critical evaluation of a comprehensive mass spectral library. Journal of the American Society for Mass Spectrometry 1999, 10, 287-299. +(11) Yates, J. R.; Morgan, S. F.; Gatlin, C. L.; Griffin, P. R.; Eng, J. K. Method to compare collision-induced dissociation spectra of peptides: potential for library searching and subtractive analysis. Analytical chemistry 1998, 70, 3557-3565. + +<--- Page Split ---> + +(12) Lam, H.; Deutsch, E. W.; Eddes, J. S.; Eng, J. K.; Stein, S. E.; Aebersold, R. Building consensus spectral libraries for peptide identification in proteomics. Nature methods 2008, 5, 873-875. + +(13) Zolg, D. P.; Wilhelm, M.; Schnatbaum, K.; Zerweck, J.; Knaute, T.; Delanghe, B.; Bailey, D. J.; Gessulat, S.; Ehrlich, H.-C.; Weininger, M., et al. Building ProteomeTools based on a complete synthetic human proteome. Nature methods 2017, 14, 259. + +(14) Craig, R.; Cortens, J.; Fenyo, D.; Beavis, R. C. Using annotated peptide mass spectrum libraries for protein identification. Journal of proteome research 2006, 5, 1843-1849. + +(15) Lam, H.; Deutsch, E. W.; Eddes, J. S.; Eng, J. K.; King, N.; Stein, S. E.; Aebersold, R. Development and validation of a spectral library searching method for peptide identification from MS/MS. Proteomics 2007, 7, 655-667. + +(16) Wang, L.; Liu, K.; Li, S.; Tang, H. A Fast and Memory-Efficient Spectral Library Search Algorithm Using Locality-Sensitive Hashing. Proteomics 2020, 20, 2000002. + +(17) Allmer, J. Algorithms for the de novo sequencing of peptides from tandem mass spectra. Expert review of proteomics 2011, 8, 645-657. + +(18) Dančik, V.; Addona, T. A.; Clauser, K. R.; Vath, J. E.; Pevzner, P. A. De novo peptide sequencing via tandem mass spectrometry. Journal of computational biology 1999, 6, 327-342. + +(19) Chen, G.-m.; Firth, M.; Rui, O. M. The dynamic relation between stock returns, trading volume, and volatility. Financial Review 2001, 36, 153-174. + +(20) Frank, A.; Pevzner, P. PepNovo: de novo peptide sequencing via probabilistic network modeling. Analytical chemistry 2005, 77, 964-973. + +(21) Zhang, J.; Xin, L.; Shan, B.; Chen, W.; Xie, M.; Yuen, D.; Zhang, W.; Zhang, Z.; Lajoie, G. A.; Ma, B. PEAKS DB: de novo sequencing assisted database search for sensitive and accurate peptide identification. Molecular & cellular proteomics 2012, 11. + +(22) Chi, H.; Chen, H.; He, K.; Wu, L.; Yang, B.; Sun, R.-X.; Liu, J.; Zeng, W.-F.; Song, C.-Q.; He, S.-M., et al. pNovo+: de novo peptide sequencing using complementary HCD and ETD tandem mass spectra. Journal of proteome research 2013, 12, 615-625. + +(23) Jeong, K.; Kim, S.; Pevzner, P. A. UniNovo: a universal tool for de novo peptide sequencing. Bioinformatics 2013, 29, 1953-1962. + +<--- Page Split ---> + +(24) Ma, B. Novor: real-time peptide de novo sequencing software. Journal of the American Society for Mass Spectrometry 2015, 26, 1885-1894. + +(25) Tran, N. H.; Zhang, X.; Xin, L.; Shan, B.; Li, M. De novo peptide sequencing by deep learning. Proceedings of the National Academy of Sciences 2017, 114, 8247-8252. + +(26) Qiao, R.; Tran, N. H.; Xin, L.; Shan, B.; Li, M.; Ghodsi, A. Deepnovov2: Better de novo peptide sequencing with deep learning. arXiv preprint arXiv:1904.08514 2019, + +(27) Tran, N. H.; Qiao, R.; Xin, L.; Chen, X.; Liu, C.; Zhang, X.; Shan, B.; Ghodsi, A.; Li, M. Deep learning enables de novo peptide sequencing from data-independent-acquisition mass spectrometry. Nature methods 2019, 16, 63-66. + +(28) Qiao, R.; Tran, N. H.; Xin, L.; Chen, X.; Li, M.; Shan, B.; Ghodsi, A. Computationally instrument-resolution-independent de novo peptide sequencing for high-resolution devices. Nature Machine Intelligence 2021, 3, 420-425. + +(29) Bai, S.; Kolter, J. Z.; Koltun, V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271 2018, + +(30) Abadi, M.; Barham, P.; Chen, J.; Chen, Z.; Davis, A.; Dean, J.; Devin, M.; Ghemawat, S.; Irving, G.; Isard, M., et al. Tensorflow: A system for large-scale machine learning. 12th {USENIX} Symposium on Operating Systems Design and Implementation ({OSDI} 16). 2016; pp 265-283. + +(31) Kingma, D. P.; Ba, J. Adam: A method for stochastic optimization. International Conference on Learning Representations (ICLR). 2015. + +(32) Cox, J.; Mann, M. MaxQuant enables high peptide identification rates, individualized ppb-range mass accuracies and proteome-wide protein quantification. Nature biotechnology 2008, 26, 1367-1372. + +(33) Müller, J. B.; Geyer, P. E.; Colaço, A. R.; Treit, P. V.; Strauss, M. T.; Oroshi, M.; Doll, S.; Winter, S. V.; Bader, J. M.; Köhler, N., et al. The proteome landscape of the kingdoms of life. Nature 2020, 582, 592-596. + +(34) Consortium, U. UniProt: a hub for protein information. Nucleic acids research 2015, 43, D204-D212. + +(35) Zhao, Y.; Tang, H.; Ye, Y. RAPSearch2: a fast and memory-efficient protein similarity search tool for next-generation sequencing data. Bioinformatics 2012, 28, 125-126. + +<--- Page Split ---> + +(36) Tran, N. H.; Qiao, R.; Xin, L.; Chen, X.; Liu, C.; Zhang, X.; Shan, B.; Ghodsi, A.; Li, M. Deep learning enables de novo peptide sequencing from data-independent-acquisition mass spectrometry. Nature methods 2019, 16, 63–66. + +(37) Maron, P.-A.; Ranjard, L.; Mougel, C.; Lemanceau, P. Metaproteomics: a new approach for studying functional microbial ecology. Microbial ecology 2007, 53, 486–493. + +(38) Liu, K.; Li, S.; Wang, L.; Ye, Y.; Tang, H. Full-spectrum prediction of peptides tandem mass spectra using deep neural network. Analytical chemistry 2020, 92, 4275–4283. + +(39) Lin, T.-Y.; Dollár, P.; Girshick, R.; He, K.; Hariharan, B.; Belongie, S. Feature pyramid networks for object detection. Proceedings of the IEEE conference on computer vision and pattern recognition. 2017; pp 2117–2125. + +(40) Yang, X.; Neta, P.; Stein, S. E. Extending a Tandem Mass Spectral Library to Include MS 2 Spectra of Fragment Ions Produced In-Source and MS n Spectra. Journal of The American Society for Mass Spectrometry 2017, 28, 2280–2287. + +(41) Wang, M.; Wang, J.; Carver, J.; Pullman, B. S.; Cha, S. W.; Bandeira, N. Assembling the Community-Scale Discoverable Human Proteome. Cell systems 2018, 7, 412–421. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef_det.mmd b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7ac8fff599176db1dce6e84d0081f4495634842d --- /dev/null +++ b/preprint/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef/preprint__1230baef4d02c20b26a0c0eb2f8467621c7d9b6c96c33ef645baed64b39412ef_det.mmd @@ -0,0 +1,437 @@ +<|ref|>title<|/ref|><|det|>[[45, 106, 902, 178]]<|/det|> +# PepNet: A Fully Convolutional Neural Network for De novo Peptide Sequencing + +<|ref|>text<|/ref|><|det|>[[44, 196, 688, 240]]<|/det|> +Kaiyuan Liu Indiana University Bloomington https://orcid.org/0000- 0002- 3404- 2802 + +<|ref|>text<|/ref|><|det|>[[44, 244, 666, 330]]<|/det|> +Yuzhen Ye School of Informatics and Computing, Indiana University Bloomington Haixu Tang ( \(\circleddash\) hatang@indiana.edu) Indiana University Bloomington + +<|ref|>sub_title<|/ref|><|det|>[[44, 370, 182, 389]]<|/det|> +## Methods Article + +<|ref|>text<|/ref|><|det|>[[44, 408, 137, 427]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 446, 336, 465]]<|/det|> +Posted Date: February 17th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 484, 475, 503]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1341615/v1 + +<|ref|>text<|/ref|><|det|>[[44, 521, 910, 564]]<|/det|> +License: © \(\circleddash\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[145, 131, 853, 204]]<|/det|> +# PepNet: A Fully Convolutional Neural Network for De novo Peptide Sequencing + +<|ref|>text<|/ref|><|det|>[[328, 230, 666, 248]]<|/det|> +Kaiyuan Liu, Yuzhen Ye, and Haixu Tang\* + +<|ref|>text<|/ref|><|det|>[[135, 270, 864, 312]]<|/det|> +Luddy School of Informatics, Computing, and Engineering, Indiana University, Bloomington, Indiana 47405, United States + +<|ref|>text<|/ref|><|det|>[[401, 334, 595, 350]]<|/det|> +E- mail: hatang@indiana.edu + +<|ref|>sub_title<|/ref|><|det|>[[465, 372, 532, 387]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[152, 397, 845, 667]]<|/det|> +The de novo peptide sequencing, which does not rely on a comprehensive target sequence database, provided us a way to identify novel peptides from tandem mass (MS/MS) spectra. However, current de novo sequencing algorithms suffer from lower accuracy and coverage, which hinders their applications in proteomics. In this paper, we present PepNet, a fully convolutional neural network (CNN) for high accuracy de novo peptide sequencing. It takes an MS/MS spectrum (represented as a high dimensional vector) as input, and outputs the optimal peptide sequence along with its confidence score. Our model was trained using a total of 30 million high- energy collisional dissociation (HCD) MS/MS spectra from multiple human peptide spectral libraries. The evaluation results show that PepNet significantly outperformed currently best- performing de novo sequencing algorithms (e.g. PointNovo and DeepNovo) at both peptide level accuracy and positional level accuracy. In addition, PepNet can sequence a large fraction of spectra that were not identified by database search engines, and thus could be used as a complementary tool of database search engines for peptide identification in proteomics. + +<|ref|>sub_title<|/ref|><|det|>[[90, 703, 262, 722]]<|/det|> +## 15 Introduction + +<|ref|>text<|/ref|><|det|>[[90, 742, 884, 860]]<|/det|> +The past decade has witnessed the great advances of mass spectrometry techniques, in particular liquid chromatography coupled tandem mass spectrometry (LC- MS/MS). With enhanced throughput and sensitivity, LC- MS/MS has become one of the most commonly used approaches to functional studies of proteins at the whole proteome scale across various physiological (e.g., diseases) conditions in higher organisms including human. + +<|ref|>text<|/ref|><|det|>[[90, 869, 884, 911]]<|/det|> +In a typical proteomics experiment, after MS/MS spectra are acquired, the first and arguably most important step is to identify the peptides from these spectra. Numerous algorithms have been developed to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 88, 884, 258]]<|/det|> +address this problem, which mostly falls into three categories: protein database searching, spectral library searching, and de novo sequencing. Pioneered the peptide sequence tag method \(^{1}\) and the Sequest algorithm, \(^{2}\) protein database searching is the predominant approach used for peptide identification. Till now many peptide searching engines have been developed, including commonly used tools such as Mascot, \(^{3}\) X!Tandem \(^{4}\) OMSSA, \(^{5}\) MyriMatch, \(^{6}\) Protein Prospector \(^{7,8}\) and MSGF \(^{+}\) . \(^{9}\) Those methods compare the experimental spectra with the theoretical spectra generated from the peptides in a protein database and report those likely true peptide- spectrum matches (PSMs). + +<|ref|>text<|/ref|><|det|>[[90, 265, 884, 410]]<|/det|> +In contrast, the spectral library search approach compares newly acquired MS/MS spectra against a library containing previously characterized experimental spectra that were used in early computational analysis. \(^{10,11}\) Thanks to the improved repeatability and reproducibility of MS/MS data as well as the increasing availability of massive experimental spectra (e.g., from the proteomics data repository \(^{12}\) and large- scale synthetic peptides projects \(^{13}\) ), the spectral library search approach has become more increasingly adopted and is implemented in software tools such as X!hunter, \(^{14}\) SpectraST \(^{15}\) and msSLASH. \(^{16}\) + +<|ref|>text<|/ref|><|det|>[[90, 416, 884, 636]]<|/det|> +Finally, de novo sequencing algorithms attempt to derive a peptide sequence directly from its MS/MS spectrum without using references such as a spectral library or a protein sequence database. \(^{17}\) Many de novo sequencing algorithms adopted a graph theoretical formulation to compute the longest path in the spectrum graph by employing a dynamic programming algorithm \(^{18,19}\) and adaptive scoring schemes. \(^{20- 22}\) With the advancement of high- resolution MS instruments, the performance of de novo sequencing algorithms improves significantly, \(^{23,24}\) in particular with more sophisticated scoring schemes. More recently, DeepNovo \(^{25- 27}\) and its successor PointNovo \(^{28}\) was developed using deep learning algorithms, which automatically learn the fragment ion patterns relevant to peptide sequences from massive MS/MS spectra of peptides and reported improved performance. + +<|ref|>sub_title<|/ref|><|det|>[[90, 672, 203, 691]]<|/det|> +## 2 Results + +<|ref|>sub_title<|/ref|><|det|>[[90, 714, 784, 734]]<|/det|> +## 2.1 Accurate HCD-MS/MS spectra de novo sequencing by deep learning + +<|ref|>text<|/ref|><|det|>[[90, 749, 884, 893]]<|/det|> +We present a deep learning algorithm, PepNet, that directly outputs the peptide sequence from a given HCD- MS/MS spectra with high accuracy. As depicted in Figure 1, the input for our model is a 20,000 dimensional vector representation of the spectrum (for details see the Method section). The input vector will go through six continuing temporal convolutional network (TCN) blocks \(^{29}\) and down- sampling layers to capture the relationships between observed peaks, as depicted in the TCN branch of Fig. 1. Apparently, those TCN blocks working on different resolution levels. Thus we add a merging branch (top- down branch + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 90, 863, 345]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[191, 354, 803, 371]]<|/det|> +
Figure 1: The neural network architecture of PepNet for de novo peptide sequencing.
+ +<|ref|>text<|/ref|><|det|>[[90, 394, 886, 562]]<|/det|> +1 in Fig. 1) that fuse the global and local information into a single feature tensor. Finally, the feature tensor is then converted by a softmax decoding layer into a probabilities matrix of size \(30 \times 23\) , in which each column contains the softmax probabilities of the target amino acid at the corresponding position in the peptide. From the final probabilities matrix, we derive the optimal peptide sequence by choosing the amino acid with the highest probability at each position. Additionally, the feature tensor is also sent to several relatively easy auxiliary tasks (auxiliary tasks branch in Fig. 1) to guide and regulate for the target de novo sequencing task. + +<|ref|>text<|/ref|><|det|>[[90, 571, 886, 715]]<|/det|> +It's worth noting that, we did not force the model to output a peptide that has the matched precursor mass; instead, we design the loss function to encourage the model to output the peptide sequence containing as many correct amino acids as possible. This approach can prevent the model from outputting a peptide with the desirable amino acid composition, but neglect the sequence information in the input MS/MS spectrum, in particular during the early stage of the training process when the model has not learned sufficient rules and patterns. + +<|ref|>text<|/ref|><|det|>[[90, 722, 886, 892]]<|/det|> +We collected the HCD spectra from multiple peptide spectral libraries (see details in the Method section) by retaining spectra with the charges of \(1+\) to \(4+\) and from peptides of length no longer than 30, which result in 3,041,570 HCD- MS/MS spectra from 1,066,296 distinct peptides. We implemented using Tensorflow30 and train the model using the Adam optimizer31 for 50 epochs (using learning rate of 0.001). The entire PepNet model contains about 10 million parameters. Notably, we did not distinguish spectra that were acquired by different types of instruments during training and testing, as we observed that the HCD spectra acquired using different instruments (e.g., Orbitrap, Fusion, or Q- Exactive) could all be sequenced at high accuracy + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 88, 884, 157]]<|/det|> +1 by a single model. PepNet is released as open- source software for de novo peptide sequencing at GitHub (https://github.com/lkytal/PepNet). We also provided an online tool (https://denovo.predfull.com/) for submitting MS/MS spectra (in MGF format) to de novo peptide sequencing. + +<|ref|>sub_title<|/ref|><|det|>[[93, 185, 303, 203]]<|/det|> +## 4 Evaluation Criteria + +<|ref|>text<|/ref|><|det|>[[90, 220, 885, 540]]<|/det|> +5 We evaluated the performance of PepNet over several proteomics datasets. For these datasets, we used the database searching results from MaxQuant \(^{32}\) with scores above the threshold of 80 and the precursor mass differences smaller than 100 ppm as the ground truth to compute the accuracy of the de novo sequencing results. Then we computed two measurements to evaluate the accuracy of the de novo peptide sequencing: the positional accuracy, defined as the fraction of correct amino acid residues reported by PepNet, and the peptide-level accuracy, defined as the fraction of spectra that the sequenced peptides are completely correct. Note that in both measurements, we do not distinguish the amino acids of Leucine and Isoleucine. The two accuracy measurements, as well as the Precision-Recall (PR) curves of PepNet, were compared with the current state- of- the- art de novo peptide sequencing algorithm PointNovo \(^{26}\) and its predecessor DeepNovo. \(^{25}\) Finally, we present here only the de novo sequencing of 2+ and 3+ HCD spectra of unmodified peptides as they are most common in shotgun proteomics data, although this model can be easily extended to sequencing spectra of other charges or sequencing those from the peptides containing common Posttranslational modifications (PTMs). + +<|ref|>sub_title<|/ref|><|det|>[[92, 567, 768, 586]]<|/det|> +## 5 Performance evaluation on a large-scale human proteomics data set + +<|ref|>text<|/ref|><|det|>[[90, 601, 885, 744]]<|/det|> +We first evaluated the performance of PepNet using the HCD spectra from a large- scale human proteomics project (ProteomExchange ID: PXD019483) in comparison with PointNovo and DeepNovo. Here, we report the de novo sequencing results on the subset of identified spectra from the original study. \(^{33}\) in which 468,266 charge 2+ and 87,977 charge 3+ spectra were identified by MaxQuant (i.e., receiving the score 80 or above and mass difference smaller than 100 ppm, as described above), respectively. We will report the de novo sequencing results on the remaining unidentified spectra in a separate section below. + +<|ref|>text<|/ref|><|det|>[[90, 752, 884, 896]]<|/det|> +As shown in Fig. 2, PepNet reported completely correct peptides on 79.1% 2+ and 51.4% 3+ identified spectra, on which the positional accuracy reaches 92.4% and 77.7%, respectively. In comparison, among the peptides reported by PointNovo, 70.4% and 44.2% are completely correct for 2+ and 3+ spectra, respectively, on which the positional accuracies are 72.6% and 55.3%, respectively; among the peptides reported by DeepNovo, only 53.2% and 24.5% are completely correct for 2+ and 3+ spectra, respectively, on which the positional accuracies are 72.5% and 48.5%, respectively. Also, the precision- recall (PR) curves (Fig. 3) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[156, 90, 842, 220]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 232, 883, 294]]<|/det|> +
Figure 2: Comparison of the performance of PepNet, PointNovo and DeepNovo on the charge \(2+\) (a) and charge \(3+\) (b) spectra in the human proteomics dataset. Here, the Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses.
+ +<|ref|>image<|/ref|><|det|>[[156, 310, 842, 455]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 465, 883, 512]]<|/det|> +
Figure 3: The precision-recall (PR) curves of the peptides sequenced by PepNet, PointNovo, and DeepNovo on the spectra of charge \(2+\) (a) and charge \(3+\) (b) in the human proteomics dataset. The two dotted lines represent the precision of 0.99 and 0.95, respectively.
+ +<|ref|>text<|/ref|><|det|>[[92, 536, 884, 576]]<|/det|> +1 showed that PepNet can sequence most peptides with higher accuracy; in particular, it can sequence the peptides from a majority of spectra (i.e., high recall) even when a high threshold of precision (95% or 99%) is used. + +<|ref|>text<|/ref|><|det|>[[92, 584, 884, 600]]<|/det|> +2 peptides from a majority of spectra (i.e., high recall) even when a high threshold of precision (95% or 99%) is used. + +<|ref|>text<|/ref|><|det|>[[92, 610, 883, 628]]<|/det|> +4 We further investigated the performance of PepNet, PointNovo, and DeepNovo on spectra from the + +<|ref|>text<|/ref|><|det|>[[92, 637, 883, 655]]<|/det|> +5 peptides of different lengths, as depicted in Fig. 4. Not surprisingly, the performance of PepNet (as well + +<|ref|>text<|/ref|><|det|>[[92, 663, 883, 681]]<|/det|> +6 as PointNovo and DeepNovo) is reduced with the increasing lengths of peptides, perhaps because 1) longer + +<|ref|>text<|/ref|><|det|>[[92, 689, 883, 707]]<|/det|> +7 peptides are more challenging to be de novo sequenced (with less peptide sequence information in their + +<|ref|>text<|/ref|><|det|>[[92, 715, 883, 732]]<|/det|> +8 MS/MS spectra), and 2) the training dataset contains relatively fewer training samples of longer peptides. + +<|ref|>text<|/ref|><|det|>[[92, 740, 883, 757]]<|/det|> +9 However, the performance of PepNet is persistently better than PointNovo and DeepNovo, especially for + +<|ref|>text<|/ref|><|det|>[[92, 766, 230, 781]]<|/det|> +10 longer peptides. + +<|ref|>sub_title<|/ref|><|det|>[[92, 809, 805, 828]]<|/det|> +## 11 Performance evaluation on proteomics data from non-human organisms + +<|ref|>text<|/ref|><|det|>[[92, 844, 883, 861]]<|/det|> +12 Next, we evaluated the performance of PepNet on the proteomics data from a variety of non- human organ + +<|ref|>text<|/ref|><|det|>[[92, 870, 883, 886]]<|/det|> +13 isms. We used the HCD spectra from a large- scale proteomics project (ProteomExchange ID: PXD014877) + +<|ref|>text<|/ref|><|det|>[[92, 895, 883, 911]]<|/det|> +14 aiming to elucidate the evolutionary landscape of the proteomes in 57 organisms including bacteria, fungi, + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[172, 91, 822, 308]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 320, 884, 383]]<|/det|> +
Figure 4: The performance of PepNet, PointNovo, and DeepNovo on the peptides with different length for the spectra of charge 2+ (top) and charge 3+ (bottom) in the human proteomics dataset. Here, filtered peptide accuracy indicates the peptide-level accuracies of the sequenced peptides after removing those with un-matched precursor masses.
+ +<|ref|>text<|/ref|><|det|>[[92, 406, 884, 425]]<|/det|> +1 plants, and animals. Similar to the results shown above, here we report the de novo sequencing results on + +<|ref|>text<|/ref|><|det|>[[92, 432, 533, 449]]<|/det|> +2 the subset of identified spectra from the original study.33 + +<|ref|>text<|/ref|><|det|>[[90, 456, 884, 675]]<|/det|> +3 As illustrated in Fig. 5, the performance of PepNet is consistent over various organisms. The positional accuracy on the peptides reported by PepNet is \(0.87 \pm 0.02\) on the sequenced peptides from \(2+\) spectra, and \(0.79 \pm 0.07\) on the sequenced peptides from \(3+\) spectra, while the peptide-level accuracy are \(0.76 \pm 0.03\) and \(0.49 \pm 0.09\) on \(2+\) and \(3+\) spectra, respectively. The performance of PointNovo is consistently lower than PepNet by about 0.1 for the positional accuracy and by about 0.15 for the peptide-level accuracy, while the performance of DeepNovo is even lower. Notably, the performance of PepNet on the proteomics data from these varieties of organisms is largely comparable with the performance on the human proteome data (see above), indicating that even though PepNet was trained using MS/MS spectra mostly from human peptides, the model is generalized well for the de novo sequencing of non- human peptides. + +<|ref|>sub_title<|/ref|><|det|>[[90, 703, 844, 723]]<|/det|> +## De novo sequencing of spectra not identified by database searching engines + +<|ref|>text<|/ref|><|det|>[[90, 737, 884, 907]]<|/det|> +12 De novo sequencing of spectra not identified by database searching enginesHere, we demonstrate that PepNet can identify a large fraction of MS/MS spectra that cannot be confidently identified by the database searching engines. We applied PepNet to the subset of MS/MS spectra that were not unidentified by MaxQuant in the human proteomics dataset (ProteomExchange ID: PXD019483, as described above). We then computed a confidence score for each sequenced peptide as the product of the probabilities of the amino acid at all positions in the peptide. After removing sequenced peptides with unmatched precursor masses or with a confidence score lower than 0.5, PepNet reported 5,772,599 unique peptides from 8,856,090 MS/MS spectra (out of the whole set of 19,360,564 MS/MS spectra). Note that + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 88, 861, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 638, 880, 671]]<|/det|> +
Figure 5: Performance of PepNet, PointNovo and DeepNovo on the spectra of charge 2+ (a) and charge 3+ (b) in the proteomics datasets acquired from different species.
+ +<|ref|>text<|/ref|><|det|>[[92, 694, 884, 886]]<|/det|> +1 here the confidence score reflects the probability of the sequenced peptide to be correctly considered by the 2 model; hence, we choose the threshold of 0.5, when the probability of the peptide to be correct is higher 3 than that to be incorrect. 4 We then searched these sequenced peptides against the human proteins in the UniProt34 database by 5 using RAPSearch235 (without distinguishing amino acids of Leu and Ile). We observed that 375,043 (4.2%; 6 including 54,644 unique peptides) of these sequenced peptides matched perfectly with human proteins, while 7 another 179,368 (2.0%; including 35,083 unique peptides) contain only one substitution comparing with hu- 8 man proteins (Fig. 6). These results suggest that the de novo sequencing by PepNet is complementary to the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 88, 884, 183]]<|/det|> +1 database searching engines (such as MaxQuant), as further analyses of the de novo sequenced peptides that were not identified by database searching engines may lead to the discovery of additional peptides/proteins expressed in the proteome samples, including those not present in the target protein database (e.g., the peptides containing substitutions). + +<|ref|>image<|/ref|><|det|>[[118, 198, 880, 352]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 363, 884, 425]]<|/det|> +
Figure 6: Comparison of de novo sequencing results of PepNet (the pulled out parts) with database searching (identified by MaxQuant) on spectra of (a) charge 2+, and (b) charge 3+. Here we only retain de novo sequencing results that with matched precursor mass and confidence score \(>0.5\) , which further searched against the Uniprot human database.
+ +<|ref|>sub_title<|/ref|><|det|>[[92, 469, 882, 520]]<|/det|> +## 5 Performance of PepNet on proteomics data from Data-Independent Acquisition (DIA) + +<|ref|>text<|/ref|><|det|>[[90, 533, 884, 701]]<|/det|> +7 We further demonstrate that PepNet is also capable of the de novo sequencing of the MS/MS spectra derived from the data acquired by using Data Independent Acquisition (DIA). We compared the performance of PepNet against PointNovo and DeepNovo-DIA on a dataset acquired from a human plasma sample (used for the evaluation of DeepNovo- DIA \(^{36}\) ), which contains a total of 57,909 MS/MS spectra derived from DIA data. Notably, DeepNovo-DIA is a refined DeepNovo model using DIA-derived MS/MS spectra as training data; in contrast, we directly applied the PepNet model trained on the HCD-MS/MS spectra from Data Dependent Acquisition (DDA) as described above without any further refinement. + +<|ref|>text<|/ref|><|det|>[[90, 710, 884, 878]]<|/det|> +It's not surprising that the performance of PepNet is significantly better than PointNovo, as PointNovo was not trained on the DIA data. Interestingly, PepNet also outperforms DeepNovo-DIA which is especially fine- tuned for the DIA- derived spectra. As shown in Figure 7, PepNet achieved comparable performance on the DIA- derived spectra as on the DDA- acquired spectra, with the positional accuracy of 0.69 and peptide- level accuracy of 0.49 on the combined set of \(2+\) and \(03+\) spectra. These results indicate the PepNet model is robust for the de novo sequencing of not only the DDA- acquired MS/MS spectra but also the MS/MS spectra derived from DIA data. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[139, 90, 861, 243]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 255, 882, 317]]<|/det|> +
Figure 7: The sequencing accuracy (left) and the PR curve (right) of PepNet, PointNovo, and DeepNovo-DIA on a dataset of DIA-derived MS/MS spectra. The Filtered Peptide Accuracy is referred to as the peptide-level accuracy on the sequenced peptides after removing those with unmatched precursor masses. The two dotted lines in the PR curve represent the precision of 0.99 and 0.95, respectively.
+ +<|ref|>sub_title<|/ref|><|det|>[[92, 340, 506, 360]]<|/det|> +## 1 Observed factors related to performance + +<|ref|>text<|/ref|><|det|>[[92, 374, 884, 569]]<|/det|> +2 We found multiple factors influencing the performance of PepNet that are worth discussing. We observed that the number of most intensive peaks retained in the input MS/MS spectrum to PepNet has a significant impact on its performance, as shown in Fig. 8; the performance of PepNet significantly increases when more peaks are retained in the spectra given as input the PepNet. This result indicates that the peaks of low intensities, including some non- backbone fragment ions that were considered as "noise" peaks and thus ignored by conventional de novo sequencing algorithms, also provide useful information for the PepNet model, especially helpful for determining the residues at some positions where the supportive backbone ions are missing. + +<|ref|>image<|/ref|><|det|>[[137, 583, 860, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 744, 883, 790]]<|/det|> +
Figure 8: The positional and peptide-level accuracy of PepNet on the input \(2+\) (a) and \(3+\) (a) spectra in the testing dataset, on which different number of most intensive peaks are retained. The error bar represent the \(99\%\) confidence interval.
+ +<|ref|>text<|/ref|><|det|>[[88, 814, 883, 909]]<|/det|> +Besides, we observed that the positional accuracy by PepNet shared a similar trend for the sequenced peptides of different lengths. As shown in Fig. 9, for the charge \(2+\) spectra, the positional accuracy is low for the first few amino acid residues at the N- terminus, and becomes very high before decreasing gradually until the last residue at the C- terminus. It's not surprising that the C- terminal residue is determined accurately + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 88, 884, 182]]<|/det|> +1 because most tested spectra are from tryptic peptides. This trend of sequencing error distribution is largely 2 due to the coverage of the observed fragment ions: the first few residues are easier to determine because the 3 \(b_{1}\) and \(b_{2}\) (and even some \(b_{3}\) ) ions are often missing in HCD spectra, while y- ions are weaker than b- ions, 4 causing the C- terminal residues are harder to be determined. + +<|ref|>image<|/ref|><|det|>[[135, 198, 861, 345]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 355, 881, 389]]<|/det|> +
Figure 9: The positional accuracy at each index of the peptides, for sequenced peptide length from 6 to 20. (a) for \(2+\) spectra. (b) for \(3+\) spectra.
+ +<|ref|>sub_title<|/ref|><|det|>[[92, 432, 218, 450]]<|/det|> +## 5 Discussion + +<|ref|>text<|/ref|><|det|>[[90, 465, 886, 660]]<|/det|> +6 In this paper, we present PepNet, a novel deep learning model for high accuracy de novo peptide sequencing 7 from HCD- MS/MS spectra. We first show that PepNet is capable of sequencing human MS/MS spectra with 8 high accuracy (with \(92.4\%\) and \(77.7\%\) positional accuracy on \(2+\) and \(3+\) human spectra from PXD019483, 9 respectively), then we show that the PepNet can also perform constantly well across the proteomes of many 10 non- human organisms (by testing on dataset PXD014877). Furthermore, the de novo results on unidentified 11 spectra show that PepNet could discover a comparable number of new identifications as the protein database 12 search engines, and thus can be used as a powerful tool for proteomic data analyses when a comprehensive 13 target protein sequence database is not available (e.g., in metaproteomics 37). + +<|ref|>text<|/ref|><|det|>[[88, 667, 886, 912]]<|/det|> +14 We believe that the ability to sequence peptides with high accuracy will enable the increasing applications 15 of de novo peptide sequencing in proteomics data analyses. In addition to the peptide sequencing for HCD 16 spectra as presented in this paper, PepNet can be extended to the MS/MS spectra acquired by using other 17 fragmentation methods, such as the electron transfer dissociation (ETD), Electron- Transfer/Higher- Energy 18 Collision Dissociation (EThcD), photodissociation (PD) and the infrared multiphoton dissociation (IRMPD). 19 These methods were often considered to result in complex MS/MS spectra, in which those rich information 20 embedded in the complex MS/MS spectra may hopefully improve the accuracy of de novo peptide sequencing. 21 Furthermore, the low positional sequencing error rate of PepNet will enable the de novo sequencing of whole 22 proteins (e.g., antibodies) based on overlapping peptides (e.g., generated from multiple protease cleavages). 23 Therefore, we anticipate that PepNet will enhance the efficiency in proteomics data analyses and will benefit + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[93, 90, 391, 105]]<|/det|> +1 the community of life science research. + +<|ref|>sub_title<|/ref|><|det|>[[93, 142, 220, 163]]<|/det|> +## 2 Methods + +<|ref|>sub_title<|/ref|><|det|>[[93, 186, 488, 205]]<|/det|> +## 3 Representation of the MS/MS spectra + +<|ref|>text<|/ref|><|det|>[[92, 220, 886, 415]]<|/det|> +4 We represent the input MS/MS spectrum as a one- dimensional (1- D) vector by binning the spectrum with a given bin width. We considered only the peaks within the range of mass- to- charge ratio (m/z) between 0 and 2000 as most experimental spectra do not contain peaks with m/z above 2,000. By default, we use a bin width of 0.1, which yields the vector representation of 20,000 dimensions. Based on our experiment, using an even smaller bin size (i.e., higher mass resolution) did not improve the performance of de novo sequencing but required longer running times. Finally, we removed the precursor peak in each spectrum, and normalized each spectrum by dividing each peak over the intensity of the maximum peak in that spectrum, similar to our previous work for MS/MS spectrum prediction. \(^{38}\) + +<|ref|>sub_title<|/ref|><|det|>[[92, 441, 458, 460]]<|/det|> +## 12 Deep Neural Network Architecture + +<|ref|>text<|/ref|><|det|>[[92, 476, 885, 570]]<|/det|> +13 As depicted in the Result section (Figure 1), the input for our model is a 20,000 dimensional vector representation of the spectrum as described in the previous section. As depicted in the TCN branch of Fig. 1, six continuing temporal convolutional network (TCN) blocks \(^{29}\) and down- sampling layers are designed to capture the relationships between observed peaks. + +<|ref|>text<|/ref|><|det|>[[90, 576, 885, 722]]<|/det|> +Although those TCN blocks are capable of extracting most information from the input MS/MS spectrum, they work at different levels of resolution, and thus retrieve complementary information: the topmost TCN block emphasizes the detailed local structures of the input spectrum whereas the bottom- most block considers mostly the global features of the spectrum. We follow the idea of the Feature Pyramid Networks \(^{39}\) by adding a top- down branch that merges output from all TCN blocks into a single feature tensor (thus fusing the global and local information together), as depicted in Fig. 1. + +<|ref|>text<|/ref|><|det|>[[89, 728, 885, 896]]<|/det|> +The above feature tensor will be concatenated with meta- information (e.g. charge of the spectra, M/z of the precursor, normalized collision energy) to yield the final feature tensor, which is then converted into a final probabilities matrix of the size \(30 \times 23\) by a softmax decoding layer. As each column in the matrix represents the probabilities of each amino acid at the corresponding position, we can derive the optimal peptide sequence by choosing the amino acid with the highest probability column by column, until we meet the position at which the ending character is of the highest probability. As a post- correction to this strategy, if the theoretical mass of the inferred peptide differs from the experimental precursor mass more than 100 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 88, 884, 183]]<|/det|> +1 ppm, we attempt to check if any sub- optimal peptide has matched precursor mass: we substitute the amino acid at each position with the one with the second- highest probability; if the resulting peptide has the matched precursor mass, it will be output as the final de novo sequencing result; otherwise, the original optimal peptide sequence will still be reported. + +<|ref|>text<|/ref|><|det|>[[92, 189, 884, 308]]<|/det|> +5 In addition to the de novo sequencing task, several relatively easy tasks (auxiliary task branch in Fig. 6. 1) are trained simultaneously for archiving better performance, including 1) whether the target peptide is a tryptic peptide; 2) the length of the target peptide; 3) the amino acid composition of the target peptide; 4) the composition of adjacent amino acid pairs in the peptide. These auxiliary tasks serve as guidance and regulations for the de novo sequencing task. + +<|ref|>text<|/ref|><|det|>[[92, 315, 884, 380]]<|/det|> +10 Because the PepNet model is fully convolutional, it decodes the entire peptide sequence simultaneously rather than one amino acid at a time as adopted by the Recurrent Neural Network (RNN) in DeepNovo 12 .25,26 + +<|ref|>sub_title<|/ref|><|det|>[[92, 411, 411, 430]]<|/det|> +## 13 Training Datasets and Process + +<|ref|>text<|/ref|><|det|>[[92, 445, 884, 565]]<|/det|> +14 To compile the training data set, we collected HCD spectra from multiple peptide spectral libraries including the NIST HCD library, \(^{40}\) the NIST Synthetic HCD library, \(^{40}\) the Human HCD library from MassIVE, \(^{41}\) and the synthetic HCD library from ProteomeTools. \(^{13}\) The number of spectra in these libraries is summarized in Supplementary Table 1. In total, by retaining spectra with the charges of \(1+\) to \(4+\) and from peptides of length no longer than 30, we collected 2,962,341 HCD-MS/MS spectra from 1,048,993 distinct peptides. + +<|ref|>text<|/ref|><|det|>[[92, 572, 884, 689]]<|/det|> +19 The whole data set was randomly split into the two subsets, one subset containing 2,908,323 (95%) spectra as the training data, and the other subset containing the remaining 133,247 (5%) spectra serves as the cross- validation set. We ensured that the training and testing data shared no spectra from the same peptide in order to avoid information leakage. The complete training and testing data sets are available in the supplementary files. + +<|ref|>text<|/ref|><|det|>[[92, 696, 884, 865]]<|/det|> +24 We use the Adam optimizer \(^{31}\) with the learning rate of 0.001 to train the model for 50 epochs (batch size of 64 spectra per GPU). We warm- up the learning rate linearly from 0.0001 to 0.001 within the first 10 epochs for training stability. The complete training process takes around 20 hours using 8 cards of NVIDIA A6000 GPU. We pick the model weight from the epoch that performs best on the cross- validation set after the training finished, which achieved over 0.9 positional accuracy on charge \(2+\) spectra and over 0.78 positional accuracy on charge \(3+\) spectra. More details can be found in the Training performance section of the Supplementary Materials. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[93, 88, 532, 108]]<|/det|> +## 1 Configuration of PointNovo and DeepNovo + +<|ref|>text<|/ref|><|det|>[[92, 121, 884, 242]]<|/det|> +1 Configuration of PointNovo and DeepNovo2 To build the PointNovo and DeepNovo for compassion, we directly used the source code provided by their original publication without modification. For DeepNovo (and also DeepNovo- DIA), we directly used the provided pre- trained weights for all testing tasks. While for PointNovo, as the authors did not provide pre- trained weights, we retrain the model using the intact training dataset provided in the paper. Both PointNovo and DeepNovo were executed by using their default configurations without modifications. + +<|ref|>sub_title<|/ref|><|det|>[[93, 277, 306, 298]]<|/det|> +## 7 Data availability + +<|ref|>text<|/ref|><|det|>[[90, 317, 884, 410]]<|/det|> +7 Data availability8 The proteomic data of used for this study were taken from previous datasets with the identifier PXD019483 and PXD014877. The MaxQuant searching results provided in these studies were directly reused. The trained models of PepNet were deposited at https://zenodo.org/record/5807120 while the experiment results were deposited at https://zenodo.org/record/6003282. + +<|ref|>sub_title<|/ref|><|det|>[[92, 445, 310, 466]]<|/det|> +## 12 Code availability + +<|ref|>text<|/ref|><|det|>[[90, 487, 732, 505]]<|/det|> +12 Code availability13 Source code and scripts are available on GitHub at https://github.com/lkytal/pepnet. + +<|ref|>sub_title<|/ref|><|det|>[[92, 540, 324, 561]]<|/det|> +## 14 Acknowledgement + +<|ref|>text<|/ref|><|det|>[[90, 580, 883, 624]]<|/det|> +14 Acknowledgement15 This work was supported by National Science Foundation (DBI- 2011271), National Institutes of Health (1R01AI143254) and Indiana University Precision Health Initiative (PHI). + +<|ref|>sub_title<|/ref|><|det|>[[91, 660, 351, 681]]<|/det|> +## 17 Competing interests + +<|ref|>text<|/ref|><|det|>[[90, 701, 530, 718]]<|/det|> +18 The authors declare that they have no conflict of interest. + +<|ref|>sub_title<|/ref|><|det|>[[90, 754, 362, 775]]<|/det|> +## 19 Author contributions + +<|ref|>text<|/ref|><|det|>[[88, 795, 884, 864]]<|/det|> +19 Author contributions20 K.L. and H.T. conceived the project. K.L. designed and implemented the deep learning model. K.L., Y.Y., and H.T. analyzed the data. K.L., Y.Y., and H.T. wrote the manuscript. All authors read and revised the manuscript. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[95, 86, 242, 106]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[88, 123, 888, 920]]<|/det|> +(1) Mann, M.; Wilm, M. Error-tolerant identification of peptides in sequence databases by peptide sequence tags. Analytical chemistry 1994, 66, 4390-4399. +(2) Eng, J. K.; McCormack, A. L.; Yates, J. R. An approach to correlate tandem mass spectral data of peptides with amino acid sequences in a protein database. Journal of the american society for mass spectrometry 1994, 5, 976-989. +(3) Hirosawa, M.; Hoshida, M.; Ishikawa, M.; Toya, T. MASCOT: multiple alignment system for protein sequences based on three-way dynamic programming. Bioinformatics 1993, 9, 161-167. +(4) Craig, R.; Beavis, R. C. TANDEM: matching proteins with tandem mass spectra. Bioinformatics 2004, 20, 1466-1467. +(5) Geer, L. Y.; Markey, S. P.; Kowalak, J. A.; Wagner, L.; Xu, M.; Maynard, D. M.; Yang, X.; Shi, W.; Bryant, S. H. Open mass spectrometry search algorithm. Journal of proteome research 2004, 3, 958-964. +(6) Tabb, D. L.; Fernando, C. G.; Chambers, M. C. MyriMatch: highly accurate tandem mass spectral peptide identification by multivariate hypergeometric analysis. Journal of proteome research 2007, 6, 654-661. +(7) Clauser, K. R.; Baker, P.; Burlingame, A. L. Role of accurate mass measurement (±10 ppm) in protein identification strategies employing MS or MS/MS and database searching. Analytical chemistry 1999, 71, 2871-2882. +(8) Chalkley, R. J.; Baker, P. R.; Medzihradszky, K. F.; Lynn, A. J.; Burlingame, A. In-depth analysis of tandem mass spectrometry data from disparate instrument types. Molecular & Cellular Proteomics 2008, 7, 2386-2398. +(9) Kim, S.; Pevzner, P. A. MS-GF+ makes progress towards a universal database search tool for proteomics. Nature communications 2014, 5, 1-10. +(10) Ausloos, P.; Clifton, C.; Lias, S.; Mikaya, A.; Stein, S. E.; Tchekhovskoi, D. V.; Sparkman, O.; Zaikin, V.; Zhu, D. The critical evaluation of a comprehensive mass spectral library. Journal of the American Society for Mass Spectrometry 1999, 10, 287-299. +(11) Yates, J. R.; Morgan, S. F.; Gatlin, C. L.; Griffin, P. R.; Eng, J. K. Method to compare collision-induced dissociation spectra of peptides: potential for library searching and subtractive analysis. Analytical chemistry 1998, 70, 3557-3565. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 88, 886, 135]]<|/det|> +(12) Lam, H.; Deutsch, E. W.; Eddes, J. S.; Eng, J. K.; Stein, S. E.; Aebersold, R. Building consensus spectral libraries for peptide identification in proteomics. Nature methods 2008, 5, 873-875. + +<|ref|>text<|/ref|><|det|>[[90, 150, 884, 219]]<|/det|> +(13) Zolg, D. P.; Wilhelm, M.; Schnatbaum, K.; Zerweck, J.; Knaute, T.; Delanghe, B.; Bailey, D. J.; Gessulat, S.; Ehrlich, H.-C.; Weininger, M., et al. Building ProteomeTools based on a complete synthetic human proteome. Nature methods 2017, 14, 259. + +<|ref|>text<|/ref|><|det|>[[90, 234, 884, 279]]<|/det|> +(14) Craig, R.; Cortens, J.; Fenyo, D.; Beavis, R. C. Using annotated peptide mass spectrum libraries for protein identification. Journal of proteome research 2006, 5, 1843-1849. + +<|ref|>text<|/ref|><|det|>[[90, 294, 884, 364]]<|/det|> +(15) Lam, H.; Deutsch, E. W.; Eddes, J. S.; Eng, J. K.; King, N.; Stein, S. E.; Aebersold, R. Development and validation of a spectral library searching method for peptide identification from MS/MS. Proteomics 2007, 7, 655-667. + +<|ref|>text<|/ref|><|det|>[[90, 379, 884, 425]]<|/det|> +(16) Wang, L.; Liu, K.; Li, S.; Tang, H. A Fast and Memory-Efficient Spectral Library Search Algorithm Using Locality-Sensitive Hashing. Proteomics 2020, 20, 2000002. + +<|ref|>text<|/ref|><|det|>[[90, 440, 884, 485]]<|/det|> +(17) Allmer, J. Algorithms for the de novo sequencing of peptides from tandem mass spectra. Expert review of proteomics 2011, 8, 645-657. + +<|ref|>text<|/ref|><|det|>[[90, 501, 884, 546]]<|/det|> +(18) Dančik, V.; Addona, T. A.; Clauser, K. R.; Vath, J. E.; Pevzner, P. A. De novo peptide sequencing via tandem mass spectrometry. Journal of computational biology 1999, 6, 327-342. + +<|ref|>text<|/ref|><|det|>[[90, 561, 884, 606]]<|/det|> +(19) Chen, G.-m.; Firth, M.; Rui, O. M. The dynamic relation between stock returns, trading volume, and volatility. Financial Review 2001, 36, 153-174. + +<|ref|>text<|/ref|><|det|>[[90, 621, 884, 666]]<|/det|> +(20) Frank, A.; Pevzner, P. PepNovo: de novo peptide sequencing via probabilistic network modeling. Analytical chemistry 2005, 77, 964-973. + +<|ref|>text<|/ref|><|det|>[[90, 682, 884, 752]]<|/det|> +(21) Zhang, J.; Xin, L.; Shan, B.; Chen, W.; Xie, M.; Yuen, D.; Zhang, W.; Zhang, Z.; Lajoie, G. A.; Ma, B. PEAKS DB: de novo sequencing assisted database search for sensitive and accurate peptide identification. Molecular & cellular proteomics 2012, 11. + +<|ref|>text<|/ref|><|det|>[[90, 767, 884, 838]]<|/det|> +(22) Chi, H.; Chen, H.; He, K.; Wu, L.; Yang, B.; Sun, R.-X.; Liu, J.; Zeng, W.-F.; Song, C.-Q.; He, S.-M., et al. pNovo+: de novo peptide sequencing using complementary HCD and ETD tandem mass spectra. Journal of proteome research 2013, 12, 615-625. + +<|ref|>text<|/ref|><|det|>[[90, 853, 884, 898]]<|/det|> +(23) Jeong, K.; Kim, S.; Pevzner, P. A. UniNovo: a universal tool for de novo peptide sequencing. Bioinformatics 2013, 29, 1953-1962. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 88, 886, 135]]<|/det|> +(24) Ma, B. Novor: real-time peptide de novo sequencing software. Journal of the American Society for Mass Spectrometry 2015, 26, 1885-1894. + +<|ref|>text<|/ref|><|det|>[[90, 150, 884, 194]]<|/det|> +(25) Tran, N. H.; Zhang, X.; Xin, L.; Shan, B.; Li, M. De novo peptide sequencing by deep learning. Proceedings of the National Academy of Sciences 2017, 114, 8247-8252. + +<|ref|>text<|/ref|><|det|>[[90, 210, 884, 254]]<|/det|> +(26) Qiao, R.; Tran, N. H.; Xin, L.; Shan, B.; Li, M.; Ghodsi, A. Deepnovov2: Better de novo peptide sequencing with deep learning. arXiv preprint arXiv:1904.08514 2019, + +<|ref|>text<|/ref|><|det|>[[90, 270, 884, 339]]<|/det|> +(27) Tran, N. H.; Qiao, R.; Xin, L.; Chen, X.; Liu, C.; Zhang, X.; Shan, B.; Ghodsi, A.; Li, M. Deep learning enables de novo peptide sequencing from data-independent-acquisition mass spectrometry. Nature methods 2019, 16, 63-66. + +<|ref|>text<|/ref|><|det|>[[90, 354, 884, 425]]<|/det|> +(28) Qiao, R.; Tran, N. H.; Xin, L.; Chen, X.; Li, M.; Shan, B.; Ghodsi, A. Computationally instrument-resolution-independent de novo peptide sequencing for high-resolution devices. Nature Machine Intelligence 2021, 3, 420-425. + +<|ref|>text<|/ref|><|det|>[[90, 440, 884, 485]]<|/det|> +(29) Bai, S.; Kolter, J. Z.; Koltun, V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271 2018, + +<|ref|>text<|/ref|><|det|>[[90, 500, 884, 571]]<|/det|> +(30) Abadi, M.; Barham, P.; Chen, J.; Chen, Z.; Davis, A.; Dean, J.; Devin, M.; Ghemawat, S.; Irving, G.; Isard, M., et al. Tensorflow: A system for large-scale machine learning. 12th {USENIX} Symposium on Operating Systems Design and Implementation ({OSDI} 16). 2016; pp 265-283. + +<|ref|>text<|/ref|><|det|>[[90, 586, 884, 630]]<|/det|> +(31) Kingma, D. P.; Ba, J. Adam: A method for stochastic optimization. International Conference on Learning Representations (ICLR). 2015. + +<|ref|>text<|/ref|><|det|>[[90, 646, 884, 691]]<|/det|> +(32) Cox, J.; Mann, M. MaxQuant enables high peptide identification rates, individualized ppb-range mass accuracies and proteome-wide protein quantification. Nature biotechnology 2008, 26, 1367-1372. + +<|ref|>text<|/ref|><|det|>[[90, 706, 884, 776]]<|/det|> +(33) Müller, J. B.; Geyer, P. E.; Colaço, A. R.; Treit, P. V.; Strauss, M. T.; Oroshi, M.; Doll, S.; Winter, S. V.; Bader, J. M.; Köhler, N., et al. The proteome landscape of the kingdoms of life. Nature 2020, 582, 592-596. + +<|ref|>text<|/ref|><|det|>[[90, 792, 884, 810]]<|/det|> +(34) Consortium, U. UniProt: a hub for protein information. Nucleic acids research 2015, 43, D204-D212. + +<|ref|>text<|/ref|><|det|>[[90, 826, 884, 871]]<|/det|> +(35) Zhao, Y.; Tang, H.; Ye, Y. RAPSearch2: a fast and memory-efficient protein similarity search tool for next-generation sequencing data. Bioinformatics 2012, 28, 125-126. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 88, 884, 130]]<|/det|> +(36) Tran, N. H.; Qiao, R.; Xin, L.; Chen, X.; Liu, C.; Zhang, X.; Shan, B.; Ghodsi, A.; Li, M. Deep learning enables de novo peptide sequencing from data-independent-acquisition mass spectrometry. Nature methods 2019, 16, 63–66. + +<|ref|>text<|/ref|><|det|>[[90, 145, 884, 185]]<|/det|> +(37) Maron, P.-A.; Ranjard, L.; Mougel, C.; Lemanceau, P. Metaproteomics: a new approach for studying functional microbial ecology. Microbial ecology 2007, 53, 486–493. + +<|ref|>text<|/ref|><|det|>[[90, 201, 884, 242]]<|/det|> +(38) Liu, K.; Li, S.; Wang, L.; Ye, Y.; Tang, H. Full-spectrum prediction of peptides tandem mass spectra using deep neural network. Analytical chemistry 2020, 92, 4275–4283. + +<|ref|>text<|/ref|><|det|>[[90, 257, 884, 300]]<|/det|> +(39) Lin, T.-Y.; Dollár, P.; Girshick, R.; He, K.; Hariharan, B.; Belongie, S. Feature pyramid networks for object detection. Proceedings of the IEEE conference on computer vision and pattern recognition. 2017; pp 2117–2125. + +<|ref|>text<|/ref|><|det|>[[90, 314, 884, 360]]<|/det|> +(40) Yang, X.; Neta, P.; Stein, S. E. Extending a Tandem Mass Spectral Library to Include MS 2 Spectra of Fragment Ions Produced In-Source and MS n Spectra. Journal of The American Society for Mass Spectrometry 2017, 28, 2280–2287. + +<|ref|>text<|/ref|><|det|>[[90, 377, 884, 448]]<|/det|> +(41) Wang, M.; Wang, J.; Carver, J.; Pullman, B. S.; Cha, S. W.; Bandeira, N. Assembling the Community-Scale Discoverable Human Proteome. Cell systems 2018, 7, 412–421. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 131, 220, 150]]<|/det|> +Supplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/images_list.json b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..bf43a7f821569baf40d6a58780c5f74ef6a80da7 --- /dev/null +++ b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: NAAA is required to induce hyperalgesic priming.", + "footnote": [], + "bbox": [ + [ + 120, + 115, + 880, + 555 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Role of peripheral NAAA and PPAR-\\(\\alpha\\) signaling in hyperalgesic priming induction.", + "footnote": [], + "bbox": [ + [ + 111, + 352, + 884, + 754 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: NAAA in CD11b+ myeloid cells initiates hyperalgesic priming.", + "footnote": [], + "bbox": [ + [ + 115, + 90, + 880, + 458 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: PPARα in CD11b+ myeloid cells counters hyperalgesic priming.", + "footnote": [], + "bbox": [ + [ + 122, + 235, + 870, + 750 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Monocyte-derived cells, not neutrophils, are required for hyperalgesic priming.", + "footnote": [], + "bbox": [ + [ + 120, + 92, + 880, + 488 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: A role for NAAA-regulated PPAR-\\(\\alpha\\) signaling in hyperalgesic priming.", + "footnote": [], + "bbox": [ + [ + 111, + 496, + 864, + 866 + ] + ], + "page_idx": 27 + } +] \ No newline at end of file diff --git a/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec.mmd b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d267510c255b0970f03860950b492a0c8ca7697c --- /dev/null +++ b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec.mmd @@ -0,0 +1,431 @@ + +# NAAA-regulated lipid signaling in monocyte-derived cells controls the induction of hyperalgesic priming + +Daniele Piomelli + +piomelli@hs.uci.edu + +University of California, Irvine + +Yannick Fotio University of California, Irvine + +Alex Mabou Tagne University of California, Irvine + +Erica Squire + +UC Irvine + +Hye- lim Lee + +UC Irvine + +Faizy Ahmed + +UC Irvine + +Andrew Greenberg + +JM- USDA Human Nutrition Research Center https://orcid.org/0000- 0001- 5062- 387X + +Gilberto Spadoni Univ. of Urbino + +Marco Mor + +University of Parma https://orcid.org/0000- 0003- 0199- 1849 + +## Article + +Keywords: + +Posted Date: August 19th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3263903/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: Yes there is potential Competing Interest. D.P. and M.M. are inventors in patents that protect ARN19702 and other NAAA inhibitors, owned by the University of California, the University of + +<--- Page Split ---> + +Parma, the University of Urbino, and the Fondazione Istituto Italiano di Tecnologia (no. 13/898,225, filed 20 May 2013, published 3 April 2014; no. 62/337,744, filed 17 May 2016, published 23 November 2017). D.P. and Y.F. are inventors in a patent application that protects the algostatic effects of NAAA inhibitors, filed by the University of California (no. 63/166,134, filed 25 March 2021, published 29 September 2022). The other authors have declared no competing interest. + +Version of Record: A version of this preprint was published at Nature Communications on February 24th, 2024. See the published version at https://doi.org/10.1038/s41467-024-46139-5. + +<--- Page Split ---> + +# NAAA-regulated lipid signaling in monocyte-derived cells controls the induction of hyperalgesic priming + +Yannick Fotio \(^{1}\) , Alex Mabou Tagne \(^{1a}\) , Erica Squire \(^{1a}\) , Hye- lim Lee \(^{1}\) , Faizy Ahmed \(^{1}\) , Andrew S. Greenberg \(^{2}\) , Gilberto Spadoni \(^{3}\) , Marco Mor \(^{4}\) , and Daniele Piomelli \(^{1,5,6,8}\) + +\(^{1}\) Department of Anatomy and Neurobiology, University of California Irvine, Irvine, CA, USA. \(^{2}\) Human Nutrition Research Center, Tufts University, Boston, MA, USA \(^{3}\) Dipartimento di Scienze Biomolecolari, Università di Urbino "Carlo Bo," Urbino, Italy. \(^{4}\) Dipartimento di Scienze degli Alimenti e del Farmaco, Università di Parma, Parma, Italy. \(^{5}\) Department of Biological Chemistry, University of California Irvine, Irvine, CA, USA. \(^{6}\) Department of Pharmaceutical Sciences, University of California Irvine, Irvine, CA, USA. + +\(^{a}\) These authors contributed equally. + +\(^{8}\) Lead author: Daniele Piomelli (piomelli@hs.uci.edu). Gillespie Neuroscience Res. Facility (room 3101), 837 Health Sciences Rd., Irvine, CA 92697. + +<--- Page Split ---> + +## Abstract + +Circulating monocytes and tissue- resident macrophages contribute in important ways to the transition from acute to chronic pain but the molecular events that cause their deployment are still unclear. Using a mouse model of hyperalgesic priming (HP), we show now that monocyte- derived cells enable the progression to pain chronicity through a cell- autonomous mechanism that requires the transient activation of the intracellular lipid hydrolase, \(N\) - acylethanolamine acid amidase (NAAA), and the consequent suppression of NAAA- regulated lipid signaling at peroxisome proliferator- activated receptor- \(\alpha\) (PPAR- \(\alpha\) ). Inhibiting peripheral NAAA in the 72 hours following administration of one of three different priming stimuli (interleukin- 6, tumor necrosis factor- \(\alpha\) , or carrageenan), but not before or after this time interval, prevented the emergence of HP in male and female mice. This effect was phenocopied by global NAAA deletion and depended on the recruitment of PPAR- \(\alpha\) by endogenous NAAA substrates. Transgenic mice selectively lacking NAAA in CD11b+ myeloid cells – which include monocytes, macrophages, and neutrophils – were resistant to HP induction. Conversely, mice overexpressing NAAA or lacking PPAR- \(\alpha\) in the same cells were constitutively primed. In vivo ablation of monocytes, but not neutrophils, by administration of clodronate liposomes or blockade of type- 1 colony- stimulating factor receptors, generated mice that were refractory to priming. The results identify NAAA- regulated lipid signaling in monocyte- derived cells as a control node in the induction of HP and the transition to pain chronicity. + +<--- Page Split ---> + +## 49 Introduction + +Chronic pain places an enormous burden on hundreds of millions of people worldwide \(^{1 - 3}\) , yet its neurobiological underpinnings remain largely unknown \(^{4,5}\) . One major challenge is the identification of molecular events that enable acute pain episodes – which typically accompany self- resolving, localized tissue injuries – to progress into persistent painful states that outlast the initial damage and can radiate outside its organ boundaries \(^{4}\) . In addition to neuroplastic changes, whose roles are well established \(^{6,7}\) , activation of the innate immune system has recently emerged as a driving factor in the progression to pain chronicity \(^{8 - 11}\) . Studies in mice have shown, for example, that blood- borne monocytes infiltrate the spinal cord \(^{12}\) and act in synergy with local microglia to promote pain chronication after nerve injury \(^{13,14}\) . Similarly, monocytes and macrophages that express the CX3CR1 receptor contribute to arthritis pain and chemotherapy- induced allodynia by interacting with nociceptive neurons in dorsal root ganglia (DRG) \(^{15,16}\) . Despite these advances, the molecular checkpoints that trigger the deployment of monocytes and macrophages during the transition from acute to chronic pain are still poorly understood \(^{8 - 10}\) . + +An experimental model developed to investigate this transition exploits a form of neuroplasticity in which administration of a proinflammatory substance sensitizes a subset of peripheral nociceptors to subsequent noxious and non- noxious stimuli \(^{17,18}\) . In the report that first described this phenomenon \(^{19}\) , termed 'nociceptive' or 'hyperalgesic priming' (HP), Levine and collaborators showed that the inflammatory reaction produced by intraplantar injection of carrageenan in rats was followed by a prolonged (≥3 weeks) enhancement of the nocifensive response to other proalgesic agents applied at the same site. Further studies by the Levine laboratory identified non- peptidergic isolectin B4+ nociceptors as cellular substrates for HP and showed that its emergence required the recruitment of protein kinase Cε- dependent signal transduction pathways \(^{19 - 21}\) . In addition to its clinical relevance \(^{18}\) , HP has three notable advantages as a model to explore the progression to pain chronicity. First, it can be induced by administering a single inflammatory agent with identified mechanism of action – e.g., interleukin 6 (IL- 6) or tumor- necrosis factor- α (TNF- α) – thus allowing control over a key experimental variable. Second, the initial nocifensive reaction to a priming stimulus can be suppressed by analgesic drugs with no detectable consequences on HP induction, a clear indication that the two processes are temporally overlapping but mechanistically distinct \(^{22,23}\) . Finally, HP unfolds with a distinctive time- course in which the immediate response to the priming trigger is separated from the appearance of the primed state by a 72 h- long interval \(^{23,24}\) . This delay, which has been attributed to axonal and nuclear processes \(^{23,25}\) , identifies three phases of HP which can be studied in relative isolation: 'initiation', which occurs in parallel with the inflammatory reaction to the priming trigger; + +<--- Page Split ---> + +incubation', which follows the resolution of such reaction and precedes the appearance of the primed state; and 'maintenance', which allows such state to persist for several weeks. + +In the present study, we used a combination of genetic and pharmacological strategies to determine whether the intracellular cysteine hydrolase, N- acetylcholamine acid amidase (NAAA) \(^{26}\) , contributes to the emergence of HP. NAAA catalyzes the hydrolytic deactivation of palmitoylethanolamide (PEA) and oleoylethanolamide (OEA), two lipid- derived agonists of the ligand- operated transcription factor, peroxisome proliferator- activated receptor- \(\alpha\) (PPAR- \(\alpha\) ) \(^{27,28}\) . In addition to its well- established roles in the modulation of acute nociception \(^{29}\) and inflammation \(^{30}\) , NAAA- regulated lipid signaling at PPAR- \(\alpha\) has been recently implicated in the induction of persistent pathological nociception by chemical or surgical damage to peripheral tissues \(^{31}\) . Building on those findings, we show now that monocytes and macrophages drive the progression to pain chronicity through a cell- autonomous mechanism that requires the suppression of intracellular NAAA- regulated PPAR- \(\alpha\) signaling during – but not before or after – the incubation phase of HP. The results identify NAAA in monocyte- derived cells as a control node in the development of chronic pain and a possible target for preventive therapy. + +## Materials and Methods + +## Study Approval + +All experimental procedures complied with the ethical regulations for the care and use of laboratory animals promulgated by the National Institutes of Health (NIH) and the International Association for the Study of Pain (IASP). Formal approval was obtained from the Animal Care and Use Committee of the University of California, Irvine. + +## Chemicals + +We purchased prostaglandin \(E_{2}\) (PGE \(_{2}\) ), recombinant mouse interleukin- 6 (IL- 6), and recombinant mouse tumor necrosis factor- alpha (TNF- \(\alpha\) ) from R&D Systems (Minneapolis, MN). Carrageenan, gabapentin, GW7647, GW6471, T0070907, and AM630 were from Sigma- Aldrich (St. Louis, MO). Anandamide, palmitoylethanolamide (PEA), oleoylethanolamide (OEA), their deuterium- containing analogs, and PLX5622 were purchased from Cayman Chemicals (Ann Arbor, MI). Levagen® PEA (a water- soluble PEA formulation) was a kind gift of GE Nutrients (Austin, TX). Liposome- encapsulated clodronate and control liposomes (containing phosphate- buffered saline, + +<--- Page Split ---> + +PBS, \(\mathsf{pH} = 7.4\) ) were from Liposoma (Amsterdam, Netherlands). ARN19702, ARN726, and ARN077 were prepared according to established protocols26. + +## Drug preparation and administration + +Immediately before injections, a stock solution of \(\mathsf{PGE}_2\) (1 mg- mL \(^{- 1}\) in ethanol) was diluted in saline to a final concentration of \(5 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (ethanol: \(0.5\%\) vol/vol). TNF- \(\alpha\) and IL- 6 were dissolved in PBS \(\mathrm{(pH = 7.4)}\) containing fatty acid- free bovine serum albumin (BSA, \(0.5\%\) , Sigma- Aldrich) and were diluted with PBS to reach final concentrations of \(5 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (TNF- \(\alpha\) ) and \(0.025 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (IL- 6). ARN19702, ARN726, GW7647, GW6471, T0070907, and AM630 were dissolved in a mixture of PEG- 400/Tween- 80/sterile saline (15:15:70, vol/vol/vol). Carrageenan, gabapentin, and Levagen® PEA were dissolved in sterile saline. Drugs and vehicles were administered by intraperitoneal (IP) or subcutaneous (SC) injection in a volume of \(10 \mathrm{mL} \cdot \mathrm{kg}^{- 1}\) . Priming stimuli (IL- 6, TNF- \(\alpha\) , carrageenan) and \(\mathsf{PGE}_2\) were administered into the mouse plantar surface using a hypodermic needle (30 gauge) attached to a Hamilton syringe (50 \(\mu \mathrm{L}\) ). Intrathecal injections of ARN726 were performed in mice anesthetized with isoflurane and held by the pelvic girdle. A Hamilton syringe (30 gauge, \(10 \mu \mathrm{L}\) ) was inserted perpendicularly to the skin at the intervertebral space L5- L6 (between the hip bones). A reflexive lateral tail flick indicated a subarachnoid puncture. A volume of \(5 \mu \mathrm{L}\) was delivered and the syringe was rotated and removed before returning the mice to their home cages. + +## Experimental animals + +We used male and female C57BL/6J mice (22- 30 g, 10- 12 weeks old) from Jackson Laboratory (Sacramento, CA) as well as the genetically modified mouse lines described below. All animals were housed in the animal facility of the University of California Irvine, using standard ventilated clear plastic cages (3- 5 per cage) and conventional wood chips bedding. Food and water were available ad libitum. The mice were maintained in a pathogen free- environment on a 12- hr light/dark cycle at controlled temperature (22°C) and humidity (50- 60%). They were randomly assigned to experimental or control groups and behavioral testing was conducted during the light phase of the light/dark cycle. All possible efforts were made to minimize the number of animals used and their discomfort. + +Global NAAA knock- out (Naaa- c) mice + +<--- Page Split ---> + +Mice constitutively lacking the Naaa gene [(Naaa- ), B6N- Atm1BrdNaaatm1a(KOMP)wtsi/WtsiH], were acquired from MRC Harwell (Didcot, UK) through the EMMA-European Mouse Mutants Archive. Their generation and phenotypic profile are reported here31,32. + +\(CD11b^{+}\) myeloid cell- specific NAAA knock- in \((Naaa^{CD11b^{+}})\) mice Transgenic mice overexpressing Naaa in CD11b+ myeloid cells \((Naaa^{CD11b^{+}})\) were produced by GenOway (France). Their generation and phenotypic profile are reported here31,33. + +\(CD11b^{+}\) myeloid cell- specific NAAA knock- out \((Naaa^{CD11b - / - })\) mice NAAA- floxed mice \((Naaa^{f/f})\) mice were generated by mating Naaa- mice with flippase (Flp) FLPo- 10 mice (Jackson Lab, #011065). Subsequent recombination of \(Naaa^{f/f}\) with CD11b- Cre mice (Jackson Lab, #019696)] yielded \(Naaa^{CD11b - / - }\) mice, which were viable and fertile. + +\(CD11b^{+}\) myeloid cell- specific PPAR- \(\alpha\) knock- out \((Ppara^{CD11b - / - })\) mice PPAR- \(\alpha\) floxed \((Ppara^{f/f})\) breeders were a kind gift of Dr Andrew S. Greenberg (Tufts University School of Medicine). \(Ppara^{f/f}\) mice were cross- bred with CD11b- Cre animals (Jackson Lab) to obtain \(Ppara^{CD11b - / - }\) mice, which were viable and fertile. + +## Depletion of monocyte-derived and other cells + +\(C57Bl6/J\) mice were fed ad libitum for 20 days either with normal chow or with chow containing the colony- stimulating factor 1 (CSF1) receptor antagonist PLX5622 (1.2 g- kg- 1 of chow)34. Other groups of C57Bl6/J mice were treated with liposomes containing PBS (control) or clodronate (16 mL- kg- 1, IP), 3 days before and immediately after IL- 6 injection13,35. + +## Fluorescence Activated Cell Sorting (FACS) + +Blood was drawn via cardiac puncture and incubated with ammonium- chloride- potassium (ACK) buffer for erythrocyte lysis. Leukocytes were stained with anti- CD11b, Zombie NIR, anti- Ly6C, and anti- Ly6G antibodies (Table S1) in antibody- staining buffer (0.2% BSA and 0.1% sodium azide in PBS) for 30 min at 4°C. Viable cell populations of interest were collected by FACS and quantified using FCS Express 7 (De Novo Software, Pasadena, CA). Gating strategies are illustrated in Figure S1. First, to separate debris from cells, we gated on the side scatter area (SSC- A) versus the forward scatter area (FSC- A) and selected the cell population with a high FSC- A/SSC- A ratio. This fraction was subjected to forward scatter height (FSC- H) versus FSC- A analysis to isolate single cells that displayed a ratio of \(\sim 1\). The single- cell population was further processed to remove dead cells. Zombie NIR signal- negative cells were collected for the next steps. All FACS processes included these initial gating steps to ensure that only visible single + +<--- Page Split ---> + +cells were analyzed. From this population, we gated cells based on the expression of cellular marker proteins to isolate cell types of interest. To focus on innate immune cells, we selected cells that were positive for CD11b by plotting the viable cell population as CD11b versus FSC- A and selecting the population that was \(\mathrm{CD11b^{+}}\) . This was plotted as Ly6C versus Ly6G with the \(\mathrm{Ly6C^{high}}\) cells being divided into groups depending on Ly6G expression. Monocytes were categorized as 'classical' ( \(\mathrm{CD11b^{+}Ly6C^{high}Ly6G^{low}}\) ) or 'non- classical' ( \(\mathrm{CD11b^{+}Ly6C^{low}Ly6G^{low}}\) ), while neutrophils were defined as \(\mathrm{CD11b^{+}Ly6C^{high}Ly6G^{high}}\) . Expression of the Naaa and Ppara genes in monocytes and neutrophils was quantified by reverse transcription- quantitative polymerase chain reaction (RT- qPCR), as described below, to confirm successful recombination. The spinal cord underwent enzymatic digestion and mechanical dissociation using a brain dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany), following the manufacturer's instructions. Subsequently, the dissociated spinal cord cells were stained with anti- CD11b and anti- CD45 antibodies (Table S1) in antibody- staining buffer (0.2% BSA and 0.1% sodium azide in PBS) for 30 min at \(4^{\circ}\mathrm{C}\) . Viable cell populations were collected by FACS and quantified using FCS Express 7 (De Novo Software, Pasadena, CA). Gating strategies are described above. Microglia was plotted as CD45 versus CD11b, which are both established markers for this cell lineage37. + +## RNA analyses + +CD11b- cells and monocytes were homogenized using QiaShredder spin columns (Qiagen, Germantown, MD) and total RNA was isolated using the RNeasy Micro Kit (Qiagen). Total RNA was isolated from whole blood using the PureLink™ RNA Mini Kit (Invitrogen, Waltham, MA), following manufacturer's instructions. Before purification, samples were rendered genomic DNA free passing isolated RNA through a gDNA Eliminator spin column (Qiagen). RNA concentration and purity were determined using a SpectraMax M5 spectrophotometer (Molecular Devices, San Jose, CA). cDNA was synthesized using 20 ng of total RNA as input for the High- Capacity cDNA RT Kit with RNase Inhibitor (Applied BioSystems, Foster City, CA) with a final reaction volume of 20 µl. First- strand cDNA was amplified using TaqMan™ Universal PCR Master Mix (Invitrogen). RT- PCR primers and fluorogenic probes were purchased from Applied Biosystems [TaqMan(R) Gene Expression Assays]. We used TaqMan gene expression assays for mouse Actb (Mm00607939_s1), Hprt (Mm00446968_m1), Gapdh (Mm99999915_g1), Ppara (Mm00440939_m1), and Naaa (Mm00508965_m1) (Applied Biosystems). RT- PCRs were performed in 96- well plates using CFX96 Real- Time System (Bio- Rad, Hercules, USA). The Bestkeeper software was used to determine expression stability and the geometric mean of three + +<--- Page Split ---> + +different housekeeping genes (Actb, Hprt, and Gapdh). The relative quantity of genes of interest was calculated by the \(2^{-\Delta \Delta \mathrm{ct}}\) method and expressed as fold change over controls. + +## Behavioral experiments + +## HP induction + +We produced HP by administering IL- 6 (0.5 ng, 20 \(\mu \mathrm{l}\) ), TNF- \(\alpha\) (100 ng, 20 \(\mu \mathrm{l}\) ), or carrageenan (1%, 5 \(\mu \mathrm{L}\) ) into the surface of the mouse right hind paw \(^{19,38}\) . HP induction was confirmed 6 days later by injecting \(\mathrm{PGE}_2\) (100 ng, 20 \(\mu \mathrm{L}\) ) at the same site. Lasting ( \(\geq 6\) days) \(\mathrm{PGE}_2\) - induced heat hypersensitivity indicated successful induction \(^{19}\) . + +## Heat hypersensitivity + +We measured thermal nociceptive thresholds using a Hargreaves plantar test apparatus (San Diego Instruments, USA), as previously described \(^{31,39,40}\) . The thresholds were defined as the time (in seconds) at which mice withdrew their paws from the heat stimulus (withdrawal latency). A 15- s cut- off time was applied. Baseline thresholds were assessed in two consecutive readings made 2 days before the administration of priming agents. + +## Nociceptive thresholds + +Nociceptive thresholds were measured using the tail immersion test, as described \(^{41}\) . Briefly, mice habituated to handling were restrained in a soft tissue pocket and tail withdrawal latency was measured by dipping the distal half of the tail into a water bath set at \(54^{\circ}\mathrm{C}\) . Cut- off time was 10 s. Two tail- withdrawal measurements (separated by 30 s) were recorded and averaged. + +## Motor coordination + +Motor coordination was evaluated using an accelerating Rotarod apparatus (Ugo Basile; rod diameter: 2 cm). Briefly, mice were familiarized to the procedure in three consecutive sessions with speed gradually increasing from 4 to 40 rpm over a 5- min period. Experiments were conducted 24 h after the last training session. + +## Pharmacokinetic (PK) and pharmacodynamic experiments + +Male mice (n = 5 per group) were given a single injection of ARN726 (10 mg- kg \(^{- 1}\) , IP) or vehicle and were euthanized under isoflurane anesthesia 15, 30, 60, and 120 min later. Blood (0.5 mL) was collected by cardiac puncture into EDTA- rinsed syringes and transferred into polypropylene + +<--- Page Split ---> + +plastic tubes (1 mL) containing spray-coated K2- EDTA. Plasma was prepared by centrifugation at \(1450 \times g\) at \(4^{\circ}C\) for 15 min and transferred into polypropylene tubes, which were immediately frozen and stored at \(- 80^{\circ}C\) . Spinal cords were removed by hydraulic extrusion, snap- frozen on dry ice, and stored at \(- 80^{\circ}C\) until processed. + +## Analyte extraction + +We extracted PEA, OEA, anandamide, and ARN726 from plasma and spinal cord tissue, as described \(^{40,42,43}\) . Briefly, plasma (0.1 mL) was transferred into 8- mL glass vials, and proteins were precipitated by the addition of 0.5 mL ice- cold acetonitrile containing 1% formic acid and the following internal standards: \([^{2}\mathrm{H}_{4}]\) - PEA (20 nM), \([^{2}\mathrm{H}_{4}]\) - OEA (1 nM), \([^{2}\mathrm{H}_{4}]\) - anandamide (10 nM), and ARN077 (200 nM). Frozen spinal cords were pulverized on dry ice. Spinal tissue samples ( \(\sim 20 \text{mg}\) each) were transferred into 2 ml Precellys CK- 14 soft tissue tubes (Bertin, Rockville, MD) and homogenized in 0.5 mL of ice- cold acetonitrile containing 1% formic acid and the internal standards listed above. Samples were homogenized for 1 min using a Bertin homogenizer at \(4^{\circ}C\) in 15 s/cycle for 2 cycles with 20 s pause between cycles. Samples were stirred vigorously for 30 s and centrifuged again at \(830 \times g\) at \(4^{\circ}C\) for 15 min. The supernatants were loaded onto pre- washed [with water/acetonitrile (1:4 vol/vol)] Enhanced Matrix Removal (EMR)- Lipid cartridges (Agilent Technologies, Santa Clara, CA) and eluted under positive pressure (3- 5 mmHg). Tissue pellets were rinsed with wash solution (0.2 mL), stirred for 30 s, and centrifuged at \(830 \times g\) for 15 min at \(4^{\circ}C\) . The supernatants were collected, transferred onto EMR cartridges, eluted, and pooled with the first eluate. The cartridges were washed again with wash solution (0.2 mL) and pressure was gradually increased to 10 mmHg for maximal analyte recovery. Eluates were dried under a gentle stream of \(N_{2}\) and reconstituted in 0.1 mL methanol containing 0.1% formic acid. Samples were transferred to deactivated glass inserts (0.2 mL) placed inside amber glass vials (2 mL, Agilent Technologies). + +## Lipid quantification + +PEA, OEA, and anandamide were fractionated using a 1260 series LC system (Agilent Technologies) consisting of a binary pump, degasser, thermostated autosampler, and column compartment coupled to a 6410B triple quadrupole mass spectrometric detector (MSD; Agilent Technologies). Analytes were separated on an Eclipse PAH column (1.8 \(\mu \text{m}\) , 2.1 x 50 mm; Agilent Technologies) with a mobile phase consisting of 0.1% formic acid in water as solvent A and 0.1% formic acid in methanol as solvent B. A linear gradient was used: 75.0% B at 0 time to 80.0% B in 5.0 min, change to 95% B at 5.01 min continuing to 6.0 min, change to 75% B at 6.01 min, and hold till 8.0 min for column re- equilibration and stop time. The column temperature was maintained + +<--- Page Split ---> + +at \(45^{\circ}C\) and the autosampler temperature at \(10^{\circ}C\) . The injection volume was \(2 \mu \mathrm{L}\) , the flow rate was \(0.3 \mathrm{mL} / \mathrm{min}\) , and the total analysis time was \(15.5 \mathrm{min}\) . To prevent carryover, the injection needle was washed three times in the autosampler port for \(30 \mathrm{s}\) before each injection, using a wash solution consisting of \(10\%\) acetone in water/methanol/isopropanol/acetonitrile \((1:1:1:1, \mathrm{vol})\) . The MSD was operated in the positive electrospray ionization (ESI) mode. Analytes were quantified by multiple reaction monitoring (MRM) of the following transitions: anandamide \(= m / z\) \(348.3 > 62.2\) , \([^{2}\mathrm{H}_{4}]\) - anandamide \(= m / z\) \(352.3 > 66.2\) , PEA \(= m / z\) \(300.3 > 62.2\) , \([^{2}\mathrm{H}_{4}]\) - PEA \(= m / z\) \(304.3 > 66.2\) , OEA \(= m / z\) \(326.3 > 62.1\) , \([^{2}\mathrm{H}_{4}]\) - OEA \(= m / z\) \(330.3 > 66.1\) . Capillary and nozzle voltages were 3,000 and 1,900 V, respectively. The drying gas temperature was \(300^{\circ}C\) with a flow of \(5 \mathrm{mL} / \mathrm{min}\) . The sheath gas temperature was \(300^{\circ}C\) with a flow of \(12 \mathrm{mL} / \mathrm{h}\) . Nebulizer pressure was set at \(40 \mathrm{psi}\) . We used the MassHunter software (Agilent Technologies) for instrument control, data acquisition, and analysis. + +## ARN726 quantification + +ARN726 was fractionated using an Eclipse Plus \(\mathrm{C_{18}}\) column \([1.8 \mu \mathrm{m}, 2.1 \times 30 \mathrm{mm}]\) (Agilent Technologies)] with a mobile phase of \(0.1\%\) formic acid in water as solvent A and \(0.1\%\) formic acid in methanol as solvent B. A linear gradient was used: start \(60\%\) B to \(80\%\) B in \(2.5 \mathrm{min}\) , changed to \(95\%\) B at \(2.51 \mathrm{min}\) and continued till \(3.0 \mathrm{min}\) , changed to \(60\%\) B at \(3.01 \mathrm{min}\) and continued till \(5.0 \mathrm{min}\) . The flow rate was \(0.3 \mathrm{mL} / \mathrm{min}\) till \(3.0 \mathrm{min}\) and then \(0.4 \mathrm{mL} / \mathrm{min}\) from \(3.01 \mathrm{min}\) to \(5.0 \mathrm{min}\) . The column temperature was maintained at \(40^{\circ}C\) and the auto- sampler temperature was \(9^{\circ}C\) . The injection volume was \(2 \mu \mathrm{L}\) . The MS was operated in positive mode. ARN726 was quantified by MRM using ARN077 as an internal standard and the following transitions: \(\mathrm{ARN}726 = m / z 269.19 > 83.1\) , \(\mathrm{ARN}077 = m / z 292.16 > 74.1\) . Capillary and nozzle voltages were \(1800\) and \(1200 \mathrm{V}\) , respectively. Nebulizer pressure was set at \(10 \mathrm{psi}\) . Drying gas temperature was \(300^{\circ}C\) , with a flow of \(12 \mathrm{L} / \mathrm{min}\) . Sheath gas temperature was \(320^{\circ}C\) with a flow of \(10 \mathrm{L} / \mathrm{min}\) . + +## PK data analysis + +Results from PK experiments were analyzed using a noncompartmental model44. Maximal concentration \((C_{max})\) and time at which maximal concentration was reached \((T_{max})\) were determined by visual inspection of averaged data. Other PK parameters – including area under the curve \((AUC)\) , elimination constant \((K_{el})\) , and half- life time of elimination \((t_{1 / 2})\) – were determined using the equations described here44. + +## Statistical analyses + +<--- Page Split ---> + +Sample size for behavioral studies was predetermined by power analysis ( \(\alpha = 0.05\) , 1- \(\beta = 0.8\) , effect size \(\sim 35\%\) , \(n \geq 7\) per group). Data analyses were performed using GraphPad Prism version 9.1 (La Jolla, CA) under blinded conditions. All data are expressed as means \(\pm\) SEM. Differences between groups were assessed by either unpaired two- tailed Student's \(t\) test or analysis of variance (ANOVA) (one- way or two- way) followed by Dunnett's or Bonferroni's post hoc test unless otherwise noted. A \(P < 0.05\) was considered statistically significant. + +## Results + +## NAAA is required for the induction of HP + +Following an established protocol \(^{19,38}\) , we evoked HP in male mice by intraplantar (i.pl.) injection of IL- 6 (0.5 ng) (Fig. 1A). As expected, the cytokine produced a state of ipsilateral heat hypersensitivity that lasted \(< 24h\) and was followed by a persistent enhancement of the nocifensive response to the proalgesic eicosanoid \(\mathrm{PGE}_2\) (100 ng, i.pl., administered 6 days later in the same paw). The initial hypersensitivity was attenuated when mice were given, 2 h before IL- 6, either the selective NAAA inhibitor ARN19702 (30 mg- kg \(^{- 1}\) , IP) \(^{45,46}\) (Fig. 1B1) or gabapentin (50 mg- kg \(^{- 1}\) , IP) (Fig. 1C1). Even though they exerted acute antinociceptive effects, the two agents did not prevent HP induction, as shown by their inability to normalize the response to \(\mathrm{PGE}_2\) in IL- 6- primed mice (Fig. 1B2, C2). The results, which were replicated in female animals (Fig. S2A- C), confirm prior studies indicating that the inflammatory reaction to a priming stimulus and the initiation of HP overlap in time but are governed by distinct mechanisms \(^{22,23}\) . + +As the initiation and maintenance phases of HP are separated by an incubation period of \(\sim 72\) \(h^{23,24}\) , we asked whether administering ARN19702 or gabapentin during this time interval might affect HP progression. We gave once- daily injections of ARN19702 (30 mg- kg \(^{- 1}\) , IP), gabapentin (50 mg- kg \(^{- 1}\) , IP) or their vehicles to male and female mice on the first, second, and third day after IL- 6 (Fig. 1A). When tested using this protocol, ARN19702 blocked HP induction (Fig. 1D, data from female mice are shown in Fig. S2D1, D2) whereas gabapentin did not (Fig. S3). The same ARN19702 regimen was also protective in male mice challenged with either of two different priming triggers, TNF- \(\alpha\) (100 ng, i.pl.) (Fig. S4A, B) and carrageenan (1%, i.pl.) (Fig. S4C). Underscoring the strict time- dependence of its effects, ARN19702 failed to prevent HP when administered in male mice for shorter time intervals (e.g., only on days 1 and 2 after IL- 6) or after the end of the incubation period (e.g., on days 4 and 5 after IL- 6) (Fig. 1E). + +To further probe NAAA's role in priming induction, we studied the response to IL- 6 in male mice constitutively lacking the enzyme. Previous studies have shown that inflammatory and acute + +<--- Page Split ---> + +nociceptive responses are blunted in homozygous Naaa- mice31,32. Consistent with those findings, the mutants' acute reaction to IL- 6 was weaker than that of their wild- type littermates (Fig. 1F1). Naaa- mice exhibited a normal acute response to carrageenan, however, possibly owing to the greater intensity of this stimulus (Fig. S4D1). Furthermore, as seen with ARN19702 treatment, Naaa- mice failed to develop HP after injection of IL- 6 (Fig. 1F2) or carrageenan (Fig. S4D2). Heterozygous Naaa- mice also showed a diminished inflammatory reaction to IL- 6 (Fig. 1F1) but not to carrageenan (Fig. S4D1) and were unable to progress to HP after administration of either stimulus (Fig. 1F2). The findings indicate that inflammatory agents induce HP through a mechanism that requires NAAA activity. Such requirement is (i) independent of NAAA's facilitatory role in the nociceptive response evoked by the agents; (ii) restricted to the 72- h incubation period that precedes the emergence of HP; and (iii) sexually monomorphic. + +## NAAA initiates HP by suppressing PPAR- \(\alpha\) signaling + +Activation of the nuclear receptor PPAR- \(\alpha\) by the endogenous NAAA substrates, PEA and OEA, underpins the anti- inflammatory and anti- nociceptive properties of NAAA inhibitors32,46 as well as the agents' ability to stop the transition to pain chronicity after sciatic nerve ligation or i.pl. formalin injection31. To determine whether a similar mechanism might be involved in halting HP induction, we challenged male mice with IL- 6 (0.5 ng, i.pl.) and treated them once daily for the subsequent three days with either ARN19702 (30 mg- kg \(^{- 1}\) , IP) or a combination of ARN19702 and a maximally effective dose of the selective PPAR- \(\alpha\) antagonist GW6471 (4 mg- kg \(^{- 1}\) , IP)31,46,47. When administered alone, ARN19702 prevented HP induction, and this effect was abolished by addition of GW6471 (Fig. 2A). By contrast, co- administration of ARN19702 with maximally effective doses of selective antagonists for PPAR- \(\gamma\) (T0070907, 1 mg- kg \(^{- 1}\) , IP)48 or CB2 cannabinoid receptors (AM630, 3 mg- kg \(^{- 1}\) , IP)49, which have been implicated in some actions of PEA50,51, did not significantly alter the response to ARN19702 (Fig. 2B, C). Further support for a role of PPAR- \(\alpha\) in mediating such response was provided by the finding that PEA (30 mg- kg \(^{- 1}\) , subcutaneous) and the synthetic PPAR- \(\alpha\) agonist GW7647 (10 mg- kg \(^{- 1}\) , IP) prevented priming when administered on days 1- 3 post IL- 6 (Fig. 2D, E). The results suggest that inhibiting NAAA activity during the incubation phase of HP stops the progression to pain chronicity by heightening PPAR- \(\alpha\) signaling. + +## Role of peripheral NAAA in HP induction + +Microglia and neurons of the central nervous system (CNS) express NAAA26,33,52. To assess whether the development of HP might involve a central pool of the enzyme, we utilized the \(\beta\) - + +<--- Page Split ---> + +lactam- based NAAA inhibitor ARN726 (Fig. 2F), which is potent and selective but has limited CNS activity \(^{53}\) . Fifteen minutes after systemic administration, ARN726 (10 mg- kg \(^{- 1}\) , IP) reached peak concentrations of 333.8 ng- mL \(^{- 1}\) in plasma and 42.7 ng- mg \(^{- 1}\) in spinal cord (Fig. 2F, other PK properties of ARN726 are reported in Table S1). The tissue- to- plasma ratio calculated from these concentrations \((= 0.13)\) is \(\sim 56\%\) lower than the brain- to- plasma ratio achieved by ARN19702 after oral administration of a 3 mg- kg \(^{- 1}\) dose \((= 0.30)^{45}\) . Concomitant with its plasma peak, ARN726 caused a transitory but substantial elevation of PEA and OEA levels in circulation (Fig. 2G, H) but had no such effect in spinal cord (Fig. 2J, K). Moreover, confirming its selectivity for NAAA \(^{53}\) , ARN726 did not affect the circulating levels of anandamide (Fig. 2I, L), an endocannabinoid mediator that is structurally related to PEA and OEA but is hydrolyzed by a different lipid amidase \(^{26}\) . Despite this lack of CNS activity, systemic administration of ARN726 (1- 30 mg- kg \(^{- 1}\) , IP) on days 1- 3 after IL- 6 blocked the development of priming with considerable potency (ED \(_{50} = 2.2\) mg- kg \(^{- 1}\) ) (Fig. 2M, Fig. S5). By contrast, such effect was observed when ARN726 (30 ng) was delivered directly to the spinal cord via intrathecal infusion (Fig. 2N). We conclude that HP induction requires peripheral NAAA activity. + +## NAAA in CD11b \(^+\) myeloid cells initiates HP induction + +Outside the CNS, NAAA is predominantly expressed in monocyte- derived cells and B- lymphocytes \(^{53,54}\) . Because monocytes and macrophages have been implicated in the progression to pain chronicity \(^{12,13}\) , we examined whether NAAA- regulated lipid signaling in this cell population might contribute to HP. We generated mutant mice that selectively lack NAAA in CD11b \(^+\) cells (Fig. 3A), which include monocytes, macrophages, and neutrophils \(^{55,56}\) . Successful recombination was confirmed by subjecting blood samples from the newly generated mouse line to fluorescence- activated cell sorting (FACS) coupled to RT- qPCR analysis (Fig. 3B, C). Naaa \(^{CD11b / - }\) mice were viable, fertile, and had normative nociceptive thresholds and motor coordination (Fig. S6A, B). When compared to Naaa \(^{fl / fl}\) littermates, which were used as controls, Naaa \(^{CD11b / - }\) mice phenocopied global Naaa knockouts in that they exhibited an attenuated acute reaction to IL- 6 (Fig. 3D) and were resistant to priming (Fig. 3E). The results suggest that NAAA- expressing CD11b \(^+\) myeloid cells participate in both the initial hypersensitivity evoked by IL- 6 and the emergence of HP but are indispensable only for the latter. Experiments with transgenic mice overexpressing NAAA in CD11b \(^+\) cells further strengthened this conclusion. Prior work suggests that resident lung macrophages in Naaa \(^{CD11b + }\) mice are inherently hyperactive \(^{33}\) , raising the possibility that NAAA overexpression in myeloid cells might promote a constitutively primed phenotype. Consistent with this prediction, the nocifensive response to PGE \(_{2}\) was heightened in + +<--- Page Split ---> + +male Naaa \(^{CD11b^{+}}\) mice that had not been previously exposed to IL- 6, but not in their wild- type littermates (Fig. 3F, G). Together, the findings indicate that NAAA activity in peripheral CD11b \(^{+}\) myeloid cells is both necessary and sufficient for the induction of HP. + +## PPAR- \(\alpha\) in CD11b \(^{+}\) myeloid cells suppresses HP induction + +To further probe the role of lipid- dependent PPAR- \(\alpha\) signaling in the initiation of HP, we generated mice that lack this nuclear receptor in CD11b \(^{+}\) cells (Fig. 4A- C), with the expectation that, like mutants overexpressing NAAA in the same cells, they would also be constitutively primed. Confirming this prediction, the nocifensive response to \(\mathrm{PGE}_{2}\) was enhanced in male IL- 6- naive \(Ppara^{CD11b - / - }\) mice, relative to their \(Ppara^{fl / fl}\) littermates (Fig. 4D, E). Of note, PPAR- \(\alpha\) deletion in CD11b \(^{+}\) cells not only sensitized mice to \(\mathrm{PGE}_{2}\) but also annulled the protective effect caused by administration of ARN19702 during HP incubation (Fig. 4F- I). The results suggest that NAAA- regulated PPAR- \(\alpha\) signaling in peripheral CD11b \(^{+}\) myeloid cells controls the transition to HP, most likely through a cell- autonomous mechanism, and is the primary target for the protective effect of NAAA inhibitors. + +## Monocytes and macrophages, but not neutrophils, are required for HP induction + +Lastly, we used two mechanistically distinct cell depletion strategies as an independent means to assess the role of monocytes and macrophages in HP induction. First, we treated wild- type mice with liposome- encapsulated clodronate, a peripherally restricted bisphosphonate- containing compound that selectively induces apoptosis in monocyte- derived cells \(^{57 - 59}\) . Male wild- type mice were given, with a 3- day interval, two IP injections of liposomes loaded with either clodronate or its vehicle (PBS) (Fig. 5A) \(^{13}\) . FACS analyses of blood and spinal cord showed that, compared to PBS liposomes, clodronate liposomes lowered the number of circulating Ly6C \(^{\text{high}}\) ('classical') monocytes without affecting Ly6C \(^{\text{low}}\) ('non- classical') monocytes or spinal cord microglia (Fig. 5B, C, Fig. S7A). Consistent with prior results \(^{60}\) , the number of neutrophils was increased by clodronate (Fig. 5D), likely owing to monocyte death \(^{61}\) . Separate groups of liposome- treated mice were exposed to IL- 6 (0.5 ng, i.p.l.) and, six days later, were challenged with \(\mathrm{PGE}_{2}\) (100 ng, i.p.l.) (Fig. 5E). As expected \(^{62}\) , administration of clodronate liposomes suppressed the initial hypersensitivity evoked by IL- 6 (Fig. 5F1). In addition to this acute effect, clodronate liposomes normalized the response to \(\mathrm{PGE}_{2}\) (Fig. 5F2), indicating that priming induction was also blocked. In a second set of experiments, we exposed for three weeks male mice to chow containing either the CSF1 receptor antagonist PLX5622 (1.2 g- kg \(^{- 1}\) chow) or its vehicle (Fig. 5G). PLX5622 treatment reduced the numbers of circulating monocytes (both 'classical' and 'non- classical'), + +<--- Page Split ---> + +neutrophils (Fig. 5H- J), and spinal cord microglia (Fig. S7B) \(^{63,64}\) , but caused no observable change in baseline nociceptive thresholds or motor coordination (Fig. S8). Closely matching the results obtained with clodronate, treatment with PLX5622 (Fig. 5K) prevented both the acute response to IL- 6 (Fig. 5L1) and the emergence of HP (Fig. 5L2). In sum, clodronate and PLX5622 lowered monocyte numbers and prevented priming but had opposite effects on neutrophil number. We interpret these findings as indicating that monocytes and/or macrophages, but not neutrophils, are required for the induction of HP. + +## Discussion + +The results identify NAAA- regulated signaling at PPAR- \(\alpha\) as a molecular switch that directs monocytes and macrophages to initiate HP in mice exposed to an inflammatory stimulus. Three independent lines of evidence support this conclusion. First, inhibiting NAAA or activating PPAR- \(\alpha\) during the incubation phase of HP stops its consolidation; global NAAA deletion has a similar effect. Second, mutant mice that selectively lack NAAA in CD11b \(^+\) cells – which include monocytes, macrophages, and neutrophils – are resistant to priming, whereas mice that overexpress NAAA or lack PPAR- \(\alpha\) in the same cell lineage are constitutively primed. Lastly, depletion of monocytes and macrophages, but not neutrophils, renders mice refractory to priming. + +The model posits that, in tissue- resident macrophages of non- primed mice, high baseline concentrations of NAAA main substrates (PEA, OEA) maintain these cells in a resting (surveillant) state by constitutively activating PPAR- \(\alpha\) - dependent transcription. In non- primed animals, PGE2 produces a short- lived ( \(\sim 2h\) ) nocifensive response by ligating protein kinase A- coupled EP2/EP4 receptors in non- peptidergic isolectin- B4 \(^+\) nociceptive neurons \(^{17,18,21}\) (Fig. 6, left panel). Priming agents such as IL- 6 have two pharmacologically distinguishable effects: on the one hand, they evoke a self- resolving inflammatory reaction whose sensory signs can be suppressed by standard analgesic drugs \(^{22,23}\) ; on the other, they initiate priming through an as- yet- unknown mechanism that is insensitive to such drugs. This initial phase is followed by a quiescent 72h- long interval that precedes stabilization of the primed state \(^{23,24}\) (Fig. 6, center panel). During this incubation period, NAAA recruitment is both necessary and sufficient to enable the functional interaction of monocytes and macrophages with non- peptidergic nociceptors. Based on our data, the neuroplastic modifications that support the maintenance of a primed state in these neurons – including the recruitment of protein kinase Cε- dependent transduction pathways \(^{19- 21}\) and the local translation and retrograde axonal transport of CREB \(^{25}\) – depend on such interaction (Fig. 6, center + +<--- Page Split ---> + +and right panels). The model emphasizes three points: first, the role of NAAA- regulated PPAR- \(\alpha\) signaling in controlling macrophage and monocyte activation; second, the cooperation between these cells and first- order nociceptors in the induction of priming; third, the subdivision of this process into mechanistically distinct components. These points are briefly discussed below. + +The regulatory functions of PPAR- \(\alpha\) in monocyte- derived cells have been the object of intensive investigation. Studies have shown that, when ligand- bound, PPAR- \(\alpha\) suppresses inflammatory signaling in macrophages \(^{65,66}\) and promotes their polarization toward an M2 phenotype associated with tissue protection \(^{67,68}\) . There is also strong evidence that NAAA's main substrates, PEA and OEA \(^{26}\) , are the ligands responsible for these effects. First, both substances are potent PPAR- \(\alpha\) agonists \(^{27,29,69}\) . Second, monocytes and macrophages express the zinc hydrolase N- acyl phosphatidylethanolamine phospholipase D (NAPE- PLD), which constitutively generates PEA and OEA \(^{70,71}\) , as well as NAAA, which deactivates them \(^{26}\) . Third, inflammatory stimuli suppress NAPE- PLD expression (via transcriptional regulation) \(^{72}\) and enhance NAAA activity (via autoproteolysis of its inactive polypeptide precursor) \(^{73}\) , causing an overall reduction in the amount of PEA and OEA available for signaling \(^{30,32,74}\) . Finally, NAAA inhibitors exert profound anti- inflammatory and anti- nociceptive effects by restoring baseline PEA and OEA levels and, consequently, enhancing PPAR- \(\alpha\) - dependent transcription \(^{30,32,46,75}\) . In sum, the available evidence supports the possibility that NAAA- dependent interruption of PPAR- \(\alpha\) signaling instructs monocytes and macrophages to initiate priming. Identifying the molecular modifications that codify these instructions is an important area for future research. + +Neuroimmune interactions are critical to the establishment of persistent painful states \(^{10,15,76}\) . For example, experiments in mouse models of arthritis and vincristine- induced toxicity have shown that the infiltration of blood- derived CXCR1- expressing monocytes into the DRG drives initiation and maintenance of persistent pathological nociception \(^{15,16}\) . Similarly, after nerve injury in mice, blood- borne monocytes migrate to the spinal cord, where they proliferate, differentiate, and act in synergy with microglia to initiate pain chonification \(^{12- 14}\) . The present study expands this body of knowledge in two ways. First, it shows that monocytes and macrophages, but not neutrophils, are necessary for priming induction. This conclusion is based on the finding that clodronate and PLX5622 exert opposite effects on neutrophil numbers yet both deplete monocyte- derived cells and prevent priming. However, the results do not exclude the possibility that neutrophils, which promote widespread persistent nociception in a mouse model of fibromyalgia \(^{77}\) , might play an ancillary role in HP. Second, and more importantly, our findings identify NAAA as a molecular switch that, when engaged, directs monocyte- derived cells to interact with first- order nociceptors + +<--- Page Split ---> + +and enable priming19- 21. The precise nature of this interaction is unknown but is likely to involve the release of chemical signals from monocytes and/or macrophages. Possible candidates for this role, which should be evaluated in future work, include lipid mediators such as sphingosine- 1- phosphate78,79, reactive oxygen species such as hydrogen peroxide15,80, and growth factors81,82. Chronic pain states are widely heterogeneous in causes, symptoms, impact on function, and temporal development3. This diversity justifies skepticism toward a simplistic view of the progression to pain chronicity as transformation of one mechanistic type of pain (e.g., acute pain associated with injury) into another (e.g., neuropathic pain)83. Nevertheless, our results do support the existence of at least one transition point which must be passed in order for lasting pathological nociception to be established. In previous work, we found that inhibiting spinal cord NAAA 48 to 72 hours after peripheral tissue damage stops the emergence of lasting pathological nociception in mice31. Similarly, in the present study, we showed that disabling peripheral NAAA in the 72 hours following administration of an inflammatory stimulus prevents priming development. In both cases, the protective effects of NAAA blockade depend on PPAR- \(\alpha\) activation, are sexually monomorphic, and are strictly restricted to a three- day time window post- injury. The last property prompted us to designate such effects with the term 'algostatic' (ancient Greek ἀλγος 'pain' and ἰστάντοι 'to stop')31 to underscore the fact that they are independent of the ability of NAAA inhibitors to attenuate ongoing nociception or inflammation. The same term may be applicable to other drugs, such as metformin, that are also able to stop the development of persistent pathological nociception in animal models84- 86. If effective in the clinic, algostatic agents might offer a strategy to prevent chronic pain after invasive surgery and other kinds of physical trauma. + +## Acknowledgments + +We are grateful to Parwinder Singh Uppal, Kayla Chang and Vitaly Perkov for their help with behavioral studies and to GE Nutrients for the kind gift of Levegan+ PEA. + +## Competing interest + +D.P. and M.M. are inventors in patents that protect ARN19702 and other NAAA inhibitors, owned by the University of California, the University of Parma, the University of Urbino, and the Fondazione Istituto Italiano di Tecnologia (no. 13/898,225, filed 20 May 2013, published 3 April 2014; no. 62/337,744, filed 17 May 2016, published 23 November 2017). D.P. and Y.F. are inventors in a patent application that protects the algostatic effects of NAAA inhibitors, filed by the University of California (no. 63/166,134, filed 25 March 2021, published 29 September 2022). The other authors have declared no competing interest. + +<--- Page Split ---> + +## Data availability + +All data supporting the findings of this study are readily available within the paper and its supplementary files. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: NAAA is required to induce hyperalgesic priming.
+ +(A) Model used in the present study. In most experiments, priming was induced in mice by intrapalantar administration of IL-6 and was assessed 6 days later by injecting \(\mathrm{PGE}_2\) at the same site. Ipsilateral heat hypersensitivity (paw withdrawal latency, seconds) was measured under baseline (BL) conditions (-2d, -1d) and immediately after administration of IL-6 (0d), \(\mathrm{PGE}_2\) (6d), or vehicle (0d and 6d). The blue bar marks the incubation period for priming. +(B, C) Self-resolving heat hypersensitivity elicited by IL-6 in mice that had received 2 h earlier (B1) ARN19702 (ARN, 30 mg-kg\(^{-1}\); green circles), (C1) gabapentin (GBP, 50 mg-kg\(^{-1}\); green triangles), or their vehicles (Veh, magenta circles). (B2, C2) Effects of \(\mathrm{PGE}_2\) in IL-6-primed mice treated with ARN19702 (B2), gabapentin (C2), or their vehicles. +(D) IL-6-primed mice were treated on 1d-3d with vehicle or ARN19702 (30 mg-kg\(^{-1}\)). (D1) Response to IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). +(D2) Effects of \(\mathrm{PGE}_2\) in IL-6-primed mice that had received vehicle (magenta circles) or ARN19702 (green triangles). Right panel, area under the curve (AUC). + +<--- Page Split ---> + +(E) IL-6-primed mice were treated with vehicle (magenta circles) or ARN19702 (30 mg·kg-1, filled triangles) for 2 days on 1d-2d, 3d-4d, or 5d-6d, and the response to PGE2 was assessed 3 days later. ARN19702 on (E1) 1d-2d, (E2) 3d-4d, and (E3) 5d-6d. + +(F) Effects of (F1) IL-6 (0d) and (F2) PGE2 (6d) in wild-type (Wt, magenta circles), heterozygous Naaa+/− (green triangles), and homozygous Naaa+/− mice (green squares). + +Data are presented as mean ± S.E.M. and were analyzed by two-way repeated measure ANOVA followed by Bonferroni's multiple comparison test, when necessary. \(^{***}P < 0.001\) , \(^{**}P < 0.01\) , \(^{*}P < 0.05\) , versus baseline; \(^{###}P < 0.001\) , \(^{##}P < 0.01\) , \(^{#}P < 0.05\) , versus vehicle or Wt. ns, nonsignificant (n = 8-16 mice per group). Dotted lines indicate baseline withdrawal latency. + +![](images/Figure_2.jpg) + +
Figure 2: Role of peripheral NAAA and PPAR-\(\alpha\) signaling in hyperalgesic priming induction.
+ +(A-C) IL-6- primed mice were treated on 1d-3d with ARN19702 (ARN, 30 mg·kg- 1) alone or in combination with GW6471 (4 mg·kg- 1), T0070907 (1 mg·kg- 1), AM630 (3 mg·kg- 1), or an appropriate vehicle. Effects of PGE2 in mice treated with the following: (A) ARN19702 alone (green triangles), ARN19702 plus GW6471 (open squares), or vehicle (Veh, magenta circles); (B) + +<--- Page Split ---> + +ARN19702 alone (green triangles), ARN19702 plus T0070907 (open squares), or vehicle (magenta circles); and (C) ARN19702 alone (green triangles), ARN19702 plus AM630 (open squares), or vehicle (magenta circles). + +squares), or vehicle (magenta circles). + +(D, E) IL- 6- primed mice were treated on 1d- 3d with PEA (30 mg- kg \(^{- 1}\) ) or GW7647 (10 mg- kg \(^{- 1}\) ). + +Effects of \(\mathsf{PG E}_{2}\) in mice treated with (D) PEA (green circles) or vehicle (magenta circles); and (E) GW7647 (green circles) or vehicle (magenta circles). + +(F) Time- course of ARN726 concentrations in plasma (green circles) and spinal cord (gray squares) after IP administration (10 mg- kg \(^{-1}\) ). Top, chemical structure of ARN726. Right panel, area under the curve (AUC). + +(G- I) Effect of ARN726 (10 mg- kg \(^{- 1}\) ) on plasma concentrations of (G) PEA, (H) OEA, and (I) anandamide (AEA). Gray bars, baseline plasma concentrations. + +(J- L) Effect of ARN726 (10 mg- kg \(^{- 1}\) ) on the spinal cord concentrations of PEA (J), OEA (K), and AEA (L). Open bars, baseline analyte concentrations. + +(M) IL- 6- primed mice were treated on 1d- 3d with ARN726 (1- 30 mg- kg \(^{- 1}\) ) and the response to \(\mathsf{PG E}_{2}\) was assessed on 6d. % MPE, percent of maximal possible effect. + +(N) Effects of intrathecal vehicle (magenta circles) or ARN726 (30 ng, administered on 1d and 3d) (green circles) on the response to \(\mathsf{PG E}_{2}\) in IL- 6 primed mice. + +Data are presented as mean ± S.E.M. and were analyzed using the Student's \(t\) test (F, AUC), one- way ANOVA (G- L), or two- way repeated measure ANOVA (A- F, N). Dunnett's or Bonferroni's post hoc test was applied as needed. ###P < 0.001, ##P < 0.01, #P < 0.05 versus vehicle. ***P < 0.001, **P < 0.01 versus baseline (time 0). ns, non- significant (n = 5- 8 mice per group). Dotted lines indicate baseline withdrawal latency. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: NAAA in CD11b+ myeloid cells initiates hyperalgesic priming.
+ +(A) Generation of NaaaCD11b-/- mice. Cross-breeding of global Naaa-/- mice with flippase (fl) FLPo-10 mice produced Naaafl/fl offspring, which was mated with CD11b-Cre mice to yield the NaaaCD11b-/- line. Yellow boxes: Naaa coding sequences; green boxes, Flp recognition target (FRT) sites; gray triangles: loxP sites flanking the deleted Naaa sequence. +(B) FACS tracings showing isolation of monocytes from blood samples of (top panels) Naaafl/fl and (bottom panels) NaaaCD11b-/- mice. Left, isolation of CD11b+ cells; right, isolation of Ly6C+ monocytes. The square denotes the Ly6Chigh ('classical') monocyte subpopulation collected for RT-qPCR analysis. 1, Ly6Clow ('non-classical') monocytes; 2, neutrophils. +(C) Naaa mRNA levels (assessed by RT-qPCR) in FACS-isolated CD11b- cells and Ly6Chigh monocytes from NaaaCD11b-/- mice (green circles) and Naaafl/fl (magenta circles). Whole blood is shown for comparison. Bars represent the average of 3 biological replicates. +(D) Self-resolving heat hypersensitivity elicited by IL-6 in NaaaCD11b-/- mice (green circles) and control Naaafl/fl mice (magenta circles). +(E) Effects of PGE2 in IL-6-primed NaaaCD11b-/- mice (green circles) and control Naaafl/fl mice (magenta circles). +(F) Normal heat sensitivity in NaaaCD11b+ mice (orange triangles) and their wild-type littermates (Wt, blue circles) after saline injection (20 μL). + +<--- Page Split ---> + +(G) Effects of \(\mathsf{PGE}_2\) in NaaaCD11b+ (orange triangles) and wild-type littermates (blue circles). + +Data are expressed as mean ± S.E.M. and were analyzed using the Student's t test (AUC) or two- way repeated measure ANOVA. Bonferroni's post hoc test was applied as appropriate. ##P < 0.001, ##P < 0.01, #P < 0.05 versus Naaafl/n or Wt. ns, non- significant (n = 7- 8 mice per group). Dotted lines indicate baseline withdrawal latency. + +![](images/Figure_4.jpg) + +
Figure 4: PPARα in CD11b+ myeloid cells counters hyperalgesic priming.
+ +(A) Generation of \(Ppara^{CD11b - / - }\) mice. Global \(Ppara^{/-}\) mice were mated with flippase (fl) FLPo-10 mice and the offspring was cross-bred with CD11b-Cre mice to yield the \(Ppara^{CD11b - / - }\) line. Gray boxes: \(Ppara\) coding sequences; green boxes, Flp recognition target (FRT) sites; gray triangles: loxP sites flanking the deleted \(Ppara\) sequence. + +<--- Page Split ---> + +(B) FACS tracings showing the isolation of monocytes from blood samples of (top) \(Ppara^{f / f}\) and (bottom) \(Ppara^{CD11b - / - }\) mice. Left, isolation of CD11b\(^+\) cells; right, isolation of Ly6C\(^+\) monocytes. + +The square denotes the Ly6C\(^high\) (classical) monocyte subpopulation collected for RT-qPCR analysis. 1, Ly6C\(^low\) (non-classical) monocytes; 2, neutrophils. + +(C) \(Ppara\) mRNA levels (assessed by RT-qPCR) in FACS-isolated CD11b\(^-\) cells and Ly6C\(^high\) monocytes from \(Ppara^{f / f}\) (open circles) and \(Ppara^{CD11b - / - }\) (magenta triangles) mice. Whole blood is shown for comparison. Bars represent the average of 2 biological replicates. + +(D) Normal heat sensitivity in \(Ppara^{f / f}\) and \(Ppara^{CD11b - / - }\) mice after saline injection (20 μL). + +(E) Effects of PGE\(_2\) in non-primed (saline-injected) \(Ppara^{f / f}\) and \(Ppara^{CD11b - / - }\) mice. Right panel, area under the curve (AUC). + +(F) IL-6-primed \(Ppara^{f / f}\) mice were treated on 1d-3d with vehicle or ARN19702 (30 mg·kg\(^{-1}\)). (F1) Effect of IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). + +(F2) Effects of PGE\(_2\) in IL-6-primed \(Ppara^{f / f}\) mice after administration of vehicle (red circles) or ARN19702 (blue circles). Right panel, AUC. + +(G) IL-6-primed \(Ppara^{CD11b - / - }\) mice were treated on 1d-3d with vehicle or ARN19702 (30 mg·kg\(^{-1}\)). + +(G1) Self-resolving heat hypersensitivity elicited by IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). (G2) Effect of PGE\(_2\) in IL-6-primed \(Ppara^{CD11b - / - }\) mice after administration of vehicle (red triangles) or ARN19702 (blue triangles). Right panel, AUC. + +Data are represented as mean ± S.E.M. and were analyzed using the Student's \(t\) test (AUC) or two-way repeated measure ### \(P < 0.001\) , ## \(P < 0.01\) , ## \(P < 0.05\) , versus vehicle or \(Ppara^{f / f}\) (n = 7- 10 mice per group). Dotted lines indicate baseline withdrawal latency. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Monocyte-derived cells, not neutrophils, are required for hyperalgesic priming.
+ +(A- D) Verification of monocyte removal by clodronate. (A) Liposomes containing clodronate or PBS were injected twice with a 3- day interval in wild- type male mice. Two hours after the second injection, blood and spinal cord were collected for FACS analysis. (B) FACS tracings of blood samples from mice treated with (top) PBS or (bottom) clodronate. Left, isolation of CD11b+ cells; right, isolation of Ly6Chigh ('classical') monocytes (1), Ly6Clow ('non- classical') monocytes (2), and neutrophils (3). The squares denote areas selected for quantification. (C, D) Number of circulating (C) monocytes and (D) neutrophils in mice treated with PBS (red bar) or clodronate (blue bar). Bars represent the average of 3 biological replicates. (E) Experimental timeline for the clodronate experiment. IL- 6 was administered in wild- type male mice immediately before the last clodronate/PBS injection. PGE2 was injected 6 days later. (F1) Self- resolving heat hypersensitivity elicited by IL- 6 in mice treated with PBS (red line) or clodronate (Clo, blue line). (F2) Effects of PGE2 in IL- 6 primed mice treated with PBS (red line) or clodronate (Clo, blue line). Right panel, area under the curve (AUC). (G- J) Verification of monocyte removal by PLX5622. (G) Mice were fed chow containing PLX5622 or vehicle for 20 days. On the last day, blood and spinal cord were collected for FACS analysis. + +<--- Page Split ---> + +(H) FACS tracings of blood samples from mice fed a chow containing vehicle (top) or PLX5622 (bottom). Left, isolation of CD11b+ cells; right, isolation of Ly6Chigh ('classical') monocytes (1), Ly6Clow ('non-classical') monocytes (2), and neutrophils (3). The squares denote areas selected for quantification. + +(I, J) Number of circulating (I) monocytes and (J) neutrophils in mice exposed to control chow (red bar) and chow containing PLX5622 (blue bar). + +(K) Experimental timeline for the PLX5622 experiment. IL- 6 was administered in wild-type male mice after 14 days of exposure to control or PLX5622-containing chow. The treatment was continued and \(\mathsf{PPE}_2\) was injected 6 days later. + +(L1) Self-resolving heat hypersensitivity elicited by IL- 6 in mice exposed to control chow (red line) or PLX5622-containing chow (blue line). + +(L2) Effects of \(\mathsf{PPE}_2\) in IL- 6 primed mice exposed to control chow (red line) or PLX5622-containing chow (blue line). Right panel, AUC. + +Data are represented as mean \(\pm\) S.E.M. and were analyzed using the Student's \(t\) test (AUC) or two- way repeated measure \(\mathrm{###P< 0.001}\) , \(\mathrm{##P< 0.01}\) , \(\mathrm{##P< 0.05}\) , versus PBS or control (n = 4- 10 mice per group). Dotted lines indicate baseline withdrawal latency. + +![](images/Figure_6.jpg) + +
Figure 6: A role for NAAA-regulated PPAR-\(\alpha\) signaling in hyperalgesic priming.
+ +<--- Page Split ---> + +**Left: non-primed state.** High baseline concentrations of NAAA substrates (e.g., PEA) in tissue-resident macrophages constitutively activate PPAR- \(\alpha\) -dependent transcriptional programs, which maintain the cells in a resting state66,87. In this non-primed state, \(\text {PGE}_{2}\) produces transient ( \(\sim 2h\) ) heat hypersensitivity by engaging protein kinase A (PKA)- coupled \(\text {EP}_{2}/\text {EP}_{4}\) receptors in isolectin- \(\text {B4}^{+}\) nociceptive neurons. PA, palmitic acid, a product of PEA cleavage by NAAA. + +**Center: initiation and incubation.** IL- 6 evokes a self-resolving nocifensive response that Isats \(\sim 24h\) and is attenuated by standard analgesics (22,23, present study). In parallel, the cytokine initiates priming through a mechanism that is insensitive to these agents. This initiation phase is followed by an incubation period, which persists for \(\sim 72h\), during which NAAA activity interrupts PPAR- \(\alpha\) signaling and drives tissue-resident macrophages and/or infiltrating monocytes into an active (or 'primed') state, which enables them to interact with non-peptidergic nociceptors - presumably via one or more unknown chemical signals (colored circles) - and to effect neuroplastic changes that initiate priming. + +**Right: primed state.** In primed animals, \(\text {EP}_{2}/\text {EP}_{4}\) receptors in non-peptidergic neurons are coupled to an additional transduction pathway that involves protein kinase C- epsilon \((\text {PKC}_{\epsilon})^{17}\) . In this state, \(\text {PGE}_{2}\) evokes a \(\text {PKC}_{\epsilon}\) - mediated heat hypersensitivity that can last \(>2\) weeks38. Figure was created with BioRender.com. + +## References + +1 Dahlhamer, J. et al. Prevalence of Chronic Pain and High- Impact Chronic Pain Among Adults - United States, 2016. MMWR Morb Mortal Wkly Rep 67, 1001- 1006, doi:10.15585/mmwr.mm6736a2 (2018). 2 Goldberg, D. S. & McGee, S. J. Pain as a global public health priority. BMC Public Health 11, 770, doi:10.1186/1471- 2458- 11- 770 (2011). 3 Cohen, S. P., Vase, L. & Hooten, W. M. Chronic pain: an update on burden, best practices, and new advances. Lancet 397, 2082- 2097, doi:10.1016/S0140- 6736(21)00393- 7 (2021). 4 Price, T. J. et al. Transition to chronic pain: opportunities for novel therapeutics. Nat Rev Neurosci 19, 383- 384, doi:10.1038/s41583- 018- 0012- 5 (2018). 5 Basbaum, A. I., Bautista, D. M., Scherrer, G. & Julius, D. Cellular and molecular mechanisms of pain. Cell 139, 267- 284, doi:10.1016/j.cell.2009.09.028 (2009). 6 Woolf, C. J. Central sensitization: implications for the diagnosis and treatment of pain. Pain 152, S2- S15, doi:10.1016/j.pain.2010.09.030 (2011). 7 Price, T. J. & Ray, P. R. Recent advances toward understanding the mysteries of the acute to chronic pain transition. Curr Opin Physiol 11, 42- 50, doi:10.1016/j.cophys.2019.05.015 (2019). 8 Ji, R. R., Chamessian, A. & Zhang, Y. Q. Pain regulation by non- neuronal cells and inflammation. Science 354, 572- 577, doi:10.1126/science.aaf8924 (2016). + +<--- Page Split ---> + +Calvo, M., Dawes, J. M. & Bennett, D. L. The role of the immune system in the generation of neuropathic pain. Lancet Neurol 11, 629- 642, doi:10.1016/S1474- 4422(12)70134- 5 (2012).Malcangio, M. Role of the immune system in neuropathic pain. Scand J Pain 20, 33- 37, doi:10.1515/sjpain- 2019- 0138 (2019).Malcangio, M. Introduction for special issue on neuroimmune interactions in chronic pain. Pain Rep 6, e894, doi:10.1097/PR9.0000000000000894 (2021).Zhang, J. et al. Expression of CCR2 in both resident and bone marrow- derived microglia plays a critical role in neuropathic pain. J Neurosci 27, 12396- 12406, doi:10.1523/JNEUROSCI.3016- 07.2007 (2007).Peng, J. et al. Microglia and monocytes synergistically promote the transition from acute to chronic pain after nerve injury. Nat Commun 7, 12029, doi:10.1038/ncomms12029 (2016).Yu, X. et al. Dorsal root ganglion macrophages contribute to both the initiation and persistence of neuropathic pain. Nat Commun 11, 264, doi:10.1038/s41467- 019- 13839- 2 (2020).Old, E. A. et al. Monocytes expressing CX3CR1 orchestrate the development of vincristine- induced pain. J Clin Invest 124, 2023- 2036, doi:10.1172/JCI71389 (2014).Oggero, S. et al. Dorsal root ganglia CX3CR1 expressing monocytes/macrophages contribute to arthritis pain. Brain Behav Immun 106, 289- 306, doi:10.1016/j.bbi.2022.09.008 (2022).Reichling, D. B. & Levine, J. D. Critical role of nociceptor plasticity in chronic pain. Trends Neurosci 32, 611- 618, doi:10.1016/j.tins.2009.07.007 (2009).Kandasamy, R. & Price, T. J. The pharmacology of nociceptor priming. Handb Exp Pharmacol 227, 15- 37, doi:10.1007/978- 3- 662- 46450- 2_2 (2015).Aley, K. O., Messing, R. O., Mochly- Rosen, D. & Levine, J. D. Chronic hypersensitivity for inflammatory nociceptor sensitization mediated by the epsilon isozyme of protein kinase C. J Neurosci 20, 4680- 4685, doi:10.1523/JNEUROSCI.20- 12- 04680.2000 (2000).Parada, C. A., Reichling, D. B. & Levine, J. D. Chronic hyperalgesic priming in the rat involves a novel interaction between cAMP and PKCepsilon second messenger pathways. Pain 113, 185- 190, doi:10.1016/j.pain.2004.10.021 (2005).Joseph, E. K. & Levine, J. D. Hyperalgesic priming is restricted to isolectin B4- positive nociceptors. Neuroscience 169, 431- 435, doi:10.1016/j.neuroscience.2010.04.082 (2010).Parada, C. A., Yeh, J. J., Joseph, E. K. & Levine, J. D. Tumor necrosis factor receptor type- 1 in sensory neurons contributes to induction of chronic enhancement of inflammatory hyperalgesia in rat. Eur J Neurosci 17, 1847- 1852, doi:10.1046/j.1460- 9568.2003.02626. x (2003).Ferrari, L. F., Bogen, O., Reichling, D. B. & Levine, J. D. Accounting for the delay in the transition from acute to chronic pain: axonal and nuclear mechanisms. J Neurosci 35, 495- 507, doi:10.1523/JNEUROSCI.5147- 13.2015 (2015).Bogen, O., Alessandri- Haber, N., Chu, C., Gear, R. W. & Levine, J. D. Generation of a pain memory in the primary afferent nociceptor triggered by PKCepsilon activation of CPEB. J Neurosci 32, 2018- 2026, doi:10.1523/JNEUROSCI.5138- 11.2012 (2012). + +<--- Page Split ---> + +781 25 Melemedjian, O. K. et al. Local translation and retrograde axonal transport of CREB regulates IL- 6- induced nociceptive plasticity. Mol Pain 10, 45, doi:10.1186/1744- 8069- 10- 45 (2014).783 26 Piomelli, D. et al. N- Acylethanolamine Acid Amidase (NAAA): Structure, Function, and Inhibition. J Med Chem 63, 7475- 7490, doi:10.1021/acs.jmedchem.0c00191 (2020).786 27 Lo Verme, J. et al. The nuclear receptor peroxisome proliferator- activated receptor- alpha mediates the anti- inflammatory actions of palmitoylethanolamide. Mol Pharmacol 67, 15- 19, doi:10.1124/mol.104.006353 (2005).788 28 LoVerme, J., La Rana, G., Russo, R., Calignano, A. & Piomelli, D. The search for the palmitoylethanolamide receptor. Life Sci 77, 1685- 1698, doi:10.1016/j.lfs.2005.05.012 (2005).792 29 LoVerme, J. et al. Rapid broad- spectrum analgesia through activation of peroxisome proliferator- activated receptor- alpha. J Pharmacol Exp Ther 319, 1051- 1061, doi:10.1124/jpet.106.111385 (2006).794 30 Solorzano, C. et al. Selective N- acylethanolamine- hydrolyzing acid amidase inhibition reveals a key role for endogenous palmitoylethanolamide in inflammation. Proc Natl Acad Sci U S A 106, 20966- 20971, doi:10.1073/pnas.0907417106 (2009).798 31 Fotio, Y. et al. NAAA- regulated lipid signaling governs the transition from acute to chronic pain. Sci Adv 7, eabi8834, doi:10.1126/sciadv.abi8834 (2021).800 32 Sasso, O. et al. The N- Acylethanolamine Acid Amidase Inhibitor ARN077 Suppresses Inflammation and Pruritus in a Mouse Model of Allergic Dermatitis. J Invest Dermatol 138, 562- 569, doi:10.1016/j.jid.2017.07.853 (2018).803 33 Pontis, S. et al. N- Acylethanolamine Acid Amidase contributes to disease progression in a mouse model of multiple sclerosis. Pharmacol Res 160, 105064, doi:10.1016/j.phrs.2020.105064 (2020).806 34 Spangenberg, E. et al. Sustained microglial depletion with CSF1R inhibitor impairs parenchymal plaque development in an Alzheimer's disease model. Nat Commun 10, 3758, doi:10.1038/s41467- 019- 11674- z (2019).809 35 Van Rooijen, N. The liposome- mediated macrophage 'suicide' technique. J Immunol Methods 124, 1- 6, doi:10.1016/0022- 1759(89)90178- 6 (1989).811 36 Lee, P. Y., Wang, J. X., Parisini, E., Dascher, C. C. & Nigrovic, P. A. Ly6 family proteins in neutrophil biology. J Leukoc Biol 94, 585- 594, doi:10.1189/jlb.0113014 (2013).813 37 Jurga, A. M., Paleczna, M. & Kuter, K. Z. Overview of General and Discriminating Markers of Differential Microglia Phenotypes. Front Cell Neurosci 14, 198, doi:10.3389/fncel.2020.00198 (2020).816 38 Paige, C. et al. Neuroendocrine Mechanisms Governing Sex Differences in Hyperalgesic Priming Involve Prolactin Receptor Sensory Neuron Signaling. J Neurosci 40, 7080- 7090, doi:10.1523/JNEUROSCI.1499- 20.2020 (2020).819 39 Mabou Tagne, A., Fotio, Y., Uppal, P. S. & Piomelli, D. Synergistic antinociceptive effects of concomitant NAAA and peripheral FAAH inhibition. Exp Neurol 357, 114194, doi:10.1016/j.expneurol.2022.114194 (2022).822 40 Mabou Tagne, A. et al. Palmitoylethanolamide and hemp oil extract exert synergistic antinociceptive effects in mouse models of acute and chronic pain. Pharmacol Res 167, 105545, doi:10.1016/j.phrs.2021.105545 (2021). + +<--- Page Split ---> + +825 41 Fotio, Y., Palese, F., Guaman Tipan, P., Ahmed, F. & Piomelli, D. Inhibition of fatty acid amide hydrolase in the CNS prevents and reverses morphine tolerance in male and female mice. Br J Pharmacol 177, 3024- 3035, doi:10.1111/bph.15031 (2020). 828 42 Fotio, Y., Mabou Tagne, A., Jung, K. M. & Piomelli, D. Fatty acid amide hydrolase inhibition alleviates anxiety- like symptoms in a rat model used to study post- traumatic stress disorder. Psychopharmacology (Berl), doi:10.1007/s00213- 023- 06358- y (2023). 831 43 Lee, H. L. et al. Frequent Low- Dose Delta(9)- Tetrahydrocannabinol in Adolescence Disrupts Microglia Homeostasis and Disables Responses to Microbial Infection and Social Stress in Young Adulthood. Biol Psychiatry 92, 845- 860, doi:10.1016/j.biopsych.2022.04.017 (2022). 834 44 Gabrielsson, J. & Weiner, D. Non- compartmental analysis. Methods Mol Biol 929, 377- 389, doi:10.1007/978- 1- 62703- 050- 2_16 (2012). 837 45 Migliore, M. D. et al. Second- Generation Non- Covalent NAAA Inhibitors are Protective in a Model of Multiple Sclerosis. Angew Chem Int Ed Engl 55, 11193- 11197, doi:10.1002/anie.201603746 (2016). 840 46 Fotio, Y., Sasso, O., Ciccocioppo, R. & Piomelli, D. Antinociceptive Profile of ARN19702, (2- Ethylsulfonylphenyl)- [(2S)-4-(6- fluoro- 1,3- benzothiazol- 2- yl)- 2- methylpiperazin- 1- yl]methanone, a Novel Orally Active N- Acylethanolamine Acid Amidase Inhibitor, in Animal Models. J Pharmacol Exp Ther 378, 70- 76, doi:10.1124/jpet.121.000674 (2021). 844 47 Xu, H. E. et al. Structural basis for antagonist- mediated recruitment of nuclear co- repressors by PPARalpha. Nature 415, 813- 817, doi:10.1038/415813a (2002). 846 48 Lee, G. et al. T0070907, a selective ligand for peroxisome proliferator- activated receptor gamma, functions as an antagonist of biochemical and cellular activities. J Biol Chem 277, 19649- 19657, doi:10.1074/jbc.M200743200 (2002). 849 49 Hosohata, Y. et al. AM630 antagonism of cannabinoid- stimulated [35S]GTP gamma S binding in the mouse brain. Eur J Pharmacol 321, R1- 3, doi:10.1016/s0014- 2999(97)00047- 2 (1997). 852 50 Paterniti, I. et al. Molecular evidence for the involvement of PPAR- delta and PPAR- gamma in anti- inflammatory and neuroprotective activities of palmitoylethanolamide after spinal cord trauma. J Neuroinflammation 10, 20, doi:10.1186/1742- 2094- 10- 20 (2013). 853 51 Guida, F. et al. Palmitoylethanolamide induces microglia changes associated with increased migration and phagocytic activity: involvement of the CB2 receptor. Sci Rep 7, 375, doi:10.1038/s41598- 017- 00342- 1 (2017). 854 52 Palese, F. et al. Targeting NAAA counters dopamine neuron loss and symptom progression in mouse models of parkinsonism. Pharmacol Res 182, 106338, doi:10.1016/j.phrs.2022.106338 (2022). 856 53 Ribeiro, A. et al. A Potent Systemically Active N- Acylethanolamine Acid Amidase Inhibitor that Suppresses Inflammation and Human Macrophage Activation. ACS Chem Biol 10, 1838- 1846, doi:10.1021/acschembio.5b00114 (2015). 857 54 Tsuboi, K. et al. Predominant expression of lysosomal N- acylethanolamine- hydrolyzing acid amidase in macrophages revealed by immunochemical studies. Biochim Biophys Acta 1771, 623- 632, doi:10.1016/j.bnalip.2007.03.005 (2007). 856 55 Gordon, S. & Taylor, P. R. Monocyte and macrophage heterogeneity. Nat Rev Immunol 5, 953- 964, doi:10.1038/nri1733 (2005). + +<--- Page Split ---> + +869 56 Mass, E., Nimmerjahn, F., Kierdorf, K. & Schlitzer, A. Tissue- specific macrophages: how 870 they develop and choreograph tissue biology. Nat Rev Immunol, 1- 17, 871 doi:10.1038/s41577- 023- 00848- y (2023). 872 57 van Rooijen, N. & Hendrikx, E. Liposomes for specific depletion of macrophages from 873 organs and tissues. Methods Mol Biol 605, 189- 203, doi:10.1007/978- 1- 60327- 360- 2_13 874 (2010). 875 58 Schmidt- Weber, C. B. et al. Apoptotic cell death in activated monocytes following 876 incorporation of clodronate- liposomes. J Leukoc Biol 60, 230- 244, 877 doi:10.1002/jlb.60.2.230 (1996). 878 59 Van Rooijen, N. & Sanders, A. Liposome mediated depletion of macrophages: mechanism 879 of action, preparation of liposomes and applications. J Immunol Methods 174, 83- 93, 880 doi:10.1016/0022- 1759(94)90012- 4 (1994). 881 60 Lee, H. H. et al. Depletion of Alveolar Macrophages Increases Pulmonary Neutrophil 882 Infiltration, Tissue Damage, and Sepsis in a Murine Model of Acinetobacter baumannii 883 Pneumonia. Infect Immun 88, doi:10.1128/AI.00128- 20 (2020). 884 61 Culemann, S. et al. Stunning of neutrophils accounts for the anti- inflammatory effects of 885 clodronate liposomes. J Exp Med 220, doi:10.1084/jem.20220525 (2023). 886 62 Willemen, H. L. et al. Monocytes/Macrophages control resolution of transient inflammatory 887 pain. J Pain 15, 496- 506, doi:10.1016/j.pain.2014.01.491 (2014). 888 63 Spiteri, A. G. et al. PLX5622 Reduces Disease Severity in Lethal CNS Infection by Off- 889 Target Inhibition of Peripheral Inflammatory Monocyte Production. Front Immunol 13, 890 851556, doi:10.3389/fimmu.2022.851556 (2022). 891 64 Lei, F. et al. CSF1R inhibition by a small- molecule inhibitor is not microglia specific; 892 affecting hematopoiesis and the function of macrophages. Proc Natl Acad Sci U S A 117, 893 23336- 23338, doi:10.1073/pnas.1922788117 (2020). 894 65 Rigamonti, E., Chinetti- Gbaguidi, G. & Staels, B. Regulation of macrophage functions by 895 PPAR- alpha, PPAR- gamma, and LXRs in mice and men. Arterioscler Thromb Vasc Biol 896 28, 1050- 1059, doi:10.1161/ATVBAHA.107.158998 (2008). 897 66 Grabacka, M., Pierzchalska, M., Plonka, P. M. & Pierzchalski, P. The Role of PPAR Alpha 898 in the Modulation of Innate Immunity. Int J Mol Sci 22, doi:10.3390/ijms221910545 (2021). 899 Penas, F. et al. Treatment in vitro with PPARalpha and PPARgamma ligands drives M1- 900 to- M2 polarization of macrophages from T. cruzi- infected mice. Biochim Biophys Acta 901 1852, 893- 904, doi:10.1016/j.bpids.2014.12.019 (2015). 902 68 Feng, X. et al. PPAR- alpha Agonist Fenofibrate Prevented Diabetic Nephropathy by 903 Inhibiting M1 Macrophages via Improving Endothelial Cell Function in db/db Mice. Front 904 Med (Lausanne) 8, 652558, doi:10.3389/fmed.2021.652558 (2021). 905 69 Fu, J. et al. Oleylethanolamide regulates feeding and body weight through activation of 906 the nuclear receptor PPAR- alpha. Nature 425, 90- 93, doi:10.1038/nature01921 (2003). 907 70 Okamoto, Y., Morishita, J., Tsuboi, K., Tonai, T. & Ueda, N. Molecular characterization of 908 a phospholipase D generating anandamide and its congeners. J Biol Chem 279, 5298- 909 5305, doi:10.1074/jbc.M306642200 (2004). 910 71 Magotti, P. et al. Structure of human N- acylphosphatidylethanolamine- hydrolyzing 911 phospholipase D: regulation of fatty acid ethanolamide biosynthesis by bile acids. 912 Structure 23, 598- 604, doi:10.1016/j.str.2014.12.018 (2015). + +<--- Page Split ---> + +Zhu, C. et al. Proinflammatory stimuli control N-acylphosphatidylethanolamine-specific phospholipase D expression in macrophages. Mol Pharmacol 79, 786- 792, doi:10.1124/mol.110.070201 (2011). Gorelik, A., Gebai, A., Illes, K., Piomelli, D. & Nagar, B. Molecular mechanism of activation of the immunoregulatory amidase NAAA. Proc Natl Acad Sci U S A 115, E10032- E10040, doi:10.1073/pnas.1811759115 (2018). Alhouayek, M. et al. N-Acylethanolamine-hydrolyzing acid amidase inhibition increases colon N-palmitoylethanolamine levels and counteracts murine colitis. FASEB J 29, 650- 661, doi:10.1096/fj.14- 255208 (2015). Bonezzi, F. T. et al. An Important Role for N-Acylethanolamine Acid Amidase in the Complete Freund's Adjuvant Rat Model of Arthritis. J Pharmacol Exp Ther 356, 656- 663, doi:10.1124/jpet.115.230516 (2016). Cairns, B. E., Arendt- Nielsen, L. & Sacerdote, P. Perspectives in Pain Research 2014: Neuroinflammation and glial cell activation: The cause of transition from acute to chronic pain? Scand J Pain 6, 3- 6, doi:10.1016/j.spain.2014.10.002 (2015). Caxaria, S. et al. Neutrophils infiltrate sensory ganglia and mediate chronic widespread pain in fibromyalgia. Proc Natl Acad Sci U S A 120, e2211631120, doi:10.1073/pnas.2211631120 (2023). Chen, Z. et al. Sphingosine- 1- phosphate receptor 1 activation in astrocytes contributes to neuropathic pain. Proc Natl Acad Sci U S A 116, 10557- 10562, doi:10.1073/pnas.1820466116 (2019). Squillace, S., Spiegel, S. & Salvemini, D. Targeting the Sphingosine- 1- Phosphate Axis for Developing Non- narcotic Pain Therapeutics. Trends Pharmacol Sci 41, 851- 867, doi:10.1016/j.tips.2020.09.006 (2020). Keeble, J. E. et al. Hydrogen peroxide is a novel mediator of inflammatory hyperalgesia, acting via transient receptor potential vanilloid 1- dependent and independent mechanisms. Pain 141, 135- 142, doi:10.1016/j.pain.2008.10.025 (2009). Reis, C., Chambel, S., Ferreira, A. & Cruz, C. D. Involvement of nerve growth factor (NGF) in chronic neuropathic pain - a systematic review. Rev Neurosci 34, 75- 84, doi:10.1515/revneuro- 2022- 0037 (2023). Cappoli, N., Tabolacci, E., Aceto, P. & Dello Russo, C. The emerging role of the BDNF- TrkB signaling pathway in the modulation of pain perception. J Neuroimmunol 349, 577406, doi:10.1016/j.jneuroim.2020.577406 (2020). Finnerup, N. B., Nikolajsen, L. & Rice, A. S. C. Transition from acute to chronic pain: a misleading concept? Pain 163, e985- e988, doi:10.1097/j.pain.0000000000002631 (2022). Shiers, S. et al. Neuropathic Pain Creates an Enduring Prefrontal Cortex Dysfunction Corrected by the Type II Diabetic Drug Metformin But Not by Gabapentin. J Neurosci 38, 7337- 7350, doi:10.1523/JNEUROSCI.0713- 18.2018 (2018). Inyang, K. E. et al. Alleviation of paclitaxel- induced mechanical hypersensitivity and hyperalgesic priming with AMPK activators in male and female mice. Neurobiol Pain 6, 100037, doi:10.1016/j.ynpai.2019.100037 (2019). Inyang, K. E. et al. Indirect AMP- Activated Protein Kinase Activators Prevent Incision- Induced Hyperalgesia and Block Hyperalgesic Priming, Whereas Positive Allosteric + +<--- Page Split ---> + +957 Modulators Block Only Priming in Mice. J Pharmacol Exp Ther 371, 138- 150, 958 doi:10.1124/jpet.119.258400 (2019). 959 87 Pontis, S., Ribeiro, A., Sasso, O. & Piomelli, D. Macrophage- derived lipid agonists of 960 PPAR- alpha as intrinsic controllers of inflammation. Crit Rev Biochem Mol Biol 51, 7- 14, 961 doi:10.3109/10409238.2015.1092944 (2016). 962 + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- FotioNatComm13082023AllSupplementalmaterial.pdf- Fotioetal.2023Figure1.xlsx- Fotioetal.2023Figure2.xlsx- Fotioetal.2023FigureS2.xlsx- Fotioetal.2023Figure4.xlsx- Fotioetal.2023Figure3.xlsx- Fotioetal.2023Figure5.xlsx- Fotioetal.2023FigureS6.xlsx- Fotioetal.2023FigureS7.xlsx- Fotioetal.2023FigureS8.xlsx- Fotioetal.2023FigureS3.xlsx- Fotioetal.2023FigureS5.xlsx- Fotioetal.2023FigureS4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec_det.mmd b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..41f4982dbfdc2744b0a12397f581115b3ec4da46 --- /dev/null +++ b/preprint/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec/preprint__123bd0ee09c9429721618645aa9aeb915f306f1aae5c4eccbeaf12f98f492aec_det.mmd @@ -0,0 +1,563 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 944, 177]]<|/det|> +# NAAA-regulated lipid signaling in monocyte-derived cells controls the induction of hyperalgesic priming + +<|ref|>text<|/ref|><|det|>[[44, 196, 184, 214]]<|/det|> +Daniele Piomelli + +<|ref|>text<|/ref|><|det|>[[52, 223, 268, 240]]<|/det|> +piomelli@hs.uci.edu + +<|ref|>text<|/ref|><|det|>[[44, 270, 312, 308]]<|/det|> +University of California, Irvine + +<|ref|>text<|/ref|><|det|>[[44, 313, 312, 351]]<|/det|> +Yannick Fotio University of California, Irvine + +<|ref|>text<|/ref|><|det|>[[44, 337, 312, 377]]<|/det|> +Alex Mabou Tagne University of California, Irvine + +<|ref|>text<|/ref|><|det|>[[44, 383, 150, 401]]<|/det|> +Erica Squire + +<|ref|>text<|/ref|><|det|>[[52, 406, 140, 423]]<|/det|> +UC Irvine + +<|ref|>text<|/ref|><|det|>[[44, 430, 148, 448]]<|/det|> +Hye- lim Lee + +<|ref|>text<|/ref|><|det|>[[52, 454, 140, 470]]<|/det|> +UC Irvine + +<|ref|>text<|/ref|><|det|>[[44, 478, 160, 496]]<|/det|> +Faizy Ahmed + +<|ref|>text<|/ref|><|det|>[[52, 501, 140, 518]]<|/det|> +UC Irvine + +<|ref|>text<|/ref|><|det|>[[44, 525, 205, 544]]<|/det|> +Andrew Greenberg + +<|ref|>text<|/ref|><|det|>[[52, 546, 794, 565]]<|/det|> +JM- USDA Human Nutrition Research Center https://orcid.org/0000- 0001- 5062- 387X + +<|ref|>text<|/ref|><|det|>[[44, 570, 196, 608]]<|/det|> +Gilberto Spadoni Univ. of Urbino + +<|ref|>text<|/ref|><|det|>[[44, 616, 142, 633]]<|/det|> +Marco Mor + +<|ref|>text<|/ref|><|det|>[[52, 637, 585, 657]]<|/det|> +University of Parma https://orcid.org/0000- 0003- 0199- 1849 + +<|ref|>sub_title<|/ref|><|det|>[[44, 700, 102, 718]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 737, 136, 756]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 775, 320, 795]]<|/det|> +Posted Date: August 19th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 813, 474, 833]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3263903/v1 + +<|ref|>text<|/ref|><|det|>[[44, 850, 909, 894]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 910, 945, 954]]<|/det|> +Additional Declarations: Yes there is potential Competing Interest. D.P. and M.M. are inventors in patents that protect ARN19702 and other NAAA inhibitors, owned by the University of California, the University of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 949, 156]]<|/det|> +Parma, the University of Urbino, and the Fondazione Istituto Italiano di Tecnologia (no. 13/898,225, filed 20 May 2013, published 3 April 2014; no. 62/337,744, filed 17 May 2016, published 23 November 2017). D.P. and Y.F. are inventors in a patent application that protects the algostatic effects of NAAA inhibitors, filed by the University of California (no. 63/166,134, filed 25 March 2021, published 29 September 2022). The other authors have declared no competing interest. + +<|ref|>text<|/ref|><|det|>[[42, 191, 946, 234]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 24th, 2024. See the published version at https://doi.org/10.1038/s41467-024-46139-5. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[250, 256, 745, 300]]<|/det|> +# NAAA-regulated lipid signaling in monocyte-derived cells controls the induction of hyperalgesic priming + +<|ref|>text<|/ref|><|det|>[[171, 350, 825, 395]]<|/det|> +Yannick Fotio \(^{1}\) , Alex Mabou Tagne \(^{1a}\) , Erica Squire \(^{1a}\) , Hye- lim Lee \(^{1}\) , Faizy Ahmed \(^{1}\) , Andrew S. Greenberg \(^{2}\) , Gilberto Spadoni \(^{3}\) , Marco Mor \(^{4}\) , and Daniele Piomelli \(^{1,5,6,8}\) + +<|ref|>text<|/ref|><|det|>[[135, 446, 860, 585]]<|/det|> +\(^{1}\) Department of Anatomy and Neurobiology, University of California Irvine, Irvine, CA, USA. \(^{2}\) Human Nutrition Research Center, Tufts University, Boston, MA, USA \(^{3}\) Dipartimento di Scienze Biomolecolari, Università di Urbino "Carlo Bo," Urbino, Italy. \(^{4}\) Dipartimento di Scienze degli Alimenti e del Farmaco, Università di Parma, Parma, Italy. \(^{5}\) Department of Biological Chemistry, University of California Irvine, Irvine, CA, USA. \(^{6}\) Department of Pharmaceutical Sciences, University of California Irvine, Irvine, CA, USA. + +<|ref|>text<|/ref|><|det|>[[115, 640, 404, 658]]<|/det|> +\(^{a}\) These authors contributed equally. + +<|ref|>text<|/ref|><|det|>[[133, 689, 863, 736]]<|/det|> +\(^{8}\) Lead author: Daniele Piomelli (piomelli@hs.uci.edu). Gillespie Neuroscience Res. Facility (room 3101), 837 Health Sciences Rd., Irvine, CA 92697. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 191, 106]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[110, 140, 886, 575]]<|/det|> +Circulating monocytes and tissue- resident macrophages contribute in important ways to the transition from acute to chronic pain but the molecular events that cause their deployment are still unclear. Using a mouse model of hyperalgesic priming (HP), we show now that monocyte- derived cells enable the progression to pain chronicity through a cell- autonomous mechanism that requires the transient activation of the intracellular lipid hydrolase, \(N\) - acylethanolamine acid amidase (NAAA), and the consequent suppression of NAAA- regulated lipid signaling at peroxisome proliferator- activated receptor- \(\alpha\) (PPAR- \(\alpha\) ). Inhibiting peripheral NAAA in the 72 hours following administration of one of three different priming stimuli (interleukin- 6, tumor necrosis factor- \(\alpha\) , or carrageenan), but not before or after this time interval, prevented the emergence of HP in male and female mice. This effect was phenocopied by global NAAA deletion and depended on the recruitment of PPAR- \(\alpha\) by endogenous NAAA substrates. Transgenic mice selectively lacking NAAA in CD11b+ myeloid cells – which include monocytes, macrophages, and neutrophils – were resistant to HP induction. Conversely, mice overexpressing NAAA or lacking PPAR- \(\alpha\) in the same cells were constitutively primed. In vivo ablation of monocytes, but not neutrophils, by administration of clodronate liposomes or blockade of type- 1 colony- stimulating factor receptors, generated mice that were refractory to priming. The results identify NAAA- regulated lipid signaling in monocyte- derived cells as a control node in the induction of HP and the transition to pain chronicity. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 223, 106]]<|/det|> +## 49 Introduction + +<|ref|>text<|/ref|><|det|>[[111, 111, 886, 420]]<|/det|> +Chronic pain places an enormous burden on hundreds of millions of people worldwide \(^{1 - 3}\) , yet its neurobiological underpinnings remain largely unknown \(^{4,5}\) . One major challenge is the identification of molecular events that enable acute pain episodes – which typically accompany self- resolving, localized tissue injuries – to progress into persistent painful states that outlast the initial damage and can radiate outside its organ boundaries \(^{4}\) . In addition to neuroplastic changes, whose roles are well established \(^{6,7}\) , activation of the innate immune system has recently emerged as a driving factor in the progression to pain chronicity \(^{8 - 11}\) . Studies in mice have shown, for example, that blood- borne monocytes infiltrate the spinal cord \(^{12}\) and act in synergy with local microglia to promote pain chronication after nerve injury \(^{13,14}\) . Similarly, monocytes and macrophages that express the CX3CR1 receptor contribute to arthritis pain and chemotherapy- induced allodynia by interacting with nociceptive neurons in dorsal root ganglia (DRG) \(^{15,16}\) . Despite these advances, the molecular checkpoints that trigger the deployment of monocytes and macrophages during the transition from acute to chronic pain are still poorly understood \(^{8 - 10}\) . + +<|ref|>text<|/ref|><|det|>[[110, 425, 886, 907]]<|/det|> +An experimental model developed to investigate this transition exploits a form of neuroplasticity in which administration of a proinflammatory substance sensitizes a subset of peripheral nociceptors to subsequent noxious and non- noxious stimuli \(^{17,18}\) . In the report that first described this phenomenon \(^{19}\) , termed 'nociceptive' or 'hyperalgesic priming' (HP), Levine and collaborators showed that the inflammatory reaction produced by intraplantar injection of carrageenan in rats was followed by a prolonged (≥3 weeks) enhancement of the nocifensive response to other proalgesic agents applied at the same site. Further studies by the Levine laboratory identified non- peptidergic isolectin B4+ nociceptors as cellular substrates for HP and showed that its emergence required the recruitment of protein kinase Cε- dependent signal transduction pathways \(^{19 - 21}\) . In addition to its clinical relevance \(^{18}\) , HP has three notable advantages as a model to explore the progression to pain chronicity. First, it can be induced by administering a single inflammatory agent with identified mechanism of action – e.g., interleukin 6 (IL- 6) or tumor- necrosis factor- α (TNF- α) – thus allowing control over a key experimental variable. Second, the initial nocifensive reaction to a priming stimulus can be suppressed by analgesic drugs with no detectable consequences on HP induction, a clear indication that the two processes are temporally overlapping but mechanistically distinct \(^{22,23}\) . Finally, HP unfolds with a distinctive time- course in which the immediate response to the priming trigger is separated from the appearance of the primed state by a 72 h- long interval \(^{23,24}\) . This delay, which has been attributed to axonal and nuclear processes \(^{23,25}\) , identifies three phases of HP which can be studied in relative isolation: 'initiation', which occurs in parallel with the inflammatory reaction to the priming trigger; + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 884, 130]]<|/det|> +incubation', which follows the resolution of such reaction and precedes the appearance of the primed state; and 'maintenance', which allows such state to persist for several weeks. + +<|ref|>text<|/ref|><|det|>[[110, 139, 884, 453]]<|/det|> +In the present study, we used a combination of genetic and pharmacological strategies to determine whether the intracellular cysteine hydrolase, N- acetylcholamine acid amidase (NAAA) \(^{26}\) , contributes to the emergence of HP. NAAA catalyzes the hydrolytic deactivation of palmitoylethanolamide (PEA) and oleoylethanolamide (OEA), two lipid- derived agonists of the ligand- operated transcription factor, peroxisome proliferator- activated receptor- \(\alpha\) (PPAR- \(\alpha\) ) \(^{27,28}\) . In addition to its well- established roles in the modulation of acute nociception \(^{29}\) and inflammation \(^{30}\) , NAAA- regulated lipid signaling at PPAR- \(\alpha\) has been recently implicated in the induction of persistent pathological nociception by chemical or surgical damage to peripheral tissues \(^{31}\) . Building on those findings, we show now that monocytes and macrophages drive the progression to pain chronicity through a cell- autonomous mechanism that requires the suppression of intracellular NAAA- regulated PPAR- \(\alpha\) signaling during – but not before or after – the incubation phase of HP. The results identify NAAA in monocyte- derived cells as a control node in the development of chronic pain and a possible target for preventive therapy. + +<|ref|>sub_title<|/ref|><|det|>[[115, 488, 311, 505]]<|/det|> +## Materials and Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 541, 250, 559]]<|/det|> +## Study Approval + +<|ref|>text<|/ref|><|det|>[[112, 564, 884, 655]]<|/det|> +All experimental procedures complied with the ethical regulations for the care and use of laboratory animals promulgated by the National Institutes of Health (NIH) and the International Association for the Study of Pain (IASP). Formal approval was obtained from the Animal Care and Use Committee of the University of California, Irvine. + +<|ref|>sub_title<|/ref|><|det|>[[115, 685, 207, 701]]<|/det|> +## Chemicals + +<|ref|>text<|/ref|><|det|>[[111, 707, 884, 871]]<|/det|> +We purchased prostaglandin \(E_{2}\) (PGE \(_{2}\) ), recombinant mouse interleukin- 6 (IL- 6), and recombinant mouse tumor necrosis factor- alpha (TNF- \(\alpha\) ) from R&D Systems (Minneapolis, MN). Carrageenan, gabapentin, GW7647, GW6471, T0070907, and AM630 were from Sigma- Aldrich (St. Louis, MO). Anandamide, palmitoylethanolamide (PEA), oleoylethanolamide (OEA), their deuterium- containing analogs, and PLX5622 were purchased from Cayman Chemicals (Ann Arbor, MI). Levagen® PEA (a water- soluble PEA formulation) was a kind gift of GE Nutrients (Austin, TX). Liposome- encapsulated clodronate and control liposomes (containing phosphate- buffered saline, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 132]]<|/det|> +PBS, \(\mathsf{pH} = 7.4\) ) were from Liposoma (Amsterdam, Netherlands). ARN19702, ARN726, and ARN077 were prepared according to established protocols26. + +<|ref|>sub_title<|/ref|><|det|>[[115, 161, 430, 179]]<|/det|> +## Drug preparation and administration + +<|ref|>text<|/ref|><|det|>[[111, 183, 886, 545]]<|/det|> +Immediately before injections, a stock solution of \(\mathsf{PGE}_2\) (1 mg- mL \(^{- 1}\) in ethanol) was diluted in saline to a final concentration of \(5 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (ethanol: \(0.5\%\) vol/vol). TNF- \(\alpha\) and IL- 6 were dissolved in PBS \(\mathrm{(pH = 7.4)}\) containing fatty acid- free bovine serum albumin (BSA, \(0.5\%\) , Sigma- Aldrich) and were diluted with PBS to reach final concentrations of \(5 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (TNF- \(\alpha\) ) and \(0.025 \mathrm{ng} \cdot \mu \mathrm{L}^{- 1}\) (IL- 6). ARN19702, ARN726, GW7647, GW6471, T0070907, and AM630 were dissolved in a mixture of PEG- 400/Tween- 80/sterile saline (15:15:70, vol/vol/vol). Carrageenan, gabapentin, and Levagen® PEA were dissolved in sterile saline. Drugs and vehicles were administered by intraperitoneal (IP) or subcutaneous (SC) injection in a volume of \(10 \mathrm{mL} \cdot \mathrm{kg}^{- 1}\) . Priming stimuli (IL- 6, TNF- \(\alpha\) , carrageenan) and \(\mathsf{PGE}_2\) were administered into the mouse plantar surface using a hypodermic needle (30 gauge) attached to a Hamilton syringe (50 \(\mu \mathrm{L}\) ). Intrathecal injections of ARN726 were performed in mice anesthetized with isoflurane and held by the pelvic girdle. A Hamilton syringe (30 gauge, \(10 \mu \mathrm{L}\) ) was inserted perpendicularly to the skin at the intervertebral space L5- L6 (between the hip bones). A reflexive lateral tail flick indicated a subarachnoid puncture. A volume of \(5 \mu \mathrm{L}\) was delivered and the syringe was rotated and removed before returning the mice to their home cages. + +<|ref|>sub_title<|/ref|><|det|>[[115, 574, 303, 590]]<|/det|> +## Experimental animals + +<|ref|>text<|/ref|><|det|>[[111, 596, 886, 807]]<|/det|> +We used male and female C57BL/6J mice (22- 30 g, 10- 12 weeks old) from Jackson Laboratory (Sacramento, CA) as well as the genetically modified mouse lines described below. All animals were housed in the animal facility of the University of California Irvine, using standard ventilated clear plastic cages (3- 5 per cage) and conventional wood chips bedding. Food and water were available ad libitum. The mice were maintained in a pathogen free- environment on a 12- hr light/dark cycle at controlled temperature (22°C) and humidity (50- 60%). They were randomly assigned to experimental or control groups and behavioral testing was conducted during the light phase of the light/dark cycle. All possible efforts were made to minimize the number of animals used and their discomfort. + +<|ref|>text<|/ref|><|det|>[[115, 820, 421, 838]]<|/det|> +Global NAAA knock- out (Naaa- c) mice + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 156]]<|/det|> +Mice constitutively lacking the Naaa gene [(Naaa- ), B6N- Atm1BrdNaaatm1a(KOMP)wtsi/WtsiH], were acquired from MRC Harwell (Didcot, UK) through the EMMA-European Mouse Mutants Archive. Their generation and phenotypic profile are reported here31,32. + +<|ref|>text<|/ref|><|det|>[[111, 166, 883, 234]]<|/det|> +\(CD11b^{+}\) myeloid cell- specific NAAA knock- in \((Naaa^{CD11b^{+}})\) mice Transgenic mice overexpressing Naaa in CD11b+ myeloid cells \((Naaa^{CD11b^{+}})\) were produced by GenOway (France). Their generation and phenotypic profile are reported here31,33. + +<|ref|>text<|/ref|><|det|>[[111, 244, 881, 338]]<|/det|> +\(CD11b^{+}\) myeloid cell- specific NAAA knock- out \((Naaa^{CD11b - / - })\) mice NAAA- floxed mice \((Naaa^{f/f})\) mice were generated by mating Naaa- mice with flippase (Flp) FLPo- 10 mice (Jackson Lab, #011065). Subsequent recombination of \(Naaa^{f/f}\) with CD11b- Cre mice (Jackson Lab, #019696)] yielded \(Naaa^{CD11b - / - }\) mice, which were viable and fertile. + +<|ref|>text<|/ref|><|det|>[[111, 350, 881, 446]]<|/det|> +\(CD11b^{+}\) myeloid cell- specific PPAR- \(\alpha\) knock- out \((Ppara^{CD11b - / - })\) mice PPAR- \(\alpha\) floxed \((Ppara^{f/f})\) breeders were a kind gift of Dr Andrew S. Greenberg (Tufts University School of Medicine). \(Ppara^{f/f}\) mice were cross- bred with CD11b- Cre animals (Jackson Lab) to obtain \(Ppara^{CD11b - / - }\) mice, which were viable and fertile. + +<|ref|>sub_title<|/ref|><|det|>[[113, 474, 514, 493]]<|/det|> +## Depletion of monocyte-derived and other cells + +<|ref|>text<|/ref|><|det|>[[112, 497, 884, 589]]<|/det|> +\(C57Bl6/J\) mice were fed ad libitum for 20 days either with normal chow or with chow containing the colony- stimulating factor 1 (CSF1) receptor antagonist PLX5622 (1.2 g- kg- 1 of chow)34. Other groups of C57Bl6/J mice were treated with liposomes containing PBS (control) or clodronate (16 mL- kg- 1, IP), 3 days before and immediately after IL- 6 injection13,35. + +<|ref|>sub_title<|/ref|><|det|>[[113, 618, 492, 636]]<|/det|> +## Fluorescence Activated Cell Sorting (FACS) + +<|ref|>text<|/ref|><|det|>[[111, 641, 884, 902]]<|/det|> +Blood was drawn via cardiac puncture and incubated with ammonium- chloride- potassium (ACK) buffer for erythrocyte lysis. Leukocytes were stained with anti- CD11b, Zombie NIR, anti- Ly6C, and anti- Ly6G antibodies (Table S1) in antibody- staining buffer (0.2% BSA and 0.1% sodium azide in PBS) for 30 min at 4°C. Viable cell populations of interest were collected by FACS and quantified using FCS Express 7 (De Novo Software, Pasadena, CA). Gating strategies are illustrated in Figure S1. First, to separate debris from cells, we gated on the side scatter area (SSC- A) versus the forward scatter area (FSC- A) and selected the cell population with a high FSC- A/SSC- A ratio. This fraction was subjected to forward scatter height (FSC- H) versus FSC- A analysis to isolate single cells that displayed a ratio of \(\sim 1\). The single- cell population was further processed to remove dead cells. Zombie NIR signal- negative cells were collected for the next steps. All FACS processes included these initial gating steps to ensure that only visible single + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 884, 468]]<|/det|> +cells were analyzed. From this population, we gated cells based on the expression of cellular marker proteins to isolate cell types of interest. To focus on innate immune cells, we selected cells that were positive for CD11b by plotting the viable cell population as CD11b versus FSC- A and selecting the population that was \(\mathrm{CD11b^{+}}\) . This was plotted as Ly6C versus Ly6G with the \(\mathrm{Ly6C^{high}}\) cells being divided into groups depending on Ly6G expression. Monocytes were categorized as 'classical' ( \(\mathrm{CD11b^{+}Ly6C^{high}Ly6G^{low}}\) ) or 'non- classical' ( \(\mathrm{CD11b^{+}Ly6C^{low}Ly6G^{low}}\) ), while neutrophils were defined as \(\mathrm{CD11b^{+}Ly6C^{high}Ly6G^{high}}\) . Expression of the Naaa and Ppara genes in monocytes and neutrophils was quantified by reverse transcription- quantitative polymerase chain reaction (RT- qPCR), as described below, to confirm successful recombination. The spinal cord underwent enzymatic digestion and mechanical dissociation using a brain dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany), following the manufacturer's instructions. Subsequently, the dissociated spinal cord cells were stained with anti- CD11b and anti- CD45 antibodies (Table S1) in antibody- staining buffer (0.2% BSA and 0.1% sodium azide in PBS) for 30 min at \(4^{\circ}\mathrm{C}\) . Viable cell populations were collected by FACS and quantified using FCS Express 7 (De Novo Software, Pasadena, CA). Gating strategies are described above. Microglia was plotted as CD45 versus CD11b, which are both established markers for this cell lineage37. + +<|ref|>sub_title<|/ref|><|det|>[[115, 498, 238, 514]]<|/det|> +## RNA analyses + +<|ref|>text<|/ref|><|det|>[[110, 520, 884, 875]]<|/det|> +CD11b- cells and monocytes were homogenized using QiaShredder spin columns (Qiagen, Germantown, MD) and total RNA was isolated using the RNeasy Micro Kit (Qiagen). Total RNA was isolated from whole blood using the PureLink™ RNA Mini Kit (Invitrogen, Waltham, MA), following manufacturer's instructions. Before purification, samples were rendered genomic DNA free passing isolated RNA through a gDNA Eliminator spin column (Qiagen). RNA concentration and purity were determined using a SpectraMax M5 spectrophotometer (Molecular Devices, San Jose, CA). cDNA was synthesized using 20 ng of total RNA as input for the High- Capacity cDNA RT Kit with RNase Inhibitor (Applied BioSystems, Foster City, CA) with a final reaction volume of 20 µl. First- strand cDNA was amplified using TaqMan™ Universal PCR Master Mix (Invitrogen). RT- PCR primers and fluorogenic probes were purchased from Applied Biosystems [TaqMan(R) Gene Expression Assays]. We used TaqMan gene expression assays for mouse Actb (Mm00607939_s1), Hprt (Mm00446968_m1), Gapdh (Mm99999915_g1), Ppara (Mm00440939_m1), and Naaa (Mm00508965_m1) (Applied Biosystems). RT- PCRs were performed in 96- well plates using CFX96 Real- Time System (Bio- Rad, Hercules, USA). The Bestkeeper software was used to determine expression stability and the geometric mean of three + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 133]]<|/det|> +different housekeeping genes (Actb, Hprt, and Gapdh). The relative quantity of genes of interest was calculated by the \(2^{-\Delta \Delta \mathrm{ct}}\) method and expressed as fold change over controls. + +<|ref|>sub_title<|/ref|><|det|>[[115, 163, 320, 180]]<|/det|> +## Behavioral experiments + +<|ref|>sub_title<|/ref|><|det|>[[115, 211, 220, 227]]<|/det|> +## HP induction + +<|ref|>text<|/ref|><|det|>[[113, 234, 884, 328]]<|/det|> +We produced HP by administering IL- 6 (0.5 ng, 20 \(\mu \mathrm{l}\) ), TNF- \(\alpha\) (100 ng, 20 \(\mu \mathrm{l}\) ), or carrageenan (1%, 5 \(\mu \mathrm{L}\) ) into the surface of the mouse right hind paw \(^{19,38}\) . HP induction was confirmed 6 days later by injecting \(\mathrm{PGE}_2\) (100 ng, 20 \(\mu \mathrm{L}\) ) at the same site. Lasting ( \(\geq 6\) days) \(\mathrm{PGE}_2\) - induced heat hypersensitivity indicated successful induction \(^{19}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 358, 285, 375]]<|/det|> +## Heat hypersensitivity + +<|ref|>text<|/ref|><|det|>[[113, 380, 884, 497]]<|/det|> +We measured thermal nociceptive thresholds using a Hargreaves plantar test apparatus (San Diego Instruments, USA), as previously described \(^{31,39,40}\) . The thresholds were defined as the time (in seconds) at which mice withdrew their paws from the heat stimulus (withdrawal latency). A 15- s cut- off time was applied. Baseline thresholds were assessed in two consecutive readings made 2 days before the administration of priming agents. + +<|ref|>sub_title<|/ref|><|det|>[[115, 526, 300, 543]]<|/det|> +## Nociceptive thresholds + +<|ref|>text<|/ref|><|det|>[[113, 548, 884, 640]]<|/det|> +Nociceptive thresholds were measured using the tail immersion test, as described \(^{41}\) . Briefly, mice habituated to handling were restrained in a soft tissue pocket and tail withdrawal latency was measured by dipping the distal half of the tail into a water bath set at \(54^{\circ}\mathrm{C}\) . Cut- off time was 10 s. Two tail- withdrawal measurements (separated by 30 s) were recorded and averaged. + +<|ref|>sub_title<|/ref|><|det|>[[115, 669, 267, 686]]<|/det|> +## Motor coordination + +<|ref|>text<|/ref|><|det|>[[113, 692, 884, 783]]<|/det|> +Motor coordination was evaluated using an accelerating Rotarod apparatus (Ugo Basile; rod diameter: 2 cm). Briefly, mice were familiarized to the procedure in three consecutive sessions with speed gradually increasing from 4 to 40 rpm over a 5- min period. Experiments were conducted 24 h after the last training session. + +<|ref|>sub_title<|/ref|><|det|>[[115, 812, 615, 830]]<|/det|> +## Pharmacokinetic (PK) and pharmacodynamic experiments + +<|ref|>text<|/ref|><|det|>[[113, 835, 884, 902]]<|/det|> +Male mice (n = 5 per group) were given a single injection of ARN726 (10 mg- kg \(^{- 1}\) , IP) or vehicle and were euthanized under isoflurane anesthesia 15, 30, 60, and 120 min later. Blood (0.5 mL) was collected by cardiac puncture into EDTA- rinsed syringes and transferred into polypropylene + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 179]]<|/det|> +plastic tubes (1 mL) containing spray-coated K2- EDTA. Plasma was prepared by centrifugation at \(1450 \times g\) at \(4^{\circ}C\) for 15 min and transferred into polypropylene tubes, which were immediately frozen and stored at \(- 80^{\circ}C\) . Spinal cords were removed by hydraulic extrusion, snap- frozen on dry ice, and stored at \(- 80^{\circ}C\) until processed. + +<|ref|>sub_title<|/ref|><|det|>[[113, 193, 262, 210]]<|/det|> +## Analyte extraction + +<|ref|>text<|/ref|><|det|>[[110, 215, 886, 666]]<|/det|> +We extracted PEA, OEA, anandamide, and ARN726 from plasma and spinal cord tissue, as described \(^{40,42,43}\) . Briefly, plasma (0.1 mL) was transferred into 8- mL glass vials, and proteins were precipitated by the addition of 0.5 mL ice- cold acetonitrile containing 1% formic acid and the following internal standards: \([^{2}\mathrm{H}_{4}]\) - PEA (20 nM), \([^{2}\mathrm{H}_{4}]\) - OEA (1 nM), \([^{2}\mathrm{H}_{4}]\) - anandamide (10 nM), and ARN077 (200 nM). Frozen spinal cords were pulverized on dry ice. Spinal tissue samples ( \(\sim 20 \text{mg}\) each) were transferred into 2 ml Precellys CK- 14 soft tissue tubes (Bertin, Rockville, MD) and homogenized in 0.5 mL of ice- cold acetonitrile containing 1% formic acid and the internal standards listed above. Samples were homogenized for 1 min using a Bertin homogenizer at \(4^{\circ}C\) in 15 s/cycle for 2 cycles with 20 s pause between cycles. Samples were stirred vigorously for 30 s and centrifuged again at \(830 \times g\) at \(4^{\circ}C\) for 15 min. The supernatants were loaded onto pre- washed [with water/acetonitrile (1:4 vol/vol)] Enhanced Matrix Removal (EMR)- Lipid cartridges (Agilent Technologies, Santa Clara, CA) and eluted under positive pressure (3- 5 mmHg). Tissue pellets were rinsed with wash solution (0.2 mL), stirred for 30 s, and centrifuged at \(830 \times g\) for 15 min at \(4^{\circ}C\) . The supernatants were collected, transferred onto EMR cartridges, eluted, and pooled with the first eluate. The cartridges were washed again with wash solution (0.2 mL) and pressure was gradually increased to 10 mmHg for maximal analyte recovery. Eluates were dried under a gentle stream of \(N_{2}\) and reconstituted in 0.1 mL methanol containing 0.1% formic acid. Samples were transferred to deactivated glass inserts (0.2 mL) placed inside amber glass vials (2 mL, Agilent Technologies). + +<|ref|>sub_title<|/ref|><|det|>[[115, 697, 267, 714]]<|/det|> +## Lipid quantification + +<|ref|>text<|/ref|><|det|>[[112, 718, 886, 907]]<|/det|> +PEA, OEA, and anandamide were fractionated using a 1260 series LC system (Agilent Technologies) consisting of a binary pump, degasser, thermostated autosampler, and column compartment coupled to a 6410B triple quadrupole mass spectrometric detector (MSD; Agilent Technologies). Analytes were separated on an Eclipse PAH column (1.8 \(\mu \text{m}\) , 2.1 x 50 mm; Agilent Technologies) with a mobile phase consisting of 0.1% formic acid in water as solvent A and 0.1% formic acid in methanol as solvent B. A linear gradient was used: 75.0% B at 0 time to 80.0% B in 5.0 min, change to 95% B at 5.01 min continuing to 6.0 min, change to 75% B at 6.01 min, and hold till 8.0 min for column re- equilibration and stop time. The column temperature was maintained + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 884, 372]]<|/det|> +at \(45^{\circ}C\) and the autosampler temperature at \(10^{\circ}C\) . The injection volume was \(2 \mu \mathrm{L}\) , the flow rate was \(0.3 \mathrm{mL} / \mathrm{min}\) , and the total analysis time was \(15.5 \mathrm{min}\) . To prevent carryover, the injection needle was washed three times in the autosampler port for \(30 \mathrm{s}\) before each injection, using a wash solution consisting of \(10\%\) acetone in water/methanol/isopropanol/acetonitrile \((1:1:1:1, \mathrm{vol})\) . The MSD was operated in the positive electrospray ionization (ESI) mode. Analytes were quantified by multiple reaction monitoring (MRM) of the following transitions: anandamide \(= m / z\) \(348.3 > 62.2\) , \([^{2}\mathrm{H}_{4}]\) - anandamide \(= m / z\) \(352.3 > 66.2\) , PEA \(= m / z\) \(300.3 > 62.2\) , \([^{2}\mathrm{H}_{4}]\) - PEA \(= m / z\) \(304.3 > 66.2\) , OEA \(= m / z\) \(326.3 > 62.1\) , \([^{2}\mathrm{H}_{4}]\) - OEA \(= m / z\) \(330.3 > 66.1\) . Capillary and nozzle voltages were 3,000 and 1,900 V, respectively. The drying gas temperature was \(300^{\circ}C\) with a flow of \(5 \mathrm{mL} / \mathrm{min}\) . The sheath gas temperature was \(300^{\circ}C\) with a flow of \(12 \mathrm{mL} / \mathrm{h}\) . Nebulizer pressure was set at \(40 \mathrm{psi}\) . We used the MassHunter software (Agilent Technologies) for instrument control, data acquisition, and analysis. + +<|ref|>sub_title<|/ref|><|det|>[[115, 401, 297, 418]]<|/det|> +## ARN726 quantification + +<|ref|>text<|/ref|><|det|>[[110, 423, 884, 705]]<|/det|> +ARN726 was fractionated using an Eclipse Plus \(\mathrm{C_{18}}\) column \([1.8 \mu \mathrm{m}, 2.1 \times 30 \mathrm{mm}]\) (Agilent Technologies)] with a mobile phase of \(0.1\%\) formic acid in water as solvent A and \(0.1\%\) formic acid in methanol as solvent B. A linear gradient was used: start \(60\%\) B to \(80\%\) B in \(2.5 \mathrm{min}\) , changed to \(95\%\) B at \(2.51 \mathrm{min}\) and continued till \(3.0 \mathrm{min}\) , changed to \(60\%\) B at \(3.01 \mathrm{min}\) and continued till \(5.0 \mathrm{min}\) . The flow rate was \(0.3 \mathrm{mL} / \mathrm{min}\) till \(3.0 \mathrm{min}\) and then \(0.4 \mathrm{mL} / \mathrm{min}\) from \(3.01 \mathrm{min}\) to \(5.0 \mathrm{min}\) . The column temperature was maintained at \(40^{\circ}C\) and the auto- sampler temperature was \(9^{\circ}C\) . The injection volume was \(2 \mu \mathrm{L}\) . The MS was operated in positive mode. ARN726 was quantified by MRM using ARN077 as an internal standard and the following transitions: \(\mathrm{ARN}726 = m / z 269.19 > 83.1\) , \(\mathrm{ARN}077 = m / z 292.16 > 74.1\) . Capillary and nozzle voltages were \(1800\) and \(1200 \mathrm{V}\) , respectively. Nebulizer pressure was set at \(10 \mathrm{psi}\) . Drying gas temperature was \(300^{\circ}C\) , with a flow of \(12 \mathrm{L} / \mathrm{min}\) . Sheath gas temperature was \(320^{\circ}C\) with a flow of \(10 \mathrm{L} / \mathrm{min}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 720, 252, 737]]<|/det|> +## PK data analysis + +<|ref|>text<|/ref|><|det|>[[113, 749, 884, 864]]<|/det|> +Results from PK experiments were analyzed using a noncompartmental model44. Maximal concentration \((C_{max})\) and time at which maximal concentration was reached \((T_{max})\) were determined by visual inspection of averaged data. Other PK parameters – including area under the curve \((AUC)\) , elimination constant \((K_{el})\) , and half- life time of elimination \((t_{1 / 2})\) – were determined using the equations described here44. + +<|ref|>sub_title<|/ref|><|det|>[[115, 886, 283, 903]]<|/det|> +## Statistical analyses + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 228]]<|/det|> +Sample size for behavioral studies was predetermined by power analysis ( \(\alpha = 0.05\) , 1- \(\beta = 0.8\) , effect size \(\sim 35\%\) , \(n \geq 7\) per group). Data analyses were performed using GraphPad Prism version 9.1 (La Jolla, CA) under blinded conditions. All data are expressed as means \(\pm\) SEM. Differences between groups were assessed by either unpaired two- tailed Student's \(t\) test or analysis of variance (ANOVA) (one- way or two- way) followed by Dunnett's or Bonferroni's post hoc test unless otherwise noted. A \(P < 0.05\) was considered statistically significant. + +<|ref|>sub_title<|/ref|><|det|>[[113, 260, 183, 277]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[113, 306, 465, 324]]<|/det|> +## NAAA is required for the induction of HP + +<|ref|>text<|/ref|><|det|>[[111, 328, 886, 589]]<|/det|> +Following an established protocol \(^{19,38}\) , we evoked HP in male mice by intraplantar (i.pl.) injection of IL- 6 (0.5 ng) (Fig. 1A). As expected, the cytokine produced a state of ipsilateral heat hypersensitivity that lasted \(< 24h\) and was followed by a persistent enhancement of the nocifensive response to the proalgesic eicosanoid \(\mathrm{PGE}_2\) (100 ng, i.pl., administered 6 days later in the same paw). The initial hypersensitivity was attenuated when mice were given, 2 h before IL- 6, either the selective NAAA inhibitor ARN19702 (30 mg- kg \(^{- 1}\) , IP) \(^{45,46}\) (Fig. 1B1) or gabapentin (50 mg- kg \(^{- 1}\) , IP) (Fig. 1C1). Even though they exerted acute antinociceptive effects, the two agents did not prevent HP induction, as shown by their inability to normalize the response to \(\mathrm{PGE}_2\) in IL- 6- primed mice (Fig. 1B2, C2). The results, which were replicated in female animals (Fig. S2A- C), confirm prior studies indicating that the inflammatory reaction to a priming stimulus and the initiation of HP overlap in time but are governed by distinct mechanisms \(^{22,23}\) . + +<|ref|>text<|/ref|><|det|>[[111, 593, 886, 857]]<|/det|> +As the initiation and maintenance phases of HP are separated by an incubation period of \(\sim 72\) \(h^{23,24}\) , we asked whether administering ARN19702 or gabapentin during this time interval might affect HP progression. We gave once- daily injections of ARN19702 (30 mg- kg \(^{- 1}\) , IP), gabapentin (50 mg- kg \(^{- 1}\) , IP) or their vehicles to male and female mice on the first, second, and third day after IL- 6 (Fig. 1A). When tested using this protocol, ARN19702 blocked HP induction (Fig. 1D, data from female mice are shown in Fig. S2D1, D2) whereas gabapentin did not (Fig. S3). The same ARN19702 regimen was also protective in male mice challenged with either of two different priming triggers, TNF- \(\alpha\) (100 ng, i.pl.) (Fig. S4A, B) and carrageenan (1%, i.pl.) (Fig. S4C). Underscoring the strict time- dependence of its effects, ARN19702 failed to prevent HP when administered in male mice for shorter time intervals (e.g., only on days 1 and 2 after IL- 6) or after the end of the incubation period (e.g., on days 4 and 5 after IL- 6) (Fig. 1E). + +<|ref|>text<|/ref|><|det|>[[112, 860, 884, 903]]<|/det|> +To further probe NAAA's role in priming induction, we studied the response to IL- 6 in male mice constitutively lacking the enzyme. Previous studies have shown that inflammatory and acute + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 348]]<|/det|> +nociceptive responses are blunted in homozygous Naaa- mice31,32. Consistent with those findings, the mutants' acute reaction to IL- 6 was weaker than that of their wild- type littermates (Fig. 1F1). Naaa- mice exhibited a normal acute response to carrageenan, however, possibly owing to the greater intensity of this stimulus (Fig. S4D1). Furthermore, as seen with ARN19702 treatment, Naaa- mice failed to develop HP after injection of IL- 6 (Fig. 1F2) or carrageenan (Fig. S4D2). Heterozygous Naaa- mice also showed a diminished inflammatory reaction to IL- 6 (Fig. 1F1) but not to carrageenan (Fig. S4D1) and were unable to progress to HP after administration of either stimulus (Fig. 1F2). The findings indicate that inflammatory agents induce HP through a mechanism that requires NAAA activity. Such requirement is (i) independent of NAAA's facilitatory role in the nociceptive response evoked by the agents; (ii) restricted to the 72- h incubation period that precedes the emergence of HP; and (iii) sexually monomorphic. + +<|ref|>sub_title<|/ref|><|det|>[[115, 377, 568, 397]]<|/det|> +## NAAA initiates HP by suppressing PPAR- \(\alpha\) signaling + +<|ref|>text<|/ref|><|det|>[[111, 400, 886, 799]]<|/det|> +Activation of the nuclear receptor PPAR- \(\alpha\) by the endogenous NAAA substrates, PEA and OEA, underpins the anti- inflammatory and anti- nociceptive properties of NAAA inhibitors32,46 as well as the agents' ability to stop the transition to pain chronicity after sciatic nerve ligation or i.pl. formalin injection31. To determine whether a similar mechanism might be involved in halting HP induction, we challenged male mice with IL- 6 (0.5 ng, i.pl.) and treated them once daily for the subsequent three days with either ARN19702 (30 mg- kg \(^{- 1}\) , IP) or a combination of ARN19702 and a maximally effective dose of the selective PPAR- \(\alpha\) antagonist GW6471 (4 mg- kg \(^{- 1}\) , IP)31,46,47. When administered alone, ARN19702 prevented HP induction, and this effect was abolished by addition of GW6471 (Fig. 2A). By contrast, co- administration of ARN19702 with maximally effective doses of selective antagonists for PPAR- \(\gamma\) (T0070907, 1 mg- kg \(^{- 1}\) , IP)48 or CB2 cannabinoid receptors (AM630, 3 mg- kg \(^{- 1}\) , IP)49, which have been implicated in some actions of PEA50,51, did not significantly alter the response to ARN19702 (Fig. 2B, C). Further support for a role of PPAR- \(\alpha\) in mediating such response was provided by the finding that PEA (30 mg- kg \(^{- 1}\) , subcutaneous) and the synthetic PPAR- \(\alpha\) agonist GW7647 (10 mg- kg \(^{- 1}\) , IP) prevented priming when administered on days 1- 3 post IL- 6 (Fig. 2D, E). The results suggest that inhibiting NAAA activity during the incubation phase of HP stops the progression to pain chronicity by heightening PPAR- \(\alpha\) signaling. + +<|ref|>sub_title<|/ref|><|det|>[[115, 818, 464, 836]]<|/det|> +## Role of peripheral NAAA in HP induction + +<|ref|>text<|/ref|><|det|>[[115, 840, 884, 886]]<|/det|> +Microglia and neurons of the central nervous system (CNS) express NAAA26,33,52. To assess whether the development of HP might involve a central pool of the enzyme, we utilized the \(\beta\) - + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 444]]<|/det|> +lactam- based NAAA inhibitor ARN726 (Fig. 2F), which is potent and selective but has limited CNS activity \(^{53}\) . Fifteen minutes after systemic administration, ARN726 (10 mg- kg \(^{- 1}\) , IP) reached peak concentrations of 333.8 ng- mL \(^{- 1}\) in plasma and 42.7 ng- mg \(^{- 1}\) in spinal cord (Fig. 2F, other PK properties of ARN726 are reported in Table S1). The tissue- to- plasma ratio calculated from these concentrations \((= 0.13)\) is \(\sim 56\%\) lower than the brain- to- plasma ratio achieved by ARN19702 after oral administration of a 3 mg- kg \(^{- 1}\) dose \((= 0.30)^{45}\) . Concomitant with its plasma peak, ARN726 caused a transitory but substantial elevation of PEA and OEA levels in circulation (Fig. 2G, H) but had no such effect in spinal cord (Fig. 2J, K). Moreover, confirming its selectivity for NAAA \(^{53}\) , ARN726 did not affect the circulating levels of anandamide (Fig. 2I, L), an endocannabinoid mediator that is structurally related to PEA and OEA but is hydrolyzed by a different lipid amidase \(^{26}\) . Despite this lack of CNS activity, systemic administration of ARN726 (1- 30 mg- kg \(^{- 1}\) , IP) on days 1- 3 after IL- 6 blocked the development of priming with considerable potency (ED \(_{50} = 2.2\) mg- kg \(^{- 1}\) ) (Fig. 2M, Fig. S5). By contrast, such effect was observed when ARN726 (30 ng) was delivered directly to the spinal cord via intrathecal infusion (Fig. 2N). We conclude that HP induction requires peripheral NAAA activity. + +<|ref|>sub_title<|/ref|><|det|>[[114, 472, 566, 490]]<|/det|> +## NAAA in CD11b \(^+\) myeloid cells initiates HP induction + +<|ref|>text<|/ref|><|det|>[[112, 495, 884, 899]]<|/det|> +Outside the CNS, NAAA is predominantly expressed in monocyte- derived cells and B- lymphocytes \(^{53,54}\) . Because monocytes and macrophages have been implicated in the progression to pain chronicity \(^{12,13}\) , we examined whether NAAA- regulated lipid signaling in this cell population might contribute to HP. We generated mutant mice that selectively lack NAAA in CD11b \(^+\) cells (Fig. 3A), which include monocytes, macrophages, and neutrophils \(^{55,56}\) . Successful recombination was confirmed by subjecting blood samples from the newly generated mouse line to fluorescence- activated cell sorting (FACS) coupled to RT- qPCR analysis (Fig. 3B, C). Naaa \(^{CD11b / - }\) mice were viable, fertile, and had normative nociceptive thresholds and motor coordination (Fig. S6A, B). When compared to Naaa \(^{fl / fl}\) littermates, which were used as controls, Naaa \(^{CD11b / - }\) mice phenocopied global Naaa knockouts in that they exhibited an attenuated acute reaction to IL- 6 (Fig. 3D) and were resistant to priming (Fig. 3E). The results suggest that NAAA- expressing CD11b \(^+\) myeloid cells participate in both the initial hypersensitivity evoked by IL- 6 and the emergence of HP but are indispensable only for the latter. Experiments with transgenic mice overexpressing NAAA in CD11b \(^+\) cells further strengthened this conclusion. Prior work suggests that resident lung macrophages in Naaa \(^{CD11b + }\) mice are inherently hyperactive \(^{33}\) , raising the possibility that NAAA overexpression in myeloid cells might promote a constitutively primed phenotype. Consistent with this prediction, the nocifensive response to PGE \(_{2}\) was heightened in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 883, 156]]<|/det|> +male Naaa \(^{CD11b^{+}}\) mice that had not been previously exposed to IL- 6, but not in their wild- type littermates (Fig. 3F, G). Together, the findings indicate that NAAA activity in peripheral CD11b \(^{+}\) myeloid cells is both necessary and sufficient for the induction of HP. + +<|ref|>sub_title<|/ref|><|det|>[[115, 185, 614, 204]]<|/det|> +## PPAR- \(\alpha\) in CD11b \(^{+}\) myeloid cells suppresses HP induction + +<|ref|>text<|/ref|><|det|>[[112, 210, 884, 450]]<|/det|> +To further probe the role of lipid- dependent PPAR- \(\alpha\) signaling in the initiation of HP, we generated mice that lack this nuclear receptor in CD11b \(^{+}\) cells (Fig. 4A- C), with the expectation that, like mutants overexpressing NAAA in the same cells, they would also be constitutively primed. Confirming this prediction, the nocifensive response to \(\mathrm{PGE}_{2}\) was enhanced in male IL- 6- naive \(Ppara^{CD11b - / - }\) mice, relative to their \(Ppara^{fl / fl}\) littermates (Fig. 4D, E). Of note, PPAR- \(\alpha\) deletion in CD11b \(^{+}\) cells not only sensitized mice to \(\mathrm{PGE}_{2}\) but also annulled the protective effect caused by administration of ARN19702 during HP incubation (Fig. 4F- I). The results suggest that NAAA- regulated PPAR- \(\alpha\) signaling in peripheral CD11b \(^{+}\) myeloid cells controls the transition to HP, most likely through a cell- autonomous mechanism, and is the primary target for the protective effect of NAAA inhibitors. + +<|ref|>sub_title<|/ref|><|det|>[[115, 479, 803, 498]]<|/det|> +## Monocytes and macrophages, but not neutrophils, are required for HP induction + +<|ref|>text<|/ref|><|det|>[[110, 503, 884, 905]]<|/det|> +Lastly, we used two mechanistically distinct cell depletion strategies as an independent means to assess the role of monocytes and macrophages in HP induction. First, we treated wild- type mice with liposome- encapsulated clodronate, a peripherally restricted bisphosphonate- containing compound that selectively induces apoptosis in monocyte- derived cells \(^{57 - 59}\) . Male wild- type mice were given, with a 3- day interval, two IP injections of liposomes loaded with either clodronate or its vehicle (PBS) (Fig. 5A) \(^{13}\) . FACS analyses of blood and spinal cord showed that, compared to PBS liposomes, clodronate liposomes lowered the number of circulating Ly6C \(^{\text{high}}\) ('classical') monocytes without affecting Ly6C \(^{\text{low}}\) ('non- classical') monocytes or spinal cord microglia (Fig. 5B, C, Fig. S7A). Consistent with prior results \(^{60}\) , the number of neutrophils was increased by clodronate (Fig. 5D), likely owing to monocyte death \(^{61}\) . Separate groups of liposome- treated mice were exposed to IL- 6 (0.5 ng, i.p.l.) and, six days later, were challenged with \(\mathrm{PGE}_{2}\) (100 ng, i.p.l.) (Fig. 5E). As expected \(^{62}\) , administration of clodronate liposomes suppressed the initial hypersensitivity evoked by IL- 6 (Fig. 5F1). In addition to this acute effect, clodronate liposomes normalized the response to \(\mathrm{PGE}_{2}\) (Fig. 5F2), indicating that priming induction was also blocked. In a second set of experiments, we exposed for three weeks male mice to chow containing either the CSF1 receptor antagonist PLX5622 (1.2 g- kg \(^{- 1}\) chow) or its vehicle (Fig. 5G). PLX5622 treatment reduced the numbers of circulating monocytes (both 'classical' and 'non- classical'), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 884, 252]]<|/det|> +neutrophils (Fig. 5H- J), and spinal cord microglia (Fig. S7B) \(^{63,64}\) , but caused no observable change in baseline nociceptive thresholds or motor coordination (Fig. S8). Closely matching the results obtained with clodronate, treatment with PLX5622 (Fig. 5K) prevented both the acute response to IL- 6 (Fig. 5L1) and the emergence of HP (Fig. 5L2). In sum, clodronate and PLX5622 lowered monocyte numbers and prevented priming but had opposite effects on neutrophil number. We interpret these findings as indicating that monocytes and/or macrophages, but not neutrophils, are required for the induction of HP. + +<|ref|>sub_title<|/ref|><|det|>[[115, 281, 213, 299]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 312, 884, 530]]<|/det|> +The results identify NAAA- regulated signaling at PPAR- \(\alpha\) as a molecular switch that directs monocytes and macrophages to initiate HP in mice exposed to an inflammatory stimulus. Three independent lines of evidence support this conclusion. First, inhibiting NAAA or activating PPAR- \(\alpha\) during the incubation phase of HP stops its consolidation; global NAAA deletion has a similar effect. Second, mutant mice that selectively lack NAAA in CD11b \(^+\) cells – which include monocytes, macrophages, and neutrophils – are resistant to priming, whereas mice that overexpress NAAA or lack PPAR- \(\alpha\) in the same cell lineage are constitutively primed. Lastly, depletion of monocytes and macrophages, but not neutrophils, renders mice refractory to priming. + +<|ref|>text<|/ref|><|det|>[[112, 540, 884, 900]]<|/det|> +The model posits that, in tissue- resident macrophages of non- primed mice, high baseline concentrations of NAAA main substrates (PEA, OEA) maintain these cells in a resting (surveillant) state by constitutively activating PPAR- \(\alpha\) - dependent transcription. In non- primed animals, PGE2 produces a short- lived ( \(\sim 2h\) ) nocifensive response by ligating protein kinase A- coupled EP2/EP4 receptors in non- peptidergic isolectin- B4 \(^+\) nociceptive neurons \(^{17,18,21}\) (Fig. 6, left panel). Priming agents such as IL- 6 have two pharmacologically distinguishable effects: on the one hand, they evoke a self- resolving inflammatory reaction whose sensory signs can be suppressed by standard analgesic drugs \(^{22,23}\) ; on the other, they initiate priming through an as- yet- unknown mechanism that is insensitive to such drugs. This initial phase is followed by a quiescent 72h- long interval that precedes stabilization of the primed state \(^{23,24}\) (Fig. 6, center panel). During this incubation period, NAAA recruitment is both necessary and sufficient to enable the functional interaction of monocytes and macrophages with non- peptidergic nociceptors. Based on our data, the neuroplastic modifications that support the maintenance of a primed state in these neurons – including the recruitment of protein kinase Cε- dependent transduction pathways \(^{19- 21}\) and the local translation and retrograde axonal transport of CREB \(^{25}\) – depend on such interaction (Fig. 6, center + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 182]]<|/det|> +and right panels). The model emphasizes three points: first, the role of NAAA- regulated PPAR- \(\alpha\) signaling in controlling macrophage and monocyte activation; second, the cooperation between these cells and first- order nociceptors in the induction of priming; third, the subdivision of this process into mechanistically distinct components. These points are briefly discussed below. + +<|ref|>text<|/ref|><|det|>[[111, 194, 884, 580]]<|/det|> +The regulatory functions of PPAR- \(\alpha\) in monocyte- derived cells have been the object of intensive investigation. Studies have shown that, when ligand- bound, PPAR- \(\alpha\) suppresses inflammatory signaling in macrophages \(^{65,66}\) and promotes their polarization toward an M2 phenotype associated with tissue protection \(^{67,68}\) . There is also strong evidence that NAAA's main substrates, PEA and OEA \(^{26}\) , are the ligands responsible for these effects. First, both substances are potent PPAR- \(\alpha\) agonists \(^{27,29,69}\) . Second, monocytes and macrophages express the zinc hydrolase N- acyl phosphatidylethanolamine phospholipase D (NAPE- PLD), which constitutively generates PEA and OEA \(^{70,71}\) , as well as NAAA, which deactivates them \(^{26}\) . Third, inflammatory stimuli suppress NAPE- PLD expression (via transcriptional regulation) \(^{72}\) and enhance NAAA activity (via autoproteolysis of its inactive polypeptide precursor) \(^{73}\) , causing an overall reduction in the amount of PEA and OEA available for signaling \(^{30,32,74}\) . Finally, NAAA inhibitors exert profound anti- inflammatory and anti- nociceptive effects by restoring baseline PEA and OEA levels and, consequently, enhancing PPAR- \(\alpha\) - dependent transcription \(^{30,32,46,75}\) . In sum, the available evidence supports the possibility that NAAA- dependent interruption of PPAR- \(\alpha\) signaling instructs monocytes and macrophages to initiate priming. Identifying the molecular modifications that codify these instructions is an important area for future research. + +<|ref|>text<|/ref|><|det|>[[112, 591, 884, 899]]<|/det|> +Neuroimmune interactions are critical to the establishment of persistent painful states \(^{10,15,76}\) . For example, experiments in mouse models of arthritis and vincristine- induced toxicity have shown that the infiltration of blood- derived CXCR1- expressing monocytes into the DRG drives initiation and maintenance of persistent pathological nociception \(^{15,16}\) . Similarly, after nerve injury in mice, blood- borne monocytes migrate to the spinal cord, where they proliferate, differentiate, and act in synergy with microglia to initiate pain chonification \(^{12- 14}\) . The present study expands this body of knowledge in two ways. First, it shows that monocytes and macrophages, but not neutrophils, are necessary for priming induction. This conclusion is based on the finding that clodronate and PLX5622 exert opposite effects on neutrophil numbers yet both deplete monocyte- derived cells and prevent priming. However, the results do not exclude the possibility that neutrophils, which promote widespread persistent nociception in a mouse model of fibromyalgia \(^{77}\) , might play an ancillary role in HP. Second, and more importantly, our findings identify NAAA as a molecular switch that, when engaged, directs monocyte- derived cells to interact with first- order nociceptors + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 886, 599]]<|/det|> +and enable priming19- 21. The precise nature of this interaction is unknown but is likely to involve the release of chemical signals from monocytes and/or macrophages. Possible candidates for this role, which should be evaluated in future work, include lipid mediators such as sphingosine- 1- phosphate78,79, reactive oxygen species such as hydrogen peroxide15,80, and growth factors81,82. Chronic pain states are widely heterogeneous in causes, symptoms, impact on function, and temporal development3. This diversity justifies skepticism toward a simplistic view of the progression to pain chronicity as transformation of one mechanistic type of pain (e.g., acute pain associated with injury) into another (e.g., neuropathic pain)83. Nevertheless, our results do support the existence of at least one transition point which must be passed in order for lasting pathological nociception to be established. In previous work, we found that inhibiting spinal cord NAAA 48 to 72 hours after peripheral tissue damage stops the emergence of lasting pathological nociception in mice31. Similarly, in the present study, we showed that disabling peripheral NAAA in the 72 hours following administration of an inflammatory stimulus prevents priming development. In both cases, the protective effects of NAAA blockade depend on PPAR- \(\alpha\) activation, are sexually monomorphic, and are strictly restricted to a three- day time window post- injury. The last property prompted us to designate such effects with the term 'algostatic' (ancient Greek ἀλγος 'pain' and ἰστάντοι 'to stop')31 to underscore the fact that they are independent of the ability of NAAA inhibitors to attenuate ongoing nociception or inflammation. The same term may be applicable to other drugs, such as metformin, that are also able to stop the development of persistent pathological nociception in animal models84- 86. If effective in the clinic, algostatic agents might offer a strategy to prevent chronic pain after invasive surgery and other kinds of physical trauma. + +<|ref|>sub_title<|/ref|><|det|>[[115, 616, 277, 633]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[115, 639, 883, 682]]<|/det|> +We are grateful to Parwinder Singh Uppal, Kayla Chang and Vitaly Perkov for their help with behavioral studies and to GE Nutrients for the kind gift of Levegan+ PEA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 714, 282, 730]]<|/det|> +## Competing interest + +<|ref|>text<|/ref|><|det|>[[113, 736, 884, 898]]<|/det|> +D.P. and M.M. are inventors in patents that protect ARN19702 and other NAAA inhibitors, owned by the University of California, the University of Parma, the University of Urbino, and the Fondazione Istituto Italiano di Tecnologia (no. 13/898,225, filed 20 May 2013, published 3 April 2014; no. 62/337,744, filed 17 May 2016, published 23 November 2017). D.P. and Y.F. are inventors in a patent application that protects the algostatic effects of NAAA inhibitors, filed by the University of California (no. 63/166,134, filed 25 March 2021, published 29 September 2022). The other authors have declared no competing interest. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 113, 255, 131]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[112, 137, 883, 180]]<|/det|> +All data supporting the findings of this study are readily available within the paper and its supplementary files. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 115, 880, 555]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 560, 622, 578]]<|/det|> +
Figure 1: NAAA is required to induce hyperalgesic priming.
+ +<|ref|>text<|/ref|><|det|>[[111, 583, 886, 890]]<|/det|> +(A) Model used in the present study. In most experiments, priming was induced in mice by intrapalantar administration of IL-6 and was assessed 6 days later by injecting \(\mathrm{PGE}_2\) at the same site. Ipsilateral heat hypersensitivity (paw withdrawal latency, seconds) was measured under baseline (BL) conditions (-2d, -1d) and immediately after administration of IL-6 (0d), \(\mathrm{PGE}_2\) (6d), or vehicle (0d and 6d). The blue bar marks the incubation period for priming. +(B, C) Self-resolving heat hypersensitivity elicited by IL-6 in mice that had received 2 h earlier (B1) ARN19702 (ARN, 30 mg-kg\(^{-1}\); green circles), (C1) gabapentin (GBP, 50 mg-kg\(^{-1}\); green triangles), or their vehicles (Veh, magenta circles). (B2, C2) Effects of \(\mathrm{PGE}_2\) in IL-6-primed mice treated with ARN19702 (B2), gabapentin (C2), or their vehicles. +(D) IL-6-primed mice were treated on 1d-3d with vehicle or ARN19702 (30 mg-kg\(^{-1}\)). (D1) Response to IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). +(D2) Effects of \(\mathrm{PGE}_2\) in IL-6-primed mice that had received vehicle (magenta circles) or ARN19702 (green triangles). Right panel, area under the curve (AUC). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 156]]<|/det|> +(E) IL-6-primed mice were treated with vehicle (magenta circles) or ARN19702 (30 mg·kg-1, filled triangles) for 2 days on 1d-2d, 3d-4d, or 5d-6d, and the response to PGE2 was assessed 3 days later. ARN19702 on (E1) 1d-2d, (E2) 3d-4d, and (E3) 5d-6d. + +<|ref|>text<|/ref|><|det|>[[111, 160, 884, 202]]<|/det|> +(F) Effects of (F1) IL-6 (0d) and (F2) PGE2 (6d) in wild-type (Wt, magenta circles), heterozygous Naaa+/− (green triangles), and homozygous Naaa+/− mice (green squares). + +<|ref|>text<|/ref|><|det|>[[111, 206, 884, 299]]<|/det|> +Data are presented as mean ± S.E.M. and were analyzed by two-way repeated measure ANOVA followed by Bonferroni's multiple comparison test, when necessary. \(^{***}P < 0.001\) , \(^{**}P < 0.01\) , \(^{*}P < 0.05\) , versus baseline; \(^{###}P < 0.001\) , \(^{##}P < 0.01\) , \(^{#}P < 0.05\) , versus vehicle or Wt. ns, nonsignificant (n = 8-16 mice per group). Dotted lines indicate baseline withdrawal latency. + +<|ref|>image<|/ref|><|det|>[[111, 352, 884, 754]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 761, 884, 805]]<|/det|> +
Figure 2: Role of peripheral NAAA and PPAR-\(\alpha\) signaling in hyperalgesic priming induction.
+ +<|ref|>text<|/ref|><|det|>[[111, 810, 884, 902]]<|/det|> +(A-C) IL-6- primed mice were treated on 1d-3d with ARN19702 (ARN, 30 mg·kg- 1) alone or in combination with GW6471 (4 mg·kg- 1), T0070907 (1 mg·kg- 1), AM630 (3 mg·kg- 1), or an appropriate vehicle. Effects of PGE2 in mice treated with the following: (A) ARN19702 alone (green triangles), ARN19702 plus GW6471 (open squares), or vehicle (Veh, magenta circles); (B) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 131]]<|/det|> +ARN19702 alone (green triangles), ARN19702 plus T0070907 (open squares), or vehicle (magenta circles); and (C) ARN19702 alone (green triangles), ARN19702 plus AM630 (open squares), or vehicle (magenta circles). + +<|ref|>text<|/ref|><|det|>[[111, 135, 884, 155]]<|/det|> +squares), or vehicle (magenta circles). + +<|ref|>text<|/ref|><|det|>[[111, 160, 884, 181]]<|/det|> +(D, E) IL- 6- primed mice were treated on 1d- 3d with PEA (30 mg- kg \(^{- 1}\) ) or GW7647 (10 mg- kg \(^{- 1}\) ). + +<|ref|>text<|/ref|><|det|>[[111, 185, 884, 227]]<|/det|> +Effects of \(\mathsf{PG E}_{2}\) in mice treated with (D) PEA (green circles) or vehicle (magenta circles); and (E) GW7647 (green circles) or vehicle (magenta circles). + +<|ref|>text<|/ref|><|det|>[[111, 231, 884, 298]]<|/det|> +(F) Time- course of ARN726 concentrations in plasma (green circles) and spinal cord (gray squares) after IP administration (10 mg- kg \(^{-1}\) ). Top, chemical structure of ARN726. Right panel, area under the curve (AUC). + +<|ref|>text<|/ref|><|det|>[[111, 303, 884, 346]]<|/det|> +(G- I) Effect of ARN726 (10 mg- kg \(^{- 1}\) ) on plasma concentrations of (G) PEA, (H) OEA, and (I) anandamide (AEA). Gray bars, baseline plasma concentrations. + +<|ref|>text<|/ref|><|det|>[[111, 350, 884, 394]]<|/det|> +(J- L) Effect of ARN726 (10 mg- kg \(^{- 1}\) ) on the spinal cord concentrations of PEA (J), OEA (K), and AEA (L). Open bars, baseline analyte concentrations. + +<|ref|>text<|/ref|><|det|>[[111, 398, 884, 441]]<|/det|> +(M) IL- 6- primed mice were treated on 1d- 3d with ARN726 (1- 30 mg- kg \(^{- 1}\) ) and the response to \(\mathsf{PG E}_{2}\) was assessed on 6d. % MPE, percent of maximal possible effect. + +<|ref|>text<|/ref|><|det|>[[111, 445, 884, 489]]<|/det|> +(N) Effects of intrathecal vehicle (magenta circles) or ARN726 (30 ng, administered on 1d and 3d) (green circles) on the response to \(\mathsf{PG E}_{2}\) in IL- 6 primed mice. + +<|ref|>text<|/ref|><|det|>[[111, 493, 884, 610]]<|/det|> +Data are presented as mean ± S.E.M. and were analyzed using the Student's \(t\) test (F, AUC), one- way ANOVA (G- L), or two- way repeated measure ANOVA (A- F, N). Dunnett's or Bonferroni's post hoc test was applied as needed. ###P < 0.001, ##P < 0.01, #P < 0.05 versus vehicle. ***P < 0.001, **P < 0.01 versus baseline (time 0). ns, non- significant (n = 5- 8 mice per group). Dotted lines indicate baseline withdrawal latency. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 90, 880, 458]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 467, 722, 487]]<|/det|> +
Figure 3: NAAA in CD11b+ myeloid cells initiates hyperalgesic priming.
+ +<|ref|>text<|/ref|><|det|>[[110, 490, 886, 896]]<|/det|> +(A) Generation of NaaaCD11b-/- mice. Cross-breeding of global Naaa-/- mice with flippase (fl) FLPo-10 mice produced Naaafl/fl offspring, which was mated with CD11b-Cre mice to yield the NaaaCD11b-/- line. Yellow boxes: Naaa coding sequences; green boxes, Flp recognition target (FRT) sites; gray triangles: loxP sites flanking the deleted Naaa sequence. +(B) FACS tracings showing isolation of monocytes from blood samples of (top panels) Naaafl/fl and (bottom panels) NaaaCD11b-/- mice. Left, isolation of CD11b+ cells; right, isolation of Ly6C+ monocytes. The square denotes the Ly6Chigh ('classical') monocyte subpopulation collected for RT-qPCR analysis. 1, Ly6Clow ('non-classical') monocytes; 2, neutrophils. +(C) Naaa mRNA levels (assessed by RT-qPCR) in FACS-isolated CD11b- cells and Ly6Chigh monocytes from NaaaCD11b-/- mice (green circles) and Naaafl/fl (magenta circles). Whole blood is shown for comparison. Bars represent the average of 3 biological replicates. +(D) Self-resolving heat hypersensitivity elicited by IL-6 in NaaaCD11b-/- mice (green circles) and control Naaafl/fl mice (magenta circles). +(E) Effects of PGE2 in IL-6-primed NaaaCD11b-/- mice (green circles) and control Naaafl/fl mice (magenta circles). +(F) Normal heat sensitivity in NaaaCD11b+ mice (orange triangles) and their wild-type littermates (Wt, blue circles) after saline injection (20 μL). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 836, 108]]<|/det|> +(G) Effects of \(\mathsf{PGE}_2\) in NaaaCD11b+ (orange triangles) and wild-type littermates (blue circles). + +<|ref|>text<|/ref|><|det|>[[113, 111, 884, 202]]<|/det|> +Data are expressed as mean ± S.E.M. and were analyzed using the Student's t test (AUC) or two- way repeated measure ANOVA. Bonferroni's post hoc test was applied as appropriate. ##P < 0.001, ##P < 0.01, #P < 0.05 versus Naaafl/n or Wt. ns, non- significant (n = 7- 8 mice per group). Dotted lines indicate baseline withdrawal latency. + +<|ref|>image<|/ref|><|det|>[[122, 235, 870, 750]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 755, 740, 775]]<|/det|> +
Figure 4: PPARα in CD11b+ myeloid cells counters hyperalgesic priming.
+ +<|ref|>text<|/ref|><|det|>[[112, 780, 884, 872]]<|/det|> +(A) Generation of \(Ppara^{CD11b - / - }\) mice. Global \(Ppara^{/-}\) mice were mated with flippase (fl) FLPo-10 mice and the offspring was cross-bred with CD11b-Cre mice to yield the \(Ppara^{CD11b - / - }\) line. Gray boxes: \(Ppara\) coding sequences; green boxes, Flp recognition target (FRT) sites; gray triangles: loxP sites flanking the deleted \(Ppara\) sequence. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 150]]<|/det|> +(B) FACS tracings showing the isolation of monocytes from blood samples of (top) \(Ppara^{f / f}\) and (bottom) \(Ppara^{CD11b - / - }\) mice. Left, isolation of CD11b\(^+\) cells; right, isolation of Ly6C\(^+\) monocytes. + +<|ref|>text<|/ref|><|det|>[[111, 135, 886, 180]]<|/det|> +The square denotes the Ly6C\(^high\) (classical) monocyte subpopulation collected for RT-qPCR analysis. 1, Ly6C\(^low\) (non-classical) monocytes; 2, neutrophils. + +<|ref|>text<|/ref|><|det|>[[111, 183, 886, 252]]<|/det|> +(C) \(Ppara\) mRNA levels (assessed by RT-qPCR) in FACS-isolated CD11b\(^-\) cells and Ly6C\(^high\) monocytes from \(Ppara^{f / f}\) (open circles) and \(Ppara^{CD11b - / - }\) (magenta triangles) mice. Whole blood is shown for comparison. Bars represent the average of 2 biological replicates. + +<|ref|>text<|/ref|><|det|>[[111, 255, 825, 277]]<|/det|> +(D) Normal heat sensitivity in \(Ppara^{f / f}\) and \(Ppara^{CD11b - / - }\) mice after saline injection (20 μL). + +<|ref|>text<|/ref|><|det|>[[111, 280, 884, 324]]<|/det|> +(E) Effects of PGE\(_2\) in non-primed (saline-injected) \(Ppara^{f / f}\) and \(Ppara^{CD11b - / - }\) mice. Right panel, area under the curve (AUC). + +<|ref|>text<|/ref|><|det|>[[111, 327, 884, 370]]<|/det|> +(F) IL-6-primed \(Ppara^{f / f}\) mice were treated on 1d-3d with vehicle or ARN19702 (30 mg·kg\(^{-1}\)). (F1) Effect of IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). + +<|ref|>text<|/ref|><|det|>[[111, 373, 884, 418]]<|/det|> +(F2) Effects of PGE\(_2\) in IL-6-primed \(Ppara^{f / f}\) mice after administration of vehicle (red circles) or ARN19702 (blue circles). Right panel, AUC. + +<|ref|>text<|/ref|><|det|>[[111, 421, 884, 444]]<|/det|> +(G) IL-6-primed \(Ppara^{CD11b - / - }\) mice were treated on 1d-3d with vehicle or ARN19702 (30 mg·kg\(^{-1}\)). + +<|ref|>text<|/ref|><|det|>[[111, 447, 884, 491]]<|/det|> +(G1) Self-resolving heat hypersensitivity elicited by IL-6 before administration of vehicle (open symbols) or ARN19702 (closed symbols). (G2) Effect of PGE\(_2\) in IL-6-primed \(Ppara^{CD11b - / - }\) mice after administration of vehicle (red triangles) or ARN19702 (blue triangles). Right panel, AUC. + +<|ref|>text<|/ref|><|det|>[[111, 494, 884, 515]]<|/det|> +Data are represented as mean ± S.E.M. and were analyzed using the Student's \(t\) test (AUC) or two-way repeated measure ### \(P < 0.001\) , ## \(P < 0.01\) , ## \(P < 0.05\) , versus vehicle or \(Ppara^{f / f}\) (n = 7- 10 mice per group). Dotted lines indicate baseline withdrawal latency. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 92, 880, 488]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 494, 872, 514]]<|/det|> +
Figure 5: Monocyte-derived cells, not neutrophils, are required for hyperalgesic priming.
+ +<|ref|>text<|/ref|><|det|>[[110, 519, 886, 899]]<|/det|> +(A- D) Verification of monocyte removal by clodronate. (A) Liposomes containing clodronate or PBS were injected twice with a 3- day interval in wild- type male mice. Two hours after the second injection, blood and spinal cord were collected for FACS analysis. (B) FACS tracings of blood samples from mice treated with (top) PBS or (bottom) clodronate. Left, isolation of CD11b+ cells; right, isolation of Ly6Chigh ('classical') monocytes (1), Ly6Clow ('non- classical') monocytes (2), and neutrophils (3). The squares denote areas selected for quantification. (C, D) Number of circulating (C) monocytes and (D) neutrophils in mice treated with PBS (red bar) or clodronate (blue bar). Bars represent the average of 3 biological replicates. (E) Experimental timeline for the clodronate experiment. IL- 6 was administered in wild- type male mice immediately before the last clodronate/PBS injection. PGE2 was injected 6 days later. (F1) Self- resolving heat hypersensitivity elicited by IL- 6 in mice treated with PBS (red line) or clodronate (Clo, blue line). (F2) Effects of PGE2 in IL- 6 primed mice treated with PBS (red line) or clodronate (Clo, blue line). Right panel, area under the curve (AUC). (G- J) Verification of monocyte removal by PLX5622. (G) Mice were fed chow containing PLX5622 or vehicle for 20 days. On the last day, blood and spinal cord were collected for FACS analysis. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 884, 180]]<|/det|> +(H) FACS tracings of blood samples from mice fed a chow containing vehicle (top) or PLX5622 (bottom). Left, isolation of CD11b+ cells; right, isolation of Ly6Chigh ('classical') monocytes (1), Ly6Clow ('non-classical') monocytes (2), and neutrophils (3). The squares denote areas selected for quantification. + +<|ref|>text<|/ref|><|det|>[[111, 184, 884, 228]]<|/det|> +(I, J) Number of circulating (I) monocytes and (J) neutrophils in mice exposed to control chow (red bar) and chow containing PLX5622 (blue bar). + +<|ref|>text<|/ref|><|det|>[[111, 232, 884, 300]]<|/det|> +(K) Experimental timeline for the PLX5622 experiment. IL- 6 was administered in wild-type male mice after 14 days of exposure to control or PLX5622-containing chow. The treatment was continued and \(\mathsf{PPE}_2\) was injected 6 days later. + +<|ref|>text<|/ref|><|det|>[[111, 303, 884, 348]]<|/det|> +(L1) Self-resolving heat hypersensitivity elicited by IL- 6 in mice exposed to control chow (red line) or PLX5622-containing chow (blue line). + +<|ref|>text<|/ref|><|det|>[[111, 351, 884, 395]]<|/det|> +(L2) Effects of \(\mathsf{PPE}_2\) in IL- 6 primed mice exposed to control chow (red line) or PLX5622-containing chow (blue line). Right panel, AUC. + +<|ref|>text<|/ref|><|det|>[[111, 399, 884, 467]]<|/det|> +Data are represented as mean \(\pm\) S.E.M. and were analyzed using the Student's \(t\) test (AUC) or two- way repeated measure \(\mathrm{###P< 0.001}\) , \(\mathrm{##P< 0.01}\) , \(\mathrm{##P< 0.05}\) , versus PBS or control (n = 4- 10 mice per group). Dotted lines indicate baseline withdrawal latency. + +<|ref|>image<|/ref|><|det|>[[111, 496, 864, 866]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 873, 789, 893]]<|/det|> +
Figure 6: A role for NAAA-regulated PPAR-\(\alpha\) signaling in hyperalgesic priming.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 204]]<|/det|> +**Left: non-primed state.** High baseline concentrations of NAAA substrates (e.g., PEA) in tissue-resident macrophages constitutively activate PPAR- \(\alpha\) -dependent transcriptional programs, which maintain the cells in a resting state66,87. In this non-primed state, \(\text {PGE}_{2}\) produces transient ( \(\sim 2h\) ) heat hypersensitivity by engaging protein kinase A (PKA)- coupled \(\text {EP}_{2}/\text {EP}_{4}\) receptors in isolectin- \(\text {B4}^{+}\) nociceptive neurons. PA, palmitic acid, a product of PEA cleavage by NAAA. + +<|ref|>text<|/ref|><|det|>[[111, 209, 884, 398]]<|/det|> +**Center: initiation and incubation.** IL- 6 evokes a self-resolving nocifensive response that Isats \(\sim 24h\) and is attenuated by standard analgesics (22,23, present study). In parallel, the cytokine initiates priming through a mechanism that is insensitive to these agents. This initiation phase is followed by an incubation period, which persists for \(\sim 72h\), during which NAAA activity interrupts PPAR- \(\alpha\) signaling and drives tissue-resident macrophages and/or infiltrating monocytes into an active (or 'primed') state, which enables them to interact with non-peptidergic nociceptors - presumably via one or more unknown chemical signals (colored circles) - and to effect neuroplastic changes that initiate priming. + +<|ref|>text<|/ref|><|det|>[[111, 403, 884, 496]]<|/det|> +**Right: primed state.** In primed animals, \(\text {EP}_{2}/\text {EP}_{4}\) receptors in non-peptidergic neurons are coupled to an additional transduction pathway that involves protein kinase C- epsilon \((\text {PKC}_{\epsilon})^{17}\) . In this state, \(\text {PGE}_{2}\) evokes a \(\text {PKC}_{\epsilon}\) - mediated heat hypersensitivity that can last \(>2\) weeks38. Figure was created with BioRender.com. + +<|ref|>sub_title<|/ref|><|det|>[[113, 526, 214, 543]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[111, 567, 884, 900]]<|/det|> +1 Dahlhamer, J. et al. Prevalence of Chronic Pain and High- Impact Chronic Pain Among Adults - United States, 2016. MMWR Morb Mortal Wkly Rep 67, 1001- 1006, doi:10.15585/mmwr.mm6736a2 (2018). 2 Goldberg, D. S. & McGee, S. J. Pain as a global public health priority. BMC Public Health 11, 770, doi:10.1186/1471- 2458- 11- 770 (2011). 3 Cohen, S. P., Vase, L. & Hooten, W. M. Chronic pain: an update on burden, best practices, and new advances. Lancet 397, 2082- 2097, doi:10.1016/S0140- 6736(21)00393- 7 (2021). 4 Price, T. J. et al. Transition to chronic pain: opportunities for novel therapeutics. Nat Rev Neurosci 19, 383- 384, doi:10.1038/s41583- 018- 0012- 5 (2018). 5 Basbaum, A. I., Bautista, D. M., Scherrer, G. & Julius, D. Cellular and molecular mechanisms of pain. Cell 139, 267- 284, doi:10.1016/j.cell.2009.09.028 (2009). 6 Woolf, C. J. Central sensitization: implications for the diagnosis and treatment of pain. Pain 152, S2- S15, doi:10.1016/j.pain.2010.09.030 (2011). 7 Price, T. J. & Ray, P. R. Recent advances toward understanding the mysteries of the acute to chronic pain transition. Curr Opin Physiol 11, 42- 50, doi:10.1016/j.cophys.2019.05.015 (2019). 8 Ji, R. R., Chamessian, A. & Zhang, Y. Q. Pain regulation by non- neuronal cells and inflammation. Science 354, 572- 577, doi:10.1126/science.aaf8924 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 90, 888, 900]]<|/det|> +Calvo, M., Dawes, J. M. & Bennett, D. L. The role of the immune system in the generation of neuropathic pain. Lancet Neurol 11, 629- 642, doi:10.1016/S1474- 4422(12)70134- 5 (2012).Malcangio, M. Role of the immune system in neuropathic pain. Scand J Pain 20, 33- 37, doi:10.1515/sjpain- 2019- 0138 (2019).Malcangio, M. Introduction for special issue on neuroimmune interactions in chronic pain. Pain Rep 6, e894, doi:10.1097/PR9.0000000000000894 (2021).Zhang, J. et al. Expression of CCR2 in both resident and bone marrow- derived microglia plays a critical role in neuropathic pain. J Neurosci 27, 12396- 12406, doi:10.1523/JNEUROSCI.3016- 07.2007 (2007).Peng, J. et al. Microglia and monocytes synergistically promote the transition from acute to chronic pain after nerve injury. Nat Commun 7, 12029, doi:10.1038/ncomms12029 (2016).Yu, X. et al. Dorsal root ganglion macrophages contribute to both the initiation and persistence of neuropathic pain. Nat Commun 11, 264, doi:10.1038/s41467- 019- 13839- 2 (2020).Old, E. A. et al. Monocytes expressing CX3CR1 orchestrate the development of vincristine- induced pain. J Clin Invest 124, 2023- 2036, doi:10.1172/JCI71389 (2014).Oggero, S. et al. Dorsal root ganglia CX3CR1 expressing monocytes/macrophages contribute to arthritis pain. Brain Behav Immun 106, 289- 306, doi:10.1016/j.bbi.2022.09.008 (2022).Reichling, D. B. & Levine, J. D. Critical role of nociceptor plasticity in chronic pain. Trends Neurosci 32, 611- 618, doi:10.1016/j.tins.2009.07.007 (2009).Kandasamy, R. & Price, T. J. The pharmacology of nociceptor priming. Handb Exp Pharmacol 227, 15- 37, doi:10.1007/978- 3- 662- 46450- 2_2 (2015).Aley, K. O., Messing, R. O., Mochly- Rosen, D. & Levine, J. D. Chronic hypersensitivity for inflammatory nociceptor sensitization mediated by the epsilon isozyme of protein kinase C. J Neurosci 20, 4680- 4685, doi:10.1523/JNEUROSCI.20- 12- 04680.2000 (2000).Parada, C. A., Reichling, D. B. & Levine, J. D. Chronic hyperalgesic priming in the rat involves a novel interaction between cAMP and PKCepsilon second messenger pathways. Pain 113, 185- 190, doi:10.1016/j.pain.2004.10.021 (2005).Joseph, E. K. & Levine, J. D. Hyperalgesic priming is restricted to isolectin B4- positive nociceptors. Neuroscience 169, 431- 435, doi:10.1016/j.neuroscience.2010.04.082 (2010).Parada, C. A., Yeh, J. J., Joseph, E. K. & Levine, J. D. Tumor necrosis factor receptor type- 1 in sensory neurons contributes to induction of chronic enhancement of inflammatory hyperalgesia in rat. Eur J Neurosci 17, 1847- 1852, doi:10.1046/j.1460- 9568.2003.02626. x (2003).Ferrari, L. F., Bogen, O., Reichling, D. B. & Levine, J. D. Accounting for the delay in the transition from acute to chronic pain: axonal and nuclear mechanisms. J Neurosci 35, 495- 507, doi:10.1523/JNEUROSCI.5147- 13.2015 (2015).Bogen, O., Alessandri- Haber, N., Chu, C., Gear, R. W. & Levine, J. D. Generation of a pain memory in the primary afferent nociceptor triggered by PKCepsilon activation of CPEB. J Neurosci 32, 2018- 2026, doi:10.1523/JNEUROSCI.5138- 11.2012 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 90, 888, 900]]<|/det|> +781 25 Melemedjian, O. K. et al. Local translation and retrograde axonal transport of CREB regulates IL- 6- induced nociceptive plasticity. Mol Pain 10, 45, doi:10.1186/1744- 8069- 10- 45 (2014).783 26 Piomelli, D. et al. N- Acylethanolamine Acid Amidase (NAAA): Structure, Function, and Inhibition. J Med Chem 63, 7475- 7490, doi:10.1021/acs.jmedchem.0c00191 (2020).786 27 Lo Verme, J. et al. The nuclear receptor peroxisome proliferator- activated receptor- alpha mediates the anti- inflammatory actions of palmitoylethanolamide. Mol Pharmacol 67, 15- 19, doi:10.1124/mol.104.006353 (2005).788 28 LoVerme, J., La Rana, G., Russo, R., Calignano, A. & Piomelli, D. The search for the palmitoylethanolamide receptor. Life Sci 77, 1685- 1698, doi:10.1016/j.lfs.2005.05.012 (2005).792 29 LoVerme, J. et al. Rapid broad- spectrum analgesia through activation of peroxisome proliferator- activated receptor- alpha. J Pharmacol Exp Ther 319, 1051- 1061, doi:10.1124/jpet.106.111385 (2006).794 30 Solorzano, C. et al. Selective N- acylethanolamine- hydrolyzing acid amidase inhibition reveals a key role for endogenous palmitoylethanolamide in inflammation. Proc Natl Acad Sci U S A 106, 20966- 20971, doi:10.1073/pnas.0907417106 (2009).798 31 Fotio, Y. et al. NAAA- regulated lipid signaling governs the transition from acute to chronic pain. Sci Adv 7, eabi8834, doi:10.1126/sciadv.abi8834 (2021).800 32 Sasso, O. et al. The N- Acylethanolamine Acid Amidase Inhibitor ARN077 Suppresses Inflammation and Pruritus in a Mouse Model of Allergic Dermatitis. J Invest Dermatol 138, 562- 569, doi:10.1016/j.jid.2017.07.853 (2018).803 33 Pontis, S. et al. N- Acylethanolamine Acid Amidase contributes to disease progression in a mouse model of multiple sclerosis. Pharmacol Res 160, 105064, doi:10.1016/j.phrs.2020.105064 (2020).806 34 Spangenberg, E. et al. Sustained microglial depletion with CSF1R inhibitor impairs parenchymal plaque development in an Alzheimer's disease model. Nat Commun 10, 3758, doi:10.1038/s41467- 019- 11674- z (2019).809 35 Van Rooijen, N. The liposome- mediated macrophage 'suicide' technique. J Immunol Methods 124, 1- 6, doi:10.1016/0022- 1759(89)90178- 6 (1989).811 36 Lee, P. Y., Wang, J. X., Parisini, E., Dascher, C. C. & Nigrovic, P. A. Ly6 family proteins in neutrophil biology. J Leukoc Biol 94, 585- 594, doi:10.1189/jlb.0113014 (2013).813 37 Jurga, A. M., Paleczna, M. & Kuter, K. Z. Overview of General and Discriminating Markers of Differential Microglia Phenotypes. Front Cell Neurosci 14, 198, doi:10.3389/fncel.2020.00198 (2020).816 38 Paige, C. et al. Neuroendocrine Mechanisms Governing Sex Differences in Hyperalgesic Priming Involve Prolactin Receptor Sensory Neuron Signaling. J Neurosci 40, 7080- 7090, doi:10.1523/JNEUROSCI.1499- 20.2020 (2020).819 39 Mabou Tagne, A., Fotio, Y., Uppal, P. S. & Piomelli, D. Synergistic antinociceptive effects of concomitant NAAA and peripheral FAAH inhibition. Exp Neurol 357, 114194, doi:10.1016/j.expneurol.2022.114194 (2022).822 40 Mabou Tagne, A. et al. Palmitoylethanolamide and hemp oil extract exert synergistic antinociceptive effects in mouse models of acute and chronic pain. Pharmacol Res 167, 105545, doi:10.1016/j.phrs.2021.105545 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 90, 888, 899]]<|/det|> +825 41 Fotio, Y., Palese, F., Guaman Tipan, P., Ahmed, F. & Piomelli, D. Inhibition of fatty acid amide hydrolase in the CNS prevents and reverses morphine tolerance in male and female mice. Br J Pharmacol 177, 3024- 3035, doi:10.1111/bph.15031 (2020). 828 42 Fotio, Y., Mabou Tagne, A., Jung, K. M. & Piomelli, D. Fatty acid amide hydrolase inhibition alleviates anxiety- like symptoms in a rat model used to study post- traumatic stress disorder. Psychopharmacology (Berl), doi:10.1007/s00213- 023- 06358- y (2023). 831 43 Lee, H. L. et al. Frequent Low- Dose Delta(9)- Tetrahydrocannabinol in Adolescence Disrupts Microglia Homeostasis and Disables Responses to Microbial Infection and Social Stress in Young Adulthood. Biol Psychiatry 92, 845- 860, doi:10.1016/j.biopsych.2022.04.017 (2022). 834 44 Gabrielsson, J. & Weiner, D. Non- compartmental analysis. Methods Mol Biol 929, 377- 389, doi:10.1007/978- 1- 62703- 050- 2_16 (2012). 837 45 Migliore, M. D. et al. Second- Generation Non- Covalent NAAA Inhibitors are Protective in a Model of Multiple Sclerosis. Angew Chem Int Ed Engl 55, 11193- 11197, doi:10.1002/anie.201603746 (2016). 840 46 Fotio, Y., Sasso, O., Ciccocioppo, R. & Piomelli, D. Antinociceptive Profile of ARN19702, (2- Ethylsulfonylphenyl)- [(2S)-4-(6- fluoro- 1,3- benzothiazol- 2- yl)- 2- methylpiperazin- 1- yl]methanone, a Novel Orally Active N- Acylethanolamine Acid Amidase Inhibitor, in Animal Models. J Pharmacol Exp Ther 378, 70- 76, doi:10.1124/jpet.121.000674 (2021). 844 47 Xu, H. E. et al. Structural basis for antagonist- mediated recruitment of nuclear co- repressors by PPARalpha. Nature 415, 813- 817, doi:10.1038/415813a (2002). 846 48 Lee, G. et al. T0070907, a selective ligand for peroxisome proliferator- activated receptor gamma, functions as an antagonist of biochemical and cellular activities. J Biol Chem 277, 19649- 19657, doi:10.1074/jbc.M200743200 (2002). 849 49 Hosohata, Y. et al. AM630 antagonism of cannabinoid- stimulated [35S]GTP gamma S binding in the mouse brain. Eur J Pharmacol 321, R1- 3, doi:10.1016/s0014- 2999(97)00047- 2 (1997). 852 50 Paterniti, I. et al. Molecular evidence for the involvement of PPAR- delta and PPAR- gamma in anti- inflammatory and neuroprotective activities of palmitoylethanolamide after spinal cord trauma. J Neuroinflammation 10, 20, doi:10.1186/1742- 2094- 10- 20 (2013). 853 51 Guida, F. et al. Palmitoylethanolamide induces microglia changes associated with increased migration and phagocytic activity: involvement of the CB2 receptor. Sci Rep 7, 375, doi:10.1038/s41598- 017- 00342- 1 (2017). 854 52 Palese, F. et al. Targeting NAAA counters dopamine neuron loss and symptom progression in mouse models of parkinsonism. Pharmacol Res 182, 106338, doi:10.1016/j.phrs.2022.106338 (2022). 856 53 Ribeiro, A. et al. A Potent Systemically Active N- Acylethanolamine Acid Amidase Inhibitor that Suppresses Inflammation and Human Macrophage Activation. ACS Chem Biol 10, 1838- 1846, doi:10.1021/acschembio.5b00114 (2015). 857 54 Tsuboi, K. et al. Predominant expression of lysosomal N- acylethanolamine- hydrolyzing acid amidase in macrophages revealed by immunochemical studies. Biochim Biophys Acta 1771, 623- 632, doi:10.1016/j.bnalip.2007.03.005 (2007). 856 55 Gordon, S. & Taylor, P. R. Monocyte and macrophage heterogeneity. Nat Rev Immunol 5, 953- 964, doi:10.1038/nri1733 (2005). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 88, 888, 899]]<|/det|> +869 56 Mass, E., Nimmerjahn, F., Kierdorf, K. & Schlitzer, A. Tissue- specific macrophages: how 870 they develop and choreograph tissue biology. Nat Rev Immunol, 1- 17, 871 doi:10.1038/s41577- 023- 00848- y (2023). 872 57 van Rooijen, N. & Hendrikx, E. Liposomes for specific depletion of macrophages from 873 organs and tissues. Methods Mol Biol 605, 189- 203, doi:10.1007/978- 1- 60327- 360- 2_13 874 (2010). 875 58 Schmidt- Weber, C. B. et al. Apoptotic cell death in activated monocytes following 876 incorporation of clodronate- liposomes. J Leukoc Biol 60, 230- 244, 877 doi:10.1002/jlb.60.2.230 (1996). 878 59 Van Rooijen, N. & Sanders, A. Liposome mediated depletion of macrophages: mechanism 879 of action, preparation of liposomes and applications. J Immunol Methods 174, 83- 93, 880 doi:10.1016/0022- 1759(94)90012- 4 (1994). 881 60 Lee, H. H. et al. Depletion of Alveolar Macrophages Increases Pulmonary Neutrophil 882 Infiltration, Tissue Damage, and Sepsis in a Murine Model of Acinetobacter baumannii 883 Pneumonia. Infect Immun 88, doi:10.1128/AI.00128- 20 (2020). 884 61 Culemann, S. et al. Stunning of neutrophils accounts for the anti- inflammatory effects of 885 clodronate liposomes. J Exp Med 220, doi:10.1084/jem.20220525 (2023). 886 62 Willemen, H. L. et al. Monocytes/Macrophages control resolution of transient inflammatory 887 pain. J Pain 15, 496- 506, doi:10.1016/j.pain.2014.01.491 (2014). 888 63 Spiteri, A. G. et al. PLX5622 Reduces Disease Severity in Lethal CNS Infection by Off- 889 Target Inhibition of Peripheral Inflammatory Monocyte Production. Front Immunol 13, 890 851556, doi:10.3389/fimmu.2022.851556 (2022). 891 64 Lei, F. et al. CSF1R inhibition by a small- molecule inhibitor is not microglia specific; 892 affecting hematopoiesis and the function of macrophages. Proc Natl Acad Sci U S A 117, 893 23336- 23338, doi:10.1073/pnas.1922788117 (2020). 894 65 Rigamonti, E., Chinetti- Gbaguidi, G. & Staels, B. Regulation of macrophage functions by 895 PPAR- alpha, PPAR- gamma, and LXRs in mice and men. Arterioscler Thromb Vasc Biol 896 28, 1050- 1059, doi:10.1161/ATVBAHA.107.158998 (2008). 897 66 Grabacka, M., Pierzchalska, M., Plonka, P. M. & Pierzchalski, P. The Role of PPAR Alpha 898 in the Modulation of Innate Immunity. Int J Mol Sci 22, doi:10.3390/ijms221910545 (2021). 899 Penas, F. et al. Treatment in vitro with PPARalpha and PPARgamma ligands drives M1- 900 to- M2 polarization of macrophages from T. cruzi- infected mice. Biochim Biophys Acta 901 1852, 893- 904, doi:10.1016/j.bpids.2014.12.019 (2015). 902 68 Feng, X. et al. PPAR- alpha Agonist Fenofibrate Prevented Diabetic Nephropathy by 903 Inhibiting M1 Macrophages via Improving Endothelial Cell Function in db/db Mice. Front 904 Med (Lausanne) 8, 652558, doi:10.3389/fmed.2021.652558 (2021). 905 69 Fu, J. et al. Oleylethanolamide regulates feeding and body weight through activation of 906 the nuclear receptor PPAR- alpha. Nature 425, 90- 93, doi:10.1038/nature01921 (2003). 907 70 Okamoto, Y., Morishita, J., Tsuboi, K., Tonai, T. & Ueda, N. Molecular characterization of 908 a phospholipase D generating anandamide and its congeners. J Biol Chem 279, 5298- 909 5305, doi:10.1074/jbc.M306642200 (2004). 910 71 Magotti, P. et al. Structure of human N- acylphosphatidylethanolamine- hydrolyzing 911 phospholipase D: regulation of fatty acid ethanolamide biosynthesis by bile acids. 912 Structure 23, 598- 604, doi:10.1016/j.str.2014.12.018 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 90, 888, 900]]<|/det|> +Zhu, C. et al. Proinflammatory stimuli control N-acylphosphatidylethanolamine-specific phospholipase D expression in macrophages. Mol Pharmacol 79, 786- 792, doi:10.1124/mol.110.070201 (2011). Gorelik, A., Gebai, A., Illes, K., Piomelli, D. & Nagar, B. Molecular mechanism of activation of the immunoregulatory amidase NAAA. Proc Natl Acad Sci U S A 115, E10032- E10040, doi:10.1073/pnas.1811759115 (2018). Alhouayek, M. et al. N-Acylethanolamine-hydrolyzing acid amidase inhibition increases colon N-palmitoylethanolamine levels and counteracts murine colitis. FASEB J 29, 650- 661, doi:10.1096/fj.14- 255208 (2015). Bonezzi, F. T. et al. An Important Role for N-Acylethanolamine Acid Amidase in the Complete Freund's Adjuvant Rat Model of Arthritis. J Pharmacol Exp Ther 356, 656- 663, doi:10.1124/jpet.115.230516 (2016). Cairns, B. E., Arendt- Nielsen, L. & Sacerdote, P. Perspectives in Pain Research 2014: Neuroinflammation and glial cell activation: The cause of transition from acute to chronic pain? Scand J Pain 6, 3- 6, doi:10.1016/j.spain.2014.10.002 (2015). Caxaria, S. et al. Neutrophils infiltrate sensory ganglia and mediate chronic widespread pain in fibromyalgia. Proc Natl Acad Sci U S A 120, e2211631120, doi:10.1073/pnas.2211631120 (2023). Chen, Z. et al. Sphingosine- 1- phosphate receptor 1 activation in astrocytes contributes to neuropathic pain. Proc Natl Acad Sci U S A 116, 10557- 10562, doi:10.1073/pnas.1820466116 (2019). Squillace, S., Spiegel, S. & Salvemini, D. Targeting the Sphingosine- 1- Phosphate Axis for Developing Non- narcotic Pain Therapeutics. Trends Pharmacol Sci 41, 851- 867, doi:10.1016/j.tips.2020.09.006 (2020). Keeble, J. E. et al. Hydrogen peroxide is a novel mediator of inflammatory hyperalgesia, acting via transient receptor potential vanilloid 1- dependent and independent mechanisms. Pain 141, 135- 142, doi:10.1016/j.pain.2008.10.025 (2009). Reis, C., Chambel, S., Ferreira, A. & Cruz, C. D. Involvement of nerve growth factor (NGF) in chronic neuropathic pain - a systematic review. Rev Neurosci 34, 75- 84, doi:10.1515/revneuro- 2022- 0037 (2023). Cappoli, N., Tabolacci, E., Aceto, P. & Dello Russo, C. The emerging role of the BDNF- TrkB signaling pathway in the modulation of pain perception. J Neuroimmunol 349, 577406, doi:10.1016/j.jneuroim.2020.577406 (2020). Finnerup, N. B., Nikolajsen, L. & Rice, A. S. C. Transition from acute to chronic pain: a misleading concept? Pain 163, e985- e988, doi:10.1097/j.pain.0000000000002631 (2022). Shiers, S. et al. Neuropathic Pain Creates an Enduring Prefrontal Cortex Dysfunction Corrected by the Type II Diabetic Drug Metformin But Not by Gabapentin. J Neurosci 38, 7337- 7350, doi:10.1523/JNEUROSCI.0713- 18.2018 (2018). Inyang, K. E. et al. Alleviation of paclitaxel- induced mechanical hypersensitivity and hyperalgesic priming with AMPK activators in male and female mice. Neurobiol Pain 6, 100037, doi:10.1016/j.ynpai.2019.100037 (2019). Inyang, K. E. et al. Indirect AMP- Activated Protein Kinase Activators Prevent Incision- Induced Hyperalgesia and Block Hyperalgesic Priming, Whereas Positive Allosteric + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 88, 886, 196]]<|/det|> +957 Modulators Block Only Priming in Mice. J Pharmacol Exp Ther 371, 138- 150, 958 doi:10.1124/jpet.119.258400 (2019). 959 87 Pontis, S., Ribeiro, A., Sasso, O. & Piomelli, D. Macrophage- derived lipid agonists of 960 PPAR- alpha as intrinsic controllers of inflammation. Crit Rev Biochem Mol Biol 51, 7- 14, 961 doi:10.3109/10409238.2015.1092944 (2016). 962 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[58, 128, 556, 470]]<|/det|> +- FotioNatComm13082023AllSupplementalmaterial.pdf- Fotioetal.2023Figure1.xlsx- Fotioetal.2023Figure2.xlsx- Fotioetal.2023FigureS2.xlsx- Fotioetal.2023Figure4.xlsx- Fotioetal.2023Figure3.xlsx- Fotioetal.2023Figure5.xlsx- Fotioetal.2023FigureS6.xlsx- Fotioetal.2023FigureS7.xlsx- Fotioetal.2023FigureS8.xlsx- Fotioetal.2023FigureS3.xlsx- Fotioetal.2023FigureS5.xlsx- Fotioetal.2023FigureS4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/images_list.json b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..9372978b17a6957656d618de777a3d248c46e30e --- /dev/null +++ b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 42, + 88, + 950, + 744 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 42, + 45, + 955, + 401 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 42, + 40, + 390, + 785 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 42, + 46, + 960, + 272 + ] + ], + "page_idx": 23 + } +] \ No newline at end of file diff --git a/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7.mmd b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7.mmd new file mode 100644 index 0000000000000000000000000000000000000000..838d6a37cf49b0dd977010687734f487e6bc02c5 --- /dev/null +++ b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7.mmd @@ -0,0 +1,347 @@ + +# Self-Consistent Grain Size Evolution controls Strain Localization during Rifting + +Jonas Ruh ( Jonas.ruh@erdw.ethz.ch ) ETH Zurich Leif Tokle ETH Zurich Whitney Behr ETH Zürich + +## Article + +Keywords: tectonic, mantle dynamics, geodynamic numerical models, rifting + +Posted Date: June 22nd, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 561920/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Geoscience on June 16th, 2022. See the published version at https://doi.org/10.1038/s41561- 022- 00964- 9. + +<--- Page Split ---> + +# Self-Consistent Grain Size Evolution controls Strain Localization during Rifting + +2 + +3 J. B. Ruh \(^{1*}\) , L. Tokle \(^{1}\) , and W. M. Behr \(^{1}\) + +\(^{1}\) Structural Geology and Tectonics group, Geological Institute, Department of Earth Sciences, ETH Zürich, Switzerland + +6 + +\(^{*}\) Corresponding author: Jonas B. Ruh (jonas.ruh@erdw.ethz.ch) + +8 + +# 9 Key Points: + +- Self-consistent grain size evolution according to the paleowattmeter is coupled to a 2D thermo-mechanical numerical model of the upper mantle- Mantle rheology is represented by a coupled grain-size-dependent diffusion and grain-size-independent dislocation creep flow law- Strain localization along lithospheric shear zones leads to reduced grain sizes and diffusion creep becomes the dominant deformation mechanism + +<--- Page Split ---> + +## Abstract + +Geodynamic numerical models often employ solely grain- size- independent dislocation creep to describe upper mantle dynamics. However, observations from nature and rock deformation experiments suggest that shear zones can transition to a grain- size- dependent creep mechanism due to dynamic grain size evolution, with important implications for the overall strength of plate boundaries. We apply a two- dimensional thermo- mechanical numerical model with a composite diffusion- dislocation creep rheology coupled to a dynamic grain size evolution model based on the paleowattmeter. Results indicate average olivine grain sizes of 3–12 cm for the upper mantle below the LAB, while in the lithosphere grain size ranges from 0.3–3 mm at the Moho to 6–15 cm at the LAB. Such a grain size distribution results in dislocation creep being the dominant deformation mechanism in the upper mantle. However, deformation- related grain size reduction below 100 μm activates diffusion creep along lithospheric- scale shear zones during rifting, affecting the overall strength of tectonic plate boundaries. + +## Main + +The Earth's lithosphere is defined by its mechanically rigid behavior in contrast to the relatively weak underlying asthenosphere. This rheological stratification, which ultimately allowed for the emergence of plate tectonics, primarily results from the apparent thermal gradient across the crust and upper mantle and the temperature- dependent activation of dislocation- and diffusion- related crystal- plastic creep of rocks and minerals1,2. Scaling of such experimentally- derived creep laws to natural strain rates allows us to estimate viscosities and strength of the lithosphere. Geophysical constraints on the elastic thickness of the continental lithosphere, that is a proxy for its strength3, led to contrasting conclusions on the uppermost mantle being either + +<--- Page Split ---> + +strong and best represented by dry dislocation creep of olivine4 or weak according to a wet olivine rheology5,6. + +Whether deformation within the upper mantle is dominated by dislocation creep or diffusion creep is still a matter of debate. The observation of crystallographic preferred orientation (CPO) in mantle xenoliths7 and evidence of strong seismic anisotropy8 has long been interpreted as an indicator for dislocation creep as the dominant deformation mechanism1. However, there is reported evidence that CPO, and therefore seismic anisotropy, may also develop as a result of diffusion creep of olivine-rich aggregates9,10. In contrast to dislocation creep, the relationship between stress and strain rate for diffusion creep is dependent on grain size, which is a crucial parameter when considering the dominant deformation mechanism in the upper mantle11- 14. A transition from dislocation to diffusion creep at depths greater than \(\sim 250 \mathrm{km}\) was proposed by Hirth and Kohlstedt12 based on theoretical estimations that olivine grain size in the upper mantle is on the order of \(10 \mathrm{mm}^{15}\). Numerical experiments of mantle convection have since implemented a composite diffusion- dislocation creep rheology and constant mantle grain size, which may result in dramatic convective instability and thermal erosion of the lithosphere16. However, the assumption of a constant upper mantle grain size is an oversimplification that appears contradictory to several observational and experimental datasets. Experimental data on wave speed and attenuation of olivine, for example, fits best with a seismological model that implies an increase in grain size from \(\sim 1 \mathrm{mm}\) to \(\sim 5 \mathrm{cm}\) between depths of \(100 - 200 \mathrm{km}^{17}\). Furthermore, natural samples of exhumed lithospheric mantle rocks show a large variety of grain sizes ranging from tens to hundreds of microns in olivine mylonites and tectonites to the centimeter- scale in weakly deformed or annealed xenoliths18- 21. + +<--- Page Split ---> + +Active plate tectonics requires mechanical weakening and strain localization along lithospheric shear zones at the plate boundaries22,23. Several studies suggest that grain size reduction and the consequent activation of diffusion creep is a viable process to initiate localization of deformation in the lithosphere24- 29, perhaps complementary to other potential weakening mechanisms such as shear heating30,31, reaction- induced weakening32, or the presence of preexisting weak zones or viscous anisotropy33,34. Here, we present a 2D thermo- mechanical numerical model with a composite diffusion- dislocation creep flow law coupled to a self- consistent grain size evolution model based on the paleowattmeter35. Such a model allows us to estimate apparent grain size distribution and the dominant deformation mechanism within the upper mantle, and to investigate the importance of grain size evolution for strain localization in the lithosphere during continental rifting. We test the influence of water content in the mantle which affects both its viscosity and rate of grain growth. Furthermore, the effect of localized grain- size- dependent weakening on the long- term strength and elastic thickness of continental lithosphere is investigated and compared to pure dislocation creep experiments. + +## Coupled grain size evolution thermo-mechanical model of upper mantle dynamics + +We apply a finite difference thermo- mechanical numerical model36,37 of the upper mantle and crust with an Eulerian domain of \(1000 \times 670 \mathrm{km}\) that undergoes horizontal divergence at a constant total rate of \(1 \mathrm{cm / yr}\) . The model employs a visco- elasto- plastic rheology where the viscous strain rate is composed of both dislocation and diffusion creep for constant water content12 and stresses are capped depending on the Drucker- Prager yield criterion (see Methods and Supplementary Methods). The applied fluid contents in the mantle are \(C_{\mathrm{OH}} = 50 - 2500 \mathrm{H / 10^6Si}\) , which cover the range of estimated values obtained from experimental studies38,39. + +<--- Page Split ---> + +Olivine grain size is calculated based on the paleowattmeter \(^{35}\) , which introduces a grain size evolution rate composed of independent growth and reduction terms (see Methods). Grain size reduction occurs by the process of dynamic recrystallization during dislocation creep, whereas grain size during diffusion creep is controlled by the process of grain growth \(^{40}\) . Based on grain sizes from experimentally deformed olivine aggregates, the fraction of work that goes into grain size reduction during dynamic recrystallization is estimated to be \(\lambda = 0.01\) (see Supplementary Methods). The grain growth parameters we implement are derived from experiments on natural olivine aggregates with in-situ water contents \(^{41}\) that predict significantly slower grain growth than previous constraints from experiments on water- saturated, synthetic olivine \(^{42}\) . Due to high temperatures and thus fast growth rates, the initial grain sizes in the lower part of the model domain rapidly adjust to a steady- state grain size. On the other hand, initial grain sizes within the lithosphere are mainly driven by the reduction term due to lower temperatures and higher deviatoric stresses. + +## Rheological implications and formation of lithospheric shear zones + +Composite diffusion- dislocation creep numerical experiments were conducted with variable water content in the mantle ( \(C_{\mathrm{OH}} = 50\) , 175, 600, and 2500 H/10 \(^{6}\) Si) that affect both viscous creep and grain growth. Mantle viscosities of the reference model ( \(C_{\mathrm{OH}} = 600\) H/10 \(^{6}\) Si) show values of \(10^{19} - 10^{21}\) Pa·s for the asthenosphere after 5 Myr of divergence (Fig. 1a). At 10 Myr, lithospheric thinning and related temperature increase below the rifted region lead to viscosities as low as \(5 \cdot 10^{17}\) Pa·s, relatively fast velocities, and gravitationally- induced lithospheric dripping. After 15 and 20 Myr of divergence, asthenospheric viscosities remain within \(10^{18} - 10^{21}\) Pa·s, with lower values where fast velocities occur due to thermally- and + +<--- Page Split ---> + +gravitationally- induced lithospheric erosion (Fig. 1a). Away from the rift, the lithosphere remains intact and strong. + +Illustrations of the dominant deformation mechanism (dislocation vs. diffusion creep) and contours of grain size in the mantle demonstrate that localization of stress in the centre of the model domain leads to grain size reduction and the activation of diffusion creep along large- scale lithospheric shear zones (Fig. 1b). The large shear zones retain relatively small grain sizes and remain dominated by diffusion creep even after 15 to 20 Myr of divergence, when a mid- ocean spreading centre is established, consuming most of the extensional velocity. + +For all experiments, grain sizes vary spatially throughout the model domain; furthermore, their values are strongly sensitive implemented mantle water content. Vertical grain size profiles along the side of the domain (at 5 Myr), away from the extensional zone, show values of 0.3–3 mm at the Moho (depending on \(C_{\mathrm{OH}}\) ) that increase to 6–15 cm at the LAB, and decrease to 2–7 cm at the base of the upper mantle (Fig. 2a). Grain sizes within localizing shear zones in the uppermost lithosphere at 40 km depth \((y = 50 \mathrm{km})\) show a rapid initial decrease to 60–250 \(\mu \mathrm{m}\) (Fig. 2b). Depending on the water content in the mantle, they are able to recover after \(\sim 15 \mathrm{Myr}\) \((C_{\mathrm{OH}} = 2500 \mathrm{H / 10^{6}Si})\) or \(\sim 20 \mathrm{Myr}\) \((C_{\mathrm{OH}} = 600 \mathrm{H / 10^{6}Si})\) . Lower water content hampers substantial grain growth within previously active shear zones before 40 Myr. Average upper mantle grain sizes below 300 km depth establish within \(\sim 2 \mathrm{Myr}\) and range in between 3–12 cm (Fig. 2c). Further undulations in average mantle grain size result from the downwelling of small- grain- size lithospheric dripplets. + +Figure 3 shows the portions of accumulated finite viscous strain within the mantle accommodated by diffusion and dislocation creep after 20 Myr of divergence. Dislocation creep is the dominant deformation mechanism in large parts of the upper mantle, independent of water + +<--- Page Split ---> + +content. Diffusion creep dominates within lithospheric shear zones that form in the early stages of rifting (Fig. 1b) and assist in lithospheric dripping (Fig. 3b- d). The continental lithospheric thickness defined by its viscosity varies between 90–150 km, depending on water content (Fig. 3). + +## Effects of grain size on lithospheric strength + +The importance of a self- consistent grain size distribution for upper mantle dynamics becomes evident when comparing our results to numerical experiments with pure dislocation creep of olivine or composite diffusion- dislocation creep with a constant grain size throughout the entire upper mantle. Experiments with dry dislocation creep result in extensive brittle- plastic deformation of the lithosphere43,44. Experiments with composite diffusion- dislocation creep and small constant grain size (1 mm) results in a lithosphere thinned by convective erosion (<90 km) driven by low asthenosphere viscosities of <1018 Pa·s16,45. For constant grain sizes larger than 1 cm, dislocation creep becomes the main deformation mechanism throughout the entire upper mantle16. These numerical experiments fail to match the effective elastic lithospheric thicknesses necessary to sustain orogens4, while brittle deformation in the lithosphere remains absent6. On the other hand, our implementation of a self- consistent grain size evolution is able to resolve this obstacle. Observed lithospheric thicknesses vary between 90–150 km (Fig. 3), while localization of deformation in the lithosphere rapidly leads to grain size reduction, diffusion creep activation, and related stress drop below the frictional yield, omitting failure. The diffusion- creep- related stress drop furthermore reduces and replaces the importance of shear heating along lithospheric shear zones30,46. + +<--- Page Split ---> + +The temporal evolution of the vertically integrated strength illustrates that experiments with composite diffusion- dislocation creep coupled to a self- consistent grain size evolution show a decrease of boundary forces below 5 TN/m within 1–2 Myr, while pure dislocation creep experiments remain above 10 TN/m for at least \(\sim 15\) Myr (Fig. 4a). Typical forces along plate boundaries are on the order of 1–5 TN/m47,48, which is sufficient to initiate continental rifting if the grain size is small enough and diffusion creep dominates deformation26. Vertical strength profiles indicate that most of the strength of coupled experiments remains within the crust with maximal values of \(\sim 200\) MPa, while pure dislocation creep experiments exhibit at least 10 km of brittle- plastic mantle lithosphere with differential stresses up to \(\sim 600–700\) MPa (Fig. 4b). Differential stresses of \(\sim 200\) MPa close to the Moho in composite diffusion- dislocation creep experiments stand in contrast to significantly lower strength along a lithospheric shear zone after 5 Myr (Fig. 4c). There, values of 1–10 MPa are defined by grain sizes as small as 100 μm and diffusion creep as the consequent deformation mechanism, efficiently weakening the entire lithospheric rift system. + +Water content in the upper mantle has important implications for the relationships between viscous flow and seismic anisotropy39, hydrous melting49, and the distribution of geochemical reservoirs50. The strength of olivine in the presence of water is significantly reduced38,51, as expressed in the flow law we apply here12. Furthermore, increased water content results in faster olivine grain growth41. The combined increase in grain growth rate and decrease in flow stress associated with higher water contents in our experiments leads to lower asthenospheric viscosity and increased thermal erosion of the lithosphere driven by diffusion creep (Fig. 3). + +<--- Page Split ---> + +The numerically predicted olivine grain size in the upper lithosphere away from shear zones (0.5–10 mm; Fig. 2a) is in agreement with naturally measured values from exhumed xenoliths18,52-54. Furthermore, recrystallized grain sizes of 10–100 μm from localized lithospheric shear zones55-58 match the grain sizes established in the diffusion-creep-dominated numerical shear zones for water contents >175 H/106Si (Fig. 2b). There are only a few constraints on grain size in the lower part of the upper mantle. However, Faul and Jackson17 suggested that the seismic signature of the upper mantle low velocity zone (LVZ) may be explained with a grain size of 5 cm together with the presence of fluids, which is consistent with our numerical results (Fig. 2a, b). + +In summary, presented numerical results are able to reproduce naturally observed distributions of olivine grain size, which indicate that dislocation creep is the dominant deformation mechanism in the upper mantle except along lithospheric shear zones, where diffusion creep is activated as a result of grain size reduction by earlier dislocation creep at high stress. The intrinsic weakness of such shear zones furthermore reduces the necessary boundary force to initiate continental rifting. Furthermore, the long- term low viscosity lithospheric shear zones allows for stretching of the continental crust and the formation of hyper- extended margins59 (Fig. 1b). + +## Methods + +Numerical experiments were conducted with a finite difference thermo- mechanical numerical code with a fully staggered Eulerian grid and a Lagrangian particle field36,37. Initial particle distribution in the Eulerian domain of 1000 x 670 km describes from top to bottom 10 km of sticky- air, 33 km of continental crust, and 627 km of upper mantle. The viscous part of the + +<--- Page Split ---> + +strain rate is composed of both dislocation and diffusion creep \(^{12}\) (see Supplementary Methods). + +We implemented viscous flow laws defined for constant fluid contents and applied values of \(C_{OH}\) + +\(= 50, 175, 600,\) or \(2500 \mathrm{H} / 10^{6} \mathrm{Si}\) , which represent experimentally observed values \(^{38,39}\) . Stresses + +are capped depending on the Drucker- Prager yield criterion to mimic brittle processes in the + +upper lithosphere (see Supplementary Methods). The initial temperature distribution describes a + +linear increase from \(0^{\circ} \mathrm{C}\) at the surface \((y = 10 \mathrm{km})\) to \(660^{\circ} \mathrm{C}\) at the Moho \((y = 43 \mathrm{km})\) , and from + +there to \(1345^{\circ} \mathrm{C}\) at the thermally- induced lithosphere- asthenosphere boundary (LAB) at \(150 \mathrm{km}\) + +depth \((y = 160 \mathrm{km})\) . Below the LAB, a static temperature increase of \(0.5^{\circ} / \mathrm{km}\) is introduced. + +Grain size of olivine is calculated based on the paleowattmeter \(^{35}\) , while all other rock + +types exhibit a constant grain size of \(1 \mathrm{mm}\) . A grain size evolution rate composed of the + +independent growth and reduction terms is introduced instead of applying a steady- state grain + +size, which would impose an immediate, time- independent equilibration of grain size. If + +dislocation creep dominates deformation, grain size is mainly defined by dynamic + +recrystallization, whereas grain size during diffusion creep is determined by grain growth \(^{40}\) . + +Therefore, only mechanical work related to dislocation creep \((\sigma \dot{\epsilon}_{dist})\) adds to grain size reduction + +rate \(^{35}\) + +\[\dot{d}_{red} = \frac{\sigma \dot{\epsilon}_{dist} \lambda d^{2}}{c \gamma}, \quad (1)\] + +where \(\sigma\) is stress, \(\dot{\epsilon}_{dist}\) is dislocation creep strain rate, \(c\) is a geometric constant \((\pi\) for spheric + +grains), \(\gamma\) is the grain boundary energy (1.4 for olivine \(^{60}\) ), and \(\lambda\) denotes the fraction of work that + +goes into grain size reduction, whereas the rest of the work goes into the shear heating term \(^{61 - 63}\) . + +<--- Page Split ---> + +A fitting of experimentally-derived olivine grain sizes versus expected grain size according to the paleowattmeter resulted in a \(\lambda\) of 0.01 (see Supplementary Methods). + +Grain growth follows a normal relationship with a grain growth rate of + +\[\dot{d}_{g r} = K_{g} f H_{2}O\exp \left(-\frac{E_{g} + P.V_{g}}{R T}\right)p^{-1}d^{1 - p}, \quad (2)\] + +where \(K_{g}\) is the rate constant, \(fH_{2}O\) is water fugacity (here as constant water content \(C_{\mathrm{OH}}\) ), \(E_{g}\) is the activation energy, \(V_{g}\) the activation volume, \(P\) is pressure, \(T\) is temperature, \(R\) is the gas constant, \(d\) is grain size, and \(p\) the growth exponent. We applied experimentally derived olivine grain growth parameters by Speciale et al. \(^{41}\) that result in significantly slower grain growth than previous constraints \(^{42}\) . + +Initial grain size distribution within the mantle logarithmically increases from 5 mm at the Moho to 10 cm at the LAB, at which size it remains farther down. High temperatures and fast growth rates in the lower part of the model domain leads to rapidly adjusting grain sizes. However, grain sizes within the lithosphere are mainly dependent on the reduction term due to lower temperatures and higher stresses. As a consequence, initial grain sizes in the lithosphere should be large enough to initially reduce. Several initial grain size distributions were tested. See Supplementary Methods for details on initial grain size distribution. + +## References + +1 Karato, S. Rheology of the deep upper mantle and its implications for the preservation of the continental roots: A review. Tectonophysics 481, 82- 98 (2010). 2 Ranalli, G. Rheology of the lithosphere in space and time. J Geol Soc London 121, 19- 37 (1997). + +<--- Page Split ---> + +Watts, A. B. Isostasy and Flexure of the Lithosphere. (Cambridge University Press, 2001). + +Burov, E. B. & Watts, A. B. The long- term strength of continental lithosphere:" jelly sandwich" or" crème brûlée"? GSA Today 16, 4 (2006). + +Jackson, J. A. Strength of the continental lithosphere: time to abandon the jelly sandwich? GSA Today 12, 4- 10 (2002). + +Maggi, A., Jackson, J. A., McKenzie, D. & Priestley, K. Earthquake focal depths, effective elastic thickness, and the strength of the continental lithosphere. Geology 28, 495- 498, doi:DOI 10.1130/0091- 7613(2000)28<495:Effdet>2.0. Co;2 (2000). + +Jin, D. H., Karato, S. & Obata, M. Mechanisms of shear localization in the continental lithosphere: Inference from the deformation microstructures of peridotites from the Ivrea zone, northwestern Italy. J Struct Geol 20, 195- 209 (1998). + +Gung, Y. C., Panning, M. & Romanowicz, B. Global anisotropy and the thickness of continents. Nature 422, 707- 711 (2003). + +Miyazaki, T., Sueyoshi, K. & Hiraga, T. Olivine crystals align during diffusion creep of Earth's upper mantle. Nature 502, 321- + (2013). + +Sundberg, M. & Cooper, R. F. Crystallographic preferred orientation produced by diffusional creep of harzburgite: Effects of chemical interactions among phases during plastic flow. J Geophys Res- Sol Ea 113 (2008). + +Faul, U. H. & Jackson, I. Diffusion creep of dry, melt- free olivine. J Geophys Res- Sol Ea 112 (2007). + +<--- Page Split ---> + +Hirth, G. & Kohlstedt, D. Rheology of the upper mantle and the mantle wedge: A view from the experimentalists. Geophysical Monograph- American Geophysical Union 138, 83- 106 (2003). + +Jain, C., Korenaga, J. & Karato, S. I. On the Grain Size Sensitivity of Olivine Rheology. J Geophys Res- Sol Ea 123, 674- 688 (2018). + +Karato, S. & Wu, P. Rheology of the Upper Mantle - a Synthesis. Science 260, 771- 778 (1993). + +Evans, B., Renner, J. & Hirth, G. A few remarks on the kinetics of static grain growth in rocks. Int J Earth Sci 90, 88- 103 (2001). + +Liao, J., Wang, Q., Gerya, T. & Ballmer, M. D. Modeling Craton Destruction by Hydration- Induced Weakening of the Upper Mantle. J Geophys Res- Sol Ea 122, 7449- 7466 (2017). + +Faul, U. H. & Jackson, I. The seismological signature of temperature and grain size variations in the upper mantle. Earth Planet Sc Lett 234, 119- 134 (2005). + +Ave Lallemant, H. G., Mercier, J. C. C., Carter, N. L. & Ross, J. V. Rheology of the Upper Mantle - Inferences from Peridotite Xenoliths. Tectonophysics 70, 85- 113 (1980). + +Bernard, R. E., Behr, W. M., Becker, T. W. & Young, D. J. Relationships Between Olivine CPO and Deformation Parameters in Naturally Deformed Rocks and Implications for Mantle Seismic Anisotropy. Geochemistry Geophysics Geosystems 20, 3469- 3494 (2019). + +Drury, M. R., Ave Lallemant, H. G., Pennock, G. M. & Palasse, L. N. Crystal preferred orientation in peridotite ultramylonites deformed by grain size sensitive creep, Etang de Lers, Pyrenees, France. J Struct Geol 33, 1776- 1789 (2011). + +<--- Page Split ---> + +Dygert, N., Bernard, R. E. & Behr, W. M. Great Basin Mantle Xenoliths Record Active Lithospheric Downwelling Beneath Central Nevada. Geochemistry Geophysics Geosystems 20, 751- 772 (2019). + +Bercovici, D., Ricard, Y. & Richards, M. The relation between mantle dynamics and plate tectonics: A primer. Geophysical Monograph- American Geophysical Union 121, 5- 46 (2000). + +Gurnis, M., Zhong, S. & Toth, J. in The History and Dynamics of Global Plate Motions Vol. Geophysical Monograph 121 73- 94 (American Geophysical Union, 2000). + +Behn, M. D., Hirth, G. & Elsenbeck, J. R. Implications of grain size evolution on the seismic structure of the oceanic upper mantle. Earth Planet Sc Lett 282, 178- 189 (2009). Braun, J. et al. A simple parameterization of strain localization in the ductile regime due to grain size reduction: A case study for olivine. J Geophys Res- Sol Ea 104, 25167- 25181 (1999). + +Hopper, J. R. & Buck, W. R. The Initiation of Rifting at Constant Tectonic Force - Role of Diffusion Creep. J Geophys Res- Sol Ea 98, 16213- 16221 (1993). + +Platt, J. P. & Behr, W. M. Grainsize evolution in ductile shear zones: Implications for strain localization and the strength of the lithosphere. J Struct Geol 33, 537- 550 (2011). Rozel, A., Ricard, Y. & Bercovici, D. A thermodynamically self- consistent damage equation for grain size evolution during dynamic recrystallization. Geophys J Int 184, 719- 728 (2011). + +Schierjott, J. C., Thielmann, M., Rozel, A. B., Golabek, G. J. & Gerya, T. V. Can Grain Size Reduction Initiate Transform Faults?- Insights From a 3- D Numerical Study. Tectonics 39 (2020). + +<--- Page Split ---> + +Hartz, E. H. & Podladchikov, Y. Y. Toasting the jelly sandwich: The effect of shear heating on lithospheric geotherms and strength. Geology 36, 331- 334 (2008). + +Willis, K., Houseman, G. A., Evans, L., Wright, T. & Hooper, A. Strain localization by shear heating and the development of lithospheric shear zones. Tectonophysics 764, 62- 76 (2019). + +Drury, M. R., Vissers, R. L. M., Vanderwal, D. & Strating, E. H. H. Shear Localization in Upper Mantle Peridotites. Pure Appl Geophys 137, 439- 460 (1991). + +Ogawa, M. Plate- like regime of a numerically modeled thermal convection in a fluid with temperature-, pressure-, and stress- history- dependent viscosity. J Geophys Res- Sol Ea 108 (2003). + +Tommasi, A. et al. Structural reactivation in plate tectonics controlled by olivine crystal anisotropy. Nat Geosci 2, 422- 426 (2009). + +Austin, N. & Evans, B. Paleowattmeters: A scaling relation for dynamically recrystallized grain size. Geology 35, 343- 346 (2007). + +Gerya, T. Introduction to Numerical Geodynamic Modelling. (Cambridge University Press, 2010). + +Ruh, J. B. Numerical modeling of tectonic underplating in accretionary wedge systems. Geosphere 16, 1385- 1407 (2020). + +Hirth, G. & Kohlstedt, D. Water in the oceanic upper mantle: Implications for rheology, melt extraction and the evolution of the lithosphere. Earth Planet Sc Lett 144, 93- 108 (1996). + +Jung, H. & Karato, S. I. Effects of water on dynamically recrystallized grain- size of olivine. J Struct Geol 23, 1337- 1344 (2001). + +<--- Page Split ---> + +Shimizu, I. Theories and applicability of grain size piezometers: The role of dynamic recrystallization mechanisms. \*J Struct Geol\* \*\*30\*\*, 899- 917 (2008). + +Speciale, P. A., Behr, W. M., Hirth, G. & Tokle, L. Rates of Olivine Grain Growth During Dynamic Recrystallization and Postdeformation Annealing. \*J Geophys Res-Sol Ea\* \*\*125\*\* (2020). + +Karato, S. Grain- Growth Kinetics in Olivine Aggregates. \*Tectonophysics\* \*\*168\*\*, 255- 273 (1989). + +Hansen, D. L. & Nielsen, S. B. Why rifts invert in compression. \*Tectonophysics\* \*\*373\*\*, 5- 24, doi:10.1016/S0040- 1951(03)00280- 4 (2003). + +Jammes, S. & Huismans, R. S. Structural styles of mountain building: Controls of lithospheric rheologic stratification and extensional inheritance. \*J Geophys Res-Sol Ea\* \*\*117\*\*, doi:Artn B10403 + +10.1029/2012jb009376 (2012). + +Candioti, L. G., Schmalholz, S. M. & Duretz, T. Impact of upper mantle convection on lithosphere hyperextension and subsequent horizontally forced subduction initiation. + +Solid Earth \*\*11\*\*, 2327- 2357 (2020). + +Kiss, D., Candioti, L. G., Duretz, T. & Schmalholz, S. M. Thermal softening induced subduction initiation at a passive margin. \*Geophys J Int\* \*\*220\*\*, 2068- 2073 (2020). + +Bird, P., Liu, Z. & Rucker, W. K. Stresses that drive the plates from below: Definitions, computational path, model optimization, and error analysis. \*J Geophys Res-Sol Ea\* \*\*113\*\* (2008). + +Gurnis, M., Hall, C. & Lavier, L. Evolving force balance during incipient subduction. \*Geochemistry Geophysics Geosystems\* \*\*5\*\* (2004). + +<--- Page Split ---> + +Katz, R. F., Spiegelman, M. & Langmuir, C. H. A new parameterization of hydrous mantle melting. Geochemistry Geophysics Geosystems 4 (2003). + +van Keken, P. E. & Ballentine, C. J. Dynamical models of mantle volatile evolution and the role of phase transitions and temperature- dependent rheology. J Geophys Res- Sol Ea 104, 7137- 7151 (1999). + +Katayama, I. & Karato, S. I. Effects of water and iron content on the rheological contrast between garnet and olivine. Phys Earth Planet In 166, 57- 66 (2008). + +Behr, W. M. & Hirth, G. Rheological properties of the mantle lid beneath the Mojave region in southern California. Earth Planet Sc Lett 393, 60- 72 (2014). + +Matysiak, A. K. & Trepmann, C. A. The deformation record of olivine in mylonitic peridotites from the Finero Complex, Ivrea Zone: Separate deformation cycles during exhumation. Tectonics 34, 2514- 2533 (2015). + +Titus, S. J., Medaris, L. G., Wang, H. F. & Tikoff, B. Continuation of the San Andreas fault system into the upper mantle: Evidence from spinel peridotite xenoliths in the Coyote Lake basalt, central California. Tectonophysics 429, 1- 20 (2007). + +Behr, W. M. & Smith, D. Deformation in the mantle wedge associated with Laramide flat- slab subduction. Geochemistry Geophysics Geosystems 17, 2643- 2660 (2016). + +Dijkstra, A. H., Drury, M. R., Vissers, R. L. M., Newman, J. & Van Roermund, H. L. M. in Flow Processes in Faults and Shear Zones Vol. Special Publications 224 (eds G. I. Alsop, R. E. Holdsworth, K. J. W. McCaffrey, & M. Hand) 11- 24 (Geological Society, 2004). + +<--- Page Split ---> + +Precigout, J., Gueydan, F., Gapais, D., Garrido, C. J. & Essaifi, A. Strain localisation in the subcontinental mantle - a ductile alternative to the brittle mantle. Tectonophysics 445, 318- 336 (2007). + +Warren, J. M. & Hirth, G. Grain size sensitive deformation mechanisms in naturally deformed peridotites. Earth Planet Sc Lett 248, 438- 450 (2006). + +Péron- Pinvidic, G. & Manatschal, G. From microcontinents to extensional allochthons: witnesses of how continents rift and break apart? 16, 189- 197, doi:10.1144/1354- 079309- 903 %J Petroleum Geoscience (2010). + +Duyster, J. & Stockhert, B. Grain boundary energies in olivine derived from natural microstructures. Contrib Mineral Petr 140, 567- 576 (2001). + +Austin, N. & Evans, B. The kinetics of microstructural evolution during deformation of calcite. J Geophys Res- Sol Ea 114 (2009). + +Poliak, E. I. & Jonas, J. J. A one- parameter approach to determining the critical conditions for the initiation of dynamic recrystallization. Acta Mater 44, 127- 136 (1996). Rosakis, P., Rosakis, A. J., Ravichandran, G. & Hodowany, J. A thermodynamic internal variable model for the partition of plastic work into heat and stored energy in metals. J Mech Phys Solids 48, 581- 607 (2000). + +## Figure captions + +Figure 1. Temporal evolution of the experiment with \(C_{\mathrm{OH}} = 600 \mathrm{H} / 10^{6} \mathrm{Si}\) . (a) Viscosity of upper mantle and marker composition of crust. White lines denote isotherms up to \(1300^{\circ} \mathrm{C}\) . (b) + +<--- Page Split ---> + +Deformation mechanism in the uppermost mantle and composition of crust. Red: Diffusion creep. White: Dislocation creep. Blue contours indicate grain size. + +Figure 2. Grain sizes in the mantle at variable water content. (a) Vertical profile at \(x = 990 \mathrm{km}\) after 5 Myr. (b) Temporal evolution lithospheric shear zones at \(y = 50 \mathrm{km}\) . (c) Temporal evolution of average lower upper mantle below \(300 \mathrm{km}\) depth. + +Figure 3. Percentage of finite strain accumulated by diffusion creep (blue) or dislocation creep (white) after 20 Myr of divergence. Red: Contour of \(\eta = 10^{21.5} \mathrm{Pa} \cdot \mathrm{s}\) indicating thickness of the elastic lithosphere. (a) \(C_{\mathrm{OH}} = 50 \mathrm{H} / 10^{6} \mathrm{Si}\) . (b) \(C_{\mathrm{OH}} = 175 \mathrm{H} / 10^{6} \mathrm{Si}\) . (c) \(C_{\mathrm{OH}} = 600 \mathrm{H} / 10^{6} \mathrm{Si}\) . (d) \(C_{\mathrm{OH}} = 2500 \mathrm{H} / 10^{6} \mathrm{Si}\) . + +Figure 4. Strength of the lithosphere. (a) Temporal evolution of the laterally averaged integrated strength (boundary force) of pure dislocation and grain- size- dependent composite diffusion- dislocation creep experiments. (b) Laterally \((x = 990 - 1000 \mathrm{km})\) averaged lithospheric strength profiles after 5 Myr. For color code see (a). (c) Strength and grain size profile along lithospheric shear zone at 5 Myr. Location of profile indicated in Fig. 1b. + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +Temporal evolution of the experiment with \(\mathrm{COH} = 600 \mathrm{H} / 106 \mathrm{Si}\) . (a) Viscosity of upper mantle and marker composition of crust. White lines denote isotherms up to \(1300^{\circ} \mathrm{C}\) . (b) Deformation mechanism in the uppermost mantle and composition of crust. Red: Diffusion creep. White: Dislocation creep. Blue contours indicate grain size. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Grain sizes in the mantle at variable water content. (a) Vertical profile at \(x = 990 \text{km}\) after 5 Myr. (b) Temporal evolution lithospheric shear zones at \(y = 50 \text{km}\) . (c) Temporal evolution of average lower upper mantle below \(300 \text{km}\) depth. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Percentage of finite strain accumulated by diffusion creep (blue) or dislocation creep (white) after 20 Myr of divergence. Red: Contour of \(\eta = 1021.5\) Pa·s indicating thickness of the elastic lithosphere. (a) \(\mathrm{COH} = 50 \mathrm{H} / 106 \mathrm{Si}\) . (b) \(\mathrm{COH} = 175 \mathrm{H} / 106 \mathrm{Si}\) . (c) \(\mathrm{COH} = 600 \mathrm{H} / 106 \mathrm{Si}\) . (d) \(\mathrm{COH} = 2500 \mathrm{H} / 106 \mathrm{Si}\) . + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Strength of the lithosphere. (a) Temporal evolution of the laterally averaged integrated strength (boundary force) of pure dislocation and grain- size- dependent composite diffusion- dislocation creep experiments. (b) Laterally ( \(x = 990 - 1000 \text{km}\) ) averaged lithospheric strength profiles after 5 Myr. For color code see (a). (c) Strength and grain size profile along lithospheric shear zone at 5 Myr. Location of profile indicated in Fig. 1b. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- RuhTokleBehrSuppMat.pdf- TableS1.xlsx- TableS2.xlsx- TableS3.xlsx- TableS4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7_det.mmd b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6e02ed77fb84f11df5570bd604ec230117935586 --- /dev/null +++ b/preprint/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7/preprint__126520c6c4bd1b20534d6fabd1749e3cd8f432b821c614e8b7b5aaa808c94bc7_det.mmd @@ -0,0 +1,469 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 936, 177]]<|/det|> +# Self-Consistent Grain Size Evolution controls Strain Localization during Rifting + +<|ref|>text<|/ref|><|det|>[[44, 195, 404, 326]]<|/det|> +Jonas Ruh ( Jonas.ruh@erdw.ethz.ch ) ETH Zurich Leif Tokle ETH Zurich Whitney Behr ETH Zürich + +<|ref|>sub_title<|/ref|><|det|>[[44, 371, 102, 389]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 408, 702, 428]]<|/det|> +Keywords: tectonic, mantle dynamics, geodynamic numerical models, rifting + +<|ref|>text<|/ref|><|det|>[[44, 446, 304, 465]]<|/det|> +Posted Date: June 22nd, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 484, 463, 504]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 561920/v1 + +<|ref|>text<|/ref|><|det|>[[44, 521, 910, 565]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 600, 953, 644]]<|/det|> +Version of Record: A version of this preprint was published at Nature Geoscience on June 16th, 2022. See the published version at https://doi.org/10.1038/s41561- 022- 00964- 9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[66, 88, 789, 110]]<|/det|> +# Self-Consistent Grain Size Evolution controls Strain Localization during Rifting + +<|ref|>text<|/ref|><|det|>[[66, 128, 88, 141]]<|/det|> +2 + +<|ref|>text<|/ref|><|det|>[[66, 159, 451, 180]]<|/det|> +3 J. B. Ruh \(^{1*}\) , L. Tokle \(^{1}\) , and W. M. Behr \(^{1}\) + +<|ref|>text<|/ref|><|det|>[[66, 192, 855, 250]]<|/det|> +\(^{1}\) Structural Geology and Tectonics group, Geological Institute, Department of Earth Sciences, ETH Zürich, Switzerland + +<|ref|>text<|/ref|><|det|>[[66, 270, 88, 283]]<|/det|> +6 + +<|ref|>text<|/ref|><|det|>[[66, 300, 622, 320]]<|/det|> +\(^{*}\) Corresponding author: Jonas B. Ruh (jonas.ruh@erdw.ethz.ch) + +<|ref|>text<|/ref|><|det|>[[66, 339, 88, 352]]<|/det|> +8 + +<|ref|>title<|/ref|><|det|>[[66, 369, 214, 388]]<|/det|> +# 9 Key Points: + +<|ref|>text<|/ref|><|det|>[[141, 404, 857, 600]]<|/det|> +- Self-consistent grain size evolution according to the paleowattmeter is coupled to a 2D thermo-mechanical numerical model of the upper mantle- Mantle rheology is represented by a coupled grain-size-dependent diffusion and grain-size-independent dislocation creep flow law- Strain localization along lithospheric shear zones leads to reduced grain sizes and diffusion creep becomes the dominant deformation mechanism + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 191, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 123, 880, 530]]<|/det|> +Geodynamic numerical models often employ solely grain- size- independent dislocation creep to describe upper mantle dynamics. However, observations from nature and rock deformation experiments suggest that shear zones can transition to a grain- size- dependent creep mechanism due to dynamic grain size evolution, with important implications for the overall strength of plate boundaries. We apply a two- dimensional thermo- mechanical numerical model with a composite diffusion- dislocation creep rheology coupled to a dynamic grain size evolution model based on the paleowattmeter. Results indicate average olivine grain sizes of 3–12 cm for the upper mantle below the LAB, while in the lithosphere grain size ranges from 0.3–3 mm at the Moho to 6–15 cm at the LAB. Such a grain size distribution results in dislocation creep being the dominant deformation mechanism in the upper mantle. However, deformation- related grain size reduction below 100 μm activates diffusion creep along lithospheric- scale shear zones during rifting, affecting the overall strength of tectonic plate boundaries. + +<|ref|>sub_title<|/ref|><|det|>[[115, 578, 162, 594]]<|/det|> +## Main + +<|ref|>text<|/ref|><|det|>[[111, 610, 884, 876]]<|/det|> +The Earth's lithosphere is defined by its mechanically rigid behavior in contrast to the relatively weak underlying asthenosphere. This rheological stratification, which ultimately allowed for the emergence of plate tectonics, primarily results from the apparent thermal gradient across the crust and upper mantle and the temperature- dependent activation of dislocation- and diffusion- related crystal- plastic creep of rocks and minerals1,2. Scaling of such experimentally- derived creep laws to natural strain rates allows us to estimate viscosities and strength of the lithosphere. Geophysical constraints on the elastic thickness of the continental lithosphere, that is a proxy for its strength3, led to contrasting conclusions on the uppermost mantle being either + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 832, 144]]<|/det|> +strong and best represented by dry dislocation creep of olivine4 or weak according to a wet olivine rheology5,6. + +<|ref|>text<|/ref|><|det|>[[110, 155, 880, 844]]<|/det|> +Whether deformation within the upper mantle is dominated by dislocation creep or diffusion creep is still a matter of debate. The observation of crystallographic preferred orientation (CPO) in mantle xenoliths7 and evidence of strong seismic anisotropy8 has long been interpreted as an indicator for dislocation creep as the dominant deformation mechanism1. However, there is reported evidence that CPO, and therefore seismic anisotropy, may also develop as a result of diffusion creep of olivine-rich aggregates9,10. In contrast to dislocation creep, the relationship between stress and strain rate for diffusion creep is dependent on grain size, which is a crucial parameter when considering the dominant deformation mechanism in the upper mantle11- 14. A transition from dislocation to diffusion creep at depths greater than \(\sim 250 \mathrm{km}\) was proposed by Hirth and Kohlstedt12 based on theoretical estimations that olivine grain size in the upper mantle is on the order of \(10 \mathrm{mm}^{15}\). Numerical experiments of mantle convection have since implemented a composite diffusion- dislocation creep rheology and constant mantle grain size, which may result in dramatic convective instability and thermal erosion of the lithosphere16. However, the assumption of a constant upper mantle grain size is an oversimplification that appears contradictory to several observational and experimental datasets. Experimental data on wave speed and attenuation of olivine, for example, fits best with a seismological model that implies an increase in grain size from \(\sim 1 \mathrm{mm}\) to \(\sim 5 \mathrm{cm}\) between depths of \(100 - 200 \mathrm{km}^{17}\). Furthermore, natural samples of exhumed lithospheric mantle rocks show a large variety of grain sizes ranging from tens to hundreds of microns in olivine mylonites and tectonites to the centimeter- scale in weakly deformed or annealed xenoliths18- 21. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 565]]<|/det|> +Active plate tectonics requires mechanical weakening and strain localization along lithospheric shear zones at the plate boundaries22,23. Several studies suggest that grain size reduction and the consequent activation of diffusion creep is a viable process to initiate localization of deformation in the lithosphere24- 29, perhaps complementary to other potential weakening mechanisms such as shear heating30,31, reaction- induced weakening32, or the presence of preexisting weak zones or viscous anisotropy33,34. Here, we present a 2D thermo- mechanical numerical model with a composite diffusion- dislocation creep flow law coupled to a self- consistent grain size evolution model based on the paleowattmeter35. Such a model allows us to estimate apparent grain size distribution and the dominant deformation mechanism within the upper mantle, and to investigate the importance of grain size evolution for strain localization in the lithosphere during continental rifting. We test the influence of water content in the mantle which affects both its viscosity and rate of grain growth. Furthermore, the effect of localized grain- size- dependent weakening on the long- term strength and elastic thickness of continental lithosphere is investigated and compared to pure dislocation creep experiments. + +<|ref|>sub_title<|/ref|><|det|>[[113, 611, 802, 632]]<|/det|> +## Coupled grain size evolution thermo-mechanical model of upper mantle dynamics + +<|ref|>text<|/ref|><|det|>[[111, 644, 877, 875]]<|/det|> +We apply a finite difference thermo- mechanical numerical model36,37 of the upper mantle and crust with an Eulerian domain of \(1000 \times 670 \mathrm{km}\) that undergoes horizontal divergence at a constant total rate of \(1 \mathrm{cm / yr}\) . The model employs a visco- elasto- plastic rheology where the viscous strain rate is composed of both dislocation and diffusion creep for constant water content12 and stresses are capped depending on the Drucker- Prager yield criterion (see Methods and Supplementary Methods). The applied fluid contents in the mantle are \(C_{\mathrm{OH}} = 50 - 2500 \mathrm{H / 10^6Si}\) , which cover the range of estimated values obtained from experimental studies38,39. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 876, 530]]<|/det|> +Olivine grain size is calculated based on the paleowattmeter \(^{35}\) , which introduces a grain size evolution rate composed of independent growth and reduction terms (see Methods). Grain size reduction occurs by the process of dynamic recrystallization during dislocation creep, whereas grain size during diffusion creep is controlled by the process of grain growth \(^{40}\) . Based on grain sizes from experimentally deformed olivine aggregates, the fraction of work that goes into grain size reduction during dynamic recrystallization is estimated to be \(\lambda = 0.01\) (see Supplementary Methods). The grain growth parameters we implement are derived from experiments on natural olivine aggregates with in-situ water contents \(^{41}\) that predict significantly slower grain growth than previous constraints from experiments on water- saturated, synthetic olivine \(^{42}\) . Due to high temperatures and thus fast growth rates, the initial grain sizes in the lower part of the model domain rapidly adjust to a steady- state grain size. On the other hand, initial grain sizes within the lithosphere are mainly driven by the reduction term due to lower temperatures and higher deviatoric stresses. + +<|ref|>sub_title<|/ref|><|det|>[[115, 579, 675, 599]]<|/det|> +## Rheological implications and formation of lithospheric shear zones + +<|ref|>text<|/ref|><|det|>[[110, 612, 866, 877]]<|/det|> +Composite diffusion- dislocation creep numerical experiments were conducted with variable water content in the mantle ( \(C_{\mathrm{OH}} = 50\) , 175, 600, and 2500 H/10 \(^{6}\) Si) that affect both viscous creep and grain growth. Mantle viscosities of the reference model ( \(C_{\mathrm{OH}} = 600\) H/10 \(^{6}\) Si) show values of \(10^{19} - 10^{21}\) Pa·s for the asthenosphere after 5 Myr of divergence (Fig. 1a). At 10 Myr, lithospheric thinning and related temperature increase below the rifted region lead to viscosities as low as \(5 \cdot 10^{17}\) Pa·s, relatively fast velocities, and gravitationally- induced lithospheric dripping. After 15 and 20 Myr of divergence, asthenospheric viscosities remain within \(10^{18} - 10^{21}\) Pa·s, with lower values where fast velocities occur due to thermally- and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 819, 144]]<|/det|> +gravitationally- induced lithospheric erosion (Fig. 1a). Away from the rift, the lithosphere remains intact and strong. + +<|ref|>text<|/ref|><|det|>[[112, 157, 884, 352]]<|/det|> +Illustrations of the dominant deformation mechanism (dislocation vs. diffusion creep) and contours of grain size in the mantle demonstrate that localization of stress in the centre of the model domain leads to grain size reduction and the activation of diffusion creep along large- scale lithospheric shear zones (Fig. 1b). The large shear zones retain relatively small grain sizes and remain dominated by diffusion creep even after 15 to 20 Myr of divergence, when a mid- ocean spreading centre is established, consuming most of the extensional velocity. + +<|ref|>text<|/ref|><|det|>[[111, 366, 875, 771]]<|/det|> +For all experiments, grain sizes vary spatially throughout the model domain; furthermore, their values are strongly sensitive implemented mantle water content. Vertical grain size profiles along the side of the domain (at 5 Myr), away from the extensional zone, show values of 0.3–3 mm at the Moho (depending on \(C_{\mathrm{OH}}\) ) that increase to 6–15 cm at the LAB, and decrease to 2–7 cm at the base of the upper mantle (Fig. 2a). Grain sizes within localizing shear zones in the uppermost lithosphere at 40 km depth \((y = 50 \mathrm{km})\) show a rapid initial decrease to 60–250 \(\mu \mathrm{m}\) (Fig. 2b). Depending on the water content in the mantle, they are able to recover after \(\sim 15 \mathrm{Myr}\) \((C_{\mathrm{OH}} = 2500 \mathrm{H / 10^{6}Si})\) or \(\sim 20 \mathrm{Myr}\) \((C_{\mathrm{OH}} = 600 \mathrm{H / 10^{6}Si})\) . Lower water content hampers substantial grain growth within previously active shear zones before 40 Myr. Average upper mantle grain sizes below 300 km depth establish within \(\sim 2 \mathrm{Myr}\) and range in between 3–12 cm (Fig. 2c). Further undulations in average mantle grain size result from the downwelling of small- grain- size lithospheric dripplets. + +<|ref|>text<|/ref|><|det|>[[112, 785, 875, 876]]<|/det|> +Figure 3 shows the portions of accumulated finite viscous strain within the mantle accommodated by diffusion and dislocation creep after 20 Myr of divergence. Dislocation creep is the dominant deformation mechanism in large parts of the upper mantle, independent of water + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 866, 213]]<|/det|> +content. Diffusion creep dominates within lithospheric shear zones that form in the early stages of rifting (Fig. 1b) and assist in lithospheric dripping (Fig. 3b- d). The continental lithospheric thickness defined by its viscosity varies between 90–150 km, depending on water content (Fig. 3). + +<|ref|>sub_title<|/ref|><|det|>[[115, 264, 485, 284]]<|/det|> +## Effects of grain size on lithospheric strength + +<|ref|>text<|/ref|><|det|>[[111, 297, 880, 842]]<|/det|> +The importance of a self- consistent grain size distribution for upper mantle dynamics becomes evident when comparing our results to numerical experiments with pure dislocation creep of olivine or composite diffusion- dislocation creep with a constant grain size throughout the entire upper mantle. Experiments with dry dislocation creep result in extensive brittle- plastic deformation of the lithosphere43,44. Experiments with composite diffusion- dislocation creep and small constant grain size (1 mm) results in a lithosphere thinned by convective erosion (<90 km) driven by low asthenosphere viscosities of <1018 Pa·s16,45. For constant grain sizes larger than 1 cm, dislocation creep becomes the main deformation mechanism throughout the entire upper mantle16. These numerical experiments fail to match the effective elastic lithospheric thicknesses necessary to sustain orogens4, while brittle deformation in the lithosphere remains absent6. On the other hand, our implementation of a self- consistent grain size evolution is able to resolve this obstacle. Observed lithospheric thicknesses vary between 90–150 km (Fig. 3), while localization of deformation in the lithosphere rapidly leads to grain size reduction, diffusion creep activation, and related stress drop below the frictional yield, omitting failure. The diffusion- creep- related stress drop furthermore reduces and replaces the importance of shear heating along lithospheric shear zones30,46. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 881, 565]]<|/det|> +The temporal evolution of the vertically integrated strength illustrates that experiments with composite diffusion- dislocation creep coupled to a self- consistent grain size evolution show a decrease of boundary forces below 5 TN/m within 1–2 Myr, while pure dislocation creep experiments remain above 10 TN/m for at least \(\sim 15\) Myr (Fig. 4a). Typical forces along plate boundaries are on the order of 1–5 TN/m47,48, which is sufficient to initiate continental rifting if the grain size is small enough and diffusion creep dominates deformation26. Vertical strength profiles indicate that most of the strength of coupled experiments remains within the crust with maximal values of \(\sim 200\) MPa, while pure dislocation creep experiments exhibit at least 10 km of brittle- plastic mantle lithosphere with differential stresses up to \(\sim 600–700\) MPa (Fig. 4b). Differential stresses of \(\sim 200\) MPa close to the Moho in composite diffusion- dislocation creep experiments stand in contrast to significantly lower strength along a lithospheric shear zone after 5 Myr (Fig. 4c). There, values of 1–10 MPa are defined by grain sizes as small as 100 μm and diffusion creep as the consequent deformation mechanism, efficiently weakening the entire lithospheric rift system. + +<|ref|>text<|/ref|><|det|>[[111, 576, 877, 841]]<|/det|> +Water content in the upper mantle has important implications for the relationships between viscous flow and seismic anisotropy39, hydrous melting49, and the distribution of geochemical reservoirs50. The strength of olivine in the presence of water is significantly reduced38,51, as expressed in the flow law we apply here12. Furthermore, increased water content results in faster olivine grain growth41. The combined increase in grain growth rate and decrease in flow stress associated with higher water contents in our experiments leads to lower asthenospheric viscosity and increased thermal erosion of the lithosphere driven by diffusion creep (Fig. 3). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 388]]<|/det|> +The numerically predicted olivine grain size in the upper lithosphere away from shear zones (0.5–10 mm; Fig. 2a) is in agreement with naturally measured values from exhumed xenoliths18,52-54. Furthermore, recrystallized grain sizes of 10–100 μm from localized lithospheric shear zones55-58 match the grain sizes established in the diffusion-creep-dominated numerical shear zones for water contents >175 H/106Si (Fig. 2b). There are only a few constraints on grain size in the lower part of the upper mantle. However, Faul and Jackson17 suggested that the seismic signature of the upper mantle low velocity zone (LVZ) may be explained with a grain size of 5 cm together with the presence of fluids, which is consistent with our numerical results (Fig. 2a, b). + +<|ref|>text<|/ref|><|det|>[[112, 402, 872, 667]]<|/det|> +In summary, presented numerical results are able to reproduce naturally observed distributions of olivine grain size, which indicate that dislocation creep is the dominant deformation mechanism in the upper mantle except along lithospheric shear zones, where diffusion creep is activated as a result of grain size reduction by earlier dislocation creep at high stress. The intrinsic weakness of such shear zones furthermore reduces the necessary boundary force to initiate continental rifting. Furthermore, the long- term low viscosity lithospheric shear zones allows for stretching of the continental crust and the formation of hyper- extended margins59 (Fig. 1b). + +<|ref|>sub_title<|/ref|><|det|>[[115, 717, 191, 734]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[112, 750, 880, 875]]<|/det|> +Numerical experiments were conducted with a finite difference thermo- mechanical numerical code with a fully staggered Eulerian grid and a Lagrangian particle field36,37. Initial particle distribution in the Eulerian domain of 1000 x 670 km describes from top to bottom 10 km of sticky- air, 33 km of continental crust, and 627 km of upper mantle. The viscous part of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 879, 110]]<|/det|> +strain rate is composed of both dislocation and diffusion creep \(^{12}\) (see Supplementary Methods). + +<|ref|>text<|/ref|><|det|>[[111, 123, 880, 145]]<|/det|> +We implemented viscous flow laws defined for constant fluid contents and applied values of \(C_{OH}\) + +<|ref|>text<|/ref|><|det|>[[111, 157, 870, 180]]<|/det|> +\(= 50, 175, 600,\) or \(2500 \mathrm{H} / 10^{6} \mathrm{Si}\) , which represent experimentally observed values \(^{38,39}\) . Stresses + +<|ref|>text<|/ref|><|det|>[[111, 192, 872, 214]]<|/det|> +are capped depending on the Drucker- Prager yield criterion to mimic brittle processes in the + +<|ref|>text<|/ref|><|det|>[[111, 227, 872, 248]]<|/det|> +upper lithosphere (see Supplementary Methods). The initial temperature distribution describes a + +<|ref|>text<|/ref|><|det|>[[111, 261, 872, 283]]<|/det|> +linear increase from \(0^{\circ} \mathrm{C}\) at the surface \((y = 10 \mathrm{km})\) to \(660^{\circ} \mathrm{C}\) at the Moho \((y = 43 \mathrm{km})\) , and from + +<|ref|>text<|/ref|><|det|>[[111, 296, 872, 318]]<|/det|> +there to \(1345^{\circ} \mathrm{C}\) at the thermally- induced lithosphere- asthenosphere boundary (LAB) at \(150 \mathrm{km}\) + +<|ref|>text<|/ref|><|det|>[[111, 330, 835, 352]]<|/det|> +depth \((y = 160 \mathrm{km})\) . Below the LAB, a static temperature increase of \(0.5^{\circ} / \mathrm{km}\) is introduced. + +<|ref|>text<|/ref|><|det|>[[111, 365, 848, 387]]<|/det|> +Grain size of olivine is calculated based on the paleowattmeter \(^{35}\) , while all other rock + +<|ref|>text<|/ref|><|det|>[[111, 400, 810, 421]]<|/det|> +types exhibit a constant grain size of \(1 \mathrm{mm}\) . A grain size evolution rate composed of the + +<|ref|>text<|/ref|><|det|>[[111, 434, 855, 456]]<|/det|> +independent growth and reduction terms is introduced instead of applying a steady- state grain + +<|ref|>text<|/ref|><|det|>[[111, 469, 812, 490]]<|/det|> +size, which would impose an immediate, time- independent equilibration of grain size. If + +<|ref|>text<|/ref|><|det|>[[111, 504, 761, 525]]<|/det|> +dislocation creep dominates deformation, grain size is mainly defined by dynamic + +<|ref|>text<|/ref|><|det|>[[111, 538, 837, 560]]<|/det|> +recrystallization, whereas grain size during diffusion creep is determined by grain growth \(^{40}\) . + +<|ref|>text<|/ref|><|det|>[[111, 573, 880, 595]]<|/det|> +Therefore, only mechanical work related to dislocation creep \((\sigma \dot{\epsilon}_{dist})\) adds to grain size reduction + +<|ref|>text<|/ref|><|det|>[[111, 609, 161, 629]]<|/det|> +rate \(^{35}\) + +<|ref|>equation<|/ref|><|det|>[[111, 644, 847, 682]]<|/det|> +\[\dot{d}_{red} = \frac{\sigma \dot{\epsilon}_{dist} \lambda d^{2}}{c \gamma}, \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[111, 695, 850, 717]]<|/det|> +where \(\sigma\) is stress, \(\dot{\epsilon}_{dist}\) is dislocation creep strain rate, \(c\) is a geometric constant \((\pi\) for spheric + +<|ref|>text<|/ref|><|det|>[[111, 730, 875, 753]]<|/det|> +grains), \(\gamma\) is the grain boundary energy (1.4 for olivine \(^{60}\) ), and \(\lambda\) denotes the fraction of work that + +<|ref|>text<|/ref|><|det|>[[111, 765, 870, 787]]<|/det|> +goes into grain size reduction, whereas the rest of the work goes into the shear heating term \(^{61 - 63}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 857, 146]]<|/det|> +A fitting of experimentally-derived olivine grain sizes versus expected grain size according to the paleowattmeter resulted in a \(\lambda\) of 0.01 (see Supplementary Methods). + +<|ref|>text<|/ref|><|det|>[[170, 160, 735, 181]]<|/det|> +Grain growth follows a normal relationship with a grain growth rate of + +<|ref|>equation<|/ref|><|det|>[[112, 193, 846, 223]]<|/det|> +\[\dot{d}_{g r} = K_{g} f H_{2}O\exp \left(-\frac{E_{g} + P.V_{g}}{R T}\right)p^{-1}d^{1 - p}, \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[112, 235, 869, 394]]<|/det|> +where \(K_{g}\) is the rate constant, \(fH_{2}O\) is water fugacity (here as constant water content \(C_{\mathrm{OH}}\) ), \(E_{g}\) is the activation energy, \(V_{g}\) the activation volume, \(P\) is pressure, \(T\) is temperature, \(R\) is the gas constant, \(d\) is grain size, and \(p\) the growth exponent. We applied experimentally derived olivine grain growth parameters by Speciale et al. \(^{41}\) that result in significantly slower grain growth than previous constraints \(^{42}\) . + +<|ref|>text<|/ref|><|det|>[[112, 408, 883, 639]]<|/det|> +Initial grain size distribution within the mantle logarithmically increases from 5 mm at the Moho to 10 cm at the LAB, at which size it remains farther down. High temperatures and fast growth rates in the lower part of the model domain leads to rapidly adjusting grain sizes. However, grain sizes within the lithosphere are mainly dependent on the reduction term due to lower temperatures and higher stresses. As a consequence, initial grain sizes in the lithosphere should be large enough to initially reduce. Several initial grain size distributions were tested. See Supplementary Methods for details on initial grain size distribution. + +<|ref|>sub_title<|/ref|><|det|>[[113, 688, 209, 705]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[112, 720, 880, 848]]<|/det|> +1 Karato, S. Rheology of the deep upper mantle and its implications for the preservation of the continental roots: A review. Tectonophysics 481, 82- 98 (2010). 2 Ranalli, G. Rheology of the lithosphere in space and time. J Geol Soc London 121, 19- 37 (1997). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 840, 143]]<|/det|> +Watts, A. B. Isostasy and Flexure of the Lithosphere. (Cambridge University Press, 2001). + +<|ref|>text<|/ref|><|det|>[[170, 157, 844, 212]]<|/det|> +Burov, E. B. & Watts, A. B. The long- term strength of continental lithosphere:" jelly sandwich" or" crème brûlée"? GSA Today 16, 4 (2006). + +<|ref|>text<|/ref|><|det|>[[170, 227, 800, 281]]<|/det|> +Jackson, J. A. Strength of the continental lithosphere: time to abandon the jelly sandwich? GSA Today 12, 4- 10 (2002). + +<|ref|>text<|/ref|><|det|>[[170, 295, 857, 386]]<|/det|> +Maggi, A., Jackson, J. A., McKenzie, D. & Priestley, K. Earthquake focal depths, effective elastic thickness, and the strength of the continental lithosphere. Geology 28, 495- 498, doi:DOI 10.1130/0091- 7613(2000)28<495:Effdet>2.0. Co;2 (2000). + +<|ref|>text<|/ref|><|det|>[[170, 401, 875, 492]]<|/det|> +Jin, D. H., Karato, S. & Obata, M. Mechanisms of shear localization in the continental lithosphere: Inference from the deformation microstructures of peridotites from the Ivrea zone, northwestern Italy. J Struct Geol 20, 195- 209 (1998). + +<|ref|>text<|/ref|><|det|>[[170, 506, 847, 561]]<|/det|> +Gung, Y. C., Panning, M. & Romanowicz, B. Global anisotropy and the thickness of continents. Nature 422, 707- 711 (2003). + +<|ref|>text<|/ref|><|det|>[[170, 575, 868, 630]]<|/det|> +Miyazaki, T., Sueyoshi, K. & Hiraga, T. Olivine crystals align during diffusion creep of Earth's upper mantle. Nature 502, 321- + (2013). + +<|ref|>text<|/ref|><|det|>[[170, 644, 859, 736]]<|/det|> +Sundberg, M. & Cooper, R. F. Crystallographic preferred orientation produced by diffusional creep of harzburgite: Effects of chemical interactions among phases during plastic flow. J Geophys Res- Sol Ea 113 (2008). + +<|ref|>text<|/ref|><|det|>[[170, 750, 877, 805]]<|/det|> +Faul, U. H. & Jackson, I. Diffusion creep of dry, melt- free olivine. J Geophys Res- Sol Ea 112 (2007). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 864, 178]]<|/det|> +Hirth, G. & Kohlstedt, D. Rheology of the upper mantle and the mantle wedge: A view from the experimentalists. Geophysical Monograph- American Geophysical Union 138, 83- 106 (2003). + +<|ref|>text<|/ref|><|det|>[[170, 192, 870, 249]]<|/det|> +Jain, C., Korenaga, J. & Karato, S. I. On the Grain Size Sensitivity of Olivine Rheology. J Geophys Res- Sol Ea 123, 674- 688 (2018). + +<|ref|>text<|/ref|><|det|>[[170, 262, 866, 318]]<|/det|> +Karato, S. & Wu, P. Rheology of the Upper Mantle - a Synthesis. Science 260, 771- 778 (1993). + +<|ref|>text<|/ref|><|det|>[[170, 331, 871, 387]]<|/det|> +Evans, B., Renner, J. & Hirth, G. A few remarks on the kinetics of static grain growth in rocks. Int J Earth Sci 90, 88- 103 (2001). + +<|ref|>text<|/ref|><|det|>[[170, 401, 860, 492]]<|/det|> +Liao, J., Wang, Q., Gerya, T. & Ballmer, M. D. Modeling Craton Destruction by Hydration- Induced Weakening of the Upper Mantle. J Geophys Res- Sol Ea 122, 7449- 7466 (2017). + +<|ref|>text<|/ref|><|det|>[[170, 505, 835, 561]]<|/det|> +Faul, U. H. & Jackson, I. The seismological signature of temperature and grain size variations in the upper mantle. Earth Planet Sc Lett 234, 119- 134 (2005). + +<|ref|>text<|/ref|><|det|>[[170, 575, 875, 640]]<|/det|> +Ave Lallemant, H. G., Mercier, J. C. C., Carter, N. L. & Ross, J. V. Rheology of the Upper Mantle - Inferences from Peridotite Xenoliths. Tectonophysics 70, 85- 113 (1980). + +<|ref|>text<|/ref|><|det|>[[170, 650, 883, 771]]<|/det|> +Bernard, R. E., Behr, W. M., Becker, T. W. & Young, D. J. Relationships Between Olivine CPO and Deformation Parameters in Naturally Deformed Rocks and Implications for Mantle Seismic Anisotropy. Geochemistry Geophysics Geosystems 20, 3469- 3494 (2019). + +<|ref|>text<|/ref|><|det|>[[170, 785, 866, 876]]<|/det|> +Drury, M. R., Ave Lallemant, H. G., Pennock, G. M. & Palasse, L. N. Crystal preferred orientation in peridotite ultramylonites deformed by grain size sensitive creep, Etang de Lers, Pyrenees, France. J Struct Geol 33, 1776- 1789 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 870, 177]]<|/det|> +Dygert, N., Bernard, R. E. & Behr, W. M. Great Basin Mantle Xenoliths Record Active Lithospheric Downwelling Beneath Central Nevada. Geochemistry Geophysics Geosystems 20, 751- 772 (2019). + +<|ref|>text<|/ref|><|det|>[[170, 191, 876, 285]]<|/det|> +Bercovici, D., Ricard, Y. & Richards, M. The relation between mantle dynamics and plate tectonics: A primer. Geophysical Monograph- American Geophysical Union 121, 5- 46 (2000). + +<|ref|>text<|/ref|><|det|>[[170, 298, 876, 360]]<|/det|> +Gurnis, M., Zhong, S. & Toth, J. in The History and Dynamics of Global Plate Motions Vol. Geophysical Monograph 121 73- 94 (American Geophysical Union, 2000). + +<|ref|>text<|/ref|><|det|>[[170, 368, 876, 528]]<|/det|> +Behn, M. D., Hirth, G. & Elsenbeck, J. R. Implications of grain size evolution on the seismic structure of the oceanic upper mantle. Earth Planet Sc Lett 282, 178- 189 (2009). Braun, J. et al. A simple parameterization of strain localization in the ductile regime due to grain size reduction: A case study for olivine. J Geophys Res- Sol Ea 104, 25167- 25181 (1999). + +<|ref|>text<|/ref|><|det|>[[170, 540, 870, 592]]<|/det|> +Hopper, J. R. & Buck, W. R. The Initiation of Rifting at Constant Tectonic Force - Role of Diffusion Creep. J Geophys Res- Sol Ea 98, 16213- 16221 (1993). + +<|ref|>text<|/ref|><|det|>[[170, 607, 866, 770]]<|/det|> +Platt, J. P. & Behr, W. M. Grainsize evolution in ductile shear zones: Implications for strain localization and the strength of the lithosphere. J Struct Geol 33, 537- 550 (2011). Rozel, A., Ricard, Y. & Bercovici, D. A thermodynamically self- consistent damage equation for grain size evolution during dynamic recrystallization. Geophys J Int 184, 719- 728 (2011). + +<|ref|>text<|/ref|><|det|>[[170, 785, 866, 876]]<|/det|> +Schierjott, J. C., Thielmann, M., Rozel, A. B., Golabek, G. J. & Gerya, T. V. Can Grain Size Reduction Initiate Transform Faults?- Insights From a 3- D Numerical Study. Tectonics 39 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 840, 149]]<|/det|> +Hartz, E. H. & Podladchikov, Y. Y. Toasting the jelly sandwich: The effect of shear heating on lithospheric geotherms and strength. Geology 36, 331- 334 (2008). + +<|ref|>text<|/ref|><|det|>[[170, 158, 866, 248]]<|/det|> +Willis, K., Houseman, G. A., Evans, L., Wright, T. & Hooper, A. Strain localization by shear heating and the development of lithospheric shear zones. Tectonophysics 764, 62- 76 (2019). + +<|ref|>text<|/ref|><|det|>[[170, 261, 883, 320]]<|/det|> +Drury, M. R., Vissers, R. L. M., Vanderwal, D. & Strating, E. H. H. Shear Localization in Upper Mantle Peridotites. Pure Appl Geophys 137, 439- 460 (1991). + +<|ref|>text<|/ref|><|det|>[[170, 331, 881, 425]]<|/det|> +Ogawa, M. Plate- like regime of a numerically modeled thermal convection in a fluid with temperature-, pressure-, and stress- history- dependent viscosity. J Geophys Res- Sol Ea 108 (2003). + +<|ref|>text<|/ref|><|det|>[[170, 436, 869, 495]]<|/det|> +Tommasi, A. et al. Structural reactivation in plate tectonics controlled by olivine crystal anisotropy. Nat Geosci 2, 422- 426 (2009). + +<|ref|>text<|/ref|><|det|>[[170, 506, 774, 565]]<|/det|> +Austin, N. & Evans, B. Paleowattmeters: A scaling relation for dynamically recrystallized grain size. Geology 35, 343- 346 (2007). + +<|ref|>text<|/ref|><|det|>[[170, 576, 850, 634]]<|/det|> +Gerya, T. Introduction to Numerical Geodynamic Modelling. (Cambridge University Press, 2010). + +<|ref|>text<|/ref|><|det|>[[170, 645, 864, 703]]<|/det|> +Ruh, J. B. Numerical modeling of tectonic underplating in accretionary wedge systems. Geosphere 16, 1385- 1407 (2020). + +<|ref|>text<|/ref|><|det|>[[170, 714, 867, 807]]<|/det|> +Hirth, G. & Kohlstedt, D. Water in the oceanic upper mantle: Implications for rheology, melt extraction and the evolution of the lithosphere. Earth Planet Sc Lett 144, 93- 108 (1996). + +<|ref|>text<|/ref|><|det|>[[170, 819, 838, 877]]<|/det|> +Jung, H. & Karato, S. I. Effects of water on dynamically recrystallized grain- size of olivine. J Struct Geol 23, 1337- 1344 (2001). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 850, 144]]<|/det|> +Shimizu, I. Theories and applicability of grain size piezometers: The role of dynamic recrystallization mechanisms. \*J Struct Geol\* \*\*30\*\*, 899- 917 (2008). + +<|ref|>text<|/ref|><|det|>[[170, 157, 861, 247]]<|/det|> +Speciale, P. A., Behr, W. M., Hirth, G. & Tokle, L. Rates of Olivine Grain Growth During Dynamic Recrystallization and Postdeformation Annealing. \*J Geophys Res-Sol Ea\* \*\*125\*\* (2020). + +<|ref|>text<|/ref|><|det|>[[170, 262, 861, 317]]<|/det|> +Karato, S. Grain- Growth Kinetics in Olivine Aggregates. \*Tectonophysics\* \*\*168\*\*, 255- 273 (1989). + +<|ref|>text<|/ref|><|det|>[[170, 331, 865, 387]]<|/det|> +Hansen, D. L. & Nielsen, S. B. Why rifts invert in compression. \*Tectonophysics\* \*\*373\*\*, 5- 24, doi:10.1016/S0040- 1951(03)00280- 4 (2003). + +<|ref|>text<|/ref|><|det|>[[170, 401, 857, 492]]<|/det|> +Jammes, S. & Huismans, R. S. Structural styles of mountain building: Controls of lithospheric rheologic stratification and extensional inheritance. \*J Geophys Res-Sol Ea\* \*\*117\*\*, doi:Artn B10403 + +<|ref|>text<|/ref|><|det|>[[170, 507, 363, 526]]<|/det|> +10.1029/2012jb009376 (2012). + +<|ref|>text<|/ref|><|det|>[[170, 541, 860, 597]]<|/det|> +Candioti, L. G., Schmalholz, S. M. & Duretz, T. Impact of upper mantle convection on lithosphere hyperextension and subsequent horizontally forced subduction initiation. + +<|ref|>text<|/ref|><|det|>[[170, 611, 448, 630]]<|/det|> +Solid Earth \*\*11\*\*, 2327- 2357 (2020). + +<|ref|>text<|/ref|><|det|>[[170, 645, 848, 700]]<|/det|> +Kiss, D., Candioti, L. G., Duretz, T. & Schmalholz, S. M. Thermal softening induced subduction initiation at a passive margin. \*Geophys J Int\* \*\*220\*\*, 2068- 2073 (2020). + +<|ref|>text<|/ref|><|det|>[[170, 714, 866, 805]]<|/det|> +Bird, P., Liu, Z. & Rucker, W. K. Stresses that drive the plates from below: Definitions, computational path, model optimization, and error analysis. \*J Geophys Res-Sol Ea\* \*\*113\*\* (2008). + +<|ref|>text<|/ref|><|det|>[[170, 820, 848, 875]]<|/det|> +Gurnis, M., Hall, C. & Lavier, L. Evolving force balance during incipient subduction. \*Geochemistry Geophysics Geosystems\* \*\*5\*\* (2004). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 880, 150]]<|/det|> +Katz, R. F., Spiegelman, M. & Langmuir, C. H. A new parameterization of hydrous mantle melting. Geochemistry Geophysics Geosystems 4 (2003). + +<|ref|>text<|/ref|><|det|>[[170, 157, 875, 250]]<|/det|> +van Keken, P. E. & Ballentine, C. J. Dynamical models of mantle volatile evolution and the role of phase transitions and temperature- dependent rheology. J Geophys Res- Sol Ea 104, 7137- 7151 (1999). + +<|ref|>text<|/ref|><|det|>[[170, 261, 875, 319]]<|/det|> +Katayama, I. & Karato, S. I. Effects of water and iron content on the rheological contrast between garnet and olivine. Phys Earth Planet In 166, 57- 66 (2008). + +<|ref|>text<|/ref|><|det|>[[170, 329, 850, 389]]<|/det|> +Behr, W. M. & Hirth, G. Rheological properties of the mantle lid beneath the Mojave region in southern California. Earth Planet Sc Lett 393, 60- 72 (2014). + +<|ref|>text<|/ref|><|det|>[[170, 400, 840, 493]]<|/det|> +Matysiak, A. K. & Trepmann, C. A. The deformation record of olivine in mylonitic peridotites from the Finero Complex, Ivrea Zone: Separate deformation cycles during exhumation. Tectonics 34, 2514- 2533 (2015). + +<|ref|>text<|/ref|><|det|>[[170, 504, 857, 600]]<|/det|> +Titus, S. J., Medaris, L. G., Wang, H. F. & Tikoff, B. Continuation of the San Andreas fault system into the upper mantle: Evidence from spinel peridotite xenoliths in the Coyote Lake basalt, central California. Tectonophysics 429, 1- 20 (2007). + +<|ref|>text<|/ref|><|det|>[[170, 610, 856, 670]]<|/det|> +Behr, W. M. & Smith, D. Deformation in the mantle wedge associated with Laramide flat- slab subduction. Geochemistry Geophysics Geosystems 17, 2643- 2660 (2016). + +<|ref|>text<|/ref|><|det|>[[170, 680, 872, 807]]<|/det|> +Dijkstra, A. H., Drury, M. R., Vissers, R. L. M., Newman, J. & Van Roermund, H. L. M. in Flow Processes in Faults and Shear Zones Vol. Special Publications 224 (eds G. I. Alsop, R. E. Holdsworth, K. J. W. McCaffrey, & M. Hand) 11- 24 (Geological Society, 2004). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[170, 88, 878, 178]]<|/det|> +Precigout, J., Gueydan, F., Gapais, D., Garrido, C. J. & Essaifi, A. Strain localisation in the subcontinental mantle - a ductile alternative to the brittle mantle. Tectonophysics 445, 318- 336 (2007). + +<|ref|>text<|/ref|><|det|>[[170, 192, 840, 248]]<|/det|> +Warren, J. M. & Hirth, G. Grain size sensitive deformation mechanisms in naturally deformed peridotites. Earth Planet Sc Lett 248, 438- 450 (2006). + +<|ref|>text<|/ref|><|det|>[[170, 262, 864, 355]]<|/det|> +Péron- Pinvidic, G. & Manatschal, G. From microcontinents to extensional allochthons: witnesses of how continents rift and break apart? 16, 189- 197, doi:10.1144/1354- 079309- 903 %J Petroleum Geoscience (2010). + +<|ref|>text<|/ref|><|det|>[[170, 368, 837, 424]]<|/det|> +Duyster, J. & Stockhert, B. Grain boundary energies in olivine derived from natural microstructures. Contrib Mineral Petr 140, 567- 576 (2001). + +<|ref|>text<|/ref|><|det|>[[170, 437, 872, 493]]<|/det|> +Austin, N. & Evans, B. The kinetics of microstructural evolution during deformation of calcite. J Geophys Res- Sol Ea 114 (2009). + +<|ref|>text<|/ref|><|det|>[[170, 506, 876, 666]]<|/det|> +Poliak, E. I. & Jonas, J. J. A one- parameter approach to determining the critical conditions for the initiation of dynamic recrystallization. Acta Mater 44, 127- 136 (1996). Rosakis, P., Rosakis, A. J., Ravichandran, G. & Hodowany, J. A thermodynamic internal variable model for the partition of plastic work into heat and stored energy in metals. J Mech Phys Solids 48, 581- 607 (2000). + +<|ref|>sub_title<|/ref|><|det|>[[115, 717, 248, 736]]<|/det|> +## Figure captions + +<|ref|>text<|/ref|><|det|>[[115, 787, 872, 840]]<|/det|> +Figure 1. Temporal evolution of the experiment with \(C_{\mathrm{OH}} = 600 \mathrm{H} / 10^{6} \mathrm{Si}\) . (a) Viscosity of upper mantle and marker composition of crust. White lines denote isotherms up to \(1300^{\circ} \mathrm{C}\) . (b) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 835, 144]]<|/det|> +Deformation mechanism in the uppermost mantle and composition of crust. Red: Diffusion creep. White: Dislocation creep. Blue contours indicate grain size. + +<|ref|>text<|/ref|><|det|>[[112, 193, 861, 283]]<|/det|> +Figure 2. Grain sizes in the mantle at variable water content. (a) Vertical profile at \(x = 990 \mathrm{km}\) after 5 Myr. (b) Temporal evolution lithospheric shear zones at \(y = 50 \mathrm{km}\) . (c) Temporal evolution of average lower upper mantle below \(300 \mathrm{km}\) depth. + +<|ref|>text<|/ref|><|det|>[[112, 333, 879, 457]]<|/det|> +Figure 3. Percentage of finite strain accumulated by diffusion creep (blue) or dislocation creep (white) after 20 Myr of divergence. Red: Contour of \(\eta = 10^{21.5} \mathrm{Pa} \cdot \mathrm{s}\) indicating thickness of the elastic lithosphere. (a) \(C_{\mathrm{OH}} = 50 \mathrm{H} / 10^{6} \mathrm{Si}\) . (b) \(C_{\mathrm{OH}} = 175 \mathrm{H} / 10^{6} \mathrm{Si}\) . (c) \(C_{\mathrm{OH}} = 600 \mathrm{H} / 10^{6} \mathrm{Si}\) . (d) \(C_{\mathrm{OH}} = 2500 \mathrm{H} / 10^{6} \mathrm{Si}\) . + +<|ref|>text<|/ref|><|det|>[[112, 506, 877, 666]]<|/det|> +Figure 4. Strength of the lithosphere. (a) Temporal evolution of the laterally averaged integrated strength (boundary force) of pure dislocation and grain- size- dependent composite diffusion- dislocation creep experiments. (b) Laterally \((x = 990 - 1000 \mathrm{km})\) averaged lithospheric strength profiles after 5 Myr. For color code see (a). (c) Strength and grain size profile along lithospheric shear zone at 5 Myr. Location of profile indicated in Fig. 1b. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 143, 70]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[42, 88, 950, 744]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 760, 115, 780]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 801, 950, 890]]<|/det|> +Temporal evolution of the experiment with \(\mathrm{COH} = 600 \mathrm{H} / 106 \mathrm{Si}\) . (a) Viscosity of upper mantle and marker composition of crust. White lines denote isotherms up to \(1300^{\circ} \mathrm{C}\) . (b) Deformation mechanism in the uppermost mantle and composition of crust. Red: Diffusion creep. White: Dislocation creep. Blue contours indicate grain size. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[42, 45, 955, 401]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 421, 117, 440]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[42, 464, 949, 530]]<|/det|> +Grain sizes in the mantle at variable water content. (a) Vertical profile at \(x = 990 \text{km}\) after 5 Myr. (b) Temporal evolution lithospheric shear zones at \(y = 50 \text{km}\) . (c) Temporal evolution of average lower upper mantle below \(300 \text{km}\) depth. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[42, 40, 390, 785]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 802, 117, 820]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 951, 909]]<|/det|> +Percentage of finite strain accumulated by diffusion creep (blue) or dislocation creep (white) after 20 Myr of divergence. Red: Contour of \(\eta = 1021.5\) Pa·s indicating thickness of the elastic lithosphere. (a) \(\mathrm{COH} = 50 \mathrm{H} / 106 \mathrm{Si}\) . (b) \(\mathrm{COH} = 175 \mathrm{H} / 106 \mathrm{Si}\) . (c) \(\mathrm{COH} = 600 \mathrm{H} / 106 \mathrm{Si}\) . (d) \(\mathrm{COH} = 2500 \mathrm{H} / 106 \mathrm{Si}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[42, 46, 960, 272]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[43, 290, 118, 309]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[40, 331, 958, 444]]<|/det|> +Strength of the lithosphere. (a) Temporal evolution of the laterally averaged integrated strength (boundary force) of pure dislocation and grain- size- dependent composite diffusion- dislocation creep experiments. (b) Laterally ( \(x = 990 - 1000 \text{km}\) ) averaged lithospheric strength profiles after 5 Myr. For color code see (a). (c) Strength and grain size profile along lithospheric shear zone at 5 Myr. Location of profile indicated in Fig. 1b. + +<|ref|>sub_title<|/ref|><|det|>[[44, 467, 311, 495]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 517, 764, 538]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 555, 320, 681]]<|/det|> +- RuhTokleBehrSuppMat.pdf- TableS1.xlsx- TableS2.xlsx- TableS3.xlsx- TableS4.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/images_list.json b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..2fe556eca9f6ddc10862f9f901a4e575de54f018 --- /dev/null +++ b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/images_list.json @@ -0,0 +1,182 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 From toroidal to supertoroidal light pulses: a,b, Isosurfaces for the electric fields of (a) the fundamental TLP \\(\\mathrm{Re}[E_{\\theta}(\\mathbf{r},t)]\\) , and (b) a STLP \\(\\mathrm{Re}[E_{\\theta}^{(\\alpha)}(\\mathbf{r},t)]\\) of \\(\\alpha = 5\\) , at amplitude levels of \\(E = \\pm 0.1\\) and the Rayleigh range of \\(q_{2} = 100q_{1}\\) , at different times of \\(t = 0\\) , \\(\\pm q_{2} / (4c)\\) , and \\(\\pm q_{2} / (2c)\\) . \\(x\\) - \\(z\\) cross-sections of the instantaneous electric field at \\(y = 0\\) . The insets in (a) and (b) are schematics of spatial topological structures of magnetic vector fields at focus \\((t = 0)\\) for the fundamental TLP and STLP, respectively. The gray dots and rings mark the distribution of singularities in magnetic field, large purple arrows mark the direction of magnetic field vector, and the smaller coloured arrows show the skyrmionic structures in magnetic field.", + "footnote": [], + "bbox": [ + [ + 82, + 60, + 920, + 365 + ] + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2 Electric field topology of toroidal and supertoroidal light pulses: a,b, The isoline plots of the electric field in the \\(x - z\\) plane for (a) the fundamental TLP, \\(\\mathrm{Re}[E_{\\theta}(\\mathbf{r},t = 0)]\\) , and (b) the STLP of \\(\\alpha = 5\\) , \\(\\mathrm{Re}[E_{\\theta}^{(\\alpha = 5)}(\\mathbf{r},t = 0)]\\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Dashed lines marked as a1-a3 and b1-b3 indicate the \\(z\\) -levels of the cross-sections on the right panels correspondingly. Panels a1-a3 and b1-b3 present isoline plots of the electric field and arrow plots of the electric field direction in the \\(x - y\\) plane, in logarithmic scale. Solid black lines and black dots mark the points and areas, where the electric field vanishes. Blue and red arrows indicate the two opposite azimuthal directions of the electric fields. Unit for coordinates: \\(q_{1}\\) .", + "footnote": [], + "bbox": [ + [ + 100, + 66, + 901, + 487 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3 Magnetic field topology of toroidal and supertoroidal light pulses: a,b, Isoline and arrow plots of the magnetic fields in the \\(x - z\\) plane for (a) the fundamental TLP and (b) the STLP of \\(\\alpha = 5\\) , in logarithmic scale. Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panels a1 and b1 present the zoom-ins of the regions highlighted by blue in (a) and (b) respectively. Panels a2 and b2 present isoline and arrow plots of the magnetic fields at \\(z = 0\\) planes, in logarithmic scale, marked by the black dashed lines in (a) and (b) respectively, where the magnetic fields vanish along the circular solid black lines with red arrows marking the styles of singularities (vortex for (a2) and saddle for (b2)). Skyrmionic structures in magnetic fields of toroidal and supertoroidal light pulses: c, Various textures of Néel-type skyrmionic structure observed at various transverse planes (see dashed purple lines in (a-b)) for the fundamental TLP (c1-c2) and the STLP of \\(\\alpha = 5\\) (c3-c6), which are demonstrated by the arrows with color-labeled longitudinal component value of magnetic field. The up-right insert of each panel shows the basic texture of the skyrmionic structure. Unit for coordinates: \\(q_{1}\\) .", + "footnote": [], + "bbox": [ + [ + 84, + 65, + 916, + 490 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4 Poynting vector topology of toroidal and supertoroidal light pulses: a,b, Contour and arrow plots of the Poynting vector fields in the \\(x - z\\) plane, in logarithmic scale, for (a) the fundamental TLP and (b) the STLP of \\(\\alpha = 5\\) . Panels (a1) and (b1) present zoom-ins of the areas highlighted by green in (a-b), respectively. Solid black lines and dots mark the zeros of the Poynting vector. Red and blue arrows indicate areas with forward and backward energy flow, respectively. Unit for coordinates: \\(q_1\\) .", + "footnote": [], + "bbox": [ + [ + 85, + 66, + 916, + 290 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5 Fractal-like pattern in electromagnetic field of supertoroidal pulses: a. The isoline plot of the electric field in the \\(x - z\\) plane for the STLP of \\(\\alpha = 20\\) , \\(\\mathrm{Re}[E_{\\theta}^{(\\alpha = 20)}(r,t = 0)]\\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Panel a1 presents the zoom-in of the region highlighted by blue in (a). b, Isoline and arrow plot of the logarithm of the magnetic field in the \\(x - z\\) plane for the STLP of \\(\\alpha = 20\\) . Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panel b1 presents the zoom-in of the region highlighted by blue in (b). Fine-scale features of skyrmionic structures: c1-c4. The skyrmionic distributions of magnetic field at several transverse planes marked by dashed lines c1-c4 in (b1). d1-d4, the distribution of normalized magnetic field and its absolute value versus \\(x\\) for the skyrmionic structures in (c1-c4). Insets illustrate the fine-scale features at the regions highlighted by gray bands. Unit for coordinates: \\(q_{1}\\) .", + "footnote": [], + "bbox": [ + [ + 84, + 60, + 920, + 545 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6 Evolution of on-axis singularity distribution versus supertoroidal order: a,b, The number (a) and positions (b) of on-axis saddle-singularities of the magnetic field of the STLP \\((q_{2} = 20q_{1})\\) within the range of \\(z\\in [0,80q_{1}]\\) , versus \\(\\alpha\\) . The blue dashed line marks where the number of singularities decreases.", + "footnote": [], + "bbox": [ + [ + 86, + 62, + 485, + 400 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 60, + 116, + 925, + 438 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 55, + 50, + 914, + 510 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 52, + 50, + 928, + 504 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 62, + 621, + 936, + 866 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 45, + 95, + 935, + 620 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 66, + 55, + 568, + 483 + ] + ], + "page_idx": 14 + } +] \ No newline at end of file diff --git a/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206.mmd b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206.mmd new file mode 100644 index 0000000000000000000000000000000000000000..583c96762b52090428ef584b201984002a186fa0 --- /dev/null +++ b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206.mmd @@ -0,0 +1,217 @@ + +# Supertoroidal light pulses: Propagating electromagnetic skyrmions in free space + +Yijie Shen ( \(\boxed{\times}\) y.shen@soton.ac.uk) + +University of Southampton + +Yaonan Hou + +University of Southampton + +Nikitas Papasimakis + +University of Southampton + +Nikolay Zheludev + +University of Southampton + +## Article + +Keywords: electrodynamics, optics, spectroscopy, light- matter interactions + +Posted Date: April 13th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 345437/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on October 8th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26037- w. + +<--- Page Split ---> + +# Supertoroidal light pulses: Propagating electromagnetic skyrmions in free space + +Yijie Shen \(^{1, *}\) , Yaonan Hou \(^{1}\) , Nikitas Papasimakis \(^{1}\) , and Nikolay I. Zheludev \(^{1,2}\) \(^{1}\) Optoelectronics Research Centre & Centre for Photonic Metamaterials, University of Southampton, Southampton SO17 1BJ, United Kingdom \(^{2}\) Centre for Disruptive Photonic Technologies, School of Physical and Mathematical Sciences and The Photonics Institute, Nanyang Technological University, Singapore 637378, Singapore (Dated: March 18, 2021) + +Topologically complex transient electromagnetic fields give access to nontrivial light- matter interactions and provide additional degrees of freedom for information transfer. An important example of such electromagnetic excitations are space- time non- separable single- cycle pulses of toroidal topology, the exact solutions of Maxwell described by Hellwarth and Nouchi in 1996 and recently observed experimentally. Here we introduce a new family of electromagnetic excitation, the supertoroidal electromagnetic pulses, in which the Hellwarth- Nouchi pulse is just the simplest member. The supertoroidal pulses exhibit skyrmionic structure of the electromagnetic fields, multiple singularities in the Poynting vector maps and fractal- like distributions of energy backflow. They are of interest for transient light- matter interactions, ultrafast optics, spectroscopy, and toroidal electrodynamics. + +Introduction - Topology of complex electromagnetic fields is attracting growing interest of the photonics and electromagnetics communities [1- 6], while topologically structured light fields find applications in super- resolution microscopy [7, 8], metrology [9, 10], and beyond [11- 13]. For example, the vortex beam with twisted phase, akin to a Mobius strips in phase domain, can carry orbital angular momentum with tuneable topological charges enabling advanced optical tweezers, machining, and communication applications [11- 13]. The complex electromagnetic strip and knots structures were also proposed as information carriers [14- 16]. Recently observed electromagnetic skyrmions are relevant to topological skyrmions quasiparticles in high- energy physics and condensed matter [8, 17- 20]. They have sophisticated vector topology [21- 23], and enable applications in super- resolution microscopy [8], ultrafast imaging [24], and give rise to new types of spin- orbit optical forces [25]. + +While a large body of work on topological properties of structured continuous light beams may be found in literatures, works on the topology of the time- dependent electromagnetic excitations and pulses only start to appear. For instance, the "Flying Doughnut" pulses, or toroidal light pulses (TLPs) first described in 1996 by Hellwarth and Nouchi [26], with unique spatiotemporal topology predicted recently [27], have only very recently observed experimentally [28]. + +Fuelled by a combination of advances in ultrafast lasers or THz emitters and in our ability to control the spatiotemporal structure of light [29, 30] together with the introduction of new experimental and theoretical pulse characterization methods [31, 32], TLPs are attracting growing attention. Indeed, TLPs exhibit their complex topological structure with vector singularities and interact with matter through coupling to toroidal and anapole localized modes [33- 35]. Generation of TLPs in the optical and THz parts of the spectrum now paves the way towards new forms of spectroscopy, sensitive to toroidal and anapole excitations, and new information transfer schemes. + +In this paper, we report that the Hellwarth and Nouchi pulses, are, in fact, the simplest example of an extended family of pulses that we will call supertoroidal light pulses (STLPs). We will show that supertoroidal light pulses introduced here exhibit complex topological structures that can be controlled by a single numerical parameter. The STLP display skyrmionlike arrangements of the transient electromagnetic fields organized in a fractal- like, self- affine manner, while the Poynting vector of the pulses feature singularities linked to the multiple energy backflow zones. + +## Results + +Supertoroidal electromagnetic pulses - Following Ziolkowski, localized finite- energy pulses can be obtained as superpositions of "electromagnetic directed- energy pulse trains" [36]. A special case of the localized finite- energy pulses was investigated by Hellwarth and Nouchi [26], who found the closed- form expression describing a single- cycle finite energy electromagnetic excitation with toroidal topology obtained from a scalar generating function \(f(\mathbf{r}, t)\) that satisfies the wave equation \((\nabla^2 - \frac{1}{c^2} \frac{\partial^2}{\partial t^2}) f(\mathbf{r}, t) = 0\) , where \(\mathbf{r} = (r, \theta , z)\) are cylindrical coordinates, \(t\) is time, \(c = 1 / \sqrt{\epsilon_0 \mu_0}\) is the speed of light, and the \(\epsilon_0\) and \(\mu_0\) are the permittivity and permeability of medium. Then, the exact solution of \(f(\mathbf{r}, t)\) can be given by the modified power spectrum method [26, 36], as \(f(\mathbf{r}, t) = f_0 / [(q_1 + i\tau)(s + q_2)^{\alpha}]\) , where \(f_0\) is a normalizing constant, \(s = r^2 /(q_1 + i\tau) - i\sigma\) , \(\tau = z - ct\) , \(\sigma = z + ct\) , \(q_1\) and \(q_2\) are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while \(\alpha\) is a real dimensionless parameter that must satisfy \(\alpha \geq 1\) to ensure finite energy solutions. Next, transverse electric (TE) and transverse magnetic (TM) solutions are readily obtained by using Hertz potentials. The electromagnetic fields for the TE solution can be derived by the potential \(\mathbf{A}(\mathbf{r}, t) = \mu_0 \nabla \times \hat{\mathbf{z}} f(\mathbf{r}, t)\) as \(\mathbf{E}(\mathbf{r}, t) = -\mu_0 \frac{\partial}{\partial t} \nabla \times \mathbf{A}\) and \(\mathbf{H}(\mathbf{r}, t) = \nabla \times (\nabla \times \mathbf{A})\) [26, 36]. Finally assuming \(\alpha = 1\) , the electromagnetic fields of the TLP are described by [26]: + +\[\begin{array}{l}{E_{\theta} = -4i f_{0}\sqrt{\frac{\mu_{0}}{\epsilon_{0}}\frac{r(q_{1} + q_{2} - 2i c t)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}}}\\ {H_{r} = 4i f_{0}\frac{r(q_{2} - q_{1} - 2i z)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}}\\ {H_{z} = -4f_{0}\frac{r^{2} - (q_{1} + i\tau)(q_{2} - i\sigma)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}} \end{array} \quad (1)\] + +where the electric field \(E_{\theta}\) is azimuthally polarized with no longitudinal or radial components, whereas the magnetic field is oriented along the radial and longitudinal directions, \(H_{r}\) and \(H_{z}\) , with no azimuthal component. Equations (1- 3) derived by Hellwarth and Nouchi for \(\alpha = 1\) show the simplest example of TLPs. Here we explore the general solution for values of + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1 From toroidal to supertoroidal light pulses: a,b, Isosurfaces for the electric fields of (a) the fundamental TLP \(\mathrm{Re}[E_{\theta}(\mathbf{r},t)]\) , and (b) a STLP \(\mathrm{Re}[E_{\theta}^{(\alpha)}(\mathbf{r},t)]\) of \(\alpha = 5\) , at amplitude levels of \(E = \pm 0.1\) and the Rayleigh range of \(q_{2} = 100q_{1}\) , at different times of \(t = 0\) , \(\pm q_{2} / (4c)\) , and \(\pm q_{2} / (2c)\) . \(x\) - \(z\) cross-sections of the instantaneous electric field at \(y = 0\) . The insets in (a) and (b) are schematics of spatial topological structures of magnetic vector fields at focus \((t = 0)\) for the fundamental TLP and STLP, respectively. The gray dots and rings mark the distribution of singularities in magnetic field, large purple arrows mark the direction of magnetic field vector, and the smaller coloured arrows show the skyrmionic structures in magnetic field.
+ +\(\alpha \geq 1\) . In the TE case, electric and magnetic fields are given by (see detailed derivation in Supplementary Information): + +\[\begin{array}{l}{{E_{\theta}^{(\alpha)}=-2\alpha if_{0}\sqrt{\frac{\mu_{0}}{\epsilon_{0}}\left\{\frac{(\alpha+1)r(q_{1}+i\tau)^{\alpha-1}(q_{1}+q_{2}-2i c t)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}-\frac{(\alpha-1)r(q_{1}+i\tau)^{\alpha-2}}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\\ {{H_{r}^{(\alpha)}=2\alpha if_{0}\left\{\frac{(\alpha+1)r(q_{1}+i\tau)^{\alpha-1}(q_{2}-q_{1}-2i z)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}-\frac{(\alpha-1)r(q_{1}+i\tau)^{\alpha-2}}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\\ {{H_{z}^{(\alpha)}=-4\alpha f_{0}\left\{\frac{(q_{1}+i\tau)^{\alpha-1}[r^{2}-\alpha(q_{1}+i\tau)(q_{2}-i\sigma)]}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}+\frac{(\alpha-1)(q_{1}+i\tau)^{\alpha-2}(q_{2}-i\sigma)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\end{array} \quad (4)\] + +For \(\alpha = 1\) , the electromagnetic fields in Eqs. (4- 6) are reduced to that of the fundamental TLP Eqs. (1- 3). Moreover, the real and imaginary parts of equations (4- 6), simultaneity fulfill Maxwell equations and therefore represent real electromagnetic pulses (see Supplementary Materials). + +While propagating in free space toroidal and supertoroidal pulses exhibit self- focusing. Figure 1 shows the evolution of the fundamental \((\alpha = 1)\) TLP and STLP \((\alpha = 5)\) upon propagation through the focal point. In the former case, the pulse is single- cycle at focus \((z = 0)\) becoming \(1\frac{1}{2}\) - cycle at the boundaries of Rayleigh range \(z = \pm q_{2} / 2\) (Fig. 1a). On the other hand, STLPs with \(\alpha >1\) (Fig. 1b, \(\alpha = 5\) ) exhibit a substantially more complex spatiotemporal evolution where the pulse is being reshaped multiple times upon propagation + +(See Video 1 and Video 2 in Supplementary Materials for the dynamic evolution of the fundamental TLP and STLPs). + +Electric field singularities - Figure 2 comparatively shows the instantaneous electric fields for the TE single- cycle fundamental TLP and STLP \((\alpha = 5)\) with \(q_{2} = 20q_{1}\) at the focus \((t = 0)\) . In all cases, the electric field vanishes on the \(z\) - axis \((r = 0)\) ; see the vertical solid black lines in Figs. 2a and 2b) owing to the azimuthal polarization and also in the \(z = 0\) plane (see the horizontal solid black lines in Figs. 2a and 2b) due to the odd symmetry of Eqs. Eqs. (1,4) with respect to \(z\) . For the fundamental TLP, the electric field vanishes on two spherical shells (indicated by the solid circles in Fig. 2a) on the positive and negative \(z\) - axis, respectively. This behavior can be more clearly observed in the transverse distributions + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2 Electric field topology of toroidal and supertoroidal light pulses: a,b, The isoline plots of the electric field in the \(x - z\) plane for (a) the fundamental TLP, \(\mathrm{Re}[E_{\theta}(\mathbf{r},t = 0)]\) , and (b) the STLP of \(\alpha = 5\) , \(\mathrm{Re}[E_{\theta}^{(\alpha = 5)}(\mathbf{r},t = 0)]\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Dashed lines marked as a1-a3 and b1-b3 indicate the \(z\) -levels of the cross-sections on the right panels correspondingly. Panels a1-a3 and b1-b3 present isoline plots of the electric field and arrow plots of the electric field direction in the \(x - y\) plane, in logarithmic scale. Solid black lines and black dots mark the points and areas, where the electric field vanishes. Blue and red arrows indicate the two opposite azimuthal directions of the electric fields. Unit for coordinates: \(q_{1}\) .
+ +at three different propagation distances, \(z = q_{1},5q_{1},35q_{1}\) of Figs. 2a1- 2a3. In accordance to Figs. 2a1 and 2a3, the electric field at \(z\) positions close to and away from the center of the pulse (cross- sections a1 and a3) rotates counter- clockwise forming a vortex around the center singularity along the propagation axis. However, at a distance of \(z = 5q_{1}\) from its center (Figs. 2a2), the electric field vanishes on a circular boundary, which corresponds to a spherical region inside which the electric field is oriented along the clockwise direction, whereas outside this region the electric field remains oriented in the opposite (counter- clockwise) direction (see cross- section a3). For the STLP case, a more complex matryoshka- like structure emerges with multiple nested singularity shells, replacing the single shell of the fundamental TLP, as Fig. 2b shows. The electric field configuration close the singularity shells can be examined in detail at transverse planes at \(z = q_{1},5q_{1},35q_{1}\) (Figs. 2b1- b3). In this case, at transverse planes close to \(z = 0\) , the electric field changes orientation from counter- clockwise close to \(r = 0\) to clockwise away from the \(z\) - axis (see Fig. 2b1). On the other hand, on transverse planes close to \(z = 5q_{1}\) (see Fig. 2b2), two singular rings (corresponding to the two singular shells) emerge as a cross- sections of the multi- layer singularity shell structure, separating space in three different regions, in which the electric field direction alternates between counter- clockwise \((r / q_{1}< 7)\) , to clockwise \((7< r / q_{1}< 15)\) , and again to counter- clockwise \((r / q_{1} > 15)\) . + +In general, the pulse of higher order of \(\alpha\) is accompanied by a more complex multi- layer singular- shell structure, see the dynamic evolution versus the order index in Video 3 in Supplementary Materials. Although the above results of electric fields are instantaneous at \(t = 0\) , we note that the multi layer shall structure propagation of supertoroidal light pulse is retained during propagation, see such dynamic process in Video 4 in Supplementary Materials. + +Magnetic field singularities – The magnetic field of STLPs has both radial and longitudinal components, \(\mathbf{H} = H_{r}\hat{\mathbf{r}} + H_{z}\hat{\mathbf{z}}\) , which lead to a topological structure more complex than the one exhibited by the electric field. Figure 3 comparatively shows the instantaneous magnetic fields for the TLP and the STLP of \(\alpha = 5\) . For the fundamental TLP (Fig. 3a), the magnetic field has ten different vector singularities on the \(x\) - \(z\) plane, including four saddle points [the longitudinal field component pointing towards (away from) and the radial com + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3 Magnetic field topology of toroidal and supertoroidal light pulses: a,b, Isoline and arrow plots of the magnetic fields in the \(x - z\) plane for (a) the fundamental TLP and (b) the STLP of \(\alpha = 5\) , in logarithmic scale. Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panels a1 and b1 present the zoom-ins of the regions highlighted by blue in (a) and (b) respectively. Panels a2 and b2 present isoline and arrow plots of the magnetic fields at \(z = 0\) planes, in logarithmic scale, marked by the black dashed lines in (a) and (b) respectively, where the magnetic fields vanish along the circular solid black lines with red arrows marking the styles of singularities (vortex for (a2) and saddle for (b2)). Skyrmionic structures in magnetic fields of toroidal and supertoroidal light pulses: c, Various textures of Néel-type skyrmionic structure observed at various transverse planes (see dashed purple lines in (a-b)) for the fundamental TLP (c1-c2) and the STLP of \(\alpha = 5\) (c3-c6), which are demonstrated by the arrows with color-labeled longitudinal component value of magnetic field. The up-right insert of each panel shows the basic texture of the skyrmionic structure. Unit for coordinates: \(q_{1}\) .
+ +ponent away from (toward) the singularity) on \(z\) - axis and six vortex rings [the surrounding vector distribution forming a vortex loop] away from the \(z\) - axis. We note that we only consider the singularities existed at an area containing \(99.9\%\) of the energy of the pulse. While the singularity existed at the region far away from the pulse center with nearly zero energy can be neglected. A zoom- in of the field structure around these singularities can be seen in Fig. 3a1. For the three off- axis singularities located at the \(x > 0\) half space, two of them (at \(z > 0\) and \(z < 0\) ) are accompanied by counter- clockwise rotating vortices, whereas the third one (at \(z = 0\) ) by a clockwise rotating vortex. Owing to the cylindrical symmetry of the pulse, the off- axis singularities correspond in essence to singularity rings. Such an example is shown in Fig. 3a2, which presents the magnetic field on the transverse plane at \(z = 0\) . Here, the magnetic field points toward the positive (negative) \(z\) - axis inside (outside) the circular region resulting in the formation of a toroidal vortex winding around the singularity ring. For the STLP (Fig. 3b), more vector singularities are unveiled in the magnetic field with six saddle points on \(z\) - axis and six off- axis singularities. A zoom- in of the field structure around these singularities can be seen in Fig. 3b1. The orientation of the magnetic field around the on- axis saddle points is alternating between "longitudinal- toward radial- outward point" and "axial- toward longitudinal- outward", similarly to the on- axis singularities of the TLP. Moreover, the off- axis singularities at \(z = 0\) become now saddle points contributing to the singularity ring in the \(z = 0\) plane. The remaining off- axis singularities are accompanied by clockwise and counterclockwise magnetic field configurations at \(x > 0\) and \(x < 0\) , respectively as shown Fig. 3b2. + +Skyrmionic structure in magnetic field – A topological feature of particular interest here is the skyrmionic structure observed in the magnetic field configuration of STLPs. The skyrmion is a topologically protected quasiparticle in condensed matter with a hedgehog- like vectorial field, that gradually changes orientation as one moves away from the + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4 Poynting vector topology of toroidal and supertoroidal light pulses: a,b, Contour and arrow plots of the Poynting vector fields in the \(x - z\) plane, in logarithmic scale, for (a) the fundamental TLP and (b) the STLP of \(\alpha = 5\) . Panels (a1) and (b1) present zoom-ins of the areas highlighted by green in (a-b), respectively. Solid black lines and dots mark the zeros of the Poynting vector. Red and blue arrows indicate areas with forward and backward energy flow, respectively. Unit for coordinates: \(q_1\) .
+ +skyrmion centre [21- 23]. Recently skyrmion- like configurations have been reported in electromagnetism, including skyrmion modes in surface plasmon polaritons [17] and the spin field of focused beams [8, 19]. Here we observe the skyrmion field configurations in the magnetic field of propagating STLPs. + +The topological properties of a skyrmionic configuration can be characterized by the skyrmion number \(s\) , which can be separated into a polarity \(p\) and vorticity number \(m\) [23]. The polarity represents the direction of the vector field, down (up) at \(r = 0\) and up (down) at \(r \to \infty\) for \(p = 1\) ( \(p = - 1\) ), the vorticity controls the distribution of the transverse field components, and another initial phase \(\gamma\) should be added for determining the helical vector distribution, see Methods for details. For the \(m = 1\) skyrmion, the cases of \(\gamma = 0\) and \(\gamma = \pi\) are classified as Neel- type, and the cases of \(\gamma = \pm \pi /2\) are classified as Bloch- type. The case for \(m = - 1\) is classified as anti- skyrmion. + +Here the vector forming skyrmionic structure is defined by the normalized magnetic field \(\mathbf{H} = \mathbf{H} / |\mathbf{H}|\) of the STLP. Two examples of two skyrmionic structures in the fundamental TLP are shown in Figs. 3c1 ( \(p = m = 1\) , \(\gamma = \pi\) ) and 3c2 ( \(p = m = 1\) , \(\gamma = 0\) ) occurs at the two transverse planes marked by purple dashed lines c1 and c2, which are both Neel- type skyrmionic structures, where the vector changes its direction from "down" at the centre to "up" away from the centre. In the case of the STLPs with more complex topology, it is possible to observe more skyrmionic structures. The STLP pulse ( \(\alpha = 5\) ) exhibits not only the clockwise ( \(p = m = 1\) , \(\gamma = \pi\) ) and counter- clockwise ( \(p = m = 1\) , \(\gamma = 0\) ) Neel- type skyrmionic structures (Fig. 3c3 and 3c4), but also those with \(p = - 1\) , \(m = 1\) , \(\gamma = \pi\) and \(p = - 1\) , \(m = 1\) , \(\gamma = 0\) , in Fig. 3c6- c6. + +In general, as the value of \(\alpha\) increases, toroidal pulses show an increasingly complex magnetic field pattern with skyrmionic structures of multiple types, see Video 3 in Supplementary Materials. We also note that the topology of the STLP is maintained during propagation, see Video 4 in Supplementary Materials. + +Energy backflow and Poynting vector singularities - The singularities of the electric and magnetic fields are linked to the complex topological behavior for the energy flow as represented by the Poynting vector \(\mathbf{S} = \mathbf{E} \times \mathbf{H}\) . An interesting effect for the fundamental TLP is the presence of energy backflow: the Poynting vector at certain regions is oriented against the prorogation direction (blue arrows in Fig. 4a) [27]. Such energy backflow effects have been predicted and discussed in the context of singular superpositions of waves [1, 37], superoscillatory light fields [9, 38], and plasmonic nanostructures [39]. The Poynting vector map reveals a complex multi- layer energy backflow structure, as shown in Fig. 4b. The energy flow vanishes at the positions of the electric and magnetic singularities and inherits their multi- layer matryoshka- like structure. Poynting vector vanishes at \(z = 0\) plane, along the \(z\) - axis, and on the dual- layer matryoshka- like singular shells (marked by the black bold lines in Fig. 4b). Importantly, energy backflow occurs at areas of relatively low energy density, and, hence, STLP as a whole still propagates forward. For the temporal evolution of the energy flow of the pulse see Videos 3 & 4 in Supplementary Materials. + +Fractal patterns hidden in electromagnetic fields - As the order \(\alpha\) of the pulse increases (see Video 3 in Supplementary Material), the topological features of the STLP appear to be organized in a hierarchical, fractal- like fashion. A characteristic case of the STLP of \(\alpha = 20\) is presented in (Fig. 5). For the electric field, the matryoshka- like singular shells involve an increasing number of layers as one examines the pulse at finer length scales, forming a self- similar pattern that seems infinitely repeated. For the magnetic field, the saddle and vortex points are distributed along the propagation axis and in two planes crossing the pulse centre, respectively. The distribution of singularities becomes increasingly dense as one approaches the centre of the pulse, resulting in a self- similar pattern. A similar pattern can be seen for the Poynting vector map (see Videos 3 & 4 in Supplementary Materials). + +Fine- scale features of skyrmionic structures - The fractal- like pattern of vectorial magnetic field of a high- order STLPs + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5 Fractal-like pattern in electromagnetic field of supertoroidal pulses: a. The isoline plot of the electric field in the \(x - z\) plane for the STLP of \(\alpha = 20\) , \(\mathrm{Re}[E_{\theta}^{(\alpha = 20)}(r,t = 0)]\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Panel a1 presents the zoom-in of the region highlighted by blue in (a). b, Isoline and arrow plot of the logarithm of the magnetic field in the \(x - z\) plane for the STLP of \(\alpha = 20\) . Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panel b1 presents the zoom-in of the region highlighted by blue in (b). Fine-scale features of skyrmionic structures: c1-c4. The skyrmionic distributions of magnetic field at several transverse planes marked by dashed lines c1-c4 in (b1). d1-d4, the distribution of normalized magnetic field and its absolute value versus \(x\) for the skyrmionic structures in (c1-c4). Insets illustrate the fine-scale features at the regions highlighted by gray bands. Unit for coordinates: \(q_{1}\) .
+ +results in skyrmionic configurations with with features changing much faster than the effective wavelength \(q_{1}\) . Figures 5c1- c4 show the four skyrmionic structures of the high- order STLP \((\alpha = 20)\) at the four transverse planes marked by the dashed lines in Fig. 5b1 at positions of \(z / q_{1} = 14,10,6,3.5\) correspondingly. The four skyrmionic structures have two different topologies with topological numbers of \((p,m,\gamma) = (1,1,\pi)\) , for c1 and c3, and \((- 1,1,0)\) , for c2 and c4. In addition, they exhibit an effect of "spin reversal", where the number reversals is given by \(\bar{p} = \frac{1}{2\pi} [\beta (r)]_{r = 0}^{r = \infty}\) , e.g. \(\bar{p} = 1,2,3,4\) for the skyrmionic structures in Figs. 5c1- c4, respectively. Each reversal corresponds to a sign change of \(H_{z}\) , which takes place over areas much smaller than the effective wavelength of the pulse \((q_{1})\) . The full width at half maximum of these areas for the four skyrmionic structures is 1/6, 1/10, 1/30, 1/50 of the effective wavelength, respectively. Conclusively, the sign reversals become increasingly rapid in transverse planes closer to the pulse center \((z = 0)\) , see Figs. 5d1- d4. Similarly, increasing the value of \(\alpha\) leads to increasingly sharper singularities. Notably, in contrast to the fundamental TLP, the skyrmionic configurations in STLP occur at areas of higher energy density, and thus we expect that they could be observed experimentally. + +The topological structure of the STLP is directly related to the distribution of on- axis saddle- points in its magnetic field. Indeed, the latter mark the intersection of the \(E\) - field singular shells with the \(z\) - axis, which in turn results in the emergence of different skyrmionic magnetic field patterns (see Fig. 5 and Videos 3 & 4). The number and position of on- axis magnetic field saddle points is defined by the supertoroidal parameter \(\alpha\) . This is illustrated in Fig. 6a, where we plot the number of on- axis \(H\) - field singularities as a function of alpha for a STLP + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6 Evolution of on-axis singularity distribution versus supertoroidal order: a,b, The number (a) and positions (b) of on-axis saddle-singularities of the magnetic field of the STLP \((q_{2} = 20q_{1})\) within the range of \(z\in [0,80q_{1}]\) , versus \(\alpha\) . The blue dashed line marks where the number of singularities decreases.
+ +with \(q_{2} = 20q_{1}\) . The number of singularities is generally increasing with increasing \(\alpha\) apart for values around \(\alpha = 5.6\) (marked by blue dashed line in Fig. 6). Moreover, the number of singularities increases in a ladder- like fashion, where only specific values of alpha lead to additional singularities. The origin of this behaviour can be traced to changes in the pulse structure as \(\alpha\) increases (see Fig. 6b). For specific values of \(\alpha\) , additional singularities appear away from the pulse center \((z = 0)\) and then move slowly towards it. On the other hand, the irregular behavior at \(\alpha = 5.6\) is a result of two singularities disappearing (see blue dashed line in Fig. 6b). + +## Discussion + +STLPs exhibit complex and unique topological structure. The electric field exhibits a matryoshka- like configuration of singularity shells, which divide the STLP into "nested" regions with opposite azimuthal polarization. The magnetic field exhibits multi- texture skyrmionic structures at various transverse planes of a single pulse, related to the distribution of multiple saddle and vortex singularities. The instantaneous Poynting vector field exhibits multiple singularities with regions of energy backflow. The singularities of the STLP appear to be hierarchically organized resulting in self- similar, fractal- like patterns for higher- order pulses. + +In conclusion, to the best of our knowledge, STLPs are so far the only known example of free- space propagating skyrmionic field configurations. Their structure contains sharp singularities of interest for super- resolution metrology and microscopy, optical information encoding, optical trapping and particle acceleration. + +## Methods + +Solving the supertoroidal pulses - The first step is to solve the scalar generating function \(f(\mathbf{r},t)\) that satisfies the wave equation \((\nabla^{2} - \frac{1}{c^{2}}\frac{\partial^{2}}{\partial t^{2}})f(\mathbf{r},t) = 0\) , where \(\mathbf{r} = (r,\theta ,z)\) are cylindrical coordinates, \(t\) is time, \(c = 1 / \sqrt{\epsilon_{0}\mu_{0}}\) is the speed of light, and the \(\epsilon_{0}\) and \(\mu_{0}\) are the permittivity and permeability of medium. The exact solution of \(f(\mathbf{r},t)\) can be given by the modified power spectrum method as \(f(\mathbf{r},t) = f_{0} / [(q_{1} + i\tau)(s + q_{2})^{\alpha}]\) , where \(f_{0}\) is a normalizing constant, \(s = r^{2} / (q_{1} + i\tau) - i\sigma\) , \(\tau = z - ct\) , \(\sigma = z + ct\) , \(q_{1}\) and \(q_{2}\) are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while \(\alpha\) is a real dimensionless parameter that must satisfy \(\alpha \geq 1\) to ensure finite energy solutions. The next step is constructing the Hertz potential. For fulfilling the toroidal symmetric and azimuthally polarized structure, the Hertz potential should be constructed as \(\mathbf{A}(\mathbf{r},t) = \mu_{0}\nabla \times \hat{\mathbf{z}} f(\mathbf{r},t)\) . Then, the exact solutions of solutions of transverse electric (TE) and transverse magnetic (TM) modes are readily obtained by using Hertz potential. The electromagnetic fields for the TE solution can be derived by the potential as \(\mathbf{E}(\mathbf{r},t) = -\mu_{0}\frac{\partial}{\partial t}\nabla \times \mathbf{A}\) and \(\mathbf{H}(\mathbf{r},t) = \nabla \times (\nabla \times \mathbf{A})\) [26, 36], see Supplementary Information for more detailed derivations. + +Characterizing topology of skyrmion - The topological properties of a skyrmionic configuration can be characterized by the skyrmion number defined by [23]: + +\[s = \frac{1}{4\pi}\iint \mathbf{n}\cdot \left(\frac{\partial\mathbf{n}}{\partial x}\times \frac{\partial\mathbf{n}}{\partial y}\right)\mathrm{d}x\mathrm{d}y \quad (7)\] + +that is an integer counting how many times the vector \(\mathbf{n}(x,y) = \mathbf{n}(r\cos \theta ,r\sin \theta)\) wraps around the unit sphere. For mapping to the unit sphere, the vector can be given by \(\mathbf{n} = (\cos \alpha (\theta)\sin \beta (r),\sin \alpha (\theta)\sin \beta (r),\cos \beta (r))\) . Also, The skyrmion number can be separated into two integers: + +\[\begin{array}{l}{s = \frac{1}{4\pi}\int_0^\infty \mathrm{d}r\int_0^{2\pi}\mathrm{d}\phi \frac{\mathrm{d}\beta(r)}{\mathrm{d}r}\frac{\mathrm{d}\alpha(\theta)}{\mathrm{d}\theta}\sin \beta (r)}\\ {= \frac{1}{4\pi} [\cos \beta (r)]_{r = 0}^{r = \infty}[\alpha (\theta)]_{\theta = 0}^{\theta = 2\pi} = p\cdot m} \end{array} \quad (8)\] + +the polarity, \(p = \frac{1}{2} [\cos \beta (r)]_{r = 0}^{r = \infty}\) , represents the direction of the vector field, down (up) at \(r = 0\) and up (down) at \(r\rightarrow \infty\) for \(p = 1\) \((p = - 1)\) . The vorticity number, \(m = \frac{1}{2\pi} [\alpha (\theta)]_{\theta = 0}^{\theta = 2\pi}\) , controls the distribution of the transverse field components. In the case of a helical distribution, an initial phase \(\gamma\) should be added, \(\alpha (\theta) = m\theta +\gamma\) . For the \(m = 1\) skyrmion, the cases of \(\gamma = 0\) and \(\gamma = \pi\) are classified as Neel- type, and the cases of \(\gamma = \pm \pi /2\) are classified as Bloch- type. The case for \(m = - 1\) is classified as anti- skyrmion. + +## Acknowledgements + +The authors acknowledge the supports of the MOE Singapore (MOE2016- T3- 1- 006), the UKs Engineering and Physical Sciences Research Council (grant EP/M009122/1, Funder Id: http://dx.doi.org/10.13039/501100000266), the European Research Council (Advanced grant FLEET- 786851, Funder Id: http://dx.doi.org/10.13039/501100000781), and the Defense Advanced Research Projects Agency (DARPA) under the Nascent Light Matter Interactions program. + +<--- Page Split ---> + +Data availability The data from this paper can be obtained from the University of Southampton ePrints research repository https://doi.org/xx.xxxx/SOTON/xxxxxx. + +\\*y.shen@soton.ac.uk + +[1] Michael Berry, "Making waves in physics," Nature 403, 21- 21 (2000). [2] Marat Soskin, Svetlana V Boriskina, Yidong Chong, Mark R Dennis, and Anton Desyatnikov, "Singular optics and topological photonics," Journal of Optics 19, 010401 (2016). [3] Ling Lu, John D Joannopoulos, and Marin Soljačić, "Topological photonics," Nature Photonics 8, 821- 829 (2014). [4] Alexander B Khanikaev and Gennady Shvets, "Two- dimensional topological photonics," Nature Photonics 11, 763- 773 (2017). [5] Tomoki Ozawa, Hannah M Price, Alberto Amo, Nathan Goldman, Mohammad Hafezi, Ling Lu, Mikael C Rechtsman, David Schuster, Jonathan Simon, Oded Zilberberg, et al., "Topological photonics," Reviews of Modern Physics 91, 015006 (2019). [6] N Asger Mortensen, Sergey I Bozhevolnyi, and Andrea Alu, "Topological nanophotonics," Nanophotonics 8, 1315- 1317 (2019). [7] Tanchao Pu, Jun- Yu Ou, Vassili Savinov, Guanghui Yuan, Nikitas Papasimakis, and Nikolay I Zheludev, "Unlabeled far- field deeply subwavelength topological microscopy (dstm)," Advanced Science 8, 2002886 (2021). [8] Luping Du, Aiping Yang, Anatoly V Zayats, and Xiaocong Yuan, "Deep- subwavelength features of photonic skyrmions in a confined electromagnetic field with orbital angular momentum," Nature Physics 15, 650- 654 (2019). [9] Guang Hui Yuan and Nikolay I Zheludev, "Detecting nanometric displacements with optical ruler metrology," Science 364, 771- 775 (2019). [10] Shawn Sederberg, Fanqi Kong, Felix Hufnagel, Chunmei Zhang, Ebrahim Karimi, and Paul B Corkum, "Vectorized optoelectronic control and metrology in a semiconductor," Nature Photonics 14, 680- 685 (2020). [11] Yijie Shen, Xuejiao Wang, Zhenwei Xie, Changjun Min, Xing Fu, Qiang Liu, Mali Gong, and Xiaocong Yuan, "Optical vortices 30 years on: Oam manipulation from topological charge to multiple singularities," Light: Science & Applications 8, 1- 29 (2019). [12] Alison M Yao and Miles J Padgett, "Orbital angular momentum: origins, behavior and applications," Advances in Optics and Photonics 3, 161- 204 (2011). [13] Rui Chen, Hong Zhou, Marco Moretti, Xiaodong Wang, and Jiandong Li, "Orbital angular momentum waves: Generation, detection and emerging applications," IEEE Communications Surveys & Tutorials (2019). [14] Thomas Bauer, Peter Banzer, Ebrahim Karimi, Sergej Orlov, Andrea Rubano, Lorenzo Marrucci, Enrico Santamato, Robert W Boyd, and Gerd Leuchs, "Observation of optical polarization mubius strips," Science 347, 964- 966 (2015). [15] Thomas Bauer, Martin Neugebauer, Gerd Leuchs, and Peter Banzer, "Optical polarization mubius strips and points of purely transverse spin density," Physical Review Letters 117, 013601 (2016). [16] Hugo Larocque, Alessio D'Errico, Manuel F Ferrer- Garcia, Avishy Carmi, Eliahu Cohen, and Ebrahim Karimi, "Optical framed knots as information carriers," Nature Communications 11, 1- 8 (2020). [17] S Tsesses, E Ostrovsky, K Cohen, B Gjonaj, NH Lindner, and G Bartal, "Optical skyrmion lattice in evanescent electromagnetic fields," Science 361, 993- 996 (2018). [18] Yanan Dai, Zhikang Zhou, Atreyie Ghosh, Roger SK Mong, Atsushi Kubo, Chen- Bin Huang, and Hrvoje Petek, "Plasmonic + +topological quasiparticle on the nanometre and femtosecond scales," Nature 588, 616- 619 (2020). [19] Sijia Gao, Fiona C Speirits, Francesco Castellucci, Sonja Franke- Arnold, Stephen M Barnett, and Jörg B Götte, "Paraxial skyrmionic beams," Physical Review A 102, 053513 (2020). [20] Rodrigo Gutiérrez Cuevas and Emilio Pisanty, "Optical polarization skyrmionic fields in free space," Journal of Optics (2021). [21] Albert Fert, Nicolas Reyren, and Vincent Cros, "Magnetic skyrmions: advances in physics and potential applications," Nature Reviews Materials 2, 1- 15 (2017). [22] XZ Yu, Yoshinori Onose, Naoya Kanazawa, JH Park, JH Han, Yoshio Matsui, Naoto Nagaosa, and Yoshinori Tokura, "Real- space observation of a two- dimensional skyrmion crystal," Nature 465, 901- 904 (2010). [23] Naoto Nagaosa and Yoshinori Tokura, "Topological properties and dynamics of magnetic skyrmions," Nature Nanotechnology 8, 899- 911 (2013). [24] Timothy J Davis, David Janoschka, Pascal Dreher, Bettina Frank, Frank- J Meyer zu Heringdorf, and Harald Giessen, "Ultrafast vector imaging of plasmonic skyrmion dynamics with deep subwavelength resolution," Science 368 (2020). [25] Chris Halcrow and Derek Harland, "Attractive spin- orbit potential from the skyrme model," Physical Review Letters 125, 042501 (2020). [26] RW Hellwarth and P Nouchi, "Focused one- cycle electromagnetic pulses," Physical Review E 54, 889 (1996). [27] Apostolos Zdagkas, Nikitas Papasimakis, Vassili Savinov, Mark R Dennis, and Nikolay I Zheludev, "Singularities in the flying electromagnetic doughnuts," Nanophotonics 8, 1379- 1385 (2019). [28] A Zdagkas, Y Shen, C McDonnell, J Deng, G Li, T Ellenbogen, N Papasimakis, and NI Zheludev, "Observation of toroidal pulses of light," arXiv preprint arXiv:2102.03636 (2021). [29] Nikitas Papasimakis, Tim Raybould, Vassili A Fedotov, Din Ping Tsai, Ian Youngs, and Nikolay I Zheludev, "Pulse generation scheme for flying electromagnetic doughnuts," Physical Review B 97, 201409 (2018). [30] Cormac McDonnell, Junhong Deng, Symeon Sideris, Tal Ellenbogen, and Guixin Li, "Functional thz emitters based on pancharatnam- berry phase nonlinear metasurfaces," Nature Communications 12, 1- 8 (2021). [31] Apostolos Zdagkas, Venkatram Nalla, Nikitas Papasimakis, and Nikolay I Zheludev, "Spatio- temporal characterization of ultrashort vector pulses," arXiv preprint arXiv:2101.11651 (2021). [32] Yijie Shen, Apostolos Zdagkas, Nikitas Papasimakis, and Nikolay I Zheludev, "Measures of space- time nonseparability of electromagnetic pulses," Physical Review Research 3, 013236 (2021). [33] Nikitas Papasimakis, VA Fedotov, Vassili Savinov, TA Raybould, and NI Zheludev, "Electromagnetic toroidal excitations in matter and free space," Nature Materials 15, 263- 271 (2016). [34] Tim Raybould, Vassili A Fedotov, Nikitas Papasimakis, Ian Youngs, and Nikolay I Zheludev, "Exciting dynamic anapoles with electromagnetic doughnut pulses," Applied Physics Letters 111, 081104 (2017). [35] Vassili Savinov, Nikitas Papasimakis, DP Tsai, and NI Zheludev, "Optical anapoles," Communications Physics 2, 1- 4 (2019). [36] Richard W Ziolkowski, "Localized transmission of electromagnetic energy," Physical Review A 39, 2005 (1989). [37] MV Berry, "Quantum backflow, negative kinetic energy, and optical retro- propagation," Journal of Physics A: Mathematical and Theoretical 43, 415302 (2010). [38] Guanghui Yuan, Edward TF Rogers, and Nikolay I Zheludev, "plasmonics" in free space: observation of giant wavevectors, vortices, and energy backflow in superoscillatory optical fields," Light: Science & Applications 8, 1- 9 (2019). [39] MV Bashevoy, VA Fedotov, and NI Zheludev, "Optical whirlpool on an absorbing metallic nanoparticle," Optics Ex + +<--- Page Split ---> + +press 13, 8372–8379 (2005). + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +From toroidal to supertoroidal light pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Electric field topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Magnetic field topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +![](images/Figure_4.jpg) + +
Figure 4
+ +<--- Page Split ---> + +Poynting vector topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +![](images/Figure_5.jpg) + +
Figure 5
+ +Fractal- like pattern in electromagnetic field of supertoroidal pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6
+ +Evolution of on- axis singularity distribution versus supertoroidal order. (See Manuscript file for full figure legend) + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Video1.mp4- Video2.mp4- Video3.mp4- Video4.mp4- SupplementaryInformationtopology.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206_det.mmd b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..8cc2f3c324c63efc0ce9e0c9ca43dcbc06970c27 --- /dev/null +++ b/preprint/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206/preprint__126886759851d7ab006726a2a5b4c6ce9d24ee1c0468e8a9a6cfd9c1f1293206_det.mmd @@ -0,0 +1,285 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 757, 176]]<|/det|> +# Supertoroidal light pulses: Propagating electromagnetic skyrmions in free space + +<|ref|>text<|/ref|><|det|>[[44, 195, 363, 216]]<|/det|> +Yijie Shen ( \(\boxed{\times}\) y.shen@soton.ac.uk) + +<|ref|>text<|/ref|><|det|>[[50, 219, 291, 238]]<|/det|> +University of Southampton + +<|ref|>text<|/ref|><|det|>[[44, 244, 140, 261]]<|/det|> +Yaonan Hou + +<|ref|>text<|/ref|><|det|>[[52, 265, 291, 283]]<|/det|> +University of Southampton + +<|ref|>text<|/ref|><|det|>[[44, 290, 225, 308]]<|/det|> +Nikitas Papasimakis + +<|ref|>text<|/ref|><|det|>[[52, 312, 291, 330]]<|/det|> +University of Southampton + +<|ref|>text<|/ref|><|det|>[[44, 336, 245, 355]]<|/det|> +Nikolay Zheludev + +<|ref|>text<|/ref|><|det|>[[52, 359, 291, 377]]<|/det|> +University of Southampton + +<|ref|>sub_title<|/ref|><|det|>[[44, 418, 102, 436]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 455, 681, 476]]<|/det|> +Keywords: electrodynamics, optics, spectroscopy, light- matter interactions + +<|ref|>text<|/ref|><|det|>[[44, 494, 296, 513]]<|/det|> +Posted Date: April 13th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 531, 463, 551]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 345437/v1 + +<|ref|>text<|/ref|><|det|>[[44, 568, 911, 611]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 647, 926, 690]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on October 8th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26037- w. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[165, 63, 835, 82]]<|/det|> +# Supertoroidal light pulses: Propagating electromagnetic skyrmions in free space + +<|ref|>text<|/ref|><|det|>[[137, 92, 865, 171]]<|/det|> +Yijie Shen \(^{1, *}\) , Yaonan Hou \(^{1}\) , Nikitas Papasimakis \(^{1}\) , and Nikolay I. Zheludev \(^{1,2}\) \(^{1}\) Optoelectronics Research Centre & Centre for Photonic Metamaterials, University of Southampton, Southampton SO17 1BJ, United Kingdom \(^{2}\) Centre for Disruptive Photonic Technologies, School of Physical and Mathematical Sciences and The Photonics Institute, Nanyang Technological University, Singapore 637378, Singapore (Dated: March 18, 2021) + +<|ref|>text<|/ref|><|det|>[[174, 179, 829, 275]]<|/det|> +Topologically complex transient electromagnetic fields give access to nontrivial light- matter interactions and provide additional degrees of freedom for information transfer. An important example of such electromagnetic excitations are space- time non- separable single- cycle pulses of toroidal topology, the exact solutions of Maxwell described by Hellwarth and Nouchi in 1996 and recently observed experimentally. Here we introduce a new family of electromagnetic excitation, the supertoroidal electromagnetic pulses, in which the Hellwarth- Nouchi pulse is just the simplest member. The supertoroidal pulses exhibit skyrmionic structure of the electromagnetic fields, multiple singularities in the Poynting vector maps and fractal- like distributions of energy backflow. They are of interest for transient light- matter interactions, ultrafast optics, spectroscopy, and toroidal electrodynamics. + +<|ref|>text<|/ref|><|det|>[[86, 300, 487, 530]]<|/det|> +Introduction - Topology of complex electromagnetic fields is attracting growing interest of the photonics and electromagnetics communities [1- 6], while topologically structured light fields find applications in super- resolution microscopy [7, 8], metrology [9, 10], and beyond [11- 13]. For example, the vortex beam with twisted phase, akin to a Mobius strips in phase domain, can carry orbital angular momentum with tuneable topological charges enabling advanced optical tweezers, machining, and communication applications [11- 13]. The complex electromagnetic strip and knots structures were also proposed as information carriers [14- 16]. Recently observed electromagnetic skyrmions are relevant to topological skyrmions quasiparticles in high- energy physics and condensed matter [8, 17- 20]. They have sophisticated vector topology [21- 23], and enable applications in super- resolution microscopy [8], ultrafast imaging [24], and give rise to new types of spin- orbit optical forces [25]. + +<|ref|>text<|/ref|><|det|>[[86, 531, 487, 639]]<|/det|> +While a large body of work on topological properties of structured continuous light beams may be found in literatures, works on the topology of the time- dependent electromagnetic excitations and pulses only start to appear. For instance, the "Flying Doughnut" pulses, or toroidal light pulses (TLPs) first described in 1996 by Hellwarth and Nouchi [26], with unique spatiotemporal topology predicted recently [27], have only very recently observed experimentally [28]. + +<|ref|>text<|/ref|><|det|>[[86, 640, 487, 803]]<|/det|> +Fuelled by a combination of advances in ultrafast lasers or THz emitters and in our ability to control the spatiotemporal structure of light [29, 30] together with the introduction of new experimental and theoretical pulse characterization methods [31, 32], TLPs are attracting growing attention. Indeed, TLPs exhibit their complex topological structure with vector singularities and interact with matter through coupling to toroidal and anapole localized modes [33- 35]. Generation of TLPs in the optical and THz parts of the spectrum now paves the way towards new forms of spectroscopy, sensitive to toroidal and anapole excitations, and new information transfer schemes. + +<|ref|>text<|/ref|><|det|>[[86, 804, 487, 912], [516, 300, 916, 328]]<|/det|> +In this paper, we report that the Hellwarth and Nouchi pulses, are, in fact, the simplest example of an extended family of pulses that we will call supertoroidal light pulses (STLPs). We will show that supertoroidal light pulses introduced here exhibit complex topological structures that can be controlled by a single numerical parameter. The STLP display skyrmionlike arrangements of the transient electromagnetic fields organized in a fractal- like, self- affine manner, while the Poynting vector of the pulses feature singularities linked to the multiple energy backflow zones. + +<|ref|>sub_title<|/ref|><|det|>[[516, 336, 565, 348]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[516, 350, 916, 696]]<|/det|> +Supertoroidal electromagnetic pulses - Following Ziolkowski, localized finite- energy pulses can be obtained as superpositions of "electromagnetic directed- energy pulse trains" [36]. A special case of the localized finite- energy pulses was investigated by Hellwarth and Nouchi [26], who found the closed- form expression describing a single- cycle finite energy electromagnetic excitation with toroidal topology obtained from a scalar generating function \(f(\mathbf{r}, t)\) that satisfies the wave equation \((\nabla^2 - \frac{1}{c^2} \frac{\partial^2}{\partial t^2}) f(\mathbf{r}, t) = 0\) , where \(\mathbf{r} = (r, \theta , z)\) are cylindrical coordinates, \(t\) is time, \(c = 1 / \sqrt{\epsilon_0 \mu_0}\) is the speed of light, and the \(\epsilon_0\) and \(\mu_0\) are the permittivity and permeability of medium. Then, the exact solution of \(f(\mathbf{r}, t)\) can be given by the modified power spectrum method [26, 36], as \(f(\mathbf{r}, t) = f_0 / [(q_1 + i\tau)(s + q_2)^{\alpha}]\) , where \(f_0\) is a normalizing constant, \(s = r^2 /(q_1 + i\tau) - i\sigma\) , \(\tau = z - ct\) , \(\sigma = z + ct\) , \(q_1\) and \(q_2\) are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while \(\alpha\) is a real dimensionless parameter that must satisfy \(\alpha \geq 1\) to ensure finite energy solutions. Next, transverse electric (TE) and transverse magnetic (TM) solutions are readily obtained by using Hertz potentials. The electromagnetic fields for the TE solution can be derived by the potential \(\mathbf{A}(\mathbf{r}, t) = \mu_0 \nabla \times \hat{\mathbf{z}} f(\mathbf{r}, t)\) as \(\mathbf{E}(\mathbf{r}, t) = -\mu_0 \frac{\partial}{\partial t} \nabla \times \mathbf{A}\) and \(\mathbf{H}(\mathbf{r}, t) = \nabla \times (\nabla \times \mathbf{A})\) [26, 36]. Finally assuming \(\alpha = 1\) , the electromagnetic fields of the TLP are described by [26]: + +<|ref|>equation<|/ref|><|det|>[[560, 707, 916, 824]]<|/det|> +\[\begin{array}{l}{E_{\theta} = -4i f_{0}\sqrt{\frac{\mu_{0}}{\epsilon_{0}}\frac{r(q_{1} + q_{2} - 2i c t)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}}}\\ {H_{r} = 4i f_{0}\frac{r(q_{2} - q_{1} - 2i z)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}}\\ {H_{z} = -4f_{0}\frac{r^{2} - (q_{1} + i\tau)(q_{2} - i\sigma)}{[r^{2} + (q_{1} + i\tau)(q_{2} - i\sigma)]^{3}}} \end{array} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[516, 832, 916, 912]]<|/det|> +where the electric field \(E_{\theta}\) is azimuthally polarized with no longitudinal or radial components, whereas the magnetic field is oriented along the radial and longitudinal directions, \(H_{r}\) and \(H_{z}\) , with no azimuthal component. Equations (1- 3) derived by Hellwarth and Nouchi for \(\alpha = 1\) show the simplest example of TLPs. Here we explore the general solution for values of + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[82, 60, 920, 365]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 380, 919, 458]]<|/det|> +
Figure 1 From toroidal to supertoroidal light pulses: a,b, Isosurfaces for the electric fields of (a) the fundamental TLP \(\mathrm{Re}[E_{\theta}(\mathbf{r},t)]\) , and (b) a STLP \(\mathrm{Re}[E_{\theta}^{(\alpha)}(\mathbf{r},t)]\) of \(\alpha = 5\) , at amplitude levels of \(E = \pm 0.1\) and the Rayleigh range of \(q_{2} = 100q_{1}\) , at different times of \(t = 0\) , \(\pm q_{2} / (4c)\) , and \(\pm q_{2} / (2c)\) . \(x\) - \(z\) cross-sections of the instantaneous electric field at \(y = 0\) . The insets in (a) and (b) are schematics of spatial topological structures of magnetic vector fields at focus \((t = 0)\) for the fundamental TLP and STLP, respectively. The gray dots and rings mark the distribution of singularities in magnetic field, large purple arrows mark the direction of magnetic field vector, and the smaller coloured arrows show the skyrmionic structures in magnetic field.
+ +<|ref|>text<|/ref|><|det|>[[85, 486, 488, 502]]<|/det|> +\(\alpha \geq 1\) . In the TE case, electric and magnetic fields are given by (see detailed derivation in Supplementary Information): + +<|ref|>equation<|/ref|><|det|>[[170, 544, 912, 675]]<|/det|> +\[\begin{array}{l}{{E_{\theta}^{(\alpha)}=-2\alpha if_{0}\sqrt{\frac{\mu_{0}}{\epsilon_{0}}\left\{\frac{(\alpha+1)r(q_{1}+i\tau)^{\alpha-1}(q_{1}+q_{2}-2i c t)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}-\frac{(\alpha-1)r(q_{1}+i\tau)^{\alpha-2}}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\\ {{H_{r}^{(\alpha)}=2\alpha if_{0}\left\{\frac{(\alpha+1)r(q_{1}+i\tau)^{\alpha-1}(q_{2}-q_{1}-2i z)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}-\frac{(\alpha-1)r(q_{1}+i\tau)^{\alpha-2}}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\\ {{H_{z}^{(\alpha)}=-4\alpha f_{0}\left\{\frac{(q_{1}+i\tau)^{\alpha-1}[r^{2}-\alpha(q_{1}+i\tau)(q_{2}-i\sigma)]}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+2}}+\frac{(\alpha-1)(q_{1}+i\tau)^{\alpha-2}(q_{2}-i\sigma)}{[r^{2}+(q_{1}+i\tau)(q_{2}-i\sigma)]^{\alpha+1}}\right\}}}\end{array} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[85, 712, 488, 783]]<|/det|> +For \(\alpha = 1\) , the electromagnetic fields in Eqs. (4- 6) are reduced to that of the fundamental TLP Eqs. (1- 3). Moreover, the real and imaginary parts of equations (4- 6), simultaneity fulfill Maxwell equations and therefore represent real electromagnetic pulses (see Supplementary Materials). + +<|ref|>text<|/ref|><|det|>[[86, 789, 488, 913]]<|/det|> +While propagating in free space toroidal and supertoroidal pulses exhibit self- focusing. Figure 1 shows the evolution of the fundamental \((\alpha = 1)\) TLP and STLP \((\alpha = 5)\) upon propagation through the focal point. In the former case, the pulse is single- cycle at focus \((z = 0)\) becoming \(1\frac{1}{2}\) - cycle at the boundaries of Rayleigh range \(z = \pm q_{2} / 2\) (Fig. 1a). On the other hand, STLPs with \(\alpha >1\) (Fig. 1b, \(\alpha = 5\) ) exhibit a substantially more complex spatiotemporal evolution where the pulse is being reshaped multiple times upon propagation + +<|ref|>text<|/ref|><|det|>[[515, 712, 916, 740]]<|/det|> +(See Video 1 and Video 2 in Supplementary Materials for the dynamic evolution of the fundamental TLP and STLPs). + +<|ref|>text<|/ref|><|det|>[[515, 747, 917, 913]]<|/det|> +Electric field singularities - Figure 2 comparatively shows the instantaneous electric fields for the TE single- cycle fundamental TLP and STLP \((\alpha = 5)\) with \(q_{2} = 20q_{1}\) at the focus \((t = 0)\) . In all cases, the electric field vanishes on the \(z\) - axis \((r = 0)\) ; see the vertical solid black lines in Figs. 2a and 2b) owing to the azimuthal polarization and also in the \(z = 0\) plane (see the horizontal solid black lines in Figs. 2a and 2b) due to the odd symmetry of Eqs. Eqs. (1,4) with respect to \(z\) . For the fundamental TLP, the electric field vanishes on two spherical shells (indicated by the solid circles in Fig. 2a) on the positive and negative \(z\) - axis, respectively. This behavior can be more clearly observed in the transverse distributions + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 66, 901, 487]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 496, 919, 576]]<|/det|> +
Figure 2 Electric field topology of toroidal and supertoroidal light pulses: a,b, The isoline plots of the electric field in the \(x - z\) plane for (a) the fundamental TLP, \(\mathrm{Re}[E_{\theta}(\mathbf{r},t = 0)]\) , and (b) the STLP of \(\alpha = 5\) , \(\mathrm{Re}[E_{\theta}^{(\alpha = 5)}(\mathbf{r},t = 0)]\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Dashed lines marked as a1-a3 and b1-b3 indicate the \(z\) -levels of the cross-sections on the right panels correspondingly. Panels a1-a3 and b1-b3 present isoline plots of the electric field and arrow plots of the electric field direction in the \(x - y\) plane, in logarithmic scale. Solid black lines and black dots mark the points and areas, where the electric field vanishes. Blue and red arrows indicate the two opposite azimuthal directions of the electric fields. Unit for coordinates: \(q_{1}\) .
+ +<|ref|>text<|/ref|><|det|>[[86, 604, 488, 905], [515, 604, 917, 673]]<|/det|> +at three different propagation distances, \(z = q_{1},5q_{1},35q_{1}\) of Figs. 2a1- 2a3. In accordance to Figs. 2a1 and 2a3, the electric field at \(z\) positions close to and away from the center of the pulse (cross- sections a1 and a3) rotates counter- clockwise forming a vortex around the center singularity along the propagation axis. However, at a distance of \(z = 5q_{1}\) from its center (Figs. 2a2), the electric field vanishes on a circular boundary, which corresponds to a spherical region inside which the electric field is oriented along the clockwise direction, whereas outside this region the electric field remains oriented in the opposite (counter- clockwise) direction (see cross- section a3). For the STLP case, a more complex matryoshka- like structure emerges with multiple nested singularity shells, replacing the single shell of the fundamental TLP, as Fig. 2b shows. The electric field configuration close the singularity shells can be examined in detail at transverse planes at \(z = q_{1},5q_{1},35q_{1}\) (Figs. 2b1- b3). In this case, at transverse planes close to \(z = 0\) , the electric field changes orientation from counter- clockwise close to \(r = 0\) to clockwise away from the \(z\) - axis (see Fig. 2b1). On the other hand, on transverse planes close to \(z = 5q_{1}\) (see Fig. 2b2), two singular rings (corresponding to the two singular shells) emerge as a cross- sections of the multi- layer singularity shell structure, separating space in three different regions, in which the electric field direction alternates between counter- clockwise \((r / q_{1}< 7)\) , to clockwise \((7< r / q_{1}< 15)\) , and again to counter- clockwise \((r / q_{1} > 15)\) . + +<|ref|>text<|/ref|><|det|>[[515, 674, 917, 782]]<|/det|> +In general, the pulse of higher order of \(\alpha\) is accompanied by a more complex multi- layer singular- shell structure, see the dynamic evolution versus the order index in Video 3 in Supplementary Materials. Although the above results of electric fields are instantaneous at \(t = 0\) , we note that the multi layer shall structure propagation of supertoroidal light pulse is retained during propagation, see such dynamic process in Video 4 in Supplementary Materials. + +<|ref|>text<|/ref|><|det|>[[515, 789, 917, 911]]<|/det|> +Magnetic field singularities – The magnetic field of STLPs has both radial and longitudinal components, \(\mathbf{H} = H_{r}\hat{\mathbf{r}} + H_{z}\hat{\mathbf{z}}\) , which lead to a topological structure more complex than the one exhibited by the electric field. Figure 3 comparatively shows the instantaneous magnetic fields for the TLP and the STLP of \(\alpha = 5\) . For the fundamental TLP (Fig. 3a), the magnetic field has ten different vector singularities on the \(x\) - \(z\) plane, including four saddle points [the longitudinal field component pointing towards (away from) and the radial com + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 65, 916, 490]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 497, 919, 619]]<|/det|> +
Figure 3 Magnetic field topology of toroidal and supertoroidal light pulses: a,b, Isoline and arrow plots of the magnetic fields in the \(x - z\) plane for (a) the fundamental TLP and (b) the STLP of \(\alpha = 5\) , in logarithmic scale. Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panels a1 and b1 present the zoom-ins of the regions highlighted by blue in (a) and (b) respectively. Panels a2 and b2 present isoline and arrow plots of the magnetic fields at \(z = 0\) planes, in logarithmic scale, marked by the black dashed lines in (a) and (b) respectively, where the magnetic fields vanish along the circular solid black lines with red arrows marking the styles of singularities (vortex for (a2) and saddle for (b2)). Skyrmionic structures in magnetic fields of toroidal and supertoroidal light pulses: c, Various textures of Néel-type skyrmionic structure observed at various transverse planes (see dashed purple lines in (a-b)) for the fundamental TLP (c1-c2) and the STLP of \(\alpha = 5\) (c3-c6), which are demonstrated by the arrows with color-labeled longitudinal component value of magnetic field. The up-right insert of each panel shows the basic texture of the skyrmionic structure. Unit for coordinates: \(q_{1}\) .
+ +<|ref|>text<|/ref|><|det|>[[86, 648, 488, 907], [515, 648, 917, 814]]<|/det|> +ponent away from (toward) the singularity) on \(z\) - axis and six vortex rings [the surrounding vector distribution forming a vortex loop] away from the \(z\) - axis. We note that we only consider the singularities existed at an area containing \(99.9\%\) of the energy of the pulse. While the singularity existed at the region far away from the pulse center with nearly zero energy can be neglected. A zoom- in of the field structure around these singularities can be seen in Fig. 3a1. For the three off- axis singularities located at the \(x > 0\) half space, two of them (at \(z > 0\) and \(z < 0\) ) are accompanied by counter- clockwise rotating vortices, whereas the third one (at \(z = 0\) ) by a clockwise rotating vortex. Owing to the cylindrical symmetry of the pulse, the off- axis singularities correspond in essence to singularity rings. Such an example is shown in Fig. 3a2, which presents the magnetic field on the transverse plane at \(z = 0\) . Here, the magnetic field points toward the positive (negative) \(z\) - axis inside (outside) the circular region resulting in the formation of a toroidal vortex winding around the singularity ring. For the STLP (Fig. 3b), more vector singularities are unveiled in the magnetic field with six saddle points on \(z\) - axis and six off- axis singularities. A zoom- in of the field structure around these singularities can be seen in Fig. 3b1. The orientation of the magnetic field around the on- axis saddle points is alternating between "longitudinal- toward radial- outward point" and "axial- toward longitudinal- outward", similarly to the on- axis singularities of the TLP. Moreover, the off- axis singularities at \(z = 0\) become now saddle points contributing to the singularity ring in the \(z = 0\) plane. The remaining off- axis singularities are accompanied by clockwise and counterclockwise magnetic field configurations at \(x > 0\) and \(x < 0\) , respectively as shown Fig. 3b2. + +<|ref|>text<|/ref|><|det|>[[515, 831, 917, 911]]<|/det|> +Skyrmionic structure in magnetic field – A topological feature of particular interest here is the skyrmionic structure observed in the magnetic field configuration of STLPs. The skyrmion is a topologically protected quasiparticle in condensed matter with a hedgehog- like vectorial field, that gradually changes orientation as one moves away from the + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[85, 66, 916, 290]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 299, 916, 348]]<|/det|> +
Figure 4 Poynting vector topology of toroidal and supertoroidal light pulses: a,b, Contour and arrow plots of the Poynting vector fields in the \(x - z\) plane, in logarithmic scale, for (a) the fundamental TLP and (b) the STLP of \(\alpha = 5\) . Panels (a1) and (b1) present zoom-ins of the areas highlighted by green in (a-b), respectively. Solid black lines and dots mark the zeros of the Poynting vector. Red and blue arrows indicate areas with forward and backward energy flow, respectively. Unit for coordinates: \(q_1\) .
+ +<|ref|>text<|/ref|><|det|>[[86, 377, 486, 459]]<|/det|> +skyrmion centre [21- 23]. Recently skyrmion- like configurations have been reported in electromagnetism, including skyrmion modes in surface plasmon polaritons [17] and the spin field of focused beams [8, 19]. Here we observe the skyrmion field configurations in the magnetic field of propagating STLPs. + +<|ref|>text<|/ref|><|det|>[[86, 460, 487, 623]]<|/det|> +The topological properties of a skyrmionic configuration can be characterized by the skyrmion number \(s\) , which can be separated into a polarity \(p\) and vorticity number \(m\) [23]. The polarity represents the direction of the vector field, down (up) at \(r = 0\) and up (down) at \(r \to \infty\) for \(p = 1\) ( \(p = - 1\) ), the vorticity controls the distribution of the transverse field components, and another initial phase \(\gamma\) should be added for determining the helical vector distribution, see Methods for details. For the \(m = 1\) skyrmion, the cases of \(\gamma = 0\) and \(\gamma = \pi\) are classified as Neel- type, and the cases of \(\gamma = \pm \pi /2\) are classified as Bloch- type. The case for \(m = - 1\) is classified as anti- skyrmion. + +<|ref|>text<|/ref|><|det|>[[86, 624, 487, 830]]<|/det|> +Here the vector forming skyrmionic structure is defined by the normalized magnetic field \(\mathbf{H} = \mathbf{H} / |\mathbf{H}|\) of the STLP. Two examples of two skyrmionic structures in the fundamental TLP are shown in Figs. 3c1 ( \(p = m = 1\) , \(\gamma = \pi\) ) and 3c2 ( \(p = m = 1\) , \(\gamma = 0\) ) occurs at the two transverse planes marked by purple dashed lines c1 and c2, which are both Neel- type skyrmionic structures, where the vector changes its direction from "down" at the centre to "up" away from the centre. In the case of the STLPs with more complex topology, it is possible to observe more skyrmionic structures. The STLP pulse ( \(\alpha = 5\) ) exhibits not only the clockwise ( \(p = m = 1\) , \(\gamma = \pi\) ) and counter- clockwise ( \(p = m = 1\) , \(\gamma = 0\) ) Neel- type skyrmionic structures (Fig. 3c3 and 3c4), but also those with \(p = - 1\) , \(m = 1\) , \(\gamma = \pi\) and \(p = - 1\) , \(m = 1\) , \(\gamma = 0\) , in Fig. 3c6- c6. + +<|ref|>text<|/ref|><|det|>[[86, 832, 487, 912]]<|/det|> +In general, as the value of \(\alpha\) increases, toroidal pulses show an increasingly complex magnetic field pattern with skyrmionic structures of multiple types, see Video 3 in Supplementary Materials. We also note that the topology of the STLP is maintained during propagation, see Video 4 in Supplementary Materials. + +<|ref|>text<|/ref|><|det|>[[515, 377, 916, 662]]<|/det|> +Energy backflow and Poynting vector singularities - The singularities of the electric and magnetic fields are linked to the complex topological behavior for the energy flow as represented by the Poynting vector \(\mathbf{S} = \mathbf{E} \times \mathbf{H}\) . An interesting effect for the fundamental TLP is the presence of energy backflow: the Poynting vector at certain regions is oriented against the prorogation direction (blue arrows in Fig. 4a) [27]. Such energy backflow effects have been predicted and discussed in the context of singular superpositions of waves [1, 37], superoscillatory light fields [9, 38], and plasmonic nanostructures [39]. The Poynting vector map reveals a complex multi- layer energy backflow structure, as shown in Fig. 4b. The energy flow vanishes at the positions of the electric and magnetic singularities and inherits their multi- layer matryoshka- like structure. Poynting vector vanishes at \(z = 0\) plane, along the \(z\) - axis, and on the dual- layer matryoshka- like singular shells (marked by the black bold lines in Fig. 4b). Importantly, energy backflow occurs at areas of relatively low energy density, and, hence, STLP as a whole still propagates forward. For the temporal evolution of the energy flow of the pulse see Videos 3 & 4 in Supplementary Materials. + +<|ref|>text<|/ref|><|det|>[[515, 672, 916, 875]]<|/det|> +Fractal patterns hidden in electromagnetic fields - As the order \(\alpha\) of the pulse increases (see Video 3 in Supplementary Material), the topological features of the STLP appear to be organized in a hierarchical, fractal- like fashion. A characteristic case of the STLP of \(\alpha = 20\) is presented in (Fig. 5). For the electric field, the matryoshka- like singular shells involve an increasing number of layers as one examines the pulse at finer length scales, forming a self- similar pattern that seems infinitely repeated. For the magnetic field, the saddle and vortex points are distributed along the propagation axis and in two planes crossing the pulse centre, respectively. The distribution of singularities becomes increasingly dense as one approaches the centre of the pulse, resulting in a self- similar pattern. A similar pattern can be seen for the Poynting vector map (see Videos 3 & 4 in Supplementary Materials). + +<|ref|>text<|/ref|><|det|>[[515, 885, 916, 912]]<|/det|> +Fine- scale features of skyrmionic structures - The fractal- like pattern of vectorial magnetic field of a high- order STLPs + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 60, 920, 545]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 555, 919, 656]]<|/det|> +
Figure 5 Fractal-like pattern in electromagnetic field of supertoroidal pulses: a. The isoline plot of the electric field in the \(x - z\) plane for the STLP of \(\alpha = 20\) , \(\mathrm{Re}[E_{\theta}^{(\alpha = 20)}(r,t = 0)]\) , in logarithmic scale. Solid black lines indicate the zeros of the electric field. Panel a1 presents the zoom-in of the region highlighted by blue in (a). b, Isoline and arrow plot of the logarithm of the magnetic field in the \(x - z\) plane for the STLP of \(\alpha = 20\) . Black dots indicate the zeros of the magnetic field with red arrows correspondingly marking the saddle or vortex style of the vector singularities. Panel b1 presents the zoom-in of the region highlighted by blue in (b). Fine-scale features of skyrmionic structures: c1-c4. The skyrmionic distributions of magnetic field at several transverse planes marked by dashed lines c1-c4 in (b1). d1-d4, the distribution of normalized magnetic field and its absolute value versus \(x\) for the skyrmionic structures in (c1-c4). Insets illustrate the fine-scale features at the regions highlighted by gray bands. Unit for coordinates: \(q_{1}\) .
+ +<|ref|>text<|/ref|><|det|>[[86, 685, 487, 904], [515, 686, 916, 779]]<|/det|> +results in skyrmionic configurations with with features changing much faster than the effective wavelength \(q_{1}\) . Figures 5c1- c4 show the four skyrmionic structures of the high- order STLP \((\alpha = 20)\) at the four transverse planes marked by the dashed lines in Fig. 5b1 at positions of \(z / q_{1} = 14,10,6,3.5\) correspondingly. The four skyrmionic structures have two different topologies with topological numbers of \((p,m,\gamma) = (1,1,\pi)\) , for c1 and c3, and \((- 1,1,0)\) , for c2 and c4. In addition, they exhibit an effect of "spin reversal", where the number reversals is given by \(\bar{p} = \frac{1}{2\pi} [\beta (r)]_{r = 0}^{r = \infty}\) , e.g. \(\bar{p} = 1,2,3,4\) for the skyrmionic structures in Figs. 5c1- c4, respectively. Each reversal corresponds to a sign change of \(H_{z}\) , which takes place over areas much smaller than the effective wavelength of the pulse \((q_{1})\) . The full width at half maximum of these areas for the four skyrmionic structures is 1/6, 1/10, 1/30, 1/50 of the effective wavelength, respectively. Conclusively, the sign reversals become increasingly rapid in transverse planes closer to the pulse center \((z = 0)\) , see Figs. 5d1- d4. Similarly, increasing the value of \(\alpha\) leads to increasingly sharper singularities. Notably, in contrast to the fundamental TLP, the skyrmionic configurations in STLP occur at areas of higher energy density, and thus we expect that they could be observed experimentally. + +<|ref|>text<|/ref|><|det|>[[515, 787, 916, 911]]<|/det|> +The topological structure of the STLP is directly related to the distribution of on- axis saddle- points in its magnetic field. Indeed, the latter mark the intersection of the \(E\) - field singular shells with the \(z\) - axis, which in turn results in the emergence of different skyrmionic magnetic field patterns (see Fig. 5 and Videos 3 & 4). The number and position of on- axis magnetic field saddle points is defined by the supertoroidal parameter \(\alpha\) . This is illustrated in Fig. 6a, where we plot the number of on- axis \(H\) - field singularities as a function of alpha for a STLP + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[86, 62, 485, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 411, 472, 473]]<|/det|> +
Figure 6 Evolution of on-axis singularity distribution versus supertoroidal order: a,b, The number (a) and positions (b) of on-axis saddle-singularities of the magnetic field of the STLP \((q_{2} = 20q_{1})\) within the range of \(z\in [0,80q_{1}]\) , versus \(\alpha\) . The blue dashed line marks where the number of singularities decreases.
+ +<|ref|>text<|/ref|><|det|>[[86, 506, 488, 657]]<|/det|> +with \(q_{2} = 20q_{1}\) . The number of singularities is generally increasing with increasing \(\alpha\) apart for values around \(\alpha = 5.6\) (marked by blue dashed line in Fig. 6). Moreover, the number of singularities increases in a ladder- like fashion, where only specific values of alpha lead to additional singularities. The origin of this behaviour can be traced to changes in the pulse structure as \(\alpha\) increases (see Fig. 6b). For specific values of \(\alpha\) , additional singularities appear away from the pulse center \((z = 0)\) and then move slowly towards it. On the other hand, the irregular behavior at \(\alpha = 5.6\) is a result of two singularities disappearing (see blue dashed line in Fig. 6b). + +<|ref|>sub_title<|/ref|><|det|>[[86, 665, 160, 678]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[86, 679, 488, 830]]<|/det|> +STLPs exhibit complex and unique topological structure. The electric field exhibits a matryoshka- like configuration of singularity shells, which divide the STLP into "nested" regions with opposite azimuthal polarization. The magnetic field exhibits multi- texture skyrmionic structures at various transverse planes of a single pulse, related to the distribution of multiple saddle and vortex singularities. The instantaneous Poynting vector field exhibits multiple singularities with regions of energy backflow. The singularities of the STLP appear to be hierarchically organized resulting in self- similar, fractal- like patterns for higher- order pulses. + +<|ref|>text<|/ref|><|det|>[[86, 831, 488, 912]]<|/det|> +In conclusion, to the best of our knowledge, STLPs are so far the only known example of free- space propagating skyrmionic field configurations. Their structure contains sharp singularities of interest for super- resolution metrology and microscopy, optical information encoding, optical trapping and particle acceleration. + +<|ref|>sub_title<|/ref|><|det|>[[515, 66, 575, 79]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[515, 80, 916, 386]]<|/det|> +Solving the supertoroidal pulses - The first step is to solve the scalar generating function \(f(\mathbf{r},t)\) that satisfies the wave equation \((\nabla^{2} - \frac{1}{c^{2}}\frac{\partial^{2}}{\partial t^{2}})f(\mathbf{r},t) = 0\) , where \(\mathbf{r} = (r,\theta ,z)\) are cylindrical coordinates, \(t\) is time, \(c = 1 / \sqrt{\epsilon_{0}\mu_{0}}\) is the speed of light, and the \(\epsilon_{0}\) and \(\mu_{0}\) are the permittivity and permeability of medium. The exact solution of \(f(\mathbf{r},t)\) can be given by the modified power spectrum method as \(f(\mathbf{r},t) = f_{0} / [(q_{1} + i\tau)(s + q_{2})^{\alpha}]\) , where \(f_{0}\) is a normalizing constant, \(s = r^{2} / (q_{1} + i\tau) - i\sigma\) , \(\tau = z - ct\) , \(\sigma = z + ct\) , \(q_{1}\) and \(q_{2}\) are parameters with dimensions of length and act as effective wavelength and Rayleigh range under the paraxial limit, while \(\alpha\) is a real dimensionless parameter that must satisfy \(\alpha \geq 1\) to ensure finite energy solutions. The next step is constructing the Hertz potential. For fulfilling the toroidal symmetric and azimuthally polarized structure, the Hertz potential should be constructed as \(\mathbf{A}(\mathbf{r},t) = \mu_{0}\nabla \times \hat{\mathbf{z}} f(\mathbf{r},t)\) . Then, the exact solutions of solutions of transverse electric (TE) and transverse magnetic (TM) modes are readily obtained by using Hertz potential. The electromagnetic fields for the TE solution can be derived by the potential as \(\mathbf{E}(\mathbf{r},t) = -\mu_{0}\frac{\partial}{\partial t}\nabla \times \mathbf{A}\) and \(\mathbf{H}(\mathbf{r},t) = \nabla \times (\nabla \times \mathbf{A})\) [26, 36], see Supplementary Information for more detailed derivations. + +<|ref|>text<|/ref|><|det|>[[515, 394, 916, 437]]<|/det|> +Characterizing topology of skyrmion - The topological properties of a skyrmionic configuration can be characterized by the skyrmion number defined by [23]: + +<|ref|>equation<|/ref|><|det|>[[592, 445, 915, 485]]<|/det|> +\[s = \frac{1}{4\pi}\iint \mathbf{n}\cdot \left(\frac{\partial\mathbf{n}}{\partial x}\times \frac{\partial\mathbf{n}}{\partial y}\right)\mathrm{d}x\mathrm{d}y \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[515, 497, 916, 565]]<|/det|> +that is an integer counting how many times the vector \(\mathbf{n}(x,y) = \mathbf{n}(r\cos \theta ,r\sin \theta)\) wraps around the unit sphere. For mapping to the unit sphere, the vector can be given by \(\mathbf{n} = (\cos \alpha (\theta)\sin \beta (r),\sin \alpha (\theta)\sin \beta (r),\cos \beta (r))\) . Also, The skyrmion number can be separated into two integers: + +<|ref|>equation<|/ref|><|det|>[[562, 575, 915, 646]]<|/det|> +\[\begin{array}{l}{s = \frac{1}{4\pi}\int_0^\infty \mathrm{d}r\int_0^{2\pi}\mathrm{d}\phi \frac{\mathrm{d}\beta(r)}{\mathrm{d}r}\frac{\mathrm{d}\alpha(\theta)}{\mathrm{d}\theta}\sin \beta (r)}\\ {= \frac{1}{4\pi} [\cos \beta (r)]_{r = 0}^{r = \infty}[\alpha (\theta)]_{\theta = 0}^{\theta = 2\pi} = p\cdot m} \end{array} \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[515, 655, 916, 779]]<|/det|> +the polarity, \(p = \frac{1}{2} [\cos \beta (r)]_{r = 0}^{r = \infty}\) , represents the direction of the vector field, down (up) at \(r = 0\) and up (down) at \(r\rightarrow \infty\) for \(p = 1\) \((p = - 1)\) . The vorticity number, \(m = \frac{1}{2\pi} [\alpha (\theta)]_{\theta = 0}^{\theta = 2\pi}\) , controls the distribution of the transverse field components. In the case of a helical distribution, an initial phase \(\gamma\) should be added, \(\alpha (\theta) = m\theta +\gamma\) . For the \(m = 1\) skyrmion, the cases of \(\gamma = 0\) and \(\gamma = \pi\) are classified as Neel- type, and the cases of \(\gamma = \pm \pi /2\) are classified as Bloch- type. The case for \(m = - 1\) is classified as anti- skyrmion. + +<|ref|>sub_title<|/ref|><|det|>[[530, 789, 661, 802]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[515, 803, 916, 912]]<|/det|> +The authors acknowledge the supports of the MOE Singapore (MOE2016- T3- 1- 006), the UKs Engineering and Physical Sciences Research Council (grant EP/M009122/1, Funder Id: http://dx.doi.org/10.13039/501100000266), the European Research Council (Advanced grant FLEET- 786851, Funder Id: http://dx.doi.org/10.13039/501100000781), and the Defense Advanced Research Projects Agency (DARPA) under the Nascent Light Matter Interactions program. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 67, 488, 108]]<|/det|> +Data availability The data from this paper can be obtained from the University of Southampton ePrints research repository https://doi.org/xx.xxxx/SOTON/xxxxxx. + +<|ref|>text<|/ref|><|det|>[[101, 118, 241, 132]]<|/det|> +\\*y.shen@soton.ac.uk + +<|ref|>text<|/ref|><|det|>[[85, 180, 490, 914]]<|/det|> +[1] Michael Berry, "Making waves in physics," Nature 403, 21- 21 (2000). [2] Marat Soskin, Svetlana V Boriskina, Yidong Chong, Mark R Dennis, and Anton Desyatnikov, "Singular optics and topological photonics," Journal of Optics 19, 010401 (2016). [3] Ling Lu, John D Joannopoulos, and Marin Soljačić, "Topological photonics," Nature Photonics 8, 821- 829 (2014). [4] Alexander B Khanikaev and Gennady Shvets, "Two- dimensional topological photonics," Nature Photonics 11, 763- 773 (2017). [5] Tomoki Ozawa, Hannah M Price, Alberto Amo, Nathan Goldman, Mohammad Hafezi, Ling Lu, Mikael C Rechtsman, David Schuster, Jonathan Simon, Oded Zilberberg, et al., "Topological photonics," Reviews of Modern Physics 91, 015006 (2019). [6] N Asger Mortensen, Sergey I Bozhevolnyi, and Andrea Alu, "Topological nanophotonics," Nanophotonics 8, 1315- 1317 (2019). [7] Tanchao Pu, Jun- Yu Ou, Vassili Savinov, Guanghui Yuan, Nikitas Papasimakis, and Nikolay I Zheludev, "Unlabeled far- field deeply subwavelength topological microscopy (dstm)," Advanced Science 8, 2002886 (2021). [8] Luping Du, Aiping Yang, Anatoly V Zayats, and Xiaocong Yuan, "Deep- subwavelength features of photonic skyrmions in a confined electromagnetic field with orbital angular momentum," Nature Physics 15, 650- 654 (2019). [9] Guang Hui Yuan and Nikolay I Zheludev, "Detecting nanometric displacements with optical ruler metrology," Science 364, 771- 775 (2019). [10] Shawn Sederberg, Fanqi Kong, Felix Hufnagel, Chunmei Zhang, Ebrahim Karimi, and Paul B Corkum, "Vectorized optoelectronic control and metrology in a semiconductor," Nature Photonics 14, 680- 685 (2020). [11] Yijie Shen, Xuejiao Wang, Zhenwei Xie, Changjun Min, Xing Fu, Qiang Liu, Mali Gong, and Xiaocong Yuan, "Optical vortices 30 years on: Oam manipulation from topological charge to multiple singularities," Light: Science & Applications 8, 1- 29 (2019). [12] Alison M Yao and Miles J Padgett, "Orbital angular momentum: origins, behavior and applications," Advances in Optics and Photonics 3, 161- 204 (2011). [13] Rui Chen, Hong Zhou, Marco Moretti, Xiaodong Wang, and Jiandong Li, "Orbital angular momentum waves: Generation, detection and emerging applications," IEEE Communications Surveys & Tutorials (2019). [14] Thomas Bauer, Peter Banzer, Ebrahim Karimi, Sergej Orlov, Andrea Rubano, Lorenzo Marrucci, Enrico Santamato, Robert W Boyd, and Gerd Leuchs, "Observation of optical polarization mubius strips," Science 347, 964- 966 (2015). [15] Thomas Bauer, Martin Neugebauer, Gerd Leuchs, and Peter Banzer, "Optical polarization mubius strips and points of purely transverse spin density," Physical Review Letters 117, 013601 (2016). [16] Hugo Larocque, Alessio D'Errico, Manuel F Ferrer- Garcia, Avishy Carmi, Eliahu Cohen, and Ebrahim Karimi, "Optical framed knots as information carriers," Nature Communications 11, 1- 8 (2020). [17] S Tsesses, E Ostrovsky, K Cohen, B Gjonaj, NH Lindner, and G Bartal, "Optical skyrmion lattice in evanescent electromagnetic fields," Science 361, 993- 996 (2018). [18] Yanan Dai, Zhikang Zhou, Atreyie Ghosh, Roger SK Mong, Atsushi Kubo, Chen- Bin Huang, and Hrvoje Petek, "Plasmonic + +<|ref|>text<|/ref|><|det|>[[512, 60, 919, 910]]<|/det|> +topological quasiparticle on the nanometre and femtosecond scales," Nature 588, 616- 619 (2020). [19] Sijia Gao, Fiona C Speirits, Francesco Castellucci, Sonja Franke- Arnold, Stephen M Barnett, and Jörg B Götte, "Paraxial skyrmionic beams," Physical Review A 102, 053513 (2020). [20] Rodrigo Gutiérrez Cuevas and Emilio Pisanty, "Optical polarization skyrmionic fields in free space," Journal of Optics (2021). [21] Albert Fert, Nicolas Reyren, and Vincent Cros, "Magnetic skyrmions: advances in physics and potential applications," Nature Reviews Materials 2, 1- 15 (2017). [22] XZ Yu, Yoshinori Onose, Naoya Kanazawa, JH Park, JH Han, Yoshio Matsui, Naoto Nagaosa, and Yoshinori Tokura, "Real- space observation of a two- dimensional skyrmion crystal," Nature 465, 901- 904 (2010). [23] Naoto Nagaosa and Yoshinori Tokura, "Topological properties and dynamics of magnetic skyrmions," Nature Nanotechnology 8, 899- 911 (2013). [24] Timothy J Davis, David Janoschka, Pascal Dreher, Bettina Frank, Frank- J Meyer zu Heringdorf, and Harald Giessen, "Ultrafast vector imaging of plasmonic skyrmion dynamics with deep subwavelength resolution," Science 368 (2020). [25] Chris Halcrow and Derek Harland, "Attractive spin- orbit potential from the skyrme model," Physical Review Letters 125, 042501 (2020). [26] RW Hellwarth and P Nouchi, "Focused one- cycle electromagnetic pulses," Physical Review E 54, 889 (1996). [27] Apostolos Zdagkas, Nikitas Papasimakis, Vassili Savinov, Mark R Dennis, and Nikolay I Zheludev, "Singularities in the flying electromagnetic doughnuts," Nanophotonics 8, 1379- 1385 (2019). [28] A Zdagkas, Y Shen, C McDonnell, J Deng, G Li, T Ellenbogen, N Papasimakis, and NI Zheludev, "Observation of toroidal pulses of light," arXiv preprint arXiv:2102.03636 (2021). [29] Nikitas Papasimakis, Tim Raybould, Vassili A Fedotov, Din Ping Tsai, Ian Youngs, and Nikolay I Zheludev, "Pulse generation scheme for flying electromagnetic doughnuts," Physical Review B 97, 201409 (2018). [30] Cormac McDonnell, Junhong Deng, Symeon Sideris, Tal Ellenbogen, and Guixin Li, "Functional thz emitters based on pancharatnam- berry phase nonlinear metasurfaces," Nature Communications 12, 1- 8 (2021). [31] Apostolos Zdagkas, Venkatram Nalla, Nikitas Papasimakis, and Nikolay I Zheludev, "Spatio- temporal characterization of ultrashort vector pulses," arXiv preprint arXiv:2101.11651 (2021). [32] Yijie Shen, Apostolos Zdagkas, Nikitas Papasimakis, and Nikolay I Zheludev, "Measures of space- time nonseparability of electromagnetic pulses," Physical Review Research 3, 013236 (2021). [33] Nikitas Papasimakis, VA Fedotov, Vassili Savinov, TA Raybould, and NI Zheludev, "Electromagnetic toroidal excitations in matter and free space," Nature Materials 15, 263- 271 (2016). [34] Tim Raybould, Vassili A Fedotov, Nikitas Papasimakis, Ian Youngs, and Nikolay I Zheludev, "Exciting dynamic anapoles with electromagnetic doughnut pulses," Applied Physics Letters 111, 081104 (2017). [35] Vassili Savinov, Nikitas Papasimakis, DP Tsai, and NI Zheludev, "Optical anapoles," Communications Physics 2, 1- 4 (2019). [36] Richard W Ziolkowski, "Localized transmission of electromagnetic energy," Physical Review A 39, 2005 (1989). [37] MV Berry, "Quantum backflow, negative kinetic energy, and optical retro- propagation," Journal of Physics A: Mathematical and Theoretical 43, 415302 (2010). [38] Guanghui Yuan, Edward TF Rogers, and Nikolay I Zheludev, "plasmonics" in free space: observation of giant wavevectors, vortices, and energy backflow in superoscillatory optical fields," Light: Science & Applications 8, 1- 9 (2019). [39] MV Bashevoy, VA Fedotov, and NI Zheludev, "Optical whirlpool on an absorbing metallic nanoparticle," Optics Ex + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[545, 67, 716, 80]]<|/det|> +press 13, 8372–8379 (2005). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 143, 69]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[60, 116, 925, 438]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 466, 115, 485]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 508, 775, 530]]<|/det|> +From toroidal to supertoroidal light pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 50, 914, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 528, 117, 548]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[42, 570, 951, 592]]<|/det|> +Electric field topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 50, 928, 504]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 525, 117, 544]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 566, 905, 610]]<|/det|> +Magnetic field topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +<|ref|>image<|/ref|><|det|>[[62, 621, 936, 866]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 889, 117, 908]]<|/det|> +
Figure 4
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 920, 88]]<|/det|> +Poynting vector topology of toroidal and supertoroidal light pulses. (See Manuscript file for full figure legend) + +<|ref|>image<|/ref|><|det|>[[45, 95, 935, 620]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 636, 116, 656]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 677, 917, 721]]<|/det|> +Fractal- like pattern in electromagnetic field of supertoroidal pulses. (See Manuscript file for full figure legend) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 55, 568, 483]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 504, 116, 523]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[42, 545, 944, 589]]<|/det|> +Evolution of on- axis singularity distribution versus supertoroidal order. (See Manuscript file for full figure legend) + +<|ref|>sub_title<|/ref|><|det|>[[44, 612, 310, 639]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 663, 765, 684]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 701, 428, 826]]<|/det|> +- Video1.mp4- Video2.mp4- Video3.mp4- Video4.mp4- SupplementaryInformationtopology.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/images_list.json b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c2e4b7cd4c9bdd47d8102bb06df11e01989750a4 --- /dev/null +++ b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "FIG. 1. Device concept of the integrated comb laser. a. Conceptual illustration of the comb generation and mode-locking principle, in which electro-optic (EO) comb generation, Kerr comb generation, and broadband optical gain all work synergistically together inside a single laser cavity for on-demand generation of mode-locked soliton comb. In addition, the laser comb output is detected and fed back to the laser cavity for resonant EO modulation to realize a self-sustained operation. b. Schematic of comb laser cavity structure formed by hybrid integration between a RSOA chip and a LN external cavity chip. Two different configurations are employed: A, cavity-enhanced (CE) comb laser structure in which the LN external cavity is formed mainly by an embedded high-Q racetrack resonator together with a broadband Sagnacloop mirror; B, Fabry-Perot (FP) comb laser in which the LN external cavity is formed by an EO phase modulation section together with an a broadband Sagnac loop mirror. c. Photo of a CE comb laser, showing that the RSOA is edge-coupled to the LN external cavity chip. d. Zoom-in photo showing the edge-coupling region between the RSOA and the LN chip. e., Photo of the racetrack resonator and the loop mirror in a CE comb laser. f. Photo of the EO phase modulator and the loop mirror in an FP comb laser.", + "footnote": [], + "bbox": [ + [ + 84, + 66, + 920, + 447 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "FIG. 2. Lasing performance of a CE comb laser. a. Schematic of the experimental setup for comb laser characterization. RSOA is powered by a stable current source and an RF signal generator is used to drive the racetrack resonator. OSA: optical spectrum analyzer; AC: autocorrelator; PD: photo-detector; ESA: electrical spectrum analyzer; OSC: real-time oscilloscope; PNA: phase noise analyzer. b. Optical spectrum of the comb laser output. Off state: single mode lasing with the RF driving signal turned off. On state: comb lasing when the RF driving signal is turned on. Inset: Autocorrelation trace of the laser output pulses, in the \"comb on\" state, in which the blue curve shows the experimental data, dotted curves show the the fitted autocorrelation profiles from individual \\(\\mathrm{sech}^2\\) pulses, and the dashed curve show the overall fitted autocorrelation trace. The autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping. c. Electrical spectrum of the beat note detected from the comb laser output. The red and blue curves show the comb-on and comb-off (single-mode lasing) states, respectively, corresponding to those in b. Gray curve shows the noise background of the optical detector, as a comparison. The inset shows the detailed spectrum of the RF beat note at \\(39.58\\mathrm{GHz}\\) . d. Phase noise spectrum of the \\(39.58\\mathrm{GHz}\\) RF beat note (red) and the RF driving signal (blue). e. Phase noise spectrum of the CE comb laser output measured with a self-heterodyne method. Red and green curves show for two different comb states, and blue curve shows for the comb-off (single-mode lasing) state. Inset shows the corresponding frequency noise spectrum of the three laser states. f. P-I-V curve of the CE comb laser, in which the red and blue curves show the L-I and I-V curves, respectively. g - j. Turnkey operation of the comb laser at two different speed of \\(2\\mathrm{Hz}\\) (g, h) and \\(200\\mathrm{Hz}\\) (i, j), respectively. Red curve shows the normalized driving RF power and blue curve shows the beating signal between the comb laser output and an external reference laser at \\(1582\\mathrm{nm}\\) . h and j show the zoomed-in signal for the on/off states, respectively.", + "footnote": [], + "bbox": [ + [ + 83, + 63, + 919, + 560 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "FIG. 3. Harmonic mode-locking of a FP comb laser with an FSR of 9.817 GHz. Third harmonics (29.45 GHz) and fourth harmonics (39.27 GHz) are separately used as driving signals. a. Schematic of the experimental setup for harmonic mode-locking of the comb laser. b. Optical spectrum of the laser output with third-harmonic mode-locking, by driving the phase modulator with a RF signal at 29.45 GHz. c. Optical spectrum of the laser output with fourth-harmonic mode-locking, by driving the phase modulator with a RF signal at 39.27 GHz. In b and c, the left insets shows the electrical spectrum of the RF beat note detected from the output laser comb, and the right insets show the autocorrelation trace of the laser output pulses with dashed curve showing the fitted individual pulses. Same as Fig. 2b, autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping.", + "footnote": [], + "bbox": [ + [ + 89, + 65, + 914, + 257 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "FIG. 4. Fast frequency tuning of the whole lasing comb. a. Left panel: Schematic of the setup for comb frequency tuning, in which a triangular-waveform electrical signal produced by an arbitrary waveform generator (AWG) is used to drive the racetrack resonator of the CE comb laser together with the 39.58-GHz mode-locking RF signal. Middle panel: Conceptual illustration of the comb frequency tuning process, showing the laser frequencies of the comb are tuned together as a whole. Right panel: Schematic showing the corresponding sideband creation around the comb lines, introduced by triangular-waveform frequency modulation. b, c, d. Time-frequency spectra of the beatnote between the comb laser output and a referenced laser operating at a fixed wavelength of 1582 nm, at the modulation speed of 1, 10, and 100 MHz, respectively. The dashed curves show the corresponding triangular-waveform EO tuning signal. Bottom panels: Corresponding relative frequency deviation at each modulation speed. e, f, g. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, at modulation speed of 1 MHz (e), 10 MHz (f), and 100 MHz (g), respectively. h. Laser frequency tuning efficiency recorded at different modulation speeds.", + "footnote": [], + "bbox": [ + [ + 84, + 65, + 914, + 460 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "FIG. 5. Self-sustained operation of the CE comb laser with feedback locking. a. Schematic of self-feedback locking of the CE comb laser. b. Optical spectrum of the comb laser output at a driving current of \\(285~\\mathrm{mA}\\) . c - e. Optical spectrum (d) and optical power (e) of the comb laser output as a function of the RSOA driving current (c). f. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, with a driving current of \\(60~\\mathrm{mA}\\) . g. Phase noise spectrum of the detected 39.58-GHz beat note.", + "footnote": [], + "bbox": [ + [ + 85, + 65, + 914, + 556 + ] + ], + "page_idx": 8 + } +] \ No newline at end of file diff --git a/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b.mmd b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b.mmd new file mode 100644 index 0000000000000000000000000000000000000000..aa25b151b8d103619f96c2155d343cd01b2c160c --- /dev/null +++ b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b.mmd @@ -0,0 +1,187 @@ + +# Electrically empowered microcomb laser + +Jingwei Ling koxuix@outlook.com + +UNIVERSITY OF ROCHESTER https://orcid.org/0000- 0002- 7576- 0878 + +Zhengdong Gao UNIVERSITY OF ROCHESTER + +Shixin Xue UNIVERSITY OF ROCHESTER + +Qili Hu https://orcid.org/0000- 0002- 2362- 7936 + +Mingxiao Li UNIVERSITY OF ROCHESTER + +Kaibo Zhang UNIVERSITY OF ROCHESTER + +Usman Javid University of Maryland College Park https://orcid.org/0000- 0001- 9266- 7395 + +Raymond Lopez- Rios University of Rochester + +Jeremy Staffa University of Rochester https://orcid.org/0000- 0003- 2140- 0516 + +Qiang Lin UNIVERSITY OF ROCHESTER + +Physical Sciences - Article + +Keywords: + +Posted Date: December 14th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3518051/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on May 17th, 2024. See the published version at https://doi.org/10.1038/s41467-024-48544-2. + +<--- Page Split ---> + +# Electrically empowered microcomb laser + +Jingwei Ling, \(^{1,*}\) Zhengdong Gao, \(^{1,*}\) Shixin Xue, \(^{1}\) Qili Hu, \(^{2}\) Mingxiao Li, \(^{1}\) Kaibo Zhang, \(^{2}\) Usman A. Javid, \(^{2}\) Raymond Lopez- Rios, \(^{2}\) Jeremy Staffa, \(^{2}\) and Qiang Lin \(^{1,2, + }\) \(^{1}\) Department of Electrical and Computer Engineering, University of Rochester, Rochester, NY 14627, USA \(^{2}\) Institute of Optics, University of Rochester, Rochester, NY 14627, USA + +Optical frequency comb underpins a wide range of applications from communication, metrology, to sensing. Its development on a chip- scale platform - so called soliton microcomb - provides a promising path towards system miniaturization and functionality integration via photonic integrated circuit (PIC) technology. Although extensively explored in recent years, challenges remain in key aspects of microcomb such as complex soliton initialization, high threshold, low power efficiency, and limited comb reconfigurability. Here we present an on- chip laser that directly outputs microcomb and resolves all these challenges, with a distinctive mechanism created from synergetic interaction among resonant electro- optic effect, optical Kerr effect, and optical gain inside the laser cavity. Realized with integration between a III- V gain chip and a thin- film lithium niobate (TFLN) PIC, the laser is able to directly emit mode- locked microcomb on demand with robust turnkey operation inherently built in, with individual comb linewidth down to \(600\mathrm{Hz}\) , whole- comb frequency tuning rate exceeding \(2.4 \times 10^{17}\mathrm{Hz / s}\) , and \(100\%\) utilization of optical power fully contributing to comb generation. The demonstrated approach unifies architecture and operation simplicity, high- speed reconfigurability, and multifunctional capability enabled by TFLN PIC, opening up a great avenue towards on- demand generation of mode- locked microcomb that is expected to have profound impact on broad applications. + +21 Optical frequency comb is a coherent light source that consists of many highly coherent single- frequency laser lines 22 equally spaced in the frequency domain. Its development has 23 revolutionized many fields including metrology, spectroscopy, 24 and clock [1]. In recent years, significant interest has been 25 attracted to the generation of phase- locked optical frequency 27 comb in on- chip nonlinear microresonators [2- 4]. The super- 28 rior coherence offered by these mode- locked microcombs has 29 rendered a variety of important applications including data 30 communication [5], spectroscopic sensing [6], optical com- 31 puting [7, 8], range measurement [9- 12], optical [13] and mi- 32 crowave [14] frequency synthesis, with many others expected 33 in the years to come. + +34 Despite these great progresses, challenges remain in the development and application of microcombs. The first is the 35 difficulty in triggering comb mode- locking due to the intrinsic device nonlinearities. Recently, self- starting operation 36 has been demonstrated to address this issue [15- 18]. Their 37 implementations, however, require sophisticated system pre- 38 configuration and careful balance of specific nonlinear dynamics, which are difficult to apply in most practical devices. 39 The second is the low power efficiency of soliton microcomb 40 generation due to the pump- laser- cavity frequency detuning 41 induced by soliton pulsing. Although pulse pumping [19] 42 or auxiliary- resonator enhancement [20, 21] can improve the 43 generation efficiency, they require delicate synchronization in 44 time or resonance frequency and the difficulty of soliton ini- 45 tialization remains the same. The third is the limitation in the 46 comb controllability due to the monolithic nature of the comb 47 generator that is difficult to change after the device is fabricated. Piezoelectric effect could be used to deform the comb 48 resonator [12], which, however, exhibits limited tuning speed + +53 and efficiency due to its slow mechanical response. To date, 54 the majority of comb generators still have to rely on external 55 laser control to adjust the microcomb state. + +56 Recently, there are significant advances in chip- scale integration of semiconductor lasers and nonlinear comb generators [16, 22- 24], in which a diode laser produces single- frequency laser emission to pump a hybridly or heterogeneously integrated external nonlinear resonator to excite microcombs. Such a fully integrated system shows great promise in improving the size, weight, and power consumption. However, the nature of soliton comb generation remains essentially the same, with all the above challenges persistent. Up to now, 65 the realization of an integrated comb source free from these 66 challenges remains elusive. + +67 Here we present a fundamentally distinctive approach to resolve all these challenges in a single device. Figure 1a shows 68 the device concept. In contrast to conventional approaches 69 that rely solely on a single mechanism - either optical Kerr 70 or electro- optic effect - for comb generation while with external pumping, we utilize the resonantly enhanced electro- 71 optic (EO) modulation to initiate the comb generation, the 72 resonantly enhanced optical Kerr effect to expand the comb 73 bandwidth and phase- lock the comb lines, and the embedded 74 III- V optical gain to sustain and stabilize the comb operation. 75 Moreover, the resulting coherent microwave (via optical detection) is fed back to the EO comb to further enhance the 76 mode- locking, leading to unique self- sustained comb operation. + +81 We realize this approach by integrating a III- V gain element 82 with a thin- film lithium- niobate (LN) photonic integrated circuit (PIC) to produce a III- V/LN comb laser (Fig. 1b). LN 83 PIC has attracted significant interest recently [25- 28] for 84 a variety of applications including high- speed modulation 85 [29, 30], frequency conversion [31- 33], optical frequency 86 comb [15, 34, 35], and single- frequency lasers [36- 39]. Here, 87 we unite active EO modulation with passive four- wave mix + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
FIG. 1. Device concept of the integrated comb laser. a. Conceptual illustration of the comb generation and mode-locking principle, in which electro-optic (EO) comb generation, Kerr comb generation, and broadband optical gain all work synergistically together inside a single laser cavity for on-demand generation of mode-locked soliton comb. In addition, the laser comb output is detected and fed back to the laser cavity for resonant EO modulation to realize a self-sustained operation. b. Schematic of comb laser cavity structure formed by hybrid integration between a RSOA chip and a LN external cavity chip. Two different configurations are employed: A, cavity-enhanced (CE) comb laser structure in which the LN external cavity is formed mainly by an embedded high-Q racetrack resonator together with a broadband Sagnacloop mirror; B, Fabry-Perot (FP) comb laser in which the LN external cavity is formed by an EO phase modulation section together with an a broadband Sagnac loop mirror. c. Photo of a CE comb laser, showing that the RSOA is edge-coupled to the LN external cavity chip. d. Zoom-in photo showing the edge-coupling region between the RSOA and the LN chip. e., Photo of the racetrack resonator and the loop mirror in a CE comb laser. f. Photo of the EO phase modulator and the loop mirror in an FP comb laser.
+ +89 ing (FWM) in a dispersion- engineered high- Q laser cavity 90 for the on- demand generation of mode- locked soliton micro- 91 comb, which naturally leads to self- starting full turnkey op- 92 eration simply by turning on/off either the RF signal driving 93 the comb resonator or the electric current driving the gain el- 94 ement. As the comb modes extract energy directly from ma- 95 terial gain, \(100\%\) of the optical power contributes directly to 96 the comb generation. Moreover, the strong electro- optic ef- 97 fect of the LN cavity enables high tunability and reconfigura- 98 bility of the produced microcomb. With this approach, we 99 are able to produce broadband highly coherent microcombs, 100 with individual comb linewidth down to \(600\mathrm{Hz}\) , frequency 101 tunability of over \(2.4\times 10^{17}\mathrm{Hz / s}\) for the entire microcomb, 102 microwave phase noise down to - 115 dBc/Hz at \(500\mathrm{kHz}\) fre- 103 quency offset, and a wall- plug efficiency exceeding \(5.6\%\) . The 104 simplicity of the demonstrated approach opens up a new path 105 for on- demand generation of mode- locked microcombs that is 106 expected to have profound impact on the broad applications in 107 high- precision metrology, telecommunications, remote sens- 108 ing, clocking, computing, and beyond. + +## Comb laser structure design + +The III- V/LN microcomb laser is formed by integrating an InP reflective semiconductor optical amplifier (RSOA) with an LN external cavity chip via facet- to- facet coupling. We employ two types of laser cavity structures for the purpose, as shown in Fig. 1b. Chip A laser structure is embedded with a dispersion- engineered EO microresonator and Chip B laser structure consists of a simple EO phase modulation waveguide section. The advantage of Chip- A resonator- type structure is that the high- Q microresonator offers strong cavity enhancement for comb generation and mode- locking, which we term as the cavity- enhanced (CE) comb laser. The benefit of the Chip- B Fabry- Perot- type structure is that it offers flexibility of mode- locking operation, which we term as the Fabry- Perot (FP) comb laser. + +In the CE comb laser, the group- velocity dispersion (GVD) of the racetrack microresonator (intrinsic optical \(Q \sim 1.6 \times 10^{6}\) ) is engineered to be small but slightly anomalous to support broadband comb generation. At the same time, the pulley + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
FIG. 2. Lasing performance of a CE comb laser. a. Schematic of the experimental setup for comb laser characterization. RSOA is powered by a stable current source and an RF signal generator is used to drive the racetrack resonator. OSA: optical spectrum analyzer; AC: autocorrelator; PD: photo-detector; ESA: electrical spectrum analyzer; OSC: real-time oscilloscope; PNA: phase noise analyzer. b. Optical spectrum of the comb laser output. Off state: single mode lasing with the RF driving signal turned off. On state: comb lasing when the RF driving signal is turned on. Inset: Autocorrelation trace of the laser output pulses, in the "comb on" state, in which the blue curve shows the experimental data, dotted curves show the the fitted autocorrelation profiles from individual \(\mathrm{sech}^2\) pulses, and the dashed curve show the overall fitted autocorrelation trace. The autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping. c. Electrical spectrum of the beat note detected from the comb laser output. The red and blue curves show the comb-on and comb-off (single-mode lasing) states, respectively, corresponding to those in b. Gray curve shows the noise background of the optical detector, as a comparison. The inset shows the detailed spectrum of the RF beat note at \(39.58\mathrm{GHz}\) . d. Phase noise spectrum of the \(39.58\mathrm{GHz}\) RF beat note (red) and the RF driving signal (blue). e. Phase noise spectrum of the CE comb laser output measured with a self-heterodyne method. Red and green curves show for two different comb states, and blue curve shows for the comb-off (single-mode lasing) state. Inset shows the corresponding frequency noise spectrum of the three laser states. f. P-I-V curve of the CE comb laser, in which the red and blue curves show the L-I and I-V curves, respectively. g - j. Turnkey operation of the comb laser at two different speed of \(2\mathrm{Hz}\) (g, h) and \(200\mathrm{Hz}\) (i, j), respectively. Red curve shows the normalized driving RF power and blue curve shows the beating signal between the comb laser output and an external reference laser at \(1582\mathrm{nm}\) . h and j show the zoomed-in signal for the on/off states, respectively.
+ +128 coupling regions are specially designed for uniform close- to- 129 critical coupling to the resonator over a broad telecom band. 130 Such a design ensures high loaded optical \(\mathrm{Q}(\sim 5\times 10^{5})\) of 131 the resonator uniformly across a wide spectral range which is 132 crucial both for enhancing the comb generation and mode- 133 locking and for efficient light coupling into/out of the mi- 134 crosensorator. Moreover, we also engineer the GVD of straight + +135 waveguide sections outside the racetrack resonator to compensate for that of the RSOA section so as to minimize the overall 136 laser cavity GVD. At the same time, the overall optical path 137 length of the laser cavity is designed to be an integer multiple of the racetrack resonator's for matching their resonance 138 mode frequencies and round-trip group delay. The GVD of 139 the FP comb laser is engineered in a similar fashion. The free + +<--- Page Split ---> + +spectral range (FSR) of the racetrack resonator is designed to be around \(40\mathrm{GHz}\) to better accommodate bandwidths of RF filter and amplifier after detection of comb beating signal. In contrast, the FP comb laser is designed to have a small FSR of around \(10\mathrm{GHz}\) for easy operation of harmonic mode locking. To support the broadband operation, the Sagnac loop mirror employs an adiabatic coupling design [40] to achieve high reflection and feedback with a reflectivity of \(>95\%\) over a broad spectral band of \(1500 - 1600\mathrm{nm}\) . On the other hand, a horn taper waveguide is designed on the LN chip to minimize the coupling loss between the LN chip and the RSOA gain chip. The RSOA exhibits a broadband gain in the telecom L- band, with a 3- dB bandwidth of \(>40\mathrm{nm}\) . Fig. 1c- f show the device structures. Details about the design parameters and the characterization of the laser structures are provided in the Supplementary Information (SI). + +## Comb laser performance + +To excite the mode- locked microcomb, we first launch a single- frequency RF signal to drive the racetrack resonator of the CE comb laser (Fig. 2a), with a frequency of \(39.58\mathrm{GHz}\) that matches its FSR. Before the RF signal is applied, the device exhibits single- mode or multi- mode lasing, with an example shown in the blue curve of Fig. 2b. However, A microcomb is readily produced as soon as the RF signal is applied, with an optical spectral bandwidth of about \(20\mathrm{nm}\) (Fig. 2b, red curve). Mode- locking of the comb is verified by the clean RF tone at \(39.58\mathrm{GHz}\) detected from the beating between comb lines (Fig. 2c), as well as the autocorrelation trace from the laser output pulses (Fig. 2b, inset). The \(39.58 - \mathrm{GHz}\) RF tone exhibits a high signal- to- noise ratio of \(79\mathrm{dB}\) (Fig. 2c, inset), whose phase noise spectrum matches identically the driving RF source (Fig. 2d), showing the preservation of the relative phase coherence between comb lines via mode- locking. Mode- locking of the comb is also clearly evident by the clean noise floor around the DC region in the RF spectrum (Fig. 2c. See SI for details), where the zero extra noise from the mode- locked comb state infers that all the comb lines of the entire comb are phase- locked together. + +The underlying mechanism responsible for mode- locking dominantly contributes from the combined resonant EO modulation and optical Kerr effect, in which the EO modulation produces EO sidebands to initiate the comb generation while the optical Kerr effect broadens the comb spectrum and phase- locks the comb lines (Fig. 1a). Indeed, the laser is able to produce mode- locked soliton pulses in the absence of EO modulation (while with a narrower spectrum), in which only the optical Kerr effect is responsible for mode- locking. The detailed theoretical modeling and testing results are provided in SI. In Fig. 2c, the small RF tone around the half- harmonic at \(19.79\mathrm{GHz}\) indicates certain comb dynamics. It can be eliminated by reconfiguring the laser and one example is shown in SI which exhibits a clean single RF beating tone and a well- defined \(\mathrm{sech}^2\) - shaped soliton pulse spectrum. The two lasers mainly differ in their overall dispersion of the laser cavity, indicating that the device dispersion plays an important role on the comb spectrum. The two- sidelobe feature of the comb spectrum in Fig. 2b implies that the output pulses are likely + +to be mode- locked two- color pulses in which the two color pulses bounds with each other via certain interpulse interaction [41]. Its exact nature, however, will require further exploration. The comb spectrum can be reconfigured by changing the power or frequency of the RF driving signal, whose details are provided in the SI. + +In addition to the high coherence between the comb lines, the comb laser also exhibits narrow linewidth on its individual comb lines. To show this feature, we employ the correlated self- heterodyne method [42, 43] to characterize the overall linewidth of the whole comb laser by launching the entire comb for linewidth measurement (rather than characterizing individual comb lines themselves) (See SI for details). The recorded phase noise spectrum is shown in Fig. 2e, which indicates a white frequency- noise floor of \(\sim 350\mathrm{Hz}^2 /\mathrm{Hz}\) (Fig. 2e, inset) that corresponds to a laser linewidth of \(\sim 2\mathrm{kHz}\) . The linewidth of the comb lines can be decreased further and an example is shown in Fig. 2e for a slightly different comb state produced from the same laser, which exhibits a white frequency- noise floor of \(\sim 100\mathrm{Hz}^2 /\mathrm{Hz}\) (Fig. 2e, inset) that corresponds to a laser linewidth as low as \(\sim 600\mathrm{Hz}\) . Note that these values represent the overall linewidth contributed from the entire comb, which indicates the upper limit of the intrinsic linewidth of individual comb lines. + +Figure 2f shows the current- dependent characteristics of the comb laser, which exhibits a low threshold current of \(50\mathrm{mA}\) , indicating the low overall loss of the integrated laser. The comb laser produces an optical output power of \(11\mathrm{mW}\) at a pumping current of \(275\mathrm{mA}\) and a pumping voltage of \(1.4\mathrm{V}\) , which corresponds to a wall- plug efficiency of \(2.8\%\) . As the laser has two output ports (Fig. 1b) that emit the same amount of optical power, the total wall- plug efficiency of the laser is thus \(5.6\%\) . This level of wall- plug efficiency is on par with other integrated external- cavity semiconductor lasers recently developed [44, 45]. Intriguingly, the comb power increases with increased driving RF power, whose details are provided in the SI. Note that the total optical power contributes fully to the generated comb, in strong contrast to conventional Kerr solitons or EO combs in which the major optical power remains in the residual pump wave with low comb generation efficiency. + +A distinctive feature of the comb laser is that the produced comb can be switched on/off on demand by simply switching on/off the driving RF signal. To show this feature, we beat the comb with a reference single- frequency laser operating at the wavelength of \(1582\mathrm{nm}\) that is inside the comb spectrum, and monitor the beating signal with the RF driving signal being turned on/off. As shown in Fig. 2g and i, the beating signal follows faithfully the driving RF signal. The coherent beating signal shows up readily when the RF driving is on, indicating the generation of the mode- locked comb. The beating signal disappears right after the RF driving is off, indicating the shut- off of the comb state. Same phenomenon is observed when the reference laser is tuned to other wavelengths within the comb spectrum. + +Similar phenomena are observed in the FP comb laser, while generally with smaller spectral extents due to the lack + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
FIG. 3. Harmonic mode-locking of a FP comb laser with an FSR of 9.817 GHz. Third harmonics (29.45 GHz) and fourth harmonics (39.27 GHz) are separately used as driving signals. a. Schematic of the experimental setup for harmonic mode-locking of the comb laser. b. Optical spectrum of the laser output with third-harmonic mode-locking, by driving the phase modulator with a RF signal at 29.45 GHz. c. Optical spectrum of the laser output with fourth-harmonic mode-locking, by driving the phase modulator with a RF signal at 39.27 GHz. In b and c, the left insets shows the electrical spectrum of the RF beat note detected from the output laser comb, and the right insets show the autocorrelation trace of the laser output pulses with dashed curve showing the fitted individual pulses. Same as Fig. 2b, autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping.
+ +257 of cavity enhancement. The FP comb laser, however, exhibits a distinctive feature in that it can be flexibly mode- locked at higher harmonics of the laser cavity FSR. Fig. 3 shows this feature. We are able to achieve third- and fourth- order harmonic mode- locking by applying an RF signal to the phase modulation section of the FP comb laser, with a frequency of 29.45 and 39.27 GHz, respectively, that are three and four times of the laser FSR (9.817 GHz). Again, mode locking of the combs is clearly verified by the detected RF beating signal from the combs with a SNR of 77 dB, as well as by the autocorrelation traces from the laser output pulses (Fig. 3b,c, insets). + +## High-speed frequency tuning of the comb laser + +Another distinctive characteristic of the comb laser is that the laser frequencies of the entire mode- locked comb can be tuned cohesively at an extremely high speed. To show this feature, we apply a triangular- waveform electric signal - to- gether with the 39.58- GHz RF driving signal - to the racetrack resonator of the CE comb laser as shown in Fig. 4a. While the 39.58- GHz RF driving signal supports the mode- locking process, the triangular- waveform electric signal will adiabatically tune the resonance frequencies of the racetrack resonator, thus tuning the laser frequencies of the entire mode- locked comb together as a whole. + +To show this feature, we beat the comb with a narrow- linewidth reference CW laser at 1582 nm that is about 15 GHz away from a comb line, and monitor the beating signal in real time. At the same time, we monitor the spectrum of the recorded 39.58- GHz RF tone from the beating between the comb lines (See SI for details of the setup). The frequency dynamics of the 15- GHz beating signal with the reference laser show the frequency tuning of the comb line nearby while the 39.58- GHz RF tone from the comb line beating indicates the quality of mode locking during the frequency tuning. Fig. 4f- h show the temporal variation of the 15- GHz beating signal at + +different modulation speeds of 1, 10, 100 MHz. They show clearly that the frequency tuning of the comb line follows faithfully the waveform of the driving triangular- waveform electric signal at all modulation speeds, with a deviation of no more than \(5\%\) . In particular, the recorded 39.58- GHz RF tone from the comb line beating (Fig. 4e- g) remains unchanged during the frequency tuning, except with created modulation sidebands that simply results from the laser frequency modulation (see also Fig. 4a, right figure). This observation confirms that the phase- locking between the comb modes is fully preserved during the high- speed frequency tuning process, indicating that the entire mode- locked comb is tuned in its frequencies as a whole, without any perturbation to the comb mode spacing. This is in strong contrast to other comb modulation approaches [11, 12, 46] where the comb mode spacing is seriously impacted by external modulation. The frequency tuning range of 1.2 GHz at the modulation speed of 100 MHz (Fig. 4h) corresponds to a frequency tuning rate as high as \(2.4 \times 10^{17} \mathrm{~Hz} / \mathrm{s}\) for the comb. Both the frequency tuning rate and tuning speed are orders of magnitudes higher than the piezoelectric tuning and the external pump modulation approaches [11, 12], which are constrained only by the photon lifetime of the high- Q racetrack resonator. As shown in Fig. 4e, the device exhibits a frequency tuning efficiency of about \(0.2 - 0.8 \mathrm{GHz} / \mathrm{V}\) depending on the modulation speed, which is more than an order of magnitude higher than the piezoelectric approach [12]. The tuning efficiency can be further doubled by employing both sets of driving electrodes of the racetrack resonator (Fig. 4a). + +## Feedback mode-locking of the comb laser + +So far, the comb laser utilizes an external RF signal to support the mode- locking. This signal, however, can be removed by feeding the coherent 39.58- GHz RF tone detected from the comb mode beating directly back to the comb laser cavity to sustain the mode- locking, resulting in unique stand- alone self + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
FIG. 4. Fast frequency tuning of the whole lasing comb. a. Left panel: Schematic of the setup for comb frequency tuning, in which a triangular-waveform electrical signal produced by an arbitrary waveform generator (AWG) is used to drive the racetrack resonator of the CE comb laser together with the 39.58-GHz mode-locking RF signal. Middle panel: Conceptual illustration of the comb frequency tuning process, showing the laser frequencies of the comb are tuned together as a whole. Right panel: Schematic showing the corresponding sideband creation around the comb lines, introduced by triangular-waveform frequency modulation. b, c, d. Time-frequency spectra of the beatnote between the comb laser output and a referenced laser operating at a fixed wavelength of 1582 nm, at the modulation speed of 1, 10, and 100 MHz, respectively. The dashed curves show the corresponding triangular-waveform EO tuning signal. Bottom panels: Corresponding relative frequency deviation at each modulation speed. e, f, g. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, at modulation speed of 1 MHz (e), 10 MHz (f), and 100 MHz (g), respectively. h. Laser frequency tuning efficiency recorded at different modulation speeds.
+ +327 sustained comb lasing operation. Figure 5a illustrates this approach. The comb laser output is detected by a high- speed optical detector whose output RF signal is amplified to an adequate amplitude, filtered to suppress excess low- frequency noises, adjusted with appropriate phase, and then fed back to drive the racetrack resonator of the CE comb laser. + +328 As shown in Fig. 5b, a broadband microcomb with a spectrum covering about \(50~\mathrm{nm}\) and an optical power of \(8.5\mathrm{mW}\) is produced on chip with a driving current of \(285~\mathrm{mA}\) . Indeed, the microcomb is readily produced on demand as soon as the driving current is turned on, with a driving current as low as \(60~\mathrm{mA}\) , as shown in Fig. 5c- e. Mode locking of the comb is clearly evident by the clean 39.58- GHz RF tone detected from the comb mode beating (Fig. 5f), which exhibits a SNR of \(65~\mathrm{dB}\) and a narrow 3- dB linewidth of \(1.5\mathrm{kHz}\) . The phase noise of the RF beating signal reaches a level of - 90 dBc/Hz at an offset frequency of \(60~\mathrm{kHz}\) , which is considerably lower than the laser heterodyne beating approach [47] and is comparable to that of free- running optical Kerr soliton microcombs [46, 48, 49]. The optical spectral bandwidth and + +329 the output power of the comb laser increase considerably with increased driving current (Fig. 5d). This is expected since the increased optical power of the mode- locked comb inside the high- Q racetrack resonator would significantly enhance the optical Kerr effect and the resulting four- wave- mixing process to broaden the comb spectrum. No saturation is observed on the comb spectral bandwidth as the current increases. + +## Discussion + +330 The attainable extent of the microcomb spectrum or soliton pulse width in current devices is primarily limited by the available optical power inside the cavity and the group delay mismatch between the enhancing resonator and the main laser cavity. For the former, it can be improved by either reducing the loss (e.g., improving the RSOA- LN chip coupling efficiency) or increasing the optical gain (e.g., using a higher- power RSOA) inside the laser cavity. For the latter, our theoretical modeling (see SI) shows that the formation of ideal ultrashort soliton pulses would require that the roundtrip time + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
FIG. 5. Self-sustained operation of the CE comb laser with feedback locking. a. Schematic of self-feedback locking of the CE comb laser. b. Optical spectrum of the comb laser output at a driving current of \(285~\mathrm{mA}\) . c - e. Optical spectrum (d) and optical power (e) of the comb laser output as a function of the RSOA driving current (c). f. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, with a driving current of \(60~\mathrm{mA}\) . g. Phase noise spectrum of the detected 39.58-GHz beat note.
+ +366 of the main laser cavity be integer times that of the enhanc- 367 ing racetrack resonator. In current devices, however, there is a 368 certain amount of mismatch which limits the comb spectrum 369 and the coherence of the mode- beating RF tone. This prob- 370 lem can be resolved by further optimization of the roundtrip 371 length of the main laser cavity and introducing tunability, after 372 which we expect that ultra- broadband highly coherent soliton 373 microcomb can be produced. + +374 To conclude, we have introduced a new type of chip- scale 375 microcomb laser that can be flexibly mode- locked with ei- 376 ther active- driving or passive- feedback approaches and that 377 can be tuned/reconfigured at an ultrafast speed, with robust 378 turnkey operation inherently built in. The demonstrated inte- 379 grated comb laser exhibits outstanding reconfigurability and 380 performance significantly beyond the reach of conventional 381 on- chip mode- locked semiconductor lasers [50- 53]. The 382 demonstrated devices combine elegantly the simplicity of in- 383 tegrated laser structure, robustness of mode- locking opera- + +tion, and electro- optically enhanced tunability and controllability, opening up a new avenue towards on- demand generation of soliton microcombs with high power efficiency that we envision to be of great promise for a wide range of applications including ranging, communication, optical and microwave synthesis, sensing, metrology, among many others. + +## Method + +Device fabrication The device fabrication begins with a con- 392 gruent x- cut thin film lithium- niobate- on- insulator (LNOI) wafer, with a \(600~\mathrm{nm}\) LN layer on a \(4.7~\mu \mathrm{m}\) silica- coated silicon substrate. E- beam lithography (EBL) and Ar- ion milling are used to etch the waveguide with ZEP- 520A as mask. Etching thickness ranges from \(300~\mathrm{nm}\) (CE comb laser) to \(350~\mathrm{nm}\) (FP comb laser) for dispersion engineering. Second EBL is applied on PMMA for deposition of \(400~\mathrm{nm}\) gold- evaporated electrodes, which are placed \(2.5~\mu \mathrm{m}\) from the waveguide. The distance between the waveguide and electrode is chosen to balance the EO modulation frequency + +<--- Page Split ---> + +402 with loss from metal absorption. Dicing and polishing of LN 404 and amplifier- to- chip coupling, with both coupling losses 403 chip are employed at last to acquire optimized fiber- to- chip 405 around 6 dB. + +407 [1] Diddams, S. A., Vahala, K. & Udem, T. Optical frequency 408 combs: Coherently uniting the electromagnetic spectrum. Sci- 409 ence 369, eaay3676 (2020). 410 [2] Kippenberg, T. J., Gaeta, A. L., Lipson, M. & Gorodetsky, 411 M. L. Dissipative Kerr solitons in optical microresonators. Sci- 412 ence 361, eaan8083 (2018). 413 [3] Gaeta, A. L., Lipson, M. & Kippenberg, T. J. Photonic- chip- 414 based frequency combs. Nature Photonics 13, 158- 169 (2019). 415 [4] Chang, L., Liu, S. & Bowers, J. E. Integrated optical frequency 416 comb technologies. Nature Photonics 16, 95- 108 (2022). 417 [5] Marin- Palomo, P. et al. Microresonator- based solitons for mas- 418 sively parallel coherent optical communications. Nature 546, 274- 279 (2017). 419 [6] Suh, M.- G., Yang, Q.- F., Yang, K. Y., Yi, X. & Vahala, K. J. 420 Microresonator soliton dual- comb spectroscopy. Science 354, 600- 603 (2016). 421 [7] Xu, X. et al. 11 TOPS photonic convolutional accelerator for 422 optical neural networks. Nature 589, 44- 51 (2021). 423 [8] Feldmann, J. et al. Parallel convolutional processing using an 424 integrated photonic tensor core. Nature 589, 52- 58 (2021). 425 [9] Suh, M.- G. & Vahala, K. J. Soliton microcomb range measure- 426 ment. Science 359, 884- 887 (2018). 427 [10] Trocha, P. et al. Ultrafast optical ranging using microresonator 428 soliton frequency combs. Science 359, 887- 891 (2018). 429 [11] Riemensberger, J. et al. Massively parallel coherent laser rang- 430 ing using a soliton microcomb. Nature 581, 164- 170 (2020). 431 [12] Liu, J. et al. Monolithic piezoelectric control of soliton micro- 432 combs. Nature 583, 385- 390 (2020). 433 [13] Spencer, D. T. et al. An optical- frequency synthesizer using 434 integrated photonics. Nature 557, 81- 85 (2018). 435 [14] Liang, W. et al. High spectral purity Kerr frequency comb radio 436 frequency photonic oscillator. Nature Communications 6, 7957 437 (2015). 438 [15] He, Y. et al. Self- starting bi- chromatic LiNbO3 soliton micro- 439 comb. Optica 6, 1138- 1144 (2019). 440 [16] Shen, B. et al. Integrated turnkey soliton microcombs. Nature 441 582, 365- 369 (2020). 442 [17] Bai, Y. et al. Brillouin- Kerr soliton frequency combs in an 443 optical microresonator. Physical Review Letters 126, 063901 444 (2021). 445 [18] Rowley, M. et al. Self- emergence of robust solitons in a micro- 446 cavity. Nature 608, 303- 309 (2022). 447 [19] Obrzud, E., Lecomte, S. & Herr, T. Temporal solitons in mi- 448 crosenators driven by optical pulses. Nature Photonics 11, 450 600- 607 (2017). 451 [20] Xue, X., Zheng, X. & Zhou, B. Super- efficient temporal soli- 452 tons in mutually coupled optical cavities. Nature Photonics 13, 453 616- 622 (2019). 454 [21] Bao, H. et al. Laser cavity- soliton microcombs. Nature Pho- 455 tonics 13, 384- 389 (2019). 456 [22] Stern, B., Ji, X., Okawachi, Y., Gaeta, A. L. & Lipson, M. 457 Battery- operated integrated frequency comb generator. Nature 458 562, 401- 405 (2018). 459 [23] Raja, A. S. et al. Electrically pumped photonic integrated soli- 460 ton microcomb. Nature Communications 10, 680 (2019). 461 [24] Xiang, C. et al. Laser soliton microcombs heterogeneously in- + +tegrated on silicon. Science 373, 99- 103 (2021). 425 [25] Boes, A., Corcoran, B., Chang, L., Bowers, J. & Mitchell, A. Status and potential of lithium niobate on insulator (LNOI) for photonic integrated circuits. Laser & Photonics Reviews 12, 1700256 (2018). 426 [26] Lin, J., Bo, F., Cheng, Y. & Xu, J. Advances in on- chip photonic devices based on lithium niobate on insulator. Photonics Research 8, 1910- 1936 (2020). 427 [27] Zhu, D. et al. Integrated photonics on thin- film lithium niobate. Advances in Optics and Photonics 13, 242- 352 (2021). 428 [28] Boes, A. et al. Lithium niobate photonics: Unlocking the electromagnetic spectrum. Science 379, eabj4396 (2023). 429 [29] Wang, C. et al. Integrated lithium niobate electro- optic modulators operating at CMOS- compatible voltages. Nature 562, 101- 104 (2018). 430 [30] He, M. et al. High- performance hybrid silicon and lithium niobate Mach- Zehnder modulators for 100 Gbit s- 1 and beyond. Nature Photonics 13, 359- 364 (2019). 431 [31] Wang, C. et al. Ultrahigh- efficiency wavelength conversion in nanophotonic periodically poled lithium niobate waveguides. Optica 5, 1438- 1441 (2018). 432 [32] Lu, J., Li, M., Zou, C.- L., Al Sayem, A. & Tang, H. X. Toward 1% single- photon anharmonicity with periodically poled lithium niobate microring resonators. Optica 7, 1654- 1659 (2020). 433 [33] McKenna, T. P. et al. Ultra- low- power second- order nonlinear optics on a chip. Nature Communications 13, 4532 (2022). 434 [34] Zhang, M. et al. Broadband electro- optic frequency comb generation in a lithium niobate microring resonator. Nature 568, 373- 377 (2019). 435 [35] Gong, Z., Liu, X., Xu, Y. & Tang, H. X. Near- octave lithium niobate soliton microcomb. Optica 7, 1275- 1278 (2020). 436 [36] de Beeck, C. O. et al. III/V- on- lithium niobate amplifiers and lasers. Optica 8, 1288- 1289 (2021). 437 [37] Han, Y. et al. Electrically pumped widely tunable O- band hybrid lithium niobite/III- V laser. Optics Letters 46, 5413- 5416 (2021). 438 [38] Li, M. et al. Integrated Pockels laser. Nature Communications 13, 5344 (2022). 439 [39] Ling, J. et al. Self- injection locked frequency conversion laser. Laser & Photonics Reviews 2200663 (2023). 440 [40] Ramadan, T. A. & Osgood, R. M. Adiabatic couplers: design rules and optimization. Journal of lightwave technology 16, 277 (1998). 441 [41] Liao, R. et al. Dual- comb generation from a single laser source: principles and spectroscopic applications towards mid- IR- A review. Journal of Physics: Photonics 2, 042006 (2020). 442 [42] Camatel, S. & Ferrero, V. Narrow linewidth CW laser phase noise characterization methods for coherent transmission system applications. Journal of Lightwave Technology 26, 3048- 3055 (2008). 443 [43] Yuan, Z. et al. Correlated self- heterodyne method for ultralow- noise laser linewidth measurements. Optics Express 30, 25147- 25161 (2022). 444 [44] Xiang, C. et al. High- performance silicon photonics using heterogeneous integration. IEEE Journal of Selected Topics in + +<--- Page Split ---> + +519 Quantum Electronics 28, 1- 15 (2021). 520 [45] Porter, C., Zeng, S., Zhao, X. & Zhu, L. Hybrid integrated chip- scale laser systems. APL Photonics 8 (2023). 521 [46] He, Y. et al. High- speed tunable microwave- rate soliton micro- comb. Nature Communications 14, 3467 (2023). 522 [47] Xiang, C. et al. 3d integration enables ultralow- noise isolator- free lasers in silicon photonics. Nature 620, 78- 85 (2023). 523 [48] Liu, J. et al. Photonic microwave generation in the X- and K- band using integrated soliton microcombs. Nature Photonics 14, 486- 491 (2020). 524 [49] Wang, B. et al. Towards high- power, high- coherence, integrated photonic mmwave platform with microcavity solitons. Light: Science & Applications 10, 4 (2021). 525 [50] Wang, Z. et al. A III- V- on- Si ultra- dense comb laser. Light: Science & Applications 6, e16260- e16260 (2017). 526 [51] Davenport, M. L., Liu, S. & Bowers, J. E. Integrated heterogeneous silicon/III- V mode- locked lasers. Photonics Research 6, 468- 478 (2018). 527 [52] Malinowski, M. et al. Towards on- chip self- referenced frequency- comb sources based on semiconductor mode- locked lasers. Micromachines 10, 391 (2019). 528 [53] Hermans, A., Van Gasse, K. & Kuyken, B. On- chip optical comb sources. APL Photonics 7 (2022). + +## Acknowledgements + +The authors would like to thank Lin Chang and William Renninger for helpful discussions. This work is supported in part by the Defense Advanced Research Projects Agency + +(DARPA) QuICC program under Agreement No. FA8650- 23- C- 7312 and LUMOS program under Agreement No. HR001- 20- 2- 0044, and the National Science Foundation (NSF) under Grant No. OMA- 2138174 and ECCS- 2231036. This work was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure (National Science Foundation, ECCS- 1542081); and at the Cornell Center for Materials Research (National Science Foundation, Grant No. DMR- 1719875). + +## Author Contributions + +J.L. and Z.G. designed and fabricated the devices. J.L. and Z.G. performed the device characterization. S.X. and Q.H. assisted in the device fabrication. S.X., K.Z., U.J., R.L. and J.S. assisted in experiments. J.L., Z.G., and Q.L. wrote the manuscript with contributions from all authors. Q.L. supervised the project. Q.L. conceived the concept. + +## Additional Informations + +Supplementary information Supplementary Information is available for this paper. Competing interests. The authors declare no competing interests. Correspondence and requests for materials should be sent to Q.L. (qiang.lin@rochester.edu). Reprints and permission. Reprints and permissions are available at www.nature.com/reprints. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformationforsolitonlasing.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b_det.mmd b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..c353c85c19f411e791eb3e3a55036d21b2b8e541 --- /dev/null +++ b/preprint/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b/preprint__12696181428465838111cfb9bc03699d1113262f9765d95778f49019f4af4f2b_det.mmd @@ -0,0 +1,254 @@ +<|ref|>title<|/ref|><|det|>[[45, 108, 755, 142]]<|/det|> +# Electrically empowered microcomb laser + +<|ref|>text<|/ref|><|det|>[[44, 162, 258, 207]]<|/det|> +Jingwei Ling koxuix@outlook.com + +<|ref|>text<|/ref|><|det|>[[44, 234, 670, 254]]<|/det|> +UNIVERSITY OF ROCHESTER https://orcid.org/0000- 0002- 7576- 0878 + +<|ref|>text<|/ref|><|det|>[[44, 260, 312, 300]]<|/det|> +Zhengdong Gao UNIVERSITY OF ROCHESTER + +<|ref|>text<|/ref|><|det|>[[44, 307, 312, 346]]<|/det|> +Shixin Xue UNIVERSITY OF ROCHESTER + +<|ref|>text<|/ref|><|det|>[[44, 353, 400, 393]]<|/det|> +Qili Hu https://orcid.org/0000- 0002- 2362- 7936 + +<|ref|>text<|/ref|><|det|>[[44, 399, 312, 439]]<|/det|> +Mingxiao Li UNIVERSITY OF ROCHESTER + +<|ref|>text<|/ref|><|det|>[[44, 446, 312, 485]]<|/det|> +Kaibo Zhang UNIVERSITY OF ROCHESTER + +<|ref|>text<|/ref|><|det|>[[44, 492, 727, 532]]<|/det|> +Usman Javid University of Maryland College Park https://orcid.org/0000- 0001- 9266- 7395 + +<|ref|>text<|/ref|><|det|>[[44, 538, 260, 577]]<|/det|> +Raymond Lopez- Rios University of Rochester + +<|ref|>text<|/ref|><|det|>[[44, 584, 620, 624]]<|/det|> +Jeremy Staffa University of Rochester https://orcid.org/0000- 0003- 2140- 0516 + +<|ref|>text<|/ref|><|det|>[[44, 630, 312, 670]]<|/det|> +Qiang Lin UNIVERSITY OF ROCHESTER + +<|ref|>text<|/ref|><|det|>[[44, 712, 275, 731]]<|/det|> +Physical Sciences - Article + +<|ref|>text<|/ref|><|det|>[[44, 750, 137, 768]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 787, 348, 806]]<|/det|> +Posted Date: December 14th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 825, 475, 844]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3518051/v1 + +<|ref|>text<|/ref|><|det|>[[44, 863, 914, 904]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 923, 535, 943]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 911, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 17th, 2024. See the published version at https://doi.org/10.1038/s41467-024-48544-2. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[330, 63, 673, 80]]<|/det|> +# Electrically empowered microcomb laser + +<|ref|>text<|/ref|><|det|>[[180, 94, 820, 160]]<|/det|> +Jingwei Ling, \(^{1,*}\) Zhengdong Gao, \(^{1,*}\) Shixin Xue, \(^{1}\) Qili Hu, \(^{2}\) Mingxiao Li, \(^{1}\) Kaibo Zhang, \(^{2}\) Usman A. Javid, \(^{2}\) Raymond Lopez- Rios, \(^{2}\) Jeremy Staffa, \(^{2}\) and Qiang Lin \(^{1,2, + }\) \(^{1}\) Department of Electrical and Computer Engineering, University of Rochester, Rochester, NY 14627, USA \(^{2}\) Institute of Optics, University of Rochester, Rochester, NY 14627, USA + +<|ref|>text<|/ref|><|det|>[[175, 164, 829, 349]]<|/det|> +Optical frequency comb underpins a wide range of applications from communication, metrology, to sensing. Its development on a chip- scale platform - so called soliton microcomb - provides a promising path towards system miniaturization and functionality integration via photonic integrated circuit (PIC) technology. Although extensively explored in recent years, challenges remain in key aspects of microcomb such as complex soliton initialization, high threshold, low power efficiency, and limited comb reconfigurability. Here we present an on- chip laser that directly outputs microcomb and resolves all these challenges, with a distinctive mechanism created from synergetic interaction among resonant electro- optic effect, optical Kerr effect, and optical gain inside the laser cavity. Realized with integration between a III- V gain chip and a thin- film lithium niobate (TFLN) PIC, the laser is able to directly emit mode- locked microcomb on demand with robust turnkey operation inherently built in, with individual comb linewidth down to \(600\mathrm{Hz}\) , whole- comb frequency tuning rate exceeding \(2.4 \times 10^{17}\mathrm{Hz / s}\) , and \(100\%\) utilization of optical power fully contributing to comb generation. The demonstrated approach unifies architecture and operation simplicity, high- speed reconfigurability, and multifunctional capability enabled by TFLN PIC, opening up a great avenue towards on- demand generation of mode- locked microcomb that is expected to have profound impact on broad applications. + +<|ref|>text<|/ref|><|det|>[[70, 383, 488, 572]]<|/det|> +21 Optical frequency comb is a coherent light source that consists of many highly coherent single- frequency laser lines 22 equally spaced in the frequency domain. Its development has 23 revolutionized many fields including metrology, spectroscopy, 24 and clock [1]. In recent years, significant interest has been 25 attracted to the generation of phase- locked optical frequency 27 comb in on- chip nonlinear microresonators [2- 4]. The super- 28 rior coherence offered by these mode- locked microcombs has 29 rendered a variety of important applications including data 30 communication [5], spectroscopic sensing [6], optical com- 31 puting [7, 8], range measurement [9- 12], optical [13] and mi- 32 crowave [14] frequency synthesis, with many others expected 33 in the years to come. + +<|ref|>text<|/ref|><|det|>[[70, 574, 488, 850]]<|/det|> +34 Despite these great progresses, challenges remain in the development and application of microcombs. The first is the 35 difficulty in triggering comb mode- locking due to the intrinsic device nonlinearities. Recently, self- starting operation 36 has been demonstrated to address this issue [15- 18]. Their 37 implementations, however, require sophisticated system pre- 38 configuration and careful balance of specific nonlinear dynamics, which are difficult to apply in most practical devices. 39 The second is the low power efficiency of soliton microcomb 40 generation due to the pump- laser- cavity frequency detuning 41 induced by soliton pulsing. Although pulse pumping [19] 42 or auxiliary- resonator enhancement [20, 21] can improve the 43 generation efficiency, they require delicate synchronization in 44 time or resonance frequency and the difficulty of soliton ini- 45 tialization remains the same. The third is the limitation in the 46 comb controllability due to the monolithic nature of the comb 47 generator that is difficult to change after the device is fabricated. Piezoelectric effect could be used to deform the comb 48 resonator [12], which, however, exhibits limited tuning speed + +<|ref|>text<|/ref|><|det|>[[500, 383, 917, 425]]<|/det|> +53 and efficiency due to its slow mechanical response. To date, 54 the majority of comb generators still have to rely on external 55 laser control to adjust the microcomb state. + +<|ref|>text<|/ref|><|det|>[[500, 429, 917, 589]]<|/det|> +56 Recently, there are significant advances in chip- scale integration of semiconductor lasers and nonlinear comb generators [16, 22- 24], in which a diode laser produces single- frequency laser emission to pump a hybridly or heterogeneously integrated external nonlinear resonator to excite microcombs. Such a fully integrated system shows great promise in improving the size, weight, and power consumption. However, the nature of soliton comb generation remains essentially the same, with all the above challenges persistent. Up to now, 65 the realization of an integrated comb source free from these 66 challenges remains elusive. + +<|ref|>text<|/ref|><|det|>[[500, 593, 917, 796]]<|/det|> +67 Here we present a fundamentally distinctive approach to resolve all these challenges in a single device. Figure 1a shows 68 the device concept. In contrast to conventional approaches 69 that rely solely on a single mechanism - either optical Kerr 70 or electro- optic effect - for comb generation while with external pumping, we utilize the resonantly enhanced electro- 71 optic (EO) modulation to initiate the comb generation, the 72 resonantly enhanced optical Kerr effect to expand the comb 73 bandwidth and phase- lock the comb lines, and the embedded 74 III- V optical gain to sustain and stabilize the comb operation. 75 Moreover, the resulting coherent microwave (via optical detection) is fed back to the EO comb to further enhance the 76 mode- locking, leading to unique self- sustained comb operation. + +<|ref|>text<|/ref|><|det|>[[500, 800, 917, 914]]<|/det|> +81 We realize this approach by integrating a III- V gain element 82 with a thin- film lithium- niobate (LN) photonic integrated circuit (PIC) to produce a III- V/LN comb laser (Fig. 1b). LN 83 PIC has attracted significant interest recently [25- 28] for 84 a variety of applications including high- speed modulation 85 [29, 30], frequency conversion [31- 33], optical frequency 86 comb [15, 34, 35], and single- frequency lasers [36- 39]. Here, 87 we unite active EO modulation with passive four- wave mix + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 66, 920, 447]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 459, 920, 592]]<|/det|> +
FIG. 1. Device concept of the integrated comb laser. a. Conceptual illustration of the comb generation and mode-locking principle, in which electro-optic (EO) comb generation, Kerr comb generation, and broadband optical gain all work synergistically together inside a single laser cavity for on-demand generation of mode-locked soliton comb. In addition, the laser comb output is detected and fed back to the laser cavity for resonant EO modulation to realize a self-sustained operation. b. Schematic of comb laser cavity structure formed by hybrid integration between a RSOA chip and a LN external cavity chip. Two different configurations are employed: A, cavity-enhanced (CE) comb laser structure in which the LN external cavity is formed mainly by an embedded high-Q racetrack resonator together with a broadband Sagnacloop mirror; B, Fabry-Perot (FP) comb laser in which the LN external cavity is formed by an EO phase modulation section together with an a broadband Sagnac loop mirror. c. Photo of a CE comb laser, showing that the RSOA is edge-coupled to the LN external cavity chip. d. Zoom-in photo showing the edge-coupling region between the RSOA and the LN chip. e., Photo of the racetrack resonator and the loop mirror in a CE comb laser. f. Photo of the EO phase modulator and the loop mirror in an FP comb laser.
+ +<|ref|>text<|/ref|><|det|>[[68, 618, 488, 910]]<|/det|> +89 ing (FWM) in a dispersion- engineered high- Q laser cavity 90 for the on- demand generation of mode- locked soliton micro- 91 comb, which naturally leads to self- starting full turnkey op- 92 eration simply by turning on/off either the RF signal driving 93 the comb resonator or the electric current driving the gain el- 94 ement. As the comb modes extract energy directly from ma- 95 terial gain, \(100\%\) of the optical power contributes directly to 96 the comb generation. Moreover, the strong electro- optic ef- 97 fect of the LN cavity enables high tunability and reconfigura- 98 bility of the produced microcomb. With this approach, we 99 are able to produce broadband highly coherent microcombs, 100 with individual comb linewidth down to \(600\mathrm{Hz}\) , frequency 101 tunability of over \(2.4\times 10^{17}\mathrm{Hz / s}\) for the entire microcomb, 102 microwave phase noise down to - 115 dBc/Hz at \(500\mathrm{kHz}\) fre- 103 quency offset, and a wall- plug efficiency exceeding \(5.6\%\) . The 104 simplicity of the demonstrated approach opens up a new path 105 for on- demand generation of mode- locked microcombs that is 106 expected to have profound impact on the broad applications in 107 high- precision metrology, telecommunications, remote sens- 108 ing, clocking, computing, and beyond. + +<|ref|>sub_title<|/ref|><|det|>[[515, 618, 737, 634]]<|/det|> +## Comb laser structure design + +<|ref|>text<|/ref|><|det|>[[501, 650, 919, 851]]<|/det|> +The III- V/LN microcomb laser is formed by integrating an InP reflective semiconductor optical amplifier (RSOA) with an LN external cavity chip via facet- to- facet coupling. We employ two types of laser cavity structures for the purpose, as shown in Fig. 1b. Chip A laser structure is embedded with a dispersion- engineered EO microresonator and Chip B laser structure consists of a simple EO phase modulation waveguide section. The advantage of Chip- A resonator- type structure is that the high- Q microresonator offers strong cavity enhancement for comb generation and mode- locking, which we term as the cavity- enhanced (CE) comb laser. The benefit of the Chip- B Fabry- Perot- type structure is that it offers flexibility of mode- locking operation, which we term as the Fabry- Perot (FP) comb laser. + +<|ref|>text<|/ref|><|det|>[[500, 854, 917, 913]]<|/det|> +In the CE comb laser, the group- velocity dispersion (GVD) of the racetrack microresonator (intrinsic optical \(Q \sim 1.6 \times 10^{6}\) ) is engineered to be small but slightly anomalous to support broadband comb generation. At the same time, the pulley + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[83, 63, 919, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 570, 919, 781]]<|/det|> +
FIG. 2. Lasing performance of a CE comb laser. a. Schematic of the experimental setup for comb laser characterization. RSOA is powered by a stable current source and an RF signal generator is used to drive the racetrack resonator. OSA: optical spectrum analyzer; AC: autocorrelator; PD: photo-detector; ESA: electrical spectrum analyzer; OSC: real-time oscilloscope; PNA: phase noise analyzer. b. Optical spectrum of the comb laser output. Off state: single mode lasing with the RF driving signal turned off. On state: comb lasing when the RF driving signal is turned on. Inset: Autocorrelation trace of the laser output pulses, in the "comb on" state, in which the blue curve shows the experimental data, dotted curves show the the fitted autocorrelation profiles from individual \(\mathrm{sech}^2\) pulses, and the dashed curve show the overall fitted autocorrelation trace. The autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping. c. Electrical spectrum of the beat note detected from the comb laser output. The red and blue curves show the comb-on and comb-off (single-mode lasing) states, respectively, corresponding to those in b. Gray curve shows the noise background of the optical detector, as a comparison. The inset shows the detailed spectrum of the RF beat note at \(39.58\mathrm{GHz}\) . d. Phase noise spectrum of the \(39.58\mathrm{GHz}\) RF beat note (red) and the RF driving signal (blue). e. Phase noise spectrum of the CE comb laser output measured with a self-heterodyne method. Red and green curves show for two different comb states, and blue curve shows for the comb-off (single-mode lasing) state. Inset shows the corresponding frequency noise spectrum of the three laser states. f. P-I-V curve of the CE comb laser, in which the red and blue curves show the L-I and I-V curves, respectively. g - j. Turnkey operation of the comb laser at two different speed of \(2\mathrm{Hz}\) (g, h) and \(200\mathrm{Hz}\) (i, j), respectively. Red curve shows the normalized driving RF power and blue curve shows the beating signal between the comb laser output and an external reference laser at \(1582\mathrm{nm}\) . h and j show the zoomed-in signal for the on/off states, respectively.
+ +<|ref|>text<|/ref|><|det|>[[66, 808, 488, 911]]<|/det|> +128 coupling regions are specially designed for uniform close- to- 129 critical coupling to the resonator over a broad telecom band. 130 Such a design ensures high loaded optical \(\mathrm{Q}(\sim 5\times 10^{5})\) of 131 the resonator uniformly across a wide spectral range which is 132 crucial both for enhancing the comb generation and mode- 133 locking and for efficient light coupling into/out of the mi- 134 crosensorator. Moreover, we also engineer the GVD of straight + +<|ref|>text<|/ref|><|det|>[[491, 808, 918, 911]]<|/det|> +135 waveguide sections outside the racetrack resonator to compensate for that of the RSOA section so as to minimize the overall 136 laser cavity GVD. At the same time, the overall optical path 137 length of the laser cavity is designed to be an integer multiple of the racetrack resonator's for matching their resonance 138 mode frequencies and round-trip group delay. The GVD of 139 the FP comb laser is engineered in a similar fashion. The free + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 68, 490, 300]]<|/det|> +spectral range (FSR) of the racetrack resonator is designed to be around \(40\mathrm{GHz}\) to better accommodate bandwidths of RF filter and amplifier after detection of comb beating signal. In contrast, the FP comb laser is designed to have a small FSR of around \(10\mathrm{GHz}\) for easy operation of harmonic mode locking. To support the broadband operation, the Sagnac loop mirror employs an adiabatic coupling design [40] to achieve high reflection and feedback with a reflectivity of \(>95\%\) over a broad spectral band of \(1500 - 1600\mathrm{nm}\) . On the other hand, a horn taper waveguide is designed on the LN chip to minimize the coupling loss between the LN chip and the RSOA gain chip. The RSOA exhibits a broadband gain in the telecom L- band, with a 3- dB bandwidth of \(>40\mathrm{nm}\) . Fig. 1c- f show the device structures. Details about the design parameters and the characterization of the laser structures are provided in the Supplementary Information (SI). + +<|ref|>sub_title<|/ref|><|det|>[[86, 309, 279, 323]]<|/det|> +## Comb laser performance + +<|ref|>text<|/ref|><|det|>[[66, 330, 490, 636]]<|/det|> +To excite the mode- locked microcomb, we first launch a single- frequency RF signal to drive the racetrack resonator of the CE comb laser (Fig. 2a), with a frequency of \(39.58\mathrm{GHz}\) that matches its FSR. Before the RF signal is applied, the device exhibits single- mode or multi- mode lasing, with an example shown in the blue curve of Fig. 2b. However, A microcomb is readily produced as soon as the RF signal is applied, with an optical spectral bandwidth of about \(20\mathrm{nm}\) (Fig. 2b, red curve). Mode- locking of the comb is verified by the clean RF tone at \(39.58\mathrm{GHz}\) detected from the beating between comb lines (Fig. 2c), as well as the autocorrelation trace from the laser output pulses (Fig. 2b, inset). The \(39.58 - \mathrm{GHz}\) RF tone exhibits a high signal- to- noise ratio of \(79\mathrm{dB}\) (Fig. 2c, inset), whose phase noise spectrum matches identically the driving RF source (Fig. 2d), showing the preservation of the relative phase coherence between comb lines via mode- locking. Mode- locking of the comb is also clearly evident by the clean noise floor around the DC region in the RF spectrum (Fig. 2c. See SI for details), where the zero extra noise from the mode- locked comb state infers that all the comb lines of the entire comb are phase- locked together. + +<|ref|>text<|/ref|><|det|>[[66, 639, 490, 911]]<|/det|> +The underlying mechanism responsible for mode- locking dominantly contributes from the combined resonant EO modulation and optical Kerr effect, in which the EO modulation produces EO sidebands to initiate the comb generation while the optical Kerr effect broadens the comb spectrum and phase- locks the comb lines (Fig. 1a). Indeed, the laser is able to produce mode- locked soliton pulses in the absence of EO modulation (while with a narrower spectrum), in which only the optical Kerr effect is responsible for mode- locking. The detailed theoretical modeling and testing results are provided in SI. In Fig. 2c, the small RF tone around the half- harmonic at \(19.79\mathrm{GHz}\) indicates certain comb dynamics. It can be eliminated by reconfiguring the laser and one example is shown in SI which exhibits a clean single RF beating tone and a well- defined \(\mathrm{sech}^2\) - shaped soliton pulse spectrum. The two lasers mainly differ in their overall dispersion of the laser cavity, indicating that the device dispersion plays an important role on the comb spectrum. The two- sidelobe feature of the comb spectrum in Fig. 2b implies that the output pulses are likely + +<|ref|>text<|/ref|><|det|>[[500, 68, 916, 153]]<|/det|> +to be mode- locked two- color pulses in which the two color pulses bounds with each other via certain interpulse interaction [41]. Its exact nature, however, will require further exploration. The comb spectrum can be reconfigured by changing the power or frequency of the RF driving signal, whose details are provided in the SI. + +<|ref|>text<|/ref|><|det|>[[500, 156, 916, 428]]<|/det|> +In addition to the high coherence between the comb lines, the comb laser also exhibits narrow linewidth on its individual comb lines. To show this feature, we employ the correlated self- heterodyne method [42, 43] to characterize the overall linewidth of the whole comb laser by launching the entire comb for linewidth measurement (rather than characterizing individual comb lines themselves) (See SI for details). The recorded phase noise spectrum is shown in Fig. 2e, which indicates a white frequency- noise floor of \(\sim 350\mathrm{Hz}^2 /\mathrm{Hz}\) (Fig. 2e, inset) that corresponds to a laser linewidth of \(\sim 2\mathrm{kHz}\) . The linewidth of the comb lines can be decreased further and an example is shown in Fig. 2e for a slightly different comb state produced from the same laser, which exhibits a white frequency- noise floor of \(\sim 100\mathrm{Hz}^2 /\mathrm{Hz}\) (Fig. 2e, inset) that corresponds to a laser linewidth as low as \(\sim 600\mathrm{Hz}\) . Note that these values represent the overall linewidth contributed from the entire comb, which indicates the upper limit of the intrinsic linewidth of individual comb lines. + +<|ref|>text<|/ref|><|det|>[[500, 431, 916, 675]]<|/det|> +Figure 2f shows the current- dependent characteristics of the comb laser, which exhibits a low threshold current of \(50\mathrm{mA}\) , indicating the low overall loss of the integrated laser. The comb laser produces an optical output power of \(11\mathrm{mW}\) at a pumping current of \(275\mathrm{mA}\) and a pumping voltage of \(1.4\mathrm{V}\) , which corresponds to a wall- plug efficiency of \(2.8\%\) . As the laser has two output ports (Fig. 1b) that emit the same amount of optical power, the total wall- plug efficiency of the laser is thus \(5.6\%\) . This level of wall- plug efficiency is on par with other integrated external- cavity semiconductor lasers recently developed [44, 45]. Intriguingly, the comb power increases with increased driving RF power, whose details are provided in the SI. Note that the total optical power contributes fully to the generated comb, in strong contrast to conventional Kerr solitons or EO combs in which the major optical power remains in the residual pump wave with low comb generation efficiency. + +<|ref|>text<|/ref|><|det|>[[500, 678, 916, 877]]<|/det|> +A distinctive feature of the comb laser is that the produced comb can be switched on/off on demand by simply switching on/off the driving RF signal. To show this feature, we beat the comb with a reference single- frequency laser operating at the wavelength of \(1582\mathrm{nm}\) that is inside the comb spectrum, and monitor the beating signal with the RF driving signal being turned on/off. As shown in Fig. 2g and i, the beating signal follows faithfully the driving RF signal. The coherent beating signal shows up readily when the RF driving is on, indicating the generation of the mode- locked comb. The beating signal disappears right after the RF driving is off, indicating the shut- off of the comb state. Same phenomenon is observed when the reference laser is tuned to other wavelengths within the comb spectrum. + +<|ref|>text<|/ref|><|det|>[[500, 881, 916, 910]]<|/det|> +Similar phenomena are observed in the FP comb laser, while generally with smaller spectral extents due to the lack + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[89, 65, 914, 257]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 271, 920, 365]]<|/det|> +
FIG. 3. Harmonic mode-locking of a FP comb laser with an FSR of 9.817 GHz. Third harmonics (29.45 GHz) and fourth harmonics (39.27 GHz) are separately used as driving signals. a. Schematic of the experimental setup for harmonic mode-locking of the comb laser. b. Optical spectrum of the laser output with third-harmonic mode-locking, by driving the phase modulator with a RF signal at 29.45 GHz. c. Optical spectrum of the laser output with fourth-harmonic mode-locking, by driving the phase modulator with a RF signal at 39.27 GHz. In b and c, the left insets shows the electrical spectrum of the RF beat note detected from the output laser comb, and the right insets show the autocorrelation trace of the laser output pulses with dashed curve showing the fitted individual pulses. Same as Fig. 2b, autocorrelation is recorded directly from the laser output pulses without dispersion compensation or pulse shaping.
+ +<|ref|>text<|/ref|><|det|>[[65, 392, 488, 565]]<|/det|> +257 of cavity enhancement. The FP comb laser, however, exhibits a distinctive feature in that it can be flexibly mode- locked at higher harmonics of the laser cavity FSR. Fig. 3 shows this feature. We are able to achieve third- and fourth- order harmonic mode- locking by applying an RF signal to the phase modulation section of the FP comb laser, with a frequency of 29.45 and 39.27 GHz, respectively, that are three and four times of the laser FSR (9.817 GHz). Again, mode locking of the combs is clearly verified by the detected RF beating signal from the combs with a SNR of 77 dB, as well as by the autocorrelation traces from the laser output pulses (Fig. 3b,c, insets). + +<|ref|>sub_title<|/ref|><|det|>[[85, 572, 450, 587]]<|/det|> +## High-speed frequency tuning of the comb laser + +<|ref|>text<|/ref|><|det|>[[65, 592, 488, 752]]<|/det|> +Another distinctive characteristic of the comb laser is that the laser frequencies of the entire mode- locked comb can be tuned cohesively at an extremely high speed. To show this feature, we apply a triangular- waveform electric signal - to- gether with the 39.58- GHz RF driving signal - to the racetrack resonator of the CE comb laser as shown in Fig. 4a. While the 39.58- GHz RF driving signal supports the mode- locking process, the triangular- waveform electric signal will adiabatically tune the resonance frequencies of the racetrack resonator, thus tuning the laser frequencies of the entire mode- locked comb together as a whole. + +<|ref|>text<|/ref|><|det|>[[65, 754, 488, 911]]<|/det|> +To show this feature, we beat the comb with a narrow- linewidth reference CW laser at 1582 nm that is about 15 GHz away from a comb line, and monitor the beating signal in real time. At the same time, we monitor the spectrum of the recorded 39.58- GHz RF tone from the beating between the comb lines (See SI for details of the setup). The frequency dynamics of the 15- GHz beating signal with the reference laser show the frequency tuning of the comb line nearby while the 39.58- GHz RF tone from the comb line beating indicates the quality of mode locking during the frequency tuning. Fig. 4f- h show the temporal variation of the 15- GHz beating signal at + +<|ref|>text<|/ref|><|det|>[[496, 392, 918, 812]]<|/det|> +different modulation speeds of 1, 10, 100 MHz. They show clearly that the frequency tuning of the comb line follows faithfully the waveform of the driving triangular- waveform electric signal at all modulation speeds, with a deviation of no more than \(5\%\) . In particular, the recorded 39.58- GHz RF tone from the comb line beating (Fig. 4e- g) remains unchanged during the frequency tuning, except with created modulation sidebands that simply results from the laser frequency modulation (see also Fig. 4a, right figure). This observation confirms that the phase- locking between the comb modes is fully preserved during the high- speed frequency tuning process, indicating that the entire mode- locked comb is tuned in its frequencies as a whole, without any perturbation to the comb mode spacing. This is in strong contrast to other comb modulation approaches [11, 12, 46] where the comb mode spacing is seriously impacted by external modulation. The frequency tuning range of 1.2 GHz at the modulation speed of 100 MHz (Fig. 4h) corresponds to a frequency tuning rate as high as \(2.4 \times 10^{17} \mathrm{~Hz} / \mathrm{s}\) for the comb. Both the frequency tuning rate and tuning speed are orders of magnitudes higher than the piezoelectric tuning and the external pump modulation approaches [11, 12], which are constrained only by the photon lifetime of the high- Q racetrack resonator. As shown in Fig. 4e, the device exhibits a frequency tuning efficiency of about \(0.2 - 0.8 \mathrm{GHz} / \mathrm{V}\) depending on the modulation speed, which is more than an order of magnitude higher than the piezoelectric approach [12]. The tuning efficiency can be further doubled by employing both sets of driving electrodes of the racetrack resonator (Fig. 4a). + +<|ref|>sub_title<|/ref|><|det|>[[516, 817, 840, 832]]<|/det|> +## Feedback mode-locking of the comb laser + +<|ref|>text<|/ref|><|det|>[[496, 838, 918, 911]]<|/det|> +So far, the comb laser utilizes an external RF signal to support the mode- locking. This signal, however, can be removed by feeding the coherent 39.58- GHz RF tone detected from the comb mode beating directly back to the comb laser cavity to sustain the mode- locking, resulting in unique stand- alone self + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 65, 914, 460]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 473, 920, 592]]<|/det|> +
FIG. 4. Fast frequency tuning of the whole lasing comb. a. Left panel: Schematic of the setup for comb frequency tuning, in which a triangular-waveform electrical signal produced by an arbitrary waveform generator (AWG) is used to drive the racetrack resonator of the CE comb laser together with the 39.58-GHz mode-locking RF signal. Middle panel: Conceptual illustration of the comb frequency tuning process, showing the laser frequencies of the comb are tuned together as a whole. Right panel: Schematic showing the corresponding sideband creation around the comb lines, introduced by triangular-waveform frequency modulation. b, c, d. Time-frequency spectra of the beatnote between the comb laser output and a referenced laser operating at a fixed wavelength of 1582 nm, at the modulation speed of 1, 10, and 100 MHz, respectively. The dashed curves show the corresponding triangular-waveform EO tuning signal. Bottom panels: Corresponding relative frequency deviation at each modulation speed. e, f, g. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, at modulation speed of 1 MHz (e), 10 MHz (f), and 100 MHz (g), respectively. h. Laser frequency tuning efficiency recorded at different modulation speeds.
+ +<|ref|>text<|/ref|><|det|>[[66, 620, 488, 706]]<|/det|> +327 sustained comb lasing operation. Figure 5a illustrates this approach. The comb laser output is detected by a high- speed optical detector whose output RF signal is amplified to an adequate amplitude, filtered to suppress excess low- frequency noises, adjusted with appropriate phase, and then fed back to drive the racetrack resonator of the CE comb laser. + +<|ref|>text<|/ref|><|det|>[[66, 710, 488, 911]]<|/det|> +328 As shown in Fig. 5b, a broadband microcomb with a spectrum covering about \(50~\mathrm{nm}\) and an optical power of \(8.5\mathrm{mW}\) is produced on chip with a driving current of \(285~\mathrm{mA}\) . Indeed, the microcomb is readily produced on demand as soon as the driving current is turned on, with a driving current as low as \(60~\mathrm{mA}\) , as shown in Fig. 5c- e. Mode locking of the comb is clearly evident by the clean 39.58- GHz RF tone detected from the comb mode beating (Fig. 5f), which exhibits a SNR of \(65~\mathrm{dB}\) and a narrow 3- dB linewidth of \(1.5\mathrm{kHz}\) . The phase noise of the RF beating signal reaches a level of - 90 dBc/Hz at an offset frequency of \(60~\mathrm{kHz}\) , which is considerably lower than the laser heterodyne beating approach [47] and is comparable to that of free- running optical Kerr soliton microcombs [46, 48, 49]. The optical spectral bandwidth and + +<|ref|>text<|/ref|><|det|>[[500, 620, 918, 720]]<|/det|> +329 the output power of the comb laser increase considerably with increased driving current (Fig. 5d). This is expected since the increased optical power of the mode- locked comb inside the high- Q racetrack resonator would significantly enhance the optical Kerr effect and the resulting four- wave- mixing process to broaden the comb spectrum. No saturation is observed on the comb spectral bandwidth as the current increases. + +<|ref|>sub_title<|/ref|><|det|>[[500, 729, 603, 744]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[500, 767, 918, 911]]<|/det|> +330 The attainable extent of the microcomb spectrum or soliton pulse width in current devices is primarily limited by the available optical power inside the cavity and the group delay mismatch between the enhancing resonator and the main laser cavity. For the former, it can be improved by either reducing the loss (e.g., improving the RSOA- LN chip coupling efficiency) or increasing the optical gain (e.g., using a higher- power RSOA) inside the laser cavity. For the latter, our theoretical modeling (see SI) shows that the formation of ideal ultrashort soliton pulses would require that the roundtrip time + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[85, 65, 914, 556]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 570, 919, 625]]<|/det|> +
FIG. 5. Self-sustained operation of the CE comb laser with feedback locking. a. Schematic of self-feedback locking of the CE comb laser. b. Optical spectrum of the comb laser output at a driving current of \(285~\mathrm{mA}\) . c - e. Optical spectrum (d) and optical power (e) of the comb laser output as a function of the RSOA driving current (c). f. Electrical spectrum of the 39.58-GHz beat note detected from the laser output comb, with a driving current of \(60~\mathrm{mA}\) . g. Phase noise spectrum of the detected 39.58-GHz beat note.
+ +<|ref|>text<|/ref|><|det|>[[65, 653, 488, 767]]<|/det|> +366 of the main laser cavity be integer times that of the enhanc- 367 ing racetrack resonator. In current devices, however, there is a 368 certain amount of mismatch which limits the comb spectrum 369 and the coherence of the mode- beating RF tone. This prob- 370 lem can be resolved by further optimization of the roundtrip 371 length of the main laser cavity and introducing tunability, after 372 which we expect that ultra- broadband highly coherent soliton 373 microcomb can be produced. + +<|ref|>text<|/ref|><|det|>[[65, 768, 488, 911]]<|/det|> +374 To conclude, we have introduced a new type of chip- scale 375 microcomb laser that can be flexibly mode- locked with ei- 376 ther active- driving or passive- feedback approaches and that 377 can be tuned/reconfigured at an ultrafast speed, with robust 378 turnkey operation inherently built in. The demonstrated inte- 379 grated comb laser exhibits outstanding reconfigurability and 380 performance significantly beyond the reach of conventional 381 on- chip mode- locked semiconductor lasers [50- 53]. The 382 demonstrated devices combine elegantly the simplicity of in- 383 tegrated laser structure, robustness of mode- locking opera- + +<|ref|>text<|/ref|><|det|>[[504, 653, 917, 738]]<|/det|> +tion, and electro- optically enhanced tunability and controllability, opening up a new avenue towards on- demand generation of soliton microcombs with high power efficiency that we envision to be of great promise for a wide range of applications including ranging, communication, optical and microwave synthesis, sensing, metrology, among many others. + +<|ref|>sub_title<|/ref|><|det|>[[504, 739, 575, 752]]<|/det|> +## Method + +<|ref|>text<|/ref|><|det|>[[504, 753, 917, 911]]<|/det|> +Device fabrication The device fabrication begins with a con- 392 gruent x- cut thin film lithium- niobate- on- insulator (LNOI) wafer, with a \(600~\mathrm{nm}\) LN layer on a \(4.7~\mu \mathrm{m}\) silica- coated silicon substrate. E- beam lithography (EBL) and Ar- ion milling are used to etch the waveguide with ZEP- 520A as mask. Etching thickness ranges from \(300~\mathrm{nm}\) (CE comb laser) to \(350~\mathrm{nm}\) (FP comb laser) for dispersion engineering. Second EBL is applied on PMMA for deposition of \(400~\mathrm{nm}\) gold- evaporated electrodes, which are placed \(2.5~\mu \mathrm{m}\) from the waveguide. The distance between the waveguide and electrode is chosen to balance the EO modulation frequency + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[65, 66, 919, 100]]<|/det|> +402 with loss from metal absorption. Dicing and polishing of LN 404 and amplifier- to- chip coupling, with both coupling losses 403 chip are employed at last to acquire optimized fiber- to- chip 405 around 6 dB. + +<|ref|>text<|/ref|><|det|>[[63, 160, 490, 905]]<|/det|> +407 [1] Diddams, S. A., Vahala, K. & Udem, T. Optical frequency 408 combs: Coherently uniting the electromagnetic spectrum. Sci- 409 ence 369, eaay3676 (2020). 410 [2] Kippenberg, T. J., Gaeta, A. L., Lipson, M. & Gorodetsky, 411 M. L. Dissipative Kerr solitons in optical microresonators. Sci- 412 ence 361, eaan8083 (2018). 413 [3] Gaeta, A. L., Lipson, M. & Kippenberg, T. J. Photonic- chip- 414 based frequency combs. Nature Photonics 13, 158- 169 (2019). 415 [4] Chang, L., Liu, S. & Bowers, J. E. Integrated optical frequency 416 comb technologies. Nature Photonics 16, 95- 108 (2022). 417 [5] Marin- Palomo, P. et al. Microresonator- based solitons for mas- 418 sively parallel coherent optical communications. Nature 546, 274- 279 (2017). 419 [6] Suh, M.- G., Yang, Q.- F., Yang, K. Y., Yi, X. & Vahala, K. J. 420 Microresonator soliton dual- comb spectroscopy. Science 354, 600- 603 (2016). 421 [7] Xu, X. et al. 11 TOPS photonic convolutional accelerator for 422 optical neural networks. Nature 589, 44- 51 (2021). 423 [8] Feldmann, J. et al. Parallel convolutional processing using an 424 integrated photonic tensor core. Nature 589, 52- 58 (2021). 425 [9] Suh, M.- G. & Vahala, K. J. Soliton microcomb range measure- 426 ment. Science 359, 884- 887 (2018). 427 [10] Trocha, P. et al. Ultrafast optical ranging using microresonator 428 soliton frequency combs. Science 359, 887- 891 (2018). 429 [11] Riemensberger, J. et al. Massively parallel coherent laser rang- 430 ing using a soliton microcomb. Nature 581, 164- 170 (2020). 431 [12] Liu, J. et al. Monolithic piezoelectric control of soliton micro- 432 combs. Nature 583, 385- 390 (2020). 433 [13] Spencer, D. T. et al. An optical- frequency synthesizer using 434 integrated photonics. Nature 557, 81- 85 (2018). 435 [14] Liang, W. et al. High spectral purity Kerr frequency comb radio 436 frequency photonic oscillator. Nature Communications 6, 7957 437 (2015). 438 [15] He, Y. et al. Self- starting bi- chromatic LiNbO3 soliton micro- 439 comb. Optica 6, 1138- 1144 (2019). 440 [16] Shen, B. et al. Integrated turnkey soliton microcombs. Nature 441 582, 365- 369 (2020). 442 [17] Bai, Y. et al. Brillouin- Kerr soliton frequency combs in an 443 optical microresonator. Physical Review Letters 126, 063901 444 (2021). 445 [18] Rowley, M. et al. Self- emergence of robust solitons in a micro- 446 cavity. Nature 608, 303- 309 (2022). 447 [19] Obrzud, E., Lecomte, S. & Herr, T. Temporal solitons in mi- 448 crosenators driven by optical pulses. Nature Photonics 11, 450 600- 607 (2017). 451 [20] Xue, X., Zheng, X. & Zhou, B. Super- efficient temporal soli- 452 tons in mutually coupled optical cavities. Nature Photonics 13, 453 616- 622 (2019). 454 [21] Bao, H. et al. Laser cavity- soliton microcombs. Nature Pho- 455 tonics 13, 384- 389 (2019). 456 [22] Stern, B., Ji, X., Okawachi, Y., Gaeta, A. L. & Lipson, M. 457 Battery- operated integrated frequency comb generator. Nature 458 562, 401- 405 (2018). 459 [23] Raja, A. S. et al. Electrically pumped photonic integrated soli- 460 ton microcomb. Nature Communications 10, 680 (2019). 461 [24] Xiang, C. et al. Laser soliton microcombs heterogeneously in- + +<|ref|>text<|/ref|><|det|>[[501, 163, 920, 905]]<|/det|> +tegrated on silicon. Science 373, 99- 103 (2021). 425 [25] Boes, A., Corcoran, B., Chang, L., Bowers, J. & Mitchell, A. Status and potential of lithium niobate on insulator (LNOI) for photonic integrated circuits. Laser & Photonics Reviews 12, 1700256 (2018). 426 [26] Lin, J., Bo, F., Cheng, Y. & Xu, J. Advances in on- chip photonic devices based on lithium niobate on insulator. Photonics Research 8, 1910- 1936 (2020). 427 [27] Zhu, D. et al. Integrated photonics on thin- film lithium niobate. Advances in Optics and Photonics 13, 242- 352 (2021). 428 [28] Boes, A. et al. Lithium niobate photonics: Unlocking the electromagnetic spectrum. Science 379, eabj4396 (2023). 429 [29] Wang, C. et al. Integrated lithium niobate electro- optic modulators operating at CMOS- compatible voltages. Nature 562, 101- 104 (2018). 430 [30] He, M. et al. High- performance hybrid silicon and lithium niobate Mach- Zehnder modulators for 100 Gbit s- 1 and beyond. Nature Photonics 13, 359- 364 (2019). 431 [31] Wang, C. et al. Ultrahigh- efficiency wavelength conversion in nanophotonic periodically poled lithium niobate waveguides. Optica 5, 1438- 1441 (2018). 432 [32] Lu, J., Li, M., Zou, C.- L., Al Sayem, A. & Tang, H. X. Toward 1% single- photon anharmonicity with periodically poled lithium niobate microring resonators. Optica 7, 1654- 1659 (2020). 433 [33] McKenna, T. P. et al. Ultra- low- power second- order nonlinear optics on a chip. Nature Communications 13, 4532 (2022). 434 [34] Zhang, M. et al. Broadband electro- optic frequency comb generation in a lithium niobate microring resonator. Nature 568, 373- 377 (2019). 435 [35] Gong, Z., Liu, X., Xu, Y. & Tang, H. X. Near- octave lithium niobate soliton microcomb. Optica 7, 1275- 1278 (2020). 436 [36] de Beeck, C. O. et al. III/V- on- lithium niobate amplifiers and lasers. Optica 8, 1288- 1289 (2021). 437 [37] Han, Y. et al. Electrically pumped widely tunable O- band hybrid lithium niobite/III- V laser. Optics Letters 46, 5413- 5416 (2021). 438 [38] Li, M. et al. Integrated Pockels laser. Nature Communications 13, 5344 (2022). 439 [39] Ling, J. et al. Self- injection locked frequency conversion laser. Laser & Photonics Reviews 2200663 (2023). 440 [40] Ramadan, T. A. & Osgood, R. M. Adiabatic couplers: design rules and optimization. Journal of lightwave technology 16, 277 (1998). 441 [41] Liao, R. et al. Dual- comb generation from a single laser source: principles and spectroscopic applications towards mid- IR- A review. Journal of Physics: Photonics 2, 042006 (2020). 442 [42] Camatel, S. & Ferrero, V. Narrow linewidth CW laser phase noise characterization methods for coherent transmission system applications. Journal of Lightwave Technology 26, 3048- 3055 (2008). 443 [43] Yuan, Z. et al. Correlated self- heterodyne method for ultralow- noise laser linewidth measurements. Optics Express 30, 25147- 25161 (2022). 444 [44] Xiang, C. et al. High- performance silicon photonics using heterogeneous integration. IEEE Journal of Selected Topics in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[65, 68, 490, 370]]<|/det|> +519 Quantum Electronics 28, 1- 15 (2021). 520 [45] Porter, C., Zeng, S., Zhao, X. & Zhu, L. Hybrid integrated chip- scale laser systems. APL Photonics 8 (2023). 521 [46] He, Y. et al. High- speed tunable microwave- rate soliton micro- comb. Nature Communications 14, 3467 (2023). 522 [47] Xiang, C. et al. 3d integration enables ultralow- noise isolator- free lasers in silicon photonics. Nature 620, 78- 85 (2023). 523 [48] Liu, J. et al. Photonic microwave generation in the X- and K- band using integrated soliton microcombs. Nature Photonics 14, 486- 491 (2020). 524 [49] Wang, B. et al. Towards high- power, high- coherence, integrated photonic mmwave platform with microcavity solitons. Light: Science & Applications 10, 4 (2021). 525 [50] Wang, Z. et al. A III- V- on- Si ultra- dense comb laser. Light: Science & Applications 6, e16260- e16260 (2017). 526 [51] Davenport, M. L., Liu, S. & Bowers, J. E. Integrated heterogeneous silicon/III- V mode- locked lasers. Photonics Research 6, 468- 478 (2018). 527 [52] Malinowski, M. et al. Towards on- chip self- referenced frequency- comb sources based on semiconductor mode- locked lasers. Micromachines 10, 391 (2019). 528 [53] Hermans, A., Van Gasse, K. & Kuyken, B. On- chip optical comb sources. APL Photonics 7 (2022). + +<|ref|>sub_title<|/ref|><|det|>[[66, 384, 240, 400]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[65, 412, 488, 457]]<|/det|> +The authors would like to thank Lin Chang and William Renninger for helpful discussions. This work is supported in part by the Defense Advanced Research Projects Agency + +<|ref|>text<|/ref|><|det|>[[494, 66, 919, 196]]<|/det|> +(DARPA) QuICC program under Agreement No. FA8650- 23- C- 7312 and LUMOS program under Agreement No. HR001- 20- 2- 0044, and the National Science Foundation (NSF) under Grant No. OMA- 2138174 and ECCS- 2231036. This work was performed in part at the Cornell NanoScale Facility, a member of the National Nanotechnology Coordinated Infrastructure (National Science Foundation, ECCS- 1542081); and at the Cornell Center for Materials Research (National Science Foundation, Grant No. DMR- 1719875). + +<|ref|>sub_title<|/ref|><|det|>[[495, 210, 680, 223]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[494, 225, 918, 315]]<|/det|> +J.L. and Z.G. designed and fabricated the devices. J.L. and Z.G. performed the device characterization. S.X. and Q.H. assisted in the device fabrication. S.X., K.Z., U.J., R.L. and J.S. assisted in experiments. J.L., Z.G., and Q.L. wrote the manuscript with contributions from all authors. Q.L. supervised the project. Q.L. conceived the concept. + +<|ref|>sub_title<|/ref|><|det|>[[495, 327, 699, 340]]<|/det|> +## Additional Informations + +<|ref|>text<|/ref|><|det|>[[494, 342, 917, 444]]<|/det|> +Supplementary information Supplementary Information is available for this paper. Competing interests. The authors declare no competing interests. Correspondence and requests for materials should be sent to Q.L. (qiang.lin@rochester.edu). Reprints and permission. Reprints and permissions are available at www.nature.com/reprints. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 488, 150]]<|/det|> +SupplementaryInformationforsolitonlasing.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/images_list.json b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..5b168d3f7c037803d477b48797a7bb38ecf2895b --- /dev/null +++ b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 40, + 90, + 950, + 680 + ] + ], + "page_idx": 38 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 40, + 203, + 955, + 448 + ] + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 40, + 40, + 960, + 515 + ] + ], + "page_idx": 40 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 44, + 40, + 844, + 787 + ] + ], + "page_idx": 41 + } +] \ No newline at end of file diff --git a/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f.mmd b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1acc1431717d4cc598fe9cea36ac070adf0961ab --- /dev/null +++ b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f.mmd @@ -0,0 +1,598 @@ + +# Crosstalk between lymphoid and myeloid cells orchestrates glioblastoma immunity through Interleukin 10 signaling + +Vidhya Ravi Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany https://orcid.org/0000- 0003- 0062- 2099 + +Nicolas Neidert Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +Paulina Will Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Kevin Joseph Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Julian Maier Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Jan Kuckelhaus Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Lea Vollmer Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Jonathan Goeldner Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +Simon Behringer Department of Neurosurgery + +Florian Scherer Department of Medicine I, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Germany + +Melanie Boerries Faculty of Medicine, Freiburg University, Germany + +Marie Follo + +<--- Page Split ---> + +University Medical Center + +Tobias Weiss University Hospital of Zurich + +Daniel Delev Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +Julius Kembach Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +Pamela Franco Department of Neurosurgery + +Nils Schallner Faculty of Medicine, Freiburg University, Germany + +Christine Dierks Faculty of Medicine, Freiburg University, Germany + +Maria Stella Carro University Medical Center Freiburg https://orcid.org/0000- 0002- 8570- 7691 + +Ulrich Hofmann Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +Christian Fung Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +Roman Sankowski University of Freiburg https://orcid.org/0000- 0001- 9215- 8021 + +Marco Prinz University of Freiburg https://orcid.org/0000- 0002- 0349- 1955 + +Jurgen Beck University Medical Center Freiburg + +Oliver Schnell University Medical Center Freiburg + +Dieter Heiland ( Dieter.henrik.heiland@uniklinik- freiburg.de ) Medical Center, University of Freiburg https://orcid.org/0000- 0002- 9258- 3033 + +## Article + +Keywords: cancer immunotherapy, Glioblastoma (GBM), T- cells, JAK/STAT inhibition + +Posted Date: February 17th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 244157/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on February 17th, 2022. See the published version at https://doi.org/10.1038/s41467-022-28523-1. + +<--- Page Split ---> + +# Crosstalk between lymphoid and myeloid cells orchestrates glioblastoma immunity through Interleukin 10 signaling + +Vidhya M. Ravi \(^{1,2,3,4,\#}\) , Nicolas Neidert \(^{1,2,3,\#}\) , Paulina Will \(^{1,2,3,\#}\) , Kevin Joseph \(^{1,2,3}\) , Julian P. Maier \(^{1,2,3}\) , Jan Kückelhaus \(^{1,2,3}\) , Lea Vollmer \(^{1,2,3}\) , Jonathan M. Goeldner \(^{1,2,3}\) , Simon P. Behringer \(^{1,2,3}\) , Florian Scherer \(^{3,5}\) , Melanie Boerries \(^{3,6,7,8}\) , Marie Follo \(^{3,5}\) , Tobias Weiss \(^{9}\) , Daniel Delev \(^{10,11}\) , Julius Kernbach \(^{10,11}\) , Pamela Franco \(^{3,4}\) , Nils Schallner \(^{3,12}\) , Christine Dierks \(^{3,5}\) , Maria Stella Carro \(^{2,3}\) , Ulrich G. Hofmann \(^{3,4}\) , Christian Fung \(^{2,3}\) , Roman Sankowski \(^{3,13}\) , Marco Prinz \(^{3,13,14,15}\) , Jürgen Beck \(^{2,3,15}\) , Oliver Schnell \(^{2,3,16,\dagger}\) , Dieter Henrik Heiland \(^{1,2,3,\dagger}\) + +\(^{1}\) Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, + +Germany + +\(^{2}\) Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +\(^{3}\) Faculty of Medicine, Freiburg University, Germany + +\(^{4}\) Neuroelectronic Systems, Medical Center, University of Freiburg, Germany + +\(^{5}\) Department of Medicine I, Medical Center - University of Freiburg, Faculty of Medicine, + +University of Freiburg, Germany. + +\(^{6}\) Institute of Medical Bioinformatics and Systems Medicine, Medical Center- University of Freiburg, + +79110 Freiburg, Germany. + +\(^{7}\) Comprehensive Cancer Center Freiburg (CCCF), Faculty of Medicine and Medical Center - University + +of Freiburg, 79106 Freiburg, Germany. + +\(^{8}\) German Cancer Consortium (DKTK), partner site Freiburg. + +\(^{9}\) Department of Neurology and Brain Tumor Center, University Hospital Zurich and University of Zurich, Switzerland + +\(^{10}\) Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +\(^{11}\) Neurosurgical Artificial Intelligence Laboratory Aachen (NAILA), Department of Neurosurgery, + +RWTH University of Aachen, Aachen, Germany + +\(^{12}\) Department of Anesthesiology and Critical Care Medicine, Medical Center- University of Freiburg, + +79106 Freiburg, Germany + +\(^{13}\) Institute of Neuropathology, Medical Center - University of Freiburg, + +\(^{14}\) Signalling Research Centres BIOSS and CIBSS, University of Freiburg, Germany + +\(^{15}\) Center for NeuroModulation (NeuroModul), University of Freiburg, Freiburg, Germany + +\(^{16}\) Translational NeuroOncology Research Group, Medical Center, University of Freiburg, Germany + +# Equal contributed first authorship + +Equal contributed last authorship + +DISCLOSURE OF CONFLICTS OF INTEREST: No potential conflicts of interest were disclosed by the authors. + +<--- Page Split ---> + +# Corresponding author: + +Corresponding author:Dieter Henrik HeilandDepartment of NeurosurgeryMedical Center University of FreiburgBreisacher Straße 6479106 Freiburg-Germany-Tel: +49 (0) 761 270 50010Fax: +49 (0) 761 270 51020E- mail: dieter.henrik.heiland@uniklinik- freiburg.de + +<--- Page Split ---> + +## Abstract + +Despite recent advances in cancer immunotherapy, its efficacy in Glioblastoma (GBM) is limited due to poor understanding of molecular states and cellular plasticity of immune cells within the tumor microenvironment. Here, we combined spatial and single- cell transcriptomics of 47.284 immune cells, to map the potential cellular interactions leading to the immunosuppressive microenvironment and dysfunction of T cells. Computational approach identified a subset of IL10 releasing HMOX1+ myeloid cells which activates transcriptional programs towards a dysfunctional state in T cells, and was found to be localized within mesenchymal dominated subregions of the tumor. These findings were further validated by a human ex- vivo neocortical GBM model (n=6) coupled with patient derived peripheral T- cells. Finally, the dysfunctional transformation of T cells was shown to be rescued by JAK/STAT inhibition in both our model and in- vivo. We strongly believe that our findings would be the stepping stone towards successful development of immunotherapeutic approaches in GBM. + +<--- Page Split ---> + +## Introduction + +Tumor infiltrating lymphocytes, along with resident and migrated myeloid cells, account for a significant part of the tumor microenvironment in glioblastoma \(^{1 - 3}\) . Most recently, the characterization of the myeloid cell population using scRNA- sequencing revealed remarkable heterogeneity with regards to cellular diversity and plasticity within the myeloid compartment \(^{1,4}\) . However, the diversity of lymphoid cell types within malignant brain tumors remains unexplored and needs to be illuminated. Insights into the heterogeneity of cell type composition and driver genes for lineage differentiation within lymphoid compartment will aid in providing successful approaches for immunotherapy in the future. In other cancer entities such as colorectal cancer \(^{5}\) , liver cancer \(^{6}\) or melanoma \(^{7}\) , different T cell states have been investigated. Situations which involve prolonged immune activation and ambiguous stimulation, such as uncontrolled tumor growth or chronic infections, impede the ability of CD8 \(^{+}\) lymphocytes to secrete proinflammatory cytokines and maintain their cytotoxic profile \(^{7 - 9}\) . This cellular state, named dysfunctional or "exhausted" CD8 \(^{+}\) lymphocytes, represents a paramount barrier to successful immune- vaccination or checkpoint therapy \(^{2,10,11}\) . T cell exhaustion is partially orchestrated by regulation of inhibitory cell surface receptors (PD- 1, CTLA- 4, LAG- 3, TIM- 3 and others), in addition to anti- inflammatory cytokines such as IL- 10 and TGF- \(\beta\) . Glioblastoma, a common and very aggressive primary brain tumor in adults, is archetypical for tumors with a strong immunosuppressive microenvironment \(^{12}\) . Current, immunotherapeutic approaches such as PDL1/PD1 checkpoint blockade \(^{13}\) or peptide vaccination \(^{14}\) , led to remarkable responses in several cancers, has failed to demonstrate its effectivity in patients suffering from glioblastoma. + +To address the sparse knowledge with respect to the lymphoid cell population in glioblastoma, we performed deep transcriptional profiling by means of scRNA- sequencing, and mapped potential cellular interactions and cytokine responses that could lead to the dysfunctional and exhausted phenotype of T cells. Pseudotime analysis revealed an increased response to Interleukin 10 (IL10) during the transformation of T cells from the effector state to the dysfunctional state. To computationally explore "connected" cells driving this transformation, we introduced a novel approach termed "nearest functionally connected neighbor (NFCN)", which identified a subset of myeloid cells marked by CD163 \(^{+}\) and HMOX1 \(^{+}\) expression. Furthermore, we performed spatially resolved transcriptomics, which confirmed the + +<--- Page Split ---> + +spatial overlap of exhausted T cells with HMOX1+ myeloid cells, within regions of the tumor enriched with mesenchymal transcriptional signatures. Finally, using human neocortical GBM model (coupled with patient derived T cells) and myeloid cell depletion, we conclusively validated that myeloid cells are key drivers for immunosuppressive microenvironment in the glioblastoma. The dysfunctional transformation of T cells were found to be rescued by the JAK- STAT inhibition due to the reduction in IL10 release, as shown before 15, in our ex- vivo model. Based on these results, we treated a single patient having recurrent glioblastoma in a neoadjuvant setting with JAK- STAT inhibitor (Ruxolitinib) and partially rescued the immunosuppressive environment. + +<--- Page Split ---> + +## Results: + +## Single cell Analysis of the Immune Cell Compartment in Glioblastoma + +In order to interrogate the diversity of the immune microenvironment in glioblastoma, we performed droplet based 10X single cell sequencing of tissue samples from 8 patients, diagnosed with Glioblastoma. Lymphoid and myeloid populations (CD45+/CD3+) were sorted from neoplastic tissue specimens (Figure 1a and Supplementary Figure 1a). The scRNA- seq data consisted of 47,284 cells, with a median number of 2,301 unique molecular identifiers (UMIs) and approximately 1023 uniquely expressed genes per cell. We corrected the data for mitochondrial genes, regressed out cell cycle effects and removed batch effects due to technical artifacts. We then decomposed the eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparing them to randomized expression values, resulting in 41 meaningful components. Shared nearest neighbor (SNN) graph clustering resulted in 21 clusters (C0- C20) containing uniquely expressed genes. The major observed cell type when using the semi- supervised subtyping algorithm of scRNA- seq (SCINA- Model) \(^{16}\) and SingleR \(^{17}\) are: microglia cells (TMEM119, CX3CR1 and P2RY12) and macrophages (AIF1, CD68, CD163 and low expression of TMEM119, CX3CR1), followed by CD8+ T cells (CD8A, CD3D), natural killer cells (KLRD1, GZMH, GZMA, NKG7 and CD52), CD4+ T cells (BCL6, CD3D, CD4, CD84 and IL6R), T- memory cells (TRBC2, LCK, L7R and SELL), granulocytes (LYZ), a minor number of oligodendrocytes and oligodendrocyte- progenitor cell (OPC's) (OLIG1, MBP, PDGFA), and endothelial cells (CD34, PCAM1, VEGFA) Figure 1b, Supplementary Figure 1b- f. To identify malignant cells, we inferred large scale copy number variations (CNVs) from scRNA- seq profiles by averaging expression over stretches of 100 genes on their respective chromosomes \(^{18}\) . With this approach, we confirmed that there was minimal contamination by tumor cells (clustered as OPC cells), based on their typical chromosomal alterations (gain in chromosome 7 and loss in chromosome 10), Supplementary Figure 2a. + +## Diversity of T cells in the Glioblastoma Microenvironment + +To investigate the diversity of the T cells present in the microenvironment, we examined them by two different but complementary methods. Firstly, T cells were isolated in- silico by means of clustering (as shown above), based on previously published marker gene expression profiles (CD3+, CD4+/CD8+). Secondly, they were + +<--- Page Split ---> + +isolated using the SCINA model, which resulted in a total of 7,547 cells Supplementary Figure 1g. Focusing on the different regulatory states of these cells, we identified 13 subclusters using SNN- clustering which were then re- embedded into a dynamic model using RNA- velocity, closely reflecting different activation states, Supplementary Figure 1h and Figure 1c. Reconstruction of lineage differentiation trajectories by means of both pseudotime and latent time provided insights into the transformation of the cells over time19, Figure 1c. Based on common marker signatures, we defined activated T cells by expressing GZMA, CCL5 and CCL4, IL7RI and other markers (Supplementary Figure 1i and 2b) as well as increased proliferation (G2M- score, Figure 1d) (C1,C10), and naive T cells by the expression of SELL, TCF7 and CCR7 (C0), Figure 1c and Supplementary Figure 1h. We further identified T cell subgroups marked by hypoxia and heat- shock signaling (HSPA8, C2, C5,C6) and by a subgroup which showed mixed expression of activation/dysfunctional/exhaustion markers (HAVCR2, GNLY) and strong enrichment IL- 10 signaling, Figure 1d. We estimated the G2M- score and identified a lower frequency of cell cycle in the differentiated/later states of T cells with the exception of the IFN- gamma subtype, Figure 1d. Additional marker plots are given in the Supplementary Figure 2b. Regulatory CD4+ T cells (FOXP3, IL2RA and CTLA4) represents a minor population in the glioblastoma microenvironment, Figure 1c and Supplementary Figure 1i-j. + +## Dysfunctional State of T cells is Driven by IL-10 Signaling + +To gain insights into the regulatory mechanism of immune cells, we reconstructed fate decisions made during T cell exhaustion using pseudotime trajectories along the estimated velocity streams, Figure 1e. Using RNA- velocity, we estimated cells with high probability of initial and terminal states which resulted in 3 major branches with unique cell fate drivers, Figure 1e. In order to determine the dynamic adaptation across all branches (Velo trajectory 1- 3), we computed pseudotime vectors using various models. Only two trajectories confirmed the predicted ascending pseudotime inference across multiple models (HAVCR2(+)- T cells and hypoxia- induced T cells), Figure 1f. We assume that terminal states without significant pseudotime connection do not inevitably arise from the determined initial state, which suggests that the lineage of IFN- gamma driven T cells coexists and most likely originated from an earlier lineage branch Figure 1g. + +<--- Page Split ---> + +Along our trajectory from effector to HAVCR2(+)T cells, we showed a latent time- dependent increase of the response to anti- inflammatory signaling such as IL- 10 and TGF- beta signaling Figure 1h. Next, we mapped the gene expression of defined exhausted and effector signatures along our differentiation trajectory, revealing an enrichment of exhausted genes within the destination cluster. Thus, our data suggests that this response to IL- 10 contributes to the dysfunctional state of T cells and affects fate decisions Figure 1i. To gain further insights into accurate downstream signaling of IL- 10, IFN- gamma, and IL2, we created a library of the 50 most highly up- and downregulated genes, Supplementary Figure 3a- b. We then extracted signatures observed within the different T cell clusters and compared them with stimulated T cells. As expected, genes upregulated by IL2 stimulation were significantly enriched in cluster 1, while signature genes from IFN- gamma stimulation was enriched in cluster 12. Signature genes from IL- 10 stimulation showed a significant enrichment in the dysfunctional cluster (Cluster 9), Supplementary Figure 3c. Furthermore, we mapped our signatures along our defined velocity trajectory 1 which confirmed our predicted pathway inference, Figure 1j- k and Supplementary Figure 3d. + +## T cell Activation and Exhaustion Reveals Spatial Heterogeneity and Association with Glioblastoma Subtypes + +Glioblastomas present a high degree of heterogeneity due to regional metabolic differences and varying composition of the tumor microenvironment20. To decipher the spatial distribution of above illustrated T cell clusters and its colocalization to defined tumor states, we performed spatial transcriptomic RNA sequencing (stRNA- seq) of 3 primary IDH1/2 wildtype glioblastoma, containing a total number of 2,352 spots, Figure 2a. We observed a median of 8 cells per spot (range: 4 to 22 cells per spot), which allows the spatial mapping of gene expression, but not at single cell resolution. However, when we compared our dataset to the latest classification of glioblastoma20, consistent results were obtained in accordance with the diversity of subtype expression, Figure 2b. In particular, Neftel and colleagues raised evidence that the mesenchymal gene expression is more likely associated with immune response and cellular interaction to myeloid cells20. In order to investigate the spatial distribution of the mesenchymal subgroup, we performed gene set enrichment analysis at spatial resolution which revealed spot- wise enrichments within all samples, Figure 2c- d. In a next step, we used a seeded non- negative matrix factorization (NMF) regression21 to + +<--- Page Split ---> + +estimate the probability of individual T cell clusters at spatial resolution, Figure 2e. We computed the spatial overlap of T cell clusters and glioblastoma states, using a Bayesian approach, which revealed a high estimated correlation of the mesenchymal and T cell cluster 3 (T HAVCR2), cluster 5- 7 (T Hypoxia), Figure 2f. Further analysis confirmed the strong overlap of dysfunctional/exhausted marker (LAG3 and HAVCR2) exclusively with the mesenchymal gene expression signature, Figure 2g, suggesting that our data support the findings from Neftel and colleagues. + +## A Subset of Microglia and Macrophages Drive IL-10 Stimulation + +In our recent investigation \(^{15}\) , the crosstalk between microglia cells and reactive astrocytes in the tumor microenvironment was found to be responsible for upregulating IL10 release. This is mediated by microglia/macrophages stimulated with IFN gamma, leading to JAK/STAT activation in tumor- associated astrocytes. In this study we introduce the "nearest functionally connected neighbor" algorithm (NFCN), an in- silico model to identify the most likely related cell pairs through divergent down- and upstream signal activity, Figure 3a. With our model, we assume that cellular interaction with distinct mutual activation implies two fundamental prerequisites. On the one hand, the ligand needs to be expressed and released, or otherwise exposed on the cell surface. To avoid the chances of randomly elevated expression or technical artifacts, we also looked at the simultaneous occurrence of ligand induction (upstream pathway signaling). On the other hand, the receptor needs to be expressed and, additionally, downstream signaling has to be activated as well. This allows us to predict the functional status of the receiver cell (Explanation of the model can be found in the Methods section, with an overview in Supplementary Figure 4). We used our in- silico model to screen for potential cells responsible for IL- 10 activation of T cells. The algorithm identified pairs of lymphoid (T cell clusters) and myeloid cells (macrophages and microglia cluster) and estimated the likelihood of mutual activation Figure 3b,c. By extraction of the nearest connected cells (top 1% ranked cells), we identified a subset of myeloid cells characterized by remarkably high IL10 expression. Most of the receiver cells in the connected cells (top 1% ranked cells) originated from the T cell cluster Figure 3d. Our predicted IL10 interactions were also supported by another computational approach (Supplementary Figure 5). In order to validate our computational model, we used SPOTlight \(^{21}\) , an algorithm to predict the spatial position of cells from scRNA-seq data, and were able to confirm a significant overlap. In order to explore the difference between connected and non- connected cells, we extracted + +<--- Page Split ---> + +both connected and non- connected cells, defined by the highest and lowest interaction- scores (quantile \(97.5\%\) ) Using differential gene expression analysis, we observed multiple genes which confirmed the non- inflammatory polarization status of highly connected cells. + +These findings are not surprising, since one of the essential markers of non- inflammatory myeloid cells is IL10, Figure 3e. We showed that the subset of most highly connected cells were marked by CD163 and heme oxygenase 1 (HMOX1) expression. In a multilayer representation, we illustrated the estimated cellular connections with respect to their most likely spatial coordinates, Figure 3f. HMOX1 is activated during inflammation and oxidative injuries and is regulated through the Nrf2/Bach1- axis, as well as through the IL10/HMOX1- axis. This gene is also well known to be upregulated in the alternative activated macrophage subtype22. Another immunosuppressive signaling marker closely related to the alternative activation of macrophages/microglia is the release of TGF- \(\beta^{23}\) , which was also found to be upregulated in highly connected cells, Figure 3g. Consistent with our findings, most downstream signals of the IL10/HMOX1- axis such as STAT3 and p38 MAPK were found to be upregulated in a gene set enrichment analysis, Figure 3d,h. In a recent investigation, the role of doublets in the detection of cell- cell interactions was described24. In our dataset the score of potential cell- cell connections was similar to HMOX1 marked macrophages, within the doublet dominated cluster. We computed the marker genes of our predicted connected myeloid and lymphoid genes and found each of the marker sets to be highly expressed in the doublet cluster, which further validated the results from our computational model, Figure 3i. + +## Loss of Myeloid Cells Increases Antitumor Immunity + +To validate our computational findings, we made use of the recently described human neocortical GBM model, where the cellular architecture of the CNS is well preserved15,25. We cultured non- infiltrated neocortical slices (defined in a recent report15,25) coupled with autografted T cells along with myeloid cell depletion, to understand the communication between myeloid cells in the tumor microenvironment along with lymphoid cells. Three days after chemical depletion of the myeloid cells, we injected a primary cell line (BTSC#233, GFP- tagged, previously characterized by RNA- seq profiling as mesenchymal)26. After 4 days of culture, peripheral T cells (same donors), tagged using CellTrace™ Far Red (CTFR) were additionally inoculated and the sections were further cultured for another 48h, Figure 4a. Immunostainings showed + +<--- Page Split ---> + +that myeloid cell depletion reduces the number of \(\mathrm{IBA1^{+}HMOX1^{+}}\) cells, Figure 4b. Using an enzyme- linked immunosorbent assay (ELISA) we found a significant reduction in IL10 when the myeloid cells were depleted, regardless of the presence of tumor cells. The strongest difference in IL10 release was observed in myeloid cell depleted sections in the presence of tumor cells, Figure 4c. Furthermore, we stained for Granzyme B (GZMB+) T cells and quantified IL2 release to examine the amount of effector T cells in both the depleted and non- depleted sections. We found an increased number of GZMB+ T cells in sections with myeloid cell depletion, Figure 4d, along with a significant increase in IL2 and no differences in IFN gamma release were observed, Figure 4e- f. We also stained for the exhaustion marker TIM3 (Gen: HAVCR2), which was found to be enriched in T cells in the presence of myeloid cell Figure 4g, suggesting that myeloid derived IL10 release (by HMOX1+ cells) leads to T cell exhaustion, which is in agreement with the computational model. In order to prove that the IL10 signaling is responsible for the induction of the expression of exhausted T cell markers, we preincubated T cells using an IL10 neutralizing inhibitory antibody Figure 4h. This resulted in a strong increase of GZMB+ T cells, followed by a significant increase of IL2, suggesting that IL10 inhibition drives T cell activity, Figure 4i- j. From our recent study, we found that a JAK/STAT inhibition causing a significant reduction of IL10 within the tumor microenvironment15. Based on this investigation, we pretreated tissue sections with Ruxolitinib, an FDA- approved JAK- inhibitor, before inoculating the sections with patient- derived T cells. We were able to confirm that JAK inhibition caused a significant decrease of IL10 and increased levels of the inflammatory marker IL2, Figure 4k- l. + +Based on our above findings, we treated a first patient with a recurrent glioblastoma in a neoadjuvant setting with Ruxolitinib for 4 weeks. After resection, we sorted CD45+ cells and performed scRNA- sequencing Figure 4m. Immunostainings of the tumor revealed a relative increase of \(\mathrm{CD8^{+}}\) and \(\mathrm{CD4^{+}}\) T cells whereas \(\mathrm{CD68^{+}}\) myeloid cells remained stable, Figure 4n- o. scRNA sequencing revealed a large number of T cells, most of which express markers for T cell activation and a few cells showing a naive signature, Figure 4p. When compared with T cells from our initial dataset, an enrichment of the T cells from the JAK treated patient was seen within cluster 1 (activated T cells) and cluster 12 (B cells), suggesting that JAK- inhibition could be a potential treatment option to boost T cell activation by reducing immunosuppressive programs in both myeloid and glial cells, Figure 4q- r. + +<--- Page Split ---> + +## Discussion: + +Although single- cell RNA- sequencing accurately maps the cellular architecture and reflects the diversity of cellular states \(^{18,20,27,28}\) , there is a lack of spatial information. Here, we combine single cell RNA sequencing of the immune compartment along with spatial transcriptomic RNA- sequencing (stRNA- seq) to gain better insights into the complex crosstalk, cellular states and cellular plasticity leading to the immunosuppressive environment found in glioblastoma (GBM). + +Recent studies have reported different subtypes of microglia and macrophages occupying glial tumors \(^{1,4,20,27,28}\) . However, detailed information about lymphoid infiltration cells is lacking. There is intense interest in T cells and their varied states due to their importance in the development of targeted therapies and to further the understanding of the immunosuppressive environment of glioblastoma. T cell states, particularly in disease, are somehow difficult to accurately classify, leading to numerous definitions and markers in recent years \(^{2,7,29 - 31}\) . Some authors use the terms "dysfunctional" and "exhausted" synonymously \(^{32}\) , whereas others differentiate between the dysfunctional and the exhausted states of T cells \(^{29,31}\) . In this study we use the definition of cellular states proposed by Singer et al., 2016 \(^{8}\) . On the basis of these gene sets, our data showed that only cells which remained activated along the pseudotime- trajectory were able to enter a state of dysfunction, and later exhaustion. The dysfunction appears to be a transient state, associated with increased proliferation, despite immunosuppressive stimulation from the tumor environment. This imbalance between pro- and anti- inflammatory signaling, dominated by IL10 secretion, leads to final exhaustion of the T cells, which is in agreement with the current literature \(^{2,33}\) . In order to find a consensus with regard to marker genes we further validated our findings on a set of exhausted marker genes recently published in an overview study \(^{34}\) . We and others have shown that the GBM microenvironment aids in the evolution of immune suppression. In this process, astrocytes and myeloid cells, both driven by STAT- 3 signaling, orchestrate the immunosuppressive environment \(^{4,15,35,36}\) . Based on the knowledge that IL10 interaction plays a crucial role in the shift from activated to exhausted T cells, we built an in- silico model that identified potential connected cells driving T cell exhaustion. + +Using this model, we identified a subset of myeloid cells marked by high expression of HMOX1 \(^{+}\) , a gene which is induced by oxidative stress and metabolic imbalance \(^{37,38}\) . HMOX1 is linked to the STAT- 3 pathway and induces IL10 production via MAPK + +<--- Page Split ---> + +activation, and all of these markers were also found to be upregulated in our connected cells, as reported above. Furthermore, we used spatial transcriptomics to confirm the spatial overlap of cells that we identified are highly connected. We were able to show that the HMOX1+- myeloid cells were spatially correlated with exhaustion and the mesenchymal state of glioblastoma. These findings are in accord with published reports, revealing that the mesenchymal cells are the component of GBM responsible for the immune crosstalk20. HMOX1 expression in GBM and IDH-WT astrocytoma was found to be increased in recurrent GBM and negatively associated with overall survival, Supplementary Figure 6a,b. In addition, we made use of a human neocortical GBM model coupled with patient derived T cells with/without myeloid cell depletion. This model helped us to simulate the function of the myeloid cells with regard to IL10 release and T cell stimulation. Fitting with our computational model, we confirmed that HMOX1+ myeloid cells cause a reduction of effector T cells, with a respective reduction in IL2 release and increased expression of our identified exhaustion marker TIM3. Following our recent investigations in which we demonstrated that JAK-inhibition is able to reduce the level of IL10 in human brain tumors15, we demonstrated that in a single patient that inhibition of the JAK- STAT axis was able to partially rescue the immunosuppressive environment. The treated subject is still alive but showed permanent disruption of the blood-brain barrier with repetitive increase of contrast enhancing lesions. Each radiologically confirmed progress was sampled without evidence of a tumor recurrence, suggesting that manipulation of the glia/myeloid environment simultaneously caused exaggerated inflammation and pseudo progression. Our single-cell RNA-seq confirmed a pronounced enrichment of activated T cells, while the number of myeloid cells remained relatively stable. In this sample, we also detected the strongest contamination with glial cells. We assumed that potential doublets or strong cell-cell connections led to glial cells being detected as CD45+, resulting in a false positive sorting. + +In conclusion, this work provides the first knowledge regarding lymphocyte population in the glioblastoma microenvironment where we showed that the functional interaction between myeloid and lymphoid cells, leads to a dysfunctional state of T cells. Using human neocortical GBM model and single patient subject we showed that the IL- 10 driven T cell exhaustion can be rescued by JAK/STAT inhibition. Thus, the results from this work can be the steppingstone towards successful immunotherapeutic approaches for GBM. + +<--- Page Split ---> + +## Methods: + +## Ethical Approval + +The local ethics committee of the University of Freiburg approved the data evaluation, imaging procedures and experimental design (protocol 100020/09 and 472/15_160880). The methods were carried out in accordance with the approved guidelines, with written informed consent obtained from all subjects. The studies were approved by an institutional review board. Further information and requests for resources, raw data and reagents should be directed and will be fulfilled by the Contact: D. H. Heiland, dieter.henrik.heiland@uniklinik- freiburg.de. A complete table of all materials used is given in the supplementary information. + +## T cell isolation and stimulation + +Blood was drawn from a healthy human individual into an EDTA (ethylenediaminetetraacetic acid) cannula. T cells were extracted in a negative selection manner using a MACSxpress® Whole Blood Pan T Cell Isolation Kit (Miltenyi Biotech). T cells were then transferred in Advanced RPMI 1640 Medium (ThermoFisher Scientific, Pinneberg, Germany) and split for cytokine treatment: Three technical replicates were used for each T cell- treatment condition. Interleukin 2 (IL- 2, Abcam, Cambridge, UK) was used at a final concentration of 1 ng/ml, Interleukin 10 (IL- 10, Abcam) at 5 ng/ml, Interferon gamma (IFN- \(\gamma\) , Abcam) at 1 ng/ml and Osteopontin (SPP- 1, Abcam) at 3 \(\mu \mathrm{g} / \mathrm{ml}\) . Cytokine treatment was performed in Advanced RPMI 1640 Medium and T cells were incubated at \(37^{\circ}\mathrm{C}\) and \(5\%\) CO2 for 24h. + +## RNA sequencing of stimulated T Cells + +The purification of mRNA from total RNA samples was achieved using the Dynabeads mRNA Purification Kit (Thermo Fisher Scientific, Carlsbad, USA). The subsequent reverse transcription reaction was performed using SuperScript IV reverse transcriptase (Thermo Fisher Scientific, Carlsbad, USA). For preparation of RNA sequencing, the Low Input by PCR Barcoding Kit and the cDNA- PCR Sequencing Kit (Oxford Nanopore Technologies, Oxford, United Kingdom) were used as recommended by the manufacturer. RNA sequencing was performed using the MinION Sequencing Device, the SpotON Flow Cell and MinKNOW software (Oxford Nanopore Technologies, Oxford, United Kingdom) according to the manufacturer's + +<--- Page Split ---> + +instructions. Samples were sequenced for 48h on two flow-cells. Basecalling was performed by Albacore implemented in the nanopore software. Only \(\mathsf{D}^2\) - Reads with a quality Score above 8 were used for further alignment. + +## Sequence trimming and Alignment + +In the framework of this study, we developed an automated pipeline for nanopore cDNA- seq data, which is available at github (https://github.com/heilandd/NanoPoreSeq). First the pipeline set up a new class termed "Poreseq" by a distinct sample description file. The analysis starts by rearranging the reads from the fastq output from the nanopore sequencer containing all of the \(\mathsf{D}^2\) - Reads. All fastq files need to be combined into one file. Multiplexed samples were separated according to their barcode and trimmed by Porechop (https://github.com/rrwick/Porechop). Alignment was performed with minimap2 (https://github.com/lh3-/minimap2) and processed with sam-tools. + +## Posthoc Analysis of Bulk-RNA-seq + +A matrix of counted genes was further prepared by the RawToVis.R (github.com/heilandd/VRSD_Lab_v1.5) script, containing normalization of Mapped reads by DESeq, batch effect removal (ComBat package) and fitting for differential gene expression. Gene set enrichment analysis was performed by transformation of the log2 fold- change of DE into a ranked z- scored matrix, which was used as the input. The expression matrix was analysed with AutoPipe (https://github.com/heilandd/AutoPipe) by a supervised machine- learning algorithm and visualized with a heatmap. Full analysis was visualized with the Visualization of RNA- Seq Data (VRSD_Lab software, github.com/heilandd/VRSD_Lab_v1.5) as a dashboard app based on shiny R- software. We extracted the 50 top up/down regulated genes respectively of each stimulation with respect to control condition to construct a stimulation library. + +## Single-Cell Suspension for scRNA-sequencing + +Tumor tissue was obtained from glioma surgery immediately after resection and was transported in phosphate- buffered saline (PBS) within approximately 5 minutes into our cell culture laboratory. Tumor tissue was processed under a laminar flow cabinet. Tissue was reduced to small pieces using two scalpels and the tissue was processed + +<--- Page Split ---> + +with the Neural Tissue Dissociation Kit (T) using C- Tubes (Miltenyi Biotech, Bergisch- Gladbach, Germany) according to the manufacturer's instructions. The Debris Removal Kit from Miltenyi was used according to the manufacturer's instructions to remove remaining myelin and extracellular debris. In order to remove the remaining erythrocytes, we resuspended the pellet in \(3.5 \text{ml ACK lysis buffer (ThermoFisher Scientific, Pinneberg, Germany)}\) and incubated the suspension for 5 minutes followed by a centrifugation step (350g, 10 min, RT). Cell quantification with a hematocytometer was performed after discarding the supernatant and resuspending the pellet in PBS. Cell suspensions were centrifuged again (350g, 10 min, RT) and resuspended in freezing medium containing \(10\%\) DMSO (Sigma- Aldrich, Schnelldorf, Germany) in FCS (PAN- Biotech, Aidenbach, Germany). Cell suspensions were immediately placed in a freezing box containing isopropanol and stored in a \(- 80^{\circ}\text{C}\) freezer for not more than 4 weeks. + +## Cell sorting by Magnetic Beads + +Four frozen single- cell suspensions, originating from one patient with an IDH- mutated glioma and three patients with an IDH- wildtype glioblastoma (GBM), were thawed and the dead cells magnetically labeled and eliminated using a Dead Cell Removal Kit (Miltenyi Biotech). The tumor immune environment in general and T cells in particular were positively selected by using \(\text{CD3 + - MACS}\) (Miltenyi Biotech). Cells were stained with trypan blue, counted using a hematocytometer and prepared at a concentration of 700 cells/μL. + +## Droplet scRNA-sequencing + +At least 16000 cells per sample were loaded on the Chromium Controller (10x Genomics, Pleasanton, CA, USA) for one reaction of the Chromium Next GEM Single Cell \(3' \text{v} 3.1\) protocol (10x Genomics), based on a droplet scRNA- sequencing approach. Library construction and sample indexing was performed according to the manufacturer's instructions. scRNA- libraries were sequenced on a NextSeq 500/550 High Output Flow Cell v2.5 (150 Cycles) on an Illumina NextSeq 550 (Illumina, San Diego, CA, USA). The bcl2fastq function and the cell ranger (v3.0) was used for quality control. + +<--- Page Split ---> + +## Postprocessing scRNA-sequencing + +We used cell ranger to detect low- quality read pairs of single- cell RNA sequencing (scRNA- seq) data. We filtered out reads which did not reach the following criteria: (1) bases with quality \(< 10\) , (2) no homopolymers (3) 'N' bases accounting for \(\geq 10\%\) of the read length. Filtered reads were mapped by STAR aligner and the resulting filtered count matrix further processed by Seurat v3.0 (R- package). We normalized gene expression values by dividing each estimated cell by the total number of transcripts and multiplied by 10,000, followed by natural- log transformation. Next, we removed batch effects and scaled data by a regression model including sample batch and percentage of ribosomal and mitochondrial gene expression. For further analysis we used the 2000 most variable expressed genes and decomposed eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparison to randomized expression values. The obtained non- trivial components were used for SNN clustering followed by dimensional reduction using the UMAP algorithm. Differently expressed genes (DE) of each cluster were obtained using a hurdle model tailored to scRNA- seq data which is part of the MAST package. Cell types were identified by 3 different methods; Classical expression of signature markers of immune cells; SingleR an automated annotation tool for single- cell RNA sequencing data obtaining signatures from the Human Primary Cell Atlas, SCINA, a semi- supervised cell type identification tool using cell- type signatures as well as a Gene- Set Variation Analysis (GSVA). Results were combined and clusters were assigned to the cell type with the highest enrichment within all models. In order to individually analyze T cells, we used the assigned cluster and filter for the following criteria. For further analysis T cells were defined by: \(\mathrm{CD3^{+}CD8^{+}}\) / \(\mathrm{CD4^{+}CD14^{+}LYZ^{- }}\) GFAP \(\cdot\) CD163 \(\cdot\) IBA. + +## Spatial Transcriptomics + +The spatial transcriptomics experiments were done using the 10X Spatial transcriptomics kit (https://spatialtranscriptomics.com/). All the instructions for Tissue Optimization and Library preparation were followed according to the manufacturer's protocol. Here, we briefly describe the methods followed using the library preparation protocol. + +<--- Page Split ---> + +## Tissue collection and RNA quality control: + +Tissue samples from three patients, diagnosed with WHO IV glioblastoma multiforme (GBM), were included in this study. Fresh tissue collected immediately post resection was quickly embedded in optimal cutting temperature compound (OCT, Sakura) and snap frozen in liquid \(\mathbb{N}_2\) . The embedded tissue was stored at \(- 80^{\circ}C\) until further processing. A total of 10 sections (10μm each) per sample were lysed using TriZOI (Invitrogen, 15596026) and used to determine RNA integrity. Total RNA was extracted using PicoPure RNA Isolation Kit (Thermo Fisher, KIT0204) according to the manufacturer's protocol. RIN values were determined using a 2100 Bioanalyzer (RNA 6000 Pico Kit, Agilent) according to the manufacturer's protocol. It is recommended to only use samples with an RNA integrity value \(>7\) . + +## Tissue staining and Imaging: + +Sections were mounted onto spatially barcoded glass slides with poly- T reverse transcription primers, with one section per array. These slides can be stored at \(- 80^{\circ}C\) until use. The slides were then warmed to \(37^{\circ}C\) , after which the sections were fixed for 10 minutes using \(4\%\) formaldehyde solution (Carl Roth, P087.1), which was then washed off using PBS. The fixed sections were covered with propan- 2- ol (VWR, 20842312). Following evaporation for 40 seconds, sections were incubated in Mayer's Hematoxylin (VWR, 1092490500) for 7 min, bluing buffer (Dako, CS70230- 2) for 90 seconds and finally in Eosin Y (Sigma, E4382) for 1 min. The glass slides were then washed using RNase/DNase free water and incubated at \(37^{\circ}C\) for 5 min or until dry. Before imaging, the glass slides were mounted with \(87\%\) glycerol (AppliChem, A3739) and covered with coverslips (R. Langenbrick, 01- 2450/1). Brightfield imaging was performed at 10x magnification with a Zeiss Axio Imager 2 Microscope, and postprocessing was performed using ImageJ software. + +The coverslips and glycerol were removed by washing the glass slides in RNase/DNase free water until the coverslips came off, after which the slides were washed using \(80\%\) ethanol to remove any remaining glycerol. + +## Permeabilization, cDNA synthesis and tissue removal: + +For each capture array, \(70\mu \mathrm{L}\) of pre- permeabilization buffer, containing \(50\mathrm{U / \mu L}\) Collagenase along with \(0.1\%\) Pepsin in HCl was added, followed by an incubation for 20 minutes at \(37^{\circ}C\) . Each array well was then carefully washed using \(100\mu \mathrm{L}0.1\times \mathrm{SSC}\) + +<--- Page Split ---> + +buffer. \(70\mu \mathrm{L}\) of Pepsin was then added and incubated for 11 minutes at \(37^{\circ}C\) . Each well was washed as previously described and \(75\mu \mathrm{L}\) of cDNA synthesis master mix containing: \(96\mu \mathrm{L}\) of 5X First strand buffer, \(24\mu \mathrm{L}\) 0.1M DTT, \(255.2\mu \mathrm{L}\) of DNase/RNase free water, \(4.8\mu \mathrm{L}\) Actinomycin, \(4.5\mu \mathrm{L}\) of \(20\mathrm{mg / mL}\) BSA, \(24\mu \mathrm{L}\) of \(10\mathrm{mM}\) dNTP, \(48\mu \mathrm{L}\) of Superscript® and \(24\mu \mathrm{L}\) of RNaseOUT™ was added to each well and incubated for 20 hours at \(42^{\circ}C\) without shaking. Cyanine 3- dCTP was used to aid in the determination of the footprint of the tissue section used. + +Since glioblastoma tissue is a fatty tissue, degradation and tissue removal was carried out using Proteinase K treatment for which \(420\mu \mathrm{L}\) Proteinase K and PKD buffer (1:7), were added to each well and then incubated at \(56^{\circ}C\) for 1hr with intermittent agitation (15 seconds / 3 minutes). After incubation, the glass slides were washed three times with \(100\mathrm{mL}\) of \(50^{\circ}C\) SSC/SDS buffer with agitation for 10 minutes, 1 minute and finally for 1 minute at 300 rpm. The glass slides were then air- dried at room temperature. Tissue cleavage was carried out by the addition of \(70\mu \mathrm{L}\) of cleavage buffer ( \(320\mu \mathrm{L}\) RNase/DNase free water, \(104\mu \mathrm{L}\) Second strand buffer, \(4.2\mu \mathrm{L}\) of \(10\mathrm{mM}\) dNTP, \(4.8\mu \mathrm{L}\) of \(20\mathrm{mg / mL}\) BSA and \(48\mu \mathrm{L}\) of USER™ Enzyme) to each well and incubation at \(37^{\circ}C\) for 2 hours with intermittent agitation. + +## Spot Hybridization: + +In order to determine the exact location and quality of each of the 1007 spots, fluorescent Cyanine- 3 A is hybridized to the 5' ends of the surface probes. \(75\mu \mathrm{L}\) of the hybridization solution ( \(20\mu \mathrm{L}\) of \(10\mu \mathrm{M}\) Cyanine- 3A probe and \(20\mu \mathrm{L}\) of \(10\mu \mathrm{M}\) Cyanine- 4 Frame probe in \(960\mu \mathrm{L}\) of 1X PBS) was added to each well and incubated for 10min at room temperature. The slides were then washed three times with \(100\mathrm{ml}\) of SSC/SDS buffer preheated to \(50^{\circ}C\) for 10min, 1min and 1min at room temperature with agitation. The slides were then air- dried and imaged after applying Slowfade® Gold Antifade medium and a coverslip. + +## Library Preparation: + +## 1. Second Strand Synthesis + +\(5\mu \mathrm{L}\) second strand synthesis mix containing \(20\mu \mathrm{L}\) of 5X First Strand Buffer, \(14\mu \mathrm{L}\) of DNA polymerase I (10U/μL) and \(3.5\mu \mathrm{L}\) Ribonuclease H (2U/μL) were added to the cleaved sample and incubated at \(16^{\circ}C\) for 2 hours. Eppendorf tubes were placed on + +<--- Page Split ---> + +ice and 5μL of T4 DNA polymerase (3U/μL) were added to each strand and incubated for 20 minutes at 16°C. 25μL of 80mM EDTA (mix 30μL of 500mM EDTA with 158μL DNase/RNase free water) was added to each sample and the samples were kept cool on ice. + +## 2. cDNA purification + +cDNA from the previous step was purified using Agencourt RNAclean XP beads and DynaMag™- 2 magnetic rack, incubated at room temperature for 5 min. Further cleansing was performed by the addition of 80% Ethanol to the sample tubes, while the samples were still placed in the magnetic rack. Sample elution was then carried out using 13μL of NTP/water mix. + +## 3. In Vitro Transcription and Purification + +cDNA transcription to aRNA was carried out by adding 4μL of reaction mix containing: 10x Reaction Buffer, T7 Enzyme mix and SUPERaseln™ RNase Inhibitor (20 U/μL) to 12μL of the eluted cDNA sample and incubated at 37°C, for 14 hours. The samples were purified using RNA clean XP beads according to the manufacturer's protocol and further eluted into 10μL DNase/RNase free water. The amount and average fragment length of amplified RNA was determined using the RNA 6000 Pico Kit (Agilent, 5067- 1513) with a 2100 Bioanalyzer according to the manufacturer's protocol. + +## 4. Adapter Ligation + +Next, 2.5μL Ligation adapter (IDT) was added to the sample and was heated for 2 min at 70°C and then placed on ice. A total of 4.5μL ligation mix containing 11.3μL of 10X T4 RNA Ligase, T4 RNA truncated Ligase 2 and 11.3μL of murine RNase inhibitor was then added to the sample. Samples were then incubated at 25°C for 1 hour. The samples were then purified using RNAClean XP beads according to the manufacturer's protocol. + +## 5. Second cDNA synthesis + +Purified samples were mixed with 1μL cDNA primer (IDT), 1μL dNTP mix up to a total volume of 12μL and incubated at 65°C for 5 min and then directly placed on ice. A 1.5ml Eppendorf tube 8μL of the sample was mixed with 30μL of First Strand Buffer(5X),), 7.5μL of DTT(0.1M), 7.5μL of DNase/RNase free water, 7.5μL of SuperScript® III Reverse transcriptase and 7.5μL of RNaseOUT™ Recombinant ribonuclease Inhibitor and incubated at 50°C for 1 hour followed by cDNA purification + +<--- Page Split ---> + +using Agencourt RNAClean XP beads according to the manufacturer's protocol. Samples were then stored at \(- 20^{\circ}C\) . + +## 6. PCR amplification + +Prior to PCR amplification, we determined that 20 cycles were required for appropriate amplification. A total reaction volume of 25μL containing 2x KAPA mix, 0.04μM PCRInPE2 (IDT), 0.4μM PCR InPE1.0 (IDT), 0.5μM PCR Index (IDT) and 5μL of purified cDNA were amplified using the following protocol: 98°C for 3 min followed by 20 cycles at 98°C for 20 seconds, 60°C for 30 seconds, 72°C for 30 seconds followed by 72°C for 5 minutes. The libraries were purified according to the manufacturer's protocol and eluted in 20μL EB (elution buffer). The samples were then stored at -20°C until used. + +## 7. Quality control of Libraries + +The average length of the prepared libraries was quantified using an Agilent DNA 1000 high sensitivity kit with a 2100 Bioanalyzer. The concentration of the libraries was determined using a Qubit dsDNA HS kit. The libraries were diluted to 4nM, pooled and denatured before sequencing on the Illumina NextSeq platform using paired-end sequencing. We used 30 cycles for read 1 and 270 cycles for read 2 during sequencing. + +Sequence + +
Ligation Adapter cDNA primer PCR primer INPE1 PCR primer INPE2 PCR index primer/5rApp/AGATCGGAAAGAGCACACGTCTGAACTCCAGTCAC/3d- dC/
GTGACTGGAGTTCAGACGTGTGCTCTTCCGA
AATGATACGGGCACCACGAGATCTACACTCTTTCCCTACACGACGCTCT- CCGATCT
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT
CAAGCAGAAGACGGCATACGAGATCGATGATGGACTGGAGTTC
+ +<--- Page Split ---> + +## Postprocessing Spatial Transcriptomics + +First, we aligned the H&E staining by the use of the st- pipeline (github.com/SpatialTranscriptomics- Research/st_pipeline). The pipeline contains the following steps: Quality trimming and removing of low quality bases (bases with quality \(< 10\) ), sanity check (reads same length, reads order, etc.), remove homopolymers, normalize for AT and GC content, mapping the read2 with STAR, demultiplexing based on read1, sort for reads (read1) with valid barcodes, annotate the reads with htsqcount, group annotated reads by barcode (spot position), gene and genomic location (with an offset) to get a read count (github.com/SpatialTranscriptomics- Research/st_pipeline). The pipeline resulted in a gene count matrix and a spatial information file containing the x and y position and the H&E image. We used the Seurat v3.0 package to normalize gene expression values by dividing each estimated cell by the total number of transcripts and multiplied by 10,000, followed by natural- log transformation. As described for sc- RNA sequencing, we removed batch effects and scaled data by a regression model including sample batch and percentage of ribosomal and mitochondrial gene expression. For further analysis we used the 2000 most variable expressed genes and decomposed eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparison to randomized expression values. The obtained non- trivial components were used for SNN clustering followed by dimensional reduction using the UMAP algorithm. Differently expressed genes (DE) of each cluster were obtained using a hurdle model tailored to scRNA- seq data which is part of the MAST package. We further build a user- friendly viewer for spatial transcriptomic data and provide tutorials on analysis of data: https://themilolab.github.io/SPATA/index.html. + +## Spatial gene expression + +For spatial expression plots, we used either normalized and scaled gene expression values (to plot single genes) or scores of a set of genes, using the 0.5 quantile of a probability distribution fitting. The x- axis and y- axis coordinates are given by the input file based on the localization at the H&E staining. We computed a matrix based on the maximum and minimum extension of the spots used (32x33) containing the gene expression or computed scores. Spots without tissue covering were set to zero. Next, + +<--- Page Split ---> + +we transformed the matrix, using the squared distance between two points divided by a given threshold, implemented in the fields package (R- software) and adapted the input values by increasing the contrast between uncovered spots. The data are illustrated either as surface plots (plotly package R- software) or as images (graphics package R- software). + +## Representation of Cellular States + +We aligned cells/spots to variable states with regard to gene sets (GS) that were selected GS(1,2,..n). First, we separated cells into GS(1+2) versus GS(2+4), using the following equation: + +\[A_{1} = \| G S_{(1)},G S_{(2)}\|_{\infty} - \| G S_{(3)},G S_{(4)}\|_{\infty}\] + +A1 defines the y- axis of the two- dimensional representation. In a next step, we calculated the x- axis separately for spots A1<0 and A1>0: + +\[A1 > 0:A_{2} = \log 2(\overline{{G S_{(1)}}} -[\overline{{G S_{(2)}}} +1])\] + +\[A1< 0:A_{2} = \log 2(\overline{{G S_{(3)}}} -[\overline{{G S_{(4)}}} ])\] + +For further visualization of the enrichment of subsets of cells according to gene set enrichment across the two- dimensional representation, we transformed the distribution to representative colors using a probability distribution fitting. This representation is an adapted method published by Neftel and colleges recently20,28. + +## Spatial correlation analysis + +The spatial correlation was performed by integrating a deep autoencoder for background noise reduction and a Bayesian correlation model. In a first step, we performed noise reduction through an autoencoder similar to recent described for single- cell RNA- sequencing studies39. The autoencoder consist of an encoder and decoder part which can be defined as transitions: + +\[(1) encoder:\phi :X\to F decoder:\psi :F\to X\] + +\[(2)\phi ,\psi = arg.min\| X - (\phi ,\psi)X\| ^2\] + +The encoder stage of an autoencoder takes the input \(x \in \mathbb{R}^d = X\) and maps it to \(z \in \mathbb{R}^p = \mathcal{F}\) at the layer position \(\phi\) : + +\[(3)x = A^{\phi = 0};z^{\phi} = ReLU(W^{\phi}\times A^{\phi -1} + b^{\phi})\] + +\(z^{\phi}\) is also referred to as latent representation, here presented as \(z^{1}, z^{2}, \ldots , z^{\phi = n}\) in which \(\phi\) describes the number of hidden layers. \(\mathbf{W}\) is the weight matrix and \(\mathbf{b}\) represent the dropout or bias vector. Our network architecture contained 32 hidden layers, as + +<--- Page Split ---> + +recommended39. In the decoder weights and biases are reconstructed through backpropagation \((\psi :\mathcal{F}\to \mathcal{X})\) and \(z\) is mapped to \(x^{\prime} = A^{0^{\prime}}\) in the shape as \(x^{\prime}\) . + +\[A^{\phi -1^{\prime}} = \sigma^{\prime}(W^{\phi^{\prime}}\times z^{\phi} + b^{\phi^{\prime}})\] + +In this context, \(W^{\prime},\sigma^{\prime},b^{\prime}\) from the decoder are unrelated to \(W,\sigma ,b\) from the encoder. We used a loss function to train the network in order to minimize reconstruction errors. + +\[\mathcal{L}(x,x^{\prime}) = \left\| x - \sigma^{\prime}(W^{\prime}(\sigma (Wx + b)) + b^{\prime})\right\|^{2}\] + +In the second step, we used the predicted gene expression matrix \((x^{\prime})\) and fitted and Bayesian correlation model (Bayesian First Aid, https://github.com/rasmusab/bayesian_first_aid). An illustration of the spatial correlation is given in the supplementary figure 7 + +## sc-RNA-sequencing integration into spatial context + +In order to integrate determined cluster into the spatial context of the spatial transcriptomic data. We used the recent published spotlighted algorithm and integrated the output into our SPATA objects for visualization21. + +## RNA-velocity and Pseudotime trajectory analysis + +In order to determine dynamic gene expression changes, we extracted spliced and unspliced genes from the bam output created by cell ranger using the velocyto.py tool40. The resulting \*.loom files were merged and transformed into .h5ad format for further processing by scVelo41 and CellRank42. The pipeline is integrated into the SPATA toolbox43. Single-cell data are reformatted into an SPATA S4 object using the UMAP coordinates as spatial coordinates. Outputs of the scVelo script (implemented in the development branch of the SPATA toolbox) are imported into the SPATA S4 object (slot:@fdata) and available for visualization. RNA- velocity streams are converted into trajectories and imported to the SPATA S4 object (slot:@trajectories). Dynamic gene expression changes along trajectories are performed by the assessTrajectoryTrends() function. + +## Gene set enrichment analysis + +Gene sets were obtained from the database MSigDB v7 and internally created gene sets are available at github.com/heilandd. For enrichment analysis of single clusters, the normalized and centered expression data were used and further transformed to z- + +<--- Page Split ---> + +scores ranging from 1 to 0. Genes were ranked in accordance to the obtained differential expression values and used as the input for GSEA. + +## Identification of cycling cells + +We used the set of genes published by Neftel and colleagues to calculate proliferation scores based on the GSVA package implemented in R- software. The analysis based on a non- parametric unsupervised approach, which transformed a classic gene matrix (gene- by- sample) into a gene set by sample matrix resulted in an enrichment score for each sample and pathway. From the output enrichment scores we set a threshold based on distribution fitting to define cycling cells. + +## Nearest Functionally Connected Neighbor (NFCN) + +To identify connected cells that interact by defined activation or inhibition of downstream signals in the responder cell, we created a novel model. Therefore, we assumed that a cell- cell interaction is given only if a receptor/ligand pair induce correspondent down- stream signaling within the responder cell (cell with expressed receptor). Furthermore, we take into account that the importance of an activator cell (cell with expressed ligand) can be ranked according to their enriched signaling, which is responsible for inducing ligand expression. Based on these assumptions we defined an algorithm to map cells along an interaction- trajectory. The algorithm was designed to identify potential activators from a defined subset of cells. + +As input for the analysis we used a normalized and scaled gene expression matrix, a string containing the subset of target cells, a list of genes defining ligand induction on the one side and receptor signaling on the other side. These genes were chosen either by the MSigDB v7 database or our stimulation library explained above. Next, we downscaled the data to 3000 representative cells including all myeloid cell types and calculated the enrichment of induction and activation of the receptor/ligand pair. Enrichment scores were calculated by singular value decomposition (SVD) over the genes in the gene set and the coefficients of the first right- singular vector defined the enrichment of induction/activation profile. Both expression values and enrichment scores were fitted by a probability distribution model and cells outside the 95% quantile were removed. Next, we fitted a model using a non- parametric kernel estimation (Gaussian or Cauchy- Kernel), on the basis of receptor/ligand expression (Aexp) and + +<--- Page Split ---> + +up/downstream signaling (Aeff) of each cell (i={1,..n}). Both input vectors were normalized and z- scored: + +\[n_{\mathrm{exp}i} = \frac{A_{\mathrm{exp}i} - min(A_{\mathrm{exp}})}{max(A_{\mathrm{exp}}) - min(A_{\mathrm{exp}})} \qquad (2) \widehat{f}_h(n_{\mathrm{exp}i}) = \frac{1}{n}\sum_{i = 1}^{n}K_h(n_{\mathrm{exp}} - n_{\mathrm{exp}i})\] + +K is the kernel and \(0.7 > h > 0.3\) is used to adjust the estimator. The model resulted in a trajectory which was defined as Ligand \((- )\) - Induction \((- )\) to cells of the target subset with Receptor \((- )\) - Activation \((- )\) . Further cells were aligned along the "interaction- trajectory". We defined connected cells by reaching the upper \(70\%\) Cl in receptor/ligand expression as well as sores of induction/activation. The process of representation is illustrated schematically in Supplementary Figure 4. Additionally we determined receptor- ligand interaction by the NicheNet software as recommended by the authors \(^{44}\) . + +## CNV estimation: + +Copy- number Variations (CNVs) were estimated by aligning genes to their chromosomal location and applying a moving average to the relative expression values, with a sliding window of 100 genes within each chromosome, as described recently \(^{18}\) . First, we arranged genes in accordance to their respective genomic localization using the CONICSmat package (R- software). As a reference set of non- malignant cells, we in- silico extracted 400 CD8 positive cells (unlikely to be expressed on tumor cells). To avoid the considerable impact of any particular gene on the moving average we limited the relative expression values [- 2.6,2.6] by replacing all values above/below \(exp_{(i)} = [2.6]\) , by using the infercnv package (R- software). This was performed only in the context of CNV estimation as previous reported \(^{11}\) . + +## Flow cytometry: + +Single- Cell suspensions were obtained after Dead- Cell Removal and CD3 MACS- enrichment. Cells were incubated with VivaFix™ 398/550 (BioRad Laboratories, CA, USA) according to the manufacturer's instructions. Cells were fixed in \(4\%\) paraformaldehyde (PFA) for 10 minutes. After centrifugation (350 g; \(4^{\circ}C\) ; 5 min) and removal of the supernatant, the cell pellet was suspended in \(0.5 \text{ml} 4^{\circ}C\) cold FACS buffer. Cell suspensions were washed and centrifuged at 350xg for 5 mins, followed by resuspension in FACS buffer. The washing step was repeated twice. Finally, cells were resuspended in at least 0.5 to \(1 \text{ml}\) of FACS buffer depending on the number of + +<--- Page Split ---> + +cells. We used a Sony SP6800 spectral analyzer in standardization mode with PMT voltage set to maximum to reach a saturation rate below \(0.1\%\) . Gating was performed by FCS Express 7 plus at the Lighthouse Core Facility, University of Freiburg. + +## Immunofluorescence + +The same protocol was followed for human neocortical slices with or without microglia and tumor cell injection. The media was removed and exchanged for \(1 \text{mL}\) of \(4\%\) paraformaldehyde (PFA) for \(1 \text{h}\) and further incubated in \(20\%\) methanol in PBS for 5 minutes. Slices were then permeabilized by incubating in PBS supplemented with \(1\%\) Triton (TX- 100) overnight at \(4^{\circ}\text{C}\) and further blocked using \(20\%\) BSA for 4 hours. The permeabilized and blocked slices were then incubated by primary antibodies in \(5\%\) BSA- PBS incubated overnight at \(4^{\circ}\text{C}\) . After washing in PBS, slices were labelled with secondary antibodies conjugated with Alexa 405, 488, 555, or 568 for 3 hours at room temperature. Finally, slices were mounted on glass slides using DAPI fluoromount (Southern Biotech, Cat. No. 0100- 20), as recently described15. + +## Human Organotypic Slice Culture + +Human neocortical slices were prepared as recently described15,25. Capillaries and damaged tissue were dissected away from the tissue block in the preparation medium containing Hibernate medium supplemented with \(13 \text{mM D + }\) Glucose and \(30 \text{mM}\) NMDG. Coronal slices of \(300 \mu \text{m}\) thickness were sectioned using a vibratome (VT1200, Leica Germany) and incubated in preparation medium for 10 minutes before plating to avoid any variability due to tissue trauma. Three to four slices were gathered per insert. The transfer of the slices was facilitated by a polished wide mouth glass pipette. Slices were maintained in growth medium containing Neurobasal (L- Glutamine) supplemented with \(2\%\) serum free B- 27, \(2\%\) Anti- Anti, \(10 \text{mM D + }\) Glucose, \(1 \text{mM MgSO4}\) , and \(1 \text{mM Glutamax}\) at \(5\% \text{CO}_2\) and \(37^{\circ}\text{C}\) . The entire medium was replaced with fresh culture medium 24 hours post plating, and every 48 hours thereafter. + +<--- Page Split ---> + +## Chemical depletion of Microglia from slice cultures + +Selective depletion of the myeloid cell compartment in human neocortical slices was performed by supplementing the growth medium with \(50~\mu \mathrm{M}\) for 24hrs and \(11\mu \mathrm{M}\) of Clodronate (Sigma, D4434) for the next 48hrs at \(37^{\circ}C\) . Subsequently, the slices were carefully rinsed with growth medium to wash away any debris. + +## Tumor/T cell injection onto tissue cultures + +ZsGreen tagged BTSC#233 cell lines cultured and prepared as described in the cell culture section. Post trypsinization, a centrifugation step was performed, following which the cells were harvested and suspended in MEM media at 20,000 cells/ul. A 10 \(\mu \mathrm{L}\) Hamilton syringe was used to inject \(1\mu \mathrm{L}\) of GBM cells into the slices and were incubated at \(37^{\circ}C\) for 3 days with a medium change every other days. + +Blood samples were collected from the same donors from whom we obtained the healthy cortex for our organotypic slice cultures. Peripheral T cells were isolated using the same MACSxpress® Whole Blood Pan T Cell Isolation Kit (Miltenyi Biotech) and erythrocytes were eliminated from the suspension using ACK- lysis buffer (Thermo Fisher Scientific). T cells were tagged using the Cell Trace Far Red dye (Thermo Fisher Scientific) prior to injection into the slices. To block endogenous IL- 10 receptor in T- cells, the neutralizing antibody anti- IL10 hAB (R&D systems) were added to the cells at the concentration of \(5\mu \mathrm{g / ml}\) for 1h and incubated at \(37^{\circ}C\) . Both IL10 knock out T cells and normal T cells were injected onto the sections and left it for another 48hrs at \(37^{\circ}C\) . + +## Enzyme linked Immunosorbent Assay + +An enzyme linked immunosorbent assay (ELISA) was performed in order to measure cytokine concentrations of IL- 2, IL- 10, IL- 13 and IFN- gamma in the cell culture medium 48h after T cell injection. The Multi- Analyte ELISAarray Kit (Qiagen, Venlo, Netherlands; MEH- 003A) was used according to the manufacturer's instructions. Absorbance was measured using the Tecan Infinite® 200 (Tecan, Männedorf, Switzerland). + +## Treatment of patient with JAK-inhibitor + +A patient with a recurrent glioblastoma was treated with a daily dose of 40mg Ruxolitinib for 4 weeks. Before treatment, we confirmed the progress by a biopsy. The + +<--- Page Split ---> + +treatment was performed as a neoadjuvant therapy. After 4 weeks, the patient underwent a gross- total surgery and adjuvant Temozolomide therapy. + +## Acknowledgement + +DHI is funded by the Else- Kröner Stiftung and BMBF. VM Ravi was partially funded by the BMBF- projects FMT (13GW0230A). We thank Manching Ku and Dietmar Pfeifer for their helpful advice. We acknowledge Biorender.com. + +## Conflict of interests + +No potential conflicts of interest were disclosed by the authors. + +## Data availability + +scRNA- Sequencing Data available: (in preparation), Accession codes: www.github.com/heilandd/.VisLabv1.5 https://github.com/heilandd/Vis_Lab1.5, NFCN Algorithm www.github.com/heilandd/NFCN, SPATA- Lab: www.github.com/heilandd/SPATA- Lab. Further information and requests for resources, raw data and reagents should be directed and will be fulfilled by the Contact: D. H. Heiland, dieter.henrik.heiland@uniklinik- freiburg.de. + +<--- Page Split ---> + +Bibliography1. Darmanis, S. et al. Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep. 21, 1399–1410 (2017).2. Woroniecka, K. et al. T-Cell Exhaustion Signatures Vary with Tumor Type and Are Severe in Glioblastoma. Clin. Cancer Res. 24, 4175–4186 (2018).3. Chen, Z. & Hambardzumyan, D. Immune microenvironment in glioblastoma subtypes. Front. Immunol. 9, 1004 (2018).4. Sankowski, R. et al. Mapping microglia states in the human brain through the integration of high-dimensional techniques. Nat. Neurosci. 22, 2098–2110 (2019).5. Zhang, L. et al. Lineage tracking reveals dynamic relationships of T cells in colorectal cancer. Nature 564, 268–272 (2018).6. Zheng, C. et al. Landscape of Infiltrating T Cells in Liver Cancer Revealed by Single-Cell Sequencing. Cell 169, 1342–1356.e16 (2017).7. Baitsch, L. et al. Exhaustion of tumor-specific CD8+ T cells in metastases from melanoma patients. J. Clin. Invest. 121, 2350–2360 (2011).8. Singer, M. et al. A Distinct Gene Module for Dysfunction Uncoupled from Activation in Tumor-Infiltrating T Cells. Cell 166, 1500–1511.e9 (2016).9. Anderson, A. C., Joller, N. & Kuchroo, V. K. Lag-3, Tim-3, and TIGIT: Co-inhibitory Receptors with Specialized Functions in Immune Regulation. Immunity 44, 989–1004 (2016).10. Im, S. J. et al. Defining CD8+ T cells that provide the proliferative burst after PD-1 therapy. Nature 537, 417–421 (2016).11. Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).12. Platten, M., Ochs, K., Lemke, D., Opitz, C. & Wick, W. Microenvironmental clues for glioma immunotherapy. Curr Neurol Neurosci Rep 14, 440 (2014).13. Filley, A. C., Henriquez, M. & Dey, M. Recurrent glioma clinical trial, CheckMate-143: the game is not over yet. Oncotarget 8, 91779–91794 (2017).14. Weller, M. et al. Rindopepimut with temozolomide for patients with newly diagnosed, EGFRvIII-expressing glioblastoma (ACT IV): a randomised, double-blind, international phase 3 trial. Lancet Oncol. 18, 1373–1385 (2017).15. Henrik Heiland, D. et al. Tumor-associated reactive astrocytes aid the evolution of + +<--- Page Split ---> + +immunosuppressive environment in glioblastoma. Nat. Commun. 10, 2541 (2019). + +16. Zhang, Z. et al. SCINA: A Semi-Supervised Subtyping Algorithm of Single Cells and Bulk Samples. Genes (Basel) 10, (2019). +17. Aran, D. et al. Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage. Nat. Immunol. 20, 163–172 (2019). +18. Patel, A. P. et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014). +19. Chen, H. et al. Single-cell trajectories reconstruction, exploration and mapping of omics data with STREAM. Nat. Commun. 10, 1903 (2019). +20. Neftel, C. et al. An integrative model of cellular states, plasticity, and genetics for glioblastoma. Cell 178, 835–849.e21 (2019). +21. Elosua, M., Nieto, P., Mereu, E., Gut, I. & Heyn, H. SPOTlight:Seeded NMF regression to Deconvolute Spatial Transcriptomics Spots with Single-Cell Transcriptomes. BioRxiv (2020). doi:10.1101/2020.06.03.131334 +22. Naito, Y., Takagi, T. & Higashimura, Y. Heme oxygenase-1 and anti-inflammatory M2 macrophages. Arch. Biochem. Biophys. 564, 83–88 (2014). +23. Qin, Y. et al. A Milieu Molecule for TGF-β Required for Microglia Function in the Nervous System. Cell 174, 156–171.e16 (2018). +24. Giladi, A. et al. Dissecting cellular crosstalk by sequencing physically interacting cells. Nat. Biotechnol. 38, 629–637 (2020). +25. Ravi, V. M. et al. Human organotypic brain slice culture: a novel framework for environmental research in neuro-oncology. Life Sci. Alliance 2, (2019). +26. Kling, T. et al. Integrative Modeling Reveals Annexin A2-mediated Epigenetic Control of Mesenchymal Glioblastoma. EBioMedicine 12, 72–85 (2016). +27. Venteicher, A. S. et al. Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science 355, (2017). +28. Tirosh, I. et al. Single-cell RNA-seq supports a developmental hierarchy in human oligodendroglioma. Nature 539, 309–313 (2016). +29. Blank, C. U. et al. Defining “T cell exhaustion”. Nat. Rev. Immunol. 19, 665–674 (2019). +30. Jiang, Y., Li, Y. & Zhu, B. T-cell exhaustion in the tumor microenvironment. Cell Death Dis. 6, e1792 (2015). +31. Wherry, E. J. T cell exhaustion. Nat. Immunol. 12, 492–499 (2011). +32. Li, H. et al. Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated + +<--- Page Split ---> + +1007 Compartment within Human Melanoma. Cell 176, 775- 789. e18 (2019). 1008 33. Woroniecka, K. I., Rhodin, K. E., Chongsathidkiet, P., Keith, K. A. & Fecci, P. E. T- cell Dysfunction in Glioblastoma: Applying a New Framework. Clin. Cancer Res. 24, 3792- 3802 (2018). 1011 34. Winkler, F. & Bengsch, B. Use of mass cytometry to profile human T cell exhaustion. Front. Immunol. 10, 3039 (2019). 1013 35. Wurm, J. et al. Astrogliosis Releases Pro- Oncogenic Chitinase 3- Like 1 Causing MAPK Signaling in Glioblastoma. Cancers (Basel) (2019). 1015 36. Priego, N. et al. STAT3 labels a subpopulation of reactive astrocytes required for brain metastasis. Nat. Med. 24, 1024- 1035 (2018). 1017 37. Kaiser, S. et al. Neuroprotection after Hemorrhagic Stroke Depends on Cerebral Heme Oxygenase- 1. Antioxidants (Basel) 8, (2019). 1019 38. Sebastián, V. P. et al. Heme Oxygenase- 1 as a Modulator of Intestinal Inflammation Development and Progression. Front. Immunol. 9, 1956 (2018). 1021 39. Eraslan, G., Simon, L. M., Mircea, M., Mueller, N. S. & Theis, F. J. Single- cell RNA- seq denoising using a deep count autoencoder. Nat. Commun. 10, 390 (2019). 1023 40. La Manno, G. et al. RNA velocity of single cells. Nature 560, 494- 498 (2018). 1024 41. Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & Theis, F. J. Generalizing RNA velocity to transient cell states through dynamical modeling. Nat. Biotechnol. 38, 1408- 1414 (2020). 1027 42. Lange, M. et al. CellRank for directed single- cell fate mapping. BioRxiv (2020). doi:10.1101/2020.10.19.345983 1029 43. Kueckelhaus, J. et al. Inferring spatially transient gene expression pattern from spatial transcriptomic studies. BioRxiv (2020). doi:10.1101/2020.10.20.346544 1031 44. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat. Methods 17, 159- 162 (2020). + +<--- Page Split ---> + +## Figures: + +Figure 1: a) Illustration of the workflow, tissue specimens were obtained from 11 glioblastoma patients. Samples from 8 patients were used for scRNA- seq and 3 for spatial transcriptomics. b) Dimensional reduction using UMAP, cell type was determined by SingleR (github.com/dviraran/SingleR). c) Dimensional reduction (UMAP) of CD3+/CD8+ cells. SNN-clustering reveal 13 different clusters. RNA- velocity based pseudotime is presented by streams obtained from dynamic modeling in scvelo. Marker expression in different subpopulations ranked by grade of activity. d) Dimensional reduction plots of estimated gene- set enrichment (left), of cell proliferation (middle) and regulatory marker genes (right). e,f) Detailed presentation of 3 major trajectories and its changes of estimated pseudotime across different models. Velocity trajectory 2 reveals almost no temporal changes, but is marked by increasing IFNg signaling along the trajectory. g) Line plot of IFNg signalling gene expression along the velocity trajectory 2 (left) and dimensional reduction (right). h) Inferring dynamic alterations along the trajectory revealed up- regulation of IL- 10 and TGF- \(\beta\) signaling, presented as a line plot (top). (i) Gene set enrichment analysis of the IL10 signaling enrichment in defined start and destination region of the trajectory (bottom). j) Mapping of gene expression along the pseudotime trajectory. Bars at the bottom indicate the enrichment of each cell to enrich effector or exhausted signatures. Line plots on the top, showed enrichment for the T cell state signatures. + +Figure 2: a) Workflow of spatial transcriptomics b) 2D representation of heterogeneous states in glioblastoma by Neftel, colors indicate the expression of cycling cells (quantile). c) H&E staining and (d) gene expression enrichment maps of the mesenchymal state defined by Neftel20. e) Enrichment maps of T cell clusters by spotlight algorithm21 in different samples. Maps are colored by a normalized spotlight score. f) Correlation heatmap of a Bayesian correlation model of T cell clusters and glioblastoma transcriptional subtypes. (g) Scatter plots of spot- wise correlation of HAVCR2 and LAG3 expression and T cells (left side), mesenchymal- like gene expression (middle) and NPC- like gene expression (right side), Rho correlation and CI95% are given at the top. + +Figure 3: a) Workflow exploring cell- cell interactions using two approaches: 1. Predict functional neighbors based on defined ligand- receptor interaction. 2. Validation in spatial transcriptomic datasets. b) Cell- cell interaction plot as explained in supplementary figure 4. Cells with an interaction- score above 0.95 are mapped to the UMAP. c) Volcano plot of differential gene expression between highly connected cells (CI>97.5%, left side) vs non- connected cells (CI<2.5%, right side), adjusted - log(p- vale) (FDR) was used at the y- axis. Red cells are defined by fold- change above 2 and FDR < 0.05. d) Spatial surface plots of gene expression and pathway enrichment. e) Multilayer representation of cellular interaction. At the top layer, a UMAP is shown colored by expression levels of HMOX1. Each red line represents a cell (upper CI>99% NFCNA- Score) and its most likely position within the spatial dataset. The second layer is a spatial transcriptomic dataset, colored by the predicted NFCNA- score. At the bottom, the third layer represents a UMAP colored accordingly to CD3 expression. Green lines showed the upper CI>99% of receiver cells. f) Violin plots of the mean NFCNA score in each cluster. g) Violin plots of gene expression between connected cells (CI>97.5%, left side) vs non- connected cells (CI<2.5%, right side). Wilcoxon Rank Sum test and FDR adjustment was used for statistical testing. h) Gene Set enrichment analysis + +<--- Page Split ---> + +of four different gene sets. i) Heatmap of differently expressed genes or connected myeloid (right) and lymphoid (middle) cells and the cluster of doublets. At the top, scatterplot of total UMIs per cell in each group. + +Figure 4: a) Experimental workflow of the neocortical GBM model autografted with patient derived T cells and with/without myeloid cell depletion. b) Immunostainings of IBA1 (Macrophages and Microglia) in magenta and HMOX1 in cyan, tumor cells are depicted in grey. In the upper panel, the control set with no myeloid cell depletion (M+) is shown, the bottom panel contains the myeloid cell depleted sections. c) ELISA measurements of IL10 d) Immunostainings of T cells (CSFE- Tagged, in red) and GZMB, a marker of T cell activation (green). e, f) ELISA measurements of IL2 and IFNg g) Immunostainings of TIM3 (gene: HAVCR2) in yellow, which was identified in the ScRNA- seq, and T cell in red. h) Illustration of the workflow. i) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) with pre- treatment in anti- IL- 10R antibodies. j) ELISA measurements of IL2 in ctr and IL10R inhibition. k) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) treated with JAK- inhibitor Ruxolitinib. l) Heatmap of interleukin intensities different environment. Information regarding the presence of myeloid cells are given at the bottom. m) Illustration of the workflow of JAK inhibition in a recurrent GBM patient. n- o) Immunohistochemistry of immune marker and its quantification, white: pre therapy, gray: post therapy p) Dimensional reduction of single- cell RNA- sequencing of the Ruxolitinib treated patient revealed a large percentage of T cells (right side). q) Comparison of T cells from the Ruxolitinib treated patient and the non- treated cohort. r) A barplot that indicates the cluster specific enrichment of JAK treated T cells. P- values are determined by one- way ANOVA (c,e,f,j) adjusted by Benjamini- Hochberger (c,e,f,j) for multiple testing. Data is given as mean ± standard deviation. + +## Supplementary Figures: + +Supplementary Figure 1: a) Workflow and representative FACS images. b) UMAP representation of all detected cell types. c) Distribution of patients across all clusters and cell types. d) UMAP representation of all determined clusters (SNN) e) The graph representing the percentage of each patient in different clusters f) Signature genes of each cluster g- i) UMAP representation of Isolated T cells spitted into CD4+ and CD8+ cells and the total number of clusters (13). j) Heatmap of signature genes + +Supplementary Figure 2: a) Copy- number alterations based on single cell data. Only a small subset of tumor cells was found in the OPC cluster. b) Gene expression maps of common marker genes. + +Supplementary Figure 3: a) Workflow to build a library of stimulated T- cells b) T cell stimulation in order to build a library for cytokine effects, illustrated is a heatmap of the 10 most significant marker genes of each stimulation state, based on PAMR algorithm implemented in the AutoPipe. c) Dimensional + +<--- Page Split ---> + +reduction (UMAP) of gene expression of the different simulation experiments. d) Z- scored expression of each stimulation signature along the velocity trajectory 1 (Figure 1). + +Supplementary Figure 4: a) Workflow and concept of the NFCN Analysis + +Supplementary Figure 5: a) Results from the Nichnet algorithm, left: The receptor- target interaction, right: the receptor- ligand network. + +Supplementary Figure 6: a) Kaplan- Meier survival estimation of HMOX1 high/low expression GBM. b- c) Expression of HMOX1 in different regions of the tumor (b) and in de- novo and recurrent stage (c). + +Supplementary Figure 7: a) Illustration of the workflow to determine the spatial correlation using a deep autoencoder for denoising followed by a Bayesian correlation model. b) Examples of the predicted overlap of T cells and CD163/HMOX1(+) myeloid cells. c) Distribution of the predicted correlation. + +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +a) Illustration of the workflow, tissue specimens were obtained from 11 glioblastoma patients. Samples from 8 patients were used for scRNA-seq and 3 for spatial transcriptomics. b) Dimensional reduction using UMAP, cell type was determined by SingleR (github.com/dviraran/SingleR). c) Dimensional reduction (UMAP) of CD3+/CD8+ cells. SNN-clustering reveal 13 different clusters. RNA- velocity based pseudotime is presented by streams obtained from dynamic modeling in scvelo. Marker expression in different subpopulations ranked by grade of activity. d) Dimensional reduction plots of estimated gene-set enrichment (left), of cell proliferation (middle) and regulatory marker genes (right). e,f) Detailed presentation of 3 major trajectories and its changes of estimated pseudotime across different models. Velocity trajectory 2 reveals almost no temporal changes, but is marked by increasing IFNg signaling + +<--- Page Split ---> + +along the trajectory. g) Line plot of IFNg signalling gene expression along the velocity trajectory 2 (left) and dimensional reduction (right). h) Inferring dynamic alterations along the trajectory revealed upregulation of IL-10 and TGF- \(\beta\) signaling, presented as a line plot (top). (i) Gene set enrichment analysis of the IL10 signaling enrichment in defined start and destination region of the trajectory (bottom). j) Mapping of gene expression along the pseudotime trajectory. Bars at the bottom indicate the enrichment of each cell to enrich effector or exhausted signatures. Line plots on the top, showed enrichment for the T cell state signatures8. + +![](images/Figure_2.jpg) + +
Figure 2
+ +a) Workflow of spatial transcriptomics b) 2D representation of heterogeneous states in glioblastoma by Neftel, colors indicate the expression of cycling cells (quantile). c) H&E staining and (d) gene expression enrichment maps of the mesenchymal state defined by Neftel20. e) Enrichment maps of T cell clusters by spotlight algorithm21 in different samples. Maps are colored by a normalized spotlight 1 score. f) Correlation heatmap of a Bayesian correlation model of T cell clusters and glioblastoma transcriptional subtypes. (g) Scatter plots of spot-wise correlation of HAVCR2 and LAG3 expression and T cells (left side), mesenchymal-like gene expression (middle) and NPC-like gene expression (right 1065 side), Rho correlation and CI95% are given at the top. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +a) Workflow exploring cell-cell interactions using two approaches: 1. Predict functional neighbors based on defined ligand-receptor interaction. 2. Validation in spatial transcriptomic datasets. b) Cell-cell interaction plot as explained in supplementary figure 4. Cells with an interaction-score above 0.95 are mapped to the UMAP. c) Volcano plot of differential gene expression between highly connected cells (CI>97.5%, left side) vs non-connected cells (CI<2.5%, right side), adjusted -log(p-vale) (FDR) was used at the y-axis. Red cells are defined by fold-change above 2 and FDR < 0.05. d) Spatial surface plots of gene expression and pathway enrichment. e) Multilayer representation of cellular interaction. At the top layer, a UMAP is shown colored by expression levels of HMOX1. Each red line represents a cell (upper CI>99% NFcNA-Score) and its most likely position within the spatial dataset. The second layer is a spatial transcriptomic dataset, colored by the predicted NFcNA-score. At the bottom, the third layer represents a UMAP colored accordingly to CD3 expression. Green lines showed the upper CI>99% of receiver cells. f) Violin plots of the mean NFcNA score in each cluster. g) Violin plots of gene expression between connected cells (CI>97.5%, left side) vs non-connected cells (CI<2.5%, right side). Wilcoxon Rank Sum test and FDR adjustment was used for statistical testing. h) Gene Set enrichment analysis of four different gene sets. i) Heatmap of differently expressed genes or connected myeloid (right) and 1lymphoid (middle) cells and the cluster of doublets. At the top, scatterplot of total UMLs per cell in each 1083 group. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +a) Experimental workflow of the neocortical GBM model autografted with patient derived T cells and with/without myeloid cell depletion. b) Immunostaining of IBA1 (Macrophages and Microglia) in magenta and HMOX1 in cyan, tumor cells are depicted in grey. In the upper panel, the control set with no myeloid cell depletion (M+) is shown, the bottom panel contains the myeloid cell depleted sections. c) ELISA measurements of IL10 d) Immunostaining of T cells (CSFE-Tagged, in red) and 1 GZMB, a marker + +<--- Page Split ---> + +of T cell activation (green).). e, f) ELISA measurements of IL2 and IFNg g) Immunostainings of TIM3 (gene: HAVCR2) in yellow, which was identified in the ScRNA- seq, and T cell in red. h) Illustration of the workflow. i) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, inred) and GZMB a marker of T cell activation (green) with pre- treatment in anti- IL- 10R antibodies. j) ELISA measurements of IL2 in ctr and IL10R inhibition. k) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) treated with JAK- inhibitor Ruxolitinib. l) Heatmap of interleukin intensities different environment. Information regarding the presence of myeloid cells are given at the bottom. m) Illustration of the workflow of JAK inhibition in a recurrent GBM patient. n- o) Immunohistochemistry of immune marker and its quantification, white: pre therapy, gray: post therapy p) Dimensional reduction of single- cell RNA- sequencing of the Ruxolitinib treated patient revealed a large percentage of T cells (right side). q) Comparison of T cells from the Ruxolitinib treated patient and the non- treated cohort. r) A barplot that indicates the cluster specific enrichment of JAK treated T cells. P- values are determined by one- way ANOVA (c,e,f,j) adjusted by Benjamini- Hochberger (c,e,f,j) for multiple testing. Data is given as mean ± standard deviation. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementary1. pdf Supplementary2. pdf Supplementary3. pdf Supplementary4. pdf Supplementary5. pdf Supplementary6. pdf Supplementary7. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f_det.mmd b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..66bb273df528c2554c61c4de0c2e219758105456 --- /dev/null +++ b/preprint/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f/preprint__126b45d3c59304c8cdcaeff2fb15da92849086ba5d613f1f0d668d95576aa34f_det.mmd @@ -0,0 +1,798 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 857, 210]]<|/det|> +# Crosstalk between lymphoid and myeloid cells orchestrates glioblastoma immunity through Interleukin 10 signaling + +<|ref|>text<|/ref|><|det|>[[44, 230, 884, 295]]<|/det|> +Vidhya Ravi Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany https://orcid.org/0000- 0003- 0062- 2099 + +<|ref|>text<|/ref|><|det|>[[44, 300, 720, 343]]<|/det|> +Nicolas Neidert Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 347, 884, 411]]<|/det|> +Paulina Will Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 415, 884, 480]]<|/det|> +Kevin Joseph Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 484, 884, 549]]<|/det|> +Julian Maier Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 553, 884, 618]]<|/det|> +Jan Kuckelhaus Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 622, 884, 686]]<|/det|> +Lea Vollmer Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 691, 884, 755]]<|/det|> +Jonathan Goeldner Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 760, 304, 803]]<|/det|> +Simon Behringer Department of Neurosurgery + +<|ref|>text<|/ref|><|det|>[[44, 808, 919, 874]]<|/det|> +Florian Scherer Department of Medicine I, Medical Center - University of Freiburg, Faculty of Medicine, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 877, 485, 920]]<|/det|> +Melanie Boerries Faculty of Medicine, Freiburg University, Germany + +<|ref|>text<|/ref|><|det|>[[44, 924, 144, 942]]<|/det|> +Marie Follo + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 46, 277, 64]]<|/det|> +University Medical Center + +<|ref|>text<|/ref|><|det|>[[44, 70, 303, 110]]<|/det|> +Tobias Weiss University Hospital of Zurich + +<|ref|>text<|/ref|><|det|>[[44, 117, 710, 159]]<|/det|> +Daniel Delev Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +<|ref|>text<|/ref|><|det|>[[44, 163, 710, 205]]<|/det|> +Julius Kembach Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +<|ref|>text<|/ref|><|det|>[[44, 210, 303, 250]]<|/det|> +Pamela Franco Department of Neurosurgery + +<|ref|>text<|/ref|><|det|>[[44, 255, 485, 296]]<|/det|> +Nils Schallner Faculty of Medicine, Freiburg University, Germany + +<|ref|>text<|/ref|><|det|>[[44, 301, 485, 343]]<|/det|> +Christine Dierks Faculty of Medicine, Freiburg University, Germany + +<|ref|>text<|/ref|><|det|>[[44, 348, 710, 390]]<|/det|> +Maria Stella Carro University Medical Center Freiburg https://orcid.org/0000- 0002- 8570- 7691 + +<|ref|>text<|/ref|><|det|>[[44, 394, 718, 436]]<|/det|> +Ulrich Hofmann Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 440, 718, 482]]<|/det|> +Christian Fung Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[44, 486, 598, 528]]<|/det|> +Roman Sankowski University of Freiburg https://orcid.org/0000- 0001- 9215- 8021 + +<|ref|>text<|/ref|><|det|>[[44, 533, 600, 575]]<|/det|> +Marco Prinz University of Freiburg https://orcid.org/0000- 0002- 0349- 1955 + +<|ref|>text<|/ref|><|det|>[[44, 580, 353, 621]]<|/det|> +Jurgen Beck University Medical Center Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 626, 738, 667]]<|/det|> +Oliver Schnell University Medical Center Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 671, 736, 714]]<|/det|> +Dieter Heiland ( Dieter.henrik.heiland@uniklinik- freiburg.de ) Medical Center, University of Freiburg https://orcid.org/0000- 0002- 9258- 3033 + +<|ref|>sub_title<|/ref|><|det|>[[44, 755, 101, 772]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 792, 773, 813]]<|/det|> +Keywords: cancer immunotherapy, Glioblastoma (GBM), T- cells, JAK/STAT inhibition + +<|ref|>text<|/ref|><|det|>[[44, 831, 333, 850]]<|/det|> +Posted Date: February 17th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 869, 463, 889]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 244157/v1 + +<|ref|>text<|/ref|><|det|>[[44, 906, 909, 949]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 944, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 17th, 2022. See the published version at https://doi.org/10.1038/s41467-022-28523-1. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[144, 83, 853, 128]]<|/det|> +# Crosstalk between lymphoid and myeloid cells orchestrates glioblastoma immunity through Interleukin 10 signaling + +<|ref|>text<|/ref|><|det|>[[113, 152, 884, 270]]<|/det|> +Vidhya M. Ravi \(^{1,2,3,4,\#}\) , Nicolas Neidert \(^{1,2,3,\#}\) , Paulina Will \(^{1,2,3,\#}\) , Kevin Joseph \(^{1,2,3}\) , Julian P. Maier \(^{1,2,3}\) , Jan Kückelhaus \(^{1,2,3}\) , Lea Vollmer \(^{1,2,3}\) , Jonathan M. Goeldner \(^{1,2,3}\) , Simon P. Behringer \(^{1,2,3}\) , Florian Scherer \(^{3,5}\) , Melanie Boerries \(^{3,6,7,8}\) , Marie Follo \(^{3,5}\) , Tobias Weiss \(^{9}\) , Daniel Delev \(^{10,11}\) , Julius Kernbach \(^{10,11}\) , Pamela Franco \(^{3,4}\) , Nils Schallner \(^{3,12}\) , Christine Dierks \(^{3,5}\) , Maria Stella Carro \(^{2,3}\) , Ulrich G. Hofmann \(^{3,4}\) , Christian Fung \(^{2,3}\) , Roman Sankowski \(^{3,13}\) , Marco Prinz \(^{3,13,14,15}\) , Jürgen Beck \(^{2,3,15}\) , Oliver Schnell \(^{2,3,16,\dagger}\) , Dieter Henrik Heiland \(^{1,2,3,\dagger}\) + +<|ref|>text<|/ref|><|det|>[[115, 297, 848, 314]]<|/det|> +\(^{1}\) Microenvironment and Immunology Research Laboratory, Medical Center, University of Freiburg, + +<|ref|>text<|/ref|><|det|>[[115, 319, 191, 333]]<|/det|> +Germany + +<|ref|>text<|/ref|><|det|>[[115, 339, 710, 355]]<|/det|> +\(^{2}\) Department of Neurosurgery, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 360, 495, 375]]<|/det|> +\(^{3}\) Faculty of Medicine, Freiburg University, Germany + +<|ref|>text<|/ref|><|det|>[[115, 380, 684, 395]]<|/det|> +\(^{4}\) Neuroelectronic Systems, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 400, 771, 416]]<|/det|> +\(^{5}\) Department of Medicine I, Medical Center - University of Freiburg, Faculty of Medicine, + +<|ref|>text<|/ref|><|det|>[[115, 421, 364, 436]]<|/det|> +University of Freiburg, Germany. + +<|ref|>text<|/ref|><|det|>[[115, 441, 841, 457]]<|/det|> +\(^{6}\) Institute of Medical Bioinformatics and Systems Medicine, Medical Center- University of Freiburg, + +<|ref|>text<|/ref|><|det|>[[115, 462, 315, 476]]<|/det|> +79110 Freiburg, Germany. + +<|ref|>text<|/ref|><|det|>[[115, 481, 881, 497]]<|/det|> +\(^{7}\) Comprehensive Cancer Center Freiburg (CCCF), Faculty of Medicine and Medical Center - University + +<|ref|>text<|/ref|><|det|>[[115, 502, 405, 517]]<|/det|> +of Freiburg, 79106 Freiburg, Germany. + +<|ref|>text<|/ref|><|det|>[[115, 522, 560, 537]]<|/det|> +\(^{8}\) German Cancer Consortium (DKTK), partner site Freiburg. + +<|ref|>text<|/ref|><|det|>[[115, 542, 831, 576]]<|/det|> +\(^{9}\) Department of Neurology and Brain Tumor Center, University Hospital Zurich and University of Zurich, Switzerland + +<|ref|>text<|/ref|><|det|>[[115, 582, 709, 597]]<|/det|> +\(^{10}\) Department of Neurosurgery, RWTH University of Aachen, Aachen, Germany + +<|ref|>text<|/ref|><|det|>[[115, 602, 830, 618]]<|/det|> +\(^{11}\) Neurosurgical Artificial Intelligence Laboratory Aachen (NAILA), Department of Neurosurgery, + +<|ref|>text<|/ref|><|det|>[[115, 623, 475, 638]]<|/det|> +RWTH University of Aachen, Aachen, Germany + +<|ref|>text<|/ref|><|det|>[[115, 643, 850, 659]]<|/det|> +\(^{12}\) Department of Anesthesiology and Critical Care Medicine, Medical Center- University of Freiburg, + +<|ref|>text<|/ref|><|det|>[[115, 664, 313, 679]]<|/det|> +79106 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 685, 633, 700]]<|/det|> +\(^{13}\) Institute of Neuropathology, Medical Center - University of Freiburg, + +<|ref|>text<|/ref|><|det|>[[115, 705, 737, 721]]<|/det|> +\(^{14}\) Signalling Research Centres BIOSS and CIBSS, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 726, 767, 741]]<|/det|> +\(^{15}\) Center for NeuroModulation (NeuroModul), University of Freiburg, Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 746, 848, 762]]<|/det|> +\(^{16}\) Translational NeuroOncology Research Group, Medical Center, University of Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[115, 767, 404, 783]]<|/det|> +# Equal contributed first authorship + +<|ref|>text<|/ref|><|det|>[[115, 788, 402, 803]]<|/det|> +Equal contributed last authorship + +<|ref|>text<|/ref|><|det|>[[115, 844, 881, 882]]<|/det|> +DISCLOSURE OF CONFLICTS OF INTEREST: No potential conflicts of interest were disclosed by the authors. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[67, 82, 321, 100]]<|/det|> +# Corresponding author: + +<|ref|>text<|/ref|><|det|>[[66, 105, 530, 303]]<|/det|> +Corresponding author:Dieter Henrik HeilandDepartment of NeurosurgeryMedical Center University of FreiburgBreisacher Straße 6479106 Freiburg-Germany-Tel: +49 (0) 761 270 50010Fax: +49 (0) 761 270 51020E- mail: dieter.henrik.heiland@uniklinik- freiburg.de + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 202, 100]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[112, 108, 885, 420]]<|/det|> +Despite recent advances in cancer immunotherapy, its efficacy in Glioblastoma (GBM) is limited due to poor understanding of molecular states and cellular plasticity of immune cells within the tumor microenvironment. Here, we combined spatial and single- cell transcriptomics of 47.284 immune cells, to map the potential cellular interactions leading to the immunosuppressive microenvironment and dysfunction of T cells. Computational approach identified a subset of IL10 releasing HMOX1+ myeloid cells which activates transcriptional programs towards a dysfunctional state in T cells, and was found to be localized within mesenchymal dominated subregions of the tumor. These findings were further validated by a human ex- vivo neocortical GBM model (n=6) coupled with patient derived peripheral T- cells. Finally, the dysfunctional transformation of T cells was shown to be rescued by JAK/STAT inhibition in both our model and in- vivo. We strongly believe that our findings would be the stepping stone towards successful development of immunotherapeutic approaches in GBM. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 108, 237, 125]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[110, 128, 885, 670]]<|/det|> +Tumor infiltrating lymphocytes, along with resident and migrated myeloid cells, account for a significant part of the tumor microenvironment in glioblastoma \(^{1 - 3}\) . Most recently, the characterization of the myeloid cell population using scRNA- sequencing revealed remarkable heterogeneity with regards to cellular diversity and plasticity within the myeloid compartment \(^{1,4}\) . However, the diversity of lymphoid cell types within malignant brain tumors remains unexplored and needs to be illuminated. Insights into the heterogeneity of cell type composition and driver genes for lineage differentiation within lymphoid compartment will aid in providing successful approaches for immunotherapy in the future. In other cancer entities such as colorectal cancer \(^{5}\) , liver cancer \(^{6}\) or melanoma \(^{7}\) , different T cell states have been investigated. Situations which involve prolonged immune activation and ambiguous stimulation, such as uncontrolled tumor growth or chronic infections, impede the ability of CD8 \(^{+}\) lymphocytes to secrete proinflammatory cytokines and maintain their cytotoxic profile \(^{7 - 9}\) . This cellular state, named dysfunctional or "exhausted" CD8 \(^{+}\) lymphocytes, represents a paramount barrier to successful immune- vaccination or checkpoint therapy \(^{2,10,11}\) . T cell exhaustion is partially orchestrated by regulation of inhibitory cell surface receptors (PD- 1, CTLA- 4, LAG- 3, TIM- 3 and others), in addition to anti- inflammatory cytokines such as IL- 10 and TGF- \(\beta\) . Glioblastoma, a common and very aggressive primary brain tumor in adults, is archetypical for tumors with a strong immunosuppressive microenvironment \(^{12}\) . Current, immunotherapeutic approaches such as PDL1/PD1 checkpoint blockade \(^{13}\) or peptide vaccination \(^{14}\) , led to remarkable responses in several cancers, has failed to demonstrate its effectivity in patients suffering from glioblastoma. + +<|ref|>text<|/ref|><|det|>[[111, 672, 885, 912]]<|/det|> +To address the sparse knowledge with respect to the lymphoid cell population in glioblastoma, we performed deep transcriptional profiling by means of scRNA- sequencing, and mapped potential cellular interactions and cytokine responses that could lead to the dysfunctional and exhausted phenotype of T cells. Pseudotime analysis revealed an increased response to Interleukin 10 (IL10) during the transformation of T cells from the effector state to the dysfunctional state. To computationally explore "connected" cells driving this transformation, we introduced a novel approach termed "nearest functionally connected neighbor (NFCN)", which identified a subset of myeloid cells marked by CD163 \(^{+}\) and HMOX1 \(^{+}\) expression. Furthermore, we performed spatially resolved transcriptomics, which confirmed the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 81, 884, 323]]<|/det|> +spatial overlap of exhausted T cells with HMOX1+ myeloid cells, within regions of the tumor enriched with mesenchymal transcriptional signatures. Finally, using human neocortical GBM model (coupled with patient derived T cells) and myeloid cell depletion, we conclusively validated that myeloid cells are key drivers for immunosuppressive microenvironment in the glioblastoma. The dysfunctional transformation of T cells were found to be rescued by the JAK- STAT inhibition due to the reduction in IL10 release, as shown before 15, in our ex- vivo model. Based on these results, we treated a single patient having recurrent glioblastoma in a neoadjuvant setting with JAK- STAT inhibitor (Ruxolitinib) and partially rescued the immunosuppressive environment. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 199, 99]]<|/det|> +## Results: + +<|ref|>sub_title<|/ref|><|det|>[[115, 106, 787, 126]]<|/det|> +## Single cell Analysis of the Immune Cell Compartment in Glioblastoma + +<|ref|>text<|/ref|><|det|>[[110, 123, 885, 770]]<|/det|> +In order to interrogate the diversity of the immune microenvironment in glioblastoma, we performed droplet based 10X single cell sequencing of tissue samples from 8 patients, diagnosed with Glioblastoma. Lymphoid and myeloid populations (CD45+/CD3+) were sorted from neoplastic tissue specimens (Figure 1a and Supplementary Figure 1a). The scRNA- seq data consisted of 47,284 cells, with a median number of 2,301 unique molecular identifiers (UMIs) and approximately 1023 uniquely expressed genes per cell. We corrected the data for mitochondrial genes, regressed out cell cycle effects and removed batch effects due to technical artifacts. We then decomposed the eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparing them to randomized expression values, resulting in 41 meaningful components. Shared nearest neighbor (SNN) graph clustering resulted in 21 clusters (C0- C20) containing uniquely expressed genes. The major observed cell type when using the semi- supervised subtyping algorithm of scRNA- seq (SCINA- Model) \(^{16}\) and SingleR \(^{17}\) are: microglia cells (TMEM119, CX3CR1 and P2RY12) and macrophages (AIF1, CD68, CD163 and low expression of TMEM119, CX3CR1), followed by CD8+ T cells (CD8A, CD3D), natural killer cells (KLRD1, GZMH, GZMA, NKG7 and CD52), CD4+ T cells (BCL6, CD3D, CD4, CD84 and IL6R), T- memory cells (TRBC2, LCK, L7R and SELL), granulocytes (LYZ), a minor number of oligodendrocytes and oligodendrocyte- progenitor cell (OPC's) (OLIG1, MBP, PDGFA), and endothelial cells (CD34, PCAM1, VEGFA) Figure 1b, Supplementary Figure 1b- f. To identify malignant cells, we inferred large scale copy number variations (CNVs) from scRNA- seq profiles by averaging expression over stretches of 100 genes on their respective chromosomes \(^{18}\) . With this approach, we confirmed that there was minimal contamination by tumor cells (clustered as OPC cells), based on their typical chromosomal alterations (gain in chromosome 7 and loss in chromosome 10), Supplementary Figure 2a. + +<|ref|>sub_title<|/ref|><|det|>[[115, 795, 671, 814]]<|/det|> +## Diversity of T cells in the Glioblastoma Microenvironment + +<|ref|>text<|/ref|><|det|>[[115, 820, 884, 913]]<|/det|> +To investigate the diversity of the T cells present in the microenvironment, we examined them by two different but complementary methods. Firstly, T cells were isolated in- silico by means of clustering (as shown above), based on previously published marker gene expression profiles (CD3+, CD4+/CD8+). Secondly, they were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 885, 575]]<|/det|> +isolated using the SCINA model, which resulted in a total of 7,547 cells Supplementary Figure 1g. Focusing on the different regulatory states of these cells, we identified 13 subclusters using SNN- clustering which were then re- embedded into a dynamic model using RNA- velocity, closely reflecting different activation states, Supplementary Figure 1h and Figure 1c. Reconstruction of lineage differentiation trajectories by means of both pseudotime and latent time provided insights into the transformation of the cells over time19, Figure 1c. Based on common marker signatures, we defined activated T cells by expressing GZMA, CCL5 and CCL4, IL7RI and other markers (Supplementary Figure 1i and 2b) as well as increased proliferation (G2M- score, Figure 1d) (C1,C10), and naive T cells by the expression of SELL, TCF7 and CCR7 (C0), Figure 1c and Supplementary Figure 1h. We further identified T cell subgroups marked by hypoxia and heat- shock signaling (HSPA8, C2, C5,C6) and by a subgroup which showed mixed expression of activation/dysfunctional/exhaustion markers (HAVCR2, GNLY) and strong enrichment IL- 10 signaling, Figure 1d. We estimated the G2M- score and identified a lower frequency of cell cycle in the differentiated/later states of T cells with the exception of the IFN- gamma subtype, Figure 1d. Additional marker plots are given in the Supplementary Figure 2b. Regulatory CD4+ T cells (FOXP3, IL2RA and CTLA4) represents a minor population in the glioblastoma microenvironment, Figure 1c and Supplementary Figure 1i-j. + +<|ref|>sub_title<|/ref|><|det|>[[115, 574, 670, 594]]<|/det|> +## Dysfunctional State of T cells is Driven by IL-10 Signaling + +<|ref|>text<|/ref|><|det|>[[111, 599, 885, 888]]<|/det|> +To gain insights into the regulatory mechanism of immune cells, we reconstructed fate decisions made during T cell exhaustion using pseudotime trajectories along the estimated velocity streams, Figure 1e. Using RNA- velocity, we estimated cells with high probability of initial and terminal states which resulted in 3 major branches with unique cell fate drivers, Figure 1e. In order to determine the dynamic adaptation across all branches (Velo trajectory 1- 3), we computed pseudotime vectors using various models. Only two trajectories confirmed the predicted ascending pseudotime inference across multiple models (HAVCR2(+)- T cells and hypoxia- induced T cells), Figure 1f. We assume that terminal states without significant pseudotime connection do not inevitably arise from the determined initial state, which suggests that the lineage of IFN- gamma driven T cells coexists and most likely originated from an earlier lineage branch Figure 1g. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 471]]<|/det|> +Along our trajectory from effector to HAVCR2(+)T cells, we showed a latent time- dependent increase of the response to anti- inflammatory signaling such as IL- 10 and TGF- beta signaling Figure 1h. Next, we mapped the gene expression of defined exhausted and effector signatures along our differentiation trajectory, revealing an enrichment of exhausted genes within the destination cluster. Thus, our data suggests that this response to IL- 10 contributes to the dysfunctional state of T cells and affects fate decisions Figure 1i. To gain further insights into accurate downstream signaling of IL- 10, IFN- gamma, and IL2, we created a library of the 50 most highly up- and downregulated genes, Supplementary Figure 3a- b. We then extracted signatures observed within the different T cell clusters and compared them with stimulated T cells. As expected, genes upregulated by IL2 stimulation were significantly enriched in cluster 1, while signature genes from IFN- gamma stimulation was enriched in cluster 12. Signature genes from IL- 10 stimulation showed a significant enrichment in the dysfunctional cluster (Cluster 9), Supplementary Figure 3c. Furthermore, we mapped our signatures along our defined velocity trajectory 1 which confirmed our predicted pathway inference, Figure 1j- k and Supplementary Figure 3d. + +<|ref|>sub_title<|/ref|><|det|>[[115, 500, 881, 543]]<|/det|> +## T cell Activation and Exhaustion Reveals Spatial Heterogeneity and Association with Glioblastoma Subtypes + +<|ref|>text<|/ref|><|det|>[[111, 549, 884, 912]]<|/det|> +Glioblastomas present a high degree of heterogeneity due to regional metabolic differences and varying composition of the tumor microenvironment20. To decipher the spatial distribution of above illustrated T cell clusters and its colocalization to defined tumor states, we performed spatial transcriptomic RNA sequencing (stRNA- seq) of 3 primary IDH1/2 wildtype glioblastoma, containing a total number of 2,352 spots, Figure 2a. We observed a median of 8 cells per spot (range: 4 to 22 cells per spot), which allows the spatial mapping of gene expression, but not at single cell resolution. However, when we compared our dataset to the latest classification of glioblastoma20, consistent results were obtained in accordance with the diversity of subtype expression, Figure 2b. In particular, Neftel and colleagues raised evidence that the mesenchymal gene expression is more likely associated with immune response and cellular interaction to myeloid cells20. In order to investigate the spatial distribution of the mesenchymal subgroup, we performed gene set enrichment analysis at spatial resolution which revealed spot- wise enrichments within all samples, Figure 2c- d. In a next step, we used a seeded non- negative matrix factorization (NMF) regression21 to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 884, 250]]<|/det|> +estimate the probability of individual T cell clusters at spatial resolution, Figure 2e. We computed the spatial overlap of T cell clusters and glioblastoma states, using a Bayesian approach, which revealed a high estimated correlation of the mesenchymal and T cell cluster 3 (T HAVCR2), cluster 5- 7 (T Hypoxia), Figure 2f. Further analysis confirmed the strong overlap of dysfunctional/exhausted marker (LAG3 and HAVCR2) exclusively with the mesenchymal gene expression signature, Figure 2g, suggesting that our data support the findings from Neftel and colleagues. + +<|ref|>sub_title<|/ref|><|det|>[[115, 274, 727, 294]]<|/det|> +## A Subset of Microglia and Macrophages Drive IL-10 Stimulation + +<|ref|>text<|/ref|><|det|>[[111, 295, 884, 938]]<|/det|> +In our recent investigation \(^{15}\) , the crosstalk between microglia cells and reactive astrocytes in the tumor microenvironment was found to be responsible for upregulating IL10 release. This is mediated by microglia/macrophages stimulated with IFN gamma, leading to JAK/STAT activation in tumor- associated astrocytes. In this study we introduce the "nearest functionally connected neighbor" algorithm (NFCN), an in- silico model to identify the most likely related cell pairs through divergent down- and upstream signal activity, Figure 3a. With our model, we assume that cellular interaction with distinct mutual activation implies two fundamental prerequisites. On the one hand, the ligand needs to be expressed and released, or otherwise exposed on the cell surface. To avoid the chances of randomly elevated expression or technical artifacts, we also looked at the simultaneous occurrence of ligand induction (upstream pathway signaling). On the other hand, the receptor needs to be expressed and, additionally, downstream signaling has to be activated as well. This allows us to predict the functional status of the receiver cell (Explanation of the model can be found in the Methods section, with an overview in Supplementary Figure 4). We used our in- silico model to screen for potential cells responsible for IL- 10 activation of T cells. The algorithm identified pairs of lymphoid (T cell clusters) and myeloid cells (macrophages and microglia cluster) and estimated the likelihood of mutual activation Figure 3b,c. By extraction of the nearest connected cells (top 1% ranked cells), we identified a subset of myeloid cells characterized by remarkably high IL10 expression. Most of the receiver cells in the connected cells (top 1% ranked cells) originated from the T cell cluster Figure 3d. Our predicted IL10 interactions were also supported by another computational approach (Supplementary Figure 5). In order to validate our computational model, we used SPOTlight \(^{21}\) , an algorithm to predict the spatial position of cells from scRNA-seq data, and were able to confirm a significant overlap. In order to explore the difference between connected and non- connected cells, we extracted + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 81, 884, 175]]<|/det|> +both connected and non- connected cells, defined by the highest and lowest interaction- scores (quantile \(97.5\%\) ) Using differential gene expression analysis, we observed multiple genes which confirmed the non- inflammatory polarization status of highly connected cells. + +<|ref|>text<|/ref|><|det|>[[111, 180, 884, 644]]<|/det|> +These findings are not surprising, since one of the essential markers of non- inflammatory myeloid cells is IL10, Figure 3e. We showed that the subset of most highly connected cells were marked by CD163 and heme oxygenase 1 (HMOX1) expression. In a multilayer representation, we illustrated the estimated cellular connections with respect to their most likely spatial coordinates, Figure 3f. HMOX1 is activated during inflammation and oxidative injuries and is regulated through the Nrf2/Bach1- axis, as well as through the IL10/HMOX1- axis. This gene is also well known to be upregulated in the alternative activated macrophage subtype22. Another immunosuppressive signaling marker closely related to the alternative activation of macrophages/microglia is the release of TGF- \(\beta^{23}\) , which was also found to be upregulated in highly connected cells, Figure 3g. Consistent with our findings, most downstream signals of the IL10/HMOX1- axis such as STAT3 and p38 MAPK were found to be upregulated in a gene set enrichment analysis, Figure 3d,h. In a recent investigation, the role of doublets in the detection of cell- cell interactions was described24. In our dataset the score of potential cell- cell connections was similar to HMOX1 marked macrophages, within the doublet dominated cluster. We computed the marker genes of our predicted connected myeloid and lymphoid genes and found each of the marker sets to be highly expressed in the doublet cluster, which further validated the results from our computational model, Figure 3i. + +<|ref|>sub_title<|/ref|><|det|>[[115, 667, 621, 687]]<|/det|> +## Loss of Myeloid Cells Increases Antitumor Immunity + +<|ref|>text<|/ref|><|det|>[[113, 691, 884, 933]]<|/det|> +To validate our computational findings, we made use of the recently described human neocortical GBM model, where the cellular architecture of the CNS is well preserved15,25. We cultured non- infiltrated neocortical slices (defined in a recent report15,25) coupled with autografted T cells along with myeloid cell depletion, to understand the communication between myeloid cells in the tumor microenvironment along with lymphoid cells. Three days after chemical depletion of the myeloid cells, we injected a primary cell line (BTSC#233, GFP- tagged, previously characterized by RNA- seq profiling as mesenchymal)26. After 4 days of culture, peripheral T cells (same donors), tagged using CellTrace™ Far Red (CTFR) were additionally inoculated and the sections were further cultured for another 48h, Figure 4a. Immunostainings showed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 885, 644]]<|/det|> +that myeloid cell depletion reduces the number of \(\mathrm{IBA1^{+}HMOX1^{+}}\) cells, Figure 4b. Using an enzyme- linked immunosorbent assay (ELISA) we found a significant reduction in IL10 when the myeloid cells were depleted, regardless of the presence of tumor cells. The strongest difference in IL10 release was observed in myeloid cell depleted sections in the presence of tumor cells, Figure 4c. Furthermore, we stained for Granzyme B (GZMB+) T cells and quantified IL2 release to examine the amount of effector T cells in both the depleted and non- depleted sections. We found an increased number of GZMB+ T cells in sections with myeloid cell depletion, Figure 4d, along with a significant increase in IL2 and no differences in IFN gamma release were observed, Figure 4e- f. We also stained for the exhaustion marker TIM3 (Gen: HAVCR2), which was found to be enriched in T cells in the presence of myeloid cell Figure 4g, suggesting that myeloid derived IL10 release (by HMOX1+ cells) leads to T cell exhaustion, which is in agreement with the computational model. In order to prove that the IL10 signaling is responsible for the induction of the expression of exhausted T cell markers, we preincubated T cells using an IL10 neutralizing inhibitory antibody Figure 4h. This resulted in a strong increase of GZMB+ T cells, followed by a significant increase of IL2, suggesting that IL10 inhibition drives T cell activity, Figure 4i- j. From our recent study, we found that a JAK/STAT inhibition causing a significant reduction of IL10 within the tumor microenvironment15. Based on this investigation, we pretreated tissue sections with Ruxolitinib, an FDA- approved JAK- inhibitor, before inoculating the sections with patient- derived T cells. We were able to confirm that JAK inhibition caused a significant decrease of IL10 and increased levels of the inflammatory marker IL2, Figure 4k- l. + +<|ref|>text<|/ref|><|det|>[[111, 648, 885, 914]]<|/det|> +Based on our above findings, we treated a first patient with a recurrent glioblastoma in a neoadjuvant setting with Ruxolitinib for 4 weeks. After resection, we sorted CD45+ cells and performed scRNA- sequencing Figure 4m. Immunostainings of the tumor revealed a relative increase of \(\mathrm{CD8^{+}}\) and \(\mathrm{CD4^{+}}\) T cells whereas \(\mathrm{CD68^{+}}\) myeloid cells remained stable, Figure 4n- o. scRNA sequencing revealed a large number of T cells, most of which express markers for T cell activation and a few cells showing a naive signature, Figure 4p. When compared with T cells from our initial dataset, an enrichment of the T cells from the JAK treated patient was seen within cluster 1 (activated T cells) and cluster 12 (B cells), suggesting that JAK- inhibition could be a potential treatment option to boost T cell activation by reducing immunosuppressive programs in both myeloid and glial cells, Figure 4q- r. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 103, 234, 120]]<|/det|> +## Discussion: + +<|ref|>text<|/ref|><|det|>[[115, 127, 884, 270]]<|/det|> +Although single- cell RNA- sequencing accurately maps the cellular architecture and reflects the diversity of cellular states \(^{18,20,27,28}\) , there is a lack of spatial information. Here, we combine single cell RNA sequencing of the immune compartment along with spatial transcriptomic RNA- sequencing (stRNA- seq) to gain better insights into the complex crosstalk, cellular states and cellular plasticity leading to the immunosuppressive environment found in glioblastoma (GBM). + +<|ref|>text<|/ref|><|det|>[[110, 270, 885, 870]]<|/det|> +Recent studies have reported different subtypes of microglia and macrophages occupying glial tumors \(^{1,4,20,27,28}\) . However, detailed information about lymphoid infiltration cells is lacking. There is intense interest in T cells and their varied states due to their importance in the development of targeted therapies and to further the understanding of the immunosuppressive environment of glioblastoma. T cell states, particularly in disease, are somehow difficult to accurately classify, leading to numerous definitions and markers in recent years \(^{2,7,29 - 31}\) . Some authors use the terms "dysfunctional" and "exhausted" synonymously \(^{32}\) , whereas others differentiate between the dysfunctional and the exhausted states of T cells \(^{29,31}\) . In this study we use the definition of cellular states proposed by Singer et al., 2016 \(^{8}\) . On the basis of these gene sets, our data showed that only cells which remained activated along the pseudotime- trajectory were able to enter a state of dysfunction, and later exhaustion. The dysfunction appears to be a transient state, associated with increased proliferation, despite immunosuppressive stimulation from the tumor environment. This imbalance between pro- and anti- inflammatory signaling, dominated by IL10 secretion, leads to final exhaustion of the T cells, which is in agreement with the current literature \(^{2,33}\) . In order to find a consensus with regard to marker genes we further validated our findings on a set of exhausted marker genes recently published in an overview study \(^{34}\) . We and others have shown that the GBM microenvironment aids in the evolution of immune suppression. In this process, astrocytes and myeloid cells, both driven by STAT- 3 signaling, orchestrate the immunosuppressive environment \(^{4,15,35,36}\) . Based on the knowledge that IL10 interaction plays a crucial role in the shift from activated to exhausted T cells, we built an in- silico model that identified potential connected cells driving T cell exhaustion. + +<|ref|>text<|/ref|><|det|>[[115, 864, 884, 932]]<|/det|> +Using this model, we identified a subset of myeloid cells marked by high expression of HMOX1 \(^{+}\) , a gene which is induced by oxidative stress and metabolic imbalance \(^{37,38}\) . HMOX1 is linked to the STAT- 3 pathway and induces IL10 production via MAPK + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 884, 748]]<|/det|> +activation, and all of these markers were also found to be upregulated in our connected cells, as reported above. Furthermore, we used spatial transcriptomics to confirm the spatial overlap of cells that we identified are highly connected. We were able to show that the HMOX1+- myeloid cells were spatially correlated with exhaustion and the mesenchymal state of glioblastoma. These findings are in accord with published reports, revealing that the mesenchymal cells are the component of GBM responsible for the immune crosstalk20. HMOX1 expression in GBM and IDH-WT astrocytoma was found to be increased in recurrent GBM and negatively associated with overall survival, Supplementary Figure 6a,b. In addition, we made use of a human neocortical GBM model coupled with patient derived T cells with/without myeloid cell depletion. This model helped us to simulate the function of the myeloid cells with regard to IL10 release and T cell stimulation. Fitting with our computational model, we confirmed that HMOX1+ myeloid cells cause a reduction of effector T cells, with a respective reduction in IL2 release and increased expression of our identified exhaustion marker TIM3. Following our recent investigations in which we demonstrated that JAK-inhibition is able to reduce the level of IL10 in human brain tumors15, we demonstrated that in a single patient that inhibition of the JAK- STAT axis was able to partially rescue the immunosuppressive environment. The treated subject is still alive but showed permanent disruption of the blood-brain barrier with repetitive increase of contrast enhancing lesions. Each radiologically confirmed progress was sampled without evidence of a tumor recurrence, suggesting that manipulation of the glia/myeloid environment simultaneously caused exaggerated inflammation and pseudo progression. Our single-cell RNA-seq confirmed a pronounced enrichment of activated T cells, while the number of myeloid cells remained relatively stable. In this sample, we also detected the strongest contamination with glial cells. We assumed that potential doublets or strong cell-cell connections led to glial cells being detected as CD45+, resulting in a false positive sorting. + +<|ref|>text<|/ref|><|det|>[[112, 744, 884, 912]]<|/det|> +In conclusion, this work provides the first knowledge regarding lymphocyte population in the glioblastoma microenvironment where we showed that the functional interaction between myeloid and lymphoid cells, leads to a dysfunctional state of T cells. Using human neocortical GBM model and single patient subject we showed that the IL- 10 driven T cell exhaustion can be rescued by JAK/STAT inhibition. Thus, the results from this work can be the steppingstone towards successful immunotherapeutic approaches for GBM. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 208, 100]]<|/det|> +## Methods: + +<|ref|>sub_title<|/ref|><|det|>[[115, 108, 279, 126]]<|/det|> +## Ethical Approval + +<|ref|>text<|/ref|><|det|>[[115, 131, 883, 323]]<|/det|> +The local ethics committee of the University of Freiburg approved the data evaluation, imaging procedures and experimental design (protocol 100020/09 and 472/15_160880). The methods were carried out in accordance with the approved guidelines, with written informed consent obtained from all subjects. The studies were approved by an institutional review board. Further information and requests for resources, raw data and reagents should be directed and will be fulfilled by the Contact: D. H. Heiland, dieter.henrik.heiland@uniklinik- freiburg.de. A complete table of all materials used is given in the supplementary information. + +<|ref|>sub_title<|/ref|><|det|>[[115, 353, 413, 370]]<|/det|> +## T cell isolation and stimulation + +<|ref|>text<|/ref|><|det|>[[113, 376, 884, 641]]<|/det|> +Blood was drawn from a healthy human individual into an EDTA (ethylenediaminetetraacetic acid) cannula. T cells were extracted in a negative selection manner using a MACSxpress® Whole Blood Pan T Cell Isolation Kit (Miltenyi Biotech). T cells were then transferred in Advanced RPMI 1640 Medium (ThermoFisher Scientific, Pinneberg, Germany) and split for cytokine treatment: Three technical replicates were used for each T cell- treatment condition. Interleukin 2 (IL- 2, Abcam, Cambridge, UK) was used at a final concentration of 1 ng/ml, Interleukin 10 (IL- 10, Abcam) at 5 ng/ml, Interferon gamma (IFN- \(\gamma\) , Abcam) at 1 ng/ml and Osteopontin (SPP- 1, Abcam) at 3 \(\mu \mathrm{g} / \mathrm{ml}\) . Cytokine treatment was performed in Advanced RPMI 1640 Medium and T cells were incubated at \(37^{\circ}\mathrm{C}\) and \(5\%\) CO2 for 24h. + +<|ref|>sub_title<|/ref|><|det|>[[115, 672, 483, 691]]<|/det|> +## RNA sequencing of stimulated T Cells + +<|ref|>text<|/ref|><|det|>[[115, 696, 884, 912]]<|/det|> +The purification of mRNA from total RNA samples was achieved using the Dynabeads mRNA Purification Kit (Thermo Fisher Scientific, Carlsbad, USA). The subsequent reverse transcription reaction was performed using SuperScript IV reverse transcriptase (Thermo Fisher Scientific, Carlsbad, USA). For preparation of RNA sequencing, the Low Input by PCR Barcoding Kit and the cDNA- PCR Sequencing Kit (Oxford Nanopore Technologies, Oxford, United Kingdom) were used as recommended by the manufacturer. RNA sequencing was performed using the MinION Sequencing Device, the SpotON Flow Cell and MinKNOW software (Oxford Nanopore Technologies, Oxford, United Kingdom) according to the manufacturer's + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 883, 150]]<|/det|> +instructions. Samples were sequenced for 48h on two flow-cells. Basecalling was performed by Albacore implemented in the nanopore software. Only \(\mathsf{D}^2\) - Reads with a quality Score above 8 were used for further alignment. + +<|ref|>sub_title<|/ref|><|det|>[[115, 180, 453, 199]]<|/det|> +## Sequence trimming and Alignment + +<|ref|>text<|/ref|><|det|>[[115, 204, 884, 420]]<|/det|> +In the framework of this study, we developed an automated pipeline for nanopore cDNA- seq data, which is available at github (https://github.com/heilandd/NanoPoreSeq). First the pipeline set up a new class termed "Poreseq" by a distinct sample description file. The analysis starts by rearranging the reads from the fastq output from the nanopore sequencer containing all of the \(\mathsf{D}^2\) - Reads. All fastq files need to be combined into one file. Multiplexed samples were separated according to their barcode and trimmed by Porechop (https://github.com/rrwick/Porechop). Alignment was performed with minimap2 (https://github.com/lh3-/minimap2) and processed with sam-tools. + +<|ref|>sub_title<|/ref|><|det|>[[115, 450, 454, 469]]<|/det|> +## Posthoc Analysis of Bulk-RNA-seq + +<|ref|>text<|/ref|><|det|>[[115, 475, 884, 764]]<|/det|> +A matrix of counted genes was further prepared by the RawToVis.R (github.com/heilandd/VRSD_Lab_v1.5) script, containing normalization of Mapped reads by DESeq, batch effect removal (ComBat package) and fitting for differential gene expression. Gene set enrichment analysis was performed by transformation of the log2 fold- change of DE into a ranked z- scored matrix, which was used as the input. The expression matrix was analysed with AutoPipe (https://github.com/heilandd/AutoPipe) by a supervised machine- learning algorithm and visualized with a heatmap. Full analysis was visualized with the Visualization of RNA- Seq Data (VRSD_Lab software, github.com/heilandd/VRSD_Lab_v1.5) as a dashboard app based on shiny R- software. We extracted the 50 top up/down regulated genes respectively of each stimulation with respect to control condition to construct a stimulation library. + +<|ref|>sub_title<|/ref|><|det|>[[115, 795, 567, 814]]<|/det|> +## Single-Cell Suspension for scRNA-sequencing + +<|ref|>text<|/ref|><|det|>[[115, 819, 883, 911]]<|/det|> +Tumor tissue was obtained from glioma surgery immediately after resection and was transported in phosphate- buffered saline (PBS) within approximately 5 minutes into our cell culture laboratory. Tumor tissue was processed under a laminar flow cabinet. Tissue was reduced to small pieces using two scalpels and the tissue was processed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 884, 397]]<|/det|> +with the Neural Tissue Dissociation Kit (T) using C- Tubes (Miltenyi Biotech, Bergisch- Gladbach, Germany) according to the manufacturer's instructions. The Debris Removal Kit from Miltenyi was used according to the manufacturer's instructions to remove remaining myelin and extracellular debris. In order to remove the remaining erythrocytes, we resuspended the pellet in \(3.5 \text{ml ACK lysis buffer (ThermoFisher Scientific, Pinneberg, Germany)}\) and incubated the suspension for 5 minutes followed by a centrifugation step (350g, 10 min, RT). Cell quantification with a hematocytometer was performed after discarding the supernatant and resuspending the pellet in PBS. Cell suspensions were centrifuged again (350g, 10 min, RT) and resuspended in freezing medium containing \(10\%\) DMSO (Sigma- Aldrich, Schnelldorf, Germany) in FCS (PAN- Biotech, Aidenbach, Germany). Cell suspensions were immediately placed in a freezing box containing isopropanol and stored in a \(- 80^{\circ}\text{C}\) freezer for not more than 4 weeks. + +<|ref|>sub_title<|/ref|><|det|>[[116, 426, 419, 445]]<|/det|> +## Cell sorting by Magnetic Beads + +<|ref|>text<|/ref|><|det|>[[113, 450, 884, 617]]<|/det|> +Four frozen single- cell suspensions, originating from one patient with an IDH- mutated glioma and three patients with an IDH- wildtype glioblastoma (GBM), were thawed and the dead cells magnetically labeled and eliminated using a Dead Cell Removal Kit (Miltenyi Biotech). The tumor immune environment in general and T cells in particular were positively selected by using \(\text{CD3 + - MACS}\) (Miltenyi Biotech). Cells were stained with trypan blue, counted using a hematocytometer and prepared at a concentration of 700 cells/μL. + +<|ref|>sub_title<|/ref|><|det|>[[116, 648, 380, 666]]<|/det|> +## Droplet scRNA-sequencing + +<|ref|>text<|/ref|><|det|>[[113, 671, 884, 862]]<|/det|> +At least 16000 cells per sample were loaded on the Chromium Controller (10x Genomics, Pleasanton, CA, USA) for one reaction of the Chromium Next GEM Single Cell \(3' \text{v} 3.1\) protocol (10x Genomics), based on a droplet scRNA- sequencing approach. Library construction and sample indexing was performed according to the manufacturer's instructions. scRNA- libraries were sequenced on a NextSeq 500/550 High Output Flow Cell v2.5 (150 Cycles) on an Illumina NextSeq 550 (Illumina, San Diego, CA, USA). The bcl2fastq function and the cell ranger (v3.0) was used for quality control. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 461, 101]]<|/det|> +## Postprocessing scRNA-sequencing + +<|ref|>text<|/ref|><|det|>[[113, 103, 884, 694]]<|/det|> +We used cell ranger to detect low- quality read pairs of single- cell RNA sequencing (scRNA- seq) data. We filtered out reads which did not reach the following criteria: (1) bases with quality \(< 10\) , (2) no homopolymers (3) 'N' bases accounting for \(\geq 10\%\) of the read length. Filtered reads were mapped by STAR aligner and the resulting filtered count matrix further processed by Seurat v3.0 (R- package). We normalized gene expression values by dividing each estimated cell by the total number of transcripts and multiplied by 10,000, followed by natural- log transformation. Next, we removed batch effects and scaled data by a regression model including sample batch and percentage of ribosomal and mitochondrial gene expression. For further analysis we used the 2000 most variable expressed genes and decomposed eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparison to randomized expression values. The obtained non- trivial components were used for SNN clustering followed by dimensional reduction using the UMAP algorithm. Differently expressed genes (DE) of each cluster were obtained using a hurdle model tailored to scRNA- seq data which is part of the MAST package. Cell types were identified by 3 different methods; Classical expression of signature markers of immune cells; SingleR an automated annotation tool for single- cell RNA sequencing data obtaining signatures from the Human Primary Cell Atlas, SCINA, a semi- supervised cell type identification tool using cell- type signatures as well as a Gene- Set Variation Analysis (GSVA). Results were combined and clusters were assigned to the cell type with the highest enrichment within all models. In order to individually analyze T cells, we used the assigned cluster and filter for the following criteria. For further analysis T cells were defined by: \(\mathrm{CD3^{+}CD8^{+}}\) / \(\mathrm{CD4^{+}CD14^{+}LYZ^{- }}\) GFAP \(\cdot\) CD163 \(\cdot\) IBA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 721, 348, 739]]<|/det|> +## Spatial Transcriptomics + +<|ref|>text<|/ref|><|det|>[[115, 744, 884, 862]]<|/det|> +The spatial transcriptomics experiments were done using the 10X Spatial transcriptomics kit (https://spatialtranscriptomics.com/). All the instructions for Tissue Optimization and Library preparation were followed according to the manufacturer's protocol. Here, we briefly describe the methods followed using the library preparation protocol. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 525, 101]]<|/det|> +## Tissue collection and RNA quality control: + +<|ref|>text<|/ref|><|det|>[[113, 107, 884, 352]]<|/det|> +Tissue samples from three patients, diagnosed with WHO IV glioblastoma multiforme (GBM), were included in this study. Fresh tissue collected immediately post resection was quickly embedded in optimal cutting temperature compound (OCT, Sakura) and snap frozen in liquid \(\mathbb{N}_2\) . The embedded tissue was stored at \(- 80^{\circ}C\) until further processing. A total of 10 sections (10μm each) per sample were lysed using TriZOI (Invitrogen, 15596026) and used to determine RNA integrity. Total RNA was extracted using PicoPure RNA Isolation Kit (Thermo Fisher, KIT0204) according to the manufacturer's protocol. RIN values were determined using a 2100 Bioanalyzer (RNA 6000 Pico Kit, Agilent) according to the manufacturer's protocol. It is recommended to only use samples with an RNA integrity value \(>7\) . + +<|ref|>sub_title<|/ref|><|det|>[[116, 380, 396, 399]]<|/det|> +## Tissue staining and Imaging: + +<|ref|>text<|/ref|><|det|>[[113, 404, 884, 740]]<|/det|> +Sections were mounted onto spatially barcoded glass slides with poly- T reverse transcription primers, with one section per array. These slides can be stored at \(- 80^{\circ}C\) until use. The slides were then warmed to \(37^{\circ}C\) , after which the sections were fixed for 10 minutes using \(4\%\) formaldehyde solution (Carl Roth, P087.1), which was then washed off using PBS. The fixed sections were covered with propan- 2- ol (VWR, 20842312). Following evaporation for 40 seconds, sections were incubated in Mayer's Hematoxylin (VWR, 1092490500) for 7 min, bluing buffer (Dako, CS70230- 2) for 90 seconds and finally in Eosin Y (Sigma, E4382) for 1 min. The glass slides were then washed using RNase/DNase free water and incubated at \(37^{\circ}C\) for 5 min or until dry. Before imaging, the glass slides were mounted with \(87\%\) glycerol (AppliChem, A3739) and covered with coverslips (R. Langenbrick, 01- 2450/1). Brightfield imaging was performed at 10x magnification with a Zeiss Axio Imager 2 Microscope, and postprocessing was performed using ImageJ software. + +<|ref|>text<|/ref|><|det|>[[115, 728, 882, 799]]<|/det|> +The coverslips and glycerol were removed by washing the glass slides in RNase/DNase free water until the coverslips came off, after which the slides were washed using \(80\%\) ethanol to remove any remaining glycerol. + +<|ref|>sub_title<|/ref|><|det|>[[115, 827, 640, 847]]<|/det|> +## Permeabilization, cDNA synthesis and tissue removal: + +<|ref|>text<|/ref|><|det|>[[115, 852, 882, 923]]<|/det|> +For each capture array, \(70\mu \mathrm{L}\) of pre- permeabilization buffer, containing \(50\mathrm{U / \mu L}\) Collagenase along with \(0.1\%\) Pepsin in HCl was added, followed by an incubation for 20 minutes at \(37^{\circ}C\) . Each array well was then carefully washed using \(100\mu \mathrm{L}0.1\times \mathrm{SSC}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 81, 885, 270]]<|/det|> +buffer. \(70\mu \mathrm{L}\) of Pepsin was then added and incubated for 11 minutes at \(37^{\circ}C\) . Each well was washed as previously described and \(75\mu \mathrm{L}\) of cDNA synthesis master mix containing: \(96\mu \mathrm{L}\) of 5X First strand buffer, \(24\mu \mathrm{L}\) 0.1M DTT, \(255.2\mu \mathrm{L}\) of DNase/RNase free water, \(4.8\mu \mathrm{L}\) Actinomycin, \(4.5\mu \mathrm{L}\) of \(20\mathrm{mg / mL}\) BSA, \(24\mu \mathrm{L}\) of \(10\mathrm{mM}\) dNTP, \(48\mu \mathrm{L}\) of Superscript® and \(24\mu \mathrm{L}\) of RNaseOUT™ was added to each well and incubated for 20 hours at \(42^{\circ}C\) without shaking. Cyanine 3- dCTP was used to aid in the determination of the footprint of the tissue section used. + +<|ref|>text<|/ref|><|det|>[[110, 264, 885, 514]]<|/det|> +Since glioblastoma tissue is a fatty tissue, degradation and tissue removal was carried out using Proteinase K treatment for which \(420\mu \mathrm{L}\) Proteinase K and PKD buffer (1:7), were added to each well and then incubated at \(56^{\circ}C\) for 1hr with intermittent agitation (15 seconds / 3 minutes). After incubation, the glass slides were washed three times with \(100\mathrm{mL}\) of \(50^{\circ}C\) SSC/SDS buffer with agitation for 10 minutes, 1 minute and finally for 1 minute at 300 rpm. The glass slides were then air- dried at room temperature. Tissue cleavage was carried out by the addition of \(70\mu \mathrm{L}\) of cleavage buffer ( \(320\mu \mathrm{L}\) RNase/DNase free water, \(104\mu \mathrm{L}\) Second strand buffer, \(4.2\mu \mathrm{L}\) of \(10\mathrm{mM}\) dNTP, \(4.8\mu \mathrm{L}\) of \(20\mathrm{mg / mL}\) BSA and \(48\mu \mathrm{L}\) of USER™ Enzyme) to each well and incubation at \(37^{\circ}C\) for 2 hours with intermittent agitation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 544, 303, 562]]<|/det|> +## Spot Hybridization: + +<|ref|>text<|/ref|><|det|>[[113, 568, 885, 767]]<|/det|> +In order to determine the exact location and quality of each of the 1007 spots, fluorescent Cyanine- 3 A is hybridized to the 5' ends of the surface probes. \(75\mu \mathrm{L}\) of the hybridization solution ( \(20\mu \mathrm{L}\) of \(10\mu \mathrm{M}\) Cyanine- 3A probe and \(20\mu \mathrm{L}\) of \(10\mu \mathrm{M}\) Cyanine- 4 Frame probe in \(960\mu \mathrm{L}\) of 1X PBS) was added to each well and incubated for 10min at room temperature. The slides were then washed three times with \(100\mathrm{ml}\) of SSC/SDS buffer preheated to \(50^{\circ}C\) for 10min, 1min and 1min at room temperature with agitation. The slides were then air- dried and imaged after applying Slowfade® Gold Antifade medium and a coverslip. + +<|ref|>sub_title<|/ref|><|det|>[[115, 799, 310, 816]]<|/det|> +## Library Preparation: + +<|ref|>sub_title<|/ref|><|det|>[[115, 822, 422, 841]]<|/det|> +## 1. Second Strand Synthesis + +<|ref|>text<|/ref|><|det|>[[115, 847, 884, 920]]<|/det|> +\(5\mu \mathrm{L}\) second strand synthesis mix containing \(20\mu \mathrm{L}\) of 5X First Strand Buffer, \(14\mu \mathrm{L}\) of DNA polymerase I (10U/μL) and \(3.5\mu \mathrm{L}\) Ribonuclease H (2U/μL) were added to the cleaved sample and incubated at \(16^{\circ}C\) for 2 hours. Eppendorf tubes were placed on + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 883, 177]]<|/det|> +ice and 5μL of T4 DNA polymerase (3U/μL) were added to each strand and incubated for 20 minutes at 16°C. 25μL of 80mM EDTA (mix 30μL of 500mM EDTA with 158μL DNase/RNase free water) was added to each sample and the samples were kept cool on ice. + +<|ref|>sub_title<|/ref|><|det|>[[115, 184, 350, 202]]<|/det|> +## 2. cDNA purification + +<|ref|>text<|/ref|><|det|>[[113, 208, 884, 328]]<|/det|> +cDNA from the previous step was purified using Agencourt RNAclean XP beads and DynaMag™- 2 magnetic rack, incubated at room temperature for 5 min. Further cleansing was performed by the addition of 80% Ethanol to the sample tubes, while the samples were still placed in the magnetic rack. Sample elution was then carried out using 13μL of NTP/water mix. + +<|ref|>sub_title<|/ref|><|det|>[[113, 334, 541, 353]]<|/det|> +## 3. In Vitro Transcription and Purification + +<|ref|>text<|/ref|><|det|>[[113, 359, 884, 532]]<|/det|> +cDNA transcription to aRNA was carried out by adding 4μL of reaction mix containing: 10x Reaction Buffer, T7 Enzyme mix and SUPERaseln™ RNase Inhibitor (20 U/μL) to 12μL of the eluted cDNA sample and incubated at 37°C, for 14 hours. The samples were purified using RNA clean XP beads according to the manufacturer's protocol and further eluted into 10μL DNase/RNase free water. The amount and average fragment length of amplified RNA was determined using the RNA 6000 Pico Kit (Agilent, 5067- 1513) with a 2100 Bioanalyzer according to the manufacturer's protocol. + +<|ref|>sub_title<|/ref|><|det|>[[115, 538, 339, 557]]<|/det|> +## 4. Adapter Ligation + +<|ref|>text<|/ref|><|det|>[[113, 563, 884, 709]]<|/det|> +Next, 2.5μL Ligation adapter (IDT) was added to the sample and was heated for 2 min at 70°C and then placed on ice. A total of 4.5μL ligation mix containing 11.3μL of 10X T4 RNA Ligase, T4 RNA truncated Ligase 2 and 11.3μL of murine RNase inhibitor was then added to the sample. Samples were then incubated at 25°C for 1 hour. The samples were then purified using RNAClean XP beads according to the manufacturer's protocol. + +<|ref|>sub_title<|/ref|><|det|>[[115, 716, 411, 734]]<|/det|> +## 5. Second cDNA synthesis + +<|ref|>text<|/ref|><|det|>[[113, 740, 884, 891]]<|/det|> +Purified samples were mixed with 1μL cDNA primer (IDT), 1μL dNTP mix up to a total volume of 12μL and incubated at 65°C for 5 min and then directly placed on ice. A 1.5ml Eppendorf tube 8μL of the sample was mixed with 30μL of First Strand Buffer(5X),), 7.5μL of DTT(0.1M), 7.5μL of DNase/RNase free water, 7.5μL of SuperScript® III Reverse transcriptase and 7.5μL of RNaseOUT™ Recombinant ribonuclease Inhibitor and incubated at 50°C for 1 hour followed by cDNA purification + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 881, 126]]<|/det|> +using Agencourt RNAClean XP beads according to the manufacturer's protocol. Samples were then stored at \(- 20^{\circ}C\) . + +<|ref|>title<|/ref|><|det|>[[115, 135, 355, 151]]<|/det|> +## 6. PCR amplification + +<|ref|>text<|/ref|><|det|>[[113, 158, 884, 358]]<|/det|> +Prior to PCR amplification, we determined that 20 cycles were required for appropriate amplification. A total reaction volume of 25μL containing 2x KAPA mix, 0.04μM PCRInPE2 (IDT), 0.4μM PCR InPE1.0 (IDT), 0.5μM PCR Index (IDT) and 5μL of purified cDNA were amplified using the following protocol: 98°C for 3 min followed by 20 cycles at 98°C for 20 seconds, 60°C for 30 seconds, 72°C for 30 seconds followed by 72°C for 5 minutes. The libraries were purified according to the manufacturer's protocol and eluted in 20μL EB (elution buffer). The samples were then stored at -20°C until used. + +<|ref|>title<|/ref|><|det|>[[115, 366, 440, 383]]<|/det|> +## 7. Quality control of Libraries + +<|ref|>text<|/ref|><|det|>[[113, 390, 884, 531]]<|/det|> +The average length of the prepared libraries was quantified using an Agilent DNA 1000 high sensitivity kit with a 2100 Bioanalyzer. The concentration of the libraries was determined using a Qubit dsDNA HS kit. The libraries were diluted to 4nM, pooled and denatured before sequencing on the Illumina NextSeq platform using paired-end sequencing. We used 30 cycles for read 1 and 270 cycles for read 2 during sequencing. + +<|ref|>table_caption<|/ref|><|det|>[[214, 539, 312, 555]]<|/det|> +Sequence + +<|ref|>table<|/ref|><|det|>[[115, 559, 940, 884]]<|/det|> +
Ligation Adapter cDNA primer PCR primer INPE1 PCR primer INPE2 PCR index primer/5rApp/AGATCGGAAAGAGCACACGTCTGAACTCCAGTCAC/3d- dC/
GTGACTGGAGTTCAGACGTGTGCTCTTCCGA
AATGATACGGGCACCACGAGATCTACACTCTTTCCCTACACGACGCTCT- CCGATCT
GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT
CAAGCAGAAGACGGCATACGAGATCGATGATGGACTGGAGTTC
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 131, 505, 150]]<|/det|> +## Postprocessing Spatial Transcriptomics + +<|ref|>text<|/ref|><|det|>[[113, 152, 884, 719]]<|/det|> +First, we aligned the H&E staining by the use of the st- pipeline (github.com/SpatialTranscriptomics- Research/st_pipeline). The pipeline contains the following steps: Quality trimming and removing of low quality bases (bases with quality \(< 10\) ), sanity check (reads same length, reads order, etc.), remove homopolymers, normalize for AT and GC content, mapping the read2 with STAR, demultiplexing based on read1, sort for reads (read1) with valid barcodes, annotate the reads with htsqcount, group annotated reads by barcode (spot position), gene and genomic location (with an offset) to get a read count (github.com/SpatialTranscriptomics- Research/st_pipeline). The pipeline resulted in a gene count matrix and a spatial information file containing the x and y position and the H&E image. We used the Seurat v3.0 package to normalize gene expression values by dividing each estimated cell by the total number of transcripts and multiplied by 10,000, followed by natural- log transformation. As described for sc- RNA sequencing, we removed batch effects and scaled data by a regression model including sample batch and percentage of ribosomal and mitochondrial gene expression. For further analysis we used the 2000 most variable expressed genes and decomposed eigenvalue frequencies of the first 100 principal components and determined the number of non- trivial components by comparison to randomized expression values. The obtained non- trivial components were used for SNN clustering followed by dimensional reduction using the UMAP algorithm. Differently expressed genes (DE) of each cluster were obtained using a hurdle model tailored to scRNA- seq data which is part of the MAST package. We further build a user- friendly viewer for spatial transcriptomic data and provide tutorials on analysis of data: https://themilolab.github.io/SPATA/index.html. + +<|ref|>sub_title<|/ref|><|det|>[[115, 746, 350, 765]]<|/det|> +## Spatial gene expression + +<|ref|>text<|/ref|><|det|>[[115, 770, 884, 913]]<|/det|> +For spatial expression plots, we used either normalized and scaled gene expression values (to plot single genes) or scores of a set of genes, using the 0.5 quantile of a probability distribution fitting. The x- axis and y- axis coordinates are given by the input file based on the localization at the H&E staining. We computed a matrix based on the maximum and minimum extension of the spots used (32x33) containing the gene expression or computed scores. Spots without tissue covering were set to zero. Next, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 884, 200]]<|/det|> +we transformed the matrix, using the squared distance between two points divided by a given threshold, implemented in the fields package (R- software) and adapted the input values by increasing the contrast between uncovered spots. The data are illustrated either as surface plots (plotly package R- software) or as images (graphics package R- software). + +<|ref|>sub_title<|/ref|><|det|>[[115, 229, 437, 248]]<|/det|> +## Representation of Cellular States + +<|ref|>text<|/ref|><|det|>[[113, 253, 884, 322]]<|/det|> +We aligned cells/spots to variable states with regard to gene sets (GS) that were selected GS(1,2,..n). First, we separated cells into GS(1+2) versus GS(2+4), using the following equation: + +<|ref|>equation<|/ref|><|det|>[[328, 327, 670, 348]]<|/det|> +\[A_{1} = \| G S_{(1)},G S_{(2)}\|_{\infty} - \| G S_{(3)},G S_{(4)}\|_{\infty}\] + +<|ref|>text<|/ref|><|det|>[[113, 354, 884, 399]]<|/det|> +A1 defines the y- axis of the two- dimensional representation. In a next step, we calculated the x- axis separately for spots A1<0 and A1>0: + +<|ref|>equation<|/ref|><|det|>[[319, 401, 678, 427]]<|/det|> +\[A1 > 0:A_{2} = \log 2(\overline{{G S_{(1)}}} -[\overline{{G S_{(2)}}} +1])\] + +<|ref|>equation<|/ref|><|det|>[[339, 430, 660, 454]]<|/det|> +\[A1< 0:A_{2} = \log 2(\overline{{G S_{(3)}}} -[\overline{{G S_{(4)}}} ])\] + +<|ref|>text<|/ref|><|det|>[[113, 458, 884, 553]]<|/det|> +For further visualization of the enrichment of subsets of cells according to gene set enrichment across the two- dimensional representation, we transformed the distribution to representative colors using a probability distribution fitting. This representation is an adapted method published by Neftel and colleges recently20,28. + +<|ref|>sub_title<|/ref|><|det|>[[115, 583, 380, 601]]<|/det|> +## Spatial correlation analysis + +<|ref|>text<|/ref|><|det|>[[113, 606, 884, 725]]<|/det|> +The spatial correlation was performed by integrating a deep autoencoder for background noise reduction and a Bayesian correlation model. In a first step, we performed noise reduction through an autoencoder similar to recent described for single- cell RNA- sequencing studies39. The autoencoder consist of an encoder and decoder part which can be defined as transitions: + +<|ref|>equation<|/ref|><|det|>[[339, 730, 694, 750]]<|/det|> +\[(1) encoder:\phi :X\to F decoder:\psi :F\to X\] + +<|ref|>equation<|/ref|><|det|>[[339, 755, 657, 776]]<|/det|> +\[(2)\phi ,\psi = arg.min\| X - (\phi ,\psi)X\| ^2\] + +<|ref|>text<|/ref|><|det|>[[113, 780, 884, 825]]<|/det|> +The encoder stage of an autoencoder takes the input \(x \in \mathbb{R}^d = X\) and maps it to \(z \in \mathbb{R}^p = \mathcal{F}\) at the layer position \(\phi\) : + +<|ref|>equation<|/ref|><|det|>[[299, 830, 700, 850]]<|/det|> +\[(3)x = A^{\phi = 0};z^{\phi} = ReLU(W^{\phi}\times A^{\phi -1} + b^{\phi})\] + +<|ref|>text<|/ref|><|det|>[[113, 855, 884, 925]]<|/det|> +\(z^{\phi}\) is also referred to as latent representation, here presented as \(z^{1}, z^{2}, \ldots , z^{\phi = n}\) in which \(\phi\) describes the number of hidden layers. \(\mathbf{W}\) is the weight matrix and \(\mathbf{b}\) represent the dropout or bias vector. Our network architecture contained 32 hidden layers, as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 881, 127]]<|/det|> +recommended39. In the decoder weights and biases are reconstructed through backpropagation \((\psi :\mathcal{F}\to \mathcal{X})\) and \(z\) is mapped to \(x^{\prime} = A^{0^{\prime}}\) in the shape as \(x^{\prime}\) . + +<|ref|>equation<|/ref|><|det|>[[350, 130, 644, 151]]<|/det|> +\[A^{\phi -1^{\prime}} = \sigma^{\prime}(W^{\phi^{\prime}}\times z^{\phi} + b^{\phi^{\prime}})\] + +<|ref|>text<|/ref|><|det|>[[115, 156, 881, 201]]<|/det|> +In this context, \(W^{\prime},\sigma^{\prime},b^{\prime}\) from the decoder are unrelated to \(W,\sigma ,b\) from the encoder. We used a loss function to train the network in order to minimize reconstruction errors. + +<|ref|>equation<|/ref|><|det|>[[287, 207, 707, 233]]<|/det|> +\[\mathcal{L}(x,x^{\prime}) = \left\| x - \sigma^{\prime}(W^{\prime}(\sigma (Wx + b)) + b^{\prime})\right\|^{2}\] + +<|ref|>text<|/ref|><|det|>[[115, 237, 883, 331]]<|/det|> +In the second step, we used the predicted gene expression matrix \((x^{\prime})\) and fitted and Bayesian correlation model (Bayesian First Aid, https://github.com/rasmusab/bayesian_first_aid). An illustration of the spatial correlation is given in the supplementary figure 7 + +<|ref|>sub_title<|/ref|><|det|>[[115, 361, 611, 380]]<|/det|> +## sc-RNA-sequencing integration into spatial context + +<|ref|>text<|/ref|><|det|>[[115, 385, 883, 455]]<|/det|> +In order to integrate determined cluster into the spatial context of the spatial transcriptomic data. We used the recent published spotlighted algorithm and integrated the output into our SPATA objects for visualization21. + +<|ref|>sub_title<|/ref|><|det|>[[115, 484, 588, 504]]<|/det|> +## RNA-velocity and Pseudotime trajectory analysis + +<|ref|>text<|/ref|><|det|>[[114, 508, 884, 772]]<|/det|> +In order to determine dynamic gene expression changes, we extracted spliced and unspliced genes from the bam output created by cell ranger using the velocyto.py tool40. The resulting \*.loom files were merged and transformed into .h5ad format for further processing by scVelo41 and CellRank42. The pipeline is integrated into the SPATA toolbox43. Single-cell data are reformatted into an SPATA S4 object using the UMAP coordinates as spatial coordinates. Outputs of the scVelo script (implemented in the development branch of the SPATA toolbox) are imported into the SPATA S4 object (slot:@fdata) and available for visualization. RNA- velocity streams are converted into trajectories and imported to the SPATA S4 object (slot:@trajectories). Dynamic gene expression changes along trajectories are performed by the assessTrajectoryTrends() function. + +<|ref|>sub_title<|/ref|><|det|>[[117, 804, 404, 822]]<|/det|> +## Gene set enrichment analysis + +<|ref|>text<|/ref|><|det|>[[115, 828, 883, 896]]<|/det|> +Gene sets were obtained from the database MSigDB v7 and internally created gene sets are available at github.com/heilandd. For enrichment analysis of single clusters, the normalized and centered expression data were used and further transformed to z- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 881, 126]]<|/det|> +scores ranging from 1 to 0. Genes were ranked in accordance to the obtained differential expression values and used as the input for GSEA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 181, 395, 200]]<|/det|> +## Identification of cycling cells + +<|ref|>text<|/ref|><|det|>[[113, 205, 884, 348]]<|/det|> +We used the set of genes published by Neftel and colleagues to calculate proliferation scores based on the GSVA package implemented in R- software. The analysis based on a non- parametric unsupervised approach, which transformed a classic gene matrix (gene- by- sample) into a gene set by sample matrix resulted in an enrichment score for each sample and pathway. From the output enrichment scores we set a threshold based on distribution fitting to define cycling cells. + +<|ref|>sub_title<|/ref|><|det|>[[115, 377, 597, 396]]<|/det|> +## Nearest Functionally Connected Neighbor (NFCN) + +<|ref|>text<|/ref|><|det|>[[112, 401, 884, 616]]<|/det|> +To identify connected cells that interact by defined activation or inhibition of downstream signals in the responder cell, we created a novel model. Therefore, we assumed that a cell- cell interaction is given only if a receptor/ligand pair induce correspondent down- stream signaling within the responder cell (cell with expressed receptor). Furthermore, we take into account that the importance of an activator cell (cell with expressed ligand) can be ranked according to their enriched signaling, which is responsible for inducing ligand expression. Based on these assumptions we defined an algorithm to map cells along an interaction- trajectory. The algorithm was designed to identify potential activators from a defined subset of cells. + +<|ref|>text<|/ref|><|det|>[[112, 621, 884, 912]]<|/det|> +As input for the analysis we used a normalized and scaled gene expression matrix, a string containing the subset of target cells, a list of genes defining ligand induction on the one side and receptor signaling on the other side. These genes were chosen either by the MSigDB v7 database or our stimulation library explained above. Next, we downscaled the data to 3000 representative cells including all myeloid cell types and calculated the enrichment of induction and activation of the receptor/ligand pair. Enrichment scores were calculated by singular value decomposition (SVD) over the genes in the gene set and the coefficients of the first right- singular vector defined the enrichment of induction/activation profile. Both expression values and enrichment scores were fitted by a probability distribution model and cells outside the 95% quantile were removed. Next, we fitted a model using a non- parametric kernel estimation (Gaussian or Cauchy- Kernel), on the basis of receptor/ligand expression (Aexp) and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 80, 881, 125]]<|/det|> +up/downstream signaling (Aeff) of each cell (i={1,..n}). Both input vectors were normalized and z- scored: + +<|ref|>equation<|/ref|><|det|>[[140, 155, 857, 190]]<|/det|> +\[n_{\mathrm{exp}i} = \frac{A_{\mathrm{exp}i} - min(A_{\mathrm{exp}})}{max(A_{\mathrm{exp}}) - min(A_{\mathrm{exp}})} \qquad (2) \widehat{f}_h(n_{\mathrm{exp}i}) = \frac{1}{n}\sum_{i = 1}^{n}K_h(n_{\mathrm{exp}} - n_{\mathrm{exp}i})\] + +<|ref|>text<|/ref|><|det|>[[115, 216, 884, 386]]<|/det|> +K is the kernel and \(0.7 > h > 0.3\) is used to adjust the estimator. The model resulted in a trajectory which was defined as Ligand \((- )\) - Induction \((- )\) to cells of the target subset with Receptor \((- )\) - Activation \((- )\) . Further cells were aligned along the "interaction- trajectory". We defined connected cells by reaching the upper \(70\%\) Cl in receptor/ligand expression as well as sores of induction/activation. The process of representation is illustrated schematically in Supplementary Figure 4. Additionally we determined receptor- ligand interaction by the NicheNet software as recommended by the authors \(^{44}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 415, 275, 433]]<|/det|> +## CNV estimation: + +<|ref|>text<|/ref|><|det|>[[115, 438, 884, 680]]<|/det|> +Copy- number Variations (CNVs) were estimated by aligning genes to their chromosomal location and applying a moving average to the relative expression values, with a sliding window of 100 genes within each chromosome, as described recently \(^{18}\) . First, we arranged genes in accordance to their respective genomic localization using the CONICSmat package (R- software). As a reference set of non- malignant cells, we in- silico extracted 400 CD8 positive cells (unlikely to be expressed on tumor cells). To avoid the considerable impact of any particular gene on the moving average we limited the relative expression values [- 2.6,2.6] by replacing all values above/below \(exp_{(i)} = [2.6]\) , by using the infercnv package (R- software). This was performed only in the context of CNV estimation as previous reported \(^{11}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 710, 273, 728]]<|/det|> +## Flow cytometry: + +<|ref|>text<|/ref|><|det|>[[115, 733, 884, 927]]<|/det|> +Single- Cell suspensions were obtained after Dead- Cell Removal and CD3 MACS- enrichment. Cells were incubated with VivaFix™ 398/550 (BioRad Laboratories, CA, USA) according to the manufacturer's instructions. Cells were fixed in \(4\%\) paraformaldehyde (PFA) for 10 minutes. After centrifugation (350 g; \(4^{\circ}C\) ; 5 min) and removal of the supernatant, the cell pellet was suspended in \(0.5 \text{ml} 4^{\circ}C\) cold FACS buffer. Cell suspensions were washed and centrifuged at 350xg for 5 mins, followed by resuspension in FACS buffer. The washing step was repeated twice. Finally, cells were resuspended in at least 0.5 to \(1 \text{ml}\) of FACS buffer depending on the number of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 883, 150]]<|/det|> +cells. We used a Sony SP6800 spectral analyzer in standardization mode with PMT voltage set to maximum to reach a saturation rate below \(0.1\%\) . Gating was performed by FCS Express 7 plus at the Lighthouse Core Facility, University of Freiburg. + +<|ref|>sub_title<|/ref|><|det|>[[115, 181, 323, 199]]<|/det|> +## Immunofluorescence + +<|ref|>text<|/ref|><|det|>[[113, 205, 883, 445]]<|/det|> +The same protocol was followed for human neocortical slices with or without microglia and tumor cell injection. The media was removed and exchanged for \(1 \text{mL}\) of \(4\%\) paraformaldehyde (PFA) for \(1 \text{h}\) and further incubated in \(20\%\) methanol in PBS for 5 minutes. Slices were then permeabilized by incubating in PBS supplemented with \(1\%\) Triton (TX- 100) overnight at \(4^{\circ}\text{C}\) and further blocked using \(20\%\) BSA for 4 hours. The permeabilized and blocked slices were then incubated by primary antibodies in \(5\%\) BSA- PBS incubated overnight at \(4^{\circ}\text{C}\) . After washing in PBS, slices were labelled with secondary antibodies conjugated with Alexa 405, 488, 555, or 568 for 3 hours at room temperature. Finally, slices were mounted on glass slides using DAPI fluoromount (Southern Biotech, Cat. No. 0100- 20), as recently described15. + +<|ref|>sub_title<|/ref|><|det|>[[115, 475, 441, 494]]<|/det|> +## Human Organotypic Slice Culture + +<|ref|>text<|/ref|><|det|>[[113, 499, 884, 789]]<|/det|> +Human neocortical slices were prepared as recently described15,25. Capillaries and damaged tissue were dissected away from the tissue block in the preparation medium containing Hibernate medium supplemented with \(13 \text{mM D + }\) Glucose and \(30 \text{mM}\) NMDG. Coronal slices of \(300 \mu \text{m}\) thickness were sectioned using a vibratome (VT1200, Leica Germany) and incubated in preparation medium for 10 minutes before plating to avoid any variability due to tissue trauma. Three to four slices were gathered per insert. The transfer of the slices was facilitated by a polished wide mouth glass pipette. Slices were maintained in growth medium containing Neurobasal (L- Glutamine) supplemented with \(2\%\) serum free B- 27, \(2\%\) Anti- Anti, \(10 \text{mM D + }\) Glucose, \(1 \text{mM MgSO4}\) , and \(1 \text{mM Glutamax}\) at \(5\% \text{CO}_2\) and \(37^{\circ}\text{C}\) . The entire medium was replaced with fresh culture medium 24 hours post plating, and every 48 hours thereafter. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 608, 100]]<|/det|> +## Chemical depletion of Microglia from slice cultures + +<|ref|>text<|/ref|><|det|>[[115, 107, 884, 199]]<|/det|> +Selective depletion of the myeloid cell compartment in human neocortical slices was performed by supplementing the growth medium with \(50~\mu \mathrm{M}\) for 24hrs and \(11\mu \mathrm{M}\) of Clodronate (Sigma, D4434) for the next 48hrs at \(37^{\circ}C\) . Subsequently, the slices were carefully rinsed with growth medium to wash away any debris. + +<|ref|>sub_title<|/ref|><|det|>[[115, 230, 525, 249]]<|/det|> +## Tumor/T cell injection onto tissue cultures + +<|ref|>text<|/ref|><|det|>[[115, 254, 884, 345]]<|/det|> +ZsGreen tagged BTSC#233 cell lines cultured and prepared as described in the cell culture section. Post trypsinization, a centrifugation step was performed, following which the cells were harvested and suspended in MEM media at 20,000 cells/ul. A 10 \(\mu \mathrm{L}\) Hamilton syringe was used to inject \(1\mu \mathrm{L}\) of GBM cells into the slices and were incubated at \(37^{\circ}C\) for 3 days with a medium change every other days. + +<|ref|>text<|/ref|><|det|>[[115, 330, 884, 616]]<|/det|> +Blood samples were collected from the same donors from whom we obtained the healthy cortex for our organotypic slice cultures. Peripheral T cells were isolated using the same MACSxpress® Whole Blood Pan T Cell Isolation Kit (Miltenyi Biotech) and erythrocytes were eliminated from the suspension using ACK- lysis buffer (Thermo Fisher Scientific). T cells were tagged using the Cell Trace Far Red dye (Thermo Fisher Scientific) prior to injection into the slices. To block endogenous IL- 10 receptor in T- cells, the neutralizing antibody anti- IL10 hAB (R&D systems) were added to the cells at the concentration of \(5\mu \mathrm{g / ml}\) for 1h and incubated at \(37^{\circ}C\) . Both IL10 knock out T cells and normal T cells were injected onto the sections and left it for another 48hrs at \(37^{\circ}C\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 648, 483, 667]]<|/det|> +## Enzyme linked Immunosorbent Assay + +<|ref|>text<|/ref|><|det|>[[115, 672, 884, 790]]<|/det|> +An enzyme linked immunosorbent assay (ELISA) was performed in order to measure cytokine concentrations of IL- 2, IL- 10, IL- 13 and IFN- gamma in the cell culture medium 48h after T cell injection. The Multi- Analyte ELISAarray Kit (Qiagen, Venlo, Netherlands; MEH- 003A) was used according to the manufacturer's instructions. Absorbance was measured using the Tecan Infinite® 200 (Tecan, Männedorf, Switzerland). + +<|ref|>sub_title<|/ref|><|det|>[[115, 820, 493, 838]]<|/det|> +## Treatment of patient with JAK-inhibitor + +<|ref|>text<|/ref|><|det|>[[115, 845, 883, 887]]<|/det|> +A patient with a recurrent glioblastoma was treated with a daily dose of 40mg Ruxolitinib for 4 weeks. Before treatment, we confirmed the progress by a biopsy. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 883, 127]]<|/det|> +treatment was performed as a neoadjuvant therapy. After 4 weeks, the patient underwent a gross- total surgery and adjuvant Temozolomide therapy. + +<|ref|>sub_title<|/ref|><|det|>[[117, 181, 299, 199]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[115, 205, 883, 273]]<|/det|> +DHI is funded by the Else- Kröner Stiftung and BMBF. VM Ravi was partially funded by the BMBF- projects FMT (13GW0230A). We thank Manching Ku and Dietmar Pfeifer for their helpful advice. We acknowledge Biorender.com. + +<|ref|>sub_title<|/ref|><|det|>[[117, 303, 309, 321]]<|/det|> +## Conflict of interests + +<|ref|>text<|/ref|><|det|>[[115, 327, 677, 347]]<|/det|> +No potential conflicts of interest were disclosed by the authors. + +<|ref|>sub_title<|/ref|><|det|>[[117, 402, 272, 420]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 426, 884, 569]]<|/det|> +scRNA- Sequencing Data available: (in preparation), Accession codes: www.github.com/heilandd/.VisLabv1.5 https://github.com/heilandd/Vis_Lab1.5, NFCN Algorithm www.github.com/heilandd/NFCN, SPATA- Lab: www.github.com/heilandd/SPATA- Lab. Further information and requests for resources, raw data and reagents should be directed and will be fulfilled by the Contact: D. H. Heiland, dieter.henrik.heiland@uniklinik- freiburg.de. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 100, 885, 920]]<|/det|> +Bibliography1. Darmanis, S. et al. Single-Cell RNA-Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep. 21, 1399–1410 (2017).2. Woroniecka, K. et al. T-Cell Exhaustion Signatures Vary with Tumor Type and Are Severe in Glioblastoma. Clin. Cancer Res. 24, 4175–4186 (2018).3. Chen, Z. & Hambardzumyan, D. Immune microenvironment in glioblastoma subtypes. Front. Immunol. 9, 1004 (2018).4. Sankowski, R. et al. Mapping microglia states in the human brain through the integration of high-dimensional techniques. Nat. Neurosci. 22, 2098–2110 (2019).5. Zhang, L. et al. Lineage tracking reveals dynamic relationships of T cells in colorectal cancer. Nature 564, 268–272 (2018).6. Zheng, C. et al. Landscape of Infiltrating T Cells in Liver Cancer Revealed by Single-Cell Sequencing. Cell 169, 1342–1356.e16 (2017).7. Baitsch, L. et al. Exhaustion of tumor-specific CD8+ T cells in metastases from melanoma patients. J. Clin. Invest. 121, 2350–2360 (2011).8. Singer, M. et al. A Distinct Gene Module for Dysfunction Uncoupled from Activation in Tumor-Infiltrating T Cells. Cell 166, 1500–1511.e9 (2016).9. Anderson, A. C., Joller, N. & Kuchroo, V. K. Lag-3, Tim-3, and TIGIT: Co-inhibitory Receptors with Specialized Functions in Immune Regulation. Immunity 44, 989–1004 (2016).10. Im, S. J. et al. Defining CD8+ T cells that provide the proliferative burst after PD-1 therapy. Nature 537, 417–421 (2016).11. Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).12. Platten, M., Ochs, K., Lemke, D., Opitz, C. & Wick, W. Microenvironmental clues for glioma immunotherapy. Curr Neurol Neurosci Rep 14, 440 (2014).13. Filley, A. C., Henriquez, M. & Dey, M. Recurrent glioma clinical trial, CheckMate-143: the game is not over yet. Oncotarget 8, 91779–91794 (2017).14. Weller, M. et al. Rindopepimut with temozolomide for patients with newly diagnosed, EGFRvIII-expressing glioblastoma (ACT IV): a randomised, double-blind, international phase 3 trial. Lancet Oncol. 18, 1373–1385 (2017).15. Henrik Heiland, D. et al. Tumor-associated reactive astrocytes aid the evolution of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 80, 885, 120]]<|/det|> +immunosuppressive environment in glioblastoma. Nat. Commun. 10, 2541 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 105, 886, 920]]<|/det|> +16. Zhang, Z. et al. SCINA: A Semi-Supervised Subtyping Algorithm of Single Cells and Bulk Samples. Genes (Basel) 10, (2019). +17. Aran, D. et al. Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage. Nat. Immunol. 20, 163–172 (2019). +18. Patel, A. P. et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014). +19. Chen, H. et al. Single-cell trajectories reconstruction, exploration and mapping of omics data with STREAM. Nat. Commun. 10, 1903 (2019). +20. Neftel, C. et al. An integrative model of cellular states, plasticity, and genetics for glioblastoma. Cell 178, 835–849.e21 (2019). +21. Elosua, M., Nieto, P., Mereu, E., Gut, I. & Heyn, H. SPOTlight:Seeded NMF regression to Deconvolute Spatial Transcriptomics Spots with Single-Cell Transcriptomes. BioRxiv (2020). doi:10.1101/2020.06.03.131334 +22. Naito, Y., Takagi, T. & Higashimura, Y. Heme oxygenase-1 and anti-inflammatory M2 macrophages. Arch. Biochem. Biophys. 564, 83–88 (2014). +23. Qin, Y. et al. A Milieu Molecule for TGF-β Required for Microglia Function in the Nervous System. Cell 174, 156–171.e16 (2018). +24. Giladi, A. et al. Dissecting cellular crosstalk by sequencing physically interacting cells. Nat. Biotechnol. 38, 629–637 (2020). +25. Ravi, V. M. et al. Human organotypic brain slice culture: a novel framework for environmental research in neuro-oncology. Life Sci. Alliance 2, (2019). +26. Kling, T. et al. Integrative Modeling Reveals Annexin A2-mediated Epigenetic Control of Mesenchymal Glioblastoma. EBioMedicine 12, 72–85 (2016). +27. Venteicher, A. S. et al. Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science 355, (2017). +28. Tirosh, I. et al. Single-cell RNA-seq supports a developmental hierarchy in human oligodendroglioma. Nature 539, 309–313 (2016). +29. Blank, C. U. et al. Defining “T cell exhaustion”. Nat. Rev. Immunol. 19, 665–674 (2019). +30. Jiang, Y., Li, Y. & Zhu, B. T-cell exhaustion in the tumor microenvironment. Cell Death Dis. 6, e1792 (2015). +31. Wherry, E. J. T cell exhaustion. Nat. Immunol. 12, 492–499 (2011). +32. Li, H. et al. Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 887, 747]]<|/det|> +1007 Compartment within Human Melanoma. Cell 176, 775- 789. e18 (2019). 1008 33. Woroniecka, K. I., Rhodin, K. E., Chongsathidkiet, P., Keith, K. A. & Fecci, P. E. T- cell Dysfunction in Glioblastoma: Applying a New Framework. Clin. Cancer Res. 24, 3792- 3802 (2018). 1011 34. Winkler, F. & Bengsch, B. Use of mass cytometry to profile human T cell exhaustion. Front. Immunol. 10, 3039 (2019). 1013 35. Wurm, J. et al. Astrogliosis Releases Pro- Oncogenic Chitinase 3- Like 1 Causing MAPK Signaling in Glioblastoma. Cancers (Basel) (2019). 1015 36. Priego, N. et al. STAT3 labels a subpopulation of reactive astrocytes required for brain metastasis. Nat. Med. 24, 1024- 1035 (2018). 1017 37. Kaiser, S. et al. Neuroprotection after Hemorrhagic Stroke Depends on Cerebral Heme Oxygenase- 1. Antioxidants (Basel) 8, (2019). 1019 38. Sebastián, V. P. et al. Heme Oxygenase- 1 as a Modulator of Intestinal Inflammation Development and Progression. Front. Immunol. 9, 1956 (2018). 1021 39. Eraslan, G., Simon, L. M., Mircea, M., Mueller, N. S. & Theis, F. J. Single- cell RNA- seq denoising using a deep count autoencoder. Nat. Commun. 10, 390 (2019). 1023 40. La Manno, G. et al. RNA velocity of single cells. Nature 560, 494- 498 (2018). 1024 41. Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & Theis, F. J. Generalizing RNA velocity to transient cell states through dynamical modeling. Nat. Biotechnol. 38, 1408- 1414 (2020). 1027 42. Lange, M. et al. CellRank for directed single- cell fate mapping. BioRxiv (2020). doi:10.1101/2020.10.19.345983 1029 43. Kueckelhaus, J. et al. Inferring spatially transient gene expression pattern from spatial transcriptomic studies. BioRxiv (2020). doi:10.1101/2020.10.20.346544 1031 44. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat. Methods 17, 159- 162 (2020). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 198, 100]]<|/det|> +## Figures: + +<|ref|>text<|/ref|><|det|>[[113, 107, 883, 430]]<|/det|> +Figure 1: a) Illustration of the workflow, tissue specimens were obtained from 11 glioblastoma patients. Samples from 8 patients were used for scRNA- seq and 3 for spatial transcriptomics. b) Dimensional reduction using UMAP, cell type was determined by SingleR (github.com/dviraran/SingleR). c) Dimensional reduction (UMAP) of CD3+/CD8+ cells. SNN-clustering reveal 13 different clusters. RNA- velocity based pseudotime is presented by streams obtained from dynamic modeling in scvelo. Marker expression in different subpopulations ranked by grade of activity. d) Dimensional reduction plots of estimated gene- set enrichment (left), of cell proliferation (middle) and regulatory marker genes (right). e,f) Detailed presentation of 3 major trajectories and its changes of estimated pseudotime across different models. Velocity trajectory 2 reveals almost no temporal changes, but is marked by increasing IFNg signaling along the trajectory. g) Line plot of IFNg signalling gene expression along the velocity trajectory 2 (left) and dimensional reduction (right). h) Inferring dynamic alterations along the trajectory revealed up- regulation of IL- 10 and TGF- \(\beta\) signaling, presented as a line plot (top). (i) Gene set enrichment analysis of the IL10 signaling enrichment in defined start and destination region of the trajectory (bottom). j) Mapping of gene expression along the pseudotime trajectory. Bars at the bottom indicate the enrichment of each cell to enrich effector or exhausted signatures. Line plots on the top, showed enrichment for the T cell state signatures. + +<|ref|>text<|/ref|><|det|>[[113, 458, 883, 618]]<|/det|> +Figure 2: a) Workflow of spatial transcriptomics b) 2D representation of heterogeneous states in glioblastoma by Neftel, colors indicate the expression of cycling cells (quantile). c) H&E staining and (d) gene expression enrichment maps of the mesenchymal state defined by Neftel20. e) Enrichment maps of T cell clusters by spotlight algorithm21 in different samples. Maps are colored by a normalized spotlight score. f) Correlation heatmap of a Bayesian correlation model of T cell clusters and glioblastoma transcriptional subtypes. (g) Scatter plots of spot- wise correlation of HAVCR2 and LAG3 expression and T cells (left side), mesenchymal- like gene expression (middle) and NPC- like gene expression (right side), Rho correlation and CI95% are given at the top. + +<|ref|>text<|/ref|><|det|>[[112, 642, 883, 926]]<|/det|> +Figure 3: a) Workflow exploring cell- cell interactions using two approaches: 1. Predict functional neighbors based on defined ligand- receptor interaction. 2. Validation in spatial transcriptomic datasets. b) Cell- cell interaction plot as explained in supplementary figure 4. Cells with an interaction- score above 0.95 are mapped to the UMAP. c) Volcano plot of differential gene expression between highly connected cells (CI>97.5%, left side) vs non- connected cells (CI<2.5%, right side), adjusted - log(p- vale) (FDR) was used at the y- axis. Red cells are defined by fold- change above 2 and FDR < 0.05. d) Spatial surface plots of gene expression and pathway enrichment. e) Multilayer representation of cellular interaction. At the top layer, a UMAP is shown colored by expression levels of HMOX1. Each red line represents a cell (upper CI>99% NFCNA- Score) and its most likely position within the spatial dataset. The second layer is a spatial transcriptomic dataset, colored by the predicted NFCNA- score. At the bottom, the third layer represents a UMAP colored accordingly to CD3 expression. Green lines showed the upper CI>99% of receiver cells. f) Violin plots of the mean NFCNA score in each cluster. g) Violin plots of gene expression between connected cells (CI>97.5%, left side) vs non- connected cells (CI<2.5%, right side). Wilcoxon Rank Sum test and FDR adjustment was used for statistical testing. h) Gene Set enrichment analysis + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 883, 140]]<|/det|> +of four different gene sets. i) Heatmap of differently expressed genes or connected myeloid (right) and lymphoid (middle) cells and the cluster of doublets. At the top, scatterplot of total UMIs per cell in each group. + +<|ref|>text<|/ref|><|det|>[[113, 203, 884, 590]]<|/det|> +Figure 4: a) Experimental workflow of the neocortical GBM model autografted with patient derived T cells and with/without myeloid cell depletion. b) Immunostainings of IBA1 (Macrophages and Microglia) in magenta and HMOX1 in cyan, tumor cells are depicted in grey. In the upper panel, the control set with no myeloid cell depletion (M+) is shown, the bottom panel contains the myeloid cell depleted sections. c) ELISA measurements of IL10 d) Immunostainings of T cells (CSFE- Tagged, in red) and GZMB, a marker of T cell activation (green). e, f) ELISA measurements of IL2 and IFNg g) Immunostainings of TIM3 (gene: HAVCR2) in yellow, which was identified in the ScRNA- seq, and T cell in red. h) Illustration of the workflow. i) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) with pre- treatment in anti- IL- 10R antibodies. j) ELISA measurements of IL2 in ctr and IL10R inhibition. k) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) treated with JAK- inhibitor Ruxolitinib. l) Heatmap of interleukin intensities different environment. Information regarding the presence of myeloid cells are given at the bottom. m) Illustration of the workflow of JAK inhibition in a recurrent GBM patient. n- o) Immunohistochemistry of immune marker and its quantification, white: pre therapy, gray: post therapy p) Dimensional reduction of single- cell RNA- sequencing of the Ruxolitinib treated patient revealed a large percentage of T cells (right side). q) Comparison of T cells from the Ruxolitinib treated patient and the non- treated cohort. r) A barplot that indicates the cluster specific enrichment of JAK treated T cells. P- values are determined by one- way ANOVA (c,e,f,j) adjusted by Benjamini- Hochberger (c,e,f,j) for multiple testing. Data is given as mean ± standard deviation. + +<|ref|>sub_title<|/ref|><|det|>[[116, 614, 349, 633]]<|/det|> +## Supplementary Figures: + +<|ref|>text<|/ref|><|det|>[[113, 660, 884, 784]]<|/det|> +Supplementary Figure 1: a) Workflow and representative FACS images. b) UMAP representation of all detected cell types. c) Distribution of patients across all clusters and cell types. d) UMAP representation of all determined clusters (SNN) e) The graph representing the percentage of each patient in different clusters f) Signature genes of each cluster g- i) UMAP representation of Isolated T cells spitted into CD4+ and CD8+ cells and the total number of clusters (13). j) Heatmap of signature genes + +<|ref|>text<|/ref|><|det|>[[113, 806, 883, 844]]<|/det|> +Supplementary Figure 2: a) Copy- number alterations based on single cell data. Only a small subset of tumor cells was found in the OPC cluster. b) Gene expression maps of common marker genes. + +<|ref|>text<|/ref|><|det|>[[113, 867, 884, 925]]<|/det|> +Supplementary Figure 3: a) Workflow to build a library of stimulated T- cells b) T cell stimulation in order to build a library for cytokine effects, illustrated is a heatmap of the 10 most significant marker genes of each stimulation state, based on PAMR algorithm implemented in the AutoPipe. c) Dimensional + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 881, 120]]<|/det|> +reduction (UMAP) of gene expression of the different simulation experiments. d) Z- scored expression of each stimulation signature along the velocity trajectory 1 (Figure 1). + +<|ref|>text<|/ref|><|det|>[[115, 142, 671, 160]]<|/det|> +Supplementary Figure 4: a) Workflow and concept of the NFCN Analysis + +<|ref|>text<|/ref|><|det|>[[115, 183, 881, 221]]<|/det|> +Supplementary Figure 5: a) Results from the Nichnet algorithm, left: The receptor- target interaction, right: the receptor- ligand network. + +<|ref|>text<|/ref|><|det|>[[115, 244, 881, 282]]<|/det|> +Supplementary Figure 6: a) Kaplan- Meier survival estimation of HMOX1 high/low expression GBM. b- c) Expression of HMOX1 in different regions of the tumor (b) and in de- novo and recurrent stage (c). + +<|ref|>text<|/ref|><|det|>[[115, 305, 881, 364]]<|/det|> +Supplementary Figure 7: a) Illustration of the workflow to determine the spatial correlation using a deep autoencoder for denoising followed by a Bayesian correlation model. b) Examples of the predicted overlap of T cells and CD163/HMOX1(+) myeloid cells. c) Distribution of the predicted correlation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[40, 90, 950, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 694, 115, 713]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[41, 735, 937, 940]]<|/det|> +a) Illustration of the workflow, tissue specimens were obtained from 11 glioblastoma patients. Samples from 8 patients were used for scRNA-seq and 3 for spatial transcriptomics. b) Dimensional reduction using UMAP, cell type was determined by SingleR (github.com/dviraran/SingleR). c) Dimensional reduction (UMAP) of CD3+/CD8+ cells. SNN-clustering reveal 13 different clusters. RNA- velocity based pseudotime is presented by streams obtained from dynamic modeling in scvelo. Marker expression in different subpopulations ranked by grade of activity. d) Dimensional reduction plots of estimated gene-set enrichment (left), of cell proliferation (middle) and regulatory marker genes (right). e,f) Detailed presentation of 3 major trajectories and its changes of estimated pseudotime across different models. Velocity trajectory 2 reveals almost no temporal changes, but is marked by increasing IFNg signaling + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 952, 201]]<|/det|> +along the trajectory. g) Line plot of IFNg signalling gene expression along the velocity trajectory 2 (left) and dimensional reduction (right). h) Inferring dynamic alterations along the trajectory revealed upregulation of IL-10 and TGF- \(\beta\) signaling, presented as a line plot (top). (i) Gene set enrichment analysis of the IL10 signaling enrichment in defined start and destination region of the trajectory (bottom). j) Mapping of gene expression along the pseudotime trajectory. Bars at the bottom indicate the enrichment of each cell to enrich effector or exhausted signatures. Line plots on the top, showed enrichment for the T cell state signatures8. + +<|ref|>image<|/ref|><|det|>[[40, 203, 955, 448]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 465, 117, 485]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[39, 508, 953, 689]]<|/det|> +a) Workflow of spatial transcriptomics b) 2D representation of heterogeneous states in glioblastoma by Neftel, colors indicate the expression of cycling cells (quantile). c) H&E staining and (d) gene expression enrichment maps of the mesenchymal state defined by Neftel20. e) Enrichment maps of T cell clusters by spotlight algorithm21 in different samples. Maps are colored by a normalized spotlight 1 score. f) Correlation heatmap of a Bayesian correlation model of T cell clusters and glioblastoma transcriptional subtypes. (g) Scatter plots of spot-wise correlation of HAVCR2 and LAG3 expression and T cells (left side), mesenchymal-like gene expression (middle) and NPC-like gene expression (right 1065 side), Rho correlation and CI95% are given at the top. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[40, 40, 960, 515]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 537, 116, 555]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[40, 576, 951, 940]]<|/det|> +a) Workflow exploring cell-cell interactions using two approaches: 1. Predict functional neighbors based on defined ligand-receptor interaction. 2. Validation in spatial transcriptomic datasets. b) Cell-cell interaction plot as explained in supplementary figure 4. Cells with an interaction-score above 0.95 are mapped to the UMAP. c) Volcano plot of differential gene expression between highly connected cells (CI>97.5%, left side) vs non-connected cells (CI<2.5%, right side), adjusted -log(p-vale) (FDR) was used at the y-axis. Red cells are defined by fold-change above 2 and FDR < 0.05. d) Spatial surface plots of gene expression and pathway enrichment. e) Multilayer representation of cellular interaction. At the top layer, a UMAP is shown colored by expression levels of HMOX1. Each red line represents a cell (upper CI>99% NFcNA-Score) and its most likely position within the spatial dataset. The second layer is a spatial transcriptomic dataset, colored by the predicted NFcNA-score. At the bottom, the third layer represents a UMAP colored accordingly to CD3 expression. Green lines showed the upper CI>99% of receiver cells. f) Violin plots of the mean NFcNA score in each cluster. g) Violin plots of gene expression between connected cells (CI>97.5%, left side) vs non-connected cells (CI<2.5%, right side). Wilcoxon Rank Sum test and FDR adjustment was used for statistical testing. h) Gene Set enrichment analysis of four different gene sets. i) Heatmap of differently expressed genes or connected myeloid (right) and 1lymphoid (middle) cells and the cluster of doublets. At the top, scatterplot of total UMLs per cell in each 1083 group. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 40, 844, 787]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 802, 117, 820]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 951, 955]]<|/det|> +a) Experimental workflow of the neocortical GBM model autografted with patient derived T cells and with/without myeloid cell depletion. b) Immunostaining of IBA1 (Macrophages and Microglia) in magenta and HMOX1 in cyan, tumor cells are depicted in grey. In the upper panel, the control set with no myeloid cell depletion (M+) is shown, the bottom panel contains the myeloid cell depleted sections. c) ELISA measurements of IL10 d) Immunostaining of T cells (CSFE-Tagged, in red) and 1 GZMB, a marker + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 955, 360]]<|/det|> +of T cell activation (green).). e, f) ELISA measurements of IL2 and IFNg g) Immunostainings of TIM3 (gene: HAVCR2) in yellow, which was identified in the ScRNA- seq, and T cell in red. h) Illustration of the workflow. i) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, inred) and GZMB a marker of T cell activation (green) with pre- treatment in anti- IL- 10R antibodies. j) ELISA measurements of IL2 in ctr and IL10R inhibition. k) Immunostainings of tumor cells (grey), T cells (CSFE- Tagged, in red) and GZMB a marker of T cell activation (green) treated with JAK- inhibitor Ruxolitinib. l) Heatmap of interleukin intensities different environment. Information regarding the presence of myeloid cells are given at the bottom. m) Illustration of the workflow of JAK inhibition in a recurrent GBM patient. n- o) Immunohistochemistry of immune marker and its quantification, white: pre therapy, gray: post therapy p) Dimensional reduction of single- cell RNA- sequencing of the Ruxolitinib treated patient revealed a large percentage of T cells (right side). q) Comparison of T cells from the Ruxolitinib treated patient and the non- treated cohort. r) A barplot that indicates the cluster specific enrichment of JAK treated T cells. P- values are determined by one- way ANOVA (c,e,f,j) adjusted by Benjamini- Hochberger (c,e,f,j) for multiple testing. Data is given as mean ± standard deviation. + +<|ref|>sub_title<|/ref|><|det|>[[44, 384, 311, 411]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 434, 765, 455]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 472, 264, 650]]<|/det|> +Supplementary1. pdf Supplementary2. pdf Supplementary3. pdf Supplementary4. pdf Supplementary5. pdf Supplementary6. pdf Supplementary7. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/images_list.json b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..add17a2c7a986bc5eb411337d1cb9b08e2cee53e --- /dev/null +++ b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 52, + 50, + 920, + 670 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 50, + 115, + 930, + 655 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 57, + 45, + 920, + 444 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 50, + 52, + 936, + 520 + ] + ], + "page_idx": 17 + } +] \ No newline at end of file diff --git a/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca.mmd b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca.mmd new file mode 100644 index 0000000000000000000000000000000000000000..567c2c89e4ae7a44948f027297c7649727f02556 --- /dev/null +++ b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca.mmd @@ -0,0 +1,293 @@ + +# Mechanistic evaluation of enhanced graphene toxicity to Bacillus induced by humic acid adsorption + +Qing Zhao zhaoqing@iae.ac.cn + +Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +Xuejiao Zhang Institute of Applied Ecology, Chinese Academy of Sciences + +Jin Zeng Institute of Applied Ecology, Chinese Academy of Sciences + +Jason White The Connecticut Agricultural Experiment Station + +Fangbai Li Guangdong Academy of Sciences + +Zhiqiang Xiong Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +Siyu Zhang Institute of Applied Ecology, Chinese Academy of Sciences + +Yuze Xu Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +Jingjing Yang Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +Weihao Tang Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +Fengchang Wu Chinese Research Academy of Environmental Sciences + +Baoshan Xing University of Massachusetts Amherst https://orcid.org/0000- 0003- 2028- 1295 + +Article + +Keywords: + +<--- Page Split ---> + +Posted Date: September 21st, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3294178/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on January 3rd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55270- 2. + +<--- Page Split ---> + +## Abstract + +AbstractThe extensive application of graphene nanosheets (GNSs) has raised concerns over risks to sensitive species in the aquatic environment. The humic acid (HA) corona is traditionally considered to reduce GNSs toxicity. Here, we evaluated the effect of sorbed HA (GNSs- HA) on the toxicity of GNSs to Gram positive \*Bacillus tropicus\*. Contrary to previous data, GNSs- HA exhibited greater toxicity than bare GNSs. Multi- omics combined with sensitive bioassays and electrochemical methods demonstrated that bare GNSs disrupted oxidative phosphorylation by causing physical membrane damage. This led to the accumulation of intracellular reactive oxygen species and inhibition of ATP production, subsequently suppressing metabolic processes and ultimately causing bacterial death. Conversely, GNSs- HA directly extracted electrons from bacteria and oxidized biomolecules due to HA- improved electron transfer. This finding suggests that the HA corona does not always mitigate the toxicity of engineered nanoscale pollutants (ENPs), thereby introducing uncertainty over the interaction between the environmental corona and ENPs during ecological risk evaluation. + +## Introduction + +Since its discovery in 2004, graphene nanosheets (GNSs) have been extensively studied for their unique properties that enable multiple applications, including in electronics, catalysis, biomedicine, and environmental fields. \(^{1,2}\) With increasing production and widespread application of GNSs, there is an increased likelihood of GNSs- based waste entering the aquatic environment, raising concerns over potential environmental damage. Previous studies have shown that GNSs exhibit significant toxicity to aquatic organisms through oxidative stress, physical disruption of cell membranes or cell walls, and electron transfer between biological molecules and GNSs. + +Humic acid (HA) is the main component of natural organic matter (NOM) and is ubiquitous in aqueous systems. When GNSs enter the aquatic environment, formation of a HA corona can be expected, which alters the physicochemical properties of GNSs, subsequently impacting their transport, fate, and toxicity. GNSs can be internalized by algae and accumulate intracellularly, interfering with biological processes. \(^{3}\) Notably, the interaction between GNSs and bacteria typically occurs at the nano- bio interface. Previous studies have demonstrated that HA reduces GNSs toxicity to algae, largely by mitigating physical damage through reduced direct contact, leading to significantly less intracellular or extracellular reactive oxygen species (ROS). \(^{4,5}\) However, the impact of HA corona on the bacterial toxicity of GNSs is unknown. + +Unlike algae, which can internalize GNSs, bacterial interactions with GNSs are largely restricted to the cell membrane. In addition to inducing physical damage to the membrane, GNSs can also cause cell membrane dysfunction through chemical reactions such as electron transfer. \(^{6}\) Previous studies have shown that GNSs can serve as strong electron acceptors and extract electrons directly from the electron tranport chain (ETC) in the cytoplasmic membrane, resulting in compromised membrane integrity and reduced ATP production. \(^{6,7}\) While the HA corona typically alleviates the physical membrane damage + +<--- Page Split ---> + +caused by 2D nanomaterials, \(^{8}\) the impact on chemical interactions between GNSs and the bacterial membrane is unclear. + +The electron extraction capacity from the ETC is influenced by the electrical conductivity or band structure of the material in question. Yeung et al. reported that manipulating the electrical conductivity of vanadium dioxide by tungsten doping enhanced antimicrobial activity by facilitating electron extraction from the bacterial respiratory chain. This process induced oxidative stress, disrupted metabolism, and ultimately led to membrane perturbation and leakage of intracellular matter. \(^{9}\) Importantly, HA can modify the band structure of semiconductive materials through physical adsorption. \(^{10, 11}\) GNSs are semimetal materials with a zero bandgap that can be modified by non- covalent functionalization through surface transfer doping. \(^{12}\) Based on this information, we hypothesize that HA can alter the band structure of GNSs and influence electron transfer between GNSs and bacteria. + +In the current study, we evaluated the effect of HA adsorption on the bacterial toxicity of GNSs by exposing GNSs and HA- sorbed GNSs (GNSs- HA) to Gram positive Bacillus tropicus (B. tropicus) at doses of \(1 \text{to} 200 \mu \text{g mL}^{- 1}\) . Measured end points included intracellular ROS, GSH oxidation, bacterial morphology, LDH release, model membrane integrity, oxidative phosphorylation, ATP synthesis, electron transfer, and biomolecule oxidation. In addition, multi- omic techniques were used to investigate the mechanism of HA- enhanced GNSs toxicity to B. tropicus. This study highlights the importance of considering the interaction between nanoparticles and electroactive NOM during ecological risk evaluation. + +## Effects of HA on the bacterial toxicity of GNSs + +Chemical adsorption was found to play a dominant role in the formation of GNSs- HA via adsorption kinetics investigation, suggesting the involvement of electron sharing or electron transfer between HA and GNSs (Figure S1, Supporting Information 2.1). The characterization of morphological and structural changes of GNSs following HA adsorption demonstrated the increased crystalline degree, which can improve conductivity (Figure S2- S7). As the concentration of GNSs increased from \(1 \text{to} 200 \mu \text{g mL}^{- 1}\) , the survival rate decreased from 45- 19.23%. Surprisingly, HA adsorption increased the toxicity of GNSs, except at \(100 \mu \text{g mL}^{- 1}\) . The survival rate after GNSs- HA treatment decreased to 30.33% \(\sim\) 15.8% at 1 to \(100 \mu \text{g mL}^{- 1}\) (Fig. 1a). However, the toxicity did not further increase at \(200 \mu \text{g mL}^{- 1}\) due to the obvious precipitation (Figure S8). The short- term growth curve exhibited that the inhibitory effect of both GNSs and GNSs- HA was largely complete within 30 min, with a greater effect for GNSs- HA (Fig. 1b). When treated with HA alone, the survival rate was 1.58 times higher than the control (Fig. 1c), excluding the effect of HA. These findings contradict the traditional understanding that HA alleviates the toxicity of 2D nanomaterials to algae and other species as reported in the literature. \(^{4, 5, 11, 13 - 15}\) + +Given that NOM in the surface- bound and dissolved states may differentially impact nanotoxicity, \(^{14}\) we investigated the survival rate of B. tropicus treated with GNSs in the presence of dissolved HA (GNSs + + +<--- Page Split ---> + +HA). Interestingly, unlike surface- bound HA, which either aggravated or did not change the toxicity, dissolved HA significantly improved bacterial survival compared to the GNSs group (Fig. 1c). Therefore, our results demonstrated that HA in the surface- bound state enhanced the GNSs toxicity to B. tropicus. + +## Oxidative stress + +Neither GNSs nor GNSs- HA increased the ROS level in B. tropicus after 2 h exposure, although GNSs induced a concentration- dependent increase after 12 h (Fig. 1d and S9). These results suggest that ROS- dependent oxidative stress was not the promoter of GNSs- or GNSs- HA- induced bacterial toxicity. More specifically, within the first 2 h, the antioxidative system of B. tropicus was capable of alleviating excess ROS produced intracellularly. However, when overloaded in the GNSs group, the antioxidative system became overwhelmed, leading to the accumulation of ROS and subsequent bacterial toxicity.16 + +Glutathione (GSH) is an antioxidant to protect cellular components from oxidative stress- induced damage.17 In the presence of free radicals, GSH can generate glutathione disulfide (GSSG), which can react with sulfhydryl (- SH) containing biomolecules, leading to toxic effects.18 In our study, the GSH/GSSG ratio in the experimental groups decreased, and this decrease was more pronounced with GNSs (Fig. 1e). However, the oxidation of GSH to GSSG should have been suppressed with GNSs due to the down- regulation of GSH oxidation- related proteins (Figure S10), indicating that the significant oxidation of GSH is likely a result of greater ROS. In contrast, GNSs- HA might have induced GSH oxidation through a non- ROS- dependent pathway. In the presence of antioxidant melatonin, the survival rate was slightly increased from 19.13–27.11% with GNSs but was unchanged with GNSs- HA (Fig. 1f). These results suggest that ROS played a minor role in the toxicity of GNSs and was not involved in GNSs- HA- induced bacterial toxicity. + +## Membrane damage + +In addition to oxidative stress, membrane damage also plays a significant role in the antibacterial behavior of 2D nanomaterials due to the "nanoknife" effect.19 Both GNSs and GNSs- HA caused significant membrane damage to B. tropicus (Fig. 1g- i and S11). Although sharp edges of the GNSs lamellar structure were passivated by HA (Figure S2), which was expected to alleviate direct physical damage,5 GNSs- HA induced greater membrane damage than did GNSs, with distorted and shriveled patterns evident by SEM. In addition, both GNSs and GNSs- HA induced concentration- dependent LDH release, the exception being GNSs at \(200\mu \mathrm{g}\mathrm{mL}^{- 1}\) (Fig. 1j), due to the reduced interaction from GNSs and bacterial cell agglomeration. Consistent with SEM observations, greater LDH release and more significant alterations in the expression of membrane integrity- related proteins were induced after GNSs- HA treatment (Figure S12), indicating more extensive membrane damage. + +Cell membrane can be disrupted by 2D nanomaterials either through physical damage or by chemical reactions.20- 22 The physical membrane damage was monitored by quartz crystal microbalance with + +<--- Page Split ---> + +dissipation (QCMD). GNSs elicited a more obvious frequency increase (5.86 Hz) compared to GNSs- HA (2.53 Hz) (Fig. 1k,l and S13), demonstrating greater physical damage to the membrane. S- layer proteins (SLPs) are cell wall- related proteins located on the outermost layer of microorganisms that are in direct contact with the external environment and protect bacterial cells against xenobiotics. \(^{23}\) The proteomics data showed that the SLPs of B. tropicus decreased more significantly with GNSs than GNSs- HA (Figure S14), suggesting that HA adsorption reduced the physical interaction- driven membrane damage of GNSs. Therefore, it is clear that chemical oxidation dominates the membrane damage caused by GNSs- HA. + +## Disturbance of oxidative phosphorylation + +The oxidative phosphorylation system plays a crucial role in generating ATP to sustain cellular activity. \(^{24 - 26}\) It was more severely disrupted by GNSs than GNSs- HA (Fig. 2a, S15, and S16). This disruption was primarily manifested through significant down- regulation of complex \(\mathbb{V}\) and \(\mathbb{V}_i\) upregulation of complex III, and interference with complex II and IV upon GNSs exposure (Fig. 2b). Conversely, GNSs- HA disruption was less severe and primarily affected complex II and IV. This impact on oxidative phosphorylation by GNSs was likely due to a disruption of electron transfer between iron- sulfur centers, as indicated in a previous study. \(^{7}\) + +## ATP synthesis + +Disturbance of oxidative phosphorylation by GNSs inhibited ATP synthesis, reducing ATP content compared to the control (Fig. 2c). Conversely, there was no significant change in ATP content with GNSs- HA. Since sufficient energy supply is required to maintain physiological function, ATP deficiency could impair bacterial metabolism. Five differentially abundant proteins (DAPs) associated with metabolism- related pathways were significantly down- regulated with GNSs, while one was up- regulated with GNSs- HA (Figure S17- S19), indicating the inhibitory effect of GNSs on B. tropicus metabolism. Since metabolic processes require ATP and are necessary for bacterial growth and reproduction, \(^{27 - 29}\) the deficiency of ATP induced by GNSs inhibited metabolism and ultimately led to a reduction in bacterial growth. + +Complex V (ATP synthase) is composed of \(\mathsf{F}_0\) and \(\mathsf{F}_1\) units, which act as the proton channel and catalyst for ATP synthesis, respectively. \(^{25}\) The synthesis of each ATP molecule requires \(2 \sim 3\) protons to be transported across the membrane from the intermembrane space to the cytoplasm by the proton channel. \(^{30}\) The down- regulation of complex V in the GNSs group interfered with proton pump function, inhibiting the delivery of protons into the cytoplasm (Fig. 2a,b). As the terminal electron acceptor of aerobic bacteria, intracellular \(\mathsf{O}_2\) is reduced to \(\mathsf{H}_2\mathsf{O}\) at the terminal of the ETC (Eq. 1). With insufficient proton pumping in the cells, incomplete reduction of \(\mathsf{O}_2\) occurred (Eq. 2- 4), resulting in the accumulation of intracellular ROS. \(^{31}\) This likely explains the delayed elevation of intracellular ROS with GNSs (Fig. 1d). + +<--- Page Split ---> + +Conversely, GNSs- HA did not interfere with the expression of complex V (Fig. 2a), which allows the occurrence of four- electron reduction of \(O_{2}\) to water (Eq. 1), thus avoiding the generation of ROS. + +\[\begin{array}{l}{{\frac{1}{2}\mathrm{O}_{2}+2\mathrm{H}^{+}+2\mathrm{e}^{-}\rightarrow\mathrm{H}_{2}\mathrm{O}}}\\ {{}}\\ {{\mathrm{O}^{2}+\mathrm{e}^{-}\rightarrow\mathrm{O}_{2}\cdot\cdot}}\\ {{}}\\ {{\mathrm{O}_{2}+2\mathrm{e}^{-}+2\mathrm{H}^{+}\rightarrow\mathrm{H}_{2}\mathrm{O}_{2}}}\\ {{}}\\ {{\mathrm{O}_{2}+3\mathrm{e}^{-}+3\mathrm{H}^{+}\rightarrow\mathrm{H}_{2}\mathrm{O}+\mathrm{OH}\bullet}}\end{array} \quad (1)\] + +## Electron transfer between B. tropicus and graphenes + +Although it is still unclear why GNSs- HA elicited higher toxicity than GNSs, more significant membrane damage with GNSs- HA from chemical oxidation, as opposed to physical disruption, provides a valuable clue. If an oxidation reaction occurs between B. tropicus and GNSs or GNSs- HA, electron transfer is inevitable and this can be characterized by an electrochemical method.32- 34 The linear sweep voltammetry (LSV) curves, Tafel slopes, Nyquist plots, and cyclic voltammetry (CV) curves demonstrated the electron transfer between GNSs- HA and B. tropicus was more efficient because HA adsorption decreased the electrochemical impedance (Fig. 2d- h, Supporting Information 2.7). Notably, the phenomenon of HA- enhanced electron transfer has been previously observed in microbial fuel cells.35 + +Kelvin probe force microscopy (KPFM) was performed to determine the electron flow direction by measuring the bacterial surface potential before and after GNSs or GNSs- HA treatment. Compared to the control, the surface potential of B. tropicus was enhanced in the presence of both GNSs and GNSs- HA, indicating bacterial cells served as electron donors (Fig. 3a- c). According to the proteomic data, electron transfer flavoprotein subunits were up- regulated in GNSs- treated cells (Figure S20), indicating more endogenous electrons were delivered via extracellular electron transfer (EET). However, more biological electrons were lost in GNSs- HA- treated cells (Fig. 3c). In addition, GNSs and GNSs- HA may further transfer the obtained electrons to extracellular \(O_{2}\) ; the oxygen consumption rate was in the order of GNSs- HA > GNSs > control (Fig. 3d). Since no extracellular ROS was detected (Fig. 3e), the four- electron reduction of \(O_{2}\) to water likely occurred (Eq. 1), indicating the oxidation of bacterial components was not derived from extracellular ROS (Section 2.9). Consequently, there should be other pathways, including direct oxidation of biological components, contributing to the loss of electrons from GNSs- HA- treated B. tropicus, as facilitated by HA- enhanced electron transfer. + +## Oxidation of bacterial components + +<--- Page Split ---> + +As a result of the loss of biological electrons, biomolecules such as lipid membrane components and intracellular macromolecules can be oxidized. After treatment with GNSs, lipid peroxidation as measured by malondialdehyde (MDA) followed a dose- dependent pattern (Fig. 3f), which was proportional to the intracellular ROS level, suggesting excessive ROS might explain GNSs- induced lipid peroxidation. Surprisingly, GNSs- HA induced even stronger lipid peroxidation, despite not leading to an increase in the intracellular or extracellular ROS (Fig. 1d and 3e). Given this, direct oxidation by GNSs- HA through electron transfer could be a plausible mechanism. \(^{33,36}\) Importantly, only GNSs- HA oxidized intracellular DNA, as determined by the level of 8- hydroxy- deoxyguanosine (8- OHdG), which is a marker of DNA oxidative damage (Fig. 3g). \(^{19}\) Several purine metabolism- related genes were significantly altered under GNSs- HA stress (Figure S21), suggesting the disturbance of purine metabolism, which might be due to oxidative damage of guanine in DNA driven by electron transfer rather than ROS. + +## Mechanisms and implications + +HA is commonly known as a detoxifying agent for nanoparticles. However, in this current study, we found surface- bound HA enhanced the toxicity of GNSs to Gram- positive B. tropicus by exerting a different toxicity mechanism (Fig. 4). Specifically, GNSs caused serious physical membrane damage, leading to the disturbance of oxidative phosphorylation that not only induced intracellular ROS accumulation but also inhibited ATP synthesis. This disruption subsequently led to energy deficiency, which inhibited metabolism and caused cell death. Conversely, GNSs- HA induced less physical membrane damage but elicited greater LDH leakage and lipid peroxidation. Additionally, despite the lack of intracellular or extracellular ROS, we observed the oxidation of GSH and DNA, indicating that GNSs- HA directly extracted electrons from B. tropicus and oxidized biomolecules. While both materials can serve as electron acceptors through EET, electron transfer from B. tropicus to GNSs- HA was more efficient than GNSs (Fig. 2d- h). The conduction band (CB) minimum of semiconductor overlapping with the biological redox potential (BRP, - 4.12 ~ - 4.84 eV) allows for electron transfer from biological couples to the CB, leading to the oxidation of biological substances. \(^{37,38}\) The CB minimum of GNSs- HA (Fig. 3h and S22) was within the scope of BRP, enabling spontaneous electron transfer from B. tropicus biomolecules to GNSs- HA. These distinct antibacterial mechanisms are thus responsible for the HA- enhanced toxicity of GNSs to B. tropicus observed in the current study. + +In order to determine if the proposed mechanism is universal, we also conducted the toxicity evaluations on Gram- negative Escherichia coli k12 (E. coli k12), Gram- positive staphylococcus aureus (S. aureus) and Bacillus subtilis (B. subtilis). Consistent with the findings in B. tropicus, B. subtilis was also found to be more susceptible to GNSs- HA, while the toxicity of GNSs to E. coli k12 and S. aureus was reduced after HA adsorption (Figure S23). Based on these results, we speculate that both bacterial membrane structure and morphology may contribute to the different responses. In fact, previous studies have indicated that Gram- positive bacteria are more susceptible to 2D nanomaterials than Gram- negative bacteria, possibly due to differences in bacterial membrane structure. \(^{39 - 43}\) The outer membrane of Gram- negative E. coli provides protection against physical damage caused by the sharp edges of + +<--- Page Split ---> + +GNSs. \(^{44}\) Despite the absence of outer membrane protection in Gram- positive S. aureus, their smaller spherical morphology allows utilization of GNSs as a shelter, reducing the exposure density of sharp edges to the membrane relative to rod- shaped cells. \(^{45}\) Additionally, the cell membrane, which has a high insulation with the conductivity of \(10^{- 7} \text{S m}^{- 1}\) , \(^{46}\) limits electron outflow from the conductive interior of E. coli and S. aureus to GNSs- HA, even with the overlapping of the CB minimum with BRP. However, for rod- shaped Gram- positive bacteria such as B. tropicus and B. subtillis, GNSs- HA disrupts membrane integrity, which enables abnormal electron transfer that leads to the oxidation of biological substances. Overall, the combination of physical membrane perturbation and disturbed electron balance contributes to the altered bacterial toxicity mechanism of GNSs by HA adsorption. + +## Conclusions + +In the current work, the effect of HA on bacterial toxicity of GNSs is strain dependent. Specifically, although GNSs- HA exhibited lower toxicity to E. coli k12 and S. aureus than bare GNSs, they were more toxic to B. tropicus and B. subtillis under the same exposure scenario, challenging the conventional view that HA mitigates ENPs toxicity. B. tropicus was used as a model to investigate these phenomena and the data showed that the HA corona altered the bacterial toxicity mechanism of GNSs. The physical damage induced by GNSs to the cell membrane disrupted oxidative phosphorylation, leading to ROS accumulation and ATP synthesis inhibition and ultimately bacterial death due to energy deficiency. Conversely, the band structure of GNSs was modulated by the HA corona, allowing GNSs- HA to extract electrons not only through EET, but also via direct biomolecule oxidation. This demonstration of enhanced ENPs' bacterial toxicity with HA attachment highlights the importance of understanding the mechanisms of eccorona activity when trying to accurately evaluate the risk of ENPs in the environment. + +## Methods + +## Bacterial toxicity + +The bacterial toxicity of GNSs, GNSs- HA, and GNSs + HA was evaluated using optical density (OD) values and the colony counting method. Briefly, \(1.0 \text{mL}\) of GNSs and GNSs- HA suspensions (0, 10, 100, 500, 1000, and \(2000 \mu \text{g mL}^{- 1}\) ), \(1.0 \text{mL}\) of bacterial suspension ( \(10^{5} \text{CFU mL}^{- 1}\) ), and \(8.0 \text{mL}\) of fresh Luria Bertani (LB) broth were mixed together, obtaining the final material concentrations at 0, 1, 10, 50, 100, and \(200 \mu \text{g mL}^{- 1}\) . For the GNSs + HA group, \(1.0 \text{mL}\) of bacterial suspension ( \(10^{5} \text{CFU mL}^{- 1}\) ), \(1.0 \text{mL}\) of HA ( \(12 \mu \text{g mL}^{- 1}\) ), \(1.0 \text{mg}\) of GNSs powder, and \(8.0 \text{mL}\) of fresh LB broth were mixed together. The mixtures were incubated at \(37^{\circ} \text{C}\) and shaken at 150 rpm. The OD values were determined every 5 min at 600 nm on a microplate reader (BMG LABTECH, Germany) for 30 min to obtain the short- term growth curve. As a comparison, the effect of dissolved HA ( \(12 \mu \text{g mL}^{- 1}\) ) on bacterial growth was evaluated following the same procedure. After 12 h incubation, the bacterial suspensions were diluted with sterilized water to a dilution series of \(10^{- 4}\) , \(10^{- 5}\) , and \(10^{- 6}\) . Then \(100 \mu \text{L}\) of each dilution was spread + +<--- Page Split ---> + +onto a LB agar plate, which was incubated at \(37^{\circ}C\) overnight. The colony forming units (CFUs) of all plates were counted and the survival rate \((\%)\) was calculated by comparing the CFUs of treatment groups with that of controls. + +Melatonin, as an antioxidant, was used to explore the role of ROS in the bacterial toxicity. Briefly, \(0.5 \mathrm{mL}\) of GNSs and GNSs- HA suspensions, \(1.0 \mathrm{mL}\) of \(B\) . tropicus suspension ( \(10^{5} \mathrm{CFU} \mathrm{mL}^{- 1}\) ), \(8.0 \mathrm{mL}\) of fresh LB broth, and \(0.5 \mathrm{mL}\) melatonin solution ( \(20 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ) were mixed together to obtain the final concentration of GNSs and GNSs- HA at \(100 \mu \mathrm{g} \mathrm{mL}^{- 1}\) . The mixtures were shaken at \(120 \mathrm{rpm}\) and incubated at \(37^{\circ}C\) for \(12 \mathrm{h}\) . The survival rate was evaluated using the colony counting method. + +## Physical membrane damage + +The interaction of GNSs and GNSs- HA with the model cell membrane was investigated by QCMD. Before the experiment, the gold- plated crystal sensor was cleaned by immersion in a mixed solution ( \(\mathrm{H}_2\mathrm{O}\) : \(30\%\) \(\mathrm{H}_2\mathrm{O}_2\) : \(25\%\) \(\mathrm{NH}_3 = 5:1:1\) ) for \(5 \mathrm{min}\) at \(75^{\circ}C\) , and then rinsed with a large amount of deionized water before drying under nitrogen. The sensor was then oxidized by UV ozone exposure for \(20 \mathrm{min}\) , followed by rinsing with Millipore water and drying with nitrogen gas. + +To prepare the model membrane, chloroform solution of 1,2- dioleoyl- sn- glycero- 3- phosphocholine (DOPC, \(25 \mathrm{mg} \mathrm{mL}^{- 1}\) , \(0.2 \mathrm{mL}\) ) was added into a conical flask and dried under nitrogen to form a DOPC film, which was vacuum dried for more than \(4 \mathrm{h}\) to remove residual chloroform. Then \(5 \mathrm{mL}\) HEPES buffer solution was added under continuous magnetic stirring for \(30 \mathrm{min}\) to form DOPC vesicles. The DOPC solution was passed through a polycarbonate film (Whatman) back and forth for more than \(15 \mathrm{times}\) by a micro lipid extruder (Avanti Polar Lipids Inc.). A single layer of vesicles with a hydrodynamic diameter of \(192 - 194 \mathrm{nm}\) was obtained. The vesicle suspension was sealed with nitrogen in a glass bottle at \(4^{\circ}C\) and used within 4 days after preparation. + +During the QCMD test, the temperature in the flow cell chamber was kept at \(37^{\circ}C\) , and the liquid flow rate was \(150 \mu \mathrm{L} \mathrm{min}^{- 1}\) . The frequency (representing the deposited mass) and energy dissipation (representing the viscoelastic properties of the deposited layer) were detected. HEPES buffer was first introduced to the system until the signal was stable. Afterwards, DOPC vesicles were introduced to form a supporting vesicle layer on the sensor surface. Then HEPES buffer was injected again to remove the unadsorbed DOPC vesicles. GNSs and GNSs- HA were then suspended in HEPES ( \(50 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ) to initiate the interaction process, followed by rinsing with HEPES to remove the unadsorbed GNSs and GNSs- HA. At the end of each experiment, \(1\%\) Triton X- 100 (a membrane solubilizer) was used to remove the adsorbed vesicles. + +## Electrochemical measurement + +GNSs or GNSs- HA powders were mixed with a polytetrafluoroethylene solution ( \(2 \mathrm{wt}\%\) ), and the mixture was pasted on the surface of carbon fiber paper (surface area of \(2.5 \mathrm{cm}^2\) ) with a loading content of \(0.5 \mathrm{mg} \mathrm{cm}^{- 2}\) . After drying at \(100^{\circ}C\) for \(3 \mathrm{h}\) in the oven, the final electrode material was obtained. A bacterial + +<--- Page Split ---> + +suspension (100 mL) with an initial concentration of \(10^{5} \text{CFU mL}^{- 1}\) was incubated at \(37^{\circ} \text{C}\) for \(12 \text{h}\) before use. \(B\) . tropicus cells were harvested by centrifugation at 10000 rpm for 3 min and washed three times with anaerobic PBS (0.1 M, pH 7.0). The \(B\) . tropicus cells were resuspended in PBS and diluted to \(\sim 10^{9} \text{CFU mL}^{- 1}\) . The bacterial suspension was deoxygenated by purging with nitrogen for 30 min before each test. Electrochemical measurements were conducted by an electrochemical workstation (CHI660E, CH Instruments, Inc., Shanghai, China). A three- electrode cell was setup, consisting of a GNSs- or GNSs- HA- loaded carbon fiber paper working electrode, a Pt wire counter electrode, and an Ag/AgCl reference electrode. CV measurements were carried out at a scan rate of \(100 \text{mV s}^{- 1}\) and over a potential range from \(- 0.6\) to \(0.1 \text{V}\) . Electrochemical impedance spectra (EIS) were obtained over frequencies ranging from \(1000 \text{kHz}\) to \(1 \text{Hz}\) , with an amplitude of \(5 \text{mV}\) . LSV was obtained at a scan rate of \(50 \text{mV s}^{- 1}\) . + +## Local surface potential measurement by KPFM + +After \(B\) . tropicus were incubated with GNSs and GNSs- HA ( \(100 \mu \text{g mL}^{- 1}\) ) for \(60 \text{min}\) at \(37^{\circ} \text{C}\) , the mixture was centrifuged at \(800 \text{rpm}\) for 5 min to remove the agglomerates, followed by dropping the suspension onto a monocrystalline silicon wafer. As soon as the sample was dried under ambient conditions, the silicon wafer with fixed samples was finally adhered to a small piece ( \(1 \text{cm} \times 1 \text{cm}\) ) of conductive double- sided tape, which was placed on a grounded microscope stage for KPFM measurement. Three bacterial cells were analyzed for each sample in order to verify that the microscopy images represented the average potential of the samples. KPFM was conducted in the tapping mode on the Agilent 5500 AFM. Platinum (Pt)/iridium (Ir)- coated silicon cantilever probes (Bruker, SCM- PIT- V2) were used as the conductive probes with a force constant of approximately \(3 \text{N m}^{- 1}\) and a nominal resonance frequency of \(75 \text{kHz}\) . + +## Declarations + +## Author contributions + +X.Z. conceived the idea and designed the research. J. Z. performed most of the experiments. Z.X., S.Z., and J.Y. analyzed the omics data. W.T., and Y.X. interpreted the data. X.Z. and Q.Z. contributed to writing the manuscript. J.C.W., F.L., B.X. and F.W. modified the manuscript. + +## Acknowledgements + +This work was supported by the National Natural Science Foundation of China (No. 42022056, 42192574, 42277423, 42077394, 22176196), GDAS' Project of Science and Technology Development (2022GDASZH- 2022010105, 2020GDASYL- 20200101002), GDAS' Project of Science and Technology Development (22022GDASZH- 2022020402- 01), GDAS' Project of Science and Technology Development (2022GDASZH- 2022020402- 02), and Guangdong Foundation for Program of Science and Technology Research (2020B1212060048). + +<--- Page Split ---> + +## References + +1. NOVOSELOV, K. S. et al. FIRSOV Electric Field Effect in Atomically Thin Carbon Films. Science 306, 666-669 (2004). +2. Khan, K. et al. Recent developments in emerging two-dimensional materials and their applications. J. Mater. Chem. C 8, 387-440 (2020). +3. Su, Y. et al. Green Algae as Carriers Enhance the Bioavailability of (14)C-Labeled Few-Layer Graphene to Freshwater Snails. Environ. Sci. Technol. 52, 1591-1601 (2018). +4. Chen, W. et al. Black Phosphorus Nanosheets as a Neuroprotective Nanomedicine for Neurodegenerative Disorder Therapy. Adv. Mater. 30, 1703458 (2018). +5. Zhao, J. et al. Humic acid mitigated toxicity of graphene-family materials to algae through reducing oxidative stress and heteroaggregation. Environ. Sci.-Nano 6, 1909-1920 (2019). +6. Jannesari, M., Akhavan, O., Madaah Hosseini, H.R. & Bakhshi, B. Graphene/CuO(2) Nanoshuttles with Controllable Release of Oxygen Nanobubbles Promoting Interruption of Bacterial Respiration. ACS Appl. Mater. Interfaces 12, 35813-35825 (2020). +7. Zhou, H. et al. The inhibition of migration and invasion of cancer cells by graphene via the impairment of mitochondrial respiration. Biomaterials 35, 1597-1607 (2014). +8. Xu, L. et al. The Crucial Role of Environmental Coronas in Determining the Biological Effects of Engineered Nanomaterials. Small 16, e2003691 (2020). +9. Li, J. et al. Temperature-responsive tungsten doped vanadium dioxide thin film starves bacteria to death. Mater. Today 22, 35-49 (2019). +10. He, X. et al. Assessing the effects of surface-bound humic acid on the phototoxicity of anatase and rutile TiO(2) nanoparticles in vitro. J. Environ. Sci.-China 42, 50-60 (2016). +11. Zou, W., Zhou, Q., Zhang, X. & Hu, X. Environmental Transformations and Algal Toxicity of Single-Layer Molybdenum Disulfide Regulated by Humic Acid. Environ. Sci. Technol. 52, 2638-2648 (2018). +12. Mao, H.Y. et al. Manipulating the electronic and chemical properties of graphene via molecular functionalization. Prog. Surf. Sci. 88, 132-159 (2013). +13. Shang, E. et al. Relative importance of humic and fulvic acid on ROS generation, dissolution, and toxicity of sulfide nanoparticles. Water Res. 124, 595-604 (2017). +14. Zhang, X., Sui, M., Yan, X., Huang, T. & Yuan, Z. Mitigation in the toxicity of graphene oxide nanosheets towards Escherichia coli in the presence of humic acid. Environ. Sci.-Proc. Imp. 18, 744-750 (2016). +15. Zhang, Y. et al. The effects of humic acid on the toxicity of graphene oxide to Scenedesmus obliquus and Daphnia magna. Sci. Total Environ. 649, 163-171 (2019). +16. Hong, Y., Zeng, J., Wang, X., Drlica, K. & Zhao, X. Post-stress bacterial cell death mediated by reactive oxygen species. Proc. Natl. Acad. Sci. USA 116, 10064-10071 (2019). +17. Shaobin Liu, T.H.Z., Mario Hofmann, Ehdi Burcombe, Jun Wei, Rongrong Jiang, Jing Kong, and Yuan Chen Antibacterial Activity of Graphite, Graphite Oxide, Graphene Oxide, and Reduced Graphene + +<--- Page Split ---> + +Oxide: Membrane and Oxidative Stress. ACS Nano 5, 6971- 6980 (2011). + +18. Oktyabrsky, G.V.S.O.N. Glutathione in Bacteria. Biochemistry-Moscow 70, 1199 (2005). + +19. Yang, S.P. et al. Influence of humic acid on titanium dioxide nanoparticle toxicity to developing zebrafish. Environ. Sci. Technol. 47, 4718-4725 (2013). + +20. Feng, Y. et al. Interaction of Graphitic Carbon Nitride with Cell Membranes: Probing Phospholipid Extraction and Lipid Bilayer Destruction. Environ. Sci. Technol. 56, 17663-17673 (2022). + +21. Hu, X., Ouyang, S., Mu, L., An, J. & Zhou, Q. Effects of Graphene Oxide and Oxidized Carbon Nanotubes on the Cellular Division, Microstructure, Uptake, Oxidative Stress, and Metabolic Profiles. Environ. Sci. Technol. 49, 10825-10833 (2015). + +22. Zhou, Q. & Hu, X. Systemic Stress and Recovery Patterns of Rice Roots in Response to Graphene Oxide Nanosheets. Environ. Sci. Technol. 51, 2022-2030 (2017). + +23. Gerbino, E., Carasi, P., Mobili, P., Serradell, M.A. & Gomez-Zavaglia, A. Role of S-layer proteins in bacteria. World J. Microb. Biot. 31, 1877-1887 (2015). + +24. BA Haddock, C.J. Bacterial respiration. Bacteriological Rev. 41 (1977). + +25. Llorente-Garcia, I. et al. Single-molecule in vivo imaging of bacterial respiratory complexes indicates delocalized oxidative phosphorylation. Biochim. Biophys. Acta 1837, 811-824 (2014). + +26. Trounce, I. Genetic control of oxidative phosphorylation and experimental models of defects. Hum. Reprod. 15, 18-27 (2000). + +27. Cui, Y. & Qu, X. Genetic mechanisms of prebiotic carbohydrate metabolism in lactic acid bacteria: Emphasis on Lacticaseibacillus casei and Lacticaseibacillus paracasei as flexible, diverse and outstanding prebiotic carbohydrate starters. Trends Food Sci. Tech. 115, 486-499 (2021). + +28. Komosinska-Vassev, K. et al. Alterations of glycosaminoglycan metabolism in the development of diabetic complications in relation to metabolic control. Clin. Chem. Lab. Med. 43, 924-929 (2005). + +29. Sobhanifar, S., King, D.T. & Strynadka, N.C.J. Fortifying the wall: synthesis, regulation and degradation of bacterial peptidoglycan. Curr. Opin. Struct. Biol. 23, 695-703 (2013). + +30. Kaila, V.R.I. & Wikstrom, M. Architecture of bacterial respiratory chains. Nat. Rev. Microbiol. 19, 319-330 (2021). + +31. Tan, J. et al. A facile and universal strategy to endow implant materials with antibacterial ability via alkalinity disturbing bacterial respiration. Biomater. Sci. 8, 1815-1829 (2020). + +32. Chen, S. et al. MoS2 Nanosheets-Cyanobacteria Interaction: Reprogrammed Carbon and Nitrogen Metabolism. ACS Nano 15, 16344-16356 (2021). + +33. Li, J. et al. Antibacterial activity of large-area monolayer graphene film manipulated by charge transfer. Sci. Rep. 4, 4359 (2014). + +34. Zhou, S., Lin, M., Zhuang, Z., Liu, P. & Chen, Z. Biosynthetic graphene enhanced extracellular electron transfer for high performance anode in microbial fuel cell. Chemosphere 232, 396-402 (2019). + +35. Huang, B. et al. Ferroferric oxide loads humic acid doped anode accelerate electron transfer process in anodic chamber of bioelectrochemical system. J. Electroanal. Chem. 851, 113464 (2019). + +<--- Page Split ---> + +36. Liu S, Z.T.H., Hofmann M Antibacterial activity of graphite, graphite oxide, graphene oxide, and reduced graphene oxide: membrane and oxidative stress. ACS Nano 5, 6971-6980. +37. Li, J. et al. Band Gap Engineering of Titania Film through Cobalt Regulation for Oxidative Damage of Bacterial Respiration and Viability. ACS Appl. Mater. Interfaces 9, 27475-27490 (2017). +38. Qu, G. et al. Property-Activity Relationship of Black Phosphorus at the Nano-Bio Interface: From Molecules to Organisms. Chem. Rev. 120, 2288-2346 (2020). +39. Deokar, A.R., Lin, L.Y., Chang, C.C. & Ling, Y.C. Single-walled carbon nanotube coated antibacterial paper: preparation and mechanistic study. J. Mater. Chem. B 1, 2639-2646 (2013). +40. Pulingam, T. et al. Graphene oxide exhibits differential mechanistic action towards Gram-positive and Gram-negative bacteria. Colloids Surf. B Biointerfaces 181, 6-15 (2019). +41. Wang, M. et al. Interaction with teichoic acids contributes to highly effective antibacterial activity of graphene oxide on Gram-positive bacteria. J. Hazard. Mater. 412, 125333 (2021). +42. Xiong, Z. et al. Transcriptome Analysis Reveals the Growth Promotion Mechanism of Enteropathogenic Escherichia coli Induced by Black Phosphorus Nanosheets. ACS Nano 17, 3574-3586 (2023). +43. Xiong, Z. et al. Bacterial toxicity of exfoliated black phosphorus nanosheets. Ecotoxicol. Environ. Saf. 161, 507-514 (2018). +44. Ghaderi, O.A.E. Toxicity of Graphene and Graphene Oxide Nanowalls Against Bacteria. ACS Nano 4, 5731-5736. +45. Vy T. H. Pham, V.K.T., Matthew D. J. Quinn, Shannon M. Notley, Yachong Guo, Vladimir A. Baulin, Mohammad Al Kobaisi, Russell J. Crawford, and Elena P. Ivanova Graphene Induces Formation of Pores That Kill Spherical and Rod-Shaped Bacteria. ACS Nano 9, 8458-8467 (2015). +46. Yang, L. Electrical impedance spectroscopy for detection of bacterial cells in suspensions using interdigitated microelectrodes. Talanta 74, 1621-1629 (2008). + +## Figures + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +(a) Survival rate of \(B\) . tropicus after treatment with GNSs and GNSs-HA for \(12 \text{h}\) ; (b) Effects of GNSs and GNSs-HA on the OD value of \(B\) . tropicus within \(30 \text{min}\) ; (c) Survival rate of \(B\) . tropicus after treated with GNSs, GNSs-HA, and GNSs+HA ( \(100 \mu \text{g mL}^{-1}\) ); (d) Intracellular ROS level after treatment with GNSs and GNSs-HA for \(2 \text{(light color)}\) and \(12 \text{h}\) (deep color); (e) GSH/GSSG ratio in \(B\) . tropicus after treatment with GNSs and GNSs-HA as compared to the value without any treatment; (f) Survival rate of \(B\) . tropicus after treated with \(100 \mu \text{g mL}^{-1}\) of GNSs and GNSs-HA in the presence and absence of melatonin; Morphology of \(B\) . tropicus without any treatment (g) or after GNSs (h) and GNSs-HA (i) treatments; (j) LDH activity of \(B\) . tropicus after treated with different concentrations of GNSs and GNSs-HA; Frequency variations of the + +<--- Page Split ---> + +DOPC vesicles after introducing \(100 \mu \mathrm{g mL}^{- 1}\) of GNSs (k) and GNSs- HA (l). Error bars represent standard deviation from triple measurements. + +![](images/Figure_2.jpg) + +
Figure 2
+ +(a) Changes in ATP synthase related proteins after GNSs or GNSs-HA treatments; (b) Schematic diagram of respiratory chain changes in the GNSs treatment group; (c) ATP content after treated with GNSs and GNSs-HA (100 \(\mu \mathrm{g mL}^{-1}\) ); Electron transfer between B. tropicus and GNSs or GNSs-HA represented by LSV curves (d), Tafel slopes (e), Nyquist plots (f), and CV curves (g, h). Error bars represent standard deviation from triple measurements. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +KPFM images of \(B\) . tropicus without any treatment (a), or treated with GNSs (b) and GNSs- HA (c); (d) Oxygen consumption rate of \(B\) . tropicus treated with GNSs and GNSs- HA within \(30 \text{min}\) ; (e) Extracellular ROS level in the bacterial culture medium after treated with GNSs, GNSs- HA, and \(H_2O_2\) ; MDA (f) and 8- OHdG content (g) in \(B\) . tropicus after exposure to GNSs and GNSs- HA; (h) Schematic illustration of the possible electron transfer pathway between \(B\) . tropicus and the materials. Error bars represent standard deviation from triple measurements. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Schematic illustration of different mechanisms of bacterial toxicity from GNSs or GNSs- HA exposure. GNSs induced physical damage to the cell membrane, which disrupted oxidative phosphorylation, leading to ROS accumulation and energy deficiency; GNSs- HA directly extracted electrons from \(B\) . tropicus across the mildly destroyed membrane, leading to the oxidation of biomolecules. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Supportinginformation2023.08.25.docx + +<--- Page Split ---> diff --git a/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca_det.mmd b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..fa08cb13e12f3039eb4976bade0e75fc19a09992 --- /dev/null +++ b/preprint/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca/preprint__126ed42b3fd6d9a339d2cd64a17c61348e2a1fa4df128380a49ace31df8377ca_det.mmd @@ -0,0 +1,387 @@ +<|ref|>title<|/ref|><|det|>[[42, 108, 848, 208]]<|/det|> +# Mechanistic evaluation of enhanced graphene toxicity to Bacillus induced by humic acid adsorption + +<|ref|>text<|/ref|><|det|>[[44, 230, 257, 275]]<|/det|> +Qing Zhao zhaoqing@iae.ac.cn + +<|ref|>text<|/ref|><|det|>[[44, 302, 780, 323]]<|/det|> +Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 328, 568, 369]]<|/det|> +Xuejiao Zhang Institute of Applied Ecology, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 374, 568, 415]]<|/det|> +Jin Zeng Institute of Applied Ecology, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 420, 477, 460]]<|/det|> +Jason White The Connecticut Agricultural Experiment Station + +<|ref|>text<|/ref|><|det|>[[44, 466, 352, 507]]<|/det|> +Fangbai Li Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 512, 780, 553]]<|/det|> +Zhiqiang Xiong Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 558, 568, 599]]<|/det|> +Siyu Zhang Institute of Applied Ecology, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 604, 780, 645]]<|/det|> +Yuze Xu Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 650, 780, 691]]<|/det|> +Jingjing Yang Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 696, 780, 737]]<|/det|> +Weihao Tang Institute of Eco- environmental and Soil Sciences, Guangdong Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 742, 780, 784]]<|/det|> +Fengchang Wu Chinese Research Academy of Environmental Sciences + +<|ref|>text<|/ref|><|det|>[[44, 789, 745, 831]]<|/det|> +Baoshan Xing University of Massachusetts Amherst https://orcid.org/0000- 0003- 2028- 1295 + +<|ref|>text<|/ref|><|det|>[[44, 873, 102, 890]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 910, 135, 928]]<|/det|> +Keywords: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 46, 353, 65]]<|/det|> +Posted Date: September 21st, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 84, 475, 103]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3294178/v1 + +<|ref|>text<|/ref|><|det|>[[42, 120, 916, 164]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 182, 535, 201]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 236, 933, 280]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 3rd, 2025. See the published version at https://doi.org/10.1038/s41467- 024- 55270- 2. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 158, 68]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[40, 82, 945, 354]]<|/det|> +AbstractThe extensive application of graphene nanosheets (GNSs) has raised concerns over risks to sensitive species in the aquatic environment. The humic acid (HA) corona is traditionally considered to reduce GNSs toxicity. Here, we evaluated the effect of sorbed HA (GNSs- HA) on the toxicity of GNSs to Gram positive \*Bacillus tropicus\*. Contrary to previous data, GNSs- HA exhibited greater toxicity than bare GNSs. Multi- omics combined with sensitive bioassays and electrochemical methods demonstrated that bare GNSs disrupted oxidative phosphorylation by causing physical membrane damage. This led to the accumulation of intracellular reactive oxygen species and inhibition of ATP production, subsequently suppressing metabolic processes and ultimately causing bacterial death. Conversely, GNSs- HA directly extracted electrons from bacteria and oxidized biomolecules due to HA- improved electron transfer. This finding suggests that the HA corona does not always mitigate the toxicity of engineered nanoscale pollutants (ENPs), thereby introducing uncertainty over the interaction between the environmental corona and ENPs during ecological risk evaluation. + +<|ref|>sub_title<|/ref|><|det|>[[44, 377, 204, 403]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[41, 415, 943, 577]]<|/det|> +Since its discovery in 2004, graphene nanosheets (GNSs) have been extensively studied for their unique properties that enable multiple applications, including in electronics, catalysis, biomedicine, and environmental fields. \(^{1,2}\) With increasing production and widespread application of GNSs, there is an increased likelihood of GNSs- based waste entering the aquatic environment, raising concerns over potential environmental damage. Previous studies have shown that GNSs exhibit significant toxicity to aquatic organisms through oxidative stress, physical disruption of cell membranes or cell walls, and electron transfer between biological molecules and GNSs. + +<|ref|>text<|/ref|><|det|>[[41, 592, 950, 800]]<|/det|> +Humic acid (HA) is the main component of natural organic matter (NOM) and is ubiquitous in aqueous systems. When GNSs enter the aquatic environment, formation of a HA corona can be expected, which alters the physicochemical properties of GNSs, subsequently impacting their transport, fate, and toxicity. GNSs can be internalized by algae and accumulate intracellularly, interfering with biological processes. \(^{3}\) Notably, the interaction between GNSs and bacteria typically occurs at the nano- bio interface. Previous studies have demonstrated that HA reduces GNSs toxicity to algae, largely by mitigating physical damage through reduced direct contact, leading to significantly less intracellular or extracellular reactive oxygen species (ROS). \(^{4,5}\) However, the impact of HA corona on the bacterial toxicity of GNSs is unknown. + +<|ref|>text<|/ref|><|det|>[[41, 816, 950, 956]]<|/det|> +Unlike algae, which can internalize GNSs, bacterial interactions with GNSs are largely restricted to the cell membrane. In addition to inducing physical damage to the membrane, GNSs can also cause cell membrane dysfunction through chemical reactions such as electron transfer. \(^{6}\) Previous studies have shown that GNSs can serve as strong electron acceptors and extract electrons directly from the electron tranport chain (ETC) in the cytoplasmic membrane, resulting in compromised membrane integrity and reduced ATP production. \(^{6,7}\) While the HA corona typically alleviates the physical membrane damage + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 46, 905, 90]]<|/det|> +caused by 2D nanomaterials, \(^{8}\) the impact on chemical interactions between GNSs and the bacterial membrane is unclear. + +<|ref|>text<|/ref|><|det|>[[41, 106, 955, 316]]<|/det|> +The electron extraction capacity from the ETC is influenced by the electrical conductivity or band structure of the material in question. Yeung et al. reported that manipulating the electrical conductivity of vanadium dioxide by tungsten doping enhanced antimicrobial activity by facilitating electron extraction from the bacterial respiratory chain. This process induced oxidative stress, disrupted metabolism, and ultimately led to membrane perturbation and leakage of intracellular matter. \(^{9}\) Importantly, HA can modify the band structure of semiconductive materials through physical adsorption. \(^{10, 11}\) GNSs are semimetal materials with a zero bandgap that can be modified by non- covalent functionalization through surface transfer doping. \(^{12}\) Based on this information, we hypothesize that HA can alter the band structure of GNSs and influence electron transfer between GNSs and bacteria. + +<|ref|>text<|/ref|><|det|>[[41, 333, 955, 517]]<|/det|> +In the current study, we evaluated the effect of HA adsorption on the bacterial toxicity of GNSs by exposing GNSs and HA- sorbed GNSs (GNSs- HA) to Gram positive Bacillus tropicus (B. tropicus) at doses of \(1 \text{to} 200 \mu \text{g mL}^{- 1}\) . Measured end points included intracellular ROS, GSH oxidation, bacterial morphology, LDH release, model membrane integrity, oxidative phosphorylation, ATP synthesis, electron transfer, and biomolecule oxidation. In addition, multi- omic techniques were used to investigate the mechanism of HA- enhanced GNSs toxicity to B. tropicus. This study highlights the importance of considering the interaction between nanoparticles and electroactive NOM during ecological risk evaluation. + +<|ref|>sub_title<|/ref|><|det|>[[44, 545, 646, 573]]<|/det|> +## Effects of HA on the bacterial toxicity of GNSs + +<|ref|>text<|/ref|><|det|>[[40, 588, 950, 892]]<|/det|> +Chemical adsorption was found to play a dominant role in the formation of GNSs- HA via adsorption kinetics investigation, suggesting the involvement of electron sharing or electron transfer between HA and GNSs (Figure S1, Supporting Information 2.1). The characterization of morphological and structural changes of GNSs following HA adsorption demonstrated the increased crystalline degree, which can improve conductivity (Figure S2- S7). As the concentration of GNSs increased from \(1 \text{to} 200 \mu \text{g mL}^{- 1}\) , the survival rate decreased from 45- 19.23%. Surprisingly, HA adsorption increased the toxicity of GNSs, except at \(100 \mu \text{g mL}^{- 1}\) . The survival rate after GNSs- HA treatment decreased to 30.33% \(\sim\) 15.8% at 1 to \(100 \mu \text{g mL}^{- 1}\) (Fig. 1a). However, the toxicity did not further increase at \(200 \mu \text{g mL}^{- 1}\) due to the obvious precipitation (Figure S8). The short- term growth curve exhibited that the inhibitory effect of both GNSs and GNSs- HA was largely complete within 30 min, with a greater effect for GNSs- HA (Fig. 1b). When treated with HA alone, the survival rate was 1.58 times higher than the control (Fig. 1c), excluding the effect of HA. These findings contradict the traditional understanding that HA alleviates the toxicity of 2D nanomaterials to algae and other species as reported in the literature. \(^{4, 5, 11, 13 - 15}\) + +<|ref|>text<|/ref|><|det|>[[42, 909, 935, 954]]<|/det|> +Given that NOM in the surface- bound and dissolved states may differentially impact nanotoxicity, \(^{14}\) we investigated the survival rate of B. tropicus treated with GNSs in the presence of dissolved HA (GNSs + + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 940, 111]]<|/det|> +HA). Interestingly, unlike surface- bound HA, which either aggravated or did not change the toxicity, dissolved HA significantly improved bacterial survival compared to the GNSs group (Fig. 1c). Therefore, our results demonstrated that HA in the surface- bound state enhanced the GNSs toxicity to B. tropicus. + +<|ref|>sub_title<|/ref|><|det|>[[44, 141, 255, 168]]<|/det|> +## Oxidative stress + +<|ref|>text<|/ref|><|det|>[[41, 184, 950, 321]]<|/det|> +Neither GNSs nor GNSs- HA increased the ROS level in B. tropicus after 2 h exposure, although GNSs induced a concentration- dependent increase after 12 h (Fig. 1d and S9). These results suggest that ROS- dependent oxidative stress was not the promoter of GNSs- or GNSs- HA- induced bacterial toxicity. More specifically, within the first 2 h, the antioxidative system of B. tropicus was capable of alleviating excess ROS produced intracellularly. However, when overloaded in the GNSs group, the antioxidative system became overwhelmed, leading to the accumulation of ROS and subsequent bacterial toxicity.16 + +<|ref|>text<|/ref|><|det|>[[40, 336, 950, 591]]<|/det|> +Glutathione (GSH) is an antioxidant to protect cellular components from oxidative stress- induced damage.17 In the presence of free radicals, GSH can generate glutathione disulfide (GSSG), which can react with sulfhydryl (- SH) containing biomolecules, leading to toxic effects.18 In our study, the GSH/GSSG ratio in the experimental groups decreased, and this decrease was more pronounced with GNSs (Fig. 1e). However, the oxidation of GSH to GSSG should have been suppressed with GNSs due to the down- regulation of GSH oxidation- related proteins (Figure S10), indicating that the significant oxidation of GSH is likely a result of greater ROS. In contrast, GNSs- HA might have induced GSH oxidation through a non- ROS- dependent pathway. In the presence of antioxidant melatonin, the survival rate was slightly increased from 19.13–27.11% with GNSs but was unchanged with GNSs- HA (Fig. 1f). These results suggest that ROS played a minor role in the toxicity of GNSs and was not involved in GNSs- HA- induced bacterial toxicity. + +<|ref|>sub_title<|/ref|><|det|>[[44, 620, 300, 647]]<|/det|> +## Membrane damage + +<|ref|>text<|/ref|><|det|>[[41, 662, 944, 896]]<|/det|> +In addition to oxidative stress, membrane damage also plays a significant role in the antibacterial behavior of 2D nanomaterials due to the "nanoknife" effect.19 Both GNSs and GNSs- HA caused significant membrane damage to B. tropicus (Fig. 1g- i and S11). Although sharp edges of the GNSs lamellar structure were passivated by HA (Figure S2), which was expected to alleviate direct physical damage,5 GNSs- HA induced greater membrane damage than did GNSs, with distorted and shriveled patterns evident by SEM. In addition, both GNSs and GNSs- HA induced concentration- dependent LDH release, the exception being GNSs at \(200\mu \mathrm{g}\mathrm{mL}^{- 1}\) (Fig. 1j), due to the reduced interaction from GNSs and bacterial cell agglomeration. Consistent with SEM observations, greater LDH release and more significant alterations in the expression of membrane integrity- related proteins were induced after GNSs- HA treatment (Figure S12), indicating more extensive membrane damage. + +<|ref|>text<|/ref|><|det|>[[42, 911, 923, 958]]<|/det|> +Cell membrane can be disrupted by 2D nanomaterials either through physical damage or by chemical reactions.20- 22 The physical membrane damage was monitored by quartz crystal microbalance with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 945, 227]]<|/det|> +dissipation (QCMD). GNSs elicited a more obvious frequency increase (5.86 Hz) compared to GNSs- HA (2.53 Hz) (Fig. 1k,l and S13), demonstrating greater physical damage to the membrane. S- layer proteins (SLPs) are cell wall- related proteins located on the outermost layer of microorganisms that are in direct contact with the external environment and protect bacterial cells against xenobiotics. \(^{23}\) The proteomics data showed that the SLPs of B. tropicus decreased more significantly with GNSs than GNSs- HA (Figure S14), suggesting that HA adsorption reduced the physical interaction- driven membrane damage of GNSs. Therefore, it is clear that chemical oxidation dominates the membrane damage caused by GNSs- HA. + +<|ref|>sub_title<|/ref|><|det|>[[44, 255, 580, 283]]<|/det|> +## Disturbance of oxidative phosphorylation + +<|ref|>text<|/ref|><|det|>[[41, 298, 951, 460]]<|/det|> +The oxidative phosphorylation system plays a crucial role in generating ATP to sustain cellular activity. \(^{24 - 26}\) It was more severely disrupted by GNSs than GNSs- HA (Fig. 2a, S15, and S16). This disruption was primarily manifested through significant down- regulation of complex \(\mathbb{V}\) and \(\mathbb{V}_i\) upregulation of complex III, and interference with complex II and IV upon GNSs exposure (Fig. 2b). Conversely, GNSs- HA disruption was less severe and primarily affected complex II and IV. This impact on oxidative phosphorylation by GNSs was likely due to a disruption of electron transfer between iron- sulfur centers, as indicated in a previous study. \(^{7}\) + +<|ref|>sub_title<|/ref|><|det|>[[44, 491, 231, 518]]<|/det|> +## ATP synthesis + +<|ref|>text<|/ref|><|det|>[[41, 533, 951, 714]]<|/det|> +Disturbance of oxidative phosphorylation by GNSs inhibited ATP synthesis, reducing ATP content compared to the control (Fig. 2c). Conversely, there was no significant change in ATP content with GNSs- HA. Since sufficient energy supply is required to maintain physiological function, ATP deficiency could impair bacterial metabolism. Five differentially abundant proteins (DAPs) associated with metabolism- related pathways were significantly down- regulated with GNSs, while one was up- regulated with GNSs- HA (Figure S17- S19), indicating the inhibitory effect of GNSs on B. tropicus metabolism. Since metabolic processes require ATP and are necessary for bacterial growth and reproduction, \(^{27 - 29}\) the deficiency of ATP induced by GNSs inhibited metabolism and ultimately led to a reduction in bacterial growth. + +<|ref|>text<|/ref|><|det|>[[41, 731, 951, 925]]<|/det|> +Complex V (ATP synthase) is composed of \(\mathsf{F}_0\) and \(\mathsf{F}_1\) units, which act as the proton channel and catalyst for ATP synthesis, respectively. \(^{25}\) The synthesis of each ATP molecule requires \(2 \sim 3\) protons to be transported across the membrane from the intermembrane space to the cytoplasm by the proton channel. \(^{30}\) The down- regulation of complex V in the GNSs group interfered with proton pump function, inhibiting the delivery of protons into the cytoplasm (Fig. 2a,b). As the terminal electron acceptor of aerobic bacteria, intracellular \(\mathsf{O}_2\) is reduced to \(\mathsf{H}_2\mathsf{O}\) at the terminal of the ETC (Eq. 1). With insufficient proton pumping in the cells, incomplete reduction of \(\mathsf{O}_2\) occurred (Eq. 2- 4), resulting in the accumulation of intracellular ROS. \(^{31}\) This likely explains the delayed elevation of intracellular ROS with GNSs (Fig. 1d). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 899, 88]]<|/det|> +Conversely, GNSs- HA did not interfere with the expression of complex V (Fig. 2a), which allows the occurrence of four- electron reduction of \(O_{2}\) to water (Eq. 1), thus avoiding the generation of ROS. + +<|ref|>equation<|/ref|><|det|>[[80, 115, 614, 333]]<|/det|> +\[\begin{array}{l}{{\frac{1}{2}\mathrm{O}_{2}+2\mathrm{H}^{+}+2\mathrm{e}^{-}\rightarrow\mathrm{H}_{2}\mathrm{O}}}\\ {{}}\\ {{\mathrm{O}^{2}+\mathrm{e}^{-}\rightarrow\mathrm{O}_{2}\cdot\cdot}}\\ {{}}\\ {{\mathrm{O}_{2}+2\mathrm{e}^{-}+2\mathrm{H}^{+}\rightarrow\mathrm{H}_{2}\mathrm{O}_{2}}}\\ {{}}\\ {{\mathrm{O}_{2}+3\mathrm{e}^{-}+3\mathrm{H}^{+}\rightarrow\mathrm{H}_{2}\mathrm{O}+\mathrm{OH}\bullet}}\end{array} \quad (1)\] + +<|ref|>sub_title<|/ref|><|det|>[[42, 365, 501, 386]]<|/det|> +## Electron transfer between B. tropicus and graphenes + +<|ref|>text<|/ref|><|det|>[[40, 402, 950, 589]]<|/det|> +Although it is still unclear why GNSs- HA elicited higher toxicity than GNSs, more significant membrane damage with GNSs- HA from chemical oxidation, as opposed to physical disruption, provides a valuable clue. If an oxidation reaction occurs between B. tropicus and GNSs or GNSs- HA, electron transfer is inevitable and this can be characterized by an electrochemical method.32- 34 The linear sweep voltammetry (LSV) curves, Tafel slopes, Nyquist plots, and cyclic voltammetry (CV) curves demonstrated the electron transfer between GNSs- HA and B. tropicus was more efficient because HA adsorption decreased the electrochemical impedance (Fig. 2d- h, Supporting Information 2.7). Notably, the phenomenon of HA- enhanced electron transfer has been previously observed in microbial fuel cells.35 + +<|ref|>text<|/ref|><|det|>[[39, 604, 950, 904]]<|/det|> +Kelvin probe force microscopy (KPFM) was performed to determine the electron flow direction by measuring the bacterial surface potential before and after GNSs or GNSs- HA treatment. Compared to the control, the surface potential of B. tropicus was enhanced in the presence of both GNSs and GNSs- HA, indicating bacterial cells served as electron donors (Fig. 3a- c). According to the proteomic data, electron transfer flavoprotein subunits were up- regulated in GNSs- treated cells (Figure S20), indicating more endogenous electrons were delivered via extracellular electron transfer (EET). However, more biological electrons were lost in GNSs- HA- treated cells (Fig. 3c). In addition, GNSs and GNSs- HA may further transfer the obtained electrons to extracellular \(O_{2}\) ; the oxygen consumption rate was in the order of GNSs- HA > GNSs > control (Fig. 3d). Since no extracellular ROS was detected (Fig. 3e), the four- electron reduction of \(O_{2}\) to water likely occurred (Eq. 1), indicating the oxidation of bacterial components was not derived from extracellular ROS (Section 2.9). Consequently, there should be other pathways, including direct oxidation of biological components, contributing to the loss of electrons from GNSs- HA- treated B. tropicus, as facilitated by HA- enhanced electron transfer. + +<|ref|>sub_title<|/ref|><|det|>[[42, 932, 496, 958]]<|/det|> +## Oxidation of bacterial components + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 44, 950, 298]]<|/det|> +As a result of the loss of biological electrons, biomolecules such as lipid membrane components and intracellular macromolecules can be oxidized. After treatment with GNSs, lipid peroxidation as measured by malondialdehyde (MDA) followed a dose- dependent pattern (Fig. 3f), which was proportional to the intracellular ROS level, suggesting excessive ROS might explain GNSs- induced lipid peroxidation. Surprisingly, GNSs- HA induced even stronger lipid peroxidation, despite not leading to an increase in the intracellular or extracellular ROS (Fig. 1d and 3e). Given this, direct oxidation by GNSs- HA through electron transfer could be a plausible mechanism. \(^{33,36}\) Importantly, only GNSs- HA oxidized intracellular DNA, as determined by the level of 8- hydroxy- deoxyguanosine (8- OHdG), which is a marker of DNA oxidative damage (Fig. 3g). \(^{19}\) Several purine metabolism- related genes were significantly altered under GNSs- HA stress (Figure S21), suggesting the disturbance of purine metabolism, which might be due to oxidative damage of guanine in DNA driven by electron transfer rather than ROS. + +<|ref|>sub_title<|/ref|><|det|>[[42, 327, 439, 355]]<|/det|> +## Mechanisms and implications + +<|ref|>text<|/ref|><|det|>[[39, 370, 955, 735]]<|/det|> +HA is commonly known as a detoxifying agent for nanoparticles. However, in this current study, we found surface- bound HA enhanced the toxicity of GNSs to Gram- positive B. tropicus by exerting a different toxicity mechanism (Fig. 4). Specifically, GNSs caused serious physical membrane damage, leading to the disturbance of oxidative phosphorylation that not only induced intracellular ROS accumulation but also inhibited ATP synthesis. This disruption subsequently led to energy deficiency, which inhibited metabolism and caused cell death. Conversely, GNSs- HA induced less physical membrane damage but elicited greater LDH leakage and lipid peroxidation. Additionally, despite the lack of intracellular or extracellular ROS, we observed the oxidation of GSH and DNA, indicating that GNSs- HA directly extracted electrons from B. tropicus and oxidized biomolecules. While both materials can serve as electron acceptors through EET, electron transfer from B. tropicus to GNSs- HA was more efficient than GNSs (Fig. 2d- h). The conduction band (CB) minimum of semiconductor overlapping with the biological redox potential (BRP, - 4.12 ~ - 4.84 eV) allows for electron transfer from biological couples to the CB, leading to the oxidation of biological substances. \(^{37,38}\) The CB minimum of GNSs- HA (Fig. 3h and S22) was within the scope of BRP, enabling spontaneous electron transfer from B. tropicus biomolecules to GNSs- HA. These distinct antibacterial mechanisms are thus responsible for the HA- enhanced toxicity of GNSs to B. tropicus observed in the current study. + +<|ref|>text<|/ref|><|det|>[[40, 750, 949, 956]]<|/det|> +In order to determine if the proposed mechanism is universal, we also conducted the toxicity evaluations on Gram- negative Escherichia coli k12 (E. coli k12), Gram- positive staphylococcus aureus (S. aureus) and Bacillus subtilis (B. subtilis). Consistent with the findings in B. tropicus, B. subtilis was also found to be more susceptible to GNSs- HA, while the toxicity of GNSs to E. coli k12 and S. aureus was reduced after HA adsorption (Figure S23). Based on these results, we speculate that both bacterial membrane structure and morphology may contribute to the different responses. In fact, previous studies have indicated that Gram- positive bacteria are more susceptible to 2D nanomaterials than Gram- negative bacteria, possibly due to differences in bacterial membrane structure. \(^{39 - 43}\) The outer membrane of Gram- negative E. coli provides protection against physical damage caused by the sharp edges of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 951, 255]]<|/det|> +GNSs. \(^{44}\) Despite the absence of outer membrane protection in Gram- positive S. aureus, their smaller spherical morphology allows utilization of GNSs as a shelter, reducing the exposure density of sharp edges to the membrane relative to rod- shaped cells. \(^{45}\) Additionally, the cell membrane, which has a high insulation with the conductivity of \(10^{- 7} \text{S m}^{- 1}\) , \(^{46}\) limits electron outflow from the conductive interior of E. coli and S. aureus to GNSs- HA, even with the overlapping of the CB minimum with BRP. However, for rod- shaped Gram- positive bacteria such as B. tropicus and B. subtillis, GNSs- HA disrupts membrane integrity, which enables abnormal electron transfer that leads to the oxidation of biological substances. Overall, the combination of physical membrane perturbation and disturbed electron balance contributes to the altered bacterial toxicity mechanism of GNSs by HA adsorption. + +<|ref|>sub_title<|/ref|><|det|>[[44, 277, 207, 303]]<|/det|> +## Conclusions + +<|ref|>text<|/ref|><|det|>[[40, 317, 944, 588]]<|/det|> +In the current work, the effect of HA on bacterial toxicity of GNSs is strain dependent. Specifically, although GNSs- HA exhibited lower toxicity to E. coli k12 and S. aureus than bare GNSs, they were more toxic to B. tropicus and B. subtillis under the same exposure scenario, challenging the conventional view that HA mitigates ENPs toxicity. B. tropicus was used as a model to investigate these phenomena and the data showed that the HA corona altered the bacterial toxicity mechanism of GNSs. The physical damage induced by GNSs to the cell membrane disrupted oxidative phosphorylation, leading to ROS accumulation and ATP synthesis inhibition and ultimately bacterial death due to energy deficiency. Conversely, the band structure of GNSs was modulated by the HA corona, allowing GNSs- HA to extract electrons not only through EET, but also via direct biomolecule oxidation. This demonstration of enhanced ENPs' bacterial toxicity with HA attachment highlights the importance of understanding the mechanisms of eccorona activity when trying to accurately evaluate the risk of ENPs in the environment. + +<|ref|>sub_title<|/ref|><|det|>[[44, 610, 161, 636]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[44, 650, 300, 680]]<|/det|> +## Bacterial toxicity + +<|ref|>text<|/ref|><|det|>[[39, 695, 951, 955]]<|/det|> +The bacterial toxicity of GNSs, GNSs- HA, and GNSs + HA was evaluated using optical density (OD) values and the colony counting method. Briefly, \(1.0 \text{mL}\) of GNSs and GNSs- HA suspensions (0, 10, 100, 500, 1000, and \(2000 \mu \text{g mL}^{- 1}\) ), \(1.0 \text{mL}\) of bacterial suspension ( \(10^{5} \text{CFU mL}^{- 1}\) ), and \(8.0 \text{mL}\) of fresh Luria Bertani (LB) broth were mixed together, obtaining the final material concentrations at 0, 1, 10, 50, 100, and \(200 \mu \text{g mL}^{- 1}\) . For the GNSs + HA group, \(1.0 \text{mL}\) of bacterial suspension ( \(10^{5} \text{CFU mL}^{- 1}\) ), \(1.0 \text{mL}\) of HA ( \(12 \mu \text{g mL}^{- 1}\) ), \(1.0 \text{mg}\) of GNSs powder, and \(8.0 \text{mL}\) of fresh LB broth were mixed together. The mixtures were incubated at \(37^{\circ} \text{C}\) and shaken at 150 rpm. The OD values were determined every 5 min at 600 nm on a microplate reader (BMG LABTECH, Germany) for 30 min to obtain the short- term growth curve. As a comparison, the effect of dissolved HA ( \(12 \mu \text{g mL}^{- 1}\) ) on bacterial growth was evaluated following the same procedure. After 12 h incubation, the bacterial suspensions were diluted with sterilized water to a dilution series of \(10^{- 4}\) , \(10^{- 5}\) , and \(10^{- 6}\) . Then \(100 \mu \text{L}\) of each dilution was spread + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 955, 110]]<|/det|> +onto a LB agar plate, which was incubated at \(37^{\circ}C\) overnight. The colony forming units (CFUs) of all plates were counted and the survival rate \((\%)\) was calculated by comparing the CFUs of treatment groups with that of controls. + +<|ref|>text<|/ref|><|det|>[[42, 127, 941, 247]]<|/det|> +Melatonin, as an antioxidant, was used to explore the role of ROS in the bacterial toxicity. Briefly, \(0.5 \mathrm{mL}\) of GNSs and GNSs- HA suspensions, \(1.0 \mathrm{mL}\) of \(B\) . tropicus suspension ( \(10^{5} \mathrm{CFU} \mathrm{mL}^{- 1}\) ), \(8.0 \mathrm{mL}\) of fresh LB broth, and \(0.5 \mathrm{mL}\) melatonin solution ( \(20 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ) were mixed together to obtain the final concentration of GNSs and GNSs- HA at \(100 \mu \mathrm{g} \mathrm{mL}^{- 1}\) . The mixtures were shaken at \(120 \mathrm{rpm}\) and incubated at \(37^{\circ}C\) for \(12 \mathrm{h}\) . The survival rate was evaluated using the colony counting method. + +<|ref|>sub_title<|/ref|><|det|>[[44, 247, 480, 278]]<|/det|> +## Physical membrane damage + +<|ref|>text<|/ref|><|det|>[[42, 292, 952, 410]]<|/det|> +The interaction of GNSs and GNSs- HA with the model cell membrane was investigated by QCMD. Before the experiment, the gold- plated crystal sensor was cleaned by immersion in a mixed solution ( \(\mathrm{H}_2\mathrm{O}\) : \(30\%\) \(\mathrm{H}_2\mathrm{O}_2\) : \(25\%\) \(\mathrm{NH}_3 = 5:1:1\) ) for \(5 \mathrm{min}\) at \(75^{\circ}C\) , and then rinsed with a large amount of deionized water before drying under nitrogen. The sensor was then oxidized by UV ozone exposure for \(20 \mathrm{min}\) , followed by rinsing with Millipore water and drying with nitrogen gas. + +<|ref|>text<|/ref|><|det|>[[41, 425, 949, 607]]<|/det|> +To prepare the model membrane, chloroform solution of 1,2- dioleoyl- sn- glycero- 3- phosphocholine (DOPC, \(25 \mathrm{mg} \mathrm{mL}^{- 1}\) , \(0.2 \mathrm{mL}\) ) was added into a conical flask and dried under nitrogen to form a DOPC film, which was vacuum dried for more than \(4 \mathrm{h}\) to remove residual chloroform. Then \(5 \mathrm{mL}\) HEPES buffer solution was added under continuous magnetic stirring for \(30 \mathrm{min}\) to form DOPC vesicles. The DOPC solution was passed through a polycarbonate film (Whatman) back and forth for more than \(15 \mathrm{times}\) by a micro lipid extruder (Avanti Polar Lipids Inc.). A single layer of vesicles with a hydrodynamic diameter of \(192 - 194 \mathrm{nm}\) was obtained. The vesicle suspension was sealed with nitrogen in a glass bottle at \(4^{\circ}C\) and used within 4 days after preparation. + +<|ref|>text<|/ref|><|det|>[[41, 623, 950, 830]]<|/det|> +During the QCMD test, the temperature in the flow cell chamber was kept at \(37^{\circ}C\) , and the liquid flow rate was \(150 \mu \mathrm{L} \mathrm{min}^{- 1}\) . The frequency (representing the deposited mass) and energy dissipation (representing the viscoelastic properties of the deposited layer) were detected. HEPES buffer was first introduced to the system until the signal was stable. Afterwards, DOPC vesicles were introduced to form a supporting vesicle layer on the sensor surface. Then HEPES buffer was injected again to remove the unadsorbed DOPC vesicles. GNSs and GNSs- HA were then suspended in HEPES ( \(50 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ) to initiate the interaction process, followed by rinsing with HEPES to remove the unadsorbed GNSs and GNSs- HA. At the end of each experiment, \(1\%\) Triton X- 100 (a membrane solubilizer) was used to remove the adsorbed vesicles. + +<|ref|>sub_title<|/ref|><|det|>[[44, 833, 512, 862]]<|/det|> +## Electrochemical measurement + +<|ref|>text<|/ref|><|det|>[[42, 877, 944, 949]]<|/det|> +GNSs or GNSs- HA powders were mixed with a polytetrafluoroethylene solution ( \(2 \mathrm{wt}\%\) ), and the mixture was pasted on the surface of carbon fiber paper (surface area of \(2.5 \mathrm{cm}^2\) ) with a loading content of \(0.5 \mathrm{mg} \mathrm{cm}^{- 2}\) . After drying at \(100^{\circ}C\) for \(3 \mathrm{h}\) in the oven, the final electrode material was obtained. A bacterial + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 953, 280]]<|/det|> +suspension (100 mL) with an initial concentration of \(10^{5} \text{CFU mL}^{- 1}\) was incubated at \(37^{\circ} \text{C}\) for \(12 \text{h}\) before use. \(B\) . tropicus cells were harvested by centrifugation at 10000 rpm for 3 min and washed three times with anaerobic PBS (0.1 M, pH 7.0). The \(B\) . tropicus cells were resuspended in PBS and diluted to \(\sim 10^{9} \text{CFU mL}^{- 1}\) . The bacterial suspension was deoxygenated by purging with nitrogen for 30 min before each test. Electrochemical measurements were conducted by an electrochemical workstation (CHI660E, CH Instruments, Inc., Shanghai, China). A three- electrode cell was setup, consisting of a GNSs- or GNSs- HA- loaded carbon fiber paper working electrode, a Pt wire counter electrode, and an Ag/AgCl reference electrode. CV measurements were carried out at a scan rate of \(100 \text{mV s}^{- 1}\) and over a potential range from \(- 0.6\) to \(0.1 \text{V}\) . Electrochemical impedance spectra (EIS) were obtained over frequencies ranging from \(1000 \text{kHz}\) to \(1 \text{Hz}\) , with an amplitude of \(5 \text{mV}\) . LSV was obtained at a scan rate of \(50 \text{mV s}^{- 1}\) . + +<|ref|>sub_title<|/ref|><|det|>[[44, 279, 760, 312]]<|/det|> +## Local surface potential measurement by KPFM + +<|ref|>text<|/ref|><|det|>[[40, 328, 955, 555]]<|/det|> +After \(B\) . tropicus were incubated with GNSs and GNSs- HA ( \(100 \mu \text{g mL}^{- 1}\) ) for \(60 \text{min}\) at \(37^{\circ} \text{C}\) , the mixture was centrifuged at \(800 \text{rpm}\) for 5 min to remove the agglomerates, followed by dropping the suspension onto a monocrystalline silicon wafer. As soon as the sample was dried under ambient conditions, the silicon wafer with fixed samples was finally adhered to a small piece ( \(1 \text{cm} \times 1 \text{cm}\) ) of conductive double- sided tape, which was placed on a grounded microscope stage for KPFM measurement. Three bacterial cells were analyzed for each sample in order to verify that the microscopy images represented the average potential of the samples. KPFM was conducted in the tapping mode on the Agilent 5500 AFM. Platinum (Pt)/iridium (Ir)- coated silicon cantilever probes (Bruker, SCM- PIT- V2) were used as the conductive probes with a force constant of approximately \(3 \text{N m}^{- 1}\) and a nominal resonance frequency of \(75 \text{kHz}\) . + +<|ref|>sub_title<|/ref|><|det|>[[44, 577, 210, 603]]<|/det|> +## Declarations + +<|ref|>sub_title<|/ref|><|det|>[[44, 617, 360, 646]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[42, 664, 944, 730]]<|/det|> +X.Z. conceived the idea and designed the research. J. Z. performed most of the experiments. Z.X., S.Z., and J.Y. analyzed the omics data. W.T., and Y.X. interpreted the data. X.Z. and Q.Z. contributed to writing the manuscript. J.C.W., F.L., B.X. and F.W. modified the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[44, 761, 348, 791]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[42, 806, 940, 942]]<|/det|> +This work was supported by the National Natural Science Foundation of China (No. 42022056, 42192574, 42277423, 42077394, 22176196), GDAS' Project of Science and Technology Development (2022GDASZH- 2022010105, 2020GDASYL- 20200101002), GDAS' Project of Science and Technology Development (22022GDASZH- 2022020402- 01), GDAS' Project of Science and Technology Development (2022GDASZH- 2022020402- 02), and Guangdong Foundation for Program of Science and Technology Research (2020B1212060048). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 42, 194, 68]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[50, 80, 941, 970]]<|/det|> +1. NOVOSELOV, K. S. et al. FIRSOV Electric Field Effect in Atomically Thin Carbon Films. Science 306, 666-669 (2004). +2. Khan, K. et al. Recent developments in emerging two-dimensional materials and their applications. J. Mater. Chem. C 8, 387-440 (2020). +3. Su, Y. et al. Green Algae as Carriers Enhance the Bioavailability of (14)C-Labeled Few-Layer Graphene to Freshwater Snails. Environ. Sci. Technol. 52, 1591-1601 (2018). +4. Chen, W. et al. Black Phosphorus Nanosheets as a Neuroprotective Nanomedicine for Neurodegenerative Disorder Therapy. Adv. Mater. 30, 1703458 (2018). +5. Zhao, J. et al. Humic acid mitigated toxicity of graphene-family materials to algae through reducing oxidative stress and heteroaggregation. Environ. Sci.-Nano 6, 1909-1920 (2019). +6. Jannesari, M., Akhavan, O., Madaah Hosseini, H.R. & Bakhshi, B. Graphene/CuO(2) Nanoshuttles with Controllable Release of Oxygen Nanobubbles Promoting Interruption of Bacterial Respiration. ACS Appl. Mater. Interfaces 12, 35813-35825 (2020). +7. Zhou, H. et al. The inhibition of migration and invasion of cancer cells by graphene via the impairment of mitochondrial respiration. Biomaterials 35, 1597-1607 (2014). +8. Xu, L. et al. The Crucial Role of Environmental Coronas in Determining the Biological Effects of Engineered Nanomaterials. Small 16, e2003691 (2020). +9. Li, J. et al. Temperature-responsive tungsten doped vanadium dioxide thin film starves bacteria to death. Mater. Today 22, 35-49 (2019). +10. He, X. et al. Assessing the effects of surface-bound humic acid on the phototoxicity of anatase and rutile TiO(2) nanoparticles in vitro. J. Environ. Sci.-China 42, 50-60 (2016). +11. Zou, W., Zhou, Q., Zhang, X. & Hu, X. Environmental Transformations and Algal Toxicity of Single-Layer Molybdenum Disulfide Regulated by Humic Acid. Environ. Sci. Technol. 52, 2638-2648 (2018). +12. Mao, H.Y. et al. Manipulating the electronic and chemical properties of graphene via molecular functionalization. Prog. Surf. Sci. 88, 132-159 (2013). +13. Shang, E. et al. Relative importance of humic and fulvic acid on ROS generation, dissolution, and toxicity of sulfide nanoparticles. Water Res. 124, 595-604 (2017). +14. Zhang, X., Sui, M., Yan, X., Huang, T. & Yuan, Z. Mitigation in the toxicity of graphene oxide nanosheets towards Escherichia coli in the presence of humic acid. Environ. Sci.-Proc. Imp. 18, 744-750 (2016). +15. Zhang, Y. et al. The effects of humic acid on the toxicity of graphene oxide to Scenedesmus obliquus and Daphnia magna. Sci. Total Environ. 649, 163-171 (2019). +16. Hong, Y., Zeng, J., Wang, X., Drlica, K. & Zhao, X. Post-stress bacterial cell death mediated by reactive oxygen species. Proc. Natl. Acad. Sci. USA 116, 10064-10071 (2019). +17. Shaobin Liu, T.H.Z., Mario Hofmann, Ehdi Burcombe, Jun Wei, Rongrong Jiang, Jing Kong, and Yuan Chen Antibacterial Activity of Graphite, Graphite Oxide, Graphene Oxide, and Reduced Graphene + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[75, 45, 704, 65]]<|/det|> +Oxide: Membrane and Oxidative Stress. ACS Nano 5, 6971- 6980 (2011). + +<|ref|>text<|/ref|><|det|>[[52, 71, 833, 92]]<|/det|> +18. Oktyabrsky, G.V.S.O.N. Glutathione in Bacteria. Biochemistry-Moscow 70, 1199 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 98, 900, 141]]<|/det|> +19. Yang, S.P. et al. Influence of humic acid on titanium dioxide nanoparticle toxicity to developing zebrafish. Environ. Sci. Technol. 47, 4718-4725 (2013). + +<|ref|>text<|/ref|><|det|>[[50, 147, 920, 191]]<|/det|> +20. Feng, Y. et al. Interaction of Graphitic Carbon Nitride with Cell Membranes: Probing Phospholipid Extraction and Lipid Bilayer Destruction. Environ. Sci. Technol. 56, 17663-17673 (2022). + +<|ref|>text<|/ref|><|det|>[[50, 197, 945, 262]]<|/det|> +21. Hu, X., Ouyang, S., Mu, L., An, J. & Zhou, Q. Effects of Graphene Oxide and Oxidized Carbon Nanotubes on the Cellular Division, Microstructure, Uptake, Oxidative Stress, and Metabolic Profiles. Environ. Sci. Technol. 49, 10825-10833 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 268, 920, 311]]<|/det|> +22. Zhou, Q. & Hu, X. Systemic Stress and Recovery Patterns of Rice Roots in Response to Graphene Oxide Nanosheets. Environ. Sci. Technol. 51, 2022-2030 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 317, 905, 360]]<|/det|> +23. Gerbino, E., Carasi, P., Mobili, P., Serradell, M.A. & Gomez-Zavaglia, A. Role of S-layer proteins in bacteria. World J. Microb. Biot. 31, 1877-1887 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 366, 690, 386]]<|/det|> +24. BA Haddock, C.J. Bacterial respiration. Bacteriological Rev. 41 (1977). + +<|ref|>text<|/ref|><|det|>[[50, 392, 945, 436]]<|/det|> +25. Llorente-Garcia, I. et al. Single-molecule in vivo imaging of bacterial respiratory complexes indicates delocalized oxidative phosphorylation. Biochim. Biophys. Acta 1837, 811-824 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 442, 936, 485]]<|/det|> +26. Trounce, I. Genetic control of oxidative phosphorylation and experimental models of defects. Hum. Reprod. 15, 18-27 (2000). + +<|ref|>text<|/ref|><|det|>[[50, 491, 928, 558]]<|/det|> +27. Cui, Y. & Qu, X. Genetic mechanisms of prebiotic carbohydrate metabolism in lactic acid bacteria: Emphasis on Lacticaseibacillus casei and Lacticaseibacillus paracasei as flexible, diverse and outstanding prebiotic carbohydrate starters. Trends Food Sci. Tech. 115, 486-499 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 564, 931, 608]]<|/det|> +28. Komosinska-Vassev, K. et al. Alterations of glycosaminoglycan metabolism in the development of diabetic complications in relation to metabolic control. Clin. Chem. Lab. Med. 43, 924-929 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 613, 852, 657]]<|/det|> +29. Sobhanifar, S., King, D.T. & Strynadka, N.C.J. Fortifying the wall: synthesis, regulation and degradation of bacterial peptidoglycan. Curr. Opin. Struct. Biol. 23, 695-703 (2013). + +<|ref|>text<|/ref|><|det|>[[50, 662, 943, 706]]<|/det|> +30. Kaila, V.R.I. & Wikstrom, M. Architecture of bacterial respiratory chains. Nat. Rev. Microbiol. 19, 319-330 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 711, 938, 755]]<|/det|> +31. Tan, J. et al. A facile and universal strategy to endow implant materials with antibacterial ability via alkalinity disturbing bacterial respiration. Biomater. Sci. 8, 1815-1829 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 760, 930, 804]]<|/det|> +32. Chen, S. et al. MoS2 Nanosheets-Cyanobacteria Interaction: Reprogrammed Carbon and Nitrogen Metabolism. ACS Nano 15, 16344-16356 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 810, 896, 853]]<|/det|> +33. Li, J. et al. Antibacterial activity of large-area monolayer graphene film manipulated by charge transfer. Sci. Rep. 4, 4359 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 859, 945, 903]]<|/det|> +34. Zhou, S., Lin, M., Zhuang, Z., Liu, P. & Chen, Z. Biosynthetic graphene enhanced extracellular electron transfer for high performance anode in microbial fuel cell. Chemosphere 232, 396-402 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 908, 950, 952]]<|/det|> +35. Huang, B. et al. Ferroferric oxide loads humic acid doped anode accelerate electron transfer process in anodic chamber of bioelectrochemical system. J. Electroanal. Chem. 851, 113464 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 43, 950, 633]]<|/det|> +36. Liu S, Z.T.H., Hofmann M Antibacterial activity of graphite, graphite oxide, graphene oxide, and reduced graphene oxide: membrane and oxidative stress. ACS Nano 5, 6971-6980. +37. Li, J. et al. Band Gap Engineering of Titania Film through Cobalt Regulation for Oxidative Damage of Bacterial Respiration and Viability. ACS Appl. Mater. Interfaces 9, 27475-27490 (2017). +38. Qu, G. et al. Property-Activity Relationship of Black Phosphorus at the Nano-Bio Interface: From Molecules to Organisms. Chem. Rev. 120, 2288-2346 (2020). +39. Deokar, A.R., Lin, L.Y., Chang, C.C. & Ling, Y.C. Single-walled carbon nanotube coated antibacterial paper: preparation and mechanistic study. J. Mater. Chem. B 1, 2639-2646 (2013). +40. Pulingam, T. et al. Graphene oxide exhibits differential mechanistic action towards Gram-positive and Gram-negative bacteria. Colloids Surf. B Biointerfaces 181, 6-15 (2019). +41. Wang, M. et al. Interaction with teichoic acids contributes to highly effective antibacterial activity of graphene oxide on Gram-positive bacteria. J. Hazard. Mater. 412, 125333 (2021). +42. Xiong, Z. et al. Transcriptome Analysis Reveals the Growth Promotion Mechanism of Enteropathogenic Escherichia coli Induced by Black Phosphorus Nanosheets. ACS Nano 17, 3574-3586 (2023). +43. Xiong, Z. et al. Bacterial toxicity of exfoliated black phosphorus nanosheets. Ecotoxicol. Environ. Saf. 161, 507-514 (2018). +44. Ghaderi, O.A.E. Toxicity of Graphene and Graphene Oxide Nanowalls Against Bacteria. ACS Nano 4, 5731-5736. +45. Vy T. H. Pham, V.K.T., Matthew D. J. Quinn, Shannon M. Notley, Yachong Guo, Vladimir A. Baulin, Mohammad Al Kobaisi, Russell J. Crawford, and Elena P. Ivanova Graphene Induces Formation of Pores That Kill Spherical and Rod-Shaped Bacteria. ACS Nano 9, 8458-8467 (2015). +46. Yang, L. Electrical impedance spectroscopy for detection of bacterial cells in suspensions using interdigitated microelectrodes. Talanta 74, 1621-1629 (2008). + +<|ref|>sub_title<|/ref|><|det|>[[45, 650, 144, 676]]<|/det|> +## Figures + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 50, 920, 670]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 692, 114, 711]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[40, 732, 950, 920]]<|/det|> +(a) Survival rate of \(B\) . tropicus after treatment with GNSs and GNSs-HA for \(12 \text{h}\) ; (b) Effects of GNSs and GNSs-HA on the OD value of \(B\) . tropicus within \(30 \text{min}\) ; (c) Survival rate of \(B\) . tropicus after treated with GNSs, GNSs-HA, and GNSs+HA ( \(100 \mu \text{g mL}^{-1}\) ); (d) Intracellular ROS level after treatment with GNSs and GNSs-HA for \(2 \text{(light color)}\) and \(12 \text{h}\) (deep color); (e) GSH/GSSG ratio in \(B\) . tropicus after treatment with GNSs and GNSs-HA as compared to the value without any treatment; (f) Survival rate of \(B\) . tropicus after treated with \(100 \mu \text{g mL}^{-1}\) of GNSs and GNSs-HA in the presence and absence of melatonin; Morphology of \(B\) . tropicus without any treatment (g) or after GNSs (h) and GNSs-HA (i) treatments; (j) LDH activity of \(B\) . tropicus after treated with different concentrations of GNSs and GNSs-HA; Frequency variations of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 46, 949, 90]]<|/det|> +DOPC vesicles after introducing \(100 \mu \mathrm{g mL}^{- 1}\) of GNSs (k) and GNSs- HA (l). Error bars represent standard deviation from triple measurements. + +<|ref|>image<|/ref|><|det|>[[50, 115, 930, 655]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 677, 117, 696]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[40, 718, 955, 833]]<|/det|> +(a) Changes in ATP synthase related proteins after GNSs or GNSs-HA treatments; (b) Schematic diagram of respiratory chain changes in the GNSs treatment group; (c) ATP content after treated with GNSs and GNSs-HA (100 \(\mu \mathrm{g mL}^{-1}\) ); Electron transfer between B. tropicus and GNSs or GNSs-HA represented by LSV curves (d), Tafel slopes (e), Nyquist plots (f), and CV curves (g, h). Error bars represent standard deviation from triple measurements. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[57, 45, 920, 444]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 466, 117, 485]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[41, 508, 945, 644]]<|/det|> +KPFM images of \(B\) . tropicus without any treatment (a), or treated with GNSs (b) and GNSs- HA (c); (d) Oxygen consumption rate of \(B\) . tropicus treated with GNSs and GNSs- HA within \(30 \text{min}\) ; (e) Extracellular ROS level in the bacterial culture medium after treated with GNSs, GNSs- HA, and \(H_2O_2\) ; MDA (f) and 8- OHdG content (g) in \(B\) . tropicus after exposure to GNSs and GNSs- HA; (h) Schematic illustration of the possible electron transfer pathway between \(B\) . tropicus and the materials. Error bars represent standard deviation from triple measurements. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 52, 936, 520]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 550, 118, 570]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 591, 920, 681]]<|/det|> +Schematic illustration of different mechanisms of bacterial toxicity from GNSs or GNSs- HA exposure. GNSs induced physical damage to the cell membrane, which disrupted oxidative phosphorylation, leading to ROS accumulation and energy deficiency; GNSs- HA directly extracted electrons from \(B\) . tropicus across the mildly destroyed membrane, leading to the oxidation of biomolecules. + +<|ref|>sub_title<|/ref|><|det|>[[44, 703, 312, 731]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 754, 768, 774]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 792, 429, 812]]<|/det|> +- Supportinginformation2023.08.25.docx + +<--- Page Split ---> diff --git a/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/images_list.json b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..1969000cb510d8f78a5c519e84f91a6ced250f43 --- /dev/null +++ b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Schematic of the experimental configuration. X-ray diffraction spectra are recorded as a function of time on a JUNGFRAU 1M detector in a Debye-Scherrer geometry (left) and are azimuthally integrated into an XRD pattern where the diffraction peaks of ice VII are clearly measured in \\(20 \\mu \\mathrm{s}\\) (bottom left). \\(\\mathrm{SrB_4O_7:Sm^{2 + }}\\) fluorescence over time is recorded on an iXon EMCCD (right) alongside images on a Photron Fast Camera. A typical system pressure response (red) to the input voltage (blue) are plotted versus time (upper right).", + "footnote": [], + "bbox": [ + [ + 117, + 88, + 880, + 415 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Dynamic compression of liquid water at \\(0.4\\mathrm{GPa / ms}\\) . The sample pressure over time is measured using the \\(\\mathrm{SrB_4O_7:Sm^{2 + }}\\) -chip fluorescence while micro-photographs are recorded at \\(40\\mathrm{kHz}\\) framing time. At the liquid-solid phase transition a characteristic pressure drop due the density difference is observed. Visual observation with the ultra-fast camera reveals a homogeneous nucleation in the sample volume. The over-compression pressure, \\(\\Delta \\mathrm{P}\\) , is defined as the pressure differential between crystallization and the associated ice VII melting at the same temperature (1.56 GPa). The quantity \\(\\Delta t\\) represents the corresponding duration over which fluid water remains in its metastable state.", + "footnote": [], + "bbox": [ + [ + 195, + 92, + 799, + 454 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Measured pressure as a function of time for compression rates of 1.6 GPa/ms and 110 GPa/ms. At the freezing pressure into ice VII, a rupture in the compression slope is observed due to ,the negative discontinuity at crystallization. Pressure is measured using the \\(\\mathrm{SrB_4O_7}:\\mathrm{Sm^{2 + }}\\) fluorescence gauge and in some cases using the volume of a piece of Cu embedded in the sample chamber. The freezing into ice VII is evidenced from the XRD pattern. After freezing, excellent agreement is observed between the pressure measured from the volume of ice VII, the luminescence gauge and the volume of Cu. Good reproducibility of the sample pressure response to a given voltage function driving the piezo-actuator is observed. The quantities \\(\\Delta \\mathrm{P}\\) and \\(\\Delta \\mathrm{t}\\) defined in the text are represented for each run. Note the break in scale for the horizontal axis.", + "footnote": [], + "bbox": [ + [ + 192, + 90, + 790, + 528 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. The time evolution of the integrated XRD image of water freezing under a 0.8 GPa/ms compression rate. Metastable water crystallizes first into ice VII as evidenced by the appearance of its (110) peak at 580 μs. \\(\\sim 200\\) μs later, diffraction peaks attributed to ice VI are observed. The horizontal red and green lines mark the apparition of the ice VII and ice VI respectively. Top: integrated diffraction pattern of the same sample held at the maximum pressure of 1.8 GPa taken after a few minutes. A mixture of ice VI and ice VII is still observed.", + "footnote": [], + "bbox": [ + [ + 226, + 87, + 768, + 732 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Over-compression ( \\(\\Delta P\\) ) as a function of compression rate (II). Our data points alongside the averaged points from refs. 14–17 are reported. Temperature at freezing is shown as a color-scale. The grey-dashed lines is the best fit result of the power function \\(\\Delta P = a\\Pi^c + b\\) , with \\(a = 2.6654\\) , \\(b = 0.0564\\) and \\(c = -2.2544\\) . Inset: Low-pressure phase diagram of \\(\\mathrm{H}_2\\mathrm{O}\\) showing how the \\(\\Delta P\\) is calculated for isentropic compressions.", + "footnote": [], + "bbox": [ + [ + 188, + 85, + 805, + 560 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. Logarithm of the nucleation time as a function of over-compression. The nucleation time is assumed as \\(t_{n} = \\alpha \\Delta t\\) , with \\(\\alpha = 1\\) for the figure. Changing \\(\\alpha\\) only changes the constant of the fit. Our data points with averaged points from refs. 14 and 16 are fitted using the CNT model described in the text, with \\(\\log (t_{min}) = -11.39\\) , a=5380.17 and b=245293534.93. Inset: Extrapolation of the CNT model at large over-compression showing an asymptotic limit at the picosecond-timescale. The error bars in pressure are shown while those in time are smaller than the symbol size.", + "footnote": [], + "bbox": [ + [ + 188, + 87, + 800, + 541 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7. Presentation of the DAC used in this study (A). The DAC is then placed in contact with the piezo actuator (B). The set screws are used to finely adjust the starting pressure.", + "footnote": [], + "bbox": [ + [ + 189, + 342, + 804, + 500 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8. Dynamic compression of liquid water at slow compression rate ( \\(\\sim 0.1 \\mathrm{GPa / ms}\\) ). The sample pressure over time is measured using the \\(\\mathrm{SrB_4O_7}:\\mathrm{Sm}^{2 + }\\) fluorescence. Two events are observed: the first one associated with the crystallisation of water into ice VI and the second with the solid-solid phase transition of ice VI into ice VII.", + "footnote": [], + "bbox": [ + [ + 186, + 85, + 802, + 541 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9. Low-pressure phase diagram of \\(\\mathrm{H}_2\\mathrm{O}\\) . The metastable melting lines of ice VI and VII have been experimentally measured as described in the text. The points correspond to the observation of freezing as described in the main text. Nucleation of ice VII in the stability domain of ice VI is observed for a compression rate below \\(0.4\\mathrm{GPa / ms}\\) at \\(20^{\\circ}\\mathrm{C}\\) . As the compression rate (as the color scale) increases, the direct liquid – ice VII transition is pushed to higher pressures.", + "footnote": [], + "bbox": [ + [ + 192, + 100, + 802, + 563 + ] + ], + "page_idx": 22 + } +] \ No newline at end of file diff --git a/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846.mmd b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9fa218eef6f6603e48f39a7fba1f43dfd2dbad19 --- /dev/null +++ b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846.mmd @@ -0,0 +1,265 @@ + +# Metastable water at several compression rates and its freezing kinetics into ice VII + +Charles Pépin charles.pepin@cea.fr + +CEA https://orcid.org/0000- 0002- 9638- 3303 + +Ramesh André CEA + +Florent Occelli CEA, DAM, DIF + +Florian Dembele CEA https://orcid.org/0009- 0001- 3626- 9250 + +Aldo Mozzanica Paul Scherrer Institute + +Viktoria Hinger PSI https://orcid.org/0000- 0002- 2616- 4084 + +Matteo Levantino ESRF - The European Synchrotron https://orcid.org/0000- 0002- 1224- 4809 + +Paul Loubeyre CEA/DIF https://orcid.org/0000- 0002- 1778- 3510 + +Article + +Keywords: + +Posted Date: April 23rd, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4234717/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on September 19th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52576-z. + +<--- Page Split ---> + +# Metastable water at several compression rates and its freezing kinetics into ice VII + +Charles M. Pépin, \(^{1,2,*}\) Ramesh André, \(^{1,2}\) Florent Occelli, \(^{1,2}\) Florian Dembele, \(^{1,2}\) Aldo Mozzanica, \(^{3}\) Viktoria Hinger, \(^{3}\) Matteo Levantino, \(^{4}\) and Paul Loubeyre \(^{1,2}\) + +\(^{1}\) CEA, DAM, DIF, F- 91297 Arpajon, France \(^{2}\) Université Paris- Saclay, CEA, Laboratoire Matière en Conditions Extrêmes, 91680 Bruyères- le- Châtel, France \(^{3}\) PSI, Forschungsstrasse 111, 5232 Villigen, Switzerland \(^{4}\) European Synchrotron Radiation Facility, 71 Avenue des Martyrs, CS 40220, 38043 Grenoble, France + +<--- Page Split ---> + +## Abstract + +Water can be dynamically over- compressed well into the stability field of ice VII. Whether water then transforms into ice VII, vitreous ice or a metastable novel crystalline phase remained uncertain. We report here the freezing of over- compressed water to ice VII by time- resolved X- ray diffraction. Quasi- isothermal dynamic compression paths are achieved using a dynamic- piezo- Diamond- Anvil- Cell, with programmable pressure rise time from 0.1 ms to 100 ms. By combining the present data set with those obtained on various ns- dynamical platforms, a complete evolution of the solidification pressure of metastable water versus the compression rate is rationalized within the classical nucleation theory framework. Also, when crystallization into ice VII occurs in between 1.56 GPa and 2.0 GPa, that is in the stability field of ice VI, a structural evolution over few ms is then observed into a mixture of ice VI and ice VII that seems to resolve apparent contradictions between previous results. + +## INTRODUCTION + +Water under pressure is both ubiquitous in natural phenomena and a complex system. The quasi- static pressure exploration of \(\mathrm{H}_2\mathrm{O}\) phase diagram, under various temperatures, has revealed a rich polymorphism, with nineteen crystalline forms of ice identified so far [1, 2]. Ice VII is a prominent phase occupying a large region of the phase diagram at pressures above 2 GPa and having the simplest ice structure with a body- centered cubic oxygen sublattice and a proton disordered arrangement [3, 4]. Ice VII is believed to play an important role in the physics of outer planetary bodies [5] and it is now considered a mineral as it has been recovered from the earth mantle as a diamond inclusion [6]. Understanding the structural changes of water under rapid temperature or pressure variation is a topic of current focus, with the observation of metastable phases and implication to impact events on planetary icy bodies. On one hand, rapid cooling of \(\mathrm{H}_2\mathrm{O}\) below its glass transition temperature of 136 K yields an amorphous phase known as low- density amorphous (LDA) ice [7]. A High density amorphous (HDA) ice has also been observed under pressure and the transition between LDA and HDA was speculated to be consistent with the existence of a liquid- liquid transition in supercooled water [8, 9]. However this analogy is still a matter of ongoing debate [10, 11]. + +<--- Page Split ---> + +The true \(\mathrm{H}_2\mathrm{O}\) liquid- liquid transition could be recently observed by fast isochoric heating of a quenched HDA sample at ambient pressure [12]. On the other hand, many studies have investigated rapidly compressed water. Quasi- isentropic pressure loading platforms, using laser drives [13, 14], magnetic drives [15, 16] and gas guns [17] have been used to investigate the freezing kinetics in over- compressed water. Loading pressure times within the nanosecond timeframe, associated to compression rate covering two orders of magnitude over the different platforms, could then be investigated. Liquid water was observed to persist up to 8 GPa, well into the stability field of ice VII and a freezing into ice VII was therefore speculated. The effects of loading pressure within the millisecond range were investigated using the dynamical diamond anvil cell (d- DAC) [18–20]. The corresponding compression rates could over- compress liquid water up to approximately 2 GPa at most. Freezing into ice VII in the stability field of ice VI was first suspected [18], then evidence of HDA was presented [19] and finally evidence of a mixture between ice VI and HDA was shown [20]. It should be noted that none of these dynamic studies could unambiguously resolve the freezing mechanism of metastable water, due to the absence of a time- resolved microscopic diagnostic. The present study aims to elucidate, at the microscopic level, the solidification process of metastable water under dynamic compression. In particular, we are investigating whether ice VII is still the freezing state of metastable water, and why the transition sequence reverses compared to rapid cooling, i.e. with an amorphous phase at low compression rates and a crystalline phase at high compression rates. Additionally, we aim to consistently interpret the observations across the entire range of compression rates accessible, using d- DAC on one end to gas- gun/laser compression on the other end. To better bridge the gap between static compression and nanosecond- dynamic compression, the d- DAC technique has been optimized to achieve a pressure rise time of about 0.1 ms, allowing four orders of compression rates to be covered on the same d- DAC platform. Structural microscopic changes along the compression path were monitored using time- resolved X- ray diffraction (XRD), complemented by optical imaging and pressure luminescence measurements, all with framing steps down to a few microseconds. + +<--- Page Split ---> + +## RESULTS + +## Instrumentation + +The experimental setup, depicted in Figure 1, comprises a piezoactuated d- DAC along with three real- time event monitoring diagnostics: X- ray diffraction, pressure luminescence measurements, and optical imaging. The DAC serves as the primary tool for static high- pressure experiments. Leveraging diamond's remarkable transparency across the electromagnetic spectrum, light is utilized to extract properties of the compressed samples. Typically, detailed characterizations of sample properties under pressures up to 100 GPa are conducted, employing primarily synchrotron X- ray diffraction completed by spectroscopy methods over the infra- red to X- ray excitation range [21]. Recently, this characterization has been extended to the TPa range [22]. The concept of operating the DAC dynamically was proposed in 2007, employing voltage- driven piezoactuators to generate pressure [18, 23]. Subsequently, pressure rise times in the 10 ms range were demonstrated, particularly in the study of water [19, 20]. An improved design employing a hollow cylinder piezoactuator has been proposed, offering access to sub- millisecond rise times [24]. Our d- DAC is based on a similar design. As illustrated in Figure 1, the sample pressure response in the sample follows the tailored voltage applied to the piezoactuator. Pressure rise times as fast as 100 us can be achieved. Importantly, the experimental conditions maintain close proximity to isothermal compression, facilitated by the high thermal conductivity of diamond anvils [25]. The equilibration time of pressure in the sample is estimated to be less than 1 us, derived from the reverberation time of compressive waves propagating at the sample's sound velocity across the chamber. The d- DAC retains the optical access of a standard DAC. The primary challenge in detailed measurements during dynamic compression has been to follow the sample's properties with sufficiently small time steps. d- DAC compression experiments initially relied on ms- time resolved micro- photography, ruby luminescence pressure measurement, and Raman spectroscopy [18- 20]. Time- resolved X- ray diffraction measurements in the d- DAC became feasible using the high brilliance, high flux X- ray beams from third- generation synchrotron sources and by the emergence of large hybrid pixel array detectors with very short readout times, such as the EIGER [26] and LAMBDA [24] detectors. Angular X- ray diffraction patterns at repetition rates up to 4 kHz have been reported [24]. In the present study, + +<--- Page Split ---> + +diffraction patterns at a maximum repetition rate of 250 kHz could be achieved, benefiting significantly from the high flux at 19 KeV in front of the ID09 beamline of the EBS- ESRF synchrotron [27] and the operation of the JUNGFRAU detector in burst mode (limited to 16 images)[28]. Pressure luminescence measurements and imaging of the sample could be performed within a similar timeframe, though not simultaneously with XRD. Typical XRD patterns and luminescence spectra are shown in Figure 1. The pressure was measured using the samarium- doped strontium borate (SrB4O7 : Sm2+) luminescence pressure gauge [29] and using XRD data from the Birch- Murnaghan equation of state (EoS) of ice VII [30] and the Vinet EoS of Cu [31]. A high reproducibility of the sample pressure evolution under a given piezoactuator voltage function acting on the piston of the d- DAC is observed. + +## Pressure response of water under compression ramps + +By applying a step load function on the piezo- actuator of the d- DAC, the pressure is expected to undergo a three- stage evolution: initially a gradual increase in pressure, followed by a rapid linear rise, and reaching a plateau. However, if a phase transition occurs, the pressure rise is disrupted by the convolution with the pressure drop associated to the negative volume discontinuity at the transition, resulting in an apparent negative compressibility. This phenomenon is depicted in Figure 2 and in Figure 8 (Supplementary Materials) when compressing water from its liquid phase across the stability domains of ice VI and ice VII. In figure Figure 8, with a compression rate of \(\sim 0.1\) GPa/ms, two events are observed: the first one associated with the crystallisation of water into ice VI and the second with the structural transition of ice VI into ice VII. If the over- compression of the low- pressure phase within the stability field of the high pressure phase is not excessive, the pressure drop aligns with the pressure of the phase transition. A pressure plateau at the melting pressure is even observed if the compression rate is slower than the crystal growth rate. As depicted in Figure 2, when crystallization occurs at higher compression rates \((\geq 0.4 \text{GPa / ms})\) , the pressure drops to 1.56 GPa, followed by the observation of a pressure plateau. Time- resolved photo- micrographs correlate optical changes in the sample with deviations from the expected linear pressure rise. These deviations begin when an homogeneous nucleation occurs, while the subsequent plateau at 1.56 GPa corresponds to crystal growth. Once the sample chamber is again filled with an ice crystal, a linear pressure rise resumes, with no further pressure + +<--- Page Split ---> + +rise anomalies observed. Notably, he solid- liquid equilibrium at 1.56 GPa corresponds to the melting point of ice VII (obtained from the melting curve of ice VII metastably extended below the Ice VI- ice VII- liquid triple point, see Supplementary Materials). Similarly, a stress release phenomenon, resulting from the interplay of compression and phase transition, was noted in the VISAR traces of the ns- dynamical quasi- isentropic compression experiments [14- 17], being interpreted as the signature of a phase transition. This characteristic was termed a kinetic van der Waals loop [15]. Its observation implied that a first order phase transition was taking place, even for the highest compression rate achieved with laser shocks, thus ruling out the transformation of super- compressed water into an amorphous phase via a glass phase transition. It was suggested that crystallization occurs into ice VII, although without time- resolved XRD measurements this proposition remains to be proven. + +In this study, over- compressed water has been investigated for various compression ramps, ranging from 0.1 ms to 20 ms in pressure rise time. The compression peak was maintained below 20 GPa, so the associated compression rates varied from 0.4 GPa/ms to 110 GPa/ms. The solidification of metastable water into ice VII was always observed by time resolved XRD. Figure 3 displays typical pressure responses of the water sample for two measurements conducted at slow and fast compression rates. Sample pressure- versus- time evolution for each compression was consistently measured using various pressure gauges: X- ray diffraction of a small metal flake (Sn or Cu) embedded in water, using their Vinet Equation of State (EoS) [31, 32]; XRD volume of ice VII post metastable water crystallization, using the Birch- Murnaghan EoS of ice VII [30]; and the strontium- doped- borate luminescence gauge [29]. An excellent agreement between the different pressure determinations was observed, along with high reproducibility of the pressure evolution. In all time- resolved compression ramps, a disruption in the pressure rise is observed correlated to the nucleation of metastable water to ice VII, as evidence by the appearance of XRD (110) and (111) peaks of ice VII. At low compression rates, a kinetic pressure loop with apparent negative compressibility can be noted while at higher rates, this feature transitions more towards a slope rupture, probably because the framing time of the measurements is too long to capture the pressure dip. + +As shown in Figures 2 and 3, two quantities are defined for future data analysis: \(\Delta t\) , representing the duration over which fluid water remained in its metastable state along a given pressure ramp; \(\Delta P\) , indicating the over- compression pressure, i.e., the pressure differential between crystallization and the associated ice VII melting at the same temperature. + +<--- Page Split ---> + +## 1. Crystallization of ice VII in the stability domain of ice VI + +Between compression rates of 0.15 GPa/ms and 0.9 GPa/ms, water remains metastable up to pressures ranging from 1.56 GPa and 2.1 GPa, respectively. This pressure range corresponds to the previous domain of investigation in d- DAC experiments focusing on the solidification of metastable water. However, the findings from these studies appear contradictory. One study suggested the solidification into ice VII [18], supported by a plateau around 1.5 GPa in the pressure response of the sample, similarly to that depicted in Figure 2. Another study proposed that ice VII germinating in the stability field of ice VI evolves into amorphous HDA ice, based on the Raman measurements taken at the end of the compression [19]. In a third study, a mixture of ice VI and vitreous ice was observed by Raman and XRD measurements taken at the end of the pressure ramp [20]. The situation is puzzling and prompts the need for time resolved XRD measurement to elucidate the structure of the nucleated ice and the mechanism of its growth and transformation over time. Time resolved XRD measurements of metastable water crystallizing at 1.75 GPa are shown in Figure 4. After azimuthal integration, the time resolved XRD data is represented as a 2D plot with diffraction angle on the x- axis, time on the y- axis and intensity displayed using a grey color scale. The XRD signal of ice VII (evidenced by the strong (110) peak) emerges first, followed by the appearance of the ice VI XRD pattern (evidenced by its 3 main peaks) after 200 us, with its intensity increasing over time. After maintaining the sample at its peak pressure of 1.8 GPa for a few minutes, the integrated diffraction pattern (Figure 4, top) indicates a mixture of ice VI and ice VII. The present observations support the hypothesis that metastable water, within the pressure range of 1.5 GPa to 2.1 GPa, nucleates into ice VII first, despite this range being the stability field of ice VI. Ice VI nucleation is also observed after a few hundred microseconds, revealing the difference in nucleation time between ice VI and ice VII. In this pressure domain, where both phases are stable, the faster nucleation of ice VII could be attributed to a similarity of the local order between the liquid and ice VII, resulting in a lower interfacial energy and thus a lower nucleation barrier, leading to earlier nucleation [18]. Here, the ice sample remaining as a mixture of ice VI and ice VII is observed at 1.8 GPa. In another XRD study, a mixture of ice VI and amorphous ice was reported at 1.7 GPa[20]. This difference could be explained by the fact that at 1.8 GPa, ice VII is approaching its thermodynamic stability, allowing for + +<--- Page Split ---> + +the stabilization of a mixture of ice VI and ice VII, whereas at 1.7 GPa, ice VII is gradually transforming, leading to the stabilization of a mixture of ice VI and amorphous ice. These results offer a clarification to the apparent contradiction of the previous findings. + +## 2. Freezing pressure of metastable water versus compression rate + +The pressure limit of metastable water is here observed to continuously increase with the compression rate, by scanning the compression rate over 4 orders of magnitude in the d- DAC. Water can be over- compressed up to 2.9 GPa at 300 K for a compression rate of 110 GPa/ms. This value is much lower than the pressure limit of metastable water in ns- dynamic measurements [14] which has been reaching 8 GPa for compression rate about 4 orders of magnitude higher than the one in d- DAC. Given that the compression ramp is isothermal in the d- DAC while it is isentropic in the ns- compression ramps, a consistent metric for the overpressure value is necessary. The differential pressure \(\Delta P\) , as previously defined, is used. For the isentropic ramps, the measured pressure limit of metastable water, \(P_{lim}\) , is converted into \(\Delta P\) by computing the difference between \(P_{lim}\) and the associated ice VII melting pressure at the temperature of the isentrope at \(P_{lim}\) , as illustrated in the inset of Figure 5. Figure 5 plots in a semi- logarithmic scale the metastability pressure limit of water as a function of compression rate, obtained here using the d- DAC and utilizing data from other various platforms [14- 17] (excluding the data of Gleason et al. due to heterogeneous nucleation [13]), so spanning six orders of compression rates. Remarkably, the various datasets align smoothly on an increasing trend. Recently, Myint et al. have proposed a scaling law for the onset of solidification at extreme over- compression of liquids [33], with \(\Delta P\) being a power function of the compression rate II. We have fitted the dataset in Figure 5 with a similar power function \(\Delta P = a\Pi^{c} + b\) . Notably, the exponent \(c\) of the fit equals to 0.056, which matches the one given by the Myint's scaling law, equal to 0.069. The validity of such a scaling law is thus extended over the whole range of compression rates leading to freezing of metastable water into ice VII. It should be noted though that the constant \(b\) is added here to take into account the need of a minimal compression rate above which freezing of metastable water into ice VII is observed. + +Over- pressurization is conceptually equivalent to isobaric cooling. When metastable water freezes under a compression rate of 110 GPa/ms, it corresponds to a cooling rate of + +<--- Page Split ---> + +\(10^{7}\mathrm{K / s}\) . Such a rate has been utilized to observe the vitrification of water by spraying water in a cryogenic medium [34]. In contrast here, a solidification into ice VII through homogeneous nucleation is observed. + +## DISCUSSION + +During the freezing of metastable water, we observe homogeneous nucleation of ice VII which uniformly fills the entire sample chamber, across the range of the compression rates investigated here with the d- DAC. Consequently, we analyze the freezing kinetics using the classical Johnson- Mehl- Avrami- Kolmogorov (JMAK) theory. Within this framework, the freezing kinetics incorporates contributions from the nucleation rate \((\gamma)\) and the growth velocity of the ice germs \((v)\) , which are the two dominant mechanisms to obtain the kinetic time scale of freezing \((\tau_{F})\) , as: \(\tau_{F} \propto (\gamma v^{3})^{- 1 / 4}\) [35]. In highly metastable states, the nucleation rate predominates over the growth rate, thereby controlling the freezing kinetics. That is evidenced by the absence of a plateau in the pressure- time evolution in the d- DAC for compression rates above a few GPa/ms. According to Classical Nucleation Theory (CNT), the nucleation rate is proportional to the probability of spontaneously overcoming a free energy barrier associated with the formation of a critical- nucleus ice crystallite, expressed in terms of \(\Delta g\) , the Gibbs free energy difference between the liquid and the solid. The nucleation time and hence the freezing time can be expressed as: + +\[t_{n} = \frac{\exp\left(\frac{B}{\Delta g^{2}}\right)}{A} \quad (1)\] + +where \(A\) and \(B\) are parameters specific to the nucleation process [36]. \(\Delta g\) can be calculated from the equation of state of solid ice VII and that of liquid extrapolated in the metastable domain. Since water and ice are quite incompressible, we estimate the volume variation over the pressure range covered here assuming a constant bulk modulus (2), that is those of the liquid [18] and the solid [30] calculated at the melting pressure. + +\[\frac{\partial V}{\partial P} \approx -KV_{0} \implies V(P) = V_{0} - KV_{0}P \quad (2)\] + +\(\Delta g\) is then straightforwardly estimated in terms of the over- compression \(\Delta P\) of the liquid and of the liquid- solid volume difference \(\Delta V_{l - s}\) by: + +\[\Delta g = \int \Delta V_{l - s} dP = a + \Delta V_{0_{l - s}} \Delta P - \frac{1}{2} \Delta (KV_{0})_{l - s} \Delta P^{2} \quad (3)\] + +<--- Page Split ---> + +From which the freezing time can be related to \(\Delta P\) through the relation: + +\[log(t_{n})\propto log(t_{min}) + \frac{b}{(a + \Delta V_{0_{l - s}}\Delta P - \frac{1}{2}\Delta (KV_{0})_{l - s}\Delta P^{2})^{2}} \quad (4)\] + +This phenomenological model serves as a fitting function to correlate the freezing time of metastable water to the metastable over- pressure. In the dynamical compression of water described above, the freezing nucleation time is intertwined with the duration of the over- compression, \(\Delta t\) , with the nucleation time being \(t_{n} = \alpha \Delta t\) . We assume that \(\alpha\) remains constant across the different experimental data, from the d- DAC to the ns- isentropic compressions. The freezing time is plotted against the over- compression in Figure 6, where the proposed fit form in equation 4 accurately reproduces the entire dataset. + +The extrapolation of the fit to larger over- compression is shown in the inset of Figure 6, revealing an intriguing asymptotic behavior at high over- compression. Beyond a \(\Delta P\) of 10 GPa, the predicted freezing time of water is only a few picoseconds. Although such experimental over- compression seems unfeasible, numerical simulations are seen to validate the asymptotic behavior. Numerically, the thermodynamic conditions of metastable water can be established in less than a ps, and a freezing time of 12 ps has been reported for an over- compression of water about 30 GPa [37]. With increasing compression rate, water tends to reach a higher over- compressed metastable state. It has been suggested, though, that there exists a pressure limit close to 11 GPa for observing metastable water [14], beyond which freezing should occur via mechanisms such as spinodal decomposition or amorphization. In contrast, we do not propose a pressure limit for metastable water here but instead suggest a freezing time limit of a few ps. + +It is assumed that the dynamic compression experiments from refs. 16 and 14 resulted in a crystalline sample of ice VII. This assumption is reinforced by the coherence observed between the current d- DAC dataset and the previously obtained nanosecond- isentropic compression data.. However, a clear confirmation through time- resolved XRD under isentropic compression is warranted, possibly utilizing a free electron- laser X- ray diffraction measurement, as in ref.13, but assuring the conditions of homogeneous nucleation. Finally, the present study illustrates the remarkable ability of using the d- DAC to vary dynamic phenomena over four orders of magnitude of compression rate on the same platform, while simultaneously enabling detailed measurements of property evolution through time- resolved visual observation and XRD measurements. + +<--- Page Split ---> + +## METHODS + +X- ray diffraction measurements: X- ray diffraction experiments were carried out at the ID09 beamline of the ESRF. Data were collected using a 19 keV X- ray pink beam (1.5% bandwidth) focused to a 30 μm (FWHM) round spot. A 100 μm pinhole a few centimeters away from the sample was used to remove any contribution from the tails of the beam and from air scattering. The beamline heat- load chopper was operated at its slowest rotation speed in order to obtain X- ray pulses up to 1.5 ms, time long enough to cover the dynamics of interest. Diffraction images were collected using a JUNGFRAU 1M detector. The detector, composed of 2 tiles with 500k pixels each, has a 75 μm pixel pitch and uses a 320 μm thick silicon sensor. The detector was horizontally offset with respect to the direct X- ray beam to avoid radiation damage. It was operated in storage cell (SC) mode, where a burst of up to 16 images can be collected quickly and stored on a local, in pixel, analog memory for a later read- out. The maximum achievable frame rate during the burst is 250 kHz, with 1.5 μs exposures; however, the timing is adaptable, allowing for the utilization of longer periods and exposures if needed. In this study, the exposure time was typically 7 μs. The detector was controlled by the beamline control software via a custom TANGO device. Beside the run themselves, detector control included dark frames collection routines, which were individual per storage cell and timing setup. The timing of the measurement was orchestrated by the beamline control hardware, which would, in sync with the heat- load chopper, open the millisecond shutter to let the 1.5 ms long X- ray pulse on the sample, trigger the detector burst and the dDAC piezo controller. Sample- detector distance ( \(\sim 170\) mm), tilt, and rotation were calibrated using a reference Bi sample with DIOPTAS [38]. The two- dimensional images from the JUNGFRAU detector were corrected for offset and gain, and then radially integrated using DIOPTAS. Individual peaks of both water and the X- ray gauge were fitted with a PseudoVoigt function for the determination of the lattice parameters. + +\(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fluorescence measurements: \(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fuorescence signal was collected in a back- scattering geometry by an iXon electron multiplying (EM) CCD coupled to a 1200 l/mm grating, allowing a spectral resolution of approximately 0.08 nm. The iXon EMCCD is used in Fast- Kinetics mode, allowing the recording of spectra with an acquisition time varying from 1 to 500 μs with a time interval of \(\sim 8\) μs between each + +<--- Page Split ---> + +spectrum. \(\mathrm{SrB_4O_7:Sm^{2 + }}\) fluorescence excitation was performed using \(10\mathrm{mW}\) of a 491 nm Cobolt laser. With nonradiative decay times for \(\mathrm{Sm^{2 + }}\) ions typically in the range of a few nanoseconds [40], \(\mathrm{SrB_4O_7:Sm^{2 + }}\) emerges as an ideal pressure gauge for time- resolved experiments with a microsecond- level time resolution. + +d- DAC experiments: Dynamic compression experiments were performed using a LeToullec DAC, i.e a modification of the membrane- DAC [39] with a piston guided by four rods, in conjunction with the Piezosystem Jena HPST 1000/35- 25/80 VS45. A waveform Tabor Electronics WW2074 and an Amplifier Piezosystem Jena RCV 1000/7 are used to deliver the current in the piezoactuator with a chosen profile. DACs were equipped with Boelher- Almax design diamond anvils with \(300\mu \mathrm{m}\) culets. Stainless steel gaskets were used for all experiments, indented to \(40\mu \mathrm{m}\) thickness with a \(100\mu \mathrm{m}\) - diameter hole for the experimental chamber. + +## ACKNOWLEDGEMENTS (NOT COMPULSORY) + +The authors acknowledge the European Synchrotron Radiation Facility for provision of synchrotron radiation facilities on beamline ID09, during beam time allocated to the proposal HC- 4895. + +## AUTHOR CONTRIBUTIONS STATEMENT + +P.L. and C.P. conceived the experiment; R.A., F.O. and C.P. developed the d- DAC; All authors participated to the ESRF HC- 4895 experiment; F.D. and C.P. performed complementary in- house experiments; A.M. and V.H. operated the Jungfrau detector; M.L. operated the beamline; C.P. processed the XRD data; C.P. and P.L. analyzed the data and wrote the manuscript; All authors revised the manuscript. + +## ADDITIONAL INFORMATION + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Schematic of the experimental configuration. X-ray diffraction spectra are recorded as a function of time on a JUNGFRAU 1M detector in a Debye-Scherrer geometry (left) and are azimuthally integrated into an XRD pattern where the diffraction peaks of ice VII are clearly measured in \(20 \mu \mathrm{s}\) (bottom left). \(\mathrm{SrB_4O_7:Sm^{2 + }}\) fluorescence over time is recorded on an iXon EMCCD (right) alongside images on a Photron Fast Camera. A typical system pressure response (red) to the input voltage (blue) are plotted versus time (upper right).
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Dynamic compression of liquid water at \(0.4\mathrm{GPa / ms}\) . The sample pressure over time is measured using the \(\mathrm{SrB_4O_7:Sm^{2 + }}\) -chip fluorescence while micro-photographs are recorded at \(40\mathrm{kHz}\) framing time. At the liquid-solid phase transition a characteristic pressure drop due the density difference is observed. Visual observation with the ultra-fast camera reveals a homogeneous nucleation in the sample volume. The over-compression pressure, \(\Delta \mathrm{P}\) , is defined as the pressure differential between crystallization and the associated ice VII melting at the same temperature (1.56 GPa). The quantity \(\Delta t\) represents the corresponding duration over which fluid water remains in its metastable state.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Measured pressure as a function of time for compression rates of 1.6 GPa/ms and 110 GPa/ms. At the freezing pressure into ice VII, a rupture in the compression slope is observed due to ,the negative discontinuity at crystallization. Pressure is measured using the \(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fluorescence gauge and in some cases using the volume of a piece of Cu embedded in the sample chamber. The freezing into ice VII is evidenced from the XRD pattern. After freezing, excellent agreement is observed between the pressure measured from the volume of ice VII, the luminescence gauge and the volume of Cu. Good reproducibility of the sample pressure response to a given voltage function driving the piezo-actuator is observed. The quantities \(\Delta \mathrm{P}\) and \(\Delta \mathrm{t}\) defined in the text are represented for each run. Note the break in scale for the horizontal axis.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. The time evolution of the integrated XRD image of water freezing under a 0.8 GPa/ms compression rate. Metastable water crystallizes first into ice VII as evidenced by the appearance of its (110) peak at 580 μs. \(\sim 200\) μs later, diffraction peaks attributed to ice VI are observed. The horizontal red and green lines mark the apparition of the ice VII and ice VI respectively. Top: integrated diffraction pattern of the same sample held at the maximum pressure of 1.8 GPa taken after a few minutes. A mixture of ice VI and ice VII is still observed.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. Over-compression ( \(\Delta P\) ) as a function of compression rate (II). Our data points alongside the averaged points from refs. 14–17 are reported. Temperature at freezing is shown as a color-scale. The grey-dashed lines is the best fit result of the power function \(\Delta P = a\Pi^c + b\) , with \(a = 2.6654\) , \(b = 0.0564\) and \(c = -2.2544\) . Inset: Low-pressure phase diagram of \(\mathrm{H}_2\mathrm{O}\) showing how the \(\Delta P\) is calculated for isentropic compressions.
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. Logarithm of the nucleation time as a function of over-compression. The nucleation time is assumed as \(t_{n} = \alpha \Delta t\) , with \(\alpha = 1\) for the figure. Changing \(\alpha\) only changes the constant of the fit. Our data points with averaged points from refs. 14 and 16 are fitted using the CNT model described in the text, with \(\log (t_{min}) = -11.39\) , a=5380.17 and b=245293534.93. Inset: Extrapolation of the CNT model at large over-compression showing an asymptotic limit at the picosecond-timescale. The error bars in pressure are shown while those in time are smaller than the symbol size.
+ +<--- Page Split ---> + +## SUPPLEMENTARY MATERIAL + +### A. Description of the DAC and of the dDAC assembly + +A schematic of the DAC used in this study is presented in figure 7. It consists of a modified LeToullec membrane DAC with four rods to precisely guide the piston. Four set screws have been implemented to finely tune the starting pressure of the sample before setting the DAC in place against the piezo actuator. The piezo actuator has a running course of \(\sim 80 \mu \mathrm{m}\) and can generate forces up to \(35 \mathrm{kN}\) . In this modified version (7B), it has been preloaded to \(3 \mathrm{kN}\) to enhance the response time. + +![](images/Figure_7.jpg) + +
Figure 7. Presentation of the DAC used in this study (A). The DAC is then placed in contact with the piezo actuator (B). The set screws are used to finely adjust the starting pressure.
+ +### B. Pressure response of water at low compression rates + +When compressed at a slow compression rate of \(0.1 \mathrm{GPa / ms}\) , the pressure response of the sample is characterized by two distinctive pressure drops corresponding first to crystallization into ice VI and then to the ice VI - ice VII phase transition. The pressure drop is associated with the volume discontinuity at each transition. + +### C. Metastable melting line of ice VII + +This metastable melting line was measured in a static diamond anvil cell by growing an ice VII crystal in equilibrium with the liquid at high temperature and by decreasing the temperature while adjusting the pressure to metastably maintain the liquid - ice VII + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8. Dynamic compression of liquid water at slow compression rate ( \(\sim 0.1 \mathrm{GPa / ms}\) ). The sample pressure over time is measured using the \(\mathrm{SrB_4O_7}:\mathrm{Sm}^{2 + }\) fluorescence. Two events are observed: the first one associated with the crystallisation of water into ice VI and the second with the solid-solid phase transition of ice VI into ice VII.
+ +equilibrium down to \(26.8^{\circ}\mathrm{C}\) . The melting line was then fitted by a Simon equation \(P = 1.41 + a\left(\left(\frac{T}{278.2}\right)^c - 1\right)\) with \(a = 0.903 \pm 0.005\) and \(c = 2.373 \pm 0.008\) . + +### D. Table data + +The data points used for Figure 5 and Figure 6 are reported in table I. + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 9. Low-pressure phase diagram of \(\mathrm{H}_2\mathrm{O}\) . The metastable melting lines of ice VI and VII have been experimentally measured as described in the text. The points correspond to the observation of freezing as described in the main text. Nucleation of ice VII in the stability domain of ice VI is observed for a compression rate below \(0.4\mathrm{GPa / ms}\) at \(20^{\circ}\mathrm{C}\) . As the compression rate (as the color scale) increases, the direct liquid – ice VII transition is pushed to higher pressures.
+ +<--- Page Split ---> + + +
ΔP(GPa)Rate(GPa/ms)Temp.(K)log(Δt)Ref.
3.2035870787.5480-8.3979[14]
4.198822172180495-[14]
2.8163108400475-[16]
2.941189066.667478-8.3979[17]
2.52100000469-[15]
0.11000000395-[13]
1.34110300-4.9586Our data
1.0514.82300-3.9208Our data
0.664.42300-3.3665Our data
0.541.6300-3.1192Our data
0.190.4300-3.1079Our data
+ +Table I. Data points used in Figure 5 and 6. + +<--- Page Split ---> + +[1] T. C. Hansen, Nature Communications 12, 3161 (2021).[2] T. Loerting, V. Fuentes- Landete, C. M. Tonauer, and T. M. Gasser, Communications Chemistry 3, 109 (2020).[3] B. Kamb and B. L. Davis, PNAS 52, 1433 (1964).[4] V. F. Petrenko and R. W. Whitworth, Physics of Ice (Oxford University Press, Oxford, 1999).[5] T. Bartels- Rausch et al., Rev. Mod. Phys. 84, 885–944 (2012).[6] O. Tschauner et al., Science 359, 1136 (2018).[7] O. Mishima and E. H. Stanley, Nature 396, 329 (1998).[8] P. Gallo et al., Chem. Rev. 116, 7463–7500 (2016).[9] K. Amann- Winkel et al., Rev. Mod. Phys. 88, 011002 (2016).[10] C. A. Tulk, J. Molaison, A. Makhluf, and D. Klug, Nature 569, 542 (2019).[11] R. Bauer, J. S. Tse, K. Komatsu, et al., Nature 585, E9 (2020).[12] K. H. Kim et al., Science 370, 978–982 (2020).[13] A. E. Gleason et al., Phys. Rev. Lett. 119, 025701 (2017).[14] M. C. Marshall et al., Phys. Rev. Lett. 127, 135701 (2021).[15] M. Bastea, S. Mastea, J. E. Reaugh, and D. B. Reisman, Phys. Rev. B 75, 172104 (2007).[16] D. H. Dolan, M. D. Knudson, C. A. Hall, and C. Deeney, Nature Physics 3, 330 (2007).[17] D. H. Dolan and Y. M. Gupta, J. Chem. Phys. 121, 9050 (2004).[18] G. W. Lee, W. J. Evans, and C. S. Yoo, Phys. Rev. B 74, 134112 (2006).[19] J. Y. Chen and C. S. Yoo, PNAS 108, 7685 (2011).[20] A. K. Shargh et al., PNAS 119, e2117281119 (2022).[21] H. K. Mao et al., Rev. of Mod. Phys. 90, 015007 (2018).[22] A. Dewaele, P. Loubeyre, F. Occelli, et al., Nature Communications 9, 2913 (2018).[23] W. J. Evans, C. S. Yoo, G. W. Lee, H. Cynn, M. J. Lipp, and K. Visbeck, Rev Sci Instrum 78, 073904 (2007).[24] Z. Jenei, H. P. Liermann, R. Husband, A. S. J. Méndez, D. Pennicard, H. Marquardt, E. F. O’Bannon, A. Pakhomova, Z. Konopkova, K. Glazyrin, M. Wendt, S. Wenz, E. E. McBride, W. Morgenroth, B. Winkler, A. Rothkirch, M. Hanfland, and W. J. Evans, Rev Sci Instrum 90, 065114 (2019). + +<--- Page Split ---> + +[25] J. Kong, K. Shi, X. Dong, X. Dong, X. Zhang, J. Zhang, L. Su, and G. Yang, Phys. Rev. Lett. 131, 266101 (2023). + +[26] J. S. Smith, S. V. Sinogeikin, C. Lin, E. Rod, L. Bai, and G. Shen, Rev. Sci. Instrum. 86, 072208 (2015). + +[27] M. Cammarata, L. Eybert, F. Ewald, W. Reichenbach, M. Wulff, P. Anfinrud, F. Schotte, A. Plech, Q. Kong, M. Lorenc, B. Lindenau, J. Rábiger, and S. Polachowski, Rev. Sci. Instrum. 80, 015101 (2009). + +[28] M. Sikorski, M. Ramilli, R. de Wijn, V. Hinger, A. Mozzanica, B. Schmitt, H. Han, R. Bean, J. Bielecki, G. Bortel, T. Dietze, G. Faigel, K. Kharitonov, C. Kim, J. C. P. Koliyadu, F. H. M. Koua, R. Letrun, L. M. Lopez, N. Reimers, A. Round, A. Sarma, T. Sato, M. Tegze, and M. Turcato, Frontiers in Physics 11, 10.3389/fphy.2023.1303247 (2023). + +[29] F. Datchi, R. LeToullec, and P. Loubeyre, Journal of Applied Physics 81, 3333 (1997). + +[30] L. Bezacier, B. Journaux, J.- P. Perrillat, H. Cardon, M. Hanfland, and I. Daniel, J. Chem. Phys. 141, 104505 (2014). + +[31] A. Dewaele, P. Loubeyre, and M. Mezouar, Phys. Rev. B 70, 094112 (2004). + +[32] A. Salamat, R. Briggs, P. Bouvier, S. Petitgirard, A. Dewaele, M. E. Cutler, F. Cora, D. Daisenberger, G. Garbarino, and P. F. McMillan, Phys. Rev. B 88, 104104 (2013). + +[33] P. C. Myint, D. M. Sterbentz, J. L. Brown, B. S. Stoltzfus, J.- P. R. Delplanque, and J. L. Belof, Phys. Rev. Lett. 131, 106101 (2023). + +[34] E. Mayer and P. Brüggeller, Nature 298, 715 (1982). + +[35] M. Avrami, The Journal of Chemical Physics 7, 1103 (1939). + +[36] A. Walton, Nucleation, edited by A. Zettlemoyer (Marcel Dekker, New York, 1969). + +[37] J. A. Hernandez, R. Caracas, and S. Labrosse, Nature Communications 13, 3303 (2022). + +[38] C. Prescher and V. B. Prakapenka, High. Press. Res. 35, 223- 230 (2015). + +[39] R. Letoullec, J. P. Pinceaux, and P. Loubeyre, High Pressure Research 1, 77 (1988). + +[40] M. Anson, M. W. McGeoch, and R. C. Smith, The Journal of Chemical Physics 59, 2143 (1973). + +<--- Page Split ---> diff --git a/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846_det.mmd b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6ed1237ce4ba2f12c4af48d386d613f0b7505e67 --- /dev/null +++ b/preprint/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846/preprint__127b4a1df348ee954b5fefb88eeed0676c975caa5b61439cfcd7368eac65b846_det.mmd @@ -0,0 +1,339 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 936, 175]]<|/det|> +# Metastable water at several compression rates and its freezing kinetics into ice VII + +<|ref|>text<|/ref|><|det|>[[44, 196, 277, 240]]<|/det|> +Charles Pépin charles.pepin@cea.fr + +<|ref|>text<|/ref|><|det|>[[44, 270, 455, 288]]<|/det|> +CEA https://orcid.org/0000- 0002- 9638- 3303 + +<|ref|>text<|/ref|><|det|>[[44, 294, 175, 332]]<|/det|> +Ramesh André CEA + +<|ref|>text<|/ref|><|det|>[[44, 340, 185, 379]]<|/det|> +Florent Occelli CEA, DAM, DIF + +<|ref|>text<|/ref|><|det|>[[44, 386, 455, 425]]<|/det|> +Florian Dembele CEA https://orcid.org/0009- 0001- 3626- 9250 + +<|ref|>text<|/ref|><|det|>[[44, 432, 250, 471]]<|/det|> +Aldo Mozzanica Paul Scherrer Institute + +<|ref|>text<|/ref|><|det|>[[44, 478, 448, 517]]<|/det|> +Viktoria Hinger PSI https://orcid.org/0000- 0002- 2616- 4084 + +<|ref|>text<|/ref|><|det|>[[44, 524, 710, 564]]<|/det|> +Matteo Levantino ESRF - The European Synchrotron https://orcid.org/0000- 0002- 1224- 4809 + +<|ref|>text<|/ref|><|det|>[[44, 570, 492, 610]]<|/det|> +Paul Loubeyre CEA/DIF https://orcid.org/0000- 0002- 1778- 3510 + +<|ref|>text<|/ref|><|det|>[[44, 655, 105, 672]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 692, 137, 710]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 729, 300, 748]]<|/det|> +Posted Date: April 23rd, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 767, 475, 786]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4234717/v1 + +<|ref|>text<|/ref|><|det|>[[44, 804, 914, 846]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 865, 535, 884]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 920, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 19th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52576-z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[135, 85, 860, 140]]<|/det|> +# Metastable water at several compression rates and its freezing kinetics into ice VII + +<|ref|>text<|/ref|><|det|>[[158, 161, 836, 210]]<|/det|> +Charles M. Pépin, \(^{1,2,*}\) Ramesh André, \(^{1,2}\) Florent Occelli, \(^{1,2}\) Florian Dembele, \(^{1,2}\) Aldo Mozzanica, \(^{3}\) Viktoria Hinger, \(^{3}\) Matteo Levantino, \(^{4}\) and Paul Loubeyre \(^{1,2}\) + +<|ref|>text<|/ref|><|det|>[[240, 221, 756, 375]]<|/det|> +\(^{1}\) CEA, DAM, DIF, F- 91297 Arpajon, France \(^{2}\) Université Paris- Saclay, CEA, Laboratoire Matière en Conditions Extrêmes, 91680 Bruyères- le- Châtel, France \(^{3}\) PSI, Forschungsstrasse 111, 5232 Villigen, Switzerland \(^{4}\) European Synchrotron Radiation Facility, 71 Avenue des Martyrs, CS 40220, 38043 Grenoble, France + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[452, 86, 541, 106]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[113, 114, 883, 409]]<|/det|> +Water can be dynamically over- compressed well into the stability field of ice VII. Whether water then transforms into ice VII, vitreous ice or a metastable novel crystalline phase remained uncertain. We report here the freezing of over- compressed water to ice VII by time- resolved X- ray diffraction. Quasi- isothermal dynamic compression paths are achieved using a dynamic- piezo- Diamond- Anvil- Cell, with programmable pressure rise time from 0.1 ms to 100 ms. By combining the present data set with those obtained on various ns- dynamical platforms, a complete evolution of the solidification pressure of metastable water versus the compression rate is rationalized within the classical nucleation theory framework. Also, when crystallization into ice VII occurs in between 1.56 GPa and 2.0 GPa, that is in the stability field of ice VI, a structural evolution over few ms is then observed into a mixture of ice VI and ice VII that seems to resolve apparent contradictions between previous results. + +<|ref|>sub_title<|/ref|><|det|>[[139, 466, 310, 484]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[113, 508, 883, 902]]<|/det|> +Water under pressure is both ubiquitous in natural phenomena and a complex system. The quasi- static pressure exploration of \(\mathrm{H}_2\mathrm{O}\) phase diagram, under various temperatures, has revealed a rich polymorphism, with nineteen crystalline forms of ice identified so far [1, 2]. Ice VII is a prominent phase occupying a large region of the phase diagram at pressures above 2 GPa and having the simplest ice structure with a body- centered cubic oxygen sublattice and a proton disordered arrangement [3, 4]. Ice VII is believed to play an important role in the physics of outer planetary bodies [5] and it is now considered a mineral as it has been recovered from the earth mantle as a diamond inclusion [6]. Understanding the structural changes of water under rapid temperature or pressure variation is a topic of current focus, with the observation of metastable phases and implication to impact events on planetary icy bodies. On one hand, rapid cooling of \(\mathrm{H}_2\mathrm{O}\) below its glass transition temperature of 136 K yields an amorphous phase known as low- density amorphous (LDA) ice [7]. A High density amorphous (HDA) ice has also been observed under pressure and the transition between LDA and HDA was speculated to be consistent with the existence of a liquid- liquid transition in supercooled water [8, 9]. However this analogy is still a matter of ongoing debate [10, 11]. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 81, 883, 824]]<|/det|> +The true \(\mathrm{H}_2\mathrm{O}\) liquid- liquid transition could be recently observed by fast isochoric heating of a quenched HDA sample at ambient pressure [12]. On the other hand, many studies have investigated rapidly compressed water. Quasi- isentropic pressure loading platforms, using laser drives [13, 14], magnetic drives [15, 16] and gas guns [17] have been used to investigate the freezing kinetics in over- compressed water. Loading pressure times within the nanosecond timeframe, associated to compression rate covering two orders of magnitude over the different platforms, could then be investigated. Liquid water was observed to persist up to 8 GPa, well into the stability field of ice VII and a freezing into ice VII was therefore speculated. The effects of loading pressure within the millisecond range were investigated using the dynamical diamond anvil cell (d- DAC) [18–20]. The corresponding compression rates could over- compress liquid water up to approximately 2 GPa at most. Freezing into ice VII in the stability field of ice VI was first suspected [18], then evidence of HDA was presented [19] and finally evidence of a mixture between ice VI and HDA was shown [20]. It should be noted that none of these dynamic studies could unambiguously resolve the freezing mechanism of metastable water, due to the absence of a time- resolved microscopic diagnostic. The present study aims to elucidate, at the microscopic level, the solidification process of metastable water under dynamic compression. In particular, we are investigating whether ice VII is still the freezing state of metastable water, and why the transition sequence reverses compared to rapid cooling, i.e. with an amorphous phase at low compression rates and a crystalline phase at high compression rates. Additionally, we aim to consistently interpret the observations across the entire range of compression rates accessible, using d- DAC on one end to gas- gun/laser compression on the other end. To better bridge the gap between static compression and nanosecond- dynamic compression, the d- DAC technique has been optimized to achieve a pressure rise time of about 0.1 ms, allowing four orders of compression rates to be covered on the same d- DAC platform. Structural microscopic changes along the compression path were monitored using time- resolved X- ray diffraction (XRD), complemented by optical imaging and pressure luminescence measurements, all with framing steps down to a few microseconds. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[140, 90, 234, 106]]<|/det|> +## RESULTS + +<|ref|>sub_title<|/ref|><|det|>[[140, 146, 287, 163]]<|/det|> +## Instrumentation + +<|ref|>text<|/ref|><|det|>[[113, 199, 882, 935]]<|/det|> +The experimental setup, depicted in Figure 1, comprises a piezoactuated d- DAC along with three real- time event monitoring diagnostics: X- ray diffraction, pressure luminescence measurements, and optical imaging. The DAC serves as the primary tool for static high- pressure experiments. Leveraging diamond's remarkable transparency across the electromagnetic spectrum, light is utilized to extract properties of the compressed samples. Typically, detailed characterizations of sample properties under pressures up to 100 GPa are conducted, employing primarily synchrotron X- ray diffraction completed by spectroscopy methods over the infra- red to X- ray excitation range [21]. Recently, this characterization has been extended to the TPa range [22]. The concept of operating the DAC dynamically was proposed in 2007, employing voltage- driven piezoactuators to generate pressure [18, 23]. Subsequently, pressure rise times in the 10 ms range were demonstrated, particularly in the study of water [19, 20]. An improved design employing a hollow cylinder piezoactuator has been proposed, offering access to sub- millisecond rise times [24]. Our d- DAC is based on a similar design. As illustrated in Figure 1, the sample pressure response in the sample follows the tailored voltage applied to the piezoactuator. Pressure rise times as fast as 100 us can be achieved. Importantly, the experimental conditions maintain close proximity to isothermal compression, facilitated by the high thermal conductivity of diamond anvils [25]. The equilibration time of pressure in the sample is estimated to be less than 1 us, derived from the reverberation time of compressive waves propagating at the sample's sound velocity across the chamber. The d- DAC retains the optical access of a standard DAC. The primary challenge in detailed measurements during dynamic compression has been to follow the sample's properties with sufficiently small time steps. d- DAC compression experiments initially relied on ms- time resolved micro- photography, ruby luminescence pressure measurement, and Raman spectroscopy [18- 20]. Time- resolved X- ray diffraction measurements in the d- DAC became feasible using the high brilliance, high flux X- ray beams from third- generation synchrotron sources and by the emergence of large hybrid pixel array detectors with very short readout times, such as the EIGER [26] and LAMBDA [24] detectors. Angular X- ray diffraction patterns at repetition rates up to 4 kHz have been reported [24]. In the present study, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 882, 345]]<|/det|> +diffraction patterns at a maximum repetition rate of 250 kHz could be achieved, benefiting significantly from the high flux at 19 KeV in front of the ID09 beamline of the EBS- ESRF synchrotron [27] and the operation of the JUNGFRAU detector in burst mode (limited to 16 images)[28]. Pressure luminescence measurements and imaging of the sample could be performed within a similar timeframe, though not simultaneously with XRD. Typical XRD patterns and luminescence spectra are shown in Figure 1. The pressure was measured using the samarium- doped strontium borate (SrB4O7 : Sm2+) luminescence pressure gauge [29] and using XRD data from the Birch- Murnaghan equation of state (EoS) of ice VII [30] and the Vinet EoS of Cu [31]. A high reproducibility of the sample pressure evolution under a given piezoactuator voltage function acting on the piston of the d- DAC is observed. + +<|ref|>sub_title<|/ref|><|det|>[[139, 392, 620, 409]]<|/det|> +## Pressure response of water under compression ramps + +<|ref|>text<|/ref|><|det|>[[112, 435, 882, 932]]<|/det|> +By applying a step load function on the piezo- actuator of the d- DAC, the pressure is expected to undergo a three- stage evolution: initially a gradual increase in pressure, followed by a rapid linear rise, and reaching a plateau. However, if a phase transition occurs, the pressure rise is disrupted by the convolution with the pressure drop associated to the negative volume discontinuity at the transition, resulting in an apparent negative compressibility. This phenomenon is depicted in Figure 2 and in Figure 8 (Supplementary Materials) when compressing water from its liquid phase across the stability domains of ice VI and ice VII. In figure Figure 8, with a compression rate of \(\sim 0.1\) GPa/ms, two events are observed: the first one associated with the crystallisation of water into ice VI and the second with the structural transition of ice VI into ice VII. If the over- compression of the low- pressure phase within the stability field of the high pressure phase is not excessive, the pressure drop aligns with the pressure of the phase transition. A pressure plateau at the melting pressure is even observed if the compression rate is slower than the crystal growth rate. As depicted in Figure 2, when crystallization occurs at higher compression rates \((\geq 0.4 \text{GPa / ms})\) , the pressure drops to 1.56 GPa, followed by the observation of a pressure plateau. Time- resolved photo- micrographs correlate optical changes in the sample with deviations from the expected linear pressure rise. These deviations begin when an homogeneous nucleation occurs, while the subsequent plateau at 1.56 GPa corresponds to crystal growth. Once the sample chamber is again filled with an ice crystal, a linear pressure rise resumes, with no further pressure + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 85, 882, 371]]<|/det|> +rise anomalies observed. Notably, he solid- liquid equilibrium at 1.56 GPa corresponds to the melting point of ice VII (obtained from the melting curve of ice VII metastably extended below the Ice VI- ice VII- liquid triple point, see Supplementary Materials). Similarly, a stress release phenomenon, resulting from the interplay of compression and phase transition, was noted in the VISAR traces of the ns- dynamical quasi- isentropic compression experiments [14- 17], being interpreted as the signature of a phase transition. This characteristic was termed a kinetic van der Waals loop [15]. Its observation implied that a first order phase transition was taking place, even for the highest compression rate achieved with laser shocks, thus ruling out the transformation of super- compressed water into an amorphous phase via a glass phase transition. It was suggested that crystallization occurs into ice VII, although without time- resolved XRD measurements this proposition remains to be proven. + +<|ref|>text<|/ref|><|det|>[[113, 380, 882, 824]]<|/det|> +In this study, over- compressed water has been investigated for various compression ramps, ranging from 0.1 ms to 20 ms in pressure rise time. The compression peak was maintained below 20 GPa, so the associated compression rates varied from 0.4 GPa/ms to 110 GPa/ms. The solidification of metastable water into ice VII was always observed by time resolved XRD. Figure 3 displays typical pressure responses of the water sample for two measurements conducted at slow and fast compression rates. Sample pressure- versus- time evolution for each compression was consistently measured using various pressure gauges: X- ray diffraction of a small metal flake (Sn or Cu) embedded in water, using their Vinet Equation of State (EoS) [31, 32]; XRD volume of ice VII post metastable water crystallization, using the Birch- Murnaghan EoS of ice VII [30]; and the strontium- doped- borate luminescence gauge [29]. An excellent agreement between the different pressure determinations was observed, along with high reproducibility of the pressure evolution. In all time- resolved compression ramps, a disruption in the pressure rise is observed correlated to the nucleation of metastable water to ice VII, as evidence by the appearance of XRD (110) and (111) peaks of ice VII. At low compression rates, a kinetic pressure loop with apparent negative compressibility can be noted while at higher rates, this feature transitions more towards a slope rupture, probably because the framing time of the measurements is too long to capture the pressure dip. + +<|ref|>text<|/ref|><|det|>[[114, 831, 881, 932]]<|/det|> +As shown in Figures 2 and 3, two quantities are defined for future data analysis: \(\Delta t\) , representing the duration over which fluid water remained in its metastable state along a given pressure ramp; \(\Delta P\) , indicating the over- compression pressure, i.e., the pressure differential between crystallization and the associated ice VII melting at the same temperature. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[140, 88, 630, 107]]<|/det|> +## 1. Crystallization of ice VII in the stability domain of ice VI + +<|ref|>text<|/ref|><|det|>[[112, 145, 883, 933]]<|/det|> +Between compression rates of 0.15 GPa/ms and 0.9 GPa/ms, water remains metastable up to pressures ranging from 1.56 GPa and 2.1 GPa, respectively. This pressure range corresponds to the previous domain of investigation in d- DAC experiments focusing on the solidification of metastable water. However, the findings from these studies appear contradictory. One study suggested the solidification into ice VII [18], supported by a plateau around 1.5 GPa in the pressure response of the sample, similarly to that depicted in Figure 2. Another study proposed that ice VII germinating in the stability field of ice VI evolves into amorphous HDA ice, based on the Raman measurements taken at the end of the compression [19]. In a third study, a mixture of ice VI and vitreous ice was observed by Raman and XRD measurements taken at the end of the pressure ramp [20]. The situation is puzzling and prompts the need for time resolved XRD measurement to elucidate the structure of the nucleated ice and the mechanism of its growth and transformation over time. Time resolved XRD measurements of metastable water crystallizing at 1.75 GPa are shown in Figure 4. After azimuthal integration, the time resolved XRD data is represented as a 2D plot with diffraction angle on the x- axis, time on the y- axis and intensity displayed using a grey color scale. The XRD signal of ice VII (evidenced by the strong (110) peak) emerges first, followed by the appearance of the ice VI XRD pattern (evidenced by its 3 main peaks) after 200 us, with its intensity increasing over time. After maintaining the sample at its peak pressure of 1.8 GPa for a few minutes, the integrated diffraction pattern (Figure 4, top) indicates a mixture of ice VI and ice VII. The present observations support the hypothesis that metastable water, within the pressure range of 1.5 GPa to 2.1 GPa, nucleates into ice VII first, despite this range being the stability field of ice VI. Ice VI nucleation is also observed after a few hundred microseconds, revealing the difference in nucleation time between ice VI and ice VII. In this pressure domain, where both phases are stable, the faster nucleation of ice VII could be attributed to a similarity of the local order between the liquid and ice VII, resulting in a lower interfacial energy and thus a lower nucleation barrier, leading to earlier nucleation [18]. Here, the ice sample remaining as a mixture of ice VI and ice VII is observed at 1.8 GPa. In another XRD study, a mixture of ice VI and amorphous ice was reported at 1.7 GPa[20]. This difference could be explained by the fact that at 1.8 GPa, ice VII is approaching its thermodynamic stability, allowing for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 87, 880, 160]]<|/det|> +the stabilization of a mixture of ice VI and ice VII, whereas at 1.7 GPa, ice VII is gradually transforming, leading to the stabilization of a mixture of ice VI and amorphous ice. These results offer a clarification to the apparent contradiction of the previous findings. + +<|ref|>sub_title<|/ref|><|det|>[[140, 206, 656, 223]]<|/det|> +## 2. Freezing pressure of metastable water versus compression rate + +<|ref|>text<|/ref|><|det|>[[112, 247, 882, 880]]<|/det|> +The pressure limit of metastable water is here observed to continuously increase with the compression rate, by scanning the compression rate over 4 orders of magnitude in the d- DAC. Water can be over- compressed up to 2.9 GPa at 300 K for a compression rate of 110 GPa/ms. This value is much lower than the pressure limit of metastable water in ns- dynamic measurements [14] which has been reaching 8 GPa for compression rate about 4 orders of magnitude higher than the one in d- DAC. Given that the compression ramp is isothermal in the d- DAC while it is isentropic in the ns- compression ramps, a consistent metric for the overpressure value is necessary. The differential pressure \(\Delta P\) , as previously defined, is used. For the isentropic ramps, the measured pressure limit of metastable water, \(P_{lim}\) , is converted into \(\Delta P\) by computing the difference between \(P_{lim}\) and the associated ice VII melting pressure at the temperature of the isentrope at \(P_{lim}\) , as illustrated in the inset of Figure 5. Figure 5 plots in a semi- logarithmic scale the metastability pressure limit of water as a function of compression rate, obtained here using the d- DAC and utilizing data from other various platforms [14- 17] (excluding the data of Gleason et al. due to heterogeneous nucleation [13]), so spanning six orders of compression rates. Remarkably, the various datasets align smoothly on an increasing trend. Recently, Myint et al. have proposed a scaling law for the onset of solidification at extreme over- compression of liquids [33], with \(\Delta P\) being a power function of the compression rate II. We have fitted the dataset in Figure 5 with a similar power function \(\Delta P = a\Pi^{c} + b\) . Notably, the exponent \(c\) of the fit equals to 0.056, which matches the one given by the Myint's scaling law, equal to 0.069. The validity of such a scaling law is thus extended over the whole range of compression rates leading to freezing of metastable water into ice VII. It should be noted though that the constant \(b\) is added here to take into account the need of a minimal compression rate above which freezing of metastable water into ice VII is observed. + +<|ref|>text<|/ref|><|det|>[[115, 886, 880, 931]]<|/det|> +Over- pressurization is conceptually equivalent to isobaric cooling. When metastable water freezes under a compression rate of 110 GPa/ms, it corresponds to a cooling rate of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 85, 880, 160]]<|/det|> +\(10^{7}\mathrm{K / s}\) . Such a rate has been utilized to observe the vitrification of water by spraying water in a cryogenic medium [34]. In contrast here, a solidification into ice VII through homogeneous nucleation is observed. + +<|ref|>sub_title<|/ref|><|det|>[[138, 196, 269, 214]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[112, 238, 882, 602]]<|/det|> +During the freezing of metastable water, we observe homogeneous nucleation of ice VII which uniformly fills the entire sample chamber, across the range of the compression rates investigated here with the d- DAC. Consequently, we analyze the freezing kinetics using the classical Johnson- Mehl- Avrami- Kolmogorov (JMAK) theory. Within this framework, the freezing kinetics incorporates contributions from the nucleation rate \((\gamma)\) and the growth velocity of the ice germs \((v)\) , which are the two dominant mechanisms to obtain the kinetic time scale of freezing \((\tau_{F})\) , as: \(\tau_{F} \propto (\gamma v^{3})^{- 1 / 4}\) [35]. In highly metastable states, the nucleation rate predominates over the growth rate, thereby controlling the freezing kinetics. That is evidenced by the absence of a plateau in the pressure- time evolution in the d- DAC for compression rates above a few GPa/ms. According to Classical Nucleation Theory (CNT), the nucleation rate is proportional to the probability of spontaneously overcoming a free energy barrier associated with the formation of a critical- nucleus ice crystallite, expressed in terms of \(\Delta g\) , the Gibbs free energy difference between the liquid and the solid. The nucleation time and hence the freezing time can be expressed as: + +<|ref|>equation<|/ref|><|det|>[[428, 607, 876, 655]]<|/det|> +\[t_{n} = \frac{\exp\left(\frac{B}{\Delta g^{2}}\right)}{A} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[113, 663, 881, 791]]<|/det|> +where \(A\) and \(B\) are parameters specific to the nucleation process [36]. \(\Delta g\) can be calculated from the equation of state of solid ice VII and that of liquid extrapolated in the metastable domain. Since water and ice are quite incompressible, we estimate the volume variation over the pressure range covered here assuming a constant bulk modulus (2), that is those of the liquid [18] and the solid [30] calculated at the melting pressure. + +<|ref|>equation<|/ref|><|det|>[[332, 796, 876, 833]]<|/det|> +\[\frac{\partial V}{\partial P} \approx -KV_{0} \implies V(P) = V_{0} - KV_{0}P \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[113, 841, 880, 888]]<|/det|> +\(\Delta g\) is then straightforwardly estimated in terms of the over- compression \(\Delta P\) of the liquid and of the liquid- solid volume difference \(\Delta V_{l - s}\) by: + +<|ref|>equation<|/ref|><|det|>[[260, 893, 876, 933]]<|/det|> +\[\Delta g = \int \Delta V_{l - s} dP = a + \Delta V_{0_{l - s}} \Delta P - \frac{1}{2} \Delta (KV_{0})_{l - s} \Delta P^{2} \quad (3)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 86, 730, 107]]<|/det|> +From which the freezing time can be related to \(\Delta P\) through the relation: + +<|ref|>equation<|/ref|><|det|>[[247, 118, 877, 161]]<|/det|> +\[log(t_{n})\propto log(t_{min}) + \frac{b}{(a + \Delta V_{0_{l - s}}\Delta P - \frac{1}{2}\Delta (KV_{0})_{l - s}\Delta P^{2})^{2}} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[113, 171, 881, 351]]<|/det|> +This phenomenological model serves as a fitting function to correlate the freezing time of metastable water to the metastable over- pressure. In the dynamical compression of water described above, the freezing nucleation time is intertwined with the duration of the over- compression, \(\Delta t\) , with the nucleation time being \(t_{n} = \alpha \Delta t\) . We assume that \(\alpha\) remains constant across the different experimental data, from the d- DAC to the ns- isentropic compressions. The freezing time is plotted against the over- compression in Figure 6, where the proposed fit form in equation 4 accurately reproduces the entire dataset. + +<|ref|>text<|/ref|><|det|>[[113, 356, 882, 667]]<|/det|> +The extrapolation of the fit to larger over- compression is shown in the inset of Figure 6, revealing an intriguing asymptotic behavior at high over- compression. Beyond a \(\Delta P\) of 10 GPa, the predicted freezing time of water is only a few picoseconds. Although such experimental over- compression seems unfeasible, numerical simulations are seen to validate the asymptotic behavior. Numerically, the thermodynamic conditions of metastable water can be established in less than a ps, and a freezing time of 12 ps has been reported for an over- compression of water about 30 GPa [37]. With increasing compression rate, water tends to reach a higher over- compressed metastable state. It has been suggested, though, that there exists a pressure limit close to 11 GPa for observing metastable water [14], beyond which freezing should occur via mechanisms such as spinodal decomposition or amorphization. In contrast, we do not propose a pressure limit for metastable water here but instead suggest a freezing time limit of a few ps. + +<|ref|>text<|/ref|><|det|>[[113, 673, 882, 931]]<|/det|> +It is assumed that the dynamic compression experiments from refs. 16 and 14 resulted in a crystalline sample of ice VII. This assumption is reinforced by the coherence observed between the current d- DAC dataset and the previously obtained nanosecond- isentropic compression data.. However, a clear confirmation through time- resolved XRD under isentropic compression is warranted, possibly utilizing a free electron- laser X- ray diffraction measurement, as in ref.13, but assuring the conditions of homogeneous nucleation. Finally, the present study illustrates the remarkable ability of using the d- DAC to vary dynamic phenomena over four orders of magnitude of compression rate on the same platform, while simultaneously enabling detailed measurements of property evolution through time- resolved visual observation and XRD measurements. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[138, 90, 247, 106]]<|/det|> +## METHODS + +<|ref|>text<|/ref|><|det|>[[112, 135, 883, 797]]<|/det|> +X- ray diffraction measurements: X- ray diffraction experiments were carried out at the ID09 beamline of the ESRF. Data were collected using a 19 keV X- ray pink beam (1.5% bandwidth) focused to a 30 μm (FWHM) round spot. A 100 μm pinhole a few centimeters away from the sample was used to remove any contribution from the tails of the beam and from air scattering. The beamline heat- load chopper was operated at its slowest rotation speed in order to obtain X- ray pulses up to 1.5 ms, time long enough to cover the dynamics of interest. Diffraction images were collected using a JUNGFRAU 1M detector. The detector, composed of 2 tiles with 500k pixels each, has a 75 μm pixel pitch and uses a 320 μm thick silicon sensor. The detector was horizontally offset with respect to the direct X- ray beam to avoid radiation damage. It was operated in storage cell (SC) mode, where a burst of up to 16 images can be collected quickly and stored on a local, in pixel, analog memory for a later read- out. The maximum achievable frame rate during the burst is 250 kHz, with 1.5 μs exposures; however, the timing is adaptable, allowing for the utilization of longer periods and exposures if needed. In this study, the exposure time was typically 7 μs. The detector was controlled by the beamline control software via a custom TANGO device. Beside the run themselves, detector control included dark frames collection routines, which were individual per storage cell and timing setup. The timing of the measurement was orchestrated by the beamline control hardware, which would, in sync with the heat- load chopper, open the millisecond shutter to let the 1.5 ms long X- ray pulse on the sample, trigger the detector burst and the dDAC piezo controller. Sample- detector distance ( \(\sim 170\) mm), tilt, and rotation were calibrated using a reference Bi sample with DIOPTAS [38]. The two- dimensional images from the JUNGFRAU detector were corrected for offset and gain, and then radially integrated using DIOPTAS. Individual peaks of both water and the X- ray gauge were fitted with a PseudoVoigt function for the determination of the lattice parameters. + +<|ref|>text<|/ref|><|det|>[[114, 805, 881, 932]]<|/det|> +\(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fluorescence measurements: \(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fuorescence signal was collected in a back- scattering geometry by an iXon electron multiplying (EM) CCD coupled to a 1200 l/mm grating, allowing a spectral resolution of approximately 0.08 nm. The iXon EMCCD is used in Fast- Kinetics mode, allowing the recording of spectra with an acquisition time varying from 1 to 500 μs with a time interval of \(\sim 8\) μs between each + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 85, 881, 187]]<|/det|> +spectrum. \(\mathrm{SrB_4O_7:Sm^{2 + }}\) fluorescence excitation was performed using \(10\mathrm{mW}\) of a 491 nm Cobolt laser. With nonradiative decay times for \(\mathrm{Sm^{2 + }}\) ions typically in the range of a few nanoseconds [40], \(\mathrm{SrB_4O_7:Sm^{2 + }}\) emerges as an ideal pressure gauge for time- resolved experiments with a microsecond- level time resolution. + +<|ref|>text<|/ref|><|det|>[[113, 193, 882, 399]]<|/det|> +d- DAC experiments: Dynamic compression experiments were performed using a LeToullec DAC, i.e a modification of the membrane- DAC [39] with a piston guided by four rods, in conjunction with the Piezosystem Jena HPST 1000/35- 25/80 VS45. A waveform Tabor Electronics WW2074 and an Amplifier Piezosystem Jena RCV 1000/7 are used to deliver the current in the piezoactuator with a chosen profile. DACs were equipped with Boelher- Almax design diamond anvils with \(300\mu \mathrm{m}\) culets. Stainless steel gaskets were used for all experiments, indented to \(40\mu \mathrm{m}\) thickness with a \(100\mu \mathrm{m}\) - diameter hole for the experimental chamber. + +<|ref|>sub_title<|/ref|><|det|>[[140, 434, 608, 453]]<|/det|> +## ACKNOWLEDGEMENTS (NOT COMPULSORY) + +<|ref|>text<|/ref|><|det|>[[114, 477, 881, 550]]<|/det|> +The authors acknowledge the European Synchrotron Radiation Facility for provision of synchrotron radiation facilities on beamline ID09, during beam time allocated to the proposal HC- 4895. + +<|ref|>sub_title<|/ref|><|det|>[[140, 588, 554, 607]]<|/det|> +## AUTHOR CONTRIBUTIONS STATEMENT + +<|ref|>text<|/ref|><|det|>[[114, 631, 881, 756]]<|/det|> +P.L. and C.P. conceived the experiment; R.A., F.O. and C.P. developed the d- DAC; All authors participated to the ESRF HC- 4895 experiment; F.D. and C.P. performed complementary in- house experiments; A.M. and V.H. operated the Jungfrau detector; M.L. operated the beamline; C.P. processed the XRD data; C.P. and P.L. analyzed the data and wrote the manuscript; All authors revised the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[140, 794, 438, 812]]<|/det|> +## ADDITIONAL INFORMATION + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 88, 880, 415]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 438, 882, 595]]<|/det|> +
Figure 1. Schematic of the experimental configuration. X-ray diffraction spectra are recorded as a function of time on a JUNGFRAU 1M detector in a Debye-Scherrer geometry (left) and are azimuthally integrated into an XRD pattern where the diffraction peaks of ice VII are clearly measured in \(20 \mu \mathrm{s}\) (bottom left). \(\mathrm{SrB_4O_7:Sm^{2 + }}\) fluorescence over time is recorded on an iXon EMCCD (right) alongside images on a Photron Fast Camera. A typical system pressure response (red) to the input voltage (blue) are plotted versus time (upper right).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[195, 92, 799, 454]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 480, 883, 692]]<|/det|> +
Figure 2. Dynamic compression of liquid water at \(0.4\mathrm{GPa / ms}\) . The sample pressure over time is measured using the \(\mathrm{SrB_4O_7:Sm^{2 + }}\) -chip fluorescence while micro-photographs are recorded at \(40\mathrm{kHz}\) framing time. At the liquid-solid phase transition a characteristic pressure drop due the density difference is observed. Visual observation with the ultra-fast camera reveals a homogeneous nucleation in the sample volume. The over-compression pressure, \(\Delta \mathrm{P}\) , is defined as the pressure differential between crystallization and the associated ice VII melting at the same temperature (1.56 GPa). The quantity \(\Delta t\) represents the corresponding duration over which fluid water remains in its metastable state.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[192, 90, 790, 528]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 551, 883, 790]]<|/det|> +
Figure 3. Measured pressure as a function of time for compression rates of 1.6 GPa/ms and 110 GPa/ms. At the freezing pressure into ice VII, a rupture in the compression slope is observed due to ,the negative discontinuity at crystallization. Pressure is measured using the \(\mathrm{SrB_4O_7}:\mathrm{Sm^{2 + }}\) fluorescence gauge and in some cases using the volume of a piece of Cu embedded in the sample chamber. The freezing into ice VII is evidenced from the XRD pattern. After freezing, excellent agreement is observed between the pressure measured from the volume of ice VII, the luminescence gauge and the volume of Cu. Good reproducibility of the sample pressure response to a given voltage function driving the piezo-actuator is observed. The quantities \(\Delta \mathrm{P}\) and \(\Delta \mathrm{t}\) defined in the text are represented for each run. Note the break in scale for the horizontal axis.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[226, 87, 768, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 750, 882, 907]]<|/det|> +
Figure 4. The time evolution of the integrated XRD image of water freezing under a 0.8 GPa/ms compression rate. Metastable water crystallizes first into ice VII as evidenced by the appearance of its (110) peak at 580 μs. \(\sim 200\) μs later, diffraction peaks attributed to ice VI are observed. The horizontal red and green lines mark the apparition of the ice VII and ice VI respectively. Top: integrated diffraction pattern of the same sample held at the maximum pressure of 1.8 GPa taken after a few minutes. A mixture of ice VI and ice VII is still observed.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[188, 85, 805, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 584, 881, 712]]<|/det|> +
Figure 5. Over-compression ( \(\Delta P\) ) as a function of compression rate (II). Our data points alongside the averaged points from refs. 14–17 are reported. Temperature at freezing is shown as a color-scale. The grey-dashed lines is the best fit result of the power function \(\Delta P = a\Pi^c + b\) , with \(a = 2.6654\) , \(b = 0.0564\) and \(c = -2.2544\) . Inset: Low-pressure phase diagram of \(\mathrm{H}_2\mathrm{O}\) showing how the \(\Delta P\) is calculated for isentropic compressions.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[188, 87, 800, 541]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 564, 883, 720]]<|/det|> +
Figure 6. Logarithm of the nucleation time as a function of over-compression. The nucleation time is assumed as \(t_{n} = \alpha \Delta t\) , with \(\alpha = 1\) for the figure. Changing \(\alpha\) only changes the constant of the fit. Our data points with averaged points from refs. 14 and 16 are fitted using the CNT model described in the text, with \(\log (t_{min}) = -11.39\) , a=5380.17 and b=245293534.93. Inset: Extrapolation of the CNT model at large over-compression showing an asymptotic limit at the picosecond-timescale. The error bars in pressure are shown while those in time are smaller than the symbol size.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[140, 88, 448, 107]]<|/det|> +## SUPPLEMENTARY MATERIAL + +<|ref|>sub_title<|/ref|><|det|>[[140, 133, 645, 153]]<|/det|> +### A. Description of the DAC and of the dDAC assembly + +<|ref|>text<|/ref|><|det|>[[113, 177, 882, 330]]<|/det|> +A schematic of the DAC used in this study is presented in figure 7. It consists of a modified LeToullec membrane DAC with four rods to precisely guide the piston. Four set screws have been implemented to finely tune the starting pressure of the sample before setting the DAC in place against the piezo actuator. The piezo actuator has a running course of \(\sim 80 \mu \mathrm{m}\) and can generate forces up to \(35 \mathrm{kN}\) . In this modified version (7B), it has been preloaded to \(3 \mathrm{kN}\) to enhance the response time. + +<|ref|>image<|/ref|><|det|>[[189, 342, 804, 500]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 526, 880, 575]]<|/det|> +
Figure 7. Presentation of the DAC used in this study (A). The DAC is then placed in contact with the piezo actuator (B). The set screws are used to finely adjust the starting pressure.
+ +<|ref|>sub_title<|/ref|><|det|>[[138, 633, 653, 653]]<|/det|> +### B. Pressure response of water at low compression rates + +<|ref|>text<|/ref|><|det|>[[113, 677, 881, 777]]<|/det|> +When compressed at a slow compression rate of \(0.1 \mathrm{GPa / ms}\) , the pressure response of the sample is characterized by two distinctive pressure drops corresponding first to crystallization into ice VI and then to the ice VI - ice VII phase transition. The pressure drop is associated with the volume discontinuity at each transition. + +<|ref|>sub_title<|/ref|><|det|>[[139, 815, 489, 834]]<|/det|> +### C. Metastable melting line of ice VII + +<|ref|>text<|/ref|><|det|>[[113, 858, 881, 932]]<|/det|> +This metastable melting line was measured in a static diamond anvil cell by growing an ice VII crystal in equilibrium with the liquid at high temperature and by decreasing the temperature while adjusting the pressure to metastably maintain the liquid - ice VII + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[186, 85, 802, 541]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 562, 881, 664]]<|/det|> +
Figure 8. Dynamic compression of liquid water at slow compression rate ( \(\sim 0.1 \mathrm{GPa / ms}\) ). The sample pressure over time is measured using the \(\mathrm{SrB_4O_7}:\mathrm{Sm}^{2 + }\) fluorescence. Two events are observed: the first one associated with the crystallisation of water into ice VI and the second with the solid-solid phase transition of ice VI into ice VII.
+ +<|ref|>text<|/ref|><|det|>[[114, 686, 880, 735]]<|/det|> +equilibrium down to \(26.8^{\circ}\mathrm{C}\) . The melting line was then fitted by a Simon equation \(P = 1.41 + a\left(\left(\frac{T}{278.2}\right)^c - 1\right)\) with \(a = 0.903 \pm 0.005\) and \(c = 2.373 \pm 0.008\) . + +<|ref|>sub_title<|/ref|><|det|>[[138, 770, 280, 788]]<|/det|> +### D. Table data + +<|ref|>text<|/ref|><|det|>[[138, 813, 736, 834]]<|/det|> +The data points used for Figure 5 and Figure 6 are reported in table I. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[192, 100, 802, 563]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 587, 882, 716]]<|/det|> +
Figure 9. Low-pressure phase diagram of \(\mathrm{H}_2\mathrm{O}\) . The metastable melting lines of ice VI and VII have been experimentally measured as described in the text. The points correspond to the observation of freezing as described in the main text. Nucleation of ice VII in the stability domain of ice VI is observed for a compression rate below \(0.4\mathrm{GPa / ms}\) at \(20^{\circ}\mathrm{C}\) . As the compression rate (as the color scale) increases, the direct liquid – ice VII transition is pushed to higher pressures.
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[265, 88, 728, 420]]<|/det|> + +
ΔP(GPa)Rate(GPa/ms)Temp.(K)log(Δt)Ref.
3.2035870787.5480-8.3979[14]
4.198822172180495-[14]
2.8163108400475-[16]
2.941189066.667478-8.3979[17]
2.52100000469-[15]
0.11000000395-[13]
1.34110300-4.9586Our data
1.0514.82300-3.9208Our data
0.664.42300-3.3665Our data
0.541.6300-3.1192Our data
0.190.4300-3.1079Our data
+ +<|ref|>table_caption<|/ref|><|det|>[[322, 432, 671, 447]]<|/det|> +Table I. Data points used in Figure 5 and 6. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 145, 883, 930]]<|/det|> +[1] T. C. Hansen, Nature Communications 12, 3161 (2021).[2] T. Loerting, V. Fuentes- Landete, C. M. Tonauer, and T. M. Gasser, Communications Chemistry 3, 109 (2020).[3] B. Kamb and B. L. Davis, PNAS 52, 1433 (1964).[4] V. F. Petrenko and R. W. Whitworth, Physics of Ice (Oxford University Press, Oxford, 1999).[5] T. Bartels- Rausch et al., Rev. Mod. Phys. 84, 885–944 (2012).[6] O. Tschauner et al., Science 359, 1136 (2018).[7] O. Mishima and E. H. Stanley, Nature 396, 329 (1998).[8] P. Gallo et al., Chem. Rev. 116, 7463–7500 (2016).[9] K. Amann- Winkel et al., Rev. Mod. Phys. 88, 011002 (2016).[10] C. A. Tulk, J. Molaison, A. Makhluf, and D. Klug, Nature 569, 542 (2019).[11] R. Bauer, J. S. Tse, K. Komatsu, et al., Nature 585, E9 (2020).[12] K. H. Kim et al., Science 370, 978–982 (2020).[13] A. E. Gleason et al., Phys. Rev. Lett. 119, 025701 (2017).[14] M. C. Marshall et al., Phys. Rev. Lett. 127, 135701 (2021).[15] M. Bastea, S. Mastea, J. E. Reaugh, and D. B. Reisman, Phys. Rev. B 75, 172104 (2007).[16] D. H. Dolan, M. D. Knudson, C. A. Hall, and C. Deeney, Nature Physics 3, 330 (2007).[17] D. H. Dolan and Y. M. Gupta, J. Chem. Phys. 121, 9050 (2004).[18] G. W. Lee, W. J. Evans, and C. S. Yoo, Phys. Rev. B 74, 134112 (2006).[19] J. Y. Chen and C. S. Yoo, PNAS 108, 7685 (2011).[20] A. K. Shargh et al., PNAS 119, e2117281119 (2022).[21] H. K. Mao et al., Rev. of Mod. Phys. 90, 015007 (2018).[22] A. Dewaele, P. Loubeyre, F. Occelli, et al., Nature Communications 9, 2913 (2018).[23] W. J. Evans, C. S. Yoo, G. W. Lee, H. Cynn, M. J. Lipp, and K. Visbeck, Rev Sci Instrum 78, 073904 (2007).[24] Z. Jenei, H. P. Liermann, R. Husband, A. S. J. Méndez, D. Pennicard, H. Marquardt, E. F. O’Bannon, A. Pakhomova, Z. Konopkova, K. Glazyrin, M. Wendt, S. Wenz, E. E. McBride, W. Morgenroth, B. Winkler, A. Rothkirch, M. Hanfland, and W. J. Evans, Rev Sci Instrum 90, 065114 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 883, 135]]<|/det|> +[25] J. Kong, K. Shi, X. Dong, X. Dong, X. Zhang, J. Zhang, L. Su, and G. Yang, Phys. Rev. Lett. 131, 266101 (2023). + +<|ref|>text<|/ref|><|det|>[[113, 142, 881, 190]]<|/det|> +[26] J. S. Smith, S. V. Sinogeikin, C. Lin, E. Rod, L. Bai, and G. Shen, Rev. Sci. Instrum. 86, 072208 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 197, 881, 273]]<|/det|> +[27] M. Cammarata, L. Eybert, F. Ewald, W. Reichenbach, M. Wulff, P. Anfinrud, F. Schotte, A. Plech, Q. Kong, M. Lorenc, B. Lindenau, J. Rábiger, and S. Polachowski, Rev. Sci. Instrum. 80, 015101 (2009). + +<|ref|>text<|/ref|><|det|>[[113, 279, 881, 383]]<|/det|> +[28] M. Sikorski, M. Ramilli, R. de Wijn, V. Hinger, A. Mozzanica, B. Schmitt, H. Han, R. Bean, J. Bielecki, G. Bortel, T. Dietze, G. Faigel, K. Kharitonov, C. Kim, J. C. P. Koliyadu, F. H. M. Koua, R. Letrun, L. M. Lopez, N. Reimers, A. Round, A. Sarma, T. Sato, M. Tegze, and M. Turcato, Frontiers in Physics 11, 10.3389/fphy.2023.1303247 (2023). + +<|ref|>text<|/ref|><|det|>[[113, 389, 833, 410]]<|/det|> +[29] F. Datchi, R. LeToullec, and P. Loubeyre, Journal of Applied Physics 81, 3333 (1997). + +<|ref|>text<|/ref|><|det|>[[113, 417, 880, 464]]<|/det|> +[30] L. Bezacier, B. Journaux, J.- P. Perrillat, H. Cardon, M. Hanfland, and I. Daniel, J. Chem. Phys. 141, 104505 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 471, 753, 492]]<|/det|> +[31] A. Dewaele, P. Loubeyre, and M. Mezouar, Phys. Rev. B 70, 094112 (2004). + +<|ref|>text<|/ref|><|det|>[[113, 498, 880, 546]]<|/det|> +[32] A. Salamat, R. Briggs, P. Bouvier, S. Petitgirard, A. Dewaele, M. E. Cutler, F. Cora, D. Daisenberger, G. Garbarino, and P. F. McMillan, Phys. Rev. B 88, 104104 (2013). + +<|ref|>text<|/ref|><|det|>[[113, 553, 880, 601]]<|/det|> +[33] P. C. Myint, D. M. Sterbentz, J. L. Brown, B. S. Stoltzfus, J.- P. R. Delplanque, and J. L. Belof, Phys. Rev. Lett. 131, 106101 (2023). + +<|ref|>text<|/ref|><|det|>[[113, 608, 568, 628]]<|/det|> +[34] E. Mayer and P. Brüggeller, Nature 298, 715 (1982). + +<|ref|>text<|/ref|><|det|>[[113, 636, 631, 656]]<|/det|> +[35] M. Avrami, The Journal of Chemical Physics 7, 1103 (1939). + +<|ref|>text<|/ref|><|det|>[[113, 663, 808, 683]]<|/det|> +[36] A. Walton, Nucleation, edited by A. Zettlemoyer (Marcel Dekker, New York, 1969). + +<|ref|>text<|/ref|><|det|>[[113, 690, 845, 710]]<|/det|> +[37] J. A. Hernandez, R. Caracas, and S. Labrosse, Nature Communications 13, 3303 (2022). + +<|ref|>text<|/ref|><|det|>[[113, 717, 728, 737]]<|/det|> +[38] C. Prescher and V. B. Prakapenka, High. Press. Res. 35, 223- 230 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 744, 812, 765]]<|/det|> +[39] R. Letoullec, J. P. Pinceaux, and P. Loubeyre, High Pressure Research 1, 77 (1988). + +<|ref|>text<|/ref|><|det|>[[113, 771, 880, 819]]<|/det|> +[40] M. Anson, M. W. McGeoch, and R. C. Smith, The Journal of Chemical Physics 59, 2143 (1973). + +<--- Page Split ---> diff --git a/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/images_list.json b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..a0c71ffe24653202e099078e26c11f2709ae9b2c --- /dev/null +++ b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/images_list.json @@ -0,0 +1,317 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Aso volcano system and observations of tilt offset synchronous with LPT. (a) The topographic relief near the Aso caldera is overlaid with collocated seismometer and tiltmeter (open triangles), LPT source in the shallow crack-like conduit (thick bar), the geodetically inferred magma chamber (open circle), and the low-velocity zone (dashed circle). The star marks the Naka-dake first crater. The upper inset displays Aso volcano located in Kyushu island, southwest Japan. The source of the tilt offset is marked as a red cross and it is referred to as a shallow magma storage zone (SMSZ). The beach ball displays the deviatoric component of the source mechanism in SMSZ. (b) Tilt waveforms at four V-net stations near the October 8, 2016 phreatomagmatic eruption. The waveforms are low-pass filtered at 0.05 Hz with a \\(4^{\\text{th}}\\) -order casual Butterworth filter. The linear trend determined in the first 10 minutes is used to remove the background trend. “Event 1” and “Event 2” denote the tilt offset that occurred concurrently with two LPT events. Redline marks the timing of the 2016 eruption.", + "footnote": [], + "bbox": [ + [ + 113, + 88, + 860, + 409 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Global inflation stack [Nov 2011-Aug 2014]", + "footnote": [], + "bbox": [ + [ + 123, + 88, + 480, + 320 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Observations of inflation (a) and deflation (b) waveform stacks against LPT and SPT. Broadband tilt, displacement, and velocity of global waveform stacks at station N.ASHV are presented against LPT velocity waveforms (10-30 s) and SPT envelopes (0.5-1 s). The bootstrap errors are shown in yellow strips. These waveforms and envelopes are aligned against the onset of SPT. To preserve the phase of the long-period waveform (> 50 s), a \\(2^{\\text{nd}}\\) -order acausal Butterworth filter is implemented. We apply a \\(2^{\\text{nd}}\\) -order causal Butterworth filter to preserve the timing of the LPT waveform (10-30 s) and SPT envelope (0.5-1 s).", + "footnote": [], + "bbox": [ + [ + 123, + 321, + 480, + 675 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Source inversion result and a new conceptual framework of magma/gas transport in the plumbing system beneath Aso volcano. (a), (b) and (c) compares data (black arrows) and synthetics (red arrows) from the inverted source location and mechanism. Red cross marks the inverted source location with a confidence level of \\(99\\%\\) . Black, red and blue circles indicate the new Mogi magma chamber40, BYA and BCU magma chambers49, respectively. Dashed circle marks the low-velocity zone39. (d) An east-west cross section showing a conceptual diagram of magma/gas transport system beneath Aso volcano. The inverted source, or the shallow magma storage zone (SMSZ), connects the LPT source in the shallow conduit from above and the magma chamber from below. SPT source is shown directly above the shallow conduit. The inset displays the volumetric and deviatoric components of the focal mechanism, respectively, looking from the south. Blue dashed line indicates the brittle-ductile transition. (e), (f) and (g) display the simplified flow conditions where the SMSZ suffers inflation, null volume change, or", + "footnote": [], + "bbox": [ + [ + 115, + 85, + 861, + 519 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Characterization of the volume/mass change inside the SMSZ during the 2011-2016 eruption cycle (a) The accumulative net volume changes in the SMSZ. Dark inverse triangles mark the onset of the 2014 Strombolian eruption, the 2015 and 2016 phreatomagmatic eruptions, whereas hatched areas mark their eruption episodes. Grey inverse triangles indicate minor eruptions. \\(\\oplus\\) , \\(\\ominus\\) and \\(\\odot\\) mark the episodes associated with observations of inflation event, deflation event and event of negligible offset, respectively (see waveform stacks in the insets). \\(①\\) and \\(②\\) mark the time windows where the mass changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively (see also Fig. 4e). (b) Outgassing potential of the shallow conduit, or equivalently the moment ratio between pressurization and depressurization LPTs \\(^{31}\\) . A lower (higher) outgassing potential corresponds to a higher proportion of pressurization (depressurization) LPT events. (c) The campaign SO \\(_2\\) emission from JMA (https://www.data.jma.go.jp/svd/vois/data/fukuoka/rovdm/Asosan_rovdm/gas/gas.html). (d) The volume change rates estimated by the inflation event (black crosses), the geodetic data (green cross), and", + "footnote": [], + "bbox": [ + [ + 113, + 85, + 861, + 528 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Comparisons of the mass flow rate and decompression rate from petrology and seismology. Open circles and triangles mark the volume estimates from waveform stacks of inflation and deflation events before the 2014 and 2016 eruptions (see details in the main text). Target crosses mark the median volume estimates from the five subsets (Fig. S13a-e). Solid circles mark previous compilations in basaltic volcanoes \\(^5\\) , where the decompression rates are estimated from water diffusion in olivine or embayment studies and the mass discharge rates are estimated from volcanic outputs or plume heights, respectively. Contour lines display a theoretical estimate of decompression rate and mass flow rate against fixed conduit radius (see equation 4 in Barth et al. \\(^5\\) ). Hatched areas show the mass discharge rate of the 2014 and 2016 eruptions (see details in the main text). Dashed line indicates the boundary of magmastatic-dominated and friction-dominated decompression after Barth et al. \\(^5\\) .", + "footnote": [], + "bbox": [ + [ + 125, + 90, + 867, + 536 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. A conceptual diagram of episodic deformation in the chamber accompanying repetitive shallow volcanic resonances, providing the glue to long-term volcanic output and short-term magma ascent/discharge rates. Green solid bars indicate the global observations of the average volcanic output rates over eruption cycles in geological time scale ( \\(\\sim 1\\) yr–1 Myr). Gray dashed bars indicate the single-event magma ascent/discharge rate over a seismic time scale ( \\(\\sim 100\\) s). Composition, viscosity, rheology and tectonic settings govern the recurrences of episodic deformation (red lines).", + "footnote": [], + "bbox": [ + [ + 115, + 88, + 860, + 458 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "Fig. S1. Examples of single-event tilt and displacement waveforms at N.ASHV. (a) Low-pass filtered displacement and tilt waveforms with the upward and east-down offsets, respectively. (b) Low-pass filtered displacement and tilt waveforms with the downward and west-down offsets, respectively. The starting time of each waveform is indicated above each trace.", + "footnote": [], + "bbox": [ + [ + 81, + 130, + 930, + 656 + ] + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_2.jpg", + "caption": "Fig. S2. Global and monthly waveform stacks against highest quality data (subset I). (a) the global waveform stacks of inflation events during the unrest (November 2011 to August 2014). (b) the global waveform stacks of deflation events (October 2014 to April 2015) (c) the monthly waveform stack of inflation events in July 2014. (d) the monthly waveform stack of deflation events in November 2014. LPT signal is present in the middle of the time window at \\(\\sim 220 - 270\\) s. Waveform stacks from stations N.ASHV and N.ASIV are shown in black lines and green lines, respectively. The orange shaded region marks the \\(99.7\\%\\) confidence interval estimated by the bootstrapped resampling (ref.36). Displacement and tilt waveform stacks are low-pass filtered at 10 s and 30 s with a 4th-order casual Butterworth filter, respectively. The number of events used in each waveform stack is shown in the upper right.", + "footnote": [], + "bbox": [ + [ + 235, + 130, + 777, + 787 + ] + ], + "page_idx": 40 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_3.jpg", + "caption": "Fig. S3. 1-hour long global and monthly waveform stacks against highest quality data (subset I). Same as Fig. S2, expect only waveform data associated with isolated LPT events are included in the stack. The number in the left panel indicates the number of the well-isolated LPT used in the stacks. (a)-(d) the global and monthly tilt-rate waveform stacks. (e)-(h) global and monthly tilt stacks. The trend defined in [-10,-2] minutes of the tilt waveform stacks is used to remove the background trend. The gray shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).", + "footnote": [], + "bbox": [ + [ + 80, + 131, + 931, + 621 + ] + ], + "page_idx": 41 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_4.jpg", + "caption": "Fig. S4. Synthetic amplitude-distance decay against static and filtered waveforms. (a) and (b) display normalized displacement amplitude-distance decay trend with a source depth \\(0.5 \\mathrm{km}\\) and \\(5.0 \\mathrm{km}\\) , respectively. (c) and (d) display normalized tilt amplitude-distance decay trend with a source depth \\(0.5 \\mathrm{km}\\) and \\(5.0 \\mathrm{km}\\) , respectively. Static, long-period (10-30 s), and ultra-long-period (100-200 s) amplitude measurements are shown in green lines, blue lines and red lines, respectively.", + "footnote": [], + "bbox": [ + [ + 220, + 131, + 787, + 560 + ] + ], + "page_idx": 42 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_5.jpg", + "caption": "Fig. S5. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. (a)-(e) the amplitude ratio in the LPT frequency band (10-30 s) against five subsets, respectively (the subset I→the highest quality and the subset V→the lowest quality). The amplitude ratio measured in each monthly stack is shown in shaded square. The color of the square denotes the number of stacked inflation events in a linear scale (dark→higher number and light→lower number). The error bar marks the uncertainty with a 99.7% confidence level estimated from the bootstrap resampling (ref.36). The red line marks the amplitude ratio measured from the global inflation waveform stack (Fig. S2a). The green diamond marks the amplitude ratio from Event 1. (f)-(j) Same as (a)-(e), except for measurements in the filtered-static-offset (FSO) frequency band (100-200 s). (k)-(o) the number of events used in each monthly stack against five subsets, respectively. The amplitude ratio in the FSO frequency band becomes more variable when the data quality is low or/and the number of events in the monthly stack is low.", + "footnote": [], + "bbox": [ + [ + 78, + 170, + 933, + 667 + ] + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_6.jpg", + "caption": "Fig. S6. Amplitude ratios between north-south and east-west tilt at N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.", + "footnote": [], + "bbox": [ + [ + 79, + 130, + 936, + 630 + ] + ], + "page_idx": 44 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_7.jpg", + "caption": "Fig. S7. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.", + "footnote": [], + "bbox": [ + [ + 79, + 130, + 936, + 630 + ] + ], + "page_idx": 45 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_8.jpg", + "caption": "Fig. S8. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S5, except for the monthly deflation waveform stacks. The red line marks the amplitude ratio measured from the global deflation waveform stack (Fig. S2b).", + "footnote": [], + "bbox": [ + [ + 80, + 130, + 933, + 630 + ] + ], + "page_idx": 46 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_9.jpg", + "caption": "Fig. S9. Amplitude ratios between north and east tilts at N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.", + "footnote": [], + "bbox": [ + [ + 80, + 130, + 936, + 631 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_10.jpg", + "caption": "Fig. S10. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.", + "footnote": [], + "bbox": [ + [ + 78, + 128, + 936, + 628 + ] + ], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_11.jpg", + "caption": "Fig. S11. Broadband displacement waveforms before and after the correction. (a-c) the original broadband displacement waveforms of the reference event in the vertical, east-west and north-south components at the four broadband seismic stations before removing the baseline error estimated from the polynomial fit. (d-f) the broadband displacement waveforms after removing the baseline errors. A \\(4^{th}\\) -order polynomial function is used to estimate the baseline error from the original displacement waveforms in the pre-event and post-event time windows. Vertical dashed lines mark the starting and ending times of the event (see also Method).", + "footnote": [], + "bbox": [ + [ + 104, + 131, + 912, + 626 + ] + ], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_12.jpg", + "caption": "Fig. S12. Posterior probability functions of the inverted source parameters. (a-c) the marginal posterior probability functions against the source location with respect to the Naka-dake first crater. (d-e) the marginal probability functions of the strike and dip angles of the shearing and tensile components. (f) the marginal probability function against the regularization parameter \\(\\beta\\) . (g-j) display the marginal probability functions against the moments of the strike-slip (Mss), dip-slip (Mds), tensile-crack (Mtc), and isotropic (Mex) components, respectively. (k-l) display the joint posterior probability functions of the coupled parameters regarding the bimodal probability functions in (d) and (e). The red line marks the preferred parameters with the maximum posterior probability. Two peaks in (d) and (e) define the orientations the fault plane and the auxiliary plane.", + "footnote": [], + "bbox": [ + [ + 65, + 137, + 949, + 518 + ] + ], + "page_idx": 50 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_13.jpg", + "caption": "Fig. S13. Measurement of tilt offset in the spectra domain. (a) Synthetic tilt waveforms with the positive (red), null (gray), and negative (blue) offsets of \\(1 \\mu m\\) . (b) Synthetic tilt waveform with a large offset of \\(50 \\mu m\\) . (c) Synthetic tilt-rate amplitude spectra normalized against the amplitude of (b) at 10,000 s. Static offsets measured directly from (a) and (b) are normalized against the one from (b), shown in squares. (d) Observed monthly tilt waveform stacks with the positive (red), near-null (gray), and negative (blue) offsets. (e) Observed tilt waveform of the reference event. (f) Observed tilt-rate amplitude spectra normalized against the amplitude of (e) at 10,000 s. Static offsets measured directly from (d) and (e) are normalized against the one from (e), shown in squares. These examples show that the amplitude plateau of the tilt-rate waveform at 10000 s provides a reasonable estimate of the static offset.", + "footnote": [], + "bbox": [ + [ + 280, + 202, + 740, + 690 + ] + ], + "page_idx": 51 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_14.jpg", + "caption": "Fig. S14. Monthly volume changes in the five subsets. (a)-(e) Histograms of the volume change per event estimated from the monthly tilt waveform stacks in five subsets, respectively. Results from the monthly inflation and deflation stacks are shown in red and blue, respectively. (f-l) Same as (a)-(e), except displaying the monthly volume changes. The shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).", + "footnote": [], + "bbox": [ + [ + 84, + 235, + 925, + 747 + ] + ], + "page_idx": 52 + } +] \ No newline at end of file diff --git a/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0.mmd b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1aebacdbf7c9ddc8edf6a9477ef8c71ea4ef1dc0 --- /dev/null +++ b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0.mmd @@ -0,0 +1,551 @@ + +# Episodic transport of discrete magma batches beneath Aso volcano + +Jieming Niu ( j.niu@ucl.ac.uk ) University College London https://orcid.org/0000- 0002- 4481- 022X + +Teh- Ru Song Department of Earth Sciences, University College London + +## Article + +Keywords: Trans- crustal Magma Plumbing System, Long- term Volcanic Output, Short- term Eruption Dynamics, Episodic Long- period Tremors, Synchronous Deformation Events, Brittle- to- ductile Transition Rheology + +Posted Date: May 26th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 408334/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on September 21st, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 25883- y. + +<--- Page Split ---> + +# Title: Episodic transport of discrete magma batches beneath Aso volcano + +Authors: Jieming Niu \(^{1*}\) , Teh- Ru Alex Song \(^{1\dagger}\) + +## Affiliations: + +\(^{1}\) Seismological Laboratory, Department of Earth Sciences, University College London, WC1E 6BT, London, United Kingdom. + +\*Correspondence to: j.niu@ucl.ac.uk. + +Summary: Magma ascent, storage, and discharge in the trans- crustal magma plumbing system are key to long- term volcanic output and short- term eruption dynamics. Petrological analytics, geodetic deformations and mechanical modeling have shaped the current understanding of magma transport. However, due to the lack of observations, how a distinct magma batch transports from a crystal- rich mush region to a crystal- poor pool with eruptible magma remains enigmatic. Through stacking of tilt and seismic waveform data, we find that episodic long- period tremors (LPTs) located near sea level beneath the Aso volcano are accompanied by a synchronous deformation event, which initiates \(\sim 50\) seconds before individual LPT event and concludes seconds after. The episodic deformation source corresponds to either an inflation or a deflation event located \(\sim 3\) km below sea level, with a major volumetric component (50- 440 cubic meters per event) and a minor high- angle normal- fault component. We suggest that these deformation events likely represent short- lived, episodic upward transport of discrete magma batches accompanied by high- angle shear failure near the roof of the inferred magma chamber at relatively high temperature, whereas their recurrences, potentially composition dependent, are regulated by the brittle- to- ductile transition rheology under low differential stress and high strain rate due to the surge of magma from below, regulating long- term volcanic output rate. The magma ascent velocity, decompression rates, and cumulative magma output deduced from the episodic deformation events before recent eruptions in Aso volcano are compatible with retrospective observations of the eruption style, tephra fallouts, and plume heights, promising real- time evaluation of upcoming eruptions. + +<--- Page Split ---> + +## Main Text: + +Theoretical considerations have shown that the ascending of a magma batch is likely dictated by the density contrast between magma and surrounding rock, the excess pressure in the magma reservoir, and magma viscosity'. More recent developments on magma transport and storage emphasize a transcrustal magmatic plumbing system consisting of relatively long- lived crystal- rich mushes and shorter time- scale crystal- poor pools that were eventually tapped by eruptions?. However, the key process of distinct magma transport from the crystal- rich mush to the crystal- poor pool is not well known, even though it is intimately linked to magma ascent and discharge, controlling short- term eruption dynamics and long- term volcanic output. + +Numerous petrological means have been utilized to interrogate magma ascent, including recent chemical diffusion based geospeedometers- 9. However, these advancements rely on retroactive analysis of eruption products, which cannot be obtained in real- time during unrest or before eruptions. Furthermore, these estimates do not resolve the volume or duration of distinct magma ascent. Magma ascent inferred from the migration of seismicity can be difficult as seismicity rarely follows simple upward movement and the progression of seismicity could indicate conduit formation or propagating of pressure through existing magma, rather than magma ascent1,10,11. + +Ground deformation detected from geodetic means (e.g., InSAR, GNSS, tiltmeter, strainmeter) provides invaluable insight into the average rate of magma supply12- 14. However, except few instances where syn-eruptive deformation of 0.1- 10 microstrain over hours or less are reported15- 17, the documented pre-eruptive ground deformation is typically on the order of millimeter to centimeter, microstrain, or microradian over days, weeks, or months and they do not offer a direct assessment on the nature of a + +<--- Page Split ---> + +distinct magma ascent, potentially operating at a much shorter duration, higher frequency or/and smaller volume. + +Seismic resonances with periods of 0.2- 200 s have been widely detected near shallow conduits or reservoirs (e.g., \(< 2 \mathrm{km}\) ) in diverse volcanic systems \(^{18,19}\) . They are often stationary and repetitive, responding to internal processes accompanying magma ascent or/and the build- up of conduit overpressure, including degassing, heat transfer and vaporization, change in the fluid pressure, or/and unsteady magma transport of a deeper origin. The repetitive nature of shallow seismic resonance permits the use of stacking to tease out small deformation signal \(^{20 - 22}\) that are otherwise undetected. + +Here we report new observations that manifest a causal relationship between repetitive shallow seismic resonances (i.e., shallow hydrothermal reservoir) and deformation within the deeper plumbing system beneath Aso, an andesitic volcano in Japan \(^{23 - 25}\) . Specifically, we focus investigation against the very long- period signal ( \(\sim 15 \mathrm{s}\) period) located in a fluid- filled crack- like conduit near sea level beneath the active Naka- dake first crater \(^{26 - 31}\) (Fig. 1a). Following the nomenclature established in earlier studies, we refer to the very long- period signal as long- period tremor (LPT) hereafter. Critically, LPT is known to occur during the quiescence and active period \(^{27,29,31}\) , offering a unique opportunity to methodically explore discrete magma ascents over the entire eruption cycle in 2011- 2016. + +## LPT and synchronous tilt/displacement offset + +We analyze seismic and tilt waveform recorded by the fundamental volcano observation network, or V- net \(^{32}\) , which is equipped with collocated surface broadband seismic sensors (Nanometrics 240, \(\sim 250 \mathrm{s}\) natural period) and borehole tiltmeters ( \(\sim 1\) nanoradian resolution) \(^{33}\) . Before the \(8^{\mathrm{th}}\) Oct 2016 phreatomagmatic eruption \(^{34}\) , we observe two anomalously large LPT events that occurred about 2- 3 + +<--- Page Split ---> + +minutes before the eruption (Fig. 1b). Tilt offsets accompanying the two LPT events can be readily observed at all four V- net stations (Fig. 1b). At station N.ASHV, the east- west tilt offset accompanying the two events reach \(\sim 58\) nrad and \(\sim 88\) nrad, respectively. + +We manually check tilt and vertical displacement waveforms of other events in the LPT catalog31 and discover similar observations in several instances (Fig. S1). In general, the tilt and displacement offsets at station N.ASHV can be either east- down and vertical- up or west- down and vertical- down, but the amplitude is typically much smaller than those accompanying the two LPT events before the 2016 eruption. As shown in Fig. 1b, there are several striking differences between the signal of LPT and the tilt offset. First, the tilt offset at station N.ASHV is much stronger in the east- west component than that in the north- south component, whereas the LPT signal in the north- south component is much stronger than that in the east- west component. Secondly, the east- west tilt offset at station N.ASHV is much stronger than that at station N.ASIV, but the LPT signal at station N.ASHV is much weaker than that at station N.ASIV. These observations strongly indicate that the source of the tilt offset is spatially separated from the LPT source, possibly closer to station N.ASHV than the source of LPT, which is near the active Nakadake first crater (see also Fig. 1a). + +Discovery of the inflation/deflation event beneath Aso volcano + +To systematically examine the presence and the stability of such synchronous signals, we implement the matched- filter technique31,35 against all LPT events during the 2011- 2016 Aso eruption cycle31. The first of the two LPT events (Event 1) before the 2016 eruption is selected as the reference template event, and we cross- correlate tilt and seismic waveform data of each LPT event against the template waveforms in the ultra- long period band of 50- 250 s, which allows us to effectively detect possible displacement or tilt offset in the near- field (see also Material and Method). Specifically, we + +<--- Page Split ---> + +gather detections with the same sign of cross- correlation coefficient in a given calendar time window to produce waveform stacks and compute bootstrap uncertainties36. + +Here we present two global waveform stacks of the highest quality at station N.ASHV during the volcanic unrest in January 2011- August 2014 (Fig. 2a) and the intermittent Strombolian eruptions in October 2014- April 2015 (Fig. 2b), respectively. During the volcanic unrest, we observe an upward displacement offset and a predominantly east- down tilt offset (Fig. 2a). During the 2014 Strombolian eruption, we observe a downward displacement offset and a predominantly west- down tilt offset (Fig. 2b). While the amplitude of the LPT differs by a factor of \(\sim 3\) between the two episodes, the waveform offset is relatively steady, i.e., \(\sim 1 \mu m\) in the vertical displacement and \(\sim 1\) nrad in the east- west tilt, respectively. Small bootstrap uncertainties of the global waveform stacks suggest the prevalence of the signal that is synchronous with LPT. Here we refer to the signal associated with an upward (downward) displacement offset as the inflation (deflation) event. + +While it is well documented that LPT is often accompanied by short- period tremors (SPTs, \(\sim 0.5 - 1\) s period) located near the top of the LPT source region18,27,37, as demonstrated in Fig. 2, the inflation/deflation event occurs \(\sim 60\) seconds earlier than SPT or LPT, providing a causal source of internal triggering. While SPT precedes LPT by \(\sim 10\) seconds, the peak energy of SPT envelopes arrives \(\sim 10\) seconds after the LPT onset. The inflation (deflation) event, LPT, and SPT end approximately at the same time. Observations of inflation/deflation events can also be made in other stations or high- quality monthly waveform stacks (Fig. S2). Notably, the tilt offsets can remain relatively constant over 30 minutes (Fig. S3). We also measure the displacement and tilt amplitude ratio between stations N.ASHV and N.ASIV in the ultra- long period band of 100- 200 s (Fig. S4) against all monthly stacks (Fig. S5- S10). The stability of these observational attributes indicates that repetitive inflation and deflation events share generally the same proximity with a relatively stationary source region. + +<--- Page Split ---> + +Following the Bayesian source inversion approach by Fukuda and Johnson38, we measure the vertical displacement, horizontal displacement (Fig. S11, see Method), and horizontal tilt offsets of the reference event (Event 1) and invert the source location and mechanism (Fig. 3a- c, see also Material and Method). We find that the source is located at a depth of \(\sim 3\) km below sea level, \(\sim 2\) km west of the Nakadake first crater (Fig. 1a, Fig. 3a- d, see also Table S2). The reference event has a moment magnitude \(\mathrm{M_w} = 3.3\) with a predominantly volumetric component ( \(\sim 80\%\) ) and a minor dip- slip (normal- fault) component of \(\sim 20\%\) (Fig. 1a, Fig. S12, Table S2). The corresponding volume change is \(\sim 25,800 \mathrm{m}^3\) (see Method). + +## The nature of inflation/deflation event in a renewed plumbing system beneath Aso volcano + +The source of the inflation/deflation event is located near the inferred magma chamber39,40 (Fig. 1a) but at a shallower depth. It is also near the upper end of a high conductivity anomaly41. We suggest that the source of the inflation/deflation event corresponds to a shallow magma storage zone (SMSZ) beneath the Aso volcano, connecting the source of LPT in the shallow conduit from above and the magma chamber from below (Fig. 3d), suggesting that these episodic deformation events likely represent instantaneous upward transport of a distinct magma batch near the roof of the inferred magma chamber. + +We consider that the upward transport of a discrete magma batch only occurs in a short episode of \(\sim 50 - 100\) s near the roof of a partially molten zone filled with crystal- rich magma, accompanied by high- angle shear failure. Since the temperature at the depth of SMSZ ( \(\sim 3\) km) beneath Aso volcano reaches \(500^{\circ} \mathrm{C}^{42,43}\) , the recurrences of these deformation events are likely regulated by the brittle- to- ductile transition, under high temperature and high strain- rate44 due to the surge of magma from below, possibly facilitated by ductile fracturing45 or/and magma mixing in the mush46. The deformation event ends as the system returns to the ductile regime as a result of the shear- stress release or/and decreasing magma supply (i.e., lower strain rate). As elaborated in the next session, the transport of discrete magma batches does not exclude volatiles that may readily exist in the top of the mush region47. + +<--- Page Split ---> + +We suggest that, when the inflow (outflow) is higher than the outflow (inflow), there is a net volume increase (decrease) in the SMSZ and we observe an inflation (deflation) event (Fig. 3e, Fig. 3g). When the inflow is approximately equal to the outflow, there is negligible net volume change in the SMSZ and we observe a negligible offset in tilt or displacement (Fig. 3f). As shown in Fig. 4a, we observe predominantly inflation events during intense unrest in July and August 2014. Right after the ash eruption at the end of August 2014, we observe events with a negligible net offset in September 2014. During the intermittent Strombolian eruptions (e.g., November 2014, March 2015), we observe predominantly deflation events. After the pyroclastic cone collapsed in early May 2015, we again observe inflation events in May and June 2015. While the deflation events dominate in late 2015 and most of 2016, two anomalously large inflation events occur just minutes before the 2016 phreatomagmatic eruption (Fig. 1b). + +These systematic observations also point to a robust and intimate link between surface volcanic activities and the state of the SMSZ. The prominence of inflation (deflation) events corresponds to a lower (higher) outgassing potential (Fig. 4b), equivalently a higher (lower) proportion of pressurization LPT in the shallow conduit31, and a lower (higher) \(\mathrm{SO}_2\) emission (Fig. 4c). Finally, the depth of the SMSZ is very consistent with the highest gas saturation pressure of melt inclusions in scoria (~80- 118 MPa, or 2- 3.5 km below sea level)23, supporting the hypothesis that the magma (and gas) are likely transferred from a deeper reservoir48 (i.e., a crystal- rich mush) and temporarily stalled in the SMSZ (i.e., a crystal- poor pool), which serves as a preparation zone for the storage and upward transport before upcoming eruptions. + +Implications of the inflation/deflation event on magma/gas transport and upcoming magmatic eruption beneath Aso volcano + +<--- Page Split ---> + +Our new observations suggest that the upward transport of magma/gas from the magma chamber toward the surface is a stepwise process in an episodic fashion. Since the source is repetitive, the volume change associated with individual inflation/deflation events can be obtained by scaling the tilt amplitude of each event against that of the reference event. For example, the east-west tilt of the reference event at station N.ASHV is \(\sim 58\) nrad, corresponding to a volume increase of \(\sim 25,800 \mathrm{m}^3\) . The typical east-west tilt in the global stacks is only about \(\sim 1\) nrad (Fig. 2), which translates to a volume change of \(\sim 440 \mathrm{m}^3\) per event, or the size of a typical swimming pool. + +We systematically measure the tilt amplitude of the monthly stacks (Fig.S13, see also Material and Method) and estimate the volume change per event, which can be \(\sim 50 \mathrm{m}^3\) in the subset of the lowest signal- to- noise ratio (Fig. S14). The monthly volume change is obtained by multiplying the volume change per event against the event number (Fig. S14) and the cumulative volume change over 2011- 2016 is illustrated in Fig. 4a. + +For a typical source duration of \(\sim 100 \mathrm{s}\) (Fig. 2), the single- event volume change rate estimated from the global stacks or the monthly stacks is \(\sim 0.5 - 4.4 \mathrm{m}^3 /\mathrm{s}\) or \(\sim 0.03 - 0.13 \mathrm{km}^3 /\mathrm{year}\) which is \(\sim 2\) orders of magnitude higher than the average volume change rate in 2011- 2013 (i.e., \(\sim 10^{- 4} \mathrm{km}^3 /\mathrm{year}\) ) and in late 2014 ( \(\sim 1.3 \times 10^{- 3} \mathrm{km}^3 /\mathrm{year}\) ), the geodetic deflation rate of the magma chamber at a greater depth (i.e., \(\sim 2.5 \times 10^{- 4} \mathrm{km}^3 /\mathrm{year}\) over several decades) \(^{49,50}\) and Aso post- caldera output rates (i.e., \(\sim 0.001 \mathrm{km}^3 /\mathrm{year}\) ) \(^{24,25,51}\) (Fig. 4d). We hypothesize that the volume change rate associated with individual inflation/deflation event marks the instantaneous magma transport rate of a distinct magma batch, whereas the recurrence interval modulates the averaged transport rate, reconciling the magma transport over multiple timescales. This hypothesis reconciles similar disparities noted in basaltic volcanic systems such as Kilauea \(^{52}\) and silicic volcanic systems such as Mount St. Helens \(^{53}\) . + +<--- Page Split ---> + +Considering a magma density of \(2500\mathrm{kg / m}^3\) , we convert the net volume change \(\Delta \mathrm{V}\) shown in Fig. 4a to the net mass change in the SMSZ. There is a good agreement between the total mass change in the SMSZ before the 2014 eruption and the mass of tephra fallout \(^{54,55}\) (Fig. 4e), consistent with a minimal effect of magma compressibility on the estimate of \(\Delta \mathrm{V}\) . While the net volume change \(\Delta \mathrm{V}\) in the SMSZ can be due to magma \((\Delta \mathrm{V}_{\mathrm{magma}})\) or/and gas \((\Delta \mathrm{V}_{\mathrm{gas}})\) , our observation shown in Fig. 4e is consistent with the scenario where \(\Delta \mathrm{V}_{\mathrm{magma}} \gg \Delta \mathrm{V}_{\mathrm{gas}}\) before the 2014 eruption. However, the net volume change in the SMSZ during/after the 2014 Strombolian eruption episode is 6- 8 times larger than that before the eruption (Fig. 4e). Such a discrepancy is consistent with \(\Delta \mathrm{V}_{\mathrm{magma}} \ll \Delta \mathrm{V}_{\mathrm{gas}}\) in the SMSZ, corroborating with a higher outgassing potential \(^{31}\) (Fig. 4b), \(\mathrm{SO}_2\) emission \(^{56}\) (Fig. 4c), and excess degassing \(^{23}\) . + +On the other hand, the mass change of the two anomalously large inflation events \(\sim 2 - 3\) minutes before the 2016 phreatomagmatic explosion is also readily comparable to the mass of juvenile magma from the ash fallout and pyroclastic current deposit (PDC) \(^{34}\) (Fig. 4e). Nevertheless, if the SMSZ buffers the upward magma/gas from the magma chamber, our observation is consistent with the scenario where the cumulative mass change in the SMSZ is closely linked to the size of upcoming magmatic eruptions. + +Contrasting magma ascent velocity and mass flow rate near the SMSZ prior to the 2014 Strombolian eruption and the 2016 phreatomagmatic explosion + +Using a simplified 1- D conduit flow model that balances the conduit radius against the overpressure modulated by the magma flow \(^{57}\) , the single- event volume change rate allows us to infer the magma ascent velocity and decompression rate near the SMSZ before the eruptions (Material and Method). For a low- viscosity magma (i.e., \(10^3\mathrm{Pa}\cdot \mathrm{s}\) ), a density difference between magma and wall rock of \(\sim 100\mathrm{kg / m}^3\) , a lithological gradient of \(0.025\mathrm{MPa / m}\) , and a volume change rate of \(\sim 0.5 - 4.4\mathrm{m}^3 /\mathrm{s}\) , we + +<--- Page Split ---> + +estimate the ascent velocity near the SMSZ of \(\sim 0.14 - 0.41 \mathrm{m / s}\) and the decompression rate of \(\sim 0.004 - 0.011 \mathrm{MPa / s}\) , which is comparable to syn-eruptive decompression rates estimated by melt embayment and water-in-olivine studies in basaltic eruptions \(^{5,6}\) and theoretical estimates for the strombolian eruption \(^{58}\) (Fig. 5). + +Using a magma density of \(2500 \mathrm{kg / m^3}\) , we convert the single- event volume change rate \((\sim 0.5 - 4.4 \mathrm{m^3 / s})\) to the mass flow rate of \(\sim 1,250 - 11,000 \mathrm{kg / s}\) (Fig. 5). Since the Strombolian eruptions in Aso are intermittent, the averaged mass discharge rate estimated by tephra deposits of the 2014 Strombolian eruption (i.e., \(\sim 25 - 430 \mathrm{kg / s})^{55}\) is likely lower than the instantaneous mass discharge rate. Assuming only small percentages (i.e., \(5\%\) ) of the entire eruption duration \(^{24}\) , the magma discharge rate of the 2014 eruption is \(\sim 500 - 8,600 \mathrm{kg / s}\) , which is reasonably comparable to the estimated single- event mass flow rate in the SMSZ. While the estimated mass flow rate in Aso is lower than those estimated in basaltic eruptions by an order of magnitude (Fig. 5), such a difference is consistent with the disparity in the average volcanic output rate between basaltic eruptions and andesitic eruptions \(^{59}\) (Fig. 4d). + +In contrast, the two anomalously large inflation events \(\sim 2 - 3\) minutes before the 2016 phreatomagmatic explosion corresponds to a much higher volume change rate of \(\sim 860 - 1,300 \mathrm{m^3 / s}\) over a 30 s duration. The estimated mass flow rate is \(\sim 2.2 - 3.3 \times 10^6 \mathrm{kg / s}\) (Fig. 5) and the estimated ascent velocity and decompression rate near the SMSZ are also much higher at \(\sim 7 \mathrm{m / s}\) and \(\sim 0.2 \mathrm{MPa / s}\) , respectively, indicating a much larger overpressure. These results are similar to other explosive basaltic eruptions \(^{6 - 8}\) (Fig. 5). Considering the tephra mass and the duration of the 2016 phreatomagmatic explosion of \(160 - 260 \mathrm{s}\) , the total mass discharge rate is \(2.7 - 4.1 \times 10^6 \mathrm{kg / s}^{60}\) , comparable to our inferred mass flow rate in the SMSZ. Using the empirical scaling between the mass discharge rate and the plume height \(^{61}\) , the mass flow rate in the SMSZ before the 2016 eruption projects a plume height of \(\sim 10 \mathrm{km}\) , consistent with observations \(^{56,62,63}\) . + +<--- Page Split ---> + +In essence, we show that repetitive shallow resonances provide a pathway to unravel magma flows deep in the magma plumbing system (Fig.6). Specifically, discrete magma transport between crystal- rich mush and crystal- poor pool could be accommodated by episodic deformation event accompanied by high- angle normal faulting in a high strain- rate, low differential stress regime governed by the brittle- ductile transition rheology. The duration of each deformation event ( \(\sim 50\) s) is much longer than what is expected for crustal earthquakes of similar size and such a slow deformation, to some extent, is analogous to slow- slip events observed worldwide in the subduction zone interface or in the deep crust64,65, where the rheological condition and fluid pressure are likely similar to that near the chamber roof, or the top of the mush region.. + +The contrast in the decompression rate (or the ascent velocity) and the mass flow rate inferred in this report are strongly correlated with the eruption style in Aso, collaborating with theoretical prediction66 and geological evidence67. As persistent shallow resonances in Aso have been documented for almost a century31,68,69, establishing the amplitude scaling between the synchronous sources in Aso will permit critical evaluations of pre- eruptive volume changes (or volume change rates) against the size and style of historical eruptions even when modern geodetic and broadband seismic data are not available. + +Since the mass discharge rate and the magma ascent velocity or decompression rate can only be obtained retroactively after the eruptions5- 10,67,70, new seismic observations reported here underscore the possibility that, through real- time signal detection/processing and source analysis in volcanic systems where repetitive shallow seismic resonances are routinely detected18,19, the volume change and volume change rate (or mass flow rate) of episodic deformation similar to observations in Aso could facilitate in- situ characterization of magma ascent and transport before upcoming eruptions. Such efforts can be coupled with improved knowledge of external loading/unloading conditions67,71, continuous monitoring of + +<--- Page Split ---> + +degassing rate \(^{72}\) and the permeability/rheology of shallow conduit \(^{31,67}\) to facilitate diagnosing short- term eruption potential. + +<--- Page Split ---> + +## References + +1. Rubin, A. M. Propagation of magma-filled cracks. \*Annual Review of Earth and Planetary Sciences\* 23, 287-336 (1995). + +2. Cashman, K. V., Sparks, R. S. J. & Blundy, J. D. Vertically extensive and unstable magmatic systems: A unified view of igneous processes. \*Science\* 355, eaag3055 (2017). + +3. Edmonds, M., Cashman, K. V., Holness, M. & Jackson, M. Architecture and dynamics of magma reservoirs. \*Phil. Trans. R. Soc. A\* 377, 20180298 (2019). + +4. Petrelli, M. & Zellmer, G. F. Rates and timescales of magma transfer, storage, emplacement, and eruption. \*Dynamic Magma Evolution\* 1-41 (2020). + +5. Barth, A. \*et al.\* Magma decompression rate correlates with explosivity at basaltic volcanoes—Constraints from water diffusion in olivine. \*Journal of Volcanology and Geothermal Research\* 387, 106664 (2019). + +6. Newcombe, M. E., Plank, T., Barth, A., Asimow, P. D. & Hauri, E. Water-in-olivine magma ascent chronometry: Every crystal is a clock. \*Journal of Volcanology and Geothermal Research\* 398, 106872 (2020). + +7. Lloyd, A. S. \*et al.\* NanoSIMS results from olivine-hosted melt embayments: Magma ascent rate during explosive basaltic eruptions. \*Journal of Volcanology and Geothermal Research\* 283, 1-18 (2014). + +8. Ferguson, D. J. \*et al.\* Magma decompression rates during explosive eruptions of Kīlauea volcano, Hawaii, recorded by melt embayments. \*Bulletin of Volcanology\* 78, 71 (2016). + +<--- Page Split ---> + +9. Costa, F., Shea, T. & Ubide, T. Diffusion chronometry and the timescales of magmatic processes. \*Nat Rev Earth Environ\* 1, 201–214 (2020). + +10. Rutherford, M. J. Magma ascent rates. \*Reviews in Mineralogy and Geochemistry\* 69, 241–271 (2008). + +11. Rivalta, E., Taisne, B., Bunger, A. & Katz, R. A review of mechanical models of dike propagation: Schools of thought, results and future directions. \*Tectonophysics\* 638, 1–42 (2015). + +12. Dzurisin, D. \*Volcano deformation: new geodetic monitoring techniques\*. (Springer Science & Business Media, 2006). + +13. Segall, P. \*Earthquake and volcano deformation\*. (Princeton University Press, 2010). + +14. Biggs, J. & Pritchard, M. E. Global Volcano Monitoring: What Does It Mean When Volcanoes Deform? \*ELEMENTS\* 13, 17–22 (2017). + +15. Linde, A. T., Agustsson, K., Sacks, I. S. & Stefansson, R. Mechanism of the 1991 eruption of Hekla from continuous borehole strain monitoring. \*Nature\* 365, 737–740 (1993). + +16. Linde, A. T. \*et al.\* Vulcanian explosion at Soufrière Hills Volcano, Montserrat on March 2004 as revealed by strain data. \*Geophysical Research Letters\* 37, (2010). + +17. Ripepe, M. \*et al.\* Ground deformation reveals the scale-invariant conduit dynamics driving explosive basaltic eruptions. \*Nature Communications\* 12, 1–8 (2021). + +18. Kawakatsu, H. & Yamamoto, M. Volcano Seismology. in \*Treatise on Geophysics\* vol. 4 389–420 (Elsevier, 2007). + +19. Chouet, B. A. & Matoza, R. S. A multi-decadal view of seismic methods for detecting precursors of magma movement and eruption. \*Journal of Volcanology and Geothermal Research\* 252, 108–175 (2013). + +<--- Page Split ---> + +20. Oikawa, J. et al. Ground deformation associated with volcanic tremor at Izu-Oshima Volcano. Geophysical Research Letters 18, 443–446 (1991). + +21. Dawson, P. & Chouet, B. Characterization of very-long-period seismicity accompanying summit activity at Kilauea Volcano, Hawai'i: 2007–2013. Journal of Volcanology and Geothermal Research 278–279, 59–85 (2014). + +22. Kobayashi, T., Ohminato, T., Ida, Y. & Fujita, E. Intermittent inflations recorded by broadband seismometers prior to caldera formation at Miyake-jima volcano in 2000. Earth and Planetary Science Letters 357, 145–151 (2012). + +23. Saito, G., Ishizuka, O., Ishizuka, Y., Hoshizumi, H. & Miyagi, I. Petrological characteristics and volatile content of magma of the 1979, 1989, and 2014 eruptions of Nakadake, Aso volcano, Japan. Earth Planets Space 70, 197 (2018). + +24. Ono, K., Watanabe, K., Hoshizumi, H. & Ikebe, S. Ash eruption of the Naka-dake crater, Aso volcano, southwestern Japan. Journal of Volcanology and Geothermal Research 66, 137–148 (1995). + +25. Miyabuchi, Y. A 90,000-year tephrostratigraphic framework of Aso Volcano, Japan. Sedimentary Geology 220, 169–189 (2009). + +26. Sassa, K. Geophysical studies on the volcano Aso. (Part 1: Volcanic micro-tremors and eruptive-earthquakes). Mem. Coll. Sci., Kyoto Imp. Univ. Series A 255–293 (1935). + +27. Kaneshima, S. et al. Mechanism of phreatic eruptions at Aso volcano inferred from near-field broadband seismic observations. Science 273, 643–645 (1996). + +28. Yamamoto, M. et al. Detection of a crack-like conduit beneath the active crater at Aso volcano Japan. Geophys. Res. Lett. 26, 3677–3680 (1999). + +<--- Page Split ---> + +29. Kawakatsu, H. et al. Aso94: Aso seismic observation with broadband instruments. Journal of Volcanology and Geothermal Research 101, 129-154 (2000). + +30. Legrand, D., Kaneshima, S. & Kawakatsu, H. Moment tensor analysis of near-field broadband waveforms observed at Aso volcano, Japan. Journal of Volcanology and Geothermal Research 101, 155-169 (2000). + +31. Niu, J. & Song, T.-R. A. Real-time and in-situ assessment of conduit permeability through diverse long-period tremors beneath Aso volcano, Japan. Journal of Volcanology and Geothermal Research 401, 106964 (2020). + +32. Tanada, T. et al. NIED's V-net, the fundamental volcano observation network in Japan. JDR 12, 926-931 (2017). + +33. Sato, H. et al. Development of the Crustal Tilt Observation Method Using Borehole-type Tiltmeters. Zisin1 33, 343-368 (1980). + +34. Miyabuchi, Y. et al. The October 7-8, 2016 eruptions of Nakadake crater, Aso Volcano, Japan and their deposits. in Japan geoscience union-American geophysical union joint meeting. Chiba, Japan, SVC47-11 vol. 22 (2017). + +35. Turin, G. An introduction to matched filters. IEEE Trans. Inform. Theory 6, 311-329 (1960). + +36. Efron, B. & Gong, G. A leisurely look at the bootstrap, the jackknife, and cross-validation. The American Statistician 37, 36-48 (1983). + +37. Mori, T., Sudo, Y., Tsutsui, T. & Yoshikawa, S. Characteristics of isolated hybrid tremor (HBT) during a calm activity period at Aso Volcano. Bull Volcanol 70, 1031-1042 (2008). + +<--- Page Split ---> + +38. Fukuda, J. & Johnson, K. M. Mixed linear—non-linear inversion of crustal deformation data: Bayesian inference of model, weighting and regularization parameters. Geophysical Journal International 181, 1441–1458 (2010). + +39. Sudo, Y. & Kong, L. Three-dimensional seismic velocity structure beneath Aso volcano, Kyushu, Japan. Bull Volcanol 63, 326–344 (2001). + +40. Ohkura, S., Yoshikawa, S., Inoue, H., Utsugi, M. & Kagiyama, T. Leveling in Aso (September-October 2008): Aso research report 2009 (report in Japanese). (2009). + +41. Hata, M. et al. Three-dimensional electrical resistivity distribution beneath the Beppu-Shimabara graben with a focus on Aso caldera, southwest Japan subduction zone. J. Geophys. Res. Solid Earth (2018) doi:10.1029/2018JB015506. + +42. Okubo, Y. & Shibuya, A. Thermal and crustal structure of the Aso volcano and surrounding regions constrained by gravity and magnetic data, Japan. Journal of volcanology and geothermal research 55, 337–350 (1993). + +43. Tanaka, Y. Eruption mechanism as inferred from geomagnetic changes with special attention to the 1989–1990 activity of Aso volcano. Journal of Volcanology and Geothermal Research 56, 319–338 (1993). + +44. Fournier, R. O. Hydrothermal processes related to movement of fluid from plastic into brittle rock in the magmatic-epithermal environment. Economic Geology 94, 1193–1211 (1999). + +45. Weinberg, R. F. & Regenauer-Lieb, K. Ductile fractures and magma migration from source. Geology 38, 363–366 (2010). + +46. Bergantz, G., Schleicher, J. & Burgisser, A. Open-system dynamics and mixing in magma mushes. Nature Geoscience 8, 793–796 (2015). + +<--- Page Split ---> + +47. Edmonds, M. & Wallace, P. J. Volatiles and exsolved vapor in volcanic systems. Elements 13, 29–34 (2017). + +48. Kawaguchi, M. et al. Persistent gas emission originating from a deep basaltic magma reservoir of an active volcano: the case of Aso volcano, Japan. Contributions to Mineralogy and Petrology 176, 1–24 (2021). + +49. Sudo, Y., Tsutsui, T. & Nakaboh, M. Ground deformation and magma reservoir at Aso volcano: Location of deflation source derived from long-term geodetic surveys (In Japanese). Bull. Volcanol. Soc. Jpn. 51, 291–309 (2006). + +50. Nobile, A. et al. Steady subsidence of a repeatedly erupting caldera through InSAR observations: Aso, Japan. Bull Volcanol 79, 32 (2017). + +51. Miyoshi, M. et al. K–Ar ages determined for post-caldera volcanic products from Aso volcano, central Kyushu, Japan. Journal of Volcanology and Geothermal Research 229–230, 64–73 (2012). + +52. Ohminato, T., Chouet, B. A., Dawson, P. & Kedar, S. Waveform inversion of very long period impulsive signals associated with magmatic injection beneath Kilauea volcano, Hawaii. J. Geophys. Res. 103, 23839–23862 (1998). + +53. Scandone, R. & Malone, S. D. Magma supply, magma discharge and readjustment of the feeding system of Mount St. Helens during 1980. Journal of Volcanology and Geothermal Research 23, 239–262 (1985). + +54. Yokoo, A. & Miyabuchi, Y. Eruption at the Nakadake 1st crater of Aso volcano started in November 2014 (In Japanese). Bull. Volcanol. Soc. Jpn. 60, 275–278 (2015). + +<--- Page Split ---> + +55. Miyabuchi, Y. & Hara, C. Temporal variations in discharge rate and component characteristics of tephra-fall deposits during the 2014–2015 eruption of Nakadake first crater, Aso Volcano, Japan. Earth Planets Space 71, 44 (2019). + +56. Japan Meterological Agency. Aso volcano monthly activity reports, October 2016. (2016). + +57. Jaupart, C. & Tait, S. Dynamics of eruptive phenomena. Reviews in Mineralogy and Geochemistry 24, 213–238 (1990). + +58. Gonnermann, H. M. & Manga, M. Dynamics of magma ascent in the. Modeling volcanic processes: The physics and mathematics of volcanism 55 (2013). + +59. White, S. M., Crisp, J. A. & Spera, F. J. Long-term volumetric eruption rates and magma budgets. Geochemistry, Geophysics, Geosystems 7, (2006). + +60. Shimbori, T. Current Status and Perspectives on Tephra Transport Models: Volcanic Ash Fall Forecasts of Aso Volcano on 8 October 2016 as an Example. PROGRAMME AND ABSTRACTS THE VOLCANOLOGICAL SOCIETY OF JAPAN 2017, 5–5 (2017). + +61. Mastin, L. G. et al. A multidisciplinary effort to assign realistic source parameters to models of volcanic ash-cloud transport and dispersion during eruptions. Journal of Volcanology and Geothermal Research 186, 10–21 (2009). + +62. Ishii, K., Hayashi, Y. & Shimbori, T. Using Himawari-8, estimation of SO2 cloud altitude at Aso volcano eruption, on October 8, 2016. Earth Planets Space 70, 19 (2018). + +63. Sato, E., Fukui, K. & Shimbori, T. Aso volcano eruption on October 8, 2016, observed by weather radars. Earth Planets Space 70, 105 (2018). + +<--- Page Split ---> + +64. Peng, Z. & Gomberg, J. An integrated perspective of the continuum between earthquakes and slow-slip phenomena. Nature geoscience 3, 599-607 (2010). + +65. Beroza, G. C. & Ide, S. Slow earthquakes and nonvolcanic tremor. Annual review of Earth and planetary sciences 39, 271-296 (2011). + +66. Woods, A. W. & Koyaguchi, T. Transitions between explosive and effusive eruptions of silicic magmas. Nature 370, 641-644 (1994). + +67. Cassidy, M., Manga, M., Cashman, K. & Bachmann, O. Controls on explosive-effusive volcanic eruption styles. Nature Communications 9, 1-16 (2018). + +68. Sassa, K. Micro-Seismometrical Study on Eruptions of the Volcano Aso (Part II, of the Geophysical Studies on the Volcano Aso). Mem. Coll. Sci., Kyoto Imp. Univ. Series A (1936). + +69. Sakaguchi, H., Sudo, Y., Sawada, Y. & Yoshikawa, S. Relationship between the explosive activities and the associated volcanic tremors observed at Nakadake summit of Aso volcano -- The temporal variation in amplitude of tremors recorded by Wiechert seismograph -- (In Japanese). Bull. Volcanol. Soc. Jpn. 53, 143-149 (2008). + +70. Toramaru, A. BND (bubble number density) decompression rate meter for explosive volcanic eruptions. Journal of Volcanology and Geothermal Research 154, 303-316 (2006). + +71. Manga, M. & Brodsky, E. Seismic triggering of eruptions in the far field: volcanoes and geysers. Annu. Rev. Earth Planet. Sci. 34, 263-291 (2006). + +72. Oppenheimer, C., Fischer, T. & Scaillet, B. Volcanic degassing: Process and impact. Treatise on Geochemistry (Second Edition) 4, 111-179 (2014). + +<--- Page Split ---> + +73. Wielandt, E. & Forbriger, T. Near-field seismic displacement and tilt associated with the explosive activity of Stromboli. Annals of Geophysics (1999) doi:10.4401/ag-3723. + +74. Graizer, V. Tilts in strong ground motion. Bulletin of the Seismological Society of America 96, 2090–2102 (2006). + +75. Graizer, V. Determination of the true ground displacement by using strong motion records. Izvestiya Phys. Solid Earth 15, 875–885 (1979). + +76. Zhu, L. Recovering permanent displacements from seismic records of the June 9, 1994 Bolivia deep earthquake. Geophysical research letters 30, (2003). + +77. Aoyama, H. & Oshima, H. Tilt change recorded by broadband seismometer prior to small phreatic explosion of Meakan-dake volcano, Hokkaido, Japan. Geophysical research letters 35, (2008). + +78. Wang, R., Schurr, B., Milkereit, C., Shao, Z. & Jin, M. An improved automatic scheme for empirical baseline correction of digital strong-motion records. Bulletin of the Seismological Society of America 101, 2029–2044 (2011). + +79. Thun, J., Lokmer, I. & Bean, C. J. New observations of displacement steps associated with volcano seismic long-period events, constrained by step table experiments. Geophysical Research Letters 42, 3855–3862 (2015). + +80. Okada, Y. Internal deformation due to shear and tensile faults in a half-space. Bulletin of the seismological society of America 82, 1018–1040 (1992). + +81. Williams, C. A. & Wadge, G. The effects of topography on magma chamber deformation models: Application to Mt. Etna and radar interferometry. Geophysical Research Letters 25, 1549–1552 (1998). + +<--- Page Split ---> + +82. Storn, R. & Price, K. Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces. Journal of global optimization 11, 341–359 (1997). + +83. Goodman, J. & Weare, J. Ensemble samplers with affine invariance. Communications in applied mathematics and computational science 5, 65–80 (2010). + +84. Foreman-Mackey, D., Hogg, D. W., Lang, D. & Goodman, J. emcee: the MCMC hammer. Publications of the Astronomical Society of the Pacific 125, 306 (2013). + +85. Aki, K. & Richards, P. G. Quantitative seismology. (2002). + +86. Tape, W. & Tape, C. A geometric setting for moment tensors. Geophysical Journal International 190, 476–498 (2012). + +Acknowledgments: We gratefully thank NIED and JMA for providing us the high-quality waveform data of V-net. We are also grateful to Prof. Mare Yamamoto and Prof. Takuhiro Ohkura for hosting J. Niu in Aso Volcanic Observatory and Tohoku University during the early stage of this work, which helps improve our understanding of the LPT and Aso volcanic system. The uses of the UCL Legion High-Performance Computing Facility (Legion@UCL), the UCL Grace High-Performance Computing Facility (Grace@UCL), the UCL Emerald High-Performance Computing Facility (Emerald@UCL), and associated support services are acknowledged. Data processing and production of figures are implemented in Python with relevant modules such as ObsPy and PyCUDA. Funding: J. Niu and T.- R. A. Song are supported by the Natural Environment Research Council, UK (NE/ P001378/1 & NE/T001372/1). Author contributions: J. Niu: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Visualization. T.- R. A. Song: Conceptualization, Methodology, Validation, Formal analysis, Writing - review & editing, Supervision, Project administration, Funding acquisition, Conceptualization.; Competing interests: + +<--- Page Split ---> + +Authors declare no competing interests. Data and materials availability: The waveform data can be downloaded from https://www.hinet.bosai.go.jp. The LPT catalog published by Niu and Song (2020) and all the codes used in this study are available upon request. + +## Supplementary Materials: + +Methods + +Figures S1- S14 + +Tables S1- S2 + +References (###- ###) + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Aso volcano system and observations of tilt offset synchronous with LPT. (a) The topographic relief near the Aso caldera is overlaid with collocated seismometer and tiltmeter (open triangles), LPT source in the shallow crack-like conduit (thick bar), the geodetically inferred magma chamber (open circle), and the low-velocity zone (dashed circle). The star marks the Naka-dake first crater. The upper inset displays Aso volcano located in Kyushu island, southwest Japan. The source of the tilt offset is marked as a red cross and it is referred to as a shallow magma storage zone (SMSZ). The beach ball displays the deviatoric component of the source mechanism in SMSZ. (b) Tilt waveforms at four V-net stations near the October 8, 2016 phreatomagmatic eruption. The waveforms are low-pass filtered at 0.05 Hz with a \(4^{\text{th}}\) -order casual Butterworth filter. The linear trend determined in the first 10 minutes is used to remove the background trend. “Event 1” and “Event 2” denote the tilt offset that occurred concurrently with two LPT events. Redline marks the timing of the 2016 eruption.
+ +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Global inflation stack [Nov 2011-Aug 2014]
+ +![](images/Figure_2.jpg) + +
Figure 2. Observations of inflation (a) and deflation (b) waveform stacks against LPT and SPT. Broadband tilt, displacement, and velocity of global waveform stacks at station N.ASHV are presented against LPT velocity waveforms (10-30 s) and SPT envelopes (0.5-1 s). The bootstrap errors are shown in yellow strips. These waveforms and envelopes are aligned against the onset of SPT. To preserve the phase of the long-period waveform (> 50 s), a \(2^{\text{nd}}\) -order acausal Butterworth filter is implemented. We apply a \(2^{\text{nd}}\) -order causal Butterworth filter to preserve the timing of the LPT waveform (10-30 s) and SPT envelope (0.5-1 s).
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Source inversion result and a new conceptual framework of magma/gas transport in the plumbing system beneath Aso volcano. (a), (b) and (c) compares data (black arrows) and synthetics (red arrows) from the inverted source location and mechanism. Red cross marks the inverted source location with a confidence level of \(99\%\) . Black, red and blue circles indicate the new Mogi magma chamber40, BYA and BCU magma chambers49, respectively. Dashed circle marks the low-velocity zone39. (d) An east-west cross section showing a conceptual diagram of magma/gas transport system beneath Aso volcano. The inverted source, or the shallow magma storage zone (SMSZ), connects the LPT source in the shallow conduit from above and the magma chamber from below. SPT source is shown directly above the shallow conduit. The inset displays the volumetric and deviatoric components of the focal mechanism, respectively, looking from the south. Blue dashed line indicates the brittle-ductile transition. (e), (f) and (g) display the simplified flow conditions where the SMSZ suffers inflation, null volume change, or
+ +<--- Page Split ---> + +deflation, respectively. Waveform stacks representative of these three conditions are shown directly below. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. Characterization of the volume/mass change inside the SMSZ during the 2011-2016 eruption cycle (a) The accumulative net volume changes in the SMSZ. Dark inverse triangles mark the onset of the 2014 Strombolian eruption, the 2015 and 2016 phreatomagmatic eruptions, whereas hatched areas mark their eruption episodes. Grey inverse triangles indicate minor eruptions. \(\oplus\) , \(\ominus\) and \(\odot\) mark the episodes associated with observations of inflation event, deflation event and event of negligible offset, respectively (see waveform stacks in the insets). \(①\) and \(②\) mark the time windows where the mass changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively (see also Fig. 4e). (b) Outgassing potential of the shallow conduit, or equivalently the moment ratio between pressurization and depressurization LPTs \(^{31}\) . A lower (higher) outgassing potential corresponds to a higher proportion of pressurization (depressurization) LPT events. (c) The campaign SO \(_2\) emission from JMA (https://www.data.jma.go.jp/svd/vois/data/fukuoka/rovdm/Asosan_rovdm/gas/gas.html). (d) The volume change rates estimated by the inflation event (black crosses), the geodetic data (green cross), and
+ +<--- Page Split ---> + +geological analysis (blue cross). Average volcanic output rates from White et al. \(^{59}\) are shown in the pink shaded regions. (e) The observed volcanic output of the 2014 Strombolian and 2016 phreatomagmatic eruptions against the estimated mass change inside the SMSZ before the eruptions. \(①\) and \(②\) mark the time windows in Fig. 4a where the mass changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. Comparisons of the mass flow rate and decompression rate from petrology and seismology. Open circles and triangles mark the volume estimates from waveform stacks of inflation and deflation events before the 2014 and 2016 eruptions (see details in the main text). Target crosses mark the median volume estimates from the five subsets (Fig. S13a-e). Solid circles mark previous compilations in basaltic volcanoes \(^5\) , where the decompression rates are estimated from water diffusion in olivine or embayment studies and the mass discharge rates are estimated from volcanic outputs or plume heights, respectively. Contour lines display a theoretical estimate of decompression rate and mass flow rate against fixed conduit radius (see equation 4 in Barth et al. \(^5\) ). Hatched areas show the mass discharge rate of the 2014 and 2016 eruptions (see details in the main text). Dashed line indicates the boundary of magmastatic-dominated and friction-dominated decompression after Barth et al. \(^5\) .
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. A conceptual diagram of episodic deformation in the chamber accompanying repetitive shallow volcanic resonances, providing the glue to long-term volcanic output and short-term magma ascent/discharge rates. Green solid bars indicate the global observations of the average volcanic output rates over eruption cycles in geological time scale ( \(\sim 1\) yr–1 Myr). Gray dashed bars indicate the single-event magma ascent/discharge rate over a seismic time scale ( \(\sim 100\) s). Composition, viscosity, rheology and tectonic settings govern the recurrences of episodic deformation (red lines).
+ +<--- Page Split ---> + +## Methods + +1. Systematic detection using the matched filter (the inflation/deflation event) + +A robust observation of the tilt or displacement offset shown in Fig. 1b and Fig. S1 requires a high signal- to- noise ratio that is not routinely achievable. Instead, we measure the energy in the ultra- long- period band and use it as a proxy for the static offset30. Here we use the matched filtered technique31,35 to systematically examine the presence and the stability of such synchronous signals against the LPT event catalog during the 2011- 2016 Aso eruption cycle31. + +The first of the two LPT events before the 2016 eruption (Event 1 in Fig.1) is selected as the reference template event. The waveforms at two long- running stations, N.ASHV and N.ASIV, are cut in a 10- minute window centered at the LPT arrival time and filtered in the period band of 50- 250 s using a 3rd- order causal Butterworth filter to obtain the template waveforms. The waveforms of each LPT event are processed in the same way as the reference event before the matched filter is applied to cross- correlate waveform data against the template waveforms. To avoid the coupling between tilt and translation in the horizontal components73, we only concern vertical- component seismic waveform data and borehole tilt waveform data in the signal detection. Depending on the cross- correlation coefficient (CC) and signal- to- noise ratio (SNR) between waveform data and the template waveforms (Table S1), the detection against the entire LPT catalog of more than 200,000 events is divided into five subsets (I- V) (Table S2). + +## 2. Waveform stacking + +After removing the mean and trend of raw seismic waveforms, we align and stack the waveforms against the LPT event time to produce the waveform stacks, which greatly enhances the signal and eliminates incoherent noise. To minimize the bias introduced by arbitrary spikes as well as the energy from large earthquakes, we normalize raw velocity waveforms of the single event against their vertical peak- to- peak velocity in the period band of 10- 30 s at station N.ASHV before stacking. After stacking the + +<--- Page Split ---> + +normalized waveforms, the waveform stacks are multiplied by the median vertical peak- to- peak velocity in the period band of 10- 30 s to recover the amplitude of the waveform stack in a given month. The bootstrap technique is used to quantify the uncertainties of the waveform stacks36. After deconvolving the instrument response up to 1000 seconds, we remove the background mean in the first 150 sec of the velocity waveform stacks, perform a causal low- pass filter, and integrate the velocity waveform stacks to obtain displacement waveform stacks. + +Regarding tilt waveform stacks, we first differentiate the raw tilt waveform to obtain the tilt- rate waveform and apply the same data processing procedure as done against velocity waveforms. This stacking procedure not only reduces noise but also helps remove unwanted bias associated with tides, pressure, temperature, or rainfall, which are nonetheless trivial in the data window of 10 minutes. + +3. Estimate the static displacement offset of the broadband seismometer + +As the seismic recording is also sensitive to sensor tilt, it is often difficult to recover true translational signal74. This tilt- related signal is often exhibited as a non- linear drift in the original displacement seismogram74. Due to the projection of gravitational force into the horizontal axis, this phenomenon is particularly severe in the horizontal components. Many approaches have been proposed to separate translational and tilt signals in seismic recordings73,75- 79. Here we estimate and remove the nonlinear drift in the displacement seismogram by least- square fitting the portion of seismogram without major signals with a polynomial function75,76. + +The pre- event window is set to 20 s and the total duration of seismograms is 120 s. A \(4^{\text{th}}\) - order polynomial function is used as suggested by Zhu76. After removing the non- linear drift (Fig. S11), the displacement offset in the individual component is calculated by differing the average displacements in the post- event (mean: \(d^{1}\) , standard deviation: \(\sigma^{1}\) ) and pre- event (mean: \(d^{0}\) , standard deviation: \(\sigma^{0}\) ) windows as \(d = d^{1} - d^{0}\) , with uncertainty, \(\sigma = \sqrt{(\sigma^{0})^{2} + (\sigma^{1})^{2}}\) . + +<--- Page Split ---> + +As indicated in the previous study \(^{78}\) , the static displacement from this approach may be affected by the selection of polynomial order and time window \(^{78}\) . In our case, the determination of the time window of the major signal of the reference event is relatively straightforward. As presented in Tab. S2, the selection of polynomial order does not significantly change the estimations of the point- source location and source geometry. + +## 4. Source properties of the inflation/deflation event + +We followed the mixed linear- nonlinear source inversion scheme in a Bayesian framework proposed by Fukuda and Johnson \(^{38}\) to obtain the location and mechanism of the reference event (Event 1 in Fig. 1b). This mixed inversion method efficiently handles non- linear inversion problems when some of the parameters are linearly related to the observations. The offsets of the vertical displacement, horizontal displacement and horizontal tilt are modeled with Okada's analytic deformation model for a point source \(^{80}\) . + +### 4.1 Problem formulation + +Considering a mixed linear- non- linear inversion problem \(^{38}\) associated with a point dislocation/force source proposed by Okada \(^{80}\) : \(\mathbf{d} = \mathbf{G}(\mathbf{m})\mathbf{s} + \epsilon\) where \(\mathbf{G}(\mathbf{m})\) is the kernel matrix relating non- linear model parameters, \(\mathbf{m} = [x_{s}, y_{s}, z_{s}, \theta , \phi ]^{T}\) (3- D Cartesian location of a point source with respect to the Naka- dake first crater, strike and dip angles of the fault plane and crack plane), and linear model parameters, \(\mathbf{s}_{1} = [m_{ss}, m_{ds}, m_{tc}, m_{ex}]^{T}\) (the scalar moments of strike- slip, dip- slip, tensile crack and explosion) or \(\mathbf{s}_{2} = [m_{ss}, m_{ds}, m_{tc}, m_{ex}, f_{E}, f_{N}, f_{U}]^{T}\) (the scalar moments of strike- slip, dip- slip, tensile crack and explosion, the magnitudes of forces towards east, north, and up directions), to the observation vector, \(\mathbf{d} = [\mathbf{d}_{1}, \mathbf{d}_{2}, \mathbf{d}_{3}]^{T}\) (vertical displacement, horizontal displacement, and horizontal tilt), with normally distributed error, \(\boldsymbol {\epsilon} = [\boldsymbol {\epsilon}_{1}, \boldsymbol {\epsilon}_{2}, \boldsymbol {\epsilon}_{3}]^{T}\) where \(\boldsymbol{\epsilon}_{i} \sim \mathcal{N}(\mathbf{0}, \mathbf{D}_{i})\) . \(J^{th}\) minimum norm constraints (or + +<--- Page Split ---> + +Tikhonov regularization) are applied on the linear parameters, \(\left\| R_{j}\mathbf{s}\right\|^{2}\) , where \(J = 2\) if forces are considered. Otherwise, \(J = 1\) . + +An optimal solution can be found by minimizing an objective function \(\Phi (\pmb {m},\pmb {s}) = \sum_{i = 1}^{I}\frac{1}{\sigma_{i}^{2}} [\pmb {d}_{i} - \pmb {G}_{i}(\pmb {m})\pmb {s}]^{\top}\pmb {D}_{i}^{- 1}[\pmb {d}_{i} - \pmb {G}_{i}(\pmb {m})\pmb {s}] + \sum_{j = 1}^{J}\frac{1}{\beta_{j}^{2}}\left\| \pmb {R}_{j}\pmb {s}\right\|^{2}\) where \(\sigma_{i}^{2}\) adjust relative weights of the multiple data sets in fitting the data; \(\beta_{j}^{2}\) adjust relative weights of regularization prior on \(\pmb{s}\) . If \(J = 1\) , \(\pmb{R}_{1} = \pmb{I}\) . If \(J = 2\) , \(\pmb{R}_{1} = \left( \begin{array}{cc}I_{4\times 4} & 0\\ 0 & 0 \end{array} \right)\) and \(\pmb{R}_{2} = \left( \begin{array}{cc}0 & 0\\ 0 & I_{3\times 3} \end{array} \right)\) . Following Fukuda and Johnson \(^{38}\) , if assuming that the prior probability density functions of the linear, non- linear and hyperparameters are uniform, the posterior probability density function of the parameters given the data, \(p(\pmb {m},\pmb {s},\pmb {\sigma},\pmb {\beta}|\pmb {d})\) , can be given according to Bayes' theorem as below: + +\[-\ln p(\pmb {m},\pmb {s},\pmb {\sigma},\pmb {\beta}|\pmb {d})\] \[\qquad \propto \sum_{i = 1}^{I}2N_{i}\ln \sigma_{i} + \sum_{j = 1}^{J}2M_{j}\ln \beta_{j} + \Phi (\pmb {m},\pmb {s}^{*})\] \[\qquad +\ln \left|\sum_{i = 1}^{I}\frac{1}{\sigma_{i}^{2}} G_{i}^{T}(\pmb {m})\pmb{D}_{i}^{-1}\pmb{G}_{i}(\pmb {m}) + \sum_{j = 1}^{J}\frac{1}{\beta_{j}^{2}} R_{j}^{T}\pmb {R}_{j}\right|\] + +where \(\pmb{s}^{*}\) is the weighted damped least- square solution of the linear parameter given the nonlinear and hyperparameters, \(\pmb {m}\) , \(\pmb{\sigma}\) and \(\beta\) ; \(N_{i}\) is the number of samples in \(i^{t h}\) data set; \(M_{j}\) is the number of parameters efficiently regulated by \(j^{t h}\) prior constraint. If \(J = 1\) , \(M_{1} = 4\) . If \(J = 2\) , \(M_{1} = 4\) and \(M_{2} = 3\) . + +To tackle the weights between displacement and tilt, we first normalize the kernel matrix \(\pmb {G}(\pmb {m})\) , the observation vector \(\pmb{d}\) and the covariance matrix \(\pmb{D}\) by the absolute of observation vector as \(\pmb {K} = \left( \begin{array}{ccc}1 / |d_{1}| & \dots & 0 \\ \vdots & \ddots & \vdots \\ 0 & \dots & 1 / |d_{N}| \end{array} \right)\) , \(\pmb {G}^{\prime}(\pmb {m}) = \pmb {K}\pmb {G}(\pmb {m})\) , \(\pmb{d}^{\prime} = \pmb {K}\pmb {d}\) , and \(\pmb{D}^{\prime} = \pmb {K}\pmb {D}\pmb{K}^{T}\) . Then, by utilizing Cholesky + +<--- Page Split ---> + +factorization, we equalize the weight difference among the observations as \(D^{\prime} = L^{T}L\) , \(d^{\prime \prime} = L d^{\prime}\) , \(G^{\prime \prime}(m) = L G^{\prime}(m)\) . Then, if let \(\sigma_{i}^{2} = 1\) , we can rewrite the objective function as \(\Phi (m, s) = (y - A s)^{T}(y - A s)\) where \(y = [d^{\prime \prime}, 0]^{T}\) and \(A = \left[G^{\prime \prime}(m), \frac{R_{0}}{\beta_{0}}, \dots , \frac{R_{J}}{\beta_{J}}\right]^{T}\) . In this case, the posterior probability density function of the parameters can be re- written as \(- \ln p(m, s, \beta | d) \propto \sum_{j = 1}^{J} 2 M_{j} \ln \beta_{j} + \Phi (m, s^{*}) + \ln |A^{T} A|\) where \(s^{*} = (A^{T} A)^{- 1} A^{T} y\) . + +### 4.2 Synthetic static displacement/tilt calculation + +Synthetic displacements and tilts from a point dislocation source in a homogeneous half space are calculated following Okada80. Following Legrand et al.30, we set the P velocity of 1500 m/s, S- velocity of 800 m/s and the density of 1700 kg/m3, respectively. To evaluate the significance of the single force component, we also calculate synthetic displacement and tilt from a point force source following equations 1- 2 in Okada80. To take into account the station elevation (i.e., typically 600 m above sea level), we set the station elevation as the free surface and the depth of the source is adjusted accordingly in the calculation of synthetics81. Since the borehole tiltmeter is typically about 100 meters below the surface, we calculate synthetic tilts at the borehole depth. + +### 4.3 Bayesian inference of model parameters + +We first use differential evolution method82 to find the global minimum of \(- \ln p(m, s, \beta | d)\) . This step is repeated for hundreds of times with the randomly assigned initial values. A consistently converged minimum is regarded as the global minimum. Subsequently, we use an ensemble Monte Carlo Markov Chain method83,84 to sample the posterior probability density function. The starting values of ensemble Markov chains of \(k^{th}\) parameter are randomly assigned according to the global minimum by a normal distribution, \(N(x_{k}, |x_{k}|)\) , where \(x_{k} \in \{m, s, \beta \}\) . This step is also repeated for multiple times to evaluate the convergence of independent Markov chains. Finally, the posterior probability density functions of the parameters are combined from multiple independent chains (Fig.S12). The moment- tensor representation + +<--- Page Split ---> + +of the source can be reconstructed from the moments of strike-slip, dip-slip, tensile crack and explosion by following equation 3.21 in Aki and Richard85. + +## 5. Estimating volume change + +### 5.1 Volume change of the reference event + +As the strike- slip and dip- slip components do not experience any volume changes, the volume change associated with the reference event is estimated by only considering the tensile crack and explosion components. Following Okada80, the eigenvalue of tensile- crack moment tensor is \(\Lambda_{\mathrm{tc}} = m_{\mathrm{tc}}[1,1,1 + 2\mu /\lambda ]^{T}\) and the eigenvalue of explosion moment tensor is \(\Lambda_{\mathrm{ex}} = m_{\mathrm{ex}}[1,1,1]^{T}\) . Following equation 16 in Kawakastu and Yamamoto18, the volume change associated with the tensile crack component is \(\Delta V_{m}^{tc} = m_{tc} / \lambda\) where \(\lambda\) is Lame constant. Following equations 10- 11 in Kawakastu and Yamamoto18, the volume change associated with the explosion component in the presence of the confining pressure of the surrounding elastic medium is \(\Delta V_{m}^{ex} = m_{ex} / (\lambda + 2\mu)\) where \(\mu\) is shear modulus. Hence, the 'Mogi volume' change associated with the reference event is \(\Delta V_{m} = \Delta V_{m}^{ex} + \Delta V_{m}^{tc}\) . If assuming \(\lambda = 2.737 GPa\) and \(\mu = 1.088 GPa\) , \(\Delta V_{m} = \sim 25840 m^{3}\) (Tab.S2, a \(4^{\mathrm{th}}\) - order polynomial function). + +### 5.2 Estimate volume change from monthly tilt waveform stacks + +Following the note in section 5.1, the Mogi volume change corresponding to a 1- nrad tilt at N.ASHV.LE is \(\sim 440 \mathrm{m}^{3}\) . However, unlike the reference event, the static tilt offset in the monthly stacks can be difficult to measure directly in the time domain. As the duration of the tilt offset is typically \(\sim 100 \mathrm{s}\) or less, the amplitude of the tilt- rate spectra plateaus in the near- field at the very long- period (i.e., \(>1000 \mathrm{s}\) ), corresponding to the amplitude of the tilt (Fig. S13). We zero- pad the stacked waveform 10,000s before the start of the tilt- rate time series and compute the tilt- rate amplitude spectra. The amplitude at the period of 10,000 s is used as a proxy for the static offset, which is scaled against the tilt of the reference + +<--- Page Split ---> + +event to obtain the volume change per event associated with each monthly stack. The monthly volume change can be obtained by multiplying the event number against the volume change per event (Fig.S14). + +## 6. Estimate magma ascent velocity and decompression rate + +Following the 1- D conduit flow model by Jaupart & Tait \(^{57}\) , the conduit radius \(R\) is balanced by the dynamic pressure of the ascended magma and lithospheric gradient \(R^{4} = G \frac{8\eta}{\pi \rho (\rho_{r} - \rho)g}\) , where \(G\) is the mass flow rate, \(\eta\) is magma viscosity, \(\rho_{r}\) is the density of wall rock, \(\rho\) is the density of magma, and \(g\) is the gravitational acceleration. To estimate the conduit radius, the mass flow rate is estimated by the volume change rate of the inflation/deflation event and magma density \(\rho\) . The magma ascent velocity \(V_{a}\) is estimated as \(V_{a} = \frac{G}{\pi \rho R^{2}}\) . Assuming a lithospheric gradient, the decompression rate \(dP / dt\) is estimated as \(dP / dt = \rho_{r}gV_{a}\) . + +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
Fig. S1. Examples of single-event tilt and displacement waveforms at N.ASHV. (a) Low-pass filtered displacement and tilt waveforms with the upward and east-down offsets, respectively. (b) Low-pass filtered displacement and tilt waveforms with the downward and west-down offsets, respectively. The starting time of each waveform is indicated above each trace.
+ +<--- Page Split ---> +![](images/Figure_unknown_2.jpg) + +
Fig. S2. Global and monthly waveform stacks against highest quality data (subset I). (a) the global waveform stacks of inflation events during the unrest (November 2011 to August 2014). (b) the global waveform stacks of deflation events (October 2014 to April 2015) (c) the monthly waveform stack of inflation events in July 2014. (d) the monthly waveform stack of deflation events in November 2014. LPT signal is present in the middle of the time window at \(\sim 220 - 270\) s. Waveform stacks from stations N.ASHV and N.ASIV are shown in black lines and green lines, respectively. The orange shaded region marks the \(99.7\%\) confidence interval estimated by the bootstrapped resampling (ref.36). Displacement and tilt waveform stacks are low-pass filtered at 10 s and 30 s with a 4th-order casual Butterworth filter, respectively. The number of events used in each waveform stack is shown in the upper right.
+ +<--- Page Split ---> +![](images/Figure_unknown_3.jpg) + +
Fig. S3. 1-hour long global and monthly waveform stacks against highest quality data (subset I). Same as Fig. S2, expect only waveform data associated with isolated LPT events are included in the stack. The number in the left panel indicates the number of the well-isolated LPT used in the stacks. (a)-(d) the global and monthly tilt-rate waveform stacks. (e)-(h) global and monthly tilt stacks. The trend defined in [-10,-2] minutes of the tilt waveform stacks is used to remove the background trend. The gray shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).
+ +<--- Page Split ---> +![](images/Figure_unknown_4.jpg) + +
Fig. S4. Synthetic amplitude-distance decay against static and filtered waveforms. (a) and (b) display normalized displacement amplitude-distance decay trend with a source depth \(0.5 \mathrm{km}\) and \(5.0 \mathrm{km}\) , respectively. (c) and (d) display normalized tilt amplitude-distance decay trend with a source depth \(0.5 \mathrm{km}\) and \(5.0 \mathrm{km}\) , respectively. Static, long-period (10-30 s), and ultra-long-period (100-200 s) amplitude measurements are shown in green lines, blue lines and red lines, respectively.
+ +<--- Page Split ---> +![](images/Figure_unknown_5.jpg) + +
Fig. S5. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. (a)-(e) the amplitude ratio in the LPT frequency band (10-30 s) against five subsets, respectively (the subset I→the highest quality and the subset V→the lowest quality). The amplitude ratio measured in each monthly stack is shown in shaded square. The color of the square denotes the number of stacked inflation events in a linear scale (dark→higher number and light→lower number). The error bar marks the uncertainty with a 99.7% confidence level estimated from the bootstrap resampling (ref.36). The red line marks the amplitude ratio measured from the global inflation waveform stack (Fig. S2a). The green diamond marks the amplitude ratio from Event 1. (f)-(j) Same as (a)-(e), except for measurements in the filtered-static-offset (FSO) frequency band (100-200 s). (k)-(o) the number of events used in each monthly stack against five subsets, respectively. The amplitude ratio in the FSO frequency band becomes more variable when the data quality is low or/and the number of events in the monthly stack is low.
+ +<--- Page Split ---> +![](images/Figure_unknown_6.jpg) + +
Fig. S6. Amplitude ratios between north-south and east-west tilt at N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.
+ +<--- Page Split ---> +![](images/Figure_unknown_7.jpg) + +
Fig. S7. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.
+ +<--- Page Split ---> +![](images/Figure_unknown_8.jpg) + +
Fig. S8. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S5, except for the monthly deflation waveform stacks. The red line marks the amplitude ratio measured from the global deflation waveform stack (Fig. S2b).
+ +<--- Page Split ---> +![](images/Figure_unknown_9.jpg) + +
Fig. S9. Amplitude ratios between north and east tilts at N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.
+ +<--- Page Split ---> +![](images/Figure_unknown_10.jpg) + +
Fig. S10. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.
+ +<--- Page Split ---> +![](images/Figure_unknown_11.jpg) + +
Fig. S11. Broadband displacement waveforms before and after the correction. (a-c) the original broadband displacement waveforms of the reference event in the vertical, east-west and north-south components at the four broadband seismic stations before removing the baseline error estimated from the polynomial fit. (d-f) the broadband displacement waveforms after removing the baseline errors. A \(4^{th}\) -order polynomial function is used to estimate the baseline error from the original displacement waveforms in the pre-event and post-event time windows. Vertical dashed lines mark the starting and ending times of the event (see also Method).
+ +<--- Page Split ---> +![](images/Figure_unknown_12.jpg) + +
Fig. S12. Posterior probability functions of the inverted source parameters. (a-c) the marginal posterior probability functions against the source location with respect to the Naka-dake first crater. (d-e) the marginal probability functions of the strike and dip angles of the shearing and tensile components. (f) the marginal probability function against the regularization parameter \(\beta\) . (g-j) display the marginal probability functions against the moments of the strike-slip (Mss), dip-slip (Mds), tensile-crack (Mtc), and isotropic (Mex) components, respectively. (k-l) display the joint posterior probability functions of the coupled parameters regarding the bimodal probability functions in (d) and (e). The red line marks the preferred parameters with the maximum posterior probability. Two peaks in (d) and (e) define the orientations the fault plane and the auxiliary plane.
+ +<--- Page Split ---> +![](images/Figure_unknown_13.jpg) + +
Fig. S13. Measurement of tilt offset in the spectra domain. (a) Synthetic tilt waveforms with the positive (red), null (gray), and negative (blue) offsets of \(1 \mu m\) . (b) Synthetic tilt waveform with a large offset of \(50 \mu m\) . (c) Synthetic tilt-rate amplitude spectra normalized against the amplitude of (b) at 10,000 s. Static offsets measured directly from (a) and (b) are normalized against the one from (b), shown in squares. (d) Observed monthly tilt waveform stacks with the positive (red), near-null (gray), and negative (blue) offsets. (e) Observed tilt waveform of the reference event. (f) Observed tilt-rate amplitude spectra normalized against the amplitude of (e) at 10,000 s. Static offsets measured directly from (d) and (e) are normalized against the one from (e), shown in squares. These examples show that the amplitude plateau of the tilt-rate waveform at 10000 s provides a reasonable estimate of the static offset.
+ +<--- Page Split ---> +![](images/Figure_unknown_14.jpg) + +
Fig. S14. Monthly volume changes in the five subsets. (a)-(e) Histograms of the volume change per event estimated from the monthly tilt waveform stacks in five subsets, respectively. Results from the monthly inflation and deflation stacks are shown in red and blue, respectively. (f-l) Same as (a)-(e), except displaying the monthly volume changes. The shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).
+ +<--- Page Split ---> + +Table S1: Classifications of the ultra-long-period detections against the LPT catalog. + +
SubsetDefinitionNumberPercentage
I\(CC\geq 0.68\)and \(\mathrm {SNR}\geq 6.9\)55622.7%
II\(CC\geq 0.58\)and \(\mathrm {SNR}\geq 5.3\)(excluding I)114165.5%
III\(CC\geq 0.42\)and \(\mathrm {SNR}\geq 3.7\)(excluding I-II)3576617.1%
IV\(CC\geq 0.30\)and \(\mathrm {SNR}\geq 2.8\)(excluding I-III)4596422.0%
V\(CC<0.30\)or \(\mathrm {SNR}<2.8\)10980952.7%
+ +<--- Page Split ---> + + +Table S2: Summary of inverted source parameters for the reference event. + +
Polynomial orderxsys [km]zsφ [°]θ [°]mssmds [×1014N·m]mtcmexfE [×108N]fN [m3]fUΔVn [m3]Mf:MDMw [°]γ [°]δ [°]
3-2.33-0.642.41324480.007-0.258-0.1731.006158023.2:13.313.272.6
4-1.96-1.022.9833160-0.021-0.283-0.0020.993258394.3:13.30.176.8
5-1.97-1.082.8433460-0.014-0.2860.0190.903247754.0:13.3-1.575.9
6-2.05-0.353.55313480.003-0.395-0.1381.583330154.2:13.47.476.4
4-1.97-1.143.0132758-0.015-0.275-0.0110.966-0.005-0.0050.005246104.2:13.30.976.7
+ +Note: \(x_{s}, y_{s}\) and \(z_{s}\) are the 3- D Cartesian location of the preferred source along the eastward, northward and downward directions referred to the Naka- dake first crater, respectively; \(\phi\) and \(\theta\) are the strike and dip angles of the shearing and tensile components, respectively; \(m_{ss}\) , \(m_{ds}\) , \(m_{tc}\) and \(m_{ex}\) are the moments of strike- slip, dip- slip, tensile crack and explosion, respectively; \(f_{E}\) , \(f_{N}\) and \(f_{U}\) are the magnitude of singe force along the eastward, northward and upward directions, respectively; \(\Delta V_{m}\) is the equivalent Mogi volume change; \(M_{I}\) and \(M_{D}\) are the isotropic and deviatoric scalar seismic moments of the reconstructed moment tensor (ref.30), respectively; \(\gamma\) and \(\delta\) are the longitude and latitude with pole (1, 1, 1) for the fundamental lune (ref.86), respectively. + +<--- Page Split ---> diff --git a/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0_det.mmd b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..23eabe3ec3cfe9250362c2b8922560c22816e044 --- /dev/null +++ b/preprint/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0/preprint__1287b0275a1ceb77692926e47e4cb244d3ebcdcf9220bb8cc464a7bb7c810ed0_det.mmd @@ -0,0 +1,704 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 855, 174]]<|/det|> +# Episodic transport of discrete magma batches beneath Aso volcano + +<|ref|>text<|/ref|><|det|>[[44, 195, 641, 238]]<|/det|> +Jieming Niu ( j.niu@ucl.ac.uk ) University College London https://orcid.org/0000- 0002- 4481- 022X + +<|ref|>text<|/ref|><|det|>[[44, 243, 550, 284]]<|/det|> +Teh- Ru Song Department of Earth Sciences, University College London + +<|ref|>sub_title<|/ref|><|det|>[[44, 325, 102, 342]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 362, 936, 428]]<|/det|> +Keywords: Trans- crustal Magma Plumbing System, Long- term Volcanic Output, Short- term Eruption Dynamics, Episodic Long- period Tremors, Synchronous Deformation Events, Brittle- to- ductile Transition Rheology + +<|ref|>text<|/ref|><|det|>[[44, 445, 295, 464]]<|/det|> +Posted Date: May 26th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 483, 463, 503]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 408334/v1 + +<|ref|>text<|/ref|><|det|>[[44, 521, 911, 564]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 600, 916, 643]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 21st, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 25883- y. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[115, 98, 677, 115]]<|/det|> +# Title: Episodic transport of discrete magma batches beneath Aso volcano + +<|ref|>text<|/ref|><|det|>[[115, 136, 454, 154]]<|/det|> +Authors: Jieming Niu \(^{1*}\) , Teh- Ru Alex Song \(^{1\dagger}\) + +<|ref|>sub_title<|/ref|><|det|>[[115, 177, 209, 193]]<|/det|> +## Affiliations: + +<|ref|>text<|/ref|><|det|>[[115, 214, 839, 264]]<|/det|> +\(^{1}\) Seismological Laboratory, Department of Earth Sciences, University College London, WC1E 6BT, London, United Kingdom. + +<|ref|>text<|/ref|><|det|>[[115, 287, 388, 304]]<|/det|> +\*Correspondence to: j.niu@ucl.ac.uk. + +<|ref|>text<|/ref|><|det|>[[111, 323, 886, 890]]<|/det|> +Summary: Magma ascent, storage, and discharge in the trans- crustal magma plumbing system are key to long- term volcanic output and short- term eruption dynamics. Petrological analytics, geodetic deformations and mechanical modeling have shaped the current understanding of magma transport. However, due to the lack of observations, how a distinct magma batch transports from a crystal- rich mush region to a crystal- poor pool with eruptible magma remains enigmatic. Through stacking of tilt and seismic waveform data, we find that episodic long- period tremors (LPTs) located near sea level beneath the Aso volcano are accompanied by a synchronous deformation event, which initiates \(\sim 50\) seconds before individual LPT event and concludes seconds after. The episodic deformation source corresponds to either an inflation or a deflation event located \(\sim 3\) km below sea level, with a major volumetric component (50- 440 cubic meters per event) and a minor high- angle normal- fault component. We suggest that these deformation events likely represent short- lived, episodic upward transport of discrete magma batches accompanied by high- angle shear failure near the roof of the inferred magma chamber at relatively high temperature, whereas their recurrences, potentially composition dependent, are regulated by the brittle- to- ductile transition rheology under low differential stress and high strain rate due to the surge of magma from below, regulating long- term volcanic output rate. The magma ascent velocity, decompression rates, and cumulative magma output deduced from the episodic deformation events before recent eruptions in Aso volcano are compatible with retrospective observations of the eruption style, tephra fallouts, and plume heights, promising real- time evaluation of upcoming eruptions. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 129, 203, 145]]<|/det|> +## Main Text: + +<|ref|>text<|/ref|><|det|>[[112, 196, 886, 440]]<|/det|> +Theoretical considerations have shown that the ascending of a magma batch is likely dictated by the density contrast between magma and surrounding rock, the excess pressure in the magma reservoir, and magma viscosity'. More recent developments on magma transport and storage emphasize a transcrustal magmatic plumbing system consisting of relatively long- lived crystal- rich mushes and shorter time- scale crystal- poor pools that were eventually tapped by eruptions?. However, the key process of distinct magma transport from the crystal- rich mush to the crystal- poor pool is not well known, even though it is intimately linked to magma ascent and discharge, controlling short- term eruption dynamics and long- term volcanic output. + +<|ref|>text<|/ref|><|det|>[[112, 488, 886, 700]]<|/det|> +Numerous petrological means have been utilized to interrogate magma ascent, including recent chemical diffusion based geospeedometers- 9. However, these advancements rely on retroactive analysis of eruption products, which cannot be obtained in real- time during unrest or before eruptions. Furthermore, these estimates do not resolve the volume or duration of distinct magma ascent. Magma ascent inferred from the migration of seismicity can be difficult as seismicity rarely follows simple upward movement and the progression of seismicity could indicate conduit formation or propagating of pressure through existing magma, rather than magma ascent1,10,11. + +<|ref|>text<|/ref|><|det|>[[112, 748, 886, 895]]<|/det|> +Ground deformation detected from geodetic means (e.g., InSAR, GNSS, tiltmeter, strainmeter) provides invaluable insight into the average rate of magma supply12- 14. However, except few instances where syn-eruptive deformation of 0.1- 10 microstrain over hours or less are reported15- 17, the documented pre-eruptive ground deformation is typically on the order of millimeter to centimeter, microstrain, or microradian over days, weeks, or months and they do not offer a direct assessment on the nature of a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +distinct magma ascent, potentially operating at a much shorter duration, higher frequency or/and smaller volume. + +<|ref|>text<|/ref|><|det|>[[112, 188, 885, 368]]<|/det|> +Seismic resonances with periods of 0.2- 200 s have been widely detected near shallow conduits or reservoirs (e.g., \(< 2 \mathrm{km}\) ) in diverse volcanic systems \(^{18,19}\) . They are often stationary and repetitive, responding to internal processes accompanying magma ascent or/and the build- up of conduit overpressure, including degassing, heat transfer and vaporization, change in the fluid pressure, or/and unsteady magma transport of a deeper origin. The repetitive nature of shallow seismic resonance permits the use of stacking to tease out small deformation signal \(^{20 - 22}\) that are otherwise undetected. + +<|ref|>text<|/ref|><|det|>[[112, 416, 885, 660]]<|/det|> +Here we report new observations that manifest a causal relationship between repetitive shallow seismic resonances (i.e., shallow hydrothermal reservoir) and deformation within the deeper plumbing system beneath Aso, an andesitic volcano in Japan \(^{23 - 25}\) . Specifically, we focus investigation against the very long- period signal ( \(\sim 15 \mathrm{s}\) period) located in a fluid- filled crack- like conduit near sea level beneath the active Naka- dake first crater \(^{26 - 31}\) (Fig. 1a). Following the nomenclature established in earlier studies, we refer to the very long- period signal as long- period tremor (LPT) hereafter. Critically, LPT is known to occur during the quiescence and active period \(^{27,29,31}\) , offering a unique opportunity to methodically explore discrete magma ascents over the entire eruption cycle in 2011- 2016. + +<|ref|>sub_title<|/ref|><|det|>[[115, 709, 466, 727]]<|/det|> +## LPT and synchronous tilt/displacement offset + +<|ref|>text<|/ref|><|det|>[[112, 770, 885, 885]]<|/det|> +We analyze seismic and tilt waveform recorded by the fundamental volcano observation network, or V- net \(^{32}\) , which is equipped with collocated surface broadband seismic sensors (Nanometrics 240, \(\sim 250 \mathrm{s}\) natural period) and borehole tiltmeters ( \(\sim 1\) nanoradian resolution) \(^{33}\) . Before the \(8^{\mathrm{th}}\) Oct 2016 phreatomagmatic eruption \(^{34}\) , we observe two anomalously large LPT events that occurred about 2- 3 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 172]]<|/det|> +minutes before the eruption (Fig. 1b). Tilt offsets accompanying the two LPT events can be readily observed at all four V- net stations (Fig. 1b). At station N.ASHV, the east- west tilt offset accompanying the two events reach \(\sim 58\) nrad and \(\sim 88\) nrad, respectively. + +<|ref|>text<|/ref|><|det|>[[112, 198, 886, 572]]<|/det|> +We manually check tilt and vertical displacement waveforms of other events in the LPT catalog31 and discover similar observations in several instances (Fig. S1). In general, the tilt and displacement offsets at station N.ASHV can be either east- down and vertical- up or west- down and vertical- down, but the amplitude is typically much smaller than those accompanying the two LPT events before the 2016 eruption. As shown in Fig. 1b, there are several striking differences between the signal of LPT and the tilt offset. First, the tilt offset at station N.ASHV is much stronger in the east- west component than that in the north- south component, whereas the LPT signal in the north- south component is much stronger than that in the east- west component. Secondly, the east- west tilt offset at station N.ASHV is much stronger than that at station N.ASIV, but the LPT signal at station N.ASHV is much weaker than that at station N.ASIV. These observations strongly indicate that the source of the tilt offset is spatially separated from the LPT source, possibly closer to station N.ASHV than the source of LPT, which is near the active Nakadake first crater (see also Fig. 1a). + +<|ref|>text<|/ref|><|det|>[[115, 620, 567, 638]]<|/det|> +Discovery of the inflation/deflation event beneath Aso volcano + +<|ref|>text<|/ref|><|det|>[[112, 688, 886, 867]]<|/det|> +To systematically examine the presence and the stability of such synchronous signals, we implement the matched- filter technique31,35 against all LPT events during the 2011- 2016 Aso eruption cycle31. The first of the two LPT events (Event 1) before the 2016 eruption is selected as the reference template event, and we cross- correlate tilt and seismic waveform data of each LPT event against the template waveforms in the ultra- long period band of 50- 250 s, which allows us to effectively detect possible displacement or tilt offset in the near- field (see also Material and Method). Specifically, we + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +gather detections with the same sign of cross- correlation coefficient in a given calendar time window to produce waveform stacks and compute bootstrap uncertainties36. + +<|ref|>text<|/ref|><|det|>[[112, 167, 886, 475]]<|/det|> +Here we present two global waveform stacks of the highest quality at station N.ASHV during the volcanic unrest in January 2011- August 2014 (Fig. 2a) and the intermittent Strombolian eruptions in October 2014- April 2015 (Fig. 2b), respectively. During the volcanic unrest, we observe an upward displacement offset and a predominantly east- down tilt offset (Fig. 2a). During the 2014 Strombolian eruption, we observe a downward displacement offset and a predominantly west- down tilt offset (Fig. 2b). While the amplitude of the LPT differs by a factor of \(\sim 3\) between the two episodes, the waveform offset is relatively steady, i.e., \(\sim 1 \mu m\) in the vertical displacement and \(\sim 1\) nrad in the east- west tilt, respectively. Small bootstrap uncertainties of the global waveform stacks suggest the prevalence of the signal that is synchronous with LPT. Here we refer to the signal associated with an upward (downward) displacement offset as the inflation (deflation) event. + +<|ref|>text<|/ref|><|det|>[[112, 502, 886, 842]]<|/det|> +While it is well documented that LPT is often accompanied by short- period tremors (SPTs, \(\sim 0.5 - 1\) s period) located near the top of the LPT source region18,27,37, as demonstrated in Fig. 2, the inflation/deflation event occurs \(\sim 60\) seconds earlier than SPT or LPT, providing a causal source of internal triggering. While SPT precedes LPT by \(\sim 10\) seconds, the peak energy of SPT envelopes arrives \(\sim 10\) seconds after the LPT onset. The inflation (deflation) event, LPT, and SPT end approximately at the same time. Observations of inflation/deflation events can also be made in other stations or high- quality monthly waveform stacks (Fig. S2). Notably, the tilt offsets can remain relatively constant over 30 minutes (Fig. S3). We also measure the displacement and tilt amplitude ratio between stations N.ASHV and N.ASIV in the ultra- long period band of 100- 200 s (Fig. S4) against all monthly stacks (Fig. S5- S10). The stability of these observational attributes indicates that repetitive inflation and deflation events share generally the same proximity with a relatively stationary source region. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 300]]<|/det|> +Following the Bayesian source inversion approach by Fukuda and Johnson38, we measure the vertical displacement, horizontal displacement (Fig. S11, see Method), and horizontal tilt offsets of the reference event (Event 1) and invert the source location and mechanism (Fig. 3a- c, see also Material and Method). We find that the source is located at a depth of \(\sim 3\) km below sea level, \(\sim 2\) km west of the Nakadake first crater (Fig. 1a, Fig. 3a- d, see also Table S2). The reference event has a moment magnitude \(\mathrm{M_w} = 3.3\) with a predominantly volumetric component ( \(\sim 80\%\) ) and a minor dip- slip (normal- fault) component of \(\sim 20\%\) (Fig. 1a, Fig. S12, Table S2). The corresponding volume change is \(\sim 25,800 \mathrm{m}^3\) (see Method). + +<|ref|>sub_title<|/ref|><|det|>[[115, 343, 809, 362]]<|/det|> +## The nature of inflation/deflation event in a renewed plumbing system beneath Aso volcano + +<|ref|>text<|/ref|><|det|>[[112, 403, 886, 584]]<|/det|> +The source of the inflation/deflation event is located near the inferred magma chamber39,40 (Fig. 1a) but at a shallower depth. It is also near the upper end of a high conductivity anomaly41. We suggest that the source of the inflation/deflation event corresponds to a shallow magma storage zone (SMSZ) beneath the Aso volcano, connecting the source of LPT in the shallow conduit from above and the magma chamber from below (Fig. 3d), suggesting that these episodic deformation events likely represent instantaneous upward transport of a distinct magma batch near the roof of the inferred magma chamber. + +<|ref|>text<|/ref|><|det|>[[112, 611, 886, 886]]<|/det|> +We consider that the upward transport of a discrete magma batch only occurs in a short episode of \(\sim 50 - 100\) s near the roof of a partially molten zone filled with crystal- rich magma, accompanied by high- angle shear failure. Since the temperature at the depth of SMSZ ( \(\sim 3\) km) beneath Aso volcano reaches \(500^{\circ} \mathrm{C}^{42,43}\) , the recurrences of these deformation events are likely regulated by the brittle- to- ductile transition, under high temperature and high strain- rate44 due to the surge of magma from below, possibly facilitated by ductile fracturing45 or/and magma mixing in the mush46. The deformation event ends as the system returns to the ductile regime as a result of the shear- stress release or/and decreasing magma supply (i.e., lower strain rate). As elaborated in the next session, the transport of discrete magma batches does not exclude volatiles that may readily exist in the top of the mush region47. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 428]]<|/det|> +We suggest that, when the inflow (outflow) is higher than the outflow (inflow), there is a net volume increase (decrease) in the SMSZ and we observe an inflation (deflation) event (Fig. 3e, Fig. 3g). When the inflow is approximately equal to the outflow, there is negligible net volume change in the SMSZ and we observe a negligible offset in tilt or displacement (Fig. 3f). As shown in Fig. 4a, we observe predominantly inflation events during intense unrest in July and August 2014. Right after the ash eruption at the end of August 2014, we observe events with a negligible net offset in September 2014. During the intermittent Strombolian eruptions (e.g., November 2014, March 2015), we observe predominantly deflation events. After the pyroclastic cone collapsed in early May 2015, we again observe inflation events in May and June 2015. While the deflation events dominate in late 2015 and most of 2016, two anomalously large inflation events occur just minutes before the 2016 phreatomagmatic eruption (Fig. 1b). + +<|ref|>text<|/ref|><|det|>[[112, 454, 886, 729]]<|/det|> +These systematic observations also point to a robust and intimate link between surface volcanic activities and the state of the SMSZ. The prominence of inflation (deflation) events corresponds to a lower (higher) outgassing potential (Fig. 4b), equivalently a higher (lower) proportion of pressurization LPT in the shallow conduit31, and a lower (higher) \(\mathrm{SO}_2\) emission (Fig. 4c). Finally, the depth of the SMSZ is very consistent with the highest gas saturation pressure of melt inclusions in scoria (~80- 118 MPa, or 2- 3.5 km below sea level)23, supporting the hypothesis that the magma (and gas) are likely transferred from a deeper reservoir48 (i.e., a crystal- rich mush) and temporarily stalled in the SMSZ (i.e., a crystal- poor pool), which serves as a preparation zone for the storage and upward transport before upcoming eruptions. + +<|ref|>text<|/ref|><|det|>[[113, 772, 883, 822]]<|/det|> +Implications of the inflation/deflation event on magma/gas transport and upcoming magmatic eruption beneath Aso volcano + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 885, 300]]<|/det|> +Our new observations suggest that the upward transport of magma/gas from the magma chamber toward the surface is a stepwise process in an episodic fashion. Since the source is repetitive, the volume change associated with individual inflation/deflation events can be obtained by scaling the tilt amplitude of each event against that of the reference event. For example, the east-west tilt of the reference event at station N.ASHV is \(\sim 58\) nrad, corresponding to a volume increase of \(\sim 25,800 \mathrm{m}^3\) . The typical east-west tilt in the global stacks is only about \(\sim 1\) nrad (Fig. 2), which translates to a volume change of \(\sim 440 \mathrm{m}^3\) per event, or the size of a typical swimming pool. + +<|ref|>text<|/ref|><|det|>[[112, 327, 885, 475]]<|/det|> +We systematically measure the tilt amplitude of the monthly stacks (Fig.S13, see also Material and Method) and estimate the volume change per event, which can be \(\sim 50 \mathrm{m}^3\) in the subset of the lowest signal- to- noise ratio (Fig. S14). The monthly volume change is obtained by multiplying the volume change per event against the event number (Fig. S14) and the cumulative volume change over 2011- 2016 is illustrated in Fig. 4a. + +<|ref|>text<|/ref|><|det|>[[112, 501, 886, 811]]<|/det|> +For a typical source duration of \(\sim 100 \mathrm{s}\) (Fig. 2), the single- event volume change rate estimated from the global stacks or the monthly stacks is \(\sim 0.5 - 4.4 \mathrm{m}^3 /\mathrm{s}\) or \(\sim 0.03 - 0.13 \mathrm{km}^3 /\mathrm{year}\) which is \(\sim 2\) orders of magnitude higher than the average volume change rate in 2011- 2013 (i.e., \(\sim 10^{- 4} \mathrm{km}^3 /\mathrm{year}\) ) and in late 2014 ( \(\sim 1.3 \times 10^{- 3} \mathrm{km}^3 /\mathrm{year}\) ), the geodetic deflation rate of the magma chamber at a greater depth (i.e., \(\sim 2.5 \times 10^{- 4} \mathrm{km}^3 /\mathrm{year}\) over several decades) \(^{49,50}\) and Aso post- caldera output rates (i.e., \(\sim 0.001 \mathrm{km}^3 /\mathrm{year}\) ) \(^{24,25,51}\) (Fig. 4d). We hypothesize that the volume change rate associated with individual inflation/deflation event marks the instantaneous magma transport rate of a distinct magma batch, whereas the recurrence interval modulates the averaged transport rate, reconciling the magma transport over multiple timescales. This hypothesis reconciles similar disparities noted in basaltic volcanic systems such as Kilauea \(^{52}\) and silicic volcanic systems such as Mount St. Helens \(^{53}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 150, 886, 438]]<|/det|> +Considering a magma density of \(2500\mathrm{kg / m}^3\) , we convert the net volume change \(\Delta \mathrm{V}\) shown in Fig. 4a to the net mass change in the SMSZ. There is a good agreement between the total mass change in the SMSZ before the 2014 eruption and the mass of tephra fallout \(^{54,55}\) (Fig. 4e), consistent with a minimal effect of magma compressibility on the estimate of \(\Delta \mathrm{V}\) . While the net volume change \(\Delta \mathrm{V}\) in the SMSZ can be due to magma \((\Delta \mathrm{V}_{\mathrm{magma}})\) or/and gas \((\Delta \mathrm{V}_{\mathrm{gas}})\) , our observation shown in Fig. 4e is consistent with the scenario where \(\Delta \mathrm{V}_{\mathrm{magma}} \gg \Delta \mathrm{V}_{\mathrm{gas}}\) before the 2014 eruption. However, the net volume change in the SMSZ during/after the 2014 Strombolian eruption episode is 6- 8 times larger than that before the eruption (Fig. 4e). Such a discrepancy is consistent with \(\Delta \mathrm{V}_{\mathrm{magma}} \ll \Delta \mathrm{V}_{\mathrm{gas}}\) in the SMSZ, corroborating with a higher outgassing potential \(^{31}\) (Fig. 4b), \(\mathrm{SO}_2\) emission \(^{56}\) (Fig. 4c), and excess degassing \(^{23}\) . + +<|ref|>text<|/ref|><|det|>[[113, 463, 886, 610]]<|/det|> +On the other hand, the mass change of the two anomalously large inflation events \(\sim 2 - 3\) minutes before the 2016 phreatomagmatic explosion is also readily comparable to the mass of juvenile magma from the ash fallout and pyroclastic current deposit (PDC) \(^{34}\) (Fig. 4e). Nevertheless, if the SMSZ buffers the upward magma/gas from the magma chamber, our observation is consistent with the scenario where the cumulative mass change in the SMSZ is closely linked to the size of upcoming magmatic eruptions. + +<|ref|>text<|/ref|><|det|>[[113, 653, 884, 705]]<|/det|> +Contrasting magma ascent velocity and mass flow rate near the SMSZ prior to the 2014 Strombolian eruption and the 2016 phreatomagmatic explosion + +<|ref|>text<|/ref|><|det|>[[112, 746, 886, 896]]<|/det|> +Using a simplified 1- D conduit flow model that balances the conduit radius against the overpressure modulated by the magma flow \(^{57}\) , the single- event volume change rate allows us to infer the magma ascent velocity and decompression rate near the SMSZ before the eruptions (Material and Method). For a low- viscosity magma (i.e., \(10^3\mathrm{Pa}\cdot \mathrm{s}\) ), a density difference between magma and wall rock of \(\sim 100\mathrm{kg / m}^3\) , a lithological gradient of \(0.025\mathrm{MPa / m}\) , and a volume change rate of \(\sim 0.5 - 4.4\mathrm{m}^3 /\mathrm{s}\) , we + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 204]]<|/det|> +estimate the ascent velocity near the SMSZ of \(\sim 0.14 - 0.41 \mathrm{m / s}\) and the decompression rate of \(\sim 0.004 - 0.011 \mathrm{MPa / s}\) , which is comparable to syn-eruptive decompression rates estimated by melt embayment and water-in-olivine studies in basaltic eruptions \(^{5,6}\) and theoretical estimates for the strombolian eruption \(^{58}\) (Fig. 5). + +<|ref|>text<|/ref|><|det|>[[112, 230, 885, 508]]<|/det|> +Using a magma density of \(2500 \mathrm{kg / m^3}\) , we convert the single- event volume change rate \((\sim 0.5 - 4.4 \mathrm{m^3 / s})\) to the mass flow rate of \(\sim 1,250 - 11,000 \mathrm{kg / s}\) (Fig. 5). Since the Strombolian eruptions in Aso are intermittent, the averaged mass discharge rate estimated by tephra deposits of the 2014 Strombolian eruption (i.e., \(\sim 25 - 430 \mathrm{kg / s})^{55}\) is likely lower than the instantaneous mass discharge rate. Assuming only small percentages (i.e., \(5\%\) ) of the entire eruption duration \(^{24}\) , the magma discharge rate of the 2014 eruption is \(\sim 500 - 8,600 \mathrm{kg / s}\) , which is reasonably comparable to the estimated single- event mass flow rate in the SMSZ. While the estimated mass flow rate in Aso is lower than those estimated in basaltic eruptions by an order of magnitude (Fig. 5), such a difference is consistent with the disparity in the average volcanic output rate between basaltic eruptions and andesitic eruptions \(^{59}\) (Fig. 4d). + +<|ref|>text<|/ref|><|det|>[[112, 533, 885, 840]]<|/det|> +In contrast, the two anomalously large inflation events \(\sim 2 - 3\) minutes before the 2016 phreatomagmatic explosion corresponds to a much higher volume change rate of \(\sim 860 - 1,300 \mathrm{m^3 / s}\) over a 30 s duration. The estimated mass flow rate is \(\sim 2.2 - 3.3 \times 10^6 \mathrm{kg / s}\) (Fig. 5) and the estimated ascent velocity and decompression rate near the SMSZ are also much higher at \(\sim 7 \mathrm{m / s}\) and \(\sim 0.2 \mathrm{MPa / s}\) , respectively, indicating a much larger overpressure. These results are similar to other explosive basaltic eruptions \(^{6 - 8}\) (Fig. 5). Considering the tephra mass and the duration of the 2016 phreatomagmatic explosion of \(160 - 260 \mathrm{s}\) , the total mass discharge rate is \(2.7 - 4.1 \times 10^6 \mathrm{kg / s}^{60}\) , comparable to our inferred mass flow rate in the SMSZ. Using the empirical scaling between the mass discharge rate and the plume height \(^{61}\) , the mass flow rate in the SMSZ before the 2016 eruption projects a plume height of \(\sim 10 \mathrm{km}\) , consistent with observations \(^{56,62,63}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 150, 886, 427]]<|/det|> +In essence, we show that repetitive shallow resonances provide a pathway to unravel magma flows deep in the magma plumbing system (Fig.6). Specifically, discrete magma transport between crystal- rich mush and crystal- poor pool could be accommodated by episodic deformation event accompanied by high- angle normal faulting in a high strain- rate, low differential stress regime governed by the brittle- ductile transition rheology. The duration of each deformation event ( \(\sim 50\) s) is much longer than what is expected for crustal earthquakes of similar size and such a slow deformation, to some extent, is analogous to slow- slip events observed worldwide in the subduction zone interface or in the deep crust64,65, where the rheological condition and fluid pressure are likely similar to that near the chamber roof, or the top of the mush region.. + +<|ref|>text<|/ref|><|det|>[[112, 453, 886, 632]]<|/det|> +The contrast in the decompression rate (or the ascent velocity) and the mass flow rate inferred in this report are strongly correlated with the eruption style in Aso, collaborating with theoretical prediction66 and geological evidence67. As persistent shallow resonances in Aso have been documented for almost a century31,68,69, establishing the amplitude scaling between the synchronous sources in Aso will permit critical evaluations of pre- eruptive volume changes (or volume change rates) against the size and style of historical eruptions even when modern geodetic and broadband seismic data are not available. + +<|ref|>text<|/ref|><|det|>[[112, 659, 886, 871]]<|/det|> +Since the mass discharge rate and the magma ascent velocity or decompression rate can only be obtained retroactively after the eruptions5- 10,67,70, new seismic observations reported here underscore the possibility that, through real- time signal detection/processing and source analysis in volcanic systems where repetitive shallow seismic resonances are routinely detected18,19, the volume change and volume change rate (or mass flow rate) of episodic deformation similar to observations in Aso could facilitate in- situ characterization of magma ascent and transport before upcoming eruptions. Such efforts can be coupled with improved knowledge of external loading/unloading conditions67,71, continuous monitoring of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 140]]<|/det|> +degassing rate \(^{72}\) and the permeability/rheology of shallow conduit \(^{31,67}\) to facilitate diagnosing short- term eruption potential. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 202, 107]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[113, 135, 884, 189]]<|/det|> +1. Rubin, A. M. Propagation of magma-filled cracks. \*Annual Review of Earth and Planetary Sciences\* 23, 287-336 (1995). + +<|ref|>text<|/ref|><|det|>[[113, 214, 884, 267]]<|/det|> +2. Cashman, K. V., Sparks, R. S. J. & Blundy, J. D. Vertically extensive and unstable magmatic systems: A unified view of igneous processes. \*Science\* 355, eaag3055 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 293, 884, 346]]<|/det|> +3. Edmonds, M., Cashman, K. V., Holness, M. & Jackson, M. Architecture and dynamics of magma reservoirs. \*Phil. Trans. R. Soc. A\* 377, 20180298 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 371, 884, 425]]<|/det|> +4. Petrelli, M. & Zellmer, G. F. Rates and timescales of magma transfer, storage, emplacement, and eruption. \*Dynamic Magma Evolution\* 1-41 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 450, 884, 535]]<|/det|> +5. Barth, A. \*et al.\* Magma decompression rate correlates with explosivity at basaltic volcanoes—Constraints from water diffusion in olivine. \*Journal of Volcanology and Geothermal Research\* 387, 106664 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 561, 884, 644]]<|/det|> +6. Newcombe, M. E., Plank, T., Barth, A., Asimow, P. D. & Hauri, E. Water-in-olivine magma ascent chronometry: Every crystal is a clock. \*Journal of Volcanology and Geothermal Research\* 398, 106872 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 672, 884, 757]]<|/det|> +7. Lloyd, A. S. \*et al.\* NanoSIMS results from olivine-hosted melt embayments: Magma ascent rate during explosive basaltic eruptions. \*Journal of Volcanology and Geothermal Research\* 283, 1-18 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 784, 884, 836]]<|/det|> +8. Ferguson, D. J. \*et al.\* Magma decompression rates during explosive eruptions of Kīlauea volcano, Hawaii, recorded by melt embayments. \*Bulletin of Volcanology\* 78, 71 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 140]]<|/det|> +9. Costa, F., Shea, T. & Ubide, T. Diffusion chronometry and the timescales of magmatic processes. \*Nat Rev Earth Environ\* 1, 201–214 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 167, 881, 188]]<|/det|> +10. Rutherford, M. J. Magma ascent rates. \*Reviews in Mineralogy and Geochemistry\* 69, 241–271 (2008). + +<|ref|>text<|/ref|><|det|>[[114, 214, 883, 266]]<|/det|> +11. Rivalta, E., Taisne, B., Bunger, A. & Katz, R. A review of mechanical models of dike propagation: Schools of thought, results and future directions. \*Tectonophysics\* 638, 1–42 (2015). + +<|ref|>text<|/ref|><|det|>[[114, 293, 883, 344]]<|/det|> +12. Dzurisin, D. \*Volcano deformation: new geodetic monitoring techniques\*. (Springer Science & Business Media, 2006). + +<|ref|>text<|/ref|><|det|>[[114, 372, 752, 393]]<|/det|> +13. Segall, P. \*Earthquake and volcano deformation\*. (Princeton University Press, 2010). + +<|ref|>text<|/ref|><|det|>[[114, 419, 883, 471]]<|/det|> +14. Biggs, J. & Pritchard, M. E. Global Volcano Monitoring: What Does It Mean When Volcanoes Deform? \*ELEMENTS\* 13, 17–22 (2017). + +<|ref|>text<|/ref|><|det|>[[114, 498, 883, 550]]<|/det|> +15. Linde, A. T., Agustsson, K., Sacks, I. S. & Stefansson, R. Mechanism of the 1991 eruption of Hekla from continuous borehole strain monitoring. \*Nature\* 365, 737–740 (1993). + +<|ref|>text<|/ref|><|det|>[[114, 577, 883, 629]]<|/det|> +16. Linde, A. T. \*et al.\* Vulcanian explosion at Soufrière Hills Volcano, Montserrat on March 2004 as revealed by strain data. \*Geophysical Research Letters\* 37, (2010). + +<|ref|>text<|/ref|><|det|>[[114, 656, 883, 708]]<|/det|> +17. Ripepe, M. \*et al.\* Ground deformation reveals the scale-invariant conduit dynamics driving explosive basaltic eruptions. \*Nature Communications\* 12, 1–8 (2021). + +<|ref|>text<|/ref|><|det|>[[114, 735, 883, 786]]<|/det|> +18. Kawakatsu, H. & Yamamoto, M. Volcano Seismology. in \*Treatise on Geophysics\* vol. 4 389–420 (Elsevier, 2007). + +<|ref|>text<|/ref|><|det|>[[114, 814, 884, 896]]<|/det|> +19. Chouet, B. A. & Matoza, R. S. A multi-decadal view of seismic methods for detecting precursors of magma movement and eruption. \*Journal of Volcanology and Geothermal Research\* 252, 108–175 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +20. Oikawa, J. et al. Ground deformation associated with volcanic tremor at Izu-Oshima Volcano. Geophysical Research Letters 18, 443–446 (1991). + +<|ref|>text<|/ref|><|det|>[[113, 167, 884, 250]]<|/det|> +21. Dawson, P. & Chouet, B. Characterization of very-long-period seismicity accompanying summit activity at Kilauea Volcano, Hawai'i: 2007–2013. Journal of Volcanology and Geothermal Research 278–279, 59–85 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 278, 884, 362]]<|/det|> +22. Kobayashi, T., Ohminato, T., Ida, Y. & Fujita, E. Intermittent inflations recorded by broadband seismometers prior to caldera formation at Miyake-jima volcano in 2000. Earth and Planetary Science Letters 357, 145–151 (2012). + +<|ref|>text<|/ref|><|det|>[[113, 389, 884, 474]]<|/det|> +23. Saito, G., Ishizuka, O., Ishizuka, Y., Hoshizumi, H. & Miyagi, I. Petrological characteristics and volatile content of magma of the 1979, 1989, and 2014 eruptions of Nakadake, Aso volcano, Japan. Earth Planets Space 70, 197 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 500, 884, 552]]<|/det|> +24. Ono, K., Watanabe, K., Hoshizumi, H. & Ikebe, S. Ash eruption of the Naka-dake crater, Aso volcano, southwestern Japan. Journal of Volcanology and Geothermal Research 66, 137–148 (1995). + +<|ref|>text<|/ref|><|det|>[[113, 579, 884, 630]]<|/det|> +25. Miyabuchi, Y. A 90,000-year tephrostratigraphic framework of Aso Volcano, Japan. Sedimentary Geology 220, 169–189 (2009). + +<|ref|>text<|/ref|><|det|>[[113, 658, 884, 709]]<|/det|> +26. Sassa, K. Geophysical studies on the volcano Aso. (Part 1: Volcanic micro-tremors and eruptive-earthquakes). Mem. Coll. Sci., Kyoto Imp. Univ. Series A 255–293 (1935). + +<|ref|>text<|/ref|><|det|>[[113, 737, 884, 788]]<|/det|> +27. Kaneshima, S. et al. Mechanism of phreatic eruptions at Aso volcano inferred from near-field broadband seismic observations. Science 273, 643–645 (1996). + +<|ref|>text<|/ref|><|det|>[[113, 816, 883, 867]]<|/det|> +28. Yamamoto, M. et al. Detection of a crack-like conduit beneath the active crater at Aso volcano Japan. Geophys. Res. Lett. 26, 3677–3680 (1999). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 886, 140]]<|/det|> +29. Kawakatsu, H. et al. Aso94: Aso seismic observation with broadband instruments. Journal of Volcanology and Geothermal Research 101, 129-154 (2000). + +<|ref|>text<|/ref|><|det|>[[113, 167, 884, 250]]<|/det|> +30. Legrand, D., Kaneshima, S. & Kawakatsu, H. Moment tensor analysis of near-field broadband waveforms observed at Aso volcano, Japan. Journal of Volcanology and Geothermal Research 101, 155-169 (2000). + +<|ref|>text<|/ref|><|det|>[[113, 278, 884, 362]]<|/det|> +31. Niu, J. & Song, T.-R. A. Real-time and in-situ assessment of conduit permeability through diverse long-period tremors beneath Aso volcano, Japan. Journal of Volcanology and Geothermal Research 401, 106964 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 388, 883, 440]]<|/det|> +32. Tanada, T. et al. NIED's V-net, the fundamental volcano observation network in Japan. JDR 12, 926-931 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 468, 883, 519]]<|/det|> +33. Sato, H. et al. Development of the Crustal Tilt Observation Method Using Borehole-type Tiltmeters. Zisin1 33, 343-368 (1980). + +<|ref|>text<|/ref|><|det|>[[113, 547, 884, 630]]<|/det|> +34. Miyabuchi, Y. et al. The October 7-8, 2016 eruptions of Nakadake crater, Aso Volcano, Japan and their deposits. in Japan geoscience union-American geophysical union joint meeting. Chiba, Japan, SVC47-11 vol. 22 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 658, 819, 678]]<|/det|> +35. Turin, G. An introduction to matched filters. IEEE Trans. Inform. Theory 6, 311-329 (1960). + +<|ref|>text<|/ref|><|det|>[[113, 706, 884, 756]]<|/det|> +36. Efron, B. & Gong, G. A leisurely look at the bootstrap, the jackknife, and cross-validation. The American Statistician 37, 36-48 (1983). + +<|ref|>text<|/ref|><|det|>[[113, 784, 883, 836]]<|/det|> +37. Mori, T., Sudo, Y., Tsutsui, T. & Yoshikawa, S. Characteristics of isolated hybrid tremor (HBT) during a calm activity period at Aso Volcano. Bull Volcanol 70, 1031-1042 (2008). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 172]]<|/det|> +38. Fukuda, J. & Johnson, K. M. Mixed linear—non-linear inversion of crustal deformation data: Bayesian inference of model, weighting and regularization parameters. Geophysical Journal International 181, 1441–1458 (2010). + +<|ref|>text<|/ref|><|det|>[[113, 199, 883, 251]]<|/det|> +39. Sudo, Y. & Kong, L. Three-dimensional seismic velocity structure beneath Aso volcano, Kyushu, Japan. Bull Volcanol 63, 326–344 (2001). + +<|ref|>text<|/ref|><|det|>[[113, 278, 883, 330]]<|/det|> +40. Ohkura, S., Yoshikawa, S., Inoue, H., Utsugi, M. & Kagiyama, T. Leveling in Aso (September-October 2008): Aso research report 2009 (report in Japanese). (2009). + +<|ref|>text<|/ref|><|det|>[[113, 357, 884, 440]]<|/det|> +41. Hata, M. et al. Three-dimensional electrical resistivity distribution beneath the Beppu-Shimabara graben with a focus on Aso caldera, southwest Japan subduction zone. J. Geophys. Res. Solid Earth (2018) doi:10.1029/2018JB015506. + +<|ref|>text<|/ref|><|det|>[[113, 468, 884, 551]]<|/det|> +42. Okubo, Y. & Shibuya, A. Thermal and crustal structure of the Aso volcano and surrounding regions constrained by gravity and magnetic data, Japan. Journal of volcanology and geothermal research 55, 337–350 (1993). + +<|ref|>text<|/ref|><|det|>[[113, 579, 884, 661]]<|/det|> +43. Tanaka, Y. Eruption mechanism as inferred from geomagnetic changes with special attention to the 1989–1990 activity of Aso volcano. Journal of Volcanology and Geothermal Research 56, 319–338 (1993). + +<|ref|>text<|/ref|><|det|>[[113, 689, 884, 741]]<|/det|> +44. Fournier, R. O. Hydrothermal processes related to movement of fluid from plastic into brittle rock in the magmatic-epithermal environment. Economic Geology 94, 1193–1211 (1999). + +<|ref|>text<|/ref|><|det|>[[113, 769, 884, 820]]<|/det|> +45. Weinberg, R. F. & Regenauer-Lieb, K. Ductile fractures and magma migration from source. Geology 38, 363–366 (2010). + +<|ref|>text<|/ref|><|det|>[[113, 848, 883, 900]]<|/det|> +46. Bergantz, G., Schleicher, J. & Burgisser, A. Open-system dynamics and mixing in magma mushes. Nature Geoscience 8, 793–796 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +47. Edmonds, M. & Wallace, P. J. Volatiles and exsolved vapor in volcanic systems. Elements 13, 29–34 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 168, 884, 250]]<|/det|> +48. Kawaguchi, M. et al. Persistent gas emission originating from a deep basaltic magma reservoir of an active volcano: the case of Aso volcano, Japan. Contributions to Mineralogy and Petrology 176, 1–24 (2021). + +<|ref|>text<|/ref|><|det|>[[113, 278, 883, 362]]<|/det|> +49. Sudo, Y., Tsutsui, T. & Nakaboh, M. Ground deformation and magma reservoir at Aso volcano: Location of deflation source derived from long-term geodetic surveys (In Japanese). Bull. Volcanol. Soc. Jpn. 51, 291–309 (2006). + +<|ref|>text<|/ref|><|det|>[[113, 389, 883, 441]]<|/det|> +50. Nobile, A. et al. Steady subsidence of a repeatedly erupting caldera through InSAR observations: Aso, Japan. Bull Volcanol 79, 32 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 469, 883, 520]]<|/det|> +51. Miyoshi, M. et al. K–Ar ages determined for post-caldera volcanic products from Aso volcano, central Kyushu, Japan. Journal of Volcanology and Geothermal Research 229–230, 64–73 (2012). + +<|ref|>text<|/ref|><|det|>[[113, 547, 884, 630]]<|/det|> +52. Ohminato, T., Chouet, B. A., Dawson, P. & Kedar, S. Waveform inversion of very long period impulsive signals associated with magmatic injection beneath Kilauea volcano, Hawaii. J. Geophys. Res. 103, 23839–23862 (1998). + +<|ref|>text<|/ref|><|det|>[[113, 658, 884, 740]]<|/det|> +53. Scandone, R. & Malone, S. D. Magma supply, magma discharge and readjustment of the feeding system of Mount St. Helens during 1980. Journal of Volcanology and Geothermal Research 23, 239–262 (1985). + +<|ref|>text<|/ref|><|det|>[[113, 769, 883, 820]]<|/det|> +54. Yokoo, A. & Miyabuchi, Y. Eruption at the Nakadake 1st crater of Aso volcano started in November 2014 (In Japanese). Bull. Volcanol. Soc. Jpn. 60, 275–278 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 172]]<|/det|> +55. Miyabuchi, Y. & Hara, C. Temporal variations in discharge rate and component characteristics of tephra-fall deposits during the 2014–2015 eruption of Nakadake first crater, Aso Volcano, Japan. Earth Planets Space 71, 44 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 199, 800, 218]]<|/det|> +56. Japan Meterological Agency. Aso volcano monthly activity reports, October 2016. (2016). + +<|ref|>text<|/ref|><|det|>[[113, 245, 884, 298]]<|/det|> +57. Jaupart, C. & Tait, S. Dynamics of eruptive phenomena. Reviews in Mineralogy and Geochemistry 24, 213–238 (1990). + +<|ref|>text<|/ref|><|det|>[[113, 325, 883, 378]]<|/det|> +58. Gonnermann, H. M. & Manga, M. Dynamics of magma ascent in the. Modeling volcanic processes: The physics and mathematics of volcanism 55 (2013). + +<|ref|>text<|/ref|><|det|>[[113, 404, 883, 457]]<|/det|> +59. White, S. M., Crisp, J. A. & Spera, F. J. Long-term volumetric eruption rates and magma budgets. Geochemistry, Geophysics, Geosystems 7, (2006). + +<|ref|>text<|/ref|><|det|>[[113, 483, 884, 567]]<|/det|> +60. Shimbori, T. Current Status and Perspectives on Tephra Transport Models: Volcanic Ash Fall Forecasts of Aso Volcano on 8 October 2016 as an Example. PROGRAMME AND ABSTRACTS THE VOLCANOLOGICAL SOCIETY OF JAPAN 2017, 5–5 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 594, 884, 678]]<|/det|> +61. Mastin, L. G. et al. A multidisciplinary effort to assign realistic source parameters to models of volcanic ash-cloud transport and dispersion during eruptions. Journal of Volcanology and Geothermal Research 186, 10–21 (2009). + +<|ref|>text<|/ref|><|det|>[[113, 705, 884, 757]]<|/det|> +62. Ishii, K., Hayashi, Y. & Shimbori, T. Using Himawari-8, estimation of SO2 cloud altitude at Aso volcano eruption, on October 8, 2016. Earth Planets Space 70, 19 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 784, 884, 836]]<|/det|> +63. Sato, E., Fukui, K. & Shimbori, T. Aso volcano eruption on October 8, 2016, observed by weather radars. Earth Planets Space 70, 105 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +64. Peng, Z. & Gomberg, J. An integrated perspective of the continuum between earthquakes and slow-slip phenomena. Nature geoscience 3, 599-607 (2010). + +<|ref|>text<|/ref|><|det|>[[113, 167, 884, 219]]<|/det|> +65. Beroza, G. C. & Ide, S. Slow earthquakes and nonvolcanic tremor. Annual review of Earth and planetary sciences 39, 271-296 (2011). + +<|ref|>text<|/ref|><|det|>[[113, 246, 883, 298]]<|/det|> +66. Woods, A. W. & Koyaguchi, T. Transitions between explosive and effusive eruptions of silicic magmas. Nature 370, 641-644 (1994). + +<|ref|>text<|/ref|><|det|>[[113, 325, 883, 377]]<|/det|> +67. Cassidy, M., Manga, M., Cashman, K. & Bachmann, O. Controls on explosive-effusive volcanic eruption styles. Nature Communications 9, 1-16 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 404, 883, 455]]<|/det|> +68. Sassa, K. Micro-Seismometrical Study on Eruptions of the Volcano Aso (Part II, of the Geophysical Studies on the Volcano Aso). Mem. Coll. Sci., Kyoto Imp. Univ. Series A (1936). + +<|ref|>text<|/ref|><|det|>[[113, 483, 884, 599]]<|/det|> +69. Sakaguchi, H., Sudo, Y., Sawada, Y. & Yoshikawa, S. Relationship between the explosive activities and the associated volcanic tremors observed at Nakadake summit of Aso volcano -- The temporal variation in amplitude of tremors recorded by Wiechert seismograph -- (In Japanese). Bull. Volcanol. Soc. Jpn. 53, 143-149 (2008). + +<|ref|>text<|/ref|><|det|>[[113, 627, 883, 679]]<|/det|> +70. Toramaru, A. BND (bubble number density) decompression rate meter for explosive volcanic eruptions. Journal of Volcanology and Geothermal Research 154, 303-316 (2006). + +<|ref|>text<|/ref|><|det|>[[113, 706, 883, 757]]<|/det|> +71. Manga, M. & Brodsky, E. Seismic triggering of eruptions in the far field: volcanoes and geysers. Annu. Rev. Earth Planet. Sci. 34, 263-291 (2006). + +<|ref|>text<|/ref|><|det|>[[113, 784, 883, 836]]<|/det|> +72. Oppenheimer, C., Fischer, T. & Scaillet, B. Volcanic degassing: Process and impact. Treatise on Geochemistry (Second Edition) 4, 111-179 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 140]]<|/det|> +73. Wielandt, E. & Forbriger, T. Near-field seismic displacement and tilt associated with the explosive activity of Stromboli. Annals of Geophysics (1999) doi:10.4401/ag-3723. + +<|ref|>text<|/ref|><|det|>[[113, 167, 883, 218]]<|/det|> +74. Graizer, V. Tilts in strong ground motion. Bulletin of the Seismological Society of America 96, 2090–2102 (2006). + +<|ref|>text<|/ref|><|det|>[[113, 246, 883, 298]]<|/det|> +75. Graizer, V. Determination of the true ground displacement by using strong motion records. Izvestiya Phys. Solid Earth 15, 875–885 (1979). + +<|ref|>text<|/ref|><|det|>[[113, 325, 883, 377]]<|/det|> +76. Zhu, L. Recovering permanent displacements from seismic records of the June 9, 1994 Bolivia deep earthquake. Geophysical research letters 30, (2003). + +<|ref|>text<|/ref|><|det|>[[113, 404, 883, 455]]<|/det|> +77. Aoyama, H. & Oshima, H. Tilt change recorded by broadband seismometer prior to small phreatic explosion of Meakan-dake volcano, Hokkaido, Japan. Geophysical research letters 35, (2008). + +<|ref|>text<|/ref|><|det|>[[113, 483, 884, 565]]<|/det|> +78. Wang, R., Schurr, B., Milkereit, C., Shao, Z. & Jin, M. An improved automatic scheme for empirical baseline correction of digital strong-motion records. Bulletin of the Seismological Society of America 101, 2029–2044 (2011). + +<|ref|>text<|/ref|><|det|>[[113, 593, 884, 676]]<|/det|> +79. Thun, J., Lokmer, I. & Bean, C. J. New observations of displacement steps associated with volcano seismic long-period events, constrained by step table experiments. Geophysical Research Letters 42, 3855–3862 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 705, 884, 756]]<|/det|> +80. Okada, Y. Internal deformation due to shear and tensile faults in a half-space. Bulletin of the seismological society of America 82, 1018–1040 (1992). + +<|ref|>text<|/ref|><|det|>[[113, 784, 884, 866]]<|/det|> +81. Williams, C. A. & Wadge, G. The effects of topography on magma chamber deformation models: Application to Mt. Etna and radar interferometry. Geophysical Research Letters 25, 1549–1552 (1998). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 140]]<|/det|> +82. Storn, R. & Price, K. Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces. Journal of global optimization 11, 341–359 (1997). + +<|ref|>text<|/ref|><|det|>[[112, 167, 884, 219]]<|/det|> +83. Goodman, J. & Weare, J. Ensemble samplers with affine invariance. Communications in applied mathematics and computational science 5, 65–80 (2010). + +<|ref|>text<|/ref|><|det|>[[112, 245, 883, 298]]<|/det|> +84. Foreman-Mackey, D., Hogg, D. W., Lang, D. & Goodman, J. emcee: the MCMC hammer. Publications of the Astronomical Society of the Pacific 125, 306 (2013). + +<|ref|>text<|/ref|><|det|>[[112, 324, 575, 344]]<|/det|> +85. Aki, K. & Richards, P. G. Quantitative seismology. (2002). + +<|ref|>text<|/ref|><|det|>[[112, 372, 884, 424]]<|/det|> +86. Tape, W. & Tape, C. A geometric setting for moment tensors. Geophysical Journal International 190, 476–498 (2012). + +<|ref|>text<|/ref|><|det|>[[111, 490, 885, 897]]<|/det|> +Acknowledgments: We gratefully thank NIED and JMA for providing us the high-quality waveform data of V-net. We are also grateful to Prof. Mare Yamamoto and Prof. Takuhiro Ohkura for hosting J. Niu in Aso Volcanic Observatory and Tohoku University during the early stage of this work, which helps improve our understanding of the LPT and Aso volcanic system. The uses of the UCL Legion High-Performance Computing Facility (Legion@UCL), the UCL Grace High-Performance Computing Facility (Grace@UCL), the UCL Emerald High-Performance Computing Facility (Emerald@UCL), and associated support services are acknowledged. Data processing and production of figures are implemented in Python with relevant modules such as ObsPy and PyCUDA. Funding: J. Niu and T.- R. A. Song are supported by the Natural Environment Research Council, UK (NE/ P001378/1 & NE/T001372/1). Author contributions: J. Niu: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing - original draft, Visualization. T.- R. A. Song: Conceptualization, Methodology, Validation, Formal analysis, Writing - review & editing, Supervision, Project administration, Funding acquisition, Conceptualization.; Competing interests: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 884, 171]]<|/det|> +Authors declare no competing interests. Data and materials availability: The waveform data can be downloaded from https://www.hinet.bosai.go.jp. The LPT catalog published by Niu and Song (2020) and all the codes used in this study are available upon request. + +<|ref|>sub_title<|/ref|><|det|>[[115, 201, 319, 218]]<|/det|> +## Supplementary Materials: + +<|ref|>text<|/ref|><|det|>[[114, 241, 180, 256]]<|/det|> +Methods + +<|ref|>text<|/ref|><|det|>[[114, 280, 230, 297]]<|/det|> +Figures S1- S14 + +<|ref|>text<|/ref|><|det|>[[114, 320, 214, 336]]<|/det|> +Tables S1- S2 + +<|ref|>text<|/ref|><|det|>[[114, 360, 256, 376]]<|/det|> +References (###- ###) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 88, 860, 409]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 415, 884, 750]]<|/det|> +
Figure 1. Aso volcano system and observations of tilt offset synchronous with LPT. (a) The topographic relief near the Aso caldera is overlaid with collocated seismometer and tiltmeter (open triangles), LPT source in the shallow crack-like conduit (thick bar), the geodetically inferred magma chamber (open circle), and the low-velocity zone (dashed circle). The star marks the Naka-dake first crater. The upper inset displays Aso volcano located in Kyushu island, southwest Japan. The source of the tilt offset is marked as a red cross and it is referred to as a shallow magma storage zone (SMSZ). The beach ball displays the deviatoric component of the source mechanism in SMSZ. (b) Tilt waveforms at four V-net stations near the October 8, 2016 phreatomagmatic eruption. The waveforms are low-pass filtered at 0.05 Hz with a \(4^{\text{th}}\) -order casual Butterworth filter. The linear trend determined in the first 10 minutes is used to remove the background trend. “Event 1” and “Event 2” denote the tilt offset that occurred concurrently with two LPT events. Redline marks the timing of the 2016 eruption.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 88, 480, 320]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[170, 87, 456, 102]]<|/det|> +
Global inflation stack [Nov 2011-Aug 2014]
+ +<|ref|>image<|/ref|><|det|>[[123, 321, 480, 675]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[496, 183, 884, 585]]<|/det|> +
Figure 2. Observations of inflation (a) and deflation (b) waveform stacks against LPT and SPT. Broadband tilt, displacement, and velocity of global waveform stacks at station N.ASHV are presented against LPT velocity waveforms (10-30 s) and SPT envelopes (0.5-1 s). The bootstrap errors are shown in yellow strips. These waveforms and envelopes are aligned against the onset of SPT. To preserve the phase of the long-period waveform (> 50 s), a \(2^{\text{nd}}\) -order acausal Butterworth filter is implemented. We apply a \(2^{\text{nd}}\) -order causal Butterworth filter to preserve the timing of the LPT waveform (10-30 s) and SPT envelope (0.5-1 s).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 85, 861, 519]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 521, 884, 860]]<|/det|> +
Figure 3. Source inversion result and a new conceptual framework of magma/gas transport in the plumbing system beneath Aso volcano. (a), (b) and (c) compares data (black arrows) and synthetics (red arrows) from the inverted source location and mechanism. Red cross marks the inverted source location with a confidence level of \(99\%\) . Black, red and blue circles indicate the new Mogi magma chamber40, BYA and BCU magma chambers49, respectively. Dashed circle marks the low-velocity zone39. (d) An east-west cross section showing a conceptual diagram of magma/gas transport system beneath Aso volcano. The inverted source, or the shallow magma storage zone (SMSZ), connects the LPT source in the shallow conduit from above and the magma chamber from below. SPT source is shown directly above the shallow conduit. The inset displays the volumetric and deviatoric components of the focal mechanism, respectively, looking from the south. Blue dashed line indicates the brittle-ductile transition. (e), (f) and (g) display the simplified flow conditions where the SMSZ suffers inflation, null volume change, or
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 883, 138]]<|/det|> +deflation, respectively. Waveform stacks representative of these three conditions are shown directly below. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 85, 861, 528]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 531, 884, 580]]<|/det|> +
Figure 4. Characterization of the volume/mass change inside the SMSZ during the 2011-2016 eruption cycle (a) The accumulative net volume changes in the SMSZ. Dark inverse triangles mark the onset of the 2014 Strombolian eruption, the 2015 and 2016 phreatomagmatic eruptions, whereas hatched areas mark their eruption episodes. Grey inverse triangles indicate minor eruptions. \(\oplus\) , \(\ominus\) and \(\odot\) mark the episodes associated with observations of inflation event, deflation event and event of negligible offset, respectively (see waveform stacks in the insets). \(①\) and \(②\) mark the time windows where the mass changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively (see also Fig. 4e). (b) Outgassing potential of the shallow conduit, or equivalently the moment ratio between pressurization and depressurization LPTs \(^{31}\) . A lower (higher) outgassing potential corresponds to a higher proportion of pressurization (depressurization) LPT events. (c) The campaign SO \(_2\) emission from JMA (https://www.data.jma.go.jp/svd/vois/data/fukuoka/rovdm/Asosan_rovdm/gas/gas.html). (d) The volume change rates estimated by the inflation event (black crosses), the geodetic data (green cross), and
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 240]]<|/det|> +geological analysis (blue cross). Average volcanic output rates from White et al. \(^{59}\) are shown in the pink shaded regions. (e) The observed volcanic output of the 2014 Strombolian and 2016 phreatomagmatic eruptions against the estimated mass change inside the SMSZ before the eruptions. \(①\) and \(②\) mark the time windows in Fig. 4a where the mass changes in SMSZ before the 2014 and 2016 eruptions are calculated, respectively. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 90, 867, 536]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 564, 884, 871]]<|/det|> +
Figure 5. Comparisons of the mass flow rate and decompression rate from petrology and seismology. Open circles and triangles mark the volume estimates from waveform stacks of inflation and deflation events before the 2014 and 2016 eruptions (see details in the main text). Target crosses mark the median volume estimates from the five subsets (Fig. S13a-e). Solid circles mark previous compilations in basaltic volcanoes \(^5\) , where the decompression rates are estimated from water diffusion in olivine or embayment studies and the mass discharge rates are estimated from volcanic outputs or plume heights, respectively. Contour lines display a theoretical estimate of decompression rate and mass flow rate against fixed conduit radius (see equation 4 in Barth et al. \(^5\) ). Hatched areas show the mass discharge rate of the 2014 and 2016 eruptions (see details in the main text). Dashed line indicates the boundary of magmastatic-dominated and friction-dominated decompression after Barth et al. \(^5\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 88, 860, 458]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 486, 886, 666]]<|/det|> +
Figure 6. A conceptual diagram of episodic deformation in the chamber accompanying repetitive shallow volcanic resonances, providing the glue to long-term volcanic output and short-term magma ascent/discharge rates. Green solid bars indicate the global observations of the average volcanic output rates over eruption cycles in geological time scale ( \(\sim 1\) yr–1 Myr). Gray dashed bars indicate the single-event magma ascent/discharge rate over a seismic time scale ( \(\sim 100\) s). Composition, viscosity, rheology and tectonic settings govern the recurrences of episodic deformation (red lines).
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 184, 107]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[115, 152, 675, 170]]<|/det|> +1. Systematic detection using the matched filter (the inflation/deflation event) + +<|ref|>text<|/ref|><|det|>[[113, 198, 884, 345]]<|/det|> +A robust observation of the tilt or displacement offset shown in Fig. 1b and Fig. S1 requires a high signal- to- noise ratio that is not routinely achievable. Instead, we measure the energy in the ultra- long- period band and use it as a proxy for the static offset30. Here we use the matched filtered technique31,35 to systematically examine the presence and the stability of such synchronous signals against the LPT event catalog during the 2011- 2016 Aso eruption cycle31. + +<|ref|>text<|/ref|><|det|>[[112, 372, 885, 680]]<|/det|> +The first of the two LPT events before the 2016 eruption (Event 1 in Fig.1) is selected as the reference template event. The waveforms at two long- running stations, N.ASHV and N.ASIV, are cut in a 10- minute window centered at the LPT arrival time and filtered in the period band of 50- 250 s using a 3rd- order causal Butterworth filter to obtain the template waveforms. The waveforms of each LPT event are processed in the same way as the reference event before the matched filter is applied to cross- correlate waveform data against the template waveforms. To avoid the coupling between tilt and translation in the horizontal components73, we only concern vertical- component seismic waveform data and borehole tilt waveform data in the signal detection. Depending on the cross- correlation coefficient (CC) and signal- to- noise ratio (SNR) between waveform data and the template waveforms (Table S1), the detection against the entire LPT catalog of more than 200,000 events is divided into five subsets (I- V) (Table S2). + +<|ref|>sub_title<|/ref|><|det|>[[115, 708, 277, 725]]<|/det|> +## 2. Waveform stacking + +<|ref|>text<|/ref|><|det|>[[113, 754, 884, 901]]<|/det|> +After removing the mean and trend of raw seismic waveforms, we align and stack the waveforms against the LPT event time to produce the waveform stacks, which greatly enhances the signal and eliminates incoherent noise. To minimize the bias introduced by arbitrary spikes as well as the energy from large earthquakes, we normalize raw velocity waveforms of the single event against their vertical peak- to- peak velocity in the period band of 10- 30 s at station N.ASHV before stacking. After stacking the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 268]]<|/det|> +normalized waveforms, the waveform stacks are multiplied by the median vertical peak- to- peak velocity in the period band of 10- 30 s to recover the amplitude of the waveform stack in a given month. The bootstrap technique is used to quantify the uncertainties of the waveform stacks36. After deconvolving the instrument response up to 1000 seconds, we remove the background mean in the first 150 sec of the velocity waveform stacks, perform a causal low- pass filter, and integrate the velocity waveform stacks to obtain displacement waveform stacks. + +<|ref|>text<|/ref|><|det|>[[113, 295, 884, 410]]<|/det|> +Regarding tilt waveform stacks, we first differentiate the raw tilt waveform to obtain the tilt- rate waveform and apply the same data processing procedure as done against velocity waveforms. This stacking procedure not only reduces noise but also helps remove unwanted bias associated with tides, pressure, temperature, or rainfall, which are nonetheless trivial in the data window of 10 minutes. + +<|ref|>text<|/ref|><|det|>[[113, 438, 633, 457]]<|/det|> +3. Estimate the static displacement offset of the broadband seismometer + +<|ref|>text<|/ref|><|det|>[[113, 484, 884, 696]]<|/det|> +As the seismic recording is also sensitive to sensor tilt, it is often difficult to recover true translational signal74. This tilt- related signal is often exhibited as a non- linear drift in the original displacement seismogram74. Due to the projection of gravitational force into the horizontal axis, this phenomenon is particularly severe in the horizontal components. Many approaches have been proposed to separate translational and tilt signals in seismic recordings73,75- 79. Here we estimate and remove the nonlinear drift in the displacement seismogram by least- square fitting the portion of seismogram without major signals with a polynomial function75,76. + +<|ref|>text<|/ref|><|det|>[[113, 723, 884, 876]]<|/det|> +The pre- event window is set to 20 s and the total duration of seismograms is 120 s. A \(4^{\text{th}}\) - order polynomial function is used as suggested by Zhu76. After removing the non- linear drift (Fig. S11), the displacement offset in the individual component is calculated by differing the average displacements in the post- event (mean: \(d^{1}\) , standard deviation: \(\sigma^{1}\) ) and pre- event (mean: \(d^{0}\) , standard deviation: \(\sigma^{0}\) ) windows as \(d = d^{1} - d^{0}\) , with uncertainty, \(\sigma = \sqrt{(\sigma^{0})^{2} + (\sigma^{1})^{2}}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 236]]<|/det|> +As indicated in the previous study \(^{78}\) , the static displacement from this approach may be affected by the selection of polynomial order and time window \(^{78}\) . In our case, the determination of the time window of the major signal of the reference event is relatively straightforward. As presented in Tab. S2, the selection of polynomial order does not significantly change the estimations of the point- source location and source geometry. + +<|ref|>sub_title<|/ref|><|det|>[[114, 264, 490, 282]]<|/det|> +## 4. Source properties of the inflation/deflation event + +<|ref|>text<|/ref|><|det|>[[112, 310, 886, 488]]<|/det|> +We followed the mixed linear- nonlinear source inversion scheme in a Bayesian framework proposed by Fukuda and Johnson \(^{38}\) to obtain the location and mechanism of the reference event (Event 1 in Fig. 1b). This mixed inversion method efficiently handles non- linear inversion problems when some of the parameters are linearly related to the observations. The offsets of the vertical displacement, horizontal displacement and horizontal tilt are modeled with Okada's analytic deformation model for a point source \(^{80}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 518, 296, 535]]<|/det|> +### 4.1 Problem formulation + +<|ref|>text<|/ref|><|det|>[[112, 562, 886, 846]]<|/det|> +Considering a mixed linear- non- linear inversion problem \(^{38}\) associated with a point dislocation/force source proposed by Okada \(^{80}\) : \(\mathbf{d} = \mathbf{G}(\mathbf{m})\mathbf{s} + \epsilon\) where \(\mathbf{G}(\mathbf{m})\) is the kernel matrix relating non- linear model parameters, \(\mathbf{m} = [x_{s}, y_{s}, z_{s}, \theta , \phi ]^{T}\) (3- D Cartesian location of a point source with respect to the Naka- dake first crater, strike and dip angles of the fault plane and crack plane), and linear model parameters, \(\mathbf{s}_{1} = [m_{ss}, m_{ds}, m_{tc}, m_{ex}]^{T}\) (the scalar moments of strike- slip, dip- slip, tensile crack and explosion) or \(\mathbf{s}_{2} = [m_{ss}, m_{ds}, m_{tc}, m_{ex}, f_{E}, f_{N}, f_{U}]^{T}\) (the scalar moments of strike- slip, dip- slip, tensile crack and explosion, the magnitudes of forces towards east, north, and up directions), to the observation vector, \(\mathbf{d} = [\mathbf{d}_{1}, \mathbf{d}_{2}, \mathbf{d}_{3}]^{T}\) (vertical displacement, horizontal displacement, and horizontal tilt), with normally distributed error, \(\boldsymbol {\epsilon} = [\boldsymbol {\epsilon}_{1}, \boldsymbol {\epsilon}_{2}, \boldsymbol {\epsilon}_{3}]^{T}\) where \(\boldsymbol{\epsilon}_{i} \sim \mathcal{N}(\mathbf{0}, \mathbf{D}_{i})\) . \(J^{th}\) minimum norm constraints (or + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 148]]<|/det|> +Tikhonov regularization) are applied on the linear parameters, \(\left\| R_{j}\mathbf{s}\right\|^{2}\) , where \(J = 2\) if forces are considered. Otherwise, \(J = 1\) . + +<|ref|>text<|/ref|><|det|>[[112, 175, 884, 430]]<|/det|> +An optimal solution can be found by minimizing an objective function \(\Phi (\pmb {m},\pmb {s}) = \sum_{i = 1}^{I}\frac{1}{\sigma_{i}^{2}} [\pmb {d}_{i} - \pmb {G}_{i}(\pmb {m})\pmb {s}]^{\top}\pmb {D}_{i}^{- 1}[\pmb {d}_{i} - \pmb {G}_{i}(\pmb {m})\pmb {s}] + \sum_{j = 1}^{J}\frac{1}{\beta_{j}^{2}}\left\| \pmb {R}_{j}\pmb {s}\right\|^{2}\) where \(\sigma_{i}^{2}\) adjust relative weights of the multiple data sets in fitting the data; \(\beta_{j}^{2}\) adjust relative weights of regularization prior on \(\pmb{s}\) . If \(J = 1\) , \(\pmb{R}_{1} = \pmb{I}\) . If \(J = 2\) , \(\pmb{R}_{1} = \left( \begin{array}{cc}I_{4\times 4} & 0\\ 0 & 0 \end{array} \right)\) and \(\pmb{R}_{2} = \left( \begin{array}{cc}0 & 0\\ 0 & I_{3\times 3} \end{array} \right)\) . Following Fukuda and Johnson \(^{38}\) , if assuming that the prior probability density functions of the linear, non- linear and hyperparameters are uniform, the posterior probability density function of the parameters given the data, \(p(\pmb {m},\pmb {s},\pmb {\sigma},\pmb {\beta}|\pmb {d})\) , can be given according to Bayes' theorem as below: + +<|ref|>equation<|/ref|><|det|>[[264, 456, 732, 616]]<|/det|> +\[-\ln p(\pmb {m},\pmb {s},\pmb {\sigma},\pmb {\beta}|\pmb {d})\] \[\qquad \propto \sum_{i = 1}^{I}2N_{i}\ln \sigma_{i} + \sum_{j = 1}^{J}2M_{j}\ln \beta_{j} + \Phi (\pmb {m},\pmb {s}^{*})\] \[\qquad +\ln \left|\sum_{i = 1}^{I}\frac{1}{\sigma_{i}^{2}} G_{i}^{T}(\pmb {m})\pmb{D}_{i}^{-1}\pmb{G}_{i}(\pmb {m}) + \sum_{j = 1}^{J}\frac{1}{\beta_{j}^{2}} R_{j}^{T}\pmb {R}_{j}\right|\] + +<|ref|>text<|/ref|><|det|>[[112, 656, 885, 746]]<|/det|> +where \(\pmb{s}^{*}\) is the weighted damped least- square solution of the linear parameter given the nonlinear and hyperparameters, \(\pmb {m}\) , \(\pmb{\sigma}\) and \(\beta\) ; \(N_{i}\) is the number of samples in \(i^{t h}\) data set; \(M_{j}\) is the number of parameters efficiently regulated by \(j^{t h}\) prior constraint. If \(J = 1\) , \(M_{1} = 4\) . If \(J = 2\) , \(M_{1} = 4\) and \(M_{2} = 3\) . + +<|ref|>text<|/ref|><|det|>[[112, 772, 884, 890]]<|/det|> +To tackle the weights between displacement and tilt, we first normalize the kernel matrix \(\pmb {G}(\pmb {m})\) , the observation vector \(\pmb{d}\) and the covariance matrix \(\pmb{D}\) by the absolute of observation vector as \(\pmb {K} = \left( \begin{array}{ccc}1 / |d_{1}| & \dots & 0 \\ \vdots & \ddots & \vdots \\ 0 & \dots & 1 / |d_{N}| \end{array} \right)\) , \(\pmb {G}^{\prime}(\pmb {m}) = \pmb {K}\pmb {G}(\pmb {m})\) , \(\pmb{d}^{\prime} = \pmb {K}\pmb {d}\) , and \(\pmb{D}^{\prime} = \pmb {K}\pmb {D}\pmb{K}^{T}\) . Then, by utilizing Cholesky + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 270]]<|/det|> +factorization, we equalize the weight difference among the observations as \(D^{\prime} = L^{T}L\) , \(d^{\prime \prime} = L d^{\prime}\) , \(G^{\prime \prime}(m) = L G^{\prime}(m)\) . Then, if let \(\sigma_{i}^{2} = 1\) , we can rewrite the objective function as \(\Phi (m, s) = (y - A s)^{T}(y - A s)\) where \(y = [d^{\prime \prime}, 0]^{T}\) and \(A = \left[G^{\prime \prime}(m), \frac{R_{0}}{\beta_{0}}, \dots , \frac{R_{J}}{\beta_{J}}\right]^{T}\) . In this case, the posterior probability density function of the parameters can be re- written as \(- \ln p(m, s, \beta | d) \propto \sum_{j = 1}^{J} 2 M_{j} \ln \beta_{j} + \Phi (m, s^{*}) + \ln |A^{T} A|\) where \(s^{*} = (A^{T} A)^{- 1} A^{T} y\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 298, 464, 316]]<|/det|> +### 4.2 Synthetic static displacement/tilt calculation + +<|ref|>text<|/ref|><|det|>[[112, 344, 886, 589]]<|/det|> +Synthetic displacements and tilts from a point dislocation source in a homogeneous half space are calculated following Okada80. Following Legrand et al.30, we set the P velocity of 1500 m/s, S- velocity of 800 m/s and the density of 1700 kg/m3, respectively. To evaluate the significance of the single force component, we also calculate synthetic displacement and tilt from a point force source following equations 1- 2 in Okada80. To take into account the station elevation (i.e., typically 600 m above sea level), we set the station elevation as the free surface and the depth of the source is adjusted accordingly in the calculation of synthetics81. Since the borehole tiltmeter is typically about 100 meters below the surface, we calculate synthetic tilts at the borehole depth. + +<|ref|>sub_title<|/ref|><|det|>[[115, 616, 435, 633]]<|/det|> +### 4.3 Bayesian inference of model parameters + +<|ref|>text<|/ref|><|det|>[[112, 661, 886, 909]]<|/det|> +We first use differential evolution method82 to find the global minimum of \(- \ln p(m, s, \beta | d)\) . This step is repeated for hundreds of times with the randomly assigned initial values. A consistently converged minimum is regarded as the global minimum. Subsequently, we use an ensemble Monte Carlo Markov Chain method83,84 to sample the posterior probability density function. The starting values of ensemble Markov chains of \(k^{th}\) parameter are randomly assigned according to the global minimum by a normal distribution, \(N(x_{k}, |x_{k}|)\) , where \(x_{k} \in \{m, s, \beta \}\) . This step is also repeated for multiple times to evaluate the convergence of independent Markov chains. Finally, the posterior probability density functions of the parameters are combined from multiple independent chains (Fig.S12). The moment- tensor representation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 139]]<|/det|> +of the source can be reconstructed from the moments of strike-slip, dip-slip, tensile crack and explosion by following equation 3.21 in Aki and Richard85. + +<|ref|>sub_title<|/ref|><|det|>[[115, 169, 328, 186]]<|/det|> +## 5. Estimating volume change + +<|ref|>sub_title<|/ref|><|det|>[[115, 215, 416, 232]]<|/det|> +### 5.1 Volume change of the reference event + +<|ref|>text<|/ref|><|det|>[[111, 260, 886, 604]]<|/det|> +As the strike- slip and dip- slip components do not experience any volume changes, the volume change associated with the reference event is estimated by only considering the tensile crack and explosion components. Following Okada80, the eigenvalue of tensile- crack moment tensor is \(\Lambda_{\mathrm{tc}} = m_{\mathrm{tc}}[1,1,1 + 2\mu /\lambda ]^{T}\) and the eigenvalue of explosion moment tensor is \(\Lambda_{\mathrm{ex}} = m_{\mathrm{ex}}[1,1,1]^{T}\) . Following equation 16 in Kawakastu and Yamamoto18, the volume change associated with the tensile crack component is \(\Delta V_{m}^{tc} = m_{tc} / \lambda\) where \(\lambda\) is Lame constant. Following equations 10- 11 in Kawakastu and Yamamoto18, the volume change associated with the explosion component in the presence of the confining pressure of the surrounding elastic medium is \(\Delta V_{m}^{ex} = m_{ex} / (\lambda + 2\mu)\) where \(\mu\) is shear modulus. Hence, the 'Mogi volume' change associated with the reference event is \(\Delta V_{m} = \Delta V_{m}^{ex} + \Delta V_{m}^{tc}\) . If assuming \(\lambda = 2.737 GPa\) and \(\mu = 1.088 GPa\) , \(\Delta V_{m} = \sim 25840 m^{3}\) (Tab.S2, a \(4^{\mathrm{th}}\) - order polynomial function). + +<|ref|>sub_title<|/ref|><|det|>[[115, 625, 570, 643]]<|/det|> +### 5.2 Estimate volume change from monthly tilt waveform stacks + +<|ref|>text<|/ref|><|det|>[[111, 671, 886, 883]]<|/det|> +Following the note in section 5.1, the Mogi volume change corresponding to a 1- nrad tilt at N.ASHV.LE is \(\sim 440 \mathrm{m}^{3}\) . However, unlike the reference event, the static tilt offset in the monthly stacks can be difficult to measure directly in the time domain. As the duration of the tilt offset is typically \(\sim 100 \mathrm{s}\) or less, the amplitude of the tilt- rate spectra plateaus in the near- field at the very long- period (i.e., \(>1000 \mathrm{s}\) ), corresponding to the amplitude of the tilt (Fig. S13). We zero- pad the stacked waveform 10,000s before the start of the tilt- rate time series and compute the tilt- rate amplitude spectra. The amplitude at the period of 10,000 s is used as a proxy for the static offset, which is scaled against the tilt of the reference + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 140]]<|/det|> +event to obtain the volume change per event associated with each monthly stack. The monthly volume change can be obtained by multiplying the event number against the volume change per event (Fig.S14). + +<|ref|>sub_title<|/ref|><|det|>[[113, 168, 543, 186]]<|/det|> +## 6. Estimate magma ascent velocity and decompression rate + +<|ref|>text<|/ref|><|det|>[[112, 213, 885, 450]]<|/det|> +Following the 1- D conduit flow model by Jaupart & Tait \(^{57}\) , the conduit radius \(R\) is balanced by the dynamic pressure of the ascended magma and lithospheric gradient \(R^{4} = G \frac{8\eta}{\pi \rho (\rho_{r} - \rho)g}\) , where \(G\) is the mass flow rate, \(\eta\) is magma viscosity, \(\rho_{r}\) is the density of wall rock, \(\rho\) is the density of magma, and \(g\) is the gravitational acceleration. To estimate the conduit radius, the mass flow rate is estimated by the volume change rate of the inflation/deflation event and magma density \(\rho\) . The magma ascent velocity \(V_{a}\) is estimated as \(V_{a} = \frac{G}{\pi \rho R^{2}}\) . Assuming a lithospheric gradient, the decompression rate \(dP / dt\) is estimated as \(dP / dt = \rho_{r}gV_{a}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[81, 130, 930, 656]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[121, 665, 892, 740]]<|/det|> +
Fig. S1. Examples of single-event tilt and displacement waveforms at N.ASHV. (a) Low-pass filtered displacement and tilt waveforms with the upward and east-down offsets, respectively. (b) Low-pass filtered displacement and tilt waveforms with the downward and west-down offsets, respectively. The starting time of each waveform is indicated above each trace.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[235, 130, 777, 787]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 795, 893, 977]]<|/det|> +
Fig. S2. Global and monthly waveform stacks against highest quality data (subset I). (a) the global waveform stacks of inflation events during the unrest (November 2011 to August 2014). (b) the global waveform stacks of deflation events (October 2014 to April 2015) (c) the monthly waveform stack of inflation events in July 2014. (d) the monthly waveform stack of deflation events in November 2014. LPT signal is present in the middle of the time window at \(\sim 220 - 270\) s. Waveform stacks from stations N.ASHV and N.ASIV are shown in black lines and green lines, respectively. The orange shaded region marks the \(99.7\%\) confidence interval estimated by the bootstrapped resampling (ref.36). Displacement and tilt waveform stacks are low-pass filtered at 10 s and 30 s with a 4th-order casual Butterworth filter, respectively. The number of events used in each waveform stack is shown in the upper right.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[80, 131, 931, 621]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 631, 892, 757]]<|/det|> +
Fig. S3. 1-hour long global and monthly waveform stacks against highest quality data (subset I). Same as Fig. S2, expect only waveform data associated with isolated LPT events are included in the stack. The number in the left panel indicates the number of the well-isolated LPT used in the stacks. (a)-(d) the global and monthly tilt-rate waveform stacks. (e)-(h) global and monthly tilt stacks. The trend defined in [-10,-2] minutes of the tilt waveform stacks is used to remove the background trend. The gray shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[220, 131, 787, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[122, 574, 892, 684]]<|/det|> +
Fig. S4. Synthetic amplitude-distance decay against static and filtered waveforms. (a) and (b) display normalized displacement amplitude-distance decay trend with a source depth \(0.5 \mathrm{km}\) and \(5.0 \mathrm{km}\) , respectively. (c) and (d) display normalized tilt amplitude-distance decay trend with a source depth \(0.5 \mathrm{km}\) and \(5.0 \mathrm{km}\) , respectively. Static, long-period (10-30 s), and ultra-long-period (100-200 s) amplitude measurements are shown in green lines, blue lines and red lines, respectively.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[78, 170, 933, 667]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 675, 893, 916]]<|/det|> +
Fig. S5. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. (a)-(e) the amplitude ratio in the LPT frequency band (10-30 s) against five subsets, respectively (the subset I→the highest quality and the subset V→the lowest quality). The amplitude ratio measured in each monthly stack is shown in shaded square. The color of the square denotes the number of stacked inflation events in a linear scale (dark→higher number and light→lower number). The error bar marks the uncertainty with a 99.7% confidence level estimated from the bootstrap resampling (ref.36). The red line marks the amplitude ratio measured from the global inflation waveform stack (Fig. S2a). The green diamond marks the amplitude ratio from Event 1. (f)-(j) Same as (a)-(e), except for measurements in the filtered-static-offset (FSO) frequency band (100-200 s). (k)-(o) the number of events used in each monthly stack against five subsets, respectively. The amplitude ratio in the FSO frequency band becomes more variable when the data quality is low or/and the number of events in the monthly stack is low.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[79, 130, 936, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[122, 639, 894, 694]]<|/det|> +
Fig. S6. Amplitude ratios between north-south and east-west tilt at N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[79, 130, 936, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 639, 894, 710]]<|/det|> +
Fig. S7. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly inflation waveform stacks in 2011-2016. Same as Fig. S5, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[80, 130, 933, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[122, 639, 892, 712]]<|/det|> +
Fig. S8. Amplitude ratios of east-west tilt between N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S5, except for the monthly deflation waveform stacks. The red line marks the amplitude ratio measured from the global deflation waveform stack (Fig. S2b).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[80, 130, 936, 631]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 639, 891, 694]]<|/det|> +
Fig. S9. Amplitude ratios between north and east tilts at N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio between north-south and east-west tilts at N.ASHV.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[78, 128, 936, 628]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 637, 893, 710]]<|/det|> +
Fig. S10. Amplitude ratios of vertical displacement between at N.ASIV and N.ASHV in the LPT and FSO frequency bands from monthly deflation waveform stacks in 2011-2016. Same as Fig. S8, except showing the amplitude ratio of vertical displacement between N.ASIV and N.ASHV.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[104, 131, 912, 626]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 637, 892, 766]]<|/det|> +
Fig. S11. Broadband displacement waveforms before and after the correction. (a-c) the original broadband displacement waveforms of the reference event in the vertical, east-west and north-south components at the four broadband seismic stations before removing the baseline error estimated from the polynomial fit. (d-f) the broadband displacement waveforms after removing the baseline errors. A \(4^{th}\) -order polynomial function is used to estimate the baseline error from the original displacement waveforms in the pre-event and post-event time windows. Vertical dashed lines mark the starting and ending times of the event (see also Method).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[65, 137, 949, 518]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[122, 529, 894, 692]]<|/det|> +
Fig. S12. Posterior probability functions of the inverted source parameters. (a-c) the marginal posterior probability functions against the source location with respect to the Naka-dake first crater. (d-e) the marginal probability functions of the strike and dip angles of the shearing and tensile components. (f) the marginal probability function against the regularization parameter \(\beta\) . (g-j) display the marginal probability functions against the moments of the strike-slip (Mss), dip-slip (Mds), tensile-crack (Mtc), and isotropic (Mex) components, respectively. (k-l) display the joint posterior probability functions of the coupled parameters regarding the bimodal probability functions in (d) and (e). The red line marks the preferred parameters with the maximum posterior probability. Two peaks in (d) and (e) define the orientations the fault plane and the auxiliary plane.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[280, 202, 740, 690]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[123, 701, 892, 884]]<|/det|> +
Fig. S13. Measurement of tilt offset in the spectra domain. (a) Synthetic tilt waveforms with the positive (red), null (gray), and negative (blue) offsets of \(1 \mu m\) . (b) Synthetic tilt waveform with a large offset of \(50 \mu m\) . (c) Synthetic tilt-rate amplitude spectra normalized against the amplitude of (b) at 10,000 s. Static offsets measured directly from (a) and (b) are normalized against the one from (b), shown in squares. (d) Observed monthly tilt waveform stacks with the positive (red), near-null (gray), and negative (blue) offsets. (e) Observed tilt waveform of the reference event. (f) Observed tilt-rate amplitude spectra normalized against the amplitude of (e) at 10,000 s. Static offsets measured directly from (d) and (e) are normalized against the one from (e), shown in squares. These examples show that the amplitude plateau of the tilt-rate waveform at 10000 s provides a reasonable estimate of the static offset.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 235, 925, 747]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[122, 759, 892, 850]]<|/det|> +
Fig. S14. Monthly volume changes in the five subsets. (a)-(e) Histograms of the volume change per event estimated from the monthly tilt waveform stacks in five subsets, respectively. Results from the monthly inflation and deflation stacks are shown in red and blue, respectively. (f-l) Same as (a)-(e), except displaying the monthly volume changes. The shaded region displays the uncertainty at the 99.7% confidence level from the bootstrap resampling (ref.36).
+ +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[156, 137, 857, 152]]<|/det|> +Table S1: Classifications of the ultra-long-period detections against the LPT catalog. + +<|ref|>table<|/ref|><|det|>[[203, 165, 810, 279]]<|/det|> + +
SubsetDefinitionNumberPercentage
I\(CC\geq 0.68\)and \(\mathrm {SNR}\geq 6.9\)55622.7%
II\(CC\geq 0.58\)and \(\mathrm {SNR}\geq 5.3\)(excluding I)114165.5%
III\(CC\geq 0.42\)and \(\mathrm {SNR}\geq 3.7\)(excluding I-II)3576617.1%
IV\(CC\geq 0.30\)and \(\mathrm {SNR}\geq 2.8\)(excluding I-III)4596422.0%
V\(CC<0.30\)or \(\mathrm {SNR}<2.8\)10980952.7%
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[70, 159, 840, 312]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[213, 125, 689, 149]]<|/det|> +Table S2: Summary of inverted source parameters for the reference event. + +
Polynomial orderxsys [km]zsφ [°]θ [°]mssmds [×1014N·m]mtcmexfE [×108N]fN [m3]fUΔVn [m3]Mf:MDMw [°]γ [°]δ [°]
3-2.33-0.642.41324480.007-0.258-0.1731.006158023.2:13.313.272.6
4-1.96-1.022.9833160-0.021-0.283-0.0020.993258394.3:13.30.176.8
5-1.97-1.082.8433460-0.014-0.2860.0190.903247754.0:13.3-1.575.9
6-2.05-0.353.55313480.003-0.395-0.1381.583330154.2:13.47.476.4
4-1.97-1.143.0132758-0.015-0.275-0.0110.966-0.005-0.0050.005246104.2:13.30.976.7
+ +<|ref|>text<|/ref|><|det|>[[42, 319, 866, 465]]<|/det|> +Note: \(x_{s}, y_{s}\) and \(z_{s}\) are the 3- D Cartesian location of the preferred source along the eastward, northward and downward directions referred to the Naka- dake first crater, respectively; \(\phi\) and \(\theta\) are the strike and dip angles of the shearing and tensile components, respectively; \(m_{ss}\) , \(m_{ds}\) , \(m_{tc}\) and \(m_{ex}\) are the moments of strike- slip, dip- slip, tensile crack and explosion, respectively; \(f_{E}\) , \(f_{N}\) and \(f_{U}\) are the magnitude of singe force along the eastward, northward and upward directions, respectively; \(\Delta V_{m}\) is the equivalent Mogi volume change; \(M_{I}\) and \(M_{D}\) are the isotropic and deviatoric scalar seismic moments of the reconstructed moment tensor (ref.30), respectively; \(\gamma\) and \(\delta\) are the longitude and latitude with pole (1, 1, 1) for the fundamental lune (ref.86), respectively. + +<--- Page Split ---> diff --git a/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/images_list.json b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ab88ab136843251813adea16fabfb589e1a6859c --- /dev/null +++ b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/images_list.json @@ -0,0 +1,107 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Single and combined mutants by zebrafish zygotic genome activators used in this study.", + "footnote": [], + "bbox": [ + [ + 115, + 131, + 870, + 666 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Changes in chromatin accessibility parallel the changes in gene expression in Pou5f3, Sox19b and Nanog triple mutant.", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 857, + 760 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Four types of TdARs by pioneer-like activity of Pou5f3, SoxB1, and Nanog.", + "footnote": [], + "bbox": [ + [ + 115, + 81, + 880, + 500 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Activator-blocker model: Pou5f3 and Nanog oppose each other effects on H3K27 acetylation and chromatin accessibility on a fraction of enhancers.", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 884, + 603 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Zygotic gene expression is balanced by synergy and competition of Pou5f3 and Nanog on common enhancers.", + "footnote": [], + "bbox": [ + [ + 115, + 78, + 880, + 750 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. Pou5f3 blocks premature transcriptional activation by Nanog and vice versa.", + "footnote": [], + "bbox": [ + [ + 115, + 75, + 880, + 744 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7. Synergistic and antagonistic Pou5f3/Nanog direct enhancers: the mechanism of action and direct target genes.", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 884, + 319 + ] + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2.mmd b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7a4a43859074f9107030a41f4e425d2939bfd123 --- /dev/null +++ b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2.mmd @@ -0,0 +1,482 @@ + +# Activator-blocker model of transcriptional regulation by pioneer-like factors + +Daria Onichtchouk (daria.onichtchouk@biologie.uni- freiburg.de) + +University of Freiburg https://orcid.org/0000- 0001- 6497- 1445 + +Aileen Riesle University of Freiburg https://orcid.org/0000- 0002- 8195- 1291 + +Meijiang Gao University of Freiburg + +Marcus Rosenblatt University of Freiburg + +Jacques Hermes University of Freiburg https://orcid.org/0000- 0003- 3608- 5969 + +Helge Hass University of Freiburg + +Anna Gebhard University of Freiburg + +Marina Veil University of Freiburg + +Bjoern Gruening University of Freiburg + +Jens Timmer University of Freiburg https://orcid.org/0000- 0003- 4517- 1383 + +## Article + +Keywords: + +Posted Date: January 25th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2485402/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on September 14th, 2023. See the published version at https://doi.org/10.1038/s41467-023-41507-z. + +<--- Page Split ---> + +# Activator-blocker model of transcriptional regulation by pioneer-like factors + +Aileen Julia Riesle\*,1,8, Meijiang Gao\*,1, Marcus Rosenblatt\*,2,4, Jacques Hermes\*,2,4 Helge Hass2,4, Anna Gebhard1, Marina Veil1, Björn Grüning5,6, Jens Timmer\*\*,2,3,4, Daria Onichtchouk\*\*1,3,7 + +# \*contributed equally + +1Department of Developmental Biology, Albert- Ludwigs- University of Freiburg, 79104 Freiburg, Germany + +2Institute of Physics, Albert- Ludwigs- University of Freiburg, 79104 Freiburg, Germany + +3Signalling Research centers BIOSS and CIBSS, 79104 Freiburg, Germany + +4Freiburg Center for Data Analysis and Modelling (FDM), 79104 Freiburg, Germany + +5Department of Computer Science, University of Freiburg, 79110 Freiburg, Germany + +6Center for Biological Systems Analysis (ZBSA), University of Freiburg, 79104 Freiburg, Germany + +7Institute of Developmental Biology RAS, 119991 Moscow, Russia + +8Current address: Epigenetics and Neurobiology Unit, European Molecular Biology Laboratory, EMBL Rome, Adriano Buzzati- Traverso Campus, Via Ramarini 32, 00015 Monterotondo (RM), Italy + +<--- Page Split ---> + +## Abstract + +Zygotic genome activation (ZGA) in the development of flies, fish, frogs and mammals depends on pioneer- like transcription factors (TFs). Those TFs create open chromatin regions, promote histone acetylation on enhancers, and activate transcription. Here, we use the panel of single, double and triple mutants for zebrafish genome activators Pou5f3, Sox19b and Nanog, multi- omics and mathematical modeling to investigate the combinatorial mechanisms of genome activation. We show that Pou5f3 and Nanog act differently on synergistic and antagonistic enhancer types. Pou5f3 and Nanog both bind as pioneers on synergistic enhancers, promote histone acetylation and activate transcription. Antagonistic enhancers are activated by pioneer binding of one of these factors. The other TF binds as non- pioneer, competes with the activator and blocks all its effects, partially or completely. This activator- blocker mechanism mutually restricts widespread transcriptional activation by Pou5f3 and Nanog and prevents premature expression of late developmental regulators in the early embryo. + +<--- Page Split ---> + +## Introduction + +Awakening of zygotic transcription during maternal- to- zygotic transition is a universal feature of multicellular organisms. The major wave of Zygotic Genome Activation (ZGA) is driven by sequence- specific transcription factors (TFs), different across species \(^{1,2}\) . Nucleosomes protect enhancers from binding of most TFs and are the main obstacles for transcriptional activation. The small group of pioneer TFs directly bind to nucleosomes and create open, nucleosome- free chromatin regions on enhancers \(^{3}\) . Zygotic genome activators act as pioneer- like factors, by opening chromatin, initiating enhancer activation, and widespread zygotic gene expression \(^{4,5}\) . Direct nucleosome binding was shown for some of the genome activators \(^{6,7}\) . + +In animals, gene products synthetized at ZGA enable gastrulation and subdivide the embryo into the three germ layers, ectoderm, mesoderm and endoderm. The regulators of later developmental programs, organogenesis and cell lineage specification, are kept silent at ZGA. They will be synthesized shortly before the beginning of appropriate developmental stages \(^{8,9}\) . The expression waves of transcripts encoding cohorts of transcriptional regulators follow each other in precise order in the developmental time course \(^{10}\) . Molecular mechanisms which keep lineage specifying genes silent at ZGA are unknown. + +Maternal transcription factors Pou5f3, Sox19b and Nanog (PSN) together activate zygotic gene expression in zebrafish \(^{11 - 13}\) . Sox19b is the only maternal member of the SoxB1 family. Shortly after ZGA, early zygotic SoxB1 factors Sox3, Sox19a and Sox2 are expressed and act redundantly with Sox19b \(^{14}\) . PSN, as well as their mammalian counterparts, frequently colocalize on the same genomic sites; therefore, they initially were thought to cooperate \(^{12,15}\) . However, recent studies in mouse and fish revealed that the factors may act alone, additively or interchangeably \(^{5,16,17}\) . PSN open chromatin on more than half of active enhancers at ZGA, but are necessary for expression of only a fraction of early zygotic genes \(^{5}\) . It is unclear how widespread pioneer- like activity of PSN on enhancers relates to early zygotic transcription. + +The purpose of this study was to understand the mechanisms of combined Pou5f3, Sox19b and Nanog action, and the connection between PSN pioneer activity, enhancer activation and selection of the early zygotic gene expression repertoire. We found that Pou5f3 and Nanog in combination not only activate part of zygotic transcripts, but reciprocally restrict or block transcriptional activation by each other on hundreds of developmental enhancers. Further, competition between Pou5f3 and Nanog on common genomic sites establishes the order of gene expression after ZGA. + +<--- Page Split ---> + +## Results + +## Study design and overview of the ATAC-seq data set + +For multi- omic assays, we used the panel of eight zebrafish genotypes: zebrafish wildtype embryos (WT), maternal- zygotic mutants for Pou5f3 (P, MZspg), Sox19b (S, MZsox19b), and Nanog (N, MZnanog), double mutants MZps, MZpn, MZsn, and the triple mutant MZtriple. Sox19b was dispensable for normal development, the absence of Pou5f3, Pou5f3/Sox19b or Nanog resulted in abnormal gastrulation as previously reported \(^{17 - 19}\) . At the absence of Pou5f3/Nanog, Sox19b/Nanog or all three factors zygotic development was arrested (Fig. 1a). + +To assess individual and combined effects of Pou5f3, Sox19b, and Nanog on enhancer regulation and transcription, we performed ATAC- seq and time- resolved RNA- seq profiling in all genotypes. We then aimed to derive enhancer groups differentially regulated by PSN using genomic data (Table S1), and differentially regulated target gene groups using transcriptome (Table S2). Finally, for cross- validation, we put together the independently obtained results of the genome and transcriptome parts (Table S3). The outline of data analysis is shown in Fig. S1. + +We started the genomic analysis by estimating the differences in the chromatin accessibility between the genotypes, on 102945 Accessible Regions (ARs) in the wild- type. We made three initial observations using Principal Component Analysis (PCA, Fig. 1b). First, chromatin accessibility profiles in MZsox19b mutants were close to the wild- type. Sox19b is the major SoxB1 factor expressed at 3.7 and 4.3 hpf, we therefore assumed that the Sox19b factors acts as a cofactor or redundantly with Pou5f3 or Nanog. Second, accessibility profiles of MZsn, MZpn, and MZtriple mutants clustered together in PCA space, corresponding to the most severe phenotypes of these mutants. Third, accessibility profiles of Pou5f3 and Nanog single mutants were quite different, suggesting that these TFs change chromatin accessibility on different regions, or in the opposite directions. + +## Analysis of MZtriple mutant suggests the presence of additional genome activators in early zebrafish embryo + +To tease apart the apparently complex regulation of chromatin accessibility by Pou5f3, SoxB1 and Nanog, we first characterized the effects of their combined activity on chromatin using MZtriple mutant. We divided all ARs to three groups: "down", where the chromatin accessibility was reduced in MZtriple compared to the wild- type (31% of all ARs), "up", where the chromatin accessibility was increased (15% of all ARs), and "same" - unchanged (Fig. 2A). Sequence- specific binding motifs for all three factors were enriched in "down" regions, while "up" regions had the highest content of G and C nucleotides and of GC- rich TF- binding motifs (Fig. 2b, c). The genes closest to "down" and "up" ARs were both enriched in transcriptional regulatory functions among the others (Fig. 2d). + +Next, we analyzed the time- resolved transcriptome in MZtriple mutant compared to the wild- type, using our previously developed R package RNA- sense (www.bioconductor.org/packages/release/bio/html/RNAsense.html \(^{17}\) ). We selected 4777 transcripts zygotically expressed in the wild- type or in MZtriple, and split them into three groups (Fig. 2e): "DOWN" (downregulated compared to the wild- type or not expressed in MZtriple), "SAME" (unchanged), and "UP" (upregulated in MZtriple compared to the wild- type, or expressed only in MZtriple). + +<--- Page Split ---> + +Out of 791 "UP" group transcripts, \(67\%\) were zygotically expressed only in MZtriple (i.e. hoxc8a, nr2f2). We wondered if lineage- specifying regulatory genes were prematurely expressed MZtriple, as we observed previously in MZspg and MZps mutants \(^{17,20}\) . To test that, we used published data \(^{10}\) to calculate the maximal expression time during normal zebrafish development for each zygotic gene. Indeed, the median expression time in the "UP" group was 24 hpf, versus 8 hpf in the groups "DOWN" and "SAME" (Fig. 2f). Similar analysis for other mutants revealed that except for MZsox19b, all of them prematurely expressed diverse sets of late regulatory genes (Fig. S2). + +Finally, we checked if the changes in chromatin accessibility on the putative regulatory regions of zygotic genes correlated with the changes in their expression in MZtriple. This was indeed the case: "down" ARs were enriched around the promoters of "DOWN" genes, and "up" ARs around "UP" genes in MZtriple (Fig. 2g). + +Taken together, our analysis confirmed previous findings that Pou5f3, Sox19b and Nanog pioneer chromatin accessibility on the enhancers of early zygotic genes \(^{5,21,22}\) . The increase in transcription in MZtriple mutants was unexpected. We concluded that additional activators were present at ZGA. These unknown activators could bind to GC- rich regulatory elements. + +Four types of regulation of chromatin accessibility by Pou5f3, SoxB1 and Nanog To investigate how Pou5f3, SoxB1 and Nanog create the regions of open chromatin, we restricted our analysis to the ARs, where any combination of the three factors was required for chromatin accessibility ("down" in MZtriple, Fig. 2a). To study only direct TF effects, we further selected 20131 TdARs (TF- bound Accessible Regions, downregulated in the MZtriple mutant), overlapping with ChIP- seq peaks for Pou5f3, SoxB1 \(^{12}\) , or Nanog \(^{23}\) . + +We then classified TdARs by non- redundant requirements for Pou5f3 and Nanog for chromatin accessibility into four groups (Fig. 3a). Accessibility of the 1. PN group TdARs was reduced in MZspg and MZnanog, so we concluded that Pou5f3 and Nanog were both required in these regions. Pou5f3 was required in the 2. P group, Nanog in the 3. N group, and none of the single factors was required in the 4. - group (i.e. several factors were required redundantly). Out of the four groups, 2. P and 3. N were the most enriched for the Pou5f3- and Nanog - motifs, and the most highly occupied by Pou5f3 and Nanog, respectively (Fig. 3b, c). + +To address if chromatin becomes accessible in all groups at the same time, we performed ATAC- seq at major ZGA (3 hpf), and compared ATAC- seq signals at 3, 3.7 and 4.3 hpf, in the wild- type and MZtriple. While all groups depended on PSN already at ZGA, chromatin in 4. - group was more accessible in MZtriple in all stages than on the other groups (Fig. 3d, dashed lines). Therefore, it was conceivable that factors other than PSN also contribute to chromatin accessibility in the group 4. - . We found that GC content increased in order \(1. \mathrm{PN} < 2. \mathrm{P} < 3. \mathrm{N} < 4. - ,\) (Fig. 3e). As our analysis of MZtriple mutants suggested that GC- binding activators were present at ZGA, we hypothesized that GC- binding factor X could be involved in opening chromatin in group 4. - regions (Fig. 3f). + +<--- Page Split ---> + +We next wondered whether the factors required for chromatin accessibility in a given group of TdARs were also sufficient to open chromatin. We used rescue experiments of Miao et al., \(2022^{5}\) to answer this question. To address which factors can restore chromatin accessibility in the triple mutants MZnpS, the authors microinjected Pou5f3, Sox19b, and Nanog mRNAs, individually and in combinations, and compared ATAC- seq signals in the injected and non- injected MZnpS embryos. Using their data, we scored the chromatin accessibility rescue in each of the four groups. Nanog alone rescued more than \(90\%\) of TdARs in the 3. N group. Pou5f3 alone rescued only \(26\%\) of TdARs in the 2. P group; most of the rest could be rescued by the Pou5f3/Sox19b or Pou5f3/Nanog combinations (Fig. S3a- f). Further, in \(59\%\) of TdARs in the 2. P group chromatin accessibility was also downregulated in the double Sox19b/Nanog mutant MZsn (Fig. S3g, h). Thus, although Pou5f3 was present in MZsn, it was apparently not sufficient to open chromatin in most of its target regions. In sum, we concluded that Nanog alone was required and sufficient for establishing chromatin accessibility in the majority of its target regions. In contrast, Pou5f3 needed assistance of either Nanog or SoxB1 to establish accessible chromatin. + +Fig. 3f summarizes four types of establishment of chromatin accessibility by pioneer- like factors. Pou5f3 and Nanog are both required on the 1. PN group TdARs. Pou5f3 and redundant contribution of Nanog or SoxB1 are required on most of the 2. P group TdARs. Nanog is required on most of the 3. N group TdARs. Pou5f3, SoxB1 or Nanog are redundantly required on 4. - group TdARs, together with a hypothetical GC- binding factor X. + +## Pou5f3 and Nanog act as Activator and Blocker and vice versa on a fraction of common enhancers + +Pou5f3, SoxB1, and Nanog together promote histone acetylation on their binding sites, which is critical for the enhancer activation5. However, individual roles of the TFs remain unclear. We used H3K27ac ChIP- seq data in the single mutants17 (and this work) and MZnpS triple mutant5 to analyze how H3K27 acetylation relates to the pioneer- like activity of the factors. + +We found that the same TF which was strictly required for chromatin accessibility was required for H3K27 acetylation: both Pou5f3 and Nanog in the 1. PN group, Pou5f3 in the 2. P group, Nanog in the 3. N group, and none of the single factors in the 4. - group. Unexpectedly, single mutant analysis revealed reciprocal antagonistic relationships between Pou5f3/Sox19b and Nanog: Nanog reduced H3K27ac in the 2. P group, and Pou5f3 reduced H3K27ac in the 3. N group (Fig. 4a). The PSN combined activity was required for H3K27 acetylation in all groups (Fig. 4b). + +To further investigate the individual roles of the TFs, we defined H3K27- acetylated TdARs as enhancers (STAR methods, Table S1). We then scored the enhancers on which H3K27ac was induced (+), unchanged (0), or reduced (- ) by each TF (Fig. 4c), or by the PSN combined activity (Fig. 4d), in each of the four groups. We noted that the percentage of enhancers activated by PSN was the highest in the 1. PN and the lowest in the 4. - group (Fig. 4c, d), and thus inversely correlated with GC content (Fig. 3e). Importantly, our analysis uncovered a large number of previously uncharacterized "antagonistic enhancers", where Pou5f3 and Nanog regulated H3K27ac in the opposite directions (total 2504 p+n- and p-n+ enhancers, boxed in Fig. 4c). + +<--- Page Split ---> + +To understand how per se transcriptional activators and pioneer- like factors could negatively regulate enhancer activity, we took a closer look on chromatin accessibility changes on the antagonistic enhancers. We found that the factor which reduced H3K27ac also reduced chromatin accessibility (Fig. 4e, Fig. S4a- d, f). We confirmed this result using MNase- seq nucleosome positioning data22 (Fig. S4e, f). Together, we have shown by two independent methods that only one of the TFs, either Pou5f3 or Nanog, acts as pioneer- like factor on the antagonistic enhancers. This observation suggested that Pou5f3 and Nanog alternate their binding mode, acting as pioneer- like factors on some genomic sites and as non- pioneers on the others. The non- pioneer TF reduces chromatin accessibility, perhaps by competing for common genomic site with the pioneer- like TF. + +Our next question was whether the positive and negative effects of Pou5f3 and Nanog on chromatin accessibility correlate with sequence features of the bound sites. As shown in Fig. S4g, Pou5f3 induced chromatin accessibility on its own motifs and reduced it on Nanog motifs, and vice versa. While common homeodomain- binding sequence in Pou5f3 and Nanog cognate motifs is recognized by both TF(Fig. 3b), exact match to the motif is necessary for the pioneer- like activity. + +Summarizing our findings, we suggest the general mechanism of antagonistic interactions between Pou5f3 and Nanog. Two TFs, A (activator) and B (blocker) compete for binding on the common motif within an enhancer (Fig. 4f). The activator acts as a pioneer- like factor on this motif: it displaces nucleosomes and promotes histone acetylation. Non- pioneer binding of the blocker protects the motif from the activator, thereby reducing nucleosome displacement and histone acetylation. Two examples illustrate the model. On her3 regulatory region, Pou5f3 together with SoxB1 act as an activator and Nanog as a blocker on sox:pou motif (Fig. 4g). On more3b regulatory region, Nanog acts as an activator and Pou5f3 as a blocker on two nearby nanog motifs (Fig. 4h). Hence, Pou5f3 and Nanog compete as pioneer- like and non- pioneer factors on common genomic sites, eliciting opposite effects on enhancer activity. + +## Mathematical modelling of transcriptional regulation by genome activators + +We wondered next, whether the antagonistic enhancers are relevant for zygotic gene expression. To answer this question, we chose the strategy to on the one hand sort zygotic transcripts into groups according to their regulation, without involving genomic information. On the other hand, we linked enhancers to transcripts and determined correlations between the enhancer and transcript groups. + +To sort the transcriptome, we used a mathematical modeling approach in two subsequent steps. In the first step, we developed a "core model" to describe the dynamic behavior of the known core components of the system: Pou5f3, Nanog, and SoxB1 group genes (including Sox19b, Sox19a, Sox3 and Sox2). In the second step, dynamics of these core components were then used to inform ODE- based mini models to formalize all possible regulatory inputs from Pou5f3, Nanog or SoxB1 on the individual target genes. + +The core model was based on ordinary differential equations (ODEs), where initial assumptions on the regulatory interactions between the core components were taken from the literature (STAR methods, Fig. S5a for model cartoon and Table S4 for model equations). To infer the model's parameters, we used the measured time- resolved + +<--- Page Split ---> + +transcriptional profiles for pou5f3, nanog, sox19b, sox19a, sox3 and sox2. After model calibration, the model trajectories resulting from the best fit parameters were in good agreement with the experimental data (Fig. S5b, c, Fig. S6a, and Table S5). + +For the second step of ODE- based mini models, we assumed that at least one factor of Pou5f3, Nanog or SoxB1 should activate the target, while the other two can activate \((+)\) , repress \((- )\) or do nothing (0). This gives 19 possible regulatory combinations. Each of the 19 mini models was separately fitted to experimental time- resolved RNA- seq data for each of the 1799 zygotic transcripts that had been found to be directly or indirectly activated by Pou5f3, Sox19b or Nanog ("DOWN" group in Fig. 2e, Table S6). The dynamics of Pou5f3, Nanog, and SoxB1 group genes (SOX) obtained in step one of the analysis were used as input to the mini models, while the remaining parameters were again calibrated based on the RNA- seq data. To simplify the analysis with respect to our goal of connecting chromatin state and transcription for Pou5f3 and Nanog, we merged the 19 mini- models down to six mini- model groups capturing all possible variants of target regulations by Pou5f3 and Nanog and the regulation by SOXB1 alone (Fig. 5a). + +For each transcript, we selected the best candidate mini- model group according to Bayesian Information Criterion (BIC, see Table S7 and Table S2). Fits of selected mini- model groups are shown for exemplary targets in Fig. 5b and Fig. S6b. 1799 transcripts were then classified into six groups according to the best fitting direct regulatory scenario (Fig.5c). + +Interestingly, three out of six groups of targets significantly associated with gene ontology terms. Groups 1. \(\mathrm{P + N + }\) and 3b. \(\mathrm{P - N + }\) were enriched in the regulators of transcription and development, while the largest group 3a. \(\mathrm{PON + }\) was enriched in the housekeeping genes (Fig. S7a). In sum, we sorted PSN- dependent zygotic transcripts to six groups with biologically relevant functions, using formal and unbiased mathematical modelling approach. + +## Competition between Pou5f3 and Nanog on antagonistic enhancers establishes the order of gene expression. + +To inquire if synergistic and antagonistic chromatin regulation by Pou5f3 and Nanog on common enhancers is directly linked to synergistic and antagonistic regulation of transcription, we put together the results of genomic and transcriptomic analyses. As shown in Fig. 5d- f (see also Fig. S7b, c), the predicted direction of transcriptional regulation of targets by Pou5f3 and Nanog (activation or repression) strongly correlated with Pou5f3 and Nanog- dependent H3K27ac changes on their enhancers (activation or blocking), respectively. Thus, cross- validation of independent genomic and transcriptomic data sets demonstrated that the transcription at ZGA is directly regulated by synergistic and antagonistic enhancers. + +We have noticed that all mutants except MZsox19b prematurely transcribe poorly overlapping sets of genes, enriched for DNA- binding and transcription regulatory functions (Fig. S2). We also demonstrated that Pou5f3 and Nanog compete on antagonistic enhancers as activator and blocker of chromatin accessibility and H3K27 acetylation. Putting these two findings together, we wondered again whether Pou5f3 could directly block premature transcriptional activation by Nanog and vice versa. To answer this question, we examined putative regulatory regions of the transcripts, + +<--- Page Split ---> + +upregulated in the Pou5f3 and Nanog mutants, for the enrichment of antagonistic enhancers of the opposite types. + +In Pou5f3 mutant MZspg, \(17\%\) of transcripts were upregulated compared to the wildtype (Fig. 6a) and contained prematurely expressed genes (Fig. 6b). Strikingly, the putative regulatory regions of these genes were highly enriched for antagonistic 3. N p- n+ enhancers (Fig. 6c, boxed). 3. N p- n+ enhancers were also enriched in the putative regulatory regions of genes upregulated in the double Pou5f3/Sox19b mutant MZps (Fig. S7d- f). Thus, Pou5f3 blocked inappropriate transcriptional activation of late developmental genes by Nanog. + +To test if the reverse was also true, we examined the transcripts upregulated in MZnanog. \(10\%\) of transcripts were upregulated compared to the wild- type in MZnanog and contained prematurely expressed genes (Fig. 6d, e). The putative regulatory regions of these genes were enriched for antagonistic 2. P p+ n- enhancers (Fig. 6f, boxed). Thus, Nanog blocked inappropriate transcriptional activation of late developmental genes by Pou5f3. + +In sum, cross- validation of independent genomic and transcriptomic data sets demonstrated that the transcription of PSN target genes at ZGA is directly regulated by synergistic and antagonistic enhancers. On the synergistic enhancers, Pou5f3 and Nanog are both required to establish chromatin accessibility and activate early zygotic genes (Fig. 7a). On the antagonistic enhancers, "blocker" TF(- ) attenuates activation of the early zygotic genes by "activator" (TF+), or prevents the activation of lineage- specific regulators by "activator" TF (Fig. 7b, c). + +## Discussion + +Our study provides the mechanistic links between the pioneer- like activity of genome activators Pou5f3, Sox19b (together with zygotic SoxB1 factors) and Nanog, and their potential to activate transcription at ZGA and during the first half of gastrulation. We demonstrate that direct binding of PSN to genome is not only responsible for transcriptional activation at ZGA, but also for adjusting the levels of early zygotic transcripts and for the timing of gene expression in development. + +Synergistic interactions were long ago suggested for zebrafish genome activators and their mammalian homologs12,15. Antagonistic interactions of Pou5f3 and Nanog on chromatin are a new finding. A reasonable question is which features define pioneer- and non- pioneer binding in each case. We found a correlation with two sequence features. One feature is GC content around TF binding motifs: on synergistic enhancers, GC content is lower than on the other enhancer types (Fig. 3e, Fig. 4c). Another feature is the different frequency of sequence- specific motifs: we found that a given TF acts more frequently as activator on its own motifs, and as blocker as it binds to the motifs of the other factor (Fig. 4f- h, Fig. S4g). In a sense, "activator" and "blocker" TFs can be compared with a precise key and a key blank to the locked door (which is an enhancer): both keys fit to the lock, but only a precise key can open the door (activate an enhancer). + +In whole embryo assays, we could not distinguish whether the ATAC- seq signals come from the whole embryo or from the subpopulation of cells. Therefore, we cannot exclude that cell- specific differences, such as local concentrations of TFs, presence of transcriptional cofactors or cross- talk with signaling pathways affect the binding mode + +<--- Page Split ---> + +cell type- specifically. In line with that, it was shown that nucleosome displacement activity of known mammalian pioneer factors, including Oct4/Pou5f1, is conditional: all of them bind to different genomic locations in different cell types24. Further, it is also possible that antagonistic enhancers act as bifunctional cis- regulatory elements, i.e. as activators or as silencers, depending on the embryonic cell type25. Single cell multi- omics and reporter assays are needed to clarify this issue. + +The dual role of Pou5f3 and Nanog in activating early zygotic genes and blocking premature transcription provides an immediate parallel to the biological roles of their mammalian homologues. Oct4/Pou5f1 and Nanog regulate the seemingly opposite processes of pluripotency maintenance and cell lineage decisions in mammalian embryos and ES cells26- 30. Although zebrafish Pou5f3 is not equivalent to Oct4/Pou5f1 in terms of nucleosome- binding and reprogramming capacity31, Pou5f3 and Oct4 recognize the same motifs and colocalize with Nanog in the respective systems. It is tempting to speculate that Oct4 and Nanog may reciprocally block some of each other activities similarly to their zebrafish homologues. Recent study supports this possibility: acute depletion of Oct4 in ES cells results in increased genomic binding of Nanog, suggesting that Oct4 outcompetes Nanog on chromatin and antagonizes its function32. + +Our results suggest that the list of zebrafish genome activators is not complete: hundreds of transcripts, enriched in transcriptional regulatory functions, are prematurely expressed in MZtriple mutant and linked to regulatory elements with high GC content (Fig. 2). Identification of putative GC- binding factors regulating these genes and dissecting their interplay with Pou5f3, SoxB1 and Nanog will further clarify the mechanisms of gene regulation at ZGA. Finally, the concepts presented in this work will assist the studies of combinatorial genome activation by TFs in other organisms and deepen our understanding of two global and currently unresolved questions: conditional activity of pioneer and pioneer- like factors in different biological settings, and of how the gene expression is regulated in developmental time. + +## Acknowledgments + +We are grateful to Rainer Duden for commenting on the manuscript, to Sabine Götter for fish care, and to Andrea Buderer and Cornelia Wagner for administrative support. This work was supported by DFG- ON86/4- 2 and DFG- ON86/6- 1 for DO, and DFG- EXC2189 - Project ID: 390939984 for D.O. and J.T. BG is funded by DFG grant SFB 992/1 2012 and BMBF grant 031 A538A RBC. + +## Author Contributions + +AJR, MG, MV, and AG - experiments; AJR, MG, BG and DO - data analysis, MR and JH - mathematical modeling, DO, JT - supervision of the study. DO wrote the manuscript, all authors edited the manuscript. + +## Declarations of Interests + +The authors declare no competing interests + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Single and combined mutants by zebrafish zygotic genome activators used in this study.
+ +a. Mutant phenotypes at blastula (dome) and midgastrula (shield) stages. White dotted line shows epiboly border; double arrows show the internalized yolk. Scale bar=100μm. b. Principal Component Analysis of ATAC-seq data in all mutants, on the genomic regions accessible in the wild-type (ARs). The data in eight genotypes clustered into four groups (boxed). + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Changes in chromatin accessibility parallel the changes in gene expression in Pou5f3, Sox19b and Nanog triple mutant.
+ +a. Three groups of accessible regions (ARs) by chromatin accessibility change in the MZtriple: "down", "same", "up". b. ARs of "up" group had the highest GC content. P-values for Tukey-Kramer test; p-value for 1-way ANOVA was \(< 2e^{-16}\) . Co - control genomic regions (dotted line). c. Pou5f3, Sox and Nanog-binding motifs (black rectangle) are enriched in "down" ARs. d. GREAT analysis. e. Three groups of zygotic transcripts by expression change in MZtripleThe heatmap shows normalized expression at eight time points, from pre-ZGA (2.5 hpf) till 6 hpf; example developmental genes at the left. f. Group of zygotic genes is prematurely expressed in + +<--- Page Split ---> + +MZtriple. Y- axis: time of maximal expression in the normal development (3 hpf - 120 hpf), schematic embryo drawings illustrate the stages. Median expression time of transcripts upregulated in MZtriple was 24 hpf (yellow dotted line), versus 8 hpf for down- or unchanged transcripts (gray dotted line). P- values in Tukey- Kramer test; p- value in 1- way ANOVA was \(< 2e^{- 16}\) . g. Down- or upregulation of chromatin accessibility in MZtriple correlates with respective transcriptional changes of linked genes. \(\chi^2\) test; positive correlations are shown in blue and negative in red. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Four types of TdARs by pioneer-like activity of Pou5f3, SoxB1, and Nanog.
+ +a. Venn diagram. Four groups of TdARs (by non-redundant requirements for Pou5f3 and/or Nanog for chromatin accessibility. b. Frequencies of PSN motifs in the 4 groups. Black box - common TAAT sequence, recognized by homeodomains. Note that 2.P and 3.N groups are the most enriched in the Pou5f3- and Nanog- specific motifs, respectively (colored boxes). c. 2.P and 3.N groups are the mostly occupied by Pou5f3/SoxB1 and Nanog, respectively (gray dashed lines). Summary ChIP-seq profiles for indicated TFs (rpkm). d. PSN establish accessible chromatin starting from major ZGA (3 hpf). ATAC-seq summary profiles in WT and MZtriple at 3, 3.7 and 4.3 hpf (rpkm). Note that chromatin in group 4.- regions in MZtriple is more accessible than in the other groups (gray dashed line). e. 4.- group has the highest GC content out of all groups. P-values for Tukey-Kramer test; p-value for 1-way ANOVA was \(< 2e^{-16}\) . Lower dashed line shows median genomic control GC content, upper line - median GC content of ARs upregulated in MZtriple. f. Schematic drawing of four types of regulation of chromatin accessibility, percentage of each group from all TdARs is shown. "OR" - logical operator. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. Activator-blocker model: Pou5f3 and Nanog oppose each other effects on H3K27 acetylation and chromatin accessibility on a fraction of enhancers.
+ +a. Summary profiles of H3K27ac histone mark in the wild-type and single mutants, in four types of TdARs: 1. PN, 2. P, 3. N, 4. - . Note the opposite effects of Pou5f3 and Nanog on H3K27 acetylation on the 2. P and the 3. N groups. b. Summary profiles of H3K27ac mark in the WT and MZ_nps. c. Effects of the single TFs on H3K27ac. Antagonistic enhancers are boxed in the legend. d. Effects of PSN combined activity on H3K27ac. e. Pou5f3 and Nanog have the Opposite effects of Pou5f3 and Nanog on chromatin accessibility on antagonistic enhancers. f. Schematic illustration of activator-blocker model (explanations in the text). (g, h) Genomic browser views show (from bottom to top) motif occurrence, TF binding, H3K27ac and ATAC-seq in the indicated genotypes. g. All three TFs bind to sox:pou motif on her3 2. P p+n- antagonistic enhancer (blue shading). h. Pou5f3 and Nanog, or all three factors bind nanog motifs on morc3b 3. N p+n+ antagonistic enhancers (yellow shading). \* - data from5. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. Zygotic gene expression is balanced by synergy and competition of Pou5f3 and Nanog on common enhancers.
+ +a. Schemes of six alternative mini-model groups for direct target regulation by Pou5f3, Nanog and SoxB1. b. foxb1a and tbx5a transcripts dynamics fitted best to the antagonistic groups 2b. P+N- and 3b. P+N+ respectively. c. Heatmap of zygotic genes, downregulated in MZtriple, sorted by best fit to one of the six model groups. d-f. \(\chi^2\) tests, positive correlations between the transcriptomics and genomics groups are shown in blue and negative in red. Vertical axis: transcriptomics groups. Direct enhancers linked to the promoters of zygotic genes were sorted by best fit transcriptional model. Synergistic and antagonistic model groups are boxed. Horizontal + +<--- Page Split ---> + +axis: direct enhancers were sorted by chromatin accessibility, and by H3K27ac regulation by sum of PSN activities (d), by Pou5f3 (e) or by Nanog (f). \*- data from \(^{5}\) . + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. Pou5f3 blocks premature transcriptional activation by Nanog and vice versa.
+ +(a- c). Pou5f3 blocks transcription from Nanog- dependent enhancers. (d- f). Nanog blocks transcription from Pou5f3- dependent enhancers. Zygotic transcripts were sorted by expression change in MZspg (a) or MZnanog (d) compared to the wild- type. Heatmaps show normalized expression at eight time points, from pre- ZGA (2.5 hpf) till 6 hpf; numbers of transcripts in each group are shown at the right. b, e. Premature expression of zygotic genes in MZspg (b) or in MZnanog (e). Median expression time in the normal development was compared in "UP" "DOWN" and "SAME" groups. P- values in Tukey- Kramer test; p- value in 1- way ANOVA was \(< 2e^{- 16}\) . c, f. \(\chi^2\) tests, positive correlations between the transcriptomics and genomics groups are shown in + +<--- Page Split ---> + +blue and negative in red, scales show Pearson residuals. Vertical axis: transcriptomics groups. Direct enhancers were linked to the promoters of zygotic genes within \(+ / - 50\) kb and sorted according to DOWN", "SAME" and "UP" transcriptional groups. Horizontal axis: genomic groups. Direct enhancers were sorted by changes in chromatin accessibility, and by H3K27ac regulation by Pou5f3 or Nanog, as indicated. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7. Synergistic and antagonistic Pou5f3/Nanog direct enhancers: the mechanism of action and direct target genes.
+ +a. Synergistic enhancers Pou5f3+Nanog+: Pou5f3 and Nanog act as activators. b. Antagonistic enhancers Pou5f3+Nanog-: Pou5f3 acts as an activator, Nanog as a blocker. c. Antagonistic enhancers Pou5f3-Nanog+: Nanog acts as an activator, Pou5f3 as a blocker. Representative early targets (expressed in the wild-type during 2.5-6 hpf time course), and late targets (prematurely expressed in the mutants by blocker TF during 2.5-6 hpf time course) are indicated by schematic early and late embryo drawings. + +<--- Page Split ---> + +## Methods + +KEY RESOURCES TABLE + +
REAGENT or RESOURCESOURCEIDENTIFIER
Antibodies
Anti-Histone H3 (acetyl K27) rabbit, 1/100 dilutionAbcam plc.,
Cambridge, UK		ab 4729
Chemicals, Peptides, and Recombinant Proteins
cOmplete™, EDTA-free Protease Inhibitor CocktailSigma-Aldrich Chemie GmbH, Germany5056489001
SYTOX GreenThermoFisher SCIENTIFICS7020
Critical Commercial Assays
Agencourt® AMPure® XP BeadsBeckmann Coulter, Krefeld, GermanyA63880
Agilent High Sensitivity DNA KitAgilent Technologies, Santa Clara, California, USA5067-4626
Agilent RNA 6000 Nano KitAgilent Technologies, Santa Clara, California, USA5067-1511
Dynabeads® Protein Ginvitrogen Dynal AS, Oslo, Norway10003D
E.Z.N.A® Cycle Pure KitOmega Biotek, Norcross, Georgia, USAD6493-02
Illumina Tagment DNA Enzyme and Buffer Small KitIllumina20034197
Microcon®-30 Centrifugal FiltersMerck Millipore, Darmstadt, GermanyMRCF0R030
mMESSAGE mMACHINE® SP6 transcription KitAmbion10086184
NEBNext Ultra DNA Library Prep Kit for IlluminaNew England Biolabs, Inc., Frankfurt a.M., GermanyE7370S
NEBNext® Multiplex Oligos for Illumina® (Index Primers Set 1)New England Biolabs, Inc., Frankfurt a.M., GermanyE7335S
RNeasy® Mini KitQIAGEN, Hilden, Germany74104
Quant-iT™ PicoGreen® dsDNA Assay Kitinvitrogen™ Molecular Probes® Waltham, Massachusetts, USAQ33120
Deposited Data
Omni ATAC-seq of the wild-type, MZsox19b, MZspg, MZps zebrafish embryos at 3.7 and 4.3 hpfGao et al., 2022GEO: GSE188364
Omni ATAC-seq of the wild-type, MZnanog, MZpn, MZsn and MZtriple zebrafish embryos at 3.0, 3.7 and 4.3 hpfthis workGEO: GSE215956
Omni ATAC-seq and H3K27ac ChIP-seq of the wild-type and MZnpss zebrafish embryos at 4 hpfMiao et al., 2022SRA: SRP355652
Time-resolved RNA-seq at 8 time points in the wild-type, MZsox19b, MZspg, MZnanog, MZps, MZpn, MZsn and MZtriple zebrafish embryosthis workGEO: GSE162415
+ +<--- Page Split ---> + +
ChIP-seq for H3K27ac in the wild-type, MZspg and MZsox19b zebrafish embryos, 4.3 hpfGao et al., 2022GEO: GSE143306
ChIP-seq for H3K27ac in the MZnanog zebrafish embryos, 4.3 hpfthis workGEO: GSE143439
ChIP-seq for Pou5f3 and Sox2 (SoxB1) TF binding, 5.3 hpf.Leichsenring et al., 2013GEO: GSE39780
ChIP-seq for Nanog TF binding, 4.3 hpfXu et al., 2012GEO: GSE34683
MNase-seq of MZsox19b embryos, 4.3 hpfGao et al., 2022GEO: GSE125945
MNase-seq of WT and MZspg embryos, 4.3 hpfVeil et al., 2019GEO: GSE109410
Zebrafish reference genome assembly danner11/GRCz11Genome reference Consortiumhttps://www.ncbi.nlm.nih.gov/grc/zebrafish
Experimental Models: Organisms/Strains
Wild-type zebrafish strain AB/TLZIRCZL1/ZL86
MZspg zebrafishLunde et al., 2004m793
MZnanog zebrafishVeil et al., 2018m1435
MZsox19b zebrafishGao et al., 2022m1434
MZps zebrafishGao et al., 2022m1434/m793
MZpn zebrafishthis workm1435/m793
MZsn zebrafishthis workm1434/m1435
MZtriple zebrafishthis workm1434/m1435/m793
Oligonucleotides
PCR primer for genotyping
Sox19b-f1 5'-ATTTGGGGTGCTTTCTTCAGC-3'		Gao et al., 2022no
PCR primer for genotyping
Sox19b-r1 5'-GTTCTCCTGGGCCATCTTCC-3'		Gao et al., 2022no
PCR primer for genotyping
Nanog1-f1 5'-CAACTTGCGGTCGCTCTAAAC-3'		this workno
PCR primer for genotyping
Nanog1-r1 5'-GCTCAGTCTTGTTGTAGGACAG-3'		this workno
PCR primer for genotyping
spg-f1 5'-GTCGTCTGACTGAACATTTTGC-3'		this workno
PCR primer for genotyping
spg-r1 5'-GCAGTGATTCTGAGGAAGAGGT-3'		this workno
PCR primer for H3K27ac ChIP-seq control, positive reference
tiparp_f_1 5' CGCTCCCAACTCCTAGTATC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, positive reference
tiparp_r_1 5'-AACGCAAGCCAAACGATCTC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, negative reference
igsf2_f_2 5'-GAACTGCATTAGAGACCCAC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, negative reference
igsf2_r_2 5'-CAATCAACTGGGAAAGCATGA-3'
Software and Algorithms
Bed ToolsQuinlan and Hall, 2010BED Tools in usegalaxy.eu
Bowtie2Langmead and Salzberg, 2012Bowtie2 in usegalaxy.eu
DeepTools2Ramirez et al., 2016deepTools in usegalaxy.eu
DESeq2Love et al., 2014DESeq2 in usegalaxy.eu
+ +<--- Page Split ---> + + +
FeatureCountsLiao et al., 2014featureCounts in
usegalaxy.eu
Galaxy serverAfgan et al., 2018https://usegalaxy.eu/
gecece utility to calculate fractional GC content of
nucleic acid sequences		emboss (version
5.0.0)		gecece in
usegalaxy.eu
GREAT: Genomic Regions Enrichment of Annotations
Tool, version 3.0.0		Hiller et al., 2013http://great.stanford.
edu/great/public-
3.0.0/html/
In-vitro nucleosome prediction programKaplan et al., 2009
and
https://github.com/bgr
uening/galaxytools		Nucleosome
Predictions in
usegalaxy.eu
MACS2Ferg et al., 2007MACS2 callpeak and
MACS2 bdgepackall
in
usegalaxy.eu
R: Version 3.6.1R Core Team (2019).
R: A language and
environment for
statistical computing.
R Foundation for
Statistical Computing,
Vienna, Austria.		https://www.R-
project.org/.
RNA StarDobin et al., 2013RNA Star in
usegalaxy.eu
RNA-senseGao et al., 2022https://bioconductor.
org/packages/releas
e/bioc/html/RNAsens
e.html
+ +# RESOURCE AVAILABILITY + +# Lead contact + +Requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de). + +# Materials availability + +Zebrafish lines generated in this study are available from lead contact, Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de) under the restrictions of German Animal Protection law (TierSchutzGesetz). + +# Data and code availability + +·Raw and processed RNA-seq and ATAC-seq data generated in this study have been deposited in GEO and are publicly available as of the date of publication.Accession numbers are listed in the key resources table. + +·The original code is deposited in https://github.com/vandensich/zebrafish-minimodels and will be made available to the date of publication. + +·Any additional information required to reanalyze the data reported in this paper is available from the lead contact Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de) upon request. + +<--- Page Split ---> + +## EXPERIMENTAL MODEL AND SUBJECT DETAILS + +## Zebrafish maintenance and generation of the double and triple mutants + +Zebrafish maintenance and generation of the double and triple mutantsWild- type fish of AB/TL and mutant strains were raised, maintained and crossed under standard conditions as described by Westerfield \(^{34}\) . Embryos were obtained by natural crossing (4 males and 4 females in 1,7 l breeding tanks, Techniplast). Wild- type and mutant embryos from natural crosses were collected in parallel in 10- 15 minutes intervals and raised in egg water at \(28.5^{\circ}C\) until the desired stage. Staging was performed following the Kimmel staging series \(^{35}\) . Stages of the mutant embryos were indirectly determined by observation of wild- type embryos born at the same time and incubated under identical conditions. All experiments were performed in accordance with German Animal Protection Law (TierSchG) and European Convention on the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (Strasburg, 1986). The generation of double and triple mutants for Pou5f3, Sox19b and Nanog was approved by the Ethics Committee for Animal Research of the Koltzov Institute of Developmental Biology RAS, protocol 26 from 14.02.2019. Maternal- Zygotic (MZ) homozygous mutant embryos MZsn were obtained in three subsequent crossings. First, MZnanog \(^{m1435}\) \(^{19}\) homozygous males were crossed with MZsox19b \(^{m143417}\) homozygous females. The double heterozygous fish were raised to sexual maturity and incrossed. The progeny developed into phenotypically normal and fertile adults. Genomic DNA from tail fin biopsies was isolated and used for genotyping. We first selected sox19b homozygous mutants, by PCR- with Sox19b- f1/Sox19b- r1 primers followed by restriction digest with Bbs1 \(^{17}\) . To select the double homozygous fish, we used the genomic DNA from sox19b homozygous mutants for by PCR with Nanog1- f1/Nanog1- r1 primers followed by restriction digest with Ndel. MZsn embryos for experiments were obtained from incrosses of double homozygous sox19b- /- ; nanog- /- fish. The line was maintained by crossing sox19b- /- ; nanog- /- ; males with sox19b- /- ; nanog+/- females. + +MZpn homozygous mutant embryos were obtained in three subsequent crossings. First, MZnanog \(^{m1435}\) homozygous males \(^{19}\) were crossed with MZspg \(^{m793}\) \(^{18}\) homozygous females. The double heterozygous fish were raised to sexual maturity and incrossed. To bypass the early requirement for Pou5f3 in the spg793 homozygous mutants, one- cell stage embryos were microinjected with 50- 100 pg synthetic Pou5f3 mRNA as previously described \(^{18}\) . The fish were raised to sexual maturity, genomic DNA from tail fin biopsies was isolated and used for genotyping. We first selected nanog homozygous mutants, by PCR with Nanog1- f1/Nanog1- r1 primers followed by restriction digest with Ndel. To select the double homozygous fish, we used the genomic DNA from nanog homozygous mutants to PCR- amplify the region, flanking the spg \(^{m793}\) allele. Spg \(^{m793}\) allele carries an A- >G point mutation in the splice acceptor site of the first intron of Pou5f3 gene, which results in the frameshift starting at the beginning of the second exon, prior to the DNA- binding domain. Spg \(^{m793}\) is considered to be null allele. We used the following PCR primers: spg- f1 \(5^{\prime}\) - + +<--- Page Split ---> + +GTCGTCTGACTGAACATTTTGC - 3' and spg-r1 5'- GCAGTGATTCTGAGGAAGAGGT - 3'. Sanger sequencing of the PCR products was performed using commercial service (Sigma). The sequencing traces were examined and the fish carrying A to G mutation were selected. MZpn embryos for experiments were obtained from incrosses of double homozygous spg- /- ; nanog- /- fish. The line was maintained by crossing spg- /- ; nanog- /- males with spg- /- ; nanog+/- females and microinjecting Pou5f3 mRNA in each generation. + +To obtain the triple Maternal- Zygotic (MZ) homozygous mutant embryos MZtriple, MZsn double homozygous males were crossed with MZps double homozygous females 17. The sox19b- /- ; spg+/- ; nanog+/- progeny was incrossed, microinjected with 50- 100 pg synthetic Pou5f3 mRNA and genotyped for spg and nanog as described above. MZtriple embryos for experiments were obtained from incrosses of triple homozygous fish. The line was maintained by crossing sox19b- /- ; spg- /- ; nanog- /- males with sox19b- /- ; spg- /- ; nanog+/- females and microinjecting Pou5f3 mRNA in each generation. + +## Genomic DNA isolation and PCR for genotyping + +Genomic DNA was isolated from individual tail fin biopsies of 3 months old fish. Tail fin biopsies or embryos were lysed in \(50\mu \mathrm{l}\) lysis buffer ( \(10\mathrm{mM}\) Tris pH 8, \(50\mathrm{mM}\) KCl, \(0.3\%\) Tween20, \(0.3\%\) NP- 40, \(1\mathrm{mM}\) EDTA) and incubated at \(98^{\circ}\mathrm{C}\) for \(10\mathrm{min}\) . After cooling down Proteinase K solution ( \(20\mathrm{mg / ml}\) , A3830, AppliChem) was added and incubated overnight at \(55^{\circ}\mathrm{C}\) . The Proteinase K was destroyed by heating up to \(98^{\circ}\mathrm{C}\) for \(10\mathrm{min}\) . The tail fin biopsies material was diluted \(20\times\) with sterile water. \(2\mu \mathrm{l}\) of was used as a template for PCR. PCR was performed in \(25 - 50\mu \mathrm{l}\) volume, using MyTag polymerase (Bioline GmbH, Germany) according to the manufacturer instructions, with 30- 35 amplification cycles. The primer sequences are listed in the key resources table. + +## METHOD DETAILS + +## ATAC-seq and library preparation + +Omni- ATAC- seq was performed as described in 17. Briefly, embryos were enzymatically dechorionated with Pronase E ( \(30\mathrm{mg / ml}\) ). 30 embryos were deyloked manually with an eyebrow needle in Danieaus medium, centrifuged ( \(500\times \mathrm{g}\) , \(5\mathrm{min}\) , \(4^{\circ}\mathrm{C}\) ), washed with \(100\mu \mathrm{l}\) ice- cold PBS, and then centrifuged again. The cells were resuspended by pipetting in \(50\mu \mathrm{l}\) ice- cold lysis buffer (ATAC- RSB ( \(10\mathrm{mM}\) TrisHCl pH 7.4, \(10\mathrm{mM}\) NaCl, \(3\mathrm{mM}\) MgCl2), with \(0.1\%\) NP40, \(0.1\%\) Tween20, and \(0.01\%\) digitonin) and incubated on ice for 5 minutes. Then, the lysate was diluted with ice- cold dilution buffer ( \(1\mathrm{ml}\) , ATAC- RSB containing \(0.1\%\) Tween- 20). Nuclei were pelleted ( \(500\times \mathrm{g}\) , \(10\mathrm{min}\) , \(4^{\circ}\mathrm{C}\) ), the supernatant carefully removed, and the nuclei resuspended in \(16.5\mu \mathrm{l}\) PBST. Tagnentation reaction was assembled according to the manufacturer instructions for Illumina small Tn5 buffer and enzyme kit in \(50\mu \mathrm{l}\) volume and incubated in a thermomixer for \(30\mathrm{min}\) ( \(37^{\circ}\mathrm{C}\) , \(800\mathrm{rpm}\) ). The sample was purified using Qiagen PCR MinElute Purification kit. The purified transposed mix was pre- amplified for 5 cycles ( \(72^{\circ}\mathrm{C}\) for \(5\mathrm{min}\) , \(98^{\circ}\mathrm{C}\) for \(30\mathrm{sec}\) , 5 cycles of \(98^{\circ}\mathrm{C}\) for \(10\mathrm{sec}\) , \(63^{\circ}\mathrm{C}\) for \(30\mathrm{sec}\) , + +<--- Page Split ---> + +\(72^{\circ}C\) for 1 min) using NEBNext 2x Master Mix (NEB) and Nextera adapters and put on ice. The number of additional cycles was determined using \(5\mu l\) of the pre- amplified mixture as in \(^{36}\) . The total number of PCR cycles was 10- 12. The amplified libraries were purified using SPRI beads (Beckmann) according to the manufacturer instructions. ATAC- seq was performed at three developmental stages, 3 hpf (1k), 3.7 hpf (oblong) and 4.3 hpf (dome). At 3 hpf, three biological replicates for WT and MZtriple were used. At 3.7 hpf, two biological replicates for MZnanog and MZpn and three biological replicates for MZtriple were used. At 4.3 hpf, three biological replicates for MZsn and MZtriple and two for MZpn and MZnanog were used. Paired- end 150 bp sequencing of resulting 23 ATAC- seq libraries was performed on NovaSeq6000 (Illumina) by Novogene company, with the sequencing depth 80 million reads per library. The raw and processed ATAC- seq data are deposited in GEO (NCBI), accession number GSE215956. + +## ATAC-seq data processing + +ATAC- seq data processingATAC- seq data processing was performed as in (Gao et al., 2022) on the european Galaxy server \(^{37}\) . Adapters were removed, the reads were cropped to 30 bp from start, the reads with average quality less than 30 and length less than 25 bp were removed using Trimmomatics. The trimmed reads were aligned to the danner11/ GRCz11 genome assembly without alternative contigs using Bowtie2 \(^{38}\) , with the parameters: '—dovetail, --very- sensitive, --no- unal'. The aligned reads were filtered using BamTools Filter \(^{39}\) with the parameters - isProperPair true, - mapQuality \(>=30\) , - isDuplicate false, - reference lchrM. The aligned reads were filtered further using the alignmentsieve tool from the deepTools2 suite \(^{40}\) , with the parameters —ATACshift, --minMappingQuality 1, --maxFragmentLength 110. Thus, only fragments with a size of 110 bp or less were kept. The filtering steps were performed for the individual replicates as well as and for the merged biological replicates BAM files. Bigwig files for each stage and genotype were obtained from merged BAM files using Bamcoverage and used for data visualization in deepTools2. Bigwig and peak files for each stage and genotype are included in GEO GSE215956 submission as processed files. + +## Definition of regulated ARs and TdARs + +The ATAC- seq data from this work at 3.7 hpf and 4.3 hpf were processed together with our previously published ATAC- seq data set: three biological replicates for the wild- type and two biological replicates for MZspg, MZsox19b and MZps at 3.7 hpf, three biological replicates for the wild- type, MZspg and MZsox19b and four for MZps at 4.3 hpf \(^{17}\) , GEO accession number GSE188364. For seven wild- type biological replicates, filtered reads were converted from BAM to BED using Bedtools \(^{41}\) and were used as inputs for peak calling. Peaks were called from individual replicates and merged replicates using MACS2 callpeak \(^{42}\) with the additional parameters '--format BED, --nomodel, --extsize 200, --shift - 100' and a 5% FDR cutoff. Regions mapping to unassembled contigs were excluded. Only peaks that were overlapping in merged and each of the single replicates were kept. ATAC- seq peaks overlapping between 3.7 and 4.3 hpf stages in the wild- type were centered on ATAC- seq peak summits (4.3 hpf) and cut to 110 bp length. This set of 102 965 regions accessible in the wild- type (ARs) + +<--- Page Split ---> + +was used in all subsequent analyses. The coverage of ATAC- seq reads on ARs was scored using MultiCovBed (Bedtools) in all ATAC- seq experiments (3.7 and 4.3 hpf). Deseq2 \(^{43}\) was used to normalize the reads and compare chromatin accessibility in each mutant to the wild- type. For each mutant genotype, we made four Deseq2 cross- normalizations: across all replicates, across the replicates of 3.7 hpf or 4.3 hpf separately, and across three samples (i.e. wild- type, MZnanog and MZtriple). The region was scored as downregulated in the mutant, if the fold change to the wild- type was negative with FDR \(5\%\) in four Deseq2 normalizations. The region was scored as upregulated in the mutant, if the fold change to the wild- type was positive with FDR \(5\%\) in four Deseq2 normalizations. The remaining ARs were considered unchanged. To define TdARs, (TF- bound Accessible Regions, downregulated in MZtriple), we selected ARs, downregulated in MZtriple and overlapping with any of Sox2, Pou5f3 \(^{12}\) or Nanog \(^{23}\) ChIP- seq peaks for at least 1 bp. The genomic coordinates of TF- binding peaks remapped to danrer11/GRCz11 assembly were taken from \(^{17}\) . Genomic coordinates of ARs and their properties are provided in the Table S1. + +## Scoring the motif enrichment + +Sequence- specific nanog1 and nanog2 motif matrices were taken from \(^{22}\) , the rest of the motif matrices from \(^{17}\) . Genomic coordinates of all occurrences for each motif in the 3 kb regions around ATAC- seq peaks were obtained using FIMO \(^{44}\) with p- value threshold \(10^{- 4}\) and converted to BED files. To calculate the background frequency of the motifs, we used the background control peak file, where the genomic coordinates of all ATAC- seq summits were shifted 1 kb downstream. The motif densities were calculated as the average number of motifs overlapping with 60 bp around the ATAC- seq or control peak summits. The log2 ratio of ATAC- seq to background motif densities was taken as a motif enrichment value. + +## GC content and in-vitro nucleosome predictions + +GC content for 110 bp around AR summits was calculated with geece program, in- vitro nucleosome prediction with the algorithm of Kaplan et al. \(^{33}\) in the european Galaxy instance usegalaxy.org. The genomic coordinates of ARs and control peaks were extended to +/- 5 kb to account for the edge effects, and in- vitro nucleosome prediction value was derived for every base pair. Nucleosome predictions were converted to bedgraph files for scoring. GC content and nucleosome prediction values per 110 bp around ARs are provided in the Supplementary Table 1. + +## Scoring the rescue of chromatin accessibility + +Track hub with Omni- ATAC rescue processed data in UCSC bigwig format was downloaded from the home page of Antonio Giraldez lab (https://www.giraldezlab.org/data/miao_et_al_2022_molecular_cell/) and used for scoring. We calculated ATAC- seq signals in all conditions for 110 bp around ARs. For ARs, downregulated in MZnpcs compared to the WT B1 replicate in \(^{5}\) , we estimated the rescue. We considered AR as rescued by TF(s) microinjection, if the log2 of ATAC- seq signals (injected MZnpcs/control MZnpcs) was exceeding log2 of ATAC- seq signals + +<--- Page Split ---> + +(control WT B1/injected MZnps). The rescue status by all combinations of TFs for ARs downregulated in MZnps is provided in the Table S1; the statistics for rescue on TdARs is provided on the Fig. S3. + +## ChIP-seq for H3K27ac in MZnanog + +The ChIP- seq was performed as in 17. Embryos were obtained from natural crossings in mass- crossing cages (4 males + 4 females). The freshly laid eggs of MZnanog mutants and wild- type were collected in 10- 15 min intervals. The embryos were incubated at \(28.5^{\circ}C\) and enzymatically dehorionated with \(30mg / ml\) Pronase shortly before 4.3 hpf. Embryos were homogenized in \(10ml0.5\%\) Danieau's (for 1L of 30X stock: \(1740mMNaCl\) , \(21mM KCl\) , \(12mM MgSO4\) , \(18mM Ca(NO3)2\) , \(150mM HEPES\) buffer, pH 7.6) containing 1x protease inhibitor cocktail (PIC, Roche) using a Dounce tissue grinder, and immediately treated with \(1\%\) (v/v) Methanol- free Formaldehyde (Pierce) for exactly 10 min shaking at room temperature. The fixation was stopped with 0.125 M Glycine by shaking for 5 min on a rotating platform. Lysate was centrifuged for 5 min, \(4700rpm\) at \(4^{\circ}C\) , the pellet was resolved in cell lysis buffer ( \(10mM\) Tris- HCl (pH 7.5), \(10mMNaCl\) , \(0.5\%\) NP- 40, 1- 4 ml/1000 embryos), followed by 5 min incubation on ice, and centrifuged 1 min, \(2700g\) at \(4^{\circ}C\) . The pellet was washed 2 times with 1 ml ice cold 1x PBST (for 1 L: \(40ml\) PO4 buffer (0.5 M), \(8gNaCl\) , \(0.2gKCl\) , \(0.1\%\) Tween20, pH 7.5) and resuspended in 1 ml PBST. 5 \(\mu l\) aliquote of nuclei suspension was stained with Sytox green, nuclei were scored under fluorescence microscope using the Neubauer counting chamber. The residual nuclei were pelleted by 1 min centrifugation at \(2700g\) at \(4^{\circ}C\) , snap frozen in liquid nitrogen and stored at - \(80^{\circ}C\) . The nuclei collected in different days were pooled together to reach the total number of 2.5- 3 million, which was used to start one ChIP experiment. + +The chromatin was thawn, pooled in 2 ml of nuclei lysis buffer ( \(50mM\) Tris- HCl (pH 7.5), \(10mM\) EDTA, \(1\%\) SDS) and incubated 1 h on ice. In order to shear the chromatin to 200 bp fragments (on average), the chromatin was sonicated using the Covaris S2 sonicator (DC \(20\%\) , Intensity 5, Cycles of Burst 200, Time \(= 3^{*}40\) cycles with 30 sec each ( \(3^{*}20\) min)). To control the sonication, \(30\mu l\) of the sheared chromatin was de- crosslinked with \(250mMNaCl\) overnight at \(65^{\circ}C\) and then analyzed using the Agilent Expert 2100 Bioanalyzer® and Agilent high sensitivity DNA Chip kit. + +The lysed and sheared samples were centrifuged for 10 min, \(4^{\circ}C\) at max. speed in table top centrifuge. \(60\mu l\) of each sample were kept as input control. The chromatin was then concentrated to \(100\mu l\) using the Microcon centrifugal filters (Merck Millipore MRCF0R030) and diluted 1:3 by adding ChIP dilution buffer ( \(16.7mM\) Tris- HCl pH 7.5, \(167.0mMNaCl\) , \(1.2mM\) EDTA) containing protease inhibitors. The H3K27ac antibody was added and incubated overnight at \(4^{\circ}C\) on a rotating wheel. \(150\mu l\) of magnetic Dynabeads coupled to protein G (Stock \(30mg / ml\) ; invitrogen Dynal 10003D) were transferred into a 2 ml eppendorf tube and placed on a magnetic rack. The beads were washed 3x with \(5mg / ml\) specially purified BSA in PBST and 1x with \(500\mu l\) ChIP dilution buffer. After removing the ChIP dilution buffer, the chromatin- antibody mix was added and incubated with the beads at \(4^{\circ}C\) overnight on a rotating wheel. Beads were pulled down by placing the eppendorf tubes on the magnetic rack. The beads were + +<--- Page Split ---> + +resuspended in \(333\mu \mathrm{l}\) RIPA buffer ( \(10\mathrm{mM}\) Tris- HCl (pH 7.6), \(1\mathrm{mM}\) EDTA, \(1\%\) NP- 40, \(0.7\%\) sodium deoxycholate, \(0.5\mathrm{M}\) LiCl) containing PIC. The Protein G- antibodychromatin complex was washed \(4\times 5\) min on a rotating platform with \(1\mathrm{ml}\) of RIPA buffer, followed by \(1\times 1\) ml TBST buffer ( \(25\mathrm{mM}\) Tris- HCl, \(150\mathrm{mM}\) NaCl, \(0.05\%\) Tween 20, pH 7.6). The beads were pulled down again. To elute the chromatin, \(260\mu \mathrm{l}\) elution buffer ( \(0.1\mathrm{M}\) NaHCO3, \(1\%\) SDS) was added and incubated for \(1\mathrm{h}\) at \(65^{\circ}\mathrm{C}\) in a water bath. The samples were vortexed every 10 - 15 min. The beads were pulled down, supernatant was transferred to a fresh eppendorf tube. \(12.5\mu \mathrm{l}5\mathrm{M}\) NaCl was added to de- crosslink the chromatin and incubated overnight at \(65^{\circ}\mathrm{C}\) in a water bath. The input samples were treated in parallel ( \(230\mu \mathrm{l}\) elution buffer per \(30\mu \mathrm{l}\) input). Purification of the de- crosslinked chromatin was performed using the QIAquick PCR Purification Kit from Qiagen. The concentration was determined using the Qubit fluorometer and Quanti- iT™ PicroGreen® dsDNA Kit according to manufacturer instructions. + +To estimate the signal to background ratio in each ChIP experiment, we used the positive and negative reference genomic regions, near tiparp and igsf2 genes, respectively enriched in or devoid of H3K27ac (Gao et al., 2022). We performed quantitative PCR in ChIP and Input control material, using the primers for these regions. PCR primers used were: tiparp_f_1 5' CGCTCCCAACTCCATGTATC- 3', tiparp_r_1 5'- AACGCAAGCCAAACGATCTC- 3', igsf2_f_2 5'- GAACTGCATTAGAGACCCAC- 3', igsf2_r_2 5'- CAATCAACTGGGAAAGCATGA- 3'. QPCR was carried out using the SsoAdvanced™ Universal SYBR® Green Supermix from BIO- RAD. ChIP and input were normalized by ddCT method, using negative reference region (igsf2). The ChIP experiment was considered successful, if the enrichment in ChIP over input control on the positive reference region (tiparp) was more than 5- fold. + +The library preparation was carried out with NEBNext® Ultra™ DNA Library Prep Kit according to manufacturer instructions, with the modifications indicated below. As the DNA outcome of individual MZnanog K27ac ChIP experiments did not reach the input DNA limit for the kit (5 ng), we pooled together the material from two successful ChIP experiments, as well as corresponding inputs. Since the DNA input was \(< 100\) ng, in the adaptor ligation step, the NEBNext Adaptor for Illumina® ( \(15\mu \mathrm{M}\) ) was diluted 10- fold in \(10\mathrm{mM}\) Tris- HCl (pH 7.8) to a final concentration of \(1.5\mu \mathrm{M}\) and used immediately. At the final clean- up step, the reaction was purified by mixing the samples with AMPure XP Beads ( \(45\mu \mathrm{l}\) ) before incubating at room temperature for 5 min. After a quick spin using the tabletop centrifuge, samples were placed on a magnetic rack and supernatant was discarded. \(200\mu \mathrm{l}\) of \(80\%\) Ethanol were added and removed after 30 sec, two times. The beads were air dried for 3 min and the DNA target was eluted by adding \(33\mu \mathrm{l}\) of \(0.1\times \mathrm{TE}\) (pH 8) and incubating at room temperature for 2 min. \(28\mu \mathrm{l}\) of the library were transferred to a fresh PCR tube and stored at - 20 °C. \(2\mu \mathrm{l}\) of the sample were diluted 5- fold with \(0.1\times \mathrm{TE}\) and used to check the size distribution of the library using Agilent Expert 2100 Bioanalyzer® and Agilent high sensitivity DNA Chip kit. The ChIP- seq library was sequenced at \(70\mathrm{mln}\) paired end 150bp reads the input library were sequenced to \(30\mathrm{mln}\) reads. Sequencing was performed by the Novogene + +<--- Page Split ---> + +company (China). The raw and processed data have been deposited in GEO under the number GSE143439. + +## Definition of regulated enhancers + +Sequenced reads were mapped to the zebrafish danrer11 assembly using Bowtie2 in the european Galaxy server. Alternate loci and unassembled scaffolds were excluded. H3K27ac data for MZnanog from this work was further processed together with H3K27ac for the wild- type, MZspg and MZsox19b17. To create bigwig files for scoring, the log2 ratio between each ChIP and merged inputs (in rpkm) was obtained using BAMcompare program in deepTools, bin size =10. The mean H3K27ac signal (log2 ChIP/input per bin) around each AR was calculated for 1 kb around AR summit. Values smaller than 0.1 were rounded to 0.1. We considered AR as H3K27- acetylated, if the mean H3K27ac signal was higher than arbitrary threshold 0.2. All H3K27- acetylated TdARs were considered as enhancers. Enhancer was considered as activated by TF, if log2 (mutant by TF/WT) ratio of H3K27ac signals was less than arbitrary threshold - 0.7, was considered as blocked by TF, if log2 (mutant by TF/WT) ratio of H3K27ac signals was more than 0.7, and was considered as unchanged by TF all other cases. Enhancers activated, blocked or unchanged by the sum of TFs were assigned by applying the same procedure and thresholds to H3K27ac ChIP- seq data on the WT and MZnps5. H3K27ac status of all ARs is provided in the Supplementary Table 1. + +## RNA-seq time curves + +RNA- seq time curves were obtained as described previously (Gao et al., 2022). Freshly laid eggs were obtained from natural crossings in mass- crossing cages (4 males + 4 females). 5- 10 cages were set up per genotype, the eggs from different cages were pooled. At least 600 freshly laid eggs per each genotype collected within 10- 15 minutes were taken for single experiment. To match the developmental curves as precisely as possible, the material was simultaneously collected for two genotypes in parallel. 45- 60 minutes after the egg collection, we ensured that the embryos of both genotypes are at 2- 4 cell stage, removed non- fertilized eggs and distributed the embryos to 8 dishes per genotype, 40- 45 embryos per dish, to obtain 8 time points per genotype: 2.5, 3, 3.5, 4, 4.5, 5, 5.5 and 6 hpf. The temperature was kept at 28.5°C throughout the experiment. The embryos from one dish per genotype were snap- frozen in liquid nitrogen every 30 minutes, starting from 2.5 hpf (pre- ZGA, 256 cell stage, 8th cell cycle) till midgastrula (6 hpf). Two biological replicates for each of MZnanog and MZsn were previously collected in parallel to the wild- type replicates B1 and B2 from (Gao et al., 2022). In addition, we collected six biological replicates of the time curve for the wild- type, and three biological replicates for MZpn and MZtriple mutants from six independent experiments. Total RNA was isolated with QIAGEN RNeasy kit following the user's manual. We checked RNA quantity and quality by using Agilent RNA 6000 nano Kit on Agilent Bioanalyzer, according to manufacturer instructions. Poly- A enriched library preparation and sequencing on Illumina platform was performed by Novogene company (China). The MZnanog and MZsn samples were single- end sequenced at 35 million reads per sample (50 bp), similarly to the previously published samples WT, MZsox19b, MZspg and MZps (Gao et al., 2022). The MZpn + +<--- Page Split ---> + +and MZtriple samples and corresponding six replicates of the wild- type were paired- end sequenced at 58 million reads per sample (150 bp). + +## RNA-seq time curves: data processing + +FASTQ files were processed on european Galaxy server usegalaxy.eu. All sequenced clean reads data were trimmed on 5' end for 5 bp by Trim Galore! program according to the sequencing quality. Trimmed reads were mapped to danner11 genome assembly using RNA STAR \(^{45}\) and ENSEMBL gene models. Number of reads for each gene was counted using Feature count \(^{46}\) . Single- end and paired- end sequencing data sets were further processed separately as the part 1 and the part 2. For the part 1, feature counts for the MZnanog and MZsn samples were cross- normalized with MZsox19b, MZspg and MZps and four biological replicates of the wild- type (Gao et al., 2022) using Deseq2. For the part 2, feature counts for the MZpn and MZtriple samples and corresponding six replicates of the wild- type were cross- normalized using Deseq2. The raw data and two normalized data tables were deposited in GEO with the accession number GSE162415. + +## Finding differentially expressed zygotic genes + +We used 3- step RNA- sense program to process normalized RNA- seq time curves as described in \(^{17}\) . Shortly, RNA- sense takes RNA- seq time curves in two conditions (WT and mutant) as an input. The transcripts expressed below a user- defined threshold are excluded from the analysis. In the first step, RNA- sense creates the lists of transcripts switching UP or DOWN in time, in one of the genotypes; the genotype and switch p- value are defined by user. We found that 4247 genes are switching UP in the wild- type, in both part1 and part 2 of the analysis (thresholds \(>100\) for expression and \(< 0.15\) for switch p- value). Switching UP transcripts were considered as zygotically expressed \(^{17}\) . 1668 transcripts were switching up in at least one mutant but not in the wild- type; we considered them as zygotically expressed in the respective mutants. The regulation status was assigned to each zygotic transcript on the second step of RNA- sense. Zygotic transcript was considered to be downregulated in the mutant, if it was expressed at least two- fold lower compared to the wild- type with p- value \(< 0.05\) in Student's t- test in at least one time point at or after the switch UP. Zygotic transcript was considered to be upregulated in the mutant, if it was expressed at least two- fold higher compared to the wild- type with p- value \(< 0.05\) in Student's t- test in at least one time point. Some transcripts were up – and down regulated in the same mutant in different time points and were considered as “undefined” (“U”). The list of zygotic genes, their switch and regulation status in all the mutants is provided in the Table S2. + +## Enrichment statistics + +Zygotic transcripts (Table S2) were linked to Chromatin Accessible Regions (ARs, Table S1) within +/- 50 kb from Transcription Start Site (TSS). Resulting 70419 AR- transcript pairs were uniquely numbered (Table S3). To estimate the over- or underrepresentation of different types of enhancers in the putative regulatory regions of zygotic genes, Chi- squared ( \(\chi^{2}\) ) test was performed using chisq.test function and visualized using corrplot function in R. We considered only strong positive correlations + +<--- Page Split ---> + +with Pearson residuals more than 4. To reproduce our results, the input files for all \(\chi^2\) tests can be easily derived from the supplementary table 3 as explained below. The input file is a tab- separated text file with three columns: "AR- transcript_pair" (which is always column A in the Table S3), "AR", and "transcript". To create the input file for \(\chi^2\) test, delete the first two rows of the Supplementary Table 3, keep column A and two columns specified below (delete all other columns), delete the rows specified below, change the headings to AR" and "transcript" and export the input file as tab- separated text. Fig. 2G: "AR" - column N, "transcript" - column AF, delete all rows with "U". Fig. 5D- F: "AR" - columns O, Q, R, respectively, delete all rows with "none"; "transcript" - column AG, delete all rows with "N/A". Fig. 6C: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column Z, delete all rows with "U". Fig. 6C: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column AB, delete all rows with "U". Fig. S7B, C: "AR" - columns G, P, respectively, delete all rows with "none"; "transcript" - column AG, delete all rows with "N/A". Fig. S7F: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column AC, delete all rows with "U". + +## Gene Ontology analysis. + +For the genomic regions we used GREAT analysis 47 at http://great.stanford.edu/great/public- 3.0.0/html/. Genomic regions were converted to zv9/danrer9 genomic assembly, which is the only assembly available for this server, using Liftover Utility from UCSC and associated with genes using 20 kb single nearest gene association rule. The categories were ranked by ascending FDR value. For the transcript groups we used DAVID 48. + +## Data visualization + +Omni- ATAC- seq: to create bigwig files for data visualization, Omni- ATAC- seq signal (in rpkm) was obtained using BAMcoverage program in deepTools2, bin size \(= 10\) . H3K27ac: to create bigwig files for data visualization, the log2 ratio between each ChIP and merged inputs (in rpkm) was obtained using BAMcompare program in DeepTools, bin size \(= 10\) . The heatmaps or mean profiles were plotted using plotheatmap or plotprofile programs in DeepTools. H3K27ac, TF ChIP- seq and omni- ATAC- seq profiles in single genes were visualized in UCSC browser. RNA- seq heatmaps were plotted using R with custom script. For the heatmaps, the transcripts were sorted by ascending RNA- sense switch p- value then by ascending RNA- sense switch time then by group. Biological replicates from all experiments were cross- normalized using Deseq2. Replicates for each genotype were averaged and the maximal expression value across the time curve in all genotypes was calculated. Expression/max ratio for each timepoint was plotted. + +## Mathematical Modeling + +The mathematical modeling process was performed in two main steps, (i) the core model and (ii) the mini models. The idea behind developing a core model was to use the resulting dynamics of relevant transcription factors or their combinations as input functions to the mini models that later describe the dynamical behavior of a huge amount of 1799 preselected target genes. Generation of ODE models, parameter + +<--- Page Split ---> + +estimation and identifiability analysis was performed by means of the R package dMod \(^{49}\) . For each model, the best parameter set was obtained by the method of maximum likelihood performing a deterministic multi- start optimization \(^{50}\) from multiple randomly chosen starting points in parameter space. The trust region optimizer trust (Geyer CJ (2004) trust: Trust Region Optimization. \(R\) package Version 0.1- 8) was used constituting the standard optimizer in the R package dMod. Model parameters were log- transformed to ensure positivity and enable optimization over a broad range of magnitudes \(^{50}\) . Parameter values were not restricted by fixed borders. Instead, in order to prevent the optimizer from finding solutions with very low or high parameter values, we constrained the model parameters with a weak \(L_{2}\) prior that contributed with one to the likelihood, if the parameter differed by three orders of magnitude from a value of zero on the logarithmic scale. For the computation of confidence intervals and to test identifiability of the model's parameters by means of the profile likelihood \(^{51}\) the contribution of these \(L_{2}\) priors was subtracted to ensure a purely data- based result. + +## Construction of the core model + +In total, the core model consists of 11 model states and 17 reactions (Fig. S5A, Table S4). Kinetic expressions of the model reactions were derived based on state- of- the- art transcription factor network modeling \(^{52}\) and prior knowledge from published literature \(^{14,17,19,20}\) . For simplicity, the degradation of Sox19b was incorporated via a switch- like input function dependent on the time t as a parameter. Model states were assumed to be directly observed, and uncertainty of the observations were estimated jointly with the kinetic parameters assuming an absolute error model for the measurements of each observed state. Model equations, as summarized in Table S4, include three boolean variables MZspg, MZnanog and MZsox that were used to distinguish between the seven experimental conditions including wild type, single, double and triple knockdowns. In total, a number of 888 RNA- seq data points for six targets was used for model calibration. The first time point of the data set was at 2.5 hpf. As an assumption, the initial values of the ODE model were defined at 2 hpf. The initial values of downstream targets Sox2, Sox3 and Sox19a were assumed to be zero, while for the values of the zygotic transcription factors Pou5f3, Sox19b and Nanog a free parameter was estimated by the model, with the constraint of being the same over all experimental conditions. + +During the model development, a couple of parameters was fixed to zero or one, following likelihood- ratio test based model reduction \(^{53}\) since they were not necessary in order to describe the available data. Hill- kinetic parameters were fixed to a value of five corresponding to a strong threshold behavior. Further parameter transformations that have been applied are summarized in https://github.com/vandensich/zebrafish- minimodels. Parameter estimation was performed using deterministic multi- start optimization \(^{50}\) starting from 192 randomly chosen sampled positions in parameter space. Parameter values of the best obtained fit are summarized in Table S5. To show reliability of the optimization, the 100 best optimization runs were displayed as a waterfall plot (Fig. S5B) sorted by their negative log- likelihood values \(^{50}\) . The global + +<--- Page Split ---> + +optimum was found in 38 of the 101 converged optimizations runs. Considering a confidence level of \(95\%\) , profile likelihoods showed finite confidence intervals for 30 out of 37 parameters (see Fig. S6A). + +## Construction of the mini models + +Assuming at least one of Pou5f3, Nanog and SOX to be activating, a set of 19 mini models was constructed. Model equations for each mini model (Table S6) were derived from the theory of transcription factor network modeling \(^{52}\) . When more than one transcription factor was either activating or repressing, the combination was modeled using appropriate logical AND- gate functions \(^{53}\) . Dynamics of the proteins Pou5f3, Nanog and the sum of SoxB1 class genes SOX were taken from the core model for the different experimental conditions and used as input functions for the mini models. To avoid numerical issues with infinite derivatives of hill- kinetic terms, a small number of 1e- 4 was added to the input functions where necessary. + +Analogously to the core model, RNA- seq data of seven experimental conditions was used for model fitting of each mini model and for each target gene. Per target gene, a total number of 192 was available. In contrast to the core model, the different experimental conditions were not incorporated via boolean variables, instead the differential equations stayed the same over all conditions. Thus, differences in dynamics were only possible via the differences in the given input dynamics of the core model. + +For each mini model and for each target gene, a total number of 64 optimization runs was performed out of which \(99.97\%\) converged and at least \(98.67\%\) reached the global optimum. Finally, using the Bayesian Information criterion (BIC), out of the 19 analyzed mini models, the best fitting one was selected for every target. Parameter values of the selected models are summarized in Table S7. + +## QUANTIFICATION AND STATISTICAL ANALYSIS + +Statistical comparisons of three or more samples were performed using one- way ANOVA, followed by Tukey- Kramer Test to evaluate the statistical significance of pairwise differences. Two samples were compared using Student 2- tailed t- test. + +<--- Page Split ---> + +## Supplemental items. + +Supplemental figures and legends.pdf + +Table S1. Accessible_regions.xlsx Related to main Fig. 1, Fig. 2, Fig. 3, Fig. 4, Fig. S3, Fig. S4, the legend in the file + +Table S2. Zygotic_transcripts.xlsx Related to main Fig. 2, Fig. 5, Fig. 6, Fig. S2, Fig. S6, Fig. S7, the legend in the file. + +Table S3. ARs_linked_to_transcripts.xlsx Related to main Fig. 2, Fig. 5, Fig. 6, Fig. 7, Fig. S7, the legend in the file. + +Table S4. Coremodel_equations.csv Related to Fig.S5 and Fig.S6 + +Table S5. Coremodel_Bestfitparameters.csv. Related to Fig.S5 and Fig.S6. + +Table S6. Minimodel_equations.csv Related to main Fig. 5, Fig. S6 and Fig. S7. + +Table S7. Minimodel_parameters.csv Related to main Fig. 5, Fig. S6 and Fig. S7 + +<--- Page Split ---> + +## References + +1 Vastenhouw, N. L., Cao, W. X. & Lipshitz, H. D. The maternal- to- zygotic transition revisited. Development 146, doi:10.1242/dev.161471 (2019).2 Schulz, K. N. & Harrison, M. M. Mechanisms regulating zygotic genome activation. Nature Reviews Genetics, doi:10.1038/s41576- 018- 0087- x (2018).3 Iwafuchi- Doi, M. & Zaret, K. S. Cell fate control by pioneer transcription factors. Development 143, 1833- 1837, doi:10.1242/dev.133900 (2016).4 Gaskill, M. M., Gibson, T. J., Larson, E. D. & Harrison, M. M. GAF is essential for zygotic genome activation and chromatin accessibility in the early Drosophila embryo. Elife 10, doi:10.7554/eLife.66668 (2021).5 Miao, L. et al. The landscape of pioneer factor activity reveals the mechanisms of chromatin reprogramming and genome activation. Mol Cell 82, 986- 1002 e1009, doi:10.1016/j.molcel.2022.01.024 (2022).6 Fernandez Garcia, M. et al. Structural Features of Transcription Factors Associating with Nucleosome Binding. Mol Cell 75, 921- 932 e926, doi:10.1016/j.molcel.2019.06.009 (2019).7 Gassler, J. et al. Zygotic genome activation by the totipotency pioneer factor Nr5a2. Science, eabn7478, doi:10.1126/science. abn7478 (2022).8 Heyn, P. et al. The earliest transcribed zygotic genes are short, newly evolved, and different across species. Cell reports 6, 285- 292, doi:10.1016/j.celrep.2013.12.030 (2014).9 Neyfakh, A. A. Radiation Investigation of Nucleo- Cytoplasmic Interrelations in Morphogenesis and Biochemical Differentiation. Nature 201, 880- 884 (1964).10 White, R. J. et al. A high- resolution mRNA expression time course of embryonic development in zebrafish. Elife 6, doi:10.7554/eLife.30860 (2017).11 Lee, M. T. et al. Nanog, Pou5f1 and SoxB1 activate zygotic gene expression during the maternal- to- zygotic transition. Nature, 360- 364., doi:10.1038/nature12632 (2013).12 Leichsenring, M., Maes, J., Mossner, R., Driever, W. & Onichtchouk, D. Pou5f1 transcription factor controls zygotic gene activation in vertebrates. Science 341, 1005- 1009, doi:10.1126/science.1242527 (2013).13 Onichtchouk, D. & Driever, W. Zygotic Genome Activators, Developmental Timing, and Pluripotency. Current topics in developmental biology 116, 273- 297, doi:10.1016/bs.ctdb.2015.12.004 (2016).14 Okuda, Y., Ogura, E., Kondoh, H. & Kamachi, Y. B1 SOX coordinate cell specification with patterning and morphogenesis in the early zebrafish embryo. PLoS Genetics 6, e1000936, doi:10.1371/journal.pgen.1000936 (2010).15 Boyer, L. A. et al. Core transcriptional regulatory circuitry in human embryonic stem cells. Cell. 122, 947- 956. (2005).16 Friman, E. T. et al. Dynamic regulation of chromatin accessibility by pluripotency transcription factors across the cell cycle. Elife 8, doi:10.7554/eLife.50087 (2019).17 Gao, M. et al. Pluripotency factors determine gene expression repertoire at zygotic genome activation. Nat Commun 13, 788, doi:10.1038/s41467- 022- 28434- 1 (2022).18 Lunde, K., Belting, H. G. & Driever, W. Zebrafish pou5f1/pou2, homolog of mammalian Oct4, functions in the endoderm specification cascade. Current biology : CB 14, 48- 55 (2004).19 Veil, M. et al. Maternal Nanog is required for zebrafish embryo architecture and for cell viability during gastrulation. Development 145, doi:10.1242/dev.155366 (2018). + +<--- Page Split ---> + +20 Onichtchouk, D. et al. Zebrafish Pou5f1- dependent transcriptional networks in temporal control of early development. Mol Syst Biol 6, 354, doi:msb20109 [pii] 10.1038/msb.2010.9 (2010). 21 Palfy, M., Schulze, G., Valen, E. & Vastenhouw, N. L. Chromatin accessibility established by Pou5f3, Sox19b and Nanog primes genes for activity during zebrafish genome activation. PLoS Genet 16, e1008546, doi:10.1371/journal.pgen.1008546 (2020). 22 Veil, M., Yampolsky, L., Gruening, B. & Onichtchouk, D. Pou5f3, SoxB1, and Nanog remodel chromatin on High Nucleosome Affinity Regions at Zygotic Genome Activation. Genome Res, doi:10.1101/gr.240572.118 (2019). 23 Xu, C. et al. Nanog- like Regulates Endoderm Formation through the Mxtx2- Nodal Pathway. Developmental cell 22, 625- 638, doi:10.1016/j.devcel.2012.01.003 (2012). 24 Donaghey, J. et al. Genetic determinants and epigenetic effects of pioneer- factor occupancy. Nature Genetics 50, 250- 258, doi:10.1038/s41588- 017- 0034- 3 (2018). 25 Segert, J. A., Gisselbrecht, S. S. & Bulyk, M. L. Transcriptional Silencers: Driving Gene Expression with the Brakes On. Trends Genet 37, 514- 527, doi:10.1016/j.tig.2021.02.002 (2021). 26 Mulas, C. et al. Oct4 regulates the embryonic axis and coordinates exit from pluripotency and germ layer specification in the mouse embryo. Development 145, doi:10.1242/dev.159103 (2018). 27 Niwa, H., Miyazaki, J. & Smith, A. G. Quantitative expression of Oct- 3/4 defines differentiation, dedifferentiation or self- renewal of ES cells. Nat Genet 24, 372- 376, doi:10.1038/74199 (2000). 28 Radzisheuskaya, A. & Silva, J. C. Do all roads lead to Oct4? the emerging concepts of induced pluripotency. Trends in cell biology 24, 275- 284, doi:10.1016/j.tcb.2013.11.010 (2014). 29 Tiana, M. et al. Pluripotency factors regulate the onset of Hox cluster activation in the early embryo. Sci Adv 8, eabo3583, doi:10.1126/sciadv.abo3583 (2022). 30 Torres- Padilla, M. E. & Chambers, I. Transcription factor heterogeneity in pluripotent stem cells: a stochastic advantage. Development 141, 2173- 2181, doi:10.1242/dev.102624 (2014). 31 Esch, D. et al. A unique Oct4 interface is crucial for reprogramming to pluripotency. Nature cell biology 15, 295- 301, doi:10.1038/ncb2680 (2013). 32 Bates, L. E., Alves, M. R. P. & Silva, J. C. R. Auxin- degron system identifies immediate mechanisms of OCT4. Stem Cell Reports 16, 1818- 1831, doi:10.1016/j.stemcr.2021.05.016 (2021). 33 Kaplan, N. et al. The DNA- encoded nucleosome organization of a eukaryotic genome. Nature 458, 362- 366, doi:10.1038/nature07667 (2009). 34 Westerfield, M. The zebrafish book. A guide for the laboratory use of zebrafish (Danio rerio). 4th ed edn, (Univ. of Oregon Press, Eugene., 2000). 35 Kimmel, C. B., Ballard, W. W., Kimmel, S. R., Ullmann, B. & Schilling, T. F. Stages of embryonic development of the zebrafish. Dev Dyn 203, 253- 310 (1995). 36 Buenrostro, J. D., Giresi, P. G., Zaba, L. C., Chang, H. Y. & Greenleaf, W. J. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA- binding proteins and nucleosome position. Nature methods 10, 1213- 1218, doi:10.1038/nmeth.2688 (2013). 37 Afgan, E. et al. The Galaxy platform for accessible, reproducible and collaborative biomedical analyses: 2016 update. Nucleic Acids Res 44, W3- W10, doi:10.1093/nar/gkw343 (2016). 38 Langmead, B. & Salzberg, S. L. Fast gapped- read alignment with Bowtie 2. Nat Methods 9, 357- 359, doi:10.1038/nmeth.1923 (2012). + +<--- Page Split ---> + +39 Barnett, D. W., Garrison, E. K., Quinlan, A. R., Stromberg, M. P. & Marth, G. T. BamTools: a C++ API and toolkit for analyzing and managing BAM files. Bioinformatics 27, 1691- 1692, doi:10.1093/bioinformatics/br174 (2011). 40 Ramirez, F. et al. deepTools2: a next generation web server for deep- sequencing data analysis. Nucleic Acids Res 44, W160- 165, doi:10.1093/nar/gkw257 (2016). 41 Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841- 842, doi:10.1093/bioinformatics/btq033 (2010). 42 Feng, J., Liu, T., Qin, B., Zhang, Y. & Liu, X. S. Identifying ChIP- seq enrichment using MACS. Nat Protoc 7, 1728- 1740, doi:10.1038/nprot.2012.101 (2012). 43 Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 15, 550, doi:10.1186/s13059- 014- 0550- 8 (2014). 44 Grant, C. E., Bailey, T. L. & Noble, W. S. FIMO: scanning for occurrences of a given motif. Bioinformatics 27, 1017- 1018, doi:10.1093/bioinformatics/br064 (2011). 45 Dobin, A. et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 29, 15- 21, doi:10.1093/bioinformatics/bt5635 (2013). 46 Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923- 930, doi:10.1093/bioinformatics/bt656 (2014). 47 Hiller, M. et al. Computational methods to detect conserved non- genic elements in phylogenetically isolated genomes: application to zebrafish. Nucleic Acids Res 41, e151, doi:10.1093/nar/gkt557 (2013). 48 Huang da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4, 44- 57, doi:10.1038/nprot.2008.211 (2009). 49 Kaschek, D., Mader, W., Fehling- Kaschek, M., Rosenblatt, M. & Timmer, J. Dynamic Modeling, Parameter Estimation, and Uncertainty Analysis in R. Journal of Statistical Software 88, 1 - 32, doi:10.18637/jss.v088.ii10 (2019). 50 Raue, A. et al. Correction: Lessons Learned from Quantitative Dynamical Modeling in Systems Biology. PLOS ONE 8, 10.1371/annotation/ea0193d1378- 1371f1377f- 1492a- b1370b1377- d877629fdcee, doi:10.1371/annotation/ea0193d8- 1f7f- 492a- b0b7- d877629fdcee (2013). 51 Raue, A. et al. Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood. Bioinformatics 25, 1923- 1929, doi:10.1093/bioinformatics/btp358 (2009). 52 Alon, U. Network motifs: theory and experimental approaches. Nature Reviews Genetics 8, 450- 461, doi:10.1038/nrg2102 (2007). 53 Maiwald, T. et al. Driving the Model to Its Limit: Profile Likelihood Based Model Reduction. PLOS ONE 11, e0162366, doi:10.1371/journal.pone.0162366 (2016). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- TableS1accessibleregions.xlsx- TableS2zygotictranscripts.xlsx- TableS3ARslinkedtotranscripts.xlsx- TableS4Coremodelequations.csv- TableS5CoremodelBestfitparameters.csv- TableS6Minimodelequations.csv- TableS7Minimodelparameters.csv- RiesleSUPPLng3.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2_det.mmd b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..2c565541e471e57d47a97422ee5e9afd45ba0722 --- /dev/null +++ b/preprint/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2/preprint__129058fcec31e99c0819f8921d0245dbb3bd7403d48f93401d336635a4c8ebb2_det.mmd @@ -0,0 +1,637 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 952, 175]]<|/det|> +# Activator-blocker model of transcriptional regulation by pioneer-like factors + +<|ref|>text<|/ref|><|det|>[[44, 195, 636, 216]]<|/det|> +Daria Onichtchouk (daria.onichtchouk@biologie.uni- freiburg.de) + +<|ref|>text<|/ref|><|det|>[[50, 219, 600, 238]]<|/det|> +University of Freiburg https://orcid.org/0000- 0001- 6497- 1445 + +<|ref|>text<|/ref|><|det|>[[44, 243, 600, 285]]<|/det|> +Aileen Riesle University of Freiburg https://orcid.org/0000- 0002- 8195- 1291 + +<|ref|>text<|/ref|><|det|>[[44, 290, 243, 331]]<|/det|> +Meijiang Gao University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 337, 243, 377]]<|/det|> +Marcus Rosenblatt University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 383, 600, 424]]<|/det|> +Jacques Hermes University of Freiburg https://orcid.org/0000- 0003- 3608- 5969 + +<|ref|>text<|/ref|><|det|>[[44, 429, 243, 470]]<|/det|> +Helge Hass University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 475, 243, 515]]<|/det|> +Anna Gebhard University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 521, 243, 561]]<|/det|> +Marina Veil University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 567, 243, 607]]<|/det|> +Bjoern Gruening University of Freiburg + +<|ref|>text<|/ref|><|det|>[[44, 613, 603, 654]]<|/det|> +Jens Timmer University of Freiburg https://orcid.org/0000- 0003- 4517- 1383 + +<|ref|>sub_title<|/ref|><|det|>[[44, 696, 101, 713]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 733, 137, 752]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 771, 330, 790]]<|/det|> +Posted Date: January 25th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 809, 475, 828]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2485402/v1 + +<|ref|>text<|/ref|><|det|>[[44, 846, 910, 889]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 907, 531, 926]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 916, 88]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 14th, 2023. See the published version at https://doi.org/10.1038/s41467-023-41507-z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[117, 82, 880, 133]]<|/det|> +# Activator-blocker model of transcriptional regulation by pioneer-like factors + +<|ref|>text<|/ref|><|det|>[[117, 146, 881, 205]]<|/det|> +Aileen Julia Riesle\*,1,8, Meijiang Gao\*,1, Marcus Rosenblatt\*,2,4, Jacques Hermes\*,2,4 Helge Hass2,4, Anna Gebhard1, Marina Veil1, Björn Grüning5,6, Jens Timmer\*\*,2,3,4, Daria Onichtchouk\*\*1,3,7 + +<|ref|>title<|/ref|><|det|>[[118, 225, 345, 245]]<|/det|> +# \*contributed equally + +<|ref|>text<|/ref|><|det|>[[175, 270, 881, 312]]<|/det|> +1Department of Developmental Biology, Albert- Ludwigs- University of Freiburg, 79104 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 308, 880, 328]]<|/det|> +2Institute of Physics, Albert- Ludwigs- University of Freiburg, 79104 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 333, 789, 352]]<|/det|> +3Signalling Research centers BIOSS and CIBSS, 79104 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 357, 870, 376]]<|/det|> +4Freiburg Center for Data Analysis and Modelling (FDM), 79104 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 380, 870, 400]]<|/det|> +5Department of Computer Science, University of Freiburg, 79110 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 405, 880, 440]]<|/det|> +6Center for Biological Systems Analysis (ZBSA), University of Freiburg, 79104 Freiburg, Germany + +<|ref|>text<|/ref|><|det|>[[175, 444, 727, 462]]<|/det|> +7Institute of Developmental Biology RAS, 119991 Moscow, Russia + +<|ref|>text<|/ref|><|det|>[[175, 467, 880, 518]]<|/det|> +8Current address: Epigenetics and Neurobiology Unit, European Molecular Biology Laboratory, EMBL Rome, Adriano Buzzati- Traverso Campus, Via Ramarini 32, 00015 Monterotondo (RM), Italy + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 242, 107]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[116, 107, 881, 321]]<|/det|> +Zygotic genome activation (ZGA) in the development of flies, fish, frogs and mammals depends on pioneer- like transcription factors (TFs). Those TFs create open chromatin regions, promote histone acetylation on enhancers, and activate transcription. Here, we use the panel of single, double and triple mutants for zebrafish genome activators Pou5f3, Sox19b and Nanog, multi- omics and mathematical modeling to investigate the combinatorial mechanisms of genome activation. We show that Pou5f3 and Nanog act differently on synergistic and antagonistic enhancer types. Pou5f3 and Nanog both bind as pioneers on synergistic enhancers, promote histone acetylation and activate transcription. Antagonistic enhancers are activated by pioneer binding of one of these factors. The other TF binds as non- pioneer, competes with the activator and blocks all its effects, partially or completely. This activator- blocker mechanism mutually restricts widespread transcriptional activation by Pou5f3 and Nanog and prevents premature expression of late developmental regulators in the early embryo. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 100, 295, 123]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[116, 123, 881, 272]]<|/det|> +Awakening of zygotic transcription during maternal- to- zygotic transition is a universal feature of multicellular organisms. The major wave of Zygotic Genome Activation (ZGA) is driven by sequence- specific transcription factors (TFs), different across species \(^{1,2}\) . Nucleosomes protect enhancers from binding of most TFs and are the main obstacles for transcriptional activation. The small group of pioneer TFs directly bind to nucleosomes and create open, nucleosome- free chromatin regions on enhancers \(^{3}\) . Zygotic genome activators act as pioneer- like factors, by opening chromatin, initiating enhancer activation, and widespread zygotic gene expression \(^{4,5}\) . Direct nucleosome binding was shown for some of the genome activators \(^{6,7}\) . + +<|ref|>text<|/ref|><|det|>[[116, 286, 881, 420]]<|/det|> +In animals, gene products synthetized at ZGA enable gastrulation and subdivide the embryo into the three germ layers, ectoderm, mesoderm and endoderm. The regulators of later developmental programs, organogenesis and cell lineage specification, are kept silent at ZGA. They will be synthesized shortly before the beginning of appropriate developmental stages \(^{8,9}\) . The expression waves of transcripts encoding cohorts of transcriptional regulators follow each other in precise order in the developmental time course \(^{10}\) . Molecular mechanisms which keep lineage specifying genes silent at ZGA are unknown. + +<|ref|>text<|/ref|><|det|>[[116, 434, 881, 599]]<|/det|> +Maternal transcription factors Pou5f3, Sox19b and Nanog (PSN) together activate zygotic gene expression in zebrafish \(^{11 - 13}\) . Sox19b is the only maternal member of the SoxB1 family. Shortly after ZGA, early zygotic SoxB1 factors Sox3, Sox19a and Sox2 are expressed and act redundantly with Sox19b \(^{14}\) . PSN, as well as their mammalian counterparts, frequently colocalize on the same genomic sites; therefore, they initially were thought to cooperate \(^{12,15}\) . However, recent studies in mouse and fish revealed that the factors may act alone, additively or interchangeably \(^{5,16,17}\) . PSN open chromatin on more than half of active enhancers at ZGA, but are necessary for expression of only a fraction of early zygotic genes \(^{5}\) . It is unclear how widespread pioneer- like activity of PSN on enhancers relates to early zygotic transcription. + +<|ref|>text<|/ref|><|det|>[[116, 614, 881, 730]]<|/det|> +The purpose of this study was to understand the mechanisms of combined Pou5f3, Sox19b and Nanog action, and the connection between PSN pioneer activity, enhancer activation and selection of the early zygotic gene expression repertoire. We found that Pou5f3 and Nanog in combination not only activate part of zygotic transcripts, but reciprocally restrict or block transcriptional activation by each other on hundreds of developmental enhancers. Further, competition between Pou5f3 and Nanog on common genomic sites establishes the order of gene expression after ZGA. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 227, 107]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[117, 124, 625, 141]]<|/det|> +## Study design and overview of the ATAC-seq data set + +<|ref|>text<|/ref|><|det|>[[116, 141, 881, 256]]<|/det|> +For multi- omic assays, we used the panel of eight zebrafish genotypes: zebrafish wildtype embryos (WT), maternal- zygotic mutants for Pou5f3 (P, MZspg), Sox19b (S, MZsox19b), and Nanog (N, MZnanog), double mutants MZps, MZpn, MZsn, and the triple mutant MZtriple. Sox19b was dispensable for normal development, the absence of Pou5f3, Pou5f3/Sox19b or Nanog resulted in abnormal gastrulation as previously reported \(^{17 - 19}\) . At the absence of Pou5f3/Nanog, Sox19b/Nanog or all three factors zygotic development was arrested (Fig. 1a). + +<|ref|>text<|/ref|><|det|>[[116, 271, 881, 386]]<|/det|> +To assess individual and combined effects of Pou5f3, Sox19b, and Nanog on enhancer regulation and transcription, we performed ATAC- seq and time- resolved RNA- seq profiling in all genotypes. We then aimed to derive enhancer groups differentially regulated by PSN using genomic data (Table S1), and differentially regulated target gene groups using transcriptome (Table S2). Finally, for cross- validation, we put together the independently obtained results of the genome and transcriptome parts (Table S3). The outline of data analysis is shown in Fig. S1. + +<|ref|>text<|/ref|><|det|>[[116, 402, 881, 583]]<|/det|> +We started the genomic analysis by estimating the differences in the chromatin accessibility between the genotypes, on 102945 Accessible Regions (ARs) in the wild- type. We made three initial observations using Principal Component Analysis (PCA, Fig. 1b). First, chromatin accessibility profiles in MZsox19b mutants were close to the wild- type. Sox19b is the major SoxB1 factor expressed at 3.7 and 4.3 hpf, we therefore assumed that the Sox19b factors acts as a cofactor or redundantly with Pou5f3 or Nanog. Second, accessibility profiles of MZsn, MZpn, and MZtriple mutants clustered together in PCA space, corresponding to the most severe phenotypes of these mutants. Third, accessibility profiles of Pou5f3 and Nanog single mutants were quite different, suggesting that these TFs change chromatin accessibility on different regions, or in the opposite directions. + +<|ref|>sub_title<|/ref|><|det|>[[116, 599, 880, 632]]<|/det|> +## Analysis of MZtriple mutant suggests the presence of additional genome activators in early zebrafish embryo + +<|ref|>text<|/ref|><|det|>[[116, 632, 881, 797]]<|/det|> +To tease apart the apparently complex regulation of chromatin accessibility by Pou5f3, SoxB1 and Nanog, we first characterized the effects of their combined activity on chromatin using MZtriple mutant. We divided all ARs to three groups: "down", where the chromatin accessibility was reduced in MZtriple compared to the wild- type (31% of all ARs), "up", where the chromatin accessibility was increased (15% of all ARs), and "same" - unchanged (Fig. 2A). Sequence- specific binding motifs for all three factors were enriched in "down" regions, while "up" regions had the highest content of G and C nucleotides and of GC- rich TF- binding motifs (Fig. 2b, c). The genes closest to "down" and "up" ARs were both enriched in transcriptional regulatory functions among the others (Fig. 2d). + +<|ref|>text<|/ref|><|det|>[[116, 812, 881, 928]]<|/det|> +Next, we analyzed the time- resolved transcriptome in MZtriple mutant compared to the wild- type, using our previously developed R package RNA- sense (www.bioconductor.org/packages/release/bio/html/RNAsense.html \(^{17}\) ). We selected 4777 transcripts zygotically expressed in the wild- type or in MZtriple, and split them into three groups (Fig. 2e): "DOWN" (downregulated compared to the wild- type or not expressed in MZtriple), "SAME" (unchanged), and "UP" (upregulated in MZtriple compared to the wild- type, or expressed only in MZtriple). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 98, 881, 248]]<|/det|> +Out of 791 "UP" group transcripts, \(67\%\) were zygotically expressed only in MZtriple (i.e. hoxc8a, nr2f2). We wondered if lineage- specifying regulatory genes were prematurely expressed MZtriple, as we observed previously in MZspg and MZps mutants \(^{17,20}\) . To test that, we used published data \(^{10}\) to calculate the maximal expression time during normal zebrafish development for each zygotic gene. Indeed, the median expression time in the "UP" group was 24 hpf, versus 8 hpf in the groups "DOWN" and "SAME" (Fig. 2f). Similar analysis for other mutants revealed that except for MZsox19b, all of them prematurely expressed diverse sets of late regulatory genes (Fig. S2). + +<|ref|>text<|/ref|><|det|>[[116, 262, 881, 330]]<|/det|> +Finally, we checked if the changes in chromatin accessibility on the putative regulatory regions of zygotic genes correlated with the changes in their expression in MZtriple. This was indeed the case: "down" ARs were enriched around the promoters of "DOWN" genes, and "up" ARs around "UP" genes in MZtriple (Fig. 2g). + +<|ref|>text<|/ref|><|det|>[[116, 344, 881, 427]]<|/det|> +Taken together, our analysis confirmed previous findings that Pou5f3, Sox19b and Nanog pioneer chromatin accessibility on the enhancers of early zygotic genes \(^{5,21,22}\) . The increase in transcription in MZtriple mutants was unexpected. We concluded that additional activators were present at ZGA. These unknown activators could bind to GC- rich regulatory elements. + +<|ref|>text<|/ref|><|det|>[[115, 442, 881, 560]]<|/det|> +Four types of regulation of chromatin accessibility by Pou5f3, SoxB1 and Nanog To investigate how Pou5f3, SoxB1 and Nanog create the regions of open chromatin, we restricted our analysis to the ARs, where any combination of the three factors was required for chromatin accessibility ("down" in MZtriple, Fig. 2a). To study only direct TF effects, we further selected 20131 TdARs (TF- bound Accessible Regions, downregulated in the MZtriple mutant), overlapping with ChIP- seq peaks for Pou5f3, SoxB1 \(^{12}\) , or Nanog \(^{23}\) . + +<|ref|>text<|/ref|><|det|>[[115, 574, 881, 707]]<|/det|> +We then classified TdARs by non- redundant requirements for Pou5f3 and Nanog for chromatin accessibility into four groups (Fig. 3a). Accessibility of the 1. PN group TdARs was reduced in MZspg and MZnanog, so we concluded that Pou5f3 and Nanog were both required in these regions. Pou5f3 was required in the 2. P group, Nanog in the 3. N group, and none of the single factors was required in the 4. - group (i.e. several factors were required redundantly). Out of the four groups, 2. P and 3. N were the most enriched for the Pou5f3- and Nanog - motifs, and the most highly occupied by Pou5f3 and Nanog, respectively (Fig. 3b, c). + +<|ref|>text<|/ref|><|det|>[[115, 721, 881, 888]]<|/det|> +To address if chromatin becomes accessible in all groups at the same time, we performed ATAC- seq at major ZGA (3 hpf), and compared ATAC- seq signals at 3, 3.7 and 4.3 hpf, in the wild- type and MZtriple. While all groups depended on PSN already at ZGA, chromatin in 4. - group was more accessible in MZtriple in all stages than on the other groups (Fig. 3d, dashed lines). Therefore, it was conceivable that factors other than PSN also contribute to chromatin accessibility in the group 4. - . We found that GC content increased in order \(1. \mathrm{PN} < 2. \mathrm{P} < 3. \mathrm{N} < 4. - ,\) (Fig. 3e). As our analysis of MZtriple mutants suggested that GC- binding activators were present at ZGA, we hypothesized that GC- binding factor X could be involved in opening chromatin in group 4. - regions (Fig. 3f). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 345]]<|/det|> +We next wondered whether the factors required for chromatin accessibility in a given group of TdARs were also sufficient to open chromatin. We used rescue experiments of Miao et al., \(2022^{5}\) to answer this question. To address which factors can restore chromatin accessibility in the triple mutants MZnpS, the authors microinjected Pou5f3, Sox19b, and Nanog mRNAs, individually and in combinations, and compared ATAC- seq signals in the injected and non- injected MZnpS embryos. Using their data, we scored the chromatin accessibility rescue in each of the four groups. Nanog alone rescued more than \(90\%\) of TdARs in the 3. N group. Pou5f3 alone rescued only \(26\%\) of TdARs in the 2. P group; most of the rest could be rescued by the Pou5f3/Sox19b or Pou5f3/Nanog combinations (Fig. S3a- f). Further, in \(59\%\) of TdARs in the 2. P group chromatin accessibility was also downregulated in the double Sox19b/Nanog mutant MZsn (Fig. S3g, h). Thus, although Pou5f3 was present in MZsn, it was apparently not sufficient to open chromatin in most of its target regions. In sum, we concluded that Nanog alone was required and sufficient for establishing chromatin accessibility in the majority of its target regions. In contrast, Pou5f3 needed assistance of either Nanog or SoxB1 to establish accessible chromatin. + +<|ref|>text<|/ref|><|det|>[[116, 361, 881, 460]]<|/det|> +Fig. 3f summarizes four types of establishment of chromatin accessibility by pioneer- like factors. Pou5f3 and Nanog are both required on the 1. PN group TdARs. Pou5f3 and redundant contribution of Nanog or SoxB1 are required on most of the 2. P group TdARs. Nanog is required on most of the 3. N group TdARs. Pou5f3, SoxB1 or Nanog are redundantly required on 4. - group TdARs, together with a hypothetical GC- binding factor X. + +<|ref|>sub_title<|/ref|><|det|>[[115, 476, 880, 509]]<|/det|> +## Pou5f3 and Nanog act as Activator and Blocker and vice versa on a fraction of common enhancers + +<|ref|>text<|/ref|><|det|>[[116, 509, 881, 590]]<|/det|> +Pou5f3, SoxB1, and Nanog together promote histone acetylation on their binding sites, which is critical for the enhancer activation5. However, individual roles of the TFs remain unclear. We used H3K27ac ChIP- seq data in the single mutants17 (and this work) and MZnpS triple mutant5 to analyze how H3K27 acetylation relates to the pioneer- like activity of the factors. + +<|ref|>text<|/ref|><|det|>[[116, 606, 881, 722]]<|/det|> +We found that the same TF which was strictly required for chromatin accessibility was required for H3K27 acetylation: both Pou5f3 and Nanog in the 1. PN group, Pou5f3 in the 2. P group, Nanog in the 3. N group, and none of the single factors in the 4. - group. Unexpectedly, single mutant analysis revealed reciprocal antagonistic relationships between Pou5f3/Sox19b and Nanog: Nanog reduced H3K27ac in the 2. P group, and Pou5f3 reduced H3K27ac in the 3. N group (Fig. 4a). The PSN combined activity was required for H3K27 acetylation in all groups (Fig. 4b). + +<|ref|>text<|/ref|><|det|>[[116, 737, 881, 886]]<|/det|> +To further investigate the individual roles of the TFs, we defined H3K27- acetylated TdARs as enhancers (STAR methods, Table S1). We then scored the enhancers on which H3K27ac was induced (+), unchanged (0), or reduced (- ) by each TF (Fig. 4c), or by the PSN combined activity (Fig. 4d), in each of the four groups. We noted that the percentage of enhancers activated by PSN was the highest in the 1. PN and the lowest in the 4. - group (Fig. 4c, d), and thus inversely correlated with GC content (Fig. 3e). Importantly, our analysis uncovered a large number of previously uncharacterized "antagonistic enhancers", where Pou5f3 and Nanog regulated H3K27ac in the opposite directions (total 2504 p+n- and p-n+ enhancers, boxed in Fig. 4c). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 82, 881, 264]]<|/det|> +To understand how per se transcriptional activators and pioneer- like factors could negatively regulate enhancer activity, we took a closer look on chromatin accessibility changes on the antagonistic enhancers. We found that the factor which reduced H3K27ac also reduced chromatin accessibility (Fig. 4e, Fig. S4a- d, f). We confirmed this result using MNase- seq nucleosome positioning data22 (Fig. S4e, f). Together, we have shown by two independent methods that only one of the TFs, either Pou5f3 or Nanog, acts as pioneer- like factor on the antagonistic enhancers. This observation suggested that Pou5f3 and Nanog alternate their binding mode, acting as pioneer- like factors on some genomic sites and as non- pioneers on the others. The non- pioneer TF reduces chromatin accessibility, perhaps by competing for common genomic site with the pioneer- like TF. + +<|ref|>text<|/ref|><|det|>[[116, 280, 881, 378]]<|/det|> +Our next question was whether the positive and negative effects of Pou5f3 and Nanog on chromatin accessibility correlate with sequence features of the bound sites. As shown in Fig. S4g, Pou5f3 induced chromatin accessibility on its own motifs and reduced it on Nanog motifs, and vice versa. While common homeodomain- binding sequence in Pou5f3 and Nanog cognate motifs is recognized by both TF(Fig. 3b), exact match to the motif is necessary for the pioneer- like activity. + +<|ref|>text<|/ref|><|det|>[[116, 393, 881, 592]]<|/det|> +Summarizing our findings, we suggest the general mechanism of antagonistic interactions between Pou5f3 and Nanog. Two TFs, A (activator) and B (blocker) compete for binding on the common motif within an enhancer (Fig. 4f). The activator acts as a pioneer- like factor on this motif: it displaces nucleosomes and promotes histone acetylation. Non- pioneer binding of the blocker protects the motif from the activator, thereby reducing nucleosome displacement and histone acetylation. Two examples illustrate the model. On her3 regulatory region, Pou5f3 together with SoxB1 act as an activator and Nanog as a blocker on sox:pou motif (Fig. 4g). On more3b regulatory region, Nanog acts as an activator and Pou5f3 as a blocker on two nearby nanog motifs (Fig. 4h). Hence, Pou5f3 and Nanog compete as pioneer- like and non- pioneer factors on common genomic sites, eliciting opposite effects on enhancer activity. + +<|ref|>sub_title<|/ref|><|det|>[[116, 607, 833, 624]]<|/det|> +## Mathematical modelling of transcriptional regulation by genome activators + +<|ref|>text<|/ref|><|det|>[[117, 624, 881, 705]]<|/det|> +We wondered next, whether the antagonistic enhancers are relevant for zygotic gene expression. To answer this question, we chose the strategy to on the one hand sort zygotic transcripts into groups according to their regulation, without involving genomic information. On the other hand, we linked enhancers to transcripts and determined correlations between the enhancer and transcript groups. + +<|ref|>text<|/ref|><|det|>[[116, 721, 881, 837]]<|/det|> +To sort the transcriptome, we used a mathematical modeling approach in two subsequent steps. In the first step, we developed a "core model" to describe the dynamic behavior of the known core components of the system: Pou5f3, Nanog, and SoxB1 group genes (including Sox19b, Sox19a, Sox3 and Sox2). In the second step, dynamics of these core components were then used to inform ODE- based mini models to formalize all possible regulatory inputs from Pou5f3, Nanog or SoxB1 on the individual target genes. + +<|ref|>text<|/ref|><|det|>[[117, 853, 881, 918]]<|/det|> +The core model was based on ordinary differential equations (ODEs), where initial assumptions on the regulatory interactions between the core components were taken from the literature (STAR methods, Fig. S5a for model cartoon and Table S4 for model equations). To infer the model's parameters, we used the measured time- resolved + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 82, 880, 133]]<|/det|> +transcriptional profiles for pou5f3, nanog, sox19b, sox19a, sox3 and sox2. After model calibration, the model trajectories resulting from the best fit parameters were in good agreement with the experimental data (Fig. S5b, c, Fig. S6a, and Table S5). + +<|ref|>text<|/ref|><|det|>[[115, 148, 881, 362]]<|/det|> +For the second step of ODE- based mini models, we assumed that at least one factor of Pou5f3, Nanog or SoxB1 should activate the target, while the other two can activate \((+)\) , repress \((- )\) or do nothing (0). This gives 19 possible regulatory combinations. Each of the 19 mini models was separately fitted to experimental time- resolved RNA- seq data for each of the 1799 zygotic transcripts that had been found to be directly or indirectly activated by Pou5f3, Sox19b or Nanog ("DOWN" group in Fig. 2e, Table S6). The dynamics of Pou5f3, Nanog, and SoxB1 group genes (SOX) obtained in step one of the analysis were used as input to the mini models, while the remaining parameters were again calibrated based on the RNA- seq data. To simplify the analysis with respect to our goal of connecting chromatin state and transcription for Pou5f3 and Nanog, we merged the 19 mini- models down to six mini- model groups capturing all possible variants of target regulations by Pou5f3 and Nanog and the regulation by SOXB1 alone (Fig. 5a). + +<|ref|>text<|/ref|><|det|>[[116, 377, 880, 461]]<|/det|> +For each transcript, we selected the best candidate mini- model group according to Bayesian Information Criterion (BIC, see Table S7 and Table S2). Fits of selected mini- model groups are shown for exemplary targets in Fig. 5b and Fig. S6b. 1799 transcripts were then classified into six groups according to the best fitting direct regulatory scenario (Fig.5c). + +<|ref|>text<|/ref|><|det|>[[116, 476, 880, 576]]<|/det|> +Interestingly, three out of six groups of targets significantly associated with gene ontology terms. Groups 1. \(\mathrm{P + N + }\) and 3b. \(\mathrm{P - N + }\) were enriched in the regulators of transcription and development, while the largest group 3a. \(\mathrm{PON + }\) was enriched in the housekeeping genes (Fig. S7a). In sum, we sorted PSN- dependent zygotic transcripts to six groups with biologically relevant functions, using formal and unbiased mathematical modelling approach. + +<|ref|>sub_title<|/ref|><|det|>[[115, 591, 880, 625]]<|/det|> +## Competition between Pou5f3 and Nanog on antagonistic enhancers establishes the order of gene expression. + +<|ref|>text<|/ref|><|det|>[[116, 625, 881, 771]]<|/det|> +To inquire if synergistic and antagonistic chromatin regulation by Pou5f3 and Nanog on common enhancers is directly linked to synergistic and antagonistic regulation of transcription, we put together the results of genomic and transcriptomic analyses. As shown in Fig. 5d- f (see also Fig. S7b, c), the predicted direction of transcriptional regulation of targets by Pou5f3 and Nanog (activation or repression) strongly correlated with Pou5f3 and Nanog- dependent H3K27ac changes on their enhancers (activation or blocking), respectively. Thus, cross- validation of independent genomic and transcriptomic data sets demonstrated that the transcription at ZGA is directly regulated by synergistic and antagonistic enhancers. + +<|ref|>text<|/ref|><|det|>[[116, 787, 880, 903]]<|/det|> +We have noticed that all mutants except MZsox19b prematurely transcribe poorly overlapping sets of genes, enriched for DNA- binding and transcription regulatory functions (Fig. S2). We also demonstrated that Pou5f3 and Nanog compete on antagonistic enhancers as activator and blocker of chromatin accessibility and H3K27 acetylation. Putting these two findings together, we wondered again whether Pou5f3 could directly block premature transcriptional activation by Nanog and vice versa. To answer this question, we examined putative regulatory regions of the transcripts, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 880, 116]]<|/det|> +upregulated in the Pou5f3 and Nanog mutants, for the enrichment of antagonistic enhancers of the opposite types. + +<|ref|>text<|/ref|><|det|>[[115, 131, 881, 248]]<|/det|> +In Pou5f3 mutant MZspg, \(17\%\) of transcripts were upregulated compared to the wildtype (Fig. 6a) and contained prematurely expressed genes (Fig. 6b). Strikingly, the putative regulatory regions of these genes were highly enriched for antagonistic 3. N p- n+ enhancers (Fig. 6c, boxed). 3. N p- n+ enhancers were also enriched in the putative regulatory regions of genes upregulated in the double Pou5f3/Sox19b mutant MZps (Fig. S7d- f). Thus, Pou5f3 blocked inappropriate transcriptional activation of late developmental genes by Nanog. + +<|ref|>text<|/ref|><|det|>[[115, 263, 881, 362]]<|/det|> +To test if the reverse was also true, we examined the transcripts upregulated in MZnanog. \(10\%\) of transcripts were upregulated compared to the wild- type in MZnanog and contained prematurely expressed genes (Fig. 6d, e). The putative regulatory regions of these genes were enriched for antagonistic 2. P p+ n- enhancers (Fig. 6f, boxed). Thus, Nanog blocked inappropriate transcriptional activation of late developmental genes by Pou5f3. + +<|ref|>text<|/ref|><|det|>[[115, 377, 881, 494]]<|/det|> +In sum, cross- validation of independent genomic and transcriptomic data sets demonstrated that the transcription of PSN target genes at ZGA is directly regulated by synergistic and antagonistic enhancers. On the synergistic enhancers, Pou5f3 and Nanog are both required to establish chromatin accessibility and activate early zygotic genes (Fig. 7a). On the antagonistic enhancers, "blocker" TF(- ) attenuates activation of the early zygotic genes by "activator" (TF+), or prevents the activation of lineage- specific regulators by "activator" TF (Fig. 7b, c). + +<|ref|>sub_title<|/ref|><|det|>[[117, 510, 279, 532]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[116, 533, 881, 632]]<|/det|> +Our study provides the mechanistic links between the pioneer- like activity of genome activators Pou5f3, Sox19b (together with zygotic SoxB1 factors) and Nanog, and their potential to activate transcription at ZGA and during the first half of gastrulation. We demonstrate that direct binding of PSN to genome is not only responsible for transcriptional activation at ZGA, but also for adjusting the levels of early zygotic transcripts and for the timing of gene expression in development. + +<|ref|>text<|/ref|><|det|>[[115, 647, 881, 844]]<|/det|> +Synergistic interactions were long ago suggested for zebrafish genome activators and their mammalian homologs12,15. Antagonistic interactions of Pou5f3 and Nanog on chromatin are a new finding. A reasonable question is which features define pioneer- and non- pioneer binding in each case. We found a correlation with two sequence features. One feature is GC content around TF binding motifs: on synergistic enhancers, GC content is lower than on the other enhancer types (Fig. 3e, Fig. 4c). Another feature is the different frequency of sequence- specific motifs: we found that a given TF acts more frequently as activator on its own motifs, and as blocker as it binds to the motifs of the other factor (Fig. 4f- h, Fig. S4g). In a sense, "activator" and "blocker" TFs can be compared with a precise key and a key blank to the locked door (which is an enhancer): both keys fit to the lock, but only a precise key can open the door (activate an enhancer). + +<|ref|>text<|/ref|><|det|>[[116, 860, 881, 927]]<|/det|> +In whole embryo assays, we could not distinguish whether the ATAC- seq signals come from the whole embryo or from the subpopulation of cells. Therefore, we cannot exclude that cell- specific differences, such as local concentrations of TFs, presence of transcriptional cofactors or cross- talk with signaling pathways affect the binding mode + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 82, 881, 181]]<|/det|> +cell type- specifically. In line with that, it was shown that nucleosome displacement activity of known mammalian pioneer factors, including Oct4/Pou5f1, is conditional: all of them bind to different genomic locations in different cell types24. Further, it is also possible that antagonistic enhancers act as bifunctional cis- regulatory elements, i.e. as activators or as silencers, depending on the embryonic cell type25. Single cell multi- omics and reporter assays are needed to clarify this issue. + +<|ref|>text<|/ref|><|det|>[[116, 197, 881, 378]]<|/det|> +The dual role of Pou5f3 and Nanog in activating early zygotic genes and blocking premature transcription provides an immediate parallel to the biological roles of their mammalian homologues. Oct4/Pou5f1 and Nanog regulate the seemingly opposite processes of pluripotency maintenance and cell lineage decisions in mammalian embryos and ES cells26- 30. Although zebrafish Pou5f3 is not equivalent to Oct4/Pou5f1 in terms of nucleosome- binding and reprogramming capacity31, Pou5f3 and Oct4 recognize the same motifs and colocalize with Nanog in the respective systems. It is tempting to speculate that Oct4 and Nanog may reciprocally block some of each other activities similarly to their zebrafish homologues. Recent study supports this possibility: acute depletion of Oct4 in ES cells results in increased genomic binding of Nanog, suggesting that Oct4 outcompetes Nanog on chromatin and antagonizes its function32. + +<|ref|>text<|/ref|><|det|>[[116, 393, 881, 560]]<|/det|> +Our results suggest that the list of zebrafish genome activators is not complete: hundreds of transcripts, enriched in transcriptional regulatory functions, are prematurely expressed in MZtriple mutant and linked to regulatory elements with high GC content (Fig. 2). Identification of putative GC- binding factors regulating these genes and dissecting their interplay with Pou5f3, SoxB1 and Nanog will further clarify the mechanisms of gene regulation at ZGA. Finally, the concepts presented in this work will assist the studies of combinatorial genome activation by TFs in other organisms and deepen our understanding of two global and currently unresolved questions: conditional activity of pioneer and pioneer- like factors in different biological settings, and of how the gene expression is regulated in developmental time. + +<|ref|>sub_title<|/ref|><|det|>[[118, 576, 386, 599]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[116, 599, 881, 682]]<|/det|> +We are grateful to Rainer Duden for commenting on the manuscript, to Sabine Götter for fish care, and to Andrea Buderer and Cornelia Wagner for administrative support. This work was supported by DFG- ON86/4- 2 and DFG- ON86/6- 1 for DO, and DFG- EXC2189 - Project ID: 390939984 for D.O. and J.T. BG is funded by DFG grant SFB 992/1 2012 and BMBF grant 031 A538A RBC. + +<|ref|>sub_title<|/ref|><|det|>[[118, 700, 425, 722]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[117, 722, 881, 772]]<|/det|> +AJR, MG, MV, and AG - experiments; AJR, MG, BG and DO - data analysis, MR and JH - mathematical modeling, DO, JT - supervision of the study. DO wrote the manuscript, all authors edited the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[117, 789, 470, 811]]<|/det|> +## Declarations of Interests + +<|ref|>text<|/ref|><|det|>[[118, 812, 507, 828]]<|/det|> +The authors declare no competing interests + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 131, 870, 666]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 666, 880, 700]]<|/det|> +
Figure 1. Single and combined mutants by zebrafish zygotic genome activators used in this study.
+ +<|ref|>text<|/ref|><|det|>[[115, 702, 881, 787]]<|/det|> +a. Mutant phenotypes at blastula (dome) and midgastrula (shield) stages. White dotted line shows epiboly border; double arrows show the internalized yolk. Scale bar=100μm. b. Principal Component Analysis of ATAC-seq data in all mutants, on the genomic regions accessible in the wild-type (ARs). The data in eight genotypes clustered into four groups (boxed). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 857, 760]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 760, 880, 794]]<|/det|> +
Figure 2. Changes in chromatin accessibility parallel the changes in gene expression in Pou5f3, Sox19b and Nanog triple mutant.
+ +<|ref|>text<|/ref|><|det|>[[115, 794, 881, 925]]<|/det|> +a. Three groups of accessible regions (ARs) by chromatin accessibility change in the MZtriple: "down", "same", "up". b. ARs of "up" group had the highest GC content. P-values for Tukey-Kramer test; p-value for 1-way ANOVA was \(< 2e^{-16}\) . Co - control genomic regions (dotted line). c. Pou5f3, Sox and Nanog-binding motifs (black rectangle) are enriched in "down" ARs. d. GREAT analysis. e. Three groups of zygotic transcripts by expression change in MZtripleThe heatmap shows normalized expression at eight time points, from pre-ZGA (2.5 hpf) till 6 hpf; example developmental genes at the left. f. Group of zygotic genes is prematurely expressed in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 199]]<|/det|> +MZtriple. Y- axis: time of maximal expression in the normal development (3 hpf - 120 hpf), schematic embryo drawings illustrate the stages. Median expression time of transcripts upregulated in MZtriple was 24 hpf (yellow dotted line), versus 8 hpf for down- or unchanged transcripts (gray dotted line). P- values in Tukey- Kramer test; p- value in 1- way ANOVA was \(< 2e^{- 16}\) . g. Down- or upregulation of chromatin accessibility in MZtriple correlates with respective transcriptional changes of linked genes. \(\chi^2\) test; positive correlations are shown in blue and negative in red. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 81, 880, 500]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 509, 880, 545]]<|/det|> +
Figure 3. Four types of TdARs by pioneer-like activity of Pou5f3, SoxB1, and Nanog.
+ +<|ref|>text<|/ref|><|det|>[[115, 544, 882, 790]]<|/det|> +a. Venn diagram. Four groups of TdARs (by non-redundant requirements for Pou5f3 and/or Nanog for chromatin accessibility. b. Frequencies of PSN motifs in the 4 groups. Black box - common TAAT sequence, recognized by homeodomains. Note that 2.P and 3.N groups are the most enriched in the Pou5f3- and Nanog- specific motifs, respectively (colored boxes). c. 2.P and 3.N groups are the mostly occupied by Pou5f3/SoxB1 and Nanog, respectively (gray dashed lines). Summary ChIP-seq profiles for indicated TFs (rpkm). d. PSN establish accessible chromatin starting from major ZGA (3 hpf). ATAC-seq summary profiles in WT and MZtriple at 3, 3.7 and 4.3 hpf (rpkm). Note that chromatin in group 4.- regions in MZtriple is more accessible than in the other groups (gray dashed line). e. 4.- group has the highest GC content out of all groups. P-values for Tukey-Kramer test; p-value for 1-way ANOVA was \(< 2e^{-16}\) . Lower dashed line shows median genomic control GC content, upper line - median GC content of ARs upregulated in MZtriple. f. Schematic drawing of four types of regulation of chromatin accessibility, percentage of each group from all TdARs is shown. "OR" - logical operator. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 884, 603]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 603, 881, 637]]<|/det|> +
Figure 4. Activator-blocker model: Pou5f3 and Nanog oppose each other effects on H3K27 acetylation and chromatin accessibility on a fraction of enhancers.
+ +<|ref|>text<|/ref|><|det|>[[115, 637, 883, 835]]<|/det|> +a. Summary profiles of H3K27ac histone mark in the wild-type and single mutants, in four types of TdARs: 1. PN, 2. P, 3. N, 4. - . Note the opposite effects of Pou5f3 and Nanog on H3K27 acetylation on the 2. P and the 3. N groups. b. Summary profiles of H3K27ac mark in the WT and MZ_nps. c. Effects of the single TFs on H3K27ac. Antagonistic enhancers are boxed in the legend. d. Effects of PSN combined activity on H3K27ac. e. Pou5f3 and Nanog have the Opposite effects of Pou5f3 and Nanog on chromatin accessibility on antagonistic enhancers. f. Schematic illustration of activator-blocker model (explanations in the text). (g, h) Genomic browser views show (from bottom to top) motif occurrence, TF binding, H3K27ac and ATAC-seq in the indicated genotypes. g. All three TFs bind to sox:pou motif on her3 2. P p+n- antagonistic enhancer (blue shading). h. Pou5f3 and Nanog, or all three factors bind nanog motifs on morc3b 3. N p+n+ antagonistic enhancers (yellow shading). \* - data from5. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 78, 880, 750]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 756, 880, 789]]<|/det|> +
Figure 5. Zygotic gene expression is balanced by synergy and competition of Pou5f3 and Nanog on common enhancers.
+ +<|ref|>text<|/ref|><|det|>[[115, 789, 881, 921]]<|/det|> +a. Schemes of six alternative mini-model groups for direct target regulation by Pou5f3, Nanog and SoxB1. b. foxb1a and tbx5a transcripts dynamics fitted best to the antagonistic groups 2b. P+N- and 3b. P+N+ respectively. c. Heatmap of zygotic genes, downregulated in MZtriple, sorted by best fit to one of the six model groups. d-f. \(\chi^2\) tests, positive correlations between the transcriptomics and genomics groups are shown in blue and negative in red. Vertical axis: transcriptomics groups. Direct enhancers linked to the promoters of zygotic genes were sorted by best fit transcriptional model. Synergistic and antagonistic model groups are boxed. Horizontal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 881, 118]]<|/det|> +axis: direct enhancers were sorted by chromatin accessibility, and by H3K27ac regulation by sum of PSN activities (d), by Pou5f3 (e) or by Nanog (f). \*- data from \(^{5}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 75, 880, 744]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 740, 880, 772]]<|/det|> +
Figure 6. Pou5f3 blocks premature transcriptional activation by Nanog and vice versa.
+ +<|ref|>text<|/ref|><|det|>[[115, 774, 881, 922]]<|/det|> +(a- c). Pou5f3 blocks transcription from Nanog- dependent enhancers. (d- f). Nanog blocks transcription from Pou5f3- dependent enhancers. Zygotic transcripts were sorted by expression change in MZspg (a) or MZnanog (d) compared to the wild- type. Heatmaps show normalized expression at eight time points, from pre- ZGA (2.5 hpf) till 6 hpf; numbers of transcripts in each group are shown at the right. b, e. Premature expression of zygotic genes in MZspg (b) or in MZnanog (e). Median expression time in the normal development was compared in "UP" "DOWN" and "SAME" groups. P- values in Tukey- Kramer test; p- value in 1- way ANOVA was \(< 2e^{- 16}\) . c, f. \(\chi^2\) tests, positive correlations between the transcriptomics and genomics groups are shown in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 166]]<|/det|> +blue and negative in red, scales show Pearson residuals. Vertical axis: transcriptomics groups. Direct enhancers were linked to the promoters of zygotic genes within \(+ / - 50\) kb and sorted according to DOWN", "SAME" and "UP" transcriptional groups. Horizontal axis: genomic groups. Direct enhancers were sorted by changes in chromatin accessibility, and by H3K27ac regulation by Pou5f3 or Nanog, as indicated. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 884, 319]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 323, 880, 357]]<|/det|> +
Figure 7. Synergistic and antagonistic Pou5f3/Nanog direct enhancers: the mechanism of action and direct target genes.
+ +<|ref|>text<|/ref|><|det|>[[115, 357, 881, 473]]<|/det|> +a. Synergistic enhancers Pou5f3+Nanog+: Pou5f3 and Nanog act as activators. b. Antagonistic enhancers Pou5f3+Nanog-: Pou5f3 acts as an activator, Nanog as a blocker. c. Antagonistic enhancers Pou5f3-Nanog+: Nanog acts as an activator, Pou5f3 as a blocker. Representative early targets (expressed in the wild-type during 2.5-6 hpf time course), and late targets (prematurely expressed in the mutants by blocker TF during 2.5-6 hpf time course) are indicated by schematic early and late embryo drawings. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 116, 245, 140]]<|/det|> +## Methods + +<|ref|>table<|/ref|><|det|>[[113, 180, 925, 916]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[116, 164, 363, 180]]<|/det|> +KEY RESOURCES TABLE + +
REAGENT or RESOURCESOURCEIDENTIFIER
Antibodies
Anti-Histone H3 (acetyl K27) rabbit, 1/100 dilutionAbcam plc.,
Cambridge, UK		ab 4729
Chemicals, Peptides, and Recombinant Proteins
cOmplete™, EDTA-free Protease Inhibitor CocktailSigma-Aldrich Chemie GmbH, Germany5056489001
SYTOX GreenThermoFisher SCIENTIFICS7020
Critical Commercial Assays
Agencourt® AMPure® XP BeadsBeckmann Coulter, Krefeld, GermanyA63880
Agilent High Sensitivity DNA KitAgilent Technologies, Santa Clara, California, USA5067-4626
Agilent RNA 6000 Nano KitAgilent Technologies, Santa Clara, California, USA5067-1511
Dynabeads® Protein Ginvitrogen Dynal AS, Oslo, Norway10003D
E.Z.N.A® Cycle Pure KitOmega Biotek, Norcross, Georgia, USAD6493-02
Illumina Tagment DNA Enzyme and Buffer Small KitIllumina20034197
Microcon®-30 Centrifugal FiltersMerck Millipore, Darmstadt, GermanyMRCF0R030
mMESSAGE mMACHINE® SP6 transcription KitAmbion10086184
NEBNext Ultra DNA Library Prep Kit for IlluminaNew England Biolabs, Inc., Frankfurt a.M., GermanyE7370S
NEBNext® Multiplex Oligos for Illumina® (Index Primers Set 1)New England Biolabs, Inc., Frankfurt a.M., GermanyE7335S
RNeasy® Mini KitQIAGEN, Hilden, Germany74104
Quant-iT™ PicoGreen® dsDNA Assay Kitinvitrogen™ Molecular Probes® Waltham, Massachusetts, USAQ33120
Deposited Data
Omni ATAC-seq of the wild-type, MZsox19b, MZspg, MZps zebrafish embryos at 3.7 and 4.3 hpfGao et al., 2022GEO: GSE188364
Omni ATAC-seq of the wild-type, MZnanog, MZpn, MZsn and MZtriple zebrafish embryos at 3.0, 3.7 and 4.3 hpfthis workGEO: GSE215956
Omni ATAC-seq and H3K27ac ChIP-seq of the wild-type and MZnpss zebrafish embryos at 4 hpfMiao et al., 2022SRA: SRP355652
Time-resolved RNA-seq at 8 time points in the wild-type, MZsox19b, MZspg, MZnanog, MZps, MZpn, MZsn and MZtriple zebrafish embryosthis workGEO: GSE162415
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[113, 78, 925, 911]]<|/det|> +
ChIP-seq for H3K27ac in the wild-type, MZspg and MZsox19b zebrafish embryos, 4.3 hpfGao et al., 2022GEO: GSE143306
ChIP-seq for H3K27ac in the MZnanog zebrafish embryos, 4.3 hpfthis workGEO: GSE143439
ChIP-seq for Pou5f3 and Sox2 (SoxB1) TF binding, 5.3 hpf.Leichsenring et al., 2013GEO: GSE39780
ChIP-seq for Nanog TF binding, 4.3 hpfXu et al., 2012GEO: GSE34683
MNase-seq of MZsox19b embryos, 4.3 hpfGao et al., 2022GEO: GSE125945
MNase-seq of WT and MZspg embryos, 4.3 hpfVeil et al., 2019GEO: GSE109410
Zebrafish reference genome assembly danner11/GRCz11Genome reference Consortiumhttps://www.ncbi.nlm.nih.gov/grc/zebrafish
Experimental Models: Organisms/Strains
Wild-type zebrafish strain AB/TLZIRCZL1/ZL86
MZspg zebrafishLunde et al., 2004m793
MZnanog zebrafishVeil et al., 2018m1435
MZsox19b zebrafishGao et al., 2022m1434
MZps zebrafishGao et al., 2022m1434/m793
MZpn zebrafishthis workm1435/m793
MZsn zebrafishthis workm1434/m1435
MZtriple zebrafishthis workm1434/m1435/m793
Oligonucleotides
PCR primer for genotyping
Sox19b-f1 5'-ATTTGGGGTGCTTTCTTCAGC-3'		Gao et al., 2022no
PCR primer for genotyping
Sox19b-r1 5'-GTTCTCCTGGGCCATCTTCC-3'		Gao et al., 2022no
PCR primer for genotyping
Nanog1-f1 5'-CAACTTGCGGTCGCTCTAAAC-3'		this workno
PCR primer for genotyping
Nanog1-r1 5'-GCTCAGTCTTGTTGTAGGACAG-3'		this workno
PCR primer for genotyping
spg-f1 5'-GTCGTCTGACTGAACATTTTGC-3'		this workno
PCR primer for genotyping
spg-r1 5'-GCAGTGATTCTGAGGAAGAGGT-3'		this workno
PCR primer for H3K27ac ChIP-seq control, positive reference
tiparp_f_1 5' CGCTCCCAACTCCTAGTATC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, positive reference
tiparp_r_1 5'-AACGCAAGCCAAACGATCTC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, negative reference
igsf2_f_2 5'-GAACTGCATTAGAGACCCAC-3'		Gao et al., 2022no
PCR primer for H3K27ac ChIP-seq control, negative reference
igsf2_r_2 5'-CAATCAACTGGGAAAGCATGA-3'
Software and Algorithms
Bed ToolsQuinlan and Hall, 2010BED Tools in usegalaxy.eu
Bowtie2Langmead and Salzberg, 2012Bowtie2 in usegalaxy.eu
DeepTools2Ramirez et al., 2016deepTools in usegalaxy.eu
DESeq2Love et al., 2014DESeq2 in usegalaxy.eu
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[114, 80, 925, 514]]<|/det|> + +
FeatureCountsLiao et al., 2014featureCounts in
usegalaxy.eu
Galaxy serverAfgan et al., 2018https://usegalaxy.eu/
gecece utility to calculate fractional GC content of
nucleic acid sequences		emboss (version
5.0.0)		gecece in
usegalaxy.eu
GREAT: Genomic Regions Enrichment of Annotations
Tool, version 3.0.0		Hiller et al., 2013http://great.stanford.
edu/great/public-
3.0.0/html/
In-vitro nucleosome prediction programKaplan et al., 2009
and
https://github.com/bgr
uening/galaxytools		Nucleosome
Predictions in
usegalaxy.eu
MACS2Ferg et al., 2007MACS2 callpeak and
MACS2 bdgepackall
in
usegalaxy.eu
R: Version 3.6.1R Core Team (2019).
R: A language and
environment for
statistical computing.
R Foundation for
Statistical Computing,
Vienna, Austria.		https://www.R-
project.org/.
RNA StarDobin et al., 2013RNA Star in
usegalaxy.eu
RNA-senseGao et al., 2022https://bioconductor.
org/packages/releas
e/bioc/html/RNAsens
e.html
+ +<|ref|>title<|/ref|><|det|>[[115, 554, 380, 567]]<|/det|> +# RESOURCE AVAILABILITY + +<|ref|>title<|/ref|><|det|>[[115, 592, 247, 605]]<|/det|> +# Lead contact + +<|ref|>text<|/ref|><|det|>[[115, 611, 880, 644]]<|/det|> +Requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de). + +<|ref|>title<|/ref|><|det|>[[115, 668, 317, 681]]<|/det|> +# Materials availability + +<|ref|>text<|/ref|><|det|>[[115, 687, 880, 740]]<|/det|> +Zebrafish lines generated in this study are available from lead contact, Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de) under the restrictions of German Animal Protection law (TierSchutzGesetz). + +<|ref|>title<|/ref|><|det|>[[115, 763, 368, 776]]<|/det|> +# Data and code availability + +<|ref|>text<|/ref|><|det|>[[146, 783, 880, 834]]<|/det|> +·Raw and processed RNA-seq and ATAC-seq data generated in this study have been deposited in GEO and are publicly available as of the date of publication.Accession numbers are listed in the key resources table. + +<|ref|>text<|/ref|><|det|>[[146, 840, 880, 873]]<|/det|> +·The original code is deposited in https://github.com/vandensich/zebrafish-minimodels and will be made available to the date of publication. + +<|ref|>text<|/ref|><|det|>[[146, 879, 880, 930]]<|/det|> +·Any additional information required to reanalyze the data reported in this paper is available from the lead contact Daria Onichtchouk (daria.onichtchouk@biologie.uni-freiburg.de) upon request. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 116, 595, 135]]<|/det|> +## EXPERIMENTAL MODEL AND SUBJECT DETAILS + +<|ref|>sub_title<|/ref|><|det|>[[116, 150, 800, 168]]<|/det|> +## Zebrafish maintenance and generation of the double and triple mutants + +<|ref|>text<|/ref|><|det|>[[115, 170, 883, 640]]<|/det|> +Zebrafish maintenance and generation of the double and triple mutantsWild- type fish of AB/TL and mutant strains were raised, maintained and crossed under standard conditions as described by Westerfield \(^{34}\) . Embryos were obtained by natural crossing (4 males and 4 females in 1,7 l breeding tanks, Techniplast). Wild- type and mutant embryos from natural crosses were collected in parallel in 10- 15 minutes intervals and raised in egg water at \(28.5^{\circ}C\) until the desired stage. Staging was performed following the Kimmel staging series \(^{35}\) . Stages of the mutant embryos were indirectly determined by observation of wild- type embryos born at the same time and incubated under identical conditions. All experiments were performed in accordance with German Animal Protection Law (TierSchG) and European Convention on the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (Strasburg, 1986). The generation of double and triple mutants for Pou5f3, Sox19b and Nanog was approved by the Ethics Committee for Animal Research of the Koltzov Institute of Developmental Biology RAS, protocol 26 from 14.02.2019. Maternal- Zygotic (MZ) homozygous mutant embryos MZsn were obtained in three subsequent crossings. First, MZnanog \(^{m1435}\) \(^{19}\) homozygous males were crossed with MZsox19b \(^{m143417}\) homozygous females. The double heterozygous fish were raised to sexual maturity and incrossed. The progeny developed into phenotypically normal and fertile adults. Genomic DNA from tail fin biopsies was isolated and used for genotyping. We first selected sox19b homozygous mutants, by PCR- with Sox19b- f1/Sox19b- r1 primers followed by restriction digest with Bbs1 \(^{17}\) . To select the double homozygous fish, we used the genomic DNA from sox19b homozygous mutants for by PCR with Nanog1- f1/Nanog1- r1 primers followed by restriction digest with Ndel. MZsn embryos for experiments were obtained from incrosses of double homozygous sox19b- /- ; nanog- /- fish. The line was maintained by crossing sox19b- /- ; nanog- /- ; males with sox19b- /- ; nanog+/- females. + +<|ref|>text<|/ref|><|det|>[[115, 658, 883, 920]]<|/det|> +MZpn homozygous mutant embryos were obtained in three subsequent crossings. First, MZnanog \(^{m1435}\) homozygous males \(^{19}\) were crossed with MZspg \(^{m793}\) \(^{18}\) homozygous females. The double heterozygous fish were raised to sexual maturity and incrossed. To bypass the early requirement for Pou5f3 in the spg793 homozygous mutants, one- cell stage embryos were microinjected with 50- 100 pg synthetic Pou5f3 mRNA as previously described \(^{18}\) . The fish were raised to sexual maturity, genomic DNA from tail fin biopsies was isolated and used for genotyping. We first selected nanog homozygous mutants, by PCR with Nanog1- f1/Nanog1- r1 primers followed by restriction digest with Ndel. To select the double homozygous fish, we used the genomic DNA from nanog homozygous mutants to PCR- amplify the region, flanking the spg \(^{m793}\) allele. Spg \(^{m793}\) allele carries an A- >G point mutation in the splice acceptor site of the first intron of Pou5f3 gene, which results in the frameshift starting at the beginning of the second exon, prior to the DNA- binding domain. Spg \(^{m793}\) is considered to be null allele. We used the following PCR primers: spg- f1 \(5^{\prime}\) - + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 881, 213]]<|/det|> +GTCGTCTGACTGAACATTTTGC - 3' and spg-r1 5'- GCAGTGATTCTGAGGAAGAGGT - 3'. Sanger sequencing of the PCR products was performed using commercial service (Sigma). The sequencing traces were examined and the fish carrying A to G mutation were selected. MZpn embryos for experiments were obtained from incrosses of double homozygous spg- /- ; nanog- /- fish. The line was maintained by crossing spg- /- ; nanog- /- males with spg- /- ; nanog+/- females and microinjecting Pou5f3 mRNA in each generation. + +<|ref|>text<|/ref|><|det|>[[115, 232, 881, 383]]<|/det|> +To obtain the triple Maternal- Zygotic (MZ) homozygous mutant embryos MZtriple, MZsn double homozygous males were crossed with MZps double homozygous females 17. The sox19b- /- ; spg+/- ; nanog+/- progeny was incrossed, microinjected with 50- 100 pg synthetic Pou5f3 mRNA and genotyped for spg and nanog as described above. MZtriple embryos for experiments were obtained from incrosses of triple homozygous fish. The line was maintained by crossing sox19b- /- ; spg- /- ; nanog- /- males with sox19b- /- ; spg- /- ; nanog+/- females and microinjecting Pou5f3 mRNA in each generation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 402, 579, 420]]<|/det|> +## Genomic DNA isolation and PCR for genotyping + +<|ref|>text<|/ref|><|det|>[[115, 421, 881, 590]]<|/det|> +Genomic DNA was isolated from individual tail fin biopsies of 3 months old fish. Tail fin biopsies or embryos were lysed in \(50\mu \mathrm{l}\) lysis buffer ( \(10\mathrm{mM}\) Tris pH 8, \(50\mathrm{mM}\) KCl, \(0.3\%\) Tween20, \(0.3\%\) NP- 40, \(1\mathrm{mM}\) EDTA) and incubated at \(98^{\circ}\mathrm{C}\) for \(10\mathrm{min}\) . After cooling down Proteinase K solution ( \(20\mathrm{mg / ml}\) , A3830, AppliChem) was added and incubated overnight at \(55^{\circ}\mathrm{C}\) . The Proteinase K was destroyed by heating up to \(98^{\circ}\mathrm{C}\) for \(10\mathrm{min}\) . The tail fin biopsies material was diluted \(20\times\) with sterile water. \(2\mu \mathrm{l}\) of was used as a template for PCR. PCR was performed in \(25 - 50\mu \mathrm{l}\) volume, using MyTag polymerase (Bioline GmbH, Germany) according to the manufacturer instructions, with 30- 35 amplification cycles. The primer sequences are listed in the key resources table. + +<|ref|>sub_title<|/ref|><|det|>[[117, 606, 298, 624]]<|/det|> +## METHOD DETAILS + +<|ref|>sub_title<|/ref|><|det|>[[117, 640, 441, 658]]<|/det|> +## ATAC-seq and library preparation + +<|ref|>text<|/ref|><|det|>[[115, 659, 881, 920]]<|/det|> +Omni- ATAC- seq was performed as described in 17. Briefly, embryos were enzymatically dechorionated with Pronase E ( \(30\mathrm{mg / ml}\) ). 30 embryos were deyloked manually with an eyebrow needle in Danieaus medium, centrifuged ( \(500\times \mathrm{g}\) , \(5\mathrm{min}\) , \(4^{\circ}\mathrm{C}\) ), washed with \(100\mu \mathrm{l}\) ice- cold PBS, and then centrifuged again. The cells were resuspended by pipetting in \(50\mu \mathrm{l}\) ice- cold lysis buffer (ATAC- RSB ( \(10\mathrm{mM}\) TrisHCl pH 7.4, \(10\mathrm{mM}\) NaCl, \(3\mathrm{mM}\) MgCl2), with \(0.1\%\) NP40, \(0.1\%\) Tween20, and \(0.01\%\) digitonin) and incubated on ice for 5 minutes. Then, the lysate was diluted with ice- cold dilution buffer ( \(1\mathrm{ml}\) , ATAC- RSB containing \(0.1\%\) Tween- 20). Nuclei were pelleted ( \(500\times \mathrm{g}\) , \(10\mathrm{min}\) , \(4^{\circ}\mathrm{C}\) ), the supernatant carefully removed, and the nuclei resuspended in \(16.5\mu \mathrm{l}\) PBST. Tagnentation reaction was assembled according to the manufacturer instructions for Illumina small Tn5 buffer and enzyme kit in \(50\mu \mathrm{l}\) volume and incubated in a thermomixer for \(30\mathrm{min}\) ( \(37^{\circ}\mathrm{C}\) , \(800\mathrm{rpm}\) ). The sample was purified using Qiagen PCR MinElute Purification kit. The purified transposed mix was pre- amplified for 5 cycles ( \(72^{\circ}\mathrm{C}\) for \(5\mathrm{min}\) , \(98^{\circ}\mathrm{C}\) for \(30\mathrm{sec}\) , 5 cycles of \(98^{\circ}\mathrm{C}\) for \(10\mathrm{sec}\) , \(63^{\circ}\mathrm{C}\) for \(30\mathrm{sec}\) , + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 326]]<|/det|> +\(72^{\circ}C\) for 1 min) using NEBNext 2x Master Mix (NEB) and Nextera adapters and put on ice. The number of additional cycles was determined using \(5\mu l\) of the pre- amplified mixture as in \(^{36}\) . The total number of PCR cycles was 10- 12. The amplified libraries were purified using SPRI beads (Beckmann) according to the manufacturer instructions. ATAC- seq was performed at three developmental stages, 3 hpf (1k), 3.7 hpf (oblong) and 4.3 hpf (dome). At 3 hpf, three biological replicates for WT and MZtriple were used. At 3.7 hpf, two biological replicates for MZnanog and MZpn and three biological replicates for MZtriple were used. At 4.3 hpf, three biological replicates for MZsn and MZtriple and two for MZpn and MZnanog were used. Paired- end 150 bp sequencing of resulting 23 ATAC- seq libraries was performed on NovaSeq6000 (Illumina) by Novogene company, with the sequencing depth 80 million reads per library. The raw and processed ATAC- seq data are deposited in GEO (NCBI), accession number GSE215956. + +<|ref|>sub_title<|/ref|><|det|>[[117, 344, 376, 362]]<|/det|> +## ATAC-seq data processing + +<|ref|>text<|/ref|><|det|>[[115, 363, 881, 644]]<|/det|> +ATAC- seq data processingATAC- seq data processing was performed as in (Gao et al., 2022) on the european Galaxy server \(^{37}\) . Adapters were removed, the reads were cropped to 30 bp from start, the reads with average quality less than 30 and length less than 25 bp were removed using Trimmomatics. The trimmed reads were aligned to the danner11/ GRCz11 genome assembly without alternative contigs using Bowtie2 \(^{38}\) , with the parameters: '—dovetail, --very- sensitive, --no- unal'. The aligned reads were filtered using BamTools Filter \(^{39}\) with the parameters - isProperPair true, - mapQuality \(>=30\) , - isDuplicate false, - reference lchrM. The aligned reads were filtered further using the alignmentsieve tool from the deepTools2 suite \(^{40}\) , with the parameters —ATACshift, --minMappingQuality 1, --maxFragmentLength 110. Thus, only fragments with a size of 110 bp or less were kept. The filtering steps were performed for the individual replicates as well as and for the merged biological replicates BAM files. Bigwig files for each stage and genotype were obtained from merged BAM files using Bamcoverage and used for data visualization in deepTools2. Bigwig and peak files for each stage and genotype are included in GEO GSE215956 submission as processed files. + +<|ref|>sub_title<|/ref|><|det|>[[117, 663, 490, 681]]<|/det|> +## Definition of regulated ARs and TdARs + +<|ref|>text<|/ref|><|det|>[[115, 682, 881, 927]]<|/det|> +The ATAC- seq data from this work at 3.7 hpf and 4.3 hpf were processed together with our previously published ATAC- seq data set: three biological replicates for the wild- type and two biological replicates for MZspg, MZsox19b and MZps at 3.7 hpf, three biological replicates for the wild- type, MZspg and MZsox19b and four for MZps at 4.3 hpf \(^{17}\) , GEO accession number GSE188364. For seven wild- type biological replicates, filtered reads were converted from BAM to BED using Bedtools \(^{41}\) and were used as inputs for peak calling. Peaks were called from individual replicates and merged replicates using MACS2 callpeak \(^{42}\) with the additional parameters '--format BED, --nomodel, --extsize 200, --shift - 100' and a 5% FDR cutoff. Regions mapping to unassembled contigs were excluded. Only peaks that were overlapping in merged and each of the single replicates were kept. ATAC- seq peaks overlapping between 3.7 and 4.3 hpf stages in the wild- type were centered on ATAC- seq peak summits (4.3 hpf) and cut to 110 bp length. This set of 102 965 regions accessible in the wild- type (ARs) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 883, 364]]<|/det|> +was used in all subsequent analyses. The coverage of ATAC- seq reads on ARs was scored using MultiCovBed (Bedtools) in all ATAC- seq experiments (3.7 and 4.3 hpf). Deseq2 \(^{43}\) was used to normalize the reads and compare chromatin accessibility in each mutant to the wild- type. For each mutant genotype, we made four Deseq2 cross- normalizations: across all replicates, across the replicates of 3.7 hpf or 4.3 hpf separately, and across three samples (i.e. wild- type, MZnanog and MZtriple). The region was scored as downregulated in the mutant, if the fold change to the wild- type was negative with FDR \(5\%\) in four Deseq2 normalizations. The region was scored as upregulated in the mutant, if the fold change to the wild- type was positive with FDR \(5\%\) in four Deseq2 normalizations. The remaining ARs were considered unchanged. To define TdARs, (TF- bound Accessible Regions, downregulated in MZtriple), we selected ARs, downregulated in MZtriple and overlapping with any of Sox2, Pou5f3 \(^{12}\) or Nanog \(^{23}\) ChIP- seq peaks for at least 1 bp. The genomic coordinates of TF- binding peaks remapped to danrer11/GRCz11 assembly were taken from \(^{17}\) . Genomic coordinates of ARs and their properties are provided in the Table S1. + +<|ref|>sub_title<|/ref|><|det|>[[117, 402, 399, 420]]<|/det|> +## Scoring the motif enrichment + +<|ref|>text<|/ref|><|det|>[[115, 421, 883, 590]]<|/det|> +Sequence- specific nanog1 and nanog2 motif matrices were taken from \(^{22}\) , the rest of the motif matrices from \(^{17}\) . Genomic coordinates of all occurrences for each motif in the 3 kb regions around ATAC- seq peaks were obtained using FIMO \(^{44}\) with p- value threshold \(10^{- 4}\) and converted to BED files. To calculate the background frequency of the motifs, we used the background control peak file, where the genomic coordinates of all ATAC- seq summits were shifted 1 kb downstream. The motif densities were calculated as the average number of motifs overlapping with 60 bp around the ATAC- seq or control peak summits. The log2 ratio of ATAC- seq to background motif densities was taken as a motif enrichment value. + +<|ref|>sub_title<|/ref|><|det|>[[117, 607, 580, 625]]<|/det|> +## GC content and in-vitro nucleosome predictions + +<|ref|>text<|/ref|><|det|>[[116, 626, 883, 757]]<|/det|> +GC content for 110 bp around AR summits was calculated with geece program, in- vitro nucleosome prediction with the algorithm of Kaplan et al. \(^{33}\) in the european Galaxy instance usegalaxy.org. The genomic coordinates of ARs and control peaks were extended to +/- 5 kb to account for the edge effects, and in- vitro nucleosome prediction value was derived for every base pair. Nucleosome predictions were converted to bedgraph files for scoring. GC content and nucleosome prediction values per 110 bp around ARs are provided in the Supplementary Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[116, 777, 553, 795]]<|/det|> +## Scoring the rescue of chromatin accessibility + +<|ref|>text<|/ref|><|det|>[[116, 796, 883, 927]]<|/det|> +Track hub with Omni- ATAC rescue processed data in UCSC bigwig format was downloaded from the home page of Antonio Giraldez lab (https://www.giraldezlab.org/data/miao_et_al_2022_molecular_cell/) and used for scoring. We calculated ATAC- seq signals in all conditions for 110 bp around ARs. For ARs, downregulated in MZnpcs compared to the WT B1 replicate in \(^{5}\) , we estimated the rescue. We considered AR as rescued by TF(s) microinjection, if the log2 of ATAC- seq signals (injected MZnpcs/control MZnpcs) was exceeding log2 of ATAC- seq signals + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 81, 881, 138]]<|/det|> +(control WT B1/injected MZnps). The rescue status by all combinations of TFs for ARs downregulated in MZnps is provided in the Table S1; the statistics for rescue on TdARs is provided on the Fig. S3. + +<|ref|>sub_title<|/ref|><|det|>[[116, 158, 450, 176]]<|/det|> +## ChIP-seq for H3K27ac in MZnanog + +<|ref|>text<|/ref|><|det|>[[115, 178, 881, 552]]<|/det|> +The ChIP- seq was performed as in 17. Embryos were obtained from natural crossings in mass- crossing cages (4 males + 4 females). The freshly laid eggs of MZnanog mutants and wild- type were collected in 10- 15 min intervals. The embryos were incubated at \(28.5^{\circ}C\) and enzymatically dehorionated with \(30mg / ml\) Pronase shortly before 4.3 hpf. Embryos were homogenized in \(10ml0.5\%\) Danieau's (for 1L of 30X stock: \(1740mMNaCl\) , \(21mM KCl\) , \(12mM MgSO4\) , \(18mM Ca(NO3)2\) , \(150mM HEPES\) buffer, pH 7.6) containing 1x protease inhibitor cocktail (PIC, Roche) using a Dounce tissue grinder, and immediately treated with \(1\%\) (v/v) Methanol- free Formaldehyde (Pierce) for exactly 10 min shaking at room temperature. The fixation was stopped with 0.125 M Glycine by shaking for 5 min on a rotating platform. Lysate was centrifuged for 5 min, \(4700rpm\) at \(4^{\circ}C\) , the pellet was resolved in cell lysis buffer ( \(10mM\) Tris- HCl (pH 7.5), \(10mMNaCl\) , \(0.5\%\) NP- 40, 1- 4 ml/1000 embryos), followed by 5 min incubation on ice, and centrifuged 1 min, \(2700g\) at \(4^{\circ}C\) . The pellet was washed 2 times with 1 ml ice cold 1x PBST (for 1 L: \(40ml\) PO4 buffer (0.5 M), \(8gNaCl\) , \(0.2gKCl\) , \(0.1\%\) Tween20, pH 7.5) and resuspended in 1 ml PBST. 5 \(\mu l\) aliquote of nuclei suspension was stained with Sytox green, nuclei were scored under fluorescence microscope using the Neubauer counting chamber. The residual nuclei were pelleted by 1 min centrifugation at \(2700g\) at \(4^{\circ}C\) , snap frozen in liquid nitrogen and stored at - \(80^{\circ}C\) . The nuclei collected in different days were pooled together to reach the total number of 2.5- 3 million, which was used to start one ChIP experiment. + +<|ref|>text<|/ref|><|det|>[[116, 553, 881, 683]]<|/det|> +The chromatin was thawn, pooled in 2 ml of nuclei lysis buffer ( \(50mM\) Tris- HCl (pH 7.5), \(10mM\) EDTA, \(1\%\) SDS) and incubated 1 h on ice. In order to shear the chromatin to 200 bp fragments (on average), the chromatin was sonicated using the Covaris S2 sonicator (DC \(20\%\) , Intensity 5, Cycles of Burst 200, Time \(= 3^{*}40\) cycles with 30 sec each ( \(3^{*}20\) min)). To control the sonication, \(30\mu l\) of the sheared chromatin was de- crosslinked with \(250mMNaCl\) overnight at \(65^{\circ}C\) and then analyzed using the Agilent Expert 2100 Bioanalyzer® and Agilent high sensitivity DNA Chip kit. + +<|ref|>text<|/ref|><|det|>[[116, 701, 881, 925]]<|/det|> +The lysed and sheared samples were centrifuged for 10 min, \(4^{\circ}C\) at max. speed in table top centrifuge. \(60\mu l\) of each sample were kept as input control. The chromatin was then concentrated to \(100\mu l\) using the Microcon centrifugal filters (Merck Millipore MRCF0R030) and diluted 1:3 by adding ChIP dilution buffer ( \(16.7mM\) Tris- HCl pH 7.5, \(167.0mMNaCl\) , \(1.2mM\) EDTA) containing protease inhibitors. The H3K27ac antibody was added and incubated overnight at \(4^{\circ}C\) on a rotating wheel. \(150\mu l\) of magnetic Dynabeads coupled to protein G (Stock \(30mg / ml\) ; invitrogen Dynal 10003D) were transferred into a 2 ml eppendorf tube and placed on a magnetic rack. The beads were washed 3x with \(5mg / ml\) specially purified BSA in PBST and 1x with \(500\mu l\) ChIP dilution buffer. After removing the ChIP dilution buffer, the chromatin- antibody mix was added and incubated with the beads at \(4^{\circ}C\) overnight on a rotating wheel. Beads were pulled down by placing the eppendorf tubes on the magnetic rack. The beads were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 881, 326]]<|/det|> +resuspended in \(333\mu \mathrm{l}\) RIPA buffer ( \(10\mathrm{mM}\) Tris- HCl (pH 7.6), \(1\mathrm{mM}\) EDTA, \(1\%\) NP- 40, \(0.7\%\) sodium deoxycholate, \(0.5\mathrm{M}\) LiCl) containing PIC. The Protein G- antibodychromatin complex was washed \(4\times 5\) min on a rotating platform with \(1\mathrm{ml}\) of RIPA buffer, followed by \(1\times 1\) ml TBST buffer ( \(25\mathrm{mM}\) Tris- HCl, \(150\mathrm{mM}\) NaCl, \(0.05\%\) Tween 20, pH 7.6). The beads were pulled down again. To elute the chromatin, \(260\mu \mathrm{l}\) elution buffer ( \(0.1\mathrm{M}\) NaHCO3, \(1\%\) SDS) was added and incubated for \(1\mathrm{h}\) at \(65^{\circ}\mathrm{C}\) in a water bath. The samples were vortexed every 10 - 15 min. The beads were pulled down, supernatant was transferred to a fresh eppendorf tube. \(12.5\mu \mathrm{l}5\mathrm{M}\) NaCl was added to de- crosslink the chromatin and incubated overnight at \(65^{\circ}\mathrm{C}\) in a water bath. The input samples were treated in parallel ( \(230\mu \mathrm{l}\) elution buffer per \(30\mu \mathrm{l}\) input). Purification of the de- crosslinked chromatin was performed using the QIAquick PCR Purification Kit from Qiagen. The concentration was determined using the Qubit fluorometer and Quanti- iT™ PicroGreen® dsDNA Kit according to manufacturer instructions. + +<|ref|>text<|/ref|><|det|>[[115, 343, 881, 569]]<|/det|> +To estimate the signal to background ratio in each ChIP experiment, we used the positive and negative reference genomic regions, near tiparp and igsf2 genes, respectively enriched in or devoid of H3K27ac (Gao et al., 2022). We performed quantitative PCR in ChIP and Input control material, using the primers for these regions. PCR primers used were: tiparp_f_1 5' CGCTCCCAACTCCATGTATC- 3', tiparp_r_1 5'- AACGCAAGCCAAACGATCTC- 3', igsf2_f_2 5'- GAACTGCATTAGAGACCCAC- 3', igsf2_r_2 5'- CAATCAACTGGGAAAGCATGA- 3'. QPCR was carried out using the SsoAdvanced™ Universal SYBR® Green Supermix from BIO- RAD. ChIP and input were normalized by ddCT method, using negative reference region (igsf2). The ChIP experiment was considered successful, if the enrichment in ChIP over input control on the positive reference region (tiparp) was more than 5- fold. + +<|ref|>text<|/ref|><|det|>[[115, 585, 881, 923]]<|/det|> +The library preparation was carried out with NEBNext® Ultra™ DNA Library Prep Kit according to manufacturer instructions, with the modifications indicated below. As the DNA outcome of individual MZnanog K27ac ChIP experiments did not reach the input DNA limit for the kit (5 ng), we pooled together the material from two successful ChIP experiments, as well as corresponding inputs. Since the DNA input was \(< 100\) ng, in the adaptor ligation step, the NEBNext Adaptor for Illumina® ( \(15\mu \mathrm{M}\) ) was diluted 10- fold in \(10\mathrm{mM}\) Tris- HCl (pH 7.8) to a final concentration of \(1.5\mu \mathrm{M}\) and used immediately. At the final clean- up step, the reaction was purified by mixing the samples with AMPure XP Beads ( \(45\mu \mathrm{l}\) ) before incubating at room temperature for 5 min. After a quick spin using the tabletop centrifuge, samples were placed on a magnetic rack and supernatant was discarded. \(200\mu \mathrm{l}\) of \(80\%\) Ethanol were added and removed after 30 sec, two times. The beads were air dried for 3 min and the DNA target was eluted by adding \(33\mu \mathrm{l}\) of \(0.1\times \mathrm{TE}\) (pH 8) and incubating at room temperature for 2 min. \(28\mu \mathrm{l}\) of the library were transferred to a fresh PCR tube and stored at - 20 °C. \(2\mu \mathrm{l}\) of the sample were diluted 5- fold with \(0.1\times \mathrm{TE}\) and used to check the size distribution of the library using Agilent Expert 2100 Bioanalyzer® and Agilent high sensitivity DNA Chip kit. The ChIP- seq library was sequenced at \(70\mathrm{mln}\) paired end 150bp reads the input library were sequenced to \(30\mathrm{mln}\) reads. Sequencing was performed by the Novogene + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 880, 118]]<|/det|> +company (China). The raw and processed data have been deposited in GEO under the number GSE143439. + +<|ref|>sub_title<|/ref|><|det|>[[117, 140, 439, 157]]<|/det|> +## Definition of regulated enhancers + +<|ref|>text<|/ref|><|det|>[[115, 158, 881, 458]]<|/det|> +Sequenced reads were mapped to the zebrafish danrer11 assembly using Bowtie2 in the european Galaxy server. Alternate loci and unassembled scaffolds were excluded. H3K27ac data for MZnanog from this work was further processed together with H3K27ac for the wild- type, MZspg and MZsox19b17. To create bigwig files for scoring, the log2 ratio between each ChIP and merged inputs (in rpkm) was obtained using BAMcompare program in deepTools, bin size =10. The mean H3K27ac signal (log2 ChIP/input per bin) around each AR was calculated for 1 kb around AR summit. Values smaller than 0.1 were rounded to 0.1. We considered AR as H3K27- acetylated, if the mean H3K27ac signal was higher than arbitrary threshold 0.2. All H3K27- acetylated TdARs were considered as enhancers. Enhancer was considered as activated by TF, if log2 (mutant by TF/WT) ratio of H3K27ac signals was less than arbitrary threshold - 0.7, was considered as blocked by TF, if log2 (mutant by TF/WT) ratio of H3K27ac signals was more than 0.7, and was considered as unchanged by TF all other cases. Enhancers activated, blocked or unchanged by the sum of TFs were assigned by applying the same procedure and thresholds to H3K27ac ChIP- seq data on the WT and MZnps5. H3K27ac status of all ARs is provided in the Supplementary Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[117, 479, 321, 496]]<|/det|> +## RNA-seq time curves + +<|ref|>text<|/ref|><|det|>[[115, 499, 881, 930]]<|/det|> +RNA- seq time curves were obtained as described previously (Gao et al., 2022). Freshly laid eggs were obtained from natural crossings in mass- crossing cages (4 males + 4 females). 5- 10 cages were set up per genotype, the eggs from different cages were pooled. At least 600 freshly laid eggs per each genotype collected within 10- 15 minutes were taken for single experiment. To match the developmental curves as precisely as possible, the material was simultaneously collected for two genotypes in parallel. 45- 60 minutes after the egg collection, we ensured that the embryos of both genotypes are at 2- 4 cell stage, removed non- fertilized eggs and distributed the embryos to 8 dishes per genotype, 40- 45 embryos per dish, to obtain 8 time points per genotype: 2.5, 3, 3.5, 4, 4.5, 5, 5.5 and 6 hpf. The temperature was kept at 28.5°C throughout the experiment. The embryos from one dish per genotype were snap- frozen in liquid nitrogen every 30 minutes, starting from 2.5 hpf (pre- ZGA, 256 cell stage, 8th cell cycle) till midgastrula (6 hpf). Two biological replicates for each of MZnanog and MZsn were previously collected in parallel to the wild- type replicates B1 and B2 from (Gao et al., 2022). In addition, we collected six biological replicates of the time curve for the wild- type, and three biological replicates for MZpn and MZtriple mutants from six independent experiments. Total RNA was isolated with QIAGEN RNeasy kit following the user's manual. We checked RNA quantity and quality by using Agilent RNA 6000 nano Kit on Agilent Bioanalyzer, according to manufacturer instructions. Poly- A enriched library preparation and sequencing on Illumina platform was performed by Novogene company (China). The MZnanog and MZsn samples were single- end sequenced at 35 million reads per sample (50 bp), similarly to the previously published samples WT, MZsox19b, MZspg and MZps (Gao et al., 2022). The MZpn + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[116, 82, 880, 119]]<|/det|> +and MZtriple samples and corresponding six replicates of the wild- type were paired- end sequenced at 58 million reads per sample (150 bp). + +<|ref|>sub_title<|/ref|><|det|>[[116, 140, 488, 157]]<|/det|> +## RNA-seq time curves: data processing + +<|ref|>text<|/ref|><|det|>[[115, 158, 882, 383]]<|/det|> +FASTQ files were processed on european Galaxy server usegalaxy.eu. All sequenced clean reads data were trimmed on 5' end for 5 bp by Trim Galore! program according to the sequencing quality. Trimmed reads were mapped to danner11 genome assembly using RNA STAR \(^{45}\) and ENSEMBL gene models. Number of reads for each gene was counted using Feature count \(^{46}\) . Single- end and paired- end sequencing data sets were further processed separately as the part 1 and the part 2. For the part 1, feature counts for the MZnanog and MZsn samples were cross- normalized with MZsox19b, MZspg and MZps and four biological replicates of the wild- type (Gao et al., 2022) using Deseq2. For the part 2, feature counts for the MZpn and MZtriple samples and corresponding six replicates of the wild- type were cross- normalized using Deseq2. The raw data and two normalized data tables were deposited in GEO with the accession number GSE162415. + +<|ref|>sub_title<|/ref|><|det|>[[116, 403, 561, 421]]<|/det|> +## Finding differentially expressed zygotic genes + +<|ref|>text<|/ref|><|det|>[[115, 422, 882, 780]]<|/det|> +We used 3- step RNA- sense program to process normalized RNA- seq time curves as described in \(^{17}\) . Shortly, RNA- sense takes RNA- seq time curves in two conditions (WT and mutant) as an input. The transcripts expressed below a user- defined threshold are excluded from the analysis. In the first step, RNA- sense creates the lists of transcripts switching UP or DOWN in time, in one of the genotypes; the genotype and switch p- value are defined by user. We found that 4247 genes are switching UP in the wild- type, in both part1 and part 2 of the analysis (thresholds \(>100\) for expression and \(< 0.15\) for switch p- value). Switching UP transcripts were considered as zygotically expressed \(^{17}\) . 1668 transcripts were switching up in at least one mutant but not in the wild- type; we considered them as zygotically expressed in the respective mutants. The regulation status was assigned to each zygotic transcript on the second step of RNA- sense. Zygotic transcript was considered to be downregulated in the mutant, if it was expressed at least two- fold lower compared to the wild- type with p- value \(< 0.05\) in Student's t- test in at least one time point at or after the switch UP. Zygotic transcript was considered to be upregulated in the mutant, if it was expressed at least two- fold higher compared to the wild- type with p- value \(< 0.05\) in Student's t- test in at least one time point. Some transcripts were up – and down regulated in the same mutant in different time points and were considered as “undefined” (“U”). The list of zygotic genes, their switch and regulation status in all the mutants is provided in the Table S2. + +<|ref|>sub_title<|/ref|><|det|>[[117, 799, 323, 815]]<|/det|> +## Enrichment statistics + +<|ref|>text<|/ref|><|det|>[[116, 816, 881, 927]]<|/det|> +Zygotic transcripts (Table S2) were linked to Chromatin Accessible Regions (ARs, Table S1) within +/- 50 kb from Transcription Start Site (TSS). Resulting 70419 AR- transcript pairs were uniquely numbered (Table S3). To estimate the over- or underrepresentation of different types of enhancers in the putative regulatory regions of zygotic genes, Chi- squared ( \(\chi^{2}\) ) test was performed using chisq.test function and visualized using corrplot function in R. We considered only strong positive correlations + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 883, 365]]<|/det|> +with Pearson residuals more than 4. To reproduce our results, the input files for all \(\chi^2\) tests can be easily derived from the supplementary table 3 as explained below. The input file is a tab- separated text file with three columns: "AR- transcript_pair" (which is always column A in the Table S3), "AR", and "transcript". To create the input file for \(\chi^2\) test, delete the first two rows of the Supplementary Table 3, keep column A and two columns specified below (delete all other columns), delete the rows specified below, change the headings to AR" and "transcript" and export the input file as tab- separated text. Fig. 2G: "AR" - column N, "transcript" - column AF, delete all rows with "U". Fig. 5D- F: "AR" - columns O, Q, R, respectively, delete all rows with "none"; "transcript" - column AG, delete all rows with "N/A". Fig. 6C: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column Z, delete all rows with "U". Fig. 6C: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column AB, delete all rows with "U". Fig. S7B, C: "AR" - columns G, P, respectively, delete all rows with "none"; "transcript" - column AG, delete all rows with "N/A". Fig. S7F: "AR" - columns G, Q, R, delete all rows with "none"; "transcript" - column AC, delete all rows with "U". + +<|ref|>sub_title<|/ref|><|det|>[[117, 384, 352, 402]]<|/det|> +## Gene Ontology analysis. + +<|ref|>text<|/ref|><|det|>[[115, 402, 883, 515]]<|/det|> +For the genomic regions we used GREAT analysis 47 at http://great.stanford.edu/great/public- 3.0.0/html/. Genomic regions were converted to zv9/danrer9 genomic assembly, which is the only assembly available for this server, using Liftover Utility from UCSC and associated with genes using 20 kb single nearest gene association rule. The categories were ranked by ascending FDR value. For the transcript groups we used DAVID 48. + +<|ref|>sub_title<|/ref|><|det|>[[117, 535, 289, 551]]<|/det|> +## Data visualization + +<|ref|>text<|/ref|><|det|>[[115, 553, 883, 798]]<|/det|> +Omni- ATAC- seq: to create bigwig files for data visualization, Omni- ATAC- seq signal (in rpkm) was obtained using BAMcoverage program in deepTools2, bin size \(= 10\) . H3K27ac: to create bigwig files for data visualization, the log2 ratio between each ChIP and merged inputs (in rpkm) was obtained using BAMcompare program in DeepTools, bin size \(= 10\) . The heatmaps or mean profiles were plotted using plotheatmap or plotprofile programs in DeepTools. H3K27ac, TF ChIP- seq and omni- ATAC- seq profiles in single genes were visualized in UCSC browser. RNA- seq heatmaps were plotted using R with custom script. For the heatmaps, the transcripts were sorted by ascending RNA- sense switch p- value then by ascending RNA- sense switch time then by group. Biological replicates from all experiments were cross- normalized using Deseq2. Replicates for each genotype were averaged and the maximal expression value across the time curve in all genotypes was calculated. Expression/max ratio for each timepoint was plotted. + +<|ref|>sub_title<|/ref|><|det|>[[117, 817, 340, 835]]<|/det|> +## Mathematical Modeling + +<|ref|>text<|/ref|><|det|>[[117, 836, 881, 929]]<|/det|> +The mathematical modeling process was performed in two main steps, (i) the core model and (ii) the mini models. The idea behind developing a core model was to use the resulting dynamics of relevant transcription factors or their combinations as input functions to the mini models that later describe the dynamical behavior of a huge amount of 1799 preselected target genes. Generation of ODE models, parameter + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 81, 883, 350]]<|/det|> +estimation and identifiability analysis was performed by means of the R package dMod \(^{49}\) . For each model, the best parameter set was obtained by the method of maximum likelihood performing a deterministic multi- start optimization \(^{50}\) from multiple randomly chosen starting points in parameter space. The trust region optimizer trust (Geyer CJ (2004) trust: Trust Region Optimization. \(R\) package Version 0.1- 8) was used constituting the standard optimizer in the R package dMod. Model parameters were log- transformed to ensure positivity and enable optimization over a broad range of magnitudes \(^{50}\) . Parameter values were not restricted by fixed borders. Instead, in order to prevent the optimizer from finding solutions with very low or high parameter values, we constrained the model parameters with a weak \(L_{2}\) prior that contributed with one to the likelihood, if the parameter differed by three orders of magnitude from a value of zero on the logarithmic scale. For the computation of confidence intervals and to test identifiability of the model's parameters by means of the profile likelihood \(^{51}\) the contribution of these \(L_{2}\) priors was subtracted to ensure a purely data- based result. + +<|ref|>sub_title<|/ref|><|det|>[[117, 373, 418, 391]]<|/det|> +## Construction of the core model + +<|ref|>text<|/ref|><|det|>[[115, 392, 883, 711]]<|/det|> +In total, the core model consists of 11 model states and 17 reactions (Fig. S5A, Table S4). Kinetic expressions of the model reactions were derived based on state- of- the- art transcription factor network modeling \(^{52}\) and prior knowledge from published literature \(^{14,17,19,20}\) . For simplicity, the degradation of Sox19b was incorporated via a switch- like input function dependent on the time t as a parameter. Model states were assumed to be directly observed, and uncertainty of the observations were estimated jointly with the kinetic parameters assuming an absolute error model for the measurements of each observed state. Model equations, as summarized in Table S4, include three boolean variables MZspg, MZnanog and MZsox that were used to distinguish between the seven experimental conditions including wild type, single, double and triple knockdowns. In total, a number of 888 RNA- seq data points for six targets was used for model calibration. The first time point of the data set was at 2.5 hpf. As an assumption, the initial values of the ODE model were defined at 2 hpf. The initial values of downstream targets Sox2, Sox3 and Sox19a were assumed to be zero, while for the values of the zygotic transcription factors Pou5f3, Sox19b and Nanog a free parameter was estimated by the model, with the constraint of being the same over all experimental conditions. + +<|ref|>text<|/ref|><|det|>[[115, 730, 883, 918]]<|/det|> +During the model development, a couple of parameters was fixed to zero or one, following likelihood- ratio test based model reduction \(^{53}\) since they were not necessary in order to describe the available data. Hill- kinetic parameters were fixed to a value of five corresponding to a strong threshold behavior. Further parameter transformations that have been applied are summarized in https://github.com/vandensich/zebrafish- minimodels. Parameter estimation was performed using deterministic multi- start optimization \(^{50}\) starting from 192 randomly chosen sampled positions in parameter space. Parameter values of the best obtained fit are summarized in Table S5. To show reliability of the optimization, the 100 best optimization runs were displayed as a waterfall plot (Fig. S5B) sorted by their negative log- likelihood values \(^{50}\) . The global + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 81, 881, 139]]<|/det|> +optimum was found in 38 of the 101 converged optimizations runs. Considering a confidence level of \(95\%\) , profile likelihoods showed finite confidence intervals for 30 out of 37 parameters (see Fig. S6A). + +<|ref|>sub_title<|/ref|><|det|>[[117, 157, 428, 175]]<|/det|> +## Construction of the mini models + +<|ref|>text<|/ref|><|det|>[[115, 177, 881, 345]]<|/det|> +Assuming at least one of Pou5f3, Nanog and SOX to be activating, a set of 19 mini models was constructed. Model equations for each mini model (Table S6) were derived from the theory of transcription factor network modeling \(^{52}\) . When more than one transcription factor was either activating or repressing, the combination was modeled using appropriate logical AND- gate functions \(^{53}\) . Dynamics of the proteins Pou5f3, Nanog and the sum of SoxB1 class genes SOX were taken from the core model for the different experimental conditions and used as input functions for the mini models. To avoid numerical issues with infinite derivatives of hill- kinetic terms, a small number of 1e- 4 was added to the input functions where necessary. + +<|ref|>text<|/ref|><|det|>[[115, 364, 881, 496]]<|/det|> +Analogously to the core model, RNA- seq data of seven experimental conditions was used for model fitting of each mini model and for each target gene. Per target gene, a total number of 192 was available. In contrast to the core model, the different experimental conditions were not incorporated via boolean variables, instead the differential equations stayed the same over all conditions. Thus, differences in dynamics were only possible via the differences in the given input dynamics of the core model. + +<|ref|>text<|/ref|><|det|>[[115, 497, 881, 591]]<|/det|> +For each mini model and for each target gene, a total number of 64 optimization runs was performed out of which \(99.97\%\) converged and at least \(98.67\%\) reached the global optimum. Finally, using the Bayesian Information criterion (BIC), out of the 19 analyzed mini models, the best fitting one was selected for every target. Parameter values of the selected models are summarized in Table S7. + +<|ref|>sub_title<|/ref|><|det|>[[115, 609, 584, 627]]<|/det|> +## QUANTIFICATION AND STATISTICAL ANALYSIS + +<|ref|>text<|/ref|><|det|>[[117, 629, 881, 680]]<|/det|> +Statistical comparisons of three or more samples were performed using one- way ANOVA, followed by Tukey- Kramer Test to evaluate the statistical significance of pairwise differences. Two samples were compared using Student 2- tailed t- test. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 113, 411, 138]]<|/det|> +## Supplemental items. + +<|ref|>text<|/ref|><|det|>[[116, 161, 483, 180]]<|/det|> +Supplemental figures and legends.pdf + +<|ref|>text<|/ref|><|det|>[[116, 202, 845, 237]]<|/det|> +Table S1. Accessible_regions.xlsx Related to main Fig. 1, Fig. 2, Fig. 3, Fig. 4, Fig. S3, Fig. S4, the legend in the file + +<|ref|>text<|/ref|><|det|>[[116, 252, 863, 286]]<|/det|> +Table S2. Zygotic_transcripts.xlsx Related to main Fig. 2, Fig. 5, Fig. 6, Fig. S2, Fig. S6, Fig. S7, the legend in the file. + +<|ref|>text<|/ref|><|det|>[[116, 300, 780, 335]]<|/det|> +Table S3. ARs_linked_to_transcripts.xlsx Related to main Fig. 2, Fig. 5, Fig. 6, Fig. 7, Fig. S7, the legend in the file. + +<|ref|>text<|/ref|><|det|>[[116, 350, 465, 384]]<|/det|> +Table S4. Coremodel_equations.csv Related to Fig.S5 and Fig.S6 + +<|ref|>text<|/ref|><|det|>[[116, 400, 545, 434]]<|/det|> +Table S5. Coremodel_Bestfitparameters.csv. Related to Fig.S5 and Fig.S6. + +<|ref|>text<|/ref|><|det|>[[116, 449, 512, 483]]<|/det|> +Table S6. Minimodel_equations.csv Related to main Fig. 5, Fig. S6 and Fig. S7. + +<|ref|>text<|/ref|><|det|>[[116, 498, 508, 532]]<|/det|> +Table S7. Minimodel_parameters.csv Related to main Fig. 5, Fig. S6 and Fig. S7 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 108, 281, 132]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[110, 160, 875, 920]]<|/det|> +1 Vastenhouw, N. L., Cao, W. X. & Lipshitz, H. D. The maternal- to- zygotic transition revisited. Development 146, doi:10.1242/dev.161471 (2019).2 Schulz, K. N. & Harrison, M. M. Mechanisms regulating zygotic genome activation. Nature Reviews Genetics, doi:10.1038/s41576- 018- 0087- x (2018).3 Iwafuchi- Doi, M. & Zaret, K. S. Cell fate control by pioneer transcription factors. Development 143, 1833- 1837, doi:10.1242/dev.133900 (2016).4 Gaskill, M. M., Gibson, T. J., Larson, E. D. & Harrison, M. M. GAF is essential for zygotic genome activation and chromatin accessibility in the early Drosophila embryo. Elife 10, doi:10.7554/eLife.66668 (2021).5 Miao, L. et al. The landscape of pioneer factor activity reveals the mechanisms of chromatin reprogramming and genome activation. Mol Cell 82, 986- 1002 e1009, doi:10.1016/j.molcel.2022.01.024 (2022).6 Fernandez Garcia, M. et al. Structural Features of Transcription Factors Associating with Nucleosome Binding. Mol Cell 75, 921- 932 e926, doi:10.1016/j.molcel.2019.06.009 (2019).7 Gassler, J. et al. Zygotic genome activation by the totipotency pioneer factor Nr5a2. Science, eabn7478, doi:10.1126/science. abn7478 (2022).8 Heyn, P. et al. The earliest transcribed zygotic genes are short, newly evolved, and different across species. Cell reports 6, 285- 292, doi:10.1016/j.celrep.2013.12.030 (2014).9 Neyfakh, A. A. Radiation Investigation of Nucleo- Cytoplasmic Interrelations in Morphogenesis and Biochemical Differentiation. Nature 201, 880- 884 (1964).10 White, R. J. et al. A high- resolution mRNA expression time course of embryonic development in zebrafish. Elife 6, doi:10.7554/eLife.30860 (2017).11 Lee, M. T. et al. Nanog, Pou5f1 and SoxB1 activate zygotic gene expression during the maternal- to- zygotic transition. Nature, 360- 364., doi:10.1038/nature12632 (2013).12 Leichsenring, M., Maes, J., Mossner, R., Driever, W. & Onichtchouk, D. Pou5f1 transcription factor controls zygotic gene activation in vertebrates. Science 341, 1005- 1009, doi:10.1126/science.1242527 (2013).13 Onichtchouk, D. & Driever, W. Zygotic Genome Activators, Developmental Timing, and Pluripotency. Current topics in developmental biology 116, 273- 297, doi:10.1016/bs.ctdb.2015.12.004 (2016).14 Okuda, Y., Ogura, E., Kondoh, H. & Kamachi, Y. B1 SOX coordinate cell specification with patterning and morphogenesis in the early zebrafish embryo. PLoS Genetics 6, e1000936, doi:10.1371/journal.pgen.1000936 (2010).15 Boyer, L. A. et al. Core transcriptional regulatory circuitry in human embryonic stem cells. Cell. 122, 947- 956. (2005).16 Friman, E. T. et al. Dynamic regulation of chromatin accessibility by pluripotency transcription factors across the cell cycle. Elife 8, doi:10.7554/eLife.50087 (2019).17 Gao, M. et al. Pluripotency factors determine gene expression repertoire at zygotic genome activation. Nat Commun 13, 788, doi:10.1038/s41467- 022- 28434- 1 (2022).18 Lunde, K., Belting, H. G. & Driever, W. Zebrafish pou5f1/pou2, homolog of mammalian Oct4, functions in the endoderm specification cascade. Current biology : CB 14, 48- 55 (2004).19 Veil, M. et al. Maternal Nanog is required for zebrafish embryo architecture and for cell viability during gastrulation. Development 145, doi:10.1242/dev.155366 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 879, 920]]<|/det|> +20 Onichtchouk, D. et al. Zebrafish Pou5f1- dependent transcriptional networks in temporal control of early development. Mol Syst Biol 6, 354, doi:msb20109 [pii] 10.1038/msb.2010.9 (2010). 21 Palfy, M., Schulze, G., Valen, E. & Vastenhouw, N. L. Chromatin accessibility established by Pou5f3, Sox19b and Nanog primes genes for activity during zebrafish genome activation. PLoS Genet 16, e1008546, doi:10.1371/journal.pgen.1008546 (2020). 22 Veil, M., Yampolsky, L., Gruening, B. & Onichtchouk, D. Pou5f3, SoxB1, and Nanog remodel chromatin on High Nucleosome Affinity Regions at Zygotic Genome Activation. Genome Res, doi:10.1101/gr.240572.118 (2019). 23 Xu, C. et al. Nanog- like Regulates Endoderm Formation through the Mxtx2- Nodal Pathway. Developmental cell 22, 625- 638, doi:10.1016/j.devcel.2012.01.003 (2012). 24 Donaghey, J. et al. Genetic determinants and epigenetic effects of pioneer- factor occupancy. Nature Genetics 50, 250- 258, doi:10.1038/s41588- 017- 0034- 3 (2018). 25 Segert, J. A., Gisselbrecht, S. S. & Bulyk, M. L. Transcriptional Silencers: Driving Gene Expression with the Brakes On. Trends Genet 37, 514- 527, doi:10.1016/j.tig.2021.02.002 (2021). 26 Mulas, C. et al. Oct4 regulates the embryonic axis and coordinates exit from pluripotency and germ layer specification in the mouse embryo. Development 145, doi:10.1242/dev.159103 (2018). 27 Niwa, H., Miyazaki, J. & Smith, A. G. Quantitative expression of Oct- 3/4 defines differentiation, dedifferentiation or self- renewal of ES cells. Nat Genet 24, 372- 376, doi:10.1038/74199 (2000). 28 Radzisheuskaya, A. & Silva, J. C. Do all roads lead to Oct4? the emerging concepts of induced pluripotency. Trends in cell biology 24, 275- 284, doi:10.1016/j.tcb.2013.11.010 (2014). 29 Tiana, M. et al. Pluripotency factors regulate the onset of Hox cluster activation in the early embryo. Sci Adv 8, eabo3583, doi:10.1126/sciadv.abo3583 (2022). 30 Torres- Padilla, M. E. & Chambers, I. Transcription factor heterogeneity in pluripotent stem cells: a stochastic advantage. Development 141, 2173- 2181, doi:10.1242/dev.102624 (2014). 31 Esch, D. et al. A unique Oct4 interface is crucial for reprogramming to pluripotency. Nature cell biology 15, 295- 301, doi:10.1038/ncb2680 (2013). 32 Bates, L. E., Alves, M. R. P. & Silva, J. C. R. Auxin- degron system identifies immediate mechanisms of OCT4. Stem Cell Reports 16, 1818- 1831, doi:10.1016/j.stemcr.2021.05.016 (2021). 33 Kaplan, N. et al. The DNA- encoded nucleosome organization of a eukaryotic genome. Nature 458, 362- 366, doi:10.1038/nature07667 (2009). 34 Westerfield, M. The zebrafish book. A guide for the laboratory use of zebrafish (Danio rerio). 4th ed edn, (Univ. of Oregon Press, Eugene., 2000). 35 Kimmel, C. B., Ballard, W. W., Kimmel, S. R., Ullmann, B. & Schilling, T. F. Stages of embryonic development of the zebrafish. Dev Dyn 203, 253- 310 (1995). 36 Buenrostro, J. D., Giresi, P. G., Zaba, L. C., Chang, H. Y. & Greenleaf, W. J. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA- binding proteins and nucleosome position. Nature methods 10, 1213- 1218, doi:10.1038/nmeth.2688 (2013). 37 Afgan, E. et al. The Galaxy platform for accessible, reproducible and collaborative biomedical analyses: 2016 update. Nucleic Acids Res 44, W3- W10, doi:10.1093/nar/gkw343 (2016). 38 Langmead, B. & Salzberg, S. L. Fast gapped- read alignment with Bowtie 2. Nat Methods 9, 357- 359, doi:10.1038/nmeth.1923 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 80, 884, 744]]<|/det|> +39 Barnett, D. W., Garrison, E. K., Quinlan, A. R., Stromberg, M. P. & Marth, G. T. BamTools: a C++ API and toolkit for analyzing and managing BAM files. Bioinformatics 27, 1691- 1692, doi:10.1093/bioinformatics/br174 (2011). 40 Ramirez, F. et al. deepTools2: a next generation web server for deep- sequencing data analysis. Nucleic Acids Res 44, W160- 165, doi:10.1093/nar/gkw257 (2016). 41 Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841- 842, doi:10.1093/bioinformatics/btq033 (2010). 42 Feng, J., Liu, T., Qin, B., Zhang, Y. & Liu, X. S. Identifying ChIP- seq enrichment using MACS. Nat Protoc 7, 1728- 1740, doi:10.1038/nprot.2012.101 (2012). 43 Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 15, 550, doi:10.1186/s13059- 014- 0550- 8 (2014). 44 Grant, C. E., Bailey, T. L. & Noble, W. S. FIMO: scanning for occurrences of a given motif. Bioinformatics 27, 1017- 1018, doi:10.1093/bioinformatics/br064 (2011). 45 Dobin, A. et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 29, 15- 21, doi:10.1093/bioinformatics/bt5635 (2013). 46 Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923- 930, doi:10.1093/bioinformatics/bt656 (2014). 47 Hiller, M. et al. Computational methods to detect conserved non- genic elements in phylogenetically isolated genomes: application to zebrafish. Nucleic Acids Res 41, e151, doi:10.1093/nar/gkt557 (2013). 48 Huang da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4, 44- 57, doi:10.1038/nprot.2008.211 (2009). 49 Kaschek, D., Mader, W., Fehling- Kaschek, M., Rosenblatt, M. & Timmer, J. Dynamic Modeling, Parameter Estimation, and Uncertainty Analysis in R. Journal of Statistical Software 88, 1 - 32, doi:10.18637/jss.v088.ii10 (2019). 50 Raue, A. et al. Correction: Lessons Learned from Quantitative Dynamical Modeling in Systems Biology. PLOS ONE 8, 10.1371/annotation/ea0193d1378- 1371f1377f- 1492a- b1370b1377- d877629fdcee, doi:10.1371/annotation/ea0193d8- 1f7f- 492a- b0b7- d877629fdcee (2013). 51 Raue, A. et al. Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood. Bioinformatics 25, 1923- 1929, doi:10.1093/bioinformatics/btp358 (2009). 52 Alon, U. Network motifs: theory and experimental approaches. Nature Reviews Genetics 8, 450- 461, doi:10.1038/nrg2102 (2007). 53 Maiwald, T. et al. Driving the Model to Its Limit: Profile Likelihood Based Model Reduction. PLOS ONE 11, e0162366, doi:10.1371/journal.pone.0162366 (2016). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 438, 338]]<|/det|> +- TableS1accessibleregions.xlsx- TableS2zygotictranscripts.xlsx- TableS3ARslinkedtotranscripts.xlsx- TableS4Coremodelequations.csv- TableS5CoremodelBestfitparameters.csv- TableS6Minimodelequations.csv- TableS7Minimodelparameters.csv- RiesleSUPPLng3.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/images_list.json b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..3e1906e2a1346179d0f9affc6fc9f0471a01f6ed --- /dev/null +++ b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/images_list.json @@ -0,0 +1,32 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 52, + 48, + 928, + 460 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 50, + 49, + 950, + 365 + ] + ], + "page_idx": 9 + } +] \ No newline at end of file diff --git a/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208.mmd b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9b9afbb55932fa7e76f42548421bdee8efd8d423 --- /dev/null +++ b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208.mmd @@ -0,0 +1,215 @@ + +# Insights into the use of cytochrome bc1 inhibitors for future therapeutic strategies for tuberculosis + +Clara Aguilar Pérez CAgui1a2@its.jn.j.com + +Janssen Pharmaceutica Anne Lenaerts Colorado State University + +Cristina Villellas Janssen Pharmaceutica Jerome Guillemont Janssen-Cilag + +Department of Infection Biology, Faculty of Infectious and Tropical Disease, London School of Hygiene and Tropical Medicine + +Hannah Painter London School of Hygiene and Tropical Medicine + +Nicole Ammerman Erasmus MC + +Anis Hassan London School of Hygiene and Tropical Medicine + +Guillaume Golovkine Evotec https://orcid.org/0000- 0001- 8388- 5892 + +Laure Brock Evotec + +Aurelie Chauffour Sorbonne Universités https://orcid.org/0000- 0002- 8846- 5069 + +Alexandra Aubry Sorbonne Université + +Thi Mai Sorbonne Université + +Sarah Wong Sorbonne Université + +Taane Clark London School of Hygiene & Tropical Medicine https://orcid.org/0000- 0001- 8985- 9265 + +Kiyean Nam Qurient Co. + +Jeongjun Kim + +<--- Page Split ---> + +Qurient Co. + +Jinho Choi Qurient Co. + +Marjolein Crabbe Janssen Pharmaceutica + +Jorge Esquivias Janssen- Cilag + +Nacer Lounis J&J + +Bart Stoops Janssen Pharmaceutica + +Katie Amssoms Janssen Research and Development + +Jose Bartolome- Nebreda Janssen- Cilag + +Veronica Gruppo Colorado State University + +Gregory Robertson Colorado State University, Fort Collins, CO + +Nicolas Veziris Sorbonne Université + +Anna Upton Evotec + +Eric Nuermberger Johns Hopkins University School of Medicine + +Vivian Cox Janssen Pharmaceutica + +Lluis Ballell Janssen Pharmaceutica + +Benny Baeten Janssen Pharmaceutica + +Anil Koul Janssen Research and Development, Johnson and Johnson + +Alexander Pym Janssen Pharmaceutica + +Richard Wall London School of Hygiene and Tropical Medicine + +Dirk Lamprecht Janssen Pharmaceutica + +Brief Communication + +Keywords: + +<--- Page Split ---> + +**Posted Date:** October 31st, 2024 + +**DOI:** https://doi.org/10.21203/rs.3.rs- 5331796/v1 + +**License:** © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +**Additional Declarations:** Yes there is potential Competing Interest. JG, CV and DAL have been named inventors in a patent application for JNJ- 2901 and JNJ- 4052 compounds. CAP, CV, JG, MC, JE, NL, BS, JMB, VC, LBallell, BB, AK, ASP and DAL were/are all full- time employees of Janssen, a Johnson & Johnson company, and/or potential stockholders of Johnson & Johnson. The other authors declare no competing interests + +**Version of Record:** A version of this preprint was published at Nature Communications on October 22nd, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 64427- 6. + +<--- Page Split ---> + +## Abstract + +Recent efforts to improve tuberculosis (TB) treatment options have focused on developing molecules with novel mechanisms of action and identifying optimal treatment regimens. Inhibition of Mycobacterium tuberculosis cytochrome bc1 oxidase has emerged as a promising therapeutic target that could potentially contribute to improved TB combination regimens. Using a relapsing mouse model, we demonstrate that cytochrome bc1 inhibitors could serve as effective partner drugs, enhancing regimen sterilisation. We propose several novel regimen strategies for both multidrug- resistant TB (MDR- TB) and drug- sensitive TB (DS- TB), where cytochrome bc1 inhibitors contribute to sterilisation and treatment shortening. Additionally, we show that clinical isolates exhibit heightened susceptibility to cytochrome bc1 inhibitors compared to laboratory- adapted strains, further supporting their translational potential. These findings suggest that cytochrome bc1 inhibitors have significant potential to improve TB treatment outcomes and highlight the need for further studies to evaluate their clinical contribution to novel treatment regimens. + +## Full Text + +The global response to the tuberculosis (TB) epidemic has been exacerbated by the emergence of multidrug- resistant TB (MDR- TB). Combination treatment for TB is essential to prevent drug resistance, ensure complete eradication of the bacteria, and improve treatment effectiveness; however, current regimens still require many months of treatment. The World Health Organisation's consolidated guidelines include new recommendations for treatment shortening \(^{1,2}\) . These recommendations suggest two six- month regimen options (compared to the previous 9 to 18- month duration): BPaLM (bedaquiline, pretomanid, linezolid and moxifloxacin) for ages 14 years and older, and BDLLfxC (bedaquiline, delamanid, linezolid, levofloxacin and clofazimine) for children, adolescents, adults and pregnant women. The central role that fluoroquinolones (FQ) and linezolid play in those regimens pose serious limitations due to drug resistance and toxicity, respectively. Moxifloxacin (M) or levofloxacin (Lfx) must be excluded in cases of fluoroquinolone resistance \(^{3}\) , modifying the regimens to BPaL and BDLC, respectively. Furthermore, the use of linezolid in MDR- TB treatment is often associated with adverse events, such as myelosuppression and peripheral neuropathy, which may negatively impact patient adherence and require regimen modification \(^{4}\) . While MDR- TB treatment options are diversifying and improving \(^{5}\) , emerging drug resistance adds additional concerns, underscoring the urgent need to discover compounds with novel modes of action (MoA) and develop novel therapeutic approaches. The discovery of Q203 (telacebec, T; Supplementary Figure S1) and its successful Phase 2A Early Bactericidal Activity (EBA) trial has highlighted the inhibition of the cytochrome \(bc_{1}\) complex of Mycobacterium tuberculosis as a promising new molecular target \(^{6 - 8}\) . The cytochrome \(bc_{1}\) complex is a crucial part of the electron transport chain, essential for ATP production, potentially making an inhibitor of cytochrome \(bc_{1}\) a good partner drug for other inhibitors of oxidative phosphorylation such as bedaquiline. Despite this potential, currently, there is limited demonstration of the specific role cytochrome \(bc_{1}\) inhibitors could play in future TB treatment regimens \(^{9}\) . Here, using relapsing mouse models, we assessed the contribution of a validated cytochrome \(bc_{1}\) inhibitor tool compound (JNJ- 2901, J) in different treatment regimen strategies with special focus on combinations devoid of FQ- resistance liability or linezolid- associated toxicity (Figure 1A, Supplementary Figure S1- 2, Supplementary Table S1- 3). + +In the TB- PRACTECAL trial, the BPaLC (bedaquiline, pretomanid, linezolid and clofazimine) regimen resulted in a higher proportion of MDR- TB patients with favourable outcomes (81%) compared to BPaL (77%) and standard of care (SoC; 52%), primarily due to early treatment discontinuation from adverse events in the SoC group \(^{10}\) . Given the + +<--- Page Split ---> + +adverse effects associated with the use of linezolid in the treatment of MDR- TB, there is growing interest in identifying alternative partner drugs11-13. Here, using a mouse model, we show that a regimen including a cytochrome \(bc_{1}\) inhibitor could serve as a promising alternative (Figure 1B- C; Supplementary Table S4- 7; Studies A and B). + +Although BPaL, the regimen recommended for fluoroquinolone- resistant MDR- TB, led to a greater decrease in colony forming units (CFU) after 8 weeks compared to BPaJ, comparable relapse rates (measured after 3 months post treatment) were observed after 8 weeks (86.6%- 100% for BPaL, 100% for BPaJ) and 12 weeks of treatment (26.6% for both regimens). Replacing linezolid with clofazimine resulted in a similar decrease in CFU after 8 weeks but led to fewer relapses post- treatment (86.6%- 100% for BPaL, 26.6- 66.6% for BPaC). BPaCJ demonstrated the best bactericidal effect, achieving an additional \(1 \log_{10}\) decrease in CFU compared to BPaC after 8 weeks, translating to a 33% relapse rate. No relapses occurred after 12 weeks of treatment with either BPaC or BPaCJ, highlighting the potential for treatment shortening compared to the BPaL SoC (Figure 1B- C; Supplementary Table S4- 7). Interestingly, addition of JNJ- 2901 to a moxifloxacin- containing regimen (BPaMJ) did not lead to a further decrease in bacterial burden or in relapse rate compared with BPaM (Figure 1B; Supplementary Table S4- 7). Our findings demonstrate that combining a cytochrome \(bc_{1}\) inhibitor with BPaC increases the sterilising activity compared to BPaC alone, suggesting that cytochrome \(bc_{1}\) inhibitors could play an important role in future MDR- TB regimens. In an additional study (Study C), JNJ- 2901 reduced the relapse rate when combined with the BPaL regimen, despite an inability to rescue mice when administered as a monotherapy in an intravenous model using the H37Rv reference strain (Supplementary Figure S3). There was a trend towards improvement of both the bacterial burden (BPaLJ: \(0.95 \pm 1.1\) vs BPaL: \(2.06 \pm 1.3 \log_{10}\) CFU lung- 1, \(p = 0.04\)) and a quantifiable improvement in relapse rates, that failed to reach statistical significance within the limitations of this trial design (50% vs 89%, \(p > 0.05\)), while BPaLJ was comparable to BPaLM and BPaMZ (Supplementary Table S8- 10). This suggests that BPaLJ could be considered as a potential treatment regimen irrespective of fluoroquinolone susceptibility. + +In a further study, an alternative dosing strategy was used where mice were treated with BPaL for 8 weeks during an initial phase, followed by an additional 8 weeks of either BPa, BJ, B or no treatment in a continuation phase. The rationale for this study design was to reduce the time and overall amount of drug required to reach bacterial sterility. Bacterial burdens in lungs were assessed after 8, 12 and 16 weeks, and relapse rates were measured after 16 weeks of treatment plus 16 weeks following treatment cessation (Figure 1D; Supplementary Table S11; Study D). Both the BPaL/BPa and BPaL/BJ regimens trended towards lower relapse rates compared to BPaL without continuation treatment, although this was not statistically significant ( \(p\) - value=0.09). No colonies were detected after 12 weeks of treatment with BPaL/BJ, whereas with BPaL/B, colonies were detected from 3/5 mice at the end of treatment. These results suggest that inclusion of a cytochrome \(bc_{1}\) inhibitor could enhance regimen effectiveness during the continuation phase of treatment with fewer drugs and a reduced overall drug burden. Effective regimens require drugs with both bactericidal and sterilising activities to rapidly reduce bacterial loads and kill drug- tolerant bacilli. Drugs in the continuation phase must target drug- tolerant bacteria that survive the initial intensive phase14. A two- phase regimen, with an initial phase of drugs that rapidly reduce bacterial burdens, followed by sterilising drugs in a continuation phase, is a proven strategy for improving treatment outcomes for patients. The inclusion of a cytochrome \(bc_{1}\) inhibitor in the continuation phase of the BPaL/BJ regimen demonstrated a sterilising effect similar to that of the BPaL/BPa regimen, suggesting that a cytochrome \(bc_{1}\) inhibitor could serve as an effective alternative to premenopausal (Pa). + +<--- Page Split ---> + +Next, we focused on an ultra- short treatment strategy for drug sensitive- TB (DS- TB) based on drugs targeting the respiratory pathway. For this purpose, we assessed the efficacy of regimens containing telacebec alongside other drugs targeting the respiratory pathway, compared to the current DS- TB SoC; isoniazid, rifampicin, pyrazinamide, and ethambutol (HRZE; Study E). BCZ, CZT and BCZT regimens all demonstrated superior efficacy over HRZE after 8 weeks of treatment in reducing lung CFU burdens (Figure 2A, Supplementary Tables S12- 14; Study E). Both BCZ and BCZT regimens achieved \(0\%\) relapse rates 12 weeks after 6- and 8- weeks of treatment, whereas CZT required at least 12 weeks of treatment to achieve \(0\%\) relapse. One of the mice treated with HRZE remained culture positive even after 20 weeks of treatment. The benefit of adding telacebec to BCZ is highlighted by the relapse rates after 4 weeks treatment: BCZ \((60\%)\) vs. BCZT \((36\%)\) , underscoring the treatment- shortening potential of cytochrome \(bc_{1}\) inhibitors in combination with other drugs targeting the electron transport chain. Based on the TRUNCATE- TB trial strategy, a 2- month BZ- containing regimen could be as effective as the current 6- month standard (HRZE) \(^{15}\) . Adding telacebec to the CZ core regimen reduced bacterial load by \(2.2 \log_{10}\) CFU in this model and, including telacebec in CZ or BCZ regimens significantly shortened treatment compared to HRZE. These findings provide strong evidence that incorporating a cytochrome \(bc_{1}\) inhibitor into new treatment regimens offers substantial benefits which should be further investigated. + +We found that the impact of cytochrome \(bc_{1}\) inhibitors may be underrepresented depending on the M. tuberculosis strain selected for investigation. Here, using another cytochrome \(bc_{1}\) inhibitor, JNJ- 4052, we demonstrate that a diverse range of clinical isolates from infected patients exhibited increased susceptibility to cytochrome \(bc_{1}\) inhibitors in both minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) assays when compared to the lab- adapted H37Rv strain(Figures 2B- E; Supplementary Table S15). This heightened susceptibility translated into increased in vivo efficacy in an acute mouse model, where a statistically significant CFU reduction \((1.6 \log_{10}\) reduction; \(p< 0.01\) ) was observed following cytochrome \(bc_{1}\) inhibitor (JNJ- 4052) treatment in mice inoculated with a clinical isolate (Figure 2F; Supplementary Tables S1- S3; Study F). Further work is needed to determine whether the routine use of clinical isolates should be incorporated into regimen design studies. Indeed, similar findings have recently been reported \(^{16}\) , showing that the clinical isolate HN878 is more susceptible to cytochrome \(bc_{1}\) inhibitors, with a notable contribution when added to BPaC, resulting in a 2- log reduction in lung bacterial burden between BPaC and BPaCT, after 4 weeks of treatment. These results further emphasise the relevance of clinical isolates for evaluating the in vivo efficacy of new compounds. + +Together, our findings suggest that cytochrome \(bc_{1}\) inhibitors could play a crucial role in future TB treatment regimens. Specifically, cytochrome \(bc_{1}\) inhibitors may serve as effective replacements for linezolid in SoC regimens for MDR- and fluoroquinolone- resistant MDR- TB. As this represents a novel MoA, it could also help raise the barrier to resistance development to other components of the regimen and could be use for treatment of DS- TB. Our experiments containing a treatment phase followed by a continuation phase may also benefit future formulation strategies such as long acting injectables (LAI) providing an alternative strategy to improve treatment outcomes. Finally, drug repurposing has proven to be an effective strategy for accelerating the discovery of new treatments; linezolid, for example, was originally developed for vancomycin- resistant enterococcal infections. Our data highlight the potential of both clofazimine- and cytochrome \(bc_{1}\) inhibitor- containing regimens to significantly shorten treatment durations for both DS- and DR- TB. + +## Declarations + +## Acknowledgements + +<--- Page Split ---> + +The authors would like to thank: Amit Kaushik for technical support at Johns Hopkins University; AT Henze and PM Petrar for medical writing support at Janssen Pharmaceutica; Annelies Wouter and Lieve Lammens for toxicology input at Janssen Pharmaceutica; Nadia Neto for her feedback on WGS at Janssen Pharmaceutica; Gregory Bancroft for his intellectual input at the London School of Hygiene & Tropical Medicine; and Courtney I. Hastings at Colorado State University for technical support. + +## Author contributions + +CAP, DAL, AJL, SS, GTR, NL, AU, NV, ELN, NCA AA, AC BB, LB, HP, JD and ASP were involved in the conception or the design of the study. JG and JMB designed and synthesised the compounds. AK, CV, DAL, BB, ASP and LBallell supervised the overall research programme. CAP, SS, ELN, AC, NCA, TCM, SW, NV, NW, HP, JD, AH, BS, LBrock, GG and GR participated in the collection or generation of the studies data. SS, EN, AC, TCM, SW, HP, JD, AH, VG and GTR performed the studies. CAP, SS, ELN, JE, TGC and GR contributed to the study with materials/analysis tools. CAP, AJL, NV, SW, GTR, SS, MC, ELN, NCA, LBallell, JD, HP, VC, BS, RJW and DAL were involved in the analysis or interpretation of the data. CAP, RJW and DAL wrote the manuscript. All authors reviewed and commented drafts of the manuscript for intellectual content and gave final approval to submit for publication. All authors attest they meet the ICMJE criteria for authorship. + +## Data availability + +All data generated or analysed during this study are included in this published article (and its supplementary information files). + +## Conflict of interest statement + +JG, CV and DAL have been named inventors in a patent application for JNJ- 2901 and JNJ- 4052 compounds. CAP, CV, JG, MC, JE, NL, BS, JMB, VC, LBallell, BB, AK, ASP and DAL were/are all full- time employees of Janssen, a Johnson & Johnson company, and/or potential stockholders of Johnson & Johnson. The other authors declare no competing interests. + +## Funding statement + +This work was supported by Janssen Pharmaceutica NV, including all costs associated with the development and publishing of the manuscript. The work at the London School of Hygiene & Tropical Medicine was supported by funding from Janssen Pharmaceutica. This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 853903 (RespiriTB). This Joint Undertaking receives support from the European Union's Horizon 2020 research and innovation programme and EFPIA. + +## References + +1. World Health Organisation, WHO consolidated guidelines on tuberculosis. Module 4: treatment - drug-resistant tuberculosis treatment. (2022). +2. World Health Organisation, Key updates to the treatment of drug-resistant tuberculosis - rapid communication. (2024). +3. Nehru, V.J., et al. Risk assessment and transmission of fluoroquinolone resistance in drug-resistant pulmonary tuberculosis: a retrospective genomic epidemiology study. Scientific Reports 14, 19719 (2024). + +<--- Page Split ---> + +4. McKee, E.E., Ferguson, M., Bentley, A.T. & Marks, T.A. Inhibition of mammalian mitochondrial protein synthesis by oxazolidinones. Antimicrob Agents Chemother 50, 2042-2049 (2006). + +5. Stop TB Partnership, Clinical Pipeline. (2024). + +6. de Jager, V.R., et al. Telacebec (Q203), a New Antituberculosis Agent. N Engl J Med 382, 1280-1281 (2020). + +7. Chandrasekera, N.S., et al. Improved Phenoxylalkylbenzimidazoles with Activity against Mycobacterium tuberculosis Appear to Target QcrB. ACS Infectious Diseases 3, 898-916 (2017). + +8. Pethe, K., et al. Discovery of Q203, a potent clinical candidate for the treatment of tuberculosis. Nature Medicine 19, 1157-1160 (2013). + +9. Wani, M.A. & Dhaked, D.K. Targeting the cytochrome bc(1) complex for drug development in M. tuberculosis: review. Mol Divers 26, 2949-2965 (2022). + +10. Nyang'wa, B.-T., et al. A 24-Week, All-Oral Regimen for Rifampin-Resistant Tuberculosis. New England Journal of Medicine 387, 2331-2343 (2022). + +11. Eimer, J., et al. Association Between Increased Linezolid Plasma Concentrations and the Development of Severe Toxicity in Multidrug-Resistant Tuberculosis Treatment. Clin Infect Dis 76, e947-e956 (2023). + +12. Vengurlekar, D., et al. Linezolid resistance in patients with drug-resistant TB. Int J Tuberc Lung Dis 27, 567-569 (2023). + +13. Wasserman, S., et al. Linezolid toxicity in patients with drug-resistant tuberculosis: a prospective cohort study. J Antimicrob Chemother 77, 1146-1154 (2022). + +14. Sotgiu, G., Centis, R., D'Ambrosio, L. & Migliori, G.B. Tuberculosis treatment and drug regimens. Cold Spring Harb Perspect Med 5, a017822 (2015). + +15. Paton, N.I., et al. Treatment Strategy for Rifampin-Susceptible Tuberculosis. New England Journal of Medicine 388, 873-887 (2023). + +16. Komm, O.D., et al. Contribution of telacebec to novel drug regimens in a murine tuberculosis model. bioRxiv, 2024.2006.2027.601059 (2024). + +17. Li, S.Y., et al. Evaluation of moxifloxacin-containing regimens in pathologically distinct murine tuberculosis models. Antimicrob Agents Chemother 59, 4026-4030 (2015). + +18. Kort, F., et al. Fully weekly antituberculosis regimen: a proof-of-concept study. Eur Respir J 56(2020). + +19. Akester, J.N., et al. Synthesis, Structure-Activity Relationship, and Mechanistic Studies of Aminoquinazolinones Displaying Antimycobacterial Activity. ACS Infect Dis 6, 1951-1964 (2020). + +20. Ray, P.C., et al. Spirocycle MmpL3 Inhibitors with Improved hERG and Cytotoxicity Profiles as Inhibitors of Mycobacterium tuberculosis Growth. ACS Omega 6, 2284-2311 (2021). + +21. Lamprecht, D.A., et al. Targeting de novo purine biosynthesis for tuberculosis treatment. Pre-print https://www.researchsquare.com/article/rs-4913610/v1 (2024). + +## Figures + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +Investigating alternative MDR- TB treatment regimens based on inclusion of a cytochrome \(bc_{1}\) inhibitor in relapsing mouse models of TB – (A) Schematic of the studies presented in this project. (B- D) Lung bacterial burden, proportions of mice with positive cultures at end of treatment and relapse in M. tuberculosis- infected mice from Studies A, B and D. CFU data is shown for week 8 (panels B- C) and week 12 (panel D). Details of treatment doses can be found in Supplementary Tables S4 and S11. Bedaquiline: B; pretomanid: Pa; clofazimine: C; linezolid: L; moxifoxacin: M; JNJ- 2901: J. SoT: Start of treatment. \*: relapse rate; 12 or 16 weeks after treatment cessation (+12 or 16 wks). Relapse rate (n/N). + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Investigating alternative DS- TB treatment regimens based on inclusion of a cytochrome \(bc_{1}\) inhibitor and bactericidal activity using a clinical isolate – (A) Lung bacterial burden and relapse in M. tuberculosis- infected mice from Study E. Details of treatment doses can be found in Supplementary Table S12. Limit of detection (dotted line) was \(0.4 \log_{10} \text{CFU} \text{lung}^{-1}\) . Bedaquiline: B; clofazimine: C; rifampicin: R; ethambutol: E; isoniazid: H; telacebec: T; pyrazinamide: Z. (B) Distribution of \(\text{MIC}_{90}\) (concentration achieving \(90\%\) inhibition) values for a diverse selection of clinical isolates (black) compared with the lab- adapted WT H37Rv (cyan) for JNJ- 4052 and bedaquiline (BDQ) (Supplementary Table S15). Clinical isolate N1283 (orange) is indicated. \(n = 2\) technical replicates. Impact of (C) JNJ- 4052, (D) JNJ- 2901 and (E) telecebec on CFU counts in WT H37Rv compared to N1283 clinical isolate. \(n = \geq 2\) biological replicates. (F) In vivo efficacy (Study F) of JNJ- 4052 (50 mg \(\text{kg}^{-1}\) ; PO) in H37Rv and N1283 after 2 weeks of treatment. SoT: Start of treatment, 7 days after inoculum with 300 CFU. \(n = 3\) mice. \(** = p< 0.01\) . + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Insightsintotheuseofcytochrome bc1inhibitorsforfuturetherapeutictrategiesfortuberculosissupplementary.docx +- NMEDBC136846epc.pdf +- NMEDBC136846rs.pdf +- Onlinemethods.docx + +<--- Page Split ---> diff --git a/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208_det.mmd b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..034673c09c9bcd7557b2c143de5b93e365de33a0 --- /dev/null +++ b/preprint/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208/preprint__1292826cb041fd67ca20b49f8643de61f387627c09dfbb550f24e651c23d0208_det.mmd @@ -0,0 +1,299 @@ +<|ref|>title<|/ref|><|det|>[[42, 106, 880, 170]]<|/det|> +# Insights into the use of cytochrome bc1 inhibitors for future therapeutic strategies for tuberculosis + +<|ref|>text<|/ref|><|det|>[[44, 186, 256, 228]]<|/det|> +Clara Aguilar Pérez CAgui1a2@its.jn.j.com + +<|ref|>text<|/ref|><|det|>[[44, 253, 255, 310]]<|/det|> +Janssen Pharmaceutica Anne Lenaerts Colorado State University + +<|ref|>text<|/ref|><|det|>[[44, 316, 250, 410]]<|/det|> +Cristina Villellas Janssen Pharmaceutica Jerome Guillemont Janssen-Cilag + +<|ref|>text<|/ref|><|det|>[[44, 414, 894, 455]]<|/det|> +Department of Infection Biology, Faculty of Infectious and Tropical Disease, London School of Hygiene and Tropical Medicine + +<|ref|>text<|/ref|><|det|>[[44, 460, 444, 500]]<|/det|> +Hannah Painter London School of Hygiene and Tropical Medicine + +<|ref|>text<|/ref|><|det|>[[44, 505, 196, 542]]<|/det|> +Nicole Ammerman Erasmus MC + +<|ref|>text<|/ref|><|det|>[[44, 548, 444, 587]]<|/det|> +Anis Hassan London School of Hygiene and Tropical Medicine + +<|ref|>text<|/ref|><|det|>[[44, 592, 437, 628]]<|/det|> +Guillaume Golovkine Evotec https://orcid.org/0000- 0001- 8388- 5892 + +<|ref|>text<|/ref|><|det|>[[44, 633, 135, 668]]<|/det|> +Laure Brock Evotec + +<|ref|>text<|/ref|><|det|>[[44, 674, 550, 711]]<|/det|> +Aurelie Chauffour Sorbonne Universités https://orcid.org/0000- 0002- 8846- 5069 + +<|ref|>text<|/ref|><|det|>[[44, 716, 216, 752]]<|/det|> +Alexandra Aubry Sorbonne Université + +<|ref|>text<|/ref|><|det|>[[44, 758, 216, 794]]<|/det|> +Thi Mai Sorbonne Université + +<|ref|>text<|/ref|><|det|>[[44, 800, 216, 835]]<|/det|> +Sarah Wong Sorbonne Université + +<|ref|>text<|/ref|><|det|>[[44, 841, 750, 877]]<|/det|> +Taane Clark London School of Hygiene & Tropical Medicine https://orcid.org/0000- 0001- 8985- 9265 + +<|ref|>text<|/ref|><|det|>[[44, 882, 155, 917]]<|/det|> +Kiyean Nam Qurient Co. + +<|ref|>text<|/ref|><|det|>[[44, 923, 155, 940]]<|/det|> +Jeongjun Kim + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 45, 145, 61]]<|/det|> +Qurient Co. + +<|ref|>text<|/ref|><|det|>[[44, 68, 145, 103]]<|/det|> +Jinho Choi Qurient Co. + +<|ref|>text<|/ref|><|det|>[[44, 109, 247, 144]]<|/det|> +Marjolein Crabbe Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 150, 171, 186]]<|/det|> +Jorge Esquivias Janssen- Cilag + +<|ref|>text<|/ref|><|det|>[[44, 192, 150, 226]]<|/det|> +Nacer Lounis J&J + +<|ref|>text<|/ref|><|det|>[[44, 234, 239, 269]]<|/det|> +Bart Stoops Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 275, 344, 311]]<|/det|> +Katie Amssoms Janssen Research and Development + +<|ref|>text<|/ref|><|det|>[[44, 317, 245, 352]]<|/det|> +Jose Bartolome- Nebreda Janssen- Cilag + +<|ref|>text<|/ref|><|det|>[[44, 359, 255, 394]]<|/det|> +Veronica Gruppo Colorado State University + +<|ref|>text<|/ref|><|det|>[[44, 400, 384, 437]]<|/det|> +Gregory Robertson Colorado State University, Fort Collins, CO + +<|ref|>text<|/ref|><|det|>[[44, 443, 216, 479]]<|/det|> +Nicolas Veziris Sorbonne Université + +<|ref|>text<|/ref|><|det|>[[44, 485, 140, 520]]<|/det|> +Anna Upton Evotec + +<|ref|>text<|/ref|><|det|>[[44, 526, 185, 562]]<|/det|> +Eric Nuermberger Johns Hopkins University School of Medicine + +<|ref|>text<|/ref|><|det|>[[44, 567, 247, 603]]<|/det|> +Vivian Cox Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 609, 247, 644]]<|/det|> +Lluis Ballell Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 650, 247, 686]]<|/det|> +Benny Baeten Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 692, 528, 728]]<|/det|> +Anil Koul Janssen Research and Development, Johnson and Johnson + +<|ref|>text<|/ref|><|det|>[[44, 733, 247, 769]]<|/det|> +Alexander Pym Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 775, 440, 811]]<|/det|> +Richard Wall London School of Hygiene and Tropical Medicine + +<|ref|>text<|/ref|><|det|>[[44, 817, 247, 853]]<|/det|> +Dirk Lamprecht Janssen Pharmaceutica + +<|ref|>text<|/ref|><|det|>[[44, 894, 213, 911]]<|/det|> +Brief Communication + +<|ref|>text<|/ref|><|det|>[[44, 929, 128, 946]]<|/det|> +Keywords: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[43, 45, 301, 62]]<|/det|> +**Posted Date:** October 31st, 2024 + +<|ref|>text<|/ref|><|det|>[[42, 79, 434, 97]]<|/det|> +**DOI:** https://doi.org/10.21203/rs.3.rs- 5331796/v1 + +<|ref|>text<|/ref|><|det|>[[42, 113, 920, 152]]<|/det|> +**License:** © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 168, 944, 248]]<|/det|> +**Additional Declarations:** Yes there is potential Competing Interest. JG, CV and DAL have been named inventors in a patent application for JNJ- 2901 and JNJ- 4052 compounds. CAP, CV, JG, MC, JE, NL, BS, JMB, VC, LBallell, BB, AK, ASP and DAL were/are all full- time employees of Janssen, a Johnson & Johnson company, and/or potential stockholders of Johnson & Johnson. The other authors declare no competing interests + +<|ref|>text<|/ref|><|det|>[[42, 280, 944, 319]]<|/det|> +**Version of Record:** A version of this preprint was published at Nature Communications on October 22nd, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 64427- 6. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 147, 65]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[41, 78, 945, 304]]<|/det|> +Recent efforts to improve tuberculosis (TB) treatment options have focused on developing molecules with novel mechanisms of action and identifying optimal treatment regimens. Inhibition of Mycobacterium tuberculosis cytochrome bc1 oxidase has emerged as a promising therapeutic target that could potentially contribute to improved TB combination regimens. Using a relapsing mouse model, we demonstrate that cytochrome bc1 inhibitors could serve as effective partner drugs, enhancing regimen sterilisation. We propose several novel regimen strategies for both multidrug- resistant TB (MDR- TB) and drug- sensitive TB (DS- TB), where cytochrome bc1 inhibitors contribute to sterilisation and treatment shortening. Additionally, we show that clinical isolates exhibit heightened susceptibility to cytochrome bc1 inhibitors compared to laboratory- adapted strains, further supporting their translational potential. These findings suggest that cytochrome bc1 inhibitors have significant potential to improve TB treatment outcomes and highlight the need for further studies to evaluate their clinical contribution to novel treatment regimens. + +<|ref|>sub_title<|/ref|><|det|>[[44, 325, 147, 348]]<|/det|> +## Full Text + +<|ref|>text<|/ref|><|det|>[[40, 355, 944, 875]]<|/det|> +The global response to the tuberculosis (TB) epidemic has been exacerbated by the emergence of multidrug- resistant TB (MDR- TB). Combination treatment for TB is essential to prevent drug resistance, ensure complete eradication of the bacteria, and improve treatment effectiveness; however, current regimens still require many months of treatment. The World Health Organisation's consolidated guidelines include new recommendations for treatment shortening \(^{1,2}\) . These recommendations suggest two six- month regimen options (compared to the previous 9 to 18- month duration): BPaLM (bedaquiline, pretomanid, linezolid and moxifloxacin) for ages 14 years and older, and BDLLfxC (bedaquiline, delamanid, linezolid, levofloxacin and clofazimine) for children, adolescents, adults and pregnant women. The central role that fluoroquinolones (FQ) and linezolid play in those regimens pose serious limitations due to drug resistance and toxicity, respectively. Moxifloxacin (M) or levofloxacin (Lfx) must be excluded in cases of fluoroquinolone resistance \(^{3}\) , modifying the regimens to BPaL and BDLC, respectively. Furthermore, the use of linezolid in MDR- TB treatment is often associated with adverse events, such as myelosuppression and peripheral neuropathy, which may negatively impact patient adherence and require regimen modification \(^{4}\) . While MDR- TB treatment options are diversifying and improving \(^{5}\) , emerging drug resistance adds additional concerns, underscoring the urgent need to discover compounds with novel modes of action (MoA) and develop novel therapeutic approaches. The discovery of Q203 (telacebec, T; Supplementary Figure S1) and its successful Phase 2A Early Bactericidal Activity (EBA) trial has highlighted the inhibition of the cytochrome \(bc_{1}\) complex of Mycobacterium tuberculosis as a promising new molecular target \(^{6 - 8}\) . The cytochrome \(bc_{1}\) complex is a crucial part of the electron transport chain, essential for ATP production, potentially making an inhibitor of cytochrome \(bc_{1}\) a good partner drug for other inhibitors of oxidative phosphorylation such as bedaquiline. Despite this potential, currently, there is limited demonstration of the specific role cytochrome \(bc_{1}\) inhibitors could play in future TB treatment regimens \(^{9}\) . Here, using relapsing mouse models, we assessed the contribution of a validated cytochrome \(bc_{1}\) inhibitor tool compound (JNJ- 2901, J) in different treatment regimen strategies with special focus on combinations devoid of FQ- resistance liability or linezolid- associated toxicity (Figure 1A, Supplementary Figure S1- 2, Supplementary Table S1- 3). + +<|ref|>text<|/ref|><|det|>[[42, 886, 949, 949]]<|/det|> +In the TB- PRACTECAL trial, the BPaLC (bedaquiline, pretomanid, linezolid and clofazimine) regimen resulted in a higher proportion of MDR- TB patients with favourable outcomes (81%) compared to BPaL (77%) and standard of care (SoC; 52%), primarily due to early treatment discontinuation from adverse events in the SoC group \(^{10}\) . Given the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 940, 130]]<|/det|> +adverse effects associated with the use of linezolid in the treatment of MDR- TB, there is growing interest in identifying alternative partner drugs11-13. Here, using a mouse model, we show that a regimen including a cytochrome \(bc_{1}\) inhibitor could serve as a promising alternative (Figure 1B- C; Supplementary Table S4- 7; Studies A and B). + +<|ref|>text<|/ref|><|det|>[[40, 144, 950, 545]]<|/det|> +Although BPaL, the regimen recommended for fluoroquinolone- resistant MDR- TB, led to a greater decrease in colony forming units (CFU) after 8 weeks compared to BPaJ, comparable relapse rates (measured after 3 months post treatment) were observed after 8 weeks (86.6%- 100% for BPaL, 100% for BPaJ) and 12 weeks of treatment (26.6% for both regimens). Replacing linezolid with clofazimine resulted in a similar decrease in CFU after 8 weeks but led to fewer relapses post- treatment (86.6%- 100% for BPaL, 26.6- 66.6% for BPaC). BPaCJ demonstrated the best bactericidal effect, achieving an additional \(1 \log_{10}\) decrease in CFU compared to BPaC after 8 weeks, translating to a 33% relapse rate. No relapses occurred after 12 weeks of treatment with either BPaC or BPaCJ, highlighting the potential for treatment shortening compared to the BPaL SoC (Figure 1B- C; Supplementary Table S4- 7). Interestingly, addition of JNJ- 2901 to a moxifloxacin- containing regimen (BPaMJ) did not lead to a further decrease in bacterial burden or in relapse rate compared with BPaM (Figure 1B; Supplementary Table S4- 7). Our findings demonstrate that combining a cytochrome \(bc_{1}\) inhibitor with BPaC increases the sterilising activity compared to BPaC alone, suggesting that cytochrome \(bc_{1}\) inhibitors could play an important role in future MDR- TB regimens. In an additional study (Study C), JNJ- 2901 reduced the relapse rate when combined with the BPaL regimen, despite an inability to rescue mice when administered as a monotherapy in an intravenous model using the H37Rv reference strain (Supplementary Figure S3). There was a trend towards improvement of both the bacterial burden (BPaLJ: \(0.95 \pm 1.1\) vs BPaL: \(2.06 \pm 1.3 \log_{10}\) CFU lung- 1, \(p = 0.04\)) and a quantifiable improvement in relapse rates, that failed to reach statistical significance within the limitations of this trial design (50% vs 89%, \(p > 0.05\)), while BPaLJ was comparable to BPaLM and BPaMZ (Supplementary Table S8- 10). This suggests that BPaLJ could be considered as a potential treatment regimen irrespective of fluoroquinolone susceptibility. + +<|ref|>text<|/ref|><|det|>[[39, 558, 950, 915]]<|/det|> +In a further study, an alternative dosing strategy was used where mice were treated with BPaL for 8 weeks during an initial phase, followed by an additional 8 weeks of either BPa, BJ, B or no treatment in a continuation phase. The rationale for this study design was to reduce the time and overall amount of drug required to reach bacterial sterility. Bacterial burdens in lungs were assessed after 8, 12 and 16 weeks, and relapse rates were measured after 16 weeks of treatment plus 16 weeks following treatment cessation (Figure 1D; Supplementary Table S11; Study D). Both the BPaL/BPa and BPaL/BJ regimens trended towards lower relapse rates compared to BPaL without continuation treatment, although this was not statistically significant ( \(p\) - value=0.09). No colonies were detected after 12 weeks of treatment with BPaL/BJ, whereas with BPaL/B, colonies were detected from 3/5 mice at the end of treatment. These results suggest that inclusion of a cytochrome \(bc_{1}\) inhibitor could enhance regimen effectiveness during the continuation phase of treatment with fewer drugs and a reduced overall drug burden. Effective regimens require drugs with both bactericidal and sterilising activities to rapidly reduce bacterial loads and kill drug- tolerant bacilli. Drugs in the continuation phase must target drug- tolerant bacteria that survive the initial intensive phase14. A two- phase regimen, with an initial phase of drugs that rapidly reduce bacterial burdens, followed by sterilising drugs in a continuation phase, is a proven strategy for improving treatment outcomes for patients. The inclusion of a cytochrome \(bc_{1}\) inhibitor in the continuation phase of the BPaL/BJ regimen demonstrated a sterilising effect similar to that of the BPaL/BPa regimen, suggesting that a cytochrome \(bc_{1}\) inhibitor could serve as an effective alternative to premenopausal (Pa). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 44, 948, 360]]<|/det|> +Next, we focused on an ultra- short treatment strategy for drug sensitive- TB (DS- TB) based on drugs targeting the respiratory pathway. For this purpose, we assessed the efficacy of regimens containing telacebec alongside other drugs targeting the respiratory pathway, compared to the current DS- TB SoC; isoniazid, rifampicin, pyrazinamide, and ethambutol (HRZE; Study E). BCZ, CZT and BCZT regimens all demonstrated superior efficacy over HRZE after 8 weeks of treatment in reducing lung CFU burdens (Figure 2A, Supplementary Tables S12- 14; Study E). Both BCZ and BCZT regimens achieved \(0\%\) relapse rates 12 weeks after 6- and 8- weeks of treatment, whereas CZT required at least 12 weeks of treatment to achieve \(0\%\) relapse. One of the mice treated with HRZE remained culture positive even after 20 weeks of treatment. The benefit of adding telacebec to BCZ is highlighted by the relapse rates after 4 weeks treatment: BCZ \((60\%)\) vs. BCZT \((36\%)\) , underscoring the treatment- shortening potential of cytochrome \(bc_{1}\) inhibitors in combination with other drugs targeting the electron transport chain. Based on the TRUNCATE- TB trial strategy, a 2- month BZ- containing regimen could be as effective as the current 6- month standard (HRZE) \(^{15}\) . Adding telacebec to the CZ core regimen reduced bacterial load by \(2.2 \log_{10}\) CFU in this model and, including telacebec in CZ or BCZ regimens significantly shortened treatment compared to HRZE. These findings provide strong evidence that incorporating a cytochrome \(bc_{1}\) inhibitor into new treatment regimens offers substantial benefits which should be further investigated. + +<|ref|>text<|/ref|><|det|>[[37, 373, 950, 653]]<|/det|> +We found that the impact of cytochrome \(bc_{1}\) inhibitors may be underrepresented depending on the M. tuberculosis strain selected for investigation. Here, using another cytochrome \(bc_{1}\) inhibitor, JNJ- 4052, we demonstrate that a diverse range of clinical isolates from infected patients exhibited increased susceptibility to cytochrome \(bc_{1}\) inhibitors in both minimum inhibitory concentration (MIC) and minimum bactericidal concentration (MBC) assays when compared to the lab- adapted H37Rv strain(Figures 2B- E; Supplementary Table S15). This heightened susceptibility translated into increased in vivo efficacy in an acute mouse model, where a statistically significant CFU reduction \((1.6 \log_{10}\) reduction; \(p< 0.01\) ) was observed following cytochrome \(bc_{1}\) inhibitor (JNJ- 4052) treatment in mice inoculated with a clinical isolate (Figure 2F; Supplementary Tables S1- S3; Study F). Further work is needed to determine whether the routine use of clinical isolates should be incorporated into regimen design studies. Indeed, similar findings have recently been reported \(^{16}\) , showing that the clinical isolate HN878 is more susceptible to cytochrome \(bc_{1}\) inhibitors, with a notable contribution when added to BPaC, resulting in a 2- log reduction in lung bacterial burden between BPaC and BPaCT, after 4 weeks of treatment. These results further emphasise the relevance of clinical isolates for evaluating the in vivo efficacy of new compounds. + +<|ref|>text<|/ref|><|det|>[[40, 666, 952, 876]]<|/det|> +Together, our findings suggest that cytochrome \(bc_{1}\) inhibitors could play a crucial role in future TB treatment regimens. Specifically, cytochrome \(bc_{1}\) inhibitors may serve as effective replacements for linezolid in SoC regimens for MDR- and fluoroquinolone- resistant MDR- TB. As this represents a novel MoA, it could also help raise the barrier to resistance development to other components of the regimen and could be use for treatment of DS- TB. Our experiments containing a treatment phase followed by a continuation phase may also benefit future formulation strategies such as long acting injectables (LAI) providing an alternative strategy to improve treatment outcomes. Finally, drug repurposing has proven to be an effective strategy for accelerating the discovery of new treatments; linezolid, for example, was originally developed for vancomycin- resistant enterococcal infections. Our data highlight the potential of both clofazimine- and cytochrome \(bc_{1}\) inhibitor- containing regimens to significantly shorten treatment durations for both DS- and DR- TB. + +<|ref|>sub_title<|/ref|><|det|>[[44, 898, 195, 920]]<|/det|> +## Declarations + +<|ref|>sub_title<|/ref|><|det|>[[44, 934, 202, 951]]<|/det|> +## Acknowledgements + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 949, 145]]<|/det|> +The authors would like to thank: Amit Kaushik for technical support at Johns Hopkins University; AT Henze and PM Petrar for medical writing support at Janssen Pharmaceutica; Annelies Wouter and Lieve Lammens for toxicology input at Janssen Pharmaceutica; Nadia Neto for her feedback on WGS at Janssen Pharmaceutica; Gregory Bancroft for his intellectual input at the London School of Hygiene & Tropical Medicine; and Courtney I. Hastings at Colorado State University for technical support. + +<|ref|>sub_title<|/ref|><|det|>[[44, 161, 207, 178]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[41, 195, 953, 378]]<|/det|> +CAP, DAL, AJL, SS, GTR, NL, AU, NV, ELN, NCA AA, AC BB, LB, HP, JD and ASP were involved in the conception or the design of the study. JG and JMB designed and synthesised the compounds. AK, CV, DAL, BB, ASP and LBallell supervised the overall research programme. CAP, SS, ELN, AC, NCA, TCM, SW, NV, NW, HP, JD, AH, BS, LBrock, GG and GR participated in the collection or generation of the studies data. SS, EN, AC, TCM, SW, HP, JD, AH, VG and GTR performed the studies. CAP, SS, ELN, JE, TGC and GR contributed to the study with materials/analysis tools. CAP, AJL, NV, SW, GTR, SS, MC, ELN, NCA, LBallell, JD, HP, VC, BS, RJW and DAL were involved in the analysis or interpretation of the data. CAP, RJW and DAL wrote the manuscript. All authors reviewed and commented drafts of the manuscript for intellectual content and gave final approval to submit for publication. All authors attest they meet the ICMJE criteria for authorship. + +<|ref|>sub_title<|/ref|><|det|>[[44, 394, 170, 411]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[42, 428, 896, 467]]<|/det|> +All data generated or analysed during this study are included in this published article (and its supplementary information files). + +<|ref|>sub_title<|/ref|><|det|>[[44, 484, 277, 501]]<|/det|> +## Conflict of interest statement + +<|ref|>text<|/ref|><|det|>[[42, 517, 939, 596]]<|/det|> +JG, CV and DAL have been named inventors in a patent application for JNJ- 2901 and JNJ- 4052 compounds. CAP, CV, JG, MC, JE, NL, BS, JMB, VC, LBallell, BB, AK, ASP and DAL were/are all full- time employees of Janssen, a Johnson & Johnson company, and/or potential stockholders of Johnson & Johnson. The other authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[44, 613, 195, 630]]<|/det|> +## Funding statement + +<|ref|>text<|/ref|><|det|>[[42, 647, 945, 747]]<|/det|> +This work was supported by Janssen Pharmaceutica NV, including all costs associated with the development and publishing of the manuscript. The work at the London School of Hygiene & Tropical Medicine was supported by funding from Janssen Pharmaceutica. This project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 853903 (RespiriTB). This Joint Undertaking receives support from the European Union's Horizon 2020 research and innovation programme and EFPIA. + +<|ref|>sub_title<|/ref|><|det|>[[44, 770, 178, 792]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[55, 808, 945, 937]]<|/det|> +1. World Health Organisation, WHO consolidated guidelines on tuberculosis. Module 4: treatment - drug-resistant tuberculosis treatment. (2022). +2. World Health Organisation, Key updates to the treatment of drug-resistant tuberculosis - rapid communication. (2024). +3. Nehru, V.J., et al. Risk assessment and transmission of fluoroquinolone resistance in drug-resistant pulmonary tuberculosis: a retrospective genomic epidemiology study. Scientific Reports 14, 19719 (2024). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 44, 940, 78]]<|/det|> +4. McKee, E.E., Ferguson, M., Bentley, A.T. & Marks, T.A. Inhibition of mammalian mitochondrial protein synthesis by oxazolidinones. Antimicrob Agents Chemother 50, 2042-2049 (2006). + +<|ref|>text<|/ref|><|det|>[[58, 88, 437, 107]]<|/det|> +5. Stop TB Partnership, Clinical Pipeline. (2024). + +<|ref|>text<|/ref|><|det|>[[55, 112, 920, 133]]<|/det|> +6. de Jager, V.R., et al. Telacebec (Q203), a New Antituberculosis Agent. N Engl J Med 382, 1280-1281 (2020). + +<|ref|>text<|/ref|><|det|>[[57, 137, 891, 177]]<|/det|> +7. Chandrasekera, N.S., et al. Improved Phenoxylalkylbenzimidazoles with Activity against Mycobacterium tuberculosis Appear to Target QcrB. ACS Infectious Diseases 3, 898-916 (2017). + +<|ref|>text<|/ref|><|det|>[[57, 181, 888, 221]]<|/det|> +8. Pethe, K., et al. Discovery of Q203, a potent clinical candidate for the treatment of tuberculosis. Nature Medicine 19, 1157-1160 (2013). + +<|ref|>text<|/ref|><|det|>[[55, 225, 930, 266]]<|/det|> +9. Wani, M.A. & Dhaked, D.K. Targeting the cytochrome bc(1) complex for drug development in M. tuberculosis: review. Mol Divers 26, 2949-2965 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 270, 936, 310]]<|/det|> +10. Nyang'wa, B.-T., et al. A 24-Week, All-Oral Regimen for Rifampin-Resistant Tuberculosis. New England Journal of Medicine 387, 2331-2343 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 314, 910, 355]]<|/det|> +11. Eimer, J., et al. Association Between Increased Linezolid Plasma Concentrations and the Development of Severe Toxicity in Multidrug-Resistant Tuberculosis Treatment. Clin Infect Dis 76, e947-e956 (2023). + +<|ref|>text<|/ref|><|det|>[[48, 359, 940, 399]]<|/det|> +12. Vengurlekar, D., et al. Linezolid resistance in patients with drug-resistant TB. Int J Tuberc Lung Dis 27, 567-569 (2023). + +<|ref|>text<|/ref|><|det|>[[48, 403, 940, 444]]<|/det|> +13. Wasserman, S., et al. Linezolid toxicity in patients with drug-resistant tuberculosis: a prospective cohort study. J Antimicrob Chemother 77, 1146-1154 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 448, 925, 489]]<|/det|> +14. Sotgiu, G., Centis, R., D'Ambrosio, L. & Migliori, G.B. Tuberculosis treatment and drug regimens. Cold Spring Harb Perspect Med 5, a017822 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 493, 940, 533]]<|/det|> +15. Paton, N.I., et al. Treatment Strategy for Rifampin-Susceptible Tuberculosis. New England Journal of Medicine 388, 873-887 (2023). + +<|ref|>text<|/ref|><|det|>[[48, 538, 927, 578]]<|/det|> +16. Komm, O.D., et al. Contribution of telacebec to novel drug regimens in a murine tuberculosis model. bioRxiv, 2024.2006.2027.601059 (2024). + +<|ref|>text<|/ref|><|det|>[[48, 582, 915, 622]]<|/det|> +17. Li, S.Y., et al. Evaluation of moxifloxacin-containing regimens in pathologically distinct murine tuberculosis models. Antimicrob Agents Chemother 59, 4026-4030 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 626, 866, 647]]<|/det|> +18. Kort, F., et al. Fully weekly antituberculosis regimen: a proof-of-concept study. Eur Respir J 56(2020). + +<|ref|>text<|/ref|><|det|>[[48, 650, 945, 691]]<|/det|> +19. Akester, J.N., et al. Synthesis, Structure-Activity Relationship, and Mechanistic Studies of Aminoquinazolinones Displaying Antimycobacterial Activity. ACS Infect Dis 6, 1951-1964 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 695, 905, 736]]<|/det|> +20. Ray, P.C., et al. Spirocycle MmpL3 Inhibitors with Improved hERG and Cytotoxicity Profiles as Inhibitors of Mycobacterium tuberculosis Growth. ACS Omega 6, 2284-2311 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 740, 844, 780]]<|/det|> +21. Lamprecht, D.A., et al. Targeting de novo purine biosynthesis for tuberculosis treatment. Pre-print https://www.researchsquare.com/article/rs-4913610/v1 (2024). + +<|ref|>sub_title<|/ref|><|det|>[[44, 835, 133, 858]]<|/det|> +## Figures + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 48, 928, 460]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 481, 108, 499]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 517, 950, 664]]<|/det|> +Investigating alternative MDR- TB treatment regimens based on inclusion of a cytochrome \(bc_{1}\) inhibitor in relapsing mouse models of TB – (A) Schematic of the studies presented in this project. (B- D) Lung bacterial burden, proportions of mice with positive cultures at end of treatment and relapse in M. tuberculosis- infected mice from Studies A, B and D. CFU data is shown for week 8 (panels B- C) and week 12 (panel D). Details of treatment doses can be found in Supplementary Tables S4 and S11. Bedaquiline: B; pretomanid: Pa; clofazimine: C; linezolid: L; moxifoxacin: M; JNJ- 2901: J. SoT: Start of treatment. \*: relapse rate; 12 or 16 weeks after treatment cessation (+12 or 16 wks). Relapse rate (n/N). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 49, 950, 365]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 394, 111, 412]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[39, 431, 950, 647]]<|/det|> +Investigating alternative DS- TB treatment regimens based on inclusion of a cytochrome \(bc_{1}\) inhibitor and bactericidal activity using a clinical isolate – (A) Lung bacterial burden and relapse in M. tuberculosis- infected mice from Study E. Details of treatment doses can be found in Supplementary Table S12. Limit of detection (dotted line) was \(0.4 \log_{10} \text{CFU} \text{lung}^{-1}\) . Bedaquiline: B; clofazimine: C; rifampicin: R; ethambutol: E; isoniazid: H; telacebec: T; pyrazinamide: Z. (B) Distribution of \(\text{MIC}_{90}\) (concentration achieving \(90\%\) inhibition) values for a diverse selection of clinical isolates (black) compared with the lab- adapted WT H37Rv (cyan) for JNJ- 4052 and bedaquiline (BDQ) (Supplementary Table S15). Clinical isolate N1283 (orange) is indicated. \(n = 2\) technical replicates. Impact of (C) JNJ- 4052, (D) JNJ- 2901 and (E) telecebec on CFU counts in WT H37Rv compared to N1283 clinical isolate. \(n = \geq 2\) biological replicates. (F) In vivo efficacy (Study F) of JNJ- 4052 (50 mg \(\text{kg}^{-1}\) ; PO) in H37Rv and N1283 after 2 weeks of treatment. SoT: Start of treatment, 7 days after inoculum with 300 CFU. \(n = 3\) mice. \(** = p< 0.01\) . + +<|ref|>sub_title<|/ref|><|det|>[[44, 666, 287, 692]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 712, 700, 731]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[57, 746, 955, 835]]<|/det|> +- Insightsintotheuseofcytochrome bc1inhibitorsforfuturetherapeutictrategiesfortuberculosissupplementary.docx +- NMEDBC136846epc.pdf +- NMEDBC136846rs.pdf +- Onlinemethods.docx + +<--- Page Split ---> diff --git a/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/images_list.json b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..df8b1d0bd58bb6022baf2e57e4cb44a62d5a5c68 --- /dev/null +++ b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. (A.) Schematic illustration of a \"metafluid\" containing encapsulated gas within multistable capsules, immersed with a second fluid. (B.) Pressure \\((p_{el})\\) vs. volume \\((v_{i})\\) relation (shown by black continuous and dashed line) and the tri-linear approximation (shown by blue line) for each individual bi-stable element \\((i)\\) in the multi-stable capsule, where phase I, spinodal, and phase II regions are indicated.", + "footnote": [], + "bbox": [ + [ + 88, + 78, + 905, + 299 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Internal energy per unit mass, \\(U_{tot} / U_{0}\\) , vs pressure, \\(p / p_{atm}\\) , where \\(U_{0} = \\Phi p_{atm}v_{atm}\\) . All parameters were taken from experimental measurements and are defined below (see caption of Fig. 4). Adiabatic and isothermal cases are shown by blue and orange lines, respectively, where dashed lines denote states with bi-stable elements in the spinodal \\((n_{s} \\neq 0)\\) . Dotted curves of the corresponding colors represent the energy-pressure relation for an ideal gas. The multiple possible internal energies per external pressure allow to store, or release, energy at atmospheric conditions. The density of possible states decreases as the distance to the global minimum increases.", + "footnote": [], + "bbox": [ + [ + 292, + 75, + 705, + 312 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Pressure \\((p / p_{atm})\\) vs. density \\((\\rho /\\rho_0)\\) for stable (solid lines) and unstable (dashed lines) equilibria states. The physical parameters used are listed below (see caption of Fig. 4). (A.) Comparison between adiabatic (blue lines) and isothermal (orange lines) processes. Dotted lines represent ideal gas. (B.) Examining the effect of \\(v_{atm}\\) , the volume of the gas encapsulated in a single capsule at atmospheric pressure, and \\(k_{\\Pi}\\) , the slope of phase II. The blue line is a reference curve based on experimental values, with \\(v_{atm} = 18 \\mathrm{ml}\\) and \\(k_{\\Pi} = 152 \\mathrm{kPa / ml}\\) . The green and purple lines are pressure-density curves for fluid with decreased and increased values of \\(v_{atm}\\) of \\(12 \\mathrm{ml}\\) and \\(24 \\mathrm{ml}\\) , respectively. The red line is a pressure-density curve for a fluid with positive slope at unstable regions, achieved by setting \\(k_{\\Pi} = 2300 \\mathrm{kPa / ml}\\) , while keeping all other parameters identical to the blue reference curve. The insets show that decreasing \\(v_{atm}\\) or increasing \\(k_{\\Pi}\\) creates regions where the fluid is stable although a bi-stable element is at the spinodal region.", + "footnote": [], + "bbox": [ + [ + 90, + 75, + 908, + 386 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. (A.) Experimental and theoretical pressure \\((p / p_{atm})\\) vs. density \\((\\rho /\\rho_{0})\\) cycles. The black, blue, and red curves represent (analytically calculated) equi-permutation lines, loading, and unloading modes, respectively. Orange and green curves represent the experimental results (each curve containing three cycles). Representative frames (circular insets) taken from our experiment correspond to points (1)–(4) on the orange curve. In the experimental data, snap-open and snap-close points are \\((v_{s}^{1},p_{s}^{\\mathrm{open}}) = (0.25\\mathrm{ml},26.24\\mathrm{kPa})\\) and \\((v_{s}^{\\mathrm{II}},p_{s}^{\\mathrm{close}}) = (1.04\\mathrm{ml},\\) \\(-16.61\\mathrm{kPa})\\) , respectively. The slopes of the tri-linear approximation are \\(k_{1} = 2300\\mathrm{kPa / ml}\\) \\(k_{\\mathrm{II}} = 152\\mathrm{kPa / ml}\\) , for phase I and II, respectively, and the number of bi-stable elements per capsule is 18. (B.) The corresponding internal energy of the system, where points (1)–(4) correspond to the frames in (A). (C.) Experimental setup: the pressure controller, which is responsible for regulating the (constant) pressure in the fluid reservoir, is connected to the metafluid tank inlet via a flow valve. The height of the fluid in the fluid reservoir is monitored in the experiment to allow measuring the change in equivalent density of the metafluid. (SI - Movies)", + "footnote": [], + "bbox": [ + [ + 90, + 230, + 909, + 563 + ] + ], + "page_idx": 7 + } +] \ No newline at end of file diff --git a/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee.mmd b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee.mmd new file mode 100644 index 0000000000000000000000000000000000000000..5ea01e31beee1066d9d16e17256263214250804e --- /dev/null +++ b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee.mmd @@ -0,0 +1,222 @@ + +# A metafluid with multistable density and internal energy states + +Ofek Peretz (ofekperetz@campus.technion.ac.il) Technion - Israel Institute of Technology https://orcid.org/0000- 0002- 4420- 8706 + +Ezra Ben Abu Technion - Israel institute of technology + +Anna Zigelman Technion - Israel Institute of Technology + +Sefi Givli Technion - Israel institute of technology + +Amir Gat Technion - Israel Institute of Technology https://orcid.org/0000- 0001- 8925- 3001 + +## Article + +Keywords: artificial fluid, metafluid, stable states + +Posted Date: July 9th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 244708/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on April 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29048- 3. + +<--- Page Split ---> + +# A metafluid with multistable density and internal energy states + +Ofek Peretz \(^{1, *}\) , Ezra Ben Abu \(^{1}\) , Anna Zigelman \(^{1}\) , Sefi Givli \(^{1}\) , and Amir D. Gat \(^{1}\) + +\(^{1}\) Faculty of Mechanical Engineering, Technion - Israel Institute of Technology, Haifa 3200003, Israel. \(^{*}\) ofekperetz@campus.technion.ac.il + +## ABSTRACT + +The efficiency, operation range, and environmental safety of energy and refrigeration cycles are determined by the thermodynamic properties of available fluids. We here suggest combining gas, liquid and multistable elastic capsules to create an artificial fluid with a multitude of stable states. We study, theoretically and experimentally, the suspension's internal energy, equilibrium pressure- density relations and their stability for both adiabatic and isothermal processes. We show that the elastic multistability of the capsules endows the fluid with multistable thermodynamic properties, including the ability of capturing and storing energy at standard atmospheric conditions, not found in naturally available fluids. + +## 1 Introduction + +The thermodynamic properties of fluids are crucial to many fields, and specifically to energy and refrigeration cycles \(^{1,2}\) . Creating a fluid with exceptional properties can thus be of great practical importance, yielding new possibilities of advancement. In this work we study the properties of a suspension composed of a multitude of lubricated multistable capsules (structures capable of transforming between different equilibrium deformation patterns \(^{3,4}\) ) enclosing a compressible gas and immersed within another fluid. The thermodynamic properties of the suspension are determined by an interaction between the external liquid, the encapsulated gas, and the characteristic elastic energy profile of the capsules. By leveraging the elastic multistability of the suspended capsules, we can produce a fluid with multiple stable density points for a given pressure and temperature states as well as unstable regions. + +Multistable structures are a special class of mechanical meta- materials. Meta- materials are architected structures made from assemblies of micro- level building blocks, usually arranged in a repeated pattern. While made from standard materials, the geometry of their building blocks can give rise to unique behaviors \(^{5}\) . For example, meta- materials have been designed to have "negative mass" \(^{6 - 8}\) , to manipulate light \(^{9,10}\) and stress waves \(^{11 - 13}\) , or to provide ultra- high stiffness to weight ratio \(^{14}\) . Multistable meta- materials feature a multitude of possible equilibrium configurations for a prescribed load. These meta- stable configurations differ in the state of each mechanical unit (building block), which leads to a highly inhomogeneous and non- affine deformation field for the whole body \(^{15}\) . Hence, by careful design of their building blocks, these multi- stable structures can be "programmed" to undergo morphological changes in response to external stimuli. The ability to manufacture multi- stable meta- materials at all scales has opened exciting possibilities in a range of applications such as vibration isolation, super- elastic behavior, sensing and actuation, soft robotics, deployable structures and more \(^{16 - 25}\) . + +In this work we adopt a similar approach for the creation of multistable metafluids. We study the pressure- density relations, and stability, of a "metafluid", composed of a multitude of multistable gas- filled capsules suspended in a liquid. A simple example of a multistable capsule, used in the current study, is presented in Fig. 1, showing multiple connected bi- stable elements, produced by combining two non- identical conical frusta in opposing orientations. Bi- stability is therefore achieved by switching between extended and collapsed states through inversion of one frustum which creates a significant change in volume. Below, we fully define our model problem (§ 2), discuss our approach for energy calculation (§ 3), present our results for stable and unstable equilibria (§ 4), present experimental data of pressure cycles (§ 5), and provide concluding remarks (§ 6). + +<--- Page Split ---> + +## 2 Problem definition + +We study the thermodynamic pressure- density relations and their stability for a fluidic suspension composed of multistable elastic capsules, enclosing a compressible gas and immersed in a liquid, as seen in Fig. 1. Each capsule is composed of \(n\) identical bi- stable elastic elements. For the sake of simplicity, body forces are hereafter neglected. We examine a control volume which includes a large number of capsules, so that homogenization may be applied, while keeping all physical properties spatially uniform. The density of the external liquid is denoted by \(\rho_{I}\) . The number of capsules per- unit- mass is denoted by \(\Phi\) . The total mass of the capsules per suspension mass \(m\) can be calculated via \(\Phi\) , as \(m\Phi m_{c}\) , where \(m_{c}\) is the mass of a single capsule. We define \(\Psi\) as the external fluid mass ratio. Thus, the mass of the external liquid is \(m_{l} = \Psi m\) and the total mass of the capsules is \(m(1 - \Psi)\) . Thus we obtain a \(\Phi - \Psi\) relation of \(\Phi = (1 - \Psi) / m_{c}\) . + +Since the external fluid is assumed to be incompressible and thus its volume is constant, the equivalent density \(\rho\) of the suspension can be expressed by + +\[\rho = \left[\frac{1 - \Phi m_{c}}{\rho_{I}} +\Phi \left(\nu_{a} + \sum_{i = 1}^{n}\nu_{i}\right)\right]^{-1} \quad (1)\] + +where \(\nu_{i}\) is the gas- filled volume contained within bi- stable element \(i\) , and \(\nu_{a}\) is all the additional gas and solid volume of the capsule. + +The volume \(\nu_{i}\) is a multi- valued function of the difference between the pressure of the internal gas, denoted as \(p_{gas}\) and the pressure in the external fluid, denoted simply as \(p\) . We thus define this pressure difference by \(p_{el} = p_{gas} - p\) . We approximate \(\nu_{i}\) by the tri- linear model (see Fig. 1B) which is a common and useful approximation that enables a simplified analytic treatment of bi- stability. More specifically, according to this model, when considering the energy relation to the pressure \(p_{el}\) and element volume \(\nu_{i}\) , the energy function may be approximated by three quadratic functionals \(^{15,26 - 30}\) leading to linear pressure- volume relation. + +The tri- linear curve is defined by the stiffnesses \(k_{\mathrm{I}}\) and \(k_{\mathrm{II}}\) , and the stability threshold points \((\nu_{s}^{\mathrm{I}},p_{s}^{\mathrm{open}})\) and \((\nu_{s}^{\mathrm{II}},p_{s}^{\mathrm{close}})\) in phases I and II, respectively (see definitions in Fig. 1B). Thus, the possible pressure range for bi- stable elements in phase I is \(p_{el}< p_{s}^{\mathrm{open}}\) , in the Spinodal region \(p_{s}^{\mathrm{close}}< p_{el}< p_{s}^{\mathrm{open}}\) , and in phase II \(p_{el} > p_{s}^{\mathrm{close}}\) . These coefficients are commonly evaluated by fitting to an experiment \(^{20,31,32}\) . Thus, the pressure- volume function for each of the bi- stable elements is given by, + +\[p_{el} = \left\{ \begin{array}{l l}{p_{el}^{I}}\\ {p_{el}^{II}}\\ {p_{el}^{III}} \end{array} \right\} = \left\{ \begin{array}{l l}{k_{\mathrm{I}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{I}}\right) + p_{\mathrm{s}}^{\mathrm{open}}}\\ {k_{\mathrm{s}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{I}}\right) + p_{\mathrm{s}}^{\mathrm{open}}}\\ {k_{\mathrm{II}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{II}}\right) + p_{\mathrm{s}}^{\mathrm{close}}} \end{array} \right\} ,\qquad \nu_{\mathrm{s}}^{\mathrm{I}}< \nu_{\mathrm{i}}< \nu_{\mathrm{s}}^{\mathrm{II}} \quad (2)\] + +where \(k_{s}\) is determined by requiring continuity between the spinodal region and I and II, yielding \(k_{s} = (p_{s}^{\mathrm{close}} - p_{s}^{\mathrm{open}}) / (\nu_{s}^{\mathrm{II}} - \nu_{s}^{\mathrm{I}}) < 0\) . + +Next, we address the total gas- filled volume of a capsule, which is composed of the sum of all volumes of the \(n\) bi- stable elements (given by \(\sum_{i = 1}^{n}\nu_{i}\) and additional gas- filled volume \(\nu_{a,gas}\) ). Since each element of the capsule can be in one of the three possible phases, the volume of a capsule is determined by the combination of all elements' states, denoted hereafter as the multistable state permutation \(\overline{p\vec{e}^{r}}\) . The permutation + +\[\overline{p\vec{e}^{r}} = \{n_{1},n_{s},n_{\mathrm{II}}\} \quad (3)\] + +is defined by the number of elements in phase I \((n_{\mathrm{I}})\) , in the spinodal region \((n_{s})\) , and in phase II \((n_{\mathrm{II}})\) , where \(n_{\mathrm{I}} + n_{s} + n_{\mathrm{II}} = n\) . Thus, a capsule with 18 bistable elements (as in the presented experiments) has 190 possible permutations. Considering that the pressure acting on all elements is identical and equal to \(p_{el}\) , we can calculate the volume of an element in each phase according to equation (2). Denoting the volume in each phase by \(\nu_{\mathrm{I}}(p_{el}),\nu_{\mathrm{s}}(p_{el}),\nu_{\mathrm{II}}(p_{el})\) , we obtain the total volume of gas within a capsule as \(\overline{p\vec{e}^{r}}\cdot \{\nu_{\mathrm{I}}(p_{el}),\nu_{\mathrm{s}}(p_{el}),\nu_{\mathrm{II}}(p_{el})\} + \nu_{a,gas}\) . + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. (A.) Schematic illustration of a "metafluid" containing encapsulated gas within multistable capsules, immersed with a second fluid. (B.) Pressure \((p_{el})\) vs. volume \((v_{i})\) relation (shown by black continuous and dashed line) and the tri-linear approximation (shown by blue line) for each individual bi-stable element \((i)\) in the multi-stable capsule, where phase I, spinodal, and phase II regions are indicated.
+ +## 3 Internal energy + +We now calculate the internal energy for all of the possible permutations \(\{n_{1},n_{s},n_{\mathrm{II}}\}\) of the suspension. For the current analysis, we neglect the compressibility of the external fluid and assume that all capsules are identical. Thus, the internal energy of the suspension is composed of the total energy of the ideal gas, denoted by \(U_{gas}\) , and the energy of the multistable elastic capsules, denoted by \(U_{el}\) . + +The elastic energy \(U_{el}\) of a bi- stable element \(i\) for the three possible phases is given by integration of (2) over \(v_{i}\) , yielding + +\[\left\{ \begin{array}{l}U_{el}^{I}\\ U_{el}^{S}\\ U_{el}^{\mathrm{II}} \end{array} \right\} = \left\{ \begin{array}{l}k_{1}\left(v_{i} - v_{s}^{\mathrm{I}}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{i}\\ k_{s}\left(v_{i} - v_{s}^{\mathrm{I}}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{i}\\ k_{\mathrm{II}}\left(v_{i} - v_{s}^{\mathrm{II}}\right)^{2} / 2 + p_{s}^{\mathrm{close}}v_{i} \end{array} \right\} ,\quad v_{s}^{\mathrm{I}}< v_{i}< v_{s}^{\mathrm{II}}\\ v_{s}^{\mathrm{II}}< v_{i} \quad (4)\] + +and the total elastic potential energy per- unit- mass is thus + +\[U_{el} = \Phi \cdot \overrightarrow{p e^{i}}\cdot \{U_{el}^{I},U_{el}^{S},U_{el}^{\mathrm{II}}\} . \quad (5)\] + +To calculate \(U_{gas}\) we assume an ideal gas model, \(p_{gas} = \rho_{gas}RT\) . For isothermal processes, we thus obtain the relation \(p_{gas} = m_{g}RT / (v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\})\) , where \(R\) is the specific gas constant, \(T\) is the constant temperature, \(m_{g}\) is the mass of the gas within the capsule. For adiabatic processes, assuming ideal gas and isentropic compression/expansion, the gas pressure is given by \(p_{gas} / p_{atm} = [v_{atm} / (v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\})]^{\gamma}\) , where \(p_{atm}\) is atmospheric pressure, and \(v_{atm}\) is the volume taken by the gas contained within a single capsule at atmospheric pressure. In the adiabatic case, the temperature of the gas within the capsules will change during pressure variations according to \(T = T_{0}(p / p_{0})^{1 - 1 / \gamma}\) , but not the temperature of the external incompressible fluid. These spatial temperature variations may yield heat transfer between the gas, solid and liquid, and more complex dynamics. To allow simple treatment of this case, we assume in the current analysis that an adiabatic process corresponds to the more strict requirement of no heat transfer from the encapsulated gas. + +The gas energy per- unit- mass of the suspension is defined as + +\[U_{gas} = -\Phi \int_{v_{atm}}^{v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\}} p_{gas}dv \quad (6)\] + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Internal energy per unit mass, \(U_{tot} / U_{0}\) , vs pressure, \(p / p_{atm}\) , where \(U_{0} = \Phi p_{atm}v_{atm}\) . All parameters were taken from experimental measurements and are defined below (see caption of Fig. 4). Adiabatic and isothermal cases are shown by blue and orange lines, respectively, where dashed lines denote states with bi-stable elements in the spinodal \((n_{s} \neq 0)\) . Dotted curves of the corresponding colors represent the energy-pressure relation for an ideal gas. The multiple possible internal energies per external pressure allow to store, or release, energy at atmospheric conditions. The density of possible states decreases as the distance to the global minimum increases.
+ +and thus the total energy of the system for an adiabatic process is given by \(U_{tot} = U_{gas} + U_{el} + U_{ex}\) , + +\[U_{tot} = \frac{\Phi p_{atm}v_{atm}}{\gamma - 1}\left[\left(\frac{v_{atm}}{v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}\right)^{\gamma - 1} - 1\right] + \Phi \left\{ \begin{array}{l}k_{1}\left(v_{1} - v_{s}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{1}\\ k_{s}\left(v_{s} - v_{s}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{s}\\ k_{\mathrm{II}}\left(v_{\mathrm{II}} - v_{\mathrm{II}}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{close}}v_{\mathrm{II}} \end{array} \right\} \cdot \overline{p e^{\hat{r}}} \\ +\Phi p\left[v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\} - v_{atm}\right] \quad (7)\] + +where the last RHS term is the external work energy \(U_{ex}\) and \(p\) is the dictated pressure in the external fluid. For an isothermal process the first RHS term is replaced with \(\Phi p_{atm}\ln \left[(v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\} \right) / v_{atm}]\) . + +Equilibrium of the system occurs when the elastic force in all bistable elements is balanced by the difference between the external pressure, and the pressure of the internal gas. In the current analysis, all bi- stable elements are identical, and can be in one of three possible states (phase I, spinodal, and phase II). Thus, the equilibrium values in each of the three possible states can be calculated from the following relations, + +\[k_{\mathrm{I}}\left(v_{\mathrm{I}} - v_{\mathrm{s}}^{1}\right) + p_{s}^{\mathrm{open}} = k_{s}\left(v_{\mathrm{s}} - v_{\mathrm{s}}^{1}\right) + p_{s}^{\mathrm{open}}\equiv k_{\mathrm{II}}\left(v_{\mathrm{II}} - v_{\mathrm{s}}^{\mathrm{II}}\right) + p_{s}^{\mathrm{close}} = p_{atm}\left[\frac{v_{atm}}{v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}\right]^{\gamma} - p. \quad (8)\] + +Equation (8) describes an adiabatic process, and for an isothermal process the last most RHS term is replaced with \((m_{g}RT) / (v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\})\) . + +Solving equations (8) for the equilibrium values \(v_{\mathrm{I}}\) , \(v_{\mathrm{s}}\) and \(v_{\mathrm{II}}\) , and substituting these into the energy equation (7) allows to calculate the energy of the suspension per given permutation \(\overline{p e^{\hat{r}}} = \{n_{1},n_{s},n_{\mathrm{II}}\}\) and external pressure \(p\) . Each permutation is only valid in a specific range of external pressures, as defined in equation (4). In Fig. 2 we present all of the available energy equilibria states, for multistable capsules composed of 18 bi- stable elements, with 190 possible permutations. Both isothermal (orange lines) and adiabatic (blue lines) processes are presented. The lines are determined by changing the external pressure, for different permutations of the phases of the bi- stable elements. The results show that the multi- stability extends the possible state of an ideal gas from a single line to a region containing multitude of possible energies. Different permutations provide different energy- pressure + +<--- Page Split ---> + +curves, corresponding to a fluid with different properties. Changing the properties of the fluid can be achieved by increasing/reducing the pressure beyond the range of a specific permutation. In addition to changing fluid properties, such transitions between different equilibrium states allow the fluid to store and release energy, as discussed below. + +## 4 Stability and the equation of state + +We examine the stability of equilibrium states by considering the Hessian of the total energy \(U_{tot}\) , given by \(H_{ij} = \partial^{2}U_{tot} / \partial v_{i}\partial v_{j}\) . An equilibrium configuration is stable if it corresponds to a local minimum of the energy, requiring positive definite Hessian. In order to obtain the Hessian in terms of all \(n\) degrees of freedom, \(v_{i}\) , we note that \(\overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\} = \sum_{j = 1}^{n}v_{j}\) . Substituting this relation into equation (7) and differentiating the total energy with respect to the degrees of freedom, \(v_{i}\) , yields the following set of \(n\) equations + +\[\frac{\partial U_{tot}}{\partial v_{i}} = p_{el}^{\alpha (i)}\left(v_{i}\right) + \left(\frac{v_{atm}}{v_{a,gas} + \sum_{j = 1}^{n}v_{j}}\right) - p = 0,\quad i = 1,\ldots ,n, \quad (9)\] + +where \(\alpha (i)\) denotes the phase of bi- stable element \(i\) , \(\alpha (i) \in \{1, s, \Pi \}\) . The set of \(n\) equations in (9) is differentiated again in order to obtain the Hessian matrix, + +\[\frac{\partial^{2}U_{tot}}{\partial v_{i}\partial v_{j}} = \delta_{ij}k_{\alpha (i)} + k_{g},\quad i,j = 1,\ldots ,n, \quad (10)\] + +where \(k_{\alpha (i)}\) \((i = 1,2,\dots,n)\) is the stiffness of the \(i\) - th bistable element and \(k_{g}\) is the (instantaneous) stiffness of the gas at the equilibrium configuration. For an adiabatic process \(k_{g} = (\gamma p_{atm} / v_{atm})[v_{atm} / (v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\})]^{\gamma +1}\) and for isothermal process \(k_{g} = v_{atm}p_{atm} / (v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\})^2\) . + +An equilibrium state is stable if and only if the Hessian is positive definite, which occurs if and only if all leading principal minors \(A_{j}\) (the determinant of the submatrix of first \(j\) rows and \(j\) columns) are positive. The principal minors of the current configuration in (10) are + +\[A_{j} = \left(\prod_{i = 1}^{j}k_{\alpha (i)}\right)\left[1 + k_{g}\sum_{i = 1}^{j}\frac{1}{k_{\alpha (i)}}\right]. \quad (11)\] + +In order to obtain conditions for positive \(A_{j}\) , we examine separately three possible permutation states. In the first state the permutation is of the form \((n_{1},0,n_{\mathrm{II}})\) and thus there are no elements in the spinodal region. In this case, since all \(k_{\alpha (i)}\) and \(k_{g}\) are positive, all minors \(A_{j}\) in (11) are necessarily positive for all \(p\) . Another possible permutation state is the case of two or more elements in the spinodal region. The symmetry in terms of the degrees of freedom allows us to arbitrarily rearrange the degrees of freedom. We re- order the Hessian so that the first two elements are in the spinodal region, and thus \(k_{\alpha (1)} = k_{\alpha (2)} = k_{s}< 0\) . The first minor in this case is given by \(A_{1} = k_{g}k_{s}(1 / k_{g} + 1 / k_{s})\) and is positive only if \(k_{g} > |k_{s}|\) . The second minor \(A_{2} = k_{g}(k_{s})^{2}(1 / k_{g} + 2 / k_{s}))\) is positive only if \(k_{g}< |k_{s}| / 2\) . Thus, since the two conditions cannot hold simultaneously, if there are at least two elements in the spinodal region, the configuration is unstable for all \(k_{g}\) . Finally, we examine the case of permutations of the form \((n_{\mathrm{I}},1,n_{\mathrm{II}})\) , meaning that only a single element is in the spinodal region. We re- order the Hessian so that the \(n\) - th element is in the spinodal region. Thus all minors \(A_{j}\) , \(j\in \{1,\ldots ,n - 1\}\) , are positive. The requirement that the minor \(A_{n}\) is positive, which is needed for the stability condition to be satisfied, may be expressed for an adiabatic process as + +\[\frac{v_{atm}}{\gamma p_{atm}}\left(\frac{v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}{\gamma_{atm}}\right)^{\gamma +1} + \overline{p e^{j}}\cdot \left\{\frac{1}{k_{1}},\frac{1}{k_{s}},\frac{1}{k_{\mathrm{II}}}\right\} < 0 \quad (12)\] + +where \(\{v_{1},v_{s},v_{\mathrm{II}}\}\) are functions of the external pressure \(p\) , and are calculated from equation (8). For isothermal process the first LHS term is replaced with \(\left(v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\}\right)^{2} / \left(v_{atm}p_{atm}\right)\) . Obtaining the stability condition + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Pressure \((p / p_{atm})\) vs. density \((\rho /\rho_0)\) for stable (solid lines) and unstable (dashed lines) equilibria states. The physical parameters used are listed below (see caption of Fig. 4). (A.) Comparison between adiabatic (blue lines) and isothermal (orange lines) processes. Dotted lines represent ideal gas. (B.) Examining the effect of \(v_{atm}\) , the volume of the gas encapsulated in a single capsule at atmospheric pressure, and \(k_{\Pi}\) , the slope of phase II. The blue line is a reference curve based on experimental values, with \(v_{atm} = 18 \mathrm{ml}\) and \(k_{\Pi} = 152 \mathrm{kPa / ml}\) . The green and purple lines are pressure-density curves for fluid with decreased and increased values of \(v_{atm}\) of \(12 \mathrm{ml}\) and \(24 \mathrm{ml}\) , respectively. The red line is a pressure-density curve for a fluid with positive slope at unstable regions, achieved by setting \(k_{\Pi} = 2300 \mathrm{kPa / ml}\) , while keeping all other parameters identical to the blue reference curve. The insets show that decreasing \(v_{atm}\) or increasing \(k_{\Pi}\) creates regions where the fluid is stable although a bi-stable element is at the spinodal region.
+ +allows us to present all possible density- pressure relations which correspond to all permutations of the form \(\overline{pe^r} = (n_1, 0, n_{\Pi})\) or permutations of the form \(\overline{pe^r} = (n_1, 1, n_{\Pi})\) along with the condition (12). + +The pressure- density relations of the fluid are presented in Fig. 3, where unstable states are marked by dashed lines and stable states are marked by solid lines. In Fig. 3A we present the external pressure \(p\) vs density \(\rho\) for adiabatic (blue lines) and isothermal (orange lines) processes. In addition, pressure- density relations for gas only are presented by dotted lines with corresponding colors. The distance between stable points of the fluid is minimal near the permutation \(\overline{pe^r} = (0, 0, n)\) and increases as we approach the reversed order permutation \(\overline{pe^r} = (n, 0, 0)\) . In Fig. 2B, we show adiabatic processes, which were obtained for three different initial gas volumes within the capsules at atmospheric conditions (blue line is the reference curve, purple line has decreased amount of gas and green have increased amount of gas). The effect of the amount of gas becomes more pronounced as the density increases and the average slope of the multi- stable curve increases with the addition of gas into the capsules. In addition, it is possible to see that permutations of the form \(\overline{pe^r} = (n_1, 1, n_{\Pi})\) become stable as the density of the fluid increases (see insets within Fig. 2B). The red curve in Fig. 2B, presents modified \(k_{\Pi}\) . In this case, a permutation with unstable element \((n_s = 1)\) results in a stable state of the system. The insets show that decreasing \(v_{atm}\) or increasing \(k_{\Pi}\) creates regions where the fluid is stable although a bi- stable element is at the spinodal region. In both panels, several stable densities are possible for a given pressure, as well as several pressures for a single density. This feature appears for unstable curves with positive slopes. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. (A.) Experimental and theoretical pressure \((p / p_{atm})\) vs. density \((\rho /\rho_{0})\) cycles. The black, blue, and red curves represent (analytically calculated) equi-permutation lines, loading, and unloading modes, respectively. Orange and green curves represent the experimental results (each curve containing three cycles). Representative frames (circular insets) taken from our experiment correspond to points (1)–(4) on the orange curve. In the experimental data, snap-open and snap-close points are \((v_{s}^{1},p_{s}^{\mathrm{open}}) = (0.25\mathrm{ml},26.24\mathrm{kPa})\) and \((v_{s}^{\mathrm{II}},p_{s}^{\mathrm{close}}) = (1.04\mathrm{ml},\) \(-16.61\mathrm{kPa})\) , respectively. The slopes of the tri-linear approximation are \(k_{1} = 2300\mathrm{kPa / ml}\) \(k_{\mathrm{II}} = 152\mathrm{kPa / ml}\) , for phase I and II, respectively, and the number of bi-stable elements per capsule is 18. (B.) The corresponding internal energy of the system, where points (1)–(4) correspond to the frames in (A). (C.) Experimental setup: the pressure controller, which is responsible for regulating the (constant) pressure in the fluid reservoir, is connected to the metafluid tank inlet via a flow valve. The height of the fluid in the fluid reservoir is monitored in the experiment to allow measuring the change in equivalent density of the metafluid. (SI - Movies)
+ +<--- Page Split ---> + +## 5 Pressure cycles and experimental demonstration + +In Fig. 4 we present analytic and experimental results for cyclical pressurization of the fluid. In panel 4A, each of the black curves denotes a stable equilibrium state corresponding to a different permutation, while the blue and red curves follow the actual loading and unloading paths, respectively, involving switching between different permutations. In order to access one of the multiple possible densities per a given pressure, we need to reach the specific permutation on which the desired density- pressure point is located. This can be achieved by increasing the pressure beyond the initial snap- close value associated with the current permutation (blue line in Fig. 4A), until reaching a stable permutation for the given pressure value. At this point, reducing the pressure will not change the permutation as long as the pressure is above the snap- open value of the current permutation (red line in Fig. 4A) and the system will remain at the same permutation (following the black line in Fig. 4A associated with that permutation). Thus, multi- stable pressure cycles are closed curves which involve intervals on constant permutation curves (black lines), intervals on the snap- close curves (blue lines), and intervals on the snap- open curves (red lines). This unique ability to choose a specific operation curve is demonstrated experimentally in what follows. + +The experimental setup (Fig 4C) consists of a cubic \(35\mathrm{x}30\mathrm{x}3.2\mathrm{cm}\) fluid- filled tank, a pressure flow controller, and a fluid reservoir. The capsules, illustrated in Fig 1, yield the following physical parameters: \(\nu_{s}^{\mathrm{I}} = 0.51\mathrm{ml}\) \(\nu_{s}^{\mathrm{II}} = 1.04\mathrm{ml}\) \(k_{\mathrm{I}} = 2300\mathrm{kPa / ml}\) \(k_{\mathrm{II}} = 152\mathrm{kPa / ml}\) and are composed of \(n = 18\) bi- stable elements. The surrounding fluid is water \((\rho_{l} = 0.997\mathrm{g / ml})\) and the internal fluid is air \((\gamma = 1.4,R = 0.286\mathrm{J / (gK)})\) . The fluid container is connected to a pressure controller (Elveflow OB1), which is connected to the tank. Using this setup, we experimentally measured the snapping pressures (see definitions in Fig. 1) as \(p_{s}^{\mathrm{open}} = 26.24\mathrm{kPa}\) for the snap- open value and \(p_{s}^{\mathrm{close}} = - 16.61\) kPa for the snap- close value, with standard deviation of \(\pm 0.1\mathrm{kPa}\) for both values. The measurements were based on averaging 6 experiments, where we used air actuation to eliminate transient fluidic effects. To obtain the change in density we measured the height of the fluid in the reservoir which is equivalent to the change in the volume of the capsules (since the surrounding fluid is assumed incompressible). Throughout the experiment, the height of the liquid in the fluid reservoir and the pressure in the tank were continuously monitored. By using the density of the surrounding fluid and the volume of the metafluid tank we calculated the equivalent density as a function of the pressure. The increase of tank volume due to elasticity of the walls was accounted for by pressurizing the tank without capsules, and measuring the volume- pressure relation. + +In Fig. 4A, the orange and green curves present experimental results of two different pressure- density cycles of a multi- stable fluid. Each experimental curve presents three identical pressure cycles, repeated in each experiment. In Experiment #1, the initial state of the capsules at atmospheric conditions yields equivalent density of \(\rho_{0} = 0.719\) \(\mathrm{g / ml}\) and permutation of \(\overline{{p e^{\prime}r_{0}}} = \{5,0,13\}\) (see subpanel A(1)). We pressurized the tank linearly from \(p_{0} = 101\mathrm{kPa}\) to \(171\mathrm{kPa}\) over a period of 30 seconds. The fluid density increased and the permutations changed until reaching a steady state permutation of \(\overline{{p e^{\prime}r}} = \{14,0,4\}\) and density of \(\rho = 1.07\rho_{0}\) (see subpanel A(2)). At this stage, we reduced the pressure back to \(p_{0}\) over a period of 30 seconds. In agreement with the model, the permutation of the fluid remained nearly unchanged \((\overline{{p e^{\prime}r}} = \{13,0,5\})\) while the density decreased to \(\rho = 1.037\rho_{0}\) (see subpanel A(3)). To return to the initial state, we applied a pressure of \(30\mathrm{kPa}\) which allowed us to change the state back to the initial fluid permutation of \(\overline{{p e^{\prime}r}} = \{5,0,13\}\) and the density of \(\rho = 0.95\rho_{0}\) (see subpanel A(4)). Finally, increasing the pressure back to \(101\mathrm{kPa}\) compressed the fluid along a constant permutation. Similarly, in Experiment #2 we applied a smaller cycle, in which points 1- 4 have the densities of \(\rho = \rho_{0},\rho = 1.04\rho_{0},\rho = 0.97\rho_{0},\rho = 1.03\rho_{0}\) and the permutations of \(\{7,0,11\} ,\{11,0,7\} ,\{11,0,7\} ,\{7,0,11\}\) , respectively. As it is evident from Fig. 4, a good agreement is observed between the theoretic stability states of the system and the experimental results for both cycles (see SI- Movies). + +In panel 4B we present the internal energy associated with the different equilibrium states. The internal energy stored within the fluid at atmospheric conditions in the initial state, point 1, is released by adding sufficient energy to the fluid reaching a different permutation, point 2. From point 2, energy is released along a single permutation until reaching point 3, which is the minimal energy at the atmospheric conditions. Similarly, storing energy within the fluid can be achieved by temporarily adding energy to change permutations, and reaching point 4. Reducing the pressure back to atmospheric pressure will not change the fluid permutation, and some of the energy will thus + +<--- Page Split ---> + +remain stored within the fluid. + +## 6 Concluding remarks + +In this study, we introduced a new concept for a multistable metafluid, and investigated theoretically and experimentally the relation between pressure and density, for both adiabatic and isothermal processes. We showed that the multitude of stable density and energy states make it possible to capture and store energy at standard atmospheric conditions, which is not possible with naturally available fluids. + +The experiments presented in this work involved large \(O(cm)\) particles. While these experiments aimed at demonstrating the concept, miniaturization of the multistable particles is, evidently, needed for creating metafluids for real applications. The results of our experiments suggest that the concept of multistable metafluids paves the way for the creation of futuristic non- polluting fluids with enhanced properties for energy and cooling cycles, leveraging multi- stability to harvest, store and release energy and heat. + +The code used in the work are available in SI Appendix Code + +## References + +1. Chen, H., Goswami, D. Y. & Stefanakos, E. K. A review of thermodynamic cycles and working fluids for the conversion of low-grade heat. Renew. Sustain. Energy Rev. 14, 3059-3067 (2010). + +2. Becker, T. M. et al. Absorption refrigeration cycles with ammonia-ionic liquid working pairs studied by molecular simulation. Ind. Eng. Chem. Res. 57, 5442-5452 (2018). + +3. Peretz, O., Mishra, A. K., Shepherd, R. F. & Gat, A. D. Underactuated fluidic control of a continuous multistable membrane. PNAS 117, 5217-5221 (2020). + +4. Guest, S. D. & Pellegrino, S. Analytical models for bistable cylindrical shells. Proc. R. Soc. A 462, 839-854 (2006). + +5. Lee, J. H., Singer, J. P. & Thomas, E. Micro-nanostructured mechanical metamaterials. Adv. Mater. 24, 4782-4810 (2012). + +6. Correa, D. M. et al. Negative stiffness honeycombs for recoverable shock isolation. Rapid Prototyp. J. 21, 193-200 (2015). + +7. Correa, D. M., Seepersad, C. C. & Haberman, M. Mechanical design of negative stiffness honeycomb materials. Integrating Mater. Manuf. Innov. 4, 10 (2015). + +8. Babaee, S. et al. 3d soft metamaterials with negative poisson's ratio. Rapid Prototyp. J. 25, 5044-5049 (2013). + +9. Arpin, K. A. et al. Multidimensional architectures for functional optical devices. Adv. Mater. 22, 1084-1101 (2010). + +10. Soukoulis, C. M. & Wegener, M. Past achievements and future challenges in the development of three-dimensional photonic metamaterials. Nat. Photonics 5, 523-530 (2011). + +11. Peng, P. et al. Acoustic tunneling through artificial structures: From phononic crystals to acoustic metamaterials. Solid State Commun. 151, 400-403 (2011). + +12. Lu, M. H., Feng, L. & Chen, Y. F. Phononic crystals and acoustic metamaterials. Mater. Today 12, 34-42 (2009). + +13. Deymier, P. Acoustic metamaterials and phononic crystals, in springer series in solid-state sciences. Springer Berlin Heidelberg: Imprint: Springer: Berlin, Heidelberg. (2013). + +14. Zheng, X. et al. Ultralight, ultrastiff mechanical metamaterials. Science 344, 1373-1377 (2014). + +15. Puglisi, G. & Truskinovsky, L. Mechanics of a discrete chain with bi-stable elements. J. Mech. Phys. Solids 48, 1-27 (2000). + +<--- Page Split ---> + +16. Frenzel, T., Findeisen, C., Kadic, M., Gumbsch, P. & Wegener, M. Tailored buckling microlattices as reusable light-weight shock absorbers. Adv. Mater. 28, 5865-5870 (2016).17. Ha, C. S., Lakes, R. S. & Plesha, M. E. Design, fabrication, and analysis of lattice exhibiting energy absorption via snap-through behavior. Mater. Des. 141, 426-437 (2018).18. Leelavanichkul, S., Chekarev, A., Adams, D. O. & Solzbacher, F. Energy absorption of a helicoidal bistable structure. J. Mech. Mater. Struct. 5, 305-321 (2010).19. Yang, H. & Ma, L. Multi-stable mechanical metamaterials by elastic buckling instability. J. Mater. Sci. 54, 3509-3526 (2019).20. Cohen, T. & Givli, S. Dynamics of a discrete chain of bi-stable elements: a biomimetic shock absorbing mechanism. J. Mech. Phys. Solids 64, 426-439 (2014).21. Katz, S. & Givli, S. Boomerons in a 1-d lattice with only nearest-neighbor interactions. EPL 131, 64002 (2020).22. Shan, S. et al. Multistable architected materials for trapping elastic strain energy. Adv. Mater. 27, 4296-4301 (2015).23. Rafsanjani, A., Akbarzadeh, A. & Pasini, D. Snapping mechanical metamaterials under tension. Adv. Mater. 27, 5931-5935 (2015).24. Pan, F. et al. 3d pixel mechanical metamaterials. Adv. Mater. 31, 1900548 (2019).25. James, K. & Waisman, H. Layout design of a bi-stable cardiovascular stent using topology optimization. Comput. Methods Appl. Mech. Eng. 305, 869-890 (2016).26. Müller, I. & Xu, H. On the pseudo-elastic hysteresis. Acta metallurgica et materialia 39, 263-271 (1991).27. Fedelich, B. & Zanzotto, G. Hysteresis in discrete systems of possibly interacting elements with a double-well energy. J. Nonlinear Sci. 2, 319-342 (1992).28. Allinger, T. L., Epstein, M. & Herzog, W. Stability of muscle fibers on the descending limb of the force-length relation. a theoretical consideration. J. biomechanics 29, 627-633 (1996).29. Müller, I. & Villaggio, P. A model for an elastic-plastic body. Arch. for Ration. Mech. Analysis 65, 25-46 (1977).30. Shaw, J. A. & Kyriakides, S. On the nucleation and propagation of phase transformation fronts in a nit alloy. Acta materialia 45, 683-700 (1997).31. Luongo, A., Casciati, S. & Zulli, D. Perturbation method for the dynamic analysis of a bistable oscillator under slow harmonic excitation. Smart structures systems 18, 183-196 (2016).32. Benichou, I., Faran, E., Shilo, D. & Givli, S. Application of a bi-stable chain model for the analysis of jerky twin boundary motion in nimnga. Appl. Phys. Lett. 102, 011912 (2013). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SIMovieX2. mp4SIMovieX1. mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee_det.mmd b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..805c5d255241d638e62ba6c6519cb132402ed14b --- /dev/null +++ b/preprint/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee/preprint__12c1e2591b902566bcf874c6e0723a71afe8a65a7daa8e0b67beac7dee3391ee_det.mmd @@ -0,0 +1,307 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 886, 175]]<|/det|> +# A metafluid with multistable density and internal energy states + +<|ref|>text<|/ref|><|det|>[[44, 195, 760, 238]]<|/det|> +Ofek Peretz (ofekperetz@campus.technion.ac.il) Technion - Israel Institute of Technology https://orcid.org/0000- 0002- 4420- 8706 + +<|ref|>text<|/ref|><|det|>[[44, 243, 400, 285]]<|/det|> +Ezra Ben Abu Technion - Israel institute of technology + +<|ref|>text<|/ref|><|det|>[[44, 291, 404, 333]]<|/det|> +Anna Zigelman Technion - Israel Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 338, 400, 379]]<|/det|> +Sefi Givli Technion - Israel institute of technology + +<|ref|>text<|/ref|><|det|>[[44, 384, 760, 426]]<|/det|> +Amir Gat Technion - Israel Institute of Technology https://orcid.org/0000- 0001- 8925- 3001 + +<|ref|>sub_title<|/ref|><|det|>[[44, 465, 102, 482]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 502, 464, 520]]<|/det|> +Keywords: artificial fluid, metafluid, stable states + +<|ref|>text<|/ref|><|det|>[[44, 540, 283, 559]]<|/det|> +Posted Date: July 9th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 578, 463, 596]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 244708/v1 + +<|ref|>text<|/ref|><|det|>[[44, 615, 910, 656]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 694, 950, 737]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on April 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29048- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[90, 74, 842, 136]]<|/det|> +# A metafluid with multistable density and internal energy states + +<|ref|>text<|/ref|><|det|>[[90, 145, 789, 167]]<|/det|> +Ofek Peretz \(^{1, *}\) , Ezra Ben Abu \(^{1}\) , Anna Zigelman \(^{1}\) , Sefi Givli \(^{1}\) , and Amir D. Gat \(^{1}\) + +<|ref|>text<|/ref|><|det|>[[90, 185, 805, 217]]<|/det|> +\(^{1}\) Faculty of Mechanical Engineering, Technion - Israel Institute of Technology, Haifa 3200003, Israel. \(^{*}\) ofekperetz@campus.technion.ac.il + +<|ref|>sub_title<|/ref|><|det|>[[90, 240, 198, 258]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[94, 275, 904, 383]]<|/det|> +The efficiency, operation range, and environmental safety of energy and refrigeration cycles are determined by the thermodynamic properties of available fluids. We here suggest combining gas, liquid and multistable elastic capsules to create an artificial fluid with a multitude of stable states. We study, theoretically and experimentally, the suspension's internal energy, equilibrium pressure- density relations and their stability for both adiabatic and isothermal processes. We show that the elastic multistability of the capsules endows the fluid with multistable thermodynamic properties, including the ability of capturing and storing energy at standard atmospheric conditions, not found in naturally available fluids. + +<|ref|>sub_title<|/ref|><|det|>[[91, 416, 226, 434]]<|/det|> +## 1 Introduction + +<|ref|>text<|/ref|><|det|>[[90, 441, 909, 595]]<|/det|> +The thermodynamic properties of fluids are crucial to many fields, and specifically to energy and refrigeration cycles \(^{1,2}\) . Creating a fluid with exceptional properties can thus be of great practical importance, yielding new possibilities of advancement. In this work we study the properties of a suspension composed of a multitude of lubricated multistable capsules (structures capable of transforming between different equilibrium deformation patterns \(^{3,4}\) ) enclosing a compressible gas and immersed within another fluid. The thermodynamic properties of the suspension are determined by an interaction between the external liquid, the encapsulated gas, and the characteristic elastic energy profile of the capsules. By leveraging the elastic multistability of the suspended capsules, we can produce a fluid with multiple stable density points for a given pressure and temperature states as well as unstable regions. + +<|ref|>text<|/ref|><|det|>[[90, 596, 910, 784]]<|/det|> +Multistable structures are a special class of mechanical meta- materials. Meta- materials are architected structures made from assemblies of micro- level building blocks, usually arranged in a repeated pattern. While made from standard materials, the geometry of their building blocks can give rise to unique behaviors \(^{5}\) . For example, meta- materials have been designed to have "negative mass" \(^{6 - 8}\) , to manipulate light \(^{9,10}\) and stress waves \(^{11 - 13}\) , or to provide ultra- high stiffness to weight ratio \(^{14}\) . Multistable meta- materials feature a multitude of possible equilibrium configurations for a prescribed load. These meta- stable configurations differ in the state of each mechanical unit (building block), which leads to a highly inhomogeneous and non- affine deformation field for the whole body \(^{15}\) . Hence, by careful design of their building blocks, these multi- stable structures can be "programmed" to undergo morphological changes in response to external stimuli. The ability to manufacture multi- stable meta- materials at all scales has opened exciting possibilities in a range of applications such as vibration isolation, super- elastic behavior, sensing and actuation, soft robotics, deployable structures and more \(^{16 - 25}\) . + +<|ref|>text<|/ref|><|det|>[[90, 785, 910, 921]]<|/det|> +In this work we adopt a similar approach for the creation of multistable metafluids. We study the pressure- density relations, and stability, of a "metafluid", composed of a multitude of multistable gas- filled capsules suspended in a liquid. A simple example of a multistable capsule, used in the current study, is presented in Fig. 1, showing multiple connected bi- stable elements, produced by combining two non- identical conical frusta in opposing orientations. Bi- stability is therefore achieved by switching between extended and collapsed states through inversion of one frustum which creates a significant change in volume. Below, we fully define our model problem (§ 2), discuss our approach for energy calculation (§ 3), present our results for stable and unstable equilibria (§ 4), present experimental data of pressure cycles (§ 5), and provide concluding remarks (§ 6). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[89, 77, 284, 96]]<|/det|> +## 2 Problem definition + +<|ref|>text<|/ref|><|det|>[[88, 102, 910, 259]]<|/det|> +We study the thermodynamic pressure- density relations and their stability for a fluidic suspension composed of multistable elastic capsules, enclosing a compressible gas and immersed in a liquid, as seen in Fig. 1. Each capsule is composed of \(n\) identical bi- stable elastic elements. For the sake of simplicity, body forces are hereafter neglected. We examine a control volume which includes a large number of capsules, so that homogenization may be applied, while keeping all physical properties spatially uniform. The density of the external liquid is denoted by \(\rho_{I}\) . The number of capsules per- unit- mass is denoted by \(\Phi\) . The total mass of the capsules per suspension mass \(m\) can be calculated via \(\Phi\) , as \(m\Phi m_{c}\) , where \(m_{c}\) is the mass of a single capsule. We define \(\Psi\) as the external fluid mass ratio. Thus, the mass of the external liquid is \(m_{l} = \Psi m\) and the total mass of the capsules is \(m(1 - \Psi)\) . Thus we obtain a \(\Phi - \Psi\) relation of \(\Phi = (1 - \Psi) / m_{c}\) . + +<|ref|>text<|/ref|><|det|>[[88, 259, 907, 294]]<|/det|> +Since the external fluid is assumed to be incompressible and thus its volume is constant, the equivalent density \(\rho\) of the suspension can be expressed by + +<|ref|>equation<|/ref|><|det|>[[133, 305, 907, 355]]<|/det|> +\[\rho = \left[\frac{1 - \Phi m_{c}}{\rho_{I}} +\Phi \left(\nu_{a} + \sum_{i = 1}^{n}\nu_{i}\right)\right]^{-1} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[88, 364, 907, 399]]<|/det|> +where \(\nu_{i}\) is the gas- filled volume contained within bi- stable element \(i\) , and \(\nu_{a}\) is all the additional gas and solid volume of the capsule. + +<|ref|>text<|/ref|><|det|>[[88, 399, 910, 504]]<|/det|> +The volume \(\nu_{i}\) is a multi- valued function of the difference between the pressure of the internal gas, denoted as \(p_{gas}\) and the pressure in the external fluid, denoted simply as \(p\) . We thus define this pressure difference by \(p_{el} = p_{gas} - p\) . We approximate \(\nu_{i}\) by the tri- linear model (see Fig. 1B) which is a common and useful approximation that enables a simplified analytic treatment of bi- stability. More specifically, according to this model, when considering the energy relation to the pressure \(p_{el}\) and element volume \(\nu_{i}\) , the energy function may be approximated by three quadratic functionals \(^{15,26 - 30}\) leading to linear pressure- volume relation. + +<|ref|>text<|/ref|><|det|>[[88, 503, 909, 589]]<|/det|> +The tri- linear curve is defined by the stiffnesses \(k_{\mathrm{I}}\) and \(k_{\mathrm{II}}\) , and the stability threshold points \((\nu_{s}^{\mathrm{I}},p_{s}^{\mathrm{open}})\) and \((\nu_{s}^{\mathrm{II}},p_{s}^{\mathrm{close}})\) in phases I and II, respectively (see definitions in Fig. 1B). Thus, the possible pressure range for bi- stable elements in phase I is \(p_{el}< p_{s}^{\mathrm{open}}\) , in the Spinodal region \(p_{s}^{\mathrm{close}}< p_{el}< p_{s}^{\mathrm{open}}\) , and in phase II \(p_{el} > p_{s}^{\mathrm{close}}\) . These coefficients are commonly evaluated by fitting to an experiment \(^{20,31,32}\) . Thus, the pressure- volume function for each of the bi- stable elements is given by, + +<|ref|>equation<|/ref|><|det|>[[133, 612, 906, 672]]<|/det|> +\[p_{el} = \left\{ \begin{array}{l l}{p_{el}^{I}}\\ {p_{el}^{II}}\\ {p_{el}^{III}} \end{array} \right\} = \left\{ \begin{array}{l l}{k_{\mathrm{I}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{I}}\right) + p_{\mathrm{s}}^{\mathrm{open}}}\\ {k_{\mathrm{s}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{I}}\right) + p_{\mathrm{s}}^{\mathrm{open}}}\\ {k_{\mathrm{II}}\left(\nu_{\mathrm{i}} - \nu_{\mathrm{s}}^{\mathrm{II}}\right) + p_{\mathrm{s}}^{\mathrm{close}}} \end{array} \right\} ,\qquad \nu_{\mathrm{s}}^{\mathrm{I}}< \nu_{\mathrm{i}}< \nu_{\mathrm{s}}^{\mathrm{II}} \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[88, 684, 907, 720]]<|/det|> +where \(k_{s}\) is determined by requiring continuity between the spinodal region and I and II, yielding \(k_{s} = (p_{s}^{\mathrm{close}} - p_{s}^{\mathrm{open}}) / (\nu_{s}^{\mathrm{II}} - \nu_{s}^{\mathrm{I}}) < 0\) . + +<|ref|>text<|/ref|><|det|>[[88, 720, 909, 789]]<|/det|> +Next, we address the total gas- filled volume of a capsule, which is composed of the sum of all volumes of the \(n\) bi- stable elements (given by \(\sum_{i = 1}^{n}\nu_{i}\) and additional gas- filled volume \(\nu_{a,gas}\) ). Since each element of the capsule can be in one of the three possible phases, the volume of a capsule is determined by the combination of all elements' states, denoted hereafter as the multistable state permutation \(\overline{p\vec{e}^{r}}\) . The permutation + +<|ref|>equation<|/ref|><|det|>[[133, 800, 906, 822]]<|/det|> +\[\overline{p\vec{e}^{r}} = \{n_{1},n_{s},n_{\mathrm{II}}\} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[88, 835, 909, 923]]<|/det|> +is defined by the number of elements in phase I \((n_{\mathrm{I}})\) , in the spinodal region \((n_{s})\) , and in phase II \((n_{\mathrm{II}})\) , where \(n_{\mathrm{I}} + n_{s} + n_{\mathrm{II}} = n\) . Thus, a capsule with 18 bistable elements (as in the presented experiments) has 190 possible permutations. Considering that the pressure acting on all elements is identical and equal to \(p_{el}\) , we can calculate the volume of an element in each phase according to equation (2). Denoting the volume in each phase by \(\nu_{\mathrm{I}}(p_{el}),\nu_{\mathrm{s}}(p_{el}),\nu_{\mathrm{II}}(p_{el})\) , we obtain the total volume of gas within a capsule as \(\overline{p\vec{e}^{r}}\cdot \{\nu_{\mathrm{I}}(p_{el}),\nu_{\mathrm{s}}(p_{el}),\nu_{\mathrm{II}}(p_{el})\} + \nu_{a,gas}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[88, 78, 905, 299]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 311, 904, 382]]<|/det|> +
Figure 1. (A.) Schematic illustration of a "metafluid" containing encapsulated gas within multistable capsules, immersed with a second fluid. (B.) Pressure \((p_{el})\) vs. volume \((v_{i})\) relation (shown by black continuous and dashed line) and the tri-linear approximation (shown by blue line) for each individual bi-stable element \((i)\) in the multi-stable capsule, where phase I, spinodal, and phase II regions are indicated.
+ +<|ref|>sub_title<|/ref|><|det|>[[89, 407, 253, 426]]<|/det|> +## 3 Internal energy + +<|ref|>text<|/ref|><|det|>[[88, 431, 910, 502]]<|/det|> +We now calculate the internal energy for all of the possible permutations \(\{n_{1},n_{s},n_{\mathrm{II}}\}\) of the suspension. For the current analysis, we neglect the compressibility of the external fluid and assume that all capsules are identical. Thus, the internal energy of the suspension is composed of the total energy of the ideal gas, denoted by \(U_{gas}\) , and the energy of the multistable elastic capsules, denoted by \(U_{el}\) . + +<|ref|>text<|/ref|><|det|>[[88, 501, 910, 536]]<|/det|> +The elastic energy \(U_{el}\) of a bi- stable element \(i\) for the three possible phases is given by integration of (2) over \(v_{i}\) , yielding + +<|ref|>equation<|/ref|><|det|>[[135, 546, 905, 611]]<|/det|> +\[\left\{ \begin{array}{l}U_{el}^{I}\\ U_{el}^{S}\\ U_{el}^{\mathrm{II}} \end{array} \right\} = \left\{ \begin{array}{l}k_{1}\left(v_{i} - v_{s}^{\mathrm{I}}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{i}\\ k_{s}\left(v_{i} - v_{s}^{\mathrm{I}}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{i}\\ k_{\mathrm{II}}\left(v_{i} - v_{s}^{\mathrm{II}}\right)^{2} / 2 + p_{s}^{\mathrm{close}}v_{i} \end{array} \right\} ,\quad v_{s}^{\mathrm{I}}< v_{i}< v_{s}^{\mathrm{II}}\\ v_{s}^{\mathrm{II}}< v_{i} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[88, 621, 504, 639]]<|/det|> +and the total elastic potential energy per- unit- mass is thus + +<|ref|>equation<|/ref|><|det|>[[133, 650, 907, 673]]<|/det|> +\[U_{el} = \Phi \cdot \overrightarrow{p e^{i}}\cdot \{U_{el}^{I},U_{el}^{S},U_{el}^{\mathrm{II}}\} . \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[88, 686, 910, 858]]<|/det|> +To calculate \(U_{gas}\) we assume an ideal gas model, \(p_{gas} = \rho_{gas}RT\) . For isothermal processes, we thus obtain the relation \(p_{gas} = m_{g}RT / (v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\})\) , where \(R\) is the specific gas constant, \(T\) is the constant temperature, \(m_{g}\) is the mass of the gas within the capsule. For adiabatic processes, assuming ideal gas and isentropic compression/expansion, the gas pressure is given by \(p_{gas} / p_{atm} = [v_{atm} / (v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\})]^{\gamma}\) , where \(p_{atm}\) is atmospheric pressure, and \(v_{atm}\) is the volume taken by the gas contained within a single capsule at atmospheric pressure. In the adiabatic case, the temperature of the gas within the capsules will change during pressure variations according to \(T = T_{0}(p / p_{0})^{1 - 1 / \gamma}\) , but not the temperature of the external incompressible fluid. These spatial temperature variations may yield heat transfer between the gas, solid and liquid, and more complex dynamics. To allow simple treatment of this case, we assume in the current analysis that an adiabatic process corresponds to the more strict requirement of no heat transfer from the encapsulated gas. + +<|ref|>text<|/ref|><|det|>[[122, 858, 552, 875]]<|/det|> +The gas energy per- unit- mass of the suspension is defined as + +<|ref|>equation<|/ref|><|det|>[[133, 886, 907, 925]]<|/det|> +\[U_{gas} = -\Phi \int_{v_{atm}}^{v_{a,gas} + \overrightarrow{p e^{i}}\cdot \{v_{I},v_{s},v_{\mathrm{II}}\}} p_{gas}dv \quad (6)\] + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[292, 75, 705, 312]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 324, 911, 429]]<|/det|> +
Figure 2. Internal energy per unit mass, \(U_{tot} / U_{0}\) , vs pressure, \(p / p_{atm}\) , where \(U_{0} = \Phi p_{atm}v_{atm}\) . All parameters were taken from experimental measurements and are defined below (see caption of Fig. 4). Adiabatic and isothermal cases are shown by blue and orange lines, respectively, where dashed lines denote states with bi-stable elements in the spinodal \((n_{s} \neq 0)\) . Dotted curves of the corresponding colors represent the energy-pressure relation for an ideal gas. The multiple possible internal energies per external pressure allow to store, or release, energy at atmospheric conditions. The density of possible states decreases as the distance to the global minimum increases.
+ +<|ref|>text<|/ref|><|det|>[[88, 452, 794, 472]]<|/det|> +and thus the total energy of the system for an adiabatic process is given by \(U_{tot} = U_{gas} + U_{el} + U_{ex}\) , + +<|ref|>equation<|/ref|><|det|>[[133, 496, 907, 584]]<|/det|> +\[U_{tot} = \frac{\Phi p_{atm}v_{atm}}{\gamma - 1}\left[\left(\frac{v_{atm}}{v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}\right)^{\gamma - 1} - 1\right] + \Phi \left\{ \begin{array}{l}k_{1}\left(v_{1} - v_{s}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{1}\\ k_{s}\left(v_{s} - v_{s}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{open}}v_{s}\\ k_{\mathrm{II}}\left(v_{\mathrm{II}} - v_{\mathrm{II}}^{1}\right)^{2} / 2 + p_{s}^{\mathrm{close}}v_{\mathrm{II}} \end{array} \right\} \cdot \overline{p e^{\hat{r}}} \\ +\Phi p\left[v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\} - v_{atm}\right] \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[88, 590, 907, 627]]<|/det|> +where the last RHS term is the external work energy \(U_{ex}\) and \(p\) is the dictated pressure in the external fluid. For an isothermal process the first RHS term is replaced with \(\Phi p_{atm}\ln \left[(v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\} \right) / v_{atm}]\) . + +<|ref|>text<|/ref|><|det|>[[88, 627, 909, 694]]<|/det|> +Equilibrium of the system occurs when the elastic force in all bistable elements is balanced by the difference between the external pressure, and the pressure of the internal gas. In the current analysis, all bi- stable elements are identical, and can be in one of three possible states (phase I, spinodal, and phase II). Thus, the equilibrium values in each of the three possible states can be calculated from the following relations, + +<|ref|>equation<|/ref|><|det|>[[133, 700, 907, 741]]<|/det|> +\[k_{\mathrm{I}}\left(v_{\mathrm{I}} - v_{\mathrm{s}}^{1}\right) + p_{s}^{\mathrm{open}} = k_{s}\left(v_{\mathrm{s}} - v_{\mathrm{s}}^{1}\right) + p_{s}^{\mathrm{open}}\equiv k_{\mathrm{II}}\left(v_{\mathrm{II}} - v_{\mathrm{s}}^{\mathrm{II}}\right) + p_{s}^{\mathrm{close}} = p_{atm}\left[\frac{v_{atm}}{v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}\right]^{\gamma} - p. \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[88, 750, 907, 785]]<|/det|> +Equation (8) describes an adiabatic process, and for an isothermal process the last most RHS term is replaced with \((m_{g}RT) / (v_{a,gas} + \overline{p e^{\hat{r}}} \cdot \{v_{1},v_{s},v_{\mathrm{II}}\})\) . + +<|ref|>text<|/ref|><|det|>[[88, 785, 910, 923]]<|/det|> +Solving equations (8) for the equilibrium values \(v_{\mathrm{I}}\) , \(v_{\mathrm{s}}\) and \(v_{\mathrm{II}}\) , and substituting these into the energy equation (7) allows to calculate the energy of the suspension per given permutation \(\overline{p e^{\hat{r}}} = \{n_{1},n_{s},n_{\mathrm{II}}\}\) and external pressure \(p\) . Each permutation is only valid in a specific range of external pressures, as defined in equation (4). In Fig. 2 we present all of the available energy equilibria states, for multistable capsules composed of 18 bi- stable elements, with 190 possible permutations. Both isothermal (orange lines) and adiabatic (blue lines) processes are presented. The lines are determined by changing the external pressure, for different permutations of the phases of the bi- stable elements. The results show that the multi- stability extends the possible state of an ideal gas from a single line to a region containing multitude of possible energies. Different permutations provide different energy- pressure + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 78, 910, 132]]<|/det|> +curves, corresponding to a fluid with different properties. Changing the properties of the fluid can be achieved by increasing/reducing the pressure beyond the range of a specific permutation. In addition to changing fluid properties, such transitions between different equilibrium states allow the fluid to store and release energy, as discussed below. + +<|ref|>sub_title<|/ref|><|det|>[[88, 147, 426, 167]]<|/det|> +## 4 Stability and the equation of state + +<|ref|>text<|/ref|><|det|>[[88, 172, 910, 260]]<|/det|> +We examine the stability of equilibrium states by considering the Hessian of the total energy \(U_{tot}\) , given by \(H_{ij} = \partial^{2}U_{tot} / \partial v_{i}\partial v_{j}\) . An equilibrium configuration is stable if it corresponds to a local minimum of the energy, requiring positive definite Hessian. In order to obtain the Hessian in terms of all \(n\) degrees of freedom, \(v_{i}\) , we note that \(\overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\} = \sum_{j = 1}^{n}v_{j}\) . Substituting this relation into equation (7) and differentiating the total energy with respect to the degrees of freedom, \(v_{i}\) , yields the following set of \(n\) equations + +<|ref|>equation<|/ref|><|det|>[[133, 269, 907, 316]]<|/det|> +\[\frac{\partial U_{tot}}{\partial v_{i}} = p_{el}^{\alpha (i)}\left(v_{i}\right) + \left(\frac{v_{atm}}{v_{a,gas} + \sum_{j = 1}^{n}v_{j}}\right) - p = 0,\quad i = 1,\ldots ,n, \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[88, 325, 907, 360]]<|/det|> +where \(\alpha (i)\) denotes the phase of bi- stable element \(i\) , \(\alpha (i) \in \{1, s, \Pi \}\) . The set of \(n\) equations in (9) is differentiated again in order to obtain the Hessian matrix, + +<|ref|>equation<|/ref|><|det|>[[133, 383, 907, 421]]<|/det|> +\[\frac{\partial^{2}U_{tot}}{\partial v_{i}\partial v_{j}} = \delta_{ij}k_{\alpha (i)} + k_{g},\quad i,j = 1,\ldots ,n, \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[88, 431, 908, 486]]<|/det|> +where \(k_{\alpha (i)}\) \((i = 1,2,\dots,n)\) is the stiffness of the \(i\) - th bistable element and \(k_{g}\) is the (instantaneous) stiffness of the gas at the equilibrium configuration. For an adiabatic process \(k_{g} = (\gamma p_{atm} / v_{atm})[v_{atm} / (v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\})]^{\gamma +1}\) and for isothermal process \(k_{g} = v_{atm}p_{atm} / (v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\})^2\) . + +<|ref|>text<|/ref|><|det|>[[88, 485, 907, 537]]<|/det|> +An equilibrium state is stable if and only if the Hessian is positive definite, which occurs if and only if all leading principal minors \(A_{j}\) (the determinant of the submatrix of first \(j\) rows and \(j\) columns) are positive. The principal minors of the current configuration in (10) are + +<|ref|>equation<|/ref|><|det|>[[132, 545, 907, 593]]<|/det|> +\[A_{j} = \left(\prod_{i = 1}^{j}k_{\alpha (i)}\right)\left[1 + k_{g}\sum_{i = 1}^{j}\frac{1}{k_{\alpha (i)}}\right]. \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[88, 601, 910, 825]]<|/det|> +In order to obtain conditions for positive \(A_{j}\) , we examine separately three possible permutation states. In the first state the permutation is of the form \((n_{1},0,n_{\mathrm{II}})\) and thus there are no elements in the spinodal region. In this case, since all \(k_{\alpha (i)}\) and \(k_{g}\) are positive, all minors \(A_{j}\) in (11) are necessarily positive for all \(p\) . Another possible permutation state is the case of two or more elements in the spinodal region. The symmetry in terms of the degrees of freedom allows us to arbitrarily rearrange the degrees of freedom. We re- order the Hessian so that the first two elements are in the spinodal region, and thus \(k_{\alpha (1)} = k_{\alpha (2)} = k_{s}< 0\) . The first minor in this case is given by \(A_{1} = k_{g}k_{s}(1 / k_{g} + 1 / k_{s})\) and is positive only if \(k_{g} > |k_{s}|\) . The second minor \(A_{2} = k_{g}(k_{s})^{2}(1 / k_{g} + 2 / k_{s}))\) is positive only if \(k_{g}< |k_{s}| / 2\) . Thus, since the two conditions cannot hold simultaneously, if there are at least two elements in the spinodal region, the configuration is unstable for all \(k_{g}\) . Finally, we examine the case of permutations of the form \((n_{\mathrm{I}},1,n_{\mathrm{II}})\) , meaning that only a single element is in the spinodal region. We re- order the Hessian so that the \(n\) - th element is in the spinodal region. Thus all minors \(A_{j}\) , \(j\in \{1,\ldots ,n - 1\}\) , are positive. The requirement that the minor \(A_{n}\) is positive, which is needed for the stability condition to be satisfied, may be expressed for an adiabatic process as + +<|ref|>equation<|/ref|><|det|>[[133, 834, 907, 876]]<|/det|> +\[\frac{v_{atm}}{\gamma p_{atm}}\left(\frac{v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\}}{\gamma_{atm}}\right)^{\gamma +1} + \overline{p e^{j}}\cdot \left\{\frac{1}{k_{1}},\frac{1}{k_{s}},\frac{1}{k_{\mathrm{II}}}\right\} < 0 \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[88, 885, 907, 923]]<|/det|> +where \(\{v_{1},v_{s},v_{\mathrm{II}}\}\) are functions of the external pressure \(p\) , and are calculated from equation (8). For isothermal process the first LHS term is replaced with \(\left(v_{a,gas} + \overline{p e^{j}}\cdot \{v_{1},v_{s},v_{\mathrm{II}}\}\right)^{2} / \left(v_{atm}p_{atm}\right)\) . Obtaining the stability condition + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 75, 908, 386]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 396, 911, 570]]<|/det|> +
Figure 3. Pressure \((p / p_{atm})\) vs. density \((\rho /\rho_0)\) for stable (solid lines) and unstable (dashed lines) equilibria states. The physical parameters used are listed below (see caption of Fig. 4). (A.) Comparison between adiabatic (blue lines) and isothermal (orange lines) processes. Dotted lines represent ideal gas. (B.) Examining the effect of \(v_{atm}\) , the volume of the gas encapsulated in a single capsule at atmospheric pressure, and \(k_{\Pi}\) , the slope of phase II. The blue line is a reference curve based on experimental values, with \(v_{atm} = 18 \mathrm{ml}\) and \(k_{\Pi} = 152 \mathrm{kPa / ml}\) . The green and purple lines are pressure-density curves for fluid with decreased and increased values of \(v_{atm}\) of \(12 \mathrm{ml}\) and \(24 \mathrm{ml}\) , respectively. The red line is a pressure-density curve for a fluid with positive slope at unstable regions, achieved by setting \(k_{\Pi} = 2300 \mathrm{kPa / ml}\) , while keeping all other parameters identical to the blue reference curve. The insets show that decreasing \(v_{atm}\) or increasing \(k_{\Pi}\) creates regions where the fluid is stable although a bi-stable element is at the spinodal region.
+ +<|ref|>text<|/ref|><|det|>[[88, 597, 907, 633]]<|/det|> +allows us to present all possible density- pressure relations which correspond to all permutations of the form \(\overline{pe^r} = (n_1, 0, n_{\Pi})\) or permutations of the form \(\overline{pe^r} = (n_1, 1, n_{\Pi})\) along with the condition (12). + +<|ref|>text<|/ref|><|det|>[[88, 634, 910, 892]]<|/det|> +The pressure- density relations of the fluid are presented in Fig. 3, where unstable states are marked by dashed lines and stable states are marked by solid lines. In Fig. 3A we present the external pressure \(p\) vs density \(\rho\) for adiabatic (blue lines) and isothermal (orange lines) processes. In addition, pressure- density relations for gas only are presented by dotted lines with corresponding colors. The distance between stable points of the fluid is minimal near the permutation \(\overline{pe^r} = (0, 0, n)\) and increases as we approach the reversed order permutation \(\overline{pe^r} = (n, 0, 0)\) . In Fig. 2B, we show adiabatic processes, which were obtained for three different initial gas volumes within the capsules at atmospheric conditions (blue line is the reference curve, purple line has decreased amount of gas and green have increased amount of gas). The effect of the amount of gas becomes more pronounced as the density increases and the average slope of the multi- stable curve increases with the addition of gas into the capsules. In addition, it is possible to see that permutations of the form \(\overline{pe^r} = (n_1, 1, n_{\Pi})\) become stable as the density of the fluid increases (see insets within Fig. 2B). The red curve in Fig. 2B, presents modified \(k_{\Pi}\) . In this case, a permutation with unstable element \((n_s = 1)\) results in a stable state of the system. The insets show that decreasing \(v_{atm}\) or increasing \(k_{\Pi}\) creates regions where the fluid is stable although a bi- stable element is at the spinodal region. In both panels, several stable densities are possible for a given pressure, as well as several pressures for a single density. This feature appears for unstable curves with positive slopes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 230, 909, 563]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 575, 911, 764]]<|/det|> +
Figure 4. (A.) Experimental and theoretical pressure \((p / p_{atm})\) vs. density \((\rho /\rho_{0})\) cycles. The black, blue, and red curves represent (analytically calculated) equi-permutation lines, loading, and unloading modes, respectively. Orange and green curves represent the experimental results (each curve containing three cycles). Representative frames (circular insets) taken from our experiment correspond to points (1)–(4) on the orange curve. In the experimental data, snap-open and snap-close points are \((v_{s}^{1},p_{s}^{\mathrm{open}}) = (0.25\mathrm{ml},26.24\mathrm{kPa})\) and \((v_{s}^{\mathrm{II}},p_{s}^{\mathrm{close}}) = (1.04\mathrm{ml},\) \(-16.61\mathrm{kPa})\) , respectively. The slopes of the tri-linear approximation are \(k_{1} = 2300\mathrm{kPa / ml}\) \(k_{\mathrm{II}} = 152\mathrm{kPa / ml}\) , for phase I and II, respectively, and the number of bi-stable elements per capsule is 18. (B.) The corresponding internal energy of the system, where points (1)–(4) correspond to the frames in (A). (C.) Experimental setup: the pressure controller, which is responsible for regulating the (constant) pressure in the fluid reservoir, is connected to the metafluid tank inlet via a flow valve. The height of the fluid in the fluid reservoir is monitored in the experiment to allow measuring the change in equivalent density of the metafluid. (SI - Movies)
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 78, 567, 98]]<|/det|> +## 5 Pressure cycles and experimental demonstration + +<|ref|>text<|/ref|><|det|>[[88, 106, 910, 312]]<|/det|> +In Fig. 4 we present analytic and experimental results for cyclical pressurization of the fluid. In panel 4A, each of the black curves denotes a stable equilibrium state corresponding to a different permutation, while the blue and red curves follow the actual loading and unloading paths, respectively, involving switching between different permutations. In order to access one of the multiple possible densities per a given pressure, we need to reach the specific permutation on which the desired density- pressure point is located. This can be achieved by increasing the pressure beyond the initial snap- close value associated with the current permutation (blue line in Fig. 4A), until reaching a stable permutation for the given pressure value. At this point, reducing the pressure will not change the permutation as long as the pressure is above the snap- open value of the current permutation (red line in Fig. 4A) and the system will remain at the same permutation (following the black line in Fig. 4A associated with that permutation). Thus, multi- stable pressure cycles are closed curves which involve intervals on constant permutation curves (black lines), intervals on the snap- close curves (blue lines), and intervals on the snap- open curves (red lines). This unique ability to choose a specific operation curve is demonstrated experimentally in what follows. + +<|ref|>text<|/ref|><|det|>[[88, 315, 910, 555]]<|/det|> +The experimental setup (Fig 4C) consists of a cubic \(35\mathrm{x}30\mathrm{x}3.2\mathrm{cm}\) fluid- filled tank, a pressure flow controller, and a fluid reservoir. The capsules, illustrated in Fig 1, yield the following physical parameters: \(\nu_{s}^{\mathrm{I}} = 0.51\mathrm{ml}\) \(\nu_{s}^{\mathrm{II}} = 1.04\mathrm{ml}\) \(k_{\mathrm{I}} = 2300\mathrm{kPa / ml}\) \(k_{\mathrm{II}} = 152\mathrm{kPa / ml}\) and are composed of \(n = 18\) bi- stable elements. The surrounding fluid is water \((\rho_{l} = 0.997\mathrm{g / ml})\) and the internal fluid is air \((\gamma = 1.4,R = 0.286\mathrm{J / (gK)})\) . The fluid container is connected to a pressure controller (Elveflow OB1), which is connected to the tank. Using this setup, we experimentally measured the snapping pressures (see definitions in Fig. 1) as \(p_{s}^{\mathrm{open}} = 26.24\mathrm{kPa}\) for the snap- open value and \(p_{s}^{\mathrm{close}} = - 16.61\) kPa for the snap- close value, with standard deviation of \(\pm 0.1\mathrm{kPa}\) for both values. The measurements were based on averaging 6 experiments, where we used air actuation to eliminate transient fluidic effects. To obtain the change in density we measured the height of the fluid in the reservoir which is equivalent to the change in the volume of the capsules (since the surrounding fluid is assumed incompressible). Throughout the experiment, the height of the liquid in the fluid reservoir and the pressure in the tank were continuously monitored. By using the density of the surrounding fluid and the volume of the metafluid tank we calculated the equivalent density as a function of the pressure. The increase of tank volume due to elasticity of the walls was accounted for by pressurizing the tank without capsules, and measuring the volume- pressure relation. + +<|ref|>text<|/ref|><|det|>[[88, 558, 910, 815]]<|/det|> +In Fig. 4A, the orange and green curves present experimental results of two different pressure- density cycles of a multi- stable fluid. Each experimental curve presents three identical pressure cycles, repeated in each experiment. In Experiment #1, the initial state of the capsules at atmospheric conditions yields equivalent density of \(\rho_{0} = 0.719\) \(\mathrm{g / ml}\) and permutation of \(\overline{{p e^{\prime}r_{0}}} = \{5,0,13\}\) (see subpanel A(1)). We pressurized the tank linearly from \(p_{0} = 101\mathrm{kPa}\) to \(171\mathrm{kPa}\) over a period of 30 seconds. The fluid density increased and the permutations changed until reaching a steady state permutation of \(\overline{{p e^{\prime}r}} = \{14,0,4\}\) and density of \(\rho = 1.07\rho_{0}\) (see subpanel A(2)). At this stage, we reduced the pressure back to \(p_{0}\) over a period of 30 seconds. In agreement with the model, the permutation of the fluid remained nearly unchanged \((\overline{{p e^{\prime}r}} = \{13,0,5\})\) while the density decreased to \(\rho = 1.037\rho_{0}\) (see subpanel A(3)). To return to the initial state, we applied a pressure of \(30\mathrm{kPa}\) which allowed us to change the state back to the initial fluid permutation of \(\overline{{p e^{\prime}r}} = \{5,0,13\}\) and the density of \(\rho = 0.95\rho_{0}\) (see subpanel A(4)). Finally, increasing the pressure back to \(101\mathrm{kPa}\) compressed the fluid along a constant permutation. Similarly, in Experiment #2 we applied a smaller cycle, in which points 1- 4 have the densities of \(\rho = \rho_{0},\rho = 1.04\rho_{0},\rho = 0.97\rho_{0},\rho = 1.03\rho_{0}\) and the permutations of \(\{7,0,11\} ,\{11,0,7\} ,\{11,0,7\} ,\{7,0,11\}\) , respectively. As it is evident from Fig. 4, a good agreement is observed between the theoretic stability states of the system and the experimental results for both cycles (see SI- Movies). + +<|ref|>text<|/ref|><|det|>[[89, 818, 909, 922]]<|/det|> +In panel 4B we present the internal energy associated with the different equilibrium states. The internal energy stored within the fluid at atmospheric conditions in the initial state, point 1, is released by adding sufficient energy to the fluid reaching a different permutation, point 2. From point 2, energy is released along a single permutation until reaching point 3, which is the minimal energy at the atmospheric conditions. Similarly, storing energy within the fluid can be achieved by temporarily adding energy to change permutations, and reaching point 4. Reducing the pressure back to atmospheric pressure will not change the fluid permutation, and some of the energy will thus + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 81, 306, 96]]<|/det|> +remain stored within the fluid. + +<|ref|>sub_title<|/ref|><|det|>[[90, 116, 301, 135]]<|/det|> +## 6 Concluding remarks + +<|ref|>text<|/ref|><|det|>[[90, 141, 909, 210]]<|/det|> +In this study, we introduced a new concept for a multistable metafluid, and investigated theoretically and experimentally the relation between pressure and density, for both adiabatic and isothermal processes. We showed that the multitude of stable density and energy states make it possible to capture and store energy at standard atmospheric conditions, which is not possible with naturally available fluids. + +<|ref|>text<|/ref|><|det|>[[90, 210, 909, 295]]<|/det|> +The experiments presented in this work involved large \(O(cm)\) particles. While these experiments aimed at demonstrating the concept, miniaturization of the multistable particles is, evidently, needed for creating metafluids for real applications. The results of our experiments suggest that the concept of multistable metafluids paves the way for the creation of futuristic non- polluting fluids with enhanced properties for energy and cooling cycles, leveraging multi- stability to harvest, store and release energy and heat. + +<|ref|>text<|/ref|><|det|>[[116, 296, 559, 312]]<|/det|> +The code used in the work are available in SI Appendix Code + +<|ref|>sub_title<|/ref|><|det|>[[90, 331, 198, 349]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[92, 355, 910, 391]]<|/det|> +1. Chen, H., Goswami, D. Y. & Stefanakos, E. K. A review of thermodynamic cycles and working fluids for the conversion of low-grade heat. Renew. Sustain. Energy Rev. 14, 3059-3067 (2010). + +<|ref|>text<|/ref|><|det|>[[95, 396, 909, 431]]<|/det|> +2. Becker, T. M. et al. Absorption refrigeration cycles with ammonia-ionic liquid working pairs studied by molecular simulation. Ind. Eng. Chem. Res. 57, 5442-5452 (2018). + +<|ref|>text<|/ref|><|det|>[[95, 436, 909, 471]]<|/det|> +3. Peretz, O., Mishra, A. K., Shepherd, R. F. & Gat, A. D. Underactuated fluidic control of a continuous multistable membrane. PNAS 117, 5217-5221 (2020). + +<|ref|>text<|/ref|><|det|>[[95, 476, 909, 512]]<|/det|> +4. Guest, S. D. & Pellegrino, S. Analytical models for bistable cylindrical shells. Proc. R. Soc. A 462, 839-854 (2006). + +<|ref|>text<|/ref|><|det|>[[95, 517, 909, 552]]<|/det|> +5. Lee, J. H., Singer, J. P. & Thomas, E. Micro-nanostructured mechanical metamaterials. Adv. Mater. 24, 4782-4810 (2012). + +<|ref|>text<|/ref|><|det|>[[95, 557, 909, 592]]<|/det|> +6. Correa, D. M. et al. Negative stiffness honeycombs for recoverable shock isolation. Rapid Prototyp. J. 21, 193-200 (2015). + +<|ref|>text<|/ref|><|det|>[[95, 597, 909, 633]]<|/det|> +7. Correa, D. M., Seepersad, C. C. & Haberman, M. Mechanical design of negative stiffness honeycomb materials. Integrating Mater. Manuf. Innov. 4, 10 (2015). + +<|ref|>text<|/ref|><|det|>[[95, 638, 909, 656]]<|/det|> +8. Babaee, S. et al. 3d soft metamaterials with negative poisson's ratio. Rapid Prototyp. J. 25, 5044-5049 (2013). + +<|ref|>text<|/ref|><|det|>[[95, 661, 909, 696]]<|/det|> +9. Arpin, K. A. et al. Multidimensional architectures for functional optical devices. Adv. Mater. 22, 1084-1101 (2010). + +<|ref|>text<|/ref|><|det|>[[95, 701, 909, 737]]<|/det|> +10. Soukoulis, C. M. & Wegener, M. Past achievements and future challenges in the development of three-dimensional photonic metamaterials. Nat. Photonics 5, 523-530 (2011). + +<|ref|>text<|/ref|><|det|>[[95, 742, 909, 777]]<|/det|> +11. Peng, P. et al. Acoustic tunneling through artificial structures: From phononic crystals to acoustic metamaterials. Solid State Commun. 151, 400-403 (2011). + +<|ref|>text<|/ref|><|det|>[[95, 782, 909, 817]]<|/det|> +12. Lu, M. H., Feng, L. & Chen, Y. F. Phononic crystals and acoustic metamaterials. Mater. Today 12, 34-42 (2009). + +<|ref|>text<|/ref|><|det|>[[95, 822, 909, 858]]<|/det|> +13. Deymier, P. Acoustic metamaterials and phononic crystals, in springer series in solid-state sciences. Springer Berlin Heidelberg: Imprint: Springer: Berlin, Heidelberg. (2013). + +<|ref|>text<|/ref|><|det|>[[95, 862, 812, 880]]<|/det|> +14. Zheng, X. et al. Ultralight, ultrastiff mechanical metamaterials. Science 344, 1373-1377 (2014). + +<|ref|>text<|/ref|><|det|>[[95, 885, 909, 920]]<|/det|> +15. Puglisi, G. & Truskinovsky, L. Mechanics of a discrete chain with bi-stable elements. J. Mech. Phys. Solids 48, 1-27 (2000). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 75, 912, 710]]<|/det|> +16. Frenzel, T., Findeisen, C., Kadic, M., Gumbsch, P. & Wegener, M. Tailored buckling microlattices as reusable light-weight shock absorbers. Adv. Mater. 28, 5865-5870 (2016).17. Ha, C. S., Lakes, R. S. & Plesha, M. E. Design, fabrication, and analysis of lattice exhibiting energy absorption via snap-through behavior. Mater. Des. 141, 426-437 (2018).18. Leelavanichkul, S., Chekarev, A., Adams, D. O. & Solzbacher, F. Energy absorption of a helicoidal bistable structure. J. Mech. Mater. Struct. 5, 305-321 (2010).19. Yang, H. & Ma, L. Multi-stable mechanical metamaterials by elastic buckling instability. J. Mater. Sci. 54, 3509-3526 (2019).20. Cohen, T. & Givli, S. Dynamics of a discrete chain of bi-stable elements: a biomimetic shock absorbing mechanism. J. Mech. Phys. Solids 64, 426-439 (2014).21. Katz, S. & Givli, S. Boomerons in a 1-d lattice with only nearest-neighbor interactions. EPL 131, 64002 (2020).22. Shan, S. et al. Multistable architected materials for trapping elastic strain energy. Adv. Mater. 27, 4296-4301 (2015).23. Rafsanjani, A., Akbarzadeh, A. & Pasini, D. Snapping mechanical metamaterials under tension. Adv. Mater. 27, 5931-5935 (2015).24. Pan, F. et al. 3d pixel mechanical metamaterials. Adv. Mater. 31, 1900548 (2019).25. James, K. & Waisman, H. Layout design of a bi-stable cardiovascular stent using topology optimization. Comput. Methods Appl. Mech. Eng. 305, 869-890 (2016).26. Müller, I. & Xu, H. On the pseudo-elastic hysteresis. Acta metallurgica et materialia 39, 263-271 (1991).27. Fedelich, B. & Zanzotto, G. Hysteresis in discrete systems of possibly interacting elements with a double-well energy. J. Nonlinear Sci. 2, 319-342 (1992).28. Allinger, T. L., Epstein, M. & Herzog, W. Stability of muscle fibers on the descending limb of the force-length relation. a theoretical consideration. J. biomechanics 29, 627-633 (1996).29. Müller, I. & Villaggio, P. A model for an elastic-plastic body. Arch. for Ration. Mech. Analysis 65, 25-46 (1977).30. Shaw, J. A. & Kyriakides, S. On the nucleation and propagation of phase transformation fronts in a nit alloy. Acta materialia 45, 683-700 (1997).31. Luongo, A., Casciati, S. & Zulli, D. Perturbation method for the dynamic analysis of a bistable oscillator under slow harmonic excitation. Smart structures systems 18, 183-196 (2016).32. Benichou, I., Faran, E., Shilo, D. & Givli, S. Application of a bi-stable chain model for the analysis of jerky twin boundary motion in nimnga. Appl. Phys. Lett. 102, 011912 (2013). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 131, 223, 178]]<|/det|> +SIMovieX2. mp4SIMovieX1. mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/images_list.json b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..55b208c42a54c14bab4e0a7b0c9563e1fc9b2653 --- /dev/null +++ b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/images_list.json @@ -0,0 +1,122 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Overview of the scSemiProfiler method. a, Initial Configuration: Bulk sequencing is first conducted on the entire cohort, followed by clustering analysis of this data. This analysis identifies representative samples, typically those closest to the cluster centroids. b, Representative Profiling: The identified representatives are then subjected to single-cell sequencing. The data obtained from this sequencing is further processed to determine gene set scores and feature importance weights, enriching the subsequent analysis steps. c, Deep Generative Inference: Utilizing a VAE-GAN-based model, the process integrates comprehensive bulk data from the cohort with the single-cell data derived from the representatives. During the model's 3-stage training, the generator aims to optimize losses \\(L_{G_{Pretrain1}}\\) , \\(L_{G_{Pretrain2}}\\) , and \\(L_{inference}\\) , respectively, whereas the discriminator focuses on minimizing the discriminator loss \\(L_{D}\\) . In \\(L_{D}\\) , \\(G\\) and \\(D\\) are the generator and discriminator respectively. The term \\(D((\\mathbf{x}_1, \\mathbf{s}_1))\\) represents the discriminator's predicted probability that a given input cell is real, under the condition that it is indeed a real cell. Conversely, \\(D(G((\\mathbf{x}_1, \\mathbf{s}_1)))\\) denotes the discriminator's predicted probability that the input cell is real, when in fact it is a cell reconstructed by the generator. d, Representative Selection Decision: Decisions on further representative selection are made, taking into account budget constraints and the effectiveness of the current representatives. An active learning algorithm, which draws on insights from the bulk data and the generative models, is employed to pinpoint additional optimal representatives. These newly selected representatives then undergo further single-cell sequencing (b) and serve as new reference points for the ongoing in silico inference process (c). e, Comprehensive Downstream Analyses: The final panel shows extensive downstream analyses enabled by the semi-profiled single-cell data. This is pivotal in demonstrating the model's capacity to provide deep and wide-ranging insights, showcasing the full potential and applicability of the semi-profiled single-cell data.", + "footnote": [], + "bbox": [ + [ + 145, + 80, + 880, + 600 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Overall comparisons of the semi-profiled and real-profiled COVID-19 dataset. a, UMAP visualization of the real-profiled data. Colors correspond to cell types and are consistent with (g). b, UMAP visualization of the semi-profiled data. c, UMAP visualization of semi-profiled data and real-profiled data together. The color differentiation signifies whether cells originate from the semi-profiled or the real-profiled dataset. Areas of overlap between the two indicate where the semi-profiled data closely resembles the real-profiled data. d, UMAP visualization of the semi-profiled cohort, displaying different colors to distinguish cells produced by a deep generative model (labeled as \"Generated\") from the representative cells obtained through real-profiling (labeled as \"Representatives\"). e, Visualization illustrates the relative activation patterns of the interferon pathway. The comparison of these values between the semi-profiled and real-profiled matrices yields a Pearson correlation coefficient of 0.849 and a p-value of \\(3.63 \\times 10^{-26}\\). f, Graph depicting the normalized error in semi-profiled data with an increasing number of representatives. The terms 'scSemiProfiler' and 'Selection-only' represent our semi-profiling method and a method that only selects representatives using an active learning algorithm, respectively. It is important to note that actual costs may vary based on the sequencing technology and specific cells sequenced. g, Stacked bar plot illustrating the proportions of cell types across various disease conditions. The upper portion represents the real-profiled data, while the lower portion depicts the semi-profiled data. Pearson correlation coefficients comparing cell type proportions between the real-profiled and semi-profiled datasets are provided for different conditions: Healthy (0.987), Asymptomatic (0.970), Mild (0.996), Moderate (0.992), Severe (0.978), and Critical (0.989), indicating a high degree of similarity between the two datasets across these conditions. h-i, Cell type deconvolution benchmarking. h, Figure displaying Pearson correlation coefficients between actual (ground truth) cell type proportions and those estimated by various deconvolution methods. Except for the first two columns, all other columns' results are based on 4 representatives' single-cell data as reference. i, Comparison of Root Mean Square Error (RMSE) across various deconvolution methods.", + "footnote": [], + "bbox": [ + [ + 147, + 80, + 870, + 565 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Comparative analyses of single-cell level downstream analysis tasks using real-profiled and semi-profiled COVID-19 datasets. a, Dot plots elucidating the expression proportion and intensity of discerned cell type signature genes. The top half showcases the real-profiled dataset, while the bottom delineates the semi-profiled version. b, RRHO plot emphasizing the congruence between the CD4 positive and negative markers in both datasets. c, Visualization of the GO term enrichment outcomes rooted in CD4 signature genes from both dataset versions. The plot accentuates the union of the top 10 enriched terms, with the Pearson correlation coefficients of between the bar lengths, which is based on the corresponding p-value and therefore represents the significant level. d, A juxtaposition of cell-cell interaction analyses stemming from real-profiled and semi-profiled cells from moderate COVID-19 patients, underscoring the similarity in interaction types and counts. e, Comparative depiction of pseudotime trajectories for CD4 cells across both datasets, highlighting their striking similarity in reconstructing dynamic cellular processes.", + "footnote": [], + "bbox": [ + [ + 120, + 80, + 846, + 650 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Detailed comparisons between the semi-profiled and real-profiled data in the heterogeneous colorectal cancer dataset. a, UMAP visualization of the real-profiled data, with colors denoting distinct cell types. Colors are consistent with (g). b, UMAP visualization of the semi-profiled data, with colors denoting distinct cell types. c, Joint UMAP visualization highlighting the close resemblance between the semi-profiled and real-profiled data. d, UMAP plot of the semi-profiled dataset, with color-coding distinguishing cells from the actual sequenced representatives and the ones generated through semi-profiling. e, \"Activation of immune response\" gene set relative activation pattern calculated for different tumor tissue types as compared to the \"Normal\" type in the real-profiled and semi-profiled datasets. Entries with fewer than 500 cells are left blank. f, Performance trajectory of the scSemiProfiler on the colorectal cancer dataset, showcasing its superiority over the selection-only approach, with costs computed similarly to Fig. 2d. g, Stacked bar plots comparing cell type compositions between the semi-profiled and real-profiled datasets across different tissues. The Pearson correlation coefficients between the real-profiled and semi-profiled tissues are LymphNode: 0.995, Normal: 0.993, iCMS2: 0.994, iCMS3: 0.988. h-i, Cell type deconvolution benchmarking. h, Pearson correlation coefficients between the actual cell type proportions and those estimated by various deconvolution methodologies. Note: DWLS failed to yield results after a week-long computation. i, Root Mean Square Error (RMSE) comparisons among different deconvolution techniques, highlighting the computational efficiency and accuracy of the scSemiProfiler, especially with an extended set of representatives.", + "footnote": [], + "bbox": [ + [ + 149, + 80, + 875, + 576 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Downstream analysis results comparisons for the colorectal cancer dataset. a, Dot plots visualizing the cell type signature genes. b, RRHO plot visualizing the comparison between semi-profiled and real-profiled markers of epithelial cells. c, Epithelial cell type signature genes GO enrichment analysis results comparison. d, Cell-cell interaction results comparison between the real-profiled tumor tissue cells and semi-profiled tumor tissue cells. e, Pseudotime results comparison using the epithelial cells.", + "footnote": [], + "bbox": [ + [ + 145, + 80, + 880, + 608 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 Comparative analyses between semi-profiled and real-profiled iMGL datasets. a, UMAP visualization of the real-profiled iMGL cohort. Different colors represent different cell types. Colors are consistent with (g). b, UMAP visualization of the real-profiled iMGL cohort. c, Combined UMAP visualization showcasing the consistent cell distribution across both data versions. d, UMAP visualization highlighting the representatives' cells alongside the semi-profiled cells within the semi-profiled dataset. e, Relative pathway activation pattern of the GO term \"activation of immune response\" calculated for cells of different treatments as compared to cell type \"C1: Homeostatic non-proliferative\" in the real-profiled and semi-profiled cohorts. Entries with fewer than 100 cells are left as blank. f, Performance evaluation of the scSemiProfiler on the iMGL dataset, emphasizing its efficiency in error reduction. g, A comparative illustration of cell type proportions under varying experimental conditions, accentuating the similarity in patterns between datasets. The Pearson correlation coefficients between the real-profiled and semi-profiled versions of cell type proportions under different conditions are: iMGL.D0: 0.999, iMGL.D1: 0.871, iMGL.D2: 0.993, iMGL.D3: 0.989, iMGL.D4: 0.992, iMGL.DMSC: 0.998, iMGL.GW30: 0.960, iMGL.GW-300: 0.999, iMGL.T-30: 0.987, iMGL.T-300: 0.999. h, i, Deconvolution performance benchmarking using Pearson correlation and RMSE.", + "footnote": [], + "bbox": [ + [ + 120, + 80, + 850, + 628 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig. 7 Similarities in downstream single-cell analyses between real-profiled and semi-profiled data for the iMGL cohort. a, Dot plots visualizing the nearly identical expression patterns of cell type signature genes across both datasets. b, RRHO plots highlighting the striking similarities between the top 50 positive and negative C3 markers from both real and semi-profiled datasets. c, Overlapping GO enrichment analysis results for the top C3 signature genes, emphasizing consistent analytical outcomes between the datasets. d, PAGA plots illustrating the consistent major cell type links observed in both datasets. e, Pseudotime plots affirming the topographical alignment between the real-profiled and semi-profiled cohort. The p-value of the Mann-Whitney U test is smaller than the smallest float number that the system can represent, which is \\(2.23 \\times 10^{-300}\\) .", + "footnote": [], + "bbox": [ + [ + 120, + 145, + 850, + 632 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8 Active learning demonstrates its prowess in selectively profiling the most informative samples at the single-cell level. The x-axis represents the number of samples selected for single-cell profiling (representatives). The y-axis shows the single-cell in silico inference difficulty of the dataset, which is quantified by the average single-cell difference from each sample to its representative, showcasing the efficiency of representative selection strategies. The marked stars signify the iterations chosen for our methodology, with the generated data underpinning the analyses detailed in previous sections. a, Results from the COVID-19 dataset. Active learning shows significantly better performance, especially in the beginning when a few representatives are selected. b, Observations derived from the colorectal cancer dataset. Active learning continues to show significantly better performance even when more representatives are selected. c, Insights from the iMGL dataset with real bulk measurements. Active learning still manages to outperform passive learning significantly.", + "footnote": [], + "bbox": [ + [ + 147, + 586, + 878, + 723 + ] + ], + "page_idx": 17 + } +] \ No newline at end of file diff --git a/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c.mmd b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c.mmd new file mode 100644 index 0000000000000000000000000000000000000000..fb9fcd06bf4a758e8257db59f77bc4012325c929 --- /dev/null +++ b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c.mmd @@ -0,0 +1,648 @@ + +# scSemiProfiler: Advancing Large-scale Single-cell Studies through Semi-profiling with Deep Generative Models and Active Learning + +Jun Ding jun.ding@mcgill.ca + +McGill University Health Centre https://orcid.org/0000- 0001- 5183- 6885 + +Jingtao Wang McGill University Health Centre + +Gregory Fonseca McGill University Health Centre + +## Article + +Keywords: Deep Generative Learning, Active Learning, Bulk Deconvolution, Single- cell sequencing, Bulk sequencing, Cohort studies, scSemiProfiler + +Posted Date: December 5th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3638900/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on July 16th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 50150- 1. + +<--- Page Split ---> + +# scSemiProfiler: Advancing Large-scale Single-cell Studies through Semi-profiling with Deep Generative Models and Active Learning + +Jingtao Wang \(^{1,2}\) , Gregory Fonseca \(^{1,2,4}\) , Jun Ding \(^{1,2,3,4,5*}\) + +\(^{1}\) Meakins- Christe Laboratories, Research Institute of McGill University Health Centre, 1001 Decarie Blvd, Montreal, H4A 3J1, Quebec, Canada. \(^{2}\) Department of Medicine, Division of Experimental Medicine, McGill University, 1001 Decarie Blvd, Montreal, H4A 3J1, Quebec, Canada. \(^{3}\) School of Computer Science, McGill University, 3480 Rue University, Montreal, H3A 2A7, Quebec, Canada. \(^{4}\) Quantitative Life Sciences, McGill University, 845 Rue Sherbrooke Ouest, Montreal, H3A 0G4, Quebec, Canada. \(^{5}\) Mila- Quebec AI Institute, 6666 Rue Saint- Urbain, Montreal, H2S 3H1, Quebec, Canada. + +\*Corresponding author(s). E- mail(s): jun.ding@mcgill.ca; Contributing authors: jingtao.wang@mail.mcgill.ca; gregory.fonseca@mcgill.ca; + +## Abstract + +Single- cell sequencing is a crucial tool for dissecting the cellular intricacies of complex diseases. Its prohibitive cost, however, hampers its application in expansive biomedical studies. Traditional cellular deconvolution approaches can infer cell type proportions from more affordable bulk sequencing data, yet they fall short in providing the detailed resolution required for single- cell- level analyses. To overcome this challenge, we introduce "scSemiProfiler", an innovative computational framework that marries deep generative model with active learning strategies. This method adeptly infers single- cell profiles across large cohorts by fusing bulk sequencing data with targeted single- cell sequencing from a few carefully chosen representatives. Extensive validation across heterogeneous datasets verifies the precision of our semi- profiling approach, aligning closely with true single- cell profiling data and empowering refined cellular analyses. Originally developed for extensive disease cohorts, "scSemiProfiler" is adaptable for broad applications. It provides a scalable, cost- effective solution for single- cell profiling, facilitating in- depth cellular investigation in various biological domains. + +Keywords: Deep Generative Learning, Active Learning, Bulk Deconvolution, Single- cell sequencing, Bulk sequencing, Cohort studies, scSemiProfiler + +## Introduction + +The advent of single- cell sequencing has dramatically reshaped the landscape of biological research, providing an unparalleled view into the cellular complexities of organisms [1, 2]. This technique has unearthed the subtle distinctions among individual cells, enabling a richer understanding of cellular dynamics [3, 4]. High- resolution data from such analyses are essential for delineating and characterizing the myriad of cellular subpopulations within patient samples, leading to transformative developments in biomarker discovery and personalized therapy strategies [5- 7]. Cohort studies, + +<--- Page Split ---> + +which offer longitudinal insights by observing specific groups over time, are particularly poised to benefit from these advances [8- 10]. However, the substantial financial cost associated with single- cell sequencing—such as the estimated \(\) 6,000\$ required to sequence 20,000 cells (costpercell)—can often be a limiting factor for extensive research endeavors. + +A range of computational strategies, particularly deconvolution methods, are available for dissecting bulk data [11] into distinct cell populations. This approach enables a harmonious balance between affordability and data resolution. Prominent among these are deconvolution techniques such as CIBERSORTx [12] and Bisque [13], which have become particularly popular. These methods estimate the proportions of different cell types within bulk sequencing samples by utilizing signature profiles from single- cell reference datasets. Conventional bulk decomposition methods, while valuable for enhancing cohort study analyses, exhibit limitations in their resolution and accuracy. Capable of dissecting bulk samples into different cell types and ascertaining their gene expression patterns, they offer beneficial insights at the cell- type level. Yet, these methods fall short in delivering true single- cell resolution. Another significant challenge they face is the precise decomposition of cell types and accurate inference of gene expression. Additionally, these methods often overlook the considerable variability that exists within individual cell types. This variability is a critical component for deciphering the complexities of disease dynamics and their response to treatments [14- 18]. Those limitations of conventional bulk decomposition methods may impede the thorough exploration of in- depth single- cell level analyses. These analyses are essential for understanding cellular heterogeneity and dynamics in datasets, including but not limited to cell pseudotime trajectory analysis [19- 22], as well as various advanced machine learning techniques [23- 27]. + +In response to the challenges previously highlighted and with the goal of offering a cost- effective approach for extensive single- cell sequencing, we introduce the single- cell Semi- profiler (scSemiProfiler). This innovative computational tool is crafted to significantly improve the precision and depth of single- cell analysis. It stands out as a more economical and scalable option for single- cell sequencing, thereby facilitating advanced single- cell analysis with greater accessibility. This tool effectively integrates active learning techniques [28] with deep generative neural network algorithms [29], aiming to provide single- cell resolution data at a more affordable price. scSemiProfiler aims to simultaneously achieve two fundamental goals in the semi- profiling process. On one hand, scSemiProfiler's active learning module integrates information from the deep learning model and bulk data, intelligently selecting the most informative samples for actual single- cell sequencing. On the other hand, scSemiProfiler's deep generative model component [30- 33] effectively merges single- cell data from representative samples with the bulk sequencing data of the cohort. This process computationally infers the single- cell data for the remaining non- representative samples. This advanced approach leads to more detailed "deconvolution" of the target bulk data into precise single- cell level measurements. Consequently, with only the budget for bulk sequencing and single- cell sequencing of representatives, scSemiProfiler outputs single- cell data for all samples in the study. To the best of our knowledge, the scSemiProfiler is the first of its kind designed for such intricate single- cell level computational decomposition from bulk sequencing data. + +Through comprehensive evaluations across a variety of datasets, scSemiProfiler has consistently delivered high- quality semi- profiled single- cell data, which not only accurately represents actual single- cell datasets but also mirrors the results of downstream tasks with remarkable fidelity. Therefore, scSemiProfiler has established itself as a pivotal tool in single- cell research. This innovative approach is poised to revolutionize access to single- cell data in large- scale studies, such as disease cohort investigations and beyond. By making large- scale single- cell studies more affordable, scSemiProfiler promises to catalyze the application of single- cell technologies in a wide array of expansive biomedical research. This advancement is set to expand the scope and enhance the depth of biological research globally. + +## Results + +## Method overview + +The scSemiProfiler approach represents a novel method for decomposing broad- scope bulk sequencing data into detailed single- cell cohorts. It achieves this by conducting single- cell profiling on only a select few representative samples and then computationally inferring single- cell data for the remainder. This approach substantially lowers the costs associated with large- scale single- cell studies. As depicted in Fig. 1, our method is designed to deliver cost- effective, semi- profiled single- cell sequencing data, + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Overview of the scSemiProfiler method. a, Initial Configuration: Bulk sequencing is first conducted on the entire cohort, followed by clustering analysis of this data. This analysis identifies representative samples, typically those closest to the cluster centroids. b, Representative Profiling: The identified representatives are then subjected to single-cell sequencing. The data obtained from this sequencing is further processed to determine gene set scores and feature importance weights, enriching the subsequent analysis steps. c, Deep Generative Inference: Utilizing a VAE-GAN-based model, the process integrates comprehensive bulk data from the cohort with the single-cell data derived from the representatives. During the model's 3-stage training, the generator aims to optimize losses \(L_{G_{Pretrain1}}\) , \(L_{G_{Pretrain2}}\) , and \(L_{inference}\) , respectively, whereas the discriminator focuses on minimizing the discriminator loss \(L_{D}\) . In \(L_{D}\) , \(G\) and \(D\) are the generator and discriminator respectively. The term \(D((\mathbf{x}_1, \mathbf{s}_1))\) represents the discriminator's predicted probability that a given input cell is real, under the condition that it is indeed a real cell. Conversely, \(D(G((\mathbf{x}_1, \mathbf{s}_1)))\) denotes the discriminator's predicted probability that the input cell is real, when in fact it is a cell reconstructed by the generator. d, Representative Selection Decision: Decisions on further representative selection are made, taking into account budget constraints and the effectiveness of the current representatives. An active learning algorithm, which draws on insights from the bulk data and the generative models, is employed to pinpoint additional optimal representatives. These newly selected representatives then undergo further single-cell sequencing (b) and serve as new reference points for the ongoing in silico inference process (c). e, Comprehensive Downstream Analyses: The final panel shows extensive downstream analyses enabled by the semi-profiled single-cell data. This is pivotal in demonstrating the model's capacity to provide deep and wide-ranging insights, showcasing the full potential and applicability of the semi-profiled single-cell data.
+ +enabling deep exploration of cellular dynamics in large cohorts. In this context, "semi- profiling" refers to the generation of single- cell data for an entire cohort, achieved either through direct single- cell sequencing of selected representative samples or via in silico inference using a deep generative model. This in silico inference process combines actual single- cell sequencing data from a representative + +<--- Page Split ---> + +sample with bulk sequencing data, encompassing both the target and the representative sample. Thus, a "semi- profiled cohort" includes real single- cell data for representative samples and inferred data for the non- representative ones. This innovative approach facilitates a thorough examination of individual cellular profiles within a larger dataset, seamlessly linking the extensive scope of bulk sequencing with the granularity of single- cell analysis. + +Initially, the semi- profiling pipeline commences with bulk sequencing of each cohort member (Fig. 1a), laying the foundational data layer for all subsequent analyses. Following this foundational step, the methodology employs a clustering analysis to form \(B\) (sample batch size) sample clusters utilizing the extensive data derived from the initial bulk sequencing and selecting a "representative" sample for each cluster (Fig. 1b). Single- cell profiling will be conducted on the selected representative samples in preparation for the following steps. The core of the scSemiProfiler involves an innovative deep generative learning model (Fig. 1c). This model is engineered to intricately meld actual single- cell data profiles with the gathered bulk sequencing data, thereby capturing complex biological patterns and nuances. Specifically, it uses a VAE- GAN [34] architecture initially pretrained on single- cell sequencing data of selected representatives for self- reconstruction. Subsequently, the VAE- GAN is further pretrained with a representative reconstruction bulk loss, aligning pseudobulk estimations from the reconstructed single- cell data with real pseudobulk. Finally, the model undergoes fine- tuning with another target bulk loss tied to the real bulk sequencing data of the target sample, facilitating precise in silico inference of the target's single- cell profile. Once the in silico single- cell inference is finished for all non- representative samples in the cohort, an active learning module can be used for selecting the next batch of potentially most informative representatives for single- cell sequencing to further improve the semi- profiling performance (Fig. 1d). When studying a smaller dataset, or when more cells per sample are required, a smaller batch size, such as 2, may be preferred. However, a batch size of 4 is set as default to maximize the usage of a 10x genomics single- cell toolkit, which can typically capture up to 20000 cells (4 samples if assuming 5,000 cells each). This dynamic, iterative process is continuously augmented with newly acquired single- cell data, ensuring that the most informative samples are selected for real single- cell profiling, leading to more accurate in silico single- cell inference for non- representative samples. This iterative process concludes when the budgetary constraint is met or when a sufficient number of representatives have been chosen to ensure satisfactory semi- profiling performance. + +When any of the stop criteria is met, the semi- profiled single- cell data can be used for a broad spectrum of downstream single- cell analyses (Fig. 1e), such as cell feature visualization, biomarker, and function enrichment analysis, tracking cell type compositions in various tissues/conditions, cell- cell interaction analysis, and pseudotime analysis. Ultimately, scSemiProfiler offers a holistic sequencing perspective, delivering nuanced single- cell insights from bulk sequencing data. The method exhibits acceptable performance in terms of runtime and memory usage (refer to Supplementary Fig. S1), making it suitable for scaling to large- scale datasets. + +## The semi-profiled COVID-19 single-cell cohort exhibits significant similarity to its real counterpart + +To test the performance of scSemiProfiler in generating semi- profiled single- cell data that resonates with the granularity and details of actual single- cell sequencing, we utilized a COVID- 19 cohort single- cell sequencing dataset [35]. After quality controls, this dataset includes 124 samples, including healthy controls and infected patients of different severity levels: asymptomatic, mild, moderate, severe, or critical. Here, we produced pseudobulk data by taking the average of the normalized count single- cell data. We then tested scSemiProfiler's ability to regenerate the single- cell cohort from the pseudo- bulk data and real- profiled single- cell representatives using semi- profiling, as the actual bulk data for those samples in the COVID- 19 cohort is absent. In generating our semi- profiled dataset, 28 representatives were selected in batches of 4 using our active learning algorithm for real single- cell profiling. Deep generative models then inferred the single- cell profiles for the remaining samples based on the representatives' real- profiled single- cell data and the bulk data of all samples. The estimated total cost of these bulk and single- cell sequencing is \(62,640. This is only 33.7\%\) of the estimated price, \(\) 186,000\(, for actually conducting single- cell sequencing for the entire cohort. Additionally, our approach offers the advantage of generating extra bulk data for the cohort, a benefit not provided by single- cell sequencing of the entire cohort. The price of bulk sequencing is estimated based on the cost at McGill Genome Centre in the year 2023. This estimation assumes the use of one NovaSeq 6000 S2 system (capable of sequencing up to 4 billion reads per run) at an approximate cost of\) \ \(7,000\) + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 Overall comparisons of the semi-profiled and real-profiled COVID-19 dataset. a, UMAP visualization of the real-profiled data. Colors correspond to cell types and are consistent with (g). b, UMAP visualization of the semi-profiled data. c, UMAP visualization of semi-profiled data and real-profiled data together. The color differentiation signifies whether cells originate from the semi-profiled or the real-profiled dataset. Areas of overlap between the two indicate where the semi-profiled data closely resembles the real-profiled data. d, UMAP visualization of the semi-profiled cohort, displaying different colors to distinguish cells produced by a deep generative model (labeled as "Generated") from the representative cells obtained through real-profiling (labeled as "Representatives"). e, Visualization illustrates the relative activation patterns of the interferon pathway. The comparison of these values between the semi-profiled and real-profiled matrices yields a Pearson correlation coefficient of 0.849 and a p-value of \(3.63 \times 10^{-26}\). f, Graph depicting the normalized error in semi-profiled data with an increasing number of representatives. The terms 'scSemiProfiler' and 'Selection-only' represent our semi-profiling method and a method that only selects representatives using an active learning algorithm, respectively. It is important to note that actual costs may vary based on the sequencing technology and specific cells sequenced. g, Stacked bar plot illustrating the proportions of cell types across various disease conditions. The upper portion represents the real-profiled data, while the lower portion depicts the semi-profiled data. Pearson correlation coefficients comparing cell type proportions between the real-profiled and semi-profiled datasets are provided for different conditions: Healthy (0.987), Asymptomatic (0.970), Mild (0.996), Moderate (0.992), Severe (0.978), and Critical (0.989), indicating a high degree of similarity between the two datasets across these conditions. h-i, Cell type deconvolution benchmarking. h, Figure displaying Pearson correlation coefficients between actual (ground truth) cell type proportions and those estimated by various deconvolution methods. Except for the first two columns, all other columns' results are based on 4 representatives' single-cell data as reference. i, Comparison of Root Mean Square Error (RMSE) across various deconvolution methods.
+ +plus an additional \(\) 110\(for library preparation per bulk sample. The cost for single - cell sequencing is based on the tool (costpercell). This tool provides a cost estimate for capturing 0.8 billion total reads from 20,000 cells across four samples in one 10x lane, equating to\) \ \(0.3\) per cell. Consequently, the estimated cost for each sample (5,000 cells) is \(\) 1,500\$ . + +<--- Page Split ---> + +Using UMAP [36] visualizations, we show a significant alignment between the semi- profiled and the real- profiled in Fig. 2a and b. We also annotated the semi- profiled COVID- 19 cohort using an unbiased approach. We trained a Multi- layer Perceptron (MLP) classifier using the annotated representatives' cells and used it to predict the cell types of the rest of the non- representative samples' cells generated by the deep learning model. As shown in Fig. 2b, cell clustering remains intact in the semi- profiled cohort, from the distinctive clusters of B cells, plasmoblasts, and platelets to the nuanced similarities of the CD14 and CD16 cells. The semi- profiled dataset illustrates its remarkable fidelity to the real- profiled version. This fidelity extends to capturing even the subtle batch effects, as observed in the twin CD4 clusters, further accentuating scSemiProfiler's robust in silico inference capabilities. The UMAP plots in Fig. 2c weave together directly sequenced samples with the semi- profiled ones, showcasing the tool's finesse. The overlapping data points found in both sequencing techniques resonate with the transformative nature of scSemiProfiler—it harmonizes accuracy and cost- efficiency seamlessly. This prompts an inquiry: Does this alignment owe its credit to the active learning mechanism that identifies the most informative representatives, or does it also hail from the deep generative model's prowess in inferring target samples' cells with finesse? To answer this, Fig.2d uses distinct colors to delineate real- profiled cells from representatives and generated cells for target samples. It shows that our representative selection strategy is able to select representatives such that their cells have a relatively good coverage of the overall cell distribution. Meanwhile, the deep learning model managed to generate cells to complement the rest and make the overall semi- profiled cohort almost identical to the original real- profiled cohort. The effectiveness of the deep learning model in the semi- profiling process is further demonstrated in Supplementary Fig. S2. This figure illustrates the model's capability in reconstructing data from single representative samples and its proficiency in inferring data for individual target samples. + +To test the fidelity of single- cell gene expression semi- profiling, we tested the interferon (IFN) pathway gene set—crucial for the innate immune response against COVID- 19 [37–40]. Through the prism of our semi- profiled dataset, Fig. 2e reveals IFN activation patterns that harmonize with the real- profiled dataset. The uniformity in IFN activation patterns across various key cell types and severity levels, as highlighted by similar heatmaps, confirms the effectiveness of the semi- profiling technique. This uniformity indicates that the critical disease- related pathways were effectively captured and maintained in the semi- profiled data. + +Further, we explored the quantitative metrics of efficacy and cost- effectiveness in semi- profiling, Fig. 2f. Our analysis centered on understanding the relationship between the number of representative samples used for single- cell sequencing and the associated semi- profiling error. While an increase in representative samples intuitively raises the costs of single- cell sequencing, scSemiProfiler effectively leverages the single- cell data of these representatives for accurate inference of target samples. As more representatives are selected, our semi- profiling method effectively reduced the normalized error. We also compared our method to a selection- only method, which uses the same representatives chosen by scSemiProfiler using the active learning algorithm. For each target sample, it performs the single- cell data inference in a naive manner: merely copying the corresponding representative's data. The dashed line shows that our semi- profiling method has a huge lead over the selection- only method. The star symbol denotes the number of representatives selected for the specific semi- profiled cohort that was utilized for subsequent analyses, along with the associated error. The vertical dashed line underscores that by using the same representatives, our method achieves substantially lower errors compared to the selection- only method. Moreover, the horizontal dashed line demonstrates that the selection- only method requires a considerably larger pool of representatives to attain the same level of semi- profiling accuracy as our method. The comparative analysis between scSemiProfiler and the selection- only method highlights the deep learning model's efficacy in reducing costs and minimizing errors. + +In further evaluating the effectiveness of semi- profiling, defined as a single- cell granularity bulk decomposition, we turned to the conventional realm of cell type proportion metrics, as anchored by Fig. 2g. Existing research [35] elucidates that PBMC cell type proportions undergo dynamic shifts with evolving disease conditions. True to this, our real- profiled dataset indicates a pronounced expansion of B cells and CD14 cells under aggravated conditions—a pattern mirrored in the semi- profiled dataset. The Pearson correlation coefficient [41] of cell type composition associated with different disease conditions between the semi- profiled and the real single- cell datasets consistently surpasses 0.9. + +<--- Page Split ---> + +In our comprehensive analysis of cell deconvolution methods, we meticulously compared scSemiProfiler with several leading- edge techniques, including CIBERSORTx [12], Bisque [13], Scaden [42], TAPE [43], and DWLS [44], as depicted in Fig. 2h and i. Each method was tested under identical conditions, using the same bulk data and single- cell reference data. Notably, DWLS was not included in our final results due to its failure to yield results within a week. A key challenge in this analysis was the memory constraints encountered by most methods, elaborated in Supplementary Fig. S1b, which hindered their ability to process the full set of 28 representative single- cell data. To maintain a level playing field, we limited the single- cell reference to an initial batch of 4 representative samples for all methods. Our results unequivocally demonstrate the superior deconvolution performance of scSemiProfiler over all benchmarked methods. Its effectiveness is not only apparent when utilizing a smaller reference set of 4 samples but also becomes increasingly pronounced with a larger set of 28 samples. This distinct superiority of scSemiProfiler is a testament to its remarkable efficiency and versatility. Capable of excelling with both compact and extensive single- cell reference datasets, scSemiProfiler stands out as the most adept tool in cell deconvolution, surpassing its contemporaries in handling diverse data scales with unparalleled precision and reliability. In Supplementary Fig. S3, we also show the high deconvolution accuracy using a side- by- side comparison between each sample's predicted cell type proportion using 28 representatives with the ground truth. + +## The semi-profiled COVID-19 single-cell cohort proves reliable for single-cell downstream analyses + +We have previously demonstrated the capability of scSemiProfiler in accurately generating semi- profiled single- cell data that closely aligns with its real- profiled counterpart. Moving beyond basic cell type proportion predictions, which is the primary focus of other methods, scSemiProfiler excels in predicting gene expression for each cell within a population, thereby more authentically mimicking true single- cell data. This advancement is crucial for more complex downstream single- cell analysis tasks. + +To illustrate the effectiveness of semi- profiled data in standard downstream single- cell analyses, we conducted a series of evaluations. A key task in these analyses is the identification of biomarkers within distinct cell clusters, highlighting genes with elevated expression levels. Utilizing the semi- profiled data generated by scSemiProfiler, we performed various single- cell level downstream analyses. The results from these analyses, as depicted in Fig. 3, demonstrate a remarkable consistency with outcomes derived from real- profiled data. For instance, we identified top cell type signature genes using the real- profiled cohort. When comparing their expression patterns in both real- profiled and semi- profiled datasets, the similarities were striking. The dot plots in Fig. 3a display these patterns, showcasing an almost indiscernible difference between the datasets. The semi- profiling at the single- cell level provides high- resolution expression data for marker genes within each cell population (cell type). Our approach reveals the distribution of marker gene expression across all cells within a specific type. While existing bulk deconvolution methods can offer average gene expression data for cell type markers, they fall short in depicting detailed gene expression distribution and variations among individual cells within the same population. This high level of cell type biomarker concordance underscores the scSemiProfiler's robustness not only in replicating single- cell data but also in ensuring the fidelity of downstream analytical processes. + +We further explored the similarities between biomarkers identified using semi- profiled data and those from real- profiled data. Biomarker discovery is possible with lower- resolution data, such as the average cell type gene expression data provided by current bulk deconvolution tools. However, these methods are less reliable compared to single- cell data. The limitation of average cell type gene expression data is its lack of replicates for each cell type, leading to reduced statistical power. The absence of replicates makes it challenging to estimate variance within a cell type, which is essential for standard differential expression tests. In contrast, our semi- profiled single- cell data supports robust biomarker discovery through rigorous statistical testing, as it includes multiple samples (i.e. cells) for each cell type. We demonstrate the similarity of biomarkers identified using real- profiled and semi- profiled datasets through a rank- rank hypergeometric overlap (RRHO) plot. An RRHO plot [45] visualizes the overlap between two ranked gene lists, highlighting the degree of similarity and the significance of the overlap between them (see the Methods section for more details). Leveraging the RRHO plots, we compared the top 50 positive and top 50 negative gene lists associated with different cell types (Fig. 3b for CD4 cells and Supplementary Fig. S4 for all other cell types. The plots show positive marker and negative marker lists from both datasets are highly similar. A + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Comparative analyses of single-cell level downstream analysis tasks using real-profiled and semi-profiled COVID-19 datasets. a, Dot plots elucidating the expression proportion and intensity of discerned cell type signature genes. The top half showcases the real-profiled dataset, while the bottom delineates the semi-profiled version. b, RRHO plot emphasizing the congruence between the CD4 positive and negative markers in both datasets. c, Visualization of the GO term enrichment outcomes rooted in CD4 signature genes from both dataset versions. The plot accentuates the union of the top 10 enriched terms, with the Pearson correlation coefficients of between the bar lengths, which is based on the corresponding p-value and therefore represents the significant level. d, A juxtaposition of cell-cell interaction analyses stemming from real-profiled and semi-profiled cells from moderate COVID-19 patients, underscoring the similarity in interaction types and counts. e, Comparative depiction of pseudotime trajectories for CD4 cells across both datasets, highlighting their striking similarity in reconstructing dynamic cellular processes.
+ +marked dissimilarity was evident between the positive and negative marker lists, which is intuitively anticipated. By definition, positive markers are genes that are higher expressed in the corresponding cell types, and negative markers are the opposite - lower expressed. Therefore, they should have no overlap. The compelling concordance demonstrated in Fig. 3a and b bolsters our claim that the semi- profiled data can viably supplant real- profiled data for the pivotal task of biomarker discovery using scSemiProfiler. + +Next, we used the biomarkers derived from the semi- profiled dataset and those from the real- profiled dataset for gene functional enrichment analysis, assessing whether the two versions of the + +<--- Page Split ---> + +analysis yield consistent results. Fig. 3c compares the Gene Ontology (GO) [46, 47] enrichment [48, 49] outcomes derived from real- profiled and semi- profiled datasets. The top 100 signature genes from both datasets are used for the enrichment analysis. We observed an overlap of 95 genes between the two lists, yielding a highly significant hypergeometric test p- value of \(4.00 \times 10^{- 200}\) (the population size of the hypergeometric test is the number of highly variable genes used for this dataset, 6030). A comparison of the top 10 overlapping terms from both versions reveals nearly identical significance (Pearson correlation coefficient of 0.998 with a p- value of \(4.13 \times 10^{- 12}\) for comparing the significant levels). The results for other cell types are in Supplementary Fig. S5. Reactome pathway [50] enrichment analysis results are in Supplementary Fig. S6. This further corroborates the reliability of semi- profiled data in downstream analyses. + +Progressing to yet another pivotal single- cell level downstream analysis task, we evaluated the congruence in cell- cell interaction analyses derived from real- profiled and semi- profiled datasets. Given the paramount role of cell- cell interactions in orchestrating a myriad of multicellular processes, their analysis often unveils pivotal biological insights [51, 52]. Fig.3d juxtaposes cell- cell interaction analyses rooted in real- profiled and semi- profiled cells from moderate COVID- 19 patients (see results for other severity levels in Supplementary Fig. S7). The evident concordance in types and counts of interactions in both renditions reinforces the reliability of our semi- profiled data ( \(R = 0.905\) , \(P = 8.46 \times 10^{- 122}\) ). We also show a comparison of partition- based graph abstraction (PAGA) plots generated using real- profiled cohort and semi- profiled cohort in Supplementary Fig. S8, which demonstrates that the semi- profiled data can accurately capture the cellular trajectories and relationships between cell types. Given that such analyses intrinsically require single- cell data, scSemiProfiler emerges as the sole contender capable of producing data apt for this task from bulk sources. + +Delving further into the capacity of semi- profiled data for other downstream single- cell level analysis tasks, we turned our attention to pseudotime analysis [19, 20]. Pseudotime is a pivotal tool in reconstructing dynamic cellular processes, ranging from differentiation pathways to developmental timelines or disease trajectories. As depicted in Fig. 3e, the pseudotime trajectories derived from real- profiled and semi- profiled CD4 cells are strikingly similar (Consistent results for the pseudotime analysis of other cell types can be found in Supplementary Fig. S9). Such compelling evidence underscores that the semi- profiled data retains its reliability even for intricate biological explorations like cell trajectory and differentiation analyses. + +## Semi-profiling maintains accuracy on a heterogeneous colorectal cancer dataset + +We further tested the effectiveness of scSemiProfiler by validating it against a notably heterogeneous colorectal cancer dataset [53], which encompassed 112 single- cell sequencing samples that passed the quality control. This collection comprised 19 normal tissues, 86 tumor tissues (including colorectal cancer subtype iCMS2 and iCMS3), and 7 lymph node tissues. Considering the inherent diversity of this dataset, achieving accurate semi- profiling could ostensibly test the limits of scSemiProfiler. Again, this study does not include paired bulk sequencing data. Therefore, we have utilized the pseudo- bulk data derived from single- cell analysis as a surrogate for actual bulk sequencing. Nevertheless, following a consistent data processing and semi- profiling protocol, and selecting 36 representatives in batches of 4, the semi- profiled dataset mirrored its real- profiled counterpart. This congruence manifested not only in visual similarity but also in cell type distributions and subsequent analyses outcomes. Using the same estimation method as applied to the COVID- 19 cohort, the total cost for both bulk and single- cell sequencing to obtain this highly similar semi- profiled single- cell cohort is approximately \$73,320. This price also includes the cost of bulk data for the cohort and represents only 43.6% of the \$168,000 estimated for conducting single- cell sequencing on the entire cohort. + +Figure 4a and b graphically highlight the remarkable similarity between the semi- profiled and real- profiled data using UMAP visualizations. These visualizations, color- coded according to cell types, show a substantial alignment between the datasets. The semi- profiled data mirrors the real- profiled data in terms of the location and shape of each cell type cluster. Notably, both datasets effectively segregate cell types such as plasma B, enteric glial, Mast, and epithelial. Additionally, a nuanced connection between fibroblast and endothelial cells is evident in both versions. Immuno- centric cells like McDC, T_NK, and B cells are also accurately positioned in close proximity in both datasets, underscoring the precision of the semi- profiled data. This similarity is further emphasized + +<--- Page Split ---> + +in Figure 4c, which showcases the significant overlap between the two datasets. Moreover, Figure 4d employs distinct color schemes to differentiate between cells generated by the deep generative learning model and those from real- profiled representative data. The cells from the representatives cover a substantial portion, indicating their well- chosen representative selection. However, numerous cells that fall outside the representatives' distribution are accurately generated by the deep learning model, highlighting the critical role of both active learning and the deep generative model in achieving effective semi- profiling. The accuracy of the deep learning model's generation can be further shown in Supplementary Fig. S10, where we present the model's single- cell inference for individual samples. + +To further justify the semi- profiled gene expression values are accurate and can be used for biological analysis, we compute the gene set activation pattern of the GO term "activation of immune response" (GO:0002253) for the two tumor tissue types the same way as we did for the COVID- 19 cohort (Fig. 4e). We chose this term because the immune response plays a significant role in the body's defense against cancer, and its activation or suppression can influence cancer progression and patient outcomes [54, 55]. The gene set activation scores are calculated and then adjusted by subtracting the score of the "Normal" tissue. The activation pattern in the real- profiled and semi- profiled datasets are highly similar, leading to a high Pearson correlation coefficient of 0.919 between them. + +We also quantitatively examined the overall performance of scSemiProfiler as different numbers of representatives are selected. As shown in Fig. 4f, while our approach trumps the selection- only method, the gap narrows in comparison to results on the COVID- 19 dataset—owing largely to the colorectal cancer dataset's inherent heterogeneity. Despite this, the deep generative model's efficacy remains conspicuous, ensuring cost- effective error reduction. Also, if we aim to achieve an error as low as the previous COVID- 19 cohort, which leads to almost identical analysis results as the real data, only half of the samples need to be selected as representatives. This still reduces the cost significantly. + +Diving deeper into the cell type proportions within the colorectal cancer cohort, one discerns variations across different tissue types—"Lymph Node", "Normal", "iCMS2", and "iCMS3". Fig. 4g illustrates these differences. For example, "Lymph Node" contains an expanded population of B cells compared with other tissue types, "Normal" is enriched with PlasmaB cells, and the two tumor subtypes have a pronounced epithelial presence. Remarkably, the semi- profiled dataset captures these nuances with precision, underlining its capability to replicate intricate analyses with fidelity. + +Lastly, the benchmarking of deconvolution results presented in Fig. 4h and i positions scSemiProfiler at the forefront, significantly outperforming existing methods such as Bisque, TAPE, and Scaden. While its performance with four representatives is on par with CIBERSORTx, scSemiProfiler excels in computational memory efficiency. Further extending the representatives to 36 dramatically boosts the deconvolution accuracy. In contrast, existing methods such as CIBERSORT falter when tasked with handling a large reference set like 36 representatives, mainly due to their computational inefficiencies. This distinction underscores the scSemiProfiler's distinct advantage unshared by its peers. To provide a more detailed perspective on our deconvolution outcomes, Supplementary Fig. S11 showcases a comparative analysis. It displays our predicted cell type proportions for each individual sample alongside the ground truth, enabling a side- by- side evaluation + +## scSemiProfiler Ensures consistent downstream analyses between semi-profiled and real single-cell data in heterogeneous colorectal cancer cohorts + +In the context of a heterogeneous dataset like this colorectal cancer one, the semi- profiled dataset stands robust, offering downstream analysis results that mirror the real- profiled data. This close resemblance is consolidated in Fig. 5. + +A notable observation is the accuracy in analyzing biomarker expression pattern and their intracluster variation using semi- profiled data. Fig. 5a showcases dot plots for top cell type signature genes derived from the real dataset. These plots reflect an identical pattern in both the real- profiled and semi- profiled datasets. Further affirmation comes from the strong Pearson correlation coefficient between the colors (0.994) and sizes (0.996) of the dots. Notably, these correlation coefficients even surpass those observed in the more homogeneous COVID- 19 dataset. The semi- profiled dataset also reproduces biomarker discovery results, establishing its credibility as a suitable stand- in for the real- profiled data in such analyses. This consistency is exemplified by genes like KRT18 and KRT8, exclusive to epithelial cells, corroborated by existing literature [56, 57]. Another illustration is the unique expression of CPA3 and TPSB2 in Mast across both datasets. Beyond these top cell type signature genes, a granular examination of epithelial cells—encompassing the top 50 positive and + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 Detailed comparisons between the semi-profiled and real-profiled data in the heterogeneous colorectal cancer dataset. a, UMAP visualization of the real-profiled data, with colors denoting distinct cell types. Colors are consistent with (g). b, UMAP visualization of the semi-profiled data, with colors denoting distinct cell types. c, Joint UMAP visualization highlighting the close resemblance between the semi-profiled and real-profiled data. d, UMAP plot of the semi-profiled dataset, with color-coding distinguishing cells from the actual sequenced representatives and the ones generated through semi-profiling. e, "Activation of immune response" gene set relative activation pattern calculated for different tumor tissue types as compared to the "Normal" type in the real-profiled and semi-profiled datasets. Entries with fewer than 500 cells are left blank. f, Performance trajectory of the scSemiProfiler on the colorectal cancer dataset, showcasing its superiority over the selection-only approach, with costs computed similarly to Fig. 2d. g, Stacked bar plots comparing cell type compositions between the semi-profiled and real-profiled datasets across different tissues. The Pearson correlation coefficients between the real-profiled and semi-profiled tissues are LymphNode: 0.995, Normal: 0.993, iCMS2: 0.994, iCMS3: 0.988. h-i, Cell type deconvolution benchmarking. h, Pearson correlation coefficients between the actual cell type proportions and those estimated by various deconvolution methodologies. Note: DWLS failed to yield results after a week-long computation. i, Root Mean Square Error (RMSE) comparisons among different deconvolution techniques, highlighting the computational efficiency and accuracy of the scSemiProfiler, especially with an extended set of representatives.
+ +negative markers—reinforces the congruence. All markers were identified using data at single- cell resolution through thorough statistical testing, a process unachievable with decomposed cell type- level data from standard bulk decomposition methods. As depicted in Fig. 5b, the preponderance of highly significant entries, with many p- values lower than \(10^{- 50}\) , strongly indicates a high degree of similarity in marker lists between the two datasets. RRHO plots for additional cell types can be found in Supplementary Fig. S12. + +Diving deeper into the gene functional enrichment analysis, Fig. 5c offers further validation. Analyzing the top 100 signature genes across both datasets reveals a staggering 96 common markers, + +<--- Page Split ---> + +yielding a hypergeometric test p- value of \(8.88 \times 10^{- 188}\) (population size = 4053). GO terms from both datasets, along with their respective p- values, showcase pronounced similarity. Although the heterogeneity of the dataset leads to a relatively lower Pearson correlation coefficient of 0.593, the overall patterns in the two plots remain statistically similar ( \(P = 0.042\) ), leading to the same scientific conclusions. Moreover, when considering the union of the top 10 Gene Ontology (GO) terms from both the semi- profiled and actual datasets (comprising a total of 12 terms), there is a significant similarity in terms of enrichment p- values. Notably, the two versions of the top 10 GO terms have 9 overlap terms. More comprehensive GO term and pathway enrichment analysis results for other cell types' signature genes are also consistent for the real- profiled and semi- profiled versions (Supplementary Figs. S13 and S14). + +Despite the increased heterogeneity of the dataset, the analysis of cell- cell interactions with the colorectal cancer semi- profiled cohort remains promising. As illustrated in Fig. 5d, the cell- cell interaction analysis, when executed on the real- profiled tumor tissue cells and the semi- profiled counterpart, reveals substantial consistency. Navigating the intricate interaction patterns characteristic of tumor tissues, the semi- profiled data astonishingly replicates the intricate layout with a robust Pearson correlation coefficient of 0.933. Both versions highlight enteric glial and fibroblast as the primary senders, while neutrophils emerge as the predominant receivers. Significantly, the most intense interactions identified across both sets involve the enteric glial with itself, the enteric glial with neutrophils, and the fibroblast with neutrophils. The cell- cell interaction results for other tissues also exhibit a high degree of similarity between the real- profiled and semi- profiled versions, as shown in Supplementary Fig. S15. Additionally, the strikingly similar PAGA plots presented in Supplementary Fig. S16 further demonstrate the utility of semi- profiled data in studying cellular trajectories and relationships between different cell types. + +Shifting the focus to pseudotime analysis, we evaluated epithelial cells across all tissues as an example demonstration (Fig. 5e). Given the presence of cells from tumor tissues, this might introduce elevated heterogeneity within the cell type. Yet, the consistency between pseudotime analysis results from both versions is significant. Both versions discern lower pseudotime values concentrated at the base of the cluster, culminating in larger values towards the upper regions, with the pinnacle being the top- right quadrant. The statistical significance of the similarity is further validated by a Mann- Whitney U test [58, 59], yielding a compelling p- value of \(2.84 \times 10^{- 197}\) . The pseudotime analyses for other cell types also demonstrate a high degree of similarity, as evidenced in Supplementary Fig. S17. Such a finding underscores the capability of the semi- profiled data to adeptly capture intra- cluster nuances in detailed analyses. + +## Semi-profiling with real bulk measurements yields a dataset nearly identical to the original single-cell data + +To further illustrate the adaptability of scSemiProfiler in real- world applications, we directed our analysis towards the iMGL dataset [60], which uniquely profiles both single- cell and real bulk RNA- seq measurements for the human inducible pluripotent stem cell (iPSC)- derived microglia- like (iMGL) cells, differing from the pseudobulk datasets previously used. There are 25 samples having both single- cell data and bulk RNA sequencing data. Samples are of different conditions (grown in cell culture for 0- 4 days and under various treatments). The availability of such datasets, which include both single- cell and bulk sequencing data on a large scale, remains very limited, partially due to the unnecessary of doing both sequencing for the same large- scale cohort and its prohibitive cost. This deviation from pseudobulk data is challenging. Pseudobulk, created by averaging out the noisy single- cell data [61], often fails to encapsulate the subtle features of actual bulk RNA- seq measurements. These real bulk measurements are inherently less noisy and often exhibit systematic differences when compared to single- cell RNA- seq data. + +To navigate this complexity, we devised a method to infer pseudobulk directly from real bulk data (refer to the Methods section for comprehensive details). This approach enables us to more effectively utilize our deep learning model in the pseudobulk data space, which often aligns more closely with single- cell data. Through this method, despite the intricate challenges of the iMGL dataset, the results in Fig. 6 illustrate that our semi- profiled data parallels the real- profiled data quite closely. By selecting only eight representative samples out of a total of 25 samples, we could notably reduce the overall cost without compromising on accuracy. For this smaller dataset, we employed active learning to select 8 representative samples in batches of 2. Using the same estimation method as in the other two studies, the total cost for acquiring both bulk and semi- profiled single- cell data through + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 Downstream analysis results comparisons for the colorectal cancer dataset. a, Dot plots visualizing the cell type signature genes. b, RRHO plot visualizing the comparison between semi-profiled and real-profiled markers of epithelial cells. c, Epithelial cell type signature genes GO enrichment analysis results comparison. d, Cell-cell interaction results comparison between the real-profiled tumor tissue cells and semi-profiled tumor tissue cells. e, Pseudotime results comparison using the epithelial cells.
+ +our method is approximately \(21,750. This amount is just 58\%\) of the estimated \(37,500 required for conducting single- cell sequencing across the entire cohort. + +The UMAP visualization, as presented in Fig. 6a and b, solidifies our findings. Here, the two versions — semi- profiled and real- profiled — show remarkable consistency. The cell distributions of various iMGL subtypes (as delineated by the dataset provider) in the UMAP follow nearly identical patterns. The detailed observation showcases that clusters like “C2: Activated, immediate- early”, “C4: Activated, non- immediate- early”, and “C5: Activated, immediate- early” are interconnected and primarily located on the right- hand side. Likewise, “C3: Homeostatic, proliferative” finds its position at the upper left, with “C1: Homeostatic, non- proliferative” and “C6: Freshly thawed” lying at the bottom left. This consistency transcends to Fig. 6c, emphasizing a consistent cell distribution across the two versions. Furthermore, Fig. 6d underlines the precision of scSemiProfiler, where a majority of cells in the semi- profiled version were accurately generated. The efficient coverage of representative cells in this figure also highlights that our active learning strategy remains robust + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6 Comparative analyses between semi-profiled and real-profiled iMGL datasets. a, UMAP visualization of the real-profiled iMGL cohort. Different colors represent different cell types. Colors are consistent with (g). b, UMAP visualization of the real-profiled iMGL cohort. c, Combined UMAP visualization showcasing the consistent cell distribution across both data versions. d, UMAP visualization highlighting the representatives' cells alongside the semi-profiled cells within the semi-profiled dataset. e, Relative pathway activation pattern of the GO term "activation of immune response" calculated for cells of different treatments as compared to cell type "C1: Homeostatic non-proliferative" in the real-profiled and semi-profiled cohorts. Entries with fewer than 100 cells are left as blank. f, Performance evaluation of the scSemiProfiler on the iMGL dataset, emphasizing its efficiency in error reduction. g, A comparative illustration of cell type proportions under varying experimental conditions, accentuating the similarity in patterns between datasets. The Pearson correlation coefficients between the real-profiled and semi-profiled versions of cell type proportions under different conditions are: iMGL.D0: 0.999, iMGL.D1: 0.871, iMGL.D2: 0.993, iMGL.D3: 0.989, iMGL.D4: 0.992, iMGL.DMSC: 0.998, iMGL.GW30: 0.960, iMGL.GW-300: 0.999, iMGL.T-30: 0.987, iMGL.T-300: 0.999. h, i, Deconvolution performance benchmarking using Pearson correlation and RMSE.
+ +even when navigating the challenges of real bulk measurements. The supplementary Fig. S18 further illustrates the precise single- cell inference achieved by our deep learning model on this dataset. + +Fig. 6e provides an in- depth look at the accuracy of semi- profiled gene expression values. Since microglia are the resident immune cells in the brain, we checked the activation pattern of GO term "activation of immune response" in each cell type under each treatment. The semi- profiled dataset presents a highly similar activation pattern as the real- profiled dataset. + +<--- Page Split ---> + +Fig. 6f offers further evidence of the scSemiProfiler's effectiveness. Here, the performance curve of the semi- profiled approach significantly undercuts the selection- only one, demonstrating its capability to limit semi- profiling errors while optimizing on costs. + +An intriguing observation from the iMGL dataset is the cell type proportion's dynamic shifts under various experimental conditions (Fig. 6g). The transitions from iMGL_D0 to iMGL_D4, for instance, reveal a progressive increase in the proportions of C1 and C4 cells. Contrastingly, "C2: Activated, immediate- early" cells peak at iMGL_D1 and then decrease steadily. Although the intricate effects of drugs iMGL_GW and iMGL_T need further investigation, preliminary data suggests that elevated doses result in a surge of "C1: Homeostatic, non- proliferative" cells. Impressively, these intricate variations are mirrored in the semi- profiled dataset. + +A juxtaposition of our deconvolution method against others on the iMGL dataset offers illuminating insights. The performance of CIBERSORTx, which was one of the best in other datasets, dramatically decreases, potentially due to the inherent challenges posed by real bulk data and the nuanced similarities among cell types. Other methods, such as Bisque, TAPE, and Scaden, are more robust to those challenges, showing decent deconvolution performance. Despite these challenges, scSemiProfiler showcases resilience and consistently outperforms all its peers except Bisque, a fact further corroborated by the Wilcoxon test [62] (see p- values in Fig. 6h and i). The marginal difference between our scSemiProfiler and the selection- only method is a testament to both approaches nearing optimal performance in this specific context. The supplementary Fig. S19 further presents our accurate deconvolution for individual samples. + +## Semi-profiling using real bulk measurement also leads to reliable downstream results + +In this more realistic setting, where scSemiProfiler is tasked with semi- profiling using real bulk data for downstream analyses, the semi- profiled data consistently mirrored results from the real- profiled version. This is particularly remarkable given the unique challenges presented by the real bulk data. We present these downstream analysis results in Fig. 7. + +The markers identified using single- cell resolution data in the real- profiled dataset were almost identical in expression patterns to those in the semi- profiled dataset. Fig. 7a visually reinforces this, showing "C1: Homeostatic, non- proliferative" and "C10: Homeostatic, non- proliferative" with virtually indistinguishable expression patterns across both data types. Additionally, unique expressions in "C11: Myeloid, progenitors", such as GAPA2 and HPGDS, were consistently observed. The overarching similarities were further quantified with impressive Pearson correlation coefficients for both dot sizes and colors, clocking in at 0.980 and 0.989, respectively. + +Further validating the congruency of our method, Fig.7b presented the RRHO plot of the top 50 positive and negative C3 markers from both single- cell datasets (see RRHO plots for other cell types in Supplementary Fig. S20). The degree of similarity is substantial, with most of the markers showcasing p- values less than \(10^{- 50}\) . Such findings strongly suggest that the scSemiProfiler is adept at producing reliable data for biomarker discovery. + +Proceeding to more in- depth downstream analysis using the top 100 signature genes for GO enrichment, we observed an overlap of 90 genes between the semi- profiled and the real single- cell datasets (Fig. 7c). The hypergeometric test revealed a significant p- value of \(1.81 \times 10^{- 183}\) . The enriched terms identified from both the semi- profiled and real datasets matched closely. The Pearson correlation coefficient between the two versions' significance is 0.995, underscoring the consistency in their analytical outcomes. Extended GO and Reactome enrichment analysis in Supplementary Figs. S21 and S22 further confirm the accuracy of finding signature genes using semi- profiled data. + +In the case of cell- cell interactions, both real and semi- profiled single- cell data did not capture any significant interactions, probably due to the similarity between cell clusters in the iMGL data. Instead, the partition- based graph abstraction (PAGA) analysis [63] showcased in Fig.7d highlighted that major cell type links were consistent across datasets. Further cementing this was the strong Pearson correlation coefficient of 0.865 between the adjacency matrices of the two networks. + +Pseudotime analysis in Fig.7e also affirmed the alignment between the two datasets. The topographical pseudotime alignment between them is almost congruent, to the extent that the Mann- Whitney U test implemented in SciPy returned a p- value smaller than the smallest float number our computer can represent, which is \(2.23 \times 10^{- 308}\) . + +<--- Page Split ---> + +In conclusion, our findings robustly demonstrate that scSemiProfiler seamlessly adapts to real-world scenarios employing real bulk data. The downstream analytical outcomes derived from the semi- profiled data are significantly consistent with those based on real data. + +![](images/Figure_7.jpg) + +
Fig. 7 Similarities in downstream single-cell analyses between real-profiled and semi-profiled data for the iMGL cohort. a, Dot plots visualizing the nearly identical expression patterns of cell type signature genes across both datasets. b, RRHO plots highlighting the striking similarities between the top 50 positive and negative C3 markers from both real and semi-profiled datasets. c, Overlapping GO enrichment analysis results for the top C3 signature genes, emphasizing consistent analytical outcomes between the datasets. d, PAGA plots illustrating the consistent major cell type links observed in both datasets. e, Pseudotime plots affirming the topographical alignment between the real-profiled and semi-profiled cohort. The p-value of the Mann-Whitney U test is smaller than the smallest float number that the system can represent, which is \(2.23 \times 10^{-300}\) .
+ +## Active learning demonstrates its prowess in selecting the most informative samples for enhanced single-cell profiling + +The crux of scSemiProfiler's strategy revolves around judiciously selecting representative samples. The rationale is straightforward: the more informative the chosen representatives, the better the semi- profiling performance. This not only enhances the fidelity of the generated profiles but also provides a cost- effective approach by minimizing the number of necessary representatives. + +Initially, our methodology bore similarities to uncertainty sampling [28, 64, 65], a heuristic active learning technique. Here, the intuition is to query samples with the most uncertainty, thereby maximizing the incremental information acquired. In our algorithm, we employed bulk data to pick out + +<--- Page Split ---> + +batches of samples that exhibited the most variance from their designated representatives. Since this method does not use information from the base learner (the deep generative model), it is still a passive learning algorithm. + +To improve the representative selection, we then turn the algorithm into an active learning algorithm by incorporating the information from the deep generative models. The algorithm also utilizes the clustering information in the cohort and is thus a type II active learning algorithm [66]. Combining these two ideas, the algorithm aims to reduce the total heterogeneity of each sample cluster, ensuring that each target sample has a similar representative, thus optimizing semi- profiling performance. + +As depicted in Fig. 8, we juxtaposed our advanced active learning algorithm against the rudimentary passive learning approach. Each panel, from Fig. 8a to Fig. 8c, encapsulates the comparative analyses derived from distinct datasets. While the x- axis maps the representatives earmarked for single- cell sequencing, the y- axis portrays the single- cell in silico inference difficulty, which is quantified using the average single- cell- level difference (see Equation 16 in the Methods section) between target samples and their representatives. This premise holds that as the dissimilarity between the target and the representatives increases, the complexity of the in silico inference task also rises. This metric is crucial as it highlights the challenges encountered by the deep generative learning model in semi- profiling, thereby illuminating the effectiveness of the strategies used for selecting representatives. + +The empirical evidence is resounding. Across all datasets, the active learning algorithm showcased its mettle by consistently pinpointing representatives that considerably reduced the total distance to other samples. This underscores the algorithm's capability to foster superior representative selection for semi- profiling. In Fig. 8a, when applying our method to the COVID- 19 dataset, active learning shows better performance than passive learning, especially in the beginning iterations of the semi- profiling. Active learning reduces the inference difficulty when the same number of representatives are selected. Also, consider the marked point where 28 representatives are selected, which is consistent with representatives we selected for our analyses in previous sections. To reach the same level of inference difficulty, passive learning needs to select 4 more representatives, i.e. the cost of 4 samples' single- cell sequencing experiment is saved. For the colorectal cancer dataset (Fig. 8b), active learning continues to perform better than passive learning, and at the point we selected, the cost for more than two batches (8) of representatives can be saved using active learning. Fig. 8c shows the results for the iMGL dataset, in which active learning is also significantly better than passive learning. In all experiments, active learning consistently outperforms passive learning in selecting representatives. With the same budget, active learning achieves lower inference difficulty. Furthermore, to reach a comparable level of inference difficulty, active learning requires a lower cost. + +![](images/Figure_8.jpg) + +
Fig. 8 Active learning demonstrates its prowess in selectively profiling the most informative samples at the single-cell level. The x-axis represents the number of samples selected for single-cell profiling (representatives). The y-axis shows the single-cell in silico inference difficulty of the dataset, which is quantified by the average single-cell difference from each sample to its representative, showcasing the efficiency of representative selection strategies. The marked stars signify the iterations chosen for our methodology, with the generated data underpinning the analyses detailed in previous sections. a, Results from the COVID-19 dataset. Active learning shows significantly better performance, especially in the beginning when a few representatives are selected. b, Observations derived from the colorectal cancer dataset. Active learning continues to show significantly better performance even when more representatives are selected. c, Insights from the iMGL dataset with real bulk measurements. Active learning still manages to outperform passive learning significantly.
+ +<--- Page Split ---> + +## Discussion + +In the present work, we introduced scSemiProfiler, an innovative computational framework designed to produce affordable single- cell measurements for a given large cohort. The allure of scSemiProfiler stems from its unparalleled capacity to mirror the outcomes of conventional single- cell sequencing, yet with remarkable cost- effectiveness. Marrying expansive sequencing with pinpoint profiling and optimizing data utilization, it charts a robust, intricate, and economical path for large- scale single- cell exploration. This tool is not merely a means of generating data but ensures the derived single- cell information remains dependable for an array of downstream analyses. The pipeline ingests bulk data, leverages an active learning module to judiciously select representatives for authentic single- cell sequencing, and employs a deep generative learning model, VAE- GAN, to infer the single- cell data of the remainder of the cohort. Through this approach, every sample in the cohort has access to single- cell data, either semi- profiled or real- profiled. We subjected scSemiProfiler to rigorous evaluation using three distinct cohorts, simulating a scenario devoid of single- cell data and then juxtaposing the output of our tool with the authentic single- cell datasets. The striking resemblance between semi- profiled and real- profiled data in facets like UMAP visualization and cell type proportions ascertains the robustness of scSemiProfiler. Furthermore, both datasets align almost perfectly in subsequent analyses, encompassing biomarkers, enrichment analyses, cell- cell interactions, and pseudotime trajectories, offering a cost- effective means of integrating single- cell data in cohort studies without compromising on analytical precision. + +The magnitude and uniqueness of our study's contributions can be primarily attributed to two game- changing innovations that stand apart in their novelty and transformative potential. Precision- driven Computational Modeling: The crowning achievement of our study lies in the design and implementation of a groundbreaking computational model for single- cell level bulk decomposition. While traditional methodologies offer a mere cell type level decomposition of bulk data, scSemiProfiler shatters these bounds with an unprecedented capability. What we've pioneered isn't just a method to interpret bulk data, but a mechanism to deconvolute it into authentic single- cell data. This offers a myriad of sophisticated and meaningful downstream analytic possibilities, previously unattainable with conventional techniques. To the best of our knowledge, scSemiProfiler is the very first in its class to provide a true single- cell level decomposition. Furthermore, when placed under rigorous scrutiny, the semi- profiled data generated by our model consistently showcases remarkable similarity to real- profiled single- cell data, underscoring its reliability and potential to reshape the paradigm of single- cell analytics, particularly for large- scale cohort studies in which real single- cell sequencing cost would be prohibitive. Intelligent Active Learning Mechanism: Our second salient innovation hinges on the power of active learning. Instead of passively selecting representatives, our model embarks on an intelligent, iterative journey. By constantly evaluating the data and learning from previous rounds of selection, this module discerns which representatives would offer the most informative insights for semi- profiling. This isn't a mere addition; it's a reinvention of how representative selection operates. The synchronization of insights from the deep learning model with this active learning module ensures that the selection process is not just data- driven, but also insight- driven. The implications of this are twofold: firstly, it enhances the quality and relevance of the selected representatives, ensuring that they truly resonate with the cohort's cellular composition. Secondly, it offers an economic advantage. By intelligently choosing representatives, the financial overheads associated with single- cell experiments can be minimized, ensuring maximum return on investment both in terms of data quality and budgetary constraints. Together, these innovations not only elevate the efficacy of scSemiProfiler but also pioneer a new direction in how single- cell data can be derived, analyzed, and utilized, holding profound implications for future biomedical research endeavors. + +Future work on scSemiProfiler is focused on expanding its capabilities from single- cell RNA sequencing (scRNA- seq) to other biological modalities, such as proteomics. The adaptability of the tool's current methodology will require adjustments in algorithms to suit the unique data characteristics of each new modality, along with comprehensive performance evaluations benchmarked against established standards. A pivotal element of this development is the utilization of cross- modal information exchange, which aims to improve the accuracy of semi- profiling. This approach could significantly reduce the costs associated with single- cell multi- omics studies beyond the scope of the current study, by minimizing the need for extensive representative samples from each modality and leveraging the inherent similarities across various biological data types. + +<--- Page Split ---> + +The scSemiProfiler represents a groundbreaking shift in single- cell analysis, particularly for large- scale and cohort studies, by offering a cost- effective yet comprehensive approach. It achieves this by selectively performing single- cell experiments on a few samples using active learning, coupled with deep generative models for in silico inference of single- cell profiles for the remaining samples, thereby creating similar semi- profiled cohorts at a substantially lower cost. This approach not only alleviates financial constraints in expansive biomedical research but also extends its utility beyond cohort studies. The tool's deep generative model enables "single- cell level deconvolution" across various studies, provided there is a single- cell reference. Initially centered on RNA sequencing datasets, scSemiProfiler's adaptability to other single- cell and bulk data modalities opens new avenues in personalized medicine and can significantly enrich global data repositories. Significantly, scSemiProfiler is designed to complement, rather than compete with, large- scale single- cell sequencing technologies, such as WT Mega by Parse Biosciences. WT Mega is notable for its ability to process up to 1 million cells across 96 samples. When scSemiProfiler is integrated with these advanced sequencing platforms, it further reduces costs, enabling single- cell level profiling of thousands of samples in large- scale studies at a more manageable expense. This synergy significantly broadens the scope and depth of possible research, enhancing the efficiency and affordability of large- scale single- cell studies. + +## Methods + +## The initial setup of scSemiProfiler + +The initial setup of the scSemiProfiler method plays a pivotal role (depicted in Fig. 1 a), serving as the foundation for subsequent semi- profiling iterations. Initially, each individual sample within the designated cohort undergoes bulk sequencing. Upon obtaining the raw count data from this sequencing, we perform library size normalization followed by a log1p transformation. From the processed data, we identify and select highly variable genes and relevant markers for further analysis. Subsequently, we employ Principal Component Analysis (PCA) [67- 69], implemented in the Python package Scikit- learn [70], to reduce the dimensionality of the processed bulk data. We then determine the batch size of representatives ( \(B\) ) representing the number of selected representatives for actual single- cell profiling in each iteration. Once the representative batch size is defined, we cluster the dimensionality- reduced bulk data to select the initial batch of representatives. The dimensionality- reduced data is clustered into \(B\) distinct clusters using the KMeans algorithm [71], the sample that is closest to the cluster centroid is designated as the representative for that cluster. This well- prepared initial setup provides the clustered bulk sequencing data and identified representatives for high- resolution single- cell sequencing. + +## Representative single-cell profiling and processing + +After the selection of representatives, the scSemiProfiler method proceeds to the single- cell profiling phase, focusing specifically on the chosen representatives (Fig. 1b). The raw count data obtained from single- cell profiling undergoes a sequence of preprocessing steps. These steps encompass traditional quality controls as well as advanced processing techniques aimed at enhancing the learning of the deep generative model. + +To initiate, expression values that are extremely low (below \(0.1\%\) of the library size) are adjusted to zero. This adjustment is based on insights from prior studies [72, 73], which have demonstrated that these minimal values are often representative of background noise and, therefore, should be excluded. The threshold for this processing is empirically determined by selecting the value that yields the most high- quality UMAP visualizations consistent with the original cell type annotations. Subsequently, a series of standard preprocessing steps commonly used by existing methods [74, 75] is applied, including the removal of low- quality cells (expressing fewer than 200 genes), genes (expressed in fewer than 3 cells), dead cells (with a high proportion of mitochondrial reads), and doublets (expressing an excessive number of genes). Afterward, the data is normalized to a library size (total count of each cell) of 10,000, followed by a log1p transformation. The data matrix is then cropped to include only the highly variable genes identified from the bulk data. Our preprocessing approach is carefully tailored to meet the unique requirements of each dataset. For instance, the gene expression data of the COVID- 19 dataset arrive in a preprocessed state, including cell quality control, normalization, and log1p transformation. We retrieved the data as normalized RNA counts, eliminated the background noise, and subsequently reapplied normalization and log1p transformation. Finally, we selected the columns corresponding to highly variable genes and important cell type markers. In contrast, the + +<--- Page Split ---> + +colorectal cancer dataset is provided in raw counts, necessitating a full spectrum of standard single- cell preprocessing procedures. With the iMGL dataset, our focus is on cells annotated by the dataset provider. We identify genes common to both bulk and single- cell datasets and apply standard preprocessing steps, excluding the removal of low- quality cells, as this task has already been performed by the dataset provider. + +After the initial preprocessing, the single- cell data is further refined by integrating two types of prior knowledge to enhance cell representations. Firstly, gene set scores are calculated using all curated gene sets from MSigDB [48] and Ernst et al [76], sourced from UNIFAN [77]. Gene set scores are determined by averaging the expression values for each gene set, followed by a log1p transformation, and then combined with the preprocessed single- cell gene expression matrix. Secondly, we adopt a feature weighting method, giving more weight to features with higher variance in each cell's input when calculating reconstruction loss. More comprehensive details of this strategy are available in the section focusing on single- cell inference. + +## Pretrain the deep generative learning model for reconstructing the single-cell data of the selected representatives + +The next phase, depicted in Fig.1c, involves training a deep generative learning model for the in silico single- cell data inference of a non- representative target sample using processed representative single- cell data and bulk data of both samples. This will be executed for all non- representative samples to ensure everyone in the study cohort will have single- cell data (real- profiled or semi- profiled) in the end. Firstly, our deep generative model aims to reconstruct the single- cell profiles of representatives. This reconstruction lays the foundation of the single- cell inference of the target samples, which can be viewed as a modified single- cell data reconstruction task. When performing the single- cell inference, the initial single- cell data reconstruction of the representative is modified by introducing the difference between the target sample and the representative based on their distinct bulk data profiling, aligning the synthetic single- cell data for the target sample closer to its actual single- cell profiling counterpart. The fundamental reasoning is that the single- cell data matrix of a specific target sample should bear resemblance to the representatives chosen from the same study. Significant differences between them can be adjusted and guided by the disparity as delineated by the bulk sequencing data. + +We designed a VAE- GAN- based model for the representative reconstruction and the target single- cell generation guided by their expression difference at the bulk level. It ingests gene expression data into an MLP encoder, which subsequently outputs parameters of a multivariate Gaussian. A random variable, \(\mathbf{z}\) , is sampled from this Gaussian and processed through an MLP decoder, resulting in parameters of a Zero- Inflated Negative Binomial (ZINB) distribution - optimal for modeling single- cell RNA- seq data [78- 80]. Training the VAE parameters involves maximizing the data likelihood and minimizing the KL divergence [81] between the latent variable distribution and a standard Gaussian distribution, following the Evidence Lower BOund (ELBO) loss framework [30]. + +Innovating upon this foundational VAE structure, we integrated the following four techniques to enhance effective cell representation learning. Gene Set Score Inclusion: As mentioned in the data preprocessing section, beyond mere gene expression reconstruction, we compute gene set scores for cells and concatenate them to the gene expression data. During the two stages of pretrain, we also compute the reconstruction loss for gene set scores. Such inclusion furnishes the model with an enriched biological context, fostering a more comprehensive learning of the input cells. Hence, input for each cell becomes the concatenation of gene expression and gene set scores: \((\mathbf{x}_1, \mathbf{s}_1)\) . Feature Importance Weight: We also compute a weight vector \(\mathbf{w}\) to weight the contribution of each feature to the VAE's reconstruction loss based on the feature's variance. Graph Convolutional Networks (GCN) [82, 83]: A GCN layer assimilates adjacent cells' information, mitigating dropout concerns and the inherent noise of single- cell sequencing. Generative Adversarial Network (GAN) [31] Dynamics: We employ a discriminator network, resonating with GAN structures, which discerns between genuine and generated cell data. This guides the generator towards producing more authentic single cell data. The resultant loss function for the generator discussed above is: + +\[L_{G_{P r e t r a i n 1}} = L_{V A E} - \lambda_{d}L_{D_{F a k e}} \quad (1)\] + +\[L_{VA E} = \sum_{i = 1}^{N}[-{\bf w}\cdot log(Z I N B(({\bf x}_{1},{\bf s}_{1})|\rho (z_{i}),\pi (z_{i}),\theta))] + K L(q({\bf z}|({\bf x}_{1},{\bf s}_{1}))||{\mathcal{N}}({\bf 0},I))] \quad (2)\] + +<--- Page Split ---> + +\[L_{D_{F a k e}} = \sum_{i = 1}^{N} - l o g(1 - D(G((\mathbf{x}_{i},\mathbf{s}_{i})))) \quad (3)\] + +\(L_{G_{P r e t r a i n 1}}\) is the loss used to train the VAE- based generator network during the first pertrain stage. It has two components, a VAE loss \(L_{VA E}\) and another loss \(L_{D_{F a k e}}\) including the feedback from the discriminator. In \(L_{V A E}\) \(N\) is the number of cells in the representative's single- cell dataset. \(\mathbf{x_{i}}\) is the vector representing the gene expression value of the \(i\) - th cell in the dataset and \(\mathbf{s_{i}}\) is the corresponding gene set score. \(\mathcal{N}(\mathbf{0},I)\) is the standard Gaussian distribution. \(\mathbf{z_{i}}\) is a low- dimensional latent variable sampled from the Gaussian distribution \(q(\mathbf{z}|\mathbf{(x_{i},s_{i})})\) whose parameters \(\mu_{i}\) and \(\pmb{\Sigma}_{i}\) are generated by the encoder network: \(M L P_{E n c o d e r}(G C N((\mathbf{x_{i}},\mathbf{s_{i}})))\) . We follow the ZINB distribution design of SCVI [78] and generate ZINB parameter using the decoder network: \(\pmb {\rho}(\mathbf{z_{i}}) = M L P_{D e c o d e r,\rho}(\mathbf{z_{i}})\) \(\pi (\mathbf{z_{i}}) = M L P_{D e c o d e r,\pi}(\mathbf{z_{i}})\) and \(\theta\) is a free parameter vector that are learned in the training process. In \(L_{D_{\mathrm{F a k e}}}\) \(G\) represents the VAE- based generator so \(G((\mathbf{x_{i}},\mathbf{s_{i}}))\) represents the \(i\) - th cell reconstructed by the generator. The reconstructed cell data is the mean of the generator's \(Z I N B\) distribution. And \(D\) is our discriminator network, so \(D(G((\mathbf{x_{i}},\mathbf{s_{i}})))\) is the discriminator's predicted probability of the \(i\) - th reconstructed cell being a real- profiled cell. \(\lambda_{d}\) denotes an empirically determined scaling factor. Meanwhile, the discriminator network will be trained using cross- entropy loss to reach higher classification performance. + +\[L_{D} = \sum_{i = 1}^{N}[-log(D((\mathbf{x_{i}},\mathbf{s_{i}}))) - log(1 - D(G((\mathbf{x_{i}},\mathbf{s_{i}}))))] \quad (4)\] + +Here, \(D((\mathbf{x_{i}},\mathbf{s_{i}}))\) represents the discriminator's predicted probability of the \(i\) - th input cell being real when it is indeed real, and \(D(G((\mathbf{x_{i}},\mathbf{s_{i}})))\) , as mentioned in a previous paragraph, represents the discriminator's predicted probability of the \(i\) - th cell being real when it is, in fact, a cell reconstructed by the generator network. When trained together, the generator will first be trained until convergence, followed by alternating training with the discriminator every three epochs. Together, they form a GAN with a min- max objective: + +\[m i n_{G}m a x_{D}\sum_{i = 1}^{N}[l o g(D((\mathbf{x_{i}},\mathbf{s_{i}}))) + l o g(1 - D(G((\mathbf{x_{i}},\mathbf{s_{i}}))))] \quad (5)\] + +In this context, the discriminator's objective is to maximize its accuracy, aiming for precise classification, while the generator's goal is to minimize discrepancies, generating realistic cells to challenge the discriminator. + +Overall, our approach involves two pretrain stages and one fine- tune stage for performing single- cell inference. In the initial pretrain stage, we employ \(L_{G_{P r e t r a i n 1}}\) and \(L_{D}\) to train the generator \(G\) and discriminator \(D\) , respectively, forming a GAN. The primary objective here is to train a generator capable of reconstructing genuine cell data from the representatives. Subsequently, the second pretraining stage closely resembles the first, but it functions in full- batch mode. This modification enables the inclusion of an extra bulk loss term in the generator's loss function. Such an addition improves the model's capacity to leverage insights from bulk data and better aligns the single- cell data with its bulk counterpart. This enhancement is crucial for priming the model for the in silico inference of single- cell data from bulk data, while also strengthening the data reconstruction process. + +\[\begin{array}{r}{L_{G_{P r e t r a i n 2}} = L_{V A E} - \lambda_{d}L_{D_{F a k e}} + \lambda_{B u l k R}L_{B u l k R}}\\ {L_{B u l k R} = \left\| \frac{\sum_{i = 1}^{N}\mathbf{x_{i}^{\prime}}}{N} - \frac{\sum_{i = 1}^{N}\mathbf{x_{i}}}{N}\right\|_{2}} \end{array} \quad (6)\] + +The term \(\lambda_{B u l k R}\) represents another empirical hyperparameter for regularization weight. \(\mathbf{x_{i}^{\prime}}\) represents the reconstructed \(i\) - th cell. The bulk loss is defined as the disparity between the pseudobulk of the real representative dataset and that of the reconstructed representative dataset. + +## Fine-tune the deep generative learning model to infer the single-cell measurements for the target samples + +After successfully pretraining our deep generative model so that it can perform accurate single- cell data reconstruction, the subsequent stage of our method leverages this foundation. In this phase, + +<--- Page Split ---> + +we introduce an extra term to the loss function that accounts for the difference between the bulk data of the representative and that of the target sample. This fine- tuning process refines the model's capabilities, enabling it to perform in silico generation of cells from the target sample while maintaining consistency with the observed bulk expression patterns. Consequently, our deep generative model transitions from being solely focused on reconstruction to generating single- cell data for the target sample. In this crucial stage, we retire the discriminator from our model architecture, as our focus shifts from generating cells that resemble the representatives' cells to adapting the model for the target samples. Consequently, the training process exclusively involves the generator. The loss function for the generator adopted in this stage bears resemblance to that of the second pretrain phase, with the distinction of the bulk loss transitioning from \(L_{BulkR}\) to \(L_{BulkT}\) : + +\[L_{\mathrm{Inference}} = L_{\mathrm{VAE}} + \lambda_{\mathrm{BulkT}}L_{\mathrm{BulkT}} \quad (8)\] + +\[L_{\mathrm{BulkT}} = \left\| \frac{\sum_{i = 1}^{N}\mathbf{x}_{1}^{i}}{N} -\frac{\sum_{i = 1}^{N}\mathbf{x}_{1}}{N}\cdot \frac{\mathbf{B}_{\mathrm{T}} + \epsilon}{\mathbf{B}_{\mathrm{R}} + \epsilon}\right\|_{2} \quad (9)\] + +Here, \(\mathbf{B}_{\mathrm{T}}\) and \(\mathbf{B}_{\mathrm{R}}\) represent the real bulk data of the target and representative samples, respectively, while \(\epsilon\) , whose elements are all set to 0.1, is introduced as a pseudocount vector to avoid division by zero. The second term of equation 9, \(\frac{\sum_{i = 1}^{N}\mathbf{x}_{1}}{N}\frac{\mathbf{B}_{\mathrm{T}} + \epsilon}{\mathbf{B}_{\mathrm{R}} + \epsilon}\) , functions as an estimation of the pseudobulk value for the target sample, derived from the actual bulk data. This term is incorporated to reduce the systematic discrepancy between the pseudobulk and the actual bulk data corresponding to the same sample. + +By incorporating the actual bulk difference alongside the pseudobulk data of the representative, this term effectively simulates the pseudobulk data for the target sample, ensuring that any disparities between the pseudobulk and actual bulk do not distort the model's performance. In cases such as the COVID- 19 and colorectal cancer dataset, where real bulk data is unavailable, the entire second term simplifies to the target's pseudobulk. Furthermore, as the bulk loss \(L_{BulkT}\) inherently propagates more gradient to higher gene expression values (given the target bulk loss is in a squared error form), we implement a strategy to gradually reduce the gradient received by larger values. See details in the "Deep Generative Learning Model Training Details" section. Concurrently, we increase the weight of the bulk loss in the loss function over the course of training. This refined approach ensures that, after adequately training for larger values, the model can fine- tune smaller values, resulting in a comprehensive and precise single- cell data generation for the target samples. To counteract the influence of potentially noisy gene expression with minuscule values, we introduced a neuron activation threshold. This threshold converts values that are exceedingly small, specifically those below 0.1% of the input library size (a threshold consistent with that used for background noise removal in preprocessing), to zero. + +## Deep generative learning model architecture + +In this study, our model integrates a Variational Autoencoder (VAE) as the generator and a Multi- Layer Perceptron (MLP) as the discriminator. We developed both components using PyTorch [84]. A concise overview of the model architecture is provided below, and for an in- depth understanding, we encourage consulting the complete details available in our GitHub repository (refer to the code availability section). + +Before sending the cell data into the model, we performed some precomputation for the cells, which is part of a Graph Convolutional Network (GCN) layer. We aggregated information from the K (15 by default) nearest neighboring cells for each input cell. The neighbor finding for each cell was based on the 'neighbors. neighbors. graph' tool in the Scikit- learn package. We preprocessed this message- passing step on the log1p transformed cell expression data, subsequently inputting the aggregated cell data into the deep learning model. The encoder of the VAE then employs linear layers and activation functions, leading to a hidden layer comprising 256 neurons. The message passing, the first linear layer, and the first activation function in the encoder form a GCN layer together. The design of the rest of the VAE structure, inclusive of both the encoder and decoder, was adapted from SCVI. We chose 32 for the latent space dimension and 256 for the decoder's hidden layer size. For the discriminator, we constructed a 4- layer MLP, with Leaky ReLU activations [85] between linear layers. The hidden layers are configured with 256, 128, and 10 neurons, respectively. The output layer employs a Sigmoid activation to produce the prediction probability of the input cell being real. Weight initialization in the model utilizes the Kaiming method [86]. + +<--- Page Split ---> + +## Deep Generative Learning Model Training Details + +Following the detailed outline of the VAE- GAN model structure, we next describe the detailed training settings, which were meticulously designed to optimize the model's performance. The training process was structured in a three- stage sequence to ensure comprehensive learning and fine- tuning of the model parameters. + +Initially, in the first pretraining stage, we focus on training the VAE generator independently until convergence (setting 100 epochs as a default value). Then, the generator and discriminator are trained jointly until the generator cannot further reduce its loss. When trained together, the generator and the discriminator are trained alternatively for 3 epochs each, and this will go for 100 iterations by default. During this phase, we employed the default SCVI learning rate of \(1 \times 10^{- 3}\) . + +The second pretraining stage mirrors the first in its basic approach but incorporates notable modifications. It is executed in full- batch mode, with an additional representative bulk loss integrated into the training process. Here, the generator is again trained until convergence (set to 50 epochs by default), and then trained jointly with the discriminator until convergence. The default setting is training the two networks for 50 iterations, where each network is trained for 3 epochs in each iteration. To facilitate more refined adjustments during this stage, the learning rate is reduced to \(1 \times 10^{- 4}\) . + +In the final stage of training, the single- cell inference fine- tuning stage, the discriminator is set aside, and training is executed through a series of mini- stages. Each mini- stage involves setting the gradients for highly expressed genes above specific thresholds to zero, a strategy aimed at ensuring equitable optimization for genes with smaller expression values. These thresholds are calculated relative to the peak expression value (max. expr) from the representative's normalized count matrix, set at No threshold, \(\frac{1}{2} \max . \exp r\) , \(\frac{1}{4} \max . \exp r\) , \(\frac{1}{6} \max . \exp r\) , and \(\frac{1}{8} \max . \exp r\) . Training is continued until convergence in each mini- stage, typically not exceeding 150 epochs, with a learning rate of \(2 \times 10^{- 4}\) . As the thresholds are lowered, the weight for the target bulk loss in the loss function is progressively quadrupled, ensuring a consistent total magnitude for this loss component. Upon completion of all mini- stages, the trained model is then utilized to generate single- cell data for the target sample. + +## Incrementally select representatives using active learning to improve single-cell inference + +Upon executing the deep generative learning models for in silico inference, every sample in the cohort possesses single- cell data, either real- profiled or semi- profiled. While researchers can opt to conclude here and proceed with downstream analysis using the single- cell data, when budget permits, our active learning module can further refine the process. By pinpointing additional informative representatives for single- cell sequencing, some target samples will be assigned a more similar representative, leading to enhanced single- cell inference performance. + +Our initial representative selection strategy is predicated on the belief that selecting the most "uncertain" samples as the representatives results in the most inference improvement and yields the richest insights for our model. Consequently, we select the sample with the largest bulk data difference compared to its representative to serve as the new representative: + +\[R_{new} = \underset {i\in C o h o r t}{\arg \max}D_{b}(R(i),i) \quad (10)\] + +Where \(R(i)\) is the representative of \(i\) , and \(D_{b}\) is the Euclidean distance between the PCA- reduced bulk data of two samples. This process is executed \(B\) (representative batch size) times. Upon choosing new representatives, we update the membership for all samples. This ensures that, if a sample aligns more closely with a new representative than their previous one, their affiliation shifts: + +\[R(i) = \underset {r\in R s}{\arg \min}D_{b}(r,i) \quad (11)\] + +where \(R(i)\) is the representative of the sample \(i\) and \(R s\) is the set of representatives. Conceptually, this approach is similar to the uncertainty sampling [64] algorithm, a subtype of active learning. However, this method is passive learning since it does not utilize the knowledge acquired by the base learners, our deep generative learning models. + +To improve the representative selection, we developed our active learning algorithm by merging the information gained by the deep generative model base learners into the selection of new + +<--- Page Split ---> + +representatives. To commence, we pinpoint the \(B\) sample clusters exhibiting the highest in- cluster heterogeneity, which is the aggregate distance from each sample to the cluster's representative. The heterogeneity for a cluster \(c\) is expressed as: \(\sum_{i\in c}[D_{b}(R_{c},i) + \lambda_{sc}D_{sc}(R_{c},i) + \lambda_{pb}D_{pb}(R_{c},i)\) , where \(i\) can be any sample in the cluster and \(R_{c}\) is the representative. Essentially, a sample's heterogeneity is the combined distance—across three distance metrics—between the sample and its representative. The total cluster heterogeneity is derived from the summation of all sample heterogeneity. \(D_{b}\) denotes the Euclidean distance between the PCA- reduced bulk data of two samples. The term \(D_{sc}(R_{c},i)\) represents the difficulty of transforming the representative single- cell data to that of a target. This is quantified by averaging the Euclidean distances between the representative cells and their K nearest neighbors in the target sample, all within a PCA- reduced space. Formally: + +\[D_{sc}(a,b) = \frac{1}{N_{a}}\sum_{i = 1}^{N_{a}}\sum_{j\in KNN(i)}ED(\mathbf{v_{a,i}},\mathbf{v_{b,j}}) \quad (12)\] + +with \(a,b\) being two samples in the semi- profiled cohort, \(v_{a},v_{b}\) being PCA reduced single- cell data matrices for the two samples, \(ED(x,y)\) computing the Euclidean distance between \(x\) and \(y\) , and \(KNN(i)\) identifying the cell's \(K\) (1 by default) nearest neighbor in another sample's data matrix. Subscript represents rows, e.g. \(\mathbf{v_{a,i}}\) is the \(i\) th row (ith cell) of the PCA reduced data of sample \(a\) . \(D_{pb}\) is the Euclidean distance of the log1p transformed pseudobulk data of two samples, which is positively related to the amount of bulk loss and thus quantifies the deep learning model's difficulty in performing the in silico inference. Finally, \(\lambda_{sc}\) and \(\lambda_{pb}\) are empirical scaling factors. + +The most heterogeneous cluster \(C_{H}\) is chosen as: + +\[C_{H} = \underset {c\in \mathrm{clusters}}{\arg \max}\sum_{i\in c}[D_{b}(R_{c},i) + \lambda_{sc}D_{sc}(R_{c},i) + \lambda_{pb}D_{pb}(R_{c},i)] \quad (13)\] + +After selecting the \(B\) most heterogeneous clusters to split, the subsequent step involves selecting a new representative for each of these \(B\) clusters to minimize total in- cluster heterogeneity. Within each cluster, every non- representative sample is a potential new representative, each representing a different way of splitting the cluster. In the cluster, samples that are closer in bulk distance \(D_{b}\) to the potential new representative than to the original representative can be reassigned to the new representative, splitting the cluster into two. The total heterogeneity of these two new clusters is then calculated based on the bulk distance. For each of the \(B\) most heterogeneous clusters, we select the sample whose corresponding split results in the minimal total in- cluster heterogeneity to be the new representative \(R_{new}\) , as shown in the equation below. + +\[R_{new} = \underset {j\in C_{H}}{\arg \min}\sum_{i\in C_{H}}min(D_{b}(j,i),D_{b}(R_{c},i)) \quad (14)\] + +Where \(R_{c}\) denotes the original representative of this cluster. Finally, the cluster membership updating will be executed to make sure each sample \(i\) is assigned to the closest representative \(R(i)\) . + +\[R(i) = \underset {r\in R_{B}}{\arg \min}D_{b}(r,i) \quad (15)\] + +With the identification of new representatives, they are subjected to real single- cell data profiling and appended to the existing collection. In subsequent semi- profiling iterations, certain target samples will be assigned with more analogous representatives, enhancing in silico single- cell inference accuracy. This iterative procedure persists until either the budget is exhausted or a sufficient number of representatives are ascertained, ensuring optimal semi- profiling results. + +## Semi-profiling pipeline stop criteria + +For our semi- profiling pipeline that iteratively selects representatives and performs in silico single- cell inference, we recommend two types of stop criteria for the users to choose according to their own needs. First, in the case that the user knows their budget and aims to achieve the best possible semi- profiling performance, they can run the pipeline and keep choosing representatives until they run out of budget. Second, in the case that the user aims to achieve an acceptable semi- profiling performance with the least amount of budget, the iterative cycle persists until an acceptable performance is reached. The performance is measured by comparing the representatives' actual single- cell data and + +<--- Page Split ---> + +the inferred data for them in previous iterations. Based on the user's needs, we recommend three levels of criteria. First, when the user is mainly studying the cell type proportion, we recommend stopping when the Pearson correlation between the ground truth and the estimated proportion reaches 0.85. Second, when the user's study includes more detailed analysis, such as biomarker discovery, we recommend stopping when the Pearson correlations between the biomarker expression patterns (dot colors and sizes in the dot plots) discovered in real- profiled data and those in semi- profiled data reach 0.85. Finally, for studies requiring more intricate single- cell level analysis, it is advised to halt when the Pearson correlation between data matrices, generated from advanced analysis tasks like trajectory inference and cell- cell interaction, for semi- profiled and real- profiled data, reaches 0.85. These guidelines are merely suggestions, and users have the flexibility to modify the stopping criteria to suit their specific requirements. The effectiveness of the semi- profiling process can be consistently evaluated by comparing the semi- profiled single- cell data (from the previous round) with the actual single- cell profiling (in the subsequent round). If the actual single- cell profiling aligns with the semi- profiling generated by our model from the previous round, this indicates a robust semi- profiling model has been effectively learned, and the iterative process can thus be concluded. + +## Semi-profiling performance evaluation + +Cell type Annotation for Semi- Profiled Data: Utilizing the annotated single- cell data from the representatives, we trained an MLP classifier available in the Scikit- learn package [70]. This classifier was then employed to categorize the cell types for cells in the semi- profiled data. + +Dimensionality Reduction: In order to juxtapose the real- profiled and semi- profiled cohorts, we merged them and executed Principal Component Analysis (PCA) to reduce it to 100 principal components. This PCA- processed data serves as the foundation for computing single- cell level difference metrics. For visualization purposes, we first used SCANPY [75] to compute the neighbor graph using the PCA- reduced data, with the local neighborhood size parameter set to 50. Subsequently, the Uniform Manifold Approximation and Projection (UMAP) functionality within SCANPY was used for the visualization process. + +Semi- profiling error quantification: To assess the accuracy of the semi- profiled cohort, we performed PCA to reduce its dimensionality, in conjunction with the real- profiled data, to 100 dimensions. For each sample within the semi- profiled dataset, we calculated a single- cell difference metric in comparison to the real- profiled single- cell data corresponding to that sample. This involves determining the Euclidean distance of each cell to its K- nearest neighbors in another dataset within the PCA- reduced space, followed by averaging these distances across all cells. The formula below outlines the difference computation between single- cell matrices \(\mathbf{m}_{\mathbf{a}}\) and \(\mathbf{m}_{\mathbf{b}}\) . + +\[D i f_{s c}(\mathbf{m}_{\mathbf{a}},\mathbf{m}_{\mathbf{b}}) = \frac{1}{N_{a} + N_{b}} (\sum_{i = 1}^{N_{a}}\sum_{j\in K N N(i)}E D(\mathbf{v}_{\mathbf{a},i},\mathbf{v}_{\mathbf{b},j}) + \sum_{i = 1}^{N_{b}}\sum_{j\in K N N(i)}E D(\mathbf{v}_{\mathbf{b},i},\mathbf{v}_{\mathbf{a},j})) \quad (16)\] + +Where \(D i f_{s c}\) denotes single- cell level difference. \(N_{a}\) and \(N_{b}\) signify the cell counts in matrices \(\mathbf{m}_{\mathbf{a}}\) and \(\mathbf{m}_{\mathbf{b}}\) respectively. The vs represent the PCA- reduced matrices. The terms \(\mathbf{v}_{\mathbf{a},i}\) , \(\mathbf{v}_{\mathbf{b},j}\) represent the \(i\) th and \(j\) th cell (rows) in the PCA- reduced matrices of sample \(a\) and \(b\) respectively. The function \(K N N\) identifies the \(K\) nearest neighbors within the alternate sample's PCA- reduced data matrix. We chose \(K = 1\) for our experiments. \(E D\) computes the Euclidean distance between two vectors. + +Referring to the comprehensive performance curves depicted in (Fig.2d, Fig.4d, Fig.6d, the normalized error \(E_{N}\) displayed on the left y- axis represents the mean single- cell difference between each sample's real- profiled version and semi- profiled version. This error is normalized using a theoretical upper bound ( \(U B\) ) and a lower bound ( \(L B\) ) of an expected single- cell difference between samples in the study. + +\[E_{N} = \frac{1}{U B - L B} (\frac{1}{P}\sum_{i = 1}^{P}D i f_{s c}(\mathbf{m}_{\mathbf{i}},\tilde{\mathbf{m}}_{\mathbf{i}}) - L B) \quad (17)\] + +and if \(i\) itself is a representative, the single- cell difference equals the lower bound. + +\[D i f_{s c}(\mathbf{m}_{\mathbf{i}},\tilde{\mathbf{m}}_{\mathbf{i}}) = L B,\quad \mathrm{if} i\in R s\] + +<--- Page Split ---> + +where \(P\) stands for the total count of samples within the cohort and \(\tilde{\mathbf{m}}_1\) refers to the semi- profiled version of \(\mathbf{m}_1\) . \(R\) s denotes the set of representatives. The upper bound \(UB\) was derived by averaging the single- cell difference of randomly paired samples based on their PCA- reduced real- profiled data. This represents the performance of the worst possible representative selection, which is random selection. For computing the lower bound \(LB\) , we first divided the comprehensive PCA matrix, which contains all single- cell samples, into two random halves. Due to the randomness of the split, these two halves have the same data probabilistic distribution. Furthermore, given the large sample size, these halves are expected to have approximately the same actual cell distribution and thus can be regarded as two replicates of the same study. The difference between the replicates should be regarded as the lower bound, as it represents the best performance that an optimal selection can achieve [87]. + +Deconvolution benchmarking: We benchmarked the accuracy of our deconvolution method by comparing its performance against established methods: CIBERSORTx, Bisque, TAPE, Scaden, and DWLS. We ran CIBERSORTx using the official web portal. Bisque, TAPE, and DWLS were installed according to the official GitHub repositories provided in the corresponding publications. We used the PyTorch version of Scaden implemented in TAPE's GitHub repository for its testing. First, we annotated our cells generated by the deep learning model using the MLP classifier, which was trained on the representatives' annotated data, as discussed previously. Using this annotated data, we computed the cell type proportion. Subsequently, we employed the Pearson correlation coefficient (calculated using the Python package SciPy [59]) and RMSE to assess the consistency with the ground truth. However, when applying the aforementioned methods to the three datasets, we encountered both time and space complexity constraints for many benchmarked methods. Specifically, CIBERSORTx and Bisque were unable to utilize as many representative samples for single- cell references as our approach. Scaden and TAPE always only sample 5,000 cells for training their models, thereby also failing to fully exploit all available representatives. To ensure a fair comparison, we also offered an alternative version of our results for these datasets. In this version, the single- cell reference is confined to just the first 4 representatives, aligning with the capacity of all other benchmarked methods. Across all three datasets, DWLS failed to produce results even after running for a week and was thus excluded from the comparison. To statistically ascertain if one method notably surpasses another in performance, we calculated the p- values using a one- sided Wilcoxon test [62], as implemented in SciPy. This test enables us to determine whether the metric of one method is significantly higher or lower compared to that of another. + +Gene set activation scores computation and heatmap plotting: We computed the interferon pathway activation scores following the same procedure in the study [35] from which we acquired the COVID- 19 dataset. To compute the activation scores depicted in the heatmaps (refer to Fig.2c), we first computed the average expression for each cell type and severity combination using the code from the COVID- 19 dataset provider, which uses the 'tl.score_genes' tool in SCANPY. Then, for each cell type, we computed the activation score of each severity level as the fold changes from the healthy condition to it. For the "activation of immune response" activation pattern in the colorectal cancer dataset depicted in Fig.4c, we first collected the genes of this GO term (GO:0002253) and used the same SCANPY tool to compute the average expressions for each cell type and tissue combination. For each cell type, we calculated the activation scores by determining the fold changes in comparison to the "Normal" tissue type across all other tissue types. For the "activation of immune response" activation pattern in the iMGL dataset depicted in Fig.6c, we first computed the average gene expression for each cell type and tissue combination similarly. The activation score was determined by calculating the fold change for each cell type, using "C1: Homeostatic non- proliferative" as the reference background. Considering the relatively smaller size of the colorectal cancer and iMGL dataset and the fact that some entries have very few cells, the values were only computed for entries with more than 500 cells for the colorectal cancer dataset and entries with more than 100 cells for the iMGL dataset. + +Biomarker discovery and enrichment analysis: For making the RRHO plots, we used SCANPY to identify each cell type's top 50 positive cell type markers and top 50 negative cell type markers for the real- profiled and semi- profiled datasets. RRHO plots are used for visualizing the overlap between two ranked gene lists. Each entry in the plot corresponds to a negative logged p- value of the hypergeometric test between two marker lists. The bottom left quadrant corresponds to the + +<--- Page Split ---> + +comparison between two positive marker lists. In this quadrant, the plot entry in the \(i\) - th row from bottom to top and \(j\) - th column from left to right corresponds to the comparison between the top \(i\) genes in the real- profiled version top negative markers and top \(j\) gene in the semi- profiled version top negative markers. Other quadrants are plotted similarly, with the marker lists always starting from the respective corners of the plot. + +For the GO and Reactome enrichment analysis, we identified the top 100 cell type signature genes for both real- profiled and semi- profiled datasets using SCANPY. The identified markers were subsequently employed for enrichment analysis with the Python package GSEApy [88]. For enrichment analysis visualization in Fig.3c, Fig.5c, Fig.7c, we employed the adjusted p- value from GSEApy's output to indicate significance. + +Cell- cell interaction analysis: We conducted cell- cell interaction analyses on both real- profiled and semi- profiled datasets using the R package CellChat [89]. For the COVID- 19 dataset, we analyzed interactions of cells from patients with each disease severity. In the colorectal cancer dataset, our focus was on interactions of cells originating from different tissues, and in the iMGL dataset, we evaluated cells across various condition groups. + +Pseudotime analysis: We carried out pseudotime analysis utilizing the Monocle3 R package [90]. The UMAP coordinates, previously generated in our UMAP comparison between real- profiled and semi- profiled datasets, served as the input for pseudotime computation. To ensure a fair comparison between the real- profiled and semi- profiled data, we consistently selected the roots at the same positions within the UMAP space. To compare the similarity between the pseudotime values computed for the two datasets, we employed the Mann- Whitney U test, as implemented in SciPy. + +Cell trajectory analysis using partition- based graph abstraction (PAGA): The PAGA plots were generated using SCANPY. First, PCA was applied independently to each dataset to reduce them to 100 principal components. Subsequently, the PCA- reduced data was utilized to compute the neighbor graphs, with the size of the local neighborhood set to 50, aligning with our UMAP settings. Based on the neighbor graphs, we generated the PAGA plots using SCANPY. + +## Data availability + +Three datasets are used for evaluating our method. The preprocessed COVID- 19 dataset is from Stephenson et al.'s study [35] and can be downloaded from Array Express under accession number E- MTAB- 10026 The colorectal cancer dataset is from Joanito et al.'s study [53]. The count expression matrices are available through Synapse under the accession codes syn26844071 The iMGL dataset is from Ramaswami et al.'s study [60]. The raw count iMGL bulk and single- cell data can be downloaded from the Gene Expression Omnibus (GEO) repository under accession number GSE226081. + +## Code availability + +Code for the models and results reproduction is publically available on GitHub: https://github.com/mcgilldinglab/scSemiProfiler. + +## Acknowledgments + +This work was funded in part by grants awarded to [JD]. We gratefully acknowledge the support from the Canadian Institutes of Health Research (CIHR) under Grant Nos. PJT- 180505; the Funds de recherche du Québec - Santé (FRQS) under Grant Nos. 295298 and 295299; the Natural Sciences and Engineering Research Council of Canada (NSERC) under Grant No. RGPIN2022- 04399; and the Meakins- Christie Chair in Respiratory Research. We also extend our gratitude to Yanshuo Chen, the lead author of TAPE, for his indispensable support in the deconvolution benchmarking experiments. This work is part of the HCA publication bundle. + +## References + +[1] Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X., Bodeau, J., Tuch, B.B., Siddiqui, A., et al.: mrna- seq whole- transcriptome analysis of a single cell. Nature methods + +<--- Page Split ---> + +6(5), 377- 382 (2009) + +[2] Shapiro, E., Biezuner, T., Linnarsson, S.: Single- cell sequencing- based technologies will revolutionize whole- organism science. Nature Reviews Genetics 14(9), 618- 630 (2013) + +[3] Li, Y., LeMaire, S.A., Shen, Y.H.: Molecular and cellular dynamics of aortic aneurysms revealed by single- cell transcriptomics. Arteriosclerosis, Thrombosis, and Vascular Biology 41(11), 2671- 2680 (2021) + +[4] Kazer, S.W., Aicher, T.P., Muema, D.M., Carroll, S.L., Ordovas- Montanes, J., Miao, V.N., Tu, A.A., Ziegler, C.G., Nyquist, S.K., Wong, E.B., et al.: Integrated single- cell analysis of multicellular immune dynamics during hyperacute hiv- 1 infection. Nature medicine 26(4), 511- 518 (2020) + +[5] Chai, R.L., Grandis, J.R.: Advances in molecular diagnostics and therapeutics in head and neck cancer. Current treatment options in oncology 7, 3- 11 (2006) + +[6] Marquette, C.- H., Boutros, J., Benzaquen, J., Ferreira, M., Pastre, J., Pison, C., Padovani, B., Bettayeb, F., Fallet, V., Guibert, N., et al.: Circulating tumour cells as a potential biomarker for lung cancer screening: a prospective cohort study. The Lancet Respiratory Medicine 8(7), 709- 716 (2020) + +[7] Wedge, D.C., Gundem, G., Mitchell, T., Woodcock, D.J., Martincorena, I., Ghori, M., Zamora, J., Butler, A., Whitaker, H., Kote- Jarai, Z., et al.: Sequencing of prostate cancers identifies new cancer genes, routes of progression and drug targets. Nature genetics 50(5), 682- 692 (2018) + +[8] Mathys, H., Davila- Velderrain, J., Peng, Z., Gao, F., Mohammadi, S., Young, J.Z., Menon, M., He, L., Abdurrob, F., Jiang, X., et al.: Single- cell transcriptomic analysis of alzheimer's disease. Nature 570(7761), 332- 337 (2019) + +[9] Pal, B., Chen, Y., Vaillant, F., Capaldo, B.D., Joyce, R., Song, X., Bryant, V.L., Penington, J.S., Di Stefano, L., Tubau Ribera, N., et al.: A single- cell rna expression atlas of normal, preneoplastic and tumorigenic states in the human breast. The EMBO journal 40(11), 107333 (2021) + +[10] Depuydt, M.A., Prange, K.H., Slenders, L., Ord, T., Elbersen, D., Boltjes, A., De Jager, S.C., Asselbergs, F.W., De Borst, G.J., Aavik, E., et al.: Microanatomy of the human atherosclerotic plaque by single- cell transcriptomics. Circulation research 127(11), 1437- 1455 (2020) + +[11] Wang, Z., Gerstein, M., Snyder, M.: Rna- seq: a revolutionary tool for transcriptomics. Nature reviews genetics 10(1), 57- 63 (2009) + +[12] Newman, A.M., Steen, C.B., Liu, C.L., Gentles, A.J., Chaudhuri, A.A., Scherer, F., Khodadoust, M.S., Esfahani, M.S., Luca, B.A., Steiner, D., et al.: Determining cell type abundance and expression from bulk tissues with digital cytometry. Nature biotechnology 37(7), 773- 782 (2019) + +[13] Jew, B., Alvarez, M., Rahmani, E., Miao, Z., Ko, A., Garske, K.M., Sul, J.H., Pietilainen, K.H., Pajukanta, P., Halperin, E.: Accurate estimation of cell composition in bulk expression through robust integration of single- cell information. Nature communications 11(1), 1971 (2020) + +[14] Nel, S., Durandt, C., Murdoch, C., Pepper, M.S.: Determinants of dental pulp stem cell heterogeneity. Journal of Endodontics 48(10), 1232- 1240 (2022) + +[15] Papalexi, E., Satija, R.: Single- cell rna sequencing to explore immune cell heterogeneity. Nature Reviews Immunology 18(1), 35- 45 (2018) + +[16] Oliveira, G., Wu, C.J.: Dynamics and specificities of t cells in cancer immunotherapy. Nature Reviews Cancer, 1- 22 (2023) + +[17] Rahimi, A.M., Cai, M., Hoyer- Fender, S.: Heterogeneity of the nih3t3 fibroblast cell line. Cells 11(17), 2677 (2022) + +<--- Page Split ---> + +[18] Liu, Y., Mi, Y., Mueller, T., Kreibich, S., Williams, E.G., Van Drogen, A., Borel, C., Frank, M., Germain, P.- L., Bludau, I., et al.: Multi- omic measurements of heterogeneity in hela cells across laboratories. Nature biotechnology 37(3), 314- 322 (2019) + +[19] Trapnell, C., Cacchiarelli, D., Grimsby, J., Pokharel, P., Li, S., Morse, M., Lennon, N.J., Livak, K.J., Mikkelsen, T.S., Rinn, J.L.: The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nature biotechnology 32(4), 381- 386 (2014) + +[20] Qiu, X., Mao, Q., Tang, Y., Wang, L., Chawla, R., Pliner, H.A., Trapnell, C.: Reversed graph embedding resolves complex single- cell trajectories. Nature methods 14(10), 979- 982 (2017) + +[21] Song, Q., Wang, J., Bar- Joseph, Z.: sctstem: clustering pseudotime ordered single- cell data. Genome Biology 23(1), 1- 17 (2022) + +[22] Ding, J., Lin, C., Bar- Joseph, Z.: Cell lineage inference from snp and scrna- seq data. Nucleic Acids Research 47(10), 56- 56 (2019) + +[23] Ma, Q., Xu, D.: Deep learning shapes single- cell data analysis. Nature Reviews Molecular Cell Biology 23(5), 303- 304 (2022) + +[24] Amodio, M., Van Dijk, D., Srinivasan, K., Chen, W.S., Mohsen, H., Moon, K.R., Campbell, A., Zhao, Y., Wang, X., Venkataswamy, M., et al.: Exploring single- cell data with deep multitasking neural networks. Nature methods 16(11), 1139- 1145 (2019) + +[25] Lin, X., Tian, T., Wei, Z., Hakonarson, H.: Clustering of single- cell multi- omics data with a multimodal deep learning method. Nature communications 13(1), 7705 (2022) + +[26] Wang, J., Ma, A., Chang, Y., Gong, J., Jiang, Y., Qi, R., Wang, C., Fu, H., Ma, Q., Xu, D.: scgnn is a novel graph neural network framework for single- cell rna- seq analyses. Nature communications 12(1), 1882 (2021) + +[27] Hu, J., Li, X., Coleman, K., Schroeder, A., Ma, N., Irwin, D.J., Lee, E.B., Shinohara, R.T., Li, M.: Spagcn: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nature methods 18(11), 1342- 1351 (2021) + +[28] Settles, B.: Active learning literature survey (2009) + +[29] LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. nature 521(7553), 436- 444 (2015) + +[30] Kingma, D.P., Welling, M.: Auto- encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013) + +[31] Creswell, A., White, T., Dumoulin, V., Arulkumaran, K., Sengupta, B., Bharath, A.A.: Generative adversarial networks: An overview. IEEE signal processing magazine 35(1), 53- 65 (2018) + +[32] Bengio, Y., Courville, A., Vincent, P.: Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence 35(8), 1798- 1828 (2013) + +[33] Bengio, Y., Courville, A.C., Vincent, P.: Unsupervised feature learning and deep learning: A review and new perspectives. CoRR, abs/1206.5538 1(2665), 2012 (2012) + +[34] Larsen, A.B.L., Sonderby, S.K., Larochelle, H., Winther, O.: Autoencoding beyond pixels using a learned similarity metric. In: International Conference on Machine Learning, pp. 1558- 1566 (2016). PMLR + +[35] Stephenson, E., Reynolds, G., Botting, R.A., Calero- Nieto, F.J., Morgan, M.D., Tuong, Z.K., Bach, K., Sungnak, W., Worlock, K.B., Yoshida, M., et al.: Single- cell multi- omics analysis of the immune response in covid- 19. Nature medicine 27(5), 904- 916 (2021) + +<--- Page Split ---> + +[36] McInnes, L., Healy, J., Melville, J.: Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426 (2018) + +[37] Hadjadj, J., Yatim, N., Barnabei, L., Corneau, A., Boussier, J., Smith, N., Peré, H., Charbit, B., Bondet, V., Chenevier- Gobeaux, C., et al.: Impaired type i interferon activity and inflammatory responses in severe covid- 19 patients. Science 369(6504), 718- 724 (2020) + +[38] Ramasamy, S., Subbian, S.: Critical determinants of cytokine storm and type i interferon response in covid- 19 pathogenesis. Clinical microbiology reviews 34(3), 10- 1128 (2021) + +[39] Lee, J.S., Park, S., Jeong, H.W., Ahn, J.Y., Choi, S.J., Lee, H., Choi, B., Nam, S.K., Sa, M., Kwon, J.- S., et al.: Immunophenotyping of covid- 19 and influenza highlights the role of type i interferons in development of severe covid- 19. Science immunology 5(49), 1554 (2020) + +[40] Lee, J.S., Shin, E.- C.: The type i interferon response in covid- 19: implications for treatment. Nature Reviews Immunology 20(10), 585- 586 (2020) + +[41] Pearson, K.: Liii. on lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin philosophical magazine and journal of science 2(11), 559- 572 (1901) + +[42] Menden, K., Marouf, M., Oller, S., Dalmia, A., Magruder, D.S., Kloiber, K., Heutink, P., Bonn, S.: Deep learning- based cell composition analysis from tissue expression profiles. Science advances 6(30), 2619 (2020) + +[43] Chen, Y., Wang, Y., Chen, Y., Cheng, Y., Wei, Y., Li, Y., Wang, J., Wei, Y., Chan, T.- F., Li, Y.: Deep autoencoder for interpretable tissue- adaptive deconvolution and cell- type- specific gene analysis. Nature Communications 13(1), 6735 (2022) + +[44] Tsoucas, D., Dong, R., Chen, H., Zhu, Q., Guo, G., Yuan, G.- C.: Accurate estimation of cell- type composition from gene expression data. Nature communications 10(1), 2975 (2019) + +[45] Plaisier, S.B., Taschereau, R., Wong, J.A., Graeber, T.G.: Rank- rank hypergeometric overlap: identification of statistically significant overlap between gene- expression signatures. Nucleic acids research 38(17), 169- 169 (2010) + +[46] Thomas, P.D., Ebert, D., Muruganujan, A., Mushayahama, T., Albou, L.- P., Mi, H.: Panther: Making genome- scale phylogenetics accessible to all. Protein Science 31(1), 8- 22 (2022) + +[47] Carbon, S., Ireland, A., Mungall, C.J., Shu, S., Marshall, B., Lewis, S., Hub, A., Group, W.P.W.: Amigo: online access to ontology and annotation data. Bioinformatics 25(2), 288- 289 (2009) + +[48] Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., et al.: Gene set enrichment analysis: a knowledge- based approach for interpreting genome- wide expression profiles. Proceedings of the National Academy of Sciences 102(43), 15545- 15550 (2005) + +[49] Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al.: Gene ontology: tool for the unification of biology. Nature genetics 25(1), 25- 29 (2000) + +[50] Gillespie, M., Jassal, B., Stephan, R., Milacic, M., Rothfels, K., Senff- Ribeiro, A., Griss, J., Sevilla, C., Matthews, L., Gong, C., et al.: The reactome pathway knowledgebase 2022. Nucleic acids research 50(D1), 687- 692 (2022) + +[51] Nishida- Aoki, N., Gujral, T.S.: Emerging approaches to study cell- cell interactions in tumor microenvironment. Oncotarget 10(7), 785 (2019) + +[52] Armingol, E., Officer, A., Harismendy, O., Lewis, N.E.: Deciphering cell- cell interactions and communication from gene expression. Nature Reviews Genetics 22(2), 71- 88 (2021) + +[53] Joanito, I., Wirapati, P., Zhao, N., Nawaz, Z., Yeo, G., Lee, F., Eng, C.L., Macalinao, D.C., + +<--- Page Split ---> + +Kahraman, M., Srinivasan, H., et al.: Single- cell and bulk transcriptome sequencing identifies two epithelial tumor cell states and refines the consensus molecular classification of colorectal cancer. Nature genetics 54(7), 963- 975 (2022) + +[54] Gao, J., Wu, Z., Zhao, M., Zhang, R., Li, M., Sun, D., Cheng, H., Qi, X., Shen, Y., Xu, Q., et al.: Allosteric inhibition reveals shp2- mediated tumor immunosuppression in colon cancer by single- cell transcriptomics. Acta Pharmaceutica Sinica B 12(1), 149- 166 (2022) + +[55] Cao, Y., Tang, W., Tang, W.: Immune cell infiltration characteristics and related core genes in lupus nephritis: results from bioinformatic analysis. BMC immunology 20, 1- 12 (2019) + +[56] Fortier, A.- M., Asselin, E., Cadrin, M.: Keratin 8 and 18 loss in epithelial cancer cells increases collective cell migration and cisplatin sensitivity through claudin1 up- regulation. Journal of Biological Chemistry 288(16), 11555- 11571 (2013) + +[57] Karantza, V.: Keratins in health and cancer: more than mere epithelial cell markers. Oncogene 30(2), 127- 138 (2011) + +[58] Mann, H.B., Whitney, D.R.: On a test of whether one of two random variables is stochastically larger than the other. The annals of mathematical statistics, 50- 60 (1947) + +[59] Virtanen, P., Gommers, R., Oliphant, T.E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., et al.: Scipy 1.0: fundamental algorithms for scientific computing in python. Nature methods 17(3), 261- 272 (2020) + +[60] Ramaswami, G., Yuva- Aydemir, Y., Akerberg, B., Matthews, B., Williams, J., Golczer, G., Huang, J., Huh, D., Burkly, L.C., Engle, S.J., et al.: Transcriptional characterization of ipsc- derived microglia as a model for therapeutic development in neurodegeneration. bioRxiv, 2023- 03 (2023) + +[61] Lahnemann, D., Köster, J., Szczurek, E., McCarthy, D.J., Hicks, S.C., Robinson, M.D., Vallejos, C.A., Campbell, K.R., Beerenwinkel, N., Mahfouz, A., et al.: Eleven grand challenges in single- cell data science. Genome biology 21(1), 1- 35 (2020) + +[62] Pratt, J.W.: Remarks on zeros and ties in the wilcoxon signed rank procedures. Journal of the American Statistical Association 54(287), 655- 667 (1959) + +[63] Wolf, F.A., Hamey, F.K., Plass, M., Solana, J., Dahlin, J.S., Göttgens, B., Rajewsky, N., Simon, L., Theis, F.J.: Paga: graph abstraction reconciles clustering with trajectory inference through a topology preserving map of single cells. Genome biology 20, 1- 9 (2019) + +[64] Nguyen, V.- L., Shaker, M.H., Hüllermeier, E.: How to measure uncertainty in uncertainty sampling for active learning. Machine Learning 111(1), 89- 122 (2022) + +[65] Lewis, D.D.: A sequential algorithm for training text classifiers: Corrigendum and additional data. In: Acm Sigir Forum, vol. 29, pp. 13- 19 (1995). ACM New York, NY, USA + +[66] Dasgupta, S.: Two faces of active learning. Theoretical computer science 412(19), 1767- 1781 (2011) + +[67] Tipping, M.E., Bishop, C.M.: Probabilistic principal component analysis. Journal of the Royal Statistical Society Series B: Statistical Methodology 61(3), 611- 622 (1999) + +[68] Halko, N., Martinsson, P.- G., Tropp, J.A.: Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions. SIAM review 53(2), 217- 288 (2011) + +[69] Martinsson, P.- G., Rokhlin, V., Tygert, M.: A randomized algorithm for the decomposition of matrices. Applied and Computational Harmonic Analysis 30(1), 47- 68 (2011) + +[70] Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., + +<--- Page Split ---> + +Prettenhofer, P., Weiss, R., Dubourg, V., et al.: Scikit- learn: Machine learning in python. the Journal of machine Learning research 12, 2825- 2830 (2011) + +[71] MacQueen, J., et al.: Some methods for classification and analysis of multivariate observations. In: Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, vol. 1, pp. 281- 297 (1967). Oakland, CA, USA + +[72] Finak, G., McDavid, A., Yajima, M., Deng, J., Gersuk, V., Shalek, A.K., Slichter, C.K., Miller, H.W., McElrath, M.J., Prlic, M., et al.: Mast: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single- cell rna sequencing data. Genome biology 16(1), 1- 13 (2015) + +[73] Shalek, A.K., Satija, R., Shuga, J., Trombetta, J.J., Gennert, D., Lu, D., Chen, P., Gertner, R.S., Gaublomme, J.T., Yosef, N., et al.: Single- cell rna- seq reveals dynamic paracrine control of cellular variation. Nature 510(7505), 363- 369 (2014) + +[74] Hasanaj, E., Wang, J., Sarathi, A., Ding, J., Bar- Joseph, Z.: Interactive single- cell data analysis using cellar. Nature communications 13(1), 1998 (2022) + +[75] Wolf, F.A., Angerer, P., Theis, F.J.: Scanpy: large- scale single- cell gene expression data analysis. Genome biology 19, 1- 5 (2018) + +[76] Ernst, J., Vainas, O., Harbison, C.T., Simon, I., Bar- Joseph, Z.: Reconstructing dynamic regulatory maps. Molecular systems biology 3(1), 74 (2007) + +[77] Li, D., Ding, J., Bar- Joseph, Z.: Unifan: A tool for unsupervised single- cell clustering and annotation. Journal of Computational Biology 29(11), 1229- 1232 (2022) + +[78] Lopez, R., Regier, J., Cole, M.B., Jordan, M.I., Yosef, N.: Deep generative modeling for single- cell transcriptomics. Nature methods 15(12), 1053- 1058 (2018) + +[79] Risso, D., Perraudeau, F., Gribkova, S., Dudoit, S., Vert, J.- P.: A general and flexible method for signal extraction from single- cell rna- seq data. Nature communications 9(1), 284 (2018) + +[80] Berge, K., Perraudeau, F., Soneson, C., Love, M.I., Risso, D., Vert, J.- P., Robinson, M.D., Dudoit, S., Clement, L.: Observation weights unlock bulk rna- seq tools for zero inflation and single- cell applications. Genome biology 19, 1- 17 (2018) + +[81] Kullback, S., Leibler, R.A.: On information and sufficiency. The annals of mathematical statistics 22(1), 79- 86 (1951) + +[82] Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., Philip, S.Y.: A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems 32(1), 4- 24 (2020) + +[83] Scarselli, F., Gori, M., Tsoi, A.C., Hagenbuchner, M., Monfardini, G.: The graph neural network model. IEEE transactions on neural networks 20(1), 61- 80 (2008) + +[84] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high- performance deep learning library. Advances in neural information processing systems 32 (2019) + +[85] Maas, A.L., Hannun, A.Y., Ng, A.Y., et al.: Rectifier nonlinearities improve neural network acoustic models. In: Proc. Icml, vol. 30, p. 3 (2013). Atlanta, GA + +[86] He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: Surpassing human- level performance on imagenet classification. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 1026- 1034 (2015) + +[87] Kleyman, M., Sefer, E., Nicola, T., Espinoza, C., Chhabra, D., Hagood, J.S., Kaminski, N., Ambalavanan, N., Bar- Joseph, Z.: Selecting the most appropriate time points to profile in high- throughput studies. Elife 6, 18541 (2017) + +<--- Page Split ---> + +[88] Fang, Z., Liu, X., Peltz, G.: Gseapy: a comprehensive package for performing gene set enrichment analysis in python. Bioinformatics 39(1), 757 (2023) + +[89] Jin, S., Guerrero- Juarez, C., Zhang, L., Chang, I., Ramos, R., Kuan, C., Myung, P., Pikus, M., Nie, Q.: Inference and analysis of cell- cell communication using CellChat. Nat. Commun. 12, 1088 (2021) + +[90] Cao, J., Spielmann, M., Qiu, X., Huang, X., Ibrahim, D.M., Hill, A.J., Zhang, F., Mundlos, S., Christiansen, L., Steemers, F.J., et al.: The single- cell transcriptional landscape of mammalian organogenesis. Nature 566(7745), 496- 502 (2019) + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +scSemiProfilersupplementary.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c_det.mmd b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b377e4a54857a53de41cda21f71d73811f3141ba --- /dev/null +++ b/preprint/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c/preprint__12ebfde9a14e45b09d9b752afc6b18e3be9634c0cb1bdc7891cb5e1d153fc58c_det.mmd @@ -0,0 +1,895 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 950, 210]]<|/det|> +# scSemiProfiler: Advancing Large-scale Single-cell Studies through Semi-profiling with Deep Generative Models and Active Learning + +<|ref|>text<|/ref|><|det|>[[44, 230, 256, 276]]<|/det|> +Jun Ding jun.ding@mcgill.ca + +<|ref|>text<|/ref|><|det|>[[50, 302, 686, 323]]<|/det|> +McGill University Health Centre https://orcid.org/0000- 0001- 5183- 6885 + +<|ref|>text<|/ref|><|det|>[[44, 328, 328, 370]]<|/det|> +Jingtao Wang McGill University Health Centre + +<|ref|>text<|/ref|><|det|>[[44, 375, 328, 416]]<|/det|> +Gregory Fonseca McGill University Health Centre + +<|ref|>sub_title<|/ref|><|det|>[[44, 456, 103, 473]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 493, 933, 536]]<|/det|> +Keywords: Deep Generative Learning, Active Learning, Bulk Deconvolution, Single- cell sequencing, Bulk sequencing, Cohort studies, scSemiProfiler + +<|ref|>text<|/ref|><|det|>[[44, 554, 338, 573]]<|/det|> +Posted Date: December 5th, 2023 + +<|ref|>text<|/ref|><|det|>[[42, 592, 475, 611]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3638900/v1 + +<|ref|>text<|/ref|><|det|>[[42, 630, 914, 673]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 690, 535, 710]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 745, 911, 789]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 16th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 50150- 1. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[181, 153, 848, 230]]<|/det|> +# scSemiProfiler: Advancing Large-scale Single-cell Studies through Semi-profiling with Deep Generative Models and Active Learning + +<|ref|>text<|/ref|><|det|>[[268, 252, 757, 270]]<|/det|> +Jingtao Wang \(^{1,2}\) , Gregory Fonseca \(^{1,2,4}\) , Jun Ding \(^{1,2,3,4,5*}\) + +<|ref|>text<|/ref|><|det|>[[166, 278, 857, 440]]<|/det|> +\(^{1}\) Meakins- Christe Laboratories, Research Institute of McGill University Health Centre, 1001 Decarie Blvd, Montreal, H4A 3J1, Quebec, Canada. \(^{2}\) Department of Medicine, Division of Experimental Medicine, McGill University, 1001 Decarie Blvd, Montreal, H4A 3J1, Quebec, Canada. \(^{3}\) School of Computer Science, McGill University, 3480 Rue University, Montreal, H3A 2A7, Quebec, Canada. \(^{4}\) Quantitative Life Sciences, McGill University, 845 Rue Sherbrooke Ouest, Montreal, H3A 0G4, Quebec, Canada. \(^{5}\) Mila- Quebec AI Institute, 6666 Rue Saint- Urbain, Montreal, H2S 3H1, Quebec, Canada. + +<|ref|>text<|/ref|><|det|>[[196, 467, 830, 500]]<|/det|> +\*Corresponding author(s). E- mail(s): jun.ding@mcgill.ca; Contributing authors: jingtao.wang@mail.mcgill.ca; gregory.fonseca@mcgill.ca; + +<|ref|>sub_title<|/ref|><|det|>[[478, 527, 546, 540]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[187, 543, 839, 715]]<|/det|> +Single- cell sequencing is a crucial tool for dissecting the cellular intricacies of complex diseases. Its prohibitive cost, however, hampers its application in expansive biomedical studies. Traditional cellular deconvolution approaches can infer cell type proportions from more affordable bulk sequencing data, yet they fall short in providing the detailed resolution required for single- cell- level analyses. To overcome this challenge, we introduce "scSemiProfiler", an innovative computational framework that marries deep generative model with active learning strategies. This method adeptly infers single- cell profiles across large cohorts by fusing bulk sequencing data with targeted single- cell sequencing from a few carefully chosen representatives. Extensive validation across heterogeneous datasets verifies the precision of our semi- profiling approach, aligning closely with true single- cell profiling data and empowering refined cellular analyses. Originally developed for extensive disease cohorts, "scSemiProfiler" is adaptable for broad applications. It provides a scalable, cost- effective solution for single- cell profiling, facilitating in- depth cellular investigation in various biological domains. + +<|ref|>text<|/ref|><|det|>[[186, 724, 818, 749]]<|/det|> +Keywords: Deep Generative Learning, Active Learning, Bulk Deconvolution, Single- cell sequencing, Bulk sequencing, Cohort studies, scSemiProfiler + +<|ref|>sub_title<|/ref|><|det|>[[147, 792, 297, 811]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[147, 820, 880, 907]]<|/det|> +The advent of single- cell sequencing has dramatically reshaped the landscape of biological research, providing an unparalleled view into the cellular complexities of organisms [1, 2]. This technique has unearthed the subtle distinctions among individual cells, enabling a richer understanding of cellular dynamics [3, 4]. High- resolution data from such analyses are essential for delineating and characterizing the myriad of cellular subpopulations within patient samples, leading to transformative developments in biomarker discovery and personalized therapy strategies [5- 7]. Cohort studies, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 141]]<|/det|> +which offer longitudinal insights by observing specific groups over time, are particularly poised to benefit from these advances [8- 10]. However, the substantial financial cost associated with single- cell sequencing—such as the estimated \(\) 6,000\$ required to sequence 20,000 cells (costpercell)—can often be a limiting factor for extensive research endeavors. + +<|ref|>text<|/ref|><|det|>[[118, 142, 850, 384]]<|/det|> +A range of computational strategies, particularly deconvolution methods, are available for dissecting bulk data [11] into distinct cell populations. This approach enables a harmonious balance between affordability and data resolution. Prominent among these are deconvolution techniques such as CIBERSORTx [12] and Bisque [13], which have become particularly popular. These methods estimate the proportions of different cell types within bulk sequencing samples by utilizing signature profiles from single- cell reference datasets. Conventional bulk decomposition methods, while valuable for enhancing cohort study analyses, exhibit limitations in their resolution and accuracy. Capable of dissecting bulk samples into different cell types and ascertaining their gene expression patterns, they offer beneficial insights at the cell- type level. Yet, these methods fall short in delivering true single- cell resolution. Another significant challenge they face is the precise decomposition of cell types and accurate inference of gene expression. Additionally, these methods often overlook the considerable variability that exists within individual cell types. This variability is a critical component for deciphering the complexities of disease dynamics and their response to treatments [14- 18]. Those limitations of conventional bulk decomposition methods may impede the thorough exploration of in- depth single- cell level analyses. These analyses are essential for understanding cellular heterogeneity and dynamics in datasets, including but not limited to cell pseudotime trajectory analysis [19- 22], as well as various advanced machine learning techniques [23- 27]. + +<|ref|>text<|/ref|><|det|>[[118, 384, 850, 640]]<|/det|> +In response to the challenges previously highlighted and with the goal of offering a cost- effective approach for extensive single- cell sequencing, we introduce the single- cell Semi- profiler (scSemiProfiler). This innovative computational tool is crafted to significantly improve the precision and depth of single- cell analysis. It stands out as a more economical and scalable option for single- cell sequencing, thereby facilitating advanced single- cell analysis with greater accessibility. This tool effectively integrates active learning techniques [28] with deep generative neural network algorithms [29], aiming to provide single- cell resolution data at a more affordable price. scSemiProfiler aims to simultaneously achieve two fundamental goals in the semi- profiling process. On one hand, scSemiProfiler's active learning module integrates information from the deep learning model and bulk data, intelligently selecting the most informative samples for actual single- cell sequencing. On the other hand, scSemiProfiler's deep generative model component [30- 33] effectively merges single- cell data from representative samples with the bulk sequencing data of the cohort. This process computationally infers the single- cell data for the remaining non- representative samples. This advanced approach leads to more detailed "deconvolution" of the target bulk data into precise single- cell level measurements. Consequently, with only the budget for bulk sequencing and single- cell sequencing of representatives, scSemiProfiler outputs single- cell data for all samples in the study. To the best of our knowledge, the scSemiProfiler is the first of its kind designed for such intricate single- cell level computational decomposition from bulk sequencing data. + +<|ref|>text<|/ref|><|det|>[[118, 640, 850, 769]]<|/det|> +Through comprehensive evaluations across a variety of datasets, scSemiProfiler has consistently delivered high- quality semi- profiled single- cell data, which not only accurately represents actual single- cell datasets but also mirrors the results of downstream tasks with remarkable fidelity. Therefore, scSemiProfiler has established itself as a pivotal tool in single- cell research. This innovative approach is poised to revolutionize access to single- cell data in large- scale studies, such as disease cohort investigations and beyond. By making large- scale single- cell studies more affordable, scSemiProfiler promises to catalyze the application of single- cell technologies in a wide array of expansive biomedical research. This advancement is set to expand the scope and enhance the depth of biological research globally. + +<|ref|>sub_title<|/ref|><|det|>[[118, 784, 207, 802]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[118, 813, 291, 829]]<|/det|> +## Method overview + +<|ref|>text<|/ref|><|det|>[[118, 836, 850, 907]]<|/det|> +The scSemiProfiler approach represents a novel method for decomposing broad- scope bulk sequencing data into detailed single- cell cohorts. It achieves this by conducting single- cell profiling on only a select few representative samples and then computationally inferring single- cell data for the remainder. This approach substantially lowers the costs associated with large- scale single- cell studies. As depicted in Fig. 1, our method is designed to deliver cost- effective, semi- profiled single- cell sequencing data, + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[145, 80, 880, 600]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 614, 880, 833]]<|/det|> +
Fig. 1 Overview of the scSemiProfiler method. a, Initial Configuration: Bulk sequencing is first conducted on the entire cohort, followed by clustering analysis of this data. This analysis identifies representative samples, typically those closest to the cluster centroids. b, Representative Profiling: The identified representatives are then subjected to single-cell sequencing. The data obtained from this sequencing is further processed to determine gene set scores and feature importance weights, enriching the subsequent analysis steps. c, Deep Generative Inference: Utilizing a VAE-GAN-based model, the process integrates comprehensive bulk data from the cohort with the single-cell data derived from the representatives. During the model's 3-stage training, the generator aims to optimize losses \(L_{G_{Pretrain1}}\) , \(L_{G_{Pretrain2}}\) , and \(L_{inference}\) , respectively, whereas the discriminator focuses on minimizing the discriminator loss \(L_{D}\) . In \(L_{D}\) , \(G\) and \(D\) are the generator and discriminator respectively. The term \(D((\mathbf{x}_1, \mathbf{s}_1))\) represents the discriminator's predicted probability that a given input cell is real, under the condition that it is indeed a real cell. Conversely, \(D(G((\mathbf{x}_1, \mathbf{s}_1)))\) denotes the discriminator's predicted probability that the input cell is real, when in fact it is a cell reconstructed by the generator. d, Representative Selection Decision: Decisions on further representative selection are made, taking into account budget constraints and the effectiveness of the current representatives. An active learning algorithm, which draws on insights from the bulk data and the generative models, is employed to pinpoint additional optimal representatives. These newly selected representatives then undergo further single-cell sequencing (b) and serve as new reference points for the ongoing in silico inference process (c). e, Comprehensive Downstream Analyses: The final panel shows extensive downstream analyses enabled by the semi-profiled single-cell data. This is pivotal in demonstrating the model's capacity to provide deep and wide-ranging insights, showcasing the full potential and applicability of the semi-profiled single-cell data.
+ +<|ref|>text<|/ref|><|det|>[[147, 850, 879, 907]]<|/det|> +enabling deep exploration of cellular dynamics in large cohorts. In this context, "semi- profiling" refers to the generation of single- cell data for an entire cohort, achieved either through direct single- cell sequencing of selected representative samples or via in silico inference using a deep generative model. This in silico inference process combines actual single- cell sequencing data from a representative + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 85, 850, 156]]<|/det|> +sample with bulk sequencing data, encompassing both the target and the representative sample. Thus, a "semi- profiled cohort" includes real single- cell data for representative samples and inferred data for the non- representative ones. This innovative approach facilitates a thorough examination of individual cellular profiles within a larger dataset, seamlessly linking the extensive scope of bulk sequencing with the granularity of single- cell analysis. + +<|ref|>text<|/ref|><|det|>[[118, 157, 850, 512]]<|/det|> +Initially, the semi- profiling pipeline commences with bulk sequencing of each cohort member (Fig. 1a), laying the foundational data layer for all subsequent analyses. Following this foundational step, the methodology employs a clustering analysis to form \(B\) (sample batch size) sample clusters utilizing the extensive data derived from the initial bulk sequencing and selecting a "representative" sample for each cluster (Fig. 1b). Single- cell profiling will be conducted on the selected representative samples in preparation for the following steps. The core of the scSemiProfiler involves an innovative deep generative learning model (Fig. 1c). This model is engineered to intricately meld actual single- cell data profiles with the gathered bulk sequencing data, thereby capturing complex biological patterns and nuances. Specifically, it uses a VAE- GAN [34] architecture initially pretrained on single- cell sequencing data of selected representatives for self- reconstruction. Subsequently, the VAE- GAN is further pretrained with a representative reconstruction bulk loss, aligning pseudobulk estimations from the reconstructed single- cell data with real pseudobulk. Finally, the model undergoes fine- tuning with another target bulk loss tied to the real bulk sequencing data of the target sample, facilitating precise in silico inference of the target's single- cell profile. Once the in silico single- cell inference is finished for all non- representative samples in the cohort, an active learning module can be used for selecting the next batch of potentially most informative representatives for single- cell sequencing to further improve the semi- profiling performance (Fig. 1d). When studying a smaller dataset, or when more cells per sample are required, a smaller batch size, such as 2, may be preferred. However, a batch size of 4 is set as default to maximize the usage of a 10x genomics single- cell toolkit, which can typically capture up to 20000 cells (4 samples if assuming 5,000 cells each). This dynamic, iterative process is continuously augmented with newly acquired single- cell data, ensuring that the most informative samples are selected for real single- cell profiling, leading to more accurate in silico single- cell inference for non- representative samples. This iterative process concludes when the budgetary constraint is met or when a sufficient number of representatives have been chosen to ensure satisfactory semi- profiling performance. + +<|ref|>text<|/ref|><|det|>[[119, 512, 850, 612]]<|/det|> +When any of the stop criteria is met, the semi- profiled single- cell data can be used for a broad spectrum of downstream single- cell analyses (Fig. 1e), such as cell feature visualization, biomarker, and function enrichment analysis, tracking cell type compositions in various tissues/conditions, cell- cell interaction analysis, and pseudotime analysis. Ultimately, scSemiProfiler offers a holistic sequencing perspective, delivering nuanced single- cell insights from bulk sequencing data. The method exhibits acceptable performance in terms of runtime and memory usage (refer to Supplementary Fig. S1), making it suitable for scaling to large- scale datasets. + +<|ref|>sub_title<|/ref|><|det|>[[119, 626, 771, 660]]<|/det|> +## The semi-profiled COVID-19 single-cell cohort exhibits significant similarity to its real counterpart + +<|ref|>text<|/ref|><|det|>[[118, 666, 850, 909]]<|/det|> +To test the performance of scSemiProfiler in generating semi- profiled single- cell data that resonates with the granularity and details of actual single- cell sequencing, we utilized a COVID- 19 cohort single- cell sequencing dataset [35]. After quality controls, this dataset includes 124 samples, including healthy controls and infected patients of different severity levels: asymptomatic, mild, moderate, severe, or critical. Here, we produced pseudobulk data by taking the average of the normalized count single- cell data. We then tested scSemiProfiler's ability to regenerate the single- cell cohort from the pseudo- bulk data and real- profiled single- cell representatives using semi- profiling, as the actual bulk data for those samples in the COVID- 19 cohort is absent. In generating our semi- profiled dataset, 28 representatives were selected in batches of 4 using our active learning algorithm for real single- cell profiling. Deep generative models then inferred the single- cell profiles for the remaining samples based on the representatives' real- profiled single- cell data and the bulk data of all samples. The estimated total cost of these bulk and single- cell sequencing is \(62,640. This is only 33.7\%\) of the estimated price, \(\) 186,000\(, for actually conducting single- cell sequencing for the entire cohort. Additionally, our approach offers the advantage of generating extra bulk data for the cohort, a benefit not provided by single- cell sequencing of the entire cohort. The price of bulk sequencing is estimated based on the cost at McGill Genome Centre in the year 2023. This estimation assumes the use of one NovaSeq 6000 S2 system (capable of sequencing up to 4 billion reads per run) at an approximate cost of\) \ \(7,000\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[147, 80, 870, 565]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 587, 880, 826]]<|/det|> +
Fig. 2 Overall comparisons of the semi-profiled and real-profiled COVID-19 dataset. a, UMAP visualization of the real-profiled data. Colors correspond to cell types and are consistent with (g). b, UMAP visualization of the semi-profiled data. c, UMAP visualization of semi-profiled data and real-profiled data together. The color differentiation signifies whether cells originate from the semi-profiled or the real-profiled dataset. Areas of overlap between the two indicate where the semi-profiled data closely resembles the real-profiled data. d, UMAP visualization of the semi-profiled cohort, displaying different colors to distinguish cells produced by a deep generative model (labeled as "Generated") from the representative cells obtained through real-profiling (labeled as "Representatives"). e, Visualization illustrates the relative activation patterns of the interferon pathway. The comparison of these values between the semi-profiled and real-profiled matrices yields a Pearson correlation coefficient of 0.849 and a p-value of \(3.63 \times 10^{-26}\). f, Graph depicting the normalized error in semi-profiled data with an increasing number of representatives. The terms 'scSemiProfiler' and 'Selection-only' represent our semi-profiling method and a method that only selects representatives using an active learning algorithm, respectively. It is important to note that actual costs may vary based on the sequencing technology and specific cells sequenced. g, Stacked bar plot illustrating the proportions of cell types across various disease conditions. The upper portion represents the real-profiled data, while the lower portion depicts the semi-profiled data. Pearson correlation coefficients comparing cell type proportions between the real-profiled and semi-profiled datasets are provided for different conditions: Healthy (0.987), Asymptomatic (0.970), Mild (0.996), Moderate (0.992), Severe (0.978), and Critical (0.989), indicating a high degree of similarity between the two datasets across these conditions. h-i, Cell type deconvolution benchmarking. h, Figure displaying Pearson correlation coefficients between actual (ground truth) cell type proportions and those estimated by various deconvolution methods. Except for the first two columns, all other columns' results are based on 4 representatives' single-cell data as reference. i, Comparison of Root Mean Square Error (RMSE) across various deconvolution methods.
+ +<|ref|>text<|/ref|><|det|>[[147, 844, 878, 900]]<|/det|> +plus an additional \(\) 110\(for library preparation per bulk sample. The cost for single - cell sequencing is based on the tool (costpercell). This tool provides a cost estimate for capturing 0.8 billion total reads from 20,000 cells across four samples in one 10x lane, equating to\) \ \(0.3\) per cell. Consequently, the estimated cost for each sample (5,000 cells) is \(\) 1,500\$ . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 85, 850, 411]]<|/det|> +Using UMAP [36] visualizations, we show a significant alignment between the semi- profiled and the real- profiled in Fig. 2a and b. We also annotated the semi- profiled COVID- 19 cohort using an unbiased approach. We trained a Multi- layer Perceptron (MLP) classifier using the annotated representatives' cells and used it to predict the cell types of the rest of the non- representative samples' cells generated by the deep learning model. As shown in Fig. 2b, cell clustering remains intact in the semi- profiled cohort, from the distinctive clusters of B cells, plasmoblasts, and platelets to the nuanced similarities of the CD14 and CD16 cells. The semi- profiled dataset illustrates its remarkable fidelity to the real- profiled version. This fidelity extends to capturing even the subtle batch effects, as observed in the twin CD4 clusters, further accentuating scSemiProfiler's robust in silico inference capabilities. The UMAP plots in Fig. 2c weave together directly sequenced samples with the semi- profiled ones, showcasing the tool's finesse. The overlapping data points found in both sequencing techniques resonate with the transformative nature of scSemiProfiler—it harmonizes accuracy and cost- efficiency seamlessly. This prompts an inquiry: Does this alignment owe its credit to the active learning mechanism that identifies the most informative representatives, or does it also hail from the deep generative model's prowess in inferring target samples' cells with finesse? To answer this, Fig.2d uses distinct colors to delineate real- profiled cells from representatives and generated cells for target samples. It shows that our representative selection strategy is able to select representatives such that their cells have a relatively good coverage of the overall cell distribution. Meanwhile, the deep learning model managed to generate cells to complement the rest and make the overall semi- profiled cohort almost identical to the original real- profiled cohort. The effectiveness of the deep learning model in the semi- profiling process is further demonstrated in Supplementary Fig. S2. This figure illustrates the model's capability in reconstructing data from single representative samples and its proficiency in inferring data for individual target samples. + +<|ref|>text<|/ref|><|det|>[[118, 412, 850, 512]]<|/det|> +To test the fidelity of single- cell gene expression semi- profiling, we tested the interferon (IFN) pathway gene set—crucial for the innate immune response against COVID- 19 [37–40]. Through the prism of our semi- profiled dataset, Fig. 2e reveals IFN activation patterns that harmonize with the real- profiled dataset. The uniformity in IFN activation patterns across various key cell types and severity levels, as highlighted by similar heatmaps, confirms the effectiveness of the semi- profiling technique. This uniformity indicates that the critical disease- related pathways were effectively captured and maintained in the semi- profiled data. + +<|ref|>text<|/ref|><|det|>[[118, 512, 850, 768]]<|/det|> +Further, we explored the quantitative metrics of efficacy and cost- effectiveness in semi- profiling, Fig. 2f. Our analysis centered on understanding the relationship between the number of representative samples used for single- cell sequencing and the associated semi- profiling error. While an increase in representative samples intuitively raises the costs of single- cell sequencing, scSemiProfiler effectively leverages the single- cell data of these representatives for accurate inference of target samples. As more representatives are selected, our semi- profiling method effectively reduced the normalized error. We also compared our method to a selection- only method, which uses the same representatives chosen by scSemiProfiler using the active learning algorithm. For each target sample, it performs the single- cell data inference in a naive manner: merely copying the corresponding representative's data. The dashed line shows that our semi- profiling method has a huge lead over the selection- only method. The star symbol denotes the number of representatives selected for the specific semi- profiled cohort that was utilized for subsequent analyses, along with the associated error. The vertical dashed line underscores that by using the same representatives, our method achieves substantially lower errors compared to the selection- only method. Moreover, the horizontal dashed line demonstrates that the selection- only method requires a considerably larger pool of representatives to attain the same level of semi- profiling accuracy as our method. The comparative analysis between scSemiProfiler and the selection- only method highlights the deep learning model's efficacy in reducing costs and minimizing errors. + +<|ref|>text<|/ref|><|det|>[[118, 769, 850, 883]]<|/det|> +In further evaluating the effectiveness of semi- profiling, defined as a single- cell granularity bulk decomposition, we turned to the conventional realm of cell type proportion metrics, as anchored by Fig. 2g. Existing research [35] elucidates that PBMC cell type proportions undergo dynamic shifts with evolving disease conditions. True to this, our real- profiled dataset indicates a pronounced expansion of B cells and CD14 cells under aggravated conditions—a pattern mirrored in the semi- profiled dataset. The Pearson correlation coefficient [41] of cell type composition associated with different disease conditions between the semi- profiled and the real single- cell datasets consistently surpasses 0.9. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 879, 327]]<|/det|> +In our comprehensive analysis of cell deconvolution methods, we meticulously compared scSemiProfiler with several leading- edge techniques, including CIBERSORTx [12], Bisque [13], Scaden [42], TAPE [43], and DWLS [44], as depicted in Fig. 2h and i. Each method was tested under identical conditions, using the same bulk data and single- cell reference data. Notably, DWLS was not included in our final results due to its failure to yield results within a week. A key challenge in this analysis was the memory constraints encountered by most methods, elaborated in Supplementary Fig. S1b, which hindered their ability to process the full set of 28 representative single- cell data. To maintain a level playing field, we limited the single- cell reference to an initial batch of 4 representative samples for all methods. Our results unequivocally demonstrate the superior deconvolution performance of scSemiProfiler over all benchmarked methods. Its effectiveness is not only apparent when utilizing a smaller reference set of 4 samples but also becomes increasingly pronounced with a larger set of 28 samples. This distinct superiority of scSemiProfiler is a testament to its remarkable efficiency and versatility. Capable of excelling with both compact and extensive single- cell reference datasets, scSemiProfiler stands out as the most adept tool in cell deconvolution, surpassing its contemporaries in handling diverse data scales with unparalleled precision and reliability. In Supplementary Fig. S3, we also show the high deconvolution accuracy using a side- by- side comparison between each sample's predicted cell type proportion using 28 representatives with the ground truth. + +<|ref|>sub_title<|/ref|><|det|>[[147, 341, 794, 375]]<|/det|> +## The semi-profiled COVID-19 single-cell cohort proves reliable for single-cell downstream analyses + +<|ref|>text<|/ref|><|det|>[[147, 381, 879, 465]]<|/det|> +We have previously demonstrated the capability of scSemiProfiler in accurately generating semi- profiled single- cell data that closely aligns with its real- profiled counterpart. Moving beyond basic cell type proportion predictions, which is the primary focus of other methods, scSemiProfiler excels in predicting gene expression for each cell within a population, thereby more authentically mimicking true single- cell data. This advancement is crucial for more complex downstream single- cell analysis tasks. + +<|ref|>text<|/ref|><|det|>[[147, 466, 879, 695]]<|/det|> +To illustrate the effectiveness of semi- profiled data in standard downstream single- cell analyses, we conducted a series of evaluations. A key task in these analyses is the identification of biomarkers within distinct cell clusters, highlighting genes with elevated expression levels. Utilizing the semi- profiled data generated by scSemiProfiler, we performed various single- cell level downstream analyses. The results from these analyses, as depicted in Fig. 3, demonstrate a remarkable consistency with outcomes derived from real- profiled data. For instance, we identified top cell type signature genes using the real- profiled cohort. When comparing their expression patterns in both real- profiled and semi- profiled datasets, the similarities were striking. The dot plots in Fig. 3a display these patterns, showcasing an almost indiscernible difference between the datasets. The semi- profiling at the single- cell level provides high- resolution expression data for marker genes within each cell population (cell type). Our approach reveals the distribution of marker gene expression across all cells within a specific type. While existing bulk deconvolution methods can offer average gene expression data for cell type markers, they fall short in depicting detailed gene expression distribution and variations among individual cells within the same population. This high level of cell type biomarker concordance underscores the scSemiProfiler's robustness not only in replicating single- cell data but also in ensuring the fidelity of downstream analytical processes. + +<|ref|>text<|/ref|><|det|>[[147, 696, 879, 910]]<|/det|> +We further explored the similarities between biomarkers identified using semi- profiled data and those from real- profiled data. Biomarker discovery is possible with lower- resolution data, such as the average cell type gene expression data provided by current bulk deconvolution tools. However, these methods are less reliable compared to single- cell data. The limitation of average cell type gene expression data is its lack of replicates for each cell type, leading to reduced statistical power. The absence of replicates makes it challenging to estimate variance within a cell type, which is essential for standard differential expression tests. In contrast, our semi- profiled single- cell data supports robust biomarker discovery through rigorous statistical testing, as it includes multiple samples (i.e. cells) for each cell type. We demonstrate the similarity of biomarkers identified using real- profiled and semi- profiled datasets through a rank- rank hypergeometric overlap (RRHO) plot. An RRHO plot [45] visualizes the overlap between two ranked gene lists, highlighting the degree of similarity and the significance of the overlap between them (see the Methods section for more details). Leveraging the RRHO plots, we compared the top 50 positive and top 50 negative gene lists associated with different cell types (Fig. 3b for CD4 cells and Supplementary Fig. S4 for all other cell types. The plots show positive marker and negative marker lists from both datasets are highly similar. A + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 80, 846, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 660, 850, 775]]<|/det|> +
Fig. 3 Comparative analyses of single-cell level downstream analysis tasks using real-profiled and semi-profiled COVID-19 datasets. a, Dot plots elucidating the expression proportion and intensity of discerned cell type signature genes. The top half showcases the real-profiled dataset, while the bottom delineates the semi-profiled version. b, RRHO plot emphasizing the congruence between the CD4 positive and negative markers in both datasets. c, Visualization of the GO term enrichment outcomes rooted in CD4 signature genes from both dataset versions. The plot accentuates the union of the top 10 enriched terms, with the Pearson correlation coefficients of between the bar lengths, which is based on the corresponding p-value and therefore represents the significant level. d, A juxtaposition of cell-cell interaction analyses stemming from real-profiled and semi-profiled cells from moderate COVID-19 patients, underscoring the similarity in interaction types and counts. e, Comparative depiction of pseudotime trajectories for CD4 cells across both datasets, highlighting their striking similarity in reconstructing dynamic cellular processes.
+ +<|ref|>text<|/ref|><|det|>[[118, 791, 850, 878]]<|/det|> +marked dissimilarity was evident between the positive and negative marker lists, which is intuitively anticipated. By definition, positive markers are genes that are higher expressed in the corresponding cell types, and negative markers are the opposite - lower expressed. Therefore, they should have no overlap. The compelling concordance demonstrated in Fig. 3a and b bolsters our claim that the semi- profiled data can viably supplant real- profiled data for the pivotal task of biomarker discovery using scSemiProfiler. + +<|ref|>text<|/ref|><|det|>[[118, 878, 850, 906]]<|/det|> +Next, we used the biomarkers derived from the semi- profiled dataset and those from the real- profiled dataset for gene functional enrichment analysis, assessing whether the two versions of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 84, 879, 226]]<|/det|> +analysis yield consistent results. Fig. 3c compares the Gene Ontology (GO) [46, 47] enrichment [48, 49] outcomes derived from real- profiled and semi- profiled datasets. The top 100 signature genes from both datasets are used for the enrichment analysis. We observed an overlap of 95 genes between the two lists, yielding a highly significant hypergeometric test p- value of \(4.00 \times 10^{- 200}\) (the population size of the hypergeometric test is the number of highly variable genes used for this dataset, 6030). A comparison of the top 10 overlapping terms from both versions reveals nearly identical significance (Pearson correlation coefficient of 0.998 with a p- value of \(4.13 \times 10^{- 12}\) for comparing the significant levels). The results for other cell types are in Supplementary Fig. S5. Reactome pathway [50] enrichment analysis results are in Supplementary Fig. S6. This further corroborates the reliability of semi- profiled data in downstream analyses. + +<|ref|>text<|/ref|><|det|>[[147, 228, 879, 411]]<|/det|> +Progressing to yet another pivotal single- cell level downstream analysis task, we evaluated the congruence in cell- cell interaction analyses derived from real- profiled and semi- profiled datasets. Given the paramount role of cell- cell interactions in orchestrating a myriad of multicellular processes, their analysis often unveils pivotal biological insights [51, 52]. Fig.3d juxtaposes cell- cell interaction analyses rooted in real- profiled and semi- profiled cells from moderate COVID- 19 patients (see results for other severity levels in Supplementary Fig. S7). The evident concordance in types and counts of interactions in both renditions reinforces the reliability of our semi- profiled data ( \(R = 0.905\) , \(P = 8.46 \times 10^{- 122}\) ). We also show a comparison of partition- based graph abstraction (PAGA) plots generated using real- profiled cohort and semi- profiled cohort in Supplementary Fig. S8, which demonstrates that the semi- profiled data can accurately capture the cellular trajectories and relationships between cell types. Given that such analyses intrinsically require single- cell data, scSemiProfiler emerges as the sole contender capable of producing data apt for this task from bulk sources. + +<|ref|>text<|/ref|><|det|>[[147, 412, 879, 527]]<|/det|> +Delving further into the capacity of semi- profiled data for other downstream single- cell level analysis tasks, we turned our attention to pseudotime analysis [19, 20]. Pseudotime is a pivotal tool in reconstructing dynamic cellular processes, ranging from differentiation pathways to developmental timelines or disease trajectories. As depicted in Fig. 3e, the pseudotime trajectories derived from real- profiled and semi- profiled CD4 cells are strikingly similar (Consistent results for the pseudotime analysis of other cell types can be found in Supplementary Fig. S9). Such compelling evidence underscores that the semi- profiled data retains its reliability even for intricate biological explorations like cell trajectory and differentiation analyses. + +<|ref|>sub_title<|/ref|><|det|>[[147, 541, 852, 573]]<|/det|> +## Semi-profiling maintains accuracy on a heterogeneous colorectal cancer dataset + +<|ref|>text<|/ref|><|det|>[[147, 580, 879, 794]]<|/det|> +We further tested the effectiveness of scSemiProfiler by validating it against a notably heterogeneous colorectal cancer dataset [53], which encompassed 112 single- cell sequencing samples that passed the quality control. This collection comprised 19 normal tissues, 86 tumor tissues (including colorectal cancer subtype iCMS2 and iCMS3), and 7 lymph node tissues. Considering the inherent diversity of this dataset, achieving accurate semi- profiling could ostensibly test the limits of scSemiProfiler. Again, this study does not include paired bulk sequencing data. Therefore, we have utilized the pseudo- bulk data derived from single- cell analysis as a surrogate for actual bulk sequencing. Nevertheless, following a consistent data processing and semi- profiling protocol, and selecting 36 representatives in batches of 4, the semi- profiled dataset mirrored its real- profiled counterpart. This congruence manifested not only in visual similarity but also in cell type distributions and subsequent analyses outcomes. Using the same estimation method as applied to the COVID- 19 cohort, the total cost for both bulk and single- cell sequencing to obtain this highly similar semi- profiled single- cell cohort is approximately \$73,320. This price also includes the cost of bulk data for the cohort and represents only 43.6% of the \$168,000 estimated for conducting single- cell sequencing on the entire cohort. + +<|ref|>text<|/ref|><|det|>[[147, 795, 879, 910]]<|/det|> +Figure 4a and b graphically highlight the remarkable similarity between the semi- profiled and real- profiled data using UMAP visualizations. These visualizations, color- coded according to cell types, show a substantial alignment between the datasets. The semi- profiled data mirrors the real- profiled data in terms of the location and shape of each cell type cluster. Notably, both datasets effectively segregate cell types such as plasma B, enteric glial, Mast, and epithelial. Additionally, a nuanced connection between fibroblast and endothelial cells is evident in both versions. Immuno- centric cells like McDC, T_NK, and B cells are also accurately positioned in close proximity in both datasets, underscoring the precision of the semi- profiled data. This similarity is further emphasized + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 200]]<|/det|> +in Figure 4c, which showcases the significant overlap between the two datasets. Moreover, Figure 4d employs distinct color schemes to differentiate between cells generated by the deep generative learning model and those from real- profiled representative data. The cells from the representatives cover a substantial portion, indicating their well- chosen representative selection. However, numerous cells that fall outside the representatives' distribution are accurately generated by the deep learning model, highlighting the critical role of both active learning and the deep generative model in achieving effective semi- profiling. The accuracy of the deep learning model's generation can be further shown in Supplementary Fig. S10, where we present the model's single- cell inference for individual samples. + +<|ref|>text<|/ref|><|det|>[[118, 200, 850, 313]]<|/det|> +To further justify the semi- profiled gene expression values are accurate and can be used for biological analysis, we compute the gene set activation pattern of the GO term "activation of immune response" (GO:0002253) for the two tumor tissue types the same way as we did for the COVID- 19 cohort (Fig. 4e). We chose this term because the immune response plays a significant role in the body's defense against cancer, and its activation or suppression can influence cancer progression and patient outcomes [54, 55]. The gene set activation scores are calculated and then adjusted by subtracting the score of the "Normal" tissue. The activation pattern in the real- profiled and semi- profiled datasets are highly similar, leading to a high Pearson correlation coefficient of 0.919 between them. + +<|ref|>text<|/ref|><|det|>[[118, 313, 850, 412]]<|/det|> +We also quantitatively examined the overall performance of scSemiProfiler as different numbers of representatives are selected. As shown in Fig. 4f, while our approach trumps the selection- only method, the gap narrows in comparison to results on the COVID- 19 dataset—owing largely to the colorectal cancer dataset's inherent heterogeneity. Despite this, the deep generative model's efficacy remains conspicuous, ensuring cost- effective error reduction. Also, if we aim to achieve an error as low as the previous COVID- 19 cohort, which leads to almost identical analysis results as the real data, only half of the samples need to be selected as representatives. This still reduces the cost significantly. + +<|ref|>text<|/ref|><|det|>[[118, 412, 850, 498]]<|/det|> +Diving deeper into the cell type proportions within the colorectal cancer cohort, one discerns variations across different tissue types—"Lymph Node", "Normal", "iCMS2", and "iCMS3". Fig. 4g illustrates these differences. For example, "Lymph Node" contains an expanded population of B cells compared with other tissue types, "Normal" is enriched with PlasmaB cells, and the two tumor subtypes have a pronounced epithelial presence. Remarkably, the semi- profiled dataset captures these nuances with precision, underlining its capability to replicate intricate analyses with fidelity. + +<|ref|>text<|/ref|><|det|>[[118, 498, 850, 640]]<|/det|> +Lastly, the benchmarking of deconvolution results presented in Fig. 4h and i positions scSemiProfiler at the forefront, significantly outperforming existing methods such as Bisque, TAPE, and Scaden. While its performance with four representatives is on par with CIBERSORTx, scSemiProfiler excels in computational memory efficiency. Further extending the representatives to 36 dramatically boosts the deconvolution accuracy. In contrast, existing methods such as CIBERSORT falter when tasked with handling a large reference set like 36 representatives, mainly due to their computational inefficiencies. This distinction underscores the scSemiProfiler's distinct advantage unshared by its peers. To provide a more detailed perspective on our deconvolution outcomes, Supplementary Fig. S11 showcases a comparative analysis. It displays our predicted cell type proportions for each individual sample alongside the ground truth, enabling a side- by- side evaluation + +<|ref|>sub_title<|/ref|><|det|>[[118, 655, 835, 704]]<|/det|> +## scSemiProfiler Ensures consistent downstream analyses between semi-profiled and real single-cell data in heterogeneous colorectal cancer cohorts + +<|ref|>text<|/ref|><|det|>[[118, 711, 849, 754]]<|/det|> +In the context of a heterogeneous dataset like this colorectal cancer one, the semi- profiled dataset stands robust, offering downstream analysis results that mirror the real- profiled data. This close resemblance is consolidated in Fig. 5. + +<|ref|>text<|/ref|><|det|>[[118, 755, 850, 911]]<|/det|> +A notable observation is the accuracy in analyzing biomarker expression pattern and their intracluster variation using semi- profiled data. Fig. 5a showcases dot plots for top cell type signature genes derived from the real dataset. These plots reflect an identical pattern in both the real- profiled and semi- profiled datasets. Further affirmation comes from the strong Pearson correlation coefficient between the colors (0.994) and sizes (0.996) of the dots. Notably, these correlation coefficients even surpass those observed in the more homogeneous COVID- 19 dataset. The semi- profiled dataset also reproduces biomarker discovery results, establishing its credibility as a suitable stand- in for the real- profiled data in such analyses. This consistency is exemplified by genes like KRT18 and KRT8, exclusive to epithelial cells, corroborated by existing literature [56, 57]. Another illustration is the unique expression of CPA3 and TPSB2 in Mast across both datasets. Beyond these top cell type signature genes, a granular examination of epithelial cells—encompassing the top 50 positive and + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[149, 80, 875, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 591, 880, 772]]<|/det|> +
Fig. 4 Detailed comparisons between the semi-profiled and real-profiled data in the heterogeneous colorectal cancer dataset. a, UMAP visualization of the real-profiled data, with colors denoting distinct cell types. Colors are consistent with (g). b, UMAP visualization of the semi-profiled data, with colors denoting distinct cell types. c, Joint UMAP visualization highlighting the close resemblance between the semi-profiled and real-profiled data. d, UMAP plot of the semi-profiled dataset, with color-coding distinguishing cells from the actual sequenced representatives and the ones generated through semi-profiling. e, "Activation of immune response" gene set relative activation pattern calculated for different tumor tissue types as compared to the "Normal" type in the real-profiled and semi-profiled datasets. Entries with fewer than 500 cells are left blank. f, Performance trajectory of the scSemiProfiler on the colorectal cancer dataset, showcasing its superiority over the selection-only approach, with costs computed similarly to Fig. 2d. g, Stacked bar plots comparing cell type compositions between the semi-profiled and real-profiled datasets across different tissues. The Pearson correlation coefficients between the real-profiled and semi-profiled tissues are LymphNode: 0.995, Normal: 0.993, iCMS2: 0.994, iCMS3: 0.988. h-i, Cell type deconvolution benchmarking. h, Pearson correlation coefficients between the actual cell type proportions and those estimated by various deconvolution methodologies. Note: DWLS failed to yield results after a week-long computation. i, Root Mean Square Error (RMSE) comparisons among different deconvolution techniques, highlighting the computational efficiency and accuracy of the scSemiProfiler, especially with an extended set of representatives.
+ +<|ref|>text<|/ref|><|det|>[[147, 789, 878, 875]]<|/det|> +negative markers—reinforces the congruence. All markers were identified using data at single- cell resolution through thorough statistical testing, a process unachievable with decomposed cell type- level data from standard bulk decomposition methods. As depicted in Fig. 5b, the preponderance of highly significant entries, with many p- values lower than \(10^{- 50}\) , strongly indicates a high degree of similarity in marker lists between the two datasets. RRHO plots for additional cell types can be found in Supplementary Fig. S12. + +<|ref|>text<|/ref|><|det|>[[147, 875, 877, 904]]<|/det|> +Diving deeper into the gene functional enrichment analysis, Fig. 5c offers further validation. Analyzing the top 100 signature genes across both datasets reveals a staggering 96 common markers, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 227]]<|/det|> +yielding a hypergeometric test p- value of \(8.88 \times 10^{- 188}\) (population size = 4053). GO terms from both datasets, along with their respective p- values, showcase pronounced similarity. Although the heterogeneity of the dataset leads to a relatively lower Pearson correlation coefficient of 0.593, the overall patterns in the two plots remain statistically similar ( \(P = 0.042\) ), leading to the same scientific conclusions. Moreover, when considering the union of the top 10 Gene Ontology (GO) terms from both the semi- profiled and actual datasets (comprising a total of 12 terms), there is a significant similarity in terms of enrichment p- values. Notably, the two versions of the top 10 GO terms have 9 overlap terms. More comprehensive GO term and pathway enrichment analysis results for other cell types' signature genes are also consistent for the real- profiled and semi- profiled versions (Supplementary Figs. S13 and S14). + +<|ref|>text<|/ref|><|det|>[[118, 228, 850, 411]]<|/det|> +Despite the increased heterogeneity of the dataset, the analysis of cell- cell interactions with the colorectal cancer semi- profiled cohort remains promising. As illustrated in Fig. 5d, the cell- cell interaction analysis, when executed on the real- profiled tumor tissue cells and the semi- profiled counterpart, reveals substantial consistency. Navigating the intricate interaction patterns characteristic of tumor tissues, the semi- profiled data astonishingly replicates the intricate layout with a robust Pearson correlation coefficient of 0.933. Both versions highlight enteric glial and fibroblast as the primary senders, while neutrophils emerge as the predominant receivers. Significantly, the most intense interactions identified across both sets involve the enteric glial with itself, the enteric glial with neutrophils, and the fibroblast with neutrophils. The cell- cell interaction results for other tissues also exhibit a high degree of similarity between the real- profiled and semi- profiled versions, as shown in Supplementary Fig. S15. Additionally, the strikingly similar PAGA plots presented in Supplementary Fig. S16 further demonstrate the utility of semi- profiled data in studying cellular trajectories and relationships between different cell types. + +<|ref|>text<|/ref|><|det|>[[118, 412, 850, 555]]<|/det|> +Shifting the focus to pseudotime analysis, we evaluated epithelial cells across all tissues as an example demonstration (Fig. 5e). Given the presence of cells from tumor tissues, this might introduce elevated heterogeneity within the cell type. Yet, the consistency between pseudotime analysis results from both versions is significant. Both versions discern lower pseudotime values concentrated at the base of the cluster, culminating in larger values towards the upper regions, with the pinnacle being the top- right quadrant. The statistical significance of the similarity is further validated by a Mann- Whitney U test [58, 59], yielding a compelling p- value of \(2.84 \times 10^{- 197}\) . The pseudotime analyses for other cell types also demonstrate a high degree of similarity, as evidenced in Supplementary Fig. S17. Such a finding underscores the capability of the semi- profiled data to adeptly capture intra- cluster nuances in detailed analyses. + +<|ref|>sub_title<|/ref|><|det|>[[118, 570, 780, 602]]<|/det|> +## Semi-profiling with real bulk measurements yields a dataset nearly identical to the original single-cell data + +<|ref|>text<|/ref|><|det|>[[118, 608, 850, 781]]<|/det|> +To further illustrate the adaptability of scSemiProfiler in real- world applications, we directed our analysis towards the iMGL dataset [60], which uniquely profiles both single- cell and real bulk RNA- seq measurements for the human inducible pluripotent stem cell (iPSC)- derived microglia- like (iMGL) cells, differing from the pseudobulk datasets previously used. There are 25 samples having both single- cell data and bulk RNA sequencing data. Samples are of different conditions (grown in cell culture for 0- 4 days and under various treatments). The availability of such datasets, which include both single- cell and bulk sequencing data on a large scale, remains very limited, partially due to the unnecessary of doing both sequencing for the same large- scale cohort and its prohibitive cost. This deviation from pseudobulk data is challenging. Pseudobulk, created by averaging out the noisy single- cell data [61], often fails to encapsulate the subtle features of actual bulk RNA- seq measurements. These real bulk measurements are inherently less noisy and often exhibit systematic differences when compared to single- cell RNA- seq data. + +<|ref|>text<|/ref|><|det|>[[118, 781, 850, 910]]<|/det|> +To navigate this complexity, we devised a method to infer pseudobulk directly from real bulk data (refer to the Methods section for comprehensive details). This approach enables us to more effectively utilize our deep learning model in the pseudobulk data space, which often aligns more closely with single- cell data. Through this method, despite the intricate challenges of the iMGL dataset, the results in Fig. 6 illustrate that our semi- profiled data parallels the real- profiled data quite closely. By selecting only eight representative samples out of a total of 25 samples, we could notably reduce the overall cost without compromising on accuracy. For this smaller dataset, we employed active learning to select 8 representative samples in batches of 2. Using the same estimation method as in the other two studies, the total cost for acquiring both bulk and semi- profiled single- cell data through + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[145, 80, 880, 608]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[145, 625, 879, 682]]<|/det|> +
Fig. 5 Downstream analysis results comparisons for the colorectal cancer dataset. a, Dot plots visualizing the cell type signature genes. b, RRHO plot visualizing the comparison between semi-profiled and real-profiled markers of epithelial cells. c, Epithelial cell type signature genes GO enrichment analysis results comparison. d, Cell-cell interaction results comparison between the real-profiled tumor tissue cells and semi-profiled tumor tissue cells. e, Pseudotime results comparison using the epithelial cells.
+ +<|ref|>text<|/ref|><|det|>[[145, 699, 877, 728]]<|/det|> +our method is approximately \(21,750. This amount is just 58\%\) of the estimated \(37,500 required for conducting single- cell sequencing across the entire cohort. + +<|ref|>text<|/ref|><|det|>[[145, 729, 879, 885]]<|/det|> +The UMAP visualization, as presented in Fig. 6a and b, solidifies our findings. Here, the two versions — semi- profiled and real- profiled — show remarkable consistency. The cell distributions of various iMGL subtypes (as delineated by the dataset provider) in the UMAP follow nearly identical patterns. The detailed observation showcases that clusters like “C2: Activated, immediate- early”, “C4: Activated, non- immediate- early”, and “C5: Activated, immediate- early” are interconnected and primarily located on the right- hand side. Likewise, “C3: Homeostatic, proliferative” finds its position at the upper left, with “C1: Homeostatic, non- proliferative” and “C6: Freshly thawed” lying at the bottom left. This consistency transcends to Fig. 6c, emphasizing a consistent cell distribution across the two versions. Furthermore, Fig. 6d underlines the precision of scSemiProfiler, where a majority of cells in the semi- profiled version were accurately generated. The efficient coverage of representative cells in this figure also highlights that our active learning strategy remains robust + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 80, 850, 628]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 643, 850, 792]]<|/det|> +
Fig. 6 Comparative analyses between semi-profiled and real-profiled iMGL datasets. a, UMAP visualization of the real-profiled iMGL cohort. Different colors represent different cell types. Colors are consistent with (g). b, UMAP visualization of the real-profiled iMGL cohort. c, Combined UMAP visualization showcasing the consistent cell distribution across both data versions. d, UMAP visualization highlighting the representatives' cells alongside the semi-profiled cells within the semi-profiled dataset. e, Relative pathway activation pattern of the GO term "activation of immune response" calculated for cells of different treatments as compared to cell type "C1: Homeostatic non-proliferative" in the real-profiled and semi-profiled cohorts. Entries with fewer than 100 cells are left as blank. f, Performance evaluation of the scSemiProfiler on the iMGL dataset, emphasizing its efficiency in error reduction. g, A comparative illustration of cell type proportions under varying experimental conditions, accentuating the similarity in patterns between datasets. The Pearson correlation coefficients between the real-profiled and semi-profiled versions of cell type proportions under different conditions are: iMGL.D0: 0.999, iMGL.D1: 0.871, iMGL.D2: 0.993, iMGL.D3: 0.989, iMGL.D4: 0.992, iMGL.DMSC: 0.998, iMGL.GW30: 0.960, iMGL.GW-300: 0.999, iMGL.T-30: 0.987, iMGL.T-300: 0.999. h, i, Deconvolution performance benchmarking using Pearson correlation and RMSE.
+ +<|ref|>text<|/ref|><|det|>[[118, 809, 850, 838]]<|/det|> +even when navigating the challenges of real bulk measurements. The supplementary Fig. S18 further illustrates the precise single- cell inference achieved by our deep learning model on this dataset. + +<|ref|>text<|/ref|><|det|>[[119, 838, 850, 896]]<|/det|> +Fig. 6e provides an in- depth look at the accuracy of semi- profiled gene expression values. Since microglia are the resident immune cells in the brain, we checked the activation pattern of GO term "activation of immune response" in each cell type under each treatment. The semi- profiled dataset presents a highly similar activation pattern as the real- profiled dataset. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 879, 127]]<|/det|> +Fig. 6f offers further evidence of the scSemiProfiler's effectiveness. Here, the performance curve of the semi- profiled approach significantly undercuts the selection- only one, demonstrating its capability to limit semi- profiling errors while optimizing on costs. + +<|ref|>text<|/ref|><|det|>[[147, 128, 879, 227]]<|/det|> +An intriguing observation from the iMGL dataset is the cell type proportion's dynamic shifts under various experimental conditions (Fig. 6g). The transitions from iMGL_D0 to iMGL_D4, for instance, reveal a progressive increase in the proportions of C1 and C4 cells. Contrastingly, "C2: Activated, immediate- early" cells peak at iMGL_D1 and then decrease steadily. Although the intricate effects of drugs iMGL_GW and iMGL_T need further investigation, preliminary data suggests that elevated doses result in a surge of "C1: Homeostatic, non- proliferative" cells. Impressively, these intricate variations are mirrored in the semi- profiled dataset. + +<|ref|>text<|/ref|><|det|>[[147, 228, 879, 370]]<|/det|> +A juxtaposition of our deconvolution method against others on the iMGL dataset offers illuminating insights. The performance of CIBERSORTx, which was one of the best in other datasets, dramatically decreases, potentially due to the inherent challenges posed by real bulk data and the nuanced similarities among cell types. Other methods, such as Bisque, TAPE, and Scaden, are more robust to those challenges, showing decent deconvolution performance. Despite these challenges, scSemiProfiler showcases resilience and consistently outperforms all its peers except Bisque, a fact further corroborated by the Wilcoxon test [62] (see p- values in Fig. 6h and i). The marginal difference between our scSemiProfiler and the selection- only method is a testament to both approaches nearing optimal performance in this specific context. The supplementary Fig. S19 further presents our accurate deconvolution for individual samples. + +<|ref|>sub_title<|/ref|><|det|>[[147, 384, 787, 417]]<|/det|> +## Semi-profiling using real bulk measurement also leads to reliable downstream results + +<|ref|>text<|/ref|><|det|>[[147, 424, 879, 481]]<|/det|> +In this more realistic setting, where scSemiProfiler is tasked with semi- profiling using real bulk data for downstream analyses, the semi- profiled data consistently mirrored results from the real- profiled version. This is particularly remarkable given the unique challenges presented by the real bulk data. We present these downstream analysis results in Fig. 7. + +<|ref|>text<|/ref|><|det|>[[147, 481, 879, 581]]<|/det|> +The markers identified using single- cell resolution data in the real- profiled dataset were almost identical in expression patterns to those in the semi- profiled dataset. Fig. 7a visually reinforces this, showing "C1: Homeostatic, non- proliferative" and "C10: Homeostatic, non- proliferative" with virtually indistinguishable expression patterns across both data types. Additionally, unique expressions in "C11: Myeloid, progenitors", such as GAPA2 and HPGDS, were consistently observed. The overarching similarities were further quantified with impressive Pearson correlation coefficients for both dot sizes and colors, clocking in at 0.980 and 0.989, respectively. + +<|ref|>text<|/ref|><|det|>[[147, 582, 879, 653]]<|/det|> +Further validating the congruency of our method, Fig.7b presented the RRHO plot of the top 50 positive and negative C3 markers from both single- cell datasets (see RRHO plots for other cell types in Supplementary Fig. S20). The degree of similarity is substantial, with most of the markers showcasing p- values less than \(10^{- 50}\) . Such findings strongly suggest that the scSemiProfiler is adept at producing reliable data for biomarker discovery. + +<|ref|>text<|/ref|><|det|>[[147, 654, 879, 752]]<|/det|> +Proceeding to more in- depth downstream analysis using the top 100 signature genes for GO enrichment, we observed an overlap of 90 genes between the semi- profiled and the real single- cell datasets (Fig. 7c). The hypergeometric test revealed a significant p- value of \(1.81 \times 10^{- 183}\) . The enriched terms identified from both the semi- profiled and real datasets matched closely. The Pearson correlation coefficient between the two versions' significance is 0.995, underscoring the consistency in their analytical outcomes. Extended GO and Reactome enrichment analysis in Supplementary Figs. S21 and S22 further confirm the accuracy of finding signature genes using semi- profiled data. + +<|ref|>text<|/ref|><|det|>[[147, 753, 879, 823]]<|/det|> +In the case of cell- cell interactions, both real and semi- profiled single- cell data did not capture any significant interactions, probably due to the similarity between cell clusters in the iMGL data. Instead, the partition- based graph abstraction (PAGA) analysis [63] showcased in Fig.7d highlighted that major cell type links were consistent across datasets. Further cementing this was the strong Pearson correlation coefficient of 0.865 between the adjacency matrices of the two networks. + +<|ref|>text<|/ref|><|det|>[[147, 824, 879, 880]]<|/det|> +Pseudotime analysis in Fig.7e also affirmed the alignment between the two datasets. The topographical pseudotime alignment between them is almost congruent, to the extent that the Mann- Whitney U test implemented in SciPy returned a p- value smaller than the smallest float number our computer can represent, which is \(2.23 \times 10^{- 308}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 85, 850, 128]]<|/det|> +In conclusion, our findings robustly demonstrate that scSemiProfiler seamlessly adapts to real-world scenarios employing real bulk data. The downstream analytical outcomes derived from the semi- profiled data are significantly consistent with those based on real data. + +<|ref|>image<|/ref|><|det|>[[120, 145, 850, 632]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 645, 850, 736]]<|/det|> +
Fig. 7 Similarities in downstream single-cell analyses between real-profiled and semi-profiled data for the iMGL cohort. a, Dot plots visualizing the nearly identical expression patterns of cell type signature genes across both datasets. b, RRHO plots highlighting the striking similarities between the top 50 positive and negative C3 markers from both real and semi-profiled datasets. c, Overlapping GO enrichment analysis results for the top C3 signature genes, emphasizing consistent analytical outcomes between the datasets. d, PAGA plots illustrating the consistent major cell type links observed in both datasets. e, Pseudotime plots affirming the topographical alignment between the real-profiled and semi-profiled cohort. The p-value of the Mann-Whitney U test is smaller than the smallest float number that the system can represent, which is \(2.23 \times 10^{-300}\) .
+ +<|ref|>sub_title<|/ref|><|det|>[[118, 772, 738, 806]]<|/det|> +## Active learning demonstrates its prowess in selecting the most informative samples for enhanced single-cell profiling + +<|ref|>text<|/ref|><|det|>[[118, 812, 850, 868]]<|/det|> +The crux of scSemiProfiler's strategy revolves around judiciously selecting representative samples. The rationale is straightforward: the more informative the chosen representatives, the better the semi- profiling performance. This not only enhances the fidelity of the generated profiles but also provides a cost- effective approach by minimizing the number of necessary representatives. + +<|ref|>text<|/ref|><|det|>[[118, 868, 850, 911]]<|/det|> +Initially, our methodology bore similarities to uncertainty sampling [28, 64, 65], a heuristic active learning technique. Here, the intuition is to query samples with the most uncertainty, thereby maximizing the incremental information acquired. In our algorithm, we employed bulk data to pick out + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 878, 128]]<|/det|> +batches of samples that exhibited the most variance from their designated representatives. Since this method does not use information from the base learner (the deep generative model), it is still a passive learning algorithm. + +<|ref|>text<|/ref|><|det|>[[147, 129, 878, 199]]<|/det|> +To improve the representative selection, we then turn the algorithm into an active learning algorithm by incorporating the information from the deep generative models. The algorithm also utilizes the clustering information in the cohort and is thus a type II active learning algorithm [66]. Combining these two ideas, the algorithm aims to reduce the total heterogeneity of each sample cluster, ensuring that each target sample has a similar representative, thus optimizing semi- profiling performance. + +<|ref|>text<|/ref|><|det|>[[147, 200, 879, 340]]<|/det|> +As depicted in Fig. 8, we juxtaposed our advanced active learning algorithm against the rudimentary passive learning approach. Each panel, from Fig. 8a to Fig. 8c, encapsulates the comparative analyses derived from distinct datasets. While the x- axis maps the representatives earmarked for single- cell sequencing, the y- axis portrays the single- cell in silico inference difficulty, which is quantified using the average single- cell- level difference (see Equation 16 in the Methods section) between target samples and their representatives. This premise holds that as the dissimilarity between the target and the representatives increases, the complexity of the in silico inference task also rises. This metric is crucial as it highlights the challenges encountered by the deep generative learning model in semi- profiling, thereby illuminating the effectiveness of the strategies used for selecting representatives. + +<|ref|>text<|/ref|><|det|>[[147, 341, 879, 569]]<|/det|> +The empirical evidence is resounding. Across all datasets, the active learning algorithm showcased its mettle by consistently pinpointing representatives that considerably reduced the total distance to other samples. This underscores the algorithm's capability to foster superior representative selection for semi- profiling. In Fig. 8a, when applying our method to the COVID- 19 dataset, active learning shows better performance than passive learning, especially in the beginning iterations of the semi- profiling. Active learning reduces the inference difficulty when the same number of representatives are selected. Also, consider the marked point where 28 representatives are selected, which is consistent with representatives we selected for our analyses in previous sections. To reach the same level of inference difficulty, passive learning needs to select 4 more representatives, i.e. the cost of 4 samples' single- cell sequencing experiment is saved. For the colorectal cancer dataset (Fig. 8b), active learning continues to perform better than passive learning, and at the point we selected, the cost for more than two batches (8) of representatives can be saved using active learning. Fig. 8c shows the results for the iMGL dataset, in which active learning is also significantly better than passive learning. In all experiments, active learning consistently outperforms passive learning in selecting representatives. With the same budget, active learning achieves lower inference difficulty. Furthermore, to reach a comparable level of inference difficulty, active learning requires a lower cost. + +<|ref|>image<|/ref|><|det|>[[147, 586, 878, 723]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 739, 878, 854]]<|/det|> +
Fig. 8 Active learning demonstrates its prowess in selectively profiling the most informative samples at the single-cell level. The x-axis represents the number of samples selected for single-cell profiling (representatives). The y-axis shows the single-cell in silico inference difficulty of the dataset, which is quantified by the average single-cell difference from each sample to its representative, showcasing the efficiency of representative selection strategies. The marked stars signify the iterations chosen for our methodology, with the generated data underpinning the analyses detailed in previous sections. a, Results from the COVID-19 dataset. Active learning shows significantly better performance, especially in the beginning when a few representatives are selected. b, Observations derived from the colorectal cancer dataset. Active learning continues to show significantly better performance even when more representatives are selected. c, Insights from the iMGL dataset with real bulk measurements. Active learning still manages to outperform passive learning significantly.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 80, 243, 99]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[118, 109, 850, 366]]<|/det|> +In the present work, we introduced scSemiProfiler, an innovative computational framework designed to produce affordable single- cell measurements for a given large cohort. The allure of scSemiProfiler stems from its unparalleled capacity to mirror the outcomes of conventional single- cell sequencing, yet with remarkable cost- effectiveness. Marrying expansive sequencing with pinpoint profiling and optimizing data utilization, it charts a robust, intricate, and economical path for large- scale single- cell exploration. This tool is not merely a means of generating data but ensures the derived single- cell information remains dependable for an array of downstream analyses. The pipeline ingests bulk data, leverages an active learning module to judiciously select representatives for authentic single- cell sequencing, and employs a deep generative learning model, VAE- GAN, to infer the single- cell data of the remainder of the cohort. Through this approach, every sample in the cohort has access to single- cell data, either semi- profiled or real- profiled. We subjected scSemiProfiler to rigorous evaluation using three distinct cohorts, simulating a scenario devoid of single- cell data and then juxtaposing the output of our tool with the authentic single- cell datasets. The striking resemblance between semi- profiled and real- profiled data in facets like UMAP visualization and cell type proportions ascertains the robustness of scSemiProfiler. Furthermore, both datasets align almost perfectly in subsequent analyses, encompassing biomarkers, enrichment analyses, cell- cell interactions, and pseudotime trajectories, offering a cost- effective means of integrating single- cell data in cohort studies without compromising on analytical precision. + +<|ref|>text<|/ref|><|det|>[[118, 366, 850, 764]]<|/det|> +The magnitude and uniqueness of our study's contributions can be primarily attributed to two game- changing innovations that stand apart in their novelty and transformative potential. Precision- driven Computational Modeling: The crowning achievement of our study lies in the design and implementation of a groundbreaking computational model for single- cell level bulk decomposition. While traditional methodologies offer a mere cell type level decomposition of bulk data, scSemiProfiler shatters these bounds with an unprecedented capability. What we've pioneered isn't just a method to interpret bulk data, but a mechanism to deconvolute it into authentic single- cell data. This offers a myriad of sophisticated and meaningful downstream analytic possibilities, previously unattainable with conventional techniques. To the best of our knowledge, scSemiProfiler is the very first in its class to provide a true single- cell level decomposition. Furthermore, when placed under rigorous scrutiny, the semi- profiled data generated by our model consistently showcases remarkable similarity to real- profiled single- cell data, underscoring its reliability and potential to reshape the paradigm of single- cell analytics, particularly for large- scale cohort studies in which real single- cell sequencing cost would be prohibitive. Intelligent Active Learning Mechanism: Our second salient innovation hinges on the power of active learning. Instead of passively selecting representatives, our model embarks on an intelligent, iterative journey. By constantly evaluating the data and learning from previous rounds of selection, this module discerns which representatives would offer the most informative insights for semi- profiling. This isn't a mere addition; it's a reinvention of how representative selection operates. The synchronization of insights from the deep learning model with this active learning module ensures that the selection process is not just data- driven, but also insight- driven. The implications of this are twofold: firstly, it enhances the quality and relevance of the selected representatives, ensuring that they truly resonate with the cohort's cellular composition. Secondly, it offers an economic advantage. By intelligently choosing representatives, the financial overheads associated with single- cell experiments can be minimized, ensuring maximum return on investment both in terms of data quality and budgetary constraints. Together, these innovations not only elevate the efficacy of scSemiProfiler but also pioneer a new direction in how single- cell data can be derived, analyzed, and utilized, holding profound implications for future biomedical research endeavors. + +<|ref|>text<|/ref|><|det|>[[118, 765, 850, 894]]<|/det|> +Future work on scSemiProfiler is focused on expanding its capabilities from single- cell RNA sequencing (scRNA- seq) to other biological modalities, such as proteomics. The adaptability of the tool's current methodology will require adjustments in algorithms to suit the unique data characteristics of each new modality, along with comprehensive performance evaluations benchmarked against established standards. A pivotal element of this development is the utilization of cross- modal information exchange, which aims to improve the accuracy of semi- profiling. This approach could significantly reduce the costs associated with single- cell multi- omics studies beyond the scope of the current study, by minimizing the need for extensive representative samples from each modality and leveraging the inherent similarities across various biological data types. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 85, 879, 313]]<|/det|> +The scSemiProfiler represents a groundbreaking shift in single- cell analysis, particularly for large- scale and cohort studies, by offering a cost- effective yet comprehensive approach. It achieves this by selectively performing single- cell experiments on a few samples using active learning, coupled with deep generative models for in silico inference of single- cell profiles for the remaining samples, thereby creating similar semi- profiled cohorts at a substantially lower cost. This approach not only alleviates financial constraints in expansive biomedical research but also extends its utility beyond cohort studies. The tool's deep generative model enables "single- cell level deconvolution" across various studies, provided there is a single- cell reference. Initially centered on RNA sequencing datasets, scSemiProfiler's adaptability to other single- cell and bulk data modalities opens new avenues in personalized medicine and can significantly enrich global data repositories. Significantly, scSemiProfiler is designed to complement, rather than compete with, large- scale single- cell sequencing technologies, such as WT Mega by Parse Biosciences. WT Mega is notable for its ability to process up to 1 million cells across 96 samples. When scSemiProfiler is integrated with these advanced sequencing platforms, it further reduces costs, enabling single- cell level profiling of thousands of samples in large- scale studies at a more manageable expense. This synergy significantly broadens the scope and depth of possible research, enhancing the efficiency and affordability of large- scale single- cell studies. + +<|ref|>sub_title<|/ref|><|det|>[[147, 326, 252, 344]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[147, 356, 495, 373]]<|/det|> +## The initial setup of scSemiProfiler + +<|ref|>text<|/ref|><|det|>[[147, 379, 879, 580]]<|/det|> +The initial setup of the scSemiProfiler method plays a pivotal role (depicted in Fig. 1 a), serving as the foundation for subsequent semi- profiling iterations. Initially, each individual sample within the designated cohort undergoes bulk sequencing. Upon obtaining the raw count data from this sequencing, we perform library size normalization followed by a log1p transformation. From the processed data, we identify and select highly variable genes and relevant markers for further analysis. Subsequently, we employ Principal Component Analysis (PCA) [67- 69], implemented in the Python package Scikit- learn [70], to reduce the dimensionality of the processed bulk data. We then determine the batch size of representatives ( \(B\) ) representing the number of selected representatives for actual single- cell profiling in each iteration. Once the representative batch size is defined, we cluster the dimensionality- reduced bulk data to select the initial batch of representatives. The dimensionality- reduced data is clustered into \(B\) distinct clusters using the KMeans algorithm [71], the sample that is closest to the cluster centroid is designated as the representative for that cluster. This well- prepared initial setup provides the clustered bulk sequencing data and identified representatives for high- resolution single- cell sequencing. + +<|ref|>sub_title<|/ref|><|det|>[[147, 592, 642, 609]]<|/det|> +## Representative single-cell profiling and processing + +<|ref|>text<|/ref|><|det|>[[147, 615, 879, 686]]<|/det|> +After the selection of representatives, the scSemiProfiler method proceeds to the single- cell profiling phase, focusing specifically on the chosen representatives (Fig. 1b). The raw count data obtained from single- cell profiling undergoes a sequence of preprocessing steps. These steps encompass traditional quality controls as well as advanced processing techniques aimed at enhancing the learning of the deep generative model. + +<|ref|>text<|/ref|><|det|>[[147, 687, 879, 913]]<|/det|> +To initiate, expression values that are extremely low (below \(0.1\%\) of the library size) are adjusted to zero. This adjustment is based on insights from prior studies [72, 73], which have demonstrated that these minimal values are often representative of background noise and, therefore, should be excluded. The threshold for this processing is empirically determined by selecting the value that yields the most high- quality UMAP visualizations consistent with the original cell type annotations. Subsequently, a series of standard preprocessing steps commonly used by existing methods [74, 75] is applied, including the removal of low- quality cells (expressing fewer than 200 genes), genes (expressed in fewer than 3 cells), dead cells (with a high proportion of mitochondrial reads), and doublets (expressing an excessive number of genes). Afterward, the data is normalized to a library size (total count of each cell) of 10,000, followed by a log1p transformation. The data matrix is then cropped to include only the highly variable genes identified from the bulk data. Our preprocessing approach is carefully tailored to meet the unique requirements of each dataset. For instance, the gene expression data of the COVID- 19 dataset arrive in a preprocessed state, including cell quality control, normalization, and log1p transformation. We retrieved the data as normalized RNA counts, eliminated the background noise, and subsequently reapplied normalization and log1p transformation. Finally, we selected the columns corresponding to highly variable genes and important cell type markers. In contrast, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 156]]<|/det|> +colorectal cancer dataset is provided in raw counts, necessitating a full spectrum of standard single- cell preprocessing procedures. With the iMGL dataset, our focus is on cells annotated by the dataset provider. We identify genes common to both bulk and single- cell datasets and apply standard preprocessing steps, excluding the removal of low- quality cells, as this task has already been performed by the dataset provider. + +<|ref|>text<|/ref|><|det|>[[118, 156, 850, 271]]<|/det|> +After the initial preprocessing, the single- cell data is further refined by integrating two types of prior knowledge to enhance cell representations. Firstly, gene set scores are calculated using all curated gene sets from MSigDB [48] and Ernst et al [76], sourced from UNIFAN [77]. Gene set scores are determined by averaging the expression values for each gene set, followed by a log1p transformation, and then combined with the preprocessed single- cell gene expression matrix. Secondly, we adopt a feature weighting method, giving more weight to features with higher variance in each cell's input when calculating reconstruction loss. More comprehensive details of this strategy are available in the section focusing on single- cell inference. + +<|ref|>sub_title<|/ref|><|det|>[[118, 284, 782, 317]]<|/det|> +## Pretrain the deep generative learning model for reconstructing the single-cell data of the selected representatives + +<|ref|>text<|/ref|><|det|>[[118, 323, 850, 510]]<|/det|> +The next phase, depicted in Fig.1c, involves training a deep generative learning model for the in silico single- cell data inference of a non- representative target sample using processed representative single- cell data and bulk data of both samples. This will be executed for all non- representative samples to ensure everyone in the study cohort will have single- cell data (real- profiled or semi- profiled) in the end. Firstly, our deep generative model aims to reconstruct the single- cell profiles of representatives. This reconstruction lays the foundation of the single- cell inference of the target samples, which can be viewed as a modified single- cell data reconstruction task. When performing the single- cell inference, the initial single- cell data reconstruction of the representative is modified by introducing the difference between the target sample and the representative based on their distinct bulk data profiling, aligning the synthetic single- cell data for the target sample closer to its actual single- cell profiling counterpart. The fundamental reasoning is that the single- cell data matrix of a specific target sample should bear resemblance to the representatives chosen from the same study. Significant differences between them can be adjusted and guided by the disparity as delineated by the bulk sequencing data. + +<|ref|>text<|/ref|><|det|>[[118, 510, 850, 624]]<|/det|> +We designed a VAE- GAN- based model for the representative reconstruction and the target single- cell generation guided by their expression difference at the bulk level. It ingests gene expression data into an MLP encoder, which subsequently outputs parameters of a multivariate Gaussian. A random variable, \(\mathbf{z}\) , is sampled from this Gaussian and processed through an MLP decoder, resulting in parameters of a Zero- Inflated Negative Binomial (ZINB) distribution - optimal for modeling single- cell RNA- seq data [78- 80]. Training the VAE parameters involves maximizing the data likelihood and minimizing the KL divergence [81] between the latent variable distribution and a standard Gaussian distribution, following the Evidence Lower BOund (ELBO) loss framework [30]. + +<|ref|>text<|/ref|><|det|>[[118, 624, 850, 823]]<|/det|> +Innovating upon this foundational VAE structure, we integrated the following four techniques to enhance effective cell representation learning. Gene Set Score Inclusion: As mentioned in the data preprocessing section, beyond mere gene expression reconstruction, we compute gene set scores for cells and concatenate them to the gene expression data. During the two stages of pretrain, we also compute the reconstruction loss for gene set scores. Such inclusion furnishes the model with an enriched biological context, fostering a more comprehensive learning of the input cells. Hence, input for each cell becomes the concatenation of gene expression and gene set scores: \((\mathbf{x}_1, \mathbf{s}_1)\) . Feature Importance Weight: We also compute a weight vector \(\mathbf{w}\) to weight the contribution of each feature to the VAE's reconstruction loss based on the feature's variance. Graph Convolutional Networks (GCN) [82, 83]: A GCN layer assimilates adjacent cells' information, mitigating dropout concerns and the inherent noise of single- cell sequencing. Generative Adversarial Network (GAN) [31] Dynamics: We employ a discriminator network, resonating with GAN structures, which discerns between genuine and generated cell data. This guides the generator towards producing more authentic single cell data. The resultant loss function for the generator discussed above is: + +<|ref|>equation<|/ref|><|det|>[[364, 835, 847, 852]]<|/det|> +\[L_{G_{P r e t r a i n 1}} = L_{V A E} - \lambda_{d}L_{D_{F a k e}} \quad (1)\] + +<|ref|>equation<|/ref|><|det|>[[175, 864, 847, 904]]<|/det|> +\[L_{VA E} = \sum_{i = 1}^{N}[-{\bf w}\cdot log(Z I N B(({\bf x}_{1},{\bf s}_{1})|\rho (z_{i}),\pi (z_{i}),\theta))] + K L(q({\bf z}|({\bf x}_{1},{\bf s}_{1}))||{\mathcal{N}}({\bf 0},I))] \quad (2)\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[370, 95, 876, 135]]<|/det|> +\[L_{D_{F a k e}} = \sum_{i = 1}^{N} - l o g(1 - D(G((\mathbf{x}_{i},\mathbf{s}_{i})))) \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[145, 133, 879, 348]]<|/det|> +\(L_{G_{P r e t r a i n 1}}\) is the loss used to train the VAE- based generator network during the first pertrain stage. It has two components, a VAE loss \(L_{VA E}\) and another loss \(L_{D_{F a k e}}\) including the feedback from the discriminator. In \(L_{V A E}\) \(N\) is the number of cells in the representative's single- cell dataset. \(\mathbf{x_{i}}\) is the vector representing the gene expression value of the \(i\) - th cell in the dataset and \(\mathbf{s_{i}}\) is the corresponding gene set score. \(\mathcal{N}(\mathbf{0},I)\) is the standard Gaussian distribution. \(\mathbf{z_{i}}\) is a low- dimensional latent variable sampled from the Gaussian distribution \(q(\mathbf{z}|\mathbf{(x_{i},s_{i})})\) whose parameters \(\mu_{i}\) and \(\pmb{\Sigma}_{i}\) are generated by the encoder network: \(M L P_{E n c o d e r}(G C N((\mathbf{x_{i}},\mathbf{s_{i}})))\) . We follow the ZINB distribution design of SCVI [78] and generate ZINB parameter using the decoder network: \(\pmb {\rho}(\mathbf{z_{i}}) = M L P_{D e c o d e r,\rho}(\mathbf{z_{i}})\) \(\pi (\mathbf{z_{i}}) = M L P_{D e c o d e r,\pi}(\mathbf{z_{i}})\) and \(\theta\) is a free parameter vector that are learned in the training process. In \(L_{D_{\mathrm{F a k e}}}\) \(G\) represents the VAE- based generator so \(G((\mathbf{x_{i}},\mathbf{s_{i}}))\) represents the \(i\) - th cell reconstructed by the generator. The reconstructed cell data is the mean of the generator's \(Z I N B\) distribution. And \(D\) is our discriminator network, so \(D(G((\mathbf{x_{i}},\mathbf{s_{i}})))\) is the discriminator's predicted probability of the \(i\) - th reconstructed cell being a real- profiled cell. \(\lambda_{d}\) denotes an empirically determined scaling factor. Meanwhile, the discriminator network will be trained using cross- entropy loss to reach higher classification performance. + +<|ref|>equation<|/ref|><|det|>[[315, 356, 876, 397]]<|/det|> +\[L_{D} = \sum_{i = 1}^{N}[-log(D((\mathbf{x_{i}},\mathbf{s_{i}}))) - log(1 - D(G((\mathbf{x_{i}},\mathbf{s_{i}}))))] \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[145, 408, 879, 495]]<|/det|> +Here, \(D((\mathbf{x_{i}},\mathbf{s_{i}}))\) represents the discriminator's predicted probability of the \(i\) - th input cell being real when it is indeed real, and \(D(G((\mathbf{x_{i}},\mathbf{s_{i}})))\) , as mentioned in a previous paragraph, represents the discriminator's predicted probability of the \(i\) - th cell being real when it is, in fact, a cell reconstructed by the generator network. When trained together, the generator will first be trained until convergence, followed by alternating training with the discriminator every three epochs. Together, they form a GAN with a min- max objective: + +<|ref|>equation<|/ref|><|det|>[[295, 501, 876, 540]]<|/det|> +\[m i n_{G}m a x_{D}\sum_{i = 1}^{N}[l o g(D((\mathbf{x_{i}},\mathbf{s_{i}}))) + l o g(1 - D(G((\mathbf{x_{i}},\mathbf{s_{i}}))))] \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[145, 540, 879, 584]]<|/det|> +In this context, the discriminator's objective is to maximize its accuracy, aiming for precise classification, while the generator's goal is to minimize discrepancies, generating realistic cells to challenge the discriminator. + +<|ref|>text<|/ref|><|det|>[[145, 584, 879, 714]]<|/det|> +Overall, our approach involves two pretrain stages and one fine- tune stage for performing single- cell inference. In the initial pretrain stage, we employ \(L_{G_{P r e t r a i n 1}}\) and \(L_{D}\) to train the generator \(G\) and discriminator \(D\) , respectively, forming a GAN. The primary objective here is to train a generator capable of reconstructing genuine cell data from the representatives. Subsequently, the second pretraining stage closely resembles the first, but it functions in full- batch mode. This modification enables the inclusion of an extra bulk loss term in the generator's loss function. Such an addition improves the model's capacity to leverage insights from bulk data and better aligns the single- cell data with its bulk counterpart. This enhancement is crucial for priming the model for the in silico inference of single- cell data from bulk data, while also strengthening the data reconstruction process. + +<|ref|>equation<|/ref|><|det|>[[330, 725, 876, 789]]<|/det|> +\[\begin{array}{r}{L_{G_{P r e t r a i n 2}} = L_{V A E} - \lambda_{d}L_{D_{F a k e}} + \lambda_{B u l k R}L_{B u l k R}}\\ {L_{B u l k R} = \left\| \frac{\sum_{i = 1}^{N}\mathbf{x_{i}^{\prime}}}{N} - \frac{\sum_{i = 1}^{N}\mathbf{x_{i}}}{N}\right\|_{2}} \end{array} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[145, 789, 879, 832]]<|/det|> +The term \(\lambda_{B u l k R}\) represents another empirical hyperparameter for regularization weight. \(\mathbf{x_{i}^{\prime}}\) represents the reconstructed \(i\) - th cell. The bulk loss is defined as the disparity between the pseudobulk of the real representative dataset and that of the reconstructed representative dataset. + +<|ref|>sub_title<|/ref|><|det|>[[145, 844, 822, 879]]<|/det|> +## Fine-tune the deep generative learning model to infer the single-cell measurements for the target samples + +<|ref|>text<|/ref|><|det|>[[145, 885, 879, 914]]<|/det|> +After successfully pretraining our deep generative model so that it can perform accurate single- cell data reconstruction, the subsequent stage of our method leverages this foundation. In this phase, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 228]]<|/det|> +we introduce an extra term to the loss function that accounts for the difference between the bulk data of the representative and that of the target sample. This fine- tuning process refines the model's capabilities, enabling it to perform in silico generation of cells from the target sample while maintaining consistency with the observed bulk expression patterns. Consequently, our deep generative model transitions from being solely focused on reconstruction to generating single- cell data for the target sample. In this crucial stage, we retire the discriminator from our model architecture, as our focus shifts from generating cells that resemble the representatives' cells to adapting the model for the target samples. Consequently, the training process exclusively involves the generator. The loss function for the generator adopted in this stage bears resemblance to that of the second pretrain phase, with the distinction of the bulk loss transitioning from \(L_{BulkR}\) to \(L_{BulkT}\) : + +<|ref|>equation<|/ref|><|det|>[[359, 240, 848, 257]]<|/det|> +\[L_{\mathrm{Inference}} = L_{\mathrm{VAE}} + \lambda_{\mathrm{BulkT}}L_{\mathrm{BulkT}} \quad (8)\] + +<|ref|>equation<|/ref|><|det|>[[325, 264, 848, 302]]<|/det|> +\[L_{\mathrm{BulkT}} = \left\| \frac{\sum_{i = 1}^{N}\mathbf{x}_{1}^{i}}{N} -\frac{\sum_{i = 1}^{N}\mathbf{x}_{1}}{N}\cdot \frac{\mathbf{B}_{\mathrm{T}} + \epsilon}{\mathbf{B}_{\mathrm{R}} + \epsilon}\right\|_{2} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[118, 303, 850, 392]]<|/det|> +Here, \(\mathbf{B}_{\mathrm{T}}\) and \(\mathbf{B}_{\mathrm{R}}\) represent the real bulk data of the target and representative samples, respectively, while \(\epsilon\) , whose elements are all set to 0.1, is introduced as a pseudocount vector to avoid division by zero. The second term of equation 9, \(\frac{\sum_{i = 1}^{N}\mathbf{x}_{1}}{N}\frac{\mathbf{B}_{\mathrm{T}} + \epsilon}{\mathbf{B}_{\mathrm{R}} + \epsilon}\) , functions as an estimation of the pseudobulk value for the target sample, derived from the actual bulk data. This term is incorporated to reduce the systematic discrepancy between the pseudobulk and the actual bulk data corresponding to the same sample. + +<|ref|>text<|/ref|><|det|>[[118, 392, 850, 606]]<|/det|> +By incorporating the actual bulk difference alongside the pseudobulk data of the representative, this term effectively simulates the pseudobulk data for the target sample, ensuring that any disparities between the pseudobulk and actual bulk do not distort the model's performance. In cases such as the COVID- 19 and colorectal cancer dataset, where real bulk data is unavailable, the entire second term simplifies to the target's pseudobulk. Furthermore, as the bulk loss \(L_{BulkT}\) inherently propagates more gradient to higher gene expression values (given the target bulk loss is in a squared error form), we implement a strategy to gradually reduce the gradient received by larger values. See details in the "Deep Generative Learning Model Training Details" section. Concurrently, we increase the weight of the bulk loss in the loss function over the course of training. This refined approach ensures that, after adequately training for larger values, the model can fine- tune smaller values, resulting in a comprehensive and precise single- cell data generation for the target samples. To counteract the influence of potentially noisy gene expression with minuscule values, we introduced a neuron activation threshold. This threshold converts values that are exceedingly small, specifically those below 0.1% of the input library size (a threshold consistent with that used for background noise removal in preprocessing), to zero. + +<|ref|>sub_title<|/ref|><|det|>[[118, 618, 560, 636]]<|/det|> +## Deep generative learning model architecture + +<|ref|>text<|/ref|><|det|>[[118, 642, 850, 714]]<|/det|> +In this study, our model integrates a Variational Autoencoder (VAE) as the generator and a Multi- Layer Perceptron (MLP) as the discriminator. We developed both components using PyTorch [84]. A concise overview of the model architecture is provided below, and for an in- depth understanding, we encourage consulting the complete details available in our GitHub repository (refer to the code availability section). + +<|ref|>text<|/ref|><|det|>[[118, 714, 850, 912]]<|/det|> +Before sending the cell data into the model, we performed some precomputation for the cells, which is part of a Graph Convolutional Network (GCN) layer. We aggregated information from the K (15 by default) nearest neighboring cells for each input cell. The neighbor finding for each cell was based on the 'neighbors. neighbors. graph' tool in the Scikit- learn package. We preprocessed this message- passing step on the log1p transformed cell expression data, subsequently inputting the aggregated cell data into the deep learning model. The encoder of the VAE then employs linear layers and activation functions, leading to a hidden layer comprising 256 neurons. The message passing, the first linear layer, and the first activation function in the encoder form a GCN layer together. The design of the rest of the VAE structure, inclusive of both the encoder and decoder, was adapted from SCVI. We chose 32 for the latent space dimension and 256 for the decoder's hidden layer size. For the discriminator, we constructed a 4- layer MLP, with Leaky ReLU activations [85] between linear layers. The hidden layers are configured with 256, 128, and 10 neurons, respectively. The output layer employs a Sigmoid activation to produce the prediction probability of the input cell being real. Weight initialization in the model utilizes the Kaiming method [86]. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[147, 82, 646, 100]]<|/det|> +## Deep Generative Learning Model Training Details + +<|ref|>text<|/ref|><|det|>[[147, 105, 879, 163]]<|/det|> +Following the detailed outline of the VAE- GAN model structure, we next describe the detailed training settings, which were meticulously designed to optimize the model's performance. The training process was structured in a three- stage sequence to ensure comprehensive learning and fine- tuning of the model parameters. + +<|ref|>text<|/ref|><|det|>[[147, 163, 879, 235]]<|/det|> +Initially, in the first pretraining stage, we focus on training the VAE generator independently until convergence (setting 100 epochs as a default value). Then, the generator and discriminator are trained jointly until the generator cannot further reduce its loss. When trained together, the generator and the discriminator are trained alternatively for 3 epochs each, and this will go for 100 iterations by default. During this phase, we employed the default SCVI learning rate of \(1 \times 10^{- 3}\) . + +<|ref|>text<|/ref|><|det|>[[147, 235, 879, 334]]<|/det|> +The second pretraining stage mirrors the first in its basic approach but incorporates notable modifications. It is executed in full- batch mode, with an additional representative bulk loss integrated into the training process. Here, the generator is again trained until convergence (set to 50 epochs by default), and then trained jointly with the discriminator until convergence. The default setting is training the two networks for 50 iterations, where each network is trained for 3 epochs in each iteration. To facilitate more refined adjustments during this stage, the learning rate is reduced to \(1 \times 10^{- 4}\) . + +<|ref|>text<|/ref|><|det|>[[147, 334, 879, 476]]<|/det|> +In the final stage of training, the single- cell inference fine- tuning stage, the discriminator is set aside, and training is executed through a series of mini- stages. Each mini- stage involves setting the gradients for highly expressed genes above specific thresholds to zero, a strategy aimed at ensuring equitable optimization for genes with smaller expression values. These thresholds are calculated relative to the peak expression value (max. expr) from the representative's normalized count matrix, set at No threshold, \(\frac{1}{2} \max . \exp r\) , \(\frac{1}{4} \max . \exp r\) , \(\frac{1}{6} \max . \exp r\) , and \(\frac{1}{8} \max . \exp r\) . Training is continued until convergence in each mini- stage, typically not exceeding 150 epochs, with a learning rate of \(2 \times 10^{- 4}\) . As the thresholds are lowered, the weight for the target bulk loss in the loss function is progressively quadrupled, ensuring a consistent total magnitude for this loss component. Upon completion of all mini- stages, the trained model is then utilized to generate single- cell data for the target sample. + +<|ref|>sub_title<|/ref|><|det|>[[147, 490, 828, 523]]<|/det|> +## Incrementally select representatives using active learning to improve single-cell inference + +<|ref|>text<|/ref|><|det|>[[147, 529, 879, 615]]<|/det|> +Upon executing the deep generative learning models for in silico inference, every sample in the cohort possesses single- cell data, either real- profiled or semi- profiled. While researchers can opt to conclude here and proceed with downstream analysis using the single- cell data, when budget permits, our active learning module can further refine the process. By pinpointing additional informative representatives for single- cell sequencing, some target samples will be assigned a more similar representative, leading to enhanced single- cell inference performance. + +<|ref|>text<|/ref|><|det|>[[147, 616, 879, 672]]<|/det|> +Our initial representative selection strategy is predicated on the belief that selecting the most "uncertain" samples as the representatives results in the most inference improvement and yields the richest insights for our model. Consequently, we select the sample with the largest bulk data difference compared to its representative to serve as the new representative: + +<|ref|>equation<|/ref|><|det|>[[412, 683, 875, 708]]<|/det|> +\[R_{new} = \underset {i\in C o h o r t}{\arg \max}D_{b}(R(i),i) \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[147, 720, 879, 777]]<|/det|> +Where \(R(i)\) is the representative of \(i\) , and \(D_{b}\) is the Euclidean distance between the PCA- reduced bulk data of two samples. This process is executed \(B\) (representative batch size) times. Upon choosing new representatives, we update the membership for all samples. This ensures that, if a sample aligns more closely with a new representative than their previous one, their affiliation shifts: + +<|ref|>equation<|/ref|><|det|>[[428, 790, 875, 814]]<|/det|> +\[R(i) = \underset {r\in R s}{\arg \min}D_{b}(r,i) \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[147, 826, 879, 884]]<|/det|> +where \(R(i)\) is the representative of the sample \(i\) and \(R s\) is the set of representatives. Conceptually, this approach is similar to the uncertainty sampling [64] algorithm, a subtype of active learning. However, this method is passive learning since it does not utilize the knowledge acquired by the base learners, our deep generative learning models. + +<|ref|>text<|/ref|><|det|>[[144, 884, 879, 913]]<|/det|> +To improve the representative selection, we developed our active learning algorithm by merging the information gained by the deep generative model base learners into the selection of new + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 84, 850, 228]]<|/det|> +representatives. To commence, we pinpoint the \(B\) sample clusters exhibiting the highest in- cluster heterogeneity, which is the aggregate distance from each sample to the cluster's representative. The heterogeneity for a cluster \(c\) is expressed as: \(\sum_{i\in c}[D_{b}(R_{c},i) + \lambda_{sc}D_{sc}(R_{c},i) + \lambda_{pb}D_{pb}(R_{c},i)\) , where \(i\) can be any sample in the cluster and \(R_{c}\) is the representative. Essentially, a sample's heterogeneity is the combined distance—across three distance metrics—between the sample and its representative. The total cluster heterogeneity is derived from the summation of all sample heterogeneity. \(D_{b}\) denotes the Euclidean distance between the PCA- reduced bulk data of two samples. The term \(D_{sc}(R_{c},i)\) represents the difficulty of transforming the representative single- cell data to that of a target. This is quantified by averaging the Euclidean distances between the representative cells and their K nearest neighbors in the target sample, all within a PCA- reduced space. Formally: + +<|ref|>equation<|/ref|><|det|>[[325, 238, 848, 281]]<|/det|> +\[D_{sc}(a,b) = \frac{1}{N_{a}}\sum_{i = 1}^{N_{a}}\sum_{j\in KNN(i)}ED(\mathbf{v_{a,i}},\mathbf{v_{b,j}}) \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[118, 293, 850, 395]]<|/det|> +with \(a,b\) being two samples in the semi- profiled cohort, \(v_{a},v_{b}\) being PCA reduced single- cell data matrices for the two samples, \(ED(x,y)\) computing the Euclidean distance between \(x\) and \(y\) , and \(KNN(i)\) identifying the cell's \(K\) (1 by default) nearest neighbor in another sample's data matrix. Subscript represents rows, e.g. \(\mathbf{v_{a,i}}\) is the \(i\) th row (ith cell) of the PCA reduced data of sample \(a\) . \(D_{pb}\) is the Euclidean distance of the log1p transformed pseudobulk data of two samples, which is positively related to the amount of bulk loss and thus quantifies the deep learning model's difficulty in performing the in silico inference. Finally, \(\lambda_{sc}\) and \(\lambda_{pb}\) are empirical scaling factors. + +<|ref|>text<|/ref|><|det|>[[144, 395, 504, 409]]<|/det|> +The most heterogeneous cluster \(C_{H}\) is chosen as: + +<|ref|>equation<|/ref|><|det|>[[260, 420, 848, 451]]<|/det|> +\[C_{H} = \underset {c\in \mathrm{clusters}}{\arg \max}\sum_{i\in c}[D_{b}(R_{c},i) + \lambda_{sc}D_{sc}(R_{c},i) + \lambda_{pb}D_{pb}(R_{c},i)] \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[118, 463, 850, 592]]<|/det|> +After selecting the \(B\) most heterogeneous clusters to split, the subsequent step involves selecting a new representative for each of these \(B\) clusters to minimize total in- cluster heterogeneity. Within each cluster, every non- representative sample is a potential new representative, each representing a different way of splitting the cluster. In the cluster, samples that are closer in bulk distance \(D_{b}\) to the potential new representative than to the original representative can be reassigned to the new representative, splitting the cluster into two. The total heterogeneity of these two new clusters is then calculated based on the bulk distance. For each of the \(B\) most heterogeneous clusters, we select the sample whose corresponding split results in the minimal total in- cluster heterogeneity to be the new representative \(R_{new}\) , as shown in the equation below. + +<|ref|>equation<|/ref|><|det|>[[319, 602, 848, 632]]<|/det|> +\[R_{new} = \underset {j\in C_{H}}{\arg \min}\sum_{i\in C_{H}}min(D_{b}(j,i),D_{b}(R_{c},i)) \quad (14)\] + +<|ref|>text<|/ref|><|det|>[[118, 635, 850, 664]]<|/det|> +Where \(R_{c}\) denotes the original representative of this cluster. Finally, the cluster membership updating will be executed to make sure each sample \(i\) is assigned to the closest representative \(R(i)\) . + +<|ref|>equation<|/ref|><|det|>[[400, 675, 848, 700]]<|/det|> +\[R(i) = \underset {r\in R_{B}}{\arg \min}D_{b}(r,i) \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[118, 703, 850, 775]]<|/det|> +With the identification of new representatives, they are subjected to real single- cell data profiling and appended to the existing collection. In subsequent semi- profiling iterations, certain target samples will be assigned with more analogous representatives, enhancing in silico single- cell inference accuracy. This iterative procedure persists until either the budget is exhausted or a sufficient number of representatives are ascertained, ensuring optimal semi- profiling results. + +<|ref|>sub_title<|/ref|><|det|>[[118, 789, 470, 806]]<|/det|> +## Semi-profiling pipeline stop criteria + +<|ref|>text<|/ref|><|det|>[[118, 812, 850, 912]]<|/det|> +For our semi- profiling pipeline that iteratively selects representatives and performs in silico single- cell inference, we recommend two types of stop criteria for the users to choose according to their own needs. First, in the case that the user knows their budget and aims to achieve the best possible semi- profiling performance, they can run the pipeline and keep choosing representatives until they run out of budget. Second, in the case that the user aims to achieve an acceptable semi- profiling performance with the least amount of budget, the iterative cycle persists until an acceptable performance is reached. The performance is measured by comparing the representatives' actual single- cell data and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 84, 880, 300]]<|/det|> +the inferred data for them in previous iterations. Based on the user's needs, we recommend three levels of criteria. First, when the user is mainly studying the cell type proportion, we recommend stopping when the Pearson correlation between the ground truth and the estimated proportion reaches 0.85. Second, when the user's study includes more detailed analysis, such as biomarker discovery, we recommend stopping when the Pearson correlations between the biomarker expression patterns (dot colors and sizes in the dot plots) discovered in real- profiled data and those in semi- profiled data reach 0.85. Finally, for studies requiring more intricate single- cell level analysis, it is advised to halt when the Pearson correlation between data matrices, generated from advanced analysis tasks like trajectory inference and cell- cell interaction, for semi- profiled and real- profiled data, reaches 0.85. These guidelines are merely suggestions, and users have the flexibility to modify the stopping criteria to suit their specific requirements. The effectiveness of the semi- profiling process can be consistently evaluated by comparing the semi- profiled single- cell data (from the previous round) with the actual single- cell profiling (in the subsequent round). If the actual single- cell profiling aligns with the semi- profiling generated by our model from the previous round, this indicates a robust semi- profiling model has been effectively learned, and the iterative process can thus be concluded. + +<|ref|>sub_title<|/ref|><|det|>[[147, 313, 525, 330]]<|/det|> +## Semi-profiling performance evaluation + +<|ref|>text<|/ref|><|det|>[[147, 336, 880, 380]]<|/det|> +Cell type Annotation for Semi- Profiled Data: Utilizing the annotated single- cell data from the representatives, we trained an MLP classifier available in the Scikit- learn package [70]. This classifier was then employed to categorize the cell types for cells in the semi- profiled data. + +<|ref|>text<|/ref|><|det|>[[146, 393, 880, 494]]<|/det|> +Dimensionality Reduction: In order to juxtapose the real- profiled and semi- profiled cohorts, we merged them and executed Principal Component Analysis (PCA) to reduce it to 100 principal components. This PCA- processed data serves as the foundation for computing single- cell level difference metrics. For visualization purposes, we first used SCANPY [75] to compute the neighbor graph using the PCA- reduced data, with the local neighborhood size parameter set to 50. Subsequently, the Uniform Manifold Approximation and Projection (UMAP) functionality within SCANPY was used for the visualization process. + +<|ref|>text<|/ref|><|det|>[[146, 507, 880, 608]]<|/det|> +Semi- profiling error quantification: To assess the accuracy of the semi- profiled cohort, we performed PCA to reduce its dimensionality, in conjunction with the real- profiled data, to 100 dimensions. For each sample within the semi- profiled dataset, we calculated a single- cell difference metric in comparison to the real- profiled single- cell data corresponding to that sample. This involves determining the Euclidean distance of each cell to its K- nearest neighbors in another dataset within the PCA- reduced space, followed by averaging these distances across all cells. The formula below outlines the difference computation between single- cell matrices \(\mathbf{m}_{\mathbf{a}}\) and \(\mathbf{m}_{\mathbf{b}}\) . + +<|ref|>equation<|/ref|><|det|>[[184, 630, 877, 674]]<|/det|> +\[D i f_{s c}(\mathbf{m}_{\mathbf{a}},\mathbf{m}_{\mathbf{b}}) = \frac{1}{N_{a} + N_{b}} (\sum_{i = 1}^{N_{a}}\sum_{j\in K N N(i)}E D(\mathbf{v}_{\mathbf{a},i},\mathbf{v}_{\mathbf{b},j}) + \sum_{i = 1}^{N_{b}}\sum_{j\in K N N(i)}E D(\mathbf{v}_{\mathbf{b},i},\mathbf{v}_{\mathbf{a},j})) \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[146, 686, 879, 758]]<|/det|> +Where \(D i f_{s c}\) denotes single- cell level difference. \(N_{a}\) and \(N_{b}\) signify the cell counts in matrices \(\mathbf{m}_{\mathbf{a}}\) and \(\mathbf{m}_{\mathbf{b}}\) respectively. The vs represent the PCA- reduced matrices. The terms \(\mathbf{v}_{\mathbf{a},i}\) , \(\mathbf{v}_{\mathbf{b},j}\) represent the \(i\) th and \(j\) th cell (rows) in the PCA- reduced matrices of sample \(a\) and \(b\) respectively. The function \(K N N\) identifies the \(K\) nearest neighbors within the alternate sample's PCA- reduced data matrix. We chose \(K = 1\) for our experiments. \(E D\) computes the Euclidean distance between two vectors. + +<|ref|>text<|/ref|><|det|>[[146, 758, 879, 830]]<|/det|> +Referring to the comprehensive performance curves depicted in (Fig.2d, Fig.4d, Fig.6d, the normalized error \(E_{N}\) displayed on the left y- axis represents the mean single- cell difference between each sample's real- profiled version and semi- profiled version. This error is normalized using a theoretical upper bound ( \(U B\) ) and a lower bound ( \(L B\) ) of an expected single- cell difference between samples in the study. + +<|ref|>equation<|/ref|><|det|>[[346, 827, 876, 867]]<|/det|> +\[E_{N} = \frac{1}{U B - L B} (\frac{1}{P}\sum_{i = 1}^{P}D i f_{s c}(\mathbf{m}_{\mathbf{i}},\tilde{\mathbf{m}}_{\mathbf{i}}) - L B) \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[146, 869, 745, 884]]<|/det|> +and if \(i\) itself is a representative, the single- cell difference equals the lower bound. + +<|ref|>equation<|/ref|><|det|>[[392, 897, 633, 912]]<|/det|> +\[D i f_{s c}(\mathbf{m}_{\mathbf{i}},\tilde{\mathbf{m}}_{\mathbf{i}}) = L B,\quad \mathrm{if} i\in R s\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 850, 243]]<|/det|> +where \(P\) stands for the total count of samples within the cohort and \(\tilde{\mathbf{m}}_1\) refers to the semi- profiled version of \(\mathbf{m}_1\) . \(R\) s denotes the set of representatives. The upper bound \(UB\) was derived by averaging the single- cell difference of randomly paired samples based on their PCA- reduced real- profiled data. This represents the performance of the worst possible representative selection, which is random selection. For computing the lower bound \(LB\) , we first divided the comprehensive PCA matrix, which contains all single- cell samples, into two random halves. Due to the randomness of the split, these two halves have the same data probabilistic distribution. Furthermore, given the large sample size, these halves are expected to have approximately the same actual cell distribution and thus can be regarded as two replicates of the same study. The difference between the replicates should be regarded as the lower bound, as it represents the best performance that an optimal selection can achieve [87]. + +<|ref|>text<|/ref|><|det|>[[118, 255, 850, 555]]<|/det|> +Deconvolution benchmarking: We benchmarked the accuracy of our deconvolution method by comparing its performance against established methods: CIBERSORTx, Bisque, TAPE, Scaden, and DWLS. We ran CIBERSORTx using the official web portal. Bisque, TAPE, and DWLS were installed according to the official GitHub repositories provided in the corresponding publications. We used the PyTorch version of Scaden implemented in TAPE's GitHub repository for its testing. First, we annotated our cells generated by the deep learning model using the MLP classifier, which was trained on the representatives' annotated data, as discussed previously. Using this annotated data, we computed the cell type proportion. Subsequently, we employed the Pearson correlation coefficient (calculated using the Python package SciPy [59]) and RMSE to assess the consistency with the ground truth. However, when applying the aforementioned methods to the three datasets, we encountered both time and space complexity constraints for many benchmarked methods. Specifically, CIBERSORTx and Bisque were unable to utilize as many representative samples for single- cell references as our approach. Scaden and TAPE always only sample 5,000 cells for training their models, thereby also failing to fully exploit all available representatives. To ensure a fair comparison, we also offered an alternative version of our results for these datasets. In this version, the single- cell reference is confined to just the first 4 representatives, aligning with the capacity of all other benchmarked methods. Across all three datasets, DWLS failed to produce results even after running for a week and was thus excluded from the comparison. To statistically ascertain if one method notably surpasses another in performance, we calculated the p- values using a one- sided Wilcoxon test [62], as implemented in SciPy. This test enables us to determine whether the metric of one method is significantly higher or lower compared to that of another. + +<|ref|>text<|/ref|><|det|>[[118, 568, 850, 826]]<|/det|> +Gene set activation scores computation and heatmap plotting: We computed the interferon pathway activation scores following the same procedure in the study [35] from which we acquired the COVID- 19 dataset. To compute the activation scores depicted in the heatmaps (refer to Fig.2c), we first computed the average expression for each cell type and severity combination using the code from the COVID- 19 dataset provider, which uses the 'tl.score_genes' tool in SCANPY. Then, for each cell type, we computed the activation score of each severity level as the fold changes from the healthy condition to it. For the "activation of immune response" activation pattern in the colorectal cancer dataset depicted in Fig.4c, we first collected the genes of this GO term (GO:0002253) and used the same SCANPY tool to compute the average expressions for each cell type and tissue combination. For each cell type, we calculated the activation scores by determining the fold changes in comparison to the "Normal" tissue type across all other tissue types. For the "activation of immune response" activation pattern in the iMGL dataset depicted in Fig.6c, we first computed the average gene expression for each cell type and tissue combination similarly. The activation score was determined by calculating the fold change for each cell type, using "C1: Homeostatic non- proliferative" as the reference background. Considering the relatively smaller size of the colorectal cancer and iMGL dataset and the fact that some entries have very few cells, the values were only computed for entries with more than 500 cells for the colorectal cancer dataset and entries with more than 100 cells for the iMGL dataset. + +<|ref|>text<|/ref|><|det|>[[118, 840, 850, 911]]<|/det|> +Biomarker discovery and enrichment analysis: For making the RRHO plots, we used SCANPY to identify each cell type's top 50 positive cell type markers and top 50 negative cell type markers for the real- profiled and semi- profiled datasets. RRHO plots are used for visualizing the overlap between two ranked gene lists. Each entry in the plot corresponds to a negative logged p- value of the hypergeometric test between two marker lists. The bottom left quadrant corresponds to the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 879, 156]]<|/det|> +comparison between two positive marker lists. In this quadrant, the plot entry in the \(i\) - th row from bottom to top and \(j\) - th column from left to right corresponds to the comparison between the top \(i\) genes in the real- profiled version top negative markers and top \(j\) gene in the semi- profiled version top negative markers. Other quadrants are plotted similarly, with the marker lists always starting from the respective corners of the plot. + +<|ref|>text<|/ref|><|det|>[[147, 157, 879, 228]]<|/det|> +For the GO and Reactome enrichment analysis, we identified the top 100 cell type signature genes for both real- profiled and semi- profiled datasets using SCANPY. The identified markers were subsequently employed for enrichment analysis with the Python package GSEApy [88]. For enrichment analysis visualization in Fig.3c, Fig.5c, Fig.7c, we employed the adjusted p- value from GSEApy's output to indicate significance. + +<|ref|>text<|/ref|><|det|>[[147, 241, 879, 313]]<|/det|> +Cell- cell interaction analysis: We conducted cell- cell interaction analyses on both real- profiled and semi- profiled datasets using the R package CellChat [89]. For the COVID- 19 dataset, we analyzed interactions of cells from patients with each disease severity. In the colorectal cancer dataset, our focus was on interactions of cells originating from different tissues, and in the iMGL dataset, we evaluated cells across various condition groups. + +<|ref|>text<|/ref|><|det|>[[147, 326, 879, 412]]<|/det|> +Pseudotime analysis: We carried out pseudotime analysis utilizing the Monocle3 R package [90]. The UMAP coordinates, previously generated in our UMAP comparison between real- profiled and semi- profiled datasets, served as the input for pseudotime computation. To ensure a fair comparison between the real- profiled and semi- profiled data, we consistently selected the roots at the same positions within the UMAP space. To compare the similarity between the pseudotime values computed for the two datasets, we employed the Mann- Whitney U test, as implemented in SciPy. + +<|ref|>text<|/ref|><|det|>[[147, 425, 879, 499]]<|/det|> +Cell trajectory analysis using partition- based graph abstraction (PAGA): The PAGA plots were generated using SCANPY. First, PCA was applied independently to each dataset to reduce them to 100 principal components. Subsequently, the PCA- reduced data was utilized to compute the neighbor graphs, with the size of the local neighborhood set to 50, aligning with our UMAP settings. Based on the neighbor graphs, we generated the PAGA plots using SCANPY. + +<|ref|>sub_title<|/ref|><|det|>[[147, 512, 343, 531]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[147, 541, 879, 627]]<|/det|> +Three datasets are used for evaluating our method. The preprocessed COVID- 19 dataset is from Stephenson et al.'s study [35] and can be downloaded from Array Express under accession number E- MTAB- 10026 The colorectal cancer dataset is from Joanito et al.'s study [53]. The count expression matrices are available through Synapse under the accession codes syn26844071 The iMGL dataset is from Ramaswami et al.'s study [60]. The raw count iMGL bulk and single- cell data can be downloaded from the Gene Expression Omnibus (GEO) repository under accession number GSE226081. + +<|ref|>sub_title<|/ref|><|det|>[[147, 641, 346, 660]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[147, 670, 879, 700]]<|/det|> +Code for the models and results reproduction is publically available on GitHub: https://github.com/mcgilldinglab/scSemiProfiler. + +<|ref|>sub_title<|/ref|><|det|>[[148, 713, 360, 732]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[147, 742, 879, 841]]<|/det|> +This work was funded in part by grants awarded to [JD]. We gratefully acknowledge the support from the Canadian Institutes of Health Research (CIHR) under Grant Nos. PJT- 180505; the Funds de recherche du Québec - Santé (FRQS) under Grant Nos. 295298 and 295299; the Natural Sciences and Engineering Research Council of Canada (NSERC) under Grant No. RGPIN2022- 04399; and the Meakins- Christie Chair in Respiratory Research. We also extend our gratitude to Yanshuo Chen, the lead author of TAPE, for his indispensable support in the deconvolution benchmarking experiments. This work is part of the HCA publication bundle. + +<|ref|>sub_title<|/ref|><|det|>[[147, 856, 275, 874]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[152, 884, 879, 914]]<|/det|> +[1] Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X., Bodeau, J., Tuch, B.B., Siddiqui, A., et al.: mrna- seq whole- transcriptome analysis of a single cell. Nature methods + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[150, 85, 306, 100]]<|/det|> +6(5), 377- 382 (2009) + +<|ref|>text<|/ref|><|det|>[[123, 110, 850, 140]]<|/det|> +[2] Shapiro, E., Biezuner, T., Linnarsson, S.: Single- cell sequencing- based technologies will revolutionize whole- organism science. Nature Reviews Genetics 14(9), 618- 630 (2013) + +<|ref|>text<|/ref|><|det|>[[125, 151, 850, 194]]<|/det|> +[3] Li, Y., LeMaire, S.A., Shen, Y.H.: Molecular and cellular dynamics of aortic aneurysms revealed by single- cell transcriptomics. Arteriosclerosis, Thrombosis, and Vascular Biology 41(11), 2671- 2680 (2021) + +<|ref|>text<|/ref|><|det|>[[125, 205, 850, 263]]<|/det|> +[4] Kazer, S.W., Aicher, T.P., Muema, D.M., Carroll, S.L., Ordovas- Montanes, J., Miao, V.N., Tu, A.A., Ziegler, C.G., Nyquist, S.K., Wong, E.B., et al.: Integrated single- cell analysis of multicellular immune dynamics during hyperacute hiv- 1 infection. Nature medicine 26(4), 511- 518 (2020) + +<|ref|>text<|/ref|><|det|>[[125, 274, 850, 303]]<|/det|> +[5] Chai, R.L., Grandis, J.R.: Advances in molecular diagnostics and therapeutics in head and neck cancer. Current treatment options in oncology 7, 3- 11 (2006) + +<|ref|>text<|/ref|><|det|>[[125, 314, 850, 371]]<|/det|> +[6] Marquette, C.- H., Boutros, J., Benzaquen, J., Ferreira, M., Pastre, J., Pison, C., Padovani, B., Bettayeb, F., Fallet, V., Guibert, N., et al.: Circulating tumour cells as a potential biomarker for lung cancer screening: a prospective cohort study. The Lancet Respiratory Medicine 8(7), 709- 716 (2020) + +<|ref|>text<|/ref|><|det|>[[125, 382, 850, 425]]<|/det|> +[7] Wedge, D.C., Gundem, G., Mitchell, T., Woodcock, D.J., Martincorena, I., Ghori, M., Zamora, J., Butler, A., Whitaker, H., Kote- Jarai, Z., et al.: Sequencing of prostate cancers identifies new cancer genes, routes of progression and drug targets. Nature genetics 50(5), 682- 692 (2018) + +<|ref|>text<|/ref|><|det|>[[125, 436, 850, 479]]<|/det|> +[8] Mathys, H., Davila- Velderrain, J., Peng, Z., Gao, F., Mohammadi, S., Young, J.Z., Menon, M., He, L., Abdurrob, F., Jiang, X., et al.: Single- cell transcriptomic analysis of alzheimer's disease. Nature 570(7761), 332- 337 (2019) + +<|ref|>text<|/ref|><|det|>[[125, 490, 850, 534]]<|/det|> +[9] Pal, B., Chen, Y., Vaillant, F., Capaldo, B.D., Joyce, R., Song, X., Bryant, V.L., Penington, J.S., Di Stefano, L., Tubau Ribera, N., et al.: A single- cell rna expression atlas of normal, preneoplastic and tumorigenic states in the human breast. The EMBO journal 40(11), 107333 (2021) + +<|ref|>text<|/ref|><|det|>[[120, 545, 850, 589]]<|/det|> +[10] Depuydt, M.A., Prange, K.H., Slenders, L., Ord, T., Elbersen, D., Boltjes, A., De Jager, S.C., Asselbergs, F.W., De Borst, G.J., Aavik, E., et al.: Microanatomy of the human atherosclerotic plaque by single- cell transcriptomics. Circulation research 127(11), 1437- 1455 (2020) + +<|ref|>text<|/ref|><|det|>[[120, 600, 850, 630]]<|/det|> +[11] Wang, Z., Gerstein, M., Snyder, M.: Rna- seq: a revolutionary tool for transcriptomics. Nature reviews genetics 10(1), 57- 63 (2009) + +<|ref|>text<|/ref|><|det|>[[120, 641, 850, 685]]<|/det|> +[12] Newman, A.M., Steen, C.B., Liu, C.L., Gentles, A.J., Chaudhuri, A.A., Scherer, F., Khodadoust, M.S., Esfahani, M.S., Luca, B.A., Steiner, D., et al.: Determining cell type abundance and expression from bulk tissues with digital cytometry. Nature biotechnology 37(7), 773- 782 (2019) + +<|ref|>text<|/ref|><|det|>[[120, 696, 850, 739]]<|/det|> +[13] Jew, B., Alvarez, M., Rahmani, E., Miao, Z., Ko, A., Garske, K.M., Sul, J.H., Pietilainen, K.H., Pajukanta, P., Halperin, E.: Accurate estimation of cell composition in bulk expression through robust integration of single- cell information. Nature communications 11(1), 1971 (2020) + +<|ref|>text<|/ref|><|det|>[[120, 750, 850, 779]]<|/det|> +[14] Nel, S., Durandt, C., Murdoch, C., Pepper, M.S.: Determinants of dental pulp stem cell heterogeneity. Journal of Endodontics 48(10), 1232- 1240 (2022) + +<|ref|>text<|/ref|><|det|>[[120, 791, 850, 819]]<|/det|> +[15] Papalexi, E., Satija, R.: Single- cell rna sequencing to explore immune cell heterogeneity. Nature Reviews Immunology 18(1), 35- 45 (2018) + +<|ref|>text<|/ref|><|det|>[[120, 831, 850, 860]]<|/det|> +[16] Oliveira, G., Wu, C.J.: Dynamics and specificities of t cells in cancer immunotherapy. Nature Reviews Cancer, 1- 22 (2023) + +<|ref|>text<|/ref|><|det|>[[120, 872, 850, 901]]<|/det|> +[17] Rahimi, A.M., Cai, M., Hoyer- Fender, S.: Heterogeneity of the nih3t3 fibroblast cell line. Cells 11(17), 2677 (2022) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 880, 130]]<|/det|> +[18] Liu, Y., Mi, Y., Mueller, T., Kreibich, S., Williams, E.G., Van Drogen, A., Borel, C., Frank, M., Germain, P.- L., Bludau, I., et al.: Multi- omic measurements of heterogeneity in hela cells across laboratories. Nature biotechnology 37(3), 314- 322 (2019) + +<|ref|>text<|/ref|><|det|>[[145, 140, 880, 184]]<|/det|> +[19] Trapnell, C., Cacchiarelli, D., Grimsby, J., Pokharel, P., Li, S., Morse, M., Lennon, N.J., Livak, K.J., Mikkelsen, T.S., Rinn, J.L.: The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nature biotechnology 32(4), 381- 386 (2014) + +<|ref|>text<|/ref|><|det|>[[145, 194, 880, 224]]<|/det|> +[20] Qiu, X., Mao, Q., Tang, Y., Wang, L., Chawla, R., Pliner, H.A., Trapnell, C.: Reversed graph embedding resolves complex single- cell trajectories. Nature methods 14(10), 979- 982 (2017) + +<|ref|>text<|/ref|><|det|>[[145, 234, 880, 264]]<|/det|> +[21] Song, Q., Wang, J., Bar- Joseph, Z.: sctstem: clustering pseudotime ordered single- cell data. Genome Biology 23(1), 1- 17 (2022) + +<|ref|>text<|/ref|><|det|>[[145, 274, 880, 304]]<|/det|> +[22] Ding, J., Lin, C., Bar- Joseph, Z.: Cell lineage inference from snp and scrna- seq data. Nucleic Acids Research 47(10), 56- 56 (2019) + +<|ref|>text<|/ref|><|det|>[[145, 315, 880, 345]]<|/det|> +[23] Ma, Q., Xu, D.: Deep learning shapes single- cell data analysis. Nature Reviews Molecular Cell Biology 23(5), 303- 304 (2022) + +<|ref|>text<|/ref|><|det|>[[145, 355, 880, 399]]<|/det|> +[24] Amodio, M., Van Dijk, D., Srinivasan, K., Chen, W.S., Mohsen, H., Moon, K.R., Campbell, A., Zhao, Y., Wang, X., Venkataswamy, M., et al.: Exploring single- cell data with deep multitasking neural networks. Nature methods 16(11), 1139- 1145 (2019) + +<|ref|>text<|/ref|><|det|>[[145, 409, 880, 440]]<|/det|> +[25] Lin, X., Tian, T., Wei, Z., Hakonarson, H.: Clustering of single- cell multi- omics data with a multimodal deep learning method. Nature communications 13(1), 7705 (2022) + +<|ref|>text<|/ref|><|det|>[[145, 450, 880, 495]]<|/det|> +[26] Wang, J., Ma, A., Chang, Y., Gong, J., Jiang, Y., Qi, R., Wang, C., Fu, H., Ma, Q., Xu, D.: scgnn is a novel graph neural network framework for single- cell rna- seq analyses. Nature communications 12(1), 1882 (2021) + +<|ref|>text<|/ref|><|det|>[[145, 505, 880, 562]]<|/det|> +[27] Hu, J., Li, X., Coleman, K., Schroeder, A., Ma, N., Irwin, D.J., Lee, E.B., Shinohara, R.T., Li, M.: Spagcn: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nature methods 18(11), 1342- 1351 (2021) + +<|ref|>text<|/ref|><|det|>[[145, 573, 558, 588]]<|/det|> +[28] Settles, B.: Active learning literature survey (2009) + +<|ref|>text<|/ref|><|det|>[[145, 599, 805, 615]]<|/det|> +[29] LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. nature 521(7553), 436- 444 (2015) + +<|ref|>text<|/ref|><|det|>[[145, 625, 880, 655]]<|/det|> +[30] Kingma, D.P., Welling, M.: Auto- encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013) + +<|ref|>text<|/ref|><|det|>[[145, 666, 880, 710]]<|/det|> +[31] Creswell, A., White, T., Dumoulin, V., Arulkumaran, K., Sengupta, B., Bharath, A.A.: Generative adversarial networks: An overview. IEEE signal processing magazine 35(1), 53- 65 (2018) + +<|ref|>text<|/ref|><|det|>[[145, 721, 880, 750]]<|/det|> +[32] Bengio, Y., Courville, A., Vincent, P.: Representation learning: A review and new perspectives. IEEE transactions on pattern analysis and machine intelligence 35(8), 1798- 1828 (2013) + +<|ref|>text<|/ref|><|det|>[[145, 761, 880, 790]]<|/det|> +[33] Bengio, Y., Courville, A.C., Vincent, P.: Unsupervised feature learning and deep learning: A review and new perspectives. CoRR, abs/1206.5538 1(2665), 2012 (2012) + +<|ref|>text<|/ref|><|det|>[[145, 801, 880, 844]]<|/det|> +[34] Larsen, A.B.L., Sonderby, S.K., Larochelle, H., Winther, O.: Autoencoding beyond pixels using a learned similarity metric. In: International Conference on Machine Learning, pp. 1558- 1566 (2016). PMLR + +<|ref|>text<|/ref|><|det|>[[145, 856, 880, 900]]<|/det|> +[35] Stephenson, E., Reynolds, G., Botting, R.A., Calero- Nieto, F.J., Morgan, M.D., Tuong, Z.K., Bach, K., Sungnak, W., Worlock, K.B., Yoshida, M., et al.: Single- cell multi- omics analysis of the immune response in covid- 19. Nature medicine 27(5), 904- 916 (2021) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 852, 114]]<|/det|> +[36] McInnes, L., Healy, J., Melville, J.: Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426 (2018) + +<|ref|>text<|/ref|><|det|>[[117, 124, 852, 168]]<|/det|> +[37] Hadjadj, J., Yatim, N., Barnabei, L., Corneau, A., Boussier, J., Smith, N., Peré, H., Charbit, B., Bondet, V., Chenevier- Gobeaux, C., et al.: Impaired type i interferon activity and inflammatory responses in severe covid- 19 patients. Science 369(6504), 718- 724 (2020) + +<|ref|>text<|/ref|><|det|>[[117, 179, 852, 209]]<|/det|> +[38] Ramasamy, S., Subbian, S.: Critical determinants of cytokine storm and type i interferon response in covid- 19 pathogenesis. Clinical microbiology reviews 34(3), 10- 1128 (2021) + +<|ref|>text<|/ref|><|det|>[[117, 219, 852, 264]]<|/det|> +[39] Lee, J.S., Park, S., Jeong, H.W., Ahn, J.Y., Choi, S.J., Lee, H., Choi, B., Nam, S.K., Sa, M., Kwon, J.- S., et al.: Immunophenotyping of covid- 19 and influenza highlights the role of type i interferons in development of severe covid- 19. Science immunology 5(49), 1554 (2020) + +<|ref|>text<|/ref|><|det|>[[117, 274, 850, 304]]<|/det|> +[40] Lee, J.S., Shin, E.- C.: The type i interferon response in covid- 19: implications for treatment. Nature Reviews Immunology 20(10), 585- 586 (2020) + +<|ref|>text<|/ref|><|det|>[[117, 315, 850, 345]]<|/det|> +[41] Pearson, K.: Liii. on lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin philosophical magazine and journal of science 2(11), 559- 572 (1901) + +<|ref|>text<|/ref|><|det|>[[117, 355, 852, 399]]<|/det|> +[42] Menden, K., Marouf, M., Oller, S., Dalmia, A., Magruder, D.S., Kloiber, K., Heutink, P., Bonn, S.: Deep learning- based cell composition analysis from tissue expression profiles. Science advances 6(30), 2619 (2020) + +<|ref|>text<|/ref|><|det|>[[117, 409, 852, 454]]<|/det|> +[43] Chen, Y., Wang, Y., Chen, Y., Cheng, Y., Wei, Y., Li, Y., Wang, J., Wei, Y., Chan, T.- F., Li, Y.: Deep autoencoder for interpretable tissue- adaptive deconvolution and cell- type- specific gene analysis. Nature Communications 13(1), 6735 (2022) + +<|ref|>text<|/ref|><|det|>[[117, 464, 850, 494]]<|/det|> +[44] Tsoucas, D., Dong, R., Chen, H., Zhu, Q., Guo, G., Yuan, G.- C.: Accurate estimation of cell- type composition from gene expression data. Nature communications 10(1), 2975 (2019) + +<|ref|>text<|/ref|><|det|>[[117, 504, 852, 548]]<|/det|> +[45] Plaisier, S.B., Taschereau, R., Wong, J.A., Graeber, T.G.: Rank- rank hypergeometric overlap: identification of statistically significant overlap between gene- expression signatures. Nucleic acids research 38(17), 169- 169 (2010) + +<|ref|>text<|/ref|><|det|>[[117, 559, 850, 589]]<|/det|> +[46] Thomas, P.D., Ebert, D., Muruganujan, A., Mushayahama, T., Albou, L.- P., Mi, H.: Panther: Making genome- scale phylogenetics accessible to all. Protein Science 31(1), 8- 22 (2022) + +<|ref|>text<|/ref|><|det|>[[117, 599, 850, 629]]<|/det|> +[47] Carbon, S., Ireland, A., Mungall, C.J., Shu, S., Marshall, B., Lewis, S., Hub, A., Group, W.P.W.: Amigo: online access to ontology and annotation data. Bioinformatics 25(2), 288- 289 (2009) + +<|ref|>text<|/ref|><|det|>[[117, 639, 852, 697]]<|/det|> +[48] Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., et al.: Gene set enrichment analysis: a knowledge- based approach for interpreting genome- wide expression profiles. Proceedings of the National Academy of Sciences 102(43), 15545- 15550 (2005) + +<|ref|>text<|/ref|><|det|>[[117, 708, 852, 752]]<|/det|> +[49] Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al.: Gene ontology: tool for the unification of biology. Nature genetics 25(1), 25- 29 (2000) + +<|ref|>text<|/ref|><|det|>[[117, 763, 852, 806]]<|/det|> +[50] Gillespie, M., Jassal, B., Stephan, R., Milacic, M., Rothfels, K., Senff- Ribeiro, A., Griss, J., Sevilla, C., Matthews, L., Gong, C., et al.: The reactome pathway knowledgebase 2022. Nucleic acids research 50(D1), 687- 692 (2022) + +<|ref|>text<|/ref|><|det|>[[117, 817, 850, 847]]<|/det|> +[51] Nishida- Aoki, N., Gujral, T.S.: Emerging approaches to study cell- cell interactions in tumor microenvironment. Oncotarget 10(7), 785 (2019) + +<|ref|>text<|/ref|><|det|>[[117, 857, 850, 887]]<|/det|> +[52] Armingol, E., Officer, A., Harismendy, O., Lewis, N.E.: Deciphering cell- cell interactions and communication from gene expression. Nature Reviews Genetics 22(2), 71- 88 (2021) + +<|ref|>text<|/ref|><|det|>[[115, 898, 850, 913]]<|/det|> +[53] Joanito, I., Wirapati, P., Zhao, N., Nawaz, Z., Yeo, G., Lee, F., Eng, C.L., Macalinao, D.C., + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[177, 84, 880, 128]]<|/det|> +Kahraman, M., Srinivasan, H., et al.: Single- cell and bulk transcriptome sequencing identifies two epithelial tumor cell states and refines the consensus molecular classification of colorectal cancer. Nature genetics 54(7), 963- 975 (2022) + +<|ref|>text<|/ref|><|det|>[[147, 138, 880, 183]]<|/det|> +[54] Gao, J., Wu, Z., Zhao, M., Zhang, R., Li, M., Sun, D., Cheng, H., Qi, X., Shen, Y., Xu, Q., et al.: Allosteric inhibition reveals shp2- mediated tumor immunosuppression in colon cancer by single- cell transcriptomics. Acta Pharmaceutica Sinica B 12(1), 149- 166 (2022) + +<|ref|>text<|/ref|><|det|>[[147, 194, 880, 224]]<|/det|> +[55] Cao, Y., Tang, W., Tang, W.: Immune cell infiltration characteristics and related core genes in lupus nephritis: results from bioinformatic analysis. BMC immunology 20, 1- 12 (2019) + +<|ref|>text<|/ref|><|det|>[[147, 234, 880, 278]]<|/det|> +[56] Fortier, A.- M., Asselin, E., Cadrin, M.: Keratin 8 and 18 loss in epithelial cancer cells increases collective cell migration and cisplatin sensitivity through claudin1 up- regulation. Journal of Biological Chemistry 288(16), 11555- 11571 (2013) + +<|ref|>text<|/ref|><|det|>[[147, 289, 880, 319]]<|/det|> +[57] Karantza, V.: Keratins in health and cancer: more than mere epithelial cell markers. Oncogene 30(2), 127- 138 (2011) + +<|ref|>text<|/ref|><|det|>[[147, 330, 880, 359]]<|/det|> +[58] Mann, H.B., Whitney, D.R.: On a test of whether one of two random variables is stochastically larger than the other. The annals of mathematical statistics, 50- 60 (1947) + +<|ref|>text<|/ref|><|det|>[[147, 370, 880, 414]]<|/det|> +[59] Virtanen, P., Gommers, R., Oliphant, T.E., Haberland, M., Reddy, T., Cournapeau, D., Burovski, E., Peterson, P., Weckesser, W., Bright, J., et al.: Scipy 1.0: fundamental algorithms for scientific computing in python. Nature methods 17(3), 261- 272 (2020) + +<|ref|>text<|/ref|><|det|>[[147, 424, 880, 482]]<|/det|> +[60] Ramaswami, G., Yuva- Aydemir, Y., Akerberg, B., Matthews, B., Williams, J., Golczer, G., Huang, J., Huh, D., Burkly, L.C., Engle, S.J., et al.: Transcriptional characterization of ipsc- derived microglia as a model for therapeutic development in neurodegeneration. bioRxiv, 2023- 03 (2023) + +<|ref|>text<|/ref|><|det|>[[147, 493, 880, 537]]<|/det|> +[61] Lahnemann, D., Köster, J., Szczurek, E., McCarthy, D.J., Hicks, S.C., Robinson, M.D., Vallejos, C.A., Campbell, K.R., Beerenwinkel, N., Mahfouz, A., et al.: Eleven grand challenges in single- cell data science. Genome biology 21(1), 1- 35 (2020) + +<|ref|>text<|/ref|><|det|>[[147, 548, 880, 577]]<|/det|> +[62] Pratt, J.W.: Remarks on zeros and ties in the wilcoxon signed rank procedures. Journal of the American Statistical Association 54(287), 655- 667 (1959) + +<|ref|>text<|/ref|><|det|>[[147, 588, 880, 632]]<|/det|> +[63] Wolf, F.A., Hamey, F.K., Plass, M., Solana, J., Dahlin, J.S., Göttgens, B., Rajewsky, N., Simon, L., Theis, F.J.: Paga: graph abstraction reconciles clustering with trajectory inference through a topology preserving map of single cells. Genome biology 20, 1- 9 (2019) + +<|ref|>text<|/ref|><|det|>[[147, 643, 880, 673]]<|/det|> +[64] Nguyen, V.- L., Shaker, M.H., Hüllermeier, E.: How to measure uncertainty in uncertainty sampling for active learning. Machine Learning 111(1), 89- 122 (2022) + +<|ref|>text<|/ref|><|det|>[[147, 684, 880, 713]]<|/det|> +[65] Lewis, D.D.: A sequential algorithm for training text classifiers: Corrigendum and additional data. In: Acm Sigir Forum, vol. 29, pp. 13- 19 (1995). ACM New York, NY, USA + +<|ref|>text<|/ref|><|det|>[[147, 724, 880, 753]]<|/det|> +[66] Dasgupta, S.: Two faces of active learning. Theoretical computer science 412(19), 1767- 1781 (2011) + +<|ref|>text<|/ref|><|det|>[[147, 764, 880, 793]]<|/det|> +[67] Tipping, M.E., Bishop, C.M.: Probabilistic principal component analysis. Journal of the Royal Statistical Society Series B: Statistical Methodology 61(3), 611- 622 (1999) + +<|ref|>text<|/ref|><|det|>[[147, 804, 880, 847]]<|/det|> +[68] Halko, N., Martinsson, P.- G., Tropp, J.A.: Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions. SIAM review 53(2), 217- 288 (2011) + +<|ref|>text<|/ref|><|det|>[[147, 858, 880, 887]]<|/det|> +[69] Martinsson, P.- G., Rokhlin, V., Tygert, M.: A randomized algorithm for the decomposition of matrices. Applied and Computational Harmonic Analysis 30(1), 47- 68 (2011) + +<|ref|>text<|/ref|><|det|>[[147, 899, 880, 913]]<|/det|> +[70] Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[150, 84, 850, 113]]<|/det|> +Prettenhofer, P., Weiss, R., Dubourg, V., et al.: Scikit- learn: Machine learning in python. the Journal of machine Learning research 12, 2825- 2830 (2011) + +<|ref|>text<|/ref|><|det|>[[118, 124, 850, 168]]<|/det|> +[71] MacQueen, J., et al.: Some methods for classification and analysis of multivariate observations. In: Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, vol. 1, pp. 281- 297 (1967). Oakland, CA, USA + +<|ref|>text<|/ref|><|det|>[[118, 179, 850, 238]]<|/det|> +[72] Finak, G., McDavid, A., Yajima, M., Deng, J., Gersuk, V., Shalek, A.K., Slichter, C.K., Miller, H.W., McElrath, M.J., Prlic, M., et al.: Mast: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single- cell rna sequencing data. Genome biology 16(1), 1- 13 (2015) + +<|ref|>text<|/ref|><|det|>[[118, 248, 850, 291]]<|/det|> +[73] Shalek, A.K., Satija, R., Shuga, J., Trombetta, J.J., Gennert, D., Lu, D., Chen, P., Gertner, R.S., Gaublomme, J.T., Yosef, N., et al.: Single- cell rna- seq reveals dynamic paracrine control of cellular variation. Nature 510(7505), 363- 369 (2014) + +<|ref|>text<|/ref|><|det|>[[118, 303, 850, 332]]<|/det|> +[74] Hasanaj, E., Wang, J., Sarathi, A., Ding, J., Bar- Joseph, Z.: Interactive single- cell data analysis using cellar. Nature communications 13(1), 1998 (2022) + +<|ref|>text<|/ref|><|det|>[[118, 343, 850, 372]]<|/det|> +[75] Wolf, F.A., Angerer, P., Theis, F.J.: Scanpy: large- scale single- cell gene expression data analysis. Genome biology 19, 1- 5 (2018) + +<|ref|>text<|/ref|><|det|>[[118, 383, 850, 412]]<|/det|> +[76] Ernst, J., Vainas, O., Harbison, C.T., Simon, I., Bar- Joseph, Z.: Reconstructing dynamic regulatory maps. Molecular systems biology 3(1), 74 (2007) + +<|ref|>text<|/ref|><|det|>[[118, 423, 850, 452]]<|/det|> +[77] Li, D., Ding, J., Bar- Joseph, Z.: Unifan: A tool for unsupervised single- cell clustering and annotation. Journal of Computational Biology 29(11), 1229- 1232 (2022) + +<|ref|>text<|/ref|><|det|>[[118, 463, 850, 492]]<|/det|> +[78] Lopez, R., Regier, J., Cole, M.B., Jordan, M.I., Yosef, N.: Deep generative modeling for single- cell transcriptomics. Nature methods 15(12), 1053- 1058 (2018) + +<|ref|>text<|/ref|><|det|>[[118, 504, 850, 533]]<|/det|> +[79] Risso, D., Perraudeau, F., Gribkova, S., Dudoit, S., Vert, J.- P.: A general and flexible method for signal extraction from single- cell rna- seq data. Nature communications 9(1), 284 (2018) + +<|ref|>text<|/ref|><|det|>[[118, 544, 850, 588]]<|/det|> +[80] Berge, K., Perraudeau, F., Soneson, C., Love, M.I., Risso, D., Vert, J.- P., Robinson, M.D., Dudoit, S., Clement, L.: Observation weights unlock bulk rna- seq tools for zero inflation and single- cell applications. Genome biology 19, 1- 17 (2018) + +<|ref|>text<|/ref|><|det|>[[118, 599, 850, 628]]<|/det|> +[81] Kullback, S., Leibler, R.A.: On information and sufficiency. The annals of mathematical statistics 22(1), 79- 86 (1951) + +<|ref|>text<|/ref|><|det|>[[118, 639, 850, 668]]<|/det|> +[82] Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., Philip, S.Y.: A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems 32(1), 4- 24 (2020) + +<|ref|>text<|/ref|><|det|>[[118, 680, 850, 708]]<|/det|> +[83] Scarselli, F., Gori, M., Tsoi, A.C., Hagenbuchner, M., Monfardini, G.: The graph neural network model. IEEE transactions on neural networks 20(1), 61- 80 (2008) + +<|ref|>text<|/ref|><|det|>[[118, 720, 850, 763]]<|/det|> +[84] Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: Pytorch: An imperative style, high- performance deep learning library. Advances in neural information processing systems 32 (2019) + +<|ref|>text<|/ref|><|det|>[[118, 775, 850, 804]]<|/det|> +[85] Maas, A.L., Hannun, A.Y., Ng, A.Y., et al.: Rectifier nonlinearities improve neural network acoustic models. In: Proc. Icml, vol. 30, p. 3 (2013). Atlanta, GA + +<|ref|>text<|/ref|><|det|>[[118, 815, 850, 858]]<|/det|> +[86] He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: Surpassing human- level performance on imagenet classification. In: Proceedings of the IEEE International Conference on Computer Vision, pp. 1026- 1034 (2015) + +<|ref|>text<|/ref|><|det|>[[118, 870, 850, 912]]<|/det|> +[87] Kleyman, M., Sefer, E., Nicola, T., Espinoza, C., Chhabra, D., Hagood, J.S., Kaminski, N., Ambalavanan, N., Bar- Joseph, Z.: Selecting the most appropriate time points to profile in high- throughput studies. Elife 6, 18541 (2017) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 880, 115]]<|/det|> +[88] Fang, Z., Liu, X., Peltz, G.: Gseapy: a comprehensive package for performing gene set enrichment analysis in python. Bioinformatics 39(1), 757 (2023) + +<|ref|>text<|/ref|><|det|>[[145, 125, 880, 169]]<|/det|> +[89] Jin, S., Guerrero- Juarez, C., Zhang, L., Chang, I., Ramos, R., Kuan, C., Myung, P., Pikus, M., Nie, Q.: Inference and analysis of cell- cell communication using CellChat. Nat. Commun. 12, 1088 (2021) + +<|ref|>text<|/ref|><|det|>[[145, 180, 880, 223]]<|/det|> +[90] Cao, J., Spielmann, M., Qiu, X., Huang, X., Ibrahim, D.M., Hill, A.J., Zhang, F., Mundlos, S., Christiansen, L., Steemers, F.J., et al.: The single- cell transcriptional landscape of mammalian organogenesis. Nature 566(7745), 496- 502 (2019) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 92, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 376, 150]]<|/det|> +scSemiProfilersupplementary.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/images_list.json b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..edc77606280e735a519d13e6530d9ff4894f1ca6 --- /dev/null +++ b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/images_list.json @@ -0,0 +1,475 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Workflow of the high throughput approach used in this work. \\(\\mathrm{Pd}_n\\mathrm{H}_n\\) ( \\(n = 0\\) , 1 or 2 and \\(m = 0\\) , 1 or 2) is used as an illustration. Starting with the zeolite material, structures are generated by identifying the unique Al position(s) in step 1 (generating structures from 1 to w), followed by identifying atoms in the QM and MM region for each structure in step 2. In step 3, possible charge-exchange sites are enumerated and then further evaluated in step 4 using QM/MM calculations at the GGA level. The 5 most favorable exchange sites from step 4 for each unique Al position are further evaluated at the hGGA level of theory in step 5. Finally, all structures from step 5 are ordered based on their energy stability, yielding the most energetically favorable sites for the exchange in the zeolite in step 6", + "footnote": [], + "bbox": [ + [ + 115, + 275, + 878, + 541 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Visual illustration of atoms positions surrounding the \\(\\mathrm{Si}^{T - atom}\\) . \\(\\mathrm{Si}^{T - atom}\\) is replaced with an Al atom to generate a univalent structure. For divalent structures, \\(\\mathrm{Si}^{NN}\\) , \\(\\mathrm{Si}^{NNN}\\) , and \\(\\mathrm{Si}^{NNNN}\\) atoms are identified and then desired structures are generated by replacing either the \\(\\mathrm{Si}^{NNN}\\) or \\(\\mathrm{Si}^{NNNN}\\) and \\(\\mathrm{Si}^{T - atom}\\) atom with Al atoms", + "footnote": [], + "bbox": [ + [ + 202, + 220, + 825, + 686 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Formation Energy of Pd+H+ on CHA. Each bar represents unique Al location(s). The color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in an NNN configuration or the isolated site and striped bars refer to the NNNN configuration.", + "footnote": [], + "bbox": [ + [ + 120, + 276, + 870, + 635 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Formation energy of \\(\\mathrm{Pd^{+2}}\\) on CHA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in NNN configuration and striped bars refer to NNNN configuration. The two most stable sites (CHA-7 and CHA-3) are illustrated in Figure 5.", + "footnote": [], + "bbox": [ + [ + 117, + 272, + 860, + 633 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: QM atoms of the optimized calculations. Color coding: red=Oxygen, white=Hydrogen, blue=Palladium and beige=Silicon", + "footnote": [], + "bbox": [ + [ + 120, + 90, + 861, + 425 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: Formation energy of \\(\\mathrm{Pd^{+2}}\\) on BEA. Each bar represents unique Al locations. Color coding refers to the type of Al pairs in the zeolite matrix. Patterned bars refer to Al pairs in NNNN positions while solid bars refer to Al pairs in NNN positions. The four most stable sites are illustrated in Figure 5.", + "footnote": [], + "bbox": [ + [ + 120, + 400, + 865, + 680 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7: Formation energy of \\(\\mathrm{Pd}^{+}\\) on BEA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. In Al pairs, patterned bars refer to Al pairs in NNNN positions and solid colors refer to Al pairs in NNN positions. The most stable site (BEA-65) is illustrated in Figure 5.", + "footnote": [], + "bbox": [ + [ + 117, + 491, + 872, + 747 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8: Comparison between favorable Al pair configurations for \\(\\mathrm{Pd}^{+2}\\) and \\(\\mathrm{Co}^{+2}\\)", + "footnote": [], + "bbox": [ + [ + 180, + 470, + 800, + 787 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9: NO Binding energy versus Pd formation energy on CHA. The color coding refers to the type of Al in the \\(n\\) -MR. The marker shapes (□ and \\(\\circ\\) ) represent \\(\\mathrm{Pd^{+2}}\\) and \\(\\mathrm{Pd^{+} / Pd^{+}H^{+}}\\) , respectively. The black and blue lines are based on fitting data for \\(\\mathrm{Pd^{+2}}\\) and \\(\\mathrm{Pd^{+} / Pd^{+}H^{+}}\\) , respectively. Filled markers are used for Al pairs in NNNN configuration and half-filled markers are for isolated Al or Al in NNN configurations", + "footnote": [], + "bbox": [ + [ + 120, + 92, + 872, + 394 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S1: Images of the QM atoms in four BEA structures. BEA-2 structure is shown in a. alongside a similar structure in b. BEA-11 is shown in c. alongside a similar structure in d. The structures similar to BEA-2 and BEA-11 (b. and d., respectively) are not exactly identical, as calculated by the nuclear repulsion energy, but share the same connectivity to surrounding Si and O atoms (types of MR). The same color coding as in Figure 5.", + "footnote": [], + "bbox": [ + [ + 235, + 95, + 780, + 512 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S2: Example of two BEA clusters (a. and b.) where Al atoms are not in the same MR and are on opposite sides of the open cage. The same color coding as in Figure 5, however, for clarity purposes, Al atoms are enlarged and are shown in green", + "footnote": [], + "bbox": [ + [ + 230, + 92, + 760, + 360 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure S3: Flow diagram for determining if Al atom is in an X MR (where X refers to the length of the MR). First, starting with an Al atom, Si/Al neighbors (NN, NNN, NNNN) are identified similar to the procedure described in Figure 2. This is done up to the X+1 neighbor. Based on the neighbors, lists are developed (an example of a list is [Al, NN1, NNN11, ..], where NN1 refers to the first NN of the starting Al and NNN11 is the first NN to NN1). For each list, the following checks are made: first and last element of the list are the same (the starting Al atom), no element in the list (with the exception of the starting Al atom) appears twice in the list, and no sub-list of the list form a MR smaller than X. If a list passes those checks, then Al is part of XMR.", + "footnote": [], + "bbox": [ + [ + 120, + 100, + 870, + 275 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Figure S4: Impact of the number of atoms selected in the QM region on the convergence of the QM/MM calculations ( \\(\\Delta E_{form}\\) is defined in equation 2). Calculations were done using \\(\\omega\\) B97X-D functional.", + "footnote": [], + "bbox": [ + [ + 216, + 100, + 761, + 424 + ] + ], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S5: Example of a. Pd cation and b. proton placement near the four oxygens neighboring Al atom. For clarity, only Si in NN position and O neighboring Al are shown. The position of cation/proton is determined by finding the middle distance between neighboring oxygen, and then adding a displacement from the Al atom (usually 1-1.5 A). The same color coding as in Figure 5.", + "footnote": [], + "bbox": [ + [ + 140, + 106, + 864, + 328 + ] + ], + "page_idx": 37 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "Figure S6: CHA-2 optimum proton locations search results. Each data represents the optimum energy based on a different initial position of the protons. The x-axis refers to the index of the structure in the database in the SI and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14). Calculations were done using the \\(\\omega\\) B97X-D functional.", + "footnote": [], + "bbox": [ + [ + 123, + 99, + 884, + 455 + ] + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_2.jpg", + "caption": "Figure S7: CHA-2 optimum proton locations search results. Each data represents a different initial position of the proton. The x-axis refers to the O-O distance (where the oxygen is the atom H adsorbs on) in the optimized structure and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14)", + "footnote": [], + "bbox": [ + [ + 213, + 100, + 768, + 479 + ] + ], + "page_idx": 40 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_3.jpg", + "caption": "Figure S8: CHA-2 optimum \\(\\mathrm{Pd^{+}H^{+}}\\) locations search results. Each data represents a different initial position of \\(\\mathrm{Pd^{+}H^{+}}\\) . The x-axis refers to the index of the structure in the database and the y-axis is the energy relative to the most stable structure (CHA-2-Pd+H+-3). Calculations were done using the \\(\\omega \\mathrm{B97X - D}\\) functional.", + "footnote": [], + "bbox": [ + [ + 117, + 98, + 872, + 465 + ] + ], + "page_idx": 41 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S9: QM images of the optimized geometry of a. CHA-3-Pd+H+-21 and b. CHA-3-Pd+H+-17. Despite only a change in the proton position, there is \\(>0.4 \\mathrm{eV}\\) in energy difference between the two optimized geometries. The same color coding as in Figure 5", + "footnote": [], + "bbox": [ + [ + 170, + 110, + 857, + 316 + ] + ], + "page_idx": 42 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S10: QM region images of the optimized geometry of a. Pd+H+ on CHA-7 b. Pd+H+ on CHA-3 c. Pd+H+ on CHA-8 d. Pd+H+ on CHA-6 e. Pd+H+ on CHA-9 f. Pd+H+ on CHA-12 g. Pd+2 on CHA-5 and h. Pd+2 on CHA-10. The same color coding as in Figure 5.", + "footnote": [], + "bbox": [ + [ + 135, + 90, + 884, + 425 + ] + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_4.jpg", + "caption": "Figure S11: \\(\\Delta E_{form}\\) as a function of the distance between Pd and H in Pd+H+ optimized geometry on CHA (□) and BEA (○)", + "footnote": [], + "bbox": [ + [ + 118, + 99, + 875, + 525 + ] + ], + "page_idx": 44 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S12: QM region images of optimized geometry of a. Pd+2 on BEA-8 b. Pd+2 on BEA-55 c. Pd+H+ on BEA-33 d. Pd+H+ on BEA-55. The same color coding as in Figure 5.", + "footnote": [], + "bbox": [ + [ + 118, + 95, + 860, + 570 + ] + ], + "page_idx": 45 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_5.jpg", + "caption": "Figure S13: Parity plot of the calculated DFT energy on Al pairs (that do not share a MR) and the predicted energy based on the respective isolated Al site (the most stable of the two). All calculations were done on BEA with \\(\\mathrm{Pd^{+}H^{+}}\\) as the cation", + "footnote": [], + "bbox": [ + [ + 125, + 108, + 855, + 494 + ] + ], + "page_idx": 46 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S14: Image of the QM region on optimized BEA calculations a. BEA-51 (Pd+H+) b. BEA-78 (Pd+H+) c. BEA-52 (Pd+H+). The same color coding as in Figure 5.", + "footnote": [], + "bbox": [ + [ + 144, + 100, + 860, + 560 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure S15: Cluster models of a. T696 CHA and b. T810 BEA. The same color coding as in Figure 5", + "footnote": [], + "bbox": [ + [ + 120, + 98, + 850, + 375 + ] + ], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 50, + 97, + 940, + 410 + ] + ], + "page_idx": 50 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 55, + 55, + 940, + 696 + ] + ], + "page_idx": 51 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 50, + 56, + 930, + 480 + ] + ], + "page_idx": 52 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 46, + 50, + 943, + 490 + ] + ], + "page_idx": 53 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 60, + 54, + 930, + 450 + ] + ], + "page_idx": 54 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 50, + 575, + 944, + 912 + ] + ], + "page_idx": 54 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7", + "footnote": [], + "bbox": [ + [ + 48, + 72, + 936, + 380 + ] + ], + "page_idx": 55 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8", + "footnote": [], + "bbox": [], + "page_idx": 55 + } +] \ No newline at end of file diff --git a/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91.mmd b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6e2e92d833f29c9c91799bdfc273b8129de93c48 --- /dev/null +++ b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91.mmd @@ -0,0 +1,514 @@ + +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput quantum chemistry workflow + +Hassan Aljama University of California, Berkeley Martin Head- Gordon ( \(\boxed{\text {mhg@cchem.berkeley.edu}}\) ) University of California, Berkeley https://orcid.org/0000- 0002- 4309- 6669 + +Alexis Bell Joint Center for Artificial Photosynthesis, Lawrence Berkeley National Laboratory https://orcid.org/0000- 0002- 5738- 4645 + +## Article + +Keywords: Cation Exchanged- zeolites, Density Functional Theory, High Throughput Screening, NOx Capture + +Posted Date: May 7th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 396201/v1 + +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on May 25th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29505- z. + +<--- Page Split ---> + +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput quantum chemistry workflow + +Hassan Aljama,† Martin Head- Gordon,\*,\$ and Alexis T. Bell\*,\$ + +†Department of Chemical and Biomolecular Engineering, University of California, Berkeley, California + +‡Department of Chemistry, University of California, Berkeley, California + +E- mail: mhg@cchem.berkeley.edu; alexbell@berkeley.edu + +## Abstract + +Cation exchanged- zeolites are functional materials with a wide range of applications from catalysis to sorbents. They present a challenge for computational studies using density functional theory due to the numerous possible active sites. From Al configuration, to placement of extra framework cation(s), to potentially different oxidation states of the cation, accounting for all these possibilities is not trivial. To make the number of calculations more tractable, most studies focus on a few active sites. We attempt to go beyond these limitations by implementing a workflow for a high throughput screening, designed to systematize the problem and exhaustively search for feasible active sites. We use Pd- exchanged CHA and BEA to illustrate the approach. After conducting thousands of individual calculations, we identify the sites most favorable for the Pd cation and discuss the results in detail. The high throughput screening identifies many energetically favorable sites that are non- trivial. Lastly, we employ these results to examine NO adsorption in Pd- exchanged CHA, which is a promising + +<--- Page Split ---> + +passive \(\mathrm{NO}_{x}\) adsorbent (PNA) during the cold start of automobiles. The results shed light on critical active sites for NOx capture that were not previously studied. + +## Introduction + +Zeolites are microporous aluminosilicates composed of corner- sharing \(\mathrm{TO}_{4}\) tetrahedra ( \(\mathrm{T} =\) Si or Al). Substitution of trivalent Al for tetravalent Si creates a charge imbalance that is compensated by a proton or a metal cation. If the charge compensating cation is a proton, the zeolites becomes a strong solid acid that is an effective catalyst for promoting a wide variety of hydrocarbons reaction as well as the synthesis of a broad range of organic compounds. Metal cation- exchanged zeolites can also serve as adsorbents and catalysts [1- 5]. + +In recent years, considerable insights into zeolite catalyzed reactions have come from the application of quantum chemical analyses of the free energy and enthalpy landscapes governing the progress of chemical reactions [6, 7]. A significant challenge for such studies is selection of the structure of the active site. To appreciate the issue, we must first recall that there are over 200 known zeolites structures, many of which have T sites occupied by Si and Al having symmetries differing from one another. Each of the charge- exchange sites is associated with a framework Al atom, but the distribution of Al is generally not known and is thought to be controlled by the kinetics of zeolite synthesis. A further complication is that the charge- exchange site involves an Al atom bonded to four oxygen atoms, any one of which can accommodate the proton. Further complexity arises when extra- framework metal cations replace protons. Monovalent cations are usually large enough that they bridge two framework oxygen atoms, while bivalent cations can interact with two or three Al sites simultaneously. + +It is, therefore, evident that the selection of a representative cation site is challenging and that full exploration of the chemistry on all possible sites is computationally formidable because of cost. For these reasons, many higher level theoretical studies of zeolite- catalyzed + +<--- Page Split ---> + +reactions[8- 11] using density functional theory (DFT) have chosen to focus on a few active sites selected on the basis of limited experimental evidence and/or physical intuition. While this choice leads to a more tractable set of calculations, the downside is that potentially important active sites might be missed because they are difficult to identify experimentally or are not physically intuitive. Indeed, growing numbers of DFT studies have progressed from studying single T- sites[12, 13] to address greater complexity[14, 15]. + +With this trend in mind, the objective of this work is to present and apply a DFT- based computational framework to identify the energetically most favorable adsorption sites (cation or proton) on any zeolite using a systematic high- throughput approach. The underlying calculations employ a hybrid quantum mechanics/molecular mechanics (QM/MM) models[16, 17] to capture extended environment around each active site, and high quality DFT[18] to attain acceptable accuracy. Our approach starts by evaluating the possible zeolite structures that arise for different Si/Al ratios, then searching for energetically favorable cation/proton(s) adsorption sites for each structure using a lower level of theory (e.g. smaller basis set and cheaper functional) as a filter. The most favorable sites are then evaluated at a higher level of theory. We note that while there are previous high throughput approaches in zeolites, they were largely focused on grand canonical Monte Carlo (GCMC) simulations [19- 23]. + +To illustrate our approach, we use the challenging example of the siting of extraframe- work Pd ions in Al- doped chabazite (CHA) and beta (BEA). Pd- CHA and Pd- BEA were chosen because they are good candidates for passive NO adsorbers (PNAs) that can be used to trap the emissions of NO from automobile exhaust during cold startup and before the three- way catalytic converter becomes effective [24- 26]. The speciation of Pd in these materials is recognized as a challenging, and still controversial problem [27, 28]. We limit our discussion to the following Pd species: Pd+ associated with isolated charge- exchange sites, Pd+H+ pairs associated with two charge- exchange sites involving next nearest neighbor (NNN) or next- next nearest neighbor (NNN) pairs of Al atoms, and Pd+2 associated + +<--- Page Split ---> + +with two charge- exchange sites also involving NNN or NNNN Al pairs. The goal is to provide a methodology that is easily applicable and transferable to any adsorption/catalysis problem with any zeolite, thereby shifting the focus of zeolite studies from a limited number of specific sites to a more systematic approach and allowing a more exhaustive exploration of the descriptor space. After establishing a large set of feasible charge exchange sites for Pd in CHA and BEA, we conclude by assessing the performance of the sites for NO adsorption on Pd- CHA, with some interesting results. + +## High Throughput Approach + +Figure 1 shows the workflow for the high throughput approach to determine which charge- exchange sites are most favorable for accommodating the charge- compensating cation and/or proton. Briefly, Al atom(s) are first introduced at different tetrahedral sites in the bare \(\mathrm{Si}_{x}\mathrm{O}_{y}\) zeolite, generating structures with unique (i.e. distinguishable) Al positions. For each unique Al arrangement, distinctions are made between atoms in the QM region, which are expected to be active in the adsorption process, and the surrounding atoms in the MM region. To determine the most energetically favorable location for accommodating the extra framework cation, potential cation- exchanged sites are first enumerated and then surveyed by QM/MM calculations at the GGA level of theory, followed by further evaluation of the top 5 most stable sites at the hGGA level of theory. Lastly, results for all structures are compared to determine the structures where Al position(s) yield the most stable charge- exchange site. Each step is discussed in detail in the next few sections. + +## Distinguishable Al locations + +The first step of the approach (Figure 1) is determining the set of distinguishable Al sites in the zeolite structure. Univalent structures could be easily generated by replacing each unique Si T- atom in the zeolite by an Al atom. However, generating Al pairs is not trivial + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: Workflow of the high throughput approach used in this work. \(\mathrm{Pd}_n\mathrm{H}_n\) ( \(n = 0\) , 1 or 2 and \(m = 0\) , 1 or 2) is used as an illustration. Starting with the zeolite material, structures are generated by identifying the unique Al position(s) in step 1 (generating structures from 1 to w), followed by identifying atoms in the QM and MM region for each structure in step 2. In step 3, possible charge-exchange sites are enumerated and then further evaluated in step 4 using QM/MM calculations at the GGA level. The 5 most favorable exchange sites from step 4 for each unique Al position are further evaluated at the hGGA level of theory in step 5. Finally, all structures from step 5 are ordered based on their energy stability, yielding the most energetically favorable sites for the exchange in the zeolite in step 6
+ +<--- Page Split ---> + +due to the 3D structure of zeolites. While this can be done manually for CHA, since it only has one unique T- site, it is more challenging to do for BEA, which has 9 T- sites. Hence, we developed an automated method to generate all possible Al pairs. We focused in this work on pairs in the NNN (where the two framework Al are separated by a single Si atom) and the NNNN (where the two framework Al are separated by two Si atoms). Al pairs in nearest neighbor (NN) position were avoided in order to satisfy the Loewenstein rule [29]. Pairs further apart than NNNN were not considered since Al pairs further away should behave as isolated Al atoms. We note that this approach is easily applicable to any zeolite material and can be extended beyond NNNN pairs. + +Our approach uses a molecular graph that identifies the connectivity of each atom in the cluster [30, 31] (visually illustrated in Figure 2). Each Al pair structure is generated by first identifying a unique T- site (Si \(T^{- atom}\) ). Based on the connectivity of the Si \(T^{- atom}\) , its NN Si atoms are identified (Si \(N^{NN}\) ). The Si atoms in NNN (Si \(N^{NN}\) ) are identified by finding the next neighbor of Si \(N^{NN}\) (excluding the original Si \(T^{- atom}\) and Si \(N^{NN}\) ). Each NNN Al pairs are generated by replacing Si \(T^{- atom}\) and one of the Si \(N^{NN}\) with Al atoms. For the NNNN Al pairs, the next neighbor of the Si \(N^{NN}\) atom are identified (Si \(N^{NNN}\) ) (excluding Si \(T^{- atom}\) , Si \(N^{NN}\) and Si \(N^{NN}\) ). Similarly, Si \(T^{- atom}\) and each one of the Si \(N^{NNN}\) atoms are then replaced with Al atoms to generate the NNNN Al pairs. In this manuscript, each structure with a unique Al location is identified by a unique index after the zeolite name (e.g. CHA- 13) (full details on the xyz coordinates for the structures are available in the Supporting Information (SI)). + +Since the procedure described above can produce duplicate structures, we relied on calculating the nuclear repulsion energy of all generated pairs to eliminate duplicates. The method was verified using a BEA unit cell (which contains 36 T- atoms) and generating 36 different structures by replacing each Si T- atom with an Al atom. By using the nuclear repulsion energy, we were able to recover the 9 unique T- sites. Overall, this results in 26 and 212 unique structures for CHA and BEA, respectively. In order to make the number + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Visual illustration of atoms positions surrounding the \(\mathrm{Si}^{T - atom}\) . \(\mathrm{Si}^{T - atom}\) is replaced with an Al atom to generate a univalent structure. For divalent structures, \(\mathrm{Si}^{NN}\) , \(\mathrm{Si}^{NNN}\) , and \(\mathrm{Si}^{NNNN}\) atoms are identified and then desired structures are generated by replacing either the \(\mathrm{Si}^{NNN}\) or \(\mathrm{Si}^{NNNN}\) and \(\mathrm{Si}^{T - atom}\) atom with Al atoms
+ +<--- Page Split ---> + +of calculations more tractable, we further reduced the number of candidates by eliminating structures that share the same connectivity (i.e., the Al atoms in structures share the same types of \(n\) - membered- ring ( \(n\) - MR)). Although in those cases structures are not exactly the same, they are structurally very similar (Figure S1). We also did not evaluate structures for BEA for which Al pairs do not share the same MR and are on opposite sides of the open cage (Figure S2). This reduced the final numbers to 82 structures for BEA and 12 structures for CHA. + +## Selection of the QM and MM regions + +The second part of the workflow (Figure 1) defines the atoms that comprise the QM region and the surrounding MM atoms in the cluster. Appropriate choice of the QM region is critical for the QM/MM calculations since the number of atoms must be sufficient to ensure the calculations converge, but not too large in order to avoid significant computational cost. For structures with an isolated Al atom, the Si and O atoms that create a 4,5 or 6MR with the Al atom are included in the QM atoms. If the structure contains an Al pair, and if the two pairs form a 4,5,6,7 or 8MR, we include all Si and O atoms that are part of the \(n\) - MR. Also, for those structures, we include Si and O atoms that are part of a 4 or 5MR with either of the two Al atoms. Atoms comprising a MR are identified following the procedure described in Figure S3. The QM region includes on average 54 atoms, excluding hydrogen atoms used to terminate the Si atoms. The QM regions used here are larger than used in previous work [11, 32- 35] and we found it to be more than sufficient for the calculations to converge (Figure S4). Finally, for the calculations of NO adsorption in Pd- CHA, the number of QM atoms was extended as needed to account for potential interactions of NO with other Si/O atoms in the framework. + +<--- Page Split ---> + +## Enumeration of charge-exchange sites + +Enumeration of charge- exchange sitesFor each structure with a unique Al arrangement, we survey the energy landscape by placing the adsorbate(s) at multiple initial positions near each oxygen atom neighboring Al atom(s), as shown in Figure S5. For structures with an isolated Al atom, this requires only 4 calculations for either \(\mathrm{Pd^{+}}\) or \(\mathrm{H^{+}}\) . For structures with an Al pair, finding the optimum position of a \(\mathrm{Pd^{+2}}\) requires 8 calculations per structure, and finding the optimum position for \(\mathrm{H^{+}H^{+}}\) requires 16 total calculations. The most complicated case is for a \(\mathrm{Pd^{+}H^{+}}\) site near an Al pair, where in addition to surveying for \(\mathrm{Pd^{+}}\) charge- exchange site, a proton must be present near the opposite Al atom to compensate for the missing charge. This situation requires a total of 32 calculations per structure. We note that these numbers reflect the maximum number of calculations attempted. For some very unfavorable initial positions, the calculations did not converge. This is mostly limited to the GGA level search and happens in less than \(10\%\) of the total calculations (for the most part, this is due to placing \(\mathrm{Pd}\) at the center of a 4MR). The extended number of initial positions in the scheme inevitably leads to some poor initial conditions. All generated structures and optimized geometries are available in the SI. + +## Survey of charge-exchange sites + +Survey of charge- exchange sitesThe major part of the computational cost in the workflow (Figure 1) is associated with searching for the global minimum energy of the charge- exchange site, steps 4 and 5. For zeolites with multiple T- sites, the number of calculations is significant. Carrying out all the calculations at the range- separated hybrid functional level of theory is intractable; however, it has been shown previously that this level of theory is needed to reach close agreement with experimental values [36]. Accordingly, we first use the B97- D3 exchange functional (which is at the GGA level) as a filter to determine the 5 most favorable exchange positions for the cation/proton(s) per each unique Al arrangement. For those 5 candidates, further calculations are done at the range- separated hybrid level using the \(\omega\) B97X- D exchange functional to determine the most energetically favorable position of the cations per each unique Al + +<--- Page Split ---> + +arrangement. We tested this approach on a number of structures by comparing the results between doing the full calculations using only \(\omega \mathrm{B97X - D}\) to the approach described earlier. We found this approach to yield virtually the same results with significant reduction in computational cost (Table S1). + +In order to compare the stability of the Pd cation at different cation- exchanged sites, we calculated the energy of reaction ( \(\Delta E_{form}\) ) for the following two reactions: + +\[\mathrm{Pd}_{(g)} + \mathrm{H}^{+}\mathrm{Z}^{-} + \frac{1}{4}\mathrm{O}_{2(g)}\rightarrow \mathrm{Pd}^{+}\mathrm{Z}^{-} + \frac{1}{2}\mathrm{H}_{2}\mathrm{O}_{(g)} \quad (1)\] + +\[\mathrm{Pd}_{(g)} + \mathrm{H}^{+}\mathrm{H}^{+}\mathrm{Z}^{-2} + \frac{x}{4}\mathrm{O}_{2(g)}\rightarrow \mathrm{Pd}^{+x}\mathrm{H}^{+}_{2 - x}\mathrm{Z}^{-2} + \frac{x}{2}\mathrm{H}_{2}\mathrm{O}_{(g)} \quad (2)\] + +Equation 1 is used for an isolated Al zeolite and equation 2 for structures with an Al pair. \((x)\) is the oxidation state of Pd (either 1 or 2), \((\mathrm{Pd}_{(g)})\) is a gas phase Pd atom, \((\mathrm{H}^{+})\) is the compensating proton, \((\mathrm{Z})\) is the charged zeolite framework, \((\mathrm{H}_{2}\mathrm{O}_{(g)})\) and \((\mathrm{O}_{2(g)})\) are water and oxygen in the gas phase, and \((\mathrm{Pd}^{+x})\) is the Pd adsorbed in the zeolite framework. Equations 1 and 2 allow comparing the relative stability of sites with different Al configurations and oxidation states by using a consistent reference. The equations also rely on using a Brønsted site as a reference. This eliminates the impact of the thermodynamics of Al placement in the zeolite (which is kinetically driven during the synthesis of the zeolite [37]). For NO adsorption on Pd- exchanged CHA, we use the following equation to calculate the NO adsorption energy ( \(\Delta E_{\mathrm{NO}}\) ): + +\[\Delta E_{\mathrm{NO}} = E_{\mathrm{Pd}*\mathrm{NO}*} - E_{\mathrm{Pd}*} - E_{\mathrm{NO}_{(g)}} \quad (3)\] + +where \((E_{\mathrm{Pd}*\mathrm{NO}*})\) is the total energy of NO adsorbed on the Pd- exchanged zeolite in the DFT calculation, \((E_{\mathrm{Pd}*})\) is the total energy of the Pd adsorbed on the zeolite framework and \((E_{\mathrm{NO}_{(g)}})\) is the total energy of NO in the gas phase. \(\mathrm{Pd}*\) refers to either \(\mathrm{Pd}^{+}, \mathrm{Pd}^{+}\mathrm{H}^{+}\) , or \(\mathrm{Pd}^{+2}\) . + +<--- Page Split ---> + +## Results and Discussion + +Results of Pd cation siting in CHA are discussed first followed by a discussion of cation- exchanged BEA. A comparison between the two types of zeolites is then made. Finally, we apply our results to NO adsorption on Pd- exchanged CHA. + +## Pd-exchanged CHA + +We start by reporting our results for the location of Pd cations exchanged into CHA [27, 28]. CHA has a single T- site, which limits the number of unique Al pairs. As mentioned previously, equations 1 and 2 are used to evaluate the stability of the sites. This requires, for each unique Al arrangement, finding the minimum energy of the compensating protons and the Pd cation. An example of the search results for the optimum proton location (based on sampling different initial positions for a given Al siting), is shown in Figure S6. Some locations can be more favorable than others by as much as 1 eV. Based on an examination of the stability of the 12 different Al arrangements in CHA, we do not find a clear indication of why certain positions stabilize protons more than others. One descriptor we find useful is the distance between the oxygen atoms where the protons adsorb ( \(\mathrm{d}_{O - O}\) ) (Figure S7). If the oxygen atoms are too close or too far (relative to Al- Al distance), the associated 2 proton configuration is not favorable. Intermediate distance almost always yields the most favorable arrangement. This descriptor can reduce the number of required calculations; however, it is not a substitute for performing the search through an approach such as the high throughput screening employed here, especially since many candidates have similar values of \(\mathrm{d}_{O - O}\) . + +In addition, sample search results for the global minimum of \(\mathrm{Pd^{+}H^{+}}\) exchange are shown in Figure S8. As for the proton case discussed above, there is a large variance of the results depending on the initial position of the Pd cation and the proton. It is important to also note that the relative energy is sensitive to the proton position. For example, CHA- 3- \(\mathrm{Pd^{+}H^{+}}\) - 17 and CHA- 3- \(\mathrm{Pd^{+}H^{+}}\) - 21 both have the Pd at the center of the 6MR; however, the latter is 0.4 + +<--- Page Split ---> + +eV less stable due to the proton occupying a different location (Figure S9). + +A summary of the energies of \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) in CHA, comparing unique Al positions, is given in Figures 3 and 4, respectively. In both cases, the stability of the Pd cation is heavily dependent on the Al positions, with energies varying by as much as several eV. The range of energies in \(\mathrm{Pd^{+}H^{+}}\) is much more closely spaced ( \(< 0.8 \mathrm{eV}\) ) compared to \(\mathrm{Pd^{+2}}\) (around \(3 \mathrm{eV}\) ). Al pairs in the 6MR arrangements, especially in the NNNN position, provide the most favorable host for the Pd cation, in which case the Pd resides at the center of the 6MR (Figure 5a and b and Figure S10a and b). This finding is consistent with other recent studies [27, 28]. This geometry provides the most number of oxygen atoms in close proximity to the cation, but not too close. Surprisingly, the Pd cation is then most stable either at the isolated site or when the two Al pairs do not share the same ring. In these cases, the Pd cation mostly resides at the center of the 6MR (Figure S10c- f). For both \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , Al pairs in the 8MR and 4MR provide unfavorable arrangements, especially for the latter. In the 8MR (Figure S10g and h), unlike the 6MR, the two Al atoms are farther apart and there is a lack of neighboring Si/O atoms to provide orbital overlap to stabilize the cation. In the case of the 4MR, the atoms are too closely spaced. + +Despite some attempts to do, we were not able to find a simple descriptor related to \(\Delta E_{form}\) (e.g. Al- Al distance and Si/O atoms in close proximity) for the data contained in Figures 3 and 4. For \(\mathrm{Pd^{+}H^{+}}\) , in almost all calculations, a minimum distance of \(4 \mathrm{\AA}\) separates the Pd cation and the proton in the optimized structure (Figure S11). This indicates repulsive interaction between the two cations at shorter distances. + +Comparing \(\Delta E_{form}\) between \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) on CHA shows all Al pairs arrangements are more favorable to \(\mathrm{Pd^{+}H^{+}}\) compared to \(\mathrm{Pd^{+2}}\) , with the exceptions of Al pairs in 6MR. This has implications for the adsorption of guest molecules, as will be discussed later on. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: Formation Energy of Pd+H+ on CHA. Each bar represents unique Al location(s). The color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in an NNN configuration or the isolated site and striped bars refer to the NNNN configuration.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: Formation energy of \(\mathrm{Pd^{+2}}\) on CHA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in NNN configuration and striped bars refer to NNNN configuration. The two most stable sites (CHA-7 and CHA-3) are illustrated in Figure 5.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: QM atoms of the optimized calculations. Color coding: red=Oxygen, white=Hydrogen, blue=Palladium and beige=Silicon
+ +## Pd-exchanged BEA + +BEA has a more diverse set of rings (4,5,6 and 12), 9 T- sites, and is denser than CHA. This makes it more challenging and nontrivial to determine the favorable sites for the cations. Similar to CHA, we carried out calculations for \(\mathrm{Pd^{+}}\) , \(\mathrm{Pd^{+}H^{+}}\) , and \(\mathrm{Pd^{+2}}\) charge- exchanged into BEA using the approach shown in Figure 1. Figure 6 summarizes the results for \(\mathrm{Pd^{+2}}\) in BEA. It is noticeable that the four most energetically favorable structures all have Al pairs in 6MR (images of the QM atoms are shown in Figure 5c- f). The four structures are separated by 0.15- 0.6 eV. Structure BEA- 65 (Figure 5c), the most favorable energetically, has Al pairs in a 6MR in NNN configuration. The Pd cation resides at the center of the 6MR in close proximity to 4 neighboring oxygen atoms. The three structures closest in energy (BEA- 80, BEA- 45 and BEA- 41) have similar configurations, but differ mainly in Al placements within the 6MR and the Si/O atoms surrounding the 6MR. These four structures are followed by a number of structures where Al pairs are in a 5MR (e.g. Figure S12a). Those 5MR structures + +<--- Page Split ---> + +are close in energy (separated by less than 0.15 eV). In all of these cases, the Pd cation is most stable at the center of the 5MR. Al pairs in 4MR (e.g. BEA- 36) are poor hosts for the Pd cation, similar to what was observed for CHA. For Al pairs not in a 4,5 or 6MR, \(\Delta E_{form}\) is considerably lower. The most stable structure for those cases (BEA- 55, Figure S12b) is 1.2 eV weaker compared to the most stable structure. This reinforces the results observed in CHA (Figure 4) where Al pairs in the same \(n\) - MR allow for additional stability of the cation. This likely stems from having neighboring oxygen atoms in positions favorable for orbital overlap with the Pd cation. However, this observation fails to explain why some of the other structures with Al pairs in a 6MR or 5MR arrangements (e.g. BEA- 62 and BEA- 59) have significantly smaller values of \(\Delta E_{form}\) . This subject will be discussed further below. + +![](images/Figure_6.jpg) + +
Figure 6: Formation energy of \(\mathrm{Pd^{+2}}\) on BEA. Each bar represents unique Al locations. Color coding refers to the type of Al pairs in the zeolite matrix. Patterned bars refer to Al pairs in NNNN positions while solid bars refer to Al pairs in NNN positions. The four most stable sites are illustrated in Figure 5.
+ +Figure 7 summarizes the formation energies for \(\mathrm{Pd^{+}}\) and \(\mathrm{Pd^{+}H^{+}}\) on BEA. Structure BEA- 65 is the most energetically favorable site in BEA for \(\mathrm{Pd^{+}H^{+}}\) . It is the same site that hosts the most stable \(\mathrm{Pd^{+2}}\) , for which the cation is located at the center of the 6MR and the Al pairs are in NNN arrangement (Figure 5g). It is \(>0.25 \mathrm{eV}\) more favorable than + +<--- Page Split ---> + +any of the other structures examined. Surprisingly, structures BEA- 33 and BEA- 55 (Figure S12c and d), in which the Al pairs are do not share a 5MR or 6MR, are more stable than the other Al pairs in 5MR or 6MR. However, after BEA- 65, most of the structures are close in energy (the difference can be \(< 0.1 \mathrm{eV}\) ). Given DFT errors and the large number of structures close in energies, it is difficult to make a conclusion on the order between many of these structures. Nevertheless, there is a lack of clear distinctions between Al pairs in the same \(n\) - MR observed for \(\mathrm{Pd}^{+2}\) in CHA and BEA. Some Al configurations (e.g. BEA- 8) can have a high \(\Delta E_{form}\) for \(\mathrm{Pd}^{+2}\) but not for \(\mathrm{Pd}^{+}\mathrm{H}^{+}\) . This indicates that results are not transferable between different oxidation states. It also sheds light on potential importance of sites where Al pairs do not share a 5MR or 6MR, a motif that does not generally receive much attention in computational studies, and the role they can play in adsorption/catalysis. In these structures, the presence of a neighboring Al atom in close proximity can significantly alter the \(\Delta E_{form}\) compared to their respective value for isolated sites (Figure S13). + +![](images/Figure_7.jpg) + +
Figure 7: Formation energy of \(\mathrm{Pd}^{+}\) on BEA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. In Al pairs, patterned bars refer to Al pairs in NNNN positions and solid colors refer to Al pairs in NNN positions. The most stable site (BEA-65) is illustrated in Figure 5.
+ +During the search for the most favorable charge- exchange sites, we found that in some cases the energetically most favorable Pd cation position may not involve situating the cation + +<--- Page Split ---> + +within a single ring, even if the Al pairs are in the same MR. Examples of this situation are BEA- 51, BEA- 78, and BEA- 52, shown in Figure S14. These positions can be \(>0.2 \mathrm{eV}\) more stable than the Pd at the center of the 6MR. This highlights the importance of the high throughout screening approach, which can find the optimum adsorption location when it is not the most physically intuitive. + +Although the energetic order of Al configurations in BEA does not correlate directly for Pd +1 and +2 oxidation states, we attempted to confirm whether this is true or not for \(\mathrm{Co}^{+2}\) cations. We performed a limited number of calculations on \(\mathrm{Co}^{+2}\) on CHA and found a similar trend in terms of favorable Al pairs (Figure 8). This indicates that the results for one cation in a zeolite matrix are can be transferable to other, chemically similar, cations exchanged into the same zeolite. This finding is important, since the UV- Vis spectrum of \(\mathrm{Co}^{2 + }\) is used to identify the location of divalent cation in zeolites [38, 39]. + +![](images/Figure_8.jpg) + +
Figure 8: Comparison between favorable Al pair configurations for \(\mathrm{Pd}^{+2}\) and \(\mathrm{Co}^{+2}\)
+ +<--- Page Split ---> + +## Comparison between Pd-exchanged CHA and BEA + +Comparison between Pd- exchanged CHA and BEAIn general, there are similarities in Pd- exchanged BEA and CHA. In both cases, the Pd cation (both as a \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+}H^{+}}\) ) is energetically most favorable in the 6MR. For \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , the site with the highest \(\Delta E_{form}\) has almost identical energy values (around - 3.3 eV and - 4.2, respectively). The 6MR in BEA is more oval shaped (Figure 5c) compared to that of CHA (Figure 5a), placing the oxygen atoms at a closer distance. The four oxygens close to the Pd cation in the 6MR are on average located 2.0 Å from the Pd in BEA and 2.2 Å in CHA. This, however, does not seem to impact the formation energy. + +## NO Adsorption + +NO AdsorptionThe location of Pd cations in CHA and BEA has practical implications for the capability of the zeolite to act as a passive \(\mathrm{NO}_x\) adsorber (PNA) during the cold start of automobiles. The nature of the oxidation state of the Pd cations and their location in the zeolite continue to be debated in the scientific literature [27, 28, 40]. Here, we attempt to shed some light on the subject based on the results of our high throughput screening. We focus on NO adsorption on Pd- exchanged- CHA, considering \(\mathrm{Pd^{+}}\) , \(\mathrm{Pd^{+}H^{+}}\) , and \(\mathrm{Pd^{+2}}\) . Figure 9 shows the correlation results between \(\Delta \mathrm{E}_{\mathrm{NO}}\) and \(\Delta E_{form}\) . Generally, a weaker \(\Delta E_{form}\) correlates with a stronger NO adsorption energy. Similar to most adsorption processes, the stronger the binding energy of the cation site, the less electron density is available to bind the guest gas specie [41]. Although there appears to be a linear correlation with a low mean absolute error (MAE) (0.1 and 0.22 eV for NO adsorption on \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , respectively), there are clear outliers (e.g. one of two 4 MR in \(\mathrm{Pd^{+2}}\) is 0.5 eV off the line). + +Figure 9 demonstrates that there is a clear distinction in NO adsorption between \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) . For a similar \(\Delta E_{form}\) , \(\mathrm{E}_{\mathrm{NO}}\) in \(\mathrm{Pd^{+}H^{+}}\) is much stronger (by as much as 1 eV). For most of the Al pair arrangements, the NO binds more strongly to \(\mathrm{Pd^{+}H^{+}}\) compared to \(\mathrm{Pd^{+2}}\) . We also find NO adsorption on \(\mathrm{Pd^{+}H^{+}}\) to be stronger when the \(\mathrm{Pd^{+}}\) and NO unpaired electrons are paired (compared to two unpaired electrons) (Table S2). This might + +<--- Page Split ---> + +be one reason for the stronger binding on \(\mathrm{Pd^{+}H^{+}}\) (given that NO adsorption on \(\mathrm{Pd^{+2}}\) has an unpaired NO electron). Experimentally, it is unclear if \(\mathrm{Pd^{+}H^{+}}\) is present under operating conditions; however, Figure 9 indicates that if present, it is a superior NO adsorption site compared to \(\mathrm{Pd^{+2}}\) . + +Figure 9 also shows that Al pairs in the 6MR arrangement (especially in NNNN configuration), which has been discussed most extensively in the literature because they hold \(\mathrm{Pd^{+2}}\) cations most stably, are weaker sites for NO adsorption compared to other Al arrangements. This is not unexpected since the more stable the Pd, the more weakly it can bind to a guest molecule. It is striking, however, that the difference in \(\Delta \mathrm{E}_{\mathrm{NO}}\) compared to the other Al arrangements (0.25- 1.5 eV). Figure 9 also highlights how many of the CHA sites are very close in energy, especially for \(\mathrm{Pd^{+}H^{+}}\) . Therefore, it is very difficult based on the small differences in energy values to discern spectroscopic data and assign them to specific sites. The results indicate that an ensemble of sites of very similar energies contribute similarly to the adsorption of NO. + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 9: NO Binding energy versus Pd formation energy on CHA. The color coding refers to the type of Al in the \(n\) -MR. The marker shapes (□ and \(\circ\) ) represent \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+} / Pd^{+}H^{+}}\) , respectively. The black and blue lines are based on fitting data for \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+} / Pd^{+}H^{+}}\) , respectively. Filled markers are used for Al pairs in NNNN configuration and half-filled markers are for isolated Al or Al in NNN configurations
+ +## Conclusion + +We developed a high throughput screening approach for enumerating unique Al arrangements in zeolites, examined possible cation- exchanged sites, calculated energies via density functional theory (DFT) for stabilizing Pd cations in different locations, and determined favorable locations for the positioning of metal cations in a zeolite. Pd adsorption on CHA and BEA was used to illustrate the approach. After conducting thousands of individual calculations, we arrived at Al configurations that allow optimum Pd cation stability. Sites with Al in 6MR bind Pd cations most strongly, consistent with prior work [27, 28]. However, high throughput screening identifies other Al configurations, especially for \(\mathrm{Pd^{+}H^{+}}\) , that bind Pd cations very stably. Many of these arrangements have not been considered previously for CHA and BEA. In addition, the most stable location for Pd is not always the most + +<--- Page Split ---> + +physically intuitive (e.g., inside a 5MR or 6MR). Lastly, we examined the impact of the high throughput approach on NO adsorption on Pd- exchanged- CHA. We generally find \(\mathrm{Pd^{+}H^{+}}\) to be more stable on CHA compared to \(\mathrm{Pd^{+2}}\) . Although Al pairs in 6MR are very stable sites for the Pd cation, NO adsorption on these sites is weaker compared to that for other Al configurations examined. This identifies sites that might play a critical role in the adsorption of NO in a PNA. Finally, we note that the DFT- based high throughput screening reported here provides a robust systematic path for identifying the most energetically favored cation exchange sites. While the method is applied here to zeolites, it could be extended to identify the energetically preferred location of cations in many other materials (e.g., MOFs). + +## Methods + +## Theoretical calculations + +A hybrid quantum mechanics/molecular mechanics (QM/MM) approach was used to model the zeolite structure. Detailed implementation of this model can be found elsewhere [16]. The QM/MM approach has proven to account for long- range Coulombic and dispersive interactions, which are critical in describing the zeolite framework interactions with adsorbates [36]. Many studies have shown that the QM/MM approach gives a good prediction of experimental data for different zeolites and adsorbates [16, 32- 34, 36, 42, 43]. This approach is also computationally more efficient than periodic calculations since it requires a smaller number of QM atoms, especially when the unit cell contains a large number of atoms (e.g. BEA). All QM/MM calculations were done with a development version of Q- Chem [44]. + +The adsorbate(s) and a cluster encompassing the active site are described by QM and the rest of the zeolite is modeled by MM using a standard force field of the CHARMM type with the P2 parameter set [17]. During structural optimization, the QM region is allowed to relax while the MM region is fixed. The B97- D3 exchange functional [45], a generalized gradient approximation (GGA) exchange functional, is used as an initial filter + +<--- Page Split ---> + +to determine the optimum site of the cation followed by calculations done with the more accurate range- separated hybrid functional, \(\omega \mathrm{B97X - D}\) [46], which was shown to be among the best performing hybrid functionals in a benchmarking study [18]. An effective core potential was used on Pd atom. For each structure, the def2- SV(P) basis set was used to obtain the optimized structure geometry, and further energy refinement was done using a single- point calculation at the def2- TZVPD level of theory [47]. + +## Zeolite model + +The crystallographic structure of CHA and BEA were obtained from the International Zeolite Association (IZA) database [48]. Cluster models containing 696 and 810 tetrahedral atoms (T696 and T810) were used to model CHA and BEA topologies, respectively (Figure S15). The CHA cluster is based on the previous work from or group[27]. While earlier work has suggested that a 100 T- atom cluster model is sufficient [32], larger cluster models are used here due to the marginal additional computational cost and the extended active site region in some of the calculations. Each cluster was terminated with hydrogen atoms replacing terminal oxygen atoms. + +## Data Availability + +The data that support the results within this paper and other findings of this study are available in the supporting information. + +<--- Page Split ---> + +## References + +1. Guisnet, M. & Gilson, J.-P. Zeolites for cleaner technologies (Imperial College Press London, 2002).2. Corma, A. From microporous to mesoporous molecular sieve materials and their use in catalysis. Chemical reviews 97, 2373-2420 (1997).3. Kulprathipanja, S. Zeolites in industrial separation and catalysis (John Wiley & Sons, 2010).4. Paolucci, C. et al. Catalysis in a cage: Condition-dependent speciation and dynamics of exchanged cu cations in ssz-13 zeolites. Journal of the American Chemical Society 138, 6028-6048. ISSN: 15205126 (2016).5. Janda, A., Vlaisavljevich, B., Lin, L.-C., Smit, B. & Bell, A. T. Effects of zeolite structural confinement on adsorption thermodynamics and reaction kinetics for monomolecular cracking and dehydrogenation of n-butane. Journal of the American Chemical Society 138, 4739-4756 (2016).6. Norskov, J. K., Studt, F., Abild-Pedersen, F. & Bligaard, T. Fundamental concepts in heterogeneous catalysis (John Wiley & Sons, 2014).7. Van Santen, R. A. Modern heterogeneous catalysis: an introduction (John Wiley & Sons, 2017).8. Olson, D. H., Khosrovani, N., Peters, A. W. & Toby, B. H. Crystal structure of dehydrated CsZSM-5 (5.8 Al): evidence for nonrandom aluminum distribution. The Journal of Physical Chemistry B 104, 4844-4848 (2000).9. Mentzen, B. & Sacerdote-Peronnet, M. Prediction of preferred proton locations in HMFI/benzene complexes by molecular mechanics calculations. Comparison with NMR, structural and calorimetric results. Materials research bulletin 29, 1341-1348 (1994). + +<--- Page Split ---> + +10. SUN, X.-L., HUANG, C.-P., ZHANG, J. & CHEN, B.-H. Location of Al and Acid Strength of Br2nsted Acid in Beta Zeolite. Acta Physico-Chimica Sinica 25, 1136–1142 (2009). + +11. Mansoor, E., Head-Gordon, M. & Bell, A. T. Computational modeling of the nature and role of Ga species for light alkane dehydrogenation catalyzed by Ga/H-MFI. ACS Catalysis 8, 6146–6162 (2018). + +12. Ghorbanpour, A., Rimer, J. D. & Grabow, L. C. Periodic, vdW-corrected density functional theory investigation of the effect of Al siting in H-ZSM-5 on chemisorption properties and site-specific acidity. Catalysis Communications 52, 98–102 (2014). + +13. Jones, A. J. & Iglesia, E. The Strength of Br2nsted Acid Sites in Microporous Aluminosilicates. ACS Catalysis 5, 5741–5755. ISSN: 21555435 (2015). + +14. Xing, B., Ma, J., Li, R. & Jiao, H. Location, distribution and acidity of Al substitution in ZSM-5 with different Si/Al ratios—a periodic DFT computation. Catalysis Science & Technology 7, 5694–5708 (2017). + +15. Zhang, N., Liu, C., Ma, J., Li, R. & Jiao, H. Determining the structures, acidity and adsorption properties of Al substituted HZSM-5. Physical Chemistry Chemical Physics 21, 18758–18768 (2019). + +16. Zimmerman, P. M., Head-Gordon, M. & Bell, A. T. Selection and validation of charge and Lennard-Jones parameters for QM/MM simulations of hydrocarbon interactions with zeolites. Journal of chemical theory and computation 7, 1695–1703 (2011). + +17. Li, Y.-P., Gomes, J., Mallikarjun Sharada, S., Bell, A. T. & Head-Gordon, M. Improved force-field parameters for QM/MM simulations of the energies of adsorption for molecules in zeolites and a free rotor correction to the rigid rotor harmonic oscillator model for adsorption enthalpies. The Journal of Physical Chemistry C 119, 1840–1850 (2015). + +<--- Page Split ---> + +18. Mardirossian, N. & Head-Gordon, M. Thirty years of density functional theory in computational chemistry: an overview and extensive assessment of 200 density functionals. Molecular Physics 115, 2315–2372 (2017). + +19. Kim, J., Abouelnasr, M., Lin, L.-C. & Smit, B. Large-scale screening of zeolite structures for CO2 membrane separations. Journal of the American Chemical Society 135, 7545–7552 (2013). + +20. Lee, Y. et al. High-throughput screening approach for nanoporous materials genome using topological data analysis: application to zeolites. Journal of chemical theory and computation 14, 4427–4437 (2018). + +21. Kim, J. et al. Large-scale computational screening of zeolites for ethane/ethene separation. Langmuir 28, 11914–11919 (2012). + +22. Slawek, A., Grzybowska, K., Vicent-Luna, J. M., Makowski, W. & Calero, S. Adsorption of Cyclohexane in Pure Silica Zeolites: High-Throughput Computational Screening Validated by Experimental Data. ChemPhysChem 19, 3364–3371 (2018). + +23. Siepmann, J. I., Bai, P., Tsapatsis, M., Knight, C. & Deem, M. W. Discovery of optimal zeolites for challenging separations and chemical conversions through predictive materials modeling. APS 2015, B16–003 (2015). + +24. Chen, H.-Y. et al. Low temperature NO storage of zeolite supported Pd for low temperature diesel engine emission control. Catalysis Letters 146, 1706–1711 (2016). + +25. Gu, Y. & Epling, W. S. Passive NOx adsorber: An overview of catalyst performance and reaction chemistry. Applied Catalysis A: General 570, 1–14 (2019). + +26. Ji, Y., Bai, S. & Crocker, M. Al2O3-based passive NOx adsorbers for low temperature applications. Applied Catalysis B: Environmental 170, 283–292 (2015). + +27. Van der Mynsbruggc, J., Head-Gordon, M. & Bell, A. T. Computational modeling predicts the stability of both Pd+ and Pd 2+ ion-exchanged into H-CHA. Journal of Materials Chemistry A 9, 2161–2174 (2021). + +<--- Page Split ---> + +28. Mandal, K. et al. Condition-Dependent Pd Speciation and NO Adsorption in Pd/Zeolites. ACS Catalysis 10, 12801-12818 (2020). + +29. Loewenstein, W. The distribution of aluminum in the tetrahedra of silicates and aluminates. American Mineralogist: Journal of Earth and Planetary Materials 39, 92-96 (1954). + +30. Verstraelen, T., Van Speybroeck, V. & Waroquier, M. ZEOBUILDER: A GUI Toolkit for the Construction of Complex Molecular Structures on the Nanoscale with Building Blocks. Journal of Chemical Information and Modeling 48, 1530-1541. ISSN: 1549-9596. http://pubs.acs.org/doi/abs/10.1021/ci8000748 (2008). + +31. Larsen, A. H. et al. The atomic simulation environment—a Python library for working with atoms. Journal of Physics: Condensed Matter 29, 273002. http://stacks.iop.org/0953-8984/29/i=27/a=273002 (2017). + +32. Gomes, J., Zimmerman, P. M., Head-Gordon, M. & Bell, A. T. Accurate prediction of hydrocarbon interactions with zeolites utilizing improved exchange-correlation functionals and QM/MM methods: benchmark calculations of adsorption enthalpies and application to ethene methylation by methanol. The Journal of Physical Chemistry C 116, 15406-15414 (2012). + +33. Mallikarjun Sharada, S., Zimmerman, P. M., Bell, A. T. & Head-Gordon, M. Insights into the kinetics of cracking and dehydrogenation reactions of light alkanes in H-MFI. The Journal of Physical Chemistry C 117, 12600-12611 (2013). + +34. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Theoretical study of 4-(hydroxymethyl) benzoic acid synthesis from ethylene and 5-(hydroxymethyl) furoic acid catalyzed by Sn-BEA. ACS Catalysis 6, 5052-5061 (2016). + +35. Dinda, S., Govindasamy, A., Genest, A. & Rösch, N. Modeling Catalytic Steps on Extra-Framework Metal Centers in Zeolites. A Case Study on Ethylene Dimerization. The Journal of Physical Chemistry C 118, 25077-25088 (2014). + +<--- Page Split ---> + +36. Mansoor, E., Van der Mynsbrugg, J., Head-Gordon, M. & Bell, A. T. Impact of long-range electrostatic and dispersive interactions on theoretical predictions of adsorption and catalysis in zeolites. Catalysis Today 312, 51-65 (2018). + +37. Sastre, G., Fornes, V. & Corma, A. On the preferential location of Al and proton siting in zeolites: a computational and infrared study. The Journal of Physical Chemistry B 106, 701-708 (2002). + +38. Dedecek, J. & Wichterlová, B. Co2+ Ion Siting in Pentasil-Containing Zeolites. I. Co2+ Ion Sites and Their Occupation in Mordenite. A Vis- NIR Diffuse Reflectance Spectroscopy Study. The Journal of Physical Chemistry B 103, 1462-1476 (1999). + +39. Dedecek, J., Sobalik, Z. & Wichterlová, B. Siting and distribution of framework aluminium atoms in silicon-rich zeolites and impact on catalysis. Catalysis Reviews 54, 135-223 (2012). + +40. Khivantsev, K. et al. Stabilization of super electrophilic Pd+ 2 cations in small-pore SSZ-13 zeolite. The Journal of Physical Chemistry C 124, 309-321 (2019). + +41. Abild-Pedersen, F. et al. Scaling properties of adsorption energies for hydrogen-containing molecules on transition-metal surfaces. Physical review letters 99, 016105 (2007). + +42. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Computational study of p-xylene synthesis from ethylene and 2, 5-dimethylfuran catalyzed by H-BEA. The Journal of Physical Chemistry C 118, 22090-22095 (2014). + +43. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Analysis of the reaction mechanism and catalytic activity of metal-substituted beta zeolite for the isomerization of glucose to fructose. Acs Catalysis 4, 1537-1545 (2014). + +44. Shao, Y. et al. Advances in molecular quantum chemistry contained in the Q-Chem 4 program package. Molecular Physics 113, 184-215 (2015). + +45. Grimme, S. Semiempirical GGA-type density functional constructed with a long-range dispersion correction. Journal of computational chemistry 27, 1787-1799 (2006). + +<--- Page Split ---> + +46. Chai, J.-D. & Head-Gordon, M. Long-range corrected hybrid density functionals with damped atom-atom dispersion corrections. Physical Chemistry Chemical Physics 10, 6615-6620 (2008). + +47. Weigend, F. & Ahlrichs, R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence quality for H to Rn: Design and assessment of accuracy. Physical Chemistry Chemical Physics 7, 3297-3305 (2005). + +48. Meier, W. M., Olson, D. H. & Baerlocher, C. Atlas of the zeolite structure types. Zeolites 17 (1996). + +<--- Page Split ---> + +## Acknowledgement + +AcknowledgementThis material is based upon work supported by the U.S. Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE) under the Vehicle Technologies Program Award Number DE- EE0008213. This research used resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE- AC02- 05CH11231. The authors also thank UC Berkeley's Molecular Graphics and Computation Facility (supported by NIH S10OD023532) for their computational resources. H.A. would like Saudi Aramco for their funding, and MHG acknowledges funding from the National Institutes of Health under Grant No. 5U01GM121667. This work used the Extreme Science and Engineering Discovery Environment (XSEDE) Comet at the service- provider through allocation TG- CE200085. + +## Author Contributions + +Author ContributionsH.A and A.T.B conceptualized the project. H.A, M.H.G and A.T.B developed the methodology. H.A. wrote the code, performed the calculations, and wrote the original draft. All authors participated in data analysis and editing the manuscript. M.H.G and A.T.B supervised the research reported in the paper. + +<--- Page Split ---> + +# Supporting Information Available + +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput workflow + +Supporting Information + +Hassan Aljama†, Martin Head- Gordon‡ and Alexis T. Bell\*,\† † Department of Chemical and Biomolecular Engineering, University of California, Berkeley, California ‡Department of Chemistry, University of California, Berkeley, California + +<--- Page Split ---> + +All energies and xyz coordinates of the optimized geometries are available in the electronic supporting information (ESI). The nomenclature used to tabulate the data is as follow: zeolite name (CHA or BEA) - a number representing structure with a unique Al arrangement - adsorbate name (Pd+2, Pd+H+, Pd+, H+ or H+H+) - a number representing an initial position of the adsorbate - level of theory (GGA or hGGA) - type of calculations (opt for optimization and sp for single point calculation). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S1: Images of the QM atoms in four BEA structures. BEA-2 structure is shown in a. alongside a similar structure in b. BEA-11 is shown in c. alongside a similar structure in d. The structures similar to BEA-2 and BEA-11 (b. and d., respectively) are not exactly identical, as calculated by the nuclear repulsion energy, but share the same connectivity to surrounding Si and O atoms (types of MR). The same color coding as in Figure 5.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S2: Example of two BEA clusters (a. and b.) where Al atoms are not in the same MR and are on opposite sides of the open cage. The same color coding as in Figure 5, however, for clarity purposes, Al atoms are enlarged and are shown in green
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure S3: Flow diagram for determining if Al atom is in an X MR (where X refers to the length of the MR). First, starting with an Al atom, Si/Al neighbors (NN, NNN, NNNN) are identified similar to the procedure described in Figure 2. This is done up to the X+1 neighbor. Based on the neighbors, lists are developed (an example of a list is [Al, NN1, NNN11, ..], where NN1 refers to the first NN of the starting Al and NNN11 is the first NN to NN1). For each list, the following checks are made: first and last element of the list are the same (the starting Al atom), no element in the list (with the exception of the starting Al atom) appears twice in the list, and no sub-list of the list form a MR smaller than X. If a list passes those checks, then Al is part of XMR.
+ +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Figure S4: Impact of the number of atoms selected in the QM region on the convergence of the QM/MM calculations ( \(\Delta E_{form}\) is defined in equation 2). Calculations were done using \(\omega\) B97X-D functional.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S5: Example of a. Pd cation and b. proton placement near the four oxygens neighboring Al atom. For clarity, only Si in NN position and O neighboring Al are shown. The position of cation/proton is determined by finding the middle distance between neighboring oxygen, and then adding a displacement from the Al atom (usually 1-1.5 A). The same color coding as in Figure 5.
+ +<--- Page Split ---> + +Table S1: Comparison of the stability order (from most stable to 5th most stable) on 3 different structures (BEA-9, BEA-52 and BEA-6) at two levels of theory (GGA and hGGA). The adsorbate is \(\mathrm{Pd^{+}H^{+}}\) . For simplicity, only the 5 most stable structures are shown out of the 32. The two results do not always yield the same order, however, the most stable structure at the hGGA level always appears among the 5 most stable structures at the GGA level. This means following the approach in Figure 1 yields the same results as doing all the calculations using \(\omega \mathrm{B97X - D}\) . + +
StabilityTheory LevelBEA-9BEA-52BEA-63
1stGGABEA-9-Pd+H+-3BEA-52-Pd+H+-24BEA-63-Pd+H+-26
hGGABEA-9-Pd+H+-12BEA-52-Pd+H+-24BEA-63-Pd+H+-25
2ndGGABEA-9-Pd+H+-11BEA-52-Pd+H+-4BEA-63-Pd+H+-25
hGGABEA-9-Pd+H+-11BEA-52-Pd+H+-23BEA-63-Pd+H+-26
3rdGGABEA-9-Pd+H+-12BEA-52-Pd+H+-3BEA-63-Pd+H+-5
hGGABEA-9-Pd+H+-3BEA-52-Pd+H+-1BEA-63-Pd+H+-5
4thGGABEA-9-Pd+H+-24BEA-52-Pd+H+-27BEA-63-Pd+H+-24
hGGABEA-9-Pd+H+-24BEA-52-Pd+H+-27BEA-63-Pd+H+-24
5thGGABEA-9-Pd+H+-27BEA-52-Pd+H+-1BEA-63-Pd+H+-16
hGGABEA-9-Pd+H+-27BEA-52-Pd+H+-4BEA-63-Pd+H+-16
+ +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
Figure S6: CHA-2 optimum proton locations search results. Each data represents the optimum energy based on a different initial position of the protons. The x-axis refers to the index of the structure in the database in the SI and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14). Calculations were done using the \(\omega\) B97X-D functional.
+ +<--- Page Split ---> +![](images/Figure_unknown_2.jpg) + +
Figure S7: CHA-2 optimum proton locations search results. Each data represents a different initial position of the proton. The x-axis refers to the O-O distance (where the oxygen is the atom H adsorbs on) in the optimized structure and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14)
+ +<--- Page Split ---> +![](images/Figure_unknown_3.jpg) + +
Figure S8: CHA-2 optimum \(\mathrm{Pd^{+}H^{+}}\) locations search results. Each data represents a different initial position of \(\mathrm{Pd^{+}H^{+}}\) . The x-axis refers to the index of the structure in the database and the y-axis is the energy relative to the most stable structure (CHA-2-Pd+H+-3). Calculations were done using the \(\omega \mathrm{B97X - D}\) functional.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S9: QM images of the optimized geometry of a. CHA-3-Pd+H+-21 and b. CHA-3-Pd+H+-17. Despite only a change in the proton position, there is \(>0.4 \mathrm{eV}\) in energy difference between the two optimized geometries. The same color coding as in Figure 5
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S10: QM region images of the optimized geometry of a. Pd+H+ on CHA-7 b. Pd+H+ on CHA-3 c. Pd+H+ on CHA-8 d. Pd+H+ on CHA-6 e. Pd+H+ on CHA-9 f. Pd+H+ on CHA-12 g. Pd+2 on CHA-5 and h. Pd+2 on CHA-10. The same color coding as in Figure 5.
+ +<--- Page Split ---> +![](images/Figure_unknown_4.jpg) + +
Figure S11: \(\Delta E_{form}\) as a function of the distance between Pd and H in Pd+H+ optimized geometry on CHA (□) and BEA (○)
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S12: QM region images of optimized geometry of a. Pd+2 on BEA-8 b. Pd+2 on BEA-55 c. Pd+H+ on BEA-33 d. Pd+H+ on BEA-55. The same color coding as in Figure 5.
+ +<--- Page Split ---> +![](images/Figure_unknown_5.jpg) + +
Figure S13: Parity plot of the calculated DFT energy on Al pairs (that do not share a MR) and the predicted energy based on the respective isolated Al site (the most stable of the two). All calculations were done on BEA with \(\mathrm{Pd^{+}H^{+}}\) as the cation
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S14: Image of the QM region on optimized BEA calculations a. BEA-51 (Pd+H+) b. BEA-78 (Pd+H+) c. BEA-52 (Pd+H+). The same color coding as in Figure 5.
+ +<--- Page Split ---> + +Table S2: Difference in NO adsorption energy ( \(\Delta \mathrm{E}_{NO}\) ) on \(\mathrm{Pd}^{+}\mathrm{H}^{+}\) in CHA when the two electrons (from \(\mathrm{Pd}^{+}\) and NO) are paired and the two electrons are unpaired. Paired electrons are used as the reference (0 eV) + +
Structure NamePaired ElectronsTwo Unpaired Electrons
CHA-500.89
CHA-800.82
CHA-701.53
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure S15: Cluster models of a. T696 CHA and b. T810 BEA. The same color coding as in Figure 5
+ +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +Please see manuscript PDF for full caption. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Please see manuscript PDF for full caption. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Please see manuscript PDF for full caption. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Please see manuscript PDF for full caption. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5
+ +QM atoms of the optimized calculations. Color coding: red=Oxygen, white=Hydrogen, blue=Palladium and beige=Silicon + +![](images/Figure_6.jpg) + +
Figure 6
+ +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7
+ +Please see manuscript PDF for full caption. + +![](images/Figure_8.jpg) + + +<--- Page Split ---> +![PLACEHOLDER_56_0] + +
Figure 8
+ +Figure 9 + +Please see manuscript PDF for full caption. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91_det.mmd b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1f8e6d11008557f45b0b93777eb67f0b3fc789d5 --- /dev/null +++ b/preprint/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91/preprint__1303a8e6573487cd8e52e445ff02034cc0cbaf9a74a9a03fae1faebe66bf0a91_det.mmd @@ -0,0 +1,633 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 925, 210]]<|/det|> +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput quantum chemistry workflow + +<|ref|>text<|/ref|><|det|>[[44, 230, 696, 316]]<|/det|> +Hassan Aljama University of California, Berkeley Martin Head- Gordon ( \(\boxed{\text {mhg@cchem.berkeley.edu}}\) ) University of California, Berkeley https://orcid.org/0000- 0002- 4309- 6669 + +<|ref|>text<|/ref|><|det|>[[44, 322, 755, 387]]<|/det|> +Alexis Bell Joint Center for Artificial Photosynthesis, Lawrence Berkeley National Laboratory https://orcid.org/0000- 0002- 5738- 4645 + +<|ref|>sub_title<|/ref|><|det|>[[44, 428, 102, 446]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 465, 899, 508]]<|/det|> +Keywords: Cation Exchanged- zeolites, Density Functional Theory, High Throughput Screening, NOx Capture + +<|ref|>text<|/ref|><|det|>[[44, 525, 283, 545]]<|/det|> +Posted Date: May 7th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 563, 463, 583]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 396201/v1 + +<|ref|>text<|/ref|><|det|>[[44, 601, 909, 644]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 679, 909, 723]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 25th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29505- z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[131, 144, 866, 280]]<|/det|> +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput quantum chemistry workflow + +<|ref|>text<|/ref|><|det|>[[201, 314, 792, 338]]<|/det|> +Hassan Aljama,† Martin Head- Gordon,\*,\$ and Alexis T. Bell\*,\$ + +<|ref|>text<|/ref|><|det|>[[152, 362, 844, 414]]<|/det|> +†Department of Chemical and Biomolecular Engineering, University of California, Berkeley, California + +<|ref|>text<|/ref|><|det|>[[189, 423, 808, 444]]<|/det|> +‡Department of Chemistry, University of California, Berkeley, California + +<|ref|>text<|/ref|><|det|>[[273, 468, 722, 488]]<|/det|> +E- mail: mhg@cchem.berkeley.edu; alexbell@berkeley.edu + +<|ref|>sub_title<|/ref|><|det|>[[458, 519, 538, 537]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[161, 553, 839, 912]]<|/det|> +Cation exchanged- zeolites are functional materials with a wide range of applications from catalysis to sorbents. They present a challenge for computational studies using density functional theory due to the numerous possible active sites. From Al configuration, to placement of extra framework cation(s), to potentially different oxidation states of the cation, accounting for all these possibilities is not trivial. To make the number of calculations more tractable, most studies focus on a few active sites. We attempt to go beyond these limitations by implementing a workflow for a high throughput screening, designed to systematize the problem and exhaustively search for feasible active sites. We use Pd- exchanged CHA and BEA to illustrate the approach. After conducting thousands of individual calculations, we identify the sites most favorable for the Pd cation and discuss the results in detail. The high throughput screening identifies many energetically favorable sites that are non- trivial. Lastly, we employ these results to examine NO adsorption in Pd- exchanged CHA, which is a promising + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[161, 90, 836, 138]]<|/det|> +passive \(\mathrm{NO}_{x}\) adsorbent (PNA) during the cold start of automobiles. The results shed light on critical active sites for NOx capture that were not previously studied. + +<|ref|>sub_title<|/ref|><|det|>[[115, 182, 291, 206]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[113, 230, 884, 403]]<|/det|> +Zeolites are microporous aluminosilicates composed of corner- sharing \(\mathrm{TO}_{4}\) tetrahedra ( \(\mathrm{T} =\) Si or Al). Substitution of trivalent Al for tetravalent Si creates a charge imbalance that is compensated by a proton or a metal cation. If the charge compensating cation is a proton, the zeolites becomes a strong solid acid that is an effective catalyst for promoting a wide variety of hydrocarbons reaction as well as the synthesis of a broad range of organic compounds. Metal cation- exchanged zeolites can also serve as adsorbents and catalysts [1- 5]. + +<|ref|>text<|/ref|><|det|>[[112, 411, 885, 796]]<|/det|> +In recent years, considerable insights into zeolite catalyzed reactions have come from the application of quantum chemical analyses of the free energy and enthalpy landscapes governing the progress of chemical reactions [6, 7]. A significant challenge for such studies is selection of the structure of the active site. To appreciate the issue, we must first recall that there are over 200 known zeolites structures, many of which have T sites occupied by Si and Al having symmetries differing from one another. Each of the charge- exchange sites is associated with a framework Al atom, but the distribution of Al is generally not known and is thought to be controlled by the kinetics of zeolite synthesis. A further complication is that the charge- exchange site involves an Al atom bonded to four oxygen atoms, any one of which can accommodate the proton. Further complexity arises when extra- framework metal cations replace protons. Monovalent cations are usually large enough that they bridge two framework oxygen atoms, while bivalent cations can interact with two or three Al sites simultaneously. + +<|ref|>text<|/ref|><|det|>[[114, 805, 884, 886]]<|/det|> +It is, therefore, evident that the selection of a representative cation site is challenging and that full exploration of the chemistry on all possible sites is computationally formidable because of cost. For these reasons, many higher level theoretical studies of zeolite- catalyzed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 263]]<|/det|> +reactions[8- 11] using density functional theory (DFT) have chosen to focus on a few active sites selected on the basis of limited experimental evidence and/or physical intuition. While this choice leads to a more tractable set of calculations, the downside is that potentially important active sites might be missed because they are difficult to identify experimentally or are not physically intuitive. Indeed, growing numbers of DFT studies have progressed from studying single T- sites[12, 13] to address greater complexity[14, 15]. + +<|ref|>text<|/ref|><|det|>[[113, 271, 885, 623]]<|/det|> +With this trend in mind, the objective of this work is to present and apply a DFT- based computational framework to identify the energetically most favorable adsorption sites (cation or proton) on any zeolite using a systematic high- throughput approach. The underlying calculations employ a hybrid quantum mechanics/molecular mechanics (QM/MM) models[16, 17] to capture extended environment around each active site, and high quality DFT[18] to attain acceptable accuracy. Our approach starts by evaluating the possible zeolite structures that arise for different Si/Al ratios, then searching for energetically favorable cation/proton(s) adsorption sites for each structure using a lower level of theory (e.g. smaller basis set and cheaper functional) as a filter. The most favorable sites are then evaluated at a higher level of theory. We note that while there are previous high throughput approaches in zeolites, they were largely focused on grand canonical Monte Carlo (GCMC) simulations [19- 23]. + +<|ref|>text<|/ref|><|det|>[[113, 632, 885, 895]]<|/det|> +To illustrate our approach, we use the challenging example of the siting of extraframe- work Pd ions in Al- doped chabazite (CHA) and beta (BEA). Pd- CHA and Pd- BEA were chosen because they are good candidates for passive NO adsorbers (PNAs) that can be used to trap the emissions of NO from automobile exhaust during cold startup and before the three- way catalytic converter becomes effective [24- 26]. The speciation of Pd in these materials is recognized as a challenging, and still controversial problem [27, 28]. We limit our discussion to the following Pd species: Pd+ associated with isolated charge- exchange sites, Pd+H+ pairs associated with two charge- exchange sites involving next nearest neighbor (NNN) or next- next nearest neighbor (NNN) pairs of Al atoms, and Pd+2 associated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 291]]<|/det|> +with two charge- exchange sites also involving NNN or NNNN Al pairs. The goal is to provide a methodology that is easily applicable and transferable to any adsorption/catalysis problem with any zeolite, thereby shifting the focus of zeolite studies from a limited number of specific sites to a more systematic approach and allowing a more exhaustive exploration of the descriptor space. After establishing a large set of feasible charge exchange sites for Pd in CHA and BEA, we conclude by assessing the performance of the sites for NO adsorption on Pd- CHA, with some interesting results. + +<|ref|>sub_title<|/ref|><|det|>[[115, 333, 507, 361]]<|/det|> +## High Throughput Approach + +<|ref|>text<|/ref|><|det|>[[113, 383, 885, 737]]<|/det|> +Figure 1 shows the workflow for the high throughput approach to determine which charge- exchange sites are most favorable for accommodating the charge- compensating cation and/or proton. Briefly, Al atom(s) are first introduced at different tetrahedral sites in the bare \(\mathrm{Si}_{x}\mathrm{O}_{y}\) zeolite, generating structures with unique (i.e. distinguishable) Al positions. For each unique Al arrangement, distinctions are made between atoms in the QM region, which are expected to be active in the adsorption process, and the surrounding atoms in the MM region. To determine the most energetically favorable location for accommodating the extra framework cation, potential cation- exchanged sites are first enumerated and then surveyed by QM/MM calculations at the GGA level of theory, followed by further evaluation of the top 5 most stable sites at the hGGA level of theory. Lastly, results for all structures are compared to determine the structures where Al position(s) yield the most stable charge- exchange site. Each step is discussed in detail in the next few sections. + +<|ref|>sub_title<|/ref|><|det|>[[115, 773, 442, 796]]<|/det|> +## Distinguishable Al locations + +<|ref|>text<|/ref|><|det|>[[114, 814, 884, 895]]<|/det|> +The first step of the approach (Figure 1) is determining the set of distinguishable Al sites in the zeolite structure. Univalent structures could be easily generated by replacing each unique Si T- atom in the zeolite by an Al atom. However, generating Al pairs is not trivial + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 275, 878, 541]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 553, 883, 718]]<|/det|> +
Figure 1: Workflow of the high throughput approach used in this work. \(\mathrm{Pd}_n\mathrm{H}_n\) ( \(n = 0\) , 1 or 2 and \(m = 0\) , 1 or 2) is used as an illustration. Starting with the zeolite material, structures are generated by identifying the unique Al position(s) in step 1 (generating structures from 1 to w), followed by identifying atoms in the QM and MM region for each structure in step 2. In step 3, possible charge-exchange sites are enumerated and then further evaluated in step 4 using QM/MM calculations at the GGA level. The 5 most favorable exchange sites from step 4 for each unique Al position are further evaluated at the hGGA level of theory in step 5. Finally, all structures from step 5 are ordered based on their energy stability, yielding the most energetically favorable sites for the exchange in the zeolite in step 6
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 885, 352]]<|/det|> +due to the 3D structure of zeolites. While this can be done manually for CHA, since it only has one unique T- site, it is more challenging to do for BEA, which has 9 T- sites. Hence, we developed an automated method to generate all possible Al pairs. We focused in this work on pairs in the NNN (where the two framework Al are separated by a single Si atom) and the NNNN (where the two framework Al are separated by two Si atoms). Al pairs in nearest neighbor (NN) position were avoided in order to satisfy the Loewenstein rule [29]. Pairs further apart than NNNN were not considered since Al pairs further away should behave as isolated Al atoms. We note that this approach is easily applicable to any zeolite material and can be extended beyond NNNN pairs. + +<|ref|>text<|/ref|><|det|>[[113, 360, 886, 714]]<|/det|> +Our approach uses a molecular graph that identifies the connectivity of each atom in the cluster [30, 31] (visually illustrated in Figure 2). Each Al pair structure is generated by first identifying a unique T- site (Si \(T^{- atom}\) ). Based on the connectivity of the Si \(T^{- atom}\) , its NN Si atoms are identified (Si \(N^{NN}\) ). The Si atoms in NNN (Si \(N^{NN}\) ) are identified by finding the next neighbor of Si \(N^{NN}\) (excluding the original Si \(T^{- atom}\) and Si \(N^{NN}\) ). Each NNN Al pairs are generated by replacing Si \(T^{- atom}\) and one of the Si \(N^{NN}\) with Al atoms. For the NNNN Al pairs, the next neighbor of the Si \(N^{NN}\) atom are identified (Si \(N^{NNN}\) ) (excluding Si \(T^{- atom}\) , Si \(N^{NN}\) and Si \(N^{NN}\) ). Similarly, Si \(T^{- atom}\) and each one of the Si \(N^{NNN}\) atoms are then replaced with Al atoms to generate the NNNN Al pairs. In this manuscript, each structure with a unique Al location is identified by a unique index after the zeolite name (e.g. CHA- 13) (full details on the xyz coordinates for the structures are available in the Supporting Information (SI)). + +<|ref|>text<|/ref|><|det|>[[113, 722, 885, 895]]<|/det|> +Since the procedure described above can produce duplicate structures, we relied on calculating the nuclear repulsion energy of all generated pairs to eliminate duplicates. The method was verified using a BEA unit cell (which contains 36 T- atoms) and generating 36 different structures by replacing each Si T- atom with an Al atom. By using the nuclear repulsion energy, we were able to recover the 9 unique T- sites. Overall, this results in 26 and 212 unique structures for CHA and BEA, respectively. In order to make the number + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[202, 220, 825, 686]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 721, 883, 797]]<|/det|> +
Figure 2: Visual illustration of atoms positions surrounding the \(\mathrm{Si}^{T - atom}\) . \(\mathrm{Si}^{T - atom}\) is replaced with an Al atom to generate a univalent structure. For divalent structures, \(\mathrm{Si}^{NN}\) , \(\mathrm{Si}^{NNN}\) , and \(\mathrm{Si}^{NNNN}\) atoms are identified and then desired structures are generated by replacing either the \(\mathrm{Si}^{NNN}\) or \(\mathrm{Si}^{NNNN}\) and \(\mathrm{Si}^{T - atom}\) atom with Al atoms
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 291]]<|/det|> +of calculations more tractable, we further reduced the number of candidates by eliminating structures that share the same connectivity (i.e., the Al atoms in structures share the same types of \(n\) - membered- ring ( \(n\) - MR)). Although in those cases structures are not exactly the same, they are structurally very similar (Figure S1). We also did not evaluate structures for BEA for which Al pairs do not share the same MR and are on opposite sides of the open cage (Figure S2). This reduced the final numbers to 82 structures for BEA and 12 structures for CHA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 326, 550, 350]]<|/det|> +## Selection of the QM and MM regions + +<|ref|>text<|/ref|><|det|>[[113, 367, 886, 813]]<|/det|> +The second part of the workflow (Figure 1) defines the atoms that comprise the QM region and the surrounding MM atoms in the cluster. Appropriate choice of the QM region is critical for the QM/MM calculations since the number of atoms must be sufficient to ensure the calculations converge, but not too large in order to avoid significant computational cost. For structures with an isolated Al atom, the Si and O atoms that create a 4,5 or 6MR with the Al atom are included in the QM atoms. If the structure contains an Al pair, and if the two pairs form a 4,5,6,7 or 8MR, we include all Si and O atoms that are part of the \(n\) - MR. Also, for those structures, we include Si and O atoms that are part of a 4 or 5MR with either of the two Al atoms. Atoms comprising a MR are identified following the procedure described in Figure S3. The QM region includes on average 54 atoms, excluding hydrogen atoms used to terminate the Si atoms. The QM regions used here are larger than used in previous work [11, 32- 35] and we found it to be more than sufficient for the calculations to converge (Figure S4). Finally, for the calculations of NO adsorption in Pd- CHA, the number of QM atoms was extended as needed to account for potential interactions of NO with other Si/O atoms in the framework. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 88, 552, 111]]<|/det|> +## Enumeration of charge-exchange sites + +<|ref|>text<|/ref|><|det|>[[112, 128, 886, 544]]<|/det|> +Enumeration of charge- exchange sitesFor each structure with a unique Al arrangement, we survey the energy landscape by placing the adsorbate(s) at multiple initial positions near each oxygen atom neighboring Al atom(s), as shown in Figure S5. For structures with an isolated Al atom, this requires only 4 calculations for either \(\mathrm{Pd^{+}}\) or \(\mathrm{H^{+}}\) . For structures with an Al pair, finding the optimum position of a \(\mathrm{Pd^{+2}}\) requires 8 calculations per structure, and finding the optimum position for \(\mathrm{H^{+}H^{+}}\) requires 16 total calculations. The most complicated case is for a \(\mathrm{Pd^{+}H^{+}}\) site near an Al pair, where in addition to surveying for \(\mathrm{Pd^{+}}\) charge- exchange site, a proton must be present near the opposite Al atom to compensate for the missing charge. This situation requires a total of 32 calculations per structure. We note that these numbers reflect the maximum number of calculations attempted. For some very unfavorable initial positions, the calculations did not converge. This is mostly limited to the GGA level search and happens in less than \(10\%\) of the total calculations (for the most part, this is due to placing \(\mathrm{Pd}\) at the center of a 4MR). The extended number of initial positions in the scheme inevitably leads to some poor initial conditions. All generated structures and optimized geometries are available in the SI. + +<|ref|>sub_title<|/ref|><|det|>[[114, 579, 483, 601]]<|/det|> +## Survey of charge-exchange sites + +<|ref|>text<|/ref|><|det|>[[113, 618, 886, 913]]<|/det|> +Survey of charge- exchange sitesThe major part of the computational cost in the workflow (Figure 1) is associated with searching for the global minimum energy of the charge- exchange site, steps 4 and 5. For zeolites with multiple T- sites, the number of calculations is significant. Carrying out all the calculations at the range- separated hybrid functional level of theory is intractable; however, it has been shown previously that this level of theory is needed to reach close agreement with experimental values [36]. Accordingly, we first use the B97- D3 exchange functional (which is at the GGA level) as a filter to determine the 5 most favorable exchange positions for the cation/proton(s) per each unique Al arrangement. For those 5 candidates, further calculations are done at the range- separated hybrid level using the \(\omega\) B97X- D exchange functional to determine the most energetically favorable position of the cations per each unique Al + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 201]]<|/det|> +arrangement. We tested this approach on a number of structures by comparing the results between doing the full calculations using only \(\omega \mathrm{B97X - D}\) to the approach described earlier. We found this approach to yield virtually the same results with significant reduction in computational cost (Table S1). + +<|ref|>text<|/ref|><|det|>[[113, 209, 884, 262]]<|/det|> +In order to compare the stability of the Pd cation at different cation- exchanged sites, we calculated the energy of reaction ( \(\Delta E_{form}\) ) for the following two reactions: + +<|ref|>equation<|/ref|><|det|>[[308, 290, 880, 328]]<|/det|> +\[\mathrm{Pd}_{(g)} + \mathrm{H}^{+}\mathrm{Z}^{-} + \frac{1}{4}\mathrm{O}_{2(g)}\rightarrow \mathrm{Pd}^{+}\mathrm{Z}^{-} + \frac{1}{2}\mathrm{H}_{2}\mathrm{O}_{(g)} \quad (1)\] + +<|ref|>equation<|/ref|><|det|>[[256, 365, 880, 401]]<|/det|> +\[\mathrm{Pd}_{(g)} + \mathrm{H}^{+}\mathrm{H}^{+}\mathrm{Z}^{-2} + \frac{x}{4}\mathrm{O}_{2(g)}\rightarrow \mathrm{Pd}^{+x}\mathrm{H}^{+}_{2 - x}\mathrm{Z}^{-2} + \frac{x}{2}\mathrm{H}_{2}\mathrm{O}_{(g)} \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[112, 415, 885, 712]]<|/det|> +Equation 1 is used for an isolated Al zeolite and equation 2 for structures with an Al pair. \((x)\) is the oxidation state of Pd (either 1 or 2), \((\mathrm{Pd}_{(g)})\) is a gas phase Pd atom, \((\mathrm{H}^{+})\) is the compensating proton, \((\mathrm{Z})\) is the charged zeolite framework, \((\mathrm{H}_{2}\mathrm{O}_{(g)})\) and \((\mathrm{O}_{2(g)})\) are water and oxygen in the gas phase, and \((\mathrm{Pd}^{+x})\) is the Pd adsorbed in the zeolite framework. Equations 1 and 2 allow comparing the relative stability of sites with different Al configurations and oxidation states by using a consistent reference. The equations also rely on using a Brønsted site as a reference. This eliminates the impact of the thermodynamics of Al placement in the zeolite (which is kinetically driven during the synthesis of the zeolite [37]). For NO adsorption on Pd- exchanged CHA, we use the following equation to calculate the NO adsorption energy ( \(\Delta E_{\mathrm{NO}}\) ): + +<|ref|>equation<|/ref|><|det|>[[354, 744, 880, 766]]<|/det|> +\[\Delta E_{\mathrm{NO}} = E_{\mathrm{Pd}*\mathrm{NO}*} - E_{\mathrm{Pd}*} - E_{\mathrm{NO}_{(g)}} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[113, 797, 884, 910]]<|/det|> +where \((E_{\mathrm{Pd}*\mathrm{NO}*})\) is the total energy of NO adsorbed on the Pd- exchanged zeolite in the DFT calculation, \((E_{\mathrm{Pd}*})\) is the total energy of the Pd adsorbed on the zeolite framework and \((E_{\mathrm{NO}_{(g)}})\) is the total energy of NO in the gas phase. \(\mathrm{Pd}*\) refers to either \(\mathrm{Pd}^{+}, \mathrm{Pd}^{+}\mathrm{H}^{+}\) , or \(\mathrm{Pd}^{+2}\) . + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 85, 435, 110]]<|/det|> +## Results and Discussion + +<|ref|>text<|/ref|><|det|>[[114, 133, 883, 215]]<|/det|> +Results of Pd cation siting in CHA are discussed first followed by a discussion of cation- exchanged BEA. A comparison between the two types of zeolites is then made. Finally, we apply our results to NO adsorption on Pd- exchanged CHA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 251, 345, 274]]<|/det|> +## Pd-exchanged CHA + +<|ref|>text<|/ref|><|det|>[[112, 291, 885, 740]]<|/det|> +We start by reporting our results for the location of Pd cations exchanged into CHA [27, 28]. CHA has a single T- site, which limits the number of unique Al pairs. As mentioned previously, equations 1 and 2 are used to evaluate the stability of the sites. This requires, for each unique Al arrangement, finding the minimum energy of the compensating protons and the Pd cation. An example of the search results for the optimum proton location (based on sampling different initial positions for a given Al siting), is shown in Figure S6. Some locations can be more favorable than others by as much as 1 eV. Based on an examination of the stability of the 12 different Al arrangements in CHA, we do not find a clear indication of why certain positions stabilize protons more than others. One descriptor we find useful is the distance between the oxygen atoms where the protons adsorb ( \(\mathrm{d}_{O - O}\) ) (Figure S7). If the oxygen atoms are too close or too far (relative to Al- Al distance), the associated 2 proton configuration is not favorable. Intermediate distance almost always yields the most favorable arrangement. This descriptor can reduce the number of required calculations; however, it is not a substitute for performing the search through an approach such as the high throughput screening employed here, especially since many candidates have similar values of \(\mathrm{d}_{O - O}\) . + +<|ref|>text<|/ref|><|det|>[[113, 745, 885, 888]]<|/det|> +In addition, sample search results for the global minimum of \(\mathrm{Pd^{+}H^{+}}\) exchange are shown in Figure S8. As for the proton case discussed above, there is a large variance of the results depending on the initial position of the Pd cation and the proton. It is important to also note that the relative energy is sensitive to the proton position. For example, CHA- 3- \(\mathrm{Pd^{+}H^{+}}\) - 17 and CHA- 3- \(\mathrm{Pd^{+}H^{+}}\) - 21 both have the Pd at the center of the 6MR; however, the latter is 0.4 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 745, 110]]<|/det|> +eV less stable due to the proton occupying a different location (Figure S9). + +<|ref|>text<|/ref|><|det|>[[112, 119, 886, 565]]<|/det|> +A summary of the energies of \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) in CHA, comparing unique Al positions, is given in Figures 3 and 4, respectively. In both cases, the stability of the Pd cation is heavily dependent on the Al positions, with energies varying by as much as several eV. The range of energies in \(\mathrm{Pd^{+}H^{+}}\) is much more closely spaced ( \(< 0.8 \mathrm{eV}\) ) compared to \(\mathrm{Pd^{+2}}\) (around \(3 \mathrm{eV}\) ). Al pairs in the 6MR arrangements, especially in the NNNN position, provide the most favorable host for the Pd cation, in which case the Pd resides at the center of the 6MR (Figure 5a and b and Figure S10a and b). This finding is consistent with other recent studies [27, 28]. This geometry provides the most number of oxygen atoms in close proximity to the cation, but not too close. Surprisingly, the Pd cation is then most stable either at the isolated site or when the two Al pairs do not share the same ring. In these cases, the Pd cation mostly resides at the center of the 6MR (Figure S10c- f). For both \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , Al pairs in the 8MR and 4MR provide unfavorable arrangements, especially for the latter. In the 8MR (Figure S10g and h), unlike the 6MR, the two Al atoms are farther apart and there is a lack of neighboring Si/O atoms to provide orbital overlap to stabilize the cation. In the case of the 4MR, the atoms are too closely spaced. + +<|ref|>text<|/ref|><|det|>[[113, 573, 885, 713]]<|/det|> +Despite some attempts to do, we were not able to find a simple descriptor related to \(\Delta E_{form}\) (e.g. Al- Al distance and Si/O atoms in close proximity) for the data contained in Figures 3 and 4. For \(\mathrm{Pd^{+}H^{+}}\) , in almost all calculations, a minimum distance of \(4 \mathrm{\AA}\) separates the Pd cation and the proton in the optimized structure (Figure S11). This indicates repulsive interaction between the two cations at shorter distances. + +<|ref|>text<|/ref|><|det|>[[113, 722, 884, 804]]<|/det|> +Comparing \(\Delta E_{form}\) between \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) on CHA shows all Al pairs arrangements are more favorable to \(\mathrm{Pd^{+}H^{+}}\) compared to \(\mathrm{Pd^{+2}}\) , with the exceptions of Al pairs in 6MR. This has implications for the adsorption of guest molecules, as will be discussed later on. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 276, 870, 635]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 657, 883, 730]]<|/det|> +
Figure 3: Formation Energy of Pd+H+ on CHA. Each bar represents unique Al location(s). The color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in an NNN configuration or the isolated site and striped bars refer to the NNNN configuration.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 272, 860, 633]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 657, 883, 732]]<|/det|> +
Figure 4: Formation energy of \(\mathrm{Pd^{+2}}\) on CHA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. Solid bars refer to Al pairs in NNN configuration and striped bars refer to NNNN configuration. The two most stable sites (CHA-7 and CHA-3) are illustrated in Figure 5.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 90, 861, 425]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 441, 882, 481]]<|/det|> +
Figure 5: QM atoms of the optimized calculations. Color coding: red=Oxygen, white=Hydrogen, blue=Palladium and beige=Silicon
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 503, 343, 525]]<|/det|> +## Pd-exchanged BEA + +<|ref|>text<|/ref|><|det|>[[112, 544, 886, 900]]<|/det|> +BEA has a more diverse set of rings (4,5,6 and 12), 9 T- sites, and is denser than CHA. This makes it more challenging and nontrivial to determine the favorable sites for the cations. Similar to CHA, we carried out calculations for \(\mathrm{Pd^{+}}\) , \(\mathrm{Pd^{+}H^{+}}\) , and \(\mathrm{Pd^{+2}}\) charge- exchanged into BEA using the approach shown in Figure 1. Figure 6 summarizes the results for \(\mathrm{Pd^{+2}}\) in BEA. It is noticeable that the four most energetically favorable structures all have Al pairs in 6MR (images of the QM atoms are shown in Figure 5c- f). The four structures are separated by 0.15- 0.6 eV. Structure BEA- 65 (Figure 5c), the most favorable energetically, has Al pairs in a 6MR in NNN configuration. The Pd cation resides at the center of the 6MR in close proximity to 4 neighboring oxygen atoms. The three structures closest in energy (BEA- 80, BEA- 45 and BEA- 41) have similar configurations, but differ mainly in Al placements within the 6MR and the Si/O atoms surrounding the 6MR. These four structures are followed by a number of structures where Al pairs are in a 5MR (e.g. Figure S12a). Those 5MR structures + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 383]]<|/det|> +are close in energy (separated by less than 0.15 eV). In all of these cases, the Pd cation is most stable at the center of the 5MR. Al pairs in 4MR (e.g. BEA- 36) are poor hosts for the Pd cation, similar to what was observed for CHA. For Al pairs not in a 4,5 or 6MR, \(\Delta E_{form}\) is considerably lower. The most stable structure for those cases (BEA- 55, Figure S12b) is 1.2 eV weaker compared to the most stable structure. This reinforces the results observed in CHA (Figure 4) where Al pairs in the same \(n\) - MR allow for additional stability of the cation. This likely stems from having neighboring oxygen atoms in positions favorable for orbital overlap with the Pd cation. However, this observation fails to explain why some of the other structures with Al pairs in a 6MR or 5MR arrangements (e.g. BEA- 62 and BEA- 59) have significantly smaller values of \(\Delta E_{form}\) . This subject will be discussed further below. + +<|ref|>image<|/ref|><|det|>[[120, 400, 865, 680]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 691, 884, 766]]<|/det|> +
Figure 6: Formation energy of \(\mathrm{Pd^{+2}}\) on BEA. Each bar represents unique Al locations. Color coding refers to the type of Al pairs in the zeolite matrix. Patterned bars refer to Al pairs in NNNN positions while solid bars refer to Al pairs in NNN positions. The four most stable sites are illustrated in Figure 5.
+ +<|ref|>text<|/ref|><|det|>[[114, 795, 884, 906]]<|/det|> +Figure 7 summarizes the formation energies for \(\mathrm{Pd^{+}}\) and \(\mathrm{Pd^{+}H^{+}}\) on BEA. Structure BEA- 65 is the most energetically favorable site in BEA for \(\mathrm{Pd^{+}H^{+}}\) . It is the same site that hosts the most stable \(\mathrm{Pd^{+2}}\) , for which the cation is located at the center of the 6MR and the Al pairs are in NNN arrangement (Figure 5g). It is \(>0.25 \mathrm{eV}\) more favorable than + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 473]]<|/det|> +any of the other structures examined. Surprisingly, structures BEA- 33 and BEA- 55 (Figure S12c and d), in which the Al pairs are do not share a 5MR or 6MR, are more stable than the other Al pairs in 5MR or 6MR. However, after BEA- 65, most of the structures are close in energy (the difference can be \(< 0.1 \mathrm{eV}\) ). Given DFT errors and the large number of structures close in energies, it is difficult to make a conclusion on the order between many of these structures. Nevertheless, there is a lack of clear distinctions between Al pairs in the same \(n\) - MR observed for \(\mathrm{Pd}^{+2}\) in CHA and BEA. Some Al configurations (e.g. BEA- 8) can have a high \(\Delta E_{form}\) for \(\mathrm{Pd}^{+2}\) but not for \(\mathrm{Pd}^{+}\mathrm{H}^{+}\) . This indicates that results are not transferable between different oxidation states. It also sheds light on potential importance of sites where Al pairs do not share a 5MR or 6MR, a motif that does not generally receive much attention in computational studies, and the role they can play in adsorption/catalysis. In these structures, the presence of a neighboring Al atom in close proximity can significantly alter the \(\Delta E_{form}\) compared to their respective value for isolated sites (Figure S13). + +<|ref|>image<|/ref|><|det|>[[117, 491, 872, 747]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 760, 883, 834]]<|/det|> +
Figure 7: Formation energy of \(\mathrm{Pd}^{+}\) on BEA. Each bar represents unique Al location(s). Color coding refers to the type of Al pairs or isolated Al in the zeolite matrix. In Al pairs, patterned bars refer to Al pairs in NNNN positions and solid colors refer to Al pairs in NNN positions. The most stable site (BEA-65) is illustrated in Figure 5.
+ +<|ref|>text<|/ref|><|det|>[[115, 862, 883, 912]]<|/det|> +During the search for the most favorable charge- exchange sites, we found that in some cases the energetically most favorable Pd cation position may not involve situating the cation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 230]]<|/det|> +within a single ring, even if the Al pairs are in the same MR. Examples of this situation are BEA- 51, BEA- 78, and BEA- 52, shown in Figure S14. These positions can be \(>0.2 \mathrm{eV}\) more stable than the Pd at the center of the 6MR. This highlights the importance of the high throughout screening approach, which can find the optimum adsorption location when it is not the most physically intuitive. + +<|ref|>text<|/ref|><|det|>[[113, 240, 885, 442]]<|/det|> +Although the energetic order of Al configurations in BEA does not correlate directly for Pd +1 and +2 oxidation states, we attempted to confirm whether this is true or not for \(\mathrm{Co}^{+2}\) cations. We performed a limited number of calculations on \(\mathrm{Co}^{+2}\) on CHA and found a similar trend in terms of favorable Al pairs (Figure 8). This indicates that the results for one cation in a zeolite matrix are can be transferable to other, chemically similar, cations exchanged into the same zeolite. This finding is important, since the UV- Vis spectrum of \(\mathrm{Co}^{2 + }\) is used to identify the location of divalent cation in zeolites [38, 39]. + +<|ref|>image<|/ref|><|det|>[[180, 470, 800, 787]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[150, 810, 845, 831]]<|/det|> +
Figure 8: Comparison between favorable Al pair configurations for \(\mathrm{Pd}^{+2}\) and \(\mathrm{Co}^{+2}\)
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 87, 710, 111]]<|/det|> +## Comparison between Pd-exchanged CHA and BEA + +<|ref|>text<|/ref|><|det|>[[113, 128, 886, 331]]<|/det|> +Comparison between Pd- exchanged CHA and BEAIn general, there are similarities in Pd- exchanged BEA and CHA. In both cases, the Pd cation (both as a \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+}H^{+}}\) ) is energetically most favorable in the 6MR. For \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , the site with the highest \(\Delta E_{form}\) has almost identical energy values (around - 3.3 eV and - 4.2, respectively). The 6MR in BEA is more oval shaped (Figure 5c) compared to that of CHA (Figure 5a), placing the oxygen atoms at a closer distance. The four oxygens close to the Pd cation in the 6MR are on average located 2.0 Å from the Pd in BEA and 2.2 Å in CHA. This, however, does not seem to impact the formation energy. + +<|ref|>sub_title<|/ref|><|det|>[[115, 367, 296, 389]]<|/det|> +## NO Adsorption + +<|ref|>text<|/ref|><|det|>[[113, 407, 886, 762]]<|/det|> +NO AdsorptionThe location of Pd cations in CHA and BEA has practical implications for the capability of the zeolite to act as a passive \(\mathrm{NO}_x\) adsorber (PNA) during the cold start of automobiles. The nature of the oxidation state of the Pd cations and their location in the zeolite continue to be debated in the scientific literature [27, 28, 40]. Here, we attempt to shed some light on the subject based on the results of our high throughput screening. We focus on NO adsorption on Pd- exchanged- CHA, considering \(\mathrm{Pd^{+}}\) , \(\mathrm{Pd^{+}H^{+}}\) , and \(\mathrm{Pd^{+2}}\) . Figure 9 shows the correlation results between \(\Delta \mathrm{E}_{\mathrm{NO}}\) and \(\Delta E_{form}\) . Generally, a weaker \(\Delta E_{form}\) correlates with a stronger NO adsorption energy. Similar to most adsorption processes, the stronger the binding energy of the cation site, the less electron density is available to bind the guest gas specie [41]. Although there appears to be a linear correlation with a low mean absolute error (MAE) (0.1 and 0.22 eV for NO adsorption on \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) , respectively), there are clear outliers (e.g. one of two 4 MR in \(\mathrm{Pd^{+2}}\) is 0.5 eV off the line). + +<|ref|>text<|/ref|><|det|>[[114, 770, 884, 913]]<|/det|> +Figure 9 demonstrates that there is a clear distinction in NO adsorption between \(\mathrm{Pd^{+}H^{+}}\) and \(\mathrm{Pd^{+2}}\) . For a similar \(\Delta E_{form}\) , \(\mathrm{E}_{\mathrm{NO}}\) in \(\mathrm{Pd^{+}H^{+}}\) is much stronger (by as much as 1 eV). For most of the Al pair arrangements, the NO binds more strongly to \(\mathrm{Pd^{+}H^{+}}\) compared to \(\mathrm{Pd^{+2}}\) . We also find NO adsorption on \(\mathrm{Pd^{+}H^{+}}\) to be stronger when the \(\mathrm{Pd^{+}}\) and NO unpaired electrons are paired (compared to two unpaired electrons) (Table S2). This might + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 200]]<|/det|> +be one reason for the stronger binding on \(\mathrm{Pd^{+}H^{+}}\) (given that NO adsorption on \(\mathrm{Pd^{+2}}\) has an unpaired NO electron). Experimentally, it is unclear if \(\mathrm{Pd^{+}H^{+}}\) is present under operating conditions; however, Figure 9 indicates that if present, it is a superior NO adsorption site compared to \(\mathrm{Pd^{+2}}\) . + +<|ref|>text<|/ref|><|det|>[[113, 209, 885, 504]]<|/det|> +Figure 9 also shows that Al pairs in the 6MR arrangement (especially in NNNN configuration), which has been discussed most extensively in the literature because they hold \(\mathrm{Pd^{+2}}\) cations most stably, are weaker sites for NO adsorption compared to other Al arrangements. This is not unexpected since the more stable the Pd, the more weakly it can bind to a guest molecule. It is striking, however, that the difference in \(\Delta \mathrm{E}_{\mathrm{NO}}\) compared to the other Al arrangements (0.25- 1.5 eV). Figure 9 also highlights how many of the CHA sites are very close in energy, especially for \(\mathrm{Pd^{+}H^{+}}\) . Therefore, it is very difficult based on the small differences in energy values to discern spectroscopic data and assign them to specific sites. The results indicate that an ensemble of sites of very similar energies contribute similarly to the adsorption of NO. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 92, 872, 394]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 406, 884, 500]]<|/det|> +
Figure 9: NO Binding energy versus Pd formation energy on CHA. The color coding refers to the type of Al in the \(n\) -MR. The marker shapes (□ and \(\circ\) ) represent \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+} / Pd^{+}H^{+}}\) , respectively. The black and blue lines are based on fitting data for \(\mathrm{Pd^{+2}}\) and \(\mathrm{Pd^{+} / Pd^{+}H^{+}}\) , respectively. Filled markers are used for Al pairs in NNNN configuration and half-filled markers are for isolated Al or Al in NNN configurations
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 563, 268, 588]]<|/det|> +## Conclusion + +<|ref|>text<|/ref|><|det|>[[113, 610, 884, 906]]<|/det|> +We developed a high throughput screening approach for enumerating unique Al arrangements in zeolites, examined possible cation- exchanged sites, calculated energies via density functional theory (DFT) for stabilizing Pd cations in different locations, and determined favorable locations for the positioning of metal cations in a zeolite. Pd adsorption on CHA and BEA was used to illustrate the approach. After conducting thousands of individual calculations, we arrived at Al configurations that allow optimum Pd cation stability. Sites with Al in 6MR bind Pd cations most strongly, consistent with prior work [27, 28]. However, high throughput screening identifies other Al configurations, especially for \(\mathrm{Pd^{+}H^{+}}\) , that bind Pd cations very stably. Many of these arrangements have not been considered previously for CHA and BEA. In addition, the most stable location for Pd is not always the most + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 885, 352]]<|/det|> +physically intuitive (e.g., inside a 5MR or 6MR). Lastly, we examined the impact of the high throughput approach on NO adsorption on Pd- exchanged- CHA. We generally find \(\mathrm{Pd^{+}H^{+}}\) to be more stable on CHA compared to \(\mathrm{Pd^{+2}}\) . Although Al pairs in 6MR are very stable sites for the Pd cation, NO adsorption on these sites is weaker compared to that for other Al configurations examined. This identifies sites that might play a critical role in the adsorption of NO in a PNA. Finally, we note that the DFT- based high throughput screening reported here provides a robust systematic path for identifying the most energetically favored cation exchange sites. While the method is applied here to zeolites, it could be extended to identify the energetically preferred location of cations in many other materials (e.g., MOFs). + +<|ref|>sub_title<|/ref|><|det|>[[115, 394, 238, 418]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 450, 392, 472]]<|/det|> +## Theoretical calculations + +<|ref|>text<|/ref|><|det|>[[113, 490, 885, 754]]<|/det|> +A hybrid quantum mechanics/molecular mechanics (QM/MM) approach was used to model the zeolite structure. Detailed implementation of this model can be found elsewhere [16]. The QM/MM approach has proven to account for long- range Coulombic and dispersive interactions, which are critical in describing the zeolite framework interactions with adsorbates [36]. Many studies have shown that the QM/MM approach gives a good prediction of experimental data for different zeolites and adsorbates [16, 32- 34, 36, 42, 43]. This approach is also computationally more efficient than periodic calculations since it requires a smaller number of QM atoms, especially when the unit cell contains a large number of atoms (e.g. BEA). All QM/MM calculations were done with a development version of Q- Chem [44]. + +<|ref|>text<|/ref|><|det|>[[113, 763, 885, 904]]<|/det|> +The adsorbate(s) and a cluster encompassing the active site are described by QM and the rest of the zeolite is modeled by MM using a standard force field of the CHARMM type with the P2 parameter set [17]. During structural optimization, the QM region is allowed to relax while the MM region is fixed. The B97- D3 exchange functional [45], a generalized gradient approximation (GGA) exchange functional, is used as an initial filter + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 262]]<|/det|> +to determine the optimum site of the cation followed by calculations done with the more accurate range- separated hybrid functional, \(\omega \mathrm{B97X - D}\) [46], which was shown to be among the best performing hybrid functionals in a benchmarking study [18]. An effective core potential was used on Pd atom. For each structure, the def2- SV(P) basis set was used to obtain the optimized structure geometry, and further energy refinement was done using a single- point calculation at the def2- TZVPD level of theory [47]. + +<|ref|>sub_title<|/ref|><|det|>[[115, 297, 275, 319]]<|/det|> +## Zeolite model + +<|ref|>text<|/ref|><|det|>[[113, 338, 886, 570]]<|/det|> +The crystallographic structure of CHA and BEA were obtained from the International Zeolite Association (IZA) database [48]. Cluster models containing 696 and 810 tetrahedral atoms (T696 and T810) were used to model CHA and BEA topologies, respectively (Figure S15). The CHA cluster is based on the previous work from or group[27]. While earlier work has suggested that a 100 T- atom cluster model is sufficient [32], larger cluster models are used here due to the marginal additional computational cost and the extended active site region in some of the calculations. Each cluster was terminated with hydrogen atoms replacing terminal oxygen atoms. + +<|ref|>sub_title<|/ref|><|det|>[[115, 607, 315, 629]]<|/det|> +## Data Availability + +<|ref|>text<|/ref|><|det|>[[115, 648, 884, 699]]<|/det|> +The data that support the results within this paper and other findings of this study are available in the supporting information. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 86, 266, 110]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[120, 130, 886, 880]]<|/det|> +1. Guisnet, M. & Gilson, J.-P. Zeolites for cleaner technologies (Imperial College Press London, 2002).2. Corma, A. From microporous to mesoporous molecular sieve materials and their use in catalysis. Chemical reviews 97, 2373-2420 (1997).3. Kulprathipanja, S. Zeolites in industrial separation and catalysis (John Wiley & Sons, 2010).4. Paolucci, C. et al. Catalysis in a cage: Condition-dependent speciation and dynamics of exchanged cu cations in ssz-13 zeolites. Journal of the American Chemical Society 138, 6028-6048. ISSN: 15205126 (2016).5. Janda, A., Vlaisavljevich, B., Lin, L.-C., Smit, B. & Bell, A. T. Effects of zeolite structural confinement on adsorption thermodynamics and reaction kinetics for monomolecular cracking and dehydrogenation of n-butane. Journal of the American Chemical Society 138, 4739-4756 (2016).6. Norskov, J. K., Studt, F., Abild-Pedersen, F. & Bligaard, T. Fundamental concepts in heterogeneous catalysis (John Wiley & Sons, 2014).7. Van Santen, R. A. Modern heterogeneous catalysis: an introduction (John Wiley & Sons, 2017).8. Olson, D. H., Khosrovani, N., Peters, A. W. & Toby, B. H. Crystal structure of dehydrated CsZSM-5 (5.8 Al): evidence for nonrandom aluminum distribution. The Journal of Physical Chemistry B 104, 4844-4848 (2000).9. Mentzen, B. & Sacerdote-Peronnet, M. Prediction of preferred proton locations in HMFI/benzene complexes by molecular mechanics calculations. Comparison with NMR, structural and calorimetric results. Materials research bulletin 29, 1341-1348 (1994). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 171]]<|/det|> +10. SUN, X.-L., HUANG, C.-P., ZHANG, J. & CHEN, B.-H. Location of Al and Acid Strength of Br2nsted Acid in Beta Zeolite. Acta Physico-Chimica Sinica 25, 1136–1142 (2009). + +<|ref|>text<|/ref|><|det|>[[113, 185, 884, 268]]<|/det|> +11. Mansoor, E., Head-Gordon, M. & Bell, A. T. Computational modeling of the nature and role of Ga species for light alkane dehydrogenation catalyzed by Ga/H-MFI. ACS Catalysis 8, 6146–6162 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 282, 884, 365]]<|/det|> +12. Ghorbanpour, A., Rimer, J. D. & Grabow, L. C. Periodic, vdW-corrected density functional theory investigation of the effect of Al siting in H-ZSM-5 on chemisorption properties and site-specific acidity. Catalysis Communications 52, 98–102 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 379, 883, 431]]<|/det|> +13. Jones, A. J. & Iglesia, E. The Strength of Br2nsted Acid Sites in Microporous Aluminosilicates. ACS Catalysis 5, 5741–5755. ISSN: 21555435 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 445, 884, 528]]<|/det|> +14. Xing, B., Ma, J., Li, R. & Jiao, H. Location, distribution and acidity of Al substitution in ZSM-5 with different Si/Al ratios—a periodic DFT computation. Catalysis Science & Technology 7, 5694–5708 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 542, 884, 623]]<|/det|> +15. Zhang, N., Liu, C., Ma, J., Li, R. & Jiao, H. Determining the structures, acidity and adsorption properties of Al substituted HZSM-5. Physical Chemistry Chemical Physics 21, 18758–18768 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 638, 884, 720]]<|/det|> +16. Zimmerman, P. M., Head-Gordon, M. & Bell, A. T. Selection and validation of charge and Lennard-Jones parameters for QM/MM simulations of hydrocarbon interactions with zeolites. Journal of chemical theory and computation 7, 1695–1703 (2011). + +<|ref|>text<|/ref|><|det|>[[113, 735, 884, 879]]<|/det|> +17. Li, Y.-P., Gomes, J., Mallikarjun Sharada, S., Bell, A. T. & Head-Gordon, M. Improved force-field parameters for QM/MM simulations of the energies of adsorption for molecules in zeolites and a free rotor correction to the rigid rotor harmonic oscillator model for adsorption enthalpies. The Journal of Physical Chemistry C 119, 1840–1850 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 171]]<|/det|> +18. Mardirossian, N. & Head-Gordon, M. Thirty years of density functional theory in computational chemistry: an overview and extensive assessment of 200 density functionals. Molecular Physics 115, 2315–2372 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 185, 884, 266]]<|/det|> +19. Kim, J., Abouelnasr, M., Lin, L.-C. & Smit, B. Large-scale screening of zeolite structures for CO2 membrane separations. Journal of the American Chemical Society 135, 7545–7552 (2013). + +<|ref|>text<|/ref|><|det|>[[113, 281, 884, 363]]<|/det|> +20. Lee, Y. et al. High-throughput screening approach for nanoporous materials genome using topological data analysis: application to zeolites. Journal of chemical theory and computation 14, 4427–4437 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 378, 883, 428]]<|/det|> +21. Kim, J. et al. Large-scale computational screening of zeolites for ethane/ethene separation. Langmuir 28, 11914–11919 (2012). + +<|ref|>text<|/ref|><|det|>[[113, 442, 884, 523]]<|/det|> +22. Slawek, A., Grzybowska, K., Vicent-Luna, J. M., Makowski, W. & Calero, S. Adsorption of Cyclohexane in Pure Silica Zeolites: High-Throughput Computational Screening Validated by Experimental Data. ChemPhysChem 19, 3364–3371 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 538, 884, 619]]<|/det|> +23. Siepmann, J. I., Bai, P., Tsapatsis, M., Knight, C. & Deem, M. W. Discovery of optimal zeolites for challenging separations and chemical conversions through predictive materials modeling. APS 2015, B16–003 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 634, 884, 685]]<|/det|> +24. Chen, H.-Y. et al. Low temperature NO storage of zeolite supported Pd for low temperature diesel engine emission control. Catalysis Letters 146, 1706–1711 (2016). + +<|ref|>text<|/ref|><|det|>[[113, 700, 884, 750]]<|/det|> +25. Gu, Y. & Epling, W. S. Passive NOx adsorber: An overview of catalyst performance and reaction chemistry. Applied Catalysis A: General 570, 1–14 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 765, 884, 816]]<|/det|> +26. Ji, Y., Bai, S. & Crocker, M. Al2O3-based passive NOx adsorbers for low temperature applications. Applied Catalysis B: Environmental 170, 283–292 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 831, 886, 911]]<|/det|> +27. Van der Mynsbruggc, J., Head-Gordon, M. & Bell, A. T. Computational modeling predicts the stability of both Pd+ and Pd 2+ ion-exchanged into H-CHA. Journal of Materials Chemistry A 9, 2161–2174 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 894, 140]]<|/det|> +28. Mandal, K. et al. Condition-Dependent Pd Speciation and NO Adsorption in Pd/Zeolites. ACS Catalysis 10, 12801-12818 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 155, 884, 238]]<|/det|> +29. Loewenstein, W. The distribution of aluminum in the tetrahedra of silicates and aluminates. American Mineralogist: Journal of Earth and Planetary Materials 39, 92-96 (1954). + +<|ref|>text<|/ref|><|det|>[[113, 252, 884, 365]]<|/det|> +30. Verstraelen, T., Van Speybroeck, V. & Waroquier, M. ZEOBUILDER: A GUI Toolkit for the Construction of Complex Molecular Structures on the Nanoscale with Building Blocks. Journal of Chemical Information and Modeling 48, 1530-1541. ISSN: 1549-9596. http://pubs.acs.org/doi/abs/10.1021/ci8000748 (2008). + +<|ref|>text<|/ref|><|det|>[[113, 379, 884, 461]]<|/det|> +31. Larsen, A. H. et al. The atomic simulation environment—a Python library for working with atoms. Journal of Physics: Condensed Matter 29, 273002. http://stacks.iop.org/0953-8984/29/i=27/a=273002 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 476, 885, 617]]<|/det|> +32. Gomes, J., Zimmerman, P. M., Head-Gordon, M. & Bell, A. T. Accurate prediction of hydrocarbon interactions with zeolites utilizing improved exchange-correlation functionals and QM/MM methods: benchmark calculations of adsorption enthalpies and application to ethene methylation by methanol. The Journal of Physical Chemistry C 116, 15406-15414 (2012). + +<|ref|>text<|/ref|><|det|>[[113, 632, 884, 715]]<|/det|> +33. Mallikarjun Sharada, S., Zimmerman, P. M., Bell, A. T. & Head-Gordon, M. Insights into the kinetics of cracking and dehydrogenation reactions of light alkanes in H-MFI. The Journal of Physical Chemistry C 117, 12600-12611 (2013). + +<|ref|>text<|/ref|><|det|>[[113, 729, 883, 812]]<|/det|> +34. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Theoretical study of 4-(hydroxymethyl) benzoic acid synthesis from ethylene and 5-(hydroxymethyl) furoic acid catalyzed by Sn-BEA. ACS Catalysis 6, 5052-5061 (2016). + +<|ref|>text<|/ref|><|det|>[[113, 826, 884, 909]]<|/det|> +35. Dinda, S., Govindasamy, A., Genest, A. & Rösch, N. Modeling Catalytic Steps on Extra-Framework Metal Centers in Zeolites. A Case Study on Ethylene Dimerization. The Journal of Physical Chemistry C 118, 25077-25088 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 170]]<|/det|> +36. Mansoor, E., Van der Mynsbrugg, J., Head-Gordon, M. & Bell, A. T. Impact of long-range electrostatic and dispersive interactions on theoretical predictions of adsorption and catalysis in zeolites. Catalysis Today 312, 51-65 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 184, 884, 267]]<|/det|> +37. Sastre, G., Fornes, V. & Corma, A. On the preferential location of Al and proton siting in zeolites: a computational and infrared study. The Journal of Physical Chemistry B 106, 701-708 (2002). + +<|ref|>text<|/ref|><|det|>[[112, 280, 884, 363]]<|/det|> +38. Dedecek, J. & Wichterlová, B. Co2+ Ion Siting in Pentasil-Containing Zeolites. I. Co2+ Ion Sites and Their Occupation in Mordenite. A Vis- NIR Diffuse Reflectance Spectroscopy Study. The Journal of Physical Chemistry B 103, 1462-1476 (1999). + +<|ref|>text<|/ref|><|det|>[[112, 376, 884, 457]]<|/det|> +39. Dedecek, J., Sobalik, Z. & Wichterlová, B. Siting and distribution of framework aluminium atoms in silicon-rich zeolites and impact on catalysis. Catalysis Reviews 54, 135-223 (2012). + +<|ref|>text<|/ref|><|det|>[[112, 472, 884, 524]]<|/det|> +40. Khivantsev, K. et al. Stabilization of super electrophilic Pd+ 2 cations in small-pore SSZ-13 zeolite. The Journal of Physical Chemistry C 124, 309-321 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 538, 896, 590]]<|/det|> +41. Abild-Pedersen, F. et al. Scaling properties of adsorption energies for hydrogen-containing molecules on transition-metal surfaces. Physical review letters 99, 016105 (2007). + +<|ref|>text<|/ref|><|det|>[[112, 604, 884, 686]]<|/det|> +42. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Computational study of p-xylene synthesis from ethylene and 2, 5-dimethylfuran catalyzed by H-BEA. The Journal of Physical Chemistry C 118, 22090-22095 (2014). + +<|ref|>text<|/ref|><|det|>[[112, 700, 884, 781]]<|/det|> +43. Li, Y.-P., Head-Gordon, M. & Bell, A. T. Analysis of the reaction mechanism and catalytic activity of metal-substituted beta zeolite for the isomerization of glucose to fructose. Acs Catalysis 4, 1537-1545 (2014). + +<|ref|>text<|/ref|><|det|>[[112, 795, 884, 847]]<|/det|> +44. Shao, Y. et al. Advances in molecular quantum chemistry contained in the Q-Chem 4 program package. Molecular Physics 113, 184-215 (2015). + +<|ref|>text<|/ref|><|det|>[[112, 861, 884, 913]]<|/det|> +45. Grimme, S. Semiempirical GGA-type density functional constructed with a long-range dispersion correction. Journal of computational chemistry 27, 1787-1799 (2006). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 170]]<|/det|> +46. Chai, J.-D. & Head-Gordon, M. Long-range corrected hybrid density functionals with damped atom-atom dispersion corrections. Physical Chemistry Chemical Physics 10, 6615-6620 (2008). + +<|ref|>text<|/ref|><|det|>[[113, 185, 884, 267]]<|/det|> +47. Weigend, F. & Ahlrichs, R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence quality for H to Rn: Design and assessment of accuracy. Physical Chemistry Chemical Physics 7, 3297-3305 (2005). + +<|ref|>text<|/ref|><|det|>[[113, 283, 884, 334]]<|/det|> +48. Meier, W. M., Olson, D. H. & Baerlocher, C. Atlas of the zeolite structure types. Zeolites 17 (1996). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 85, 366, 110]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[112, 132, 886, 457]]<|/det|> +AcknowledgementThis material is based upon work supported by the U.S. Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE) under the Vehicle Technologies Program Award Number DE- EE0008213. This research used resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE- AC02- 05CH11231. The authors also thank UC Berkeley's Molecular Graphics and Computation Facility (supported by NIH S10OD023532) for their computational resources. H.A. would like Saudi Aramco for their funding, and MHG acknowledges funding from the National Institutes of Health under Grant No. 5U01GM121667. This work used the Extreme Science and Engineering Discovery Environment (XSEDE) Comet at the service- provider through allocation TG- CE200085. + +<|ref|>sub_title<|/ref|><|det|>[[116, 500, 420, 525]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[113, 549, 884, 660]]<|/det|> +Author ContributionsH.A and A.T.B conceptualized the project. H.A, M.H.G and A.T.B developed the methodology. H.A. wrote the code, performed the calculations, and wrote the original draft. All authors participated in data analysis and editing the manuscript. M.H.G and A.T.B supervised the research reported in the paper. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[114, 85, 585, 111]]<|/det|> +# Supporting Information Available + +<|ref|>title<|/ref|><|det|>[[115, 164, 882, 221]]<|/det|> +# Assessing the stability of Pd-exchanged sites in zeolites with the aid of a high throughput workflow + +<|ref|>text<|/ref|><|det|>[[377, 254, 619, 277]]<|/det|> +Supporting Information + +<|ref|>text<|/ref|><|det|>[[137, 343, 857, 469]]<|/det|> +Hassan Aljama†, Martin Head- Gordon‡ and Alexis T. Bell\*,\† † Department of Chemical and Biomolecular Engineering, University of California, Berkeley, California ‡Department of Chemistry, University of California, Berkeley, California + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 261]]<|/det|> +All energies and xyz coordinates of the optimized geometries are available in the electronic supporting information (ESI). The nomenclature used to tabulate the data is as follow: zeolite name (CHA or BEA) - a number representing structure with a unique Al arrangement - adsorbate name (Pd+2, Pd+H+, Pd+, H+ or H+H+) - a number representing an initial position of the adsorbate - level of theory (GGA or hGGA) - type of calculations (opt for optimization and sp for single point calculation). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[235, 95, 780, 512]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 528, 883, 621]]<|/det|> +
Figure S1: Images of the QM atoms in four BEA structures. BEA-2 structure is shown in a. alongside a similar structure in b. BEA-11 is shown in c. alongside a similar structure in d. The structures similar to BEA-2 and BEA-11 (b. and d., respectively) are not exactly identical, as calculated by the nuclear repulsion energy, but share the same connectivity to surrounding Si and O atoms (types of MR). The same color coding as in Figure 5.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[230, 92, 760, 360]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 377, 882, 435]]<|/det|> +
Figure S2: Example of two BEA clusters (a. and b.) where Al atoms are not in the same MR and are on opposite sides of the open cage. The same color coding as in Figure 5, however, for clarity purposes, Al atoms are enlarged and are shown in green
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 100, 870, 275]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 287, 883, 455]]<|/det|> +
Figure S3: Flow diagram for determining if Al atom is in an X MR (where X refers to the length of the MR). First, starting with an Al atom, Si/Al neighbors (NN, NNN, NNNN) are identified similar to the procedure described in Figure 2. This is done up to the X+1 neighbor. Based on the neighbors, lists are developed (an example of a list is [Al, NN1, NNN11, ..], where NN1 refers to the first NN of the starting Al and NNN11 is the first NN to NN1). For each list, the following checks are made: first and last element of the list are the same (the starting Al atom), no element in the list (with the exception of the starting Al atom) appears twice in the list, and no sub-list of the list form a MR smaller than X. If a list passes those checks, then Al is part of XMR.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[216, 100, 761, 424]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 445, 884, 500]]<|/det|> +
Figure S4: Impact of the number of atoms selected in the QM region on the convergence of the QM/MM calculations ( \(\Delta E_{form}\) is defined in equation 2). Calculations were done using \(\omega\) B97X-D functional.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 106, 864, 328]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 361, 883, 455]]<|/det|> +
Figure S5: Example of a. Pd cation and b. proton placement near the four oxygens neighboring Al atom. For clarity, only Si in NN position and O neighboring Al are shown. The position of cation/proton is determined by finding the middle distance between neighboring oxygen, and then adding a displacement from the Al atom (usually 1-1.5 A). The same color coding as in Figure 5.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 883, 216]]<|/det|> +Table S1: Comparison of the stability order (from most stable to 5th most stable) on 3 different structures (BEA-9, BEA-52 and BEA-6) at two levels of theory (GGA and hGGA). The adsorbate is \(\mathrm{Pd^{+}H^{+}}\) . For simplicity, only the 5 most stable structures are shown out of the 32. The two results do not always yield the same order, however, the most stable structure at the hGGA level always appears among the 5 most stable structures at the GGA level. This means following the approach in Figure 1 yields the same results as doing all the calculations using \(\omega \mathrm{B97X - D}\) . + +<|ref|>table<|/ref|><|det|>[[120, 226, 877, 432]]<|/det|> + +
StabilityTheory LevelBEA-9BEA-52BEA-63
1stGGABEA-9-Pd+H+-3BEA-52-Pd+H+-24BEA-63-Pd+H+-26
hGGABEA-9-Pd+H+-12BEA-52-Pd+H+-24BEA-63-Pd+H+-25
2ndGGABEA-9-Pd+H+-11BEA-52-Pd+H+-4BEA-63-Pd+H+-25
hGGABEA-9-Pd+H+-11BEA-52-Pd+H+-23BEA-63-Pd+H+-26
3rdGGABEA-9-Pd+H+-12BEA-52-Pd+H+-3BEA-63-Pd+H+-5
hGGABEA-9-Pd+H+-3BEA-52-Pd+H+-1BEA-63-Pd+H+-5
4thGGABEA-9-Pd+H+-24BEA-52-Pd+H+-27BEA-63-Pd+H+-24
hGGABEA-9-Pd+H+-24BEA-52-Pd+H+-27BEA-63-Pd+H+-24
5thGGABEA-9-Pd+H+-27BEA-52-Pd+H+-1BEA-63-Pd+H+-16
hGGABEA-9-Pd+H+-27BEA-52-Pd+H+-4BEA-63-Pd+H+-16
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 99, 884, 455]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 470, 883, 545]]<|/det|> +
Figure S6: CHA-2 optimum proton locations search results. Each data represents the optimum energy based on a different initial position of the protons. The x-axis refers to the index of the structure in the database in the SI and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14). Calculations were done using the \(\omega\) B97X-D functional.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[213, 100, 768, 479]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 490, 883, 565]]<|/det|> +
Figure S7: CHA-2 optimum proton locations search results. Each data represents a different initial position of the proton. The x-axis refers to the O-O distance (where the oxygen is the atom H adsorbs on) in the optimized structure and the y-axis is the energy relative to the most stable structure (CHA-2-H+H+-14)
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 98, 872, 465]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 479, 883, 553]]<|/det|> +
Figure S8: CHA-2 optimum \(\mathrm{Pd^{+}H^{+}}\) locations search results. Each data represents a different initial position of \(\mathrm{Pd^{+}H^{+}}\) . The x-axis refers to the index of the structure in the database and the y-axis is the energy relative to the most stable structure (CHA-2-Pd+H+-3). Calculations were done using the \(\omega \mathrm{B97X - D}\) functional.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[170, 110, 857, 316]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 344, 882, 400]]<|/det|> +
Figure S9: QM images of the optimized geometry of a. CHA-3-Pd+H+-21 and b. CHA-3-Pd+H+-17. Despite only a change in the proton position, there is \(>0.4 \mathrm{eV}\) in energy difference between the two optimized geometries. The same color coding as in Figure 5
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 90, 884, 425]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 444, 883, 500]]<|/det|> +
Figure S10: QM region images of the optimized geometry of a. Pd+H+ on CHA-7 b. Pd+H+ on CHA-3 c. Pd+H+ on CHA-8 d. Pd+H+ on CHA-6 e. Pd+H+ on CHA-9 f. Pd+H+ on CHA-12 g. Pd+2 on CHA-5 and h. Pd+2 on CHA-10. The same color coding as in Figure 5.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 99, 875, 525]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 533, 881, 571]]<|/det|> +
Figure S11: \(\Delta E_{form}\) as a function of the distance between Pd and H in Pd+H+ optimized geometry on CHA (□) and BEA (○)
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 95, 860, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 583, 883, 636]]<|/det|> +
Figure S12: QM region images of optimized geometry of a. Pd+2 on BEA-8 b. Pd+2 on BEA-55 c. Pd+H+ on BEA-33 d. Pd+H+ on BEA-55. The same color coding as in Figure 5.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 108, 855, 494]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 506, 883, 562]]<|/det|> +
Figure S13: Parity plot of the calculated DFT energy on Al pairs (that do not share a MR) and the predicted energy based on the respective isolated Al site (the most stable of the two). All calculations were done on BEA with \(\mathrm{Pd^{+}H^{+}}\) as the cation
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 100, 860, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 584, 880, 622]]<|/det|> +
Figure S14: Image of the QM region on optimized BEA calculations a. BEA-51 (Pd+H+) b. BEA-78 (Pd+H+) c. BEA-52 (Pd+H+). The same color coding as in Figure 5.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 143]]<|/det|> +Table S2: Difference in NO adsorption energy ( \(\Delta \mathrm{E}_{NO}\) ) on \(\mathrm{Pd}^{+}\mathrm{H}^{+}\) in CHA when the two electrons (from \(\mathrm{Pd}^{+}\) and NO) are paired and the two electrons are unpaired. Paired electrons are used as the reference (0 eV) + +<|ref|>table<|/ref|><|det|>[[228, 153, 768, 231]]<|/det|> + +
Structure NamePaired ElectronsTwo Unpaired Electrons
CHA-500.89
CHA-800.82
CHA-701.53
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 98, 850, 375]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 399, 883, 438]]<|/det|> +
Figure S15: Cluster models of a. T696 CHA and b. T810 BEA. The same color coding as in Figure 5
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 45, 143, 70]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[50, 97, 940, 410]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 440, 115, 460]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 484, 420, 504]]<|/det|> +Please see manuscript PDF for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 55, 940, 696]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 727, 118, 748]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[44, 770, 420, 790]]<|/det|> +Please see manuscript PDF for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 56, 930, 480]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 510, 117, 529]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 552, 420, 571]]<|/det|> +Please see manuscript PDF for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[46, 50, 943, 490]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 519, 118, 538]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 561, 420, 580]]<|/det|> +Please see manuscript PDF for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 54, 930, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 477, 118, 496]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 519, 920, 561]]<|/det|> +QM atoms of the optimized calculations. Color coding: red=Oxygen, white=Hydrogen, blue=Palladium and beige=Silicon + +<|ref|>image<|/ref|><|det|>[[50, 575, 944, 912]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 942, 118, 960]]<|/det|> +
Figure 6
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[48, 72, 936, 380]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 404, 116, 422]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[44, 447, 419, 466]]<|/det|> +Please see manuscript PDF for full caption. + +<|ref|>image<|/ref|><|det|>[[60, 490, 930, 925]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 123, 933, 479]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 43, 116, 61]]<|/det|> +
Figure 8
+ +<|ref|>text<|/ref|><|det|>[[42, 490, 116, 509]]<|/det|> +Figure 9 + +<|ref|>text<|/ref|><|det|>[[42, 549, 420, 567]]<|/det|> +Please see manuscript PDF for full caption. + +<|ref|>sub_title<|/ref|><|det|>[[44, 590, 311, 617]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 640, 765, 660]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 679, 137, 696]]<|/det|> +Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/images_list.json b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..f6ab0b3c11d9a7a1a273db9ea97dbcd151b86e3d --- /dev/null +++ b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/images_list.json @@ -0,0 +1,56 @@ +[ + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Table 1. Exploration and optimization of reaction conditions.", + "footnote": [], + "bbox": [], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "b. Late stage modification", + "footnote": [], + "bbox": [], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Dearomatization of fused arenes. a, Scope of naphthalenes. b, Scope of anthracenes and phenanathrenes. c, Large scale synthesis. d, Application in synthesis of valuable targets. aReaction condition: unless specified, see footnote a of Table 1 and the ESI; bRatio of 6-methoxyanphthalene to pyrazole: 1 : 2; cRatio of naphthalene to pyrazole is 2:1; dReaction time: 4 d; eCondition B using10.0 equiv of acid; fCondition B using 5.0 equiv of acid.", + "footnote": [], + "bbox": [], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Mechanistic studies. a, Proposed catalytic cycle. b, Chemoselectivity studies. c, Study of roles of HAT reagent. d, Study of regioselectivity of isoquinoline engaged reaction.", + "footnote": [], + "bbox": [ + [ + 115, + 412, + 883, + 821 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Calculated free energy diagram (in kcal/mol) at the ωB97X-D/def2-TZVPP, SMD (CH₂Cl₂)//ωB97X-D/def2-SVP level of theory.", + "footnote": [], + "bbox": [ + [ + 150, + 85, + 848, + 662 + ] + ], + "page_idx": 6 + } +] \ No newline at end of file diff --git a/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e.mmd b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e.mmd new file mode 100644 index 0000000000000000000000000000000000000000..05786d6daf31eabd1f0acafca2920eac4d4d7e77 --- /dev/null +++ b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e.mmd @@ -0,0 +1,238 @@ + +# Selective skeletal editing of polycyclic arenes using organophotoredox dearmative functionalization + +Wei Wang ( wwang@pharmacy.arizona.edu ) University of Arizona https://orcid.org/0000- 0001- 6043- 0860 + +Peng Ji University of Arizona + +Cassandra Davies Oberlin College + +Feng Gao University of Arizona + +Jing Chen University of Arizona + +Xiang Meng University of Arizona + +Kendall Houk University of California Los Angeles https://orcid.org/0000- 0002- 8387- 5261 + +Shuming Chen Oberlin College https://orcid.org/0000- 0003- 1897- 2249 + +Article + +Keywords: + +Posted Date: January 5th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1139594/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on August 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32201- 7. + +<--- Page Split ---> + +# Selective skeletal editing of polycyclic arenes using organophotoredox dearomatic functionalization + +Peng Ji, \(^{1}\) Cassandra C. Davies, \(^{2}\) Feng Gao, \(^{1}\) Jing Chen, \(^{1}\) Xiang Meng, \(^{1}\) K. N. Houk, \(^{3*}\) Shuming Chen \(^{2*}\) and Wei Wang \(^{1*}\) + +\(^{1}\) Departments of Pharmacology and Toxicology and Chemistry and Biochemistry, University of Arizona, Tucson, AZ 85721- 0207, USA + +\(^{2}\) Department of Chemistry and Biochemistry, Oberlin College, Oberlin, OH 44074, USA + +\(^{3}\) Department of Chemistry and Biochemistry, University of California, Los Angeles, CA 90095- 1569, USA + +## Abstract + +Reactions that lead to destruction of aromatic ring systems often require harsh conditions and, thus, take place with poor selectivities. Selective partial dearomatization of fused arenes is even more challenging but it can be a strategic approach to creating versatile, complex polycyclic frameworks. Herein we describe a general organophotoredox approach for the chemo- and regioselective dearomatization of structurally diverse polycyclic aromatics, including quinolines, isoquinolines, quinoxalines, naphthalenes, anthracenes and phenanthrenes. The success of the new method for chemoselective oxidative rupture of aromatic moieties relies on precise manipulation of the electronic nature of the fused polycyclic arenes. Experimental and computational results show that the key to overcoming the intrinsic thermodynamic and kinetic unfavorability of the dearomatization process is an ultimate hydrogen atom transfer (HAT) step, which enables dearomatization to predominate over the otherwise favorable aromatization pathway. We show that this strategy can be applied to rapid synthesis of biologically valued targets and late- stage skeletal remodeling en route to complex structures. + +<--- Page Split ---> + +Polycyclic scaffolds bearing partially dearomatized fused arenes are commonly encountered in natural products, pharmaceuticals and bioactive molecules1-5. While these molecular frameworks lead to great structural diversity and intriguing biological properties, they engender synthetic challenges as their assembly often requires prefunctionalized substrates and multi- step sequences. Direct dearomatization of fused arenes constitutes an outstandingly efficient approach for the construction of polycyclic scaffolds due to its high atom and step economy1-5. However, selective dearomatization and functionalization of fused arenes is difficult because harsh conditions are often required to disrupt aromaticity, leading to poor selectivities. As a result, dearomative functionalizations of fused arenes has been largely limited to activated arenes such as indoles6-15 and naphthols16-20. + +![](images/Figure_unknown_0.jpg) + + +<--- Page Split ---> + +Figure 1. Methods for selective dearomative functionalization of fused arenes. a, Examples of partially dearomatic natural products and bioactive compounds. b, Selective dearomatization of quinolines. c, Selective dearomatization of naphthalenes. d, Selective dearomatization of structurally diverse fused arenes (this work). + +Issues of chemo- and regioselectivity also add to the difficulty of dearomative functionalizations of unactivated fused arenes such as quinolines and naphthalenes. Selective dearomatization of pyridine moieties in fused arenes is feasible21- 25 owing to the electron deficient nature of pyridine and assistance provided by Lewis acid complexation. Selective dearomatization of phenyl moieties in fused arenes, on the other hand, is significantly more challenging. To our knowledge, only one example has been described in a recent report by Brown, Houk and Glorius involving a photochemical [4+2] cycloaddition between quinolines and alkenes (Figure 1b)26 enabled by energy transfer and Lewis acid activation. A similar [4+2] cycloaddition strategy initiated by energy transfer was also employed in the dearomatization of naphthalene by Sarlah et al. (Figure 1c)27- 32. + +We wondered if a photoredox strategy could be used to selectively dearomatize the phenyl moieties in quinolines, isoquinolines, quinoxalines, naphthalenes and other fused arenes. In reported photoredox processes, radicals add selectively to the pyridine ring in quinolines33,34, and indiscriminately to the phenyl rings in naphthalenes. In contrast to Sarlah's [4+2] cycloaddition strategy, in which regioselectivity is governed primarily by steric effects27,32, we envisaged taking advantage of the relative electron richness of the phenyl ring to achieve selective dearomative functionalization (Figure 1d). Photocatalytically promoted single electron transfer (SET) oxidation of quinolines would produce radical cation I, in which the positive charge is mostly localized on the phenyl ring. We anticipated nucleophiles to add to intermediate I selectively to form radical II. Direct arene \(\mathrm{C}(sp^2)\) - H functionalization with nucleophiles under photoredox catalysis or electrophotochemical conditions have been elegantly leveraged by Nicewicz35- 38, Lambert39, Hu40, and Wickens41 to generate aromatic functionalized products. We postulated that it would be possible to direct the reaction toward the less thermodynamically stable dearomatized products by capturing radical II with a sufficiently activated hydrogen atom transfer (HAT) agent. Below, we describe the successful development of a general chemo- and regioselective method for the dearomative functionalization of diverse fused arenes using this strategy. + +<--- Page Split ---> + +## Dearomatization of quinolines, isoquinolines, and quinoxalines + +In order to undergo selective photoredox promoted dearomative functionalization, fused arenes must have oxidation potentials that enable them to be oxidized by excited states of photocatalysts (PCs) through SET. The oxidation potential of unsubstituted quinoline ( \(E^{\mathrm{ox}} = 2.23 \mathrm{~V}\) vs SCE, Supplementary Figure S1 and S2) suggests that its oxidation would be very difficult using common PCs. However, introduction of an OMe group onto the phenyl moiety of quinoline lowers the oxidation potential sufficiently (e.g., 6- methoxyquinoline 1a, \(E^{\mathrm{ox}} = 1.83 \mathrm{~V}\) vs SCE, Table 1 and Supplementary Figure S2), making it possible for it to participate in thermodynamically driven SETs with excited states of conventional organophotoredox catalysts such as acridinium and triphenylpyrrylium (TPT) salts that have excited state reduction potentials ( \(E^{\mathrm{red}}\) ) greater than 2.0 V (Table 1). + +![](images/Figure_unknown_1.jpg) + +
Table 1. Exploration and optimization of reaction conditions.
+ +
EntryVariation from the “Standard Conditions”aYieldb (%)
1None81, 76c (&lt;5 of 4a)b
2Mes-Acr1 instead of Mes-Acr275
3Mes-Acr3 instead of Mes-Acr267
4DCA instead of Mes-Acr234
5TPT instead of Mes-Acr218
64CzIPN instead of Mes-Acr2trace
7DCM (0.5 M) with 2,6-lutidine (0.2 equiv)20
8DCM (0.2 M) with 2,6-lutidine (0.2 equiv)49
+ +<--- Page Split ---> + + +
9DCM (0.1 M) with 2,6-lutidine (0.2 equiv)63
10DCM (0.05 M) with 2,6-lutidine (0.2 equiv)76
11DCM (0.025 M) with 2,6-lutidine (0.2 equiv)74
12without Ph3SiSH53 (23 of 4a)b
13without Mes-Acr2trace
14without lighttrace
15390 nm light40
16O2 instead of N2trace (56 of 4a)b
+ +aStandard conditions: unless otherwise specified, a mixture of 6-methoxyquinoline 1a (0.3 mmol, 0.05 M),pyrazole 1b (0.2 mmol), Ph3SiSH (0.04 mmol), and Mes-Acr2 (0.005 mmol) in DCM (under a N2 atmosphere at room temperature was irradiated with 40 W Kessil blue LEDs for 48 h. bYield determined by using 1H NMR spectroscopy using 1,3,5-trimethoxynbenzene as an internal standard. cIsolated yield. + +We assessed the viability of the proposed dearomatic functionalization protocol by reacting 6-methoxyquinoline 1a with pyrazole 2a as a nucleophile under irradiation with blue LEDs in DCM solutions containing HAT agents and commercially available organic photosensitizers (Table 1). The desired dearomatization product 3a is generated in the highest yield using 1.5 equiv of 1a, 1.0 equiv of 2a (0.05 M), 2.5 mol% N-methylmeso-acridinium tetrafluoroborate (Mes-Acr1), 0.2 equiv of the HAT agent and 0.2 equiv of 2,6-lutidine. Nucleophile addition occurred exclusively at the 5-position. The efficiency of the reaction is dependent on the HAT agent (Supplementary Table S1). Among the 17 different thiols and benzeneselenols screened, Ph3SiSH was found to be the superior HAT agent, leading to regioselective formation of 3a in 81% yield (Table 1, entry 1). While photocatalysts such as TPT (E*red=2.55 V vs SCE, 18%, entry 4) with excited state reduction potentials higher than the oxidation potential of 1a were found to promote this reaction, Mes-Acr2 (2.5 mol%,=2.2 V vs SCE) was identified as the best photocatalyst (3a formed 81% 1H NMR yield, entry 1). Solvent has a significant effect on the reaction (Supplementary Table S2). Furthermore, substrate ratio (Supplementary Table S3) and concentration has a marked impact on the yield (entries 7-11) with low concentrations of 1a being more beneficial (entries 7 vs 10). The absence of 2,6-lutidine does not affect the reaction outcome (entries 1 vs 10) as quinoline 1a can serve as the base. Lastly, control experiments confirmed that the HAT agent, light and photocatalyst are all required for selective dearomatization. In particular, the HAT agent is critical for minimizing the formation of the aromatization product 4a (entries 1 vs 12). + +<--- Page Split ---> +![](images/Figure_3.jpg) + + +![](images/Figure_4.jpg) + +
b. Late stage modification
+ +![](images/Figure_5.jpg) + + +<--- Page Split ---> + +Figure 2. Scope of photoredox dearomatic functionalization of quinolines, isoquinolines, and quinoxalines with azoles. a, Scope of quinolines and quinoxalines with azoles. b, Late-stage functionalization. aReaction conditions: unless otherwise specified, see footnote a of Table 1 and supplementary materials; bRatio of isomers was determined by using \(^1\mathrm{H}\) NMR analysis of the crude reaction mixture; s.m.: starting materials; ddr value was determined by \(^1\mathrm{H}\) NMR analysis of the crude reaction mixture; Condition B without 2,6- lutidine; Condition A using 3,6- di- tert- butyl- 9- mesityl- 10- phenylacridin- 10- ium tetrafluoroborate as photocatalyst for 5 d. + +Having established optimal conditions for the new photoredox promoted dearomatic functionalization reaction, we next explored the scope with fused arenes as substrates and azoles as nucleophiles (Figure 2a). Our results show that the reaction proceeds with high chemo- and regioselectivity for a wide range of quinolines and azoles. Both electron- donating and electron- withdrawing groups are tolerated on the pyrazole. Halogenated pyrazoles reacted with 6- methoxyquinoline with Ph3SiSH as the HAT agent (condition A giving 3b- 3d). For 5- fluoropyrazole, benzeneselenol, which is 10- fold more reactive than thiols in HAT42,43, is required along with 2,6- lutidine (condition B) to yield 3e. Other electron- deficient pyrazoles also reacted in the presence of benzeneselenol to produce 3h and 3i (condition C, without 2,6- lutidine). Notably, the dearomatization method can be applied to a pyrazole boronate, which generates 3j (51%) containing a useful handle for subsequent coupling reactions. A disubstituted pyrazole also reacted smoothly to form 3k (51%), as did other common azoles including benzotriazole (3l), 1,2,3- triazoles (3m), and tetrazole (3n) under condition B or C. + +6- Alkoxy- substituted quinolines are particularly effective substrates for the dearomatic functionalization reaction, producing diverse substituted 5,8- dihydroquinolines as products (Figure 2a). A wide range of alkoxy substituents, including those containing methyl (3o, 3q, 3r, and 3v), ester (3p, 3ai), bromo (3t), amide (3s), phthalamide (3w), alkene (3x), benzyl (3y, 3z), TBS (3aa, 3ab), MOM (3ac), phenyl (3ad), ketone (3ae), alkyne (3aj), nitrile (3y, 3ak), thiophene (3af) and benzofuran (3ag) moieties, are well tolerated. Moreover, the process promotes selective dearomatization and functionalization of quinoline ring systems even when other arene moieties including benzene (3w, 3y, 3z, 3ad, 3ae), thiophene (3af) and benzofuran (3ah) are present. Notably, the protocol enables highly selective dearomatization of a quinoline ( \(E^{ox} = 1.86 \mathrm{~V}\) vs SCE) over a naphthalene ( \(E^{ox} = 1.92 \mathrm{~V}\) vs SCE) ring (3ag) despite their similar oxidation potentials. In addition to controlling chemoselectivity, 6- alkyloxy groups also direct site selective formation of + +<--- Page Split ---> + +5- nucleophile- substituted- 5,8- dihydroquinolines. Furthermore, dearomatization of 7- methoxyquinolines is also highly regioselective generating only 8- substituted 5,8- dihydroquinoline products (3u, 3v). Isoquinolines can participate in the process (3al, 3am). However, regioisomeric 7,8- dihydroisoquinolines are formed. Finally, quinoxalines are also viable substrate for the dearomatization process (3an). + +Owing to the mild conditions required, the organophotoredox- promoted dearomatic functionalization process is applicable to late- stage functionalizations of complex pharmaceutically relevant quinoline structures including viqualine (3ao), Oppolzer's camphorsultam (3ap), galactose (3aq, 3ar), menthol (3as), probalan (3at), tyrosine (3au), cholesterol (3av), and estrone (3aw) groups (Figure 2b). It is particularly impressive that a grazoprevir analog (3ax) containing various functional groups was selectively dearomatized utilizing this process. These results underscore the mildness, high chemo- and regioselectivity, and practicality of the new protocol. + +## Dearomatization of fused arenes + +Encouraged by the success of the new photoredox reaction of quinolines, we explored its utility for the selective dearomatization of fused non- heteroaromatics. The results show that selective dearomatization of fused arenes is more challenging. As with quinolines, electron- donating substituents can be used to lower the oxidation potentials and enhance reactivity. A 2- methoxy substituent in naphthalene, for instance, lowers the oxidation potential to 1.8 V (vs SCE, supplementary Figure S2) and enables selective oxidation by the excited state of photocatalyst Mes- Acr2. Selective dearomatization and C- 1 functionalization of 2- methoxynaphthalene by pyrazole using condition B led to 3ay (Figure 3a). The efficiency of this process is also dependent on the base and the HAT agent, as are shown in 3az, 3ba, and 3bi (condition C), 3bb, 3bf, 3bg, 3bh, and 3bj (condition B) and 3be (condition D). The dearomatization strategy can be applied to methyl substituted naphthalene (3bd). Finally, unsubstituted naphthalene (3bd) also participates in the process to produce the 1- substituted product regioselectively. + +<--- Page Split ---> +![PLACEHOLDER_9_0] + +
Figure 3. Dearomatization of fused arenes. a, Scope of naphthalenes. b, Scope of anthracenes and phenanathrenes. c, Large scale synthesis. d, Application in synthesis of valuable targets. aReaction condition: unless specified, see footnote a of Table 1 and the ESI; bRatio of 6-methoxyanphthalene to pyrazole: 1 : 2; cRatio of naphthalene to pyrazole is 2:1; dReaction time: 4 d; eCondition B using10.0 equiv of acid; fCondition B using 5.0 equiv of acid.
+ +The photoredox reaction can also be applied to the dearomatization functionalization of other common polycyclic aromatic hydrocarbons including anthracene (3bk, \(87\%\) ) and phenanthrene (3bl, \(73\%\) ) (Figure 3b). Notably, 6- methoxyphenanthrene reacts to form the trans substitution + +<--- Page Split ---> + +product 3bm \((84\%)\) . Moreover, nucleophiles other than azoles, such as trifluoromethanesulfonamide (3bn) and imidazole (3bo), can also participate in the process. Furthermore, carboxylic acids, which are often used as radical precursors in photoredox reactions44, serve as effective nucleophiles for dearmative functionalization of phenanthrenes. These reactions proceed in moderate to high yields (3bp- 3cb, \(25 - 89\%\) ) and display broad functional group tolerance including phenyl (3br, 3bt), azide (3bs), alkenyl (3bu), alkynyl (3bv), cyclobutyl (3bw), and hydroxyl acid groups (3cb). Due to their weaker nucleophilicity and steric effects, benzoic acid (3bx) and secondary carboxylic acids (3bw) are less effective nucleophiles for this process. In addition, the reaction can be utilized for the direct modification of biologically relevant structures such as aspirin (3bx) and amino acids (3by- 3ca). + +Studies with the silyloxquinoline 1b show that the synthetic protocol can be scaled up without loss of yield (Figure 3c). Moreover, the reaction can be adapted for use in a flow system, an emerging technology in organic synthesis45. Using the flow system, the reaction time is reduced dramatically to 3 h without compromising the yield. The product silylenol ether 3aa can be converted to the corresponding alcohol 3cc. The preparative power of the new method was also demonstrated by its use in the cost- effective synthesis of important biologically active targets, including a JAK inhibitor (Figure 3d)46. Our new synthetic route starts from naphthalene (\$0.04/g) instead of the more expensive 1- bromo- 1,2,3,4- tetrahydronaphthalene (\$370/g) used in the earlier preparative pathway46. Dearomatization of naphthalene with pyrazol- 4- ylboronic acid pinacol ester produces key intermediate 3cd, which is then reduced using Rh/Al\(_2\)O\(_3\) catalyzed hydrogenation to form 3ce. The JAK inhibitor is then readily prepared from 3ce using reported procedures46. These studies clearly demonstrate the synthetic value of the our dearmative functionalization protocol, which provides direct and efficient access to polycyclic partially dearomatized frameworks that were previously inaccessible or required strenuous synthetic efforts. + +Mechanistic studies. To gain insight into the reaction mechanism, chemo- and regioselectivity and the role played by the HAT agent, we conducted computational studies and additional experiments. The excited state \(\mathrm{PC}^{+*}\) Mes- Acr2\* possesses strong oxidizing power ( \(E^{\mathrm{red}} = 2.2 \mathrm{~V}\) ) and can oxidize the 6- methoxyquinoline (1a) ( \(E^{\mathrm{ox}} = 1.83 \mathrm{~V}\) vs SCE, supplementary Figure S1) to give radical cation 5a (Figure 4a and Stern- Volmer quenching experiments, supplementary Figure S3). The oxidation potential of the fused arene substrate is crucial for the success of this process, + +<--- Page Split ---> + +as demonstrated by the high reactivity of 1a in contrast to the lack of reactivity of unsubstituted quinoline \((E^{\mathrm{ox}} = 2.23 \mathrm{V}\) vs SCE, Supplementary Figure S2). In addition, when a mixture of 1a and 6-acetoxyquinoline (1c) \((E^{\mathrm{ox}} = 2.2 \mathrm{V}\) vs SCE, supplementary Figure S2) is subjected to the reaction conditions, only 1a undergoes selective dearomatization to form 3a, indicating that electron-withdrawing substituents inhibit the reaction. In contrast, 1a and 2- methoxynaphthalene (1e, \(E^{\mathrm{ox}} = 1.8 \mathrm{V}\) vs SCE, Supplementary Figure S2), which have similar oxidation potentials, both react under the same conditions to produce the corresponding dearomatized products 3a and 3ay (Figure 4b). Likewise, phenanthrene 1d \((E^{\mathrm{ox}} = 1.91 \mathrm{V}\) vs SCE, Supplementary Figure S2) is an effective substrate for this process (Figure 4b). Notably, the pyrazole nucleophile \((E^{\mathrm{ox}} = 2.15 \mathrm{V}\) vs SCE, Supplementary Figure S2) is not oxidized by Mes- Acr2\\*. This is supported by luminescence quenching study of Mes- Acr2 with varying concentrations of 6- methoxyquinoline 1a and pyrazole 2a (Supplementary Figure S4). + +![PLACEHOLDER_11_0] + +
Figure 4. Mechanistic studies. a, Proposed catalytic cycle. b, Chemoselectivity studies. c, Study of roles of HAT reagent. d, Study of regioselectivity of isoquinoline engaged reaction.
+ +<--- Page Split ---> + +Following the oxidation of 1a through SET, pyrazole reacts with the electrophilic radical cation 5a to form the radical cation 6a (Figure 4a). Calculated Hirshfeld charge and spin densities (Figure 4a; see Supplementary Figure S5 for calculated charge and spin densities for other substrates) indicate that C5 has the highest spin density (bold face) in 5a and is therefore the preferred site for nucleophilic addition. The calculated free energies also support the observation that reaction at C5 position is kinetically favored over C8 or C7 by at least 3 kcal/mol (Supplementary Figure S6). Subsequent exothermic deprotonation of 6a by 2,6- lutidine gives the neutral radical 7a. Calculated free energies suggest that sequential SET oxidation and deprotonation of 7a to form the aromatic substitution product 4a is more thermodynamically favorable (- 24.9 kcal/mol, Figure 5). This is reflected in the outcomes of processes leading to fully aromatic products developed by Nicewicz \(^{35 - 38}\) , Lambert \(^{39}\) , \(\mathrm{Hu}^{40}\) and Wickens \(^{41}\) . However, the presence of a reactive HAT agent (e.g., Ph \(_3\) SiSH) redirects the course of the reaction toward the dearomatized product 3a rather than the more stable aromatic product 4a (Figure 5). The transformation of 7a into 3a is facilitated by the exergonic nature of the HAT step (- 17.4 kcal/mol change in free energy). Thus, success of the dearomatization process depends on the use of thiol and selenol HAT agents that have weak S- H and Se- H bonds (S- H: 82 kcal/mol, Se- H: 73 kcal/mol) and correspondingly high H- atom transfer rates (PhSH: \(\mathrm{K}_{20} = 9.0 \times 10^{7} \mathrm{M}^{- 1} \mathrm{S}^{- 1}\) and PhSeH: \(\mathrm{K}_{20} = 1.3 \times 10^{9} \mathrm{M}^{- 1} \mathrm{S}^{- 1}\) ) \(^{42,43,47}\) . Furthermore, the process is highly regioselective. We did not observe the formation of regioisomeric 3a'. The observation is supported by the calculation that its formation is unfavorable by 3.8 kcal/mol due to the steric effect (Figure 5). Control experiments showed that in the absence of the HAT agent Ph \(_3\) SiSH, the yield of the dearomatized product dropped to 46% (Figure 4c). After the HAT step, the Ph \(_3\) Si' radical is reduced by the PC' Mes- Acr2 radical to generate the Ph \(_3\) Si' anion, which is protonated to give Ph \(_3\) SiH and the PC Mes- Acr2 (Figure 4a). + +The effect of the HAT agent is particularly noticeable in reactions of isoquinolines and naphthalenes with pyrazole. Only when PhSeH is present do these processes produce the desired products 3al and 3ay (Figure 4c). Furthermore, the polarity match \(^{48}\) between the protic S- H/Se- H and the nucleophilic carbon radical 7a might further facilitate the HAT process. Unexpectedly, reactions between isoquinolines and pyrazole give regioisomeric 7,8- dihydroisoquinolines. Our experiments suggest that the reaction initially forms 5,8- dihydroisoquinoline 3al', which is then converted to the more stable 7,8- dihydroisoquinoline 3al (Figure 4d). The 5,8- dihydroisoquinoline 3al' was obtained in the presence of \(\mathrm{H}_2\mathrm{O}\) . Two possible pathways could lead to 5,8- + +<--- Page Split ---> + +dihydroisoquinoline 3al. However, the deprotonation/C=C isomerization pathway was ruled out as we did not observe the formation of 3al when 3al' was treated with 2,6- lutidine, a base used in the dearomatization process. In contrast, 3al was formed under the photoredox reaction conditions through SET oxidation of the C=C bond, deprotonation and HAT processes. This pathway was also supported by calculated reaction energies (Supplementary Figure S7). + +An alternative pathway for the formation of the functionalized dearomatized product 3a is through reduction of radical 7a by SET from PC' to form anion 8a, which then undergoes protonation. However, calculations suggest that the formation of 8a is highly unfavorable (Figure 5). Moreover, preliminary experiments showed that dearomatized product 3a ( \(E^{ox} = 1.60 \mathrm{~V}\) vs SCE, Supplementary Figure S2) could be converted back to starting material 1a via PC'\\* (Mes- Acr2\\*) mediated oxidation under the standard reaction conditions (Figure 4c). Unexpectedly, we found that the HAT agent Ph3SiSH retards the efficiency of this process (Figure 4c). Specifically, photoinduced reaction of 3a in the presence of Ph3SiSH leads to \(61\%\) recovery of 3aa and formation of \(28\%\) of 1b. In contrast, in the absence of the HAT agent, 1b was produced in \(49\%\) yield. Thus, these observations indicate that the HAT agent not only facilitates the formation of dearomatized products, it also inhibits the conversion of these products back to aromatic starting materials. + +<--- Page Split ---> +![PLACEHOLDER_14_0] + +
Figure 5. Calculated free energy diagram (in kcal/mol) at the ωB97X-D/def2-TZVPP, SMD (CH₂Cl₂)//ωB97X-D/def2-SVP level of theory.
+ +In summary, we developed a conceptually unique organophotoredox catalytic dearomative functionalization strategy for the selective disruption of aromaticity in fused arenes. A reactive HAT agent is utilized to overcome the competing aromatization pathway, which is typically more favorable, and achieve the selective formation of dearomatizatized products. Experimental and computational mechanistic studies support the key role played by HAT agents in the unprecedented process. The preparative power of the protocol was demonstrated by applications to structurally + +<--- Page Split ---> + +diverse fused arenes including quinolines, isoquinolines, quinoxalines, naphthalenes, anthracenes and phenanthrenes. The method can also be applied to the synthesis and late- stage skeletal editing of complex pharmaceutically valued structures. We anticipate that the new process will enable facile access to a wide range of synthetically versatile frameworks and accelerate the construction of new molecular architectures for drug discovery. + +## References + +1. Pape, A. R., Kaliappan, K. P. & Kündig, E. P. Transition-metal-mediated dearomatization reactions. Chem. Rev. 100, 2917-2940 (2000). +2. López Ortiz, F., Iglesias, M. J., Fernández, I., Andújar Sánchez, C. M. & Ruiz Gómez, G. Nucleophilic dearomatizing (DNAr) reactions of aromatic C-H systems. A mature paradigm in organic synthesis. Chem. Rev. 107, 1580-1691 (2007). +3. Roche, S. P. & Porco, J. A. Dearomatization strategies in the synthesis of complex natural products. Angew. Chem. Int. Ed. 50, 4068-4093 (2011). +4. Zheng, C. & You, S.-L. Advances in catalytic asymmetric dearomatization. ACS Cent. Sci. 7, 432-444 (2021). +5. Huck, C. J. & Sarlah, D. Shaping molecular landscapes: recent advances, opportunities, and challenges in dearomatization. Chem 6, 1589-1603 (2020). +6. Roche, S. P., Tendoung, J.-J. Y. & Tréguier, B. Advances in dearomatization strategies of indoles. Tetrahedron 71, 3549-3591 (2015). +7. Zheng, C. & You, S. L. Catalytic asymmetric dearomatization (CADA) reaction-enabled total synthesis of indole-based natural products. Nat. Prod. Rep. 36, 1589-1605 (2019). +8. Gentry, E. C., Rono, L. J., Hale, M. E., Matsuura, R. & Knowles, R. R. Enantioselective synthesis of pyrroloindolines via noncovalent stabilization of indole radical cations and applications to the synthesis of alkaloid natural products. J. Am. Chem. Soc. 140, 3394-3402 (2018). +9. An, J., Zou, Y.-Q., Yang, Q.-Q., Wang, Q. & Xiao, W.-J. Visible light-induced aerobic oxymidation of indoles: a photocatalytic strategy for the preparation of tetrahydro-5h-indolo[2,3-b]quinolinols. Adv. Synth. Catal. 355, 1483-1489 (2013). + +<--- Page Split ---> + +10. Zhu, M., Zheng, C., Zhang, X. & You, S.-L. Synthesis of cyclobutane-fused angular tetracyclic spiroindolines via visible-light-promoted intramolecular dearomatization of indole derivatives. J. Am. Chem. Soc. 141, 2636–2644 (2019). +11. Ma, J., Schäfers, F., Daniliuc, C., Bergander, K., Strassert, C. A. & Frank Glorius Gadolinium photocatalysis: Dearomative [2+2] cycloaddition/ring-expansion sequence with indoles. Angew. Chem. Int. Ed. 59, 9639-9645 (2020). +12. Liu, K. et al Electrooxidation enables highly regioselectivedearomative annulation of indole and benzofuran derivatives. Nat. Commun. 11, https://doi.org/10.1038/s41467-019-13829-4 (2020). +13. Wu, J., Dou, Y., Guillot, R., Cyrille Kouklovsky, C., & Vincent, G. Electrochemical dearomative 2,3-difunctionalization of indoles. J. Am. Chem. Soc. 141, 2832–2837 (2019). +14. Studies from our group and ref 15: Zhang, Y., Ji, P., Gao, F., Dong, Y., Huang, H., Wang, C., Zhou, Z., & Wang, W. Organophotocatalytic dearomatization of indoles, pyrroles and benzo(thio)furans via a Giese-type transformation. Commun. Chem. 4, 20 (2021). +15. Zhang, Y., Ji, P., Gao, F., Huang, H., Zeng, F., & Wang, W. Photorecox asymmetric nucleophilic dearomatization of indoles with neutral radicals. ACS Catal. 11, 998-1007 (2021) +16. Pouységu, L., Deffieux, D., & Quideau, S. Hypervalent iodine-mediated phenol dearomatization in natural product synthesis. Tetrahedron, 66, 2235-2261 (2010). +17. Dhineshkumar, J., Samaddar, P., & Prabhu, K. R. A copper catalyzed azidation and peroxidation of β-naphthols via an oxidative dearomatization strategy. Chem. Commun. 52, 11084-11087 (2016). +18. Kulish, K., Boldrini, C., Reis, M. C., Pérez, J. M., & Harutyunyan, S. R. Lewis acid promoted dearomatization of naphthols. Chem. Eur. J. 26, 15843 (2020). +19. Sun, W., Li, G., Hong, L., & Wang, R. Asymmetric dearomatization of phenols. Org. Biomol. Chem. 2164-2176 (2016) +20. Zhu, G., Bao, G., Li, Y., Yang, J., Sun, W., Li, J., Hong, L., & Wang, R. Chiral phosphoric acid catalyzed asymmetric oxidative dearomatization of naphthols with quinones. Org. Lett. 18, 5288-5291 (2016). +21. Takamura, M., Funabashi, K., Kanai, M., & Shibasaki, M. Asymmetric Reissert-type reaction promoted by bifunctional catalyst. J. Am. Chem. Soc. 122, 6327-6328 (2000). + +<--- Page Split ---> + +22. Zurro, M., Asmus, S., Beckendorf, S., Mück-Lichtenfeld, C., & Mancheño, O. G. Chiral helical oligotriazoles: new class of anion-binding catalysts for the asymmetric dearomatization of electron-deficient \(N\)-heteroarenes. J. Am. Chem. Soc. 136, 13999-14002 (2014). +23. Pappoppula, M., Cardoso, F. S. P., Garrett, B. O., & Aponick, A. Enantioselective copper-catalyzed quinoline alkynylation. Angew. Chem. Int. Ed. 54, 15202-15206 (2015). +24. Yan, X., Ge, L., Castiñeira Reis, M., & Harutyunyan, S. R. Nucleophilic dearomatization of \(N\)-heteroaromatics enabled by Lewis acids and copper catalysis. J. Am. Chem. Soc. 142, 20247-20256 (2020). +25. Leitch, J. A., Rogova, T., Duarte, F., & Dixon, D. J. Dearomatic photocatalytic construction of bridged 1,3-diazepanes. Angew. Chem. Int. Ed. 59, 4121-4130 (2020). +26. Ma, J., et al Photochemical intermolecular dearomatic cycloaddition of bicyclic azaarenes with alkenes. Science 371, 1338-1345 (2021). +27. Southgate, E. H., Pospech, J., Fu, J., Holycross, D. R., & Sarlah, D. Dearomatic dihydroxylation with arenophiles. Nat. Chem. 8, 922-928 (2016). +28. Okumura, M., Shved, A. S., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-carboamination. J. Am. Chem. Soc. 139, 17787-17790 (2017). +29. Hernandez, L. W., Klockner, U., Pospech, J., Hauss, L., & Sarlah, D. Nickel-catalyzed dearomatic trans-1,2-carboamination. J. Am. Chem. Soc. 140, 4503-4507 (2018). +30. Tang, C., Okumura, M., Deng, H., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-oxymination. Angew. Chem. Int. Ed. 58, 15762-15766 (2019). +31. Wertjes, W. C., Okumura, M., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-diamination. J. Am. Chem. Soc. 141, 163-167; (2019). +32. Siddiqi, Z., Wertjes, W. C. & Sarlah, D. Chemical equivalent of arene monooxygenases: dearomatic synthesis of arene oxides and oxepines. J. Am. Chem. Soc. 142, 10125-10131 (2020). +33. Duncton, M. A. J., Minisci reactions: Versatile CH-functionalizations for medicinal chemists. Med. Chem. Commun. 2, 1135-1161 (2011). +34. Proctor, R. S. J. & Phipps, R. J. Recent advances in Minisci-type reactions. Angew. Chem. Int. Ed. 58, 13666-13699 (2019). + +<--- Page Split ---> + +35. Romero, N. A., Margrey, K. A., Tay, N. E., & Nicewicz, D. A. Site-selective arene C-H amination via photoredox catalysis. Science 349, 1326-1330 (2015). +36. Margrey, K. A., McManus, J. B., Bonazzi, S., Zecri, F., & Nicewicz, D. A. Predictive model for site-selective aryl and heteroaryl C-H functionalization via organic photoredox catalysis. J. Am. Chem. Soc. 139, 11288-11299 (2017). +37. McManus, J. B. & Nicewicz, D. A. Direct C-H cyanation of arenes via organic photoredox catalysis. J. Am. Chem. Soc. 139, 2880-2883 (2017). +38. Margrey, K. A., Levens, A., & Nicewicz, D. A. Direct aryl C-H amination with primary amines using organic ohotoredox catalysis. Angew. Chem. Int. Ed. 56, 15644-15648 (2017). +39. Huang, H., Strater, Z. M., Rauch, M. Shee, J., Sisto, T. J., Nuckolls, C., & Lambert, T. H. Electrophotocatalysis with a trisaminocyclopropenium radical dication. Angew. Chem. Int. Ed. 58, 13318-13322 (2019). +40. Zhang, L., Liardet, L., Luo, J., Ren, D., Grätzel, M., & Hu, X. Photoelectrocatalytic arene C-H amination. Nat. Catal. 2, 366-373 (2019). +41. Targos, K., Williams, O. P., & Wickens, Z. K. Unveiling Potent Photooxidation Behavior of Catalytic Photoreductants. J. Am. Chem. Soc. 143, 4125-4132 (2021). +42. Newcomb, M., Choi, S.-Y., & Horner, J. H. Adjusting the Top End of the Alkyl Radical Kinetic Scale. Laser Flash Photolysis Calibrations of Fast Radical Clocks and Rate Constants for Reactions of Benzeneselenol. J. Org. Chem. 64, 1225-1231; (1999). +43. Newcomb, M., Varick, T. R., Ha, C., Manek, M. B., & Yue, X. Picosecond radical kinetics. Rate constants for reaction of benzeneselenol with primary alkyl radicals and calibration of the 6-cyano-5-hexenyl radical cyclization. J. Am. Chem. Soc. 114, 8158-8163 (1992). +44. Xuan, J., Zhang, Z.-G., & Xiao, W.-J. Visible-light-induced decarboxylative functionalization of carboxylic acids and their derivatives. Angew. Chem. Int. Ed. 54, 15632-15641 (2015). +45. Plutschack, M. B., Pieber, B., Gilmore, K., & Seeberger, P. H. The Hitchhiker's guide to flow chemistry. Chem. Rev. 17, 11796-11893 (2017). +46. Rodgers, J. D., et al. Heteroaryl substituted pyrrolo[2,3-b]pyridines and pyrrolo[2,3-b]pyrimidines as Janus kinase inhibitors. US2009181959A1. +47. Mayer, J. M. Understanding hydrogen atom transfer: from bond strengths to Marcus theory. Acc. Chem. Res. 44, 36-46 (2011). + +<--- Page Split ---> + +48. Roberts, B. P. Polarity-reversal catalysis of hydrogen-atom abstraction reactions: concepts and applications in organic chemistry. Chem. Soc. Rev. 28, 25-35 (1999). + +Acknowledgements W.W. acknowledges the NIH (5R01GM125920) and University of Arizona for financial support and the NSF MRI for acquisition of 500 MHz NMR spectrometer (1920234). S.C. is grateful to Oberlin College for financial support. K.N.H thanks the National Science Foundation (Grant CHE-1764328). DFT calculations were performed using the SCIURus, the Oberlin College HPC cluster (NSF MRI 1427949). + +Author Information P. J., F. G., J. C., and X. M. conducted and analyzed the synthetic experiments, C. C. D. performed DFT calculations under the direction of S. C., K. N. H. and W. W planned, designed and directed the project and P. J., S. C., K. N. H and W. W. wrote the manuscript. + +Competing interests The authors declare no competing interests. + +Additional Information Supplementary information and chemical compound information accompany this paper at www.nature.com/naturec. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermission/. Correspondence and requests for materials should be addressed to K. N. H. (houk@chem.ucla.edu), S. C. (shuming.chen@oberlin.edu) and W. W. (wwang@pharmacy.arizona.edu). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e_det.mmd b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..56f308d3f8fe893a32eb4432e7f0375ac473f686 --- /dev/null +++ b/preprint/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e/preprint__131ccad170451a3cf6bab1fd53b1cd08074dccc9d714c075e0204302fcbbbd7e_det.mmd @@ -0,0 +1,281 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 944, 177]]<|/det|> +# Selective skeletal editing of polycyclic arenes using organophotoredox dearmative functionalization + +<|ref|>text<|/ref|><|det|>[[44, 195, 595, 238]]<|/det|> +Wei Wang ( wwang@pharmacy.arizona.edu ) University of Arizona https://orcid.org/0000- 0001- 6043- 0860 + +<|ref|>text<|/ref|><|det|>[[44, 243, 238, 285]]<|/det|> +Peng Ji University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 291, 204, 330]]<|/det|> +Cassandra Davies Oberlin College + +<|ref|>text<|/ref|><|det|>[[44, 336, 238, 377]]<|/det|> +Feng Gao University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 383, 238, 424]]<|/det|> +Jing Chen University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 430, 238, 470]]<|/det|> +Xiang Meng University of Arizona + +<|ref|>text<|/ref|><|det|>[[44, 475, 722, 517]]<|/det|> +Kendall Houk University of California Los Angeles https://orcid.org/0000- 0002- 8387- 5261 + +<|ref|>text<|/ref|><|det|>[[44, 521, 546, 562]]<|/det|> +Shuming Chen Oberlin College https://orcid.org/0000- 0003- 1897- 2249 + +<|ref|>text<|/ref|><|det|>[[44, 603, 102, 620]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 641, 137, 659]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 678, 320, 698]]<|/det|> +Posted Date: January 5th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 716, 475, 736]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1139594/v1 + +<|ref|>text<|/ref|><|det|>[[44, 753, 911, 797]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 832, 920, 876]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on August 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32201- 7. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[157, 89, 839, 141]]<|/det|> +# Selective skeletal editing of polycyclic arenes using organophotoredox dearomatic functionalization + +<|ref|>text<|/ref|><|det|>[[123, 157, 875, 204]]<|/det|> +Peng Ji, \(^{1}\) Cassandra C. Davies, \(^{2}\) Feng Gao, \(^{1}\) Jing Chen, \(^{1}\) Xiang Meng, \(^{1}\) K. N. Houk, \(^{3*}\) Shuming Chen \(^{2*}\) and Wei Wang \(^{1*}\) + +<|ref|>text<|/ref|><|det|>[[121, 221, 879, 269]]<|/det|> +\(^{1}\) Departments of Pharmacology and Toxicology and Chemistry and Biochemistry, University of Arizona, Tucson, AZ 85721- 0207, USA + +<|ref|>text<|/ref|><|det|>[[144, 283, 849, 305]]<|/det|> +\(^{2}\) Department of Chemistry and Biochemistry, Oberlin College, Oberlin, OH 44074, USA + +<|ref|>text<|/ref|><|det|>[[120, 321, 879, 368]]<|/det|> +\(^{3}\) Department of Chemistry and Biochemistry, University of California, Los Angeles, CA 90095- 1569, USA + +<|ref|>sub_title<|/ref|><|det|>[[115, 420, 191, 437]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[112, 454, 886, 790]]<|/det|> +Reactions that lead to destruction of aromatic ring systems often require harsh conditions and, thus, take place with poor selectivities. Selective partial dearomatization of fused arenes is even more challenging but it can be a strategic approach to creating versatile, complex polycyclic frameworks. Herein we describe a general organophotoredox approach for the chemo- and regioselective dearomatization of structurally diverse polycyclic aromatics, including quinolines, isoquinolines, quinoxalines, naphthalenes, anthracenes and phenanthrenes. The success of the new method for chemoselective oxidative rupture of aromatic moieties relies on precise manipulation of the electronic nature of the fused polycyclic arenes. Experimental and computational results show that the key to overcoming the intrinsic thermodynamic and kinetic unfavorability of the dearomatization process is an ultimate hydrogen atom transfer (HAT) step, which enables dearomatization to predominate over the otherwise favorable aromatization pathway. We show that this strategy can be applied to rapid synthesis of biologically valued targets and late- stage skeletal remodeling en route to complex structures. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 344]]<|/det|> +Polycyclic scaffolds bearing partially dearomatized fused arenes are commonly encountered in natural products, pharmaceuticals and bioactive molecules1-5. While these molecular frameworks lead to great structural diversity and intriguing biological properties, they engender synthetic challenges as their assembly often requires prefunctionalized substrates and multi- step sequences. Direct dearomatization of fused arenes constitutes an outstandingly efficient approach for the construction of polycyclic scaffolds due to its high atom and step economy1-5. However, selective dearomatization and functionalization of fused arenes is difficult because harsh conditions are often required to disrupt aromaticity, leading to poor selectivities. As a result, dearomative functionalizations of fused arenes has been largely limited to activated arenes such as indoles6-15 and naphthols16-20. + +<|ref|>image<|/ref|><|det|>[[207, 350, 737, 870]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 886, 187]]<|/det|> +Figure 1. Methods for selective dearomative functionalization of fused arenes. a, Examples of partially dearomatic natural products and bioactive compounds. b, Selective dearomatization of quinolines. c, Selective dearomatization of naphthalenes. d, Selective dearomatization of structurally diverse fused arenes (this work). + +<|ref|>text<|/ref|><|det|>[[113, 217, 886, 475]]<|/det|> +Issues of chemo- and regioselectivity also add to the difficulty of dearomative functionalizations of unactivated fused arenes such as quinolines and naphthalenes. Selective dearomatization of pyridine moieties in fused arenes is feasible21- 25 owing to the electron deficient nature of pyridine and assistance provided by Lewis acid complexation. Selective dearomatization of phenyl moieties in fused arenes, on the other hand, is significantly more challenging. To our knowledge, only one example has been described in a recent report by Brown, Houk and Glorius involving a photochemical [4+2] cycloaddition between quinolines and alkenes (Figure 1b)26 enabled by energy transfer and Lewis acid activation. A similar [4+2] cycloaddition strategy initiated by energy transfer was also employed in the dearomatization of naphthalene by Sarlah et al. (Figure 1c)27- 32. + +<|ref|>text<|/ref|><|det|>[[113, 480, 886, 894]]<|/det|> +We wondered if a photoredox strategy could be used to selectively dearomatize the phenyl moieties in quinolines, isoquinolines, quinoxalines, naphthalenes and other fused arenes. In reported photoredox processes, radicals add selectively to the pyridine ring in quinolines33,34, and indiscriminately to the phenyl rings in naphthalenes. In contrast to Sarlah's [4+2] cycloaddition strategy, in which regioselectivity is governed primarily by steric effects27,32, we envisaged taking advantage of the relative electron richness of the phenyl ring to achieve selective dearomative functionalization (Figure 1d). Photocatalytically promoted single electron transfer (SET) oxidation of quinolines would produce radical cation I, in which the positive charge is mostly localized on the phenyl ring. We anticipated nucleophiles to add to intermediate I selectively to form radical II. Direct arene \(\mathrm{C}(sp^2)\) - H functionalization with nucleophiles under photoredox catalysis or electrophotochemical conditions have been elegantly leveraged by Nicewicz35- 38, Lambert39, Hu40, and Wickens41 to generate aromatic functionalized products. We postulated that it would be possible to direct the reaction toward the less thermodynamically stable dearomatized products by capturing radical II with a sufficiently activated hydrogen atom transfer (HAT) agent. Below, we describe the successful development of a general chemo- and regioselective method for the dearomative functionalization of diverse fused arenes using this strategy. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 641, 108]]<|/det|> +## Dearomatization of quinolines, isoquinolines, and quinoxalines + +<|ref|>text<|/ref|><|det|>[[113, 125, 886, 380]]<|/det|> +In order to undergo selective photoredox promoted dearomative functionalization, fused arenes must have oxidation potentials that enable them to be oxidized by excited states of photocatalysts (PCs) through SET. The oxidation potential of unsubstituted quinoline ( \(E^{\mathrm{ox}} = 2.23 \mathrm{~V}\) vs SCE, Supplementary Figure S1 and S2) suggests that its oxidation would be very difficult using common PCs. However, introduction of an OMe group onto the phenyl moiety of quinoline lowers the oxidation potential sufficiently (e.g., 6- methoxyquinoline 1a, \(E^{\mathrm{ox}} = 1.83 \mathrm{~V}\) vs SCE, Table 1 and Supplementary Figure S2), making it possible for it to participate in thermodynamically driven SETs with excited states of conventional organophotoredox catalysts such as acridinium and triphenylpyrrylium (TPT) salts that have excited state reduction potentials ( \(E^{\mathrm{red}}\) ) greater than 2.0 V (Table 1). + +<|ref|>image<|/ref|><|det|>[[198, 430, 805, 655]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 397, 560, 415]]<|/det|> +
Table 1. Exploration and optimization of reaction conditions.
+ +<|ref|>table<|/ref|><|det|>[[115, 672, 888, 895]]<|/det|> + +
EntryVariation from the “Standard Conditions”aYieldb (%)
1None81, 76c (&lt;5 of 4a)b
2Mes-Acr1 instead of Mes-Acr275
3Mes-Acr3 instead of Mes-Acr267
4DCA instead of Mes-Acr234
5TPT instead of Mes-Acr218
64CzIPN instead of Mes-Acr2trace
7DCM (0.5 M) with 2,6-lutidine (0.2 equiv)20
8DCM (0.2 M) with 2,6-lutidine (0.2 equiv)49
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[135, 85, 857, 279]]<|/det|> + +
9DCM (0.1 M) with 2,6-lutidine (0.2 equiv)63
10DCM (0.05 M) with 2,6-lutidine (0.2 equiv)76
11DCM (0.025 M) with 2,6-lutidine (0.2 equiv)74
12without Ph3SiSH53 (23 of 4a)b
13without Mes-Acr2trace
14without lighttrace
15390 nm light40
16O2 instead of N2trace (56 of 4a)b
+ +<|ref|>text<|/ref|><|det|>[[113, 281, 884, 371]]<|/det|> +aStandard conditions: unless otherwise specified, a mixture of 6-methoxyquinoline 1a (0.3 mmol, 0.05 M),pyrazole 1b (0.2 mmol), Ph3SiSH (0.04 mmol), and Mes-Acr2 (0.005 mmol) in DCM (under a N2 atmosphere at room temperature was irradiated with 40 W Kessil blue LEDs for 48 h. bYield determined by using 1H NMR spectroscopy using 1,3,5-trimethoxynbenzene as an internal standard. cIsolated yield. + +<|ref|>text<|/ref|><|det|>[[113, 388, 884, 902]]<|/det|> +We assessed the viability of the proposed dearomatic functionalization protocol by reacting 6-methoxyquinoline 1a with pyrazole 2a as a nucleophile under irradiation with blue LEDs in DCM solutions containing HAT agents and commercially available organic photosensitizers (Table 1). The desired dearomatization product 3a is generated in the highest yield using 1.5 equiv of 1a, 1.0 equiv of 2a (0.05 M), 2.5 mol% N-methylmeso-acridinium tetrafluoroborate (Mes-Acr1), 0.2 equiv of the HAT agent and 0.2 equiv of 2,6-lutidine. Nucleophile addition occurred exclusively at the 5-position. The efficiency of the reaction is dependent on the HAT agent (Supplementary Table S1). Among the 17 different thiols and benzeneselenols screened, Ph3SiSH was found to be the superior HAT agent, leading to regioselective formation of 3a in 81% yield (Table 1, entry 1). While photocatalysts such as TPT (E*red=2.55 V vs SCE, 18%, entry 4) with excited state reduction potentials higher than the oxidation potential of 1a were found to promote this reaction, Mes-Acr2 (2.5 mol%,=2.2 V vs SCE) was identified as the best photocatalyst (3a formed 81% 1H NMR yield, entry 1). Solvent has a significant effect on the reaction (Supplementary Table S2). Furthermore, substrate ratio (Supplementary Table S3) and concentration has a marked impact on the yield (entries 7-11) with low concentrations of 1a being more beneficial (entries 7 vs 10). The absence of 2,6-lutidine does not affect the reaction outcome (entries 1 vs 10) as quinoline 1a can serve as the base. Lastly, control experiments confirmed that the HAT agent, light and photocatalyst are all required for selective dearomatization. In particular, the HAT agent is critical for minimizing the formation of the aromatization product 4a (entries 1 vs 12). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[165, 90, 825, 152]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[163, 175, 830, 625]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[165, 630, 277, 639]]<|/det|> +
b. Late stage modification
+ +<|ref|>image<|/ref|><|det|>[[172, 640, 812, 905]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 884, 251]]<|/det|> +Figure 2. Scope of photoredox dearomatic functionalization of quinolines, isoquinolines, and quinoxalines with azoles. a, Scope of quinolines and quinoxalines with azoles. b, Late-stage functionalization. aReaction conditions: unless otherwise specified, see footnote a of Table 1 and supplementary materials; bRatio of isomers was determined by using \(^1\mathrm{H}\) NMR analysis of the crude reaction mixture; s.m.: starting materials; ddr value was determined by \(^1\mathrm{H}\) NMR analysis of the crude reaction mixture; Condition B without 2,6- lutidine; Condition A using 3,6- di- tert- butyl- 9- mesityl- 10- phenylacridin- 10- ium tetrafluoroborate as photocatalyst for 5 d. + +<|ref|>text<|/ref|><|det|>[[113, 266, 886, 600]]<|/det|> +Having established optimal conditions for the new photoredox promoted dearomatic functionalization reaction, we next explored the scope with fused arenes as substrates and azoles as nucleophiles (Figure 2a). Our results show that the reaction proceeds with high chemo- and regioselectivity for a wide range of quinolines and azoles. Both electron- donating and electron- withdrawing groups are tolerated on the pyrazole. Halogenated pyrazoles reacted with 6- methoxyquinoline with Ph3SiSH as the HAT agent (condition A giving 3b- 3d). For 5- fluoropyrazole, benzeneselenol, which is 10- fold more reactive than thiols in HAT42,43, is required along with 2,6- lutidine (condition B) to yield 3e. Other electron- deficient pyrazoles also reacted in the presence of benzeneselenol to produce 3h and 3i (condition C, without 2,6- lutidine). Notably, the dearomatization method can be applied to a pyrazole boronate, which generates 3j (51%) containing a useful handle for subsequent coupling reactions. A disubstituted pyrazole also reacted smoothly to form 3k (51%), as did other common azoles including benzotriazole (3l), 1,2,3- triazoles (3m), and tetrazole (3n) under condition B or C. + +<|ref|>text<|/ref|><|det|>[[113, 615, 888, 899]]<|/det|> +6- Alkoxy- substituted quinolines are particularly effective substrates for the dearomatic functionalization reaction, producing diverse substituted 5,8- dihydroquinolines as products (Figure 2a). A wide range of alkoxy substituents, including those containing methyl (3o, 3q, 3r, and 3v), ester (3p, 3ai), bromo (3t), amide (3s), phthalamide (3w), alkene (3x), benzyl (3y, 3z), TBS (3aa, 3ab), MOM (3ac), phenyl (3ad), ketone (3ae), alkyne (3aj), nitrile (3y, 3ak), thiophene (3af) and benzofuran (3ag) moieties, are well tolerated. Moreover, the process promotes selective dearomatization and functionalization of quinoline ring systems even when other arene moieties including benzene (3w, 3y, 3z, 3ad, 3ae), thiophene (3af) and benzofuran (3ah) are present. Notably, the protocol enables highly selective dearomatization of a quinoline ( \(E^{ox} = 1.86 \mathrm{~V}\) vs SCE) over a naphthalene ( \(E^{ox} = 1.92 \mathrm{~V}\) vs SCE) ring (3ag) despite their similar oxidation potentials. In addition to controlling chemoselectivity, 6- alkyloxy groups also direct site selective formation of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 213]]<|/det|> +5- nucleophile- substituted- 5,8- dihydroquinolines. Furthermore, dearomatization of 7- methoxyquinolines is also highly regioselective generating only 8- substituted 5,8- dihydroquinoline products (3u, 3v). Isoquinolines can participate in the process (3al, 3am). However, regioisomeric 7,8- dihydroisoquinolines are formed. Finally, quinoxalines are also viable substrate for the dearomatization process (3an). + +<|ref|>text<|/ref|><|det|>[[114, 229, 884, 432]]<|/det|> +Owing to the mild conditions required, the organophotoredox- promoted dearomatic functionalization process is applicable to late- stage functionalizations of complex pharmaceutically relevant quinoline structures including viqualine (3ao), Oppolzer's camphorsultam (3ap), galactose (3aq, 3ar), menthol (3as), probalan (3at), tyrosine (3au), cholesterol (3av), and estrone (3aw) groups (Figure 2b). It is particularly impressive that a grazoprevir analog (3ax) containing various functional groups was selectively dearomatized utilizing this process. These results underscore the mildness, high chemo- and regioselectivity, and practicality of the new protocol. + +<|ref|>sub_title<|/ref|><|det|>[[115, 485, 388, 504]]<|/det|> +## Dearomatization of fused arenes + +<|ref|>text<|/ref|><|det|>[[113, 520, 884, 828]]<|/det|> +Encouraged by the success of the new photoredox reaction of quinolines, we explored its utility for the selective dearomatization of fused non- heteroaromatics. The results show that selective dearomatization of fused arenes is more challenging. As with quinolines, electron- donating substituents can be used to lower the oxidation potentials and enhance reactivity. A 2- methoxy substituent in naphthalene, for instance, lowers the oxidation potential to 1.8 V (vs SCE, supplementary Figure S2) and enables selective oxidation by the excited state of photocatalyst Mes- Acr2. Selective dearomatization and C- 1 functionalization of 2- methoxynaphthalene by pyrazole using condition B led to 3ay (Figure 3a). The efficiency of this process is also dependent on the base and the HAT agent, as are shown in 3az, 3ba, and 3bi (condition C), 3bb, 3bf, 3bg, 3bh, and 3bj (condition B) and 3be (condition D). The dearomatization strategy can be applied to methyl substituted naphthalene (3bd). Finally, unsubstituted naphthalene (3bd) also participates in the process to produce the 1- substituted product regioselectively. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[216, 88, 783, 690]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 699, 884, 814]]<|/det|> +
Figure 3. Dearomatization of fused arenes. a, Scope of naphthalenes. b, Scope of anthracenes and phenanathrenes. c, Large scale synthesis. d, Application in synthesis of valuable targets. aReaction condition: unless specified, see footnote a of Table 1 and the ESI; bRatio of 6-methoxyanphthalene to pyrazole: 1 : 2; cRatio of naphthalene to pyrazole is 2:1; dReaction time: 4 d; eCondition B using10.0 equiv of acid; fCondition B using 5.0 equiv of acid.
+ +<|ref|>text<|/ref|><|det|>[[114, 829, 883, 901]]<|/det|> +The photoredox reaction can also be applied to the dearomatization functionalization of other common polycyclic aromatic hydrocarbons including anthracene (3bk, \(87\%\) ) and phenanthrene (3bl, \(73\%\) ) (Figure 3b). Notably, 6- methoxyphenanthrene reacts to form the trans substitution + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 886, 345]]<|/det|> +product 3bm \((84\%)\) . Moreover, nucleophiles other than azoles, such as trifluoromethanesulfonamide (3bn) and imidazole (3bo), can also participate in the process. Furthermore, carboxylic acids, which are often used as radical precursors in photoredox reactions44, serve as effective nucleophiles for dearmative functionalization of phenanthrenes. These reactions proceed in moderate to high yields (3bp- 3cb, \(25 - 89\%\) ) and display broad functional group tolerance including phenyl (3br, 3bt), azide (3bs), alkenyl (3bu), alkynyl (3bv), cyclobutyl (3bw), and hydroxyl acid groups (3cb). Due to their weaker nucleophilicity and steric effects, benzoic acid (3bx) and secondary carboxylic acids (3bw) are less effective nucleophiles for this process. In addition, the reaction can be utilized for the direct modification of biologically relevant structures such as aspirin (3bx) and amino acids (3by- 3ca). + +<|ref|>text<|/ref|><|det|>[[113, 359, 885, 720]]<|/det|> +Studies with the silyloxquinoline 1b show that the synthetic protocol can be scaled up without loss of yield (Figure 3c). Moreover, the reaction can be adapted for use in a flow system, an emerging technology in organic synthesis45. Using the flow system, the reaction time is reduced dramatically to 3 h without compromising the yield. The product silylenol ether 3aa can be converted to the corresponding alcohol 3cc. The preparative power of the new method was also demonstrated by its use in the cost- effective synthesis of important biologically active targets, including a JAK inhibitor (Figure 3d)46. Our new synthetic route starts from naphthalene (\$0.04/g) instead of the more expensive 1- bromo- 1,2,3,4- tetrahydronaphthalene (\$370/g) used in the earlier preparative pathway46. Dearomatization of naphthalene with pyrazol- 4- ylboronic acid pinacol ester produces key intermediate 3cd, which is then reduced using Rh/Al\(_2\)O\(_3\) catalyzed hydrogenation to form 3ce. The JAK inhibitor is then readily prepared from 3ce using reported procedures46. These studies clearly demonstrate the synthetic value of the our dearmative functionalization protocol, which provides direct and efficient access to polycyclic partially dearomatized frameworks that were previously inaccessible or required strenuous synthetic efforts. + +<|ref|>text<|/ref|><|det|>[[113, 735, 885, 887]]<|/det|> +Mechanistic studies. To gain insight into the reaction mechanism, chemo- and regioselectivity and the role played by the HAT agent, we conducted computational studies and additional experiments. The excited state \(\mathrm{PC}^{+*}\) Mes- Acr2\* possesses strong oxidizing power ( \(E^{\mathrm{red}} = 2.2 \mathrm{~V}\) ) and can oxidize the 6- methoxyquinoline (1a) ( \(E^{\mathrm{ox}} = 1.83 \mathrm{~V}\) vs SCE, supplementary Figure S1) to give radical cation 5a (Figure 4a and Stern- Volmer quenching experiments, supplementary Figure S3). The oxidation potential of the fused arene substrate is crucial for the success of this process, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 397]]<|/det|> +as demonstrated by the high reactivity of 1a in contrast to the lack of reactivity of unsubstituted quinoline \((E^{\mathrm{ox}} = 2.23 \mathrm{V}\) vs SCE, Supplementary Figure S2). In addition, when a mixture of 1a and 6-acetoxyquinoline (1c) \((E^{\mathrm{ox}} = 2.2 \mathrm{V}\) vs SCE, supplementary Figure S2) is subjected to the reaction conditions, only 1a undergoes selective dearomatization to form 3a, indicating that electron-withdrawing substituents inhibit the reaction. In contrast, 1a and 2- methoxynaphthalene (1e, \(E^{\mathrm{ox}} = 1.8 \mathrm{V}\) vs SCE, Supplementary Figure S2), which have similar oxidation potentials, both react under the same conditions to produce the corresponding dearomatized products 3a and 3ay (Figure 4b). Likewise, phenanthrene 1d \((E^{\mathrm{ox}} = 1.91 \mathrm{V}\) vs SCE, Supplementary Figure S2) is an effective substrate for this process (Figure 4b). Notably, the pyrazole nucleophile \((E^{\mathrm{ox}} = 2.15 \mathrm{V}\) vs SCE, Supplementary Figure S2) is not oxidized by Mes- Acr2\\*. This is supported by luminescence quenching study of Mes- Acr2 with varying concentrations of 6- methoxyquinoline 1a and pyrazole 2a (Supplementary Figure S4). + +<|ref|>image<|/ref|><|det|>[[115, 412, 883, 821]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 837, 883, 883]]<|/det|> +
Figure 4. Mechanistic studies. a, Proposed catalytic cycle. b, Chemoselectivity studies. c, Study of roles of HAT reagent. d, Study of regioselectivity of isoquinoline engaged reaction.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 80, 888, 688]]<|/det|> +Following the oxidation of 1a through SET, pyrazole reacts with the electrophilic radical cation 5a to form the radical cation 6a (Figure 4a). Calculated Hirshfeld charge and spin densities (Figure 4a; see Supplementary Figure S5 for calculated charge and spin densities for other substrates) indicate that C5 has the highest spin density (bold face) in 5a and is therefore the preferred site for nucleophilic addition. The calculated free energies also support the observation that reaction at C5 position is kinetically favored over C8 or C7 by at least 3 kcal/mol (Supplementary Figure S6). Subsequent exothermic deprotonation of 6a by 2,6- lutidine gives the neutral radical 7a. Calculated free energies suggest that sequential SET oxidation and deprotonation of 7a to form the aromatic substitution product 4a is more thermodynamically favorable (- 24.9 kcal/mol, Figure 5). This is reflected in the outcomes of processes leading to fully aromatic products developed by Nicewicz \(^{35 - 38}\) , Lambert \(^{39}\) , \(\mathrm{Hu}^{40}\) and Wickens \(^{41}\) . However, the presence of a reactive HAT agent (e.g., Ph \(_3\) SiSH) redirects the course of the reaction toward the dearomatized product 3a rather than the more stable aromatic product 4a (Figure 5). The transformation of 7a into 3a is facilitated by the exergonic nature of the HAT step (- 17.4 kcal/mol change in free energy). Thus, success of the dearomatization process depends on the use of thiol and selenol HAT agents that have weak S- H and Se- H bonds (S- H: 82 kcal/mol, Se- H: 73 kcal/mol) and correspondingly high H- atom transfer rates (PhSH: \(\mathrm{K}_{20} = 9.0 \times 10^{7} \mathrm{M}^{- 1} \mathrm{S}^{- 1}\) and PhSeH: \(\mathrm{K}_{20} = 1.3 \times 10^{9} \mathrm{M}^{- 1} \mathrm{S}^{- 1}\) ) \(^{42,43,47}\) . Furthermore, the process is highly regioselective. We did not observe the formation of regioisomeric 3a'. The observation is supported by the calculation that its formation is unfavorable by 3.8 kcal/mol due to the steric effect (Figure 5). Control experiments showed that in the absence of the HAT agent Ph \(_3\) SiSH, the yield of the dearomatized product dropped to 46% (Figure 4c). After the HAT step, the Ph \(_3\) Si' radical is reduced by the PC' Mes- Acr2 radical to generate the Ph \(_3\) Si' anion, which is protonated to give Ph \(_3\) SiH and the PC Mes- Acr2 (Figure 4a). + +<|ref|>text<|/ref|><|det|>[[113, 699, 886, 903]]<|/det|> +The effect of the HAT agent is particularly noticeable in reactions of isoquinolines and naphthalenes with pyrazole. Only when PhSeH is present do these processes produce the desired products 3al and 3ay (Figure 4c). Furthermore, the polarity match \(^{48}\) between the protic S- H/Se- H and the nucleophilic carbon radical 7a might further facilitate the HAT process. Unexpectedly, reactions between isoquinolines and pyrazole give regioisomeric 7,8- dihydroisoquinolines. Our experiments suggest that the reaction initially forms 5,8- dihydroisoquinoline 3al', which is then converted to the more stable 7,8- dihydroisoquinoline 3al (Figure 4d). The 5,8- dihydroisoquinoline 3al' was obtained in the presence of \(\mathrm{H}_2\mathrm{O}\) . Two possible pathways could lead to 5,8- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 214]]<|/det|> +dihydroisoquinoline 3al. However, the deprotonation/C=C isomerization pathway was ruled out as we did not observe the formation of 3al when 3al' was treated with 2,6- lutidine, a base used in the dearomatization process. In contrast, 3al was formed under the photoredox reaction conditions through SET oxidation of the C=C bond, deprotonation and HAT processes. This pathway was also supported by calculated reaction energies (Supplementary Figure S7). + +<|ref|>text<|/ref|><|det|>[[113, 228, 886, 536]]<|/det|> +An alternative pathway for the formation of the functionalized dearomatized product 3a is through reduction of radical 7a by SET from PC' to form anion 8a, which then undergoes protonation. However, calculations suggest that the formation of 8a is highly unfavorable (Figure 5). Moreover, preliminary experiments showed that dearomatized product 3a ( \(E^{ox} = 1.60 \mathrm{~V}\) vs SCE, Supplementary Figure S2) could be converted back to starting material 1a via PC'\\* (Mes- Acr2\\*) mediated oxidation under the standard reaction conditions (Figure 4c). Unexpectedly, we found that the HAT agent Ph3SiSH retards the efficiency of this process (Figure 4c). Specifically, photoinduced reaction of 3a in the presence of Ph3SiSH leads to \(61\%\) recovery of 3aa and formation of \(28\%\) of 1b. In contrast, in the absence of the HAT agent, 1b was produced in \(49\%\) yield. Thus, these observations indicate that the HAT agent not only facilitates the formation of dearomatized products, it also inhibits the conversion of these products back to aromatic starting materials. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 85, 848, 662]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 677, 883, 724]]<|/det|> +
Figure 5. Calculated free energy diagram (in kcal/mol) at the ωB97X-D/def2-TZVPP, SMD (CH₂Cl₂)//ωB97X-D/def2-SVP level of theory.
+ +<|ref|>text<|/ref|><|det|>[[114, 739, 884, 893]]<|/det|> +In summary, we developed a conceptually unique organophotoredox catalytic dearomative functionalization strategy for the selective disruption of aromaticity in fused arenes. A reactive HAT agent is utilized to overcome the competing aromatization pathway, which is typically more favorable, and achieve the selective formation of dearomatizatized products. Experimental and computational mechanistic studies support the key role played by HAT agents in the unprecedented process. The preparative power of the protocol was demonstrated by applications to structurally + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 884, 213]]<|/det|> +diverse fused arenes including quinolines, isoquinolines, quinoxalines, naphthalenes, anthracenes and phenanthrenes. The method can also be applied to the synthesis and late- stage skeletal editing of complex pharmaceutically valued structures. We anticipate that the new process will enable facile access to a wide range of synthetically versatile frameworks and accelerate the construction of new molecular architectures for drug discovery. + +<|ref|>sub_title<|/ref|><|det|>[[114, 230, 210, 248]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[110, 264, 888, 840]]<|/det|> +1. Pape, A. R., Kaliappan, K. P. & Kündig, E. P. Transition-metal-mediated dearomatization reactions. Chem. Rev. 100, 2917-2940 (2000). +2. López Ortiz, F., Iglesias, M. J., Fernández, I., Andújar Sánchez, C. M. & Ruiz Gómez, G. Nucleophilic dearomatizing (DNAr) reactions of aromatic C-H systems. A mature paradigm in organic synthesis. Chem. Rev. 107, 1580-1691 (2007). +3. Roche, S. P. & Porco, J. A. Dearomatization strategies in the synthesis of complex natural products. Angew. Chem. Int. Ed. 50, 4068-4093 (2011). +4. Zheng, C. & You, S.-L. Advances in catalytic asymmetric dearomatization. ACS Cent. Sci. 7, 432-444 (2021). +5. Huck, C. J. & Sarlah, D. Shaping molecular landscapes: recent advances, opportunities, and challenges in dearomatization. Chem 6, 1589-1603 (2020). +6. Roche, S. P., Tendoung, J.-J. Y. & Tréguier, B. Advances in dearomatization strategies of indoles. Tetrahedron 71, 3549-3591 (2015). +7. Zheng, C. & You, S. L. Catalytic asymmetric dearomatization (CADA) reaction-enabled total synthesis of indole-based natural products. Nat. Prod. Rep. 36, 1589-1605 (2019). +8. Gentry, E. C., Rono, L. J., Hale, M. E., Matsuura, R. & Knowles, R. R. Enantioselective synthesis of pyrroloindolines via noncovalent stabilization of indole radical cations and applications to the synthesis of alkaloid natural products. J. Am. Chem. Soc. 140, 3394-3402 (2018). +9. An, J., Zou, Y.-Q., Yang, Q.-Q., Wang, Q. & Xiao, W.-J. Visible light-induced aerobic oxymidation of indoles: a photocatalytic strategy for the preparation of tetrahydro-5h-indolo[2,3-b]quinolinols. Adv. Synth. Catal. 355, 1483-1489 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 888, 870]]<|/det|> +10. Zhu, M., Zheng, C., Zhang, X. & You, S.-L. Synthesis of cyclobutane-fused angular tetracyclic spiroindolines via visible-light-promoted intramolecular dearomatization of indole derivatives. J. Am. Chem. Soc. 141, 2636–2644 (2019). +11. Ma, J., Schäfers, F., Daniliuc, C., Bergander, K., Strassert, C. A. & Frank Glorius Gadolinium photocatalysis: Dearomative [2+2] cycloaddition/ring-expansion sequence with indoles. Angew. Chem. Int. Ed. 59, 9639-9645 (2020). +12. Liu, K. et al Electrooxidation enables highly regioselectivedearomative annulation of indole and benzofuran derivatives. Nat. Commun. 11, https://doi.org/10.1038/s41467-019-13829-4 (2020). +13. Wu, J., Dou, Y., Guillot, R., Cyrille Kouklovsky, C., & Vincent, G. Electrochemical dearomative 2,3-difunctionalization of indoles. J. Am. Chem. Soc. 141, 2832–2837 (2019). +14. Studies from our group and ref 15: Zhang, Y., Ji, P., Gao, F., Dong, Y., Huang, H., Wang, C., Zhou, Z., & Wang, W. Organophotocatalytic dearomatization of indoles, pyrroles and benzo(thio)furans via a Giese-type transformation. Commun. Chem. 4, 20 (2021). +15. Zhang, Y., Ji, P., Gao, F., Huang, H., Zeng, F., & Wang, W. Photorecox asymmetric nucleophilic dearomatization of indoles with neutral radicals. ACS Catal. 11, 998-1007 (2021) +16. Pouységu, L., Deffieux, D., & Quideau, S. Hypervalent iodine-mediated phenol dearomatization in natural product synthesis. Tetrahedron, 66, 2235-2261 (2010). +17. Dhineshkumar, J., Samaddar, P., & Prabhu, K. R. A copper catalyzed azidation and peroxidation of β-naphthols via an oxidative dearomatization strategy. Chem. Commun. 52, 11084-11087 (2016). +18. Kulish, K., Boldrini, C., Reis, M. C., Pérez, J. M., & Harutyunyan, S. R. Lewis acid promoted dearomatization of naphthols. Chem. Eur. J. 26, 15843 (2020). +19. Sun, W., Li, G., Hong, L., & Wang, R. Asymmetric dearomatization of phenols. Org. Biomol. Chem. 2164-2176 (2016) +20. Zhu, G., Bao, G., Li, Y., Yang, J., Sun, W., Li, J., Hong, L., & Wang, R. Chiral phosphoric acid catalyzed asymmetric oxidative dearomatization of naphthols with quinones. Org. Lett. 18, 5288-5291 (2016). +21. Takamura, M., Funabashi, K., Kanai, M., & Shibasaki, M. Asymmetric Reissert-type reaction promoted by bifunctional catalyst. J. Am. Chem. Soc. 122, 6327-6328 (2000). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 888, 870]]<|/det|> +22. Zurro, M., Asmus, S., Beckendorf, S., Mück-Lichtenfeld, C., & Mancheño, O. G. Chiral helical oligotriazoles: new class of anion-binding catalysts for the asymmetric dearomatization of electron-deficient \(N\)-heteroarenes. J. Am. Chem. Soc. 136, 13999-14002 (2014). +23. Pappoppula, M., Cardoso, F. S. P., Garrett, B. O., & Aponick, A. Enantioselective copper-catalyzed quinoline alkynylation. Angew. Chem. Int. Ed. 54, 15202-15206 (2015). +24. Yan, X., Ge, L., Castiñeira Reis, M., & Harutyunyan, S. R. Nucleophilic dearomatization of \(N\)-heteroaromatics enabled by Lewis acids and copper catalysis. J. Am. Chem. Soc. 142, 20247-20256 (2020). +25. Leitch, J. A., Rogova, T., Duarte, F., & Dixon, D. J. Dearomatic photocatalytic construction of bridged 1,3-diazepanes. Angew. Chem. Int. Ed. 59, 4121-4130 (2020). +26. Ma, J., et al Photochemical intermolecular dearomatic cycloaddition of bicyclic azaarenes with alkenes. Science 371, 1338-1345 (2021). +27. Southgate, E. H., Pospech, J., Fu, J., Holycross, D. R., & Sarlah, D. Dearomatic dihydroxylation with arenophiles. Nat. Chem. 8, 922-928 (2016). +28. Okumura, M., Shved, A. S., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-carboamination. J. Am. Chem. Soc. 139, 17787-17790 (2017). +29. Hernandez, L. W., Klockner, U., Pospech, J., Hauss, L., & Sarlah, D. Nickel-catalyzed dearomatic trans-1,2-carboamination. J. Am. Chem. Soc. 140, 4503-4507 (2018). +30. Tang, C., Okumura, M., Deng, H., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-oxymination. Angew. Chem. Int. Ed. 58, 15762-15766 (2019). +31. Wertjes, W. C., Okumura, M., & Sarlah, D. Palladium-catalyzed dearomatic \(syn\)-1,4-diamination. J. Am. Chem. Soc. 141, 163-167; (2019). +32. Siddiqi, Z., Wertjes, W. C. & Sarlah, D. Chemical equivalent of arene monooxygenases: dearomatic synthesis of arene oxides and oxepines. J. Am. Chem. Soc. 142, 10125-10131 (2020). +33. Duncton, M. A. J., Minisci reactions: Versatile CH-functionalizations for medicinal chemists. Med. Chem. Commun. 2, 1135-1161 (2011). +34. Proctor, R. S. J. & Phipps, R. J. Recent advances in Minisci-type reactions. Angew. Chem. Int. Ed. 58, 13666-13699 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 888, 899]]<|/det|> +35. Romero, N. A., Margrey, K. A., Tay, N. E., & Nicewicz, D. A. Site-selective arene C-H amination via photoredox catalysis. Science 349, 1326-1330 (2015). +36. Margrey, K. A., McManus, J. B., Bonazzi, S., Zecri, F., & Nicewicz, D. A. Predictive model for site-selective aryl and heteroaryl C-H functionalization via organic photoredox catalysis. J. Am. Chem. Soc. 139, 11288-11299 (2017). +37. McManus, J. B. & Nicewicz, D. A. Direct C-H cyanation of arenes via organic photoredox catalysis. J. Am. Chem. Soc. 139, 2880-2883 (2017). +38. Margrey, K. A., Levens, A., & Nicewicz, D. A. Direct aryl C-H amination with primary amines using organic ohotoredox catalysis. Angew. Chem. Int. Ed. 56, 15644-15648 (2017). +39. Huang, H., Strater, Z. M., Rauch, M. Shee, J., Sisto, T. J., Nuckolls, C., & Lambert, T. H. Electrophotocatalysis with a trisaminocyclopropenium radical dication. Angew. Chem. Int. Ed. 58, 13318-13322 (2019). +40. Zhang, L., Liardet, L., Luo, J., Ren, D., Grätzel, M., & Hu, X. Photoelectrocatalytic arene C-H amination. Nat. Catal. 2, 366-373 (2019). +41. Targos, K., Williams, O. P., & Wickens, Z. K. Unveiling Potent Photooxidation Behavior of Catalytic Photoreductants. J. Am. Chem. Soc. 143, 4125-4132 (2021). +42. Newcomb, M., Choi, S.-Y., & Horner, J. H. Adjusting the Top End of the Alkyl Radical Kinetic Scale. Laser Flash Photolysis Calibrations of Fast Radical Clocks and Rate Constants for Reactions of Benzeneselenol. J. Org. Chem. 64, 1225-1231; (1999). +43. Newcomb, M., Varick, T. R., Ha, C., Manek, M. B., & Yue, X. Picosecond radical kinetics. Rate constants for reaction of benzeneselenol with primary alkyl radicals and calibration of the 6-cyano-5-hexenyl radical cyclization. J. Am. Chem. Soc. 114, 8158-8163 (1992). +44. Xuan, J., Zhang, Z.-G., & Xiao, W.-J. Visible-light-induced decarboxylative functionalization of carboxylic acids and their derivatives. Angew. Chem. Int. Ed. 54, 15632-15641 (2015). +45. Plutschack, M. B., Pieber, B., Gilmore, K., & Seeberger, P. H. The Hitchhiker's guide to flow chemistry. Chem. Rev. 17, 11796-11893 (2017). +46. Rodgers, J. D., et al. Heteroaryl substituted pyrrolo[2,3-b]pyridines and pyrrolo[2,3-b]pyrimidines as Janus kinase inhibitors. US2009181959A1. +47. Mayer, J. M. Understanding hydrogen atom transfer: from bond strengths to Marcus theory. Acc. Chem. Res. 44, 36-46 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 883, 135]]<|/det|> +48. Roberts, B. P. Polarity-reversal catalysis of hydrogen-atom abstraction reactions: concepts and applications in organic chemistry. Chem. Soc. Rev. 28, 25-35 (1999). + +<|ref|>text<|/ref|><|det|>[[114, 151, 884, 276]]<|/det|> +Acknowledgements W.W. acknowledges the NIH (5R01GM125920) and University of Arizona for financial support and the NSF MRI for acquisition of 500 MHz NMR spectrometer (1920234). S.C. is grateful to Oberlin College for financial support. K.N.H thanks the National Science Foundation (Grant CHE-1764328). DFT calculations were performed using the SCIURus, the Oberlin College HPC cluster (NSF MRI 1427949). + +<|ref|>text<|/ref|><|det|>[[114, 304, 884, 415]]<|/det|> +Author Information P. J., F. G., J. C., and X. M. conducted and analyzed the synthetic experiments, C. C. D. performed DFT calculations under the direction of S. C., K. N. H. and W. W planned, designed and directed the project and P. J., S. C., K. N. H and W. W. wrote the manuscript. + +<|ref|>text<|/ref|><|det|>[[115, 428, 632, 448]]<|/det|> +Competing interests The authors declare no competing interests. + +<|ref|>text<|/ref|><|det|>[[114, 469, 884, 610]]<|/det|> +Additional Information Supplementary information and chemical compound information accompany this paper at www.nature.com/naturec. Reprints and permission information is available online at http://npg.nature.com/reprintsandpermission/. Correspondence and requests for materials should be addressed to K. N. H. (houk@chem.ucla.edu), S. C. (shuming.chen@oberlin.edu) and W. W. (wwang@pharmacy.arizona.edu). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 130, 137, 149]]<|/det|> +- Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/images_list.json b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..20ab00c8e3bdc419d393c394b128447d5a44f21e --- /dev/null +++ b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 Acquisition and characterization of antibodies. Serum antibody titers of six SARS- CoV- 2 convalescent patients to the SARS- CoV- 2 S (a) and N (b) proteins measured by ELISA. Sorting of single plasma cells (c) with CD38 and CD27 double- positive B cells and single N and S protein specific memory B cells (d) by", + "footnote": [], + "bbox": [ + [ + 44, + 92, + 950, + 840 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 50, + 140, + 825, + 870 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 50, + 188, + 820, + 920 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 50, + 304, + 949, + 910 + ] + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2.mmd b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2.mmd new file mode 100644 index 0000000000000000000000000000000000000000..3c898e6c1d8c23b6dae0b90631b9ed9a5b2ecf28 --- /dev/null +++ b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2.mmd @@ -0,0 +1,426 @@ + +# A COVID-19 antibody curbs SARS-CoV-2 nucleocapsid protein-induced complement hyperactivation + +Sisi Kang Sun Yat- sen University + +Mei Yang The Fifth Affiliated Hospital of Sun Yat- sen University + +Suhua He The Fifth Affiliated Hospital of Sun Yat- sen University + +Yueming Wang Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +Xiaoxue Chen Department of Experimental Medicine, Guangdong Provincial Key Laboratory of Biomedical Imaging, The Fifth affiliated Hospital, Sun Yat- sen University, Zhuhai, 519000 + +Yao- Qing chen Sun Yat- sen University + +Zhongsi Hong Fifth Affiliated Hospital of Sun Yat- sen University + +Jing Liu The Fifth Affiliated Hospital of Sun Yat- sen University + +Guanmin Jiang Fifth Affiliated Hospital of Sun Yat- sen University + +Qiuyue Chen Sun Yat- sen University + +Ziliang Zhou Sun Yat- sen University https://orcid.org/0000- 0002- 6801- 4180 + +Zhechong Zhou Sun Yat- sen University + +Zhaoxia Huang Sun Yat- sen University + +Xi Huang Fifth Affiliated Hospital of Sun Yat- sen University + +Huanhuan He Guangdong Provincial Engineering Research Center of Molecular Imaging + +<--- Page Split ---> + +Weihong Zheng Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +Hua- Xin Liao Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +Fei Xiao Sun Yat- sen University + +Hong Shan Sun Yat- sen University + +Shoudeng Chen ( chenshd5@mail.sysu.edu.cn ) Sun Yat- sen University https://orcid.org/0000- 0002- 7634- 2141 + +## Article + +Keywords: human monoclonal antibody, COVID- 19, SARS- CoV- 2, nucleocapsid protein, crystal structure, complement hyperactivation, viral protein targeting therapy, MASP- 2 + +Posted Date: December 2nd, 2020 + +DOI: https://doi.org/10.21203/rs.3.rs- 106760/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on May 11th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 23036- 9. + +<--- Page Split ---> + +# A SARS-CoV-2 antibody curbs viral N protein-induced + +# complement hyperactivation + +3 Sisi Kang1\*, Mei Yang1\*, Suhua He1\*, Yueming Wang2,3\*, Xiaoxue Chen1, Yao- 4 Qing Chen4, Zhongsi Hong5, Jing Liu6, Guanmin Jiang7, Qiuyue Chen1, Ziliang 5 Zhou1, Zhechong Zhou1, Zhaoxia Huang1, Xi Huang8, Huanhuan He1, Weihong 6 Zheng2,3, Hua-Xin Liao2,3,#, Fei Xiao1,5,#, Hong Shan1,9,#, Shoudeng Chen1,10,# 7 1. Molecular Imaging Center, Guangdong Provincial Key Laboratory of 8 Biomedical Imaging, The Fifth Affiliated Hospital, Sun Yat- sen University, 9 Zhuhai, 519000, China 10 2. Institute of Biomedicine, Jinan University, Guangzhou, 510632, China 11 3. Zhuhai Trinomab Biotechnology Co., Ltd., Zhuhai, 519040, China 12 4. School of Public Health (Shenzhen), Sun Yat- sen University, Shenzhen, 13 China 14 5. Department of Infectious Disease, The Fifth Affiliated Hospital, Sun Yat- sen 15 University, Zhuhai, 519000, China 16 6. Department of Respiratory Disease, The Fifth Affiliated Hospital, Sun Yat- sen 17 University, Zhuhai, 519000, China 18 7. Department of Clinical laboratory, The Fifth Affiliated Hospital of Sun Yat- sen 19 University, Zhuhai, 519000, China 20 8. Center for Infection and Immunity, The Fifth Affiliated Hospital, Sun Yat- sen 21 University, Zhuhai, 519000, China + +<--- Page Split ---> + +22 9. Department of Intervention Medicine, The Fifth Affiliated Hospital, Sun Yatsen University, Zhuhai, 519000, China +24 10. Department of Experimental Medicine, The Fifth Affiliated Hospital, Sun Yatsen University, Zhuhai, 51 9000, China +26 \* These authors contributed equally to this work +27 # Co-correspondence: Shoudeng Chen (chenshd5@mail.sysu.edu.cn); Hong Shan (shanhong@mail.sysu.edu.cn); Fei Xiao (xiaof35@mail.sysu.edu.cn); Hua-Xin Liao (tliao805@jnu.edu.cn) + +<--- Page Split ---> + +## Summary + +Although human antibodies elicited by the severe acute respiratory syndrome coronavirus 2 (SARS- CoV- 2) nucleocapsid (N) protein are profoundly boosted upon infection, little is known about the function of N- reactive antibodies. Herein, we isolated and profiled a panel of 32 N protein- specific monoclonal antibodies (mAbs) from a quick recovery coronavirus disease- 19 (COVID- 19) convalescent patient who had dominant antibody responses to the SARS- CoV- 2 N protein rather than to the SARS- CoV- 2 spike (S) protein. The complex structure of the N protein RNA binding domain with the mAb with the highest binding affinity (nCoV396) revealed changes in the epitopes and antigen's allosteric regulation. Functionally, a virus- free complement hyperactivation analysis demonstrated that nCoV396 specifically compromises the N protein- induced complement hyperactivation, which is a risk factor for the morbidity and mortality of COVID- 19 patients, thus laying the foundation for the identification of functional anti- N protein mAbs. + +Keywords: human monoclonal antibody, COVID- 19, SARS- CoV- 2, nucleocapsid protein, crystal structure, complement hyperactivation, viral protein targeting therapy, MASP- 2 + +<--- Page Split ---> + +## Main + +The fatality rate of critical condition coronavirus disease 2019 (COVID- 19) patients is exceptionally high (40% - 49%)1,2. Acute respiratory failure and generalized coagulopathy are significant aspects associated with morbidity and mortality3- 5. A subset of severe COVID- 19 patients has distinct clinical features compared to classic acute respiratory distress syndrome (ARDS), with delayed onset of respiratory distress6 and relatively well- preserved lung mechanics despite the severity of hypoxemia7. It has been reported that complement- mediated thrombotic microvascular injury in the lung may contribute to atypical ARDS features of COVID- 19, accompanied by extensive deposition of the alternative pathway (AP) and lectin pathway (LP) complement components8. Indeed, complement activation is found in multiple organs of severe COVID- 19 patients in several other studies9,10, as well as in patients with severe acute respiratory syndrome (SARS)11,12. A recent retrospective observational study of 11,116 patients revealed that complement disorder was associated with the morbidity and mortality of COVID- 1913. + +The nucleocapsid (N) protein of severe acute respiratory syndrome coronavirus (SARS- CoV- 2), the etiology agent of COVID- 19, is one of the most abundant viral structural proteins with multiple functions inside the viral particles, the host cellular environment, and in ex vivo experiments14- 20. Among these functions, a recent preprint study found that the SARS- CoV- 2 N protein bound to MBL (mannan- binding lectin)- associated serine protease 2 (MASP- 2) and resulted in + +<--- Page Split ---> + +complement hyperactivation and aggravated inflammatory lung injury19. Consistently, the highly pathogenic SARS- CoV N protein was also found to bind with MAP19, an alternative product of MASP- 221. + +Although systemic activation of complement plays a pivotal role in protective immunity against pathogens, hyperactivation of complement may lead to collateral tissue injury. Thus, how to precisely regulate virus- induced dysfunctional complement activation in COVID- 19 patients remains to be elucidated. The SARS- CoV- 2 N protein is a highly immunopathogenic viral protein that elicits high titers of binding antibodies in humoral immune responses22- 24. Several studies have reported the isolation of human monoclonal antibodies (mAbs) targeting the SARS- CoV- 2 spike (S) protein, helping explain the possible developing therapeutic interventions for COVID- 1922,25- 29. However, little is known about the potential therapeutic applications of N protein- targeting mAbs in the convalescent B cell repertoire. Herein, we report a human mAb derived from the COVID- 19 convalescent patient that specifically targets the SARS- CoV- 2 N protein and functionally compromises complement hyperactivation ex vivo. + +## Isolation of N protein-reactive mAbs + +To profile the antibody response to the SARS- CoV- 2 N protein in patients during the early recovery phase, we collected blood samples from six convalescent patients seven to 25 days after the onset of the disease symptoms. All patients recovered from COVID- 19 during the outbreak in Zhuhai, Guangdong Province, + +<--- Page Split ---> + +China, with ages ranging from 23 to 66 years (Extended Data Table 1). Our work and use of patients' samples is in accordance with the declaration of Helsinki, medical ethics standards and China's laws. Our study was approved by the Ethics Committee of The Fifth Affiliated Hospital, Sun Yat- sen University, and all patients signed informed consent forms. SARS- CoV- 2 nasal swabs reverse transcription polymerase chain reaction (RT- PCR) tests were used to confirm that all 6 COVID- 19 patients were negative for SARS- CoV- 2 at the time of blood collection. Plasma samples and peripheral blood mononuclear cells (PBMCs) were isolated for serological analysis and antibody isolation. Serum antibody titers to SARS- CoV- 2 S and N proteins were measured by enzyme- linked immunosorbent assay (ELISA) (Figure 1a, 1b, and Extended Data Table 1). Serologic analysis demonstrated that serum antibody titers to the N protein were substantially higher than those to the S protein in most of the patients. For example, ZD004 and ZD006 had only minimal levels of antibody response to the S protein, while they had much higher antibody titers to the N protein. Notably, the time from disease onset to complete recovery from clinical symptoms of COVID- 19 patient ZD006 was only 9 days (Extended Data Table 1). + +To take advantage of patient ZD006, who was still in the early recovery phase with a high possibility of a high percentage of antigen- specific plasma cells, single plasma cells (Figure 1c) with the phenotype of CD3\*/CD14\*/CD16\*/CD235a\*/CD19\*/CD20low- neg\*/CD27hi\*/CD38hi as well as antigen- specific memory B cells with the phenotype of CD19\*/CD27\* (Figure 1d) were sorted from PBMCs of patient ZD006 by fluorescence- activated cell sorting (FACS). To ensure an + +<--- Page Split ---> + +unbiased assessment, the sorting of antigen- specific memory B cells was carried out with combined probes for both fluorophore- labeled S and N recombinant proteins. Variable regions of immunoglobulin (Ig) heavy and light chain gene segment ( \(\mathrm{V}_{\mathrm{H}}\) and \(\mathrm{V}_{\mathrm{L}}\) , respectively) pairs from the sorted single cells were amplified by RT- PCR, sequenced, annotated and expressed as recombinant mAbs using the methods described previously30. Recombinant mAbs were screened against SARS- CoV- 2 S and N proteins. In total, we identified 32 mAbs that reacted with the SARS- CoV- 2 N protein, including 20 mAbs from plasma cells and 12 mAbs from memory B cells (Extended Data Table 2). We found that IgG1 is the predominant isotype at \(46.9\%\) , followed by IgG3 ( \(25.0\%\) ), IgA ( \(18.8\%\) ), IgG2 ( \(6.3\%\) ) and IgM (3.1) (Figure 1e). \(\mathrm{V}_{\mathrm{H}}\) gene family usage in SARS- CoV- 2 N protein- reactive antibodies was \(18.8\%\) for \(\mathrm{V}_{\mathrm{H}1}\) , \(62.5\%\) for \(\mathrm{V}_{\mathrm{H}3}\) , \(9.4\%\) for \(\mathrm{V}_{\mathrm{H}4}\) , \(6.2\%\) for \(\mathrm{V}_{\mathrm{H}5}\) and \(3.1\%\) for \(\mathrm{V}_{\mathrm{H}7}\) , respectively (Figure 1f), which was similar to the distribution of \(\mathrm{V}_{\mathrm{H}}\) families collected in the NCBI database. Nine of 32 SARS- CoV- 2 N protein- reactive antibodies had no mutations from their germline \(\mathrm{V}_{\mathrm{H}}\) and \(\mathrm{V}_{\mathrm{H}}\) gene segments (Figure 1f, Extended Data Table 2). The average mutation frequency of the remaining mutated antibodies was \(5.3\%\) (+/- 3.6%) in \(\mathrm{V}_{\mathrm{H}}\) and \(3.5\%\) (+/- 2.7%) in \(\mathrm{V}_{\mathrm{L}}\) . + +Consistent with the lower serum antibody titers to the SARS- CoV- 2 S protein, we identified only eight SARS- CoV- 2 S protein- reactive mAbs, including 5 antibodies from plasma cells and three antibodies from memory B cells. The \(\mathrm{V}_{\mathrm{H}}\) gene segment of the S protein- reactive antibodies had either no (6/8) or minimal (1/300) mutations (Figure 1g). There were no significant differences in + +<--- Page Split ---> + +complementarity- determining region 3 (CDR3) length in amino acid residues between the N- (Figure 1h) and S- reactive antibodies (Figure 1i). + +Approximately a quarter of antibodies directed to the N protein (Figure 1f) and almost all of antibodies to the S protein that had no or minimal mutations from their germlines (Figure 1g) had a primary antibody response similar to other typical primary viral infections. However, relatively high \(V_H\) mutation frequencies (mean of \(5.7\%\) ) of the majority of antibodies to the N proteins were more similar to the mutation frequencies of antibodies from the secondary responses to the influenza vaccination reported previously31. Although patient ZD006 was hospitalized for only nine days after the onset of COVID- 19 symptoms, the patient had high serum antibody titers, and the majority of the isolated N- reactive antibodies had a high mutation frequency, whereas the S- reactive antibodies had no or minimal mutations. These results reflect a much stronger antigen stimulation to the host driven by the SARS- CoV- 2 N protein than by the S protein. + +## Binding characterizations of anti-N mAbs + +To determine the antigenic targets by the N- reactive antibodies, we next analyzed the binding activities by ELISA with variant constructs of the N protein (full length N protein (N- FL): 1- 400; N protein N- terminal domain (N- NTD): 41- 174; and N protein C- terminal domain (N- CTD): 250- 364) (Figure 2a). Among the 32 mAbs that bound to N- FL, 13 antibodies bound to N- NTD, and one antibody bound to N- CTD (Figure 2b). A total of nine antibodies, including one antibody (nCoV400) that bound to N- CTD, seven mAbs (nCoV396, nCoV416, + +<--- Page Split ---> + +nCoV424, nCoV425, nCoV433, nCoV454, and nCoV457) that bound to N- NTD and one mAb (nCoV402) that bound only to N- FL but not to the other variant N proteins, were chosen as representative antibodies for further studies. Purified antibodies were confirmed to bind the N- FL protein by ELISA (Figure 2c). The affinity of these antibodies to the N- FL protein was measured by surface plasmon resonance (SPR) (Figure 2d). In an effort to further characterize the functional and structural relationships, three antibodies, nCoV396, nCoV416 and nCoV457, were selected for the production of recombinant fragment antigen- binding (Fab) antibodies based on their unique characteristics. The mAb nCoV396 has a \(\mathrm{V_H}\) mutation frequency of \(2.8\%\) but a high binding affinity with a \(\mathrm{K_D}\) of 1.02 nM (Figure 2d) to the N protein. The mAbs nCOV416 and nCOV457 have high \(\mathrm{V_H}\) mutations at \(11.1\%\) and \(8.7\%\) , respectively and have a binding affinity to the N protein with \(\mathrm{K_D}\) values of 7.26 nM and 12.6 nM, respectively (Figure 2d, Extended Data Table 3). + +## Complex structure of mAb with N-NDT + +To investigate the molecular interaction mechanism of the mAb nCoV396 with the N protein, we next solved the complex structure of the SARS- CoV- 2 N- NTD with the nCoV396 Fab nCoV396Fab at a 2.1 Å resolution by X- ray crystallography. The final structure was fitted with the visible electron density spanning residues 49- 173 (SARS- CoV- 2 N- NTD), 1- 220 (nCoV396Fab, the heavy chain of the Fab), and 1- 213 (nCoV396Fab, the light chain of Fab, except for the residues ranging from 136- 141). The complete statistics for the data collection, phasing, and refinement are presented in Extended Data Table 4. + +<--- Page Split ---> + +With the help of the high- resolution structure, we were able to designate all complementarity- determining regions (CDRs) in nCoV396Fab as the following: light chain CDR1, residues 23- 32 (L- CDR1), light chain CDR2, residues 51- 54 (L- CDR2), light chain CDR3, residues 94- 100 (L- CDR3), heavy chain CDR1, residues 26- 33 (H- CDR1), heavy chain CDR2, residues 51- 57 (H- CDR2), and heavy chain CDR3, residues 99- 108 (H- CDR3). Among them, we identified the interaction interface between N- NTD and L- CDR1, L- CDR3, H- CDR1, H- CDR2, and H- CDR3 of nCoV396Fab with an unambiguous electron density map (Figure 3a, Extended Data Figure 1a). + +The interacting CDRs pinch the CT tail of the SARS- CoV- 2 N- NTD (residues ranging from 159 to 172), with extensive binding contacts and a buried surface area of 1,079 Ų (Extended Data Table 5). Light chain L- CDR1 and L- CDR3 of nCoV396Fab interact with residues ranging from 159- 163 of N- NTD via numerous hydrophilic and hydrophobic contacts (Figure 3b, Extended Data Figure 1b). Notably, the SARS- CoV- 2 N- NTD residue Q163 is recognized by the L- CDR3 residue T95 via a hydrogen bond and simultaneously stacks with the L- CDR3 residue W96 and the L- CDR1 residue Y31 (Figure 3c). In addition, a network of interactions from the heavy chain H- CDR2 and H- CDR3 of nCoV396Fab to residues 165- 172 of N- NTD suggests that the conserved residue K169 of SARS- CoV- 2 N- NTD has a critical role in nCoV396 antibody binding. K169 is recognized via hydrogen bonds with the E99 δ- carboxyl group and the T100, D102, S105 main- chain carbonyl groups inside the H- CDR3 of nCoV396Fab (Figure 3d). In addition, SARS- CoV- 2 N- NTD L167 also interacts + +<--- Page Split ---> + +with I33, V50, N57, and A59 of H- CDR1 and H- CDR2 of nCoV396Fab through hydrophobic interactions (Figure 3e). Interestingly, all three residues (Q163, L167, and K169) of SARS- CoV- 2 N- NTD are relatively conserved in the highly pathogenic betacoronavirus N protein (Extended Data Figure 2a), which implies that nCoV396 may cross- interact with the SARS- CoV N protein or the Middle East respiratory syndrome coronavirus (MERS- CoV) N protein. Indeed, the binding affinities measured by SPR analysis demonstrate that nCoV396 interacts with the SARS- CoV N protein and the MERS- CoV N protein with a \(K_{\mathrm{D}}\) of 7.4 nM (Extended Data Figure 2b - c). + +To discover the conformational changes between the SARS- CoV- 2 N- NTD apo- state and the antibody- bound state, we next superimposed the complex structure with the N- NTD structure (PDB ID: 6M3M) \(^{14}\) . The superimposition results suggest that the CT tail of SARS- CoV- 2 N- NTD unfolds from the basic palm region upon nCoV396Fab binding (Figure 3f), which likely contributes to the allosteric regulation of the normal full- length N protein function. Additionally, nCoV396Fab binding results in a 7.4 Å movement of the \(\beta\) - finger region outward from the RNA binding pocket, which may enlarge the RNA binding pocket of the N protein (Figure 3f). + +In summary, our crystal structural data demonstrated that the human mAb nCoV396 recognizes the SARS- CoV- 2 N protein via a pinching model, resulting in a dramatic conformational change in residues 159 to 172, which is the linker region of N- NTD that is connected with other domains. + +<--- Page Split ---> + +## MAb curbs N-induced complement activation + +Although a recent study suggests that the complement cascade is hyperactive by the N protein in the lungs of COVID- 19 patients via the lectin pathway19, it is unclear how to develop a virus- free and an effective system for analyzing the role of the SARS- CoV- 2 N protein on complement hyperactivation. To this end, we developed a clinical autoimmune disease serum- based protease enzymatic approach to assess complement activation levels in the presence of the SARS- CoV- 2 N protein. Since complement activation initiated by the lectin pathway features MASP- 2 proteases by specific activity for cleaving complement components 2 and 4 (C2 and C4)32, we designed a C2 internal quenched fluorescent peptide- based analysis route for ex vivo complement hyperactivation (Figure 4a). Briefly, serum was collected from the peripheral blood of volunteers with an autoimmune disease, as their serum contains the necessary components for complement activation characterized by elevated C3 levels (Extended Data Table 6). Next, we collected the fluorescence signal from cleaved C2 synthetic peptide substrates (2Abz- SLGRKIQI- Lys(Dnp)- NH2) in reaction mixtures containing autoimmune disease serum in the absence or presence of the SARS- CoV- 2 N protein with or without the mAb nCoV396. The initial reaction rate (v0) was estimated at a single concentration of individual sera from duplicate measurements over a range of substrate concentrations. The steady- state reaction constants maximal velocity (Vmax) and Michaelis constant (Km) were determined for comparisons (Figure 4a). + +<--- Page Split ---> + +As shown in Figure 4b, the calculated \(V_{\max}\) of reactions without any other exogenous proteins is 1.49 response units (RU) \(\cdot \mathrm{s}^{- 1}\) . Additions of the SARS- CoV- + +2 N protein (concentrations ranging from 0.5 \(\mu \mathrm{M}\) to \(10 \mu \mathrm{M}\) ) in the reactions + +remarkably elevate the \(V_{\max}\) up to 2- fold, ranging from \(2.37 \sim 3.02 \mathrm{RU} \cdot \mathrm{s}^{- 1}\) . + +Similarly, additions of the SARS- CoV- 2 N protein led to an approximate 1.8- fold + +increase in the \(V_{\max} / K_{\mathrm{m}}\) values, which suggested that the specificity constant + +\((K_{\mathrm{cat}} / K_{\mathrm{m}})\) of MASP- 2 to substrates is increased in the presence of the viral N + +protein as the enzyme concentrations are equivalent among the reactions + +(Extended Data Table 7 - 8). To confirm the kinetic analyses, Hanes plots ([S]/V + +versus [S]) were also drawn and found to be linear (Figure 4c). Therefore, the + +addition of the SARS- CoV- 2 N protein does not change the single substrate + +binding site characterization of the enzymatic reactions. To assess the + +suppression ability of nCoV396 to the SARS- CoV- 2 N protein- induced + +complement hyperactivation function, we next conducted complement + +hyperactivation analyses at various N protein: nCoV396 ratios. As shown in + +Figure 4d, the addition of the N protein elevates the \(V_{\max}\) value up to 40- fold (1:0 + +ratio), whereas the addition of the antibody nCoV396 decreases the \(V_{\max}\) in a + +dose- dependent manner (Extended Data Table 9). To further validate the + +function of nCoV396, we next performed complement hyperactivation analyses in + +five other serum samples from autoimmune disease donors. Consistently, the + +\(V_{\max}\) of reactions was boosted in the presence of the N protein in all samples but + +declined in the presence of both the mAb nCoV396 and the N protein (Figure 4e). + +In conclusion, these results demonstrate that the SARS- CoV- 2 N protein is + +<--- Page Split ---> + +capable of inducing complement hyperactivation ex vivo, not only by facilitating the \(V_{\max}\) of MASP- 2 catalytic activity but also by enhancing the substrate binding specificity in the reactions. The N- reactive mAb nCoV396 specifically compromises the SARS- CoV- 2 N protein- induced complement hyperactivation within clinical serum samples. + +## Discussion + +From a quickly recovered COVID- 19 patient, we isolated 32 mAbs specifically targeting the SARS- CoV- 2 N protein. The binding affinity of mAbs ranged from 1 nM to 25 nM, which is comparable with the binding affinity of mature S protein- reactive antibodies22,25- 29 and the other mature antibodies identified during acute infections33,34. The characteristics of the isolated N- reactive mAbs are different from those of the isolated S- reactive mAbs in COVID- 19 patients during the early recovery phase, suggesting that sampling time is pivotal for identifying differential immune responses to different SARS- CoV- 2 viral proteins. + +The crystal structure of nCoV396 bound to SARS- CoV- 2 N- NTD elucidates the interaction mechanism of the complex between the first reported N protein- reactive human mAb and its targeted N protein. Three conserved amino acids (Q163, L167, and K169) in the N protein are responsible for nCoV396 recognition, which provides evidence of cross- reactivity of nCoV396 to the N protein of SARS- CoV or MERS- CoV. Intriguingly, the nCoV396 binding of SARS- CoV- 2 N- NTD undergoes several conformational changes, resulting in an enlargement of the N- NTD RNA binding pocket enlargement and partial unfolding of the basic + +<--- Page Split ---> + +palm region. More importantly, this conformational change occurs in the CT tail of the N- NTD, which may alter the positioning of individual domains in the context of the full- length protein and lead to a potential allosteric effect for protein functions. + +Complement is one of the first lines of defense in innate immunity and is essential for cellular integrity and tissue homeostasis and for modifying the adaptive immune response35. Emerging evidence suggests that the complement system plays a vital role in a subset of critical COVID- 19 patients, with features of atypical acute respiratory distress syndrome, disseminated intravascular coagulation, and multiple organ failure9,10,36. A few pieces of evidence show that the N protein of highly pathogenic coronaviruses (SARS- CoV- 2 and SARS- CoV) is involved in the initiation of MASP- 2- dependent complement activation19,21. Encouragingly, critical COVID- 19 patients treated with complement inhibitors, including small molecules to the complement component C3 and an antibody targeting the complement component C5, show remarkable therapeutic outcomes19. Currently, there are 11 clinical trials related to targeting the complement pathway (https://clinicaltrials.gov). To avoid adverse effects of human complement component- targeting therapy, a viral protein- specific approach is warranted. The antibody nCoV396 isolated from COVID- 19 convalescent patients is an excellent potential candidate with a high binding affinity to the N protein and high potency to inhibit complement hyperactivation. As revealed by atomic structural information, the binding may allosterically change the full- length N protein conformation. To determine the role of nCoV396 in the suppression of complement hyperactivation, we monitored MASP- 2 + +<--- Page Split ---> + +protease activity based on its specific fluorescence- quenched C2 substrate in sera from autoimmune disease patients. The complete complement components in the sera of patients with autoimmune disorders allow us to monitor the activating effects of the SARS- CoV- 2 N protein and its specific mAbs. Although we cannot calculate the other steady- state enzymatic reaction constants as the precise concentration of MASP- 2 in serum is unknown, we identified the \(V_{\mathrm{max}}\) of the specific C2 substrate for the enzymatic reaction. We demonstrated that the SARS- CoV- 2 N protein elevated the \(V_{\mathrm{max}}\) of the reaction, up to 40- fold, in the sera of all 7 individuals tested, while nCoV396 effectively suppressed the \(V_{\mathrm{max}}\) of the reaction mixtures. These results indicated that serum- based complement activation analysis of autoimmune disease patients is a virus- free and an effective method for examining complement activation mediated by the SARS- CoV- 2 N protein. + +Although the precise interaction of the SARS- CoV- 2 N protein with MASP- 2 remains to be elucidated, our work defined the region on the SARS- CoV- 2 N protein recognized by the mAb nCoV396 that plays an important role in complement hyperactivation and indicates that human mAbs from convalescents could be a promising potential therapeutic candidate for the treatment of COVID- 19. + +## References + +1 Epidemiology Working Group for Ncip Epidemic Response, Chinese Center for Disease Control and Prevention. [The epidemiological + +<--- Page Split ---> + +characteristics of an outbreak of 2019 novel coronavirus diseases (COVID- 19) in China]. Zhonghua Liu Xing Bing Xue Za Zhi 41, 145- 151 (2020) + +Wiersinga, W. J., Rhodes, A., Cheng, A. C., Peacock, S. J. & Prescott, H. C. Pathophysiology, Transmission, Diagnosis, and Treatment of Coronavirus Disease 2019 (COVID- 19): A Review. Jama- J. Am. Med. Assoc. doi:10.1001/jama.2020.12839 (2020). + +Tang, N., Li, D. J., Wang, X. & Sun, Z. Y. Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia. J Thromb. Haemost. 18, 844- 847 (2020). + +Wang, D. W. et al. Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus- Infected Pneumonia in Wuhan, China. Jama- J. Am. Med. Assoc. 323, 1061- 1069 (2020). + +Zhu, N. et al. A Novel Coronavirus from Patients with Pneumonia in China, 2019. N. Engl. J. Med. 382, 727- 733 (2020). + +Zhou, F. et al. Clinical course and risk factors for mortality of adult inpatients with COVID- 19 in Wuhan, China: a retrospective cohort study. Lancet 395, 1054- 1062 (2020). + +Gattinoni, L. et al. COVID- 19 Does Not Lead to a "Typical" Acute Respiratory Distress Syndrome. Am. J. Respir. Crit. Care Med 201, 1299- 1300, (2020). + +<--- Page Split ---> + +Magro, C. et al. Complement associated microvascular injury and thrombosis in the pathogenesis of severe COVID- 19 infection: A report of five cases. Transl. Res. 220, 1- 13 (2020). + +Cugno, M. et al. Complement activation in patients with COVID- 19: A novel therapeutic target. J. Allergy Clin Immunol 146, 215- 217, (2020). + +Noris, M., Benigni, A. & Remuzzi, G. The case of complement activation in COVID- 19 multiorgan impact. Kidney Int. 98, 314- 322 (2020). + +Pang, R. T. et al. Serum proteomic fingerprints of adult patients with severe acute respiratory syndrome. Clin. Chem. 52, 421- 429, (2006). + +Chen, J. H. et al. Plasma proteome of severe acute respiratory syndrome analyzed by two- dimensional gel electrophoresis and mass spectrometry. Proc. Natl. Acad. Sci. U. S. A. 101, 17039- 17044 (2004). + +Ramlall, V. et al. Immune complement and coagulation dysfunction in adverse outcomes of SARS- CoV- 2 infection. Nat. Med. (2020). + +Kang, S. et al. Crystal structure of SARS- CoV- 2 nucleocapsid protein RNA binding domain reveals potential unique drug targeting sites. Acta Pharm. Sin. B, doi:10.1016/j.apsb.2020.04.009 (2020). + +Ye, Q., West, A. M. V., Silletti, S. & Corbett, K. D. Architecture and self- assembly of the SARS- CoV- 2 nucleocapsid protein. Protein Sci., doi:10.1002/pro.3909 (2020). + +Iserman, C. et al. Specific viral RNA drives the SARS CoV- 2 nucleocapsid to phase separate. bioRxiv, doi:10.1101/2020.06.11.147199 (2020). + +<--- Page Split ---> + +17 Li, J. Y. et al. The ORF6, ORF8 and nucleocapsid proteins of SARS-CoV- 2 inhibit type I interferon signaling pathway. Virus Res. 286, 198074, (2020). 18 Perdikari, T. M. et al. SARS-CoV-2 nucleocapsid protein undergoes liquid- liquid phase separation stimulated by RNA and partitions into phases of human ribonucleoproteins. bioRxiv, doi:10.1101/2020.06.09.141101 (2020). 19 Gao, T. et al. Highly pathogenic coronavirus N protein aggravates lung injury by MASP- 2- mediated complement over- activation. medRxiv, 2020.2003.2029.20041962, doi:10.1101/2020.03.29.20041962 (2020). 20 Guo, Y. R. et al. The origin, transmission and clinical therapies on coronavirus disease 2019 (COVID- 19) outbreak - an update on the status. Military Med. Res. 7, 11 (2020). 21 Liu, J. L., Cao, C. & Ma, Q. J. Study on interaction between SARS- CoV N and MAP19. Xi bao yu fen zi mian yi xue za zhi = Chinese journal of cellular and molecular immunology 25, 777- 779 (2009). 22 Chi, X. et al. A neutralizing human antibody binds to the N- terminal domain of the Spike protein of SARS- CoV- 2. Science 369, 650- 655 (2020). 23 Klasse, P. J. & Moore, J. P. Antibodies to SARS- CoV- 2 and their potential for therapeutic passive immunization. Elife 9, e57877 (2020). 24 Kreer, C. et al. Longitudinal Isolation of Potent Near- Germline SARS- CoV- 2- Neutralizing Antibodies from COVID- 19 Patients. Cell S0092- 8674(20), 30821- 30827 (2020). + +<--- Page Split ---> + +Zost, S. J. et al. Potently neutralizing and protective human antibodies against SARS- CoV- 2. Nature, doi:10.1038/s41586- 020- 2548- 6 (2020). + +Wu, Y. et al. A noncompeting pair of human neutralizing antibodies block COVID- 19 virus binding to its receptor ACE2. Science 368, 1274- 1278, (2020). + +Wang, C. et al. A human monoclonal antibody blocking SARS- CoV- 2 infection. Nat. Commun. 11, 2251 (2020). + +Ju, B. et al. Human neutralizing antibodies elicited by SARS- CoV- 2 infection. Nature 584, 115- 119 (2020). + +Cao, Y. et al. Potent Neutralizing Antibodies against SARS- CoV- 2 Identified by High- Throughput Single- Cell Sequencing of Convalescent Patients' B Cells. Cell 182, 73- 84 (2020). + +Liao, H. X. et al. High- throughput isolation of immunoglobulin genes from single human B cells and expression as monoclonal antibodies. J. Virol. Methods 158, 171- 179 (2009). + +Moody, M. A. et al. H3N2 influenza infection elicits more cross- reactive and less clonally expanded anti- hemagglutinin antibodies than influenza vaccination. PLoS One 6 (10): e25797 (2011). + +Duncan, R. C. et al. Multiple domains of MASP- 2, an initiating complement protease, are required for interaction with its substrate C4. Mol. Immunol. 49, 593- 600 (2012). + +Stettler, K. et al. Specificity, cross- reactivity, and function of antibodies elicited by Zika virus infection. Science 353, 823- 826 (2016). + +<--- Page Split ---> + +34 Yu, L. et al. Delineating antibody recognition against Zika virus during natural infection. JCI Insight 2 (12):e93042 (2017). 35 Zipfel, P. F. & Skerka, C. Complement regulators and inhibitory proteins. Nat. Rev. Immunol. 9, 729-740 (2009). 36 Lo, M. W., Kemper, C. & Woodruff, T. M. COVID-19: Complement, Coagulation, and Collateral Damage. J. Immunol, doi: 10.4049/jimmunol.2000644 (2020). + +<--- Page Split ---> + +## Methods + +## Recombinant SARS-CoV-2 S-ECD and N proteins. + +Recombinant SARS- CoV- 2 S protein (extracellular domain of the S protein (ECD) with His and FLAG Tags, Z03481) was purchased from GenScript. SARS- CoV- 2 (formerly known as 2019- nCoV, recombinant full- length N protein with a CT 6x His tag (His tag, 40588- V08B) was purchased from Sino Biological. The SARS- CoV- 2 N protein expression plasmid (SARS- CoV N- FL) was a gift from the Guangdong Medical Laboratory Animal Center. SARS- CoV and MERS- CoV N- FL were purchased from RuiBiotech. SARS- CoV- 2 N- FL (residues 1 to 419), SARS- CoV- 2 N- NTD domain (residues 41 to 174), SARS- CoV- 2 N- CTD domain (residues 250 to 364), SARS- CoV N- FL and MERS- CoV N- FL were cloned into the pET- 28a vector and expressed in the Rosetta E. coli strain. Expression of SARS- CoV- 2 N- FL and variants in E. coli was induced with 0.1 mM isopropylthiob- galactoside (IPTG) and cultured overnight at 16 °C in Terrific Broth media. Expressed recombinant N proteins were initially purified by using nickel column chromatography and further purified via size- exclusion chromatography. + +## PBMCs from COVID-19 patients and sorting of single plasma cells and memory B cells by FACS + +Blood samples were collected 9 - 25 days after the onset of the disease from patients who had recovered from COVID- 19 infection. PBMCs and plasma were isolated from blood samples by Ficoll- Paque PLUS (GE, 17- 1440- 02) density gradient centrifugation. All work related to human subjects was carried out in + +<--- Page Split ---> + +compliance with Institutional Review Board protocols approved by the Institutional Review Board of the Fifth Affiliated Hospital of Sun Yat- sen University. Single plasma cells with the surface markers CD3+, CD14+, CD16+, CD235a+, CD19+, CD20low- neg, CD27hi and CD38hi and memory B cells with the surface markers CD3+, CD14+, CD16+, CD235a+, CD20+, CD19+, CD27+, SARS- CoV- 2 S+ and SARS- CoV- 2 N+ (BD Biosciences and Invitrogen) were sorted into individual wells in 96- well microtiter plates containing cell lysis and RT buffer for Ig gene amplification by fluorescence activated cell sorting (FACS) as previously described37 on a BD FACS Aria SORP. Data were analyzed using BD FACS Diva 8.0.1 software. + +## Isolation and expression of Ig \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes + +Genes encoding \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) were amplified by reverse transcription (RT) and nested primer chain reaction (PCR) and nested PCR using the method previously described38. PCR products of Ig \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes were purified using a PCR purification kit (QIAGEN), sequenced in forward and reverse directions (Thermo Fisher scientific) and annotated by using IMGT/V- QUEST (www.imgt.org/IMGT_vquest). Functional \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes were used for assembling full- length Ig heavy and light chain linear expression cassettes by one- step overlapping PCR38. HEK- 293T cells in 12- well plates were transfected with the assembled Ig heavy and light chain pairs derived from the same single individual plasma cells using Effectene (QIAGEN) as the transfection reagent38. + +## Production of recombinant IgG and Fab antibodies + +<--- Page Split ---> + +For the production of purified full- length IgG1 antibodies, the \(\mathrm{V_H}\) and \(\mathrm{V_L}\) genes were cloned into the pCDNA3.1+ (Invitrogen)mammalian expression vector containing either the human IgG1 constant region gene, the human kappa light chain constant region gene or the lambda light chain constant region gene using standard recombinant DNA technology38. For the production of the purified nCoV396Fab antibody, a stop codon TGA was introduced after the sequence (5'- TCTTGTGACAAA- 3'), which encodes the amino acid residues SCDK, just before the hinge of the human IgG1 constant region39. Recombinant IgG1 antibodies and the nCoV396Fab antibody were produced in 293F cells cultured in serum- free medium by cotransfection with the generated IgG1 full- length or Fab heavy- and light chain gene expression plasmid pairs using polyethyleneimine40. Full- length IgG1 antibodies were purified by using Protein A column chromatography as described previously38. The nCoV396Fab antibody used for the crystal structure was purified by Lambda FabSelect, an affinity resin designed for the purification of human Fab with a lambda light chain (GE Healthcare)39. + +## Analysis of the binding of plasma antibodies and isolated mAbs to the S and N proteins by ELISA + +We collected plasma from 6 patients and measured serum antibody titers using recombinant SARS- CoV- 2 S and N proteins as antigens to coat ELISA plates. Antibodies in the supernatant of the transfected 293T cultures harvested 3 days after transfection were screened by ELISA as described previously38. The binding of purified antibodies to the N or S protein was also analyzed by ELISA. Briefly, all protein antigens (S, N- FL, N- NTD (41- 174), and N- CTD (250- 364)) were used + +<--- Page Split ---> + +at 200 ng/well to coat 96- well high- binding ELISA plates (Nunc 442404) using carbonate- bicarbonate buffer at pH 9.6. Plates were incubated overnight at 4 °C and blocked at room temperature for 2 h with PBS blocking buffer containing 5% w/v goat serum and 0.05% Tween- 20. Plasma or supernatant of transfected 293T cell cultures or purified mAbs in serial dilutions in PBS were incubated at 37 °C for 1 h. Goat anti- human IgG- horseradish peroxidase (HRP, 1:10,000 dilution) (Promega, W4031) as the secondary antibody diluted in blocking buffer was added and incubated at 37 °C for 1 h. These plates were then washed 5 times with PBS and developed with 100 μL of 3,3',5,5'- Tetramethylbenzidine (TMB) substrate/well (Solarbio PR1200). The reaction of the plates was stopped with 50 μL of 2 M H₂SO₄/well and read at a wavelength of 450 nm by an ELISA reader. The relative affinity of mAbs to the N protein antigen was determined as the effective concentration (EC₅₀) of the antibody resulting in half maximal binding to the antigen by curve fitting with GraphPad Prism software. + +## Affinity and kinetic measurements by SPR + +The binding affinity (K₀), association rate (kₐ) and dissociation rate (kₐ) of purified mAbs to the N protein were determined by SPR using a Biacore X100 System. Anti- human Fc IgG antibody was first immobilized on a CM5- chip to approximately 6,000 RU by covalent amine coupling using a human antibody capture kit (GE Healthcare). Purified mAbs were captured on channel 2 of the CM5 chip to approximately 200 RU. Five 2- fold serial dilutions of the N protein starting at 40 μg/ml, 20 μg/ml, 10 μg/ml or 4 μg/ml were injected at a rate of 30 μL/min for 90 s with a 600 s dissociation. The chip was regenerated by injection + +<--- Page Split ---> + +of 3 M MgCl₂ for 30 s. All experiments were performed at room temperature, and data were analyzed using Biacore X100 Evaluation Software (version: 2.0.1). Curves were fitted to a 1:1 binding model to determine the kinetic rate constants (kₐ and kₐ). K₀ values were calculated from these rate constants. + +## Crystallization and data collection + +SARS- CoV- 2 N- NTD (41–174) was cloned into the pRSF- Duet- 1 vector with an NT 6x His- SUMO tag, recombinantly expressed in E. coli and purified as an NT 6x His- Sumo- tagged protein. After Ni column chromatography followed by ulp1 digestion for tag removal, the SARS- CoV- 2 N- NTD (41–174) protein was further purified via size- exclusion chromatography. Prior to crystallization, the SARS- CoV- 2 N- NTD (41–174) sample was mixed with nCoV396Fab at a 1:1.5 molar ratio for approximately half an hour and then further purified via size- exclusion chromatography. Crystals were grown by the sitting drop method using a Mosquito LCP crystallization robot with 0.3 μL of protein (6 mg/ml) mixed with 0.3 μL of well solution at 16 °C. Better crystals were obtained in 0.01 M calcium chloride dihydrate, 0.05 M sodium cacodylate trihydrate (pH 7.2), 1.675 M ammonium sulfate and 0.5 mM spermine. The crystals were harvested after 3 days. Crystals were frozen in liquid nitrogen in the reservoir solution supplemented with 25% glycerol (v/v) as a cryoprotectant. X- ray diffraction data were collected at the Shanghai Synchrotron Radiation Facility BL18U at a wavelength of 0.979 Å and a temperature of 100 K. The complex structure of SARS- CoV- 2 N- NTD with the mAb nCoV396 was determined by Phenix molecular replacement using the SARS- CoV- 2 N- NTD structure (PDB ID: 6M3M) + +<--- Page Split ---> + +and the monoclonal antibody omalizumab Fab (PDB ID: 6TCN) as the search models. The X- ray diffraction and structure refinement statistics are summarized in Extended Data Table 4. The final Ramachandran statistics are 97.1% favored, 2.9% allowed and 0.0% outliers. + +## Fluorescence-quenched substrate assays + +The fluorescence- quenched substrate (FQS) [C2 P4- P4' (2Abz- SLGRKIQL- Lys(Dnp)- NH2)] was synthesized and purified in greater than 90% purity by Sangon Biotech (Shanghai, China). The FQS was solubilized in 50% (v/v) dimethylformamide (DMF). Assays were carried out in fluorescence assay buffer (FAB) (0.05 M Tris- HCl, 0.15 M NaCl, 0.2% (w/v) PEG 8,000, and pH 7.4) at 37 °C using final substrate concentrations in the range of 2.8125- 90 μM. A range of substrate dilutions, performed in triplicate, were prepared in FAB and were added to the wells at 100 μl/well in the assay plates. The plates were incubated at 37 °C for 10 minutes. Five microliters of serum from autoimmune patients was added to the mixture of the SARS- CoV- 2 N protein with nCoV396 at a 1:1.2 molar ratio in the assay plates. Fluorescence intensity was measured on an EnVision 2015 Multimode plate reader (PerkinElmer) at an excitation wavelength of 320 nm and an emission wavelength of 420 nm for the FQS. The initial reaction rate was estimated at a single concentration of enzyme from duplicate measurements over a range of substrate concentrations. To determine steady- state reaction constants ( \(V_{\text{max}}\) , half saturation constant ( \(K_{0.5}\) ) and Hill coefficient(h)), the experimental results were fitted using GraphPad Prism Version 8.0 (GraphPad Software, San Diego, CA) to an equation describing + +<--- Page Split ---> + +574 positive cooperativity: \(Y = V_{max} \times \frac{x^n}{K \circ \delta^b + x^n}\) , where 0.5 defines the relationship between the reaction rate (V) and the substrate concentration ([S]) when more than one binding site applies. + +## Methods references + +578 37. Morris, L. et al. Isolation of a human anti- HIV gp41 membrane proximal region neutralizing antibody by antigen- specific single B cell sorting. PloS one 6, e23532 (2011). + +581 38. Liao, H. X. et al. High-throughput isolation of immunoglobulin genes from single human B cells and expression as monoclonal antibodies. J. Virol. Methods 158, 171- 179 (2009). + +584 39. Nicely, N. I. et al. Crystal structure of a non- neutralizing antibody to the HIV- 1 gp41 membrane- proximal external region. Nat. Struct. Mol. Biol. 17, 1492- 1494 (2010). + +587 40. Smith, K. et al. Rapid generation of fully human monoclonal antibodies specific to a vaccinating antigen. Nat. Protoc. 4, 372- 384 (2009). + +<--- Page Split ---> + +## Acknowledgement + +This work is supported by the COVID- 19 Emerging Prevention Products, Research Special Fund of Zhuhai City (ZH22036302200016PWC to S.C.; ZH22036302200028PWC to F. X.; ZH22046301200011PWC to H- X. L.); Emergency Fund from Key Realm R&D Program of Guangdong Province (2020B111113001) to H.S.; Zhuhai Innovative and Entrepreneurial Research Team Program (ZH01110405160015PWC, ZH01110405180040PWC) to H- X. L; We thank the staffs of the BL18U/19U/17U beamlines at SSRF for their help with the X- ray diffraction data screening and collections. We thank Junlang Liang, Tong Liu, Nan Li, Xiaoli Wang, Zhenxing Jia, and Jiaqi Li from Zhuhai Trinomab Biotechnology Co., Ltd. for technical assistants of mAbs isolation, production and characterization. + +## Author Contributions + +S. C., H. S., F. X. and H-X. L. contributed the conception of the study and established the construction of the article. S. C. and H-X. L. designed the experiments and wrote the manuscript. S. K., M. Y., S. H. contributed to protein purification and crystallization, in vitro protein-protein interaction analysis, and complement activation analysis. Y. W. contributed to mAbs isolation, in vitro protein-protein interaction analysis. S. C., S. K. M. Y., and S. H. performed structural determination and validation. S. C., S. K., Y. W. drew figures. X. C., Y. C., Q. C., Z. Z., Z. Z., Z. H., X. H., H. S., W. Z., and H. H. contributed to + +<--- Page Split ---> + +610 interpretation of data. Z. H., J. L., G. J., and F. X. contributed to clinical samples collections. S.K., M.Y., S. H., Y.W. contributed equally to this work. + +## Conflict of Interest + +The authors declare no conflict of interest. + +## Data availability statement + +The structure in this paper is deposited to the Protein Data Bank with 7CR5 access code. + +## Additional Information + +Correspondence and requests for materials should be addressed to Shoudeng Chen (chenshd5@mail.sysu.edu.cn) and/or Hua- Xin Liao (tliao805@jnu.edu.cn). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +Acquisition and characterization of antibodies. Serum antibody titers of six SARS- CoV- 2 convalescent patients to the SARS- CoV- 2 S (a) and N (b) proteins measured by ELISA. Sorting of single plasma cells (c) with CD38 and CD27 double- positive B cells and single N and S protein specific memory B cells (d) by + +<--- Page Split ---> + +FACS. (e) Percentage of different isotypes, VH and VL gene families of 32 isolated N-reactive antibodies. (f) Number of mutations in nucleotides and amino acids in VH and VL (Vk and Vλ) of 32 N-reactive antibodies and eight S-reactive antibodies (g). Length of the 32 Nreactive antibodies (h) and eight S-reactive antibodies (i) in H-CDR3. + +![](images/Figure_2.jpg) + + + +
Figure 2
+ +<--- Page Split ---> + +Reactivity and affinity of the isolated antibodies to the N protein antigens. (a) Schematic presentation of the SARS-CoV-2 N protein and two variants. (b) Antibodies expressed in transfected 293 cells were evaluated for binding to N-FL, N-NTD and N-CTD by ELISA. Plasma from patient ZD006 and an irrelevant mAb TRN006 were used as the positive control and negative control, respectively. (c) The ability of nine purified antibodies to the N-FL protein was determined by ELISA. (d) Binding affinity of nine selected antibodies to the N protein was measured by SPR. KD values are shown above the individual plots. + +![](images/Figure_3.jpg) + + + +
Figure 3
+ +<--- Page Split ---> + +Complex structure of mAb nCoV396 with SARS-CoV-2 N-NTD(a) Overall structure of the mAb SARS-CoV-2 N-NTD complex. The light chain (pink) and heavy chain (blue) of mAb nCoV396 are illustrated with the ribbon representation. SARS-CoV-2 N-NTD is illustrated with an electrostatic surface, in which blue denotes a positive charge potential, while red indicates a negative charge potential. (b) The N-NDD epitope recognized by mAb nCoV396. The interacting residues of N-NTD and nCoV396 are highlighted with the strict representation. Recognition of Q163 (c), K169 (d) and L167 (e) in N-NTD by mAb nCoV396. The dashed blue line represents hydrogen bonds. Hydrophobic interactions are illustrated with the dot representation. (f) Conformational changes in N-NTD upon mAb nCoV396 binding. The apo structure of N-NTD is colored gray. Antibody-bound N-NDD is colored green. The N-terminus and C-terminus of the N-NTD are labeled with circles. mAb nCoV396 is illustrated with surface representation. All figures were prepared by PyMol. + +![](images/Figure_4.jpg) + + + +
Figure 4
+ +<--- Page Split ---> + +Antibody nCoV396 compromises SARS- CoV- 2 N protein- induced complement hyperactivation. (a) Flow scheme of the SARS- CoV- 2 N protein and nCoV396 influencing the protease activity of MASP- 2 in the serum of autoimmune disease patients. The Michaelis- Menten curve shows the effect of increasing the N protein concentration (b) and antibody concentration (d) on the substrate C2 cleavage of MASP- 2 in the serum of patient 49 and patient 20. (c) A Hanes plot where C2 concentrationN0 is plotted against C2 concentration with the addition of 5 μM N protein. (e) The mAb nCoV396 inhibits the N protein- induced excessive cleavage of C2 in the serum of six autoimmune disease patients, and the last panel shows a summary of Vmax for all patients. Negative control (Negative Ctrl) and blank control (Blank Ctrl) represent reactions containing bovine serum albumin (BSA) instead of N or N+mAb and without exogenous protein, respectively. The mean and standard deviation (SD) values of three technical replicates are shown. P values: \(*P < 0.05\) ; \(**P < 0.01\) ; "-" indicates that the experimental kinetics did not conform to Michaelis- Menten kinetics. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- 5ExTFigTablesNC.pdf- 4PDBNC.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2_det.mmd b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..52a88a9b1f94b2aa9731233b93fc9f579cb92444 --- /dev/null +++ b/preprint/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2/preprint__131d794d0a449bac37a46f99668131f02dc2ee676420c8e424ccf014742270b2_det.mmd @@ -0,0 +1,553 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 796, 209]]<|/det|> +# A COVID-19 antibody curbs SARS-CoV-2 nucleocapsid protein-induced complement hyperactivation + +<|ref|>text<|/ref|><|det|>[[44, 230, 210, 270]]<|/det|> +Sisi Kang Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 277, 519, 317]]<|/det|> +Mei Yang The Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 324, 520, 365]]<|/det|> +Suhua He The Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 370, 560, 410]]<|/det|> +Yueming Wang Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +<|ref|>text<|/ref|><|det|>[[44, 415, 930, 480]]<|/det|> +Xiaoxue Chen Department of Experimental Medicine, Guangdong Provincial Key Laboratory of Biomedical Imaging, The Fifth affiliated Hospital, Sun Yat- sen University, Zhuhai, 519000 + +<|ref|>text<|/ref|><|det|>[[44, 485, 250, 526]]<|/det|> +Yao- Qing chen Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 532, 480, 572]]<|/det|> +Zhongsi Hong Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 578, 520, 619]]<|/det|> +Jing Liu The Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 624, 480, 665]]<|/det|> +Guanmin Jiang Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 671, 250, 711]]<|/det|> +Qiuyue Chen Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 717, 610, 758]]<|/det|> +Ziliang Zhou Sun Yat- sen University https://orcid.org/0000- 0002- 6801- 4180 + +<|ref|>text<|/ref|><|det|>[[44, 763, 250, 803]]<|/det|> +Zhechong Zhou Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 810, 250, 849]]<|/det|> +Zhaoxia Huang Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 856, 480, 896]]<|/det|> +Xi Huang Fifth Affiliated Hospital of Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[44, 902, 700, 943]]<|/det|> +Huanhuan He Guangdong Provincial Engineering Research Center of Molecular Imaging + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 42, 559, 87]]<|/det|> +Weihong Zheng Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +<|ref|>text<|/ref|><|det|>[[42, 91, 559, 133]]<|/det|> +Hua- Xin Liao Jinan University; Zhuhai Trinomab Biotechnology Co., Ltd. + +<|ref|>text<|/ref|><|det|>[[42, 137, 250, 178]]<|/det|> +Fei Xiao Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[42, 183, 250, 224]]<|/det|> +Hong Shan Sun Yat- sen University + +<|ref|>text<|/ref|><|det|>[[42, 228, 608, 270]]<|/det|> +Shoudeng Chen ( chenshd5@mail.sysu.edu.cn ) Sun Yat- sen University https://orcid.org/0000- 0002- 7634- 2141 + +<|ref|>sub_title<|/ref|><|det|>[[42, 311, 102, 329]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 348, 936, 393]]<|/det|> +Keywords: human monoclonal antibody, COVID- 19, SARS- CoV- 2, nucleocapsid protein, crystal structure, complement hyperactivation, viral protein targeting therapy, MASP- 2 + +<|ref|>text<|/ref|><|det|>[[44, 409, 340, 429]]<|/det|> +Posted Date: December 2nd, 2020 + +<|ref|>text<|/ref|><|det|>[[42, 447, 463, 467]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 106760/v1 + +<|ref|>text<|/ref|><|det|>[[42, 484, 910, 528]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 563, 909, 607]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 11th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 23036- 9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[104, 104, 857, 130]]<|/det|> +# A SARS-CoV-2 antibody curbs viral N protein-induced + +<|ref|>title<|/ref|><|det|>[[102, 152, 501, 177]]<|/det|> +# complement hyperactivation + +<|ref|>text<|/ref|><|det|>[[101, 207, 860, 880]]<|/det|> +3 Sisi Kang1\*, Mei Yang1\*, Suhua He1\*, Yueming Wang2,3\*, Xiaoxue Chen1, Yao- 4 Qing Chen4, Zhongsi Hong5, Jing Liu6, Guanmin Jiang7, Qiuyue Chen1, Ziliang 5 Zhou1, Zhechong Zhou1, Zhaoxia Huang1, Xi Huang8, Huanhuan He1, Weihong 6 Zheng2,3, Hua-Xin Liao2,3,#, Fei Xiao1,5,#, Hong Shan1,9,#, Shoudeng Chen1,10,# 7 1. Molecular Imaging Center, Guangdong Provincial Key Laboratory of 8 Biomedical Imaging, The Fifth Affiliated Hospital, Sun Yat- sen University, 9 Zhuhai, 519000, China 10 2. Institute of Biomedicine, Jinan University, Guangzhou, 510632, China 11 3. Zhuhai Trinomab Biotechnology Co., Ltd., Zhuhai, 519040, China 12 4. School of Public Health (Shenzhen), Sun Yat- sen University, Shenzhen, 13 China 14 5. Department of Infectious Disease, The Fifth Affiliated Hospital, Sun Yat- sen 15 University, Zhuhai, 519000, China 16 6. Department of Respiratory Disease, The Fifth Affiliated Hospital, Sun Yat- sen 17 University, Zhuhai, 519000, China 18 7. Department of Clinical laboratory, The Fifth Affiliated Hospital of Sun Yat- sen 19 University, Zhuhai, 519000, China 20 8. Center for Infection and Immunity, The Fifth Affiliated Hospital, Sun Yat- sen 21 University, Zhuhai, 519000, China + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[95, 87, 859, 355]]<|/det|> +22 9. Department of Intervention Medicine, The Fifth Affiliated Hospital, Sun Yatsen University, Zhuhai, 519000, China +24 10. Department of Experimental Medicine, The Fifth Affiliated Hospital, Sun Yatsen University, Zhuhai, 51 9000, China +26 \* These authors contributed equally to this work +27 # Co-correspondence: Shoudeng Chen (chenshd5@mail.sysu.edu.cn); Hong Shan (shanhong@mail.sysu.edu.cn); Fei Xiao (xiaof35@mail.sysu.edu.cn); Hua-Xin Liao (tliao805@jnu.edu.cn) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 105, 266, 130]]<|/det|> +## Summary + +<|ref|>text<|/ref|><|det|>[[140, 163, 860, 675]]<|/det|> +Although human antibodies elicited by the severe acute respiratory syndrome coronavirus 2 (SARS- CoV- 2) nucleocapsid (N) protein are profoundly boosted upon infection, little is known about the function of N- reactive antibodies. Herein, we isolated and profiled a panel of 32 N protein- specific monoclonal antibodies (mAbs) from a quick recovery coronavirus disease- 19 (COVID- 19) convalescent patient who had dominant antibody responses to the SARS- CoV- 2 N protein rather than to the SARS- CoV- 2 spike (S) protein. The complex structure of the N protein RNA binding domain with the mAb with the highest binding affinity (nCoV396) revealed changes in the epitopes and antigen's allosteric regulation. Functionally, a virus- free complement hyperactivation analysis demonstrated that nCoV396 specifically compromises the N protein- induced complement hyperactivation, which is a risk factor for the morbidity and mortality of COVID- 19 patients, thus laying the foundation for the identification of functional anti- N protein mAbs. + +<|ref|>text<|/ref|><|det|>[[142, 703, 858, 793]]<|/det|> +Keywords: human monoclonal antibody, COVID- 19, SARS- CoV- 2, nucleocapsid protein, crystal structure, complement hyperactivation, viral protein targeting therapy, MASP- 2 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 106, 207, 128]]<|/det|> +## Main + +<|ref|>text<|/ref|><|det|>[[140, 160, 860, 678]]<|/det|> +The fatality rate of critical condition coronavirus disease 2019 (COVID- 19) patients is exceptionally high (40% - 49%)1,2. Acute respiratory failure and generalized coagulopathy are significant aspects associated with morbidity and mortality3- 5. A subset of severe COVID- 19 patients has distinct clinical features compared to classic acute respiratory distress syndrome (ARDS), with delayed onset of respiratory distress6 and relatively well- preserved lung mechanics despite the severity of hypoxemia7. It has been reported that complement- mediated thrombotic microvascular injury in the lung may contribute to atypical ARDS features of COVID- 19, accompanied by extensive deposition of the alternative pathway (AP) and lectin pathway (LP) complement components8. Indeed, complement activation is found in multiple organs of severe COVID- 19 patients in several other studies9,10, as well as in patients with severe acute respiratory syndrome (SARS)11,12. A recent retrospective observational study of 11,116 patients revealed that complement disorder was associated with the morbidity and mortality of COVID- 1913. + +<|ref|>text<|/ref|><|det|>[[141, 703, 860, 899]]<|/det|> +The nucleocapsid (N) protein of severe acute respiratory syndrome coronavirus (SARS- CoV- 2), the etiology agent of COVID- 19, is one of the most abundant viral structural proteins with multiple functions inside the viral particles, the host cellular environment, and in ex vivo experiments14- 20. Among these functions, a recent preprint study found that the SARS- CoV- 2 N protein bound to MBL (mannan- binding lectin)- associated serine protease 2 (MASP- 2) and resulted in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 857, 180]]<|/det|> +complement hyperactivation and aggravated inflammatory lung injury19. Consistently, the highly pathogenic SARS- CoV N protein was also found to bind with MAP19, an alternative product of MASP- 221. + +<|ref|>text<|/ref|><|det|>[[140, 206, 860, 684]]<|/det|> +Although systemic activation of complement plays a pivotal role in protective immunity against pathogens, hyperactivation of complement may lead to collateral tissue injury. Thus, how to precisely regulate virus- induced dysfunctional complement activation in COVID- 19 patients remains to be elucidated. The SARS- CoV- 2 N protein is a highly immunopathogenic viral protein that elicits high titers of binding antibodies in humoral immune responses22- 24. Several studies have reported the isolation of human monoclonal antibodies (mAbs) targeting the SARS- CoV- 2 spike (S) protein, helping explain the possible developing therapeutic interventions for COVID- 1922,25- 29. However, little is known about the potential therapeutic applications of N protein- targeting mAbs in the convalescent B cell repertoire. Herein, we report a human mAb derived from the COVID- 19 convalescent patient that specifically targets the SARS- CoV- 2 N protein and functionally compromises complement hyperactivation ex vivo. + +<|ref|>sub_title<|/ref|><|det|>[[142, 712, 483, 731]]<|/det|> +## Isolation of N protein-reactive mAbs + +<|ref|>text<|/ref|><|det|>[[141, 760, 860, 886]]<|/det|> +To profile the antibody response to the SARS- CoV- 2 N protein in patients during the early recovery phase, we collected blood samples from six convalescent patients seven to 25 days after the onset of the disease symptoms. All patients recovered from COVID- 19 during the outbreak in Zhuhai, Guangdong Province, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[137, 87, 860, 670]]<|/det|> +China, with ages ranging from 23 to 66 years (Extended Data Table 1). Our work and use of patients' samples is in accordance with the declaration of Helsinki, medical ethics standards and China's laws. Our study was approved by the Ethics Committee of The Fifth Affiliated Hospital, Sun Yat- sen University, and all patients signed informed consent forms. SARS- CoV- 2 nasal swabs reverse transcription polymerase chain reaction (RT- PCR) tests were used to confirm that all 6 COVID- 19 patients were negative for SARS- CoV- 2 at the time of blood collection. Plasma samples and peripheral blood mononuclear cells (PBMCs) were isolated for serological analysis and antibody isolation. Serum antibody titers to SARS- CoV- 2 S and N proteins were measured by enzyme- linked immunosorbent assay (ELISA) (Figure 1a, 1b, and Extended Data Table 1). Serologic analysis demonstrated that serum antibody titers to the N protein were substantially higher than those to the S protein in most of the patients. For example, ZD004 and ZD006 had only minimal levels of antibody response to the S protein, while they had much higher antibody titers to the N protein. Notably, the time from disease onset to complete recovery from clinical symptoms of COVID- 19 patient ZD006 was only 9 days (Extended Data Table 1). + +<|ref|>text<|/ref|><|det|>[[140, 695, 859, 890]]<|/det|> +To take advantage of patient ZD006, who was still in the early recovery phase with a high possibility of a high percentage of antigen- specific plasma cells, single plasma cells (Figure 1c) with the phenotype of CD3\*/CD14\*/CD16\*/CD235a\*/CD19\*/CD20low- neg\*/CD27hi\*/CD38hi as well as antigen- specific memory B cells with the phenotype of CD19\*/CD27\* (Figure 1d) were sorted from PBMCs of patient ZD006 by fluorescence- activated cell sorting (FACS). To ensure an + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 87, 860, 707]]<|/det|> +unbiased assessment, the sorting of antigen- specific memory B cells was carried out with combined probes for both fluorophore- labeled S and N recombinant proteins. Variable regions of immunoglobulin (Ig) heavy and light chain gene segment ( \(\mathrm{V}_{\mathrm{H}}\) and \(\mathrm{V}_{\mathrm{L}}\) , respectively) pairs from the sorted single cells were amplified by RT- PCR, sequenced, annotated and expressed as recombinant mAbs using the methods described previously30. Recombinant mAbs were screened against SARS- CoV- 2 S and N proteins. In total, we identified 32 mAbs that reacted with the SARS- CoV- 2 N protein, including 20 mAbs from plasma cells and 12 mAbs from memory B cells (Extended Data Table 2). We found that IgG1 is the predominant isotype at \(46.9\%\) , followed by IgG3 ( \(25.0\%\) ), IgA ( \(18.8\%\) ), IgG2 ( \(6.3\%\) ) and IgM (3.1) (Figure 1e). \(\mathrm{V}_{\mathrm{H}}\) gene family usage in SARS- CoV- 2 N protein- reactive antibodies was \(18.8\%\) for \(\mathrm{V}_{\mathrm{H}1}\) , \(62.5\%\) for \(\mathrm{V}_{\mathrm{H}3}\) , \(9.4\%\) for \(\mathrm{V}_{\mathrm{H}4}\) , \(6.2\%\) for \(\mathrm{V}_{\mathrm{H}5}\) and \(3.1\%\) for \(\mathrm{V}_{\mathrm{H}7}\) , respectively (Figure 1f), which was similar to the distribution of \(\mathrm{V}_{\mathrm{H}}\) families collected in the NCBI database. Nine of 32 SARS- CoV- 2 N protein- reactive antibodies had no mutations from their germline \(\mathrm{V}_{\mathrm{H}}\) and \(\mathrm{V}_{\mathrm{H}}\) gene segments (Figure 1f, Extended Data Table 2). The average mutation frequency of the remaining mutated antibodies was \(5.3\%\) (+/- 3.6%) in \(\mathrm{V}_{\mathrm{H}}\) and \(3.5\%\) (+/- 2.7%) in \(\mathrm{V}_{\mathrm{L}}\) . + +<|ref|>text<|/ref|><|det|>[[141, 730, 859, 891]]<|/det|> +Consistent with the lower serum antibody titers to the SARS- CoV- 2 S protein, we identified only eight SARS- CoV- 2 S protein- reactive mAbs, including 5 antibodies from plasma cells and three antibodies from memory B cells. The \(\mathrm{V}_{\mathrm{H}}\) gene segment of the S protein- reactive antibodies had either no (6/8) or minimal (1/300) mutations (Figure 1g). There were no significant differences in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 88, 857, 144]]<|/det|> +complementarity- determining region 3 (CDR3) length in amino acid residues between the N- (Figure 1h) and S- reactive antibodies (Figure 1i). + +<|ref|>text<|/ref|><|det|>[[140, 171, 860, 580]]<|/det|> +Approximately a quarter of antibodies directed to the N protein (Figure 1f) and almost all of antibodies to the S protein that had no or minimal mutations from their germlines (Figure 1g) had a primary antibody response similar to other typical primary viral infections. However, relatively high \(V_H\) mutation frequencies (mean of \(5.7\%\) ) of the majority of antibodies to the N proteins were more similar to the mutation frequencies of antibodies from the secondary responses to the influenza vaccination reported previously31. Although patient ZD006 was hospitalized for only nine days after the onset of COVID- 19 symptoms, the patient had high serum antibody titers, and the majority of the isolated N- reactive antibodies had a high mutation frequency, whereas the S- reactive antibodies had no or minimal mutations. These results reflect a much stronger antigen stimulation to the host driven by the SARS- CoV- 2 N protein than by the S protein. + +<|ref|>sub_title<|/ref|><|det|>[[143, 607, 532, 627]]<|/det|> +## Binding characterizations of anti-N mAbs + +<|ref|>text<|/ref|><|det|>[[140, 655, 860, 886]]<|/det|> +To determine the antigenic targets by the N- reactive antibodies, we next analyzed the binding activities by ELISA with variant constructs of the N protein (full length N protein (N- FL): 1- 400; N protein N- terminal domain (N- NTD): 41- 174; and N protein C- terminal domain (N- CTD): 250- 364) (Figure 2a). Among the 32 mAbs that bound to N- FL, 13 antibodies bound to N- NTD, and one antibody bound to N- CTD (Figure 2b). A total of nine antibodies, including one antibody (nCoV400) that bound to N- CTD, seven mAbs (nCoV396, nCoV416, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[137, 87, 861, 562]]<|/det|> +nCoV424, nCoV425, nCoV433, nCoV454, and nCoV457) that bound to N- NTD and one mAb (nCoV402) that bound only to N- FL but not to the other variant N proteins, were chosen as representative antibodies for further studies. Purified antibodies were confirmed to bind the N- FL protein by ELISA (Figure 2c). The affinity of these antibodies to the N- FL protein was measured by surface plasmon resonance (SPR) (Figure 2d). In an effort to further characterize the functional and structural relationships, three antibodies, nCoV396, nCoV416 and nCoV457, were selected for the production of recombinant fragment antigen- binding (Fab) antibodies based on their unique characteristics. The mAb nCoV396 has a \(\mathrm{V_H}\) mutation frequency of \(2.8\%\) but a high binding affinity with a \(\mathrm{K_D}\) of 1.02 nM (Figure 2d) to the N protein. The mAbs nCOV416 and nCOV457 have high \(\mathrm{V_H}\) mutations at \(11.1\%\) and \(8.7\%\) , respectively and have a binding affinity to the N protein with \(\mathrm{K_D}\) values of 7.26 nM and 12.6 nM, respectively (Figure 2d, Extended Data Table 3). + +<|ref|>sub_title<|/ref|><|det|>[[144, 591, 504, 612]]<|/det|> +## Complex structure of mAb with N-NDT + +<|ref|>text<|/ref|><|det|>[[140, 640, 861, 906]]<|/det|> +To investigate the molecular interaction mechanism of the mAb nCoV396 with the N protein, we next solved the complex structure of the SARS- CoV- 2 N- NTD with the nCoV396 Fab nCoV396Fab at a 2.1 Å resolution by X- ray crystallography. The final structure was fitted with the visible electron density spanning residues 49- 173 (SARS- CoV- 2 N- NTD), 1- 220 (nCoV396Fab, the heavy chain of the Fab), and 1- 213 (nCoV396Fab, the light chain of Fab, except for the residues ranging from 136- 141). The complete statistics for the data collection, phasing, and refinement are presented in Extended Data Table 4. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 87, 860, 388]]<|/det|> +With the help of the high- resolution structure, we were able to designate all complementarity- determining regions (CDRs) in nCoV396Fab as the following: light chain CDR1, residues 23- 32 (L- CDR1), light chain CDR2, residues 51- 54 (L- CDR2), light chain CDR3, residues 94- 100 (L- CDR3), heavy chain CDR1, residues 26- 33 (H- CDR1), heavy chain CDR2, residues 51- 57 (H- CDR2), and heavy chain CDR3, residues 99- 108 (H- CDR3). Among them, we identified the interaction interface between N- NTD and L- CDR1, L- CDR3, H- CDR1, H- CDR2, and H- CDR3 of nCoV396Fab with an unambiguous electron density map (Figure 3a, Extended Data Figure 1a). + +<|ref|>text<|/ref|><|det|>[[140, 415, 860, 896]]<|/det|> +The interacting CDRs pinch the CT tail of the SARS- CoV- 2 N- NTD (residues ranging from 159 to 172), with extensive binding contacts and a buried surface area of 1,079 Ų (Extended Data Table 5). Light chain L- CDR1 and L- CDR3 of nCoV396Fab interact with residues ranging from 159- 163 of N- NTD via numerous hydrophilic and hydrophobic contacts (Figure 3b, Extended Data Figure 1b). Notably, the SARS- CoV- 2 N- NTD residue Q163 is recognized by the L- CDR3 residue T95 via a hydrogen bond and simultaneously stacks with the L- CDR3 residue W96 and the L- CDR1 residue Y31 (Figure 3c). In addition, a network of interactions from the heavy chain H- CDR2 and H- CDR3 of nCoV396Fab to residues 165- 172 of N- NTD suggests that the conserved residue K169 of SARS- CoV- 2 N- NTD has a critical role in nCoV396 antibody binding. K169 is recognized via hydrogen bonds with the E99 δ- carboxyl group and the T100, D102, S105 main- chain carbonyl groups inside the H- CDR3 of nCoV396Fab (Figure 3d). In addition, SARS- CoV- 2 N- NTD L167 also interacts + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 87, 860, 388]]<|/det|> +with I33, V50, N57, and A59 of H- CDR1 and H- CDR2 of nCoV396Fab through hydrophobic interactions (Figure 3e). Interestingly, all three residues (Q163, L167, and K169) of SARS- CoV- 2 N- NTD are relatively conserved in the highly pathogenic betacoronavirus N protein (Extended Data Figure 2a), which implies that nCoV396 may cross- interact with the SARS- CoV N protein or the Middle East respiratory syndrome coronavirus (MERS- CoV) N protein. Indeed, the binding affinities measured by SPR analysis demonstrate that nCoV396 interacts with the SARS- CoV N protein and the MERS- CoV N protein with a \(K_{\mathrm{D}}\) of 7.4 nM (Extended Data Figure 2b - c). + +<|ref|>text<|/ref|><|det|>[[140, 415, 860, 719]]<|/det|> +To discover the conformational changes between the SARS- CoV- 2 N- NTD apo- state and the antibody- bound state, we next superimposed the complex structure with the N- NTD structure (PDB ID: 6M3M) \(^{14}\) . The superimposition results suggest that the CT tail of SARS- CoV- 2 N- NTD unfolds from the basic palm region upon nCoV396Fab binding (Figure 3f), which likely contributes to the allosteric regulation of the normal full- length N protein function. Additionally, nCoV396Fab binding results in a 7.4 Å movement of the \(\beta\) - finger region outward from the RNA binding pocket, which may enlarge the RNA binding pocket of the N protein (Figure 3f). + +<|ref|>text<|/ref|><|det|>[[140, 747, 860, 873]]<|/det|> +In summary, our crystal structural data demonstrated that the human mAb nCoV396 recognizes the SARS- CoV- 2 N protein via a pinching model, resulting in a dramatic conformational change in residues 159 to 172, which is the linker region of N- NTD that is connected with other domains. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[142, 88, 568, 108]]<|/det|> +## MAb curbs N-induced complement activation + +<|ref|>text<|/ref|><|det|>[[140, 135, 860, 860]]<|/det|> +Although a recent study suggests that the complement cascade is hyperactive by the N protein in the lungs of COVID- 19 patients via the lectin pathway19, it is unclear how to develop a virus- free and an effective system for analyzing the role of the SARS- CoV- 2 N protein on complement hyperactivation. To this end, we developed a clinical autoimmune disease serum- based protease enzymatic approach to assess complement activation levels in the presence of the SARS- CoV- 2 N protein. Since complement activation initiated by the lectin pathway features MASP- 2 proteases by specific activity for cleaving complement components 2 and 4 (C2 and C4)32, we designed a C2 internal quenched fluorescent peptide- based analysis route for ex vivo complement hyperactivation (Figure 4a). Briefly, serum was collected from the peripheral blood of volunteers with an autoimmune disease, as their serum contains the necessary components for complement activation characterized by elevated C3 levels (Extended Data Table 6). Next, we collected the fluorescence signal from cleaved C2 synthetic peptide substrates (2Abz- SLGRKIQI- Lys(Dnp)- NH2) in reaction mixtures containing autoimmune disease serum in the absence or presence of the SARS- CoV- 2 N protein with or without the mAb nCoV396. The initial reaction rate (v0) was estimated at a single concentration of individual sera from duplicate measurements over a range of substrate concentrations. The steady- state reaction constants maximal velocity (Vmax) and Michaelis constant (Km) were determined for comparisons (Figure 4a). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 88, 860, 142]]<|/det|> +As shown in Figure 4b, the calculated \(V_{\max}\) of reactions without any other exogenous proteins is 1.49 response units (RU) \(\cdot \mathrm{s}^{- 1}\) . Additions of the SARS- CoV- + +<|ref|>text<|/ref|><|det|>[[140, 152, 860, 180]]<|/det|> +2 N protein (concentrations ranging from 0.5 \(\mu \mathrm{M}\) to \(10 \mu \mathrm{M}\) ) in the reactions + +<|ref|>text<|/ref|><|det|>[[140, 190, 860, 214]]<|/det|> +remarkably elevate the \(V_{\max}\) up to 2- fold, ranging from \(2.37 \sim 3.02 \mathrm{RU} \cdot \mathrm{s}^{- 1}\) . + +<|ref|>text<|/ref|><|det|>[[140, 225, 860, 249]]<|/det|> +Similarly, additions of the SARS- CoV- 2 N protein led to an approximate 1.8- fold + +<|ref|>text<|/ref|><|det|>[[140, 260, 860, 283]]<|/det|> +increase in the \(V_{\max} / K_{\mathrm{m}}\) values, which suggested that the specificity constant + +<|ref|>text<|/ref|><|det|>[[140, 295, 860, 319]]<|/det|> +\((K_{\mathrm{cat}} / K_{\mathrm{m}})\) of MASP- 2 to substrates is increased in the presence of the viral N + +<|ref|>text<|/ref|><|det|>[[140, 329, 860, 352]]<|/det|> +protein as the enzyme concentrations are equivalent among the reactions + +<|ref|>text<|/ref|><|det|>[[140, 363, 860, 387]]<|/det|> +(Extended Data Table 7 - 8). To confirm the kinetic analyses, Hanes plots ([S]/V + +<|ref|>text<|/ref|><|det|>[[140, 398, 860, 421]]<|/det|> +versus [S]) were also drawn and found to be linear (Figure 4c). Therefore, the + +<|ref|>text<|/ref|><|det|>[[140, 432, 860, 456]]<|/det|> +addition of the SARS- CoV- 2 N protein does not change the single substrate + +<|ref|>text<|/ref|><|det|>[[140, 467, 860, 490]]<|/det|> +binding site characterization of the enzymatic reactions. To assess the + +<|ref|>text<|/ref|><|det|>[[140, 501, 860, 524]]<|/det|> +suppression ability of nCoV396 to the SARS- CoV- 2 N protein- induced + +<|ref|>text<|/ref|><|det|>[[140, 535, 860, 558]]<|/det|> +complement hyperactivation function, we next conducted complement + +<|ref|>text<|/ref|><|det|>[[140, 569, 860, 592]]<|/det|> +hyperactivation analyses at various N protein: nCoV396 ratios. As shown in + +<|ref|>text<|/ref|><|det|>[[140, 603, 860, 627]]<|/det|> +Figure 4d, the addition of the N protein elevates the \(V_{\max}\) value up to 40- fold (1:0 + +<|ref|>text<|/ref|><|det|>[[140, 638, 860, 662]]<|/det|> +ratio), whereas the addition of the antibody nCoV396 decreases the \(V_{\max}\) in a + +<|ref|>text<|/ref|><|det|>[[140, 673, 860, 696]]<|/det|> +dose- dependent manner (Extended Data Table 9). To further validate the + +<|ref|>text<|/ref|><|det|>[[140, 707, 860, 730]]<|/det|> +function of nCoV396, we next performed complement hyperactivation analyses in + +<|ref|>text<|/ref|><|det|>[[140, 741, 860, 765]]<|/det|> +five other serum samples from autoimmune disease donors. Consistently, the + +<|ref|>text<|/ref|><|det|>[[140, 776, 860, 799]]<|/det|> +\(V_{\max}\) of reactions was boosted in the presence of the N protein in all samples but + +<|ref|>text<|/ref|><|det|>[[140, 810, 860, 833]]<|/det|> +declined in the presence of both the mAb nCoV396 and the N protein (Figure 4e). + +<|ref|>text<|/ref|><|det|>[[140, 844, 860, 868]]<|/det|> +In conclusion, these results demonstrate that the SARS- CoV- 2 N protein is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 859, 248]]<|/det|> +capable of inducing complement hyperactivation ex vivo, not only by facilitating the \(V_{\max}\) of MASP- 2 catalytic activity but also by enhancing the substrate binding specificity in the reactions. The N- reactive mAb nCoV396 specifically compromises the SARS- CoV- 2 N protein- induced complement hyperactivation within clinical serum samples. + +<|ref|>sub_title<|/ref|><|det|>[[144, 279, 287, 303]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[141, 338, 860, 604]]<|/det|> +From a quickly recovered COVID- 19 patient, we isolated 32 mAbs specifically targeting the SARS- CoV- 2 N protein. The binding affinity of mAbs ranged from 1 nM to 25 nM, which is comparable with the binding affinity of mature S protein- reactive antibodies22,25- 29 and the other mature antibodies identified during acute infections33,34. The characteristics of the isolated N- reactive mAbs are different from those of the isolated S- reactive mAbs in COVID- 19 patients during the early recovery phase, suggesting that sampling time is pivotal for identifying differential immune responses to different SARS- CoV- 2 viral proteins. + +<|ref|>text<|/ref|><|det|>[[141, 632, 863, 899]]<|/det|> +The crystal structure of nCoV396 bound to SARS- CoV- 2 N- NTD elucidates the interaction mechanism of the complex between the first reported N protein- reactive human mAb and its targeted N protein. Three conserved amino acids (Q163, L167, and K169) in the N protein are responsible for nCoV396 recognition, which provides evidence of cross- reactivity of nCoV396 to the N protein of SARS- CoV or MERS- CoV. Intriguingly, the nCoV396 binding of SARS- CoV- 2 N- NTD undergoes several conformational changes, resulting in an enlargement of the N- NTD RNA binding pocket enlargement and partial unfolding of the basic + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 88, 859, 178]]<|/det|> +palm region. More importantly, this conformational change occurs in the CT tail of the N- NTD, which may alter the positioning of individual domains in the context of the full- length protein and lead to a potential allosteric effect for protein functions. + +<|ref|>text<|/ref|><|det|>[[140, 201, 860, 895]]<|/det|> +Complement is one of the first lines of defense in innate immunity and is essential for cellular integrity and tissue homeostasis and for modifying the adaptive immune response35. Emerging evidence suggests that the complement system plays a vital role in a subset of critical COVID- 19 patients, with features of atypical acute respiratory distress syndrome, disseminated intravascular coagulation, and multiple organ failure9,10,36. A few pieces of evidence show that the N protein of highly pathogenic coronaviruses (SARS- CoV- 2 and SARS- CoV) is involved in the initiation of MASP- 2- dependent complement activation19,21. Encouragingly, critical COVID- 19 patients treated with complement inhibitors, including small molecules to the complement component C3 and an antibody targeting the complement component C5, show remarkable therapeutic outcomes19. Currently, there are 11 clinical trials related to targeting the complement pathway (https://clinicaltrials.gov). To avoid adverse effects of human complement component- targeting therapy, a viral protein- specific approach is warranted. The antibody nCoV396 isolated from COVID- 19 convalescent patients is an excellent potential candidate with a high binding affinity to the N protein and high potency to inhibit complement hyperactivation. As revealed by atomic structural information, the binding may allosterically change the full- length N protein conformation. To determine the role of nCoV396 in the suppression of complement hyperactivation, we monitored MASP- 2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 87, 860, 530]]<|/det|> +protease activity based on its specific fluorescence- quenched C2 substrate in sera from autoimmune disease patients. The complete complement components in the sera of patients with autoimmune disorders allow us to monitor the activating effects of the SARS- CoV- 2 N protein and its specific mAbs. Although we cannot calculate the other steady- state enzymatic reaction constants as the precise concentration of MASP- 2 in serum is unknown, we identified the \(V_{\mathrm{max}}\) of the specific C2 substrate for the enzymatic reaction. We demonstrated that the SARS- CoV- 2 N protein elevated the \(V_{\mathrm{max}}\) of the reaction, up to 40- fold, in the sera of all 7 individuals tested, while nCoV396 effectively suppressed the \(V_{\mathrm{max}}\) of the reaction mixtures. These results indicated that serum- based complement activation analysis of autoimmune disease patients is a virus- free and an effective method for examining complement activation mediated by the SARS- CoV- 2 N protein. + +<|ref|>text<|/ref|><|det|>[[140, 556, 860, 750]]<|/det|> +Although the precise interaction of the SARS- CoV- 2 N protein with MASP- 2 remains to be elucidated, our work defined the region on the SARS- CoV- 2 N protein recognized by the mAb nCoV396 that plays an important role in complement hyperactivation and indicates that human mAbs from convalescents could be a promising potential therapeutic candidate for the treatment of COVID- 19. + +<|ref|>sub_title<|/ref|><|det|>[[144, 775, 288, 799]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[140, 826, 859, 882]]<|/det|> +1 Epidemiology Working Group for Ncip Epidemic Response, Chinese Center for Disease Control and Prevention. [The epidemiological + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[198, 87, 857, 178]]<|/det|> +characteristics of an outbreak of 2019 novel coronavirus diseases (COVID- 19) in China]. Zhonghua Liu Xing Bing Xue Za Zhi 41, 145- 151 (2020) + +<|ref|>text<|/ref|><|det|>[[198, 191, 857, 319]]<|/det|> +Wiersinga, W. J., Rhodes, A., Cheng, A. C., Peacock, S. J. & Prescott, H. C. Pathophysiology, Transmission, Diagnosis, and Treatment of Coronavirus Disease 2019 (COVID- 19): A Review. Jama- J. Am. Med. Assoc. doi:10.1001/jama.2020.12839 (2020). + +<|ref|>text<|/ref|><|det|>[[198, 331, 857, 423]]<|/det|> +Tang, N., Li, D. J., Wang, X. & Sun, Z. Y. Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia. J Thromb. Haemost. 18, 844- 847 (2020). + +<|ref|>text<|/ref|><|det|>[[198, 435, 857, 529]]<|/det|> +Wang, D. W. et al. Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus- Infected Pneumonia in Wuhan, China. Jama- J. Am. Med. Assoc. 323, 1061- 1069 (2020). + +<|ref|>text<|/ref|><|det|>[[198, 541, 857, 595]]<|/det|> +Zhu, N. et al. A Novel Coronavirus from Patients with Pneumonia in China, 2019. N. Engl. J. Med. 382, 727- 733 (2020). + +<|ref|>text<|/ref|><|det|>[[198, 608, 857, 700]]<|/det|> +Zhou, F. et al. Clinical course and risk factors for mortality of adult inpatients with COVID- 19 in Wuhan, China: a retrospective cohort study. Lancet 395, 1054- 1062 (2020). + +<|ref|>text<|/ref|><|det|>[[198, 714, 857, 806]]<|/det|> +Gattinoni, L. et al. COVID- 19 Does Not Lead to a "Typical" Acute Respiratory Distress Syndrome. Am. J. Respir. Crit. Care Med 201, 1299- 1300, (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[201, 88, 858, 178]]<|/det|> +Magro, C. et al. Complement associated microvascular injury and thrombosis in the pathogenesis of severe COVID- 19 infection: A report of five cases. Transl. Res. 220, 1- 13 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 192, 858, 250]]<|/det|> +Cugno, M. et al. Complement activation in patients with COVID- 19: A novel therapeutic target. J. Allergy Clin Immunol 146, 215- 217, (2020). + +<|ref|>text<|/ref|><|det|>[[201, 262, 858, 319]]<|/det|> +Noris, M., Benigni, A. & Remuzzi, G. The case of complement activation in COVID- 19 multiorgan impact. Kidney Int. 98, 314- 322 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 331, 858, 389]]<|/det|> +Pang, R. T. et al. Serum proteomic fingerprints of adult patients with severe acute respiratory syndrome. Clin. Chem. 52, 421- 429, (2006). + +<|ref|>text<|/ref|><|det|>[[201, 401, 858, 495]]<|/det|> +Chen, J. H. et al. Plasma proteome of severe acute respiratory syndrome analyzed by two- dimensional gel electrophoresis and mass spectrometry. Proc. Natl. Acad. Sci. U. S. A. 101, 17039- 17044 (2004). + +<|ref|>text<|/ref|><|det|>[[201, 507, 858, 564]]<|/det|> +Ramlall, V. et al. Immune complement and coagulation dysfunction in adverse outcomes of SARS- CoV- 2 infection. Nat. Med. (2020). + +<|ref|>text<|/ref|><|det|>[[201, 576, 858, 666]]<|/det|> +Kang, S. et al. Crystal structure of SARS- CoV- 2 nucleocapsid protein RNA binding domain reveals potential unique drug targeting sites. Acta Pharm. Sin. B, doi:10.1016/j.apsb.2020.04.009 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 679, 858, 771]]<|/det|> +Ye, Q., West, A. M. V., Silletti, S. & Corbett, K. D. Architecture and self- assembly of the SARS- CoV- 2 nucleocapsid protein. Protein Sci., doi:10.1002/pro.3909 (2020). + +<|ref|>text<|/ref|><|det|>[[201, 784, 858, 841]]<|/det|> +Iserman, C. et al. Specific viral RNA drives the SARS CoV- 2 nucleocapsid to phase separate. bioRxiv, doi:10.1101/2020.06.11.147199 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 88, 860, 870]]<|/det|> +17 Li, J. Y. et al. The ORF6, ORF8 and nucleocapsid proteins of SARS-CoV- 2 inhibit type I interferon signaling pathway. Virus Res. 286, 198074, (2020). 18 Perdikari, T. M. et al. SARS-CoV-2 nucleocapsid protein undergoes liquid- liquid phase separation stimulated by RNA and partitions into phases of human ribonucleoproteins. bioRxiv, doi:10.1101/2020.06.09.141101 (2020). 19 Gao, T. et al. Highly pathogenic coronavirus N protein aggravates lung injury by MASP- 2- mediated complement over- activation. medRxiv, 2020.2003.2029.20041962, doi:10.1101/2020.03.29.20041962 (2020). 20 Guo, Y. R. et al. The origin, transmission and clinical therapies on coronavirus disease 2019 (COVID- 19) outbreak - an update on the status. Military Med. Res. 7, 11 (2020). 21 Liu, J. L., Cao, C. & Ma, Q. J. Study on interaction between SARS- CoV N and MAP19. Xi bao yu fen zi mian yi xue za zhi = Chinese journal of cellular and molecular immunology 25, 777- 779 (2009). 22 Chi, X. et al. A neutralizing human antibody binds to the N- terminal domain of the Spike protein of SARS- CoV- 2. Science 369, 650- 655 (2020). 23 Klasse, P. J. & Moore, J. P. Antibodies to SARS- CoV- 2 and their potential for therapeutic passive immunization. Elife 9, e57877 (2020). 24 Kreer, C. et al. Longitudinal Isolation of Potent Near- Germline SARS- CoV- 2- Neutralizing Antibodies from COVID- 19 Patients. Cell S0092- 8674(20), 30821- 30827 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[199, 88, 857, 149]]<|/det|> +Zost, S. J. et al. Potently neutralizing and protective human antibodies against SARS- CoV- 2. Nature, doi:10.1038/s41586- 020- 2548- 6 (2020). + +<|ref|>text<|/ref|><|det|>[[199, 158, 857, 249]]<|/det|> +Wu, Y. et al. A noncompeting pair of human neutralizing antibodies block COVID- 19 virus binding to its receptor ACE2. Science 368, 1274- 1278, (2020). + +<|ref|>text<|/ref|><|det|>[[199, 262, 857, 320]]<|/det|> +Wang, C. et al. A human monoclonal antibody blocking SARS- CoV- 2 infection. Nat. Commun. 11, 2251 (2020). + +<|ref|>text<|/ref|><|det|>[[199, 332, 857, 390]]<|/det|> +Ju, B. et al. Human neutralizing antibodies elicited by SARS- CoV- 2 infection. Nature 584, 115- 119 (2020). + +<|ref|>text<|/ref|><|det|>[[199, 402, 857, 492]]<|/det|> +Cao, Y. et al. Potent Neutralizing Antibodies against SARS- CoV- 2 Identified by High- Throughput Single- Cell Sequencing of Convalescent Patients' B Cells. Cell 182, 73- 84 (2020). + +<|ref|>text<|/ref|><|det|>[[199, 505, 857, 599]]<|/det|> +Liao, H. X. et al. High- throughput isolation of immunoglobulin genes from single human B cells and expression as monoclonal antibodies. J. Virol. Methods 158, 171- 179 (2009). + +<|ref|>text<|/ref|><|det|>[[199, 611, 857, 701]]<|/det|> +Moody, M. A. et al. H3N2 influenza infection elicits more cross- reactive and less clonally expanded anti- hemagglutinin antibodies than influenza vaccination. PLoS One 6 (10): e25797 (2011). + +<|ref|>text<|/ref|><|det|>[[199, 714, 857, 805]]<|/det|> +Duncan, R. C. et al. Multiple domains of MASP- 2, an initiating complement protease, are required for interaction with its substrate C4. Mol. Immunol. 49, 593- 600 (2012). + +<|ref|>text<|/ref|><|det|>[[199, 819, 857, 876]]<|/det|> +Stettler, K. et al. Specificity, cross- reactivity, and function of antibodies elicited by Zika virus infection. Science 353, 823- 826 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 87, 858, 328]]<|/det|> +34 Yu, L. et al. Delineating antibody recognition against Zika virus during natural infection. JCI Insight 2 (12):e93042 (2017). 35 Zipfel, P. F. & Skerka, C. Complement regulators and inhibitory proteins. Nat. Rev. Immunol. 9, 729-740 (2009). 36 Lo, M. W., Kemper, C. & Woodruff, T. M. COVID-19: Complement, Coagulation, and Collateral Damage. J. Immunol, doi: 10.4049/jimmunol.2000644 (2020). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 106, 255, 129]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[144, 156, 608, 176]]<|/det|> +## Recombinant SARS-CoV-2 S-ECD and N proteins. + +<|ref|>text<|/ref|><|det|>[[140, 193, 861, 675]]<|/det|> +Recombinant SARS- CoV- 2 S protein (extracellular domain of the S protein (ECD) with His and FLAG Tags, Z03481) was purchased from GenScript. SARS- CoV- 2 (formerly known as 2019- nCoV, recombinant full- length N protein with a CT 6x His tag (His tag, 40588- V08B) was purchased from Sino Biological. The SARS- CoV- 2 N protein expression plasmid (SARS- CoV N- FL) was a gift from the Guangdong Medical Laboratory Animal Center. SARS- CoV and MERS- CoV N- FL were purchased from RuiBiotech. SARS- CoV- 2 N- FL (residues 1 to 419), SARS- CoV- 2 N- NTD domain (residues 41 to 174), SARS- CoV- 2 N- CTD domain (residues 250 to 364), SARS- CoV N- FL and MERS- CoV N- FL were cloned into the pET- 28a vector and expressed in the Rosetta E. coli strain. Expression of SARS- CoV- 2 N- FL and variants in E. coli was induced with 0.1 mM isopropylthiob- galactoside (IPTG) and cultured overnight at 16 °C in Terrific Broth media. Expressed recombinant N proteins were initially purified by using nickel column chromatography and further purified via size- exclusion chromatography. + +<|ref|>sub_title<|/ref|><|det|>[[144, 691, 804, 744]]<|/det|> +## PBMCs from COVID-19 patients and sorting of single plasma cells and memory B cells by FACS + +<|ref|>text<|/ref|><|det|>[[143, 765, 860, 892]]<|/det|> +Blood samples were collected 9 - 25 days after the onset of the disease from patients who had recovered from COVID- 19 infection. PBMCs and plasma were isolated from blood samples by Ficoll- Paque PLUS (GE, 17- 1440- 02) density gradient centrifugation. All work related to human subjects was carried out in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 859, 421]]<|/det|> +compliance with Institutional Review Board protocols approved by the Institutional Review Board of the Fifth Affiliated Hospital of Sun Yat- sen University. Single plasma cells with the surface markers CD3+, CD14+, CD16+, CD235a+, CD19+, CD20low- neg, CD27hi and CD38hi and memory B cells with the surface markers CD3+, CD14+, CD16+, CD235a+, CD20+, CD19+, CD27+, SARS- CoV- 2 S+ and SARS- CoV- 2 N+ (BD Biosciences and Invitrogen) were sorted into individual wells in 96- well microtiter plates containing cell lysis and RT buffer for Ig gene amplification by fluorescence activated cell sorting (FACS) as previously described37 on a BD FACS Aria SORP. Data were analyzed using BD FACS Diva 8.0.1 software. + +<|ref|>sub_title<|/ref|><|det|>[[144, 443, 580, 464]]<|/det|> +## Isolation and expression of Ig \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes + +<|ref|>text<|/ref|><|det|>[[141, 482, 860, 819]]<|/det|> +Genes encoding \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) were amplified by reverse transcription (RT) and nested primer chain reaction (PCR) and nested PCR using the method previously described38. PCR products of Ig \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes were purified using a PCR purification kit (QIAGEN), sequenced in forward and reverse directions (Thermo Fisher scientific) and annotated by using IMGT/V- QUEST (www.imgt.org/IMGT_vquest). Functional \(\mathsf{V}_{\mathsf{H}}\) and \(\mathsf{V}_{\mathsf{L}}\) genes were used for assembling full- length Ig heavy and light chain linear expression cassettes by one- step overlapping PCR38. HEK- 293T cells in 12- well plates were transfected with the assembled Ig heavy and light chain pairs derived from the same single individual plasma cells using Effectene (QIAGEN) as the transfection reagent38. + +<|ref|>sub_title<|/ref|><|det|>[[144, 840, 620, 860]]<|/det|> +## Production of recombinant IgG and Fab antibodies + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[137, 87, 860, 600]]<|/det|> +For the production of purified full- length IgG1 antibodies, the \(\mathrm{V_H}\) and \(\mathrm{V_L}\) genes were cloned into the pCDNA3.1+ (Invitrogen)mammalian expression vector containing either the human IgG1 constant region gene, the human kappa light chain constant region gene or the lambda light chain constant region gene using standard recombinant DNA technology38. For the production of the purified nCoV396Fab antibody, a stop codon TGA was introduced after the sequence (5'- TCTTGTGACAAA- 3'), which encodes the amino acid residues SCDK, just before the hinge of the human IgG1 constant region39. Recombinant IgG1 antibodies and the nCoV396Fab antibody were produced in 293F cells cultured in serum- free medium by cotransfection with the generated IgG1 full- length or Fab heavy- and light chain gene expression plasmid pairs using polyethyleneimine40. Full- length IgG1 antibodies were purified by using Protein A column chromatography as described previously38. The nCoV396Fab antibody used for the crystal structure was purified by Lambda FabSelect, an affinity resin designed for the purification of human Fab with a lambda light chain (GE Healthcare)39. + +<|ref|>sub_title<|/ref|><|det|>[[142, 617, 822, 672]]<|/det|> +## Analysis of the binding of plasma antibodies and isolated mAbs to the S and N proteins by ELISA + +<|ref|>text<|/ref|><|det|>[[141, 692, 860, 888]]<|/det|> +We collected plasma from 6 patients and measured serum antibody titers using recombinant SARS- CoV- 2 S and N proteins as antigens to coat ELISA plates. Antibodies in the supernatant of the transfected 293T cultures harvested 3 days after transfection were screened by ELISA as described previously38. The binding of purified antibodies to the N or S protein was also analyzed by ELISA. Briefly, all protein antigens (S, N- FL, N- NTD (41- 174), and N- CTD (250- 364)) were used + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 87, 860, 565]]<|/det|> +at 200 ng/well to coat 96- well high- binding ELISA plates (Nunc 442404) using carbonate- bicarbonate buffer at pH 9.6. Plates were incubated overnight at 4 °C and blocked at room temperature for 2 h with PBS blocking buffer containing 5% w/v goat serum and 0.05% Tween- 20. Plasma or supernatant of transfected 293T cell cultures or purified mAbs in serial dilutions in PBS were incubated at 37 °C for 1 h. Goat anti- human IgG- horseradish peroxidase (HRP, 1:10,000 dilution) (Promega, W4031) as the secondary antibody diluted in blocking buffer was added and incubated at 37 °C for 1 h. These plates were then washed 5 times with PBS and developed with 100 μL of 3,3',5,5'- Tetramethylbenzidine (TMB) substrate/well (Solarbio PR1200). The reaction of the plates was stopped with 50 μL of 2 M H₂SO₄/well and read at a wavelength of 450 nm by an ELISA reader. The relative affinity of mAbs to the N protein antigen was determined as the effective concentration (EC₅₀) of the antibody resulting in half maximal binding to the antigen by curve fitting with GraphPad Prism software. + +<|ref|>sub_title<|/ref|><|det|>[[143, 583, 541, 603]]<|/det|> +## Affinity and kinetic measurements by SPR + +<|ref|>text<|/ref|><|det|>[[141, 622, 860, 890]]<|/det|> +The binding affinity (K₀), association rate (kₐ) and dissociation rate (kₐ) of purified mAbs to the N protein were determined by SPR using a Biacore X100 System. Anti- human Fc IgG antibody was first immobilized on a CM5- chip to approximately 6,000 RU by covalent amine coupling using a human antibody capture kit (GE Healthcare). Purified mAbs were captured on channel 2 of the CM5 chip to approximately 200 RU. Five 2- fold serial dilutions of the N protein starting at 40 μg/ml, 20 μg/ml, 10 μg/ml or 4 μg/ml were injected at a rate of 30 μL/min for 90 s with a 600 s dissociation. The chip was regenerated by injection + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 859, 214]]<|/det|> +of 3 M MgCl₂ for 30 s. All experiments were performed at room temperature, and data were analyzed using Biacore X100 Evaluation Software (version: 2.0.1). Curves were fitted to a 1:1 binding model to determine the kinetic rate constants (kₐ and kₐ). K₀ values were calculated from these rate constants. + +<|ref|>sub_title<|/ref|><|det|>[[144, 243, 465, 263]]<|/det|> +## Crystallization and data collection + +<|ref|>text<|/ref|><|det|>[[140, 288, 860, 910]]<|/det|> +SARS- CoV- 2 N- NTD (41–174) was cloned into the pRSF- Duet- 1 vector with an NT 6x His- SUMO tag, recombinantly expressed in E. coli and purified as an NT 6x His- Sumo- tagged protein. After Ni column chromatography followed by ulp1 digestion for tag removal, the SARS- CoV- 2 N- NTD (41–174) protein was further purified via size- exclusion chromatography. Prior to crystallization, the SARS- CoV- 2 N- NTD (41–174) sample was mixed with nCoV396Fab at a 1:1.5 molar ratio for approximately half an hour and then further purified via size- exclusion chromatography. Crystals were grown by the sitting drop method using a Mosquito LCP crystallization robot with 0.3 μL of protein (6 mg/ml) mixed with 0.3 μL of well solution at 16 °C. Better crystals were obtained in 0.01 M calcium chloride dihydrate, 0.05 M sodium cacodylate trihydrate (pH 7.2), 1.675 M ammonium sulfate and 0.5 mM spermine. The crystals were harvested after 3 days. Crystals were frozen in liquid nitrogen in the reservoir solution supplemented with 25% glycerol (v/v) as a cryoprotectant. X- ray diffraction data were collected at the Shanghai Synchrotron Radiation Facility BL18U at a wavelength of 0.979 Å and a temperature of 100 K. The complex structure of SARS- CoV- 2 N- NTD with the mAb nCoV396 was determined by Phenix molecular replacement using the SARS- CoV- 2 N- NTD structure (PDB ID: 6M3M) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 858, 213]]<|/det|> +and the monoclonal antibody omalizumab Fab (PDB ID: 6TCN) as the search models. The X- ray diffraction and structure refinement statistics are summarized in Extended Data Table 4. The final Ramachandran statistics are 97.1% favored, 2.9% allowed and 0.0% outliers. + +<|ref|>sub_title<|/ref|><|det|>[[144, 243, 537, 263]]<|/det|> +## Fluorescence-quenched substrate assays + +<|ref|>text<|/ref|><|det|>[[140, 285, 860, 911]]<|/det|> +The fluorescence- quenched substrate (FQS) [C2 P4- P4' (2Abz- SLGRKIQL- Lys(Dnp)- NH2)] was synthesized and purified in greater than 90% purity by Sangon Biotech (Shanghai, China). The FQS was solubilized in 50% (v/v) dimethylformamide (DMF). Assays were carried out in fluorescence assay buffer (FAB) (0.05 M Tris- HCl, 0.15 M NaCl, 0.2% (w/v) PEG 8,000, and pH 7.4) at 37 °C using final substrate concentrations in the range of 2.8125- 90 μM. A range of substrate dilutions, performed in triplicate, were prepared in FAB and were added to the wells at 100 μl/well in the assay plates. The plates were incubated at 37 °C for 10 minutes. Five microliters of serum from autoimmune patients was added to the mixture of the SARS- CoV- 2 N protein with nCoV396 at a 1:1.2 molar ratio in the assay plates. Fluorescence intensity was measured on an EnVision 2015 Multimode plate reader (PerkinElmer) at an excitation wavelength of 320 nm and an emission wavelength of 420 nm for the FQS. The initial reaction rate was estimated at a single concentration of enzyme from duplicate measurements over a range of substrate concentrations. To determine steady- state reaction constants ( \(V_{\text{max}}\) , half saturation constant ( \(K_{0.5}\) ) and Hill coefficient(h)), the experimental results were fitted using GraphPad Prism Version 8.0 (GraphPad Software, San Diego, CA) to an equation describing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 88, 858, 193]]<|/det|> +574 positive cooperativity: \(Y = V_{max} \times \frac{x^n}{K \circ \delta^b + x^n}\) , where 0.5 defines the relationship between the reaction rate (V) and the substrate concentration ([S]) when more than one binding site applies. + +<|ref|>sub_title<|/ref|><|det|>[[144, 220, 395, 244]]<|/det|> +## Methods references + +<|ref|>text<|/ref|><|det|>[[141, 280, 859, 373]]<|/det|> +578 37. Morris, L. et al. Isolation of a human anti- HIV gp41 membrane proximal region neutralizing antibody by antigen- specific single B cell sorting. PloS one 6, e23532 (2011). + +<|ref|>text<|/ref|><|det|>[[141, 399, 859, 490]]<|/det|> +581 38. Liao, H. X. et al. High-throughput isolation of immunoglobulin genes from single human B cells and expression as monoclonal antibodies. J. Virol. Methods 158, 171- 179 (2009). + +<|ref|>text<|/ref|><|det|>[[141, 519, 859, 612]]<|/det|> +584 39. Nicely, N. I. et al. Crystal structure of a non- neutralizing antibody to the HIV- 1 gp41 membrane- proximal external region. Nat. Struct. Mol. Biol. 17, 1492- 1494 (2010). + +<|ref|>text<|/ref|><|det|>[[141, 639, 859, 696]]<|/det|> +587 40. Smith, K. et al. Rapid generation of fully human monoclonal antibodies specific to a vaccinating antigen. Nat. Protoc. 4, 372- 384 (2009). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 105, 379, 130]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[141, 163, 860, 535]]<|/det|> +This work is supported by the COVID- 19 Emerging Prevention Products, Research Special Fund of Zhuhai City (ZH22036302200016PWC to S.C.; ZH22036302200028PWC to F. X.; ZH22046301200011PWC to H- X. L.); Emergency Fund from Key Realm R&D Program of Guangdong Province (2020B111113001) to H.S.; Zhuhai Innovative and Entrepreneurial Research Team Program (ZH01110405160015PWC, ZH01110405180040PWC) to H- X. L; We thank the staffs of the BL18U/19U/17U beamlines at SSRF for their help with the X- ray diffraction data screening and collections. We thank Junlang Liang, Tong Liu, Nan Li, Xiaoli Wang, Zhenxing Jia, and Jiaqi Li from Zhuhai Trinomab Biotechnology Co., Ltd. for technical assistants of mAbs isolation, production and characterization. + +<|ref|>sub_title<|/ref|><|det|>[[144, 565, 411, 589]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[140, 623, 860, 890]]<|/det|> +S. C., H. S., F. X. and H-X. L. contributed the conception of the study and established the construction of the article. S. C. and H-X. L. designed the experiments and wrote the manuscript. S. K., M. Y., S. H. contributed to protein purification and crystallization, in vitro protein-protein interaction analysis, and complement activation analysis. Y. W. contributed to mAbs isolation, in vitro protein-protein interaction analysis. S. C., S. K. M. Y., and S. H. performed structural determination and validation. S. C., S. K., Y. W. drew figures. X. C., Y. C., Q. C., Z. Z., Z. Z., Z. H., X. H., H. S., W. Z., and H. H. contributed to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 88, 858, 144]]<|/det|> +610 interpretation of data. Z. H., J. L., G. J., and F. X. contributed to clinical samples collections. S.K., M.Y., S. H., Y.W. contributed equally to this work. + +<|ref|>sub_title<|/ref|><|det|>[[144, 174, 380, 198]]<|/det|> +## Conflict of Interest + +<|ref|>text<|/ref|><|det|>[[144, 235, 512, 255]]<|/det|> +The authors declare no conflict of interest. + +<|ref|>sub_title<|/ref|><|det|>[[144, 286, 477, 310]]<|/det|> +## Data availability statement + +<|ref|>text<|/ref|><|det|>[[144, 346, 857, 401]]<|/det|> +The structure in this paper is deposited to the Protein Data Bank with 7CR5 access code. + +<|ref|>sub_title<|/ref|><|det|>[[144, 432, 426, 456]]<|/det|> +## Additional Information + +<|ref|>text<|/ref|><|det|>[[88, 492, 857, 549]]<|/det|> +Correspondence and requests for materials should be addressed to Shoudeng Chen (chenshd5@mail.sysu.edu.cn) and/or Hua- Xin Liao (tliao805@jnu.edu.cn). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 92, 950, 840]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 849, 115, 869]]<|/det|> +
Figure 1
+ +Acquisition and characterization of antibodies. Serum antibody titers of six SARS- CoV- 2 convalescent patients to the SARS- CoV- 2 S (a) and N (b) proteins measured by ELISA. Sorting of single plasma cells (c) with CD38 and CD27 double- positive B cells and single N and S protein specific memory B cells (d) by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 940, 134]]<|/det|> +FACS. (e) Percentage of different isotypes, VH and VL gene families of 32 isolated N-reactive antibodies. (f) Number of mutations in nucleotides and amino acids in VH and VL (Vk and Vλ) of 32 N-reactive antibodies and eight S-reactive antibodies (g). Length of the 32 Nreactive antibodies (h) and eight S-reactive antibodies (i) in H-CDR3. + +<|ref|>image<|/ref|><|det|>[[50, 140, 825, 870]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[44, 895, 120, 913]]<|/det|> +
Figure 2
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 949, 180]]<|/det|> +Reactivity and affinity of the isolated antibodies to the N protein antigens. (a) Schematic presentation of the SARS-CoV-2 N protein and two variants. (b) Antibodies expressed in transfected 293 cells were evaluated for binding to N-FL, N-NTD and N-CTD by ELISA. Plasma from patient ZD006 and an irrelevant mAb TRN006 were used as the positive control and negative control, respectively. (c) The ability of nine purified antibodies to the N-FL protein was determined by ELISA. (d) Binding affinity of nine selected antibodies to the N protein was measured by SPR. KD values are shown above the individual plots. + +<|ref|>image<|/ref|><|det|>[[50, 188, 820, 920]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[42, 939, 120, 957]]<|/det|> +
Figure 3
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 949, 292]]<|/det|> +Complex structure of mAb nCoV396 with SARS-CoV-2 N-NTD(a) Overall structure of the mAb SARS-CoV-2 N-NTD complex. The light chain (pink) and heavy chain (blue) of mAb nCoV396 are illustrated with the ribbon representation. SARS-CoV-2 N-NTD is illustrated with an electrostatic surface, in which blue denotes a positive charge potential, while red indicates a negative charge potential. (b) The N-NDD epitope recognized by mAb nCoV396. The interacting residues of N-NTD and nCoV396 are highlighted with the strict representation. Recognition of Q163 (c), K169 (d) and L167 (e) in N-NTD by mAb nCoV396. The dashed blue line represents hydrogen bonds. Hydrophobic interactions are illustrated with the dot representation. (f) Conformational changes in N-NTD upon mAb nCoV396 binding. The apo structure of N-NTD is colored gray. Antibody-bound N-NDD is colored green. The N-terminus and C-terminus of the N-NTD are labeled with circles. mAb nCoV396 is illustrated with surface representation. All figures were prepared by PyMol. + +<|ref|>image<|/ref|><|det|>[[50, 304, 949, 910]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[42, 925, 120, 944]]<|/det|> +
Figure 4
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 950, 316]]<|/det|> +Antibody nCoV396 compromises SARS- CoV- 2 N protein- induced complement hyperactivation. (a) Flow scheme of the SARS- CoV- 2 N protein and nCoV396 influencing the protease activity of MASP- 2 in the serum of autoimmune disease patients. The Michaelis- Menten curve shows the effect of increasing the N protein concentration (b) and antibody concentration (d) on the substrate C2 cleavage of MASP- 2 in the serum of patient 49 and patient 20. (c) A Hanes plot where C2 concentrationN0 is plotted against C2 concentration with the addition of 5 μM N protein. (e) The mAb nCoV396 inhibits the N protein- induced excessive cleavage of C2 in the serum of six autoimmune disease patients, and the last panel shows a summary of Vmax for all patients. Negative control (Negative Ctrl) and blank control (Blank Ctrl) represent reactions containing bovine serum albumin (BSA) instead of N or N+mAb and without exogenous protein, respectively. The mean and standard deviation (SD) values of three technical replicates are shown. P values: \(*P < 0.05\) ; \(**P < 0.01\) ; "-" indicates that the experimental kinetics did not conform to Michaelis- Menten kinetics. + +<|ref|>sub_title<|/ref|><|det|>[[44, 338, 311, 366]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 388, 765, 409]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 425, 275, 472]]<|/det|> +- 5ExTFigTablesNC.pdf- 4PDBNC.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/images_list.json b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..05ebbcb935c0a6d1beb459ee940f8af33a3c0f74 --- /dev/null +++ b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 - A regulatory-driven clustering method.", + "footnote": [], + "bbox": [ + [ + 50, + 0, + 944, + 777 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2 - Method validation and analysis tools.", + "footnote": [], + "bbox": [ + [ + 100, + 8, + 911, + 805 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3 - scRegClust predicts a way of potentiating temozolomide treatment in U3065MG cells.", + "footnote": [], + "bbox": [ + [ + 50, + 0, + 940, + 950 + ] + ], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4 - The regulatory landscape of neuro-oncology.", + "footnote": [], + "bbox": [ + [ + 55, + 40, + 940, + 790 + ] + ], + "page_idx": 37 + } +] \ No newline at end of file diff --git a/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9.mmd b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9.mmd new file mode 100644 index 0000000000000000000000000000000000000000..641068c93c7b6679446f4a32b23649e1bb847504 --- /dev/null +++ b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9.mmd @@ -0,0 +1,701 @@ + +# Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers + +Sven Nelander + +sven.nelander@igp.uu.se + +Uppsala University https://orcid.org/0000- 0003- 1758- 1262 + +Ida Larsson + +Uppsala University https://orcid.org/0000- 0001- 5422- 4243 + +Felix Held + +Chalmers University of Technology https://orcid.org/0000- 0002- 7679- 7752 + +Gergana Popova + +Uppsala University https://orcid.org/0000- 0001- 9804- 4041 + +Alper Koc + +Uppsala University https://orcid.org/0009- 0009- 8011- 9952 + +Soumi Kundu + +Uppsala University https://orcid.org/0000- 0002- 0759- 3051 + +Rebecka Jörnsten + +University of Gothenburg and Chalmers University of Technology + +## Article + +Keywords: regulatory programs, regulatory- driven clustering, cell state, phenotypic plasticity + +Posted Date: January 11th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 3658321/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on November 9th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 53954- 3. + +<--- Page Split ---> + +# Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers + +Ida Larsson \(^{1,\times}\) , Felix Held \(^{2,\times}\) , Gergana Popova \(^{1}\) , Alper Koc \(^{1}\) , Soumi Kundu \(^{1}\) , Rebecka Jörnsten \(^{2}\) , Sven Nelander \(^{1,\ast}\) + +1: Department of Immunology, Genetics and Pathology, Uppsala University, SE- 751 85 Uppsala, Sweden. + +2: Mathematical Sciences, Chalmers University of Technology, SE- 412 96 Gothenburg, Sweden. + +\(\times\) : Indicates equal contribution. + +\* Correspondence: sven.nelander@igp.uu.se + +Keywords: regulatory programs, regulatory- driven clustering, cell state, phenotypic plasticity + +<--- Page Split ---> + +## Abstract + +Nervous system cancers contain a large spectrum of transcriptional cell states, reflecting processes active during normal development, injury response and growth. However, we lack a good understanding of these states' regulation and pharmacological importance. Here, we describe the integrated reconstruction of such cellular regulatory programs and their therapeutic targets from extensive collections of single- cell RNA sequencing data (scRNA- seq) from both tumors and developing tissues. Our method, termed single- cell Regulatory- driven Clustering (scRegClust), splits target genes into modules and predicts essential kinases and transcription factors in little computational time thanks to a new efficient optimization strategy. Using this method, we analyze scRNA- seq data from both adult and childhood brain cancers to identify transcription factors and kinases that regulate distinct tumor cell states. In adult glioblastoma, our model predicts that blocking the activity of PDGFRA, DDR1, ERBB3 or SOX6, or increasing YBX1- activity, would potentiate temozolomide treatment. We further perform an integrative study of scRNA- seq data from both cancer and the developing brain to uncover the regulation of emerging meta- modules. We find a meta- module regulated by the transcription factors SPI1 and IRF8 and link it to an immune- mediated mesenchymal- like state. Our algorithm is available as an easy- to- use R package and companion visualization tool that help uncover the regulatory programs underlying cell plasticity in cancer and other diseases. + +<--- Page Split ---> + +## Introduction + +Nervous system cancers in adults and children share a common trait: tumor cells exist in multiple transcriptional states, which partly resemble the diversification of cells during normal embryonic development. In adult glioblastoma (GBM), cells resemble neural progenitor cells (NPCs), oligodendrocyte progenitor cells (OPCs) or astrocytes (ACs) (Neftel et al, 2019). In addition to these states, a substantial fraction of cells display a mesenchymal (MES)/injury response- like profile that has been linked to monocyte populations (Richards et al, 2021; Wang et al, 2017; Gangoso et al, 2021; Schmitt et al, 2021; Hara et al, 2021). In childhood cancers such as diffuse midline glioma, tumor cells move from a proliferative, OPC- like state to more differentiated states resembling either AC- or oligodendrocyte (OC)- like lineages (Filbin et al, 2018), and in medulloblastoma (MB) the four main subgroups differ in the proportion of undifferentiated vs differentiated tumor cells and their lineage resemblance (Hovestadt et al, 2019). + +There are strong reasons to believe that the cell states found in a tumor have implications for disease progression. For instance, invasive GBM cells mimic the migration mechanisms of neuronal cells (Cuddapah et al, 2014; Venkataramani et al, 2022), whereas tumor initiation is mediated by quiescent stem- like cells (Xie et al, 2022). Further, it's been reported that recurrent therapy- resistant GBM tumors display a higher fraction of MES cells (Bhat et al, 2013), possibly due to a phenotypic shift in the cells from non- MES to MES (Wang et al, 2022). However, despite their role in disease recurrence, progression, and drug resistance, we have limited knowledge about the mechanisms by which cell states are regulated. In particular, to understand the transcriptional regulation it is essential to identify sets of transcription factors (TFs) whose activity have a direct impact on a specific state phenotype. Furthermore, to identify drug targeting opportunities, it is important to identify sets of drugable proteins, particularly kinases, that can be linked to states. + +An increasing amount of available repositories of scRNA- seq data presents new opportunities to uncover such regulation with high precision. However, existing data analysis methods are not developed with this goal in mind. The frequently used clustering methods do not contain regulatory predictions; they are only concerned with grouping cells or genes. Previously, a standard method to identify regulators of gene signatures has been to first estimate a gene regulatory network (GRN), followed by post- processing to identify regulators linked to signature genes. This approach can, for instance, predict regulators of mesenchymal transformation from bulk transcriptional (Carro et al, 2010) and bulk multi- omic (Kling et al, 2016) data, but transferring it to the scRNA- seq setting + +<--- Page Split ---> + +has proven unsatisfactory as individual links are hard to reproduce (Pratapa et al, 2020; Chen and Mar, 2018). Adding additional layers of data, primarily ATACseq or CHIPseq measurements (Aibar et al, 2017; Kamimoto et al, 2020; Kamal et al, 2022), is one way of alleviating the problem but comes with its own limitations. Existing methods are not fast enough to process millions of cells, and restricting the model to only TFs makes it less amenable to modeling the impact of pharmacologically more relevant gene classes, such as kinases. Thus, the development of fast and accurate methods to reconstruct gene regulatory programs from scRNA-seq data will be important, if not essential. + +Here, we describe a new method for fast construction of regulatory programs that is well- suited for large data sets and for the detection of critical regulators of cell states. Applied to scRNA- seq data, it jointly detects modules (sets, clusters) of co- expressed genes and groups of regulators, such as transcription factors and kinases, linked to each module. Compared to methods of similar scope it is both faster and more flexible. As proof- of- principle, we show the applicability of our algorithm through three use cases. First, we apply it to a data set from peripheral blood mononuclear cells (PBMC) and benchmark it against the most comparable algorithm we can find, SCENIC+ (González- Blas et al (2023)). Second, we apply it to scRNA- seq data previously generated by our group (Larsson et al, 2021) to predict regulatory interventions to potentiate temozolomide (TMZ) treatment in GBM. Third, we integrate 13 data sets from the developing brain and cancers of the nervous system to investigate the regulatory landscape of neuro- oncology. We find a meta- module regulated by the transcription factors SPI1 and IRF8 which shows strong resemblance with the previously described immune cell- induced MES- like state. The algorithm we developed, scRegClust, is available as an easy- to- use R- package. + +## Results + +## A regulatory-driven clustering method + +We present a new approach to detect regulatory programs from scRNA- seq data alone, based on a mathematical framework that directly models the interaction between regulators and responding target gene sets (Figure 1A). Our model assumes that genes belong to either of two categories: tentative regulators and tentative targets. Given a volume of data \(\mathbf{Z}\) , organized as cells \(\times\) genes, we split the data in two parts along genes, a regulatory part \(\mathbf{Z}_r\) , and a target part \(\mathbf{Z}_t\) . These parts are further randomly split along cells into two sets of training and assessment cells. We then perform a + +<--- Page Split ---> + +clustering task by identifying groups of genes in \(\mathbf{Z}_{t}\) that correspond to sets of co- regulated genes, and for each such set, we perform a regulatory program re- construction task by identifying a small number of genes in \(\mathbf{Z}_{r}\) that are the likely regulators. These steps are repeated until configurations stabilize. To fit regulatory programs from data computationally, we developed an alternating two- step scheme iterating between determining the most predictive regulators (regulatory program re- construction) for each of a pre- specified number \(K\) of modules and using these optimal regulators to allocate target genes into \(K\) modules (clustering). Step 1 is approached using cooperative- Lasso (coop- Lasso, Chiquet et al, 2012) on the training set. In this step, we search for a sparse set of regulators that can be linked to most genes in a module, each regulator with the same sign (positive vs negative). A sparsity penalty parameter controls the number of regulators assigned to each module. In section "Regulatory- driven clustering improves accuracy" we provide details on how to choose this penalty parameter. To solve the optimization problem efficiently, we apply over- relaxed Alternating Direction Method of Multipliers (ADMM, Boyd et al, 2011; Xu et al, 2017) to minimize the coop- Lasso objective function. Once regulators have been assigned to modules, we move on to Step 2 where the task is to refine the target gene sets. We approach this by re- estimating coefficients per module using sign- constrained non- negative least squares (NNLS, Meinshausen, 2013) on the training set and an observation- based allocation scheme using the assessment set. Target genes whose predictive \(R^{2}\) is below a threshold across all modules are marked as noise and placed in a rag- bag cluster. This ensures that outliers do not overly distort the clustering result. There are several optional inputs that can guide the algorithm. Of note is the possibility to provide prior information on the relationship between genes in the form of a gene \(\times\) gene matrix where a non- zero entry indicates that there is a known link between the genes. A known link can for example be that both genes appear in the same biological process. Algorithmic details can be found in "Methods". The resulting regulatory programs are visualized using either a coupling matrix view, where predicted regulators are columns and target gene modules are rows, or a module network view (Figure 1A). In addition, the algorithm returns goodness- of- fit measures which can be used to guide the selection of optimization parameters (Figure 2A, Figure S6). + +## Comparison with SCENIC+ shows good agreement between methods + +We benchmarked our entire workflow on a publicly available data set from peripheral blood mononuclear cells (PBMC). Starting from a matrix of raw gene counts, the data was normalized and formatted + +<--- Page Split ---> + +(see details in “Methods”). As previously mentioned, potential regulators can be all TFs, all kinases, or any custom assignment of interest. For a direct comparison with the newly released SCENIC+ (González- Blas et al, 2023) (combining scRNA-seq and scATAC-seq), we used all TFs as potential regulators. We did not provide the algorithm with any prior information. Once the algorithm converged, it had clustered the PBMC data into nine gene modules and 62 active TFs interacting with one or several of these gene modules. + +From previous work on the PBMC data, we know that the cells can be divided into 8 clusters representing various types of immune cells (Figure 1B, Figure S1A). To functionally annotate the gene modules defined by scRegClust, we scored these against the gene signatures of the immune cell types (Figure S1B- C). Our algorithm can successfully distinguish modules upregulated in CD8+ T cells, CD4+ T cells, B cells, NK cells, FCGR3A+ monocytes, CD14+ monocytes and plasmacytoid dendritic cells (pDCs). Among the strongest regulators we find many examples of known regulators, e.g. PAX5 (B cells), TBX21 (NK cells), CUX2 (dendritic cells) and LEF1 (T cells), to mention a few (Figure 1C). As a point of comparison, SCENIC+ identified 53 (activator) TFs, out of which 30 overlap with our identified regulators ( \(p < 10^{- 5}\) , Fisher’s exact test; Figure 1D). To get a more detailed understanding of our output, we looked into the functional annotation for the predicted regulators with a particular focus on the 32 TFs that do not overlap with SCENIC+. For the regulators that had an annotation in the Human Protein Atlas (HPA, Uhlen et al, 2015), 76.3 % are classified as being immune cell enriched/enhanced compared to a baseline 22 % for all genes in the atlas. For the non-overlapping genes, 55 % are enhanced or enriched in immune cells and have an immune cell type annotation that agrees with the one predicted by scRegClust (Figure 1E, Supplementary Table 1). + +Jointly, the comparison between scRegClust and SCENIC+ on PBMC data showed a good level of agreement, despite the fact that scRegClust uses scRNA- seq data alone. In contrast to SCENIC+, which relies on pre- defined cell types for each cell in the primary data, scRegClust performs clustering on target genes, which can then be functionally annotated to match cell types. As previously mentioned, SCENIC+ is restricted to identifying TFs as regulators, due to its reliance on scATAC- seq data, while scRegClust in principle can consider any category of genes as potential regulators. The speed of the two algorithms is also remarkably different, scRegClust being three times faster than SCENIC+ (Figure 1F). + +<--- Page Split ---> + +## Regulatory-driven clustering improves accuracy + +We also benchmarked our workflow on synthetic data to explore the effect of associating regulators with modules on clustering quality, the robustness of our modeling assumptions, and correct regulator recovery. In addition, we developed tools and strategies to guide the selection of penalty parameters and the correct number of modules. The quality of the clustering with respect to the groundtruth is measured using the adjusted Rand index (Hubert and Arabie, 1985). Two measures are introduced to evaluate the quality of the selected regulatory model associated with each module. Predictive \(R^2\) per module measures the predictive performance on the assessment set of the regulators associated with each module. Regulator importance measures the change in predictive \(R^2\) per module when a single regulator is omitted from the full regulator model for a module. In addition, we considered the true and false positive rates, presented as a ROC curve, of estimating the correct regulators associated with each target gene. See the "Methods" section for details on the computed metrics. + +Synthetic data was generated to mimic characteristics of our in- house scRNA- seq data as well as characteristics of the multi- study comparison presented in the upcoming sections. We based our simulations on a modified scRNA- seq data set (see "Supplementary Methods") and generated four different setups capturing different aspects of real data scenarios. Below, a summary of the most important findings is presented. Consult the Supplementary Material for details on simulations and further results. + +If no initial clustering is provided, our algorithm is initialized by k- means++ (Arthur and Vassilvitskii, 2007) on the cross- correlation matrix of the target genes and regulators in the training set. When the penalization parameter was chosen appropriately, associating regulatory models with each module improved the clustering quality substantially (Figure 2B, adjusted Rand index increases from 0.33 to 0.97). scRegClust showed excellent regulator recovery properties in our simulations (see Figure 2B and C). For appropriate penalization, the correct regulators were selected and only few true regulators were disregarded. scRegClust performed well even when model assumptions were violated (see the Supplementary Material for details). Notably, scRegClust achieved a large improvement in clustering quality in a scenario with unequal module sizes. There, clustering on the cross- correlation matrix gave an essentially random result, whereas our algorithm managed to nearly correctly recover the simulated modules, when penalization was chosen appropriately (see Figure S5B, adjusted Rand index increases from 0.06 to 0.95). This shows that the integrated approach of structure modeling and clustering is superior to clustering on the correlation structure alone. + +<--- Page Split ---> + +Predictive \(R^{2}\) per module and regulator importance can be used as guidance for the selection of the appropriate range of penalty parameters. As a general rule, at least 5 to 10 different penalization parameters should be tested. Increasing the penalty parameter reduces the number of active regulators associated with each module. This will typically reduce predictive \(R^{2}\) gradually until the point when penalization becomes too strong and important regulators are not associated with a module any longer. A similar but reversed trend can be observed for regulator importance. Due to many regulators being included in the models for low penalization, importance will typically be low at first. Increasing the penalty parameter and therefore including fewer regulators in the model will increase regulator importance until it becomes inflated due to the inclusion of too few regulators. As illustrated in Figure 2A, finding this change point from gradual to rapid decrease or increase, respectively, indicates a range of appropriate penalty parameters. + +A common problem in clustering is the selection of the correct number of modules \(K\) . Contrary to methods like k- means, our algorithm can produce empty modules. This can be enforced more strongly by specifying a minimum non- empty cluster size. We find in our simulations that scRegClust tends to combine groundtruth modules with regulatory overlap when \(K\) is underspecified and tends to split groundtruth modules into smaller ones when \(K\) is overspecified (see Supplementary Table 3). To provide further guidance on the selection of the number of modules, we introduce silhouette scores (see "Methods" section). These can be interpreted similarly to standard silhouette values (Rousseeuw, 1987) as a measure of how well a target gene is located within its own module compared to how well it would be located within the nearest module. High average silhouette scores indicate that target genes are well located within their module. Analysis of the average silhouette score should be combined with analysis of the predictive \(R^{2}\) per module (see Figure S6). A good clustering achieves high values in both scores. Our recommendation is to start with an initial guess for the number of modules \(K_{0}\) and to run scRegClust on a range of module counts above and below \(K_{0}\) . Analysis of the silhouette scores and predictive \(R^{2}\) per module as illustrated in Figure S6 can give guidance on how to choose the optimal number of modules. + +To summarize, through an extensive simulation study where several data simulation setups reflecting various real data scenarios were tested, we demonstrated that scRegClust performs excellent both in terms of improving clustering accuracy and recovering regulatory interactions. + +<--- Page Split ---> + +## Targeting regulators of an OPC-like state potentiates temozolomide treatment + +Encouraged by the results from the benchmarking studies, we continued by applying scRegClust to complex data sets from nervous system cancers to explore two key questions. First, we asked if the algorithm could predict interventions, such as drugs or gene perturbations, that would push GBM cells into a temozolomide (TMZ)- sensitive state. In subsequent passages, we leverage the power of our algorithm to integrate scRNA- seq data across nervous system cancers and the developing brain to identify meta- modules with shared modes of regulation and biological function. + +One of the questions raised during our previous work on cell state transitions in GBM was how to successfully target an entire cell state, and not just individual target genes of these states. The question was prompted by our prediction that the minimal intervention needed to potentiate TMZ treatment is to block transitions to what we call "state 5", a state with an OPC- like, invasive profile (Larsson et al, 2021). To investigate this, we applied scRegClust to the previously mentioned scRNA- seq data from U3065MG cells (Larsson et al, 2021; Xie et al, 2015) using either TFs or kinases as potential regulators (Figure 3A). The algorithm was run without giving prior information on cluster assignment for the genes, which resulted in scRegClust clustering the genes in 9 (TF) or 6 (kinase) modules, respectively (Figure 3B). + +We found that scRegClust defines a group of modules that correspond to state 2, a highly proliferative progenitor state. By studying the functional profile of the modules further (Figure 3B), it is evident that the modules are related to proliferation and the cell cycle and that they diverge on cell cycle phase - where two of the modules represent the earlier phases of cell cycle (G1, G1S, S) while remaining modules represent the later phases (S, G2M, M). Several regulators identified for these modules agree with the module profiles, e.g. E2F1 and TK1 regulating the earlier phases of cell cycle (Martínez- Balbás et al (2000); Chang et al (1998)) while AURKA and CENPA are mostly active during G2M/M- phase (Marumoto et al (2003); Zeitlin et al (2001)). Another interesting observation is the astrocyte- like state 4, which is predicted to be positively regulated by ID3, known to induce astrocyte differentiation through the BMP pathway signaling (Bohrer et al (2015)). + +Addressing our initial question on how to suppress state 5 to potentiate TMZ treatment we focus on the regulation of the modules corresponding to this state (Figure 3C). The top kinase regulator is PDGFRA, consistent with state 5 having an OPC- like profile as PDGFRA is both a known regulator of OPCs in normal development (Ellison and de Vellis, 1994) and frequently amplified in the OPC- like + +<--- Page Split ---> + +GBM state (Neftel et al, 2019). Other strong kinase regulators are DDR1, a receptor tyrosine kinase expressed in oligodendrocytes in the developing brain (Vilella et al, 2019), and ERBB3, also implicated in the oligodendrocyte lineage development and the OPC- like state in GBM (Wang et al, 2020). The top TF regulators are SOX6 (positive regulation) and YBX1 (negative regulation). SOX6 belongs to the group D family of SOX TFs, which have been shown to regulate several stages of oligodendrocyte development. The combined predictions from Larsson et al (2021) and scRegClust suggest that blocking the activity of PDGFRA, DDR1, ERBB3 or SOX6, or increasing YBX1- activity, would potentiate TMZ treatment. We tested these predictions experimentally using two approaches. First, using CRISPR/Cas9- mediated knockdown we reduced the expression of DDR1 or SOX6 in U3065MG cells, followed by TMZ treatment. Viability measurements of the treated cells confirmed our predictions; both DDR1- KD cells and SOX6- KD cells respond better to TMZ treatment, which is seen through a dose- dependent shift in the viability curve compared to non- targeting control cells (Figure 3D). The effect is most pronounced in DDR1- KD cells, where we achieved a halving in viability in the KD population compared to control at the highest drug dose tested. + +Second, we combined TMZ with the tyrosine- kinase inhibitor (TKI) dasatinib (Figure 3E), which had an inhibition score of \(>0.85\) in the STITCH database for PDGFRA and ERBB3, and an interaction score of 0.889 with DDR1. In the U3065MG cell line, simultaneous treatment with TMZ and dasatinib showed no synergistic effect, neither did pre- treatment with TMZ followed by simultaneous treatment with both drugs. A synergistic effect was, however achieved when pre- treating cells with dasatinib for 72 hours, followed by combined treatment with both drugs (BSS \(>10\) ) (Figure 3F). Relating this to previous predictions, the results imply that pre- treatment with dasatinib depletes the state 5 population, while subsequent combined treatment inhibits cells from escaping to state 5 and therefore leaves the tumor population vulnerable to TMZ. + +Building on these results, we broadened our investigation's scope to explore regulators of cell plasticity in nervous system cancers. To that end, we included more GBM single- cell data sets, as well as data from other cancers of the nervous system and corresponding healthy tissue. + +## The regulatory landscape of neuro-oncology + +We used scRegClust to integrate the regulatory programs from the developing brain (Couturier et al, 2020; Eze et al, 2021; Hamed et al, 2022; Luo et al, 2021; La Manno et al, 2021; Weng et al, 2019), normal adrenal gland (Kildisuite et al, 2021), GBM (Couturier et al, 2020; Darmanis et al, 2017; + +<--- Page Split ---> + +LeBlanc et al, 2022; Neftel et al, 2019; Wang et al, 2019), MB (Hovestadt et al, 2019; Luo et al, 2021; Ocasio et al, 2019) and neuroblastoma (NB) (Kildisuite et al, 2021). We ran the algorithm for each data set individually and merged the resulting regulatory tables (Figure 4A). In the merged regulatory table (Figure 4B), the rows correspond to regulators (TFs) and the columns to gene modules derived from each individual study. The top annotation bars indicate the disease type and study that each module originate from. It is evident that certain meta- modules emerge (Box 1), consisting of several modules with similar gene content from different studies that are regulated by the same regulators. + +Two meta- modules, 1 and 7, represented actively proliferating cells from all types (normal tissue, GBM, MB and NB). Meta- module 1 was primarily driven by positive regulation by CENPA and HMGB2, while meta- module 7 was more strongly regulated by HMGB1 and DEK. Our model suggests a subdivision of MB and normal samples along a gradient regulated by HMGB1/2 and DEK vs NEUROD1, where NEUROD1 promotes non- proliferating neuronal- like NPC2- like cells (meta- module 3), and HMGB1/2 and DEK promotes neural progenitors with active proliferation. + +Two meta- modules, 2 and 6, were enriched for hallmark signatures of hypoxia and apoptosis, as well as the Ivy atlas signature pseudopalisading cells around necrosis (CT pan), probably representing a general stress response in the cells. These were regulated by YBX1 (#2) and members of the Fos and Jun family of TFs (#6). An astrocyte- like meta- module (#4) in normal and MB cells was linked to ID3, ID4 and HOPX. In GBM, these cells were clustered in meta- module 2 due to their additional enrichment for stress response signatures. An OPC- like meta- module (#13) was identified that was driven by e.g. OLIG1, MYRF and SOX10. + +There were two distinct mesenchymal meta- modules, 5 and 15. One mesenchymal meta- module (#5) coincided with an enrichment of microglial signatures and was primarily driven by SPI1 and IRF8 in normal, GBM and MB cells. A possible explanation to this meta- module is that it captures the phenotype of cells undergoing a mesenchymal shift induced by tumor- associated macrophages (TAMs) or microglia, and that the enrichment of microglia- related gene signatures in disease modules are due to this previous interaction between tumor and immune cells. SPI1 and IRF8 could therefore be candidate regulators of the observed mesenchymal shift in GBM cells. SPI1 is a pioneer TF, i.e. one that can bind directly to condensed chromatin and thereafter recruit other, non- pioneer TFs, to the site. The known normal function of SPI1 is in controlling the cell fate decision of hematopoietic cells, e.g. in macrophage differentiation. After binding to regulatory elements in the genome, it regulates + +<--- Page Split ---> + +gene expression by recruiting other transcription factors, such as interferon regulatory factors, e.g. IRF8 (Pham et al, 2013; Le Coz et al, 2021). IRF8 has been previously identified as a candidate gene involved in the immune evasion of GBM cells. Gangoso et al (2021) observed that IRF8 was upregulated in GSC cells in an immuno- competent mouse model following immune attack, and since IRF8 is normally a myeloid- specific master transcription factor, they termed this immune evasion strategy “myeloid- mimicry”. + +The other mesenchymal metamodule (#15) showed no enrichment of microglial signatures, but instead showed enrichment for the Ivy atlas signature microvascular proliferation. This meta- module was interesting in the sense that it had regulators that depended on disease category. GBM samples were regulated by FOX51 and HEY1, while MB samples were regulated by AEBP1 and TBX18. Normal tissue samples were driven by previously mentioned regulators but also MEF2C and TSC22D4. A third meta- module (#12) showed a slightly weaker enrichment for the microvascular proliferation signature and included all types. This was driven by several TFs from the ERG family (ERG, FLI, ETS1) and SOX17, to name a few. Several of these regulator interactions are consistent with their described role in literature, e.g. FLI1 which is a prognostic marker in astrocytoma (Hu et al, 2022) and is predicted to be a regulator of perivascular- like glioma cells (Hu et al, 2022) and the endothelial potential of myogenic progenitors (Ferdous et al, 2021). Similarly, the role of SOX17 as a promotor of tumor angiogenesis has been described in mice (Yang et al, 2013). + +Our model also suggests a subdivision of NB samples (meta- modules 9 and 14) along an adrenal- to- mesenchymal axis, driven by ZNF90, MEIS2 and MYC, each of which was positively linked to mesenchymal- like signatures and negatively linked to ADRN- like signatures. Remaining meta- modules (#8, 10, 11) were specific to one type and driven by just one or two regulators. + +Similarly to the TF regulatory landscape, the kinase map (Figure S2) revealed several likely links and made interesting predictions. For instance, along the oligodendrocytic lineage, AATK is linked to a meta- module in both GBM and normal developing brain that is enriched for oligodendrocyte differentiation. PDGFRA and DGKB is linked to a meta- module with an OPC- like profile, whereas PDGFRB is linked to invasiveness and microvascular proliferation signatures. CKB regulates a meta- module of NPC signature targets in normal developing brain, GBM and MB, and suppresses cell cycle genes in MB and GBM. CCND1 suppresses NPC markers. Mature astrocytes were primarily linked to NTRK2 activity. CCL2 drives a microglia- like signature in both normal developing brain and MB. As in the TF map, known cell cycle driving kinases converge on similar modules in GBM, + +<--- Page Split ---> + +MB and normal developing brain, e.g. AURKA, PLK1 and CCNB2. Still there was an MB- specific TK1- driven cell cycle module. AK4 links to a GBM- specific module enriched for cellular tumor genes. We also found some hits of unclear significance, like SGK1 regulating a common meta- module in GBM, MB and normal developing brain. + +## Discussion + +The high extent of intratumoral heterogeneity in nervous system cancers and the plastic behavior of tumor cells are major obstacles in the search for more efficient therapies against these often deadly diseases. The enormous amount of scRNA- seq data that has been generated to map intratumoral heterogeneity presents great opportunities to understand the regulation of transcriptional cell states and cell state transitions, but it also challenges us to develop suitable methods to properly analyze the data. In this work, we have addressed the lack of computational methods to identify such regulators of intratumoral heterogeneity and present a method that simultaneously detects gene modules and their regulators from scRNA- seq data. The utility of our method is demonstrated through three use cases, where it is applied to real data, and through a thorough investigation of model properties using synthetic data. + +By associating a regulator model with each module, scRegClust improves substantially on initial clustering results derived from the correlation of target genes and regulators. We further develop scores, such as \(R^2\) per module, regulator importance, and a silhouette score, to help the user with parameter choices and to support the interpretation of the resulting regulatory network. In our first use case, we apply scRegClust to the PBMC data set and compare our results to those of SCENIC+. We show that we can, in an unsupervised fashion, recapitulate signatures of the known cell types present in the data as well as several known regulators of each cell type. The result is comparable to that of SCENIC+, but obtained in shorter time and with only one data source. We do however emphasize that although the methods produce a similar output, they differ in several important technical aspects, and should be treated as complementing rather than competing methods. + +In our second and third use case, scRegClust is applied to data sets from nervous system cancers (GBM, MB and NB) and the developing brain and used to derive insights about the diseases that warrant further investigation. We identify several regulators of an OPC- like, invasive GBM state and combine TMZ and dasatinib, a tyrosine kinase inhibitor targeting the state regulators, to potentiate + +<--- Page Split ---> + +TMZ treatment. Dasatinib has in previous studies shown anti- migratory effects (Li et al, 2015), consistent with state 5 having an invasive profile, and were tested in combination with TMZ in a clinical trial as treatment against GBM (Milano et al, 2009). However, this trial was terminated before proceeding to phase II. We find that the combination treatment is only synergistic in our U3065MG cell line when we first pre- treat the cells with dasatinib, followed by combined treatment with both drugs, and speculate whether this means that state 5 first needs to be depleted for the combination treatment to be synergistic. These speculations need to be more carefully investigated by measuring whether the protein levels of state 5 markers decrease following dasatinib treatment. Second, by performing an integrative study of the regulatory programs of GBM, MB, NB and the developing brain, we find that SP11 and IRF8 are candidate regulators of the mesenchymal shift induced by TAMs and/or microglia in GBM. These regulators can be an important piece of the puzzle to understand the mechanism behind the observed mesenchymal transition, which appears to be a strategy for the tumor cells to evade the immune system and become increasingly resistant to treatment. + +For future developments, we see several possible extensions of the algorithm. At present, scRegClust has the functionality to include prior knowledge of gene- gene relationships to guide the clustering of target genes. It would however be interesting to explore the possibility of including prior knowledge in the regulator- target gene relationship as well, e.g. by favouring connections that have support in an external data set (ATAC- seq, CHIP- seq etc). In addition to this, we see extensions related to non- additive contributions to the regulation of a module, extended regulation modeling to allow for feedback regulation, or automatic partitioning in regulator- and target genes. + +To conclude, we present an algorithm that operates on scRNA- seq data alone to construct regulatory programs consisting of regulators and target gene modules, and use it to understand the regulatory mechanisms of cell state plasticity. The algorithm is provided as an easy- to- use R- package and can be applied to any scRNA- seq data set. + +<--- Page Split ---> + +## Methods + +## Datasets + +All publicly available data sets used in this publication are listed in Table 1, including the data repository from where they were downloaded. The notation "3CA" refers to the Curated Cancer Cell Atlas provided by the Tirosh Lab (Gavish et al, 2021). + +Table 1: Summary of data sets used in this publication. + +
StudyData repositoryDisease
Couturier et al. 20203CA/brainDeveloping brain
Couturier et al. 20203CA/brianGlioblastoma
Darmanis et al. 20173CA/brainGlioblastoma
Eze et al. 2021UCSC Cell BrowserDeveloping brain
Hamed et al. 2022GEO: GSE200202Developing brain
Hovestadt et al. 20193CA/brainMedulloblastoma
Kildisuite et al. 2021Neuroblastoma Cell AtlasNeuroblastoma
Kildisuite et al. 2021Neuroblastoma Cells AtlasNormal fetal adrenal gland
Larsson, Dalmo et alArrayExpress E-MTAB-9296Glioblastoma
Leblanc et al. 2022GEO: GSE173278, GSE193884Glioblastoma
Luo et al. 2021Short Read Archive: GSE156633Developing cerebellar granule
Luo et al. 2021Short Read Archive: GSC156633Medulloblastoma
Manno et al. 2021Sequence Read Archive: PRJNA637987Developing brain
Neftel et al. 20193CA/brainGlioblastoma
Ocasio et al. 2019GEO: GSE129730Medulloblastoma
Ocasio et al. 2019GEO:GSE129730Cerebellum
Wang et al. 20203CA/brainGlioblastoma
Weng et al. 2019GEO: GSE122871Developing brain
PBMC10X websitehuman blood cells
+ +## Data processing and formatting + +If available, raw counts were downloaded from the indicated source in Table 1. The data was processed before further analysis in R (R Core Team, 2022) using the Seurat package (Butler et al, 2018). Cells containing less than 200 genes and genes present in less than 3 cells were filtered out. Data generated from a UMI- based single- cell sequencing protocol (10X Chromium, CEL- seq2) were normalized using sctransform (Hafemeister and Satija, 2019), while expression levels for remaining data sets were quantified using \(\log_{2}(\mathrm{TPM} / 10 + 1)\) , as suggested by Tirosh et al (2016). When needed, non- malignant cells were filtered out from the data set based on the provided metadata from the + +<--- Page Split ---> + +authors or the 3CA- portal (Gavish et al, 2021). We provide functionality to format the normalized data matrix to conform with the desired input of our scRegClust- algorithm (more details below). + +## Clustering Algorithm + +We developed a two- step alternating algorithm to simultaneously perform clustering on a set of target genes as well as to associate a set of regulators with each cluster (module). In a first step, given an initial clustering of the data, the algorithm determines the linearly most predictive regulators for each cluster as well as whether the regulator acts stimulating or repressing. Then, in a second step, these optimal regulators are used to allocate target genes into clusters. Prior knowledge about target gene relationships can be included to guide cluster allocation. These two steps are repeated until configurations stabilize or a maximum number of steps is reached. The algorithm was implemented as an R- package and can be found on GitHub at https://github.com/sven- nelander/scregclust. + +## Algorithmic outline of scRegClust + +The algorithm takes as its input two matrices \(\mathbf{Z}_{t}\) and \(\mathbf{Z}_{r}\) containing \(n\) rows of matching observations, typically cells, on \(p_{t}\) target genes and \(p_{r}\) tentative regulator genes in their columns, respectively. In addition, the algorithm requires the number \(K\) of desired clusters and an initial clustering of the target genes into \(K\) clusters. If the latter is not supplied, an initial clustering is produced as described below under "Preprocessing and initialization". + +The motivation behind our method is gene \(j\) in cluster \(i\) is described by a sign- constrained linear model + +\[\mathbf{Z}_{t}^{(:,j)} = \mathbf{Z}_{r}^{(:,R_{i})}\mathrm{diag}(\mathbf{s}_{i})\mathbf{B}_{i}^{(:,j)} + \sigma_{i,j}\mathbf{e}_{i,j}\] + +where \(\mathrm{R}_{i}\) is a subset of regulators associated with cluster \(i\) , \(\mathbf{s}_{i}\) is a vector of size \(|\mathrm{R}_{i}|\) containing the sign of each regulator associated with cluster \(i\) , \(\mathbf{B}_{i}\) are non- negative coefficients of dimension \(|\mathrm{R}_{i}|\times p_{t}\) , \(\sigma_{i,j}^{2}\) is a positive residual variance for each target gene \(j\) and cluster \(i\) , and \(\mathbf{e}_{i,j}\) are additive errors that are assumed to be normally distributed. Each cluster is therefore described by a selection of regulators \(\mathrm{R}_{i}\) and their signs \(\mathbf{s}_{i}\) encoding the assumption that a regulator acts stimulating (positive sign) or repressing (negative sign) on all target genes in the cluster. Coefficient values and residual variances are considered to be target gene- specific. + +To cluster target genes and to select regulators associated with each cluster we consider the partitioning + +<--- Page Split ---> + +problem + +\[\begin{array}{r l} & {\arg \min_{\Pi ,\mathbf{R},\mathbf{B},\mathbf{s},\pmb {\sigma}}\quad \frac{1}{2}\sum_{i = 1}^{K}\sum_{j = 1}^{p_{t}}\pmb {\Pi}^{(i,j)}\left(n\log \left(\sigma_{i,j}^{2}\right) + \frac{1}{\sigma_{i,j}^{2}}\left\| \pmb{Z}_{t}^{(i,j)} - \pmb{Z}_{r}^{(i,\mathbf{R})}\mathrm{diag}(\pmb {s}_{i})\pmb{B}_{i}^{(i,j)}\right\|_{F}^{2}\right)}\\ & {\quad \mathrm{such~that}\quad |R_{i}|\leq N_{i,r},\pmb{s}_{i}^{(k)}\in \{-1,1\} ,\pmb {B}_{i}\geq \mathbf{0}\mathrm{~and~}\sigma_{i,j} > 0}\\ & {\quad \mathrm{for~all}\quad i = 1,\ldots ,K,\ j = 1,\ldots ,p_{t},\ k = 1,\ldots ,p_{r},} \end{array} \quad (1)\] + +where \(\mathbf{\Pi}\) is a \(K\times p_{t}\) cluster membership matrix such that \(\mathbf{\Pi}^{(i,j)} = 1\) if target gene \(j\) is in cluster \(i\) and 0 otherwise. The size constraint \(|{\bf R}_{i}|\leq N_{i,r}\) ensures that only a subset of regulator is associated with each cluster. Note, that we do allow regulators to appear in multiple clusters. + +The direct solution of this optimization problem, however, requires to solve a combinatorial number of regression problems, which is computationally too demanding in practice. Instead, we used a data- driven approximation to solve Eq. (1). + +Table 2 contains an overview of all symbols used in the description of the scRegClust algorithm. + +## Data-driven approximation + +We proceed in the fashion of alternating clustering algorithms, such as k- means, by alternating between (1) determining the structure of clusters ( \(\mathrm{R}_{i}\) and \(\mathbf{s}_{i}\) ), and (2) re- assigning cluster membership (II). + +Preprocessing and initialization Observations of target genes and regulators (or a randomly reduced subset) are randomly split into two sets to allow for unbiased estimation of the model quality during cluster allocation below. The ratio is user- defined but 50- 50 by default. If the observations are stratified, the sample assignment of each observation can be supplied and splitting is performed within each sub- group. We write \(\mathbf{Z}_{t,d}\) and \(\mathbf{Z}_{r,d}\) for the \(d\) - th data split containing \(n_{d}\) observations \((d = 1,2)\) . The first data split is regarded as a training set, whereas the second data split functions as an assessment set. We therefore compute the preprocessing below on the training set and apply the same values to the assessment set. + +During data splitting, the user can choose whether or not to center the target genes within each sub- group defined by the sample stratification. In addition, after splitting, both target and regulatory genes are centered individually (without regard to stratification) and regulatory genes are scaled to standard deviation 1. + +<--- Page Split ---> + +Algorithm 1 High- level overview of the scRegClust algorithm. + +Input: Pre- processed expression data for target genes \(\mathbf{Z}_{t}\) and regulators \(\mathbf{Z}_{r}\) . Optionally, an initial clustering \(\mathbf{\Pi}\) and an indicator matrix \(\mathbf{J}\) describing prior knowledge. + +Initialization: Split data randomly into two sets \((\mathbf{Z}_{t,1},\mathbf{Z}_{r,1})\) and \((\mathbf{Z}_{t,2},\mathbf{Z}_{r,2})\) , and, if not provided in input, find initial clustering \(\mathbf{\Pi}\) by applying kmeans++ to the cross- correlation matrix of \(\mathbf{Z}_{t,1}\) and \(\mathbf{Z}_{r,1}\) . + +for cycle \(c\) from 1 to maximum cycle do + +Store current clustering \(\mathbf{\Pi}\) + +for each cluster \(i\) do + +Step 1: Selection + +Given the current clustering \(\mathbf{\Pi}\) : + +Identify active regulators \(\mathbf{R}_{i}\) and determine their signs \(s_{i}\) . + +end for + +for each cluster \(i\) do + +Step 2a: Re- estimation + +Given active regulators \(\mathbf{R}_{i}\) and their signs \(s_{i}\) : + +Re- estimate coefficients for each target gene. + +Compute normalized likelihood \(\mathbf{L}_{j}^{(i,l)}\) for each gene \(j\) and observation \(l\) . + +Incorporate prior information if supplied. + +end for + +for each target gene \(j\) do + +Step 2b: Allocation + +Compute \(R_{i,j}^{2}\) for each cluster \(i\) and perform one of the following. + +a) Sort out noisy genes into the rag-bag cluster if \(R_{i,j}^{2}\) across all clusters is too low. + +b) Update the cluster allocation of \(j\) according to a majority vote across observations. + +end for + +if \(\mathbf{\Pi}\) is equal to any previously stored clustering then + +Stop iteration. + +end if + +end for + +<--- Page Split ---> + + +Table 2: Overview of symbols used in the description of the scRegClust algorithm + +
SymbolDescription
Problem setup:
nTotal number of cells in the input matrix
ptNumber of target genes
prNumber of regulators
Zt,Zrtarget gene and regulator expression (n × pt resp. n × pr matrices)
KDesired number of clusters
ΠK × pt cluster membership matrix for target genes with Π(i,j) ∈ {0,1} and ∑K=1 Π(i,j) ≤ 1
Jpt × pt indicator matrix describing prior knowledge of biological relationships between target genes (e.g. pathway co-occurence)
For data splits d = 1, 2:
ndNumber of cells in the d-th data split
Zt,dtarget gene expression used in the d-th data split (nd × pt matrix)
Zr,dregulator expression used in the d-th data split (nd × pr matrix)
For each cluster i = 1, ..., K:
CiSet of target genes in cluster i
RiSet of regulators associated with cluster i
Ni,rMaximum number of regulators associated with cluster i
siSign vectors containing one sign for each regulator in Ri
Binon-negative regression coefficients (|Ri| × pt matrix)
σi,j2positive variance parameter for each target gene and cluster
Optimization-related (i = 1, ..., K, j = 1, ..., pt):
λPositive penalization parameter used in coop-Lasso
wiPositive weight vector of length pr for each cluster i in coop-Lasso
Bi,OLSOrdinary least squares estimates of the regression coefficients in cluster i (pr × |Ci| matrix)
Bi,CLCoop-lasso estimates of the regression coefficients in cluster i (pr × |Ci| matrix)
τNon-negative threshold for rag-bag clustering
pi,jPrior probability of target gene j being in cluster i
LjK × n2 likelihood matrix for target gene j
vjVector of n2 votes for the cluster assignment of target gene j
μPrior strength in [0,1] for trade-off between likelihood and prior in cluster allocation
Validation measures (i = 1, ..., K, j = 1, ..., pt, k = 1, ..., pr):
R2iPredictive R2 for cluster i
R2i,-kPredictive R2 for cluster i with regulator k omitted
Ii,kImportance of regulator k in cluster i
R2i,jPredictive R2 for target gene j predicted by regulators in Ri
SjSilhouette score for target gene j
+ +Output: T Regulatory table providing a summary of regulator strength in each cluster \((p_{r}\times K\) matrix) + +If no initial cluster membership is provided, the cross- correlation matrix of \(\mathbf{Z}_{t,1}\) and \(\mathbf{Z}_{r,1}\) is computed. The k- means++ algorithm (Arthur and Vassilvitskii, 2007) with the cross- correlation matrix as an + +<--- Page Split ---> + +input and multiple restarts is then used to find an initial clustering of the target genes into \(K\) clusters. Note that this step takes regulators into account and is therefore not equivalent to the clustering of only target genes. + +Determining a regulatory model for each cluster. Given the cluster membership of target genes, the goal of Step 1 is to determine which regulators are linearly most predictive for the target genes in each cluster. If cluster membership is considered to be known, the optimization problem in Eq. (1) can be solved separately for each cluster. However, due to the computational complexity of the optimization problem, selection of at most \(N_{i,r}\) out of \(p_{t}\) regulators as well as their signs, it is not possible to solve it exactly. In our scenario for cluster \(i\) , there are \(\sum_{k = 0}^{N_{i,r}}\binom{p_{r}}{k}\cdot 2^{k}\) candidate sets of regulators and their signs to consider. + +The size constraint \(|\mathbf{R}_{i}|\leq N_{i,r}\) in (1) is equivalent to replacing \(\mathbf{Z}_{r}^{(:, \mathbf{R}_{i})}\) with \(\mathbf{Z}_{r}\) , considering \(\mathbf{B}_{i}\) to be a \(p_{r} \times p_{t}\) matrix, and introducing \(\ell_{0}\) regularization on rows of \(\mathbf{B}_{i}\) . This approach ensures that at most \(N_{i,r}\) rows of \(\mathbf{B}_{i}\) are non- zero and the indices of these rows correspond to \(\mathbf{R}_{i}\) . A common approach to ensure computability of \(\ell_{0}\) penalized problems is relaxation of the constraint to a convex norm instead. Here, groups of coefficients are considered that are either zero or non- zero simultaneously. This is a typical scenario covered by the group- Lasso (Yuan and Lin, 2006). In addition, to ensure sign- consistency of the selected regulators, an extension of the group- Lasso known as the cooperative- Lasso (coop- Lasso, Chiquet et al, 2012) is used to solve the regulator selection problem. + +Applied to our case, the coop- Lasso solves the following optimization problem + +\[\mathbf{B}_{i,\mathrm{CL}} = \arg \min_{\mathbf{B}}\frac{1}{2}\sum_{j\in \mathcal{C}_{i}}\frac{1}{n_{1}\sigma_{i,j}^{2}}\left\| \mathbf{Z}_{t,1}^{(:,j)} - \mathbf{Z}_{r,1}\mathbf{B}^{(:,j)}\right\|_{F}^{2} + \lambda \sum_{k = 1}^{p_{r}}\pmb{w}_{i}^{(k)}(\left\| \mathbf{B}_{+}^{(k,:)}\right\|_{2} + \left\| \mathbf{B}_{-}^{(k,:)}\right\|_{2}), \quad (2)\] + +where \(C_{i}\) is the set of all target genes in cluster \(i\) , \(\lambda\) is a penalty parameter related to \(N_{i,r}\) above, controlling the amount of regulators that will be selected, \(\pmb{w}_{i}\) is a vector of weights that can be cluster- specific and will be described below, \(\mathbf{B}^{(k,:)}\) refers to the \(k\) - th row in \(\mathbf{B}\) , and \(\mathbf{B}_{+}^{(k,:)}\) as well as \(\mathbf{B}_{- }^{(k,:)}\) are defined by setting all negative or all positive elements in \(\mathbf{B}^{(k,:)}\) to zero, respectively. + +The coop- Lasso selects which regulators, i.e. which rows of \(\mathbf{B}\) , are included in the model by setting the coefficients of the deselected groups to zero. In addition, the coop- Lasso aims for sign- coherence in each group and induces sparsity within groups, deselecting target genes not affected by some + +<--- Page Split ---> + +of the regulators. This means that typically the rows of the estimated coefficient matrix \(\mathbf{B}_{i,\mathrm{CL}}\) will have all positive or all negative sign. To determine the sign of each regulator, we assign \(\pmb{s}_{i}^{(k)} = \mathrm{sgn}\left(\sum_{j}\mathbf{B}_{i,\mathrm{CL}}^{(k,j)} / |C_{i}|\right)\) . The set of active regulators \(\mathrm{R}_{i}\) is equal to the indices of non- zero rows in \(\mathbf{B}_{i,\mathrm{CL}}\) . This way, the coop- Lasso in Eq. (2) provides an approximation to the optimization problem in Eq. (1) for fixed cluster membership. + +To make computations on a matrix of coefficients efficient, we apply over- relaxed Alternating Direction Method of Multipliers (ADMM) (Boyd et al, 2011) to the coop- Lasso problem, splitting the variables such that one set is specific to the loss and the other is specific to the penalty, leading to simple solutions for each separate sub- problem. To solve the sub- problem corresponding to the penalty, we use an explicit form of the proximal operator of the coop- Lasso given in Chiquet et al (2012). To speed- up convergence we compute ADMM step- length and over- relaxation parameters in an adaptive fashion as described in Xu et al (2017). + +Only the coefficients \(\mathbf{B}_{i,\mathrm{CL}}\) are optimized in Eq. (2). The variance parameters \(\sigma_{i,j}^{2}\) for \(j \in C_{i}\) are treated as known and weights are assumed to be given. We use a plug- in estimate for the variance parameters, which is computed using an ordinary least squares (OLS) estimate \(\mathbf{B}_{i,\mathrm{OLS}}\) of the regression coefficients of \(\mathbf{Z}_{t,1}^{(:,C_{i})}\) on \(\mathbf{Z}_{r,1}\) . Variances are then estimated in an unbiased way as + +\[\sigma_{i,j}^{2} = \frac{1}{n_{1} - p_{r}}\left\| \mathbf{Z}_{t,1}^{(:,j)} - \mathbf{Z}_{r,1}\mathbf{B}_{i,\mathrm{OLS}}^{(:,j)}\right\|_{2}^{2} \quad (3)\] + +To debias the estimated coefficients in Eq. (2), weights are selected in a fashion similar to the adaptive Lasso (Zou, 2006) before estimation of \(\mathbf{B}_{i}\) . Setting the weights to + +\[\pmb{w}_{i}^{(k)} = \sqrt{|C_{i}| / ||\mathbf{B}_{i,\mathrm{OLS}}^{(k,:)}||_{2}} \quad (4)\] + +improves the selection of regulators. + +Determining cluster membership. In Step 2, cluster membership is re- allocated, based on the updated cluster structure determined in Step 1. To do so requires the estimation of non- negative coefficients \(\mathbf{B}_{i}\) and residual variances \(\sigma_{i,j}^{2}\) for all clusters and target genes. Previously, these were only estimated for the target genes contained in each cluster. In addition, re- estimation of the coefficients without penalization for the selected regulators and their signs removes the bias the coop lasso introduced and enforces the chosen signs. Finally, the updated cluster membership matrix \(\mathbf{\Pi}\) is + +<--- Page Split ---> + +computed. + +Given \(\mathbf{R}_{i}\) and \(\pmb{s}_{i}\) , we used sign- constrained linear regression using non- negative least squares (NNLS) (Meinshausen, 2013) to determine the coefficients \(\mathbf{B}_{i}\) for each cluster. To do so, the following optimization problem was solved + +\[\mathbf{B}_{i} = \arg \min_{\mathbf{B}}\frac{1}{2}\left\| \mathbf{Z}_{t,1} - \mathbf{Z}_{r,1}^{(\cdot ,\mathbf{R}_{i})}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}\right\|_{F}^{2}\mathrm{~such~that~}\mathbf{B}\geq \mathbf{0}, \quad (5)\] + +where the inequality is considered element- wise. To compute the NNLS coefficients efficiently, we implemented the algorithm described in Nguyen and Ho (2017). We modified the algorithm to be able to perform computations on a matrix of responses instead of on a single vector. To avoid unnecessary computation, responses are excluded from the computations once they reach the desired convergence criterion. + +The variance parameters \(\sigma_{i,j}^{2}\) for each \(i = 1,\ldots ,K\) and all \(j = 1,\ldots ,p_{t}\) are re- estimated as + +\[\sigma_{i,j}^{2} = \frac{1}{n_{1} - |\mathbf{R}_{i}|}\left\| \mathbf{Z}_{t,1}^{(\cdot ,j)} - \mathbf{Z}_{r,1}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}_{i}^{(\cdot ,j)}\right\|_{2}^{2} \quad (6)\] + +In the following, computations were performed on the second data split \(\mathbf{Z}_{t,2}\) and \(\mathbf{Z}_{r,2}\) to avoid bias towards the most complex regulatory programs. + +Rag bag clustering is used to identify target genes that do not fit well in any cluster. To do so, the predictive \(R^{2}\) - value, denoted as \(R_{i,j}^{2}\) , is computed for each target gene and cluster from the residuals of predicting \(\mathbf{Z}_{t,2}\) from \(\mathbf{Z}_{r,2}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}_{i}\) . The best predictive \(R_{i,j}^{2}\) across clusters for each target gene is recorded. If this best value is below a user- specified threshold \(\tau\) , then the gene is considered noise and badly predicted within all clusters. It is then placed in a noise cluster/rag bag. Only the remaining target genes are considered in the following steps. + +To include prior knowledge of target genes that have a biological relationship, the algorithm allows the user to supply an indicator matrix \(\mathbf{J}\) of size \(q\times q\) such that \(\mathbf{J}^{(i,j)} = 1\) if genes \(i\) and \(j\) have a biological relationship, and zero otherwise. It is assumed that \(\mathbf{J}^{(i,i)} = 0\) to simplify computations below. By providing gene symbols for the target genes in \(\mathbf{Z}_{t}\) and the genes in \(\mathbf{J}\) , the algorithm determines the genes for which prior information is available. It is not required that the provided sets are equal as long as there is overlap. Assume for sake of notation that prior information is provided for all target genes in \(\mathbf{Z}_{t}\) . + +<--- Page Split ---> + +For a fixed target gene \(j\) , \(\mathbf{J}^{(j,:)}\) contains 1's for those target genes which have a biological relationship with gene \(j\) . Given the current cluster membership matrix \(\mathbf{\Pi}\) , compute fractions \(f_{i,j} = \mathbf{J}^{(j,:)}\mathbf{\Pi}^{(i,:)^{\top}} / \sum_{g}\mathbf{\Pi}^{(i,g)}\) to encode the biological evidence supporting gene \(j\) to be in cluster \(i\) . In case cluster \(i\) is empty, set \(f_{i,j} = 0\) . These fractions are then normalized across clusters as \(p_{i,j} = f_{i,j} / \sum_{c}f_{c,j}\) . For numerical reasons, log- probabilities are used below. To avoid taking the log of zero, a small baseline parameter \(\alpha = 10^{- 6}\) is added to each \(f_{i,j}\) before normalization. + +Given \(\mathrm{R}_{i}\) , \(\mathbf{s}_{i}\) , \(\mathbf{B}_{i}\) , and \(\sigma_{i,j}^{2}\) , the likelihood for target gene \(j\) across clusters is computed for observation \(l\) as + +\[\mathbf{L}_{j}^{(i,l)} = \frac{1}{\sqrt{2\pi\sigma_{i,j}^{2}}}\exp \left(-\frac{1}{2\sigma_{i,j}^{2}}\left(\mathbf{Z}_{t,2}^{(l,j)} - \mathbf{Z}_{r,2}^{(l,\mathrm{R}_{i})}\mathrm{diag}(\mathbf{s}_{i})\mathbf{B}_{i}^{(:,j)}\right)^{2}\right). \quad (7)\] + +These are then normalized by setting \(\mathbf{L}_{j}^{(i,l)}\leftarrow \mathbf{L}_{j}^{(i,l)} / \sum_{c}\mathbf{L}_{j}^{(c,l)}\) . To update the cluster membership for target gene \(j\) we then compute votes + +\[\pmb{v}_{j}^{(l)} = \arg \max_{i}\left[(1 - \mu)\log \mathbf{L}_{j}^{(i,l)} + \mu \log p_{i,j}\right] \quad (8)\] + +for each observation \(l\) and assign target gene \(j\) to the cluster which receives the majority of votes. The parameter \(\mu \in [0,1]\) can be used to control the strength of the prior on the allocation process. All target genes are processed in a random ordering to avoid introducing bias. Prior fractions for each gene are computed in each iteration using the previously updated cluster assignments as well as old cluster assignments for genes that have not been updated yet. + +Determining convergence A history of the cluster membership matrices is kept and the algorithm is stopped when the current \(\mathbf{\Pi}\) at cycle \(c\) computed in Step 2 is equal to the cluster membership matrix in a previous cycle \(c_{0}\) . At this point, the algorithm would enter a loop of length \(c - c_{0}\) and is therefore exited. We consider the algorithm as converged if \(c - c_{0} = 1\) . If \(c - c_{0} > 1\) , the algorithm has found an unstable cluster configuration and results for each possible configuration within the loop are returned. + +<--- Page Split ---> + +Regulatory table In addition to all estimated quantities, the algorithm returns a \(p_{r}\times K\) regulatory table \(\mathbf{T}\) . It is computed as follows for \(k = 1,\ldots ,p_{r}\) and \(i = 1,\ldots ,K\) + +\[\mathbf{T}^{(k,i)} = \frac{1}{|C_{i}|}\sum_{j\in C_{i}}\mathbf{B}_{i,\mathrm{CL}}^{(k,j)}\qquad \mathrm{if}\quad \mathrm{median}\left\{\mathrm{corr}(\mathbf{Z}_{t,1}^{(:,j)},\mathbf{Z}_{r,1}^{(:,k)})\right\} >0.025 \quad (9)\] + +or \(\mathbf{T}^{(k,i)} = 0\) otherwise. It summarizes the average effect of an active regulator within a module, given that the regulator is sufficiently correlated with the target genes in that module. + +## Internal validation measures + +In the following, we will consider three different types of predictive \(R^{2}\) - values. First, predictive \(R^{2}\) per module, written \(R_{i}^{2}\) for cluster \(i\) , is the predictive \(R^{2}\) for a given module \(i\) computed on the second data split. Second, predictive \(R^{2}\) per module excluding regulator \(k\) , written \(R_{i, - k}^{2}\) , is the predictive \(R^{2}\) for a given module \(i\) conditional on regulator \(k\) not being associated with the module. Thirdly, predictive \(R^{2}\) per module and target gene, written \(R^{2}i,j\) , is the predictive \(R^{2}\) for target gene \(j\) computed within the regulatory program of module \(i\) . The latter is also called cross- cluster predictive \(R^{2}\) for target gene \(j\) . + +Guidance on selection of the penalty parameter To evaluate the clustering results we introduce two performance scores. The aim of our algorithm is to associate modules with their linearly most predictive regulators. As a measure of clustering quality, it is therefore natural to consider the predictive \(R^{2}\) per module, computed on the second data split with the coefficients \(\mathbf{B}_{i}\) for the selected regulators re- estimated on the first data split as in Eq. (5). We therefore compute + +\[R_{i}^{2} = \left\| \mathbf{Z}_{t,2}^{(:,C_{i})} - \mathbf{Z}_{r,2}^{(:,R_{i})}\mathrm{diag}(\pmb {s}_{i})\mathbf{B}_{i}\right\|_{F}^{2} / \sum_{j\in C_{i}}\left\| \mathbf{Z}_{t,2}^{(:,j)} - \frac{1}{n_{2}}\sum_{l = 1}^{n_{2}}\mathbf{Z}_{t,2}^{(l,j)}\right\|_{2}^{2}. \quad (10)\] + +In addition, we compute the importance of each regulator within a module as follows. The regulator and sign sets without regulator \(k\) , denoted as \(\mathbf{R}_{i, - k} = :\{g:g\in \mathbf{R}_{i},g\neq k\}\) and \(\pmb{s}_{i, - k}\) , are used to re- estimate the coefficients for module \(i\) . We then compute \(R_{i, - k}^{2}\) as the predictive \(R^{2}\) value on this reduced regulatory program and compute the importance of regulator \(k\) in module \(i\) as \(I_{i,k} = (R_{i}^{2} - R_{i, - k}^{2}) / R_{i}^{2} = 1 - R_{i, - k}^{2} / R_{i}^{2}\) . This is the ratio of the semi- partial correlation of the expression profile of regulator \(k\) with the expression profiles in module \(i\) to the overall \(R^{2}\) of module \(i\) + +<--- Page Split ---> + +with all regulators. A regulator that is more influential in predicting the target genes in module \(i\) will have larger importance. Importance values are typically less or equal to 1 and can be negative, which indicates that a regulator was introducing more noise than it compensated for in predictive strength. A decrease in \(R^{2}\) per module with an increase in the penalization parameter \(\lambda\) is expected, due to selection of less regulators and therefore less degrees of freedom in the linear models associated with each module. Importance is the ratio of the squared marginal correlation of a regulator with the target genes in a module to the overall \(R^{2}\) of that module. This implies that an increase of importance values is expected to concur with a decrease in number of selected regulators. Therefore, importance is expected to increase with an increase in the penalization parameter \(\lambda\) . Selection of the penalty parameter can therefore be guided by balancing these two scores. Predictive \(R^{2}\) should be high while importance should neither be too low nor too high, since the latter is typically indicative of too few selected regulators. + +Guidance on selection of the number of modules Selection of the number of modules can be aided by cross- cluster predictive \(R^{2}\) . For each final module's selected regulators and signs, coefficients are re- estimated for all target genes. These are then used to compute the predictive \(R^{2}\) for each target gene in all available modules. Denote these values as \(R_{i,j}^{2}\) for target gene \(j\) when computed with regulators from module \(i\) . If there are \(K\) modules, this will result in a \(K \times p_{t}\) matrix. We then define the following score for each target gene \(j\) which has been clustered in module \(i\) + +\[a_{j} = R_{i,j}^{2},\quad b_{j} = \max (0,\max_{c\neq i}R_{c,j}^{2}),\quad S_{j} = \frac{a_{j} - b_{j}}{\max (a_{j},b_{j})}.\] + +Due to the similarity with the well- known silhouette value (Rousseeuw, 1987), we call \(S_{j}\) the silhouette score for target gene \(j\) . The average silhouette score across all target genes gives a rough measure for clustering quality and can be used as a straight- forward tool to compare different module counts. A larger average silhouette score indicates that target genes on average are better located within modules. To determine the optimal number of modules, average silhouette score should be considered jointly with predictive \(R^{2}\) per module. A good clustering achieves high values in both scores. + +<--- Page Split ---> + +## External validation measures + +To evaluate the performance of simulations, we compared the groundtruth to the estimated clustering in multiple ways. + +Adjusted Rand index To determine overall clustering performance, we use the adjusted Rand index (Hubert and Arabie, 1985), a well- known similarity measure between two clusterings. The standard adjusted Rand index ARI does not account for rag bag clustering. We therefore consider a modified version of the adjusted Rand index + +\[\mathrm{ARI}\cdot \frac{\#\mathrm{non - noise~target~genes}}{\#\mathrm{all~target~genes}}. \quad (11)\] + +Regulator selection To answer whether scRegClust selects the right regulators for each module we cannot compare regulators associated with each module directly, since estimated modules might not match clearly with the groundtruth. We therefore check correct selection of regulators for each target gene, as these are easily comparable with the groundtruth. For easier comparison we compute two measures: + +1. True positive rate, as the proportion of correctly selected regulators, + +2. False positive rate, as the proportion of incorrectly selected regulators. + +Both of these measures are between 0 and 1 with larger values being better. + +Implementation The package was implemented in R (R Core Team, 2022). Computationally expensive parts were written in C++ using the packages Rcpp (Eddelbuettel and Francois, 2011) and RcppEigen (Bates and Eddelbuettel, 2013). The packages igraph (Csardi and Nepusz, 2006) and ggplot2 (Wickham, 2016) were used for visualization. The R package clusterCrit was used to compute the Rand index. + +## Knockdown experiments + +Target gene- editing by RNP complex delivery. As regulators of state 5, we chose DDR1 and SOX6. To disrupt gene function we delivered U3065MG cells between passage 16- 20 with SpCas9 + +<--- Page Split ---> + +2NLS and target specific multiguide (Synthego Gene Knockout Kit V2). The RNP complex was delivered using 4D-NucleoporatorTM X-unit and an optimized protocol with SF Cell line nucleofection kit S (Lonza, V4XC-2032). RNP complex formation was carried out as per the manufacturer's protocol. Briefly, RNP complex formation was carried out by mixing Cas9-2NLS and sgRNA in a molar ratio of 1:9 at room temperature, followed by adding the estimated amount of U3065MG cells dissolved in nucleofection solution and supplement 1. Experimental control used included non-targeting scrambled, Cas9 only, and mock (nucleofection of cells+RNP complex with no electric pulse). Nucleofection for all experiments was done using optimized settings in SF solution and program CA-137, and no major cell death was observed using these conditions. + +Table 3: sgRNA sequence information. + +
TargetSpeciessgRNA sequences (multi-guide)
DDR1HumanUGGAGAGCAGUGACGGGGAU
CUGCAAGUACUCCUCCUCCU
SOX6HumanGGCCCCCGCAUGCCGUCCC
UAUGGGGUGCAGAGGCAGAU
UUCCCUUGAGGUAAAUCUU
Non-targeting scrambled controlAAUUGGAGAGUGGGCUUGCU
GCACUACCAGAGCUAAACUCA
+ +Genotyping and Sanger sequencing. Seven days post- nucleoporation, edited cells and non non- targeting scrambled control group were harvested. To evaluate gene editing, genomic DNA was isolated from all samples, and PCR amplification of the edited region was performed, followed by Sanger sequencing of the amplicon. Genotyping PCR was performed using Phusion hot start II high- fidelity PCR master mix (Thermo Scientific, F- 565L), on an Applied Biosystems miniAmp thermocycler. PCR cycling conditions are: Initial denaturation 980C (5min, 1cycle); amplification (30 cycles)- denaturation 980C (1min), annealing 620C (30s), extension 720C (30s); final extension (1cycle) 720C (10min). PCR amplicons from non- targeting and target- specific samples were purified using the Macherey- Nagel PCR clean- up kit (NucleoSpinTM PCR Clean- up, Cat 740609.250). PCR amplicons were sequenced using a single- end read at Eurofins Genomics (TubeSeq Supreme service, Germany). Analysis of the Sanger sequencing traces was performed using the freely available web tool "Inference of CRISPR edits" ("ICE", https://ice.synthego.com/) by entering the trace files (.ab1 files) from the edited samples and the non- targeting scrambled control. + +<--- Page Split ---> + + +Table 4: Primer sequences used. + +
Target:DDR1
Primer pair for ampliconForward 5'-GGACCTCTACTTCCCCTCCA-3'
Reverse 5'- GTACAGCTGGAAGTGAGGAGA-3'
Sanger sequencing primer5'-GTACAGCTGGAAGTGAGGAGA-3'
Target: SOX6
Primer pair for ampliconForward 5'-GGTAGTTGTTTCATGATCTACACAA-3'
Reverse 5'-GGGTTTAGAGTGGACCAC-3'
Sanger sequencing primer5'-CACAAAAATAACTCTAAGGTCATCTATTTTC-3'
+ +Temozolomide treatment. Cells from nontargeting scrambled control and target- specific edited samples were seeded to 384 well plates coated (Greiner blackwell optically clear bottom) with laminin, 7 days post nucleoporation (the editing from different batches of the experiment confirmed efficient knockdown on day 7). Twenty- four hours after seeding, cells were treated with temozolomide over a dose range from 750uM to 1.25uM. Treated cells were placed back into the incubator and an Alamar blue reading was taken 96 hours after treatment using a CLARIOstar plate reader (BMG Labtech). To know the effect of TMZ on viability fluorescence reading was measured, background subtracted values were normalized to DMSO control, and dose- response curves were plotted using GraphPad Prism. + +## Drug combination treatments + +Cell culture The human glioblastoma cells U3065MG were obtained from the Human Glioma Cell Culture (HGCC) Biobank at Uppsala University (Xie et al, 2015). The cells were cultured in a mixture of Neurabasa1 (Gibco, #21103- 049) and DMEM/F12 (Gibco, #31331- 028) growth medium in 1:1 ratio, 1x B27, without Vitamin A (Gibco, #12587001), 1x N2 (Gibco, #17502001) and \(1\%\) Pen/Strep (Sigma- Aldrich, #P0781). The cells were grown as adherent culture on a laminin coated flasks (Sigma Aldrich, #L2020- 1MG), and were detached for splitting and seeding by TypIE without phenol red (Gibco, #12604039). The growth medium was supplemented with \(10\mathrm{ng / mL}\) recombinant human FGF- basic (Peprotech, #100- 18B) and \(10\mathrm{ng / mL}\) recombinant human EGF (Peprotech, #AF- 100- 15) each passage. The cells were regularly tested for Mycoplasma by either MycoAlert assay (Lonza, #LT07- 418) or MycoStrip (InvivoGen, #rep- mys- 50). + +Cell viability assay U3065MG cells were seeded in a laminin coated 384- well plates (Thermo Scientific, #142761) at a density of 2,000 cells/well at a volume of 0.04 mL/well. The plates were + +<--- Page Split ---> + +centrifuged briefly at \(200\mathrm{xg}\) to collect the cells at the bottom of the plates and incubated for 24 h at \(37^{\circ}\mathrm{C}\) in humidified atmosphere with \(5\%\) CO2. Afterwards, the cells were either treated with temozolomide (Selleckchem, #S1237) or one of dasatinib (Selleckchem, #S1021). All compounds were dissolved in DMSO to a \(10\mathrm{mM}\) stock solution. The drugs were dispensed with D300e digital dispenser (Tecan) without additional dilution of the stock solution. Afterwards, the plates were centrifuged briefly at \(200\mathrm{xg}\) , and were incubated for 48 or 72 h at \(37^{\circ}\mathrm{C}\) in humidified atmosphere with \(5\%\) CO2. Cell viability was measured by the Alamar Blue assay (Invitrogen, #DAL1100), which was added 16 h prior the readout. After the 72 h time point, the growth medium was aspirated, and the cells were washed once with sterile PBS before addition of fresh medium. The cells were treated for additional 72 h with a combination between teamozolomide and dasatinib in a \(7\times 7\) dose- response matrix. 16 h prior the end point of the reaction, alamar blue was added into each well of the plates. The reduction of resazurin to resorufin was used as a proxy for determination of the cell viability. The raw measurements of the treatments were normalized to the DMSO control. The BLISS coefficient and the sensitivity of the combinations were calculated with SynergyFinder (Zheng et al, 2022) in R (R Core Team, 2022). + +## Data availability + +No data was generated for this work. The R- package can be found on GitHub at https://github.com/sven- nelander/scregclust. + +## Acknowledgements + +We thank the Swedish Cancer Society, Swedish Childhood Cancer Foundation, Swedish Research Council and the Swedish Strategic Research Foundation for financial support. Some of the computations were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at the National Supercomputer Centre (NSC), Linköping University partially funded by the Swedish Research Council through grant agreement no. 2022- 06725. + +<--- Page Split ---> + +## Author contributions + +SN conceived the study. IL and FH developed the algorithm. FH wrote the R- package and IL performed the computational analyses, with support from RJ and SN. GP and AK planned and conducted the combination treatment experiments. SK planned and conducted the knockdown experiments. IL and FH wrote the first version of the paper and all authors assisted in editing. + +## Conflict of interest + +The authors declare no conflict of interest. + +<--- Page Split ---> + +## Figures and Figure Legends + +## Figure 1. A regulatory-driven clustering method. + +A A schematic overview of the algorithm pipeline from gene expression data to regulatory networks. B UMAP embedding of the PBMC cells colored according to immune cell type assignment. C The regulatory table inferred by scRegClust. Rows correspond to immune cell types and columns to regulators (TFs). The scale goes from high positive regulation (red) to high negative regulation (blue), as indicated by the key. D Venn diagram showing the overlap in identified regulators between SCENIC+ and scRegClust. E Regulator enrichment in the Human Protein Atlas compared to a baseline of all genes in the atlas. F Comparison between scRegClust and SCENIC+. + +## Figure 2. Method validation and analysis tools. + +Results from a representative run of scRegClust on the data from simulation setup A (see the Supplementary Material for details on data generation). + +A Boxplots of predictive \(R^2\) per module and importance per regulator shown across a progression of penalization parameters. Dashed lines indicate a region of solutions that demonstrates our selection rule. B Average true positive rate (TPR) shown against average false positive rate (FPR) in a ROC curve, illustrating the quality of groundtruth regulator identification. TPR and FPR are computed per target gene and then averaged. The corresponding penalization parameter is shown with each average. A table of ajusted Rand indices between the estimated clustering and the true clustering is shown as an inset. C Relative importance of the regulators within each module as estimated by scRegClust within each penalization level. Each plot represents one of the five estimated modules and each line shows the result for a penalization parameter. Colored tiles indicate that the regulator was included in the final model for the respective module. Green dots above the regulators for each module indicate regulators used during simulation of the data. The dashed lines correspond to the dashed lines in part A. They illustrate which regulators would be selected when the penalization parameter was chosen by our selection rule. + +<--- Page Split ---> + +Figure 3. scRegClust predicts a way of potentiating temozolomide treatment in U3065MG cells + +A Flowchart of analysis. + +B The merged regulatory table from scRegClust, with gene modules as columns and regulators as rows. Bottom panel display enrichments for each module against gene sets from (Larsson et al, 2021) and (Neftel et al, 2019)). + +C Schematic of the predicted regulation of state 5. PDGFRA, DDR1, ERBB3 and SOX6 positively regulated state 5 and should be knocked down to block transitions to state 5. YBX1 negatively regulated transitions to state 5 and should be overexpressed to block transitions to state 5. + +D Dose response curves for TMZ- treated cells with CRISPR/Cas9- mediated knockdown of DDR1 (top) and SOX6 (bottom). + +E Schematic of the three combination treatment arms. Cells were either pre- treated with tyrosine kinase inhibitors (TKIs), TMZ or no pre- treatment, followed by combination treatment with TMZ and TKIs. + +F Boxplot showing the mean bliss synergy score (BSS) for the three treatment arms, all replicates (top) and synergy landscape for the two combination experiments where the highest synergy scores were obtained (bottom). + +## Figure 4. The regulatory landscape of neuro-oncology. + +A Schematic overview of the analysis. + +B Middle panel is the regulatory table from scRegClust, with modules as columns and regulators (TFs) as rows. Top panel are annotation bars indicating what type and study each module originate from. Meta- modules (Box 1) are indicated by color in the bar below the middle panel. Bottom panel display enrichments for each module against a database of neuro- oncology related gene sets, derived from studies indicated by PubMed ID (PMID). Abbreviations: CT = cellular tumor, MVP = microvascular proliferation, GPM = glycolytic/plurimetabolic, pan = pseudopalisading cells around necrosis, NPC = neural progenitor cells, PPR = proliferating progenitor cells, IT = infiltrative tumor, OPC = oligodendrocyte progenitor cells, tRG = truncated radial glia. + +<--- Page Split ---> + +## Box 1. The regulatory meta-modules. + +Description of the meta- modules defined in Figure 4. The colors to the left correspond to the middle color bar in Figure 4. The regulators listed regulate at least two of the individual modules included in the meta- module and all diseases represented in the meta- module are indicated by a filled box. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1 - A regulatory-driven clustering method.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2 - Method validation and analysis tools.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3 - scRegClust predicts a way of potentiating temozolomide treatment in U3065MG cells.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4 - The regulatory landscape of neuro-oncology.
+ +<--- Page Split ---> + + +Box 1 - The regulatory meta-modules + +
Box 1 - The regulatory meta-modules
Meta-moduleInterpretationRegulatorsNormaGBMNBNB
1Cycling cellsHMGB1/2+, CENPA+, DEK+
2Necrosis, hypoxia, apoptosisYBX1-, ID3+
3Non-proliferating NPC2-like cellsHMGB1/2-, NEUROD1/2+, DEK+, SOX11+, SOX4+, TCF4+
4Astrocyte-like cellsID3+, ID4+, HOPX+, ZFP36L1+, MEF2C+
5Microglial-like mesenchymal cellsSPI1+, IRF8+, MAF+, ZFP36+, NME2+
6Stress responseFOXJ1+, JUN+, JUND+, FOS+, JUNB+, HMGN3+, NR4A1+
7Cycling cellsHMGB1/2+, CENPA+/-, DEK+, NEUROD1-
8FOXJ1-drivenFOXJ1+
9NB non-adrenergic cellsZNF90+, MEIS2+, MYC+, STAT1+
10FOXP2, RORA-driven infiltrative tumorFOXP2+, RORA+
11MYT1L-driven infiltrative tumorMYT1L+
12SOX17-driven mvpERG+, FLI+, ETS1+, EPAS1+, FOXQ1+, SOX17+, ZFP36+, ID3+
13OPC-like cellsOLIG1+, MYRF+, SOX10+, TSC22D4+, NKX6-2+/(-)/SOX6+
14NB adrenergic cellsZNF90-, MEIS2-, MYC-
15Mvp-signature mesenchymal cellsFOXS1+, HEY1+, AEBP1+, TBX18+, NME2+, MEF2C+, TSC22D4+
+ +Box 1 - The regulatory meta- modules. + +<--- Page Split ---> + +## References + +Aibar S, lez Blas CB, Moerman T, Huynh- Thu VA, Imrichova H, Hulselmans G, Rambow F, Marine JC, Geurts P, Aerts J, van den Oord J, Atak ZK, Wouters J, Aerts S (2017) SCENIC: single- cell regulatory network inference and clustering. Nat Methods 14: 1083–1086Arthur D, Vassilvitskii S (2007) K- means++: the advantages of careful seeding. In Proceedings of the 18th Annual ACM- SIAM Symposium on Discrete AlgorithmsBates D, Eddelbuettel D (2013) Fast and Elegant Numerical Linear Algebra Using the RcppEigen Package. Journal of Statistical Software 52: 1–24Bhat KPL, Balasubramaniyan V, Vaillant B, Ezhilarasan R, Hummelink K, Hollingsworth F, Wani K, Heathcock L, James JD, Goodman LD, Conroy S, Long L, Lelic N, Wang S, Gumin J, Raj D, Kodama Y, Raghunathan A, Olar A, Joshi K, et al (2013) B promotes radiation resistance in glioblastoma. Cancer Cell 24: 331–346Bohrer C, Pfurr S, Mammadzada K, Schildge S, Plappert L, Hils M, Pous L, Rauch KS, Dumit VI, Pfeifer D, Dengjel J, Kirsch M, Schachtrup K, Schachtrup C (2015) The balance of Id3 and E47 determines neural stem/precursor cell differentiation into astrocytes. EMBO J 34: 2804–2819Boyd S, Parikh N, Chu E, Peleato B, Eckstein J (2011) Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers. Foundations and Trends in Machine Learning 3: 1–122Butler A, Hoffman P, Smibert P, Papalexi E, Satija R (2018) Integrating single- cell transcriptomic data across different conditions, technologies, and species. Nature BiotechnologyCarro MS, Lim WK, Alvarez MJ, Bollo RJ, Zhao X, Snyder EY, Sulman EP, Anne SL, Doetsch F, Colman H, Lasorella A, Aldape K, Califano A, Iavarone A (2010) The transcriptional network for mesenchymal transformation of brain tumours. Nature 463: 318–325Chang ZF, Huang DY, Chi LM (1998) Serine 13 is the site of mitotic phosphorylation of human thymidine kinase. J Biol Chem 273: 12095–12100Chen S, Mar JC (2018) Evaluating methods of inferring gene regulatory networks highlights their lack of performance for single cell gene expression data. BMC Bioinformatics 19: 232 + +<--- Page Split ---> + +Chiquet J, Grandvalet Y, Charbonnier C (2012) Sparsity with sign- coherent groups of variables via the cooperative- Lasso. The Annals of Applied Statistics 6 + +Couturier CP, Ayyadhury S, Le PU, Nadaf J, Monlong J, Riva G, Allache R, Baig S, Yan X, Bourgey M, Lee C, Wang YCD, Wee Yong V, Guiot MC, Najafabadi H, Misic B, Antel J, Bourque G, Ragoussis J, Petrecca K (2020) Single- cell RNA- seq reveals that glioblastoma recapitulates a normal neurodevelopmental hierarchy. Nat Commun 11: 3406 + +Csardi G, Nepusz T (2006) The igraph software package for complex network research. InterJournal Complex Systems: 1695 + +Cuddapah VA, Robel S, Watkins S, Sontheimer H (2014) A neurocentric perspective on glioma invasion. Nat Rev Neurosci 15: 455- 465 + +Darmanis S, Sloan SA, Croote D, Mignardi M, Chernikova S, Samghababi P, Zhang Y, Neff N, Kowarsky M, Caneda C, Li G, Chang SD, Connolly ID, Li Y, Barres BA, Gephart MH, Quake SR (2017) Single- Cell RNA- Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep 21: 1399- 1410 + +Eddelbuettel D, Francois R (2011) Rcpp: Seamless R and C++ Integration. Journal of Statistical Software 40: 1- 18 + +Ellison JA, de Vellis J (1994) Platelet- derived growth factor receptor is expressed by cells in the early oligodendrocyte lineage. J Neurosci Res 37: 116- 128 + +Eze UC, Bhaduri A, Haeussler M, Nowakowski TJ, Kriegstein AR (2021) Single- cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia. Nat Neurosci 24: 584- 594 + +Ferdous A, Singh S, Luo Y, Abedin MJ, Jiang N, Perry CE, Evers BM, Gillette TG, Kyba M, Trojanowska M, Hill JA (2021) Fli1 Promotes Vascular Morphogenesis by Regulating Endothelial Potential of Multipotent Myogenic Progenitors. Circ Res 129: 949- 964 + +Filbin MG, Tirosh I, Hovestadt V, Shaw ML, Escalante LE, Mathewson ND, Neftel C, Frank N, Pelton K, Hebert CM, Haberler C, Yizhak K, Gojo J, Egervari K, Mount C, van Galen P, Bonal DM, Nguyen QD, Beck A, Sinai C, et al (2018) Developmental and oncogenic programs in H3K27M gliomas dissected by single- cell RNA- seq. Science 360: 331- 335 + +<--- Page Split ---> + +Gangoso E, Southgate B, Bradley L, Rus S, Galvez- Cancino F, McGivern N, \(\zeta\) E, Kapourani CA, Byron A, Ferguson KM, Alfazema N, Morrison G, Grant V, Blin C, Sou I, Marques- Torrejon MA, Conde L, Parrinello S, Herrero J, Beck S, et al (2021) Glioblastomas acquire myeloid- affiliated transcriptional programs via epigenetic immunoediting to elicit immune evasion. Cell 184: 2454- 2470 + +Gavish A, Tyler M, Simkin D, Kovarsky D, Gonzalez Castro LN, Halder D, Chanoch- Myers R, Laffy J, Mints M, Greenwald AR, Wider A, Tal R, Spitzer A, Hara T, Tirosh A, Puram SV, Suva ML, Tirosh I (2021) The transcriptional hallmarks of intra- tumor heterogeneity across a thousand tumors. bioRxiv + +González- Blas C, De Winter S, Hulseimans G, Hecker N, Matetovici I, Christiaens V, Poovathingal S, Wouters J, Aibar S, Aerts S (2023) SCENIC+: single- cell multionic inference of enhancers and gene regulatory networks. Nat Methods 20: 1355- 1367 + +Hafemeister C, Satija R (2019) Normalization and variance stabilization of single- cell RNA- seq data using regularized negative binomial regression. Genome Biol 20: 296 + +Hamed AA, Kunz DJ, El- Hamamy I, Trinh QM, Subedar OD, Richards LM, Foltz W, Bullivant G, Ware M, Vladoiu MC, Zhang J, Raj AM, Pugh TJ, Taylor MD, Teichmann SA, Stein LD, Simons BD, Dirks PB (2022) A brain precursor atlas reveals the acquisition of developmental- like states in adult cerebral tumours. Nat Commun 13: 4178 + +Hara T, Chanoch- Myers R, Mathewson ND, Myskiw C, Atta L, Bussema L, Eichhorn SW, Greenwald AC, Kinker GS, Rodman C, Gonzalez Castro LN, Wakimoto H, Rozenblatt- Rosen O, Zhuang X, Fan J, Hunter T, Verma IM, Wucherpfennig KW, Regev A, Suva ML, et al (2021) Interactions between cancer cells and immune cells drive transitions to mesenchymal- like states in glioblastoma. Cancer Cell 39: 779- 792 + +Hovestadt V, Smith KS, Bihannic L, Filbin MG, Shaw ML, Baumgartner A, DeWitt JC, Groves A, Mayr L, Weisman HR, Richman AR, Shore ME, Goumnerova L, Rosencrance C, Carter RA, Phoenix TN, Hadley JL, Tong Y, Houston J, Ashmun RA, et al (2019) Resolving medulloblastoma cellular architecture by single- cell genomics. Nature 572: 74- 79 + +Hu Y, Jiang Y, Behnam J, Ribeiro MM, Kalantzi C, Zhang MD, Lou D, ring M, Sharma N, Okawa S, Del Sol A, Adameyko I, Svensson M, Persson O, Ernfors P (2022) Neural network learning defines glioblastoma features to be of neural crest perivascular or radial glia lineages. Sci Adv 8: eabm6340 + +<--- Page Split ---> + +Hubert L, Arabie P (1985) Comparing partitions. Journal of Classification 2: 193- 218 + +Kamal A, Arnold C, Claringbould A, Moussa R, Servaas NH, Kholmatov M, Daga N, Nogina D, Mueller- Dott S, Reyes- Palomares A, Palla G, Sigalova O, Bunina D, Pabst C, Zaugg JB (2022) GRaNIE and GRaNPA: Inference and evaluation of enhancer- mediated gene regulatory networks applied to study macrophages. bioRxiv + +Kamimoto K, Hoffmann CM, Morris SA (2020) CellOracle: Dissecting cell identity via network inference and in silico gene perturbation. bioRxiv + +Kildisuite G, Kholosy WM, Young MD, Roberts K, Elmentaite R, van Hooff SR, Pacyna CN, Khabirova E, Piapi A, Thevanesan C, Bugallo- Blanco E, Burke C, Mamanova L, Keller KM, Langenberg- Ververgaert KPS, Lijnzaad P, Margaritis T, Holstege FCP, Tas ML, Wijnen MHWA, et al (2021) Tumor to normal single- cell mRNA comparisons reveal a pan- neuroblastoma cancer cell. Sci Adv 7 + +Kling T, Ferrarese R, n D, Johansson P, Heiland DH, Dai F, Vasilikos I, Weyerbrock A, rnsten R, Carro MS, Nelander S (2016) Integrative Modeling Reveals Annexin A2- mediated Epigenetic Control of Mesenchymal Glioblastoma. EBioMedicine 12: 72- 85 + +La Manno G, Siletti K, Furlan A, Gyllborg D, Vinsland E, Mossi Albiac A, Mattsson Langseth C, Khven I, Lederer AR, Dratva LM, Johnsson A, Nilsson M, Lönnerberg P, Linnarsson S (2021) Molecular architecture of the developing mouse brain. Nature 596: 92- 96 + +Larsson I, Dalmo E, Elgendy R, Niklasson M, Doroszko M, Segerman A, Jörnsten R, Westermark B, Nelander S (2021) Modeling glioblastoma heterogeneity as a dynamic network of cell states. Mol Syst Biol 17: e10105 + +Le Coz C, Nguyen DN, Su C, Nolan BE, Albrecht AV, Xhani S, Sun D, Demaree B, Pillarisetti P, Khanna C, Wright F, Chen PA, Yoon S, Stiegler AL, Maurer K, Garifallou JP, Rymaszewski A, Kroft SH, Olson TS, Seif AE, et al (2021) Constrained chromatin accessibility in PU.1- mutated agammaglobulinemia patients. J Exp Med 218 + +LeBlanc VG, Trinh DL, Aslanpour S, Hughes M, Livingstone D, Jin D, Ahn BY, Blough MD, Cairn- cross JG, Chan JA, Kelly JJP, Marra MA (2022) Single- cell landscapes of primary glioblastomas and matched explants and cell lines show variable retention of inter- and intratumor heterogeneity. Cancer Cell 40: 379- 392 + +<--- Page Split ---> + +Li YJ, He YF, Han XH, Hu B (2015) Dasatinib suppresses invasion and induces apoptosis in nasopharyngeal carcinoma. Int J Clin Exp Pathol 8: 7818- 7824 + +Luo W, Lin GN, Song W, Zhang Y, Lai H, Zhang M, Miao J, Cheng X, Wang Y, Li W, Wei W, Gao WQ, Yang R, Wang J (2021) Single- cell spatial transcriptomic analysis reveals common and divergent features of developing postnatal granule cerebellar cells and medulloblastoma. BMC Biol 19: 135 + +Martínez- Balbás MA, Bauer UM, Nielsen SJ, Brehm A, Kouzarides T (2000) Regulation of E2F1 activity by acetylation. EMBO J 19: 662- 671 + +Marumoto T, Honda S, Hara T, Nitta M, Hirota T, Kohmura E, Saya H (2003) Aurora- A kinase maintains the fidelity of early and late mitotic events in HeLa cells. J Biol Chem 278: 51786- 51795 + +Meinshausen N (2013) Sign- constrained least squares estimation for high- dimensional regression. Electronic Journal of Statistics 7 + +Milano V, Piao Y, LaFortune T, de Groot J (2009) Dasatinib- induced autophagy is enhanced in combination with temozolomide in glioma. Mol Cancer Ther 8: 394- 406 + +Neftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, Richman AR, Silverbush D, Shaw ML, Hebert CM, Dewitt J, Gritsch S, Perez EM, Gonzalez Castro LN, Lan X, Druck N, Rodman C, Dionne D, Kaplan A, Bertalan MS, et al (2019) An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell 178: 835- 849 + +Nguyen DK, Ho TB (2017) Accelerated anti- lopsided algorithm for nonnegative least squares. International Journal of Data Science and Analytics 3: 23- 34 + +Ocasio JK, Babcock B, Malawsky D, Weir SJ, Loo L, Simon JM, Zylka MJ, Hwang D, Dismuke T, Sokolsky M, Rosen EP, Vibhakar R, Zhang J, Saulnier O, Vladoiu M, El- Hamamy I, Stein LD, Taylor MD, Smith KS, Northcott PA, et al (2019) scRNA- seq in medulloblastoma shows cellular heterogeneity and lineage expansion support resistance to SHH inhibitor therapy. Nat Commun 10: 5829 + +Pham TH, Minderjahn J, Schmidl C, Hoffmeister H, Schmidhofer S, Chen W, Längst G, Benner C, Rehli M (2013) Mechanisms of in vivo binding site selection of the hematopoietic master transcription factor PU.1. Nucleic Acids Res 41: 6391- 6402 + +<--- Page Split ---> + +Pratapa A, Jalihal AP, Law JN, Bharadwaj A, Murali TM (2020) Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat Methods 17: 147–154 + +R Core Team (2022) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria + +Richards LM, Whitley OKN, MacLeod G, Cavalli FMG, Coutinho FJ, Jaramillo JE, Svergun N, Riverin M, Croucher DC, Kushida M, Yu K, Guilhamon P, Rastegar N, Ahmadi M, Bhatti JK, Bozek DA, Li N, Lee L, Che C, Luis E, et al (2021) Gradient of Developmental and Injury Response transcriptional states defines functional vulnerabilities underpinning glioblastoma heterogeneity. Nat Cancer 2: 157–173 + +Rousseeuw PJ (1987) Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20: 53–65 + +Schmitt MJ, Company C, Dramaretska Y, Barozzi I, hrig A, Kertalli S, mann M, Naumann H, Sanchez- Bailon MP, Hulsman D, Glass R, Squatrito M, Serresi M, Gargiulo G (2021) Phenotypic Mapping of Pathologic Cross- Talk between Glioblastoma and Innate Immune Cells by Synthetic Genetic Tracing. Cancer Discov 11: 754–777 + +Tirosh I, Venteicher AS, Hebert C, Escalante LE, Patel AP, Yizhak K, Fisher JM, Rodman C, Mount C, Filbin MG, Neftel C, Desai N, Nyman J, Izar B, Luo CC, Francis JM, Patel AA, Onozato ML, Riggi N, Livak KJ, et al (2016) Single- cell RNA- seq supports a developmental hierarchy in human oligodendroglioma. Nature 539: 309–313 + +Uhlen M, Fagerberg L, Hallstrom BM, Lindskog C, Oksvold P, Mardinoglu A, Sivertsson A, Kampf C, Sjostedt E, Asplund A, Olsson I, Edlund K, Lundberg E, Navani S, Szigyarto CA, Odeberg J, Djureinovic D, Takanen JO, Hober S, Alm T, et al (2015) Proteomics. Tissue- based map of the human proteome. Science 347: 1260419 + +Venkataramani V, Yang Y, Schubert MC, Reyhan E, Tetzlaff SK, mann N, Botz M, Soyka SJ, Beretta CA, Pramatarov RL, Fankhauser L, Garofano L, Freudenberg A, Wagner J, Tanev DI, Ratliff M, Xie R, Kessler T, Hoffmann DC, Hai L, et al (2022) Glioblastoma hijacks neuronal mechanisms for brain invasion. Cell 185: 2899–2917 + +Vilella E, Gas C, Garcia- Ruiz B, Rivera FJ (2019) Expression of DDR1 in the CNS and in myelinating oligodendrocytes. Biochim Biophys Acta Mol Cell Res 1866: 118483 + +<--- Page Split ---> + +Wang L, Babikir H, Müller S, Yagnik G, Shamardani K, Catalan F, Kohanbash G, Alvarado B, Di Lullo E, Kriegstein A, Shah S, Wadhwa H, Chang SM, Phillips JJ, Aghi MK, Diaz AA (2019) The Phenotypes of Proliferating Glioblastoma Cells Reside on a Single Axis of Variation. Cancer Discov 9: 1708–1719 + +Wang L, Jung J, Babikir H, Shamardani K, Jain S, Feng X, Gupta N, Rosi S, Chang S, Raleigh D, Solomon D, Phillips JJ, Diaz AA (2022) A single- cell atlas of glioblastoma evolution under therapy reveals cell- intrinsic and cell- extrinsic therapeutic targets. Nat Cancer 3: 1534–1552 + +Wang Q, Hu B, Hu X, Kim H, Squatrito M, Scarpace L, deCarvalho AC, Lyu S, Li P, Li Y, Barthel F, Cho HJ, Lin YH, Satani N, Martinez- Ledesma E, Zheng S, Chang E, é CG, Olar A, Lan ZD, et al (2017) Tumor Evolution of Glioma- Intrinsic Gene Expression Subtypes Associates with Immunological Changes in the Microenvironment. Cancer Cell 32: 42–56 + +Wang Z, Sun D, Chen YJ, Xie X, Shi Y, Tabar V, Brennan CW, Bale TA, Jayewickreme CD, Laks DR, Alcantara Llaguno S, Parada LF (2020) Cell Lineage- Based Stratification for Glioblastoma. Cancer Cell 38: 366–379 + +Weng Q, Wang J, Wang J, He D, Cheng Z, Zhang F, Verma R, Xu L, Dong X, Liao Y, He X, Potter A, Zhang L, Zhao C, Xin M, Zhou Q, Aronow BJ, Blackshear PJ, Rich JN, He Q, et al (2019) Single- Cell Transcriptomics Uncovers Glial Progenitor Diversity and Cell Fate Determinants during Development and Gliomagenesis. Cell Stem Cell 24: 707–723 + +Wickham H (2016) ggplot2: Elegant Graphics for Data Analysis. New York, NY: Springer Science+Business Media, LLC, 2 edition + +Xie XP, Laks DR, Sun D, Ganbold M, Wang Z, Pedraza AM, Bale T, Tabar V, Brennan C, Zhou X, Parada LF (2022) Quiescent human glioblastoma cancer stem cells drive tumor initiation, expansion, and recurrence following chemotherapy. Dev Cell 57: 32–46 + +Xie Y, Bergström T, Jiang Y, Johansson P, Marinescu VD, Lindberg N, Segerman A, Wicher G, Niklasson M, Baskaran S, Sreedharan S, Everlien I, Kastemar M, Hermansson A, Elfneh L, Libard S, Holland EC, Hesselager G, Alafuzoff I, Westermark B, et al (2015) The Human Glioblastoma Cell Culture Resource: Validated Cell Models Representing All Molecular Subtypes. EBioMedicine 2: 1351–1363 + +<--- Page Split ---> + +Xu Z, Figueiredo MAT, Yuan X, Studer C, Goldstein T (2017) Adaptive Relaxed ADMM: Convergence Theory and Practical Implementation. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) + +Yang H, Lee S, Lee S, Kim K, Yang Y, Kim JH, Adams RH, Wells JM, Morrison SJ, Koh GY, Kim I (2013) Sox17 promotes tumor angiogenesis and destabilizes tumor vessels in mice. J Clin Invest 123: 418- 431 + +Yuan M, Lin Y (2006) Model selection and estimation in regression with grouped variables. Journal of the Royal Statistical Society Series B Statistical Methodology 68: 49- 67 + +Zeitlin SG, Shelby RD, Sullivan KF (2001) CENP- A is phosphorylated by Aurora B kinase and plays an unexpected role in completion of cytokinesis. J Cell Biol 155: 1147- 1157 + +Zheng S, Wang W, Aldahdooh J, Malyutina A, Shadbahr T, Tanoli Z, Pessia A, Tang J (2022) SynergyFinder Plus: Toward Better Interpretation and Annotation of Drug Combination Screening Datasets. Genomics Proteomics Bioinformatics + +Zou H (2006) The Adaptive Lasso and Its Oracle Properties. Journal of the American Statistical Association 101: 1418- 1429 + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryTable1. xlsx supplementarytable2. xlsx supplementarytable3. xlsx appendixversion20231123. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9_det.mmd b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..85286115933cf2ea12643408fd3ad62eeecfab63 --- /dev/null +++ b/preprint/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9/preprint__132662067fd4339398b4b0c2aa91a488cb53e3b3cb0af5e8f911aa65e93e02e9_det.mmd @@ -0,0 +1,949 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 928, 175]]<|/det|> +# Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers + +<|ref|>text<|/ref|><|det|>[[44, 196, 171, 213]]<|/det|> +Sven Nelander + +<|ref|>text<|/ref|><|det|>[[55, 223, 312, 240]]<|/det|> +sven.nelander@igp.uu.se + +<|ref|>text<|/ref|><|det|>[[44, 269, 580, 290]]<|/det|> +Uppsala University https://orcid.org/0000- 0003- 1758- 1262 + +<|ref|>text<|/ref|><|det|>[[44, 295, 149, 311]]<|/det|> +Ida Larsson + +<|ref|>text<|/ref|><|det|>[[55, 315, 580, 334]]<|/det|> +Uppsala University https://orcid.org/0000- 0001- 5422- 4243 + +<|ref|>text<|/ref|><|det|>[[44, 340, 133, 356]]<|/det|> +Felix Held + +<|ref|>text<|/ref|><|det|>[[55, 361, 718, 380]]<|/det|> +Chalmers University of Technology https://orcid.org/0000- 0002- 7679- 7752 + +<|ref|>text<|/ref|><|det|>[[44, 386, 190, 403]]<|/det|> +Gergana Popova + +<|ref|>text<|/ref|><|det|>[[55, 407, 576, 426]]<|/det|> +Uppsala University https://orcid.org/0000- 0001- 9804- 4041 + +<|ref|>text<|/ref|><|det|>[[44, 432, 131, 448]]<|/det|> +Alper Koc + +<|ref|>text<|/ref|><|det|>[[55, 454, 578, 472]]<|/det|> +Uppsala University https://orcid.org/0009- 0009- 8011- 9952 + +<|ref|>text<|/ref|><|det|>[[44, 479, 161, 495]]<|/det|> +Soumi Kundu + +<|ref|>text<|/ref|><|det|>[[55, 500, 576, 519]]<|/det|> +Uppsala University https://orcid.org/0000- 0002- 0759- 3051 + +<|ref|>text<|/ref|><|det|>[[44, 525, 202, 542]]<|/det|> +Rebecka Jörnsten + +<|ref|>text<|/ref|><|det|>[[55, 547, 617, 565]]<|/det|> +University of Gothenburg and Chalmers University of Technology + +<|ref|>sub_title<|/ref|><|det|>[[44, 608, 103, 625]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 645, 840, 666]]<|/det|> +Keywords: regulatory programs, regulatory- driven clustering, cell state, phenotypic plasticity + +<|ref|>text<|/ref|><|det|>[[44, 683, 330, 702]]<|/det|> +Posted Date: January 11th, 2024 + +<|ref|>text<|/ref|><|det|>[[42, 721, 475, 740]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3658321/v1 + +<|ref|>text<|/ref|><|det|>[[42, 758, 912, 800]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 819, 534, 838]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 874, 950, 917]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 9th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 53954- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[117, 85, 880, 141]]<|/det|> +# Reconstructing the regulatory programs underlying the phenotypic plasticity of neural cancers + +<|ref|>text<|/ref|><|det|>[[118, 167, 880, 211]]<|/det|> +Ida Larsson \(^{1,\times}\) , Felix Held \(^{2,\times}\) , Gergana Popova \(^{1}\) , Alper Koc \(^{1}\) , Soumi Kundu \(^{1}\) , Rebecka Jörnsten \(^{2}\) , Sven Nelander \(^{1,\ast}\) + +<|ref|>text<|/ref|><|det|>[[123, 228, 875, 272]]<|/det|> +1: Department of Immunology, Genetics and Pathology, Uppsala University, SE- 751 85 Uppsala, Sweden. + +<|ref|>text<|/ref|><|det|>[[125, 287, 870, 305]]<|/det|> +2: Mathematical Sciences, Chalmers University of Technology, SE- 412 96 Gothenburg, Sweden. + +<|ref|>text<|/ref|><|det|>[[374, 321, 626, 338]]<|/det|> +\(\times\) : Indicates equal contribution. + +<|ref|>text<|/ref|><|det|>[[113, 440, 454, 457]]<|/det|> +\* Correspondence: sven.nelander@igp.uu.se + +<|ref|>text<|/ref|><|det|>[[112, 473, 840, 491]]<|/det|> +Keywords: regulatory programs, regulatory- driven clustering, cell state, phenotypic plasticity + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 86, 214, 106]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 133, 889, 540]]<|/det|> +Nervous system cancers contain a large spectrum of transcriptional cell states, reflecting processes active during normal development, injury response and growth. However, we lack a good understanding of these states' regulation and pharmacological importance. Here, we describe the integrated reconstruction of such cellular regulatory programs and their therapeutic targets from extensive collections of single- cell RNA sequencing data (scRNA- seq) from both tumors and developing tissues. Our method, termed single- cell Regulatory- driven Clustering (scRegClust), splits target genes into modules and predicts essential kinases and transcription factors in little computational time thanks to a new efficient optimization strategy. Using this method, we analyze scRNA- seq data from both adult and childhood brain cancers to identify transcription factors and kinases that regulate distinct tumor cell states. In adult glioblastoma, our model predicts that blocking the activity of PDGFRA, DDR1, ERBB3 or SOX6, or increasing YBX1- activity, would potentiate temozolomide treatment. We further perform an integrative study of scRNA- seq data from both cancer and the developing brain to uncover the regulation of emerging meta- modules. We find a meta- module regulated by the transcription factors SPI1 and IRF8 and link it to an immune- mediated mesenchymal- like state. Our algorithm is available as an easy- to- use R package and companion visualization tool that help uncover the regulatory programs underlying cell plasticity in cancer and other diseases. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 86, 257, 107]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[112, 133, 888, 410]]<|/det|> +Nervous system cancers in adults and children share a common trait: tumor cells exist in multiple transcriptional states, which partly resemble the diversification of cells during normal embryonic development. In adult glioblastoma (GBM), cells resemble neural progenitor cells (NPCs), oligodendrocyte progenitor cells (OPCs) or astrocytes (ACs) (Neftel et al, 2019). In addition to these states, a substantial fraction of cells display a mesenchymal (MES)/injury response- like profile that has been linked to monocyte populations (Richards et al, 2021; Wang et al, 2017; Gangoso et al, 2021; Schmitt et al, 2021; Hara et al, 2021). In childhood cancers such as diffuse midline glioma, tumor cells move from a proliferative, OPC- like state to more differentiated states resembling either AC- or oligodendrocyte (OC)- like lineages (Filbin et al, 2018), and in medulloblastoma (MB) the four main subgroups differ in the proportion of undifferentiated vs differentiated tumor cells and their lineage resemblance (Hovestadt et al, 2019). + +<|ref|>text<|/ref|><|det|>[[111, 424, 888, 700]]<|/det|> +There are strong reasons to believe that the cell states found in a tumor have implications for disease progression. For instance, invasive GBM cells mimic the migration mechanisms of neuronal cells (Cuddapah et al, 2014; Venkataramani et al, 2022), whereas tumor initiation is mediated by quiescent stem- like cells (Xie et al, 2022). Further, it's been reported that recurrent therapy- resistant GBM tumors display a higher fraction of MES cells (Bhat et al, 2013), possibly due to a phenotypic shift in the cells from non- MES to MES (Wang et al, 2022). However, despite their role in disease recurrence, progression, and drug resistance, we have limited knowledge about the mechanisms by which cell states are regulated. In particular, to understand the transcriptional regulation it is essential to identify sets of transcription factors (TFs) whose activity have a direct impact on a specific state phenotype. Furthermore, to identify drug targeting opportunities, it is important to identify sets of drugable proteins, particularly kinases, that can be linked to states. + +<|ref|>text<|/ref|><|det|>[[111, 714, 888, 913]]<|/det|> +An increasing amount of available repositories of scRNA- seq data presents new opportunities to uncover such regulation with high precision. However, existing data analysis methods are not developed with this goal in mind. The frequently used clustering methods do not contain regulatory predictions; they are only concerned with grouping cells or genes. Previously, a standard method to identify regulators of gene signatures has been to first estimate a gene regulatory network (GRN), followed by post- processing to identify regulators linked to signature genes. This approach can, for instance, predict regulators of mesenchymal transformation from bulk transcriptional (Carro et al, 2010) and bulk multi- omic (Kling et al, 2016) data, but transferring it to the scRNA- seq setting + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 262]]<|/det|> +has proven unsatisfactory as individual links are hard to reproduce (Pratapa et al, 2020; Chen and Mar, 2018). Adding additional layers of data, primarily ATACseq or CHIPseq measurements (Aibar et al, 2017; Kamimoto et al, 2020; Kamal et al, 2022), is one way of alleviating the problem but comes with its own limitations. Existing methods are not fast enough to process millions of cells, and restricting the model to only TFs makes it less amenable to modeling the impact of pharmacologically more relevant gene classes, such as kinases. Thus, the development of fast and accurate methods to reconstruct gene regulatory programs from scRNA-seq data will be important, if not essential. + +<|ref|>text<|/ref|><|det|>[[111, 275, 888, 628]]<|/det|> +Here, we describe a new method for fast construction of regulatory programs that is well- suited for large data sets and for the detection of critical regulators of cell states. Applied to scRNA- seq data, it jointly detects modules (sets, clusters) of co- expressed genes and groups of regulators, such as transcription factors and kinases, linked to each module. Compared to methods of similar scope it is both faster and more flexible. As proof- of- principle, we show the applicability of our algorithm through three use cases. First, we apply it to a data set from peripheral blood mononuclear cells (PBMC) and benchmark it against the most comparable algorithm we can find, SCENIC+ (González- Blas et al (2023)). Second, we apply it to scRNA- seq data previously generated by our group (Larsson et al, 2021) to predict regulatory interventions to potentiate temozolomide (TMZ) treatment in GBM. Third, we integrate 13 data sets from the developing brain and cancers of the nervous system to investigate the regulatory landscape of neuro- oncology. We find a meta- module regulated by the transcription factors SPI1 and IRF8 which shows strong resemblance with the previously described immune cell- induced MES- like state. The algorithm we developed, scRegClust, is available as an easy- to- use R- package. + +<|ref|>sub_title<|/ref|><|det|>[[113, 668, 198, 688]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[113, 717, 488, 737]]<|/det|> +## A regulatory-driven clustering method + +<|ref|>text<|/ref|><|det|>[[112, 759, 886, 907]]<|/det|> +We present a new approach to detect regulatory programs from scRNA- seq data alone, based on a mathematical framework that directly models the interaction between regulators and responding target gene sets (Figure 1A). Our model assumes that genes belong to either of two categories: tentative regulators and tentative targets. Given a volume of data \(\mathbf{Z}\) , organized as cells \(\times\) genes, we split the data in two parts along genes, a regulatory part \(\mathbf{Z}_r\) , and a target part \(\mathbf{Z}_t\) . These parts are further randomly split along cells into two sets of training and assessment cells. We then perform a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 80, 888, 777]]<|/det|> +clustering task by identifying groups of genes in \(\mathbf{Z}_{t}\) that correspond to sets of co- regulated genes, and for each such set, we perform a regulatory program re- construction task by identifying a small number of genes in \(\mathbf{Z}_{r}\) that are the likely regulators. These steps are repeated until configurations stabilize. To fit regulatory programs from data computationally, we developed an alternating two- step scheme iterating between determining the most predictive regulators (regulatory program re- construction) for each of a pre- specified number \(K\) of modules and using these optimal regulators to allocate target genes into \(K\) modules (clustering). Step 1 is approached using cooperative- Lasso (coop- Lasso, Chiquet et al, 2012) on the training set. In this step, we search for a sparse set of regulators that can be linked to most genes in a module, each regulator with the same sign (positive vs negative). A sparsity penalty parameter controls the number of regulators assigned to each module. In section "Regulatory- driven clustering improves accuracy" we provide details on how to choose this penalty parameter. To solve the optimization problem efficiently, we apply over- relaxed Alternating Direction Method of Multipliers (ADMM, Boyd et al, 2011; Xu et al, 2017) to minimize the coop- Lasso objective function. Once regulators have been assigned to modules, we move on to Step 2 where the task is to refine the target gene sets. We approach this by re- estimating coefficients per module using sign- constrained non- negative least squares (NNLS, Meinshausen, 2013) on the training set and an observation- based allocation scheme using the assessment set. Target genes whose predictive \(R^{2}\) is below a threshold across all modules are marked as noise and placed in a rag- bag cluster. This ensures that outliers do not overly distort the clustering result. There are several optional inputs that can guide the algorithm. Of note is the possibility to provide prior information on the relationship between genes in the form of a gene \(\times\) gene matrix where a non- zero entry indicates that there is a known link between the genes. A known link can for example be that both genes appear in the same biological process. Algorithmic details can be found in "Methods". The resulting regulatory programs are visualized using either a coupling matrix view, where predicted regulators are columns and target gene modules are rows, or a module network view (Figure 1A). In addition, the algorithm returns goodness- of- fit measures which can be used to guide the selection of optimization parameters (Figure 2A, Figure S6). + +<|ref|>sub_title<|/ref|><|det|>[[114, 808, 787, 829]]<|/det|> +## Comparison with SCENIC+ shows good agreement between methods + +<|ref|>text<|/ref|><|det|>[[112, 850, 886, 895]]<|/det|> +We benchmarked our entire workflow on a publicly available data set from peripheral blood mononuclear cells (PBMC). Starting from a matrix of raw gene counts, the data was normalized and formatted + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 888, 234]]<|/det|> +(see details in “Methods”). As previously mentioned, potential regulators can be all TFs, all kinases, or any custom assignment of interest. For a direct comparison with the newly released SCENIC+ (González- Blas et al, 2023) (combining scRNA-seq and scATAC-seq), we used all TFs as potential regulators. We did not provide the algorithm with any prior information. Once the algorithm converged, it had clustered the PBMC data into nine gene modules and 62 active TFs interacting with one or several of these gene modules. + +<|ref|>text<|/ref|><|det|>[[110, 248, 888, 654]]<|/det|> +From previous work on the PBMC data, we know that the cells can be divided into 8 clusters representing various types of immune cells (Figure 1B, Figure S1A). To functionally annotate the gene modules defined by scRegClust, we scored these against the gene signatures of the immune cell types (Figure S1B- C). Our algorithm can successfully distinguish modules upregulated in CD8+ T cells, CD4+ T cells, B cells, NK cells, FCGR3A+ monocytes, CD14+ monocytes and plasmacytoid dendritic cells (pDCs). Among the strongest regulators we find many examples of known regulators, e.g. PAX5 (B cells), TBX21 (NK cells), CUX2 (dendritic cells) and LEF1 (T cells), to mention a few (Figure 1C). As a point of comparison, SCENIC+ identified 53 (activator) TFs, out of which 30 overlap with our identified regulators ( \(p < 10^{- 5}\) , Fisher’s exact test; Figure 1D). To get a more detailed understanding of our output, we looked into the functional annotation for the predicted regulators with a particular focus on the 32 TFs that do not overlap with SCENIC+. For the regulators that had an annotation in the Human Protein Atlas (HPA, Uhlen et al, 2015), 76.3 % are classified as being immune cell enriched/enhanced compared to a baseline 22 % for all genes in the atlas. For the non-overlapping genes, 55 % are enhanced or enriched in immune cells and have an immune cell type annotation that agrees with the one predicted by scRegClust (Figure 1E, Supplementary Table 1). + +<|ref|>text<|/ref|><|det|>[[111, 667, 888, 867]]<|/det|> +Jointly, the comparison between scRegClust and SCENIC+ on PBMC data showed a good level of agreement, despite the fact that scRegClust uses scRNA- seq data alone. In contrast to SCENIC+, which relies on pre- defined cell types for each cell in the primary data, scRegClust performs clustering on target genes, which can then be functionally annotated to match cell types. As previously mentioned, SCENIC+ is restricted to identifying TFs as regulators, due to its reliance on scATAC- seq data, while scRegClust in principle can consider any category of genes as potential regulators. The speed of the two algorithms is also remarkably different, scRegClust being three times faster than SCENIC+ (Figure 1F). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 88, 574, 108]]<|/det|> +## Regulatory-driven clustering improves accuracy + +<|ref|>text<|/ref|><|det|>[[112, 129, 886, 406]]<|/det|> +We also benchmarked our workflow on synthetic data to explore the effect of associating regulators with modules on clustering quality, the robustness of our modeling assumptions, and correct regulator recovery. In addition, we developed tools and strategies to guide the selection of penalty parameters and the correct number of modules. The quality of the clustering with respect to the groundtruth is measured using the adjusted Rand index (Hubert and Arabie, 1985). Two measures are introduced to evaluate the quality of the selected regulatory model associated with each module. Predictive \(R^2\) per module measures the predictive performance on the assessment set of the regulators associated with each module. Regulator importance measures the change in predictive \(R^2\) per module when a single regulator is omitted from the full regulator model for a module. In addition, we considered the true and false positive rates, presented as a ROC curve, of estimating the correct regulators associated with each target gene. See the "Methods" section for details on the computed metrics. + +<|ref|>text<|/ref|><|det|>[[112, 419, 886, 566]]<|/det|> +Synthetic data was generated to mimic characteristics of our in- house scRNA- seq data as well as characteristics of the multi- study comparison presented in the upcoming sections. We based our simulations on a modified scRNA- seq data set (see "Supplementary Methods") and generated four different setups capturing different aspects of real data scenarios. Below, a summary of the most important findings is presented. Consult the Supplementary Material for details on simulations and further results. + +<|ref|>text<|/ref|><|det|>[[111, 580, 886, 907]]<|/det|> +If no initial clustering is provided, our algorithm is initialized by k- means++ (Arthur and Vassilvitskii, 2007) on the cross- correlation matrix of the target genes and regulators in the training set. When the penalization parameter was chosen appropriately, associating regulatory models with each module improved the clustering quality substantially (Figure 2B, adjusted Rand index increases from 0.33 to 0.97). scRegClust showed excellent regulator recovery properties in our simulations (see Figure 2B and C). For appropriate penalization, the correct regulators were selected and only few true regulators were disregarded. scRegClust performed well even when model assumptions were violated (see the Supplementary Material for details). Notably, scRegClust achieved a large improvement in clustering quality in a scenario with unequal module sizes. There, clustering on the cross- correlation matrix gave an essentially random result, whereas our algorithm managed to nearly correctly recover the simulated modules, when penalization was chosen appropriately (see Figure S5B, adjusted Rand index increases from 0.06 to 0.95). This shows that the integrated approach of structure modeling and clustering is superior to clustering on the correlation structure alone. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 888, 364]]<|/det|> +Predictive \(R^{2}\) per module and regulator importance can be used as guidance for the selection of the appropriate range of penalty parameters. As a general rule, at least 5 to 10 different penalization parameters should be tested. Increasing the penalty parameter reduces the number of active regulators associated with each module. This will typically reduce predictive \(R^{2}\) gradually until the point when penalization becomes too strong and important regulators are not associated with a module any longer. A similar but reversed trend can be observed for regulator importance. Due to many regulators being included in the models for low penalization, importance will typically be low at first. Increasing the penalty parameter and therefore including fewer regulators in the model will increase regulator importance until it becomes inflated due to the inclusion of too few regulators. As illustrated in Figure 2A, finding this change point from gradual to rapid decrease or increase, respectively, indicates a range of appropriate penalty parameters. + +<|ref|>text<|/ref|><|det|>[[111, 377, 888, 757]]<|/det|> +A common problem in clustering is the selection of the correct number of modules \(K\) . Contrary to methods like k- means, our algorithm can produce empty modules. This can be enforced more strongly by specifying a minimum non- empty cluster size. We find in our simulations that scRegClust tends to combine groundtruth modules with regulatory overlap when \(K\) is underspecified and tends to split groundtruth modules into smaller ones when \(K\) is overspecified (see Supplementary Table 3). To provide further guidance on the selection of the number of modules, we introduce silhouette scores (see "Methods" section). These can be interpreted similarly to standard silhouette values (Rousseeuw, 1987) as a measure of how well a target gene is located within its own module compared to how well it would be located within the nearest module. High average silhouette scores indicate that target genes are well located within their module. Analysis of the average silhouette score should be combined with analysis of the predictive \(R^{2}\) per module (see Figure S6). A good clustering achieves high values in both scores. Our recommendation is to start with an initial guess for the number of modules \(K_{0}\) and to run scRegClust on a range of module counts above and below \(K_{0}\) . Analysis of the silhouette scores and predictive \(R^{2}\) per module as illustrated in Figure S6 can give guidance on how to choose the optimal number of modules. + +<|ref|>text<|/ref|><|det|>[[112, 770, 886, 840]]<|/det|> +To summarize, through an extensive simulation study where several data simulation setups reflecting various real data scenarios were tested, we demonstrated that scRegClust performs excellent both in terms of improving clustering accuracy and recovering regulatory interactions. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 88, 870, 108]]<|/det|> +## Targeting regulators of an OPC-like state potentiates temozolomide treatment + +<|ref|>text<|/ref|><|det|>[[112, 129, 886, 279]]<|/det|> +Encouraged by the results from the benchmarking studies, we continued by applying scRegClust to complex data sets from nervous system cancers to explore two key questions. First, we asked if the algorithm could predict interventions, such as drugs or gene perturbations, that would push GBM cells into a temozolomide (TMZ)- sensitive state. In subsequent passages, we leverage the power of our algorithm to integrate scRNA- seq data across nervous system cancers and the developing brain to identify meta- modules with shared modes of regulation and biological function. + +<|ref|>text<|/ref|><|det|>[[111, 291, 886, 516]]<|/det|> +One of the questions raised during our previous work on cell state transitions in GBM was how to successfully target an entire cell state, and not just individual target genes of these states. The question was prompted by our prediction that the minimal intervention needed to potentiate TMZ treatment is to block transitions to what we call "state 5", a state with an OPC- like, invasive profile (Larsson et al, 2021). To investigate this, we applied scRegClust to the previously mentioned scRNA- seq data from U3065MG cells (Larsson et al, 2021; Xie et al, 2015) using either TFs or kinases as potential regulators (Figure 3A). The algorithm was run without giving prior information on cluster assignment for the genes, which resulted in scRegClust clustering the genes in 9 (TF) or 6 (kinase) modules, respectively (Figure 3B). + +<|ref|>text<|/ref|><|det|>[[111, 529, 886, 781]]<|/det|> +We found that scRegClust defines a group of modules that correspond to state 2, a highly proliferative progenitor state. By studying the functional profile of the modules further (Figure 3B), it is evident that the modules are related to proliferation and the cell cycle and that they diverge on cell cycle phase - where two of the modules represent the earlier phases of cell cycle (G1, G1S, S) while remaining modules represent the later phases (S, G2M, M). Several regulators identified for these modules agree with the module profiles, e.g. E2F1 and TK1 regulating the earlier phases of cell cycle (Martínez- Balbás et al (2000); Chang et al (1998)) while AURKA and CENPA are mostly active during G2M/M- phase (Marumoto et al (2003); Zeitlin et al (2001)). Another interesting observation is the astrocyte- like state 4, which is predicted to be positively regulated by ID3, known to induce astrocyte differentiation through the BMP pathway signaling (Bohrer et al (2015)). + +<|ref|>text<|/ref|><|det|>[[112, 794, 885, 890]]<|/det|> +Addressing our initial question on how to suppress state 5 to potentiate TMZ treatment we focus on the regulation of the modules corresponding to this state (Figure 3C). The top kinase regulator is PDGFRA, consistent with state 5 having an OPC- like profile as PDGFRA is both a known regulator of OPCs in normal development (Ellison and de Vellis, 1994) and frequently amplified in the OPC- like + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 888, 440]]<|/det|> +GBM state (Neftel et al, 2019). Other strong kinase regulators are DDR1, a receptor tyrosine kinase expressed in oligodendrocytes in the developing brain (Vilella et al, 2019), and ERBB3, also implicated in the oligodendrocyte lineage development and the OPC- like state in GBM (Wang et al, 2020). The top TF regulators are SOX6 (positive regulation) and YBX1 (negative regulation). SOX6 belongs to the group D family of SOX TFs, which have been shown to regulate several stages of oligodendrocyte development. The combined predictions from Larsson et al (2021) and scRegClust suggest that blocking the activity of PDGFRA, DDR1, ERBB3 or SOX6, or increasing YBX1- activity, would potentiate TMZ treatment. We tested these predictions experimentally using two approaches. First, using CRISPR/Cas9- mediated knockdown we reduced the expression of DDR1 or SOX6 in U3065MG cells, followed by TMZ treatment. Viability measurements of the treated cells confirmed our predictions; both DDR1- KD cells and SOX6- KD cells respond better to TMZ treatment, which is seen through a dose- dependent shift in the viability curve compared to non- targeting control cells (Figure 3D). The effect is most pronounced in DDR1- KD cells, where we achieved a halving in viability in the KD population compared to control at the highest drug dose tested. + +<|ref|>text<|/ref|><|det|>[[111, 454, 887, 678]]<|/det|> +Second, we combined TMZ with the tyrosine- kinase inhibitor (TKI) dasatinib (Figure 3E), which had an inhibition score of \(>0.85\) in the STITCH database for PDGFRA and ERBB3, and an interaction score of 0.889 with DDR1. In the U3065MG cell line, simultaneous treatment with TMZ and dasatinib showed no synergistic effect, neither did pre- treatment with TMZ followed by simultaneous treatment with both drugs. A synergistic effect was, however achieved when pre- treating cells with dasatinib for 72 hours, followed by combined treatment with both drugs (BSS \(>10\) ) (Figure 3F). Relating this to previous predictions, the results imply that pre- treatment with dasatinib depletes the state 5 population, while subsequent combined treatment inhibits cells from escaping to state 5 and therefore leaves the tumor population vulnerable to TMZ. + +<|ref|>text<|/ref|><|det|>[[112, 692, 886, 762]]<|/det|> +Building on these results, we broadened our investigation's scope to explore regulators of cell plasticity in nervous system cancers. To that end, we included more GBM single- cell data sets, as well as data from other cancers of the nervous system and corresponding healthy tissue. + +<|ref|>sub_title<|/ref|><|det|>[[113, 796, 537, 816]]<|/det|> +## The regulatory landscape of neuro-oncology + +<|ref|>text<|/ref|><|det|>[[112, 839, 888, 909]]<|/det|> +We used scRegClust to integrate the regulatory programs from the developing brain (Couturier et al, 2020; Eze et al, 2021; Hamed et al, 2022; Luo et al, 2021; La Manno et al, 2021; Weng et al, 2019), normal adrenal gland (Kildisuite et al, 2021), GBM (Couturier et al, 2020; Darmanis et al, 2017; + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 888, 286]]<|/det|> +LeBlanc et al, 2022; Neftel et al, 2019; Wang et al, 2019), MB (Hovestadt et al, 2019; Luo et al, 2021; Ocasio et al, 2019) and neuroblastoma (NB) (Kildisuite et al, 2021). We ran the algorithm for each data set individually and merged the resulting regulatory tables (Figure 4A). In the merged regulatory table (Figure 4B), the rows correspond to regulators (TFs) and the columns to gene modules derived from each individual study. The top annotation bars indicate the disease type and study that each module originate from. It is evident that certain meta- modules emerge (Box 1), consisting of several modules with similar gene content from different studies that are regulated by the same regulators. + +<|ref|>text<|/ref|><|det|>[[111, 300, 888, 450]]<|/det|> +Two meta- modules, 1 and 7, represented actively proliferating cells from all types (normal tissue, GBM, MB and NB). Meta- module 1 was primarily driven by positive regulation by CENPA and HMGB2, while meta- module 7 was more strongly regulated by HMGB1 and DEK. Our model suggests a subdivision of MB and normal samples along a gradient regulated by HMGB1/2 and DEK vs NEUROD1, where NEUROD1 promotes non- proliferating neuronal- like NPC2- like cells (meta- module 3), and HMGB1/2 and DEK promotes neural progenitors with active proliferation. + +<|ref|>text<|/ref|><|det|>[[111, 461, 887, 635]]<|/det|> +Two meta- modules, 2 and 6, were enriched for hallmark signatures of hypoxia and apoptosis, as well as the Ivy atlas signature pseudopalisading cells around necrosis (CT pan), probably representing a general stress response in the cells. These were regulated by YBX1 (#2) and members of the Fos and Jun family of TFs (#6). An astrocyte- like meta- module (#4) in normal and MB cells was linked to ID3, ID4 and HOPX. In GBM, these cells were clustered in meta- module 2 due to their additional enrichment for stress response signatures. An OPC- like meta- module (#13) was identified that was driven by e.g. OLIG1, MYRF and SOX10. + +<|ref|>text<|/ref|><|det|>[[111, 648, 888, 900]]<|/det|> +There were two distinct mesenchymal meta- modules, 5 and 15. One mesenchymal meta- module (#5) coincided with an enrichment of microglial signatures and was primarily driven by SPI1 and IRF8 in normal, GBM and MB cells. A possible explanation to this meta- module is that it captures the phenotype of cells undergoing a mesenchymal shift induced by tumor- associated macrophages (TAMs) or microglia, and that the enrichment of microglia- related gene signatures in disease modules are due to this previous interaction between tumor and immune cells. SPI1 and IRF8 could therefore be candidate regulators of the observed mesenchymal shift in GBM cells. SPI1 is a pioneer TF, i.e. one that can bind directly to condensed chromatin and thereafter recruit other, non- pioneer TFs, to the site. The known normal function of SPI1 is in controlling the cell fate decision of hematopoietic cells, e.g. in macrophage differentiation. After binding to regulatory elements in the genome, it regulates + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 888, 235]]<|/det|> +gene expression by recruiting other transcription factors, such as interferon regulatory factors, e.g. IRF8 (Pham et al, 2013; Le Coz et al, 2021). IRF8 has been previously identified as a candidate gene involved in the immune evasion of GBM cells. Gangoso et al (2021) observed that IRF8 was upregulated in GSC cells in an immuno- competent mouse model following immune attack, and since IRF8 is normally a myeloid- specific master transcription factor, they termed this immune evasion strategy “myeloid- mimicry”. + +<|ref|>text<|/ref|><|det|>[[112, 249, 888, 552]]<|/det|> +The other mesenchymal metamodule (#15) showed no enrichment of microglial signatures, but instead showed enrichment for the Ivy atlas signature microvascular proliferation. This meta- module was interesting in the sense that it had regulators that depended on disease category. GBM samples were regulated by FOX51 and HEY1, while MB samples were regulated by AEBP1 and TBX18. Normal tissue samples were driven by previously mentioned regulators but also MEF2C and TSC22D4. A third meta- module (#12) showed a slightly weaker enrichment for the microvascular proliferation signature and included all types. This was driven by several TFs from the ERG family (ERG, FLI, ETS1) and SOX17, to name a few. Several of these regulator interactions are consistent with their described role in literature, e.g. FLI1 which is a prognostic marker in astrocytoma (Hu et al, 2022) and is predicted to be a regulator of perivascular- like glioma cells (Hu et al, 2022) and the endothelial potential of myogenic progenitors (Ferdous et al, 2021). Similarly, the role of SOX17 as a promotor of tumor angiogenesis has been described in mice (Yang et al, 2013). + +<|ref|>text<|/ref|><|det|>[[112, 564, 886, 662]]<|/det|> +Our model also suggests a subdivision of NB samples (meta- modules 9 and 14) along an adrenal- to- mesenchymal axis, driven by ZNF90, MEIS2 and MYC, each of which was positively linked to mesenchymal- like signatures and negatively linked to ADRN- like signatures. Remaining meta- modules (#8, 10, 11) were specific to one type and driven by just one or two regulators. + +<|ref|>text<|/ref|><|det|>[[112, 675, 888, 900]]<|/det|> +Similarly to the TF regulatory landscape, the kinase map (Figure S2) revealed several likely links and made interesting predictions. For instance, along the oligodendrocytic lineage, AATK is linked to a meta- module in both GBM and normal developing brain that is enriched for oligodendrocyte differentiation. PDGFRA and DGKB is linked to a meta- module with an OPC- like profile, whereas PDGFRB is linked to invasiveness and microvascular proliferation signatures. CKB regulates a meta- module of NPC signature targets in normal developing brain, GBM and MB, and suppresses cell cycle genes in MB and GBM. CCND1 suppresses NPC markers. Mature astrocytes were primarily linked to NTRK2 activity. CCL2 drives a microglia- like signature in both normal developing brain and MB. As in the TF map, known cell cycle driving kinases converge on similar modules in GBM, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 185]]<|/det|> +MB and normal developing brain, e.g. AURKA, PLK1 and CCNB2. Still there was an MB- specific TK1- driven cell cycle module. AK4 links to a GBM- specific module enriched for cellular tumor genes. We also found some hits of unclear significance, like SGK1 regulating a common meta- module in GBM, MB and normal developing brain. + +<|ref|>sub_title<|/ref|><|det|>[[113, 225, 232, 245]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 273, 886, 524]]<|/det|> +The high extent of intratumoral heterogeneity in nervous system cancers and the plastic behavior of tumor cells are major obstacles in the search for more efficient therapies against these often deadly diseases. The enormous amount of scRNA- seq data that has been generated to map intratumoral heterogeneity presents great opportunities to understand the regulation of transcriptional cell states and cell state transitions, but it also challenges us to develop suitable methods to properly analyze the data. In this work, we have addressed the lack of computational methods to identify such regulators of intratumoral heterogeneity and present a method that simultaneously detects gene modules and their regulators from scRNA- seq data. The utility of our method is demonstrated through three use cases, where it is applied to real data, and through a thorough investigation of model properties using synthetic data. + +<|ref|>text<|/ref|><|det|>[[112, 538, 886, 788]]<|/det|> +By associating a regulator model with each module, scRegClust improves substantially on initial clustering results derived from the correlation of target genes and regulators. We further develop scores, such as \(R^2\) per module, regulator importance, and a silhouette score, to help the user with parameter choices and to support the interpretation of the resulting regulatory network. In our first use case, we apply scRegClust to the PBMC data set and compare our results to those of SCENIC+. We show that we can, in an unsupervised fashion, recapitulate signatures of the known cell types present in the data as well as several known regulators of each cell type. The result is comparable to that of SCENIC+, but obtained in shorter time and with only one data source. We do however emphasize that although the methods produce a similar output, they differ in several important technical aspects, and should be treated as complementing rather than competing methods. + +<|ref|>text<|/ref|><|det|>[[112, 802, 886, 898]]<|/det|> +In our second and third use case, scRegClust is applied to data sets from nervous system cancers (GBM, MB and NB) and the developing brain and used to derive insights about the diseases that warrant further investigation. We identify several regulators of an OPC- like, invasive GBM state and combine TMZ and dasatinib, a tyrosine kinase inhibitor targeting the state regulators, to potentiate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 888, 440]]<|/det|> +TMZ treatment. Dasatinib has in previous studies shown anti- migratory effects (Li et al, 2015), consistent with state 5 having an invasive profile, and were tested in combination with TMZ in a clinical trial as treatment against GBM (Milano et al, 2009). However, this trial was terminated before proceeding to phase II. We find that the combination treatment is only synergistic in our U3065MG cell line when we first pre- treat the cells with dasatinib, followed by combined treatment with both drugs, and speculate whether this means that state 5 first needs to be depleted for the combination treatment to be synergistic. These speculations need to be more carefully investigated by measuring whether the protein levels of state 5 markers decrease following dasatinib treatment. Second, by performing an integrative study of the regulatory programs of GBM, MB, NB and the developing brain, we find that SP11 and IRF8 are candidate regulators of the mesenchymal shift induced by TAMs and/or microglia in GBM. These regulators can be an important piece of the puzzle to understand the mechanism behind the observed mesenchymal transition, which appears to be a strategy for the tumor cells to evade the immune system and become increasingly resistant to treatment. + +<|ref|>text<|/ref|><|det|>[[111, 454, 888, 629]]<|/det|> +For future developments, we see several possible extensions of the algorithm. At present, scRegClust has the functionality to include prior knowledge of gene- gene relationships to guide the clustering of target genes. It would however be interesting to explore the possibility of including prior knowledge in the regulator- target gene relationship as well, e.g. by favouring connections that have support in an external data set (ATAC- seq, CHIP- seq etc). In addition to this, we see extensions related to non- additive contributions to the regulation of a module, extended regulation modeling to allow for feedback regulation, or automatic partitioning in regulator- and target genes. + +<|ref|>text<|/ref|><|det|>[[112, 642, 886, 738]]<|/det|> +To conclude, we present an algorithm that operates on scRNA- seq data alone to construct regulatory programs consisting of regulators and target gene modules, and use it to understand the regulatory mechanisms of cell state plasticity. The algorithm is provided as an easy- to- use R- package and can be applied to any scRNA- seq data set. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[112, 87, 214, 105]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[112, 136, 199, 152]]<|/det|> +## Datasets + +<|ref|>text<|/ref|><|det|>[[112, 178, 884, 248]]<|/det|> +All publicly available data sets used in this publication are listed in Table 1, including the data repository from where they were downloaded. The notation "3CA" refers to the Curated Cancer Cell Atlas provided by the Tirosh Lab (Gavish et al, 2021). + +<|ref|>table<|/ref|><|det|>[[115, 285, 880, 646]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[291, 260, 702, 276]]<|/det|> +Table 1: Summary of data sets used in this publication. + +
StudyData repositoryDisease
Couturier et al. 20203CA/brainDeveloping brain
Couturier et al. 20203CA/brianGlioblastoma
Darmanis et al. 20173CA/brainGlioblastoma
Eze et al. 2021UCSC Cell BrowserDeveloping brain
Hamed et al. 2022GEO: GSE200202Developing brain
Hovestadt et al. 20193CA/brainMedulloblastoma
Kildisuite et al. 2021Neuroblastoma Cell AtlasNeuroblastoma
Kildisuite et al. 2021Neuroblastoma Cells AtlasNormal fetal adrenal gland
Larsson, Dalmo et alArrayExpress E-MTAB-9296Glioblastoma
Leblanc et al. 2022GEO: GSE173278, GSE193884Glioblastoma
Luo et al. 2021Short Read Archive: GSE156633Developing cerebellar granule
Luo et al. 2021Short Read Archive: GSC156633Medulloblastoma
Manno et al. 2021Sequence Read Archive: PRJNA637987Developing brain
Neftel et al. 20193CA/brainGlioblastoma
Ocasio et al. 2019GEO: GSE129730Medulloblastoma
Ocasio et al. 2019GEO:GSE129730Cerebellum
Wang et al. 20203CA/brainGlioblastoma
Weng et al. 2019GEO: GSE122871Developing brain
PBMC10X websitehuman blood cells
+ +<|ref|>sub_title<|/ref|><|det|>[[113, 690, 421, 708]]<|/det|> +## Data processing and formatting + +<|ref|>text<|/ref|><|det|>[[112, 731, 886, 904]]<|/det|> +If available, raw counts were downloaded from the indicated source in Table 1. The data was processed before further analysis in R (R Core Team, 2022) using the Seurat package (Butler et al, 2018). Cells containing less than 200 genes and genes present in less than 3 cells were filtered out. Data generated from a UMI- based single- cell sequencing protocol (10X Chromium, CEL- seq2) were normalized using sctransform (Hafemeister and Satija, 2019), while expression levels for remaining data sets were quantified using \(\log_{2}(\mathrm{TPM} / 10 + 1)\) , as suggested by Tirosh et al (2016). When needed, non- malignant cells were filtered out from the data set based on the provided metadata from the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 132]]<|/det|> +authors or the 3CA- portal (Gavish et al, 2021). We provide functionality to format the normalized data matrix to conform with the desired input of our scRegClust- algorithm (more details below). + +<|ref|>sub_title<|/ref|><|det|>[[113, 165, 321, 185]]<|/det|> +## Clustering Algorithm + +<|ref|>text<|/ref|><|det|>[[111, 208, 886, 408]]<|/det|> +We developed a two- step alternating algorithm to simultaneously perform clustering on a set of target genes as well as to associate a set of regulators with each cluster (module). In a first step, given an initial clustering of the data, the algorithm determines the linearly most predictive regulators for each cluster as well as whether the regulator acts stimulating or repressing. Then, in a second step, these optimal regulators are used to allocate target genes into clusters. Prior knowledge about target gene relationships can be included to guide cluster allocation. These two steps are repeated until configurations stabilize or a maximum number of steps is reached. The algorithm was implemented as an R- package and can be found on GitHub at https://github.com/sven- nelander/scregclust. + +<|ref|>sub_title<|/ref|><|det|>[[113, 440, 417, 459]]<|/det|> +## Algorithmic outline of scRegClust + +<|ref|>text<|/ref|><|det|>[[111, 481, 887, 604]]<|/det|> +The algorithm takes as its input two matrices \(\mathbf{Z}_{t}\) and \(\mathbf{Z}_{r}\) containing \(n\) rows of matching observations, typically cells, on \(p_{t}\) target genes and \(p_{r}\) tentative regulator genes in their columns, respectively. In addition, the algorithm requires the number \(K\) of desired clusters and an initial clustering of the target genes into \(K\) clusters. If the latter is not supplied, an initial clustering is produced as described below under "Preprocessing and initialization". + +<|ref|>text<|/ref|><|det|>[[111, 618, 886, 662]]<|/det|> +The motivation behind our method is gene \(j\) in cluster \(i\) is described by a sign- constrained linear model + +<|ref|>equation<|/ref|><|det|>[[352, 666, 644, 690]]<|/det|> +\[\mathbf{Z}_{t}^{(:,j)} = \mathbf{Z}_{r}^{(:,R_{i})}\mathrm{diag}(\mathbf{s}_{i})\mathbf{B}_{i}^{(:,j)} + \sigma_{i,j}\mathbf{e}_{i,j}\] + +<|ref|>text<|/ref|><|det|>[[111, 707, 887, 882]]<|/det|> +where \(\mathrm{R}_{i}\) is a subset of regulators associated with cluster \(i\) , \(\mathbf{s}_{i}\) is a vector of size \(|\mathrm{R}_{i}|\) containing the sign of each regulator associated with cluster \(i\) , \(\mathbf{B}_{i}\) are non- negative coefficients of dimension \(|\mathrm{R}_{i}|\times p_{t}\) , \(\sigma_{i,j}^{2}\) is a positive residual variance for each target gene \(j\) and cluster \(i\) , and \(\mathbf{e}_{i,j}\) are additive errors that are assumed to be normally distributed. Each cluster is therefore described by a selection of regulators \(\mathrm{R}_{i}\) and their signs \(\mathbf{s}_{i}\) encoding the assumption that a regulator acts stimulating (positive sign) or repressing (negative sign) on all target genes in the cluster. Coefficient values and residual variances are considered to be target gene- specific. + +<|ref|>text<|/ref|><|det|>[[110, 895, 884, 914]]<|/det|> +To cluster target genes and to select regulators associated with each cluster we consider the partitioning + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 91, 180, 106]]<|/det|> +problem + +<|ref|>equation<|/ref|><|det|>[[133, 125, 882, 210]]<|/det|> +\[\begin{array}{r l} & {\arg \min_{\Pi ,\mathbf{R},\mathbf{B},\mathbf{s},\pmb {\sigma}}\quad \frac{1}{2}\sum_{i = 1}^{K}\sum_{j = 1}^{p_{t}}\pmb {\Pi}^{(i,j)}\left(n\log \left(\sigma_{i,j}^{2}\right) + \frac{1}{\sigma_{i,j}^{2}}\left\| \pmb{Z}_{t}^{(i,j)} - \pmb{Z}_{r}^{(i,\mathbf{R})}\mathrm{diag}(\pmb {s}_{i})\pmb{B}_{i}^{(i,j)}\right\|_{F}^{2}\right)}\\ & {\quad \mathrm{such~that}\quad |R_{i}|\leq N_{i,r},\pmb{s}_{i}^{(k)}\in \{-1,1\} ,\pmb {B}_{i}\geq \mathbf{0}\mathrm{~and~}\sigma_{i,j} > 0}\\ & {\quad \mathrm{for~all}\quad i = 1,\ldots ,K,\ j = 1,\ldots ,p_{t},\ k = 1,\ldots ,p_{r},} \end{array} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[112, 231, 886, 302]]<|/det|> +where \(\mathbf{\Pi}\) is a \(K\times p_{t}\) cluster membership matrix such that \(\mathbf{\Pi}^{(i,j)} = 1\) if target gene \(j\) is in cluster \(i\) and 0 otherwise. The size constraint \(|{\bf R}_{i}|\leq N_{i,r}\) ensures that only a subset of regulator is associated with each cluster. Note, that we do allow regulators to appear in multiple clusters. + +<|ref|>text<|/ref|><|det|>[[112, 315, 886, 387]]<|/det|> +The direct solution of this optimization problem, however, requires to solve a combinatorial number of regression problems, which is computationally too demanding in practice. Instead, we used a data- driven approximation to solve Eq. (1). + +<|ref|>text<|/ref|><|det|>[[112, 400, 861, 420]]<|/det|> +Table 2 contains an overview of all symbols used in the description of the scRegClust algorithm. + +<|ref|>sub_title<|/ref|><|det|>[[113, 454, 358, 472]]<|/det|> +## Data-driven approximation + +<|ref|>text<|/ref|><|det|>[[112, 495, 886, 566]]<|/det|> +We proceed in the fashion of alternating clustering algorithms, such as k- means, by alternating between (1) determining the structure of clusters ( \(\mathrm{R}_{i}\) and \(\mathbf{s}_{i}\) ), and (2) re- assigning cluster membership (II). + +<|ref|>text<|/ref|><|det|>[[112, 597, 886, 797]]<|/det|> +Preprocessing and initialization Observations of target genes and regulators (or a randomly reduced subset) are randomly split into two sets to allow for unbiased estimation of the model quality during cluster allocation below. The ratio is user- defined but 50- 50 by default. If the observations are stratified, the sample assignment of each observation can be supplied and splitting is performed within each sub- group. We write \(\mathbf{Z}_{t,d}\) and \(\mathbf{Z}_{r,d}\) for the \(d\) - th data split containing \(n_{d}\) observations \((d = 1,2)\) . The first data split is regarded as a training set, whereas the second data split functions as an assessment set. We therefore compute the preprocessing below on the training set and apply the same values to the assessment set. + +<|ref|>text<|/ref|><|det|>[[112, 811, 886, 907]]<|/det|> +During data splitting, the user can choose whether or not to center the target genes within each sub- group defined by the sample stratification. In addition, after splitting, both target and regulatory genes are centered individually (without regard to stratification) and regulatory genes are scaled to standard deviation 1. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 228, 606, 247]]<|/det|> +Algorithm 1 High- level overview of the scRegClust algorithm. + +<|ref|>text<|/ref|><|det|>[[128, 248, 884, 285]]<|/det|> +Input: Pre- processed expression data for target genes \(\mathbf{Z}_{t}\) and regulators \(\mathbf{Z}_{r}\) . Optionally, an initial clustering \(\mathbf{\Pi}\) and an indicator matrix \(\mathbf{J}\) describing prior knowledge. + +<|ref|>text<|/ref|><|det|>[[128, 282, 884, 336]]<|/det|> +Initialization: Split data randomly into two sets \((\mathbf{Z}_{t,1},\mathbf{Z}_{r,1})\) and \((\mathbf{Z}_{t,2},\mathbf{Z}_{r,2})\) , and, if not provided in input, find initial clustering \(\mathbf{\Pi}\) by applying kmeans++ to the cross- correlation matrix of \(\mathbf{Z}_{t,1}\) and \(\mathbf{Z}_{r,1}\) . + +<|ref|>text<|/ref|><|det|>[[130, 352, 448, 369]]<|/det|> +for cycle \(c\) from 1 to maximum cycle do + +<|ref|>text<|/ref|><|det|>[[153, 370, 365, 386]]<|/det|> +Store current clustering \(\mathbf{\Pi}\) + +<|ref|>text<|/ref|><|det|>[[155, 387, 320, 402]]<|/det|> +for each cluster \(i\) do + +<|ref|>text<|/ref|><|det|>[[732, 388, 883, 404]]<|/det|> +Step 1: Selection + +<|ref|>text<|/ref|><|det|>[[185, 404, 432, 420]]<|/det|> +Given the current clustering \(\mathbf{\Pi}\) : + +<|ref|>text<|/ref|><|det|>[[183, 421, 650, 439]]<|/det|> +Identify active regulators \(\mathbf{R}_{i}\) and determine their signs \(s_{i}\) . + +<|ref|>text<|/ref|><|det|>[[155, 439, 220, 454]]<|/det|> +end for + +<|ref|>text<|/ref|><|det|>[[155, 472, 320, 488]]<|/det|> +for each cluster \(i\) do + +<|ref|>text<|/ref|><|det|>[[686, 472, 883, 488]]<|/det|> +Step 2a: Re- estimation + +<|ref|>text<|/ref|><|det|>[[185, 490, 537, 506]]<|/det|> +Given active regulators \(\mathbf{R}_{i}\) and their signs \(s_{i}\) : + +<|ref|>text<|/ref|><|det|>[[185, 507, 549, 523]]<|/det|> +Re- estimate coefficients for each target gene. + +<|ref|>text<|/ref|><|det|>[[185, 523, 755, 543]]<|/det|> +Compute normalized likelihood \(\mathbf{L}_{j}^{(i,l)}\) for each gene \(j\) and observation \(l\) . + +<|ref|>text<|/ref|><|det|>[[185, 543, 523, 560]]<|/det|> +Incorporate prior information if supplied. + +<|ref|>text<|/ref|><|det|>[[155, 560, 220, 575]]<|/det|> +end for + +<|ref|>text<|/ref|><|det|>[[155, 593, 360, 610]]<|/det|> +for each target gene \(j\) do + +<|ref|>text<|/ref|><|det|>[[712, 593, 883, 610]]<|/det|> +Step 2b: Allocation + +<|ref|>text<|/ref|><|det|>[[185, 610, 710, 628]]<|/det|> +Compute \(R_{i,j}^{2}\) for each cluster \(i\) and perform one of the following. + +<|ref|>text<|/ref|><|det|>[[185, 628, 840, 647]]<|/det|> +a) Sort out noisy genes into the rag-bag cluster if \(R_{i,j}^{2}\) across all clusters is too low. + +<|ref|>text<|/ref|><|det|>[[185, 646, 853, 664]]<|/det|> +b) Update the cluster allocation of \(j\) according to a majority vote across observations. + +<|ref|>text<|/ref|><|det|>[[155, 663, 220, 679]]<|/det|> +end for + +<|ref|>text<|/ref|><|det|>[[155, 697, 581, 715]]<|/det|> +if \(\mathbf{\Pi}\) is equal to any previously stored clustering then + +<|ref|>text<|/ref|><|det|>[[185, 715, 297, 730]]<|/det|> +Stop iteration. + +<|ref|>text<|/ref|><|det|>[[155, 732, 210, 747]]<|/det|> +end if + +<|ref|>text<|/ref|><|det|>[[130, 749, 196, 764]]<|/det|> +end for + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[110, 116, 886, 730]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[201, 87, 796, 103]]<|/det|> +Table 2: Overview of symbols used in the description of the scRegClust algorithm + +
SymbolDescription
Problem setup:
nTotal number of cells in the input matrix
ptNumber of target genes
prNumber of regulators
Zt,Zrtarget gene and regulator expression (n × pt resp. n × pr matrices)
KDesired number of clusters
ΠK × pt cluster membership matrix for target genes with Π(i,j) ∈ {0,1} and ∑K=1 Π(i,j) ≤ 1
Jpt × pt indicator matrix describing prior knowledge of biological relationships between target genes (e.g. pathway co-occurence)
For data splits d = 1, 2:
ndNumber of cells in the d-th data split
Zt,dtarget gene expression used in the d-th data split (nd × pt matrix)
Zr,dregulator expression used in the d-th data split (nd × pr matrix)
For each cluster i = 1, ..., K:
CiSet of target genes in cluster i
RiSet of regulators associated with cluster i
Ni,rMaximum number of regulators associated with cluster i
siSign vectors containing one sign for each regulator in Ri
Binon-negative regression coefficients (|Ri| × pt matrix)
σi,j2positive variance parameter for each target gene and cluster
Optimization-related (i = 1, ..., K, j = 1, ..., pt):
λPositive penalization parameter used in coop-Lasso
wiPositive weight vector of length pr for each cluster i in coop-Lasso
Bi,OLSOrdinary least squares estimates of the regression coefficients in cluster i (pr × |Ci| matrix)
Bi,CLCoop-lasso estimates of the regression coefficients in cluster i (pr × |Ci| matrix)
τNon-negative threshold for rag-bag clustering
pi,jPrior probability of target gene j being in cluster i
LjK × n2 likelihood matrix for target gene j
vjVector of n2 votes for the cluster assignment of target gene j
μPrior strength in [0,1] for trade-off between likelihood and prior in cluster allocation
Validation measures (i = 1, ..., K, j = 1, ..., pt, k = 1, ..., pr):
R2iPredictive R2 for cluster i
R2i,-kPredictive R2 for cluster i with regulator k omitted
Ii,kImportance of regulator k in cluster i
R2i,jPredictive R2 for target gene j predicted by regulators in Ri
SjSilhouette score for target gene j
+ +<|ref|>table_footnote<|/ref|><|det|>[[113, 797, 885, 837]]<|/det|> +Output: T Regulatory table providing a summary of regulator strength in each cluster \((p_{r}\times K\) matrix) + +<|ref|>text<|/ref|><|det|>[[112, 862, 886, 906]]<|/det|> +If no initial cluster membership is provided, the cross- correlation matrix of \(\mathbf{Z}_{t,1}\) and \(\mathbf{Z}_{r,1}\) is computed. The k- means++ algorithm (Arthur and Vassilvitskii, 2007) with the cross- correlation matrix as an + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 159]]<|/det|> +input and multiple restarts is then used to find an initial clustering of the target genes into \(K\) clusters. Note that this step takes regulators into account and is therefore not equivalent to the clustering of only target genes. + +<|ref|>text<|/ref|><|det|>[[111, 190, 886, 365]]<|/det|> +Determining a regulatory model for each cluster. Given the cluster membership of target genes, the goal of Step 1 is to determine which regulators are linearly most predictive for the target genes in each cluster. If cluster membership is considered to be known, the optimization problem in Eq. (1) can be solved separately for each cluster. However, due to the computational complexity of the optimization problem, selection of at most \(N_{i,r}\) out of \(p_{t}\) regulators as well as their signs, it is not possible to solve it exactly. In our scenario for cluster \(i\) , there are \(\sum_{k = 0}^{N_{i,r}}\binom{p_{r}}{k}\cdot 2^{k}\) candidate sets of regulators and their signs to consider. + +<|ref|>text<|/ref|><|det|>[[111, 375, 886, 602]]<|/det|> +The size constraint \(|\mathbf{R}_{i}|\leq N_{i,r}\) in (1) is equivalent to replacing \(\mathbf{Z}_{r}^{(:, \mathbf{R}_{i})}\) with \(\mathbf{Z}_{r}\) , considering \(\mathbf{B}_{i}\) to be a \(p_{r} \times p_{t}\) matrix, and introducing \(\ell_{0}\) regularization on rows of \(\mathbf{B}_{i}\) . This approach ensures that at most \(N_{i,r}\) rows of \(\mathbf{B}_{i}\) are non- zero and the indices of these rows correspond to \(\mathbf{R}_{i}\) . A common approach to ensure computability of \(\ell_{0}\) penalized problems is relaxation of the constraint to a convex norm instead. Here, groups of coefficients are considered that are either zero or non- zero simultaneously. This is a typical scenario covered by the group- Lasso (Yuan and Lin, 2006). In addition, to ensure sign- consistency of the selected regulators, an extension of the group- Lasso known as the cooperative- Lasso (coop- Lasso, Chiquet et al, 2012) is used to solve the regulator selection problem. + +<|ref|>text<|/ref|><|det|>[[112, 616, 721, 636]]<|/det|> +Applied to our case, the coop- Lasso solves the following optimization problem + +<|ref|>equation<|/ref|><|det|>[[165, 653, 882, 700]]<|/det|> +\[\mathbf{B}_{i,\mathrm{CL}} = \arg \min_{\mathbf{B}}\frac{1}{2}\sum_{j\in \mathcal{C}_{i}}\frac{1}{n_{1}\sigma_{i,j}^{2}}\left\| \mathbf{Z}_{t,1}^{(:,j)} - \mathbf{Z}_{r,1}\mathbf{B}^{(:,j)}\right\|_{F}^{2} + \lambda \sum_{k = 1}^{p_{r}}\pmb{w}_{i}^{(k)}(\left\| \mathbf{B}_{+}^{(k,:)}\right\|_{2} + \left\| \mathbf{B}_{-}^{(k,:)}\right\|_{2}), \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[111, 717, 886, 814]]<|/det|> +where \(C_{i}\) is the set of all target genes in cluster \(i\) , \(\lambda\) is a penalty parameter related to \(N_{i,r}\) above, controlling the amount of regulators that will be selected, \(\pmb{w}_{i}\) is a vector of weights that can be cluster- specific and will be described below, \(\mathbf{B}^{(k,:)}\) refers to the \(k\) - th row in \(\mathbf{B}\) , and \(\mathbf{B}_{+}^{(k,:)}\) as well as \(\mathbf{B}_{- }^{(k,:)}\) are defined by setting all negative or all positive elements in \(\mathbf{B}^{(k,:)}\) to zero, respectively. + +<|ref|>text<|/ref|><|det|>[[111, 826, 886, 899]]<|/det|> +The coop- Lasso selects which regulators, i.e. which rows of \(\mathbf{B}\) , are included in the model by setting the coefficients of the deselected groups to zero. In addition, the coop- Lasso aims for sign- coherence in each group and induces sparsity within groups, deselecting target genes not affected by some + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 886, 210]]<|/det|> +of the regulators. This means that typically the rows of the estimated coefficient matrix \(\mathbf{B}_{i,\mathrm{CL}}\) will have all positive or all negative sign. To determine the sign of each regulator, we assign \(\pmb{s}_{i}^{(k)} = \mathrm{sgn}\left(\sum_{j}\mathbf{B}_{i,\mathrm{CL}}^{(k,j)} / |C_{i}|\right)\) . The set of active regulators \(\mathrm{R}_{i}\) is equal to the indices of non- zero rows in \(\mathbf{B}_{i,\mathrm{CL}}\) . This way, the coop- Lasso in Eq. (2) provides an approximation to the optimization problem in Eq. (1) for fixed cluster membership. + +<|ref|>text<|/ref|><|det|>[[111, 223, 886, 397]]<|/det|> +To make computations on a matrix of coefficients efficient, we apply over- relaxed Alternating Direction Method of Multipliers (ADMM) (Boyd et al, 2011) to the coop- Lasso problem, splitting the variables such that one set is specific to the loss and the other is specific to the penalty, leading to simple solutions for each separate sub- problem. To solve the sub- problem corresponding to the penalty, we use an explicit form of the proximal operator of the coop- Lasso given in Chiquet et al (2012). To speed- up convergence we compute ADMM step- length and over- relaxation parameters in an adaptive fashion as described in Xu et al (2017). + +<|ref|>text<|/ref|><|det|>[[111, 409, 886, 507]]<|/det|> +Only the coefficients \(\mathbf{B}_{i,\mathrm{CL}}\) are optimized in Eq. (2). The variance parameters \(\sigma_{i,j}^{2}\) for \(j \in C_{i}\) are treated as known and weights are assumed to be given. We use a plug- in estimate for the variance parameters, which is computed using an ordinary least squares (OLS) estimate \(\mathbf{B}_{i,\mathrm{OLS}}\) of the regression coefficients of \(\mathbf{Z}_{t,1}^{(:,C_{i})}\) on \(\mathbf{Z}_{r,1}\) . Variances are then estimated in an unbiased way as + +<|ref|>equation<|/ref|><|det|>[[355, 525, 881, 562]]<|/det|> +\[\sigma_{i,j}^{2} = \frac{1}{n_{1} - p_{r}}\left\| \mathbf{Z}_{t,1}^{(:,j)} - \mathbf{Z}_{r,1}\mathbf{B}_{i,\mathrm{OLS}}^{(:,j)}\right\|_{2}^{2} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[111, 580, 886, 625]]<|/det|> +To debias the estimated coefficients in Eq. (2), weights are selected in a fashion similar to the adaptive Lasso (Zou, 2006) before estimation of \(\mathbf{B}_{i}\) . Setting the weights to + +<|ref|>equation<|/ref|><|det|>[[400, 644, 881, 675]]<|/det|> +\[\pmb{w}_{i}^{(k)} = \sqrt{|C_{i}| / ||\mathbf{B}_{i,\mathrm{OLS}}^{(k,:)}||_{2}} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[111, 700, 394, 717]]<|/det|> +improves the selection of regulators. + +<|ref|>text<|/ref|><|det|>[[111, 751, 886, 899]]<|/det|> +Determining cluster membership. In Step 2, cluster membership is re- allocated, based on the updated cluster structure determined in Step 1. To do so requires the estimation of non- negative coefficients \(\mathbf{B}_{i}\) and residual variances \(\sigma_{i,j}^{2}\) for all clusters and target genes. Previously, these were only estimated for the target genes contained in each cluster. In addition, re- estimation of the coefficients without penalization for the selected regulators and their signs removes the bias the coop lasso introduced and enforces the chosen signs. Finally, the updated cluster membership matrix \(\mathbf{\Pi}\) is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 195, 106]]<|/det|> +computed. + +<|ref|>text<|/ref|><|det|>[[111, 120, 886, 192]]<|/det|> +Given \(\mathbf{R}_{i}\) and \(\pmb{s}_{i}\) , we used sign- constrained linear regression using non- negative least squares (NNLS) (Meinshausen, 2013) to determine the coefficients \(\mathbf{B}_{i}\) for each cluster. To do so, the following optimization problem was solved + +<|ref|>equation<|/ref|><|det|>[[261, 207, 882, 247]]<|/det|> +\[\mathbf{B}_{i} = \arg \min_{\mathbf{B}}\frac{1}{2}\left\| \mathbf{Z}_{t,1} - \mathbf{Z}_{r,1}^{(\cdot ,\mathbf{R}_{i})}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}\right\|_{F}^{2}\mathrm{~such~that~}\mathbf{B}\geq \mathbf{0}, \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[111, 264, 886, 387]]<|/det|> +where the inequality is considered element- wise. To compute the NNLS coefficients efficiently, we implemented the algorithm described in Nguyen and Ho (2017). We modified the algorithm to be able to perform computations on a matrix of responses instead of on a single vector. To avoid unnecessary computation, responses are excluded from the computations once they reach the desired convergence criterion. + +<|ref|>text<|/ref|><|det|>[[111, 400, 821, 422]]<|/det|> +The variance parameters \(\sigma_{i,j}^{2}\) for each \(i = 1,\ldots ,K\) and all \(j = 1,\ldots ,p_{t}\) are re- estimated as + +<|ref|>equation<|/ref|><|det|>[[322, 437, 882, 476]]<|/det|> +\[\sigma_{i,j}^{2} = \frac{1}{n_{1} - |\mathbf{R}_{i}|}\left\| \mathbf{Z}_{t,1}^{(\cdot ,j)} - \mathbf{Z}_{r,1}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}_{i}^{(\cdot ,j)}\right\|_{2}^{2} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[111, 501, 886, 546]]<|/det|> +In the following, computations were performed on the second data split \(\mathbf{Z}_{t,2}\) and \(\mathbf{Z}_{r,2}\) to avoid bias towards the most complex regulatory programs. + +<|ref|>text<|/ref|><|det|>[[111, 559, 886, 708]]<|/det|> +Rag bag clustering is used to identify target genes that do not fit well in any cluster. To do so, the predictive \(R^{2}\) - value, denoted as \(R_{i,j}^{2}\) , is computed for each target gene and cluster from the residuals of predicting \(\mathbf{Z}_{t,2}\) from \(\mathbf{Z}_{r,2}\mathrm{diag}(\pmb{s}_{i})\mathbf{B}_{i}\) . The best predictive \(R_{i,j}^{2}\) across clusters for each target gene is recorded. If this best value is below a user- specified threshold \(\tau\) , then the gene is considered noise and badly predicted within all clusters. It is then placed in a noise cluster/rag bag. Only the remaining target genes are considered in the following steps. + +<|ref|>text<|/ref|><|det|>[[111, 721, 886, 896]]<|/det|> +To include prior knowledge of target genes that have a biological relationship, the algorithm allows the user to supply an indicator matrix \(\mathbf{J}\) of size \(q\times q\) such that \(\mathbf{J}^{(i,j)} = 1\) if genes \(i\) and \(j\) have a biological relationship, and zero otherwise. It is assumed that \(\mathbf{J}^{(i,i)} = 0\) to simplify computations below. By providing gene symbols for the target genes in \(\mathbf{Z}_{t}\) and the genes in \(\mathbf{J}\) , the algorithm determines the genes for which prior information is available. It is not required that the provided sets are equal as long as there is overlap. Assume for sake of notation that prior information is provided for all target genes in \(\mathbf{Z}_{t}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 85, 888, 235]]<|/det|> +For a fixed target gene \(j\) , \(\mathbf{J}^{(j,:)}\) contains 1's for those target genes which have a biological relationship with gene \(j\) . Given the current cluster membership matrix \(\mathbf{\Pi}\) , compute fractions \(f_{i,j} = \mathbf{J}^{(j,:)}\mathbf{\Pi}^{(i,:)^{\top}} / \sum_{g}\mathbf{\Pi}^{(i,g)}\) to encode the biological evidence supporting gene \(j\) to be in cluster \(i\) . In case cluster \(i\) is empty, set \(f_{i,j} = 0\) . These fractions are then normalized across clusters as \(p_{i,j} = f_{i,j} / \sum_{c}f_{c,j}\) . For numerical reasons, log- probabilities are used below. To avoid taking the log of zero, a small baseline parameter \(\alpha = 10^{- 6}\) is added to each \(f_{i,j}\) before normalization. + +<|ref|>text<|/ref|><|det|>[[111, 248, 886, 290]]<|/det|> +Given \(\mathrm{R}_{i}\) , \(\mathbf{s}_{i}\) , \(\mathbf{B}_{i}\) , and \(\sigma_{i,j}^{2}\) , the likelihood for target gene \(j\) across clusters is computed for observation \(l\) as + +<|ref|>equation<|/ref|><|det|>[[248, 285, 882, 340]]<|/det|> +\[\mathbf{L}_{j}^{(i,l)} = \frac{1}{\sqrt{2\pi\sigma_{i,j}^{2}}}\exp \left(-\frac{1}{2\sigma_{i,j}^{2}}\left(\mathbf{Z}_{t,2}^{(l,j)} - \mathbf{Z}_{r,2}^{(l,\mathrm{R}_{i})}\mathrm{diag}(\mathbf{s}_{i})\mathbf{B}_{i}^{(:,j)}\right)^{2}\right). \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[111, 348, 886, 398]]<|/det|> +These are then normalized by setting \(\mathbf{L}_{j}^{(i,l)}\leftarrow \mathbf{L}_{j}^{(i,l)} / \sum_{c}\mathbf{L}_{j}^{(c,l)}\) . To update the cluster membership for target gene \(j\) we then compute votes + +<|ref|>equation<|/ref|><|det|>[[327, 419, 882, 453]]<|/det|> +\[\pmb{v}_{j}^{(l)} = \arg \max_{i}\left[(1 - \mu)\log \mathbf{L}_{j}^{(i,l)} + \mu \log p_{i,j}\right] \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[111, 469, 888, 600]]<|/det|> +for each observation \(l\) and assign target gene \(j\) to the cluster which receives the majority of votes. The parameter \(\mu \in [0,1]\) can be used to control the strength of the prior on the allocation process. All target genes are processed in a random ordering to avoid introducing bias. Prior fractions for each gene are computed in each iteration using the previously updated cluster assignments as well as old cluster assignments for genes that have not been updated yet. + +<|ref|>text<|/ref|><|det|>[[111, 632, 886, 781]]<|/det|> +Determining convergence A history of the cluster membership matrices is kept and the algorithm is stopped when the current \(\mathbf{\Pi}\) at cycle \(c\) computed in Step 2 is equal to the cluster membership matrix in a previous cycle \(c_{0}\) . At this point, the algorithm would enter a loop of length \(c - c_{0}\) and is therefore exited. We consider the algorithm as converged if \(c - c_{0} = 1\) . If \(c - c_{0} > 1\) , the algorithm has found an unstable cluster configuration and results for each possible configuration within the loop are returned. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 133]]<|/det|> +Regulatory table In addition to all estimated quantities, the algorithm returns a \(p_{r}\times K\) regulatory table \(\mathbf{T}\) . It is computed as follows for \(k = 1,\ldots ,p_{r}\) and \(i = 1,\ldots ,K\) + +<|ref|>equation<|/ref|><|det|>[[232, 152, 882, 195]]<|/det|> +\[\mathbf{T}^{(k,i)} = \frac{1}{|C_{i}|}\sum_{j\in C_{i}}\mathbf{B}_{i,\mathrm{CL}}^{(k,j)}\qquad \mathrm{if}\quad \mathrm{median}\left\{\mathrm{corr}(\mathbf{Z}_{t,1}^{(:,j)},\mathbf{Z}_{r,1}^{(:,k)})\right\} >0.025 \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[112, 214, 886, 261]]<|/det|> +or \(\mathbf{T}^{(k,i)} = 0\) otherwise. It summarizes the average effect of an active regulator within a module, given that the regulator is sufficiently correlated with the target genes in that module. + +<|ref|>sub_title<|/ref|><|det|>[[112, 294, 369, 312]]<|/det|> +## Internal validation measures + +<|ref|>text<|/ref|><|det|>[[111, 334, 887, 509]]<|/det|> +In the following, we will consider three different types of predictive \(R^{2}\) - values. First, predictive \(R^{2}\) per module, written \(R_{i}^{2}\) for cluster \(i\) , is the predictive \(R^{2}\) for a given module \(i\) computed on the second data split. Second, predictive \(R^{2}\) per module excluding regulator \(k\) , written \(R_{i, - k}^{2}\) , is the predictive \(R^{2}\) for a given module \(i\) conditional on regulator \(k\) not being associated with the module. Thirdly, predictive \(R^{2}\) per module and target gene, written \(R^{2}i,j\) , is the predictive \(R^{2}\) for target gene \(j\) computed within the regulatory program of module \(i\) . The latter is also called cross- cluster predictive \(R^{2}\) for target gene \(j\) . + +<|ref|>text<|/ref|><|det|>[[111, 541, 886, 665]]<|/det|> +Guidance on selection of the penalty parameter To evaluate the clustering results we introduce two performance scores. The aim of our algorithm is to associate modules with their linearly most predictive regulators. As a measure of clustering quality, it is therefore natural to consider the predictive \(R^{2}\) per module, computed on the second data split with the coefficients \(\mathbf{B}_{i}\) for the selected regulators re- estimated on the first data split as in Eq. (5). We therefore compute + +<|ref|>equation<|/ref|><|det|>[[241, 682, 882, 730]]<|/det|> +\[R_{i}^{2} = \left\| \mathbf{Z}_{t,2}^{(:,C_{i})} - \mathbf{Z}_{r,2}^{(:,R_{i})}\mathrm{diag}(\pmb {s}_{i})\mathbf{B}_{i}\right\|_{F}^{2} / \sum_{j\in C_{i}}\left\| \mathbf{Z}_{t,2}^{(:,j)} - \frac{1}{n_{2}}\sum_{l = 1}^{n_{2}}\mathbf{Z}_{t,2}^{(l,j)}\right\|_{2}^{2}. \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[111, 746, 886, 896]]<|/det|> +In addition, we compute the importance of each regulator within a module as follows. The regulator and sign sets without regulator \(k\) , denoted as \(\mathbf{R}_{i, - k} = :\{g:g\in \mathbf{R}_{i},g\neq k\}\) and \(\pmb{s}_{i, - k}\) , are used to re- estimate the coefficients for module \(i\) . We then compute \(R_{i, - k}^{2}\) as the predictive \(R^{2}\) value on this reduced regulatory program and compute the importance of regulator \(k\) in module \(i\) as \(I_{i,k} = (R_{i}^{2} - R_{i, - k}^{2}) / R_{i}^{2} = 1 - R_{i, - k}^{2} / R_{i}^{2}\) . This is the ratio of the semi- partial correlation of the expression profile of regulator \(k\) with the expression profiles in module \(i\) to the overall \(R^{2}\) of module \(i\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 888, 397]]<|/det|> +with all regulators. A regulator that is more influential in predicting the target genes in module \(i\) will have larger importance. Importance values are typically less or equal to 1 and can be negative, which indicates that a regulator was introducing more noise than it compensated for in predictive strength. A decrease in \(R^{2}\) per module with an increase in the penalization parameter \(\lambda\) is expected, due to selection of less regulators and therefore less degrees of freedom in the linear models associated with each module. Importance is the ratio of the squared marginal correlation of a regulator with the target genes in a module to the overall \(R^{2}\) of that module. This implies that an increase of importance values is expected to concur with a decrease in number of selected regulators. Therefore, importance is expected to increase with an increase in the penalization parameter \(\lambda\) . Selection of the penalty parameter can therefore be guided by balancing these two scores. Predictive \(R^{2}\) should be high while importance should neither be too low nor too high, since the latter is typically indicative of too few selected regulators. + +<|ref|>text<|/ref|><|det|>[[112, 429, 886, 578]]<|/det|> +Guidance on selection of the number of modules Selection of the number of modules can be aided by cross- cluster predictive \(R^{2}\) . For each final module's selected regulators and signs, coefficients are re- estimated for all target genes. These are then used to compute the predictive \(R^{2}\) for each target gene in all available modules. Denote these values as \(R_{i,j}^{2}\) for target gene \(j\) when computed with regulators from module \(i\) . If there are \(K\) modules, this will result in a \(K \times p_{t}\) matrix. We then define the following score for each target gene \(j\) which has been clustered in module \(i\) + +<|ref|>equation<|/ref|><|det|>[[283, 600, 713, 634]]<|/det|> +\[a_{j} = R_{i,j}^{2},\quad b_{j} = \max (0,\max_{c\neq i}R_{c,j}^{2}),\quad S_{j} = \frac{a_{j} - b_{j}}{\max (a_{j},b_{j})}.\] + +<|ref|>text<|/ref|><|det|>[[111, 650, 886, 799]]<|/det|> +Due to the similarity with the well- known silhouette value (Rousseeuw, 1987), we call \(S_{j}\) the silhouette score for target gene \(j\) . The average silhouette score across all target genes gives a rough measure for clustering quality and can be used as a straight- forward tool to compare different module counts. A larger average silhouette score indicates that target genes on average are better located within modules. To determine the optimal number of modules, average silhouette score should be considered jointly with predictive \(R^{2}\) per module. A good clustering achieves high values in both scores. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 374, 107]]<|/det|> +## External validation measures + +<|ref|>text<|/ref|><|det|>[[112, 130, 885, 175]]<|/det|> +To evaluate the performance of simulations, we compared the groundtruth to the estimated clustering in multiple ways. + +<|ref|>text<|/ref|><|det|>[[112, 208, 886, 304]]<|/det|> +Adjusted Rand index To determine overall clustering performance, we use the adjusted Rand index (Hubert and Arabie, 1985), a well- known similarity measure between two clusterings. The standard adjusted Rand index ARI does not account for rag bag clustering. We therefore consider a modified version of the adjusted Rand index + +<|ref|>equation<|/ref|><|det|>[[371, 323, 882, 360]]<|/det|> +\[\mathrm{ARI}\cdot \frac{\#\mathrm{non - noise~target~genes}}{\#\mathrm{all~target~genes}}. \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[112, 385, 886, 507]]<|/det|> +Regulator selection To answer whether scRegClust selects the right regulators for each module we cannot compare regulators associated with each module directly, since estimated modules might not match clearly with the groundtruth. We therefore check correct selection of regulators for each target gene, as these are easily comparable with the groundtruth. For easier comparison we compute two measures: + +<|ref|>text<|/ref|><|det|>[[133, 532, 692, 551]]<|/det|> +1. True positive rate, as the proportion of correctly selected regulators, + +<|ref|>text<|/ref|><|det|>[[133, 570, 707, 588]]<|/det|> +2. False positive rate, as the proportion of incorrectly selected regulators. + +<|ref|>text<|/ref|><|det|>[[112, 614, 705, 632]]<|/det|> +Both of these measures are between 0 and 1 with larger values being better. + +<|ref|>text<|/ref|><|det|>[[112, 666, 886, 787]]<|/det|> +Implementation The package was implemented in R (R Core Team, 2022). Computationally expensive parts were written in C++ using the packages Rcpp (Eddelbuettel and Francois, 2011) and RcppEigen (Bates and Eddelbuettel, 2013). The packages igraph (Csardi and Nepusz, 2006) and ggplot2 (Wickham, 2016) were used for visualization. The R package clusterCrit was used to compute the Rand index. + +<|ref|>sub_title<|/ref|><|det|>[[113, 821, 352, 840]]<|/det|> +## Knockdown experiments + +<|ref|>text<|/ref|><|det|>[[112, 864, 886, 909]]<|/det|> +Target gene- editing by RNP complex delivery. As regulators of state 5, we chose DDR1 and SOX6. To disrupt gene function we delivered U3065MG cells between passage 16- 20 with SpCas9 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 312]]<|/det|> +2NLS and target specific multiguide (Synthego Gene Knockout Kit V2). The RNP complex was delivered using 4D-NucleoporatorTM X-unit and an optimized protocol with SF Cell line nucleofection kit S (Lonza, V4XC-2032). RNP complex formation was carried out as per the manufacturer's protocol. Briefly, RNP complex formation was carried out by mixing Cas9-2NLS and sgRNA in a molar ratio of 1:9 at room temperature, followed by adding the estimated amount of U3065MG cells dissolved in nucleofection solution and supplement 1. Experimental control used included non-targeting scrambled, Cas9 only, and mock (nucleofection of cells+RNP complex with no electric pulse). Nucleofection for all experiments was done using optimized settings in SF solution and program CA-137, and no major cell death was observed using these conditions. + +<|ref|>table<|/ref|><|det|>[[191, 350, 805, 507]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[354, 323, 640, 340]]<|/det|> +Table 3: sgRNA sequence information. + +
TargetSpeciessgRNA sequences (multi-guide)
DDR1HumanUGGAGAGCAGUGACGGGGAU
CUGCAAGUACUCCUCCUCCU
SOX6HumanGGCCCCCGCAUGCCGUCCC
UAUGGGGUGCAGAGGCAGAU
UUCCCUUGAGGUAAAUCUU
Non-targeting scrambled controlAAUUGGAGAGUGGGCUUGCU
GCACUACCAGAGCUAAACUCA
+ +<|ref|>text<|/ref|><|det|>[[111, 548, 886, 876]]<|/det|> +Genotyping and Sanger sequencing. Seven days post- nucleoporation, edited cells and non non- targeting scrambled control group were harvested. To evaluate gene editing, genomic DNA was isolated from all samples, and PCR amplification of the edited region was performed, followed by Sanger sequencing of the amplicon. Genotyping PCR was performed using Phusion hot start II high- fidelity PCR master mix (Thermo Scientific, F- 565L), on an Applied Biosystems miniAmp thermocycler. PCR cycling conditions are: Initial denaturation 980C (5min, 1cycle); amplification (30 cycles)- denaturation 980C (1min), annealing 620C (30s), extension 720C (30s); final extension (1cycle) 720C (10min). PCR amplicons from non- targeting and target- specific samples were purified using the Macherey- Nagel PCR clean- up kit (NucleoSpinTM PCR Clean- up, Cat 740609.250). PCR amplicons were sequenced using a single- end read at Eurofins Genomics (TubeSeq Supreme service, Germany). Analysis of the Sanger sequencing traces was performed using the freely available web tool "Inference of CRISPR edits" ("ICE", https://ice.synthego.com/) by entering the trace files (.ab1 files) from the edited samples and the non- targeting scrambled control. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[210, 112, 789, 262]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[377, 87, 618, 102]]<|/det|> +Table 4: Primer sequences used. + +
Target:DDR1
Primer pair for ampliconForward 5'-GGACCTCTACTTCCCCTCCA-3'
Reverse 5'- GTACAGCTGGAAGTGAGGAGA-3'
Sanger sequencing primer5'-GTACAGCTGGAAGTGAGGAGA-3'
Target: SOX6
Primer pair for ampliconForward 5'-GGTAGTTGTTTCATGATCTACACAA-3'
Reverse 5'-GGGTTTAGAGTGGACCAC-3'
Sanger sequencing primer5'-CACAAAAATAACTCTAAGGTCATCTATTTTC-3'
+ +<|ref|>text<|/ref|><|det|>[[112, 285, 888, 510]]<|/det|> +Temozolomide treatment. Cells from nontargeting scrambled control and target- specific edited samples were seeded to 384 well plates coated (Greiner blackwell optically clear bottom) with laminin, 7 days post nucleoporation (the editing from different batches of the experiment confirmed efficient knockdown on day 7). Twenty- four hours after seeding, cells were treated with temozolomide over a dose range from 750uM to 1.25uM. Treated cells were placed back into the incubator and an Alamar blue reading was taken 96 hours after treatment using a CLARIOstar plate reader (BMG Labtech). To know the effect of TMZ on viability fluorescence reading was measured, background subtracted values were normalized to DMSO control, and dose- response curves were plotted using GraphPad Prism. + +<|ref|>sub_title<|/ref|><|det|>[[114, 544, 401, 563]]<|/det|> +## Drug combination treatments + +<|ref|>text<|/ref|><|det|>[[112, 586, 886, 836]]<|/det|> +Cell culture The human glioblastoma cells U3065MG were obtained from the Human Glioma Cell Culture (HGCC) Biobank at Uppsala University (Xie et al, 2015). The cells were cultured in a mixture of Neurabasa1 (Gibco, #21103- 049) and DMEM/F12 (Gibco, #31331- 028) growth medium in 1:1 ratio, 1x B27, without Vitamin A (Gibco, #12587001), 1x N2 (Gibco, #17502001) and \(1\%\) Pen/Strep (Sigma- Aldrich, #P0781). The cells were grown as adherent culture on a laminin coated flasks (Sigma Aldrich, #L2020- 1MG), and were detached for splitting and seeding by TypIE without phenol red (Gibco, #12604039). The growth medium was supplemented with \(10\mathrm{ng / mL}\) recombinant human FGF- basic (Peprotech, #100- 18B) and \(10\mathrm{ng / mL}\) recombinant human EGF (Peprotech, #AF- 100- 15) each passage. The cells were regularly tested for Mycoplasma by either MycoAlert assay (Lonza, #LT07- 418) or MycoStrip (InvivoGen, #rep- mys- 50). + +<|ref|>text<|/ref|><|det|>[[113, 870, 886, 914]]<|/det|> +Cell viability assay U3065MG cells were seeded in a laminin coated 384- well plates (Thermo Scientific, #142761) at a density of 2,000 cells/well at a volume of 0.04 mL/well. The plates were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 87, 888, 467]]<|/det|> +centrifuged briefly at \(200\mathrm{xg}\) to collect the cells at the bottom of the plates and incubated for 24 h at \(37^{\circ}\mathrm{C}\) in humidified atmosphere with \(5\%\) CO2. Afterwards, the cells were either treated with temozolomide (Selleckchem, #S1237) or one of dasatinib (Selleckchem, #S1021). All compounds were dissolved in DMSO to a \(10\mathrm{mM}\) stock solution. The drugs were dispensed with D300e digital dispenser (Tecan) without additional dilution of the stock solution. Afterwards, the plates were centrifuged briefly at \(200\mathrm{xg}\) , and were incubated for 48 or 72 h at \(37^{\circ}\mathrm{C}\) in humidified atmosphere with \(5\%\) CO2. Cell viability was measured by the Alamar Blue assay (Invitrogen, #DAL1100), which was added 16 h prior the readout. After the 72 h time point, the growth medium was aspirated, and the cells were washed once with sterile PBS before addition of fresh medium. The cells were treated for additional 72 h with a combination between teamozolomide and dasatinib in a \(7\times 7\) dose- response matrix. 16 h prior the end point of the reaction, alamar blue was added into each well of the plates. The reduction of resazurin to resorufin was used as a proxy for determination of the cell viability. The raw measurements of the treatments were normalized to the DMSO control. The BLISS coefficient and the sensitivity of the combinations were calculated with SynergyFinder (Zheng et al, 2022) in R (R Core Team, 2022). + +<|ref|>sub_title<|/ref|><|det|>[[113, 507, 303, 528]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[112, 556, 888, 601]]<|/det|> +No data was generated for this work. The R- package can be found on GitHub at https://github.com/sven- nelander/scregclust. + +<|ref|>sub_title<|/ref|><|det|>[[114, 641, 330, 662]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[112, 689, 888, 814]]<|/det|> +We thank the Swedish Cancer Society, Swedish Childhood Cancer Foundation, Swedish Research Council and the Swedish Strategic Research Foundation for financial support. Some of the computations were enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at the National Supercomputer Centre (NSC), Linköping University partially funded by the Swedish Research Council through grant agreement no. 2022- 06725. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 86, 355, 107]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[112, 134, 886, 232]]<|/det|> +SN conceived the study. IL and FH developed the algorithm. FH wrote the R- package and IL performed the computational analyses, with support from RJ and SN. GP and AK planned and conducted the combination treatment experiments. SK planned and conducted the knockdown experiments. IL and FH wrote the first version of the paper and all authors assisted in editing. + +<|ref|>sub_title<|/ref|><|det|>[[113, 271, 327, 292]]<|/det|> +## Conflict of interest + +<|ref|>text<|/ref|><|det|>[[112, 322, 445, 339]]<|/det|> +The authors declare no conflict of interest. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 86, 430, 108]]<|/det|> +## Figures and Figure Legends + +<|ref|>sub_title<|/ref|><|det|>[[113, 135, 588, 155]]<|/det|> +## Figure 1. A regulatory-driven clustering method. + +<|ref|>text<|/ref|><|det|>[[130, 170, 888, 390]]<|/det|> +A A schematic overview of the algorithm pipeline from gene expression data to regulatory networks. B UMAP embedding of the PBMC cells colored according to immune cell type assignment. C The regulatory table inferred by scRegClust. Rows correspond to immune cell types and columns to regulators (TFs). The scale goes from high positive regulation (red) to high negative regulation (blue), as indicated by the key. D Venn diagram showing the overlap in identified regulators between SCENIC+ and scRegClust. E Regulator enrichment in the Human Protein Atlas compared to a baseline of all genes in the atlas. F Comparison between scRegClust and SCENIC+. + +<|ref|>sub_title<|/ref|><|det|>[[113, 426, 571, 446]]<|/det|> +## Figure 2. Method validation and analysis tools. + +<|ref|>text<|/ref|><|det|>[[113, 469, 885, 514]]<|/det|> +Results from a representative run of scRegClust on the data from simulation setup A (see the Supplementary Material for details on data generation). + +<|ref|>text<|/ref|><|det|>[[131, 533, 886, 905]]<|/det|> +A Boxplots of predictive \(R^2\) per module and importance per regulator shown across a progression of penalization parameters. Dashed lines indicate a region of solutions that demonstrates our selection rule. B Average true positive rate (TPR) shown against average false positive rate (FPR) in a ROC curve, illustrating the quality of groundtruth regulator identification. TPR and FPR are computed per target gene and then averaged. The corresponding penalization parameter is shown with each average. A table of ajusted Rand indices between the estimated clustering and the true clustering is shown as an inset. C Relative importance of the regulators within each module as estimated by scRegClust within each penalization level. Each plot represents one of the five estimated modules and each line shows the result for a penalization parameter. Colored tiles indicate that the regulator was included in the final model for the respective module. Green dots above the regulators for each module indicate regulators used during simulation of the data. The dashed lines correspond to the dashed lines in part A. They illustrate which regulators would be selected when the penalization parameter was chosen by our selection rule. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 133]]<|/det|> +Figure 3. scRegClust predicts a way of potentiating temozolomide treatment in U3065MG cells + +<|ref|>text<|/ref|><|det|>[[133, 149, 330, 167]]<|/det|> +A Flowchart of analysis. + +<|ref|>text<|/ref|><|det|>[[133, 175, 886, 245]]<|/det|> +B The merged regulatory table from scRegClust, with gene modules as columns and regulators as rows. Bottom panel display enrichments for each module against gene sets from (Larsson et al, 2021) and (Neftel et al, 2019)). + +<|ref|>text<|/ref|><|det|>[[133, 252, 886, 321]]<|/det|> +C Schematic of the predicted regulation of state 5. PDGFRA, DDR1, ERBB3 and SOX6 positively regulated state 5 and should be knocked down to block transitions to state 5. YBX1 negatively regulated transitions to state 5 and should be overexpressed to block transitions to state 5. + +<|ref|>text<|/ref|><|det|>[[133, 329, 886, 373]]<|/det|> +D Dose response curves for TMZ- treated cells with CRISPR/Cas9- mediated knockdown of DDR1 (top) and SOX6 (bottom). + +<|ref|>text<|/ref|><|det|>[[133, 380, 886, 450]]<|/det|> +E Schematic of the three combination treatment arms. Cells were either pre- treated with tyrosine kinase inhibitors (TKIs), TMZ or no pre- treatment, followed by combination treatment with TMZ and TKIs. + +<|ref|>text<|/ref|><|det|>[[133, 457, 886, 527]]<|/det|> +F Boxplot showing the mean bliss synergy score (BSS) for the three treatment arms, all replicates (top) and synergy landscape for the two combination experiments where the highest synergy scores were obtained (bottom). + +<|ref|>sub_title<|/ref|><|det|>[[113, 560, 637, 581]]<|/det|> +## Figure 4. The regulatory landscape of neuro-oncology. + +<|ref|>text<|/ref|><|det|>[[133, 596, 434, 613]]<|/det|> +A Schematic overview of the analysis. + +<|ref|>text<|/ref|><|det|>[[133, 621, 886, 846]]<|/det|> +B Middle panel is the regulatory table from scRegClust, with modules as columns and regulators (TFs) as rows. Top panel are annotation bars indicating what type and study each module originate from. Meta- modules (Box 1) are indicated by color in the bar below the middle panel. Bottom panel display enrichments for each module against a database of neuro- oncology related gene sets, derived from studies indicated by PubMed ID (PMID). Abbreviations: CT = cellular tumor, MVP = microvascular proliferation, GPM = glycolytic/plurimetabolic, pan = pseudopalisading cells around necrosis, NPC = neural progenitor cells, PPR = proliferating progenitor cells, IT = infiltrative tumor, OPC = oligodendrocyte progenitor cells, tRG = truncated radial glia. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 88, 477, 107]]<|/det|> +## Box 1. The regulatory meta-modules. + +<|ref|>text<|/ref|><|det|>[[112, 129, 886, 201]]<|/det|> +Description of the meta- modules defined in Figure 4. The colors to the left correspond to the middle color bar in Figure 4. The regulators listed regulate at least two of the individual modules included in the meta- module and all diseases represented in the meta- module are indicated by a filled box. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 0, 944, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[56, 861, 525, 882]]<|/det|> +
Figure 1 - A regulatory-driven clustering method.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 8, 911, 805]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[58, 861, 517, 882]]<|/det|> +
Figure 2 - Method validation and analysis tools.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 0, 940, 950]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[52, 954, 920, 992]]<|/det|> +
Figure 3 - scRegClust predicts a way of potentiating temozolomide treatment in U3065MG cells.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 40, 940, 790]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[57, 802, 587, 822]]<|/det|> +
Figure 4 - The regulatory landscape of neuro-oncology.
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[98, 48, 900, 396]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[335, 52, 662, 67]]<|/det|> +Box 1 - The regulatory meta-modules + +
Box 1 - The regulatory meta-modules
Meta-moduleInterpretationRegulatorsNormaGBMNBNB
1Cycling cellsHMGB1/2+, CENPA+, DEK+
2Necrosis, hypoxia, apoptosisYBX1-, ID3+
3Non-proliferating NPC2-like cellsHMGB1/2-, NEUROD1/2+, DEK+, SOX11+, SOX4+, TCF4+
4Astrocyte-like cellsID3+, ID4+, HOPX+, ZFP36L1+, MEF2C+
5Microglial-like mesenchymal cellsSPI1+, IRF8+, MAF+, ZFP36+, NME2+
6Stress responseFOXJ1+, JUN+, JUND+, FOS+, JUNB+, HMGN3+, NR4A1+
7Cycling cellsHMGB1/2+, CENPA+/-, DEK+, NEUROD1-
8FOXJ1-drivenFOXJ1+
9NB non-adrenergic cellsZNF90+, MEIS2+, MYC+, STAT1+
10FOXP2, RORA-driven infiltrative tumorFOXP2+, RORA+
11MYT1L-driven infiltrative tumorMYT1L+
12SOX17-driven mvpERG+, FLI+, ETS1+, EPAS1+, FOXQ1+, SOX17+, ZFP36+, ID3+
13OPC-like cellsOLIG1+, MYRF+, SOX10+, TSC22D4+, NKX6-2+/(-)/SOX6+
14NB adrenergic cellsZNF90-, MEIS2-, MYC-
15Mvp-signature mesenchymal cellsFOXS1+, HEY1+, AEBP1+, TBX18+, NME2+, MEF2C+, TSC22D4+
+ +<|ref|>text<|/ref|><|det|>[[56, 423, 423, 442]]<|/det|> +Box 1 - The regulatory meta- modules. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 86, 236, 106]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[110, 124, 888, 900]]<|/det|> +Aibar S, lez Blas CB, Moerman T, Huynh- Thu VA, Imrichova H, Hulselmans G, Rambow F, Marine JC, Geurts P, Aerts J, van den Oord J, Atak ZK, Wouters J, Aerts S (2017) SCENIC: single- cell regulatory network inference and clustering. Nat Methods 14: 1083–1086Arthur D, Vassilvitskii S (2007) K- means++: the advantages of careful seeding. In Proceedings of the 18th Annual ACM- SIAM Symposium on Discrete AlgorithmsBates D, Eddelbuettel D (2013) Fast and Elegant Numerical Linear Algebra Using the RcppEigen Package. Journal of Statistical Software 52: 1–24Bhat KPL, Balasubramaniyan V, Vaillant B, Ezhilarasan R, Hummelink K, Hollingsworth F, Wani K, Heathcock L, James JD, Goodman LD, Conroy S, Long L, Lelic N, Wang S, Gumin J, Raj D, Kodama Y, Raghunathan A, Olar A, Joshi K, et al (2013) B promotes radiation resistance in glioblastoma. Cancer Cell 24: 331–346Bohrer C, Pfurr S, Mammadzada K, Schildge S, Plappert L, Hils M, Pous L, Rauch KS, Dumit VI, Pfeifer D, Dengjel J, Kirsch M, Schachtrup K, Schachtrup C (2015) The balance of Id3 and E47 determines neural stem/precursor cell differentiation into astrocytes. EMBO J 34: 2804–2819Boyd S, Parikh N, Chu E, Peleato B, Eckstein J (2011) Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers. Foundations and Trends in Machine Learning 3: 1–122Butler A, Hoffman P, Smibert P, Papalexi E, Satija R (2018) Integrating single- cell transcriptomic data across different conditions, technologies, and species. Nature BiotechnologyCarro MS, Lim WK, Alvarez MJ, Bollo RJ, Zhao X, Snyder EY, Sulman EP, Anne SL, Doetsch F, Colman H, Lasorella A, Aldape K, Califano A, Iavarone A (2010) The transcriptional network for mesenchymal transformation of brain tumours. Nature 463: 318–325Chang ZF, Huang DY, Chi LM (1998) Serine 13 is the site of mitotic phosphorylation of human thymidine kinase. J Biol Chem 273: 12095–12100Chen S, Mar JC (2018) Evaluating methods of inferring gene regulatory networks highlights their lack of performance for single cell gene expression data. BMC Bioinformatics 19: 232 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 133]]<|/det|> +Chiquet J, Grandvalet Y, Charbonnier C (2012) Sparsity with sign- coherent groups of variables via the cooperative- Lasso. The Annals of Applied Statistics 6 + +<|ref|>text<|/ref|><|det|>[[112, 151, 886, 248]]<|/det|> +Couturier CP, Ayyadhury S, Le PU, Nadaf J, Monlong J, Riva G, Allache R, Baig S, Yan X, Bourgey M, Lee C, Wang YCD, Wee Yong V, Guiot MC, Najafabadi H, Misic B, Antel J, Bourque G, Ragoussis J, Petrecca K (2020) Single- cell RNA- seq reveals that glioblastoma recapitulates a normal neurodevelopmental hierarchy. Nat Commun 11: 3406 + +<|ref|>text<|/ref|><|det|>[[112, 265, 886, 310]]<|/det|> +Csardi G, Nepusz T (2006) The igraph software package for complex network research. InterJournal Complex Systems: 1695 + +<|ref|>text<|/ref|><|det|>[[112, 327, 886, 372]]<|/det|> +Cuddapah VA, Robel S, Watkins S, Sontheimer H (2014) A neurocentric perspective on glioma invasion. Nat Rev Neurosci 15: 455- 465 + +<|ref|>text<|/ref|><|det|>[[112, 389, 886, 486]]<|/det|> +Darmanis S, Sloan SA, Croote D, Mignardi M, Chernikova S, Samghababi P, Zhang Y, Neff N, Kowarsky M, Caneda C, Li G, Chang SD, Connolly ID, Li Y, Barres BA, Gephart MH, Quake SR (2017) Single- Cell RNA- Seq Analysis of Infiltrating Neoplastic Cells at the Migrating Front of Human Glioblastoma. Cell Rep 21: 1399- 1410 + +<|ref|>text<|/ref|><|det|>[[112, 503, 886, 548]]<|/det|> +Eddelbuettel D, Francois R (2011) Rcpp: Seamless R and C++ Integration. Journal of Statistical Software 40: 1- 18 + +<|ref|>text<|/ref|><|det|>[[112, 566, 886, 611]]<|/det|> +Ellison JA, de Vellis J (1994) Platelet- derived growth factor receptor is expressed by cells in the early oligodendrocyte lineage. J Neurosci Res 37: 116- 128 + +<|ref|>text<|/ref|><|det|>[[112, 629, 886, 699]]<|/det|> +Eze UC, Bhaduri A, Haeussler M, Nowakowski TJ, Kriegstein AR (2021) Single- cell atlas of early human brain development highlights heterogeneity of human neuroepithelial cells and early radial glia. Nat Neurosci 24: 584- 594 + +<|ref|>text<|/ref|><|det|>[[112, 717, 886, 788]]<|/det|> +Ferdous A, Singh S, Luo Y, Abedin MJ, Jiang N, Perry CE, Evers BM, Gillette TG, Kyba M, Trojanowska M, Hill JA (2021) Fli1 Promotes Vascular Morphogenesis by Regulating Endothelial Potential of Multipotent Myogenic Progenitors. Circ Res 129: 949- 964 + +<|ref|>text<|/ref|><|det|>[[112, 805, 886, 901]]<|/det|> +Filbin MG, Tirosh I, Hovestadt V, Shaw ML, Escalante LE, Mathewson ND, Neftel C, Frank N, Pelton K, Hebert CM, Haberler C, Yizhak K, Gojo J, Egervari K, Mount C, van Galen P, Bonal DM, Nguyen QD, Beck A, Sinai C, et al (2018) Developmental and oncogenic programs in H3K27M gliomas dissected by single- cell RNA- seq. Science 360: 331- 335 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 888, 210]]<|/det|> +Gangoso E, Southgate B, Bradley L, Rus S, Galvez- Cancino F, McGivern N, \(\zeta\) E, Kapourani CA, Byron A, Ferguson KM, Alfazema N, Morrison G, Grant V, Blin C, Sou I, Marques- Torrejon MA, Conde L, Parrinello S, Herrero J, Beck S, et al (2021) Glioblastomas acquire myeloid- affiliated transcriptional programs via epigenetic immunoediting to elicit immune evasion. Cell 184: 2454- 2470 + +<|ref|>text<|/ref|><|det|>[[113, 225, 886, 320]]<|/det|> +Gavish A, Tyler M, Simkin D, Kovarsky D, Gonzalez Castro LN, Halder D, Chanoch- Myers R, Laffy J, Mints M, Greenwald AR, Wider A, Tal R, Spitzer A, Hara T, Tirosh A, Puram SV, Suva ML, Tirosh I (2021) The transcriptional hallmarks of intra- tumor heterogeneity across a thousand tumors. bioRxiv + +<|ref|>text<|/ref|><|det|>[[113, 336, 886, 408]]<|/det|> +González- Blas C, De Winter S, Hulseimans G, Hecker N, Matetovici I, Christiaens V, Poovathingal S, Wouters J, Aibar S, Aerts S (2023) SCENIC+: single- cell multionic inference of enhancers and gene regulatory networks. Nat Methods 20: 1355- 1367 + +<|ref|>text<|/ref|><|det|>[[113, 422, 886, 468]]<|/det|> +Hafemeister C, Satija R (2019) Normalization and variance stabilization of single- cell RNA- seq data using regularized negative binomial regression. Genome Biol 20: 296 + +<|ref|>text<|/ref|><|det|>[[113, 483, 886, 580]]<|/det|> +Hamed AA, Kunz DJ, El- Hamamy I, Trinh QM, Subedar OD, Richards LM, Foltz W, Bullivant G, Ware M, Vladoiu MC, Zhang J, Raj AM, Pugh TJ, Taylor MD, Teichmann SA, Stein LD, Simons BD, Dirks PB (2022) A brain precursor atlas reveals the acquisition of developmental- like states in adult cerebral tumours. Nat Commun 13: 4178 + +<|ref|>text<|/ref|><|det|>[[113, 594, 887, 716]]<|/det|> +Hara T, Chanoch- Myers R, Mathewson ND, Myskiw C, Atta L, Bussema L, Eichhorn SW, Greenwald AC, Kinker GS, Rodman C, Gonzalez Castro LN, Wakimoto H, Rozenblatt- Rosen O, Zhuang X, Fan J, Hunter T, Verma IM, Wucherpfennig KW, Regev A, Suva ML, et al (2021) Interactions between cancer cells and immune cells drive transitions to mesenchymal- like states in glioblastoma. Cancer Cell 39: 779- 792 + +<|ref|>text<|/ref|><|det|>[[113, 732, 886, 828]]<|/det|> +Hovestadt V, Smith KS, Bihannic L, Filbin MG, Shaw ML, Baumgartner A, DeWitt JC, Groves A, Mayr L, Weisman HR, Richman AR, Shore ME, Goumnerova L, Rosencrance C, Carter RA, Phoenix TN, Hadley JL, Tong Y, Houston J, Ashmun RA, et al (2019) Resolving medulloblastoma cellular architecture by single- cell genomics. Nature 572: 74- 79 + +<|ref|>text<|/ref|><|det|>[[113, 843, 886, 915]]<|/det|> +Hu Y, Jiang Y, Behnam J, Ribeiro MM, Kalantzi C, Zhang MD, Lou D, ring M, Sharma N, Okawa S, Del Sol A, Adameyko I, Svensson M, Persson O, Ernfors P (2022) Neural network learning defines glioblastoma features to be of neural crest perivascular or radial glia lineages. Sci Adv 8: eabm6340 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 788, 107]]<|/det|> +Hubert L, Arabie P (1985) Comparing partitions. Journal of Classification 2: 193- 218 + +<|ref|>text<|/ref|><|det|>[[113, 124, 886, 220]]<|/det|> +Kamal A, Arnold C, Claringbould A, Moussa R, Servaas NH, Kholmatov M, Daga N, Nogina D, Mueller- Dott S, Reyes- Palomares A, Palla G, Sigalova O, Bunina D, Pabst C, Zaugg JB (2022) GRaNIE and GRaNPA: Inference and evaluation of enhancer- mediated gene regulatory networks applied to study macrophages. bioRxiv + +<|ref|>text<|/ref|><|det|>[[113, 238, 886, 282]]<|/det|> +Kamimoto K, Hoffmann CM, Morris SA (2020) CellOracle: Dissecting cell identity via network inference and in silico gene perturbation. bioRxiv + +<|ref|>text<|/ref|><|det|>[[113, 300, 886, 421]]<|/det|> +Kildisuite G, Kholosy WM, Young MD, Roberts K, Elmentaite R, van Hooff SR, Pacyna CN, Khabirova E, Piapi A, Thevanesan C, Bugallo- Blanco E, Burke C, Mamanova L, Keller KM, Langenberg- Ververgaert KPS, Lijnzaad P, Margaritis T, Holstege FCP, Tas ML, Wijnen MHWA, et al (2021) Tumor to normal single- cell mRNA comparisons reveal a pan- neuroblastoma cancer cell. Sci Adv 7 + +<|ref|>text<|/ref|><|det|>[[113, 439, 886, 510]]<|/det|> +Kling T, Ferrarese R, n D, Johansson P, Heiland DH, Dai F, Vasilikos I, Weyerbrock A, rnsten R, Carro MS, Nelander S (2016) Integrative Modeling Reveals Annexin A2- mediated Epigenetic Control of Mesenchymal Glioblastoma. EBioMedicine 12: 72- 85 + +<|ref|>text<|/ref|><|det|>[[113, 527, 886, 597]]<|/det|> +La Manno G, Siletti K, Furlan A, Gyllborg D, Vinsland E, Mossi Albiac A, Mattsson Langseth C, Khven I, Lederer AR, Dratva LM, Johnsson A, Nilsson M, Lönnerberg P, Linnarsson S (2021) Molecular architecture of the developing mouse brain. Nature 596: 92- 96 + +<|ref|>text<|/ref|><|det|>[[113, 615, 886, 685]]<|/det|> +Larsson I, Dalmo E, Elgendy R, Niklasson M, Doroszko M, Segerman A, Jörnsten R, Westermark B, Nelander S (2021) Modeling glioblastoma heterogeneity as a dynamic network of cell states. Mol Syst Biol 17: e10105 + +<|ref|>text<|/ref|><|det|>[[113, 703, 886, 799]]<|/det|> +Le Coz C, Nguyen DN, Su C, Nolan BE, Albrecht AV, Xhani S, Sun D, Demaree B, Pillarisetti P, Khanna C, Wright F, Chen PA, Yoon S, Stiegler AL, Maurer K, Garifallou JP, Rymaszewski A, Kroft SH, Olson TS, Seif AE, et al (2021) Constrained chromatin accessibility in PU.1- mutated agammaglobulinemia patients. J Exp Med 218 + +<|ref|>text<|/ref|><|det|>[[113, 816, 886, 912]]<|/det|> +LeBlanc VG, Trinh DL, Aslanpour S, Hughes M, Livingstone D, Jin D, Ahn BY, Blough MD, Cairn- cross JG, Chan JA, Kelly JJP, Marra MA (2022) Single- cell landscapes of primary glioblastomas and matched explants and cell lines show variable retention of inter- and intratumor heterogeneity. Cancer Cell 40: 379- 392 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 888, 133]]<|/det|> +Li YJ, He YF, Han XH, Hu B (2015) Dasatinib suppresses invasion and induces apoptosis in nasopharyngeal carcinoma. Int J Clin Exp Pathol 8: 7818- 7824 + +<|ref|>text<|/ref|><|det|>[[110, 150, 888, 245]]<|/det|> +Luo W, Lin GN, Song W, Zhang Y, Lai H, Zhang M, Miao J, Cheng X, Wang Y, Li W, Wei W, Gao WQ, Yang R, Wang J (2021) Single- cell spatial transcriptomic analysis reveals common and divergent features of developing postnatal granule cerebellar cells and medulloblastoma. BMC Biol 19: 135 + +<|ref|>text<|/ref|><|det|>[[110, 263, 886, 308]]<|/det|> +Martínez- Balbás MA, Bauer UM, Nielsen SJ, Brehm A, Kouzarides T (2000) Regulation of E2F1 activity by acetylation. EMBO J 19: 662- 671 + +<|ref|>text<|/ref|><|det|>[[110, 326, 887, 371]]<|/det|> +Marumoto T, Honda S, Hara T, Nitta M, Hirota T, Kohmura E, Saya H (2003) Aurora- A kinase maintains the fidelity of early and late mitotic events in HeLa cells. J Biol Chem 278: 51786- 51795 + +<|ref|>text<|/ref|><|det|>[[110, 389, 886, 434]]<|/det|> +Meinshausen N (2013) Sign- constrained least squares estimation for high- dimensional regression. Electronic Journal of Statistics 7 + +<|ref|>text<|/ref|><|det|>[[110, 452, 886, 498]]<|/det|> +Milano V, Piao Y, LaFortune T, de Groot J (2009) Dasatinib- induced autophagy is enhanced in combination with temozolomide in glioma. Mol Cancer Ther 8: 394- 406 + +<|ref|>text<|/ref|><|det|>[[110, 514, 888, 611]]<|/det|> +Neftel C, Laffy J, Filbin MG, Hara T, Shore ME, Rahme GJ, Richman AR, Silverbush D, Shaw ML, Hebert CM, Dewitt J, Gritsch S, Perez EM, Gonzalez Castro LN, Lan X, Druck N, Rodman C, Dionne D, Kaplan A, Bertalan MS, et al (2019) An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell 178: 835- 849 + +<|ref|>text<|/ref|><|det|>[[110, 629, 886, 674]]<|/det|> +Nguyen DK, Ho TB (2017) Accelerated anti- lopsided algorithm for nonnegative least squares. International Journal of Data Science and Analytics 3: 23- 34 + +<|ref|>text<|/ref|><|det|>[[110, 691, 888, 812]]<|/det|> +Ocasio JK, Babcock B, Malawsky D, Weir SJ, Loo L, Simon JM, Zylka MJ, Hwang D, Dismuke T, Sokolsky M, Rosen EP, Vibhakar R, Zhang J, Saulnier O, Vladoiu M, El- Hamamy I, Stein LD, Taylor MD, Smith KS, Northcott PA, et al (2019) scRNA- seq in medulloblastoma shows cellular heterogeneity and lineage expansion support resistance to SHH inhibitor therapy. Nat Commun 10: 5829 + +<|ref|>text<|/ref|><|det|>[[110, 831, 886, 902]]<|/det|> +Pham TH, Minderjahn J, Schmidl C, Hoffmeister H, Schmidhofer S, Chen W, Längst G, Benner C, Rehli M (2013) Mechanisms of in vivo binding site selection of the hematopoietic master transcription factor PU.1. Nucleic Acids Res 41: 6391- 6402 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 888, 133]]<|/det|> +Pratapa A, Jalihal AP, Law JN, Bharadwaj A, Murali TM (2020) Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat Methods 17: 147–154 + +<|ref|>text<|/ref|><|det|>[[111, 151, 886, 195]]<|/det|> +R Core Team (2022) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria + +<|ref|>text<|/ref|><|det|>[[111, 212, 888, 334]]<|/det|> +Richards LM, Whitley OKN, MacLeod G, Cavalli FMG, Coutinho FJ, Jaramillo JE, Svergun N, Riverin M, Croucher DC, Kushida M, Yu K, Guilhamon P, Rastegar N, Ahmadi M, Bhatti JK, Bozek DA, Li N, Lee L, Che C, Luis E, et al (2021) Gradient of Developmental and Injury Response transcriptional states defines functional vulnerabilities underpinning glioblastoma heterogeneity. Nat Cancer 2: 157–173 + +<|ref|>text<|/ref|><|det|>[[111, 352, 886, 397]]<|/det|> +Rousseeuw PJ (1987) Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20: 53–65 + +<|ref|>text<|/ref|><|det|>[[111, 414, 886, 510]]<|/det|> +Schmitt MJ, Company C, Dramaretska Y, Barozzi I, hrig A, Kertalli S, mann M, Naumann H, Sanchez- Bailon MP, Hulsman D, Glass R, Squatrito M, Serresi M, Gargiulo G (2021) Phenotypic Mapping of Pathologic Cross- Talk between Glioblastoma and Innate Immune Cells by Synthetic Genetic Tracing. Cancer Discov 11: 754–777 + +<|ref|>text<|/ref|><|det|>[[111, 527, 886, 623]]<|/det|> +Tirosh I, Venteicher AS, Hebert C, Escalante LE, Patel AP, Yizhak K, Fisher JM, Rodman C, Mount C, Filbin MG, Neftel C, Desai N, Nyman J, Izar B, Luo CC, Francis JM, Patel AA, Onozato ML, Riggi N, Livak KJ, et al (2016) Single- cell RNA- seq supports a developmental hierarchy in human oligodendroglioma. Nature 539: 309–313 + +<|ref|>text<|/ref|><|det|>[[111, 640, 886, 737]]<|/det|> +Uhlen M, Fagerberg L, Hallstrom BM, Lindskog C, Oksvold P, Mardinoglu A, Sivertsson A, Kampf C, Sjostedt E, Asplund A, Olsson I, Edlund K, Lundberg E, Navani S, Szigyarto CA, Odeberg J, Djureinovic D, Takanen JO, Hober S, Alm T, et al (2015) Proteomics. Tissue- based map of the human proteome. Science 347: 1260419 + +<|ref|>text<|/ref|><|det|>[[111, 754, 886, 850]]<|/det|> +Venkataramani V, Yang Y, Schubert MC, Reyhan E, Tetzlaff SK, mann N, Botz M, Soyka SJ, Beretta CA, Pramatarov RL, Fankhauser L, Garofano L, Freudenberg A, Wagner J, Tanev DI, Ratliff M, Xie R, Kessler T, Hoffmann DC, Hai L, et al (2022) Glioblastoma hijacks neuronal mechanisms for brain invasion. Cell 185: 2899–2917 + +<|ref|>text<|/ref|><|det|>[[111, 868, 886, 913]]<|/det|> +Vilella E, Gas C, Garcia- Ruiz B, Rivera FJ (2019) Expression of DDR1 in the CNS and in myelinating oligodendrocytes. Biochim Biophys Acta Mol Cell Res 1866: 118483 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 888, 185]]<|/det|> +Wang L, Babikir H, Müller S, Yagnik G, Shamardani K, Catalan F, Kohanbash G, Alvarado B, Di Lullo E, Kriegstein A, Shah S, Wadhwa H, Chang SM, Phillips JJ, Aghi MK, Diaz AA (2019) The Phenotypes of Proliferating Glioblastoma Cells Reside on a Single Axis of Variation. Cancer Discov 9: 1708–1719 + +<|ref|>text<|/ref|><|det|>[[112, 201, 888, 273]]<|/det|> +Wang L, Jung J, Babikir H, Shamardani K, Jain S, Feng X, Gupta N, Rosi S, Chang S, Raleigh D, Solomon D, Phillips JJ, Diaz AA (2022) A single- cell atlas of glioblastoma evolution under therapy reveals cell- intrinsic and cell- extrinsic therapeutic targets. Nat Cancer 3: 1534–1552 + +<|ref|>text<|/ref|><|det|>[[112, 289, 886, 388]]<|/det|> +Wang Q, Hu B, Hu X, Kim H, Squatrito M, Scarpace L, deCarvalho AC, Lyu S, Li P, Li Y, Barthel F, Cho HJ, Lin YH, Satani N, Martinez- Ledesma E, Zheng S, Chang E, é CG, Olar A, Lan ZD, et al (2017) Tumor Evolution of Glioma- Intrinsic Gene Expression Subtypes Associates with Immunological Changes in the Microenvironment. Cancer Cell 32: 42–56 + +<|ref|>text<|/ref|><|det|>[[112, 404, 886, 475]]<|/det|> +Wang Z, Sun D, Chen YJ, Xie X, Shi Y, Tabar V, Brennan CW, Bale TA, Jayewickreme CD, Laks DR, Alcantara Llaguno S, Parada LF (2020) Cell Lineage- Based Stratification for Glioblastoma. Cancer Cell 38: 366–379 + +<|ref|>text<|/ref|><|det|>[[112, 491, 886, 589]]<|/det|> +Weng Q, Wang J, Wang J, He D, Cheng Z, Zhang F, Verma R, Xu L, Dong X, Liao Y, He X, Potter A, Zhang L, Zhao C, Xin M, Zhou Q, Aronow BJ, Blackshear PJ, Rich JN, He Q, et al (2019) Single- Cell Transcriptomics Uncovers Glial Progenitor Diversity and Cell Fate Determinants during Development and Gliomagenesis. Cell Stem Cell 24: 707–723 + +<|ref|>text<|/ref|><|det|>[[112, 606, 886, 652]]<|/det|> +Wickham H (2016) ggplot2: Elegant Graphics for Data Analysis. New York, NY: Springer Science+Business Media, LLC, 2 edition + +<|ref|>text<|/ref|><|det|>[[112, 669, 886, 739]]<|/det|> +Xie XP, Laks DR, Sun D, Ganbold M, Wang Z, Pedraza AM, Bale T, Tabar V, Brennan C, Zhou X, Parada LF (2022) Quiescent human glioblastoma cancer stem cells drive tumor initiation, expansion, and recurrence following chemotherapy. Dev Cell 57: 32–46 + +<|ref|>text<|/ref|><|det|>[[112, 756, 886, 877]]<|/det|> +Xie Y, Bergström T, Jiang Y, Johansson P, Marinescu VD, Lindberg N, Segerman A, Wicher G, Niklasson M, Baskaran S, Sreedharan S, Everlien I, Kastemar M, Hermansson A, Elfneh L, Libard S, Holland EC, Hesselager G, Alafuzoff I, Westermark B, et al (2015) The Human Glioblastoma Cell Culture Resource: Validated Cell Models Representing All Molecular Subtypes. EBioMedicine 2: 1351–1363 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 87, 888, 160]]<|/det|> +Xu Z, Figueiredo MAT, Yuan X, Studer C, Goldstein T (2017) Adaptive Relaxed ADMM: Convergence Theory and Practical Implementation. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) + +<|ref|>text<|/ref|><|det|>[[110, 177, 886, 248]]<|/det|> +Yang H, Lee S, Lee S, Kim K, Yang Y, Kim JH, Adams RH, Wells JM, Morrison SJ, Koh GY, Kim I (2013) Sox17 promotes tumor angiogenesis and destabilizes tumor vessels in mice. J Clin Invest 123: 418- 431 + +<|ref|>text<|/ref|><|det|>[[110, 264, 886, 311]]<|/det|> +Yuan M, Lin Y (2006) Model selection and estimation in regression with grouped variables. Journal of the Royal Statistical Society Series B Statistical Methodology 68: 49- 67 + +<|ref|>text<|/ref|><|det|>[[110, 327, 886, 373]]<|/det|> +Zeitlin SG, Shelby RD, Sullivan KF (2001) CENP- A is phosphorylated by Aurora B kinase and plays an unexpected role in completion of cytokinesis. J Cell Biol 155: 1147- 1157 + +<|ref|>text<|/ref|><|det|>[[110, 390, 886, 461]]<|/det|> +Zheng S, Wang W, Aldahdooh J, Malyutina A, Shadbahr T, Tanoli Z, Pessia A, Tang J (2022) SynergyFinder Plus: Toward Better Interpretation and Annotation of Drug Combination Screening Datasets. Genomics Proteomics Bioinformatics + +<|ref|>text<|/ref|><|det|>[[110, 478, 886, 523]]<|/det|> +Zou H (2006) The Adaptive Lasso and Its Oracle Properties. Journal of the American Statistical Association 101: 1418- 1429 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 352, 230]]<|/det|> +SupplementaryTable1. xlsx supplementarytable2. xlsx supplementarytable3. xlsx appendixversion20231123. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/images_list.json b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..f372c6b08995cb59bbd8627f74fcda764eab3399 --- /dev/null +++ b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 142, + 118, + 780, + 444 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 123, + 110, + 880, + 650 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 112, + 103, + 884, + 884 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 113, + 101, + 881, + 730 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_1.jpg", + "caption": "Supplementary Figure 1", + "footnote": [], + "bbox": [ + [ + 123, + 110, + 880, + 641 + ] + ], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Chong et al. Allele Counts", + "footnote": [], + "bbox": [ + [ + 175, + 120, + 666, + 767 + ] + ], + "page_idx": 37 + } +] \ No newline at end of file diff --git a/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386.mmd b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386.mmd new file mode 100644 index 0000000000000000000000000000000000000000..c116190e3b658a1be38a25ec3272cb9431ff6cc6 --- /dev/null +++ b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386.mmd @@ -0,0 +1,432 @@ + +# MARS: Improved De Novo Peptide Candidate Selection for Non-Canonical Antigen Target Discovery in Cancer + +Hanqing Liao University of Oxford + +Carolina Barra Technical University of Denmark https://orcid.org/0000- 0002- 6836- 4906 + +Xu Peng University of Oxford + +Isaac Woodhouse University of Oxford https://orcid.org/0000- 0002- 8221- 7697 + +Arun Tailor University of Oxford + +Robert Parker University of Oxford + +Alexia Carre Institute Cochin, INSERM U1016, CNRS UMR8104 https://orcid.org/0000- 0001- 8144- 7194 + +Roberto Mallone Institute Cochin, INSERM U1016, CNRS UMR8104 https://orcid.org/0000- 0002- 9846- 8861 + +Michael Hogan Children's Hospital of Philadelphia https://orcid.org/0000- 0002- 6244- 2993 + +Wayne Paes University of Oxford https://orcid.org/0000- 0002- 0529- 2765 + +Laurence Eisenlohr University of Pennsylvania https://orcid.org/0000- 0002- 8475- 7910 + +Morten Nielsen Technical University of Denmark https://orcid.org/0000- 0001- 7885- 4311 + +Nicola Temette ( \(\square\) nicola.termette@ndm.ox.ac.uk) University of Oxford https://orcid.org/0000- 0002- 9283- 0743 + +Article + +Keywords: + +<--- Page Split ---> + +Posted Date: August 3rd, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1890352/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +1 MARS: Improved De Novo Peptide Candidate Selection for Non- + +2 Canonical Antigen Target Discovery in Cancer + +3 Hanqing Liao \(^{1,2}\) , Carolina Barra \(^{3}\) , Xu Peng \(^{1}\) , Isaac Woodhouse \(^{1,2}\) , Arun Tailor \(^{1,2}\) , Robert Parker \(^{1,2}\) , Alexia + +4 Carré \(^{4}\) , Roberto Mallone \(^{4}\) , Persephone Borrow \(^{2}\) , Michael J Hogan \(^{5}\) , Wayne Paes \(^{1,2}\) , Laurence C. Eisenlohr \(^{5,6}\) , + +5 Morten Nielsen \(^{3*}\) , Nicola Ternette \(^{1,2*}\) # + +6 + +7 1 The Jenner Institute, University of Oxford, OX37BN, UK + +8 2 Centre for Immuno-Oncology, Nuffield Department of Medicine, University of Oxford, OX37DQ, UK + +9 3 Technical University Denmark, Copenhagen, Denmark + +10 4 Université de Paris Cité, Institut Cochin, CNRS, INSERM, 75014 Paris, France + +11 5 Department of Pathology and Laboratory Medicine, Children's Hospital of Philadelphia, Philadelphia, PA, 19104. + +12 6 Department of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, + +13 19104. + +14 \*equal contribution + +15 \# to whom correspondence may be addressed: nicola.ternette@ndm.ox.ac.uk + +<--- Page Split ---> + +## ABSTRACT + +Cryptic Human Leukocyte Antigen (HLA)- presented peptide identification from unannotated genome sources is a priority for target antigen discovery for development of next generation immunotherapies in cancer. Current immunopeptidomic approaches utilize the integration of transcriptomics data to inform spectral interpretation, however, recent observations that tumour- associated antigen- encoding RNA levels are often low highlights limitations of such proteogenomic approaches1. + +levels are often low highlights limitations of such proteogenomic approaches1. + +We here employ a de novo sequencing approach with a refined, MHC- centric analysis strategy to detect non- canonical HLA- associated peptide sequences (HLAP) in cancer without integration of transcript sequence information. Our strategy integrates HLA binding prediction, peptide retention time prediction, and average local confidence scores culminating in the machine learning model MARS (MHC binding prediction, Average Local Confidence Score, and Retention time integration for improved de novo candidate Selection). + +We demonstrate increased HLA- I peptide identification sensitivity by benchmarking our model against de novo sequencing alone with a large synthetic HLA- I peptide library dataset. We further define the sensitivity of MARS by reanalysis of a published dataset of high- quality non- canonical HLAP identifications in human cancer cell line and tissue datasets and achieve almost 2- fold improvement of the full sequence recall (FSR) for high quality spectral assignments in comparison to de novo sequencing alone2. We minimize the false discovery rate (FDR) through a step- wise peptide sequence mapping strategy and are able to expand the reported non- canonical peptide space with an assignment accuracy above 85.7%. + +Finally, we utilize MARS to detect and validate IncRNA- derived peptides in human cervical tumour resections, demonstrating its suitability to discover novel, non- canonical peptide sequences in primary tumour tissue at reduced FDR, in the absence of transcriptomic sequencing data. + +<--- Page Split ---> + +## INTRODUCTION + +The analysis of HLA- bound peptide ligands using liquid chromatography mass spectrometry (LC- MS) technology (immunopeptidomics) is a rapidly evolving field that has advanced identification of T cell antigens including clinically relevant epitopes in autoimmunity, pathogen infection and cancer \(^{4 - 6}\) . These approaches have recently been expanded to identification of peptides from unannotated open reading frames (ORFs), or out- of- frame translated sequences, that are valuable for development of immunotherapeutic approaches to cancer, using the integration of RNA sequence information \(^{2, 4, 7 - 9}\) . However, recent observations that many HLA- presented peptides exhibit minimal RNA expression highlight the need for transcriptomics data- independent interrogation of the immunopeptidome for cancer antigen discovery. + +Software tools to interrogate LC- MS spectral data have widely been adopted from the field of proteomics, and extensive differences in performance have been observed for HLA- derived peptide data \(^{2, 10, 11}\) . Spectra interpretation software is based on three main strategies. The most common approach, the "database search" approach \(^{12 - 14}\) , relies on matching the masses of experimentally obtained peptide fragment spectra to all theoretically possible peptide fragments of sequences in a protein sequence list or "database". This approach is limited to identifications in the sequence database, and can therefore not be used for discovery of non- canonical peptides unless such information is added to known canonical protein databases. For increased sensitivity, this approach has been extended by integration of spectral matching assignment, which uses advanced deep learning and peptide fragmentation prediction to annotate the experimentally obtained peptide fragment spectra with high accuracy \(^{15 - 19}\) . Thirdly, de novo algorithms interpret the MS spectra without comparison to a protein sequence database, annotating the amino acid sequence directly from the spectral information available \(^{20 - 22}\) . Integration of de novo approaches and database searching (de novo- assisted database approach) has been shown to be particularly advantageous for HLA- peptide datasets \(^{10, 11, 23}\) . + +<--- Page Split ---> + +For the interrogation of immunopeptidome datasets with a focus on cancer- specific peptide ligands, proteogenomics approaches have been used to refine the database- search approach in order to identify peptide sequences that contain tumour- specific mutations or other non- canonical sources like functional RNA genes, intronic sequences, and/or reading frames in unannotated genomic origins that are generally not present in universal protein databases \(^{2, 7, 23 - 25}\) . Here, genomics and/or transcriptomics data is integrated to achieve a highly accurate, sample- specific database that can be used as template for a "database search" analysis approach. + +The incorporation of "personalised" binding prediction (MSRescue \(^{26}\) ), or difference between observed and predicted retention time (DeepRTplus \(^{27}\) , Deep Rescore \(^{19}\) ) to database- search based peptide identification workflows can also improve performance. Recently, machine learning has been applied in a personalized fashion in order to identify patient neoantigens \(^{28}\) . + +The purpose of this study is to increase the power of the de novo approach for identification of non- canonical HLA- I- associated peptides (HLA- Ip) without the use of a personalized template protein database. We report that by integration of (i) HLA binding affinity predictions and (ii) differences between observed and predicted retention time (RT) of candidate peptide sequences as additional factors, HLA- I peptide identification sensitivity can be improved over the baseline de novo sequencing annotation. We validate our findings by benchmarking our algorithm termed MARS on LC- MS data from synthetic HLA- I peptide libraries, and published non- canonical peptide sequences that were obtained from proteogenomic analyses. We then proceed and develop a hierarchical strategy to map MARS sequence candidates to their likely origin in the genome, and conclude in a final false discovery rate below 14.3% - offering opportunity for improved shortlisting of accurate sequence identifications from the generally observed global 65% FDR in de novo sequence data \(^{3}\) . Through this approach, we are able to expand the non- canonical peptidome of published data by hundreds of peptides. We finally detect and validate + +<--- Page Split ---> + +IncRNA derived peptides in primary human tumour tissue, demonstrating applicability for antigen discovery in cancer. + +## RESULTS + +## Training Data Selection and Processing + +Our aim was to construct and validate our computational model for improved de novo (template- independent) identification of HLA- I- associated peptides. We selected a panel of in- house single allele (CD4.221 cells expressing either A\*01:01, A\*02:01, A\*03:01, A\*11:01, B\*08:01, B\*44:02, B\*57:03, C\*03:03, or C\*03:04) and multi allele (Jurkat cell line: A\*03:01, B\*07:02, B35:03, C\*04:01, C\*07:02; and the C1866 cell line: A\*24:02. B\*40:02, B\*51:01, C\*03:03, C\*14:02) data sets to develop models and to evaluate model performances. + +We analyzed these datasets using Peaks DeNovo and also the Peaks DeNovo- assisted Database search (DB) using the SwissProt human canonical proteome 29. From the Peaks DeNovo search, we obtained up to 100 sequence candidates per spectrum (independent from the provided sequence database), with associated average local confidence (ALC) scores for each candidate sequence as a measure for the confidence of each sequence assignment. The Peaks DB search results were returned as a list of sequence assignments that matched to the SwissProt database. This latter search results in one sequence assignment per spectrum (with exception of L/I isomers) with an associated score (- 10/gP), which reflects the probability of a correct peptide spectrum match. The results are controlled by a simultaneous decoy database search which provides an accurate estimation of the false discovery rate to be expected in the result 30. + +In order to define a subset of "true sequences" to use as a training dataset, we defined spectra as confidently assigned if they were identified with Peaks DB with high confidence - 10/logP ≥ 20 resulting in an FDR of less than 1% on peptide- spectrum match (PSM) level. In addition, we only included peptides + +<--- Page Split ---> + +with a length of 8- 13 amino acids, and those that were assigned a low NetMHCpan 4.1 Eluted Ligand Rank (ELR) score \((ELR< 10)\) , aiming to remove peptides that are likely originating from co- purification during biochemical enrichment. All spectra that fulfilled these criteria were defined as the training dataset. + +Next, within the Peaks DeNovo search results, the "true" peptide was considered the Peaks DB assigned sequence identification with all constituent isoleucine (I) replaced by leucine (L), because Peaks DeNovo algorithm does not distinguish I from L. Only those spectra with the "true" (DB) peptides present in the list of the de novo peptide candidates were selected for model development and evaluation. In order to confirm that the I to L replacement does not negatively affect the NetMHCpan binding prediction, we tested NetMHCpan predictions within the test set for both L and I variants, and found that replacement of all I with L incurred significantly less divergence from the original result than vice versa (Supplementary Figure 1). + +All selected spectra were then split into 5 equal parts to form a five- fold cross- validation scheme, such that no two parts shared any common 8- mer substring training peptide sequence. The model trainings and evaluations were carried out using a five- fold cross- validation scheme, where each time a different combination of four parts was used as model- training set, while the performances were evaluated on the fifth (test) partition. Finally, all five- fold test set predictions were concatenated to obtain the final model performance. + +## Two factor model evaluation: Incorporation of NetMHCpan prediction improves de novo candidate selection + +We used the Peaks DeNovo result output as the baseline model. Accordingly, for each spectrum, the peptide candidate of the highest ALC was taken as the "identified" peptide further referred to as "ALC" result. In order to understand whether we could re- rank the de novo candidates by selecting the best candidate based on their binding prediction to the respective MHC allele of origin, we determined the + +<--- Page Split ---> + +NetMHCpan binding percentile Rank score for each of the 100 sequence candidates in the PEAKS DeNovo list, and calculated for each peptide candidate an affine combination of the ALC and NetMHCpan predicted Rank (ELR) scores to produce a new composite score. Then, for each spectrum, the peptide candidate of the highest composite score was taken as the identified peptide. Because this affine combination was controlled by a single weight parameter, alpha, and modelled ALC and ELR as two peptide identity determining factors, we named it "two- factor alpha model". We calculated the F- ranks which describe the rank of the true peptide sequence in descending order, divided by the total number of candidates, for each spectrum as an identification performance measure. We applied a grid search to obtain the optimal alpha value for a minimal mean F- rank within the training set. Then this alpha value was applied to the test for performance evaluation. All five test sets yielded one- sided paired Wilcoxon signed- rank test p- values <2.2E10- 16, indicating that the reduction of F- ranks by applying the two- factor alpha model in comparison to the ALC baseline model was highly significant (Supplementary Table 1, Figure 1A, B). + +To examine whether the improved F- rank led to higher identification sensitivity when top scoring peptide candidate per spectrum was taken as its identity, we employed another measurement, the full sequence recall (FSR), which depicts the proportion of identified (top one scoring) peptides being correct. A paired T- test using the mean FSRs from each of the five partitions also suggested that two- factor alpha model (M=0.782, SD=0.00461) significantly improved FSR over the baseline ALC model (M=0.731, SD=0.00554); t(4)=36.74, p=3.28E10- 6 (Figure 1C). + +Lastly, the optimal alpha obtained in the five- fold cross- validations were highly concordant, suggesting the two- factor alpha model was robust (Supplementary Table 1). In summary, these results suggested that incorporating ELR as a factor can improve PSM candidate selection, resulting in improved HLAP peptide identification compared to ALC alone. + +<--- Page Split ---> + +# Incorporation of RT prediction further improves de novo candidate selection: Final model evaluation + +To further improve PSM candidate selection, we incorporated DeepRTplus \(^{27}\) to predict the RT of input peptides. We trained the DeepRTplus model with our training data, and applied the learned model to the training data itself to derive an absolute value of differences between predicted and observed retention time (DRT). We then defined an affine combination of the ALC, ELR and DRT scores to define a three- factor composite model and used the grid search to find optimal alpha (weight on ELR) and beta (weight on DRT) values to minimize the mean F- rank in the training data. Finally, the optimal values of alpha and beta were used to assign combinatory scores to each peptide in the test data. + +We then proceeded with our evaluation strategy as before: a paired Wilcoxon test revealed that the F- ranks of the three- factor model were significantly lower than those of the two- factor alpha model (maximal of five p- values \(< 1.2 \times 10E- 10\) (Supplementary Table 1, Figure 1B)), suggesting that the three- factor model outperformed the two- factor model in promoting the true peptide identity amongst all other candidates in each spectrum. From the perspective of FSR, the performance gain was consistent with the F- Rank: the FSR by the three- factor model (M=0.786, SD=0.00601) was higher than FSR of two- factor alpha model (M=0.781, SD=0.00462) t(4)=3.14, p=0.0349 (Figure 1C). Further, the model “training” was found to be robust in that the optimal alpha and beta were stable across test sets (Supplementary Table 1). + +novo sequencing candidates. + +## MARS improves HLAP identification sensitivity across HLA-A and HLA-B alleles + +To substantiate our findings, we applied MARS to synthetic HLAP standard data for sensitivity evaluation. We hypothesized that the improvement of the re- ranking of de novo HLAP identification by the MARS + +<--- Page Split ---> + +model is robust in the sense that it can be applied to different datasets with fixed factor coefficients. The only requirement is a transfer learning scheme needed to calibrate the DeepRTplus model in order to accommodate the discrepancy introduced by the LC system setups and gradient conditions between different datasets. + +We utilized a large dataset published by Wilhelm, Zolg et al., of \(\sim 169,000\) synthetic HLA class I peptides acquired by liquid chromatography mass spectrometry (LC- MS) \(^{17}\) . Out of this dataset, we selected the 116 most frequent HLA alleles \(^{31}\) , comprising 35 HLA- A, 59 HLA- B, and 22 HLA- C alleles (Supplementary Table 2). We here included spectra from the 3xHCD method only, which was reported the best performing in the original study. Each peptide's HLA binding information was extracted from IEDB \(^{32,33}\) . We further applied a filter to exclude peptide- HLA associations with NetMHCapn ELR>10, which we deemed likely inaccurate assignments. This resulted in a collection of 28,919 MS2 spectra from 24,193 unique peptide sequences, of which the majority of sequences (26,452 spectra/22,324 peptide sequences; 91.5%) were listed in the Peaks DeNovo result candidate list (hence forward referred to as "recoverable") (Supplementary Table 2). We selected sequence- RT pairs from peptide candidates with ALC > 95 to calibrate the DeepRTplus model, while keeping the alpha and beta values constant. + +MARS outperformed ALC and the average F- ranks were significantly lower for MARS than ALC by paired Wilcoxon test ( \(p < 2.2 \times 10E- 16\) , Figure 2A). MARS FSR ( \(M = 0.907\) , \(SD = 0.0344\) ) was also significantly higher than ALC FSR ( \(M = 0.859\) , \(SD = 0.0744\) ): \(t(115) = 8.4295\) , \(p = 1.147e- 13\) (Figure 2B). MARS could again increase the number of identified peptide sequences over ALC (Figure 2C), and overall improvement was 7.23%. By grouping HLA alleles based on gene locus (A, B, and C), it can be seen that for HLA- A and B loci, the peptide FSRs can be significantly improved by MARS, but for HLA- C, the difference between ALC FSR and MARS FSR did not reach significance, possibly due to a lower number of spectra included for HLA- C (Figure 2D, Supplementary Table 2). When we further looked at the improvement of sequence identifications for all individual alleles, we noted that for HLA- A, only HLA- A\*01:01 and HLA- A\*31:01 did not benefit from + +<--- Page Split ---> + +MARS over ALC, while MARS improved results for all other HLA- A assigned peptide sequences. A high variability in performance across different HLA- B alleles was observed, with striking improvement in peptide identifications using MARS for HLA- B\*27:05 and other individual HLA- B alleles, while HLA- B35 did not benefit from MARS rescoring (Figure 2D). + +## MARS identifies previously identified HLAP of non-canonical origins without integration of sample-specific RNA sequencing data + +To further assess the identification performance of the MARS model, and in particular the performance in the non- canonical HLA peptide space, we utilized a published dataset from Chong et al. comprising nine LC- MS datasets for five melanoma cell lines and two sets of lung cancer and healthy control tissues spanning a wide range of HLA alleles 2. The authors identify and report 508 non- canonical, HLA- presented peptides mostly sourced from IncRNA genes, which are not present in the canonical human proteome. The authors achieved identification of these peptides sequences using an immunopeptidomics approach which carefully integrates RNA sequencing and ribosome profiling data for generation of personalized, sample- specific transcript assemblies. These were then used for in- silico translation into protein databases and mapping of non- canonical, HLA- presented sequences in the immunopeptidomics data using a standard database- search. + +Here, we wanted to understand whether MARS could recover these sequences without the integration of any of the additional RNA sequencing datasets used in the original publication. Firstly, we searched the Chong et al. datasets against the SwissProt human protein database using Peaks software in order to exclude spectra that were assigned to the canonical human proteome from our search. N- terminal acetylation and oxidation were included in the search parameters in order to match the analysis settings used in the original publication. We identified a total of 83,295 canonical peptides using the PEAKS + +<--- Page Split ---> + +database search then applied the MARS model to the remaining spectra, returning an additional 333,011 unique peptide identifications. + +Overall, MARS was able to recover 227 of the 508 (45%) reported IncRNA derived peptides reported by Chong et al.. To note that only 272 out of the 508 peptides were listed as a candidate in the Peaks DeNovo search results, therefore the other sequences were "non- discoverable" by MARS. MARS therefore recovered \(83\%\) (227/272) of the possible "de novo- identifiable" space. + +A clear separation of the number of peptides recovered with increasing false discovery rate was observed when applying different MARS score thresholds across all nine samples in comparison to ALC alone (Figure 3A). We therefore decided to define three major score confidence regions across this space, defining a MARS score threshold for increasing FDR: region 1: \(< 30\%\) FDR (MARS score \(>94.7\) ), region2: \(>30\%\) FDR (MARS score \(>89.7\) ), and region 3: \(>45\%\) FDR (MARS score \(>80\) ) (Figure 3A). + +With a stringent MARS score threshold of \(>94.7\) , and an estimated \(30\%\) FDR, we were able to recover 175/272 (64%) of sequences, while ALC alone could recover only 101/272 (37%) peptides (Figure 3B, Table S2, Supplementary Table 3), resembling an almost 2- fold improvement in FSR. + +Integration of a sequence mapping strategy for MARS peptide candidates confirms non- canonical HLAP origin. + +We next aimed to assign the 333,011 MARS identified sequences in the interrogated dataset to their respective likely gene origin. We therefore developed a stratified search approach to associate MARS identified HLAP to a wide range of possible origins across the human genome. We defined peptides mapping to (i) canonical origins mapping to known protein isoforms as "Human Protein", (ii) single amino acid substitutions as "1AA substitution", and (iii) other locations in the human genome as "non- canonical" (Figure 3C, and Methods section). + +<--- Page Split ---> + +Further to the 83,295 peptides that had initially been identified by the PEAKS database search, 19,619 sequences were additionally identified as isoforms of the known proteome space as assigned and reported by SwissProt, Tremble and Ensemble by MARS. A further 22,934 MARS sequences mapped to canonical human proteins with exactly 1 aa substitution, which we considered as likely products of germline or tumour- specific mutations (neoantigens). A total of 11,339 peptides were linearly mapped to non- canonical sequences in the human genome. For 279,119 MARS peptides we could not find an accurate match within the human genome with less than 1 aa distance, and we therefore considered these sequences as distant from the human proteome ("unmatched", Figure 3D, left panel, see Discussion). + +193/272 of the Chong et al. published de novo- identifiable peptides were as expected part of the "noncanonical" peptide category as defined here. In addition, with our search strategy, we found 16 peptides reported in our DB database search, an additional 23 peptides matched to known human isoforms reported by SwissProt, Trembl and/or Ensemble, and 30 peptides to human proteins with 1 AA substitution (Figure 3D, right panel). + +The 194 peptides identified by both Chong et al. and 194 MARS had an overall higher MARS score (mainly falling into regions 1 and 2) in comparison to the 13,724 additional non- canonical peptide candidates MARS recovered from the published datasets (Figure 3D, E), indicating that these peptides were recovered with high confidence and that a higher MARS score cut- off can likely limit the FDR significantly. + +## MARS extends the non-canonical peptide space by hundreds of peptides at an FDR <14.3% + +In order to determine the actual FDR experimentally for the novel MARS non- canonical peptide identifications, we selected 97 peptides that closely resembled the observed MARS score distribution and synthesized peptides for spectral validation and FDR evaluation (Figure 4A, Supplementary Figure 3). We also included 6 peptides identified by both Chong et al. and MARS, and 3 peptides matching to the human + +<--- Page Split ---> + +proteome as (likely positive) controls. We found 36 out of 97 sequences to be assigned correctly, resulting in an FDR of \(14.3\%\) (region 1, MARS score \(>94.8\) ), \(45.7\%\) (region 2, MARS score \(>89.7\) ), and \(66.0\%\) (region 3, MARS score \(>80\) ) (Figure 3F,G, Supplementary Table 6). All 9 positive control peptides were confirmed. When considering only peptides that were matched to a genome origin (excluding all "unmatched" identifications), there were 42 peptides selected, the FDR could further be reduced to \(0\%\) (region 1, MARS score \(>94.8\) ), \(39.5\%\) (region 2, MARS score \(>89.7\) ), and \(63.0\%\) (region 3, MARS score \(>80\) ) (Figure 3G). This meant that all peptides with a MARS score of \(>94.8\) that mapped to a genome origin were validated in this experiment. + +We then proceeded to further stratify the non- canonical peptides using GENECODE "Biotype" definitions, including functional RNA genes and IncRNA genes, protein coding regions, unannotated regions, and pseudogenes as detailed in the Method section (Figure 3H). For this, we did not apply a hierarchical strategy that would consider certain origins more likely than others, but argue that without sample specific genome/RNA level evidence, peptides with multiple possible origins cannot be faithfully assigned to a single source. + +Out of the 508 peptides originally published by Chong et al., 399 peptides are defined as "non- canonical" according to our categorization, which excludes peptides with single amino acid substitutions (57 peptides, "1 AA Substitutions" category) and those that map to canonical sequence databases in our workflow (52 peptides, "Human Protein" category). MARS was able to expand the originally published, in our definition "non- canonical" peptide identifications from 399 originally published peptides in this category to 1870 peptides from non- canonical origins at a an FDR between \(0< \text{FDR}< 14.3\%\) (Figure 3H). + +MARS uniquely identifies novel non- canonical IncRNA- derived HLA peptide sequences in primary tumour tissue + +<--- Page Split ---> + +In order to establish whether MARS could identify novel non- canonical, IncRNA- derived peptides in primary tumour tissue, we applied MARS to a set of in- house immunopeptidomics data from ten primary cervical tumour resections spanning 35 HLA alleles (Supplementary Figure 2, Supplementary Table 5). + +Apart from 32,366 Peaks database search identifications using the Peaks DeNovo- assisted database search, we could expand identifications using MARS by a further 77,129 assigned sequences. We mapped 6,551 further peptides to human source proteins, 5,828 peptides with a single amino- acid substitution, and 3,980 peptides to non- canonical origins (Figure 4A). 60,770 peptides remained unmatched. + +We then further stratified the obtained non- canonical HLAP according to the annotation of their origins as described in the last section (Figure 4B). Peptides uniquely mapped to protein coding regions formed the largest subcategory, comprising of 1150 HLAP, while 811 peptides remained unmatched. We found overall 1,423 peptides associated with IncRNA origin. In order to validate that the FDR of these identifications was as expected from the Chong et al. dataset, we again selected a set of peptides for spectral validation. We chose 19 peptide candidates that had a MARS score of >89.7 (region 1 and 2), and we further prioritized peptides if more than one sequence supported the identification of an IncRNA gene. We validated 11/19 spectra (42.2% FDR), which was in the range of the expected FDR of 39.5% from previous results for Region 1 and 2 (MARS score >89.7). For Region 1 only (MARS score >94.7) we again yielded high accuracy in annotation with 7/8 correct assignments and an FDR of 12.5% (Figure 4C, Table 1). + +In summary, MARS identifies thousands of peptides in addition to peptides identified with standard database- search approaches at an accuracy of above 85.7%, as confirmed in this independent primary tumour dataset, paving the way for transcriptomics- independent antigen discovery workflows to inform the development of novel cancer immunotherapies. + +<--- Page Split ---> + +A main limitation of non- canonical antigen discovery is the requirement of prior knowledge of expected protein sequences translated in a given sample. We here describe how we can rescue identification of non- canonical IncRNA- derived HLA- associated peptides from LC- MS/MS data without the requirement for integration of RNA sequencing data or, in fact, any other sample- specific sequence information other than HLA type, using a discovery engine that improves candidate selection from de novo sequencing results. + +Since in immunopeptidomics the lengths of peptides and positions of individual amino acids in the peptides are crucial for the HLA binding and T- cell response, we here used a study point- wise- full- sequence- recall (PSTR) to assess the development of MARS instead of improvements of single tag recall STR which assesses the percentage of individual amino acids being correctly identified \(^{3,34}\) . + +We evaluated the increased sensitivity using an external dataset from Wilhelm et al. \(^{17}\) , and confirmed an overall increase in performance of MARS de novo candidate selection in comparison to the best ALC scoring peptide as reported by Peaks DeNovo by over \(7\%\) . Interestingly, we saw outstanding performance for B27, B40 and B41 alleles. B27 has the requirement of an arginine at position two within the peptide sequence, while B40 and B41 are known to require a glutamic acid in this position. Hence, chemical properties of the peptide ligands are expected to be rather distinct, and we were not able to explain this much increased level of performance. + +We did not observe a significant improvement for identification of HLAP originating from HLA- C. Since peptide data was acquired in equimolar ratios for the HLAP library datasets, this could not be due to lower expression of HLA- C in comparison to HLA- A and - B as generally observed across different datasets. Since NetMHCpan- 4.1 performs equally well across all three loci, we conclude that most likely we did not evaluate a sufficient number of HLA- C peptide ligands to reach a significantly better performance. + +We then proceeded to evaluate how many peptide identifications could be rescued from a list of published non- canonical HLAP that had been generated using a proteogenomics approach. MARS was + +<--- Page Split ---> + +able to identify \(45\%\) of the reported sequences highlighting the power of MARS to identify non- canonical antigens in the absence of transcriptomics and RiboSeq datasets. + +One limitation remains the high false discovery rate associated with de novo sequencing results. With MARS, the FDR could be further limited by assigning the peptide sequences to a possible origin in the human genome and transcriptome, resulting in a manageable FDR range below \(14.3\%\) for MARS scores above \(94.8\) , compared to generally anticipated FDR of \(65\%\) for de novo sequencing results. Furthermore, considering only those sequences that mapped to a genome origin resulted in a substantial decrease of the FDR (to \(0\%\) in our experimental validation), demonstrating that MARS offers identification of non- canonical peptide sequences at low FDR, and can therefore expand the peptide identification beyond sophisticated proteogenomics approaches alone. + +We here focused our analysis on non- canonical peptides derived specifically from IncRNA genes. The question remains what the origin is of the parge proportion of non- assigned peptide sequences identified by MARS. Post- translationally modified peptides, peptides derived from proteasomal fusion events, and peptides of non- human origin are all potential candidates to explain this space, which remains to be further investigated. Finally, the observed higher proportion of sequences with lower MARS score within the unmatched peptides indicates a higher FDR in this subcategory, and further suggests that these spectra may not contain sufficient information for accurate sequence matching. This observation suggests that MARS performance could further be improved by measurement of higher quality spectra for each peptide ion species, i.e. by increased ion accumulation times. + +We could not validate any cysteine containing peptide assignments, confirming the known observed underrepresentation of cysteines in immunopeptidomics data \(^{35,36}\) , and highlighting the need for specific modification of the otherwise highly reactive amino acid. + +Finally, we were able to shortlist novel IncRNA- derived peptide candidates from primary human tumour tissue. We shortlisted HLAP if more than one sequence supported the identification of a specific IncRNA + +<--- Page Split ---> + +gene. We validated the sequence identifications with spectral matching to their synthetic counterpart, and we validated the FDRs measured in our previous evaluation. Note that we did not set a threshold for number of I and L in the peptide sequence, which could theoretically further reduce the FDR. + +Finally, we identified and validated 11 peptides originating from IncRNA origins in human cervical cancer tissue. Peptide KVHVFLVKKT, which originated from IncRNA HLA- F- AS1, plays an oncogenic role in both colorectal cancer and breast cancer \(^{37,38}\) . The identified peptides MTMSTILSK and MTMSTILSKK were derived from IncRNA Inckb.42318 also known as MIR4458HG (ENSG00000247516), which was reported as highly expressed IncRNA in 407 TCGA evaluated ovarian tumors \(^{39}\) . It has further been reported that, together with CYTOR, and MAPTAS1, MIR4458HG was an independent prognostic factor in breast cancer patients (https://doi.org/10.21203/rs.2.24062/v1, preprint), with high expression associated with lower hazard ratios. + +We conclude that MARS is a suitable platform for non- canonical antigen discovery in cancer immunopeptidomics datasets in the absence of high quality RNA sequencing data. We further highlight that the increases and unbiased results reported provide an exciting opportunity to expand non- canonical cancer antigen discovery to post- translationally modified peptides, peptides derived from proteasomal fusion events, and non- human sources, i.e. the microbiome. + +<--- Page Split ---> + +## METHODS + +## Human tissue material + +Tissue samples were obtained from Tissue Solutions Ltd.. Ethical approval was granted by the Central University Research Ethics Committee (CUREC) of the University of Oxford under reference R68126/RE001. + +## Cell lines + +All cell lines were mycoplasma- negative and cultured at \(37^{\circ}C\) in the presence of \(5\% \mathrm{CO_2}\) . The 721.221 HLA- class I deficient cell line that only expresses residual amounts of \(\mathsf{C}^{*}01:01^{40}\) transfected with CD4 (CD4.221) and selected HLA class I alleles ( \(\mathsf{A}^{*}01:01\) , \(\mathsf{A}^{*}02:01\) , \(\mathsf{A}^{*}03:01\) , \(\mathsf{A}^{*}11:01\) , \(\mathsf{B}^{*}08:01\) , \(\mathsf{B}^{*}44:02\) , \(\mathsf{B}^{*}57:03\) , \(\mathsf{C}^{*}03:03\) \(\mathsf{C}^{*}03:04\) ) was kindly provided by Prof. Masafumi Takaguchi, Kumamoto University, Japan. Cells were grown in RPMI 1640 medium (Thermo Fisher) containing \(10\%\) fetal bovine serum (FBS), \(2 \mathrm{mM}\) l- glutamine, \(100 \mathrm{U / mL}\) penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin (R10). R10 medium was supplemented with \(0.15 \mathrm{mg / mL}\) hygromycin B (Thermo Fisher) to maintain HLA- I expression as previously described \(^{40,41}\) . + +## Baseline ALC Model, or One-Factor Model + +Peaks DeNovo sequencing algorithm was performed using Peaks X (Bioinformatics Solutions) on each MS2 spectrum in the studied data to produce peptide- spectrum matches (PSMs). The quality of PSMs are measured by the average local confidence (ALC, represented by symbol \(A\) in the formula), which ranged in [0,100], with 100 being the best PSM quality. We selected PSMS with ALC in range [50, 100] to ensure reasonable input quality. The top- ALC peptide of each spectrum was used as its identity. No training was required for this one- factor model. + +<--- Page Split ---> + +## Integration of HLA Binding Affinity Ranking Factor (R): Two-Factor \(\alpha\) Model + +To obtain numerical characterization of HLA- binding affinities, The PSM candidates were input into NetMHcpan 4.1 model, and the output eluted ligand percentile rank (ERL, represented by symbol \(R\) in the formula) predictions were used to resemble the likelihood of HLA presentation of the corresponding peptides. The ELR predictions were considered robust across different biological samples and HLAs. The output ELR ranged between [0, 100], with the smaller ranks having higher likelihood of HLA presentation. We only selected PSMs with ERL in the range [0,10] to exclude non- binding peptides. By incorporating ELR into PSM candidates selection, a two- factor \(\alpha\) model can be derived as: \((1 - \alpha)A + \alpha (100 - R)\) which is a positive affine combination of ALC (A) and ELR (R). The parameter \(\alpha\) is in the range of [0,1], by the property of affine combination, the model outputs range is kept within [0,100] to ensure factors are normalized and of comparable scale. To obtain numerical characterization of HLA- binding affinities, The PSM candidates were input into NetMHcpan 4.1 model, and the output ELR (eluted ligand percentile rank) predictions used to resemble the likelihood of HLA presentation of the corresponding peptides. + +## Integration of the Retention Time Factor (T), Three-Factor MARS Model + +We used an adapted version of DeepRTplus, which is a special type of deep neural network model termed capsule neural networks (CapsNets), to predict the RT of input peptides. Before the trained DeepRTplus model can be used for predicting into a new set of peptides, it needs to be recalibrated for each chromatographic setup, since peptides" RT observations are known to be chromatography dependent. We used the sequence- RT pairs from de- novo peptides with \(\mathsf{ALC} > 95\) as a small but reliable set of identified peptides to re- calibrate the pre- trained DeepRTplus model, forming a transfer learning strategy to increase prediction accuracy. To develop the three- factor model by incorporating the difference in peptide chromatographic retention time (DRT, represented by symbol \(T\) in the formula) out of the DeepRTplus output as the input, firstly the + +<--- Page Split ---> + +absolute difference between observed and predicted RT: \(\Delta T = |T_{predicted} - T_{observed}|\) was computed. Secondly, T was computed as the \(\Delta T\) s normalised to the interval of [0,100] to match the range of ALC and ELR: \(T = 100(\Delta T_{max} - \Delta T) / (\Delta T_{max} - \Delta T_{min})\) . By this normalisation, T was transformed to a reciprocal score range, with high values being better than low values. Thirdly, to take both ELR and DRT into selecting de- novo PSM candidates, a three- factor model was defined as: \((1 - \alpha - \delta)A + \alpha (100 - R) + \delta T\) . + +## Performance Evaluation: Fractional Ranks (F-Ranks) + +Fractional Ranks (F- Ranks) Fractional rank (F- Rank) of a correct peptide sequence for each spectrum is defined as the rank by descending order of this peptide divided by the total number of candidates of that spectrum. For peptides with tie scores, average rank was used as the numerator. It measures peptide identification performance improvement/deterioration at individual spectrum level. The reduction of mean F- rank of all correct peptides by multi- factor model outputs over the baseline ALC model was to suggest the promotion of the correct peptides. Differences in average F rank scores were evaluated with a paired T test. + +## Performance evaluation: Full-Sequence Recall (FSR) + +FSR is defined as the ratio between the total number of correctly identified peptides and the total number of peptides in the given dataset. It measures the peptide identification sensitivity at the spectra ensemble level. For each factor model, the top- scoring PSM of each MS2 spectrum was used as an identified peptide sequence, whereas the database peptide sequence of the same MS2 spectrum was considered the correct peptide sequence. A peptide was considered correctly identified if the peptide predicted by the given model was identical (up to I/L indistinguishability) to the database peptide. Differences in FSR were evaluated with a paired T test. + +<--- Page Split ---> + +## Strategy for MARS candidate HLAp origin annotation and database curation + +We coarsely stratified the HLAP origins into three categories: "human protein", "single amino acid substitution", and "non- canonical". Besides, we put all HLAP not strongly associated with these three categories into another "unmatched" category. Seqkit (Shen et al. 2016) was employed for locating MARS peptides in the curated databases. For these searches, we excluded all database search against SwissProt assigned peptides with a score cut- off of - 10lgP>20, and searching the remaining de novo candidates as shortlisted by the MARS identification module. + +Firstly, we searched Uniprot and Ensembl (only human protein entries) with isoforms as the "human proteome". MARS peptides matching exactly to human proteome were considered "Human Proteome" originated HLAP, matching human protein with 1AA difference were considered "1AA substitution". These two categories of peptides were considered of canonical origin, and excluded from non- canonical category. To further examine the possible alternative origins of the peptides of non- canonical origin, we used GENCODE v40 six- frame translated reference genome and three- frame translated reference transcriptome to form the protein database. All annotations were combined into Genecode "Biotypes" as follows: "Protein coding": protein_coding; "IncRNA": IncRNA; "other RNAs": miRNA, misc_RNA, Mt_rRNA, Mt_tRNA, ribozyme, rRNA, rRNA_pseudogene, scaRNA, scRNA, snoRNA, snRNA, sRNA, vault_RNA; "pseudogene": all biotypes including "pseudogene" (polymorphic, processed, transcribed_processed, transcribed_unitary, transcribed_unprocessed, translated_processed, translated_unprocessed, unitary, unprocessed); "IG": IG_C_gene, IG_C_pseudogene, IG_D_gene, IG_J_gene, IG_J_pseudogene, IG_pseudogene, IG_V_gene, IG_V_pseudogene; "TR": TR_C_gene, TR_D_gene, TR_J_gene, TR_J_pseudogene, TR_V_gene, TR_V_pseudogene and "TEC": TEC. We integrated the following databases for IncRNA mapping: LNCipedia \(^{42}\) , IncRNAKB \(^{43}\) , and added a separate "TE" category using annotations from Dfam \(^{44}\) , and an Erv database \(^{45}\) . We report our annotations in Upset plots to include multiple assignments. + +<--- Page Split ---> + +## LC-MS acquisition + +All synthetic peptides were obtained from Genscript as a library service and crude purity. We measured the peptide libraries by LC- MS on an Ultimate 3000 RSLcnano System coupled with an Q Exactive™ HF Hybrid Quadrupole- Orbitrap™ Mass Spectrometer (Thermo Scientific). Peptides were loaded onto the analytical column (PepMap C18 column, \(2\mu \mathrm{m}\) particle size, \(75\mu \mathrm{m}\times 50\mathrm{cm}\) ; Thermo Scientific) and eluted in a 60 min linear gradient from \(3\%\) to \(25\%\) ACN in \(1\%\) DMSO/0.1% formic acid at a flow rate of \(250~\mathrm{nl / min}\) . Peptides were introduced to the mass spectrometer using an EasySpray source at 2000 V and \(45^{\circ}C\) , and the transfer tube temperature was set to \(305^{\circ}C\) . Mass spectrometry (MS) detection was performed with a resolution of 120,000 for full MS (320- 1600m/z scan range) and AGC target of 300,000. Precursors were isolated using an isolation width of 1.6 amu and fragmented using high- energy collisional dissociation (HCD) set at 25 for peptides with 2- 4 charges. Fragments resolution was set at 60,000 with AGC target of 50000. + +## LC-MS datasets and data analysis + +We utilized the following datasets: Data for the C1866 cell line (A\*01:01, B\*08:01, B\*44:02, C\*05:01, and C\*07:01) and for single allele transfectant cell lines is partially available in PXD015489 40. Data for the Jurkat cell line (A\*03:01, B\*07:02, B\*35:03, C\*04:01, C07:02) is available in PXD011723 46. Wilhelm et al. peptide standard datasets are available at PXD021013 17, and Chong et al. datasets are available at PXD013649 2. + +Peaks vX (Bioinformatics Solutions) was used to identify to search against a database containing the synthetic standard sequences. For database search, we used no enzyme specificity (setting: unspecific). The precursor mass error range was set within 5 ppm. Only oxidation was selected as variable modification, unless otherwise indicated in the main text. For the de novo sequencing, the precursor mass + +<--- Page Split ---> + +error range was set within 4.5 ppm, and only oxidation was selected as a variable modification unless otherwise indicated. + +## Spectral Matching + +Spectral comparisons were performed manually, with the help of the Universal Spectrum Explorer (USE)47. If a synthetic peptide was not detected by LC- MS measurement, we considered the peptide sequence identification as "false", as it is equally unlikely that the identical peptide molecule was successfully ionized and detected during the experiment. + +## Trinity De-Novo RNA Assembly + +Raw RNA sequencing reads had adaptors removed and low quality bases trimmed using Trim_Galore v0.6.2 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore) and quality control metrics assessed using FastQC v0.11.9 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc). A custom STAR (v2.7.3a)48 genome reference was created utilising UCSC hg38 genome sequence. Reads were then mapped to this custom genome reference using STAR with two passes, sorted and indexed by Samtools v1.10 49. Duplicates were marked using GATK v4.1.7.0 50. Unmapped reads from STAR were used as inputs for Trinity v2.11.0 51 for de novo transcript assembly. + +<--- Page Split ---> + +## ACKNOWLEDGEMENT + +We thank Prof. Masafumi Takiguchi, Kumamoto University, Japan for provision of CD4- expressing cell lines transfected with single HLA- I alleles. This work was funded by the NIH project #1R21AI153978- 01 (LCE), The Leona M. and Harry B. Helmsley Charitable Trust (RM), the European Association for the Study of Diabetes FSD/JDRF/Lilly Programme on Type 1 Diabetes Research 2019 (NT), Cancer Research UK Cancer Immunology project award C55884/A21045 (NT), and Cancer Research UK RadNet Centre Award C6078/A28736 (NT). Some immunopeptidomic datasets employed for this study were generated as part of studies supported by Medical Research Council (MRC) programme grant MR/K012037, grants from the National Institutes of Health (UM1 AI 100645 and R01 AI 118549) and by AMED grant 17FK0410302H0003 for AIDS research (PB). The computational aspects of this research were supported by the Wellcome Trust Core Award Grant Number 203141/Z/16/Z and the NIHR Oxford BRC. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. PB is a Jenner Institute Investigator. + +## DATA AND CODE AVAILABILITY + +Primary cervical tumour mass spectrometry data will be published in the Proteomics Identifications Database (https://www.ebi.ac.uk/pride/) upon acceptance of the manuscript. Code will be provided to academic researchers upon reasonable request. + +<--- Page Split ---> + +545 + +546 Table 1: IncRNA peptide validation + +547 Red highlighted: MARS identification not reported by Peaks DeNovo as first candidate + +
PEPTIDEMARS SCOREHLA ALLELEBINDING RANKGENE LOCATIONLNC-RNAALT NAMESAMPLESPECRAL MATCH
IIYFCLHKI93.4HLA-A*02:010.629chr5:164989217-164989191Inckb.44609CTC-340A15.24,5No
LPYIKWTI93.6HLA-B*51:010.076chr5:164805831-164805808Inckb.44609CTC-340A15.27Yes
LSSTLNKQI90.1HLA-C*02:021.143chr5:164743628-164743602Inckb.44609CTC-340A15.21No
IISASKVIL92.3HLA-A*02:010.617chr11:90809963-90809989Inckb.9710DISC1FP17,9No
TELKAWKI95.2HLA-B*37010.138chr11:90693569-90693592Inckb.9710DISC1FP13No
IRNVKIYLI91.2HLA-C*06:020.09Inckb.45851HLA-F-AS110No
KVHVFLVKK99.7HLA-A*03:010.029chr6:29750586-29750560Inckb.45851HLA-F-AS11,8Yes
RQAPALRHL92.5HLA-C*06:020.133chr13:94909331-94909357Inckb.14510LOC1019272849Yes
VIFSGIRSL95.4HLA-A*02:010.081chr13:94926124-94926150Inckb.14510LOC1019272845Yes
VLLSIFIEHL94.2HLA-A*02:011.15chr13:94845116-94845145Inckb.14510LOC1019272847No
ALRAVTLTAK97.9HLA-A*03:010.099chr3:35175054-35175080Inckb.36918LOC1019281358,10Yes
KQLNKQLIK98.3HLA-A*03010.053chr8:78259721-78259747Inckb.53745LOC1053759114,9Yes
VTIFQNRVK92HLA-A*11:010.575chr8:78413273-78413247Inckb.53745LOC1053759118Yes
KTKIIGTSK94.1HLA-A*03:010.192chrX:75675106-75675132Inckb.58303LOC1079856648Yes
SLSILSLKV92.9HLA-B*13:020.164chrX:75543586-75543560Inckb.58303LOC1079856641No
IPHGEIPDTSA94.9HLA-B*56:010.064chr2:172323392-172323424Inckb.31346LOC1079859605Yes
LPNFARII92.6HLA-B*07:020.29chr2:172280278-172280255Inckb.31346LOC1079859604,8No
MTMSTILSK99.2HLA-A*11:010.008chr5:8460009-8460035Inckb.42318MIR4458HG8Yes
MTMSTILSKK99.8HLA-A*03:010.199chr5:8460009-8460038Inckb.42318MIR4458HG4,8,9Yes
+ +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +549 Figure 1: Schematic of MARS workflow and performance evaluation in a five- fold cross- validation. (A) Schematic of integrative modules in the MARS pipeline (B) Mean F- rank score of all partitions, and (C) Full sequence recall (FSR) achieved across the partitions in the target peptide space. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Figure 2: MARS sensitivity evaluation in Wilhelm et al. large HLA peptide library datasets. (A) The average F- rank distribution across all five partitions and (B) Venn diagram depicting the number of peptide sequences identified by ALC (green) and MARS (red), respectively, within those sequences that were present in the 20 de novo candidates as reported by Peaks ("de novo identifiable", dark grey). Total sequence space is shown in light grey. (C) Box plots indicating the number of peptide sequences uniquely identified by MARS for all HLA- A, - B, and - C alleles combined (D) Full sequence recall (FSR) for individual HLA alleles for ALC (green) and MARS (red). + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +<--- Page Split ---> + +Figure 3: Non- canonical peptide identification using MARS in a published dataset from Chong et al. \(^{2}\) . (A) Left panel: Sequence recall for IncRNA peptide sequences identified by Chong at all across all samples plotted against the false discovery rate. The three regions are defined by FDR cut- offs at \(30\%\) and \(45\%\) , respectively. The right panel shows these regions in relation to the MARS score. (B) Venn diagrams demonstrating the overlap of sequence identifications for each of the three defined regions from (A). (C) Overview of MARS search engine and blast search strategy for identification of source origin. (D) Left panel: Bar graph showing the categorization of the MARS hits as indicated. Right panel: Mutually identified peptides by MARS and Chong et al. from the 508 reported non- canonical identifications assigned to the MARS categories as indicated. Colored subsections of the graph indicate the original alternative classification as applied by Chong et al. in the original publication (E) Score distribution of all MARS identified non- canonical peptides (11339) in comparison to those mutually identified by Chong et al. (193). (F) Experimental validation of the FDR using a synthetic peptide standard. Histogram summarizing the peptide score distribution of the synthetic peptide standard. Shown are number of peptides identified by MARS only for indicated MARS score bins (upper panel), and FDP progression in relation to declining MARS score (lower panel) calculated on the spectral matching outcome. (G) Number and proportion of all confirmed peptides for all synthetic standard sequences and for the subset of peptides that were mapped to the human genome (“excluding unmatched”). (H) Upset plot of the MARS non- canonical peptide top 10 categories for all hits in region 1, 2, and 3 (lower panel), and mutual identifications as reported in Chong et al. (upper graph). The Venn diagram indicates the overall number of non- canonical sequences that were identified with MARS at a FDR \(\leq 14.3\%\) . + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Figure 4: Identification of novel non- canonical HLAP in primary tumour tissue using MARS. (A) Bar graph showing the categorization of the MARS hits as indicated. (B) Upset plot of the MARS non- canonical peptide category for all hits (C) Bar graph depicting the level of evidence found in trinity assemblies of RNAseq data for each IncRNA category. (D) Score distribution of all MARS peptide hits in comparison of those that are validated with additional RNA sequence evidence. (E) Spectral matching validation for indicated sequences identified in the primary tumour tissue. + +<--- Page Split ---> + +589 1. Jaeger, A.M. et al. Deciphering the immunopeptidome in vivo reveals new tumour antigens. Nature (2022). 591 2. Chong, C. et al. Integrated proteogenomic deep sequencing and analytics accurately identify non-canonical peptides in tumor immunopeptidomes. Nat Commun 11, 1293 (2020). 594 3. Muth, T. & Renard, B.Y. Evaluating de novo sequencing in proteomics: already an accurate alternative to database-driven peptide identification? Brief Bioinform 19, 954- 970 (2018). 597 4. Bassani-Sternberg, M. et al. Direct identification of clinically relevant neoepitopes presented on native human melanoma tissue by mass spectrometry. Nat Commun 7, 13404 (2016). 600 5. Ternette, N. et al. Defining the HLA class I- associated viral antigen repertoire from HIV- 1- infected human cells. Eur J Immunol 46, 60- 69 (2016). 602 6. Bettencourt, P. et al. Identification of antigens presented by MHC for vaccines against tuberculosis. NPJ Vaccines 5, 2 (2020). 604 7. Attig, J. et al. LTR retroelement expansion of the human cancer transcriptome and immunopeptidome revealed by de novo transcript assembly. Genome Res 29, 1578- 1590 (2019). 607 8. Kracht, M.J. et al. Autoimmunity against a defective ribosomal insulin gene product in type 1 diabetes. Nat Med 23, 501- 507 (2017). 609 9. Zanker, D.J. et al. Influenza A Virus Infection Induces Viral and Cellular Defective Ribosomal Products Encoded by Alternative Reading Frames. J Immunol 202, 3370- 3380 (2019). 612 10. Parker, R. et al. The Choice of Search Engine Affects Sequencing Depth and HLA Class I Allele- Specific Peptide Repertoires. Mol Cell Proteomics 20, 100124 (2021). 614 11. Bichmann, L. et al. MHCquant: Automated and Reproducible Data Analysis for Immunopeptidomics. J Proteome Res 18, 3876- 3884 (2019). 616 12. Mann, M. & Wilm, M. Error- tolerant identification of peptides in sequence databases by peptide sequence tags. Anal Chem 66, 4390- 4399 (1994). 618 13. Yates, J.R., 3rd, Eng, J.K., McCormack, A.L. & Schieltz, D. Method to correlate tandem mass spectra of modified peptides to amino acid sequences in the protein database. Anal Chem 67, 1426- 1436 (1995). 621 14. Nesvizhskii, A.I. Protein identification by tandem mass spectrometry and sequence database searching. Methods Mol Biol 367, 87- 119 (2007). 623 15. AS, C.S., Bouwmeester, R., Martens, L. & Degroeve, S. Accurate peptide fragmentation predictions allow data driven approaches to replace and improve upon proteomics search engine scoring functions. Bioinformatics 35, 5243- 5248 (2019). 626 16. Gessulat, S. et al. Prosit: proteome- wide prediction of peptide tandem mass spectra by deep learning. Nat Methods 16, 509- 518 (2019). 627 17. Wilhelm, M. et al. Deep learning boosts sensitivity of mass spectrometry- based immunopeptidomics. Nat Commun 12, 3346 (2021). 630 18. Kong, A.T., Leprevost, F.V., Avtonomov, D.M., Mellacheruvu, D. & Nesvizhskii, A.I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry- based proteomics. Nat Methods 14, 513- 520 (2017). + +<--- Page Split ---> + +633 19. Li, K., Jain, A., Malovannaya, A., Wen, B. & Zhang, B. DeepRescore: Leveraging Deep Learning to Improve Peptide Identification in Immunopeptidomics. \*Proteomics\* 20, e1900334 (2020).636 20. Tran, N.H., Zhang, X., Xin, L., Shan, B. & Li, M. De novo peptide sequencing by deep learning. \*Proc Natl Acad Sci U S A\* 114, 8247- 8252 (2017).637 21. Yang, H., Chi, H., Zeng, W.F., Zhou, W.J. & He, S.M. pNovo 3: precise de novo peptide sequencing using a learning- to- rank framework. \*Bioinformatics\* 35, i183- i190 (2019).640 22. Ma, B. et al. PEAKS: powerful software for peptide de novo sequencing by tandem mass spectrometry. \*Rapid Commun Mass Spectrom\* 17, 2337- 2342 (2003).641 23. Erhard, F., Dolken, L., Schilling, B. & Schlosser, A. Identification of the Cryptic HLA- I Immunopeptidome. \*Cancer Immunol Res\* 8, 1018- 1026 (2020).642 24. Laumont, C.M. et al. Global proteogenomic analysis of human MHC class I- associated peptides derived from non- canonical reading frames. \*Nat Commun\* 7, 10238 (2016).643 25. Ouspenskaja, T. et al. Unannotated proteins expand the MHC- I- restricted immunopeptidome in cancer. \*Nat Biotechnol\* 40, 209- 217 (2022).644 26. Andreatta, M. et al. MS- rescue: A Computational Pipeline to Increase the Quality and Yield of Immunopeptidomics Experiments. \*Proteomics\*, e1800357 (2018).645 27. Ma, C. et al. Improved Peptide Retention Time Prediction in Liquid Chromatography through Deep Learning. \*Anal Chem\* 90, 10881- 10888 (2018).646 28. Tran, N.H. et al. Personalized deep learning of individual immunopeptidomes to identify neoantigens for cancer vaccines. \*Nat Mach Intell\* 2 (2020).647 29. Zhang, J. et al. PEAKS DB: de novo sequencing assisted database search for sensitive and accurate peptide identification. \*Mol Cell Proteomics\* 11, M111 010587 (2012).648 30. Tran, N.H. et al. Deep learning enables de novo peptide sequencing from data- independent- acquisition mass spectrometry. \*Nat Methods\* 16, 63- 66 (2019).649 31. Gonzalez- Galarza, F.F. et al. Allele frequency net database (AFND) 2020 update: gold- standard data classification, open access genotype data and new query tools. \*Nucleic Acids Res\* 48, D783- D788 (2020).650 32. Martini, S., Nielsen, M., Peters, B. & Sette, A. The Immune Epitope Database and Analysis Resource Program 2003- 2018: reflections and outlook. \*Immunogenetics\* 72, 57- 76 (2020).651 33. Vita, R. et al. The Immune Epitope Database (IEDB): 2018 update. \*Nucleic Acids Res\* 47, D339- D343 (2019).652 34. Xu, C. & Ma, B. Complexity and scoring function of MS/MS peptide de novo sequencing. \*Comput Syst Bioinformatics Conf\*, 361- 369 (2006).653 35. Meadows, L. et al. The HLA- A\*0201- restricted H- Y antigen contains a posttranslationally modified cysteine that significantly affects T cell recognition. \*Immunity\* 6, 273- 281 (1997).654 36. Trujillo, J.A. et al. The cellular redox environment alters antigen presentation. \*J Biol Chem\* 289, 27979- 27991 (2014).655 37. Huang, Y. et al. HLA- F- AS1/miR- 330- 3p/ PFN1 axis promotes colorectal cancer progression. \*Life Sci\* 254, 117180 (2020).656 38. Wu, D., Jia, H., Zhang, Z. & Li, S. STAT3- induced HLA- F- AS1 promotes cell proliferation and stemness characteristics in triple negative breast cancer cells by upregulating TRABD. \*Bioorg Chem\* 109, 104722 (2021).657 39. Akrami, R. et al. Comprehensive analysis of long non- coding RNAs in ovarian cancer reveals global patterns and targeted DNA amplification. \*PLoS One\* 8, e80306 (2013).658 40. Partridge, T. et al. Discrimination Between Human Leukocyte Antigen Class I- Bound and Co- Purified HIV- Derived Peptides in Immunopeptidomics Workflows. \*Front Immunol\* 9, 912 (2018).659 41. Paes, W. et al. Contribution of proteasome- catalyzed peptide cis- splicing to viral targeting by CD8(+) T cells in HIV- 1 infection. \*Proc Natl Acad Sci U S A\* 116, 24748- 24759 (2019). + +<--- Page Split ---> + +683 42. Volders, P.J. et al. LNCipedia 5: towards a reference set of human long non- coding RNAs. 684 Nucleic Acids Res 47, D135- D139 (2019). 685 43. Seifuddin, F. et al. lncRNAKB, a knowledgebase of tissue- specific functional annotation 686 and trait association of long noncoding RNA. Sci Data 7, 326 (2020). 687 44. Storer, J., Hubley, R., Rosen, J., Wheeler, T.J. & Smit, A.F. The Dfam community resource 688 of transposable element families, sequence models, and genome annotations. Mob DNA 689 12, 2 (2021). 690 45. Saini, S.K. et al. Human endogenous retroviruses form a reservoir of T cell targets in 691 hematological cancers. Nat Commun 11, 5660 (2020). 692 46. Nicastri, A., Liao, H., Muller, J., Purcell, A.W. & Ternette, N. The Choice of HLA- Associated 693 Peptide Enrichment and Purification Strategy Affects Peptide Yields and Creates a Bias 694 in Detected Sequence Repertoire. Proteomics 20, e1900401 (2020). 695 47. Schmidt, T. et al. Universal Spectrum Explorer: A Standalone (Web-)Application for Cross- 696 Resource Spectrum Comparison. J Proteome Res 20, 3388- 3394 (2021). 697 48. Dobin, A. et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 29, 15- 21 698 (2013). 699 49. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078- 700 2079 (2009). 701 50. McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing 702 next- generation DNA sequencing data. Genome Res 20, 1297- 1303 (2010). 703 51. Grabherr, M.G. et al. Full- length transcriptome assembly from RNA- Seq data without a 704 reference genome. Nature Biotechnology 29, 644- U130 (2011). + +<--- Page Split ---> + +SUPPLEMENTARY INFORMATION + +Supplementary Figure 1: I replacement by L maintains validity of NetMHCpan 4.1 prediction across distinct HLA alleles + +Supplementary Figure 2: HLA allele distributions for Chong et al. dataset, and cervical tumour patient cohort. + +Supplementary Figure 3 (provided separately): Spectral matching for FDR determination + +Supplementary Data Table 1: Five- Fold Cross- Validations using Baseline ALC model, Two- Factor alpha model, Two- Factor beta model, and Three- Factor MARS model. Column 2- 5 shows the optimal parameters learned by grid search for each part, suggesting the models are robust. Column 6- 8 gives the results of Paired Wilcoxon signed- rank tests on F- ranks of true peptide sequences; significant drop of F- ranks indicates improved identification power. + +Supplementary Data Table 2: Number of peptide sequences reported to bind to each respective HLA allele in the Wilhelm et al. dataset + +Supplementary Data Table 3: FSR rates for Chong et al. data for each sample + +Supplementary Data Table 4: Overview of MARS peptide identifications for each sample in the Chong et al. dataset + +Supplementary Data Table 5: HLA- typing results for cervical tumour patients + +Supplementary Data Table 6: FDR determination: Peptide Validation Summary + +<--- Page Split ---> +![](images/Supplementary_Figure_1.jpg) + +
Supplementary Figure 1
+ +Supplementary Figure 1: I replacement by L maintains validity of NetMHCan 4.1 prediction across distinct HLA alleles. (A) K- L divergences were computed as statistics to characterize the similarity of two distributions D(Original || Only I) and D(Original || Only L). KLDiv_Boxplot shows that Only L had consistently very small K- L divergences, and a paired Wilcoxon signed rank test suggested that D(Original || Only I) is significantly greater than D(Original || Only L) with \(\mathsf{p} = 0.0156\) . (B) Rank score distributions for example alleles for original peptide sequence, all I replaced by L (Only L), and vice versa (Only I). + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Chong et al. Allele Counts
+ +740 + +741 + +742 + +743 + +744 + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryTable1. csv SupplementaryTable2. csv SupplementaryTable3. csv SupplementaryTable4. csv SupplementaryTable5. csv SupplementaryTable6. csv + +<--- Page Split ---> diff --git a/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386_det.mmd b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..2f04b798c86ea79449a9f0f9b976311ddbd8d7d1 --- /dev/null +++ b/preprint/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386/preprint__132c5c25872672e27d75bc3b05aab904da731729c0591a5589f0e86f9e21a386_det.mmd @@ -0,0 +1,567 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 828, 210]]<|/det|> +# MARS: Improved De Novo Peptide Candidate Selection for Non-Canonical Antigen Target Discovery in Cancer + +<|ref|>text<|/ref|><|det|>[[44, 230, 234, 271]]<|/det|> +Hanqing Liao University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 277, 696, 319]]<|/det|> +Carolina Barra Technical University of Denmark https://orcid.org/0000- 0002- 6836- 4906 + +<|ref|>text<|/ref|><|det|>[[44, 324, 231, 365]]<|/det|> +Xu Peng University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 371, 588, 411]]<|/det|> +Isaac Woodhouse University of Oxford https://orcid.org/0000- 0002- 8221- 7697 + +<|ref|>text<|/ref|><|det|>[[44, 417, 231, 457]]<|/det|> +Arun Tailor University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 463, 231, 503]]<|/det|> +Robert Parker University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 509, 842, 550]]<|/det|> +Alexia Carre Institute Cochin, INSERM U1016, CNRS UMR8104 https://orcid.org/0000- 0001- 8144- 7194 + +<|ref|>text<|/ref|><|det|>[[44, 555, 840, 597]]<|/det|> +Roberto Mallone Institute Cochin, INSERM U1016, CNRS UMR8104 https://orcid.org/0000- 0002- 9846- 8861 + +<|ref|>text<|/ref|><|det|>[[44, 602, 712, 642]]<|/det|> +Michael Hogan Children's Hospital of Philadelphia https://orcid.org/0000- 0002- 6244- 2993 + +<|ref|>text<|/ref|><|det|>[[44, 648, 588, 689]]<|/det|> +Wayne Paes University of Oxford https://orcid.org/0000- 0002- 0529- 2765 + +<|ref|>text<|/ref|><|det|>[[44, 694, 645, 735]]<|/det|> +Laurence Eisenlohr University of Pennsylvania https://orcid.org/0000- 0002- 8475- 7910 + +<|ref|>text<|/ref|><|det|>[[44, 740, 694, 781]]<|/det|> +Morten Nielsen Technical University of Denmark https://orcid.org/0000- 0001- 7885- 4311 + +<|ref|>text<|/ref|><|det|>[[44, 786, 588, 827]]<|/det|> +Nicola Temette ( \(\square\) nicola.termette@ndm.ox.ac.uk) University of Oxford https://orcid.org/0000- 0002- 9283- 0743 + +<|ref|>text<|/ref|><|det|>[[44, 870, 101, 887]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 907, 135, 924]]<|/det|> +Keywords: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 46, 309, 64]]<|/det|> +Posted Date: August 3rd, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 84, 474, 102]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1890352/v1 + +<|ref|>text<|/ref|><|det|>[[44, 120, 910, 163]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[75, 105, 884, 130]]<|/det|> +1 MARS: Improved De Novo Peptide Candidate Selection for Non- + +<|ref|>text<|/ref|><|det|>[[75, 155, 598, 180]]<|/det|> +2 Canonical Antigen Target Discovery in Cancer + +<|ref|>text<|/ref|><|det|>[[75, 201, 884, 222]]<|/det|> +3 Hanqing Liao \(^{1,2}\) , Carolina Barra \(^{3}\) , Xu Peng \(^{1}\) , Isaac Woodhouse \(^{1,2}\) , Arun Tailor \(^{1,2}\) , Robert Parker \(^{1,2}\) , Alexia + +<|ref|>text<|/ref|><|det|>[[75, 237, 884, 257]]<|/det|> +4 Carré \(^{4}\) , Roberto Mallone \(^{4}\) , Persephone Borrow \(^{2}\) , Michael J Hogan \(^{5}\) , Wayne Paes \(^{1,2}\) , Laurence C. Eisenlohr \(^{5,6}\) , + +<|ref|>text<|/ref|><|det|>[[75, 272, 398, 290]]<|/det|> +5 Morten Nielsen \(^{3*}\) , Nicola Ternette \(^{1,2*}\) # + +<|ref|>text<|/ref|><|det|>[[75, 305, 88, 319]]<|/det|> +6 + +<|ref|>text<|/ref|><|det|>[[75, 333, 454, 350]]<|/det|> +7 1 The Jenner Institute, University of Oxford, OX37BN, UK + +<|ref|>text<|/ref|><|det|>[[75, 360, 717, 377]]<|/det|> +8 2 Centre for Immuno-Oncology, Nuffield Department of Medicine, University of Oxford, OX37DQ, UK + +<|ref|>text<|/ref|><|det|>[[75, 388, 448, 404]]<|/det|> +9 3 Technical University Denmark, Copenhagen, Denmark + +<|ref|>text<|/ref|><|det|>[[75, 415, 576, 431]]<|/det|> +10 4 Université de Paris Cité, Institut Cochin, CNRS, INSERM, 75014 Paris, France + +<|ref|>text<|/ref|><|det|>[[75, 442, 790, 459]]<|/det|> +11 5 Department of Pathology and Laboratory Medicine, Children's Hospital of Philadelphia, Philadelphia, PA, 19104. + +<|ref|>text<|/ref|><|det|>[[75, 470, 884, 486]]<|/det|> +12 6 Department of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, + +<|ref|>text<|/ref|><|det|>[[75, 498, 158, 513]]<|/det|> +13 19104. + +<|ref|>text<|/ref|><|det|>[[75, 526, 235, 542]]<|/det|> +14 \*equal contribution + +<|ref|>text<|/ref|><|det|>[[75, 554, 585, 570]]<|/det|> +15 \# to whom correspondence may be addressed: nicola.ternette@ndm.ox.ac.uk + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 91, 196, 106]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[110, 121, 886, 222]]<|/det|> +Cryptic Human Leukocyte Antigen (HLA)- presented peptide identification from unannotated genome sources is a priority for target antigen discovery for development of next generation immunotherapies in cancer. Current immunopeptidomic approaches utilize the integration of transcriptomics data to inform spectral interpretation, however, recent observations that tumour- associated antigen- encoding RNA levels are often low highlights limitations of such proteogenomic approaches1. + +<|ref|>text<|/ref|><|det|>[[110, 232, 886, 277]]<|/det|> +levels are often low highlights limitations of such proteogenomic approaches1. + +<|ref|>text<|/ref|><|det|>[[110, 291, 886, 477]]<|/det|> +We here employ a de novo sequencing approach with a refined, MHC- centric analysis strategy to detect non- canonical HLA- associated peptide sequences (HLAP) in cancer without integration of transcript sequence information. Our strategy integrates HLA binding prediction, peptide retention time prediction, and average local confidence scores culminating in the machine learning model MARS (MHC binding prediction, Average Local Confidence Score, and Retention time integration for improved de novo candidate Selection). + +<|ref|>text<|/ref|><|det|>[[110, 494, 886, 730]]<|/det|> +We demonstrate increased HLA- I peptide identification sensitivity by benchmarking our model against de novo sequencing alone with a large synthetic HLA- I peptide library dataset. We further define the sensitivity of MARS by reanalysis of a published dataset of high- quality non- canonical HLAP identifications in human cancer cell line and tissue datasets and achieve almost 2- fold improvement of the full sequence recall (FSR) for high quality spectral assignments in comparison to de novo sequencing alone2. We minimize the false discovery rate (FDR) through a step- wise peptide sequence mapping strategy and are able to expand the reported non- canonical peptide space with an assignment accuracy above 85.7%. + +<|ref|>text<|/ref|><|det|>[[110, 735, 885, 820]]<|/det|> +Finally, we utilize MARS to detect and validate IncRNA- derived peptides in human cervical tumour resections, demonstrating its suitability to discover novel, non- canonical peptide sequences in primary tumour tissue at reduced FDR, in the absence of transcriptomic sequencing data. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 236, 107]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[110, 120, 886, 415]]<|/det|> +The analysis of HLA- bound peptide ligands using liquid chromatography mass spectrometry (LC- MS) technology (immunopeptidomics) is a rapidly evolving field that has advanced identification of T cell antigens including clinically relevant epitopes in autoimmunity, pathogen infection and cancer \(^{4 - 6}\) . These approaches have recently been expanded to identification of peptides from unannotated open reading frames (ORFs), or out- of- frame translated sequences, that are valuable for development of immunotherapeutic approaches to cancer, using the integration of RNA sequence information \(^{2, 4, 7 - 9}\) . However, recent observations that many HLA- presented peptides exhibit minimal RNA expression highlight the need for transcriptomics data- independent interrogation of the immunopeptidome for cancer antigen discovery. + +<|ref|>text<|/ref|><|det|>[[110, 425, 886, 888]]<|/det|> +Software tools to interrogate LC- MS spectral data have widely been adopted from the field of proteomics, and extensive differences in performance have been observed for HLA- derived peptide data \(^{2, 10, 11}\) . Spectra interpretation software is based on three main strategies. The most common approach, the "database search" approach \(^{12 - 14}\) , relies on matching the masses of experimentally obtained peptide fragment spectra to all theoretically possible peptide fragments of sequences in a protein sequence list or "database". This approach is limited to identifications in the sequence database, and can therefore not be used for discovery of non- canonical peptides unless such information is added to known canonical protein databases. For increased sensitivity, this approach has been extended by integration of spectral matching assignment, which uses advanced deep learning and peptide fragmentation prediction to annotate the experimentally obtained peptide fragment spectra with high accuracy \(^{15 - 19}\) . Thirdly, de novo algorithms interpret the MS spectra without comparison to a protein sequence database, annotating the amino acid sequence directly from the spectral information available \(^{20 - 22}\) . Integration of de novo approaches and database searching (de novo- assisted database approach) has been shown to be particularly advantageous for HLA- peptide datasets \(^{10, 11, 23}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 310]]<|/det|> +For the interrogation of immunopeptidome datasets with a focus on cancer- specific peptide ligands, proteogenomics approaches have been used to refine the database- search approach in order to identify peptide sequences that contain tumour- specific mutations or other non- canonical sources like functional RNA genes, intronic sequences, and/or reading frames in unannotated genomic origins that are generally not present in universal protein databases \(^{2, 7, 23 - 25}\) . Here, genomics and/or transcriptomics data is integrated to achieve a highly accurate, sample- specific database that can be used as template for a "database search" analysis approach. + +<|ref|>text<|/ref|><|det|>[[111, 325, 884, 448]]<|/det|> +The incorporation of "personalised" binding prediction (MSRescue \(^{26}\) ), or difference between observed and predicted retention time (DeepRTplus \(^{27}\) , Deep Rescore \(^{19}\) ) to database- search based peptide identification workflows can also improve performance. Recently, machine learning has been applied in a personalized fashion in order to identify patient neoantigens \(^{28}\) . + +<|ref|>text<|/ref|><|det|>[[110, 461, 884, 856]]<|/det|> +The purpose of this study is to increase the power of the de novo approach for identification of non- canonical HLA- I- associated peptides (HLA- Ip) without the use of a personalized template protein database. We report that by integration of (i) HLA binding affinity predictions and (ii) differences between observed and predicted retention time (RT) of candidate peptide sequences as additional factors, HLA- I peptide identification sensitivity can be improved over the baseline de novo sequencing annotation. We validate our findings by benchmarking our algorithm termed MARS on LC- MS data from synthetic HLA- I peptide libraries, and published non- canonical peptide sequences that were obtained from proteogenomic analyses. We then proceed and develop a hierarchical strategy to map MARS sequence candidates to their likely origin in the genome, and conclude in a final false discovery rate below 14.3% - offering opportunity for improved shortlisting of accurate sequence identifications from the generally observed global 65% FDR in de novo sequence data \(^{3}\) . Through this approach, we are able to expand the non- canonical peptidome of published data by hundreds of peptides. We finally detect and validate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[63, 88, 883, 143]]<|/det|> +IncRNA derived peptides in primary human tumour tissue, demonstrating applicability for antigen discovery in cancer. + +<|ref|>sub_title<|/ref|><|det|>[[63, 192, 183, 208]]<|/det|> +## RESULTS + +<|ref|>sub_title<|/ref|><|det|>[[113, 225, 408, 243]]<|/det|> +## Training Data Selection and Processing + +<|ref|>text<|/ref|><|det|>[[110, 255, 886, 447]]<|/det|> +Our aim was to construct and validate our computational model for improved de novo (template- independent) identification of HLA- I- associated peptides. We selected a panel of in- house single allele (CD4.221 cells expressing either A\*01:01, A\*02:01, A\*03:01, A\*11:01, B\*08:01, B\*44:02, B\*57:03, C\*03:03, or C\*03:04) and multi allele (Jurkat cell line: A\*03:01, B\*07:02, B35:03, C\*04:01, C\*07:02; and the C1866 cell line: A\*24:02. B\*40:02, B\*51:01, C\*03:03, C\*14:02) data sets to develop models and to evaluate model performances. + +<|ref|>text<|/ref|><|det|>[[110, 460, 886, 786]]<|/det|> +We analyzed these datasets using Peaks DeNovo and also the Peaks DeNovo- assisted Database search (DB) using the SwissProt human canonical proteome 29. From the Peaks DeNovo search, we obtained up to 100 sequence candidates per spectrum (independent from the provided sequence database), with associated average local confidence (ALC) scores for each candidate sequence as a measure for the confidence of each sequence assignment. The Peaks DB search results were returned as a list of sequence assignments that matched to the SwissProt database. This latter search results in one sequence assignment per spectrum (with exception of L/I isomers) with an associated score (- 10/gP), which reflects the probability of a correct peptide spectrum match. The results are controlled by a simultaneous decoy database search which provides an accurate estimation of the false discovery rate to be expected in the result 30. + +<|ref|>text<|/ref|><|det|>[[110, 800, 885, 888]]<|/det|> +In order to define a subset of "true sequences" to use as a training dataset, we defined spectra as confidently assigned if they were identified with Peaks DB with high confidence - 10/logP ≥ 20 resulting in an FDR of less than 1% on peptide- spectrum match (PSM) level. In addition, we only included peptides + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 150]]<|/det|> +with a length of 8- 13 amino acids, and those that were assigned a low NetMHCpan 4.1 Eluted Ligand Rank (ELR) score \((ELR< 10)\) , aiming to remove peptides that are likely originating from co- purification during biochemical enrichment. All spectra that fulfilled these criteria were defined as the training dataset. + +<|ref|>text<|/ref|><|det|>[[111, 163, 886, 450]]<|/det|> +Next, within the Peaks DeNovo search results, the "true" peptide was considered the Peaks DB assigned sequence identification with all constituent isoleucine (I) replaced by leucine (L), because Peaks DeNovo algorithm does not distinguish I from L. Only those spectra with the "true" (DB) peptides present in the list of the de novo peptide candidates were selected for model development and evaluation. In order to confirm that the I to L replacement does not negatively affect the NetMHCpan binding prediction, we tested NetMHCpan predictions within the test set for both L and I variants, and found that replacement of all I with L incurred significantly less divergence from the original result than vice versa (Supplementary Figure 1). + +<|ref|>text<|/ref|><|det|>[[111, 461, 886, 650]]<|/det|> +All selected spectra were then split into 5 equal parts to form a five- fold cross- validation scheme, such that no two parts shared any common 8- mer substring training peptide sequence. The model trainings and evaluations were carried out using a five- fold cross- validation scheme, where each time a different combination of four parts was used as model- training set, while the performances were evaluated on the fifth (test) partition. Finally, all five- fold test set predictions were concatenated to obtain the final model performance. + +<|ref|>sub_title<|/ref|><|det|>[[111, 700, 883, 750]]<|/det|> +## Two factor model evaluation: Incorporation of NetMHCpan prediction improves de novo candidate selection + +<|ref|>text<|/ref|><|det|>[[111, 766, 886, 888]]<|/det|> +We used the Peaks DeNovo result output as the baseline model. Accordingly, for each spectrum, the peptide candidate of the highest ALC was taken as the "identified" peptide further referred to as "ALC" result. In order to understand whether we could re- rank the de novo candidates by selecting the best candidate based on their binding prediction to the respective MHC allele of origin, we determined the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 520]]<|/det|> +NetMHCpan binding percentile Rank score for each of the 100 sequence candidates in the PEAKS DeNovo list, and calculated for each peptide candidate an affine combination of the ALC and NetMHCpan predicted Rank (ELR) scores to produce a new composite score. Then, for each spectrum, the peptide candidate of the highest composite score was taken as the identified peptide. Because this affine combination was controlled by a single weight parameter, alpha, and modelled ALC and ELR as two peptide identity determining factors, we named it "two- factor alpha model". We calculated the F- ranks which describe the rank of the true peptide sequence in descending order, divided by the total number of candidates, for each spectrum as an identification performance measure. We applied a grid search to obtain the optimal alpha value for a minimal mean F- rank within the training set. Then this alpha value was applied to the test for performance evaluation. All five test sets yielded one- sided paired Wilcoxon signed- rank test p- values <2.2E10- 16, indicating that the reduction of F- ranks by applying the two- factor alpha model in comparison to the ALC baseline model was highly significant (Supplementary Table 1, Figure 1A, B). + +<|ref|>text<|/ref|><|det|>[[110, 529, 886, 720]]<|/det|> +To examine whether the improved F- rank led to higher identification sensitivity when top scoring peptide candidate per spectrum was taken as its identity, we employed another measurement, the full sequence recall (FSR), which depicts the proportion of identified (top one scoring) peptides being correct. A paired T- test using the mean FSRs from each of the five partitions also suggested that two- factor alpha model (M=0.782, SD=0.00461) significantly improved FSR over the baseline ALC model (M=0.731, SD=0.00554); t(4)=36.74, p=3.28E10- 6 (Figure 1C). + +<|ref|>text<|/ref|><|det|>[[110, 732, 886, 853]]<|/det|> +Lastly, the optimal alpha obtained in the five- fold cross- validations were highly concordant, suggesting the two- factor alpha model was robust (Supplementary Table 1). In summary, these results suggested that incorporating ELR as a factor can improve PSM candidate selection, resulting in improved HLAP peptide identification compared to ALC alone. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[110, 89, 867, 108]]<|/det|> +# Incorporation of RT prediction further improves de novo candidate selection: Final model evaluation + +<|ref|>text<|/ref|><|det|>[[110, 120, 885, 345]]<|/det|> +To further improve PSM candidate selection, we incorporated DeepRTplus \(^{27}\) to predict the RT of input peptides. We trained the DeepRTplus model with our training data, and applied the learned model to the training data itself to derive an absolute value of differences between predicted and observed retention time (DRT). We then defined an affine combination of the ALC, ELR and DRT scores to define a three- factor composite model and used the grid search to find optimal alpha (weight on ELR) and beta (weight on DRT) values to minimize the mean F- rank in the training data. Finally, the optimal values of alpha and beta were used to assign combinatory scores to each peptide in the test data. + +<|ref|>text<|/ref|><|det|>[[110, 358, 885, 712]]<|/det|> +We then proceeded with our evaluation strategy as before: a paired Wilcoxon test revealed that the F- ranks of the three- factor model were significantly lower than those of the two- factor alpha model (maximal of five p- values \(< 1.2 \times 10E- 10\) (Supplementary Table 1, Figure 1B)), suggesting that the three- factor model outperformed the two- factor model in promoting the true peptide identity amongst all other candidates in each spectrum. From the perspective of FSR, the performance gain was consistent with the F- Rank: the FSR by the three- factor model (M=0.786, SD=0.00601) was higher than FSR of two- factor alpha model (M=0.781, SD=0.00462) t(4)=3.14, p=0.0349 (Figure 1C). Further, the model “training” was found to be robust in that the optimal alpha and beta were stable across test sets (Supplementary Table 1). + +<|ref|>text<|/ref|><|det|>[[112, 723, 880, 750]]<|/det|> +novo sequencing candidates. + +<|ref|>sub_title<|/ref|><|det|>[[112, 802, 697, 820]]<|/det|> +## MARS improves HLAP identification sensitivity across HLA-A and HLA-B alleles + +<|ref|>text<|/ref|><|det|>[[112, 834, 883, 888]]<|/det|> +To substantiate our findings, we applied MARS to synthetic HLAP standard data for sensitivity evaluation. We hypothesized that the improvement of the re- ranking of de novo HLAP identification by the MARS + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 210]]<|/det|> +model is robust in the sense that it can be applied to different datasets with fixed factor coefficients. The only requirement is a transfer learning scheme needed to calibrate the DeepRTplus model in order to accommodate the discrepancy introduced by the LC system setups and gradient conditions between different datasets. + +<|ref|>text<|/ref|><|det|>[[111, 223, 885, 584]]<|/det|> +We utilized a large dataset published by Wilhelm, Zolg et al., of \(\sim 169,000\) synthetic HLA class I peptides acquired by liquid chromatography mass spectrometry (LC- MS) \(^{17}\) . Out of this dataset, we selected the 116 most frequent HLA alleles \(^{31}\) , comprising 35 HLA- A, 59 HLA- B, and 22 HLA- C alleles (Supplementary Table 2). We here included spectra from the 3xHCD method only, which was reported the best performing in the original study. Each peptide's HLA binding information was extracted from IEDB \(^{32,33}\) . We further applied a filter to exclude peptide- HLA associations with NetMHCapn ELR>10, which we deemed likely inaccurate assignments. This resulted in a collection of 28,919 MS2 spectra from 24,193 unique peptide sequences, of which the majority of sequences (26,452 spectra/22,324 peptide sequences; 91.5%) were listed in the Peaks DeNovo result candidate list (hence forward referred to as "recoverable") (Supplementary Table 2). We selected sequence- RT pairs from peptide candidates with ALC > 95 to calibrate the DeepRTplus model, while keeping the alpha and beta values constant. + +<|ref|>text<|/ref|><|det|>[[111, 596, 885, 888]]<|/det|> +MARS outperformed ALC and the average F- ranks were significantly lower for MARS than ALC by paired Wilcoxon test ( \(p < 2.2 \times 10E- 16\) , Figure 2A). MARS FSR ( \(M = 0.907\) , \(SD = 0.0344\) ) was also significantly higher than ALC FSR ( \(M = 0.859\) , \(SD = 0.0744\) ): \(t(115) = 8.4295\) , \(p = 1.147e- 13\) (Figure 2B). MARS could again increase the number of identified peptide sequences over ALC (Figure 2C), and overall improvement was 7.23%. By grouping HLA alleles based on gene locus (A, B, and C), it can be seen that for HLA- A and B loci, the peptide FSRs can be significantly improved by MARS, but for HLA- C, the difference between ALC FSR and MARS FSR did not reach significance, possibly due to a lower number of spectra included for HLA- C (Figure 2D, Supplementary Table 2). When we further looked at the improvement of sequence identifications for all individual alleles, we noted that for HLA- A, only HLA- A\*01:01 and HLA- A\*31:01 did not benefit from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 210]]<|/det|> +MARS over ALC, while MARS improved results for all other HLA- A assigned peptide sequences. A high variability in performance across different HLA- B alleles was observed, with striking improvement in peptide identifications using MARS for HLA- B\*27:05 and other individual HLA- B alleles, while HLA- B35 did not benefit from MARS rescoring (Figure 2D). + +<|ref|>sub_title<|/ref|><|det|>[[111, 258, 882, 310]]<|/det|> +## MARS identifies previously identified HLAP of non-canonical origins without integration of sample-specific RNA sequencing data + +<|ref|>text<|/ref|><|det|>[[110, 325, 885, 650]]<|/det|> +To further assess the identification performance of the MARS model, and in particular the performance in the non- canonical HLA peptide space, we utilized a published dataset from Chong et al. comprising nine LC- MS datasets for five melanoma cell lines and two sets of lung cancer and healthy control tissues spanning a wide range of HLA alleles 2. The authors identify and report 508 non- canonical, HLA- presented peptides mostly sourced from IncRNA genes, which are not present in the canonical human proteome. The authors achieved identification of these peptides sequences using an immunopeptidomics approach which carefully integrates RNA sequencing and ribosome profiling data for generation of personalized, sample- specific transcript assemblies. These were then used for in- silico translation into protein databases and mapping of non- canonical, HLA- presented sequences in the immunopeptidomics data using a standard database- search. + +<|ref|>text<|/ref|><|det|>[[110, 664, 885, 854]]<|/det|> +Here, we wanted to understand whether MARS could recover these sequences without the integration of any of the additional RNA sequencing datasets used in the original publication. Firstly, we searched the Chong et al. datasets against the SwissProt human protein database using Peaks software in order to exclude spectra that were assigned to the canonical human proteome from our search. N- terminal acetylation and oxidation were included in the search parameters in order to match the analysis settings used in the original publication. We identified a total of 83,295 canonical peptides using the PEAKS + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 883, 143]]<|/det|> +database search then applied the MARS model to the remaining spectra, returning an additional 333,011 unique peptide identifications. + +<|ref|>text<|/ref|><|det|>[[111, 155, 884, 277]]<|/det|> +Overall, MARS was able to recover 227 of the 508 (45%) reported IncRNA derived peptides reported by Chong et al.. To note that only 272 out of the 508 peptides were listed as a candidate in the Peaks DeNovo search results, therefore the other sequences were "non- discoverable" by MARS. MARS therefore recovered \(83\%\) (227/272) of the possible "de novo- identifiable" space. + +<|ref|>text<|/ref|><|det|>[[111, 289, 884, 448]]<|/det|> +A clear separation of the number of peptides recovered with increasing false discovery rate was observed when applying different MARS score thresholds across all nine samples in comparison to ALC alone (Figure 3A). We therefore decided to define three major score confidence regions across this space, defining a MARS score threshold for increasing FDR: region 1: \(< 30\%\) FDR (MARS score \(>94.7\) ), region2: \(>30\%\) FDR (MARS score \(>89.7\) ), and region 3: \(>45\%\) FDR (MARS score \(>80\) ) (Figure 3A). + +<|ref|>text<|/ref|><|det|>[[111, 460, 884, 548]]<|/det|> +With a stringent MARS score threshold of \(>94.7\) , and an estimated \(30\%\) FDR, we were able to recover 175/272 (64%) of sequences, while ALC alone could recover only 101/272 (37%) peptides (Figure 3B, Table S2, Supplementary Table 3), resembling an almost 2- fold improvement in FSR. + +<|ref|>text<|/ref|><|det|>[[111, 598, 883, 650]]<|/det|> +Integration of a sequence mapping strategy for MARS peptide candidates confirms non- canonical HLAP origin. + +<|ref|>text<|/ref|><|det|>[[111, 663, 884, 855]]<|/det|> +We next aimed to assign the 333,011 MARS identified sequences in the interrogated dataset to their respective likely gene origin. We therefore developed a stratified search approach to associate MARS identified HLAP to a wide range of possible origins across the human genome. We defined peptides mapping to (i) canonical origins mapping to known protein isoforms as "Human Protein", (ii) single amino acid substitutions as "1AA substitution", and (iii) other locations in the human genome as "non- canonical" (Figure 3C, and Methods section). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 884, 380]]<|/det|> +Further to the 83,295 peptides that had initially been identified by the PEAKS database search, 19,619 sequences were additionally identified as isoforms of the known proteome space as assigned and reported by SwissProt, Tremble and Ensemble by MARS. A further 22,934 MARS sequences mapped to canonical human proteins with exactly 1 aa substitution, which we considered as likely products of germline or tumour- specific mutations (neoantigens). A total of 11,339 peptides were linearly mapped to non- canonical sequences in the human genome. For 279,119 MARS peptides we could not find an accurate match within the human genome with less than 1 aa distance, and we therefore considered these sequences as distant from the human proteome ("unmatched", Figure 3D, left panel, see Discussion). + +<|ref|>text<|/ref|><|det|>[[110, 393, 884, 550]]<|/det|> +193/272 of the Chong et al. published de novo- identifiable peptides were as expected part of the "noncanonical" peptide category as defined here. In addition, with our search strategy, we found 16 peptides reported in our DB database search, an additional 23 peptides matched to known human isoforms reported by SwissProt, Trembl and/or Ensemble, and 30 peptides to human proteins with 1 AA substitution (Figure 3D, right panel). + +<|ref|>text<|/ref|><|det|>[[110, 563, 884, 685]]<|/det|> +The 194 peptides identified by both Chong et al. and 194 MARS had an overall higher MARS score (mainly falling into regions 1 and 2) in comparison to the 13,724 additional non- canonical peptide candidates MARS recovered from the published datasets (Figure 3D, E), indicating that these peptides were recovered with high confidence and that a higher MARS score cut- off can likely limit the FDR significantly. + +<|ref|>sub_title<|/ref|><|det|>[[110, 732, 792, 752]]<|/det|> +## MARS extends the non-canonical peptide space by hundreds of peptides at an FDR <14.3% + +<|ref|>text<|/ref|><|det|>[[110, 765, 884, 888]]<|/det|> +In order to determine the actual FDR experimentally for the novel MARS non- canonical peptide identifications, we selected 97 peptides that closely resembled the observed MARS score distribution and synthesized peptides for spectral validation and FDR evaluation (Figure 4A, Supplementary Figure 3). We also included 6 peptides identified by both Chong et al. and MARS, and 3 peptides matching to the human + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 885, 344]]<|/det|> +proteome as (likely positive) controls. We found 36 out of 97 sequences to be assigned correctly, resulting in an FDR of \(14.3\%\) (region 1, MARS score \(>94.8\) ), \(45.7\%\) (region 2, MARS score \(>89.7\) ), and \(66.0\%\) (region 3, MARS score \(>80\) ) (Figure 3F,G, Supplementary Table 6). All 9 positive control peptides were confirmed. When considering only peptides that were matched to a genome origin (excluding all "unmatched" identifications), there were 42 peptides selected, the FDR could further be reduced to \(0\%\) (region 1, MARS score \(>94.8\) ), \(39.5\%\) (region 2, MARS score \(>89.7\) ), and \(63.0\%\) (region 3, MARS score \(>80\) ) (Figure 3G). This meant that all peptides with a MARS score of \(>94.8\) that mapped to a genome origin were validated in this experiment. + +<|ref|>text<|/ref|><|det|>[[111, 359, 885, 549]]<|/det|> +We then proceeded to further stratify the non- canonical peptides using GENECODE "Biotype" definitions, including functional RNA genes and IncRNA genes, protein coding regions, unannotated regions, and pseudogenes as detailed in the Method section (Figure 3H). For this, we did not apply a hierarchical strategy that would consider certain origins more likely than others, but argue that without sample specific genome/RNA level evidence, peptides with multiple possible origins cannot be faithfully assigned to a single source. + +<|ref|>text<|/ref|><|det|>[[111, 563, 885, 751]]<|/det|> +Out of the 508 peptides originally published by Chong et al., 399 peptides are defined as "non- canonical" according to our categorization, which excludes peptides with single amino acid substitutions (57 peptides, "1 AA Substitutions" category) and those that map to canonical sequence databases in our workflow (52 peptides, "Human Protein" category). MARS was able to expand the originally published, in our definition "non- canonical" peptide identifications from 399 originally published peptides in this category to 1870 peptides from non- canonical origins at a an FDR between \(0< \text{FDR}< 14.3\%\) (Figure 3H). + +<|ref|>text<|/ref|><|det|>[[111, 800, 884, 853]]<|/det|> +MARS uniquely identifies novel non- canonical IncRNA- derived HLA peptide sequences in primary tumour tissue + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 181]]<|/det|> +In order to establish whether MARS could identify novel non- canonical, IncRNA- derived peptides in primary tumour tissue, we applied MARS to a set of in- house immunopeptidomics data from ten primary cervical tumour resections spanning 35 HLA alleles (Supplementary Figure 2, Supplementary Table 5). + +<|ref|>text<|/ref|><|det|>[[111, 191, 886, 310]]<|/det|> +Apart from 32,366 Peaks database search identifications using the Peaks DeNovo- assisted database search, we could expand identifications using MARS by a further 77,129 assigned sequences. We mapped 6,551 further peptides to human source proteins, 5,828 peptides with a single amino- acid substitution, and 3,980 peptides to non- canonical origins (Figure 4A). 60,770 peptides remained unmatched. + +<|ref|>text<|/ref|><|det|>[[111, 325, 886, 682]]<|/det|> +We then further stratified the obtained non- canonical HLAP according to the annotation of their origins as described in the last section (Figure 4B). Peptides uniquely mapped to protein coding regions formed the largest subcategory, comprising of 1150 HLAP, while 811 peptides remained unmatched. We found overall 1,423 peptides associated with IncRNA origin. In order to validate that the FDR of these identifications was as expected from the Chong et al. dataset, we again selected a set of peptides for spectral validation. We chose 19 peptide candidates that had a MARS score of >89.7 (region 1 and 2), and we further prioritized peptides if more than one sequence supported the identification of an IncRNA gene. We validated 11/19 spectra (42.2% FDR), which was in the range of the expected FDR of 39.5% from previous results for Region 1 and 2 (MARS score >89.7). For Region 1 only (MARS score >94.7) we again yielded high accuracy in annotation with 7/8 correct assignments and an FDR of 12.5% (Figure 4C, Table 1). + +<|ref|>text<|/ref|><|det|>[[111, 697, 885, 820]]<|/det|> +In summary, MARS identifies thousands of peptides in addition to peptides identified with standard database- search approaches at an accuracy of above 85.7%, as confirmed in this independent primary tumour dataset, paving the way for transcriptomics- independent antigen discovery workflows to inform the development of novel cancer immunotherapies. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 181]]<|/det|> +A main limitation of non- canonical antigen discovery is the requirement of prior knowledge of expected protein sequences translated in a given sample. We here describe how we can rescue identification of non- canonical IncRNA- derived HLA- associated peptides from LC- MS/MS data without the requirement for integration of RNA sequencing data or, in fact, any other sample- specific sequence information other than HLA type, using a discovery engine that improves candidate selection from de novo sequencing results. + +<|ref|>text<|/ref|><|det|>[[111, 193, 886, 375]]<|/det|> +Since in immunopeptidomics the lengths of peptides and positions of individual amino acids in the peptides are crucial for the HLA binding and T- cell response, we here used a study point- wise- full- sequence- recall (PSTR) to assess the development of MARS instead of improvements of single tag recall STR which assesses the percentage of individual amino acids being correctly identified \(^{3,34}\) . + +<|ref|>text<|/ref|><|det|>[[111, 392, 886, 616]]<|/det|> +We evaluated the increased sensitivity using an external dataset from Wilhelm et al. \(^{17}\) , and confirmed an overall increase in performance of MARS de novo candidate selection in comparison to the best ALC scoring peptide as reported by Peaks DeNovo by over \(7\%\) . Interestingly, we saw outstanding performance for B27, B40 and B41 alleles. B27 has the requirement of an arginine at position two within the peptide sequence, while B40 and B41 are known to require a glutamic acid in this position. Hence, chemical properties of the peptide ligands are expected to be rather distinct, and we were not able to explain this much increased level of performance. + +<|ref|>text<|/ref|><|det|>[[111, 631, 886, 789]]<|/det|> +We did not observe a significant improvement for identification of HLAP originating from HLA- C. Since peptide data was acquired in equimolar ratios for the HLAP library datasets, this could not be due to lower expression of HLA- C in comparison to HLA- A and - B as generally observed across different datasets. Since NetMHCpan- 4.1 performs equally well across all three loci, we conclude that most likely we did not evaluate a sufficient number of HLA- C peptide ligands to reach a significantly better performance. + +<|ref|>text<|/ref|><|det|>[[111, 802, 886, 853]]<|/det|> +We then proceeded to evaluate how many peptide identifications could be rescued from a list of published non- canonical HLAP that had been generated using a proteogenomics approach. MARS was + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 141]]<|/det|> +able to identify \(45\%\) of the reported sequences highlighting the power of MARS to identify non- canonical antigens in the absence of transcriptomics and RiboSeq datasets. + +<|ref|>text<|/ref|><|det|>[[111, 155, 884, 413]]<|/det|> +One limitation remains the high false discovery rate associated with de novo sequencing results. With MARS, the FDR could be further limited by assigning the peptide sequences to a possible origin in the human genome and transcriptome, resulting in a manageable FDR range below \(14.3\%\) for MARS scores above \(94.8\) , compared to generally anticipated FDR of \(65\%\) for de novo sequencing results. Furthermore, considering only those sequences that mapped to a genome origin resulted in a substantial decrease of the FDR (to \(0\%\) in our experimental validation), demonstrating that MARS offers identification of non- canonical peptide sequences at low FDR, and can therefore expand the peptide identification beyond sophisticated proteogenomics approaches alone. + +<|ref|>text<|/ref|><|det|>[[111, 427, 885, 717]]<|/det|> +We here focused our analysis on non- canonical peptides derived specifically from IncRNA genes. The question remains what the origin is of the parge proportion of non- assigned peptide sequences identified by MARS. Post- translationally modified peptides, peptides derived from proteasomal fusion events, and peptides of non- human origin are all potential candidates to explain this space, which remains to be further investigated. Finally, the observed higher proportion of sequences with lower MARS score within the unmatched peptides indicates a higher FDR in this subcategory, and further suggests that these spectra may not contain sufficient information for accurate sequence matching. This observation suggests that MARS performance could further be improved by measurement of higher quality spectra for each peptide ion species, i.e. by increased ion accumulation times. + +<|ref|>text<|/ref|><|det|>[[112, 732, 884, 820]]<|/det|> +We could not validate any cysteine containing peptide assignments, confirming the known observed underrepresentation of cysteines in immunopeptidomics data \(^{35,36}\) , and highlighting the need for specific modification of the otherwise highly reactive amino acid. + +<|ref|>text<|/ref|><|det|>[[112, 835, 884, 888]]<|/det|> +Finally, we were able to shortlist novel IncRNA- derived peptide candidates from primary human tumour tissue. We shortlisted HLAP if more than one sequence supported the identification of a specific IncRNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 147]]<|/det|> +gene. We validated the sequence identifications with spectral matching to their synthetic counterpart, and we validated the FDRs measured in our previous evaluation. Note that we did not set a threshold for number of I and L in the peptide sequence, which could theoretically further reduce the FDR. + +<|ref|>text<|/ref|><|det|>[[111, 191, 884, 444]]<|/det|> +Finally, we identified and validated 11 peptides originating from IncRNA origins in human cervical cancer tissue. Peptide KVHVFLVKKT, which originated from IncRNA HLA- F- AS1, plays an oncogenic role in both colorectal cancer and breast cancer \(^{37,38}\) . The identified peptides MTMSTILSK and MTMSTILSKK were derived from IncRNA Inckb.42318 also known as MIR4458HG (ENSG00000247516), which was reported as highly expressed IncRNA in 407 TCGA evaluated ovarian tumors \(^{39}\) . It has further been reported that, together with CYTOR, and MAPTAS1, MIR4458HG was an independent prognostic factor in breast cancer patients (https://doi.org/10.21203/rs.2.24062/v1, preprint), with high expression associated with lower hazard ratios. + +<|ref|>text<|/ref|><|det|>[[111, 460, 884, 617]]<|/det|> +We conclude that MARS is a suitable platform for non- canonical antigen discovery in cancer immunopeptidomics datasets in the absence of high quality RNA sequencing data. We further highlight that the increases and unbiased results reported provide an exciting opportunity to expand non- canonical cancer antigen discovery to post- translationally modified peptides, peptides derived from proteasomal fusion events, and non- human sources, i.e. the microbiome. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 91, 195, 106]]<|/det|> +## METHODS + +<|ref|>sub_title<|/ref|><|det|>[[113, 125, 287, 141]]<|/det|> +## Human tissue material + +<|ref|>text<|/ref|><|det|>[[112, 156, 884, 245]]<|/det|> +Tissue samples were obtained from Tissue Solutions Ltd.. Ethical approval was granted by the Central University Research Ethics Committee (CUREC) of the University of Oxford under reference R68126/RE001. + +<|ref|>sub_title<|/ref|><|det|>[[113, 294, 184, 309]]<|/det|> +## Cell lines + +<|ref|>text<|/ref|><|det|>[[111, 325, 884, 583]]<|/det|> +All cell lines were mycoplasma- negative and cultured at \(37^{\circ}C\) in the presence of \(5\% \mathrm{CO_2}\) . The 721.221 HLA- class I deficient cell line that only expresses residual amounts of \(\mathsf{C}^{*}01:01^{40}\) transfected with CD4 (CD4.221) and selected HLA class I alleles ( \(\mathsf{A}^{*}01:01\) , \(\mathsf{A}^{*}02:01\) , \(\mathsf{A}^{*}03:01\) , \(\mathsf{A}^{*}11:01\) , \(\mathsf{B}^{*}08:01\) , \(\mathsf{B}^{*}44:02\) , \(\mathsf{B}^{*}57:03\) , \(\mathsf{C}^{*}03:03\) \(\mathsf{C}^{*}03:04\) ) was kindly provided by Prof. Masafumi Takaguchi, Kumamoto University, Japan. Cells were grown in RPMI 1640 medium (Thermo Fisher) containing \(10\%\) fetal bovine serum (FBS), \(2 \mathrm{mM}\) l- glutamine, \(100 \mathrm{U / mL}\) penicillin, and \(100 \mu \mathrm{g / mL}\) streptomycin (R10). R10 medium was supplemented with \(0.15 \mathrm{mg / mL}\) hygromycin B (Thermo Fisher) to maintain HLA- I expression as previously described \(^{40,41}\) . + +<|ref|>sub_title<|/ref|><|det|>[[113, 633, 430, 650]]<|/det|> +## Baseline ALC Model, or One-Factor Model + +<|ref|>text<|/ref|><|det|>[[111, 666, 884, 852]]<|/det|> +Peaks DeNovo sequencing algorithm was performed using Peaks X (Bioinformatics Solutions) on each MS2 spectrum in the studied data to produce peptide- spectrum matches (PSMs). The quality of PSMs are measured by the average local confidence (ALC, represented by symbol \(A\) in the formula), which ranged in [0,100], with 100 being the best PSM quality. We selected PSMS with ALC in range [50, 100] to ensure reasonable input quality. The top- ALC peptide of each spectrum was used as its identity. No training was required for this one- factor model. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[112, 90, 675, 108]]<|/det|> +## Integration of HLA Binding Affinity Ranking Factor (R): Two-Factor \(\alpha\) Model + +<|ref|>text<|/ref|><|det|>[[111, 123, 886, 516]]<|/det|> +To obtain numerical characterization of HLA- binding affinities, The PSM candidates were input into NetMHcpan 4.1 model, and the output eluted ligand percentile rank (ERL, represented by symbol \(R\) in the formula) predictions were used to resemble the likelihood of HLA presentation of the corresponding peptides. The ELR predictions were considered robust across different biological samples and HLAs. The output ELR ranged between [0, 100], with the smaller ranks having higher likelihood of HLA presentation. We only selected PSMs with ERL in the range [0,10] to exclude non- binding peptides. By incorporating ELR into PSM candidates selection, a two- factor \(\alpha\) model can be derived as: \((1 - \alpha)A + \alpha (100 - R)\) which is a positive affine combination of ALC (A) and ELR (R). The parameter \(\alpha\) is in the range of [0,1], by the property of affine combination, the model outputs range is kept within [0,100] to ensure factors are normalized and of comparable scale. To obtain numerical characterization of HLA- binding affinities, The PSM candidates were input into NetMHcpan 4.1 model, and the output ELR (eluted ligand percentile rank) predictions used to resemble the likelihood of HLA presentation of the corresponding peptides. + +<|ref|>sub_title<|/ref|><|det|>[[113, 563, 650, 581]]<|/det|> +## Integration of the Retention Time Factor (T), Three-Factor MARS Model + +<|ref|>text<|/ref|><|det|>[[111, 596, 886, 902]]<|/det|> +We used an adapted version of DeepRTplus, which is a special type of deep neural network model termed capsule neural networks (CapsNets), to predict the RT of input peptides. Before the trained DeepRTplus model can be used for predicting into a new set of peptides, it needs to be recalibrated for each chromatographic setup, since peptides" RT observations are known to be chromatography dependent. We used the sequence- RT pairs from de- novo peptides with \(\mathsf{ALC} > 95\) as a small but reliable set of identified peptides to re- calibrate the pre- trained DeepRTplus model, forming a transfer learning strategy to increase prediction accuracy. To develop the three- factor model by incorporating the difference in peptide chromatographic retention time (DRT, represented by symbol \(T\) in the formula) out of the DeepRTplus output as the input, firstly the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 245]]<|/det|> +absolute difference between observed and predicted RT: \(\Delta T = |T_{predicted} - T_{observed}|\) was computed. Secondly, T was computed as the \(\Delta T\) s normalised to the interval of [0,100] to match the range of ALC and ELR: \(T = 100(\Delta T_{max} - \Delta T) / (\Delta T_{max} - \Delta T_{min})\) . By this normalisation, T was transformed to a reciprocal score range, with high values being better than low values. Thirdly, to take both ELR and DRT into selecting de- novo PSM candidates, a three- factor model was defined as: \((1 - \alpha - \delta)A + \alpha (100 - R) + \delta T\) . + +<|ref|>sub_title<|/ref|><|det|>[[113, 293, 505, 311]]<|/det|> +## Performance Evaluation: Fractional Ranks (F-Ranks) + +<|ref|>text<|/ref|><|det|>[[111, 325, 885, 550]]<|/det|> +Fractional Ranks (F- Ranks) Fractional rank (F- Rank) of a correct peptide sequence for each spectrum is defined as the rank by descending order of this peptide divided by the total number of candidates of that spectrum. For peptides with tie scores, average rank was used as the numerator. It measures peptide identification performance improvement/deterioration at individual spectrum level. The reduction of mean F- rank of all correct peptides by multi- factor model outputs over the baseline ALC model was to suggest the promotion of the correct peptides. Differences in average F rank scores were evaluated with a paired T test. + +<|ref|>sub_title<|/ref|><|det|>[[113, 597, 500, 616]]<|/det|> +## Performance evaluation: Full-Sequence Recall (FSR) + +<|ref|>text<|/ref|><|det|>[[111, 630, 885, 852]]<|/det|> +FSR is defined as the ratio between the total number of correctly identified peptides and the total number of peptides in the given dataset. It measures the peptide identification sensitivity at the spectra ensemble level. For each factor model, the top- scoring PSM of each MS2 spectrum was used as an identified peptide sequence, whereas the database peptide sequence of the same MS2 spectrum was considered the correct peptide sequence. A peptide was considered correctly identified if the peptide predicted by the given model was identical (up to I/L indistinguishability) to the database peptide. Differences in FSR were evaluated with a paired T test. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[111, 89, 680, 108]]<|/det|> +## Strategy for MARS candidate HLAp origin annotation and database curation + +<|ref|>text<|/ref|><|det|>[[111, 121, 886, 311]]<|/det|> +We coarsely stratified the HLAP origins into three categories: "human protein", "single amino acid substitution", and "non- canonical". Besides, we put all HLAP not strongly associated with these three categories into another "unmatched" category. Seqkit (Shen et al. 2016) was employed for locating MARS peptides in the curated databases. For these searches, we excluded all database search against SwissProt assigned peptides with a score cut- off of - 10lgP>20, and searching the remaining de novo candidates as shortlisted by the MARS identification module. + +<|ref|>text<|/ref|><|det|>[[110, 325, 886, 888]]<|/det|> +Firstly, we searched Uniprot and Ensembl (only human protein entries) with isoforms as the "human proteome". MARS peptides matching exactly to human proteome were considered "Human Proteome" originated HLAP, matching human protein with 1AA difference were considered "1AA substitution". These two categories of peptides were considered of canonical origin, and excluded from non- canonical category. To further examine the possible alternative origins of the peptides of non- canonical origin, we used GENCODE v40 six- frame translated reference genome and three- frame translated reference transcriptome to form the protein database. All annotations were combined into Genecode "Biotypes" as follows: "Protein coding": protein_coding; "IncRNA": IncRNA; "other RNAs": miRNA, misc_RNA, Mt_rRNA, Mt_tRNA, ribozyme, rRNA, rRNA_pseudogene, scaRNA, scRNA, snoRNA, snRNA, sRNA, vault_RNA; "pseudogene": all biotypes including "pseudogene" (polymorphic, processed, transcribed_processed, transcribed_unitary, transcribed_unprocessed, translated_processed, translated_unprocessed, unitary, unprocessed); "IG": IG_C_gene, IG_C_pseudogene, IG_D_gene, IG_J_gene, IG_J_pseudogene, IG_pseudogene, IG_V_gene, IG_V_pseudogene; "TR": TR_C_gene, TR_D_gene, TR_J_gene, TR_J_pseudogene, TR_V_gene, TR_V_pseudogene and "TEC": TEC. We integrated the following databases for IncRNA mapping: LNCipedia \(^{42}\) , IncRNAKB \(^{43}\) , and added a separate "TE" category using annotations from Dfam \(^{44}\) , and an Erv database \(^{45}\) . We report our annotations in Upset plots to include multiple assignments. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 125, 250, 141]]<|/det|> +## LC-MS acquisition + +<|ref|>text<|/ref|><|det|>[[111, 155, 886, 515]]<|/det|> +All synthetic peptides were obtained from Genscript as a library service and crude purity. We measured the peptide libraries by LC- MS on an Ultimate 3000 RSLcnano System coupled with an Q Exactive™ HF Hybrid Quadrupole- Orbitrap™ Mass Spectrometer (Thermo Scientific). Peptides were loaded onto the analytical column (PepMap C18 column, \(2\mu \mathrm{m}\) particle size, \(75\mu \mathrm{m}\times 50\mathrm{cm}\) ; Thermo Scientific) and eluted in a 60 min linear gradient from \(3\%\) to \(25\%\) ACN in \(1\%\) DMSO/0.1% formic acid at a flow rate of \(250~\mathrm{nl / min}\) . Peptides were introduced to the mass spectrometer using an EasySpray source at 2000 V and \(45^{\circ}C\) , and the transfer tube temperature was set to \(305^{\circ}C\) . Mass spectrometry (MS) detection was performed with a resolution of 120,000 for full MS (320- 1600m/z scan range) and AGC target of 300,000. Precursors were isolated using an isolation width of 1.6 amu and fragmented using high- energy collisional dissociation (HCD) set at 25 for peptides with 2- 4 charges. Fragments resolution was set at 60,000 with AGC target of 50000. + +<|ref|>sub_title<|/ref|><|det|>[[115, 565, 365, 581]]<|/det|> +## LC-MS datasets and data analysis + +<|ref|>text<|/ref|><|det|>[[111, 596, 886, 750]]<|/det|> +We utilized the following datasets: Data for the C1866 cell line (A\*01:01, B\*08:01, B\*44:02, C\*05:01, and C\*07:01) and for single allele transfectant cell lines is partially available in PXD015489 40. Data for the Jurkat cell line (A\*03:01, B\*07:02, B\*35:03, C\*04:01, C07:02) is available in PXD011723 46. Wilhelm et al. peptide standard datasets are available at PXD021013 17, and Chong et al. datasets are available at PXD013649 2. + +<|ref|>text<|/ref|><|det|>[[111, 767, 886, 888]]<|/det|> +Peaks vX (Bioinformatics Solutions) was used to identify to search against a database containing the synthetic standard sequences. For database search, we used no enzyme specificity (setting: unspecific). The precursor mass error range was set within 5 ppm. Only oxidation was selected as variable modification, unless otherwise indicated in the main text. For the de novo sequencing, the precursor mass + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 883, 141]]<|/det|> +error range was set within 4.5 ppm, and only oxidation was selected as a variable modification unless otherwise indicated. + +<|ref|>sub_title<|/ref|><|det|>[[115, 192, 254, 209]]<|/det|> +## Spectral Matching + +<|ref|>text<|/ref|><|det|>[[110, 224, 884, 345]]<|/det|> +Spectral comparisons were performed manually, with the help of the Universal Spectrum Explorer (USE)47. If a synthetic peptide was not detected by LC- MS measurement, we considered the peptide sequence identification as "false", as it is equally unlikely that the identical peptide molecule was successfully ionized and detected during the experiment. + +<|ref|>sub_title<|/ref|><|det|>[[115, 395, 349, 412]]<|/det|> +## Trinity De-Novo RNA Assembly + +<|ref|>text<|/ref|><|det|>[[110, 427, 885, 650]]<|/det|> +Raw RNA sequencing reads had adaptors removed and low quality bases trimmed using Trim_Galore v0.6.2 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore) and quality control metrics assessed using FastQC v0.11.9 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc). A custom STAR (v2.7.3a)48 genome reference was created utilising UCSC hg38 genome sequence. Reads were then mapped to this custom genome reference using STAR with two passes, sorted and indexed by Samtools v1.10 49. Duplicates were marked using GATK v4.1.7.0 50. Unmapped reads from STAR were used as inputs for Trinity v2.11.0 51 for de novo transcript assembly. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 91, 281, 107]]<|/det|> +## ACKNOWLEDGEMENT + +<|ref|>text<|/ref|><|det|>[[111, 123, 886, 515]]<|/det|> +We thank Prof. Masafumi Takiguchi, Kumamoto University, Japan for provision of CD4- expressing cell lines transfected with single HLA- I alleles. This work was funded by the NIH project #1R21AI153978- 01 (LCE), The Leona M. and Harry B. Helmsley Charitable Trust (RM), the European Association for the Study of Diabetes FSD/JDRF/Lilly Programme on Type 1 Diabetes Research 2019 (NT), Cancer Research UK Cancer Immunology project award C55884/A21045 (NT), and Cancer Research UK RadNet Centre Award C6078/A28736 (NT). Some immunopeptidomic datasets employed for this study were generated as part of studies supported by Medical Research Council (MRC) programme grant MR/K012037, grants from the National Institutes of Health (UM1 AI 100645 and R01 AI 118549) and by AMED grant 17FK0410302H0003 for AIDS research (PB). The computational aspects of this research were supported by the Wellcome Trust Core Award Grant Number 203141/Z/16/Z and the NIHR Oxford BRC. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. PB is a Jenner Institute Investigator. + +<|ref|>sub_title<|/ref|><|det|>[[113, 564, 350, 580]]<|/det|> +## DATA AND CODE AVAILABILITY + +<|ref|>text<|/ref|><|det|>[[112, 597, 884, 682]]<|/det|> +Primary cervical tumour mass spectrometry data will be published in the Proteomics Identifications Database (https://www.ebi.ac.uk/pride/) upon acceptance of the manuscript. Code will be provided to academic researchers upon reasonable request. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 123, 70, 140]]<|/det|> +545 + +<|ref|>text<|/ref|><|det|>[[45, 163, 288, 184]]<|/det|> +546 Table 1: IncRNA peptide validation + +<|ref|>text<|/ref|><|det|>[[45, 205, 576, 226]]<|/det|> +547 Red highlighted: MARS identification not reported by Peaks DeNovo as first candidate + +<|ref|>table<|/ref|><|det|>[[88, 244, 920, 680]]<|/det|> +
PEPTIDEMARS SCOREHLA ALLELEBINDING RANKGENE LOCATIONLNC-RNAALT NAMESAMPLESPECRAL MATCH
IIYFCLHKI93.4HLA-A*02:010.629chr5:164989217-164989191Inckb.44609CTC-340A15.24,5No
LPYIKWTI93.6HLA-B*51:010.076chr5:164805831-164805808Inckb.44609CTC-340A15.27Yes
LSSTLNKQI90.1HLA-C*02:021.143chr5:164743628-164743602Inckb.44609CTC-340A15.21No
IISASKVIL92.3HLA-A*02:010.617chr11:90809963-90809989Inckb.9710DISC1FP17,9No
TELKAWKI95.2HLA-B*37010.138chr11:90693569-90693592Inckb.9710DISC1FP13No
IRNVKIYLI91.2HLA-C*06:020.09Inckb.45851HLA-F-AS110No
KVHVFLVKK99.7HLA-A*03:010.029chr6:29750586-29750560Inckb.45851HLA-F-AS11,8Yes
RQAPALRHL92.5HLA-C*06:020.133chr13:94909331-94909357Inckb.14510LOC1019272849Yes
VIFSGIRSL95.4HLA-A*02:010.081chr13:94926124-94926150Inckb.14510LOC1019272845Yes
VLLSIFIEHL94.2HLA-A*02:011.15chr13:94845116-94845145Inckb.14510LOC1019272847No
ALRAVTLTAK97.9HLA-A*03:010.099chr3:35175054-35175080Inckb.36918LOC1019281358,10Yes
KQLNKQLIK98.3HLA-A*03010.053chr8:78259721-78259747Inckb.53745LOC1053759114,9Yes
VTIFQNRVK92HLA-A*11:010.575chr8:78413273-78413247Inckb.53745LOC1053759118Yes
KTKIIGTSK94.1HLA-A*03:010.192chrX:75675106-75675132Inckb.58303LOC1079856648Yes
SLSILSLKV92.9HLA-B*13:020.164chrX:75543586-75543560Inckb.58303LOC1079856641No
IPHGEIPDTSA94.9HLA-B*56:010.064chr2:172323392-172323424Inckb.31346LOC1079859605Yes
LPNFARII92.6HLA-B*07:020.29chr2:172280278-172280255Inckb.31346LOC1079859604,8No
MTMSTILSK99.2HLA-A*11:010.008chr5:8460009-8460035Inckb.42318MIR4458HG8Yes
MTMSTILSKK99.8HLA-A*03:010.199chr5:8460009-8460038Inckb.42318MIR4458HG4,8,9Yes
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[142, 118, 780, 444]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 87, 210, 109]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[56, 454, 884, 549]]<|/det|> +549 Figure 1: Schematic of MARS workflow and performance evaluation in a five- fold cross- validation. (A) Schematic of integrative modules in the MARS pipeline (B) Mean F- rank score of all partitions, and (C) Full sequence recall (FSR) achieved across the partitions in the target peptide space. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 110, 880, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 87, 199, 106]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[112, 660, 884, 830]]<|/det|> +Figure 2: MARS sensitivity evaluation in Wilhelm et al. large HLA peptide library datasets. (A) The average F- rank distribution across all five partitions and (B) Venn diagram depicting the number of peptide sequences identified by ALC (green) and MARS (red), respectively, within those sequences that were present in the 20 de novo candidates as reported by Peaks ("de novo identifiable", dark grey). Total sequence space is shown in light grey. (C) Box plots indicating the number of peptide sequences uniquely identified by MARS for all HLA- A, - B, and - C alleles combined (D) Full sequence recall (FSR) for individual HLA alleles for ALC (green) and MARS (red). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 103, 884, 884]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 87, 194, 105]]<|/det|> +
Figure 3
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 884, 636]]<|/det|> +Figure 3: Non- canonical peptide identification using MARS in a published dataset from Chong et al. \(^{2}\) . (A) Left panel: Sequence recall for IncRNA peptide sequences identified by Chong at all across all samples plotted against the false discovery rate. The three regions are defined by FDR cut- offs at \(30\%\) and \(45\%\) , respectively. The right panel shows these regions in relation to the MARS score. (B) Venn diagrams demonstrating the overlap of sequence identifications for each of the three defined regions from (A). (C) Overview of MARS search engine and blast search strategy for identification of source origin. (D) Left panel: Bar graph showing the categorization of the MARS hits as indicated. Right panel: Mutually identified peptides by MARS and Chong et al. from the 508 reported non- canonical identifications assigned to the MARS categories as indicated. Colored subsections of the graph indicate the original alternative classification as applied by Chong et al. in the original publication (E) Score distribution of all MARS identified non- canonical peptides (11339) in comparison to those mutually identified by Chong et al. (193). (F) Experimental validation of the FDR using a synthetic peptide standard. Histogram summarizing the peptide score distribution of the synthetic peptide standard. Shown are number of peptides identified by MARS only for indicated MARS score bins (upper panel), and FDP progression in relation to declining MARS score (lower panel) calculated on the spectral matching outcome. (G) Number and proportion of all confirmed peptides for all synthetic standard sequences and for the subset of peptides that were mapped to the human genome (“excluding unmatched”). (H) Upset plot of the MARS non- canonical peptide top 10 categories for all hits in region 1, 2, and 3 (lower panel), and mutual identifications as reported in Chong et al. (upper graph). The Venn diagram indicates the overall number of non- canonical sequences that were identified with MARS at a FDR \(\leq 14.3\%\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 101, 881, 730]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 87, 194, 105]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[111, 737, 884, 876]]<|/det|> +Figure 4: Identification of novel non- canonical HLAP in primary tumour tissue using MARS. (A) Bar graph showing the categorization of the MARS hits as indicated. (B) Upset plot of the MARS non- canonical peptide category for all hits (C) Bar graph depicting the level of evidence found in trinity assemblies of RNAseq data for each IncRNA category. (D) Score distribution of all MARS peptide hits in comparison of those that are validated with additional RNA sequence evidence. (E) Spectral matching validation for indicated sequences identified in the primary tumour tissue. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 150, 886, 865]]<|/det|> +589 1. Jaeger, A.M. et al. Deciphering the immunopeptidome in vivo reveals new tumour antigens. Nature (2022). 591 2. Chong, C. et al. Integrated proteogenomic deep sequencing and analytics accurately identify non-canonical peptides in tumor immunopeptidomes. Nat Commun 11, 1293 (2020). 594 3. Muth, T. & Renard, B.Y. Evaluating de novo sequencing in proteomics: already an accurate alternative to database-driven peptide identification? Brief Bioinform 19, 954- 970 (2018). 597 4. Bassani-Sternberg, M. et al. Direct identification of clinically relevant neoepitopes presented on native human melanoma tissue by mass spectrometry. Nat Commun 7, 13404 (2016). 600 5. Ternette, N. et al. Defining the HLA class I- associated viral antigen repertoire from HIV- 1- infected human cells. Eur J Immunol 46, 60- 69 (2016). 602 6. Bettencourt, P. et al. Identification of antigens presented by MHC for vaccines against tuberculosis. NPJ Vaccines 5, 2 (2020). 604 7. Attig, J. et al. LTR retroelement expansion of the human cancer transcriptome and immunopeptidome revealed by de novo transcript assembly. Genome Res 29, 1578- 1590 (2019). 607 8. Kracht, M.J. et al. Autoimmunity against a defective ribosomal insulin gene product in type 1 diabetes. Nat Med 23, 501- 507 (2017). 609 9. Zanker, D.J. et al. Influenza A Virus Infection Induces Viral and Cellular Defective Ribosomal Products Encoded by Alternative Reading Frames. J Immunol 202, 3370- 3380 (2019). 612 10. Parker, R. et al. The Choice of Search Engine Affects Sequencing Depth and HLA Class I Allele- Specific Peptide Repertoires. Mol Cell Proteomics 20, 100124 (2021). 614 11. Bichmann, L. et al. MHCquant: Automated and Reproducible Data Analysis for Immunopeptidomics. J Proteome Res 18, 3876- 3884 (2019). 616 12. Mann, M. & Wilm, M. Error- tolerant identification of peptides in sequence databases by peptide sequence tags. Anal Chem 66, 4390- 4399 (1994). 618 13. Yates, J.R., 3rd, Eng, J.K., McCormack, A.L. & Schieltz, D. Method to correlate tandem mass spectra of modified peptides to amino acid sequences in the protein database. Anal Chem 67, 1426- 1436 (1995). 621 14. Nesvizhskii, A.I. Protein identification by tandem mass spectrometry and sequence database searching. Methods Mol Biol 367, 87- 119 (2007). 623 15. AS, C.S., Bouwmeester, R., Martens, L. & Degroeve, S. Accurate peptide fragmentation predictions allow data driven approaches to replace and improve upon proteomics search engine scoring functions. Bioinformatics 35, 5243- 5248 (2019). 626 16. Gessulat, S. et al. Prosit: proteome- wide prediction of peptide tandem mass spectra by deep learning. Nat Methods 16, 509- 518 (2019). 627 17. Wilhelm, M. et al. Deep learning boosts sensitivity of mass spectrometry- based immunopeptidomics. Nat Commun 12, 3346 (2021). 630 18. Kong, A.T., Leprevost, F.V., Avtonomov, D.M., Mellacheruvu, D. & Nesvizhskii, A.I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry- based proteomics. Nat Methods 14, 513- 520 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 90, 886, 890]]<|/det|> +633 19. Li, K., Jain, A., Malovannaya, A., Wen, B. & Zhang, B. DeepRescore: Leveraging Deep Learning to Improve Peptide Identification in Immunopeptidomics. \*Proteomics\* 20, e1900334 (2020).636 20. Tran, N.H., Zhang, X., Xin, L., Shan, B. & Li, M. De novo peptide sequencing by deep learning. \*Proc Natl Acad Sci U S A\* 114, 8247- 8252 (2017).637 21. Yang, H., Chi, H., Zeng, W.F., Zhou, W.J. & He, S.M. pNovo 3: precise de novo peptide sequencing using a learning- to- rank framework. \*Bioinformatics\* 35, i183- i190 (2019).640 22. Ma, B. et al. PEAKS: powerful software for peptide de novo sequencing by tandem mass spectrometry. \*Rapid Commun Mass Spectrom\* 17, 2337- 2342 (2003).641 23. Erhard, F., Dolken, L., Schilling, B. & Schlosser, A. Identification of the Cryptic HLA- I Immunopeptidome. \*Cancer Immunol Res\* 8, 1018- 1026 (2020).642 24. Laumont, C.M. et al. Global proteogenomic analysis of human MHC class I- associated peptides derived from non- canonical reading frames. \*Nat Commun\* 7, 10238 (2016).643 25. Ouspenskaja, T. et al. Unannotated proteins expand the MHC- I- restricted immunopeptidome in cancer. \*Nat Biotechnol\* 40, 209- 217 (2022).644 26. Andreatta, M. et al. MS- rescue: A Computational Pipeline to Increase the Quality and Yield of Immunopeptidomics Experiments. \*Proteomics\*, e1800357 (2018).645 27. Ma, C. et al. Improved Peptide Retention Time Prediction in Liquid Chromatography through Deep Learning. \*Anal Chem\* 90, 10881- 10888 (2018).646 28. Tran, N.H. et al. Personalized deep learning of individual immunopeptidomes to identify neoantigens for cancer vaccines. \*Nat Mach Intell\* 2 (2020).647 29. Zhang, J. et al. PEAKS DB: de novo sequencing assisted database search for sensitive and accurate peptide identification. \*Mol Cell Proteomics\* 11, M111 010587 (2012).648 30. Tran, N.H. et al. Deep learning enables de novo peptide sequencing from data- independent- acquisition mass spectrometry. \*Nat Methods\* 16, 63- 66 (2019).649 31. Gonzalez- Galarza, F.F. et al. Allele frequency net database (AFND) 2020 update: gold- standard data classification, open access genotype data and new query tools. \*Nucleic Acids Res\* 48, D783- D788 (2020).650 32. Martini, S., Nielsen, M., Peters, B. & Sette, A. The Immune Epitope Database and Analysis Resource Program 2003- 2018: reflections and outlook. \*Immunogenetics\* 72, 57- 76 (2020).651 33. Vita, R. et al. The Immune Epitope Database (IEDB): 2018 update. \*Nucleic Acids Res\* 47, D339- D343 (2019).652 34. Xu, C. & Ma, B. Complexity and scoring function of MS/MS peptide de novo sequencing. \*Comput Syst Bioinformatics Conf\*, 361- 369 (2006).653 35. Meadows, L. et al. The HLA- A\*0201- restricted H- Y antigen contains a posttranslationally modified cysteine that significantly affects T cell recognition. \*Immunity\* 6, 273- 281 (1997).654 36. Trujillo, J.A. et al. The cellular redox environment alters antigen presentation. \*J Biol Chem\* 289, 27979- 27991 (2014).655 37. Huang, Y. et al. HLA- F- AS1/miR- 330- 3p/ PFN1 axis promotes colorectal cancer progression. \*Life Sci\* 254, 117180 (2020).656 38. Wu, D., Jia, H., Zhang, Z. & Li, S. STAT3- induced HLA- F- AS1 promotes cell proliferation and stemness characteristics in triple negative breast cancer cells by upregulating TRABD. \*Bioorg Chem\* 109, 104722 (2021).657 39. Akrami, R. et al. Comprehensive analysis of long non- coding RNAs in ovarian cancer reveals global patterns and targeted DNA amplification. \*PLoS One\* 8, e80306 (2013).658 40. Partridge, T. et al. Discrimination Between Human Leukocyte Antigen Class I- Bound and Co- Purified HIV- Derived Peptides in Immunopeptidomics Workflows. \*Front Immunol\* 9, 912 (2018).659 41. Paes, W. et al. Contribution of proteasome- catalyzed peptide cis- splicing to viral targeting by CD8(+) T cells in HIV- 1 infection. \*Proc Natl Acad Sci U S A\* 116, 24748- 24759 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 90, 884, 448]]<|/det|> +683 42. Volders, P.J. et al. LNCipedia 5: towards a reference set of human long non- coding RNAs. 684 Nucleic Acids Res 47, D135- D139 (2019). 685 43. Seifuddin, F. et al. lncRNAKB, a knowledgebase of tissue- specific functional annotation 686 and trait association of long noncoding RNA. Sci Data 7, 326 (2020). 687 44. Storer, J., Hubley, R., Rosen, J., Wheeler, T.J. & Smit, A.F. The Dfam community resource 688 of transposable element families, sequence models, and genome annotations. Mob DNA 689 12, 2 (2021). 690 45. Saini, S.K. et al. Human endogenous retroviruses form a reservoir of T cell targets in 691 hematological cancers. Nat Commun 11, 5660 (2020). 692 46. Nicastri, A., Liao, H., Muller, J., Purcell, A.W. & Ternette, N. The Choice of HLA- Associated 693 Peptide Enrichment and Purification Strategy Affects Peptide Yields and Creates a Bias 694 in Detected Sequence Repertoire. Proteomics 20, e1900401 (2020). 695 47. Schmidt, T. et al. Universal Spectrum Explorer: A Standalone (Web-)Application for Cross- 696 Resource Spectrum Comparison. J Proteome Res 20, 3388- 3394 (2021). 697 48. Dobin, A. et al. STAR: ultrafast universal RNA- seq aligner. Bioinformatics 29, 15- 21 698 (2013). 699 49. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078- 700 2079 (2009). 701 50. McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing 702 next- generation DNA sequencing data. Genome Res 20, 1297- 1303 (2010). 703 51. Grabherr, M.G. et al. Full- length transcriptome assembly from RNA- Seq data without a 704 reference genome. Nature Biotechnology 29, 644- U130 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 88, 365, 105]]<|/det|> +SUPPLEMENTARY INFORMATION + +<|ref|>text<|/ref|><|det|>[[60, 123, 884, 175]]<|/det|> +Supplementary Figure 1: I replacement by L maintains validity of NetMHCpan 4.1 prediction across distinct HLA alleles + +<|ref|>text<|/ref|><|det|>[[60, 191, 864, 228]]<|/det|> +Supplementary Figure 2: HLA allele distributions for Chong et al. dataset, and cervical tumour patient cohort. + +<|ref|>text<|/ref|><|det|>[[60, 250, 770, 269]]<|/det|> +Supplementary Figure 3 (provided separately): Spectral matching for FDR determination + +<|ref|>text<|/ref|><|det|>[[60, 285, 879, 384]]<|/det|> +Supplementary Data Table 1: Five- Fold Cross- Validations using Baseline ALC model, Two- Factor alpha model, Two- Factor beta model, and Three- Factor MARS model. Column 2- 5 shows the optimal parameters learned by grid search for each part, suggesting the models are robust. Column 6- 8 gives the results of Paired Wilcoxon signed- rank tests on F- ranks of true peptide sequences; significant drop of F- ranks indicates improved identification power. + +<|ref|>text<|/ref|><|det|>[[60, 404, 860, 444]]<|/det|> +Supplementary Data Table 2: Number of peptide sequences reported to bind to each respective HLA allele in the Wilhelm et al. dataset + +<|ref|>text<|/ref|><|det|>[[60, 463, 680, 482]]<|/det|> +Supplementary Data Table 3: FSR rates for Chong et al. data for each sample + +<|ref|>text<|/ref|><|det|>[[60, 500, 880, 539]]<|/det|> +Supplementary Data Table 4: Overview of MARS peptide identifications for each sample in the Chong et al. dataset + +<|ref|>text<|/ref|><|det|>[[60, 558, 680, 577]]<|/det|> +Supplementary Data Table 5: HLA- typing results for cervical tumour patients + +<|ref|>text<|/ref|><|det|>[[60, 596, 694, 614]]<|/det|> +Supplementary Data Table 6: FDR determination: Peptide Validation Summary + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 110, 880, 641]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 87, 355, 108]]<|/det|> +
Supplementary Figure 1
+ +<|ref|>text<|/ref|><|det|>[[111, 655, 884, 842]]<|/det|> +Supplementary Figure 1: I replacement by L maintains validity of NetMHCan 4.1 prediction across distinct HLA alleles. (A) K- L divergences were computed as statistics to characterize the similarity of two distributions D(Original || Only I) and D(Original || Only L). KLDiv_Boxplot shows that Only L had consistently very small K- L divergences, and a paired Wilcoxon signed rank test suggested that D(Original || Only I) is significantly greater than D(Original || Only L) with \(\mathsf{p} = 0.0156\) . (B) Rank score distributions for example alleles for original peptide sequence, all I replaced by L (Only L), and vice versa (Only I). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[175, 120, 666, 767]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[202, 121, 384, 136]]<|/det|> +
Chong et al. Allele Counts
+ +<|ref|>text<|/ref|><|det|>[[60, 750, 90, 764]]<|/det|> +740 + +<|ref|>text<|/ref|><|det|>[[60, 768, 90, 782]]<|/det|> +741 + +<|ref|>text<|/ref|><|det|>[[60, 788, 86, 802]]<|/det|> +742 + +<|ref|>text<|/ref|><|det|>[[60, 808, 86, 821]]<|/det|> +743 + +<|ref|>text<|/ref|><|det|>[[60, 828, 90, 841]]<|/det|> +744 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 131, 312, 283]]<|/det|> +SupplementaryTable1. csv SupplementaryTable2. csv SupplementaryTable3. csv SupplementaryTable4. csv SupplementaryTable5. csv SupplementaryTable6. csv + +<--- Page Split ---> diff --git a/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/images_list.json b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d1962a389e4a73663ea5fb3eb10bf641693cad8f --- /dev/null +++ b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/images_list.json @@ -0,0 +1,130 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 – Structure and Assumptions of Temporal Bases for Temporal Difference", + "footnote": [], + "bbox": [ + [ + 100, + 150, + 744, + 567 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2– A fixed RNN as a basis function generator. a) Schematic of a fixed recurrent neural network as a temporal basis. The network receives external inputs and generates states s(t), which act as a basis for learning the value function V(t). Compare to figure 1c. b) Schematic of the task protocol. Every presentation of C is followed by a reward at a fixed delay of 1000ms. However, any combination or sequence of irrelevant stimuli may precede the conditioned stimulus C (they might also come after the CS, e.g. A,C,B). c) Network activity, plotted along its first two principal components, for a given initial state s0 and a sequence of presented stimuli A-B-C (red letter is displayed at the time of a given stimulus' presentation). d) Same as c) but for input sequence B-A-C. e) Overlay of the A-B-C ad B-A-C network trajectories, starting from the state at the time of the presentation of C (state sc). The trajectory of network activity differs in these two cases, so the RNN state does not provide a consistent temporal basis that tracks the time since the presentation of stimulus C.", + "footnote": [], + "bbox": [ + [ + 95, + 262, + 870, + 576 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3- Certain features of experimental results run counter to predictions of TD. a) Putative temporal basis functions observed in experiments \\(^{16,19}\\) after training are plastic, shown here schematically. If, after training, the interval between CS and US is scaled, the basis-functions also change. Recordings in striatum \\(^{16}\\) show these basis-functions scale with the modified interval (top), while in recordings from hippocampus \\(^{19}\\) (bottom), they are observed to redistribute to fill up the new interval. b) According to the \\(TD(0)\\) theory, RPE neuron activity during learning exhibits a backward moving bump (left). For \\(TD(\\lambda)\\) the bump no longer appears (right). c) A schematic depiction of experiments where there is no backward shifting bump \\(^{22,24}\\) . d) The integral of DA neuron activity according to TD theory (left) should be constant over training (for \\(\\gamma = 1\\) ) or decreasing monotonically for \\((\\gamma < 1)\\) . We reanalyzed existing experimental data (right) and found that the integral can transiently significantly increase. The inset shows the mean and standard errors for early (1-2), intermediate (3-4) and late (8-10) days. (Data from Amo et al 2022 \\(^{23,25}\\) , see Supplemental Figure 1).", + "footnote": [], + "bbox": [ + [ + 100, + 70, + 910, + 530 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4- Potential Architectures for a Flexible Temporal Basis. Three example networks which could implement a FLEX theory. Top, network architecture. Middle, activity of one example neuron in the network. Bottom, network activity before and after training. Each network initially only has transient responses to stimuli, modifying plastic connections (in blue) during association to develop a specific temporal basis to reward-predictive stimuli. a) Synfire chains could support a FLEX model, if the chain could recruit more members during learning, exclusive to reward-predictive stimuli. b) A population of neurons with homogenous recurrent connections have a characteristic decay time that is related to the strength of the weights. The cue-relative time can then be read out by the mean level of activity in the network. c) A population of neurons with heterogenous and large recurrent connections (liquid state machine) can represent cue-relative time by treating the activity vector at time t as the \"microstate\" representing time t (as opposed to b) where only mean activity is used).", + "footnote": [], + "bbox": [ + [ + 100, + 180, + 890, + 560 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5 – Biophysically Inspired Architecture Allows for Flexible Encoding of Time a) Diagram of model architecture. CNAs (visualized here as columns) located in PFC are selective to certain sensory stimuli (indicated here by color atop the column) via fixed excitatory inputs (CS). VTA DA neurons receive fixed input from naturally positive valence stimuli, such as food or water reward (US). DA neuron firing releases dopamine, which acts as a learning signal for both PFC and VTA. Solid lines indicate fixed connections, while dotted lines indicate learned connections. b,c) Schematic representation of data adapted from Liu et al. 201542. b) Timers learn to characteristically decay at the time of cue-predicted reward. c) Messengers learn to have a firing peak at the time of cue-predicted reward.", + "footnote": [], + "bbox": [ + [ + 95, + 144, + 888, + 468 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6- CS-evoked and US-evoked Model Dopamine Responses Evolve on Different Timescales. The model is trained for 40 trials while being presented with a CS at 100ms and a reward at 1100ms. a) Mean firing rates for the CNA (see inset for colors), for three different stages of learning. b) Mean firing rate over all DA neurons taken at the same three stages of learning. Firing above or below thresholds (dotted lines) evokes positive or negative D(t) in PFC. c) Evolution of mean synaptic weights over the course of learning. Top, middle, and bottom, mean strength of \\(\\mathrm{T}\\rightarrow \\mathrm{T}\\) , CS \\(\\rightarrow\\) DA, and M \\(\\rightarrow\\) GABA synapses, respectively. d) Area under receiver operating characteristic (auROC, see Methods) for all VTA neurons in our model for 15 trials before (left, US only), 15 trials during (middle, CS+US), and 15 trials after (right, CS+US) learning. Values above and below 0.5 indicate firing rates above and below the baseline distribution.", + "footnote": [], + "bbox": [ + [ + 113, + 60, + 923, + 560 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7 – FLEX Model Dynamics Diverges from those of TD Learning Dynamics of both TD(λ) and FLEX when trained with the same conditioning protocol as shown in Figure 6. a) Dopaminergic release D(t) for FLEX (left), and RPE for TD(λ) (right), over the course of training. b) Sum total (i.e., the sum of each row in a) is a point on the line of b)) dopaminergic reinforcement D(t) during a given trial of training in FLEX (black), and sum total RPE in three instances of TD(λ) with discounting parameters \\(\\gamma = 1\\) , \\(\\gamma = .99\\) , and \\(\\gamma = .95\\) (red, orange, and yellow, respectively). Shaded areas indicate functionally different stages of learning in FLEX. The learning rate in our model is reduced in this example, as to make direct comparison with TD(λ). 100 trials of US-only presentation (before learning) are included for comparison with subsequent stages.", + "footnote": [], + "bbox": [ + [ + 103, + 280, + 840, + 711 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8 – FLEX model Reconciles Differing Experimental Phenomena Observed during Sequential Conditioning Results from “sequential conditioning”, where sequential neutral stimuli CS1 and CS2 and are paired with delayed reward US. a) Visualization of protocol. In this example, the US is presented starting at 1500ms, with CS1 is presented starting at 100ms, and CS2 is presented starting at 800ms. b) Mean firing rates over all DA neurons, for four distinctive stages in learning – initialization(i), acquisition(ii), reward depression(iii), and serial transfer of activation(iv). c) Schematic illustration of experimental results from recorded dopamine neurons, labeled with the matching stage of learning in our model. c) DA neuron firing before (top), during (middle) and after (bottom) training, wherein two cues (0s and 4s) were followed by a single reward (6s). Adapted from Pan et al. 200522. d) DA neuron firing after training wherein two cues (instruction, trigger) were followed by a single reward. Adapted from Schultz et al. 199349.", + "footnote": [], + "bbox": [ + [ + 139, + 66, + 875, + 485 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Supplemental Figure 6 – Expected Rewards Block Learning of New Cue-Reward Associations Unless Reward Magnitude is Increased", + "footnote": [], + "bbox": [], + "page_idx": 29 + } +] \ No newline at end of file diff --git a/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3.mmd b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3.mmd new file mode 100644 index 0000000000000000000000000000000000000000..671cbdc322ffde8e83bfa20ebc5d0a5dd02b418b --- /dev/null +++ b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3.mmd @@ -0,0 +1,571 @@ + +# Learning to Express Reward Prediction Error-like Dopaminergic Activity Requires Plastic Representations of Time + +Harel Shouval + +harel.shouval@uth.tmc.edu + +University of Texas Health https://orcid.org/0000- 0003- 2799- 1337 + +lan Cone Imperial College London Claudia Clopath Imperial College London + +## Article + +Keywords: + +Posted Date: September 19th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3289985/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on July 12th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 50205- 3. + +<--- Page Split ---> + +# Learning to Express Reward Prediction Error-like Dopaminergic Activity Requires Plastic Representations of Time + +Abbreviated title: RPE via Plastic Representations of Time + +Ian Cone1,2,3, Claudia Clopath1, Harel Z. Shouval2,4 1Department of Bioengineering, Imperial College London, London, United Kingdom; 2Department of Neurobiology and Anatomy, University of Texas Medical School at Houston, Houston, TX; 3Applied Physics Program, Rice University, Houston, TX; 4Department of Electrical and Computer Engineering, Rice University, Houston, TX + +Corresponding Autor: Harel Shouval, email: harel.shouval@uth.tmc.edu + +No financial conflict of interest + +<--- Page Split ---> + +## Abstract + +The dominant theoretical framework to account for reinforcement learning in the brain is temporal difference (TD) reinforcement learning. The TD framework predicts that some neuronal elements should represent the reward prediction error (RPE), which means they signal the difference between the expected future rewards and the actual rewards. The prominence of the TD theory arises from the observation that firing properties of dopaminergic neurons in the ventral tegmental area appear similar to those of RPE model- neurons in TD learning. Previous implementations of TD learning assume a fixed temporal basis for each stimulus that might eventually predict a reward. Here we show that such a fixed temporal basis is implausible and that certain predictions of TD learning are inconsistent with experiments. We propose instead an alternative theoretical framework, coined FLEX (Flexibly Learned Errors in Expected Reward). In FLEX, feature specific representations of time are learned, allowing for neural representations of stimuli to adjust their timing and relation to rewards in an online manner. In FLEX dopamine acts as an instructive signal which helps build temporal models of the environment. FLEX is a general theoretical framework that has many possible biophysical implementations. In order to show that FLEX is a feasible approach, we present a specific biophysically plausible model which implements the principles of FLEX. We show that this implementation can account for various reinforcement learning paradigms, and that its results and predictions are consistent with a preponderance of both existing and reanalyzed experimental data. + +<--- Page Split ---> + +The term reinforcement learning is used in machine learning \(^{1}\) , in behavioral science \(^{2}\) and in neurobiology \(^{3}\) , to denote learning on the basis of rewards or punishment. One type of reinforcement learning is temporal difference (TD) learning, which was designed for machine learning purposes. It has the normative goal of estimating future rewards when rewards can be delayed in time with respect to the actions or cues that engendered these rewards \(^{1}\) . + +One of the variables in TD algorithms is called reward prediction error (RPE), and it is simply the difference between the discounted predicted reward at the current state and the discounted predicted reward + actual reward at the next state. The concept of TD learning became prominent in neuroscience once it was demonstrated that firing patterns of dopaminergic neurons in ventral tegmental area (VTA) during reinforcement learning resemble RPE \(^{4 - 6}\) . + +Implementations of TD using computer algorithms are straightforward but are more complex when they are mapped onto plausible neural machinery \(^{7 - 9}\) . Current implementations of neural TD assume a set of temporal basis- functions \(^{9,10}\) . These basis functions are activated by external cues. For this assumption to hold, each possible external cue must activate a separate set of basis- functions, and these basis- functions must tile all possible learnable intervals between stimulus and reward. + +In this paper we demonstrate that 1) these assumptions are implausible from a fundamental conceptual level, and 2) predictions of such algorithms are inconsistent with various established experimental results. Instead, we propose that temporal basis functions used by the brain are themselves learned. We call this theoretical framework: Flexibly Learned Errors in Expected Reward, or FLEX for short. We also propose a biophysically plausible implementation of FLEX, as a proof- of- concept model. We show that key predictions of this model are consistent with actual experimental results but are inconsistent with some key predictions of the TD theory. + +## Results + +## TD learning with a fixed feature specific temporal-basis + +The original TD learning algorithms assumed that agents can be in a set of discrete labeled states (s) which are stored in memory. The goal of TD is to learn a value function such that each state becomes associated with a unique value \((V(s))\) that estimates future discounted rewards. Learning is driven by the difference between value at two subsequent states, and hence such algorithms are called temporal difference algorithms. Mathematically this is captured by the update algorithm: \(V(s) \leftarrow V(s) + \alpha (r(s') + \gamma V(s') - V(s))\) , where \(s'\) is the next state and \(r(s')\) is the reward in next state, \(\gamma\) is an optional discount factor and \(\alpha\) is the learning rate. + +The term in the brackets in the right- hand side of the equation is called the RPE. It represents the different between the estimated value at the current state and the estimated discounted value at the next state in addition to the actual reward at the next state. If RPE is zero for every state, the value function no longer changes, and learning reaches a stable state. In experiments that linked RPE to the firing patterns of dopaminergic neurons in VTA, a transient conditioned stimulus (CS) is presented to a naive animal followed by a delayed reward (also called unconditioned stimulus or US, Figure 1a). It was found that VTA neurons initially respond at the time of reward, but once the association between stimulus and reward is learned, neurons stop firing at the time of the reward and start firing at the time of the stimulus (Figure 1b). This response pattern is what one would expect from TD learning if VTA neurons represent RPE \(^{5}\) . + +Learning algorithms similar to TD have been very successful in machine learning \(^{11,12}\) . In such implementations the state (s) could, for example, represent the state of the chess board, or the coordinates in space in a navigational task. Each of these states could be associated + +<--- Page Split ---> + +with a value. The state space in such examples might be very large, but the values of all these different states could be feasibly stored in a computer's memory. In some cases, a similar formulation seems feasible for a biological system as well. For example, consider a 2- D navigation problem, where each state is a location in space. One could imagine that each state would be represented by the set of hippocampal place cells activated in this location \(^{13}\) , and that another set of neurons would encode the value function, while a third population of neurons (the "RPE") neurons would compare the value at the current and subsequent state. On its face, this seems to be a reasonable assumption. + +However, in contrast to cases where a discrete set of states might have straightforward biological implementation, there are many cases in which this machine learning inspired algorithm cannot be implemented simply in biological machinery. For example, in experiments where reward is delivered with a temporal delay with respect to the stimulus offset (Figure 1), an additional assumption of a preexisting temporal basis is required \(^{8}\) . + +## Why is it implausible to assume a fixed temporal-basis in the brain? + +Consider the simple canonical example of Figure 1. In the time interval between the stimulus and the reward, the animal does not change its location, nor does its sensory environment change in any predictable manner. The only thing that changes consistently within this interval is time itself. Hence, in order to mark the states between stimulus and reward, the brain must have some internal representation of time, an internal clock which tracks the time since the start of the stimulus. Note however that before the conditioning starts, the animal has no way of knowing that this specific sensory stimulus has unique significance and therefore each specific stimulus must a priori be assigned its own specific temporal representation. + +This is the main hurdle of implementing TD in a biophysically realistic manner - figuring out how to represent the temporal basis upon which the association between cue and reward occurs (Figure 1a). Previous attempts were based on the assumption that there is a fixed cue- specific temporal basis, an assumption which has previously been termed "a useful fiction" \(^{8}\) . The specific implementations include the commonly utilized tapped delay lines \(^{5,14,15}\) (or the so- called complete serial compound), which are neurons triggered by the sensory cue, but that are active only at a specific delay, or alternatively, a set of cue specific neuronal responses which are also delayed but have a broader temporal support which increases with an increasing delay (the so called "microstimuli") \(^{8,10}\) (Figure 1c). + +For this class of temporal representations, the delay time between cue and reward is tiled by a chain of neurons, with each neuron representing a cue- specific time (sometimes referred to as a "microstate") (Figure 1c). In the simple case of the complete serial compound, the temporal basis is simply a set of neurons that have non- overlapping responses that start responding at the cue- relative times: \(t_{cue}\) , \(t_{cue} + dt\) , \(t_{cue} + 2dt\) , ..., \(t_{reward}\) . The learned value function (and in turn the RPE) assigned to a given cue at time \(t\) is then given by a learned weighted sum of the activations of these microstates at time \(t\) (Figure 1d). + +We argue that the conception of a fixed cue- dependent temporal basis makes biologically unrealistic assumptions. First, since one does not know a priori whether presentation of a cue will be followed by a reward, these models assume implicitly that every single environmental cue (or combination of environmental cues) must trigger its own sequence of neural microstates, prior to potential cue- reward pairing (Figure 1e). Further, since one does not know a priori when presentation of a cue may or may not be followed by a reward, these models also assume that microstate sequences are arbitrarily long to account for all possible (or a range of possible) cue- reward delays (Figure 1f). Finally, these microstates are assumed to be reliably reactivated upon subsequent presentations of the cue, e.g., a neuron that represents \(t_{cue} + 3dt\) must always represent \(t_{cue} + 3dt\) - across trials, sessions, days, etc. However, implementation of models that generate a chain- like structure of activity can be fragile to biologically relevant factors such as noise, neural + +<--- Page Split ---> + +dropout, and neural drift, all of which suggest that the assumption of reliability is problematic as well (Figure 1g). The totality of these observations imply that on the basis of first principles, it is hard to justify the idea of the fixed feature- specific temporal basis, a mechanism which is required for current supposedly biophysical implementations of TD learning. + +![](images/Figure_1.jpg) + +
Figure 1 – Structure and Assumptions of Temporal Bases for Temporal Difference
+ +Learning a) Diagram of a simple trace conditioning task. A conditioned stimulus (CS) such as a visual grating is paired, after a delay \(\Delta T\) , with an unconditioned stimulus (US) such as a water reward. b) According to the canonical view, neurons in VTA respond only to the US before training, and only to the CS after training. c) In order to represent the delay period, models generally assume neural "microstates" which span the time in between cue and reward. In the simplest case of the complete serial compound (left) the microstimuli do not overlap, and each one uniquely represents a different interval. In general, though (e.g.: microstimuli, right), these microstates can overlap with each other and decay over time. d) A weighted sum of these microstates determines the learned value function V(t). e) An agent does not know a priori which cue will subsequently be paired with reward. In turn, microstate TD models implicitly assume that all N unique cues or experiences in an environment each have their own independent chain of microstates before learning. f) Rewards delivered after the end of a particular cue-specific chain cannot be paired with the cue in question. The chosen length of the chain therefore determines the temporal window of possible associations. g) Microstate chains are assumed to be reliable and robust, but realistic levels of neural noise, drift, and variability can interrupt their propagation, thereby disrupting their ability to associate cue and reward. + +Although a fixed set of basis- functions for every possible stimulus is untenable, one could assume that it is possible to replace this assumption with a single set of fixed, universal basis + +<--- Page Split ---> + +functions. An example of mechanism that can generate such general basis functions is a fixed recurrent neural network (RNN). Instead of the firing of an individual neuron representing a particular time, here the entire network state can be thought of as a representation of a cue specific time. This setup is illustrated in Figure 2a. + +To understand the consequences of this setup, we assume a simple environment in which one specific stimulus (denoted as stimulus C) is always followed 1000ms later by a reward; this stimulus is the CS. However, it is reasonable to assume for the natural world that this stimulus exists among many other stimuli that do not predict a reward. For simplicity we consider 3 stimuli, A, B and C, which can appear at any possible order, as shown in Figure 2b, but in which stimulus C always predicts a reward with a delay of 1000ms. + +![](images/Figure_2.jpg) + +
Figure 2– A fixed RNN as a basis function generator. a) Schematic of a fixed recurrent neural network as a temporal basis. The network receives external inputs and generates states s(t), which act as a basis for learning the value function V(t). Compare to figure 1c. b) Schematic of the task protocol. Every presentation of C is followed by a reward at a fixed delay of 1000ms. However, any combination or sequence of irrelevant stimuli may precede the conditioned stimulus C (they might also come after the CS, e.g. A,C,B). c) Network activity, plotted along its first two principal components, for a given initial state s0 and a sequence of presented stimuli A-B-C (red letter is displayed at the time of a given stimulus' presentation). d) Same as c) but for input sequence B-A-C. e) Overlay of the A-B-C ad B-A-C network trajectories, starting from the state at the time of the presentation of C (state sc). The trajectory of network activity differs in these two cases, so the RNN state does not provide a consistent temporal basis that tracks the time since the presentation of stimulus C.
+ +We simulated the responses of such a fixed RNN to different stimulus combinations (see Methods). The complex RNN activity can be viewed as a projection to a subspace spanned by the first two principal components of the data. In Figure 2c,d we show a projection of the RNN response for two different sequences, A- B- C and B- A- C respectively, aligned to the time + +<--- Page Split ---> + +of reward. In Figure 2e we show the two trajectories side by side in the same subspace, starting with the presentation of stimulus C. + +What these results show is that every time stimulus C appears, it generates a different temporal response in the RNN depending on the preceding stimuli. These temporal patterns can also be changed by a subsequent stimulus that may appear between the CS and US. These results mean that such a fixed RNN cannot serve as a universal basis function because its responses are not repeatable. + +There are potential workarounds, such as to force the network states representing the time since stimulus C to be the same across trials. This is equivalent to learning the weights of the network such that all possible "distractor" cues pass through the network's null space. This means that the stimulus resets the network and erases its memory, but that other stimuli have no effect on the network. Generally, one would have to train the RNN to reproduce a given dynamical pattern representing C- >reward, while also being invariant to noise or task- irrelevant dimensions, the latter of which can be arbitrarily high and would have to be sampled over during training. + +However, this approach requires a priori knowledge that C is the conditioned stimuli (since C- > reward is the dynamical pattern we want to preserve) and that the other stimuli are nuisance stimuli. This leaves us with quite a conundrum. In the prospective view of temporal associations assumed by TD, to learn that C is associated with reward, we require a steady and repeatable labeled temporal basis (i.e. the network tracks the time since stimulus C). However, to train an RNN to robustly produce this basis, we need to have previously learned that C is associated with reward, and that the other stimuli are not. As such, these modifications to the RNN, while mathematically convenient, are based on unreasonable assumptions. + +## Models of TD learning with a fixed temporal-basis are inconsistent with data + +Apart from being based on unrealistic assumptions, models of TD learning also make predictions that are inconsistent with experimental data. In recent years, several experiments presented evidence of neurons with temporal response profiles that resemble temporal basis functions \(^{16 - 21}\) , as depicted schematically in Figure 3a. While there is indeed evidence of sequential activity in the brain spanning the delay between cues and rewards (such as in the striatum and hippocampus), these sequences are generally observed after association learning between a stimulus and a delayed reward \(^{16,19,20}\) . Some of these experiments have further shown that if the interval between stimulus and reward is extended, the response profiles either remap \(^{19}\) , or stretch to fill out the new extended interval \(^{16}\) , as depicted in Figure 3a. The fact that these sequences are observed after training and that the temporal response profiles are modified when the interval is changed supports the notion of plastic stimulus specific basis functions, rather than of a fixed set of basis function for each possible stimulus. Mechanistically, these results suggest that the naive network might generate generic temporal response profile to novel stimuli before learning, resulting from the networks initial connectivity. + +In the canonical version of TD learning (TD(0)), RPE neurons exhibit a bump of activity that moves back in time from the US to the CS during the learning process (Figure 3b- left). Subsequent versions of TD learning, called TD(λ), which speed up learning by the use of a memory trace, have a much smaller, or no noticeable moving bump (Figure 3b, center and right), depending on the length of the memory trace, denoted by λ. Most published experiments have not shown VTA neuron responses during the learning process. In one prominent example by Pan et al. in which VTA neurons are observed over the learning process \(^{22}\) , (depicted schematically in Figure 2c) no moving bump is observed, prompting the authors to deduce that such memory traces exist. In a more recent paper by Amo et al. a moving bump is reported \(^{23}\) . In contrast, in another recently published paper no moving bump + +<--- Page Split ---> + +is observed \(^{24}\) . Taken together, these different results suggest that at least in some cases a moving bump is not observed. However, since a moving bump is not predicted in TD( \(\lambda\) ) for sufficiently large \(\lambda\) , these results do not invalidate the TD framework in general, but rather suggest that in some cases at least the TD(0) variant is inconsistent with the data \(^{22}\) . + +While the moving bump prediction is parameter dependent, another prediction common to all TD variants is that the integrated RPE, obtained by summing response magnitudes over the whole trial duration, does not exceed the initial US response on the first trial. This prediction is robust because the normative basis of TD is to evaluate expected reward or discounted expected reward. In versions of TD where non- discounted reward is evaluated ( \(\gamma = 1\) ) the integral of RPE activity should remain constant throughout learning. Commonly TD estimates discounted reward ( \(\gamma < 1\) ), where the discount means that rewards that come with a small delay are worth more than rewards that arrive with a large delay. With discounted rewards the integral of RPE activity will decrease with learning and become smaller than the initial US response. In contrast, we reanalyzed data from a recent experiment (Amo et al. 2022) \(^{23,25}\) and found that the integrated response can transiently increase over the course of learning (Figure 3d). + +An additional prediction of TD learning which holds across many variants is that when learning converges, and if a reward is always delivered (100% reward delivery schedule), the response of RPE neurons at the time of reward is zero. Even in the case where a small number of rewards are omitted (e.g. 10%), TD predicts that the response of RPE neurons at the time of reward is very small, much smaller than at the time of the stimulus. This seems to be indeed the case for several example neurons shown in the original experiments of dopaminergic VTA neurons \(^{5}\) . However, additional data obtained more recently indicates this might not always be the case and that significant response to reward persists throughout training \(^{22,23,26}\) . This discrepancy between TD and the neural data is observed both for experiments in which responses throughout learning are presented \(^{22,23}\) as well as in experiments that only show results after training \(^{26}\) . + +In experimental approaches affording large ensembles of DA neurons to be simultaneously recorded, a diversity of responses has been reported. Some DA neurons are observed to become fully unresponsive at time of reward, while others exhibit a robust response at time of reward that is no weaker than the initial US response of these cells. This is clearly exhibited by one class of dopaminergic cells (type I) that Cohen et. al \(^{26}\) recorded in VTA. This diversity implies that TD is inconsistent with the results of some of the recorded neurons, but it is possible that TD does apply in some sense to the whole population. One complicating factor is that in most experiments we have no way of ascertaining that learning has reached its final state. + +The original conception of TD is clean, elegant, and based on a simple normative foundation of estimating expected rewards. Over the years, various experimental results that do not fully conform with the predictions of TD have been interpreted as consistent with theory by making ad- hoc modifications \(^{27,28}\) . Such modifications might include an assumption of different variants of the learning rule for each neuron, such that each dopaminergic neuron no longer represents RPE \(^{27}\) , or an assumption of additional inputs such that even when the expectation of reward is learned and fully expected dopaminergic neurons still respond at the time of reward \(^{28}\) . + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3- Certain features of experimental results run counter to predictions of TD. a) Putative temporal basis functions observed in experiments \(^{16,19}\) after training are plastic, shown here schematically. If, after training, the interval between CS and US is scaled, the basis-functions also change. Recordings in striatum \(^{16}\) show these basis-functions scale with the modified interval (top), while in recordings from hippocampus \(^{19}\) (bottom), they are observed to redistribute to fill up the new interval. b) According to the \(TD(0)\) theory, RPE neuron activity during learning exhibits a backward moving bump (left). For \(TD(\lambda)\) the bump no longer appears (right). c) A schematic depiction of experiments where there is no backward shifting bump \(^{22,24}\) . d) The integral of DA neuron activity according to TD theory (left) should be constant over training (for \(\gamma = 1\) ) or decreasing monotonically for \((\gamma < 1)\) . We reanalyzed existing experimental data (right) and found that the integral can transiently significantly increase. The inset shows the mean and standard errors for early (1-2), intermediate (3-4) and late (8-10) days. (Data from Amo et al 2022 \(^{23,25}\) , see Supplemental Figure 1).
+ +The various modifications that are added to account for new data, and which diverge from a straightforward implementation of TD, raise the subtle question: when does TD stop being TD? We propose that changes to TD in which its normative basis of maximizing future - discounted reward no longer holds are no longer part of the TD framework. The types of modifications added to theory to account for the experimental data are quite common in science in general. Scientific theories that no longer account for the data are often repeatedly modified before there is eventually a paradigm shift, and they are reluctantly abandoned \(^{29}\) . + +<--- Page Split ---> + +Towards this end, a recent paper has also shown that dopamine release, as recorded with photometry, seems to be inconsistent with RPE24. This paper has shown many experimental results that are at odds with those expected by a RPE, and specifically, these experiments show that dopamine release at least partially represents retrospective probability of stimulus given reward. Other work has suggested that dopamine signaling is more consistent with direct learning of behavioral policy than a value- based RPE30. + +![](images/Figure_4.jpg) + +
Figure 4- Potential Architectures for a Flexible Temporal Basis. Three example networks which could implement a FLEX theory. Top, network architecture. Middle, activity of one example neuron in the network. Bottom, network activity before and after training. Each network initially only has transient responses to stimuli, modifying plastic connections (in blue) during association to develop a specific temporal basis to reward-predictive stimuli. a) Synfire chains could support a FLEX model, if the chain could recruit more members during learning, exclusive to reward-predictive stimuli. b) A population of neurons with homogenous recurrent connections have a characteristic decay time that is related to the strength of the weights. The cue-relative time can then be read out by the mean level of activity in the network. c) A population of neurons with heterogenous and large recurrent connections (liquid state machine) can represent cue-relative time by treating the activity vector at time t as the "microstate" representing time t (as opposed to b) where only mean activity is used).
+ +## The FLEX theory of reinforcement learning: A theoretical framework based on a plastic temporal basis. + +The FLEX theory assumes that there is a plastic (as opposed to fixed) temporal basis that evolves alongside the changing response of reward dependent neurons (such as DA neurons in VTA). The theory in general is agnostic about the functional form of the temporal basis, and several possible examples are shown in Figure 4 (top, schematic; middle, + +<--- Page Split ---> + +characteristic single unit activity; bottom, population activity before and after learning). Synfire chains (Figure 4a), homogenous recurrent networks (Figure 4b), and heterogeneous recurrent networks (Figure 4c) could all plausibly support the temporal basis. Before learning, these basis- functions do not exist, though some neurons do respond transiently to the CS. Over learning, such basis- functions develop (Figure 4, bottom) in response to the rewarded stimulus, and not to unrewarded stimuli. In FLEX, we do not need a separate, predeveloped basis for every possible stimulus that spans an arbitrary amount of time. Instead, basis functions only form in the process of learning, develop only to stimuli which are tied to reward, and only span the relevant temporal interval for the behavior. + +In the following, we demonstrate that such a framework can be implemented in a biophysically plausible model, and that such a model not only agrees with many existing experimental observations, but also can reconcile seemingly incongruent results pertaining to sequential conditioning. The aim of this model is to show that the FLEX theoretical framework is possible and plausible given the available data, not to claim that this implementation is a perfectly validated model of reinforcement learning in the brain. Previous models concerning hippocampus and prefrontal cortex have also considered cue memories with adaptive durations, but not explicitly in the context of challenging the fundamental idea of a fixed temporal basis31,32. + +## A biophysically plausible implementation of FLEX, proof-of-concept + +Here we present a biophysically plausible proof- of- concept model that implements FLEX. This model is motivated by previous experimental results17,18 and previous theoretical work in our lab33- 37. The network's full architecture (visualized in Figure 5a) consists of two separate modules, a basis function module, and a reward module, here mapped onto distinct brain areas. We treat the reward module as an analogue of the VTA and the basis- function module akin to a cortical region such as the mPFC or OFC (although other cortical or subcortical regions, notably striatum16,38 might support temporal basis functions). All cells in these regions are modeled as spiking integrate and fire neurons (see Methods). + +We assume that within our basis function module are sub- populations of neurons tuned to certain external inputs, visualized in Figure 5a as set of discrete "columns", each responding to a specific stimulus. Within each column there are both excitatory and inhibitory cells, with a connectivity structure that includes both plastic (dashed lines, Figure 5a) and fixed synaptic connections (solid lines, Figure 5a). The VTA is composed of dopaminergic (DA) and inhibitory GABAergic cells. The VTA neurons have a background firing rate of \(\sim 5\) Hz, and the DA neurons have preexisting inputs from "intrinsically rewarding" stimuli (such as a water reward). The plastic and fixed connections between the modules and from both the CS and US to these modules are also depicted in Figure 5a. + +The model's structure is motivated by observations of distinct classes of temporally- sensitive cell responses that have evolve during trace conditioning experiments in medial prefrontal cortex (mPFC) orbitofrontal cortex (OFC) and primary visual cortex (V1)17,18,36,39,40. The architecture described above allows us to incorporate these observed cell classes into our basis- function module (Figure 5b). The first class of neurons ("Timers") are feature- specific and learn to maintain persistently elevated activity that spans the delay period between cue and reward, eventually decaying at the time of reward (real or expected). The second class, the "Messengers", have an activity profile that peaks at the time of real or expected reward. This set of cells form what has been coined a "Core Neural Architecture" (CNA)35, a potentially canonical package of neural temporal representations. A slew of previous studies have shown these cell classes within the CNA to be a robust phenomenon experimentally18,36,39- 43, and computational work has demonstrated that the CNA can be used to learn and recall single temporal intervals35, Markovian and non- Markovian sequences34,44. For simplicity, our model treats connections between populations within a single CNA as fixed + +<--- Page Split ---> + +(previous work has shown that such a construction is robust to perturbation of these weights34,35). + +![](images/Figure_5.jpg) + +
Figure 5 – Biophysically Inspired Architecture Allows for Flexible Encoding of Time a) Diagram of model architecture. CNAs (visualized here as columns) located in PFC are selective to certain sensory stimuli (indicated here by color atop the column) via fixed excitatory inputs (CS). VTA DA neurons receive fixed input from naturally positive valence stimuli, such as food or water reward (US). DA neuron firing releases dopamine, which acts as a learning signal for both PFC and VTA. Solid lines indicate fixed connections, while dotted lines indicate learned connections. b,c) Schematic representation of data adapted from Liu et al. 201542. b) Timers learn to characteristically decay at the time of cue-predicted reward. c) Messengers learn to have a firing peak at the time of cue-predicted reward.
+ +Learning in the model is dictated by the interaction of eligibility traces and dopaminergic reinforcement. We use a previously established two- trace learning rule34,36,37,45 (TTL), which assumes two Hebbian activated eligibility traces, one associated with LTP and one associated with LTD (see Methods). We use this rule because it solves the temporal credit assignment problem inherent in trace conditioning, reaches stable fixed points, and since such traces have been experimentally observed in trace conditioning tasks36. In theory, other methods capable of solving the temporal credit assignment problem (such as a rule with a single eligibility trace33) could also be used to facilitate learning in FLEX, but owing to its functionality and experimental support, we choose to utilize TTL for this work. See the Methods section for details of the implementation of TTL used here. + +Now, we will use this implementation of FLEX to simulate several experimental paradigms, showing that it can account for reported results. Importantly, some of the predictions of the model are categorically different than those produced by TD, which allows us to distinguish between the two theories based on experimental evidence. + +<--- Page Split ---> + +## CS-evoked and US-evoked Dopamine Responses Evolve on Different Timescales + +First, we test FLEX on a basic trace conditioning task, where a single conditioned stimulus is presented, followed by an unconditioned stimulus at a fixed delay of one second (Figure 6a). The evolution of FLEX over training is mediated by reinforcement learning (via TTL) in three sets of weights: Timer \(\rightarrow\) Timer, Messenger \(\rightarrow\) VTA GABA neurons, and CS \(\rightarrow\) VTA DA neurons. These learned connections encode the feature- specific cue- reward delay, the temporally specific suppression of US- evoked dopamine, and the emergence of CS- evoked dopamine, respectively. + +Upon presentation of the cue, cue neurons (CS) and feature- specific Timers are excited, producing Hebbian- activated eligibility traces at their CS- \(>\) DA and T- \(>\) T synapses, respectively. When the reward is subsequently presented one second later, the excess dopamine it triggers acts as a reinforcement signal for these eligibility traces (which we model as a function of the DA neuron firing rate, D(t), see Methods) causing both the cue neurons' feed- forward connections and the Timers' recurrent connections to increase (Figure 6b). + +Over repeated trials of this cue- reward pairing, the Timers' recurrent connections continue to increase until they approach their fixed points, which corresponds to the Timers' persistent firing duration increasing until it spans the delay between CS and US (Supplemental Figure 2 and Methods). These mature Timers then provide a feature- specific representation of the expected cue- reward delay. + +The increase of feed- forward connections from the CS to the DA neurons (6c) causes the model to develop a CS- evoked dopamine response. Again, this feed- forward learning uses dopamine release at \(t_{US}\) as the reinforcement signal to convert the Hebbian activated CS \(\rightarrow\) DA eligibility traces into synaptic changes (Supplemental Figure 3). The emergence of excess dopamine at the time of the CS \((t_{CS})\) owing to these potentiated connections also acts to maintain them at a non- zero fixed point, so CS- evoked dopamine persists long after US- evoked dopamine has been suppressed to baseline (see Methods). + +As the Timer population modifies its timing to span the delay period, the Messengers are "dragged along", since, owing to the dynamics of the Messengers' inputs (T and Inh), the Messengers themselves selectively fire at the end of the Timers' firing envelope. Eventually, the Messengers overlap with tonic background activity of VTA GABAergic neurons at the time of the US \((t_{US})\) (Figure 6b). When combined with the dopamine release at \(t_{US}\) , this overlap triggers Hebbian learning at the Messenger \(\rightarrow\) VTA GABA synapses (see Figure 6c, Methods), which indirectly suppresses the DA neurons. Because of the temporal specificity of the Messengers, this learned inhibition of the DA neurons (through excitation of the VTA GABAergic neurons) is effectively restricted to a short time window around the US and acts to suppress DA neural activity at \(t_{US}\) back towards baseline. + +As a result of these processes, our model recaptures the traditional picture of DA neuron activity before and after learning a trace conditioning task (Figure 6b). While the classical single neuron results of Schultz and others suggested that DA neurons are almost completely lacking excess firing at the time of expected reward5, more recent calcium imaging studies have revealed that a complete suppression of the US response is not universal. Rather, many optogenetically identified dopamine neurons maintain a response to the US and show varying development of a response to the CS22,23,26. This diversity is also exhibited in our implementation of FLEX (Figure 6d) due to the connectivity structure which is based on sparse random projections from the CS to the VTA and from the US to VTA. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6- CS-evoked and US-evoked Model Dopamine Responses Evolve on Different Timescales. The model is trained for 40 trials while being presented with a CS at 100ms and a reward at 1100ms. a) Mean firing rates for the CNA (see inset for colors), for three different stages of learning. b) Mean firing rate over all DA neurons taken at the same three stages of learning. Firing above or below thresholds (dotted lines) evokes positive or negative D(t) in PFC. c) Evolution of mean synaptic weights over the course of learning. Top, middle, and bottom, mean strength of \(\mathrm{T}\rightarrow \mathrm{T}\) , CS \(\rightarrow\) DA, and M \(\rightarrow\) GABA synapses, respectively. d) Area under receiver operating characteristic (auROC, see Methods) for all VTA neurons in our model for 15 trials before (left, US only), 15 trials during (middle, CS+US), and 15 trials after (right, CS+US) learning. Values above and below 0.5 indicate firing rates above and below the baseline distribution.
+ +During trace conditioning in FLEX, the inhibition of the US- evoked dopamine response (via M \(\rightarrow\) GABA learning) occurs only after the Timers have learned the delay period (since M and GABA firing must overlap to trigger learning), giving the potentiation of the CS response time to occur first. At intermediate learning stages (e.g. trial 5, Figure 6a,b), the CS- evoked dopamine response (or equivalently, the CS \(\rightarrow\) DA weights) already exhibits significant potentiation while the US- evoked dopamine response (or equivalently, inverse of the M \(\rightarrow\) US weights) has only been slightly depressed. While this phenomenon has been occasionally observed in certain experimental paradigms \(^{22,46 - 48}\) , it has not been widely commented on – + +<--- Page Split ---> + +in FLEX, this is a fundamental property of the dynamics of learning (in particular, very early learning). + +If an expected reward is omitted in FLEX, the resulting DA neuron firing will be inhibited at that time, demonstrating the characteristic dopamine "dip" seen in experiments (Supplemental Figure 4) \(^{5}\) . This phenomenon occurs in our model because the previous balance between excitation and inhibition (from the US and GABA neurons, respectively) is disrupted when the US is not presented. The remaining input at \(t_{US}\) is therefore largely inhibitory, resulting in a transient drop in firing rates. If the CS is consistently presented without being paired with the US, the association between cue and reward is unlearned, since the consistent negative D(t) at the time of the US causes depression of CS \(\rightarrow\) DA weights (Supplemental Figure 4). + +![](images/Figure_7.jpg) + +
Figure 7 – FLEX Model Dynamics Diverges from those of TD Learning Dynamics of both TD(λ) and FLEX when trained with the same conditioning protocol as shown in Figure 6. a) Dopaminergic release D(t) for FLEX (left), and RPE for TD(λ) (right), over the course of training. b) Sum total (i.e., the sum of each row in a) is a point on the line of b)) dopaminergic reinforcement D(t) during a given trial of training in FLEX (black), and sum total RPE in three instances of TD(λ) with discounting parameters \(\gamma = 1\) , \(\gamma = .99\) , and \(\gamma = .95\) (red, orange, and yellow, respectively). Shaded areas indicate functionally different stages of learning in FLEX. The learning rate in our model is reduced in this example, as to make direct comparison with TD(λ). 100 trials of US-only presentation (before learning) are included for comparison with subsequent stages.
+ +<--- Page Split ---> + +## Dynamics of FLEX Diverge from those of TD During Conditioning + +FLEX's property that the evolution of the CS responses can occur independently of (and before) depression of the US response underlies a much more fundamental and general departure of our model from TD based models. In our model, DA activity does not "travel" backwards over trials \(^{1,5}\) as in TD(0), nor is DA activity transferred from one time to the other in an equal and opposite manner as in TD( \(\lambda\) ) \(^{7,22}\) . This is because our DA activity is not a strict RPE signal. Instead, while the DA neural firing in FLEX may resemble RPE following successful learning, the DA neural firing and RPE are not equivalent, as evidenced during the learning period. + +To demonstrate this, we compare the putative DA responses in FLEX to the RPEs in TD( \(\lambda\) ) (Figure 7a), training on the previously described trace conditioning task (see Figure 6a). We set the parameters of our TD( \(\lambda\) ) model to match those in earlier work \(^{22}\) and approximate the cue and reward as Gaussians centered at \(t_{\mathrm{CS1}}\) and \(t_{\mathrm{CS2}}\) , respectively. In TD models, by definition, integral of the RPE over the course of the trial is always less than or equal to the original total RPE provided by the initial unexpected presentation of reward. In other words, the error in reward expectation cannot be larger than the initial reward. In both TD(0) and TD( \(\lambda\) ), this quantity of "integrated RPE" is conserved; for versions of TD with a temporal discounting factor \(\gamma\) (which acts such that a reward with value \(r_0\) presented \(n\) timesteps in the future is only worth \(r_0 \gamma^n\) where \(\gamma \leq 1\) ), this quantity decreases as learning progresses. + +In FLEX, by contrast, integrated dopaminergic release D(t) during a given trial can be greater than that evoked by the original unexpected US (see Figure 7b), and therefore during training the DA signal in FLEX diverges from a reward prediction error. This property has not been explicitly investigated, and most published experiments do not provide continuous data during the training phase. However, to test our prediction, we re- analyzed recently published data which does cover the training phase \(^{23}\) , and found that there is indeed a significant transient increase in the dopamine release during training (Figure 3d and Supplemental Figure 1). Another recent publication found that initial DA response to the US was uncorrelated with the final DA response to the CS \(^{30}\) , which also supports the idea that integrated dopamine release is not conserved. + +## FLEX Unifies Sequential Conditioning Results + +Standard trace conditioning experiments with multiple cues (CS1 \(\rightarrow\) CS2 \(\rightarrow\) US) have generally reported the so- called "serial transfer of activation" – that dopamine neurons learn to fire at the time of the earliest reward- predictive cue, "transferring" their initial activation at \(t_{\mathrm{US}}\) back to the time of the first conditioned stimulus \(^{49,50}\) . However, other results have shown that the DA neural responses at the times of the first and second conditioned stimuli ( \(t_{\mathrm{CS1}}\) and \(t_{\mathrm{CS2}}\) , respectively) evolve together, with both CS1 and CS2 predictive of the reward \(^{22,24}\) . Surprisingly, FLEX can reconcile these seemingly contradictory results. In Figure 8 we show simulations of sequential conditioning using FLEX. In early training we observe an emerging response to both CS1 and CS2, as well as to the US (Figure 8b, ii). Later on the response to the US is suppressed (Figure 8b, iii). During late training (Figure 8b, iv) the response to CS2 is suppressed and all activation is transferred to the earliest predictive stimulus, CS1. These evolving dynamics can be compared to the different experimental results. Early training is similar to the results of Pan et al. (2005) \(^{22}\) and Jeong et al. (2022) \(^{24}\) , as seen in Figure 8c, while the late training results are similar to the results of Schultz (1993) \(^{49}\) as seen in Figure 8d. + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8 – FLEX model Reconciles Differing Experimental Phenomena Observed during Sequential Conditioning Results from “sequential conditioning”, where sequential neutral stimuli CS1 and CS2 and are paired with delayed reward US. a) Visualization of protocol. In this example, the US is presented starting at 1500ms, with CS1 is presented starting at 100ms, and CS2 is presented starting at 800ms. b) Mean firing rates over all DA neurons, for four distinctive stages in learning – initialization(i), acquisition(ii), reward depression(iii), and serial transfer of activation(iv). c) Schematic illustration of experimental results from recorded dopamine neurons, labeled with the matching stage of learning in our model. c) DA neuron firing before (top), during (middle) and after (bottom) training, wherein two cues (0s and 4s) were followed by a single reward (6s). Adapted from Pan et al. 200522. d) DA neuron firing after training wherein two cues (instruction, trigger) were followed by a single reward. Adapted from Schultz et al. 199349.
+ +After training, in FLEX, both sequential cues still affect the dopamine release at the time of reward, as removal of either cue results in a partial recovery of the dopamine response at \(t_{\mathrm{US}}\) (Supplemental Figure 5). This is the case even in late training in FLEX, when there is a positive dopamine response only to first cue – our model predicts that removal of the second cue will result in a positive dopamine response at the time of expected reward. In contrast, the RPE hypothesis would posit that after extended training, the value function would eventually be maximized following the first cue, and therefore removal of the subsequent cue would not change dopamine release at \(t_{\mathrm{US}}\) . + +FLEX is also capable of replicating results of a different set of sequential learning paradigms (Supplemental Figure 6). In these protocols, the network is initially trained on a standard trace conditioning task with a single CS. Once the cue- reward association is learned completely, a second cue is inserted in between the initial cue and the reward, and learning is repeated. As in experiments, this intermediate cue on its own does not become reward predictive, a phenomenon called “blocking”51–53. However, if reward magnitude is increased + +<--- Page Split ---> + +or additional dopamine is introduced to the system, a response to the intermediate CS (CS2) emerges, a phenomenon termed "unblocking" \(^{54,55}\) . Each of these phenomena can be replicated in FLEX (Supplemental Figure 6). + +## Discussion + +TD has established itself as one of the most successful models of neural function to date, as its predictions regarding RPE have, to a large extent, matched experimental results. However, two key factors make it reasonable to consider alternatives to TD as a modeling framework for how midbrain dopamine neurons could learn RPE. First, attempts for biologically plausible implementations of TD have previously assumed that even before learning, each possible cue triggers a separate chain of neurons which tile an arbitrary period of time relative to the cue start. The a priori existence of such an immense set of fixed, arbitrarily long cue- dependent basis functions is both implausible and inconsistent with experimental evidence. Second, different conditioning paradigms have revealed dopamine dynamics that are incompatible with the predictions of models based on the TD framework \(^{22,24,56,57}\) . + +To overcome these problems, we suggest that the temporal basis itself is not fixed, but instead plastic, and is learned only for specific cues that lead to reward. We call this theoretical framework FLEX. We also presented a biophysically plausible implementation of FLEX, and we have shown that it can generate flexible basis functions, and that it produces dopamine cell dynamics that are consistent with experimental results. Our implementation should be seen as a proof- of- concept model. It shows that FLEX can be implemented with biophysical components, and that such an implementation is consistent with much of the data. It does not show that the specific details of this implementation, (including the brain regions in which the temporal- basis is developed, the specific dynamics of the temporal basis functions, and the learning rules used) are those used by the brain. + +One of the appealing aspects of TD learning is that it arises from simple normative assumption that the brain needs to estimate future expected reward. Does FLEX have a similar normative basis? Indeed, in the final state of learning, the responses of DA neurons in FLEX resemble RPE neurons in TD, however in FLEX there is no analog for the value neurons assumed in TD. Unlike in TD, activity of FLEX DA neurons in response to the cue represent an association with future expected rewards, independent of a valuation. + +Instead, the goal of FLEX is to learn the association between cue and reward, develop the temporal basis functions that span the period between the two, and transfer DA signaling from the time of the reward to the time of the cue. These basis functions could then be used as a mechanistic foundation for brain activities, including the timing of actions. In essence, DA in FLEX acts to create internal models of associations and timings. Recent experimental evidence \(^{30}\) which suggests that DA correlates more with the direct learning of behavioral policies rather than value- encoded prediction errors is broadly consistent with this view of dopamine's function within the FLEX framework. + +Another recent publication \(^{24}\) has questioned the claim that DA neurons indeed represent RPE, instead hypothesizing that DA release is also dependent on retrospective probabilities \(^{24,58}\) . The design of most historical experiments cannot distinguish between these competing hypotheses. In this recent research project, a set of new experiments were designed specifically to test these competing hypotheses, and the results obtained are inconsistent with the common interpretation that DA neurons simply represent RPE \(^{24}\) . More generally, while the idea that the brain does and should estimate economic value seems intuitive, it has been recently questioned \(^{59}\) . This challenge to the prevailing normative view is motivated by behavioral experimental results which instead suggest that a heuristic process, which does not faithfully represent value, often guides decisions. + +Recent papers have questioned the common view of the response of DA neurons in the brain and their relation to value estimation \(^{9,24,30,59}\) . Here we survey additional problems + +<--- Page Split ---> + +with implementations of TD algorithms in neuronal machinery, and propose an alternative theoretical formulation, FLEX, along with a computational implementation of this theory. The fundamental difference from previous work is that FLEX postulates that the temporal basis- functions necessary for learning are themselves learned, and that neuromodulator activity in the brain is an instructive signal for learning these basis functions. Our computational implementation has predictions that are different than those of TD and are consistent with many experimental results. Further, we tested a unique prediction of FLEX (that integrated DA release across a trial can change over learning) by re- analyzing experimental data, showing that the data was consistent with FLEX but not the TD framework. + +## Methods + +## Network Dynamics + +The membrane dynamics for each neuron \(i\) are described by the following equations: + +## Equation 1 - Membrane Dynamics + +\[\mathrm{C}\frac{\mathrm{d}\nu_{\mathrm{i}}}{\mathrm{d}\mathrm{t}} = \mathrm{g}_{\mathrm{L}}(\mathrm{E}_{\mathrm{L}} - \nu_{\mathrm{i}}) + \mathrm{g}_{\mathrm{E},\mathrm{i}}(\mathrm{E}_{\mathrm{E}} - \nu_{\mathrm{i}}) + \mathrm{g}_{\mathrm{I},\mathrm{i}}(\mathrm{E}_{\mathrm{I}} - \nu_{\mathrm{i}}) + \sigma\] + +## Equation 2 - Synaptic Activation Dynamics + +\[\frac{\mathrm{d}s_{\mathrm{i}}}{\mathrm{d}t} = -\frac{s_{\mathrm{i}}}{\tau_{\mathrm{s}}} +\rho (1 - s_{\mathrm{i}})\sum_{\mathrm{k}}\delta (\mathrm{t} - \mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\] + +The membrane potential and synaptic activation of neuron \(i\) are notated \(\nu_{i}\) and \(s_{i}\) . \(g\) refers to the conductance, and \(E\) to the reversal potentials indicated by the appropriate subscript, where leak, excitatory, and inhibitory are indicated by subscripts \(L\) , \(E\) , and \(I\) , respectively. \(\sigma\) is a mean zero noise term. The neuron spikes upon it crossing its membrane threshold potential \(\nu_{th}\) , after which it enters a refractory period \(t_{ref}\) . The synaptic activation \(s_{\mathrm{i}}\) is updated by an amount \(\rho (1 - s_{\mathrm{i}})\) , at each time \((\mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\) the neuron spikes, and decays exponentially when there is no spike. + +The conductance \(g\) is the product of the incoming synaptic weights and their respective presynaptic neurons: + +## Equation 3 - Conductance + +\[\mathrm{g}_{\alpha ,\mathrm{i}} = \sum_{j}\mathrm{W}_{ij}^{\alpha}\mathrm{s}_{j}\] + +\(W_{ij}^{\alpha}\) are the connection strengths from neuron \(j\) to neuron \(i\) , where the superscript \(\alpha\) can either indicate \(E\) (excitatory) or \(I\) (inhibitory). A firing rate estimate for each neuron \(\mathrm{r}_{\mathrm{i}}\) is calculated as an exponential filter of the spikes, with a time constant \(\tau_{\mathrm{r}}\) . + +## Equation 4 - Firing Rate + +\[\tau_{\mathrm{r}}\frac{\mathrm{d}\mathrm{r}_{\mathrm{i}}}{\mathrm{d}t} = -\mathrm{r}_{\mathrm{i}} + \sum_{\mathrm{k}}\delta (\mathrm{t} - \mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\] + +## Fixed RNN + +For the network in Figure 2, the dynamics of the units \(\mathrm{u}_{\mathrm{i}}\) in the RNN are described by the equation below: + +## Equation 5 - RNN Dynamics + +\[\tau_{\mathrm{net}}\frac{\mathrm{d}\mathrm{u}_{\mathrm{i}}}{\mathrm{d}t} = -\mathrm{u}_{\mathrm{i}} + \sum_{k}\mathrm{W}_{\mathrm{ik}}\phi (\mathrm{u}_{\mathrm{k}})\] + +<--- Page Split ---> + +Where \(\mathbf{u}_{\mathrm{i}}\) are the firing rates of the units in the RNN. \(W_{\mathrm{ik}}\) are the recurrent weights of the RNN, each of which is drawn from a normal distribution \(\mathrm{N}(0,\frac{g}{\sqrt{K}})\) , where \(g\) is the "gain" of the network \(^{60}\) and \(\phi\) is a sigmoidal activation function. Each of the inputs given to the network (A, B, or C) is a unique, normally distributed projection of a 100ms step function. + +## Dopaminergic Two-Trace Learning (dTTL) + +Rather than using temporal difference learning, FLEX uses a previously established learning rule based on competitive eligibility traces, known as "two- trace learning" or TTL \(^{34,36,37}\) . However, we replace the general reinforcement signal \(\mathrm{R(t)}\) of previous implementations by a dopaminergic reinforcement \(\mathrm{D(t)}\) . We repeat here the description of \(\mathrm{D(t)}\) from the main text: + +## Equation 6 - Dopamine reinforcement + +\[\mathrm{D(t)} = \left\{ \begin{array}{ll}\mathrm{r_{DA}(t)} - (\mathrm{r_0} - \theta), & \mathrm{r_{DA}(t)}\leq \mathrm{r_0} - \theta \\ 0, & \mathrm{r_0} - \theta < \mathrm{r_{DA}(t)}< \mathrm{r_0} + \theta \\ \mathrm{r_{DA}(t)} - (\mathrm{r_0} + \theta), & \mathrm{r_{DA}(t)}\geq \mathrm{r_0} + \theta \end{array} \right.\] + +Note that the dopaminergic reinforcement can be both positive and negative, even though actual DA neuron firing (and the subsequent release of dopamine neurotransmitter) can itself only be positive. The bipolar nature of \(\mathrm{D(t)}\) implicitly assumes that background tonic levels of dopamine do not modify weights, and that changes in synaptic efficacies are a result of a departure (positive or negative) from this background level. The neutral region around \(\mathrm{r_0}\) provides robustness to small fluctuations in firing rates which are inherent in spiking networks. + +The eligibility traces, which are synapse specific, act as long- lasting markers of Hebbian activity. The two traces are separated into LTP- and LTD- associated varieties via distinct dynamics, which are described in the equations below. + +## Equation 7 - LTP Trace Dynamics + +\[\tau^{\mathrm{p}}\frac{\mathrm{d}\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}}{\mathrm{d}t} = -\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}} + \eta^{\mathrm{p}}\mathrm{H}_{\mathrm{ij}}(\mathrm{T}_{\mathrm{max}}^{\mathrm{p}} - \mathrm{T}_{\mathrm{ij}}^{\mathrm{p}})\] + +## Equation 8 - LTD Trace Dynamics + +\[\tau^{\mathrm{d}}\frac{\mathrm{d}\mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}}{\mathrm{d}t} = -\mathrm{T}_{\mathrm{ij}}^{\mathrm{d}} + \eta^{\mathrm{d}}\mathrm{H}_{\mathrm{ij}}(\mathrm{T}_{\mathrm{max}}^{\mathrm{d}} - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}})\] + +Here, \(T_{ij}^{\alpha}\) (where \(a\in (p,d))\) is the LTP \((p\) superscript) or LTD ( \(d\) superscript) eligibility trace located at the synapse between the j- th presynaptic cell and the i- th postsynaptic cell. The Hebbian activity, \(\mathrm{H_{ij}}\) , is a simple multiplication \(r_i\cdot r_j\) for application of this rule in VTA, where \(r_i\) and \(r_i\) are the time- averaged firing rates at the pre- and post- synaptic cells. Experimentally, the "Hebbian" terms \((\mathrm{H_{ij}})\) which impact LTP and LTD trace generation are complex \(^{36}\) , but in VTA we approximate with the simple multiplication \(r_i\cdot r_j\) . For synapses in PFC, we make the alteration that \(\mathrm{H_{ij}} = \frac{r_i\cdot r_j}{1 + \alpha D(t)}\) , acting to restrict PFC trace generation for large positive RPEs. The alteration of \(\mathrm{H_{ij}}\) by large positive RPEs is inspired by recent experimental work showing that large positive RPEs act as event boundaries and disrupt across- boundary (but not within- boundary) associations and timing \(^{61}\) . Functionally, this altered \(\mathrm{H_{ij}}\) biases Timers in PFC towards encoding a single delay period (cue to cue or cue to reward) and disrupts their ability to encode across- boundary delays. + +The LTP and LTD traces activate (via activation constant \(\eta^{\mathrm{a}}\) ), saturate (at a level \(\mathrm{T}_{\mathrm{max}}^{\mathrm{a}}\) ) and decay (with time constant \(\tau^{\mathrm{a}}\) ) at different rates. \(\mathrm{D(t)}\) binds to these eligibility traces, converting them into changes in synaptic weights. This conversion into synaptic changes is "competitive", being determined by the difference in the product: + +<--- Page Split ---> + +## Equation 9 - Weight Update + +\[\frac{\mathrm{d}W_{\mathrm{ij}}}{\mathrm{dt}} = \eta \mathrm{D(t)}\left(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\right)\] + +where \(\eta\) is the learning rate. + +The above eligibility trace learning rules, with the inclusion of dopamine, are referred to as dopaminergic "two- trace" learning or dTTL. This rule is pertinent not only because it can solve the temporal credit assignment problem, allowing the network to associate events distal in time, but also because dTTL is supported by recent experiments which have found eligibility traces for in multiple brain regions36,62- 65. Notably, in such experiments, the eligibility traces in prefrontal cortex were found to convert into positive synaptic changes via delayed application of dopamine36, which is the main assumption behind dTTL. + +For simplicity and to reduce computational time, in the simulations shown, M- - GABA connections are learned via a simple dopamine modulated Hebbian rule, \(\frac{\mathrm{d}W_{\mathrm{ij}}}{\mathrm{dt}} = \eta \mathrm{D(t)}\mathrm{r}_{\mathrm{i}}\mathrm{r}_{\mathrm{j}}\) . Since these connections are responsible for inhibiting the DA neurons at the time of reward, this learning rule imposes its own fixed point by suppressing \(\mathrm{D(t)}\) down to 0. For any appropriate selection of feed- forward learning parameters in dTTL (Equation 9), the fixed- point \(\mathrm{D(t)} = 0\) is reached well before the fixed point \(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) = \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\) . This is because, by construction, the function of M- - GABA learning is to suppress \(\mathrm{D(t)}\) down to zero. Therefore, the fixed point \(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) = \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\) needs to be placed (via choosing trace parameters) beyond the fixed- point \(\mathrm{D(t)} = 0\) . Functionally, then, both rules act to potentiate M- - GABA connections monotonically until \(\mathrm{D(t)} = 0\) . As a result, the dopamine modulated Hebbian rule is in practice equivalent to dTTL in this case. + +## Network Architecture + +Our network architecture consists of two regions, VTA and PFC, each of which consist of subpopulations of leaky- integrate- and- fire neurons. Both fixed and learned connections exist between certain populations to facilitate the functionality of our model. + +To model VTA, we include 100 dopaminergic and 100 GABAergic neurons, both of which receive tonic noisy input to establish baseline firing rates of \(\sim 5\mathrm{Hz}\) . Naturally appetitive stimuli, such as food or water, are assumed to have fixed connections to DA neurons via the gustatory system. Dopamine release is determined by DA neuron firing above or below a threshold \(\theta\) , and dopamine release acts as reinforcement for all learned connections in the model. + +Our model of PFC is comprised of different feature- specific 'columns'. Within each column there is a CNA microcircuit, with each subpopulation (Timers, Inhibitory, Messengers) consisting of 100 LIF neurons. Previous work has shown that these subpopulations can emerge from randomly distributed connections35, and further that a single mean field neuron can well approximate the activity of each of these subpopulations of spiking neurons66. + +The two- trace learning rule we utilize for our model is described in further detail in previous work34,36,37,45. However, we will attempt to clarify how it functions below. As a first approximation, the traces from the two- trace learning rule we utilize effectively act such that when they interact with dopamine above baseline, the learning rule will favor potentiation, and when they encounter dopamine below baseline, the learning rule will favor depression. Formally, the rule has fixed points which depend on both the dynamics of the traces and the dopamine release \(\mathrm{D(t)}\) : + +## Equation 10 - Fixed Point + +\[\int_{0}^{T_{trial}}dt\mathrm{D(t)}\left(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\right) = 0\] + +<--- Page Split ---> + +A trivial fixed point exists when \(\mathrm{D(t)} = 0\) for all t. Another simple fixed point exists in the limit that \(\mathrm{D(t)} = \delta (t_R - t)\) , where \(t_R\) is the time of reward, as Equation 10 then reduces to \(\mathrm{T_{ij}^p(t_R) = T_{ij}^d(t_R)}\) . In this case, the weights have reached their fixed point when the traces cross at the time of reward. In practice, the true fixed points of the model are a combination of these two factors (suppression of dopamine and crossing dynamics of the traces). In reality, \(\mathrm{D(t)}\) is not a delta function (and may have multiple peaks during the trial), so to truly calculate the fixed points, one must use Equation 10 as a whole. However, the delta approximation used above gives a functional intuition for the dynamics of learning in the model. + +Supplemental Figure 2 and Supplemental Figure 3 demonstrate examples of fixed points for both recurrent and feed- forward learning, respectively. Note that in these examples the two "bumps" of excess dopamine (CS- evoked and US- evoked) are the only instances of nonzero \(\mathrm{D(t)}\) . As such, we can take the integral in Equation 10 and split it into two parts: + +Equation 11 - Two- part Fixed Point + +\[t_{CS,ent} \qquad \int_{t_{CS,start}}^{t_{CS,end}}dt\mathrm{D(t)}\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right) + \int_{t_{US,start}}^{t_{US,end}}dt\mathrm{D(t)}\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right) = 0\] + +For recurrent learning, the dynamics evolve as follows. At the beginning, only the integral over \(\Delta t_{US}\) exists, as \(\mathrm{D(t)}\) is initially zero over \(\Delta t_{CS}\) (Trial 1 in Supplemental Figure 2). As a result, the learning rule evolves to approach the fixed point mediated by \(\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right)\) (Trial 20 in Supplemental Figure 2). After the recurrent weights have reached this fixed point and the Timer neurons encode the cue- reward delay (Trial 30 in Supplemental Figure 2), M→GABA learning acts to suppress \(\mathrm{D(t)}\) down to zero as well (Trial 40 in Supplemental Figure 2). Note again that we make the assumption that trace generation in PFC is inhibited during large positive RPEs. This acts to encourage the Timers to encode a single "duration" (whether cue- cue or cue- reward). In line with our assumption, experimental evidence has shown these large positive RPEs act as event boundaries and disrupt across- boundary (but not within- boundary) reports of timing61. + +For feed- forward learning, the weights initially evolve identically to the recurrent weights (Trial 1 in Supplemental Figure 3). Again, only the integral over \(\Delta t_{US}\) exists, so the feed- forward weights evolve according to \(\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right)\) . However, soon the potentiation of these feed- forward CS→DA weights themselves cause release of CS- evoked dopamine, and therefore we must consider both integrals to explain the learning dynamics (Trial 5 in Supplemental Figure 3). This stage of learning is harder to intuit, but an intermediate fixed point is reached when the positive \(\Delta W\) produced by the traces' overlap with US- evoked dopamine is equal and opposite to the negative \(\Delta W\) produced by the traces' overlap with CS- evoked dopamine (Trial 20 in Supplemental Figure 3). Finally, after US- evoked dopamine has been suppressed to baseline, the feed- forward weights reach a final fixed point where both positive and negative contributions to \(\Delta W\) over the course of the CS offset each other (Trial 50 in Supplemental Figure 3). + +The measure of "area under receiver operating characteristic" (auROC) is used throughout this paper, for the purpose of making direct comparison to a slew of calcium imaging results that use auROC as a measure of statistical significance. Following the methods of Cohen et al. (2012)26, time is tiled into 50ms bins. For a single neuron, within each 50ms bin, the distribution of spike counts for 25 trials of baseline spontaneous firing (no external stimuli) is compared to the distribution of spike counts during the same time bin for 25 trials of the learning phase in question. For example, in Figure 6d, left, the baseline distributions of spike counts are compared to the distributions of spike counts when US only is presented. ROC is calculated for each bin by sliding the criteria from zero to the max + +<--- Page Split ---> + +spike count within the bin, and then plotting P(active>criteria) versus P(baseline>criteria). The area under this curve is then a measure of discriminability between the two distributions, with an auROC of 1 demonstrating a maximally discriminable increase in spikes compared to baseline, 0 demonstrating a maximally discriminable decrease of spikes compared to baseline, and .5 demonstrating an inability to discriminate between the two distributions. + +Data from Amo et al. 2022 was used for Supplemental Figure \(1^{23,25}\) . The data from 7 animals (437- 440, 444- 446) is shown here. This data is already z- scored as described by Amo et al. On every training day only the first 40 trials are used, unless there was a smaller number of rewarded trials, in which case that number was used. For every animal per every day, the integral from the time of the CS to the end of the trial is calculated and averaged over all trials in that day. An example of the integral for one animal and one day is shown in Supplemental Figure 1a. These are the data points in Supplemental Figure 1b, with each animal represented by a different color. The blue line is the average over animals. The bar graphs in Supplemental Figure 1c represent the average over all animals in early (day 1- 2), intermediate (day 3- 4) and late (day 8- 10) periods. Wilcoxon rank sum tests find that intermediate is significantly higher than early (p=0.004) and that late is significantly higher than early (p=0.006). Late is not significantly lower than intermediate. We also find that late is significantly higher than early if we take the average per animal over the early and intermediate days (p=0.02). A MATLAB code that carries out this analysis is posted on https://github.com/ianconehed/FLEX and will also be posted on modelDB together with the code for FLEX. + +## Code Availability + +Code for both simulations and analysis is posted on: https://github.com/ianconehed/FLEX and will be available on ModelDB upon the publication of the paper. + +## Acknowledgements + +This work was supported by Biotechnology and Biological Sciences Research Council (BBSRC) Grants BB/N013956/1 and BB/N019008/1, Wellcome Trust 200790/Z/16/Z, Simons Foundation Engineering and Physical Sciences Research Council (EPSRC) EP/R035806/1, National Institute of Biomedical Imaging and Bioengineering 1R01EB022891- 01 and Office of Naval Research N00014- 16- R- BA01. + +## Competing Interests + +The authors declare no competing interests + +<--- Page Split ---> + +## References + +1. Sutton, R. S. & Barto, A. G. Reinforcement Learning, second edition: An Introduction. (MIT Press, 2018). + +2. Glickman, S. E. & Schiff, B. B. A biological theory of reinforcement. Psychol. Rev. 74, 81-109 (1967). + +3. Lee, D., Seo, H. & Jung, M. W. Neural Basis of Reinforcement Learning and Decision Making. Annu. Rev. Neurosci. 35, 287-308 (2012). + +4. A Model of How the Basal Ganglia Generate and Use Neural Signals That Predict Reinforcement. in Models of Information Processing in the Basal Ganglia (eds. Houk, J. C., Davis, J. L. & Beiser, D. G.) (The MIT Press, 1994). doi:10.7551/mitpress/4708.003.0020. + +5. Schultz, W., Dayan, P. & Montague, P. R. A Neural Substrate of Prediction and Reward. Science 275, 1593-1599 (1997). + +6. Montague, P. R., Dayan, P. & Sejnowski, T. J. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. J. Neurosci. 16, 1936-1947 (1996). + +7. Sutton, R. S. Learning to predict by the methods of temporal differences. Mach. Learn. 3, 9-44 (1988). + +8. Ludvig, E. A., Sutton, R. S. & Kehoe, E. J. Evaluating the TD model of classical conditioning. Learn. Behav. 40, 305-319 (2012). + +9. Namboodiri, V. M. K. How do real animals account for the passage of time during associative learning? Behav. Neurosci. 136, 383-391 (2022). + +10. Ludvig, E. A., Sutton, R. S. & Kehoe, E. J. Stimulus Representation and the Timing of Reward-Prediction Errors in Models of the Dopamine System. Neural Comput. 20, 3034-3054 (2008). + +11. Tesauro, G. TD-Gammon, a self-teaching backgammon program, achieves master-level play. Neural Comput. 6, 215-219 (1994). + +12. Arulkumaran, K., Deisenroth, M. P., Brundage, M. & Bharath, A. A. Deep Reinforcement Learning: A Brief Survey. IEEE Signal Process. Mag. 34, 26-38 (2017). + +13. Foster, D. J., Morris, R. G. & Dayan, P. A model of hippocampal dependent navigation, using the temporal difference learning rule. Hippocampus 10, 1-16 (2000). + +14. Montague, P., Dayan, P. & Sejnowski, T. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. J. Neurosci. 16, 1936-1947 (1996). + +<--- Page Split ---> + +15. Schultz, W. Behavioral Theories and the Neurophysiology of Reward. Annu. Rev. Psychol. 57, 87-115 (2006). + +16. Mello, G. B. M., Soares, S. & Paton, J. J. A Scalable Population Code for Time in the Striatum. Curr. Biol. 25, 1113-1122 (2015). + +17. Namboodiri, V. M. K. et al. Single-cell activity tracking reveals that orbitofrontal neurons acquire and maintain a long-term memory to guide behavioral adaptation. Nat. Neurosci. 22, 1110-1121 (2019). + +18. Kim, E., Bari, B. A. & Cohen, J. Y. Subthreshold basis for reward-predictive persistent activity in mouse prefrontal cortex. Cell Rep. 35, 109082 (2021). + +19. Taxidis, J. et al. Differential Emergence and Stability of Sensory and Temporal Representations in Context-Specific Hippocampal Sequences. Neuron 108, 984-998.e9 (2020). + +20. MacDonald, C. J., Lepage, K. Q., Eden, U. T. & Eichenbaum, H. Hippocampal "Time Cells" Bridge the Gap in Memory for Discontiguous Events. Neuron 71, 737-749 (2011). + +21. Parker, N. F. et al. Choice-selective sequences dominate in cortical relative to thalamic inputs to NAc to support reinforcement learning. Cell Rep. 39, 110756 (2022). + +22. Pan, W.-X. Dopamine Cells Respond to Predicted Events during Classical Conditioning: Evidence for Eligibility Traces in the Reward-Learning Network. J. Neurosci. 25, 6235-6242 (2005). + +23. Amo, R. et al. A gradual temporal shift of dopamine responses mirrors the progression of temporal difference error in machine learning. Nat. Neurosci. 25, 1082-1092 (2022). + +24. Jeong, H. et al. Mesolimbic dopamine release conveys causal associations. Science 0, eabq6740 (2022). + +24. Amo, R., Matias, S., Uchida, N. & Watabe-Uchida, M. Population and single dopamine neuron activity during classical conditioning. Dryad, Dataset (2022) doi:10.5061/DRYAD.HHMGQNKJW. + +26. Cohen, J. Y., Haesler, S., Vong, L., Lowell, B. B. & Uchida, N. Neuron-type-specific signals for reward and punishment in the ventral tegmental area. nature 482, 85-88 (2012). + +27. Dabney, W. et al. A distributional code for value in dopamine-based reinforcement learning. Nature 577, 671-675 (2020). + +28. Mikhael, J. G., Kim, H. R., Uchida, N. & Gershman, S. J. The role of state uncertainty in the dynamics of dopamine. Curr. Biol. 32, 1077-1087 (2022). + +29. Kuhn Thomas. The structure of scientific revolutions. vol. 111 (University of Chicago Press:, 1970). + +<--- Page Split ---> + +30. Coddington, L. T., Lindo, S. E. & Dudman, J. T. Mesolimbic dopamine adapts the rate of learning from action. Nature 614, 294-302 (2023). + +31. Wang, J. X. et al. Prefrontal cortex as a meta-reinforcement learning system. Nat. Neurosci. 21, 860-868 (2018). + +32. O'Reilly, R. C., Frank, M. J., Hazy, T. E. & Watz, B. PVLV: The Primary Value and Learned Value Pavlovian Learning Algorithm. Behav. Neurosci. 121, 31-49 (2007). + +33. Gavornik, J. P., Shuler, M. G. H., Loewenstein, Y., Bear, M. F. & Shouval, H. Z. Learning reward timing in cortex through reward dependent expression of synaptic plasticity. Proc. Natl. Acad. Sci. 106, 6826-6831 (2009). + +34. Cone, I. & Shouval, H. Z. Learning precise spatiotemporal sequences via biophysically realistic learning rules in a modular, spiking network. eLife 10, e63751 (2021). + +35. Huertas, M. A., Hussain Shuler, M. G. & Shouval, H. Z. A Simple Network Architecture Accounts for Diverse Reward Time Responses in Primary Visual Cortex. J. Neurosci. 35, 12659-12672 (2015). + +36. He, K. et al. Distinct Eligibility Traces for LTP and LTD in Cortical Synapses. Neuron 88, 528-538 (2015). + +37. Huertas, M. A., Schwettmann, S. E. & Shouval, H. Z. The Role of Multiple Neuromodulators in Reinforcement Learning That Is Based on Competition between Eligibility Traces. Front. Synaptic Neurosci. 8, (2016). + +38. Aosaki, T. et al. Responses of tonically active neurons in the primate's striatum undergo systematic changes during behavioral sensorimotor conditioning. J. Neurosci. 14, 3969-3984 (1994). + +39. Chubykin, A. A., Roach, E. B., Bear, M. F. & Shuler, M. G. H. A Cholinergic Mechanism for Reward Timing within Primary Visual Cortex. Neuron 77, 723-735 (2013). + +40. Shuler, M. G. Reward Timing in the Primary Visual Cortex. Science 311, 1606-1609 (2006). + +41. Namboodiri, V. M. K., Huertas, M. A., Monk, K. J., Shouval, H. Z. & Hussain Shuler, M. G. Visually Cued Action Timing in the Primary Visual Cortex. Neuron 86, 319-330 (2015). + +42. Liu, C.-H., Coleman, J. E., Davoudi, H., Zhang, K. & Hussain Shuler, M. G. Selective Activation of a Putative Reinforcement Signal Conditions Cued Interval Timing in Primary Visual Cortex. Curr. Biol. 25, 1551-1561 (2015). + +<--- Page Split ---> + +43. Monk, K. J., Allard, S. & Hussain Shuler, M. G. Reward Timing and Its Expression by Inhibitory Interneurons in the Mouse Primary Visual Cortex. Cereb. Cortex N. Y. NY 30, 4662–4676 (2020). + +44. Zajzon, B., Duarte, R. & Morrison, A. Towards reproducible models of sequence learning: replication and analysis of a modular spiking network with reward-based learning. bioRxiv 2023.01.18.524604 (2023) doi:10.1101/2023.01.18.524604. + +45. Cone, I. & Shouval, H. Z. Behavioral Time Scale Plasticity of Place Fields: Mathematical Analysis. Front. Comput. Neurosci. 15, (2021). + +46. Ljungberg, T., Apicella, P. & Schultz, W. Responses of monkey dopamine neurons during learning of behavioral reactions. J. Neurophysiol. 67, 145–163 (1992). + +47. Clark, J. J., Collins, A. L., Sanford, C. A. & Phillips, P. E. M. Dopamine Encoding of Pavlovian Incentive Stimuli Diminishes with Extended Training. J. Neurosci. 33, 3526–3532 (2013). + +48. Coddington, L. T. & Dudman, J. T. The timing of action determines reward prediction signals in identified midbrain dopamine neurons. Nat. Neurosci. 21, 1563–1573 (2018). + +49. Schultz, W., Apicella, P. & Ljungberg, T. Responses of monkey dopamine neurons to reward and conditioned stimuli during successive steps of learning a delayed response task. J. Neurosci. 13, 900–913 (1993). + +50. Schultz, W. Predictive Reward Signal of Dopamine Neurons. J. Neurophysiol. 80, 1–27 (1998). + +51. Kamin, L. J. Predictability, surprise, attention, and conditioning. B Campbell R M Church Eds Punishm. Aversive Behav. 279–296 (1969). + +52. Mackintosh, N. J. & Turner, C. Blocking as a function of novelty of CS and predictability of UCS. Q. J. Exp. Psychol. 23, 359–366 (1971). + +53. Waelti, P., Dickinson, A. & Schultz, W. Dopamine responses comply with basic assumptions of formal learning theory. Nature 412, 43–48 (2001). + +54. Holland, P. C. Unblocking in Pavlovian appetitive conditioning. J. Exp. Psychol. Anim. Behav. Process. 10, 476–497 (1984). + +55. Steinberg, E. E. et al. A causal link between prediction errors, dopamine neurons and learning. Nat. Neurosci. 16, 966–973 (2013). + +56. Starkweather, C. K., Babayan, B. M., Uchida, N. & Gershman, S. J. Dopamine reward prediction errors reflect hidden-state inference across time. Nat. Neurosci. 20, 581–589 (2017). + +<--- Page Split ---> + +57. Gardner, M. P. H., Schoenbaum, G. & Gershman, S. J. Rethinking dopamine as generalized prediction error. Proc. R. Soc. B Biol. Sci. 285, 20181645 (2018). + +58. Namboodiri, V. M. K. & Stuber, G. D. The learning of prospective and retrospective cognitive maps within neural circuits. Neuron 109, 3552-3575 (2021). + +59. Hayden, B. Y. & Niv, Y. The case against economic values in the orbitofrontal cortex (or anywhere else in the brain). Behav. Neurosci. 135, 192 (2021). + +60. Rajan, K., Abbott, L. F. & Sompolinsky, H. Stimulus-Dependent Suppression of Chaos in Recurrent Neural Networks. Phys. Rev. E 82, 011903 (2010). + +61. Rouhani, N., Norman, K. A., Niv, Y. & Bornstein, A. M. Reward prediction errors create event boundaries in memory. Cognition 203, 104269 (2020). + +62. Yagishita, S. et al. A critical time window for dopamine actions on the structural plasticity of dendritic spines. Science 345, 1616-1620 (2014). + +63. Hong, S. Z. et al. Norepinephrine potentiates and serotonin depresses visual cortical responses by transforming eligibility traces. Nat. Commun. 13, 3202 (2022). + +64. Gerstner, W., Lehmann, M., Liakoni, V., Corneil, D. & Brea, J. Eligibility Traces and Plasticity on Behavioral Time Scales: Experimental Support of NeoHebbian Three-Factor Learning Rules. Front. Neural Circuits 12, (2018). + +65. Brzosko, Z., Schultz, W. & Paulsen, O. Retroactive modulation of spike timing-dependent plasticity by dopamine. eLife 4, e09685 (2015). + +66. Gavornik, J. P. & Shouval, H. Z. A network of spiking neurons that can represent interval timing: mean field analysis. J. Comput. Neurosci. 30, 501-513 (2011). + +<--- Page Split ---> + +## Supplemental Figures + +![](images/Figure_6.jpg) + + +## Supplemental Figure 1 - Dynamics of the DA integral. + +a) Time course of Z scored signal per one animal and one day from the Amo et al data (2022)23 (blue curve). The yellow shaded area illustrates the definition of the integrated signal. b) Re-analyzed data from Amo et al (2022)23. Here we plot the time evolution of the integral of DA response over training days. The different color symbols are the average DA integral (averaged over trials) per each training day, per each animal. Different animals are color coded. The blue line is the mean over animals. c) Average integrals for early (days 1-2), intermediate (days 3-4) and late (days 8-10) of training. Early is significantly lower than intermediate (Wilcoxon rank sum test \(p = 0.004\) ) and late \((p = 0.006)\) , but intermediate is not significantly higher than late. + +<--- Page Split ---> +![PLACEHOLDER_30_0] + + +Supplemental Figure 2 – Two-Trace Learning Flexibly Encodes Cue-Reward Delay in Recurrent Connections of Timer Neurons Demonstration of the dynamics of two- trace learning for recurrent connections. Rows consist of different trials. Left column, mean firing rate of Timer neurons (black) and dopamine reinforcement D(t) (light blue) for a given trial. Middle column, LTP (red) and LTD (blue) associated eligibility traces triggered by the Hebbian overlap \(\frac{r_i \cdot r_j}{1 + \alpha D(t)}\) . Right column, \(\frac{dW}{dt}\) , calculated at a given time as the difference between the two traces \((T_{ij}^p - T_{ij}^d)\) multiplied by the dopamine reinforcement, D(t). For all trials, the increase in recurrent weights is mediated by dopamine release at \(t_{US}\) . Since dopamine acts to suppress trace generation in PFC, the CS- evoked dopamine response (trial 20) has little effect on the recurrent learning. The weights increase until \(\Delta W\) is zero (trial 30) and remain at their fixed point after US- evoked dopamine has been suppressed (trial 40). + +<--- Page Split ---> +![PLACEHOLDER_31_0] + + +## Supplemental Figure 3 – Two-Trace Learning Encodes Reward Predictive Cues via Feed-Forward Connections to DA Neurons + +Demonstration of the dynamics of two- trace learning for feed- forward connections. Rows consist of different trials. Left column, input from external conditioned stimulus (orange) and mean firing rate of dopamine neurons (light blue) for a given trial. Middle column, LTP (red) and LTD (blue) associated eligibility traces triggered by the Hebbian overlap \(r_{CS} * r_{DA}\) . Right column, \(\frac{dW}{dt}\) , calculated at a given time as the difference between the two traces \((T_{ij}^p - T_{ij}^d)\) multiplied by the dopamine reinforcement, D(t). Initially (trial 1), increase in feed forward weights is mediated by dopamine release at \(t_{US}\) . However, as a CS- evoked dopamine response begins to develop (trial 5), the weights are bounded at a fixed point, constrained by the relative positive (from the US) and negative (from the CS) contributions (trial 20). After the expected reward at \(t_{US}\) has been depressed, the DA neuron firing at \(t_{CS}\) decreases slightly until \(\frac{dW}{dt}\) reaches a fixed point maintained by the CS dopamine alone. + +<--- Page Split ---> +![PLACEHOLDER_32_0] + + +## Supplemental Figure 4 – Consistent Unpairing Leads to Negative RPEs and Unlearning + +a) Mean firing rate over all VTA DA neurons for a given trial after learning reward (i), upon omission of the learned reward (ii), and finally, following unlearning (iii). The omission of the expected reward produces a characteristic dopamine “dip”, which in turn acts as a negative RPE and facilitates unlearning of the cue-reward association. b) Mean Timer→Timer (top), CS→DA (middle), and M→GABA (bottom) synaptic weights over the course of pairing and unpairing. The cue is presented for all trials, while the reward is only presented for the first 45 trials (shaded grey). For trials ~30-45, M→GABA weights increase, suppressing the amount of DA firing at the time of the reward. In response, the fixed point of CS→DA weights (which is determined, in part, by the dynamics of the dopamine reinforcement D(t)) decreases before stabilizing in trials ~40-45. Following omission, both CS→DA and M→GABA weights decrease to zero, but T→T weights are maintained, since the fixed point for these weights (see Supplemental Figure 2) is agnostic to changes in to D(t). + +<--- Page Split ---> +![PLACEHOLDER_33_0] + + +## Supplemental Figure 5 – Both Cues Contribute to Reward Prediction Following Initial Training + +a) Mean firing rate of all VTA DA neurons for presentation of CS1+CS2+US following sequential conditioning. b) Spike raster of Timer neurons (1-100, 201-300), Messenger neurons (101-200, 301-400), and VTA DA neurons (401-500) for the same trial. c) Mean firing rate of all VTA DA neurons for presentation of CS1+US, omitting CS2 following simultaneous conditioning. Notably, CS1 only partially predicts (and in turn partially inhibits) the dopamine response at \(t_{US}\) . d) Spike raster of Timer neurons (1-100, 201-300), Messenger neurons (101-200, 301-400), and VTA DA neurons (401-500) for the same trial. + +<--- Page Split ---> +![PLACEHOLDER_34_0] + +
Supplemental Figure 6 – Expected Rewards Block Learning of New Cue-Reward Associations Unless Reward Magnitude is Increased
+ +Each column marks a different phase of conditioning in a blocking/unblocking paradigm. Results shown are averaged over 25 trials. Top row, visual representation of protocol in given column. Middle row, mean over all DA neurons and trials for given presentation protocol. Bottom row, a selected single unit response averaged over trials of the given presentation protocol. Before learning, unpredicted rewards trigger dopamine release at \(t_{US}\) . In phase 1, CS1→US is learned, and DA neurons develop a response to CS1 and have suppressed their response to the US. In phase 2, CS1→CS2→US is presented, but a dopamine response fails to develop to CS2, as the US is already fully predicted by CS1 (and therefore CS2 is "blocked"). Phase 3 results in "unblocking" via increasing dopamine release at \(t_{US}\) – doing so recovers the CS2-evoked response and results in both stimuli becoming reward-predictive. + +<--- Page Split ---> diff --git a/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3_det.mmd b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f31532b87043292a83f6702b707056418ac7b5d4 --- /dev/null +++ b/preprint/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3/preprint__133ea09548ba1407528a6fcf831fcdfaf4efa7804742b01e2a033ea449a740b3_det.mmd @@ -0,0 +1,768 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 880, 208]]<|/det|> +# Learning to Express Reward Prediction Error-like Dopaminergic Activity Requires Plastic Representations of Time + +<|ref|>text<|/ref|><|det|>[[44, 230, 185, 248]]<|/det|> +Harel Shouval + +<|ref|>text<|/ref|><|det|>[[52, 258, 333, 274]]<|/det|> +harel.shouval@uth.tmc.edu + +<|ref|>text<|/ref|><|det|>[[44, 303, 644, 323]]<|/det|> +University of Texas Health https://orcid.org/0000- 0003- 2799- 1337 + +<|ref|>text<|/ref|><|det|>[[44, 328, 270, 414]]<|/det|> +lan Cone Imperial College London Claudia Clopath Imperial College London + +<|ref|>sub_title<|/ref|><|det|>[[44, 456, 103, 473]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 494, 137, 512]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 531, 354, 550]]<|/det|> +Posted Date: September 19th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 570, 474, 588]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3289985/v1 + +<|ref|>text<|/ref|><|det|>[[44, 607, 912, 650]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 668, 533, 687]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[44, 723, 910, 766]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 12th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 50205- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[88, 62, 872, 97]]<|/det|> +# Learning to Express Reward Prediction Error-like Dopaminergic Activity Requires Plastic Representations of Time + +<|ref|>text<|/ref|><|det|>[[95, 130, 617, 148]]<|/det|> +Abbreviated title: RPE via Plastic Representations of Time + +<|ref|>text<|/ref|><|det|>[[95, 181, 842, 267]]<|/det|> +Ian Cone1,2,3, Claudia Clopath1, Harel Z. Shouval2,4 1Department of Bioengineering, Imperial College London, London, United Kingdom; 2Department of Neurobiology and Anatomy, University of Texas Medical School at Houston, Houston, TX; 3Applied Physics Program, Rice University, Houston, TX; 4Department of Electrical and Computer Engineering, Rice University, Houston, TX + +<|ref|>text<|/ref|><|det|>[[95, 299, 738, 317]]<|/det|> +Corresponding Autor: Harel Shouval, email: harel.shouval@uth.tmc.edu + +<|ref|>text<|/ref|><|det|>[[96, 351, 367, 368]]<|/det|> +No financial conflict of interest + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[455, 88, 540, 105]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[87, 114, 910, 427]]<|/det|> +The dominant theoretical framework to account for reinforcement learning in the brain is temporal difference (TD) reinforcement learning. The TD framework predicts that some neuronal elements should represent the reward prediction error (RPE), which means they signal the difference between the expected future rewards and the actual rewards. The prominence of the TD theory arises from the observation that firing properties of dopaminergic neurons in the ventral tegmental area appear similar to those of RPE model- neurons in TD learning. Previous implementations of TD learning assume a fixed temporal basis for each stimulus that might eventually predict a reward. Here we show that such a fixed temporal basis is implausible and that certain predictions of TD learning are inconsistent with experiments. We propose instead an alternative theoretical framework, coined FLEX (Flexibly Learned Errors in Expected Reward). In FLEX, feature specific representations of time are learned, allowing for neural representations of stimuli to adjust their timing and relation to rewards in an online manner. In FLEX dopamine acts as an instructive signal which helps build temporal models of the environment. FLEX is a general theoretical framework that has many possible biophysical implementations. In order to show that FLEX is a feasible approach, we present a specific biophysically plausible model which implements the principles of FLEX. We show that this implementation can account for various reinforcement learning paradigms, and that its results and predictions are consistent with a preponderance of both existing and reanalyzed experimental data. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 62, 910, 145]]<|/det|> +The term reinforcement learning is used in machine learning \(^{1}\) , in behavioral science \(^{2}\) and in neurobiology \(^{3}\) , to denote learning on the basis of rewards or punishment. One type of reinforcement learning is temporal difference (TD) learning, which was designed for machine learning purposes. It has the normative goal of estimating future rewards when rewards can be delayed in time with respect to the actions or cues that engendered these rewards \(^{1}\) . + +<|ref|>text<|/ref|><|det|>[[80, 145, 910, 243]]<|/det|> +One of the variables in TD algorithms is called reward prediction error (RPE), and it is simply the difference between the discounted predicted reward at the current state and the discounted predicted reward + actual reward at the next state. The concept of TD learning became prominent in neuroscience once it was demonstrated that firing patterns of dopaminergic neurons in ventral tegmental area (VTA) during reinforcement learning resemble RPE \(^{4 - 6}\) . + +<|ref|>text<|/ref|><|det|>[[80, 243, 910, 342]]<|/det|> +Implementations of TD using computer algorithms are straightforward but are more complex when they are mapped onto plausible neural machinery \(^{7 - 9}\) . Current implementations of neural TD assume a set of temporal basis- functions \(^{9,10}\) . These basis functions are activated by external cues. For this assumption to hold, each possible external cue must activate a separate set of basis- functions, and these basis- functions must tile all possible learnable intervals between stimulus and reward. + +<|ref|>text<|/ref|><|det|>[[80, 342, 910, 473]]<|/det|> +In this paper we demonstrate that 1) these assumptions are implausible from a fundamental conceptual level, and 2) predictions of such algorithms are inconsistent with various established experimental results. Instead, we propose that temporal basis functions used by the brain are themselves learned. We call this theoretical framework: Flexibly Learned Errors in Expected Reward, or FLEX for short. We also propose a biophysically plausible implementation of FLEX, as a proof- of- concept model. We show that key predictions of this model are consistent with actual experimental results but are inconsistent with some key predictions of the TD theory. + +<|ref|>sub_title<|/ref|><|det|>[[88, 500, 175, 516]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[88, 518, 619, 535]]<|/det|> +## TD learning with a fixed feature specific temporal-basis + +<|ref|>text<|/ref|><|det|>[[88, 535, 900, 669]]<|/det|> +The original TD learning algorithms assumed that agents can be in a set of discrete labeled states (s) which are stored in memory. The goal of TD is to learn a value function such that each state becomes associated with a unique value \((V(s))\) that estimates future discounted rewards. Learning is driven by the difference between value at two subsequent states, and hence such algorithms are called temporal difference algorithms. Mathematically this is captured by the update algorithm: \(V(s) \leftarrow V(s) + \alpha (r(s') + \gamma V(s') - V(s))\) , where \(s'\) is the next state and \(r(s')\) is the reward in next state, \(\gamma\) is an optional discount factor and \(\alpha\) is the learning rate. + +<|ref|>text<|/ref|><|det|>[[80, 680, 910, 860]]<|/det|> +The term in the brackets in the right- hand side of the equation is called the RPE. It represents the different between the estimated value at the current state and the estimated discounted value at the next state in addition to the actual reward at the next state. If RPE is zero for every state, the value function no longer changes, and learning reaches a stable state. In experiments that linked RPE to the firing patterns of dopaminergic neurons in VTA, a transient conditioned stimulus (CS) is presented to a naive animal followed by a delayed reward (also called unconditioned stimulus or US, Figure 1a). It was found that VTA neurons initially respond at the time of reward, but once the association between stimulus and reward is learned, neurons stop firing at the time of the reward and start firing at the time of the stimulus (Figure 1b). This response pattern is what one would expect from TD learning if VTA neurons represent RPE \(^{5}\) . + +<|ref|>text<|/ref|><|det|>[[80, 861, 910, 910]]<|/det|> +Learning algorithms similar to TD have been very successful in machine learning \(^{11,12}\) . In such implementations the state (s) could, for example, represent the state of the chess board, or the coordinates in space in a navigational task. Each of these states could be associated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 62, 910, 195]]<|/det|> +with a value. The state space in such examples might be very large, but the values of all these different states could be feasibly stored in a computer's memory. In some cases, a similar formulation seems feasible for a biological system as well. For example, consider a 2- D navigation problem, where each state is a location in space. One could imagine that each state would be represented by the set of hippocampal place cells activated in this location \(^{13}\) , and that another set of neurons would encode the value function, while a third population of neurons (the "RPE") neurons would compare the value at the current and subsequent state. On its face, this seems to be a reasonable assumption. + +<|ref|>text<|/ref|><|det|>[[80, 194, 910, 277]]<|/det|> +However, in contrast to cases where a discrete set of states might have straightforward biological implementation, there are many cases in which this machine learning inspired algorithm cannot be implemented simply in biological machinery. For example, in experiments where reward is delivered with a temporal delay with respect to the stimulus offset (Figure 1), an additional assumption of a preexisting temporal basis is required \(^{8}\) . + +<|ref|>sub_title<|/ref|><|det|>[[87, 287, 740, 305]]<|/det|> +## Why is it implausible to assume a fixed temporal-basis in the brain? + +<|ref|>text<|/ref|><|det|>[[80, 304, 910, 452]]<|/det|> +Consider the simple canonical example of Figure 1. In the time interval between the stimulus and the reward, the animal does not change its location, nor does its sensory environment change in any predictable manner. The only thing that changes consistently within this interval is time itself. Hence, in order to mark the states between stimulus and reward, the brain must have some internal representation of time, an internal clock which tracks the time since the start of the stimulus. Note however that before the conditioning starts, the animal has no way of knowing that this specific sensory stimulus has unique significance and therefore each specific stimulus must a priori be assigned its own specific temporal representation. + +<|ref|>text<|/ref|><|det|>[[80, 451, 910, 600]]<|/det|> +This is the main hurdle of implementing TD in a biophysically realistic manner - figuring out how to represent the temporal basis upon which the association between cue and reward occurs (Figure 1a). Previous attempts were based on the assumption that there is a fixed cue- specific temporal basis, an assumption which has previously been termed "a useful fiction" \(^{8}\) . The specific implementations include the commonly utilized tapped delay lines \(^{5,14,15}\) (or the so- called complete serial compound), which are neurons triggered by the sensory cue, but that are active only at a specific delay, or alternatively, a set of cue specific neuronal responses which are also delayed but have a broader temporal support which increases with an increasing delay (the so called "microstimuli") \(^{8,10}\) (Figure 1c). + +<|ref|>text<|/ref|><|det|>[[80, 599, 910, 715]]<|/det|> +For this class of temporal representations, the delay time between cue and reward is tiled by a chain of neurons, with each neuron representing a cue- specific time (sometimes referred to as a "microstate") (Figure 1c). In the simple case of the complete serial compound, the temporal basis is simply a set of neurons that have non- overlapping responses that start responding at the cue- relative times: \(t_{cue}\) , \(t_{cue} + dt\) , \(t_{cue} + 2dt\) , ..., \(t_{reward}\) . The learned value function (and in turn the RPE) assigned to a given cue at time \(t\) is then given by a learned weighted sum of the activations of these microstates at time \(t\) (Figure 1d). + +<|ref|>text<|/ref|><|det|>[[87, 714, 907, 912]]<|/det|> +We argue that the conception of a fixed cue- dependent temporal basis makes biologically unrealistic assumptions. First, since one does not know a priori whether presentation of a cue will be followed by a reward, these models assume implicitly that every single environmental cue (or combination of environmental cues) must trigger its own sequence of neural microstates, prior to potential cue- reward pairing (Figure 1e). Further, since one does not know a priori when presentation of a cue may or may not be followed by a reward, these models also assume that microstate sequences are arbitrarily long to account for all possible (or a range of possible) cue- reward delays (Figure 1f). Finally, these microstates are assumed to be reliably reactivated upon subsequent presentations of the cue, e.g., a neuron that represents \(t_{cue} + 3dt\) must always represent \(t_{cue} + 3dt\) - across trials, sessions, days, etc. However, implementation of models that generate a chain- like structure of activity can be fragile to biologically relevant factors such as noise, neural + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 62, 910, 146]]<|/det|> +dropout, and neural drift, all of which suggest that the assumption of reliability is problematic as well (Figure 1g). The totality of these observations imply that on the basis of first principles, it is hard to justify the idea of the fixed feature- specific temporal basis, a mechanism which is required for current supposedly biophysical implementations of TD learning. + +<|ref|>image<|/ref|><|det|>[[100, 150, 744, 567]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 582, 884, 611]]<|/det|> +
Figure 1 – Structure and Assumptions of Temporal Bases for Temporal Difference
+ +<|ref|>text<|/ref|><|det|>[[95, 601, 890, 839]]<|/det|> +Learning a) Diagram of a simple trace conditioning task. A conditioned stimulus (CS) such as a visual grating is paired, after a delay \(\Delta T\) , with an unconditioned stimulus (US) such as a water reward. b) According to the canonical view, neurons in VTA respond only to the US before training, and only to the CS after training. c) In order to represent the delay period, models generally assume neural "microstates" which span the time in between cue and reward. In the simplest case of the complete serial compound (left) the microstimuli do not overlap, and each one uniquely represents a different interval. In general, though (e.g.: microstimuli, right), these microstates can overlap with each other and decay over time. d) A weighted sum of these microstates determines the learned value function V(t). e) An agent does not know a priori which cue will subsequently be paired with reward. In turn, microstate TD models implicitly assume that all N unique cues or experiences in an environment each have their own independent chain of microstates before learning. f) Rewards delivered after the end of a particular cue-specific chain cannot be paired with the cue in question. The chosen length of the chain therefore determines the temporal window of possible associations. g) Microstate chains are assumed to be reliable and robust, but realistic levels of neural noise, drift, and variability can interrupt their propagation, thereby disrupting their ability to associate cue and reward. + +<|ref|>text<|/ref|><|det|>[[87, 879, 910, 914]]<|/det|> +Although a fixed set of basis- functions for every possible stimulus is untenable, one could assume that it is possible to replace this assumption with a single set of fixed, universal basis + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 62, 910, 129]]<|/det|> +functions. An example of mechanism that can generate such general basis functions is a fixed recurrent neural network (RNN). Instead of the firing of an individual neuron representing a particular time, here the entire network state can be thought of as a representation of a cue specific time. This setup is illustrated in Figure 2a. + +<|ref|>text<|/ref|><|det|>[[87, 129, 910, 228]]<|/det|> +To understand the consequences of this setup, we assume a simple environment in which one specific stimulus (denoted as stimulus C) is always followed 1000ms later by a reward; this stimulus is the CS. However, it is reasonable to assume for the natural world that this stimulus exists among many other stimuli that do not predict a reward. For simplicity we consider 3 stimuli, A, B and C, which can appear at any possible order, as shown in Figure 2b, but in which stimulus C always predicts a reward with a delay of 1000ms. + +<|ref|>image<|/ref|><|det|>[[95, 262, 870, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[130, 588, 860, 817]]<|/det|> +
Figure 2– A fixed RNN as a basis function generator. a) Schematic of a fixed recurrent neural network as a temporal basis. The network receives external inputs and generates states s(t), which act as a basis for learning the value function V(t). Compare to figure 1c. b) Schematic of the task protocol. Every presentation of C is followed by a reward at a fixed delay of 1000ms. However, any combination or sequence of irrelevant stimuli may precede the conditioned stimulus C (they might also come after the CS, e.g. A,C,B). c) Network activity, plotted along its first two principal components, for a given initial state s0 and a sequence of presented stimuli A-B-C (red letter is displayed at the time of a given stimulus' presentation). d) Same as c) but for input sequence B-A-C. e) Overlay of the A-B-C ad B-A-C network trajectories, starting from the state at the time of the presentation of C (state sc). The trajectory of network activity differs in these two cases, so the RNN state does not provide a consistent temporal basis that tracks the time since the presentation of stimulus C.
+ +<|ref|>text<|/ref|><|det|>[[87, 845, 910, 913]]<|/det|> +We simulated the responses of such a fixed RNN to different stimulus combinations (see Methods). The complex RNN activity can be viewed as a projection to a subspace spanned by the first two principal components of the data. In Figure 2c,d we show a projection of the RNN response for two different sequences, A- B- C and B- A- C respectively, aligned to the time + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 63, 907, 95]]<|/det|> +of reward. In Figure 2e we show the two trajectories side by side in the same subspace, starting with the presentation of stimulus C. + +<|ref|>text<|/ref|><|det|>[[88, 96, 909, 177]]<|/det|> +What these results show is that every time stimulus C appears, it generates a different temporal response in the RNN depending on the preceding stimuli. These temporal patterns can also be changed by a subsequent stimulus that may appear between the CS and US. These results mean that such a fixed RNN cannot serve as a universal basis function because its responses are not repeatable. + +<|ref|>text<|/ref|><|det|>[[88, 179, 909, 309]]<|/det|> +There are potential workarounds, such as to force the network states representing the time since stimulus C to be the same across trials. This is equivalent to learning the weights of the network such that all possible "distractor" cues pass through the network's null space. This means that the stimulus resets the network and erases its memory, but that other stimuli have no effect on the network. Generally, one would have to train the RNN to reproduce a given dynamical pattern representing C- >reward, while also being invariant to noise or task- irrelevant dimensions, the latter of which can be arbitrarily high and would have to be sampled over during training. + +<|ref|>text<|/ref|><|det|>[[88, 309, 909, 456]]<|/det|> +However, this approach requires a priori knowledge that C is the conditioned stimuli (since C- > reward is the dynamical pattern we want to preserve) and that the other stimuli are nuisance stimuli. This leaves us with quite a conundrum. In the prospective view of temporal associations assumed by TD, to learn that C is associated with reward, we require a steady and repeatable labeled temporal basis (i.e. the network tracks the time since stimulus C). However, to train an RNN to robustly produce this basis, we need to have previously learned that C is associated with reward, and that the other stimuli are not. As such, these modifications to the RNN, while mathematically convenient, are based on unreasonable assumptions. + +<|ref|>sub_title<|/ref|><|det|>[[90, 484, 818, 501]]<|/det|> +## Models of TD learning with a fixed temporal-basis are inconsistent with data + +<|ref|>text<|/ref|><|det|>[[88, 502, 909, 746]]<|/det|> +Apart from being based on unrealistic assumptions, models of TD learning also make predictions that are inconsistent with experimental data. In recent years, several experiments presented evidence of neurons with temporal response profiles that resemble temporal basis functions \(^{16 - 21}\) , as depicted schematically in Figure 3a. While there is indeed evidence of sequential activity in the brain spanning the delay between cues and rewards (such as in the striatum and hippocampus), these sequences are generally observed after association learning between a stimulus and a delayed reward \(^{16,19,20}\) . Some of these experiments have further shown that if the interval between stimulus and reward is extended, the response profiles either remap \(^{19}\) , or stretch to fill out the new extended interval \(^{16}\) , as depicted in Figure 3a. The fact that these sequences are observed after training and that the temporal response profiles are modified when the interval is changed supports the notion of plastic stimulus specific basis functions, rather than of a fixed set of basis function for each possible stimulus. Mechanistically, these results suggest that the naive network might generate generic temporal response profile to novel stimuli before learning, resulting from the networks initial connectivity. + +<|ref|>text<|/ref|><|det|>[[88, 747, 909, 911]]<|/det|> +In the canonical version of TD learning (TD(0)), RPE neurons exhibit a bump of activity that moves back in time from the US to the CS during the learning process (Figure 3b- left). Subsequent versions of TD learning, called TD(λ), which speed up learning by the use of a memory trace, have a much smaller, or no noticeable moving bump (Figure 3b, center and right), depending on the length of the memory trace, denoted by λ. Most published experiments have not shown VTA neuron responses during the learning process. In one prominent example by Pan et al. in which VTA neurons are observed over the learning process \(^{22}\) , (depicted schematically in Figure 2c) no moving bump is observed, prompting the authors to deduce that such memory traces exist. In a more recent paper by Amo et al. a moving bump is reported \(^{23}\) . In contrast, in another recently published paper no moving bump + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 62, 910, 130]]<|/det|> +is observed \(^{24}\) . Taken together, these different results suggest that at least in some cases a moving bump is not observed. However, since a moving bump is not predicted in TD( \(\lambda\) ) for sufficiently large \(\lambda\) , these results do not invalidate the TD framework in general, but rather suggest that in some cases at least the TD(0) variant is inconsistent with the data \(^{22}\) . + +<|ref|>text<|/ref|><|det|>[[87, 129, 910, 327]]<|/det|> +While the moving bump prediction is parameter dependent, another prediction common to all TD variants is that the integrated RPE, obtained by summing response magnitudes over the whole trial duration, does not exceed the initial US response on the first trial. This prediction is robust because the normative basis of TD is to evaluate expected reward or discounted expected reward. In versions of TD where non- discounted reward is evaluated ( \(\gamma = 1\) ) the integral of RPE activity should remain constant throughout learning. Commonly TD estimates discounted reward ( \(\gamma < 1\) ), where the discount means that rewards that come with a small delay are worth more than rewards that arrive with a large delay. With discounted rewards the integral of RPE activity will decrease with learning and become smaller than the initial US response. In contrast, we reanalyzed data from a recent experiment (Amo et al. 2022) \(^{23,25}\) and found that the integrated response can transiently increase over the course of learning (Figure 3d). + +<|ref|>text<|/ref|><|det|>[[87, 326, 910, 508]]<|/det|> +An additional prediction of TD learning which holds across many variants is that when learning converges, and if a reward is always delivered (100% reward delivery schedule), the response of RPE neurons at the time of reward is zero. Even in the case where a small number of rewards are omitted (e.g. 10%), TD predicts that the response of RPE neurons at the time of reward is very small, much smaller than at the time of the stimulus. This seems to be indeed the case for several example neurons shown in the original experiments of dopaminergic VTA neurons \(^{5}\) . However, additional data obtained more recently indicates this might not always be the case and that significant response to reward persists throughout training \(^{22,23,26}\) . This discrepancy between TD and the neural data is observed both for experiments in which responses throughout learning are presented \(^{22,23}\) as well as in experiments that only show results after training \(^{26}\) . + +<|ref|>text<|/ref|><|det|>[[87, 507, 910, 655]]<|/det|> +In experimental approaches affording large ensembles of DA neurons to be simultaneously recorded, a diversity of responses has been reported. Some DA neurons are observed to become fully unresponsive at time of reward, while others exhibit a robust response at time of reward that is no weaker than the initial US response of these cells. This is clearly exhibited by one class of dopaminergic cells (type I) that Cohen et. al \(^{26}\) recorded in VTA. This diversity implies that TD is inconsistent with the results of some of the recorded neurons, but it is possible that TD does apply in some sense to the whole population. One complicating factor is that in most experiments we have no way of ascertaining that learning has reached its final state. + +<|ref|>text<|/ref|><|det|>[[87, 655, 910, 787]]<|/det|> +The original conception of TD is clean, elegant, and based on a simple normative foundation of estimating expected rewards. Over the years, various experimental results that do not fully conform with the predictions of TD have been interpreted as consistent with theory by making ad- hoc modifications \(^{27,28}\) . Such modifications might include an assumption of different variants of the learning rule for each neuron, such that each dopaminergic neuron no longer represents RPE \(^{27}\) , or an assumption of additional inputs such that even when the expectation of reward is learned and fully expected dopaminergic neurons still respond at the time of reward \(^{28}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 70, 910, 530]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 537, 905, 770]]<|/det|> +
Figure 3- Certain features of experimental results run counter to predictions of TD. a) Putative temporal basis functions observed in experiments \(^{16,19}\) after training are plastic, shown here schematically. If, after training, the interval between CS and US is scaled, the basis-functions also change. Recordings in striatum \(^{16}\) show these basis-functions scale with the modified interval (top), while in recordings from hippocampus \(^{19}\) (bottom), they are observed to redistribute to fill up the new interval. b) According to the \(TD(0)\) theory, RPE neuron activity during learning exhibits a backward moving bump (left). For \(TD(\lambda)\) the bump no longer appears (right). c) A schematic depiction of experiments where there is no backward shifting bump \(^{22,24}\) . d) The integral of DA neuron activity according to TD theory (left) should be constant over training (for \(\gamma = 1\) ) or decreasing monotonically for \((\gamma < 1)\) . We reanalyzed existing experimental data (right) and found that the integral can transiently significantly increase. The inset shows the mean and standard errors for early (1-2), intermediate (3-4) and late (8-10) days. (Data from Amo et al 2022 \(^{23,25}\) , see Supplemental Figure 1).
+ +<|ref|>text<|/ref|><|det|>[[88, 786, 911, 902]]<|/det|> +The various modifications that are added to account for new data, and which diverge from a straightforward implementation of TD, raise the subtle question: when does TD stop being TD? We propose that changes to TD in which its normative basis of maximizing future - discounted reward no longer holds are no longer part of the TD framework. The types of modifications added to theory to account for the experimental data are quite common in science in general. Scientific theories that no longer account for the data are often repeatedly modified before there is eventually a paradigm shift, and they are reluctantly abandoned \(^{29}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 63, 910, 163]]<|/det|> +Towards this end, a recent paper has also shown that dopamine release, as recorded with photometry, seems to be inconsistent with RPE24. This paper has shown many experimental results that are at odds with those expected by a RPE, and specifically, these experiments show that dopamine release at least partially represents retrospective probability of stimulus given reward. Other work has suggested that dopamine signaling is more consistent with direct learning of behavioral policy than a value- based RPE30. + +<|ref|>image<|/ref|><|det|>[[100, 180, 890, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 577, 900, 791]]<|/det|> +
Figure 4- Potential Architectures for a Flexible Temporal Basis. Three example networks which could implement a FLEX theory. Top, network architecture. Middle, activity of one example neuron in the network. Bottom, network activity before and after training. Each network initially only has transient responses to stimuli, modifying plastic connections (in blue) during association to develop a specific temporal basis to reward-predictive stimuli. a) Synfire chains could support a FLEX model, if the chain could recruit more members during learning, exclusive to reward-predictive stimuli. b) A population of neurons with homogenous recurrent connections have a characteristic decay time that is related to the strength of the weights. The cue-relative time can then be read out by the mean level of activity in the network. c) A population of neurons with heterogenous and large recurrent connections (liquid state machine) can represent cue-relative time by treating the activity vector at time t as the "microstate" representing time t (as opposed to b) where only mean activity is used).
+ +<|ref|>sub_title<|/ref|><|det|>[[88, 818, 852, 852]]<|/det|> +## The FLEX theory of reinforcement learning: A theoretical framework based on a plastic temporal basis. + +<|ref|>text<|/ref|><|det|>[[88, 852, 910, 919]]<|/det|> +The FLEX theory assumes that there is a plastic (as opposed to fixed) temporal basis that evolves alongside the changing response of reward dependent neurons (such as DA neurons in VTA). The theory in general is agnostic about the functional form of the temporal basis, and several possible examples are shown in Figure 4 (top, schematic; middle, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 62, 910, 211]]<|/det|> +characteristic single unit activity; bottom, population activity before and after learning). Synfire chains (Figure 4a), homogenous recurrent networks (Figure 4b), and heterogeneous recurrent networks (Figure 4c) could all plausibly support the temporal basis. Before learning, these basis- functions do not exist, though some neurons do respond transiently to the CS. Over learning, such basis- functions develop (Figure 4, bottom) in response to the rewarded stimulus, and not to unrewarded stimuli. In FLEX, we do not need a separate, predeveloped basis for every possible stimulus that spans an arbitrary amount of time. Instead, basis functions only form in the process of learning, develop only to stimuli which are tied to reward, and only span the relevant temporal interval for the behavior. + +<|ref|>text<|/ref|><|det|>[[88, 211, 910, 358]]<|/det|> +In the following, we demonstrate that such a framework can be implemented in a biophysically plausible model, and that such a model not only agrees with many existing experimental observations, but also can reconcile seemingly incongruent results pertaining to sequential conditioning. The aim of this model is to show that the FLEX theoretical framework is possible and plausible given the available data, not to claim that this implementation is a perfectly validated model of reinforcement learning in the brain. Previous models concerning hippocampus and prefrontal cortex have also considered cue memories with adaptive durations, but not explicitly in the context of challenging the fundamental idea of a fixed temporal basis31,32. + +<|ref|>sub_title<|/ref|><|det|>[[88, 370, 741, 387]]<|/det|> +## A biophysically plausible implementation of FLEX, proof-of-concept + +<|ref|>text<|/ref|><|det|>[[88, 387, 910, 518]]<|/det|> +Here we present a biophysically plausible proof- of- concept model that implements FLEX. This model is motivated by previous experimental results17,18 and previous theoretical work in our lab33- 37. The network's full architecture (visualized in Figure 5a) consists of two separate modules, a basis function module, and a reward module, here mapped onto distinct brain areas. We treat the reward module as an analogue of the VTA and the basis- function module akin to a cortical region such as the mPFC or OFC (although other cortical or subcortical regions, notably striatum16,38 might support temporal basis functions). All cells in these regions are modeled as spiking integrate and fire neurons (see Methods). + +<|ref|>text<|/ref|><|det|>[[88, 518, 910, 664]]<|/det|> +We assume that within our basis function module are sub- populations of neurons tuned to certain external inputs, visualized in Figure 5a as set of discrete "columns", each responding to a specific stimulus. Within each column there are both excitatory and inhibitory cells, with a connectivity structure that includes both plastic (dashed lines, Figure 5a) and fixed synaptic connections (solid lines, Figure 5a). The VTA is composed of dopaminergic (DA) and inhibitory GABAergic cells. The VTA neurons have a background firing rate of \(\sim 5\) Hz, and the DA neurons have preexisting inputs from "intrinsically rewarding" stimuli (such as a water reward). The plastic and fixed connections between the modules and from both the CS and US to these modules are also depicted in Figure 5a. + +<|ref|>text<|/ref|><|det|>[[88, 665, 910, 894]]<|/det|> +The model's structure is motivated by observations of distinct classes of temporally- sensitive cell responses that have evolve during trace conditioning experiments in medial prefrontal cortex (mPFC) orbitofrontal cortex (OFC) and primary visual cortex (V1)17,18,36,39,40. The architecture described above allows us to incorporate these observed cell classes into our basis- function module (Figure 5b). The first class of neurons ("Timers") are feature- specific and learn to maintain persistently elevated activity that spans the delay period between cue and reward, eventually decaying at the time of reward (real or expected). The second class, the "Messengers", have an activity profile that peaks at the time of real or expected reward. This set of cells form what has been coined a "Core Neural Architecture" (CNA)35, a potentially canonical package of neural temporal representations. A slew of previous studies have shown these cell classes within the CNA to be a robust phenomenon experimentally18,36,39- 43, and computational work has demonstrated that the CNA can be used to learn and recall single temporal intervals35, Markovian and non- Markovian sequences34,44. For simplicity, our model treats connections between populations within a single CNA as fixed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 63, 910, 97]]<|/det|> +(previous work has shown that such a construction is robust to perturbation of these weights34,35). + +<|ref|>image<|/ref|><|det|>[[95, 144, 888, 468]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 480, 898, 647]]<|/det|> +
Figure 5 – Biophysically Inspired Architecture Allows for Flexible Encoding of Time a) Diagram of model architecture. CNAs (visualized here as columns) located in PFC are selective to certain sensory stimuli (indicated here by color atop the column) via fixed excitatory inputs (CS). VTA DA neurons receive fixed input from naturally positive valence stimuli, such as food or water reward (US). DA neuron firing releases dopamine, which acts as a learning signal for both PFC and VTA. Solid lines indicate fixed connections, while dotted lines indicate learned connections. b,c) Schematic representation of data adapted from Liu et al. 201542. b) Timers learn to characteristically decay at the time of cue-predicted reward. c) Messengers learn to have a firing peak at the time of cue-predicted reward.
+ +<|ref|>text<|/ref|><|det|>[[87, 674, 910, 838]]<|/det|> +Learning in the model is dictated by the interaction of eligibility traces and dopaminergic reinforcement. We use a previously established two- trace learning rule34,36,37,45 (TTL), which assumes two Hebbian activated eligibility traces, one associated with LTP and one associated with LTD (see Methods). We use this rule because it solves the temporal credit assignment problem inherent in trace conditioning, reaches stable fixed points, and since such traces have been experimentally observed in trace conditioning tasks36. In theory, other methods capable of solving the temporal credit assignment problem (such as a rule with a single eligibility trace33) could also be used to facilitate learning in FLEX, but owing to its functionality and experimental support, we choose to utilize TTL for this work. See the Methods section for details of the implementation of TTL used here. + +<|ref|>text<|/ref|><|det|>[[88, 838, 909, 904]]<|/det|> +Now, we will use this implementation of FLEX to simulate several experimental paradigms, showing that it can account for reported results. Importantly, some of the predictions of the model are categorically different than those produced by TD, which allows us to distinguish between the two theories based on experimental evidence. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 62, 866, 80]]<|/det|> +## CS-evoked and US-evoked Dopamine Responses Evolve on Different Timescales + +<|ref|>text<|/ref|><|det|>[[88, 80, 910, 195]]<|/det|> +First, we test FLEX on a basic trace conditioning task, where a single conditioned stimulus is presented, followed by an unconditioned stimulus at a fixed delay of one second (Figure 6a). The evolution of FLEX over training is mediated by reinforcement learning (via TTL) in three sets of weights: Timer \(\rightarrow\) Timer, Messenger \(\rightarrow\) VTA GABA neurons, and CS \(\rightarrow\) VTA DA neurons. These learned connections encode the feature- specific cue- reward delay, the temporally specific suppression of US- evoked dopamine, and the emergence of CS- evoked dopamine, respectively. + +<|ref|>text<|/ref|><|det|>[[88, 194, 910, 293]]<|/det|> +Upon presentation of the cue, cue neurons (CS) and feature- specific Timers are excited, producing Hebbian- activated eligibility traces at their CS- \(>\) DA and T- \(>\) T synapses, respectively. When the reward is subsequently presented one second later, the excess dopamine it triggers acts as a reinforcement signal for these eligibility traces (which we model as a function of the DA neuron firing rate, D(t), see Methods) causing both the cue neurons' feed- forward connections and the Timers' recurrent connections to increase (Figure 6b). + +<|ref|>text<|/ref|><|det|>[[88, 292, 910, 375]]<|/det|> +Over repeated trials of this cue- reward pairing, the Timers' recurrent connections continue to increase until they approach their fixed points, which corresponds to the Timers' persistent firing duration increasing until it spans the delay between CS and US (Supplemental Figure 2 and Methods). These mature Timers then provide a feature- specific representation of the expected cue- reward delay. + +<|ref|>text<|/ref|><|det|>[[88, 374, 910, 490]]<|/det|> +The increase of feed- forward connections from the CS to the DA neurons (6c) causes the model to develop a CS- evoked dopamine response. Again, this feed- forward learning uses dopamine release at \(t_{US}\) as the reinforcement signal to convert the Hebbian activated CS \(\rightarrow\) DA eligibility traces into synaptic changes (Supplemental Figure 3). The emergence of excess dopamine at the time of the CS \((t_{CS})\) owing to these potentiated connections also acts to maintain them at a non- zero fixed point, so CS- evoked dopamine persists long after US- evoked dopamine has been suppressed to baseline (see Methods). + +<|ref|>text<|/ref|><|det|>[[88, 489, 910, 654]]<|/det|> +As the Timer population modifies its timing to span the delay period, the Messengers are "dragged along", since, owing to the dynamics of the Messengers' inputs (T and Inh), the Messengers themselves selectively fire at the end of the Timers' firing envelope. Eventually, the Messengers overlap with tonic background activity of VTA GABAergic neurons at the time of the US \((t_{US})\) (Figure 6b). When combined with the dopamine release at \(t_{US}\) , this overlap triggers Hebbian learning at the Messenger \(\rightarrow\) VTA GABA synapses (see Figure 6c, Methods), which indirectly suppresses the DA neurons. Because of the temporal specificity of the Messengers, this learned inhibition of the DA neurons (through excitation of the VTA GABAergic neurons) is effectively restricted to a short time window around the US and acts to suppress DA neural activity at \(t_{US}\) back towards baseline. + +<|ref|>text<|/ref|><|det|>[[88, 654, 910, 802]]<|/det|> +As a result of these processes, our model recaptures the traditional picture of DA neuron activity before and after learning a trace conditioning task (Figure 6b). While the classical single neuron results of Schultz and others suggested that DA neurons are almost completely lacking excess firing at the time of expected reward5, more recent calcium imaging studies have revealed that a complete suppression of the US response is not universal. Rather, many optogenetically identified dopamine neurons maintain a response to the US and show varying development of a response to the CS22,23,26. This diversity is also exhibited in our implementation of FLEX (Figure 6d) due to the connectivity structure which is based on sparse random projections from the CS to the VTA and from the US to VTA. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 60, 923, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 570, 916, 753]]<|/det|> +
Figure 6- CS-evoked and US-evoked Model Dopamine Responses Evolve on Different Timescales. The model is trained for 40 trials while being presented with a CS at 100ms and a reward at 1100ms. a) Mean firing rates for the CNA (see inset for colors), for three different stages of learning. b) Mean firing rate over all DA neurons taken at the same three stages of learning. Firing above or below thresholds (dotted lines) evokes positive or negative D(t) in PFC. c) Evolution of mean synaptic weights over the course of learning. Top, middle, and bottom, mean strength of \(\mathrm{T}\rightarrow \mathrm{T}\) , CS \(\rightarrow\) DA, and M \(\rightarrow\) GABA synapses, respectively. d) Area under receiver operating characteristic (auROC, see Methods) for all VTA neurons in our model for 15 trials before (left, US only), 15 trials during (middle, CS+US), and 15 trials after (right, CS+US) learning. Values above and below 0.5 indicate firing rates above and below the baseline distribution.
+ +<|ref|>text<|/ref|><|det|>[[88, 769, 911, 901]]<|/det|> +During trace conditioning in FLEX, the inhibition of the US- evoked dopamine response (via M \(\rightarrow\) GABA learning) occurs only after the Timers have learned the delay period (since M and GABA firing must overlap to trigger learning), giving the potentiation of the CS response time to occur first. At intermediate learning stages (e.g. trial 5, Figure 6a,b), the CS- evoked dopamine response (or equivalently, the CS \(\rightarrow\) DA weights) already exhibits significant potentiation while the US- evoked dopamine response (or equivalently, inverse of the M \(\rightarrow\) US weights) has only been slightly depressed. While this phenomenon has been occasionally observed in certain experimental paradigms \(^{22,46 - 48}\) , it has not been widely commented on – + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 62, 909, 97]]<|/det|> +in FLEX, this is a fundamental property of the dynamics of learning (in particular, very early learning). + +<|ref|>text<|/ref|><|det|>[[87, 102, 910, 252]]<|/det|> +If an expected reward is omitted in FLEX, the resulting DA neuron firing will be inhibited at that time, demonstrating the characteristic dopamine "dip" seen in experiments (Supplemental Figure 4) \(^{5}\) . This phenomenon occurs in our model because the previous balance between excitation and inhibition (from the US and GABA neurons, respectively) is disrupted when the US is not presented. The remaining input at \(t_{US}\) is therefore largely inhibitory, resulting in a transient drop in firing rates. If the CS is consistently presented without being paired with the US, the association between cue and reward is unlearned, since the consistent negative D(t) at the time of the US causes depression of CS \(\rightarrow\) DA weights (Supplemental Figure 4). + +<|ref|>image<|/ref|><|det|>[[103, 280, 840, 711]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[96, 716, 850, 902]]<|/det|> +
Figure 7 – FLEX Model Dynamics Diverges from those of TD Learning Dynamics of both TD(λ) and FLEX when trained with the same conditioning protocol as shown in Figure 6. a) Dopaminergic release D(t) for FLEX (left), and RPE for TD(λ) (right), over the course of training. b) Sum total (i.e., the sum of each row in a) is a point on the line of b)) dopaminergic reinforcement D(t) during a given trial of training in FLEX (black), and sum total RPE in three instances of TD(λ) with discounting parameters \(\gamma = 1\) , \(\gamma = .99\) , and \(\gamma = .95\) (red, orange, and yellow, respectively). Shaded areas indicate functionally different stages of learning in FLEX. The learning rate in our model is reduced in this example, as to make direct comparison with TD(λ). 100 trials of US-only presentation (before learning) are included for comparison with subsequent stages.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 62, 714, 80]]<|/det|> +## Dynamics of FLEX Diverge from those of TD During Conditioning + +<|ref|>text<|/ref|><|det|>[[88, 80, 911, 211]]<|/det|> +FLEX's property that the evolution of the CS responses can occur independently of (and before) depression of the US response underlies a much more fundamental and general departure of our model from TD based models. In our model, DA activity does not "travel" backwards over trials \(^{1,5}\) as in TD(0), nor is DA activity transferred from one time to the other in an equal and opposite manner as in TD( \(\lambda\) ) \(^{7,22}\) . This is because our DA activity is not a strict RPE signal. Instead, while the DA neural firing in FLEX may resemble RPE following successful learning, the DA neural firing and RPE are not equivalent, as evidenced during the learning period. + +<|ref|>text<|/ref|><|det|>[[88, 211, 911, 393]]<|/det|> +To demonstrate this, we compare the putative DA responses in FLEX to the RPEs in TD( \(\lambda\) ) (Figure 7a), training on the previously described trace conditioning task (see Figure 6a). We set the parameters of our TD( \(\lambda\) ) model to match those in earlier work \(^{22}\) and approximate the cue and reward as Gaussians centered at \(t_{\mathrm{CS1}}\) and \(t_{\mathrm{CS2}}\) , respectively. In TD models, by definition, integral of the RPE over the course of the trial is always less than or equal to the original total RPE provided by the initial unexpected presentation of reward. In other words, the error in reward expectation cannot be larger than the initial reward. In both TD(0) and TD( \(\lambda\) ), this quantity of "integrated RPE" is conserved; for versions of TD with a temporal discounting factor \(\gamma\) (which acts such that a reward with value \(r_0\) presented \(n\) timesteps in the future is only worth \(r_0 \gamma^n\) where \(\gamma \leq 1\) ), this quantity decreases as learning progresses. + +<|ref|>text<|/ref|><|det|>[[88, 392, 911, 556]]<|/det|> +In FLEX, by contrast, integrated dopaminergic release D(t) during a given trial can be greater than that evoked by the original unexpected US (see Figure 7b), and therefore during training the DA signal in FLEX diverges from a reward prediction error. This property has not been explicitly investigated, and most published experiments do not provide continuous data during the training phase. However, to test our prediction, we re- analyzed recently published data which does cover the training phase \(^{23}\) , and found that there is indeed a significant transient increase in the dopamine release during training (Figure 3d and Supplemental Figure 1). Another recent publication found that initial DA response to the US was uncorrelated with the final DA response to the CS \(^{30}\) , which also supports the idea that integrated dopamine release is not conserved. + +<|ref|>sub_title<|/ref|><|det|>[[80, 572, 523, 589]]<|/det|> +## FLEX Unifies Sequential Conditioning Results + +<|ref|>text<|/ref|><|det|>[[85, 589, 911, 833]]<|/det|> +Standard trace conditioning experiments with multiple cues (CS1 \(\rightarrow\) CS2 \(\rightarrow\) US) have generally reported the so- called "serial transfer of activation" – that dopamine neurons learn to fire at the time of the earliest reward- predictive cue, "transferring" their initial activation at \(t_{\mathrm{US}}\) back to the time of the first conditioned stimulus \(^{49,50}\) . However, other results have shown that the DA neural responses at the times of the first and second conditioned stimuli ( \(t_{\mathrm{CS1}}\) and \(t_{\mathrm{CS2}}\) , respectively) evolve together, with both CS1 and CS2 predictive of the reward \(^{22,24}\) . Surprisingly, FLEX can reconcile these seemingly contradictory results. In Figure 8 we show simulations of sequential conditioning using FLEX. In early training we observe an emerging response to both CS1 and CS2, as well as to the US (Figure 8b, ii). Later on the response to the US is suppressed (Figure 8b, iii). During late training (Figure 8b, iv) the response to CS2 is suppressed and all activation is transferred to the earliest predictive stimulus, CS1. These evolving dynamics can be compared to the different experimental results. Early training is similar to the results of Pan et al. (2005) \(^{22}\) and Jeong et al. (2022) \(^{24}\) , as seen in Figure 8c, while the late training results are similar to the results of Schultz (1993) \(^{49}\) as seen in Figure 8d. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[139, 66, 875, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 491, 875, 689]]<|/det|> +
Figure 8 – FLEX model Reconciles Differing Experimental Phenomena Observed during Sequential Conditioning Results from “sequential conditioning”, where sequential neutral stimuli CS1 and CS2 and are paired with delayed reward US. a) Visualization of protocol. In this example, the US is presented starting at 1500ms, with CS1 is presented starting at 100ms, and CS2 is presented starting at 800ms. b) Mean firing rates over all DA neurons, for four distinctive stages in learning – initialization(i), acquisition(ii), reward depression(iii), and serial transfer of activation(iv). c) Schematic illustration of experimental results from recorded dopamine neurons, labeled with the matching stage of learning in our model. c) DA neuron firing before (top), during (middle) and after (bottom) training, wherein two cues (0s and 4s) were followed by a single reward (6s). Adapted from Pan et al. 200522. d) DA neuron firing after training wherein two cues (instruction, trigger) were followed by a single reward. Adapted from Schultz et al. 199349.
+ +<|ref|>text<|/ref|><|det|>[[88, 689, 910, 821]]<|/det|> +After training, in FLEX, both sequential cues still affect the dopamine release at the time of reward, as removal of either cue results in a partial recovery of the dopamine response at \(t_{\mathrm{US}}\) (Supplemental Figure 5). This is the case even in late training in FLEX, when there is a positive dopamine response only to first cue – our model predicts that removal of the second cue will result in a positive dopamine response at the time of expected reward. In contrast, the RPE hypothesis would posit that after extended training, the value function would eventually be maximized following the first cue, and therefore removal of the subsequent cue would not change dopamine release at \(t_{\mathrm{US}}\) . + +<|ref|>text<|/ref|><|det|>[[88, 821, 910, 920]]<|/det|> +FLEX is also capable of replicating results of a different set of sequential learning paradigms (Supplemental Figure 6). In these protocols, the network is initially trained on a standard trace conditioning task with a single CS. Once the cue- reward association is learned completely, a second cue is inserted in between the initial cue and the reward, and learning is repeated. As in experiments, this intermediate cue on its own does not become reward predictive, a phenomenon called “blocking”51–53. However, if reward magnitude is increased + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 62, 909, 113]]<|/det|> +or additional dopamine is introduced to the system, a response to the intermediate CS (CS2) emerges, a phenomenon termed "unblocking" \(^{54,55}\) . Each of these phenomena can be replicated in FLEX (Supplemental Figure 6). + +<|ref|>sub_title<|/ref|><|det|>[[88, 120, 216, 137]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[87, 138, 910, 318]]<|/det|> +TD has established itself as one of the most successful models of neural function to date, as its predictions regarding RPE have, to a large extent, matched experimental results. However, two key factors make it reasonable to consider alternatives to TD as a modeling framework for how midbrain dopamine neurons could learn RPE. First, attempts for biologically plausible implementations of TD have previously assumed that even before learning, each possible cue triggers a separate chain of neurons which tile an arbitrary period of time relative to the cue start. The a priori existence of such an immense set of fixed, arbitrarily long cue- dependent basis functions is both implausible and inconsistent with experimental evidence. Second, different conditioning paradigms have revealed dopamine dynamics that are incompatible with the predictions of models based on the TD framework \(^{22,24,56,57}\) . + +<|ref|>text<|/ref|><|det|>[[87, 318, 910, 483]]<|/det|> +To overcome these problems, we suggest that the temporal basis itself is not fixed, but instead plastic, and is learned only for specific cues that lead to reward. We call this theoretical framework FLEX. We also presented a biophysically plausible implementation of FLEX, and we have shown that it can generate flexible basis functions, and that it produces dopamine cell dynamics that are consistent with experimental results. Our implementation should be seen as a proof- of- concept model. It shows that FLEX can be implemented with biophysical components, and that such an implementation is consistent with much of the data. It does not show that the specific details of this implementation, (including the brain regions in which the temporal- basis is developed, the specific dynamics of the temporal basis functions, and the learning rules used) are those used by the brain. + +<|ref|>text<|/ref|><|det|>[[87, 482, 900, 580]]<|/det|> +One of the appealing aspects of TD learning is that it arises from simple normative assumption that the brain needs to estimate future expected reward. Does FLEX have a similar normative basis? Indeed, in the final state of learning, the responses of DA neurons in FLEX resemble RPE neurons in TD, however in FLEX there is no analog for the value neurons assumed in TD. Unlike in TD, activity of FLEX DA neurons in response to the cue represent an association with future expected rewards, independent of a valuation. + +<|ref|>text<|/ref|><|det|>[[87, 580, 910, 712]]<|/det|> +Instead, the goal of FLEX is to learn the association between cue and reward, develop the temporal basis functions that span the period between the two, and transfer DA signaling from the time of the reward to the time of the cue. These basis functions could then be used as a mechanistic foundation for brain activities, including the timing of actions. In essence, DA in FLEX acts to create internal models of associations and timings. Recent experimental evidence \(^{30}\) which suggests that DA correlates more with the direct learning of behavioral policies rather than value- encoded prediction errors is broadly consistent with this view of dopamine's function within the FLEX framework. + +<|ref|>text<|/ref|><|det|>[[87, 712, 910, 876]]<|/det|> +Another recent publication \(^{24}\) has questioned the claim that DA neurons indeed represent RPE, instead hypothesizing that DA release is also dependent on retrospective probabilities \(^{24,58}\) . The design of most historical experiments cannot distinguish between these competing hypotheses. In this recent research project, a set of new experiments were designed specifically to test these competing hypotheses, and the results obtained are inconsistent with the common interpretation that DA neurons simply represent RPE \(^{24}\) . More generally, while the idea that the brain does and should estimate economic value seems intuitive, it has been recently questioned \(^{59}\) . This challenge to the prevailing normative view is motivated by behavioral experimental results which instead suggest that a heuristic process, which does not faithfully represent value, often guides decisions. + +<|ref|>text<|/ref|><|det|>[[87, 876, 907, 909]]<|/det|> +Recent papers have questioned the common view of the response of DA neurons in the brain and their relation to value estimation \(^{9,24,30,59}\) . Here we survey additional problems + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 62, 910, 212]]<|/det|> +with implementations of TD algorithms in neuronal machinery, and propose an alternative theoretical formulation, FLEX, along with a computational implementation of this theory. The fundamental difference from previous work is that FLEX postulates that the temporal basis- functions necessary for learning are themselves learned, and that neuromodulator activity in the brain is an instructive signal for learning these basis functions. Our computational implementation has predictions that are different than those of TD and are consistent with many experimental results. Further, we tested a unique prediction of FLEX (that integrated DA release across a trial can change over learning) by re- analyzing experimental data, showing that the data was consistent with FLEX but not the TD framework. + +<|ref|>sub_title<|/ref|><|det|>[[88, 223, 188, 240]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[88, 241, 270, 258]]<|/det|> +## Network Dynamics + +<|ref|>text<|/ref|><|det|>[[88, 264, 848, 283]]<|/det|> +The membrane dynamics for each neuron \(i\) are described by the following equations: + +<|ref|>sub_title<|/ref|><|det|>[[88, 288, 415, 306]]<|/det|> +## Equation 1 - Membrane Dynamics + +<|ref|>equation<|/ref|><|det|>[[268, 305, 728, 340]]<|/det|> +\[\mathrm{C}\frac{\mathrm{d}\nu_{\mathrm{i}}}{\mathrm{d}\mathrm{t}} = \mathrm{g}_{\mathrm{L}}(\mathrm{E}_{\mathrm{L}} - \nu_{\mathrm{i}}) + \mathrm{g}_{\mathrm{E},\mathrm{i}}(\mathrm{E}_{\mathrm{E}} - \nu_{\mathrm{i}}) + \mathrm{g}_{\mathrm{I},\mathrm{i}}(\mathrm{E}_{\mathrm{I}} - \nu_{\mathrm{i}}) + \sigma\] + +<|ref|>sub_title<|/ref|><|det|>[[87, 367, 499, 386]]<|/det|> +## Equation 2 - Synaptic Activation Dynamics + +<|ref|>equation<|/ref|><|det|>[[344, 383, 650, 430]]<|/det|> +\[\frac{\mathrm{d}s_{\mathrm{i}}}{\mathrm{d}t} = -\frac{s_{\mathrm{i}}}{\tau_{\mathrm{s}}} +\rho (1 - s_{\mathrm{i}})\sum_{\mathrm{k}}\delta (\mathrm{t} - \mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\] + +<|ref|>text<|/ref|><|det|>[[87, 432, 910, 553]]<|/det|> +The membrane potential and synaptic activation of neuron \(i\) are notated \(\nu_{i}\) and \(s_{i}\) . \(g\) refers to the conductance, and \(E\) to the reversal potentials indicated by the appropriate subscript, where leak, excitatory, and inhibitory are indicated by subscripts \(L\) , \(E\) , and \(I\) , respectively. \(\sigma\) is a mean zero noise term. The neuron spikes upon it crossing its membrane threshold potential \(\nu_{th}\) , after which it enters a refractory period \(t_{ref}\) . The synaptic activation \(s_{\mathrm{i}}\) is updated by an amount \(\rho (1 - s_{\mathrm{i}})\) , at each time \((\mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\) the neuron spikes, and decays exponentially when there is no spike. + +<|ref|>text<|/ref|><|det|>[[87, 559, 907, 594]]<|/det|> +The conductance \(g\) is the product of the incoming synaptic weights and their respective presynaptic neurons: + +<|ref|>sub_title<|/ref|><|det|>[[88, 599, 339, 617]]<|/det|> +## Equation 3 - Conductance + +<|ref|>equation<|/ref|><|det|>[[428, 614, 567, 657]]<|/det|> +\[\mathrm{g}_{\alpha ,\mathrm{i}} = \sum_{j}\mathrm{W}_{ij}^{\alpha}\mathrm{s}_{j}\] + +<|ref|>text<|/ref|><|det|>[[87, 663, 910, 716]]<|/det|> +\(W_{ij}^{\alpha}\) are the connection strengths from neuron \(j\) to neuron \(i\) , where the superscript \(\alpha\) can either indicate \(E\) (excitatory) or \(I\) (inhibitory). A firing rate estimate for each neuron \(\mathrm{r}_{\mathrm{i}}\) is calculated as an exponential filter of the spikes, with a time constant \(\tau_{\mathrm{r}}\) . + +<|ref|>sub_title<|/ref|><|det|>[[88, 722, 317, 740]]<|/det|> +## Equation 4 - Firing Rate + +<|ref|>equation<|/ref|><|det|>[[379, 737, 617, 782]]<|/det|> +\[\tau_{\mathrm{r}}\frac{\mathrm{d}\mathrm{r}_{\mathrm{i}}}{\mathrm{d}t} = -\mathrm{r}_{\mathrm{i}} + \sum_{\mathrm{k}}\delta (\mathrm{t} - \mathrm{t}_{\mathrm{k}}^{\mathrm{i}})\] + +<|ref|>sub_title<|/ref|><|det|>[[88, 788, 193, 804]]<|/det|> +## Fixed RNN + +<|ref|>text<|/ref|><|det|>[[87, 804, 907, 838]]<|/det|> +For the network in Figure 2, the dynamics of the units \(\mathrm{u}_{\mathrm{i}}\) in the RNN are described by the equation below: + +<|ref|>sub_title<|/ref|><|det|>[[88, 845, 360, 864]]<|/det|> +## Equation 5 - RNN Dynamics + +<|ref|>equation<|/ref|><|det|>[[364, 871, 633, 918]]<|/det|> +\[\tau_{\mathrm{net}}\frac{\mathrm{d}\mathrm{u}_{\mathrm{i}}}{\mathrm{d}t} = -\mathrm{u}_{\mathrm{i}} + \sum_{k}\mathrm{W}_{\mathrm{ik}}\phi (\mathrm{u}_{\mathrm{k}})\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 60, 910, 138]]<|/det|> +Where \(\mathbf{u}_{\mathrm{i}}\) are the firing rates of the units in the RNN. \(W_{\mathrm{ik}}\) are the recurrent weights of the RNN, each of which is drawn from a normal distribution \(\mathrm{N}(0,\frac{g}{\sqrt{K}})\) , where \(g\) is the "gain" of the network \(^{60}\) and \(\phi\) is a sigmoidal activation function. Each of the inputs given to the network (A, B, or C) is a unique, normally distributed projection of a 100ms step function. + +<|ref|>sub_title<|/ref|><|det|>[[88, 145, 490, 163]]<|/det|> +## Dopaminergic Two-Trace Learning (dTTL) + +<|ref|>text<|/ref|><|det|>[[87, 162, 910, 231]]<|/det|> +Rather than using temporal difference learning, FLEX uses a previously established learning rule based on competitive eligibility traces, known as "two- trace learning" or TTL \(^{34,36,37}\) . However, we replace the general reinforcement signal \(\mathrm{R(t)}\) of previous implementations by a dopaminergic reinforcement \(\mathrm{D(t)}\) . We repeat here the description of \(\mathrm{D(t)}\) from the main text: + +<|ref|>sub_title<|/ref|><|det|>[[88, 235, 455, 253]]<|/det|> +## Equation 6 - Dopamine reinforcement + +<|ref|>equation<|/ref|><|det|>[[250, 250, 747, 310]]<|/det|> +\[\mathrm{D(t)} = \left\{ \begin{array}{ll}\mathrm{r_{DA}(t)} - (\mathrm{r_0} - \theta), & \mathrm{r_{DA}(t)}\leq \mathrm{r_0} - \theta \\ 0, & \mathrm{r_0} - \theta < \mathrm{r_{DA}(t)}< \mathrm{r_0} + \theta \\ \mathrm{r_{DA}(t)} - (\mathrm{r_0} + \theta), & \mathrm{r_{DA}(t)}\geq \mathrm{r_0} + \theta \end{array} \right.\] + +<|ref|>text<|/ref|><|det|>[[87, 312, 910, 413]]<|/det|> +Note that the dopaminergic reinforcement can be both positive and negative, even though actual DA neuron firing (and the subsequent release of dopamine neurotransmitter) can itself only be positive. The bipolar nature of \(\mathrm{D(t)}\) implicitly assumes that background tonic levels of dopamine do not modify weights, and that changes in synaptic efficacies are a result of a departure (positive or negative) from this background level. The neutral region around \(\mathrm{r_0}\) provides robustness to small fluctuations in firing rates which are inherent in spiking networks. + +<|ref|>text<|/ref|><|det|>[[87, 417, 910, 469]]<|/det|> +The eligibility traces, which are synapse specific, act as long- lasting markers of Hebbian activity. The two traces are separated into LTP- and LTD- associated varieties via distinct dynamics, which are described in the equations below. + +<|ref|>sub_title<|/ref|><|det|>[[88, 474, 411, 492]]<|/det|> +## Equation 7 - LTP Trace Dynamics + +<|ref|>equation<|/ref|><|det|>[[344, 488, 647, 530]]<|/det|> +\[\tau^{\mathrm{p}}\frac{\mathrm{d}\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}}{\mathrm{d}t} = -\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}} + \eta^{\mathrm{p}}\mathrm{H}_{\mathrm{ij}}(\mathrm{T}_{\mathrm{max}}^{\mathrm{p}} - \mathrm{T}_{\mathrm{ij}}^{\mathrm{p}})\] + +<|ref|>sub_title<|/ref|><|det|>[[88, 535, 411, 553]]<|/det|> +## Equation 8 - LTD Trace Dynamics + +<|ref|>equation<|/ref|><|det|>[[344, 552, 647, 592]]<|/det|> +\[\tau^{\mathrm{d}}\frac{\mathrm{d}\mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}}{\mathrm{d}t} = -\mathrm{T}_{\mathrm{ij}}^{\mathrm{d}} + \eta^{\mathrm{d}}\mathrm{H}_{\mathrm{ij}}(\mathrm{T}_{\mathrm{max}}^{\mathrm{d}} - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}})\] + +<|ref|>text<|/ref|><|det|>[[87, 594, 910, 820]]<|/det|> +Here, \(T_{ij}^{\alpha}\) (where \(a\in (p,d))\) is the LTP \((p\) superscript) or LTD ( \(d\) superscript) eligibility trace located at the synapse between the j- th presynaptic cell and the i- th postsynaptic cell. The Hebbian activity, \(\mathrm{H_{ij}}\) , is a simple multiplication \(r_i\cdot r_j\) for application of this rule in VTA, where \(r_i\) and \(r_i\) are the time- averaged firing rates at the pre- and post- synaptic cells. Experimentally, the "Hebbian" terms \((\mathrm{H_{ij}})\) which impact LTP and LTD trace generation are complex \(^{36}\) , but in VTA we approximate with the simple multiplication \(r_i\cdot r_j\) . For synapses in PFC, we make the alteration that \(\mathrm{H_{ij}} = \frac{r_i\cdot r_j}{1 + \alpha D(t)}\) , acting to restrict PFC trace generation for large positive RPEs. The alteration of \(\mathrm{H_{ij}}\) by large positive RPEs is inspired by recent experimental work showing that large positive RPEs act as event boundaries and disrupt across- boundary (but not within- boundary) associations and timing \(^{61}\) . Functionally, this altered \(\mathrm{H_{ij}}\) biases Timers in PFC towards encoding a single delay period (cue to cue or cue to reward) and disrupts their ability to encode across- boundary delays. + +<|ref|>text<|/ref|><|det|>[[87, 824, 902, 893]]<|/det|> +The LTP and LTD traces activate (via activation constant \(\eta^{\mathrm{a}}\) ), saturate (at a level \(\mathrm{T}_{\mathrm{max}}^{\mathrm{a}}\) ) and decay (with time constant \(\tau^{\mathrm{a}}\) ) at different rates. \(\mathrm{D(t)}\) binds to these eligibility traces, converting them into changes in synaptic weights. This conversion into synaptic changes is "competitive", being determined by the difference in the product: + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 62, 352, 80]]<|/det|> +## Equation 9 - Weight Update + +<|ref|>equation<|/ref|><|det|>[[368, 77, 627, 115]]<|/det|> +\[\frac{\mathrm{d}W_{\mathrm{ij}}}{\mathrm{dt}} = \eta \mathrm{D(t)}\left(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\right)\] + +<|ref|>text<|/ref|><|det|>[[88, 120, 337, 137]]<|/det|> +where \(\eta\) is the learning rate. + +<|ref|>text<|/ref|><|det|>[[87, 142, 911, 260]]<|/det|> +The above eligibility trace learning rules, with the inclusion of dopamine, are referred to as dopaminergic "two- trace" learning or dTTL. This rule is pertinent not only because it can solve the temporal credit assignment problem, allowing the network to associate events distal in time, but also because dTTL is supported by recent experiments which have found eligibility traces for in multiple brain regions36,62- 65. Notably, in such experiments, the eligibility traces in prefrontal cortex were found to convert into positive synaptic changes via delayed application of dopamine36, which is the main assumption behind dTTL. + +<|ref|>text<|/ref|><|det|>[[87, 264, 911, 467]]<|/det|> +For simplicity and to reduce computational time, in the simulations shown, M- - GABA connections are learned via a simple dopamine modulated Hebbian rule, \(\frac{\mathrm{d}W_{\mathrm{ij}}}{\mathrm{dt}} = \eta \mathrm{D(t)}\mathrm{r}_{\mathrm{i}}\mathrm{r}_{\mathrm{j}}\) . Since these connections are responsible for inhibiting the DA neurons at the time of reward, this learning rule imposes its own fixed point by suppressing \(\mathrm{D(t)}\) down to 0. For any appropriate selection of feed- forward learning parameters in dTTL (Equation 9), the fixed- point \(\mathrm{D(t)} = 0\) is reached well before the fixed point \(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) = \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\) . This is because, by construction, the function of M- - GABA learning is to suppress \(\mathrm{D(t)}\) down to zero. Therefore, the fixed point \(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) = \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\) needs to be placed (via choosing trace parameters) beyond the fixed- point \(\mathrm{D(t)} = 0\) . Functionally, then, both rules act to potentiate M- - GABA connections monotonically until \(\mathrm{D(t)} = 0\) . As a result, the dopamine modulated Hebbian rule is in practice equivalent to dTTL in this case. + +<|ref|>sub_title<|/ref|><|det|>[[88, 473, 295, 489]]<|/det|> +## Network Architecture + +<|ref|>text<|/ref|><|det|>[[87, 490, 911, 539]]<|/det|> +Our network architecture consists of two regions, VTA and PFC, each of which consist of subpopulations of leaky- integrate- and- fire neurons. Both fixed and learned connections exist between certain populations to facilitate the functionality of our model. + +<|ref|>text<|/ref|><|det|>[[87, 539, 909, 636]]<|/det|> +To model VTA, we include 100 dopaminergic and 100 GABAergic neurons, both of which receive tonic noisy input to establish baseline firing rates of \(\sim 5\mathrm{Hz}\) . Naturally appetitive stimuli, such as food or water, are assumed to have fixed connections to DA neurons via the gustatory system. Dopamine release is determined by DA neuron firing above or below a threshold \(\theta\) , and dopamine release acts as reinforcement for all learned connections in the model. + +<|ref|>text<|/ref|><|det|>[[87, 637, 900, 736]]<|/det|> +Our model of PFC is comprised of different feature- specific 'columns'. Within each column there is a CNA microcircuit, with each subpopulation (Timers, Inhibitory, Messengers) consisting of 100 LIF neurons. Previous work has shown that these subpopulations can emerge from randomly distributed connections35, and further that a single mean field neuron can well approximate the activity of each of these subpopulations of spiking neurons66. + +<|ref|>text<|/ref|><|det|>[[87, 736, 905, 850]]<|/det|> +The two- trace learning rule we utilize for our model is described in further detail in previous work34,36,37,45. However, we will attempt to clarify how it functions below. As a first approximation, the traces from the two- trace learning rule we utilize effectively act such that when they interact with dopamine above baseline, the learning rule will favor potentiation, and when they encounter dopamine below baseline, the learning rule will favor depression. Formally, the rule has fixed points which depend on both the dynamics of the traces and the dopamine release \(\mathrm{D(t)}\) : + +<|ref|>sub_title<|/ref|><|det|>[[88, 850, 337, 866]]<|/det|> +## Equation 10 - Fixed Point + +<|ref|>equation<|/ref|><|det|>[[355, 864, 640, 922]]<|/det|> +\[\int_{0}^{T_{trial}}dt\mathrm{D(t)}\left(\mathrm{T}_{\mathrm{ij}}^{\mathrm{p}}(\mathrm{t}) - \mathrm{T}_{\mathrm{ij}}^{\mathrm{d}}(\mathrm{t})\right) = 0\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 60, 910, 216]]<|/det|> +A trivial fixed point exists when \(\mathrm{D(t)} = 0\) for all t. Another simple fixed point exists in the limit that \(\mathrm{D(t)} = \delta (t_R - t)\) , where \(t_R\) is the time of reward, as Equation 10 then reduces to \(\mathrm{T_{ij}^p(t_R) = T_{ij}^d(t_R)}\) . In this case, the weights have reached their fixed point when the traces cross at the time of reward. In practice, the true fixed points of the model are a combination of these two factors (suppression of dopamine and crossing dynamics of the traces). In reality, \(\mathrm{D(t)}\) is not a delta function (and may have multiple peaks during the trial), so to truly calculate the fixed points, one must use Equation 10 as a whole. However, the delta approximation used above gives a functional intuition for the dynamics of learning in the model. + +<|ref|>text<|/ref|><|det|>[[87, 231, 910, 299]]<|/det|> +Supplemental Figure 2 and Supplemental Figure 3 demonstrate examples of fixed points for both recurrent and feed- forward learning, respectively. Note that in these examples the two "bumps" of excess dopamine (CS- evoked and US- evoked) are the only instances of nonzero \(\mathrm{D(t)}\) . As such, we can take the integral in Equation 10 and split it into two parts: + +<|ref|>text<|/ref|><|det|>[[88, 298, 428, 314]]<|/det|> +Equation 11 - Two- part Fixed Point + +<|ref|>equation<|/ref|><|det|>[[192, 312, 800, 375]]<|/det|> +\[t_{CS,ent} \qquad \int_{t_{CS,start}}^{t_{CS,end}}dt\mathrm{D(t)}\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right) + \int_{t_{US,start}}^{t_{US,end}}dt\mathrm{D(t)}\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right) = 0\] + +<|ref|>text<|/ref|><|det|>[[87, 377, 911, 569]]<|/det|> +For recurrent learning, the dynamics evolve as follows. At the beginning, only the integral over \(\Delta t_{US}\) exists, as \(\mathrm{D(t)}\) is initially zero over \(\Delta t_{CS}\) (Trial 1 in Supplemental Figure 2). As a result, the learning rule evolves to approach the fixed point mediated by \(\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right)\) (Trial 20 in Supplemental Figure 2). After the recurrent weights have reached this fixed point and the Timer neurons encode the cue- reward delay (Trial 30 in Supplemental Figure 2), M→GABA learning acts to suppress \(\mathrm{D(t)}\) down to zero as well (Trial 40 in Supplemental Figure 2). Note again that we make the assumption that trace generation in PFC is inhibited during large positive RPEs. This acts to encourage the Timers to encode a single "duration" (whether cue- cue or cue- reward). In line with our assumption, experimental evidence has shown these large positive RPEs act as event boundaries and disrupt across- boundary (but not within- boundary) reports of timing61. + +<|ref|>text<|/ref|><|det|>[[87, 568, 911, 776]]<|/det|> +For feed- forward learning, the weights initially evolve identically to the recurrent weights (Trial 1 in Supplemental Figure 3). Again, only the integral over \(\Delta t_{US}\) exists, so the feed- forward weights evolve according to \(\left(\mathrm{T_{ij}^p(t) - T_{ij}^d(t)}\right)\) . However, soon the potentiation of these feed- forward CS→DA weights themselves cause release of CS- evoked dopamine, and therefore we must consider both integrals to explain the learning dynamics (Trial 5 in Supplemental Figure 3). This stage of learning is harder to intuit, but an intermediate fixed point is reached when the positive \(\Delta W\) produced by the traces' overlap with US- evoked dopamine is equal and opposite to the negative \(\Delta W\) produced by the traces' overlap with CS- evoked dopamine (Trial 20 in Supplemental Figure 3). Finally, after US- evoked dopamine has been suppressed to baseline, the feed- forward weights reach a final fixed point where both positive and negative contributions to \(\Delta W\) over the course of the CS offset each other (Trial 50 in Supplemental Figure 3). + +<|ref|>text<|/ref|><|det|>[[87, 775, 910, 923]]<|/det|> +The measure of "area under receiver operating characteristic" (auROC) is used throughout this paper, for the purpose of making direct comparison to a slew of calcium imaging results that use auROC as a measure of statistical significance. Following the methods of Cohen et al. (2012)26, time is tiled into 50ms bins. For a single neuron, within each 50ms bin, the distribution of spike counts for 25 trials of baseline spontaneous firing (no external stimuli) is compared to the distribution of spike counts during the same time bin for 25 trials of the learning phase in question. For example, in Figure 6d, left, the baseline distributions of spike counts are compared to the distributions of spike counts when US only is presented. ROC is calculated for each bin by sliding the criteria from zero to the max + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 63, 897, 161]]<|/det|> +spike count within the bin, and then plotting P(active>criteria) versus P(baseline>criteria). The area under this curve is then a measure of discriminability between the two distributions, with an auROC of 1 demonstrating a maximally discriminable increase in spikes compared to baseline, 0 demonstrating a maximally discriminable decrease of spikes compared to baseline, and .5 demonstrating an inability to discriminate between the two distributions. + +<|ref|>text<|/ref|><|det|>[[87, 168, 911, 430]]<|/det|> +Data from Amo et al. 2022 was used for Supplemental Figure \(1^{23,25}\) . The data from 7 animals (437- 440, 444- 446) is shown here. This data is already z- scored as described by Amo et al. On every training day only the first 40 trials are used, unless there was a smaller number of rewarded trials, in which case that number was used. For every animal per every day, the integral from the time of the CS to the end of the trial is calculated and averaged over all trials in that day. An example of the integral for one animal and one day is shown in Supplemental Figure 1a. These are the data points in Supplemental Figure 1b, with each animal represented by a different color. The blue line is the average over animals. The bar graphs in Supplemental Figure 1c represent the average over all animals in early (day 1- 2), intermediate (day 3- 4) and late (day 8- 10) periods. Wilcoxon rank sum tests find that intermediate is significantly higher than early (p=0.004) and that late is significantly higher than early (p=0.006). Late is not significantly lower than intermediate. We also find that late is significantly higher than early if we take the average per animal over the early and intermediate days (p=0.02). A MATLAB code that carries out this analysis is posted on https://github.com/ianconehed/FLEX and will also be posted on modelDB together with the code for FLEX. + +<|ref|>sub_title<|/ref|><|det|>[[88, 459, 252, 475]]<|/det|> +## Code Availability + +<|ref|>text<|/ref|><|det|>[[88, 476, 886, 510]]<|/det|> +Code for both simulations and analysis is posted on: https://github.com/ianconehed/FLEX and will be available on ModelDB upon the publication of the paper. + +<|ref|>sub_title<|/ref|><|det|>[[88, 522, 279, 538]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[88, 538, 857, 620]]<|/det|> +This work was supported by Biotechnology and Biological Sciences Research Council (BBSRC) Grants BB/N013956/1 and BB/N019008/1, Wellcome Trust 200790/Z/16/Z, Simons Foundation Engineering and Physical Sciences Research Council (EPSRC) EP/R035806/1, National Institute of Biomedical Imaging and Bioengineering 1R01EB022891- 01 and Office of Naval Research N00014- 16- R- BA01. + +<|ref|>sub_title<|/ref|><|det|>[[88, 633, 285, 648]]<|/det|> +## Competing Interests + +<|ref|>text<|/ref|><|det|>[[89, 650, 479, 666]]<|/det|> +The authors declare no competing interests + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 63, 200, 80]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[87, 90, 884, 138]]<|/det|> +1. Sutton, R. S. & Barto, A. G. Reinforcement Learning, second edition: An Introduction. (MIT Press, 2018). + +<|ref|>text<|/ref|><|det|>[[87, 149, 888, 168]]<|/det|> +2. Glickman, S. E. & Schiff, B. B. A biological theory of reinforcement. Psychol. Rev. 74, 81-109 (1967). + +<|ref|>text<|/ref|><|det|>[[87, 180, 880, 228]]<|/det|> +3. Lee, D., Seo, H. & Jung, M. W. Neural Basis of Reinforcement Learning and Decision Making. Annu. Rev. Neurosci. 35, 287-308 (2012). + +<|ref|>text<|/ref|><|det|>[[87, 240, 884, 317]]<|/det|> +4. A Model of How the Basal Ganglia Generate and Use Neural Signals That Predict Reinforcement. in Models of Information Processing in the Basal Ganglia (eds. Houk, J. C., Davis, J. L. & Beiser, D. G.) (The MIT Press, 1994). doi:10.7551/mitpress/4708.003.0020. + +<|ref|>text<|/ref|><|det|>[[87, 330, 881, 378]]<|/det|> +5. Schultz, W., Dayan, P. & Montague, P. R. A Neural Substrate of Prediction and Reward. Science 275, 1593-1599 (1997). + +<|ref|>text<|/ref|><|det|>[[87, 390, 900, 439]]<|/det|> +6. Montague, P. R., Dayan, P. & Sejnowski, T. J. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. J. Neurosci. 16, 1936-1947 (1996). + +<|ref|>text<|/ref|><|det|>[[87, 450, 886, 469]]<|/det|> +7. Sutton, R. S. Learning to predict by the methods of temporal differences. Mach. Learn. 3, 9-44 (1988). + +<|ref|>text<|/ref|><|det|>[[87, 481, 880, 529]]<|/det|> +8. Ludvig, E. A., Sutton, R. S. & Kehoe, E. J. Evaluating the TD model of classical conditioning. Learn. Behav. 40, 305-319 (2012). + +<|ref|>text<|/ref|><|det|>[[87, 541, 896, 589]]<|/det|> +9. Namboodiri, V. M. K. How do real animals account for the passage of time during associative learning? Behav. Neurosci. 136, 383-391 (2022). + +<|ref|>text<|/ref|><|det|>[[87, 600, 908, 648]]<|/det|> +10. Ludvig, E. A., Sutton, R. S. & Kehoe, E. J. Stimulus Representation and the Timing of Reward-Prediction Errors in Models of the Dopamine System. Neural Comput. 20, 3034-3054 (2008). + +<|ref|>text<|/ref|><|det|>[[87, 660, 872, 707]]<|/det|> +11. Tesauro, G. TD-Gammon, a self-teaching backgammon program, achieves master-level play. Neural Comput. 6, 215-219 (1994). + +<|ref|>text<|/ref|><|det|>[[87, 720, 886, 768]]<|/det|> +12. Arulkumaran, K., Deisenroth, M. P., Brundage, M. & Bharath, A. A. Deep Reinforcement Learning: A Brief Survey. IEEE Signal Process. Mag. 34, 26-38 (2017). + +<|ref|>text<|/ref|><|det|>[[87, 781, 863, 829]]<|/det|> +13. Foster, D. J., Morris, R. G. & Dayan, P. A model of hippocampal dependent navigation, using the temporal difference learning rule. Hippocampus 10, 1-16 (2000). + +<|ref|>text<|/ref|><|det|>[[87, 841, 884, 889]]<|/det|> +14. Montague, P., Dayan, P. & Sejnowski, T. A framework for mesencephalic dopamine systems based on predictive Hebbian learning. J. Neurosci. 16, 1936-1947 (1996). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 61, 885, 108]]<|/det|> +15. Schultz, W. Behavioral Theories and the Neurophysiology of Reward. Annu. Rev. Psychol. 57, 87-115 (2006). + +<|ref|>text<|/ref|><|det|>[[88, 120, 877, 169]]<|/det|> +16. Mello, G. B. M., Soares, S. & Paton, J. J. A Scalable Population Code for Time in the Striatum. Curr. Biol. 25, 1113-1122 (2015). + +<|ref|>text<|/ref|><|det|>[[88, 181, 884, 230]]<|/det|> +17. Namboodiri, V. M. K. et al. Single-cell activity tracking reveals that orbitofrontal neurons acquire and maintain a long-term memory to guide behavioral adaptation. Nat. Neurosci. 22, 1110-1121 (2019). + +<|ref|>text<|/ref|><|det|>[[88, 241, 900, 290]]<|/det|> +18. Kim, E., Bari, B. A. & Cohen, J. Y. Subthreshold basis for reward-predictive persistent activity in mouse prefrontal cortex. Cell Rep. 35, 109082 (2021). + +<|ref|>text<|/ref|><|det|>[[88, 301, 855, 350]]<|/det|> +19. Taxidis, J. et al. Differential Emergence and Stability of Sensory and Temporal Representations in Context-Specific Hippocampal Sequences. Neuron 108, 984-998.e9 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 362, 895, 410]]<|/det|> +20. MacDonald, C. J., Lepage, K. Q., Eden, U. T. & Eichenbaum, H. Hippocampal "Time Cells" Bridge the Gap in Memory for Discontiguous Events. Neuron 71, 737-749 (2011). + +<|ref|>text<|/ref|><|det|>[[88, 422, 895, 471]]<|/det|> +21. Parker, N. F. et al. Choice-selective sequences dominate in cortical relative to thalamic inputs to NAc to support reinforcement learning. Cell Rep. 39, 110756 (2022). + +<|ref|>text<|/ref|><|det|>[[88, 483, 883, 531]]<|/det|> +22. Pan, W.-X. Dopamine Cells Respond to Predicted Events during Classical Conditioning: Evidence for Eligibility Traces in the Reward-Learning Network. J. Neurosci. 25, 6235-6242 (2005). + +<|ref|>text<|/ref|><|det|>[[88, 543, 860, 591]]<|/det|> +23. Amo, R. et al. A gradual temporal shift of dopamine responses mirrors the progression of temporal difference error in machine learning. Nat. Neurosci. 25, 1082-1092 (2022). + +<|ref|>text<|/ref|><|det|>[[88, 603, 890, 621]]<|/det|> +24. Jeong, H. et al. Mesolimbic dopamine release conveys causal associations. Science 0, eabq6740 (2022). + +<|ref|>text<|/ref|><|det|>[[88, 633, 890, 680]]<|/det|> +24. Amo, R., Matias, S., Uchida, N. & Watabe-Uchida, M. Population and single dopamine neuron activity during classical conditioning. Dryad, Dataset (2022) doi:10.5061/DRYAD.HHMGQNKJW. + +<|ref|>text<|/ref|><|det|>[[88, 693, 895, 740]]<|/det|> +26. Cohen, J. Y., Haesler, S., Vong, L., Lowell, B. B. & Uchida, N. Neuron-type-specific signals for reward and punishment in the ventral tegmental area. nature 482, 85-88 (2012). + +<|ref|>text<|/ref|><|det|>[[88, 752, 902, 800]]<|/det|> +27. Dabney, W. et al. A distributional code for value in dopamine-based reinforcement learning. Nature 577, 671-675 (2020). + +<|ref|>text<|/ref|><|det|>[[88, 812, 904, 860]]<|/det|> +28. Mikhael, J. G., Kim, H. R., Uchida, N. & Gershman, S. J. The role of state uncertainty in the dynamics of dopamine. Curr. Biol. 32, 1077-1087 (2022). + +<|ref|>text<|/ref|><|det|>[[86, 873, 866, 891]]<|/det|> +29. Kuhn Thomas. The structure of scientific revolutions. vol. 111 (University of Chicago Press:, 1970). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 60, 896, 110]]<|/det|> +30. Coddington, L. T., Lindo, S. E. & Dudman, J. T. Mesolimbic dopamine adapts the rate of learning from action. Nature 614, 294-302 (2023). + +<|ref|>text<|/ref|><|det|>[[85, 121, 900, 170]]<|/det|> +31. Wang, J. X. et al. Prefrontal cortex as a meta-reinforcement learning system. Nat. Neurosci. 21, 860-868 (2018). + +<|ref|>text<|/ref|><|det|>[[85, 181, 870, 231]]<|/det|> +32. O'Reilly, R. C., Frank, M. J., Hazy, T. E. & Watz, B. PVLV: The Primary Value and Learned Value Pavlovian Learning Algorithm. Behav. Neurosci. 121, 31-49 (2007). + +<|ref|>text<|/ref|><|det|>[[85, 241, 904, 320]]<|/det|> +33. Gavornik, J. P., Shuler, M. G. H., Loewenstein, Y., Bear, M. F. & Shouval, H. Z. Learning reward timing in cortex through reward dependent expression of synaptic plasticity. Proc. Natl. Acad. Sci. 106, 6826-6831 (2009). + +<|ref|>text<|/ref|><|det|>[[85, 331, 900, 381]]<|/det|> +34. Cone, I. & Shouval, H. Z. Learning precise spatiotemporal sequences via biophysically realistic learning rules in a modular, spiking network. eLife 10, e63751 (2021). + +<|ref|>text<|/ref|><|det|>[[85, 391, 886, 441]]<|/det|> +35. Huertas, M. A., Hussain Shuler, M. G. & Shouval, H. Z. A Simple Network Architecture Accounts for Diverse Reward Time Responses in Primary Visual Cortex. J. Neurosci. 35, 12659-12672 (2015). + +<|ref|>text<|/ref|><|det|>[[85, 451, 866, 501]]<|/det|> +36. He, K. et al. Distinct Eligibility Traces for LTP and LTD in Cortical Synapses. Neuron 88, 528-538 (2015). + +<|ref|>text<|/ref|><|det|>[[85, 512, 857, 591]]<|/det|> +37. Huertas, M. A., Schwettmann, S. E. & Shouval, H. Z. The Role of Multiple Neuromodulators in Reinforcement Learning That Is Based on Competition between Eligibility Traces. Front. Synaptic Neurosci. 8, (2016). + +<|ref|>text<|/ref|><|det|>[[85, 601, 864, 651]]<|/det|> +38. Aosaki, T. et al. Responses of tonically active neurons in the primate's striatum undergo systematic changes during behavioral sensorimotor conditioning. J. Neurosci. 14, 3969-3984 (1994). + +<|ref|>text<|/ref|><|det|>[[85, 661, 880, 711]]<|/det|> +39. Chubykin, A. A., Roach, E. B., Bear, M. F. & Shuler, M. G. H. A Cholinergic Mechanism for Reward Timing within Primary Visual Cortex. Neuron 77, 723-735 (2013). + +<|ref|>text<|/ref|><|det|>[[85, 721, 812, 741]]<|/det|> +40. Shuler, M. G. Reward Timing in the Primary Visual Cortex. Science 311, 1606-1609 (2006). + +<|ref|>text<|/ref|><|det|>[[85, 752, 880, 801]]<|/det|> +41. Namboodiri, V. M. K., Huertas, M. A., Monk, K. J., Shouval, H. Z. & Hussain Shuler, M. G. Visually Cued Action Timing in the Primary Visual Cortex. Neuron 86, 319-330 (2015). + +<|ref|>text<|/ref|><|det|>[[85, 811, 910, 891]]<|/det|> +42. Liu, C.-H., Coleman, J. E., Davoudi, H., Zhang, K. & Hussain Shuler, M. G. Selective Activation of a Putative Reinforcement Signal Conditions Cued Interval Timing in Primary Visual Cortex. Curr. Biol. 25, 1551-1561 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 60, 850, 110]]<|/det|> +43. Monk, K. J., Allard, S. & Hussain Shuler, M. G. Reward Timing and Its Expression by Inhibitory Interneurons in the Mouse Primary Visual Cortex. Cereb. Cortex N. Y. NY 30, 4662–4676 (2020). + +<|ref|>text<|/ref|><|det|>[[86, 121, 911, 200]]<|/det|> +44. Zajzon, B., Duarte, R. & Morrison, A. Towards reproducible models of sequence learning: replication and analysis of a modular spiking network with reward-based learning. bioRxiv 2023.01.18.524604 (2023) doi:10.1101/2023.01.18.524604. + +<|ref|>text<|/ref|><|det|>[[86, 211, 863, 259]]<|/det|> +45. Cone, I. & Shouval, H. Z. Behavioral Time Scale Plasticity of Place Fields: Mathematical Analysis. Front. Comput. Neurosci. 15, (2021). + +<|ref|>text<|/ref|><|det|>[[86, 271, 881, 320]]<|/det|> +46. Ljungberg, T., Apicella, P. & Schultz, W. Responses of monkey dopamine neurons during learning of behavioral reactions. J. Neurophysiol. 67, 145–163 (1992). + +<|ref|>text<|/ref|><|det|>[[86, 331, 910, 380]]<|/det|> +47. Clark, J. J., Collins, A. L., Sanford, C. A. & Phillips, P. E. M. Dopamine Encoding of Pavlovian Incentive Stimuli Diminishes with Extended Training. J. Neurosci. 33, 3526–3532 (2013). + +<|ref|>text<|/ref|><|det|>[[86, 391, 840, 441]]<|/det|> +48. Coddington, L. T. & Dudman, J. T. The timing of action determines reward prediction signals in identified midbrain dopamine neurons. Nat. Neurosci. 21, 1563–1573 (2018). + +<|ref|>text<|/ref|><|det|>[[86, 451, 904, 530]]<|/det|> +49. Schultz, W., Apicella, P. & Ljungberg, T. Responses of monkey dopamine neurons to reward and conditioned stimuli during successive steps of learning a delayed response task. J. Neurosci. 13, 900–913 (1993). + +<|ref|>text<|/ref|><|det|>[[86, 541, 831, 561]]<|/det|> +50. Schultz, W. Predictive Reward Signal of Dopamine Neurons. J. Neurophysiol. 80, 1–27 (1998). + +<|ref|>text<|/ref|><|det|>[[86, 572, 894, 620]]<|/det|> +51. Kamin, L. J. Predictability, surprise, attention, and conditioning. B Campbell R M Church Eds Punishm. Aversive Behav. 279–296 (1969). + +<|ref|>text<|/ref|><|det|>[[86, 631, 892, 680]]<|/det|> +52. Mackintosh, N. J. & Turner, C. Blocking as a function of novelty of CS and predictability of UCS. Q. J. Exp. Psychol. 23, 359–366 (1971). + +<|ref|>text<|/ref|><|det|>[[86, 691, 890, 740]]<|/det|> +53. Waelti, P., Dickinson, A. & Schultz, W. Dopamine responses comply with basic assumptions of formal learning theory. Nature 412, 43–48 (2001). + +<|ref|>text<|/ref|><|det|>[[86, 751, 891, 800]]<|/det|> +54. Holland, P. C. Unblocking in Pavlovian appetitive conditioning. J. Exp. Psychol. Anim. Behav. Process. 10, 476–497 (1984). + +<|ref|>text<|/ref|><|det|>[[86, 811, 860, 860]]<|/det|> +55. Steinberg, E. E. et al. A causal link between prediction errors, dopamine neurons and learning. Nat. Neurosci. 16, 966–973 (2013). + +<|ref|>text<|/ref|><|det|>[[86, 872, 890, 921]]<|/det|> +56. Starkweather, C. K., Babayan, B. M., Uchida, N. & Gershman, S. J. Dopamine reward prediction errors reflect hidden-state inference across time. Nat. Neurosci. 20, 581–589 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 60, 886, 108]]<|/det|> +57. Gardner, M. P. H., Schoenbaum, G. & Gershman, S. J. Rethinking dopamine as generalized prediction error. Proc. R. Soc. B Biol. Sci. 285, 20181645 (2018). + +<|ref|>text<|/ref|><|det|>[[86, 120, 870, 169]]<|/det|> +58. Namboodiri, V. M. K. & Stuber, G. D. The learning of prospective and retrospective cognitive maps within neural circuits. Neuron 109, 3552-3575 (2021). + +<|ref|>text<|/ref|><|det|>[[86, 181, 896, 230]]<|/det|> +59. Hayden, B. Y. & Niv, Y. The case against economic values in the orbitofrontal cortex (or anywhere else in the brain). Behav. Neurosci. 135, 192 (2021). + +<|ref|>text<|/ref|><|det|>[[86, 241, 870, 290]]<|/det|> +60. Rajan, K., Abbott, L. F. & Sompolinsky, H. Stimulus-Dependent Suppression of Chaos in Recurrent Neural Networks. Phys. Rev. E 82, 011903 (2010). + +<|ref|>text<|/ref|><|det|>[[86, 301, 831, 350]]<|/det|> +61. Rouhani, N., Norman, K. A., Niv, Y. & Bornstein, A. M. Reward prediction errors create event boundaries in memory. Cognition 203, 104269 (2020). + +<|ref|>text<|/ref|><|det|>[[86, 362, 885, 410]]<|/det|> +62. Yagishita, S. et al. A critical time window for dopamine actions on the structural plasticity of dendritic spines. Science 345, 1616-1620 (2014). + +<|ref|>text<|/ref|><|det|>[[86, 422, 847, 471]]<|/det|> +63. Hong, S. Z. et al. Norepinephrine potentiates and serotonin depresses visual cortical responses by transforming eligibility traces. Nat. Commun. 13, 3202 (2022). + +<|ref|>text<|/ref|><|det|>[[86, 482, 865, 560]]<|/det|> +64. Gerstner, W., Lehmann, M., Liakoni, V., Corneil, D. & Brea, J. Eligibility Traces and Plasticity on Behavioral Time Scales: Experimental Support of NeoHebbian Three-Factor Learning Rules. Front. Neural Circuits 12, (2018). + +<|ref|>text<|/ref|><|det|>[[86, 572, 888, 620]]<|/det|> +65. Brzosko, Z., Schultz, W. & Paulsen, O. Retroactive modulation of spike timing-dependent plasticity by dopamine. eLife 4, e09685 (2015). + +<|ref|>text<|/ref|><|det|>[[86, 632, 890, 680]]<|/det|> +66. Gavornik, J. P. & Shouval, H. Z. A network of spiking neurons that can represent interval timing: mean field analysis. J. Comput. Neurosci. 30, 501-513 (2011). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 91, 301, 108]]<|/det|> +## Supplemental Figures + +<|ref|>image<|/ref|><|det|>[[100, 120, 873, 277]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[88, 290, 602, 309]]<|/det|> +## Supplemental Figure 1 - Dynamics of the DA integral. + +<|ref|>text<|/ref|><|det|>[[117, 319, 911, 470]]<|/det|> +a) Time course of Z scored signal per one animal and one day from the Amo et al data (2022)23 (blue curve). The yellow shaded area illustrates the definition of the integrated signal. b) Re-analyzed data from Amo et al (2022)23. Here we plot the time evolution of the integral of DA response over training days. The different color symbols are the average DA integral (averaged over trials) per each training day, per each animal. Different animals are color coded. The blue line is the mean over animals. c) Average integrals for early (days 1-2), intermediate (days 3-4) and late (days 8-10) of training. Early is significantly lower than intermediate (Wilcoxon rank sum test \(p = 0.004\) ) and late \((p = 0.006)\) , but intermediate is not significantly higher than late. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[85, 90, 830, 488]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[85, 525, 900, 754]]<|/det|> +Supplemental Figure 2 – Two-Trace Learning Flexibly Encodes Cue-Reward Delay in Recurrent Connections of Timer Neurons Demonstration of the dynamics of two- trace learning for recurrent connections. Rows consist of different trials. Left column, mean firing rate of Timer neurons (black) and dopamine reinforcement D(t) (light blue) for a given trial. Middle column, LTP (red) and LTD (blue) associated eligibility traces triggered by the Hebbian overlap \(\frac{r_i \cdot r_j}{1 + \alpha D(t)}\) . Right column, \(\frac{dW}{dt}\) , calculated at a given time as the difference between the two traces \((T_{ij}^p - T_{ij}^d)\) multiplied by the dopamine reinforcement, D(t). For all trials, the increase in recurrent weights is mediated by dopamine release at \(t_{US}\) . Since dopamine acts to suppress trace generation in PFC, the CS- evoked dopamine response (trial 20) has little effect on the recurrent learning. The weights increase until \(\Delta W\) is zero (trial 30) and remain at their fixed point after US- evoked dopamine has been suppressed (trial 40). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 63, 873, 472]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[122, 472, 825, 505]]<|/det|> +## Supplemental Figure 3 – Two-Trace Learning Encodes Reward Predictive Cues via Feed-Forward Connections to DA Neurons + +<|ref|>text<|/ref|><|det|>[[120, 505, 872, 737]]<|/det|> +Demonstration of the dynamics of two- trace learning for feed- forward connections. Rows consist of different trials. Left column, input from external conditioned stimulus (orange) and mean firing rate of dopamine neurons (light blue) for a given trial. Middle column, LTP (red) and LTD (blue) associated eligibility traces triggered by the Hebbian overlap \(r_{CS} * r_{DA}\) . Right column, \(\frac{dW}{dt}\) , calculated at a given time as the difference between the two traces \((T_{ij}^p - T_{ij}^d)\) multiplied by the dopamine reinforcement, D(t). Initially (trial 1), increase in feed forward weights is mediated by dopamine release at \(t_{US}\) . However, as a CS- evoked dopamine response begins to develop (trial 5), the weights are bounded at a fixed point, constrained by the relative positive (from the US) and negative (from the CS) contributions (trial 20). After the expected reward at \(t_{US}\) has been depressed, the DA neuron firing at \(t_{CS}\) decreases slightly until \(\frac{dW}{dt}\) reaches a fixed point maintained by the CS dopamine alone. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[133, 67, 840, 356]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[122, 369, 845, 403]]<|/det|> +## Supplemental Figure 4 – Consistent Unpairing Leads to Negative RPEs and Unlearning + +<|ref|>text<|/ref|><|det|>[[120, 403, 868, 633]]<|/det|> +a) Mean firing rate over all VTA DA neurons for a given trial after learning reward (i), upon omission of the learned reward (ii), and finally, following unlearning (iii). The omission of the expected reward produces a characteristic dopamine “dip”, which in turn acts as a negative RPE and facilitates unlearning of the cue-reward association. b) Mean Timer→Timer (top), CS→DA (middle), and M→GABA (bottom) synaptic weights over the course of pairing and unpairing. The cue is presented for all trials, while the reward is only presented for the first 45 trials (shaded grey). For trials ~30-45, M→GABA weights increase, suppressing the amount of DA firing at the time of the reward. In response, the fixed point of CS→DA weights (which is determined, in part, by the dynamics of the dopamine reinforcement D(t)) decreases before stabilizing in trials ~40-45. Following omission, both CS→DA and M→GABA weights decrease to zero, but T→T weights are maintained, since the fixed point for these weights (see Supplemental Figure 2) is agnostic to changes in to D(t). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 118, 866, 394]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[124, 398, 783, 433]]<|/det|> +## Supplemental Figure 5 – Both Cues Contribute to Reward Prediction Following Initial Training + +<|ref|>text<|/ref|><|det|>[[124, 432, 863, 565]]<|/det|> +a) Mean firing rate of all VTA DA neurons for presentation of CS1+CS2+US following sequential conditioning. b) Spike raster of Timer neurons (1-100, 201-300), Messenger neurons (101-200, 301-400), and VTA DA neurons (401-500) for the same trial. c) Mean firing rate of all VTA DA neurons for presentation of CS1+US, omitting CS2 following simultaneous conditioning. Notably, CS1 only partially predicts (and in turn partially inhibits) the dopamine response at \(t_{US}\) . d) Spike raster of Timer neurons (1-100, 201-300), Messenger neurons (101-200, 301-400), and VTA DA neurons (401-500) for the same trial. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 88, 835, 303]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[95, 303, 792, 338]]<|/det|> +
Supplemental Figure 6 – Expected Rewards Block Learning of New Cue-Reward Associations Unless Reward Magnitude is Increased
+ +<|ref|>text<|/ref|><|det|>[[95, 338, 831, 533]]<|/det|> +Each column marks a different phase of conditioning in a blocking/unblocking paradigm. Results shown are averaged over 25 trials. Top row, visual representation of protocol in given column. Middle row, mean over all DA neurons and trials for given presentation protocol. Bottom row, a selected single unit response averaged over trials of the given presentation protocol. Before learning, unpredicted rewards trigger dopamine release at \(t_{US}\) . In phase 1, CS1→US is learned, and DA neurons develop a response to CS1 and have suppressed their response to the US. In phase 2, CS1→CS2→US is presented, but a dopamine response fails to develop to CS2, as the US is already fully predicted by CS1 (and therefore CS2 is "blocked"). Phase 3 results in "unblocking" via increasing dopamine release at \(t_{US}\) – doing so recovers the CS2-evoked response and results in both stimuli becoming reward-predictive. + +<--- Page Split ---> diff --git a/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/images_list.json b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..4a5858f3e58a9db614c139198aca6c8a81715513 --- /dev/null +++ b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: (a) Image of the DYNAP-SE2 chip, showing the four cores. An E-I network is implemented on chip. Excitatory (Pyr) and Inhibitory (PV) neurons occupy different cores, and are connected by four weight classes, Pyr-to-Pyr \\((w_{ee})\\) , Pyr-to-PV \\((w_{ie})\\) , PV-to-PV \\((w_{ii})\\) , and PV-to-Pyr \\((w_{ei})\\) . (b) Circuit diagram depicting a DPI neuron, with each component within the circuit replicating various sub-threshold characteristics (such as adaptation, leakage, refractory period, etc.) inherent to an adaptive exponential neuron [30]. (c) Neuron membrane potential traces over time as a response to DC injection. Blue traces show two example cells from the excitatory chip core in response to four current levels. Red traces show two sample PV cells from the inhibitory chip core. The relationship between the measured voltage and the internal current-mode circuit output that represents the membrane potential is given by an exponential function that governs the subthreshold transistor operation mode [31]. In each case, the bottom two traces show neuronal dynamics in the subthreshold range. The top two traces show neuronal dynamics as the membrane potential surpasses the spike threshold (spike added as grey line to indicate when the voltage surpasses threshold). DC injection values are indicated on every trace. (d) Input-response curves for the resulting Pyr and PV neurons on-chip. Inset: zoomed-in view at low firing rates. In (c), it is evident that a higher input current was required to elicit spikes in PV neurons. This phenomenon can be attributed to the higher threshold of PV neurons.", + "footnote": [], + "bbox": [ + [ + 103, + 71, + 905, + 540 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: (a) A representative experiment of the convergence of excitatory (red) and inhibitory (blue) firing rates during learning. The gray dashed lines show the desired set-point targets. (b) A representative experiment (the same as in panel a) of the convergence of weight values during learning. The plotted value is nominal, inferred from eq. (1) (see Methods). (c) Raster plot exemplifying on-chip firing activity during a single emulation run at the end of training. Both Pyr and PV neurons successfully converge to an asynchronous-irregular firing pattern at the desired FRs. (d) Initial and final firing rates across different chips (n=57, across two chips) and different initial conditions. For each initial condition, network connectivity was randomized. (e) Left: relationship between the final \\(w_{ee}\\) and \\(w_{ei}\\) values across different iterations. Orange and green colors correspond to two different chips. Right: same for \\(w_{ii}\\) and \\(w_{ie}\\) . (f) Distributions for the 5 ms correlation coefficients between neurons (left) and coefficient of variation ( \\(CV^2\\) ) (right), demonstrating low regularity and low synchrony. The line indicates the mean. Shaded region indicate standard deviation.", + "footnote": [], + "bbox": [ + [ + 95, + 78, + 920, + 730 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: (a). The paradoxical effect is a well known phenomena in cortical circuits. In the awake resting cortex of mice (upper figure), when inhibitory neurons are optogenetically activated, their firing rates (in red) show a paradoxical decrease in activity during the stimulation, indicative of an inhibition-stabilized network. Adapted from Sanzeni et al. [21]. Similar to the experimental case (bottom figure), numerical simulations of firing rate models with sufficient excitatory gain and balanced by inhibition show the paradoxical effect when the inhibitory population is excited via an external current. (b). The paradoxical effect can be demonstrated in analog neuromorphic circuits by applying a depolarizing current to the inhibitory units of networks converged to self-sustained activity via cross-homeostasis. The resulting decrease in the firing rate of the inhibitory units demonstrates that the on-chip network is in the inhibition-stabilized regime. No. of trials \\(= 14\\) , shading region represents standard deviation in firing rate across trials.", + "footnote": [], + "bbox": [ + [ + 115, + 223, + 888, + 466 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: (a) Example network with training results on \"fixed\" inhibitory synapses. Convergence is impaired if inhibitory plasticity is turned off. (b) Random networks ran with different initialization of all weights (n=8). The firing rates do not converge to the targets as precisely as in Fig. 2a, having a much higher root mean square error from the target (RMS on \\(\\bar{E} 6.39\\) , on \\(\\bar{I} 35.11\\) ).", + "footnote": [], + "bbox": [ + [ + 95, + 73, + 864, + 290 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Training result with a memory implanted in the network. (a) Random recurrent E/I network with a neural ensemble (size = 32, green neurons). (b) Convergence of both excitatory (Pyr) and inhibitory (PV) populations to their respective set-point. A subset of the weights is manually changed to represent an ensemble at iteration 250, well after the network has converged to its set-point. A jump in activity is observed in both Pyr and PV cells due to implanted memory. (c) Diagram illustrating the process of rebalancing the weight matrix after implanting a single memory. Due to the chip's limited observable and shared parameters, it's not possible to record the actual weight matrix for visualization. (d) The weight \\((w_{ee})\\) value for 32 neurons is increased by a factor of 0.0088 \\(\\mu \\mathrm{A}\\) , illustrated by a dotted line. All weights undergo alteration again to compensate for the ensemble, bringing the network back to the set point after the memory is implanted. However, the memory remains intact, as shown by higher recurrent weight \\((w_{memory})\\) compared to the baseline excitatory connectivity \\((w_{ee})\\) .", + "footnote": [], + "bbox": [ + [ + 169, + 73, + 825, + 464 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: Simultaneous training of multiple clusters. (a) Schematic of clusters on the chip. One core contains excitatory populations, and another contains inhibitory populations. We implemented five E/I networks, with each network comprising 50 excitatory and 15 inhibitory neurons. As per table 1, the connection probability within these networks stands at 0.35. To account for the network's scaling, we defined higher set points. It's worth noting that these clusters do not interconnect with one another. b) Rate convergence for 5 clusters sharing the same nominal weights. Each color represents one cluster.", + "footnote": [], + "bbox": [ + [ + 109, + 300, + 904, + 530 + ] + ], + "page_idx": 9 + } +] \ No newline at end of file diff --git a/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4.mmd b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4.mmd new file mode 100644 index 0000000000000000000000000000000000000000..362d78c810a5fc48f08ec3a3ba32f312a4e9f340 --- /dev/null +++ b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4.mmd @@ -0,0 +1,343 @@ + +# Orchestrated excitatory and inhibitory plasticity produces stable dynamics in heterogeneous neuromorphic computing systems + +Maryada Maryada maryada@ini.uzh.ch + +Institute of Neuroinformatics, University of Zurich and ETH Zurich Saray Soldado- Magraner University of California, Los Angeles https://orcid.org/0000- 0003- 4890- 2671 Martino Sorbaro ETH AI Center, ETH Zurich https://orcid.org/0000- 0002- 0182- 7443 Rodrigo Laje University of Quilmes https://orcid.org/0000- 0003- 1481- 303X Dean Buonomano University of California, Los Angeles https://orcid.org/0000- 0002- 8528- 9231 Giacomo Indiveri SynSense AG Corporation https://orcid.org/0000- 0002- 7109- 1689 + +## Article + +## Keywords: + +Posted Date: October 20th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3449716/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 60697- 2. + +<--- Page Split ---> + +# Orchestrated excitatory and inhibitory plasticity produces stable dynamics in heterogeneous neuromorphic computing systems + +Maryada \(^{1}\) , Saray Soldado- Magraner \(^{2}\) , Martino Sorbaro \(^{1,3}\) , Rodrigo Laje \(^{4,5}\) , Dean V. Buonomano \(^{2}\) , and Giacomo Indiveri \(^{1}\) + +\(^{1}\) Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland \(^{2}\) Department of Neurobiology, University of California, Los Angeles, USA \(^{3}\) ETH AI Center, ETH Zurich, Switzerland \(^{4}\) Department of Science and Technology, Universidad Nacional de Quilmes, Bernal, Argentina \(^{5}\) CONICET, Argentina + +## Abstract + +Many neural computations emerge from self- sustained patterns of activity in recurrent neural circuits, which rely on balanced excitation and inhibition. Neuromorphic electronic circuits that use the physics of silicon to emulate neuronal dynamics represent a promising approach for implementing the brain's computational primitives, including self- sustained neural activity. However, achieving the same robustness of biological networks in neuromorphic computing systems remains a challenge, due to the high degree of heterogeneity and variability of their analog components. Inspired by the strategies used by real cortical networks, we apply a biologically- plausible cross- homeostatic learning rule to balance excitation and inhibition in neuromorphic implementations of spiking neural networks. We demonstrate how this learning rule allows the neuromorphic system to overcome device mismatch and to autonomously tune the spiking network to produce robust, self- sustained attractor dynamics in an inhibition- stabilized regime. We also show that this rule can implement a stable working memory, and that the electronic circuits can reproduce biologically relevant emergent neural dynamics, including the so- called "paradoxical effect". In addition to validating neuroscience models on a substrate that shares many similar properties and limitations with biological systems, this work enables the construction of ultra- low power, mixed- signal neuromorphic technologies that can be automatically configured to compute reliably, despite the large on- chip and chip- to- chip variability of their analog components. + +## Introduction + +Animal brains can perform complex computations including sensory processing, motor control, and working memory, as well as higher cognitive functions, such as decision- making and reasoning, in an efficient and reliable manner. At the neural network level, these processes are implemented using a variety of computational primitives that rely on neural dynamics within recurrent neocortical microcircuits [1- 6]. Translating the computational primitives observed in the brain into novel technologies can potentially lead to radical innovations in artificial intelligence and edge- computing applications. A promising technology that can implement these primitives with compact and low- power devices is that of mixed- signal neuromorphic systems, which employ analog electronic circuits to emulate the biophysics of real neurons [7- 9]. As opposed to software simulations, the direct emulation performed by these electronic circuits relies on the use of the physics of the silicon substrate to faithfully reproduce the dynamics of neurons and synapses asynchronously in real time. While systems designed following this approach have the advantage of ultra- low power consumption, they have limitations and constraints similar to those found in biology. These include a high degree of variability, heterogeneity, and sensitivity to noise [10, 11]. Due to these constraints, implementing recurrent networks in silicon with the same stability and robustness observed in biological circuits remains an open challenge. + +Previous studies have already demonstrated neuromorphic implementations of recurrent computational primitives such as winner- take- all or state- dependent networks [12- 15]. However, these systems required + +<--- Page Split ---> + +manual tuning and exhibited dynamics that would not always produce stable self- sustained regimes. Manually tuning on- chip recurrent networks with large numbers of parameters is challenging, tedious, and not scalable. Furthermore, because of cross- chip variability, this process requires re- tuning for each individual chip. Even when a stable regime is achieved with manual tuning, changes in the network conditions (e.g., due to temperature variations, increase or decrease of input signals, etc.) would require re- tuning. Therefore, mixed- signal neuromorphic technology can greatly benefit from automatic stabilizing and tuning mechanisms that can overcome the challenges imposed by the analog substrate. + +Neocortical computations rely heavily on positive feedback imposed by recurrent connections between excitatory neurons, which allows networks to perform complex time- dependent computations and actively maintain information about past events. Recurrent excitation, however, also makes neocortical circuits vulnerable to "runaway excitation" and epileptic activity [16, 17]. In order to harness the computational power of recurrent excitation and avoid pathological regimes, the neocortical circuits operate in an inhibition- stabilized regime, in which positive feedback is held in check by recurrent inhibition [18- 21]. Indeed, there is evidence that the default awake cortical dynamic regime may be inhibition- stabilized [20, 22]. It is generally accepted that neocortical microcircuits have synaptic learning rules in place to homeostatically balance excitation and inhibition in order to generate dynamic regimes capable of self- sustained activity [23- 26]. Recently, a family of synaptic learning rules that differentially operate at excitatory and inhibitory synapses has been proposed, which can drive simulated neural networks to self- sustained and inhibition- stabilized regimes in a self- organizing manner [27]. This family of learning rules is referred to as being "cross- homeostatic" because synaptic plasticity at excitatory and inhibitory synapses is dependent on both the inhibitory and excitatory setpoints. + +Taking inspiration from self- calibration principles in cortical networks, we use these cross- homeostatic plasticity rules to guide neuromorphic circuits to balanced excitatory- inhibitory regimes in a self- organizing manner. + +We show that these rules can be successfully employed to autonomously calibrate analog spiking recurrent networks in silicon. Specifically, by automatically tuning all synaptic weight classes in parallel, the networks of silicon neurons converge to fully self- sustained attractor dynamics, and configure themselves into the inhibition- stabilized regime. In addition, the emergent neural dynamics obtained in this way express the "paradoxical effect", a signature of inhibition- stabilized networks, which is observed in cortical circuits [18, 20]. The plasticity rules proposed prove resilient to hardware variability and noise, also across different chips, as well as to different parameter initializations. Importantly, we demonstrate that inhibitory plasticity (often neglected in neuromorphic electronic systems) is necessary for successful convergence. Finally, we also show that by using such plasticity rules, long- term memories can be implemented without disrupting the learned dynamics, and that multiple such networks can be implemented on a single chip robustly. + +From a computational neuroscience perspective, these results validate the robustness of cross- homeostatic plasticity in a physical substrate that presents similar challenges to those of biological networks, unlike idealized digital simulations. From a neuromorphic perspective, this approach provides the community with a reliable method to autonomously and robustly calibrate recurrent neural networks in future mixed- signal analog/digital systems that are affected by device variability (also including memristive devices [28, 29]). This will open the door to designing autonomous systems that can interact in real time with the environment, and compute reliably with low- latency at extremely low- power using attractor dynamics, as observed in biological circuits. + +## Results + +In the cortex, excitatory Pyramidal (Pyr) and inhibitory Parvalbumin- positive interneurons (PV) constitute the main neuronal subtypes and are primarily responsible for excitatory/inhibitory (E/I) balance [32]. Pyr and PV neurons have different intrinsic biophysical properties: in terms of excitability, PV cells have a higher threshold and gain compared to Pyr neurons [33]. These properties, along with other neuronal characteristics, such as refractory period and membrane time constants, contribute to the characteristic regular- spiking patterns of pyramidal cells and fast- spiking patterns of PV interneurons. + +In our hardware experiments, we reproduce the properties of both Pyr and PV neuron types on a mixed- signal analog/digital Dynamic Asynchronous Neuromorphic Processor, called DYNAP- SE2 [30] (see Fig. 1a), which implements an adaptive exponential integrate- and- fire neuronal model. The dynamics of the different neuron + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: (a) Image of the DYNAP-SE2 chip, showing the four cores. An E-I network is implemented on chip. Excitatory (Pyr) and Inhibitory (PV) neurons occupy different cores, and are connected by four weight classes, Pyr-to-Pyr \((w_{ee})\) , Pyr-to-PV \((w_{ie})\) , PV-to-PV \((w_{ii})\) , and PV-to-Pyr \((w_{ei})\) . (b) Circuit diagram depicting a DPI neuron, with each component within the circuit replicating various sub-threshold characteristics (such as adaptation, leakage, refractory period, etc.) inherent to an adaptive exponential neuron [30]. (c) Neuron membrane potential traces over time as a response to DC injection. Blue traces show two example cells from the excitatory chip core in response to four current levels. Red traces show two sample PV cells from the inhibitory chip core. The relationship between the measured voltage and the internal current-mode circuit output that represents the membrane potential is given by an exponential function that governs the subthreshold transistor operation mode [31]. In each case, the bottom two traces show neuronal dynamics in the subthreshold range. The top two traces show neuronal dynamics as the membrane potential surpasses the spike threshold (spike added as grey line to indicate when the voltage surpasses threshold). DC injection values are indicated on every trace. (d) Input-response curves for the resulting Pyr and PV neurons on-chip. Inset: zoomed-in view at low firing rates. In (c), it is evident that a higher input current was required to elicit spikes in PV neurons. This phenomenon can be attributed to the higher threshold of PV neurons.
+ +classes were reproduced by tuning the refractory period, time constants, spiking threshold, and neuron gain parameters (Fig. 1b) to match the corresponding biologically measured values. Since parameters are shared by all neurons in a chip core, we assign each neuron population type to a different core (Fig. 1a). Even though all neurons in a core share the same parameters, variations in their intrinsic characteristics arise as a result of analog chip mismatch. For instance, one can observe differences between two excitatory or inhibitory neurons recorded within the same core, as illustrated in Fig. 1c. Overall, the neural dynamics and Input- Frequency (FI) curve of + +<--- Page Split ---> + +recorded excitatory and inhibitory neurons on chip (Fig. 1c- d show that the choice of parameters used for the silicon neurons is compatible with the behavior of biological PV and Pyr cells [33]. In the following section, we employ cross- homeostatic plasticity [27] to self- tune the connections between PV and Pyr populations so that spontaneous activity in the inhibition- stabilized regime emerges in the network. + +## The network converges to stable self-sustained dynamics + +As an overview of the experimental paradigm, the training procedure begins with random initialization of the network weights. Then, at each iteration, we record the resulting firing rates, calculate the weight update following the cross- homeostatic equations for the four weight classes \((w_{ee},w_{ei},w_{ie},\) and \(w_{ii}\) , equation 2, see Methods) and apply the new weight values to the chip. + +Fig. 2a illustrates an example of the evolution of firing activity during the course of training for Pyr and PV cells. Starting from a random value determined by the initialization of weights, the average rates converge to their target values. Fig. 2b plots the evolution of weights during training. On chip, the weights are controlled by two parameters, a coarse \(C_w\) and a fine \(F_w\) value (See Methods, eq. 4). On most iterations, only the fine value \(F_w\) is updated, but updates to the coarse value \(C_w\) occur occasionally. Even though the coarse updates cause abrupt jumps in the weight value (due to the nature of the bias- generator circuit implementation), the cross- homeostatic rule is robust enough to guide the network to a stable regime. Fig. 2c shows a raster plot from the network in Fig. 2a- b at a converged state. It should be emphasized that, while an initial "kick" was provided to the network to engage recurrent activity (40 ms at \(100\mathrm{Hz}\) ), no further external input was introduced during the trial; hence, any observed spiking activity originates entirely from internal mechanisms. + +When repeating the training process from different initial conditions of randomly sampled weights, the weights converge to different values, which however produce the same desired network average firing rate [27]. Fig. 2d illustrates the rate space with multiple initialization and convergence to respective target set- points (dashed lines). The root mean square error from the target is \(1.02\mathrm{Hz}\) for \(\bar{E}\) and \(6.32\mathrm{Hz}\) for \(\bar{I}\) . + +The converged weight values are approximately aligned to a linear manifold (Fig. 2e), where the sets of excitatory and inhibitory weights \(w_{ee},w_{ei}\) and \(w_{ie},w_{ii}\) , are correlated (correlation coefficients and p- values are as follows: \(w_{ee}\) and \(w_{ei}\) : \(\rho_{green} = 0.81\) and \(p = 1.55\cdot 10^{- 06}\) ; \(\rho_{orange} = 0.88\) , \(p = 6.6\cdot 10^{- 12}\) ; \(w_{ie}\) and \(w_{ii}\) : \(\rho_{green} = 0.63\) \(p = 0.0009\) ; \(\rho_{orange} = 0.75\) , \(p = 3.9\cdot 10^{- 07}\) , where green and orange refer to two different chips). This is well aligned with the theoretical solution derived for rate E- I networks at their set- points, when the neuron transfer function is linear or threshold- linear [27, eqn. 4- 5]. In our case, the linear transfer function assumption does not hold, because the AdEx silicon neuron models saturate at a rate that is inversely proportional to the refractory period parameter. However, the linear approximation holds well in the region of operation forced by the learning rule, set at relatively low firing rates (see Fig. 1d). + +## The network is in an inhibition-stabilized asynchronous-irregular firing regime + +Next we investigated the properties of the final network dynamics after learning. Fig. 2c shows a sample of activity of the converged network, as recorded from the chip, in which the excitatory and inhibitory populations exhibit asynchronous- irregular activity patterns. The minimal correlations among neurons (averages \(\langle \rho_{pyr}\rangle = 0.037\) , \(\langle \rho_{pv}\rangle = 0.069\) ) serve as evidence of their asynchronous activity (Fig. 2f, left). Irregularity is shown by the coefficient of variation ( \(\mathrm{CV}^2\) ) of most neurons being close to 1 (averages \(\langle CV_{pyr}^2\rangle = 0.949\) , \(\langle CV_{pv}^2\rangle = 0.945\) ) (Fig. 2f, right). The \(\mathrm{CV}^2\) is computed as the variance of the inter- spike interval divided by its squared mean, and equals 0 for perfectly regular firing, and 1 for Poisson firing [34]. It is worth noting that during the 1- second simulations, we do find instances of synchronization, which can be attributed to the small population size of the network and increased firing rate in comparison to the baseline spontaneous activity. As an average however, we conclude that cross- homeostatic plasticity brings the network to an asynchronous- irregular firing regime, what is typical in realistic cortical networks [35]. + +As further evidence of the inhibition- stabilized regime of the network, we demonstrate the occurrence of the "paradoxical effect" in our deployed network on chip [18, 20]. The paradoxical effect is a hallmark of inhibition- stabilized networks that has been observed in cortical circuits. When inhibitory neurons are excited (either optogenetically in vivo, or via an external current in computational models) their firing rates show a paradoxical decrease in activity during the stimulation, providing clear evidence that that the network is in an + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: (a) A representative experiment of the convergence of excitatory (red) and inhibitory (blue) firing rates during learning. The gray dashed lines show the desired set-point targets. (b) A representative experiment (the same as in panel a) of the convergence of weight values during learning. The plotted value is nominal, inferred from eq. (1) (see Methods). (c) Raster plot exemplifying on-chip firing activity during a single emulation run at the end of training. Both Pyr and PV neurons successfully converge to an asynchronous-irregular firing pattern at the desired FRs. (d) Initial and final firing rates across different chips (n=57, across two chips) and different initial conditions. For each initial condition, network connectivity was randomized. (e) Left: relationship between the final \(w_{ee}\) and \(w_{ei}\) values across different iterations. Orange and green colors correspond to two different chips. Right: same for \(w_{ii}\) and \(w_{ie}\) . (f) Distributions for the 5 ms correlation coefficients between neurons (left) and coefficient of variation ( \(CV^2\) ) (right), demonstrating low regularity and low synchrony. The line indicates the mean. Shaded region indicate standard deviation.
+ +<--- Page Split ---> + +inhibition- stabilized regime (Fig. 3a). Analogously, when we inject an external depolarizing pulse to all PV neurons on chip for \(0.2\mathrm{s}\) we observe a decrease in the firing rate of both cell populations for the duration of the stimulus, returning back to the target FR when the stimulus ends (Fig. 3b). Across all 14 iterations we use pairwise t- tests to assess the impact of stimulating PV neurons. For PV neurons, stimulation significantly reduces activity during stimulation compared to pre- and post- stimulation states \((p< 10^{- 18}\) in both cases). Conversely, there is no significant difference between pre- and post- stimulation activity, indicating recovery. Five trials, which were excluded from the analysis, do not exhibit recovery because the activity is entirely suppressed by the strength of the paradoxical effect. A similar trend is observed for Pyramidal neurons \((p< 10^{- 20})\) . + +![](images/Figure_3.jpg) + +
Figure 3: (a). The paradoxical effect is a well known phenomena in cortical circuits. In the awake resting cortex of mice (upper figure), when inhibitory neurons are optogenetically activated, their firing rates (in red) show a paradoxical decrease in activity during the stimulation, indicative of an inhibition-stabilized network. Adapted from Sanzeni et al. [21]. Similar to the experimental case (bottom figure), numerical simulations of firing rate models with sufficient excitatory gain and balanced by inhibition show the paradoxical effect when the inhibitory population is excited via an external current. (b). The paradoxical effect can be demonstrated in analog neuromorphic circuits by applying a depolarizing current to the inhibitory units of networks converged to self-sustained activity via cross-homeostasis. The resulting decrease in the firing rate of the inhibitory units demonstrates that the on-chip network is in the inhibition-stabilized regime. No. of trials \(= 14\) , shading region represents standard deviation in firing rate across trials.
+ +## Inhibitory plasticity is necessary to achieve reliable convergence + +The equations for the weights in absence of noise [27, eqn. 4- 5] and the experimental results in Fig. 2 e indicate that two of the four weight parameters, for example, the inhibitory weights \(w_{ii}\) and \(w_{ei}\) , can be chosen arbitrarily, since one can always find a solution by learning the other two. This appears to suggest that homeostatic plasticity of the inhibitory weights might not be required in an ideal case. However, we experimentally show that excitatory plasticity alone is not sufficient to reliably make the network converge in the presence of the noise and non- idealities of the analog substrate (see Fig. 4). In the absence of inhibitory plasticity, both the excitatory and inhibitory populations approach their respective set- points, but they fail to converge to stable activity levels. (Fig. 4a). In contrast, the full cross- homeostatic plasticity rule robustly drives the network activity towards the set- points after just a few iterations (Fig. 2a). These results hold for multiple initializations (Fig. 4b), indicating that inhibitory plasticity is necessary for robust convergence. + +Homeostatic plasticity in inhibitory neurons is less studied than in excitatory neurons [24, 36], although there is evidence that inhibitory neurons also regulate their activity levels based on activity set- points [37, 38]. Considering our on- chip results, we therefore speculate that homeostatic plasticity in all excitatory and inhibitory connections plays an important stabilization role also in biological systems. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: (a) Example network with training results on "fixed" inhibitory synapses. Convergence is impaired if inhibitory plasticity is turned off. (b) Random networks ran with different initialization of all weights (n=8). The firing rates do not converge to the targets as precisely as in Fig. 2a, having a much higher root mean square error from the target (RMS on \(\bar{E} 6.39\) , on \(\bar{I} 35.11\) ).
+ +## Neuronal ensembles remain intact under cross-homeostasis + +The effect of homeostasis is to maintain a set level of activity in neurons by scaling synaptic efficacy, as done in this work, or by changing the intrinsic properties of the neuron. This raises concerns about the stability- plasticity dilemma, i.e., the competition between the homeostatic stability mechanism and the associative forms of synaptic plasticity that contribute to the formation of functional neural ensembles, as they may counteract each other causing either forgetting or the inability to form new memories. A neural ensemble corresponds to a group of interconnected neurons (e.g., that increased their synaptic weights via learning) that contributes to the representation of a particular feature or concept. Typically, neurons forming an ensemble exhibit shared selectivity [39, 40]. + +The ensemble could potentially be affected by the homeostatic scaling process, which attempts to equalize the activity levels across the network. To verify that the cross- homeostatic plasticity used here does not interfere or disrupt the role of neural ensembles we artificially formed a "crafted memory" into the network during the homeostatic training process (Fig. 5). This memory is implemented as a neural ensemble within the excitatory population of the network. Specifically, the ensemble consists of 32 neurons, in which each neuron has a connection probability with other neurons of the same ensemble of 0.5, much higher than the rest of the network (Fig. 5a). + +After 250 cross- homeostatic plasticity training iterations, we manually form the ensemble by introducing the additional connections in it. We observe an immediate increase in the overall activity of both excitatory and inhibitory neurons (Fig. 5b) after ensemble implantation, which is however reduced as the homeostatic training continues. The change in weights (Fig. 5d- c) illustrates how inhibitory plasticity (specially \(w_{ei}\) ) compensates for increased activity, bringing the firing rates back to the target point. Because the values of weights within and outside of the ensemble remain differentiated, even after homeostatic plasticity converges, the neural ensemble remains intact and retains its selectivity for the specific feature it represents. We conclude that the cross- homeostatic learning rule brings the activity back to the target while preserving the implanted memory. + +## Multiple subnetworks can be emulated on a single chip + +In the network discussed so far, we modeled a single stable point attractor in firing rate space, using an E/I- balanced network. We used 200 out of the 256 neurons available on a single core of our multi- core neuromorphic chip (DYNAP- SE2). + +Even though it is still a relatively small- scale network, for computational purposes, having a single attractor formed from 200 neurons is not an optimal utilization of the resources on the chip. Employing smaller networks would allow us to implement multiple attractors on a single chip, whereby each subnetwork could represent, for + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Training result with a memory implanted in the network. (a) Random recurrent E/I network with a neural ensemble (size = 32, green neurons). (b) Convergence of both excitatory (Pyr) and inhibitory (PV) populations to their respective set-point. A subset of the weights is manually changed to represent an ensemble at iteration 250, well after the network has converged to its set-point. A jump in activity is observed in both Pyr and PV cells due to implanted memory. (c) Diagram illustrating the process of rebalancing the weight matrix after implanting a single memory. Due to the chip's limited observable and shared parameters, it's not possible to record the actual weight matrix for visualization. (d) The weight \((w_{ee})\) value for 32 neurons is increased by a factor of 0.0088 \(\mu \mathrm{A}\) , illustrated by a dotted line. All weights undergo alteration again to compensate for the ensemble, bringing the network back to the set point after the memory is implanted. However, the memory remains intact, as shown by higher recurrent weight \((w_{memory})\) compared to the baseline excitatory connectivity \((w_{ee})\) .
+ +example, a distinct choice in a decision- making task. The model could then be extended to introduce competition among the clusters leading to winner- take- all dynamics to facilitate decision- making processes [12- 15]. + +Therefore, we also run scaled down variations of the previous network, and assess whether cross- homeostatic plasticity can still bring activity to desired levels. Table 1 summarizes the population count, connectivity scaling, and adjustment in the target firing rate that we adopt for each experiment. + +Table 1: Variation in network size, connection scaling, and stable target firing rate (set-points) for reliable self-sustained activity. + +
Number of neurons (E/I)Connection ProbabilityTarget (E/I)
200/500.120/40 Hz
100/250.120/40 Hz
64/160.3525/50 Hz
48/120.3540/60 Hz
+ +<--- Page Split ---> + +Furthermore, we experiment with implementing multiple copies of these scaled- down networks on the same chip. We create 5 instances of different E/I networks, each with 50 excitatory and 15 inhibitory neurons. The connectivities within each sub- network are drawn from the same distribution. Each subnetwork does not cross- talk with others. The configuration is illustrated in Fig. 6a. To implement this configuration on the chip, we group the excitatory neuron populations into a single core, and use another core for the inhibitory populations. All parameter values, including weights, are shared between sub- networks. The challenge is to check if homeostatic plasticity leads to convergence for all subnetworks, since the effect of the weights is different for each E/I network due to the inhomogeneity of the chip. + +During the training process, we sequentially stimulate one sub- network at a time during each iteration. We then compute the change in synaptic weights, denoted as \(\Delta w\) , based on the activity of the neuron population within that E/I network. The weight is configured on the chip after each iteration based on the average \(\Delta w\) across all sub- networks. Fig. 6b shows that all five subnetworks, with shared weights, can indeed converge to mean firing rate regimes close to the target firing rate. + +![](images/Figure_6.jpg) + +
Figure 6: Simultaneous training of multiple clusters. (a) Schematic of clusters on the chip. One core contains excitatory populations, and another contains inhibitory populations. We implemented five E/I networks, with each network comprising 50 excitatory and 15 inhibitory neurons. As per table 1, the connection probability within these networks stands at 0.35. To account for the network's scaling, we defined higher set points. It's worth noting that these clusters do not interconnect with one another. b) Rate convergence for 5 clusters sharing the same nominal weights. Each color represents one cluster.
+ +## Discussion + +Hardware implementations of spiking neural networks are being proposed as a promising "neuromorphic" technology that can complement standard computation for sensory- processing applications that require limited resource usage (such as power, memory, and size), low latency, and that cannot resort to sending the recorded data to off- line computing centers (i.e., "the cloud") [29, 41, 42]. Among the proposed solutions, those that exploit the analog properties of their computing substrate have the highest potential of minimizing power consumption and carrying out "always- on" processing [9, 43]. However, the challenge of using this technology is to understand how to carry out reliable processing in the presence of the high variability, heterogeneity and sensitivity to noise of the analog substrate. This challenge becomes even more daunting when taking into account the variability of memristive devices, which are expected to allow for the storage and processing of information directly on chip [44]. + +In this paper, we showed how resorting to strategies used by animal brains to solve the same challenge can lead to robust solutions: we demonstrated that cross- homeostatic plasticity can autonomously bring analog neuromorphic hardware to a fully self- sustained inhibition- stabilized regime, the regime that is thought to be the + +<--- Page Split ---> + +default state of the awake neocortex [19- 22]. These are the first results to show how biologically inspired learning rules can be used to self- tune neuromorphic hardware into computationally powerful states while simultaneously solving the problem of heterogeneity and variability in the hardware. + +Our results also validate the robustness of the cross- homeostatic plasticity hypothesis [27] in a physical substrate that presents similar challenges to those of faced by biological networks. We also highlight the crucial contribution of inhibitory plasticity, specifically, that homeostatic plasticity of inhibitory connections is necessary to maintain stability in both biological and silicon networks. Notably, we constructed an on- chip stable attractor neural network that effectively maintains a long- term memory represented by a neural ensemble, and demonstrates that multiple networks, each exhibiting attractor dynamics, can coexist within the hardware [18, 35]. In the future, we can extend this stable multi- network implementation to simulate decision- making processes, along with self- generated dynamics that are thought to underlie a number of computations including motor control, timing, and sensory integration [45- 47]. + +By taking inspiration from the self- calibration features of biological networks, we achieve a remarkable level of resilience to hardware variability, enabling recurrent networks to maintain stability and self- sustained activity across different analog neuromorphic chips. Cross- homeostatic plasticity allows us to instantiate neural networks on different chips without the need for chip- specific tuning, thereby addressing the inherent mismatch in analog devices. Such bio- plausible algorithms provide a promising avenue for the development of robust and scalable hardware implementations of recurrent neural networks, while mitigating the challenges associated with the intrinsic mismatch in analog devices. + +In short, these results not only provide insights into the underlying mechanisms configuring biological circuits, but also hold great promise for the development of neuromorphic applications that unlock the power of recurrent computations. This research presents a substantial step forward in our understanding of neural dynamics and paves new possibilities for practical deployment of neuromorphic devices that compute, like biological circuits, with extremely low- power and low- latency using attractor dynamics. + +## Methods + +## Neuromorphic hardware: the spiking neural network chip + +In this study, we use a recently- developed mixed- signal spiking neural network chip, called Dynamic Neuromorphic Asynchronous Processor (DYNAP- SE2) [30]. This chip comprises 1024 adaptive- exponential integrate- and- fire (AdEx I&F) neurons [8, 48], distributed across four cores. The neuron parameters consist of its spiking threshold, refractory period, membrane time constant, and gain. On hardware, the inputs to each neuron are limited to 64 incoming connections, implemented using differential pair integrator (DPI) synapse circuits [49]. Each synaptic input can be configured to express one of four possible temporal dynamics (via four different DPI circuits): AMPA, NMDA, GABA- A, and GABA- B. Besides expressing different time constants, the excitatory synapse circuits (AMPA and NMDA) differ in that the NMDA circuit also incorporates a voltage gating mechanism. Similarly, the inhibitory synapses (GABA- A, and GABA- B) differ in their subtractive (GABA- B) versus shunting (GABA- A) effect. + +The bias parameters that determine the weights, neuronal and synaptic properties are set by an on- chip digital to analog (DAC) bias- generator circuit [50] and are shared globally within each core. These parameters are expressed as currents that drive the neuromorphic circuits and are represented internally through a "coarse" and a "fine" value, according to the following equation: + +\[I = I_{C}(C_{w})\frac{F_{w}}{256}, \quad (1)\] + +where \(F_{w}\) represents the fine value and \(C_{w}\) the coarse one. The variable \(I_{C}\) represents a current estimated from the bias- generator circuit simulations, which depends on \(C_{w}\) as specified in Table 2. + +Before training, the network is initialized with \(C_{w}\) randomly taking a value of 3, 4, or 5, and \(F_{w}\) uniformly distributed between 20 and 200. The corresponding currents will therefore range between \(2.7\mathrm{nA}\) and \(1757\mathrm{nA}\) . However, due to device mismatch and variability in the analog circuits, even when many circuits share the same nominal biases, the effective values will vary. On the DYNAP- SE2 and other mixed- signal neuromorphic processors implemented using similar technology nodes and circuit design styles, the coefficient of variation of + +<--- Page Split ---> + + +
Cw012345
IC(nA)0.070.554.4535.02802250
+ +Table 2: Nominal estimated coarse current values of the on-chip bias generator circuit for synaptic weights. + +these parameters is typically around 20% [10]. However, the learning rules proposed in this work make these networks robust to such variability and lead to a stable activity regime. + +# Network emulation + +We create an on-chip E/I network composed of 200 Pyr (excitatory) and 50 PV (inhibitory) neurons. Each neuron type is implemented by tuning the refractory period, time constants, spiking threshold, and neuron gain parameters to match biologically measured values (Fig. 1c, [33]). The two populations are sparsely connected with a 10% connection probability for each pair of neurons (both, within population and between populations).Synaptic weights take four sets of values shared globally among all Pyr-to-Pyr \((w_{ee})\) , Pyr-to-PV \((w_{ie})\) , PV-to-PV \((w_{ii})\) , and PV-to-Pyr \((w_{ei})\) connections respectively. All excitatory weights represent synapses modeled as AMPA receptors, and inhibitory weights as GABA-A (somatic). Here, \(w_{ii}\) and \(w_{ei}\) are somatic inhibitory connections.Even though the weight parameters are shared, the synaptic efficacy of each synapse is subject to variability due to device mismatch in the analog circuits [10]. + +During training, synaptic weights are adjusted such that the average activity for both Pyr and PV population converge to their respective target firing rates (or set-points), and can thus maintain self-sustained activity in the absence of any external input or bias current. To kickstart activity in the network, we provide a brief 40ms external input of four spikes at 10 ms intervals to 80% of the Pyr cells, with random delays for each postsynaptic target neuron to introduce temporal sparsity to the initial response of the population. The network is emulated in real-time on the chip (i.e., the dynamics of the analog circuits evolve through physical time reproducing the dynamics of the biological neural circuits). Each experiment is run for 1 s and the spiking activity of the whole network is streamed to a computer, for data logging and analysis purposes. + +In the analysis, the network response is measured by calculating the average firing rate of both neural populations, which is then used to calculate the error (i.e., the difference between set-point and actual activity)that drives the weight update (see next section). As we are interested in the network's self-sustained behavior, we discard the response induced by the transient external input while calculating the firing rates (i.e., we ignore the first 60 ms of activity, which comprises the first 40 ms of external stimulation and an additional stabilization window of 20 ms). During the initial stages of the training procedure, before E/I balance is reached, the network could generate non-sustained bursts of activity. Therefore, we calculate the firing rate by considering only the time window when the neural populations are active, i.e. the in-burst firing rate, as opposed to the average over the whole trial. We empirically find that this choice leads to the desired sustained activity. + +# Learning rule + +To bring the recurrent dynamics to a fully self-sustained inhibition-stabilized regime we employ the recently proposed cross-homeostatic plasticity rule [27], which tunes all four synaptic weight classes in parallel. + +In classic forms of homeostatic plasticity, both excitatory and inhibitory populations self-regulate calcium levels based on their own activity [24, 36]. In contrast to these classic forms, where a neuron tunes its weights to maintain its own average output of activity (a firing rate set-point), the cross-homeostatic weight update rule aims at bringing the neuron's presynaptic partners of opposite polarity to an average population firing rate setpoint,according to the following equations: + +\[\begin{array}{l}\Delta w_{ee} = +\alpha \bar {E}(t)(I_{target} - \bar {I}(t)),\\ \Delta w_{ei} = -\alpha \bar {I}(t)(I_{target} - \bar {I}(t)),\end{array}\qquad \begin{array}{l}\Delta w_{ie} = -\alpha \bar {E}(t)(E_{target} - \bar {E}(t)),\\ \Delta w_{ii} = +\alpha \bar {I}(t)(E_{target} - \bar {E}(t)),\end{array}\tag{2}\] + +where \(\alpha\) is the learning rate \((\alpha =0.05)\) , \(\bar {E}(t)\) and \(\bar {I}(t)\) are the average measured firing rates during a trial t and \(E_{target}=20\) Hz, \(I_{target}=40\) Hz are the desired firing rate set-points, for the Pyr and PV populations + +<--- Page Split ---> + +respectively. This type of plasticity can be interpreted biologically as having a target on the input current that a given neuron receives. For example, an excitatory neuron could have indirect access to the average firing rate of their presynaptic inhibitory neurons based on a running average of the total GABA it receives. + +While biologically plausible target activities are typically lower (e.g., \(5\mathrm{Hz}\) for the \(E\) population and \(14\mathrm{Hz}\) for the \(I\) one as proposed in [27]), in our experiments we had to use higher set- point rates to attain reliable behavior, because of the relatively small network size and the low probability of connections used. In Table 1 we quantify the effects of network size and connectivity with additional experiments. + +As the weights are shared globally within each population, the learning rules are implemented at the whole population level. Therefore the network maintains a given firing rate at the population level, rather than at the individual neuron level (see [27] for a local implementation of the rules). However, the heterogeneity in the analog circuits and their inherent noise produce diverse behaviors among neurons, for example leading some of the neurons to fire intermittently at slightly increased rates (while maintaining the desired set- point firing rate at the full population level). In the following section, we explain the training procedure on the hardware and how the changes in weight \(\Delta w\) are translated to the fine \(F_{w}\) and course \(C_{w}\) synaptic bias parameters on chip. + +## Chip-in-the-loop training + +Although the DYNAP- SE2 chip can be tuned to exhibit biologically plausible dynamics, it does not incorporate on- chip synaptic plasticity circuits that are capable of learning. To overcome this limitation we used a "chip- in- the- loop" learning paradigm: the activity produced by the chip is transmitted to a computer, which analyzes the data at the software level and calculates the parameter updates, which are then sent back to the chip. + +In particular, the network is emulated for 1 s on the chip. Each emulation is repeated five times to reduce the influence of noise and obtain more reliable results. The output spikes are sent back to the computer, which averages the results, calculates the appropriate weight update, applies stochastic rounding (see below), and sends the updated weights to the chip for the next iteration (i.e., after 5 repetitions). After each iteration, we drain the accumulated residual current from both the neuron and synapse circuits. We stop the procedure after 400 iterations, as we found it to be sufficient for the network to converge to the target firing rates regardless of the initial conditions. Most simulations converge to target firing rate before iteration 200, which shows the long- term stability of the cross- homeostasis rules in a converged network. + +For all experiments, we applied the same set of parameters to neurons and synapses at the beginning of the training, with the exception of the random initialization of weights. The training produced stable and reliable activity around the programmed set- points \(100\%\) of the time, without having to re- tune any of the neuron or synapse parameters (except the synaptic weights), across multiple days, and for multiple chips. + +Stochastic rounding As the values of the synaptic weight parameters are set by the bias- generator DAC, the weight updates can only take discrete values \(\Delta F_{w} = 0,\pm 1,\pm 2,\dots\) . For very large learning rates, this could lead to unstable learning, and for very low ones to no changes at all. To overcome the constraints of limited- resolution weight updates, a common strategy used is to use stochastic rounding (SR) [51, 52]. The SR technique consists of interpreting the \(\Delta w\) calculated by the learning rule as a probability to increase or decrease the weight, as follows: + +\[\mathrm{SR}(x) = \left\{ \begin{array}{ll}[x] & \mathrm{with~prob.~}x - \lfloor x\rfloor \\ \lfloor x\rfloor & \mathrm{otherwise.} \end{array} \right. \quad (3)\] + +where \([\cdot ]\) indicates the ceiling operation and \(\lfloor \cdot \rfloor\) the floor operation. + +Values for the desired weight updates \(\Delta w\) are given by the learning rules (2). We also define upper and lower bounds for \(F_{w}\) , respectively \(F_{- } = 20\) and \(F_{+} = 250\) . We then obtain the new coarse and fine values \(C_{w}^{\prime},F_{w}^{\prime}\) of the weight as follows: + +\[\left\{ \begin{array}{ll}C_{w}^{\prime} = C_{w}, & F_{w}^{\prime} = F_{w} + \mathrm{SR}(\Delta w) & \mathrm{if} F_{-}\leq F_{w} + \mathrm{SR}(\Delta w)\leq F_{+} \\ C_{w}^{\prime} = C_{w} - 1, & F_{w}^{\prime} = F_{+} & \mathrm{if} F_{w} + \mathrm{SR}(\Delta w)< F_{-} \\ C_{w}^{\prime} = C_{w} + 1, & F_{w}^{\prime} = F_{-} & \mathrm{if} F_{w} + \mathrm{SR}(\Delta w) > F_{+}. \end{array} \right. \quad (4)\] + +In other words, the fine value is updated with the stochastically- rounded version of \(\Delta w\) ; when it oversteps the bound, the fine value is reset to the bound, and the coarse value is changed instead. + +<--- Page Split ---> + +Note that this SR strategy is compatible with neuromorphic hardware implementations, as demonstrated in [53]. Our setup using a chip- in- the- loop enables the evaluation of different learning mechanisms and test the robustness of different design choices. The design of future on- chip learning circuit can therefore be informed by our results, following a hardware- software co- design approach. + +## Author Contributions + +Conceptualization: M, SSM and RL. Investigation: M, MS and SSM. Methodology, Software, Data curation and Visualization: M and MS. Supervision SSM, RL, DB and GI. Writing Original Draft: M and MS. Writing and editing: SSM, RL, DB and GI. Funding acquisition: SSM, MS, DB and GI. + +## Competing Interests + +The authors declare that they have no conflict of interest. + +## Acknowledgments + +We thank the organizers of the CapoCaccia Workshop Towards Neuromorphic Intelligence 2022, where this work was conceived. We thank Chenxi Wu, Ole Richter and German Koestinger for DYNAP- SE2 support, and Matthew Cook for useful discussions. SSM was supported by the Swiss National Science Foundation (SNSF) grant nos. P2ZHP3- 187943 and P500PB- 203133. M was supported by the SNSF Sinergia Project (CRSII5- 180316). The work of MS was made possible by a postdoctoral fellowship from the ETH AI Center. GI was supported by the EU ERC Grant "NeuroAgents" (No. 724295), DVB and RL were supported by the National Institutes of Health grant no. NS116589. RL was supported by Universidad Nacional de Quilmes (Argentina), CONICET (Argentina), and NeurotechEU (Faculty Scholarship). + +## References + +[1] Guillaume Hennequin, Tim P Vogels, and Wulfram Gerstner. "Optimal control of transient dynamics in balanced networks supports generation of complex movements". In: Neuron 82.6 (2014), pp. 1394- 1406. [2] Dean V Buonomano and Wolfgang Maass. "State- dependent computations: spatiotemporal processing in cortical networks". In: Nature Reviews Neuroscience 10.2 (2009), pp. 113- 125. [3] Saurabh Vyas, Matthew D Golub, David Sussillo, and Krishna V Shenoy. "Computation through neural population dynamics". In: Annual review of neuroscience 43 (2020), pp. 249- 275. [4] Rodney J. Douglas and Kevan A.C. Martin. "Recurrent neuronal circuits in the neocortex". In: Current Biology 17.13 (2007), R496- R500. doi: 10.1016/j.cub.2007.04.024. [5] Rodney J. Douglas, Christof Koch, Misha Mahowald, Kevan A. C. Martin, and Humbert H. Suarez. "Recurrent Excitation in Neocortical Circuits". In: Science 269.5226 (1995), pp. 981- 985. doi: 10.1126/science.7638624. [6] Xiao- Jing Wang. "Synaptic reverberation underlying mnemonic persistent activity". In: Trends in neurosciences 24.8 (2001), pp. 455- 463. [7] Carver Mead. "Neuromorphic Engineering: In Memory of Misha Mahowald". In: Neural Computation 35 (2023), pp. 343- 383. doi: 10.1162/neco_a_01553. [8] E. Chicca, F. Stefanini, C. Bartolozzi, and G. Indiveri. "Neuromorphic electronic circuits for building autonomous cognitive systems". In: Proceedings of the IEEE 102.9 (Sept. 2014), pp. 1367- 1388. issn: 0018- 9219. doi: 10.1109/JPROC.2014.2313954. [9] C. Mead. "Neuromorphic Electronic Systems". In: Proceedings of the IEEE 78.10 (1990), pp. 1629- 36. doi: 10.1109/5.58356. + +<--- Page Split ---> + +[10] Dmitrii Zendrikov, Sergio Solinas, and Giacomo Indiveri. "Brain-inspired methods for achieving robust computation in heterogeneous mixed- signal neuromorphic processing systems". In: Neuromorphic Computing and Engineering 3.3 (July 2023), p. 034002. issn: 2634- 4386. doi: 10.1088/2634- 4386/ac6c4c. + +[11] Simon B. Laughlin and Terrence J. Sejnowski. "Communication in Neuronal Networks". In: Science 301.5641 (2003), pp. 1870- 1874. doi: 10.1126/science.1089662. + +[12] U. Rutishauser and R. Douglas. "State- Dependent Computation Using Coupled Recurrent Networks". In: Neural Computation 21 (2009), pp. 478- 509. + +[13] E. Neftci et al. "Synthesizing cognition in neuromorphic electronic systems". In: Proceedings of the National Academy of Sciences 110.37 (2013), E3468- E3476. + +[14] M. Giulioni et al. "Robust working memory in an asynchronously spiking neural network realized in neuromorphic VLSI". In: Frontiers in Neuroscience 5.149 (2012). issn: 1662- 453X. doi: 10.3389/fnins.2011.00149. + +[15] T. Pfeil et al. "Six networks on a universal neuromorphic computing substrate". In: Frontiers in Neuroscience 7 (2013). issn: 1662- 453X. doi: 10.3389/fnins.2013.00011. + +[16] Mircea Steriade, David A. McCormick, and Terrence J. Sejnowski. "Thalamocortical oscillations in the sleeping and aroused brain". In: Science 262.5134 (1993), pp. 679- 685. + +[17] DAVID A McCormick. "GABA as an inhibitory neurotransmitter in human cerebral cortex". In: Journal of neurophysiology 62.5 (1989), pp. 1018- 1027. + +[18] Misha V Tsodyks, William E Skaggs, Terrence J Sejnowski, and Bruce L McNaughton. "Paradoxical effects of external modulation of inhibitory interneurons". In: Journal of neuroscience 17.11 (1997), pp. 4382- 4388. + +[19] Hirofumi Ozeki, Ian M Finn, Evan S Schaffer, Kenneth D Miller, and David Ferster. "Inhibitory stabilization of the cortical network underlies visual surround suppression". In: Neuron 62.4 (2009), pp. 578- 592. + +[20] Alessandro Sanzeni et al. "Inhibition stabilization is a widespread property of cortical networks". In: Elife 9 (2020), e54875. + +[21] Sadra Sadeh and Claudia Clopath. "Inhibitory stabilization and cortical computation". In: Nature Reviews Neuroscience 22.1 (2021), pp. 21- 37. + +[22] Alain Destexhe, Stuart W Hughes, Michelle Rudolph, and Vincenzo Crunelli. "Are corticothalamic 'up' states fragments of wakefulness?" In: Trends in neurosciences 30.7 (2007), pp. 334- 342. + +[23] G.G. Turrigiano and S.N. Nelson. "Homeostatic plasticity in the developing nervous system". In: Nature Reviews Neuroscience 5 (Feb. 2004), pp. 97- 107. + +[24] Gina Turrigiano. "Homeostatic synaptic plasticity: local and global mechanisms for stabilizing neuronal function". In: Cold Spring Harbor perspectives in biology 4.1 (2012), a005736. + +[25] Guillaume Hennequin, Everton J. Agnes, and Tim P. Vogels. "Inhibitory Plasticity: Balance, Control, and Codependence". In: Annual Review of Neuroscience 40 (2017), pp. 557- 579. issn: 15454126. doi: 10.1146/annurev-neuro- 072116- 031005. + +[26] Robert C Froemke. "Plasticity of cortical excitatory- inhibitory balance". In: Annual review of neuroscience 38 (2015), pp. 195- 219. + +[27] Saray Soldado- Magraner, Michael J Seay, Rodrigo Laje, and Dean V Buonomano. "Paradoxical self- sustained dynamics emerge from orchestrated excitatory and inhibitory homeostatic plasticity rules". In: Proceedings of the National Academy of Sciences 119.43 (2022), e2200621119. + +[28] Bipin Rajendran, Damien Querlioz, Sabina Spiga, and Abu Sebastian. "Memristive devices for spiking neural networks". In: Memristive Devices for Brain- Inspired Computing. Elsevier, 2020, pp. 399- 405. + +[29] Dennis Valbjorn Christensen et al. "2022 roadmap on neuromorphic computing and engineering". In: Neuromorphic Computing and Engineering (2022). doi: 10.1088/2634- 4386/ac4a83. + +[30] Ole Richter et al. DYNAP- SE2: a scalable multi- core dynamic neuromorphic asynchronous spiking neural network processor. 2023. doi: 10.48550/arXiv.2310.00564. + +<--- Page Split ---> + +[31] S.- C. Liu, J. Kramer, G. Indiveri, T. Delbruck, and R.J. Douglas. Analog VLSI: Circuits and Principles. MIT Press, 2002. + +[32] Brielle R. Ferguson and Wen- Jun Gao. "PV Interneurons: Critical Regulators of E/I Balance for Prefrontal Cortex- Dependent Behavior and Psychiatric Disorders". In: Frontiers in Neural Circuits 12 (2018), p. 37. DOI: 10.3389/fncir.2018.00037. + +[33] Juan L Romero- Sosa, Helen Motanis, and Dean V Buonomano. "Differential excitability of PV and SST neurons results in distinct functional roles in inhibition stabilization of up states". In: Journal of Neuroscience 41.34 (2021), pp. 7182- 7196. + +[34] Arvind Kumar, Sven Schrader, Ad Aertsen, and Stefan Rotter. "The high- conductance state of cortical networks". In: Neural computation 20.1 (2008), pp. 1- 43. + +[35] Alfonso Renart et al. "The asynchronous state in cortical circuits". In: science 327.5965 (2010), pp. 587- 590. + +[36] Gina G Turrigiano and Sacha B Nelson. "Homeostatic plasticity in the developing nervous system". In: Nature reviews neuroscience 5.2 (2004), pp. 97- 107. + +[37] Sandra J Kuhlman et al. "A disinhibitory microcircuit initiates critical- period plasticity in the visual cortex". In: Nature 501.7468 (2013), pp. 543- 546. + +[38] Zhengyu Ma, Gina G Turrigiano, Ralf Wessel, and Keith B Hengen. "Cortical circuit dynamics are homeostatically tuned to criticality in vivo". In: Neuron 104.4 (2019), pp. 655- 664. + +[39] Tomoki Fukai and Shigeru Tanaka. "A simple neural network exhibiting selective activation of neuronal ensembles: from winner- take- all to winners- share- all". In: Neural computation 9.1 (1997), pp. 77- 97. + +[40] Jordan P Hamm, Yuriy Shymkiv, Shuting Han, Weijian Yang, and Rafael Yuste. "Cortical ensembles selective for context". In: Proceedings of the National Academy of Sciences 118.14 (2021), e2026179118. + +[41] G. Indiveri and S.- C. Liu. "Memory and information processing in neuromorphic systems". In: Proceedings of the IEEE 103.8 (2015), pp. 1379- 1397. DOI: 10.1109/JPROC.2015.2444094. + +[42] Adnan Mehonic and Anthony J Kenyon. "Brain- inspired computing: We need a master plan". In: Nature 604.7905 (2022), pp. 255- 260. DOI: 10.1038/s41586- 021- 04362- w. + +[43] Giacomo Indiveri and Yulia Sandamirskaya. "The importance of space and time for signal processing in neuromorphic agents". In: IEEE Signal Processing Magazine 36.6 (2019), pp. 16- 28. DOI: 10.1109/MSP.2019.2928376. + +[44] Wen Sun et al. "Understanding memristive switching via in situ characterization and device modeling". In: Nature communications 10.1 (2019), pp. 1- 13. DOI: 10.1038/s41467- 019- 11411- 6. + +[45] Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. "Context- dependent computation by recurrent dynamics in prefrontal cortex". In: Nature 503.7474 (2013), pp. 78- 84. issn: 00092665. DOI: 10.1038/nmeth.2835. Multiplexed. + +[46] Vishwa Goudar and Dean V Buonomano. "Encoding sensory and motor patterns as time- invariant trajectories in recurrent neural networks". In: Elife 7 (2018), e31134. + +[47] Brian DePasquale, Christopher J Cueva, Kanaka Rajan, G Sean Escola, and LF Abbott. "full- FORCE: A target- based method for training recurrent networks". In: PloS one 13.2 (2018), e0191527. + +[48] G. Indiveri et al. "Neuromorphic silicon neuron circuits". In: Frontiers in Neuroscience 5 (2011), pp. 1- 23. issn: 1662- 453X. DOI: 10.3389/fnins.2011.00073. + +[49] C. Bartolozzi and G. Indiveri. "Synaptic dynamics in analog VLSI". In: Neural Computation 19.10 (Oct. 2007), pp. 2581- 2603. DOI: 10.1162/neco.2007.19.10.2581. + +[50] T. Delbruck and A. Van Schaik. "Bias Current Generators with Wide Dynamic Range". In: Analog Integrated Circuits and Signal Processing 43.3 (2005), pp. 247- 268. + +[51] Lorenz K Muller and Giacomo Indiveri. "Rounding methods for neural networks with low resolution synaptic weights". In: arXiv preprint arXiv:1504.05767 (2015), pp. 1- 11. + +<--- Page Split ---> + +[52] Matteo Croci, Massimiliano Fasi, Nicholas J Higham, Theo Mary, and Mantas Mikaitis. “Stochastic rounding: implementation, error analysis and applications”. In: Royal Society Open Science 9.3 (2022), p. 211631. [53] Matteo Cartiglia et al. “Stochastic dendrites enable online learning in mixed-signal neuromorphic processing systems”. In: 2022 IEEE International Symposium on Circuits and Systems (ISCAS) (May 2022). doi: 10.1109/iscas48785.2022.9937833. + +<--- Page Split ---> diff --git a/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4_det.mmd b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a9039ef78b0058ec82cb283feb6b0256e87163b9 --- /dev/null +++ b/preprint/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4/preprint__1340fc445c5c5e83cf45d6deb62f7e51a44088cbf8510934a1ce63e2f09036b4_det.mmd @@ -0,0 +1,474 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 866, 209]]<|/det|> +# Orchestrated excitatory and inhibitory plasticity produces stable dynamics in heterogeneous neuromorphic computing systems + +<|ref|>text<|/ref|><|det|>[[44, 230, 256, 275]]<|/det|> +Maryada Maryada maryada@ini.uzh.ch + +<|ref|>text<|/ref|><|det|>[[44, 303, 730, 560]]<|/det|> +Institute of Neuroinformatics, University of Zurich and ETH Zurich Saray Soldado- Magraner University of California, Los Angeles https://orcid.org/0000- 0003- 4890- 2671 Martino Sorbaro ETH AI Center, ETH Zurich https://orcid.org/0000- 0002- 0182- 7443 Rodrigo Laje University of Quilmes https://orcid.org/0000- 0003- 1481- 303X Dean Buonomano University of California, Los Angeles https://orcid.org/0000- 0002- 8528- 9231 Giacomo Indiveri SynSense AG Corporation https://orcid.org/0000- 0002- 7109- 1689 + +<|ref|>sub_title<|/ref|><|det|>[[44, 595, 103, 613]]<|/det|> +## Article + +<|ref|>sub_title<|/ref|><|det|>[[44, 633, 136, 651]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 670, 328, 690]]<|/det|> +Posted Date: October 20th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 709, 475, 728]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3449716/v1 + +<|ref|>text<|/ref|><|det|>[[44, 746, 914, 789]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 807, 535, 826]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 862, 950, 905]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 60697- 2. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[128, 120, 870, 171]]<|/det|> +# Orchestrated excitatory and inhibitory plasticity produces stable dynamics in heterogeneous neuromorphic computing systems + +<|ref|>text<|/ref|><|det|>[[196, 193, 800, 227]]<|/det|> +Maryada \(^{1}\) , Saray Soldado- Magraner \(^{2}\) , Martino Sorbaro \(^{1,3}\) , Rodrigo Laje \(^{4,5}\) , Dean V. Buonomano \(^{2}\) , and Giacomo Indiveri \(^{1}\) + +<|ref|>text<|/ref|><|det|>[[118, 240, 880, 325]]<|/det|> +\(^{1}\) Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland \(^{2}\) Department of Neurobiology, University of California, Los Angeles, USA \(^{3}\) ETH AI Center, ETH Zurich, Switzerland \(^{4}\) Department of Science and Technology, Universidad Nacional de Quilmes, Bernal, Argentina \(^{5}\) CONICET, Argentina + +<|ref|>sub_title<|/ref|><|det|>[[466, 366, 530, 380]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[138, 386, 860, 615]]<|/det|> +Many neural computations emerge from self- sustained patterns of activity in recurrent neural circuits, which rely on balanced excitation and inhibition. Neuromorphic electronic circuits that use the physics of silicon to emulate neuronal dynamics represent a promising approach for implementing the brain's computational primitives, including self- sustained neural activity. However, achieving the same robustness of biological networks in neuromorphic computing systems remains a challenge, due to the high degree of heterogeneity and variability of their analog components. Inspired by the strategies used by real cortical networks, we apply a biologically- plausible cross- homeostatic learning rule to balance excitation and inhibition in neuromorphic implementations of spiking neural networks. We demonstrate how this learning rule allows the neuromorphic system to overcome device mismatch and to autonomously tune the spiking network to produce robust, self- sustained attractor dynamics in an inhibition- stabilized regime. We also show that this rule can implement a stable working memory, and that the electronic circuits can reproduce biologically relevant emergent neural dynamics, including the so- called "paradoxical effect". In addition to validating neuroscience models on a substrate that shares many similar properties and limitations with biological systems, this work enables the construction of ultra- low power, mixed- signal neuromorphic technologies that can be automatically configured to compute reliably, despite the large on- chip and chip- to- chip variability of their analog components. + +<|ref|>sub_title<|/ref|><|det|>[[93, 637, 225, 656]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[92, 668, 905, 895]]<|/det|> +Animal brains can perform complex computations including sensory processing, motor control, and working memory, as well as higher cognitive functions, such as decision- making and reasoning, in an efficient and reliable manner. At the neural network level, these processes are implemented using a variety of computational primitives that rely on neural dynamics within recurrent neocortical microcircuits [1- 6]. Translating the computational primitives observed in the brain into novel technologies can potentially lead to radical innovations in artificial intelligence and edge- computing applications. A promising technology that can implement these primitives with compact and low- power devices is that of mixed- signal neuromorphic systems, which employ analog electronic circuits to emulate the biophysics of real neurons [7- 9]. As opposed to software simulations, the direct emulation performed by these electronic circuits relies on the use of the physics of the silicon substrate to faithfully reproduce the dynamics of neurons and synapses asynchronously in real time. While systems designed following this approach have the advantage of ultra- low power consumption, they have limitations and constraints similar to those found in biology. These include a high degree of variability, heterogeneity, and sensitivity to noise [10, 11]. Due to these constraints, implementing recurrent networks in silicon with the same stability and robustness observed in biological circuits remains an open challenge. + +<|ref|>text<|/ref|><|det|>[[92, 895, 904, 927]]<|/det|> +Previous studies have already demonstrated neuromorphic implementations of recurrent computational primitives such as winner- take- all or state- dependent networks [12- 15]. However, these systems required + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 75, 905, 188]]<|/det|> +manual tuning and exhibited dynamics that would not always produce stable self- sustained regimes. Manually tuning on- chip recurrent networks with large numbers of parameters is challenging, tedious, and not scalable. Furthermore, because of cross- chip variability, this process requires re- tuning for each individual chip. Even when a stable regime is achieved with manual tuning, changes in the network conditions (e.g., due to temperature variations, increase or decrease of input signals, etc.) would require re- tuning. Therefore, mixed- signal neuromorphic technology can greatly benefit from automatic stabilizing and tuning mechanisms that can overcome the challenges imposed by the analog substrate. + +<|ref|>text<|/ref|><|det|>[[92, 189, 905, 396]]<|/det|> +Neocortical computations rely heavily on positive feedback imposed by recurrent connections between excitatory neurons, which allows networks to perform complex time- dependent computations and actively maintain information about past events. Recurrent excitation, however, also makes neocortical circuits vulnerable to "runaway excitation" and epileptic activity [16, 17]. In order to harness the computational power of recurrent excitation and avoid pathological regimes, the neocortical circuits operate in an inhibition- stabilized regime, in which positive feedback is held in check by recurrent inhibition [18- 21]. Indeed, there is evidence that the default awake cortical dynamic regime may be inhibition- stabilized [20, 22]. It is generally accepted that neocortical microcircuits have synaptic learning rules in place to homeostatically balance excitation and inhibition in order to generate dynamic regimes capable of self- sustained activity [23- 26]. Recently, a family of synaptic learning rules that differentially operate at excitatory and inhibitory synapses has been proposed, which can drive simulated neural networks to self- sustained and inhibition- stabilized regimes in a self- organizing manner [27]. This family of learning rules is referred to as being "cross- homeostatic" because synaptic plasticity at excitatory and inhibitory synapses is dependent on both the inhibitory and excitatory setpoints. + +<|ref|>text<|/ref|><|det|>[[92, 397, 905, 444]]<|/det|> +Taking inspiration from self- calibration principles in cortical networks, we use these cross- homeostatic plasticity rules to guide neuromorphic circuits to balanced excitatory- inhibitory regimes in a self- organizing manner. + +<|ref|>text<|/ref|><|det|>[[92, 445, 905, 606]]<|/det|> +We show that these rules can be successfully employed to autonomously calibrate analog spiking recurrent networks in silicon. Specifically, by automatically tuning all synaptic weight classes in parallel, the networks of silicon neurons converge to fully self- sustained attractor dynamics, and configure themselves into the inhibition- stabilized regime. In addition, the emergent neural dynamics obtained in this way express the "paradoxical effect", a signature of inhibition- stabilized networks, which is observed in cortical circuits [18, 20]. The plasticity rules proposed prove resilient to hardware variability and noise, also across different chips, as well as to different parameter initializations. Importantly, we demonstrate that inhibitory plasticity (often neglected in neuromorphic electronic systems) is necessary for successful convergence. Finally, we also show that by using such plasticity rules, long- term memories can be implemented without disrupting the learned dynamics, and that multiple such networks can be implemented on a single chip robustly. + +<|ref|>text<|/ref|><|det|>[[92, 607, 905, 718]]<|/det|> +From a computational neuroscience perspective, these results validate the robustness of cross- homeostatic plasticity in a physical substrate that presents similar challenges to those of biological networks, unlike idealized digital simulations. From a neuromorphic perspective, this approach provides the community with a reliable method to autonomously and robustly calibrate recurrent neural networks in future mixed- signal analog/digital systems that are affected by device variability (also including memristive devices [28, 29]). This will open the door to designing autonomous systems that can interact in real time with the environment, and compute reliably with low- latency at extremely low- power using attractor dynamics, as observed in biological circuits. + +<|ref|>sub_title<|/ref|><|det|>[[93, 741, 169, 759]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[92, 772, 905, 869]]<|/det|> +In the cortex, excitatory Pyramidal (Pyr) and inhibitory Parvalbumin- positive interneurons (PV) constitute the main neuronal subtypes and are primarily responsible for excitatory/inhibitory (E/I) balance [32]. Pyr and PV neurons have different intrinsic biophysical properties: in terms of excitability, PV cells have a higher threshold and gain compared to Pyr neurons [33]. These properties, along with other neuronal characteristics, such as refractory period and membrane time constants, contribute to the characteristic regular- spiking patterns of pyramidal cells and fast- spiking patterns of PV interneurons. + +<|ref|>text<|/ref|><|det|>[[92, 870, 905, 917]]<|/det|> +In our hardware experiments, we reproduce the properties of both Pyr and PV neuron types on a mixed- signal analog/digital Dynamic Asynchronous Neuromorphic Processor, called DYNAP- SE2 [30] (see Fig. 1a), which implements an adaptive exponential integrate- and- fire neuronal model. The dynamics of the different neuron + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[103, 71, 905, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 550, 905, 793]]<|/det|> +
Figure 1: (a) Image of the DYNAP-SE2 chip, showing the four cores. An E-I network is implemented on chip. Excitatory (Pyr) and Inhibitory (PV) neurons occupy different cores, and are connected by four weight classes, Pyr-to-Pyr \((w_{ee})\) , Pyr-to-PV \((w_{ie})\) , PV-to-PV \((w_{ii})\) , and PV-to-Pyr \((w_{ei})\) . (b) Circuit diagram depicting a DPI neuron, with each component within the circuit replicating various sub-threshold characteristics (such as adaptation, leakage, refractory period, etc.) inherent to an adaptive exponential neuron [30]. (c) Neuron membrane potential traces over time as a response to DC injection. Blue traces show two example cells from the excitatory chip core in response to four current levels. Red traces show two sample PV cells from the inhibitory chip core. The relationship between the measured voltage and the internal current-mode circuit output that represents the membrane potential is given by an exponential function that governs the subthreshold transistor operation mode [31]. In each case, the bottom two traces show neuronal dynamics in the subthreshold range. The top two traces show neuronal dynamics as the membrane potential surpasses the spike threshold (spike added as grey line to indicate when the voltage surpasses threshold). DC injection values are indicated on every trace. (d) Input-response curves for the resulting Pyr and PV neurons on-chip. Inset: zoomed-in view at low firing rates. In (c), it is evident that a higher input current was required to elicit spikes in PV neurons. This phenomenon can be attributed to the higher threshold of PV neurons.
+ +<|ref|>text<|/ref|><|det|>[[92, 817, 905, 914]]<|/det|> +classes were reproduced by tuning the refractory period, time constants, spiking threshold, and neuron gain parameters (Fig. 1b) to match the corresponding biologically measured values. Since parameters are shared by all neurons in a chip core, we assign each neuron population type to a different core (Fig. 1a). Even though all neurons in a core share the same parameters, variations in their intrinsic characteristics arise as a result of analog chip mismatch. For instance, one can observe differences between two excitatory or inhibitory neurons recorded within the same core, as illustrated in Fig. 1c. Overall, the neural dynamics and Input- Frequency (FI) curve of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 73, 904, 139]]<|/det|> +recorded excitatory and inhibitory neurons on chip (Fig. 1c- d show that the choice of parameters used for the silicon neurons is compatible with the behavior of biological PV and Pyr cells [33]. In the following section, we employ cross- homeostatic plasticity [27] to self- tune the connections between PV and Pyr populations so that spontaneous activity in the inhibition- stabilized regime emerges in the network. + +<|ref|>sub_title<|/ref|><|det|>[[92, 157, 576, 175]]<|/det|> +## The network converges to stable self-sustained dynamics + +<|ref|>text<|/ref|><|det|>[[92, 183, 905, 248]]<|/det|> +As an overview of the experimental paradigm, the training procedure begins with random initialization of the network weights. Then, at each iteration, we record the resulting firing rates, calculate the weight update following the cross- homeostatic equations for the four weight classes \((w_{ee},w_{ei},w_{ie},\) and \(w_{ii}\) , equation 2, see Methods) and apply the new weight values to the chip. + +<|ref|>text<|/ref|><|det|>[[92, 248, 905, 408]]<|/det|> +Fig. 2a illustrates an example of the evolution of firing activity during the course of training for Pyr and PV cells. Starting from a random value determined by the initialization of weights, the average rates converge to their target values. Fig. 2b plots the evolution of weights during training. On chip, the weights are controlled by two parameters, a coarse \(C_w\) and a fine \(F_w\) value (See Methods, eq. 4). On most iterations, only the fine value \(F_w\) is updated, but updates to the coarse value \(C_w\) occur occasionally. Even though the coarse updates cause abrupt jumps in the weight value (due to the nature of the bias- generator circuit implementation), the cross- homeostatic rule is robust enough to guide the network to a stable regime. Fig. 2c shows a raster plot from the network in Fig. 2a- b at a converged state. It should be emphasized that, while an initial "kick" was provided to the network to engage recurrent activity (40 ms at \(100\mathrm{Hz}\) ), no further external input was introduced during the trial; hence, any observed spiking activity originates entirely from internal mechanisms. + +<|ref|>text<|/ref|><|det|>[[92, 408, 905, 473]]<|/det|> +When repeating the training process from different initial conditions of randomly sampled weights, the weights converge to different values, which however produce the same desired network average firing rate [27]. Fig. 2d illustrates the rate space with multiple initialization and convergence to respective target set- points (dashed lines). The root mean square error from the target is \(1.02\mathrm{Hz}\) for \(\bar{E}\) and \(6.32\mathrm{Hz}\) for \(\bar{I}\) . + +<|ref|>text<|/ref|><|det|>[[92, 473, 905, 618]]<|/det|> +The converged weight values are approximately aligned to a linear manifold (Fig. 2e), where the sets of excitatory and inhibitory weights \(w_{ee},w_{ei}\) and \(w_{ie},w_{ii}\) , are correlated (correlation coefficients and p- values are as follows: \(w_{ee}\) and \(w_{ei}\) : \(\rho_{green} = 0.81\) and \(p = 1.55\cdot 10^{- 06}\) ; \(\rho_{orange} = 0.88\) , \(p = 6.6\cdot 10^{- 12}\) ; \(w_{ie}\) and \(w_{ii}\) : \(\rho_{green} = 0.63\) \(p = 0.0009\) ; \(\rho_{orange} = 0.75\) , \(p = 3.9\cdot 10^{- 07}\) , where green and orange refer to two different chips). This is well aligned with the theoretical solution derived for rate E- I networks at their set- points, when the neuron transfer function is linear or threshold- linear [27, eqn. 4- 5]. In our case, the linear transfer function assumption does not hold, because the AdEx silicon neuron models saturate at a rate that is inversely proportional to the refractory period parameter. However, the linear approximation holds well in the region of operation forced by the learning rule, set at relatively low firing rates (see Fig. 1d). + +<|ref|>sub_title<|/ref|><|det|>[[92, 636, 770, 654]]<|/det|> +## The network is in an inhibition-stabilized asynchronous-irregular firing regime + +<|ref|>text<|/ref|><|det|>[[92, 661, 905, 840]]<|/det|> +Next we investigated the properties of the final network dynamics after learning. Fig. 2c shows a sample of activity of the converged network, as recorded from the chip, in which the excitatory and inhibitory populations exhibit asynchronous- irregular activity patterns. The minimal correlations among neurons (averages \(\langle \rho_{pyr}\rangle = 0.037\) , \(\langle \rho_{pv}\rangle = 0.069\) ) serve as evidence of their asynchronous activity (Fig. 2f, left). Irregularity is shown by the coefficient of variation ( \(\mathrm{CV}^2\) ) of most neurons being close to 1 (averages \(\langle CV_{pyr}^2\rangle = 0.949\) , \(\langle CV_{pv}^2\rangle = 0.945\) ) (Fig. 2f, right). The \(\mathrm{CV}^2\) is computed as the variance of the inter- spike interval divided by its squared mean, and equals 0 for perfectly regular firing, and 1 for Poisson firing [34]. It is worth noting that during the 1- second simulations, we do find instances of synchronization, which can be attributed to the small population size of the network and increased firing rate in comparison to the baseline spontaneous activity. As an average however, we conclude that cross- homeostatic plasticity brings the network to an asynchronous- irregular firing regime, what is typical in realistic cortical networks [35]. + +<|ref|>text<|/ref|><|det|>[[92, 840, 905, 920]]<|/det|> +As further evidence of the inhibition- stabilized regime of the network, we demonstrate the occurrence of the "paradoxical effect" in our deployed network on chip [18, 20]. The paradoxical effect is a hallmark of inhibition- stabilized networks that has been observed in cortical circuits. When inhibitory neurons are excited (either optogenetically in vivo, or via an external current in computational models) their firing rates show a paradoxical decrease in activity during the stimulation, providing clear evidence that that the network is in an + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 78, 920, 730]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 735, 907, 912]]<|/det|> +
Figure 2: (a) A representative experiment of the convergence of excitatory (red) and inhibitory (blue) firing rates during learning. The gray dashed lines show the desired set-point targets. (b) A representative experiment (the same as in panel a) of the convergence of weight values during learning. The plotted value is nominal, inferred from eq. (1) (see Methods). (c) Raster plot exemplifying on-chip firing activity during a single emulation run at the end of training. Both Pyr and PV neurons successfully converge to an asynchronous-irregular firing pattern at the desired FRs. (d) Initial and final firing rates across different chips (n=57, across two chips) and different initial conditions. For each initial condition, network connectivity was randomized. (e) Left: relationship between the final \(w_{ee}\) and \(w_{ei}\) values across different iterations. Orange and green colors correspond to two different chips. Right: same for \(w_{ii}\) and \(w_{ie}\) . (f) Distributions for the 5 ms correlation coefficients between neurons (left) and coefficient of variation ( \(CV^2\) ) (right), demonstrating low regularity and low synchrony. The line indicates the mean. Shaded region indicate standard deviation.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 74, 905, 204]]<|/det|> +inhibition- stabilized regime (Fig. 3a). Analogously, when we inject an external depolarizing pulse to all PV neurons on chip for \(0.2\mathrm{s}\) we observe a decrease in the firing rate of both cell populations for the duration of the stimulus, returning back to the target FR when the stimulus ends (Fig. 3b). Across all 14 iterations we use pairwise t- tests to assess the impact of stimulating PV neurons. For PV neurons, stimulation significantly reduces activity during stimulation compared to pre- and post- stimulation states \((p< 10^{- 18}\) in both cases). Conversely, there is no significant difference between pre- and post- stimulation activity, indicating recovery. Five trials, which were excluded from the analysis, do not exhibit recovery because the activity is entirely suppressed by the strength of the paradoxical effect. A similar trend is observed for Pyramidal neurons \((p< 10^{- 20})\) . + +<|ref|>image<|/ref|><|det|>[[115, 223, 888, 466]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 479, 906, 640]]<|/det|> +
Figure 3: (a). The paradoxical effect is a well known phenomena in cortical circuits. In the awake resting cortex of mice (upper figure), when inhibitory neurons are optogenetically activated, their firing rates (in red) show a paradoxical decrease in activity during the stimulation, indicative of an inhibition-stabilized network. Adapted from Sanzeni et al. [21]. Similar to the experimental case (bottom figure), numerical simulations of firing rate models with sufficient excitatory gain and balanced by inhibition show the paradoxical effect when the inhibitory population is excited via an external current. (b). The paradoxical effect can be demonstrated in analog neuromorphic circuits by applying a depolarizing current to the inhibitory units of networks converged to self-sustained activity via cross-homeostasis. The resulting decrease in the firing rate of the inhibitory units demonstrates that the on-chip network is in the inhibition-stabilized regime. No. of trials \(= 14\) , shading region represents standard deviation in firing rate across trials.
+ +<|ref|>sub_title<|/ref|><|det|>[[92, 674, 640, 692]]<|/det|> +## Inhibitory plasticity is necessary to achieve reliable convergence + +<|ref|>text<|/ref|><|det|>[[92, 699, 905, 860]]<|/det|> +The equations for the weights in absence of noise [27, eqn. 4- 5] and the experimental results in Fig. 2 e indicate that two of the four weight parameters, for example, the inhibitory weights \(w_{ii}\) and \(w_{ei}\) , can be chosen arbitrarily, since one can always find a solution by learning the other two. This appears to suggest that homeostatic plasticity of the inhibitory weights might not be required in an ideal case. However, we experimentally show that excitatory plasticity alone is not sufficient to reliably make the network converge in the presence of the noise and non- idealities of the analog substrate (see Fig. 4). In the absence of inhibitory plasticity, both the excitatory and inhibitory populations approach their respective set- points, but they fail to converge to stable activity levels. (Fig. 4a). In contrast, the full cross- homeostatic plasticity rule robustly drives the network activity towards the set- points after just a few iterations (Fig. 2a). These results hold for multiple initializations (Fig. 4b), indicating that inhibitory plasticity is necessary for robust convergence. + +<|ref|>text<|/ref|><|det|>[[92, 861, 905, 926]]<|/det|> +Homeostatic plasticity in inhibitory neurons is less studied than in excitatory neurons [24, 36], although there is evidence that inhibitory neurons also regulate their activity levels based on activity set- points [37, 38]. Considering our on- chip results, we therefore speculate that homeostatic plasticity in all excitatory and inhibitory connections plays an important stabilization role also in biological systems. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 73, 864, 290]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 300, 905, 365]]<|/det|> +
Figure 4: (a) Example network with training results on "fixed" inhibitory synapses. Convergence is impaired if inhibitory plasticity is turned off. (b) Random networks ran with different initialization of all weights (n=8). The firing rates do not converge to the targets as precisely as in Fig. 2a, having a much higher root mean square error from the target (RMS on \(\bar{E} 6.39\) , on \(\bar{I} 35.11\) ).
+ +<|ref|>sub_title<|/ref|><|det|>[[92, 388, 600, 405]]<|/det|> +## Neuronal ensembles remain intact under cross-homeostasis + +<|ref|>text<|/ref|><|det|>[[92, 414, 905, 544]]<|/det|> +The effect of homeostasis is to maintain a set level of activity in neurons by scaling synaptic efficacy, as done in this work, or by changing the intrinsic properties of the neuron. This raises concerns about the stability- plasticity dilemma, i.e., the competition between the homeostatic stability mechanism and the associative forms of synaptic plasticity that contribute to the formation of functional neural ensembles, as they may counteract each other causing either forgetting or the inability to form new memories. A neural ensemble corresponds to a group of interconnected neurons (e.g., that increased their synaptic weights via learning) that contributes to the representation of a particular feature or concept. Typically, neurons forming an ensemble exhibit shared selectivity [39, 40]. + +<|ref|>text<|/ref|><|det|>[[92, 544, 905, 655]]<|/det|> +The ensemble could potentially be affected by the homeostatic scaling process, which attempts to equalize the activity levels across the network. To verify that the cross- homeostatic plasticity used here does not interfere or disrupt the role of neural ensembles we artificially formed a "crafted memory" into the network during the homeostatic training process (Fig. 5). This memory is implemented as a neural ensemble within the excitatory population of the network. Specifically, the ensemble consists of 32 neurons, in which each neuron has a connection probability with other neurons of the same ensemble of 0.5, much higher than the rest of the network (Fig. 5a). + +<|ref|>text<|/ref|><|det|>[[92, 656, 905, 785]]<|/det|> +After 250 cross- homeostatic plasticity training iterations, we manually form the ensemble by introducing the additional connections in it. We observe an immediate increase in the overall activity of both excitatory and inhibitory neurons (Fig. 5b) after ensemble implantation, which is however reduced as the homeostatic training continues. The change in weights (Fig. 5d- c) illustrates how inhibitory plasticity (specially \(w_{ei}\) ) compensates for increased activity, bringing the firing rates back to the target point. Because the values of weights within and outside of the ensemble remain differentiated, even after homeostatic plasticity converges, the neural ensemble remains intact and retains its selectivity for the specific feature it represents. We conclude that the cross- homeostatic learning rule brings the activity back to the target while preserving the implanted memory. + +<|ref|>sub_title<|/ref|><|det|>[[92, 803, 562, 821]]<|/det|> +## Multiple subnetworks can be emulated on a single chip + +<|ref|>text<|/ref|><|det|>[[92, 829, 905, 876]]<|/det|> +In the network discussed so far, we modeled a single stable point attractor in firing rate space, using an E/I- balanced network. We used 200 out of the 256 neurons available on a single core of our multi- core neuromorphic chip (DYNAP- SE2). + +<|ref|>text<|/ref|><|det|>[[92, 878, 905, 925]]<|/det|> +Even though it is still a relatively small- scale network, for computational purposes, having a single attractor formed from 200 neurons is not an optimal utilization of the resources on the chip. Employing smaller networks would allow us to implement multiple attractors on a single chip, whereby each subnetwork could represent, for + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[169, 73, 825, 464]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 474, 905, 650]]<|/det|> +
Figure 5: Training result with a memory implanted in the network. (a) Random recurrent E/I network with a neural ensemble (size = 32, green neurons). (b) Convergence of both excitatory (Pyr) and inhibitory (PV) populations to their respective set-point. A subset of the weights is manually changed to represent an ensemble at iteration 250, well after the network has converged to its set-point. A jump in activity is observed in both Pyr and PV cells due to implanted memory. (c) Diagram illustrating the process of rebalancing the weight matrix after implanting a single memory. Due to the chip's limited observable and shared parameters, it's not possible to record the actual weight matrix for visualization. (d) The weight \((w_{ee})\) value for 32 neurons is increased by a factor of 0.0088 \(\mu \mathrm{A}\) , illustrated by a dotted line. All weights undergo alteration again to compensate for the ensemble, bringing the network back to the set point after the memory is implanted. However, the memory remains intact, as shown by higher recurrent weight \((w_{memory})\) compared to the baseline excitatory connectivity \((w_{ee})\) .
+ +<|ref|>text<|/ref|><|det|>[[92, 676, 904, 708]]<|/det|> +example, a distinct choice in a decision- making task. The model could then be extended to introduce competition among the clusters leading to winner- take- all dynamics to facilitate decision- making processes [12- 15]. + +<|ref|>text<|/ref|><|det|>[[92, 710, 905, 756]]<|/det|> +Therefore, we also run scaled down variations of the previous network, and assess whether cross- homeostatic plasticity can still bring activity to desired levels. Table 1 summarizes the population count, connectivity scaling, and adjustment in the target firing rate that we adopt for each experiment. + +<|ref|>table<|/ref|><|det|>[[231, 770, 764, 865]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[90, 875, 904, 907]]<|/det|> +Table 1: Variation in network size, connection scaling, and stable target firing rate (set-points) for reliable self-sustained activity. + +
Number of neurons (E/I)Connection ProbabilityTarget (E/I)
200/500.120/40 Hz
100/250.120/40 Hz
64/160.3525/50 Hz
48/120.3540/60 Hz
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 75, 905, 203]]<|/det|> +Furthermore, we experiment with implementing multiple copies of these scaled- down networks on the same chip. We create 5 instances of different E/I networks, each with 50 excitatory and 15 inhibitory neurons. The connectivities within each sub- network are drawn from the same distribution. Each subnetwork does not cross- talk with others. The configuration is illustrated in Fig. 6a. To implement this configuration on the chip, we group the excitatory neuron populations into a single core, and use another core for the inhibitory populations. All parameter values, including weights, are shared between sub- networks. The challenge is to check if homeostatic plasticity leads to convergence for all subnetworks, since the effect of the weights is different for each E/I network due to the inhomogeneity of the chip. + +<|ref|>text<|/ref|><|det|>[[92, 205, 905, 285]]<|/det|> +During the training process, we sequentially stimulate one sub- network at a time during each iteration. We then compute the change in synaptic weights, denoted as \(\Delta w\) , based on the activity of the neuron population within that E/I network. The weight is configured on the chip after each iteration based on the average \(\Delta w\) across all sub- networks. Fig. 6b shows that all five subnetworks, with shared weights, can indeed converge to mean firing rate regimes close to the target firing rate. + +<|ref|>image<|/ref|><|det|>[[109, 300, 904, 530]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 541, 905, 637]]<|/det|> +
Figure 6: Simultaneous training of multiple clusters. (a) Schematic of clusters on the chip. One core contains excitatory populations, and another contains inhibitory populations. We implemented five E/I networks, with each network comprising 50 excitatory and 15 inhibitory neurons. As per table 1, the connection probability within these networks stands at 0.35. To account for the network's scaling, we defined higher set points. It's worth noting that these clusters do not interconnect with one another. b) Rate convergence for 5 clusters sharing the same nominal weights. Each color represents one cluster.
+ +<|ref|>sub_title<|/ref|><|det|>[[93, 677, 202, 696]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[92, 709, 905, 869]]<|/det|> +Hardware implementations of spiking neural networks are being proposed as a promising "neuromorphic" technology that can complement standard computation for sensory- processing applications that require limited resource usage (such as power, memory, and size), low latency, and that cannot resort to sending the recorded data to off- line computing centers (i.e., "the cloud") [29, 41, 42]. Among the proposed solutions, those that exploit the analog properties of their computing substrate have the highest potential of minimizing power consumption and carrying out "always- on" processing [9, 43]. However, the challenge of using this technology is to understand how to carry out reliable processing in the presence of the high variability, heterogeneity and sensitivity to noise of the analog substrate. This challenge becomes even more daunting when taking into account the variability of memristive devices, which are expected to allow for the storage and processing of information directly on chip [44]. + +<|ref|>text<|/ref|><|det|>[[92, 871, 904, 918]]<|/det|> +In this paper, we showed how resorting to strategies used by animal brains to solve the same challenge can lead to robust solutions: we demonstrated that cross- homeostatic plasticity can autonomously bring analog neuromorphic hardware to a fully self- sustained inhibition- stabilized regime, the regime that is thought to be the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 75, 904, 123]]<|/det|> +default state of the awake neocortex [19- 22]. These are the first results to show how biologically inspired learning rules can be used to self- tune neuromorphic hardware into computationally powerful states while simultaneously solving the problem of heterogeneity and variability in the hardware. + +<|ref|>text<|/ref|><|det|>[[92, 124, 905, 268]]<|/det|> +Our results also validate the robustness of the cross- homeostatic plasticity hypothesis [27] in a physical substrate that presents similar challenges to those of faced by biological networks. We also highlight the crucial contribution of inhibitory plasticity, specifically, that homeostatic plasticity of inhibitory connections is necessary to maintain stability in both biological and silicon networks. Notably, we constructed an on- chip stable attractor neural network that effectively maintains a long- term memory represented by a neural ensemble, and demonstrates that multiple networks, each exhibiting attractor dynamics, can coexist within the hardware [18, 35]. In the future, we can extend this stable multi- network implementation to simulate decision- making processes, along with self- generated dynamics that are thought to underlie a number of computations including motor control, timing, and sensory integration [45- 47]. + +<|ref|>text<|/ref|><|det|>[[92, 269, 905, 380]]<|/det|> +By taking inspiration from the self- calibration features of biological networks, we achieve a remarkable level of resilience to hardware variability, enabling recurrent networks to maintain stability and self- sustained activity across different analog neuromorphic chips. Cross- homeostatic plasticity allows us to instantiate neural networks on different chips without the need for chip- specific tuning, thereby addressing the inherent mismatch in analog devices. Such bio- plausible algorithms provide a promising avenue for the development of robust and scalable hardware implementations of recurrent neural networks, while mitigating the challenges associated with the intrinsic mismatch in analog devices. + +<|ref|>text<|/ref|><|det|>[[92, 381, 905, 461]]<|/det|> +In short, these results not only provide insights into the underlying mechanisms configuring biological circuits, but also hold great promise for the development of neuromorphic applications that unlock the power of recurrent computations. This research presents a substantial step forward in our understanding of neural dynamics and paves new possibilities for practical deployment of neuromorphic devices that compute, like biological circuits, with extremely low- power and low- latency using attractor dynamics. + +<|ref|>sub_title<|/ref|><|det|>[[92, 484, 184, 502]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[92, 515, 593, 532]]<|/det|> +## Neuromorphic hardware: the spiking neural network chip + +<|ref|>text<|/ref|><|det|>[[92, 540, 905, 701]]<|/det|> +In this study, we use a recently- developed mixed- signal spiking neural network chip, called Dynamic Neuromorphic Asynchronous Processor (DYNAP- SE2) [30]. This chip comprises 1024 adaptive- exponential integrate- and- fire (AdEx I&F) neurons [8, 48], distributed across four cores. The neuron parameters consist of its spiking threshold, refractory period, membrane time constant, and gain. On hardware, the inputs to each neuron are limited to 64 incoming connections, implemented using differential pair integrator (DPI) synapse circuits [49]. Each synaptic input can be configured to express one of four possible temporal dynamics (via four different DPI circuits): AMPA, NMDA, GABA- A, and GABA- B. Besides expressing different time constants, the excitatory synapse circuits (AMPA and NMDA) differ in that the NMDA circuit also incorporates a voltage gating mechanism. Similarly, the inhibitory synapses (GABA- A, and GABA- B) differ in their subtractive (GABA- B) versus shunting (GABA- A) effect. + +<|ref|>text<|/ref|><|det|>[[92, 703, 905, 766]]<|/det|> +The bias parameters that determine the weights, neuronal and synaptic properties are set by an on- chip digital to analog (DAC) bias- generator circuit [50] and are shared globally within each core. These parameters are expressed as currents that drive the neuromorphic circuits and are represented internally through a "coarse" and a "fine" value, according to the following equation: + +<|ref|>equation<|/ref|><|det|>[[433, 776, 903, 807]]<|/det|> +\[I = I_{C}(C_{w})\frac{F_{w}}{256}, \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[92, 812, 904, 843]]<|/det|> +where \(F_{w}\) represents the fine value and \(C_{w}\) the coarse one. The variable \(I_{C}\) represents a current estimated from the bias- generator circuit simulations, which depends on \(C_{w}\) as specified in Table 2. + +<|ref|>text<|/ref|><|det|>[[92, 844, 905, 923]]<|/det|> +Before training, the network is initialized with \(C_{w}\) randomly taking a value of 3, 4, or 5, and \(F_{w}\) uniformly distributed between 20 and 200. The corresponding currents will therefore range between \(2.7\mathrm{nA}\) and \(1757\mathrm{nA}\) . However, due to device mismatch and variability in the analog circuits, even when many circuits share the same nominal biases, the effective values will vary. On the DYNAP- SE2 and other mixed- signal neuromorphic processors implemented using similar technology nodes and circuit design styles, the coefficient of variation of + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[300, 72, 696, 120]]<|/det|> + +
Cw012345
IC(nA)0.070.554.4535.02802250
+ +<|ref|>table_caption<|/ref|><|det|>[[103, 134, 891, 147]]<|/det|> +Table 2: Nominal estimated coarse current values of the on-chip bias generator circuit for synaptic weights. + +<|ref|>text<|/ref|><|det|>[[92, 175, 905, 204]]<|/det|> +these parameters is typically around 20% [10]. However, the learning rules proposed in this work make these networks robust to such variability and lead to a stable activity regime. + +<|ref|>title<|/ref|><|det|>[[92, 226, 261, 237]]<|/det|> +# Network emulation + +<|ref|>text<|/ref|><|det|>[[92, 251, 905, 393]]<|/det|> +We create an on-chip E/I network composed of 200 Pyr (excitatory) and 50 PV (inhibitory) neurons. Each neuron type is implemented by tuning the refractory period, time constants, spiking threshold, and neuron gain parameters to match biologically measured values (Fig. 1c, [33]). The two populations are sparsely connected with a 10% connection probability for each pair of neurons (both, within population and between populations).Synaptic weights take four sets of values shared globally among all Pyr-to-Pyr \((w_{ee})\) , Pyr-to-PV \((w_{ie})\) , PV-to-PV \((w_{ii})\) , and PV-to-Pyr \((w_{ei})\) connections respectively. All excitatory weights represent synapses modeled as AMPA receptors, and inhibitory weights as GABA-A (somatic). Here, \(w_{ii}\) and \(w_{ei}\) are somatic inhibitory connections.Even though the weight parameters are shared, the synaptic efficacy of each synapse is subject to variability due to device mismatch in the analog circuits [10]. + +<|ref|>text<|/ref|><|det|>[[92, 397, 905, 523]]<|/det|> +During training, synaptic weights are adjusted such that the average activity for both Pyr and PV population converge to their respective target firing rates (or set-points), and can thus maintain self-sustained activity in the absence of any external input or bias current. To kickstart activity in the network, we provide a brief 40ms external input of four spikes at 10 ms intervals to 80% of the Pyr cells, with random delays for each postsynaptic target neuron to introduce temporal sparsity to the initial response of the population. The network is emulated in real-time on the chip (i.e., the dynamics of the analog circuits evolve through physical time reproducing the dynamics of the biological neural circuits). Each experiment is run for 1 s and the spiking activity of the whole network is streamed to a computer, for data logging and analysis purposes. + +<|ref|>text<|/ref|><|det|>[[92, 526, 905, 667]]<|/det|> +In the analysis, the network response is measured by calculating the average firing rate of both neural populations, which is then used to calculate the error (i.e., the difference between set-point and actual activity)that drives the weight update (see next section). As we are interested in the network's self-sustained behavior, we discard the response induced by the transient external input while calculating the firing rates (i.e., we ignore the first 60 ms of activity, which comprises the first 40 ms of external stimulation and an additional stabilization window of 20 ms). During the initial stages of the training procedure, before E/I balance is reached, the network could generate non-sustained bursts of activity. Therefore, we calculate the firing rate by considering only the time window when the neural populations are active, i.e. the in-burst firing rate, as opposed to the average over the whole trial. We empirically find that this choice leads to the desired sustained activity. + +<|ref|>title<|/ref|><|det|>[[92, 689, 216, 702]]<|/det|> +# Learning rule + +<|ref|>text<|/ref|><|det|>[[92, 715, 905, 743]]<|/det|> +To bring the recurrent dynamics to a fully self-sustained inhibition-stabilized regime we employ the recently proposed cross-homeostatic plasticity rule [27], which tunes all four synaptic weight classes in parallel. + +<|ref|>text<|/ref|><|det|>[[92, 747, 905, 824]]<|/det|> +In classic forms of homeostatic plasticity, both excitatory and inhibitory populations self-regulate calcium levels based on their own activity [24, 36]. In contrast to these classic forms, where a neuron tunes its weights to maintain its own average output of activity (a firing rate set-point), the cross-homeostatic weight update rule aims at bringing the neuron's presynaptic partners of opposite polarity to an average population firing rate setpoint,according to the following equations: + +<|ref|>equation<|/ref|><|det|>[[225, 835, 905, 876]]<|/det|> +\[\begin{array}{l}\Delta w_{ee} = +\alpha \bar {E}(t)(I_{target} - \bar {I}(t)),\\ \Delta w_{ei} = -\alpha \bar {I}(t)(I_{target} - \bar {I}(t)),\end{array}\qquad \begin{array}{l}\Delta w_{ie} = -\alpha \bar {E}(t)(E_{target} - \bar {E}(t)),\\ \Delta w_{ii} = +\alpha \bar {I}(t)(E_{target} - \bar {E}(t)),\end{array}\tag{2}\] + +<|ref|>text<|/ref|><|det|>[[92, 886, 905, 918]]<|/det|> +where \(\alpha\) is the learning rate \((\alpha =0.05)\) , \(\bar {E}(t)\) and \(\bar {I}(t)\) are the average measured firing rates during a trial t and \(E_{target}=20\) Hz, \(I_{target}=40\) Hz are the desired firing rate set-points, for the Pyr and PV populations + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 74, 905, 123]]<|/det|> +respectively. This type of plasticity can be interpreted biologically as having a target on the input current that a given neuron receives. For example, an excitatory neuron could have indirect access to the average firing rate of their presynaptic inhibitory neurons based on a running average of the total GABA it receives. + +<|ref|>text<|/ref|><|det|>[[92, 123, 905, 188]]<|/det|> +While biologically plausible target activities are typically lower (e.g., \(5\mathrm{Hz}\) for the \(E\) population and \(14\mathrm{Hz}\) for the \(I\) one as proposed in [27]), in our experiments we had to use higher set- point rates to attain reliable behavior, because of the relatively small network size and the low probability of connections used. In Table 1 we quantify the effects of network size and connectivity with additional experiments. + +<|ref|>text<|/ref|><|det|>[[92, 188, 905, 301]]<|/det|> +As the weights are shared globally within each population, the learning rules are implemented at the whole population level. Therefore the network maintains a given firing rate at the population level, rather than at the individual neuron level (see [27] for a local implementation of the rules). However, the heterogeneity in the analog circuits and their inherent noise produce diverse behaviors among neurons, for example leading some of the neurons to fire intermittently at slightly increased rates (while maintaining the desired set- point firing rate at the full population level). In the following section, we explain the training procedure on the hardware and how the changes in weight \(\Delta w\) are translated to the fine \(F_{w}\) and course \(C_{w}\) synaptic bias parameters on chip. + +<|ref|>sub_title<|/ref|><|det|>[[93, 319, 312, 336]]<|/det|> +## Chip-in-the-loop training + +<|ref|>text<|/ref|><|det|>[[92, 343, 905, 408]]<|/det|> +Although the DYNAP- SE2 chip can be tuned to exhibit biologically plausible dynamics, it does not incorporate on- chip synaptic plasticity circuits that are capable of learning. To overcome this limitation we used a "chip- in- the- loop" learning paradigm: the activity produced by the chip is transmitted to a computer, which analyzes the data at the software level and calculates the parameter updates, which are then sent back to the chip. + +<|ref|>text<|/ref|><|det|>[[92, 408, 905, 537]]<|/det|> +In particular, the network is emulated for 1 s on the chip. Each emulation is repeated five times to reduce the influence of noise and obtain more reliable results. The output spikes are sent back to the computer, which averages the results, calculates the appropriate weight update, applies stochastic rounding (see below), and sends the updated weights to the chip for the next iteration (i.e., after 5 repetitions). After each iteration, we drain the accumulated residual current from both the neuron and synapse circuits. We stop the procedure after 400 iterations, as we found it to be sufficient for the network to converge to the target firing rates regardless of the initial conditions. Most simulations converge to target firing rate before iteration 200, which shows the long- term stability of the cross- homeostasis rules in a converged network. + +<|ref|>text<|/ref|><|det|>[[92, 538, 905, 602]]<|/det|> +For all experiments, we applied the same set of parameters to neurons and synapses at the beginning of the training, with the exception of the random initialization of weights. The training produced stable and reliable activity around the programmed set- points \(100\%\) of the time, without having to re- tune any of the neuron or synapse parameters (except the synaptic weights), across multiple days, and for multiple chips. + +<|ref|>text<|/ref|><|det|>[[92, 619, 905, 700]]<|/det|> +Stochastic rounding As the values of the synaptic weight parameters are set by the bias- generator DAC, the weight updates can only take discrete values \(\Delta F_{w} = 0,\pm 1,\pm 2,\dots\) . For very large learning rates, this could lead to unstable learning, and for very low ones to no changes at all. To overcome the constraints of limited- resolution weight updates, a common strategy used is to use stochastic rounding (SR) [51, 52]. The SR technique consists of interpreting the \(\Delta w\) calculated by the learning rule as a probability to increase or decrease the weight, as follows: + +<|ref|>equation<|/ref|><|det|>[[364, 707, 904, 749]]<|/det|> +\[\mathrm{SR}(x) = \left\{ \begin{array}{ll}[x] & \mathrm{with~prob.~}x - \lfloor x\rfloor \\ \lfloor x\rfloor & \mathrm{otherwise.} \end{array} \right. \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[92, 755, 590, 771]]<|/det|> +where \([\cdot ]\) indicates the ceiling operation and \(\lfloor \cdot \rfloor\) the floor operation. + +<|ref|>text<|/ref|><|det|>[[92, 772, 905, 819]]<|/det|> +Values for the desired weight updates \(\Delta w\) are given by the learning rules (2). We also define upper and lower bounds for \(F_{w}\) , respectively \(F_{- } = 20\) and \(F_{+} = 250\) . We then obtain the new coarse and fine values \(C_{w}^{\prime},F_{w}^{\prime}\) of the weight as follows: + +<|ref|>equation<|/ref|><|det|>[[261, 827, 904, 889]]<|/det|> +\[\left\{ \begin{array}{ll}C_{w}^{\prime} = C_{w}, & F_{w}^{\prime} = F_{w} + \mathrm{SR}(\Delta w) & \mathrm{if} F_{-}\leq F_{w} + \mathrm{SR}(\Delta w)\leq F_{+} \\ C_{w}^{\prime} = C_{w} - 1, & F_{w}^{\prime} = F_{+} & \mathrm{if} F_{w} + \mathrm{SR}(\Delta w)< F_{-} \\ C_{w}^{\prime} = C_{w} + 1, & F_{w}^{\prime} = F_{-} & \mathrm{if} F_{w} + \mathrm{SR}(\Delta w) > F_{+}. \end{array} \right. \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[90, 895, 904, 926]]<|/det|> +In other words, the fine value is updated with the stochastically- rounded version of \(\Delta w\) ; when it oversteps the bound, the fine value is reset to the bound, and the coarse value is changed instead. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 75, 905, 139]]<|/det|> +Note that this SR strategy is compatible with neuromorphic hardware implementations, as demonstrated in [53]. Our setup using a chip- in- the- loop enables the evaluation of different learning mechanisms and test the robustness of different design choices. The design of future on- chip learning circuit can therefore be informed by our results, following a hardware- software co- design approach. + +<|ref|>sub_title<|/ref|><|det|>[[94, 161, 320, 180]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[92, 194, 905, 242]]<|/det|> +Conceptualization: M, SSM and RL. Investigation: M, MS and SSM. Methodology, Software, Data curation and Visualization: M and MS. Supervision SSM, RL, DB and GI. Writing Original Draft: M and MS. Writing and editing: SSM, RL, DB and GI. Funding acquisition: SSM, MS, DB and GI. + +<|ref|>sub_title<|/ref|><|det|>[[93, 266, 303, 286]]<|/det|> +## Competing Interests + +<|ref|>text<|/ref|><|det|>[[92, 299, 508, 314]]<|/det|> +The authors declare that they have no conflict of interest. + +<|ref|>sub_title<|/ref|><|det|>[[93, 337, 280, 356]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[92, 369, 905, 499]]<|/det|> +We thank the organizers of the CapoCaccia Workshop Towards Neuromorphic Intelligence 2022, where this work was conceived. We thank Chenxi Wu, Ole Richter and German Koestinger for DYNAP- SE2 support, and Matthew Cook for useful discussions. SSM was supported by the Swiss National Science Foundation (SNSF) grant nos. P2ZHP3- 187943 and P500PB- 203133. M was supported by the SNSF Sinergia Project (CRSII5- 180316). The work of MS was made possible by a postdoctoral fellowship from the ETH AI Center. GI was supported by the EU ERC Grant "NeuroAgents" (No. 724295), DVB and RL were supported by the National Institutes of Health grant no. NS116589. RL was supported by Universidad Nacional de Quilmes (Argentina), CONICET (Argentina), and NeurotechEU (Faculty Scholarship). + +<|ref|>sub_title<|/ref|><|det|>[[92, 521, 205, 540]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[100, 552, 907, 916]]<|/det|> +[1] Guillaume Hennequin, Tim P Vogels, and Wulfram Gerstner. "Optimal control of transient dynamics in balanced networks supports generation of complex movements". In: Neuron 82.6 (2014), pp. 1394- 1406. [2] Dean V Buonomano and Wolfgang Maass. "State- dependent computations: spatiotemporal processing in cortical networks". In: Nature Reviews Neuroscience 10.2 (2009), pp. 113- 125. [3] Saurabh Vyas, Matthew D Golub, David Sussillo, and Krishna V Shenoy. "Computation through neural population dynamics". In: Annual review of neuroscience 43 (2020), pp. 249- 275. [4] Rodney J. Douglas and Kevan A.C. Martin. "Recurrent neuronal circuits in the neocortex". In: Current Biology 17.13 (2007), R496- R500. doi: 10.1016/j.cub.2007.04.024. [5] Rodney J. Douglas, Christof Koch, Misha Mahowald, Kevan A. C. Martin, and Humbert H. Suarez. "Recurrent Excitation in Neocortical Circuits". In: Science 269.5226 (1995), pp. 981- 985. doi: 10.1126/science.7638624. [6] Xiao- Jing Wang. "Synaptic reverberation underlying mnemonic persistent activity". In: Trends in neurosciences 24.8 (2001), pp. 455- 463. [7] Carver Mead. "Neuromorphic Engineering: In Memory of Misha Mahowald". In: Neural Computation 35 (2023), pp. 343- 383. doi: 10.1162/neco_a_01553. [8] E. Chicca, F. Stefanini, C. Bartolozzi, and G. Indiveri. "Neuromorphic electronic circuits for building autonomous cognitive systems". In: Proceedings of the IEEE 102.9 (Sept. 2014), pp. 1367- 1388. issn: 0018- 9219. doi: 10.1109/JPROC.2014.2313954. [9] C. Mead. "Neuromorphic Electronic Systems". In: Proceedings of the IEEE 78.10 (1990), pp. 1629- 36. doi: 10.1109/5.58356. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 75, 907, 120]]<|/det|> +[10] Dmitrii Zendrikov, Sergio Solinas, and Giacomo Indiveri. "Brain-inspired methods for achieving robust computation in heterogeneous mixed- signal neuromorphic processing systems". In: Neuromorphic Computing and Engineering 3.3 (July 2023), p. 034002. issn: 2634- 4386. doi: 10.1088/2634- 4386/ac6c4c. + +<|ref|>text<|/ref|><|det|>[[90, 125, 905, 160]]<|/det|> +[11] Simon B. Laughlin and Terrence J. Sejnowski. "Communication in Neuronal Networks". In: Science 301.5641 (2003), pp. 1870- 1874. doi: 10.1126/science.1089662. + +<|ref|>text<|/ref|><|det|>[[90, 164, 905, 198]]<|/det|> +[12] U. Rutishauser and R. Douglas. "State- Dependent Computation Using Coupled Recurrent Networks". In: Neural Computation 21 (2009), pp. 478- 509. + +<|ref|>text<|/ref|><|det|>[[90, 202, 905, 236]]<|/det|> +[13] E. Neftci et al. "Synthesizing cognition in neuromorphic electronic systems". In: Proceedings of the National Academy of Sciences 110.37 (2013), E3468- E3476. + +<|ref|>text<|/ref|><|det|>[[90, 240, 905, 288]]<|/det|> +[14] M. Giulioni et al. "Robust working memory in an asynchronously spiking neural network realized in neuromorphic VLSI". In: Frontiers in Neuroscience 5.149 (2012). issn: 1662- 453X. doi: 10.3389/fnins.2011.00149. + +<|ref|>text<|/ref|><|det|>[[90, 292, 905, 326]]<|/det|> +[15] T. Pfeil et al. "Six networks on a universal neuromorphic computing substrate". In: Frontiers in Neuroscience 7 (2013). issn: 1662- 453X. doi: 10.3389/fnins.2013.00011. + +<|ref|>text<|/ref|><|det|>[[90, 330, 905, 364]]<|/det|> +[16] Mircea Steriade, David A. McCormick, and Terrence J. Sejnowski. "Thalamocortical oscillations in the sleeping and aroused brain". In: Science 262.5134 (1993), pp. 679- 685. + +<|ref|>text<|/ref|><|det|>[[90, 367, 905, 401]]<|/det|> +[17] DAVID A McCormick. "GABA as an inhibitory neurotransmitter in human cerebral cortex". In: Journal of neurophysiology 62.5 (1989), pp. 1018- 1027. + +<|ref|>text<|/ref|><|det|>[[90, 404, 905, 454]]<|/det|> +[18] Misha V Tsodyks, William E Skaggs, Terrence J Sejnowski, and Bruce L McNaughton. "Paradoxical effects of external modulation of inhibitory interneurons". In: Journal of neuroscience 17.11 (1997), pp. 4382- 4388. + +<|ref|>text<|/ref|><|det|>[[90, 457, 905, 491]]<|/det|> +[19] Hirofumi Ozeki, Ian M Finn, Evan S Schaffer, Kenneth D Miller, and David Ferster. "Inhibitory stabilization of the cortical network underlies visual surround suppression". In: Neuron 62.4 (2009), pp. 578- 592. + +<|ref|>text<|/ref|><|det|>[[90, 494, 905, 527]]<|/det|> +[20] Alessandro Sanzeni et al. "Inhibition stabilization is a widespread property of cortical networks". In: Elife 9 (2020), e54875. + +<|ref|>text<|/ref|><|det|>[[90, 531, 905, 565]]<|/det|> +[21] Sadra Sadeh and Claudia Clopath. "Inhibitory stabilization and cortical computation". In: Nature Reviews Neuroscience 22.1 (2021), pp. 21- 37. + +<|ref|>text<|/ref|><|det|>[[90, 568, 905, 602]]<|/det|> +[22] Alain Destexhe, Stuart W Hughes, Michelle Rudolph, and Vincenzo Crunelli. "Are corticothalamic 'up' states fragments of wakefulness?" In: Trends in neurosciences 30.7 (2007), pp. 334- 342. + +<|ref|>text<|/ref|><|det|>[[90, 605, 905, 639]]<|/det|> +[23] G.G. Turrigiano and S.N. Nelson. "Homeostatic plasticity in the developing nervous system". In: Nature Reviews Neuroscience 5 (Feb. 2004), pp. 97- 107. + +<|ref|>text<|/ref|><|det|>[[90, 642, 905, 676]]<|/det|> +[24] Gina Turrigiano. "Homeostatic synaptic plasticity: local and global mechanisms for stabilizing neuronal function". In: Cold Spring Harbor perspectives in biology 4.1 (2012), a005736. + +<|ref|>text<|/ref|><|det|>[[90, 679, 905, 729]]<|/det|> +[25] Guillaume Hennequin, Everton J. Agnes, and Tim P. Vogels. "Inhibitory Plasticity: Balance, Control, and Codependence". In: Annual Review of Neuroscience 40 (2017), pp. 557- 579. issn: 15454126. doi: 10.1146/annurev-neuro- 072116- 031005. + +<|ref|>text<|/ref|><|det|>[[90, 732, 905, 766]]<|/det|> +[26] Robert C Froemke. "Plasticity of cortical excitatory- inhibitory balance". In: Annual review of neuroscience 38 (2015), pp. 195- 219. + +<|ref|>text<|/ref|><|det|>[[90, 769, 905, 819]]<|/det|> +[27] Saray Soldado- Magraner, Michael J Seay, Rodrigo Laje, and Dean V Buonomano. "Paradoxical self- sustained dynamics emerge from orchestrated excitatory and inhibitory homeostatic plasticity rules". In: Proceedings of the National Academy of Sciences 119.43 (2022), e2200621119. + +<|ref|>text<|/ref|><|det|>[[90, 822, 905, 856]]<|/det|> +[28] Bipin Rajendran, Damien Querlioz, Sabina Spiga, and Abu Sebastian. "Memristive devices for spiking neural networks". In: Memristive Devices for Brain- Inspired Computing. Elsevier, 2020, pp. 399- 405. + +<|ref|>text<|/ref|><|det|>[[90, 859, 905, 893]]<|/det|> +[29] Dennis Valbjorn Christensen et al. "2022 roadmap on neuromorphic computing and engineering". In: Neuromorphic Computing and Engineering (2022). doi: 10.1088/2634- 4386/ac4a83. + +<|ref|>text<|/ref|><|det|>[[90, 896, 905, 929]]<|/det|> +[30] Ole Richter et al. DYNAP- SE2: a scalable multi- core dynamic neuromorphic asynchronous spiking neural network processor. 2023. doi: 10.48550/arXiv.2310.00564. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 75, 907, 110]]<|/det|> +[31] S.- C. Liu, J. Kramer, G. Indiveri, T. Delbruck, and R.J. Douglas. Analog VLSI: Circuits and Principles. MIT Press, 2002. + +<|ref|>text<|/ref|><|det|>[[90, 113, 907, 161]]<|/det|> +[32] Brielle R. Ferguson and Wen- Jun Gao. "PV Interneurons: Critical Regulators of E/I Balance for Prefrontal Cortex- Dependent Behavior and Psychiatric Disorders". In: Frontiers in Neural Circuits 12 (2018), p. 37. DOI: 10.3389/fncir.2018.00037. + +<|ref|>text<|/ref|><|det|>[[90, 166, 907, 215]]<|/det|> +[33] Juan L Romero- Sosa, Helen Motanis, and Dean V Buonomano. "Differential excitability of PV and SST neurons results in distinct functional roles in inhibition stabilization of up states". In: Journal of Neuroscience 41.34 (2021), pp. 7182- 7196. + +<|ref|>text<|/ref|><|det|>[[90, 220, 905, 253]]<|/det|> +[34] Arvind Kumar, Sven Schrader, Ad Aertsen, and Stefan Rotter. "The high- conductance state of cortical networks". In: Neural computation 20.1 (2008), pp. 1- 43. + +<|ref|>text<|/ref|><|det|>[[90, 258, 905, 290]]<|/det|> +[35] Alfonso Renart et al. "The asynchronous state in cortical circuits". In: science 327.5965 (2010), pp. 587- 590. + +<|ref|>text<|/ref|><|det|>[[90, 295, 905, 328]]<|/det|> +[36] Gina G Turrigiano and Sacha B Nelson. "Homeostatic plasticity in the developing nervous system". In: Nature reviews neuroscience 5.2 (2004), pp. 97- 107. + +<|ref|>text<|/ref|><|det|>[[90, 333, 905, 365]]<|/det|> +[37] Sandra J Kuhlman et al. "A disinhibitory microcircuit initiates critical- period plasticity in the visual cortex". In: Nature 501.7468 (2013), pp. 543- 546. + +<|ref|>text<|/ref|><|det|>[[90, 370, 905, 402]]<|/det|> +[38] Zhengyu Ma, Gina G Turrigiano, Ralf Wessel, and Keith B Hengen. "Cortical circuit dynamics are homeostatically tuned to criticality in vivo". In: Neuron 104.4 (2019), pp. 655- 664. + +<|ref|>text<|/ref|><|det|>[[90, 407, 905, 440]]<|/det|> +[39] Tomoki Fukai and Shigeru Tanaka. "A simple neural network exhibiting selective activation of neuronal ensembles: from winner- take- all to winners- share- all". In: Neural computation 9.1 (1997), pp. 77- 97. + +<|ref|>text<|/ref|><|det|>[[90, 444, 905, 477]]<|/det|> +[40] Jordan P Hamm, Yuriy Shymkiv, Shuting Han, Weijian Yang, and Rafael Yuste. "Cortical ensembles selective for context". In: Proceedings of the National Academy of Sciences 118.14 (2021), e2026179118. + +<|ref|>text<|/ref|><|det|>[[90, 481, 905, 514]]<|/det|> +[41] G. Indiveri and S.- C. Liu. "Memory and information processing in neuromorphic systems". In: Proceedings of the IEEE 103.8 (2015), pp. 1379- 1397. DOI: 10.1109/JPROC.2015.2444094. + +<|ref|>text<|/ref|><|det|>[[90, 519, 905, 552]]<|/det|> +[42] Adnan Mehonic and Anthony J Kenyon. "Brain- inspired computing: We need a master plan". In: Nature 604.7905 (2022), pp. 255- 260. DOI: 10.1038/s41586- 021- 04362- w. + +<|ref|>text<|/ref|><|det|>[[90, 557, 905, 604]]<|/det|> +[43] Giacomo Indiveri and Yulia Sandamirskaya. "The importance of space and time for signal processing in neuromorphic agents". In: IEEE Signal Processing Magazine 36.6 (2019), pp. 16- 28. DOI: 10.1109/MSP.2019.2928376. + +<|ref|>text<|/ref|><|det|>[[90, 610, 905, 642]]<|/det|> +[44] Wen Sun et al. "Understanding memristive switching via in situ characterization and device modeling". In: Nature communications 10.1 (2019), pp. 1- 13. DOI: 10.1038/s41467- 019- 11411- 6. + +<|ref|>text<|/ref|><|det|>[[90, 648, 905, 696]]<|/det|> +[45] Valerio Mante, David Sussillo, Krishna V Shenoy, and William T Newsome. "Context- dependent computation by recurrent dynamics in prefrontal cortex". In: Nature 503.7474 (2013), pp. 78- 84. issn: 00092665. DOI: 10.1038/nmeth.2835. Multiplexed. + +<|ref|>text<|/ref|><|det|>[[90, 701, 905, 733]]<|/det|> +[46] Vishwa Goudar and Dean V Buonomano. "Encoding sensory and motor patterns as time- invariant trajectories in recurrent neural networks". In: Elife 7 (2018), e31134. + +<|ref|>text<|/ref|><|det|>[[90, 738, 905, 771]]<|/det|> +[47] Brian DePasquale, Christopher J Cueva, Kanaka Rajan, G Sean Escola, and LF Abbott. "full- FORCE: A target- based method for training recurrent networks". In: PloS one 13.2 (2018), e0191527. + +<|ref|>text<|/ref|><|det|>[[90, 776, 905, 808]]<|/det|> +[48] G. Indiveri et al. "Neuromorphic silicon neuron circuits". In: Frontiers in Neuroscience 5 (2011), pp. 1- 23. issn: 1662- 453X. DOI: 10.3389/fnins.2011.00073. + +<|ref|>text<|/ref|><|det|>[[90, 813, 905, 846]]<|/det|> +[49] C. Bartolozzi and G. Indiveri. "Synaptic dynamics in analog VLSI". In: Neural Computation 19.10 (Oct. 2007), pp. 2581- 2603. DOI: 10.1162/neco.2007.19.10.2581. + +<|ref|>text<|/ref|><|det|>[[90, 851, 905, 884]]<|/det|> +[50] T. Delbruck and A. Van Schaik. "Bias Current Generators with Wide Dynamic Range". In: Analog Integrated Circuits and Signal Processing 43.3 (2005), pp. 247- 268. + +<|ref|>text<|/ref|><|det|>[[90, 889, 905, 921]]<|/det|> +[51] Lorenz K Muller and Giacomo Indiveri. "Rounding methods for neural networks with low resolution synaptic weights". In: arXiv preprint arXiv:1504.05767 (2015), pp. 1- 11. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 74, 905, 180]]<|/det|> +[52] Matteo Croci, Massimiliano Fasi, Nicholas J Higham, Theo Mary, and Mantas Mikaitis. “Stochastic rounding: implementation, error analysis and applications”. In: Royal Society Open Science 9.3 (2022), p. 211631. [53] Matteo Cartiglia et al. “Stochastic dendrites enable online learning in mixed-signal neuromorphic processing systems”. In: 2022 IEEE International Symposium on Circuits and Systems (ISCAS) (May 2022). doi: 10.1109/iscas48785.2022.9937833. + +<--- Page Split ---> diff --git a/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/images_list.json b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c93311f956a9f78af6d04f3567e9a6acbaba4850 --- /dev/null +++ b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/images_list.json @@ -0,0 +1,235 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig.1. Flowchart of WAAM real-time temperature field prediction system.", + "footnote": [], + "bbox": [ + [ + 203, + 85, + 826, + 396 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig.2. Schematic of the FEM simulation model. (a) top view of the model. (b) main view of the model. (c) temperature field at a certain time.", + "footnote": [], + "bbox": [ + [ + 159, + 90, + 838, + 319 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig 3. Workflow for FEM simulation data collection and processing.", + "footnote": [], + "bbox": [ + [ + 181, + 85, + 820, + 350 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Proposed PIGeoRNN model architecture.", + "footnote": [], + "bbox": [ + [ + 222, + 444, + 775, + 780 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Hard-encoding of BCs.", + "footnote": [], + "bbox": [ + [ + 228, + 534, + 757, + 725 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6. Experiment setup for WAAM system.", + "footnote": [], + "bbox": [ + [ + 225, + 105, + 770, + 351 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig. 7. The future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models at two time points are compared with the true temperature field.", + "footnote": [], + "bbox": [ + [ + 175, + 292, + 833, + 731 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8. Comparison of the errors of the future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models with the true temperature fields at two time points.", + "footnote": [], + "bbox": [ + [ + 210, + 90, + 790, + 411 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Fig. 9. (a) The temperature history and (b) Error of the middle node on the tenth layer, predicted by the PIGeoRNN model over five time.", + "footnote": [], + "bbox": [ + [ + 167, + 92, + 840, + 280 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_10.jpg", + "caption": "Fig. 10. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models for loss and (b) MSE per time in prediction in FEM data training.", + "footnote": [], + "bbox": [ + [ + 150, + 92, + 848, + 283 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_11.jpg", + "caption": "Fig. 11. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on FEM data.", + "footnote": [], + "bbox": [ + [ + 150, + 661, + 845, + 859 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_12.jpg", + "caption": "Fig. 12. Tested thin-wall structure.", + "footnote": [], + "bbox": [ + [ + 377, + 463, + 617, + 559 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_14.jpg", + "caption": "Fig. 14. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models trained on experiment data in terms of loss and (b) MSE at each time of test.", + "footnote": [], + "bbox": [], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_15.jpg", + "caption": "Fig. 15. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on experiment data.", + "footnote": [], + "bbox": [ + [ + 156, + 92, + 792, + 282 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_16.jpg", + "caption": "Fig. 16. (a) Comparison of True and Predicted Temperature Fields for TL and No TL Models at", + "footnote": [], + "bbox": [ + [ + 174, + 677, + 825, + 878 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_17.jpg", + "caption": "Fig. 17. (a) Comparison of model training loss history TL or No TL and (b) MSE of predicted results at each time of test.", + "footnote": [], + "bbox": [ + [ + 155, + 125, + 842, + 303 + ] + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa.mmd b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d4be4ea0949488522ef47ab04d30641705ec3f95 --- /dev/null +++ b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa.mmd @@ -0,0 +1,424 @@ + +# PIML-based Real-time Long-horizon temperature Fields prediction in metallic additive manufacturing + +Donghong Ding donghong@n.jtech.edu.cn + +Nanjing Tech University mingxuan tian Nanjing Tech University Haochen Mu Nanjing Tech University Tao Liu Nanjing Tech University Mengjiao Li Nanjing Tech University Jianping Zhao Nanjing Tech University + +## Article + +Keywords: DED- arc, Temperature field prediction, Physics- informed machine learning, Geometric, Transfer learning + +Posted Date: October 31st, 2025 + +DOI: https://doi.org/10.21203/rs.3.rs- 6293181/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Communications Engineering on September 29th, 2025. See the published version at https://doi.org/10.1038/s44172- 025- 00501- 7. + +<--- Page Split ---> + +# PIML-based Real-time Long-horizon temperature Fields prediction in metallic additive manufacturing + +## Abstract + +The real- time long- horizon temperature field prediction during the Wire Arc Additive Manufacturing (WAAM) process is crucial for controlling heat accumulation, optimizing process parameters, and ensuring the quality of the manufactured parts. However, due to the time- consuming nature of finite element methods (FEM) and the process- informed neglect by existing data- driven models, real- time long- horizon temperature field prediction remains a technical challenge for control systems in metallic AM. Despite the interpretability offered by Physics- Informed Machine Learning (PIML), which still suffers from relatively long training times and a lack of long- horizon prediction capabilities for dynamic systems. To address this issue, the study proposes a Physics- informed Geometric Recurrent Neural Network (PIGeoRNN) model that fuses geometric characteristic and physical information through an encoder, extracts spatiotemporal characteristic using convolutional long short- term memory cells, and ensures that the prediction results conform to physical laws by incorporating hard- encoding initial/boundary conditions and a PI loss function. The model is capable of predicting the temperature field for the future 1.25 s based on data from the current 1.25 s, and it has also been evaluated for more long- horizon predictions. Transfer learning was used for enhance the model's efficiency in practical applications. Results demonstrate that the proposed PIGeoRNN model achieves a maximum prediction error of only \(4.5\%\) to \(13.9\%\) for both simulation and experiment. The inclusion of geometric features and physical information reduces the maximum error by approximately \(1\%\) , while the integrated model can lower the maximum error by \(4\%\) . Furthermore, transfer learning training reduces the model's training time by approximately \(50\%\) while achieving the same loss level. These findings indicate that the proposed method holds significant potential for process feed- forward control and digital twin in WAAM. It also represents an important first step toward real- time long- horizon physical fields prediction in metallic AM. + +Keywords: DED- arc; Temperature field prediction; Physics- informed machine learning; Geometric; Transfer learning + +## 1. Introduction + +With the accelerated advancement of Industry 4.0 and intelligent manufacturing, Additive Manufacturing (AM) field is entering a critical phase of technological innovation and application expansion. Against the backdrop of intelligent manufacturing, AM integrates cutting- edge technologies such as digital twin and Machine Learning (ML) to promote the automation, intelligentization, and digital transformation of production processes[1- 3]. Wire Arc Additive Manufacturing (WAAM), also known as Wire- Arc Directed Energy Deposition (Wire- Arc DED), produces near- net- shape parts by melting wire feedstock using an electric arc[4]. Characterized by high deposition efficiency and material utilization, as well as low equipment costs, this process is capable of processing high- strength materials. Therefore, it has been widely applied in fields such as aerospace, maritime, energy, and heavy industry, providing feasible solutions for the production of complex components[5]. + +Although WAAM has certain advantages compared with other AM technologies, minimizing + +<--- Page Split ---> + +the impact of the temperature field during the WAAM process on the final properties of the parts remains a challenge [6]. The spatiotemporal distribution of the temperature field during the arc additive manufacturing process directly affects the quality and performance of the printed parts. High temperatures and rapid cooling lead to the formation of different microstructures during material solidification, such as coarse columnar grains or fine equiaxed grains, which in turn affect the mechanical properties of the material[7]. The non- uniformity of the temperature field can cause uneven thermal expansion, leading to thermal deformation and the generation of residual stress, potentially inducing defects such as cracks[8]. Therefore, studying real- time long- horizon temperature field prediction helps to achieve feed- forward process control and digital twin, thereby reducing defects in WAAM parts. + +To accurately predict the temperature field in the WAAM process, both physics- based and data- driven approaches have been developed. Physics- based methods can be categorized into numerical and analytical approaches. The numerical methods for predicting the temperature field in WAAM mainly include the finite element method (FEM) [9] and finite volume method (FVM) [10], these methods discretize the continuous heat conduction problem into a finite number of elements or control volumes and use numerical solution techniques to calculate the temperature distribution within each element or volume. Although FEM can achieve long- horizon modeling, its high computational cost and time- consuming nature limit its application in real- time prediction. Compared with physics- based methods, data- driven approaches have the advantage of predicting complex temperature fields with high precision and efficiency while requiring only limited fundamental knowledge of physics [11]. Several ML, e.g. artificial neural networks (ANN) [12], feedforward neural networks (FFNN) [13] and recurrent neural network (RNN) [14], have been used for temperature field prediction in metal AM processes [15]. However, purely data- driven approaches lack physical interpretability and struggle to fully capture complex spatiotemporal sequence characteristic and process- informed. While the advent of Physics- Informed Machine Learning (PIML) has provided physically interpretable methods for additive manufacturing modeling and prediction, it is still characterized by relatively long training times and lack of long- horizon prediction capabilities for dynamic systems [16]. Real- time long- horizon temperature field prediction system for the WAAM should possess characteristics such as accuracy, adaptability, and interpretability: + +1. Process-informed: The WAAM deposition process encompasses a wealth of process information that the system can effectively identify and learn. +2. Transfer learning (TL): FEM simulation data are used to initialize model parameters, enabling the system to rapidly learn the fundamental characteristics and physical laws of the temperature field. +3. Adaptiveness: The system should be capable of calibrating prediction results in real-time based on geometric feedback. +4. Interpretability: The physics-informed (PI) loss function is combined with a hard constraint on the IBCs, and the system prediction results should be physically constrained. + +This study aims to improve the accuracy, spatiotemporal correlation, and real- time long- horizon performance of the model for long- horizon prediction of the temperature field in WAAM by combining PIML and RNN methods. The rest of the paper is organized as follows: in Section 2, this study reviews existing related work on prediction of temperature field in metal AM + +<--- Page Split ---> + +processes. In Section 3, the methodology for the system is presented. Then, the experiment system setup and practical experiments is presented in Section 4. Section 5 discusses the results of the model trained on experiment and simulation data, with or without TL training. Finally, Section 6 summarizes the findings of this paper and future work. + +## 2. Related work + +2.1. Physical modeling for temperature field simulation in WAAM + +Recently, analytical models of metal AM have been extensively studied. Batut et al. [17] developed an analytical model based on the assumption of a moving heat source for temperature prediction in the Direct Metal Deposition (DMD) process for titanium alloy Ti6Al4V. Daniel et al. [18] proposed a simplified macroscopic semi- analytical thermal analysis method to simulate temperature and phase transformations during the DED process. This method enables efficient computational simulation of the entire process, taking into account the effects of solidification and solid- state phase transformations. Ansari et al. [19] established an analytical method to predict transient temperature field variations during the Laser DED process. This study compared three heat source models and found that the two- dimensional (2D) Gaussian heat source model had the best accuracy. Subsequently, they developed a physics- based analytical model to investigate the impact of different scanning strategies on the temperature field [20]. Although analytical models are used for in- situ monitoring in AM processes, the effectiveness of their performance is still questioned. On the one hand, analytical models based on physical principles have limitations in dealing with uncertainties and variances that arise during the manufacturing process; on the other hand, these models often struggle to accurately reflect complex thermal behaviors under all boundary conditions. + +Numerical models constructed for metal AM processes have been proven to more effectively predict thermal distribution under given boundary conditions, demonstrating greater adaptability and accuracy. The FEM as a commonly used numerical approach, has been extensively employed for temperature field modeling in metal AM. To investigate the effects of scanning paths on temperature distribution and thermal gradients in laser- assisted AM, Ren et al. [21] developed a three- dimensional (3D) thermodynamic model. Cadiou et al. [22] constructed a 3D numerical model for temperature field prediction in the Cold Metal Transfer- Wire Arc Additive Manufacturing (CMT- WAAM) process. Lu et al. [23] developed thermodynamic calculations and thermal simulations for multi- layer and multi- pass WAAM processes using IN718 superalloy, employing the simultaneous transformation kinetics (STK) model derived from the Johnson- Mehl- Avrami- Kolmogorov (JMAK) theory for numerical simulation. Zhang et al. [24] established a numerical model for Inconel 718 during the WAAM deposition process, analyzing its position- dependent thermal history and temperature variation patterns. However, FEM require lengthy simulation times, leading to high computational costs. To accelerate the computational time for WAAM temperature field simulation, Fabbri et al. [10] proposed an explicit FDM with a novel adaptive heat flux calculation for modeling the temperature field in DED processes. Although these methods can meet the needs of long- horizon prediction, they still struggle to achieve real- time prediction. Moreover, computing high- fidelity models typically requires high- performance hardware and extensive computational time, which is computationally expensive. Therefore, it is necessary to predict temperature field by reducing computational time and costs. + +<--- Page Split ---> + +### 2.2. Data-driven temperature field prediction + +#### 2.2.1. Traditional machine learning methods + +2.2. Data- driven temperature field prediction2.2.1. Traditional machine learning methodsIn recent years, data- driven methods have been widely applied in temperature field prediction for metal AM to reduce the computational costs associated with physical modeling methods [25]. ML, as a type of data- driven approach, often utilizes simulation data for model training, providing an efficient means for temperature field prediction in metal AM [26- 29]. For instance, Multilayer Perceptron (MLP) have been used to online thermal field prediction [28], Extremely Randomized Trees models learn the relationship between designed characteristic from simulation data and the thermal field [30], FFNN have been used for replicate temperature field generated by FEM [13]. Some researchers have been dedicated to predicting temperature field using in- situ sensor data, the emergence of surrogate models based on experiment data. Tang et al. [12] proposed an online temperature field prediction method that uses ANN for thermal field mapping and reduced- order models (ROM) for thermal field reconstruction, demonstrating its effectiveness in various experiments and simulations. Kozjek et al. [31] utilized a ML- based regression model, trained on high- resolution coaxial temperature measurement data from the previous few layers, to predict the melt pool temperature along the tool path for the subsequent layer. Mozaffar and Chen et al. [32, 33] used graph neural networks(GNN) to predicted thermal responses by capturing spatiotemporal dependencies. Perumal et al. [34] proposed the use of Temporal Convolutional Networks (TCNs) for rapid prediction of thermal history. Traditional ML methods possess good time- efficiency and accuracy, enabling effective real- time temperature field prediction. However, they are limited by the model structure when dealing with long- horizon prediction tasks, and the prediction results lack physical interpretability. + +#### 2.2.1. Recurrent neural network methods + +Identification of time series characteristic is crucial in the metal AM process, as they reflect the dynamic changes during manufacturing, such as fluctuations in melt pool temperature and shape[35]. For instance, Ren et al.[14] developed an RNN- DNN model that inputs laser deposition states in the form of a 2D matrix and outputs the temperature field, suitable for single- layer deposition. Zhou et al.[36] designed a temperature state matrix to supplement the prediction of interlayer heat transfer effects in gas metal WAAM, but calculating the average temperature of nodes relies on adjacent node temperatures that cannot be measured in real time, thus making this method unsuitable for online control during actual printing. Cao et al.[37] combined GNN with RNN to predict the temperature field for complex structures of multi- layer parts, but the model is highly complex and demands significant computational resources. Wu et al.[38] constructed a surrogate model based on RNN, generating data through multi- physics simulations and experimental design to predict the temperature and geometry of the melt pool. However, RNN face gradient issues when processing long sequence data, which affects accuracy. Hu et al.[39] combined FEM models with Bi- GRU to improve prediction accuracy, but the approach requires long training times and relies on the accuracy of the FEM models. Karkaria et al. [40] developed a Bayesian LSTM model to predict temperatures at different spatial locations in DED- manufactured parts. These studies indicate that, although RNN and their variants have made progress in temperature field prediction, they still face limitations such as high data requirements, limited ability to capture long- horizon dependencies, high computational resource consumption, lack of + +<--- Page Split ---> + +explicit physical process modeling, and excessive model complexity. These challenges limit their applicability and real- time capability in complex manufacturing processes. + +#### 2.2.3. Physics-informed Machine Learning Methods + +To ensure that model predictions comply with physical laws and to enhance the generalizability, interpretability, and robustness of models, some researchers have applied PIML to the field of metal AM[41- 45]. For instance, Zhu et al.[46] applied physics- informed neural networks(PINN) to predict temperature and melt pool fluid dynamics during the laser powder bed fusion (L- PBF) process, training the model with data from FEM. Xie et al.[47] proposed a PINN for predicting 3D temperature field in single or multi- layer DED processes. This model embeds the heat conduction PDE into the loss function, allowing the model to be trained on data while also being constrained by the laws of physics. Liao et al.[48] applied PINN to design a hybrid physics data- driven thermal model for the metal AM process to predict the full- field temperature history and identify material or process parameters from partially observed temperature data. Based on partial experiment temperature data collected by a coaxial infrared radiometer, the established model can predict the entire temperature field of thin- wall. However, the time required for model training limits its application in real- time monitoring. To address the dependency of data- driven ML models on large labeled datasets, Li et al.[49] proposed a PINN framework that does not require any labeled dataset to predict the 3D temperature field in the laser metal deposition (LMD) process. This model replaces the original data loss with physical losses of heat conduction, convection, and radiation. However, it requires a significant amount of training time and cannot achieve real- time prediction. To address the above problems, this study proposes a model that fuses PIML with RNN. The model can not only effectively learn geometric characteristic and physical information, but also accurately recognize the spatiotemporal correlation information of the data, while ensuring that the prediction results strictly follow the physical laws. + +## 3. Methodology + +### 3.1. System overview + +Figure 1 illustrates the overall workflow of the real- time long- horizon temperature field prediction system for WAAM based on PIML. It primarily consists of five stages: data generation, data collection, data processing, basic training, and TL. The system first provides the material parameters, geometric characteristic, and process parameters required for data generation, followed by generating process data from FEM and practical experiments. Simulation data is exported via command streams, while experiment data is collected using a thermal camera. Data processing stage refines the collected data from both modes into temperature field, geometric characteristic, and physical information using post- processing techniques. Basic training stage involves training the model with simulation data and saving the model parameters. TL stage uses the saved model parameters to train with experiment data, thereby enhancing learning efficiency and performance. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig.1. Flowchart of WAAM real-time temperature field prediction system.
+ +### 3.2. WAAM FEM modeling + +To establish a FEM simulation dataset for the temperature field, a detailed command stream for the temperature field prediction of a thin- wall structures was executed. Simulation structure utilized ER70s- 6 low- carbon steel wire, with a base plate made of Q235b low- carbon steel. Heat source followed Goldak's double- ellipsoidal model[50]. The shape parameters of the heat source were calibrated using the wire feeding speed (WFS) and travel speed (TS). + +Figure 2 presents the established FEM simulation model, with meshing detailed in Figures 2(a) and 2(b). The model consists of a \(300 \times 300 \times 10 \mathrm{~mm}^3\) base plate and a \(100 \mathrm{~mm}\) thin- wall structures, with deposition height and width obtained from experiment data. The model is discretized into a series of elements of specified dimensions, and the deposition process is discretized into continuous simulation steps. The base plate is meshed using a tetrahedral free mesh. The thin- wall structures is discretized with a uniform hexahedral mesh of \(2 \times 2 \times 2.5 \mathrm{~mm}\) , also considering the moving area of the double- ellipsoidal heat source to complete the modeling. Results of the simulation are shown in Figure 2(c), clearly depicting the thermal gradients and melt pool of the temperature field. The unidirectional deposition sequence for each layer is in the opposite direction, with odd layers deposited from right to left and even layers from left to right. This deposition sequence is adopted because non- reverse deposition processes can lead to imbalances in heat accumulation (i.e., the temperature at the arc ignition point is higher than on the opposite side) and loss of dimensional control. To ensure that the interlayer temperature remains around \(250^{\circ} \mathrm{C}\) , a \(60 \mathrm{~s}\) interlayer cooling time is set after each layer is printed. FEM simulation takes approximately 1 hour. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig.2. Schematic of the FEM simulation model. (a) top view of the model. (b) main view of the model. (c) temperature field at a certain time.
+ +## 3.3. Data collection and processing + +After completing the temperature field simulation of the thin- wall structures, this study collects and post- processes the data for training and testing the proposed model. Elements number for a single thin- wall structures is \(\mathrm{n}_{\mathrm{x}}\times \mathrm{n}_{\mathrm{y}}\times \mathrm{n}_{\mathrm{z}}\) , and nodes number is \((\mathrm{n}_{\mathrm{x}} + 1)\times (\mathrm{n}_{\mathrm{y}} + 1)\times (\mathrm{n}_{\mathrm{z}} + 1)\) . Experiment collects data at each node at every time (time resolution 0.25s). To facilitate data extraction and unify data dimensions, a 2D grid capable of accommodating the geometric shape of the structures in the dataset construction is designed, with the positions of the grid nodes set to match those of the FEM model nodes, ensuring each grid node corresponds to a FEM node. Grid nodes in the 2D grid that do not correspond to any FEM nodes can be considered as corresponding to nodes that will never be deposited. The size of the 2D grid is \(\mathrm{Ly}\times \mathrm{Lz}\) ( \(\mathrm{Ly} = \mathrm{Lz} = \mathrm{n}_{\mathrm{max}} + 1\) ), where \(\mathrm{n}_{\mathrm{max}}\) is the maximum number of elements along the \(\mathrm{n}_{\mathrm{y}}\) or \(\mathrm{n}_{\mathrm{z}}\) axes. To fully collect data from the WAAM process, it is necessary to maximize the information obtained by the 2D matrix. This study uses planes defined by the Y- axis and Z- axis for data extraction. By dividing into \(\mathrm{n}_{\mathrm{x}}\) elements along the X- axis, data from \(\mathrm{nx} + 1\) nodes can be collected. Therefore, for each time, \(3\times (\mathrm{n}_{\mathrm{x}} + 1)\) 2D matrices can be obtained for each type of matrix. The processing procedure for FEM simulation data processing procedure is shown in Figure 3, where temperature field matrices (Tt), geometric characteristic matrices (Gt), and PI matrices (St) can be collected. During the temperature field simulation, the activation of elements and temperature changes can be regarded as the state and temperature changes of the 2D grid nodes. Thus, the deposition phase and its corresponding temperature field can be described by describing the states of all 2D grid nodes. Furthermore, additional data processing is required before the collected data can be input into the model. The final form of the input dataset for this study is as follows: + +\[X_{t} = [T_{t}^{*}G_{t}^{*}S_{t}^{*}]\# (1)\] + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig 3. Workflow for FEM simulation data collection and processing.
+ +This study designs a geometric characteristic matrix Gt to describe the geometric characteristic of the additive structure at time t. Each node in the matrix represents the state of a cell in the FEM model, where "1" indicates deposited geometry and "0" indicates non- deposited geometry. To enable the model to learn geometric characteristic of future time sequences, the predicted sequence of geometric characteristics was normalized by mean processing. + +\[G_{t}^{*} = \frac{G_{1} + G_{2} + \ldots + G_{m - 1} + G_{m}}{m - 1}\# (2)\] + +Where \(G_{1}\) represents the geometric characteristic matrix at the second moment, \(G_{m}\) represents the geometric characteristic matrix at the second- to- last moment, and \(G_{t}^{*}\) denotes the matrix resulting from the mean normalization process applied to the input geometric characteristic matrices from the second to the second- to- last sequence. Consequently, the original geometric characteristic matrices from \(G_{1}\) to \(G_{m}\) are transformed into \(G_{t}^{*}\) . + +This study designs a 2D matrix Tt to describe the temperature field of the additive structure at time t. The temperature field matrix Tt stores the temperature values of each node at each time, with temperatures at non- deposited geometric locations set to ambient temperature. Additionally, a PI matrix St is designed to describe the physical information of the additive structure at time t. Arc power, heat source shape parameters, and heat source position are considered important physical parameters affecting WAAM temperature field modeling. This study does not directly input these parameters as raw characteristic into the model; instead, it creates an auxiliary PI input, embedding physical knowledge related to heat input into the input layer of the neural network. This intermediate characteristic integrates the effects of process parameters to enhance the model's performance and interpretability. Before inputting the temperature field matrix and PI matrix into the model, they are normalized to a range of 0 to 1: + +\[T_{t}^{*} = \frac{T_{t} - T_{\mathrm{min}}}{T_{\mathrm{max}} - T_{\mathrm{min}}}\# (3)\] \[S_{t}^{*} = \frac{S_{t} - S_{\mathrm{min}}}{S_{\mathrm{max}} - S_{\mathrm{min}}}\# (4)\] + +<--- Page Split ---> + +Where \(T_{\mathrm{max}}\) and \(T_{\mathrm{min}}\) represent the highest and lowest temperature across all time in the temperature field, respectively, and \(T_{\mathrm{r}}^{*}\) denotes the normalized temperature field matrix; \(S_{\mathrm{max}}\) and \(S_{\mathrm{min}}\) represent the maximum and minimum heat fluxes across all time in the PI matrix, respectively, and \(S_{\mathrm{r}}^{*}\) denotes the normalized PI matrix. + +## 3.4. Physics-informed machine learning model + +## 3.4.1. Model structure + +As a variant of RNN, Convolutional Long Short- Term Memory (ConvLSTM) cell employ a hybrid architecture that seamlessly integrates the spatial characteristic extraction capabilities of CNNs with the memory and sequential modeling capabilities of LSTM, enabling the processing of both temporal and spatial data[51]. The task in this study is to take the temperature field, geometric characteristic, and physical information as inputs, with the goal of predicting the long- horizon temperature field at future. The structure of the constructed PIGeoRNN model is shown in Figure 4. It uses two ConvLSTM cells to extract spatiotemporal characteristic. The encoding layer performs characteristic extraction on the input characteristic matrices, and the decoding layer transforms the output of the ConvLSTM cells back to their original dimensions. \(X_{0}\) to \(X_{\mathrm{n}}\) use real input data, while subsequent times use the model's trained output as the input for the next time. With the aid of ConvLSTM cells, the model can allow both standard temporal memory and spatiotemporal memory to flow simultaneously. + +![](images/Figure_4.jpg) + +
Fig. 4. Proposed PIGeoRNN model architecture.
+ +Hard- encoding of I/BCs aims to formulate a well- posed optimization problem during network training, enhancing solution accuracy and promoting convergence, with the key being the precise embedding of known I/BCs information into the network. ICs are imposed through model inputs. The designed data type features a uniformly discretized solution domain, facilitating + +<--- Page Split ---> + +node- by- node filling and transfer. Figure 5 illustrates the approach to hard- encoding of BCs. To facilitate the calculation of gradients using the FDM, Ghost nodes are introduced for boundary filling. The upper boundary employs Dirichlet BCs, where the known temperatures on the boundary are used to fill the Ghost nodes. + +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i} - d)\# (5)\] + +Where \(T_{\mathrm{ghost}}\) and \(T_{\mathrm{boundary}}\) represent the ghost and boundary nodes temperature respectively, \(d\) is the size of the grid space distance. The lower boundary employs Robin BCs, and the values of the Ghost nodes can be derived from the temperature values of the lower boundary nodes using the FDM. + +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i} + d) - \frac{d h_{c}}{k}\big(T_{\mathrm{boundary}}(x_{i},y_{i} + d) - T_{0}\big)\# (6)\] + +Where \(k\) is the thermal conductivity, \(h_{c}\) is thermal conductivity of substrate material, \(T_{0}\) is ambient temperature. Considering the effects of convective and radiative heat transfer, the left and right boundaries employ convective- radiative BCs, and the values of the Ghost nodes are derived from the temperature values of the left and right boundary nodes. + +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i}\pm d)\pm \frac{d}{k} (\sigma \epsilon \big(T_{\mathrm{boundary}}(x_{i},y_{i}\pm d)^{4} - T_{0}^{4}\big)\] \[\qquad +h_{a}(T_{\mathrm{boundary}}(x_{i},y_{i}\pm d) - T_{0})\big)\# (7)\] + +Where \(\sigma\) is thermal emissivity, \(\epsilon\) is Stefan- boltzmann constant, and \(h_{a}\) is heat convection coefficient of substrate and air interaction. Although the relationship between the Ghost nodes and the internal nodes is fixed, their specific values change as the computation progresses because they depend on the values of the internal nodes. Purpose of boundary filling operations is to ensure the minimization of the residual of the PDE across the entire solution domain without compromising any boundary nodes. + +![](images/Figure_5.jpg) + +
Fig. 5. Hard-encoding of BCs.
+ +In this study, physical information is not directly input into the model as raw features, but is instead embedded through an intermediate feature related to heat input. This intermediate feature integrates the effects of process parameters and is merged with other feature matrices. The higher- order parameter selected in this study is the arc heat source term, which aims to integrate arc heat source information into the historical temperature field to enhance the modeling effect. Additionally, to enable the model to learn the geometric characteristic of the structure, a characteristic matrix of the same size as the temperature field matrix is constructed, + +<--- Page Split ---> + +with its values defined by the deposition state of the model. The geometric matrix and the PI matrix are defined in detail in Section 3.3. + +#### 3.4.2. Loss function design + +The loss function of a PIML model consists of two core elements: data- based loss and physics- based loss. The total loss of the model can be expressed as: + +\[L_{total} = w_{p}L_{pde} + w_{d}L_{data}\# (\Theta)\] + +Where \(w_{p}\) and \(w_{d}\) are the weights of \(L_{pde}\) and \(L_{data}\) , and the default values are set to 0.3 and 0.7. + +Due to the hard enforcement of I/BCs, the physical loss only needs to control the loss function of the partial differential equation (PDE). To compute the temperature gradients across time and space, the computational domain is discretized into \(m \times n\) nodes with a spatial distance of \(d\) . Inspired by the FDM, differential approximations are used to calculate the derivative terms of the heat conduction governing equations. The loss of the PDE can be represented by a difference equation, the difference form is: + +\[R_{t}(x_{i},y_{j}) = \rho C_{p}\frac{\left(\widehat{T}_{t + 1}(x_{i},y_{j}) - \widehat{T}_{t - 1}(x_{i},y_{j})\right)}{2\Delta t}\] \[-\frac{1}{d^{2}} k(\widehat{T}_{t}(x_{i} + d,y_{j}) + \widehat{T}_{t}(x_{i} - d,y_{j}) + \widehat{T}_{t}(x_{i},y_{j} + d)\] \[\qquad +\widehat{T}_{t}(x_{i},y_{i} - d) - 4\widehat{T}_{t}(x_{i},y_{i})) - \dot{Q}\# (9)\] + +Where \(\widehat{T} (x_{i},y_{j})\) denotes the predicted temperature field at position \((x_{i},y_{j})\) at time \(t\) , and \(\widehat{T}_{t + 1}(x_{i},y_{j})\) and \(\widehat{T}_{t - 1}(x_{i},y_{j})\) represent the temperature at the same node at \(t + 1\) and \(t - 1\) , respectively, \(\rho\) is the density of the part material, \(C_{p}\) is the specific heat, \(\dot{Q}\) is the arc heat source term. Since the model's output is a 2D matrix, it is necessary to calculate the physical residuals of the predicted values at each position. Ideally, Model's predicted temperature should be such that \(R_{t}(x_{i},y_{j})\) is close to zero. Physical constraints can be imposed on the model by minimizing the PDE loss function \(L_{pde}\) , whose value corresponds to the squared mean of the residuals of the PDE over the spatio- temporal discretization, which can be expressed as: + +\[L_{pde} = \frac{1}{nmT}\sum_{i = 1}^{n}\sum_{i = 1}^{m}\sum_{t = 1}^{T}R_{t}(x_{i},y_{j})^{2}\# (10)\] + +Where \(n\) and \(m\) represent the number of nodes in the spatial grid in the direction of height and width, respectively, and \(t\) is the total time for prediction. The data loss \(L_{data}\) can be expressed as: + +\[L_{data} = \frac{1}{nmT}\sum_{i = 1}^{n}\sum_{i = 1}^{m}\sum_{t = 1}^{T}\left[\widehat{T}_{t}(x_{i},y_{j}) - T_{t}(x_{i},y_{j})\right]^{2}\# (11)\] + +Where \(T_{t}(x_{i},y_{j})\) is the true temperature. + +#### 3.4.3. Training and test methods + +Initially, the real- time long- horizon temperature field prediction capabilities of four models were evaluated using FEM simulation data: the Baseline model relying solely on ConvLSTM, the GeoRNN model incorporating geometric features, the PIRNN model integrating physical information, and the PIGeoRNN model that fuses both geometric features and physical + +<--- Page Split ---> + +information. Subsequently, the performance of the proposed models was trained and tested on experiment data. Building on models trained with FEM simulation, further TL training was conducted using experiment data[52]. Experiment data provide more realistic and specific scenarios, helping the model adapt to the complexities of actual manufacturing processes and enhancing its predictive accuracy and adaptability. The dataset was split into training and testing sets in a 7:3 ratio. The learning rate was set to 0.01, the batch size to 8, and training was halted after 3000 epochs. The test set was evaluated every 500 epochs, recording the mean squared error (MSE), saving the trained model parameters, and the test results. For all models, the patch size was set to \(3 \times 3\) so that each \(51 \times 51\) characteristic matrix is converted \(17 \times 17 \times 9\) tensor. The same hyperparameters were considered for the all models proposed in this study, with the networks trained using the stochastic gradient descent Adam optimizer. + +In order to assess the performance of the proposed model based on the prediction results, the use of MSE as an evaluation metric for the model can effectively assess the prediction performance of the model at each point in time. This metrics approach ensures a nuanced understanding of precision, error magnitude, and accuracy percentage, thus enhancing the overall assessment of the model. When evaluating the long- horizon temperature field prediction capabilities of different models, the average Root Mean Square Error (RMSE) is used to quantify the differences between predicted and true temperatures. The Kullback- Leibler (KL) divergence is used to evaluate the difference between the predicted and true probability distributions. The true and predicted temperatures are normalized to convert them into probability distributions, resulting in the predicted probability distribution \(Q_t\) and the true probability distribution \(P_t\) . + +\[Q_{t}(x_{i},y_{j}) = \frac{\widehat{T}_{t}(x_{i},y_{j})}{\sum_{i^{\prime} = 1}^{n}\sum_{j^{\prime} = 1}^{m}\widehat{T}_{t}(x_{i^{\prime}},y_{j^{\prime}})}\# (12)\] \[P_{t}(x_{i},y_{j}) = \frac{T_{t}(x_{i},y_{j})}{\sum_{i^{\prime} = 1}^{n}\sum_{j^{\prime} = 1}^{m}T_{t}(x_{i^{\prime}},y_{j^{\prime}})}\# (13)\] + +Therefore, the spatial difference between the predicted temperature field and the true temperature field is defined as the KL divergence between \(Q_t\) and \(P_t\) . In this study, a smaller KL divergence indicates that the distribution of the true temperature field is closer to that of the predicted temperature field. + +\[\mathrm{D}_{\mathrm{KL}}(P_t||Q_t) = \sum_{t = 1}^{T}P_t\log \frac{P_t}{Q_t}\# (14)\] + +## 4. System setup and data collection + +### 4.1. Experiment system + +The experiment system consists of an ABB IRB 2600 robot, table, Fronius CMT welder, CMT wire feeder, robot controller, computer, FLIR a655sc thermal imager, and protective gas tank. The welding torch is held by an ABB IRB2600 industrial robot and moves along a pre- programmed trajectory. A Fronius CMT Advanced 4000 welder was used as the welding power source. A FLIR a655sc thermal imager was used for temperature field in- situ monitoring and data collection tasks. A computer was used as the main controller to coordinate the industrial robot, welder and thermal imager to perform the deposition tasks. The welding technique was cold + +<--- Page Split ---> + +metal transition (CMT). + +![](images/Figure_6.jpg) + +
Fig. 6. Experiment setup for WAAM system.
+ +\(①\) ABB IRB 2600 industrial robot, \(②\) worktable, \(③\) Feonius CMT Advanced 4000 welding machine, \(④\) CMT wire feeder, \(⑤\) robot controller, \(⑥\) control computer, \(⑦\) FLIR A655sc IR camera, \(⑧\) shielding + +### 4.2. Practical experiments + +In this study, experiments were carried out to deposit thin- wall structures by WAAM. A 1.2 mm welding wire made of ER70S- 6 was used for deposition. The substrate was selected as \(300 \times 300 \times 10 \mathrm{~mm}3\) mild steel plate made of Q235b. The main chemical compositions of the wire and the substrate are shown in Table 1. + +Table 1 Main chemical composition of ER70S-6 and Q235b. + +
ElementsCMnSiCuNiCr
ER70s-60.151.851.150.500.0250.50
Q235b0.21.40.350.30.30.3
+ +The experiments aim to evaluate the accuracy and adaptability of the temperature field prediction model. A total of 10 thin- wall structures were designed, each 100 mm long, composed of 20 deposition layers, using different WFS and TS groups, simulated 5 and deposited 4 thin- walled structures for training and deposited 1 thin- walled structure for test. The thin- wall structures were deposited using a bidirectional path. Due to thermal accumulation, the average temperature of each layer increased. To eliminate the impact of thermal accumulation on part formation, the dwell time for each layer was controlled at 60 s, including the time for the welding gun to move on each layer. Detailed process parameters for the actual experiment are shown in Table 2. The arc current in the experiment was 109A, the arc voltage was 10.9V, and a mixed gas with a flow rate of 17 L/min was used, mainly composed of 50% argon and 50% carbon dioxide. The experiment used a FLIR A655s thermal camera for in- situ monitoring and data collection of the temperature field during the WAAM process. + +The model was trained using the first 15 layers of data from five FEM simulations, with the last 5 layers reserved for testing. The same strategy was applied to five experiment datasets, and + +<--- Page Split ---> + +the prediction results were evaluated using MSE to assess model performance. The geometric characteristic and physical information of the experiment data were designed to match those of the simulation data, its defined by real-time captured geometric characteristics and welding gun positions. Experiment data, after being collected and cropped by the thermal camera, were processed to the same size as the FEM simulation data. Building on the training with FEM simulation data, the model was further trained with actual experiment data using TL, and the performance of the model before and after TL training was compared; the results will be discussed in Section 5.2. + +Table 2 Process parameters of the experiments. + +
WFS(m/min)TS(m/min)
Thin-wall training 140.48
Thin-wall training 250.48
Thin-wall training 360.48
Thin-wall training 470.48
Thin-wall training 580.48
Thin-wall training 64.90.5
Thin-wall training 74.90.45
Thin-wall training 85.50.45
Thin-wall training 960.4
Thin-wall test 160.45
+ +# 5. Results and discussion + +## 5.1. Temperature field prediction based on FEM data + +### 5.1.1. Comparison of prediction results + +For the given deposition geometry, the true temperature field from FEM simulation data is compared with the model's predicted results. As shown in Figure 7, the true temperature fields and the predicted results of the four models are compared for two sets of data, illustrating the temperature fields predicted for the future at 0.25 s and 1.25 s. The model-based prediction results demonstrate that geometric and PI features can serve as input datasets for WAAM temperature field prediction. Figure 8 presents a visual comparison of the prediction results and errors for the two sets of data across the four models. The Baseline, GeoRNN, PIRNN, and PIGeoRNN models predict future temperature fields with maximum errors of 8.8%-9.7%, 7.9%-9.5%, 7.6%-9.3%, and 4.6%-8.9%, respectively, showing significant differences in error magnitude. The Baseline model, relying solely on temperature field data and lacking understanding of geometric features and physical information, shows a clear deficiency in capturing the dynamic changes of the temperature field, especially in the melt pool area with large temperature gradients. This lack of information makes it difficult for the Baseline model to accurately predict the complex variations in the temperature field, leading to relatively larger errors in these critical areas. The GeoRNN model enhances the model's ability to learn real-time geometric changes by introducing a geometric feature matrix, dynamically adjusting the weights of temperature data in the deposition area, and correcting the geometric bias in the model's temperature predictions. The PIRNN model combines PIML with RNN, introducing PI inputs and a PI loss function to strictly adhere to physical laws. The GeoRNN and PIRNN models show similar performance in their predictions, with the maximum error reduced by approximately 1% compared to the Baseline model. The + +<--- Page Split ---> + +PIGeoRNN model combines the advantages of the GeoRNN and PIRNN models, reducing the maximum error by about \(4\%\) and providing the most accurate predictions. + +In terms of time efficiency, Simulation time of the FEM model is about 1 hour, while the training time of Baseline, GeoRNN, PIRNN and PIGeoRNN models are 5 min, 8 min, 9 min and 18 min, respectively, which indicates that the increase of model complexity requires more parameters to be optimized, which makes it need to spend more training time. The model training time in this study is short (compared to the automatic differentiation used in PINNs), primarily because the PDE is solved using the FDM. The main reason is that convolution operations with shift- invariant filters can update the entire field in each iteration, which is much more efficient than point- by- point training. In addition, the prediction time of the proposed model is about 12 ms, which can meet the real- time prediction demand. + +![](images/Figure_7.jpg) + +
Fig. 7. The future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models at two time points are compared with the true temperature field.
+ +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Fig. 8. Comparison of the errors of the future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models with the true temperature fields at two time points.
+ +To evaluate the accuracy of the PIGeoRNN model's prediction for individual nodes at each time, the temperature history of the middle node on the tenth layer was extracted, as shown in Figure 9. As shown in Figure 9(a), the model's predicted values are highly consistent with the true values in terms of overall trend and can effectively capture the changes in temperature increases and decreases. In regions where the temperature fluctuates gently, the predicted values are extremely close to the true values. However, during periods of rapid temperature change, there is a noticeable deviation between the two. Nevertheless, the model is still able to generally follow the direction of the true value changes. As shown in Figure 9(b), the errors across all time are controlled within the range of \(- 100^{\circ}\mathrm{C}\) to \(100^{\circ}\mathrm{C}\) , with the majority of the time falling within \(- 25^{\circ}\mathrm{C}\) to \(25^{\circ}\mathrm{C}\) . This indicates that the model's prediction errors are within an acceptable range and that the overall reliability of the model is high. However, the error at time 5 shows significant fluctuation and deviation. This may be due to the increased prediction time, which requires the model to process more historical information and complex spatiotemporal dependencies, leading to the accumulation and amplification of errors, thereby affecting the prediction accuracy. + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Fig. 9. (a) The temperature history and (b) Error of the middle node on the tenth layer, predicted by the PIGeoRNN model over five time.
+ +#### 5.1.2. Ablation experiment + +The training loss history of the four models on FEM simulation data and the MSE of all prediction results at each time are compared in Figure 10. Overall, the loss decreases with the decrease in epochs, indicating that the models can learn the dataset well. It can be observed from Figure 10(a) that all losses fluctuate greatly in the first 1000 epochs because the models have not yet learned the complex patterns in the data, especially for high- dimensional data. After 1000 epochs, the loss fluctuates around 5e- 5. The PIGeoRNN model shows a more gradual decrease in training loss, which is related to the complexity of the model. The PDE loss history during the training of the PIGeoRNN model tends to fluctuate within a small range after about 500 epochs, indicating that the model can learn physical information within a short number of epochs. Figure 10(b) shows that the MSE of all prediction results at each time exhibits a trend of increasing error with time, indicating that errors are accumulated over time. Long- horizon prediction may lead to further accumulation of errors, but experiments have shown that predicting the temperature field for the next 1.25 s results in lower errors. The Baseline model performs poorly in temperature field prediction at each time point, mainly because the model only learns the spatiotemporal characteristic of the temperature field data. The GeoRNN model performs better than the Baseline in temperature field prediction at each time point, as the model can learn real- time geometric characteristic to enhance robustness against environmental noise in the data. Due to physical constraints, PIRNN models are more accurate than GeoRNN models in long time predictions. The PIGeoRNN model is proven to have the best performance in prediction, with a more gradual increase in error at each time point. This is related to the model's integration of geometric characteristic inputs, PI inputs, and PI loss functions, which allows the model to learn more process knowledge and be subject to physical constraints. + +<--- Page Split ---> +![](images/Figure_10.jpg) + +
Fig. 10. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models for loss and (b) MSE per time in prediction in FEM data training.
+ +#### 5.1.3. Long-horizon predictive evaluation + +To further explore the models' performance in long- horizon prediction, this study evaluated the prediction results of the four models over 1.25s, 3.75s, 6.25s, 8.75, and 11.25 s (time resolution of 0.25 s) into the future, as shown in Figure 11. As shown in Figure 11(a), the Baseline model has the largest average RMSE in long- horizon predictions. The GeoRNN and PIRNN models show similar performance in predictions up to 6.25 s, with the PIRNN model having a smaller average RMSE beyond 6.25 s. For predictions within 6.25 s, the average RMSE of the PIGeoRNN model increases with time. However, for predictions beyond 6.25 s, the average RMSE at 8.75 s and 11.25 s is lower than that at 6.25 s, indicating that the PIGeoRNN model can perform well in long- horizon predictions. Additionally, the KL divergence was used to evaluate the temperature field distributions of the four models in long- horizon predictions, as shown in Figure 11(b). It can be observed that the Baseline and PIRNN models exhibited relatively larger KL divergence in long- horizon predictions, primarily due to the lack of long- horizon geometric feature inputs. Consequently, the GeoRNN model, which incorporates geometric features, showed a relatively smaller KL divergence. After integrating geometric features and physical information, the PIGeoRNN model was able to learn the geometric features and melt pool distribution in long- horizon predictions, resulting in the lowest KL divergence. + +![](images/Figure_11.jpg) + +
Fig. 11. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on FEM data.
+ +<--- Page Split ---> + +### 5.2. Temperature field prediction based on experiment data + +#### 5.2.1. Comparison of prediction results + +In order to test the adaptability of the real- time temperature field prediction model, a thin- walled structure was deposited using different process parameters, as shown in Figure 12. Figure 13 presents a visual comparison of the prediction results and errors for the experiment thin- wall structure data across the three models. The true temperature field reveal differences in temperature gradients in the melt pool area between experiment and simulation data, which is related to the emissivity of the thermal camera. The melt pool appears liquid during deposition, whereas the setting of 0.8 corresponds to the solid- state emissivity of the material. Consequently, the issue with the thermal camera's emissivity setting leads to larger errors in the melt pool area of the temperature field and may also result in the physical information not being well learned. The Baseline, GeoRNN, PIRNN, and PIGeoRNN models predicted the temperature field for the next 1.25 s with maximum errors of \(14.2\% - 16.5\%\) , \(13.3\% - 15.2\%\) , \(13\% - 15.2\%\) , and \(13\% - 14.2\%\) , respectively, with the PIGeoRNN still demonstrating superior performance. During the WAAM deposition process, defects such as spatter and lack of fusion can affect the distribution of the temperature field, leading to uneven temperature distribution, which in turn impacts the model's learning of geometric characteristic and physical information. Due to differences between the layer height of each deposition in the thin- wall structures and the ideal case, the geometric characteristic learned by the model may deviate from reality. + +![](images/Figure_12.jpg) + +
Fig. 12. Tested thin-wall structure.
+ +![](images/Figure_14.jpg) + + +<--- Page Split ---> + +Fig. 13. (a) Comparison of True and Predicted Temperature Fields for Baseline, GeoRNN, PIRNN, and PIGeoRNN Models at Future Times 0.25s and 1.25s; (b) Analysis of Predicted Temperature Field Error. + +#### 5.2.2. Ablation experiment + +The training loss history comparison of the four models on experiment data and the MSE of all prediction results at each time are shown in Figure 14. Overall, the loss decreases with the decrease in epochs, indicating that the models can learn the experiment data well. It can be observed from Figure 14(a) that all losses also fluctuate significantly over the first 1000 epochs. The PIGeoRNN model shows a more gradual decrease in training loss and still requires more time for training. Figure 14(b) shows that the MSE of all prediction results at each time exhibits a trend of increasing error with time, indicating that errors are accumulated over time. The PIGeoRNN model does not show a significant improvement in prediction accuracy compared to the GeoRNN and the PIRNN model. + +![](images/Figure_15.jpg) + +
Fig. 14. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models trained on experiment data in terms of loss and (b) MSE at each time of test.
+ +#### 5.2.3. Long-horizon predictive evaluation + +The long- horizon prediction performance of the four models on experimental data was further evaluated, as shown in Figure 15. As shown in Figure 15(a), The PIGeoRNN model has the lowest average RMSE across the entire time range, demonstrating the highest prediction accuracy and stability. The GeoRNN and PIRNN models perform slightly worse, while the Baseline model has the highest average RMSE and the poorest prediction accuracy. This indicates that the PIGeoRNN model has a significant advantage in handling long- horizon temperature prediction field tasks. Furthermore, Figure 15(b) shows the evaluation of the prediction results using KL divergence. It can be observed that after incorporating geometric features, the GeoRNN and PIGeoRNN models have relatively close KL divergence in long- horizon predictions, indicating better spatial distribution. + +<--- Page Split ---> +![](images/Figure_16.jpg) + +
Fig. 15. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on experiment data.
+ +#### 5.2.4. Transfer learning training + +By comparing the performance of the four models on FEM simulation data and experiment data, it can be demonstrated that the proposed PIGeoRNN model exhibits the best performance. To leverage existing knowledge to accelerate learning from experiment data, aiding the model in learning the fundamental characteristics and physical laws of the temperature field, thereby enhancing learning efficiency and performance. Building on the model trained with FEM simulation data, further TL training is conducted on experiment data using actual data. Figure 16 compares the true temperature field and the predicted temperature field with and without TL, along with the visualization of errors. The experiments indicate that the model's predictions exhibit a temperature distribution that is relatively consistent with the true temperature field. The temperature field distributions in the melt pool area and near the deposition boundaries are clearly displayed. The models without TL (No TL) and with TL predict future temperature field with maximum errors of \(13 - 14.2\%\) and \(11.2 - 14\%\) , respectively. The loss history of models with and without TL training is shown in Figure 17, proving that initializing with model parameters trained on simulation data leads to much faster loss reduction. The total loss for models with TL training can be reduced to approximately \(1\mathrm{e}^{- 5}\) in just 500 epochs, whereas models without TL require 1000 epochs to achieve the same MSE, reducing training time by \(50\%\) . + +![](images/Figure_17.jpg) + +
Fig. 16. (a) Comparison of True and Predicted Temperature Fields for TL and No TL Models at
+ +<--- Page Split ---> +![PLACEHOLDER_22_0] + +
Fig. 17. (a) Comparison of model training loss history TL or No TL and (b) MSE of predicted results at each time of test.
+ +### 5.3. Discussion + +This study aims to propose a real- time long- horizon temperature field predicted method for the WAAM process using interpretable, adaptive, and robust data- driven approaches. The main innovation of the model lies in its PIML architecture that integrates geometric features and physical information inputs, enabling long- horizon temperature field prediction. Compared with models that integrate information at different levels, this model exhibits greater interpretability and robustness when trained and used to predicted temperature field at future times using the same dataset, as demonstrated by the comparison of the model's training and prediction results on different datasets in Sections 5.1 and 5.2. + +The varying performance among different models highlights the importance of integrating geometric characteristic and physical information, which significantly impacts the model's performance. The temperature field simulation results based on FEM and experiment data indicate that the information described by inputting temperature field data alone is limited; the models primarily rely on learning patterns from historical data. However, in the WAAM process, real- time geometric characteristic affects the path and efficiency of heat conduction. Thicker deposited layers require more time to cool because heat must pass through thicker material layers before it can dissipate into the environment. Failing to adequately consider the changes in geometric characteristic during the WAAM process leads to poor performance of the Baseline model in temperature field predicted. Therefore, the GeoRNN model is constructed to utilize a real- time geometric characteristic matrix as an additional input to extract geometric characteristic information from the AM process. It dynamically adjusts the model's weights for temperature data in the deposition area, enabling the model to learn and correct predictive biases. At the cost of slightly reduced temporal efficiency, it improves the accuracy and adaptability of predictions over the Baseline model. + +During the WAAM process, changes in the temperature field are influenced by various physical mechanisms, such as heat conduction, convection, and radiation. If the model fails to adequately understand these physical mechanisms, it may lead to discrepancies between the predicted results and the actual physical processes. The PIGeoRNN model proposed in this study integrates PI inputs based on the combination of GeoRNN and PIML, enabling the model to learn + +<--- Page Split ---> + +melt pool (heat source) characteristics and adjust input weights, thus accelerating gradient descent. This model is constrained by I/BCs and a PI loss function, compelling the model to minimize the physical discrepancies between predictions and true values. This endows the model with interpretability and robustness, and it has been proven to possess the best performance. + +This study utilizes a model trained on FEM simulation data for TL on experiment data, which can effectively improve computational efficiency and accelerate model convergence. The model parameters obtained from simulation data provide an excellent starting point for the training of experiment data, thereby reducing the number of iterations and time required for training on experiment data. Since simulation data typically cover a wide range of process conditions and geometric characteristic, the model has already learned the fundamental patterns under these conditions without needing to learn all characteristic and patterns from scratch. Moreover, models trained on simulation data can capture complex physical phenomena in the WAAM process, such as the effects of heat accumulation and the trends in thermal cycling curves; this information can be directly utilized in the TL process, further improving the efficiency of model training. + +The PIGeoRNN model proposed in this study holds potential application value in feed- forward control and digital twin for the WAAM process. In feedforward control, the temperature field results predicted by the model can be directly used to guide the adjustment of process parameters (such as wire feeding speed and heat source power), to preemptively optimize deposition paths or cooling strategies, thereby actively avoiding the generation of thermal deformation and residual stress during the deposition process, enhancing the quality and performance of the formed parts. In the construction of digital twin, the model can serve as an integration interface between the physical engine and real- time data, enabling dynamic prediction of the process by establishing multi- physics field mappings of geometry- temperature- stress relationships. However, these potential applications require further experiment validation to assess their performance in actual operating conditions. + +## 6. Conclusion and future work + +The PIGeoRNN model proposed in this paper demonstrates the feasibility and effectiveness of real- time long- horizon temperature field predicted in the WAAM process, with considerable accuracy. By integrating PIML and TL strategies, this model significantly enhances prediction time- efficiency, interpretability, and robustness. The real- time long- horizon temperature field prediction model is based on RNN architecture and achieves long- horizon temperature field prediction through the fusion of multimodal spatiotemporal characteristic. The model includes an encoder- decoder module, dual- layer ConvLSTM cells, and a physical constraint module, integrating geometric characteristic and physical information to predict the temperature field for the next 1.25s based on the current 1.25s, with a prediction time of only 12 ms. The model was trained and tested based on FEM simulation data, and the predictive performance of four different hierarchical models was compared. Compared with FEM models, the PIGeoRNN model improves predictive accuracy and interpretability by integrating PI and geometric characteristic inputs while maintaining low computational costs. In addition, the experimental evaluation assessed the model's predictive performance on experimental data by additionally fabricating a thin- walled structure for test. To improve the model's learning efficiency and performance, a TL strategy was employed for the training and testing of the experiment data temperature field, and the results demonstrated the model's practicality and effectiveness in real applications, meeting the needs for + +<--- Page Split ---> + +real- time long- horizon prediction. The main contributions of this study are as follows: + +1. The PIGeoRNN model, which integrates geometric characteristic and PI inputs, a PI loss function, and ConvLSTM cells, is used for real-time long-horizon temperature field prediction in the WAAM process and holds great promise for applications requiring feed-forward control and digital twin. +2. The ConvLSTM cell can simultaneously extract spatial temperature distributions and temporal change trends, which allows for more accurate prediction and understanding of the dynamic behavior of the temperature field, thus enabling long-horizon temperature field prediction. +3. The model employs a strategy that can solve spatiotemporal PDE, imposing initial and boundary conditions as hard constraints on the model, which enhances the model's predictive accuracy and robustness. +4. Compared with Baseline, GeoRNN and PIRNN model, the PIGeoRNN model demonstrates the best performance in both simulation and experiment data temperature field prediction achieving a maximum error of only \(4.5\%\) at the cost of slightly reduced temporal efficiency. +5. Models trained using simulation data are trained in TL on experiment data, which improves the utility and efficiency of predictions and reduces the training time of the models by \(50\%\) . + +Future work can further develop the model to consider more geometrical characteristic and process parameters and optimize the AM process based on accurate and fast temperature field prediction results. Also, the construction of long- horizon predictive models for stress fields and deformations can be considered to provide a feasible solution for depositing high- quality parts. Although the prediction model has been validated only for WAAM, the model can be applied to any thermal imager- based AM process. + +## Acknowledgement + +This work is financially supported in part by the National Natural Science Foundation of China (NSFC) under Grant 52375344. + +## References + +[1] MU H, HE F, YUAN L, COMMINS P, WANG H, PAN Z. Toward a smart wire arc additive manufacturing system: A review on current developments and a framework of digital twin [J]. Journal of Manufacturing Systems, 2023, 67: 174- 89. [2] XIONG Y, TANG Y, ZHOU Q, MA Y, ROSEN D W. Intelligent additive manufacturing and design: state of the art and future perspectives [J]. Additive Manufacturing, 2022, 59. [3] NG W L, GOH G L, GOH G D, TEN J S J, YEONG W Y. Progress and Opportunities for Machine Learning in Materials and Processes of Additive Manufacturing [J]. Advanced Materials, 2024, 36(34). [4] FRAZIER W E. Metal Additive Manufacturing: A Review [J]. Journal of Materials Engineering and Performance, 2014, 23(6): 1917- 28. [5] BLAKEY- MILNER B, GRADL P, SNEDDEN G, BROOKS M, PITOT J, LOPEZ E, LEARY M, BERTO F, DU PLESSIS A. Metal additive manufacturing in aerospace: A review [J]. Materials & Design, 2021, 209. + +<--- Page Split ---> + +[6] ZHANG H, LI R, LIU J, WANG K, WEIJIAN Q, SHI L, LEI L, HE W, WU S. State-of-art review on the process-structure-properties-performance linkage in wire arc additive manufacturing [J]. Virtual and Physical Prototyping, 2024, 19(1).[7] CHANDRA S, RADHAKRISHNAN J, HUANG S, WEI S, RAMAMURTY U. Solidification in metal additive manufacturing: challenges, solutions, and opportunities [J]. Progress in Materials Science, 2025, 148. [8] WU B, PAN Z, DING D, CUIURI D, LI H, XU J, NORRISH J. A review of the wire arc additive manufacturing of metals: properties, defects and quality improvement [J]. Journal of Manufacturing Processes, 2018, 35: 127-39. [9] ZHANG J, LIOU F, SEUFZER W, TAMINGER K. A coupled finite element cellular automaton model to predict thermal history and grain morphology of Ti- 6Al- 4V during direct metal deposition (DMD) [J]. Additive Manufacturing, 2016, 11: 32-9. [10] FABBRI M, ASCHWANDEN I, WEGENER K, BAMBACH M. A finite volume scheme with an adaptive heat flux calculation for fast temperature prediction for wire arc additive manufacturing [J]. Journal of Materials Processing Technology, 2024, 324. [11] HU Z, YAN W. Data-driven modeling of process-structure-property relationships in metal additive manufacturing [J]. npj Advanced Manufacturing, 2024, 1(1).[12] TANG Y, RAHMANI DEHAGHANI M, SAJADI P, BALANI S B, DHALPE A, PANICKER S, WU D, COATANEA E, WANG G G. Online thermal field prediction for metal additive manufacturing of thin walls [J]. Journal of Manufacturing Processes, 2023, 108: 529-50. [13] PHAM T Q D, HOANG T V, VAN TRAN X, PHAM Q T, FETNI S, DUCHÈNE L, TRAN H S, HABRAKEN A- M. Fast and accurate prediction of temperature evolutions in additive manufacturing process using deep learning [J]. Journal of Intelligent Manufacturing, 2022, 34(4): 1701-19. [14] REN K, CHEW Y, ZHANG Y F, FUH J Y H, BI G J. Thermal field prediction for laser scanning paths in laser aided additive manufacturing by physics- based machine learning [J]. Computer Methods in Applied Mechanics and Engineering, 2020, 362. [15] HAMRANI A, AGARWAL A, ALLOUHI A, MCDANIEL D. Applying machine learning to wire arc additive manufacturing: a systematic data- driven literature review [J]. Journal of Intelligent Manufacturing, 2023, 35(6): 2407-39. [16] KARNIADAKIS G E, KEVREKIDIS I G, LU L, PERDIKARIS P, WANG S, YANG L. Physics- informed machine learning [J]. Nature Reviews Physics, 2021, 3(6): 422-40. [17] DE LA BATUT B, FERGANI O, BROTAN V, BAMBACH M, EL MANSOURI M. Analytical and Numerical Temperature Prediction in Direct Metal Deposition of Ti6Al4V [J]. Journal of Manufacturing and Materials Processing, 2017, 1(1).[18] WEISZ- PATRAULT D. Fast simulation of temperature and phase transitions in directed energy deposition additive manufacturing [J]. Additive Manufacturing, 2020, 31. [19] ANSARI M, KHAMOOSHI M, HUANG Y, TOYSERKANI E. Analytical solutions for rapid prediction of transient temperature field in powder- fed laser directed energy deposition based on different heat source models [J]. Applied Physics A, 2021, 127(6).[20] ANSARI M, MARTINEZ- MARCHESE A, KHAMOOSHI M, KESHAVARZKERMANI A, ESMAEILIZADEH R, TOYSERKANI E. Analytical modeling of multi- track powder- fed laser directed energy deposition: On the relationships among process, deposition dimensions, and solidification microstructure in additively manufactured near- \(\beta\) titanium alloy [J]. Journal + +<--- Page Split ---> + +of Materials Processing Technology, 2022, 306. + +[21] REN K, CHEW Y, FUH J Y H, ZHANG Y F, BI G J. Thermo-mechanical analyses for optimized path planning in laser aided additive manufacturing processes [J]. Materials & Design, 2019, 162: 80- 93. + +[22] CADIOU S, COURTOIS M, CARIN M, BERCKMANS W, LE MASSON P. 3D heat transfer, fluid flow and electromagnetic model for cold metal transfer wire arc additive manufacturing (Cmt- Waam) [J]. Additive Manufacturing, 2020, 36. + +[23] LU X, LI M V, YANG H. Simulation of precipitates evolution driven by non- isothermal cyclic thermal history during wire and arc additive manufacturing of IN718 superalloy [J]. Journal of Manufacturing Processes, 2021, 65: 258- 70. + +[24] ZHANG T, LI H, GONG H, WU Y, CHEN X, ZHANG X. Study on location- related thermal cycles and microstructure variation of additively manufactured inconel 718 [J]. Journal of Materials Research and Technology, 2022, 18: 3056- 72. + +[25] CHEN F, YANG M, YAN W. Data- driven prognostic model for temperature field in additive manufacturing based on the high- fidelity thermal- fluid flow simulation [J]. Computer Methods in Applied Mechanics and Engineering, 2022, 392. + +[26] ANANDAN KUMAR H, KUMARAGURU S, PAUL C P, BINDRA K S. Faster temperature prediction in the powder bed fusion process through the development of a surrogate model [J]. Optics & Laser Technology, 2021, 141. + +[27] LI J, JIN R, YU H Z. Integration of physically- based and data- driven approaches for thermal field prediction in additive manufacturing [J]. Materials & Design, 2018, 139: 473- 85. + +[28] XU S, RAI R, MOORE R D, ORLANDI G, ABDELJAWAD F. MPTP- Net: melt pool temperature profile network for thermal field modeling in beam shaping of laser powder bed fusion [J]. Journal of Intelligent Manufacturing, 2024. + +[29] GUO S, GUO W, BIAN L, GUO Y B. A Deep- Learning- Based Surrogate Model for Thermal Signature Prediction in Laser Metal Deposition [J]. IEEE Transactions on Automation Science and Engineering, 2023, 20(1): 482- 94. + +[30] NESS K L, PAUL A, SUN L, ZHANG Z. Towards a generic physics- based machine learning model for geometry invariant thermal history prediction in additive manufacturing [J]. Journal of Materials Processing Technology, 2022, 302. + +[31] KOZJEK D, CARTER F M, PORTER C, MOGONYE J- E, EHMANN K, CAO J. Data- driven prediction of next- layer melt pool temperatures in laser powder bed fusion based on co- axial high- resolution Planck thermometry measurements [J]. Journal of Manufacturing Processes, 2022, 79: 81- 90. + +[32] MOZAFFAR M, LIAO S, LIN H, EHMANN K, CAO J. Geometry- agnostic data- driven thermal modeling of additive manufacturing processes using graph neural networks [J]. Additive Manufacturing, 2021, 48. + +[33] CHEN J, KHRENOV M, JIN J, NARRA S P, MCCOMB C. Data- driven inpainting for full- part temperature monitoring in additive manufacturing [J]. Journal of Manufacturing Systems, 2024, 77: 558- 75. + +[34] PERUMAL V, ABUEIDDA D, KORIC S, KONTSOS A. Temporal convolutional networks for data- driven thermal modeling of directed energy deposition [J]. Journal of Manufacturing Processes, 2023, 85: 405- 16. + +[35] ZHANG Z, LIU Z, WU D. Prediction of melt pool temperature in directed energy deposition + +<--- Page Split ---> + +using machine learning [J]. Additive Manufacturing, 2021, 37. [36] ZHOU Z, SHEN H, LIU B, DU W, JIN J. Thermal field prediction for welding paths in multi-layer gas metal arc welding-based additive manufacturing: A machine learning approach [J]. Journal of Manufacturing Processes, 2021, 64: 960- 71. [37] CAO X, DUAN C, LUO X, ZHENG S, XU H, HAO X, ZHANG Z. Deep learning- based rapid prediction of temperature field and intelligent control of molten pool during directed energy deposition process [J]. Additive Manufacturing, 2024, 94. [38] WU S- H, TARIQ U, JOY R, MAHMOOD M A, MALIK A W, LIOU F. A Robust Recurrent Neural Networks- Based Surrogate Model for Thermal History and Melt Pool Characteristics in Directed Energy Deposition [J]. Materials, 2024, 17(17).[39] HU K- X, GUO K, LI W- D, WANG Y- H. Temperature evolution prediction for laser directed energy deposition enabled by finite element modelling and bi- directional gated recurrent unit [J]. Advances in Manufacturing, 2024. [40] KARKARIA V, GOECKNER A, ZHA R, CHEN J, ZHANG J, ZHU Q, CAO J, GAO R X, CHEN W. Towards a digital twin framework in additive manufacturing: Machine learning and bayesian optimization for time series process optimization [J]. Journal of Manufacturing Systems, 2024, 75: 322- 32. [41] SHENGHAN G, MOHIT A, CLAYTON C, QI T, X. G R, WEIHONG G, Y.B. G. Machine learning for metal additive manufacturing: Towards a physics- informed data- driven paradigm [J]. Journal of Manufacturing Systems, 2022, 62: 145- 63. [42] HOSSEINI E, SCHEEL P, MULLER O, MOLINARO R, MISHRA S. Single- track thermal analysis of laser powder bed fusion process: Parametric solution through physics- informed neural networks [J]. Computer Methods in Applied Mechanics and Engineering, 2023, 410. [43] ZHU Q, LU Z, HU Y. A reality- augmented adaptive physics informed machine learning method for efficient heat transfer prediction in laser melting [J]. Journal of Manufacturing Processes, 2024, 124: 444- 57. [44] TANG J, SCHEEL P, MOHEBBI M S, LEINENBACH C, DE LORENZIS L, HOSSEINI E. On the calibration of thermo- microstructural simulation models for Laser Powder Bed Fusion process: Integrating physics- informed neural networks with cellular automata [J]. Additive Manufacturing, 2024, 96. [45] PENG B, PANESAR A. Multi- layer thermal simulation using physics- informed neural network [J]. Additive Manufacturing, 2024, 95. [46] ZHU Q, LIU Z, YAN J. Machine learning for metal additive manufacturing: predicting temperature and melt pool fluid dynamics using physics- informed neural networks [J]. Computational Mechanics, 2021, 67(2): 619- 35. [47] XIE J, CHAI Z, XU L, REN X, LIU S, CHEN X. 3D temperature field prediction in direct energy deposition of metals using physics informed neural network [J]. The International Journal of Advanced Manufacturing Technology, 2022, 119(5- 6): 3449- 68. [48] SHUHENG L, TIANJU X, JIHOON J, SAMANTHA W, KORNEL E, JIAN C. Hybrid thermal modeling of additive manufacturing processes using physics- informed neural networks for temperature prediction and parameter identification [J]. Computational Mechanics, 2023, 72(3): 499- 512. [49] SHILIN L, GANG W, YUELAN D, LIPING W, HAIDOU W, QINGJUN Z. A physics- informed neural network framework to predict 3D temperature field without labeled + +<--- Page Split ---> + +data in process of laser metal deposition [J]. Engineering Applications of Artificial Intelligence, 2023, 120. [50] GOLDALK J, CHAKRAVARTI A, BIBBY M. A new finite element model for welding heat sources [J]. Metallurgical Transactions B, 1984, 15(2): 299- 305. [51] SHI X, CHEN Z, WANG H, YEUNG D- Y, WONG W- K, WOO W- C. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting [J]. Computer Science, 2015. [52] TANG Y, RAHMANI DEHAGHANI M, WANG G G. Review of transfer learning in modeling additive manufacturing processes [J]. Additive Manufacturing, 2023, 61. + +<--- Page Split ---> diff --git a/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa_det.mmd b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..de00f0d6e706f013a78440575608cab986f7e4a7 --- /dev/null +++ b/preprint/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa/preprint__134bafd5a9a92851d7638ba2ac7f57e0b5777a76e617f095c7d3c22fbc1577fa_det.mmd @@ -0,0 +1,552 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 936, 177]]<|/det|> +# PIML-based Real-time Long-horizon temperature Fields prediction in metallic additive manufacturing + +<|ref|>text<|/ref|><|det|>[[44, 196, 300, 242]]<|/det|> +Donghong Ding donghong@n.jtech.edu.cn + +<|ref|>text<|/ref|><|det|>[[42, 269, 264, 520]]<|/det|> +Nanjing Tech University mingxuan tian Nanjing Tech University Haochen Mu Nanjing Tech University Tao Liu Nanjing Tech University Mengjiao Li Nanjing Tech University Jianping Zhao Nanjing Tech University + +<|ref|>sub_title<|/ref|><|det|>[[44, 562, 104, 579]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 599, 884, 641]]<|/det|> +Keywords: DED- arc, Temperature field prediction, Physics- informed machine learning, Geometric, Transfer learning + +<|ref|>text<|/ref|><|det|>[[44, 659, 329, 678]]<|/det|> +Posted Date: October 31st, 2025 + +<|ref|>text<|/ref|><|det|>[[42, 697, 474, 717]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 6293181/v1 + +<|ref|>text<|/ref|><|det|>[[42, 735, 914, 777]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 796, 535, 815]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 852, 895, 895]]<|/det|> +Version of Record: A version of this preprint was published at Communications Engineering on September 29th, 2025. See the published version at https://doi.org/10.1038/s44172- 025- 00501- 7. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[157, 85, 838, 120]]<|/det|> +# PIML-based Real-time Long-horizon temperature Fields prediction in metallic additive manufacturing + +<|ref|>sub_title<|/ref|><|det|>[[149, 143, 217, 156]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[148, 160, 852, 584]]<|/det|> +The real- time long- horizon temperature field prediction during the Wire Arc Additive Manufacturing (WAAM) process is crucial for controlling heat accumulation, optimizing process parameters, and ensuring the quality of the manufactured parts. However, due to the time- consuming nature of finite element methods (FEM) and the process- informed neglect by existing data- driven models, real- time long- horizon temperature field prediction remains a technical challenge for control systems in metallic AM. Despite the interpretability offered by Physics- Informed Machine Learning (PIML), which still suffers from relatively long training times and a lack of long- horizon prediction capabilities for dynamic systems. To address this issue, the study proposes a Physics- informed Geometric Recurrent Neural Network (PIGeoRNN) model that fuses geometric characteristic and physical information through an encoder, extracts spatiotemporal characteristic using convolutional long short- term memory cells, and ensures that the prediction results conform to physical laws by incorporating hard- encoding initial/boundary conditions and a PI loss function. The model is capable of predicting the temperature field for the future 1.25 s based on data from the current 1.25 s, and it has also been evaluated for more long- horizon predictions. Transfer learning was used for enhance the model's efficiency in practical applications. Results demonstrate that the proposed PIGeoRNN model achieves a maximum prediction error of only \(4.5\%\) to \(13.9\%\) for both simulation and experiment. The inclusion of geometric features and physical information reduces the maximum error by approximately \(1\%\) , while the integrated model can lower the maximum error by \(4\%\) . Furthermore, transfer learning training reduces the model's training time by approximately \(50\%\) while achieving the same loss level. These findings indicate that the proposed method holds significant potential for process feed- forward control and digital twin in WAAM. It also represents an important first step toward real- time long- horizon physical fields prediction in metallic AM. + +<|ref|>text<|/ref|><|det|>[[149, 604, 836, 638]]<|/det|> +Keywords: DED- arc; Temperature field prediction; Physics- informed machine learning; Geometric; Transfer learning + +<|ref|>sub_title<|/ref|><|det|>[[149, 660, 264, 674]]<|/det|> +## 1. Introduction + +<|ref|>text<|/ref|><|det|>[[148, 678, 851, 879]]<|/det|> +With the accelerated advancement of Industry 4.0 and intelligent manufacturing, Additive Manufacturing (AM) field is entering a critical phase of technological innovation and application expansion. Against the backdrop of intelligent manufacturing, AM integrates cutting- edge technologies such as digital twin and Machine Learning (ML) to promote the automation, intelligentization, and digital transformation of production processes[1- 3]. Wire Arc Additive Manufacturing (WAAM), also known as Wire- Arc Directed Energy Deposition (Wire- Arc DED), produces near- net- shape parts by melting wire feedstock using an electric arc[4]. Characterized by high deposition efficiency and material utilization, as well as low equipment costs, this process is capable of processing high- strength materials. Therefore, it has been widely applied in fields such as aerospace, maritime, energy, and heavy industry, providing feasible solutions for the production of complex components[5]. + +<|ref|>text<|/ref|><|det|>[[180, 883, 848, 899]]<|/det|> +Although WAAM has certain advantages compared with other AM technologies, minimizing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 850, 268]]<|/det|> +the impact of the temperature field during the WAAM process on the final properties of the parts remains a challenge [6]. The spatiotemporal distribution of the temperature field during the arc additive manufacturing process directly affects the quality and performance of the printed parts. High temperatures and rapid cooling lead to the formation of different microstructures during material solidification, such as coarse columnar grains or fine equiaxed grains, which in turn affect the mechanical properties of the material[7]. The non- uniformity of the temperature field can cause uneven thermal expansion, leading to thermal deformation and the generation of residual stress, potentially inducing defects such as cracks[8]. Therefore, studying real- time long- horizon temperature field prediction helps to achieve feed- forward process control and digital twin, thereby reducing defects in WAAM parts. + +<|ref|>text<|/ref|><|det|>[[147, 272, 851, 638]]<|/det|> +To accurately predict the temperature field in the WAAM process, both physics- based and data- driven approaches have been developed. Physics- based methods can be categorized into numerical and analytical approaches. The numerical methods for predicting the temperature field in WAAM mainly include the finite element method (FEM) [9] and finite volume method (FVM) [10], these methods discretize the continuous heat conduction problem into a finite number of elements or control volumes and use numerical solution techniques to calculate the temperature distribution within each element or volume. Although FEM can achieve long- horizon modeling, its high computational cost and time- consuming nature limit its application in real- time prediction. Compared with physics- based methods, data- driven approaches have the advantage of predicting complex temperature fields with high precision and efficiency while requiring only limited fundamental knowledge of physics [11]. Several ML, e.g. artificial neural networks (ANN) [12], feedforward neural networks (FFNN) [13] and recurrent neural network (RNN) [14], have been used for temperature field prediction in metal AM processes [15]. However, purely data- driven approaches lack physical interpretability and struggle to fully capture complex spatiotemporal sequence characteristic and process- informed. While the advent of Physics- Informed Machine Learning (PIML) has provided physically interpretable methods for additive manufacturing modeling and prediction, it is still characterized by relatively long training times and lack of long- horizon prediction capabilities for dynamic systems [16]. Real- time long- horizon temperature field prediction system for the WAAM should possess characteristics such as accuracy, adaptability, and interpretability: + +<|ref|>text<|/ref|><|det|>[[182, 640, 850, 825]]<|/det|> +1. Process-informed: The WAAM deposition process encompasses a wealth of process information that the system can effectively identify and learn. +2. Transfer learning (TL): FEM simulation data are used to initialize model parameters, enabling the system to rapidly learn the fundamental characteristics and physical laws of the temperature field. +3. Adaptiveness: The system should be capable of calibrating prediction results in real-time based on geometric feedback. +4. Interpretability: The physics-informed (PI) loss function is combined with a hard constraint on the IBCs, and the system prediction results should be physically constrained. + +<|ref|>text<|/ref|><|det|>[[148, 828, 850, 900]]<|/det|> +This study aims to improve the accuracy, spatiotemporal correlation, and real- time long- horizon performance of the model for long- horizon prediction of the temperature field in WAAM by combining PIML and RNN methods. The rest of the paper is organized as follows: in Section 2, this study reviews existing related work on prediction of temperature field in metal AM + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 85, 850, 157]]<|/det|> +processes. In Section 3, the methodology for the system is presented. Then, the experiment system setup and practical experiments is presented in Section 4. Section 5 discusses the results of the model trained on experiment and simulation data, with or without TL training. Finally, Section 6 summarizes the findings of this paper and future work. + +<|ref|>sub_title<|/ref|><|det|>[[148, 179, 270, 193]]<|/det|> +## 2. Related work + +<|ref|>text<|/ref|><|det|>[[148, 197, 620, 212]]<|/det|> +2.1. Physical modeling for temperature field simulation in WAAM + +<|ref|>text<|/ref|><|det|>[[147, 215, 851, 508]]<|/det|> +Recently, analytical models of metal AM have been extensively studied. Batut et al. [17] developed an analytical model based on the assumption of a moving heat source for temperature prediction in the Direct Metal Deposition (DMD) process for titanium alloy Ti6Al4V. Daniel et al. [18] proposed a simplified macroscopic semi- analytical thermal analysis method to simulate temperature and phase transformations during the DED process. This method enables efficient computational simulation of the entire process, taking into account the effects of solidification and solid- state phase transformations. Ansari et al. [19] established an analytical method to predict transient temperature field variations during the Laser DED process. This study compared three heat source models and found that the two- dimensional (2D) Gaussian heat source model had the best accuracy. Subsequently, they developed a physics- based analytical model to investigate the impact of different scanning strategies on the temperature field [20]. Although analytical models are used for in- situ monitoring in AM processes, the effectiveness of their performance is still questioned. On the one hand, analytical models based on physical principles have limitations in dealing with uncertainties and variances that arise during the manufacturing process; on the other hand, these models often struggle to accurately reflect complex thermal behaviors under all boundary conditions. + +<|ref|>text<|/ref|><|det|>[[147, 512, 851, 899]]<|/det|> +Numerical models constructed for metal AM processes have been proven to more effectively predict thermal distribution under given boundary conditions, demonstrating greater adaptability and accuracy. The FEM as a commonly used numerical approach, has been extensively employed for temperature field modeling in metal AM. To investigate the effects of scanning paths on temperature distribution and thermal gradients in laser- assisted AM, Ren et al. [21] developed a three- dimensional (3D) thermodynamic model. Cadiou et al. [22] constructed a 3D numerical model for temperature field prediction in the Cold Metal Transfer- Wire Arc Additive Manufacturing (CMT- WAAM) process. Lu et al. [23] developed thermodynamic calculations and thermal simulations for multi- layer and multi- pass WAAM processes using IN718 superalloy, employing the simultaneous transformation kinetics (STK) model derived from the Johnson- Mehl- Avrami- Kolmogorov (JMAK) theory for numerical simulation. Zhang et al. [24] established a numerical model for Inconel 718 during the WAAM deposition process, analyzing its position- dependent thermal history and temperature variation patterns. However, FEM require lengthy simulation times, leading to high computational costs. To accelerate the computational time for WAAM temperature field simulation, Fabbri et al. [10] proposed an explicit FDM with a novel adaptive heat flux calculation for modeling the temperature field in DED processes. Although these methods can meet the needs of long- horizon prediction, they still struggle to achieve real- time prediction. Moreover, computing high- fidelity models typically requires high- performance hardware and extensive computational time, which is computationally expensive. Therefore, it is necessary to predict temperature field by reducing computational time and costs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[149, 104, 468, 120]]<|/det|> +### 2.2. Data-driven temperature field prediction + +<|ref|>title<|/ref|><|det|>[[149, 123, 466, 138]]<|/det|> +#### 2.2.1. Traditional machine learning methods + +<|ref|>text<|/ref|><|det|>[[148, 141, 851, 510]]<|/det|> +2.2. Data- driven temperature field prediction2.2.1. Traditional machine learning methodsIn recent years, data- driven methods have been widely applied in temperature field prediction for metal AM to reduce the computational costs associated with physical modeling methods [25]. ML, as a type of data- driven approach, often utilizes simulation data for model training, providing an efficient means for temperature field prediction in metal AM [26- 29]. For instance, Multilayer Perceptron (MLP) have been used to online thermal field prediction [28], Extremely Randomized Trees models learn the relationship between designed characteristic from simulation data and the thermal field [30], FFNN have been used for replicate temperature field generated by FEM [13]. Some researchers have been dedicated to predicting temperature field using in- situ sensor data, the emergence of surrogate models based on experiment data. Tang et al. [12] proposed an online temperature field prediction method that uses ANN for thermal field mapping and reduced- order models (ROM) for thermal field reconstruction, demonstrating its effectiveness in various experiments and simulations. Kozjek et al. [31] utilized a ML- based regression model, trained on high- resolution coaxial temperature measurement data from the previous few layers, to predict the melt pool temperature along the tool path for the subsequent layer. Mozaffar and Chen et al. [32, 33] used graph neural networks(GNN) to predicted thermal responses by capturing spatiotemporal dependencies. Perumal et al. [34] proposed the use of Temporal Convolutional Networks (TCNs) for rapid prediction of thermal history. Traditional ML methods possess good time- efficiency and accuracy, enabling effective real- time temperature field prediction. However, they are limited by the model structure when dealing with long- horizon prediction tasks, and the prediction results lack physical interpretability. + +<|ref|>title<|/ref|><|det|>[[149, 531, 440, 546]]<|/det|> +#### 2.2.1. Recurrent neural network methods + +<|ref|>text<|/ref|><|det|>[[148, 549, 851, 900]]<|/det|> +Identification of time series characteristic is crucial in the metal AM process, as they reflect the dynamic changes during manufacturing, such as fluctuations in melt pool temperature and shape[35]. For instance, Ren et al.[14] developed an RNN- DNN model that inputs laser deposition states in the form of a 2D matrix and outputs the temperature field, suitable for single- layer deposition. Zhou et al.[36] designed a temperature state matrix to supplement the prediction of interlayer heat transfer effects in gas metal WAAM, but calculating the average temperature of nodes relies on adjacent node temperatures that cannot be measured in real time, thus making this method unsuitable for online control during actual printing. Cao et al.[37] combined GNN with RNN to predict the temperature field for complex structures of multi- layer parts, but the model is highly complex and demands significant computational resources. Wu et al.[38] constructed a surrogate model based on RNN, generating data through multi- physics simulations and experimental design to predict the temperature and geometry of the melt pool. However, RNN face gradient issues when processing long sequence data, which affects accuracy. Hu et al.[39] combined FEM models with Bi- GRU to improve prediction accuracy, but the approach requires long training times and relies on the accuracy of the FEM models. Karkaria et al. [40] developed a Bayesian LSTM model to predict temperatures at different spatial locations in DED- manufactured parts. These studies indicate that, although RNN and their variants have made progress in temperature field prediction, they still face limitations such as high data requirements, limited ability to capture long- horizon dependencies, high computational resource consumption, lack of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[149, 85, 850, 120]]<|/det|> +explicit physical process modeling, and excessive model complexity. These challenges limit their applicability and real- time capability in complex manufacturing processes. + +<|ref|>title<|/ref|><|det|>[[149, 142, 520, 157]]<|/det|> +#### 2.2.3. Physics-informed Machine Learning Methods + +<|ref|>text<|/ref|><|det|>[[148, 160, 851, 547]]<|/det|> +To ensure that model predictions comply with physical laws and to enhance the generalizability, interpretability, and robustness of models, some researchers have applied PIML to the field of metal AM[41- 45]. For instance, Zhu et al.[46] applied physics- informed neural networks(PINN) to predict temperature and melt pool fluid dynamics during the laser powder bed fusion (L- PBF) process, training the model with data from FEM. Xie et al.[47] proposed a PINN for predicting 3D temperature field in single or multi- layer DED processes. This model embeds the heat conduction PDE into the loss function, allowing the model to be trained on data while also being constrained by the laws of physics. Liao et al.[48] applied PINN to design a hybrid physics data- driven thermal model for the metal AM process to predict the full- field temperature history and identify material or process parameters from partially observed temperature data. Based on partial experiment temperature data collected by a coaxial infrared radiometer, the established model can predict the entire temperature field of thin- wall. However, the time required for model training limits its application in real- time monitoring. To address the dependency of data- driven ML models on large labeled datasets, Li et al.[49] proposed a PINN framework that does not require any labeled dataset to predict the 3D temperature field in the laser metal deposition (LMD) process. This model replaces the original data loss with physical losses of heat conduction, convection, and radiation. However, it requires a significant amount of training time and cannot achieve real- time prediction. To address the above problems, this study proposes a model that fuses PIML with RNN. The model can not only effectively learn geometric characteristic and physical information, but also accurately recognize the spatiotemporal correlation information of the data, while ensuring that the prediction results strictly follow the physical laws. + +<|ref|>sub_title<|/ref|><|det|>[[148, 567, 271, 581]]<|/det|> +## 3. Methodology + +<|ref|>sub_title<|/ref|><|det|>[[148, 586, 303, 600]]<|/det|> +### 3.1. System overview + +<|ref|>text<|/ref|><|det|>[[148, 604, 851, 806]]<|/det|> +Figure 1 illustrates the overall workflow of the real- time long- horizon temperature field prediction system for WAAM based on PIML. It primarily consists of five stages: data generation, data collection, data processing, basic training, and TL. The system first provides the material parameters, geometric characteristic, and process parameters required for data generation, followed by generating process data from FEM and practical experiments. Simulation data is exported via command streams, while experiment data is collected using a thermal camera. Data processing stage refines the collected data from both modes into temperature field, geometric characteristic, and physical information using post- processing techniques. Basic training stage involves training the model with simulation data and saving the model parameters. TL stage uses the saved model parameters to train with experiment data, thereby enhancing learning efficiency and performance. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[203, 85, 826, 396]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[232, 401, 760, 417]]<|/det|> +
Fig.1. Flowchart of WAAM real-time temperature field prediction system.
+ +<|ref|>sub_title<|/ref|><|det|>[[149, 439, 350, 454]]<|/det|> +### 3.2. WAAM FEM modeling + +<|ref|>text<|/ref|><|det|>[[148, 457, 850, 547]]<|/det|> +To establish a FEM simulation dataset for the temperature field, a detailed command stream for the temperature field prediction of a thin- wall structures was executed. Simulation structure utilized ER70s- 6 low- carbon steel wire, with a base plate made of Q235b low- carbon steel. Heat source followed Goldak's double- ellipsoidal model[50]. The shape parameters of the heat source were calibrated using the wire feeding speed (WFS) and travel speed (TS). + +<|ref|>text<|/ref|><|det|>[[147, 549, 851, 825]]<|/det|> +Figure 2 presents the established FEM simulation model, with meshing detailed in Figures 2(a) and 2(b). The model consists of a \(300 \times 300 \times 10 \mathrm{~mm}^3\) base plate and a \(100 \mathrm{~mm}\) thin- wall structures, with deposition height and width obtained from experiment data. The model is discretized into a series of elements of specified dimensions, and the deposition process is discretized into continuous simulation steps. The base plate is meshed using a tetrahedral free mesh. The thin- wall structures is discretized with a uniform hexahedral mesh of \(2 \times 2 \times 2.5 \mathrm{~mm}\) , also considering the moving area of the double- ellipsoidal heat source to complete the modeling. Results of the simulation are shown in Figure 2(c), clearly depicting the thermal gradients and melt pool of the temperature field. The unidirectional deposition sequence for each layer is in the opposite direction, with odd layers deposited from right to left and even layers from left to right. This deposition sequence is adopted because non- reverse deposition processes can lead to imbalances in heat accumulation (i.e., the temperature at the arc ignition point is higher than on the opposite side) and loss of dimensional control. To ensure that the interlayer temperature remains around \(250^{\circ} \mathrm{C}\) , a \(60 \mathrm{~s}\) interlayer cooling time is set after each layer is printed. FEM simulation takes approximately 1 hour. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[159, 90, 838, 319]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[159, 327, 838, 362]]<|/det|> +
Fig.2. Schematic of the FEM simulation model. (a) top view of the model. (b) main view of the model. (c) temperature field at a certain time.
+ +<|ref|>sub_title<|/ref|><|det|>[[149, 382, 400, 398]]<|/det|> +## 3.3. Data collection and processing + +<|ref|>text<|/ref|><|det|>[[147, 400, 852, 789]]<|/det|> +After completing the temperature field simulation of the thin- wall structures, this study collects and post- processes the data for training and testing the proposed model. Elements number for a single thin- wall structures is \(\mathrm{n}_{\mathrm{x}}\times \mathrm{n}_{\mathrm{y}}\times \mathrm{n}_{\mathrm{z}}\) , and nodes number is \((\mathrm{n}_{\mathrm{x}} + 1)\times (\mathrm{n}_{\mathrm{y}} + 1)\times (\mathrm{n}_{\mathrm{z}} + 1)\) . Experiment collects data at each node at every time (time resolution 0.25s). To facilitate data extraction and unify data dimensions, a 2D grid capable of accommodating the geometric shape of the structures in the dataset construction is designed, with the positions of the grid nodes set to match those of the FEM model nodes, ensuring each grid node corresponds to a FEM node. Grid nodes in the 2D grid that do not correspond to any FEM nodes can be considered as corresponding to nodes that will never be deposited. The size of the 2D grid is \(\mathrm{Ly}\times \mathrm{Lz}\) ( \(\mathrm{Ly} = \mathrm{Lz} = \mathrm{n}_{\mathrm{max}} + 1\) ), where \(\mathrm{n}_{\mathrm{max}}\) is the maximum number of elements along the \(\mathrm{n}_{\mathrm{y}}\) or \(\mathrm{n}_{\mathrm{z}}\) axes. To fully collect data from the WAAM process, it is necessary to maximize the information obtained by the 2D matrix. This study uses planes defined by the Y- axis and Z- axis for data extraction. By dividing into \(\mathrm{n}_{\mathrm{x}}\) elements along the X- axis, data from \(\mathrm{nx} + 1\) nodes can be collected. Therefore, for each time, \(3\times (\mathrm{n}_{\mathrm{x}} + 1)\) 2D matrices can be obtained for each type of matrix. The processing procedure for FEM simulation data processing procedure is shown in Figure 3, where temperature field matrices (Tt), geometric characteristic matrices (Gt), and PI matrices (St) can be collected. During the temperature field simulation, the activation of elements and temperature changes can be regarded as the state and temperature changes of the 2D grid nodes. Thus, the deposition phase and its corresponding temperature field can be described by describing the states of all 2D grid nodes. Furthermore, additional data processing is required before the collected data can be input into the model. The final form of the input dataset for this study is as follows: + +<|ref|>equation<|/ref|><|det|>[[425, 790, 570, 808]]<|/det|> +\[X_{t} = [T_{t}^{*}G_{t}^{*}S_{t}^{*}]\# (1)\] + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[181, 85, 820, 350]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[253, 364, 741, 380]]<|/det|> +
Fig 3. Workflow for FEM simulation data collection and processing.
+ +<|ref|>text<|/ref|><|det|>[[147, 400, 850, 492]]<|/det|> +This study designs a geometric characteristic matrix Gt to describe the geometric characteristic of the additive structure at time t. Each node in the matrix represents the state of a cell in the FEM model, where "1" indicates deposited geometry and "0" indicates non- deposited geometry. To enable the model to learn geometric characteristic of future time sequences, the predicted sequence of geometric characteristics was normalized by mean processing. + +<|ref|>equation<|/ref|><|det|>[[357, 496, 639, 529]]<|/det|> +\[G_{t}^{*} = \frac{G_{1} + G_{2} + \ldots + G_{m - 1} + G_{m}}{m - 1}\# (2)\] + +<|ref|>text<|/ref|><|det|>[[147, 530, 850, 621]]<|/det|> +Where \(G_{1}\) represents the geometric characteristic matrix at the second moment, \(G_{m}\) represents the geometric characteristic matrix at the second- to- last moment, and \(G_{t}^{*}\) denotes the matrix resulting from the mean normalization process applied to the input geometric characteristic matrices from the second to the second- to- last sequence. Consequently, the original geometric characteristic matrices from \(G_{1}\) to \(G_{m}\) are transformed into \(G_{t}^{*}\) . + +<|ref|>text<|/ref|><|det|>[[147, 623, 850, 825]]<|/det|> +This study designs a 2D matrix Tt to describe the temperature field of the additive structure at time t. The temperature field matrix Tt stores the temperature values of each node at each time, with temperatures at non- deposited geometric locations set to ambient temperature. Additionally, a PI matrix St is designed to describe the physical information of the additive structure at time t. Arc power, heat source shape parameters, and heat source position are considered important physical parameters affecting WAAM temperature field modeling. This study does not directly input these parameters as raw characteristic into the model; instead, it creates an auxiliary PI input, embedding physical knowledge related to heat input into the input layer of the neural network. This intermediate characteristic integrates the effects of process parameters to enhance the model's performance and interpretability. Before inputting the temperature field matrix and PI matrix into the model, they are normalized to a range of 0 to 1: + +<|ref|>equation<|/ref|><|det|>[[412, 828, 582, 899]]<|/det|> +\[T_{t}^{*} = \frac{T_{t} - T_{\mathrm{min}}}{T_{\mathrm{max}} - T_{\mathrm{min}}}\# (3)\] \[S_{t}^{*} = \frac{S_{t} - S_{\mathrm{min}}}{S_{\mathrm{max}} - S_{\mathrm{min}}}\# (4)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 85, 850, 158]]<|/det|> +Where \(T_{\mathrm{max}}\) and \(T_{\mathrm{min}}\) represent the highest and lowest temperature across all time in the temperature field, respectively, and \(T_{\mathrm{r}}^{*}\) denotes the normalized temperature field matrix; \(S_{\mathrm{max}}\) and \(S_{\mathrm{min}}\) represent the maximum and minimum heat fluxes across all time in the PI matrix, respectively, and \(S_{\mathrm{r}}^{*}\) denotes the normalized PI matrix. + +<|ref|>sub_title<|/ref|><|det|>[[149, 178, 480, 194]]<|/det|> +## 3.4. Physics-informed machine learning model + +<|ref|>sub_title<|/ref|><|det|>[[149, 198, 308, 212]]<|/det|> +## 3.4.1. Model structure + +<|ref|>text<|/ref|><|det|>[[147, 215, 851, 435]]<|/det|> +As a variant of RNN, Convolutional Long Short- Term Memory (ConvLSTM) cell employ a hybrid architecture that seamlessly integrates the spatial characteristic extraction capabilities of CNNs with the memory and sequential modeling capabilities of LSTM, enabling the processing of both temporal and spatial data[51]. The task in this study is to take the temperature field, geometric characteristic, and physical information as inputs, with the goal of predicting the long- horizon temperature field at future. The structure of the constructed PIGeoRNN model is shown in Figure 4. It uses two ConvLSTM cells to extract spatiotemporal characteristic. The encoding layer performs characteristic extraction on the input characteristic matrices, and the decoding layer transforms the output of the ConvLSTM cells back to their original dimensions. \(X_{0}\) to \(X_{\mathrm{n}}\) use real input data, while subsequent times use the model's trained output as the input for the next time. With the aid of ConvLSTM cells, the model can allow both standard temporal memory and spatiotemporal memory to flow simultaneously. + +<|ref|>image<|/ref|><|det|>[[222, 444, 775, 780]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[324, 790, 670, 806]]<|/det|> +
Fig. 4. Proposed PIGeoRNN model architecture.
+ +<|ref|>text<|/ref|><|det|>[[147, 826, 850, 899]]<|/det|> +Hard- encoding of I/BCs aims to formulate a well- posed optimization problem during network training, enhancing solution accuracy and promoting convergence, with the key being the precise embedding of known I/BCs information into the network. ICs are imposed through model inputs. The designed data type features a uniformly discretized solution domain, facilitating + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 850, 157]]<|/det|> +node- by- node filling and transfer. Figure 5 illustrates the approach to hard- encoding of BCs. To facilitate the calculation of gradients using the FDM, Ghost nodes are introduced for boundary filling. The upper boundary employs Dirichlet BCs, where the known temperatures on the boundary are used to fill the Ghost nodes. + +<|ref|>equation<|/ref|><|det|>[[350, 159, 646, 177]]<|/det|> +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i} - d)\# (5)\] + +<|ref|>text<|/ref|><|det|>[[147, 178, 850, 250]]<|/det|> +Where \(T_{\mathrm{ghost}}\) and \(T_{\mathrm{boundary}}\) represent the ghost and boundary nodes temperature respectively, \(d\) is the size of the grid space distance. The lower boundary employs Robin BCs, and the values of the Ghost nodes can be derived from the temperature values of the lower boundary nodes using the FDM. + +<|ref|>equation<|/ref|><|det|>[[223, 255, 772, 287]]<|/det|> +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i} + d) - \frac{d h_{c}}{k}\big(T_{\mathrm{boundary}}(x_{i},y_{i} + d) - T_{0}\big)\# (6)\] + +<|ref|>text<|/ref|><|det|>[[147, 290, 850, 362]]<|/det|> +Where \(k\) is the thermal conductivity, \(h_{c}\) is thermal conductivity of substrate material, \(T_{0}\) is ambient temperature. Considering the effects of convective and radiative heat transfer, the left and right boundaries employ convective- radiative BCs, and the values of the Ghost nodes are derived from the temperature values of the left and right boundary nodes. + +<|ref|>equation<|/ref|><|det|>[[231, 365, 763, 420]]<|/det|> +\[T_{\mathrm{ghost}}(x_{i},y_{i}) = T_{\mathrm{boundary}}(x_{i},y_{i}\pm d)\pm \frac{d}{k} (\sigma \epsilon \big(T_{\mathrm{boundary}}(x_{i},y_{i}\pm d)^{4} - T_{0}^{4}\big)\] \[\qquad +h_{a}(T_{\mathrm{boundary}}(x_{i},y_{i}\pm d) - T_{0})\big)\# (7)\] + +<|ref|>text<|/ref|><|det|>[[147, 418, 850, 529]]<|/det|> +Where \(\sigma\) is thermal emissivity, \(\epsilon\) is Stefan- boltzmann constant, and \(h_{a}\) is heat convection coefficient of substrate and air interaction. Although the relationship between the Ghost nodes and the internal nodes is fixed, their specific values change as the computation progresses because they depend on the values of the internal nodes. Purpose of boundary filling operations is to ensure the minimization of the residual of the PDE across the entire solution domain without compromising any boundary nodes. + +<|ref|>image<|/ref|><|det|>[[228, 534, 757, 725]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[389, 735, 606, 751]]<|/det|> +
Fig. 5. Hard-encoding of BCs.
+ +<|ref|>text<|/ref|><|det|>[[147, 770, 850, 900]]<|/det|> +In this study, physical information is not directly input into the model as raw features, but is instead embedded through an intermediate feature related to heat input. This intermediate feature integrates the effects of process parameters and is merged with other feature matrices. The higher- order parameter selected in this study is the arc heat source term, which aims to integrate arc heat source information into the historical temperature field to enhance the modeling effect. Additionally, to enable the model to learn the geometric characteristic of the structure, a characteristic matrix of the same size as the temperature field matrix is constructed, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 848, 120]]<|/det|> +with its values defined by the deposition state of the model. The geometric matrix and the PI matrix are defined in detail in Section 3.3. + +<|ref|>title<|/ref|><|det|>[[148, 142, 343, 157]]<|/det|> +#### 3.4.2. Loss function design + +<|ref|>text<|/ref|><|det|>[[147, 160, 848, 194]]<|/det|> +The loss function of a PIML model consists of two core elements: data- based loss and physics- based loss. The total loss of the model can be expressed as: + +<|ref|>equation<|/ref|><|det|>[[383, 198, 612, 214]]<|/det|> +\[L_{total} = w_{p}L_{pde} + w_{d}L_{data}\# (\Theta)\] + +<|ref|>text<|/ref|><|det|>[[147, 215, 848, 250]]<|/det|> +Where \(w_{p}\) and \(w_{d}\) are the weights of \(L_{pde}\) and \(L_{data}\) , and the default values are set to 0.3 and 0.7. + +<|ref|>text<|/ref|><|det|>[[147, 252, 850, 362]]<|/det|> +Due to the hard enforcement of I/BCs, the physical loss only needs to control the loss function of the partial differential equation (PDE). To compute the temperature gradients across time and space, the computational domain is discretized into \(m \times n\) nodes with a spatial distance of \(d\) . Inspired by the FDM, differential approximations are used to calculate the derivative terms of the heat conduction governing equations. The loss of the PDE can be represented by a difference equation, the difference form is: + +<|ref|>equation<|/ref|><|det|>[[300, 366, 697, 453]]<|/det|> +\[R_{t}(x_{i},y_{j}) = \rho C_{p}\frac{\left(\widehat{T}_{t + 1}(x_{i},y_{j}) - \widehat{T}_{t - 1}(x_{i},y_{j})\right)}{2\Delta t}\] \[-\frac{1}{d^{2}} k(\widehat{T}_{t}(x_{i} + d,y_{j}) + \widehat{T}_{t}(x_{i} - d,y_{j}) + \widehat{T}_{t}(x_{i},y_{j} + d)\] \[\qquad +\widehat{T}_{t}(x_{i},y_{i} - d) - 4\widehat{T}_{t}(x_{i},y_{i})) - \dot{Q}\# (9)\] + +<|ref|>text<|/ref|><|det|>[[147, 455, 850, 602]]<|/det|> +Where \(\widehat{T} (x_{i},y_{j})\) denotes the predicted temperature field at position \((x_{i},y_{j})\) at time \(t\) , and \(\widehat{T}_{t + 1}(x_{i},y_{j})\) and \(\widehat{T}_{t - 1}(x_{i},y_{j})\) represent the temperature at the same node at \(t + 1\) and \(t - 1\) , respectively, \(\rho\) is the density of the part material, \(C_{p}\) is the specific heat, \(\dot{Q}\) is the arc heat source term. Since the model's output is a 2D matrix, it is necessary to calculate the physical residuals of the predicted values at each position. Ideally, Model's predicted temperature should be such that \(R_{t}(x_{i},y_{j})\) is close to zero. Physical constraints can be imposed on the model by minimizing the PDE loss function \(L_{pde}\) , whose value corresponds to the squared mean of the residuals of the PDE over the spatio- temporal discretization, which can be expressed as: + +<|ref|>equation<|/ref|><|det|>[[340, 606, 653, 653]]<|/det|> +\[L_{pde} = \frac{1}{nmT}\sum_{i = 1}^{n}\sum_{i = 1}^{m}\sum_{t = 1}^{T}R_{t}(x_{i},y_{j})^{2}\# (10)\] + +<|ref|>text<|/ref|><|det|>[[147, 659, 850, 712]]<|/det|> +Where \(n\) and \(m\) represent the number of nodes in the spatial grid in the direction of height and width, respectively, and \(t\) is the total time for prediction. The data loss \(L_{data}\) can be expressed as: + +<|ref|>equation<|/ref|><|det|>[[290, 717, 704, 764]]<|/det|> +\[L_{data} = \frac{1}{nmT}\sum_{i = 1}^{n}\sum_{i = 1}^{m}\sum_{t = 1}^{T}\left[\widehat{T}_{t}(x_{i},y_{j}) - T_{t}(x_{i},y_{j})\right]^{2}\# (11)\] + +<|ref|>text<|/ref|><|det|>[[183, 770, 479, 787]]<|/det|> +Where \(T_{t}(x_{i},y_{j})\) is the true temperature. + +<|ref|>title<|/ref|><|det|>[[148, 808, 378, 823]]<|/det|> +#### 3.4.3. Training and test methods + +<|ref|>text<|/ref|><|det|>[[147, 826, 850, 899]]<|/det|> +Initially, the real- time long- horizon temperature field prediction capabilities of four models were evaluated using FEM simulation data: the Baseline model relying solely on ConvLSTM, the GeoRNN model incorporating geometric features, the PIRNN model integrating physical information, and the PIGeoRNN model that fuses both geometric features and physical + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 85, 852, 287]]<|/det|> +information. Subsequently, the performance of the proposed models was trained and tested on experiment data. Building on models trained with FEM simulation, further TL training was conducted using experiment data[52]. Experiment data provide more realistic and specific scenarios, helping the model adapt to the complexities of actual manufacturing processes and enhancing its predictive accuracy and adaptability. The dataset was split into training and testing sets in a 7:3 ratio. The learning rate was set to 0.01, the batch size to 8, and training was halted after 3000 epochs. The test set was evaluated every 500 epochs, recording the mean squared error (MSE), saving the trained model parameters, and the test results. For all models, the patch size was set to \(3 \times 3\) so that each \(51 \times 51\) characteristic matrix is converted \(17 \times 17 \times 9\) tensor. The same hyperparameters were considered for the all models proposed in this study, with the networks trained using the stochastic gradient descent Adam optimizer. + +<|ref|>text<|/ref|><|det|>[[147, 289, 852, 473]]<|/det|> +In order to assess the performance of the proposed model based on the prediction results, the use of MSE as an evaluation metric for the model can effectively assess the prediction performance of the model at each point in time. This metrics approach ensures a nuanced understanding of precision, error magnitude, and accuracy percentage, thus enhancing the overall assessment of the model. When evaluating the long- horizon temperature field prediction capabilities of different models, the average Root Mean Square Error (RMSE) is used to quantify the differences between predicted and true temperatures. The Kullback- Leibler (KL) divergence is used to evaluate the difference between the predicted and true probability distributions. The true and predicted temperatures are normalized to convert them into probability distributions, resulting in the predicted probability distribution \(Q_t\) and the true probability distribution \(P_t\) . + +<|ref|>equation<|/ref|><|det|>[[337, 475, 658, 580]]<|/det|> +\[Q_{t}(x_{i},y_{j}) = \frac{\widehat{T}_{t}(x_{i},y_{j})}{\sum_{i^{\prime} = 1}^{n}\sum_{j^{\prime} = 1}^{m}\widehat{T}_{t}(x_{i^{\prime}},y_{j^{\prime}})}\# (12)\] \[P_{t}(x_{i},y_{j}) = \frac{T_{t}(x_{i},y_{j})}{\sum_{i^{\prime} = 1}^{n}\sum_{j^{\prime} = 1}^{m}T_{t}(x_{i^{\prime}},y_{j^{\prime}})}\# (13)\] + +<|ref|>text<|/ref|><|det|>[[147, 585, 850, 658]]<|/det|> +Therefore, the spatial difference between the predicted temperature field and the true temperature field is defined as the KL divergence between \(Q_t\) and \(P_t\) . In this study, a smaller KL divergence indicates that the distribution of the true temperature field is closer to that of the predicted temperature field. + +<|ref|>equation<|/ref|><|det|>[[369, 663, 626, 709]]<|/det|> +\[\mathrm{D}_{\mathrm{KL}}(P_t||Q_t) = \sum_{t = 1}^{T}P_t\log \frac{P_t}{Q_t}\# (14)\] + +<|ref|>sub_title<|/ref|><|det|>[[149, 735, 413, 750]]<|/det|> +## 4. System setup and data collection + +<|ref|>sub_title<|/ref|><|det|>[[149, 754, 318, 768]]<|/det|> +### 4.1. Experiment system + +<|ref|>text<|/ref|><|det|>[[147, 771, 850, 900]]<|/det|> +The experiment system consists of an ABB IRB 2600 robot, table, Fronius CMT welder, CMT wire feeder, robot controller, computer, FLIR a655sc thermal imager, and protective gas tank. The welding torch is held by an ABB IRB2600 industrial robot and moves along a pre- programmed trajectory. A Fronius CMT Advanced 4000 welder was used as the welding power source. A FLIR a655sc thermal imager was used for temperature field in- situ monitoring and data collection tasks. A computer was used as the main controller to coordinate the industrial robot, welder and thermal imager to perform the deposition tasks. The welding technique was cold + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 87, 319, 100]]<|/det|> +metal transition (CMT). + +<|ref|>image<|/ref|><|det|>[[225, 105, 770, 351]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[338, 420, 657, 436]]<|/det|> +
Fig. 6. Experiment setup for WAAM system.
+ +<|ref|>text<|/ref|><|det|>[[185, 360, 800, 409]]<|/det|> +\(①\) ABB IRB 2600 industrial robot, \(②\) worktable, \(③\) Feonius CMT Advanced 4000 welding machine, \(④\) CMT wire feeder, \(⑤\) robot controller, \(⑥\) control computer, \(⑦\) FLIR A655sc IR camera, \(⑧\) shielding + +<|ref|>sub_title<|/ref|><|det|>[[148, 456, 335, 471]]<|/det|> +### 4.2. Practical experiments + +<|ref|>text<|/ref|><|det|>[[147, 475, 850, 546]]<|/det|> +In this study, experiments were carried out to deposit thin- wall structures by WAAM. A 1.2 mm welding wire made of ER70S- 6 was used for deposition. The substrate was selected as \(300 \times 300 \times 10 \mathrm{~mm}3\) mild steel plate made of Q235b. The main chemical compositions of the wire and the substrate are shown in Table 1. + +<|ref|>table<|/ref|><|det|>[[236, 565, 760, 630]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[280, 549, 714, 564]]<|/det|> +Table 1 Main chemical composition of ER70S-6 and Q235b. + +
ElementsCMnSiCuNiCr
ER70s-60.151.851.150.500.0250.50
Q235b0.21.40.350.30.30.3
+ +<|ref|>text<|/ref|><|det|>[[147, 648, 851, 867]]<|/det|> +The experiments aim to evaluate the accuracy and adaptability of the temperature field prediction model. A total of 10 thin- wall structures were designed, each 100 mm long, composed of 20 deposition layers, using different WFS and TS groups, simulated 5 and deposited 4 thin- walled structures for training and deposited 1 thin- walled structure for test. The thin- wall structures were deposited using a bidirectional path. Due to thermal accumulation, the average temperature of each layer increased. To eliminate the impact of thermal accumulation on part formation, the dwell time for each layer was controlled at 60 s, including the time for the welding gun to move on each layer. Detailed process parameters for the actual experiment are shown in Table 2. The arc current in the experiment was 109A, the arc voltage was 10.9V, and a mixed gas with a flow rate of 17 L/min was used, mainly composed of 50% argon and 50% carbon dioxide. The experiment used a FLIR A655s thermal camera for in- situ monitoring and data collection of the temperature field during the WAAM process. + +<|ref|>text<|/ref|><|det|>[[147, 869, 850, 904]]<|/det|> +The model was trained using the first 15 layers of data from five FEM simulations, with the last 5 layers reserved for testing. The same strategy was applied to five experiment datasets, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 86, 850, 230]]<|/det|> +the prediction results were evaluated using MSE to assess model performance. The geometric characteristic and physical information of the experiment data were designed to match those of the simulation data, its defined by real-time captured geometric characteristics and welding gun positions. Experiment data, after being collected and cropped by the thermal camera, were processed to the same size as the FEM simulation data. Building on the training with FEM simulation data, the model was further trained with actual experiment data using TL, and the performance of the model before and after TL training was compared; the results will be discussed in Section 5.2. + +<|ref|>table_caption<|/ref|><|det|>[[346, 237, 685, 250]]<|/det|> +Table 2 Process parameters of the experiments. + +<|ref|>table<|/ref|><|det|>[[245, 250, 750, 467]]<|/det|> +
WFS(m/min)TS(m/min)
Thin-wall training 140.48
Thin-wall training 250.48
Thin-wall training 360.48
Thin-wall training 470.48
Thin-wall training 580.48
Thin-wall training 64.90.5
Thin-wall training 74.90.45
Thin-wall training 85.50.45
Thin-wall training 960.4
Thin-wall test 160.45
+ +<|ref|>title<|/ref|><|det|>[[147, 489, 343, 502]]<|/det|> +# 5. Results and discussion + +<|ref|>sub_title<|/ref|><|det|>[[147, 507, 530, 520]]<|/det|> +## 5.1. Temperature field prediction based on FEM data + +<|ref|>sub_title<|/ref|><|det|>[[147, 526, 433, 539]]<|/det|> +### 5.1.1. Comparison of prediction results + +<|ref|>text<|/ref|><|det|>[[147, 544, 850, 910]]<|/det|> +For the given deposition geometry, the true temperature field from FEM simulation data is compared with the model's predicted results. As shown in Figure 7, the true temperature fields and the predicted results of the four models are compared for two sets of data, illustrating the temperature fields predicted for the future at 0.25 s and 1.25 s. The model-based prediction results demonstrate that geometric and PI features can serve as input datasets for WAAM temperature field prediction. Figure 8 presents a visual comparison of the prediction results and errors for the two sets of data across the four models. The Baseline, GeoRNN, PIRNN, and PIGeoRNN models predict future temperature fields with maximum errors of 8.8%-9.7%, 7.9%-9.5%, 7.6%-9.3%, and 4.6%-8.9%, respectively, showing significant differences in error magnitude. The Baseline model, relying solely on temperature field data and lacking understanding of geometric features and physical information, shows a clear deficiency in capturing the dynamic changes of the temperature field, especially in the melt pool area with large temperature gradients. This lack of information makes it difficult for the Baseline model to accurately predict the complex variations in the temperature field, leading to relatively larger errors in these critical areas. The GeoRNN model enhances the model's ability to learn real-time geometric changes by introducing a geometric feature matrix, dynamically adjusting the weights of temperature data in the deposition area, and correcting the geometric bias in the model's temperature predictions. The PIRNN model combines PIML with RNN, introducing PI inputs and a PI loss function to strictly adhere to physical laws. The GeoRNN and PIRNN models show similar performance in their predictions, with the maximum error reduced by approximately 1% compared to the Baseline model. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 85, 850, 120]]<|/det|> +PIGeoRNN model combines the advantages of the GeoRNN and PIRNN models, reducing the maximum error by about \(4\%\) and providing the most accurate predictions. + +<|ref|>text<|/ref|><|det|>[[147, 122, 851, 288]]<|/det|> +In terms of time efficiency, Simulation time of the FEM model is about 1 hour, while the training time of Baseline, GeoRNN, PIRNN and PIGeoRNN models are 5 min, 8 min, 9 min and 18 min, respectively, which indicates that the increase of model complexity requires more parameters to be optimized, which makes it need to spend more training time. The model training time in this study is short (compared to the automatic differentiation used in PINNs), primarily because the PDE is solved using the FDM. The main reason is that convolution operations with shift- invariant filters can update the entire field in each iteration, which is much more efficient than point- by- point training. In addition, the prediction time of the proposed model is about 12 ms, which can meet the real- time prediction demand. + +<|ref|>image<|/ref|><|det|>[[175, 292, 833, 731]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[161, 734, 834, 769]]<|/det|> +
Fig. 7. The future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models at two time points are compared with the true temperature field.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[210, 90, 790, 411]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[152, 419, 841, 472]]<|/det|> +
Fig. 8. Comparison of the errors of the future 0.25s and 1.25s temperature fields predicted by Baseline, GeoRNN, PIRNN, and PIGeoRNN models with the true temperature fields at two time points.
+ +<|ref|>text<|/ref|><|det|>[[147, 492, 852, 750]]<|/det|> +To evaluate the accuracy of the PIGeoRNN model's prediction for individual nodes at each time, the temperature history of the middle node on the tenth layer was extracted, as shown in Figure 9. As shown in Figure 9(a), the model's predicted values are highly consistent with the true values in terms of overall trend and can effectively capture the changes in temperature increases and decreases. In regions where the temperature fluctuates gently, the predicted values are extremely close to the true values. However, during periods of rapid temperature change, there is a noticeable deviation between the two. Nevertheless, the model is still able to generally follow the direction of the true value changes. As shown in Figure 9(b), the errors across all time are controlled within the range of \(- 100^{\circ}\mathrm{C}\) to \(100^{\circ}\mathrm{C}\) , with the majority of the time falling within \(- 25^{\circ}\mathrm{C}\) to \(25^{\circ}\mathrm{C}\) . This indicates that the model's prediction errors are within an acceptable range and that the overall reliability of the model is high. However, the error at time 5 shows significant fluctuation and deviation. This may be due to the increased prediction time, which requires the model to process more historical information and complex spatiotemporal dependencies, leading to the accumulation and amplification of errors, thereby affecting the prediction accuracy. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[167, 92, 840, 280]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[156, 289, 841, 324]]<|/det|> +
Fig. 9. (a) The temperature history and (b) Error of the middle node on the tenth layer, predicted by the PIGeoRNN model over five time.
+ +<|ref|>title<|/ref|><|det|>[[150, 346, 345, 361]]<|/det|> +#### 5.1.2. Ablation experiment + +<|ref|>text<|/ref|><|det|>[[148, 364, 851, 768]]<|/det|> +The training loss history of the four models on FEM simulation data and the MSE of all prediction results at each time are compared in Figure 10. Overall, the loss decreases with the decrease in epochs, indicating that the models can learn the dataset well. It can be observed from Figure 10(a) that all losses fluctuate greatly in the first 1000 epochs because the models have not yet learned the complex patterns in the data, especially for high- dimensional data. After 1000 epochs, the loss fluctuates around 5e- 5. The PIGeoRNN model shows a more gradual decrease in training loss, which is related to the complexity of the model. The PDE loss history during the training of the PIGeoRNN model tends to fluctuate within a small range after about 500 epochs, indicating that the model can learn physical information within a short number of epochs. Figure 10(b) shows that the MSE of all prediction results at each time exhibits a trend of increasing error with time, indicating that errors are accumulated over time. Long- horizon prediction may lead to further accumulation of errors, but experiments have shown that predicting the temperature field for the next 1.25 s results in lower errors. The Baseline model performs poorly in temperature field prediction at each time point, mainly because the model only learns the spatiotemporal characteristic of the temperature field data. The GeoRNN model performs better than the Baseline in temperature field prediction at each time point, as the model can learn real- time geometric characteristic to enhance robustness against environmental noise in the data. Due to physical constraints, PIRNN models are more accurate than GeoRNN models in long time predictions. The PIGeoRNN model is proven to have the best performance in prediction, with a more gradual increase in error at each time point. This is related to the model's integration of geometric characteristic inputs, PI inputs, and PI loss functions, which allows the model to learn more process knowledge and be subject to physical constraints. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 92, 848, 283]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[156, 289, 840, 324]]<|/det|> +
Fig. 10. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models for loss and (b) MSE per time in prediction in FEM data training.
+ +<|ref|>title<|/ref|><|det|>[[150, 345, 459, 360]]<|/det|> +#### 5.1.3. Long-horizon predictive evaluation + +<|ref|>text<|/ref|><|det|>[[147, 363, 851, 658]]<|/det|> +To further explore the models' performance in long- horizon prediction, this study evaluated the prediction results of the four models over 1.25s, 3.75s, 6.25s, 8.75, and 11.25 s (time resolution of 0.25 s) into the future, as shown in Figure 11. As shown in Figure 11(a), the Baseline model has the largest average RMSE in long- horizon predictions. The GeoRNN and PIRNN models show similar performance in predictions up to 6.25 s, with the PIRNN model having a smaller average RMSE beyond 6.25 s. For predictions within 6.25 s, the average RMSE of the PIGeoRNN model increases with time. However, for predictions beyond 6.25 s, the average RMSE at 8.75 s and 11.25 s is lower than that at 6.25 s, indicating that the PIGeoRNN model can perform well in long- horizon predictions. Additionally, the KL divergence was used to evaluate the temperature field distributions of the four models in long- horizon predictions, as shown in Figure 11(b). It can be observed that the Baseline and PIRNN models exhibited relatively larger KL divergence in long- horizon predictions, primarily due to the lack of long- horizon geometric feature inputs. Consequently, the GeoRNN model, which incorporates geometric features, showed a relatively smaller KL divergence. After integrating geometric features and physical information, the PIGeoRNN model was able to learn the geometric features and melt pool distribution in long- horizon predictions, resulting in the lowest KL divergence. + +<|ref|>image<|/ref|><|det|>[[150, 661, 845, 859]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 863, 832, 899]]<|/det|> +
Fig. 11. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on FEM data.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 104, 572, 120]]<|/det|> +### 5.2. Temperature field prediction based on experiment data + +<|ref|>title<|/ref|><|det|>[[149, 123, 432, 138]]<|/det|> +#### 5.2.1. Comparison of prediction results + +<|ref|>text<|/ref|><|det|>[[147, 141, 852, 455]]<|/det|> +In order to test the adaptability of the real- time temperature field prediction model, a thin- walled structure was deposited using different process parameters, as shown in Figure 12. Figure 13 presents a visual comparison of the prediction results and errors for the experiment thin- wall structure data across the three models. The true temperature field reveal differences in temperature gradients in the melt pool area between experiment and simulation data, which is related to the emissivity of the thermal camera. The melt pool appears liquid during deposition, whereas the setting of 0.8 corresponds to the solid- state emissivity of the material. Consequently, the issue with the thermal camera's emissivity setting leads to larger errors in the melt pool area of the temperature field and may also result in the physical information not being well learned. The Baseline, GeoRNN, PIRNN, and PIGeoRNN models predicted the temperature field for the next 1.25 s with maximum errors of \(14.2\% - 16.5\%\) , \(13.3\% - 15.2\%\) , \(13\% - 15.2\%\) , and \(13\% - 14.2\%\) , respectively, with the PIGeoRNN still demonstrating superior performance. During the WAAM deposition process, defects such as spatter and lack of fusion can affect the distribution of the temperature field, leading to uneven temperature distribution, which in turn impacts the model's learning of geometric characteristic and physical information. Due to differences between the layer height of each deposition in the thin- wall structures and the ideal case, the geometric characteristic learned by the model may deviate from reality. + +<|ref|>image<|/ref|><|det|>[[377, 463, 617, 559]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[375, 567, 621, 583]]<|/det|> +
Fig. 12. Tested thin-wall structure.
+ +<|ref|>image<|/ref|><|det|>[[198, 610, 792, 910]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 85, 853, 138]]<|/det|> +Fig. 13. (a) Comparison of True and Predicted Temperature Fields for Baseline, GeoRNN, PIRNN, and PIGeoRNN Models at Future Times 0.25s and 1.25s; (b) Analysis of Predicted Temperature Field Error. + +<|ref|>title<|/ref|><|det|>[[148, 160, 346, 175]]<|/det|> +#### 5.2.2. Ablation experiment + +<|ref|>text<|/ref|><|det|>[[147, 178, 852, 342]]<|/det|> +The training loss history comparison of the four models on experiment data and the MSE of all prediction results at each time are shown in Figure 14. Overall, the loss decreases with the decrease in epochs, indicating that the models can learn the experiment data well. It can be observed from Figure 14(a) that all losses also fluctuate significantly over the first 1000 epochs. The PIGeoRNN model shows a more gradual decrease in training loss and still requires more time for training. Figure 14(b) shows that the MSE of all prediction results at each time exhibits a trend of increasing error with time, indicating that errors are accumulated over time. The PIGeoRNN model does not show a significant improvement in prediction accuracy compared to the GeoRNN and the PIRNN model. + +<|ref|>image<|/ref|><|det|>[[152, 351, 847, 525]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[174, 532, 822, 566]]<|/det|> +
Fig. 14. (a) Comparison of Baseline, GeoRNN, PIRNN, and PIGeoRNN models trained on experiment data in terms of loss and (b) MSE at each time of test.
+ +<|ref|>title<|/ref|><|det|>[[148, 587, 459, 602]]<|/det|> +#### 5.2.3. Long-horizon predictive evaluation + +<|ref|>text<|/ref|><|det|>[[147, 605, 851, 787]]<|/det|> +The long- horizon prediction performance of the four models on experimental data was further evaluated, as shown in Figure 15. As shown in Figure 15(a), The PIGeoRNN model has the lowest average RMSE across the entire time range, demonstrating the highest prediction accuracy and stability. The GeoRNN and PIRNN models perform slightly worse, while the Baseline model has the highest average RMSE and the poorest prediction accuracy. This indicates that the PIGeoRNN model has a significant advantage in handling long- horizon temperature prediction field tasks. Furthermore, Figure 15(b) shows the evaluation of the prediction results using KL divergence. It can be observed that after incorporating geometric features, the GeoRNN and PIGeoRNN models have relatively close KL divergence in long- horizon predictions, indicating better spatial distribution. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[156, 92, 792, 282]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 289, 832, 324]]<|/det|> +
Fig. 15. (a) The average RMSE and (b) KL divergence were used to evaluate the long-horizon predictions of Baseline, GeoRNN, PIRNN, and PIGeoRNN models on experiment data.
+ +<|ref|>title<|/ref|><|det|>[[150, 345, 389, 360]]<|/det|> +#### 5.2.4. Transfer learning training + +<|ref|>text<|/ref|><|det|>[[148, 363, 851, 658]]<|/det|> +By comparing the performance of the four models on FEM simulation data and experiment data, it can be demonstrated that the proposed PIGeoRNN model exhibits the best performance. To leverage existing knowledge to accelerate learning from experiment data, aiding the model in learning the fundamental characteristics and physical laws of the temperature field, thereby enhancing learning efficiency and performance. Building on the model trained with FEM simulation data, further TL training is conducted on experiment data using actual data. Figure 16 compares the true temperature field and the predicted temperature field with and without TL, along with the visualization of errors. The experiments indicate that the model's predictions exhibit a temperature distribution that is relatively consistent with the true temperature field. The temperature field distributions in the melt pool area and near the deposition boundaries are clearly displayed. The models without TL (No TL) and with TL predict future temperature field with maximum errors of \(13 - 14.2\%\) and \(11.2 - 14\%\) , respectively. The loss history of models with and without TL training is shown in Figure 17, proving that initializing with model parameters trained on simulation data leads to much faster loss reduction. The total loss for models with TL training can be reduced to approximately \(1\mathrm{e}^{- 5}\) in just 500 epochs, whereas models without TL require 1000 epochs to achieve the same MSE, reducing training time by \(50\%\) . + +<|ref|>image<|/ref|><|det|>[[174, 677, 825, 878]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[160, 883, 835, 899]]<|/det|> +
Fig. 16. (a) Comparison of True and Predicted Temperature Fields for TL and No TL Models at
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[155, 125, 842, 303]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 308, 832, 343]]<|/det|> +
Fig. 17. (a) Comparison of model training loss history TL or No TL and (b) MSE of predicted results at each time of test.
+ +<|ref|>sub_title<|/ref|><|det|>[[148, 364, 273, 378]]<|/det|> +### 5.3. Discussion + +<|ref|>text<|/ref|><|det|>[[148, 382, 851, 528]]<|/det|> +This study aims to propose a real- time long- horizon temperature field predicted method for the WAAM process using interpretable, adaptive, and robust data- driven approaches. The main innovation of the model lies in its PIML architecture that integrates geometric features and physical information inputs, enabling long- horizon temperature field prediction. Compared with models that integrate information at different levels, this model exhibits greater interpretability and robustness when trained and used to predicted temperature field at future times using the same dataset, as demonstrated by the comparison of the model's training and prediction results on different datasets in Sections 5.1 and 5.2. + +<|ref|>text<|/ref|><|det|>[[148, 530, 851, 806]]<|/det|> +The varying performance among different models highlights the importance of integrating geometric characteristic and physical information, which significantly impacts the model's performance. The temperature field simulation results based on FEM and experiment data indicate that the information described by inputting temperature field data alone is limited; the models primarily rely on learning patterns from historical data. However, in the WAAM process, real- time geometric characteristic affects the path and efficiency of heat conduction. Thicker deposited layers require more time to cool because heat must pass through thicker material layers before it can dissipate into the environment. Failing to adequately consider the changes in geometric characteristic during the WAAM process leads to poor performance of the Baseline model in temperature field predicted. Therefore, the GeoRNN model is constructed to utilize a real- time geometric characteristic matrix as an additional input to extract geometric characteristic information from the AM process. It dynamically adjusts the model's weights for temperature data in the deposition area, enabling the model to learn and correct predictive biases. At the cost of slightly reduced temporal efficiency, it improves the accuracy and adaptability of predictions over the Baseline model. + +<|ref|>text<|/ref|><|det|>[[148, 809, 850, 899]]<|/det|> +During the WAAM process, changes in the temperature field are influenced by various physical mechanisms, such as heat conduction, convection, and radiation. If the model fails to adequately understand these physical mechanisms, it may lead to discrepancies between the predicted results and the actual physical processes. The PIGeoRNN model proposed in this study integrates PI inputs based on the combination of GeoRNN and PIML, enabling the model to learn + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 85, 850, 157]]<|/det|> +melt pool (heat source) characteristics and adjust input weights, thus accelerating gradient descent. This model is constrained by I/BCs and a PI loss function, compelling the model to minimize the physical discrepancies between predictions and true values. This endows the model with interpretability and robustness, and it has been proven to possess the best performance. + +<|ref|>text<|/ref|><|det|>[[148, 160, 851, 343]]<|/det|> +This study utilizes a model trained on FEM simulation data for TL on experiment data, which can effectively improve computational efficiency and accelerate model convergence. The model parameters obtained from simulation data provide an excellent starting point for the training of experiment data, thereby reducing the number of iterations and time required for training on experiment data. Since simulation data typically cover a wide range of process conditions and geometric characteristic, the model has already learned the fundamental patterns under these conditions without needing to learn all characteristic and patterns from scratch. Moreover, models trained on simulation data can capture complex physical phenomena in the WAAM process, such as the effects of heat accumulation and the trends in thermal cycling curves; this information can be directly utilized in the TL process, further improving the efficiency of model training. + +<|ref|>text<|/ref|><|det|>[[148, 345, 851, 546]]<|/det|> +The PIGeoRNN model proposed in this study holds potential application value in feed- forward control and digital twin for the WAAM process. In feedforward control, the temperature field results predicted by the model can be directly used to guide the adjustment of process parameters (such as wire feeding speed and heat source power), to preemptively optimize deposition paths or cooling strategies, thereby actively avoiding the generation of thermal deformation and residual stress during the deposition process, enhancing the quality and performance of the formed parts. In the construction of digital twin, the model can serve as an integration interface between the physical engine and real- time data, enabling dynamic prediction of the process by establishing multi- physics field mappings of geometry- temperature- stress relationships. However, these potential applications require further experiment validation to assess their performance in actual operating conditions. + +<|ref|>sub_title<|/ref|><|det|>[[149, 567, 380, 581]]<|/det|> +## 6. Conclusion and future work + +<|ref|>text<|/ref|><|det|>[[148, 585, 851, 899]]<|/det|> +The PIGeoRNN model proposed in this paper demonstrates the feasibility and effectiveness of real- time long- horizon temperature field predicted in the WAAM process, with considerable accuracy. By integrating PIML and TL strategies, this model significantly enhances prediction time- efficiency, interpretability, and robustness. The real- time long- horizon temperature field prediction model is based on RNN architecture and achieves long- horizon temperature field prediction through the fusion of multimodal spatiotemporal characteristic. The model includes an encoder- decoder module, dual- layer ConvLSTM cells, and a physical constraint module, integrating geometric characteristic and physical information to predict the temperature field for the next 1.25s based on the current 1.25s, with a prediction time of only 12 ms. The model was trained and tested based on FEM simulation data, and the predictive performance of four different hierarchical models was compared. Compared with FEM models, the PIGeoRNN model improves predictive accuracy and interpretability by integrating PI and geometric characteristic inputs while maintaining low computational costs. In addition, the experimental evaluation assessed the model's predictive performance on experimental data by additionally fabricating a thin- walled structure for test. To improve the model's learning efficiency and performance, a TL strategy was employed for the training and testing of the experiment data temperature field, and the results demonstrated the model's practicality and effectiveness in real applications, meeting the needs for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 86, 759, 101]]<|/det|> +real- time long- horizon prediction. The main contributions of this study are as follows: + +<|ref|>text<|/ref|><|det|>[[181, 105, 852, 435]]<|/det|> +1. The PIGeoRNN model, which integrates geometric characteristic and PI inputs, a PI loss function, and ConvLSTM cells, is used for real-time long-horizon temperature field prediction in the WAAM process and holds great promise for applications requiring feed-forward control and digital twin. +2. The ConvLSTM cell can simultaneously extract spatial temperature distributions and temporal change trends, which allows for more accurate prediction and understanding of the dynamic behavior of the temperature field, thus enabling long-horizon temperature field prediction. +3. The model employs a strategy that can solve spatiotemporal PDE, imposing initial and boundary conditions as hard constraints on the model, which enhances the model's predictive accuracy and robustness. +4. Compared with Baseline, GeoRNN and PIRNN model, the PIGeoRNN model demonstrates the best performance in both simulation and experiment data temperature field prediction achieving a maximum error of only \(4.5\%\) at the cost of slightly reduced temporal efficiency. +5. Models trained using simulation data are trained in TL on experiment data, which improves the utility and efficiency of predictions and reduces the training time of the models by \(50\%\) . + +<|ref|>text<|/ref|><|det|>[[147, 438, 851, 547]]<|/det|> +Future work can further develop the model to consider more geometrical characteristic and process parameters and optimize the AM process based on accurate and fast temperature field prediction results. Also, the construction of long- horizon predictive models for stress fields and deformations can be considered to provide a feasible solution for depositing high- quality parts. Although the prediction model has been validated only for WAAM, the model can be applied to any thermal imager- based AM process. + +<|ref|>sub_title<|/ref|><|det|>[[149, 568, 290, 583]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[149, 595, 851, 631]]<|/det|> +This work is financially supported in part by the National Natural Science Foundation of China (NSFC) under Grant 52375344. + +<|ref|>sub_title<|/ref|><|det|>[[148, 654, 234, 668]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[145, 672, 852, 910]]<|/det|> +[1] MU H, HE F, YUAN L, COMMINS P, WANG H, PAN Z. Toward a smart wire arc additive manufacturing system: A review on current developments and a framework of digital twin [J]. Journal of Manufacturing Systems, 2023, 67: 174- 89. [2] XIONG Y, TANG Y, ZHOU Q, MA Y, ROSEN D W. Intelligent additive manufacturing and design: state of the art and future perspectives [J]. Additive Manufacturing, 2022, 59. [3] NG W L, GOH G L, GOH G D, TEN J S J, YEONG W Y. Progress and Opportunities for Machine Learning in Materials and Processes of Additive Manufacturing [J]. Advanced Materials, 2024, 36(34). [4] FRAZIER W E. Metal Additive Manufacturing: A Review [J]. Journal of Materials Engineering and Performance, 2014, 23(6): 1917- 28. [5] BLAKEY- MILNER B, GRADL P, SNEDDEN G, BROOKS M, PITOT J, LOPEZ E, LEARY M, BERTO F, DU PLESSIS A. Metal additive manufacturing in aerospace: A review [J]. Materials & Design, 2021, 209. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 87, 852, 900]]<|/det|> +[6] ZHANG H, LI R, LIU J, WANG K, WEIJIAN Q, SHI L, LEI L, HE W, WU S. State-of-art review on the process-structure-properties-performance linkage in wire arc additive manufacturing [J]. Virtual and Physical Prototyping, 2024, 19(1).[7] CHANDRA S, RADHAKRISHNAN J, HUANG S, WEI S, RAMAMURTY U. Solidification in metal additive manufacturing: challenges, solutions, and opportunities [J]. Progress in Materials Science, 2025, 148. [8] WU B, PAN Z, DING D, CUIURI D, LI H, XU J, NORRISH J. A review of the wire arc additive manufacturing of metals: properties, defects and quality improvement [J]. Journal of Manufacturing Processes, 2018, 35: 127-39. [9] ZHANG J, LIOU F, SEUFZER W, TAMINGER K. A coupled finite element cellular automaton model to predict thermal history and grain morphology of Ti- 6Al- 4V during direct metal deposition (DMD) [J]. Additive Manufacturing, 2016, 11: 32-9. [10] FABBRI M, ASCHWANDEN I, WEGENER K, BAMBACH M. A finite volume scheme with an adaptive heat flux calculation for fast temperature prediction for wire arc additive manufacturing [J]. Journal of Materials Processing Technology, 2024, 324. [11] HU Z, YAN W. Data-driven modeling of process-structure-property relationships in metal additive manufacturing [J]. npj Advanced Manufacturing, 2024, 1(1).[12] TANG Y, RAHMANI DEHAGHANI M, SAJADI P, BALANI S B, DHALPE A, PANICKER S, WU D, COATANEA E, WANG G G. Online thermal field prediction for metal additive manufacturing of thin walls [J]. Journal of Manufacturing Processes, 2023, 108: 529-50. [13] PHAM T Q D, HOANG T V, VAN TRAN X, PHAM Q T, FETNI S, DUCHÈNE L, TRAN H S, HABRAKEN A- M. Fast and accurate prediction of temperature evolutions in additive manufacturing process using deep learning [J]. Journal of Intelligent Manufacturing, 2022, 34(4): 1701-19. [14] REN K, CHEW Y, ZHANG Y F, FUH J Y H, BI G J. Thermal field prediction for laser scanning paths in laser aided additive manufacturing by physics- based machine learning [J]. Computer Methods in Applied Mechanics and Engineering, 2020, 362. [15] HAMRANI A, AGARWAL A, ALLOUHI A, MCDANIEL D. Applying machine learning to wire arc additive manufacturing: a systematic data- driven literature review [J]. Journal of Intelligent Manufacturing, 2023, 35(6): 2407-39. [16] KARNIADAKIS G E, KEVREKIDIS I G, LU L, PERDIKARIS P, WANG S, YANG L. Physics- informed machine learning [J]. Nature Reviews Physics, 2021, 3(6): 422-40. [17] DE LA BATUT B, FERGANI O, BROTAN V, BAMBACH M, EL MANSOURI M. Analytical and Numerical Temperature Prediction in Direct Metal Deposition of Ti6Al4V [J]. Journal of Manufacturing and Materials Processing, 2017, 1(1).[18] WEISZ- PATRAULT D. Fast simulation of temperature and phase transitions in directed energy deposition additive manufacturing [J]. Additive Manufacturing, 2020, 31. [19] ANSARI M, KHAMOOSHI M, HUANG Y, TOYSERKANI E. Analytical solutions for rapid prediction of transient temperature field in powder- fed laser directed energy deposition based on different heat source models [J]. Applied Physics A, 2021, 127(6).[20] ANSARI M, MARTINEZ- MARCHESE A, KHAMOOSHI M, KESHAVARZKERMANI A, ESMAEILIZADEH R, TOYSERKANI E. Analytical modeling of multi- track powder- fed laser directed energy deposition: On the relationships among process, deposition dimensions, and solidification microstructure in additively manufactured near- \(\beta\) titanium alloy [J]. Journal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[205, 85, 530, 101]]<|/det|> +of Materials Processing Technology, 2022, 306. + +<|ref|>text<|/ref|><|det|>[[145, 103, 850, 157]]<|/det|> +[21] REN K, CHEW Y, FUH J Y H, ZHANG Y F, BI G J. Thermo-mechanical analyses for optimized path planning in laser aided additive manufacturing processes [J]. Materials & Design, 2019, 162: 80- 93. + +<|ref|>text<|/ref|><|det|>[[145, 160, 850, 213]]<|/det|> +[22] CADIOU S, COURTOIS M, CARIN M, BERCKMANS W, LE MASSON P. 3D heat transfer, fluid flow and electromagnetic model for cold metal transfer wire arc additive manufacturing (Cmt- Waam) [J]. Additive Manufacturing, 2020, 36. + +<|ref|>text<|/ref|><|det|>[[145, 216, 850, 269]]<|/det|> +[23] LU X, LI M V, YANG H. Simulation of precipitates evolution driven by non- isothermal cyclic thermal history during wire and arc additive manufacturing of IN718 superalloy [J]. Journal of Manufacturing Processes, 2021, 65: 258- 70. + +<|ref|>text<|/ref|><|det|>[[145, 272, 850, 325]]<|/det|> +[24] ZHANG T, LI H, GONG H, WU Y, CHEN X, ZHANG X. Study on location- related thermal cycles and microstructure variation of additively manufactured inconel 718 [J]. Journal of Materials Research and Technology, 2022, 18: 3056- 72. + +<|ref|>text<|/ref|><|det|>[[145, 328, 850, 381]]<|/det|> +[25] CHEN F, YANG M, YAN W. Data- driven prognostic model for temperature field in additive manufacturing based on the high- fidelity thermal- fluid flow simulation [J]. Computer Methods in Applied Mechanics and Engineering, 2022, 392. + +<|ref|>text<|/ref|><|det|>[[145, 384, 850, 437]]<|/det|> +[26] ANANDAN KUMAR H, KUMARAGURU S, PAUL C P, BINDRA K S. Faster temperature prediction in the powder bed fusion process through the development of a surrogate model [J]. Optics & Laser Technology, 2021, 141. + +<|ref|>text<|/ref|><|det|>[[145, 440, 850, 474]]<|/det|> +[27] LI J, JIN R, YU H Z. Integration of physically- based and data- driven approaches for thermal field prediction in additive manufacturing [J]. Materials & Design, 2018, 139: 473- 85. + +<|ref|>text<|/ref|><|det|>[[145, 476, 850, 529]]<|/det|> +[28] XU S, RAI R, MOORE R D, ORLANDI G, ABDELJAWAD F. MPTP- Net: melt pool temperature profile network for thermal field modeling in beam shaping of laser powder bed fusion [J]. Journal of Intelligent Manufacturing, 2024. + +<|ref|>text<|/ref|><|det|>[[145, 532, 850, 585]]<|/det|> +[29] GUO S, GUO W, BIAN L, GUO Y B. A Deep- Learning- Based Surrogate Model for Thermal Signature Prediction in Laser Metal Deposition [J]. IEEE Transactions on Automation Science and Engineering, 2023, 20(1): 482- 94. + +<|ref|>text<|/ref|><|det|>[[145, 588, 850, 640]]<|/det|> +[30] NESS K L, PAUL A, SUN L, ZHANG Z. Towards a generic physics- based machine learning model for geometry invariant thermal history prediction in additive manufacturing [J]. Journal of Materials Processing Technology, 2022, 302. + +<|ref|>text<|/ref|><|det|>[[145, 643, 850, 714]]<|/det|> +[31] KOZJEK D, CARTER F M, PORTER C, MOGONYE J- E, EHMANN K, CAO J. Data- driven prediction of next- layer melt pool temperatures in laser powder bed fusion based on co- axial high- resolution Planck thermometry measurements [J]. Journal of Manufacturing Processes, 2022, 79: 81- 90. + +<|ref|>text<|/ref|><|det|>[[145, 717, 850, 770]]<|/det|> +[32] MOZAFFAR M, LIAO S, LIN H, EHMANN K, CAO J. Geometry- agnostic data- driven thermal modeling of additive manufacturing processes using graph neural networks [J]. Additive Manufacturing, 2021, 48. + +<|ref|>text<|/ref|><|det|>[[145, 773, 850, 826]]<|/det|> +[33] CHEN J, KHRENOV M, JIN J, NARRA S P, MCCOMB C. Data- driven inpainting for full- part temperature monitoring in additive manufacturing [J]. Journal of Manufacturing Systems, 2024, 77: 558- 75. + +<|ref|>text<|/ref|><|det|>[[145, 829, 850, 881]]<|/det|> +[34] PERUMAL V, ABUEIDDA D, KORIC S, KONTSOS A. Temporal convolutional networks for data- driven thermal modeling of directed energy deposition [J]. Journal of Manufacturing Processes, 2023, 85: 405- 16. + +<|ref|>text<|/ref|><|det|>[[145, 884, 850, 900]]<|/det|> +[35] ZHANG Z, LIU Z, WU D. Prediction of melt pool temperature in directed energy deposition + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[156, 85, 852, 900]]<|/det|> +using machine learning [J]. Additive Manufacturing, 2021, 37. [36] ZHOU Z, SHEN H, LIU B, DU W, JIN J. Thermal field prediction for welding paths in multi-layer gas metal arc welding-based additive manufacturing: A machine learning approach [J]. Journal of Manufacturing Processes, 2021, 64: 960- 71. [37] CAO X, DUAN C, LUO X, ZHENG S, XU H, HAO X, ZHANG Z. Deep learning- based rapid prediction of temperature field and intelligent control of molten pool during directed energy deposition process [J]. Additive Manufacturing, 2024, 94. [38] WU S- H, TARIQ U, JOY R, MAHMOOD M A, MALIK A W, LIOU F. A Robust Recurrent Neural Networks- Based Surrogate Model for Thermal History and Melt Pool Characteristics in Directed Energy Deposition [J]. Materials, 2024, 17(17).[39] HU K- X, GUO K, LI W- D, WANG Y- H. Temperature evolution prediction for laser directed energy deposition enabled by finite element modelling and bi- directional gated recurrent unit [J]. Advances in Manufacturing, 2024. [40] KARKARIA V, GOECKNER A, ZHA R, CHEN J, ZHANG J, ZHU Q, CAO J, GAO R X, CHEN W. Towards a digital twin framework in additive manufacturing: Machine learning and bayesian optimization for time series process optimization [J]. Journal of Manufacturing Systems, 2024, 75: 322- 32. [41] SHENGHAN G, MOHIT A, CLAYTON C, QI T, X. G R, WEIHONG G, Y.B. G. Machine learning for metal additive manufacturing: Towards a physics- informed data- driven paradigm [J]. Journal of Manufacturing Systems, 2022, 62: 145- 63. [42] HOSSEINI E, SCHEEL P, MULLER O, MOLINARO R, MISHRA S. Single- track thermal analysis of laser powder bed fusion process: Parametric solution through physics- informed neural networks [J]. Computer Methods in Applied Mechanics and Engineering, 2023, 410. [43] ZHU Q, LU Z, HU Y. A reality- augmented adaptive physics informed machine learning method for efficient heat transfer prediction in laser melting [J]. Journal of Manufacturing Processes, 2024, 124: 444- 57. [44] TANG J, SCHEEL P, MOHEBBI M S, LEINENBACH C, DE LORENZIS L, HOSSEINI E. On the calibration of thermo- microstructural simulation models for Laser Powder Bed Fusion process: Integrating physics- informed neural networks with cellular automata [J]. Additive Manufacturing, 2024, 96. [45] PENG B, PANESAR A. Multi- layer thermal simulation using physics- informed neural network [J]. Additive Manufacturing, 2024, 95. [46] ZHU Q, LIU Z, YAN J. Machine learning for metal additive manufacturing: predicting temperature and melt pool fluid dynamics using physics- informed neural networks [J]. Computational Mechanics, 2021, 67(2): 619- 35. [47] XIE J, CHAI Z, XU L, REN X, LIU S, CHEN X. 3D temperature field prediction in direct energy deposition of metals using physics informed neural network [J]. The International Journal of Advanced Manufacturing Technology, 2022, 119(5- 6): 3449- 68. [48] SHUHENG L, TIANJU X, JIHOON J, SAMANTHA W, KORNEL E, JIAN C. Hybrid thermal modeling of additive manufacturing processes using physics- informed neural networks for temperature prediction and parameter identification [J]. Computational Mechanics, 2023, 72(3): 499- 512. [49] SHILIN L, GANG W, YUELAN D, LIPING W, HAIDOU W, QINGJUN Z. A physics- informed neural network framework to predict 3D temperature field without labeled + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[156, 85, 850, 250]]<|/det|> +data in process of laser metal deposition [J]. Engineering Applications of Artificial Intelligence, 2023, 120. [50] GOLDALK J, CHAKRAVARTI A, BIBBY M. A new finite element model for welding heat sources [J]. Metallurgical Transactions B, 1984, 15(2): 299- 305. [51] SHI X, CHEN Z, WANG H, YEUNG D- Y, WONG W- K, WOO W- C. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting [J]. Computer Science, 2015. [52] TANG Y, RAHMANI DEHAGHANI M, WANG G G. Review of transfer learning in modeling additive manufacturing processes [J]. Additive Manufacturing, 2023, 61. + +<--- Page Split ---> diff --git a/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/images_list.json b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..32435e13d11e78be9c72b5c7d4c90ea289d82f75 --- /dev/null +++ b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. SMiTH: Sampling Migration Trees with Homomorphisms. A: Input phylogenetic tree. B: Distribution of possible migration trees. Parallel rectangles depict a probability density function, with each rectangle's width proportional to the corresponding probability. In practice, the distribution is represented either by a random graph model or by a stochastic graph generation procedure. C: Candidate migration trees sampled from the distribution B. D: Homomorphisms from the phylogeny to three sampled trees. In the phylogenies, nodes are color-coded by their homomorphic images in a migration tree. The phylogeny layouts in the middle of each subfigure showcases how homomorphism transforms them into sampled trees. E: consensus solution derived from homomorphisms in D. The solution is shown as potential color distributions for the phylogeny's nodes (left) or as a graph where possible migration edges are weighted according to the number of supporting solutions (right), with the edge thickness indicating weight.", + "footnote": [], + "bbox": [ + [ + 84, + 97, + 910, + 520 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Solutions for the Migration History Inference Problem under various constraints. For each scenario, a phylogenetic tree is illustrated in two different layouts on the left and in the middle, with the corresponding migration tree displayed on the right. In the phylogenies, nodes are color-coded by their homomorphic images in the migration tree. The layout in the middle showcases how homomorphism transforms a phylogeny into a migration tree. (a) Unconstrained solution. Subtrees formed by blue and red nodes are not connected, indicating a violation of the convexity constraint. (b) Convex solution. Each color-coded subtree is connected, but compactness is violated in the subtree rooted at node 7. (c) Convex and compact solution.", + "footnote": [], + "bbox": [ + [ + 72, + 95, + 468, + 426 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Overview of the Dynamic Programming Algorithm for Detecting Convex Label-Distinctive Homomorphisms. A. A phylogenetic subtree, \\(\\Psi_{\\alpha}\\) , rooted at node \\(\\alpha\\) with two children, \\(\\beta\\) and \\(\\gamma\\) . B. Input candidate migration tree \\(T\\) . The goal is to produce convex homomorphisms from \\(\\Psi_{\\alpha}\\) to induced subtrees of \\(T\\) from such homomorphisms for \\(\\Psi_{\\beta}\\) and \\(\\Psi_{\\gamma}\\) . C. Convex homomorphisms from \\(\\Psi_{\\beta}\\) (top row) and \\(\\Psi_{\\gamma}\\) (bottom row) to induced subtrees of \\(T\\) . For instance, the top figure depicts a homomorphism to an induced 1-tree \\(T[1, \\{4, 9, 10, 11\\}]\\) that consists of the vertex 1, its neighbors 4, 9, 10, 11 and all vertices connected to 1 via paths that intersect these neighbors. Nodes of subtrees are colored by their homomorphic images. Homomorphisms are organized into a bipartite graph \\(\\mathcal{G}\\) , where edges connect homomorphism pairs \\(f_{1}f_{2}\\) and \\(f_{1}f_{4}\\) that are compatible, and there is no edge between homomorphisms \\(f_{1}\\) and \\(f_{3}\\) , that are not compatible. D. Homomorphisms \\(f_{5}\\) and \\(f_{6}\\) obtained by combining compatible homomorphisms \\(f_{1}, f_{2}\\) and \\(f_{1}, f_{4}\\) .", + "footnote": [], + "bbox": [ + [ + 110, + 95, + 890, + 744 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. (a)-(c) Percent of sampled trees that are compatible with given phylogenies. (d-f) Area under the precision-recall curve (AUC) calculated by varying the support threshold.", + "footnote": [], + "bbox": [ + [ + 170, + 92, + 830, + 450 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. (a) Summary statistics of methods performance on all simulated datasets. (b) Median \\(f\\) -scores of different methods on simulated datasets with varying numbers of populations. (c) Median running times of SMiTH for construction of a constrained homomorphism for a given phylogenetic tree and candidate migration tree as a function of number of migration sites", + "footnote": [], + "bbox": [ + [ + 190, + 92, + 805, + 555 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. (a) Phylogenetic trees of HCV variants from two outbreaks. Variants sampled from different hosts are highlighted in different colors. (b) Graphs formed by edges corresponding to adjacent tree nodes with unique Fitch labels. True edges are highlighted in red. (c) Consensus networks of top \\(1\\%\\) of sampled trees with respect to the objective (10). Edge thicknesses are proportional to their frequencies, true edges are highlighted in red.", + "footnote": [], + "bbox": [ + [ + 58, + 118, + 884, + 490 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7. Distributions of migration numbers for trees with different primary tumor sites.", + "footnote": [], + "bbox": [ + [ + 142, + 98, + 852, + 540 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8. Comparison of sampled compatible trees with true trees under different constraints.", + "footnote": [], + "bbox": [ + [ + 99, + 130, + 890, + 339 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9. Clone tree for Patient 3", + "footnote": [], + "bbox": [ + [ + 198, + 404, + 785, + 740 + ] + ], + "page_idx": 32 + } +] \ No newline at end of file diff --git a/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250.mmd b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250.mmd new file mode 100644 index 0000000000000000000000000000000000000000..8878a77d0415c3f89fe4b26f07772b4618447831 --- /dev/null +++ b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250.mmd @@ -0,0 +1,736 @@ + +# Outbreaks, Metastases and Homomorphisms: Phylogenetic Inference of Migration Histories of Heterogeneous Populations under Evolutionary and Structural Constraints + +Kiril Kuzmin kkuzmin1@gsu.edu + +Georgia State University + +Henri Schmidt Princeton University + +Sagi Snir University of Haifa + +Benjamin Raphael Princeton University + +Pavel Skums University of Connecticut https://orcid.org/0000- 0003- 4007- 5624 + +## Article + +Keywords: migration tree, phylogenetic inference, viral transmission, cancer metastasis + +Posted Date: September 13th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 5040045/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on August 28th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 63411- 4. + +<--- Page Split ---> + +# Outbreaks, Metastases and Homomorphisms: Phylogenetic Inference of Migration Histories of Heterogeneous Populations under Evolutionary and Structural Constraints + +Kiril Kuzmin \(^{a,\ast}\) , Henri Schmidt \(^{b}\) , Sagi Snir \(^{c}\) , Ben Raphael \(^{b}\) , Pavel Skums \(^{d,\ast}\) + +\(^{a}\) Department of Computer Science, Georgia State University, 25 Park Place, Atlanta, GA 30303, USA \(^{b}\) Department of Computer Science, Princeton University, 35 Olden St, Princeton, NJ 08540, USA \(^{c}\) Department of Evolutionary and Environmental Biology, University of Haifa, 199 Aba Khoushy Ave., Mount Carmel, Haifa 3498838, Israel \(^{d}\) School of Computing, University of Connecticut, 371 Fairfield Way, Storrs, CT 06269, USA + +## Abstract + +Many human diseases, including viral infections and cancers, are driven by the evolutionary dynamics of heterogeneous populations of genomic variants. A major type of evolutionary behavior of these populations is migration including viral transmissions and cancer metastatic spread. A common strategy for migration pathways reconstruction involves constructing a phylogenetic tree of observed genotypes and inferring its ancestral states corresponding to migration sites. Key challenges here include determining the conditions when a phylogenetic tree topology reflects the underlying migration tree structure, and balancing computational tractability, flexibility, and biological realism of inference algorithms and models. + +In this study, we address these challenges using the powerful machinery of graph homomorphisms, a mathematical concept that describes how one graph can be mapped onto another while preserving its structure. We investigate how structural constraints on migration patterns and migration tree topologies influence the relationship between phylogenies and migration trees, characterize trees compatible with a given phylogeny and propose a series of algorithms to assess whether given phylogenetic and migration trees are compatible under various migration scenarios. + +Leveraging our findings, we present a framework for inferring migration trees by sampling potential trees from a prior random tree distribution and identifying a subsample compatible with a given phylogeny. By varying prior tree distributions, this approach expands upon several existing models, offering a versatile strategy applicable to a variety of biological processes. We validate our methodology using simulated datasets and real data from studies of viral outbreaks and cancer metastasis, demonstrating its effectiveness across different contexts. + +Keywords: migration tree, phylogenetic inference, viral transmission, cancer metastasis + +## 1. Introduction + +Many human diseases are essentially evolutionary processes. This includes viral infections, driven by evolving populations of viral variants [1], as well as cancers associated with diversifying intra- tumor subclonal lineages [2]. Although the biological mechanisms of these diseases are different, both are fueled by highly mutable populations of disease- causing agents, whose + +extreme genomic diversity originates from error- prone replication processes, whether due to the lack of a proofreading mechanism in RNA- dependent RNA polymerase or retroviral reverse transcriptase in RNA viruses [3, 4, 5], or from the genetic instability of tumor cells manifesting itself in somatic mutations, chromosomal gain/loss/translocation, and aneuploidy [6, 7, 8, 9]. Consequently, the general phylogenetic methodologies applied to these populations exhibit many similarities. From a methodological standpoint, they form a unique segment of phylogenetics and phylodynamics, fostering a mutual exchange of + +<--- Page Split ---> + +18 concepts that enhances all areas of application [10, 11, 12]. + +19 A major type of evolutionary behaviour of highly mutable 20 populations is migration, wherein their members spread from 21 initial sites, seeding new populations at newly invaded sites. For 22 infectious diseases, such migrations equate to pathogen trans- 23 missions, whereas in cancer this process is identified with me- 24 tastatic spread. Thus, accurate inference of migration networks 25 of heterogeneous populations is crucial for public health and 26 medical research [13, 14, 15, 16]. + +27 The study of highly mutable population migration has been 28 significantly enhanced by groundbreaking advancements in se- 29 quenching technologies. State- of- the- art high- throughput target- 30 ted sequencing and single- cell DNA sequencing enable the cap- 31 ture of detailed population snapshots at exceptionally high res- 32olutions, facilitating fine- grained analysis down to the level of 33 individual genotypes [16, 17, 18, 19]. In particular, this allows 34 to examine population migration on the level of individual mi- 35 gration events [20, 17, 21, 22]. + +36 A wide array of methods has been developed specifically 37 for reconstructing viral and bacterial transmission trees, reflect- 38 ing the substantial interest in tracking the spread of infectious 39 diseases. The arsenal of tools available for reconstructing trans- 40 mission networks is extensive, including but not limited to Out- 41 breaker, Outbreaker 2 [23, 24], SeqTrack [25], SCOTTI [26], 42 SOPHIE [27], Phybreak [28], Bitrugs [29], BadTriP [30], Phy- 43 loscanner [31], StrainHub [32], TransPhylo [33, 34] (along with 44 its extension TransPhyloMulti [35]), STraTUS [36], TreeFix- 45 TP [37], QUENTIN [38], VOICE [39], HIVTrace [40], GHOST 46 [41], MicrobeTrace [42], SharpTNI [43], TiTUS [12], TNeT 47 [44], AutoNet [45], and others [46, 47, 34, 48, 49, 50, 51, 52, 48 53]. These tools have been instrumental in investigation of out- 49 breaks and monitoring the transmission dynamics of pathogens 50 like HIV, hepatitis C (HCV), SARS, MERS and SARS- CoV- 2 51 [54, 55, 56, 57, 20, 58]. + +Similarly, the development of methods for deducing metastatic spread histories is burgeoning, driven by advancements in single- cell DNA sequencing and CRISPR- based lineage tracing technologies. Currently, the repertoire of tools in this domain includes MACHINA [22], FitchCount (as part of the Cassiopeia suite) [21, 59], PathFinder [60], TCC, PCC, and PCCH [61]. Notwithstanding the relatively shorter list, the impact of these tools is growing, with several studies published recently leveraging these methods to gain insights into the mechanisms of metastatic spread [21, 62, 63, 64]. + +Despite significant progress in the field, the wide variety of existing methods underscores that the challenge of accurately inferring heterogeneous population migration remains unresolved. This diversity of approaches indicates both the complexity of the problem and the ongoing efforts to refine and improve upon existing methods. Additionally, a major barrier to advancement in the field is the relative isolation of viral and cancer genomics fields. Some of the aforementioned tools are based on conceptually similar techniques - this applies, for example, to STraTUS (which focus on viral transmission) and FitchCount (which addresses metastatic spread) that, as demonstrated in this paper, yield virtually identical results when applied to the same datasets. This lack of interdisciplinary exchange of ideas often leads researchers to inadvertently duplicate efforts, thereby impeding progress in both fields. + +Phylogenetics and phylodynamics provide the most widely used methodological frameworks for migration tree/network reconstruction [35, 22]. However, their application in this context is not straightforward. A phylogenetic tree does not directly equate to a migration tree [35], as the nodes in a phylogenetic tree represent divergences of lineages rather than specific migration events [35, 65]. While some of these divergences may result from migration, others occur within previously invaded sites. Therefore, deriving a migration tree from a phyloge + +<--- Page Split ---> + +86 netic tree essentially requires solving an ancestral trait infer- 120ence problem, wherein the internal nodes of a phylogenetic tree are annotated with labels that indicate whether each divergence event occurred within a site or as a result of seeding of a new site upon migration. Furthermore, it is crucial to effectively leverage the full spectrum of intra-site (within-host or within- tumor) population diversity uncovered by high-throughput sequencing, that often provide a strong signal for migration inference. For example, the paraphyletic relationships between populations suggest recent migrations between corresponding sites [66, 67, 39, 68]. + +The challenges mentioned above highlight several crucial questions that remain unresolved and warrant further exploration. These questions include: + +1) To what extent and under which conditions does the topol-ology of a phylogenetic tree reflect the structure of the underlying migration tree? This question becomes particularly important when the level of paraphyly in the labeled phylogeny is low, a situation not uncommon as the paraphyletic signal tends to diminish over time and with smaller sample sizes [66]. A- number of studies have looked at this subject but their conclusions were mixed. Some studies have found that certain migration patterns, such as super-spreading, migration chains, or more generally, migrations within networks formed by different models like Erdos-Renyi [69], preferential attachment [70], or Watts-Strogatz models [71], lead to quantitatively distinct phylogenetic tree topologies [72, 73, 74]. Additionally, the spatial structure and dynamics of heterogeneous populations, which are directly related to migrations pathways, have been shown to affect the phylogeny structure of both viral and tumor populations [75, 76, 77, 78]. On the other hand, other studies report that the direct impact of migration patterns on phylogenetic trees ranges from minimal to moderate [79, 80, 81, 82]. Finally, certain studies have drawn mixed conclusions, indicat + +ing that while some migration characteristics are reflected in the phylogeny, others are not [83]. + +2) How can we limit the solution space to balance computational feasibility, accuracy of inference, generalizability, and biological realism? The space of migration trees compatible with given phylogenetic trees is often vast, and its properties are not well understood [36, 12]. A sampling-consensus approach is one method to address solution ambiguity, where feasible migration trees are sampled and summarized in a weighted consensus graph, with weights reflecting posterior probabilities of edges [44, 12, 43, 26, 21]. However, the size of solution space may restrict the depth of sampling. As a response, it is common practice to narrow down the solution space to a set of plausible migration trees optimizing a specific objective function under evolutionary-based constraints. Employing constrained models also aids in preventing overfitting in presence of missing data and errors. + +Various objectives and constraints have been implemented by existing methods. Limiting number of migration events, sizes of bottlenecks or numbers of back-migrations [31, 12, 44, 22, 59, 21, 61, 36, 25] is more computationally efficient and scalable due to utilization of dynamic programming [84, 85], making such approaches practical in both molecular epidemiology and computational oncology. These can also be formulated as Integer Linear Programming (ILP) problems [86] and solved with reasonable efficiency using existing ILP solvers. Models with more complex Bayesian objectives with constraints regularized as priors [35, 28, 26, 23, 30] offer a richer, biologically nuanced perspective but suffer from scalability issues and usually rely on generic methods like Markov Chain Monte Carlo (MCMC) sampling, which may not yield optimal solutions, in part due to a lack of problem-specific mathematical strategies. Balancing computational efficiency with biological comprehensiveness presents a notable challenge, compounded by the un + +<--- Page Split ---> + +certainty of how much constraints and objectives truly limit the solution space [36, 12]. + +3) How to incorporate a variety of models for phylogenetic inference of migrations into a unified modular computational framework? + +Migration inference models draw on varied biological or epidemiological assumptions. For instance, viral transmission inference often incorporates case-specific temporal data like infectious periods, exposure intervals, symptom onset, diagnosis or sample collection dates to establish order of infections and eliminate unlikely transmission links [23, 47, 28, 33, 12, 26, 29]. In some rare cases, contact networks are known and can be used for the same purpose [87, 58]. While effective, such data is often unavailable, non-informative, or sensitive, particularly for endemic and pandemic diseases caused by HIV, Hepatitis C, SARS-CoV- 2, or Influenza [27, 38, 53]. In situations where case-specific data cannot be used, genomic epidemiology tools resort to broader assumptions, like the expected degree distribution of transmission networks implied by a structure of a susceptible population [38, 27]. Similarly, methods for inferring metastatic spread use constraints defined by so- called migration patterns that reflect realistic cancer migration scenarios, such as monoclonal, polyclonal or multi- source seeding [22, 61]. A commonality across all these methods is the use of structural constraints on feasible transmission networks that are considered as subgraphs of a larger "pattern" graph. It suggests need for a versatile, modular migration inference framework that integrates these varied approaches on a unified algorithmic and mathematical basis, akin e.g. to the BEAST framework for Bayesian evolutionary analysis [88]. + +Addressing these challenges requires a comprehensive investigation into the mathematical properties of migration trees compatible with a given phylogeny, a topic that is not yet fully understood [36]. Our study aims to advance this area by devel + +oping a novel methodology based on powerful techniques from graph theory and combinatorics. Additionally, we will use this methodology to introduce novel algorithmic approaches for inferring migration trees. + +A number of earlier studies has achieved important progress in this area. Several studies noticed that migration trees compatible with a given phylogeny correspond to partitions of the phylogeny's node set or to coloring of its branches [51, 36, 33, 34]. These observations have informed the development of methods to enumerate and sample these trees, assuming a complete migration bottleneck [36] and a known sequence of migrations [89]. In fact, as we argue in this paper, the relation between a phylogenetic tree and a migration tree is described by the concept of a graph homomorphism that generalize both partitions and colorings. + +Graph homomorphism is essentially a mapping between the vertices of two graphs that preserve their structure [90]. The theory of graph homomorphisms is well- established area of discrete mathematics, with deep results and rich methodology. All types of migration trees discussed so far can be described by a graph homomorphism with specific constraints. For example, migration trees compatible with a given phylogeny under the assumptions of complete sampling and complete bottleneck, that has been studied in [51, 36] are minors [91, 92] of the phylogeny or, equivalently, its homomorphic images such that inverse images of all vertices are connected subtrees. + +We use mathematical and algorithmic machinery of graph homomorphism theory to delve into the details of migration inference. Specifically, we provide necessary and sufficient conditions describing trees compatible with a given phylogeny and propose a series of algorithms that evaluate the compatibility of phylogenetic and migration trees under various evolutionary scenarios through the construction of corresponding homomorphisms. We examine particular structural constraints on migra + +<--- Page Split ---> + +tion patterns and migration tree topologies to understand their influence on the relationship between a phylogeny and a migration tree. + +Based on aforementioned insights, we propose a general framework for migration inference that samples candidate migration trees from a chosen prior random tree distribution, and identifies a subsample of trees compatible with a given phylogeny. By varying prior tree distributions, this approach expands upon and generalizes several existing models, offering versatile and computationally efficient strategy applicable to a variety of biological processes associated with heterogeneous population migration. Crucially, the proposed framework is computationally fast, enabling biomedical and public health researchers to quickly test different tree priors that represent common migration models. Beyond migration reconstruction, proposed methods can be used for investigating how phylogenies constrain the space of possible migration trees, for inferring an order of known or suspected migration events, and for determining potential migration events that are definitively ruled out by a phylogeny [36]. + +Proposed methodology was validated using both simulated datasets and real experimental data gathered from studies of real outbreaks and cancer metastasis, demonstrating its effectiveness and applicability across different contexts. + +## 2. Methods + +### 2.1. Basic definitions + +Throughout this paper, we consider a pair of trees: a phylogenetic tree and a migration tree. For clarity, we refer to elements of a phylogeny as nodes, and to elements of a migration network as vertices, and denote them by Greek and Latin letters, respectively. + +The problem of migration network inference is set up as follows. The input is a phylogenetic tree \(\Psi = (V(\Psi), E(\Psi))\) + +with the leaf set \(L(\Psi)\) representing genomic variants belonging to different subpopulations (or demes), denoted by \(\mathcal{L}\) . The tree \(\Psi\) can be a standard binary phylogeny or non- binary mutation tree used in most cancer studies. Each leaf \(\lambda \in L(\Psi)\) has an assigned site label (or color) \(l_{\lambda} \in \mathcal{L}\) . The aim is to expand this labeling from the leaves to all nodes in the tree, creating a full labeling \(f: V(\Psi) \to \mathcal{L}\) . In this model, any multi- colored tree edge \(\alpha \beta\) represents a migration of genomic variants between demes \(f(\alpha)\) and \(f(\beta)\) . The migration tree \(T = T(\Psi, I)\) and with vertices \(V(T) = \mathcal{L}\) , is then formed by contracting the nodes with the same color [51]. + +As mentioned in the introduction, researchers often seek migration trees satisfying particular constraints restricting types of migration or tree topologies. These constraints can be encoded using a transition pattern graph \(G\) that describes permissible patterns of migration (specific examples are provided in the following subsection). We will first consider the situation when \(G\) is a simple graph; later on, it will be extended to the cases when \(G\) is a random graph characterized by some probability distribution. In this model, a migration tree should be isomorphic (i.e. identical up to relabeling of vertices) to a subgraph of the transition pattern. Any corresponding labeling will be called feasible. + +The relations between the phylogeny \(\Psi\) , the migration tree \(T\) and the transition pattern \(G\) can be captured using the concept of a graph homomorphism. A homomorphism \(f: \Psi \to G\) [90] is an adjacency- preserving mapping between vertex sets of these graphs, i.e. \(f(u)f(v) \in E(G)\) if \(uv \in E(\Psi)\) . For the sake of mathematical rigor, here and throughout this paper we assume that a transition pattern is reflexive, i.e. every vertex is adjacent to itself. With this condition in place, any feasible labeling \(f\) is a homomorphism from \(\Psi\) to \(G\) , making the migration inference problem essentially a problem of finding such a homomorphism. In graph theory, this type of problems is sometimes + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. SMiTH: Sampling Migration Trees with Homomorphisms. A: Input phylogenetic tree. B: Distribution of possible migration trees. Parallel rectangles depict a probability density function, with each rectangle's width proportional to the corresponding probability. In practice, the distribution is represented either by a random graph model or by a stochastic graph generation procedure. C: Candidate migration trees sampled from the distribution B. D: Homomorphisms from the phylogeny to three sampled trees. In the phylogenies, nodes are color-coded by their homomorphic images in a migration tree. The phylogeny layouts in the middle of each subfigure showcases how homomorphism transforms them into sampled trees. E: consensus solution derived from homomorphisms in D. The solution is shown as potential color distributions for the phylogeny's nodes (left) or as a graph where possible migration edges are weighted according to the number of supporting solutions (right), with the edge thickness indicating weight.
+ +referred to as an \(G\) - coloring of the tree \(\Psi\) [90]. + +Throughout this paper, we use standard graph theory no- 301. tations. We denote by \(\Psi_{\alpha}\) a subtree of \(\Psi\) rooted at the node302. \(\alpha\) . For any graph \(G\) , \(G[X]\) represents the subgraph induced by303. the subset of vertices \(X\) . We use \(u\sim v\) to indicate that ver- 304. tices \(u\) and \(v\) are adjacent. The set of neighbors of a vertex \(u_{305}\) in \(G\) is denoted as \(N_{G}(u)\) , \(\deg_{G}(u) = |N_{G}(u)|\) is the degree of306. \(u\) , \(N_{G}[u] = N_{G}(u)\cup \{u\}\) and \(N_{G}[X]\) is the union of the neigh- 307. borhoods of all vertices in a set \(X\) . Additionally, the distance between any two vertices \(u\) and \(v\) in \(G\) is denoted by \(d_{G}(u,v)\) , and \(P_{G}(u,v)\) refers to the corresponding shortest path between + +them. When the graph is clear from a context, we may omit the subscripts in these notations. In the case of a phylogeny \(\Psi\) , all distances and paths are undirected. + +Additionally, to simplify the notation, we will apply settheoretical operations (e.g. intersection and union) directly to subgraph of \(G\) , with the understanding that the resulting subgraph is induced by the set obtained from applying these operations to the vertex sets of the original subgraphs. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Solutions for the Migration History Inference Problem under various constraints. For each scenario, a phylogenetic tree is illustrated in two different layouts on the left and in the middle, with the corresponding migration tree displayed on the right. In the phylogenies, nodes are color-coded by their homomorphic images in the migration tree. The layout in the middle showcases how homomorphism transforms a phylogeny into a migration tree. (a) Unconstrained solution. Subtrees formed by blue and red nodes are not connected, indicating a violation of the convexity constraint. (b) Convex solution. Each color-coded subtree is connected, but compactness is violated in the subtree rooted at node 7. (c) Convex and compact solution.
+ +### 2.2. Migration inference under structural constraints + +The simplest form of the migration inference challenge appears when the vertices of the transition pattern graph \(G\) directly correspond to the labels of the phylogenetic tree leaves, that is, \(V(G) = \mathcal{L}\) . This variant is referred to as labeled inference. The transition pattern \(G\) can be an unweighted graph or a random graph with specified edge probabilities. Several scenarios exemplify this problem: + +- For viral transmissions, \(G\) could reflect a contact network of potential hosts [87, 58], where an infection spreads through direct interactions within this network. +- Another viral transmission model assumes that transmission is only possible between hosts with overlapping ex- + +posure intervals [30, 12]. In this case \(G\) is an interval graph [93], i.e. a graph with vertices representing time intervals and edges connecting vertices whose intervals intersect. + +- For metastatic spread inference, \(G\) may represent a circulatory network [94, 95, 96], with vertices representing organs and edges reflecting the prior probabilities of cancer spreading between organs. + +## Problem 1 (labeled migration inference) + +Input: + +(a) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) forming the label set \(\mathcal{L}\) ; + +(b) a transition pattern \(G\) with \(V(G) = \mathcal{L}\) ; + +Output: + +(c) a homomorphism \(f: \Psi \to G\) such that \(f(\lambda) = l_{\lambda}\) for every \(\lambda \in L(\Psi)\) . + +In practical settings, however, Problem 1 might be too restrictive or not fully reflective of reality. The main issue is the absence of a known mapping between the leaf labels and the vertices of the transition pattern. In other words, the pattern \(G\) represents the permissible topology of migration rather than specific allowed migration links. For instance, we may expect that a viral transmission network likely includes a super-spreader, but the exact subpopulation associated with this super-spreader is not known. The following models to this variant of the problem: + +- Viral transmission: The transition pattern \(G\) could represent a random graph that describes expected characteristics of transmission trees, like being scale-free or having a particular expected degree distribution. Such models draw on known expected properties of contact networks + +<--- Page Split ---> + +essential for infection spread [97, 98, 99, 100] without requiring actual host contact information, and have been explored in several studies [38, 27]. + +Metastatic spread inference: In this case, a transition pattern \(G\) may describe plausible evolutionary scenarios of cancer migration, such as monoclonal or polyclonal seeding from single or multiple sources [22, 61]. A related idea has been applied in studies analyzing CRISPR- based lineage tracing phylogenies, where \(G\) specifies so- called star homoplasy model [101]. + +In this version of migration inference problem, is no explicitly given mapping between the set of leaf labels and the vertices of the transition pattern. Instead, a feasible homomorphism should map leaves with distinct labels to distinct vertices of \(G\) and vice versa, i.e. for any \(\lambda_1, \lambda_2 \in L(\Psi)\) we have \(f(\lambda_1) \neq f(\lambda_2)\) whenever \(l_{\lambda_1} \neq l_{\lambda_2}\) (Fig. 2 (a)). We will call homomorphisms satisfying this requirement a label- distinctiveness homomorphism. It should be noted that finding such a homomorphism seems to be a non- standard variant of a graph homomorphism problem, that, to the best of our knowledge, has not been studied previously. + +In such context, the first question to be asked is whether given subtree \(T\) of the transition pattern \(G\) is compatible with the phylogeny \(\Psi\) , i.e. whether there exist a label- distinctive homomorphism \(\Psi \to T\) . + +## Problem 2 (unlabeled migration inference) + +Input: + +(a2) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) + +(b2) a tree \(T\) + +Output: + +(c2) a label- distinctive homomorphism \(f: \Psi \to T\) . + +A more restricted version of Problem 3 includes an additional convexity constraint [102, 103], where tree nodes mapping to the same vertex of \(G\) form a connected subtree of \(\Psi\) (Fig. 2 (b)): + +## Problem 3 (convex unlabeled migration inference) + +Input: (a2) and (b2) + +Output: + +(c3) a label- distinctive homomorphism \(f: \Psi \to T\) such that induced subgraphs \(\Psi [f^{-1}(v)]\) are connected for all \(v \in V(T)\) . + +We will refer to a homomorphism satisfying (c3) as a convex homomorphism. Convex homomorphisms to trees describe migrations with a complete bottleneck; such migration trees were the focus of extensive research in previous studies [51, 36, 89]. + +To define another type of constraints, consider a labeling \(l\) of nodes of the phylogeny \(\Psi\) . The labeling \(l\) is compact if the label of the most recent common ancestor (MRCA) for any group of leaves matches one of the labels within that group. + +The rationale for this is based on the understanding that migrations within any given subtree could follow one of the following scenarios: either (i) migrations involve only the demes represented by the leaves of the subtree, or (ii) migrations include external demes, but lineages from these demes were not sampled due to extinction or incomplete collection. Although both scenarios are feasible, the first is more parsimonious, especially when sampling is dense and migration events span a short period of time. Hence, homomorphisms that align with this compact labeling are referred to as compact. + +## Problem 4 (compact unlabeled migration inference) + +Input: (a2) and (b2) + +Output: + +(c4) a label- distinctive homomorphism \(f: \Psi \to T\) such that for any node \(\alpha \in V(\Psi)\) we have \(f(\alpha) \in f(L(\Psi_\alpha))\) . + +<--- Page Split ---> + +Finally, it is often desirable to produce a sample of potential solutions rather than a single solution. Such approach offers statistical backing for potential migration network edges and logically shifts the migration inference problem to a Bayesian paradigm. Potential solutions can be sampled from a random graph distribution or, if a transition pattern is a deterministic graph, from the uniform distribution of its subgraphs with specified properties (e.g. subtrees). The sampling-based version of the migration inference problem can be formulated as follows: + +## Problem 5 (unlabeled migration sampling) + +Input: + +(a5) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) ; +(b5) a random transition pattern \(G\) with edge probabilities \(p\) : \(E(G) \to [0,1]\) that define a distribution \(\mathcal{D}\) of subtrees of \(G\) . + +Output: + +(c5) a sample \(S = \{T_1, \ldots , T_m\}\) from the distribution \(\mathcal{D}\) , where each subtree \(T_i\) is a homeomorphic image of \(\Psi\) , and the corresponding label-distinctive homomorphisms \(f_i: \Psi \to T_i\) are possibly convex and/or compact. + +### 2.3. Labeled migration inference + +Problem 1 can be efficiently solved in polynomial time using dynamic programming, as detailed in Algorithm 1. Despite its simplicity, we include the algorithm here due to its relevance for subsequent, more complex methods. + +### 2.4. Unconstrained unlabeled migration inference + +A possible strategy to solve Problem 2 involves identifying a bijection \(g: \mathcal{L} \to V(T)\) that can be extended to a homomorphism \(\Psi \to T\) using Algorithm 1. We will describe such bijections as feasible. + +## Algorithm 1 Unlabeled migration inference + +1: Let \(\rho\) be the root of \(\Psi\) . Perform a post- order traversal of the tree \(\Psi\) . +2: for every node \(\alpha\) of the traversal do +3: construct a set of potential images \(I(\alpha) \subseteq \mathcal{L}\) : +4: if \(\alpha\) is a leaf then +5: \(I(\alpha) \leftarrow \{I(\alpha)\}\) ; +6: else +7: Suppose that \(\beta_1, \ldots , \beta_k\) are children of \(\alpha\) . Then \(I(\alpha)\) consists of vertices \(x \in V(G)\) such that there exist vertices \(y_i \in I(\beta_i)\) , \(i = 1, k\) such that \(y_i \sim x\) . +8: end if +9: end for +10: if \(I(\rho) = \emptyset\) then +11: homomorphism \(f\) does not exist +12: else +13: perform a pre- order traversal of \(\Psi\) and construct a homomorphism \(f\) as follows +14: for every node \(\alpha\) of the pre- order do +15: if \(\alpha = \rho\) then +16: select any \(v \in I(\rho)\) and set \(f(\rho) \leftarrow v\) ; +17: else +18: Let \(\omega\) be the parent of \(\alpha\) . Choose \(v \in I(\alpha)\) such that \(v \sim f(\omega)\) and set \(f(\alpha) \leftarrow v\) . +19: end if +20: end for +21: end if + +Let \(L_g(u) = \{\lambda \in L(\Psi): g(l_\lambda) = u\}\) be the set of leaves in \(\Psi\) whose labels map to \(u\) . The following theorem establishes a necessary and sufficient condition for the bijection feasibility. + +Theorem 1. The bijection \(g: \mathcal{L} \to V(T)\) is feasible if and only if + +\[d_T(u_1, u_2) \leq d_{\Psi}(\lambda_1, \lambda_2) \quad (1)\] + +for all \(u_1, u_2 \in V(T)\) , \(\lambda_1 \in L_g(u_1)\) , \(\lambda_2 \in L_g(u_2)\) . + +Proof. The necessity of the condition stated in the theorem is a known property of graph homomorphisms [90]. However, it is generally not sufficient [90]. For our specific type of the homomorphism problem, we will demonstrate that it indeed suffices. + +Consider a bijection \(g\) satisfying the condition (1). Define \(C_\alpha \subseteq V(T)\) as the image set of its clade, that is, \(C_\alpha = g(l_\lambda): \lambda \in L(\Psi_\alpha)\) . Additionally, for a node \(\alpha \in V(\Psi)\) and a vertex \(u \in V(T)\) , define \(B_\alpha (u)\) as a ball centered at \(u\) in \(T\) , with + +<--- Page Split ---> + +460 radius + +\[r_{\alpha ,u} = \min_{\lambda \in L_{\kappa}(u)\cap L(\Psi_{\alpha})}d_{\Psi}(\alpha ,\lambda) \quad (2)\] + +461 i.e., \(B_{\alpha}(u) = \{v\in V(T):d_{T}(u,v)\leq r_{\alpha ,u}\}\) + +462 We proceed by establishing two auxiliary facts. The first follows directly from the properties of a tree: + +464 Lemma 1. Let \(X_{1},\ldots ,X_{k}\) be subsets of vertices of the tree \(\frac{486}{487}\) \(T\) such that each subsets induces a connected subgraph and \(\bigcap_{i = 1}^{k}X_{i}\neq \emptyset\) . Then \(N[\bigcap_{i = 1}^{k}X_{i})] = \bigcap_{i = 1}^{k}N[X_{i}]\) + +467 Suppose now that \(I(\alpha):\alpha \in V(\Psi)\) are the sets of potential 489 node images produced by Algorithm 1, given the matching of 490 leaf labels in \(\Psi\) and vertices of \(T\) through the bijection \(g\) . These 491 sets are described by the following lemma. + +471 Lemma 2. For every node \(\alpha \in V(\Psi)\) \(I(\alpha) = \bigcap_{u\in C_{\alpha}}B_{\alpha}(u)\) + +472 Proof. The proof of the lemma proceeds by induction. Consider a node \(\alpha \in V(\Psi)\) . The lemma's assertion is trivially true 495 when \(\alpha\) is a leaf. Assume now that \(\alpha\) is an internal node with 496 children \(\beta_{1},\ldots ,\beta_{k}\) , each set \(I(\beta_{i})\) is non- empty and, by the in- 497 ductive assumption, + +\[I(\beta_{i}) = \bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u). \quad (3)\] + +477 Then \(C_{\alpha} = \bigcup_{i = 1}^{k}C_{\beta_{i}}\) ; furthermore, according to Algorithm 1502 and the equality (3) we have + +\[I(\alpha) = \bigcap_{i = 1}^{k}N[I(\beta_{i})] = \bigcap_{i = 1}^{k}N\left[\bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u)\right]. \quad (4)\] + +479 Now consider a vertex \(u\in C_{\alpha}\) . Let \(\beta (u)\) be the child of \(\alpha^{504}\) 480 such that \(r_{\beta (u),u} = \min_{i = 1}^{k}r_{\beta ,u}\) ; in cases where there are multiple505 481 such children, we pick any of them. Consequently, we have506 \(r_{\alpha ,u} = r_{\beta (u),u} + 1\) , leading to the relation \(B_{\alpha}(u) = N[B_{\beta (u)}(u)]\) 507 483 Furthermore, it is obvious that \(B_{\beta (u)}(u)\subseteq B_{\beta_{i}}(u)\) for all nodes508 484 \(\beta_{i}\) such that \(u\in B_{\beta_{i}}(u)\) . Together, these observations imply the + +following sequence of equalities: + +\[\bigcap_{u\in C_{\alpha}}B_{\alpha}(u) = \bigcap_{u\in C_{\alpha}}N[B_{\beta (u)}(u)] = \bigcap_{i = 1}^{k}\bigcap_{u:\beta (u) = \beta_{i}}N[B_{\beta_{i}}(u)] =\] \[\qquad = \bigcap_{i = 1}^{k}\bigcap_{u\in C_{\beta_{i}}}N[B_{\beta_{i}}(u)] \quad (5)\] + +By Lemma 1, \(N\left[\bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u)\right] = \bigcap_{u\in C_{\beta_{i}}}N[B_{\beta_{i}}(u)]\) , and thus the expressions (4) and (5) are equal. This completes the proof of Lemma 2. + +According to Lemma 2, Algorithm 1 succeeds whenever \(\bigcap_{u\in C_{\alpha}}B_{\alpha}(u)\neq \emptyset\) for every node \(\alpha \in V(\Psi)\) . To establish that this condition holds, we invoke so- called Helly property of subtrees. A family of sets \(S_{1},\ldots ,S_{k}\) has a Helly property [104] if \(\bigcap_{i = 1}^{k}S_{i}\neq \emptyset\) whenever \(S_{i}\cap S_{j}\neq \emptyset\) for every \(i,j\in [k]\) ; in other words, the existence of non- empty pairwise intersections guarantees a non- empty total intersection. + +Subtrees of a given tree are known to have the Helly property [105]. The subsets \(B_{\alpha}(u)\) obviously induce subtrees of \(T\) . Therefore, to prove the theorem, it is sufficient to demonstrate that for every node \(\alpha \in V(\Psi)\) and for every pair of vertices \(u_{1},u_{2}\in C_{\alpha}\) , the intersection \(B_{\alpha}(u_{1})\cap B_{\alpha}(u_{2})\) is non- empty. + +Select two leafs \(\lambda_{1},\lambda_{2}\in L(\Psi_{\alpha})\) that minimize the distance between nodes of the sets \(L_{g}(u_{1})\cap L(\Psi_{\alpha})\) and \(L_{g}(u_{2})\cap L(\Psi_{\alpha})\) . According to the theorem's conditions, we have: + +\[d_{T}(u_{1},u_{2})\leq d_{\Psi}(\lambda_{1},\lambda_{2})\leq d_{\Psi}(\alpha ,\lambda_{1}) + d_{\Psi}(\alpha ,\lambda_{2}). \quad (6)\] + +This implies the existence of a path between \(u_{1}\) and \(u_{2}\) in \(T\) with a length at most \(d_{\Psi}(\alpha ,\lambda_{1}) + d_{\Psi}(\alpha ,\lambda_{2})\) . On this path, there is at least one vertex \(v\) such that \(d_{T}(u_{1},v)\leq d_{\Psi}(\alpha ,\lambda_{1})\) and \(d_{T}(u_{2},v)\leq d_{\Psi}(\alpha ,\lambda_{2})\) . Consequently, \(v\) belongs to both \(B_{\alpha}(u_{1})\) and \(B_{\alpha}(u_{2})\) , thereby completing the proof. + +Theorem 1 establishes that the Unlabeled Migration Inference problem is algorithmically equivalent to the problem of + +<--- Page Split ---> + +finding a bijection that satisfies (1). First of all, this allows us to demonstrate that Problem 2 is \(\mathcal{NP}\) - hard. We will prove its through a reduction from the following problem: + +## Graph Bandwidth problem. + +Given: A graph \(G\) + +Find: The minimal integer \(K = bw(G)\) for which there exists is a bijection \(f:V(G)\to \{1,\ldots ,|V(G)|\}\) such that + +\[|f(u) - f(v)|\leq K\mathrm{~for~all~}u\sim v. \quad (7)\] + +The Graph Bandwidth problem is \(\mathcal{NP}\) - hard [106], and, moreover, it cannot be approximated within any constant factor unless \(\mathcal{P} = \mathcal{NP}\) , even when the input graph \(G\) is a tree [107]. + +Theorem 2. The Unlabeled Migration Inference problem is \(\mathcal{NP}\) - hard, even when all leaf labels in the phylogeny \(\Psi\) are unique. + +Proof. For our purposes, it is more convenient to use an equiv + +alent condition for Graph Bandwidth problem: + +\[|f(u) - f(v)|\leq Kd_G(u,v)\mathrm{~for~all~}u,v\in V(G). \quad (8)\] + +The fact that (8) implies (7) is obvious. To demonstrate that (7) implies (8), consider the shortest \((u,v)\) - path in \(G\) \((u = 553\) \(x_{0},x_{1},\ldots ,x_{d - 1},x_{d} = v)\) , where \(d = d_G(u,v)\) . Then we have + +\[\begin{array}{r l r}{{|f(u)-f(v)|=|\sum_{i=1}^{d}(f(x_{i-1})-f(x_{i}))|\leq}}&{{}}&{{}}\\ &{}&{{}\leq\sum_{i=1}^{d}|(f(x_{i-1})-f(x_{i}))|\leq K d.}}\end{array} \quad (9)\] + +Now suppose that \(T^{\prime}\) is an input tree of Graph Bandwidth problem. We can assume that \(bw(T^{\prime})\geq 3\) , since graphs where \(bw(T^{\prime})\leq 2\) are recognizable in linear time [108]. To construct an instance for Problem 2, for an integer \(K\geq 3\) , we proceed as follows: + +1) The input phylogeny \(\Psi_K\) is constructed by (a) subdivid + +ing every edge of \(T^{\prime}\) into \(K\) edges; (b) attaching a leaf labeled \(u^{\prime}\) to every node \(u\in V(T^{\prime})\) + +2) The input tree \(T\) is an \(n\) -vertex path \(P_{n}\) with \(V(T) = \{1,\ldots ,n\}\) + +For the trees constructed in this manner we have: + +\[(a)L(\Psi_K) = \{u':u\in V(T')\} ;\] \[(b)d_{\Psi_K}(u',v') = Kd_S(u,v) + 2;\] \[(c)d_T(i,j) = |i - j|.\] + +We will demonstrate that a polynomial- time algorithm for Problem 2 leads to a \(\frac{5}{3}\) - approximation algorithm for the Graph Bandwidth problem. Assume the existence of such an algorithm for Problem 2. Let \(K^{*}\) be the smallest integer for which this algorithm produces a sought- for homomorphism \(\Psi_{K^{*}}\to T\) . To establish the \(\frac{5}{3}\) approximation factor, we need to demonstrate the following relationship: + +\[K^{*}\leq bw(T^{\prime})\leq \frac{5}{3} K^{*} \quad (9)\] + +To establish an upper bound, let us consider the feasible bijection \(g^{*}:L(\Psi_{K^{*}})\to \{1,\ldots ,n\}\) . We extend this bijection to the nodes of \(V(T^{\prime})\) by setting \(g^{*}(u) = g^{*}(u^{\prime})\) . Given the inequality (1) and assuming that \(K^{*}\geq 3\) , we have: + +\[\begin{array}{r l r} & {} & {|g^{*}(u) - g^{*}(v)| = d_{T}(g^{*}(u^{\prime}),g^{*}(v^{\prime}))\leq d_{\Psi_{K^{*}}}(u^{\prime},v^{\prime}) =}\\ & {} & {= K^{*}\cdot d_{T^{\prime}}(u,v) + 2\leq \frac{5}{3} K^{*}d_{T^{\prime}}(u,v).} \end{array} \quad (9)\] + +This implies that \(bw(S)\leq \frac{5}{3} K^{*}\) + +Conversely, if \(bw(T^{\prime}) = K< K^{*}\) , and the mapping \(g:V(T^{\prime})\to \{1,\ldots ,n\}\) is the bijection reflecting this bandwidth, then we can extend it to \(L(\Psi_K)\) by setting \(g(u^{\prime}) = g(u)\) . This yields: + +<--- Page Split ---> + +\[d_{T}(g(u^{\prime}),g(v^{\prime})) = |g(u) - g(v)|\leq Kd_{T^{\prime}}(u,v)<\] \[< Kd_{T^{\prime}}(u,v) + 2 = d_{\Psi_{K}}(u^{\prime},v^{\prime}).\] + +This inequality contradicts the assumption that \(K^{*}\) is minimal. Therefore, we must have \(bw(S) \geq K^{*}\) . This completes the proof. + +Although the Unlabeled Migration Inference problem is \(\mathcal{NP}\) - hard, we can use Theorem 1 to approach it using Integer Linear Programming (ILP). We define a feasible bijection \(g: \mathcal{L} \to S_{88}\) \(V(T)\) using binary variables \(x_{iu}\) , where \(x_{iu} = 1\) if \(g(i) = u\) , for each \(i \in \mathcal{L}\) and \(u \in V(T)\) . The ILP formulation to find such a bijection is as follows: + +\[\sum_{i \in \mathcal{L}} \sum_{u \in V(T)} \delta (i) \deg (u) x_{iu} \to \max \quad (10)\] + +S.t. + +\[\sum_{u \in V(T)} x_{iu} = 1, \quad i \in \mathcal{L}; \quad (11)\] + +\[\sum_{i \in \mathcal{L}} x_{iu} = 1, \quad u \in V(T); \quad (12)\] + +\[x_{iu} + x_{jv} \leq 1, \quad i, j \in \mathcal{L}, u, v \in V(T) \quad (13)\] + +\[x_{iu} + x_{iv} + x_{ju} + x_{jv} \leq y_{ij} + 1 \quad i, j \in \mathcal{L}, uv \in E(T); \quad (14)\] + +\[\sum_{i,j \in \mathcal{L}} y_{ij} = n - 1. \quad (15)\] + +In this formulation, constraints (11) and (12) ensure that \(x\) encodes a bijection, while constraints (13) guarantee that the bijection adheres to the conditions of Theorem 1. The auxiliary variables \(y_{ij}\) in constraints (14) indicate whether a pair of leaf labels map to adjacent vertices in \(T\) , with constraint (15) ensuring the inferred migration network forms a tree. The objective function (10) facilitates the search for a solution by leveraging the relationship between population diversity and popula + +tion age [109, 110, 111], that suggests that more diverse populations, which are likely older, are also more probable origins of migration [38, 31, 66]. Consequently, it is more likely that such populations correspond to high- degree vertices in the tree \(T\) . Here a coefficient \(\delta (i)\) represents the genetic diversity of the \(i\) th subpopulation, measured as allelic entropy averaged over all allelic positions. + +### 2.5. Migration inference under convexity constraints + +In this section, a convex label- distinctive homomorphism \(\Psi \to T\) will be called feasible. When such homomorphism exists, \(T\) can be obtained from \(\Psi\) by a series of edge contractions, making \(T\) a minor of \(\Psi\) . Generally, a graph \(G_{1}\) is a minor of a graph \(G_{2}\) if \(G_{1}\) can be obtained from \(G_{2}\) by edge contractions, edge removals and node removals [112]. It is known that the problem of detecting whether a given graph is a minor of another graph is \(\mathcal{NP}\) - hard, even when input graphs are trees [113, 114]. However, minors associated with our problem satisfy a more stringent set of conditions than general graph minors: in our case, only edge contractions are allowed and, in addition, contractions of edges between labeled nodes with different labels are forbidden. We suggest that in practical settings feasible homomorphisms can be efficiently found and enumerated using dynamic programming. + +The following simple property of convex homomorphisms will be useful for our subsequent analysis: + +Lemma 3. Suppose that \(f: \Psi \to T\) is a convex homomorphism. Then \(T'\) with is a subtree of \(T\) if and only if \(\Psi ' = f^{- 1}(T')\) is a subtree of \(\Psi\) . + +Proof. Homomorphic image of a connected subgraph is connected, as implied by the definition of a homomorphism. The converse is also true, if \(|T'| = 1\) . Suppose that \(|T'| \geq 2\) and the subgraph \(f^{- 1}(T')\) is not connected. Consider two of its connected components, \(\Psi_1'\) and \(\Psi_2'\) , such that the unique path \(P\) + +<--- Page Split ---> + +between the nodes \(\alpha \in \Psi_1'\) and \(\beta \in \Psi_2'\) is shortest among all such paths between different components. Then \(|P| \geq 3\) and \(T'' = f(P \setminus \{\alpha , \beta \}) \cap T' = \emptyset\) . Moreover, the vertices \(f(\alpha)\) and \(f(\beta)\) are adjacent to vertices of \(T''\) . This leads to two distinct paths between \(f(\alpha)\) and \(f(\beta)\) – one in \(T'\) and another passing through \(T''\) . This contradicts the fact that \(T\) is a tree. \(\square\) + +Next, we describe the proposed algorithmic approach. Initially, we simplify the original phylogenetic tree \(\Psi\) by collaps ing paths between leaves sharing identical labels into a single node, a step made feasible by the convexity constraint. The resulting tree, still referred to as \(\Psi\) , may become non- binary and contains uniquely labeled leaves and possibly some labeled internal nodes. + +The algorithm performs a post- order traversal of the phylogeny \(\Psi\) and, for each node \(\alpha \in V(\Psi)\) , calculates a set \(H_{\alpha}\) describing possible homomorphisms from the subtree \(\Psi_{\alpha}\) to subtrees of \(T\) . At the root node \(\rho\) , the set \(H_{\rho}\) thus describes all homomorphisms from \(\Psi\) to \(T\) . Upon completing the traversal, the algorithm either concludes that no feasible homomorphism exists (when \(H_{\rho} = \emptyset\) ), or initiates a pre- order traversal of \(\Psi\) . During this second traversal, it reconstructs feasible homomorphisms using the information from the sets \(H_{\alpha}\) . + +Formally, let \(\Lambda_{\alpha}\) be the set of labeled nodes in the subtree \(\Psi_{\alpha}\) . A subtree \(T[v, X]\) of \(T\) is termed an induced \(v\) - subtree if it includes the vertex \(v\) , a subset \(X\) of \(v\) 's neighbors, and all vertices that are connected to \(v\) via paths that intersect with \(X\) (Fig. 3). + +For a vertex \(\alpha \in V(\Psi)\) , the set \(H_{\alpha}\) consists of triples \((v, X, C)\) called partial homomorphism tokens or simply tokens. In each token, (i) \(v \in V(T)\) , (ii) \(X \subseteq N_T(v)\) , (iii) \(C\) is a subset of vertices of an induced \(v\) - subtree \(T[v, X]\) such that there exists a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) with \(f(\alpha) = v\) and \(f(\Lambda_{\alpha}) = C\) . + +The algorithm is initialized by setting \(H_{\lambda} = \{(v, \emptyset , \{v\}) : v \in V(T)\}\) for all leafs \(\lambda\) . For an internal node \(\alpha\) , the set \(H_{\alpha}\) is constructed based on the sets from its children nodes \(\beta_1, \ldots , \beta_k\) . The construction utilizes the following lemma: + +\(v \in V(T)\) for all leafs \(\lambda\) . For an internal node \(\alpha\) , the set \(H_{\alpha}\) is constructed based on the sets from its children nodes \(\beta_1, \ldots , \beta_k\) . The construction utilizes the following lemma: + +Lemma 4. Let \(T[v, X]\) be an induced \(v\) - subtree. Then there exist a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) with \(f(\alpha) = v\) and \(f(\Lambda_{\alpha}) = C\) if and only if there exist tokens \((v_1, X_1, C_1) \in H_{\beta_1}, \ldots , (v_k, X_k, C_k) \in H_{\beta_k}\) satisfying the following conditions: + +(a1) \(v \sim v_i\) or \(v = v_i\) for all \(i \in \{1, \ldots , k\}\) ; (b1) \((V(T[v_i, X_i]) \setminus \{v\}) \cap (V(T[v_j, X_j]) \setminus \{v\}) = \emptyset\) for all \(i, j \in \{1, \ldots , k\}\) , \(i \neq j\) ; (c1) \(v \in V(T[v_i, X_i])\) if and only if \(v_i = v\) . (d1) \(v \notin C_i\) , if \(a\) is labeled. (e1) \(X_i = N_T(v_i) \setminus \{v\}\) , if \(v_i \neq v\) . (f1) \(X = \{v_1, \ldots , v_k\} \setminus \{v\}\) ; (g1) \(C = C_1 \cup \dots \cup C_k\) , if \(\alpha\) is unlabeled, and \(C = C_1 \cup \dots \cup C_k \cup \{v\}\) , if \(\alpha\) is labeled. + +Proof. Let us prove the necessity of conditions (a1) - (g1). Suppose that there exists a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) such that \(f(\alpha) = v\) . Define \(T_i\) as the image of the subtree \(\Psi_{\beta_i}\) under \(f\) , denoted by \(f(\Psi_{\beta_i})\) , and let \(v_i = f(\beta_i)\) . Also, let \(X_i = N_T(v_i) \cap V(T_i)\) and \(C_i = f(\Lambda_{\beta_i})\) . + +We aim to demonstrate that \(T_i = T[v_i, X_i]\) . According to Lemma 3, \(T_i\) is connected, which suggests that \(T_i\) must be a subgraph of \(T[v_i, X_i]\) . Furthermore, Lemma 3 also indicates that \(f^{- 1}(T[v_i, X_i])\) is connected. Given the surjectivity of \(f\) , it follows that \(f^{- 1}(T[v_i, X_i]) \subseteq \Psi_{\beta_i}\) , leading to the conclusion that \(T[v_i, X_i] \subseteq T_i\) . + +Consequently, the restriction of \(f\) to \(\Psi_{\beta_i}\) is a surjective homomorphism from \(\Psi_{\beta_i}\) to \(T[v_i, X_i]\) . Thus, the token \((v_i, X_i, C_i)\) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Overview of the Dynamic Programming Algorithm for Detecting Convex Label-Distinctive Homomorphisms. A. A phylogenetic subtree, \(\Psi_{\alpha}\) , rooted at node \(\alpha\) with two children, \(\beta\) and \(\gamma\) . B. Input candidate migration tree \(T\) . The goal is to produce convex homomorphisms from \(\Psi_{\alpha}\) to induced subtrees of \(T\) from such homomorphisms for \(\Psi_{\beta}\) and \(\Psi_{\gamma}\) . C. Convex homomorphisms from \(\Psi_{\beta}\) (top row) and \(\Psi_{\gamma}\) (bottom row) to induced subtrees of \(T\) . For instance, the top figure depicts a homomorphism to an induced 1-tree \(T[1, \{4, 9, 10, 11\}]\) that consists of the vertex 1, its neighbors 4, 9, 10, 11 and all vertices connected to 1 via paths that intersect these neighbors. Nodes of subtrees are colored by their homomorphic images. Homomorphisms are organized into a bipartite graph \(\mathcal{G}\) , where edges connect homomorphism pairs \(f_{1}f_{2}\) and \(f_{1}f_{4}\) that are compatible, and there is no edge between homomorphisms \(f_{1}\) and \(f_{3}\) , that are not compatible. D. Homomorphisms \(f_{5}\) and \(f_{6}\) obtained by combining compatible homomorphisms \(f_{1}, f_{2}\) and \(f_{1}, f_{4}\) .
+ +belongs to \(H_{\beta_{i}}\) . The condition (a1) follows from the homomor- 679 (b1)- (g1). + +\(\alpha\) 78 phism definition, while the convexity of \(f\) yields the conditionsse8 + +Conversely, suppose for each \(i \in \{1, \ldots , k\}\) , we have feasi + +<--- Page Split ---> + +681 ble homomorphisms \(f_{i}:\Psi_{\beta_{i}}\to T[v_{i},X_{i},C_{i}]\) that meet condi- 710 tions (a1) - (e1). We can construct a homomorphism \(f:\Psi_{\alpha}\to 711\) \(T[v,X,C]\) by defining \(f\) as follows: + +\[f(\gamma) = \left\{ \begin{array}{ll} f_{i}(\gamma), & \mathrm{if~}\gamma \in V(\Psi_{\beta_{i}})\\ v, & \mathrm{if~}\gamma = \alpha \end{array} \right. \quad (16)\] + +684 Condition (a1) implies that \(f\) is a homomorphism, condition716 (d1) ensures that labeled nodes map to distinct vertices of \(T_{,717}\) and (e1) guarantees that \(f\) is surjective. + +687 It remains to show that \(f\) is convex. Suppose that \(f(\gamma_{1}) = _{719}\) \(f(\gamma_{2}) = w\) . If both \(\gamma_{1}\) and \(\gamma_{2}\) belong to the same subtree \(\Psi_{\beta_{1},720}\) then the entire path \(P_{\Psi}(\gamma_{1},\gamma_{2})\) maps to \(w\) due to the convexity of \(f_{721}\) \(f_{i}\) . If, on the other hand, \(\gamma_{1}\in \Psi_{\beta_{i}}\) and \(\gamma_{2}\in \Psi_{\beta_{j}}\) , then condition722 (b1) implies that \(w = v\) . Furthermore, by the condition (c1) we \(_{723}\) have \(v_{i} = v_{j} = v\) . Therefore convexity of \(f_{i}\) and \(f_{j}\) , as well as \(_{724}\) the fact that \(f(\alpha) = v\) , imply that \(f(P_{\Psi}(\gamma_{1},\gamma_{2})) = v\) . Together, \(_{725}\) these two facts prove that \(f\) is convex. + +695 To convert Lemma 4 into an algorithm constructing the set 726 of tokens for a node \(\alpha\) using the tokens of its children, we 728 first need to identify tokens of children that satisfy conditions 729 (a1)- (g1). This can be achieved through the following steps. 730 For each vertex \(v\in V(T)\) , we construct a multipartite graph 731 \(\mathcal{G} = \mathcal{G}(\alpha ,v)\) (i.e. a graph partitioned into \(k\) independent sets or 732 parts), as described below: + +702 (i) The parts \(A_{1},\ldots ,A_{k}\) correspond to children of \(\alpha\) 734 (ii) The vertices of the set \(A_{i}\) are tokens from the set \(\Psi_{\beta_{i}}\) satisfying the conditions (a1),(c1), (d1) and (e1). 736 (iii) Two tokens from the sets \(A_{i}\) and \(A_{j}\) are adjacent whenever 737 they satisfy the condition (b1). + +In the constructed graph, sets of partial homomorphism tokens that satisfy Lemma 4 can be identified as \(k\) - vertex cliques. We employ the Bron- Kerbosch algorithm [115] to generate the + +se cliques. For each identified clique, we use conditions (f1) and (g1) to construct a new token \((v,X,C)\) for the node \(\alpha\) + +Additionally, for each token \((v,X,C)\in H_{\alpha}\) , we maintain pointers \(p(v,X,C)\) that link to the children tokens used in its construction. These pointers are used in the subsequent phase of the algorithm, which aims to reconstruct full feasible homomorphisms \(f\) . + +During this phase, the algorithm executes a pre- order traversal of \(\Psi\) . As it progresses, it recursively assigns a specific token to each node. When a token \(t = (v,X,C)\) is assigned to node \(\alpha\) , the algorithm sets \(f(\alpha) = v\) . It then retrieves tokens for \(t\) via the pointers \(p(t)\) and assigns them to children, \(\beta_{1},\ldots ,\beta_{k}\) . This ensures that by the end of the traversal, each node in \(\Psi\) has been assigned a homomorphic image, completing the construction of the homomorphism \(f\) . + +In general, the set \(H_{\rho}\) at the root node \(\rho\) may include multiple tokens, each representing a feasible homomorphism \(f:\Psi \to T\) . When multiple feasible homomorphisms are available, the algorithm selects the one that minimizes violations of the compactness constraint (to be discussed in the next subsection). In case of ties, the selection criterion shifts to minimizing the quantity + +\[D(f) = \sum_{\alpha \in \Lambda (T)}\delta (\alpha)\cdot d(f(\alpha)), \quad (17)\] + +where \(\delta (\alpha)\) is the number of labeled children of a node \(\alpha\) . This approach prioritizes homomorphisms that map high- degree nodes to high- degree vertices. + +The outlined method is formalized in Algorithm 2. Its efficiency can be improved by contracting sibling leaves in both \(\Psi\) and \(T\) into a single node (vertex). The algorithm maintains a count of copies of contracted nodes used by tokens, and a new token is generated from children tokens only if the total count of each leaf in the children tokens does not exceed its overall count. This modification markedly enhances the dynamic programming algorithm's runtime. The adjustments to Lemma + +<--- Page Split ---> + +# Algorithm 2 Convex homomorphism minimizing compactness violations + +Input: phylogenetic tree \(\Psi\) with the root \(\rho\) and a candidate migration tree \(T\) Output: a feasible homomorphism \(f:\Psi \to T\) or the answer that it does not exist. 1: modify \(\Psi\) by contracting paths between leafs with the same label. 2: perform a post- order traversal of \(\Psi\) 3: for every node \(\alpha\) of the post- order do 4: construct a set of partial homomorphism tokens \(H_{\alpha}\) 5: if \(\alpha\) is a leaf then 6: \(H_{\alpha}\leftarrow \{(v,0,\{v\}):v\in V(T)\}\) 7: end if 8: if \(\alpha\) is an internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 9: Let \(H(\beta_{j}) = \{(v_{i}^{j},X_{i}^{j},C_{i}^{j}):i = 1,\ldots ,l_{j}\}\) , \(j = 1,\ldots ,k\) 10: for \(v\in V(T)\) do 11: construct the multipartite graph \(\mathcal{G}(\alpha ,v)\) as described in (i)- (iii); 12: Generate the set \(K\) of \(k\) - vertex cliques of \(\mathcal{G}(\alpha ,v)\) using Bron- Kerbosch algorithm. 13: for each clique \(\{(u_{i_{1}}^{l},X_{i_{1}}^{l},C_{i_{1}}^{l}),\ldots ,(u_{i_{k}}^{l},X_{i_{k}}^{l},C_{i_{k}}^{l})\} \in K\) do 14: Construct the sets \(X\) and \(C\) using formulas (f1) and (g1). 15: Set \(H_{\alpha}\leftarrow H_{\alpha}\cup \{(v,X,C)\}\) and \(p(v,X,C)\leftarrow p(v,X,C)\cup \{(i_{1},\ldots ,i_{k})\}\) 16: end for 17: end for 18: end if 19: end for 20: if \(H_{\rho}\neq \emptyset\) then 21: perform a pre- order traversal of \(\Psi\) 22: for each token \(t_{1},\ldots ,t_{R}\in H_{\rho}\) do 23: assign the token \(t_{r}\) to \(\rho :AS_{\rho}\leftarrow t_{r}\) 24: for every node \(\alpha\) of the pre- order do 25: \(f_{r}(\alpha)\leftarrow v\) , where \((v,X,C) = AS_{\alpha}\) 26: if \(\alpha\) is an internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 27: Let \(H(\beta_{j}) = \{(v_{i}^{j},X_{i}^{j},C_{i}^{j}):i = 1,\ldots ,l_{j}\}\) , \(j = 1,\ldots ,k\) and \(p(v,X,C) = (i_{1},\ldots ,i_{k})\) 28: \(AS_{\beta_{j}}\leftarrow (v_{i_{j}}^{j},X_{i_{j}}^{j},C_{i_{j}}^{j})\) , \(j = 1,\ldots ,k\) 29: end if 30: end for 31: end for 32: among generated homomorphisms \(f_{1},\ldots ,f_{R}\) , output the homomorphism \(f_{r}\) with the minimal number of compactness violations and, in case of ties, with the minimal \(D(f_{r})\) 33: else 34: \(f\) does not exist 35: end if + +743 4 and Algorithm 3 in Supplementary Material are straightforward- 754 ward, but involve numerous minor technical details; hence a 755 formal description is omitted. One particular detail, however, 756 should be mentioned: if \(\Psi^{\prime}\) and \(T^{\prime}\) represent the leaf- contracted 757 versions of \(\Psi\) and \(T\) , respectively, and \(f^{\prime}:\Psi^{\prime}\to T^{\prime}\) is a fea- 758 sible homomorphism, there may be cases where \(f(\alpha) = l\) for 759 an internal node \(\alpha\) of \(\Psi\) and a leaf \(l\) in \(T^{\prime}\) that results from the 760 contraction of leaves \(l_{1}\) and \(l_{2}\) . In such instances, \(f^{\prime}\) can be ex- 761 tended to a homomorphism \(f:\Psi \to T\) by designating \(f(\alpha)\) as 762 either \(l_{1}\) or \(l_{2}\) . To resolve this ambiguity, the leaf corresponding 763 to the population with higher diversity is chosen. + +Finally, it should be noted that, strictly speaking, Algorithm 2 is not polynomial, since the number of tokens for a node of \(\Psi\) theoretically can be exponential. In practical settings, however, the algorithm is extremely fast, and require split seconds to finish. + +### 2.6. Migration inference with convexity and compactness constraints + +A similar approach to the one outlined in Subsection 2.5 can be employed to identify homomorphisms that are both convex and compact. However, the dynamic programming algorithm can be further optimized by taking advantage of the specific nature + +<--- Page Split ---> + +765 of these constraints. + +766 As with the earlier approach, homomorphisms that are both 776 convex and compact will be referred to as feasible. Following 778 a similar methodology to that used in Algorithm 2, we begin 779 by contracting the paths in the phylogeny \(\Psi\) . We then construct 800 partial homomorphism tokens similar to those used previously, 801 with the exception that subsets \(C\) of images of labeled nodes 802 are not required. Thus, the tokens are simplified to pairs \((v,X)\) 803 where \(v\in V(T)\) and \(X\subseteq N_{T}(v)\) . A pair \((v,X)\) is included in \(H_{\alpha}\) 804 if there exists a feasible surjective homomorphism \(f:\Psi_{\alpha}\to\) 805 \(T[v,X]\) such that \(f(\alpha) = v\) , and it also satisfies the following 806 condition: + +\[|\Lambda_{\alpha}| = |T[v,X]|. \quad (18)\] + +777 This condition is necessary for a partial homomorphism to be 807 extendable to a full compact homomorphism \(\Psi \to T\) + +779 The algorithm is initialized by setting \(H_{\lambda} = \{(v,\emptyset):v\in_{809}\) \(V(T))\) for leafs \(\lambda\) . For an internal node \(\alpha \in V(\Psi)\) , its token set 810 \(H_{\alpha}\) is constructed from the tokens of its children \(\beta_{1},\ldots ,\beta_{k}\) using 811 Lemma 5. + +783 Lemma 5. \((v,X)\in H_{\alpha}\) if and only if one of the following con- 813 ditions hold: + +785 1) \(\alpha\) is not labeled and there exist \(w\in X\) such that \((v,X\backslash\) \(\{w\} \in H_{\beta_{1}}\) and \((w,N_{T}(w)\backslash \{v\})\in H_{\beta_{2}}\) 816 2) \(\alpha\) is labeled, \(|X| = k\) , and there exist a permutation \((v_{1},\ldots ,v_{k})\) \(T[v,X\backslash \{w\} ]\) and \(f_{2}:\Psi_{\beta_{2}}\to T[w,N_{T}(w)\backslash \{v\} ]\) . By defining \(f\) as of elements of \(X\) such that \(v\sim v_{1},\ldots ,v_{k}\) , and \((v_{1},N_{T}(v_{1})\backslash \{v\})\) \(\{v\} \in H_{\beta_{1}},\ldots ,(v_{k},N_{T}(v_{k})\backslash \{v\})\in H_{\beta_{k}}\) + +790 Proof. We present the proof for the case where \(\alpha\) is unlabeled; the argument for labeled \(\alpha\) follows a similar rationale. In this case, \(\alpha\) was not involved in path contraction, and thus \(k = 2\) + +793 Suppose that \((v,X)\in H_{\alpha}\) , i.e. \(|\Lambda_{\alpha}| = |T[v,X]|\) and there exists a feasible surjective homomorphism \(f:\Psi_{\alpha}\to T[v,X]_{\mathrm{in}}\) such that \(f(\alpha) = v\) . Consequently, \(f(\Lambda_{\alpha}) = V(T[v,X])\) , and 82 + +therefore there must be a labeled node \(\gamma \in V(\Psi_{\beta_{1}})\) such that \(f(\gamma) = v\) . Given the connectivity constraint, this implies that \(f(\beta_{1}) = v\) . Meanwhile, the compactness constraint necessitates \(f(\beta_{2}) = w\neq v\) . + +Following Lemma 3 and considering the connectivity constraint, it can be shown that: + +\[f(\Psi_{\beta_1}) = T[v,X\backslash \{v\} ]\mathrm{and}f(\Psi_{\beta_2}) = T[w,N(w)\backslash \{v\} ] \quad (19)\] + +Let us now establish the second equality; the method for proving the first is analogous. Let \(T_{2} = f(\Psi_{\beta_{2}})\) . We know that \(v\notin V(T_{2})\) and, according to 3, \(T_{2}\) is connected. These observations imply that \(T_{2}\subseteq T[w,N(w)\backslash \{v\} ]\) . Conversely, Lemma 3 suggests that \(f^{- 1}(T[w,N(w)\backslash \{v\} ])\) is connected. Given the surjectivity of \(f\) , this implies \(f^{- 1}(T[w,N(w)\backslash \{v\} ])\subseteq \Psi_{\beta_{2}}\) , leading to \(T[w,N(w)\backslash \{v\} ]\subseteq T_{2}\) . Thus, both \(T_{2}\subseteq T[w,N(w)\backslash \{v\} ]\) and \(T[w,N(w)\backslash \{v\} ]\subseteq T_{2}\) are true, confirming the second equality. + +So, the restrictions \(f|_{\Psi_{\beta_1}}\) and \(f|_{\Psi_{\beta_2}}\) are both feasible surjective homomorphisms. Additionally, \(\Lambda_{\alpha} = \Lambda_{\beta_{1}}\cup \Lambda_{\beta_{2}}\) , \(f(\Lambda_{\beta_{1}}) = f(\Psi_{\beta_{1}}) = T[v,X\backslash \{w\} ]\) and \(f(\Lambda_{\beta_{2}}) = f(\Psi_{\beta_{2}}) = T[w,N_{T}(w)\backslash \{v\} ]\) , thus confirming that the equality (18) holds for the tokens \((v,X\backslash \{w\})\) and \((w,N_{T}(w)\backslash \{v\} ]\) . This proves the necessity of condition 1). + +To demonstrate the sufficiency of condition 1), we assume that there exist feasible surjective homomorphisms \(f_{1}:\Psi_{\beta_{1}}\to\) \(T[v,X\backslash \{w\} ]\) and \(f_{2}:\Psi_{\beta_{2}}\to T[w,N_{T}(w)\backslash \{v\} ]\) . By defining \(f\) as follows, we can establish a combined feasible homomorphism: + +\[f(\chi) = \left\{ \begin{array}{ll}f_{1}(\chi) & \mathrm{if}\chi \in V(\Psi_{\beta_{1}})\\ f_{2}(\chi) & \mathrm{if}\chi \in V(\Psi_{\beta_{2}})\\ \nu & \mathrm{if}\chi = \alpha \end{array} \right. \quad (20)\] + +Lemma 5 can be directly applied to construct tokens for an unlabeled node \(\alpha\) using the tokens from its children. For + +<--- Page Split ---> + +a labeled node \(\alpha\) , however, the process of finding a permuta- 854 tion \((v_{1},\ldots ,v_{k})\) of the tokens is more complex. The method to achieve this is detailed in the following approach. + +Lemma 5 can be straightforwardly use to construct tokens of \(\alpha\) from tokens of its children, if \(\alpha\) is unlabeled. When \(\alpha_{859}\) is labeled, then finding a permutation \((v_{1},\ldots ,v_{k})\) required by Lemma 5 is more complicated and can be achieved using the approach described next. + +Suppose that \(\mathcal{S} = (S_{1},\ldots ,S_{k})\) is a collection of sets. A vector \((x_{1},\ldots ,x_{k})\) is termed a transversal of \(\mathcal{S}\) [116] if \(x_{i}\in S_{i}\) and all \(x_{i}\) are distinct. Let now \(S_{i} = \{u:(u,Y)\} \in H_{\beta_{i}}\) and \(v\sim\) \(u]\) . Then the vector \((v_{1},\ldots ,v_{k})\) satisfies the condition 2) of Lemma 5 if and only if it is a transversal of \(S\) + +Given this, the set \(H_{\alpha}\) can be obtained by generating all transversals of \(S\) . This process involves the following steps: + +Construct a bipartite graph \(\mathcal{B}(S)\) with parts \(\mathcal{I}\) and \(\mathcal{I}\) where \(\mathcal{I} = 1,\ldots ,k\) represents the set indices, and \(\mathcal{I} = \bigcup_{i = 1}^{k}S_{i}\) , represents all elements in the sets. In this graph, a vertex \(i\in \mathcal{I}\) is adjacent to a vertex \(j\in \mathcal{I}\) if \(j\) belongs to \(S_{i}\) + +In \(\mathcal{B}\) , each transversal corresponds to a maximal matching of size \(k\) . To generate these matchings, construct a line graph \(L(\mathcal{B})\) , where each vertex represents an edge of \(\mathcal{B}\) , and two vertices are adjacent if their corresponding edges in \(\mathcal{B}\) share a common vertex. Maximal matchings of \(\mathcal{B}\) correspond to maximal independent sets in \(L(\mathcal{B})\) , that can be produced using the Bron- Kerbosch algorithm [115]. + +The entire method is detailed in Supplementary Material, Algorithm 3. Like Algorithm 2, efficiency can be significantly improved by contracting sibling leaves in both \(\Psi\) and \(T\) . + +### 2.7. SMiTH: Sampling Migration Trees via Homomorphisms + +The algorithms discussed can be effectively integrated into an Unlabeled Migration Sampling framework (Problem 5). It allows for the identification of homeomorphic images of a given phylogeny \(\Psi\) within a collection of candidate migration trees sampled from a given migration pattern represented by a specified tree distribution. + +The obtained sample of migration trees can be directly analyzed to estimate the probabilities of specific migration routes or to obtain summary statistics and confidence intervals for derivative evolutionary parameters. It can be also synthesized into a single weighted consensus graph, where each edge is weighted by the number of candidate trees that support it. When the homomorphism reconstruction includes an objective function, the consensus graph is constructed from a subsample comprising the top \(\kappa \%\) of trees ranked by their objective values. For applications requiring a specific output tree – such as for benchmarking and comparison with other methods described in Subsection 3.1 – the tree is determined by calculating the maximum- weight spanning tree of the consensus graph. The entire algorithmic pipeline, named SMiTH (Sampling Migration Trees via Homomorphisms), is illustrated in Figure 1. + +## 3. Results + +### 3.1. Simulated data + +To generate synthetic data, we used FAVITES [117], a tool capable of simulating genomes, phylogenies, and migration networks under various evolutionary scenarios. Although originally designed to simulate viral outbreaks, FAVITES supports general phylogenetic and population genetics models, making it suitable for simulating migrations of heterogeneous populations besides viruses. It also should be noted that, to the best of our knowledge, specialized simulation tools for metastatic + +<--- Page Split ---> + +spread with capabilities comparable to FAVITES are currently not available. 920 + +We simulated the migration of a heterogeneous population over a network of sites formed according to the Barabasi- Albert model [70]. This assumption can be valid for both viral [98, 118] and cancer [119, 95] spread. Migrations occur at a constant rate along each network edge (in viral context, this corresponds to the network- based Susceptible- Infected (SI) transmission model). Within each site, phylogenies evolved under the exponential coalescent, a model previously used to simulate intra- host [66] and intra- tumor [120] evolution. Genotypes were assumed to evolve under the GTR+ \(\Gamma\) substitution model, and were sampled simultaneously at the end of the simulation. In total, 275 simulated datasets were generated, encompassing 5- 30 demes with 100 sequences sampled per deme. + +In the first series of experiments, we sampled candidate migration trees from 3 distinct prior distributions of tree topologies and evaluated their compatibility with simulated phylogenies under three different types of constraints. The prior distributions included: + +T1) Degenerate distribution consisting of the true topology. Even though this scenario is unrealistic for actual migration inference, it serves as a test to determine if migration links can be accurately reconstructed when the topology is known but sites need to be correctly mapped to migration network vertices. To avoid bias linked to correlations between vertex IDs and migration times, that can be potentially introduced by the simulation methods, we produced multiple samples with randomly permuted vertex IDs. + +T2) Random scale-free trees produced by the preferential attachment procedure. + +T3) Uniformly distributed trees of a given size. Sampling was + +performed by generating random Prufer codes [121], integer sequences of length \(n - 2\) that uniquely define \(n\) - vertex trees. + +The following types of constraints were used: + +H1) unconstrained homomorphism; H2) convex homomorphism minimizing the number of compactness constraint violations; H3) convex and compact homomorphism. + +For each simulated dataset, we produced 9 samples of candidate migration trees corresponding to all combinations of conditions T1)- T3) and H1)- H3). The sample sizes ranged from 1,000 for the degenerate distribution without constraints, up to 1,000,000 for the uniform distribution with convexity and compactness constraints. This variation in sample size was necessary because stricter constraints require larger samples to ensure that a sufficient number of feasible trees are produced. We assessed the compatibility of these sampled trees with the given phylogenies under the respective constraints, using methods detailed in Subsections 2.4- 2.6. Using these assessments, subsamples of compatible tree were extracted. + +Sampled trees were compared with true migration trees produced by FAVITES. Individual trees were compared by measuring recall, defined as the fraction of inferred transmission edges among true transmission edges; precision, the fraction of true transmission edges among inferred transmission edges; and the \(f\) - score, i.e., the harmonic mean of precision and recall. + +Additionally, we summarized each subsample of compatible migration trees using a consensus graph, where each edge is weighted by the proportion of candidate trees that support that edge [26, 34, 122]. A solution can be extracted from the consensus graph by discarding edges with support below a predefined threshold. For each graph, we estimated the area under + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. (a)-(c) Percent of sampled trees that are compatible with given phylogenies. (d-f) Area under the precision-recall curve (AUC) calculated by varying the support threshold.
+ +the precision- recall curve, which was calculated by varying the support threshold. We opted for the precision- recall curve in- 969 stead of the more common ROC curve due to the imbalance between the classes of true and false migration edges. + +We found that the relationship between phylogenetic trees and migration trees is heavily influenced by structural constraints. In the absence of constraints, there is considerable ambiguity in the possible migration histories that align with a given phylogenetic tree topology, even when prior knowledge about true migration tree topology is available. Notably, almost every sampled scale- free network proved compatible with the given phylogenies, with a median fraction of compatible trees at \(\mu = 1.79\) (Fig. 4b). It is not entirely unexpected in light of Theorem 1, which suggests that trees with a low diameter - a common feature of scale- free networks - are more likely to be compatible with a given phylogenetic tree. However, even among uniformly sampled trees, a high compatibility rate was observed + +when no constraints are applied (median \(\mu = 0.84\) , Fig. 4c). + +Furthermore, without constraints individual compatible trees display only marginal agreement with true migration trees. This holds not only for scale- free and uniformly sampled trees, but even for the degenerate distribution, with median \(f\) - score within the range 0.33- 0.36 for all three tree priors (Supplementary Material, Fig. 8). In other words, even if the topology of a true migration tree is known, numerous labelings of that topology are compatible with the original phylogeny, most of which substantially diverge from the true labeling. Consequently, the true labeling is not immediately distinguishable among the alternatives without additional information. Combining a compatible tree subsample into a consensus graph, however, brings it closer to the true migration tree, with the median AUC values ranging from 0.56 to 0.59 for all three tree priors (see Fig. 4). + +The introduction of convexity and compactness constraints significantly reduces the percentage of trees deemed compati + +<--- Page Split ---> + +ble, as can be expected (Fig. 4abc). This reduction primarily eliminates incidental solutions, enhancing the alignment of the trees that meet these constraints with the true migration trees (Fig. 4def). In particular, for the degenerate distribution, introducing constraints effectively filters out ambiguous vertex mappings, nearly always recovering the true mapping, provided that the solutions satisfying the constraints exist. + +Interestingly, without constraints, the use of tree priors does not enhance accuracy, as demonstrated by the lack of significant differences in AUC distributions among the tree priors \((p = 0.083\) , Kruskal- Wallis test). In contrast, when constraints are applied, prior knowledge of the migration tree structure comes beneficial, with AUCs improving as the tree prior comes tighter. In particular, under constraints, AUCs for the scale- free prior are significantly higher than those for the form prior \((p = 4.4 \cdot 10^{- 6}\) and \(p = 1.76 \cdot 10^{- 16}\) for convex and both convex and compact cases, respectively, Kruskal- Wallis test). Given that true migration trees are generated by the preferential attachment, this suggests that under constraints, phylogeny to a certain degree reflects the properties of the underlying migration network. + +Taken together, these observations indicate that phylogeny topologies do indeed reflect underlying migration tree structures, but the extent of this reflection is influenced by evolutionary constraints. Moreover, the correspondence between geneins and migration trees is primarily discernible when analyzed statistically across a large sample of feasible migration trees that are compatible with the phylogeny. A single compatible tree may be arbitrary, and thus relying on a single solution may lead to misleading conclusions. + +Based on those observations, we have developed a method named SMiTH (Sampling MIGration Trees using Homomorphisms) for the constrained inference of migration trees with expected general properties. This method involves sampling + +candidate migration trees from a designated random tree distribution, identifying convex homomorphisms from the given phylogeny to these sampled trees while minimizing an objective function defined by the number of compactness constraint violations, constructing a consensus graph from trees with top objective values, and ultimately inferring the final migration tree as the minimal spanning tree of this consensus graph. + +We benchmarked SMiTH against several existing tools designed to infer migration networks from phylogenetic tree topologies. For the sake of fairness, tools that use dated phylogenies and/or case- specific epidemiological information were not considered. The tools selected for this comparison include Casispeia [21, 59], MACHINA [22], Phyloscanner [31], STraTUS [36], and TNet [44]. For MACHINA, we ran all four migration models provided by the tool and report the best result, which was achieved using the single- source seeding model. STraTUS generates a sample of migration trees rather than a single tree; thus, similarly to SMiTH, we used the minimum spanning tree of the consensus graph for benchmarking purposes. + +The results of the algorithm comparison are shown in Fig. 5. It was found that both variants of SMiTH – with uniform and scale- free tree priors – allow for a statistically significant improvement over other tools \((p < 10^{- 9}\) , multiple comparison of \(f\) - score distributions by Kruskal- Wallis test). SMiTH is followed by Cassiopeia and STraTUS – two other sampling- based methods, whose accuracies were statistically indiscernible \((p = 0.58\) , Kruskal- Wallis test). These tools indeed both produce samples of convex solutions, albeit using different algorithms. While STraTUS is doing it directly, while Cassiopeia’s module FitchCount samples most parsimonious solutions that in our examples were almost always convex. These tools were followed by MACHINA that, similarly to our approach, imposes structural constraints on plausible migration trees by considering them as subgraphs of so- called transition patterns. However, + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. (a) Summary statistics of methods performance on all simulated datasets. (b) Median \(f\) -scores of different methods on simulated datasets with varying numbers of populations. (c) Median running times of SMiTH for construction of a constrained homomorphism for a given phylogenetic tree and candidate migration tree as a function of number of migration sites
+ +MACHINA produces a single solution optimizing particular \(\mathrm{ob}_{1065}\) 1054jectives rather than summarizes a sample of such solutions; this1066 in light of the observations described above, likely hampered its1067 performance vis- à- vis sampling- based methods. Similar rea1068 soning can be applied to Phyloscanner, that also produces a1069 single most parsimonious solution. In addition, Phyloscanner1070 is specifically designed to make use of paraphyly, that usually1071 provides a strong signal for migration [66] when present; con1072 sequently, its accuracy can be affected when, as in analyzed1073 test cases, the number of paraphyletic clades is limited. Fur1074 thermore, Phyloscanner usually assumes that when the popula1075 tions sampled from different cites are monophyletic, then they1076 + +all have a common source [66] - the assumption that is opposite to compactness and that seems to be not always valid. The relation between algorithms' performances is mostly stable with regard to the number of migration sites (Fig. 5b). + +The median running time of our method for construction of a constrained homomorphism for a given phylogenetic tree and candidate migration tree was within 0.7 seconds for all tests and tree priors (Fig. 5c), even though theoretically our algorithms can be exponential in the worst case. It allows us, using straightforward parallelization, to produce and process samples consisting in hundreds of thousands of candidate trees in reasonable time. + +<--- Page Split ---> + +### 3.2. Experimental viral data + +Analysis of simulated data highlights the role of structural con- 1112 strains in migration tree inference, particularly when paraphyly113 is limited. Interestingly, these constraints prove just as essential114 in scenarios with a high degree of paraphyly, albeit for different115 reasons. This is evidenced by the analysis of data of Hepatitis116 C (HCV) outbreaks, which have been considered in previous117 studies [38, 39, 44, 27]. The data comprises intra- host HCV118 populations from several outbreaks investigated by the Centers119 for Disease Control and Prevention, each population consisting120 of sequences covering Hypervariable Region 1 (HVR1) of the121 HCV genome. In each outbreak, a single primary host infected122 all other hosts, rendering the migration tree in graph- theoretical123 terms a star. + +We analyzed two largest outbreaks involving 15 and 19 in125 fected hosts. Phylogenetic trees for each outbreak were con126 structed using RAxML [123]. All transmissions occurred with127 in a short time frame and, as a result, intra- host populations are128 highly intermixed (Fig. 6a). This makes paraphylytic signal129 strong, but oversaturated, thus impeding its use to reconstruct true transmission history. + +This effect can be demonstrated by examining internal node labels generated by Fitch algorithm, that serves as a basis for several methods considered in the previous section. In the trees analyzed, \(32 - 34\%\) of internal nodes were assigned a single provisional label during the post- order traversal step of the dynamic programming algorithm, indicating that these labels appear in all most parsimonious solutions (or solutions with the minimal migration number in terms of [22]). Many of these nodes are adjacent, suggesting that the transmission links they represent will be identified by any parsimony- based sampling approach similar to those employed by existing tools [44, 12, 21]. For one outbreak, these links form a connected graph, whereas in the other, only one host does not integrate into this + +single connected component (see Fig. 6b). These resulting graphs are relatively dense and include not only true edges but also a significant number of false positives (see Fig. 6b). Consequently, even if all true positive edges are correctly identified using nodes with multiple Fitch labels, the \(f\) - scores would not exceed 0.54 and 0.51, respectively. Enhancing the accuracy of this approach requires the filtering out of false positive edges, achievable only through the integration of additional prior information or constraints. + +In contrast, sampling unconstrained candidate transmission trees from the scale- free tree distribution produce the results that are significantly closer to true transmission histories (Fig. 6c). For consensus networks derived from these samples, areas under precision- recall curve are estimated at 0.76 and 0.70. Furthermore, \(f\) - scores of solutions obtained as minimal spanning trees of consensus networks produced from top \(1\%\) of sampled trees according to the objective (10) are 0.86 and 0.94. Comparable results - \(f = 0.79\) and \(f = 0.89\) - are obtained if we use top \(1\%\) of sampled trees based on the parsimony score. + +### 3.3. Experimental cancer data + +We employed SMiTH to analyze the migration history of metastatic ovarian cancer using the data published in [124]. The dataset comprised whole- genome and targeted sequencing data from samples collected at various anatomical sites, including the left ovary (LOv), the right ovary (ROv), and several metastases. In the original study, migration networks were inferred using hierarchical clustering trees and a Dollo parsimony model. Subsequent re- analysis using MACHINA [22] revealed several additional, more parsimonious migration histories. + +We focused on the data from Patients 1, 3, and 7, that included the highest number of anatomical sites (7- 8 sites) and that were thoroughly analyzed in [22]. We used clone trees shared by the authors of [22]. For each patient, we sampled candidate migration trees from a uniform distribution using an + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. (a) Phylogenetic trees of HCV variants from two outbreaks. Variants sampled from different hosts are highlighted in different colors. (b) Graphs formed by edges corresponding to adjacent tree nodes with unique Fitch labels. True edges are highlighted in red. (c) Consensus networks of top \(1\%\) of sampled trees with respect to the objective (10). Edge thicknesses are proportional to their frequencies, true edges are highlighted in red.
+ +unconstrained model. Following the methodology described in [159] [22], we resolved polytomies in clone trees to match the solution reported there. In instances where the resolution of polytomies was ambiguous, we applied a random resolution, generating a new random resolution for each sampled candidate migration tree. + +For Patient 1, [124] identified a complex migration history, designating ROV as the primary tumor site. MACHINA was able to find several more parsimonious histories that suggested either LOV or ROV as the primary tumor location, but was not able to distinguish between them. These histories shared parsimony scores (referred to as "migration numbers" [22]) and the same number of migration events (named "co- migration numbers" [22]), which left the primary tumor's location am + +biguous. In contrast, SMITH enabled the comparison of migration number distributions for different potential primary tumor sources (Fig. 7) rather than making the decision based on single most parsimonious solutions. Migration numbers associated with LOV and ROV were significantly lower than those of other potential sources ( \(p < 10^{- 65}\) , Mann- Whitney U test). Among these two, the lowest numbers were observed for ROV ( \(p < 10^{- 38}\) , Mann- Whitney U test), indicating a stronger statistical support for ROV as the primary tumor source. + +A similar situation was observed for Patient 7. Here, [124] suggested the right uterosacral ligament (RUt) as the primary tumor location, while MACHINA identified several alternative migration histories with either the left ovary (LOv) or the right ovary (ROv) as the source, each sharing identical migration and + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7. Distributions of migration numbers for trees with different primary tumor sites.
+ +co- migration numbers. Based on these results, [22] argued that available data provides no evidence for the assertion that the primary tumor is located in the RUT as opposed to the ovaries. However, SMiTH provided such statistical evidence (Fig. 7) showing that migration trees with RUT as the source generally exhibited lower migration numbers \((p < 10^{- 154}\) , Mann- Whitney U test). + +Patient 3 presents a different scenario. Here, MACHINA identified several migration histories with LOV or ROV as primary tumor sources. There is also an alternative history with the omentum (Om) as the source, which has a lower migration number. The latter hypothesis appears preferable if judged solely by the single most parsimonious solution. Yet, this con + +clusion is not reliable due to the highly symmetric distribution of clones from different sites in the clone tree. Many clones form polytomies, offering insufficient data to clearly differentiate between the corresponding sites (e.g., all clones from LOV and LFTC are siblings, rendering these sites indistinguishable; see Supplementary Material, Fig. 9). The apparently lower migration number for Om, compared to other sites, is simply due to its representation by five clones, versus four for several other sites – a difference that could stem from sampling bias given the small number of clones involved. + +SMiTH was able to capture and quantify this uncertainty. Specifically, it did not find significant differences in the distribution of migration numbers among potential primary sources, + +<--- Page Split ---> + +with \(p\) - values ranging from 0.09 to 0.96 in pairwise Mann- 1232 Whitney U tests and a \(p = 0.65\) in a joint Kruskal- Wallis1233 test (Fig. 7). This provides a statistical support for suggestion234 that the existing data is insufficient to draw reliable conclusions235 about the migration history. + +In total, these examples illustrate how SMiTH can be used237 to provide statistical support for hypotheses regarding metat1236 static spread pathways. + +## 4. Discussion + +This study is dedicated to in- depth mathematical exploration of the relationships between phylogenies and migration trees of heterogeneous genomic populations. Although it is established that phylogenetic trees impose some restrictions on migration pathways [36], the exact nature and extent of these constraints are still not well understood, despite a considerable amount of research dedicated to this problem [73, 72, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 36]. Our approach adds both depth and rigor to this area by utilizing the powerful theoretical and algorithmic framework of theory of graph homomorphisms. This framework allowed us to derive necessary and sufficient conditions for the compatibility of phylogenetic and migration trees, and to develop efficient algorithms for analyzing this compatibility through numerical experiments. + +Based on our findings, we propose a general and flexible computational framework that can be used to infer migration networks under various assumptions, quantitatively assess competing hypotheses about migration dynamics, investigate the influence of phylogenies on the migration tree space, and to determine whether a potential migration history is definitively contradicted by a phylogeny or set of phylogenies. + +Methodologically, our approach balances the advantages of probabilistic and parsimony methods. It incorporates scalability and the use of advanced combinatorial optimization techniques + +from the latter, along with the biological plausibility of the former that comes from employing appropriate prior random tree distributions. + +This study aligns well with the context of previous research. Several earlier studies have conceptualized migration inference as a coloring problem [35, 34, 36], employing this framework both to develop efficient inference algorithms and to explore the structure of migration tree space. The methodology introduced in this paper advances these prior approaches by incorporating a more comprehensive mathematical model and applying more sophisticated mathematical techniques. Similarly, the concept of constraining the tree space to random trees from a specified distribution was first introduced in our earlier studies on viral transmissions [38, 27], while a related concept of restricting the tree space to subgraphs of specific migration patterns has been utilized in computational cancer genomics [22, 61]. This paper not only refines and extends these methodologies but also integrates them into a cohesive modeling and computational framework. + +The proposed approach certainly has both advantages and disadvantages. We recognize that uniform sampling from the space of candidate migration trees may not be the most optimal tool for migration dynamics inference, and employing more precise optimization or sampling techniques to navigate the tree space could substantially enhance the method's accuracy and efficiency. This work lays a foundation for further theoretical and algorithmic development in this direction, equipping researchers with the tools needed to expand the use of graph homomorphism methodologies. On the other hand, uniform sampling may be more suitable for hypothesis testing and comparison. It also should be noted that a key aspect of our methodology is that it involves sampling from the space of subtrees of a transition pattern, rather than from the space of phylogeny node labelings as was done in other studies [36, 44, 12]. + +<--- Page Split ---> + +The former space is considerably smaller in practical scenar- ies which often makes the uniform sampling computationally- ies feasible. Furthermore, it is likely that migration models tailored to the specific characteristics of underlying populations could potentially yield more accurate insights into the biological processes involved. In particular, the biological mechanisms driving viral and cancer migrations are certainly very different. Nonetheless, employing a unified phylogenetic approach to study highly mutable populations offers several advantages. First, it decouples the initial phylogenetic reconstruction from its biological interpretation, thereby minimizing the risk of overfitting and ensuring that the results are less biased by underlying models. Additionally, such methods are significantly more computationally efficient and scalable compared to parameter-rich models. Finally, more general phylogenetic models offer greater flexibility and versatility, and usually can be readily extended to more specific settings through the integration of suitable priors. These features make them an excellent foundation for more detailed analyses. + +## 5. Acknowledgements + +K.K. has been supported by a Georgia State University Molecular Basis of Disease fellowship. P.S. has been supported by the NSF grants 2047828 and 2212508. The authors thank the organizers of the Computational Genomics Summer Institute (UCLA, July 12 - August 4, 2023 and July 10 - August 2024), where several ideas of this paper were conceived. + +## 6. Code availability + +The code developed and used in this study is available at https://github.com/compbel/SMiTH. + +## References + +[1] E. Domingo, J. Sheldon, C. Perales, Viral quasispecies evolution, Microbiology and Molecular Biology Reviews 76 (2) (2012) 159- 216. [2] R. A. Burrell, N. McGranahan, J. Bartek, C. Swanton, The causes and consequences of genetic heterogeneity in cancer evolution, Nature 501 (7467) (2013) 338- 345. [3] E. C. Smith, The not- so- infinite malleability of rna viruses: Viral and cellular determinants of rna virus mutation rates, PLoS pathogens 13 (4) (2017) e1006254. [4] S. Duffy, Why are rna virus mutation rates so damn high?, PLoS biology 16 (8) (2018) e3000003. [5] R. Sanjuan, P. Domingo- Calap, Mechanisms of viral mutation, Cellular and molecular life sciences 73 (2016) 4433- 4448. [6] I. P. Tomlinson, M. Novelli, W. Bodmer, The mutation rate and cancer, Proceedings of the National Academy of Sciences 93 (25) (1996) 14800- 14803. [7] M. Greaves, Nothing in cancer makes sense except..., BMC biology 16 (1) (2018) 1- 8. [8] W. M. Grady, Genomic instability and colon cancer, Cancer and metastasis reviews 23 (1- 2) (2004) 11- 27. [9] G. S. Charames, B. Bapat, Genomic instability and cancer, Current molecular medicine 3 (7) (2003) 589- 596. [10] F. Utro, C. Levovitz, K. Hrissorrakrai, L. Parida, A common methodological phylogenomics framework for intra- patient heteroplasms in infer sars- cov- 2 sublineages and tumor clones, BMC genomics 22 (5) (2021) 1- 13. [11] T. Stadler, O. G. Pybus, M. P. Stumpf, Phylodynamics for cell biologists, Science 371 (6526) (2021) eah6266. [12] P. Sashittal, M. El- Kebir, Sampling and summarizing transmission trees with multi- strain infections, Bioinformatics 36 (Supplement_1) (2020) i362- i370. [13] J. Z. Sanborn, J. Chung, E. Purdom, N. J. Wang, H. Kakavand, J. S. Wilmott, T. Butler, J. F. Thompson, G. J. Mann, L. E. Haydu, et al., Phylogenetic analyses of melanoma reveal complex patterns of metastatic dissemination, Proceedings of the National Academy of Sciences 112 (35) (2015) 10995- 11000. [14] D. X. Nguyen, J. Massague, Genetic determinants of cancer metastasis, Nature Reviews Genetics 8 (5) (2007) 341- 352. [15] N. D. Grubaugh, J. T. Ladner, P. Lemey, O. G. Pybus, A. Rambaut, E. C. Holmes, K. G. Andersen, Tracking virus outbreaks in the twenty- first century, Nature microbiology 4 (1) (2019) 10- 19. [16] G. L. Armstrong, D. R. MacCannell, J. Taylor, H. A. Carleton, E. B. Neuhaus, R. S. Bradbury, J. E. Posey, M. Gwinn, Pathogen genomics in + +<--- Page Split ---> + +et al., Ghost: global hepatitis outbreak and surveillance technology1468 BMC genomics 18 (10) (2017) 916. [42] E. M. Campbell, A. Boyles, A. Shankar, J. Kim, S. Knyazev, R. Cim470 tron, W. M. Switzer, Microbreature: retooling molecular epidemiology471 for rapid public health response, PLoS computational biology 17 (9)472 (2021) e1009300. [43] P. Sashittal, M. El-Kebir, Sharptni: counting and sampling parsimonious474 transmission networks under a weak bottleneck, bioRxiv (2019) 8422371475. [44] S. Dhar, C. Zhang, I. Mandoiu, M. S. Bansal, Tnet: Transmission net476 work inference using within-host strain diversity and its application to477 geographical tracking of covid-19 spread, IEEE/ACM Transactions on478 Computational Biology and Bioinformatics. 1479 [45] Z. Ke, H. Vikalo, Graph-based reconstruction and analysis of disease480 transmission networks using viral genomic data, Journal of Computa481 tional Biology. 1482 [46] R. J. Ypma, W. M. van Ballegooijen, J. Wallinga, Relating phyloge483 netic trees to transmission trees of infectious disease outbreaks, Genetics484 195 (3) (2013) 1055- 1062. 1485 [47] N. Mollentze, L. H. Nel, S. Townsend, K. Le Roux, K. Hampson, D. T486 Haydon, S. Soubeyrand, A bayesian approach for inferring the dynam487 ics of partially observed endemic infectious diseases from space-time488 genetic data, Proceedings of the Royal Society of London B: Biological489 Sciences 281 (1782) (2014) 20133251. 1490 [48] M. J. Morelli, G. Thebaud, J. Chadueuf, D. P. King, D. T. Haydon491 S. Soubeyrand, A bayesian inference framework to reconstruct transmis492 sion trees using epidemiological and genetic data, PLoS Comput Biol493 8 (11) (2012) e1002768. 1494 [49] E. M. Cottam, G. Thebaud, J. Wadsworth, J. Gloster, L. Mansley, D. J495 Paton, D. P. King, D. T. Haydon, Integrating genetic and epidemiologi496 cal data to determine transmission pathways of foot-and-mouth disease497 virus, Proceedings of the Royal Society of London B: Biological S498 ences 275 (1637) (2008) 887- 895. 1499 [50] F. Campbell, A. Cori, N. Ferguson, T. Jombart, Bayesian inference ofso500 transmission chains using timing of symptoms, pathogen genomes and501 contact data, PLoS computational biology 15 (3) (2019) e1006930. 1502 [51] M. Hall, M. Woolhouse, A. Rambaut, Epidemic reconstruction in a phy503 1genetics framework: transmission trees as partitions of the node set504 PLoS computational biology 11 (12) (2015) e1004613. 1505 [52] M. Senghore, H. Read, P. Oza, S. Johnson, H. Passarelli-Araujo, B. P506 Taylor, S. Ashley, A. Grey, A. Callendrello, R. Lee, et al., Inferring bac507 terial transmission dynamics using deep sequencing genomic surveil508 lance data, Nature Communications 14 (1) (2023) 6397. 1509 [53] D. S. Campo, G.- L. Xia, Z. Dimitrova, Y. Lin, J. C. Forbi, L. Ganova510 + +Raeya, L. Punkova, S. Ramachandran, H. Thai, P. Skums, et al., Accurate genetic detection of hepatitis c virus transmissions in outbreak settings, The Journal of infectious diseases 213 (6) (2016) 957- 965. [54] J. O. Wertheim, S. L. Kosakovsky Pond, L. A. Forgione, S. R. Mehta, B. Murrell, S. Shah, D. M. Smith, K. Scheffler, L. V. Torian, Social and genetic networks of hiv- 1 transmission in new york city, PLoS pathogens 13 (1) (2017) e1006000. [55] O. Ratmann, M. K. Grabowski, M. Hall, T. Golubchik, C. Wymant, L. Abeler- Dorner, D. Bonsall, A. Hoppe, A. L. Brown, T. de Oliveira, et al., Inferring hiv- 1 transmission networks and sources of epidemic spread in africa with deep- sequence phylogenetic analysis, Nature communications 10 (1) (2019) 1- 13. [56] Y. Zhang, C. Wymant, O. Laeyendecker, M. K. Grabowski, M. Hall, S. Hudelson, E. Piwowar- Manning, M. McCauley, T. Gamble, M. C. Hosseinipour, et al., Evaluation of phylogenetic methods for inferring the direction of human immunodeficiency virus (hiv) transmission: Hiv prevention trials network (hptn) 052, Clinical Infectious Diseases. [57] S. Ramachandran, H. Thai, J. C. Forbi, R. R. Galang, Z. Dimitrova, G.- I. Xia, Y. Lin, L. T. Punkova, P. R. Pontones, J. Gentry, et al., A large hcv transmission network enabled a fast- growing hiv outbreak in rural indiana, 2015, EBioMedicine 37 (2018) 374- 381. [58] E. M. Campbell, H. Jia, A. Shankar, D. Hanson, W. Luo, S. Masciotra, S. M. Owen, A. M. Oster, R. R. Galang, M. W. Spiller, et al., Detailed transmission network analysis of a large opiate- driven outbreak of hiv infection in the united states, The Journal of infectious diseases 216 (9) (2017) 1053- 1062. [59] M. G. Jones, A. Khodaverdian, J. J. Quinn, M. M. Chan, J. A. Hussmann, R. Wang, C. Xu, J. S. Weissman, N. Yosef, Inference of single- cell phylogenies from lineage tracing data using cassiopeia, Genome biology 21 (1) (2020) 1- 27. [60] S. Kumar, A. Chroni, K. Tamura, M. Sanderford, O. Oladeinde, V. Aly, T. Vu, S. Miura, Pathfinder: Bayesian inference of clone migration histories in cancer, Bioinformatics 36 (Supplement_2) (2020) i675- i683. [61] M. S. Roddur, S. Snir, M. El- Kebir, Inferring temporally consistent migration histories, in: 23rd International Workshop on Algorithms in Bioinformatics (WABI 2023), Schloss Dagstuhl- Leibniz- Zentrum fur Informatik, 2023, pp. 9.1- 9.22. [62] S. Hessey, P. Fessas, S. Zaccaria, M. Jamal- Hanjani, C. Swanton, Insights into the metastatic cascade through research autopsies, Trends in Cancer. [63] M. Al Bakir, A. Huebner, C. Martinez- Ruiz, K. Grigoriadis, T. B. Watkins, O. Pich, D. A. Moore, S. Veeriah, S. Ward, J. Laycock, et al., The evolution of non- small cell lung cancer metastases in tracers, Na + +<--- Page Split ---> + +[64] H.- N. Chen, Y. Shu, F. Liao, X. Liao, H. Zhang, Y. Qin, Z. Wang, M. Luo, Q. Liu, Z. Xue, et al., Genomic evolution and diverse models of systemic metastases in colorectal cancer, Gut 71 (2) (2022) 322- 332. 157 [65] R. Schwartz, A. A. Schaffer, The evolution of tumour phylogenetics principles and practice, Nature Reviews Genetics 18 (4) (2017) 213- 159 229. 1560 [66] E. O. Romero- Severson, I. Bulla, T. Leitner, Phylogenetically resolving epidemiological linkage, Proceedings of the National Academy of Sciences (2016) 201522930. 1563 [67] T. Leitner, W. Fitch, The phylogenetics of known transmission histories, The evolution of HIV. Johns Hopkins University Press, Baltimore, Md, 1999) 315- 345. 1566 [68] G. E. Fischer, M. K. Schaefer, B. J. Labus, L. Sands, P. Rowley, I. Alstor, Azzam, P. Armour, Y. E. Khudyakov, Y. Lin, G. Xia, et al., Hepatitis e virus infections from unsafe injection practices at an endoscopy clinic in las vegas, nevada, 2007- 2008, Clinical Infectious Diseases 51 (3) (2010) 267- 273. 1571 [69] P. Erdős, A. Rényi, et al., On the evolution of random graphs, Publ. math, 1932 inst. hung. acad. sci 5 (1) (1960) 17- 60. 1573 [70] A.- L. Barabási, R. Albert, Emergence of scaling in random networks, science 286 (5439) (1999) 509- 512. 1575 [71] D. J. Watts, S. H. Strogatz, Collective dynamics of 'small' world' networks, nature 393 (6684) (1998) 440. 1577 [72] C. Colijn, J. Gardy, Phylogenetic tree shapes resolve disease transmission patterns, Evolution, medicine, and public health 2014 (1) (2014) 96- 108. 1580 [73] G. E. Leventhal, R. Kouyos, T. Stadler, V. Von Wyl, S. Yerly, J. Boni, C. Cellerai, T. Klimkait, H. F. Gunthard, S. Bonhoeffer, Inferring epidemic contact structure from phylogenetic trees, PLoS computational biology 8 (3) (2012) e1002413. 1584 [74] N. B. Carnegie, Effects of contact network structure on epidemic transmission: implications for data required to estimate network structure, Statistics in medicine 37 (2) (2018) 236- 248. 1587 [75] B. T. Grenfell, O. G. Pybus, J. R. Gog, J. L. Wood, J. M. Daly, J. Alsmundorf, E. C. Holmes, Unifying the epidemiological and evolutionary dynamics of pathogens, science 303 (5656) (2004) 327- 332. 1590 [76] E. M. Volz, K. Koelle, T. Bedford, Viral phylodynamics, PLoS computational biology 9 (3) (2013) e1002947. 1592 [77] R. Noble, D. Burri, C. Le Sueur, J. Lemant, Y. Viossat, J. N. Kathen, N. Beerenwinkel, Spatial structure governs the mode of tumour evolution, Nature ecology & evolution 6 (2) (2022) 207- 217. 1595 [78] M. A. Lewinsohn, T. Bedford, N. F. Müller, A. F. Feder, State- dependent + +evolutionary models reveal modes of solid tumour growth, Nature Ecology & Evolution 7 (4) (2023) 581- 596. [79] K. Robinson, N. Fyson, T. Cohen, C. Fraser, C. Colijn, How the dynamics and structure of sexual contact networks shape pathogen phylogenies, PLoS computational biology 9 (6) (2013) e1003105. [80] L. Villandre, D. A. Stephens, A. Labbe, H. F. Gunthard, R. Kouyos, T. Stadler, S. H. C. Study, et al., Assessment of overlap of phylogenetic transmission clusters and communities in simple sexual contact networks: applications to hiv- 1, PloS one 11 (2) (2016) e0148459. [81] D. Welch, Is network clustering detectable in transmission trees?, Viruses 3 (6) (2011) 659- 676. [82] F. Giardina, E. O. Romero- Severson, J. Albert, T. Britton, T. Leitner, Inference of transmission network structure from hiv phylogenetic trees, PLoS computational biology 13 (1) (2017) e1005316. [83] R. M. McCloskey, R. H. Liang, A. F. Poon, Reconstructing contact network parameters from viral phylogenies, Virus evolution 2 (2) (2016) vev029. [84] D. Sankoff, Minimal mutation trees of sequences, SIAM Journal on Applied Mathematics 28 (1) (1975) 35- 42. [85] W. Fitch, Towards defining the course of evolution: minimum change for a specified tree topology, Systematic Zoology 20 (1971) 406- 416. [86] D. Gusfield, Integer linear programming in computational and systems biology: an entry- level text and course, Cambridge University Press, 2019. [87] W. M. Switzer, A. Shankar, H. Jia, S. Knyazev, F. Ambrosio, R. Kelly, H. Zheng, E. M. Campbell, R. Cintron, Y. Pan, et al., High hiv diversity, recombination, and superinfection revealed in a large outbreak among persons who inject drugs in kentucky and ohio, usa, Virus Evolution (2024) vaea015. [88] A. J. Drummond, A. Rambaut, Beast: Bayesian evolutionary analysis by sampling trees, BMC evolutionary biology 7 (1) (2007) 214. [89] E. Kenah, T. Britton, M. E. Halloran, I. M. Longini Jr, Molecular infectious disease epidemiology: survival analysis and algorithms linking phylogenies to transmission trees, PLoS computational biology 12 (4) (2016) e1004869. [90] P. Hell, J. Neestril, Graphs and homomorphisms, volume 28 of Oxford Lecture Series in Mathematics and its Applications, Oxford University Press, 2004. [91] L. Lovász, Graph minor theory, Bulletin of the American Mathematical Society 43 (1) (2006) 75- 86. [92] N. Robertson, P. D. Seymour, Graph minors. xx. wagner's conjecture, Journal of Combinatorial Theory, Series B 92 (2) (2004) 325- 357. [93] A. Brandstädt, V. B. Le, J. P. Spinrad, Graph classes: a survey, SIAM, + +<--- Page Split ---> + +1597 1999. 1640 1598 [94] J. Scott, P. Kuhn, A. R. Anderson, Unifying metastasis—integratings41 1599 intravasation, circulation and end-organ colonization, Nature Reviews462 1600 Cancer 12 (7) (2012) 445- 446. 1643 1601 [95] P. K. Newton, J. Mason, K. Bethel, L. Bazhenova, J. Nieva, L. Nor464 1602 ton, P. Kuhn, Spreaders and sponges define metastasis in lung cancer: as45 1603 markov chain monte carlo mathematical model, Cancer research 73 (9)646 1604 (2013) 2760- 2769. 1647 1605 [96] P. Gerlee, M. Johansson, Inferring rates of metastatic dissemination us464 1606 ing stochastic network models, PLoS Computational Biology 15 (4)649 1607 (2019) e1006868. 1650 1608 [97] F. Liljeros, C. R. Edling, L. A. N. Amaral, H. E. Stanley, Y. Aberg, The651 1609 web of human sexual contacts, Nature 411 (6840) (2001) 907- 908. 1652 1610 [98] J. O. Wertheim, A. J. Leigh Brown, N. L. Hepler, S. R. Mehta, D. D653 1611 Richman, D. M. Smith, S. L. Kosakovsky Pond, The global transmissions64 1612 network of hiv- 1, The Journal of infectious diseases 209 (2) (2014) 304- 655 1613 313. 1656 1614 [99] G. J. Hughes, E. Fearnhill, D. Dunn, S. J. Lycett, A. Rambaut, A. J. L1657 1615 Brown, U. H. D. R. Collaboration, Molecular phylodynamics of the het658 1616 erosexual hiv epidemic in the united kingdom, PLoS pathogens 5 (9)659 1617 (2009) e1000590. 1660 1618 [100] C. M. Romano, I. M. G. de Carvalho- Mello, L. F. Jamal, F. L. de Melo1661 1619 A. Iamarino, M. Motoki, J. R. R. Pinho, E. C. Holmes, P. M. de An4662 1620 drade Zanotto, V. Consortium, et al., Social networks shape the trans4663 mission dynamics of hepatitis c virus, PLoS One 5 (6) (2010) e11170. 1664 1621 [101] P. Sashittal, H. Schmidt, M. M. Chan, B. J. Raphael, Startle: a star homo4665 1623 plasy approach for crisp- cas9 lineage tracing, bioRxiv (2022) 2022- 1266 1624 [102] C. A. Meacham, T. Duncan, The necessity of convex groups in biologi4667 cal classification, Systematic Botany (1987) 78- 90. 1668 1626 [103] S. Moran, S. Snir, Convex recolorings of strings and trees: Definitions1669 hardness results and algorithms, Journal of Computer and System Sci4670 ences 74 (5) (2008) 850- 869. 1671 1629 [104] B. Bollobás, Combinatorics: set systems, hypergraphs, families of vec4672 tors, and combinatorial probability, Cambridge University Press, 1986. 1673 1631 [105] M. C. Golumbic, Algorithmic graph theory and perfect graphs, Vol. 57 1674 Elsevier, 2004. 1675 1633 [106] M. R. Garey, R. L. Graham, D. S. Johnson, D. E. Knuth, Complexity1676 results for bandwidth minimization, SIAM Journal on Applied Mat1677 matics 34 (3) (1978) 477- 495. 1678 1636 [107] C. Dubey, U. Feige, W. Unger, Hardness results for approximating the1679 bandwidth, Journal of Computer and System Sciences 77 (1) (2011) 62- 1680 1638 90. 1681 1639 [108] A. Caprara, F. Malucelli, D. Pretolani, On bandwidth- 2 graphs, Discretese82 + +Applied Mathematics 117 (1- 3) (2002) 1- 13. [109] V. Puller, R. Neher, J. Albert, Estimating time of hiv- 1 infection from next- generation sequence diversity, PLOS Computational Biology 13 (10) (2017) e1005775. [110] M. L. Russell, C. S. Fish, S. Drescher, N. A. Cassidy, P. Chanana, S. Benki- Nugent, J. Slyker, D. Mbori- Ngacha, R. Bosire, B. Richardson, et al., Using viral sequence diversity to estimate time of hiv infection in infants, PLoS Pathogens 19 (12) (2023) e1011861. [111] P. B. I. Baykal, J. Lara, Y. Khudyakov, A. Zelikovsky, P. Skums, Quantitative differences between intra- host hcv populations from persons with recently established and persistent infections, Virus Evolution 6 (2) (2021) vea103. [112] R. Diestel, Graph theory, volume 173 of, Graduate texts in mathematics (2012) 7. [113] P. Kilpeläinen, H. Mannila, Ordered and unordered tree inclusion, SIAM Journal on Computing 24 (2) (1995) 340- 356. [114] J. Matoušek, R. Thomas, On the complexity of finding iso- and other morphisms for partial k- trees, Discrete Mathematics 108 (1- 3) (1992) 343- 364. [115] C. Bron, J. Kerbosch, Algorithm 457: finding all cliques of an undirected graph, Communications of the ACM 16 (9) (1973) 575- 577. [116] L. Lovász, M. D. Plummer, Matching theory, Vol. 367, American Mathematical Soc., 2009. [117] N. Moshiri, M. Ragonnet- Cronin, J. O. Wertheim, S. Mirarab, Favorites: simultaneous simulation of transmission networks, phylogenetic trees and sequences, Bioinformatics 35 (11) (2019) 1852- 1861. [118] A. J. L. Brown, S. J. Lycett, L. Weinert, G. J. Hughes, E. Fearnhill, D. T. Dunn, Transmission network parameters estimated from hiv sequences for a nationwide epidemic, Journal of Infectious Diseases (2011) jir550. [119] S. P. Castillo, R. A. Rebolledo, M. Arim, M. E. Hochberg, P. A. Marquet, Metastatic cells exploit their stoichiometric niche in the network of cancer ecosystems, Science Advances 9 (50) (2023) ead17902. [120] D. Posada, Cellcoal: coalescent simulation of single- cell sequencing samples, Molecular biology and evolution 37 (5) (2020) 1535- 1542. [121] S. Caminiti, I. Finocchi, R. Petreschi, On coding labeled trees, Theoretical computer science 382 (2) (2007) 97- 108. [122] C. Wymant, F. Blanquart, T. Golubchik, A. Gall, M. Bakker, D. Bezemer, N. J. Croucher, M. Hall, M. Hillebrégt, S. H. Ong, et al., Easy and accurate reconstruction of whole hiv genomes from short- read sequence data with shiver, Virus evolution 4 (1) (2018) vev007. [123] A. Stamatakis, Raxml version 8: a tool for phylogenetic analysis and post- analysis of large phylogenies, Bioinformatics 30 (9) (2014) 1312- 1313. + +<--- Page Split ---> + +1683 [124] A. McPherson, A. Roth, E. Laks, T. Masud, A. Bashashati, A. W. Zhang, G. Ha, J. Biele, D. Yap, A. Wan, et al., Divergent modes of clonal spread and intraperitoneal mixing in high-grade serous ovarian cancer, Nature genetics 48 (7) (2016) 758- 767. + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8. Comparison of sampled compatible trees with true trees under different constraints.
+ +![](images/Figure_9.jpg) + +
Figure 9. Clone tree for Patient 3
+ +<--- Page Split ---> + +# Algorithm 3 Homomorphism under convexity and sampling parsimony constraints + +Input: phylogenetic tree \(\Psi\) with the root \(\rho\) and a candidate migration tree \(T\) Output: a feasible homomorphism \(f:\Psi \to T\) or the answer that it does not exist. 1: modify \(\Psi\) by contracting paths between leafs with the same label. 2: perform a post- order traversal of the tree \(\Psi\) . 3: for every node \(\alpha\) of the post- order do 4: construct a set of partial homomorphism tokens \(H_{\alpha}\) . 5: if \(\alpha\) is a leaf then 6: \(H_{\alpha}\leftarrow \{(v,0):v\in V(T)\}\) ; 7: end if 8: if \(\alpha\) is an unlabeled internal node with children \(\beta_{1}\) and \(\beta_{2}\) then 9: for all \((v,X)\in H_{\beta_{1}}\) and \((u,Y)\in H_{\beta_{2}}\) do 10: if \(v\sim u\) , \(u\notin X\) and \(N_{T}(u) = Y\cup \{v\}\) then 11: Add \((v,X\cup \{u\})\) to \(H_{\alpha}\) 12: end if 13: if \(v\sim u\) , \(v\notin Y\) and \(N_{T}(v) = X\cup \{u\}\) then 14: Add \((u,Y\cup \{v\})\) to \(H_{\alpha}\) 15: end if 16: end for 17: end if 18: if \(\alpha\) is a labeled internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 19: \(S_{i}\leftarrow \{v:(v,X)\in H_{\beta_{i}}\) and \(|X| = \deg (v) - 1\}\) , \(i = 1,k\) ; 20: Generate the set \(Tr\) of transversals of the set system \((S_{1},\ldots ,S_{k})\) ; 21: for transversals \(V = (v_{1},\ldots ,v_{k})\in Tr\) do 22: if there exists \(v\sim v_{1},\ldots ,v_{k}\) then 23: Add \((v,V)\) to \(H_{\alpha}\) 24: end if 25: end for 26: end if 27: end for 28: if \(H_{\rho}\neq \emptyset\) then 29: perform a pre- order traversal of \(\Psi\) ; 30: for each token \(t_{1},\ldots ,t_{R}\in H_{\rho}\) do 31: assign the token \(t_{r}\) to \(\rho :A S_{\rho}\leftarrow t_{r}\) . 32: for every node \(\alpha\) of the pre- order do 33: \(f(\alpha)\leftarrow v\) , where \((v,X) = A S_{\alpha}\) . 34: if \(\alpha\) is an unlabeled internal node with children \(\beta_{1}\) and \(\beta_{2}\) then 35: select a vertex \(u\in X\) such that \((v,X\setminus \{u\})\in H_{\beta_{1}}\) and \((u,N_{T}(u)\setminus \{v\})\in H_{\beta_{2}}\) . 36: \(A S_{\beta_{1}}\leftarrow (v,X\setminus \{u\})\) and \(A S_{\beta_{2}}\leftarrow (u,N_{T}(u)\setminus \{v\})\) 37: end if 38: if \(\alpha\) is a labeled internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 39: \(A S_{\beta_{i}}\leftarrow (v_{i},N_{T}(v_{i})\setminus \{v\})\) , \(i = 1,\ldots ,k\) , where \((v,\{v_{1},\ldots ,v_{k}\}) = A S_{\alpha}\) 40: end if 41: end for 42: among generated homomorphisms \(f_{1},\ldots ,f_{R}\) , output the homomorphism \(f_{r}\) with the minimal \(D(f_{r})\) . 43: end for 44: else 45: \(f\) does not exist 46: end if + +<--- Page Split ---> diff --git a/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250_det.mmd b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..93623c1e12efde39d5af6740566f33b880b9de2b --- /dev/null +++ b/preprint/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250/preprint__1350b53bf886cd503454b4711e204323220c216cfb889145903d9db022737250_det.mmd @@ -0,0 +1,1019 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 940, 242]]<|/det|> +# Outbreaks, Metastases and Homomorphisms: Phylogenetic Inference of Migration Histories of Heterogeneous Populations under Evolutionary and Structural Constraints + +<|ref|>text<|/ref|><|det|>[[44, 263, 191, 310]]<|/det|> +Kiril Kuzmin kkuzmin1@gsu.edu + +<|ref|>text<|/ref|><|det|>[[44, 338, 268, 355]]<|/det|> +Georgia State University + +<|ref|>text<|/ref|><|det|>[[44, 361, 231, 400]]<|/det|> +Henri Schmidt Princeton University + +<|ref|>text<|/ref|><|det|>[[44, 406, 220, 444]]<|/det|> +Sagi Snir University of Haifa + +<|ref|>text<|/ref|><|det|>[[44, 451, 231, 490]]<|/det|> +Benjamin Raphael Princeton University + +<|ref|>text<|/ref|><|det|>[[44, 497, 656, 540]]<|/det|> +Pavel Skums University of Connecticut https://orcid.org/0000- 0003- 4007- 5624 + +<|ref|>sub_title<|/ref|><|det|>[[44, 582, 103, 600]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 620, 800, 640]]<|/det|> +Keywords: migration tree, phylogenetic inference, viral transmission, cancer metastasis + +<|ref|>text<|/ref|><|det|>[[44, 658, 355, 678]]<|/det|> +Posted Date: September 13th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 696, 474, 716]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 5040045/v1 + +<|ref|>text<|/ref|><|det|>[[44, 733, 915, 776]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 794, 535, 814]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 849, 936, 893]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on August 28th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 63411- 4. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[75, 201, 924, 271]]<|/det|> +# Outbreaks, Metastases and Homomorphisms: Phylogenetic Inference of Migration Histories of Heterogeneous Populations under Evolutionary and Structural Constraints + +<|ref|>text<|/ref|><|det|>[[171, 296, 825, 315]]<|/det|> +Kiril Kuzmin \(^{a,\ast}\) , Henri Schmidt \(^{b}\) , Sagi Snir \(^{c}\) , Ben Raphael \(^{b}\) , Pavel Skums \(^{d,\ast}\) + +<|ref|>text<|/ref|><|det|>[[130, 325, 877, 372]]<|/det|> +\(^{a}\) Department of Computer Science, Georgia State University, 25 Park Place, Atlanta, GA 30303, USA \(^{b}\) Department of Computer Science, Princeton University, 35 Olden St, Princeton, NJ 08540, USA \(^{c}\) Department of Evolutionary and Environmental Biology, University of Haifa, 199 Aba Khoushy Ave., Mount Carmel, Haifa 3498838, Israel \(^{d}\) School of Computing, University of Connecticut, 371 Fairfield Way, Storrs, CT 06269, USA + +<|ref|>sub_title<|/ref|><|det|>[[61, 424, 125, 437]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[60, 444, 936, 523]]<|/det|> +Many human diseases, including viral infections and cancers, are driven by the evolutionary dynamics of heterogeneous populations of genomic variants. A major type of evolutionary behavior of these populations is migration including viral transmissions and cancer metastatic spread. A common strategy for migration pathways reconstruction involves constructing a phylogenetic tree of observed genotypes and inferring its ancestral states corresponding to migration sites. Key challenges here include determining the conditions when a phylogenetic tree topology reflects the underlying migration tree structure, and balancing computational tractability, flexibility, and biological realism of inference algorithms and models. + +<|ref|>text<|/ref|><|det|>[[60, 522, 936, 574]]<|/det|> +In this study, we address these challenges using the powerful machinery of graph homomorphisms, a mathematical concept that describes how one graph can be mapped onto another while preserving its structure. We investigate how structural constraints on migration patterns and migration tree topologies influence the relationship between phylogenies and migration trees, characterize trees compatible with a given phylogeny and propose a series of algorithms to assess whether given phylogenetic and migration trees are compatible under various migration scenarios. + +<|ref|>text<|/ref|><|det|>[[60, 573, 936, 625]]<|/det|> +Leveraging our findings, we present a framework for inferring migration trees by sampling potential trees from a prior random tree distribution and identifying a subsample compatible with a given phylogeny. By varying prior tree distributions, this approach expands upon several existing models, offering a versatile strategy applicable to a variety of biological processes. We validate our methodology using simulated datasets and real data from studies of viral outbreaks and cancer metastasis, demonstrating its effectiveness across different contexts. + +<|ref|>text<|/ref|><|det|>[[60, 632, 593, 645]]<|/det|> +Keywords: migration tree, phylogenetic inference, viral transmission, cancer metastasis + +<|ref|>sub_title<|/ref|><|det|>[[60, 675, 174, 688]]<|/det|> +## 1. Introduction + +<|ref|>text<|/ref|><|det|>[[39, 707, 486, 840]]<|/det|> +Many human diseases are essentially evolutionary processes. This includes viral infections, driven by evolving populations of viral variants [1], as well as cancers associated with diversifying intra- tumor subclonal lineages [2]. Although the biological mechanisms of these diseases are different, both are fueled by highly mutable populations of disease- causing agents, whose + +<|ref|>text<|/ref|><|det|>[[508, 675, 936, 901]]<|/det|> +extreme genomic diversity originates from error- prone replication processes, whether due to the lack of a proofreading mechanism in RNA- dependent RNA polymerase or retroviral reverse transcriptase in RNA viruses [3, 4, 5], or from the genetic instability of tumor cells manifesting itself in somatic mutations, chromosomal gain/loss/translocation, and aneuploidy [6, 7, 8, 9]. Consequently, the general phylogenetic methodologies applied to these populations exhibit many similarities. From a methodological standpoint, they form a unique segment of phylogenetics and phylodynamics, fostering a mutual exchange of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[33, 95, 500, 110]]<|/det|> +18 concepts that enhances all areas of application [10, 11, 12]. + +<|ref|>text<|/ref|><|det|>[[33, 119, 500, 303]]<|/det|> +19 A major type of evolutionary behaviour of highly mutable 20 populations is migration, wherein their members spread from 21 initial sites, seeding new populations at newly invaded sites. For 22 infectious diseases, such migrations equate to pathogen trans- 23 missions, whereas in cancer this process is identified with me- 24 tastatic spread. Thus, accurate inference of migration networks 25 of heterogeneous populations is crucial for public health and 26 medical research [13, 14, 15, 16]. + +<|ref|>text<|/ref|><|det|>[[33, 310, 500, 515]]<|/det|> +27 The study of highly mutable population migration has been 28 significantly enhanced by groundbreaking advancements in se- 29 quenching technologies. State- of- the- art high- throughput target- 30 ted sequencing and single- cell DNA sequencing enable the cap- 31 ture of detailed population snapshots at exceptionally high res- 32olutions, facilitating fine- grained analysis down to the level of 33 individual genotypes [16, 17, 18, 19]. In particular, this allows 34 to examine population migration on the level of individual mi- 35 gration events [20, 17, 21, 22]. + +<|ref|>text<|/ref|><|det|>[[33, 522, 500, 892]]<|/det|> +36 A wide array of methods has been developed specifically 37 for reconstructing viral and bacterial transmission trees, reflect- 38 ing the substantial interest in tracking the spread of infectious 39 diseases. The arsenal of tools available for reconstructing trans- 40 mission networks is extensive, including but not limited to Out- 41 breaker, Outbreaker 2 [23, 24], SeqTrack [25], SCOTTI [26], 42 SOPHIE [27], Phybreak [28], Bitrugs [29], BadTriP [30], Phy- 43 loscanner [31], StrainHub [32], TransPhylo [33, 34] (along with 44 its extension TransPhyloMulti [35]), STraTUS [36], TreeFix- 45 TP [37], QUENTIN [38], VOICE [39], HIVTrace [40], GHOST 46 [41], MicrobeTrace [42], SharpTNI [43], TiTUS [12], TNeT 47 [44], AutoNet [45], and others [46, 47, 34, 48, 49, 50, 51, 52, 48 53]. These tools have been instrumental in investigation of out- 49 breaks and monitoring the transmission dynamics of pathogens 50 like HIV, hepatitis C (HCV), SARS, MERS and SARS- CoV- 2 51 [54, 55, 56, 57, 20, 58]. + +<|ref|>text<|/ref|><|det|>[[511, 95, 939, 325]]<|/det|> +Similarly, the development of methods for deducing metastatic spread histories is burgeoning, driven by advancements in single- cell DNA sequencing and CRISPR- based lineage tracing technologies. Currently, the repertoire of tools in this domain includes MACHINA [22], FitchCount (as part of the Cassiopeia suite) [21, 59], PathFinder [60], TCC, PCC, and PCCH [61]. Notwithstanding the relatively shorter list, the impact of these tools is growing, with several studies published recently leveraging these methods to gain insights into the mechanisms of metastatic spread [21, 62, 63, 64]. + +<|ref|>text<|/ref|><|det|>[[511, 333, 939, 680]]<|/det|> +Despite significant progress in the field, the wide variety of existing methods underscores that the challenge of accurately inferring heterogeneous population migration remains unresolved. This diversity of approaches indicates both the complexity of the problem and the ongoing efforts to refine and improve upon existing methods. Additionally, a major barrier to advancement in the field is the relative isolation of viral and cancer genomics fields. Some of the aforementioned tools are based on conceptually similar techniques - this applies, for example, to STraTUS (which focus on viral transmission) and FitchCount (which addresses metastatic spread) that, as demonstrated in this paper, yield virtually identical results when applied to the same datasets. This lack of interdisciplinary exchange of ideas often leads researchers to inadvertently duplicate efforts, thereby impeding progress in both fields. + +<|ref|>text<|/ref|><|det|>[[511, 688, 939, 892]]<|/det|> +Phylogenetics and phylodynamics provide the most widely used methodological frameworks for migration tree/network reconstruction [35, 22]. However, their application in this context is not straightforward. A phylogenetic tree does not directly equate to a migration tree [35], as the nodes in a phylogenetic tree represent divergences of lineages rather than specific migration events [35, 65]. While some of these divergences may result from migration, others occur within previously invaded sites. Therefore, deriving a migration tree from a phyloge + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[35, 95, 491, 350]]<|/det|> +86 netic tree essentially requires solving an ancestral trait infer- 120ence problem, wherein the internal nodes of a phylogenetic tree are annotated with labels that indicate whether each divergence event occurred within a site or as a result of seeding of a new site upon migration. Furthermore, it is crucial to effectively leverage the full spectrum of intra-site (within-host or within- tumor) population diversity uncovered by high-throughput sequencing, that often provide a strong signal for migration inference. For example, the paraphyletic relationships between populations suggest recent migrations between corresponding sites [66, 67, 39, 68]. + +<|ref|>text<|/ref|><|det|>[[35, 356, 491, 419]]<|/det|> +The challenges mentioned above highlight several crucial questions that remain unresolved and warrant further exploration. These questions include: + +<|ref|>text<|/ref|><|det|>[[35, 433, 494, 899]]<|/det|> +1) To what extent and under which conditions does the topol-ology of a phylogenetic tree reflect the structure of the underlying migration tree? This question becomes particularly important when the level of paraphyly in the labeled phylogeny is low, a situation not uncommon as the paraphyletic signal tends to diminish over time and with smaller sample sizes [66]. A- number of studies have looked at this subject but their conclusions were mixed. Some studies have found that certain migration patterns, such as super-spreading, migration chains, or more generally, migrations within networks formed by different models like Erdos-Renyi [69], preferential attachment [70], or Watts-Strogatz models [71], lead to quantitatively distinct phylogenetic tree topologies [72, 73, 74]. Additionally, the spatial structure and dynamics of heterogeneous populations, which are directly related to migrations pathways, have been shown to affect the phylogeny structure of both viral and tumor populations [75, 76, 77, 78]. On the other hand, other studies report that the direct impact of migration patterns on phylogenetic trees ranges from minimal to moderate [79, 80, 81, 82]. Finally, certain studies have drawn mixed conclusions, indicat + +<|ref|>text<|/ref|><|det|>[[510, 95, 937, 135]]<|/det|> +ing that while some migration characteristics are reflected in the phylogeny, others are not [83]. + +<|ref|>text<|/ref|><|det|>[[510, 150, 939, 497]]<|/det|> +2) How can we limit the solution space to balance computational feasibility, accuracy of inference, generalizability, and biological realism? The space of migration trees compatible with given phylogenetic trees is often vast, and its properties are not well understood [36, 12]. A sampling-consensus approach is one method to address solution ambiguity, where feasible migration trees are sampled and summarized in a weighted consensus graph, with weights reflecting posterior probabilities of edges [44, 12, 43, 26, 21]. However, the size of solution space may restrict the depth of sampling. As a response, it is common practice to narrow down the solution space to a set of plausible migration trees optimizing a specific objective function under evolutionary-based constraints. Employing constrained models also aids in preventing overfitting in presence of missing data and errors. + +<|ref|>text<|/ref|><|det|>[[510, 505, 939, 899]]<|/det|> +Various objectives and constraints have been implemented by existing methods. Limiting number of migration events, sizes of bottlenecks or numbers of back-migrations [31, 12, 44, 22, 59, 21, 61, 36, 25] is more computationally efficient and scalable due to utilization of dynamic programming [84, 85], making such approaches practical in both molecular epidemiology and computational oncology. These can also be formulated as Integer Linear Programming (ILP) problems [86] and solved with reasonable efficiency using existing ILP solvers. Models with more complex Bayesian objectives with constraints regularized as priors [35, 28, 26, 23, 30] offer a richer, biologically nuanced perspective but suffer from scalability issues and usually rely on generic methods like Markov Chain Monte Carlo (MCMC) sampling, which may not yield optimal solutions, in part due to a lack of problem-specific mathematical strategies. Balancing computational efficiency with biological comprehensiveness presents a notable challenge, compounded by the un + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 95, 492, 133]]<|/det|> +certainty of how much constraints and objectives truly limit the solution space [36, 12]. + +<|ref|>text<|/ref|><|det|>[[30, 147, 492, 212]]<|/det|> +3) How to incorporate a variety of models for phylogenetic inference of migrations into a unified modular computational framework? + +<|ref|>text<|/ref|><|det|>[[30, 220, 494, 808]]<|/det|> +Migration inference models draw on varied biological or epidemiological assumptions. For instance, viral transmission inference often incorporates case-specific temporal data like infectious periods, exposure intervals, symptom onset, diagnosis or sample collection dates to establish order of infections and eliminate unlikely transmission links [23, 47, 28, 33, 12, 26, 29]. In some rare cases, contact networks are known and can be used for the same purpose [87, 58]. While effective, such data is often unavailable, non-informative, or sensitive, particularly for endemic and pandemic diseases caused by HIV, Hepatitis C, SARS-CoV- 2, or Influenza [27, 38, 53]. In situations where case-specific data cannot be used, genomic epidemiology tools resort to broader assumptions, like the expected degree distribution of transmission networks implied by a structure of a susceptible population [38, 27]. Similarly, methods for inferring metastatic spread use constraints defined by so- called migration patterns that reflect realistic cancer migration scenarios, such as monoclonal, polyclonal or multi- source seeding [22, 61]. A commonality across all these methods is the use of structural constraints on feasible transmission networks that are considered as subgraphs of a larger "pattern" graph. It suggests need for a versatile, modular migration inference framework that integrates these varied approaches on a unified algorithmic and mathematical basis, akin e.g. to the BEAST framework for Bayesian evolutionary analysis [88]. + +<|ref|>text<|/ref|><|det|>[[30, 814, 492, 899]]<|/det|> +Addressing these challenges requires a comprehensive investigation into the mathematical properties of migration trees compatible with a given phylogeny, a topic that is not yet fully understood [36]. Our study aims to advance this area by devel + +<|ref|>text<|/ref|><|det|>[[510, 95, 937, 182]]<|/det|> +oping a novel methodology based on powerful techniques from graph theory and combinatorics. Additionally, we will use this methodology to introduce novel algorithmic approaches for inferring migration trees. + +<|ref|>text<|/ref|><|det|>[[510, 191, 937, 443]]<|/det|> +A number of earlier studies has achieved important progress in this area. Several studies noticed that migration trees compatible with a given phylogeny correspond to partitions of the phylogeny's node set or to coloring of its branches [51, 36, 33, 34]. These observations have informed the development of methods to enumerate and sample these trees, assuming a complete migration bottleneck [36] and a known sequence of migrations [89]. In fact, as we argue in this paper, the relation between a phylogenetic tree and a migration tree is described by the concept of a graph homomorphism that generalize both partitions and colorings. + +<|ref|>text<|/ref|><|det|>[[510, 451, 937, 701]]<|/det|> +Graph homomorphism is essentially a mapping between the vertices of two graphs that preserve their structure [90]. The theory of graph homomorphisms is well- established area of discrete mathematics, with deep results and rich methodology. All types of migration trees discussed so far can be described by a graph homomorphism with specific constraints. For example, migration trees compatible with a given phylogeny under the assumptions of complete sampling and complete bottleneck, that has been studied in [51, 36] are minors [91, 92] of the phylogeny or, equivalently, its homomorphic images such that inverse images of all vertices are connected subtrees. + +<|ref|>text<|/ref|><|det|>[[510, 710, 937, 892]]<|/det|> +We use mathematical and algorithmic machinery of graph homomorphism theory to delve into the details of migration inference. Specifically, we provide necessary and sufficient conditions describing trees compatible with a given phylogeny and propose a series of algorithms that evaluate the compatibility of phylogenetic and migration trees under various evolutionary scenarios through the construction of corresponding homomorphisms. We examine particular structural constraints on migra + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 95, 496, 157]]<|/det|> +tion patterns and migration tree topologies to understand their influence on the relationship between a phylogeny and a migration tree. + +<|ref|>text<|/ref|><|det|>[[57, 166, 500, 561]]<|/det|> +Based on aforementioned insights, we propose a general framework for migration inference that samples candidate migration trees from a chosen prior random tree distribution, and identifies a subsample of trees compatible with a given phylogeny. By varying prior tree distributions, this approach expands upon and generalizes several existing models, offering versatile and computationally efficient strategy applicable to a variety of biological processes associated with heterogeneous population migration. Crucially, the proposed framework is computationally fast, enabling biomedical and public health researchers to quickly test different tree priors that represent common migration models. Beyond migration reconstruction, proposed methods can be used for investigating how phylogenies constrain the space of possible migration trees, for inferring an order of known or suspected migration events, and for determining potential migration events that are definitively ruled out by a phylogeny [36]. + +<|ref|>text<|/ref|><|det|>[[58, 568, 500, 653]]<|/det|> +Proposed methodology was validated using both simulated datasets and real experimental data gathered from studies of real outbreaks and cancer metastasis, demonstrating its effectiveness and applicability across different contexts. + +<|ref|>sub_title<|/ref|><|det|>[[58, 684, 147, 698]]<|/det|> +## 2. Methods + +<|ref|>sub_title<|/ref|><|det|>[[58, 718, 208, 732]]<|/det|> +### 2.1. Basic definitions + +<|ref|>text<|/ref|><|det|>[[58, 744, 500, 855]]<|/det|> +Throughout this paper, we consider a pair of trees: a phylogenetic tree and a migration tree. For clarity, we refer to elements of a phylogeny as nodes, and to elements of a migration network as vertices, and denote them by Greek and Latin letters, respectively. + +<|ref|>text<|/ref|><|det|>[[58, 864, 500, 902]]<|/det|> +The problem of migration network inference is set up as follows. The input is a phylogenetic tree \(\Psi = (V(\Psi), E(\Psi))\) + +<|ref|>text<|/ref|><|det|>[[511, 95, 937, 348]]<|/det|> +with the leaf set \(L(\Psi)\) representing genomic variants belonging to different subpopulations (or demes), denoted by \(\mathcal{L}\) . The tree \(\Psi\) can be a standard binary phylogeny or non- binary mutation tree used in most cancer studies. Each leaf \(\lambda \in L(\Psi)\) has an assigned site label (or color) \(l_{\lambda} \in \mathcal{L}\) . The aim is to expand this labeling from the leaves to all nodes in the tree, creating a full labeling \(f: V(\Psi) \to \mathcal{L}\) . In this model, any multi- colored tree edge \(\alpha \beta\) represents a migration of genomic variants between demes \(f(\alpha)\) and \(f(\beta)\) . The migration tree \(T = T(\Psi, I)\) and with vertices \(V(T) = \mathcal{L}\) , is then formed by contracting the nodes with the same color [51]. + +<|ref|>text<|/ref|><|det|>[[511, 355, 937, 629]]<|/det|> +As mentioned in the introduction, researchers often seek migration trees satisfying particular constraints restricting types of migration or tree topologies. These constraints can be encoded using a transition pattern graph \(G\) that describes permissible patterns of migration (specific examples are provided in the following subsection). We will first consider the situation when \(G\) is a simple graph; later on, it will be extended to the cases when \(G\) is a random graph characterized by some probability distribution. In this model, a migration tree should be isomorphic (i.e. identical up to relabeling of vertices) to a subgraph of the transition pattern. Any corresponding labeling will be called feasible. + +<|ref|>text<|/ref|><|det|>[[511, 638, 937, 891]]<|/det|> +The relations between the phylogeny \(\Psi\) , the migration tree \(T\) and the transition pattern \(G\) can be captured using the concept of a graph homomorphism. A homomorphism \(f: \Psi \to G\) [90] is an adjacency- preserving mapping between vertex sets of these graphs, i.e. \(f(u)f(v) \in E(G)\) if \(uv \in E(\Psi)\) . For the sake of mathematical rigor, here and throughout this paper we assume that a transition pattern is reflexive, i.e. every vertex is adjacent to itself. With this condition in place, any feasible labeling \(f\) is a homomorphism from \(\Psi\) to \(G\) , making the migration inference problem essentially a problem of finding such a homomorphism. In graph theory, this type of problems is sometimes + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 97, 910, 520]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[59, 532, 937, 612]]<|/det|> +
Figure 1. SMiTH: Sampling Migration Trees with Homomorphisms. A: Input phylogenetic tree. B: Distribution of possible migration trees. Parallel rectangles depict a probability density function, with each rectangle's width proportional to the corresponding probability. In practice, the distribution is represented either by a random graph model or by a stochastic graph generation procedure. C: Candidate migration trees sampled from the distribution B. D: Homomorphisms from the phylogeny to three sampled trees. In the phylogenies, nodes are color-coded by their homomorphic images in a migration tree. The phylogeny layouts in the middle of each subfigure showcases how homomorphism transforms them into sampled trees. E: consensus solution derived from homomorphisms in D. The solution is shown as potential color distributions for the phylogeny's nodes (left) or as a graph where possible migration edges are weighted according to the number of supporting solutions (right), with the edge thickness indicating weight.
+ +<|ref|>text<|/ref|><|det|>[[60, 647, 374, 662]]<|/det|> +referred to as an \(G\) - coloring of the tree \(\Psi\) [90]. + +<|ref|>text<|/ref|><|det|>[[59, 672, 494, 899]]<|/det|> +Throughout this paper, we use standard graph theory no- 301. tations. We denote by \(\Psi_{\alpha}\) a subtree of \(\Psi\) rooted at the node302. \(\alpha\) . For any graph \(G\) , \(G[X]\) represents the subgraph induced by303. the subset of vertices \(X\) . We use \(u\sim v\) to indicate that ver- 304. tices \(u\) and \(v\) are adjacent. The set of neighbors of a vertex \(u_{305}\) in \(G\) is denoted as \(N_{G}(u)\) , \(\deg_{G}(u) = |N_{G}(u)|\) is the degree of306. \(u\) , \(N_{G}[u] = N_{G}(u)\cup \{u\}\) and \(N_{G}[X]\) is the union of the neigh- 307. borhoods of all vertices in a set \(X\) . Additionally, the distance between any two vertices \(u\) and \(v\) in \(G\) is denoted by \(d_{G}(u,v)\) , and \(P_{G}(u,v)\) refers to the corresponding shortest path between + +<|ref|>text<|/ref|><|det|>[[510, 647, 936, 709]]<|/det|> +them. When the graph is clear from a context, we may omit the subscripts in these notations. In the case of a phylogeny \(\Psi\) , all distances and paths are undirected. + +<|ref|>text<|/ref|><|det|>[[510, 719, 936, 828]]<|/det|> +Additionally, to simplify the notation, we will apply settheoretical operations (e.g. intersection and union) directly to subgraph of \(G\) , with the understanding that the resulting subgraph is induced by the set obtained from applying these operations to the vertex sets of the original subgraphs. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[72, 95, 468, 426]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 438, 491, 552]]<|/det|> +
Figure 2. Solutions for the Migration History Inference Problem under various constraints. For each scenario, a phylogenetic tree is illustrated in two different layouts on the left and in the middle, with the corresponding migration tree displayed on the right. In the phylogenies, nodes are color-coded by their homomorphic images in the migration tree. The layout in the middle showcases how homomorphism transforms a phylogeny into a migration tree. (a) Unconstrained solution. Subtrees formed by blue and red nodes are not connected, indicating a violation of the convexity constraint. (b) Convex solution. Each color-coded subtree is connected, but compactness is violated in the subtree rooted at node 7. (c) Convex and compact solution.
+ +<|ref|>sub_title<|/ref|><|det|>[[60, 586, 421, 600]]<|/det|> +### 2.2. Migration inference under structural constraints + +<|ref|>text<|/ref|><|det|>[[60, 611, 488, 770]]<|/det|> +The simplest form of the migration inference challenge appears when the vertices of the transition pattern graph \(G\) directly correspond to the labels of the phylogenetic tree leaves, that is, \(V(G) = \mathcal{L}\) . This variant is referred to as labeled inference. The transition pattern \(G\) can be an unweighted graph or a random graph with specified edge probabilities. Several scenarios exemplify this problem: + +<|ref|>text<|/ref|><|det|>[[86, 787, 488, 900]]<|/det|> +- For viral transmissions, \(G\) could reflect a contact network of potential hosts [87, 58], where an infection spreads through direct interactions within this network. +- Another viral transmission model assumes that transmission is only possible between hosts with overlapping ex- + +<|ref|>text<|/ref|><|det|>[[550, 95, 937, 182]]<|/det|> +posure intervals [30, 12]. In this case \(G\) is an interval graph [93], i.e. a graph with vertices representing time intervals and edges connecting vertices whose intervals intersect. + +<|ref|>text<|/ref|><|det|>[[536, 199, 937, 286]]<|/det|> +- For metastatic spread inference, \(G\) may represent a circulatory network [94, 95, 96], with vertices representing organs and edges reflecting the prior probabilities of cancer spreading between organs. + +<|ref|>sub_title<|/ref|><|det|>[[512, 305, 799, 320]]<|/det|> +## Problem 1 (labeled migration inference) + +<|ref|>text<|/ref|><|det|>[[512, 330, 553, 343]]<|/det|> +Input: + +<|ref|>text<|/ref|><|det|>[[515, 362, 936, 400]]<|/det|> +(a) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) forming the label set \(\mathcal{L}\) ; + +<|ref|>text<|/ref|><|det|>[[515, 419, 809, 435]]<|/det|> +(b) a transition pattern \(G\) with \(V(G) = \mathcal{L}\) ; + +<|ref|>text<|/ref|><|det|>[[512, 456, 565, 470]]<|/det|> +Output: + +<|ref|>text<|/ref|><|det|>[[515, 488, 936, 528]]<|/det|> +(c) a homomorphism \(f: \Psi \to G\) such that \(f(\lambda) = l_{\lambda}\) for every \(\lambda \in L(\Psi)\) . + +<|ref|>text<|/ref|><|det|>[[510, 548, 937, 776]]<|/det|> +In practical settings, however, Problem 1 might be too restrictive or not fully reflective of reality. The main issue is the absence of a known mapping between the leaf labels and the vertices of the transition pattern. In other words, the pattern \(G\) represents the permissible topology of migration rather than specific allowed migration links. For instance, we may expect that a viral transmission network likely includes a super-spreader, but the exact subpopulation associated with this super-spreader is not known. The following models to this variant of the problem: + +<|ref|>text<|/ref|><|det|>[[536, 796, 937, 907]]<|/det|> +- Viral transmission: The transition pattern \(G\) could represent a random graph that describes expected characteristics of transmission trees, like being scale-free or having a particular expected degree distribution. Such models draw on known expected properties of contact networks + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[100, 95, 500, 159]]<|/det|> +essential for infection spread [97, 98, 99, 100] without requiring actual host contact information, and have been explored in several studies [38, 27]. + +<|ref|>text<|/ref|><|det|>[[85, 176, 500, 335]]<|/det|> +Metastatic spread inference: In this case, a transition pattern \(G\) may describe plausible evolutionary scenarios of cancer migration, such as monoclonal or polyclonal seeding from single or multiple sources [22, 61]. A related idea has been applied in studies analyzing CRISPR- based lineage tracing phylogenies, where \(G\) specifies so- called star homoplasy model [101]. + +<|ref|>text<|/ref|><|det|>[[28, 353, 500, 606]]<|/det|> +In this version of migration inference problem, is no explicitly given mapping between the set of leaf labels and the vertices of the transition pattern. Instead, a feasible homomorphism should map leaves with distinct labels to distinct vertices of \(G\) and vice versa, i.e. for any \(\lambda_1, \lambda_2 \in L(\Psi)\) we have \(f(\lambda_1) \neq f(\lambda_2)\) whenever \(l_{\lambda_1} \neq l_{\lambda_2}\) (Fig. 2 (a)). We will call homomorphisms satisfying this requirement a label- distinctiveness homomorphism. It should be noted that finding such a homomorphism seems to be a non- standard variant of a graph homomorphism problem, that, to the best of our knowledge, has not been studied previously. + +<|ref|>text<|/ref|><|det|>[[28, 613, 500, 700]]<|/det|> +In such context, the first question to be asked is whether given subtree \(T\) of the transition pattern \(G\) is compatible with the phylogeny \(\Psi\) , i.e. whether there exist a label- distinctive homomorphism \(\Psi \to T\) . + +<|ref|>sub_title<|/ref|><|det|>[[28, 716, 367, 731]]<|/det|> +## Problem 2 (unlabeled migration inference) + +<|ref|>text<|/ref|><|det|>[[28, 740, 102, 754]]<|/det|> +Input: + +<|ref|>text<|/ref|><|det|>[[28, 774, 425, 790]]<|/det|> +(a2) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) + +<|ref|>text<|/ref|><|det|>[[28, 808, 161, 823]]<|/det|> +(b2) a tree \(T\) + +<|ref|>text<|/ref|><|det|>[[28, 844, 113, 858]]<|/det|> +Output: + +<|ref|>text<|/ref|><|det|>[[28, 878, 415, 894]]<|/det|> +(c2) a label- distinctive homomorphism \(f: \Psi \to T\) . + +<|ref|>text<|/ref|><|det|>[[510, 95, 937, 183]]<|/det|> +A more restricted version of Problem 3 includes an additional convexity constraint [102, 103], where tree nodes mapping to the same vertex of \(G\) form a connected subtree of \(\Psi\) (Fig. 2 (b)): + +<|ref|>sub_title<|/ref|><|det|>[[512, 195, 870, 210]]<|/det|> +## Problem 3 (convex unlabeled migration inference) + +<|ref|>text<|/ref|><|det|>[[512, 220, 647, 234]]<|/det|> +Input: (a2) and (b2) + +<|ref|>text<|/ref|><|det|>[[512, 244, 566, 258]]<|/det|> +Output: + +<|ref|>text<|/ref|><|det|>[[530, 274, 937, 339]]<|/det|> +(c3) a label- distinctive homomorphism \(f: \Psi \to T\) such that induced subgraphs \(\Psi [f^{-1}(v)]\) are connected for all \(v \in V(T)\) . + +<|ref|>text<|/ref|><|det|>[[512, 354, 937, 440]]<|/det|> +We will refer to a homomorphism satisfying (c3) as a convex homomorphism. Convex homomorphisms to trees describe migrations with a complete bottleneck; such migration trees were the focus of extensive research in previous studies [51, 36, 89]. + +<|ref|>text<|/ref|><|det|>[[512, 448, 937, 534]]<|/det|> +To define another type of constraints, consider a labeling \(l\) of nodes of the phylogeny \(\Psi\) . The labeling \(l\) is compact if the label of the most recent common ancestor (MRCA) for any group of leaves matches one of the labels within that group. + +<|ref|>text<|/ref|><|det|>[[512, 542, 937, 771]]<|/det|> +The rationale for this is based on the understanding that migrations within any given subtree could follow one of the following scenarios: either (i) migrations involve only the demes represented by the leaves of the subtree, or (ii) migrations include external demes, but lineages from these demes were not sampled due to extinction or incomplete collection. Although both scenarios are feasible, the first is more parsimonious, especially when sampling is dense and migration events span a short period of time. Hence, homomorphisms that align with this compact labeling are referred to as compact. + +<|ref|>sub_title<|/ref|><|det|>[[512, 785, 882, 800]]<|/det|> +## Problem 4 (compact unlabeled migration inference) + +<|ref|>text<|/ref|><|det|>[[512, 809, 647, 823]]<|/det|> +Input: (a2) and (b2) + +<|ref|>text<|/ref|><|det|>[[512, 835, 566, 848]]<|/det|> +Output: + +<|ref|>text<|/ref|><|det|>[[512, 867, 937, 905]]<|/det|> +(c4) a label- distinctive homomorphism \(f: \Psi \to T\) such that for any node \(\alpha \in V(\Psi)\) we have \(f(\alpha) \in f(L(\Psi_\alpha))\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 95, 485, 300]]<|/det|> +Finally, it is often desirable to produce a sample of potential solutions rather than a single solution. Such approach offers statistical backing for potential migration network edges and logically shifts the migration inference problem to a Bayesian paradigm. Potential solutions can be sampled from a random graph distribution or, if a transition pattern is a deterministic graph, from the uniform distribution of its subgraphs with specified properties (e.g. subtrees). The sampling-based version of the migration inference problem can be formulated as follows: + +<|ref|>sub_title<|/ref|><|det|>[[32, 316, 365, 331]]<|/det|> +## Problem 5 (unlabeled migration sampling) + +<|ref|>text<|/ref|><|det|>[[32, 340, 102, 354]]<|/det|> +Input: + +<|ref|>text<|/ref|><|det|>[[30, 374, 485, 468]]<|/det|> +(a5) a phylogenetic tree \(\Psi\) with leaf labels \((l_{\lambda})_{\lambda \in L(\Psi)}\) ; +(b5) a random transition pattern \(G\) with edge probabilities \(p\) : \(E(G) \to [0,1]\) that define a distribution \(\mathcal{D}\) of subtrees of \(G\) . + +<|ref|>text<|/ref|><|det|>[[30, 491, 113, 505]]<|/det|> +Output: + +<|ref|>text<|/ref|><|det|>[[30, 525, 485, 615]]<|/det|> +(c5) a sample \(S = \{T_1, \ldots , T_m\}\) from the distribution \(\mathcal{D}\) , where each subtree \(T_i\) is a homeomorphic image of \(\Psi\) , and the corresponding label-distinctive homomorphisms \(f_i: \Psi \to T_i\) are possibly convex and/or compact. + +<|ref|>sub_title<|/ref|><|det|>[[32, 636, 285, 650]]<|/det|> +### 2.3. Labeled migration inference + +<|ref|>text<|/ref|><|det|>[[30, 660, 485, 750]]<|/det|> +Problem 1 can be efficiently solved in polynomial time using dynamic programming, as detailed in Algorithm 1. Despite its simplicity, we include the algorithm here due to its relevance for subsequent, more complex methods. + +<|ref|>sub_title<|/ref|><|det|>[[32, 772, 401, 787]]<|/det|> +### 2.4. Unconstrained unlabeled migration inference + +<|ref|>text<|/ref|><|det|>[[30, 797, 485, 884]]<|/det|> +A possible strategy to solve Problem 2 involves identifying a bijection \(g: \mathcal{L} \to V(T)\) that can be extended to a homomorphism \(\Psi \to T\) using Algorithm 1. We will describe such bijections as feasible. + +<|ref|>sub_title<|/ref|><|det|>[[515, 96, 810, 110]]<|/det|> +## Algorithm 1 Unlabeled migration inference + +<|ref|>text<|/ref|><|det|>[[515, 111, 936, 464]]<|/det|> +1: Let \(\rho\) be the root of \(\Psi\) . Perform a post- order traversal of the tree \(\Psi\) . +2: for every node \(\alpha\) of the traversal do +3: construct a set of potential images \(I(\alpha) \subseteq \mathcal{L}\) : +4: if \(\alpha\) is a leaf then +5: \(I(\alpha) \leftarrow \{I(\alpha)\}\) ; +6: else +7: Suppose that \(\beta_1, \ldots , \beta_k\) are children of \(\alpha\) . Then \(I(\alpha)\) consists of vertices \(x \in V(G)\) such that there exist vertices \(y_i \in I(\beta_i)\) , \(i = 1, k\) such that \(y_i \sim x\) . +8: end if +9: end for +10: if \(I(\rho) = \emptyset\) then +11: homomorphism \(f\) does not exist +12: else +13: perform a pre- order traversal of \(\Psi\) and construct a homomorphism \(f\) as follows +14: for every node \(\alpha\) of the pre- order do +15: if \(\alpha = \rho\) then +16: select any \(v \in I(\rho)\) and set \(f(\rho) \leftarrow v\) ; +17: else +18: Let \(\omega\) be the parent of \(\alpha\) . Choose \(v \in I(\alpha)\) such that \(v \sim f(\omega)\) and set \(f(\alpha) \leftarrow v\) . +19: end if +20: end for +21: end if + +<|ref|>text<|/ref|><|det|>[[512, 487, 937, 550]]<|/det|> +Let \(L_g(u) = \{\lambda \in L(\Psi): g(l_\lambda) = u\}\) be the set of leaves in \(\Psi\) whose labels map to \(u\) . The following theorem establishes a necessary and sufficient condition for the bijection feasibility. + +<|ref|>text<|/ref|><|det|>[[512, 568, 937, 604]]<|/det|> +Theorem 1. The bijection \(g: \mathcal{L} \to V(T)\) is feasible if and only if + +<|ref|>equation<|/ref|><|det|>[[644, 615, 936, 631]]<|/det|> +\[d_T(u_1, u_2) \leq d_{\Psi}(\lambda_1, \lambda_2) \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[512, 651, 822, 668]]<|/det|> +for all \(u_1, u_2 \in V(T)\) , \(\lambda_1 \in L_g(u_1)\) , \(\lambda_2 \in L_g(u_2)\) . + +<|ref|>text<|/ref|><|det|>[[512, 686, 937, 770]]<|/det|> +Proof. The necessity of the condition stated in the theorem is a known property of graph homomorphisms [90]. However, it is generally not sufficient [90]. For our specific type of the homomorphism problem, we will demonstrate that it indeed suffices. + +<|ref|>text<|/ref|><|det|>[[512, 779, 937, 864]]<|/det|> +Consider a bijection \(g\) satisfying the condition (1). Define \(C_\alpha \subseteq V(T)\) as the image set of its clade, that is, \(C_\alpha = g(l_\lambda): \lambda \in L(\Psi_\alpha)\) . Additionally, for a node \(\alpha \in V(\Psi)\) and a vertex \(u \in V(T)\) , define \(B_\alpha (u)\) as a ball centered at \(u\) in \(T\) , with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 97, 105, 110]]<|/det|> +460 radius + +<|ref|>equation<|/ref|><|det|>[[182, 118, 480, 142]]<|/det|> +\[r_{\alpha ,u} = \min_{\lambda \in L_{\kappa}(u)\cap L(\Psi_{\alpha})}d_{\Psi}(\alpha ,\lambda) \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[30, 155, 333, 172]]<|/det|> +461 i.e., \(B_{\alpha}(u) = \{v\in V(T):d_{T}(u,v)\leq r_{\alpha ,u}\}\) + +<|ref|>text<|/ref|><|det|>[[30, 179, 484, 218]]<|/det|> +462 We proceed by establishing two auxiliary facts. The first follows directly from the properties of a tree: + +<|ref|>text<|/ref|><|det|>[[30, 232, 491, 300]]<|/det|> +464 Lemma 1. Let \(X_{1},\ldots ,X_{k}\) be subsets of vertices of the tree \(\frac{486}{487}\) \(T\) such that each subsets induces a connected subgraph and \(\bigcap_{i = 1}^{k}X_{i}\neq \emptyset\) . Then \(N[\bigcap_{i = 1}^{k}X_{i})] = \bigcap_{i = 1}^{k}N[X_{i}]\) + +<|ref|>text<|/ref|><|det|>[[30, 311, 491, 399]]<|/det|> +467 Suppose now that \(I(\alpha):\alpha \in V(\Psi)\) are the sets of potential 489 node images produced by Algorithm 1, given the matching of 490 leaf labels in \(\Psi\) and vertices of \(T\) through the bijection \(g\) . These 491 sets are described by the following lemma. + +<|ref|>text<|/ref|><|det|>[[30, 410, 489, 428]]<|/det|> +471 Lemma 2. For every node \(\alpha \in V(\Psi)\) \(I(\alpha) = \bigcap_{u\in C_{\alpha}}B_{\alpha}(u)\) + +<|ref|>text<|/ref|><|det|>[[30, 444, 491, 558]]<|/det|> +472 Proof. The proof of the lemma proceeds by induction. Consider a node \(\alpha \in V(\Psi)\) . The lemma's assertion is trivially true 495 when \(\alpha\) is a leaf. Assume now that \(\alpha\) is an internal node with 496 children \(\beta_{1},\ldots ,\beta_{k}\) , each set \(I(\beta_{i})\) is non- empty and, by the in- 497 ductive assumption, + +<|ref|>equation<|/ref|><|det|>[[207, 584, 489, 615]]<|/det|> +\[I(\beta_{i}) = \bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u). \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[30, 627, 491, 670]]<|/det|> +477 Then \(C_{\alpha} = \bigcup_{i = 1}^{k}C_{\beta_{i}}\) ; furthermore, according to Algorithm 1502 and the equality (3) we have + +<|ref|>equation<|/ref|><|det|>[[130, 685, 489, 727]]<|/det|> +\[I(\alpha) = \bigcap_{i = 1}^{k}N[I(\beta_{i})] = \bigcap_{i = 1}^{k}N\left[\bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u)\right]. \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[30, 737, 491, 875]]<|/det|> +479 Now consider a vertex \(u\in C_{\alpha}\) . Let \(\beta (u)\) be the child of \(\alpha^{504}\) 480 such that \(r_{\beta (u),u} = \min_{i = 1}^{k}r_{\beta ,u}\) ; in cases where there are multiple505 481 such children, we pick any of them. Consequently, we have506 \(r_{\alpha ,u} = r_{\beta (u),u} + 1\) , leading to the relation \(B_{\alpha}(u) = N[B_{\beta (u)}(u)]\) 507 483 Furthermore, it is obvious that \(B_{\beta (u)}(u)\subseteq B_{\beta_{i}}(u)\) for all nodes508 484 \(\beta_{i}\) such that \(u\in B_{\beta_{i}}(u)\) . Together, these observations imply the + +<|ref|>text<|/ref|><|det|>[[510, 96, 736, 111]]<|/det|> +following sequence of equalities: + +<|ref|>equation<|/ref|><|det|>[[530, 130, 936, 216]]<|/det|> +\[\bigcap_{u\in C_{\alpha}}B_{\alpha}(u) = \bigcap_{u\in C_{\alpha}}N[B_{\beta (u)}(u)] = \bigcap_{i = 1}^{k}\bigcap_{u:\beta (u) = \beta_{i}}N[B_{\beta_{i}}(u)] =\] \[\qquad = \bigcap_{i = 1}^{k}\bigcap_{u\in C_{\beta_{i}}}N[B_{\beta_{i}}(u)] \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[510, 234, 937, 300]]<|/det|> +By Lemma 1, \(N\left[\bigcap_{u\in C_{\beta_{i}}}B_{\beta_{i}}(u)\right] = \bigcap_{u\in C_{\beta_{i}}}N[B_{\beta_{i}}(u)]\) , and thus the expressions (4) and (5) are equal. This completes the proof of Lemma 2. + +<|ref|>text<|/ref|><|det|>[[510, 317, 937, 475]]<|/det|> +According to Lemma 2, Algorithm 1 succeeds whenever \(\bigcap_{u\in C_{\alpha}}B_{\alpha}(u)\neq \emptyset\) for every node \(\alpha \in V(\Psi)\) . To establish that this condition holds, we invoke so- called Helly property of subtrees. A family of sets \(S_{1},\ldots ,S_{k}\) has a Helly property [104] if \(\bigcap_{i = 1}^{k}S_{i}\neq \emptyset\) whenever \(S_{i}\cap S_{j}\neq \emptyset\) for every \(i,j\in [k]\) ; in other words, the existence of non- empty pairwise intersections guarantees a non- empty total intersection. + +<|ref|>text<|/ref|><|det|>[[510, 482, 937, 592]]<|/det|> +Subtrees of a given tree are known to have the Helly property [105]. The subsets \(B_{\alpha}(u)\) obviously induce subtrees of \(T\) . Therefore, to prove the theorem, it is sufficient to demonstrate that for every node \(\alpha \in V(\Psi)\) and for every pair of vertices \(u_{1},u_{2}\in C_{\alpha}\) , the intersection \(B_{\alpha}(u_{1})\cap B_{\alpha}(u_{2})\) is non- empty. + +<|ref|>text<|/ref|><|det|>[[510, 599, 937, 664]]<|/det|> +Select two leafs \(\lambda_{1},\lambda_{2}\in L(\Psi_{\alpha})\) that minimize the distance between nodes of the sets \(L_{g}(u_{1})\cap L(\Psi_{\alpha})\) and \(L_{g}(u_{2})\cap L(\Psi_{\alpha})\) . According to the theorem's conditions, we have: + +<|ref|>equation<|/ref|><|det|>[[560, 691, 936, 710]]<|/det|> +\[d_{T}(u_{1},u_{2})\leq d_{\Psi}(\lambda_{1},\lambda_{2})\leq d_{\Psi}(\alpha ,\lambda_{1}) + d_{\Psi}(\alpha ,\lambda_{2}). \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[510, 733, 937, 846]]<|/det|> +This implies the existence of a path between \(u_{1}\) and \(u_{2}\) in \(T\) with a length at most \(d_{\Psi}(\alpha ,\lambda_{1}) + d_{\Psi}(\alpha ,\lambda_{2})\) . On this path, there is at least one vertex \(v\) such that \(d_{T}(u_{1},v)\leq d_{\Psi}(\alpha ,\lambda_{1})\) and \(d_{T}(u_{2},v)\leq d_{\Psi}(\alpha ,\lambda_{2})\) . Consequently, \(v\) belongs to both \(B_{\alpha}(u_{1})\) and \(B_{\alpha}(u_{2})\) , thereby completing the proof. + +<|ref|>text<|/ref|><|det|>[[510, 862, 937, 900]]<|/det|> +Theorem 1 establishes that the Unlabeled Migration Inference problem is algorithmically equivalent to the problem of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 95, 492, 159]]<|/det|> +finding a bijection that satisfies (1). First of all, this allows us to demonstrate that Problem 2 is \(\mathcal{NP}\) - hard. We will prove its through a reduction from the following problem: + +<|ref|>sub_title<|/ref|><|det|>[[60, 166, 261, 181]]<|/det|> +## Graph Bandwidth problem. + +<|ref|>text<|/ref|><|det|>[[60, 190, 188, 204]]<|/det|> +Given: A graph \(G\) + +<|ref|>text<|/ref|><|det|>[[30, 214, 492, 255]]<|/det|> +Find: The minimal integer \(K = bw(G)\) for which there exists is a bijection \(f:V(G)\to \{1,\ldots ,|V(G)|\}\) such that + +<|ref|>equation<|/ref|><|det|>[[168, 284, 490, 301]]<|/det|> +\[|f(u) - f(v)|\leq K\mathrm{~for~all~}u\sim v. \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[30, 320, 492, 384]]<|/det|> +The Graph Bandwidth problem is \(\mathcal{NP}\) - hard [106], and, moreover, it cannot be approximated within any constant factor unless \(\mathcal{P} = \mathcal{NP}\) , even when the input graph \(G\) is a tree [107]. + +<|ref|>text<|/ref|><|det|>[[30, 395, 492, 460]]<|/det|> +Theorem 2. The Unlabeled Migration Inference problem is \(\mathcal{NP}\) - hard, even when all leaf labels in the phylogeny \(\Psi\) are unique. + +<|ref|>text<|/ref|><|det|>[[30, 475, 480, 491]]<|/det|> +Proof. For our purposes, it is more convenient to use an equiv + +<|ref|>text<|/ref|><|det|>[[30, 500, 375, 515]]<|/det|> +alent condition for Graph Bandwidth problem: + +<|ref|>equation<|/ref|><|det|>[[120, 540, 492, 559]]<|/det|> +\[|f(u) - f(v)|\leq Kd_G(u,v)\mathrm{~for~all~}u,v\in V(G). \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[30, 585, 492, 650]]<|/det|> +The fact that (8) implies (7) is obvious. To demonstrate that (7) implies (8), consider the shortest \((u,v)\) - path in \(G\) \((u = 553\) \(x_{0},x_{1},\ldots ,x_{d - 1},x_{d} = v)\) , where \(d = d_G(u,v)\) . Then we have + +<|ref|>equation<|/ref|><|det|>[[75, 666, 490, 750]]<|/det|> +\[\begin{array}{r l r}{{|f(u)-f(v)|=|\sum_{i=1}^{d}(f(x_{i-1})-f(x_{i}))|\leq}}&{{}}&{{}}\\ &{}&{{}\leq\sum_{i=1}^{d}|(f(x_{i-1})-f(x_{i}))|\leq K d.}}\end{array} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[30, 761, 492, 870]]<|/det|> +Now suppose that \(T^{\prime}\) is an input tree of Graph Bandwidth problem. We can assume that \(bw(T^{\prime})\geq 3\) , since graphs where \(bw(T^{\prime})\leq 2\) are recognizable in linear time [108]. To construct an instance for Problem 2, for an integer \(K\geq 3\) , we proceed as follows: + +<|ref|>text<|/ref|><|det|>[[30, 889, 483, 905]]<|/det|> +1) The input phylogeny \(\Psi_K\) is constructed by (a) subdivid + +<|ref|>text<|/ref|><|det|>[[550, 95, 937, 135]]<|/det|> +ing every edge of \(T^{\prime}\) into \(K\) edges; (b) attaching a leaf labeled \(u^{\prime}\) to every node \(u\in V(T^{\prime})\) + +<|ref|>text<|/ref|><|det|>[[530, 150, 937, 191]]<|/det|> +2) The input tree \(T\) is an \(n\) -vertex path \(P_{n}\) with \(V(T) = \{1,\ldots ,n\}\) + +<|ref|>text<|/ref|><|det|>[[530, 212, 868, 226]]<|/det|> +For the trees constructed in this manner we have: + +<|ref|>equation<|/ref|><|det|>[[506, 234, 722, 300]]<|/det|> +\[(a)L(\Psi_K) = \{u':u\in V(T')\} ;\] \[(b)d_{\Psi_K}(u',v') = Kd_S(u,v) + 2;\] \[(c)d_T(i,j) = |i - j|.\] + +<|ref|>text<|/ref|><|det|>[[506, 306, 937, 464]]<|/det|> +We will demonstrate that a polynomial- time algorithm for Problem 2 leads to a \(\frac{5}{3}\) - approximation algorithm for the Graph Bandwidth problem. Assume the existence of such an algorithm for Problem 2. Let \(K^{*}\) be the smallest integer for which this algorithm produces a sought- for homomorphism \(\Psi_{K^{*}}\to T\) . To establish the \(\frac{5}{3}\) approximation factor, we need to demonstrate the following relationship: + +<|ref|>equation<|/ref|><|det|>[[652, 483, 937, 512]]<|/det|> +\[K^{*}\leq bw(T^{\prime})\leq \frac{5}{3} K^{*} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[506, 534, 937, 620]]<|/det|> +To establish an upper bound, let us consider the feasible bijection \(g^{*}:L(\Psi_{K^{*}})\to \{1,\ldots ,n\}\) . We extend this bijection to the nodes of \(V(T^{\prime})\) by setting \(g^{*}(u) = g^{*}(u^{\prime})\) . Given the inequality (1) and assuming that \(K^{*}\geq 3\) , we have: + +<|ref|>equation<|/ref|><|det|>[[530, 650, 920, 700]]<|/det|> +\[\begin{array}{r l r} & {} & {|g^{*}(u) - g^{*}(v)| = d_{T}(g^{*}(u^{\prime}),g^{*}(v^{\prime}))\leq d_{\Psi_{K^{*}}}(u^{\prime},v^{\prime}) =}\\ & {} & {= K^{*}\cdot d_{T^{\prime}}(u,v) + 2\leq \frac{5}{3} K^{*}d_{T^{\prime}}(u,v).} \end{array} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[510, 720, 725, 737]]<|/det|> +This implies that \(bw(S)\leq \frac{5}{3} K^{*}\) + +<|ref|>text<|/ref|><|det|>[[510, 744, 937, 830]]<|/det|> +Conversely, if \(bw(T^{\prime}) = K< K^{*}\) , and the mapping \(g:V(T^{\prime})\to \{1,\ldots ,n\}\) is the bijection reflecting this bandwidth, then we can extend it to \(L(\Psi_K)\) by setting \(g(u^{\prime}) = g(u)\) . This yields: + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[75, 99, 490, 147]]<|/det|> +\[d_{T}(g(u^{\prime}),g(v^{\prime})) = |g(u) - g(v)|\leq Kd_{T^{\prime}}(u,v)<\] \[< Kd_{T^{\prime}}(u,v) + 2 = d_{\Psi_{K}}(u^{\prime},v^{\prime}).\] + +<|ref|>text<|/ref|><|det|>[[30, 170, 491, 235]]<|/det|> +This inequality contradicts the assumption that \(K^{*}\) is minimal. Therefore, we must have \(bw(S) \geq K^{*}\) . This completes the proof. + +<|ref|>text<|/ref|><|det|>[[30, 252, 491, 386]]<|/det|> +Although the Unlabeled Migration Inference problem is \(\mathcal{NP}\) - hard, we can use Theorem 1 to approach it using Integer Linear Programming (ILP). We define a feasible bijection \(g: \mathcal{L} \to S_{88}\) \(V(T)\) using binary variables \(x_{iu}\) , where \(x_{iu} = 1\) if \(g(i) = u\) , for each \(i \in \mathcal{L}\) and \(u \in V(T)\) . The ILP formulation to find such a bijection is as follows: + +<|ref|>equation<|/ref|><|det|>[[165, 404, 490, 440]]<|/det|> +\[\sum_{i \in \mathcal{L}} \sum_{u \in V(T)} \delta (i) \deg (u) x_{iu} \to \max \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[30, 459, 80, 470]]<|/det|> +S.t. + +<|ref|>equation<|/ref|><|det|>[[185, 476, 490, 512]]<|/det|> +\[\sum_{u \in V(T)} x_{iu} = 1, \quad i \in \mathcal{L}; \quad (11)\] + +<|ref|>equation<|/ref|><|det|>[[185, 517, 490, 552]]<|/det|> +\[\sum_{i \in \mathcal{L}} x_{iu} = 1, \quad u \in V(T); \quad (12)\] + +<|ref|>equation<|/ref|><|det|>[[75, 562, 490, 599]]<|/det|> +\[x_{iu} + x_{jv} \leq 1, \quad i, j \in \mathcal{L}, u, v \in V(T) \quad (13)\] + +<|ref|>equation<|/ref|><|det|>[[88, 630, 490, 647]]<|/det|> +\[x_{iu} + x_{iv} + x_{ju} + x_{jv} \leq y_{ij} + 1 \quad i, j \in \mathcal{L}, uv \in E(T); \quad (14)\] + +<|ref|>equation<|/ref|><|det|>[[216, 664, 490, 698]]<|/det|> +\[\sum_{i,j \in \mathcal{L}} y_{ij} = n - 1. \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[30, 707, 491, 888]]<|/det|> +In this formulation, constraints (11) and (12) ensure that \(x\) encodes a bijection, while constraints (13) guarantee that the bijection adheres to the conditions of Theorem 1. The auxiliary variables \(y_{ij}\) in constraints (14) indicate whether a pair of leaf labels map to adjacent vertices in \(T\) , with constraint (15) ensuring the inferred migration network forms a tree. The objective function (10) facilitates the search for a solution by leveraging the relationship between population diversity and popula + +<|ref|>text<|/ref|><|det|>[[510, 95, 937, 253]]<|/det|> +tion age [109, 110, 111], that suggests that more diverse populations, which are likely older, are also more probable origins of migration [38, 31, 66]. Consequently, it is more likely that such populations correspond to high- degree vertices in the tree \(T\) . Here a coefficient \(\delta (i)\) represents the genetic diversity of the \(i\) th subpopulation, measured as allelic entropy averaged over all allelic positions. + +<|ref|>sub_title<|/ref|><|det|>[[512, 274, 870, 290]]<|/det|> +### 2.5. Migration inference under convexity constraints + +<|ref|>text<|/ref|><|det|>[[510, 301, 937, 650]]<|/det|> +In this section, a convex label- distinctive homomorphism \(\Psi \to T\) will be called feasible. When such homomorphism exists, \(T\) can be obtained from \(\Psi\) by a series of edge contractions, making \(T\) a minor of \(\Psi\) . Generally, a graph \(G_{1}\) is a minor of a graph \(G_{2}\) if \(G_{1}\) can be obtained from \(G_{2}\) by edge contractions, edge removals and node removals [112]. It is known that the problem of detecting whether a given graph is a minor of another graph is \(\mathcal{NP}\) - hard, even when input graphs are trees [113, 114]. However, minors associated with our problem satisfy a more stringent set of conditions than general graph minors: in our case, only edge contractions are allowed and, in addition, contractions of edges between labeled nodes with different labels are forbidden. We suggest that in practical settings feasible homomorphisms can be efficiently found and enumerated using dynamic programming. + +<|ref|>text<|/ref|><|det|>[[510, 656, 936, 695]]<|/det|> +The following simple property of convex homomorphisms will be useful for our subsequent analysis: + +<|ref|>text<|/ref|><|det|>[[510, 713, 936, 775]]<|/det|> +Lemma 3. Suppose that \(f: \Psi \to T\) is a convex homomorphism. Then \(T'\) with is a subtree of \(T\) if and only if \(\Psi ' = f^{- 1}(T')\) is a subtree of \(\Psi\) . + +<|ref|>text<|/ref|><|det|>[[510, 794, 937, 904]]<|/det|> +Proof. Homomorphic image of a connected subgraph is connected, as implied by the definition of a homomorphism. The converse is also true, if \(|T'| = 1\) . Suppose that \(|T'| \geq 2\) and the subgraph \(f^{- 1}(T')\) is not connected. Consider two of its connected components, \(\Psi_1'\) and \(\Psi_2'\) , such that the unique path \(P\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 94, 491, 231]]<|/det|> +between the nodes \(\alpha \in \Psi_1'\) and \(\beta \in \Psi_2'\) is shortest among all such paths between different components. Then \(|P| \geq 3\) and \(T'' = f(P \setminus \{\alpha , \beta \}) \cap T' = \emptyset\) . Moreover, the vertices \(f(\alpha)\) and \(f(\beta)\) are adjacent to vertices of \(T''\) . This leads to two distinct paths between \(f(\alpha)\) and \(f(\beta)\) – one in \(T'\) and another passing through \(T''\) . This contradicts the fact that \(T\) is a tree. \(\square\) + +<|ref|>text<|/ref|><|det|>[[30, 247, 491, 404]]<|/det|> +Next, we describe the proposed algorithmic approach. Initially, we simplify the original phylogenetic tree \(\Psi\) by collaps ing paths between leaves sharing identical labels into a single node, a step made feasible by the convexity constraint. The resulting tree, still referred to as \(\Psi\) , may become non- binary and contains uniquely labeled leaves and possibly some labeled internal nodes. + +<|ref|>text<|/ref|><|det|>[[30, 410, 491, 618]]<|/det|> +The algorithm performs a post- order traversal of the phylogeny \(\Psi\) and, for each node \(\alpha \in V(\Psi)\) , calculates a set \(H_{\alpha}\) describing possible homomorphisms from the subtree \(\Psi_{\alpha}\) to subtrees of \(T\) . At the root node \(\rho\) , the set \(H_{\rho}\) thus describes all homomorphisms from \(\Psi\) to \(T\) . Upon completing the traversal, the algorithm either concludes that no feasible homomorphism exists (when \(H_{\rho} = \emptyset\) ), or initiates a pre- order traversal of \(\Psi\) . During this second traversal, it reconstructs feasible homomorphisms using the information from the sets \(H_{\alpha}\) . + +<|ref|>text<|/ref|><|det|>[[30, 624, 491, 737]]<|/det|> +Formally, let \(\Lambda_{\alpha}\) be the set of labeled nodes in the subtree \(\Psi_{\alpha}\) . A subtree \(T[v, X]\) of \(T\) is termed an induced \(v\) - subtree if it includes the vertex \(v\) , a subset \(X\) of \(v\) 's neighbors, and all vertices that are connected to \(v\) via paths that intersect with \(X\) (Fig. 3). + +<|ref|>text<|/ref|><|det|>[[30, 742, 491, 877]]<|/det|> +For a vertex \(\alpha \in V(\Psi)\) , the set \(H_{\alpha}\) consists of triples \((v, X, C)\) called partial homomorphism tokens or simply tokens. In each token, (i) \(v \in V(T)\) , (ii) \(X \subseteq N_T(v)\) , (iii) \(C\) is a subset of vertices of an induced \(v\) - subtree \(T[v, X]\) such that there exists a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) with \(f(\alpha) = v\) and \(f(\Lambda_{\alpha}) = C\) . + +<|ref|>text<|/ref|><|det|>[[70, 884, 489, 901]]<|/det|> +The algorithm is initialized by setting \(H_{\lambda} = \{(v, \emptyset , \{v\}) : v \in V(T)\}\) for all leafs \(\lambda\) . For an internal node \(\alpha\) , the set \(H_{\alpha}\) is constructed based on the sets from its children nodes \(\beta_1, \ldots , \beta_k\) . The construction utilizes the following lemma: + +<|ref|>text<|/ref|><|det|>[[510, 94, 937, 159]]<|/det|> +\(v \in V(T)\) for all leafs \(\lambda\) . For an internal node \(\alpha\) , the set \(H_{\alpha}\) is constructed based on the sets from its children nodes \(\beta_1, \ldots , \beta_k\) . The construction utilizes the following lemma: + +<|ref|>text<|/ref|><|det|>[[510, 175, 937, 283]]<|/det|> +Lemma 4. Let \(T[v, X]\) be an induced \(v\) - subtree. Then there exist a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) with \(f(\alpha) = v\) and \(f(\Lambda_{\alpha}) = C\) if and only if there exist tokens \((v_1, X_1, C_1) \in H_{\beta_1}, \ldots , (v_k, X_k, C_k) \in H_{\beta_k}\) satisfying the following conditions: + +<|ref|>text<|/ref|><|det|>[[515, 304, 937, 570]]<|/det|> +(a1) \(v \sim v_i\) or \(v = v_i\) for all \(i \in \{1, \ldots , k\}\) ; (b1) \((V(T[v_i, X_i]) \setminus \{v\}) \cap (V(T[v_j, X_j]) \setminus \{v\}) = \emptyset\) for all \(i, j \in \{1, \ldots , k\}\) , \(i \neq j\) ; (c1) \(v \in V(T[v_i, X_i])\) if and only if \(v_i = v\) . (d1) \(v \notin C_i\) , if \(a\) is labeled. (e1) \(X_i = N_T(v_i) \setminus \{v\}\) , if \(v_i \neq v\) . (f1) \(X = \{v_1, \ldots , v_k\} \setminus \{v\}\) ; (g1) \(C = C_1 \cup \dots \cup C_k\) , if \(\alpha\) is unlabeled, and \(C = C_1 \cup \dots \cup C_k \cup \{v\}\) , if \(\alpha\) is labeled. + +<|ref|>text<|/ref|><|det|>[[510, 586, 937, 700]]<|/det|> +Proof. Let us prove the necessity of conditions (a1) - (g1). Suppose that there exists a feasible surjective homomorphism \(f: \Psi_{\alpha} \to T[v, X]\) such that \(f(\alpha) = v\) . Define \(T_i\) as the image of the subtree \(\Psi_{\beta_i}\) under \(f\) , denoted by \(f(\Psi_{\beta_i})\) , and let \(v_i = f(\beta_i)\) . Also, let \(X_i = N_T(v_i) \cap V(T_i)\) and \(C_i = f(\Lambda_{\beta_i})\) . + +<|ref|>text<|/ref|><|det|>[[510, 707, 937, 839]]<|/det|> +We aim to demonstrate that \(T_i = T[v_i, X_i]\) . According to Lemma 3, \(T_i\) is connected, which suggests that \(T_i\) must be a subgraph of \(T[v_i, X_i]\) . Furthermore, Lemma 3 also indicates that \(f^{- 1}(T[v_i, X_i])\) is connected. Given the surjectivity of \(f\) , it follows that \(f^{- 1}(T[v_i, X_i]) \subseteq \Psi_{\beta_i}\) , leading to the conclusion that \(T[v_i, X_i] \subseteq T_i\) . + +<|ref|>text<|/ref|><|det|>[[510, 846, 937, 884]]<|/det|> +Consequently, the restriction of \(f\) to \(\Psi_{\beta_i}\) is a surjective homomorphism from \(\Psi_{\beta_i}\) to \(T[v_i, X_i]\) . Thus, the token \((v_i, X_i, C_i)\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[110, 95, 890, 744]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[59, 749, 937, 830]]<|/det|> +
Figure 3. Overview of the Dynamic Programming Algorithm for Detecting Convex Label-Distinctive Homomorphisms. A. A phylogenetic subtree, \(\Psi_{\alpha}\) , rooted at node \(\alpha\) with two children, \(\beta\) and \(\gamma\) . B. Input candidate migration tree \(T\) . The goal is to produce convex homomorphisms from \(\Psi_{\alpha}\) to induced subtrees of \(T\) from such homomorphisms for \(\Psi_{\beta}\) and \(\Psi_{\gamma}\) . C. Convex homomorphisms from \(\Psi_{\beta}\) (top row) and \(\Psi_{\gamma}\) (bottom row) to induced subtrees of \(T\) . For instance, the top figure depicts a homomorphism to an induced 1-tree \(T[1, \{4, 9, 10, 11\}]\) that consists of the vertex 1, its neighbors 4, 9, 10, 11 and all vertices connected to 1 via paths that intersect these neighbors. Nodes of subtrees are colored by their homomorphic images. Homomorphisms are organized into a bipartite graph \(\mathcal{G}\) , where edges connect homomorphism pairs \(f_{1}f_{2}\) and \(f_{1}f_{4}\) that are compatible, and there is no edge between homomorphisms \(f_{1}\) and \(f_{3}\) , that are not compatible. D. Homomorphisms \(f_{5}\) and \(f_{6}\) obtained by combining compatible homomorphisms \(f_{1}, f_{2}\) and \(f_{1}, f_{4}\) .
+ +<|ref|>text<|/ref|><|det|>[[30, 864, 580, 880]]<|/det|> +belongs to \(H_{\beta_{i}}\) . The condition (a1) follows from the homomor- 679 (b1)- (g1). + +<|ref|>text<|/ref|><|det|>[[30, 889, 498, 904]]<|/det|> +\(\alpha\) 78 phism definition, while the convexity of \(f\) yields the conditionsse8 + +<|ref|>text<|/ref|><|det|>[[536, 889, 936, 904]]<|/det|> +Conversely, suppose for each \(i \in \{1, \ldots , k\}\) , we have feasi + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[28, 95, 490, 159]]<|/det|> +681 ble homomorphisms \(f_{i}:\Psi_{\beta_{i}}\to T[v_{i},X_{i},C_{i}]\) that meet condi- 710 tions (a1) - (e1). We can construct a homomorphism \(f:\Psi_{\alpha}\to 711\) \(T[v,X,C]\) by defining \(f\) as follows: + +<|ref|>equation<|/ref|><|det|>[[165, 179, 490, 227]]<|/det|> +\[f(\gamma) = \left\{ \begin{array}{ll} f_{i}(\gamma), & \mathrm{if~}\gamma \in V(\Psi_{\beta_{i}})\\ v, & \mathrm{if~}\gamma = \alpha \end{array} \right. \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[28, 236, 490, 301]]<|/det|> +684 Condition (a1) implies that \(f\) is a homomorphism, condition716 (d1) ensures that labeled nodes map to distinct vertices of \(T_{,717}\) and (e1) guarantees that \(f\) is surjective. + +<|ref|>text<|/ref|><|det|>[[28, 308, 490, 489]]<|/det|> +687 It remains to show that \(f\) is convex. Suppose that \(f(\gamma_{1}) = _{719}\) \(f(\gamma_{2}) = w\) . If both \(\gamma_{1}\) and \(\gamma_{2}\) belong to the same subtree \(\Psi_{\beta_{1},720}\) then the entire path \(P_{\Psi}(\gamma_{1},\gamma_{2})\) maps to \(w\) due to the convexity of \(f_{721}\) \(f_{i}\) . If, on the other hand, \(\gamma_{1}\in \Psi_{\beta_{i}}\) and \(\gamma_{2}\in \Psi_{\beta_{j}}\) , then condition722 (b1) implies that \(w = v\) . Furthermore, by the condition (c1) we \(_{723}\) have \(v_{i} = v_{j} = v\) . Therefore convexity of \(f_{i}\) and \(f_{j}\) , as well as \(_{724}\) the fact that \(f(\alpha) = v\) , imply that \(f(P_{\Psi}(\gamma_{1},\gamma_{2})) = v\) . Together, \(_{725}\) these two facts prove that \(f\) is convex. + +<|ref|>text<|/ref|><|det|>[[28, 504, 490, 664]]<|/det|> +695 To convert Lemma 4 into an algorithm constructing the set 726 of tokens for a node \(\alpha\) using the tokens of its children, we 728 first need to identify tokens of children that satisfy conditions 729 (a1)- (g1). This can be achieved through the following steps. 730 For each vertex \(v\in V(T)\) , we construct a multipartite graph 731 \(\mathcal{G} = \mathcal{G}(\alpha ,v)\) (i.e. a graph partitioned into \(k\) independent sets or 732 parts), as described below: + +<|ref|>text<|/ref|><|det|>[[28, 683, 490, 812]]<|/det|> +702 (i) The parts \(A_{1},\ldots ,A_{k}\) correspond to children of \(\alpha\) 734 (ii) The vertices of the set \(A_{i}\) are tokens from the set \(\Psi_{\beta_{i}}\) satisfying the conditions (a1),(c1), (d1) and (e1). 736 (iii) Two tokens from the sets \(A_{i}\) and \(A_{j}\) are adjacent whenever 737 they satisfy the condition (b1). + +<|ref|>text<|/ref|><|det|>[[28, 831, 490, 895]]<|/det|> +In the constructed graph, sets of partial homomorphism tokens that satisfy Lemma 4 can be identified as \(k\) - vertex cliques. We employ the Bron- Kerbosch algorithm [115] to generate the + +<|ref|>text<|/ref|><|det|>[[510, 95, 936, 135]]<|/det|> +se cliques. For each identified clique, we use conditions (f1) and (g1) to construct a new token \((v,X,C)\) for the node \(\alpha\) + +<|ref|>text<|/ref|><|det|>[[510, 142, 937, 254]]<|/det|> +Additionally, for each token \((v,X,C)\in H_{\alpha}\) , we maintain pointers \(p(v,X,C)\) that link to the children tokens used in its construction. These pointers are used in the subsequent phase of the algorithm, which aims to reconstruct full feasible homomorphisms \(f\) . + +<|ref|>text<|/ref|><|det|>[[510, 261, 937, 444]]<|/det|> +During this phase, the algorithm executes a pre- order traversal of \(\Psi\) . As it progresses, it recursively assigns a specific token to each node. When a token \(t = (v,X,C)\) is assigned to node \(\alpha\) , the algorithm sets \(f(\alpha) = v\) . It then retrieves tokens for \(t\) via the pointers \(p(t)\) and assigns them to children, \(\beta_{1},\ldots ,\beta_{k}\) . This ensures that by the end of the traversal, each node in \(\Psi\) has been assigned a homomorphic image, completing the construction of the homomorphism \(f\) . + +<|ref|>text<|/ref|><|det|>[[510, 451, 937, 608]]<|/det|> +In general, the set \(H_{\rho}\) at the root node \(\rho\) may include multiple tokens, each representing a feasible homomorphism \(f:\Psi \to T\) . When multiple feasible homomorphisms are available, the algorithm selects the one that minimizes violations of the compactness constraint (to be discussed in the next subsection). In case of ties, the selection criterion shifts to minimizing the quantity + +<|ref|>equation<|/ref|><|det|>[[625, 614, 936, 646]]<|/det|> +\[D(f) = \sum_{\alpha \in \Lambda (T)}\delta (\alpha)\cdot d(f(\alpha)), \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[510, 653, 936, 715]]<|/det|> +where \(\delta (\alpha)\) is the number of labeled children of a node \(\alpha\) . This approach prioritizes homomorphisms that map high- degree nodes to high- degree vertices. + +<|ref|>text<|/ref|><|det|>[[510, 723, 937, 905]]<|/det|> +The outlined method is formalized in Algorithm 2. Its efficiency can be improved by contracting sibling leaves in both \(\Psi\) and \(T\) into a single node (vertex). The algorithm maintains a count of copies of contracted nodes used by tokens, and a new token is generated from children tokens only if the total count of each leaf in the children tokens does not exceed its overall count. This modification markedly enhances the dynamic programming algorithm's runtime. The adjustments to Lemma + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[62, 96, 558, 110]]<|/det|> +# Algorithm 2 Convex homomorphism minimizing compactness violations + +<|ref|>text<|/ref|><|det|>[[62, 111, 780, 610]]<|/det|> +Input: phylogenetic tree \(\Psi\) with the root \(\rho\) and a candidate migration tree \(T\) Output: a feasible homomorphism \(f:\Psi \to T\) or the answer that it does not exist. 1: modify \(\Psi\) by contracting paths between leafs with the same label. 2: perform a post- order traversal of \(\Psi\) 3: for every node \(\alpha\) of the post- order do 4: construct a set of partial homomorphism tokens \(H_{\alpha}\) 5: if \(\alpha\) is a leaf then 6: \(H_{\alpha}\leftarrow \{(v,0,\{v\}):v\in V(T)\}\) 7: end if 8: if \(\alpha\) is an internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 9: Let \(H(\beta_{j}) = \{(v_{i}^{j},X_{i}^{j},C_{i}^{j}):i = 1,\ldots ,l_{j}\}\) , \(j = 1,\ldots ,k\) 10: for \(v\in V(T)\) do 11: construct the multipartite graph \(\mathcal{G}(\alpha ,v)\) as described in (i)- (iii); 12: Generate the set \(K\) of \(k\) - vertex cliques of \(\mathcal{G}(\alpha ,v)\) using Bron- Kerbosch algorithm. 13: for each clique \(\{(u_{i_{1}}^{l},X_{i_{1}}^{l},C_{i_{1}}^{l}),\ldots ,(u_{i_{k}}^{l},X_{i_{k}}^{l},C_{i_{k}}^{l})\} \in K\) do 14: Construct the sets \(X\) and \(C\) using formulas (f1) and (g1). 15: Set \(H_{\alpha}\leftarrow H_{\alpha}\cup \{(v,X,C)\}\) and \(p(v,X,C)\leftarrow p(v,X,C)\cup \{(i_{1},\ldots ,i_{k})\}\) 16: end for 17: end for 18: end if 19: end for 20: if \(H_{\rho}\neq \emptyset\) then 21: perform a pre- order traversal of \(\Psi\) 22: for each token \(t_{1},\ldots ,t_{R}\in H_{\rho}\) do 23: assign the token \(t_{r}\) to \(\rho :AS_{\rho}\leftarrow t_{r}\) 24: for every node \(\alpha\) of the pre- order do 25: \(f_{r}(\alpha)\leftarrow v\) , where \((v,X,C) = AS_{\alpha}\) 26: if \(\alpha\) is an internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 27: Let \(H(\beta_{j}) = \{(v_{i}^{j},X_{i}^{j},C_{i}^{j}):i = 1,\ldots ,l_{j}\}\) , \(j = 1,\ldots ,k\) and \(p(v,X,C) = (i_{1},\ldots ,i_{k})\) 28: \(AS_{\beta_{j}}\leftarrow (v_{i_{j}}^{j},X_{i_{j}}^{j},C_{i_{j}}^{j})\) , \(j = 1,\ldots ,k\) 29: end if 30: end for 31: end for 32: among generated homomorphisms \(f_{1},\ldots ,f_{R}\) , output the homomorphism \(f_{r}\) with the minimal number of compactness violations and, in case of ties, with the minimal \(D(f_{r})\) 33: else 34: \(f\) does not exist 35: end if + +<|ref|>text<|/ref|><|det|>[[30, 635, 490, 888]]<|/det|> +743 4 and Algorithm 3 in Supplementary Material are straightforward- 754 ward, but involve numerous minor technical details; hence a 755 formal description is omitted. One particular detail, however, 756 should be mentioned: if \(\Psi^{\prime}\) and \(T^{\prime}\) represent the leaf- contracted 757 versions of \(\Psi\) and \(T\) , respectively, and \(f^{\prime}:\Psi^{\prime}\to T^{\prime}\) is a fea- 758 sible homomorphism, there may be cases where \(f(\alpha) = l\) for 759 an internal node \(\alpha\) of \(\Psi\) and a leaf \(l\) in \(T^{\prime}\) that results from the 760 contraction of leaves \(l_{1}\) and \(l_{2}\) . In such instances, \(f^{\prime}\) can be ex- 761 tended to a homomorphism \(f:\Psi \to T\) by designating \(f(\alpha)\) as 762 either \(l_{1}\) or \(l_{2}\) . To resolve this ambiguity, the leaf corresponding 763 to the population with higher diversity is chosen. + +<|ref|>text<|/ref|><|det|>[[510, 635, 937, 745]]<|/det|> +Finally, it should be noted that, strictly speaking, Algorithm 2 is not polynomial, since the number of tokens for a node of \(\Psi\) theoretically can be exponential. In practical settings, however, the algorithm is extremely fast, and require split seconds to finish. + +<|ref|>sub_title<|/ref|><|det|>[[510, 768, 936, 802]]<|/det|> +### 2.6. Migration inference with convexity and compactness constraints + +<|ref|>text<|/ref|><|det|>[[510, 820, 937, 905]]<|/det|> +A similar approach to the one outlined in Subsection 2.5 can be employed to identify homomorphisms that are both convex and compact. However, the dynamic programming algorithm can be further optimized by taking advantage of the specific nature + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 95, 196, 110]]<|/det|> +765 of these constraints. + +<|ref|>text<|/ref|><|det|>[[30, 119, 490, 370]]<|/det|> +766 As with the earlier approach, homomorphisms that are both 776 convex and compact will be referred to as feasible. Following 778 a similar methodology to that used in Algorithm 2, we begin 779 by contracting the paths in the phylogeny \(\Psi\) . We then construct 800 partial homomorphism tokens similar to those used previously, 801 with the exception that subsets \(C\) of images of labeled nodes 802 are not required. Thus, the tokens are simplified to pairs \((v,X)\) 803 where \(v\in V(T)\) and \(X\subseteq N_{T}(v)\) . A pair \((v,X)\) is included in \(H_{\alpha}\) 804 if there exists a feasible surjective homomorphism \(f:\Psi_{\alpha}\to\) 805 \(T[v,X]\) such that \(f(\alpha) = v\) , and it also satisfies the following 806 condition: + +<|ref|>equation<|/ref|><|det|>[[216, 377, 488, 395]]<|/det|> +\[|\Lambda_{\alpha}| = |T[v,X]|. \quad (18)\] + +<|ref|>text<|/ref|><|det|>[[30, 413, 490, 454]]<|/det|> +777 This condition is necessary for a partial homomorphism to be 807 extendable to a full compact homomorphism \(\Psi \to T\) + +<|ref|>text<|/ref|><|det|>[[30, 461, 490, 549]]<|/det|> +779 The algorithm is initialized by setting \(H_{\lambda} = \{(v,\emptyset):v\in_{809}\) \(V(T))\) for leafs \(\lambda\) . For an internal node \(\alpha \in V(\Psi)\) , its token set 810 \(H_{\alpha}\) is constructed from the tokens of its children \(\beta_{1},\ldots ,\beta_{k}\) using 811 Lemma 5. + +<|ref|>text<|/ref|><|det|>[[30, 566, 490, 606]]<|/det|> +783 Lemma 5. \((v,X)\in H_{\alpha}\) if and only if one of the following con- 813 ditions hold: + +<|ref|>text<|/ref|><|det|>[[30, 624, 490, 745]]<|/det|> +785 1) \(\alpha\) is not labeled and there exist \(w\in X\) such that \((v,X\backslash\) \(\{w\} \in H_{\beta_{1}}\) and \((w,N_{T}(w)\backslash \{v\})\in H_{\beta_{2}}\) 816 2) \(\alpha\) is labeled, \(|X| = k\) , and there exist a permutation \((v_{1},\ldots ,v_{k})\) \(T[v,X\backslash \{w\} ]\) and \(f_{2}:\Psi_{\beta_{2}}\to T[w,N_{T}(w)\backslash \{v\} ]\) . By defining \(f\) as of elements of \(X\) such that \(v\sim v_{1},\ldots ,v_{k}\) , and \((v_{1},N_{T}(v_{1})\backslash \{v\})\) \(\{v\} \in H_{\beta_{1}},\ldots ,(v_{k},N_{T}(v_{k})\backslash \{v\})\in H_{\beta_{k}}\) + +<|ref|>text<|/ref|><|det|>[[30, 763, 490, 825]]<|/det|> +790 Proof. We present the proof for the case where \(\alpha\) is unlabeled; the argument for labeled \(\alpha\) follows a similar rationale. In this case, \(\alpha\) was not involved in path contraction, and thus \(k = 2\) + +<|ref|>text<|/ref|><|det|>[[30, 833, 490, 900]]<|/det|> +793 Suppose that \((v,X)\in H_{\alpha}\) , i.e. \(|\Lambda_{\alpha}| = |T[v,X]|\) and there exists a feasible surjective homomorphism \(f:\Psi_{\alpha}\to T[v,X]_{\mathrm{in}}\) such that \(f(\alpha) = v\) . Consequently, \(f(\Lambda_{\alpha}) = V(T[v,X])\) , and 82 + +<|ref|>text<|/ref|><|det|>[[506, 94, 937, 183]]<|/det|> +therefore there must be a labeled node \(\gamma \in V(\Psi_{\beta_{1}})\) such that \(f(\gamma) = v\) . Given the connectivity constraint, this implies that \(f(\beta_{1}) = v\) . Meanwhile, the compactness constraint necessitates \(f(\beta_{2}) = w\neq v\) . + +<|ref|>text<|/ref|><|det|>[[506, 191, 936, 228]]<|/det|> +Following Lemma 3 and considering the connectivity constraint, it can be shown that: + +<|ref|>equation<|/ref|><|det|>[[530, 255, 936, 273]]<|/det|> +\[f(\Psi_{\beta_1}) = T[v,X\backslash \{v\} ]\mathrm{and}f(\Psi_{\beta_2}) = T[w,N(w)\backslash \{v\} ] \quad (19)\] + +<|ref|>text<|/ref|><|det|>[[506, 300, 937, 483]]<|/det|> +Let us now establish the second equality; the method for proving the first is analogous. Let \(T_{2} = f(\Psi_{\beta_{2}})\) . We know that \(v\notin V(T_{2})\) and, according to 3, \(T_{2}\) is connected. These observations imply that \(T_{2}\subseteq T[w,N(w)\backslash \{v\} ]\) . Conversely, Lemma 3 suggests that \(f^{- 1}(T[w,N(w)\backslash \{v\} ])\) is connected. Given the surjectivity of \(f\) , this implies \(f^{- 1}(T[w,N(w)\backslash \{v\} ])\subseteq \Psi_{\beta_{2}}\) , leading to \(T[w,N(w)\backslash \{v\} ]\subseteq T_{2}\) . Thus, both \(T_{2}\subseteq T[w,N(w)\backslash \{v\} ]\) and \(T[w,N(w)\backslash \{v\} ]\subseteq T_{2}\) are true, confirming the second equality. + +<|ref|>text<|/ref|><|det|>[[506, 490, 937, 623]]<|/det|> +So, the restrictions \(f|_{\Psi_{\beta_1}}\) and \(f|_{\Psi_{\beta_2}}\) are both feasible surjective homomorphisms. Additionally, \(\Lambda_{\alpha} = \Lambda_{\beta_{1}}\cup \Lambda_{\beta_{2}}\) , \(f(\Lambda_{\beta_{1}}) = f(\Psi_{\beta_{1}}) = T[v,X\backslash \{w\} ]\) and \(f(\Lambda_{\beta_{2}}) = f(\Psi_{\beta_{2}}) = T[w,N_{T}(w)\backslash \{v\} ]\) , thus confirming that the equality (18) holds for the tokens \((v,X\backslash \{w\})\) and \((w,N_{T}(w)\backslash \{v\} ]\) . This proves the necessity of condition 1). + +<|ref|>text<|/ref|><|det|>[[506, 631, 937, 717]]<|/det|> +To demonstrate the sufficiency of condition 1), we assume that there exist feasible surjective homomorphisms \(f_{1}:\Psi_{\beta_{1}}\to\) \(T[v,X\backslash \{w\} ]\) and \(f_{2}:\Psi_{\beta_{2}}\to T[w,N_{T}(w)\backslash \{v\} ]\) . By defining \(f\) as follows, we can establish a combined feasible homomorphism: + +<|ref|>equation<|/ref|><|det|>[[608, 735, 936, 808]]<|/det|> +\[f(\chi) = \left\{ \begin{array}{ll}f_{1}(\chi) & \mathrm{if}\chi \in V(\Psi_{\beta_{1}})\\ f_{2}(\chi) & \mathrm{if}\chi \in V(\Psi_{\beta_{2}})\\ \nu & \mathrm{if}\chi = \alpha \end{array} \right. \quad (20)\] + +<|ref|>text<|/ref|><|det|>[[506, 860, 937, 898]]<|/det|> +Lemma 5 can be directly applied to construct tokens for an unlabeled node \(\alpha\) using the tokens from its children. For + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 95, 500, 160]]<|/det|> +a labeled node \(\alpha\) , however, the process of finding a permuta- 854 tion \((v_{1},\ldots ,v_{k})\) of the tokens is more complex. The method to achieve this is detailed in the following approach. + +<|ref|>text<|/ref|><|det|>[[60, 166, 500, 278]]<|/det|> +Lemma 5 can be straightforwardly use to construct tokens of \(\alpha\) from tokens of its children, if \(\alpha\) is unlabeled. When \(\alpha_{859}\) is labeled, then finding a permutation \((v_{1},\ldots ,v_{k})\) required by Lemma 5 is more complicated and can be achieved using the approach described next. + +<|ref|>text<|/ref|><|det|>[[60, 284, 500, 397]]<|/det|> +Suppose that \(\mathcal{S} = (S_{1},\ldots ,S_{k})\) is a collection of sets. A vector \((x_{1},\ldots ,x_{k})\) is termed a transversal of \(\mathcal{S}\) [116] if \(x_{i}\in S_{i}\) and all \(x_{i}\) are distinct. Let now \(S_{i} = \{u:(u,Y)\} \in H_{\beta_{i}}\) and \(v\sim\) \(u]\) . Then the vector \((v_{1},\ldots ,v_{k})\) satisfies the condition 2) of Lemma 5 if and only if it is a transversal of \(S\) + +<|ref|>text<|/ref|><|det|>[[60, 403, 500, 443]]<|/det|> +Given this, the set \(H_{\alpha}\) can be obtained by generating all transversals of \(S\) . This process involves the following steps: + +<|ref|>text<|/ref|><|det|>[[90, 459, 500, 572]]<|/det|> +Construct a bipartite graph \(\mathcal{B}(S)\) with parts \(\mathcal{I}\) and \(\mathcal{I}\) where \(\mathcal{I} = 1,\ldots ,k\) represents the set indices, and \(\mathcal{I} = \bigcup_{i = 1}^{k}S_{i}\) , represents all elements in the sets. In this graph, a vertex \(i\in \mathcal{I}\) is adjacent to a vertex \(j\in \mathcal{I}\) if \(j\) belongs to \(S_{i}\) + +<|ref|>text<|/ref|><|det|>[[90, 589, 500, 772]]<|/det|> +In \(\mathcal{B}\) , each transversal corresponds to a maximal matching of size \(k\) . To generate these matchings, construct a line graph \(L(\mathcal{B})\) , where each vertex represents an edge of \(\mathcal{B}\) , and two vertices are adjacent if their corresponding edges in \(\mathcal{B}\) share a common vertex. Maximal matchings of \(\mathcal{B}\) correspond to maximal independent sets in \(L(\mathcal{B})\) , that can be produced using the Bron- Kerbosch algorithm [115]. + +<|ref|>text<|/ref|><|det|>[[60, 789, 500, 852]]<|/det|> +The entire method is detailed in Supplementary Material, Algorithm 3. Like Algorithm 2, efficiency can be significantly improved by contracting sibling leaves in both \(\Psi\) and \(T\) . + +<|ref|>sub_title<|/ref|><|det|>[[512, 96, 926, 111]]<|/det|> +### 2.7. SMiTH: Sampling Migration Trees via Homomorphisms + +<|ref|>text<|/ref|><|det|>[[512, 123, 937, 255]]<|/det|> +The algorithms discussed can be effectively integrated into an Unlabeled Migration Sampling framework (Problem 5). It allows for the identification of homeomorphic images of a given phylogeny \(\Psi\) within a collection of candidate migration trees sampled from a given migration pattern represented by a specified tree distribution. + +<|ref|>text<|/ref|><|det|>[[512, 263, 937, 610]]<|/det|> +The obtained sample of migration trees can be directly analyzed to estimate the probabilities of specific migration routes or to obtain summary statistics and confidence intervals for derivative evolutionary parameters. It can be also synthesized into a single weighted consensus graph, where each edge is weighted by the number of candidate trees that support it. When the homomorphism reconstruction includes an objective function, the consensus graph is constructed from a subsample comprising the top \(\kappa \%\) of trees ranked by their objective values. For applications requiring a specific output tree – such as for benchmarking and comparison with other methods described in Subsection 3.1 – the tree is determined by calculating the maximum- weight spanning tree of the consensus graph. The entire algorithmic pipeline, named SMiTH (Sampling Migration Trees via Homomorphisms), is illustrated in Figure 1. + +<|ref|>sub_title<|/ref|><|det|>[[512, 640, 587, 654]]<|/det|> +## 3. Results + +<|ref|>sub_title<|/ref|><|det|>[[512, 675, 650, 689]]<|/det|> +### 3.1. Simulated data + +<|ref|>text<|/ref|><|det|>[[512, 701, 937, 882]]<|/det|> +To generate synthetic data, we used FAVITES [117], a tool capable of simulating genomes, phylogenies, and migration networks under various evolutionary scenarios. Although originally designed to simulate viral outbreaks, FAVITES supports general phylogenetic and population genetics models, making it suitable for simulating migrations of heterogeneous populations besides viruses. It also should be noted that, to the best of our knowledge, specialized simulation tools for metastatic + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 95, 490, 133]]<|/det|> +spread with capabilities comparable to FAVITES are currently not available. 920 + +<|ref|>text<|/ref|><|det|>[[30, 142, 490, 448]]<|/det|> +We simulated the migration of a heterogeneous population over a network of sites formed according to the Barabasi- Albert model [70]. This assumption can be valid for both viral [98, 118] and cancer [119, 95] spread. Migrations occur at a constant rate along each network edge (in viral context, this corresponds to the network- based Susceptible- Infected (SI) transmission model). Within each site, phylogenies evolved under the exponential coalescent, a model previously used to simulate intra- host [66] and intra- tumor [120] evolution. Genotypes were assumed to evolve under the GTR+ \(\Gamma\) substitution model, and were sampled simultaneously at the end of the simulation. In total, 275 simulated datasets were generated, encompassing 5- 30 demes with 100 sequences sampled per deme. + +<|ref|>text<|/ref|><|det|>[[30, 451, 490, 561]]<|/det|> +In the first series of experiments, we sampled candidate migration trees from 3 distinct prior distributions of tree topologies and evaluated their compatibility with simulated phylogenies under three different types of constraints. The prior distributions included: + +<|ref|>text<|/ref|><|det|>[[30, 580, 490, 810]]<|/det|> +T1) Degenerate distribution consisting of the true topology. Even though this scenario is unrealistic for actual migration inference, it serves as a test to determine if migration links can be accurately reconstructed when the topology is known but sites need to be correctly mapped to migration network vertices. To avoid bias linked to correlations between vertex IDs and migration times, that can be potentially introduced by the simulation methods, we produced multiple samples with randomly permuted vertex IDs. + +<|ref|>text<|/ref|><|det|>[[30, 828, 490, 865]]<|/det|> +T2) Random scale-free trees produced by the preferential attachment procedure. + +<|ref|>text<|/ref|><|det|>[[30, 884, 490, 900]]<|/det|> +T3) Uniformly distributed trees of a given size. Sampling was + +<|ref|>text<|/ref|><|det|>[[550, 95, 937, 160]]<|/det|> +performed by generating random Prufer codes [121], integer sequences of length \(n - 2\) that uniquely define \(n\) - vertex trees. + +<|ref|>text<|/ref|><|det|>[[533, 179, 845, 194]]<|/det|> +The following types of constraints were used: + +<|ref|>text<|/ref|><|det|>[[518, 213, 937, 320]]<|/det|> +H1) unconstrained homomorphism; H2) convex homomorphism minimizing the number of compactness constraint violations; H3) convex and compact homomorphism. + +<|ref|>text<|/ref|><|det|>[[512, 338, 937, 615]]<|/det|> +For each simulated dataset, we produced 9 samples of candidate migration trees corresponding to all combinations of conditions T1)- T3) and H1)- H3). The sample sizes ranged from 1,000 for the degenerate distribution without constraints, up to 1,000,000 for the uniform distribution with convexity and compactness constraints. This variation in sample size was necessary because stricter constraints require larger samples to ensure that a sufficient number of feasible trees are produced. We assessed the compatibility of these sampled trees with the given phylogenies under the respective constraints, using methods detailed in Subsections 2.4- 2.6. Using these assessments, subsamples of compatible tree were extracted. + +<|ref|>text<|/ref|><|det|>[[512, 623, 937, 757]]<|/det|> +Sampled trees were compared with true migration trees produced by FAVITES. Individual trees were compared by measuring recall, defined as the fraction of inferred transmission edges among true transmission edges; precision, the fraction of true transmission edges among inferred transmission edges; and the \(f\) - score, i.e., the harmonic mean of precision and recall. + +<|ref|>text<|/ref|><|det|>[[512, 765, 937, 900]]<|/det|> +Additionally, we summarized each subsample of compatible migration trees using a consensus graph, where each edge is weighted by the proportion of candidate trees that support that edge [26, 34, 122]. A solution can be extracted from the consensus graph by discarding edges with support below a predefined threshold. For each graph, we estimated the area under + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[170, 92, 830, 450]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 464, 936, 488]]<|/det|> +
Figure 4. (a)-(c) Percent of sampled trees that are compatible with given phylogenies. (d-f) Area under the precision-recall curve (AUC) calculated by varying the support threshold.
+ +<|ref|>text<|/ref|><|det|>[[60, 510, 500, 598]]<|/det|> +the precision- recall curve, which was calculated by varying the support threshold. We opted for the precision- recall curve in- 969 stead of the more common ROC curve due to the imbalance between the classes of true and false migration edges. + +<|ref|>text<|/ref|><|det|>[[60, 605, 501, 905]]<|/det|> +We found that the relationship between phylogenetic trees and migration trees is heavily influenced by structural constraints. In the absence of constraints, there is considerable ambiguity in the possible migration histories that align with a given phylogenetic tree topology, even when prior knowledge about true migration tree topology is available. Notably, almost every sampled scale- free network proved compatible with the given phylogenies, with a median fraction of compatible trees at \(\mu = 1.79\) (Fig. 4b). It is not entirely unexpected in light of Theorem 1, which suggests that trees with a low diameter - a common feature of scale- free networks - are more likely to be compatible with a given phylogenetic tree. However, even among uniformly sampled trees, a high compatibility rate was observed + +<|ref|>text<|/ref|><|det|>[[512, 510, 912, 526]]<|/det|> +when no constraints are applied (median \(\mu = 0.84\) , Fig. 4c). + +<|ref|>text<|/ref|><|det|>[[512, 535, 937, 857]]<|/det|> +Furthermore, without constraints individual compatible trees display only marginal agreement with true migration trees. This holds not only for scale- free and uniformly sampled trees, but even for the degenerate distribution, with median \(f\) - score within the range 0.33- 0.36 for all three tree priors (Supplementary Material, Fig. 8). In other words, even if the topology of a true migration tree is known, numerous labelings of that topology are compatible with the original phylogeny, most of which substantially diverge from the true labeling. Consequently, the true labeling is not immediately distinguishable among the alternatives without additional information. Combining a compatible tree subsample into a consensus graph, however, brings it closer to the true migration tree, with the median AUC values ranging from 0.56 to 0.59 for all three tree priors (see Fig. 4). + +<|ref|>text<|/ref|><|det|>[[512, 866, 936, 904]]<|/det|> +The introduction of convexity and compactness constraints significantly reduces the percentage of trees deemed compati + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 95, 500, 253]]<|/det|> +ble, as can be expected (Fig. 4abc). This reduction primarily eliminates incidental solutions, enhancing the alignment of the trees that meet these constraints with the true migration trees (Fig. 4def). In particular, for the degenerate distribution, introducing constraints effectively filters out ambiguous vertex mappings, nearly always recovering the true mapping, provided that the solutions satisfying the constraints exist. + +<|ref|>text<|/ref|><|det|>[[59, 261, 500, 584]]<|/det|> +Interestingly, without constraints, the use of tree priors does not enhance accuracy, as demonstrated by the lack of significant differences in AUC distributions among the tree priors \((p = 0.083\) , Kruskal- Wallis test). In contrast, when constraints are applied, prior knowledge of the migration tree structure comes beneficial, with AUCs improving as the tree prior comes tighter. In particular, under constraints, AUCs for the scale- free prior are significantly higher than those for the form prior \((p = 4.4 \cdot 10^{- 6}\) and \(p = 1.76 \cdot 10^{- 16}\) for convex and both convex and compact cases, respectively, Kruskal- Wallis test). Given that true migration trees are generated by the preferential attachment, this suggests that under constraints, phylogeny to a certain degree reflects the properties of the underlying migration network. + +<|ref|>text<|/ref|><|det|>[[59, 592, 500, 796]]<|/det|> +Taken together, these observations indicate that phylogeny topologies do indeed reflect underlying migration tree structures, but the extent of this reflection is influenced by evolutionary constraints. Moreover, the correspondence between geneins and migration trees is primarily discernible when analyzed statistically across a large sample of feasible migration trees that are compatible with the phylogeny. A single compatible tree may be arbitrary, and thus relying on a single solution may lead to misleading conclusions. + +<|ref|>text<|/ref|><|det|>[[59, 805, 500, 890]]<|/det|> +Based on those observations, we have developed a method named SMiTH (Sampling MIGration Trees using Homomorphisms) for the constrained inference of migration trees with expected general properties. This method involves sampling + +<|ref|>text<|/ref|><|det|>[[510, 95, 939, 254]]<|/det|> +candidate migration trees from a designated random tree distribution, identifying convex homomorphisms from the given phylogeny to these sampled trees while minimizing an objective function defined by the number of compactness constraint violations, constructing a consensus graph from trees with top objective values, and ultimately inferring the final migration tree as the minimal spanning tree of this consensus graph. + +<|ref|>text<|/ref|><|det|>[[510, 261, 939, 538]]<|/det|> +We benchmarked SMiTH against several existing tools designed to infer migration networks from phylogenetic tree topologies. For the sake of fairness, tools that use dated phylogenies and/or case- specific epidemiological information were not considered. The tools selected for this comparison include Casispeia [21, 59], MACHINA [22], Phyloscanner [31], STraTUS [36], and TNet [44]. For MACHINA, we ran all four migration models provided by the tool and report the best result, which was achieved using the single- source seeding model. STraTUS generates a sample of migration trees rather than a single tree; thus, similarly to SMiTH, we used the minimum spanning tree of the consensus graph for benchmarking purposes. + +<|ref|>text<|/ref|><|det|>[[510, 545, 939, 892]]<|/det|> +The results of the algorithm comparison are shown in Fig. 5. It was found that both variants of SMiTH – with uniform and scale- free tree priors – allow for a statistically significant improvement over other tools \((p < 10^{- 9}\) , multiple comparison of \(f\) - score distributions by Kruskal- Wallis test). SMiTH is followed by Cassiopeia and STraTUS – two other sampling- based methods, whose accuracies were statistically indiscernible \((p = 0.58\) , Kruskal- Wallis test). These tools indeed both produce samples of convex solutions, albeit using different algorithms. While STraTUS is doing it directly, while Cassiopeia’s module FitchCount samples most parsimonious solutions that in our examples were almost always convex. These tools were followed by MACHINA that, similarly to our approach, imposes structural constraints on plausible migration trees by considering them as subgraphs of so- called transition patterns. However, + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[190, 92, 805, 555]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 567, 937, 601]]<|/det|> +
Figure 5. (a) Summary statistics of methods performance on all simulated datasets. (b) Median \(f\) -scores of different methods on simulated datasets with varying numbers of populations. (c) Median running times of SMiTH for construction of a constrained homomorphism for a given phylogenetic tree and candidate migration tree as a function of number of migration sites
+ +<|ref|>text<|/ref|><|det|>[[59, 624, 500, 900]]<|/det|> +MACHINA produces a single solution optimizing particular \(\mathrm{ob}_{1065}\) 1054jectives rather than summarizes a sample of such solutions; this1066 in light of the observations described above, likely hampered its1067 performance vis- à- vis sampling- based methods. Similar rea1068 soning can be applied to Phyloscanner, that also produces a1069 single most parsimonious solution. In addition, Phyloscanner1070 is specifically designed to make use of paraphyly, that usually1071 provides a strong signal for migration [66] when present; con1072 sequently, its accuracy can be affected when, as in analyzed1073 test cases, the number of paraphyletic clades is limited. Fur1074 thermore, Phyloscanner usually assumes that when the popula1075 tions sampled from different cites are monophyletic, then they1076 + +<|ref|>text<|/ref|><|det|>[[512, 624, 937, 710]]<|/det|> +all have a common source [66] - the assumption that is opposite to compactness and that seems to be not always valid. The relation between algorithms' performances is mostly stable with regard to the number of migration sites (Fig. 5b). + +<|ref|>text<|/ref|><|det|>[[512, 718, 937, 899]]<|/det|> +The median running time of our method for construction of a constrained homomorphism for a given phylogenetic tree and candidate migration tree was within 0.7 seconds for all tests and tree priors (Fig. 5c), even though theoretically our algorithms can be exponential in the worst case. It allows us, using straightforward parallelization, to produce and process samples consisting in hundreds of thousands of candidate trees in reasonable time. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[64, 96, 257, 110]]<|/det|> +### 3.2. Experimental viral data + +<|ref|>text<|/ref|><|det|>[[30, 121, 492, 421]]<|/det|> +Analysis of simulated data highlights the role of structural con- 1112 strains in migration tree inference, particularly when paraphyly113 is limited. Interestingly, these constraints prove just as essential114 in scenarios with a high degree of paraphyly, albeit for different115 reasons. This is evidenced by the analysis of data of Hepatitis116 C (HCV) outbreaks, which have been considered in previous117 studies [38, 39, 44, 27]. The data comprises intra- host HCV118 populations from several outbreaks investigated by the Centers119 for Disease Control and Prevention, each population consisting120 of sequences covering Hypervariable Region 1 (HVR1) of the121 HCV genome. In each outbreak, a single primary host infected122 all other hosts, rendering the migration tree in graph- theoretical123 terms a star. + +<|ref|>text<|/ref|><|det|>[[30, 428, 492, 590]]<|/det|> +We analyzed two largest outbreaks involving 15 and 19 in125 fected hosts. Phylogenetic trees for each outbreak were con126 structed using RAxML [123]. All transmissions occurred with127 in a short time frame and, as a result, intra- host populations are128 highly intermixed (Fig. 6a). This makes paraphylytic signal129 strong, but oversaturated, thus impeding its use to reconstruct true transmission history. + +<|ref|>text<|/ref|><|det|>[[30, 595, 492, 895]]<|/det|> +This effect can be demonstrated by examining internal node labels generated by Fitch algorithm, that serves as a basis for several methods considered in the previous section. In the trees analyzed, \(32 - 34\%\) of internal nodes were assigned a single provisional label during the post- order traversal step of the dynamic programming algorithm, indicating that these labels appear in all most parsimonious solutions (or solutions with the minimal migration number in terms of [22]). Many of these nodes are adjacent, suggesting that the transmission links they represent will be identified by any parsimony- based sampling approach similar to those employed by existing tools [44, 12, 21]. For one outbreak, these links form a connected graph, whereas in the other, only one host does not integrate into this + +<|ref|>text<|/ref|><|det|>[[510, 96, 939, 301]]<|/det|> +single connected component (see Fig. 6b). These resulting graphs are relatively dense and include not only true edges but also a significant number of false positives (see Fig. 6b). Consequently, even if all true positive edges are correctly identified using nodes with multiple Fitch labels, the \(f\) - scores would not exceed 0.54 and 0.51, respectively. Enhancing the accuracy of this approach requires the filtering out of false positive edges, achievable only through the integration of additional prior information or constraints. + +<|ref|>text<|/ref|><|det|>[[510, 308, 939, 538]]<|/det|> +In contrast, sampling unconstrained candidate transmission trees from the scale- free tree distribution produce the results that are significantly closer to true transmission histories (Fig. 6c). For consensus networks derived from these samples, areas under precision- recall curve are estimated at 0.76 and 0.70. Furthermore, \(f\) - scores of solutions obtained as minimal spanning trees of consensus networks produced from top \(1\%\) of sampled trees according to the objective (10) are 0.86 and 0.94. Comparable results - \(f = 0.79\) and \(f = 0.89\) - are obtained if we use top \(1\%\) of sampled trees based on the parsimony score. + +<|ref|>sub_title<|/ref|><|det|>[[512, 557, 722, 571]]<|/det|> +### 3.3. Experimental cancer data + +<|ref|>text<|/ref|><|det|>[[510, 582, 939, 787]]<|/det|> +We employed SMiTH to analyze the migration history of metastatic ovarian cancer using the data published in [124]. The dataset comprised whole- genome and targeted sequencing data from samples collected at various anatomical sites, including the left ovary (LOv), the right ovary (ROv), and several metastases. In the original study, migration networks were inferred using hierarchical clustering trees and a Dollo parsimony model. Subsequent re- analysis using MACHINA [22] revealed several additional, more parsimonious migration histories. + +<|ref|>text<|/ref|><|det|>[[510, 795, 939, 905]]<|/det|> +We focused on the data from Patients 1, 3, and 7, that included the highest number of anatomical sites (7- 8 sites) and that were thoroughly analyzed in [22]. We used clone trees shared by the authors of [22]. For each patient, we sampled candidate migration trees from a uniform distribution using an + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 118, 884, 490]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[58, 503, 937, 538]]<|/det|> +
Figure 6. (a) Phylogenetic trees of HCV variants from two outbreaks. Variants sampled from different hosts are highlighted in different colors. (b) Graphs formed by edges corresponding to adjacent tree nodes with unique Fitch labels. True edges are highlighted in red. (c) Consensus networks of top \(1\%\) of sampled trees with respect to the objective (10). Edge thicknesses are proportional to their frequencies, true edges are highlighted in red.
+ +<|ref|>text<|/ref|><|det|>[[60, 560, 490, 693]]<|/det|> +unconstrained model. Following the methodology described in [159] [22], we resolved polytomies in clone trees to match the solution reported there. In instances where the resolution of polytomies was ambiguous, we applied a random resolution, generating a new random resolution for each sampled candidate migration tree. + +<|ref|>text<|/ref|><|det|>[[60, 702, 491, 883]]<|/det|> +For Patient 1, [124] identified a complex migration history, designating ROV as the primary tumor site. MACHINA was able to find several more parsimonious histories that suggested either LOV or ROV as the primary tumor location, but was not able to distinguish between them. These histories shared parsimony scores (referred to as "migration numbers" [22]) and the same number of migration events (named "co- migration numbers" [22]), which left the primary tumor's location am + +<|ref|>text<|/ref|><|det|>[[510, 560, 937, 765]]<|/det|> +biguous. In contrast, SMITH enabled the comparison of migration number distributions for different potential primary tumor sources (Fig. 7) rather than making the decision based on single most parsimonious solutions. Migration numbers associated with LOV and ROV were significantly lower than those of other potential sources ( \(p < 10^{- 65}\) , Mann- Whitney U test). Among these two, the lowest numbers were observed for ROV ( \(p < 10^{- 38}\) , Mann- Whitney U test), indicating a stronger statistical support for ROV as the primary tumor source. + +<|ref|>text<|/ref|><|det|>[[510, 773, 937, 883]]<|/det|> +A similar situation was observed for Patient 7. Here, [124] suggested the right uterosacral ligament (RUt) as the primary tumor location, while MACHINA identified several alternative migration histories with either the left ovary (LOv) or the right ovary (ROv) as the source, each sharing identical migration and + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[142, 98, 852, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[255, 555, 735, 567]]<|/det|> +
Figure 7. Distributions of migration numbers for trees with different primary tumor sites.
+ +<|ref|>text<|/ref|><|det|>[[17, 590, 493, 747]]<|/det|> +co- migration numbers. Based on these results, [22] argued that available data provides no evidence for the assertion that the primary tumor is located in the RUT as opposed to the ovaries. However, SMiTH provided such statistical evidence (Fig. 7) showing that migration trees with RUT as the source generally exhibited lower migration numbers \((p < 10^{- 154}\) , Mann- Whitney U test). + +<|ref|>text<|/ref|><|det|>[[17, 756, 493, 890]]<|/det|> +Patient 3 presents a different scenario. Here, MACHINA identified several migration histories with LOV or ROV as primary tumor sources. There is also an alternative history with the omentum (Om) as the source, which has a lower migration number. The latter hypothesis appears preferable if judged solely by the single most parsimonious solution. Yet, this con + +<|ref|>text<|/ref|><|det|>[[510, 590, 937, 818]]<|/det|> +clusion is not reliable due to the highly symmetric distribution of clones from different sites in the clone tree. Many clones form polytomies, offering insufficient data to clearly differentiate between the corresponding sites (e.g., all clones from LOV and LFTC are siblings, rendering these sites indistinguishable; see Supplementary Material, Fig. 9). The apparently lower migration number for Om, compared to other sites, is simply due to its representation by five clones, versus four for several other sites – a difference that could stem from sampling bias given the small number of clones involved. + +<|ref|>text<|/ref|><|det|>[[510, 827, 936, 890]]<|/det|> +SMiTH was able to capture and quantify this uncertainty. Specifically, it did not find significant differences in the distribution of migration numbers among potential primary sources, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 96, 500, 206]]<|/det|> +with \(p\) - values ranging from 0.09 to 0.96 in pairwise Mann- 1232 Whitney U tests and a \(p = 0.65\) in a joint Kruskal- Wallis1233 test (Fig. 7). This provides a statistical support for suggestion234 that the existing data is insufficient to draw reliable conclusions235 about the migration history. + +<|ref|>text<|/ref|><|det|>[[60, 216, 500, 277]]<|/det|> +In total, these examples illustrate how SMiTH can be used237 to provide statistical support for hypotheses regarding metat1236 static spread pathways. + +<|ref|>sub_title<|/ref|><|det|>[[60, 306, 159, 320]]<|/det|> +## 4. Discussion + +<|ref|>text<|/ref|><|det|>[[60, 339, 500, 664]]<|/det|> +This study is dedicated to in- depth mathematical exploration of the relationships between phylogenies and migration trees of heterogeneous genomic populations. Although it is established that phylogenetic trees impose some restrictions on migration pathways [36], the exact nature and extent of these constraints are still not well understood, despite a considerable amount of research dedicated to this problem [73, 72, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 36]. Our approach adds both depth and rigor to this area by utilizing the powerful theoretical and algorithmic framework of theory of graph homomorphisms. This framework allowed us to derive necessary and sufficient conditions for the compatibility of phylogenetic and migration trees, and to develop efficient algorithms for analyzing this compatibility through numerical experiments. + +<|ref|>text<|/ref|><|det|>[[60, 672, 500, 830]]<|/det|> +Based on our findings, we propose a general and flexible computational framework that can be used to infer migration networks under various assumptions, quantitatively assess competing hypotheses about migration dynamics, investigate the influence of phylogenies on the migration tree space, and to determine whether a potential migration history is definitively contradicted by a phylogeny or set of phylogenies. + +<|ref|>text<|/ref|><|det|>[[60, 838, 500, 900]]<|/det|> +Methodologically, our approach balances the advantages of probabilistic and parsimony methods. It incorporates scalability and the use of advanced combinatorial optimization techniques + +<|ref|>text<|/ref|><|det|>[[510, 96, 937, 159]]<|/det|> +from the latter, along with the biological plausibility of the former that comes from employing appropriate prior random tree distributions. + +<|ref|>text<|/ref|><|det|>[[510, 168, 939, 536]]<|/det|> +This study aligns well with the context of previous research. Several earlier studies have conceptualized migration inference as a coloring problem [35, 34, 36], employing this framework both to develop efficient inference algorithms and to explore the structure of migration tree space. The methodology introduced in this paper advances these prior approaches by incorporating a more comprehensive mathematical model and applying more sophisticated mathematical techniques. Similarly, the concept of constraining the tree space to random trees from a specified distribution was first introduced in our earlier studies on viral transmissions [38, 27], while a related concept of restricting the tree space to subgraphs of specific migration patterns has been utilized in computational cancer genomics [22, 61]. This paper not only refines and extends these methodologies but also integrates them into a cohesive modeling and computational framework. + +<|ref|>text<|/ref|><|det|>[[510, 545, 939, 892]]<|/det|> +The proposed approach certainly has both advantages and disadvantages. We recognize that uniform sampling from the space of candidate migration trees may not be the most optimal tool for migration dynamics inference, and employing more precise optimization or sampling techniques to navigate the tree space could substantially enhance the method's accuracy and efficiency. This work lays a foundation for further theoretical and algorithmic development in this direction, equipping researchers with the tools needed to expand the use of graph homomorphism methodologies. On the other hand, uniform sampling may be more suitable for hypothesis testing and comparison. It also should be noted that a key aspect of our methodology is that it involves sampling from the space of subtrees of a transition pattern, rather than from the space of phylogeny node labelings as was done in other studies [36, 44, 12]. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[25, 95, 504, 563]]<|/det|> +The former space is considerably smaller in practical scenar- ies which often makes the uniform sampling computationally- ies feasible. Furthermore, it is likely that migration models tailored to the specific characteristics of underlying populations could potentially yield more accurate insights into the biological processes involved. In particular, the biological mechanisms driving viral and cancer migrations are certainly very different. Nonetheless, employing a unified phylogenetic approach to study highly mutable populations offers several advantages. First, it decouples the initial phylogenetic reconstruction from its biological interpretation, thereby minimizing the risk of overfitting and ensuring that the results are less biased by underlying models. Additionally, such methods are significantly more computationally efficient and scalable compared to parameter-rich models. Finally, more general phylogenetic models offer greater flexibility and versatility, and usually can be readily extended to more specific settings through the integration of suitable priors. These features make them an excellent foundation for more detailed analyses. + +<|ref|>sub_title<|/ref|><|det|>[[27, 590, 222, 604]]<|/det|> +## 5. Acknowledgements + +<|ref|>text<|/ref|><|det|>[[25, 622, 504, 758]]<|/det|> +K.K. has been supported by a Georgia State University Molecular Basis of Disease fellowship. P.S. has been supported by the NSF grants 2047828 and 2212508. The authors thank the organizers of the Computational Genomics Summer Institute (UCLA, July 12 - August 4, 2023 and July 10 - August 2024), where several ideas of this paper were conceived. + +<|ref|>sub_title<|/ref|><|det|>[[27, 788, 204, 803]]<|/det|> +## 6. Code availability + +<|ref|>text<|/ref|><|det|>[[25, 821, 504, 860]]<|/det|> +The code developed and used in this study is available at https://github.com/compbel/SMiTH. + +<|ref|>sub_title<|/ref|><|det|>[[513, 96, 592, 110]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[517, 120, 940, 907]]<|/det|> +[1] E. Domingo, J. Sheldon, C. Perales, Viral quasispecies evolution, Microbiology and Molecular Biology Reviews 76 (2) (2012) 159- 216. [2] R. A. Burrell, N. McGranahan, J. Bartek, C. Swanton, The causes and consequences of genetic heterogeneity in cancer evolution, Nature 501 (7467) (2013) 338- 345. [3] E. C. Smith, The not- so- infinite malleability of rna viruses: Viral and cellular determinants of rna virus mutation rates, PLoS pathogens 13 (4) (2017) e1006254. [4] S. Duffy, Why are rna virus mutation rates so damn high?, PLoS biology 16 (8) (2018) e3000003. [5] R. Sanjuan, P. Domingo- Calap, Mechanisms of viral mutation, Cellular and molecular life sciences 73 (2016) 4433- 4448. [6] I. P. Tomlinson, M. Novelli, W. Bodmer, The mutation rate and cancer, Proceedings of the National Academy of Sciences 93 (25) (1996) 14800- 14803. [7] M. Greaves, Nothing in cancer makes sense except..., BMC biology 16 (1) (2018) 1- 8. [8] W. M. Grady, Genomic instability and colon cancer, Cancer and metastasis reviews 23 (1- 2) (2004) 11- 27. [9] G. S. Charames, B. Bapat, Genomic instability and cancer, Current molecular medicine 3 (7) (2003) 589- 596. [10] F. Utro, C. Levovitz, K. Hrissorrakrai, L. Parida, A common methodological phylogenomics framework for intra- patient heteroplasms in infer sars- cov- 2 sublineages and tumor clones, BMC genomics 22 (5) (2021) 1- 13. [11] T. Stadler, O. G. Pybus, M. P. Stumpf, Phylodynamics for cell biologists, Science 371 (6526) (2021) eah6266. [12] P. Sashittal, M. El- Kebir, Sampling and summarizing transmission trees with multi- strain infections, Bioinformatics 36 (Supplement_1) (2020) i362- i370. [13] J. Z. Sanborn, J. Chung, E. Purdom, N. J. Wang, H. Kakavand, J. S. Wilmott, T. Butler, J. F. Thompson, G. J. Mann, L. E. Haydu, et al., Phylogenetic analyses of melanoma reveal complex patterns of metastatic dissemination, Proceedings of the National Academy of Sciences 112 (35) (2015) 10995- 11000. [14] D. X. Nguyen, J. Massague, Genetic determinants of cancer metastasis, Nature Reviews Genetics 8 (5) (2007) 341- 352. [15] N. D. Grubaugh, J. T. Ladner, P. Lemey, O. G. Pybus, A. Rambaut, E. C. Holmes, K. G. Andersen, Tracking virus outbreaks in the twenty- first century, Nature microbiology 4 (1) (2019) 10- 19. [16] G. L. Armstrong, D. R. MacCannell, J. Taylor, H. A. Carleton, E. B. Neuhaus, R. S. Bradbury, J. E. Posey, M. Gwinn, Pathogen genomics in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[25, 99, 500, 900]]<|/det|> +et al., Ghost: global hepatitis outbreak and surveillance technology1468 BMC genomics 18 (10) (2017) 916. [42] E. M. Campbell, A. Boyles, A. Shankar, J. Kim, S. Knyazev, R. Cim470 tron, W. M. Switzer, Microbreature: retooling molecular epidemiology471 for rapid public health response, PLoS computational biology 17 (9)472 (2021) e1009300. [43] P. Sashittal, M. El-Kebir, Sharptni: counting and sampling parsimonious474 transmission networks under a weak bottleneck, bioRxiv (2019) 8422371475. [44] S. Dhar, C. Zhang, I. Mandoiu, M. S. Bansal, Tnet: Transmission net476 work inference using within-host strain diversity and its application to477 geographical tracking of covid-19 spread, IEEE/ACM Transactions on478 Computational Biology and Bioinformatics. 1479 [45] Z. Ke, H. Vikalo, Graph-based reconstruction and analysis of disease480 transmission networks using viral genomic data, Journal of Computa481 tional Biology. 1482 [46] R. J. Ypma, W. M. van Ballegooijen, J. Wallinga, Relating phyloge483 netic trees to transmission trees of infectious disease outbreaks, Genetics484 195 (3) (2013) 1055- 1062. 1485 [47] N. Mollentze, L. H. Nel, S. Townsend, K. Le Roux, K. Hampson, D. T486 Haydon, S. Soubeyrand, A bayesian approach for inferring the dynam487 ics of partially observed endemic infectious diseases from space-time488 genetic data, Proceedings of the Royal Society of London B: Biological489 Sciences 281 (1782) (2014) 20133251. 1490 [48] M. J. Morelli, G. Thebaud, J. Chadueuf, D. P. King, D. T. Haydon491 S. Soubeyrand, A bayesian inference framework to reconstruct transmis492 sion trees using epidemiological and genetic data, PLoS Comput Biol493 8 (11) (2012) e1002768. 1494 [49] E. M. Cottam, G. Thebaud, J. Wadsworth, J. Gloster, L. Mansley, D. J495 Paton, D. P. King, D. T. Haydon, Integrating genetic and epidemiologi496 cal data to determine transmission pathways of foot-and-mouth disease497 virus, Proceedings of the Royal Society of London B: Biological S498 ences 275 (1637) (2008) 887- 895. 1499 [50] F. Campbell, A. Cori, N. Ferguson, T. Jombart, Bayesian inference ofso500 transmission chains using timing of symptoms, pathogen genomes and501 contact data, PLoS computational biology 15 (3) (2019) e1006930. 1502 [51] M. Hall, M. Woolhouse, A. Rambaut, Epidemic reconstruction in a phy503 1genetics framework: transmission trees as partitions of the node set504 PLoS computational biology 11 (12) (2015) e1004613. 1505 [52] M. Senghore, H. Read, P. Oza, S. Johnson, H. Passarelli-Araujo, B. P506 Taylor, S. Ashley, A. Grey, A. Callendrello, R. Lee, et al., Inferring bac507 terial transmission dynamics using deep sequencing genomic surveil508 lance data, Nature Communications 14 (1) (2023) 6397. 1509 [53] D. S. Campo, G.- L. Xia, Z. Dimitrova, Y. Lin, J. C. Forbi, L. Ganova510 + +<|ref|>text<|/ref|><|det|>[[515, 99, 940, 900]]<|/det|> +Raeya, L. Punkova, S. Ramachandran, H. Thai, P. Skums, et al., Accurate genetic detection of hepatitis c virus transmissions in outbreak settings, The Journal of infectious diseases 213 (6) (2016) 957- 965. [54] J. O. Wertheim, S. L. Kosakovsky Pond, L. A. Forgione, S. R. Mehta, B. Murrell, S. Shah, D. M. Smith, K. Scheffler, L. V. Torian, Social and genetic networks of hiv- 1 transmission in new york city, PLoS pathogens 13 (1) (2017) e1006000. [55] O. Ratmann, M. K. Grabowski, M. Hall, T. Golubchik, C. Wymant, L. Abeler- Dorner, D. Bonsall, A. Hoppe, A. L. Brown, T. de Oliveira, et al., Inferring hiv- 1 transmission networks and sources of epidemic spread in africa with deep- sequence phylogenetic analysis, Nature communications 10 (1) (2019) 1- 13. [56] Y. Zhang, C. Wymant, O. Laeyendecker, M. K. Grabowski, M. Hall, S. Hudelson, E. Piwowar- Manning, M. McCauley, T. Gamble, M. C. Hosseinipour, et al., Evaluation of phylogenetic methods for inferring the direction of human immunodeficiency virus (hiv) transmission: Hiv prevention trials network (hptn) 052, Clinical Infectious Diseases. [57] S. Ramachandran, H. Thai, J. C. Forbi, R. R. Galang, Z. Dimitrova, G.- I. Xia, Y. Lin, L. T. Punkova, P. R. Pontones, J. Gentry, et al., A large hcv transmission network enabled a fast- growing hiv outbreak in rural indiana, 2015, EBioMedicine 37 (2018) 374- 381. [58] E. M. Campbell, H. Jia, A. Shankar, D. Hanson, W. Luo, S. Masciotra, S. M. Owen, A. M. Oster, R. R. Galang, M. W. Spiller, et al., Detailed transmission network analysis of a large opiate- driven outbreak of hiv infection in the united states, The Journal of infectious diseases 216 (9) (2017) 1053- 1062. [59] M. G. Jones, A. Khodaverdian, J. J. Quinn, M. M. Chan, J. A. Hussmann, R. Wang, C. Xu, J. S. Weissman, N. Yosef, Inference of single- cell phylogenies from lineage tracing data using cassiopeia, Genome biology 21 (1) (2020) 1- 27. [60] S. Kumar, A. Chroni, K. Tamura, M. Sanderford, O. Oladeinde, V. Aly, T. Vu, S. Miura, Pathfinder: Bayesian inference of clone migration histories in cancer, Bioinformatics 36 (Supplement_2) (2020) i675- i683. [61] M. S. Roddur, S. Snir, M. El- Kebir, Inferring temporally consistent migration histories, in: 23rd International Workshop on Algorithms in Bioinformatics (WABI 2023), Schloss Dagstuhl- Leibniz- Zentrum fur Informatik, 2023, pp. 9.1- 9.22. [62] S. Hessey, P. Fessas, S. Zaccaria, M. Jamal- Hanjani, C. Swanton, Insights into the metastatic cascade through research autopsies, Trends in Cancer. [63] M. Al Bakir, A. Huebner, C. Martinez- Ruiz, K. Grigoriadis, T. B. Watkins, O. Pich, D. A. Moore, S. Veeriah, S. Ward, J. Laycock, et al., The evolution of non- small cell lung cancer metastases in tracers, Na + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[25, 100, 497, 900]]<|/det|> +[64] H.- N. Chen, Y. Shu, F. Liao, X. Liao, H. Zhang, Y. Qin, Z. Wang, M. Luo, Q. Liu, Z. Xue, et al., Genomic evolution and diverse models of systemic metastases in colorectal cancer, Gut 71 (2) (2022) 322- 332. 157 [65] R. Schwartz, A. A. Schaffer, The evolution of tumour phylogenetics principles and practice, Nature Reviews Genetics 18 (4) (2017) 213- 159 229. 1560 [66] E. O. Romero- Severson, I. Bulla, T. Leitner, Phylogenetically resolving epidemiological linkage, Proceedings of the National Academy of Sciences (2016) 201522930. 1563 [67] T. Leitner, W. Fitch, The phylogenetics of known transmission histories, The evolution of HIV. Johns Hopkins University Press, Baltimore, Md, 1999) 315- 345. 1566 [68] G. E. Fischer, M. K. Schaefer, B. J. Labus, L. Sands, P. Rowley, I. Alstor, Azzam, P. Armour, Y. E. Khudyakov, Y. Lin, G. Xia, et al., Hepatitis e virus infections from unsafe injection practices at an endoscopy clinic in las vegas, nevada, 2007- 2008, Clinical Infectious Diseases 51 (3) (2010) 267- 273. 1571 [69] P. Erdős, A. Rényi, et al., On the evolution of random graphs, Publ. math, 1932 inst. hung. acad. sci 5 (1) (1960) 17- 60. 1573 [70] A.- L. Barabási, R. Albert, Emergence of scaling in random networks, science 286 (5439) (1999) 509- 512. 1575 [71] D. J. Watts, S. H. Strogatz, Collective dynamics of 'small' world' networks, nature 393 (6684) (1998) 440. 1577 [72] C. Colijn, J. Gardy, Phylogenetic tree shapes resolve disease transmission patterns, Evolution, medicine, and public health 2014 (1) (2014) 96- 108. 1580 [73] G. E. Leventhal, R. Kouyos, T. Stadler, V. Von Wyl, S. Yerly, J. Boni, C. Cellerai, T. Klimkait, H. F. Gunthard, S. Bonhoeffer, Inferring epidemic contact structure from phylogenetic trees, PLoS computational biology 8 (3) (2012) e1002413. 1584 [74] N. B. Carnegie, Effects of contact network structure on epidemic transmission: implications for data required to estimate network structure, Statistics in medicine 37 (2) (2018) 236- 248. 1587 [75] B. T. Grenfell, O. G. Pybus, J. R. Gog, J. L. Wood, J. M. Daly, J. Alsmundorf, E. C. Holmes, Unifying the epidemiological and evolutionary dynamics of pathogens, science 303 (5656) (2004) 327- 332. 1590 [76] E. M. Volz, K. Koelle, T. Bedford, Viral phylodynamics, PLoS computational biology 9 (3) (2013) e1002947. 1592 [77] R. Noble, D. Burri, C. Le Sueur, J. Lemant, Y. Viossat, J. N. Kathen, N. Beerenwinkel, Spatial structure governs the mode of tumour evolution, Nature ecology & evolution 6 (2) (2022) 207- 217. 1595 [78] M. A. Lewinsohn, T. Bedford, N. F. Müller, A. F. Feder, State- dependent + +<|ref|>text<|/ref|><|det|>[[515, 95, 940, 900]]<|/det|> +evolutionary models reveal modes of solid tumour growth, Nature Ecology & Evolution 7 (4) (2023) 581- 596. [79] K. Robinson, N. Fyson, T. Cohen, C. Fraser, C. Colijn, How the dynamics and structure of sexual contact networks shape pathogen phylogenies, PLoS computational biology 9 (6) (2013) e1003105. [80] L. Villandre, D. A. Stephens, A. Labbe, H. F. Gunthard, R. Kouyos, T. Stadler, S. H. C. Study, et al., Assessment of overlap of phylogenetic transmission clusters and communities in simple sexual contact networks: applications to hiv- 1, PloS one 11 (2) (2016) e0148459. [81] D. Welch, Is network clustering detectable in transmission trees?, Viruses 3 (6) (2011) 659- 676. [82] F. Giardina, E. O. Romero- Severson, J. Albert, T. Britton, T. Leitner, Inference of transmission network structure from hiv phylogenetic trees, PLoS computational biology 13 (1) (2017) e1005316. [83] R. M. McCloskey, R. H. Liang, A. F. Poon, Reconstructing contact network parameters from viral phylogenies, Virus evolution 2 (2) (2016) vev029. [84] D. Sankoff, Minimal mutation trees of sequences, SIAM Journal on Applied Mathematics 28 (1) (1975) 35- 42. [85] W. Fitch, Towards defining the course of evolution: minimum change for a specified tree topology, Systematic Zoology 20 (1971) 406- 416. [86] D. Gusfield, Integer linear programming in computational and systems biology: an entry- level text and course, Cambridge University Press, 2019. [87] W. M. Switzer, A. Shankar, H. Jia, S. Knyazev, F. Ambrosio, R. Kelly, H. Zheng, E. M. Campbell, R. Cintron, Y. Pan, et al., High hiv diversity, recombination, and superinfection revealed in a large outbreak among persons who inject drugs in kentucky and ohio, usa, Virus Evolution (2024) vaea015. [88] A. J. Drummond, A. Rambaut, Beast: Bayesian evolutionary analysis by sampling trees, BMC evolutionary biology 7 (1) (2007) 214. [89] E. Kenah, T. Britton, M. E. Halloran, I. M. Longini Jr, Molecular infectious disease epidemiology: survival analysis and algorithms linking phylogenies to transmission trees, PLoS computational biology 12 (4) (2016) e1004869. [90] P. Hell, J. Neestril, Graphs and homomorphisms, volume 28 of Oxford Lecture Series in Mathematics and its Applications, Oxford University Press, 2004. [91] L. Lovász, Graph minor theory, Bulletin of the American Mathematical Society 43 (1) (2006) 75- 86. [92] N. Robertson, P. D. Seymour, Graph minors. xx. wagner's conjecture, Journal of Combinatorial Theory, Series B 92 (2) (2004) 325- 357. [93] A. Brandstädt, V. B. Le, J. P. Spinrad, Graph classes: a survey, SIAM, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[20, 100, 497, 900]]<|/det|> +1597 1999. 1640 1598 [94] J. Scott, P. Kuhn, A. R. Anderson, Unifying metastasis—integratings41 1599 intravasation, circulation and end-organ colonization, Nature Reviews462 1600 Cancer 12 (7) (2012) 445- 446. 1643 1601 [95] P. K. Newton, J. Mason, K. Bethel, L. Bazhenova, J. Nieva, L. Nor464 1602 ton, P. Kuhn, Spreaders and sponges define metastasis in lung cancer: as45 1603 markov chain monte carlo mathematical model, Cancer research 73 (9)646 1604 (2013) 2760- 2769. 1647 1605 [96] P. Gerlee, M. Johansson, Inferring rates of metastatic dissemination us464 1606 ing stochastic network models, PLoS Computational Biology 15 (4)649 1607 (2019) e1006868. 1650 1608 [97] F. Liljeros, C. R. Edling, L. A. N. Amaral, H. E. Stanley, Y. Aberg, The651 1609 web of human sexual contacts, Nature 411 (6840) (2001) 907- 908. 1652 1610 [98] J. O. Wertheim, A. J. Leigh Brown, N. L. Hepler, S. R. Mehta, D. D653 1611 Richman, D. M. Smith, S. L. Kosakovsky Pond, The global transmissions64 1612 network of hiv- 1, The Journal of infectious diseases 209 (2) (2014) 304- 655 1613 313. 1656 1614 [99] G. J. Hughes, E. Fearnhill, D. Dunn, S. J. Lycett, A. Rambaut, A. J. L1657 1615 Brown, U. H. D. R. Collaboration, Molecular phylodynamics of the het658 1616 erosexual hiv epidemic in the united kingdom, PLoS pathogens 5 (9)659 1617 (2009) e1000590. 1660 1618 [100] C. M. Romano, I. M. G. de Carvalho- Mello, L. F. Jamal, F. L. de Melo1661 1619 A. Iamarino, M. Motoki, J. R. R. Pinho, E. C. Holmes, P. M. de An4662 1620 drade Zanotto, V. Consortium, et al., Social networks shape the trans4663 mission dynamics of hepatitis c virus, PLoS One 5 (6) (2010) e11170. 1664 1621 [101] P. Sashittal, H. Schmidt, M. M. Chan, B. J. Raphael, Startle: a star homo4665 1623 plasy approach for crisp- cas9 lineage tracing, bioRxiv (2022) 2022- 1266 1624 [102] C. A. Meacham, T. Duncan, The necessity of convex groups in biologi4667 cal classification, Systematic Botany (1987) 78- 90. 1668 1626 [103] S. Moran, S. Snir, Convex recolorings of strings and trees: Definitions1669 hardness results and algorithms, Journal of Computer and System Sci4670 ences 74 (5) (2008) 850- 869. 1671 1629 [104] B. Bollobás, Combinatorics: set systems, hypergraphs, families of vec4672 tors, and combinatorial probability, Cambridge University Press, 1986. 1673 1631 [105] M. C. Golumbic, Algorithmic graph theory and perfect graphs, Vol. 57 1674 Elsevier, 2004. 1675 1633 [106] M. R. Garey, R. L. Graham, D. S. Johnson, D. E. Knuth, Complexity1676 results for bandwidth minimization, SIAM Journal on Applied Mat1677 matics 34 (3) (1978) 477- 495. 1678 1636 [107] C. Dubey, U. Feige, W. Unger, Hardness results for approximating the1679 bandwidth, Journal of Computer and System Sciences 77 (1) (2011) 62- 1680 1638 90. 1681 1639 [108] A. Caprara, F. Malucelli, D. Pretolani, On bandwidth- 2 graphs, Discretese82 + +<|ref|>text<|/ref|><|det|>[[510, 100, 939, 899]]<|/det|> +Applied Mathematics 117 (1- 3) (2002) 1- 13. [109] V. Puller, R. Neher, J. Albert, Estimating time of hiv- 1 infection from next- generation sequence diversity, PLOS Computational Biology 13 (10) (2017) e1005775. [110] M. L. Russell, C. S. Fish, S. Drescher, N. A. Cassidy, P. Chanana, S. Benki- Nugent, J. Slyker, D. Mbori- Ngacha, R. Bosire, B. Richardson, et al., Using viral sequence diversity to estimate time of hiv infection in infants, PLoS Pathogens 19 (12) (2023) e1011861. [111] P. B. I. Baykal, J. Lara, Y. Khudyakov, A. Zelikovsky, P. Skums, Quantitative differences between intra- host hcv populations from persons with recently established and persistent infections, Virus Evolution 6 (2) (2021) vea103. [112] R. Diestel, Graph theory, volume 173 of, Graduate texts in mathematics (2012) 7. [113] P. Kilpeläinen, H. Mannila, Ordered and unordered tree inclusion, SIAM Journal on Computing 24 (2) (1995) 340- 356. [114] J. Matoušek, R. Thomas, On the complexity of finding iso- and other morphisms for partial k- trees, Discrete Mathematics 108 (1- 3) (1992) 343- 364. [115] C. Bron, J. Kerbosch, Algorithm 457: finding all cliques of an undirected graph, Communications of the ACM 16 (9) (1973) 575- 577. [116] L. Lovász, M. D. Plummer, Matching theory, Vol. 367, American Mathematical Soc., 2009. [117] N. Moshiri, M. Ragonnet- Cronin, J. O. Wertheim, S. Mirarab, Favorites: simultaneous simulation of transmission networks, phylogenetic trees and sequences, Bioinformatics 35 (11) (2019) 1852- 1861. [118] A. J. L. Brown, S. J. Lycett, L. Weinert, G. J. Hughes, E. Fearnhill, D. T. Dunn, Transmission network parameters estimated from hiv sequences for a nationwide epidemic, Journal of Infectious Diseases (2011) jir550. [119] S. P. Castillo, R. A. Rebolledo, M. Arim, M. E. Hochberg, P. A. Marquet, Metastatic cells exploit their stoichiometric niche in the network of cancer ecosystems, Science Advances 9 (50) (2023) ead17902. [120] D. Posada, Cellcoal: coalescent simulation of single- cell sequencing samples, Molecular biology and evolution 37 (5) (2020) 1535- 1542. [121] S. Caminiti, I. Finocchi, R. Petreschi, On coding labeled trees, Theoretical computer science 382 (2) (2007) 97- 108. [122] C. Wymant, F. Blanquart, T. Golubchik, A. Gall, M. Bakker, D. Bezemer, N. J. Croucher, M. Hall, M. Hillebrégt, S. H. Ong, et al., Easy and accurate reconstruction of whole hiv genomes from short- read sequence data with shiver, Virus evolution 4 (1) (2018) vev007. [123] A. Stamatakis, Raxml version 8: a tool for phylogenetic analysis and post- analysis of large phylogenies, Bioinformatics 30 (9) (2014) 1312- 1313. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[25, 99, 485, 167]]<|/det|> +1683 [124] A. McPherson, A. Roth, E. Laks, T. Masud, A. Bashashati, A. W. Zhang, G. Ha, J. Biele, D. Yap, A. Wan, et al., Divergent modes of clonal spread and intraperitoneal mixing in high-grade serous ovarian cancer, Nature genetics 48 (7) (2016) 758- 767. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[99, 130, 890, 339]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[243, 351, 746, 364]]<|/det|> +
Figure 8. Comparison of sampled compatible trees with true trees under different constraints.
+ +<|ref|>image<|/ref|><|det|>[[198, 404, 785, 740]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[405, 753, 586, 765]]<|/det|> +
Figure 9. Clone tree for Patient 3
+ +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[61, 177, 620, 191]]<|/det|> +# Algorithm 3 Homomorphism under convexity and sampling parsimony constraints + +<|ref|>text<|/ref|><|det|>[[60, 193, 720, 825]]<|/det|> +Input: phylogenetic tree \(\Psi\) with the root \(\rho\) and a candidate migration tree \(T\) Output: a feasible homomorphism \(f:\Psi \to T\) or the answer that it does not exist. 1: modify \(\Psi\) by contracting paths between leafs with the same label. 2: perform a post- order traversal of the tree \(\Psi\) . 3: for every node \(\alpha\) of the post- order do 4: construct a set of partial homomorphism tokens \(H_{\alpha}\) . 5: if \(\alpha\) is a leaf then 6: \(H_{\alpha}\leftarrow \{(v,0):v\in V(T)\}\) ; 7: end if 8: if \(\alpha\) is an unlabeled internal node with children \(\beta_{1}\) and \(\beta_{2}\) then 9: for all \((v,X)\in H_{\beta_{1}}\) and \((u,Y)\in H_{\beta_{2}}\) do 10: if \(v\sim u\) , \(u\notin X\) and \(N_{T}(u) = Y\cup \{v\}\) then 11: Add \((v,X\cup \{u\})\) to \(H_{\alpha}\) 12: end if 13: if \(v\sim u\) , \(v\notin Y\) and \(N_{T}(v) = X\cup \{u\}\) then 14: Add \((u,Y\cup \{v\})\) to \(H_{\alpha}\) 15: end if 16: end for 17: end if 18: if \(\alpha\) is a labeled internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 19: \(S_{i}\leftarrow \{v:(v,X)\in H_{\beta_{i}}\) and \(|X| = \deg (v) - 1\}\) , \(i = 1,k\) ; 20: Generate the set \(Tr\) of transversals of the set system \((S_{1},\ldots ,S_{k})\) ; 21: for transversals \(V = (v_{1},\ldots ,v_{k})\in Tr\) do 22: if there exists \(v\sim v_{1},\ldots ,v_{k}\) then 23: Add \((v,V)\) to \(H_{\alpha}\) 24: end if 25: end for 26: end if 27: end for 28: if \(H_{\rho}\neq \emptyset\) then 29: perform a pre- order traversal of \(\Psi\) ; 30: for each token \(t_{1},\ldots ,t_{R}\in H_{\rho}\) do 31: assign the token \(t_{r}\) to \(\rho :A S_{\rho}\leftarrow t_{r}\) . 32: for every node \(\alpha\) of the pre- order do 33: \(f(\alpha)\leftarrow v\) , where \((v,X) = A S_{\alpha}\) . 34: if \(\alpha\) is an unlabeled internal node with children \(\beta_{1}\) and \(\beta_{2}\) then 35: select a vertex \(u\in X\) such that \((v,X\setminus \{u\})\in H_{\beta_{1}}\) and \((u,N_{T}(u)\setminus \{v\})\in H_{\beta_{2}}\) . 36: \(A S_{\beta_{1}}\leftarrow (v,X\setminus \{u\})\) and \(A S_{\beta_{2}}\leftarrow (u,N_{T}(u)\setminus \{v\})\) 37: end if 38: if \(\alpha\) is a labeled internal node with children \(\beta_{1},\ldots ,\beta_{k}\) then 39: \(A S_{\beta_{i}}\leftarrow (v_{i},N_{T}(v_{i})\setminus \{v\})\) , \(i = 1,\ldots ,k\) , where \((v,\{v_{1},\ldots ,v_{k}\}) = A S_{\alpha}\) 40: end if 41: end for 42: among generated homomorphisms \(f_{1},\ldots ,f_{R}\) , output the homomorphism \(f_{r}\) with the minimal \(D(f_{r})\) . 43: end for 44: else 45: \(f\) does not exist 46: end if + +<--- Page Split ---> diff --git a/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/images_list.json b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..92c317a385ca605ea02eb6a02576c34838e1985a --- /dev/null +++ b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Trade effects of the RCEP tariff elimination on member countries (%). The figure presents the change rates in multilateral trade for the situation in which trade in goods among the RCEP members have zero tariffs. Two ASEAN countries (Laos and Myanmar) are classified into the rest of the world because input-output tables for Laos and Myanmar are not available. We thus cannot provide results for these two countries. The results for the cases in which the tariffs decline to the committed level in year 1, 5, 10, and 20 after the RCEP enters into force are provided in Supplementary Table 1. The results for any other years are also available upon request.", + "footnote": [], + "bbox": [ + [ + 147, + 272, + 852, + 765 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Sectoral contribution to the welfare changes in RCEP members (%). The left graph presents the sectoral contribution to the aggregate changes in the volume of trade in the case that all trade in goods among RCEP members are duty-free, while the right graph presents the results for terms of trade. For a country, the column-wise summation of all sectors' contributions equals \\(100\\%\\) . The description of the tradable sectors is provided in Table A1 of Supplementary note 1.", + "footnote": [], + "bbox": [ + [ + 147, + 321, + 822, + 650 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 \\(\\mathbf{CO}_2\\) emission burdens of RCEP's tariff reductions. The subplots A, B, and C present the changes and change rates in the amount of \\(\\mathrm{CO_2}\\) emissions emitted by different economies for the cases in which the tariffs within RCEP bloc decline to the level in year 1, 5, and 10 after the RCEP enters into force, respectively. Subplot D presents the corresponding results for the case in which trade in goods among RCEP members is ultimately duty-free. Subplot E gives the ratio of economic welfare change to the \\(\\mathrm{CO_2}\\) emission change rate for RCEP members. Here, Laos and Myanmar are still classified into the rest of the world (RoW) due to the same reason as given in the note of Fig.1.", + "footnote": [], + "bbox": [ + [ + 168, + 351, + 840, + 759 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Increased bilateral \\(\\mathrm{CO_2}\\) emission flows in trade. The graph distinguishes seven regions as the origin (the left) and destination (the right). These are: China, the ASEAN (not including Laos and Myanmar), Japan, South Korea, Australia, New Zealand, and the rest of the world (RoW). The graph presents the increased amount of \\(\\mathrm{CO_2}\\) emitted in the region of origin for its production of exports to a destination.", + "footnote": [], + "bbox": [ + [ + 156, + 170, + 840, + 440 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Sectoral contribution to \\(\\mathbf{CO}_2\\) emissions in trade. The pie charts present for RCEP members the sectoral contribution to their \\(\\mathbf{CO}_2\\) emission changes caused by trade changes. The sector classification is the same as in Fig.2, but here we aggregate the three mining-related sectors (mining energy, other mining, and mining service) into one mining sector.", + "footnote": [], + "bbox": [ + [ + 148, + 163, + 853, + 560 + ] + ], + "page_idx": 17 + } +] \ No newline at end of file diff --git a/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782.mmd b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782.mmd new file mode 100644 index 0000000000000000000000000000000000000000..0b0dc2e558797cceb8542b50074dec24a5ea3bf9 --- /dev/null +++ b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782.mmd @@ -0,0 +1,388 @@ + +# Regional trade agreement burdens global carbon emission mitigation + +Kailan Tian Academy of Mathematics and Systems Science + +Yu Zhang Academy of Mathematics and Systems Science + +Yuze Li Boston University + +Xi Ming University of Chinese Academy of Sciences + +Shangrong Jiang University of Chinese Academy of Sciences https://orcid.org/0000- 0002- 2780- 9075 + +Hongbo Duan (hbduan@ucas.ac.cn) + +University of Chinese Academy of Sciences https://orcid.org/0000- 0001- 5143- 4413 + +Cuihong Yang Institute of Systems Science, China + +Shouyang Wang + +Academy of Mathematics and Systems Science + +## Article + +Keywords: RCEP, Trade and welfare, Carbon mitigation, Input- output analysis, Sectoral linkage + +Posted Date: August 26th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 800470/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on January 20th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28004- 5. + +<--- Page Split ---> + +# Regional trade agreement burdens global carbon emission mitigation + +Kailan Tian1, Yu Zhang1, Yuze Li2, Xi Ming3, Shangrong Jiang3, Hongbo Duan3\*, Cuihong Yang1,3\*, Shouyang Wang1,3\* + +1. Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China +2. Questrom School of Business, Boston University, Boston 02215, USA +3. School of Economics and Management, University of Chinese Academy of Sciences, Beijing 100086, China + +## Abstract + +Regional trade agreements (RTAs) have been widely adopted to facilitate international trade and cross- border investment and ultimately promote economic development. However, ex ante measurements of the environmental effects of RTAs to date have not been well conducted. Here, we estimate the \(\mathrm{CO_2}\) emission burdens of the Regional Comprehensive Economic Partnership (RCEP) after evaluating its economic effects. We find that trade among RCEP member countries will increase significantly and economic output will expand with the reduction of regional tariffs. However, the results show that complete tariff elimination among RCEP members would increase the yearly global \(\mathrm{CO_2}\) emissions from fuel combustion by about \(3.12\%\) , which doubles the annual average growth rate of global \(\mathrm{CO_2}\) emissions in the last decade. The emissions in some developing members will surge. We therefore stress the necessity of balancing carbon mitigation and the pursuit of economic profitability. The technological advancement of emission mitigation and more effective climate policies for international trade are urgently required to avoid undermining international efforts to reduce global emissions. + +<--- Page Split ---> + +## Keywords + +RCEP; Trade and welfare; Carbon mitigation; Input- output analysis; Sectoral linkage + +<--- Page Split ---> + +## Introduction + +Regional trade agreements (RTAs) have been sweeping the world and have become ubiquitous in efforts to facilitate international trade and investment \(^{1 - 3}\) . After an 8- year- long negotiation, the Regional Comprehensive Economic Partnership (RCEP) finally concluded in November 2020 and became the largest RTA in the world in terms of both economic size and population. According to the Schedule of Tariff Commitments in the RCEP Agreement, over \(90\%\) of the trade in goods will eventually have zero tariffs, and most of them will be duty- free immediately or within ten years after the agreement enters into force. Tariff elimination in the region will reduce trade and production costs, resulting in considerable trade- creation and production- boosting effects. + +However, increased international production fragmentation has raised concerns about the trade- climate dilemma (or pollution haven effect) of international trade \(^{4 - 10}\) . That is, international trade increases global or regional emissions if developed economies with cleaner production technology and more stringent environmental policies transfer their polluting industries or production activities to developing countries, leading to emission leakage. Most of the RCEP member countries are typical developing economies that are less emission efficient in their manufacturing industries. In 2018, the amount of \(\mathrm{CO_2}\) emitted by RCEP member countries accounted for a high share \((39.13\%^{11})\) of global \(\mathrm{CO_2}\) emissions from fuel combustion. Therefore, the rapid growth of production activities and trade in an increasing number of less developed nations could impose non- negligible burdens on global and national emission + +<--- Page Split ---> + +mitigation. Previous work deals with the economic effects of RTAs, and the other deals with the environmental side effects of international trade. The first strand of the literature focuses on quantifying the economic welfare effects of RTAs \(^{12 - 17}\) and has largely neglected environmental problems. Conversely, the second strand has substantially accounted ex post for the large carbon flows between countries via international trade \(^{18 - 27}\) . + +In fact, few studies have ex ante quantified the environmental effects of a trade agreement. In this study, we aim to estimate such burdens after evaluating the economic effects of RCEP tariff reductions. The quantification of both RCEP economic gains and environmental burdens has important policy implications for balancing economic development and the responsibility of carbon emission mitigation when implementing RTAs. We estimate how and to what extent RCEP tariff reductions affect trade and economic welfare using a multi- sector and multi- country general equilibrium model \(^{12}\) . It captures sectoral heterogeneity, inter- sectoral linkages, and trade in intermediate products. Due to such features and its tractability, this quantitative trade model has become the workhorse for evaluating the possible consequences of tariff changes in an era of global production networks or global value chains \(^{29,30}\) . + +We find that the RCEP tariff reduction will unleash trade- creation effects, which are determined by the magnitude of tariff reduction, export bundles, trade elasticity, and inter- sectoral linkages in each country. In addition, all member countries' economic welfares will improve. To reveal through what channel welfares are improved, we further decompose the total welfare effects into volume of trade effect and terms of + +<--- Page Split ---> + +trade effect, which are further decomposed separately into the results of trade with RCEP members versus trade with the rest of the world. However, our results also show that complete tariff elimination within the RCEP bloc would increase the yearly global \(\mathrm{CO_2}\) emissions by \(3.12\%\) if the emission intensities ( \(\mathrm{CO_2}\) emissions per unit of output) keep unchanged. Given the fact that the annual average growth rate of global \(\mathrm{CO_2}\) emissions from fuel combustion in the last decade was about \(1.53\%^{11}\) , \(3.12\%\) represents a substantial burden on global \(\mathrm{CO_2}\) emission mitigation. The increased carbon emissions are driven by the rise of production for both domestic expenditures and trade. Trade changes would also increase considerable \(\mathrm{CO_2}\) emissions in the RCEP members. The environmental burdens on some developing countries such as Vietnam and Thailand will surge. As a result, we emphasize that technological advancements in reducing pollutant intensity are urgently required in more developing countries to offset the extra emissions caused by the RCEP. We also suggest that more effective climate policies for international trade should be designed and implemented. + +## Results + +## Economic effects from RCEP tariff elimination + +In this subsection, we evaluate how and to what extent trade and welfare are affected by tariff reductions committed in the RCEP Agreement. Our estimation draws on three types of datasets: input- output (I.O.) tables, bilateral trade flows, and bilateral tariff data. Supplementary note 1 provides detailed descriptions of all the data used in our evaluation. We set the effective applied ad valorem tariff rates in 2019 as the base world + +<--- Page Split ---> + +tariff structure before the RCEP enters into force. The effects of the RCEP tariff changes are evaluated by changing the tariff structure among RCEP member countries and leaving the tariff structure unchanged for countries outside the agreement. + +Fig. 1 presents the aggregate changes in multilateral trade for RCEP member countries when tariffs within the bloc decline to zero after the agreement enter into force. Unsurprisingly, tariff elimination will gradually unleash the agreement's trade- creation effect and substantially strengthen the trade linkages between these countries before the RCEP. We also observe that the trade effects vary across members. First, the RCEP will significantly increase trade between China, Japan, and South Korea. Specifically, China's exports to Japan and South Korea will increase by \(17.61\%\) and \(33.85\%\) , and Japan's exports to China and South Korea will increase by \(29.14\%\) and \(58.62\%\) , respectively. Second, the RCEP will boost more trade for some ASEAN economies. For example, Indonesia's exports to China, South Korea, Thailand, and Vietnam will increase quite markedly by \(109.75\%\) , \(118.79\%\) , \(91.41\%\) , and \(103.03\%\) , respectively. Similar effects will occur for certain other ASEAN economies, such as Malaysia, the Philippines, Thailand, and Vietnam. It is apparent that tariff elimination in the RCEP bloc will improve not only South- North but also South- South trade. The equilibrium model indicates that the trade effects are determined by the magnitude of tariff reduction, export bundles, trade elasticity, and the intermediate input structure for the production of each sector in each country. The divergent effects shown in Fig. 1 are the complete results of these differently weighted factors. Another observation is that some bilateral trade (e.g., the exports of Singapore to Australia and Japan) will decline + +<--- Page Split ---> + +after the RCEP enters into force. This negative effect arises from trade diversion. For example, Singapore was already an almost entirely free-trade country before the RCEP. The RCEP tariff reduction substantially reduces the costs of trade with other members and raises their relative competitiveness. This process facilitates trade diversion from Singapore. + +![](images/Figure_1.jpg) + +
Fig. 1 Trade effects of the RCEP tariff elimination on member countries (%). The figure presents the change rates in multilateral trade for the situation in which trade in goods among the RCEP members have zero tariffs. Two ASEAN countries (Laos and Myanmar) are classified into the rest of the world because input-output tables for Laos and Myanmar are not available. We thus cannot provide results for these two countries. The results for the cases in which the tariffs decline to the committed level in year 1, 5, 10, and 20 after the RCEP enters into force are provided in Supplementary Table 1. The results for any other years are also available upon request.
+ +<--- Page Split ---> + +Table 1 presents the results of welfare effects for RCEP members. It shows that all members benefit from the RCEP tariff reductions, with most small countries gaining more than large ones. The first column presents that all RCEP members' real wages will increase, with Cambodia increasing the most. The second column shows that the welfare (the summation of labour income, tariff revenues, and trade deficits) of Vietnam, Cambodia, and Singapore will increase the most, by \(15.63\%\) , \(8.54\%\) , and \(3.79\%\) , respectively. The effects for large economies such as China and Japan are smaller. + +Table 1 Welfare effects of the RCEP tariff elimination on member countries + +
MembersReal wageWelfare
TotalVolume of trade
RCEPRoWTerms of trade RCEP RoW
Australia100.25%0.20%0.00%0.00%0.12%0.07%
China100.36%0.34%0.16%0.01%0.08%0.09%
Japan100.28%0.28%0.05%0.00%0.09%0.13%
South Korea102.38%3.28%4.51%-0.53%-0.43%-0.28%
New Zealand100.38%0.47%0.01%0.01%0.27%0.18%
ASEAN
Brunei100.47%0.59%0.00%0.00%0.48%0.11%
Cambodia111.86%8.54%5.19%-0.53%1.75%2.13%
Indonesia100.88%0.81%0.36%0.03%0.23%0.19%
Malaysia104.68%1.31%1.18%-0.13%0.11%0.14%
Philippines102.01%0.52%0.53%-0.03%-0.04%0.06%
Singapore103.52%3.79%0.00%0.00%1.71%2.08%
Thailand103.15%1.82%2.50%-0.19%-0.42%-0.07%
Vietnam105.60%15.63%10.17%-0.38%3.14%2.70%
+ +Note: The table presents the changes in real wage and welfare for RCEP members if trade in goods among the members becomes duty-free. The total welfare effects are decomposed into volume of trade effect and terms of trade effect, which are further decomposed separately into the results of trade with RCEP members (columns 3 and 5) versus trade with the rest of the world (RoW, column 4 and 6). + +To reveal how each member's welfare is improved, we decompose the total welfare effects into volume of trade effect (columns 3 and 4) and terms of trade effect (columns 5 and 6), which are further decomposed separately into the results of trade + +<--- Page Split ---> + +with RCEP members versus trade with other economies outside the RCEP Agreement (the rest of the world, RoW). Columns 3 and 4 show that trade with RCEP members rather than the RoW is the most important contributor to the increase in all members' volume of trade. Comparing with the results in columns 5 and 6, we can find that creating more trade within the RCEP bloc also makes the most significant contribution to the increase in welfare for member countries—South Korea, Cambodia, Malaysia, the Philippines, Thailand, and Vietnam. Additionally, trade reductions with the RoW generate slightly negative welfare effects for most members. The reason for such negative effects stems from the RCEP diverting trade from non- RCEP member countries. + +Columns 5 and 6 show that the aggregate terms of trade for almost all members improve, whereas South Korea's and Thailand's terms of trade deteriorate slightly. This differential performance can be attributed to the divergent changes in the export prices of each country, as terms of trade compare the price of a country's export with the price of its import. The model shows that export prices are determined by the unit costs of input bundles, namely, the combination of labour costs (i.e., wages) and the prices of intermediate inputs. Column 1 shows that the RCEP will increase the real wages of all members, which increases export prices. However, other things being equal, the prices of intermediate inputs will decline with the reductions in tariffs on imported intermediates. Such effects can further be propagated through input- output linkages. As a result, the change in the prices of intermediate inputs decreases export prices. Ultimately, for most RCEP members, the increase in real wages is larger than the + +<--- Page Split ---> + +decrease in the prices of intermediate inputs, which results in a positive effect on terms of trade. For other economies, the contrary is the case. + +Supplementary Table 2 provides the welfare effects of RCEP tariff reductions on economies outside this agreement. The effects are twofold. On the one hand, as explained above, the agreement generates trade diversion towards RCEP members. On the other hand, in an era of increased international fragmentation, the production activities of trade products in countries outside the RCEP require intermediate inputs from RCEP members. The decreased prices of such intermediate products due to RCEP tariff reductions also change the production costs, export prices, and terms of trade in economies outside the RCEP. As a result, the effects are negative for some non- RCEP economies but positive for others. However, the impact is very small. + +For each RCEP member, we calculate the sectoral contribution to its aggregated change in volume of trade and terms of trade. First, as shown in Fig. 2, the contribution varies considerably across countries and sectors. For example, mining is the sector with the most significant contribution to the change in Australia's terms of trade, whereas electrical equipment is the greatest contributor for China and Japan. Second, agriculture (and food products for some countries) makes a significant contribution to changes in the volume of trade in many countries, including China, Japan, South Korea, New Zealand, the Philippines, and Thailand. The reason is that agriculture is strongly protected in most countries, as the current tariffs are relatively high. For example, the average tariffs applied by Japan to its imported food products and agriculture in 2019 were \(10.74\%\) and \(5.84\%\) , respectively, which were the largest and second- largest import + +<--- Page Split ---> + +tariffs among all Japanese goods sectors. Agriculture (and food products) is a homogeneous goods sector with high import tariff trade elasticity. A slight reduction in the tariffs for this sector can improve the trade volume considerably because it is relatively easy to change suppliers. Petroleum in some countries is similarly affected for similar reasons. + +![](images/Figure_2.jpg) + +
Fig. 2 Sectoral contribution to the welfare changes in RCEP members (%). The left graph presents the sectoral contribution to the aggregate changes in the volume of trade in the case that all trade in goods among RCEP members are duty-free, while the right graph presents the results for terms of trade. For a country, the column-wise summation of all sectors' contributions equals \(100\%\) . The description of the tradable sectors is provided in Table A1 of Supplementary note 1.
+ +Another notable observation is that a handful of sectors – electrical equipment, machinery and equipment, and motor vehicles – explain a high proportion of the changes in terms of trade in many countries (China, Japan, South Korea, and some ASEAN economies). This is the combined result of tariff reduction, the share of + +<--- Page Split ---> + +intermediate inputs required in the production process, and inter- sectoral linkages. Although the tariff reductions in these technology- intensive sectors are not the largest, they are considerable for some countries. More importantly, these sectors use a considerably larger share of intermediate inputs in production than other sectors. They also have stronger input- output linkages with other sectors. Therefore, a decrease in the unit production costs in these sectors has a larger multiplicative effect and thus a larger impact on terms of trade. + +## Carbon emission burdens of RCEP tariff reductions + +The synergy of tariff elimination within the RCEP bloc substantially reduces the costs of intra- regional trade and production and thus increases the outputs of the member countries. As a result, carbon emissions will also increase significantly if the emission intensities (emissions per unit of output) do not decrease enough to offset the extra emissions caused by the increase in production outputs. Assuming the emission intensities of all countries stay at the same level as that in the year 2015, the global \(\mathrm{CO_2}\) emissions from fuel combustion would increase by 251.35 million tonnes (Mt; \(0.75\%\) compared to the amount in 2018) when the tariff structure changes to that specified for the first year after the RCEP enters into force. When the tariffs continue to decline to the level in the fifth and tenth years and then to zero, global \(\mathrm{CO_2}\) emissions will increase by 463.72 Mt (1.38%), 756.38 Mt (2.26%), and 1046.45 Mt (3.12%), respectively. Recalling that global \(\mathrm{CO_2}\) emissions grew at an annual average rate of \(1.53\%^{11}\) in the last decade, these results indicate substantial burdens on carbon emission mitigation. + +<--- Page Split ---> + +The increased \(\mathrm{CO_2}\) will be emitted mainly by RCEP member countries. In the situation of all trade in goods in the RCEP region becoming duty- free, the production of RCEP members would increase emissions by 789.05 Mt \(\mathrm{CO_2}\) (subplot D in Fig. 3), accounting for \(75.40\%\) of the increased global emissions. Amongst, Mainland China will be the largest contributor in terms of absolute value (495.73 Mt \(\mathrm{CO_2}\) , \(47.37\%\) of the increased global emissions), followed by the ASEAN economies (164.72 Mt \(\mathrm{CO_2}\) , \(15.74\%\) ) and Japan (52.66 Mt \(\mathrm{CO_2}\) ). In terms of magnitude, Vietnam will increase the most ( \(16.51\%\) ), followed by Malaysia ( \(16.09\%\) ) and Thailand ( \(13.58\%\) ). The results indicate that the emission intensities in these countries must decrease by the same magnitude to ensure that the \(\mathrm{CO_2}\) emissions do not increase. The magnitude of the decrease should be larger if countries aim to reduce their emissions. Such an ambitious target could be a non- negligible burden, especially for some developing ASEAN economies. + +In subplot E of Fig. 3, we also present the ratio of welfare change to the \(\mathrm{CO_2}\) emission change rate for RCEP members. The ratio gives the welfare gains at the cost of a \(1\%\) increase in carbon emissions. In terms of this ratio, Vietnam, Singapore, and Cambodia are the greatest gainers, whereas China and Japan rank at the lower end. + +The increased carbon emissions are driven by the rise of production for both domestic expenditures and trade. As mentioned above, the RCEP tariff reductions bring mainly trade- creation effects within the RCEP bloc and cause trade diversion between RCEP members and non- RCEP economies. An RCEP member may emit more \(\mathrm{CO_2}\) in its increased trade with other RCEP members and reduce carbon emissions because of + +<--- Page Split ---> + +its decreased trade with non- RCEP economies. We employ the environmentally extended inter- country input- output (ICIO) model to account for the comprehensive carbon emission changes due to trade changes. The results show that trade changes in the case of all trade in goods within the RCEP bloc becoming duty- free would increase \(\mathrm{CO_2}\) emissions for China, the ASEAN countries, South Korea, Japan, Australia, and New Zealand by 130.17 Mt, 70.36 Mt, 27.24 Mt, 22.52 Mt, 4.82 Mt, and 0.52 Mt, respectively. + +![](images/Figure_3.jpg) + +
Fig. 3 \(\mathbf{CO}_2\) emission burdens of RCEP's tariff reductions. The subplots A, B, and C present the changes and change rates in the amount of \(\mathrm{CO_2}\) emissions emitted by different economies for the cases in which the tariffs within RCEP bloc decline to the level in year 1, 5, and 10 after the RCEP enters into force, respectively. Subplot D presents the corresponding results for the case in which trade in goods among RCEP members is ultimately duty-free. Subplot E gives the ratio of economic welfare change to the \(\mathrm{CO_2}\) emission change rate for RCEP members. Here, Laos and Myanmar are still classified into the rest of the world (RoW) due to the same reason as given in the note of Fig.1.
+ +<--- Page Split ---> + +We also find that trade changes slightly increase the amount of \(\mathrm{CO_2}\) (61.19 Mt) emitted by RoW (non- RCEP economies). The emission effects of RCEP on non- RCEP economies are twofold. On the one hand, RCEP tariff reductions reduce the direct exports of some non- RCEP economies to RCEP members for reasons we discussed in the previous section. This will reduce the emissions of some non- RCEP economies. On the other hand, the increased production for trade within the RCEP bloc requires more intermediate inputs from some economies outside the RCEP. The economic activities associated with the increased production of such intermediate products generate more \(\mathrm{CO_2}\) emissions in non- RCEP economies. Due to the increased international production fragmentation, the emissions generated by these indirect linkages can be substantial. The fact that the RCEP increases overall emissions by economies outside indicates that the indirect effects are larger than the direct effects. + +Fig. 4 visualizes increased \(\mathrm{CO_2}\) emissions due to changes in bilateral trade flows, which tells us who emits increased \(\mathrm{CO_2}\) for whom. It shows the amount of \(\mathrm{CO_2}\) emitted by a region of origin for the production of its increased exports to the destination. The largest flow is 44.28 Mt \(\mathrm{CO_2}\) for China's increased exports to the ASEAN countries. Other large flows include emissions for the increased exports of the ASEAN countries to China (40.24 Mt \(\mathrm{CO_2}\) ), China to Japan (36.07 Mt \(\mathrm{CO_2}\) ), and China to South Korea (27.83 Mt \(\mathrm{CO_2}\) ). Fig. 4 also shows that the increase in exports of Japan (20.40 Mt \(\mathrm{CO_2}\) ) and South Korea (29.33 Mt \(\mathrm{CO_2}\) ) generates a relatively small increase in the \(\mathrm{CO_2}\) emissions of these countries. However, their imports generate considerable \(\mathrm{CO_2}\) emission increases (57.83 and 60.29 Mt \(\mathrm{CO_2}\) ) in the source countries. The results + +<--- Page Split ---> + +indicate that developing RCEP members including China and some ASEAN economies will emit increasing \(\mathrm{CO_2}\) for the developed members. + +![](images/Figure_4.jpg) + +
Fig. 4 Increased bilateral \(\mathrm{CO_2}\) emission flows in trade. The graph distinguishes seven regions as the origin (the left) and destination (the right). These are: China, the ASEAN (not including Laos and Myanmar), Japan, South Korea, Australia, New Zealand, and the rest of the world (RoW). The graph presents the increased amount of \(\mathrm{CO_2}\) emitted in the region of origin for its production of exports to a destination.
+ +Fig. 5 shows the sectoral contribution to aggregate \(\mathrm{CO_2}\) emission changes generated by trade changes for RCEP members. The exports of electrical equipment contribute the most to China \((14.53\%)\) , the ASEAN countries \((12.73\%)\) , Japan \((18.72\%)\) , and South Korea \((23.65\%)\) . The other two large contributors are machinery and equipment (second largest for China and the ASEAN countries, third- largest for Japan and South Korea) and computer, electronic, and optical products (second for South Korea, third for China, and fifth for Japan). The motor vehicle industry makes the second- largest contribution \((17.75\%)\) to Japan, but its contribution to other countries is relatively small, indicating Japan's strong comparative advantage in this industry. For Australia and New Zealand, the increased trade generates a very small increase in + +<--- Page Split ---> + +emissions. The major contributors are mining for Australia and agriculture for New Zealand. + +![](images/Figure_5.jpg) + +
Fig. 5 Sectoral contribution to \(\mathbf{CO}_2\) emissions in trade. The pie charts present for RCEP members the sectoral contribution to their \(\mathbf{CO}_2\) emission changes caused by trade changes. The sector classification is the same as in Fig.2, but here we aggregate the three mining-related sectors (mining energy, other mining, and mining service) into one mining sector.
+ +The finding that RTAs increase participants' economic welfare at the cost of environmental burdens can be explained by the fact that in real economic interactions, what, how much, and with whom to trade are still determined based on economic profitability rather than environmental considerations. In a Ricardian world, a country exports more products in which it has a comparative advantage in terms of production. The advantage is defined as using fewer of the resources under consideration. Traditionally, labour was such a resource, and additional trade generated from lower + +<--- Page Split ---> + +trade costs leads to increased welfare and emissions in each of the trading countries. Alternatively, if we consider environmental aspects, the story changes. For instance, if we define the advantage as generating fewer emissions in producing a certain product34, increased trade in this product will reduce emissions in both countries. + +## Discussion + +The world trade system has been seriously undermined due to huge shocks, such as the COVID- 19 global pandemic, the United States- China trade conflict, and Brexit, which to some extent have stimulated the formation of more RTAs. In this paper, we estimate the economic gains and the corresponding carbon emission burdens of the RCEP. The results show that RCEP tariff elimination will substantially reduce intra- regional trade costs and product prices, increase the comparative advantage of regional products, and ultimately improve the welfare of all member countries. Meanwhile, we point out that policymakers should pay attention to the environmental impacts of the RCEP since our results indicate non- negligible increases in potential \(\mathrm{CO_2}\) emissions caused by the RCEP. However, we emphasize that anti- globalization is far from a possible strategy for global emission mitigation. Despite de- globalization could reduce international trade and the corresponding embodied carbon emissions in the short term35, 36, it harms the economic welfare of all countries and threatens international efforts to fight climate change in the longer run. Returning to autarky cuts off developing countries' opportunities to participate in global production networks and then upgrade their emission technologies through learning- by- doing. In addition, anti- globalization will + +<--- Page Split ---> + +hinder international cooperation that aims to mitigate global emissions \(^{37}\) . + +Instead, we emphasize that technological advancements in reducing pollutant intensity are urgently required in more developing countries. We find that the carbon emission intensities (CO₂ emissions per unit of output) in all member countries should decline to offset the extra emissions caused by the RCEP, particularly for the developing member countries. Our calculations based on data for 2015 show that the emission intensities of China (2.83 times) and most ASEAN economies, such as Vietnam (2.25 times), Malaysia (2.11 times), and Thailand (2.04 times), were more than twice that of Japan. On the one hand, developing nations should make efforts (e.g., by strengthening technological research and development) to reduce their emission intensity gap with developed nations. For example, at the Paris Climate Change Conference in 2015, China committed to achieving a peak in its carbon emissions by 2030 and reducing its emission intensity substantially. In 2021, China pledged to achieve carbon neutrality before 2060, which is largely consistent with the \(1.5^{\circ}\mathrm{C}\) warming limit \(^{38}\) . The country is making greater efforts and taking more effective measures to achieve this challenging goal, including improving technologies, transforming its energy structure, and developing a circular economy. On the other hand, it is also crucial for developed RCEP members to facilitate the transfer of cleaner production technologies to the developing members when implementing the trade agreement. This process will help accelerate improvements in emission performance in developing countries. + +Our findings also suggest that more effective climate policies for international trade should be designed and implemented as we find that some members will emit + +<--- Page Split ---> + +increasing \(\mathrm{CO_2}\) for other countries. The premise is a robust and fair accounting system to assign responsibility for internationally traded emissions. The production- based accounting (PBA) system is the currently widely adopted scheme in practice, but it ignores carbon leakages through international trade. Consumption- based accounting (CBA) considers the emissions embodied in trade but fails to stimulate the producing countries to clean up their export industries. \(^{39}\) Researchers \(^{34, 39 - 42}\) have proposed adjusted accounting systems to share producer and consumer responsibility \(^{43 - 45}\) by crediting trade that reduces global emissions and penalizing trade that increases global emissions. Despite these efforts in academia, national and global climate policies have so far not adopted such systems of credits and penalties. Among a long list of researchers, we call for the idea of countries working together to design robust and environmentally fair accounting tools, develop and implement effective global and regional climate policies, and share the responsibility of global emission mitigation \(^{46 - 49}\) . + +This study has potential extensions that are worthy of pursuit. We currently do not consider the influence mechanism by which the barriers in services trade and investment in the RCEP region will also decrease under the agreement. Both international trade and cross- border investment can influence a country's economic welfare and environmental issues. Although global flows of foreign direct investment (FDI) shrank in recent years \(^{50, 51}\) and fell sharply by one third \(^{52}\) in 2020 as a result of the COVID- 19 pandemic, FDI among RCEP members will likely continuously increase after the RCEP enters into force. Increasing FDI may facilitate relocating some climate- + +<--- Page Split ---> + +unfriendly industries or production activities from industrialized RCEP members to developing countries \(^{23,24}\) . As a result, FDI may magnify the environmental burdens on developing RCEP members. Future studies are expected to provide quantitative analyses of the effects of qualitative cross- border investment rules in the RCEP Agreement on the volume and direction of FDI flows. More in- depth analyses are also expected to allocate the carbon footprints of FDI flows and explore the scheme of sharing environmental responsibility between FDI home and host countries. + +## Methods + +## Quantifying the economic effects + +This section outlines the model that we employ to quantify the effects of RCEP tariff reductions on trade and welfare. Here, we provide a brief introduction and the main equations. A more detailed description of the model can be found in Caliendo and Parro (2015) \(^{12}\) . The world consists of \(n\) countries and there are \(m\) sectors in each country. Countries are denoted by \(s\) and \(r\) and sectors by \(i\) and \(j\) . The households in country \(r\) derive utility from consuming final products \(C_{r}^{j}\) . The function is Cobb- Douglas and given by + +\[u(C_{r}) = \prod_{j = 1}^{m}C_{r}^{j\alpha_{r}^{j}},\mathrm{where}\Sigma_{j = 1}^{m}\alpha_{r}^{j} = 1. \quad (1)\] + +There is a continuum of intermediate products \(\omega^{j}\) produced in sector \(j\) . Primary inputs (labour) and a bundle of intermediate inputs from all sectors are used for the production of \(\omega^{j}\) in country \(r\) . The production technology is + +\[q_{r}^{j}(\omega^{j}) = z_{r}^{j}(\omega^{j})[lab_{r}^{j}(\omega^{j})]^{r / r}\prod_{i = 1}^{m}[int_{i}^{r,j}(\omega^{j})]^{r / r},\] + +where \(z_{r}^{j}(\omega^{j})\) denotes the efficiency in producing \(\omega^{j}\) in country \(r\) . \(lab_{r}^{j}(\omega^{j})\) is labour and + +<--- Page Split ---> + +\(i n t_{r}^{i,j}(\omega^{j})\) are the intermediate inputs from sector \(i\) required in the production of \(\omega^{j}\) . \(\gamma_{r}^{j}\) denotes the share of value- added in production output, and \(\gamma_{r}^{i,j}\) denotes the share of products from sector \(i\) used as intermediate inputs in the production of \(\omega^{j}\) , with \(\sum_{i = 1}^{m}\gamma_{r}^{i,j} + \gamma_{r}^{j} = 1\) . The cost of an input bundle is given by + +\[c_{r}^{j} = B_{r}^{j}w_{r}\gamma_{r}^{j}\prod_{i = 1}^{m}P_{i}^{j\gamma_{r}^{i,j}}, \quad (2)\] + +where \(w_{r}\) denotes the wage rate, \(P_{i}^{j}\) gives the price of intermediate inputs from sector \(i\) , and \(B_{n}^{j}\) is a constant. \(c_{r}^{j}\) clearly incorporates all the inter- sectoral linkages which can be obtained from input- output tables. + +In an open economy, producers minimize their production costs and purchase intermediate products from suppliers across countries. However, trade is costly. We denote \(k_{r s}^{j}\) as the bilateral trade cost for country \(r\) 's imports of sector \(j\) products shipped from country \(s\) . It consists of an ad- . valorem tariff \((\tau_{r s}^{j})\) and iceberg trade cost \((d_{r s}^{j})\) \(k_{r s}^{j} = (1 + \tau_{r s}^{j})d_{r s}^{j}\) . Using Eaton and Kortum's (2002)28 representation of technologies which allows production efficiency to distribute Frechet, we can derive the price of the intermediate product as + +\[P_{r}^{j} = G^{j}\left[\sum_{s = 1}^{n}\lambda_{s}^{j}\left(c_{s}^{j}k_{r s}^{j}\right)^{-\theta^{j}}\right]^{-1 / \theta^{j}}, \quad (3)\] + +where \(G^{j}\) is a constant, \(\lambda_{s}^{j}\) reflects absolute advantage as a higher value indicates more likely a draw of high efficiency, and \(\theta^{j}\) captures comparative advantage as a lower value indicates a higher dispersion of efficiency. \(\lambda_{s}^{j}\) and \(\theta^{j}\) reflect Ricardian trade53. + +The properties of Frechet distribution further enable us to derive the bilateral trade share as + +\[\pi_{r s}^{j} = \frac{\lambda_{s}^{j}\left[c_{r}^{j}k_{r s}^{j}\right]^{-\theta^{j}}}{\sum_{h = 1}^{n}\lambda_{h}^{j}\left[c_{h}^{j}k_{r h}^{j}\right]^{-\theta^{j}}}. \quad (4)\] + +As shown, any changes in tariffs \((\tau_{r s}^{j})\) can affect trade costs \((k_{r s}^{j})\) and thus directly affect trade shares. Equations (2) and (3) show that changes in tariffs also affect the cost of input bundle \((c_{s}^{j})\) + +<--- Page Split ---> + +and thus have an indirect effect on trade. + +The total expenditure on the products of sector \(j\) in country \(r\) is the summation of firms' expenditures on intermediate products and households' expenditures on final products. It is given by + +\[X_{r}^{j} = \sum_{i = 1}^{m}\gamma_{r}^{j,i}\sum_{s = 1}^{n}X_{s}^{i}\frac{\pi_{sr}^{i}}{1 + \tau_{sr}^{i}} +\alpha_{r}^{j}I_{r}, \quad (5)\] + +where + +\[I_{r} = w_{r}L_{r} + R_{r} + D_{r} \quad (6)\] + +represents the total household income in country \(r\) , i.e., the sum of labour income \((w_{r}L_{r})\) , tariff revenues \((R_{r})\) and trade deficits \((D_{r})\) . In particular, \(R_{n} = \sum_{j = 1}^{m}\sum_{s = 1}^{n}\tau_{rs}^{j}M_{rs}^{j}\) , where \(M_{rs}^{j} = X_{r}^{j}\frac{\pi_{rs}^{j}}{1 + \tau_{rs}^{j}}\) is country \(r\) 's import of sector \(j\) products from country \(s\) . The trade deficit of a country is the summation of sectoral deficits, \(D_{r} = \sum_{j = 1}^{m}D_{r}^{j}\) , and sectoral deficit is given by \(D_{r}^{j} = \sum_{s = 1}^{n}M_{rs}^{j} - \sum_{s = 1}^{n}E_{rs}^{j}\) , where \(E_{rs}^{j} = X_{s}^{j}\frac{\pi_{sr}^{j}}{1 + \tau_{sr}^{j}}\) . + +The next step is to solve for changes in wages and prices given that the tariff structure \(\tau\) is changed to \(\tau '\) . Instead of solving for two equilibria under \(\tau\) and \(\tau '\) , Caliendo and Parro (2015) \(^{12}\) propose solving for an equilibrium in relative changes so that it is not necessary to estimate some parameters that are difficult to identify. Let a variable with a circumflex “\(\hat{x}\) ” denote its relative change. The equilibrium in relative changes satisfies the following conditions: + +\[\begin{array}{c}{\hat{k}_{r s}^{j} = (1 + \tau_{r s}^{j}) / (1 + \tau_{r s}^{j})}\\ {\hat{c}_{r}^{j} = \hat{w}_{r}\gamma_{r}^{j}\prod_{i = 1}^{m}\hat{P}_{r}^{i}\gamma_{r}^{i,j}}\\ {\hat{P}_{r}^{j} = \left[\sum_{s = 1}^{n}\pi_{r s}^{j}\left(\hat{c}_{s}^{j}\hat{k}_{r s}^{j}\right)^{-\theta j}\right]^{-1 / \theta j}}\\ {\hat{\pi}_{r s}^{j} = \left[\frac{\hat{c}_{s}^{j}\hat{k}_{r s}^{j}}{\hat{P}_{r}^{j}}\right]^{-\theta j}} \end{array} \quad (7)\] + +<--- Page Split ---> + +\[\Sigma_{j = 1}^{m}\Sigma_{s = 1}^{n}X_{r}^{j^{\prime}}\frac{\pi_{r s}^{j^{\prime}}}{1 + \tau_{r s}^{j^{\prime}}} - D_{r}^{\prime} = \Sigma_{j = 1}^{m}\Sigma_{s = 1}^{n}X_{s}^{j^{\prime}}\frac{\pi_{r s}^{j^{\prime}}}{1 + \tau_{r s}^{j^{\prime}}} \quad (12)\] + +\[I_{r}^{\prime} = \widehat{w}_{r}w_{r}L_{r} + R_{r}^{\prime} + D_{r}^{\prime}. \quad (13)\] + +Introducing into the model the changes in tariff structure, we can solve for changes in (total and bilateral) trade flows, real wages \((w_{r} / P_{r})\) , welfare \((I_{r} / P_{r})\) , and production output for each country. The change in welfare can be decomposed into volume of trade effect and terms of trade effect. The welfare change can also be calculated at both the bilateral and sectoral levels. We can calculate the change in volume of trade and terms of trade between country \(s\) and \(r\) , and the change in a specific sector \(j\) of country \(r\) . + +## Accounting for the carbon emission changes + +The RCEP tariff reductions lead to changes in multilateral trade flows resulting in trade- related carbon emissions changes. We adopt the environmentally extended ICIO model (see Table A2 in Supplementary note 1 for the stylized table) to account for the carbon emission changes. Denote the following as the flows of final products among different countries: + + + +And, + +\[\mathbf{A} = \left[ \begin{array}{llll}\mathbf{A}^{11} & \dots & \mathbf{A}^{1r} & \dots & \mathbf{A}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \mathbf{A}^{r1} & \dots & \mathbf{A}^{rr} & \dots & \mathbf{A}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \mathbf{A}^{n1} & \dots & \mathbf{A}^{nr} & \dots & \mathbf{A}^{nn} \end{array} \right]\] + +is the global direct input- output coefficient matrix. Its typical element \(a_{ij}^{sr}\) provides the intermediate input from sector \(i (= 1,\dots ,m)\) in country \(s (= 1,\dots ,n)\) used by sector \(j (= 1,\dots ,m)\) in country \(r (= 1,\dots ,n)\) for producing one unit of output. \(\mathbf{A}^{rr}\) and \(\mathbf{f}^{rr}\) provide intra + +<--- Page Split ---> + +country flows of intermediate products and final products. \(\mathbf{A}^{sr}\) and \(\mathbf{f}^{sr}\) \((s\neq r)\) represent trade in intermediate products and trade in final products, respectively. According to the standard input-output model \(^{31}\) , the gross output vector \(\mathbf{y}\) is + +\[\mathbf{y} = (\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}, \quad (14)\] + +where \(\mathbf{I}\) is an \((nm \times nm)\) identity matrix, and \(\mathbf{u}\) is a summation vector of appropriate length with all elements being ones. + +Let \(\mathbf{w}\) be the \(\mathrm{CO}_{2}\) emission coefficient vector, the elements of which provide the emissions per unit of output. Then, the \(\mathrm{CO}_{2}\) emission vector \(\mathbf{e}\) can be written as + +\[\mathbf{e}\mathbf{u} = \mathbf{w}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}. \quad (15)\] + +The left side of (15) equals the summation of \(\mathrm{CO}_{2}\) emissions in all sectors in the world, which equals global \(\mathrm{CO}_{2}\) emissions. This equation can be adapted to allow us to calculate the \(\mathrm{CO}_{2}\) emissions \((\mathbf{e}^{r})\) in a specific country \(r\) . This can be obtained by replacing the vector \(\mathbf{w}\) in (15) with a vector \(\mathbf{w}^{r}\) . The new vector has equal length, but only the \(\mathrm{CO}_{2}\) emission coefficients for the sectors in country \(r\) are retained while all other elements are set as zeros. This yields + +\[\mathbf{e}^{r} = \mathbf{w}^{r}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}. \quad (16)\] + +Equation (16) allows us to calculate, for example, the part of Thailand's \(\mathrm{CO}_{2}\) emissions that are generated by the exports of final products from Japan to final users in China. The production processes for trade between Japan and China may consume intermediate products from Thailand, of which the production emits \(\mathrm{CO}_{2}\) in Thailand. Using the global Leontief inverse, we can take fully into account these indirect effects in equation (16). + +To calculate the \(\mathrm{CO}_{2}\) emission changes in country \(r\) , we compare two situations. The first is the actual situation, and the second is the case in which multilateral trade flows are changed due to + +<--- Page Split ---> + +RCEP tariff reductions. Moving forward from equation (16), we develop the following equation: + +\[\Delta \mathbf{e}^{r} = \mathbf{w}^{r}\big(\mathbf{I} - \widetilde{\mathbf{A}}\big)^{-1}\widetilde{\mathbf{F}}\mathbf{u} - \mathbf{w}^{r}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}, \quad (17)\] + +where + +\[\widetilde{\mathbf{A}} = \left[ \begin{array}{cccc}\mathbf{A}^{11} & \dots & \widetilde{\mathbf{A}}^{1r} & \dots & \widetilde{\mathbf{A}}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{A}}^{r1} & \dots & \mathbf{A}^{rr} & \dots & \widetilde{\mathbf{A}}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{A}}^{n1} & \dots & \widetilde{\mathbf{A}}^{nr} & \dots & \mathbf{A}^{nn} \end{array} \right]\mathrm{and}\widetilde{\mathbf{F}} = \left[ \begin{array}{cccc}\mathbf{f}^{11} & \dots & \widetilde{\mathbf{f}}^{1r} & \dots & \widetilde{\mathbf{f}}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{f}}^{r1} & \dots & \mathbf{f}^{rr} & \dots & \widetilde{\mathbf{f}}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{f}}^{n1} & \dots & \widetilde{\mathbf{f}}^{nr} & \dots & \mathbf{f}^{nn} \end{array} \right].\] + +Equation (17) enables us to consider the changes in both trade in final products and trade in intermediate products. The changes in bilateral trade flows can be obtained at the sectoral level after solving the Caliendo and Parro (2015) \(^{12}\) model described above. However, this model does not enable us to calculate the respective magnitudes of change for final product trade and intermediate product trade. We thus use the same magnitude when conducting the calculation. + +## Data availability + +National and ICIO tables are from the most recent OECD Input- Output Database \(^{54}\) (2018 edition, https://stats.oecd.org/), and bilateral trade flows are obtained from the United Nations Commodity Trade (U.N. Comtrade) database (https://comtrade.un.org/data/). Bilateral tariff data are from World Integrated Trade Solution (WITS) software (https://wits.worldbank.org/). The committed tariff reductions among RCEP parties come from the Schedule of Tariff Commitments in the RCEP Agreement. The schedule provides detailed data for each RCEP party's commitments to tariff reduction for each year after the date of the entry into force of the RCEP Agreement. International Energy Agency (https://www.iea.org/data- and- statistics) provides sectoral \(\mathrm{CO}_{2}\) emissions data \(^{11}\) , and the most recent available data are for 2018. We include maximum number of economies, conditional on obtaining reliable data. We ultimately obtain 60 economies and a constructed RoW + +<--- Page Split ---> + +with 36 sectors in each economy. Trade and tariff data are for 2019, and we employ the most recent available input- output tables for 2015, assuming that input- output coefficients in 2019 are not much different than those in 2015. Supplementary note 1 provides more detailed descriptions of all data used in our evaluation. Supplementary Tables 1 and 2 provide additional results. All datasets generated in this study are available upon request. + +## References + +1. Baldwin, R. & Jaimovich, D. Are free trade agreements contagious? J. Int. Econ. 88(1), 1–16 (2012). +2. Mahadevan, R. & Nugroho, A. Can the Regional Comprehensive Economic Partnership minimise the harm from the United States–China trade war? World Econ. 42(11), 3148–3167 (2019). +3. Bhagwati, J. Termites in the trading system: How preferential agreements undermine free trade (Oxford University Press, 2008). +4. Liu, Z. et al. Targeted opportunities to address the climate–trade dilemma in China. Nat. Clim. Chang. 145, 143–145 (2015). +5. Zhang, Q. et al. Transboundary health impacts of transported global air pollution and international trade. Nature 543, 705–709 (2017). +6. Peters, G. P., Minx, J. C., Weber, C. L. & Edenhofer, O. Growth in emission transfers via international trade from 1990 to 2008. Proc. Natl. Acad. Sci. USA 108(21), 8903–8908 (2011). +7. Guan, D. B. et al. Structural decline in China’s CO₂ emissions through transitions in industry and energy systems. Nat. Geosci. 11(8), 551–555 (2018). +8. Mi, Z. F. et al. Chinese CO₂ emission flows have reversed since the global financial crisis. Nat. commun. 8(1), 1–10 (2017). +9. Davis, S. J. & Caldeira, K. Consumption-based accounting of CO₂ emissions. Proc. Natl. Acad. Sci. USA 107(12), 5687–5692 (2010). +10. Tong, D. et al. Committed emissions from existing energy infrastructure jeopardize 1.5°C climate target. Nature 572 (7769), 373–377 (2019). +11. CO₂ Emissions from Fuel Combustion (International Energy Agency, 2020). +12. Caliendo, L. & Parro, F. Estimates of the trade and welfare effects of NAFTA. Rev. Econ. Stud. 82(1), 1–44 (2015). +13. Trefler, D. The long and short of the Canada-U.S. Free Trade Agreement. Am. Econ. Rev. 94 (4), 870–895 (2004). +14. Baier, S. L. & Bergstrand, J. H. Estimating the effects of Free Trade Agreements on trade flows using matching econometrics. J. Int. Econ. 77, 77–85 (2009). +15. Romalis, J. NAFTA’s and CUSTFA’s impact on international trade. Rev. Econ. Stat. 89, 416–435 (2007). + +<--- Page Split ---> + +16. Ramondo, N. & Rodríguez-Clare, A. Trade, multinational production, and the gains from openness. J. Polit. Econ. 121(2), 273–322 (2013).17. Hertel, T., Hummels, D., Ivanic, M. & Keeney, R. How confident can we be of CGE-based assessments of Free Trade Agreements? Econ. Model. 24(4), 611–635 (2007).18. Guan, D. B. & Reiner, D. M. Emissions affected by trade among developing countries. Nature 462, 159–159 (2009).19. Meng, J. et al. The rise of South–South trade and its effect on global \(\mathrm{CO_2}\) emissions. Nat. Commun. 9(1), 1–7 (2018).20. Lin, J. et al. China’s international trade and air pollution in the United States. Proc. Natl. Acad. Sci. USA 111, 1736–1741 (2014).21. Meng, B., Peters, G. P., Wang, Z. & Li, M. Tracing \(\mathrm{CO_2}\) emissions in global value chains. Energy Econ. 73, 24–42 (2018).22. Wiedmann, T. & Lenzen, M. Environmental and social footprints of international trade. Nat. Geosci. 11, 314–321 (2018).23. López, L.-A., Cadarso, M.-Á., Zafrilla, J. & Arce, G. The carbon footprint of the U.S. multinationals’ foreign affiliates. Nat. Commun. 10, 1672 (2019).24. Zhang, Z. K. et al. Embodied carbon emissions in the supply chains of multinational enterprises. Nat. Clim. Chang. 10, 1096–1101 (2020).25. Zhang, Z. K., Zhu, K. F. & Hewings, G. J. D. A multi-regional input—output analysis of the pollution haven hypothesis from the perspective of global production fragmentation. Energy Econ. 64, 13–23 (2017).26. Zhang, Z. K., Zhang, Z. & Zhu, K. F. Allocating carbon responsibility: the role of spatial production fragmentation. Energy Econ. 87, 104491 (2020).27. Duan, Y. W. & Yan, B. Q. Economic gains and environmental losses from international trade: A decomposition of pollution intensity in China’s value-added trade. Energy Econ. 83, 540–554 (2019).28. Eaton, J. & Kortum, S. Technology, geography, and trade. Econometrica 70(5), 1741–1779 (2002).29. Koopman, R., Wang, Z. & Wei, S. J. Tracing value added and double counting in gross exports. Am. Econ. Rev. 104, 459–494 (2014).30. Antràs, P. & de Gortari, A. On the geography of global value chains. Econometrica 88(4), 1553–1598 (2020).31. Miller, R.E. & Blair, P.D in Input-output analysis: Foundations and extensions. (Cambridge University Press, 2009).32. Wiedmann, T. A review of recent multi-region input–output models used for consumption-based emission and resource accounting. Ecol. Econ. 69, 211–222 (2009).33. Nie, J.J., Shi, C.L., Xiong, Y., Xia, S.M. & Liang, J.M. Downside of a carbon tax for environment: Impact of information sharing. Adv. Clim. Chang. Res. 11(2), 92-101 (2020).34. Dietzenbacher, E., Cazcarro, I. & Arto, I. Towards a more effective climate policy on international trade. Nat. Commun. 11, 1130 (2020).35. Duan, Y. W. & Jiang, X. M. A Re-evaluation on the role of multinational enterprises in global \(\mathrm{CO_2}\) emissions. Energy Econ. 97, 105181 (2021).36. Zhu, K. F. & Jiang, X. M. Slowing down of globalization and global \(\mathrm{CO_2}\) emissions – A causal or casual association? Energy Econ. 84, 104483 (2019). + +<--- Page Split ---> + +37. Zhang, Z. K. et al. Production globalization makes China's exports cleaner. One Earth 2 (5), 468-478 (2020). + +38. Duan, H. B. et al. Assessing China's efforts to pursue the \(1.5^{\circ}\mathrm{C}\) warming limit. Science 372 (6540), 378-385 (2021). + +39. Kander, A., Jiborn, M., Moran, D. D. & Wiedmann, T. O. National greenhouse-gas accounting for effective climate policy on international trade. Nat. Clim. Chang. 5, 431-435 (2015). + +40. Domingos, T., Zafrilla, J. E. & López, L. A. Consistency of technology-adjusted consumption-based accounting. Nat. Clim. Chang. 6, 729-730 (2016). + +41. Rodrigues, J., Domingos, T., Giljum, S. & Schneider, F. Designing an indicator of environmental responsibility. Ecol. Econ. 59, 256-266 (2006). + +42. Peters, G. P. et al. Key indicators to track current progress and future ambition of the Paris Agreement. Nat. Clim. Chang. 7, 118-122 (2017). + +43. Gallego, B. & Lenzen, M. A consistent input-output formulation of shared producer and consumer responsibility. Econ. Syst. Res. 17, 365-391 (2005). + +44. Lenzen, M., Murray, J., Sack, F. & Wiedmann, T. Shared producer and consumer responsibility - theory and practice. Ecol. Econ. 61, 27-42 (2007). + +45. Lenzen, M. Consumer and producer environmental responsibility: a reply. Ecol. Econ. 66, 547-550 (2008). + +46. Rogelj, J. et al. Paris Agreement climate proposals need a boost to keep warming well below \(2^{\circ}\mathrm{C}\) . Nature 534, 631 (2016). + +47. Keen, M. & Kotsogiannis, C. Coordinating climate and trade policies: Pareto efficiency and the role of border tax adjustments. J. Int. Econ. 94(1): 119-128 (2014). + +48. Höhne, N. et al. The Paris Agreement: resolving the inconsistency between global goals and national contributions. Clim. Policy 17, 16-32 (2017). + +49. Raupach, M. R. et al. Sharing a quota on cumulative carbon emissions. Nat. Clim. Chang. 4, 873-879 (2014). + +50. World Investment Report 2018 (UNCTAD, 2018). https://unctad.org/webflyer/world-investment-report-2018 + +51. World Investment Report 2019 (UNCTAD, 2019). https://unctad.org/webflyer/world-investment-report-2019 + +52. World Investment Report 2021 (UNCTAD, 2021). https://unctad.org/webflyer/world-investment-report-2021 + +53. Eaton, J. & Kortum, S. Putting Ricardo to work. J. Econ. Perspect. 26(2), 65-90 (2012). + +54. OECD. OECD Input-Output tables. (2018). https://stats.oecd.org/ + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementaryinformation.docx + +<--- Page Split ---> diff --git a/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782_det.mmd b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9cf22061a04e07788f6c0f10f337d13612424cfd --- /dev/null +++ b/preprint/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782/preprint__1380986f3e757cfdfc73c1442e96447663548934fcec18411a7e59342d950782_det.mmd @@ -0,0 +1,501 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 896, 175]]<|/det|> +# Regional trade agreement burdens global carbon emission mitigation + +<|ref|>text<|/ref|><|det|>[[44, 196, 475, 238]]<|/det|> +Kailan Tian Academy of Mathematics and Systems Science + +<|ref|>text<|/ref|><|det|>[[44, 243, 475, 285]]<|/det|> +Yu Zhang Academy of Mathematics and Systems Science + +<|ref|>text<|/ref|><|det|>[[44, 290, 211, 330]]<|/det|> +Yuze Li Boston University + +<|ref|>text<|/ref|><|det|>[[44, 336, 432, 377]]<|/det|> +Xi Ming University of Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 382, 790, 425]]<|/det|> +Shangrong Jiang University of Chinese Academy of Sciences https://orcid.org/0000- 0002- 2780- 9075 + +<|ref|>text<|/ref|><|det|>[[44, 428, 400, 448]]<|/det|> +Hongbo Duan (hbduan@ucas.ac.cn) + +<|ref|>text<|/ref|><|det|>[[44, 450, 790, 470]]<|/det|> +University of Chinese Academy of Sciences https://orcid.org/0000- 0001- 5143- 4413 + +<|ref|>text<|/ref|><|det|>[[44, 475, 360, 514]]<|/det|> +Cuihong Yang Institute of Systems Science, China + +<|ref|>text<|/ref|><|det|>[[44, 520, 190, 540]]<|/det|> +Shouyang Wang + +<|ref|>text<|/ref|><|det|>[[52, 543, 475, 562]]<|/det|> +Academy of Mathematics and Systems Science + +<|ref|>sub_title<|/ref|><|det|>[[44, 604, 102, 621]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 641, 855, 662]]<|/det|> +Keywords: RCEP, Trade and welfare, Carbon mitigation, Input- output analysis, Sectoral linkage + +<|ref|>text<|/ref|><|det|>[[44, 679, 318, 699]]<|/det|> +Posted Date: August 26th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 717, 463, 737]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 800470/v1 + +<|ref|>text<|/ref|><|det|>[[42, 754, 910, 797]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 832, 940, 876]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 20th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28004- 5. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[149, 99, 848, 164]]<|/det|> +# Regional trade agreement burdens global carbon emission mitigation + +<|ref|>text<|/ref|><|det|>[[149, 185, 848, 225]]<|/det|> +Kailan Tian1, Yu Zhang1, Yuze Li2, Xi Ming3, Shangrong Jiang3, Hongbo Duan3\*, Cuihong Yang1,3\*, Shouyang Wang1,3\* + +<|ref|>text<|/ref|><|det|>[[148, 243, 850, 336]]<|/det|> +1. Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China +2. Questrom School of Business, Boston University, Boston 02215, USA +3. School of Economics and Management, University of Chinese Academy of Sciences, Beijing 100086, China + +<|ref|>sub_title<|/ref|><|det|>[[149, 365, 226, 381]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[147, 399, 852, 904]]<|/det|> +Regional trade agreements (RTAs) have been widely adopted to facilitate international trade and cross- border investment and ultimately promote economic development. However, ex ante measurements of the environmental effects of RTAs to date have not been well conducted. Here, we estimate the \(\mathrm{CO_2}\) emission burdens of the Regional Comprehensive Economic Partnership (RCEP) after evaluating its economic effects. We find that trade among RCEP member countries will increase significantly and economic output will expand with the reduction of regional tariffs. However, the results show that complete tariff elimination among RCEP members would increase the yearly global \(\mathrm{CO_2}\) emissions from fuel combustion by about \(3.12\%\) , which doubles the annual average growth rate of global \(\mathrm{CO_2}\) emissions in the last decade. The emissions in some developing members will surge. We therefore stress the necessity of balancing carbon mitigation and the pursuit of economic profitability. The technological advancement of emission mitigation and more effective climate policies for international trade are urgently required to avoid undermining international efforts to reduce global emissions. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 95, 238, 111]]<|/det|> +## Keywords + +<|ref|>text<|/ref|><|det|>[[147, 131, 839, 149]]<|/det|> +RCEP; Trade and welfare; Carbon mitigation; Input- output analysis; Sectoral linkage + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[148, 98, 279, 117]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[147, 157, 853, 475]]<|/det|> +Regional trade agreements (RTAs) have been sweeping the world and have become ubiquitous in efforts to facilitate international trade and investment \(^{1 - 3}\) . After an 8- year- long negotiation, the Regional Comprehensive Economic Partnership (RCEP) finally concluded in November 2020 and became the largest RTA in the world in terms of both economic size and population. According to the Schedule of Tariff Commitments in the RCEP Agreement, over \(90\%\) of the trade in goods will eventually have zero tariffs, and most of them will be duty- free immediately or within ten years after the agreement enters into force. Tariff elimination in the region will reduce trade and production costs, resulting in considerable trade- creation and production- boosting effects. + +<|ref|>text<|/ref|><|det|>[[147, 491, 853, 884]]<|/det|> +However, increased international production fragmentation has raised concerns about the trade- climate dilemma (or pollution haven effect) of international trade \(^{4 - 10}\) . That is, international trade increases global or regional emissions if developed economies with cleaner production technology and more stringent environmental policies transfer their polluting industries or production activities to developing countries, leading to emission leakage. Most of the RCEP member countries are typical developing economies that are less emission efficient in their manufacturing industries. In 2018, the amount of \(\mathrm{CO_2}\) emitted by RCEP member countries accounted for a high share \((39.13\%^{11})\) of global \(\mathrm{CO_2}\) emissions from fuel combustion. Therefore, the rapid growth of production activities and trade in an increasing number of less developed nations could impose non- negligible burdens on global and national emission + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 852, 298]]<|/det|> +mitigation. Previous work deals with the economic effects of RTAs, and the other deals with the environmental side effects of international trade. The first strand of the literature focuses on quantifying the economic welfare effects of RTAs \(^{12 - 17}\) and has largely neglected environmental problems. Conversely, the second strand has substantially accounted ex post for the large carbon flows between countries via international trade \(^{18 - 27}\) . + +<|ref|>text<|/ref|><|det|>[[147, 315, 852, 707]]<|/det|> +In fact, few studies have ex ante quantified the environmental effects of a trade agreement. In this study, we aim to estimate such burdens after evaluating the economic effects of RCEP tariff reductions. The quantification of both RCEP economic gains and environmental burdens has important policy implications for balancing economic development and the responsibility of carbon emission mitigation when implementing RTAs. We estimate how and to what extent RCEP tariff reductions affect trade and economic welfare using a multi- sector and multi- country general equilibrium model \(^{12}\) . It captures sectoral heterogeneity, inter- sectoral linkages, and trade in intermediate products. Due to such features and its tractability, this quantitative trade model has become the workhorse for evaluating the possible consequences of tariff changes in an era of global production networks or global value chains \(^{29,30}\) . + +<|ref|>text<|/ref|><|det|>[[147, 724, 852, 890]]<|/det|> +We find that the RCEP tariff reduction will unleash trade- creation effects, which are determined by the magnitude of tariff reduction, export bundles, trade elasticity, and inter- sectoral linkages in each country. In addition, all member countries' economic welfares will improve. To reveal through what channel welfares are improved, we further decompose the total welfare effects into volume of trade effect and terms of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 92, 852, 595]]<|/det|> +trade effect, which are further decomposed separately into the results of trade with RCEP members versus trade with the rest of the world. However, our results also show that complete tariff elimination within the RCEP bloc would increase the yearly global \(\mathrm{CO_2}\) emissions by \(3.12\%\) if the emission intensities ( \(\mathrm{CO_2}\) emissions per unit of output) keep unchanged. Given the fact that the annual average growth rate of global \(\mathrm{CO_2}\) emissions from fuel combustion in the last decade was about \(1.53\%^{11}\) , \(3.12\%\) represents a substantial burden on global \(\mathrm{CO_2}\) emission mitigation. The increased carbon emissions are driven by the rise of production for both domestic expenditures and trade. Trade changes would also increase considerable \(\mathrm{CO_2}\) emissions in the RCEP members. The environmental burdens on some developing countries such as Vietnam and Thailand will surge. As a result, we emphasize that technological advancements in reducing pollutant intensity are urgently required in more developing countries to offset the extra emissions caused by the RCEP. We also suggest that more effective climate policies for international trade should be designed and implemented. + +<|ref|>sub_title<|/ref|><|det|>[[148, 637, 224, 655]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[148, 695, 553, 712]]<|/det|> +## Economic effects from RCEP tariff elimination + +<|ref|>text<|/ref|><|det|>[[147, 744, 853, 910]]<|/det|> +In this subsection, we evaluate how and to what extent trade and welfare are affected by tariff reductions committed in the RCEP Agreement. Our estimation draws on three types of datasets: input- output (I.O.) tables, bilateral trade flows, and bilateral tariff data. Supplementary note 1 provides detailed descriptions of all the data used in our evaluation. We set the effective applied ad valorem tariff rates in 2019 as the base world + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 851, 188]]<|/det|> +tariff structure before the RCEP enters into force. The effects of the RCEP tariff changes are evaluated by changing the tariff structure among RCEP member countries and leaving the tariff structure unchanged for countries outside the agreement. + +<|ref|>text<|/ref|><|det|>[[147, 201, 853, 896]]<|/det|> +Fig. 1 presents the aggregate changes in multilateral trade for RCEP member countries when tariffs within the bloc decline to zero after the agreement enter into force. Unsurprisingly, tariff elimination will gradually unleash the agreement's trade- creation effect and substantially strengthen the trade linkages between these countries before the RCEP. We also observe that the trade effects vary across members. First, the RCEP will significantly increase trade between China, Japan, and South Korea. Specifically, China's exports to Japan and South Korea will increase by \(17.61\%\) and \(33.85\%\) , and Japan's exports to China and South Korea will increase by \(29.14\%\) and \(58.62\%\) , respectively. Second, the RCEP will boost more trade for some ASEAN economies. For example, Indonesia's exports to China, South Korea, Thailand, and Vietnam will increase quite markedly by \(109.75\%\) , \(118.79\%\) , \(91.41\%\) , and \(103.03\%\) , respectively. Similar effects will occur for certain other ASEAN economies, such as Malaysia, the Philippines, Thailand, and Vietnam. It is apparent that tariff elimination in the RCEP bloc will improve not only South- North but also South- South trade. The equilibrium model indicates that the trade effects are determined by the magnitude of tariff reduction, export bundles, trade elasticity, and the intermediate input structure for the production of each sector in each country. The divergent effects shown in Fig. 1 are the complete results of these differently weighted factors. Another observation is that some bilateral trade (e.g., the exports of Singapore to Australia and Japan) will decline + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 852, 260]]<|/det|> +after the RCEP enters into force. This negative effect arises from trade diversion. For example, Singapore was already an almost entirely free-trade country before the RCEP. The RCEP tariff reduction substantially reduces the costs of trade with other members and raises their relative competitiveness. This process facilitates trade diversion from Singapore. + +<|ref|>image<|/ref|><|det|>[[147, 272, 852, 765]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 771, 852, 900]]<|/det|> +
Fig. 1 Trade effects of the RCEP tariff elimination on member countries (%). The figure presents the change rates in multilateral trade for the situation in which trade in goods among the RCEP members have zero tariffs. Two ASEAN countries (Laos and Myanmar) are classified into the rest of the world because input-output tables for Laos and Myanmar are not available. We thus cannot provide results for these two countries. The results for the cases in which the tariffs decline to the committed level in year 1, 5, 10, and 20 after the RCEP enters into force are provided in Supplementary Table 1. The results for any other years are also available upon request.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 95, 855, 335]]<|/det|> +Table 1 presents the results of welfare effects for RCEP members. It shows that all members benefit from the RCEP tariff reductions, with most small countries gaining more than large ones. The first column presents that all RCEP members' real wages will increase, with Cambodia increasing the most. The second column shows that the welfare (the summation of labour income, tariff revenues, and trade deficits) of Vietnam, Cambodia, and Singapore will increase the most, by \(15.63\%\) , \(8.54\%\) , and \(3.79\%\) , respectively. The effects for large economies such as China and Japan are smaller. + +<|ref|>table_caption<|/ref|><|det|>[[149, 347, 765, 363]]<|/det|> +Table 1 Welfare effects of the RCEP tariff elimination on member countries + +<|ref|>table<|/ref|><|det|>[[149, 363, 848, 688]]<|/det|> +
MembersReal wageWelfare
TotalVolume of trade
RCEPRoWTerms of trade RCEP RoW
Australia100.25%0.20%0.00%0.00%0.12%0.07%
China100.36%0.34%0.16%0.01%0.08%0.09%
Japan100.28%0.28%0.05%0.00%0.09%0.13%
South Korea102.38%3.28%4.51%-0.53%-0.43%-0.28%
New Zealand100.38%0.47%0.01%0.01%0.27%0.18%
ASEAN
Brunei100.47%0.59%0.00%0.00%0.48%0.11%
Cambodia111.86%8.54%5.19%-0.53%1.75%2.13%
Indonesia100.88%0.81%0.36%0.03%0.23%0.19%
Malaysia104.68%1.31%1.18%-0.13%0.11%0.14%
Philippines102.01%0.52%0.53%-0.03%-0.04%0.06%
Singapore103.52%3.79%0.00%0.00%1.71%2.08%
Thailand103.15%1.82%2.50%-0.19%-0.42%-0.07%
Vietnam105.60%15.63%10.17%-0.38%3.14%2.70%
+ +<|ref|>text<|/ref|><|det|>[[147, 689, 852, 779]]<|/det|> +Note: The table presents the changes in real wage and welfare for RCEP members if trade in goods among the members becomes duty-free. The total welfare effects are decomposed into volume of trade effect and terms of trade effect, which are further decomposed separately into the results of trade with RCEP members (columns 3 and 5) versus trade with the rest of the world (RoW, column 4 and 6). + +<|ref|>text<|/ref|><|det|>[[147, 789, 850, 881]]<|/det|> +To reveal how each member's welfare is improved, we decompose the total welfare effects into volume of trade effect (columns 3 and 4) and terms of trade effect (columns 5 and 6), which are further decomposed separately into the results of trade + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 852, 447]]<|/det|> +with RCEP members versus trade with other economies outside the RCEP Agreement (the rest of the world, RoW). Columns 3 and 4 show that trade with RCEP members rather than the RoW is the most important contributor to the increase in all members' volume of trade. Comparing with the results in columns 5 and 6, we can find that creating more trade within the RCEP bloc also makes the most significant contribution to the increase in welfare for member countries—South Korea, Cambodia, Malaysia, the Philippines, Thailand, and Vietnam. Additionally, trade reductions with the RoW generate slightly negative welfare effects for most members. The reason for such negative effects stems from the RCEP diverting trade from non- RCEP member countries. + +<|ref|>text<|/ref|><|det|>[[147, 465, 853, 891]]<|/det|> +Columns 5 and 6 show that the aggregate terms of trade for almost all members improve, whereas South Korea's and Thailand's terms of trade deteriorate slightly. This differential performance can be attributed to the divergent changes in the export prices of each country, as terms of trade compare the price of a country's export with the price of its import. The model shows that export prices are determined by the unit costs of input bundles, namely, the combination of labour costs (i.e., wages) and the prices of intermediate inputs. Column 1 shows that the RCEP will increase the real wages of all members, which increases export prices. However, other things being equal, the prices of intermediate inputs will decline with the reductions in tariffs on imported intermediates. Such effects can further be propagated through input- output linkages. As a result, the change in the prices of intermediate inputs decreases export prices. Ultimately, for most RCEP members, the increase in real wages is larger than the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[148, 94, 850, 149]]<|/det|> +decrease in the prices of intermediate inputs, which results in a positive effect on terms of trade. For other economies, the contrary is the case. + +<|ref|>text<|/ref|><|det|>[[147, 168, 852, 485]]<|/det|> +Supplementary Table 2 provides the welfare effects of RCEP tariff reductions on economies outside this agreement. The effects are twofold. On the one hand, as explained above, the agreement generates trade diversion towards RCEP members. On the other hand, in an era of increased international fragmentation, the production activities of trade products in countries outside the RCEP require intermediate inputs from RCEP members. The decreased prices of such intermediate products due to RCEP tariff reductions also change the production costs, export prices, and terms of trade in economies outside the RCEP. As a result, the effects are negative for some non- RCEP economies but positive for others. However, the impact is very small. + +<|ref|>text<|/ref|><|det|>[[147, 502, 852, 892]]<|/det|> +For each RCEP member, we calculate the sectoral contribution to its aggregated change in volume of trade and terms of trade. First, as shown in Fig. 2, the contribution varies considerably across countries and sectors. For example, mining is the sector with the most significant contribution to the change in Australia's terms of trade, whereas electrical equipment is the greatest contributor for China and Japan. Second, agriculture (and food products for some countries) makes a significant contribution to changes in the volume of trade in many countries, including China, Japan, South Korea, New Zealand, the Philippines, and Thailand. The reason is that agriculture is strongly protected in most countries, as the current tariffs are relatively high. For example, the average tariffs applied by Japan to its imported food products and agriculture in 2019 were \(10.74\%\) and \(5.84\%\) , respectively, which were the largest and second- largest import + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 852, 260]]<|/det|> +tariffs among all Japanese goods sectors. Agriculture (and food products) is a homogeneous goods sector with high import tariff trade elasticity. A slight reduction in the tariffs for this sector can improve the trade volume considerably because it is relatively easy to change suppliers. Petroleum in some countries is similarly affected for similar reasons. + +<|ref|>image<|/ref|><|det|>[[147, 321, 822, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 660, 851, 750]]<|/det|> +
Fig. 2 Sectoral contribution to the welfare changes in RCEP members (%). The left graph presents the sectoral contribution to the aggregate changes in the volume of trade in the case that all trade in goods among RCEP members are duty-free, while the right graph presents the results for terms of trade. For a country, the column-wise summation of all sectors' contributions equals \(100\%\) . The description of the tradable sectors is provided in Table A1 of Supplementary note 1.
+ +<|ref|>text<|/ref|><|det|>[[147, 761, 851, 891]]<|/det|> +Another notable observation is that a handful of sectors – electrical equipment, machinery and equipment, and motor vehicles – explain a high proportion of the changes in terms of trade in many countries (China, Japan, South Korea, and some ASEAN economies). This is the combined result of tariff reduction, the share of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 852, 336]]<|/det|> +intermediate inputs required in the production process, and inter- sectoral linkages. Although the tariff reductions in these technology- intensive sectors are not the largest, they are considerable for some countries. More importantly, these sectors use a considerably larger share of intermediate inputs in production than other sectors. They also have stronger input- output linkages with other sectors. Therefore, a decrease in the unit production costs in these sectors has a larger multiplicative effect and thus a larger impact on terms of trade. + +<|ref|>sub_title<|/ref|><|det|>[[148, 366, 597, 385]]<|/det|> +## Carbon emission burdens of RCEP tariff reductions + +<|ref|>text<|/ref|><|det|>[[147, 415, 852, 882]]<|/det|> +The synergy of tariff elimination within the RCEP bloc substantially reduces the costs of intra- regional trade and production and thus increases the outputs of the member countries. As a result, carbon emissions will also increase significantly if the emission intensities (emissions per unit of output) do not decrease enough to offset the extra emissions caused by the increase in production outputs. Assuming the emission intensities of all countries stay at the same level as that in the year 2015, the global \(\mathrm{CO_2}\) emissions from fuel combustion would increase by 251.35 million tonnes (Mt; \(0.75\%\) compared to the amount in 2018) when the tariff structure changes to that specified for the first year after the RCEP enters into force. When the tariffs continue to decline to the level in the fifth and tenth years and then to zero, global \(\mathrm{CO_2}\) emissions will increase by 463.72 Mt (1.38%), 756.38 Mt (2.26%), and 1046.45 Mt (3.12%), respectively. Recalling that global \(\mathrm{CO_2}\) emissions grew at an annual average rate of \(1.53\%^{11}\) in the last decade, these results indicate substantial burdens on carbon emission mitigation. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 853, 559]]<|/det|> +The increased \(\mathrm{CO_2}\) will be emitted mainly by RCEP member countries. In the situation of all trade in goods in the RCEP region becoming duty- free, the production of RCEP members would increase emissions by 789.05 Mt \(\mathrm{CO_2}\) (subplot D in Fig. 3), accounting for \(75.40\%\) of the increased global emissions. Amongst, Mainland China will be the largest contributor in terms of absolute value (495.73 Mt \(\mathrm{CO_2}\) , \(47.37\%\) of the increased global emissions), followed by the ASEAN economies (164.72 Mt \(\mathrm{CO_2}\) , \(15.74\%\) ) and Japan (52.66 Mt \(\mathrm{CO_2}\) ). In terms of magnitude, Vietnam will increase the most ( \(16.51\%\) ), followed by Malaysia ( \(16.09\%\) ) and Thailand ( \(13.58\%\) ). The results indicate that the emission intensities in these countries must decrease by the same magnitude to ensure that the \(\mathrm{CO_2}\) emissions do not increase. The magnitude of the decrease should be larger if countries aim to reduce their emissions. Such an ambitious target could be a non- negligible burden, especially for some developing ASEAN economies. + +<|ref|>text<|/ref|><|det|>[[147, 575, 852, 706]]<|/det|> +In subplot E of Fig. 3, we also present the ratio of welfare change to the \(\mathrm{CO_2}\) emission change rate for RCEP members. The ratio gives the welfare gains at the cost of a \(1\%\) increase in carbon emissions. In terms of this ratio, Vietnam, Singapore, and Cambodia are the greatest gainers, whereas China and Japan rank at the lower end. + +<|ref|>text<|/ref|><|det|>[[147, 724, 852, 890]]<|/det|> +The increased carbon emissions are driven by the rise of production for both domestic expenditures and trade. As mentioned above, the RCEP tariff reductions bring mainly trade- creation effects within the RCEP bloc and cause trade diversion between RCEP members and non- RCEP economies. An RCEP member may emit more \(\mathrm{CO_2}\) in its increased trade with other RCEP members and reduce carbon emissions because of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 852, 333]]<|/det|> +its decreased trade with non- RCEP economies. We employ the environmentally extended inter- country input- output (ICIO) model to account for the comprehensive carbon emission changes due to trade changes. The results show that trade changes in the case of all trade in goods within the RCEP bloc becoming duty- free would increase \(\mathrm{CO_2}\) emissions for China, the ASEAN countries, South Korea, Japan, Australia, and New Zealand by 130.17 Mt, 70.36 Mt, 27.24 Mt, 22.52 Mt, 4.82 Mt, and 0.52 Mt, respectively. + +<|ref|>image<|/ref|><|det|>[[168, 351, 840, 759]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 772, 852, 899]]<|/det|> +
Fig. 3 \(\mathbf{CO}_2\) emission burdens of RCEP's tariff reductions. The subplots A, B, and C present the changes and change rates in the amount of \(\mathrm{CO_2}\) emissions emitted by different economies for the cases in which the tariffs within RCEP bloc decline to the level in year 1, 5, and 10 after the RCEP enters into force, respectively. Subplot D presents the corresponding results for the case in which trade in goods among RCEP members is ultimately duty-free. Subplot E gives the ratio of economic welfare change to the \(\mathrm{CO_2}\) emission change rate for RCEP members. Here, Laos and Myanmar are still classified into the rest of the world (RoW) due to the same reason as given in the note of Fig.1.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 853, 521]]<|/det|> +We also find that trade changes slightly increase the amount of \(\mathrm{CO_2}\) (61.19 Mt) emitted by RoW (non- RCEP economies). The emission effects of RCEP on non- RCEP economies are twofold. On the one hand, RCEP tariff reductions reduce the direct exports of some non- RCEP economies to RCEP members for reasons we discussed in the previous section. This will reduce the emissions of some non- RCEP economies. On the other hand, the increased production for trade within the RCEP bloc requires more intermediate inputs from some economies outside the RCEP. The economic activities associated with the increased production of such intermediate products generate more \(\mathrm{CO_2}\) emissions in non- RCEP economies. Due to the increased international production fragmentation, the emissions generated by these indirect linkages can be substantial. The fact that the RCEP increases overall emissions by economies outside indicates that the indirect effects are larger than the direct effects. + +<|ref|>text<|/ref|><|det|>[[147, 538, 853, 891]]<|/det|> +Fig. 4 visualizes increased \(\mathrm{CO_2}\) emissions due to changes in bilateral trade flows, which tells us who emits increased \(\mathrm{CO_2}\) for whom. It shows the amount of \(\mathrm{CO_2}\) emitted by a region of origin for the production of its increased exports to the destination. The largest flow is 44.28 Mt \(\mathrm{CO_2}\) for China's increased exports to the ASEAN countries. Other large flows include emissions for the increased exports of the ASEAN countries to China (40.24 Mt \(\mathrm{CO_2}\) ), China to Japan (36.07 Mt \(\mathrm{CO_2}\) ), and China to South Korea (27.83 Mt \(\mathrm{CO_2}\) ). Fig. 4 also shows that the increase in exports of Japan (20.40 Mt \(\mathrm{CO_2}\) ) and South Korea (29.33 Mt \(\mathrm{CO_2}\) ) generates a relatively small increase in the \(\mathrm{CO_2}\) emissions of these countries. However, their imports generate considerable \(\mathrm{CO_2}\) emission increases (57.83 and 60.29 Mt \(\mathrm{CO_2}\) ) in the source countries. The results + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 849, 150]]<|/det|> +indicate that developing RCEP members including China and some ASEAN economies will emit increasing \(\mathrm{CO_2}\) for the developed members. + +<|ref|>image<|/ref|><|det|>[[156, 170, 840, 440]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 455, 852, 547]]<|/det|> +
Fig. 4 Increased bilateral \(\mathrm{CO_2}\) emission flows in trade. The graph distinguishes seven regions as the origin (the left) and destination (the right). These are: China, the ASEAN (not including Laos and Myanmar), Japan, South Korea, Australia, New Zealand, and the rest of the world (RoW). The graph presents the increased amount of \(\mathrm{CO_2}\) emitted in the region of origin for its production of exports to a destination.
+ +<|ref|>text<|/ref|><|det|>[[147, 558, 852, 909]]<|/det|> +Fig. 5 shows the sectoral contribution to aggregate \(\mathrm{CO_2}\) emission changes generated by trade changes for RCEP members. The exports of electrical equipment contribute the most to China \((14.53\%)\) , the ASEAN countries \((12.73\%)\) , Japan \((18.72\%)\) , and South Korea \((23.65\%)\) . The other two large contributors are machinery and equipment (second largest for China and the ASEAN countries, third- largest for Japan and South Korea) and computer, electronic, and optical products (second for South Korea, third for China, and fifth for Japan). The motor vehicle industry makes the second- largest contribution \((17.75\%)\) to Japan, but its contribution to other countries is relatively small, indicating Japan's strong comparative advantage in this industry. For Australia and New Zealand, the increased trade generates a very small increase in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 850, 149]]<|/det|> +emissions. The major contributors are mining for Australia and agriculture for New Zealand. + +<|ref|>image<|/ref|><|det|>[[148, 163, 853, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 567, 852, 640]]<|/det|> +
Fig. 5 Sectoral contribution to \(\mathbf{CO}_2\) emissions in trade. The pie charts present for RCEP members the sectoral contribution to their \(\mathbf{CO}_2\) emission changes caused by trade changes. The sector classification is the same as in Fig.2, but here we aggregate the three mining-related sectors (mining energy, other mining, and mining service) into one mining sector.
+ +<|ref|>text<|/ref|><|det|>[[147, 650, 852, 891]]<|/det|> +The finding that RTAs increase participants' economic welfare at the cost of environmental burdens can be explained by the fact that in real economic interactions, what, how much, and with whom to trade are still determined based on economic profitability rather than environmental considerations. In a Ricardian world, a country exports more products in which it has a comparative advantage in terms of production. The advantage is defined as using fewer of the resources under consideration. Traditionally, labour was such a resource, and additional trade generated from lower + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 852, 223]]<|/det|> +trade costs leads to increased welfare and emissions in each of the trading countries. Alternatively, if we consider environmental aspects, the story changes. For instance, if we define the advantage as generating fewer emissions in producing a certain product34, increased trade in this product will reduce emissions in both countries. + +<|ref|>sub_title<|/ref|><|det|>[[148, 266, 256, 285]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[147, 320, 852, 904]]<|/det|> +The world trade system has been seriously undermined due to huge shocks, such as the COVID- 19 global pandemic, the United States- China trade conflict, and Brexit, which to some extent have stimulated the formation of more RTAs. In this paper, we estimate the economic gains and the corresponding carbon emission burdens of the RCEP. The results show that RCEP tariff elimination will substantially reduce intra- regional trade costs and product prices, increase the comparative advantage of regional products, and ultimately improve the welfare of all member countries. Meanwhile, we point out that policymakers should pay attention to the environmental impacts of the RCEP since our results indicate non- negligible increases in potential \(\mathrm{CO_2}\) emissions caused by the RCEP. However, we emphasize that anti- globalization is far from a possible strategy for global emission mitigation. Despite de- globalization could reduce international trade and the corresponding embodied carbon emissions in the short term35, 36, it harms the economic welfare of all countries and threatens international efforts to fight climate change in the longer run. Returning to autarky cuts off developing countries' opportunities to participate in global production networks and then upgrade their emission technologies through learning- by- doing. In addition, anti- globalization will + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 738, 112]]<|/det|> +hinder international cooperation that aims to mitigate global emissions \(^{37}\) . + +<|ref|>text<|/ref|><|det|>[[147, 128, 853, 821]]<|/det|> +Instead, we emphasize that technological advancements in reducing pollutant intensity are urgently required in more developing countries. We find that the carbon emission intensities (CO₂ emissions per unit of output) in all member countries should decline to offset the extra emissions caused by the RCEP, particularly for the developing member countries. Our calculations based on data for 2015 show that the emission intensities of China (2.83 times) and most ASEAN economies, such as Vietnam (2.25 times), Malaysia (2.11 times), and Thailand (2.04 times), were more than twice that of Japan. On the one hand, developing nations should make efforts (e.g., by strengthening technological research and development) to reduce their emission intensity gap with developed nations. For example, at the Paris Climate Change Conference in 2015, China committed to achieving a peak in its carbon emissions by 2030 and reducing its emission intensity substantially. In 2021, China pledged to achieve carbon neutrality before 2060, which is largely consistent with the \(1.5^{\circ}\mathrm{C}\) warming limit \(^{38}\) . The country is making greater efforts and taking more effective measures to achieve this challenging goal, including improving technologies, transforming its energy structure, and developing a circular economy. On the other hand, it is also crucial for developed RCEP members to facilitate the transfer of cleaner production technologies to the developing members when implementing the trade agreement. This process will help accelerate improvements in emission performance in developing countries. + +<|ref|>text<|/ref|><|det|>[[148, 835, 850, 890]]<|/det|> +Our findings also suggest that more effective climate policies for international trade should be designed and implemented as we find that some members will emit + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 90, 853, 592]]<|/det|> +increasing \(\mathrm{CO_2}\) for other countries. The premise is a robust and fair accounting system to assign responsibility for internationally traded emissions. The production- based accounting (PBA) system is the currently widely adopted scheme in practice, but it ignores carbon leakages through international trade. Consumption- based accounting (CBA) considers the emissions embodied in trade but fails to stimulate the producing countries to clean up their export industries. \(^{39}\) Researchers \(^{34, 39 - 42}\) have proposed adjusted accounting systems to share producer and consumer responsibility \(^{43 - 45}\) by crediting trade that reduces global emissions and penalizing trade that increases global emissions. Despite these efforts in academia, national and global climate policies have so far not adopted such systems of credits and penalties. Among a long list of researchers, we call for the idea of countries working together to design robust and environmentally fair accounting tools, develop and implement effective global and regional climate policies, and share the responsibility of global emission mitigation \(^{46 - 49}\) . + +<|ref|>text<|/ref|><|det|>[[147, 612, 853, 890]]<|/det|> +This study has potential extensions that are worthy of pursuit. We currently do not consider the influence mechanism by which the barriers in services trade and investment in the RCEP region will also decrease under the agreement. Both international trade and cross- border investment can influence a country's economic welfare and environmental issues. Although global flows of foreign direct investment (FDI) shrank in recent years \(^{50, 51}\) and fell sharply by one third \(^{52}\) in 2020 as a result of the COVID- 19 pandemic, FDI among RCEP members will likely continuously increase after the RCEP enters into force. Increasing FDI may facilitate relocating some climate- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 92, 852, 336]]<|/det|> +unfriendly industries or production activities from industrialized RCEP members to developing countries \(^{23,24}\) . As a result, FDI may magnify the environmental burdens on developing RCEP members. Future studies are expected to provide quantitative analyses of the effects of qualitative cross- border investment rules in the RCEP Agreement on the volume and direction of FDI flows. More in- depth analyses are also expected to allocate the carbon footprints of FDI flows and explore the scheme of sharing environmental responsibility between FDI home and host countries. + +<|ref|>sub_title<|/ref|><|det|>[[148, 378, 226, 395]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[149, 436, 397, 453]]<|/det|> +## Quantifying the economic effects + +<|ref|>text<|/ref|><|det|>[[147, 485, 852, 688]]<|/det|> +This section outlines the model that we employ to quantify the effects of RCEP tariff reductions on trade and welfare. Here, we provide a brief introduction and the main equations. A more detailed description of the model can be found in Caliendo and Parro (2015) \(^{12}\) . The world consists of \(n\) countries and there are \(m\) sectors in each country. Countries are denoted by \(s\) and \(r\) and sectors by \(i\) and \(j\) . The households in country \(r\) derive utility from consuming final products \(C_{r}^{j}\) . The function is Cobb- Douglas and given by + +<|ref|>equation<|/ref|><|det|>[[344, 703, 848, 731]]<|/det|> +\[u(C_{r}) = \prod_{j = 1}^{m}C_{r}^{j\alpha_{r}^{j}},\mathrm{where}\Sigma_{j = 1}^{m}\alpha_{r}^{j} = 1. \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[147, 744, 851, 837]]<|/det|> +There is a continuum of intermediate products \(\omega^{j}\) produced in sector \(j\) . Primary inputs (labour) and a bundle of intermediate inputs from all sectors are used for the production of \(\omega^{j}\) in country \(r\) . The production technology is + +<|ref|>equation<|/ref|><|det|>[[310, 852, 686, 880]]<|/det|> +\[q_{r}^{j}(\omega^{j}) = z_{r}^{j}(\omega^{j})[lab_{r}^{j}(\omega^{j})]^{r / r}\prod_{i = 1}^{m}[int_{i}^{r,j}(\omega^{j})]^{r / r},\] + +<|ref|>text<|/ref|><|det|>[[147, 894, 851, 912]]<|/det|> +where \(z_{r}^{j}(\omega^{j})\) denotes the efficiency in producing \(\omega^{j}\) in country \(r\) . \(lab_{r}^{j}(\omega^{j})\) is labour and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[146, 92, 852, 223]]<|/det|> +\(i n t_{r}^{i,j}(\omega^{j})\) are the intermediate inputs from sector \(i\) required in the production of \(\omega^{j}\) . \(\gamma_{r}^{j}\) denotes the share of value- added in production output, and \(\gamma_{r}^{i,j}\) denotes the share of products from sector \(i\) used as intermediate inputs in the production of \(\omega^{j}\) , with \(\sum_{i = 1}^{m}\gamma_{r}^{i,j} + \gamma_{r}^{j} = 1\) . The cost of an input bundle is given by + +<|ref|>equation<|/ref|><|det|>[[400, 237, 848, 265]]<|/det|> +\[c_{r}^{j} = B_{r}^{j}w_{r}\gamma_{r}^{j}\prod_{i = 1}^{m}P_{i}^{j\gamma_{r}^{i,j}}, \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[146, 279, 850, 372]]<|/det|> +where \(w_{r}\) denotes the wage rate, \(P_{i}^{j}\) gives the price of intermediate inputs from sector \(i\) , and \(B_{n}^{j}\) is a constant. \(c_{r}^{j}\) clearly incorporates all the inter- sectoral linkages which can be obtained from input- output tables. + +<|ref|>text<|/ref|><|det|>[[146, 390, 852, 592]]<|/det|> +In an open economy, producers minimize their production costs and purchase intermediate products from suppliers across countries. However, trade is costly. We denote \(k_{r s}^{j}\) as the bilateral trade cost for country \(r\) 's imports of sector \(j\) products shipped from country \(s\) . It consists of an ad- . valorem tariff \((\tau_{r s}^{j})\) and iceberg trade cost \((d_{r s}^{j})\) \(k_{r s}^{j} = (1 + \tau_{r s}^{j})d_{r s}^{j}\) . Using Eaton and Kortum's (2002)28 representation of technologies which allows production efficiency to distribute Frechet, we can derive the price of the intermediate product as + +<|ref|>equation<|/ref|><|det|>[[364, 604, 848, 639]]<|/det|> +\[P_{r}^{j} = G^{j}\left[\sum_{s = 1}^{n}\lambda_{s}^{j}\left(c_{s}^{j}k_{r s}^{j}\right)^{-\theta^{j}}\right]^{-1 / \theta^{j}}, \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[147, 650, 850, 744]]<|/det|> +where \(G^{j}\) is a constant, \(\lambda_{s}^{j}\) reflects absolute advantage as a higher value indicates more likely a draw of high efficiency, and \(\theta^{j}\) captures comparative advantage as a lower value indicates a higher dispersion of efficiency. \(\lambda_{s}^{j}\) and \(\theta^{j}\) reflect Ricardian trade53. + +<|ref|>text<|/ref|><|det|>[[181, 761, 830, 779]]<|/det|> +The properties of Frechet distribution further enable us to derive the bilateral trade share as + +<|ref|>equation<|/ref|><|det|>[[406, 792, 848, 840]]<|/det|> +\[\pi_{r s}^{j} = \frac{\lambda_{s}^{j}\left[c_{r}^{j}k_{r s}^{j}\right]^{-\theta^{j}}}{\sum_{h = 1}^{n}\lambda_{h}^{j}\left[c_{h}^{j}k_{r h}^{j}\right]^{-\theta^{j}}}. \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[147, 853, 850, 909]]<|/det|> +As shown, any changes in tariffs \((\tau_{r s}^{j})\) can affect trade costs \((k_{r s}^{j})\) and thus directly affect trade shares. Equations (2) and (3) show that changes in tariffs also affect the cost of input bundle \((c_{s}^{j})\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 95, 438, 111]]<|/det|> +and thus have an indirect effect on trade. + +<|ref|>text<|/ref|><|det|>[[147, 131, 851, 223]]<|/det|> +The total expenditure on the products of sector \(j\) in country \(r\) is the summation of firms' expenditures on intermediate products and households' expenditures on final products. It is given by + +<|ref|>equation<|/ref|><|det|>[[370, 237, 848, 264]]<|/det|> +\[X_{r}^{j} = \sum_{i = 1}^{m}\gamma_{r}^{j,i}\sum_{s = 1}^{n}X_{s}^{i}\frac{\pi_{sr}^{i}}{1 + \tau_{sr}^{i}} +\alpha_{r}^{j}I_{r}, \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[148, 280, 196, 295]]<|/det|> +where + +<|ref|>equation<|/ref|><|det|>[[405, 316, 848, 333]]<|/det|> +\[I_{r} = w_{r}L_{r} + R_{r} + D_{r} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[147, 352, 852, 528]]<|/det|> +represents the total household income in country \(r\) , i.e., the sum of labour income \((w_{r}L_{r})\) , tariff revenues \((R_{r})\) and trade deficits \((D_{r})\) . In particular, \(R_{n} = \sum_{j = 1}^{m}\sum_{s = 1}^{n}\tau_{rs}^{j}M_{rs}^{j}\) , where \(M_{rs}^{j} = X_{r}^{j}\frac{\pi_{rs}^{j}}{1 + \tau_{rs}^{j}}\) is country \(r\) 's import of sector \(j\) products from country \(s\) . The trade deficit of a country is the summation of sectoral deficits, \(D_{r} = \sum_{j = 1}^{m}D_{r}^{j}\) , and sectoral deficit is given by \(D_{r}^{j} = \sum_{s = 1}^{n}M_{rs}^{j} - \sum_{s = 1}^{n}E_{rs}^{j}\) , where \(E_{rs}^{j} = X_{s}^{j}\frac{\pi_{sr}^{j}}{1 + \tau_{sr}^{j}}\) . + +<|ref|>text<|/ref|><|det|>[[147, 539, 852, 705]]<|/det|> +The next step is to solve for changes in wages and prices given that the tariff structure \(\tau\) is changed to \(\tau '\) . Instead of solving for two equilibria under \(\tau\) and \(\tau '\) , Caliendo and Parro (2015) \(^{12}\) propose solving for an equilibrium in relative changes so that it is not necessary to estimate some parameters that are difficult to identify. Let a variable with a circumflex “\(\hat{x}\) ” denote its relative change. The equilibrium in relative changes satisfies the following conditions: + +<|ref|>equation<|/ref|><|det|>[[390, 720, 848, 900]]<|/det|> +\[\begin{array}{c}{\hat{k}_{r s}^{j} = (1 + \tau_{r s}^{j}) / (1 + \tau_{r s}^{j})}\\ {\hat{c}_{r}^{j} = \hat{w}_{r}\gamma_{r}^{j}\prod_{i = 1}^{m}\hat{P}_{r}^{i}\gamma_{r}^{i,j}}\\ {\hat{P}_{r}^{j} = \left[\sum_{s = 1}^{n}\pi_{r s}^{j}\left(\hat{c}_{s}^{j}\hat{k}_{r s}^{j}\right)^{-\theta j}\right]^{-1 / \theta j}}\\ {\hat{\pi}_{r s}^{j} = \left[\frac{\hat{c}_{s}^{j}\hat{k}_{r s}^{j}}{\hat{P}_{r}^{j}}\right]^{-\theta j}} \end{array} \quad (7)\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[333, 85, 848, 120]]<|/det|> +\[\Sigma_{j = 1}^{m}\Sigma_{s = 1}^{n}X_{r}^{j^{\prime}}\frac{\pi_{r s}^{j^{\prime}}}{1 + \tau_{r s}^{j^{\prime}}} - D_{r}^{\prime} = \Sigma_{j = 1}^{m}\Sigma_{s = 1}^{n}X_{s}^{j^{\prime}}\frac{\pi_{r s}^{j^{\prime}}}{1 + \tau_{r s}^{j^{\prime}}} \quad (12)\] + +<|ref|>equation<|/ref|><|det|>[[416, 131, 848, 150]]<|/det|> +\[I_{r}^{\prime} = \widehat{w}_{r}w_{r}L_{r} + R_{r}^{\prime} + D_{r}^{\prime}. \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[147, 168, 852, 373]]<|/det|> +Introducing into the model the changes in tariff structure, we can solve for changes in (total and bilateral) trade flows, real wages \((w_{r} / P_{r})\) , welfare \((I_{r} / P_{r})\) , and production output for each country. The change in welfare can be decomposed into volume of trade effect and terms of trade effect. The welfare change can also be calculated at both the bilateral and sectoral levels. We can calculate the change in volume of trade and terms of trade between country \(s\) and \(r\) , and the change in a specific sector \(j\) of country \(r\) . + +<|ref|>sub_title<|/ref|><|det|>[[149, 404, 483, 420]]<|/det|> +## Accounting for the carbon emission changes + +<|ref|>text<|/ref|><|det|>[[147, 452, 852, 582]]<|/det|> +The RCEP tariff reductions lead to changes in multilateral trade flows resulting in trade- related carbon emissions changes. We adopt the environmentally extended ICIO model (see Table A2 in Supplementary note 1 for the stylized table) to account for the carbon emission changes. Denote the following as the flows of final products among different countries: + +<|ref|>equation<|/ref|><|det|>[[383, 593, 613, 665]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[148, 678, 187, 693]]<|/det|> +And, + +<|ref|>equation<|/ref|><|det|>[[378, 704, 619, 777]]<|/det|> +\[\mathbf{A} = \left[ \begin{array}{llll}\mathbf{A}^{11} & \dots & \mathbf{A}^{1r} & \dots & \mathbf{A}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \mathbf{A}^{r1} & \dots & \mathbf{A}^{rr} & \dots & \mathbf{A}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \mathbf{A}^{n1} & \dots & \mathbf{A}^{nr} & \dots & \mathbf{A}^{nn} \end{array} \right]\] + +<|ref|>text<|/ref|><|det|>[[147, 785, 851, 880]]<|/det|> +is the global direct input- output coefficient matrix. Its typical element \(a_{ij}^{sr}\) provides the intermediate input from sector \(i (= 1,\dots ,m)\) in country \(s (= 1,\dots ,n)\) used by sector \(j (= 1,\dots ,m)\) in country \(r (= 1,\dots ,n)\) for producing one unit of output. \(\mathbf{A}^{rr}\) and \(\mathbf{f}^{rr}\) provide intra + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 93, 850, 186]]<|/det|> +country flows of intermediate products and final products. \(\mathbf{A}^{sr}\) and \(\mathbf{f}^{sr}\) \((s\neq r)\) represent trade in intermediate products and trade in final products, respectively. According to the standard input-output model \(^{31}\) , the gross output vector \(\mathbf{y}\) is + +<|ref|>equation<|/ref|><|det|>[[415, 205, 848, 223]]<|/det|> +\[\mathbf{y} = (\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}, \quad (14)\] + +<|ref|>text<|/ref|><|det|>[[147, 243, 850, 297]]<|/det|> +where \(\mathbf{I}\) is an \((nm \times nm)\) identity matrix, and \(\mathbf{u}\) is a summation vector of appropriate length with all elements being ones. + +<|ref|>text<|/ref|><|det|>[[147, 316, 850, 371]]<|/det|> +Let \(\mathbf{w}\) be the \(\mathrm{CO}_{2}\) emission coefficient vector, the elements of which provide the emissions per unit of output. Then, the \(\mathrm{CO}_{2}\) emission vector \(\mathbf{e}\) can be written as + +<|ref|>equation<|/ref|><|det|>[[415, 390, 848, 408]]<|/det|> +\[\mathbf{e}\mathbf{u} = \mathbf{w}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}. \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[147, 427, 852, 592]]<|/det|> +The left side of (15) equals the summation of \(\mathrm{CO}_{2}\) emissions in all sectors in the world, which equals global \(\mathrm{CO}_{2}\) emissions. This equation can be adapted to allow us to calculate the \(\mathrm{CO}_{2}\) emissions \((\mathbf{e}^{r})\) in a specific country \(r\) . This can be obtained by replacing the vector \(\mathbf{w}\) in (15) with a vector \(\mathbf{w}^{r}\) . The new vector has equal length, but only the \(\mathrm{CO}_{2}\) emission coefficients for the sectors in country \(r\) are retained while all other elements are set as zeros. This yields + +<|ref|>equation<|/ref|><|det|>[[415, 611, 848, 629]]<|/det|> +\[\mathbf{e}^{r} = \mathbf{w}^{r}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}. \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[147, 650, 852, 816]]<|/det|> +Equation (16) allows us to calculate, for example, the part of Thailand's \(\mathrm{CO}_{2}\) emissions that are generated by the exports of final products from Japan to final users in China. The production processes for trade between Japan and China may consume intermediate products from Thailand, of which the production emits \(\mathrm{CO}_{2}\) in Thailand. Using the global Leontief inverse, we can take fully into account these indirect effects in equation (16). + +<|ref|>text<|/ref|><|det|>[[148, 836, 850, 889]]<|/det|> +To calculate the \(\mathrm{CO}_{2}\) emission changes in country \(r\) , we compare two situations. The first is the actual situation, and the second is the case in which multilateral trade flows are changed due to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 94, 834, 111]]<|/det|> +RCEP tariff reductions. Moving forward from equation (16), we develop the following equation: + +<|ref|>equation<|/ref|><|det|>[[344, 128, 848, 150]]<|/det|> +\[\Delta \mathbf{e}^{r} = \mathbf{w}^{r}\big(\mathbf{I} - \widetilde{\mathbf{A}}\big)^{-1}\widetilde{\mathbf{F}}\mathbf{u} - \mathbf{w}^{r}(\mathbf{I} - \mathbf{A})^{-1}\mathbf{F}\mathbf{u}, \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[148, 170, 195, 184]]<|/det|> +where + +<|ref|>equation<|/ref|><|det|>[[245, 195, 750, 270]]<|/det|> +\[\widetilde{\mathbf{A}} = \left[ \begin{array}{cccc}\mathbf{A}^{11} & \dots & \widetilde{\mathbf{A}}^{1r} & \dots & \widetilde{\mathbf{A}}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{A}}^{r1} & \dots & \mathbf{A}^{rr} & \dots & \widetilde{\mathbf{A}}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{A}}^{n1} & \dots & \widetilde{\mathbf{A}}^{nr} & \dots & \mathbf{A}^{nn} \end{array} \right]\mathrm{and}\widetilde{\mathbf{F}} = \left[ \begin{array}{cccc}\mathbf{f}^{11} & \dots & \widetilde{\mathbf{f}}^{1r} & \dots & \widetilde{\mathbf{f}}^{1n}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{f}}^{r1} & \dots & \mathbf{f}^{rr} & \dots & \widetilde{\mathbf{f}}^{rn}\\ \vdots & \ddots & \vdots & \ddots & \vdots \\ \widetilde{\mathbf{f}}^{n1} & \dots & \widetilde{\mathbf{f}}^{nr} & \dots & \mathbf{f}^{nn} \end{array} \right].\] + +<|ref|>text<|/ref|><|det|>[[147, 279, 852, 445]]<|/det|> +Equation (17) enables us to consider the changes in both trade in final products and trade in intermediate products. The changes in bilateral trade flows can be obtained at the sectoral level after solving the Caliendo and Parro (2015) \(^{12}\) model described above. However, this model does not enable us to calculate the respective magnitudes of change for final product trade and intermediate product trade. We thus use the same magnitude when conducting the calculation. + +<|ref|>sub_title<|/ref|><|det|>[[148, 490, 275, 504]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[147, 548, 852, 900]]<|/det|> +National and ICIO tables are from the most recent OECD Input- Output Database \(^{54}\) (2018 edition, https://stats.oecd.org/), and bilateral trade flows are obtained from the United Nations Commodity Trade (U.N. Comtrade) database (https://comtrade.un.org/data/). Bilateral tariff data are from World Integrated Trade Solution (WITS) software (https://wits.worldbank.org/). The committed tariff reductions among RCEP parties come from the Schedule of Tariff Commitments in the RCEP Agreement. The schedule provides detailed data for each RCEP party's commitments to tariff reduction for each year after the date of the entry into force of the RCEP Agreement. International Energy Agency (https://www.iea.org/data- and- statistics) provides sectoral \(\mathrm{CO}_{2}\) emissions data \(^{11}\) , and the most recent available data are for 2018. We include maximum number of economies, conditional on obtaining reliable data. We ultimately obtain 60 economies and a constructed RoW + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 95, 852, 261]]<|/det|> +with 36 sectors in each economy. Trade and tariff data are for 2019, and we employ the most recent available input- output tables for 2015, assuming that input- output coefficients in 2019 are not much different than those in 2015. Supplementary note 1 provides more detailed descriptions of all data used in our evaluation. Supplementary Tables 1 and 2 provide additional results. All datasets generated in this study are available upon request. + +<|ref|>sub_title<|/ref|><|det|>[[148, 304, 234, 319]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[147, 352, 852, 910]]<|/det|> +1. Baldwin, R. & Jaimovich, D. Are free trade agreements contagious? J. Int. Econ. 88(1), 1–16 (2012). +2. Mahadevan, R. & Nugroho, A. Can the Regional Comprehensive Economic Partnership minimise the harm from the United States–China trade war? World Econ. 42(11), 3148–3167 (2019). +3. Bhagwati, J. Termites in the trading system: How preferential agreements undermine free trade (Oxford University Press, 2008). +4. Liu, Z. et al. Targeted opportunities to address the climate–trade dilemma in China. Nat. Clim. Chang. 145, 143–145 (2015). +5. Zhang, Q. et al. Transboundary health impacts of transported global air pollution and international trade. Nature 543, 705–709 (2017). +6. Peters, G. P., Minx, J. C., Weber, C. L. & Edenhofer, O. Growth in emission transfers via international trade from 1990 to 2008. Proc. Natl. Acad. Sci. USA 108(21), 8903–8908 (2011). +7. Guan, D. B. et al. Structural decline in China’s CO₂ emissions through transitions in industry and energy systems. Nat. Geosci. 11(8), 551–555 (2018). +8. Mi, Z. F. et al. Chinese CO₂ emission flows have reversed since the global financial crisis. Nat. commun. 8(1), 1–10 (2017). +9. Davis, S. J. & Caldeira, K. Consumption-based accounting of CO₂ emissions. Proc. Natl. Acad. Sci. USA 107(12), 5687–5692 (2010). +10. Tong, D. et al. Committed emissions from existing energy infrastructure jeopardize 1.5°C climate target. Nature 572 (7769), 373–377 (2019). +11. CO₂ Emissions from Fuel Combustion (International Energy Agency, 2020). +12. Caliendo, L. & Parro, F. Estimates of the trade and welfare effects of NAFTA. Rev. Econ. Stud. 82(1), 1–44 (2015). +13. Trefler, D. The long and short of the Canada-U.S. Free Trade Agreement. Am. Econ. Rev. 94 (4), 870–895 (2004). +14. Baier, S. L. & Bergstrand, J. H. Estimating the effects of Free Trade Agreements on trade flows using matching econometrics. J. Int. Econ. 77, 77–85 (2009). +15. Romalis, J. NAFTA’s and CUSTFA’s impact on international trade. Rev. Econ. Stat. 89, 416–435 (2007). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 85, 852, 900]]<|/det|> +16. Ramondo, N. & Rodríguez-Clare, A. Trade, multinational production, and the gains from openness. J. Polit. Econ. 121(2), 273–322 (2013).17. Hertel, T., Hummels, D., Ivanic, M. & Keeney, R. How confident can we be of CGE-based assessments of Free Trade Agreements? Econ. Model. 24(4), 611–635 (2007).18. Guan, D. B. & Reiner, D. M. Emissions affected by trade among developing countries. Nature 462, 159–159 (2009).19. Meng, J. et al. The rise of South–South trade and its effect on global \(\mathrm{CO_2}\) emissions. Nat. Commun. 9(1), 1–7 (2018).20. Lin, J. et al. China’s international trade and air pollution in the United States. Proc. Natl. Acad. Sci. USA 111, 1736–1741 (2014).21. Meng, B., Peters, G. P., Wang, Z. & Li, M. Tracing \(\mathrm{CO_2}\) emissions in global value chains. Energy Econ. 73, 24–42 (2018).22. Wiedmann, T. & Lenzen, M. Environmental and social footprints of international trade. Nat. Geosci. 11, 314–321 (2018).23. López, L.-A., Cadarso, M.-Á., Zafrilla, J. & Arce, G. The carbon footprint of the U.S. multinationals’ foreign affiliates. Nat. Commun. 10, 1672 (2019).24. Zhang, Z. K. et al. Embodied carbon emissions in the supply chains of multinational enterprises. Nat. Clim. Chang. 10, 1096–1101 (2020).25. Zhang, Z. K., Zhu, K. F. & Hewings, G. J. D. A multi-regional input—output analysis of the pollution haven hypothesis from the perspective of global production fragmentation. Energy Econ. 64, 13–23 (2017).26. Zhang, Z. K., Zhang, Z. & Zhu, K. F. Allocating carbon responsibility: the role of spatial production fragmentation. Energy Econ. 87, 104491 (2020).27. Duan, Y. W. & Yan, B. Q. Economic gains and environmental losses from international trade: A decomposition of pollution intensity in China’s value-added trade. Energy Econ. 83, 540–554 (2019).28. Eaton, J. & Kortum, S. Technology, geography, and trade. Econometrica 70(5), 1741–1779 (2002).29. Koopman, R., Wang, Z. & Wei, S. J. Tracing value added and double counting in gross exports. Am. Econ. Rev. 104, 459–494 (2014).30. Antràs, P. & de Gortari, A. On the geography of global value chains. Econometrica 88(4), 1553–1598 (2020).31. Miller, R.E. & Blair, P.D in Input-output analysis: Foundations and extensions. (Cambridge University Press, 2009).32. Wiedmann, T. A review of recent multi-region input–output models used for consumption-based emission and resource accounting. Ecol. Econ. 69, 211–222 (2009).33. Nie, J.J., Shi, C.L., Xiong, Y., Xia, S.M. & Liang, J.M. Downside of a carbon tax for environment: Impact of information sharing. Adv. Clim. Chang. Res. 11(2), 92-101 (2020).34. Dietzenbacher, E., Cazcarro, I. & Arto, I. Towards a more effective climate policy on international trade. Nat. Commun. 11, 1130 (2020).35. Duan, Y. W. & Jiang, X. M. A Re-evaluation on the role of multinational enterprises in global \(\mathrm{CO_2}\) emissions. Energy Econ. 97, 105181 (2021).36. Zhu, K. F. & Jiang, X. M. Slowing down of globalization and global \(\mathrm{CO_2}\) emissions – A causal or casual association? Energy Econ. 84, 104483 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 85, 852, 123]]<|/det|> +37. Zhang, Z. K. et al. Production globalization makes China's exports cleaner. One Earth 2 (5), 468-478 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 123, 850, 158]]<|/det|> +38. Duan, H. B. et al. Assessing China's efforts to pursue the \(1.5^{\circ}\mathrm{C}\) warming limit. Science 372 (6540), 378-385 (2021). + +<|ref|>text<|/ref|><|det|>[[145, 160, 850, 195]]<|/det|> +39. Kander, A., Jiborn, M., Moran, D. D. & Wiedmann, T. O. National greenhouse-gas accounting for effective climate policy on international trade. Nat. Clim. Chang. 5, 431-435 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 197, 850, 232]]<|/det|> +40. Domingos, T., Zafrilla, J. E. & López, L. A. Consistency of technology-adjusted consumption-based accounting. Nat. Clim. Chang. 6, 729-730 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 234, 850, 268]]<|/det|> +41. Rodrigues, J., Domingos, T., Giljum, S. & Schneider, F. Designing an indicator of environmental responsibility. Ecol. Econ. 59, 256-266 (2006). + +<|ref|>text<|/ref|><|det|>[[145, 270, 850, 305]]<|/det|> +42. Peters, G. P. et al. Key indicators to track current progress and future ambition of the Paris Agreement. Nat. Clim. Chang. 7, 118-122 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 307, 850, 342]]<|/det|> +43. Gallego, B. & Lenzen, M. A consistent input-output formulation of shared producer and consumer responsibility. Econ. Syst. Res. 17, 365-391 (2005). + +<|ref|>text<|/ref|><|det|>[[145, 344, 850, 379]]<|/det|> +44. Lenzen, M., Murray, J., Sack, F. & Wiedmann, T. Shared producer and consumer responsibility - theory and practice. Ecol. Econ. 61, 27-42 (2007). + +<|ref|>text<|/ref|><|det|>[[145, 381, 850, 416]]<|/det|> +45. Lenzen, M. Consumer and producer environmental responsibility: a reply. Ecol. Econ. 66, 547-550 (2008). + +<|ref|>text<|/ref|><|det|>[[145, 418, 850, 453]]<|/det|> +46. Rogelj, J. et al. Paris Agreement climate proposals need a boost to keep warming well below \(2^{\circ}\mathrm{C}\) . Nature 534, 631 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 455, 850, 490]]<|/det|> +47. Keen, M. & Kotsogiannis, C. Coordinating climate and trade policies: Pareto efficiency and the role of border tax adjustments. J. Int. Econ. 94(1): 119-128 (2014). + +<|ref|>text<|/ref|><|det|>[[145, 492, 850, 527]]<|/det|> +48. Höhne, N. et al. The Paris Agreement: resolving the inconsistency between global goals and national contributions. Clim. Policy 17, 16-32 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 529, 850, 564]]<|/det|> +49. Raupach, M. R. et al. Sharing a quota on cumulative carbon emissions. Nat. Clim. Chang. 4, 873-879 (2014). + +<|ref|>text<|/ref|><|det|>[[145, 566, 850, 601]]<|/det|> +50. World Investment Report 2018 (UNCTAD, 2018). https://unctad.org/webflyer/world-investment-report-2018 + +<|ref|>text<|/ref|><|det|>[[145, 603, 850, 638]]<|/det|> +51. World Investment Report 2019 (UNCTAD, 2019). https://unctad.org/webflyer/world-investment-report-2019 + +<|ref|>text<|/ref|><|det|>[[145, 640, 850, 675]]<|/det|> +52. World Investment Report 2021 (UNCTAD, 2021). https://unctad.org/webflyer/world-investment-report-2021 + +<|ref|>text<|/ref|><|det|>[[145, 677, 802, 694]]<|/det|> +53. Eaton, J. & Kortum, S. Putting Ricardo to work. J. Econ. Perspect. 26(2), 65-90 (2012). + +<|ref|>text<|/ref|><|det|>[[145, 696, 645, 712]]<|/det|> +54. OECD. OECD Input-Output tables. (2018). https://stats.oecd.org/ + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 131, 366, 150]]<|/det|> +Supplementaryinformation.docx + +<--- Page Split ---> diff --git a/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/images_list.json b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..1420f76371f262ddc6640e49f7cebd408a61b345 --- /dev/null +++ b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Figures:", + "footnote": [], + "bbox": [ + [ + 75, + 113, + 921, + 448 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Spatial distribution of changes in European summer heatwave (EuSHW) intensity under global warming levels. Multi-model mean of regression coefficients for the period 2014-2100 and for SSP2-4.5 and SSP5-8.5 scenarios between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of non-detrended EuSHW cumulative heat; b) the range (i.e., ensemble standard deviation) of non-detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the 95th confidence level.", + "footnote": [], + "bbox": [ + [ + 93, + 81, + 905, + 285 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability. a) Detrended observed, historical and projected EuSHW intensity expressed as cumulative heat in the left y-axis and scaled it to the observed historical mean (1950-2022) in the right y-axis. Solid lines indicate the ensemble mean, the shading indicates the 5th and 95th percentile, and the right vertical bars indicate the 5th and 95th percentile for the last 20 years of the historical simulation and each shared socioeconomic pathway (SSP) scenario for ACCESS-ESM-1.5; b) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat scaled to global mean temperature anomalies (GMTA) relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; c) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability scaled to GMTA relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; d-f) same as a-c) but for CanESM5; g-i) same as a-c) but for MIROC6; j-l) same as a-c) but for MPI-GE CMIP6.", + "footnote": [], + "bbox": [ + [ + 90, + 80, + 907, + 405 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Spatial distribution of changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability under global warming levels. Multi-model mean of regression coefficients for the period 2014–2100 and for SSP2-4.5 andSSP5-8.5 scenarios between global mean temperature anomalies relative to1850-1880 and; a) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \\(95^{\\text{th}}\\) confidence level.", + "footnote": [], + "bbox": [ + [ + 93, + 81, + 904, + 284 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Spatial distribution of changes in soil moisture under global warming levels. Multimodel mean of regression coefficients between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of summer (June, July, August) mean soil moisture; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of summer mean soil moisture. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \\(95^{\\text{th}}\\) confidence level.", + "footnote": [], + "bbox": [ + [ + 93, + 97, + 905, + 292 + ] + ], + "page_idx": 18 + } +] \ No newline at end of file diff --git a/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020.mmd b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a0d9af0d7a576ab057d6ead4028181fcda4c1eb9 --- /dev/null +++ b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020.mmd @@ -0,0 +1,280 @@ + +# Increasing central and northern European summer heatwave intensity due to forced internal variability changes + +Goratz Beobide Arsuaga goratz.beobide.arsuaga@uni- hamburg.de + +Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universitat Hamburg, Hamburg, Germany https://orcid.org/0000- 0002- 7775- 4906 + +Laura Suarez Gutierrez + +Laboratoire des Sciences du Climat et de l'Environnement, Institut Pierre- Simon Laplace, Paris, France / Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland + +Armineh Barkhordarian + +Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universitat Hamburg, Hamburg, Germany https://orcid.org/0000- 0001- 9786- 8081 + +Dirk Olonscheck + +Max Planck Institute for Meteorology https://orcid.org/0000- 0002- 8962- 1406 + +Johanna Baehr + +https://orcid.org/0000- 0003- 4696- 8941 + +## Article + +Keywords: + +Posted Date: April 17th, 2025 + +DOI: https://doi.org/10.21203/rs.3.rs- 6377558/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on October 30th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 65392- w. + +<--- Page Split ---> + +# Increasing central and northern European summer heatwave intensity due to forced internal variability changes + +Goratz Beobide- Arsuaga1 (goratz.beobide.arsuaga@uni- hamburg.de), Laura Suarez- Gutierrez2,3, + +Armineh Barkhordarian1, Dirk Olonscheck4, Johanna Baehr1 + +1Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universität Hamburg, Hamburg, Germany + +2Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland + +3Laboratoire des Sciences du Climat et de l'Environnement, Institut Pierre- Simon Laplace, Paris, France + +4Max Planck Institute for Meteorology, Hamburg, Germany + +## Abstract + +In recent years European summer heatwaves have strongly intensified due to rising anthropogenic emissions. While European summer heatwaves will continue to intensify due to the warming of summer temperatures, the effects of the changes in internal variability under global warming remain unknown. Employing four single- model initial condition large ensembles, we find that the forced internal variability change is projected to intensify central and northern European summer heatwaves. Central and northern Europe will experience frequent moisture limitations, enhancing land- atmospheric feedback and increasing heatwave intensity and variability. In contrast, the forced internal variability change will contribute to weaken southern European summer heatwaves. Southern Europe is projected to face a more stable moisture- depleted environment that reduces extreme temperature variability and heatwave intensity. Our findings imply that while adaptation to increasing mean temperatures in southern Europe should suffice to reduce the vulnerability to increasing heatwave intensity, in central and northern Europe adaptation to increased temperature variability will also be needed. + +## Introduction + +With human- induced rising global mean temperatures, the intensity, frequency and duration of extreme heat events have increased almost everywhere since the 1950s, and the positive trend is expected to continue in the upcoming decades1- 4. Europe, in particular, has been identified as a heatwave hotspot with increasing intensity trends up to four times higher than the rest of the northern midlatitudes5,6. During the last two decades, the occurrence of multiple unprecedented European summer heatwaves (EuSHWs) (e.g. years 2003, 2015, 2018 and 2022) has caused large economic, ecosystem, and humanitarian losses7- 10. + +The recent increase in EuSHW intensity has been partly explained by the forced shift in climatology towards warmer values11- 14. Under rising global mean temperatures, European summer temperatures are also rising, altering the background mean state and shifting summer temperature distributions. Assuming a limited adaptation towards increasing European summer mean temperatures, the threshold at which temperatures are felt extreme will be time- invariant, and consequently, the positive shift in the temperature distribution leads to an increase in EuSHW intensity15. For instance, considering an ecosystem that historically lived with average summer temperatures of \(20^{\circ}\mathrm{C}\) and extreme temperatures of \(30^{\circ}\mathrm{C}\) , a positive shift in temperature distributions of \(5^{\circ}\mathrm{C}\) will increase the mean to \(25^{\circ}\mathrm{C}\) and extreme temperatures to \(35^{\circ}\mathrm{C}\) . Although the difference between the new averaged summer temperatures and the new extreme temperatures has not changed from the historical period, the ecosystem will experience an increase of extreme heat if it does not adapt to the new climatology. + +In addition to the shift in the temperature distribution, the forced changes in internal variability of the climate system are expected to increase European summer temperature variability14,16- 24 which + +<--- Page Split ---> + +could have serious implications for reducing EuSHW vulnerability. The variability in sea surface temperatures, soil moisture content, atmospheric dynamics, and the consequent positive feedbacks involved have been related to historical EuSHWs \(^{5,23,25 - 30}\) . Under rising global mean temperatures, internal climate variability will be forced by altering the mean state and variability of the ocean, soil moisture, atmosphere, and their interactions, affecting the variability of European summer temperatures and the intensity of EuSHWs. An increasing internal variability would lead to a faster increase of extreme temperatures relative to the mean, which could enhance the increase in the heatwave intensity caused by the shift in temperature distributions and alter the intensity range of possible future EuSHWs increasing the risk of unprecedented EuSHWs \(^{31,32}\) . Following the previous example of the ecosystem, if the forced changes in internal variability induce an additional \(2^{\circ}\mathrm{C}\) increase in extreme temperatures, the difference between the new average summer temperatures and the new extreme temperatures will increase from historical times. Therefore, even if the ecosystem adapts to increasing mean temperatures and limit the effects of increasing heatwave intensity due to the distribution shift, this might not be enough to reduce the vulnerability of upcoming EuSHWs. + +While it is well established in the literature that EuSHW intensity will continue to increase due to the shift in summer temperature distributions, how the forced changes in internal climate variability will affect EuSHW intensity under increasing global warming levels remains unknown. Previous studies were limited by the short observational record that did not allow a robust sampling of EuSHWs, or by coarse temporal resolution of climate models that did not allow the identification of heatwave events but rather an assessment of summer mean temperature metrics. Here, we employ four single model initial condition large ensembles (SMILEs): ACCESS- ESM- 1.5 \(^{33}\) (40 members), CanESM5 \(^{34}\) (50 members), MIROC6 \(^{35}\) (50 members), and MPI- GE CMIP6 \(^{36}\) (50 members). All SMILEs considered here provide a daily temporal resolution for the historical period (1850- 2014) and two shared socioeconomic pathway scenarios (SSP2- 4.5, and SSP5- 8.5) that span until the end of the \(21^{\text{st}}\) century. The high temporal resolution enables us to identify heatwave events and quantify their intensity, and the multiple ensemble members for each SMILE enable us to robustly differentiate the effects of changing internal climate variability (i.e. forced changes in internal variability) from shifting temperature distributions (i.e. forced distribution shift) \(^{37,38}\) . Here, for the first time, we investigate the changes in EuSHW intensity and their range due to forced changes in internal variability and forced distribution shift under different global warming levels. + +## Results + +Historical and projected European summer heatwave intensity. The intensity of non- detrended EuSHWs (see methods for details) has been increasing over the past two decades leading to the three most extreme observed events in 2022, 2003, and 2015, in the respective order, and the increase is projected to continue until approximately 2040 regardless of the SSP scenarios for all climate models (Fig. 1a,d,g,j). Thereafter, the forced signal in EuSHW intensity, computed as the ensemble mean, and the range of possible future EuSHW intensity due to the internal climate variability will strongly depend on the SSP scenarios. The four climate models indicate that for the SSP2- 4.5 scenario the increase in EuSHW intensity will be relatively constant, and for the SSP5- 8.5 scenario the increase will be non- linear. By the end of the century, CanESM5 shows the highest forced signal in EuSHW intensity (29 times higher than the mean observed EuSHW intensity for SSP2- 4.5, and 125 times higher for SSP5- 8.5; Fig. 1d), followed by ACCESS- ESM- 1.5 (28 times for SSP2- 4.5, and 88 times for SSP5- 8.5; Fig. 1a), MIROC6 (17 times for SSP2- 4.5, and 66 times for SSP5- 8.5; Fig. 1g) and MPI- GE CMIP6 (10 times for SSP2- 4.5, and 55 times for SSP5- 8.5; Fig. 1j), in the respective order. + +The range of possible future EuSHW intensity due to the internal climate variability computed as the difference between the \(95^{\text{th}}\) and the \(5^{\text{th}}\) ensemble percentile follows a similar evolution (Fig. 1a,d,g,j). For each model, the range increases similarly in both forcing scenarios until approximately 2040. However, the range of EuSHWs due to internal climate variability will diverge + +<--- Page Split ---> + +towards the end of the \(21^{\mathrm{st}}\) century with the largest increase shown by the SSP5- 8.5 scenario. ACCESS- ESM- 1.5 shows the largest range that is 68 times larger than the mean observed EuSHW intensity for SSP5- 8.5 (Fig. 1a), followed by MPI- GE CMIP6 (59 times larger; Fig. 1j), CanESM5 (47 times larger; Fig. 1d), and MIROC6 (36 times larger; Fig. 1g). The qualitative time evolution of the forced signal and the range of possible EuSHW intensity due to internal climate variability reflects the evolution of the underlying radiative forcing, and hence, the evolution of global mean temperature anomalies. + +Sensitivity of EuSHW intensity to global warming levels. We find a robust relationship between global warming levels and the forced signal in four climate models (Fig. 1b,e,h,k), as expected, as well as a robust relationship between global warming levels and the range of possible EuSHW intensity (Fig. 1c,f,i,l). We find that on the one hand, the forced signal in EuSHW intensity increases non- linearly under increasing global mean temperature anomalies. On the other hand, the range of possible EuSHWs computed as ensemble standard deviation increases linearly under increasing global mean temperature anomalies. MPI- GE CMIP6 shows the largest increase in the range of EuSHW intensity under global warming (an increase of 4.5 times the mean observed EuSHW intensity per one degree Celsius increase of global mean temperature anomalies; Fig. 1l), followed by ACCESS- ESM- 1.5 (4.1 times; Fig. 1c), MIROC6 (3.3 times; Fig. 1i) and CanESM5 (2.3 times; Fig. 1f). + +The scaling of the forced signal and the range of EuSHW intensity due to internal variability with global warming anomalies are not spatially uniform (Fig. 2). Although the increase in the forced signal (Fig. 2a) and in the range of EuSHW intensity (Fig. 2b) are projected for entire Europe, the strongest increase is projected for southern Europe and the increase will gradually reduce northward. All models agree on the positive relationship between the forced signal and the range of EuSHW intensity, and global warming anomalies for the entire European domain (Supplementary Fig. 1). Yet, models differ on the magnitude of the increase. MIROC6 shows the largest increase across models in the forced signal and in the range of EuSHW intensity in southern Europe, but it also shows the most abrupt reduction of the increase northward (Supplementary Fig. 1e,f). In contrast, CanESM5 shows the least abrupt reduction of the increase northward (Supplementary Fig. 1c,d), with ACCESS- ESM- 1.5 (Supplementary Fig. 1a,b) and MPI- GE CMIP6 (Supplementary Fig. 1g,h) in between. + +The increase in the forced signal in non- detrended EuSHW intensity and its range is the result of the forced distribution shift and the forced changes in internal variability. The forced distribution shift implies that temperature distributions will shift towards positive values and thus more frequent exceeding the fixed threshold. With a fixed percentile- based heatwave threshold (i.e. assuming a constant perception of what extreme temperatures are for a given day of the summer and grid point) a positive shift in temperature distribution will lead to an increased EuSHW intensity together with an increase in their possible range. The forced changes in internal variability imply a change in the variability of the climate system and hence, of temperatures, which could enhance or suppress the forced signal and the range of EuSHW intensity caused by the shift in temperature distributions. We next disentangle the contribution of the forced changes in internal variability from the forced distribution shift. + +The effects of the forced changes in internal variability. We detrend daily maximum \(2\mathrm{m}\) air temperature by subtracting the ensemble mean to remove the effects of the forced distribution shift, and focus on the effects of forced changes in internal variability (Fig. 3, see methods for details). The recent and projected increase in non- detrended EuSHW intensity seen in Fig. 1 is considerably + +<--- Page Split ---> + +reduced when removing the effect of mean temperature rise, which implies that the forced distribution shift is the main contributor to the increase in EuSHW intensity (Fig. 3a,d,g,j). Yet, we find a consistent positive linear relationship between global warming levels, and the forced signal (Fig. 3b,e,h,k) and the range of detrended EuSHW intensity (Fig. 3c,f,i,l) in all models. The positive linear relationship implies that the forced changes in internal variability will enhance the intensity and variability of EuSHWs caused by the temperature distribution shift, increasing the risk of unprecedented EuSHWs. MPI- GE CMIP6 shows the largest increase in the forced signal and the range of EuSHW intensity across models (an increase of 0.19 and 0.22 times the mean observed EuSHW intensity per one degree Celsius increase of global mean temperature anomalies, respectively; Fig. 3k,l), followed by ACCESS- ESM- 1.5 (0.18 and 0.12 times; Fig. 3b,c), MIROC6 (0.9 and 0.11 times; Fig. 3h,i) and CanESM5 (0.02 and 0.04 times; Fig. 3e,f). + +We find a robust increase in the forced signal and the range of EuSHW intensity with global warming levels emerging from central and northern Europe (Fig. 4a,b). Considering that individual models largely disagree on the regional patterns (Supplementary Fig. 2), we define a robust change in the forced signal and the range of EuSHW intensity (non- dashed regions in Fig. 4a,b) when at least three out of the four models agree on the sign of the change and three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level. All models excluding MIROC6 show an increase in the forced signal and the range of EuSHW intensity in central Europe, and all models excluding CanESM5 show an increase in the forced signal and the range of EuSHW intensity in northern Europe (Supplementary Fig. 2). In central and northern Europe the forced internal variability change will cause a faster increase of the right tail of daily maximum temperatures than the mean under increasing global warming levels (Supplementary Fig. 3), that is, an increase in temperature variability in agreement with ref. \(^{22}\) and ref. \(^{23}\) . A faster increase of extreme summer temperatures compared to the mean leads to the positive trend of the forced signal in EuSHW intensity after detrending daily maximum temperatures. In addition, the increased temperature variability due to the forced internal variability change will lead to a higher fluctuation of extreme temperatures (Supplementary Fig. 4) leading to a larger range of EuSHW intensity. + +In contrast, we find a robust decrease in the forced signal and the range of EuSHWs with global warming levels along the Mediterranean coastline (Fig. 4a,b). While MIROC6 and CanESM5 show a decrease confined to the southern Iberian Peninsula, Italy and Greece, MPI- GE CMIP6 and ACCESS- ESM- 1.5 show also a decrease in the Mediterranean coast and the Balkans (Supplementary Fig. 2). In southern Europe the forced changes in internal variability will cause a slower increase of the right tail of daily maximum temperatures than mean temperatures under global warming levels (Supplementary Fig. 3), that is, a decrease in temperature variability in agreement with ref. \(^{39}\) . A slower increase of extreme temperatures compared to the mean leads to a negative trend of the forced signal in EuSHW intensity after detrending daily maximum temperatures. Furthermore, the decreased temperature variability due to the forced internal variability change will lead to a smaller fluctuation of extreme temperatures (Supplementary Fig. 4) leading to a reduced range of EuSHW intensity. + +The effects of the forced changes in internal variability remain uncertain over other parts of Europe, mainly in eastern Europe and the Balkans (dashed areas in Fig. 4a,b). Climate models considered here show large spatial differences, both on the sign and amplitude of the changes in EuSHWs (Supplementary Fig. 2). For example, while ACCESS- ESM- 1.5 and MIROC6 project a decrease in EuSHW intensity and the range in eastern Europe, CanESM5 and MPI- GE CMIP6 project an increase. Similarly, while ACCESS- ESM- 1.5 and MPI- GE CMIP6 project a decrease in EuSHW intensity and the range in the Balkans, CanESM5 and MIROC6 project an increase. + +Our results imply that while large model uncertainties exist, the forced changes in internal variability will enhance the increase in central and northern EuSHW caused by the forced + +<--- Page Split ---> + +distribution shift increasing the risk of unprecedented central and northern EuSHWs. In contrast, the forced changes in internal variability will partly mitigate the increase in southern EuSHW intensity caused by the forced distribution shift decreasing the risk of unprecedented southern EuSHWs. Therefore, while an adaptation to increasing mean daily maximum temperatures in southern Europe should be enough to reduce the vulnerability to upcoming EuSHWs, in central and northern Europe an adaptation to increased variability will also be needed. + +Relating the changes in EuSHW intensity to the changes in soil moisture. The regional differences in the effects of the forced distribution shift and the forced internal variability change on EuSHW intensity are partly related to the regional differences in soil moisture changes. We find a robust decrease in summer soil moisture content with increasing global warming levels in all Europe excluding some areas southeast of the Baltic Sea (Fig. 5a) which can be related to the robust projected increase in non- detrended EuSHW intensity (Fig. 2). A decrease in soil moisture would generally reduce the cooling effect of latent heat- flux and the heating effect of sensible heat- flux \(^{40,41}\) , leading to the overall increase in EuSHW intensity. + +However, the decrease in European summer soil moisture has an opposite effect on soil moisture variability changes in different European regions (Fig. 5b). We find a robust increase in soil moisture variability in central and northern Europe, where we also find a robust increase in EuSHW intensity and its range due to the forced changes in internal variability (Fig. 4). We also find a robust decrease in soil moisture variability in the southernmost European grid- points coherent with the areas of decreased EuSHW intensity and range due to forced changes in internal variability. Soil moisture in southern Europe is historically lower than in central and northern Europe \(^{42}\) . A further decrease in soil moisture over southern Europe will lead to a decreased variability of soil moisture conditions, resulting in a more constant state of moisture depletion state, which means less often switching between moisture limited and energy limited states, decreasing summer daily maximum temperature variability (Supplementary Fig. 3,4) and mitigating EuSHW intensity and its range due to the forced internal variability change. In contrast, a decrease in soil moisture over central and northern Europe will push the region to switch more often between moisture limited and energy limited states, resulting in an enhancement of land- atmospheric feedback and an increase in soil moisture variability (Fig. 5b), increasing summer daily maximum temperature variability (Supplementary Fig. 3,4) and increasing the amplitude and variability of EuSHW intensity due to the forced internal variability change. + +The inter- model differences in EuSHW intensity change due to the forced changes in internal variability (Fig. 4a,b; Supplementary Fig. 2) are partly related to the inter- model differences in soil moisture variability changes (Fig. 5b). We find an uncertain change of soil moisture variability with global warming levels in eastern Europe and the Balkans, where we also find uncertain changes in EuSHW intensity and its range due to the forced changes in internal variability (Fig. 4). While ACCESS- ESM- 1.5 and MIRO6 project a decrease in soil moisture variability in eastern Europe, CanESM5 and MPI- GE CMIP6 project an increase (Supplementary Fig. 5b,d,f,h). Similarly, while MPI- GE CMIP6 project a decrease in soil moisture variability in the Balkans, MIROC6 projects an increase, and ACCESS- ESM- 1.5 and CanESM5 project an insignificant change. Yet, we also find EuSHW intensity changes that can not be explained with soil moisture changes. The decreased intensity simulated by CanESM5 north of the Baltic Sea (Supplementary Fig. 2c,d) can not be explained by the decrease in soil moisture (Supplementary Fig. 5c) and increase in variability (Supplementary Fig. 5d) under global warming. Similarly, the increased intensity simulated by MPI- GE CMIP6 in northern Europe (Supplementary Fig. 1g,h and Supplementary Fig. 2g,h) can + +<--- Page Split ---> + +not be explained by the increase in soil moisture (Supplementary Fig. 5g) and decrease in variability (Supplementary Fig. 5h) under global warming. It is likely that other factors (e.g. changes in the North Atlantic sea surface temperature variability or atmospheric variability) also contribute to the forced internal variability changes in central and northern EuSHW intensity. + +## Conclusion and discussion + +Anthropogenic forcing will cause a non- linear increase in EuSHW intensity and a linear increase in their intensity range with increasing global warming anomalies, leading to larger EuSHW intensity fluctuations and risk of unprecedented EuSHWs. The main contributor to the increase in EuSHW intensity and the range is the positive shift in summer temperature distributions, which will increase the intensity and range in the entirety of Europe with the strongest impact in the south, in agreement with ref. 11 and ref. 12. Considering a limited adaptation towards increasing European summer temperatures, the threshold at which temperatures are felt extreme will be time- invariant. In consequence, with a positive shift in summer temperature distribution, temperatures will exceed more often the heatwave threshold leading to an increase in EuSHW intensity and in their range. + +In addition, the forced changes in internal variability will enhance the increase in central and northern EuSHW intensity and its range. In central and northern Europe extreme maximum temperatures are projected to increase faster than the mean, which means that temperature variability will increase, enhancing the increase in EuSHW intensity due to the forced shift in temperature distribution. Furthermore, the variability of extreme maximum temperatures is projected to increase, which will increase the range of central and northern EuSHW intensity. In contrast, the forced changes in internal variability will partly mitigate the increase in southern EuSHWs. In southern Europe extreme maximum temperatures are projected to increase slower than the mean, which means that temperature variability will decrease, diminishing the increase in EuSHW intensity due to the forced shift in temperature distribution. Furthermore, the variability of extreme maximum temperatures is projected to decrease, which will decrease the range of southern EuSHW intensity. + +The opposing effects of the forced internal variability change in central- north and southern EuSHWs can be explained by the non- linear effects of decreasing soil moisture. In central and northern Europe a decrease in soil moisture will cause the region to switch more often between moisture limited and energy limited states, amplifying land- atmospheric feedbacks, increasing temperature variability, and enhancing the intensity and variability of EuSHWs. In contrast, in southern Europe a decrease in soil moisture will cause the region to be in a moisture- depletion state more often, which will reduce temperature variability, and the intensity and range of EuSHWs. + +Although we find a robust decrease in EuSHW intensity and their range with global warming in southernmost Europe due to the forced changes in internal variability, and a robust increase in central and northern Europe, models show large differences on the regional patterns and amplitude of the changes in agreement with ref. 43. The model differences highlight large uncertainties for regional assessment of the forced changes in internal variability and the need of more SMILES with high temporal frequency to contrast our results. Model differences partly arise from different projected soil moisture variability changes. For example, the latitude at which the contribution of the forced internal variability change converts from negative to positive is sensitive to the selection of the climate model. While for CanESM5 and MIROC6 the negative effect is constrained to the southernmost European areas, for ACCESS- ESM- 1.5 and MPI- GE CMIP6 the negative contribution extends further north covering the Mediterranean coast and the Balkans. It is likely that the model differences in representing the effect of the forced internal variability partly arise from different historical and projected locations of the transitional zone, or from different soil moisture- vegetation dynamics under global warming. + +<--- Page Split ---> + +We also find regions with projected changes in EuSHW intensity due to forced changes in internal variability that can not be explained by soil moisture variability changes. For example, MPI- GE CMIP6 simulates an increase in soil moisture and a decrease in soil moisture variability in northern Europe which can not lead to an increase in northern EuSHW intensity. Similarly, CanESM5 simulates a decrease in soil moisture and an increase in soil moisture variability which can not lead to a decrease in the intensity of heatwaves north of the Baltic Sea. The discrepancy between the changes in soil moisture and EuSHW intensity under increasing global mean temperatures suggests that other factors will play a role in forcing internal variability changes and consequently affecting the intensity of future EuSHWs. The variability of sea surface temperatures in the North Atlantic and of the jet stream over the North Atlantic and Eurasia have been shown to affect the occurrence and amplitude of historical EuSHWs \(^{5,25,30,44 - 46}\) . It is likely that while global warming affects North Atlantic sea surface temperatures and atmospheric dynamics, these changes will be expressed as forced internal variability changes affecting European summer temperature variability and the intensity of EuSHWs. + +Our findings emphasize the role of soil moisture and moisture- temperature feedbacks in modulating the intensity and variability of EuSHWs under increasing global warming levels with implications for adaptation strategies. Here we show that while an adaptation to increasing mean daily maximum temperatures in southern Europe should be enough to reduce the vulnerability to increasing EuSHW intensity, in central and northern Europe adaptation to increasing mean daily maximum temperatures will not be enough due to the forced changes in internal climate variability. Furthermore, the increased summer soil moisture variability in central and northern Europe can have implications for fire regimes. Wet years that fuel vegetation growth will be followed more frequently by moisture- limited years leading to a drying and burning of the vegetation \(^{32,47,48}\) . Yet, the critical role of soil moisture brings an opportunity as a mitigation strategy for future EuSHWs. In central and northern Europe, the enhancement of increasing summer heatwave intensity due to forced internal variability change could be mitigated by altering the soil moisture content. Crop irrigation management or groundwater conservation policies, for instance, could aid in providing moisture to the soil, suppress land- atmospheric feedback, and reduce the intensity and variability of upcoming EuSHWs. + +## Methods + +## Data + +We use four single model initial condition large ensembles (SMILES): ACCESS- ESM- 1.5 (40 members \(^{33}\) ), CanESM5 (50 members \(^{34}\) ), MIROC6 (50 members \(^{35}\) ), and MPI- GE CMIP6 (50 members \(^{36}\) ). All SMILES considered here provide daily temporal resolution for the historical period (1850- 2014) and two shared socioeconomic pathway scenarios (SSP2- 4.5, and SSP5- 8.5) that span until the end of the \(21^{\text{st}}\) century. The ensemble members have identical natural and anthropogenic forcing but start from different initial conditions allowing us to robustly identify the effects of forced changes in the shift of temperature distribution and internal climate variability \(^{37,38}\) . We test the capability of climate models to simulate the intensity of recent EuSHWs by comparing it to the daily gridded land observations E- OBS dataset for the period 1950- 2022 \(^{49}\) . All datasets are linearly interpolated to the MPI- GE CMIP6 grid using Climate Data Operator (CDO) command line tool. + +We compute two sets (i.e. non- detrended and detrended) of daily maximum 2m air temperature (T2max) anomalies from 1850 until 2099 to remove any bias that might arise when computing the intensity of EuSHWs due to the seasonal cycle \(^{50}\) . The non- detrended T2max anomalies are computed in reference to a centered 15- day running window and 1985- 2014 reference period. The detrended T2max anomalies are computed in reference to a centered 15- day running window over all ensemble members for each year and each model. For E- OBS, we compute the detrended T2max anomalies in reference to a centered 15- day running window over all ensemble members from all models for each year. While the non- detrended set maintains an increasing mean trend in + +<--- Page Split ---> + +temperatures due to global warming, in the detrended set the trend is removed and the effects of forced distribution shift in EuSHW intensity are excluded. + +We use monthly 2m air temperature to compute spatially weighted global mean surface temperature anomalies (GMTA) in reference to the 1850- 1880 period. We compute boreal summer (June, July, August) mean soil moisture as the fraction of water accumulated in the root zone relative to the water capacity for each grid point. + +## EuSHW definition and intensity quantification + +We apply a percentile- based heatwave definition to identify boreal summer (June, July, August) heatwave days \(^{51}\) . Percentile- based heatwave definitions consist of a temperature threshold defined by a percentile that represents the perception of extreme temperatures, and in a temporal constraint to capture the persistent nature of heatwaves. For each calendar day and European land grid point (10°W- 30°E, 35- 70°N), a heatwave day is identified when T2max exceeds the 90th percentile based on a centered 15- day running window and 1985- 2014 reference period for at least six consecutive days. + +We quantify the intensity of EuSHWs using the cumulative heat metric \(^{1}\) . The cumulative heat is computed as a seasonal integration of heat exceeding the threshold during heatwave days. In addition, we integrate the cumulative heat over the European domain after weighting each grid point by the cosine of the latitude \(^{25}\) . Hence, with a single number the cumulative heat accounts for the duration, spatial extent, and amplitude of all EuSHWs that occur in each summer. We refer to non- detrended and detrended EuSHW intensity to the cumulative heat computed with non- detrended T2max and detrended T2max, respectively. + +## Sensitivity of the forced signal and ensemble spread to global warming levels + +For each summer the forced signal is given by the ensemble mean, and the range due to internal climate variability is given by the ensemble spread computed as ensemble standard deviation. We linearly regress the forced signal in EuSHW intensity and soil moisture to global mean surface temperature anomalies. Similarly, we regress the ensemble spread in EuSHW intensity and soil moisture to global mean surface temperature anomalies. We apply a significance test at the 95th confidence level. We define a robust change when at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the 95th confidence level. + +## Data availability + +The gridded land observations E- OBS dataset is publicly available at Copernicus Climate Data Store (https://cds.climate.copernicus.eu/). The climate model data are publicly available at Earth System Grid Federation (ESGF) nodes (https://esgf.llnl.gov/nodes.html) + +## Code availability + +The code used in this study is publicly available in the GitHub repository https://github.com/gbeobidearsuaga/EuSHW projection IV 2025. + +## Acknowledgements + +G.B.A. is supported by the Horizon Europe project EXPECT (Towards an Integrated Capability to Explain and Predict Regional Climate Changes) under Grant Agreement 101137656. L.S.G. received funding from the European Union's Horizon Europe Framework Programme under the Marie Sklodowska- Curie grant agreement No 101064940. A.B and J.B are funded by the Deutsche Forschungsgemeinschaft under Germany's Excellence Strategy - EXC 2037 'CLICCS - Climate, Climatic Change, and Society' - Project Number: 390683824, contribution to the Center for Earth System Research and Sustainability (CEN) of University of Hamburg. D.O. is supported by the + +<--- Page Split ---> + +404 Max Planck Society for the Advancement of Science. We acknowledge the Large Ensemble 405 community for providing the climate model simulations used in this study and the German Climate 406 Computing Centre (DKRZ) for the computational resources. + +## Author contributions + +409 G.B.A. conceived the idea. G.B.A. and L.S.G. designed the study. G.B.A. performed all analysis, 410 prepared the figures and wrote the first version of the manuscript. G.B.A., L.S.G., A.B., D.O. and 411 J.B. contributed to the interpretation of the results and the final version of the manuscript. + +## Competing interests + +The authors declare no competing interests. + +## Corresponding author + +Correspondence to Goratz Beobide-Arsuaga + +<--- Page Split ---> + +419 1. Perkins-Kirkpatrick, S. E. & Lewis, S. C. Increasing trends in regional heatwaves. Nat. Commun. 11, 1-8 (2020). 421 2. Russo, S., Sillmann, J. & Fischer, E. M. Top ten European heatwaves since 1950 and their occurrence in the coming decades. Environ. Res. Lett. 10, (2015). 423 3. Felsche, E., Böhnich, A., Poschlod, B. & Ludwig, R. European hot and dry summers are projected to become more frequent and expand northwards. Commun. Earth Environ. 5, 1-11 (2024). 426 4. IPCC. Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. [Core Writ. Team, H. Lee J. Romero (eds.)]. IPCC, Geneva, Switz. 35-115 (2023) doi:10.59327/IPCC/AR6-9789291691647. 430 5. Rousi, E., Kornhuber, K., Beobide-Arsuaga, G., Luo, F. & Coumou, D. Accelerated western European heatwave trends linked to more-persistent double jets over Eurasia. Nat. Commun. 13, 1-11 (2022). 433 6. Vautard, R. et al. Heat extremes in Western Europe increasing faster than simulated due to atmospheric circulation trends. Nat. Commun. 14, 1-9 (2023). 435 7. Robine, J. M. et al. Death toll exceeded 70,000 in Europe during the summer of 2003. Comptes Rendus - Biol. 331, 171-178 (2008). 437 8. Ballester, J. et al. Heat-related mortality in Europe during the summer of 2022. Nat. Med. 29, 1857-1866 (2023). 439 9. Rodrigues, M. et al. Drivers and implications of the extreme 2022 wildfire season in Southwest Europe. Sci. Total Environ. 859, 160320 (2023). 441 10. García-León, D. et al. Current and projected regional economic impacts of heatwaves in Europe. Nat. Commun. 12, 1-10 (2021). 443 11. Lau, N. C. & Nath, M. J. Model simulation and projection of European heat waves in present-day and future climates. J. Clim. 27, 3713-3730 (2014). 445 12. Lhotka, O., Kysely, J. & Farda, A. Climate change scenarios of heat waves in Central Europe and their uncertainties. Theor. Appl. Climatol. 131, 1043-1054 (2018). 447 13. Patel, R. N., Bonan, D. B. & Schneider, T. Changes in the Frequency of Observed Temperature Extremes Largely Driven by a Distribution Shift. Geophys. Res. Lett. 51, (2024). 450 14. van der Wiel, K. & Bintanja, R. Contribution of climatic changes in mean and variability to monthly temperature and precipitation extremes. Commun. Earth Environ. 2, 1-11 (2021). 452 15. Vogel, M. M., Zscheischler, J., Fischer, E. M. & Seneviratne, S. I. Development of future heatwaves for different hazard thresholds. J. Geophys. Res. Atmos. (2020) doi:10.1029/2019jd032070. + +<--- Page Split ---> + +16. Schar Christoph et al. The role of increasing temperature variability in European summer heatwaves. Nature 427, 328–332 (2004). +17. Bathiany, S., Dakos, V., Scheffer, M. & Lenton, T. M. Climate models predict increasing temperature variability in poor countries. Sci. Adv. 4, 1–10 (2018). +18. Cattiaux, J., Douville, H., Schoetter, R., Parey, S. & Yiou, P. Projected increase in diurnal and interdiurnal variations of European summer temperatures. Geophys. Res. Lett. 42, 899–907 (2015). +19. Lustenberger, A., Knutti, R. & Fischer, E. M. Sensitivity of European extreme daily temperature return levels to projected changes in mean and variance. J. Geophys. Res. 119, 3032–3044 (2014). +20. Olonscheck, D., Schurer, A. P., Lücke, L. & Hegerl, G. C. Large-scale emergence of regional changes in year-to-year temperature variability by the end of the 21st century. Nat. Commun. 12, 2–11 (2021). +21. Suarez-Gutierrez, L., Li, C., Müller, W. A. & Marotzke, J. Internal variability in European summer temperatures at 1.5 °C and 2 °C of global warming. Environ. Res. Lett. 13, (2018). +22. Seneviratne, S. I., Donat, M. G., Pitman, A. J., Knutti, R. & Wilby, R. L. Allowable CO2 emissions based on regional and impact-related climate targets. Nature 529, 477–483 (2016). +23. Suarez-Gutierrez, L., Müller, W. A., Li, C. & Marotzke, J. Dynamical and thermodynamical drivers of variability in European summer heat extremes. Clim. Dyn. 54, 4351–4366 (2020). +24. Simolo, C. & Corti, S. Enhanced risk of hot extremes revealed by observation-constrained model projections. Commun. Earth Environ. 1–11 (2025) doi:10.1038/s43247-025-02133-3. +25. Beobide-Arsuaga, G., Düsterhus, A., Müller, W. A., Barnes, E. A. & Baehr, J. Spring Regional Sea Surface Temperatures as a Precursor of European Summer Heatwaves. Geophys. Res. Lett. 50, 1–8 (2023). +26. Seneviratne, S. I. et al. Investigating soil moisture-climate interactions in a changing climate: A review. Earth-Science Rev. 99, 125–161 (2010). +27. Rousi, E. et al. The extremely hot and dry 2018 summer in central and northern Europe from a multi-faceted weather and climate perspective. Nat. Hazards Earth Syst. Sci. 23, 1699–1718 (2023). +28. Domeisen, D. I. V. et al. Prediction and projection of heatwaves. Nat. Rev. Earth Environ. 4, 36–50 (2023). +29. Della-Marta, P. M. et al. Summer heat waves over western Europe 1880-2003, their relationship to large-scale forcings and predictability. Clim. Dyn. 29, 251–275 (2007). +30. Li, X., Zheng, J., Wang, C., Lin, X. & Yao, Z. Unraveling the roles of jet streams on the unprecedented hot July in Western Europe in 2022. npj Clim. Atmos. Sci. 7, 1–11 (2024). +31. Fischer, E. M., Sippel, S. & Knutti, R. Increasing probability of record-shattering climate extremes. Nat. Clim. Chang. 11, 689–695 (2021). + +<--- Page Split ---> + +32. Suarez-Gutierrez, L., Müller, W. A. & Marotzke, J. Extreme heat and drought typical of an end-of-century climate could occur over Europe soon and repeatedly. Commun. Earth Environ. 4, 1–11 (2023). + +33. Ziehn, T. et al. The Australian Earth System Model: ACCESS-ESM1.5. J. South. Hemisp. Earth Syst. Sci. 70, 193–214 (2020). + +34. Swart, N. C. et al. The Canadian Earth System Model version 5 (CanESM5.0.3). Geosci. Model Dev. 12, 4823–4873 (2019). + +35. Tatebe, H. et al. Description and basic evaluation of simulated mean state, internal variability, and climate sensitivity in MIROC6. Geosci. Model Dev. 12, 2727–2765 (2019). + +36. Olonscheck, D. et al. The New Max Planck Institute Grand Ensemble With CMIP6 Forcing and High-Frequency Model Output. J. Adv. Model. Earth Syst. 15, 1–21 (2023). + +37. Maher, N., Milinski, S. & Ludwig, R. Large ensemble climate model simulations: Introduction, overview, and future prospects for utilising multiple types of large ensemble. Earth Syst. Dyn. 12, 401–418 (2021). + +38. Maher, N. et al. The Max Planck Institute Grand Ensemble: Enabling the Exploration of Climate System Variability. J. Adv. Model. Earth Syst. 11, 2050–2069 (2019). + +39. Gross, M. H., Donat, M. G. & Alexander, L. V. Changes in daily temperature extremes relative to the mean in Coupled Model Intercomparison Project Phase 5 models and observations. Int. J. Climatol. 39, 5273–5291 (2019). + +40. Fischer, E. M., Seneviratne, S. I., Vidale, P. L., Lüthi, D. & Schär, C. Soil moisture-atmosphere interactions during the 2003 European summer heat wave. J. Clim. 20, 5081–5099 (2007). + +41. Seneviratne, S. I., Lüthi, D., Litschi, M. & Schär, C. Land-atmosphere coupling and climate change in Europe. Nature 443, 205–209 (2006). + +42. O, S., Orth, R., Weber, U. & Park, S. K. High-resolution European daily soil moisture derived with machine learning (2003–2020). Sci. Data 9, 1–13 (2022). + +43. Marau, D., Schiemann, R., Ossó, A. & Jury, M. Changes in event soil moisture-temperature coupling can intensify very extreme heat beyond expectations. Nat. Commun. 16, 734 (2025). + +44. Duchez, A. et al. Drivers of exceptionally cold North Atlantic Ocean temperatures and their link to the 2015 European heat wave. Environ. Res. Lett. 11, (2016). + +45. Feudale, L. & Shukla, J. Influence of sea surface temperature on the European heat wave of 2003 summer. Part I: an observational study. Clim. Dyn. 36, 1691–1703 (2011). + +46. Wulff, C. O., Greatbatch, R. J., Domeisen, D. I. V., Gollan, G. & Hansen, F. Tropical Forcing of the Summer East Atlantic Pattern. Geophys. Res. Lett. 44, 11,166-11,173 (2017). + +47. O, S., Hou, X. & Orth, R. Observational evidence of wildfire-promoting soil moisture anomalies. Sci. Rep. 10, 1–8 (2020). + +<--- Page Split ---> + +528 48. Aghakouchak, A. et al. Climate Extremes and Compound Hazards in a Warming World. 529 Annu. Rev. Earth Planet. Sci. 48, 519-548 (2020). 530 49. Cornes, R. C., van der Schrier, G., van den Besselaar, E. J. M. & Jones, P. D. An Ensemble 531 Version of the E-OBS Temperature and Precipitation Data Sets. J. Geophys. Res. Atmos. 123, 532 9391-9409 (2018). 533 50. Brunner, L. & Voigt, A. Pitfalls in diagnosing temperature extremes. Nat. Commun. 15, 1-9 534 (2024). 535 51. Perkins, S. E. & Alexander, L. V. On the measurement of heat waves. J. Clim. 26, 4500- 536 4517 (2013). 537 + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Figures:
+ +Figure 1: Changes in European summer heatwave (EuSHW) intensity. a) Observed, historical and projected non- detrended EuSHW intensity expressed as cumulative heat in the left y- axis and scaled it to the observed historical mean (1950- 2022) in the right y- axis. Solid lines indicate the ensemble mean, the shading indicates the 5th and 95th percentile, and the right vertical bars indicate the 5th and 95th percentile for the last 20 years of the historical simulation and each shared socioeconomic pathway (SSP) scenario for ACCESS- ESM- 1.5; b) the forced signal (i.e., ensemble mean) of non- detrended EuSHW cumulative heat scaled to global mean temperature anomalies (GMT) relative to 1850- 1880 for SSP2- 4.5 and SSP5- 8.5 scenarios (2014- 2100) for ACCESS- ESM- 1.5; c) the range (i.e., ensemble spread computed as ensemble standard deviation) of non- detrended EuSHW cumulative heat due to internal variability scaled to GMT relative to 1850- 1880 for SSP2- 4.5 and SSP5- 8.5 scenarios (2014- 2100) for ACCESS- ESM- 1.5; d- f) same as a- c) but for CanESM5; g- i) same as a- c) but for MIROC6; j- l) same as a- c) but for MPI- GE CMIP6. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Spatial distribution of changes in European summer heatwave (EuSHW) intensity under global warming levels. Multi-model mean of regression coefficients for the period 2014-2100 and for SSP2-4.5 and SSP5-8.5 scenarios between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of non-detrended EuSHW cumulative heat; b) the range (i.e., ensemble standard deviation) of non-detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the 95th confidence level.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: Changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability. a) Detrended observed, historical and projected EuSHW intensity expressed as cumulative heat in the left y-axis and scaled it to the observed historical mean (1950-2022) in the right y-axis. Solid lines indicate the ensemble mean, the shading indicates the 5th and 95th percentile, and the right vertical bars indicate the 5th and 95th percentile for the last 20 years of the historical simulation and each shared socioeconomic pathway (SSP) scenario for ACCESS-ESM-1.5; b) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat scaled to global mean temperature anomalies (GMTA) relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; c) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability scaled to GMTA relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; d-f) same as a-c) but for CanESM5; g-i) same as a-c) but for MIROC6; j-l) same as a-c) but for MPI-GE CMIP6.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: Spatial distribution of changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability under global warming levels. Multi-model mean of regression coefficients for the period 2014–2100 and for SSP2-4.5 andSSP5-8.5 scenarios between global mean temperature anomalies relative to1850-1880 and; a) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Spatial distribution of changes in soil moisture under global warming levels. Multimodel mean of regression coefficients between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of summer (June, July, August) mean soil moisture; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of summer mean soil moisture. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +beobidearsuagaetalNatCommunsupplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020_det.mmd b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a618ac7c5c2fbba8932d357f7e7e7c54e1d2c1db --- /dev/null +++ b/preprint/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020/preprint__1395e2b7867fb4d64d3daa1a998a50868ba93a94f712ce4dea2001e1b858e020_det.mmd @@ -0,0 +1,370 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 930, 209]]<|/det|> +# Increasing central and northern European summer heatwave intensity due to forced internal variability changes + +<|ref|>text<|/ref|><|det|>[[44, 230, 465, 277]]<|/det|> +Goratz Beobide Arsuaga goratz.beobide.arsuaga@uni- hamburg.de + +<|ref|>text<|/ref|><|det|>[[44, 302, 904, 345]]<|/det|> +Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universitat Hamburg, Hamburg, Germany https://orcid.org/0000- 0002- 7775- 4906 + +<|ref|>text<|/ref|><|det|>[[44, 350, 245, 369]]<|/det|> +Laura Suarez Gutierrez + +<|ref|>text<|/ref|><|det|>[[44, 371, 951, 413]]<|/det|> +Laboratoire des Sciences du Climat et de l'Environnement, Institut Pierre- Simon Laplace, Paris, France / Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland + +<|ref|>text<|/ref|><|det|>[[44, 418, 248, 437]]<|/det|> +Armineh Barkhordarian + +<|ref|>text<|/ref|><|det|>[[44, 440, 904, 483]]<|/det|> +Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universitat Hamburg, Hamburg, Germany https://orcid.org/0000- 0001- 9786- 8081 + +<|ref|>text<|/ref|><|det|>[[44, 487, 188, 505]]<|/det|> +Dirk Olonscheck + +<|ref|>text<|/ref|><|det|>[[50, 508, 738, 528]]<|/det|> +Max Planck Institute for Meteorology https://orcid.org/0000- 0002- 8962- 1406 + +<|ref|>text<|/ref|><|det|>[[44, 534, 179, 551]]<|/det|> +Johanna Baehr + +<|ref|>text<|/ref|><|det|>[[50, 556, 397, 575]]<|/det|> +https://orcid.org/0000- 0003- 4696- 8941 + +<|ref|>sub_title<|/ref|><|det|>[[44, 617, 103, 635]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 654, 135, 673]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 692, 300, 712]]<|/det|> +Posted Date: April 17th, 2025 + +<|ref|>text<|/ref|><|det|>[[44, 730, 474, 750]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 6377558/v1 + +<|ref|>text<|/ref|><|det|>[[44, 767, 914, 810]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 828, 535, 848]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 884, 943, 927]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on October 30th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 65392- w. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[90, 65, 890, 100]]<|/det|> +# Increasing central and northern European summer heatwave intensity due to forced internal variability changes + +<|ref|>text<|/ref|><|det|>[[90, 113, 875, 131]]<|/det|> +Goratz Beobide- Arsuaga1 (goratz.beobide.arsuaga@uni- hamburg.de), Laura Suarez- Gutierrez2,3, + +<|ref|>text<|/ref|><|det|>[[90, 133, 583, 149]]<|/det|> +Armineh Barkhordarian1, Dirk Olonscheck4, Johanna Baehr1 + +<|ref|>text<|/ref|><|det|>[[90, 152, 835, 181]]<|/det|> +1Institute of Oceanography, Center for Earth System Research and Sustainability (CEN), Universität Hamburg, Hamburg, Germany + +<|ref|>text<|/ref|><|det|>[[90, 182, 630, 196]]<|/det|> +2Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland + +<|ref|>text<|/ref|><|det|>[[90, 198, 789, 212]]<|/det|> +3Laboratoire des Sciences du Climat et de l'Environnement, Institut Pierre- Simon Laplace, Paris, France + +<|ref|>text<|/ref|><|det|>[[90, 213, 491, 226]]<|/det|> +4Max Planck Institute for Meteorology, Hamburg, Germany + +<|ref|>sub_title<|/ref|><|det|>[[92, 248, 183, 264]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[90, 265, 905, 505]]<|/det|> +In recent years European summer heatwaves have strongly intensified due to rising anthropogenic emissions. While European summer heatwaves will continue to intensify due to the warming of summer temperatures, the effects of the changes in internal variability under global warming remain unknown. Employing four single- model initial condition large ensembles, we find that the forced internal variability change is projected to intensify central and northern European summer heatwaves. Central and northern Europe will experience frequent moisture limitations, enhancing land- atmospheric feedback and increasing heatwave intensity and variability. In contrast, the forced internal variability change will contribute to weaken southern European summer heatwaves. Southern Europe is projected to face a more stable moisture- depleted environment that reduces extreme temperature variability and heatwave intensity. Our findings imply that while adaptation to increasing mean temperatures in southern Europe should suffice to reduce the vulnerability to increasing heatwave intensity, in central and northern Europe adaptation to increased temperature variability will also be needed. + +<|ref|>sub_title<|/ref|><|det|>[[92, 535, 222, 550]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[92, 551, 904, 667]]<|/det|> +With human- induced rising global mean temperatures, the intensity, frequency and duration of extreme heat events have increased almost everywhere since the 1950s, and the positive trend is expected to continue in the upcoming decades1- 4. Europe, in particular, has been identified as a heatwave hotspot with increasing intensity trends up to four times higher than the rest of the northern midlatitudes5,6. During the last two decades, the occurrence of multiple unprecedented European summer heatwaves (EuSHWs) (e.g. years 2003, 2015, 2018 and 2022) has caused large economic, ecosystem, and humanitarian losses7- 10. + +<|ref|>text<|/ref|><|det|>[[92, 682, 904, 880]]<|/det|> +The recent increase in EuSHW intensity has been partly explained by the forced shift in climatology towards warmer values11- 14. Under rising global mean temperatures, European summer temperatures are also rising, altering the background mean state and shifting summer temperature distributions. Assuming a limited adaptation towards increasing European summer mean temperatures, the threshold at which temperatures are felt extreme will be time- invariant, and consequently, the positive shift in the temperature distribution leads to an increase in EuSHW intensity15. For instance, considering an ecosystem that historically lived with average summer temperatures of \(20^{\circ}\mathrm{C}\) and extreme temperatures of \(30^{\circ}\mathrm{C}\) , a positive shift in temperature distributions of \(5^{\circ}\mathrm{C}\) will increase the mean to \(25^{\circ}\mathrm{C}\) and extreme temperatures to \(35^{\circ}\mathrm{C}\) . Although the difference between the new averaged summer temperatures and the new extreme temperatures has not changed from the historical period, the ecosystem will experience an increase of extreme heat if it does not adapt to the new climatology. + +<|ref|>text<|/ref|><|det|>[[90, 896, 904, 929]]<|/det|> +In addition to the shift in the temperature distribution, the forced changes in internal variability of the climate system are expected to increase European summer temperature variability14,16- 24 which + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 65, 904, 297]]<|/det|> +could have serious implications for reducing EuSHW vulnerability. The variability in sea surface temperatures, soil moisture content, atmospheric dynamics, and the consequent positive feedbacks involved have been related to historical EuSHWs \(^{5,23,25 - 30}\) . Under rising global mean temperatures, internal climate variability will be forced by altering the mean state and variability of the ocean, soil moisture, atmosphere, and their interactions, affecting the variability of European summer temperatures and the intensity of EuSHWs. An increasing internal variability would lead to a faster increase of extreme temperatures relative to the mean, which could enhance the increase in the heatwave intensity caused by the shift in temperature distributions and alter the intensity range of possible future EuSHWs increasing the risk of unprecedented EuSHWs \(^{31,32}\) . Following the previous example of the ecosystem, if the forced changes in internal variability induce an additional \(2^{\circ}\mathrm{C}\) increase in extreme temperatures, the difference between the new average summer temperatures and the new extreme temperatures will increase from historical times. Therefore, even if the ecosystem adapts to increasing mean temperatures and limit the effects of increasing heatwave intensity due to the distribution shift, this might not be enough to reduce the vulnerability of upcoming EuSHWs. + +<|ref|>text<|/ref|><|det|>[[90, 312, 904, 575]]<|/det|> +While it is well established in the literature that EuSHW intensity will continue to increase due to the shift in summer temperature distributions, how the forced changes in internal climate variability will affect EuSHW intensity under increasing global warming levels remains unknown. Previous studies were limited by the short observational record that did not allow a robust sampling of EuSHWs, or by coarse temporal resolution of climate models that did not allow the identification of heatwave events but rather an assessment of summer mean temperature metrics. Here, we employ four single model initial condition large ensembles (SMILEs): ACCESS- ESM- 1.5 \(^{33}\) (40 members), CanESM5 \(^{34}\) (50 members), MIROC6 \(^{35}\) (50 members), and MPI- GE CMIP6 \(^{36}\) (50 members). All SMILEs considered here provide a daily temporal resolution for the historical period (1850- 2014) and two shared socioeconomic pathway scenarios (SSP2- 4.5, and SSP5- 8.5) that span until the end of the \(21^{\text{st}}\) century. The high temporal resolution enables us to identify heatwave events and quantify their intensity, and the multiple ensemble members for each SMILE enable us to robustly differentiate the effects of changing internal climate variability (i.e. forced changes in internal variability) from shifting temperature distributions (i.e. forced distribution shift) \(^{37,38}\) . Here, for the first time, we investigate the changes in EuSHW intensity and their range due to forced changes in internal variability and forced distribution shift under different global warming levels. + +<|ref|>sub_title<|/ref|><|det|>[[92, 593, 168, 608]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[90, 610, 904, 840]]<|/det|> +Historical and projected European summer heatwave intensity. The intensity of non- detrended EuSHWs (see methods for details) has been increasing over the past two decades leading to the three most extreme observed events in 2022, 2003, and 2015, in the respective order, and the increase is projected to continue until approximately 2040 regardless of the SSP scenarios for all climate models (Fig. 1a,d,g,j). Thereafter, the forced signal in EuSHW intensity, computed as the ensemble mean, and the range of possible future EuSHW intensity due to the internal climate variability will strongly depend on the SSP scenarios. The four climate models indicate that for the SSP2- 4.5 scenario the increase in EuSHW intensity will be relatively constant, and for the SSP5- 8.5 scenario the increase will be non- linear. By the end of the century, CanESM5 shows the highest forced signal in EuSHW intensity (29 times higher than the mean observed EuSHW intensity for SSP2- 4.5, and 125 times higher for SSP5- 8.5; Fig. 1d), followed by ACCESS- ESM- 1.5 (28 times for SSP2- 4.5, and 88 times for SSP5- 8.5; Fig. 1a), MIROC6 (17 times for SSP2- 4.5, and 66 times for SSP5- 8.5; Fig. 1g) and MPI- GE CMIP6 (10 times for SSP2- 4.5, and 55 times for SSP5- 8.5; Fig. 1j), in the respective order. + +<|ref|>text<|/ref|><|det|>[[90, 855, 904, 922]]<|/det|> +The range of possible future EuSHW intensity due to the internal climate variability computed as the difference between the \(95^{\text{th}}\) and the \(5^{\text{th}}\) ensemble percentile follows a similar evolution (Fig. 1a,d,g,j). For each model, the range increases similarly in both forcing scenarios until approximately 2040. However, the range of EuSHWs due to internal climate variability will diverge + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 67, 904, 185]]<|/det|> +towards the end of the \(21^{\mathrm{st}}\) century with the largest increase shown by the SSP5- 8.5 scenario. ACCESS- ESM- 1.5 shows the largest range that is 68 times larger than the mean observed EuSHW intensity for SSP5- 8.5 (Fig. 1a), followed by MPI- GE CMIP6 (59 times larger; Fig. 1j), CanESM5 (47 times larger; Fig. 1d), and MIROC6 (36 times larger; Fig. 1g). The qualitative time evolution of the forced signal and the range of possible EuSHW intensity due to internal climate variability reflects the evolution of the underlying radiative forcing, and hence, the evolution of global mean temperature anomalies. + +<|ref|>text<|/ref|><|det|>[[92, 199, 905, 383]]<|/det|> +Sensitivity of EuSHW intensity to global warming levels. We find a robust relationship between global warming levels and the forced signal in four climate models (Fig. 1b,e,h,k), as expected, as well as a robust relationship between global warming levels and the range of possible EuSHW intensity (Fig. 1c,f,i,l). We find that on the one hand, the forced signal in EuSHW intensity increases non- linearly under increasing global mean temperature anomalies. On the other hand, the range of possible EuSHWs computed as ensemble standard deviation increases linearly under increasing global mean temperature anomalies. MPI- GE CMIP6 shows the largest increase in the range of EuSHW intensity under global warming (an increase of 4.5 times the mean observed EuSHW intensity per one degree Celsius increase of global mean temperature anomalies; Fig. 1l), followed by ACCESS- ESM- 1.5 (4.1 times; Fig. 1c), MIROC6 (3.3 times; Fig. 1i) and CanESM5 (2.3 times; Fig. 1f). + +<|ref|>text<|/ref|><|det|>[[92, 396, 905, 619]]<|/det|> +The scaling of the forced signal and the range of EuSHW intensity due to internal variability with global warming anomalies are not spatially uniform (Fig. 2). Although the increase in the forced signal (Fig. 2a) and in the range of EuSHW intensity (Fig. 2b) are projected for entire Europe, the strongest increase is projected for southern Europe and the increase will gradually reduce northward. All models agree on the positive relationship between the forced signal and the range of EuSHW intensity, and global warming anomalies for the entire European domain (Supplementary Fig. 1). Yet, models differ on the magnitude of the increase. MIROC6 shows the largest increase across models in the forced signal and in the range of EuSHW intensity in southern Europe, but it also shows the most abrupt reduction of the increase northward (Supplementary Fig. 1e,f). In contrast, CanESM5 shows the least abrupt reduction of the increase northward (Supplementary Fig. 1c,d), with ACCESS- ESM- 1.5 (Supplementary Fig. 1a,b) and MPI- GE CMIP6 (Supplementary Fig. 1g,h) in between. + +<|ref|>text<|/ref|><|det|>[[92, 629, 905, 836]]<|/det|> +The increase in the forced signal in non- detrended EuSHW intensity and its range is the result of the forced distribution shift and the forced changes in internal variability. The forced distribution shift implies that temperature distributions will shift towards positive values and thus more frequent exceeding the fixed threshold. With a fixed percentile- based heatwave threshold (i.e. assuming a constant perception of what extreme temperatures are for a given day of the summer and grid point) a positive shift in temperature distribution will lead to an increased EuSHW intensity together with an increase in their possible range. The forced changes in internal variability imply a change in the variability of the climate system and hence, of temperatures, which could enhance or suppress the forced signal and the range of EuSHW intensity caused by the shift in temperature distributions. We next disentangle the contribution of the forced changes in internal variability from the forced distribution shift. + +<|ref|>text<|/ref|><|det|>[[92, 862, 904, 929]]<|/det|> +The effects of the forced changes in internal variability. We detrend daily maximum \(2\mathrm{m}\) air temperature by subtracting the ensemble mean to remove the effects of the forced distribution shift, and focus on the effects of forced changes in internal variability (Fig. 3, see methods for details). The recent and projected increase in non- detrended EuSHW intensity seen in Fig. 1 is considerably + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 66, 904, 247]]<|/det|> +reduced when removing the effect of mean temperature rise, which implies that the forced distribution shift is the main contributor to the increase in EuSHW intensity (Fig. 3a,d,g,j). Yet, we find a consistent positive linear relationship between global warming levels, and the forced signal (Fig. 3b,e,h,k) and the range of detrended EuSHW intensity (Fig. 3c,f,i,l) in all models. The positive linear relationship implies that the forced changes in internal variability will enhance the intensity and variability of EuSHWs caused by the temperature distribution shift, increasing the risk of unprecedented EuSHWs. MPI- GE CMIP6 shows the largest increase in the forced signal and the range of EuSHW intensity across models (an increase of 0.19 and 0.22 times the mean observed EuSHW intensity per one degree Celsius increase of global mean temperature anomalies, respectively; Fig. 3k,l), followed by ACCESS- ESM- 1.5 (0.18 and 0.12 times; Fig. 3b,c), MIROC6 (0.9 and 0.11 times; Fig. 3h,i) and CanESM5 (0.02 and 0.04 times; Fig. 3e,f). + +<|ref|>text<|/ref|><|det|>[[92, 262, 904, 526]]<|/det|> +We find a robust increase in the forced signal and the range of EuSHW intensity with global warming levels emerging from central and northern Europe (Fig. 4a,b). Considering that individual models largely disagree on the regional patterns (Supplementary Fig. 2), we define a robust change in the forced signal and the range of EuSHW intensity (non- dashed regions in Fig. 4a,b) when at least three out of the four models agree on the sign of the change and three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level. All models excluding MIROC6 show an increase in the forced signal and the range of EuSHW intensity in central Europe, and all models excluding CanESM5 show an increase in the forced signal and the range of EuSHW intensity in northern Europe (Supplementary Fig. 2). In central and northern Europe the forced internal variability change will cause a faster increase of the right tail of daily maximum temperatures than the mean under increasing global warming levels (Supplementary Fig. 3), that is, an increase in temperature variability in agreement with ref. \(^{22}\) and ref. \(^{23}\) . A faster increase of extreme summer temperatures compared to the mean leads to the positive trend of the forced signal in EuSHW intensity after detrending daily maximum temperatures. In addition, the increased temperature variability due to the forced internal variability change will lead to a higher fluctuation of extreme temperatures (Supplementary Fig. 4) leading to a larger range of EuSHW intensity. + +<|ref|>text<|/ref|><|det|>[[92, 540, 904, 739]]<|/det|> +In contrast, we find a robust decrease in the forced signal and the range of EuSHWs with global warming levels along the Mediterranean coastline (Fig. 4a,b). While MIROC6 and CanESM5 show a decrease confined to the southern Iberian Peninsula, Italy and Greece, MPI- GE CMIP6 and ACCESS- ESM- 1.5 show also a decrease in the Mediterranean coast and the Balkans (Supplementary Fig. 2). In southern Europe the forced changes in internal variability will cause a slower increase of the right tail of daily maximum temperatures than mean temperatures under global warming levels (Supplementary Fig. 3), that is, a decrease in temperature variability in agreement with ref. \(^{39}\) . A slower increase of extreme temperatures compared to the mean leads to a negative trend of the forced signal in EuSHW intensity after detrending daily maximum temperatures. Furthermore, the decreased temperature variability due to the forced internal variability change will lead to a smaller fluctuation of extreme temperatures (Supplementary Fig. 4) leading to a reduced range of EuSHW intensity. + +<|ref|>text<|/ref|><|det|>[[92, 753, 904, 868]]<|/det|> +The effects of the forced changes in internal variability remain uncertain over other parts of Europe, mainly in eastern Europe and the Balkans (dashed areas in Fig. 4a,b). Climate models considered here show large spatial differences, both on the sign and amplitude of the changes in EuSHWs (Supplementary Fig. 2). For example, while ACCESS- ESM- 1.5 and MIROC6 project a decrease in EuSHW intensity and the range in eastern Europe, CanESM5 and MPI- GE CMIP6 project an increase. Similarly, while ACCESS- ESM- 1.5 and MPI- GE CMIP6 project a decrease in EuSHW intensity and the range in the Balkans, CanESM5 and MIROC6 project an increase. + +<|ref|>text<|/ref|><|det|>[[92, 884, 904, 917]]<|/det|> +Our results imply that while large model uncertainties exist, the forced changes in internal variability will enhance the increase in central and northern EuSHW caused by the forced + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 65, 904, 166]]<|/det|> +distribution shift increasing the risk of unprecedented central and northern EuSHWs. In contrast, the forced changes in internal variability will partly mitigate the increase in southern EuSHW intensity caused by the forced distribution shift decreasing the risk of unprecedented southern EuSHWs. Therefore, while an adaptation to increasing mean daily maximum temperatures in southern Europe should be enough to reduce the vulnerability to upcoming EuSHWs, in central and northern Europe an adaptation to increased variability will also be needed. + +<|ref|>text<|/ref|><|det|>[[92, 180, 904, 331]]<|/det|> +Relating the changes in EuSHW intensity to the changes in soil moisture. The regional differences in the effects of the forced distribution shift and the forced internal variability change on EuSHW intensity are partly related to the regional differences in soil moisture changes. We find a robust decrease in summer soil moisture content with increasing global warming levels in all Europe excluding some areas southeast of the Baltic Sea (Fig. 5a) which can be related to the robust projected increase in non- detrended EuSHW intensity (Fig. 2). A decrease in soil moisture would generally reduce the cooling effect of latent heat- flux and the heating effect of sensible heat- flux \(^{40,41}\) , leading to the overall increase in EuSHW intensity. + +<|ref|>text<|/ref|><|det|>[[92, 340, 904, 660]]<|/det|> +However, the decrease in European summer soil moisture has an opposite effect on soil moisture variability changes in different European regions (Fig. 5b). We find a robust increase in soil moisture variability in central and northern Europe, where we also find a robust increase in EuSHW intensity and its range due to the forced changes in internal variability (Fig. 4). We also find a robust decrease in soil moisture variability in the southernmost European grid- points coherent with the areas of decreased EuSHW intensity and range due to forced changes in internal variability. Soil moisture in southern Europe is historically lower than in central and northern Europe \(^{42}\) . A further decrease in soil moisture over southern Europe will lead to a decreased variability of soil moisture conditions, resulting in a more constant state of moisture depletion state, which means less often switching between moisture limited and energy limited states, decreasing summer daily maximum temperature variability (Supplementary Fig. 3,4) and mitigating EuSHW intensity and its range due to the forced internal variability change. In contrast, a decrease in soil moisture over central and northern Europe will push the region to switch more often between moisture limited and energy limited states, resulting in an enhancement of land- atmospheric feedback and an increase in soil moisture variability (Fig. 5b), increasing summer daily maximum temperature variability (Supplementary Fig. 3,4) and increasing the amplitude and variability of EuSHW intensity due to the forced internal variability change. + +<|ref|>text<|/ref|><|det|>[[92, 668, 904, 931]]<|/det|> +The inter- model differences in EuSHW intensity change due to the forced changes in internal variability (Fig. 4a,b; Supplementary Fig. 2) are partly related to the inter- model differences in soil moisture variability changes (Fig. 5b). We find an uncertain change of soil moisture variability with global warming levels in eastern Europe and the Balkans, where we also find uncertain changes in EuSHW intensity and its range due to the forced changes in internal variability (Fig. 4). While ACCESS- ESM- 1.5 and MIRO6 project a decrease in soil moisture variability in eastern Europe, CanESM5 and MPI- GE CMIP6 project an increase (Supplementary Fig. 5b,d,f,h). Similarly, while MPI- GE CMIP6 project a decrease in soil moisture variability in the Balkans, MIROC6 projects an increase, and ACCESS- ESM- 1.5 and CanESM5 project an insignificant change. Yet, we also find EuSHW intensity changes that can not be explained with soil moisture changes. The decreased intensity simulated by CanESM5 north of the Baltic Sea (Supplementary Fig. 2c,d) can not be explained by the decrease in soil moisture (Supplementary Fig. 5c) and increase in variability (Supplementary Fig. 5d) under global warming. Similarly, the increased intensity simulated by MPI- GE CMIP6 in northern Europe (Supplementary Fig. 1g,h and Supplementary Fig. 2g,h) can + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 66, 904, 139]]<|/det|> +not be explained by the increase in soil moisture (Supplementary Fig. 5g) and decrease in variability (Supplementary Fig. 5h) under global warming. It is likely that other factors (e.g. changes in the North Atlantic sea surface temperature variability or atmospheric variability) also contribute to the forced internal variability changes in central and northern EuSHW intensity. + +<|ref|>sub_title<|/ref|><|det|>[[93, 150, 359, 168]]<|/det|> +## Conclusion and discussion + +<|ref|>text<|/ref|><|det|>[[92, 169, 904, 318]]<|/det|> +Anthropogenic forcing will cause a non- linear increase in EuSHW intensity and a linear increase in their intensity range with increasing global warming anomalies, leading to larger EuSHW intensity fluctuations and risk of unprecedented EuSHWs. The main contributor to the increase in EuSHW intensity and the range is the positive shift in summer temperature distributions, which will increase the intensity and range in the entirety of Europe with the strongest impact in the south, in agreement with ref. 11 and ref. 12. Considering a limited adaptation towards increasing European summer temperatures, the threshold at which temperatures are felt extreme will be time- invariant. In consequence, with a positive shift in summer temperature distribution, temperatures will exceed more often the heatwave threshold leading to an increase in EuSHW intensity and in their range. + +<|ref|>text<|/ref|><|det|>[[92, 332, 904, 530]]<|/det|> +In addition, the forced changes in internal variability will enhance the increase in central and northern EuSHW intensity and its range. In central and northern Europe extreme maximum temperatures are projected to increase faster than the mean, which means that temperature variability will increase, enhancing the increase in EuSHW intensity due to the forced shift in temperature distribution. Furthermore, the variability of extreme maximum temperatures is projected to increase, which will increase the range of central and northern EuSHW intensity. In contrast, the forced changes in internal variability will partly mitigate the increase in southern EuSHWs. In southern Europe extreme maximum temperatures are projected to increase slower than the mean, which means that temperature variability will decrease, diminishing the increase in EuSHW intensity due to the forced shift in temperature distribution. Furthermore, the variability of extreme maximum temperatures is projected to decrease, which will decrease the range of southern EuSHW intensity. + +<|ref|>text<|/ref|><|det|>[[92, 544, 904, 660]]<|/det|> +The opposing effects of the forced internal variability change in central- north and southern EuSHWs can be explained by the non- linear effects of decreasing soil moisture. In central and northern Europe a decrease in soil moisture will cause the region to switch more often between moisture limited and energy limited states, amplifying land- atmospheric feedbacks, increasing temperature variability, and enhancing the intensity and variability of EuSHWs. In contrast, in southern Europe a decrease in soil moisture will cause the region to be in a moisture- depletion state more often, which will reduce temperature variability, and the intensity and range of EuSHWs. + +<|ref|>text<|/ref|><|det|>[[92, 675, 904, 905]]<|/det|> +Although we find a robust decrease in EuSHW intensity and their range with global warming in southernmost Europe due to the forced changes in internal variability, and a robust increase in central and northern Europe, models show large differences on the regional patterns and amplitude of the changes in agreement with ref. 43. The model differences highlight large uncertainties for regional assessment of the forced changes in internal variability and the need of more SMILES with high temporal frequency to contrast our results. Model differences partly arise from different projected soil moisture variability changes. For example, the latitude at which the contribution of the forced internal variability change converts from negative to positive is sensitive to the selection of the climate model. While for CanESM5 and MIROC6 the negative effect is constrained to the southernmost European areas, for ACCESS- ESM- 1.5 and MPI- GE CMIP6 the negative contribution extends further north covering the Mediterranean coast and the Balkans. It is likely that the model differences in representing the effect of the forced internal variability partly arise from different historical and projected locations of the transitional zone, or from different soil moisture- vegetation dynamics under global warming. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 65, 904, 296]]<|/det|> +We also find regions with projected changes in EuSHW intensity due to forced changes in internal variability that can not be explained by soil moisture variability changes. For example, MPI- GE CMIP6 simulates an increase in soil moisture and a decrease in soil moisture variability in northern Europe which can not lead to an increase in northern EuSHW intensity. Similarly, CanESM5 simulates a decrease in soil moisture and an increase in soil moisture variability which can not lead to a decrease in the intensity of heatwaves north of the Baltic Sea. The discrepancy between the changes in soil moisture and EuSHW intensity under increasing global mean temperatures suggests that other factors will play a role in forcing internal variability changes and consequently affecting the intensity of future EuSHWs. The variability of sea surface temperatures in the North Atlantic and of the jet stream over the North Atlantic and Eurasia have been shown to affect the occurrence and amplitude of historical EuSHWs \(^{5,25,30,44 - 46}\) . It is likely that while global warming affects North Atlantic sea surface temperatures and atmospheric dynamics, these changes will be expressed as forced internal variability changes affecting European summer temperature variability and the intensity of EuSHWs. + +<|ref|>text<|/ref|><|det|>[[92, 311, 904, 558]]<|/det|> +Our findings emphasize the role of soil moisture and moisture- temperature feedbacks in modulating the intensity and variability of EuSHWs under increasing global warming levels with implications for adaptation strategies. Here we show that while an adaptation to increasing mean daily maximum temperatures in southern Europe should be enough to reduce the vulnerability to increasing EuSHW intensity, in central and northern Europe adaptation to increasing mean daily maximum temperatures will not be enough due to the forced changes in internal climate variability. Furthermore, the increased summer soil moisture variability in central and northern Europe can have implications for fire regimes. Wet years that fuel vegetation growth will be followed more frequently by moisture- limited years leading to a drying and burning of the vegetation \(^{32,47,48}\) . Yet, the critical role of soil moisture brings an opportunity as a mitigation strategy for future EuSHWs. In central and northern Europe, the enhancement of increasing summer heatwave intensity due to forced internal variability change could be mitigated by altering the soil moisture content. Crop irrigation management or groundwater conservation policies, for instance, could aid in providing moisture to the soil, suppress land- atmospheric feedback, and reduce the intensity and variability of upcoming EuSHWs. + +<|ref|>sub_title<|/ref|><|det|>[[92, 574, 183, 590]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[92, 594, 137, 607]]<|/det|> +## Data + +<|ref|>text<|/ref|><|det|>[[92, 608, 904, 774]]<|/det|> +We use four single model initial condition large ensembles (SMILES): ACCESS- ESM- 1.5 (40 members \(^{33}\) ), CanESM5 (50 members \(^{34}\) ), MIROC6 (50 members \(^{35}\) ), and MPI- GE CMIP6 (50 members \(^{36}\) ). All SMILES considered here provide daily temporal resolution for the historical period (1850- 2014) and two shared socioeconomic pathway scenarios (SSP2- 4.5, and SSP5- 8.5) that span until the end of the \(21^{\text{st}}\) century. The ensemble members have identical natural and anthropogenic forcing but start from different initial conditions allowing us to robustly identify the effects of forced changes in the shift of temperature distribution and internal climate variability \(^{37,38}\) . We test the capability of climate models to simulate the intensity of recent EuSHWs by comparing it to the daily gridded land observations E- OBS dataset for the period 1950- 2022 \(^{49}\) . All datasets are linearly interpolated to the MPI- GE CMIP6 grid using Climate Data Operator (CDO) command line tool. + +<|ref|>text<|/ref|><|det|>[[92, 789, 904, 920]]<|/det|> +We compute two sets (i.e. non- detrended and detrended) of daily maximum 2m air temperature (T2max) anomalies from 1850 until 2099 to remove any bias that might arise when computing the intensity of EuSHWs due to the seasonal cycle \(^{50}\) . The non- detrended T2max anomalies are computed in reference to a centered 15- day running window and 1985- 2014 reference period. The detrended T2max anomalies are computed in reference to a centered 15- day running window over all ensemble members for each year and each model. For E- OBS, we compute the detrended T2max anomalies in reference to a centered 15- day running window over all ensemble members from all models for each year. While the non- detrended set maintains an increasing mean trend in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[92, 66, 904, 99]]<|/det|> +temperatures due to global warming, in the detrended set the trend is removed and the effects of forced distribution shift in EuSHW intensity are excluded. + +<|ref|>text<|/ref|><|det|>[[92, 115, 904, 181]]<|/det|> +We use monthly 2m air temperature to compute spatially weighted global mean surface temperature anomalies (GMTA) in reference to the 1850- 1880 period. We compute boreal summer (June, July, August) mean soil moisture as the fraction of water accumulated in the root zone relative to the water capacity for each grid point. + +<|ref|>sub_title<|/ref|><|det|>[[92, 198, 495, 213]]<|/det|> +## EuSHW definition and intensity quantification + +<|ref|>text<|/ref|><|det|>[[92, 214, 904, 329]]<|/det|> +We apply a percentile- based heatwave definition to identify boreal summer (June, July, August) heatwave days \(^{51}\) . Percentile- based heatwave definitions consist of a temperature threshold defined by a percentile that represents the perception of extreme temperatures, and in a temporal constraint to capture the persistent nature of heatwaves. For each calendar day and European land grid point (10°W- 30°E, 35- 70°N), a heatwave day is identified when T2max exceeds the 90th percentile based on a centered 15- day running window and 1985- 2014 reference period for at least six consecutive days. + +<|ref|>text<|/ref|><|det|>[[92, 344, 904, 460]]<|/det|> +We quantify the intensity of EuSHWs using the cumulative heat metric \(^{1}\) . The cumulative heat is computed as a seasonal integration of heat exceeding the threshold during heatwave days. In addition, we integrate the cumulative heat over the European domain after weighting each grid point by the cosine of the latitude \(^{25}\) . Hence, with a single number the cumulative heat accounts for the duration, spatial extent, and amplitude of all EuSHWs that occur in each summer. We refer to non- detrended and detrended EuSHW intensity to the cumulative heat computed with non- detrended T2max and detrended T2max, respectively. + +<|ref|>sub_title<|/ref|><|det|>[[92, 475, 752, 492]]<|/det|> +## Sensitivity of the forced signal and ensemble spread to global warming levels + +<|ref|>text<|/ref|><|det|>[[92, 492, 904, 621]]<|/det|> +For each summer the forced signal is given by the ensemble mean, and the range due to internal climate variability is given by the ensemble spread computed as ensemble standard deviation. We linearly regress the forced signal in EuSHW intensity and soil moisture to global mean surface temperature anomalies. Similarly, we regress the ensemble spread in EuSHW intensity and soil moisture to global mean surface temperature anomalies. We apply a significance test at the 95th confidence level. We define a robust change when at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the 95th confidence level. + +<|ref|>sub_title<|/ref|><|det|>[[92, 644, 260, 660]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[92, 661, 904, 709]]<|/det|> +The gridded land observations E- OBS dataset is publicly available at Copernicus Climate Data Store (https://cds.climate.copernicus.eu/). The climate model data are publicly available at Earth System Grid Federation (ESGF) nodes (https://esgf.llnl.gov/nodes.html) + +<|ref|>sub_title<|/ref|><|det|>[[92, 727, 265, 744]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[92, 745, 904, 778]]<|/det|> +The code used in this study is publicly available in the GitHub repository https://github.com/gbeobidearsuaga/EuSHW projection IV 2025. + +<|ref|>sub_title<|/ref|><|det|>[[92, 796, 290, 812]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[92, 814, 904, 929]]<|/det|> +G.B.A. is supported by the Horizon Europe project EXPECT (Towards an Integrated Capability to Explain and Predict Regional Climate Changes) under Grant Agreement 101137656. L.S.G. received funding from the European Union's Horizon Europe Framework Programme under the Marie Sklodowska- Curie grant agreement No 101064940. A.B and J.B are funded by the Deutsche Forschungsgemeinschaft under Germany's Excellence Strategy - EXC 2037 'CLICCS - Climate, Climatic Change, and Society' - Project Number: 390683824, contribution to the Center for Earth System Research and Sustainability (CEN) of University of Hamburg. D.O. is supported by the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 65, 904, 118]]<|/det|> +404 Max Planck Society for the Advancement of Science. We acknowledge the Large Ensemble 405 community for providing the climate model simulations used in this study and the German Climate 406 Computing Centre (DKRZ) for the computational resources. + +<|ref|>sub_title<|/ref|><|det|>[[92, 134, 308, 151]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[92, 152, 904, 202]]<|/det|> +409 G.B.A. conceived the idea. G.B.A. and L.S.G. designed the study. G.B.A. performed all analysis, 410 prepared the figures and wrote the first version of the manuscript. G.B.A., L.S.G., A.B., D.O. and 411 J.B. contributed to the interpretation of the results and the final version of the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[92, 220, 296, 237]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[92, 239, 446, 255]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[92, 273, 320, 290]]<|/det|> +## Corresponding author + +<|ref|>text<|/ref|><|det|>[[92, 292, 452, 308]]<|/det|> +Correspondence to Goratz Beobide-Arsuaga + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[35, 68, 904, 907]]<|/det|> +419 1. Perkins-Kirkpatrick, S. E. & Lewis, S. C. Increasing trends in regional heatwaves. Nat. Commun. 11, 1-8 (2020). 421 2. Russo, S., Sillmann, J. & Fischer, E. M. Top ten European heatwaves since 1950 and their occurrence in the coming decades. Environ. Res. Lett. 10, (2015). 423 3. Felsche, E., Böhnich, A., Poschlod, B. & Ludwig, R. European hot and dry summers are projected to become more frequent and expand northwards. Commun. Earth Environ. 5, 1-11 (2024). 426 4. IPCC. Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. [Core Writ. Team, H. Lee J. Romero (eds.)]. IPCC, Geneva, Switz. 35-115 (2023) doi:10.59327/IPCC/AR6-9789291691647. 430 5. Rousi, E., Kornhuber, K., Beobide-Arsuaga, G., Luo, F. & Coumou, D. Accelerated western European heatwave trends linked to more-persistent double jets over Eurasia. Nat. Commun. 13, 1-11 (2022). 433 6. Vautard, R. et al. Heat extremes in Western Europe increasing faster than simulated due to atmospheric circulation trends. Nat. Commun. 14, 1-9 (2023). 435 7. Robine, J. M. et al. Death toll exceeded 70,000 in Europe during the summer of 2003. Comptes Rendus - Biol. 331, 171-178 (2008). 437 8. Ballester, J. et al. Heat-related mortality in Europe during the summer of 2022. Nat. Med. 29, 1857-1866 (2023). 439 9. Rodrigues, M. et al. Drivers and implications of the extreme 2022 wildfire season in Southwest Europe. Sci. Total Environ. 859, 160320 (2023). 441 10. García-León, D. et al. Current and projected regional economic impacts of heatwaves in Europe. Nat. Commun. 12, 1-10 (2021). 443 11. Lau, N. C. & Nath, M. J. Model simulation and projection of European heat waves in present-day and future climates. J. Clim. 27, 3713-3730 (2014). 445 12. Lhotka, O., Kysely, J. & Farda, A. Climate change scenarios of heat waves in Central Europe and their uncertainties. Theor. Appl. Climatol. 131, 1043-1054 (2018). 447 13. Patel, R. N., Bonan, D. B. & Schneider, T. Changes in the Frequency of Observed Temperature Extremes Largely Driven by a Distribution Shift. Geophys. Res. Lett. 51, (2024). 450 14. van der Wiel, K. & Bintanja, R. Contribution of climatic changes in mean and variability to monthly temperature and precipitation extremes. Commun. Earth Environ. 2, 1-11 (2021). 452 15. Vogel, M. M., Zscheischler, J., Fischer, E. M. & Seneviratne, S. I. Development of future heatwaves for different hazard thresholds. J. Geophys. Res. Atmos. (2020) doi:10.1029/2019jd032070. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[35, 60, 902, 920]]<|/det|> +16. Schar Christoph et al. The role of increasing temperature variability in European summer heatwaves. Nature 427, 328–332 (2004). +17. Bathiany, S., Dakos, V., Scheffer, M. & Lenton, T. M. Climate models predict increasing temperature variability in poor countries. Sci. Adv. 4, 1–10 (2018). +18. Cattiaux, J., Douville, H., Schoetter, R., Parey, S. & Yiou, P. Projected increase in diurnal and interdiurnal variations of European summer temperatures. Geophys. Res. Lett. 42, 899–907 (2015). +19. Lustenberger, A., Knutti, R. & Fischer, E. M. Sensitivity of European extreme daily temperature return levels to projected changes in mean and variance. J. Geophys. Res. 119, 3032–3044 (2014). +20. Olonscheck, D., Schurer, A. P., Lücke, L. & Hegerl, G. C. Large-scale emergence of regional changes in year-to-year temperature variability by the end of the 21st century. Nat. Commun. 12, 2–11 (2021). +21. Suarez-Gutierrez, L., Li, C., Müller, W. A. & Marotzke, J. Internal variability in European summer temperatures at 1.5 °C and 2 °C of global warming. Environ. Res. Lett. 13, (2018). +22. Seneviratne, S. I., Donat, M. G., Pitman, A. J., Knutti, R. & Wilby, R. L. Allowable CO2 emissions based on regional and impact-related climate targets. Nature 529, 477–483 (2016). +23. Suarez-Gutierrez, L., Müller, W. A., Li, C. & Marotzke, J. Dynamical and thermodynamical drivers of variability in European summer heat extremes. Clim. Dyn. 54, 4351–4366 (2020). +24. Simolo, C. & Corti, S. Enhanced risk of hot extremes revealed by observation-constrained model projections. Commun. Earth Environ. 1–11 (2025) doi:10.1038/s43247-025-02133-3. +25. Beobide-Arsuaga, G., Düsterhus, A., Müller, W. A., Barnes, E. A. & Baehr, J. Spring Regional Sea Surface Temperatures as a Precursor of European Summer Heatwaves. Geophys. Res. Lett. 50, 1–8 (2023). +26. Seneviratne, S. I. et al. Investigating soil moisture-climate interactions in a changing climate: A review. Earth-Science Rev. 99, 125–161 (2010). +27. Rousi, E. et al. The extremely hot and dry 2018 summer in central and northern Europe from a multi-faceted weather and climate perspective. Nat. Hazards Earth Syst. Sci. 23, 1699–1718 (2023). +28. Domeisen, D. I. V. et al. Prediction and projection of heatwaves. Nat. Rev. Earth Environ. 4, 36–50 (2023). +29. Della-Marta, P. M. et al. Summer heat waves over western Europe 1880-2003, their relationship to large-scale forcings and predictability. Clim. Dyn. 29, 251–275 (2007). +30. Li, X., Zheng, J., Wang, C., Lin, X. & Yao, Z. Unraveling the roles of jet streams on the unprecedented hot July in Western Europe in 2022. npj Clim. Atmos. Sci. 7, 1–11 (2024). +31. Fischer, E. M., Sippel, S. & Knutti, R. Increasing probability of record-shattering climate extremes. Nat. Clim. Chang. 11, 689–695 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[35, 67, 888, 103]]<|/det|> +32. Suarez-Gutierrez, L., Müller, W. A. & Marotzke, J. Extreme heat and drought typical of an end-of-century climate could occur over Europe soon and repeatedly. Commun. Earth Environ. 4, 1–11 (2023). + +<|ref|>text<|/ref|><|det|>[[35, 132, 884, 169]]<|/det|> +33. Ziehn, T. et al. The Australian Earth System Model: ACCESS-ESM1.5. J. South. Hemisp. Earth Syst. Sci. 70, 193–214 (2020). + +<|ref|>text<|/ref|><|det|>[[35, 180, 866, 216]]<|/det|> +34. Swart, N. C. et al. The Canadian Earth System Model version 5 (CanESM5.0.3). Geosci. Model Dev. 12, 4823–4873 (2019). + +<|ref|>text<|/ref|><|det|>[[35, 228, 884, 265]]<|/det|> +35. Tatebe, H. et al. Description and basic evaluation of simulated mean state, internal variability, and climate sensitivity in MIROC6. Geosci. Model Dev. 12, 2727–2765 (2019). + +<|ref|>text<|/ref|><|det|>[[35, 276, 888, 313]]<|/det|> +36. Olonscheck, D. et al. The New Max Planck Institute Grand Ensemble With CMIP6 Forcing and High-Frequency Model Output. J. Adv. Model. Earth Syst. 15, 1–21 (2023). + +<|ref|>text<|/ref|><|det|>[[35, 323, 879, 380]]<|/det|> +37. Maher, N., Milinski, S. & Ludwig, R. Large ensemble climate model simulations: Introduction, overview, and future prospects for utilising multiple types of large ensemble. Earth Syst. Dyn. 12, 401–418 (2021). + +<|ref|>text<|/ref|><|det|>[[35, 391, 861, 428]]<|/det|> +38. Maher, N. et al. The Max Planck Institute Grand Ensemble: Enabling the Exploration of Climate System Variability. J. Adv. Model. Earth Syst. 11, 2050–2069 (2019). + +<|ref|>text<|/ref|><|det|>[[35, 439, 849, 495]]<|/det|> +39. Gross, M. H., Donat, M. G. & Alexander, L. V. Changes in daily temperature extremes relative to the mean in Coupled Model Intercomparison Project Phase 5 models and observations. Int. J. Climatol. 39, 5273–5291 (2019). + +<|ref|>text<|/ref|><|det|>[[35, 506, 872, 562]]<|/det|> +40. Fischer, E. M., Seneviratne, S. I., Vidale, P. L., Lüthi, D. & Schär, C. Soil moisture-atmosphere interactions during the 2003 European summer heat wave. J. Clim. 20, 5081–5099 (2007). + +<|ref|>text<|/ref|><|det|>[[35, 574, 890, 610]]<|/det|> +41. Seneviratne, S. I., Lüthi, D., Litschi, M. & Schär, C. Land-atmosphere coupling and climate change in Europe. Nature 443, 205–209 (2006). + +<|ref|>text<|/ref|><|det|>[[35, 621, 840, 658]]<|/det|> +42. O, S., Orth, R., Weber, U. & Park, S. K. High-resolution European daily soil moisture derived with machine learning (2003–2020). Sci. Data 9, 1–13 (2022). + +<|ref|>text<|/ref|><|det|>[[35, 669, 904, 705]]<|/det|> +43. Marau, D., Schiemann, R., Ossó, A. & Jury, M. Changes in event soil moisture-temperature coupling can intensify very extreme heat beyond expectations. Nat. Commun. 16, 734 (2025). + +<|ref|>text<|/ref|><|det|>[[35, 716, 895, 752]]<|/det|> +44. Duchez, A. et al. Drivers of exceptionally cold North Atlantic Ocean temperatures and their link to the 2015 European heat wave. Environ. Res. Lett. 11, (2016). + +<|ref|>text<|/ref|><|det|>[[35, 764, 892, 800]]<|/det|> +45. Feudale, L. & Shukla, J. Influence of sea surface temperature on the European heat wave of 2003 summer. Part I: an observational study. Clim. Dyn. 36, 1691–1703 (2011). + +<|ref|>text<|/ref|><|det|>[[35, 811, 900, 848]]<|/det|> +46. Wulff, C. O., Greatbatch, R. J., Domeisen, D. I. V., Gollan, G. & Hansen, F. Tropical Forcing of the Summer East Atlantic Pattern. Geophys. Res. Lett. 44, 11,166-11,173 (2017). + +<|ref|>text<|/ref|><|det|>[[35, 859, 843, 895]]<|/det|> +47. O, S., Hou, X. & Orth, R. Observational evidence of wildfire-promoting soil moisture anomalies. Sci. Rep. 10, 1–8 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[35, 65, 900, 277]]<|/det|> +528 48. Aghakouchak, A. et al. Climate Extremes and Compound Hazards in a Warming World. 529 Annu. Rev. Earth Planet. Sci. 48, 519-548 (2020). 530 49. Cornes, R. C., van der Schrier, G., van den Besselaar, E. J. M. & Jones, P. D. An Ensemble 531 Version of the E-OBS Temperature and Precipitation Data Sets. J. Geophys. Res. Atmos. 123, 532 9391-9409 (2018). 533 50. Brunner, L. & Voigt, A. Pitfalls in diagnosing temperature extremes. Nat. Commun. 15, 1-9 534 (2024). 535 51. Perkins, S. E. & Alexander, L. V. On the measurement of heat waves. J. Clim. 26, 4500- 536 4517 (2013). 537 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[75, 113, 921, 448]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[90, 66, 161, 82]]<|/det|> +
Figures:
+ +<|ref|>text<|/ref|><|det|>[[77, 451, 918, 650]]<|/det|> +Figure 1: Changes in European summer heatwave (EuSHW) intensity. a) Observed, historical and projected non- detrended EuSHW intensity expressed as cumulative heat in the left y- axis and scaled it to the observed historical mean (1950- 2022) in the right y- axis. Solid lines indicate the ensemble mean, the shading indicates the 5th and 95th percentile, and the right vertical bars indicate the 5th and 95th percentile for the last 20 years of the historical simulation and each shared socioeconomic pathway (SSP) scenario for ACCESS- ESM- 1.5; b) the forced signal (i.e., ensemble mean) of non- detrended EuSHW cumulative heat scaled to global mean temperature anomalies (GMT) relative to 1850- 1880 for SSP2- 4.5 and SSP5- 8.5 scenarios (2014- 2100) for ACCESS- ESM- 1.5; c) the range (i.e., ensemble spread computed as ensemble standard deviation) of non- detrended EuSHW cumulative heat due to internal variability scaled to GMT relative to 1850- 1880 for SSP2- 4.5 and SSP5- 8.5 scenarios (2014- 2100) for ACCESS- ESM- 1.5; d- f) same as a- c) but for CanESM5; g- i) same as a- c) but for MIROC6; j- l) same as a- c) but for MPI- GE CMIP6. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[93, 81, 905, 285]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 288, 886, 420]]<|/det|> +
Figure 2: Spatial distribution of changes in European summer heatwave (EuSHW) intensity under global warming levels. Multi-model mean of regression coefficients for the period 2014-2100 and for SSP2-4.5 and SSP5-8.5 scenarios between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of non-detrended EuSHW cumulative heat; b) the range (i.e., ensemble standard deviation) of non-detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the 95th confidence level.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 80, 907, 405]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 409, 902, 622]]<|/det|> +
Figure 3: Changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability. a) Detrended observed, historical and projected EuSHW intensity expressed as cumulative heat in the left y-axis and scaled it to the observed historical mean (1950-2022) in the right y-axis. Solid lines indicate the ensemble mean, the shading indicates the 5th and 95th percentile, and the right vertical bars indicate the 5th and 95th percentile for the last 20 years of the historical simulation and each shared socioeconomic pathway (SSP) scenario for ACCESS-ESM-1.5; b) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat scaled to global mean temperature anomalies (GMTA) relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; c) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability scaled to GMTA relative to 1850-1880 for SSP2-4.5 and SSP5-8.5 scenarios (2014-2100) for ACCESS-ESM-1.5; d-f) same as a-c) but for CanESM5; g-i) same as a-c) but for MIROC6; j-l) same as a-c) but for MPI-GE CMIP6.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[93, 81, 904, 284]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[91, 286, 901, 419]]<|/det|> +
Figure 4: Spatial distribution of changes in European summer heatwave (EuSHW) intensity due to forced changes in internal variability under global warming levels. Multi-model mean of regression coefficients for the period 2014–2100 and for SSP2-4.5 andSSP5-8.5 scenarios between global mean temperature anomalies relative to1850-1880 and; a) the forced signal (i.e., ensemble mean) of detrended EuSHW cumulative heat; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of detrended EuSHW cumulative heat due to internal variability. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[93, 97, 905, 292]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[92, 293, 900, 409]]<|/det|> +
Figure 5: Spatial distribution of changes in soil moisture under global warming levels. Multimodel mean of regression coefficients between global mean temperature anomalies relative to 1850-1880 and; a) the forced signal (i.e., ensemble mean) of summer (June, July, August) mean soil moisture; b) the range (i.e., ensemble spread computed as ensemble standard deviation) of summer mean soil moisture. In the non-dashed regions at least three out of the four models agree on the regression coefficient sign, and at least three out of the four models show significant changes at the \(95^{\text{th}}\) confidence level.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 507, 150]]<|/det|> +beobidearsuagaetalNatCommunsupplement.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/images_list.json b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..22e71f0e017a7690d0ca0fa79db0d01fe9ad3dd1 --- /dev/null +++ b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/images_list.json @@ -0,0 +1,152 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1: Schematic of an example network with DERs (with and without grid-forming ability), and sectionalizing/tie switches. Outage scenario 1 isolates a small portion of the network, which can be restored by closing the tie line connecting the affected section to the main grid (substation). In the subsequently connected network formed after reconfiguration, the substation serves as the source bus. In scenario 2, a larger part of the network is disconnected without any means to reroute power from the substation. By selecting suitable switching actions, an intentional island is created around the grid-forming DER in scenario 2. The grid-forming DER is the virtual source in the reconfigured network. In scenario 3 with extensive outage, although an intentional island may be formed around the grid-forming DER, a section of the network remains unsupplied despite the presence of a DER (not grid-forming).", + "footnote": [], + "bbox": [ + [ + 212, + 163, + 824, + 433 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: The learning framework developed which includes the environment and the policy network architecture with GNN-based feature abstraction. The environment is composed of a DN model interface (DSS circuit) using a python-based API and the graph replica of the DSS circuit. Voltage sources are introduced at virtual slacks for intentional islanding scenarios. The policy network uses the state information from the environment to compute graph node embeddings and context node embeddings using GNN and feedforward networks, respectively. A final feature vector that encompasses the two embeddings is computed by an MLP.", + "footnote": [], + "bbox": [ + [ + 218, + 85, + 822, + 494 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: The training convergence plots for the policy models. Our GCAPS-based RL model is compared with the MLP model which does not utilize graph abstraction. (a) Convergence plot of GCAPS and MLP for the 13-bus network. (b) Convergence plot for the 34-bus network.", + "footnote": [], + "bbox": [ + [ + 164, + 95, + 761, + 290 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4: Test networks used to validate the proposed GCAPS model for real-time resilient control. The test networks are IEEE distribution networks modified by the addition of switches and DERs. Switches in the networks include sectionalizing and tie switches. Both types of DERs, i.e., grid-forming and grid-feeding, are considered. (a) Modified 13-bus IEEE test network. (b) Modified 34-bus IEEE test network.", + "footnote": [], + "bbox": [ + [ + 196, + 81, + 834, + 280 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5: Status of decision variables acquired from the proposed model and baselines for the 13-bus test network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outage '670-671'. (b) Status of decision variables for test scenario 2 with line outages '632-670', '671-692', and '646-684'.", + "footnote": [], + "bbox": [ + [ + 159, + 91, + 789, + 200 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6: Voltage plot of the 13-bus test network with the GCAPS outage management solution for varying test scenarios. The active phases of the network are only marked in the plot and the grey-shaded nodes represent those disconnected from the network. (a) Voltage plot for test scenario 1 with the GCAPS solution implemented in the network during outages. (b) Voltage plot for test scenario 2 with the GCAPS solution implemented in the network during outages.", + "footnote": [], + "bbox": [ + [ + 150, + 315, + 788, + 485 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig. 7: Status of decision variables acquired from the proposed model and baselines for the 34-bus network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outages '812-814' and '824-848'. (b) Status of decision variables for test scenario 2 with line outages '832-858', '834-860', and '854-852'.", + "footnote": [], + "bbox": [ + [ + 198, + 407, + 835, + 508 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig. 8: Voltage plot of the 34-bus test network for varying outage scenarios with the GCAPS outage management solution. The active phases of the network are only included in the plot. The grey-shaded nodes are disconnected from the network. (a) Outage scenario 1. (b) Outage scenario 2.", + "footnote": [], + "bbox": [ + [ + 238, + 92, + 715, + 469 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Fig. 9: Comparison of the resilience improvement for different models with varying outage scenarios. The resilience improvement is measured in terms of equivalent energy served when implementing control actions during outages. (a) Testing on the 13-bus network, where an empty bar for MLP in scenario 2 indicates invalidity due to the operation of the outage switch. (b) Testing on the 34-bus network, with MLP model for scenario 1 invalid due to operation of outage switches.", + "footnote": [], + "bbox": [ + [ + 184, + 90, + 788, + 226 + ] + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/Figure_10.jpg", + "caption": "Fig. 10: The comparison of the proposed GCAPS model with baseline models. The baseline models include an MLP-based model which does not utilize graph abstraction and the (much slower to compute) conventional mathematical programming technique (here MISOCP). The network is in the disrupted state at time instant '0'. (a) Voltage measured at the bus '652' (most stressed bus) is used as an indicator for 13-bus network operational characteristics in outage scenario 1. (b) Equivalent power supplied in the network evolving with switching decisions for 13 bus in outage scenario 1. (c) Voltage measured at the bus '890' (most stressed bus) is used for comparing 34-bus network operational limits in outage scenario 2. (d) Equivalent power supplied in the 34-bus network by different models for scenario 2 and their response with time is illustrated.", + "footnote": [], + "bbox": [ + [ + 238, + 91, + 795, + 385 + ] + ], + "page_idx": 20 + } +] \ No newline at end of file diff --git a/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05.mmd b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05.mmd new file mode 100644 index 0000000000000000000000000000000000000000..67a0c3cdde6310976bfe4b1f679ab8ae883b898b --- /dev/null +++ b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05.mmd @@ -0,0 +1,503 @@ + +# Real-Time Outage Management in Active Distribution Networks Using Reinforcement Learning over Graphs + +Jie Zhang + +jiezhang@utdallas.edu + +The University of Texas at Dallas https://orcid.org/0000- 0001- 7478- 5670 + +Roshni Jacob The University of Texas at Dallas + +Steve Paul University at Buffalo https://orcid.org/0000- 0002- 8138- 5242 + +Souma Chowdhury University at Buffalo + +Yulia Gel The University of Texas at Dallas + +## Article + +Keywords: Network reconfiguration, Graph neural networks, Reinforcement learning, Topology, Resilience, Outage management + +Posted Date: September 6th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3276125/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on June 4th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49207- y. + +<--- Page Split ---> + +# Real-Time Outage Management in Active Distribution Networks Using Reinforcement Learning over Graphs + +Roshni Anna Jacob \(^{1}\) , Steve Paul \(^{2}\) , Souma Chowdhury \(^{2}\) , Yulia R. Gel \(^{3}\) , Jie Zhang \(^{1,4*}\) + +\(^{1}\) Department of Electrical and Computer Engineering, The University of Texas at Dallas, Richardson, 75080, Texas, USA. \(^{2}\) Department of Mechanical and Aerospace Engineering, University at Buffalo, Buffalo, 14260, New York, USA. \(^{3}\) Department of Mathematical Sciences, The University of Texas at Dallas, Richardson, 75080, Texas, USA. \(^{4}\) Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, 75080, Texas, USA. + +## Abstract + +Self- healing "smart grids" are characterized by fast- acting, intelligent control mechanisms that minimize power disruptions during outages. The corrective actions adopted during outages in power distribution networks include reconfiguration through switching control and emergency load shedding. The conventional decision- making models for outage mitigation are, however, not suitable for "smart grids" due to their slow response and computational inefficiency. Here, we present a new reinforcement learning (RL) model for outage management in the distribution network to enhance its resilience. The distinctive characteristic of our approach is that it explicitly accounts for the underlying network topology and its variations with switching control, while also capturing the complex interdependencies between state variables (along nodes and edges) by modeling the task as a graph learning problem. Our model learns the optimal control policy for power restoration using a Capsule- based graph neural network. We validate our model on two test networks, namely the 13 and 34- bus modified IEEE networks where it is shown to achieve near- optimal, real- time performance with up to 5 orders of magnitude improvement in computational speed. The resilience improvement of our model in terms of loss of energy is 4.204 MWs and 13.522 MWs for 13 and 34 + +<--- Page Split ---> + +buses, respectively. Our model also demonstrates generalizability across a broad range of outage scenarios. + +Keywords: Network reconfiguration, Graph neural networks, Reinforcement learning, Topology, Resilience, Outage management. + +## 1 Introduction + +Resilience enhancement of power distribution networks (DNs) has been gaining considerable recognition in recent years, which has been often overlooked before due to the perception of DNs as merely a link between the transmission networks and consumers. A key factor for this shift is the realization that \(90\%\) of customer disruptions during extreme events can be attributed to the failure of components within the distribution network itself [1]. Additionally, the increasing presence of distributed energy resources (DERs) and the resulting decentralization of power generation have spurred the notion of DNs as autonomous entities that can operate independently from the main grid [2]. Consequently, the DN is now considered capable of retaining its functionality even during the loss of connectivity to the transmission network. + +Concurrently, modernization of the power grid and the shift towards "smart grids" have been driving the deployment of intelligent and automated technologies in the DN [3]. The distribution automation has been implemented through the deployment of line monitors, fault indicators, remote- controlled switches, and reclosers in the DN [4]. An important characteristic of the "smart grid" is its self- healing capability, which includes implementing intelligent control actions through automation to minimize power disruptions, thus enabling the recovery of network operations during outages in real time [5]. Therefore, the key requirements of a self- healing tool include autonomy, quick response, and online adaptability, which are indeed the salient features of our model discussed in this paper. + +In the face of power disruptions caused by extreme weather events or cyber- physical attacks, a self- healing DN warrants the automatic detection of faulty components, their isolation, and system restoration (fully or partially) using intelligent control algorithms. This process is referred to as FLISR, which stands for fault location, isolation, and service restoration [5], and is addressed using task- specific techniques. Restoration or the recovery of DN operation can be achieved using different control actions, such as network reconfiguration, load management, DER control, energy storage control, and reactive power resource control. The preliminary control action often adopted in such circumstances is reconfiguration (or switching control), followed by load shedding [6, 7]. Distribution network reconfiguration (DNR) by controlling the status of the network switches is a commonly used strategy to control DN operation for varying objectives such as loss minimization, reliability enhancement, load balancing, increasing penetration of renewable resources, improvement of voltage profile, and service restoration [8- 10]. The purpose of feeder reconfiguration is two- fold: (i) to quickly and efficiently reroute power from the functional part of the DN to the isolated section [11, 12], and (ii) to form intentional islands around the grid forming + +<--- Page Split ---> + +DERs when there exists no connectivity to the main grid [2, 13]. In the existing body of knowledge, these two reconfiguration strategies have been addressed separately and have been largely considered as two distinct domains. However, a comprehensive restoration strategy suitable for various outage scenarios must efficiently utilize both grid- forming and grid- feeding DERs and consider all possible reconfiguration (or switching) options [14]. Hence, in our framework, we consider both DN characterizations through switching by including simultaneously grid- connected and off- grid modes of operation. Additionally, DNR alone may not be sufficient as a restorative action during catastrophic events, as the network remains vulnerable to voltage collapse and system blackouts [7, 15]. Therefore, load shedding becomes necessary as an emergency control mechanism [16] to minimize voltage violations in the DN. + +Furthermore, power distribution networks are typically unbalanced and radial in nature, with a unidirectional power flow from the substation to the consumers. Besides the non- linearity in power flow, the optimization of modern- day DN operation has also been made challenging by the integration of DERs [17]. The DN restoration is an NP- hard, non- linear combinatorial optimization problem that aims to maximize energy supply while considering network connectivity and operational constraints [18]. Various methods have been used in the literature to solve the traditional reconfiguration problem, falling into heuristic [19, 20], meta- heuristic [21- 23], and mixed- integer programming [24- 26] techniques. In line with the increasing penetration of DERs, researchers have also explored islanding strategies using mixed- integer programming models to expand the zone of DER operation [2, 27]. Load management during outages has also been previously investigated as an emergency control strategy [14, 28]. Despite these efforts, a solution incorporating both the grid- connected and islanding (off- grid) reconfiguration schemes for outage management is limited in literature, and presents a complex and challenging problem to solve. The multitude of restorative options depends on the number of controllable devices (switches, loads) and the operational modes of DERs. Although the proliferation of remote- controlled elements in the DNs widens the horizons of automated network control, it also increases the complexity of the underlying non- linear combinatorial optimization problem [29]. The commonly used mixed- integer non- linear programming (MINLP) methodologies for restoration problems face issues of scalability, computational tractability, and real- time decision- making capability [14]. Apart from these, the existing linear programming approximation models in the literature are not designed to address restoration in three- phase unbalanced DNs with sectionalizing, tie switches, and various types of DERs (grid- forming and grid- feeding). Heuristic and meta- heuristic techniques, although explored, tend to be computationally expensive and time- consuming. Moreover, traditional methods heavily rely on a comprehensive description of the DN model and network parameters, making them model- dependent. Considering the uncertainty in network conditions during outages, it is desirable to develop a model capable of adapting to varying circumstances and is deployable online. Here, we present a model based on reinforcement learning to provide online decision support during outages. + +Reinforcement learning (RL) methods have been increasingly adopted in recent years for power system applications that require autonomous control [30]. This is because RL methods are quite effective in solving high- dimensional, combinatorial, + +<--- Page Split ---> + +stochastic optimization problems, besides providing fast- acting control. The latter is imperative to rapid responsiveness during outages, otherwise not possible with conventional optimization based decision support. Particularly with regards to reconfiguration, RL- based models [31, 32] have been developed to perform dynamic DNR during normal operation for loss minimization and voltage improvement. These methods specifically used deep Q- learning with neural networks and trained the off- policy RL network using a historical network operation dataset. The exploration problem which may arise in these models has been addressed by a Noisy- Net Q- learning model [33] developed to perform DNR for similar objectives. Another approach [34], utilized a batch- constrained soft actor- critic algorithm to learn the control policy for loss minimization during normal DN operation. As opposed to the DNR during normal operation considered in these studies, extreme operating conditions are more challenging considering the high- impact, low- probability occurrence of such events. Therefore, availing historical datasets for network operation may also not be possible as in previous studies. Although researchers have explored using RL models for DNR [35, 36] to improve network resilience, such works do not consider the feasibility of network operation based on voltage monitoring and DER operational modes during reconfiguration. Additionally, in methods based on the Q- learning approach, the policy network determines the optimal/near- optimal configuration or the spanning forest, rather than individually controlling each switch. This approach would require enumerating all feasible configurations to define a Q- probability matrix, which is impractical due to the exponential increase in state and action space with network size, possible outage scenarios, and the number of devices. Since these methods are not scalable and require significant storage and computational capabilities for exploration, policy gradient methods are more suitable for learning in outage conditions [34]. We, therefore, employ the proximal policy optimization (PPO), which is a policy gradient method for learning DN outage management in DN. + +In this model, our idea is based on the intrinsic graph representation of power distribution networks. The DN is viewed as a graph where nodes are the buses (i.e., substation, load, or DERs) and edges are the lines or transformers. The state variables of the DN, including demand/generation estimates and voltage/current measurements, can be considered as data superimposed on a graph. The state variables exhibit complex interdependencies, necessitating the extraction of meaningful representations that accurately capture the structure of the DN connectivity. Moreover, outage management, particularly reconfiguration, involves altering the DN connectivity by switching on/off network lines (binary actions) and hence, requires consideration of the underlying combinatorial network structure. Therefore, we present a new graph RL approach for simultaneous real- time control of network topology and loads, ensuring sustained network operations during failures that are caused by extreme events. Graph RL uses a graph neural network or GNN as a policy model (as is the case here) and/or "value" model, as it allows more effective capturing of the combinatorial nature of network- based state information (involving both binary and continuous variables). This advantage is demonstrated in our case studies through comparison with baseline RL- based solutions that use a standard multi- layered perceptron (MLP) based policy model. + +<--- Page Split ---> + +Specifically, we use a Graph Capsule (GCAPS) neural network to learn optimal control policies which is a novel application in power network resilience problems. Compared to other GNNs such as Graph Convolutional Networks (GCN), the capsule- based GNN has been shown by [37] and [38] to better capture the structural information of a graph (the DN in this work) as a graph embedding, where the individual intermediate features of the state are represented as a vector (in GCAPS) as compared to that of a scalar for example in GCN and Graph Attention Networks (GAT), thus giving an enhanced state representation. This enhanced state representation helps in computing better actions compared to other simple feature abstraction networks such as Multi- Layered Perceptron (MLP). Experimental validation of our trained GCAPS based model on test networks demonstrates the generalizability and real- time control capability with near- optimal performance which is desirable in a self- healing tool for DNs. + +## 2 Results + +### 2.1 Reconfiguration & Load Shedding as Emergency Response + +During extreme events in the DN, the occurrence of outages due to component failures can be addressed by a combination of control actions, including reconfiguration and load shedding. We assume that real- time outage detection and protection system responsible for detecting, locating, and isolating faulty components is a preliminary step to the work discussed in this paper. + +Line switches in the DN are typically divided into two categories: switches associated with normally- closed sectionalizing lines and those with normally- open tie lines. During emergency conditions, when component failures disrupt the power supply to the network loads, reconfiguring the DN through control actions on these switches can help maintain network functionality. The objective in such situations is to maximize (or minimize) the energy supplied (or loss of energy) to the loads, despite the network failure, while ensuring operational stability. The optimal switching control depends on factors such as the network state (voltage, branch flow, etc.), network operational limits, and the location and extent of the outage in the network. + +Besides this, the presence of DERs, particularly grid- forming DERs, plays a pivotal role in providing uninterrupted supply to loads following outages. In the off- grid mode, the formation of a self- sustained entity comprising loads and DERs is only possible with the assistance of grid- forming DERs. These grid- forming DERs generate the reference voltage and frequency for the isolated network section while grid- feeding DERs follow this reference and inject active/reactive power into the grid [39]. While the detailed modeling of these DERs is beyond the scope of this work, they are represented as voltage sources when operating in the grid- forming mode, and this characterization is incorporated in the DN model within the environment. + +Reconfiguration is often used as an umbrella term for any change in normal operating network topology using switching control. On the other hand, intentional islanding has long been recognized as a resilience enhancement technique and is a subset of the reconfiguration problem. In scenarios where the outage is extensive and the availability of tie switches is limited, intentional islanding around grid- forming DERs may be + +<--- Page Split ---> + +adopted to ensure a continuous power supply. Figure 1 illustrates different switching actions that may be employed based on the extent of the outage. Different outage scenarios that are portrayed here with mitigation strategies represent the possible solutions considered while designing the environment. + +![](images/Figure_1.jpg) + +
Fig. 1: Schematic of an example network with DERs (with and without grid-forming ability), and sectionalizing/tie switches. Outage scenario 1 isolates a small portion of the network, which can be restored by closing the tie line connecting the affected section to the main grid (substation). In the subsequently connected network formed after reconfiguration, the substation serves as the source bus. In scenario 2, a larger part of the network is disconnected without any means to reroute power from the substation. By selecting suitable switching actions, an intentional island is created around the grid-forming DER in scenario 2. The grid-forming DER is the virtual source in the reconfigured network. In scenario 3 with extensive outage, although an intentional island may be formed around the grid-forming DER, a section of the network remains unsupplied despite the presence of a DER (not grid-forming).
+ +Network topology control through switching actions alone cannot guarantee the operational feasibility of the energized sections in the network. Therefore, to ensure sustainable network operation, emergency load shedding is also considered to maintain network voltage within safe operational limits. The loads are modeled as equivalent load at the distribution transformer in the primary distribution system and can be disconnected from the network through switching actions. + +<--- Page Split ---> + +### 2.2 DN Representation as A Graph + +Outage management in DN using switching control can be largely viewed as a task of learning the associated network topology, which is our motivation to reformulate the problem in graph- theoretic terms. Consequently, we represent the DN as a graph \(\mathcal{G} = (\mathbf{N},\mathbf{E})\) , with an \(\mathbf{N}\) set of nodes interconnected by an \(\mathbf{E}\) set of edges. The nodes in the graph represent the buses in the DN, including the substation, load, DER, and zero- power injection buses. The edges represent the distribution lines and inline transformers. These lines (edges) consist of both switchable (sectionalizing and tie) and non- switchable lines. The node variables comprise both forecasted or estimated variables and measured variables. These variables include the estimated or forecasted values for active power demand (or generation), reactive power demand (or generation), and the three- phase voltage measured at each bus. The edge variable considered is the measured power flow through the branches. To obtain these measured signals, we utilize a power flow simulator in our synthetic approach. + +Network reconfiguration in the graph domain essentially involves determining the status (open or closed) of the switchable edges in the DN. Emergency load shedding at the primary DN level is indicated using a binary variable associated with the nodes representing switchable loads. + +### 2.3 Formulating a Markov Decision Process over Graphs + +The emergency response during outages in the DN is formulated as a Markov Decision Process (MDP) in the graph domain, denoted as \(\mathcal{M} = (\mathcal{S},\mathcal{A},\mathcal{P}_{tr},\mathcal{R})\) . The tuple denotes the state, action, transition probability, and reward (in the respective order), which are defined as follows: + +1. State \((\mathcal{S})\) : The state is composed of relevant observations from the DN that represent the current operating condition of the network. It includes node variables, edge variables, network topology, and other system variables, denoted as \(\mathcal{S} = [P_{d}^{N},Q_{d}^{N},P_{g}^{N},Q_{g}^{N},V^{N},V_{\mathrm{vol}},I^{E},\mathcal{T},E_{supp},\mathcal{O},\mu ]\) . Here, \(P_{d}^{N},Q_{d}^{N}\) represents the estimated or forecasted active and reactive power demand at the nodes, while \(P_{g}^{N},Q_{g}^{N}\) corresponds to the active and reactive power generation at the nodes. The three- phase voltage measured at the buses (graph nodes) is represented as \(V^{N}\) , and \(V_{\mathrm{vol}}\) indicates the voltage violation in the network. The edge variable includes the power flow through the network branches, denoted as \(I^{E}\) . The operating topology of the network is \(\mathcal{T}\) , and the total energy supplied in the network is represented by \(E_{supp}\) . The variable \(\mathcal{O}\) encapsulates the outage scenario, i.e., the multi-line failures in the network, including switch outages. To account for the outage of switches, we use a masking mechanism that suppresses the corresponding switching action, represented by the variable \(\mu\) . + +2. Action \((\mathcal{A})\) : The control actions for emergency response include switching and load shedding. Therefore, the action space is represented as \(\mathcal{A} = [\delta_{1}^{sw},\delta_{2}^{sw},\dots,\delta_{N^{s}}^{sw},\delta_{1}^{ld},\delta_{2}^{ld},\dots\delta_{N_{L}}^{ld}]\) . Here \(N\) represent the number of switchable lines, which includes both the sectionalizing and tie lines. The number of switchable loads in the network is denoted as \(N_{L}\) . Line switching is represented by a binary variable \(\delta^{sw}\) where 0 and 1 represent the opening and closing of the switch, respectively. The + +<--- Page Split ---> + +status of the loads is also represented by a binary variable \(\delta^{td}\) , where load served and load shed respectively corresponds to 1 and 0. + +3. Transition probability \((P_{tr})\) : The transition probability captures the dynamic nature of the network with emergency response, denoted as \(\mathcal{P}(s_{t + 1}^{\prime}|s_{t},a_{t})\) . This represents the transition from network state \(s\) at time step \(t\) to state \(s^{\prime}\) at step \(t + 1\) given that action \(a\) is implemented at time step \(t\) . The transition probability is learned by the agent from its interactions with the environment. + +4. Reward \((R)\) : The reward guides the RL algorithm to take optimal control actions for mitigating outages in the DN, which is formulated as follows: + +\[r(s,a) = \left\{ \begin{array}{ll}E_{supp} - V_{\mathrm{vol}}, & \mathrm{if} C_{viol} = 0,\\ 0, & \mathrm{otherwise}. \end{array} \right. \quad (1)\] + +The reward reflects the goal of improving resilience in the DN by maximizing the energy supplied \(E_{supp}\) while minimizing violations of network operational constraints. The reward takes into account both power flow convergence and its violation (i.e., \(C_{viol}\) ) to determine if an action is rewarded. This is to prevent the selection of any non- feasible or invalid control actions. Additionally, a penalty term is included to account for voltage violations \(V_{\mathrm{vol}}\) in the network. The voltage violations for each bus \(i \in \mathbf{N}\) beyond its upper limit \((\overline{V})\) and lower limit \((\underline{V})\) are evaluated after power flow estimation as follows: + +\[\Delta V_{max}^{i} = \left\{ \begin{array}{ll}\sum_{j\in \phi}V_{j}^{i} - \overline{V}, & \mathrm{if} V_{j}^{i} > \overline{V}\\ 0, & \mathrm{otherwise} \end{array} \right. \quad (2)\] + +\[\Delta V_{min}^{i} = \left\{ \begin{array}{ll}\sum_{j\in \phi}\underline{V} -V_{j}^{i}, & \mathrm{if} V_{j}^{i}< \underline{V}\\ 0, & \mathrm{otherwise}. \end{array} \right. \quad (3)\] + +where \(\phi\) denotes the set of phase connections for the bus. The voltage measurements and the energy supplied are estimated in per- units (pu) and calculated with respect to the base voltage, \(kV_{base}\) , and base power \(MVA_{base}\) of the corresponding network. The per- unit calculations in power systems eliminate the issue of units and is equivalent to normalizing them using their base values. : + +\[V_{\mathrm{vol}} = \frac{\sum_{i\in \mathbf{N}}(\Delta V_{max}^{i} + \Delta V_{min}^{i})}{3|\mathbf{N}|}, \quad (4)\] + +where \(|\mathbf{N}|\) is the cardinality of the set of network buses, and \(\Delta V_{max}\) and \(\Delta V_{min}\) represent the violations over maximum and minimum desirable voltage limits, respectively. + +### 2.4 Design of the Environment + +The distribution network models are implemented and simulated using the open- source distribution system simulator (OpenDSS) [40]. DERs are modeled using a generic + +<--- Page Split ---> + +generator and solar photovoltaic (PV) elements in OpenDSS. Switches are defined on lines with associated switching controls, while the disable/enable property of the loads is used for shedding or picking up load. OpenDSSDirect [41] is employed as the Python- based API to maneuver circuit modifications, I/O operations, and network topology extraction. The equivalent graph is constructed for the circuit using the NetworkX module. The overall framework of the environment is presented in Fig. 2. The implementation of specific switching actions may lead to the formation of multiple components within the network. These components are then translated into isolated DN sections within the DSS circuit. Furthermore, as discussed earlier, intentional islands created by grid- forming DERs are considered a potential solution to tackle outages. To enable power flow evaluation in the isolated DSS circuit section, a virtual slack or reference bus is defined at the location of the grid- forming DERs. This requires assigning a voltage source element to the selected buses (i.e., nodes). + +### 2.5 Learning Architecture + +The learning architecture utilizes a policy gradient- based RL algorithm, where the policy network is derived from a Graph Neural Network (GNN). Each node \(i\) in the DN graph has properties such as active/reactive power demand, generation, and three- phase voltage measurements, denoted as \(\gamma_{i} = [P_{d}^{i},Q_{d}^{i},P_{g}^{i},Q_{g}^{i},V^{i}]\) . The policy network takes the state information (detailed in Section 2.3) as input and produces an action. The policy network consists of three main components: 1) A GNN which is used to compute the graph node embeddings for the DN graph. 2) A feedforward network that is used to compute a feature vector, referred to as context embedding. This vector incorporates information that cannot be naturally represented in the graph structure, such as the energy supplied, voltage violations, and power flow through the edges. 3) A MLP that takes the node embeddings from the GNN and the context embeddings from the feedforward network as input. It computes a final feature vector that encompasses the entire state space information. Figure 2 shows the overall structure of the policy network, which includes the GNN- based feature abstraction. + +Initially, the node properties \(\gamma_{i},i\in N\) , are projected to a higher- dimensional space using linear transformation: \(F_{\mathrm{init}}^{i} = W_{\mathrm{init}}\times \gamma_{i} + b_{\mathrm{init}}\) , where \(W_{\mathrm{init}}\in \mathbb{R}^{|\gamma_{i}|\times h_{0}}\) and \(b_{\mathrm{init}}\) are learnable weights and biases, respectively. The cardinality of a vector or set is denoted by \(|.|\) , and \(h_{0}\) represents the projection length. Let \(F_{\mathrm{init}}\) be a matrix ( \(\in \mathbb{R}^{|N|\times h_{0}}\) ) that represents all \(F_{\mathrm{init}}^{i},i\in N\) , \((F_{\mathrm{init}} = [F_{\mathrm{init}}^{1},F_{\mathrm{init}}^{2},\dots F_{\mathrm{init}}^{|N|}]\) ) + +Node embeddings: Each feature vector \(F_{\mathrm{init}}^{i},i\in \mathbf{N}\) , is then passed through a series of Graph capsule layers. These layers utilize a graph convolutional filter of polynomial form to compute a matrix \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\) , defined as: + +\[f_{p}^{(l)}(\mathcal{X},\mathcal{L}) = \sum_{k = 0}^{K}\mathcal{L}^{k}(F_{(l - 1)}(\mathcal{X},\mathcal{L})^{\circ p})W_{pk}^{(l)}. \quad (5)\] + +Here, \(\mathcal{L}\) represents the graph Laplacian, \(p\) is the order of the statistical moment, \(K\) is the degree of the convolutional filter, \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\) denotes the output from layer \(l - 1\) , and \(F_{(l - 1)}(\mathcal{X},\mathcal{L})^{\circ p}\) represents \(p\) times element- wise multiplication of \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\) . + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: The learning framework developed which includes the environment and the policy network architecture with GNN-based feature abstraction. The environment is composed of a DN model interface (DSS circuit) using a python-based API and the graph replica of the DSS circuit. Voltage sources are introduced at virtual slacks for intentional islanding scenarios. The policy network uses the state information from the environment to compute graph node embeddings and context node embeddings using GNN and feedforward networks, respectively. A final feature vector that encompasses the two embeddings is computed by an MLP.
+ +Here, \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\in \mathbb{R}^{N_{n}\times h_{l - 1}p}\) \(W_{pk}^{(l)}\in \mathbb{R}^{h_{l - 1}p\times h_{l}}\) . The variable \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\in \mathbb{R}^{N_{n}\times h_{l}}\) is a matrix, where each row is an intermediate feature vector for each node \(i\in \mathbf{N}\) , infusing nodal information from \(L_{e}\times K\) hop neighbors, for a value of \(p\) . The output of layer \(l\) is obtained by concatenating all \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\) , as given by: + +\[F_{l}(\mathcal{X},\mathcal{L}) = [f_{1}^{(l)}(\mathcal{X},\mathcal{L}),f_{2}^{(l)}(\mathcal{X},\mathcal{L}),\dots ,f_{p}^{(l)}(\mathcal{X},\mathcal{L})]. \quad (6)\] + +<--- Page Split ---> + +Here, \(\mathcal{P}\) is the highest order of statistical moment, and \(h_{l}\) is the node embedding length of layer \(l\) . We consider all the values of \(h_{l}\) , \(l \in [0, L_{e}]\) , to be the same throughout the paper Equations (5) and (6) are computed for \(L_{e}\) layers, where each layer uses the output from the previous layer \((F_{l - 1}(\mathcal{X}, \mathcal{L}))\) . Increasing the number of layers \((L_{e})\) and raising the value of \(K\) can enhance the learning of the overall structure of the graph by aggregating nodal neighborhood features from \(L_{e} \times K\) neighbors. However, this improvement comes at the expense of having more learnable parameters in the policy, which becomes a drawback as the problem size increases. A larger value of \(h_{l}\) is beneficial as it enables the computation of a more detailed and comprehensive nodal state representation, both at the final stage and in intermediate steps. Similarly, a larger value of \(P\) assists in a better encoding of intermediate states using a vector representation (described in Eq. 6) for each intermediate feature. This richer structural embedding is expected to be more effective than the scalar embedding used in GCN (Graph Convolutional Networks). However, it is important to note that both higher \(h_{l}\) and \(P\) come with additional training costs. The final node embeddings are computed using a linear transformation of \(F_{l = L_{e}}(\mathcal{X}, \mathcal{L})\) . + +\[F_{\mathrm{Nodes}} = F_{l = L_{e}}(\mathcal{X}, \mathcal{L}).W_{F}, \quad (7)\] + +where \(W_{F}\) is a learnable weight matrix of size \(h_{L_{e}} \mathcal{P} \times h_{L_{e}}\) . + +The final graph embedding is computed by passing the node embeddings matrix \(F_{\mathrm{Nodes}}\) through a series of Linear layers, followed by taking the mean + +\[F_{\mathrm{graph}} = \mathrm{Mean}(W_{g2} \times (W_{g1} \times F_{\mathrm{Nodes}})), \quad (8)\] + +where \(W_{g1} \in \mathbb{R}^{h_{L_{e}} \times |N|}\) and \(W_{g2} \in \mathbb{R}^{h_{L_{e}} \times h_{L_{e}}}\) , and \(F_{\mathrm{graph}} \in \mathbb{R}^{h_{L_{e}}}\) , for ease of representation, the bias terms are omitted here. + +Context: In addition to the graph- based information, certain state space variables cannot be directly represented as nodes in the graph. These variables include energy supplied \(E_{supp}\) , voltage violation \(V_{\mathrm{viol}}\) , and power flow through the edges \(l^{E}\) . To incorporate this information, a feature vector called the context is constructed. + +\[F_{\mathrm{context}} = \mathrm{Feedforward}(\mathrm{Concat}([E_{supp}, V_{\mathrm{viol}}, l^{E}])) \quad (9)\] + +Final MLP layer: The final state embedding \(F_{\mathrm{final}}(\mathbb{R}^{h_{L_{e}}})\) is computed by adding \(F_{\mathrm{graph}}\) and \(F_{\mathrm{context}}\) and passing it through an MLP layer: + +\[F_{\mathrm{final}} = \mathrm{MLP}(F_{\mathrm{graph}} + F_{\mathrm{context}}). \quad (10)\] + +The Logits \(\in \mathbb{R}^{|\mathcal{A}|}\) across all available actions are computed by passing \(F_{\mathrm{final}}\) through a Feedforward layer. The Logits of the switches that need to be masked are set to negative infinity. Using the Logits, a Bernoulli probability distribution is computed for all available actions, with the probabilities computed using a Sigmoid function as \(e^{\mathrm{Logits}} / (1 + e^{\mathrm{Logits}})\) . The final switching action is determined using a greedy policy. If the mean of an action element (switch) is greater than 0.5, the switch position is set as on (or a value of 1). + +<--- Page Split ---> + +The predicted value of the state is computed by passing \(F_{\mathrm{final}}\) through another feedforward layer, which approximates the value of the state. + +### 2.6 Training Process + +The training process involves generating samples on the distribution network to simulate different outage scenarios. This is accomplished by introducing line failures, adjusting load and generation operating points, and considering various outage scenarios. The outage events in the network are primarily caused by distribution line failures, which are simulated using a graph- based approach (discussed in 4.1). The power network operating points (i.e., the load demand and power generation) are randomly drawn out of an annual profile made available in OpenDSS. To train the policy network, we employ Proximal Policy Optimization (PPO) [42]. The training process involves collecting experience in the form of tuples containing the state, action, reward, and the next state. PPO operates based on rollout operations, where each operation consists of a fixed number of steps, denoted as \(N_{\mathrm{steps}}\) . The weight updates occur after completing a rollout operation, in batches of size \(N_{\mathrm{batch}}(\leq N_{\mathrm{steps}})\) . The weight update is performed via backpropagation, aiming to minimize a cost function comprising the policy gradient loss and the state value approximation loss. The policy network was trained for a total of \(N_{\mathrm{total}}\) number of steps. To evaluate the performance of the proposed model, as well as to assess the impact of local and global structural information in the encoding process, we conducted comparative experiments with another learning- based framework called MLP. This framework utilizes the PPO algorithm, with a policy network based on a simple Multi- Layer Perceptron (MLP) architecture. To ensure a fair and unbiased comparison, MLP was trained using the same settings as GCAPS. Analyzing the convergence curve depicted in Fig. 3, it becomes evident that GCAPS consistently achieves a higher reward compared to MLP for both the 13- bus and 34- bus systems. This observation demonstrates the superior performance of GCAPS in effectively managing outages and optimizing the distribution network's operational state. + +### 2.7 Case Study on 13-bus Network: Scenario 1 + +The proposed model for outage management is validated using a modified version of the IEEE 13- bus distribution test network. This network incorporates switches and DERs and serves as the basis for validating the effectiveness of the proposed model, as shown in Fig. 4. Two grid- forming DERs of \(1000\mathrm{kW}\) are considered at buses '634' and '680', while the buses '645', '675', and '684' are equipped with grid- feeding DERs rated at \(40\mathrm{kW}\) , \(500\mathrm{kW}\) , and \(100\mathrm{kW}\) , respectively. The total connected load of the network is \(3.5\mathrm{MW}\) . In the normal configuration of the network, the sectionalizing switches are closed, while the tie switches remain open. This initial setup establishes the baseline operational state for the network. To systematically evaluate the developed model and its performance, two traditional optimization techniques, namely the mixed integer second- order conic programming (MISOCP) and binary particle swarm optimization (BPSO), are employed for all case studies in addition to the previously discussed MLP model. In the testing phase of the models, we rationally select the number and location + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: The training convergence plots for the policy models. Our GCAPS-based RL model is compared with the MLP model which does not utilize graph abstraction. (a) Convergence plot of GCAPS and MLP for the 13-bus network. (b) Convergence plot for the 34-bus network.
+ +of the line outages as opposed to the graph- based approach used during training. Additionally, the load and generating points are not drawn out of the representative annual profile discussed in training, rather a randomly generated multiplying factor is used to set the network operating point. + +Scenario 1 in 13 bus network involves the failure of a single line of importance, determined by its high edge- betweenness in normal configuration. Specifically, this scenario represents the outage of the line connecting buses '670'- '671'. The status of the decision variables, which includes both the switches and dispatchable loads, obtained from the different models for scenario 1 is depicted in Fig. 5. Notably, both the traditional optimization models, namely the MISOCP and the BPSO, yield the same solution for scenario 1. An important observation from analyzing the statuses of the switches and loads is that the reinforcement learning models demonstrate generalizability by providing distinct solutions for the two different scenarios. The voltage plot of the 13- bus network, after implementing the GCAPS solution for managing outage scenario 1, is illustrated in Fig. 6. Due to the switching actions in scenario 1, the resulting network configuration maintains a robust connection to the substation, ensuring that the voltages at all active phases of connected buses are within 0.95 and 1.10 pu, thus operating well within the desirable bounds. The voltage plot highlights the buses shaded in grey, which represent the buses disconnected from the main network due to the control actions. + +### 2.8 Case Study on 13-bus Network: Scenario 2 + +Scenario 2 involves the outage of three switchable lines connecting '632'- '670', '671'- '692', and '646'- '684'. This scenario aims to test the capability of the proposed model + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4: Test networks used to validate the proposed GCAPS model for real-time resilient control. The test networks are IEEE distribution networks modified by the addition of switches and DERs. Switches in the networks include sectionalizing and tie switches. Both types of DERs, i.e., grid-forming and grid-feeding, are considered. (a) Modified 13-bus IEEE test network. (b) Modified 34-bus IEEE test network.
+ +to enforce the inoperability of the outage switch in decision support. The status of decision variables, including the switches and dispatchable loads, obtained from the different models for scenario 2, is shown in Fig. 5. Once again, the MISOCP and the BPSO solutions for scenario 2 are identical. Upon inspecting the decision variables, it is noticeable that the MLP- based RL model violates the non- switchable condition of the outage line '671'- '692' (sw2) for scenario 2, as it mistakenly closes the switch. The voltage plot of the 13- bus network, after implementing the GCAPS solution for managing outage scenario 2, is shown in Fig. 6. In scenario 2, the GCAPS outage mitigation solution ensures a functional network with voltages at all active phases of connected buses ranging from 1.10 pu to 0.8 pu. Although there is a wider margin from the lower limit of 1 pu in scenario 2, it can be attributed to the outage of multiple switches (sectionalizing and tie) and the resulting formation of an island around bus '680' due to the control action. As a consequence of isolating the downstream section from the stronger substation bus, the phase voltages at the downstream buses slightly decrease. + +### 2.9 Case Study on 34-bus Network: Scenario 1 + +The validation of the proposed model and baselines is conducted on a modified 34- bus distribution test network, which incorporates switches and DERs. The total connected load of the network is 2.04 MW. Three grid- forming DERs with capacities of 146 kW, 144 kW, and 1000 kW are connected at buses '890', '844', and '816', respectively while a grid- feeding DER with a capacity of 96 kW is connected at bus '820'. Under normal operating conditions, the five sectionalizing switches are closed, while the four tie switches are open. Scenario 1 involves line outages at the connections between + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5: Status of decision variables acquired from the proposed model and baselines for the 13-bus test network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outage '670-671'. (b) Status of decision variables for test scenario 2 with line outages '632-670', '671-692', and '646-684'.
+ +![](images/Figure_6.jpg) + +
Fig. 6: Voltage plot of the 13-bus test network with the GCAPS outage management solution for varying test scenarios. The active phases of the network are only marked in the plot and the grey-shaded nodes represent those disconnected from the network. (a) Voltage plot for test scenario 1 with the GCAPS solution implemented in the network during outages. (b) Voltage plot for test scenario 2 with the GCAPS solution implemented in the network during outages.
+ +buses '812'- '814' and '824'- '848'. The line connecting '812'- '814' is upstream in the network and closer to the substation, indicating a severe impact on the network due to its outage. The line '824'- '848', on the other hand, is a switchable tie line (sw7). Fig. 7 presents the status of the decision variables, including switchable lines and loads, obtained from the different models for scenario 1. Both the MISOCP and BPSO yield similar results for scenario 1 on the 34-bus network. The results demonstrate the ability of RL models to differentiate between various scenarios and generalize during decision- making. However, the MLP- based RL model produces an invalid control action in scenario 1 by closing switch sw7 on the outage line. The switching action obtained from + +<--- Page Split ---> + +the GCAPS outage mitigation model results in the formation of an intentional island around DER at bus '816' for scenario 1. Consequently, the section downstream of '812'- '814' becomes a self- sustained network component disconnected from the substation. The voltage plot for the 34- bus network, derived by implementing the GCAPS solution during outages for scenario 1, is presented in Fig. 8. Scenario 1, being severe due to the outage of an upstream line, requires the GCAPS solution to intentionally isolate bus '816' to mitigate the outage. Consequently, the phase voltages at buses drop significantly, although the network remains functional. + +## 2.10 Case Study on 34-bus Network: Scenario 2 + +Scenario 2 considers multiple line failures at '832'- '858', '834'- '860', and '854'- '852' in the network. The lines '832'- '858' and '854'- '852' are in close proximity, while the line '834'- '860' is a switchable sectionalizing line (sw2). Figure 7 presents the status of the decision variables, including the switchable lines and loads, obtained from the different models for scenario 2. In scenario 2, the switching action by the GCAPS model results in a configuration that remains connected to the substation, with a small section disconnected (inactive) from the main network. The voltage plot for the 34- bus network, derived by implementing the GCAPS solution for scenario 2, is presented in Fig. 8. It is observed that the GCAPS solution for scenario 2 ensures voltages at all active phases of connected buses are well within the range of 0.95 to 1.06 pu. + +![](images/Figure_7.jpg) + +
Fig. 7: Status of decision variables acquired from the proposed model and baselines for the 34-bus network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outages '812-814' and '824-848'. (b) Status of decision variables for test scenario 2 with line outages '832-858', '834-860', and '854-852'.
+ +### 2.11 Comparison of the Proposed Model with Baselines + +We compare the developed GCAPS- based RL model with the baseline models to evaluate the performance speed and the estimated energy served during outage conditions. Figure 9 presents the estimated equivalent energy served when implementing the control decisions in the 13- bus distribution test network for scenarios 1 and 2 using the different models. As expected, the energy supplied is optimal for the MISOCP and + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Fig. 8: Voltage plot of the 34-bus test network for varying outage scenarios with the GCAPS outage management solution. The active phases of the network are only included in the plot. The grey-shaded nodes are disconnected from the network. (a) Outage scenario 1. (b) Outage scenario 2.
+ +BPSO models. Our GCAPS model shows near- optimal decision- making capability for both scenarios. In scenario 1, the MLP model is inferior as it provides the minimum energy supply among all the models, while it becomes invalid in scenario 2 due to the operation of the outage switch. The estimation of equivalent energy served obtained by implementing the control actions derived from the different models in the 34- bus distribution test network for scenarios 1 and 2 is shown in Fig. 9. Our GCAPS model exhibits near- optimal performance, closely approaching the optimal energy supply estimated by the MISOCP and BPSO models. On the other hand, the MLP model performs inferiorly compared to the other models and also produces an invalid control action for scenario 1. + +The performance of our GCAPS model is compared with the baselines by testing different scenarios in 13- bus and 34- bus networks. The computation time required to obtain the outage mitigation solution is presented in Table 1. It can be observed that + +<--- Page Split ---> + +the response time for the two RL- based models, namely GCAPS and MLP, is in the order of milliseconds, and is mostly agnostic to the increase in the size of the network from 13 to 34 bus system, demonstrating real- time performance. In comparison, the optimization- based methods, BPSO and MISOCP respectively require seconds or 10's of minutes to compute those decisions. Specifically, BPSO and MISOCP are respectively about 5 and 3 orders of magnitude more expensive than the learned RL- based policies. + +In Fig. 10, we illustrate the performance of the DN and its evolution with time when implementing the decisions provided by the different models during outages. Specifically, the proposed GCAPS- based RL model is compared with the MLP- based RL model which does not consider the underlying topology and the MISOCP method (conventionally used for solving such problems). The BPSO despite producing similar results as the MISOCP is not suitable for resilience decision support as is evident from the prohibitively delayed response shown in Table 1. Outage scenario 1 in the 13- bus network and outage scenario 2 in the 34- bus network are used to exemplify the impact of the model response on DN performance. The excluded scenarios in the two networks are not suitable for comparison owing to the invalid switching decisions provided by the MLP model as discussed earlier. As observed in Fig. 10, the voltages at the buses '652' and '890' in the 13 and 34 bus networks respectively are under voltage due to disruption. The voltage violation exists for about 100's and 1000's of cycles in the 13 and 34 bus DNs when MISOCP is used for decision support (as it takes 5- 20 sec to compute the decisions). While the RL models mitigate the voltage violation through outage management almost instantaneously. The continued operation of the network in the disrupted state also increases the risk of cascaded failures and widespread blackouts. Meanwhile, the loss of energy due to delayed decision- making by the MISOCP with respect to the GCAPS is 4,204.50 kWs and 13,522.18 kWs for 13 and 34 buses respectively. + +Table 1: Performance comparison of different models for scenarios in the test networks + +
NetworkMethodScenario 1
Mean time (s)		Scenario 2
Mean time (s)
13-busGCAPS0.00490.0056
MLP0.00390.0054
BPSO500.15540.20
MISOCP3.125.23
34-busGCAPS0.00300.0025
MLP0.00220.0020
BPSO2580.152540.20
MISOCP25.2020.20
+ +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Fig. 9: Comparison of the resilience improvement for different models with varying outage scenarios. The resilience improvement is measured in terms of equivalent energy served when implementing control actions during outages. (a) Testing on the 13-bus network, where an empty bar for MLP in scenario 2 indicates invalidity due to the operation of the outage switch. (b) Testing on the 34-bus network, with MLP model for scenario 1 invalid due to operation of outage switches.
+ +## 3 Discussion + +We have presented a real- time outage management model for distribution networks based on a reinforcement learning over graphs framework. In our outage management model, we have considered the grid- forming and feeding modes of the DER, and hence both grid- connected and islanding reconfiguration schemes have been incorporated into the solution. The load shedding adopted in the mitigation strategy ensures that the network has operational feasibility and is not vulnerable to voltage collapse. The learning model employs an on- policy RL algorithm and adopts the novel Graph Capsule (GCAPS) neural networks for integrating information about the DN topology into the learning framework. By leveraging GCAPS neural networks, the model has been shown to effectively integrate nodal properties, and local and global structural information into the learning process. + +We have evaluated our model on modified versions of the IEEE 13- bus and 34- bus distribution test networks, which include distributed energy resources (DERs) and sectionalizing/tie switches. Two traditional models based on MISOCP and BPSO, and the RL with MLP as policy network have been used as baselines to compare the real- time decision- making and network resilience improvement capability, where the energy served under disruption (see Fig. 9) can be perceived as a measure of resilience. The results have demonstrated that the proposed model achieves near- optimal performance in real- time outage management for different networks and outage scenarios. Additionally, the model has been found to effectively capture the DN topology in decision- making as indicated by the improved performance and constraint adherence when compared with the MLP- based approach. Above all, our model has also provided time- sensitive decision support for outage mitigation, thereby making it a suitable self- healing tool in the current "smart- grid" landscape. + +<--- Page Split ---> +![](images/Figure_10.jpg) + +
Fig. 10: The comparison of the proposed GCAPS model with baseline models. The baseline models include an MLP-based model which does not utilize graph abstraction and the (much slower to compute) conventional mathematical programming technique (here MISOCP). The network is in the disrupted state at time instant '0'. (a) Voltage measured at the bus '652' (most stressed bus) is used as an indicator for 13-bus network operational characteristics in outage scenario 1. (b) Equivalent power supplied in the network evolving with switching decisions for 13 bus in outage scenario 1. (c) Voltage measured at the bus '890' (most stressed bus) is used for comparing 34-bus network operational limits in outage scenario 2. (d) Equivalent power supplied in the 34-bus network by different models for scenario 2 and their response with time is illustrated.
+ +Currently, the proposed method have been applied to relatively small- medium sized networks, and applicability to larger networks remains to be explored. One challenge to accomplishing that is the associated training cost in terms of memory and computing time. From the results in our prior studies on applying related graph- based RL for Multi- Robot Task Allocation [38, 43], we have found that the computational memory requirement for training on larger graphs (more than 200 nodes) is very high and often hinders the training task. Our prior results [38, 43] have demonstrated the capability of the GNN- based policy network to learn policies that can be applied to a larger- sized mostly homogeneous networks with simple near- linear state transitions (without training), while still demonstrating comparable performance with respect to more traditional approaches. More work is required to explore if these advantages will + +<--- Page Split ---> + +also translate to applications such as the DN topology reconfiguration that involves heterogeneous networks and non- linear flow properties that affect the state transition. + +## 4 Methods + +## 4.1 Graph-based Scenario Generation + +The training scenarios used for RL over the graphs model is generated from the graph equivalent of the DN. The failure of the components, such as lines, can be approximated by disconnecting them from the DN [44]. The model developed is not specific to any particular type of extreme weather event, and hence a generalized and intuitive approach is adopted for simulating outages during training. The outages in the DN often originate from localized failures that can lead to cascading effects. To emulate this behavior, a subgraph method for randomized edge removal is employed, similar to the approach described in [45]. This method involves randomly selecting nodes \(N_{s} \in N\) from the graph representation of the DN, and creating subgraphs centered around these nodes with varying radii \(R_{s} \leq R_{max}\) (maximum radius). We consider \(R_{max} = \frac{G_{\mathrm{dia}}}{2}\) , where \(G_{\mathrm{dia}}\) is the diameter of the graph. Within each selected subgraph, a fraction of the edges \(F_{s} \in E\) is randomly removed to simulate the localized impact of contingencies. The fraction of edge failures is gradually increased from 0 to 50%. By varying \(N_{s}, R_{s}\) , and \(F_{s}\) , scenarios with multi- line failures can be generated for training the model. Furthermore, within each scenario, load multipliers and generating points are varied by randomly selecting multipliers from an annual profile available in OpenDSS package with an hourly resolution. + +### 4.2 Mixed Integer Programming Formulation + +Outage management in an unbalanced distribution network is an optimization problem that combines combinatorial and non- linear nature. The problem can be effectively formulated as an optimal power flow problem, leveraging branch flow equations with angle and conic relaxations as in [17]. The decision variables include switching and load shedding, while the control variables corresponding to power flow are also considered in the problem formulation. + +For the distribution network with \(\mathbb{L}\) set of loads and \(\tilde{\mathbb{L}}\) set of switchable loads, the active/reactive power consumption with load pickup or shedding is modeled using \(\delta^{L}\) as follows: + +\[\begin{array}{r l} & {P_{i}^{L} = \left\{ \begin{array}{l l}{\delta_{i}^{L}P_{i}^{D}} & {\mathrm{if~}i\in \widetilde{\mathbb{L}}}\\ {P_{i}^{D}} & {\mathrm{otherwise}} \end{array} \right.;\forall i\in \mathbb{L}}\\ & {Q_{i}^{L} = \left\{ \begin{array}{l l}{\delta_{i}^{L}Q_{i}^{D}} & {\mathrm{if~}i\in \widetilde{\mathbb{L}}}\\ {Q_{i}^{D}} & {\mathrm{otherwise}} \end{array} \right.;\forall i\in \mathbb{L}} \end{array} \quad (11b)\] + +where \(P_{i}^{D}\) and \(Q_{i}^{D}\) represent the active and reactive power demand of the load \(i\) , respectively. + +<--- Page Split ---> + +On the other hand, considering the set of grid- feeding generators \(\mathbb{G}_{fd}\) in the DN, the active and reactive power generation is estimated using: + +\[P_{(j,k)}^{G} = P_{avail}^{G} / |G_{ph}|~;\forall j\in \mathbb{G}_{fd},k\in \theta^{*} \quad (12a)\] + +\[Q_{(j,k)}^{G} = Q_{avail}^{G} / |G_{ph}|~;\forall j\in \mathbb{G}_{fd},k\in \theta^{*} \quad (12b)\] + +where \(P_{avail}^{G}\) , \(Q_{avail}^{G}\) is the total generation power available for the generator with \(|G_{ph}|\) number of phase connections, and \(\theta^{*}\) is the set of active phases of the generator, considering \(\theta = (a,b,c)\) . + +The total active and reactive power consumption by loads is constrained by the total generation in the DN as follows: + +\[\sum_{i\in \mathbb{L}}P_{i}^{L}\leq P_{tot}^{G}~,\sum_{i\in \mathbb{L}}Q_{i}^{L}\leq Q_{tot}^{G} \quad (13)\] + +The total power generation in the DN is given as follows: + +\[P_{tot}^{G} = \sum_{k\in \theta}\big(\sum_{j\in \mathbb{G}_{fd}}P_{(j,k)}^{G} + \sum_{h\in \mathbb{G}_{s}}P_{(h,k)}^{G}\big) \quad (14a)\] + +\[Q_{tot}^{G} = \sum_{k\in \theta}\big(\sum_{j\in \mathbb{G}_{fd}}Q_{(j,k)}^{G} + \sum_{h\in \mathbb{G}_{s}}Q_{(h,k)}^{G}\big) \quad (14b)\] + +where in addition to the grid- feeding generators, the set of grid- forming generators \(\mathbb{G}_{s}\) , including the substation, are considered. + +The power supplied by the grid- forming generators and the substation is constrained to be within its maximum capacity as follows: + +\[\sum_{k\in \theta^{*}}P_{(h,k)}^{G}\leq \overline{{P_{h}^{G}}};\forall h\in \mathbb{G}_{s}. \quad (15)\] + +Adopting three- phase branch flow formulations with relaxations as in [17], \(\nu\) and \(\mathcal{I}\) are used to denote the square of voltage and current, respectively. The voltages at all buses except the slack buses are constrained within upper and lower limits as follows: + +\[\mathcal{V}\leq \mathcal{V}_{r,k}\leq \overline{\mathcal{V}};\forall r\in \mathbb{B}\setminus \mathbb{B}_{s},k\in \theta \quad (16)\] + +Here, \(\mathbb{B}\) and \(\mathbb{B}_{s}\) denote the set of buses and the set of slack buses in the network, respectively. The voltage square at the substation (or slack) bus on the other hand is equated to 1.04 per unit. + +For the set of power delivery elements \(\mathbb{E}\) , the set of switchable elements (lines) \(\mathbb{E}_{sw}\) , and the line switch status \(\delta^{sw}\) , the power flow \(P^{E}\) through the elements are constrained as follows: + +\[\underline{{P}}_{(b,k)}^{E}\leq \underline{{P}}_{(b,k)}^{E}\leq \overline{{P}}_{(b,k)}^{E};\forall b\in \mathbb{E}\setminus \mathbb{E}_{sw},k\in \theta \quad (17a)\] + +<--- Page Split ---> + +\[\underline{{P}}_{(l,k)}^{E}\delta_{l}^{s w}\leq P_{(l,k)}^{E}\leq \overline{{P}}_{(l,k)}^{E}\delta_{l}^{s w};\forall l\in \mathbb{E}_{s w},k\in \theta \quad (17b)\] + +In a similar manner, the reactive power flow \(Q^{E}\) and the square of branch current square \(\mathcal{I}^{E}\) through the elements are also constrained within its limits. The power flow through the outage lines defined in set \(\mathbb{O}\) is, however, equated to zero as shown below: + +\[P_{(b,k)}^{E} = 0; \forall b\in \mathbb{O},k\in \theta \quad (18)\] + +The reactive power flow and the square of branch current through outage lines are also equated to zero. The balance of active and reactive power flow through the elements is formulated as follows: + +\[\begin{array}{r l r}{{P_{(b,k)}^{E}=\sum_{q\in\mathbb{R}_{L}(b)}P_{(q,k)}^{L}-\sum_{w\in\mathbb{R}_{G}(b)}P_{(w,k)}^{G}+\sum_{h\in\mathbb{C}(b)}P_{(h,k)}^{E}}}\\ &{}&{+R_{(b,k)}\mathcal{I}_{(b,k)};\forall b\in\mathbb{E},k\in\theta}\end{array} \quad (19)\] + +\[Q_{(b,k)}^{E} = \sum_{q\in \mathbb{R}_{L}(b)}Q_{(q,k)}^{L} - \sum_{w\in \mathbb{R}_{G}(b)}Q_{(w,k)}^{G} + \sum_{h\in \mathbb{C}(b)}Q_{(h,k)}^{E} \quad (20)\] + +where \(\mathbb{R}_{L}(b)\) is the set of loads and \(\mathbb{R}_{G}(b)\) is the set of generators connected to the receiving bus of element \(b\) . In (19) and (20), \(\mathbb{C}(b)\) represent the elements that are children elements to \(b\) . + +Additionally, Kirchhoff's voltage equation is modeled as: + +\[\begin{array}{r l r} & {} & {\mathcal{V}_{(\mathbb{R}(b),k)} = \mathcal{V}_{(\mathbb{S}(b),k)} - 2(\hat{R}_{(b,k)}P_{(b,k)}^{E} + \hat{X}_{(b,k)}Q_{(b,k)}^{E})}\\ & {} & {+\hat{Z}_{(b,k)}\mathcal{I}_{(b,k)});\forall b\in \mathbb{E}\setminus \mathbb{E}_{s w},k\in \theta} \end{array} \quad (21)\] + +Here, the parameters \(\hat{R}\) and \(\hat{X}\) denote the element's modified resistance and reactance, respectively. In (21), \(\mathbb{S}(b)\) and \(\mathbb{R}(b)\) denote the sending and receiving bus of the element \(b\) , respectively. For elements (lines) with switch, Eq. (21) is modified to an inequality constraint using the big M method [17] and bound within \(- (1 - \delta_{l}^{s w})M\) and \((1 - \delta_{l}^{s w})M\) . + +The second- order conic inequality constraint using convex relaxation is formulated as follows: + +\[\mathcal{I}_{(b,k)}*\mathcal{V}_{(\mathbb{S}(b),k)}\geq [(P_{(b,k)}^{E})^{2} + (Q_{(b,k)}^{E})^{2}];\forall b\in \mathbb{E},k\in \theta \quad (22)\] + +The radiality of the network structure is preserved using the spanning tree constraints as follows: + +\[\begin{array}{c}{\beta_{l}^{1} + \beta_{l}^{2} = \delta_{l}^{s w};\forall l\in \mathbb{E}_{s w}}\\ {\beta_{b}^{1} + \beta_{b}^{2} = 1;\forall b\in \mathbb{E}\setminus \mathbb{E}_{s w}}\\ {\beta_{b}^{1} + \beta_{b}^{2} = 0;\forall b\in \mathbb{O}}\\ {\beta_{b(s)}^{1} = 0;\forall s\in \mathbb{B}_{s}} \end{array} \quad (23)\] + +<--- Page Split ---> + +\[\sum_{x\in \mathbb{P}(y)}\beta_{(x,y)}^{1} + \sum_{z\in \mathbb{S}(y)}\beta_{(y,z)}^{2}\leq 1;\forall y\in \mathbb{B} \quad (25)\] + +In the above equations, the two edge directions (flows) possible for an element are represented as \(\beta^{1}\) and \(\beta^{2}\) . In Eq. 23, the flow through a branch is constrained to one direction based on the conditionalities: without/with switches, and outage of the element. The branches connected to the source bus indicated by \(b(s)\) are limited to one direction of flow as in Eq. 24. Additionally, each bus is ensured to have at most one parent as in Eq. 25, where \(\mathbb{P}(y)\) and \(\mathbb{S}(y)\) represent the set of preceding and succeeding buses in the neighborhood of bus \(y\) , respectively. + +The objective function maximizes the total power supply in the network with control actions during outages and is formulated as follows: + +\[min.\sum_{i\in \mathbb{L}}P_{i}^{L} \quad (26)\] + +### 4.3 Training Details + +The training process was allocated a maximum of 36 hours, and the total number of steps is set to 2 million. For the 13- bus systems, both GCAPS and MLP successfully completed the training with 2 million steps. However, for the 34- bus systems, MLP could only be trained for 1.5 million steps within the 36- hour time frame. To ensure a fair comparison, we utilize the trained weights of GCAPS and MLP at 1.5 million steps for the 34- bus system. Table 2 shows the training details, including the hyperparameter setting for PPO. + +Table 2: Training details + +
ParameterValue
AlgorithmPPO
Total steps (Ntotal)2.0e6 (13 bus), 1.5e6 (34 bus)
Rollout buffer size (Nsteps)5e4
Batch size (Nbatch)2e4
OptimizerAdam
Learning step size1e-6
Value coefficient0.5
Discount factor0.0
Epochs100
+ +### 4.4 Simulation Setup + +The proposed model and all the other baselines are tested on a system with Intel Core i7- 8565U 1.80GHz with 16 GB memory. The OpenDSSDirect API along with Python version 3.9.12, and Networkx version 2.8.4 are used in our simulations. The mixed integer programming is performed with Gurobiy using a Gurobi optimizer version 9.5.2. + +<--- Page Split ---> + +Acknowledgments. This material is based upon work sponsored by the Department of the Navy, Office of Naval Research under ONR award number N00014- 21- 1- 2530. The United States Government has a royalty- free license throughout the world in all copyrightable material contained herein. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Office of Naval Research. + +Data availability. The data that support the plots within this article and other findings of this study are available from the corresponding author upon reasonable request. + +Code availability. Code for this article is available from the corresponding author upon reasonable request. + +## References + +[1] Campbell, R. J. & Lowry, S. Weather- related power outages and electric system resiliency (Congressional Research Service, Library of Congress Washington, DC, 2012). + +[2] Kirthiga, M. V., Daniel, S. A. & Gurunathan, S. A methodology for transforming an existing distribution network into a sustainable autonomous micro- grid. IEEE Transactions on Sustainable Energy 4, 31- 41 (2012). + +[3] Bouhouras, A. S., Andreou, G. T., Labridis, D. P. & Bakirtzis, A. G. Selective automation upgrade in distribution networks towards a smarter grid. IEEE Transactions on Smart Grid 1, 278- 285 (2010). + +[4] U.S. Department of Energy. 2020 Smart Grid System Report (2022). + +[5] Arefifar, S. A., Alam, M. S. & Hamadi, A. A review on self- healing in modern power distribution systems. Journal of Modern Power Systems and Clean Energy (2023). + +[6] Wang, Y. et al. Coordinating multiple sources for service restoration to enhance resilience of distribution systems. IEEE Transactions on Smart Grid 10, 5781- 5793 (2019). + +[7] Fan, D. et al. Restoration of smart grids: Current status, challenges, and opportunities. Renewable and Sustainable Energy Reviews 143, 110909 (2021). + +[8] Baran, M. E. & Wu, F. F. Network reconfiguration in distribution systems for loss reduction and load balancing. IEEE Power Engineering Review 9, 101- 102 (1989). + +[9] Jacob, R. A. & Zhang, J. Distribution network reconfiguration to increase photovoltaic hosting capacity, 1- 5 (IEEE, 2020). + +<--- Page Split ---> + +[10] Jacob, R. A. & Zhang, J. Outage management in active distribution network with distributed energy resources, 1–6 (IEEE, 2021). + +[11] Al Owaifeer, M. & Al- Muhaini, M. MILP- based technique for smart self- healing grids. IET Generation, Transmission & Distribution 12, 2307–2316 (2018). + +[12] Botea, A., Rintanen, J. & Banerjee, D. Optimal reconfiguration for supply restoration with informed A\* search. IEEE Transactions on Smart Grid 3, 583–593 (2012). + +[13] Xu, Y., Liu, C.- C., Schneider, K. P., Tuffner, F. K. & Ton, D. T. Microgrids for service restoration to critical load in a resilient distribution system. IEEE Transactions on Smart Grid 9, 426–437 (2016). + +[14] Poudel, S., Dubey, A. & Schneider, K. P. A generalized framework for service restoration in a resilient power distribution system. IEEE Systems Journal 16, 252–263 (2020). + +[15] Bakar, N. N. A., Hassan, M. Y., Sulaima, M. F., Na'im Mohd Nasir, M. & Khamis, A. Microgrid and load shedding scheme during islanded mode: A review. Renewable and Sustainable Energy Reviews 71, 161–169 (2017). + +[16] Liu, H., Chen, X., Yu, K. & Hou, Y. The control and analysis of self- healing urban power grid. IEEE Transactions on smart grid 3, 1119–1129 (2012). + +[17] Farivar, M. & Low, S. H. Branch flow model: Relaxations and convexification—part i. IEEE Transactions on Power Systems 28, 2554–2564 (2013). + +[18] Sekhavatmanesh, H. & Cherkaoui, R. A novel decomposition solution approach for the restoration problem in distribution networks. IEEE Transactions on Power Systems 35, 3810–3824 (2020). + +[19] Shirmohammadi, D. Service restoration in distribution networks via network reconfiguration. IEEE Transactions on Power Delivery 7, 952–958 (1992). + +[20] Zidan, A. & El- Saadany, E. Network reconfiguration in balanced and unbalanced distribution systems with variable load demand for loss reduction and service restoration, 1–8 (IEEE, 2012). + +[21] Rao, R. S., Narasimham, S. V. L., Raju, M. R. & Rao, A. S. Optimal network reconfiguration of large- scale distribution system using harmony search algorithm. IEEE Transactions on power systems 26, 1080–1088 (2010). + +[22] Wu, Y.- K., Lee, C.- Y., Liu, L.- C. & Tsai, S.- H. Study of reconfiguration for the distribution system with distributed generators. IEEE transactions on Power Delivery 25, 1678–1685 (2010). + +<--- Page Split ---> + +[23] Pathan, M. I., Al- Muhaini, M. & Djokic, S. Z. Optimal reconfiguration and supply restoration of distribution networks with hybrid microgrids. Electric Power Systems Research 187, 106458 (2020). + +[24] Sekhavatmanesh, H. & Cherkaoui, R. Analytical approach for active distribution network restoration including optimal voltage regulation. IEEE Transactions on Power Systems 34, 1716- 1728 (2018). + +[25] de Quevedo, P. M., Contreras, J., Rider, M. J. & Allahdadian, J. Contingency assessment and network reconfiguration in distribution grids including wind power and energy storage. IEEE Transactions on Sustainable Energy 6, 1524- 1533 (2015). + +[26] Li, Y., Xiao, J., Chen, C., Tan, Y. & Cao, Y. Service restoration model with mixed- integer second- order cone programming for distribution network with distributed generations. IEEE Transactions on Smart Grid 10, 4138- 4150 (2018). + +[27] Chen, C., Wang, J., Qiu, F. & Zhao, D. Resilient distribution system by microgrids formation after natural disasters. IEEE Transactions on smart grid 7, 958- 966 (2015). + +[28] Wang, F. et al. A multi- stage restoration method for medium- voltage distribution system with dgs. IEEE transactions on smart grid 8, 2627- 2636 (2016). + +[29] Sultana, B., Mustafa, M., Sultana, U. & Bhatti, A. R. Review on reliability improvement and power loss reduction in distribution system via network reconfiguration. Renewable and sustainable energy reviews 66, 297- 310 (2016). + +[30] Cao, D. et al. Reinforcement learning and its applications in modern power and energy systems: A review. Journal of modern power systems and clean energy 8, 1029- 1042 (2020). + +[31] Gao, Y., Shi, J., Wang, W. & Yu, N. Dynamic distribution network reconfiguration using reinforcement learning, 1- 7 (IEEE, 2019). + +[32] Kundacina, O. B., Vidovic, P. M. & Petkovic, M. R. Solving dynamic distribution network reconfiguration using deep reinforcement learning. Electrical Engineering 1- 15 (2021). + +[33] Wang, B., Zhu, H., Xu, H., Bao, Y. & Di, H. Distribution network reconfiguration based on noisynet deep q- learning network. IEEE Access 9, 90358- 90365 (2021). + +[34] Gao, Y., Wang, W., Shi, J. & Yu, N. Batch- constrained reinforcement learning for dynamic distribution network reconfiguration. IEEE Transactions on Smart Grid 11, 5357- 5369 (2020). + +<--- Page Split ---> + +[35] Abdelmalak, M. et al. Network reconfiguration for enhanced operational resilience using reinforcement learning, 1–6 (IEEE, 2022). + +[36] Gautam, M., Abdelmalak, M., MansourLakouraj, M., Benidris, M. & Livani, H. Reconfiguration of distribution networks for resilience enhancement: A deep reinforcement learning- based approach, 1–6 (IEEE, 2022). + +[37] Verma, S. & Zhang, Z. L. Graph capsule convolutional neural networks (2018). 1805.08090. + +[38] Paul, S., Ghassemi, P. & Chowdhury, S. Learning scalable policies over graphs for multi- robot task allocation using capsule attention networks, 8815–8822 (IEEE, 2022). + +[39] Vinayagam, A., Swarna, K., Khoo, S. Y., Maung Than Oo, A. & Stojcevski, A. Pv based microgrid with grid- support grid- forming inverter control- (simulation and analysis) (2017). + +[40] Dugan, R. C. & McDermott, T. Reference guide. The Open Distribution System Simulator (OpenDSS). EPRI (2016). + +[41] Krishnamurthy, D. Opendssdirect. py. Tech. Rep., National Renewable Energy Lab.(NREL), Golden, CO (United States) (2017). + +[42] Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms (2017). 1707.06347. + +[43] Paul, S. et al. Efficient planning of multi- robot collective transport using graph reinforcement learning with higher order topological abstraction. arXiv preprint arXiv:2303.08933 (2023). + +[44] Danielsson, A. M. Deep learning for power system restoration. Ph.D. thesis (2018). + +[45] Bush, B., Chen, Y., Ofori- Boateng, D. & Gel, Y. R. Topological machine learning methods for power system responses to contingencies, Vol. 35, 15278–15285 (2021). + +<--- Page Split ---> diff --git a/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05_det.mmd b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..dfb82c2bd228a26d80af9cca87b71f5ed094ac5a --- /dev/null +++ b/preprint/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05/preprint__13a7ef15686e7c3c477325bdaa50c16d051ad5af065da60d8c02216a320d5f05_det.mmd @@ -0,0 +1,687 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 800, 210]]<|/det|> +# Real-Time Outage Management in Active Distribution Networks Using Reinforcement Learning over Graphs + +<|ref|>text<|/ref|><|det|>[[44, 230, 138, 249]]<|/det|> +Jie Zhang + +<|ref|>text<|/ref|><|det|>[[54, 258, 290, 275]]<|/det|> +jiezhang@utdallas.edu + +<|ref|>text<|/ref|><|det|>[[50, 303, 700, 323]]<|/det|> +The University of Texas at Dallas https://orcid.org/0000- 0001- 7478- 5670 + +<|ref|>text<|/ref|><|det|>[[44, 328, 343, 368]]<|/det|> +Roshni Jacob The University of Texas at Dallas + +<|ref|>text<|/ref|><|det|>[[44, 374, 592, 415]]<|/det|> +Steve Paul University at Buffalo https://orcid.org/0000- 0002- 8138- 5242 + +<|ref|>text<|/ref|><|det|>[[44, 420, 234, 460]]<|/det|> +Souma Chowdhury University at Buffalo + +<|ref|>text<|/ref|><|det|>[[44, 466, 343, 507]]<|/det|> +Yulia Gel The University of Texas at Dallas + +<|ref|>sub_title<|/ref|><|det|>[[44, 550, 103, 567]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[42, 588, 955, 630]]<|/det|> +Keywords: Network reconfiguration, Graph neural networks, Reinforcement learning, Topology, Resilience, Outage management + +<|ref|>text<|/ref|><|det|>[[44, 648, 343, 667]]<|/det|> +Posted Date: September 6th, 2023 + +<|ref|>text<|/ref|><|det|>[[42, 686, 475, 704]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3276125/v1 + +<|ref|>text<|/ref|><|det|>[[42, 723, 914, 765]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 784, 535, 803]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 839, 905, 881]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on June 4th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49207- y. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[225, 155, 732, 234]]<|/det|> +# Real-Time Outage Management in Active Distribution Networks Using Reinforcement Learning over Graphs + +<|ref|>text<|/ref|><|det|>[[230, 254, 722, 290]]<|/det|> +Roshni Anna Jacob \(^{1}\) , Steve Paul \(^{2}\) , Souma Chowdhury \(^{2}\) , Yulia R. Gel \(^{3}\) , Jie Zhang \(^{1,4*}\) + +<|ref|>text<|/ref|><|det|>[[189, 298, 767, 425]]<|/det|> +\(^{1}\) Department of Electrical and Computer Engineering, The University of Texas at Dallas, Richardson, 75080, Texas, USA. \(^{2}\) Department of Mechanical and Aerospace Engineering, University at Buffalo, Buffalo, 14260, New York, USA. \(^{3}\) Department of Mathematical Sciences, The University of Texas at Dallas, Richardson, 75080, Texas, USA. \(^{4}\) Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, 75080, Texas, USA. + +<|ref|>sub_title<|/ref|><|det|>[[444, 482, 512, 494]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[206, 498, 751, 721]]<|/det|> +Self- healing "smart grids" are characterized by fast- acting, intelligent control mechanisms that minimize power disruptions during outages. The corrective actions adopted during outages in power distribution networks include reconfiguration through switching control and emergency load shedding. The conventional decision- making models for outage mitigation are, however, not suitable for "smart grids" due to their slow response and computational inefficiency. Here, we present a new reinforcement learning (RL) model for outage management in the distribution network to enhance its resilience. The distinctive characteristic of our approach is that it explicitly accounts for the underlying network topology and its variations with switching control, while also capturing the complex interdependencies between state variables (along nodes and edges) by modeling the task as a graph learning problem. Our model learns the optimal control policy for power restoration using a Capsule- based graph neural network. We validate our model on two test networks, namely the 13 and 34- bus modified IEEE networks where it is shown to achieve near- optimal, real- time performance with up to 5 orders of magnitude improvement in computational speed. The resilience improvement of our model in terms of loss of energy is 4.204 MWs and 13.522 MWs for 13 and 34 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[245, 88, 790, 115]]<|/det|> +buses, respectively. Our model also demonstrates generalizability across a broad range of outage scenarios. + +<|ref|>text<|/ref|><|det|>[[245, 125, 786, 149]]<|/det|> +Keywords: Network reconfiguration, Graph neural networks, Reinforcement learning, Topology, Resilience, Outage management. + +<|ref|>sub_title<|/ref|><|det|>[[208, 193, 384, 211]]<|/det|> +## 1 Introduction + +<|ref|>text<|/ref|><|det|>[[207, 220, 831, 365]]<|/det|> +Resilience enhancement of power distribution networks (DNs) has been gaining considerable recognition in recent years, which has been often overlooked before due to the perception of DNs as merely a link between the transmission networks and consumers. A key factor for this shift is the realization that \(90\%\) of customer disruptions during extreme events can be attributed to the failure of components within the distribution network itself [1]. Additionally, the increasing presence of distributed energy resources (DERs) and the resulting decentralization of power generation have spurred the notion of DNs as autonomous entities that can operate independently from the main grid [2]. Consequently, the DN is now considered capable of retaining its functionality even during the loss of connectivity to the transmission network. + +<|ref|>text<|/ref|><|det|>[[207, 365, 831, 507]]<|/det|> +Concurrently, modernization of the power grid and the shift towards "smart grids" have been driving the deployment of intelligent and automated technologies in the DN [3]. The distribution automation has been implemented through the deployment of line monitors, fault indicators, remote- controlled switches, and reclosers in the DN [4]. An important characteristic of the "smart grid" is its self- healing capability, which includes implementing intelligent control actions through automation to minimize power disruptions, thus enabling the recovery of network operations during outages in real time [5]. Therefore, the key requirements of a self- healing tool include autonomy, quick response, and online adaptability, which are indeed the salient features of our model discussed in this paper. + +<|ref|>text<|/ref|><|det|>[[207, 507, 831, 735]]<|/det|> +In the face of power disruptions caused by extreme weather events or cyber- physical attacks, a self- healing DN warrants the automatic detection of faulty components, their isolation, and system restoration (fully or partially) using intelligent control algorithms. This process is referred to as FLISR, which stands for fault location, isolation, and service restoration [5], and is addressed using task- specific techniques. Restoration or the recovery of DN operation can be achieved using different control actions, such as network reconfiguration, load management, DER control, energy storage control, and reactive power resource control. The preliminary control action often adopted in such circumstances is reconfiguration (or switching control), followed by load shedding [6, 7]. Distribution network reconfiguration (DNR) by controlling the status of the network switches is a commonly used strategy to control DN operation for varying objectives such as loss minimization, reliability enhancement, load balancing, increasing penetration of renewable resources, improvement of voltage profile, and service restoration [8- 10]. The purpose of feeder reconfiguration is two- fold: (i) to quickly and efficiently reroute power from the functional part of the DN to the isolated section [11, 12], and (ii) to form intentional islands around the grid forming + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 243]]<|/det|> +DERs when there exists no connectivity to the main grid [2, 13]. In the existing body of knowledge, these two reconfiguration strategies have been addressed separately and have been largely considered as two distinct domains. However, a comprehensive restoration strategy suitable for various outage scenarios must efficiently utilize both grid- forming and grid- feeding DERs and consider all possible reconfiguration (or switching) options [14]. Hence, in our framework, we consider both DN characterizations through switching by including simultaneously grid- connected and off- grid modes of operation. Additionally, DNR alone may not be sufficient as a restorative action during catastrophic events, as the network remains vulnerable to voltage collapse and system blackouts [7, 15]. Therefore, load shedding becomes necessary as an emergency control mechanism [16] to minimize voltage violations in the DN. + +<|ref|>text<|/ref|><|det|>[[165, 244, 790, 684]]<|/det|> +Furthermore, power distribution networks are typically unbalanced and radial in nature, with a unidirectional power flow from the substation to the consumers. Besides the non- linearity in power flow, the optimization of modern- day DN operation has also been made challenging by the integration of DERs [17]. The DN restoration is an NP- hard, non- linear combinatorial optimization problem that aims to maximize energy supply while considering network connectivity and operational constraints [18]. Various methods have been used in the literature to solve the traditional reconfiguration problem, falling into heuristic [19, 20], meta- heuristic [21- 23], and mixed- integer programming [24- 26] techniques. In line with the increasing penetration of DERs, researchers have also explored islanding strategies using mixed- integer programming models to expand the zone of DER operation [2, 27]. Load management during outages has also been previously investigated as an emergency control strategy [14, 28]. Despite these efforts, a solution incorporating both the grid- connected and islanding (off- grid) reconfiguration schemes for outage management is limited in literature, and presents a complex and challenging problem to solve. The multitude of restorative options depends on the number of controllable devices (switches, loads) and the operational modes of DERs. Although the proliferation of remote- controlled elements in the DNs widens the horizons of automated network control, it also increases the complexity of the underlying non- linear combinatorial optimization problem [29]. The commonly used mixed- integer non- linear programming (MINLP) methodologies for restoration problems face issues of scalability, computational tractability, and real- time decision- making capability [14]. Apart from these, the existing linear programming approximation models in the literature are not designed to address restoration in three- phase unbalanced DNs with sectionalizing, tie switches, and various types of DERs (grid- forming and grid- feeding). Heuristic and meta- heuristic techniques, although explored, tend to be computationally expensive and time- consuming. Moreover, traditional methods heavily rely on a comprehensive description of the DN model and network parameters, making them model- dependent. Considering the uncertainty in network conditions during outages, it is desirable to develop a model capable of adapting to varying circumstances and is deployable online. Here, we present a model based on reinforcement learning to provide online decision support during outages. + +<|ref|>text<|/ref|><|det|>[[165, 685, 790, 727]]<|/det|> +Reinforcement learning (RL) methods have been increasingly adopted in recent years for power system applications that require autonomous control [30]. This is because RL methods are quite effective in solving high- dimensional, combinatorial, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[206, 87, 831, 472]]<|/det|> +stochastic optimization problems, besides providing fast- acting control. The latter is imperative to rapid responsiveness during outages, otherwise not possible with conventional optimization based decision support. Particularly with regards to reconfiguration, RL- based models [31, 32] have been developed to perform dynamic DNR during normal operation for loss minimization and voltage improvement. These methods specifically used deep Q- learning with neural networks and trained the off- policy RL network using a historical network operation dataset. The exploration problem which may arise in these models has been addressed by a Noisy- Net Q- learning model [33] developed to perform DNR for similar objectives. Another approach [34], utilized a batch- constrained soft actor- critic algorithm to learn the control policy for loss minimization during normal DN operation. As opposed to the DNR during normal operation considered in these studies, extreme operating conditions are more challenging considering the high- impact, low- probability occurrence of such events. Therefore, availing historical datasets for network operation may also not be possible as in previous studies. Although researchers have explored using RL models for DNR [35, 36] to improve network resilience, such works do not consider the feasibility of network operation based on voltage monitoring and DER operational modes during reconfiguration. Additionally, in methods based on the Q- learning approach, the policy network determines the optimal/near- optimal configuration or the spanning forest, rather than individually controlling each switch. This approach would require enumerating all feasible configurations to define a Q- probability matrix, which is impractical due to the exponential increase in state and action space with network size, possible outage scenarios, and the number of devices. Since these methods are not scalable and require significant storage and computational capabilities for exploration, policy gradient methods are more suitable for learning in outage conditions [34]. We, therefore, employ the proximal policy optimization (PPO), which is a policy gradient method for learning DN outage management in DN. + +<|ref|>text<|/ref|><|det|>[[207, 472, 831, 728]]<|/det|> +In this model, our idea is based on the intrinsic graph representation of power distribution networks. The DN is viewed as a graph where nodes are the buses (i.e., substation, load, or DERs) and edges are the lines or transformers. The state variables of the DN, including demand/generation estimates and voltage/current measurements, can be considered as data superimposed on a graph. The state variables exhibit complex interdependencies, necessitating the extraction of meaningful representations that accurately capture the structure of the DN connectivity. Moreover, outage management, particularly reconfiguration, involves altering the DN connectivity by switching on/off network lines (binary actions) and hence, requires consideration of the underlying combinatorial network structure. Therefore, we present a new graph RL approach for simultaneous real- time control of network topology and loads, ensuring sustained network operations during failures that are caused by extreme events. Graph RL uses a graph neural network or GNN as a policy model (as is the case here) and/or "value" model, as it allows more effective capturing of the combinatorial nature of network- based state information (involving both binary and continuous variables). This advantage is demonstrated in our case studies through comparison with baseline RL- based solutions that use a standard multi- layered perceptron (MLP) based policy model. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 271]]<|/det|> +Specifically, we use a Graph Capsule (GCAPS) neural network to learn optimal control policies which is a novel application in power network resilience problems. Compared to other GNNs such as Graph Convolutional Networks (GCN), the capsule- based GNN has been shown by [37] and [38] to better capture the structural information of a graph (the DN in this work) as a graph embedding, where the individual intermediate features of the state are represented as a vector (in GCAPS) as compared to that of a scalar for example in GCN and Graph Attention Networks (GAT), thus giving an enhanced state representation. This enhanced state representation helps in computing better actions compared to other simple feature abstraction networks such as Multi- Layered Perceptron (MLP). Experimental validation of our trained GCAPS based model on test networks demonstrates the generalizability and real- time control capability with near- optimal performance which is desirable in a self- healing tool for DNs. + +<|ref|>sub_title<|/ref|><|det|>[[165, 286, 281, 305]]<|/det|> +## 2 Results + +<|ref|>sub_title<|/ref|><|det|>[[165, 316, 787, 333]]<|/det|> +### 2.1 Reconfiguration & Load Shedding as Emergency Response + +<|ref|>text<|/ref|><|det|>[[165, 339, 790, 410]]<|/det|> +During extreme events in the DN, the occurrence of outages due to component failures can be addressed by a combination of control actions, including reconfiguration and load shedding. We assume that real- time outage detection and protection system responsible for detecting, locating, and isolating faulty components is a preliminary step to the work discussed in this paper. + +<|ref|>text<|/ref|><|det|>[[165, 410, 790, 540]]<|/det|> +Line switches in the DN are typically divided into two categories: switches associated with normally- closed sectionalizing lines and those with normally- open tie lines. During emergency conditions, when component failures disrupt the power supply to the network loads, reconfiguring the DN through control actions on these switches can help maintain network functionality. The objective in such situations is to maximize (or minimize) the energy supplied (or loss of energy) to the loads, despite the network failure, while ensuring operational stability. The optimal switching control depends on factors such as the network state (voltage, branch flow, etc.), network operational limits, and the location and extent of the outage in the network. + +<|ref|>text<|/ref|><|det|>[[165, 540, 790, 667]]<|/det|> +Besides this, the presence of DERs, particularly grid- forming DERs, plays a pivotal role in providing uninterrupted supply to loads following outages. In the off- grid mode, the formation of a self- sustained entity comprising loads and DERs is only possible with the assistance of grid- forming DERs. These grid- forming DERs generate the reference voltage and frequency for the isolated network section while grid- feeding DERs follow this reference and inject active/reactive power into the grid [39]. While the detailed modeling of these DERs is beyond the scope of this work, they are represented as voltage sources when operating in the grid- forming mode, and this characterization is incorporated in the DN model within the environment. + +<|ref|>text<|/ref|><|det|>[[165, 668, 790, 739]]<|/det|> +Reconfiguration is often used as an umbrella term for any change in normal operating network topology using switching control. On the other hand, intentional islanding has long been recognized as a resilience enhancement technique and is a subset of the reconfiguration problem. In scenarios where the outage is extensive and the availability of tie switches is limited, intentional islanding around grid- forming DERs may be + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 832, 144]]<|/det|> +adopted to ensure a continuous power supply. Figure 1 illustrates different switching actions that may be employed based on the extent of the outage. Different outage scenarios that are portrayed here with mitigation strategies represent the possible solutions considered while designing the environment. + +<|ref|>image<|/ref|><|det|>[[212, 163, 824, 433]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 435, 832, 592]]<|/det|> +
Fig. 1: Schematic of an example network with DERs (with and without grid-forming ability), and sectionalizing/tie switches. Outage scenario 1 isolates a small portion of the network, which can be restored by closing the tie line connecting the affected section to the main grid (substation). In the subsequently connected network formed after reconfiguration, the substation serves as the source bus. In scenario 2, a larger part of the network is disconnected without any means to reroute power from the substation. By selecting suitable switching actions, an intentional island is created around the grid-forming DER in scenario 2. The grid-forming DER is the virtual source in the reconfigured network. In scenario 3 with extensive outage, although an intentional island may be formed around the grid-forming DER, a section of the network remains unsupplied despite the presence of a DER (not grid-forming).
+ +<|ref|>text<|/ref|><|det|>[[207, 610, 832, 696]]<|/det|> +Network topology control through switching actions alone cannot guarantee the operational feasibility of the energized sections in the network. Therefore, to ensure sustainable network operation, emergency load shedding is also considered to maintain network voltage within safe operational limits. The loads are modeled as equivalent load at the distribution transformer in the primary distribution system and can be disconnected from the network through switching actions. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[165, 84, 521, 101]]<|/det|> +### 2.2 DN Representation as A Graph + +<|ref|>text<|/ref|><|det|>[[165, 107, 790, 294]]<|/det|> +Outage management in DN using switching control can be largely viewed as a task of learning the associated network topology, which is our motivation to reformulate the problem in graph- theoretic terms. Consequently, we represent the DN as a graph \(\mathcal{G} = (\mathbf{N},\mathbf{E})\) , with an \(\mathbf{N}\) set of nodes interconnected by an \(\mathbf{E}\) set of edges. The nodes in the graph represent the buses in the DN, including the substation, load, DER, and zero- power injection buses. The edges represent the distribution lines and inline transformers. These lines (edges) consist of both switchable (sectionalizing and tie) and non- switchable lines. The node variables comprise both forecasted or estimated variables and measured variables. These variables include the estimated or forecasted values for active power demand (or generation), reactive power demand (or generation), and the three- phase voltage measured at each bus. The edge variable considered is the measured power flow through the branches. To obtain these measured signals, we utilize a power flow simulator in our synthetic approach. + +<|ref|>text<|/ref|><|det|>[[165, 293, 790, 350]]<|/det|> +Network reconfiguration in the graph domain essentially involves determining the status (open or closed) of the switchable edges in the DN. Emergency load shedding at the primary DN level is indicated using a binary variable associated with the nodes representing switchable loads. + +<|ref|>sub_title<|/ref|><|det|>[[165, 364, 730, 382]]<|/det|> +### 2.3 Formulating a Markov Decision Process over Graphs + +<|ref|>text<|/ref|><|det|>[[165, 387, 790, 445]]<|/det|> +The emergency response during outages in the DN is formulated as a Markov Decision Process (MDP) in the graph domain, denoted as \(\mathcal{M} = (\mathcal{S},\mathcal{A},\mathcal{P}_{tr},\mathcal{R})\) . The tuple denotes the state, action, transition probability, and reward (in the respective order), which are defined as follows: + +<|ref|>text<|/ref|><|det|>[[165, 445, 790, 647]]<|/det|> +1. State \((\mathcal{S})\) : The state is composed of relevant observations from the DN that represent the current operating condition of the network. It includes node variables, edge variables, network topology, and other system variables, denoted as \(\mathcal{S} = [P_{d}^{N},Q_{d}^{N},P_{g}^{N},Q_{g}^{N},V^{N},V_{\mathrm{vol}},I^{E},\mathcal{T},E_{supp},\mathcal{O},\mu ]\) . Here, \(P_{d}^{N},Q_{d}^{N}\) represents the estimated or forecasted active and reactive power demand at the nodes, while \(P_{g}^{N},Q_{g}^{N}\) corresponds to the active and reactive power generation at the nodes. The three- phase voltage measured at the buses (graph nodes) is represented as \(V^{N}\) , and \(V_{\mathrm{vol}}\) indicates the voltage violation in the network. The edge variable includes the power flow through the network branches, denoted as \(I^{E}\) . The operating topology of the network is \(\mathcal{T}\) , and the total energy supplied in the network is represented by \(E_{supp}\) . The variable \(\mathcal{O}\) encapsulates the outage scenario, i.e., the multi-line failures in the network, including switch outages. To account for the outage of switches, we use a masking mechanism that suppresses the corresponding switching action, represented by the variable \(\mu\) . + +<|ref|>text<|/ref|><|det|>[[165, 647, 790, 732]]<|/det|> +2. Action \((\mathcal{A})\) : The control actions for emergency response include switching and load shedding. Therefore, the action space is represented as \(\mathcal{A} = [\delta_{1}^{sw},\delta_{2}^{sw},\dots,\delta_{N^{s}}^{sw},\delta_{1}^{ld},\delta_{2}^{ld},\dots\delta_{N_{L}}^{ld}]\) . Here \(N\) represent the number of switchable lines, which includes both the sectionalizing and tie lines. The number of switchable loads in the network is denoted as \(N_{L}\) . Line switching is represented by a binary variable \(\delta^{sw}\) where 0 and 1 represent the opening and closing of the switch, respectively. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[225, 85, 831, 115]]<|/det|> +status of the loads is also represented by a binary variable \(\delta^{td}\) , where load served and load shed respectively corresponds to 1 and 0. + +<|ref|>text<|/ref|><|det|>[[207, 115, 832, 187]]<|/det|> +3. Transition probability \((P_{tr})\) : The transition probability captures the dynamic nature of the network with emergency response, denoted as \(\mathcal{P}(s_{t + 1}^{\prime}|s_{t},a_{t})\) . This represents the transition from network state \(s\) at time step \(t\) to state \(s^{\prime}\) at step \(t + 1\) given that action \(a\) is implemented at time step \(t\) . The transition probability is learned by the agent from its interactions with the environment. + +<|ref|>text<|/ref|><|det|>[[207, 186, 831, 215]]<|/det|> +4. Reward \((R)\) : The reward guides the RL algorithm to take optimal control actions for mitigating outages in the DN, which is formulated as follows: + +<|ref|>equation<|/ref|><|det|>[[377, 226, 829, 265]]<|/det|> +\[r(s,a) = \left\{ \begin{array}{ll}E_{supp} - V_{\mathrm{vol}}, & \mathrm{if} C_{viol} = 0,\\ 0, & \mathrm{otherwise}. \end{array} \right. \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[225, 277, 832, 392]]<|/det|> +The reward reflects the goal of improving resilience in the DN by maximizing the energy supplied \(E_{supp}\) while minimizing violations of network operational constraints. The reward takes into account both power flow convergence and its violation (i.e., \(C_{viol}\) ) to determine if an action is rewarded. This is to prevent the selection of any non- feasible or invalid control actions. Additionally, a penalty term is included to account for voltage violations \(V_{\mathrm{vol}}\) in the network. The voltage violations for each bus \(i \in \mathbf{N}\) beyond its upper limit \((\overline{V})\) and lower limit \((\underline{V})\) are evaluated after power flow estimation as follows: + +<|ref|>equation<|/ref|><|det|>[[380, 401, 829, 444]]<|/det|> +\[\Delta V_{max}^{i} = \left\{ \begin{array}{ll}\sum_{j\in \phi}V_{j}^{i} - \overline{V}, & \mathrm{if} V_{j}^{i} > \overline{V}\\ 0, & \mathrm{otherwise} \end{array} \right. \quad (2)\] + +<|ref|>equation<|/ref|><|det|>[[380, 455, 829, 496]]<|/det|> +\[\Delta V_{min}^{i} = \left\{ \begin{array}{ll}\sum_{j\in \phi}\underline{V} -V_{j}^{i}, & \mathrm{if} V_{j}^{i}< \underline{V}\\ 0, & \mathrm{otherwise}. \end{array} \right. \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[206, 496, 832, 568]]<|/det|> +where \(\phi\) denotes the set of phase connections for the bus. The voltage measurements and the energy supplied are estimated in per- units (pu) and calculated with respect to the base voltage, \(kV_{base}\) , and base power \(MVA_{base}\) of the corresponding network. The per- unit calculations in power systems eliminate the issue of units and is equivalent to normalizing them using their base values. : + +<|ref|>equation<|/ref|><|det|>[[393, 578, 829, 611]]<|/det|> +\[V_{\mathrm{vol}} = \frac{\sum_{i\in \mathbf{N}}(\Delta V_{max}^{i} + \Delta V_{min}^{i})}{3|\mathbf{N}|}, \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[206, 624, 832, 668]]<|/det|> +where \(|\mathbf{N}|\) is the cardinality of the set of network buses, and \(\Delta V_{max}\) and \(\Delta V_{min}\) represent the violations over maximum and minimum desirable voltage limits, respectively. + +<|ref|>sub_title<|/ref|><|det|>[[207, 682, 515, 699]]<|/det|> +### 2.4 Design of the Environment + +<|ref|>text<|/ref|><|det|>[[206, 705, 832, 735]]<|/det|> +The distribution network models are implemented and simulated using the open- source distribution system simulator (OpenDSS) [40]. DERs are modeled using a generic + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 272]]<|/det|> +generator and solar photovoltaic (PV) elements in OpenDSS. Switches are defined on lines with associated switching controls, while the disable/enable property of the loads is used for shedding or picking up load. OpenDSSDirect [41] is employed as the Python- based API to maneuver circuit modifications, I/O operations, and network topology extraction. The equivalent graph is constructed for the circuit using the NetworkX module. The overall framework of the environment is presented in Fig. 2. The implementation of specific switching actions may lead to the formation of multiple components within the network. These components are then translated into isolated DN sections within the DSS circuit. Furthermore, as discussed earlier, intentional islands created by grid- forming DERs are considered a potential solution to tackle outages. To enable power flow evaluation in the isolated DSS circuit section, a virtual slack or reference bus is defined at the location of the grid- forming DERs. This requires assigning a voltage source element to the selected buses (i.e., nodes). + +<|ref|>sub_title<|/ref|><|det|>[[165, 286, 426, 303]]<|/det|> +### 2.5 Learning Architecture + +<|ref|>text<|/ref|><|det|>[[165, 309, 790, 508]]<|/det|> +The learning architecture utilizes a policy gradient- based RL algorithm, where the policy network is derived from a Graph Neural Network (GNN). Each node \(i\) in the DN graph has properties such as active/reactive power demand, generation, and three- phase voltage measurements, denoted as \(\gamma_{i} = [P_{d}^{i},Q_{d}^{i},P_{g}^{i},Q_{g}^{i},V^{i}]\) . The policy network takes the state information (detailed in Section 2.3) as input and produces an action. The policy network consists of three main components: 1) A GNN which is used to compute the graph node embeddings for the DN graph. 2) A feedforward network that is used to compute a feature vector, referred to as context embedding. This vector incorporates information that cannot be naturally represented in the graph structure, such as the energy supplied, voltage violations, and power flow through the edges. 3) A MLP that takes the node embeddings from the GNN and the context embeddings from the feedforward network as input. It computes a final feature vector that encompasses the entire state space information. Figure 2 shows the overall structure of the policy network, which includes the GNN- based feature abstraction. + +<|ref|>text<|/ref|><|det|>[[165, 508, 790, 584]]<|/det|> +Initially, the node properties \(\gamma_{i},i\in N\) , are projected to a higher- dimensional space using linear transformation: \(F_{\mathrm{init}}^{i} = W_{\mathrm{init}}\times \gamma_{i} + b_{\mathrm{init}}\) , where \(W_{\mathrm{init}}\in \mathbb{R}^{|\gamma_{i}|\times h_{0}}\) and \(b_{\mathrm{init}}\) are learnable weights and biases, respectively. The cardinality of a vector or set is denoted by \(|.|\) , and \(h_{0}\) represents the projection length. Let \(F_{\mathrm{init}}\) be a matrix ( \(\in \mathbb{R}^{|N|\times h_{0}}\) ) that represents all \(F_{\mathrm{init}}^{i},i\in N\) , \((F_{\mathrm{init}} = [F_{\mathrm{init}}^{1},F_{\mathrm{init}}^{2},\dots F_{\mathrm{init}}^{|N|}]\) ) + +<|ref|>text<|/ref|><|det|>[[165, 583, 790, 627]]<|/det|> +Node embeddings: Each feature vector \(F_{\mathrm{init}}^{i},i\in \mathbf{N}\) , is then passed through a series of Graph capsule layers. These layers utilize a graph convolutional filter of polynomial form to compute a matrix \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\) , defined as: + +<|ref|>equation<|/ref|><|det|>[[325, 637, 788, 680]]<|/det|> +\[f_{p}^{(l)}(\mathcal{X},\mathcal{L}) = \sum_{k = 0}^{K}\mathcal{L}^{k}(F_{(l - 1)}(\mathcal{X},\mathcal{L})^{\circ p})W_{pk}^{(l)}. \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[164, 692, 790, 736]]<|/det|> +Here, \(\mathcal{L}\) represents the graph Laplacian, \(p\) is the order of the statistical moment, \(K\) is the degree of the convolutional filter, \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\) denotes the output from layer \(l - 1\) , and \(F_{(l - 1)}(\mathcal{X},\mathcal{L})^{\circ p}\) represents \(p\) times element- wise multiplication of \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[218, 85, 822, 494]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 497, 832, 610]]<|/det|> +
Fig. 2: The learning framework developed which includes the environment and the policy network architecture with GNN-based feature abstraction. The environment is composed of a DN model interface (DSS circuit) using a python-based API and the graph replica of the DSS circuit. Voltage sources are introduced at virtual slacks for intentional islanding scenarios. The policy network uses the state information from the environment to compute graph node embeddings and context node embeddings using GNN and feedforward networks, respectively. A final feature vector that encompasses the two embeddings is computed by an MLP.
+ +<|ref|>text<|/ref|><|det|>[[206, 634, 832, 700]]<|/det|> +Here, \(F_{(l - 1)}(\mathcal{X},\mathcal{L})\in \mathbb{R}^{N_{n}\times h_{l - 1}p}\) \(W_{pk}^{(l)}\in \mathbb{R}^{h_{l - 1}p\times h_{l}}\) . The variable \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\in \mathbb{R}^{N_{n}\times h_{l}}\) is a matrix, where each row is an intermediate feature vector for each node \(i\in \mathbf{N}\) , infusing nodal information from \(L_{e}\times K\) hop neighbors, for a value of \(p\) . The output of layer \(l\) is obtained by concatenating all \(f_{p}^{(l)}(\mathcal{X},\mathcal{L})\) , as given by: + +<|ref|>equation<|/ref|><|det|>[[340, 710, 829, 730]]<|/det|> +\[F_{l}(\mathcal{X},\mathcal{L}) = [f_{1}^{(l)}(\mathcal{X},\mathcal{L}),f_{2}^{(l)}(\mathcal{X},\mathcal{L}),\dots ,f_{p}^{(l)}(\mathcal{X},\mathcal{L})]. \quad (6)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[164, 85, 790, 317]]<|/det|> +Here, \(\mathcal{P}\) is the highest order of statistical moment, and \(h_{l}\) is the node embedding length of layer \(l\) . We consider all the values of \(h_{l}\) , \(l \in [0, L_{e}]\) , to be the same throughout the paper Equations (5) and (6) are computed for \(L_{e}\) layers, where each layer uses the output from the previous layer \((F_{l - 1}(\mathcal{X}, \mathcal{L}))\) . Increasing the number of layers \((L_{e})\) and raising the value of \(K\) can enhance the learning of the overall structure of the graph by aggregating nodal neighborhood features from \(L_{e} \times K\) neighbors. However, this improvement comes at the expense of having more learnable parameters in the policy, which becomes a drawback as the problem size increases. A larger value of \(h_{l}\) is beneficial as it enables the computation of a more detailed and comprehensive nodal state representation, both at the final stage and in intermediate steps. Similarly, a larger value of \(P\) assists in a better encoding of intermediate states using a vector representation (described in Eq. 6) for each intermediate feature. This richer structural embedding is expected to be more effective than the scalar embedding used in GCN (Graph Convolutional Networks). However, it is important to note that both higher \(h_{l}\) and \(P\) come with additional training costs. The final node embeddings are computed using a linear transformation of \(F_{l = L_{e}}(\mathcal{X}, \mathcal{L})\) . + +<|ref|>equation<|/ref|><|det|>[[379, 328, 787, 345]]<|/det|> +\[F_{\mathrm{Nodes}} = F_{l = L_{e}}(\mathcal{X}, \mathcal{L}).W_{F}, \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[165, 356, 598, 372]]<|/det|> +where \(W_{F}\) is a learnable weight matrix of size \(h_{L_{e}} \mathcal{P} \times h_{L_{e}}\) . + +<|ref|>text<|/ref|><|det|>[[165, 371, 790, 400]]<|/det|> +The final graph embedding is computed by passing the node embeddings matrix \(F_{\mathrm{Nodes}}\) through a series of Linear layers, followed by taking the mean + +<|ref|>equation<|/ref|><|det|>[[332, 412, 787, 430]]<|/det|> +\[F_{\mathrm{graph}} = \mathrm{Mean}(W_{g2} \times (W_{g1} \times F_{\mathrm{Nodes}})), \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[165, 440, 790, 471]]<|/det|> +where \(W_{g1} \in \mathbb{R}^{h_{L_{e}} \times |N|}\) and \(W_{g2} \in \mathbb{R}^{h_{L_{e}} \times h_{L_{e}}}\) , and \(F_{\mathrm{graph}} \in \mathbb{R}^{h_{L_{e}}}\) , for ease of representation, the bias terms are omitted here. + +<|ref|>text<|/ref|><|det|>[[165, 470, 790, 528]]<|/det|> +Context: In addition to the graph- based information, certain state space variables cannot be directly represented as nodes in the graph. These variables include energy supplied \(E_{supp}\) , voltage violation \(V_{\mathrm{viol}}\) , and power flow through the edges \(l^{E}\) . To incorporate this information, a feature vector called the context is constructed. + +<|ref|>equation<|/ref|><|det|>[[293, 539, 787, 557]]<|/det|> +\[F_{\mathrm{context}} = \mathrm{Feedforward}(\mathrm{Concat}([E_{supp}, V_{\mathrm{viol}}, l^{E}])) \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[165, 568, 790, 600]]<|/det|> +Final MLP layer: The final state embedding \(F_{\mathrm{final}}(\mathbb{R}^{h_{L_{e}}})\) is computed by adding \(F_{\mathrm{graph}}\) and \(F_{\mathrm{context}}\) and passing it through an MLP layer: + +<|ref|>equation<|/ref|><|det|>[[358, 611, 787, 628]]<|/det|> +\[F_{\mathrm{final}} = \mathrm{MLP}(F_{\mathrm{graph}} + F_{\mathrm{context}}). \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[165, 639, 790, 742]]<|/det|> +The Logits \(\in \mathbb{R}^{|\mathcal{A}|}\) across all available actions are computed by passing \(F_{\mathrm{final}}\) through a Feedforward layer. The Logits of the switches that need to be masked are set to negative infinity. Using the Logits, a Bernoulli probability distribution is computed for all available actions, with the probabilities computed using a Sigmoid function as \(e^{\mathrm{Logits}} / (1 + e^{\mathrm{Logits}})\) . The final switching action is determined using a greedy policy. If the mean of an action element (switch) is greater than 0.5, the switch position is set as on (or a value of 1). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[206, 87, 830, 115]]<|/det|> +The predicted value of the state is computed by passing \(F_{\mathrm{final}}\) through another feedforward layer, which approximates the value of the state. + +<|ref|>sub_title<|/ref|><|det|>[[207, 130, 415, 147]]<|/det|> +### 2.6 Training Process + +<|ref|>text<|/ref|><|det|>[[206, 153, 832, 510]]<|/det|> +The training process involves generating samples on the distribution network to simulate different outage scenarios. This is accomplished by introducing line failures, adjusting load and generation operating points, and considering various outage scenarios. The outage events in the network are primarily caused by distribution line failures, which are simulated using a graph- based approach (discussed in 4.1). The power network operating points (i.e., the load demand and power generation) are randomly drawn out of an annual profile made available in OpenDSS. To train the policy network, we employ Proximal Policy Optimization (PPO) [42]. The training process involves collecting experience in the form of tuples containing the state, action, reward, and the next state. PPO operates based on rollout operations, where each operation consists of a fixed number of steps, denoted as \(N_{\mathrm{steps}}\) . The weight updates occur after completing a rollout operation, in batches of size \(N_{\mathrm{batch}}(\leq N_{\mathrm{steps}})\) . The weight update is performed via backpropagation, aiming to minimize a cost function comprising the policy gradient loss and the state value approximation loss. The policy network was trained for a total of \(N_{\mathrm{total}}\) number of steps. To evaluate the performance of the proposed model, as well as to assess the impact of local and global structural information in the encoding process, we conducted comparative experiments with another learning- based framework called MLP. This framework utilizes the PPO algorithm, with a policy network based on a simple Multi- Layer Perceptron (MLP) architecture. To ensure a fair and unbiased comparison, MLP was trained using the same settings as GCAPS. Analyzing the convergence curve depicted in Fig. 3, it becomes evident that GCAPS consistently achieves a higher reward compared to MLP for both the 13- bus and 34- bus systems. This observation demonstrates the superior performance of GCAPS in effectively managing outages and optimizing the distribution network's operational state. + +<|ref|>sub_title<|/ref|><|det|>[[207, 525, 670, 541]]<|/det|> +### 2.7 Case Study on 13-bus Network: Scenario 1 + +<|ref|>text<|/ref|><|det|>[[207, 547, 832, 733]]<|/det|> +The proposed model for outage management is validated using a modified version of the IEEE 13- bus distribution test network. This network incorporates switches and DERs and serves as the basis for validating the effectiveness of the proposed model, as shown in Fig. 4. Two grid- forming DERs of \(1000\mathrm{kW}\) are considered at buses '634' and '680', while the buses '645', '675', and '684' are equipped with grid- feeding DERs rated at \(40\mathrm{kW}\) , \(500\mathrm{kW}\) , and \(100\mathrm{kW}\) , respectively. The total connected load of the network is \(3.5\mathrm{MW}\) . In the normal configuration of the network, the sectionalizing switches are closed, while the tie switches remain open. This initial setup establishes the baseline operational state for the network. To systematically evaluate the developed model and its performance, two traditional optimization techniques, namely the mixed integer second- order conic programming (MISOCP) and binary particle swarm optimization (BPSO), are employed for all case studies in addition to the previously discussed MLP model. In the testing phase of the models, we rationally select the number and location + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[164, 95, 761, 290]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 306, 790, 364]]<|/det|> +
Fig. 3: The training convergence plots for the policy models. Our GCAPS-based RL model is compared with the MLP model which does not utilize graph abstraction. (a) Convergence plot of GCAPS and MLP for the 13-bus network. (b) Convergence plot for the 34-bus network.
+ +<|ref|>text<|/ref|><|det|>[[165, 389, 790, 446]]<|/det|> +of the line outages as opposed to the graph- based approach used during training. Additionally, the load and generating points are not drawn out of the representative annual profile discussed in training, rather a randomly generated multiplying factor is used to set the network operating point. + +<|ref|>text<|/ref|><|det|>[[165, 447, 790, 674]]<|/det|> +Scenario 1 in 13 bus network involves the failure of a single line of importance, determined by its high edge- betweenness in normal configuration. Specifically, this scenario represents the outage of the line connecting buses '670'- '671'. The status of the decision variables, which includes both the switches and dispatchable loads, obtained from the different models for scenario 1 is depicted in Fig. 5. Notably, both the traditional optimization models, namely the MISOCP and the BPSO, yield the same solution for scenario 1. An important observation from analyzing the statuses of the switches and loads is that the reinforcement learning models demonstrate generalizability by providing distinct solutions for the two different scenarios. The voltage plot of the 13- bus network, after implementing the GCAPS solution for managing outage scenario 1, is illustrated in Fig. 6. Due to the switching actions in scenario 1, the resulting network configuration maintains a robust connection to the substation, ensuring that the voltages at all active phases of connected buses are within 0.95 and 1.10 pu, thus operating well within the desirable bounds. The voltage plot highlights the buses shaded in grey, which represent the buses disconnected from the main network due to the control actions. + +<|ref|>sub_title<|/ref|><|det|>[[165, 689, 630, 705]]<|/det|> +### 2.8 Case Study on 13-bus Network: Scenario 2 + +<|ref|>text<|/ref|><|det|>[[165, 712, 790, 740]]<|/det|> +Scenario 2 involves the outage of three switchable lines connecting '632'- '670', '671'- '692', and '646'- '684'. This scenario aims to test the capability of the proposed model + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[196, 81, 834, 280]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 283, 833, 355]]<|/det|> +
Fig. 4: Test networks used to validate the proposed GCAPS model for real-time resilient control. The test networks are IEEE distribution networks modified by the addition of switches and DERs. Switches in the networks include sectionalizing and tie switches. Both types of DERs, i.e., grid-forming and grid-feeding, are considered. (a) Modified 13-bus IEEE test network. (b) Modified 34-bus IEEE test network.
+ +<|ref|>text<|/ref|><|det|>[[207, 380, 832, 595]]<|/det|> +to enforce the inoperability of the outage switch in decision support. The status of decision variables, including the switches and dispatchable loads, obtained from the different models for scenario 2, is shown in Fig. 5. Once again, the MISOCP and the BPSO solutions for scenario 2 are identical. Upon inspecting the decision variables, it is noticeable that the MLP- based RL model violates the non- switchable condition of the outage line '671'- '692' (sw2) for scenario 2, as it mistakenly closes the switch. The voltage plot of the 13- bus network, after implementing the GCAPS solution for managing outage scenario 2, is shown in Fig. 6. In scenario 2, the GCAPS outage mitigation solution ensures a functional network with voltages at all active phases of connected buses ranging from 1.10 pu to 0.8 pu. Although there is a wider margin from the lower limit of 1 pu in scenario 2, it can be attributed to the outage of multiple switches (sectionalizing and tie) and the resulting formation of an island around bus '680' due to the control action. As a consequence of isolating the downstream section from the stronger substation bus, the phase voltages at the downstream buses slightly decrease. + +<|ref|>sub_title<|/ref|><|det|>[[207, 609, 670, 626]]<|/det|> +### 2.9 Case Study on 34-bus Network: Scenario 1 + +<|ref|>text<|/ref|><|det|>[[207, 633, 832, 733]]<|/det|> +The validation of the proposed model and baselines is conducted on a modified 34- bus distribution test network, which incorporates switches and DERs. The total connected load of the network is 2.04 MW. Three grid- forming DERs with capacities of 146 kW, 144 kW, and 1000 kW are connected at buses '890', '844', and '816', respectively while a grid- feeding DER with a capacity of 96 kW is connected at bus '820'. Under normal operating conditions, the five sectionalizing switches are closed, while the four tie switches are open. Scenario 1 involves line outages at the connections between + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[159, 91, 789, 200]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 209, 791, 283]]<|/det|> +
Fig. 5: Status of decision variables acquired from the proposed model and baselines for the 13-bus test network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outage '670-671'. (b) Status of decision variables for test scenario 2 with line outages '632-670', '671-692', and '646-684'.
+ +<|ref|>image<|/ref|><|det|>[[150, 315, 788, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 488, 791, 574]]<|/det|> +
Fig. 6: Voltage plot of the 13-bus test network with the GCAPS outage management solution for varying test scenarios. The active phases of the network are only marked in the plot and the grey-shaded nodes represent those disconnected from the network. (a) Voltage plot for test scenario 1 with the GCAPS solution implemented in the network during outages. (b) Voltage plot for test scenario 2 with the GCAPS solution implemented in the network during outages.
+ +<|ref|>text<|/ref|><|det|>[[165, 599, 791, 728]]<|/det|> +buses '812'- '814' and '824'- '848'. The line connecting '812'- '814' is upstream in the network and closer to the substation, indicating a severe impact on the network due to its outage. The line '824'- '848', on the other hand, is a switchable tie line (sw7). Fig. 7 presents the status of the decision variables, including switchable lines and loads, obtained from the different models for scenario 1. Both the MISOCP and BPSO yield similar results for scenario 1 on the 34-bus network. The results demonstrate the ability of RL models to differentiate between various scenarios and generalize during decision- making. However, the MLP- based RL model produces an invalid control action in scenario 1 by closing switch sw7 on the outage line. The switching action obtained from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 832, 201]]<|/det|> +the GCAPS outage mitigation model results in the formation of an intentional island around DER at bus '816' for scenario 1. Consequently, the section downstream of '812'- '814' becomes a self- sustained network component disconnected from the substation. The voltage plot for the 34- bus network, derived by implementing the GCAPS solution during outages for scenario 1, is presented in Fig. 8. Scenario 1, being severe due to the outage of an upstream line, requires the GCAPS solution to intentionally isolate bus '816' to mitigate the outage. Consequently, the phase voltages at buses drop significantly, although the network remains functional. + +<|ref|>sub_title<|/ref|><|det|>[[207, 215, 684, 232]]<|/det|> +## 2.10 Case Study on 34-bus Network: Scenario 2 + +<|ref|>text<|/ref|><|det|>[[207, 238, 832, 380]]<|/det|> +Scenario 2 considers multiple line failures at '832'- '858', '834'- '860', and '854'- '852' in the network. The lines '832'- '858' and '854'- '852' are in close proximity, while the line '834'- '860' is a switchable sectionalizing line (sw2). Figure 7 presents the status of the decision variables, including the switchable lines and loads, obtained from the different models for scenario 2. In scenario 2, the switching action by the GCAPS model results in a configuration that remains connected to the substation, with a small section disconnected (inactive) from the main network. The voltage plot for the 34- bus network, derived by implementing the GCAPS solution for scenario 2, is presented in Fig. 8. It is observed that the GCAPS solution for scenario 2 ensures voltages at all active phases of connected buses are well within the range of 0.95 to 1.06 pu. + +<|ref|>image<|/ref|><|det|>[[198, 407, 835, 508]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[206, 517, 832, 602]]<|/det|> +
Fig. 7: Status of decision variables acquired from the proposed model and baselines for the 34-bus network. The light entries denote '0' indicating an open status while the darker entries denote '1' representing a closed status. (a) Status of decision variables for test scenario 1 with line outages '812-814' and '824-848'. (b) Status of decision variables for test scenario 2 with line outages '832-858', '834-860', and '854-852'.
+ +<|ref|>sub_title<|/ref|><|det|>[[207, 646, 762, 664]]<|/det|> +### 2.11 Comparison of the Proposed Model with Baselines + +<|ref|>text<|/ref|><|det|>[[207, 669, 832, 742]]<|/det|> +We compare the developed GCAPS- based RL model with the baseline models to evaluate the performance speed and the estimated energy served during outage conditions. Figure 9 presents the estimated equivalent energy served when implementing the control decisions in the 13- bus distribution test network for scenarios 1 and 2 using the different models. As expected, the energy supplied is optimal for the MISOCP and + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[238, 92, 715, 469]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[164, 472, 791, 530]]<|/det|> +
Fig. 8: Voltage plot of the 34-bus test network for varying outage scenarios with the GCAPS outage management solution. The active phases of the network are only included in the plot. The grey-shaded nodes are disconnected from the network. (a) Outage scenario 1. (b) Outage scenario 2.
+ +<|ref|>text<|/ref|><|det|>[[165, 554, 790, 698]]<|/det|> +BPSO models. Our GCAPS model shows near- optimal decision- making capability for both scenarios. In scenario 1, the MLP model is inferior as it provides the minimum energy supply among all the models, while it becomes invalid in scenario 2 due to the operation of the outage switch. The estimation of equivalent energy served obtained by implementing the control actions derived from the different models in the 34- bus distribution test network for scenarios 1 and 2 is shown in Fig. 9. Our GCAPS model exhibits near- optimal performance, closely approaching the optimal energy supply estimated by the MISOCP and BPSO models. On the other hand, the MLP model performs inferiorly compared to the other models and also produces an invalid control action for scenario 1. + +<|ref|>text<|/ref|><|det|>[[165, 698, 790, 741]]<|/det|> +The performance of our GCAPS model is compared with the baselines by testing different scenarios in 13- bus and 34- bus networks. The computation time required to obtain the outage mitigation solution is presented in Table 1. It can be observed that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[207, 87, 831, 186]]<|/det|> +the response time for the two RL- based models, namely GCAPS and MLP, is in the order of milliseconds, and is mostly agnostic to the increase in the size of the network from 13 to 34 bus system, demonstrating real- time performance. In comparison, the optimization- based methods, BPSO and MISOCP respectively require seconds or 10's of minutes to compute those decisions. Specifically, BPSO and MISOCP are respectively about 5 and 3 orders of magnitude more expensive than the learned RL- based policies. + +<|ref|>text<|/ref|><|det|>[[207, 187, 831, 471]]<|/det|> +In Fig. 10, we illustrate the performance of the DN and its evolution with time when implementing the decisions provided by the different models during outages. Specifically, the proposed GCAPS- based RL model is compared with the MLP- based RL model which does not consider the underlying topology and the MISOCP method (conventionally used for solving such problems). The BPSO despite producing similar results as the MISOCP is not suitable for resilience decision support as is evident from the prohibitively delayed response shown in Table 1. Outage scenario 1 in the 13- bus network and outage scenario 2 in the 34- bus network are used to exemplify the impact of the model response on DN performance. The excluded scenarios in the two networks are not suitable for comparison owing to the invalid switching decisions provided by the MLP model as discussed earlier. As observed in Fig. 10, the voltages at the buses '652' and '890' in the 13 and 34 bus networks respectively are under voltage due to disruption. The voltage violation exists for about 100's and 1000's of cycles in the 13 and 34 bus DNs when MISOCP is used for decision support (as it takes 5- 20 sec to compute the decisions). While the RL models mitigate the voltage violation through outage management almost instantaneously. The continued operation of the network in the disrupted state also increases the risk of cascaded failures and widespread blackouts. Meanwhile, the loss of energy due to delayed decision- making by the MISOCP with respect to the GCAPS is 4,204.50 kWs and 13,522.18 kWs for 13 and 34 buses respectively. + +<|ref|>table<|/ref|><|det|>[[312, 530, 722, 646]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[313, 498, 722, 526]]<|/det|> +Table 1: Performance comparison of different models for scenarios in the test networks + +
NetworkMethodScenario 1
Mean time (s)		Scenario 2
Mean time (s)
13-busGCAPS0.00490.0056
MLP0.00390.0054
BPSO500.15540.20
MISOCP3.125.23
34-busGCAPS0.00300.0025
MLP0.00220.0020
BPSO2580.152540.20
MISOCP25.2020.20
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[184, 90, 788, 226]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[165, 234, 792, 321]]<|/det|> +
Fig. 9: Comparison of the resilience improvement for different models with varying outage scenarios. The resilience improvement is measured in terms of equivalent energy served when implementing control actions during outages. (a) Testing on the 13-bus network, where an empty bar for MLP in scenario 2 indicates invalidity due to the operation of the outage switch. (b) Testing on the 34-bus network, with MLP model for scenario 1 invalid due to operation of outage switches.
+ +<|ref|>sub_title<|/ref|><|det|>[[166, 344, 317, 363]]<|/det|> +## 3 Discussion + +<|ref|>text<|/ref|><|det|>[[165, 371, 791, 529]]<|/det|> +We have presented a real- time outage management model for distribution networks based on a reinforcement learning over graphs framework. In our outage management model, we have considered the grid- forming and feeding modes of the DER, and hence both grid- connected and islanding reconfiguration schemes have been incorporated into the solution. The load shedding adopted in the mitigation strategy ensures that the network has operational feasibility and is not vulnerable to voltage collapse. The learning model employs an on- policy RL algorithm and adopts the novel Graph Capsule (GCAPS) neural networks for integrating information about the DN topology into the learning framework. By leveraging GCAPS neural networks, the model has been shown to effectively integrate nodal properties, and local and global structural information into the learning process. + +<|ref|>text<|/ref|><|det|>[[165, 529, 791, 715]]<|/det|> +We have evaluated our model on modified versions of the IEEE 13- bus and 34- bus distribution test networks, which include distributed energy resources (DERs) and sectionalizing/tie switches. Two traditional models based on MISOCP and BPSO, and the RL with MLP as policy network have been used as baselines to compare the real- time decision- making and network resilience improvement capability, where the energy served under disruption (see Fig. 9) can be perceived as a measure of resilience. The results have demonstrated that the proposed model achieves near- optimal performance in real- time outage management for different networks and outage scenarios. Additionally, the model has been found to effectively capture the DN topology in decision- making as indicated by the improved performance and constraint adherence when compared with the MLP- based approach. Above all, our model has also provided time- sensitive decision support for outage mitigation, thereby making it a suitable self- healing tool in the current "smart- grid" landscape. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[238, 91, 795, 385]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[205, 389, 832, 546]]<|/det|> +
Fig. 10: The comparison of the proposed GCAPS model with baseline models. The baseline models include an MLP-based model which does not utilize graph abstraction and the (much slower to compute) conventional mathematical programming technique (here MISOCP). The network is in the disrupted state at time instant '0'. (a) Voltage measured at the bus '652' (most stressed bus) is used as an indicator for 13-bus network operational characteristics in outage scenario 1. (b) Equivalent power supplied in the network evolving with switching decisions for 13 bus in outage scenario 1. (c) Voltage measured at the bus '890' (most stressed bus) is used for comparing 34-bus network operational limits in outage scenario 2. (d) Equivalent power supplied in the 34-bus network by different models for scenario 2 and their response with time is illustrated.
+ +<|ref|>text<|/ref|><|det|>[[206, 572, 832, 730]]<|/det|> +Currently, the proposed method have been applied to relatively small- medium sized networks, and applicability to larger networks remains to be explored. One challenge to accomplishing that is the associated training cost in terms of memory and computing time. From the results in our prior studies on applying related graph- based RL for Multi- Robot Task Allocation [38, 43], we have found that the computational memory requirement for training on larger graphs (more than 200 nodes) is very high and often hinders the training task. Our prior results [38, 43] have demonstrated the capability of the GNN- based policy network to learn policies that can be applied to a larger- sized mostly homogeneous networks with simple near- linear state transitions (without training), while still demonstrating comparable performance with respect to more traditional approaches. More work is required to explore if these advantages will + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 115]]<|/det|> +also translate to applications such as the DN topology reconfiguration that involves heterogeneous networks and non- linear flow properties that affect the state transition. + +<|ref|>sub_title<|/ref|><|det|>[[165, 130, 298, 148]]<|/det|> +## 4 Methods + +<|ref|>sub_title<|/ref|><|det|>[[165, 159, 541, 176]]<|/det|> +## 4.1 Graph-based Scenario Generation + +<|ref|>text<|/ref|><|det|>[[165, 183, 790, 425]]<|/det|> +The training scenarios used for RL over the graphs model is generated from the graph equivalent of the DN. The failure of the components, such as lines, can be approximated by disconnecting them from the DN [44]. The model developed is not specific to any particular type of extreme weather event, and hence a generalized and intuitive approach is adopted for simulating outages during training. The outages in the DN often originate from localized failures that can lead to cascading effects. To emulate this behavior, a subgraph method for randomized edge removal is employed, similar to the approach described in [45]. This method involves randomly selecting nodes \(N_{s} \in N\) from the graph representation of the DN, and creating subgraphs centered around these nodes with varying radii \(R_{s} \leq R_{max}\) (maximum radius). We consider \(R_{max} = \frac{G_{\mathrm{dia}}}{2}\) , where \(G_{\mathrm{dia}}\) is the diameter of the graph. Within each selected subgraph, a fraction of the edges \(F_{s} \in E\) is randomly removed to simulate the localized impact of contingencies. The fraction of edge failures is gradually increased from 0 to 50%. By varying \(N_{s}, R_{s}\) , and \(F_{s}\) , scenarios with multi- line failures can be generated for training the model. Furthermore, within each scenario, load multipliers and generating points are varied by randomly selecting multipliers from an annual profile available in OpenDSS package with an hourly resolution. + +<|ref|>sub_title<|/ref|><|det|>[[165, 439, 617, 457]]<|/det|> +### 4.2 Mixed Integer Programming Formulation + +<|ref|>text<|/ref|><|det|>[[165, 462, 790, 548]]<|/det|> +Outage management in an unbalanced distribution network is an optimization problem that combines combinatorial and non- linear nature. The problem can be effectively formulated as an optimal power flow problem, leveraging branch flow equations with angle and conic relaxations as in [17]. The decision variables include switching and load shedding, while the control variables corresponding to power flow are also considered in the problem formulation. + +<|ref|>text<|/ref|><|det|>[[165, 548, 790, 589]]<|/det|> +For the distribution network with \(\mathbb{L}\) set of loads and \(\tilde{\mathbb{L}}\) set of switchable loads, the active/reactive power consumption with load pickup or shedding is modeled using \(\delta^{L}\) as follows: + +<|ref|>equation<|/ref|><|det|>[[348, 585, 788, 666]]<|/det|> +\[\begin{array}{r l} & {P_{i}^{L} = \left\{ \begin{array}{l l}{\delta_{i}^{L}P_{i}^{D}} & {\mathrm{if~}i\in \widetilde{\mathbb{L}}}\\ {P_{i}^{D}} & {\mathrm{otherwise}} \end{array} \right.;\forall i\in \mathbb{L}}\\ & {Q_{i}^{L} = \left\{ \begin{array}{l l}{\delta_{i}^{L}Q_{i}^{D}} & {\mathrm{if~}i\in \widetilde{\mathbb{L}}}\\ {Q_{i}^{D}} & {\mathrm{otherwise}} \end{array} \right.;\forall i\in \mathbb{L}} \end{array} \quad (11b)\] + +<|ref|>text<|/ref|><|det|>[[165, 668, 790, 698]]<|/det|> +where \(P_{i}^{D}\) and \(Q_{i}^{D}\) represent the active and reactive power demand of the load \(i\) , respectively. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[205, 86, 830, 116]]<|/det|> +On the other hand, considering the set of grid- feeding generators \(\mathbb{G}_{fd}\) in the DN, the active and reactive power generation is estimated using: + +<|ref|>equation<|/ref|><|det|>[[372, 127, 829, 145]]<|/det|> +\[P_{(j,k)}^{G} = P_{avail}^{G} / |G_{ph}|~;\forall j\in \mathbb{G}_{fd},k\in \theta^{*} \quad (12a)\] + +<|ref|>equation<|/ref|><|det|>[[368, 158, 829, 176]]<|/det|> +\[Q_{(j,k)}^{G} = Q_{avail}^{G} / |G_{ph}|~;\forall j\in \mathbb{G}_{fd},k\in \theta^{*} \quad (12b)\] + +<|ref|>text<|/ref|><|det|>[[206, 178, 831, 222]]<|/det|> +where \(P_{avail}^{G}\) , \(Q_{avail}^{G}\) is the total generation power available for the generator with \(|G_{ph}|\) number of phase connections, and \(\theta^{*}\) is the set of active phases of the generator, considering \(\theta = (a,b,c)\) . + +<|ref|>text<|/ref|><|det|>[[206, 223, 831, 251]]<|/det|> +The total active and reactive power consumption by loads is constrained by the total generation in the DN as follows: + +<|ref|>equation<|/ref|><|det|>[[400, 262, 829, 294]]<|/det|> +\[\sum_{i\in \mathbb{L}}P_{i}^{L}\leq P_{tot}^{G}~,\sum_{i\in \mathbb{L}}Q_{i}^{L}\leq Q_{tot}^{G} \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[230, 307, 653, 322]]<|/det|> +The total power generation in the DN is given as follows: + +<|ref|>equation<|/ref|><|det|>[[385, 332, 829, 366]]<|/det|> +\[P_{tot}^{G} = \sum_{k\in \theta}\big(\sum_{j\in \mathbb{G}_{fd}}P_{(j,k)}^{G} + \sum_{h\in \mathbb{G}_{s}}P_{(h,k)}^{G}\big) \quad (14a)\] + +<|ref|>equation<|/ref|><|det|>[[379, 377, 829, 410]]<|/det|> +\[Q_{tot}^{G} = \sum_{k\in \theta}\big(\sum_{j\in \mathbb{G}_{fd}}Q_{(j,k)}^{G} + \sum_{h\in \mathbb{G}_{s}}Q_{(h,k)}^{G}\big) \quad (14b)\] + +<|ref|>text<|/ref|><|det|>[[206, 414, 831, 442]]<|/det|> +where in addition to the grid- feeding generators, the set of grid- forming generators \(\mathbb{G}_{s}\) , including the substation, are considered. + +<|ref|>text<|/ref|><|det|>[[206, 444, 830, 471]]<|/det|> +The power supplied by the grid- forming generators and the substation is constrained to be within its maximum capacity as follows: + +<|ref|>equation<|/ref|><|det|>[[414, 481, 829, 513]]<|/det|> +\[\sum_{k\in \theta^{*}}P_{(h,k)}^{G}\leq \overline{{P_{h}^{G}}};\forall h\in \mathbb{G}_{s}. \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[206, 525, 831, 569]]<|/det|> +Adopting three- phase branch flow formulations with relaxations as in [17], \(\nu\) and \(\mathcal{I}\) are used to denote the square of voltage and current, respectively. The voltages at all buses except the slack buses are constrained within upper and lower limits as follows: + +<|ref|>equation<|/ref|><|det|>[[391, 581, 829, 597]]<|/det|> +\[\mathcal{V}\leq \mathcal{V}_{r,k}\leq \overline{\mathcal{V}};\forall r\in \mathbb{B}\setminus \mathbb{B}_{s},k\in \theta \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[206, 610, 831, 654]]<|/det|> +Here, \(\mathbb{B}\) and \(\mathbb{B}_{s}\) denote the set of buses and the set of slack buses in the network, respectively. The voltage square at the substation (or slack) bus on the other hand is equated to 1.04 per unit. + +<|ref|>text<|/ref|><|det|>[[206, 655, 831, 698]]<|/det|> +For the set of power delivery elements \(\mathbb{E}\) , the set of switchable elements (lines) \(\mathbb{E}_{sw}\) , and the line switch status \(\delta^{sw}\) , the power flow \(P^{E}\) through the elements are constrained as follows: + +<|ref|>equation<|/ref|><|det|>[[350, 708, 829, 728]]<|/det|> +\[\underline{{P}}_{(b,k)}^{E}\leq \underline{{P}}_{(b,k)}^{E}\leq \overline{{P}}_{(b,k)}^{E};\forall b\in \mathbb{E}\setminus \mathbb{E}_{sw},k\in \theta \quad (17a)\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[305, 85, 788, 106]]<|/det|> +\[\underline{{P}}_{(l,k)}^{E}\delta_{l}^{s w}\leq P_{(l,k)}^{E}\leq \overline{{P}}_{(l,k)}^{E}\delta_{l}^{s w};\forall l\in \mathbb{E}_{s w},k\in \theta \quad (17b)\] + +<|ref|>text<|/ref|><|det|>[[163, 107, 790, 151]]<|/det|> +In a similar manner, the reactive power flow \(Q^{E}\) and the square of branch current square \(\mathcal{I}^{E}\) through the elements are also constrained within its limits. The power flow through the outage lines defined in set \(\mathbb{O}\) is, however, equated to zero as shown below: + +<|ref|>equation<|/ref|><|det|>[[383, 161, 788, 180]]<|/det|> +\[P_{(b,k)}^{E} = 0; \forall b\in \mathbb{O},k\in \theta \quad (18)\] + +<|ref|>text<|/ref|><|det|>[[163, 191, 790, 235]]<|/det|> +The reactive power flow and the square of branch current through outage lines are also equated to zero. The balance of active and reactive power flow through the elements is formulated as follows: + +<|ref|>equation<|/ref|><|det|>[[285, 243, 788, 300]]<|/det|> +\[\begin{array}{r l r}{{P_{(b,k)}^{E}=\sum_{q\in\mathbb{R}_{L}(b)}P_{(q,k)}^{L}-\sum_{w\in\mathbb{R}_{G}(b)}P_{(w,k)}^{G}+\sum_{h\in\mathbb{C}(b)}P_{(h,k)}^{E}}}\\ &{}&{+R_{(b,k)}\mathcal{I}_{(b,k)};\forall b\in\mathbb{E},k\in\theta}\end{array} \quad (19)\] + +<|ref|>equation<|/ref|><|det|>[[280, 325, 788, 379]]<|/det|> +\[Q_{(b,k)}^{E} = \sum_{q\in \mathbb{R}_{L}(b)}Q_{(q,k)}^{L} - \sum_{w\in \mathbb{R}_{G}(b)}Q_{(w,k)}^{G} + \sum_{h\in \mathbb{C}(b)}Q_{(h,k)}^{E} \quad (20)\] + +<|ref|>text<|/ref|><|det|>[[163, 389, 790, 433]]<|/det|> +where \(\mathbb{R}_{L}(b)\) is the set of loads and \(\mathbb{R}_{G}(b)\) is the set of generators connected to the receiving bus of element \(b\) . In (19) and (20), \(\mathbb{C}(b)\) represent the elements that are children elements to \(b\) . + +<|ref|>text<|/ref|><|det|>[[188, 433, 608, 448]]<|/det|> +Additionally, Kirchhoff's voltage equation is modeled as: + +<|ref|>equation<|/ref|><|det|>[[295, 457, 788, 500]]<|/det|> +\[\begin{array}{r l r} & {} & {\mathcal{V}_{(\mathbb{R}(b),k)} = \mathcal{V}_{(\mathbb{S}(b),k)} - 2(\hat{R}_{(b,k)}P_{(b,k)}^{E} + \hat{X}_{(b,k)}Q_{(b,k)}^{E})}\\ & {} & {+\hat{Z}_{(b,k)}\mathcal{I}_{(b,k)});\forall b\in \mathbb{E}\setminus \mathbb{E}_{s w},k\in \theta} \end{array} \quad (21)\] + +<|ref|>text<|/ref|><|det|>[[163, 511, 790, 570]]<|/det|> +Here, the parameters \(\hat{R}\) and \(\hat{X}\) denote the element's modified resistance and reactance, respectively. In (21), \(\mathbb{S}(b)\) and \(\mathbb{R}(b)\) denote the sending and receiving bus of the element \(b\) , respectively. For elements (lines) with switch, Eq. (21) is modified to an inequality constraint using the big M method [17] and bound within \(- (1 - \delta_{l}^{s w})M\) and \((1 - \delta_{l}^{s w})M\) . + +<|ref|>text<|/ref|><|det|>[[163, 570, 790, 597]]<|/det|> +The second- order conic inequality constraint using convex relaxation is formulated as follows: + +<|ref|>equation<|/ref|><|det|>[[285, 596, 788, 614]]<|/det|> +\[\mathcal{I}_{(b,k)}*\mathcal{V}_{(\mathbb{S}(b),k)}\geq [(P_{(b,k)}^{E})^{2} + (Q_{(b,k)}^{E})^{2}];\forall b\in \mathbb{E},k\in \theta \quad (22)\] + +<|ref|>text<|/ref|><|det|>[[163, 616, 790, 644]]<|/det|> +The radiality of the network structure is preserved using the spanning tree constraints as follows: + +<|ref|>equation<|/ref|><|det|>[[378, 653, 788, 744]]<|/det|> +\[\begin{array}{c}{\beta_{l}^{1} + \beta_{l}^{2} = \delta_{l}^{s w};\forall l\in \mathbb{E}_{s w}}\\ {\beta_{b}^{1} + \beta_{b}^{2} = 1;\forall b\in \mathbb{E}\setminus \mathbb{E}_{s w}}\\ {\beta_{b}^{1} + \beta_{b}^{2} = 0;\forall b\in \mathbb{O}}\\ {\beta_{b(s)}^{1} = 0;\forall s\in \mathbb{B}_{s}} \end{array} \quad (23)\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[378, 81, 829, 115]]<|/det|> +\[\sum_{x\in \mathbb{P}(y)}\beta_{(x,y)}^{1} + \sum_{z\in \mathbb{S}(y)}\beta_{(y,z)}^{2}\leq 1;\forall y\in \mathbb{B} \quad (25)\] + +<|ref|>text<|/ref|><|det|>[[206, 118, 832, 220]]<|/det|> +In the above equations, the two edge directions (flows) possible for an element are represented as \(\beta^{1}\) and \(\beta^{2}\) . In Eq. 23, the flow through a branch is constrained to one direction based on the conditionalities: without/with switches, and outage of the element. The branches connected to the source bus indicated by \(b(s)\) are limited to one direction of flow as in Eq. 24. Additionally, each bus is ensured to have at most one parent as in Eq. 25, where \(\mathbb{P}(y)\) and \(\mathbb{S}(y)\) represent the set of preceding and succeeding buses in the neighborhood of bus \(y\) , respectively. + +<|ref|>text<|/ref|><|det|>[[206, 220, 831, 248]]<|/det|> +The objective function maximizes the total power supply in the network with control actions during outages and is formulated as follows: + +<|ref|>equation<|/ref|><|det|>[[471, 258, 829, 290]]<|/det|> +\[min.\sum_{i\in \mathbb{L}}P_{i}^{L} \quad (26)\] + +<|ref|>sub_title<|/ref|><|det|>[[207, 302, 409, 319]]<|/det|> +### 4.3 Training Details + +<|ref|>text<|/ref|><|det|>[[206, 325, 832, 425]]<|/det|> +The training process was allocated a maximum of 36 hours, and the total number of steps is set to 2 million. For the 13- bus systems, both GCAPS and MLP successfully completed the training with 2 million steps. However, for the 34- bus systems, MLP could only be trained for 1.5 million steps within the 36- hour time frame. To ensure a fair comparison, we utilize the trained weights of GCAPS and MLP at 1.5 million steps for the 34- bus system. Table 2 shows the training details, including the hyperparameter setting for PPO. + +<|ref|>table<|/ref|><|det|>[[319, 468, 716, 586]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[424, 451, 614, 466]]<|/det|> +Table 2: Training details + +
ParameterValue
AlgorithmPPO
Total steps (Ntotal)2.0e6 (13 bus), 1.5e6 (34 bus)
Rollout buffer size (Nsteps)5e4
Batch size (Nbatch)2e4
OptimizerAdam
Learning step size1e-6
Value coefficient0.5
Discount factor0.0
Epochs100
+ +<|ref|>sub_title<|/ref|><|det|>[[207, 621, 418, 638]]<|/det|> +### 4.4 Simulation Setup + +<|ref|>text<|/ref|><|det|>[[207, 644, 832, 716]]<|/det|> +The proposed model and all the other baselines are tested on a system with Intel Core i7- 8565U 1.80GHz with 16 GB memory. The OpenDSSDirect API along with Python version 3.9.12, and Networkx version 2.8.4 are used in our simulations. The mixed integer programming is performed with Gurobiy using a Gurobi optimizer version 9.5.2. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[165, 86, 790, 172]]<|/det|> +Acknowledgments. This material is based upon work sponsored by the Department of the Navy, Office of Naval Research under ONR award number N00014- 21- 1- 2530. The United States Government has a royalty- free license throughout the world in all copyrightable material contained herein. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Office of Naval Research. + +<|ref|>text<|/ref|><|det|>[[165, 180, 790, 223]]<|/det|> +Data availability. The data that support the plots within this article and other findings of this study are available from the corresponding author upon reasonable request. + +<|ref|>text<|/ref|><|det|>[[165, 229, 790, 258]]<|/det|> +Code availability. Code for this article is available from the corresponding author upon reasonable request. + +<|ref|>sub_title<|/ref|><|det|>[[165, 273, 294, 291]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[171, 300, 792, 331]]<|/det|> +[1] Campbell, R. J. & Lowry, S. Weather- related power outages and electric system resiliency (Congressional Research Service, Library of Congress Washington, DC, 2012). + +<|ref|>text<|/ref|><|det|>[[172, 355, 792, 398]]<|/det|> +[2] Kirthiga, M. V., Daniel, S. A. & Gurunathan, S. A methodology for transforming an existing distribution network into a sustainable autonomous micro- grid. IEEE Transactions on Sustainable Energy 4, 31- 41 (2012). + +<|ref|>text<|/ref|><|det|>[[172, 409, 792, 452]]<|/det|> +[3] Bouhouras, A. S., Andreou, G. T., Labridis, D. P. & Bakirtzis, A. G. Selective automation upgrade in distribution networks towards a smarter grid. IEEE Transactions on Smart Grid 1, 278- 285 (2010). + +<|ref|>text<|/ref|><|det|>[[172, 464, 708, 480]]<|/det|> +[4] U.S. Department of Energy. 2020 Smart Grid System Report (2022). + +<|ref|>text<|/ref|><|det|>[[172, 491, 792, 533]]<|/det|> +[5] Arefifar, S. A., Alam, M. S. & Hamadi, A. A review on self- healing in modern power distribution systems. Journal of Modern Power Systems and Clean Energy (2023). + +<|ref|>text<|/ref|><|det|>[[172, 545, 790, 588]]<|/det|> +[6] Wang, Y. et al. Coordinating multiple sources for service restoration to enhance resilience of distribution systems. IEEE Transactions on Smart Grid 10, 5781- 5793 (2019). + +<|ref|>text<|/ref|><|det|>[[172, 600, 790, 629]]<|/det|> +[7] Fan, D. et al. Restoration of smart grids: Current status, challenges, and opportunities. Renewable and Sustainable Energy Reviews 143, 110909 (2021). + +<|ref|>text<|/ref|><|det|>[[172, 641, 790, 684]]<|/det|> +[8] Baran, M. E. & Wu, F. F. Network reconfiguration in distribution systems for loss reduction and load balancing. IEEE Power Engineering Review 9, 101- 102 (1989). + +<|ref|>text<|/ref|><|det|>[[172, 695, 790, 724]]<|/det|> +[9] Jacob, R. A. & Zhang, J. Distribution network reconfiguration to increase photovoltaic hosting capacity, 1- 5 (IEEE, 2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[205, 87, 833, 117]]<|/det|> +[10] Jacob, R. A. & Zhang, J. Outage management in active distribution network with distributed energy resources, 1–6 (IEEE, 2021). + +<|ref|>text<|/ref|><|det|>[[206, 127, 833, 157]]<|/det|> +[11] Al Owaifeer, M. & Al- Muhaini, M. MILP- based technique for smart self- healing grids. IET Generation, Transmission & Distribution 12, 2307–2316 (2018). + +<|ref|>text<|/ref|><|det|>[[206, 168, 833, 211]]<|/det|> +[12] Botea, A., Rintanen, J. & Banerjee, D. Optimal reconfiguration for supply restoration with informed A\* search. IEEE Transactions on Smart Grid 3, 583–593 (2012). + +<|ref|>text<|/ref|><|det|>[[206, 222, 833, 265]]<|/det|> +[13] Xu, Y., Liu, C.- C., Schneider, K. P., Tuffner, F. K. & Ton, D. T. Microgrids for service restoration to critical load in a resilient distribution system. IEEE Transactions on Smart Grid 9, 426–437 (2016). + +<|ref|>text<|/ref|><|det|>[[206, 276, 833, 319]]<|/det|> +[14] Poudel, S., Dubey, A. & Schneider, K. P. A generalized framework for service restoration in a resilient power distribution system. IEEE Systems Journal 16, 252–263 (2020). + +<|ref|>text<|/ref|><|det|>[[206, 330, 833, 373]]<|/det|> +[15] Bakar, N. N. A., Hassan, M. Y., Sulaima, M. F., Na'im Mohd Nasir, M. & Khamis, A. Microgrid and load shedding scheme during islanded mode: A review. Renewable and Sustainable Energy Reviews 71, 161–169 (2017). + +<|ref|>text<|/ref|><|det|>[[206, 384, 833, 413]]<|/det|> +[16] Liu, H., Chen, X., Yu, K. & Hou, Y. The control and analysis of self- healing urban power grid. IEEE Transactions on smart grid 3, 1119–1129 (2012). + +<|ref|>text<|/ref|><|det|>[[206, 425, 832, 454]]<|/det|> +[17] Farivar, M. & Low, S. H. Branch flow model: Relaxations and convexification—part i. IEEE Transactions on Power Systems 28, 2554–2564 (2013). + +<|ref|>text<|/ref|><|det|>[[206, 466, 833, 509]]<|/det|> +[18] Sekhavatmanesh, H. & Cherkaoui, R. A novel decomposition solution approach for the restoration problem in distribution networks. IEEE Transactions on Power Systems 35, 3810–3824 (2020). + +<|ref|>text<|/ref|><|det|>[[206, 521, 833, 550]]<|/det|> +[19] Shirmohammadi, D. Service restoration in distribution networks via network reconfiguration. IEEE Transactions on Power Delivery 7, 952–958 (1992). + +<|ref|>text<|/ref|><|det|>[[206, 561, 833, 604]]<|/det|> +[20] Zidan, A. & El- Saadany, E. Network reconfiguration in balanced and unbalanced distribution systems with variable load demand for loss reduction and service restoration, 1–8 (IEEE, 2012). + +<|ref|>text<|/ref|><|det|>[[206, 615, 833, 658]]<|/det|> +[21] Rao, R. S., Narasimham, S. V. L., Raju, M. R. & Rao, A. S. Optimal network reconfiguration of large- scale distribution system using harmony search algorithm. IEEE Transactions on power systems 26, 1080–1088 (2010). + +<|ref|>text<|/ref|><|det|>[[206, 669, 833, 712]]<|/det|> +[22] Wu, Y.- K., Lee, C.- Y., Liu, L.- C. & Tsai, S.- H. Study of reconfiguration for the distribution system with distributed generators. IEEE transactions on Power Delivery 25, 1678–1685 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 87, 792, 130]]<|/det|> +[23] Pathan, M. I., Al- Muhaini, M. & Djokic, S. Z. Optimal reconfiguration and supply restoration of distribution networks with hybrid microgrids. Electric Power Systems Research 187, 106458 (2020). + +<|ref|>text<|/ref|><|det|>[[164, 141, 790, 184]]<|/det|> +[24] Sekhavatmanesh, H. & Cherkaoui, R. Analytical approach for active distribution network restoration including optimal voltage regulation. IEEE Transactions on Power Systems 34, 1716- 1728 (2018). + +<|ref|>text<|/ref|><|det|>[[164, 196, 790, 253]]<|/det|> +[25] de Quevedo, P. M., Contreras, J., Rider, M. J. & Allahdadian, J. Contingency assessment and network reconfiguration in distribution grids including wind power and energy storage. IEEE Transactions on Sustainable Energy 6, 1524- 1533 (2015). + +<|ref|>text<|/ref|><|det|>[[164, 263, 790, 307]]<|/det|> +[26] Li, Y., Xiao, J., Chen, C., Tan, Y. & Cao, Y. Service restoration model with mixed- integer second- order cone programming for distribution network with distributed generations. IEEE Transactions on Smart Grid 10, 4138- 4150 (2018). + +<|ref|>text<|/ref|><|det|>[[164, 318, 790, 361]]<|/det|> +[27] Chen, C., Wang, J., Qiu, F. & Zhao, D. Resilient distribution system by microgrids formation after natural disasters. IEEE Transactions on smart grid 7, 958- 966 (2015). + +<|ref|>text<|/ref|><|det|>[[164, 373, 790, 402]]<|/det|> +[28] Wang, F. et al. A multi- stage restoration method for medium- voltage distribution system with dgs. IEEE transactions on smart grid 8, 2627- 2636 (2016). + +<|ref|>text<|/ref|><|det|>[[164, 414, 790, 457]]<|/det|> +[29] Sultana, B., Mustafa, M., Sultana, U. & Bhatti, A. R. Review on reliability improvement and power loss reduction in distribution system via network reconfiguration. Renewable and sustainable energy reviews 66, 297- 310 (2016). + +<|ref|>text<|/ref|><|det|>[[164, 468, 790, 511]]<|/det|> +[30] Cao, D. et al. Reinforcement learning and its applications in modern power and energy systems: A review. Journal of modern power systems and clean energy 8, 1029- 1042 (2020). + +<|ref|>text<|/ref|><|det|>[[164, 523, 790, 552]]<|/det|> +[31] Gao, Y., Shi, J., Wang, W. & Yu, N. Dynamic distribution network reconfiguration using reinforcement learning, 1- 7 (IEEE, 2019). + +<|ref|>text<|/ref|><|det|>[[164, 563, 790, 606]]<|/det|> +[32] Kundacina, O. B., Vidovic, P. M. & Petkovic, M. R. Solving dynamic distribution network reconfiguration using deep reinforcement learning. Electrical Engineering 1- 15 (2021). + +<|ref|>text<|/ref|><|det|>[[164, 618, 790, 647]]<|/det|> +[33] Wang, B., Zhu, H., Xu, H., Bao, Y. & Di, H. Distribution network reconfiguration based on noisynet deep q- learning network. IEEE Access 9, 90358- 90365 (2021). + +<|ref|>text<|/ref|><|det|>[[164, 658, 790, 700]]<|/det|> +[34] Gao, Y., Wang, W., Shi, J. & Yu, N. Batch- constrained reinforcement learning for dynamic distribution network reconfiguration. IEEE Transactions on Smart Grid 11, 5357- 5369 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[205, 86, 833, 116]]<|/det|> +[35] Abdelmalak, M. et al. Network reconfiguration for enhanced operational resilience using reinforcement learning, 1–6 (IEEE, 2022). + +<|ref|>text<|/ref|><|det|>[[205, 126, 833, 170]]<|/det|> +[36] Gautam, M., Abdelmalak, M., MansourLakouraj, M., Benidris, M. & Livani, H. Reconfiguration of distribution networks for resilience enhancement: A deep reinforcement learning- based approach, 1–6 (IEEE, 2022). + +<|ref|>text<|/ref|><|det|>[[205, 181, 831, 210]]<|/det|> +[37] Verma, S. & Zhang, Z. L. Graph capsule convolutional neural networks (2018). 1805.08090. + +<|ref|>text<|/ref|><|det|>[[205, 221, 833, 265]]<|/det|> +[38] Paul, S., Ghassemi, P. & Chowdhury, S. Learning scalable policies over graphs for multi- robot task allocation using capsule attention networks, 8815–8822 (IEEE, 2022). + +<|ref|>text<|/ref|><|det|>[[205, 276, 833, 320]]<|/det|> +[39] Vinayagam, A., Swarna, K., Khoo, S. Y., Maung Than Oo, A. & Stojcevski, A. Pv based microgrid with grid- support grid- forming inverter control- (simulation and analysis) (2017). + +<|ref|>text<|/ref|><|det|>[[205, 330, 833, 360]]<|/det|> +[40] Dugan, R. C. & McDermott, T. Reference guide. The Open Distribution System Simulator (OpenDSS). EPRI (2016). + +<|ref|>text<|/ref|><|det|>[[205, 371, 831, 401]]<|/det|> +[41] Krishnamurthy, D. Opendssdirect. py. Tech. Rep., National Renewable Energy Lab.(NREL), Golden, CO (United States) (2017). + +<|ref|>text<|/ref|><|det|>[[205, 411, 831, 441]]<|/det|> +[42] Schulman, J., Wolski, F., Dhariwal, P., Radford, A. & Klimov, O. Proximal policy optimization algorithms (2017). 1707.06347. + +<|ref|>text<|/ref|><|det|>[[205, 451, 833, 496]]<|/det|> +[43] Paul, S. et al. Efficient planning of multi- robot collective transport using graph reinforcement learning with higher order topological abstraction. arXiv preprint arXiv:2303.08933 (2023). + +<|ref|>text<|/ref|><|det|>[[205, 507, 831, 536]]<|/det|> +[44] Danielsson, A. M. Deep learning for power system restoration. Ph.D. thesis (2018). + +<|ref|>text<|/ref|><|det|>[[205, 547, 833, 576]]<|/det|> +[45] Bush, B., Chen, Y., Ofori- Boateng, D. & Gel, Y. R. Topological machine learning methods for power system responses to contingencies, Vol. 35, 15278–15285 (2021). + +<--- Page Split ---> diff --git a/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/images_list.json b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..241f8cb2c58e1f92ece34844e5664d80f71908a1 --- /dev/null +++ b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Workflow of DiffScan. a Taking raw reactivities as input, DiffScan first normalizes them relative to one another in the Normalization module (b) to correct for systematic bias, and then identifies SVRs in the Scan module (c). Different colors denote reactivity replicates in each condition. b The Normalization module transforms raw reactivities into normalized reactivities to remove systematic bias. The normalized reactivities are comparable as far as possible across different cellular conditions. Red points indicate nucleotide positions in SVRs. c Taking normalized reactivities as input, the Scan module first calculates the significance of any differential signals for each nucleotide position with Wilcoxon test, and then concatenates positional p values into a regional signal via scan statistic. The significance of the scan statistic for each enumerated region is evaluated by Monto Carlo sampling, and those regions crossing a specified significance threshold are reported as SVRs.", + "footnote": [], + "bbox": [ + [ + 143, + 90, + 857, + 312 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Comparison of DiffScan and other SVR detection methods with simulated datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 1 nt, 5 nt, and 11 nt. a Distances between simulated SVRs and the top-ranked 1,000 nucleotide positions in predicted SVRs. b Precision-Recall curves for the prediction results from the various SVR detection methods. Rows: three types of reactivity generative models. Columns: three levels of strength of differential signals at simulated SVRs. Note deltaSHAPE does not allow external thresholding, and therefore it is represented as dots instead of curves.", + "footnote": [], + "bbox": [ + [ + 143, + 90, + 630, + 675 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Comparison of DiffScan and existing SVR detection methods with negative control datasets and benchmark datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 5 nt following the original article of the method. a Position-level false positive rate at a significance level of 0.05 in six negative control datasets (i.e., datasets having no SVRs). b Average distance from the top-10 ranked nucleotide positions in the predicted SVRs to the annotated SVRs in the benchmark datasets. c Precision-Recall curves for the benchmark datasets. deltaSHAPE and PARCEL are represented as dots. \"X\" indicates that the corresponding method was not applicable for the dataset. \"*\" indicates that the corresponding method did not report any region.", + "footnote": [], + "bbox": [ + [ + 147, + 140, + 808, + 570 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Application of DiffScan to explore the roles of RNA structural variation in regulation of mRNA abundance in an icSHAPE dataset mapping RNA structure across human cellular compartments. Ch: chromatin, Np: nucleoplasm, Cy: cytoplasm. a Predicted SVRs between Ch and Np were enriched with protein binding sites and RNA modification sites. \\*p value (Fisher's exact test) \\(< 0.05\\) , \\*\\*\\*p value \\(< 1e-6\\) . b", + "footnote": [], + "bbox": [ + [ + 171, + 245, + 570, + 722 + ] + ], + "page_idx": 19 + } +] \ No newline at end of file diff --git a/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0.mmd b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0.mmd new file mode 100644 index 0000000000000000000000000000000000000000..61d2f610fe08c8724709f1a8128ff533a9c6e079 --- /dev/null +++ b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0.mmd @@ -0,0 +1,349 @@ + +# Differential Analysis of RNA Structure Probing Experiments at Nucleotide Resolution: Uncovering Regulatory Functions of RNA Structure + +Bo Yu Tsinghua University Pan Li Tsinghua University Qiangfeng Zhang School of Life Sciences, Tsinghua University https://orcid.org/0000- 0002- 4913- 0338 Lin Hou ( houl@tsinghua.edu.cn ) Tsinghua University https://orcid.org/0000- 0002- 4283- 8501 + +## Article + +Keywords: RNA structure, structure probing experiments, differential analysis, RNA secondary structure + +Posted Date: August 26th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 800283/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on July 22nd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 31875- 3. + +<--- Page Split ---> + +# Differential Analysis of RNA Structure Probing Experiments at Nucleotide Resolution: + +# Uncovering Regulatory Functions of RNA Structure + +Bo Yu \(^{1}\) , Pan Li \(^{2,3}\) , Qiangfeng Cliff Zhang \(^{2,3*}\) , Lin Hou \(^{1,2*}\) + +\(^{1}\) Center for Statistical Science, Department of Industrial Engineering, Tsinghua University, Beijing 100084, China + +\(^{2}\) MOE Key Laboratory of Bioinformatics, Tsinghua University, Beijing 100084, China + +\(^{3}\) Center for Synthetic and Systems Biology, Beijing Advanced Innovation Center for Structural Biology, Tsinghua- Peking Joint Center for Life Sciences, School of Life Sciences, Tsinghua University, Beijing 100084, China + +\* Correspondence: houl@tsinghua.edu.cn, qczhang@tsinghua.edu.cn + +<--- Page Split ---> + +## Abstract + +RNAs perform their function by forming specific structures, which can change across cellular conditions. Structure probing experiments combined with next generation sequencing technology have enabled transcriptome- wide analysis of RNA secondary structure in various cellular conditions. Differential analysis of structure probing data in different conditions can reveal the RNA structurally variable regions (SVRs), which is important for understanding RNA functions. Here, we propose DiffScan, a computational framework for normalization and differential analysis of structure probing data in high resolution. DiffScan preprocesses structure probing datasets to remove systematic bias, and then scans the transcripts to identify SVRs and adaptively determines their lengths and locations. The proposed approach is compatible with most structure probing platforms (e.g., icSHAPE, DMS- seq). When evaluated with simulated and benchmark datasets, DiffScan identifies structurally variable regions at nucleotide resolution, with substantial improvement in accuracy compared with existing SVR detection methods. Moreover, the improvement is robust when tested in multiple structure probing platforms. Application of DiffScan in a dataset of multi- subcellular RNA structureome identified multiple regions that form different structures in nucleus and cytoplasm, linking RNA structural variation to regulation of mRNAs encoding mitochondria- associated proteins. This work provides an effective tool for differential analysis of RNA secondary structure, reinforcing the power of structure probing experiments in deciphering the dynamic RNA structureome. + +<--- Page Split ---> + +## Introduction + +RNA molecules play important roles in myriad cellular processes by forming specific structures \(^{1 - 3}\) . Deciphering RNA structure is informative for understanding RNA functions. In recent years, diverse structure probing (SP) methods have been developed to study RNA secondary structure in various cellular contexts, which utilize chemicals that react differentially to nucleotides according to their pairing status \(^{4,5}\) . Coupled with next generation sequencing technologies, SP experiments can be performed at high throughput, providing a transcriptome level view of RNA secondary structure \(^{4 - 6}\) , i.e., RNA structure. There are various mature high- throughput SP platforms, such as SHAPE- Seq \(^{7,8}\) , DMS- Seq \(^{9}\) , and icSHAPE \(^{10}\) . These platforms offer flexible options for tackling different biological problems, and have achieved success in uncovering pervasive links between RNA structure and RNA function \(^{11}\) . + +Studies have examined how the structures of particular RNA molecules change across multiple cellular conditions, revealing explicit connections between RNA structure and RNA function \(^{12}\) . For example, the ubiquitous yybP- ykoY motif has been shown to adopt distinct structures in response to manganese ions, thus exerting regulatory consequences on protein translation in both Escherichia coli and Bacillus subtilis \(^{13}\) . RNA structural variations have also been reported to regulate the binding of trans- acting factors and RNA stability in multiple organisms, including human, yeast, and zebrafish \(^{14 - 18}\) . + +To explore the SP experiments to uncover dynamic RNA structures, quantitative tools that contrast SP experiments to identify structurally variable regions (SVRs) are in great demand. Several methods have been proposed to identify SVRs. PARCEL \(^{19}\) and RASA \(^{20}\) directly model and compare raw read counts at each nucleotide position, and then identify regions enriched for position- level signals. However, their models are tailored for specific experimental protocols, and + +<--- Page Split ---> + +it is not straightforward to extend them to accommodate more emerging SP techniques. StrucDiff \(^{21}\) , deltaSHAPE \(^{22}\) , and dStruct \(^{23}\) take reactivities as input, which are estimated from raw read counts to summarize the pairing status at each nucleotide position. They search for SVRs with pre- specified search lengths to aggregate differential signals \(^{21 - 23}\) . However, the choice of search length is arbitrary. + +Despite the success of existing methods, differential analysis of SP data remains challenging in many aspects. First, SVRs manifest great variation in length, ranging from a few to several dozens of nucleotide positions \(^{19, 23, 24}\) . As a result, searching with fixed search length can lead to insufficient detection power and inaccurate boundary mapping, when the prespecified search length deviates greatly from the true length. Second, SP platforms differ in utilization of additives, specific experimental protocols, and preprocessing pipelines, yielding distinct data types and distributions in output \(^{4, 5, 25}\) . As a result, it is desirable but usually not easy to extend methods developed for one platform to another. Third, SP data can be confounded by systematic bias \(^{4, 5}\) , which should be removed from reactivities of the two compared conditions prior to differential analysis \(^{26}\) . However, to our knowledge, normalization techniques that are compatible with multiple SP platforms are still lacking. Finally, rigorous error control is also highly impactful on the accuracy and biological relevance of output from SVR detection \(^{23}\) , especially for transcriptome- scale analyses. + +To address the unmet needs, we propose DiffScan, a computational framework for differential analysis of SP data at nucleotide resolution. DiffScan normalizes SP reactivities via a built- in Normalization module, which is compatible with various platforms, and then looks for SVRs via a Scan module. The Scan module locates SVRs at nucleotide resolution, and rigorously controls family- wise error rate. We demonstrated with large- scale simulated datasets and benchmark datasets that DiffScan achieves the best power and accuracy in SVR detection across + +<--- Page Split ---> + +various platforms, compared to state- of- art methods, We applied DiffScan to a recent icSHAPE dataset of RNA structurome in multiple subcellular compartments (e.g., nucleoplasm and cytoplasm), and revealed the potential roles of SVRs in regulating mRNAs encoding mitochondria- associated proteins. + +## Results + +## Model description + +DiffScan comprises a Normalization module and a Scan module (Fig. 1a). SP reactivities of RNA transcripts for two conditions (e.g., in vitro, in vivo) are taken as input, and the reactivities are normalized relative to one another in the Normalization module (Fig. 1b) to correct for systematic bias. The corrected reactivities are comparable as far as possible across different cellular conditions. In the Scan module (Fig. 1c), structural variations are initially evaluated at each nucleotide position, and then the algorithm scans through the transcripts with variable search lengths to identify SVRs. + +## Normalization + +It is known that the reactivities of particular RNA nucleotide positions can be affected by a variety of confounding factors in SP experiments26, including transcript abundance, sequencing depth, and signal- to- noise ratio (see Methods). Owing to discrepancies in these factors, the SP reactivities obtained from two conditions lacking substantial biological differences may show substantial differences27 (see Supplementary Fig. 1 for an example). The unwanted variations will then persist throughout the routine preprocessing steps. Thus, in practice careful adjustment between and within conditions is essential when comparing SP data obtained from samples in + +<--- Page Split ---> + +distinct conditions. Related challenges have been widely reported for other high- throughput sequencing experiments, such as ChIP- Seq28 and ATAC- Seq29. Inspired by normalization approaches for other high- throughput sequencing based methods, we propose the following normalization procedure for differential analysis of SP data. + +To account for a potentially wide range of reactivity levels, we first rescale the reactivity values into the interval [0,1] after a \(90\%\) winsorization (see Methods). Next, the reactivities of within- condition replicates, if available, are processed by quantile normalization30, so that the quantiles of each replicate are matched within conditions. We subsequently normalize between- condition reactivities using an approach similar to MAnorm28. Briefly, a structurally invariant set of nucleotide positions \(S\) is determined as a "pivot" for normalization, and the reactivities are transformed so that the transformed reactivities are at the same level for the pivot set between conditions. The basic idea is to learn transformation rules from \(S\) and then extrapolate the learned transformation to all nucleotide positions of the transcripts. In particular, the transformation is learned from training a linear model from \(S\) , which takes reactivities in one condition as response and reactivities in the other condition as predictor. The adjusted reactivities from the Normalization module are then ready for differential analysis in the Scan module (Supplementary Fig. 2). We show in theory that the transformation explicitly corrects for differences in sequencing depth and signal- to- noise ratio. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Workflow of DiffScan. a Taking raw reactivities as input, DiffScan first normalizes them relative to one another in the Normalization module (b) to correct for systematic bias, and then identifies SVRs in the Scan module (c). Different colors denote reactivity replicates in each condition. b The Normalization module transforms raw reactivities into normalized reactivities to remove systematic bias. The normalized reactivities are comparable as far as possible across different cellular conditions. Red points indicate nucleotide positions in SVRs. c Taking normalized reactivities as input, the Scan module first calculates the significance of any differential signals for each nucleotide position with Wilcoxon test, and then concatenates positional p values into a regional signal via scan statistic. The significance of the scan statistic for each enumerated region is evaluated by Monto Carlo sampling, and those regions crossing a specified significance threshold are reported as SVRs.
+ +## Scan + +Formally, "scan statistics" refer to statistical methods for cluster detection in time and space31, 32. These methods can accurately map regional signals and control family- wise error rates during multiple testing of interrelated hypotheses. Scan statistics have been successfully applied to many areas including molecular biology33 and human genetics34. The Scan module of DiffScan was developed with the goal to identify SVRs at nucleotide resolution in a data- adaptive fashion. In + +<--- Page Split ---> + +brief, the Scan module slides through the transcripts to enumerate overlapping regions of different lengths, identifies regions with maximal differential signals, and evaluates their statistical significance. In detail, we first quantify positional differential signals by calculating a p value for each nucleotide position by Wilcoxon test, which contrasts reactivities between conditions in a small window surrounding the nucleotide position. Note that the Normalization module does not enforce any specific distributions of the normalized reactivities (Supplementary Fig. 3), and the nonparametric test we use guarantees robust evaluation of differential signals for various SP platforms. + +Second, for each region R, we propose the following scan statistic to quantify differential signals for regions, + +\[\mathrm{Q}(\mathrm{R}) = \frac{-\Sigma_{\mathrm{j}\in \mathrm{R}}\log(\mathrm{p}_{\mathrm{j}})}{\sqrt{|\mathrm{R}|}}.\] + +The sum in the numerator aggregates positional differential signals, while the denominator penalizes the extension of a candidate region. Intuitively, regions that are enriched with structurally variable nucleotides will obtain high Q(R) scores. + +We search in the transcripts with sliding windows of different lengths, and calculate the scan statistic for each candidate region (Fig. 1c). A Monte Carlo approach is then implemented to evaluate the statistical significance of the scan statistics, which addresses the multiple testing problem for the overlapping regions by controlling the family- wise error rate (See Supplementary Methods). Accordingly, we can identify SVRs with accurately mapped boundaries based on the significance of the scan statistics (Fig. 1c). + +# Validation of DiffScan using simulated SP datasets + +<--- Page Split ---> + +We simulated RNA secondary structures in two conditions, and generated SP reactivities based on the simulated secondary structures in each condition using three types of empirical models representing different SP platforms, including two types of SHAPE reactivity25, 35, 36 and one type of icSHAPE reactivity17 (see Supplementary Methods). Biologically, it is often the case that a particular RNA molecule will occur as a mixture of structural conformations in a given cellular condition37, 38, so SVRs detected between conditions represent altered proportions comprising these mixtures of structural conformations. Thus, the between- condition differences in RNA secondary structures can be very subtle in reality. To mimic the complexity of the landscape of RNA secondary structures, we sampled multiple structural conformations from an ensemble of energy- function- based predictions, and mixed them in varying proportions in the simulated datasets (see Methods). The reactivities were thereafter simulated based on the pairing status of each nucleotide. + +In detail, we sampled 10 conformations for each of the 100 RNA transcripts we randomly selected from transcripts in human embryonic kidney (HEK293) cells17. The lengths of the RNA transcripts ranged between 61 nt and 4,810 nt, and the simulated SVRs covered 3% to 55% of nucleotide positions of the RNA transcripts, with lengths between 1 nt and 16 nt. These data included a total of 13,162 simulated SVRs, with 6,587 being single nucleotide structural variations, 6,081 having lengths between 1 nt and 5 nt, and 494 SVRs with lengths greater than 5 nt (see Supplementary Fig. 4 for further details). Note that these simulated data echo the real world knowledge that the lengths of SVRs vary extensively13, 19, 23, 24, 39. We varied the strength of differential signals of simulated SVRs between “high”, “medium”, and “low”. Combining different levels of signal strength and the three types of generative models of reactivity, we have in total 9 simulation settings. + +<--- Page Split ---> + +We compared the performance of DiffScan with two other SVR detection methods, deltaSHAPE and dStruct. Note that DiffScan adaptively determines the lengths of SVRs from data, whereas search length is a prespecified parameter for deltaSHAPE and dStruct. We set the search length to 5 nt for deltaSHAPE (the default setting suggested by the authors \(^{22}\) ) and to three different values (1 nt, 5 nt, and 11 nt) for dStruct. + +To evaluate the accuracy of SVR boundaries mapping, we calculated the distance between each nucleotide position in the predicted SVRs and the true SVRs. In the ideal case, if a predicted SVR sits within a simulated SVR, the distance for each nucleotide position in the predicted SVR is zero. Oppositely, if a predicted SVR is off- target, i.e., containing many nucleotide positions that are neither in or close to the simulated SVRs, the distances for such nucleotide positions in the predicted SVR will be large. + +The top- ranked SVRs identified by DiffScan are closer to the simulated SVRs than those by deltaSHAPE or dStruct in all 9 settings (Fig. 2a), demonstrating the superior performance of DiffScan in accurate identification of SVRs. Given that dStruct only considers nonoverlapping regions and relies on a pre- specified search length, and deltaSHAPE considers regions of fixed length, their performance is sensitive to whether or not pre- specified value of search length fits the actual lengths of the SVRs in a dataset. In contrast, DiffScan determines boundaries of SVRs by optimizing its scan statistic, thereby effectively distinguishing SVRs amongst overlapping regions of different lengths, leading to a finer level of granularity relative to deltaSHAPE and dStruct. Moreover, the superiority of DiffScan is consistent across different reactivity models and signal levels. + +We also evaluated the precision and recall rate at the nucleotide position level (see Methods) of the SVR detection methods. Among all three methods, DiffScan achieves the best + +<--- Page Split ---> + +precision at the same recall rate (Fig. 2b). Using optimization based on its scan statistic, DiffScan effectively identifies nonconsecutive SVR segments with nucleotide resolution. Although dStruct also sensitively identifies SVRs, it tends to output long, contiguous regions, which include a substantial proportion of nucleotide positions without structural variations, hampering the precision. The output of deltaSHAPE is a fixed set of regions that the threshold is internally decided, and it is represented as dots rather than curves in Fig. 2b. Although deltaSHAPE works well for the SHAPE reactivity model from Cordero et al. \(^{35}\) , which is reasonable since the method was originally developed for SHAPE- MaP experiments, its precision is much lower than dStruct and DiffScan for other types of reactivity models. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 Comparison of DiffScan and other SVR detection methods with simulated datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 1 nt, 5 nt, and 11 nt. a Distances between simulated SVRs and the top-ranked 1,000 nucleotide positions in predicted SVRs. b Precision-Recall curves for the prediction results from the various SVR detection methods. Rows: three types of reactivity generative models. Columns: three levels of strength of differential signals at simulated SVRs. Note deltaSHAPE does not allow external thresholding, and therefore it is represented as dots instead of curves.
+ +<--- Page Split ---> + +## Sensitivity analysis + +There are two tuning parameters in DiffScan: the radius of local windows (denoted by r) for smoothed calculation of positional differential signal, and the parameter associated with the penalty term (denoted by y) in the scan statistic. In the above analysis of the simulated datasets, we set r to 2 nt and y to 0.5 as their default values. We also conducted a sensitivity analysis wherein we tested DiffScan with the simulated datasets using eight other settings of \((r, \gamma)\) . In each setting, we calculated the mean precision of the predicted SVRs at recall values below 0.05. The coefficient of variation for the mean precisions corresponding to different tuning parameters is below 0.15 for all of the examined scenarios in the simulations (Supplementary Table 1), showing the robustness of DiffScan regarding to the tuning parameters. Notably, DiffScan consistently outperformed dStruct and deltaSHAPE in these tested parameter settings (Supplementary Fig. 5). + +## Model validation with negative control datasets + +To assess the extent of false positive discoveries of DiffScan and other SVR detection methods, we constructed six negative control datasets (Control 1- 6, see Methods) from multiple SP platforms, including SHAPE- Seq and icSHAPE. In these datasets, no SVRs are present between the compared conditions. Thus, any significant SVRs identified by the detection methods would represent false positives. These comparisons included the PARCEL and RASA methods for the cases where raw read counts are available (Control 1, 2, 5, and 6). As dStruct requires within- condition replicates, it is not applicable for Control 5 and 6. + +<--- Page Split ---> + +The false positive rates for DiffScan, dStruct, and PARCEL were below 0.05 for all tested datasets (Fig. 3a). It bears emphasis that the test space of DiffScan is tremendously larger than the other methods because it contains overlapping regions of differing lengths. Nevertheless, its false positive rates were well controlled. The error rate of RASA is slightly inflated in Control 2, but is well- controlled for the other datasets. Consistent inflation is observed for deltaSHAPE across different datasets, reporting false positive results covering \(\sim 15\%\) of nucleotide positions in the datasets. The fact that neither deltaSHAPE nor RASA control for the multiple testing problem can likely explain the observed inflation. + +## Evaluation using benchmark datasets + +We used two benchmark datasets with explicitly annotated SVRs between the compared conditions, curated by Choudhary et al. \(^{23}\) to support evaluation of different SVR detection methods. The Flu dataset \(^{24}\) is from a SHAPE- Seq experiment that measured the secondary structure of a Bacillus cereus crcB fluoride riboswitch in vitro in the presence or absence of fluoride. The transcript (100 nt in length) has five annotated SVRs, ranging in length between 1 nt and 8 nt. The RRE dataset \(^{39}\) is from a SHAPE- Seq experiment that measured the secondary structure of the HIV Rev- response element in the presence or absence of the Rev protein. There are seven annotated SVRs for Rev- RRE interactions in the transcript (369 nt in length), with lengths ranging between 7 nt and 39 nt. + +We evaluated the average distance between predicted SVRs and the annotated SVRs and Precision- Recall curves output by DiffScan, deltaSHAPE, PARCEL, RASA, and dStruct (Fig. 3b- c). Consistent with our results when evaluating using simulated datasets and showcasing DiffScan's superior performance for accurately locating SVRs, DiffScan achieved the lowest average distance + +<--- Page Split ---> + +from the annotated SVRs of the benchmark datasets. For the Flu dataset, DiffScan identified all the five annotated SVRs with their boundaries accurately mapped (Supplementary Fig. 6). In contrast, PARCEL predicted a long region, covering \(72\%\) of all nucleotide positions in the transcript, that spanned all five annotated SVRs. However, it did not distinguish annotated SVRs from flanking nucleotides between SVRs and falsely entailed many nucleotide positions ( \(67\%\) of the nucleotide positions in the predicted SVRs) without structural variations. Similarly, dStruct identified a long region covering \(37\%\) of all nucleotide positions, which covered two annotated SVRs, while \(62\%\) of the nucleotide positions in the predicted SVRs had no structural variation. RASA did not report any significant region for the Flu dataset. deltaSHAPE reported six regions, missing an annotated SVR while containing a false positive discovery. For the RRE dataset, DiffScan detected three regions which are located within three annotated SVRs. PARCEL and RASA were not applicable for the RRE dataset because raw count data were not available. deltaSHAPE and dStruct both reported four regions, which overlapped with two and four annotated SVRs respectively. Compared to DiffScan, they reported regions with much larger average distance to annotated SVRs, since they included many nucleotide positions ( \(53\%\) and \(35\%\) of the nucleotide positions in the predicted SVRs for deltaSHAPE and dStruct) without structural variation. + +In terms of precision and recall measures, DiffScan also achieved the best area under the curve (AUC) values with both datasets (Fig. 3c). Note that because deltaSHAPE and PARCEL do not allow threshold tunings, so they are represented as dots rather than curves in Fig. 3c. deltaSHAPE performed comparably to DiffScan for the Flu dataset. However, the precision of deltaSHAPE was much lower than DiffScan and dStruct for the RRE dataset, since the predicted SVRs of deltaSHAPE included many nucleotide positions without structural variation. In general, dStruct had lower precision compared to DiffScan when the recall rate was set to the same level, consistent to the trends we observed when analyzing the simulated datasets. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Comparison of DiffScan and existing SVR detection methods with negative control datasets and benchmark datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 5 nt following the original article of the method. a Position-level false positive rate at a significance level of 0.05 in six negative control datasets (i.e., datasets having no SVRs). b Average distance from the top-10 ranked nucleotide positions in the predicted SVRs to the annotated SVRs in the benchmark datasets. c Precision-Recall curves for the benchmark datasets. deltaSHAPE and PARCEL are represented as dots. "X" indicates that the corresponding method was not applicable for the dataset. "*" indicates that the corresponding method did not report any region.
+ +<--- Page Split ---> + +## Roles of SVRs in regulating mRNA transcripts for mitochondria-associated proteins + +We applied DiffScan to a recently reported icSHAPE dataset that mapped RNA secondary structure across human cellular compartments covering chromatin (Ch), nucleoplasm (Np), and cytoplasm (Cy)17. The DiffScan- predicted SVRs for the Ch versus Np (Fig. 4a) and for the Np versus Cy (Supplementary Fig. 7) comparisons mostly involved protein binding sites and RNA modification sites. As we also had data for RNA abundance in the Ch, Np, and Cy samples, we were interested in the potential impacts of SVRs on regulating mRNA abundance. Comparison of mRNA abundance in the Np versus Cy samples clearly placed the mRNAs into two subgroups, with mRNA levels in subgroup II decreased sharply in the Cy fraction (Fig. 4b). A manual check revealed that all 12 mRNAs of subgroup II encode mitochondria- associated proteins. In addition, DiffScan identified SVRs in all of these 12 transcripts (p value = 8.4e- 3), suggesting an association between RNA structural variation and the abundance of mRNAs encoding mitochondria- associated proteins. + +To further investigate this association, we used the FIMO module from the MEME suite40 to search for RBPs with binding motifs enriched in the predicted SVRs from the Np versus Cy comparison (see Methods). Three enriched RBPs were identified, including quaking (QKI), insulin- like growth factor 2 mRNA- binding protein 3 (IGF2BP3), and polyadenylate- binding protein 1 (PABPC1). QKI had two significantly enriched motifs, ACUAACA (corrected p value = 1.4e- 5) and AUCUAAC (corrected p value = 7.8e- 4). QKI predominantly localizes in the Np41, 42, and it has been reported to regulate various biological processes such as myelinization43, mRNA stability, and mRNA export42. Of particular note, a recent study reported that QKI binds mRNAs encoding mitochondrial proteins and transcriptional coactivators known to regulate mitochondrial biogenesis and function: QKI decreases the stability and nuclear export of these RNAs44. Therefore, the binding of QKI to the mRNA transcripts of subgroup II, which all encode mitochondria- associated proteins, may in part explain their reduced abundance in the Cy samples. + +<--- Page Split ---> + +Considering the biological evidence and the potential roles of SVRs, we enumerated four possible causal models to describe the relationships among SVRs, QKI binding, and changes in mRNA abundance (Model 1- 4 in Supplementary Fig. 8; see Methods). In Model 1, QKI binding causes SVRs in the mRNA transcripts of subgroup II and also regulates their abundance in a separate pathway, and thus SVRs have no contribution to the observed decline of mRNA abundance in the Cy. In Model 2- 4, SVRs either directly or indirectly (mediated by QKI binding) regulate mRNA abundance. The observed data rejects Model 1 and Model 2 (p value = 2.5e- 91), and therefore supports the direct roles of SVRs in regulation of mRNA abundance. However, Model 3 and 4 are statistically equivalent, i.e., not distinguishable. In another word, SVRs have direct effect in regulating mRNA abundance, in addition to QKI binding (Fig. 4c); however, the causal relationships between SVR and QKI binding cannot be determined with current data. + +The second RBP, IGF2BP3, had one motif (ACAAACA) enriched among the predicted SVRs (corrected p value = 5.4e- 3). IGF2BP3 is almost exclusively cytoplasmic, and has been conceptualized to "cage" its target mRNAs into cytoplasmic protein- RNA complexes as they enter the Cy45. Recently, the upregulation of IGF2BP3 has been associated with mitochondrial dysfunction46. We reason that IGF2BP3 cages mRNA transcripts encoding mitochondria- associated proteins as they enter the Cy, accounting for the observed decline in the abundance of such mRNA transcripts. We enumerated possible causal models (Supplementary Fig. 8) and performed hypothesis testing to explore the relationships among SVRs, IGF2BP3 binding, and changes in mRNA abundance. According to our inference, there is a direct effect from the detected SVRs to the changes in mRNA abundance (Fig. 4c; p value = 7.4e- 49), although it is not clear whether the SVRs' regulation of mRNA abundance is also mediated by IGF2BP3 binding. + +<--- Page Split ---> + +In summary, the DiffScan- predicted SVRs provide insight into the functional roles of RNA secondary structure and structural variation in regulating mRNAs encoding mitochondria- associated proteins. + +![](images/Figure_4.jpg) + +
Fig. 4 Application of DiffScan to explore the roles of RNA structural variation in regulation of mRNA abundance in an icSHAPE dataset mapping RNA structure across human cellular compartments. Ch: chromatin, Np: nucleoplasm, Cy: cytoplasm. a Predicted SVRs between Ch and Np were enriched with protein binding sites and RNA modification sites. \*p value (Fisher's exact test) \(< 0.05\) , \*\*\*p value \(< 1e-6\) . b
+ +<--- Page Split ---> + +RPKM of transcripts and the prediction results of DiffScan for Np versus Cy. c Statistical inference supports the direct roles of SVRs in regulation of mRNA abundance. + +## Discussion + +We have developed DiffScan, a computational framework for differential analysis of RNA secondary structure measured in multiple SP platforms. Our method provides several advantages. First, it adaptively estimates the lengths and locations of SVRs with single- nucleotide resolution. Compared to existing SVR detection methods, predicted SVRs by DiffScan are closer to true SVRs, both from simulated and benchmark datasets. Second, DiffScan is compatible with multiple SP platforms with robust performance. Existing SVR detection methods are usually tailored for a specific SP platform, and their generalization to other SP platforms can be prohibitive. Our method flexibly accommodates multiple SP platforms through its Normalization module and using the nonparametric test we implemented. This enables flexible analysis that is adaptable to suitable data types reflecting particular biological problems of interest. For example, the SHAPE- Seq platform with fast- acting reagent is suitable for in vitro studies of RNA folding dynamics47, while the icSHAPE platform utilizing slow- acting reagent allows in vivo transcriptome- wide structure probing16, 17. As demonstrated with simulated and benchmark datasets, the superior performance of DiffScan in terms of statistical power and accuracy for identifying SVRs is robust across different SP platforms. Third, DiffScan rigorously controls for the family- wise error rate in SVR detection, which is particularly influential for transcriptome level analyses. At last, the SVR simulation framework we devised aptly reflects the complexity of SVR detection problems in the real world. We have developed a “simulation” module in our software, which can facilitate evaluation of more upcoming SVR detection methods in the future. + +<--- Page Split ---> + +DiffScan has several limitations. First, the Normalization module in DiffScan relies on a structurally invariant set of nucleotide positions, which is identified via a built- in data- driven strategy. The strategy would fail in extreme cases when almost all nucleotide positions in the studied transcripts exhibit structural variations, although we expect it is rarely the case in practice4, 16, 17, 48. In addition, when the structurally invariant set can be specified by prior knowledge, the normalization step can be easily adapted. Second, we recognize the statistical power of DiffScan is not optimized for a particular SP platform, so when the distribution of reactivities is known \(a\) priori, use of an appropriate parametric test would yield increased power. Third, we note that the ultimate goal of studying SVRs is to understand the functional roles of RNA structure, which often involves cross- examination of other data sources, including motif analysis of RBPs, multi- omics datasets, etc. Incorporation of these multi- source data may help to accurately annotate SVRs11. + +## Acknowledgements + +This work was supported by the National Natural Science Foundation of China (Grants No. 91740204 and 91940306) and the State Key Research Development Program of China (Grant No. 2019YFA0110002) to Q.C.Z. LH acknowledges research support from the National Science Foundation of China (Grant No. 12071243). + +## Author contributions + +B.Y., Q.C.Z., and L.H. conceived the study and wrote the manuscript. B.Y. and L.H. developed the computational framework. B.Y. and P.L. performed the data preparation and statistical analysis. B.Y., Q.C.Z., and L.H. interpreted the results. All authors approved the manuscript. + +<--- Page Split ---> + +## Competing interests + +The authors declare no competing interests. + +## Methods + +## Structure Probing Datasets + +We evaluated the power of DiffScan and other SVR detection methods in two SHAPE- seq datasets, of which the SVRs were curated in Choudhary et al.23. The Flu dataset probed the Bacillus cereus crcB fluoride riboswitch in vitro with and without fluoride ions, with four replicates for each condition. The transcript is 100 nt in length, and nucleotide positions 12- 17, 22- 27, 38- 40, 48 and 67- 74 were identified as SVRs between conditions. The RRE dataset studied the HIV Rev- response element, with three replicates measured in the presence and absence of Rev. The Rev- RRE interaction sites are considered SVRs. We obtained raw read counts and reactivities for the Flu dataset and reactivities for the RRE dataset, based on data availability. + +Synthetic negative control datasets were constructed from real SP data to assess the specificity of DiffScan and existing methods. For construction, we randomly split replicates in a single condition into two groups, and identified SVRs by contrasting the two groups. Any reported SVRs were considered false positive results. In particular, control datasets 1- 4 were obtained by randomly splitting samples in the Flu dataset in the absence (Control 1) and presence (Control 2) of fluoride, and the RRE dataset with (Control 3) and without Rev (Control 4). To include more SP platforms in evaluation, we constructed another two control datasets from an icSHAPE dataset of + +<--- Page Split ---> + +mouse SRP RNA (272 nt) \(^{10}\) , by contrasting the two in vitro samples (Control 5) and the two in vivo samples (Control 6). + +To apply DiffScan in transcriptome level analysis, we downloaded the icSHAPE dataset \(^{17}\) in Sun et al., which mapped RNA secondary structure across HEK293 cellular compartments including chromatin, nucleoplasm, and cytoplasm. Transcripts with RPKM less than 5 or RT stop less than 2 for all nucleotide positions were excluded \(^{17,49}\) . Then nucleotide positions with coverage less than 200 were removed. As a result, 1,352 transcripts (covering 826,917 nucleotide positions) were compared between chromatin and nucleoplasm, and 1,851 transcripts (covering 1,220,273 nucleotide positions) were compared between nucleoplasm and cytoplasm. + +## Normalization module + +Suppose that there are \(\mathsf{n}_{\mathrm{A}}\) replicates from condition A, and the reactivity of nucleotide position j in replicate i is \(\mathrm{r}_{\mathrm{ij}}^{\mathrm{A}},1\leq \mathrm{i}\leq \mathrm{n}_{\mathrm{A}},1\leq \mathrm{j}\leq \mathrm{n}\) , where n is the length of the transcript. Similarly, for \(\mathsf{n}_{\mathrm{B}}\) replicates from condition B, the reactivity of nucleotide position j in replicate k is \(\mathrm{r}_{\mathrm{kj}}^{\mathrm{B}},1\leq \mathrm{k}\leq \mathrm{n}_{\mathrm{B}},1\leq \mathrm{j}\leq \mathrm{n}\) . We normalized reactivities using the following steps. + +Step 1: \(90\%\) winsorization is applied to the reactivities in each replicate separately to remove outliers, i.e., the bottom \(5\%\) of reactivities are set to the 5th percentile while the upper \(5\%\) of reactivities are set to the 95th percentile. After that, reactivities are scaled into range [0,1] by subtracting the minimum and then dividing the result by the new maximum. + +Step 2: Quantile normalization is applied to within- condition replicates to match the quantiles, which is frequently used in normalization analysis of genome- wide assays \(^{30,50}\) . + +Step 3: To make between- condition reactivities comparable, we first determine a structurally invariant set \(S\) , i.e., nucleotide positions that do not exhibit structural variation across conditions. + +<--- Page Split ---> + +The reactivities are normalized so that the adjusted reactivities are similar between conditions in the structurally invariant set. The determination of \(S\) is described in Supplementary Methods. Then we learn a linear transformation rule from \(S\) , which converts reactivities of one condition to the same level of those of the other condition in \(S\) , and extrapolate the transformation to all nucleotide positions to obtain normalized reactivities. + +We learn how to transform between- condition reactivities to the same level in \(S\) by fitting the following robust regression utilizing iterated re- weighted least squares with Huber's \(M\) estimate \(^{51}\) . (For the sake of simplicity, we will still use \(r_{ij}^{\mathrm{A}}, r_{ij}^{\mathrm{B}}\) to denote reactivities processed by Step 1 and Step 2. ) + +\[\log \overline{r_{j}^{\mathrm{A}}}\sim \log \alpha +\beta \log \overline{r_{j}^{\mathrm{B}}},j\in S,\] + +where \(\log \alpha\) corrects for the difference in sequencing depth and \(\beta\) corrects for the difference in signal- to- noise ratio. Given that within- condition replicates are normalized by quantile normalization, it follows that \(\overline{r_{ij}^{\mathrm{A}}}\approx \overline{r_{kj}^{\mathrm{B}}}\) in \(S\) if we transform \(r_{kj}^{\mathrm{B}}\) into \(\overline{r_{kj}^{\mathrm{B}}} = \exp \left(\overline{\log\alpha} +\beta \log r_{kj}^{\mathrm{B}}\right)\) and let \(\overline{r_{ij}^{\mathrm{A}}} = r_{ij}^{\mathrm{A}},j\in S\) . Based on this, we extrapolate the fitted model to all nucleotide positions to obtain normalized reactivities \(\overline{r_{ij}^{\mathrm{A}}} = r_{ij}^{\mathrm{A}}\) and \(\overline{r_{kj}^{\mathrm{B}}} = \exp \left(\overline{\log\alpha} +\beta \log r_{kj}^{\mathrm{B}}\right),1\leq j\leq n\) . + +## Differential Analysis at the Position Level + +With normalized reactivities as input, we first test position- level differences between conditions to derive positional p values. To encompass various SP platforms, a nonparametric Wilcoxon test is used, which evaluates the structural variation at nucleotide position \(j\) by contrasting reactivities between conditions in a small window surrounding \(j\) . Technically, we define + +\[\mathrm{p}_{\mathrm{j}} = \mathrm{Wilcoxon~test}\left(\{\overline{r_{\mathrm{it}}^{\mathrm{A}}},1\leq \mathrm{i}\leq \mathrm{n}_{\mathrm{A}},\mathrm{t}\in \mathrm{C}_{\mathrm{j}}\} ,\{\overline{r_{\mathrm{kt}}^{\mathrm{B}}},1\leq \mathrm{k}\leq \mathrm{n}_{\mathrm{B}},\mathrm{t}\in \mathrm{C}_{\mathrm{j}}\} \right),\] + +<--- Page Split ---> + +where \(\overrightarrow{\mathrm{r}}_{\mathrm{it}}^{\mathrm{A}},\overrightarrow{\mathrm{r}}_{\mathrm{kt}}^{\mathrm{B}}\) are normalized reactivities, \(\mathrm{C_j}\) is a small window centering at nucleotide position j with radius r, and \(\mathsf{p}_{\mathrm{j}}\) is the two- sided p value. By default, we set r to 2 nt. + +## Summarizing Position-level Signals via Scan Statistics + +To concatenate position- level signals into regional signals, a scan statistic is defined for each region R, + +\[\mathrm{Q}(\mathrm{R}) = \frac{-\Sigma_{\mathrm{j}\in \mathrm{R}}\log(\mathrm{p}_{\mathrm{j}})}{\sqrt{|\mathrm{R}|}}.\] + +With penalty of region length \(|\mathrm{R}|\) , extending a candidate region will accumulate position- level signals at the expense of penalization. As a result, SVRs with accurately mapped boundaries can be distinguished based on the magnitude of the scan statistic. + +## Statistical Inference for Detection of SVRs + +In view of the dynamic locations and lengths of SVRs, we scan the transcripts by enumerating all candidate regions with suitable lengths, i.e., every region in \(\mathcal{R} = \{\mathrm{R} | \mathrm{L}_{\mathrm{min}} \leq |\mathrm{R}| \leq \mathrm{L}_{\mathrm{max}}\}\) . To determine whether a region \(\mathrm{R}\) is selected as an SVR, we test + +\[\mathrm{H}_0^{\mathrm{R}}\colon \mathrm{p}_{\mathrm{j}}\sim \mathrm{i.i.d.U}(0,1),\mathrm{j}\in \mathrm{R}\] + +based on the magnitude of \(\mathrm{Q}(\mathrm{R})\) . We implemented a Monte Carlo approach to control the familywise error rate for DiffScan (see Supplementary Methods). + +## Simulations + +We simulated reactivities for two conditions through the following three steps. + +Step 1. 100 human RNA sequences in an icSHAPE dataset17 were randomly selected. For each RNA sequence, we sampled 10 conformations from the Boltzmann distribution of secondary structure52 utilizing the RNAsubopt program in ViennaRNA version 2.4.1553. + +<--- Page Split ---> + +Step 2. Conditional on the pairing status of each nucleotide in a transcript, the reactivities can be sampled from pre- trained reactivity distributions. Three distributions were used: Cordero et al. \(^{35}\) and Sükösd et al. \(^{36}\) as fitted from SHAPE data, and also reactivity distributions we fitted from icSHAPE data (Supplementary Methods). + +Step 3. Different biological conditions were characterized by differential compositions of the 10 conformations generated in Step 1. The reactivities of each conformation were linearly combined accordingly to generate the observed reactivities in each condition. Specifically, for condition A, the 10 conformations of a transcript were assigned with weights \(w_{\mathrm{C}}^{\mathrm{A}}\) , \(1 \leq c \leq 10\) , with \(w_{\mathrm{C}}^{\mathrm{A}}\) corresponding to the weight of the \(c^{\mathrm{th}}\) conformation in condition A, with \(0 \leq w_{\mathrm{C}}^{\mathrm{A}} \leq 1\) , \(\Sigma_{c = 1}^{10} w_{\mathrm{C}}^{\mathrm{A}} = 1\) . For condition B, the weight of the \(c^{\mathrm{th}}\) conformation was \(w_{\mathrm{B}}^{\mathrm{B}}\) , with \(0 \leq w_{\mathrm{B}}^{\mathrm{B}} \leq 1\) , \(\Sigma_{c = 1}^{10} w_{\mathrm{B}}^{\mathrm{B}} = 1\) . To simulate SVRs that cover a reasonable proportion (approximately \(20\%\) \(^{16,17}\) ) of all nucleotide positions of the 100 transcripts, we allocated \(90\%\) of the weight to the first two conformations and randomly distributed the remaining \(10\%\) to the other eight conformations. We then generated reactivities in condition A and B separately by changing the weights of the first two conformations. Thus, nucleotide positions with different pairing status between the first two conformations form SVRs. Furthermore, replicates were simulated by adding random noise \((N(0,0.1^{2}))\) to the weights of the first two conformations. Finally, simulations were conducted at three levels of signal strength respectively by setting (i) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.2, 0.7)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.7, 0.2)\) , (ii) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.1, 0.8)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.8, 0.1)\) , and (iii) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.9, 0.9)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.9, 0)\) . + +## Performance Metrics for Validation of DiffScan + +In the simulated datasets, we evaluated the performance of different methods with two metrics. First, to evaluate how accurately simulated SVRs (i.e., true SVRs to detect between the compared + +<--- Page Split ---> + +conditions in the simulated datasets) were mapped by predicted SVRs by a method, we took out 1,000 nucleotide positions in the top- ranked SVRs by the method. The distance from each of the nucleotide positions to simulated SVRs was calculated. Second, we calculated the Precision- Recall curve. At varying significance levels, the value of precision was calculated as the number of correctly predicted nucleotide positions divided by the number of all predicted nucleotide positions. The value of recall rate was calculated as the number of correctly predicted nucleotide positions divided by the number of all nucleotide positions in simulated SVRs. + +In the negative control datasets (Control 1- 6), to access whether the false positive discoveries of a method can be controlled at a nominal level, we calculated the position- level false positive rate as the number of predicted nucleotide positions at significance level 0.05 divided by the length of the transcript. + +In the benchmark datasets with annotated SVRs (Flu and RRE)—similar to our processing of the simulated datasets—we calculated (1) the average distance from 10 nucleotide positions in top predicted SVRs to annotated SVRs and (2) the Precision- Recall curve. + +## Implementation of existing methods + +The software deltaSHAPE version 1.0 available at the Weeks lab website22 was utilized to implement deltaSHAPE. As far as we know, there is no official software release for PARCEL or RASA; and we implemented them utilizing custom scripts from existing literature23. The R package dStruct version 1.0.0 was utilized to implement dStruct. + +## Application to icSHAPE data + +## Reactivity calculation + +<--- Page Split ---> + +To calculate reactivities, we assumed that \(\mathrm{RT_j \sim Binomial(N_j, p_j)}\) , in which \(\mathrm{RT_j}\) is the RT stop at position j in the case experiment, \(\mathrm{N_j}\) is the number of times that position j is exposed to the probing molecules, and \(\mathrm{p_j}\) is the probability that position j is modified. The maximum likelihood estimator for \(\mathrm{p_j}\) is \(\hat{\mathrm{p_j}} = \frac{\mathrm{RT_j}}{\mathrm{N_j}}\) . On the other hand, the coverage at position j in the control experiment (coveragejcontrol) is approximately proportional to \(\mathrm{N_j}\) . Therefore, we calculated the reactivity at position j as \(\mathrm{r_j} = \frac{\mathrm{RT_j}}{\mathrm{coverage_j^{control}}} ^{54}\) . Based on this, two count replicates from case experiments were paired with two count replicates from control experiments to produce two reactivity replicates under each condition. + +## Enrichment analysis of RBP binding motifs + +We collected 154 RBP binding motifs from the CISBP- RNA Database \(^{55}\) and the ATTRACT database \(^{56}\) . For each motif, we searched in the predicted SVRs for significant motif hits utilizing the FIMO module (--norc --thresh 0.001) from the MEME suite \(^{40}\) , and the number of significant hits in each SVR were counted. + +To evaluate the significance of motif enrichment, we randomly sampled null regions in non- SVR regions in the same transcripts, with region length matched to the predicted SVRs. The number of significant hits in each null region is recorded. After that, we performed a one- sided Wilcoxon signed- rank test for the two count vectors of significant hits to evaluate the significance of enrichment. Motifs with corrected p value (Benjamini- Hochberg method) less than 0.05 were considered significantly enriched. We further consolidated the result by filtering out motifs that lost enrichment, i.e., with corrected p value (Benjamini- Hochberg method) greater than 0.01, when the null regions were sampled from the non- SVR regions of all transcripts with SVRs. + +Inference for the relationships among SVRs, RBP binding, and mRNA abundance + +<--- Page Split ---> + +We considered four models of which each explicitly describes the relationships among SVRs, changes in RBP binding, and changes in mRNA abundance (Model 1- 4 in Supplementary Fig. 8). For model selection, three variables of each mRNA transcript in the Np versus Cy comparison were used: (1) whether DiffScan predicted any SVRs in the Np versus Cy comparison for the transcript (svr; \(svr = "yes"\) or "no"), which quantifies its structural variations between Np and Cy, (2) \(\log_{10}\) (fold change of RPKM; Cy/Np) of the transcript (log10- fold- change- of- RPKM), which quantifies the change in its abundance between Np and Cy, and (3) a change- in- RBP- binding variable quantifying the changes in the binding of a specified RBP between Np and Cy for the transcript. Because QKI predominantly localizes in the Np, the change- in- RBP- binding variable for QKI of a transcript can be approximated by a score quantifying the level of QKI binding in the Np for the transcript. To calculate the score, we searched in the transcript for significant hits of QKI motifs utilizing the FIMO module (--norc --thresh 0.001), and the number of significant motif hits was taken as the score quantifying the level of QKI binding in the Np for the transcript. Similar processing was implemented for IGF2BP3, because it is almost exclusively cytoplasmic. + +Given the three variables of each transcript, i.e., \(svr\) , \(\log_{10}\) - fold- change- of- RPKM, and change- in- RBP binding for a specified RBP, we tested conditional independence between \(svr\) and \(\log_{10}\) - fold- change- of- RPKM given change- in- RBP- binding utilizing the conditional independence test for hybrid Bayesian networks implemented in the bnlearn R package57. In Model 1- 2, \(svr\) and \(\log_{10}\) - fold- change- of- RPKM are conditionally independent given change- in- RBP- binding, whereas \(svr\) regulates \(\log_{10}\) - fold- change- of- RPKM given change- in- RBP- binding in Model 3- 4. In cases where Model 1- 2 are rejected by the conditional independence test, Model 3- 4 are supported. However, Model 3 and Model 4 are statistically equivalent with current data. + +<--- Page Split ---> + +## Data availability + +Data availabilityAll data used in the manuscript were publicly available. The benchmark datasets were downloaded from https://doi.org/10.5281/zenodo.2536501. The icSHAPE dataset for transcriptome analysis were downloaded from the GEO database with accession code GSE117840 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE117840). + +## Code availability + +Code availabilityAll codes and the developed software were publicly available at https://github.com/yub18/DiffScan. + +## References + +References1. Licatalosi, D.D. & Darnell, R.B. RNA processing and its regulation: global insights into biological networks. \*Nature Reviews Genetics\* 11, 75- 87 (2010).2. Cech, T.R. The RNA worlds in context. \*Cold Spring Harb Perspect Biol\* 4, a006742 (2012).3. Cech, T.R. & Steitz, J.A. The noncoding RNA revolution- trashing old rules to forge new ones. \*Cell\* 157, 77- 94 (2014).4. Yu, B., Lu, Y., Zhang, Q.C. & Hou, L. Prediction and differential analysis of RNA secondary structure. \*Quantitative Biology\* 8, 109- 118 (2020).5. Strobel, E.J., Yu, A.M. & Lucks, J.B. High- throughput determination of RNA structures. \*Nat Rev Genet\* 19, 615- 634 (2018).6. Li, P., Zhou, X., Xu, K. & Zhang, Q.C. RASP: an atlas of transcriptome- wide RNA secondary structure probing data. \*Nucleic Acids Research\* (2020).7. Loughrey, D., Watters, K.E., Settle, A.H. & Lucks, J.B. SHAPE- Seq 2.0: systematic optimization and extension of high- throughput chemical probing of RNA secondary structure with next generation sequencing. \*Nucleic Acids Research\* 42, 10 (2014).8. Lucks, J.B. et al. Multiplexed RNA structure characterization with selective 2 '- hydroxyl acylation analyzed by primer extension sequencing (SHAPE- Seq). \*Proc. Natl. Acad. Sci. U. S. A.\* 108, 11063- 11068 (2011).9. Rouskin, S., Zubradt, M., Washietl, S., Kellis, M. & Weissman, J.S. Genome- wide probing of RNA structure reveals active unfolding of mRNA structures in vivo. \*Nature\* 505, 701- 705 (2014). + +<--- Page Split ---> + +10. Spitale, R.C. et al. Structural imprints in vivo decode RNA regulatory mechanisms. Nature 519, 486-490 (2015).11. Andrzejewska, A., Zawadzka, M. & Pachulska-Wieczorek, K. On the Way to Understanding the Interplay between the RNA Structure and Functions in Cells: A Genome-Wide Perspective. Int J Mol Sci 21 (2020).12. Mustoe, A.M., Brooks, C.L. & Al-Hashimi, H.M. Hierarchy of RNA functional dynamics. Annu Rev Biochem 83, 441-466 (2014).13. Dambach, M. et al. The Ubiquitous yybP-ykoY Riboswitch Is a Manganese-Responsive Regulatory Element. Mol. Cell 57, 1099-1109 (2015).14. Mayr, C. Regulation by 3'-Untranslated Regions. Annu Rev Genet 51, 171-194 (2017).15. Mayya, V.K. & Duchaine, T.F. Ciphers and Executioners: How 3'-Untranslated Regions Determine the Fate of Messenger RNAs. Frontiers in Genetics 10 (2019).16. Shi, B. et al. RNA structural dynamics regulate early embryogenesis through controlling transcriptome fate and function. Genome Biol 21, 120 (2020).17. Sun, L. et al. RNA structure maps across mammalian cellular compartments. Nat Struct Mol Biol 26, 322-330 (2019).18. Aw, Jong Ghut A. et al. In Vivo Mapping of Eukaryotic RNA Interactomes Reveals Principles of Higher-Order Organization and Regulation. Mol. Cell 62, 603-617 (2016).19. Tapsin, S. et al. Genome-wide identification of natural RNA aptamers in prokaryotes and eukaryotes. Nat Commun 9, 1289 (2018).20. Mizrahi, O. et al. Virus-Induced Changes in mRNA Secondary Structure Uncover cis-Regulatory Elements that Directly Control Gene Expression. Mol Cell 72, 862-874 e865 (2018).21. Wan, Y. et al. Landscape and variation of RNA secondary structure across the human transcriptome. Nature 505, 706-709 (2014).22. Smola, M.J., Calabrese, J.M. & Weeks, K.M. Detection of RNA-Protein Interactions in Living Cells with SHAPE. Biochemistry 54, 6867-6875 (2015).23. Choudhary, K., Lai, Y.H., Tran, E.J. & Aviar, S. dStruct: identifying differentially reactive regions from RNA structureome profiling data. Genome Biol 20, 40 (2019).24. Strobel, E.J., Watters, K.E., Nedialkov, Y., Artsimovitch, I. & Lucks, J.B. Distributed biotin-streptavidin transcription roadblocks for mapping cotranscriptional RNA folding. Nucleic Acids Research 45, e109-e109 (2017).25. Kutchko, K.M. & Laederach, A. Transcending the prediction paradigm: novel applications of SHAPE to RNA function and evolution. Wiley Interdisciplinary Reviews: RNA 8 (2017).26. Choudhary, K., Deng, F. & Aviar, S. Comparative and integrative analysis of RNA structural profiling data: current practices and emerging questions. Quant Biol 5, 3-24 (2017).27. Wang, X.W., Liu, C.X., Chen, L.L. & Zhang, Q.C. RNA structure probing uncovers RNA structure-dependent biological functions. Nat Chem Biol 17, 755-766 (2021).28. Shao, Z., Zhang, Y., Yuan, G.C., Orkin, S.H. & Waxman, D.J. MAnorm: a robust model for quantitative comparison of ChIP-Seq data sets. Genome Biol 13, R16 (2012).29. Xiang, G. et al. S3norm: simultaneous normalization of sequencing depth and signal-to-noise ratio in epigenomic data. Nucleic Acids Res (2020).30. Bolstad, B.M., Irizarry, R.A., Astrand, M. & Speed, T.P. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185-193 (2003).31. Kulldorff, M. A spatial scan statistic. Communications in Statistics - Theory and Methods 26, 1481-1496 (1997). + +<--- Page Split ---> + +32. Glaz, J. Scan statistics. (New York : Springer, New York; 2001).33. Durand, D. & Sankoff, D. Tests for gene clustering. J Comput Biol 10, 453-482 (2003).34. Guo, H., Li, J.J., Lu, Q. & Hou, L. Detecting local genetic correlations with scan statistics. Nature Communications 12, 2033 (2021).35. Cordero, P., Kladwang, W., VanLang, C.C. & Das, R. Quantitative dimethyl sulfate mapping for automated RNA secondary structure inference. Biochemistry 51, 7037-7039 (2012).36. Sukosd, Z., Swenson, M.S., Kjems, J. & Heitsch, C.E. Evaluating the accuracy of SHAPE-directed RNA secondary structure predictions. Nucleic Acids Res 41, 2807-2816 (2013).37. Kortmann, J. & Narberhaus, F. Bacterial RNA thermometers: molecular zippers and switches. Nature Reviews Microbiology 10, 255-265 (2012).38. Li, H. & Aviiran, S. Statistical modeling of RNA structure profiling experiments enables parsimonious reconstruction of structure landscapes. Nat Commun 9, 606 (2018).39. Bai, Y., Tambe, A., Zhou, K. & Doudna, J.A. RNA-guided assembly of Rev-RRE nuclear export complexes. Elife 3, e03656 (2014).40. Bailey, T.L. et al. MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res 37, W202-208 (2009).41. Thul, P.J. et al. A subcellular map of the human proteome. Science 356 (2017).42. Teplova, M. et al. Structure-function studies of STAR family Quaking proteins bound to their in vivo RNA target sites. Genes Dev 27, 928-940 (2013).43. Aberg, K., Saetre, P., Jareborg, N. & Jazin, E. Human QKI, a potential regulator of mRNA expression of human oligodendrocyte-related genes involved in schizophrenia. Proc Natl Acad Sci U S A 103, 7482-7487 (2006).44. Lu, H. et al. QKI regulates adipose tissue metabolism by acting as a brake on thermogenesis and promoting obesity. EMBO Rep 21, e47929 (2020).45. Wachter, K., Kohn, M., Stohr, N. & Huttelmaier, S. Subcellular localization and RNP formation of IGF2BPs (IGF2 mRNA-binding proteins) is modulated by distinct RNA-binding domains. Biol Chem 394, 1077-1090 (2013).46. Li, H. et al. Runting and Stunting Syndrome Is Associated With Mitochondrial Dysfunction in Sex-Linked Dwarf Chicken. Frontiers in Genetics 10 (2020).47. Watters, K.E., Strobel, E.J., Yu, A.M., Lis, J.T. & Lucks, J.B. Cotranscriptional folding of a riboswitch at nucleotide resolution. Nat Struct Mol Biol 23, 1124-1131 (2016).48. Morandi, E. et al. Genome-scale deconvolution of RNA structure ensembles. Nat. Methods 18, 249-252 (2021).49. Flynn, R.A. et al. Transcriptome-wide interrogation of RNA secondary structure in living cells with icSHAPE. Nat. Protoc. 11, 273-290 (2016).50. Amaratunga, D. & Cabrera, J. Analysis of Data From Viral DNA Microchips. Journal of the American Statistical Association 96, 1161-1170 (2001).51. McKean, J.W. Robust Analysis of Linear Models. Statist. Sci. 19, 562-570 (2004).52. Ding, Y. A statistical sampling algorithm for RNA secondary structure prediction. Nucleic Acids Research 31, 7280-7301 (2003).53. Lorenz, R. et al. ViennaRNA Package 2.0. Algorithms. Mol. Biol. 6, 26 (2011).54. Selega, A., Sirocchi, C., Iosub, I., Granneman, S. & Sanguinetti, G. Robust statistical modeling improves sensitivity of high-throughput RNA structure probing experiments. Nat Methods 14, 83-89 (2017).55. Ray, D. et al. A compendium of RNA-binding motifs for decoding gene regulation. Nature 499, 172-177 (2013). + +<--- Page Split ---> + +56. Giudice, G., Sánchez-Cabo, F., Torroja, C. & Lara-Pezzi, E. ATTRACT-a database of RNA-binding proteins and associated motifs. Database (Oxford) 2016 (2016).57. Scutari, M. Bayesian Network Constraint-Based Structure Learning Algorithms: Parallel and Optimized Implementations in the bnlearn R Package. 2017 77, 20 (2017). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0_det.mmd b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b3b8bd70652e4f7c4272cfb5d763b23191a87dea --- /dev/null +++ b/preprint/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0/preprint__13b4ad41a4b6d76a4f1b69d9923f0eec1605b4ae5f7cb340e3f659fe15509df0_det.mmd @@ -0,0 +1,450 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 922, 209]]<|/det|> +# Differential Analysis of RNA Structure Probing Experiments at Nucleotide Resolution: Uncovering Regulatory Functions of RNA Structure + +<|ref|>text<|/ref|><|det|>[[44, 230, 799, 410]]<|/det|> +Bo Yu Tsinghua University Pan Li Tsinghua University Qiangfeng Zhang School of Life Sciences, Tsinghua University https://orcid.org/0000- 0002- 4913- 0338 Lin Hou ( houl@tsinghua.edu.cn ) Tsinghua University https://orcid.org/0000- 0002- 4283- 8501 + +<|ref|>sub_title<|/ref|><|det|>[[44, 451, 102, 469]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 490, 930, 510]]<|/det|> +Keywords: RNA structure, structure probing experiments, differential analysis, RNA secondary structure + +<|ref|>text<|/ref|><|det|>[[44, 528, 319, 547]]<|/det|> +Posted Date: August 26th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 566, 463, 586]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 800283/v1 + +<|ref|>text<|/ref|><|det|>[[44, 603, 910, 645]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 680, 912, 724]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 22nd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 31875- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[144, 90, 853, 110]]<|/det|> +# Differential Analysis of RNA Structure Probing Experiments at Nucleotide Resolution: + +<|ref|>title<|/ref|><|det|>[[144, 129, 559, 147]]<|/det|> +# Uncovering Regulatory Functions of RNA Structure + +<|ref|>text<|/ref|><|det|>[[144, 174, 530, 192]]<|/det|> +Bo Yu \(^{1}\) , Pan Li \(^{2,3}\) , Qiangfeng Cliff Zhang \(^{2,3*}\) , Lin Hou \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[144, 218, 855, 265]]<|/det|> +\(^{1}\) Center for Statistical Science, Department of Industrial Engineering, Tsinghua University, Beijing 100084, China + +<|ref|>text<|/ref|><|det|>[[144, 289, 700, 306]]<|/det|> +\(^{2}\) MOE Key Laboratory of Bioinformatics, Tsinghua University, Beijing 100084, China + +<|ref|>text<|/ref|><|det|>[[144, 330, 855, 405]]<|/det|> +\(^{3}\) Center for Synthetic and Systems Biology, Beijing Advanced Innovation Center for Structural Biology, Tsinghua- Peking Joint Center for Life Sciences, School of Life Sciences, Tsinghua University, Beijing 100084, China + +<|ref|>text<|/ref|><|det|>[[144, 433, 610, 449]]<|/det|> +\* Correspondence: houl@tsinghua.edu.cn, qczhang@tsinghua.edu.cn + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 92, 217, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[143, 131, 858, 737]]<|/det|> +RNAs perform their function by forming specific structures, which can change across cellular conditions. Structure probing experiments combined with next generation sequencing technology have enabled transcriptome- wide analysis of RNA secondary structure in various cellular conditions. Differential analysis of structure probing data in different conditions can reveal the RNA structurally variable regions (SVRs), which is important for understanding RNA functions. Here, we propose DiffScan, a computational framework for normalization and differential analysis of structure probing data in high resolution. DiffScan preprocesses structure probing datasets to remove systematic bias, and then scans the transcripts to identify SVRs and adaptively determines their lengths and locations. The proposed approach is compatible with most structure probing platforms (e.g., icSHAPE, DMS- seq). When evaluated with simulated and benchmark datasets, DiffScan identifies structurally variable regions at nucleotide resolution, with substantial improvement in accuracy compared with existing SVR detection methods. Moreover, the improvement is robust when tested in multiple structure probing platforms. Application of DiffScan in a dataset of multi- subcellular RNA structureome identified multiple regions that form different structures in nucleus and cytoplasm, linking RNA structural variation to regulation of mRNAs encoding mitochondria- associated proteins. This work provides an effective tool for differential analysis of RNA secondary structure, reinforcing the power of structure probing experiments in deciphering the dynamic RNA structureome. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 92, 250, 108]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[143, 135, 857, 461]]<|/det|> +RNA molecules play important roles in myriad cellular processes by forming specific structures \(^{1 - 3}\) . Deciphering RNA structure is informative for understanding RNA functions. In recent years, diverse structure probing (SP) methods have been developed to study RNA secondary structure in various cellular contexts, which utilize chemicals that react differentially to nucleotides according to their pairing status \(^{4,5}\) . Coupled with next generation sequencing technologies, SP experiments can be performed at high throughput, providing a transcriptome level view of RNA secondary structure \(^{4 - 6}\) , i.e., RNA structure. There are various mature high- throughput SP platforms, such as SHAPE- Seq \(^{7,8}\) , DMS- Seq \(^{9}\) , and icSHAPE \(^{10}\) . These platforms offer flexible options for tackling different biological problems, and have achieved success in uncovering pervasive links between RNA structure and RNA function \(^{11}\) . + +<|ref|>text<|/ref|><|det|>[[143, 486, 857, 707]]<|/det|> +Studies have examined how the structures of particular RNA molecules change across multiple cellular conditions, revealing explicit connections between RNA structure and RNA function \(^{12}\) . For example, the ubiquitous yybP- ykoY motif has been shown to adopt distinct structures in response to manganese ions, thus exerting regulatory consequences on protein translation in both Escherichia coli and Bacillus subtilis \(^{13}\) . RNA structural variations have also been reported to regulate the binding of trans- acting factors and RNA stability in multiple organisms, including human, yeast, and zebrafish \(^{14 - 18}\) . + +<|ref|>text<|/ref|><|det|>[[143, 732, 857, 888]]<|/det|> +To explore the SP experiments to uncover dynamic RNA structures, quantitative tools that contrast SP experiments to identify structurally variable regions (SVRs) are in great demand. Several methods have been proposed to identify SVRs. PARCEL \(^{19}\) and RASA \(^{20}\) directly model and compare raw read counts at each nucleotide position, and then identify regions enriched for position- level signals. However, their models are tailored for specific experimental protocols, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 88, 857, 245]]<|/det|> +it is not straightforward to extend them to accommodate more emerging SP techniques. StrucDiff \(^{21}\) , deltaSHAPE \(^{22}\) , and dStruct \(^{23}\) take reactivities as input, which are estimated from raw read counts to summarize the pairing status at each nucleotide position. They search for SVRs with pre- specified search lengths to aggregate differential signals \(^{21 - 23}\) . However, the choice of search length is arbitrary. + +<|ref|>text<|/ref|><|det|>[[143, 269, 858, 696]]<|/det|> +Despite the success of existing methods, differential analysis of SP data remains challenging in many aspects. First, SVRs manifest great variation in length, ranging from a few to several dozens of nucleotide positions \(^{19, 23, 24}\) . As a result, searching with fixed search length can lead to insufficient detection power and inaccurate boundary mapping, when the prespecified search length deviates greatly from the true length. Second, SP platforms differ in utilization of additives, specific experimental protocols, and preprocessing pipelines, yielding distinct data types and distributions in output \(^{4, 5, 25}\) . As a result, it is desirable but usually not easy to extend methods developed for one platform to another. Third, SP data can be confounded by systematic bias \(^{4, 5}\) , which should be removed from reactivities of the two compared conditions prior to differential analysis \(^{26}\) . However, to our knowledge, normalization techniques that are compatible with multiple SP platforms are still lacking. Finally, rigorous error control is also highly impactful on the accuracy and biological relevance of output from SVR detection \(^{23}\) , especially for transcriptome- scale analyses. + +<|ref|>text<|/ref|><|det|>[[143, 720, 857, 908]]<|/det|> +To address the unmet needs, we propose DiffScan, a computational framework for differential analysis of SP data at nucleotide resolution. DiffScan normalizes SP reactivities via a built- in Normalization module, which is compatible with various platforms, and then looks for SVRs via a Scan module. The Scan module locates SVRs at nucleotide resolution, and rigorously controls family- wise error rate. We demonstrated with large- scale simulated datasets and benchmark datasets that DiffScan achieves the best power and accuracy in SVR detection across + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 857, 210]]<|/det|> +various platforms, compared to state- of- art methods, We applied DiffScan to a recent icSHAPE dataset of RNA structurome in multiple subcellular compartments (e.g., nucleoplasm and cytoplasm), and revealed the potential roles of SVRs in regulating mRNAs encoding mitochondria- associated proteins. + +<|ref|>sub_title<|/ref|><|det|>[[145, 281, 207, 298]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[145, 327, 284, 344]]<|/det|> +## Model description + +<|ref|>text<|/ref|><|det|>[[143, 370, 857, 594]]<|/det|> +DiffScan comprises a Normalization module and a Scan module (Fig. 1a). SP reactivities of RNA transcripts for two conditions (e.g., in vitro, in vivo) are taken as input, and the reactivities are normalized relative to one another in the Normalization module (Fig. 1b) to correct for systematic bias. The corrected reactivities are comparable as far as possible across different cellular conditions. In the Scan module (Fig. 1c), structural variations are initially evaluated at each nucleotide position, and then the algorithm scans through the transcripts with variable search lengths to identify SVRs. + +<|ref|>sub_title<|/ref|><|det|>[[145, 620, 254, 636]]<|/det|> +## Normalization + +<|ref|>text<|/ref|><|det|>[[143, 661, 857, 885]]<|/det|> +It is known that the reactivities of particular RNA nucleotide positions can be affected by a variety of confounding factors in SP experiments26, including transcript abundance, sequencing depth, and signal- to- noise ratio (see Methods). Owing to discrepancies in these factors, the SP reactivities obtained from two conditions lacking substantial biological differences may show substantial differences27 (see Supplementary Fig. 1 for an example). The unwanted variations will then persist throughout the routine preprocessing steps. Thus, in practice careful adjustment between and within conditions is essential when comparing SP data obtained from samples in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 88, 857, 210]]<|/det|> +distinct conditions. Related challenges have been widely reported for other high- throughput sequencing experiments, such as ChIP- Seq28 and ATAC- Seq29. Inspired by normalization approaches for other high- throughput sequencing based methods, we propose the following normalization procedure for differential analysis of SP data. + +<|ref|>text<|/ref|><|det|>[[143, 234, 858, 696]]<|/det|> +To account for a potentially wide range of reactivity levels, we first rescale the reactivity values into the interval [0,1] after a \(90\%\) winsorization (see Methods). Next, the reactivities of within- condition replicates, if available, are processed by quantile normalization30, so that the quantiles of each replicate are matched within conditions. We subsequently normalize between- condition reactivities using an approach similar to MAnorm28. Briefly, a structurally invariant set of nucleotide positions \(S\) is determined as a "pivot" for normalization, and the reactivities are transformed so that the transformed reactivities are at the same level for the pivot set between conditions. The basic idea is to learn transformation rules from \(S\) and then extrapolate the learned transformation to all nucleotide positions of the transcripts. In particular, the transformation is learned from training a linear model from \(S\) , which takes reactivities in one condition as response and reactivities in the other condition as predictor. The adjusted reactivities from the Normalization module are then ready for differential analysis in the Scan module (Supplementary Fig. 2). We show in theory that the transformation explicitly corrects for differences in sequencing depth and signal- to- noise ratio. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[143, 90, 857, 312]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[143, 333, 857, 627]]<|/det|> +
Fig. 1 Workflow of DiffScan. a Taking raw reactivities as input, DiffScan first normalizes them relative to one another in the Normalization module (b) to correct for systematic bias, and then identifies SVRs in the Scan module (c). Different colors denote reactivity replicates in each condition. b The Normalization module transforms raw reactivities into normalized reactivities to remove systematic bias. The normalized reactivities are comparable as far as possible across different cellular conditions. Red points indicate nucleotide positions in SVRs. c Taking normalized reactivities as input, the Scan module first calculates the significance of any differential signals for each nucleotide position with Wilcoxon test, and then concatenates positional p values into a regional signal via scan statistic. The significance of the scan statistic for each enumerated region is evaluated by Monto Carlo sampling, and those regions crossing a specified significance threshold are reported as SVRs.
+ +<|ref|>sub_title<|/ref|><|det|>[[146, 695, 183, 708]]<|/det|> +## Scan + +<|ref|>text<|/ref|><|det|>[[144, 734, 857, 890]]<|/det|> +Formally, "scan statistics" refer to statistical methods for cluster detection in time and space31, 32. These methods can accurately map regional signals and control family- wise error rates during multiple testing of interrelated hypotheses. Scan statistics have been successfully applied to many areas including molecular biology33 and human genetics34. The Scan module of DiffScan was developed with the goal to identify SVRs at nucleotide resolution in a data- adaptive fashion. In + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 857, 346]]<|/det|> +brief, the Scan module slides through the transcripts to enumerate overlapping regions of different lengths, identifies regions with maximal differential signals, and evaluates their statistical significance. In detail, we first quantify positional differential signals by calculating a p value for each nucleotide position by Wilcoxon test, which contrasts reactivities between conditions in a small window surrounding the nucleotide position. Note that the Normalization module does not enforce any specific distributions of the normalized reactivities (Supplementary Fig. 3), and the nonparametric test we use guarantees robust evaluation of differential signals for various SP platforms. + +<|ref|>text<|/ref|><|det|>[[144, 371, 855, 424]]<|/det|> +Second, for each region R, we propose the following scan statistic to quantify differential signals for regions, + +<|ref|>equation<|/ref|><|det|>[[410, 449, 589, 494]]<|/det|> +\[\mathrm{Q}(\mathrm{R}) = \frac{-\Sigma_{\mathrm{j}\in \mathrm{R}}\log(\mathrm{p}_{\mathrm{j}})}{\sqrt{|\mathrm{R}|}}.\] + +<|ref|>text<|/ref|><|det|>[[144, 515, 855, 603]]<|/det|> +The sum in the numerator aggregates positional differential signals, while the denominator penalizes the extension of a candidate region. Intuitively, regions that are enriched with structurally variable nucleotides will obtain high Q(R) scores. + +<|ref|>text<|/ref|><|det|>[[143, 627, 857, 818]]<|/det|> +We search in the transcripts with sliding windows of different lengths, and calculate the scan statistic for each candidate region (Fig. 1c). A Monte Carlo approach is then implemented to evaluate the statistical significance of the scan statistics, which addresses the multiple testing problem for the overlapping regions by controlling the family- wise error rate (See Supplementary Methods). Accordingly, we can identify SVRs with accurately mapped boundaries based on the significance of the scan statistics (Fig. 1c). + +<|ref|>title<|/ref|><|det|>[[144, 887, 522, 904]]<|/det|> +# Validation of DiffScan using simulated SP datasets + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 859, 483]]<|/det|> +We simulated RNA secondary structures in two conditions, and generated SP reactivities based on the simulated secondary structures in each condition using three types of empirical models representing different SP platforms, including two types of SHAPE reactivity25, 35, 36 and one type of icSHAPE reactivity17 (see Supplementary Methods). Biologically, it is often the case that a particular RNA molecule will occur as a mixture of structural conformations in a given cellular condition37, 38, so SVRs detected between conditions represent altered proportions comprising these mixtures of structural conformations. Thus, the between- condition differences in RNA secondary structures can be very subtle in reality. To mimic the complexity of the landscape of RNA secondary structures, we sampled multiple structural conformations from an ensemble of energy- function- based predictions, and mixed them in varying proportions in the simulated datasets (see Methods). The reactivities were thereafter simulated based on the pairing status of each nucleotide. + +<|ref|>text<|/ref|><|det|>[[143, 506, 859, 864]]<|/det|> +In detail, we sampled 10 conformations for each of the 100 RNA transcripts we randomly selected from transcripts in human embryonic kidney (HEK293) cells17. The lengths of the RNA transcripts ranged between 61 nt and 4,810 nt, and the simulated SVRs covered 3% to 55% of nucleotide positions of the RNA transcripts, with lengths between 1 nt and 16 nt. These data included a total of 13,162 simulated SVRs, with 6,587 being single nucleotide structural variations, 6,081 having lengths between 1 nt and 5 nt, and 494 SVRs with lengths greater than 5 nt (see Supplementary Fig. 4 for further details). Note that these simulated data echo the real world knowledge that the lengths of SVRs vary extensively13, 19, 23, 24, 39. We varied the strength of differential signals of simulated SVRs between “high”, “medium”, and “low”. Combining different levels of signal strength and the three types of generative models of reactivity, we have in total 9 simulation settings. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 856, 245]]<|/det|> +We compared the performance of DiffScan with two other SVR detection methods, deltaSHAPE and dStruct. Note that DiffScan adaptively determines the lengths of SVRs from data, whereas search length is a prespecified parameter for deltaSHAPE and dStruct. We set the search length to 5 nt for deltaSHAPE (the default setting suggested by the authors \(^{22}\) ) and to three different values (1 nt, 5 nt, and 11 nt) for dStruct. + +<|ref|>text<|/ref|><|det|>[[143, 269, 857, 457]]<|/det|> +To evaluate the accuracy of SVR boundaries mapping, we calculated the distance between each nucleotide position in the predicted SVRs and the true SVRs. In the ideal case, if a predicted SVR sits within a simulated SVR, the distance for each nucleotide position in the predicted SVR is zero. Oppositely, if a predicted SVR is off- target, i.e., containing many nucleotide positions that are neither in or close to the simulated SVRs, the distances for such nucleotide positions in the predicted SVR will be large. + +<|ref|>text<|/ref|><|det|>[[143, 482, 857, 806]]<|/det|> +The top- ranked SVRs identified by DiffScan are closer to the simulated SVRs than those by deltaSHAPE or dStruct in all 9 settings (Fig. 2a), demonstrating the superior performance of DiffScan in accurate identification of SVRs. Given that dStruct only considers nonoverlapping regions and relies on a pre- specified search length, and deltaSHAPE considers regions of fixed length, their performance is sensitive to whether or not pre- specified value of search length fits the actual lengths of the SVRs in a dataset. In contrast, DiffScan determines boundaries of SVRs by optimizing its scan statistic, thereby effectively distinguishing SVRs amongst overlapping regions of different lengths, leading to a finer level of granularity relative to deltaSHAPE and dStruct. Moreover, the superiority of DiffScan is consistent across different reactivity models and signal levels. + +<|ref|>text<|/ref|><|det|>[[144, 831, 855, 884]]<|/det|> +We also evaluated the precision and recall rate at the nucleotide position level (see Methods) of the SVR detection methods. Among all three methods, DiffScan achieves the best + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 857, 380]]<|/det|> +precision at the same recall rate (Fig. 2b). Using optimization based on its scan statistic, DiffScan effectively identifies nonconsecutive SVR segments with nucleotide resolution. Although dStruct also sensitively identifies SVRs, it tends to output long, contiguous regions, which include a substantial proportion of nucleotide positions without structural variations, hampering the precision. The output of deltaSHAPE is a fixed set of regions that the threshold is internally decided, and it is represented as dots rather than curves in Fig. 2b. Although deltaSHAPE works well for the SHAPE reactivity model from Cordero et al. \(^{35}\) , which is reasonable since the method was originally developed for SHAPE- MaP experiments, its precision is much lower than dStruct and DiffScan for other types of reactivity models. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[143, 90, 630, 675]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[143, 697, 857, 899]]<|/det|> +
Fig. 2 Comparison of DiffScan and other SVR detection methods with simulated datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 1 nt, 5 nt, and 11 nt. a Distances between simulated SVRs and the top-ranked 1,000 nucleotide positions in predicted SVRs. b Precision-Recall curves for the prediction results from the various SVR detection methods. Rows: three types of reactivity generative models. Columns: three levels of strength of differential signals at simulated SVRs. Note deltaSHAPE does not allow external thresholding, and therefore it is represented as dots instead of curves.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 135, 288, 152]]<|/det|> +## Sensitivity analysis + +<|ref|>text<|/ref|><|det|>[[143, 177, 857, 536]]<|/det|> +There are two tuning parameters in DiffScan: the radius of local windows (denoted by r) for smoothed calculation of positional differential signal, and the parameter associated with the penalty term (denoted by y) in the scan statistic. In the above analysis of the simulated datasets, we set r to 2 nt and y to 0.5 as their default values. We also conducted a sensitivity analysis wherein we tested DiffScan with the simulated datasets using eight other settings of \((r, \gamma)\) . In each setting, we calculated the mean precision of the predicted SVRs at recall values below 0.05. The coefficient of variation for the mean precisions corresponding to different tuning parameters is below 0.15 for all of the examined scenarios in the simulations (Supplementary Table 1), showing the robustness of DiffScan regarding to the tuning parameters. Notably, DiffScan consistently outperformed dStruct and deltaSHAPE in these tested parameter settings (Supplementary Fig. 5). + +<|ref|>sub_title<|/ref|><|det|>[[144, 605, 507, 622]]<|/det|> +## Model validation with negative control datasets + +<|ref|>text<|/ref|><|det|>[[143, 648, 857, 870]]<|/det|> +To assess the extent of false positive discoveries of DiffScan and other SVR detection methods, we constructed six negative control datasets (Control 1- 6, see Methods) from multiple SP platforms, including SHAPE- Seq and icSHAPE. In these datasets, no SVRs are present between the compared conditions. Thus, any significant SVRs identified by the detection methods would represent false positives. These comparisons included the PARCEL and RASA methods for the cases where raw read counts are available (Control 1, 2, 5, and 6). As dStruct requires within- condition replicates, it is not applicable for Control 5 and 6. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 857, 345]]<|/det|> +The false positive rates for DiffScan, dStruct, and PARCEL were below 0.05 for all tested datasets (Fig. 3a). It bears emphasis that the test space of DiffScan is tremendously larger than the other methods because it contains overlapping regions of differing lengths. Nevertheless, its false positive rates were well controlled. The error rate of RASA is slightly inflated in Control 2, but is well- controlled for the other datasets. Consistent inflation is observed for deltaSHAPE across different datasets, reporting false positive results covering \(\sim 15\%\) of nucleotide positions in the datasets. The fact that neither deltaSHAPE nor RASA control for the multiple testing problem can likely explain the observed inflation. + +<|ref|>sub_title<|/ref|><|det|>[[146, 416, 426, 434]]<|/det|> +## Evaluation using benchmark datasets + +<|ref|>text<|/ref|><|det|>[[143, 457, 858, 748]]<|/det|> +We used two benchmark datasets with explicitly annotated SVRs between the compared conditions, curated by Choudhary et al. \(^{23}\) to support evaluation of different SVR detection methods. The Flu dataset \(^{24}\) is from a SHAPE- Seq experiment that measured the secondary structure of a Bacillus cereus crcB fluoride riboswitch in vitro in the presence or absence of fluoride. The transcript (100 nt in length) has five annotated SVRs, ranging in length between 1 nt and 8 nt. The RRE dataset \(^{39}\) is from a SHAPE- Seq experiment that measured the secondary structure of the HIV Rev- response element in the presence or absence of the Rev protein. There are seven annotated SVRs for Rev- RRE interactions in the transcript (369 nt in length), with lengths ranging between 7 nt and 39 nt. + +<|ref|>text<|/ref|><|det|>[[144, 773, 856, 895]]<|/det|> +We evaluated the average distance between predicted SVRs and the annotated SVRs and Precision- Recall curves output by DiffScan, deltaSHAPE, PARCEL, RASA, and dStruct (Fig. 3b- c). Consistent with our results when evaluating using simulated datasets and showcasing DiffScan's superior performance for accurately locating SVRs, DiffScan achieved the lowest average distance + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 85, 859, 620]]<|/det|> +from the annotated SVRs of the benchmark datasets. For the Flu dataset, DiffScan identified all the five annotated SVRs with their boundaries accurately mapped (Supplementary Fig. 6). In contrast, PARCEL predicted a long region, covering \(72\%\) of all nucleotide positions in the transcript, that spanned all five annotated SVRs. However, it did not distinguish annotated SVRs from flanking nucleotides between SVRs and falsely entailed many nucleotide positions ( \(67\%\) of the nucleotide positions in the predicted SVRs) without structural variations. Similarly, dStruct identified a long region covering \(37\%\) of all nucleotide positions, which covered two annotated SVRs, while \(62\%\) of the nucleotide positions in the predicted SVRs had no structural variation. RASA did not report any significant region for the Flu dataset. deltaSHAPE reported six regions, missing an annotated SVR while containing a false positive discovery. For the RRE dataset, DiffScan detected three regions which are located within three annotated SVRs. PARCEL and RASA were not applicable for the RRE dataset because raw count data were not available. deltaSHAPE and dStruct both reported four regions, which overlapped with two and four annotated SVRs respectively. Compared to DiffScan, they reported regions with much larger average distance to annotated SVRs, since they included many nucleotide positions ( \(53\%\) and \(35\%\) of the nucleotide positions in the predicted SVRs for deltaSHAPE and dStruct) without structural variation. + +<|ref|>text<|/ref|><|det|>[[143, 643, 859, 899]]<|/det|> +In terms of precision and recall measures, DiffScan also achieved the best area under the curve (AUC) values with both datasets (Fig. 3c). Note that because deltaSHAPE and PARCEL do not allow threshold tunings, so they are represented as dots rather than curves in Fig. 3c. deltaSHAPE performed comparably to DiffScan for the Flu dataset. However, the precision of deltaSHAPE was much lower than DiffScan and dStruct for the RRE dataset, since the predicted SVRs of deltaSHAPE included many nucleotide positions without structural variation. In general, dStruct had lower precision compared to DiffScan when the recall rate was set to the same level, consistent to the trends we observed when analyzing the simulated datasets. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[147, 140, 808, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 596, 857, 830]]<|/det|> +
Fig. 3 Comparison of DiffScan and existing SVR detection methods with negative control datasets and benchmark datasets. For deltaSHAPE, we used the default search length of the method of 5 nt; for dStruct we used search length 5 nt following the original article of the method. a Position-level false positive rate at a significance level of 0.05 in six negative control datasets (i.e., datasets having no SVRs). b Average distance from the top-10 ranked nucleotide positions in the predicted SVRs to the annotated SVRs in the benchmark datasets. c Precision-Recall curves for the benchmark datasets. deltaSHAPE and PARCEL are represented as dots. "X" indicates that the corresponding method was not applicable for the dataset. "*" indicates that the corresponding method did not report any region.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 762, 109]]<|/det|> +## Roles of SVRs in regulating mRNA transcripts for mitochondria-associated proteins + +<|ref|>text<|/ref|><|det|>[[143, 133, 858, 493]]<|/det|> +We applied DiffScan to a recently reported icSHAPE dataset that mapped RNA secondary structure across human cellular compartments covering chromatin (Ch), nucleoplasm (Np), and cytoplasm (Cy)17. The DiffScan- predicted SVRs for the Ch versus Np (Fig. 4a) and for the Np versus Cy (Supplementary Fig. 7) comparisons mostly involved protein binding sites and RNA modification sites. As we also had data for RNA abundance in the Ch, Np, and Cy samples, we were interested in the potential impacts of SVRs on regulating mRNA abundance. Comparison of mRNA abundance in the Np versus Cy samples clearly placed the mRNAs into two subgroups, with mRNA levels in subgroup II decreased sharply in the Cy fraction (Fig. 4b). A manual check revealed that all 12 mRNAs of subgroup II encode mitochondria- associated proteins. In addition, DiffScan identified SVRs in all of these 12 transcripts (p value = 8.4e- 3), suggesting an association between RNA structural variation and the abundance of mRNAs encoding mitochondria- associated proteins. + +<|ref|>text<|/ref|><|det|>[[143, 515, 858, 909]]<|/det|> +To further investigate this association, we used the FIMO module from the MEME suite40 to search for RBPs with binding motifs enriched in the predicted SVRs from the Np versus Cy comparison (see Methods). Three enriched RBPs were identified, including quaking (QKI), insulin- like growth factor 2 mRNA- binding protein 3 (IGF2BP3), and polyadenylate- binding protein 1 (PABPC1). QKI had two significantly enriched motifs, ACUAACA (corrected p value = 1.4e- 5) and AUCUAAC (corrected p value = 7.8e- 4). QKI predominantly localizes in the Np41, 42, and it has been reported to regulate various biological processes such as myelinization43, mRNA stability, and mRNA export42. Of particular note, a recent study reported that QKI binds mRNAs encoding mitochondrial proteins and transcriptional coactivators known to regulate mitochondrial biogenesis and function: QKI decreases the stability and nuclear export of these RNAs44. Therefore, the binding of QKI to the mRNA transcripts of subgroup II, which all encode mitochondria- associated proteins, may in part explain their reduced abundance in the Cy samples. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 858, 450]]<|/det|> +Considering the biological evidence and the potential roles of SVRs, we enumerated four possible causal models to describe the relationships among SVRs, QKI binding, and changes in mRNA abundance (Model 1- 4 in Supplementary Fig. 8; see Methods). In Model 1, QKI binding causes SVRs in the mRNA transcripts of subgroup II and also regulates their abundance in a separate pathway, and thus SVRs have no contribution to the observed decline of mRNA abundance in the Cy. In Model 2- 4, SVRs either directly or indirectly (mediated by QKI binding) regulate mRNA abundance. The observed data rejects Model 1 and Model 2 (p value = 2.5e- 91), and therefore supports the direct roles of SVRs in regulation of mRNA abundance. However, Model 3 and 4 are statistically equivalent, i.e., not distinguishable. In another word, SVRs have direct effect in regulating mRNA abundance, in addition to QKI binding (Fig. 4c); however, the causal relationships between SVR and QKI binding cannot be determined with current data. + +<|ref|>text<|/ref|><|det|>[[143, 472, 858, 832]]<|/det|> +The second RBP, IGF2BP3, had one motif (ACAAACA) enriched among the predicted SVRs (corrected p value = 5.4e- 3). IGF2BP3 is almost exclusively cytoplasmic, and has been conceptualized to "cage" its target mRNAs into cytoplasmic protein- RNA complexes as they enter the Cy45. Recently, the upregulation of IGF2BP3 has been associated with mitochondrial dysfunction46. We reason that IGF2BP3 cages mRNA transcripts encoding mitochondria- associated proteins as they enter the Cy, accounting for the observed decline in the abundance of such mRNA transcripts. We enumerated possible causal models (Supplementary Fig. 8) and performed hypothesis testing to explore the relationships among SVRs, IGF2BP3 binding, and changes in mRNA abundance. According to our inference, there is a direct effect from the detected SVRs to the changes in mRNA abundance (Fig. 4c; p value = 7.4e- 49), although it is not clear whether the SVRs' regulation of mRNA abundance is also mediated by IGF2BP3 binding. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 855, 175]]<|/det|> +In summary, the DiffScan- predicted SVRs provide insight into the functional roles of RNA secondary structure and structural variation in regulating mRNAs encoding mitochondria- associated proteins. + +<|ref|>image<|/ref|><|det|>[[171, 245, 570, 722]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 744, 856, 860]]<|/det|> +
Fig. 4 Application of DiffScan to explore the roles of RNA structural variation in regulation of mRNA abundance in an icSHAPE dataset mapping RNA structure across human cellular compartments. Ch: chromatin, Np: nucleoplasm, Cy: cytoplasm. a Predicted SVRs between Ch and Np were enriched with protein binding sites and RNA modification sites. \*p value (Fisher's exact test) \(< 0.05\) , \*\*\*p value \(< 1e-6\) . b
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 854, 137]]<|/det|> +RPKM of transcripts and the prediction results of DiffScan for Np versus Cy. c Statistical inference supports the direct roles of SVRs in regulation of mRNA abundance. + +<|ref|>sub_title<|/ref|><|det|>[[145, 207, 233, 223]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[143, 245, 857, 888]]<|/det|> +We have developed DiffScan, a computational framework for differential analysis of RNA secondary structure measured in multiple SP platforms. Our method provides several advantages. First, it adaptively estimates the lengths and locations of SVRs with single- nucleotide resolution. Compared to existing SVR detection methods, predicted SVRs by DiffScan are closer to true SVRs, both from simulated and benchmark datasets. Second, DiffScan is compatible with multiple SP platforms with robust performance. Existing SVR detection methods are usually tailored for a specific SP platform, and their generalization to other SP platforms can be prohibitive. Our method flexibly accommodates multiple SP platforms through its Normalization module and using the nonparametric test we implemented. This enables flexible analysis that is adaptable to suitable data types reflecting particular biological problems of interest. For example, the SHAPE- Seq platform with fast- acting reagent is suitable for in vitro studies of RNA folding dynamics47, while the icSHAPE platform utilizing slow- acting reagent allows in vivo transcriptome- wide structure probing16, 17. As demonstrated with simulated and benchmark datasets, the superior performance of DiffScan in terms of statistical power and accuracy for identifying SVRs is robust across different SP platforms. Third, DiffScan rigorously controls for the family- wise error rate in SVR detection, which is particularly influential for transcriptome level analyses. At last, the SVR simulation framework we devised aptly reflects the complexity of SVR detection problems in the real world. We have developed a “simulation” module in our software, which can facilitate evaluation of more upcoming SVR detection methods in the future. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 90, 858, 450]]<|/det|> +DiffScan has several limitations. First, the Normalization module in DiffScan relies on a structurally invariant set of nucleotide positions, which is identified via a built- in data- driven strategy. The strategy would fail in extreme cases when almost all nucleotide positions in the studied transcripts exhibit structural variations, although we expect it is rarely the case in practice4, 16, 17, 48. In addition, when the structurally invariant set can be specified by prior knowledge, the normalization step can be easily adapted. Second, we recognize the statistical power of DiffScan is not optimized for a particular SP platform, so when the distribution of reactivities is known \(a\) priori, use of an appropriate parametric test would yield increased power. Third, we note that the ultimate goal of studying SVRs is to understand the functional roles of RNA structure, which often involves cross- examination of other data sources, including motif analysis of RBPs, multi- omics datasets, etc. Incorporation of these multi- source data may help to accurately annotate SVRs11. + +<|ref|>sub_title<|/ref|><|det|>[[146, 522, 308, 540]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[144, 566, 857, 687]]<|/det|> +This work was supported by the National Natural Science Foundation of China (Grants No. 91740204 and 91940306) and the State Key Research Development Program of China (Grant No. 2019YFA0110002) to Q.C.Z. LH acknowledges research support from the National Science Foundation of China (Grant No. 12071243). + +<|ref|>sub_title<|/ref|><|det|>[[146, 762, 319, 779]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[144, 806, 857, 893]]<|/det|> +B.Y., Q.C.Z., and L.H. conceived the study and wrote the manuscript. B.Y. and L.H. developed the computational framework. B.Y. and P.L. performed the data preparation and statistical analysis. B.Y., Q.C.Z., and L.H. interpreted the results. All authors approved the manuscript. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 139, 312, 155]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[145, 185, 470, 201]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[145, 276, 221, 293]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[145, 323, 348, 339]]<|/det|> +## Structure Probing Datasets + +<|ref|>text<|/ref|><|det|>[[143, 365, 858, 622]]<|/det|> +We evaluated the power of DiffScan and other SVR detection methods in two SHAPE- seq datasets, of which the SVRs were curated in Choudhary et al.23. The Flu dataset probed the Bacillus cereus crcB fluoride riboswitch in vitro with and without fluoride ions, with four replicates for each condition. The transcript is 100 nt in length, and nucleotide positions 12- 17, 22- 27, 38- 40, 48 and 67- 74 were identified as SVRs between conditions. The RRE dataset studied the HIV Rev- response element, with three replicates measured in the presence and absence of Rev. The Rev- RRE interaction sites are considered SVRs. We obtained raw read counts and reactivities for the Flu dataset and reactivities for the RRE dataset, based on data availability. + +<|ref|>text<|/ref|><|det|>[[143, 648, 858, 869]]<|/det|> +Synthetic negative control datasets were constructed from real SP data to assess the specificity of DiffScan and existing methods. For construction, we randomly split replicates in a single condition into two groups, and identified SVRs by contrasting the two groups. Any reported SVRs were considered false positive results. In particular, control datasets 1- 4 were obtained by randomly splitting samples in the Flu dataset in the absence (Control 1) and presence (Control 2) of fluoride, and the RRE dataset with (Control 3) and without Rev (Control 4). To include more SP platforms in evaluation, we constructed another two control datasets from an icSHAPE dataset of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 855, 143]]<|/det|> +mouse SRP RNA (272 nt) \(^{10}\) , by contrasting the two in vitro samples (Control 5) and the two in vivo samples (Control 6). + +<|ref|>text<|/ref|><|det|>[[143, 166, 857, 390]]<|/det|> +To apply DiffScan in transcriptome level analysis, we downloaded the icSHAPE dataset \(^{17}\) in Sun et al., which mapped RNA secondary structure across HEK293 cellular compartments including chromatin, nucleoplasm, and cytoplasm. Transcripts with RPKM less than 5 or RT stop less than 2 for all nucleotide positions were excluded \(^{17,49}\) . Then nucleotide positions with coverage less than 200 were removed. As a result, 1,352 transcripts (covering 826,917 nucleotide positions) were compared between chromatin and nucleoplasm, and 1,851 transcripts (covering 1,220,273 nucleotide positions) were compared between nucleoplasm and cytoplasm. + +<|ref|>sub_title<|/ref|><|det|>[[145, 416, 315, 433]]<|/det|> +## Normalization module + +<|ref|>text<|/ref|><|det|>[[143, 457, 857, 586]]<|/det|> +Suppose that there are \(\mathsf{n}_{\mathrm{A}}\) replicates from condition A, and the reactivity of nucleotide position j in replicate i is \(\mathrm{r}_{\mathrm{ij}}^{\mathrm{A}},1\leq \mathrm{i}\leq \mathrm{n}_{\mathrm{A}},1\leq \mathrm{j}\leq \mathrm{n}\) , where n is the length of the transcript. Similarly, for \(\mathsf{n}_{\mathrm{B}}\) replicates from condition B, the reactivity of nucleotide position j in replicate k is \(\mathrm{r}_{\mathrm{kj}}^{\mathrm{B}},1\leq \mathrm{k}\leq \mathrm{n}_{\mathrm{B}},1\leq \mathrm{j}\leq \mathrm{n}\) . We normalized reactivities using the following steps. + +<|ref|>text<|/ref|><|det|>[[143, 609, 857, 731]]<|/det|> +Step 1: \(90\%\) winsorization is applied to the reactivities in each replicate separately to remove outliers, i.e., the bottom \(5\%\) of reactivities are set to the 5th percentile while the upper \(5\%\) of reactivities are set to the 95th percentile. After that, reactivities are scaled into range [0,1] by subtracting the minimum and then dividing the result by the new maximum. + +<|ref|>text<|/ref|><|det|>[[144, 755, 855, 809]]<|/det|> +Step 2: Quantile normalization is applied to within- condition replicates to match the quantiles, which is frequently used in normalization analysis of genome- wide assays \(^{30,50}\) . + +<|ref|>text<|/ref|><|det|>[[144, 833, 855, 887]]<|/det|> +Step 3: To make between- condition reactivities comparable, we first determine a structurally invariant set \(S\) , i.e., nucleotide positions that do not exhibit structural variation across conditions. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 857, 245]]<|/det|> +The reactivities are normalized so that the adjusted reactivities are similar between conditions in the structurally invariant set. The determination of \(S\) is described in Supplementary Methods. Then we learn a linear transformation rule from \(S\) , which converts reactivities of one condition to the same level of those of the other condition in \(S\) , and extrapolate the transformation to all nucleotide positions to obtain normalized reactivities. + +<|ref|>text<|/ref|><|det|>[[143, 269, 844, 394]]<|/det|> +We learn how to transform between- condition reactivities to the same level in \(S\) by fitting the following robust regression utilizing iterated re- weighted least squares with Huber's \(M\) estimate \(^{51}\) . (For the sake of simplicity, we will still use \(r_{ij}^{\mathrm{A}}, r_{ij}^{\mathrm{B}}\) to denote reactivities processed by Step 1 and Step 2. ) + +<|ref|>equation<|/ref|><|det|>[[381, 406, 615, 431]]<|/det|> +\[\log \overline{r_{j}^{\mathrm{A}}}\sim \log \alpha +\beta \log \overline{r_{j}^{\mathrm{B}}},j\in S,\] + +<|ref|>text<|/ref|><|det|>[[143, 444, 848, 620]]<|/det|> +where \(\log \alpha\) corrects for the difference in sequencing depth and \(\beta\) corrects for the difference in signal- to- noise ratio. Given that within- condition replicates are normalized by quantile normalization, it follows that \(\overline{r_{ij}^{\mathrm{A}}}\approx \overline{r_{kj}^{\mathrm{B}}}\) in \(S\) if we transform \(r_{kj}^{\mathrm{B}}\) into \(\overline{r_{kj}^{\mathrm{B}}} = \exp \left(\overline{\log\alpha} +\beta \log r_{kj}^{\mathrm{B}}\right)\) and let \(\overline{r_{ij}^{\mathrm{A}}} = r_{ij}^{\mathrm{A}},j\in S\) . Based on this, we extrapolate the fitted model to all nucleotide positions to obtain normalized reactivities \(\overline{r_{ij}^{\mathrm{A}}} = r_{ij}^{\mathrm{A}}\) and \(\overline{r_{kj}^{\mathrm{B}}} = \exp \left(\overline{\log\alpha} +\beta \log r_{kj}^{\mathrm{B}}\right),1\leq j\leq n\) . + +<|ref|>sub_title<|/ref|><|det|>[[145, 644, 455, 662]]<|/det|> +## Differential Analysis at the Position Level + +<|ref|>text<|/ref|><|det|>[[143, 686, 857, 807]]<|/det|> +With normalized reactivities as input, we first test position- level differences between conditions to derive positional p values. To encompass various SP platforms, a nonparametric Wilcoxon test is used, which evaluates the structural variation at nucleotide position \(j\) by contrasting reactivities between conditions in a small window surrounding \(j\) . Technically, we define + +<|ref|>equation<|/ref|><|det|>[[233, 828, 763, 858]]<|/det|> +\[\mathrm{p}_{\mathrm{j}} = \mathrm{Wilcoxon~test}\left(\{\overline{r_{\mathrm{it}}^{\mathrm{A}}},1\leq \mathrm{i}\leq \mathrm{n}_{\mathrm{A}},\mathrm{t}\in \mathrm{C}_{\mathrm{j}}\} ,\{\overline{r_{\mathrm{kt}}^{\mathrm{B}}},1\leq \mathrm{k}\leq \mathrm{n}_{\mathrm{B}},\mathrm{t}\in \mathrm{C}_{\mathrm{j}}\} \right),\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 855, 148]]<|/det|> +where \(\overrightarrow{\mathrm{r}}_{\mathrm{it}}^{\mathrm{A}},\overrightarrow{\mathrm{r}}_{\mathrm{kt}}^{\mathrm{B}}\) are normalized reactivities, \(\mathrm{C_j}\) is a small window centering at nucleotide position j with radius r, and \(\mathsf{p}_{\mathrm{j}}\) is the two- sided p value. By default, we set r to 2 nt. + +<|ref|>sub_title<|/ref|><|det|>[[144, 174, 542, 192]]<|/det|> +## Summarizing Position-level Signals via Scan Statistics + +<|ref|>text<|/ref|><|det|>[[144, 217, 855, 271]]<|/det|> +To concatenate position- level signals into regional signals, a scan statistic is defined for each region R, + +<|ref|>equation<|/ref|><|det|>[[411, 294, 589, 340]]<|/det|> +\[\mathrm{Q}(\mathrm{R}) = \frac{-\Sigma_{\mathrm{j}\in \mathrm{R}}\log(\mathrm{p}_{\mathrm{j}})}{\sqrt{|\mathrm{R}|}}.\] + +<|ref|>text<|/ref|><|det|>[[144, 362, 855, 450]]<|/det|> +With penalty of region length \(|\mathrm{R}|\) , extending a candidate region will accumulate position- level signals at the expense of penalization. As a result, SVRs with accurately mapped boundaries can be distinguished based on the magnitude of the scan statistic. + +<|ref|>sub_title<|/ref|><|det|>[[144, 476, 460, 494]]<|/det|> +## Statistical Inference for Detection of SVRs + +<|ref|>text<|/ref|><|det|>[[144, 519, 856, 610]]<|/det|> +In view of the dynamic locations and lengths of SVRs, we scan the transcripts by enumerating all candidate regions with suitable lengths, i.e., every region in \(\mathcal{R} = \{\mathrm{R} | \mathrm{L}_{\mathrm{min}} \leq |\mathrm{R}| \leq \mathrm{L}_{\mathrm{max}}\}\) . To determine whether a region \(\mathrm{R}\) is selected as an SVR, we test + +<|ref|>equation<|/ref|><|det|>[[395, 621, 603, 643]]<|/det|> +\[\mathrm{H}_0^{\mathrm{R}}\colon \mathrm{p}_{\mathrm{j}}\sim \mathrm{i.i.d.U}(0,1),\mathrm{j}\in \mathrm{R}\] + +<|ref|>text<|/ref|><|det|>[[144, 657, 855, 710]]<|/det|> +based on the magnitude of \(\mathrm{Q}(\mathrm{R})\) . We implemented a Monte Carlo approach to control the familywise error rate for DiffScan (see Supplementary Methods). + +<|ref|>sub_title<|/ref|><|det|>[[144, 735, 236, 751]]<|/det|> +## Simulations + +<|ref|>text<|/ref|><|det|>[[144, 778, 724, 797]]<|/det|> +We simulated reactivities for two conditions through the following three steps. + +<|ref|>text<|/ref|><|det|>[[144, 821, 856, 910]]<|/det|> +Step 1. 100 human RNA sequences in an icSHAPE dataset17 were randomly selected. For each RNA sequence, we sampled 10 conformations from the Boltzmann distribution of secondary structure52 utilizing the RNAsubopt program in ViennaRNA version 2.4.1553. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 857, 211]]<|/det|> +Step 2. Conditional on the pairing status of each nucleotide in a transcript, the reactivities can be sampled from pre- trained reactivity distributions. Three distributions were used: Cordero et al. \(^{35}\) and Sükösd et al. \(^{36}\) as fitted from SHAPE data, and also reactivity distributions we fitted from icSHAPE data (Supplementary Methods). + +<|ref|>text<|/ref|><|det|>[[142, 234, 857, 777]]<|/det|> +Step 3. Different biological conditions were characterized by differential compositions of the 10 conformations generated in Step 1. The reactivities of each conformation were linearly combined accordingly to generate the observed reactivities in each condition. Specifically, for condition A, the 10 conformations of a transcript were assigned with weights \(w_{\mathrm{C}}^{\mathrm{A}}\) , \(1 \leq c \leq 10\) , with \(w_{\mathrm{C}}^{\mathrm{A}}\) corresponding to the weight of the \(c^{\mathrm{th}}\) conformation in condition A, with \(0 \leq w_{\mathrm{C}}^{\mathrm{A}} \leq 1\) , \(\Sigma_{c = 1}^{10} w_{\mathrm{C}}^{\mathrm{A}} = 1\) . For condition B, the weight of the \(c^{\mathrm{th}}\) conformation was \(w_{\mathrm{B}}^{\mathrm{B}}\) , with \(0 \leq w_{\mathrm{B}}^{\mathrm{B}} \leq 1\) , \(\Sigma_{c = 1}^{10} w_{\mathrm{B}}^{\mathrm{B}} = 1\) . To simulate SVRs that cover a reasonable proportion (approximately \(20\%\) \(^{16,17}\) ) of all nucleotide positions of the 100 transcripts, we allocated \(90\%\) of the weight to the first two conformations and randomly distributed the remaining \(10\%\) to the other eight conformations. We then generated reactivities in condition A and B separately by changing the weights of the first two conformations. Thus, nucleotide positions with different pairing status between the first two conformations form SVRs. Furthermore, replicates were simulated by adding random noise \((N(0,0.1^{2}))\) to the weights of the first two conformations. Finally, simulations were conducted at three levels of signal strength respectively by setting (i) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.2, 0.7)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.7, 0.2)\) , (ii) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.1, 0.8)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.8, 0.1)\) , and (iii) \((w_{1}^{\mathrm{A}}, w_{2}^{\mathrm{A}}) = (0.9, 0.9)\) , \((w_{1}^{\mathrm{B}}, w_{2}^{\mathrm{B}}) = (0.9, 0)\) . + +<|ref|>sub_title<|/ref|><|det|>[[144, 797, 497, 815]]<|/det|> +## Performance Metrics for Validation of DiffScan + +<|ref|>text<|/ref|><|det|>[[144, 840, 855, 893]]<|/det|> +In the simulated datasets, we evaluated the performance of different methods with two metrics. First, to evaluate how accurately simulated SVRs (i.e., true SVRs to detect between the compared + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 88, 857, 312]]<|/det|> +conditions in the simulated datasets) were mapped by predicted SVRs by a method, we took out 1,000 nucleotide positions in the top- ranked SVRs by the method. The distance from each of the nucleotide positions to simulated SVRs was calculated. Second, we calculated the Precision- Recall curve. At varying significance levels, the value of precision was calculated as the number of correctly predicted nucleotide positions divided by the number of all predicted nucleotide positions. The value of recall rate was calculated as the number of correctly predicted nucleotide positions divided by the number of all nucleotide positions in simulated SVRs. + +<|ref|>text<|/ref|><|det|>[[144, 336, 857, 458]]<|/det|> +In the negative control datasets (Control 1- 6), to access whether the false positive discoveries of a method can be controlled at a nominal level, we calculated the position- level false positive rate as the number of predicted nucleotide positions at significance level 0.05 divided by the length of the transcript. + +<|ref|>text<|/ref|><|det|>[[144, 483, 856, 569]]<|/det|> +In the benchmark datasets with annotated SVRs (Flu and RRE)—similar to our processing of the simulated datasets—we calculated (1) the average distance from 10 nucleotide positions in top predicted SVRs to annotated SVRs and (2) the Precision- Recall curve. + +<|ref|>sub_title<|/ref|><|det|>[[146, 595, 421, 612]]<|/det|> +## Implementation of existing methods + +<|ref|>text<|/ref|><|det|>[[144, 637, 857, 760]]<|/det|> +The software deltaSHAPE version 1.0 available at the Weeks lab website22 was utilized to implement deltaSHAPE. As far as we know, there is no official software release for PARCEL or RASA; and we implemented them utilizing custom scripts from existing literature23. The R package dStruct version 1.0.0 was utilized to implement dStruct. + +<|ref|>sub_title<|/ref|><|det|>[[146, 785, 355, 802]]<|/det|> +## Application to icSHAPE data + +<|ref|>sub_title<|/ref|><|det|>[[146, 829, 308, 845]]<|/det|> +## Reactivity calculation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 380]]<|/det|> +To calculate reactivities, we assumed that \(\mathrm{RT_j \sim Binomial(N_j, p_j)}\) , in which \(\mathrm{RT_j}\) is the RT stop at position j in the case experiment, \(\mathrm{N_j}\) is the number of times that position j is exposed to the probing molecules, and \(\mathrm{p_j}\) is the probability that position j is modified. The maximum likelihood estimator for \(\mathrm{p_j}\) is \(\hat{\mathrm{p_j}} = \frac{\mathrm{RT_j}}{\mathrm{N_j}}\) . On the other hand, the coverage at position j in the control experiment (coveragejcontrol) is approximately proportional to \(\mathrm{N_j}\) . Therefore, we calculated the reactivity at position j as \(\mathrm{r_j} = \frac{\mathrm{RT_j}}{\mathrm{coverage_j^{control}}} ^{54}\) . Based on this, two count replicates from case experiments were paired with two count replicates from control experiments to produce two reactivity replicates under each condition. + +<|ref|>sub_title<|/ref|><|det|>[[145, 404, 461, 422]]<|/det|> +## Enrichment analysis of RBP binding motifs + +<|ref|>text<|/ref|><|det|>[[143, 446, 856, 570]]<|/det|> +We collected 154 RBP binding motifs from the CISBP- RNA Database \(^{55}\) and the ATTRACT database \(^{56}\) . For each motif, we searched in the predicted SVRs for significant motif hits utilizing the FIMO module (--norc --thresh 0.001) from the MEME suite \(^{40}\) , and the number of significant hits in each SVR were counted. + +<|ref|>text<|/ref|><|det|>[[143, 594, 857, 851]]<|/det|> +To evaluate the significance of motif enrichment, we randomly sampled null regions in non- SVR regions in the same transcripts, with region length matched to the predicted SVRs. The number of significant hits in each null region is recorded. After that, we performed a one- sided Wilcoxon signed- rank test for the two count vectors of significant hits to evaluate the significance of enrichment. Motifs with corrected p value (Benjamini- Hochberg method) less than 0.05 were considered significantly enriched. We further consolidated the result by filtering out motifs that lost enrichment, i.e., with corrected p value (Benjamini- Hochberg method) greater than 0.01, when the null regions were sampled from the non- SVR regions of all transcripts with SVRs. + +<|ref|>text<|/ref|><|det|>[[144, 876, 742, 895]]<|/det|> +Inference for the relationships among SVRs, RBP binding, and mRNA abundance + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 551]]<|/det|> +We considered four models of which each explicitly describes the relationships among SVRs, changes in RBP binding, and changes in mRNA abundance (Model 1- 4 in Supplementary Fig. 8). For model selection, three variables of each mRNA transcript in the Np versus Cy comparison were used: (1) whether DiffScan predicted any SVRs in the Np versus Cy comparison for the transcript (svr; \(svr = "yes"\) or "no"), which quantifies its structural variations between Np and Cy, (2) \(\log_{10}\) (fold change of RPKM; Cy/Np) of the transcript (log10- fold- change- of- RPKM), which quantifies the change in its abundance between Np and Cy, and (3) a change- in- RBP- binding variable quantifying the changes in the binding of a specified RBP between Np and Cy for the transcript. Because QKI predominantly localizes in the Np, the change- in- RBP- binding variable for QKI of a transcript can be approximated by a score quantifying the level of QKI binding in the Np for the transcript. To calculate the score, we searched in the transcript for significant hits of QKI motifs utilizing the FIMO module (--norc --thresh 0.001), and the number of significant motif hits was taken as the score quantifying the level of QKI binding in the Np for the transcript. Similar processing was implemented for IGF2BP3, because it is almost exclusively cytoplasmic. + +<|ref|>text<|/ref|><|det|>[[142, 574, 857, 830]]<|/det|> +Given the three variables of each transcript, i.e., \(svr\) , \(\log_{10}\) - fold- change- of- RPKM, and change- in- RBP binding for a specified RBP, we tested conditional independence between \(svr\) and \(\log_{10}\) - fold- change- of- RPKM given change- in- RBP- binding utilizing the conditional independence test for hybrid Bayesian networks implemented in the bnlearn R package57. In Model 1- 2, \(svr\) and \(\log_{10}\) - fold- change- of- RPKM are conditionally independent given change- in- RBP- binding, whereas \(svr\) regulates \(\log_{10}\) - fold- change- of- RPKM given change- in- RBP- binding in Model 3- 4. In cases where Model 1- 2 are rejected by the conditional independence test, Model 3- 4 are supported. However, Model 3 and Model 4 are statistically equivalent with current data. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 92, 279, 108]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[144, 136, 857, 258]]<|/det|> +Data availabilityAll data used in the manuscript were publicly available. The benchmark datasets were downloaded from https://doi.org/10.5281/zenodo.2536501. The icSHAPE dataset for transcriptome analysis were downloaded from the GEO database with accession code GSE117840 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE117840). + +<|ref|>sub_title<|/ref|><|det|>[[145, 328, 282, 345]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[144, 373, 857, 425]]<|/det|> +Code availabilityAll codes and the developed software were publicly available at https://github.com/yub18/DiffScan. + +<|ref|>sub_title<|/ref|><|det|>[[145, 497, 237, 513]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[142, 542, 855, 882]]<|/det|> +References1. Licatalosi, D.D. & Darnell, R.B. RNA processing and its regulation: global insights into biological networks. \*Nature Reviews Genetics\* 11, 75- 87 (2010).2. Cech, T.R. The RNA worlds in context. \*Cold Spring Harb Perspect Biol\* 4, a006742 (2012).3. Cech, T.R. & Steitz, J.A. The noncoding RNA revolution- trashing old rules to forge new ones. \*Cell\* 157, 77- 94 (2014).4. Yu, B., Lu, Y., Zhang, Q.C. & Hou, L. Prediction and differential analysis of RNA secondary structure. \*Quantitative Biology\* 8, 109- 118 (2020).5. Strobel, E.J., Yu, A.M. & Lucks, J.B. High- throughput determination of RNA structures. \*Nat Rev Genet\* 19, 615- 634 (2018).6. Li, P., Zhou, X., Xu, K. & Zhang, Q.C. RASP: an atlas of transcriptome- wide RNA secondary structure probing data. \*Nucleic Acids Research\* (2020).7. Loughrey, D., Watters, K.E., Settle, A.H. & Lucks, J.B. SHAPE- Seq 2.0: systematic optimization and extension of high- throughput chemical probing of RNA secondary structure with next generation sequencing. \*Nucleic Acids Research\* 42, 10 (2014).8. Lucks, J.B. et al. Multiplexed RNA structure characterization with selective 2 '- hydroxyl acylation analyzed by primer extension sequencing (SHAPE- Seq). \*Proc. Natl. Acad. Sci. U. S. A.\* 108, 11063- 11068 (2011).9. Rouskin, S., Zubradt, M., Washietl, S., Kellis, M. & Weissman, J.S. Genome- wide probing of RNA structure reveals active unfolding of mRNA structures in vivo. \*Nature\* 505, 701- 705 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 849, 904]]<|/det|> +10. Spitale, R.C. et al. Structural imprints in vivo decode RNA regulatory mechanisms. Nature 519, 486-490 (2015).11. Andrzejewska, A., Zawadzka, M. & Pachulska-Wieczorek, K. On the Way to Understanding the Interplay between the RNA Structure and Functions in Cells: A Genome-Wide Perspective. Int J Mol Sci 21 (2020).12. Mustoe, A.M., Brooks, C.L. & Al-Hashimi, H.M. Hierarchy of RNA functional dynamics. Annu Rev Biochem 83, 441-466 (2014).13. Dambach, M. et al. The Ubiquitous yybP-ykoY Riboswitch Is a Manganese-Responsive Regulatory Element. Mol. Cell 57, 1099-1109 (2015).14. Mayr, C. Regulation by 3'-Untranslated Regions. Annu Rev Genet 51, 171-194 (2017).15. Mayya, V.K. & Duchaine, T.F. Ciphers and Executioners: How 3'-Untranslated Regions Determine the Fate of Messenger RNAs. Frontiers in Genetics 10 (2019).16. Shi, B. et al. RNA structural dynamics regulate early embryogenesis through controlling transcriptome fate and function. Genome Biol 21, 120 (2020).17. Sun, L. et al. RNA structure maps across mammalian cellular compartments. Nat Struct Mol Biol 26, 322-330 (2019).18. Aw, Jong Ghut A. et al. In Vivo Mapping of Eukaryotic RNA Interactomes Reveals Principles of Higher-Order Organization and Regulation. Mol. Cell 62, 603-617 (2016).19. Tapsin, S. et al. Genome-wide identification of natural RNA aptamers in prokaryotes and eukaryotes. Nat Commun 9, 1289 (2018).20. Mizrahi, O. et al. Virus-Induced Changes in mRNA Secondary Structure Uncover cis-Regulatory Elements that Directly Control Gene Expression. Mol Cell 72, 862-874 e865 (2018).21. Wan, Y. et al. Landscape and variation of RNA secondary structure across the human transcriptome. Nature 505, 706-709 (2014).22. Smola, M.J., Calabrese, J.M. & Weeks, K.M. Detection of RNA-Protein Interactions in Living Cells with SHAPE. Biochemistry 54, 6867-6875 (2015).23. Choudhary, K., Lai, Y.H., Tran, E.J. & Aviar, S. dStruct: identifying differentially reactive regions from RNA structureome profiling data. Genome Biol 20, 40 (2019).24. Strobel, E.J., Watters, K.E., Nedialkov, Y., Artsimovitch, I. & Lucks, J.B. Distributed biotin-streptavidin transcription roadblocks for mapping cotranscriptional RNA folding. Nucleic Acids Research 45, e109-e109 (2017).25. Kutchko, K.M. & Laederach, A. Transcending the prediction paradigm: novel applications of SHAPE to RNA function and evolution. Wiley Interdisciplinary Reviews: RNA 8 (2017).26. Choudhary, K., Deng, F. & Aviar, S. Comparative and integrative analysis of RNA structural profiling data: current practices and emerging questions. Quant Biol 5, 3-24 (2017).27. Wang, X.W., Liu, C.X., Chen, L.L. & Zhang, Q.C. RNA structure probing uncovers RNA structure-dependent biological functions. Nat Chem Biol 17, 755-766 (2021).28. Shao, Z., Zhang, Y., Yuan, G.C., Orkin, S.H. & Waxman, D.J. MAnorm: a robust model for quantitative comparison of ChIP-Seq data sets. Genome Biol 13, R16 (2012).29. Xiang, G. et al. S3norm: simultaneous normalization of sequencing depth and signal-to-noise ratio in epigenomic data. Nucleic Acids Res (2020).30. Bolstad, B.M., Irizarry, R.A., Astrand, M. & Speed, T.P. A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185-193 (2003).31. Kulldorff, M. A spatial scan statistic. Communications in Statistics - Theory and Methods 26, 1481-1496 (1997). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 856, 888]]<|/det|> +32. Glaz, J. Scan statistics. (New York : Springer, New York; 2001).33. Durand, D. & Sankoff, D. Tests for gene clustering. J Comput Biol 10, 453-482 (2003).34. Guo, H., Li, J.J., Lu, Q. & Hou, L. Detecting local genetic correlations with scan statistics. Nature Communications 12, 2033 (2021).35. Cordero, P., Kladwang, W., VanLang, C.C. & Das, R. Quantitative dimethyl sulfate mapping for automated RNA secondary structure inference. Biochemistry 51, 7037-7039 (2012).36. Sukosd, Z., Swenson, M.S., Kjems, J. & Heitsch, C.E. Evaluating the accuracy of SHAPE-directed RNA secondary structure predictions. Nucleic Acids Res 41, 2807-2816 (2013).37. Kortmann, J. & Narberhaus, F. Bacterial RNA thermometers: molecular zippers and switches. Nature Reviews Microbiology 10, 255-265 (2012).38. Li, H. & Aviiran, S. Statistical modeling of RNA structure profiling experiments enables parsimonious reconstruction of structure landscapes. Nat Commun 9, 606 (2018).39. Bai, Y., Tambe, A., Zhou, K. & Doudna, J.A. RNA-guided assembly of Rev-RRE nuclear export complexes. Elife 3, e03656 (2014).40. Bailey, T.L. et al. MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res 37, W202-208 (2009).41. Thul, P.J. et al. A subcellular map of the human proteome. Science 356 (2017).42. Teplova, M. et al. Structure-function studies of STAR family Quaking proteins bound to their in vivo RNA target sites. Genes Dev 27, 928-940 (2013).43. Aberg, K., Saetre, P., Jareborg, N. & Jazin, E. Human QKI, a potential regulator of mRNA expression of human oligodendrocyte-related genes involved in schizophrenia. Proc Natl Acad Sci U S A 103, 7482-7487 (2006).44. Lu, H. et al. QKI regulates adipose tissue metabolism by acting as a brake on thermogenesis and promoting obesity. EMBO Rep 21, e47929 (2020).45. Wachter, K., Kohn, M., Stohr, N. & Huttelmaier, S. Subcellular localization and RNP formation of IGF2BPs (IGF2 mRNA-binding proteins) is modulated by distinct RNA-binding domains. Biol Chem 394, 1077-1090 (2013).46. Li, H. et al. Runting and Stunting Syndrome Is Associated With Mitochondrial Dysfunction in Sex-Linked Dwarf Chicken. Frontiers in Genetics 10 (2020).47. Watters, K.E., Strobel, E.J., Yu, A.M., Lis, J.T. & Lucks, J.B. Cotranscriptional folding of a riboswitch at nucleotide resolution. Nat Struct Mol Biol 23, 1124-1131 (2016).48. Morandi, E. et al. Genome-scale deconvolution of RNA structure ensembles. Nat. Methods 18, 249-252 (2021).49. Flynn, R.A. et al. Transcriptome-wide interrogation of RNA secondary structure in living cells with icSHAPE. Nat. Protoc. 11, 273-290 (2016).50. Amaratunga, D. & Cabrera, J. Analysis of Data From Viral DNA Microchips. Journal of the American Statistical Association 96, 1161-1170 (2001).51. McKean, J.W. Robust Analysis of Linear Models. Statist. Sci. 19, 562-570 (2004).52. Ding, Y. A statistical sampling algorithm for RNA secondary structure prediction. Nucleic Acids Research 31, 7280-7301 (2003).53. Lorenz, R. et al. ViennaRNA Package 2.0. Algorithms. Mol. Biol. 6, 26 (2011).54. Selega, A., Sirocchi, C., Iosub, I., Granneman, S. & Sanguinetti, G. Robust statistical modeling improves sensitivity of high-throughput RNA structure probing experiments. Nat Methods 14, 83-89 (2017).55. Ray, D. et al. A compendium of RNA-binding motifs for decoding gene regulation. Nature 499, 172-177 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 839, 160]]<|/det|> +56. Giudice, G., Sánchez-Cabo, F., Torroja, C. & Lara-Pezzi, E. ATTRACT-a database of RNA-binding proteins and associated motifs. Database (Oxford) 2016 (2016).57. Scutari, M. Bayesian Network Constraint-Based Structure Learning Algorithms: Parallel and Optimized Implementations in the bnlearn R Package. 2017 77, 20 (2017). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 354, 150]]<|/det|> +SupplementaryInformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/images_list.json b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..8e40faf09da8ed153a44a6924c72a0f294c6630d --- /dev/null +++ b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 Overall structure of EndoS in complex with the Fc region. A EndoS, SpeB and IdeS are secreted enzymes from S. pyogenes that can act on IgG antibodies. SpeB and IdeS are proteases that hydrolyze the flexible hinge region between residues 220 to 248 of the IgG antibody \\(^{79 - 81}\\) . EndoS is an", + "footnote": [], + "bbox": [ + [ + 120, + 70, + 884, + 860 + ] + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 The Fc binding site of EndoS. Two views of the interaction interface between the GH domain of EndoS (yellow) and FcP1 of the Fc region (orange) (A), the \\(N\\) -glycan (blue and green) of the Fc region and the active site of EndoS (yellow) (B) the Cy2-Cy3 joint region of the Fc (orange) and the b-sandwich domain of EndoS (blue) (C).", + "footnote": [], + "bbox": [ + [ + 120, + 95, + 880, + 582 + ] + ], + "page_idx": 45 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Alanine scanning mutagenesis of the EndoS-Fc interface. A Structure of the EndoS (grey) complex with Fc (pink) in which the regions with alanine mutants are color coded based on the relative rates of removal of the \\(N\\) -glycans from Fc alanine mutants by EndoS and vice versa. The rates were measured via LC-MS kinetic analyses. Red indicates the largest increase in hydrolytic activity while blue indicates no hydrolytic activity. B Rates of removal of the first glycan of Fc and Fc alanine mutants by EndoS alanine mutants and EndoS respectively, normalized to EndoS vs. Fc. All rates were statistically significant (multiple comparison test, Tukey method, \\(p< 0.0001\\) ).", + "footnote": [], + "bbox": [ + [ + 95, + 75, + 830, + 644 + ] + ], + "page_idx": 46 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Molecular dynamics (MD) simulations studies. Overlay of 25 frames evenly spaced along the \\(1\\mu \\mathrm{s}\\) MD trajectory for (A) EndoS-Fc, (B) mutant E293A and (C) mutant W803A, along with the average distance between the center of the aromatic ring of the residues involved in the stabilizing interactions derived from the MD simulations. In the case of L314, S298 and E354 Cδ, the oxygen of -OH and the oxygens of the -CO₂⁻ are considered, respectively. Errors are given as SD. The EndoS protein is shown as blue ribbons and the antibody as gray and orange ribbons.", + "footnote": [], + "bbox": [ + [ + 111, + 100, + 870, + 492 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Molecular mechanism of antibody recognition by EndoS and EndoS2. Kinetic modeling of deglycosylation of (A) FcWT and (B) IgGWT by EndoSWT and (C) IgGWT by EndoS2WT. LC-MS kinetic analysis of deglycosylation of FcWT and IgGWT by EndoSWT- showing the hydrolysis over time of the two \\(N\\) -glycans linked to Asn297. The deglycosylation process occurs sequentially as seen by the initial increase then decrease in monoglycosylated population (dark green) over time while the diglycosylated (purple) and deglycosylated (orange) species increase over time. D Schematic representation of Fc \\(N\\) -glycan processing by EndoS. In a first step, EndoS binds the diglycosylated Fc and hydrolyzes one of the \\(N\\) -glycan of the Fc which causes the release of monoglycosylated Fc. In a second step, EndoS engages the monoglycosylated Fc and hydrolyzes the remaining \\(N\\) -glycan of the Fc to produced completely deglycosylated Fc.", + "footnote": [], + "bbox": [ + [ + 120, + 68, + 880, + 602 + ] + ], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 Binding affinity and enzyme kinetics experiments studied by NMR. A Crystal structure of EndoS2 bound to CT \\(N\\) -glycan and \\(\\mathrm{Ca}^{2 + }\\) (PDB code 6MDS) showing the location of the \\(\\left[^{13}\\mathrm{C},\\mathrm{H}_{3}\\right]\\) - methyl labeled probes as black spheres. Glycoside hydrolase (GH) and \\(\\beta\\) -sandwich (BS) domains are colored in yellow and pale blue, respectively. Mutated amino acid E186L, CT glycan and \\(\\mathrm{Ca}^{2 + }\\) ion are depicted in pink, dark blue and pale green. B Superimposition of selected regions of methyl-TROSY spectra showing CSPs upon mutagenesis or at saturation concentrations of different ligands. Color code corresponds to: EndoS2 (violet), EndoS2E186L in the absence (black) and in the presence of CT (red), \\(\\mathrm{CT}_{\\mathrm{n - 1}}\\) (orange), IgG1 Fc (blue) and \\(\\mathrm{Ca}^{2 + }\\) (yellow). Addition of \\(\\mathrm{Fc}_{\\mathrm{aglyc}}\\) (green) induced to CSPs, suggesting very weak protein-protein interactions. Arrows indicate the extent of CSPs observed between the different protein states. C Ligands used for NMR titration experiments. IgG1 Fc region (Fc), complex-type carbohydrate (CT) and the products of glycan hydrolysis \\(\\mathrm{Fc}_{\\mathrm{aglyc}}\\) and \\(\\mathrm{CT}_{\\mathrm{n - 1}}\\) , respectively. Color code like in Fig. 1. D Quantum mechanics-based 2D lineshape fitting of", + "footnote": [], + "bbox": [ + [ + 115, + 68, + 880, + 691 + ] + ], + "page_idx": 49 + } +] \ No newline at end of file diff --git a/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b.mmd b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b.mmd new file mode 100644 index 0000000000000000000000000000000000000000..aa447b35dcb97d81b6bb2fe016a7d131d3b549ee --- /dev/null +++ b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b.mmd @@ -0,0 +1,454 @@ + +# Mechanism of antibody-specific deglycosylation and immune evasion by Streptococcus IgG-specific endoglycosidases + +Beatrix Trastoy CIC bioGUNE https://orcid.org/0000- 0002- 2178- 732X + +Jonathan Du Emory University + +Javier Cifuente IIS Biocruces https://orcid.org/0000- 0001- 5420- 121X + +Lorena Rudolph Universitat zu Lubeck + +Mikel Garcia- Alija CSIC- UPV + +Erik Klontz University of Maryland, Baltimore + +Daniel Deredge University of Maryland School of Pharmacy https://orcid.org/0000- 0002- 6897- 6523 + +Nazneen Sultana Emory University https://orcid.org/0000- 0001- 9866- 4443 + +Chau Huynh Emory University + +Maria Flowers Emory University + +Chao Li University of Maryland + +Diego Sastre Emory University + +Lai- Xi Wang University of Maryland College Park https://orcid.org/0000- 0003- 4293- 5819 + +Francisco Corzana University of La Rioja https://orcid.org/0000- 0001- 5597- 8127 + +Alvaro Mallagaray Institute of Chemistry and Metabolomics, Center of Structural and Cell Biology in Medicine (CSCM), University of Lubeck https://orcid.org/0000- 0001- 5825- 4407 + +<--- Page Split ---> + +Eric J. Sundberg Emory University https://orcid.org/0000- 0003- 0478- 3033 Dr. Marcelo E. Guerin (☑ mrcguerin@gmail.com ) IIS BioCruces Bizkaia https://orcid.org/0000- 0001- 9524- 3184 + +## Article + +# Keywords: + +Posted Date: June 22nd, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1774503/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Mechanism of antibody-specific deglycosylation and immune evasion by Streptococcus IgG-specific endoglycosidases + +Beatrix Trastoy, \(^{1,2,\# *}\) Jonathan J. Du, \(^{3,\#}\) Javier O. Cifuente, \(^{1,2,\#}\) Lorena Rudolph, \(^{4}\) Mikel García- Alija, \(^{1,2}\) Erik H. Klontz, \(^{5,6}\) Daniel Deredge, \(^{7}\) Nazneen Sultana, \(^{3}\) Chau G. Huynh, \(^{3}\) Maria Flowers, \(^{3}\) Chao Li, \(^{8}\) Diego E. Sastre, \(^{3}\) Lai- Xi Wang, \(^{8}\) Francisco Corzana, \(^{9}\) Alvaro Mallagaray \(^{4*}\) , Eric J. Sundberg \(^{3*}\) and Marcelo E. Guerin \(^{1,2,10*}\) + +\(^{1}\) Structural Glycobiology Laboratory, Biocrudes Health Research Institute, Barakaldo, Bizkaia, 48903, Spain \(^{2}\) Structural Glycobiology Laboratory, Center for Cooperative Research in Biosciences (CIC bioGUNE), Basque Research and Technology Alliance (BRTA), Bizkaia Technology Park, Building 801A, 48160 Derio, Spain \(^{3}\) Department of Biochemistry, Emory University School of Medicine, Atlanta, GA 30322, USA \(^{4}\) University of Lübeck, Center of Structural and Cell Biology in Medicine (CSCM), Institute of Chemistry and Metabolomics, Ratzeburger Allee 160, 23562 Lübeck, Germany \(^{5}\) Department of Microbiology and Immunology, University of Maryland School of Medicine, Baltimore, MD 21201, USA \(^{6}\) Institute of Human Virology, University of Maryland School of Medicine, Baltimore, MD 21201, USA \(^{7}\) Department of Pharmaceutical Sciences, University of Maryland School of Pharmacy, Baltimore MD 21201, USA \(^{8}\) Department of Chemistry and Biochemistry, University of Maryland, College Park, MD 20742, USA \(^{9}\) Departamento Química and Centro de Investigación en Síntesis Química, Universidad de La Rioja, 26006 Rioja, Spain \(^{10}\) Ikerbasque, Basque Foundation for Science, 48009 Bilbao, Spain + +\(^{#}\) These authors contributed equally to this work + +\*Corresponding authors. Email: beatriz.trastoy@gmail.com, alvaro.mallagaraydebenito@uniluebeck.de, eric.sundberg@emory.edu and mrcguerin@gmail.com + +<--- Page Split ---> + +## Abstract + +Bacterial pathogens have evolved diverse and intricate mechanisms to evade the human immune system, including the production of immunomodulatory enzymes. Streptococcus pyogenes secretes two multimodular endo- \(\beta\) - N- acetylglucosaminidases, EndoS and EndoS2, that specifically deglycosylate the conserved \(N\) - linked glycan at Asn297 on IgG Fc, disabling antibody- mediated effector functions. Amongst thousands of known carbohydrate- active enzymes, EndoS and EndoS2 represent just a handful of enzymes that are specific to the protein portion of the glycoprotein substrate, not just the glycan component. Here, we present the cryoEM structure of EndoS in complex with the IgG1 Fc fragment. In combination with small- angle X- ray scattering, alanine scanning mutagenesis, hydrolytic activity measurements, enzyme kinetics, nuclear magnetic resonance and molecular dynamics analysis, we establish the mechanisms of recognition and specific deglycosylation of IgG antibodies by EndoS and EndoS2. Our results provide a rational basis from which to engineer novel enzymes with antibody and glycan selectivity for clinical and biotechnological applications. + +<--- Page Split ---> + +## Introduction + +Streptococcus pyogenes (group A Streptococcus; GAS) is a Gram- positive pathogenic bacterium that causes life- threating conditions and post- infection immune- related diseases, in addition to mild skin and upper respiratory tract infections \(^{1}\) . In order to evade the immune system and facilitate the colonization of the host, GAS employs a large variety of mechanisms, most notably the secretion of enzymes that selectively inactivate key molecules of the immune system, such as immunoglobulin G (IgG) antibodies (Fig. 1A) \(^{2}\) . EndoS \(^{3}\) and EndoS \(^{2}\) \(^{4}\) are endo- \(\beta\) - N- acetylglucosaminidases (ENGases) produced by GAS serotypes M1 and M49, respectively, that hydrolyze the Asn297- linked \(N\) - glycan on the fragment crystallizable (Fc) region of IgG antibodies. These ENGases are the most prominent members of a rare class of carbohydrate- active enzymes that are specific to the protein component of the glycoprotein substrate and not just the glycan component; no mechanism of protein specificity by this class of enzymes has yet been described. The Asn297- linked \(N\) - glycan is required for IgG binding to Fc \(\gamma\) receptors (Fc \(\gamma\) Rs) and complement C1q, which trigger antibody- mediated effector functions \(^{5 - 8}\) . The activity of these enzymes thereby abolishes the effector functions mediated by the Fc region of IgG antibodies, increasing the survival and virulence of the bacteria \(^{9}\) . Accordingly, IgG- specific ENGases have been shown to ameliorate autoimmune disease in diverse animal models \(^{10 - 13}\) . In addition, EndoS and EndoS2 are powerful tools for the chemoenzymatic synthesis of antibodies with homogenous glycoforms \(^{14 - 17}\) and drug conjugates \(^{18,19}\) . The Fc region \(N\) - glycans of IgG antibodies in serum are composed of more than 33 distinct glycoforms \(^{20}\) and their unique chemical structures modulate the effector functions by altering Fc binding to Fc \(\gamma\) Rs, as well as the stability and half- life of the antibody \(^{21}\) . Antibody- based therapeutics, one of the most prominent and expanding class of drugs used for the treatment of diverse human disorders, including cancer, autoimmunity, and infectious diseases \(^{22}\) , are produced with heterogenous glycoforms that impact their safety and efficacy \(^{21}\) . EndoS and EndoS2 are particularly useful for Fc glycan remodeling due to the strict specificity of both enzymes for intact IgG antibodies and the discovery + +<--- Page Split ---> + +of their respective glycosynthase mutants, that are capable of transferring a homogenous \(N\) - glycan to IgG antibodies using a chemoenzymatic approach (Fig. 1B; \(^{23,24}\) ). The generation of antibodies with defined glycoforms is critical not only to understand the role of each glycoform on antibody function but also for the production of antibodies with defined therapeutic, diagnostic and pharmacokinetic properties \(^{25}\) . + +EndoS and EndoS2 share \(37\%\) sequence identity and, although they are both IgG- specific, exhibit distinct \(N\) - glycan substrate specificities. EndoS processes exclusively complex type (CT) \(N\) - glycans on IgG antibodies, while EndoS2 is active on CT, high mannose (HM) type and hybrid type (Hy) glycans on IgG. EndoS is a 108 kDa multi- modular protein composed of six domains which adopt a "V"- shaped structure, from the N- to the C- terminus: (i) an N- terminal three- helix bundle domain (N- 3HB; residues 37- 97); (ii) a glycosidase hydrolase domain (GH; residues 98- 445); (iii) a leucine- rich repeat domain (LRR; residues 446- 631); (iv) a hybrid Ig domain (hIg; residues 632- 764); (v) a \(\beta\) - sandwich domain (residues 765- 923), and (vi) a C- terminal three- helix bundle domain (C- 3HB; residues 924- 995) \(^{26,27}\) . EndoS2 is a 95 kDa multi- modular protein that shares a similar overall architecture to EndoS but lacks the two 3HB domains on its N- and C- termini \(^{28}\) . We previously described the structural determinants that govern the \(N\) - glycan specificities of EndoS \(^{26,27}\) and EndoS2 \(^{28}\) . However, the mechanism by which these enzymes specifically process the \(N\) - glycan on the Fc region of IgG antibodies but no other glycoproteins, has not yet been described. Here, we present the cryoEM structure of EndoS in complex with the IgG1 Fc region. In combination with small- angle X- ray scattering (SAXS), alanine scanning mutagenesis, hydrolytic activity measurements, nuclear magnetic resonance (NMR) and molecular dynamics (MD) analysis, we describe the molecular mechanisms of recognition and specificity by EndoS and EndoS2 for IgG antibodies. Our findings will enable novel strategies for the chemoenzymatic synthesis of monoclonal antibodies with improved therapeutic properties, as well as the engineering of new enzymes to treat diseases of the adaptive immune system. + +<--- Page Split ---> + +## Results + +## Architecture of the EndoS-IgG1 Fc region complex + +To elucidate the molecular mechanism of antibody recognition by EndoS we determined the single- particle cryoEM structure of the catalytically- inactive mutant EndoS \(\mathrm{E_{233A}}\) in complex with the Fc region of an IgG1 monoclonal antibody (EndoS \(\mathrm{E_{235A}}\) - Fc) at 4.5 Å resolution (Fig. 1C,D; Supplementary Table 1; Supplementary Figs. 1, 2, 3 and 4; Methods). Although the purification of the native enzyme- substrate complex by size exclusion chromatography (SEC) was heterogeneous, we were able to purify the cross- linked EndoS \(\mathrm{E_{235A}}\) - Fc using the Grafix method (20; Supplementary Fig. 1). The SDS- PAGE of the fractions obtained by stabilization of the EndoS \(\mathrm{E_{235A}}\) - Fc complex with mild conditions of glutaraldehyde mainly shows the formation of a band corresponding to the MW of the EndoS \(\mathrm{E_{235A}}\) - Fc 1:1 complex but also other bands at higher MW that could correspond with the EndoS \(\mathrm{E_{235A}}\) - Fc 2:1 (250 kDa) (Supplementary Fig. 1A). We isolated the fractions of the EndoS \(\mathrm{E_{235A}}\) - Fc 1:1 complex and further purified the sample by SEC (Supplementary Fig. 2; Methods). In the EndoS \(\mathrm{E_{235A}}\) - Fc complex structure, the Fc region is located at the cleft of the "V"- shaped structure of EndoS, between the GH and the \(\beta\) - sandwich domains. The Fc region is a homodimer of the C-terminal domains of the antibody heavy chain. Each protomer compromises \(\mathrm{C\gamma 2}\) and \(\mathrm{C\gamma 3}\) domains, which interact with the equivalent domains of the other protomer via disulfide bonds in the hinge region and non- covalent interactions between the \(\mathrm{C\gamma 3}\) domains. The \(N\) - glycan is linked to each Asn297 of both \(\mathrm{C\gamma 2}\) domains. One of the Fc protomers (FcP1) interacts with the GH and the \(\beta\) - sandwich domains of EndoS through an extensive contact area of ca. 874 Å 30, while weak electron density of the cryoEM map can be ascribed to the second Fc protomer (FcP2) which does not make substantial interactions with EndoS and the N- 3HB domain. However, multibody refinement and motion analysis showed that FcP2 transiently interacts with residues 403- 413 of the GH domain of EndoS and the top of the N- 3HB domain (Supplementary Video 1, 2 and 3). Structural comparison of the EndoS- Fc complex and unliganded EndoS showed no marked conformational changes + +<--- Page Split ---> + +(PDB code 4NUZ; r.m.s.d. of 2.5 Å from 880 residues; Supplementary Fig. 5). The dimensions of EndoS in the EndoSE235A- Fc 3D reconstruction complex (123.1 Å × 75.1 Å × 60.5 Å) are similar to those found in the X- ray crystal structure of EndoS (129.2 Å × 80.3 Å × 60.8 Å), indicating that the V shape of EndoS is not appreciably distorted in the active complexes. This is also supported by our SAXS data (Supplementary Table 2; Supplementary Fig. 6; Supplementary Note 1). EndoS in complex with the Fc region exhibits a narrower cleft between the GH and \(\beta\) - sandwich domains than does unliganded EndoS (Supplementary Fig. 5). In addition, the overall structure of FcP1 did not change after binding to the EndoS (PDB code 1H3X; r.m.s.d. of 1.6 Å from 345 residues). The main structural changes occur on the C' \(\beta\) - strand, the C'E loop (residues 295- 300) that bears the \(N\) - glycan, and the BC loop (residues 264- 272) within the C\(\gamma\)2 domain of FcP1 (Fig. 1C,D and Supplementary Fig. 6). + +## The EndoS GH domain interacts with the \(N\) -glycan and the C\(\gamma\)2 domain of the Fc region + +We have previously shown that the CT \(N\) - glycan product is accommodated in the binding site of the GH domain of EndoS that is formed by several loops whose conformations governs the strict substrate specificity of the enzyme (27; PDB code 6EN3). The EndoS GH domain adopts a conserved \((\beta /\alpha)_8\) - barrel with a defined \(N\) - glycan binding site delineated by the N- 3HB domain and the connecting loops \(\beta 1 - \beta 2\) (loop 1; residues 120–145), \(\beta 2 - \alpha 4\) (loop 2; residues 151–158), \(\beta 3 - \alpha 5\) (loop 3; residues 185–206), \(\beta 4 - \alpha 6\) (loop 4; residues 235–247), \(\beta 5 - \alpha 7\) (loop 5; residues 281–289), \(\beta 6 - \alpha 8\) (loop 6; residues 304–306), \(\beta 9 - \alpha 10\) (loop 7; residues 347–380), \(\beta 10 - \alpha 11\) (loop 8; residues 403–413), and \(\alpha 11 - \alpha 12\) (loop 9; residues 420–434). The cryoEM EndoSE235A- Fc structure shows that loop 6' (residues 312- 324) of the GH domain interacts with the C' \(\beta\) - strand and the BC loop of the C\(\gamma\)2 of FcP1 (Fig. 2). This loop adopts a \(\beta\) - hairpin architecture, which is unique amongst GH18 enzyme structures, and creates an opening towards the active site of the EndoS GH domain (26). In contrast, EndoS2 exhibits a shorter loop at this position comprising different amino acid residues (28 (Supplementary Fig. 7). Specifically, K323 of EndoS + +<--- Page Split ---> + +interacts with E294 of FcP1 by hydrogen bonds, while E315 of EndoS makes a salt bridge with R292 of FcP1. Moreover, W314 of EndoS interacts with the side chain of P271 and E272 of FcP1 by hydrophobic interactions and a hydrogen bond, respectively. The side chain of H236 from loop 4 of the EndoS GH domain interacts with Y296 of C'E loop by hydrogen bonds, while E354 makes hydrogen bonds with S298 of the same loop. These interactions between FcP1 and the GH domain of EndoS cause a distortion of the C' \(\beta\) - strand and conformational changes in the C'E loop. This leads to the conformational change of the C'E loop carrying the \(N\) - glycan and the side chain rotamer of N297 that promotes the displacement of the \(N\) - glycan from the center of the Fc region towards the active site of the enzyme (Fig. 2). Careful inspection of the electron density map in the active site allowed us to model the two first GlcNAc residues and the central mannose core saccharide (Supplementary Fig. 8). The O3 and the oxygen of the acetamide group of GlcNAc (+1) makes hydrogen bonds with the side chain of N356 of EndoS while O6 interacts with the side chain of Y305. The OD2 atom of D233 that participates in the orientation of the GlcNAc (- 1) is 5.4 Å from the nitrogen of the acetamide this sugar. The O3 of Man (- 2) interacts with the side chain of D152 by hydrogen bonds. + +## The EndoS \(\beta\) -sandwich domain confers IgG specificity through a protein-protein interaction with the Fc region + +The \(\beta\) - sandwich domain of EndoS is composed of eight \(\beta\) - strands connected by short loops and arranged in two opposing antiparallel \(\beta\) - sheets, including \(\beta 1 - \beta 2 - \beta 7 - \beta 4 - \beta 5\) , and \(\beta 6 - \beta 3 - \beta 8\) (Fig. 2). Both EndoS and EndoS2 \(\beta\) - sandwich domains were previously classified and suggested to function as carbohydrate- binding modules (CBM) from family 32 based on structural homology \(^{26 - 28,31}\) . However, the cryoEM EndoS \(_{\mathrm{E235A}}\) - Fc structure clearly reveals that the loops of the \(\beta\) - sandwich domain bind to FcP1 through protein- protein interactions. Moreover, N297 that bears the \(N\) - glycan points towards the GH domain and in the opposite direction of the \(\beta\) - sandwich domain. The \(\beta\) - sandwich domain interacts exclusively with + +<--- Page Split ---> + +the \(\mathrm{Cy2 - Cy3}\) joint of FcP1. Specifically, the EndoS residue W803, previously identified as a key residue for EndoS hydrolytic activity and binding to IgG1 Fc \(^{26}\) , makes a \(\pi - \pi\) interaction with H435 of \(\mathrm{Cy3}\) domain and hydrophobic interactions with I253 of \(\mathrm{Cy2}\) domain, while Y909 of EndoS interacts with the side chains of Q311, L314 and N315 of the \(\mathrm{Cy3}\) domain. In addition, W803 together with S800, Y909, S910 and S911 were proposed to form a putative \(N\) - glycan binding pocket \(^{28}\) but the EndoS \(_{E235A}\) - Fc cryoEM structure shows that the main function of W803 and Y909 is to mediate protein- protein interactions between the enzyme and the Fc region of the substrate. Structural comparison of the EndoS \(_{E235A}\) - Fc complex with that of EndoS2 supports a common binding mode to Fc for both enzymes. EndoS2 also bears aromatic residues, W712 and Y820, at equivalent positions to W803 and Y909 in EndoS, respectively, which are indispensable for substrate binding and catalysis \(^{28}\) ; Supplementary Fig. 7). Thus, the \(\beta\) - sandwich domains of EndoS and EndoS2, previously proposed to be CBMs, act instead to mediate a protein- protein interface with its glycoprotein in which no glycan participates. + +## EndoS-Fc protein-protein interactions are critical for glycoside hydrolase activity + +To further investigate the molecular determinants of EndoS- Fc complex formation, we performed single and multiple alanine mutations in both EndoS and Fc, as well as loop deletions in EndoS to determine the key residues affecting hydrolytic activity in the EndoS- Fc interface. The residues in EndoS and Fc were selected based on the cryo- EM structure. We determined the rates by which the EndoS mutants hydrolyzed glycans from Fc and by which EndoS hydrolyzed glycans from Fc mutants by tracking the proportions of glycosylated versus deglycosylated Fc over time by intact mass spectrometry (Fig. 3; Supplementary Figs. 9 and 10). Specifically, in the Fc region we mutated residues of the C'E loop of \(\mathrm{Cy2}\) domain (residues R292, E293, E294, Q295, Y296 and S298), the \(\mathrm{Cy2 - Cy3}\) joint region (residues I253, H310, Q311, L314, N315, E430 and H435), the BC loop of \(\mathrm{Cy2}\) domain (H268, E269, D270, E272), and the FG loop of the \(\mathrm{Cy2}\) domain (residues 325- 331) that faces the N- 3HB domain of EndoS to alanine. + +<--- Page Split ---> + +For positions in this loop that are encoded by alanine residues in the Fc region, we made mutations to glycine and tryptophan in order to determine the effect of removing or extending the side chains at those positions, respectively. In EndoS, we mutated residues in the N- 3HB loop (residues K70, Q73, E74, Q76 and K77), the GH domain (residues H236, W314, E315, N319, K323, E354, K414, Y416, K418, Q319, K420, E421, F422, K423 and loops 6' and 9), the LRR loop (residues Y541, K543, D544, N545 and K546) and the \(\beta\) - sandwich domain (residues W803, Y909 and loop 828- 836) to alanine and assessed their hydrolytic activities. As shown in Fig. 3, mutations in the C'E loop exhibited pleiotropic effects on hydrolytic activity; mutations in residues E293A and Q295A that did not interact with the enzyme, resulted in faster deglycosylation by EndoS, while the remaining mutations reduced the rate of glycan hydrolysis. Similar results were obtained for mutations in the N- 3HB facing loop on the Fc, with some variants (e.g., A330W and NKALPAP \(\rightarrow\) GGGsGGG) resulting in faster glycan removal while the remainder reduced the rate of \(N\) - glycan hydrolysis. Mutations in the \(\mathrm{C\gamma 2 - C\gamma 3}\) joint residues that interact with the \(\beta\) - sandwich domain of EndoS significantly reduced the rate of glycan hydrolysis. In EndoS, mutations in the N- 3HB, GH, LRR and \(\beta\) - sandwich domains all significantly reduced the rate of glycan hydrolysis from Fc, with the removal of loop 9 in the GH domain and the W803A mutation in the \(\beta\) - sandwich domain completely abolished EndoS activity. + +The results of our mutational analysis indicate that the most important region driving EndoS- Fc complex formation and \(N\) - glycan hydrolysis involves residues on the \(\mathrm{C\gamma 2 - C\gamma 3}\) joint region of the Fc which interact with the \(\beta\) - sandwich domain of EndoS. The most extreme examples are (i) the EndoS W803A mutation and (ii) the Fc I253A/H310A double mutation, which completely abolish hydrolytic activity (Fig. 3). The variable activities observed in the Asn297- loop and N- 3HB facing loops in Fc and in the N- 3HB in EndoS suggest that conformations adopted by these loops play an important role in recognition of Fc by EndoS. Mutations in the EndoS GH domain significantly reduced enzymatic activity suggesting that these mutations adversely affect the ability of the \(N\) - glycan to enter the active site. The + +<--- Page Split ---> + +reduction of the hydrolytic activity observed for mutations in the LRR loop of EndoS suggests that these mutations may adversely affect the conformation of this loop which stabilizes the conformation of the GH domain loop 6' to effectively recognize the glycan loop of the antibody as shown by the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure \(^{27}\) . + +We further evaluated the conformational behavior of the IgG1 Fc region by molecular dynamics (MD) simulations. We modelled the structure of wild type and mutant EndoS- Fc complex structures, including mutations in the Fc region (E293A, Y296A, E430A- H435A, and H310A- L314A) or in EndoS (W314A, W803A, and Y909A) corresponding to select mutations in the hydrolytic activity experiments described above. We observed that the vast majority of the interactions between the Fc region and EndoS in the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure in these MD simulations, except for the interaction between H236 of EndoS and Y296 of FcP1 and the salt- bridge formed between K323 and E294, which were rarely or negligibly populated. As we observed in the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure, W803 of EndoS was engaged in a \(\pi - \pi\) interaction with H435 of FcP1. Similarly, W314 and Y909 of EndoS were involved in hydrophobic contacts with P271 and L314 of the Fc. The side chain of E354 formed a hydrogen bond with the hydroxyl group of S298 of FcP1 and the salt- bridge formed between E315 of EndoS and R292 of FcP1 was displayed in approximately 80 percent of the total trajectory time (Fig. 4A). We observed a similar scenario for the complex with the FcP1 E293A mutation (Fig. 4B). Notably, and in contrast to the wild- type antibody, the salt- bridge between K323 of EndoS and E294 of FcP1 was highly occupied in solution. These computational data could explain, at least in part, the higher activity of this mutant related to wild type. Of note, a loss of one or more of these stabilizing interactions is observed in the remaining complexes, resulting in a different 3D structure of the complex compared with the wild- type complex (Fig. 4C; Supplementary Figs. 11, 12, 13 and 14). + +<--- Page Split ---> + +## EndoS and EndoS2 deglycosylate the two glycans of the IgG Fc homodimer sequentially + +Our EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure revealed a 1:1 EndoS:Fc complex in which EndoS engages a single Fc protomer through a protein- protein interaction formed by the EndoS \(\beta\) - sandwich domain and the Fc \(\mathrm{C\gamma 2 - C\gamma 3}\) joint, while the EndoS GH domain engages the N297- linked glycan on the same Fc protomer. However, EndoS hydrolyzes the glycans from both Fc protomers and no evidence of partially deglycosylated IgG has been reported to date. In order to define the mechanism of full deglycosylation of IgG by EndoS, we analyzed the deglycosylation reaction over time. We mixed wild type EndoS with Fc in a 1:5000 enzyme:substrate ratio and analyzed changes in intact Fc masses by LC- MS approximately every 5 minutes (Fig. 5A). As expected, the di- glycosylated Fc species decreased while the fully deglycosylated species increased over time. As the reaction proceeded, however, we observed that a mono- glycosylated species initially increased to approximately 50 percent of all Fc masses, prior to decreasing along with di- glycosylated Fc to near zero. From these data, we calculated a rate of hydrolysis for the transition from di- glycosylated Fc to mono- glycosylated Fc of \(3.8~\mu \mathrm{M}^{- 1}\mathrm{s}^{- 1}\) and for the transition from mono- glycosylated Fc to fully deglycosylated Fc of \(0.8~\mu \mathrm{M}^{- 1}\mathrm{s}^{- 1}\) . Thus, the hydrolysis of the first glycan was approximately 5- fold faster than the hydrolysis of the second glycan. We conducted an analogous experiment with EndoS and Rituximab, a full- length IgG1 antibody and observed identical kinetic parameters for hydrolysis of the first and second glycans (Fig 5B). We also conducted an analogous experiment with EndoS2 and Rituximab and found similar kinetic behavior (Fig 5C); although each of the reaction rates was slower, hydrolysis of the first glycan was also 5- fold faster than that of the second glycan, as for EndoS. These data indicate that EndoS- and EndoS2- mediated hydrolysis of the two glycans on Fc or IgG occurs sequentially, concomitant with dissociation of the enzyme- substrate complex after hydrolysis of the first glycan before re- associating to hydrolyze the second glycan. + +<--- Page Split ---> + +## The sterically constrained glycan binding site of EndoS2 prevents deglycosylation of potential non-IgG glycoprotein substrates + +To shed more light on the interaction of EndoS and EndoS2 with their natural ligands, we employed NMR chemical shift perturbations (CSP) \(^{32}\) . Generally speaking, the method consisted of following changes in chemical shifts from signals in NMR spectra during ligand titrations, and using them to determine the location of binding sites, affinity of ligands and/or conformational equilibria. We used CSPs for the quantification of the overall thermodynamic and kinetic binding parameters of EndoS2 to IgG1 Fc and its reaction products. We extracted CSPs from methyl- TROSY NMR spectra \(^{33}\) and fitted quantum mechanically simulated spectra using the TITAN (TITration ANalysis) software package \(^{34}\) . This approach offers unique access to binding thermodynamics and kinetics parameters \(^{33,35}\) , providing a link to the conformational changes observed by crystallography, H/D exchange MS \(^{28}\) , cryo- EM and MD, assisting our mechanistic understanding of EndoS2 enzymatic activity (Supplementary Note 2: Supplementary Figs. 15 to 33; Methods). + +While were unable to express EndoS in minimal media in quantities required for NMR experiments, we were able to express and purify isotopically \([U^{- 15}N,^{2}H]\) MIL \(^{pros}V^{pros}[^{13}C,^{1}H_{3}]\) - methyl labeled \(^{35 - 37}\) EndoS2 and catalytically- inactive EndoS2 \(_{E186L}\) in order to perform these experiments. This labeling scheme produced methyl- TROSY spectra with excellent signal- to- noise (Supplementary Fig. 15). In addition, selected methyl probes were well distributed around the binding pockets, allowing for the extraction of CSPs upon ligand titration (Fig. 6A,B). We measured the binding kinetics of EndoS2 \(_{E186L}\) to its CT glycan substrate, the IgG1 Fc region, the reaction products Fc aglycan and CT \(_{n - 1}\) , and Ca \(^{2 + }\) ions (Fig. 6C). We also studied the interaction of catalytically- inactive EndoBT- 3987 \(_{D312A/E314}\) , a GH18 family ENGase that hydrolyzes HM glycans from myriad glycoproteins but is not specific for IgG antibodies or any other protein (Supplementary Fig. 16) \(^{38,39}\) . Quantitative lineshape analysis of methyl- TROSY spectra showed that EndoS2 \(_{D186L}\) binds Rituximab- Fc with a dissociation constant of 3.1 + +<--- Page Split ---> + +\(\mu \mathrm{M}\) (Table 1, entry 1), similar to the previously reported \(K_{D}\) of \(\sim 9 \mu \mathrm{M}\) for the binding of EndoS2D186L to IgG1- CT obtained by surface plasmon resonance \(^{28}\) . A representative example of lineshape fitting can be seen in Fig. 6D. We measured the binding affinity of EndoBT- 3987D312A/E314 to its HM \(N\) - glycan substrate to be \(K_{D} = 1.7 \mu \mathrm{M}\) (Table 1, entry 2). We measured the \(K_{D}\) s of EndoS2D186L for soluble CT glycan and the reaction product \(\mathrm{CT_{n - 1}}\) to be \(106.9 \mu \mathrm{M}\) and \(150.7 \mu \mathrm{M}\) , respectively, which corresponds to affinities two orders of magnitude weaker than the EndoS2E186L- Fc and EndoBT- 3987D312A/E314 affinities. We measured on- rate constants for Fc, CT and \(\mathrm{CT_{n - 1}}\) to be \(1.3 \times 10^{6} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) , \(1.1 \times 10^{5} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) and \(6.2 \times 10^{4} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) , respectively, all much slower than expected for a diffusion- controlled process, indicating that these interactions involve solvent reorientation and possibly conformational changes. Dissociation rate constants \(k_{off}\) were similar for these three ligands, and go down into the range of a few Hz (Table 1, entries 3,4). The faster on- rate constant observed for the interaction of EndoS2D186L with Fc, similar to the on- rate constant of EndoBTD312A/E314 for just the \(N\) - glycan, suggests that a protein- protein interaction favors a “correct” orientation of the EndoS2 GH domain relative to the glycan during the initial EndoS2- Fc fragment encounter event. The energetic contribution of such interactions can be approximated as the change in the difference of the standard free energy \(\Delta \Delta G_{Fc - CT}^{\circ} = \Delta G_{Fc}^{\circ} - \Delta G_{CT}^{\circ} = -8.7 \mathrm{KJ}\) , considering \(\Delta G^{\circ} = RT \ln (K_{D})\) . We found that EndoS2- mediated hydrolysis of Fc \(N\) - glycans abrogated recognition of Fc by EndoS2D186L at otherwise saturating Fc concentrations, highlighting the central role of the carbohydrate for the recognition of IgG by EndoS2 and preventing re- binding of reaction products (Table 1, entry 5). We also extracted the catalytic rate, \(k_{cat}\) , and the Michaelis- Menten constant, \(K_{\mathrm{M}}\) , from the catalytic cleavage of CT \(N\) - glycans from the Fc region by EndoS2 by measuring the concentration of enzymatically released glycans by \(^1\mathrm{H}\) NMR spectra (Supplementary Note 2; Methods). Under the conditions used in this study, we observed a catalytic rate \(k_{cat}\) of \(5.9 \mathrm{s}^{- 1}\) and a Michaelis- Menten constant \(K_{\mathrm{M}}\) of \(26.7 \mu \mathrm{M}\) Fig. 6E). Combined with lineshape analysis, these results provide a first comprehensive + +<--- Page Split ---> + +picture of the catalytic mechanics of the enzymatic processing of IgG1 Fc \(N\) - glycans by EndoS2 (Fig. 6F). + +<--- Page Split ---> + +## Discussion + +EndoS and EndoS2 are the most prominent members of a rare subset of carbohydrate- active enzymes that are specific not only to the glycan type but also to the protein component of the glycoprotein from which they hydrolyze glycans. To date, no molecular mechanism has been described to explain how these enzymes are protein- specific. Our data now provide the mechanism by which EndoS and EndoS2 hydrolyze glycans from IgG antibodies in a highly specific manner. This mechanism for protein- specific deglycosylation by EndoS and EndoS2 relies on four principles. These enzymes: (i) avoid all other potential non- IgG glycoprotein substrates; (ii) form a protein- protein interaction with a non- enzymatic domain in order to create an anchor point on the IgG substrate; (iii) leverage their unique geometry to reach from the anchor point to orient the active site of the GH domain directly adjacent to the Asn297- linked glycan that they hydrolyze; and (iv) hydrolyze the two glycans on the IgG Fc region homodimer sequentially by a catch- and- release process. + +The first mechanistic principle, reduced distraction by potential non- IgG substrates, is accomplished by EndoS and EndoS2 constructing conformationally constrained glycan binding sites. This distinctive morphological property of their active sites results in the bound glycan needing to adopt a conformation in which the branches of a biantenary glycan are nearly parallel to one another \(^{26 - 28}\) . Although such a glycan conformation is commonly rendered schematically, it is exceedingly rare in nature due to the high degree of conformational flexibility inherent to \(N\) - linked glycans. Even when bound in the glycan binding sites of endoglycosidases that are not protein- specific, \(N\) - linked glycans adopt diverse structures that are decidedly unlike the parallel- branched glycans in EndoS and EndoS2. This is the case for the \(N\) - linked glycans that are the substrates for EndoBT- 3987 \(^{38,39}\) (Supplementary Fig. 34), EndoF \(^{3,40}\) , as well as many others. The adoption of the constrained glycan conformation in EndoS and EndoS2, as well as the flexible glycan conformation in EndoBT- 3987, can be observed in the relative on- rates of substrates binding to catalytically- inactive variants of these endoglycosidases. Indeed, our NMR data indicate that the on- rates for glycans alone are two orders of magnitude slower for EndoS + +<--- Page Split ---> + +and EndoS2 than they are for EndoBT- 3987. However, these enzymes have all evolved to have comparable affinities and binding kinetics to their biologically relevant carbohydrates – linked to IgG for EndoS and EndoS2, while free in solution or exposed on the surface of any protein for EndoBT- 3987 – a property clearly related to the distinct structural features of the active sites of these endoglycosidases and the relative entropic costs of their glycan substrates adopting their bound conformations. + +The second mechanistic principle, the establishment of an enzyme- substrate protein- protein interaction, is fulfilled by the \(\beta\) - sandwich domain of EndoS and EndoS2. While these domains have high structural similarity to carbohydrate binding modules (CBMs), they are not, in fact, bona fide CBMs in that they do not bind carbohydrates. Instead, these \(\beta\) - sandwich domains recognize a strictly protein- based molecular surface on the IgG Fc region at the joint between its \(\mathrm{Cy2}\) and \(\mathrm{Cy3}\) domains. This Fc surface is a common binding spot for myriad proteins, including Protein A \(^{41}\) and Protein G \(^{42}\) from bacteria, as well as the human neonatal Fc receptor (FcRn; \(^{43}\) ). The structural similarity of this molecular surface on the Fc region amongst all IgG subtypes of both human and mouse antibodies explains why these glycoproteins comprise all of the known substrates for EndoS and EndoS2. Furthermore, this enzyme- substrate protein- protein interaction is absolutely required for enzymatic activity, as mutations on either side of this interface can completely abrogate hydrolysis of Asn297- linked Fc glycans by EndoS \(^{26}\) and EndoS2. Accordingly, for EndoS and EndoS2, a protein- protein interaction drives hydrolytic activity of these endoglycosidases, a property not previously observed in carbohydrate- active enzymes. + +The third mechanistic principle, orientation of the enzyme active site adjacent to the substrate glycan, is a result of the unique V- shaped molecular architecture of EndoS and EndoS2 in which the \(\beta\) - sandwich and GH domains are situated on either tip of the V, interspersed by an Ig- like domain and a leucine- rich repeat. As a consequence of anchoring the enzyme on the substrate via the \(\beta\) - sandwich domain/Fc region protein- protein interface described above and the architectures of these multi- domain enzymes, the active sites of EndoS and EndoS2 become ideally positioned for engagement with and + +<--- Page Split ---> + +hydrolysis of the Fc glycan. The predominant conformation of the IgG Fc region is one in which the Asn297- linked glycans are pointed inward towards the center of the homodimer, although these antibodies are known to exhibit many conformations, including those in which these glycans are more exposed to solvent \(^{44,45}\) . Our cryoEM structure shows that the Fc loop on which the Asn297- linked glycan adopts just such a conformation; it is not exposed to solvent in the complex, though, but rather bound in the EndoS active site. Thus, when anchored by the enzyme- substrate protein- protein interface, the distinctive molecular architecture positions the GH domain to stabilize an otherwise infrequently populated conformation amongst the entire ensemble of Fc conformations so as to increase the sampling of its glycan, resulting in efficient glycan hydrolysis. It must be noted that the appendage of one or more non- enzymatic domains to GH domains is not uncommon in carbohydrate- active enzymes and is thought to guide the enzymatic domains towards their targets, as in the case of CBMs, and/or to provide sufficient spacing from the bacterial outer membrane for hydrolyzing glycans in the extracellular environment. However, the structural arrangement of the enzymatic and non- enzymatic domains of EndoS and EndoS2 is unique amongst known carbohydrate- active enzymes, as is the requirement for two of these connected domains to be absolutely required for activity. + +The fourth mechanistic principle, sequential deglycosylation of the two Asn297- linked glycans on the IgG homodimer, is due to the biologically relevant state being a 1:1 complex of EndoS: Fc, as we observed in our cryoEM structure. When anchored by their protein- protein interfaces, the GH domains of EndoS and EndoS2 can reach only to the Asn297- linked glycan on the same Fc protomer that their \(\beta\) - sandwich domains engage. In order to hydrolyze the glycan on the other Fc protomer, the enzyme must unbind the substrate entirely, and rebind it through a protein- protein interaction on the opposite Fc protomer in order to properly reposition its GH domain. Indeed, by intact mass spectrometry, we observed the accumulation of a mono- glycosylated antibody species, followed by its diminution. If the biologically relevant state were instead a 2:1 complex of EndoS: Fc, no such mono- glycosylated antibody + +<--- Page Split ---> + +species would exist, even transiently, as two EndoS enzymes would attack the antibody from each side simultaneously. Also, in our NMR experiments we observed that re- binding of the reaction product, a deglycosylated antibody, was highly disfavored. Despite this, a 2:1 complex formed between the enzyme and a di- glycosylated substrate can exist at high concentrations, as we observed as minor populations by both SAXS and cryoEM analysis (Supplementary Fig. 35). + +With the molecular mechanism of IgG- specific deglycosylation established, opportunities for engineering EndoS- based enzymes to treat diseases of the adaptive immune system abound. It is now possible to rationally conceptualize engineered EndoS enzymes that selectively deglycosylate certain subsets of IgG antibodies, such as auto- antibodies, in order to defeat their effector functions and treat autoimmunity without wholesale IgG disablement that would render the host globally immunosuppressed and vulnerable to opportunistic infections. + +<--- Page Split ---> + +## Methods + +## Expression and purification of EndoS and EndoS2 wild type and EndoS and EndoS2 mutants + +EndoS and EndoS2 wild- type and EndoS and EndoS2 mutants were purified as previously described with the following modifications \(^{26 - 28}\) . Single- point mutations were developed by PCR- based site- directed mutagenesis, and full sequences were confirmed by GeneWiz (https://www.genewiz.com). EndoS wild type and mutants were produced in Escherichia coli BL21(DE3) cells (Novagen) grown in LB medium supplemented with \(100 \mu \mathrm{g} \mathrm{ml}^{- 1}\) of ampicillin. Cultures were grown at \(37^{\circ} \mathrm{C}\) to an \(\mathrm{OD}_{600}\) of 0.6–0.8, at which point the temperature was lowered to \(22^{\circ} \mathrm{C}\) for 1 h. Induction was triggered with \(1 \mathrm{mM}\) isopropyl \(\beta\) - D- 1- thio- galactopyranoside (IPTG) at \(22^{\circ} \mathrm{C}\) overnight. Cells were harvested by centrifugation and lysed by sonication using \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 7.5\) , \(500 \mathrm{mM}\) NaCl, \(10\%\) glycerol. CPD fusion proteins were purified by \(\mathrm{Ni}^{2 + }\) - immobilized metal- affinity chromatography followed by overnight treatment with \(1 \mathrm{mM}\) phytic acid at \(4^{\circ} \mathrm{C}\) and elution the next day with the lysis buffer. The proteins used for hydrolytic activity were further purified by size exclusion chromatography (Superdex 200 Increase 10/300 GL) in PBS as running buffer. \(\mathrm{EndoS}_{\mathrm{E235A}}\) were further purified by anion exchange chromatography using as binding buffer \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 8.5\) and elution buffer \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 8.50\) , \(500 \mathrm{mM}\) NaCl. The protein was eluted using a gradient of \(5\%\) of elution buffer for \(2 \mathrm{min}\) . The fractions were pooled and aliquot to flash freeze at \(- 80^{\circ} \mathrm{C}\) . For EM sample preparation, \(\mathrm{EndoS}_{\mathrm{E235A}}\) were thaw and concentrated protein was purified by size exclusion chromatography (Superdex 200 Increase 10/300 GL) in PBS as running buffer. + +## Purification of Fc-Rituximab + +Rituximab (RITUXAN, Genentech) was kindly provided courtesy of the University of Maryland Greenebaum Comprehensive Cancer Center. A solution of \(20 \mathrm{mg} \mathrm{mL}^{- 1}\) of Rituximab was dialyzed against \(20 \mathrm{mM}\) sodium phosphate, \(10 \mathrm{mM}\) EDTA; \(\mathrm{pH} 7.0\) . A mixture of \(0.5 \mathrm{mL}\) of the Rituximab solution + +<--- Page Split ---> + +and \(0.5 \mathrm{mL}\) of the digestion buffer (20 mM sodium phosphate, 10mM EDTA, 20 mM cysteine- HCl pH 7.0), is added to the \(0.5 \mathrm{mL}\) of the \(50\%\) immobilized papain slurry (Thermo Scientific) previously equilibrated with the digestion buffer. The reaction was incubated overnight in a shaker water bath at 37 \(^\circ \mathrm{C}\) at high speed overnight. The reaction was stopped by adding \(1.5 \mathrm{mL}\) of \(10 \mathrm{mM}\) Tris- HCl pH 7.5 to the digest before centrifugation. After removing the supernatant, which contains the IgG fragments, the resin was washed with \(1.5 \mathrm{mL}\) of \(10 \mathrm{mM}\) Tris- HCl pH 7.5. The digest and wash fractions were combined and the Fc fragments were separated from the Fc fragments using an immobilized Protein A column (Thermo Scientific) and \(100 \mathrm{mM}\) glycine pH 3.0, as elution buffer. The eluted Fc was further purify by size exclusion chromatography using PBS to separate the Fc fragments from undigested IgG. + +## Preparation and purification of the cross-linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) -Fc-Rituximab complex + +The cross- linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) - Fc- Rituximab complex was prepared using a standard GraFix method \(^{29,46}\) . Gradients for GraFix were prepared by mixing GraFix buffer 1 (PBS, pH 7.4, \(5\%\) v/v glycerol) and GraFix buffer 2 (PBS, pH 7.4, \(20\%\) v/v glycerol, \(0.25\%\) glutaraldehyde) using a gradient mixer (Gradient Master ip, BioComp Instruments), following the parameters determined by the manufacturer for the glycerol content. A mixture of \(100 \mu \mathrm{L}\) of EndoS \(\mathbf{E}_{235\mathrm{A}}\) ( \(12 \mu \mathrm{M}\) ) and \(100 \mu \mathrm{L}\) of Fc- Rituximab (60 \(\mu \mathrm{M}\) ) in PBS was incubated at room temperature for 10 min. The \(200 \mu \mathrm{L}\) solution mixture was loaded onto the top of the GraFix gradients and ultracentrifugation was carried out at \(40,000 \mathrm{rpm}\) for \(25 \mathrm{hs}\) at 4 \(^\circ \mathrm{C}\) (SW60 rotor, Beckmann). After ultracentrifugation, the gradients of each tube were fractionated using a Piston Gradient Fractionator \(^\mathrm{TM}\) Base Unit (Sciences service) and a FC 203 Fraction Collector (Gilson) to pump the gradient out from bottom to top at room temperature. Each fraction of \(250 \mu \mathrm{L}\) was immediately quenched by \(50 \mu \mathrm{L}\) of \(1 \mathrm{M}\) Tris- HCl pH 8.0, to stop further crosslinking. The presence of the cross- linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) - Fc- Rituximab complex was confirmed by SDS- PAGE. Pooled fractions were flash frozen and storage at \(- 80^\circ \mathrm{C}\) . + +<--- Page Split ---> + +## Negative-stained electron microscopy + +Negative- stained electron microscopyInitial inspection of the GraFix EndoSE235L- Fc- Rituximab complex by negative stain was performed. Specifically, an aliquot from a single GraFix tube were thawed and samples of \(50~\mu \mathrm{L}\) were injected into an analytical size exclusion chromatography column (Shodex KW403- 4F) using PBS pH 7.4, as running buffer. The single peak corresponding to the complex MW was collected in \(50~\mu \mathrm{L}\) fractions for subsequent standard negative stain grid preparation. Continuous glow discharged carbon electron microscopy grids CF300- Ni (EMS, USA). Briefly, drops of \(8~\mu \mathrm{L}\) of diluted complex sample were placed on the grid, incubated for 90 s and then blotted to remove the excess with filter paper (Whatman®). The grids were washed by placing them onto a drop of PBS, pH 7.4 for 30 s and bottled dry. The complex was stained by transferring the grid to a drop of \(1\%\) w/v uranyl formate for 45 s, then dried with a filter paper and stored at room temperature. Samples were imaged at \(\times 50,000\) magnification and using an inhouse JEOL 2200, \(200\mathrm{kV}\) FEG equipped with an UltraScan 4000 CCD camera (Gatan, USA). Images were used to analyze particles by single particle analysis using RELION. + +## Cryo-EM specimen preparation and screening + +Cryo- EM specimen preparation and screeningCross- linked EndoSE235L- Fc- Rituximab complexes from four GraFix tubes were thawed, pooled, and concentrated using Pierce Protein Concentrators (PES 10K MWCO \(0.5\mathrm{mL}\) ) to obtain a final volume of \(60~\mu \mathrm{L}\) . The sample was injected into an analytical size exclusion chromatography column (Shodex KW403- 4F) using PBS pH 7.4, as running buffer. The complex peak was collected in collected in \(50~\mu \mathrm{L}\) fractions with to use directly for vitrification. Briefly, \(4\mu \mathrm{l}\) of the fractions were applied onto copper Quantifoil R 1.2/1.3 grids previously glow discharged. Grids were plunge- freeze into liquid ethane using a Vitrobot MkII (FEI) at \(10^{\circ}\mathrm{C}\) and \(80\mathrm{HR}\%\) humidity in the chamber for each concentration condition by triplicate. One of the triplicates for each condition was used for cryoEM screening performed in- house using a JEOL 2200, \(200\mathrm{kV}\) FEG equipped with an UltraScan 4000 CCD camera (Gatan, USA). + +<--- Page Split ---> + +## Cryo-EM data collection + +Cryo- EM data collection was carried out on a \(300\mathrm{kV}\) Titan Krios microscope equipped with a K2 Summit direct electron- counting camera and a GIF Quantum energy filter (Gatan) in ESRF CM01 beamline at Grenoble, France \(^{47}\) . Micrographs were recorded with EPU software (FEI) at a nominal magnification of \(\times 130,000\) , with a pixel size of \(1.052\mathrm{\AA}\) . A total of 5,546 movies of the cross- linked EndoSE235L- Fc- Rituximab grid sample were acquired for \(9\mathrm{s}\) in counting mode at a flux of 7.3 electrons per pixel \(\mathrm{s}^{- 1}\) , giving a total exposure of 59.37 electrons per \(\mathrm{\AA}^2\) and fractioned into 60 frames. A defocus range from \(- 1.3\mu \mathrm{m}\) to \(- 2.5\mu \mathrm{m}\) was used. + +## Cryo-EM single-particle data processing + +EndoSE235- Fc complex data set were processed using the general protocol described in the following lines. Data processing was initiated using the RELION 3.1.2 \(^{48}\) . The 60 movie frames of each movie were aligned using MotionCor2 (5x5 patches and Bfactor 150). The CTF of the aligned image was assessed in the non- dose weighted image using CtfFind 4.1 with parameters amplitude contrast \(10\%\) and FFT box size of 512 pixels. Image CTF was visually inspected, and images with low quality discarded. RELION reference- free automatic particle picking was performed considering a particle diameter between \(95\mathrm{\AA}\) and \(110\mathrm{\AA}\) , imposing a minimum inter- particle distance of \(90\mathrm{\AA}\) and a low picking threshold (0.05), leading to an initial number of particles (797351). Particles were extracted using a square box of 180 pixels side. Particles were subjected to several rounds and extensive 2D classification selecting the best classes. An initial first 3D auto- refinement was performed using the map obtained from the negative stain images. This initial reconstruction was used to re- extract centered particles. Afterward, several steps of 3D classification were performed to discarding classes until models reached resolutions \(7\mathrm{\AA}\) and visual inspection revealed a well- defined structure. The selected particles (140,184) were used to reconstruct a model using RELION 3D auto- refinement protocol, which led to a map resolving to \(6\mathrm{\AA}\) . At this point, the model was used to perform one round of RELION CTF refinement and particle Polishing, and further + +<--- Page Split ---> + +3D classification leading to a set of selected particles (721,336) for a final RELION refinement reaching \(4.8\mathrm{\AA}\) resolution. Polished and CTF refined particles from RELION were imported into cryoSPARC v3.2.0 \(^{49}\) and a final 2D classification of 200 classes was performed and the best- defined classes selected (Supplementary Fig. 3B) containing 339,309 particles. Final 3D reconstructions were performed in cryoSPARC using the homogeneous refinement and applying dynamic masking, using the RELION reconstructions low- pass at \(30\mathrm{\AA}\) as starting model. Finally, a final step of Non- Uniform refinement for the reconstruction \(^{50}\) , on the corresponding symmetry, was performed applying a calculated global mask from the reconstruction. The reconstruction obtained from the non- uniform cryoSPARC protocol is reported. Reported auto sharp- maps were obtained from the cryoSPARC non- uniform refinements (Supplementary Fig. 2C) and DeepEMhancer \(^{51}\) (Supplementary Fig. 2D). Local resolution was measured in cryoSPARC LocRes program implementation \(^{52}\) , using a threshold for local FSC \(= 0.5\) for local resolution assessment (Supplementary Fig. 2C,D). For quality graphs as FSC plots, angular distribution and precision see (Supplementary Fig. 2E,F). For other statistics please see Supplementary Table 1. + +## Model building and refinement + +A model of the EndoS X- ray crystal structure (PDB code 6EN3) and the Fc region of IgG1- Fc (H1X3), was initially rigid- body fitted inside the cryoEM reconstruction using UCSF- Chimera \(^{53}\) . The initial refinement of the models using Phenix real space refinement \(^{54,55}\) was performed as an individual chain rigid body group and later performing local minimization, annealing, and ADPs, using initial model as reference model to aid in keeping the geometry observed in the crystal structures. The model was improved iteratively manually in Coot \(^{56}\) using the DeepEMhancer sharp maps, following with rounds of Phenix real space refinement. Remodeling based on the interpretation of density due to the movement from loops in the interphase between EndoS and IgG1- Fc, as well as placing of carbohydrate in the density at the active site was performed. In later stages of the modelling, the side- chains whose positioning could not be defined unambiguously in the map density have been removed. The quality of + +<--- Page Split ---> + +models was checked during refinements using Molprobity and validate- PDB server57. For model statistics and quality details please see Supplementary Table 1. + +## EndoSE235-Fc map principal component analysis and motion movies + +A two- body multibody refinement of cryoEM images was performed as a continuation of the final RELION 3D auto- refinement58. The density corresponding to catalytic inactive EndoS was defined as body 1 and the Fc region as body 2. Masks were created for the top and bottom halves of the consensus reconstruction using the volume eraser tool in UCSF- Chimera53 and RELION34 and used for a multibody refinement in RELION3. Principal component analysis was performed on the orientations of both bodies, and movies for the first three principal components (PCs) were written out as a series of volumes in MRC format59 describing the relative motion of the two bodies as described by these PCs (Supplementary Videos 1- 3). + +## SEC-SAXS experiments + +A \(50~\mu \mathrm{M}\) solution of EndoSE235A or EndoSE186L were incubated with \(250~\mu \mathrm{M}\) solution of Fc- Rituximab in \(50~\mathrm{mM}\) Tris- HCl pH 7.5, \(100~\mathrm{mM}\) NaCl, \(2\%\) \(\nu /\nu\) glycerol for \(10\mathrm{min}\) at room temperature to form the EndoS(2)- Fc- Rituximab complexes. Small- Angle X- ray Scattering coupled with Size Exclusion Chromatography (SEC- SAXS) data for purified EndoSE235A, EndoSE186L, Fc- Rituximab and the complex mixtures of EndoSE235A- Fc- Rituximab and EndoSE186L- Fc- Rituximab in \(50~\mathrm{mM}\) Tris- HCl pH 7.5, \(100~\mathrm{mM}\) NaCl, \(2\%\) \(\nu /\nu\) glycerol were collected on the B21 beamline of the Diamond Light Source, UK. \(50~\mu \mathrm{L}\) of the protein samples or the complex mixtures were injected into a Shodex KW403- 4F column and eluted at a flow rate of \(150~\mu \mathrm{L}\mathrm{min}^{- 1}\) . Data were collected using a Pilatus2M detector (Dectris, CH) at a sample- detector distance of \(3,914\mathrm{mm}\) and a wavelength of \(\lambda = 1\mathrm{\AA}\) . The range of momentum transfer of \(0.1< s< 5\mathrm{nm}^{- 1}\) was covered \((s = 4\pi \sin \theta /\lambda\) , where \(\theta\) is the scattering angle). Data were + +<--- Page Split ---> + +processed and merged using standard procedures by the program package ScAter \(^{60}\) and PRIMUS \(^{61}\) . The maximum dimensions \((D_{max})\) , the interatomic distance distribution functions \((P(r))\) , and the radii of gyration \((Rg)\) were computed using GNOM \(^{62}\) . The molecular mass was determined using ScAter \(^{60}\) . The ab initio multiphase reconstruction of the SAXS data of the EndoS \(_{E235A}\) - Fc- Rituximab and the EndoS \(_{E186L}\) - Rituximab complexes were generated using the MONSA algorithm from ATSAS \(^{63}\) . The results and statistics are summarized in Supplementary Table 2. + +## Cloning, expression and purification of EndoS mutants for alanine scan experiments + +Primers to introduce single or multiple mutations or deletions into the EndoS- CPD construct were designed using the NebBaseChanger (https://nebasechanger.neb.com/) and PCR was carried out using the NEB Q5 High Fidelity polymerase and manufacturer's instructions. For some of the alanine scan mutants, gene strings were ordered from Twist Bioscience and restriction cloned into the EndoS- CPD vector. All sequences were confirmed by Sanger sequencing at Genewiz (https://www.genewiz.com). For expression, the plasmids were transformed into Escherichia coli BL21 (DE3) cells and grown in 100 mL of LB medium supplemented by \(100 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ampicillin T 37C. When the culture reached an OD of 0.6- 0.8, the culture was induced with 0.5mM IPTG and the culture was allowed to grow overnight at \(22^{\circ} \mathrm{C}\) . The cells were harvested by centrifugation at \(3,000 \times \mathrm{g}\) for 20 min and resuspended in 3mL of PBS pH 7.4. The cells were lysed with BugBuster according to manufacturer's instructions and the lysate clarified by spinning at \(18000 \times \mathrm{g}\) for 45 mins. The clarified supernatant was incubated with NiNTA resin for 3 hs before being washed with 10 column volumes (CV) of PBS pH 7.4. The EndoS was then eluted with \(100 \mu \mathrm{M}\) phytic acid in PBS pH 7.4. Protein purity was assessed by SDS- PAGE. All proteins were flash- frozen and stored at \(- 80^{\circ} \mathrm{C}\) until ready for use. + +<--- Page Split ---> + +## Cloning expression and purification of Fc mutants for alanine scan experiments + +The sequence for the heavy chain of Rituximab was ordered from Thermo Fisher Scientific in the pcDNA3.4- TOPO vector for expression in HEK293 cells. The Fc plasmid was subcloned out of this plasmid via PCR. Primers to introduce single or multiple mutations or deletions into the Fc plasmid were designed using the NebBaseChanger (https://nebasechanger.neb.com/) and PCR was carried out using the NEB Q5 High Fidelity polymerase and manufacturer's instructions All sequences were confirmed by Sanger sequencing at Genewiz (https://www.genewiz.com). The Fc mutants were expressed in either HEK293T or HEK293F cells. For HEK293T cells, the plasmids were transfected using polyethyleneimine as a transfection agent. After transfection, cells were cultured for 96h in Free- style F17 medium supplemented with GlutaMAX and Geneticin (Thermo Fisher Scientific). For Expj293 cells, the plasmids were transfected as per manufacturers protocol (MAN0007814, Thermo Fisher Scientific) with the addition of Penicillin/Streptamycin mix 24 hs after transfection. The cells were cultured for 96 hs before harvesting. The Fc mutants were purified using Protein A chromatography with PBS pH 7.4 being used as the binding buffer and 100 mM sodium citrate buffer pH 3.0 as the elution buffer. The fractions were neutralized with 1M Tris- HCl pH 9.3. SDS- PAGE was used to assess protein purity. The proteins were concentrated, flash- frozen and stored at \(-80^{0}\mathrm{C}\) until ready for use. + +## Hydrolytic activity assays + +For the EndoS- Fc alanine scan, LC- MS kinetic analysis was used to determine the reaction rate of deglycosylation by EndoS against Fc mutants and by EndoS mutants against Fc WT. \(30~\mu \mathrm{L}\) reactions were setup containing \(5\mu \mathrm{M}\) of the Fc variant and \(1 - 20~\mathrm{nM}\) of the EndoS mutants in PBS pH 7.4. The reactions were analyzed by LC- MS using an Agilent 1290 Infinity II LC System equipped with a \(50\mathrm{mm}\) PLRP- S column from Agilent with \(1000\mathrm{\AA}\) pore size. The LC system is attached to either an Agilent 6545XT quadrupole- time of flight (Q- TOF) or Agilent 6560 Ion Mobility (IM) Q- TOF mass + +<--- Page Split ---> + +spectrometer (Agilent, Santa Clara, CA). The reactions were setup and placed in the LC- MS and the reactions were sampled approximately every 15 minutes till the reactions reached completion. All reactions were performed in triplicate. Relative amounts of the substrate and hydrolysis products were quantified after deconvolution of the raw data and identification of the corresponding peaks using BioConfirm (Agilent, Santa Clara, CA). + +To obtain the rate of deglycosylation of Fc by the EndoS variants or Fc variants by EndoS, the relative proportions of the diglycosylated, monoglycosylated and deglycosylated products was imported into Kintek Global Kinetic Explorer64. This software fits the data to a model via the use of nonlinear regression analysis, where parameters are searched iteratively to determine a set of parameters which yields a minimum \(\chi^2\) value. The following model was used for all the fitted reactions: + +\[E + pGG\underset {k - 1}{\overset{k1}{\rightleftharpoons}}EpGG\underset {k - 2}{\overset{k2}{\rightleftharpoons}}EpG + G\] + +\[E + pG\underset {k - 3}{\overset{k3}{\rightleftharpoons}}EpG\underset {k - 4}{\overset{k4}{\rightleftharpoons}}Ep + G\] + +\[Ep\underset {k - 5}{\overset{k5}{\rightleftharpoons}}E + p\] + +The rates were fitted in a stepwise manner starting from k1, and once the value which yielded a minimum \(\chi^2\) was determined the value was locked. All values except for k2 and k4 were locked during the fitting process. Statistical significance was determined using a multiple comparisons test (Tukey method) in GraphPad (Graphpad Software, La Jolla, CA). + +<--- Page Split ---> + +## Molecular dynamics (MD) simulations + +The EndoS \(_{\mathrm{E235A}}\) - Fc complex obtained by cryoEM was used as starting coordinates for all MD simulations. The different mutants were generated starting form this structure and mutating the corresponding residues with PyMOL 2.5. The coordinates of the carbohydrate shown in Fig. 4 of the manuscript was generated using the GLYCAM WEB (https://glycam.org) and was attached to the side chain of Asn297 of all simulated complexes. MD simulations were performed with AMBER 20 package \(^{65}\) , implemented with ff14SB \(^{66}\) , and GLYCAM 06j- 1 force fields \(^{67}\) . Each complex was immersed in a water box with a \(10 \mathrm{\AA}\) buffer of TIP3P water molecules \(^{68}\) . The system was neutralized by adding explicit counter ions. A two- stage geometry optimization approach was performed. The first stage minimizes only the positions of solvent molecules and ions, and the second stage is an unrestrained minimization of all the atoms in the simulation cell. The systems were then heated by incrementing the temperature from 0 to \(300 \mathrm{K}\) under a constant pressure of 1 atm and periodic boundary conditions. Harmonic restraints of \(10 \mathrm{kcal} \mathrm{mol}^{- 1}\) were applied to the solute, and the Andersen temperature coupling scheme \(^{69}\) was used to control and equalize the temperature. The time step was kept at 1 fs during the heating stages, allowing potential inhomogeneities to self- adjust. Hydrogen atoms were kept fixed through the simulations using the SHAKE algorithm \(^{70}\) . Long- range electrostatic effects were modelled using the particle- mesh- Ewald method \(^{71}\) . An \(8 \mathrm{\AA}\) cutoff was applied to Lennard- Jones interactions. Each system was equilibrated for 2 ns with a 2- fs time step at a constant volume and temperature of \(300 \mathrm{K}\) . Production trajectories were then run for additional \(1 \mu \mathrm{s}\) under the same simulation conditions. + +## Protein expression and purification for NMR experiments + +\([U_{- }^{15}\mathrm{N},^{2}\mathrm{H}]\) , \(\epsilon - [^{13}\mathrm{C},^{1}\mathrm{H}_{3}]\) - Met, \(\delta 1 - [^{13}\mathrm{C}^{1}\mathrm{H}_{3}]\) - Ile, \(\delta 2 - [^{13}\mathrm{C},^{1}\mathrm{H}^{3}]\) - Leu, \(\gamma 2 - [^{13}\mathrm{C}^{1}\mathrm{H}_{3}]\) - Val- labeled (MILV) EndoS2, EndoS2 \(_{\mathrm{E186L}}\) , EndoBT- 3987 and EndoBT- 3987 \(_{\mathrm{D312A / E314L}}\) were expressed following an adapted version from previously reported protocol \(^{72}\) . Briefly, \(E\) . coli BL21(DE3) containing the gene of interest were grown in \(20 \mathrm{ml}\) of LB Lennox medium (Roth) until an optical density of \(600 \mathrm{nm}\) ( \(\mathrm{OD}_{600}) > 1.5\) was + +<--- Page Split ---> + +reached. Ampicillin (100 \(\mu \mathrm{g / mL}\) ) was used as selecting agent through the expression. Unless otherwise stated, bacteria were grown at \(37^{\circ}\mathrm{C}\) under shaking (220 rpm). Cells for inoculation of \(10\mathrm{mL}\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium with a starting \(\mathrm{OD}_{600}\) of 0.1 were harvested by centrifugation, and excess of TB medium was removed. In all \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal media, \(3\mathrm{g / L}\) of \(^{15}\mathrm{N}\) - ammonium chloride (Deutero) and \(3\mathrm{g / L}\) of deuterated \(^{12}\mathrm{C}\) - glucose (1,2,3,4,5,6,6- d7, Deutero) were used as the principal nitrogen and carbon sources, respectively. The next morning \(2\mathrm{ml}\) of the starter culture were spin- down, supernatant was removed and cells were transferred into \(20\mathrm{mL}\) of freshly prepared \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium. When an \(\mathrm{OD}_{600}\) of 0.4 was reached, the culture volume was increased to \(90\mathrm{ml}\) and cells were grown until an \(\mathrm{OD}_{600}\) of 0.6- 0.8 was reached. At this point, the temperature of the incubator was reduced to \(16^{\circ}\mathrm{C}\) and \(10\mathrm{mL}\) of \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium containing the desired labeled precursors and amino acids were added \(^{72}\) . Amounts of isotopically labelled precursors for selective methyl labeling are given in Supplementary Table 4. Protein expression was induced with the addition of \(1\mathrm{mM}\) or \(0.5\mathrm{mM}\) isopropyl \(\beta\) - D- 1- thio- galactopyranoside (IPTG) for EndoS2 or EndoBT constructs, respectively. Cells were harvested when the maximal cell density was reached and stored at \(- 20^{\circ}\mathrm{C}\) . Labeled protein was purified as previously described \(^{28,38}\) and stored at \(4^{\circ}\mathrm{C}\) . + +## Sample preparation and NMR spectroscopy + +All NMR experiments were acquired on a \(600\mathrm{MHz}\) Avance III spectrometer equipped with TCI cryogenic probe at \(298\mathrm{K}\) , unless otherwise stated. NMR spectra were processed with TopSpin 4.0.6 (Bruker) or with NMRPipe \(^{73}\) prior to analysis. \(^{1}\mathrm{H}\) chemical shifts were referenced to the DSS- \(\mathrm{d}_6\) peak, and \(^{13}\mathrm{C}\) and \(^{15}\mathrm{N}\) signals were referenced indirectly. + +## Enzyme kinetics by NMR spectroscopy + +Rituximab- Fc was concentrated using \(10\mathrm{kDa}\) MWCO Amicon Ultra- 4 filters (Merck) and the buffer was exchanged against NMR kinetics buffer, which contained \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\) (Eurisotop) pH 7.50, \(50\mathrm{mM}\) + +<--- Page Split ---> + +NaCl, \(0.1 \mathrm{mM} 2,2\) - Dimethyl- 2- silapentane- 5- sulfonate- \(\mathrm{d}_{6}\) (DSS- \(\mathrm{d}_{6}\) , Sigma- Aldrich) and \(0.02\%\) \(\mathrm{NaN}_{3}\) in \(8\% \mathrm{D}_{2}\mathrm{O}\) (Eurisotop, \(99.96\%\) ). Samples were prepared per duplicate at 3, 5, 10, 20 and \(40 \mu \mathrm{M}\) Rituximab- Fc. All samples contained a final \(160 \mu \mathrm{l}\) volume in \(3 \mathrm{mm}\) NMR tubes. + +For each kinetics experiment, the sample was introduced in the magnet, locked, matched, tuned, shimmed and the \(90^{\circ}\) pulse was calibrated. The temperature was allowed to equilibrate for 5 min and one perfect CPMG was acquired as reference (for details on the pulse sequence see Supplementary Fig. 17). The sample was removed from the magnet, \(1 \mathrm{nM}\) EndoS \(2_{\mathrm{wt}}(2 \mu \mathrm{l})\) was added directly into the NMR tube, mixed \(3 \mathrm{x}\) by inversion and the tube was immediately inserted into the magnet. After locking and shimming, a series of perfect CPMG spectra were recorded for a maximum of \(24 \mathrm{h}\) . From enzyme addition to the beginning of the first acquisition a max. of \(3 \mathrm{min} 20 \mathrm{s}\) were measured, although \(< 3 \mathrm{min}\) were normally required. Perfect CPMGs were acquired with an echo time \(\tau\) of \(0.3 \mathrm{ms}\) and a total \(T_{2}\) relaxation time of \(39.6 \mathrm{ms}\) , \(T_{1}\) relaxation delay (D1) of \(1 \mathrm{s}\) , \(32\) transients, \(4, 32\) or \(64 \mathrm{dummy}\) scans, an FID size of \(32 \mathrm{k}\) and \(12 \mathrm{ppm}\) sweep width, with a total experimental time of \(2 \mathrm{min}\) . After completion of the reaction, a perfect CPMGs was acquired with a \(T_{1}\) relaxation delay (D1) of \(20 \mathrm{s}\) to correct for insufficient \(T_{1}\) relaxation. + +Reaction time- courses were analyzed from the corrected integrals \((I_{cor})\) obtained from the composite signal between 2.08- 2.11 ppm, which corresponds to \(N\) - acetylglucosamine and \(N\) - acetylneuraminic acid moieties of cleaved \(N\) - glycans (Supplementary Fig. 18A). \(I_{cor}\) were calculated according to Eq. 2. + +\[I_{cor} = (I_{n} - I_{ref})C \quad (Eq. 2)\] + +where \(I_{n}\) corresponds to the integral of the \(n^{th}\) experiment after the addition of enzyme, \(I_{ref}\) indicates the integral of the reference experiment (previous to enzyme addition) and \(C\) denotes a correction factor for + +<--- Page Split ---> + +insufficient \(T_{l}\) relaxation, calculated as the ratio between the integrals of the experiment acquired with \(\mathrm{D1} = 20\mathrm{s}\) and the last spectrum recorded with \(\mathrm{D1} = 1\mathrm{s}\) . \(I_{cor}\) were plotted against time to generate reaction time- courses (Supplementary Fig. 18B), which were then fitted to a single exponential decay Eq. 3. Initial enzyme cycling velocity \(\nu_{0}\) was approximated as the first derivative of Eq. 3 evaluated at time \(= 0\) (Eq. 4). + +\[I_{c o r}(t) = 1 - e^{(-t\cdot a)}\] \[\nu_{0} = \frac{d(1 - e^{-t\cdot a})}{a t}\bigg|_{t = 0} = a\] + +Initial enzyme cycling velocities \(\nu_{0}\) were plotted against the initial concentration of complex glycan covalently attached to Fc or [S] \(_{0}\) , which corresponds to two times the Fc concentration. Fitting to the Michaelis- Menten equation Eq. 5 afforded the substrate concentration- dependence of enzyme activation \((K_{\mathrm{M}})\) and the maximum velocity \((V_{m a x})\) . The maximum enzyme catalytic rate \((k_{c a t})\) was calculated as \(V_{m a x} / [\mathrm{E n z y m e}]_{0}\) . Fittings were performed using in- house Matlab R2019b script, and errors are given as one standard deviation. + +\[\nu_{0} = \frac{V_{m a x}[S]_{0}}{K_{M} + [S]_{0}} \quad (Eq. 5)\] + +where + +\[K_{M} = \frac{k_{o f f} + k_{c a t}}{k_{o n}} \quad (Eq. 6)\] + +## Synthesis and purification of Rituximab-Fc aglycan and \(\mathbf{CT_{n - 1}}\) glycan + +Rituximab- Fc aglycan was obtained by pooling the products of the enzymatic reactions of Rituximab- Fc with EndoS2 used for enzyme kinetics. Rituximab- Fc aglycan was separated from EndoS2 and free + +<--- Page Split ---> + +glycans by size exclusion chromatography (HiLoad 16/600 Superdex 200 pg, GE), using \(20\mathrm{mM}\) Tris- HCl buffer pH 7.5, \(50\mathrm{mM}\) NaCl as buffer. Fractions showing desired protein in UV (280 nm) were pooled, concentrated and the buffer was exchanged against NMR titration buffer as explained above. The final stock solution contained a concentration of \(323\mu \mathrm{M}\) Rituximab- Fc aglycan. + +Glycan \(\mathrm{CT_{n - 1}}\) was obtained by digestion of CT with EndoS2 as explained in the previous chapter (enzyme kinetics by NMR spectroscopy). After reaction completion samples were lyophilized, dissolved in \(\mathrm{H}_2\mathrm{O}\) and the reaction product \(\mathrm{CT_{n - 1}}\) was isolated by HPLC (Jasco Extrema) using a semipreparative VP 250/21 Nucleosil 100- 7 C18 column (Macherey- Nagel, cat. No.: 715352) applying as solvents \(\mathrm{H}_2\mathrm{O}\) and MeOH (both with \(0.01\%\) TFA) with a flow of \(10\mathrm{ml / min}\) and the following gradients: \(20\mathrm{min}0\%\) MeOH followed by \(70\mathrm{min}0\%\) to \(20\%\) MeOH. Fractions of \(10\mathrm{ml}\) were collected, and the desired sugar eluted in fractions 10 to 12, which were pulled together and lyophilized. + +## Assignment of \(^{1}\mathrm{H},^{13}\mathrm{C}\) NMR signals from CT, \(\mathbf{CT_{n - 1}}\) and HM carbohydrates + +The identity and purity of CT, \(\mathrm{CT_{n - 1}}\) and HM carbohydrates was confirmed by NMR prior titrations. For this end, all NMR signals from \(^{1}\mathrm{H},^{13}\mathrm{C}\) - HSQC spectra from samples were assigned. Samples contained \(0.25\mathrm{mM}\) to \(2.0\mathrm{mM}\) carbohydrate concentrations in \(50\mathrm{mM}\) natrium phosphate buffer \(\mathrm{pH}^{*}7.30\) , \(50\mathrm{mM}\) NaCl, \(100\mu \mathrm{M}\) TSP- d4, \(0.02\%\) \(\mathrm{NaN}_3\) in \(\mathrm{D}_2\mathrm{O}\) in \(3\mathrm{mm}\) NMR tubes at \(160\mu \mathrm{l}\) total sample volume. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted in the indirect dimension prior Fourier- transformation (128 LP coefficients), affording a \(4096\mathrm{x}4096\) data matrix. Spectra were manually phased and peak peaking was performed using CCPNMR Analysis 2.4.2 software suit \(^{72}\) . Supplementary Fig. 19 shows a cartoon representation and the chemical structure of the assigned carbohydrates. Experimental details used for the assignment and \(^{1}\mathrm{H}/^{13}\mathrm{C}\) chemical shifts are listed in Supplementary Table 5. Assignments are shown in Supplementary Fig. 20 and Supplementary Table 6. Assignments were in excellent agreement with previously reported frequencies \(^{74,75}\) . All experiments for the assignment of carbohydrates were acquired at \(310\mathrm{K}\) . + +<--- Page Split ---> + +## Titration of selectively MILV methyl-labeled enzymes with carbohydrates, Rituximab-Fc, Rituximab-Fc aglycan and CaCl2 + +MILV methyl- labeled proteins were transferred into NMR titration buffer using \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns (Thermo Fischer Scientific), which contained \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\) pH\\* 7.50, \(50~\mathrm{mM}\) NaCl, \(0.1\mathrm{mM}\) DSS- \(\mathrm{d}_6\) and \(0.02\%\) \(\mathrm{NaN}_3\) in \(\mathrm{D}_2\mathrm{O}\) . Protein samples were prepared at \(45 - 128\mu \mathrm{M}\) protein concentration at \(160~\mu \mathrm{l}\) final volume in \(3\mathrm{mm}\) NMR tubes. Protein concentrations were determined after buffer exchange by UV absorbance at \(280\mathrm{nm}\) with \(\epsilon = 104.17\mathrm{M}^{- 1}\mathrm{cm}^{- 1}\) for EndoS2 and \(\mathrm{EndoS2_{E186L}}\) , and with \(\epsilon = 62.23\mathrm{M}^{- 1}\mathrm{cm}^{- 1}\) for EndoBT- 3987 and EndoBT- 3987D312A/E314L. + +CT, \(\mathrm{CT_{n - 1}}\) , HM and \(\mathrm{CaCl}_2\) used for titrations were dissolved in \(\mathrm{D}_2\mathrm{O}\) and lyophilized twice. Final stock solutions were prepared at high concentrations in NMR titration buffer, and the \(\mathrm{pH}^*\) was carefully readjusted to 7.50. Rituximab- Fc was buffer exchanged against NMR titration buffer using \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns and concentrated to a final \(380~\mu \mathrm{M}\) concentration stock solution. + +\(^{1}\mathrm{H}\) , \(^{13}\mathrm{C}\) HMQC spectra (methyl TROSY) \(^{33}\) were acquired with 107 ms acquisition time and a spectral window of \(4\mathrm{ppm}\) in the direct dimension. In the indirect dimension, the spectral window was set to 22 or 24 ppm with 512 or 256 increments. The relaxation delay was set to 1.5 s and 4 to 16 transients were acquired. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted (64 LP coefficients) prior Fourier- transformation, affording a 2048x2048 data matrix with a spectral resolution of \(1.17\mathrm{Hz}\) and \(1.62\mathrm{Hz}\) in the direct and indirect dimensions, respectively. + +Lineshape analysis of cross peaks of \(^{1}\mathrm{H}\) , \(^{13}\mathrm{C}\) HMQC spectra of MILV- labeled samples with the program TITAN \(^{34}\) was performed using Matlab R2019b for titrations shown in Table 1. Spectra were processed in NMRPipe \(^{73}\) prior to analysis. A representative example of the shell scripts used is compiled in Supplementary Table 7. We selected a two states binding model as the simplest model appropriately describing the experimental results (Eq. 1). For each titration series the fitting algorithm was as follows: first, linewidths and chemical shift were fitted using the spectrum from the apo form. Next, linewidths + +<--- Page Split ---> + +and chemical shift of the fully bound state were fitted using the spectrum of the protein at highest ligand concentration. Linewidths, off- rate constant \(k_{off}\) and dissociation constant \(K_{D}\) were finally fitted from spectra of all titration points and using previously determined linewidths and chemical shifts as starting values. Parameter uncertainties were obtained by bootstrap error analysis with 100 iterations. + +Titrations of \(\mathrm{CaCl_2}\) showed fast- exchange in the NMR time- scale between the unbound and bound forms. Therefore, signal position were extracted with CCPNMR Analysis v2.4.2 software package \(^{76}\) and used obtain Euclidean chemical shift perturbances (CSPs) measured as Euclidian distances \(\Delta \nu_{\mathrm{Eucl}}\) according to Eq. 6: + +\[\Delta \nu_{Eucl} = \sqrt{\Delta\nu_{H}^{2} + \Delta\nu_{C}^{2}} \quad (Eq. 6)\] + +with \(\Delta \nu_{H}\) and \(\Delta \nu_{C}\) being the CSPs in the respective dimension in Hz. In a simple two states model (Eq. 1), observed CSPs \(\Delta \nu_{obs}\) at a given total ligand concentration \(L_{t}\) are linked to the dissociation constant \(K_{D}\) via the law of mass action \(^{32}\) (Eq. 7). + +\[\Delta \nu_{obs} = \frac{(P_{t} + L_{t} + K_{D}) - \sqrt{(P_{t} + L_{t} + K_{D})^{2} - 4P_{t}L_{t}}}{2P_{t}}\Delta \nu_{max} \quad (Eq. 7)\] + +where \(P_{t}\) is the total protein concentration, and \(\Delta \nu_{max}\) is the maximum CSP at ligand saturation for each signal. Global fittings were performed using in- house Matlab R2019b script. Errors were determined from a Monte Carlo approach with 100 iterations as previously described \(^{77}\) , and are given as one standard deviation. + +<--- Page Split ---> + +## \(^{1}\mathbf{H},^{15}\mathbf{N}\) HSQC-TROSY spectra and determination of rotational correlation times \(\tau_{c}\) + +\([U_{- }^{15}\mathrm{N},^{2}\mathrm{H}]\) MILV methyl- labeled proteins were buffer exchanged \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns against the following buffer: \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\mathrm{pH}^{*}7.50,50\mathrm{mM}\) NaCl, \(0.1\mathrm{mM}\) DSS- \(\mathrm{d}_6\) and \(0.02\%\) \(\mathrm{NaN}_3\) in \(8\%\) \(\mathrm{D}_2\mathrm{O}\) . Two samples containing \(\mathrm{EndoS2_{E186L}}\) and \(\mathrm{EndoBT_{D312A / E314L}}\) were prepared at \(71.5\mu \mathrm{M}\) or \(74.4\mu \mathrm{M}\) protein concentrations, respectively. Samples were allowed to deuterium- hydrogen back exchange for 5 days at \(6^{\circ}\mathrm{C}\) , and then measured in \(3\mathrm{mm}\) NMR tubes at a final \(160\mu \mathrm{l}\) volume. \(^{1}\mathrm{H},^{15}\mathrm{N}\) HSQC- TROSY spectra were acquired were acquired with \(128\mathrm{ms}\) acquisition time and a spectral window of \(13.3\mathrm{ppm}\) in the direct dimension. In the indirect dimension, the spectral window was set to \(45\mathrm{ppm}\) with 512 increments. The relaxation delay was set to \(1.5\mathrm{s}\) and 16 (EndoBT- 3987 \(\mathrm{D_{312A / E314L}}\) ) or 48 (EndoS2 \(\mathrm{E_{186L}}\) ) transients were acquired. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted (16 LP coefficients) prior Fourier- transformation, affording a \(4096\times 1024\) data matrix with a spectral resolution of \(1.95\mathrm{Hz}\) and \(2.67\mathrm{Hz}\) in the direct and indirect dimensions, respectively. + +Rotational correlation times \(\tau_{c}\) of MILV \(\mathrm{EndoS2_{E186L}}\) and \(\mathrm{EndoBT_{D312A / E314L}}\) proteins have been estimated from the samples prepared above using \(^{1}\mathrm{H},^{15}\mathrm{N}\) - TRACT experiments \(^{78}\) . The relaxation delay was set to 2 s. Experiments were measured for 40 transients with 30 increasing delays of up to 1.5 s. Data were integrated from 8- 10 ppm, normalized and fitted to an exponential decay model for determination of average \(^{15}\mathrm{N} R_{\alpha}\) and \(R_{\beta}\) . + +## Data availability + +The atomic coordinates and structure factors have been deposited with the Protein Data Bank (PDB), accession code 8A64 (EndoS \(\mathrm{E_{235A - Fc}}\) ). Previously published PDB structures used in this study are available under the accession codes: 6MDS, 4NUZ, 1H3X and 6EN3. All other data are available from the corresponding authors upon reasonable request. Source data are provided with this paper. + +<--- Page Split ---> + +## References + +<--- Page Split ---> + +Specific Labeling and Conjugation of Antibodies: Facile Synthesis of Homogeneous Antibody- Drug Conjugates. ACS Chem. Biol. 16, 2502- 2514 (2021).19. Shi, W. et al. One- step synthesis of site- specific antibody- drug conjugates by reprogramming IgG glycoengineering with LacNAc- based substrates. Acta Pharm. Sin. B 12, 2417- 2428 (2022).20. Pučić, M. et al. High throughput isolation and glycosylation analysis of IgG- variability and heritability of the IgG glycome in three isolated human populations. Mol. Cell. Proteomics (2011). doi:10.1074/mcp.M111.01009021. Mimura, Y. et al. Glycosylation engineering of therapeutic IgG antibodies: challenges for the safety, functionality and efficacy. Protein Cell 9, 47- 62 (2018).22. Lu, R.- M. et al. Development of therapeutic antibodies for the treatment of diseases. J. Biomed. Sci. 27, 1 (2020).23. Huang, W., Giddens, J., Fan, S.- Q. Q., Toonstra, C. & Wang, L.- X. X. Chemoenzymatic glycoengineering of intact IgG antibodies for gain of functions. J. Am. Chem. Soc. 134, 12308- 12318 (2012).24. Li, T., Tong, X., Yang, Q., Giddens, J. P. & Wang, L. X. Glycosynthase mutants of endoglycosidase S2 show potent transglycosylation activity and remarkably relaxed substrate specificity for antibody glycosylation remodeling. J. Biol. Chem. 29, 16508- 16518 (2016).25. Majewska, N. I., Tejada, M. L., Betenbaugh, M. J. & Agarwal, N. N- glycosylation of IgG and IgG- like recombinant therapeutic proteins: Why is it important and how can we control it? Annu Rev Chem Biomol Eng 7, 311- 338 (2020).26. Trastoy, B. et al. Crystal structure of Streptococcus pyogenes EndoS, an immunomodulatory endoglycosidase specific for human IgG antibodies. Proc. Natl. Acad. Sci. 111, 6714- 9 (2014).27. Trastoy, B. et al. Structural basis for the recognition of complex- type N- glycans by Endoglycosidase S. Nat. Commun. 9, 1874 (2018).28. Klontz, E. H. et al. Molecular Basis of Broad Spectrum N- Glycan Specificity and Processing of Therapeutic IgG Monoclonal Antibodies by Endoglycosidase S2. ACS Cent. Sci. 5, 524- 538 (2019).29. Kastner, B. et al. GraFix: Sample preparation for single- particle electron cryomicroscopy. Nat. Methods 5, 53- 55 (2008).30. Krissinel, E. & Henrick, K. Inference of Macromolecular Assemblies from Crystalline State. J. Mol. Biol. (2007). doi:10.1016/j.jmb.2007.05.02231. Dixon, E. V. et al. Fragments of bacterial endoglycosidase S and immunoglobulin G reveal subdomains of each that contribute to deglycosylation. J. Biol. Chem. 289, 13876- 13889 (2014).32. Williamson, M. P. Using chemical shift perturbation to characterise ligand binding. Prog. Nucl. Magn. Reson. Spectrosc. 73, 1- 16 (2013).33. Tugarinov, V., Hwang, P. M., Ollerenshaw, J. E. & Kay, L. E. Cross- correlated relaxation enhanced 1H[bond]13C NMR spectroscopy of methyl groups in very high molecular weight proteins and protein complexes. J. Am. Chem. Soc. 125, 10420- 10428 (2003).34. Waudby, C. A., Ramos, A., Cabrita, L. D. & Christodoulou, J. Two- Dimensional NMR Lineshape Analysis. Sci. Rep. 6, 24826 (2016).35. Schütz, S. & Sprangers, R. Methyl TROSY spectroscopy: A versatile NMR approach to study + +<--- Page Split ---> + +challenging biological systems. Prog. Nucl. Magn. Reson. Spectrosc. 116, 56- 84 (2020).36. Tugarinov, V., Kanelis, V. & Kay, L. E. Isotope labeling strategies for the study of high- molecular- weight proteins by solution NMR spectroscopy. Nat. Protoc. 1, 749- 754 (2006).37. Tugarinov, V. & Kay, L. E. An isotope labeling strategy for methyl TROSY spectroscopy. J. Biomol. NMR 28, 165- 172 (2004).38. Trastoy, B. et al. Structural basis of mammalian high- mannose N- glycan processing by human gut Bacteroides. Nat. Commun. 11, 899 (2020).39. Trastoy, B. et al. GH18 endo- \(\beta\) - N- acetylglucosaminidases use distinct mechanisms to process hybrid- type N- linked glycans. J. Biol. Chem. 297, 101011 (2021).40. Waddling, C. A., Plummer, T. H. J., Tarentino, A. L. & Van Roey, P. Structural basis for the substrate specificity of endo- beta- N- acetylglucosaminidase F(3). Biochemistry 39, 7878- 7885 (2000).41. Deisenhofer, J. Crystallographic refinement and atomic models of a human Fc fragment and its complex with fragment B of protein A from Staphylococcus aureus at 2.9- and 2.8- A resolution. Biochemistry 20, 2361- 2370 (1981).42. Sauer- Eriksson, A. E., Kleywegt, G. J., Uhlen, M. & Jones, T. A. Crystal structure of the C2 fragment of streptococcal protein G in complex with the Fc domain of human IgG. Structure 3, 265- 278 (1995).43. Burmeister, W. P., Huber, A. H. & Bjorkman, P. J. Crystal structure of the complex of rat neonatal Fc receptor with Fc. Nature 372, 379- 383 (1994).44. Barb, A. W. & Prestegard, J. H. NMR analysis demonstrates immunoglobulin G N- glycans are accessible and dynamic. Nat. Chem. Biol. (2011). doi:10.1038/nchembio.51145. Yamaguchi, Y. & Barb, A. W. A synopsis of recent developments defining how N- glycosylation impacts immunoglobulin G structure and function. Glycobiology 30, 214- 225 (2020).46. Stark, H. GraFix: Stabilization of fragile macromolecular complexes for single particle Cryo- EM. Methods in Enzymology 481, (Elsevier Inc., 2010).47. Kandiah, E. et al. CM01: A facility for cryo- electron microscopy at the European synchrotron. Acta Crystallogr. Sect. D Struct. Biol. 75, 528- 535 (2019).48. Zivanov, J., Nakane, T. & Scheres, S. H. W. Estimation of high- order aberrations and anisotropic magnification from cryo- EM data sets in RELION- 3.1. IUcrJ 7, 253- 267 (2020).49. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nat. Methods 14, 290- 296 (2017).50. Punjani, A., Zhang, H. & Fleet, D. J. Non- uniform refinement: adaptive regularization improves single- particle cryo- EM reconstruction. Nat. Methods 17, 1214- 1221 (2020).51. Sanchez- Garcia, R. et al. DeepEMhancer: a deep learning solution for cryo- EM volume post- processing. Commun. Biol. 4, 874 (2021).52. Cardone, G., Heymann, J. B. & Steven, A. C. One number does not fit all: mapping local variations in resolution in cryo- EM reconstructions. J. Struct. Biol. 184, 226- 236 (2013).53. Pettersen, E. F. et al. UCSF Chimera - A visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605- 1612 (2004). + +<--- Page Split ---> + +54. Adams, P. D. et al. PHENIX: A comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 213–221 (2010).55. Afonine, P. V et al. Real-space refinement in PHENIX for cryo-EM and crystallography. Acta Crystallogr. Sect. D, Struct. Biol. 74, 531–544 (2018).56. Emsley, P. & Cowtan, K. Coot: Model-building tools for molecular graphics. Acta Crystallogr. Sect. D Biol. Crystallogr. (2004). doi:10.1107/S090744490401915857. Chen, V. B. et al. MolProbity: All-atom structure validation for macromolecular crystallography. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 12–21 (2010).58. Nakane, T., Kimanius, D., Lindahl, E. & Scheres, S. H. Characterisation of molecular motions in cryo-EM single-particle data by multi-body refinement in RELION. Elife 7, (2018).59. Crowther, R. A., Henderson, R. & Smith, J. M. MRC image processing programs. J. Struct. Biol. 116, 9–16 (1996).60. Rambo, R. P. & Tainer, J. A. Characterizing flexible and intrinsically unstructured biological macromolecules by SAS using the Porod-Debye law. Biopolymers 95, 559–571 (2011).61. Petoukhov, M. V. et al. New developments in the ATSAS program package for small-angle scattering data analysis. J. Appl. Crystallogr. 45, 342–350 (2012).62. Svergun, D. I. Determination of the regularization parameter in indirect-transform methods using perceptual criteria. J. Appl. Crystallogr. 25, 495–503 (1992).63. Tuukkanen, A. T., Kleywegt, G. J. & Svergun, D. I. Resolution of ab initio shapes determined from small-angle scattering. IUCrJ 3, 440–447 (2016).64. Johnson, K. A. Fitting enzyme kinetic data with KinTek Global Kinetic Explorer. Methods Enzymol. 467, 601–626 (2009).65. Case, D. A. et al. AMBER 2020. (2020).66. Maier, J. A. et al. ff14SB: Improving the Accuracy of Protein Side Chain and Backbone Parameters from ff99SB. J. Chem. Theory Comput. 11, 3696–3713 (2015).67. Kirschner, K. N. et al. GLYCAM06: a generalizable biomolecular force field. Carbohydrates. J. Comput. Chem. 29, 622–655 (2008).68. Jorgensen, W. L., Chandrasekhar, J. & Madura, J. D. Comparison of simple potential functions for simulating liquid water. J. Chem. Phys. 79, 926–935 (1983).69. Andersen, H. C. No TitleMolecular dynamics simulations at constant pressure and/or temperature. J. Chem. Phys. 72, 2384–2393 (1980).70. Miyamoto, S. & Kollman, P. A. Settle: An analytical version of the SHAKE and RATTLE algorithm for rigid water model. J. Comput. Chem. 13, 952–962 (1992).71. Darden, T. & York, D. Particle mesh Ewald: An N·log(N) method for Ewald sums in large systems. J. Chem. Phys. 98, 10089–10092 (1993).72. Müller-Hermes, C., Creutznacher, R. & Mallagaray, A. Complete assignment of Ala, Ile, Leu(ProS), Met and Val(ProS) methyl groups of the protruding domain from human norovirus GII.4 Saga. Biomol. NMR Assign. 14, 123–130 (2020).73. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995). + +<--- Page Split ---> + +74. Unione, L. et al. Glycoprofile Analysis of an Intact Glycoprotein As Inferred by NMR Spectroscopy. ACS Cent. Sci. 5, 1554–1561 (2019).75. Schubert, M., Walczak, M. J., Aebi, M. & Wider, G. Posttranslational modifications of intact proteins detected by NMR spectroscopy: application to glycosylation. Angew. Chem. Int. Ed. Engl. 54, 7096–7100 (2015).76. Vranken, W. F. et al. The CCPN data model for NMR spectroscopy: development of a software pipeline. Proteins 59, 687–696 (2005).77. Arai, M., Ferreon, J. C. & Wright, P. E. Quantitative analysis of multisite protein-ligand interactions by NMR: binding of intrinsically disordered p53 transactivation subdomains with the TAZ2 domain of CBP. J. Am. Chem. Soc. 134, 3792–3803 (2012).78. Lee, D., Hilty, C., Wider, G. & Wüthrich, K. Effective rotational correlation times of proteins from NMR relaxation interference. J. Magn. Reson. 178, 72–76 (2006).79. Collin, M. & Olsén, A. Extracellular enzymes with immunomodulating activities: variations on a theme in Streptococcus pyogenes. Infect. Immun. 71, 2983–2992 (2003).80. Kagawa, T. F. et al. Crystal structure of the zymogen form of the group A Streptococcus virulence factor SpeB: an integrin-binding cysteine protease. Proc. Natl. Acad. Sci. U. S. A. 97, 2235–2240 (2000).81. Wenig, K. et al. Structure of the streptococcal endopeptidase IdeS, a cysteine proteinase with strict specificity for IgG. Proc. Natl. Acad. Sci. U. S. A. 101, 17371–17376 (2004). + +<--- Page Split ---> + +## Acknowledgments + +This work was supported by the MICINN/FEDER EU grant PID2019- 105649RB- I00 (MEG), MINECO/FEDER EU grant Severo Ochoa Excellence Accreditation grant SEV- 2016- 0644 (MEG), Basque Government contract KK- 2021- 00034 (MEG), National Institutes of Health grant R01AI149297 (EJS, LXW, MEG), National Institutes of Health grant R01GM096973 (LXW), European Union Horizon 2020, Marie Sklodowska- Curie grant 844905 and "Ramón y Cajal" fellow from the Spanish Ministry of Economy and Competitiveness (BT), La Caixa Foundation grant LCF/BQ/DR19/11740011 (MGA), European Fonds for Regional Development, LPW- E/1.1.2/857 (AM), Agencia Estatal Investigación of Spain (AEI, RTI- 2018- 099592- B- C21, FC), European Union's Horizon 2020 research and innovation program under the Marie Sklodowska Curie grant agreement no. 956544 (FC). + +## Author contributions + +BT, JJD, LXW, FC, AM, EJS and MEG, conceived the project. BT, JJD, JOC, LR, MGA, EHK, DD, CL, DES, FC and AM, performed the experiments. BT, JJD, JOC, FC, AM, EJS and MEG, analyzed the results. BT, JJD, FC, AM, EJS and MEG, wrote the paper. + +## Competing interests + +Authors declare that they have no competing interests. + +## Additional information + +The online version contains supplementary material available at https://www.nature.com/ncomms/. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 Overall structure of EndoS in complex with the Fc region. A EndoS, SpeB and IdeS are secreted enzymes from S. pyogenes that can act on IgG antibodies. SpeB and IdeS are proteases that hydrolyze the flexible hinge region between residues 220 to 248 of the IgG antibody \(^{79 - 81}\) . EndoS is an
+ +<--- Page Split ---> + +ENGase that hydrolyze the \(\beta - 1,4\) linkage between the first two GlcNAcs of the CT \(N\) - glycans on the Fc region of IgG antibodies. The three enzymes inactivate IgG antibodies against S. pyogenes. B EndoS and EndoS2 can be used to remodel the \(N\) - glycan on therapeutic IgG antibodies using a chemoenzymatic approach. First, wild type EndoS and/or EndoS2 hydrolyze a variety of glycoforms present in the Fc region according to their \(N\) - glycan specificity. Next, other enzymes can be used to hydrolyze specific carbohydrate moieties, e.g. fucose which absence improves the binding to FcγRIIIa and the antibody- dependent cellular cytotoxicity (ADCC) properties of the antibody. Last, EndoS \(D_{233Q}\) and EndoS \(2_{D184M}\) glycosynthase mutants facilitates the transfer of a glycan- oxazoline donor with a defined glycoform to the Fc region of IgG antibodies. The dotted shapes represent that carbohydrate may or may not be present, indicating the high heterogeneity of the \(N\) - glycan which impact the functionality, immunogenicity and pharmacokinetic of the antibody. C CryoEM maps of the EndoS \(E_{235A}\) - Fc complex, including a schematic representation of the EndoS domains and the Fc region in the complex structure. D Cartoon representation showing the overall fold and the secondary structure organization of the EndoS- Fc complex. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 The Fc binding site of EndoS. Two views of the interaction interface between the GH domain of EndoS (yellow) and FcP1 of the Fc region (orange) (A), the \(N\) -glycan (blue and green) of the Fc region and the active site of EndoS (yellow) (B) the Cy2-Cy3 joint region of the Fc (orange) and the b-sandwich domain of EndoS (blue) (C).
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Alanine scanning mutagenesis of the EndoS-Fc interface. A Structure of the EndoS (grey) complex with Fc (pink) in which the regions with alanine mutants are color coded based on the relative rates of removal of the \(N\) -glycans from Fc alanine mutants by EndoS and vice versa. The rates were measured via LC-MS kinetic analyses. Red indicates the largest increase in hydrolytic activity while blue indicates no hydrolytic activity. B Rates of removal of the first glycan of Fc and Fc alanine mutants by EndoS alanine mutants and EndoS respectively, normalized to EndoS vs. Fc. All rates were statistically significant (multiple comparison test, Tukey method, \(p< 0.0001\) ).
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 Molecular dynamics (MD) simulations studies. Overlay of 25 frames evenly spaced along the \(1\mu \mathrm{s}\) MD trajectory for (A) EndoS-Fc, (B) mutant E293A and (C) mutant W803A, along with the average distance between the center of the aromatic ring of the residues involved in the stabilizing interactions derived from the MD simulations. In the case of L314, S298 and E354 Cδ, the oxygen of -OH and the oxygens of the -CO₂⁻ are considered, respectively. Errors are given as SD. The EndoS protein is shown as blue ribbons and the antibody as gray and orange ribbons.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 Molecular mechanism of antibody recognition by EndoS and EndoS2. Kinetic modeling of deglycosylation of (A) FcWT and (B) IgGWT by EndoSWT and (C) IgGWT by EndoS2WT. LC-MS kinetic analysis of deglycosylation of FcWT and IgGWT by EndoSWT- showing the hydrolysis over time of the two \(N\) -glycans linked to Asn297. The deglycosylation process occurs sequentially as seen by the initial increase then decrease in monoglycosylated population (dark green) over time while the diglycosylated (purple) and deglycosylated (orange) species increase over time. D Schematic representation of Fc \(N\) -glycan processing by EndoS. In a first step, EndoS binds the diglycosylated Fc and hydrolyzes one of the \(N\) -glycan of the Fc which causes the release of monoglycosylated Fc. In a second step, EndoS engages the monoglycosylated Fc and hydrolyzes the remaining \(N\) -glycan of the Fc to produced completely deglycosylated Fc.
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6 Binding affinity and enzyme kinetics experiments studied by NMR. A Crystal structure of EndoS2 bound to CT \(N\) -glycan and \(\mathrm{Ca}^{2 + }\) (PDB code 6MDS) showing the location of the \(\left[^{13}\mathrm{C},\mathrm{H}_{3}\right]\) - methyl labeled probes as black spheres. Glycoside hydrolase (GH) and \(\beta\) -sandwich (BS) domains are colored in yellow and pale blue, respectively. Mutated amino acid E186L, CT glycan and \(\mathrm{Ca}^{2 + }\) ion are depicted in pink, dark blue and pale green. B Superimposition of selected regions of methyl-TROSY spectra showing CSPs upon mutagenesis or at saturation concentrations of different ligands. Color code corresponds to: EndoS2 (violet), EndoS2E186L in the absence (black) and in the presence of CT (red), \(\mathrm{CT}_{\mathrm{n - 1}}\) (orange), IgG1 Fc (blue) and \(\mathrm{Ca}^{2 + }\) (yellow). Addition of \(\mathrm{Fc}_{\mathrm{aglyc}}\) (green) induced to CSPs, suggesting very weak protein-protein interactions. Arrows indicate the extent of CSPs observed between the different protein states. C Ligands used for NMR titration experiments. IgG1 Fc region (Fc), complex-type carbohydrate (CT) and the products of glycan hydrolysis \(\mathrm{Fc}_{\mathrm{aglyc}}\) and \(\mathrm{CT}_{\mathrm{n - 1}}\) , respectively. Color code like in Fig. 1. D Quantum mechanics-based 2D lineshape fitting of
+ +<--- Page Split ---> + +methyl- TROSY spectra of a representative signal for the titration of EndoS2E186L with Fc. Experimental and fitted spectra are shown as black and pink surfaces, respectively. Inserts correspond to 0, 5.5, 9.8, 20.3, 29.6 and \(56~\mu \mathrm{M}\) Fc concentrations. \(\mathbf{E}\) [S]0- dependence of the initial enzyme cycling velocity. The solid line represents the best- fit to Michaelis Menten Eq. 5, and dotted lines correspond to one standard deviation. \(\mathbf{F}\) Schematic mechanism of Fc \(N\) - glycan processing by EndoS2 as determined from lineshape analysis. The affinity of EndoS2 (here E) for Fc is two orders of magnitude higher as compared to the reaction product \(\mathrm{CT_{n - 1}}\) , promoting substrate depletion. On- rates are responsible for increased affinity towards Fc as compared to \(\mathrm{CT_{n - 1}}\) , suggesting differences in the initial enzyme- ligand encounter event. All experiments were acquired at \(600\mathrm{MHz}\) and \(298\mathrm{K}\) . + +<--- Page Split ---> + +Table 1. Binding affinity of EndoS2 for IgG antibodies and N-glycans. Dissociation constants $K_{D}$ , on- and off-rate constants $k_{on}$ and $k_{off}$ for Rituximab-Fc, CT, $\text {CT}_{\text {n-1}}$ HM and $\text {Ca}^{2+}$ binding to EndoS2E186L, EndoS2 and EndoBT-3987D312A/E314L from fitting a two-state binding model to methyl-TROSY titration data using TITAN algorithm. + +
\(N^{\circ }\)EnzymeLigand\(K_{D}(\mu M)\)\(k_{off}(s^{-1})\)\(k_{on}(M^{-1}s^{-1})\)
1EndoS2E186LFc\(3.1\pm 0.6\)\(4.0\pm 1.5\)\(1.3\times 10^{6}\pm 5.5\times 10^{5}\)
2EndoBT-3987D312A/E314LHM\(1.7\pm 0.9\)\(5.5\pm 1.4\)\(3.2\times 10^{6}\pm 1.9\times 10^{5}\)
3EndoS2E186LCT\(106.9\pm 6.7\)\(11.4\pm 1.5\)\(1.1\times 10^{5}\pm 1.6\times 10^{4}\)
4EndoS2E186L\(\text {CT}_{\text {n-1}}\)\(150.7\pm 18.0\)\(9.4\pm 3.7\)\(6.2\times 10^{4}\pm 2.6\times 10^{4}\)
5EndoS2E186L\(\text {Fc}_{\text {aglyc}}\)---
6EndoS2E186L\(\text {CaCl}_{2}\)\(3415\pm 480\)--
7EndoS2\(\text {CaCl}_{2}\)\(4000\pm 100\)--
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +2022TRASTOYFCNCSIDEF.pdfSupplementaryVideo1.movSupplementaryVideo2.movSupplementaryVideo3.mov + +<--- Page Split ---> diff --git a/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b_det.mmd b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..ebd033c212d9d351a68c0ad5184494eca2d4f500 --- /dev/null +++ b/preprint/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b/preprint__13b8efb04fff3bd1cfd3187896b6c69c5161fe46c9d80e042c825dae0c51e57b_det.mmd @@ -0,0 +1,573 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 931, 210]]<|/det|> +# Mechanism of antibody-specific deglycosylation and immune evasion by Streptococcus IgG-specific endoglycosidases + +<|ref|>text<|/ref|><|det|>[[44, 230, 530, 272]]<|/det|> +Beatrix Trastoy CIC bioGUNE https://orcid.org/0000- 0002- 2178- 732X + +<|ref|>text<|/ref|><|det|>[[44, 278, 204, 318]]<|/det|> +Jonathan Du Emory University + +<|ref|>text<|/ref|><|det|>[[44, 325, 530, 365]]<|/det|> +Javier Cifuente IIS Biocruces https://orcid.org/0000- 0001- 5420- 121X + +<|ref|>text<|/ref|><|det|>[[44, 371, 245, 410]]<|/det|> +Lorena Rudolph Universitat zu Lubeck + +<|ref|>text<|/ref|><|det|>[[44, 416, 198, 455]]<|/det|> +Mikel Garcia- Alija CSIC- UPV + +<|ref|>text<|/ref|><|det|>[[44, 462, 348, 503]]<|/det|> +Erik Klontz University of Maryland, Baltimore + +<|ref|>text<|/ref|><|det|>[[44, 509, 792, 550]]<|/det|> +Daniel Deredge University of Maryland School of Pharmacy https://orcid.org/0000- 0002- 6897- 6523 + +<|ref|>text<|/ref|><|det|>[[44, 556, 560, 596]]<|/det|> +Nazneen Sultana Emory University https://orcid.org/0000- 0001- 9866- 4443 + +<|ref|>text<|/ref|><|det|>[[44, 602, 204, 641]]<|/det|> +Chau Huynh Emory University + +<|ref|>text<|/ref|><|det|>[[44, 648, 204, 687]]<|/det|> +Maria Flowers Emory University + +<|ref|>text<|/ref|><|det|>[[44, 694, 254, 733]]<|/det|> +Chao Li University of Maryland + +<|ref|>text<|/ref|><|det|>[[44, 740, 204, 780]]<|/det|> +Diego Sastre Emory University + +<|ref|>text<|/ref|><|det|>[[44, 787, 722, 827]]<|/det|> +Lai- Xi Wang University of Maryland College Park https://orcid.org/0000- 0003- 4293- 5819 + +<|ref|>text<|/ref|><|det|>[[44, 833, 600, 874]]<|/det|> +Francisco Corzana University of La Rioja https://orcid.org/0000- 0001- 5597- 8127 + +<|ref|>text<|/ref|><|det|>[[44, 880, 914, 943]]<|/det|> +Alvaro Mallagaray Institute of Chemistry and Metabolomics, Center of Structural and Cell Biology in Medicine (CSCM), University of Lubeck https://orcid.org/0000- 0001- 5825- 4407 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 43, 600, 130]]<|/det|> +Eric J. Sundberg Emory University https://orcid.org/0000- 0003- 0478- 3033 Dr. Marcelo E. Guerin (☑ mrcguerin@gmail.com ) IIS BioCruces Bizkaia https://orcid.org/0000- 0001- 9524- 3184 + +<|ref|>sub_title<|/ref|><|det|>[[44, 172, 102, 190]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 210, 135, 228]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 247, 308, 266]]<|/det|> +Posted Date: June 22nd, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 285, 474, 304]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1774503/v1 + +<|ref|>text<|/ref|><|det|>[[44, 322, 910, 365]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[145, 77, 850, 120]]<|/det|> +# Mechanism of antibody-specific deglycosylation and immune evasion by Streptococcus IgG-specific endoglycosidases + +<|ref|>text<|/ref|><|det|>[[95, 147, 904, 221]]<|/det|> +Beatrix Trastoy, \(^{1,2,\# *}\) Jonathan J. Du, \(^{3,\#}\) Javier O. Cifuente, \(^{1,2,\#}\) Lorena Rudolph, \(^{4}\) Mikel García- Alija, \(^{1,2}\) Erik H. Klontz, \(^{5,6}\) Daniel Deredge, \(^{7}\) Nazneen Sultana, \(^{3}\) Chau G. Huynh, \(^{3}\) Maria Flowers, \(^{3}\) Chao Li, \(^{8}\) Diego E. Sastre, \(^{3}\) Lai- Xi Wang, \(^{8}\) Francisco Corzana, \(^{9}\) Alvaro Mallagaray \(^{4*}\) , Eric J. Sundberg \(^{3*}\) and Marcelo E. Guerin \(^{1,2,10*}\) + +<|ref|>text<|/ref|><|det|>[[87, 242, 911, 545]]<|/det|> +\(^{1}\) Structural Glycobiology Laboratory, Biocrudes Health Research Institute, Barakaldo, Bizkaia, 48903, Spain \(^{2}\) Structural Glycobiology Laboratory, Center for Cooperative Research in Biosciences (CIC bioGUNE), Basque Research and Technology Alliance (BRTA), Bizkaia Technology Park, Building 801A, 48160 Derio, Spain \(^{3}\) Department of Biochemistry, Emory University School of Medicine, Atlanta, GA 30322, USA \(^{4}\) University of Lübeck, Center of Structural and Cell Biology in Medicine (CSCM), Institute of Chemistry and Metabolomics, Ratzeburger Allee 160, 23562 Lübeck, Germany \(^{5}\) Department of Microbiology and Immunology, University of Maryland School of Medicine, Baltimore, MD 21201, USA \(^{6}\) Institute of Human Virology, University of Maryland School of Medicine, Baltimore, MD 21201, USA \(^{7}\) Department of Pharmaceutical Sciences, University of Maryland School of Pharmacy, Baltimore MD 21201, USA \(^{8}\) Department of Chemistry and Biochemistry, University of Maryland, College Park, MD 20742, USA \(^{9}\) Departamento Química and Centro de Investigación en Síntesis Química, Universidad de La Rioja, 26006 Rioja, Spain \(^{10}\) Ikerbasque, Basque Foundation for Science, 48009 Bilbao, Spain + +<|ref|>text<|/ref|><|det|>[[88, 556, 464, 575]]<|/det|> +\(^{#}\) These authors contributed equally to this work + +<|ref|>text<|/ref|><|det|>[[88, 591, 830, 628]]<|/det|> +\*Corresponding authors. Email: beatriz.trastoy@gmail.com, alvaro.mallagaraydebenito@uniluebeck.de, eric.sundberg@emory.edu and mrcguerin@gmail.com + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 166, 88]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[86, 102, 912, 510]]<|/det|> +Bacterial pathogens have evolved diverse and intricate mechanisms to evade the human immune system, including the production of immunomodulatory enzymes. Streptococcus pyogenes secretes two multimodular endo- \(\beta\) - N- acetylglucosaminidases, EndoS and EndoS2, that specifically deglycosylate the conserved \(N\) - linked glycan at Asn297 on IgG Fc, disabling antibody- mediated effector functions. Amongst thousands of known carbohydrate- active enzymes, EndoS and EndoS2 represent just a handful of enzymes that are specific to the protein portion of the glycoprotein substrate, not just the glycan component. Here, we present the cryoEM structure of EndoS in complex with the IgG1 Fc fragment. In combination with small- angle X- ray scattering, alanine scanning mutagenesis, hydrolytic activity measurements, enzyme kinetics, nuclear magnetic resonance and molecular dynamics analysis, we establish the mechanisms of recognition and specific deglycosylation of IgG antibodies by EndoS and EndoS2. Our results provide a rational basis from which to engineer novel enzymes with antibody and glycan selectivity for clinical and biotechnological applications. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 199, 88]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[86, 100, 912, 901]]<|/det|> +Streptococcus pyogenes (group A Streptococcus; GAS) is a Gram- positive pathogenic bacterium that causes life- threating conditions and post- infection immune- related diseases, in addition to mild skin and upper respiratory tract infections \(^{1}\) . In order to evade the immune system and facilitate the colonization of the host, GAS employs a large variety of mechanisms, most notably the secretion of enzymes that selectively inactivate key molecules of the immune system, such as immunoglobulin G (IgG) antibodies (Fig. 1A) \(^{2}\) . EndoS \(^{3}\) and EndoS \(^{2}\) \(^{4}\) are endo- \(\beta\) - N- acetylglucosaminidases (ENGases) produced by GAS serotypes M1 and M49, respectively, that hydrolyze the Asn297- linked \(N\) - glycan on the fragment crystallizable (Fc) region of IgG antibodies. These ENGases are the most prominent members of a rare class of carbohydrate- active enzymes that are specific to the protein component of the glycoprotein substrate and not just the glycan component; no mechanism of protein specificity by this class of enzymes has yet been described. The Asn297- linked \(N\) - glycan is required for IgG binding to Fc \(\gamma\) receptors (Fc \(\gamma\) Rs) and complement C1q, which trigger antibody- mediated effector functions \(^{5 - 8}\) . The activity of these enzymes thereby abolishes the effector functions mediated by the Fc region of IgG antibodies, increasing the survival and virulence of the bacteria \(^{9}\) . Accordingly, IgG- specific ENGases have been shown to ameliorate autoimmune disease in diverse animal models \(^{10 - 13}\) . In addition, EndoS and EndoS2 are powerful tools for the chemoenzymatic synthesis of antibodies with homogenous glycoforms \(^{14 - 17}\) and drug conjugates \(^{18,19}\) . The Fc region \(N\) - glycans of IgG antibodies in serum are composed of more than 33 distinct glycoforms \(^{20}\) and their unique chemical structures modulate the effector functions by altering Fc binding to Fc \(\gamma\) Rs, as well as the stability and half- life of the antibody \(^{21}\) . Antibody- based therapeutics, one of the most prominent and expanding class of drugs used for the treatment of diverse human disorders, including cancer, autoimmunity, and infectious diseases \(^{22}\) , are produced with heterogenous glycoforms that impact their safety and efficacy \(^{21}\) . EndoS and EndoS2 are particularly useful for Fc glycan remodeling due to the strict specificity of both enzymes for intact IgG antibodies and the discovery + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 911, 195]]<|/det|> +of their respective glycosynthase mutants, that are capable of transferring a homogenous \(N\) - glycan to IgG antibodies using a chemoenzymatic approach (Fig. 1B; \(^{23,24}\) ). The generation of antibodies with defined glycoforms is critical not only to understand the role of each glycoform on antibody function but also for the production of antibodies with defined therapeutic, diagnostic and pharmacokinetic properties \(^{25}\) . + +<|ref|>text<|/ref|><|det|>[[86, 205, 912, 864]]<|/det|> +EndoS and EndoS2 share \(37\%\) sequence identity and, although they are both IgG- specific, exhibit distinct \(N\) - glycan substrate specificities. EndoS processes exclusively complex type (CT) \(N\) - glycans on IgG antibodies, while EndoS2 is active on CT, high mannose (HM) type and hybrid type (Hy) glycans on IgG. EndoS is a 108 kDa multi- modular protein composed of six domains which adopt a "V"- shaped structure, from the N- to the C- terminus: (i) an N- terminal three- helix bundle domain (N- 3HB; residues 37- 97); (ii) a glycosidase hydrolase domain (GH; residues 98- 445); (iii) a leucine- rich repeat domain (LRR; residues 446- 631); (iv) a hybrid Ig domain (hIg; residues 632- 764); (v) a \(\beta\) - sandwich domain (residues 765- 923), and (vi) a C- terminal three- helix bundle domain (C- 3HB; residues 924- 995) \(^{26,27}\) . EndoS2 is a 95 kDa multi- modular protein that shares a similar overall architecture to EndoS but lacks the two 3HB domains on its N- and C- termini \(^{28}\) . We previously described the structural determinants that govern the \(N\) - glycan specificities of EndoS \(^{26,27}\) and EndoS2 \(^{28}\) . However, the mechanism by which these enzymes specifically process the \(N\) - glycan on the Fc region of IgG antibodies but no other glycoproteins, has not yet been described. Here, we present the cryoEM structure of EndoS in complex with the IgG1 Fc region. In combination with small- angle X- ray scattering (SAXS), alanine scanning mutagenesis, hydrolytic activity measurements, nuclear magnetic resonance (NMR) and molecular dynamics (MD) analysis, we describe the molecular mechanisms of recognition and specificity by EndoS and EndoS2 for IgG antibodies. Our findings will enable novel strategies for the chemoenzymatic synthesis of monoclonal antibodies with improved therapeutic properties, as well as the engineering of new enzymes to treat diseases of the adaptive immune system. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 154, 88]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[90, 105, 520, 125]]<|/det|> +## Architecture of the EndoS-IgG1 Fc region complex + +<|ref|>text<|/ref|><|det|>[[86, 135, 912, 907]]<|/det|> +To elucidate the molecular mechanism of antibody recognition by EndoS we determined the single- particle cryoEM structure of the catalytically- inactive mutant EndoS \(\mathrm{E_{233A}}\) in complex with the Fc region of an IgG1 monoclonal antibody (EndoS \(\mathrm{E_{235A}}\) - Fc) at 4.5 Å resolution (Fig. 1C,D; Supplementary Table 1; Supplementary Figs. 1, 2, 3 and 4; Methods). Although the purification of the native enzyme- substrate complex by size exclusion chromatography (SEC) was heterogeneous, we were able to purify the cross- linked EndoS \(\mathrm{E_{235A}}\) - Fc using the Grafix method (20; Supplementary Fig. 1). The SDS- PAGE of the fractions obtained by stabilization of the EndoS \(\mathrm{E_{235A}}\) - Fc complex with mild conditions of glutaraldehyde mainly shows the formation of a band corresponding to the MW of the EndoS \(\mathrm{E_{235A}}\) - Fc 1:1 complex but also other bands at higher MW that could correspond with the EndoS \(\mathrm{E_{235A}}\) - Fc 2:1 (250 kDa) (Supplementary Fig. 1A). We isolated the fractions of the EndoS \(\mathrm{E_{235A}}\) - Fc 1:1 complex and further purified the sample by SEC (Supplementary Fig. 2; Methods). In the EndoS \(\mathrm{E_{235A}}\) - Fc complex structure, the Fc region is located at the cleft of the "V"- shaped structure of EndoS, between the GH and the \(\beta\) - sandwich domains. The Fc region is a homodimer of the C-terminal domains of the antibody heavy chain. Each protomer compromises \(\mathrm{C\gamma 2}\) and \(\mathrm{C\gamma 3}\) domains, which interact with the equivalent domains of the other protomer via disulfide bonds in the hinge region and non- covalent interactions between the \(\mathrm{C\gamma 3}\) domains. The \(N\) - glycan is linked to each Asn297 of both \(\mathrm{C\gamma 2}\) domains. One of the Fc protomers (FcP1) interacts with the GH and the \(\beta\) - sandwich domains of EndoS through an extensive contact area of ca. 874 Å 30, while weak electron density of the cryoEM map can be ascribed to the second Fc protomer (FcP2) which does not make substantial interactions with EndoS and the N- 3HB domain. However, multibody refinement and motion analysis showed that FcP2 transiently interacts with residues 403- 413 of the GH domain of EndoS and the top of the N- 3HB domain (Supplementary Video 1, 2 and 3). Structural comparison of the EndoS- Fc complex and unliganded EndoS showed no marked conformational changes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 912, 411]]<|/det|> +(PDB code 4NUZ; r.m.s.d. of 2.5 Å from 880 residues; Supplementary Fig. 5). The dimensions of EndoS in the EndoSE235A- Fc 3D reconstruction complex (123.1 Å × 75.1 Å × 60.5 Å) are similar to those found in the X- ray crystal structure of EndoS (129.2 Å × 80.3 Å × 60.8 Å), indicating that the V shape of EndoS is not appreciably distorted in the active complexes. This is also supported by our SAXS data (Supplementary Table 2; Supplementary Fig. 6; Supplementary Note 1). EndoS in complex with the Fc region exhibits a narrower cleft between the GH and \(\beta\) - sandwich domains than does unliganded EndoS (Supplementary Fig. 5). In addition, the overall structure of FcP1 did not change after binding to the EndoS (PDB code 1H3X; r.m.s.d. of 1.6 Å from 345 residues). The main structural changes occur on the C' \(\beta\) - strand, the C'E loop (residues 295- 300) that bears the \(N\) - glycan, and the BC loop (residues 264- 272) within the C\(\gamma\)2 domain of FcP1 (Fig. 1C,D and Supplementary Fig. 6). + +<|ref|>sub_title<|/ref|><|det|>[[92, 460, 830, 482]]<|/det|> +## The EndoS GH domain interacts with the \(N\) -glycan and the C\(\gamma\)2 domain of the Fc region + +<|ref|>text<|/ref|><|det|>[[86, 494, 912, 909]]<|/det|> +We have previously shown that the CT \(N\) - glycan product is accommodated in the binding site of the GH domain of EndoS that is formed by several loops whose conformations governs the strict substrate specificity of the enzyme (27; PDB code 6EN3). The EndoS GH domain adopts a conserved \((\beta /\alpha)_8\) - barrel with a defined \(N\) - glycan binding site delineated by the N- 3HB domain and the connecting loops \(\beta 1 - \beta 2\) (loop 1; residues 120–145), \(\beta 2 - \alpha 4\) (loop 2; residues 151–158), \(\beta 3 - \alpha 5\) (loop 3; residues 185–206), \(\beta 4 - \alpha 6\) (loop 4; residues 235–247), \(\beta 5 - \alpha 7\) (loop 5; residues 281–289), \(\beta 6 - \alpha 8\) (loop 6; residues 304–306), \(\beta 9 - \alpha 10\) (loop 7; residues 347–380), \(\beta 10 - \alpha 11\) (loop 8; residues 403–413), and \(\alpha 11 - \alpha 12\) (loop 9; residues 420–434). The cryoEM EndoSE235A- Fc structure shows that loop 6' (residues 312- 324) of the GH domain interacts with the C' \(\beta\) - strand and the BC loop of the C\(\gamma\)2 of FcP1 (Fig. 2). This loop adopts a \(\beta\) - hairpin architecture, which is unique amongst GH18 enzyme structures, and creates an opening towards the active site of the EndoS GH domain (26). In contrast, EndoS2 exhibits a shorter loop at this position comprising different amino acid residues (28 (Supplementary Fig. 7). Specifically, K323 of EndoS + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 912, 548]]<|/det|> +interacts with E294 of FcP1 by hydrogen bonds, while E315 of EndoS makes a salt bridge with R292 of FcP1. Moreover, W314 of EndoS interacts with the side chain of P271 and E272 of FcP1 by hydrophobic interactions and a hydrogen bond, respectively. The side chain of H236 from loop 4 of the EndoS GH domain interacts with Y296 of C'E loop by hydrogen bonds, while E354 makes hydrogen bonds with S298 of the same loop. These interactions between FcP1 and the GH domain of EndoS cause a distortion of the C' \(\beta\) - strand and conformational changes in the C'E loop. This leads to the conformational change of the C'E loop carrying the \(N\) - glycan and the side chain rotamer of N297 that promotes the displacement of the \(N\) - glycan from the center of the Fc region towards the active site of the enzyme (Fig. 2). Careful inspection of the electron density map in the active site allowed us to model the two first GlcNAc residues and the central mannose core saccharide (Supplementary Fig. 8). The O3 and the oxygen of the acetamide group of GlcNAc (+1) makes hydrogen bonds with the side chain of N356 of EndoS while O6 interacts with the side chain of Y305. The OD2 atom of D233 that participates in the orientation of the GlcNAc (- 1) is 5.4 Å from the nitrogen of the acetamide this sugar. The O3 of Man (- 2) interacts with the side chain of D152 by hydrogen bonds. + +<|ref|>sub_title<|/ref|><|det|>[[88, 595, 907, 652]]<|/det|> +## The EndoS \(\beta\) -sandwich domain confers IgG specificity through a protein-protein interaction with the Fc region + +<|ref|>text<|/ref|><|det|>[[86, 666, 912, 907]]<|/det|> +The \(\beta\) - sandwich domain of EndoS is composed of eight \(\beta\) - strands connected by short loops and arranged in two opposing antiparallel \(\beta\) - sheets, including \(\beta 1 - \beta 2 - \beta 7 - \beta 4 - \beta 5\) , and \(\beta 6 - \beta 3 - \beta 8\) (Fig. 2). Both EndoS and EndoS2 \(\beta\) - sandwich domains were previously classified and suggested to function as carbohydrate- binding modules (CBM) from family 32 based on structural homology \(^{26 - 28,31}\) . However, the cryoEM EndoS \(_{\mathrm{E235A}}\) - Fc structure clearly reveals that the loops of the \(\beta\) - sandwich domain bind to FcP1 through protein- protein interactions. Moreover, N297 that bears the \(N\) - glycan points towards the GH domain and in the opposite direction of the \(\beta\) - sandwich domain. The \(\beta\) - sandwich domain interacts exclusively with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 68, 912, 487]]<|/det|> +the \(\mathrm{Cy2 - Cy3}\) joint of FcP1. Specifically, the EndoS residue W803, previously identified as a key residue for EndoS hydrolytic activity and binding to IgG1 Fc \(^{26}\) , makes a \(\pi - \pi\) interaction with H435 of \(\mathrm{Cy3}\) domain and hydrophobic interactions with I253 of \(\mathrm{Cy2}\) domain, while Y909 of EndoS interacts with the side chains of Q311, L314 and N315 of the \(\mathrm{Cy3}\) domain. In addition, W803 together with S800, Y909, S910 and S911 were proposed to form a putative \(N\) - glycan binding pocket \(^{28}\) but the EndoS \(_{E235A}\) - Fc cryoEM structure shows that the main function of W803 and Y909 is to mediate protein- protein interactions between the enzyme and the Fc region of the substrate. Structural comparison of the EndoS \(_{E235A}\) - Fc complex with that of EndoS2 supports a common binding mode to Fc for both enzymes. EndoS2 also bears aromatic residues, W712 and Y820, at equivalent positions to W803 and Y909 in EndoS, respectively, which are indispensable for substrate binding and catalysis \(^{28}\) ; Supplementary Fig. 7). Thus, the \(\beta\) - sandwich domains of EndoS and EndoS2, previously proposed to be CBMs, act instead to mediate a protein- protein interface with its glycoprotein in which no glycan participates. + +<|ref|>sub_title<|/ref|><|det|>[[90, 533, 773, 555]]<|/det|> +## EndoS-Fc protein-protein interactions are critical for glycoside hydrolase activity + +<|ref|>text<|/ref|><|det|>[[86, 567, 912, 912]]<|/det|> +To further investigate the molecular determinants of EndoS- Fc complex formation, we performed single and multiple alanine mutations in both EndoS and Fc, as well as loop deletions in EndoS to determine the key residues affecting hydrolytic activity in the EndoS- Fc interface. The residues in EndoS and Fc were selected based on the cryo- EM structure. We determined the rates by which the EndoS mutants hydrolyzed glycans from Fc and by which EndoS hydrolyzed glycans from Fc mutants by tracking the proportions of glycosylated versus deglycosylated Fc over time by intact mass spectrometry (Fig. 3; Supplementary Figs. 9 and 10). Specifically, in the Fc region we mutated residues of the C'E loop of \(\mathrm{Cy2}\) domain (residues R292, E293, E294, Q295, Y296 and S298), the \(\mathrm{Cy2 - Cy3}\) joint region (residues I253, H310, Q311, L314, N315, E430 and H435), the BC loop of \(\mathrm{Cy2}\) domain (H268, E269, D270, E272), and the FG loop of the \(\mathrm{Cy2}\) domain (residues 325- 331) that faces the N- 3HB domain of EndoS to alanine. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 912, 626]]<|/det|> +For positions in this loop that are encoded by alanine residues in the Fc region, we made mutations to glycine and tryptophan in order to determine the effect of removing or extending the side chains at those positions, respectively. In EndoS, we mutated residues in the N- 3HB loop (residues K70, Q73, E74, Q76 and K77), the GH domain (residues H236, W314, E315, N319, K323, E354, K414, Y416, K418, Q319, K420, E421, F422, K423 and loops 6' and 9), the LRR loop (residues Y541, K543, D544, N545 and K546) and the \(\beta\) - sandwich domain (residues W803, Y909 and loop 828- 836) to alanine and assessed their hydrolytic activities. As shown in Fig. 3, mutations in the C'E loop exhibited pleiotropic effects on hydrolytic activity; mutations in residues E293A and Q295A that did not interact with the enzyme, resulted in faster deglycosylation by EndoS, while the remaining mutations reduced the rate of glycan hydrolysis. Similar results were obtained for mutations in the N- 3HB facing loop on the Fc, with some variants (e.g., A330W and NKALPAP \(\rightarrow\) GGGsGGG) resulting in faster glycan removal while the remainder reduced the rate of \(N\) - glycan hydrolysis. Mutations in the \(\mathrm{C\gamma 2 - C\gamma 3}\) joint residues that interact with the \(\beta\) - sandwich domain of EndoS significantly reduced the rate of glycan hydrolysis. In EndoS, mutations in the N- 3HB, GH, LRR and \(\beta\) - sandwich domains all significantly reduced the rate of glycan hydrolysis from Fc, with the removal of loop 9 in the GH domain and the W803A mutation in the \(\beta\) - sandwich domain completely abolished EndoS activity. + +<|ref|>text<|/ref|><|det|>[[87, 636, 912, 909]]<|/det|> +The results of our mutational analysis indicate that the most important region driving EndoS- Fc complex formation and \(N\) - glycan hydrolysis involves residues on the \(\mathrm{C\gamma 2 - C\gamma 3}\) joint region of the Fc which interact with the \(\beta\) - sandwich domain of EndoS. The most extreme examples are (i) the EndoS W803A mutation and (ii) the Fc I253A/H310A double mutation, which completely abolish hydrolytic activity (Fig. 3). The variable activities observed in the Asn297- loop and N- 3HB facing loops in Fc and in the N- 3HB in EndoS suggest that conformations adopted by these loops play an important role in recognition of Fc by EndoS. Mutations in the EndoS GH domain significantly reduced enzymatic activity suggesting that these mutations adversely affect the ability of the \(N\) - glycan to enter the active site. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 911, 194]]<|/det|> +reduction of the hydrolytic activity observed for mutations in the LRR loop of EndoS suggests that these mutations may adversely affect the conformation of this loop which stabilizes the conformation of the GH domain loop 6' to effectively recognize the glycan loop of the antibody as shown by the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure \(^{27}\) . + +<|ref|>text<|/ref|><|det|>[[86, 205, 912, 830]]<|/det|> +We further evaluated the conformational behavior of the IgG1 Fc region by molecular dynamics (MD) simulations. We modelled the structure of wild type and mutant EndoS- Fc complex structures, including mutations in the Fc region (E293A, Y296A, E430A- H435A, and H310A- L314A) or in EndoS (W314A, W803A, and Y909A) corresponding to select mutations in the hydrolytic activity experiments described above. We observed that the vast majority of the interactions between the Fc region and EndoS in the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure in these MD simulations, except for the interaction between H236 of EndoS and Y296 of FcP1 and the salt- bridge formed between K323 and E294, which were rarely or negligibly populated. As we observed in the EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure, W803 of EndoS was engaged in a \(\pi - \pi\) interaction with H435 of FcP1. Similarly, W314 and Y909 of EndoS were involved in hydrophobic contacts with P271 and L314 of the Fc. The side chain of E354 formed a hydrogen bond with the hydroxyl group of S298 of FcP1 and the salt- bridge formed between E315 of EndoS and R292 of FcP1 was displayed in approximately 80 percent of the total trajectory time (Fig. 4A). We observed a similar scenario for the complex with the FcP1 E293A mutation (Fig. 4B). Notably, and in contrast to the wild- type antibody, the salt- bridge between K323 of EndoS and E294 of FcP1 was highly occupied in solution. These computational data could explain, at least in part, the higher activity of this mutant related to wild type. Of note, a loss of one or more of these stabilizing interactions is observed in the remaining complexes, resulting in a different 3D structure of the complex compared with the wild- type complex (Fig. 4C; Supplementary Figs. 11, 12, 13 and 14). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 826, 91]]<|/det|> +## EndoS and EndoS2 deglycosylate the two glycans of the IgG Fc homodimer sequentially + +<|ref|>text<|/ref|><|det|>[[86, 98, 912, 835]]<|/det|> +Our EndoS \(_{\mathrm{E235A}}\) - Fc cryoEM structure revealed a 1:1 EndoS:Fc complex in which EndoS engages a single Fc protomer through a protein- protein interaction formed by the EndoS \(\beta\) - sandwich domain and the Fc \(\mathrm{C\gamma 2 - C\gamma 3}\) joint, while the EndoS GH domain engages the N297- linked glycan on the same Fc protomer. However, EndoS hydrolyzes the glycans from both Fc protomers and no evidence of partially deglycosylated IgG has been reported to date. In order to define the mechanism of full deglycosylation of IgG by EndoS, we analyzed the deglycosylation reaction over time. We mixed wild type EndoS with Fc in a 1:5000 enzyme:substrate ratio and analyzed changes in intact Fc masses by LC- MS approximately every 5 minutes (Fig. 5A). As expected, the di- glycosylated Fc species decreased while the fully deglycosylated species increased over time. As the reaction proceeded, however, we observed that a mono- glycosylated species initially increased to approximately 50 percent of all Fc masses, prior to decreasing along with di- glycosylated Fc to near zero. From these data, we calculated a rate of hydrolysis for the transition from di- glycosylated Fc to mono- glycosylated Fc of \(3.8~\mu \mathrm{M}^{- 1}\mathrm{s}^{- 1}\) and for the transition from mono- glycosylated Fc to fully deglycosylated Fc of \(0.8~\mu \mathrm{M}^{- 1}\mathrm{s}^{- 1}\) . Thus, the hydrolysis of the first glycan was approximately 5- fold faster than the hydrolysis of the second glycan. We conducted an analogous experiment with EndoS and Rituximab, a full- length IgG1 antibody and observed identical kinetic parameters for hydrolysis of the first and second glycans (Fig 5B). We also conducted an analogous experiment with EndoS2 and Rituximab and found similar kinetic behavior (Fig 5C); although each of the reaction rates was slower, hydrolysis of the first glycan was also 5- fold faster than that of the second glycan, as for EndoS. These data indicate that EndoS- and EndoS2- mediated hydrolysis of the two glycans on Fc or IgG occurs sequentially, concomitant with dissociation of the enzyme- substrate complex after hydrolysis of the first glycan before re- associating to hydrolyze the second glycan. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 70, 907, 125]]<|/det|> +## The sterically constrained glycan binding site of EndoS2 prevents deglycosylation of potential non-IgG glycoprotein substrates + +<|ref|>text<|/ref|><|det|>[[86, 137, 912, 509]]<|/det|> +To shed more light on the interaction of EndoS and EndoS2 with their natural ligands, we employed NMR chemical shift perturbations (CSP) \(^{32}\) . Generally speaking, the method consisted of following changes in chemical shifts from signals in NMR spectra during ligand titrations, and using them to determine the location of binding sites, affinity of ligands and/or conformational equilibria. We used CSPs for the quantification of the overall thermodynamic and kinetic binding parameters of EndoS2 to IgG1 Fc and its reaction products. We extracted CSPs from methyl- TROSY NMR spectra \(^{33}\) and fitted quantum mechanically simulated spectra using the TITAN (TITration ANalysis) software package \(^{34}\) . This approach offers unique access to binding thermodynamics and kinetics parameters \(^{33,35}\) , providing a link to the conformational changes observed by crystallography, H/D exchange MS \(^{28}\) , cryo- EM and MD, assisting our mechanistic understanding of EndoS2 enzymatic activity (Supplementary Note 2: Supplementary Figs. 15 to 33; Methods). + +<|ref|>text<|/ref|><|det|>[[86, 522, 912, 893]]<|/det|> +While were unable to express EndoS in minimal media in quantities required for NMR experiments, we were able to express and purify isotopically \([U^{- 15}N,^{2}H]\) MIL \(^{pros}V^{pros}[^{13}C,^{1}H_{3}]\) - methyl labeled \(^{35 - 37}\) EndoS2 and catalytically- inactive EndoS2 \(_{E186L}\) in order to perform these experiments. This labeling scheme produced methyl- TROSY spectra with excellent signal- to- noise (Supplementary Fig. 15). In addition, selected methyl probes were well distributed around the binding pockets, allowing for the extraction of CSPs upon ligand titration (Fig. 6A,B). We measured the binding kinetics of EndoS2 \(_{E186L}\) to its CT glycan substrate, the IgG1 Fc region, the reaction products Fc aglycan and CT \(_{n - 1}\) , and Ca \(^{2 + }\) ions (Fig. 6C). We also studied the interaction of catalytically- inactive EndoBT- 3987 \(_{D312A/E314}\) , a GH18 family ENGase that hydrolyzes HM glycans from myriad glycoproteins but is not specific for IgG antibodies or any other protein (Supplementary Fig. 16) \(^{38,39}\) . Quantitative lineshape analysis of methyl- TROSY spectra showed that EndoS2 \(_{D186L}\) binds Rituximab- Fc with a dissociation constant of 3.1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 912, 870]]<|/det|> +\(\mu \mathrm{M}\) (Table 1, entry 1), similar to the previously reported \(K_{D}\) of \(\sim 9 \mu \mathrm{M}\) for the binding of EndoS2D186L to IgG1- CT obtained by surface plasmon resonance \(^{28}\) . A representative example of lineshape fitting can be seen in Fig. 6D. We measured the binding affinity of EndoBT- 3987D312A/E314 to its HM \(N\) - glycan substrate to be \(K_{D} = 1.7 \mu \mathrm{M}\) (Table 1, entry 2). We measured the \(K_{D}\) s of EndoS2D186L for soluble CT glycan and the reaction product \(\mathrm{CT_{n - 1}}\) to be \(106.9 \mu \mathrm{M}\) and \(150.7 \mu \mathrm{M}\) , respectively, which corresponds to affinities two orders of magnitude weaker than the EndoS2E186L- Fc and EndoBT- 3987D312A/E314 affinities. We measured on- rate constants for Fc, CT and \(\mathrm{CT_{n - 1}}\) to be \(1.3 \times 10^{6} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) , \(1.1 \times 10^{5} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) and \(6.2 \times 10^{4} \mathrm{M}^{- 1} \mathrm{s}^{- 1}\) , respectively, all much slower than expected for a diffusion- controlled process, indicating that these interactions involve solvent reorientation and possibly conformational changes. Dissociation rate constants \(k_{off}\) were similar for these three ligands, and go down into the range of a few Hz (Table 1, entries 3,4). The faster on- rate constant observed for the interaction of EndoS2D186L with Fc, similar to the on- rate constant of EndoBTD312A/E314 for just the \(N\) - glycan, suggests that a protein- protein interaction favors a “correct” orientation of the EndoS2 GH domain relative to the glycan during the initial EndoS2- Fc fragment encounter event. The energetic contribution of such interactions can be approximated as the change in the difference of the standard free energy \(\Delta \Delta G_{Fc - CT}^{\circ} = \Delta G_{Fc}^{\circ} - \Delta G_{CT}^{\circ} = -8.7 \mathrm{KJ}\) , considering \(\Delta G^{\circ} = RT \ln (K_{D})\) . We found that EndoS2- mediated hydrolysis of Fc \(N\) - glycans abrogated recognition of Fc by EndoS2D186L at otherwise saturating Fc concentrations, highlighting the central role of the carbohydrate for the recognition of IgG by EndoS2 and preventing re- binding of reaction products (Table 1, entry 5). We also extracted the catalytic rate, \(k_{cat}\) , and the Michaelis- Menten constant, \(K_{\mathrm{M}}\) , from the catalytic cleavage of CT \(N\) - glycans from the Fc region by EndoS2 by measuring the concentration of enzymatically released glycans by \(^1\mathrm{H}\) NMR spectra (Supplementary Note 2; Methods). Under the conditions used in this study, we observed a catalytic rate \(k_{cat}\) of \(5.9 \mathrm{s}^{- 1}\) and a Michaelis- Menten constant \(K_{\mathrm{M}}\) of \(26.7 \mu \mathrm{M}\) Fig. 6E). Combined with lineshape analysis, these results provide a first comprehensive + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 908, 125]]<|/det|> +picture of the catalytic mechanics of the enzymatic processing of IgG1 Fc \(N\) - glycans by EndoS2 (Fig. 6F). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 181, 88]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[87, 103, 912, 475]]<|/det|> +EndoS and EndoS2 are the most prominent members of a rare subset of carbohydrate- active enzymes that are specific not only to the glycan type but also to the protein component of the glycoprotein from which they hydrolyze glycans. To date, no molecular mechanism has been described to explain how these enzymes are protein- specific. Our data now provide the mechanism by which EndoS and EndoS2 hydrolyze glycans from IgG antibodies in a highly specific manner. This mechanism for protein- specific deglycosylation by EndoS and EndoS2 relies on four principles. These enzymes: (i) avoid all other potential non- IgG glycoprotein substrates; (ii) form a protein- protein interaction with a non- enzymatic domain in order to create an anchor point on the IgG substrate; (iii) leverage their unique geometry to reach from the anchor point to orient the active site of the GH domain directly adjacent to the Asn297- linked glycan that they hydrolyze; and (iv) hydrolyze the two glycans on the IgG Fc region homodimer sequentially by a catch- and- release process. + +<|ref|>text<|/ref|><|det|>[[86, 487, 912, 928]]<|/det|> +The first mechanistic principle, reduced distraction by potential non- IgG substrates, is accomplished by EndoS and EndoS2 constructing conformationally constrained glycan binding sites. This distinctive morphological property of their active sites results in the bound glycan needing to adopt a conformation in which the branches of a biantenary glycan are nearly parallel to one another \(^{26 - 28}\) . Although such a glycan conformation is commonly rendered schematically, it is exceedingly rare in nature due to the high degree of conformational flexibility inherent to \(N\) - linked glycans. Even when bound in the glycan binding sites of endoglycosidases that are not protein- specific, \(N\) - linked glycans adopt diverse structures that are decidedly unlike the parallel- branched glycans in EndoS and EndoS2. This is the case for the \(N\) - linked glycans that are the substrates for EndoBT- 3987 \(^{38,39}\) (Supplementary Fig. 34), EndoF \(^{3,40}\) , as well as many others. The adoption of the constrained glycan conformation in EndoS and EndoS2, as well as the flexible glycan conformation in EndoBT- 3987, can be observed in the relative on- rates of substrates binding to catalytically- inactive variants of these endoglycosidases. Indeed, our NMR data indicate that the on- rates for glycans alone are two orders of magnitude slower for EndoS + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 912, 229]]<|/det|> +and EndoS2 than they are for EndoBT- 3987. However, these enzymes have all evolved to have comparable affinities and binding kinetics to their biologically relevant carbohydrates – linked to IgG for EndoS and EndoS2, while free in solution or exposed on the surface of any protein for EndoBT- 3987 – a property clearly related to the distinct structural features of the active sites of these endoglycosidases and the relative entropic costs of their glycan substrates adopting their bound conformations. + +<|ref|>text<|/ref|><|det|>[[87, 243, 912, 690]]<|/det|> +The second mechanistic principle, the establishment of an enzyme- substrate protein- protein interaction, is fulfilled by the \(\beta\) - sandwich domain of EndoS and EndoS2. While these domains have high structural similarity to carbohydrate binding modules (CBMs), they are not, in fact, bona fide CBMs in that they do not bind carbohydrates. Instead, these \(\beta\) - sandwich domains recognize a strictly protein- based molecular surface on the IgG Fc region at the joint between its \(\mathrm{Cy2}\) and \(\mathrm{Cy3}\) domains. This Fc surface is a common binding spot for myriad proteins, including Protein A \(^{41}\) and Protein G \(^{42}\) from bacteria, as well as the human neonatal Fc receptor (FcRn; \(^{43}\) ). The structural similarity of this molecular surface on the Fc region amongst all IgG subtypes of both human and mouse antibodies explains why these glycoproteins comprise all of the known substrates for EndoS and EndoS2. Furthermore, this enzyme- substrate protein- protein interaction is absolutely required for enzymatic activity, as mutations on either side of this interface can completely abrogate hydrolysis of Asn297- linked Fc glycans by EndoS \(^{26}\) and EndoS2. Accordingly, for EndoS and EndoS2, a protein- protein interaction drives hydrolytic activity of these endoglycosidases, a property not previously observed in carbohydrate- active enzymes. + +<|ref|>text<|/ref|><|det|>[[87, 703, 912, 902]]<|/det|> +The third mechanistic principle, orientation of the enzyme active site adjacent to the substrate glycan, is a result of the unique V- shaped molecular architecture of EndoS and EndoS2 in which the \(\beta\) - sandwich and GH domains are situated on either tip of the V, interspersed by an Ig- like domain and a leucine- rich repeat. As a consequence of anchoring the enzyme on the substrate via the \(\beta\) - sandwich domain/Fc region protein- protein interface described above and the architectures of these multi- domain enzymes, the active sites of EndoS and EndoS2 become ideally positioned for engagement with and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 68, 912, 580]]<|/det|> +hydrolysis of the Fc glycan. The predominant conformation of the IgG Fc region is one in which the Asn297- linked glycans are pointed inward towards the center of the homodimer, although these antibodies are known to exhibit many conformations, including those in which these glycans are more exposed to solvent \(^{44,45}\) . Our cryoEM structure shows that the Fc loop on which the Asn297- linked glycan adopts just such a conformation; it is not exposed to solvent in the complex, though, but rather bound in the EndoS active site. Thus, when anchored by the enzyme- substrate protein- protein interface, the distinctive molecular architecture positions the GH domain to stabilize an otherwise infrequently populated conformation amongst the entire ensemble of Fc conformations so as to increase the sampling of its glycan, resulting in efficient glycan hydrolysis. It must be noted that the appendage of one or more non- enzymatic domains to GH domains is not uncommon in carbohydrate- active enzymes and is thought to guide the enzymatic domains towards their targets, as in the case of CBMs, and/or to provide sufficient spacing from the bacterial outer membrane for hydrolyzing glycans in the extracellular environment. However, the structural arrangement of the enzymatic and non- enzymatic domains of EndoS and EndoS2 is unique amongst known carbohydrate- active enzymes, as is the requirement for two of these connected domains to be absolutely required for activity. + +<|ref|>text<|/ref|><|det|>[[87, 591, 912, 893]]<|/det|> +The fourth mechanistic principle, sequential deglycosylation of the two Asn297- linked glycans on the IgG homodimer, is due to the biologically relevant state being a 1:1 complex of EndoS: Fc, as we observed in our cryoEM structure. When anchored by their protein- protein interfaces, the GH domains of EndoS and EndoS2 can reach only to the Asn297- linked glycan on the same Fc protomer that their \(\beta\) - sandwich domains engage. In order to hydrolyze the glycan on the other Fc protomer, the enzyme must unbind the substrate entirely, and rebind it through a protein- protein interaction on the opposite Fc protomer in order to properly reposition its GH domain. Indeed, by intact mass spectrometry, we observed the accumulation of a mono- glycosylated antibody species, followed by its diminution. If the biologically relevant state were instead a 2:1 complex of EndoS: Fc, no such mono- glycosylated antibody + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 912, 230]]<|/det|> +species would exist, even transiently, as two EndoS enzymes would attack the antibody from each side simultaneously. Also, in our NMR experiments we observed that re- binding of the reaction product, a deglycosylated antibody, was highly disfavored. Despite this, a 2:1 complex formed between the enzyme and a di- glycosylated substrate can exist at high concentrations, as we observed as minor populations by both SAXS and cryoEM analysis (Supplementary Fig. 35). + +<|ref|>text<|/ref|><|det|>[[87, 242, 912, 438]]<|/det|> +With the molecular mechanism of IgG- specific deglycosylation established, opportunities for engineering EndoS- based enzymes to treat diseases of the adaptive immune system abound. It is now possible to rationally conceptualize engineered EndoS enzymes that selectively deglycosylate certain subsets of IgG antibodies, such as auto- antibodies, in order to defeat their effector functions and treat autoimmunity without wholesale IgG disablement that would render the host globally immunosuppressed and vulnerable to opportunistic infections. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 166, 88]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[90, 104, 875, 125]]<|/det|> +## Expression and purification of EndoS and EndoS2 wild type and EndoS and EndoS2 mutants + +<|ref|>text<|/ref|><|det|>[[87, 134, 912, 725]]<|/det|> +EndoS and EndoS2 wild- type and EndoS and EndoS2 mutants were purified as previously described with the following modifications \(^{26 - 28}\) . Single- point mutations were developed by PCR- based site- directed mutagenesis, and full sequences were confirmed by GeneWiz (https://www.genewiz.com). EndoS wild type and mutants were produced in Escherichia coli BL21(DE3) cells (Novagen) grown in LB medium supplemented with \(100 \mu \mathrm{g} \mathrm{ml}^{- 1}\) of ampicillin. Cultures were grown at \(37^{\circ} \mathrm{C}\) to an \(\mathrm{OD}_{600}\) of 0.6–0.8, at which point the temperature was lowered to \(22^{\circ} \mathrm{C}\) for 1 h. Induction was triggered with \(1 \mathrm{mM}\) isopropyl \(\beta\) - D- 1- thio- galactopyranoside (IPTG) at \(22^{\circ} \mathrm{C}\) overnight. Cells were harvested by centrifugation and lysed by sonication using \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 7.5\) , \(500 \mathrm{mM}\) NaCl, \(10\%\) glycerol. CPD fusion proteins were purified by \(\mathrm{Ni}^{2 + }\) - immobilized metal- affinity chromatography followed by overnight treatment with \(1 \mathrm{mM}\) phytic acid at \(4^{\circ} \mathrm{C}\) and elution the next day with the lysis buffer. The proteins used for hydrolytic activity were further purified by size exclusion chromatography (Superdex 200 Increase 10/300 GL) in PBS as running buffer. \(\mathrm{EndoS}_{\mathrm{E235A}}\) were further purified by anion exchange chromatography using as binding buffer \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 8.5\) and elution buffer \(50 \mathrm{mM}\) Tris- HCl \(\mathrm{pH} 8.50\) , \(500 \mathrm{mM}\) NaCl. The protein was eluted using a gradient of \(5\%\) of elution buffer for \(2 \mathrm{min}\) . The fractions were pooled and aliquot to flash freeze at \(- 80^{\circ} \mathrm{C}\) . For EM sample preparation, \(\mathrm{EndoS}_{\mathrm{E235A}}\) were thaw and concentrated protein was purified by size exclusion chromatography (Superdex 200 Increase 10/300 GL) in PBS as running buffer. + +<|ref|>sub_title<|/ref|><|det|>[[90, 772, 336, 790]]<|/det|> +## Purification of Fc-Rituximab + +<|ref|>text<|/ref|><|det|>[[88, 805, 911, 895]]<|/det|> +Rituximab (RITUXAN, Genentech) was kindly provided courtesy of the University of Maryland Greenebaum Comprehensive Cancer Center. A solution of \(20 \mathrm{mg} \mathrm{mL}^{- 1}\) of Rituximab was dialyzed against \(20 \mathrm{mM}\) sodium phosphate, \(10 \mathrm{mM}\) EDTA; \(\mathrm{pH} 7.0\) . A mixture of \(0.5 \mathrm{mL}\) of the Rituximab solution + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 68, 912, 369]]<|/det|> +and \(0.5 \mathrm{mL}\) of the digestion buffer (20 mM sodium phosphate, 10mM EDTA, 20 mM cysteine- HCl pH 7.0), is added to the \(0.5 \mathrm{mL}\) of the \(50\%\) immobilized papain slurry (Thermo Scientific) previously equilibrated with the digestion buffer. The reaction was incubated overnight in a shaker water bath at 37 \(^\circ \mathrm{C}\) at high speed overnight. The reaction was stopped by adding \(1.5 \mathrm{mL}\) of \(10 \mathrm{mM}\) Tris- HCl pH 7.5 to the digest before centrifugation. After removing the supernatant, which contains the IgG fragments, the resin was washed with \(1.5 \mathrm{mL}\) of \(10 \mathrm{mM}\) Tris- HCl pH 7.5. The digest and wash fractions were combined and the Fc fragments were separated from the Fc fragments using an immobilized Protein A column (Thermo Scientific) and \(100 \mathrm{mM}\) glycine pH 3.0, as elution buffer. The eluted Fc was further purify by size exclusion chromatography using PBS to separate the Fc fragments from undigested IgG. + +<|ref|>sub_title<|/ref|><|det|>[[90, 416, 789, 438]]<|/det|> +## Preparation and purification of the cross-linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) -Fc-Rituximab complex + +<|ref|>text<|/ref|><|det|>[[86, 450, 912, 900]]<|/det|> +The cross- linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) - Fc- Rituximab complex was prepared using a standard GraFix method \(^{29,46}\) . Gradients for GraFix were prepared by mixing GraFix buffer 1 (PBS, pH 7.4, \(5\%\) v/v glycerol) and GraFix buffer 2 (PBS, pH 7.4, \(20\%\) v/v glycerol, \(0.25\%\) glutaraldehyde) using a gradient mixer (Gradient Master ip, BioComp Instruments), following the parameters determined by the manufacturer for the glycerol content. A mixture of \(100 \mu \mathrm{L}\) of EndoS \(\mathbf{E}_{235\mathrm{A}}\) ( \(12 \mu \mathrm{M}\) ) and \(100 \mu \mathrm{L}\) of Fc- Rituximab (60 \(\mu \mathrm{M}\) ) in PBS was incubated at room temperature for 10 min. The \(200 \mu \mathrm{L}\) solution mixture was loaded onto the top of the GraFix gradients and ultracentrifugation was carried out at \(40,000 \mathrm{rpm}\) for \(25 \mathrm{hs}\) at 4 \(^\circ \mathrm{C}\) (SW60 rotor, Beckmann). After ultracentrifugation, the gradients of each tube were fractionated using a Piston Gradient Fractionator \(^\mathrm{TM}\) Base Unit (Sciences service) and a FC 203 Fraction Collector (Gilson) to pump the gradient out from bottom to top at room temperature. Each fraction of \(250 \mu \mathrm{L}\) was immediately quenched by \(50 \mu \mathrm{L}\) of \(1 \mathrm{M}\) Tris- HCl pH 8.0, to stop further crosslinking. The presence of the cross- linked EndoS \(\mathbf{E}_{235\mathrm{A}}\) - Fc- Rituximab complex was confirmed by SDS- PAGE. Pooled fractions were flash frozen and storage at \(- 80^\circ \mathrm{C}\) . + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 405, 90]]<|/det|> +## Negative-stained electron microscopy + +<|ref|>text<|/ref|><|det|>[[87, 102, 912, 510]]<|/det|> +Negative- stained electron microscopyInitial inspection of the GraFix EndoSE235L- Fc- Rituximab complex by negative stain was performed. Specifically, an aliquot from a single GraFix tube were thawed and samples of \(50~\mu \mathrm{L}\) were injected into an analytical size exclusion chromatography column (Shodex KW403- 4F) using PBS pH 7.4, as running buffer. The single peak corresponding to the complex MW was collected in \(50~\mu \mathrm{L}\) fractions for subsequent standard negative stain grid preparation. Continuous glow discharged carbon electron microscopy grids CF300- Ni (EMS, USA). Briefly, drops of \(8~\mu \mathrm{L}\) of diluted complex sample were placed on the grid, incubated for 90 s and then blotted to remove the excess with filter paper (Whatman®). The grids were washed by placing them onto a drop of PBS, pH 7.4 for 30 s and bottled dry. The complex was stained by transferring the grid to a drop of \(1\%\) w/v uranyl formate for 45 s, then dried with a filter paper and stored at room temperature. Samples were imaged at \(\times 50,000\) magnification and using an inhouse JEOL 2200, \(200\mathrm{kV}\) FEG equipped with an UltraScan 4000 CCD camera (Gatan, USA). Images were used to analyze particles by single particle analysis using RELION. + +<|ref|>sub_title<|/ref|><|det|>[[88, 558, 483, 577]]<|/det|> +## Cryo-EM specimen preparation and screening + +<|ref|>text<|/ref|><|det|>[[87, 590, 912, 891]]<|/det|> +Cryo- EM specimen preparation and screeningCross- linked EndoSE235L- Fc- Rituximab complexes from four GraFix tubes were thawed, pooled, and concentrated using Pierce Protein Concentrators (PES 10K MWCO \(0.5\mathrm{mL}\) ) to obtain a final volume of \(60~\mu \mathrm{L}\) . The sample was injected into an analytical size exclusion chromatography column (Shodex KW403- 4F) using PBS pH 7.4, as running buffer. The complex peak was collected in collected in \(50~\mu \mathrm{L}\) fractions with to use directly for vitrification. Briefly, \(4\mu \mathrm{l}\) of the fractions were applied onto copper Quantifoil R 1.2/1.3 grids previously glow discharged. Grids were plunge- freeze into liquid ethane using a Vitrobot MkII (FEI) at \(10^{\circ}\mathrm{C}\) and \(80\mathrm{HR}\%\) humidity in the chamber for each concentration condition by triplicate. One of the triplicates for each condition was used for cryoEM screening performed in- house using a JEOL 2200, \(200\mathrm{kV}\) FEG equipped with an UltraScan 4000 CCD camera (Gatan, USA). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 300, 88]]<|/det|> +## Cryo-EM data collection + +<|ref|>text<|/ref|><|det|>[[87, 103, 912, 335]]<|/det|> +Cryo- EM data collection was carried out on a \(300\mathrm{kV}\) Titan Krios microscope equipped with a K2 Summit direct electron- counting camera and a GIF Quantum energy filter (Gatan) in ESRF CM01 beamline at Grenoble, France \(^{47}\) . Micrographs were recorded with EPU software (FEI) at a nominal magnification of \(\times 130,000\) , with a pixel size of \(1.052\mathrm{\AA}\) . A total of 5,546 movies of the cross- linked EndoSE235L- Fc- Rituximab grid sample were acquired for \(9\mathrm{s}\) in counting mode at a flux of 7.3 electrons per pixel \(\mathrm{s}^{- 1}\) , giving a total exposure of 59.37 electrons per \(\mathrm{\AA}^2\) and fractioned into 60 frames. A defocus range from \(- 1.3\mu \mathrm{m}\) to \(- 2.5\mu \mathrm{m}\) was used. + +<|ref|>sub_title<|/ref|><|det|>[[89, 383, 433, 403]]<|/det|> +## Cryo-EM single-particle data processing + +<|ref|>text<|/ref|><|det|>[[86, 415, 912, 930]]<|/det|> +EndoSE235- Fc complex data set were processed using the general protocol described in the following lines. Data processing was initiated using the RELION 3.1.2 \(^{48}\) . The 60 movie frames of each movie were aligned using MotionCor2 (5x5 patches and Bfactor 150). The CTF of the aligned image was assessed in the non- dose weighted image using CtfFind 4.1 with parameters amplitude contrast \(10\%\) and FFT box size of 512 pixels. Image CTF was visually inspected, and images with low quality discarded. RELION reference- free automatic particle picking was performed considering a particle diameter between \(95\mathrm{\AA}\) and \(110\mathrm{\AA}\) , imposing a minimum inter- particle distance of \(90\mathrm{\AA}\) and a low picking threshold (0.05), leading to an initial number of particles (797351). Particles were extracted using a square box of 180 pixels side. Particles were subjected to several rounds and extensive 2D classification selecting the best classes. An initial first 3D auto- refinement was performed using the map obtained from the negative stain images. This initial reconstruction was used to re- extract centered particles. Afterward, several steps of 3D classification were performed to discarding classes until models reached resolutions \(7\mathrm{\AA}\) and visual inspection revealed a well- defined structure. The selected particles (140,184) were used to reconstruct a model using RELION 3D auto- refinement protocol, which led to a map resolving to \(6\mathrm{\AA}\) . At this point, the model was used to perform one round of RELION CTF refinement and particle Polishing, and further + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 68, 912, 512]]<|/det|> +3D classification leading to a set of selected particles (721,336) for a final RELION refinement reaching \(4.8\mathrm{\AA}\) resolution. Polished and CTF refined particles from RELION were imported into cryoSPARC v3.2.0 \(^{49}\) and a final 2D classification of 200 classes was performed and the best- defined classes selected (Supplementary Fig. 3B) containing 339,309 particles. Final 3D reconstructions were performed in cryoSPARC using the homogeneous refinement and applying dynamic masking, using the RELION reconstructions low- pass at \(30\mathrm{\AA}\) as starting model. Finally, a final step of Non- Uniform refinement for the reconstruction \(^{50}\) , on the corresponding symmetry, was performed applying a calculated global mask from the reconstruction. The reconstruction obtained from the non- uniform cryoSPARC protocol is reported. Reported auto sharp- maps were obtained from the cryoSPARC non- uniform refinements (Supplementary Fig. 2C) and DeepEMhancer \(^{51}\) (Supplementary Fig. 2D). Local resolution was measured in cryoSPARC LocRes program implementation \(^{52}\) , using a threshold for local FSC \(= 0.5\) for local resolution assessment (Supplementary Fig. 2C,D). For quality graphs as FSC plots, angular distribution and precision see (Supplementary Fig. 2E,F). For other statistics please see Supplementary Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[90, 558, 354, 577]]<|/det|> +## Model building and refinement + +<|ref|>text<|/ref|><|det|>[[87, 590, 912, 928]]<|/det|> +A model of the EndoS X- ray crystal structure (PDB code 6EN3) and the Fc region of IgG1- Fc (H1X3), was initially rigid- body fitted inside the cryoEM reconstruction using UCSF- Chimera \(^{53}\) . The initial refinement of the models using Phenix real space refinement \(^{54,55}\) was performed as an individual chain rigid body group and later performing local minimization, annealing, and ADPs, using initial model as reference model to aid in keeping the geometry observed in the crystal structures. The model was improved iteratively manually in Coot \(^{56}\) using the DeepEMhancer sharp maps, following with rounds of Phenix real space refinement. Remodeling based on the interpretation of density due to the movement from loops in the interphase between EndoS and IgG1- Fc, as well as placing of carbohydrate in the density at the active site was performed. In later stages of the modelling, the side- chains whose positioning could not be defined unambiguously in the map density have been removed. The quality of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 909, 125]]<|/det|> +models was checked during refinements using Molprobity and validate- PDB server57. For model statistics and quality details please see Supplementary Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[89, 174, 658, 195]]<|/det|> +## EndoSE235-Fc map principal component analysis and motion movies + +<|ref|>text<|/ref|><|det|>[[87, 208, 912, 475]]<|/det|> +A two- body multibody refinement of cryoEM images was performed as a continuation of the final RELION 3D auto- refinement58. The density corresponding to catalytic inactive EndoS was defined as body 1 and the Fc region as body 2. Masks were created for the top and bottom halves of the consensus reconstruction using the volume eraser tool in UCSF- Chimera53 and RELION34 and used for a multibody refinement in RELION3. Principal component analysis was performed on the orientations of both bodies, and movies for the first three principal components (PCs) were written out as a series of volumes in MRC format59 describing the relative motion of the two bodies as described by these PCs (Supplementary Videos 1- 3). + +<|ref|>sub_title<|/ref|><|det|>[[89, 522, 295, 541]]<|/det|> +## SEC-SAXS experiments + +<|ref|>text<|/ref|><|det|>[[87, 556, 912, 896]]<|/det|> +A \(50~\mu \mathrm{M}\) solution of EndoSE235A or EndoSE186L were incubated with \(250~\mu \mathrm{M}\) solution of Fc- Rituximab in \(50~\mathrm{mM}\) Tris- HCl pH 7.5, \(100~\mathrm{mM}\) NaCl, \(2\%\) \(\nu /\nu\) glycerol for \(10\mathrm{min}\) at room temperature to form the EndoS(2)- Fc- Rituximab complexes. Small- Angle X- ray Scattering coupled with Size Exclusion Chromatography (SEC- SAXS) data for purified EndoSE235A, EndoSE186L, Fc- Rituximab and the complex mixtures of EndoSE235A- Fc- Rituximab and EndoSE186L- Fc- Rituximab in \(50~\mathrm{mM}\) Tris- HCl pH 7.5, \(100~\mathrm{mM}\) NaCl, \(2\%\) \(\nu /\nu\) glycerol were collected on the B21 beamline of the Diamond Light Source, UK. \(50~\mu \mathrm{L}\) of the protein samples or the complex mixtures were injected into a Shodex KW403- 4F column and eluted at a flow rate of \(150~\mu \mathrm{L}\mathrm{min}^{- 1}\) . Data were collected using a Pilatus2M detector (Dectris, CH) at a sample- detector distance of \(3,914\mathrm{mm}\) and a wavelength of \(\lambda = 1\mathrm{\AA}\) . The range of momentum transfer of \(0.1< s< 5\mathrm{nm}^{- 1}\) was covered \((s = 4\pi \sin \theta /\lambda\) , where \(\theta\) is the scattering angle). Data were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 66, 912, 266]]<|/det|> +processed and merged using standard procedures by the program package ScAter \(^{60}\) and PRIMUS \(^{61}\) . The maximum dimensions \((D_{max})\) , the interatomic distance distribution functions \((P(r))\) , and the radii of gyration \((Rg)\) were computed using GNOM \(^{62}\) . The molecular mass was determined using ScAter \(^{60}\) . The ab initio multiphase reconstruction of the SAXS data of the EndoS \(_{E235A}\) - Fc- Rituximab and the EndoS \(_{E186L}\) - Rituximab complexes were generated using the MONSA algorithm from ATSAS \(^{63}\) . The results and statistics are summarized in Supplementary Table 2. + +<|ref|>sub_title<|/ref|><|det|>[[90, 313, 799, 334]]<|/det|> +## Cloning, expression and purification of EndoS mutants for alanine scan experiments + +<|ref|>text<|/ref|><|det|>[[86, 345, 912, 824]]<|/det|> +Primers to introduce single or multiple mutations or deletions into the EndoS- CPD construct were designed using the NebBaseChanger (https://nebasechanger.neb.com/) and PCR was carried out using the NEB Q5 High Fidelity polymerase and manufacturer's instructions. For some of the alanine scan mutants, gene strings were ordered from Twist Bioscience and restriction cloned into the EndoS- CPD vector. All sequences were confirmed by Sanger sequencing at Genewiz (https://www.genewiz.com). For expression, the plasmids were transformed into Escherichia coli BL21 (DE3) cells and grown in 100 mL of LB medium supplemented by \(100 \mu \mathrm{g} \mathrm{mL}^{- 1}\) ampicillin T 37C. When the culture reached an OD of 0.6- 0.8, the culture was induced with 0.5mM IPTG and the culture was allowed to grow overnight at \(22^{\circ} \mathrm{C}\) . The cells were harvested by centrifugation at \(3,000 \times \mathrm{g}\) for 20 min and resuspended in 3mL of PBS pH 7.4. The cells were lysed with BugBuster according to manufacturer's instructions and the lysate clarified by spinning at \(18000 \times \mathrm{g}\) for 45 mins. The clarified supernatant was incubated with NiNTA resin for 3 hs before being washed with 10 column volumes (CV) of PBS pH 7.4. The EndoS was then eluted with \(100 \mu \mathrm{M}\) phytic acid in PBS pH 7.4. Protein purity was assessed by SDS- PAGE. All proteins were flash- frozen and stored at \(- 80^{\circ} \mathrm{C}\) until ready for use. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 758, 90]]<|/det|> +## Cloning expression and purification of Fc mutants for alanine scan experiments + +<|ref|>text<|/ref|><|det|>[[86, 99, 912, 616]]<|/det|> +The sequence for the heavy chain of Rituximab was ordered from Thermo Fisher Scientific in the pcDNA3.4- TOPO vector for expression in HEK293 cells. The Fc plasmid was subcloned out of this plasmid via PCR. Primers to introduce single or multiple mutations or deletions into the Fc plasmid were designed using the NebBaseChanger (https://nebasechanger.neb.com/) and PCR was carried out using the NEB Q5 High Fidelity polymerase and manufacturer's instructions All sequences were confirmed by Sanger sequencing at Genewiz (https://www.genewiz.com). The Fc mutants were expressed in either HEK293T or HEK293F cells. For HEK293T cells, the plasmids were transfected using polyethyleneimine as a transfection agent. After transfection, cells were cultured for 96h in Free- style F17 medium supplemented with GlutaMAX and Geneticin (Thermo Fisher Scientific). For Expj293 cells, the plasmids were transfected as per manufacturers protocol (MAN0007814, Thermo Fisher Scientific) with the addition of Penicillin/Streptamycin mix 24 hs after transfection. The cells were cultured for 96 hs before harvesting. The Fc mutants were purified using Protein A chromatography with PBS pH 7.4 being used as the binding buffer and 100 mM sodium citrate buffer pH 3.0 as the elution buffer. The fractions were neutralized with 1M Tris- HCl pH 9.3. SDS- PAGE was used to assess protein purity. The proteins were concentrated, flash- frozen and stored at \(-80^{0}\mathrm{C}\) until ready for use. + +<|ref|>sub_title<|/ref|><|det|>[[90, 663, 308, 682]]<|/det|> +## Hydrolytic activity assays + +<|ref|>text<|/ref|><|det|>[[87, 696, 912, 893]]<|/det|> +For the EndoS- Fc alanine scan, LC- MS kinetic analysis was used to determine the reaction rate of deglycosylation by EndoS against Fc mutants and by EndoS mutants against Fc WT. \(30~\mu \mathrm{L}\) reactions were setup containing \(5\mu \mathrm{M}\) of the Fc variant and \(1 - 20~\mathrm{nM}\) of the EndoS mutants in PBS pH 7.4. The reactions were analyzed by LC- MS using an Agilent 1290 Infinity II LC System equipped with a \(50\mathrm{mm}\) PLRP- S column from Agilent with \(1000\mathrm{\AA}\) pore size. The LC system is attached to either an Agilent 6545XT quadrupole- time of flight (Q- TOF) or Agilent 6560 Ion Mobility (IM) Q- TOF mass + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 912, 230]]<|/det|> +spectrometer (Agilent, Santa Clara, CA). The reactions were setup and placed in the LC- MS and the reactions were sampled approximately every 15 minutes till the reactions reached completion. All reactions were performed in triplicate. Relative amounts of the substrate and hydrolysis products were quantified after deconvolution of the raw data and identification of the corresponding peaks using BioConfirm (Agilent, Santa Clara, CA). + +<|ref|>text<|/ref|><|det|>[[87, 242, 912, 405]]<|/det|> +To obtain the rate of deglycosylation of Fc by the EndoS variants or Fc variants by EndoS, the relative proportions of the diglycosylated, monoglycosylated and deglycosylated products was imported into Kintek Global Kinetic Explorer64. This software fits the data to a model via the use of nonlinear regression analysis, where parameters are searched iteratively to determine a set of parameters which yields a minimum \(\chi^2\) value. The following model was used for all the fitted reactions: + +<|ref|>equation<|/ref|><|det|>[[350, 450, 647, 490]]<|/det|> +\[E + pGG\underset {k - 1}{\overset{k1}{\rightleftharpoons}}EpGG\underset {k - 2}{\overset{k2}{\rightleftharpoons}}EpG + G\] + +<|ref|>equation<|/ref|><|det|>[[368, 534, 627, 572]]<|/det|> +\[E + pG\underset {k - 3}{\overset{k3}{\rightleftharpoons}}EpG\underset {k - 4}{\overset{k4}{\rightleftharpoons}}Ep + G\] + +<|ref|>equation<|/ref|><|det|>[[434, 618, 562, 656]]<|/det|> +\[Ep\underset {k - 5}{\overset{k5}{\rightleftharpoons}}E + p\] + +<|ref|>text<|/ref|><|det|>[[87, 703, 911, 832]]<|/det|> +The rates were fitted in a stepwise manner starting from k1, and once the value which yielded a minimum \(\chi^2\) was determined the value was locked. All values except for k2 and k4 were locked during the fitting process. Statistical significance was determined using a multiple comparisons test (Tukey method) in GraphPad (Graphpad Software, La Jolla, CA). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 415, 90]]<|/det|> +## Molecular dynamics (MD) simulations + +<|ref|>text<|/ref|><|det|>[[86, 99, 912, 722]]<|/det|> +The EndoS \(_{\mathrm{E235A}}\) - Fc complex obtained by cryoEM was used as starting coordinates for all MD simulations. The different mutants were generated starting form this structure and mutating the corresponding residues with PyMOL 2.5. The coordinates of the carbohydrate shown in Fig. 4 of the manuscript was generated using the GLYCAM WEB (https://glycam.org) and was attached to the side chain of Asn297 of all simulated complexes. MD simulations were performed with AMBER 20 package \(^{65}\) , implemented with ff14SB \(^{66}\) , and GLYCAM 06j- 1 force fields \(^{67}\) . Each complex was immersed in a water box with a \(10 \mathrm{\AA}\) buffer of TIP3P water molecules \(^{68}\) . The system was neutralized by adding explicit counter ions. A two- stage geometry optimization approach was performed. The first stage minimizes only the positions of solvent molecules and ions, and the second stage is an unrestrained minimization of all the atoms in the simulation cell. The systems were then heated by incrementing the temperature from 0 to \(300 \mathrm{K}\) under a constant pressure of 1 atm and periodic boundary conditions. Harmonic restraints of \(10 \mathrm{kcal} \mathrm{mol}^{- 1}\) were applied to the solute, and the Andersen temperature coupling scheme \(^{69}\) was used to control and equalize the temperature. The time step was kept at 1 fs during the heating stages, allowing potential inhomogeneities to self- adjust. Hydrogen atoms were kept fixed through the simulations using the SHAKE algorithm \(^{70}\) . Long- range electrostatic effects were modelled using the particle- mesh- Ewald method \(^{71}\) . An \(8 \mathrm{\AA}\) cutoff was applied to Lennard- Jones interactions. Each system was equilibrated for 2 ns with a 2- fs time step at a constant volume and temperature of \(300 \mathrm{K}\) . Production trajectories were then run for additional \(1 \mu \mathrm{s}\) under the same simulation conditions. + +<|ref|>sub_title<|/ref|><|det|>[[89, 766, 577, 787]]<|/det|> +## Protein expression and purification for NMR experiments + +<|ref|>text<|/ref|><|det|>[[88, 797, 911, 927]]<|/det|> +\([U_{- }^{15}\mathrm{N},^{2}\mathrm{H}]\) , \(\epsilon - [^{13}\mathrm{C},^{1}\mathrm{H}_{3}]\) - Met, \(\delta 1 - [^{13}\mathrm{C}^{1}\mathrm{H}_{3}]\) - Ile, \(\delta 2 - [^{13}\mathrm{C},^{1}\mathrm{H}^{3}]\) - Leu, \(\gamma 2 - [^{13}\mathrm{C}^{1}\mathrm{H}_{3}]\) - Val- labeled (MILV) EndoS2, EndoS2 \(_{\mathrm{E186L}}\) , EndoBT- 3987 and EndoBT- 3987 \(_{\mathrm{D312A / E314L}}\) were expressed following an adapted version from previously reported protocol \(^{72}\) . Briefly, \(E\) . coli BL21(DE3) containing the gene of interest were grown in \(20 \mathrm{ml}\) of LB Lennox medium (Roth) until an optical density of \(600 \mathrm{nm}\) ( \(\mathrm{OD}_{600}) > 1.5\) was + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 67, 912, 580]]<|/det|> +reached. Ampicillin (100 \(\mu \mathrm{g / mL}\) ) was used as selecting agent through the expression. Unless otherwise stated, bacteria were grown at \(37^{\circ}\mathrm{C}\) under shaking (220 rpm). Cells for inoculation of \(10\mathrm{mL}\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium with a starting \(\mathrm{OD}_{600}\) of 0.1 were harvested by centrifugation, and excess of TB medium was removed. In all \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal media, \(3\mathrm{g / L}\) of \(^{15}\mathrm{N}\) - ammonium chloride (Deutero) and \(3\mathrm{g / L}\) of deuterated \(^{12}\mathrm{C}\) - glucose (1,2,3,4,5,6,6- d7, Deutero) were used as the principal nitrogen and carbon sources, respectively. The next morning \(2\mathrm{ml}\) of the starter culture were spin- down, supernatant was removed and cells were transferred into \(20\mathrm{mL}\) of freshly prepared \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium. When an \(\mathrm{OD}_{600}\) of 0.4 was reached, the culture volume was increased to \(90\mathrm{ml}\) and cells were grown until an \(\mathrm{OD}_{600}\) of 0.6- 0.8 was reached. At this point, the temperature of the incubator was reduced to \(16^{\circ}\mathrm{C}\) and \(10\mathrm{mL}\) of \(\mathrm{M9}^{+} / \mathrm{D}_2\mathrm{O}\) minimal medium containing the desired labeled precursors and amino acids were added \(^{72}\) . Amounts of isotopically labelled precursors for selective methyl labeling are given in Supplementary Table 4. Protein expression was induced with the addition of \(1\mathrm{mM}\) or \(0.5\mathrm{mM}\) isopropyl \(\beta\) - D- 1- thio- galactopyranoside (IPTG) for EndoS2 or EndoBT constructs, respectively. Cells were harvested when the maximal cell density was reached and stored at \(- 20^{\circ}\mathrm{C}\) . Labeled protein was purified as previously described \(^{28,38}\) and stored at \(4^{\circ}\mathrm{C}\) . + +<|ref|>sub_title<|/ref|><|det|>[[89, 626, 461, 647]]<|/det|> +## Sample preparation and NMR spectroscopy + +<|ref|>text<|/ref|><|det|>[[88, 660, 911, 787]]<|/det|> +All NMR experiments were acquired on a \(600\mathrm{MHz}\) Avance III spectrometer equipped with TCI cryogenic probe at \(298\mathrm{K}\) , unless otherwise stated. NMR spectra were processed with TopSpin 4.0.6 (Bruker) or with NMRPipe \(^{73}\) prior to analysis. \(^{1}\mathrm{H}\) chemical shifts were referenced to the DSS- \(\mathrm{d}_6\) peak, and \(^{13}\mathrm{C}\) and \(^{15}\mathrm{N}\) signals were referenced indirectly. + +<|ref|>sub_title<|/ref|><|det|>[[90, 835, 420, 856]]<|/det|> +## Enzyme kinetics by NMR spectroscopy + +<|ref|>text<|/ref|><|det|>[[89, 869, 910, 925]]<|/det|> +Rituximab- Fc was concentrated using \(10\mathrm{kDa}\) MWCO Amicon Ultra- 4 filters (Merck) and the buffer was exchanged against NMR kinetics buffer, which contained \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\) (Eurisotop) pH 7.50, \(50\mathrm{mM}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 910, 160]]<|/det|> +NaCl, \(0.1 \mathrm{mM} 2,2\) - Dimethyl- 2- silapentane- 5- sulfonate- \(\mathrm{d}_{6}\) (DSS- \(\mathrm{d}_{6}\) , Sigma- Aldrich) and \(0.02\%\) \(\mathrm{NaN}_{3}\) in \(8\% \mathrm{D}_{2}\mathrm{O}\) (Eurisotop, \(99.96\%\) ). Samples were prepared per duplicate at 3, 5, 10, 20 and \(40 \mu \mathrm{M}\) Rituximab- Fc. All samples contained a final \(160 \mu \mathrm{l}\) volume in \(3 \mathrm{mm}\) NMR tubes. + +<|ref|>text<|/ref|><|det|>[[87, 173, 912, 579]]<|/det|> +For each kinetics experiment, the sample was introduced in the magnet, locked, matched, tuned, shimmed and the \(90^{\circ}\) pulse was calibrated. The temperature was allowed to equilibrate for 5 min and one perfect CPMG was acquired as reference (for details on the pulse sequence see Supplementary Fig. 17). The sample was removed from the magnet, \(1 \mathrm{nM}\) EndoS \(2_{\mathrm{wt}}(2 \mu \mathrm{l})\) was added directly into the NMR tube, mixed \(3 \mathrm{x}\) by inversion and the tube was immediately inserted into the magnet. After locking and shimming, a series of perfect CPMG spectra were recorded for a maximum of \(24 \mathrm{h}\) . From enzyme addition to the beginning of the first acquisition a max. of \(3 \mathrm{min} 20 \mathrm{s}\) were measured, although \(< 3 \mathrm{min}\) were normally required. Perfect CPMGs were acquired with an echo time \(\tau\) of \(0.3 \mathrm{ms}\) and a total \(T_{2}\) relaxation time of \(39.6 \mathrm{ms}\) , \(T_{1}\) relaxation delay (D1) of \(1 \mathrm{s}\) , \(32\) transients, \(4, 32\) or \(64 \mathrm{dummy}\) scans, an FID size of \(32 \mathrm{k}\) and \(12 \mathrm{ppm}\) sweep width, with a total experimental time of \(2 \mathrm{min}\) . After completion of the reaction, a perfect CPMGs was acquired with a \(T_{1}\) relaxation delay (D1) of \(20 \mathrm{s}\) to correct for insufficient \(T_{1}\) relaxation. + +<|ref|>text<|/ref|><|det|>[[88, 591, 911, 717]]<|/det|> +Reaction time- courses were analyzed from the corrected integrals \((I_{cor})\) obtained from the composite signal between 2.08- 2.11 ppm, which corresponds to \(N\) - acetylglucosamine and \(N\) - acetylneuraminic acid moieties of cleaved \(N\) - glycans (Supplementary Fig. 18A). \(I_{cor}\) were calculated according to Eq. 2. + +<|ref|>equation<|/ref|><|det|>[[88, 768, 664, 789]]<|/det|> +\[I_{cor} = (I_{n} - I_{ref})C \quad (Eq. 2)\] + +<|ref|>text<|/ref|><|det|>[[88, 836, 910, 893]]<|/det|> +where \(I_{n}\) corresponds to the integral of the \(n^{th}\) experiment after the addition of enzyme, \(I_{ref}\) indicates the integral of the reference experiment (previous to enzyme addition) and \(C\) denotes a correction factor for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 68, 911, 230]]<|/det|> +insufficient \(T_{l}\) relaxation, calculated as the ratio between the integrals of the experiment acquired with \(\mathrm{D1} = 20\mathrm{s}\) and the last spectrum recorded with \(\mathrm{D1} = 1\mathrm{s}\) . \(I_{cor}\) were plotted against time to generate reaction time- courses (Supplementary Fig. 18B), which were then fitted to a single exponential decay Eq. 3. Initial enzyme cycling velocity \(\nu_{0}\) was approximated as the first derivative of Eq. 3 evaluated at time \(= 0\) (Eq. 4). + +<|ref|>equation<|/ref|><|det|>[[88, 277, 664, 352]]<|/det|> +\[I_{c o r}(t) = 1 - e^{(-t\cdot a)}\] \[\nu_{0} = \frac{d(1 - e^{-t\cdot a})}{a t}\bigg|_{t = 0} = a\] + +<|ref|>text<|/ref|><|det|>[[87, 399, 911, 594]]<|/det|> +Initial enzyme cycling velocities \(\nu_{0}\) were plotted against the initial concentration of complex glycan covalently attached to Fc or [S] \(_{0}\) , which corresponds to two times the Fc concentration. Fitting to the Michaelis- Menten equation Eq. 5 afforded the substrate concentration- dependence of enzyme activation \((K_{\mathrm{M}})\) and the maximum velocity \((V_{m a x})\) . The maximum enzyme catalytic rate \((k_{c a t})\) was calculated as \(V_{m a x} / [\mathrm{E n z y m e}]_{0}\) . Fittings were performed using in- house Matlab R2019b script, and errors are given as one standard deviation. + +<|ref|>equation<|/ref|><|det|>[[88, 642, 664, 675]]<|/det|> +\[\nu_{0} = \frac{V_{m a x}[S]_{0}}{K_{M} + [S]_{0}} \quad (Eq. 5)\] + +<|ref|>text<|/ref|><|det|>[[88, 692, 142, 708]]<|/det|> +where + +<|ref|>equation<|/ref|><|det|>[[88, 723, 664, 758]]<|/det|> +\[K_{M} = \frac{k_{o f f} + k_{c a t}}{k_{o n}} \quad (Eq. 6)\] + +<|ref|>sub_title<|/ref|><|det|>[[88, 806, 670, 828]]<|/det|> +## Synthesis and purification of Rituximab-Fc aglycan and \(\mathbf{CT_{n - 1}}\) glycan + +<|ref|>text<|/ref|><|det|>[[88, 841, 911, 896]]<|/det|> +Rituximab- Fc aglycan was obtained by pooling the products of the enzymatic reactions of Rituximab- Fc with EndoS2 used for enzyme kinetics. Rituximab- Fc aglycan was separated from EndoS2 and free + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 911, 195]]<|/det|> +glycans by size exclusion chromatography (HiLoad 16/600 Superdex 200 pg, GE), using \(20\mathrm{mM}\) Tris- HCl buffer pH 7.5, \(50\mathrm{mM}\) NaCl as buffer. Fractions showing desired protein in UV (280 nm) were pooled, concentrated and the buffer was exchanged against NMR titration buffer as explained above. The final stock solution contained a concentration of \(323\mu \mathrm{M}\) Rituximab- Fc aglycan. + +<|ref|>text<|/ref|><|det|>[[87, 208, 912, 439]]<|/det|> +Glycan \(\mathrm{CT_{n - 1}}\) was obtained by digestion of CT with EndoS2 as explained in the previous chapter (enzyme kinetics by NMR spectroscopy). After reaction completion samples were lyophilized, dissolved in \(\mathrm{H}_2\mathrm{O}\) and the reaction product \(\mathrm{CT_{n - 1}}\) was isolated by HPLC (Jasco Extrema) using a semipreparative VP 250/21 Nucleosil 100- 7 C18 column (Macherey- Nagel, cat. No.: 715352) applying as solvents \(\mathrm{H}_2\mathrm{O}\) and MeOH (both with \(0.01\%\) TFA) with a flow of \(10\mathrm{ml / min}\) and the following gradients: \(20\mathrm{min}0\%\) MeOH followed by \(70\mathrm{min}0\%\) to \(20\%\) MeOH. Fractions of \(10\mathrm{ml}\) were collected, and the desired sugar eluted in fractions 10 to 12, which were pulled together and lyophilized. + +<|ref|>sub_title<|/ref|><|det|>[[90, 486, 719, 508]]<|/det|> +## Assignment of \(^{1}\mathrm{H},^{13}\mathrm{C}\) NMR signals from CT, \(\mathbf{CT_{n - 1}}\) and HM carbohydrates + +<|ref|>text<|/ref|><|det|>[[86, 520, 912, 927]]<|/det|> +The identity and purity of CT, \(\mathrm{CT_{n - 1}}\) and HM carbohydrates was confirmed by NMR prior titrations. For this end, all NMR signals from \(^{1}\mathrm{H},^{13}\mathrm{C}\) - HSQC spectra from samples were assigned. Samples contained \(0.25\mathrm{mM}\) to \(2.0\mathrm{mM}\) carbohydrate concentrations in \(50\mathrm{mM}\) natrium phosphate buffer \(\mathrm{pH}^{*}7.30\) , \(50\mathrm{mM}\) NaCl, \(100\mu \mathrm{M}\) TSP- d4, \(0.02\%\) \(\mathrm{NaN}_3\) in \(\mathrm{D}_2\mathrm{O}\) in \(3\mathrm{mm}\) NMR tubes at \(160\mu \mathrm{l}\) total sample volume. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted in the indirect dimension prior Fourier- transformation (128 LP coefficients), affording a \(4096\mathrm{x}4096\) data matrix. Spectra were manually phased and peak peaking was performed using CCPNMR Analysis 2.4.2 software suit \(^{72}\) . Supplementary Fig. 19 shows a cartoon representation and the chemical structure of the assigned carbohydrates. Experimental details used for the assignment and \(^{1}\mathrm{H}/^{13}\mathrm{C}\) chemical shifts are listed in Supplementary Table 5. Assignments are shown in Supplementary Fig. 20 and Supplementary Table 6. Assignments were in excellent agreement with previously reported frequencies \(^{74,75}\) . All experiments for the assignment of carbohydrates were acquired at \(310\mathrm{K}\) . + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[88, 104, 908, 159]]<|/det|> +## Titration of selectively MILV methyl-labeled enzymes with carbohydrates, Rituximab-Fc, Rituximab-Fc aglycan and CaCl2 + +<|ref|>text<|/ref|><|det|>[[87, 172, 912, 370]]<|/det|> +MILV methyl- labeled proteins were transferred into NMR titration buffer using \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns (Thermo Fischer Scientific), which contained \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\) pH\\* 7.50, \(50~\mathrm{mM}\) NaCl, \(0.1\mathrm{mM}\) DSS- \(\mathrm{d}_6\) and \(0.02\%\) \(\mathrm{NaN}_3\) in \(\mathrm{D}_2\mathrm{O}\) . Protein samples were prepared at \(45 - 128\mu \mathrm{M}\) protein concentration at \(160~\mu \mathrm{l}\) final volume in \(3\mathrm{mm}\) NMR tubes. Protein concentrations were determined after buffer exchange by UV absorbance at \(280\mathrm{nm}\) with \(\epsilon = 104.17\mathrm{M}^{- 1}\mathrm{cm}^{- 1}\) for EndoS2 and \(\mathrm{EndoS2_{E186L}}\) , and with \(\epsilon = 62.23\mathrm{M}^{- 1}\mathrm{cm}^{- 1}\) for EndoBT- 3987 and EndoBT- 3987D312A/E314L. + +<|ref|>text<|/ref|><|det|>[[87, 382, 911, 507]]<|/det|> +CT, \(\mathrm{CT_{n - 1}}\) , HM and \(\mathrm{CaCl}_2\) used for titrations were dissolved in \(\mathrm{D}_2\mathrm{O}\) and lyophilized twice. Final stock solutions were prepared at high concentrations in NMR titration buffer, and the \(\mathrm{pH}^*\) was carefully readjusted to 7.50. Rituximab- Fc was buffer exchanged against NMR titration buffer using \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns and concentrated to a final \(380~\mu \mathrm{M}\) concentration stock solution. + +<|ref|>text<|/ref|><|det|>[[87, 518, 912, 717]]<|/det|> +\(^{1}\mathrm{H}\) , \(^{13}\mathrm{C}\) HMQC spectra (methyl TROSY) \(^{33}\) were acquired with 107 ms acquisition time and a spectral window of \(4\mathrm{ppm}\) in the direct dimension. In the indirect dimension, the spectral window was set to 22 or 24 ppm with 512 or 256 increments. The relaxation delay was set to 1.5 s and 4 to 16 transients were acquired. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted (64 LP coefficients) prior Fourier- transformation, affording a 2048x2048 data matrix with a spectral resolution of \(1.17\mathrm{Hz}\) and \(1.62\mathrm{Hz}\) in the direct and indirect dimensions, respectively. + +<|ref|>text<|/ref|><|det|>[[87, 729, 912, 926]]<|/det|> +Lineshape analysis of cross peaks of \(^{1}\mathrm{H}\) , \(^{13}\mathrm{C}\) HMQC spectra of MILV- labeled samples with the program TITAN \(^{34}\) was performed using Matlab R2019b for titrations shown in Table 1. Spectra were processed in NMRPipe \(^{73}\) prior to analysis. A representative example of the shell scripts used is compiled in Supplementary Table 7. We selected a two states binding model as the simplest model appropriately describing the experimental results (Eq. 1). For each titration series the fitting algorithm was as follows: first, linewidths and chemical shift were fitted using the spectrum from the apo form. Next, linewidths + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 68, 911, 195]]<|/det|> +and chemical shift of the fully bound state were fitted using the spectrum of the protein at highest ligand concentration. Linewidths, off- rate constant \(k_{off}\) and dissociation constant \(K_{D}\) were finally fitted from spectra of all titration points and using previously determined linewidths and chemical shifts as starting values. Parameter uncertainties were obtained by bootstrap error analysis with 100 iterations. + +<|ref|>text<|/ref|><|det|>[[88, 207, 911, 337]]<|/det|> +Titrations of \(\mathrm{CaCl_2}\) showed fast- exchange in the NMR time- scale between the unbound and bound forms. Therefore, signal position were extracted with CCPNMR Analysis v2.4.2 software package \(^{76}\) and used obtain Euclidean chemical shift perturbances (CSPs) measured as Euclidian distances \(\Delta \nu_{\mathrm{Eucl}}\) according to Eq. 6: + +<|ref|>equation<|/ref|><|det|>[[88, 385, 723, 411]]<|/det|> +\[\Delta \nu_{Eucl} = \sqrt{\Delta\nu_{H}^{2} + \Delta\nu_{C}^{2}} \quad (Eq. 6)\] + +<|ref|>text<|/ref|><|det|>[[88, 458, 910, 553]]<|/det|> +with \(\Delta \nu_{H}\) and \(\Delta \nu_{C}\) being the CSPs in the respective dimension in Hz. In a simple two states model (Eq. 1), observed CSPs \(\Delta \nu_{obs}\) at a given total ligand concentration \(L_{t}\) are linked to the dissociation constant \(K_{D}\) via the law of mass action \(^{32}\) (Eq. 7). + +<|ref|>equation<|/ref|><|det|>[[88, 602, 728, 636]]<|/det|> +\[\Delta \nu_{obs} = \frac{(P_{t} + L_{t} + K_{D}) - \sqrt{(P_{t} + L_{t} + K_{D})^{2} - 4P_{t}L_{t}}}{2P_{t}}\Delta \nu_{max} \quad (Eq. 7)\] + +<|ref|>text<|/ref|><|det|>[[88, 684, 911, 810]]<|/det|> +where \(P_{t}\) is the total protein concentration, and \(\Delta \nu_{max}\) is the maximum CSP at ligand saturation for each signal. Global fittings were performed using in- house Matlab R2019b script. Errors were determined from a Monte Carlo approach with 100 iterations as previously described \(^{77}\) , and are given as one standard deviation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 68, 785, 91]]<|/det|> +## \(^{1}\mathbf{H},^{15}\mathbf{N}\) HSQC-TROSY spectra and determination of rotational correlation times \(\tau_{c}\) + +<|ref|>text<|/ref|><|det|>[[87, 100, 912, 510]]<|/det|> +\([U_{- }^{15}\mathrm{N},^{2}\mathrm{H}]\) MILV methyl- labeled proteins were buffer exchanged \(2\mathrm{mL}\) ZebaTM Spin Desalting Columns against the following buffer: \(20\mathrm{mM}\) Tris- \(\mathrm{d}_{11}\mathrm{pH}^{*}7.50,50\mathrm{mM}\) NaCl, \(0.1\mathrm{mM}\) DSS- \(\mathrm{d}_6\) and \(0.02\%\) \(\mathrm{NaN}_3\) in \(8\%\) \(\mathrm{D}_2\mathrm{O}\) . Two samples containing \(\mathrm{EndoS2_{E186L}}\) and \(\mathrm{EndoBT_{D312A / E314L}}\) were prepared at \(71.5\mu \mathrm{M}\) or \(74.4\mu \mathrm{M}\) protein concentrations, respectively. Samples were allowed to deuterium- hydrogen back exchange for 5 days at \(6^{\circ}\mathrm{C}\) , and then measured in \(3\mathrm{mm}\) NMR tubes at a final \(160\mu \mathrm{l}\) volume. \(^{1}\mathrm{H},^{15}\mathrm{N}\) HSQC- TROSY spectra were acquired were acquired with \(128\mathrm{ms}\) acquisition time and a spectral window of \(13.3\mathrm{ppm}\) in the direct dimension. In the indirect dimension, the spectral window was set to \(45\mathrm{ppm}\) with 512 increments. The relaxation delay was set to \(1.5\mathrm{s}\) and 16 (EndoBT- 3987 \(\mathrm{D_{312A / E314L}}\) ) or 48 (EndoS2 \(\mathrm{E_{186L}}\) ) transients were acquired. Data was apodized with a QSINE window function, FIDs were zero- filled and forward linear- predicted (16 LP coefficients) prior Fourier- transformation, affording a \(4096\times 1024\) data matrix with a spectral resolution of \(1.95\mathrm{Hz}\) and \(2.67\mathrm{Hz}\) in the direct and indirect dimensions, respectively. + +<|ref|>text<|/ref|><|det|>[[87, 522, 911, 683]]<|/det|> +Rotational correlation times \(\tau_{c}\) of MILV \(\mathrm{EndoS2_{E186L}}\) and \(\mathrm{EndoBT_{D312A / E314L}}\) proteins have been estimated from the samples prepared above using \(^{1}\mathrm{H},^{15}\mathrm{N}\) - TRACT experiments \(^{78}\) . The relaxation delay was set to 2 s. Experiments were measured for 40 transients with 30 increasing delays of up to 1.5 s. Data were integrated from 8- 10 ppm, normalized and fitted to an exponential decay model for determination of average \(^{15}\mathrm{N} R_{\alpha}\) and \(R_{\beta}\) . + +<|ref|>sub_title<|/ref|><|det|>[[90, 733, 231, 752]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[88, 766, 911, 893]]<|/det|> +The atomic coordinates and structure factors have been deposited with the Protein Data Bank (PDB), accession code 8A64 (EndoS \(\mathrm{E_{235A - Fc}}\) ). Previously published PDB structures used in this study are available under the accession codes: 6MDS, 4NUZ, 1H3X and 6EN3. All other data are available from the corresponding authors upon reasonable request. Source data are provided with this paper. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 71, 184, 88]]<|/det|> +## References + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 68, 904, 923]]<|/det|> +Specific Labeling and Conjugation of Antibodies: Facile Synthesis of Homogeneous Antibody- Drug Conjugates. ACS Chem. Biol. 16, 2502- 2514 (2021).19. Shi, W. et al. One- step synthesis of site- specific antibody- drug conjugates by reprogramming IgG glycoengineering with LacNAc- based substrates. Acta Pharm. Sin. B 12, 2417- 2428 (2022).20. Pučić, M. et al. High throughput isolation and glycosylation analysis of IgG- variability and heritability of the IgG glycome in three isolated human populations. Mol. Cell. Proteomics (2011). doi:10.1074/mcp.M111.01009021. Mimura, Y. et al. Glycosylation engineering of therapeutic IgG antibodies: challenges for the safety, functionality and efficacy. Protein Cell 9, 47- 62 (2018).22. Lu, R.- M. et al. Development of therapeutic antibodies for the treatment of diseases. J. Biomed. Sci. 27, 1 (2020).23. Huang, W., Giddens, J., Fan, S.- Q. Q., Toonstra, C. & Wang, L.- X. X. Chemoenzymatic glycoengineering of intact IgG antibodies for gain of functions. J. Am. Chem. Soc. 134, 12308- 12318 (2012).24. Li, T., Tong, X., Yang, Q., Giddens, J. P. & Wang, L. X. Glycosynthase mutants of endoglycosidase S2 show potent transglycosylation activity and remarkably relaxed substrate specificity for antibody glycosylation remodeling. J. Biol. Chem. 29, 16508- 16518 (2016).25. Majewska, N. I., Tejada, M. L., Betenbaugh, M. J. & Agarwal, N. N- glycosylation of IgG and IgG- like recombinant therapeutic proteins: Why is it important and how can we control it? Annu Rev Chem Biomol Eng 7, 311- 338 (2020).26. Trastoy, B. et al. Crystal structure of Streptococcus pyogenes EndoS, an immunomodulatory endoglycosidase specific for human IgG antibodies. Proc. Natl. Acad. Sci. 111, 6714- 9 (2014).27. Trastoy, B. et al. Structural basis for the recognition of complex- type N- glycans by Endoglycosidase S. Nat. Commun. 9, 1874 (2018).28. Klontz, E. H. et al. Molecular Basis of Broad Spectrum N- Glycan Specificity and Processing of Therapeutic IgG Monoclonal Antibodies by Endoglycosidase S2. ACS Cent. Sci. 5, 524- 538 (2019).29. Kastner, B. et al. GraFix: Sample preparation for single- particle electron cryomicroscopy. Nat. Methods 5, 53- 55 (2008).30. Krissinel, E. & Henrick, K. Inference of Macromolecular Assemblies from Crystalline State. J. Mol. Biol. (2007). doi:10.1016/j.jmb.2007.05.02231. Dixon, E. V. et al. Fragments of bacterial endoglycosidase S and immunoglobulin G reveal subdomains of each that contribute to deglycosylation. J. Biol. Chem. 289, 13876- 13889 (2014).32. Williamson, M. P. Using chemical shift perturbation to characterise ligand binding. Prog. Nucl. Magn. Reson. Spectrosc. 73, 1- 16 (2013).33. Tugarinov, V., Hwang, P. M., Ollerenshaw, J. E. & Kay, L. E. Cross- correlated relaxation enhanced 1H[bond]13C NMR spectroscopy of methyl groups in very high molecular weight proteins and protein complexes. J. Am. Chem. Soc. 125, 10420- 10428 (2003).34. Waudby, C. A., Ramos, A., Cabrita, L. D. & Christodoulou, J. Two- Dimensional NMR Lineshape Analysis. Sci. Rep. 6, 24826 (2016).35. Schütz, S. & Sprangers, R. Methyl TROSY spectroscopy: A versatile NMR approach to study + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[84, 68, 900, 905]]<|/det|> +challenging biological systems. Prog. Nucl. Magn. Reson. Spectrosc. 116, 56- 84 (2020).36. Tugarinov, V., Kanelis, V. & Kay, L. E. Isotope labeling strategies for the study of high- molecular- weight proteins by solution NMR spectroscopy. Nat. Protoc. 1, 749- 754 (2006).37. Tugarinov, V. & Kay, L. E. An isotope labeling strategy for methyl TROSY spectroscopy. J. Biomol. NMR 28, 165- 172 (2004).38. Trastoy, B. et al. Structural basis of mammalian high- mannose N- glycan processing by human gut Bacteroides. Nat. Commun. 11, 899 (2020).39. Trastoy, B. et al. GH18 endo- \(\beta\) - N- acetylglucosaminidases use distinct mechanisms to process hybrid- type N- linked glycans. J. Biol. Chem. 297, 101011 (2021).40. Waddling, C. A., Plummer, T. H. J., Tarentino, A. L. & Van Roey, P. Structural basis for the substrate specificity of endo- beta- N- acetylglucosaminidase F(3). Biochemistry 39, 7878- 7885 (2000).41. Deisenhofer, J. Crystallographic refinement and atomic models of a human Fc fragment and its complex with fragment B of protein A from Staphylococcus aureus at 2.9- and 2.8- A resolution. Biochemistry 20, 2361- 2370 (1981).42. Sauer- Eriksson, A. E., Kleywegt, G. J., Uhlen, M. & Jones, T. A. Crystal structure of the C2 fragment of streptococcal protein G in complex with the Fc domain of human IgG. Structure 3, 265- 278 (1995).43. Burmeister, W. P., Huber, A. H. & Bjorkman, P. J. Crystal structure of the complex of rat neonatal Fc receptor with Fc. Nature 372, 379- 383 (1994).44. Barb, A. W. & Prestegard, J. H. NMR analysis demonstrates immunoglobulin G N- glycans are accessible and dynamic. Nat. Chem. Biol. (2011). doi:10.1038/nchembio.51145. Yamaguchi, Y. & Barb, A. W. A synopsis of recent developments defining how N- glycosylation impacts immunoglobulin G structure and function. Glycobiology 30, 214- 225 (2020).46. Stark, H. GraFix: Stabilization of fragile macromolecular complexes for single particle Cryo- EM. Methods in Enzymology 481, (Elsevier Inc., 2010).47. Kandiah, E. et al. CM01: A facility for cryo- electron microscopy at the European synchrotron. Acta Crystallogr. Sect. D Struct. Biol. 75, 528- 535 (2019).48. Zivanov, J., Nakane, T. & Scheres, S. H. W. Estimation of high- order aberrations and anisotropic magnification from cryo- EM data sets in RELION- 3.1. IUcrJ 7, 253- 267 (2020).49. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nat. Methods 14, 290- 296 (2017).50. Punjani, A., Zhang, H. & Fleet, D. J. Non- uniform refinement: adaptive regularization improves single- particle cryo- EM reconstruction. Nat. Methods 17, 1214- 1221 (2020).51. Sanchez- Garcia, R. et al. DeepEMhancer: a deep learning solution for cryo- EM volume post- processing. Commun. Biol. 4, 874 (2021).52. Cardone, G., Heymann, J. B. & Steven, A. C. One number does not fit all: mapping local variations in resolution in cryo- EM reconstructions. J. Struct. Biol. 184, 226- 236 (2013).53. Pettersen, E. F. et al. UCSF Chimera - A visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605- 1612 (2004). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 68, 910, 920]]<|/det|> +54. Adams, P. D. et al. PHENIX: A comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 213–221 (2010).55. Afonine, P. V et al. Real-space refinement in PHENIX for cryo-EM and crystallography. Acta Crystallogr. Sect. D, Struct. Biol. 74, 531–544 (2018).56. Emsley, P. & Cowtan, K. Coot: Model-building tools for molecular graphics. Acta Crystallogr. Sect. D Biol. Crystallogr. (2004). doi:10.1107/S090744490401915857. Chen, V. B. et al. MolProbity: All-atom structure validation for macromolecular crystallography. Acta Crystallogr. Sect. D Biol. Crystallogr. 66, 12–21 (2010).58. Nakane, T., Kimanius, D., Lindahl, E. & Scheres, S. H. Characterisation of molecular motions in cryo-EM single-particle data by multi-body refinement in RELION. Elife 7, (2018).59. Crowther, R. A., Henderson, R. & Smith, J. M. MRC image processing programs. J. Struct. Biol. 116, 9–16 (1996).60. Rambo, R. P. & Tainer, J. A. Characterizing flexible and intrinsically unstructured biological macromolecules by SAS using the Porod-Debye law. Biopolymers 95, 559–571 (2011).61. Petoukhov, M. V. et al. New developments in the ATSAS program package for small-angle scattering data analysis. J. Appl. Crystallogr. 45, 342–350 (2012).62. Svergun, D. I. Determination of the regularization parameter in indirect-transform methods using perceptual criteria. J. Appl. Crystallogr. 25, 495–503 (1992).63. Tuukkanen, A. T., Kleywegt, G. J. & Svergun, D. I. Resolution of ab initio shapes determined from small-angle scattering. IUCrJ 3, 440–447 (2016).64. Johnson, K. A. Fitting enzyme kinetic data with KinTek Global Kinetic Explorer. Methods Enzymol. 467, 601–626 (2009).65. Case, D. A. et al. AMBER 2020. (2020).66. Maier, J. A. et al. ff14SB: Improving the Accuracy of Protein Side Chain and Backbone Parameters from ff99SB. J. Chem. Theory Comput. 11, 3696–3713 (2015).67. Kirschner, K. N. et al. GLYCAM06: a generalizable biomolecular force field. Carbohydrates. J. Comput. Chem. 29, 622–655 (2008).68. Jorgensen, W. L., Chandrasekhar, J. & Madura, J. D. Comparison of simple potential functions for simulating liquid water. J. Chem. Phys. 79, 926–935 (1983).69. Andersen, H. C. No TitleMolecular dynamics simulations at constant pressure and/or temperature. J. Chem. Phys. 72, 2384–2393 (1980).70. Miyamoto, S. & Kollman, P. A. Settle: An analytical version of the SHAKE and RATTLE algorithm for rigid water model. J. Comput. Chem. 13, 952–962 (1992).71. Darden, T. & York, D. Particle mesh Ewald: An N·log(N) method for Ewald sums in large systems. J. Chem. Phys. 98, 10089–10092 (1993).72. Müller-Hermes, C., Creutznacher, R. & Mallagaray, A. Complete assignment of Ala, Ile, Leu(ProS), Met and Val(ProS) methyl groups of the protruding domain from human norovirus GII.4 Saga. Biomol. NMR Assign. 14, 123–130 (2020).73. Delaglio, F. et al. NMRPipe: a multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 70, 900, 457]]<|/det|> +74. Unione, L. et al. Glycoprofile Analysis of an Intact Glycoprotein As Inferred by NMR Spectroscopy. ACS Cent. Sci. 5, 1554–1561 (2019).75. Schubert, M., Walczak, M. J., Aebi, M. & Wider, G. Posttranslational modifications of intact proteins detected by NMR spectroscopy: application to glycosylation. Angew. Chem. Int. Ed. Engl. 54, 7096–7100 (2015).76. Vranken, W. F. et al. The CCPN data model for NMR spectroscopy: development of a software pipeline. Proteins 59, 687–696 (2005).77. Arai, M., Ferreon, J. C. & Wright, P. E. Quantitative analysis of multisite protein-ligand interactions by NMR: binding of intrinsically disordered p53 transactivation subdomains with the TAZ2 domain of CBP. J. Am. Chem. Soc. 134, 3792–3803 (2012).78. Lee, D., Hilty, C., Wider, G. & Wüthrich, K. Effective rotational correlation times of proteins from NMR relaxation interference. J. Magn. Reson. 178, 72–76 (2006).79. Collin, M. & Olsén, A. Extracellular enzymes with immunomodulating activities: variations on a theme in Streptococcus pyogenes. Infect. Immun. 71, 2983–2992 (2003).80. Kagawa, T. F. et al. Crystal structure of the zymogen form of the group A Streptococcus virulence factor SpeB: an integrin-binding cysteine protease. Proc. Natl. Acad. Sci. U. S. A. 97, 2235–2240 (2000).81. Wenig, K. et al. Structure of the streptococcal endopeptidase IdeS, a cysteine proteinase with strict specificity for IgG. Proc. Natl. Acad. Sci. U. S. A. 101, 17371–17376 (2004). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 70, 247, 88]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[87, 103, 912, 405]]<|/det|> +This work was supported by the MICINN/FEDER EU grant PID2019- 105649RB- I00 (MEG), MINECO/FEDER EU grant Severo Ochoa Excellence Accreditation grant SEV- 2016- 0644 (MEG), Basque Government contract KK- 2021- 00034 (MEG), National Institutes of Health grant R01AI149297 (EJS, LXW, MEG), National Institutes of Health grant R01GM096973 (LXW), European Union Horizon 2020, Marie Sklodowska- Curie grant 844905 and "Ramón y Cajal" fellow from the Spanish Ministry of Economy and Competitiveness (BT), La Caixa Foundation grant LCF/BQ/DR19/11740011 (MGA), European Fonds for Regional Development, LPW- E/1.1.2/857 (AM), Agencia Estatal Investigación of Spain (AEI, RTI- 2018- 099592- B- C21, FC), European Union's Horizon 2020 research and innovation program under the Marie Sklodowska Curie grant agreement no. 956544 (FC). + +<|ref|>sub_title<|/ref|><|det|>[[90, 454, 270, 472]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[88, 486, 890, 576]]<|/det|> +BT, JJD, LXW, FC, AM, EJS and MEG, conceived the project. BT, JJD, JOC, LR, MGA, EHK, DD, CL, DES, FC and AM, performed the experiments. BT, JJD, JOC, FC, AM, EJS and MEG, analyzed the results. BT, JJD, FC, AM, EJS and MEG, wrote the paper. + +<|ref|>sub_title<|/ref|><|det|>[[90, 628, 261, 646]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[89, 662, 520, 681]]<|/det|> +Authors declare that they have no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[90, 747, 286, 765]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[90, 789, 870, 809]]<|/det|> +The online version contains supplementary material available at https://www.nature.com/ncomms/. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 70, 884, 860]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 866, 910, 920]]<|/det|> +
Fig. 1 Overall structure of EndoS in complex with the Fc region. A EndoS, SpeB and IdeS are secreted enzymes from S. pyogenes that can act on IgG antibodies. SpeB and IdeS are proteases that hydrolyze the flexible hinge region between residues 220 to 248 of the IgG antibody \(^{79 - 81}\) . EndoS is an
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 70, 909, 317]]<|/det|> +ENGase that hydrolyze the \(\beta - 1,4\) linkage between the first two GlcNAcs of the CT \(N\) - glycans on the Fc region of IgG antibodies. The three enzymes inactivate IgG antibodies against S. pyogenes. B EndoS and EndoS2 can be used to remodel the \(N\) - glycan on therapeutic IgG antibodies using a chemoenzymatic approach. First, wild type EndoS and/or EndoS2 hydrolyze a variety of glycoforms present in the Fc region according to their \(N\) - glycan specificity. Next, other enzymes can be used to hydrolyze specific carbohydrate moieties, e.g. fucose which absence improves the binding to FcγRIIIa and the antibody- dependent cellular cytotoxicity (ADCC) properties of the antibody. Last, EndoS \(D_{233Q}\) and EndoS \(2_{D184M}\) glycosynthase mutants facilitates the transfer of a glycan- oxazoline donor with a defined glycoform to the Fc region of IgG antibodies. The dotted shapes represent that carbohydrate may or may not be present, indicating the high heterogeneity of the \(N\) - glycan which impact the functionality, immunogenicity and pharmacokinetic of the antibody. C CryoEM maps of the EndoS \(E_{235A}\) - Fc complex, including a schematic representation of the EndoS domains and the Fc region in the complex structure. D Cartoon representation showing the overall fold and the secondary structure organization of the EndoS- Fc complex. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 95, 880, 582]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 603, 911, 678]]<|/det|> +
Fig. 2 The Fc binding site of EndoS. Two views of the interaction interface between the GH domain of EndoS (yellow) and FcP1 of the Fc region (orange) (A), the \(N\) -glycan (blue and green) of the Fc region and the active site of EndoS (yellow) (B) the Cy2-Cy3 joint region of the Fc (orange) and the b-sandwich domain of EndoS (blue) (C).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 75, 830, 644]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 670, 910, 796]]<|/det|> +
Fig. 3 Alanine scanning mutagenesis of the EndoS-Fc interface. A Structure of the EndoS (grey) complex with Fc (pink) in which the regions with alanine mutants are color coded based on the relative rates of removal of the \(N\) -glycans from Fc alanine mutants by EndoS and vice versa. The rates were measured via LC-MS kinetic analyses. Red indicates the largest increase in hydrolytic activity while blue indicates no hydrolytic activity. B Rates of removal of the first glycan of Fc and Fc alanine mutants by EndoS alanine mutants and EndoS respectively, normalized to EndoS vs. Fc. All rates were statistically significant (multiple comparison test, Tukey method, \(p< 0.0001\) ).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[111, 100, 870, 492]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 506, 910, 616]]<|/det|> +
Fig. 4 Molecular dynamics (MD) simulations studies. Overlay of 25 frames evenly spaced along the \(1\mu \mathrm{s}\) MD trajectory for (A) EndoS-Fc, (B) mutant E293A and (C) mutant W803A, along with the average distance between the center of the aromatic ring of the residues involved in the stabilizing interactions derived from the MD simulations. In the case of L314, S298 and E354 Cδ, the oxygen of -OH and the oxygens of the -CO₂⁻ are considered, respectively. Errors are given as SD. The EndoS protein is shown as blue ribbons and the antibody as gray and orange ribbons.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 68, 880, 602]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 620, 910, 797]]<|/det|> +
Fig. 5 Molecular mechanism of antibody recognition by EndoS and EndoS2. Kinetic modeling of deglycosylation of (A) FcWT and (B) IgGWT by EndoSWT and (C) IgGWT by EndoS2WT. LC-MS kinetic analysis of deglycosylation of FcWT and IgGWT by EndoSWT- showing the hydrolysis over time of the two \(N\) -glycans linked to Asn297. The deglycosylation process occurs sequentially as seen by the initial increase then decrease in monoglycosylated population (dark green) over time while the diglycosylated (purple) and deglycosylated (orange) species increase over time. D Schematic representation of Fc \(N\) -glycan processing by EndoS. In a first step, EndoS binds the diglycosylated Fc and hydrolyzes one of the \(N\) -glycan of the Fc which causes the release of monoglycosylated Fc. In a second step, EndoS engages the monoglycosylated Fc and hydrolyzes the remaining \(N\) -glycan of the Fc to produced completely deglycosylated Fc.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 68, 880, 691]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 704, 910, 916]]<|/det|> +
Fig. 6 Binding affinity and enzyme kinetics experiments studied by NMR. A Crystal structure of EndoS2 bound to CT \(N\) -glycan and \(\mathrm{Ca}^{2 + }\) (PDB code 6MDS) showing the location of the \(\left[^{13}\mathrm{C},\mathrm{H}_{3}\right]\) - methyl labeled probes as black spheres. Glycoside hydrolase (GH) and \(\beta\) -sandwich (BS) domains are colored in yellow and pale blue, respectively. Mutated amino acid E186L, CT glycan and \(\mathrm{Ca}^{2 + }\) ion are depicted in pink, dark blue and pale green. B Superimposition of selected regions of methyl-TROSY spectra showing CSPs upon mutagenesis or at saturation concentrations of different ligands. Color code corresponds to: EndoS2 (violet), EndoS2E186L in the absence (black) and in the presence of CT (red), \(\mathrm{CT}_{\mathrm{n - 1}}\) (orange), IgG1 Fc (blue) and \(\mathrm{Ca}^{2 + }\) (yellow). Addition of \(\mathrm{Fc}_{\mathrm{aglyc}}\) (green) induced to CSPs, suggesting very weak protein-protein interactions. Arrows indicate the extent of CSPs observed between the different protein states. C Ligands used for NMR titration experiments. IgG1 Fc region (Fc), complex-type carbohydrate (CT) and the products of glycan hydrolysis \(\mathrm{Fc}_{\mathrm{aglyc}}\) and \(\mathrm{CT}_{\mathrm{n - 1}}\) , respectively. Color code like in Fig. 1. D Quantum mechanics-based 2D lineshape fitting of
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 70, 910, 228]]<|/det|> +methyl- TROSY spectra of a representative signal for the titration of EndoS2E186L with Fc. Experimental and fitted spectra are shown as black and pink surfaces, respectively. Inserts correspond to 0, 5.5, 9.8, 20.3, 29.6 and \(56~\mu \mathrm{M}\) Fc concentrations. \(\mathbf{E}\) [S]0- dependence of the initial enzyme cycling velocity. The solid line represents the best- fit to Michaelis Menten Eq. 5, and dotted lines correspond to one standard deviation. \(\mathbf{F}\) Schematic mechanism of Fc \(N\) - glycan processing by EndoS2 as determined from lineshape analysis. The affinity of EndoS2 (here E) for Fc is two orders of magnitude higher as compared to the reaction product \(\mathrm{CT_{n - 1}}\) , promoting substrate depletion. On- rates are responsible for increased affinity towards Fc as compared to \(\mathrm{CT_{n - 1}}\) , suggesting differences in the initial enzyme- ligand encounter event. All experiments were acquired at \(600\mathrm{MHz}\) and \(298\mathrm{K}\) . + +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[88, 106, 909, 175]]<|/det|> +Table 1. Binding affinity of EndoS2 for IgG antibodies and N-glycans. Dissociation constants $K_{D}$ , on- and off-rate constants $k_{on}$ and $k_{off}$ for Rituximab-Fc, CT, $\text {CT}_{\text {n-1}}$ HM and $\text {Ca}^{2+}$ binding to EndoS2E186L, EndoS2 and EndoBT-3987D312A/E314L from fitting a two-state binding model to methyl-TROSY titration data using TITAN algorithm. + +<|ref|>table<|/ref|><|det|>[[100, 192, 844, 330]]<|/det|> + +
\(N^{\circ }\)EnzymeLigand\(K_{D}(\mu M)\)\(k_{off}(s^{-1})\)\(k_{on}(M^{-1}s^{-1})\)
1EndoS2E186LFc\(3.1\pm 0.6\)\(4.0\pm 1.5\)\(1.3\times 10^{6}\pm 5.5\times 10^{5}\)
2EndoBT-3987D312A/E314LHM\(1.7\pm 0.9\)\(5.5\pm 1.4\)\(3.2\times 10^{6}\pm 1.9\times 10^{5}\)
3EndoS2E186LCT\(106.9\pm 6.7\)\(11.4\pm 1.5\)\(1.1\times 10^{5}\pm 1.6\times 10^{4}\)
4EndoS2E186L\(\text {CT}_{\text {n-1}}\)\(150.7\pm 18.0\)\(9.4\pm 3.7\)\(6.2\times 10^{4}\pm 2.6\times 10^{4}\)
5EndoS2E186L\(\text {Fc}_{\text {aglyc}}\)---
6EndoS2E186L\(\text {CaCl}_{2}\)\(3415\pm 480\)--
7EndoS2\(\text {CaCl}_{2}\)\(4000\pm 100\)--
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 351, 230]]<|/det|> +2022TRASTOYFCNCSIDEF.pdfSupplementaryVideo1.movSupplementaryVideo2.movSupplementaryVideo3.mov + +<--- Page Split ---> diff --git a/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/images_list.json b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..0158699f5e9d33c9cf381728cf76f175779e269f --- /dev/null +++ b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/images_list.json @@ -0,0 +1,101 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 | Analysis pipeline for elucidation of GPCR activation mechanisms. Pipeline for analysis of universal and distinct activation macro-/microswitches spanning helix repacking to sidechain rotation and connection to ligand binding, G protein coupling and signal transduction sites (see Methods).", + "footnote": [], + "bbox": [ + [ + 226, + 312, + 770, + 655 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 | Transmembrane helix movement upon activation and universal TM3 and TM6 helix 'macroswitches'. a, Movements (A) over 1.0 Å at the extracellular end, membrane mid (determined using \\(^{35}\\) ) and intracellular end of the transmembrane helices, TM1-7 upon comparison of all available receptor inactive- and active-state structure pairs (Extended Data Table 1). Red intensity denotes the number of classes with a consensus movement (for class C, priority is given to the Gi-GABA \\(_{B2}\\) over the non-coupled mGluR5 structure). b, Movement and conserved hinges of TM6 and the adjacent TM5 and TM7. c, TM3 cytosolic tilt and overall rotation. b-c, GPCR class representative inactive/active receptor structure pairs: A: \\(\\beta_{2}^{36,37}\\) , B1: GLP-1 \\(^{18,22}\\) , C: GABA \\(_{B2}^{38}\\) and F: Smoothened \\(^{39,40}\\) receptors. Proline and glycine residues that increase helix plasticity are shown along with their percent conservation in the GPCR class.", + "footnote": [], + "bbox": [ + [ + 90, + 95, + 907, + 504 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Common and unique state determinants across the human GPCR superfamily", + "footnote": [], + "bbox": [], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_1.jpg", + "caption": "Extended Data Fig. 1 | TM1-7 movement of class A receptor pairs upon activation. Transmembrane helix movement at the extracellular face, membrane mid and intracellular face based on comparison of representative inactive and active state structure pairs of class A GPCRs (Extended Data Table 1).", + "footnote": [], + "bbox": [], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 44, + 42, + 652, + 777 + ] + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 55, + 50, + 800, + 777 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 58, + 123, + 520, + 844 + ] + ], + "page_idx": 26 + } +] \ No newline at end of file diff --git a/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068.mmd b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d465d5a1cf65e4eb963530e5b31f1d7a685c68ab --- /dev/null +++ b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068.mmd @@ -0,0 +1,506 @@ + +# GPCR activation mechanisms across classes and macro/microscales + +David Gloriam ( \(\boxed{ \begin{array}{r l} \end{array} }\) david.gloriam@sund.ku.dk) University of Copenhagen https://orcid.org/0000- 0002- 4299- 7561 + +Alexander Hauser University of Copenhagen + +Albert Kooistra University of Copenhagen https://orcid.org/0000- 0001- 5514- 6021 + +Christian Munk University of Copenhagen + +M. Madan Babu St. Jude Children's Research Hospital + +## Article + +Keywords: Endogenous Ligands, Extracellular Binding, Microswitch Residue Positions, Molecular Mechanistic Maps, Functional Determinants, Conformational Selection, Allosteric Communication + +Posted Date: April 1st, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 354879/v1 + +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Structural & Molecular Biology on November 10th, 2021. See the published version at https://doi.org/10.1038/s41594- 021- 00674- 7. + +<--- Page Split ---> + +# GPCR activation mechanisms + +# across classes and macro/microscales + +Alexander S. Hauser1,4, Albert J. Kooistra1,4, Christian Munk3, M. Madan Babu2 and David E. Gloriam1\* + +1Department of Drug Design and Pharmacology, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark. 2Department of Structural Biology and Center for Data Driven Discovery, St. Jude Children's Research Hospital, Memphis, TN, USA. 3Present address: Novozymes A/S, Biologiens Vej 2, 2800 Kongens Lyngby Copenhagen, Denmark 4These authors contributed equally *Correspondence: david.gloriam@sund.ku.dk (D.E.G.) + +Two- thirds of human hormones and one- third of clinical drugs activate \(\sim 350 \mathrm{G}\) protein- coupled receptors belonging to four classes: A, B1, C and F. Whereas a model of activation has been described for class A, very little is known about the activation of the other classes which differ by being activated by endogenous ligands bound mainly or entirely extracellularly. Here, we show that although they use the same structural scaffold and share several helix macroswitches, the GPCR classes differ in their microswitch residue positions and contacts. We present molecular mechanistic maps of activation for each GPCR class and new methods for contact analysis applicable for any functional determinants. This is the first superfamily residue- level rationale for conformational selection and allosteric communication by ligands and G proteins laying the foundation for receptor- function studies and drugs with the desired modality. + +<--- Page Split ---> + +G protein- coupled receptors (GPCRs) are nature's primary transmembrane transducers for carrying signals from extracellular ligands to intracellular effectors to regulate numerous physiological processes. The most widely used nomenclature designates GPCR classes A- F and was introduced in the first version of the GPCR database, GPCRdb1 based on conserved sequence fingerprints2,3. The human GPCRs have also been classified based on phylogenetic analysis into families (class): Glutamate (C), Rhodopsin (A), Adhesion (B2), Frizzled (F), Secretin (B1)4 and Taste 2 (T, re- classified as a separate family in5). The human GPCRs are activated from different endogenous ligand binding sites in the transmembrane (class A) or extracellular domain (class C) or both (class B and F) and have very low sequence similarity (cross- class pairs mean 23%). This raises the question to what extent the GPCR superfamily utilises universal or unique activation mechanism. Considering that GPCRs mediate the actions of two- thirds of endogenous hormones and neurotransmitters6 and over one- third of the drugs7-9, mapping their activation mechanisms is important to understand human physiology, disease aetiology and to rationally design drug. + +Comparison of inactive and active structures of class A GPCRs have uncovered common activation mechanisms within the seven transmembrane helices (7TM) that have been shown to tilt, rotate, elongate or switch residue sidechain rotamers to create contact networks stabilising the receptors in a specific state10- 13. These contacts converge near the G protein site14 and are conserved regardless of the subtypes of intracellular effectors11. Site- directed mutagenesis of contact residues has revealed differences in pharmacological responses, including constitutive activity11 and signalling bias15, and naturally occurring mutations have been associated with disease or change of response11,16,17. However, the classes B, C and F are largely unexplored with respect to common activation mechanisms and until now such studies have not been possible due to a lack of structures across the activation states. + +In this work, we conducted a comprehensive comparative structural analysis of inactive/active state structures in each GPCR class by analysing all available 488 structures from different GPCR classes (Fig. 1). We present the first, GPCR superfamily- wide molecular mechanistic maps of activation and link determinants to ligand binding, G protein coupling, transduction and allosteric sites. This can serve as a roadmap to assess the + +<--- Page Split ---> + +activation of any GPCR and to understand how common or distinct determinants can fine- tune physiological signalling responses and contribute to a desired drug efficacy. + +## Results + +## New dataset and methods for comparative structure analysis + +To present the best maps possible to date, we made a comprehensive annotation of all available 406 structures of 74 class A GPCRs (twice as many as in any previous report11). These were applied to stringent quality filters to select the representative inactive/active state structures from each class (Methods and Supplementary Spreadsheet 1). Out of 68 templates used, 45 have a resolution of 3.0 Å or better. It should be noted, that in lower resolution structures (>3.0Å) mainly backbone movements can be discerned while residue contacts can only be underrepresented (i.e., not lead to false positive contacts). The largest number of structures belong to classes A and B1 and cover 22 out of 58 and all five receptor families (share endogenous ligand), respectively. This diversity ensures that any artefacts observed in a specific position or structure will only have a marginal effect on the overall frequency of movements and contacts. + +We introduce a definition of state- specific contacts based on their relative frequency (%) in a given inactive/active state. Like the scores calculated in11, this allows for the identification of many more determinants than requiring 100% presence and absence, respectively in opposite states, such as the four inactive- and two active- state stabilising contacts identified in the first landmark study comparing five class A GPCRs across the states14. Importantly, the use of frequencies with (different) sets of inactive and active state structures opens up for the utilisation of also different receptors across the states, as the same helix backbone movements and contacts can be mediated by different amino acids. This is critical to the identification of a comparable number of state determinant residues in classes C and F for which the structural coverage is so far limited: four and three receptors, representing 18% and 27% of all members of these relatively small classes, respectively. Furthermore, we uniquely apply residue pair conservation cut- offs (Methods). Together, the combined contact + +<--- Page Split ---> + +frequencies and conservation cut- offs address the class- representativeness allowing analyses across the GPCR superfamily. + +# GPCR activation engages all transmembrane helices with unique contact patterns + +To investigate how the seven transmembrane helices, TM1- 7 rearrange upon activation, we compared the 13 GPCRs among our representative receptors for which both an inactive and active state structure are available (Extended Data Table 1 and Methods). We find that each transmembrane helix rearranges \(>1.0 \mathrm{\AA}\) in a majority of receptors - demonstrating unappreciated conformational dynamics (class consensus is red in Fig. 2, individual receptor plots in Supplementary Figs. 1- 2). Notably, the extracellular region alone exhibits a relocation of all seven transmembrane helices in class B1 wherein the N- terminal domain restricts the conformation the 7TM before activation \(^{18}\) . GLP- 1, which is the best template having a full- length inactive structure \(^{18}\) , has a \(2.5 - 10 \mathrm{\AA}\) relocation of TM1- 7 - all of which also move in CRF1. Furthermore, we analysed the conformational change of each class across the mid, extracellular and cytosolic regions where the rearrangements total \(6 \mathrm{\AA}\) , \(12 \mathrm{\AA}\) and \(28 \mathrm{\AA}\) and involving \(24\%\) , \(45\%\) and \(65\%\) of receptor helices. This is the first quantitated characterisation of magnitude and abundance of helix rearrangements at the mid, extracellular and cytosolic regions of the transmembrane domain. It supports opposite functional roles in i) maintaining a stable GPCR fold, ii) adapting to ligands of diverse sizes or iii) coupling to the much larger G proteins, respectively. This connects structure- function and reveals a wide engagement of the GPCR fold - across its helices and domains. + +We next investigated how the rearrangements of the transmembrane helix bundle change the residue contact networks between receptor segments by comparing all 43 inactive and 25 active state representative structures (Extended Data Table 1). We find that in addition to TM1- 7, state- specific contacts are also formed to helix 8 (H8) in classes A and F, intracellular loop 1 (ICL1) in class C and extracellular loop 2 (ECL2) in class F (Figs. 3- 4). Classes A, C and F have a majority of inactivating contacts (66%, 56% and 62%, respectively, Fig. 3a). Of note, each contact typically spans a majority and at least 30% of all receptors in each class based on residue- residue conservation (Methods). Class B1 has 48% inactivating contacts spanning segment contacts (excluding + +<--- Page Split ---> + +four intra- helix contacts). This characterisation of contacts by state – where the majority, or close to half for B1, of contacts restrain receptors into an inactive conformation – provides a structural rationale to why most receptors have no or little activity without the prior stimulation by an agonist. Furthermore, it demonstrates that any TM helices can switch from mainly inactive to active state contacts (Figure 3B). Notably, this rewiring leads to markedly different patterns of segment contacts across GPCR classes that combine several unique and some common contacts (below), as even high- homology receptors sharing endogenous ligand, such as the A₁ and A₂A adenosine or muscarinic acetylcholine M₁ and M₂ receptors, display unique helix movements (Fig. 2). + +and A₂A adenosine or muscarinic acetylcholine M₁ and M₂ receptors, display unique helix movements (Fig. 2). + +Together, the existence of both common and unique movements and contacts (below) provides a structural rationale for shared overall functions throughout the GPCR superfamily, e.g. ligand- dependent activation, signal transduction across the cell membrane and G protein coupling, that yet are diversified with respect to the specific ligand scaffold, G protein profile and functional response kinetics and efficacy. + +TM6 is a universal helix ‘macroswitch’ but toggles, rotates and unwinds differently across GPCR classes We next single helix rearrangements across 13 receptor inactive/active state structure pairs confirms. We find that outward movement and rotation of TM6 on the cytosolic side – opening for G protein coupling – is a universal feature of activation throughout the GPCR superfamily (Fig. 2a), as first suggested for class A₁⁹. Intracellular TM6 movement is observed in all 13 investigated receptors and is the largest in classes A (7- 13 Å), B1 (14- 19 Å) and F (7 Å) where is combined with a substantial average rotation: 38°, 39° and 38°, respectively. The opposite, extracellular end of TM6 also moves by on average 2.0, 3.4 and 6.2 Å, respectively, in these classes. The movement of TM6 gives it the largest (B1 and F) or second largest (A) number of contacts stabilising an inactive or active state in these classes (Fig. 3). However, our analysis shows that TM6 also displays large mechanistic differences across all the classes. Class B1 has a unique unwinding of the extracellular- facing half of TM6 in all three activated receptors²⁰- ²². This is in agreement with a recent study, comparing the class B1 Glucagon and class A β₂ receptors, linking the weaker ability of the agonist to induce the outward movement of cytoplasmic TM6 to a slower G protein activation²¹. Furthermore, in class C, TM6 + +<--- Page Split ---> + +uniquely has no extracellular movement and only a small (2.8 Å) movement and rotation (13°). Consequently, TM6 has only one state- specific contacts to other helices whereas TM5 has five and TM7, with two such contacts, is the only switch region with contacts across both states (Fig. 3b). The smaller role of TM6 in the activation throughout the class is supported by a markedly lower conservation of its proline, 55% compared to 93- 100% in other classes A- C while the importance of TM7 is emphasised by a 95% conserved proline (Fig. 2b). The non- canonical small TM6 tilt with a less conserved kink and atypical contacts to the adjacent TM5 and TM7 are associated with a different overall mechanism, as class C GPCRs work as an asymmetric homo- or hetero- dimer in which only one of two subunits couples to a G protein. + +These findings show that the GPCR activation machinery utilises TM6 as universal yet differentiated switches - combining helix toggling, rotation and unwinding. Such commonality across classes in the activation mechanism at the level of transmembrane helices but diversity and nuances at the type of movements provides an important structural rationale for drug discovery and in the future design of experiments to elucidate effects of mutations, ligand efficacy and G protein selectivity23 mechanisms. + +## TM5 is a new universal switch, TM5-7 act in concert and TM3 is a hub for stabilisation of inactive or active receptors + +TM5, like TM6, can be considered a universal switch for GPCR activation, as it moves on the intracellular side in all four classes (with on average A: 2.1, B1: 2.4, C: 2.0 and F: 1.8 Å), including a majority of class A and all class B1, C and F receptors (Fig. 2b and (Extended Data Fig. 2). TM7 also moves in all classes except class C, either on the cytosolic side (all class A receptors), the extracellular side (e.g. 10 Å movement and 100° rotation in GLP- 1, the class B1 receptor with full- length templates) or on both sides (class F). These movements are possible due to a number of conserved proline and glycine residues that induce helix plasticity (Fig. 2b). Class A GPCRs have triple proline kinks in TM5- 7, that allow TM5 and TM7 to close- in on and stabilise TM6. Class B1 TM6 unwinding is facilitated by both a proline kink (P6x47) and a glycine (G6x50). Class B1 GPCRs also feature a proline kink (P5x42) near the extracellular end of TM5, which moves 3.3 Å in GLP- 1, and a glycine kink (G7x50) in TM7. Class F combines the TM6 switch with movements of cytosolic + +<--- Page Split ---> + +TM5 (with one Pro and two Gly residues) and TM7 on both sides. This demonstrates that TM6 does not act on its own but is supported by TM5 and TM7 in a concerted movement and that the determinants of this plasticity are conserved throughout the classes. + +We find that TM3 contacts the largest number (3- 5) other receptor segments in each GPCR class (Fig. 3), revealing that TM3 functions as a stabilisation hub – not just to maintain the common transmembrane fold24 – but to stabilise a specific inactive or active state across the GPCR superfamily. Furthermore, whereas an early report based on two class A GPCRs suggested an activation mechanism involving an upward movement of TM325, our analysis of TM helix movements in 15 receptors across classes instead points to rotation as the main mechanism (Extended Data Fig. 1). In 11 out of 13 the investigated receptor pairs (all except A2A and CRF1), TM3 rotates at either the cytosolic (most frequent for class A, average 16°) or extracellular end (classes B1, C and F, average 19°, 12° and 17°, respectively) – while no receptor rotates in both ends or in the mid (Figs. S1- 2). Lateral movement of TM3 contributes to a different extent across classes spanning from both ends (B1), cytosolic end (C), either end for a minority of receptors (A) to no movement (F). In contrast, TM4 – which is peripherally located in the transmembrane helix bundle – has few or no contacts (class A:1, B1:0, C:3 and F:0). This shows that the abundant helix packing has not immobilised TM3 which instead, by an array of mainly local rotation or movement, contributes to GPCR activation in several places. + +Contact state frequency approach uncovers residue ‘microswitches’ in all GPCR classes and expands class A activation model + +To investigate state determinants on the residue ‘microswitch’ level, we indexed topologically corresponding receptor positions with generic residue numbers26 and classified them into ‘inactivators’, ‘activators’ and ‘switches’ based on frequent contacts in inactive, active and both states, respectively (Methods). We uncover a comparable number of state determinants across classes: A: 41, B1: 33, C: 28 and F: 33 (Fig. 4). 86% of determinants (A: 80%, B1: 72%, C: 100% and F: 91%) are inactivators or activators whereas only 16 determinants are switches. Only 9 switches undergo sidechain rotamer shifts (residue text rotation in Fig. 4a) revealing that that the rotamer microswitches – described as major state determinants for class A GPCRs12,13 + +<--- Page Split ---> + +- play a small role in the GPCR superfamily. Importantly, many determinant positions contain the same highly conserved amino acid (grey scaling in Fig. 4a) and the amino acid pairs observed for each contact are conserved in at least 40% of all receptors (Methods). Notably, this includes the reference positions for generic residue numbers (index x50) in 5 out of 7 transmembrane helices, helix 8 and ICL1 and all major previously known class A state determinants (sequence motifs and microswitches with magenta border in Fig. 4a)11,12. This demonstrates that our new approach (Discussion) identifies both known and new conserved determinants – even where the structural coverage is so far limited including the active state of classes C and F. + +In class A, W6x48, initially described as a rotamer toggle switch12 does not itself undergo a rotamer but a helix rotational shift (of \(10^{\circ}\) ) and is approached by I3x40 which has both types of rotations. This provides an alternative to a reported ‘PIF’ motif27- 29 which is here found to consist most frequently of ‘PIW’ (P5x50 being the third residue). The PIFs motifs last residue, F6x44 is instead here shown to act as a switch in another new triplet ‘LLF’ located on the same helices: TM3, TM5 and TM6. The two residues D2x50 and N7x49 which coordinate a sodium ion30 are here found to have a direct interaction stabilising the inactive state. Notably, N7x49 contacts P7x50 uncovering a concerted stabilisation across the sodium ion site and TM7 helix kink around the ‘NPxxY’ motif. The final residue of the NPxxY motif, a Y7x53 switch11- 13, has three inactivating and two activating state-specific contacts with over 40% frequency difference, including to another switch: I3x46. The TM3 ‘DRY’ motif includes an intra- helical ionic lock from D3x49 to R3x50, which upon activation swings to interact with the G protein as well as the switch Y5x58 on TM5. Of note, the restraining of R3x50 is strengthened by contacts to TM2 and TM6 (A6x34) – however typically not to E6x30, which was part of the first ionic lock reported in Rhodopsin31 but less conserved (E: 25% or D/E: 31% compared to D: 65% or D/E: 86% for position 4x49). On the intracellular side, helix 8 and ICL1 contain three and two determinants, respectively, including the novel switch F8x50 contacting another new switch on TM7, A7x54. These findings corroborate the findings of previous studies on class A10- 14. They also, together with the concerted movement of TM5- 7 and role of TM3 as a state stabilisation hub (above), substantially expand the activation model of class A receptors – the largest class of GPCRs in human. + +<--- Page Split ---> + +Class B1 shares many class A state determinant residues while having a unique concentration in and around the TM6 helix switch + +The first comparison of unique and common state determinants across the GPCR superfamily shows that nearly half, 16 out of 33, of the B1 determinants maps to equivalent topological positions as in class A compared to 10 for class F and 8 for class C (leftmost in Fig. 4b). Furthermore, the classes B1 and A have four common switches (second to rightmost in Fig. 4b). Notably, this includes the two known microswitches (A/B1 residue number): Y5x58/F5x54 and Y7x53/Y7x57 as well as the pair I3x46/E3x50 and F6x44/L6x49 (magenta in Extended Data Fig. 3). We also find that the class A 'toggle switch' activator W6x48 is also conserved in class B1 as a smaller aromatic residue Y6x53 (Extended Data Fig. 3). Together, these commonalities between classes B1 and A are indicative of an, in part, shared activation mechanism. + +However, there are also markedly unique features in class B1. Strikingly, 14 out of 33 of state determinants in class B1 are clustered on one helix, TM6 which engages 10 additional determinants – of which all except one are located in TM5 and TM7 (Fig. 4) which move together with TM6 (Fig. 2). The high concentration of state determinants in TM6 and packing to adjacent helices, may provide a plausible structural rationale for a recent report demonstrating a higher energy barrier for the formation of the kinked and partially unwound TM6 in the class B1 Glucagon receptor compared to the class A \(\beta_{2}\) - adrenoceptor21. Another unique feature in class B1 are two additional switches on TM7 (Q7x49 and L7x56). Class B1, like class A, has a large structural coverage (10 out of 15 receptors) and contains several major drug targets7. The first map of state determinants in class B1 gives a better understanding of the basic receptor activation mechanisms and presents a foundation for the targeting of determinant networks in structure- based design of new drugs stabilising receptors in desired states. + +## The GPCR superfamily lacks a universal residue-level microswitch mechanism + +Class C is atypical in that it contains as many inactivating and activating contacts and no switches in the transmembrane domain (compared to 3- 8 in the other classes, two- way Venn diagram in Fig. 4a). This is in concordance with its – by far – smallest conformational change (Fig. 2). Another characteristic of class C is + +<--- Page Split ---> + +that it has the highest share of determinants, \(36\%\) (compared to A: \(20\%\) , B1: \(24\%\) and F: \(0\%\) ) that form contacts within the same transmembrane helix. As class C is the only GPCR class that binds endogenous ligands entirely in the N- terminus, the transmembrane domain solely transduces signals. On the cytosolic side this involves two determinants in ICL1 (as in class A). Class F uniquely has frequent state stabilising contacts to ECL2, Y45x51 and V45x52 – both of which stabilise the active state by interaction with the same residue, F3x29 in TM3. The two ECL2 positions follow a conserved cysteine C45x50 which forms a covalent disulphide bridge, also to TM3, across the GPCR classes \(^{32}\) . Class F has the highest conservation of determinant consensus amino acids (on average \(90.5\%\) ) showing that the vast majority of contacts identified in the three structural templates: FZD4, FZD5 and Smoothened are likely shared by its remaining eight receptors. Class F has three switches, M3x37, W3x50 and M6x30 of which the former two have a large sidechain rotamer shift ( \(49^{\circ}\) and \(47^{\circ}\) , respectively). Interestingly, like class A, class F has a switch in position 3x50 although occupied by a tryptophan (A has arginine). + +By comparing all GPCR classes, we find that 57 out of 91 determinant positions reside in a distinct topological position and are therefore unique, whereas 25 are found in two classes, 8 in three classes and only 1 throughout the GPCR superfamily (leftmost Venn in Fig. 4b). Of note, the universal determinant position (numbered 3x46 in class A/F and 3x50 in class B1/C) plays a different role as inactivator (C), activator (F) or switch (A/B1). Considering such different roles in all determinant positions reveals an even smaller commonality – presence in a maximum of two GPCR classes for 12 inactivators (max 4 in A-F), 5 activators (max 2 in A-B1 and B1-F) and 5 switches (max 4 A-B1) (second to rightmost Venn in Fig. 4b and Extended Data Fig. 3). These findings demonstrate that although they belong to the same superfamily, use the same structural scaffold and share several macroswitches (helices), the GPCR classes differ in their microswitches (residues). This suggests that there are an ensemble of structural/mechanistic solutions that is available for the GPCR superfamily during evolution that different receptor classes have explored and utilised. It also means that while thinking about developing drugs with different modalities – especially for classes C and F – one should aim to interact with + +<--- Page Split ---> + +or modulate class specific state determinants and residue- level micro- switches rather than use the same state determinant and switch as in class A. + +State determinant mapping reveals where ligands and G proteins can sense and stabilise receptor states + +To obtain a topological and functional mapping of the state determinants, we mapped their location in relation to 408 ligand and 125 Gα protein interacting positions across the GPCR classes (grey and orange in Fig. 5) extracted from all 488 available GPCR structures. We find that all GPCR classes have the largest share, 60- 68% of state determinants in ligand binding positions (Supplementary Spreadsheet 2). Class B1 is the only GPCR class with an even share across orthosteric and allosteric ligand positions (above/below membrane "Mid" in hexagon residue on TM1- 7). Class A has only three (2%) determinants (W6x48, L7x40 and S7x46) in orthosteric ligand- binding positions and other state stabilisers are located below the membrane mid. Whereas class F has 18% of such determinants, class C lack orthosteric determinants altogether in the 7TM, as it binds its endogenous ligands solely in the N- terminus. When instead considering the proportion of state determinants in G protein binding positions, we find that these account for around one- third in classes A (29%) and B1 (36%) but a much smaller fraction in classes C and F (6- 7%). For class C this may reflect its uniquely shallow G protein binding but the figures in both classes is likely underrepresented due to few G protein complex structures. The remaining determinants (at non- interacting positions) account for 21- 36% on the GPCR class level. + +Given their important role in modulating GPCR activity, we made a specific mapping of switches i.e. determinant residues that alternate contacts across the inactive/active states. In class A, two switches (R3x50 and 5x58) have direct G protein interactions whereas the remaining six are located near or in between ligand and G protein positions facilitating the allosteric signal transduction across the extra- and intra- cellular sides. Class B1 switches are distributed across orthosteric ligand (6x49), G protein (3x50, 7x56 and 7x57) and two other (5x54 and 7x49) positions. Class C uniquely has no switches in the transmembrane domain. Finally, class F shares one class A switch with G protein interaction (3x50) and has two switches in positions without + +<--- Page Split ---> + +a known interaction. Together, this functional mapping shows that ligands and G proteins can sense and stabilise receptor states by direct interactions and that their signals are largely transduced via interconnecting pathways while state determinants including switches are spread throughout the 7TM bundle. This may help to explain observed effects on ligand or G protein affinity from remote mutations and presents a residue- level rationale for conformational selection33 and allosteric modulation34. + +## Discussion + +We present the first molecular mechanistic maps for activation across the GPCR superfamily and helix macro- /residue micro- scales while extending beyond the transmembrane region to helix 8 and structurally conserved loop segments. This study has answered the fundamental question of whether the activation of other classes can be modelled based on class A – they cannot. Our findings demonstrate that although they belong to the same superfamily, use the same structural scaffold and share several helix macroswitches, the GPCR classes differ in their microswitch residue positions, contacts and amino acids. This applies also to class B1 which shares about half of the determinant positions in class A, as the similarity is very small if considering the type of determinant – stabilising an inactive, active or both states – and their specific consensus amino acids. This highlights the need to elucidate each GPCR class’ restraints and diverse activation mechanisms separately and in more detail to adequately capture structure- function relationships. Determinants of receptor activity are tightly tied to the molecular mechanisms of conformational selection (Kenakin, 2019) and allosteric modulation34, and influence an array of functions and responses, including ligand affinity, basal activity, efficacy and G protein coupling. Therefore, the contact maps (Fig. 4) presented here provide an actionable foundation for the field to design and interpretate experiments across structural, biophysics (e.g. fluorescence and double electron- electron resonance), molecular dynamics and mutagenesis studies. The maps also inform drug discovery of which inactivating and activating state determinants are located in the orthosteric and allosteric ligand sites and may therefore facilitating effective design of inverse agonists, neutral antagonists or agonists for different receptors. + +<--- Page Split ---> + +The extensive engineering and limited resolution of some structures is an inherent limitation for all structure- based studies. Therefore, more structures and advances in their determination are of outmost value and will serve to continuously refine structure- function relationships generally. For example, the universal TM5 switch reported herein was not discernible until cryo- electron microscopy allowed for more structures with a native TM5- ICL3- TM6 region (often subjected to deletion and protein fusion in crystallography fusions not in cryo- EM). Our combined contact frequency and residue pair conservation cut- offs uniquely addresses the class- representativeness and allowed for the GPCR superfamily to be described. Class F has a \(90.5\%\) average conservation of determinant consensus amino acids. For these reasons, the common conserved determinants identified herein should still apply as our knowledge expands from new structures, which could also allow additional determinants fulfil these cut- offs. Furthermore, although specific receptors and subsets thereof could have additional activation mechanisms not conserved in the whole class, and such specific and general mechanisms, respectively could act in concert. + +## Acknowledgements + +This work was supported by the Lundbeck Foundation (grants R163- 2013- 16327 and R218- 2016- 1266), the Novo Nordisk Foundation (grant NNF18OC0031226) and Independent Research Fund Denmark | Natural Sciences (grant 8021- 00173B) to D.E.G and by the American Lebanese Syrian Associated Charities (ALSAC). + +## Author Contributions + +Conceptualization, D.E.G.; Methodology, D.E.G., C.M. and A.J.K.; Data Curation, A.J.K.; Investigation, D.E.G.; Validation: D.E.G. and A.J.K.; Writing – Original Draft, D.E.G.; Writing – Review & Editing, A.S.H. and M.M.B.; Visualization, D.E.G., A.S.H. and A.J.K., Funding Acquisition, D.E.G. and M.M.B.; Supervision, D.E.G. + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> + +## 1 Online content + +2 Methods, additional references, Nature Research reporting summaries, source data, extended data, peer review3 information; and statements of data and code availability are available at [URL]. + +## 6 Main figures + +![](images/Figure_1.jpg) + +
Fig. 1 | Analysis pipeline for elucidation of GPCR activation mechanisms. Pipeline for analysis of universal and distinct activation macro-/microswitches spanning helix repacking to sidechain rotation and connection to ligand binding, G protein coupling and signal transduction sites (see Methods).
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 | Transmembrane helix movement upon activation and universal TM3 and TM6 helix 'macroswitches'. a, Movements (A) over 1.0 Å at the extracellular end, membrane mid (determined using \(^{35}\) ) and intracellular end of the transmembrane helices, TM1-7 upon comparison of all available receptor inactive- and active-state structure pairs (Extended Data Table 1). Red intensity denotes the number of classes with a consensus movement (for class C, priority is given to the Gi-GABA \(_{B2}\) over the non-coupled mGluR5 structure). b, Movement and conserved hinges of TM6 and the adjacent TM5 and TM7. c, TM3 cytosolic tilt and overall rotation. b-c, GPCR class representative inactive/active receptor structure pairs: A: \(\beta_{2}^{36,37}\) , B1: GLP-1 \(^{18,22}\) , C: GABA \(_{B2}^{38}\) and F: Smoothened \(^{39,40}\) receptors. Proline and glycine residues that increase helix plasticity are shown along with their percent conservation in the GPCR class.
+ +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + + +<--- Page Split ---> + +Fig. 3 | GPCR stablise inactive and active states by rerouting contacts between TM1- 7, H8, ICL1 and ECL2. a, State- specific residue- residue contacts in each GPCR class visualised as lines within representative inactive/active receptor structure pairs (same as in Fig. 2b- c). Numbers indicate the total, inactivating and activating contacts in each GPCR class. b, Contact networks between the seven GPCR transmembrane helices, TM1- 7 and the first intracellular (ICL1) and second extracellular (ECL2) loops (intra- segment contacts not shown). Line thickness represents the number of classes (topmost) or contact frequency differences between the inactive and active states. Line colour indicates inactivating (blue) and activating (red) contacts. Receptor segments with a magenta border are “switches”, i.e. have contacts across both states. Contacts are identified based on a higher frequency (%) in one set of 43 inactive and 25 active state receptors. a- b, The frequency difference thresholds was set according to the structural coverage in each GPCR class (no. members and inactive/active state templates): A: 40% (285, 33/14), B1: 67% (15, 3/10), C: 75% (22, 4/1) and F: 100% (11, 3/1) – yielding comparable numbers of determinants and contacts. To ensure that the identified determinants are applicable for throughout each class, we applied a cut- off requiring at least 30% of all its receptors to contain one of the amino acid pairs observed to form the given state specific contact. Contact definitions are explained in the settings menu of the online ‘Comparative structure analysis’ tool (https://review.gpcrdb.org/structure_comparison/comparative_analysis). + +<--- Page Split ---> +![](images/Extended_Data_Figure_1.jpg) + +
Common and unique state determinants across the human GPCR superfamily
+ +![](images/Figure_2.jpg) + + +<--- Page Split ---> + +1 Fig. 4 | State stabilising contact maps and differences at the residue-level 'microswitches'. a, Contact networks 2 visualise the wiring of state determinants from the extracellular (top) to intracellular (bottom) sides. Contact frequency 3 differences between the inactive and active states are shown as varied line thickness and residue rotamers as rotation of 4 the consensus amino acid in analysed structures and its generic residue number. Two-way Venn diagrams depict the 5 number and percentages of inactivator (blue), activator (red) and switch (magenta) state determinant positions. Bar 6 diagrams show their distribution across the TM helices, H8 and loops. b, Comparison of common and unique state 7 determinant positions across all investigated classes. + +<--- Page Split ---> +![](images/Figure_3.jpg) + + +<--- Page Split ---> + +Fig. 5 | Residue positions stabilising an inactive and/or active receptor state. GPCR snakeplots mapping the residue positions that form distinct contacts between state determinants classified as inactivators (blue), activators (red) and switches (magenta). Residues are denoted with the consensus amino acid of the investigated receptor structures (Extended Data Table 1) and their generic residue number26. Filled positions map ligand (grey) and G protein (orange) interaction frequency among all GPCRs in the given class that have such data (Supplementary Spreadsheet 2). Border grayscale denotes the frequency of mutations changing the ligand affinity or activity over 5-fold. The label “Mid” within hexagonal-shaped positions denote the membrane mid, above and below which the ligand positions are considered orthosteric and allosteric, respectively, in class A, B1 and F. In class C, all ligand interacting positions in the seven transmembrane helices are allosteric, as its endogenous ligands bind solely in the N-terminal domain. + +## References + +Kolakowski, L. F., Jr. GCRDb: a G-protein-coupled receptor database. \*Recept Channels\* 2, 1- 71994). + +Attwood, T. K. & Findlay, J. B. Design of a discriminating fingerprint for G-protein-coupled receptors. \*Protein Eng.\* 6, 167- 176, doi:10.1093/protein/6.2.167 (1993). + +Attwood, T. K. & Findlay, J. B. Fingerprinting G-protein-coupled receptors. \*Protein Eng.\* 7, 195- 203, doi:10.1093/protein/7.2.195 (1994). + +Fredriksson, R., Lagerstrom, M. C., Lundin, L. G. & Schioth, H. B. The G-protein-coupled receptors in the human genome form five main families. Phylogenetic analysis, paralogon groups, and fingerprints. \*Molecular pharmacology\* 63, 1256- 1272, doi:10.1124/mol.63.6.1256 (2003). + +Nordstrom, K. J., Sallman Almen, M., Edstam, M. M., Fredriksson, R. & Schioth, H. B. Independent HHsearch, Needleman- Wunsch- based, and motif analyses reveal the overall hierarchy for most of the G protein- coupled receptor families. \*Mol Biol Evol\* 28, 2471- 2480, doi:10.1093/molbev/msr061 (2011). + +Foster, S. R. et al. Discovery of Human Signaling Systems: Pairing Peptides to G Protein- Coupled Receptors. \*Cell\* 179, 895- 908 e821, doi:10.1016/j.cell.2019.10.010 (2019). + +Hauser, A. S., Attwood, M. M., Rask- Andersen, M., Schioth, H. B. & Gloriam, D. E. Trends in GPCR drug discovery: new agents, targets and indications. \*Nat. Rev. Drug Discov.\* 16, 829- 842, doi:10.1038/nrd.2017.178 (2017). + +<--- Page Split ---> + +Sriram, K. & Insel, P. A. G Protein-Coupled Receptors as Targets for Approved Drugs: How Many Targets and How Many Drugs? Molecular pharmacology 93, 251- 258, doi:10.1124/mol.117.111062 (2018). + +Santos, R. et al. A comprehensive map of molecular drug targets. Nat. Rev. Drug Discov. 16, 19-34, doi:10.1038/nrd.2016.230 (2017). + +Flock, T. et al. Selectivity determinants of GPCR-G-protein binding. Nature 545, 317-322, doi:10.1038/nature22070 (2017). + +Zhou, Q. et al. Common activation mechanism of class A GPCRs. Elife 8, e50279, doi:10.7554/eLife.50279 (2019). + +Nygaard, R., Frimurer, T. M., Holst, B., Rosenkilde, M. M. & Schwartz, T. W. Ligand binding and micro-switches in 7TM receptor structures. Trends in pharmacological sciences 30, 249-259, doi:10.1016/j.tips.2009.02.006 (2009). + +Carpenter, B. & Tate, C. G. Active state structures of G protein-coupled receptors highlight the similarities and differences in the G protein and arrestin coupling interfaces. Current opinion in structural biology 45, 124-132, doi:10.1016/j.sbi.2017.04.010 (2017). + +Venkatakrishnan, A. J. et al. Diverse activation pathways in class A GPCRs converge near the G-protein-coupling region. Nature 536, 484-487, doi:10.1038/nature19107 (2016). + +Schonegge, A. M. et al. Evolutionary action and structural basis of the allosteric switch controlling beta2AR functional selectivity. Nat Commun 8, 2169, doi:10.1038/s41467-017-02257-x (2017). + +Hauser, A. S. et al. Pharmacogenomics of GPCR Drug Targets. Cell 172, 41-54 e19, doi:10.1016/j.cell.2017.11.033 (2018). + +Schoneberg, T. & Liebscher, I. Mutations in G Protein-Coupled Receptors: Mechanisms, Pathophysiology and Potential Therapeutic Approaches. Pharmacol. Rev. 73, 89-119, doi:10.1124/pharmrev.120.000011 (2021). + +Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). + +Schoneberg, T., Schultz, G. & Gudermann, T. Structural basis of G protein-coupled receptor function. Molecular and cellular endocrinology 151, 181-193, doi:10.1016/s0303-7207(99)00017-9 (1999). + +Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). + +<--- Page Split ---> + +1 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. 2 Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 3 22 Zhang, X. et al. Differential GLP- 1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 4 80, 485- 500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 5 23 Capper, M. J. & Wacker, D. How the ubiquitous GPCR receptor family selectively activates signalling pathways. 6 Nature 558, 529- 530, doi:10.1038/d41586- 018- 05503- 4 (2018). 7 24 Venkatakrishnan, A. J. et al. Molecular signatures of G- protein- coupled receptors. Nature 494, 185- 194, 8 doi:10.1038/nature11896 (2013). 9 25 Tehan, B. G., Bortolato, A., Blaney, F. E., Weir, M. P. & Mason, J. S. Unifying family A GPCR theories of 10 activation. Pharmacology & therapeutics 143, 51- 60, doi:10.1016/j.pharmthera.2014.02.004 (2014). 11 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in 12 pharmacological sciences 36, 22- 31, doi:10.1016/j.tips.2014.11.001 (2015). 13 Rasmussen, S. G. et al. Structure of a nanobody- stabilized active state of the beta(2) adrenoceptor. Nature 469, 14 175- 180, doi:10.1038/nature09648 (2011). 15 Wacker, D. et al. Structural features for functional selectivity at serotonin receptors. Science 340, 615- 619, 16 doi:10.1126/science.1232808 (2013). 17 Huang, W. et al. Structural insights into micro- opioid receptor activation. Nature 524, 315- 321, 18 doi:10.1038/nature14886 (2015). 19 White, K. L. et al. Structural Connection between Activation Microswitch and Allosteric Sodium Site in GPCR 20 Signaling. Structure 26, 259- 269 e255, doi:10.1016/j.str.2017.12.013 (2018). 21 Palczewski, K. et al. Crystal structure of rhodopsin: A G protein- coupled receptor. Science 289, 739- 745, 22 doi:10.1126/science.289.5480.739 (2000). 23 de Graaf, C., Foata, N., Engkvist, O. & Rognan, D. Molecular modeling of the second extracellular loop of G- 24 protein coupled receptors and its implication on structure- based virtual screening. Proteins 71, 599- 620, 25 doi:10.1002/prot.21724 (2008). 26 Kenakin, T. Biased Receptor Signaling in Drug Discovery. Pharmacol. Rev. 71, 267- 315, 27 doi:10.1124/pr.118.016790 (2019). 28 DeVree, B. T. et al. Allosteric coupling from G protein to the agonist- binding pocket in GPCRs. Nature 535, 29 182- 186, doi:10.1038/nature18324 (2016). + +<--- Page Split ---> + +1 35 Lomize, M. A., Pogozheva, I. D., Joo, H., Mosberg, H. I. & Lomize, A. L. OPM database and PPM web server: resources for positioning of proteins in membranes. Nucleic Acids Res. 40, D370-376, doi:10.1093/nar/gkr703 (2012). 4 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 6 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 8 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 9 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 12 40 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Extended Data Fig. 1 | TM1-7 movement of class A receptor pairs upon activation. Transmembrane helix movement at the extracellular face, membrane mid and intracellular face based on comparison of representative inactive and active state structure pairs of class A GPCRs (Extended Data Table 1).
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + + +<--- Page Split ---> + +
State determinantsConsensus AARepresentative receptors
AB1CFSumA B1C Fβ2GLP-1GABAB2SMO
TM11x501x472NTN51L159A496T245
1x531x502VTV54A162F499T248
TM22x402x472LF158L166N503F252
2x472x482NNN69N177P517A264
2x502x572ASA76F1841524N271
3x373x332DFD79F187G527F274
TM33x403x442QYV114Q234T561Y322
3x433x432LYI121Y241F568V329
3x493x452WLL124L244M571F332
3x463x503x504IEKV126V246A573V334
TM53x503x543RLWI127E247K574L335
5x505x502PLR131L251V578W339
5x515x475x513LIMP211A316L667V411
5x585x542YFL212I317M668L412
TM65x575x652IGY219F324A675L419
6x386x352LEV222V327T678G422
6x346x302AMA271L349D688T448
6x416x322SRL272A350S689M449
TM76x376x422LTT274S352Y691R451
6x386x432LLL275T3531692L452
6x406x456x363VLG276L354G693G453
6x416x462VII278L356S695F455
TM86x446x496x403FLYM279I357V696G456
6x516x482IIF282L360V699A459
6x476x526x493CHL284T362I701G461
6x486x532WYC285H363M702F462
TM97x427x412LKW286E364C703V463
7x407x447x433LFN312L388V724E518
7x457x442GMI314F390L726I520
7x497x392QLG315T391V727N521
TM107x527x567x553ILWN318Q394C731M525
7x537x572YYI325L401L738M532
7x547x487x573AFWY326Y402F740S533
C327C403V741T534
+ +1 + +2 + +3 **Extended Data Fig. 3 | State determinants conserved across classes.** Heatmap of state determinants shared by at least two GPCR classes along with their consensus amino acid and residues in representative receptors. Each row contains corresponding positions denoted with the generic residue numbers in each class26. Key class A state determinants are shown in bold. + +3 + +4 + +5 + +6 + +<--- Page Split ---> + +1 Extended Data Table 1 | Representative structures of GPCRs and inactive and active states used as templates. The + +2 structure annotation and criteria-based filtering to select these templates is shown in Supplementary Spreadsheet 1. + +
CIReceptor familyUniProtIUPHARInactive PDB IDActive PDB IDInactive Ref.Active Ref.
AAdrenoceptorsADA2Bα2B6K4141
ACannabinoidCNR1CB16N4B42
AChemokineCCR6CCR66WWZ43
AFormylpeptideFPR2FPR2/ALX6OMM44
ANeurotensinNTR1NTS16OS945
AAdrenoceptorsADRB2β22RH13SN63637
ADopamineDRD3D33PBL46
AHistamineHRH1H13RZE47
ALysophospholipid (S1P)S1PR1S1P13V2Y48
AOpioidOPRKκ4DJH49
AOpioidOPRMμ4DKL6DDE5051
AOpioidOPRDδ4N6H52
AAcetylcholine (muscarinic)ACM3M34U1553
ALysophospholipid (LPA)LPAR1LPA14Z3654
AAngiotensinAGTR1AT14ZUD55
AAcetylcholine (muscarinic)ACM1M15CXV6OIJ5657
AOpioidOPRXNOP5DHH58
AAcetylcholine (muscarinic)ACM4M45DSG56
AAdenosineAA2ARA2A5NM45G535960
AAdenosineAA1RA15UEN6D9H6162
ADopamineDRD4D45WIU63
AOrexinOX2ROX25WQC64
AEndothelinEDNRBETB5X9365
ANeuropeptide YNPY1RY15ZBQ66
AAcetylcholine (muscarinic)ACM2M25ZKC6OIK6757
ACannabinoidCNR2CB25ZTY6KPF6869
A5-Hydroxytryptamine5HT2A5-HT2A6A9470
A5-Hydroxytryptamine5HT2C5-HT2C6BQH71
AComplement peptideC5AR1C5a16C1R72
ADopamineDRD2D26CM473
ATachykininNK1RNK16HLP74
AGhrelinGHSRGhrelin6KO575
AAdrenoceptorsADA2Cα2C6KUW76
AChemokineCXCR2CXCR26LFL6LFO7777
AAcetylcholine (muscarinic)ACM5M56OL978
AMelatoninMTR1AMT16PS879
AOrexinOX1ROX16TOD80
AVasopressin and oxytocinOXYROT6TPK81
B1CalcitoninCALCRCT6NIY82
B1CalcitoninCALRLCalcitonin-like6UVA83
B1Corticotropin-releasing factorCRFR2CRF26PB184
B1GlucagonSCTRSecretin6WZG85
B1VIP and PACAPPACRPAC16M1I86
B1VIP and PACAPVIPR1VPAC16VN787
B1Corticotropin-releasing factorCRFR1CRF14K5Y6P9X8420
B1GlucagonGLRGlucagon5EE76WPW8821
B1Parathyroid hormonePTH1RPTH16NBH89
B1GlucagonGLP1RGLP-16LN26X181822
CMetabotropic glutamateGRM1mGlu14OR290
CMetabotropic glutamateGRM5mGlu56FFI91
CGABABGABR1GABAB17C7S38
CGABABGABR2GABAB27C7S7C7Q3838
FFrizzledSMOSMO4JKV6XBK3940
FFrizzledFZD4FZD46BD492
FFrizzledFZD5FZD56WW293
+ +<--- Page Split ---> + +# Supplemental Excel spreadsheets + +# Supplementary Spreadsheet 1. GPCR structure annotation to define activation states and representative templates + +Excel file containing an annotation of all 488 GPCR structures released in the Protein Data Bank94 before November 1st 2020. Representative structures of each GPCR and inactive or active state were selected by applying comprehensive filter criteria spanning completeness of receptor and G protein (≥86% and ≥43% of generic residue positions, respectively), sequence identity >90% to human, resolution ≤4.0 Å, degree active (≤20% and ≥90% for inactive and active structures, respectively) and consistent ligand modality and state (inverse agonist/antagonist-inactive and agonist-active). A continuous annotation is provided in the GPCRdb structure browser (http://review.gpcrdb.org/structure) and the structure comparison webserver (Extended Data Fig. 1). + +# Supplementary Spreadsheet 2. Topological and functional mapping of the state determinant positions + +Mapping of state determinant 309 ligand and 103 G protein interacting positions (grey and orange fill, respectively) extracted here from 423 GPCR crystal and cryo- EM structures. Each residue position has been assigned one main role defined by ligand and G protein interacting positions located within the intracellular and extracellular TM helix bundle pockets, respectively, and signal transduction pathway residue bridging such positions on the same helix in at least two classes (Methods). + +<--- Page Split ---> + +## Methods + +Structure annotation and selection of representative template dataset. We extended the GPCRdb structures (http://review.gpcrdb.org/structure, all 488 GPCR structures released in the Protein Data Bank95 before November 1st 2020) with additional annotation. We selected a representative structure of each GPCR and inactive or active state by applying comprehensive filter criteria spanning completeness of receptor and G protein (≥83% and ≥43% of generic residue positions, respectively), sequence identity >90% to human, resolution ≤3.7 Å, degree active96 (≤20% and ≥90% for inactive and active structures, respectively) and consistent ligand modality and state96 (inverse agonist/antagonist-inactive and agonist-active). For representative templates of the active state, a G protein complex was required. (Supplementary Spreadsheet 1) + +Transmembrane helix rearrangement analysis. 2D plots for TM1- 7 segment movement at the extracellular end, cytosolic end and membrane mid, respectively, were produced using our new webserver for comparative structure analysis (http://review.gpcrdb.org/structure_comparison/comparative_analysis and 96). Class counts of moving TM1- 7 were calculated as the sum of consensus movements above 1.0 Å (for class C, priority is given to the Gi- GABAB2 over the non- coupled mGluR5 structure). + +Segment (TM1- 7, H8 and loops) contact analysis. Activation- dependent changes in segment networks were determined using 2D network plots of segment contacts96. Segments switches (or helix switches) were defined as the segments with distinct contacts (see next section) in both states. + +Generic residue numbering. Corresponding residue positions in each class were indexed with the structure- based GPCRdb generic residue numbering system36. This builds on the sequence- based generic residue numbering systems for class A (Ballesteros- Weinstein), B1 (Wootten), C (Pin) and F (Wang), but preserves gaps from a structural alignment of two receptors, caused by a unique helix bulge or constriction, in the sequence alignment thereby avoiding offset of such and following residues. All schemes assign residue numbers relative to the most conserved amino acid residue, which is given the number 50, and prefixed with + +<--- Page Split ---> + +the TM helix number (e.g. 3x32 is on TM3 and 18 positions before the reference). This generic residue numbering scheme also uniquely indexes helix 8 and structurally conserved loop segments, which are numbered by the preceding and following TM helix (e.g. 45 is ECL2 located between TM4- 5). + +State stabilising contact identification. State stabilising contacts were identified using the Structure comparison tool96. The most distinct contact in class A (1x49- 7x50 contact) has 80% higher frequency in the inactive than active state, and the specificity of state- specific contacts depends on the number of members and templates in each class. To obtain comparable number of networks and contacts, we therefore adjusted the class frequency difference thresholds accordingly (no. members and inactive/active state templates); A: 40% (285, 33/14), B1: 67% (15, 3/10), C: 75% (22, 4/1) and F: 100% (11, 3/1). To ensure that the identified determinants are have a wide role in each class, we applied a sequence conservation cut- off. This cut- off requires the amino acids pairs that make up a state specific contact to be conserved, and therefore be able to form, in at least 30% of all receptors in the given GPCR class. The remaining residues, referred to as “state determinants” or “state stabilisers” were further classified into ‘inactivators’, ‘activators’ and ‘switches’ based on if their most frequent contact occurs in inactive, active and both states, respectively. + +State determinant topological mapping in relation to ligand and G protein sites. We mapped state stabilisers to functional sites to ligand and G protein sites using GPCRdb’s comparative structure analysis tool96. For positions with both an allosteric ligand and G protein interaction, precedence is given to the latter in the snakeplot data mapping (but both are counted in Fig. 5). Determinants located in positions with no known ligand or G protein interactions were categorised as “Other” in our analysis. + +## Data availability + +All data are available in GPCRdb (https://review.gpcrdb.org), GitHub (https://github.com/protwis/gpcrdb_data) and Supplementary Spreadsheets 1- 2. + +<--- Page Split ---> + +## Code availability + +Code availabilityNo code was developed for this manuscript which instead used the existing GPCRdb97 resources, including a new comparative structure analysis platform96. All open-source code can be obtained from GitHub (https://github.com/protovis/protovis) under the permissive Apache 2.0 License (https://www.apache.org/licenses/LICENSE- 2.0). A complete list of the software used for data analysis is available from the Nature Research Reporting Summary. + +## References + +Wu, F. et al. Full- length human GLP- 1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467- 020- 14934- 5 (2020). + +Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656- 668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). + +Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). + +Zhang, X. et al. Differential GLP- 1R Binding and Activation by Peptide and Non- peptide Agonists. Mol. Cell 80, 485- 500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). + +Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22- 31, doi:10.1016/j.tips.2014.11.001 (2015). + +Cherezov, V. et al. High- resolution crystal structure of an engineered human beta2- adrenergic G protein- coupled receptor. Science 318, 1258- 1265, doi:10.1126/science.1150577 (2007). + +Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor- Gs protein complex. Nature 477, 549- 555, doi:10.1038/nature10361 (2011). + +Mao, C. et al. Cryo- EM structures of inactive and active GABAB receptor. Cell Res. 30, 564- 573, doi:10.1038/s41422- 020- 0350- 5 (2020). + +Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338- 343, doi:10.1038/nature12167 (2013). + +Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368- 1375, doi:10.1038/s41589- 020- 0646- 2 (2020). + +<--- Page Split ---> + +1 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 2 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 3 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 4 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 5 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). 6 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 7 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, doi:10.1038/nature10236 (2011). 8 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, doi:10.1126/science.1215904 (2012). 9 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, doi:10.1038/nature10939 (2012). 10 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 11 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, doi:10.1038/s41586-018-0219-7 (2018). 12 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, doi:10.1038/nature12944 (2014). 13 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 14 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). + +<--- Page Split ---> + +Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. The Journal of biological chemistry 290, 29127- 29139, doi:10.1074/jbc.M115.689000 (2015). + +Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335- 340, doi:10.1038/nature17188 (2016). + +Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic acetylcholine receptor/G-protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). + +Miller, R. L. et al. The Importance of Ligand- Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein- Coupled Receptor. Structure 23, 2291- 2299, doi:10.1016/j.str.2015.07.024 (2015). + +Weinert, T. et al. Serial millisecond crystallography for routine room- temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467- 017- 00630- 4 (2017). + +Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). + +Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). + +Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). + +Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). + +Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). + +Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). + +Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). + +Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). + +Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). + +<--- Page Split ---> + +1 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor-Gi Complex Structures. Cell 180, 655-665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 3 70 Kimura, K. T. et al. Structures of the 5-HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121-128, doi:10.1038/s41594-018-0180-z (2019). 5 71 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 7 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472-481, doi:10.1038/s41594-018-0067-z (2018). 9 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269-273, doi:10.1038/nature25758 (2018). 10 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467-018-07939-8 (2019). 11 75 Shiimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467-020-17554-1 (2020). 12 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936-2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 13 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135-140, doi:10.1038/s41586-020-2492-5 (2020). 14 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001-26007, doi:10.1073/pnas.1914446116 (2019). 15 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure-based drug design using X-ray lasers. IUCrJ 6, 1106-1119, doi:10.1107/S2052252519013137 (2019). 16 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X-ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528-1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 17 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 18 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31-51, doi:10.1021/acsptsci.8b00056 (2019). + +<--- Page Split ---> + +1 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key 2 Mechanisms in the Control of Receptor Phenotype by Receptor Activity-Modifying Proteins. ACS Pharmacol 3 Transl Sci 3, 263-284, doi:10.1021/acsptsci.9b00080 (2020). 4 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438-443, 5 doi:10.1038/nature12357 (2013). 6 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 7 4137, doi:10.1038/s41467-020-17791-4 (2020). 8 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, 9 doi:10.1038/s41422-020-0280-2 (2020). 10 87 Duan, J. et al. Cryo-EM structure of an activated VIP1 receptor-G protein complex revealed by a NanoBiT 11 tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467-020-17933-8 (2020). 12 88 Jazayeri, A. et al. Extra-helical binding site of a glucagon receptor antagonist. Nature 533, 274-277, 13 doi:10.1038/nature17414 (2016). 14 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor-1. Science 364, 15 148-153, doi:10.1126/science.aav7942 (2019). 16 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. 17 Science 344, 58-64, doi:10.1126/science.1249489 (2014). 18 91 Christopher, J. A. et al. Structure-Based Optimization Strategies for G Protein-Coupled Receptor (GPCR) 19 Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- 20 ray Structures. Journal of medicinal chemistry 62, 207-222, doi:10.1021/acs.jmedchem.7b01722 (2019). 21 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand-free state. Nature 560, 666-670, 22 doi:10.1038/s41586-018-0447-x (2018). 23 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial-assisted cryo-EM supports a heterodimeric 24 mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 25 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and 26 education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464-D474, 27 doi:10.1093/nar/gky1004 (2019). 28 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids 29 Res. 48, D335-D343, doi:10.1093/nar/gkz990 (2020). + +<--- Page Split ---> + +1 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online GPCR structure analysis platform. Submitted. 3 97 Kooistra, A. J. et al. GPCRdb in 2021: integrating GPCR sequence, structure and function. Nucleic Acids Res. 49, D335-D343, doi:10.1093/nar/gkaa1080 (2021). 5 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). 7 20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 9 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 11 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 13 26 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 15 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 17 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549- 555, doi:10.1038/nature10361 (2011). 19 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 21 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338- 343, doi:10.1038/nature12167 (2013). 23 40 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 25 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 27 42 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448- 458 e412, doi:10.1016/j.cell.2018.11.040 (2019). + +<--- Page Split ---> + +43 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 44 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 45 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, doi:10.1038/nature10236 (2011). 48 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, doi:10.1126/science.1215904 (2012). 49 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, doi:10.1038/nature10939 (2012). 50 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 51 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, doi:10.1038/s41586-018-0219-7 (2018). 52 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, doi:10.1038/nature12944 (2014). 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. The Journal of biological chemistry 290, 29127-29139, doi:10.1074/jbc.M115.689000 (2015). 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335-340, doi:10.1038/nature17188 (2016). + +<--- Page Split ---> + +1 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic acetylcholine receptor/G-protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). 2 58 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein-Coupled Receptor. Structure 23, 2291- 2299, doi:10.1016/j.str.2015.07.024 (2015). 3 59 Weinert, T. et al. Serial millisecond crystallography for routine room- temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467- 017- 00630- 4 (2017). 4 60 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). 5 61 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 6 62 Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). 7 63 Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). 8 64 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). 9 65 Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). 10 66 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). 11 67 Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). 12 68 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 13 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor- Gi Complex Structures. Cell 180, 655- 665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 14 70 Kimura, K. T. et al. Structures of the 5- HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121- 128, doi:10.1038/s41594- 018- 0180- z (2019). + +<--- Page Split ---> + +1 71 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 3 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 5 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). 6 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467- 018- 07939- 8 (2019). 9 75 Shiimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467- 020- 17554- 1 (2020). 10 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936- 2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 11 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135- 140, doi:10.1038/s41586- 020- 2492- 5 (2020). 12 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001- 26007, doi:10.1073/pnas.1914446116 (2019). 13 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure- based drug design using X- ray lasers. IUcrJ 6, 1106- 1119, doi:10.1107/S2052252519013137 (2019). 14 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X- ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528- 1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 15 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 16 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31- 51, doi:10.1021/acsptsci.8b00056 (2019). 17 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity- Modifying Proteins. ACS Pharmacol Transl Sci 3, 263- 284, doi:10.1021/acstpsci.9b00080 (2020). + +<--- Page Split ---> + +84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438- 443, doi:10.1038/nature12357 (2013). 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467- 020- 17791- 4 (2020). 86 Wang, J. et al. Cryo- EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436- 445, doi:10.1038/s41422- 020- 0280- 2 (2020). 87 Duan, J. et al. Cryo- EM structure of an activated VIP1 receptor- G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467- 020- 17933- 8 (2020). 88 Jazayeri, A. et al. Extra- helical binding site of a glucagon receptor antagonist. Nature 533, 274- 277, doi:10.1038/nature17414 (2016). 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor- 1. Science 364, 148- 153, doi:10.1126/science.aav7942 (2019). 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58- 64, doi:10.1126/science.1249489 (2014). 91 Christopher, J. A. et al. Structure- Based Optimization Strategies for G Protein- Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207- 222, doi:10.1021/acs.jmedchem.7b01722 (2019). 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand- free state. Nature 560, 666- 670, doi:10.1038/s41586- 018- 0447- x (2018). 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial- assisted cryo- EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464- D474, doi:10.1093/nar/gky1004 (2019). 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335- D343, doi:10.1093/nar/gkz990 (2020). 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online platform for comparative GPCR structure analysis. Submitted. + +<--- Page Split ---> + +97 Kooistra, A. J. et al. GPCRdb in 2021: integrating GPCR sequence, structure and function. Nucleic Acids Res. 49, D335-D343, doi:10.1093/nar/gkaa1080 (2021). 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). 20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 26 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 40 Qi, X., Friedberg, L., De Bose-Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 42 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 43 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). + +<--- Page Split ---> + +1 44 Zhuang, Y. et al. Structure of formylpeptide receptor 2- Gi complex reveals insights into ligand recognition and 2 signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 3 45 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1- Gi1 complex. Nature 572, 80- 85, 4 doi:10.1038/s41586- 019- 1337- 6 (2019). 5 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. 6 Science 330, 1091- 1095, doi:10.1126/science.1197410 (2010). 7 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65- 70, 8 doi:10.1038/nature10236 (2011). 9 48 Hanson, M. A. et al. Crystal structure of a lipid G protein- coupled receptor. Science 335, 851- 855, 10 doi:10.1126/science.1215904 (2012). 11 49 Wu, H. et al. Structure of the human kappa- opioid receptor in complex with JDTic. Nature 485, 327- 332, 12 doi:10.1038/nature10939 (2012). 13 50 Manglik, A. et al. Crystal structure of the micro- opioid receptor bound to a morphinan antagonist. Nature 485, 14 321- 326, doi:10.1038/nature10954 (2012). 15 51 Koehl, A. et al. Structure of the micro- opioid receptor- Gi protein complex. Nature 558, 547- 552, 16 doi:10.1038/s41586- 018- 0219- 7 (2018). 17 52 Fenalti, G. et al. Molecular control of delta- opioid receptor signalling. Nature 506, 191- 196, 18 doi:10.1038/nature12944 (2014). 19 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G 20 Protein- Coupled Receptor Crystallogenesis. Structure 22, 1657- 1664, doi:10.1016/j.str.2014.08.022 (2014). 21 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 22 161, 1633- 1643, doi:10.1016/j.cell.2015.06.002 (2015). 23 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. 24 The Journal of biological chemistry 290, 29127- 29139, doi:10.1074/jbc.M115.689000 (2015). 25 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335- 340, 26 doi:10.1038/nature17188 (2016). 27 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic 28 acetylcholine receptor/G- protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). + +<--- Page Split ---> + +1 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein-Coupled Receptor. Structure 23, 2291-2299, doi:10.1016/j.str.2015.07.024 (2015). 2 Weinert, T. et al. Serial millisecond crystallography for routine room-temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467-017-00630-4 (2017). 3 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104-107, doi:10.1038/nature18966 (2016). 4 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867-877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 5 Draper-Joyce, C. J. et al. Structure of the adenosine-bound human adenosine A1 receptor-Gi complex. Nature 558, 559-563, doi:10.1038/s41586-018-0236-6 (2018). 6 Wang, S. et al. D4 dopamine receptor high-resolution structures enable the discovery of selective agonists. Science 358, 381-386, doi:10.1126/science.aan5468 (2017). 7 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype-Selective Antagonist EMPA. Structure 26, 7-19 e15, doi:10.1016/j.str.2017.11.005 (2018). 8 Shihoya, W. et al. X-ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758-764, doi:10.1038/nsmb.3450 (2017). 9 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520-524, doi:10.1038/s41586-018-0046-x (2018). 10 Suno, R. et al. Structural insights into the subtype-selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150-1158, doi:10.1038/s41589-018-0152-y (2018). 11 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459-467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 12 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor-Gi Complex Structures. Cell 180, 655-665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 13 Kimura, K. T. et al. Structures of the 5-HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121-128, doi:10.1038/s41594-018-0180-z (2019). 14 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). + +<--- Page Split ---> + +72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467- 018- 07939- 8 (2019). 75 Shiimura, Y. et al. Structure of an antagonist- bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467- 020- 17554- 1 (2020). 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936- 2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135- 140, doi:10.1038/s41586- 020- 2492- 5 (2020). 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001- 26007, doi:10.1073/pnas.1914446116 (2019). 79 Ishchenko, A. et al. Toward G protein- coupled receptor structure- based drug design using X- ray lasers. IUCrJ 6, 1106- 1119, doi:10.1107/S2052252519013137 (2019). 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X- ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528- 1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31- 51, doi:10.1021/acsptsci.8b00056 (2019). 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity- Modifying Proteins. ACS Pharmacol Transl Sci 3, 263- 284, doi:10.1021/acsptsci.9b00080 (2020). 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin- releasing factor receptor 1. Nature 499, 438- 443, doi:10.1038/nature12357 (2013). + +<--- Page Split ---> + +1 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467-020-17791-4 (2020). 3 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, doi:10.1038/s41422-020-0280-2 (2020). 5 87 Duan, J. et al. Cryo-EM structure of an activated VIP1 receptor-G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467-020-17933-8 (2020). 7 88 Jazayeri, A. et al. Extra-helical binding site of a glucagon receptor antagonist. Nature 533, 274-277, doi:10.1038/nature17414 (2016). 9 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor-1. Science 364, 148-153, doi:10.1126/science.aav7942 (2019). 10 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58-64, doi:10.1126/science.1249489 (2014). 13 91 Christopher, J. A. et al. Structure-Based Optimization Strategies for G Protein-Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207-222, doi:10.1021/acs.jmedchem.7b01722 (2019). 16 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand-free state. Nature 560, 666-670, doi:10.1038/s41586-018-0447-x (2018). 18 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial-assisted cryo-EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 20 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464-D474, doi:10.1093/nar/gky1004 (2019). 23 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335-D343, doi:10.1093/nar/gkz990 (2020). 25 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online resource for comparative GPCR structure analysis. Submitted. 27 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). + +<--- Page Split ---> + +20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 23 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 24 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 25 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 26 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 27 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 28 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 29 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 30 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 31 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 32 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 33 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). + +<--- Page Split ---> + +1 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. 2 Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 3 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, 4 doi:10.1038/nature10236 (2011). 5 48 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, 6 doi:10.1126/science.1215904 (2012). 7 49 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, 8 doi:10.1038/nature10939 (2012). 9 50 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 11 51 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, 12 doi:10.1038/s41586-018-0219-7 (2018). 13 52 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, 14 doi:10.1038/nature12944 (2014). 15 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G 16 Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 17 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 18 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). 19 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. 20 The Journal of biological chemistry 290, 29127-29139, doi:10.1074/jbc.M115.689000 (2015). 21 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335-340, 22 doi:10.1038/nature17188 (2016). 23 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic 24 acetylcholine receptor/G-protein complexes. Science 364, 552-557, doi:10.1126/science.aaw5188 (2019). 25 58 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the 26 N/OFQ G Protein-Coupled Receptor. Structure 23, 2291-2299, doi:10.1016/j.str.2015.07.024 (2015). 27 59 Weinert, T. et al. Serial millisecond crystallography for routine room-temperature structure determination at 28 synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467-017-00630-4 (2017). + +<--- Page Split ---> + +1 60 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). 3 61 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 5 62 Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). 7 63 Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). 9 64 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). 11 65 Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). 13 66 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). 15 67 Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). 17 68 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 19 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor- Gi Complex Structures. Cell 180, 655- 665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 21 70 Kimura, K. T. et al. Structures of the 5- HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121- 128, doi:10.1038/s41594- 018- 0180- z (2019). 23 71 Peng, Y. et al. 5- HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719- 730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 25 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 27 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). + +<--- Page Split ---> + +1 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467-018-07939-8 (2019). 2 75 Shimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467-020-17554-1 (2020). 3 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936-2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 4 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135-140, doi:10.1038/s41586-020-2492-5 (2020). 5 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001-26007, doi:10.1073/pnas.1914446116 (2019). 6 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure-based drug design using X-ray lasers. IUCrJ 6, 1106-1119, doi:10.1107/S2052252519013137 (2019). 7 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X-ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528-1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 8 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 9 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31-51, doi:10.1021/acsptsci.8b00056 (2019). 10 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity-Modifying Proteins. ACS Pharmacol Transl Sci 3, 263-284, doi:10.1021/acsptsci.9b00080 (2020). 11 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438-443, doi:10.1038/nature12357 (2013). 12 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467-020-17791-4 (2020). 13 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, doi:10.1038/s41422-020-0280-2 (2020). + +<--- Page Split ---> + +1 87 Duan, J. et al. Cryo- EM structure of an activated VIP1 receptor- G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467- 020- 17933- 8 (2020). 2 88 Jazayeri, A. et al. Extra- helical binding site of a glucagon receptor antagonist. Nature 533, 274- 277, doi:10.1038/nature17414 (2016). 3 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor- 1. Science 364, 148- 153, doi:10.1126/science.aav7942 (2019). 4 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58- 64, doi:10.1126/science.1249489 (2014). 5 91 Christopher, J. A. et al. Structure- Based Optimization Strategies for G Protein- Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207- 222, doi:10.1021/acs.jmedchem.7b01722 (2019). 6 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand- free state. Nature 560, 666- 670, doi:10.1038/s41586- 018- 0447- x (2018). 7 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial- assisted cryo- EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 8 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464- D474, doi:10.1093/nar/gky1004 (2019). 9 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335- D343, doi:10.1093/nar/gkz990 (2020). 0 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online resource for comparative GPCR structure analysis. Submitted. + +<--- Page Split ---> + +## Figures + +![PLACEHOLDER_52_0] + + +Activation engages all transmembrane helices, helix 8 and loops TM6 and TM3 are universal helix switches 1- 7 residue switches are pivotal changing the state More determinants arise by helix movement or rotation than sidechain rotamer shift Web resource for comparative structure analysis allow uncovering of determinants + +## Figure 1 + +Analysis pipeline for elucidation of GPCR activation mechanisms. Pipeline for analysis of universal and distinct activation macro-/microswitches spanning helix repacking to sidechain rotation and connection to ligand binding, G protein coupling and signal transduction sites (see Methods). + +<--- Page Split ---> +![PLACEHOLDER_53_0] + +
Figure 2
+ +Transmembrane helix movement upon activation and universal TM3 and TM6 helix 'macroswitches'. a, Movements (A) over 1.0 Å at the extracellular end, membrane mid (determined using35) and intracellular end of the transmembrane helices, TM1- 7 upon comparison of all available receptor inactive- and active- state structure pairs (Extended Data Table 1). Red intensity denotes the number of classes with a consensus movement (for class C, priority is given to the Gi- GABAB2 over the non- coupled mGluR5 structure). b, Movement and conserved hinges of TM6 and the adjacent TM5 and TM7. c, TM3 cytosolic tilt and overall rotation. b- c, GPCR class representative inactive/active receptor structure pairs: A: β236,37, B1: GLP- 118,22, C: GABAB238 and F: Smoothened39,40 receptors. Proline and glycine residues that increase helix plasticity are shown along with their percent conservation in the GPCR class. + +<--- Page Split ---> +![PLACEHOLDER_54_0] + +
Figure 3
+ +GPCR stabilise inactive and active states by rerouting contacts between TM1- 7, H8, ICL1 and ECL2. a, State- specific residue- residue contacts in each GPCR class visualised as lines within representative inactive/active receptor structure pairs (same as in Fig. 2b- c). Numbers indicate the total, inactivating and activating contacts in each GPCR class. b, Contact networks between the seven GPCR transmembrane helices, TM1- 7 and the first intracellular (ICL1) and second extracellular (ECL2) loops (intra- segment + +<--- Page Split ---> + +contacts not shown). Line thickness represents the number of classes (topmost) or contact frequency differences between the inactive and active states. Line colour indicates inactivating (blue) and activating (red) contacts. Receptor segments with a magenta border are "switches", i.e. have contacts across both states. Contacts are identified based on a higher frequency (%) in one set of 43 inactive and 25 active state receptors. a- b, The frequency difference thresholds was set according to the structural coverage in each GPCR class' (no. members and inactive/active state templates): A: \(40\%\) (285, 33/14), B1: \(67\%\) (15, 3/10), C: \(75\%\) (22, 4/1) and F: \(100\%\) (11, 3/1) – yielding comparable numbers of determinants and contacts. To ensure that the identified determinants are applicable for throughout each class, we applied a cut- off requiring at least \(30\%\) of all its receptors to contain one of the amino acid pairs observed to form the given state specific contact. Contact definitions are explained in the settings menu of the online 'Comparative structure analysis' tool (https://review.gpcrdb.org/structure_comparison/comparative_analysis). + +<--- Page Split ---> +![PLACEHOLDER_56_0] + +
Figure 4
+ +State stabilising contact maps and differences at the residue- level 'microswitches'. a, Contact networks visualise the wiring of state determinants from the extracellular (top) to intracellular (bottom) sides. Contact frequency differences between the inactive and active states are shown as varied line thickness and residue rotamers as rotation of the consensus amino acid in analysed structures and its generic residue number. Two- way Venn diagrams depict the number and percentages of inactivator (blue), + +<--- Page Split ---> + +activator (red) and switch (magenta) state determinant positions. Bar diagrams show their distribution across the TM helices, H8 and loops. b, Comparison of common and unique state determinant positions across all investigated classes. + +![PLACEHOLDER_57_0] + +
Figure 5
+ +Residue positions stabilising an inactive and/or active receptor state. GPCR snakeplots mapping the residue positions that form distinct contacts between state determinants classified as inactivators (blue), + +<--- Page Split ---> + +activators (red) and switches (magenta). Residues are denoted with the consensus amino acid of the investigated receptor structures (Extended Data Table 1) and their generic residue number26. Filled positions map ligand (grey) and G protein (orange) interaction frequency among all GPCRs in the given class that have such data (Supplementary Spreadsheet 2). Border grayscale denotes the frequency of mutations changing the ligand affinity or activity over 5- fold. The label "Mid" within hexagon- shaped positions denote the membrane mid, above and below which the ligand positions are considered orthosteric and allosteric, respectively, in class A, B1 and F. In class C, all ligand interacting positions in the seven transmembrane helices are allosteric, as its endogenous ligands bind solely in the N- terminal domain. + +<--- Page Split ---> diff --git a/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068_det.mmd b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..fef06caf15ead13fabc45c3f0c9854fbaa53d725 --- /dev/null +++ b/preprint/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068/preprint__13bf6458fcf02e570f84c76bf84696729113a9f983cbf367b27e04ad99d1e068_det.mmd @@ -0,0 +1,626 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 914, 175]]<|/det|> +# GPCR activation mechanisms across classes and macro/microscales + +<|ref|>text<|/ref|><|det|>[[44, 195, 636, 238]]<|/det|> +David Gloriam ( \(\boxed{ \begin{array}{r l} \end{array} }\) david.gloriam@sund.ku.dk) University of Copenhagen https://orcid.org/0000- 0002- 4299- 7561 + +<|ref|>text<|/ref|><|det|>[[44, 243, 280, 283]]<|/det|> +Alexander Hauser University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[44, 290, 636, 333]]<|/det|> +Albert Kooistra University of Copenhagen https://orcid.org/0000- 0001- 5514- 6021 + +<|ref|>text<|/ref|><|det|>[[44, 337, 280, 377]]<|/det|> +Christian Munk University of Copenhagen + +<|ref|>text<|/ref|><|det|>[[44, 383, 380, 424]]<|/det|> +M. Madan Babu St. Jude Children's Research Hospital + +<|ref|>sub_title<|/ref|><|det|>[[44, 464, 102, 481]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 501, 890, 545]]<|/det|> +Keywords: Endogenous Ligands, Extracellular Binding, Microswitch Residue Positions, Molecular Mechanistic Maps, Functional Determinants, Conformational Selection, Allosteric Communication + +<|ref|>text<|/ref|><|det|>[[44, 562, 284, 581]]<|/det|> +Posted Date: April 1st, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 600, 463, 619]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 354879/v1 + +<|ref|>text<|/ref|><|det|>[[44, 637, 909, 680]]<|/det|> +License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 717, 930, 760]]<|/det|> +Version of Record: A version of this preprint was published at Nature Structural & Molecular Biology on November 10th, 2021. See the published version at https://doi.org/10.1038/s41594- 021- 00674- 7. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[259, 102, 737, 130]]<|/det|> +# GPCR activation mechanisms + +<|ref|>title<|/ref|><|det|>[[203, 163, 792, 191]]<|/det|> +# across classes and macro/microscales + +<|ref|>text<|/ref|><|det|>[[110, 323, 886, 341]]<|/det|> +Alexander S. Hauser1,4, Albert J. Kooistra1,4, Christian Munk3, M. Madan Babu2 and David E. Gloriam1\* + +<|ref|>text<|/ref|><|det|>[[90, 384, 905, 500]]<|/det|> +1Department of Drug Design and Pharmacology, University of Copenhagen, Universitetsparken 2, 2100 Copenhagen, Denmark. 2Department of Structural Biology and Center for Data Driven Discovery, St. Jude Children's Research Hospital, Memphis, TN, USA. 3Present address: Novozymes A/S, Biologiens Vej 2, 2800 Kongens Lyngby Copenhagen, Denmark 4These authors contributed equally *Correspondence: david.gloriam@sund.ku.dk (D.E.G.) + +<|ref|>text<|/ref|><|det|>[[89, 567, 908, 858]]<|/det|> +Two- thirds of human hormones and one- third of clinical drugs activate \(\sim 350 \mathrm{G}\) protein- coupled receptors belonging to four classes: A, B1, C and F. Whereas a model of activation has been described for class A, very little is known about the activation of the other classes which differ by being activated by endogenous ligands bound mainly or entirely extracellularly. Here, we show that although they use the same structural scaffold and share several helix macroswitches, the GPCR classes differ in their microswitch residue positions and contacts. We present molecular mechanistic maps of activation for each GPCR class and new methods for contact analysis applicable for any functional determinants. This is the first superfamily residue- level rationale for conformational selection and allosteric communication by ligands and G proteins laying the foundation for receptor- function studies and drugs with the desired modality. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 98, 909, 450]]<|/det|> +G protein- coupled receptors (GPCRs) are nature's primary transmembrane transducers for carrying signals from extracellular ligands to intracellular effectors to regulate numerous physiological processes. The most widely used nomenclature designates GPCR classes A- F and was introduced in the first version of the GPCR database, GPCRdb1 based on conserved sequence fingerprints2,3. The human GPCRs have also been classified based on phylogenetic analysis into families (class): Glutamate (C), Rhodopsin (A), Adhesion (B2), Frizzled (F), Secretin (B1)4 and Taste 2 (T, re- classified as a separate family in5). The human GPCRs are activated from different endogenous ligand binding sites in the transmembrane (class A) or extracellular domain (class C) or both (class B and F) and have very low sequence similarity (cross- class pairs mean 23%). This raises the question to what extent the GPCR superfamily utilises universal or unique activation mechanism. Considering that GPCRs mediate the actions of two- thirds of endogenous hormones and neurotransmitters6 and over one- third of the drugs7-9, mapping their activation mechanisms is important to understand human physiology, disease aetiology and to rationally design drug. + +<|ref|>text<|/ref|><|det|>[[88, 488, 909, 748]]<|/det|> +Comparison of inactive and active structures of class A GPCRs have uncovered common activation mechanisms within the seven transmembrane helices (7TM) that have been shown to tilt, rotate, elongate or switch residue sidechain rotamers to create contact networks stabilising the receptors in a specific state10- 13. These contacts converge near the G protein site14 and are conserved regardless of the subtypes of intracellular effectors11. Site- directed mutagenesis of contact residues has revealed differences in pharmacological responses, including constitutive activity11 and signalling bias15, and naturally occurring mutations have been associated with disease or change of response11,16,17. However, the classes B, C and F are largely unexplored with respect to common activation mechanisms and until now such studies have not been possible due to a lack of structures across the activation states. + +<|ref|>text<|/ref|><|det|>[[88, 790, 908, 898]]<|/det|> +In this work, we conducted a comprehensive comparative structural analysis of inactive/active state structures in each GPCR class by analysing all available 488 structures from different GPCR classes (Fig. 1). We present the first, GPCR superfamily- wide molecular mechanistic maps of activation and link determinants to ligand binding, G protein coupling, transduction and allosteric sites. This can serve as a roadmap to assess the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 99, 905, 145]]<|/det|> +activation of any GPCR and to understand how common or distinct determinants can fine- tune physiological signalling responses and contribute to a desired drug efficacy. + +<|ref|>sub_title<|/ref|><|det|>[[92, 194, 159, 209]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[92, 225, 616, 243]]<|/det|> +## New dataset and methods for comparative structure analysis + +<|ref|>text<|/ref|><|det|>[[90, 255, 908, 516]]<|/det|> +To present the best maps possible to date, we made a comprehensive annotation of all available 406 structures of 74 class A GPCRs (twice as many as in any previous report11). These were applied to stringent quality filters to select the representative inactive/active state structures from each class (Methods and Supplementary Spreadsheet 1). Out of 68 templates used, 45 have a resolution of 3.0 Å or better. It should be noted, that in lower resolution structures (>3.0Å) mainly backbone movements can be discerned while residue contacts can only be underrepresented (i.e., not lead to false positive contacts). The largest number of structures belong to classes A and B1 and cover 22 out of 58 and all five receptor families (share endogenous ligand), respectively. This diversity ensures that any artefacts observed in a specific position or structure will only have a marginal effect on the overall frequency of movements and contacts. + +<|ref|>text<|/ref|><|det|>[[89, 556, 908, 846]]<|/det|> +We introduce a definition of state- specific contacts based on their relative frequency (%) in a given inactive/active state. Like the scores calculated in11, this allows for the identification of many more determinants than requiring 100% presence and absence, respectively in opposite states, such as the four inactive- and two active- state stabilising contacts identified in the first landmark study comparing five class A GPCRs across the states14. Importantly, the use of frequencies with (different) sets of inactive and active state structures opens up for the utilisation of also different receptors across the states, as the same helix backbone movements and contacts can be mediated by different amino acids. This is critical to the identification of a comparable number of state determinant residues in classes C and F for which the structural coverage is so far limited: four and three receptors, representing 18% and 27% of all members of these relatively small classes, respectively. Furthermore, we uniquely apply residue pair conservation cut- offs (Methods). Together, the combined contact + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 100, 905, 145]]<|/det|> +frequencies and conservation cut- offs address the class- representativeness allowing analyses across the GPCR superfamily. + +<|ref|>title<|/ref|><|det|>[[90, 191, 741, 208]]<|/det|> +# GPCR activation engages all transmembrane helices with unique contact patterns + +<|ref|>text<|/ref|><|det|>[[88, 220, 909, 660]]<|/det|> +To investigate how the seven transmembrane helices, TM1- 7 rearrange upon activation, we compared the 13 GPCRs among our representative receptors for which both an inactive and active state structure are available (Extended Data Table 1 and Methods). We find that each transmembrane helix rearranges \(>1.0 \mathrm{\AA}\) in a majority of receptors - demonstrating unappreciated conformational dynamics (class consensus is red in Fig. 2, individual receptor plots in Supplementary Figs. 1- 2). Notably, the extracellular region alone exhibits a relocation of all seven transmembrane helices in class B1 wherein the N- terminal domain restricts the conformation the 7TM before activation \(^{18}\) . GLP- 1, which is the best template having a full- length inactive structure \(^{18}\) , has a \(2.5 - 10 \mathrm{\AA}\) relocation of TM1- 7 - all of which also move in CRF1. Furthermore, we analysed the conformational change of each class across the mid, extracellular and cytosolic regions where the rearrangements total \(6 \mathrm{\AA}\) , \(12 \mathrm{\AA}\) and \(28 \mathrm{\AA}\) and involving \(24\%\) , \(45\%\) and \(65\%\) of receptor helices. This is the first quantitated characterisation of magnitude and abundance of helix rearrangements at the mid, extracellular and cytosolic regions of the transmembrane domain. It supports opposite functional roles in i) maintaining a stable GPCR fold, ii) adapting to ligands of diverse sizes or iii) coupling to the much larger G proteins, respectively. This connects structure- function and reveals a wide engagement of the GPCR fold - across its helices and domains. + +<|ref|>text<|/ref|><|det|>[[88, 685, 909, 882]]<|/det|> +We next investigated how the rearrangements of the transmembrane helix bundle change the residue contact networks between receptor segments by comparing all 43 inactive and 25 active state representative structures (Extended Data Table 1). We find that in addition to TM1- 7, state- specific contacts are also formed to helix 8 (H8) in classes A and F, intracellular loop 1 (ICL1) in class C and extracellular loop 2 (ECL2) in class F (Figs. 3- 4). Classes A, C and F have a majority of inactivating contacts (66%, 56% and 62%, respectively, Fig. 3a). Of note, each contact typically spans a majority and at least 30% of all receptors in each class based on residue- residue conservation (Methods). Class B1 has 48% inactivating contacts spanning segment contacts (excluding + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 99, 909, 250]]<|/det|> +four intra- helix contacts). This characterisation of contacts by state – where the majority, or close to half for B1, of contacts restrain receptors into an inactive conformation – provides a structural rationale to why most receptors have no or little activity without the prior stimulation by an agonist. Furthermore, it demonstrates that any TM helices can switch from mainly inactive to active state contacts (Figure 3B). Notably, this rewiring leads to markedly different patterns of segment contacts across GPCR classes that combine several unique and some common contacts (below), as even high- homology receptors sharing endogenous ligand, such as the A₁ and A₂A adenosine or muscarinic acetylcholine M₁ and M₂ receptors, display unique helix movements (Fig. 2). + +<|ref|>text<|/ref|><|det|>[[90, 260, 907, 305]]<|/det|> +and A₂A adenosine or muscarinic acetylcholine M₁ and M₂ receptors, display unique helix movements (Fig. 2). + +<|ref|>text<|/ref|><|det|>[[90, 340, 908, 447]]<|/det|> +Together, the existence of both common and unique movements and contacts (below) provides a structural rationale for shared overall functions throughout the GPCR superfamily, e.g. ligand- dependent activation, signal transduction across the cell membrane and G protein coupling, that yet are diversified with respect to the specific ligand scaffold, G protein profile and functional response kinetics and efficacy. + +<|ref|>text<|/ref|><|det|>[[88, 490, 909, 875]]<|/det|> +TM6 is a universal helix ‘macroswitch’ but toggles, rotates and unwinds differently across GPCR classes We next single helix rearrangements across 13 receptor inactive/active state structure pairs confirms. We find that outward movement and rotation of TM6 on the cytosolic side – opening for G protein coupling – is a universal feature of activation throughout the GPCR superfamily (Fig. 2a), as first suggested for class A₁⁹. Intracellular TM6 movement is observed in all 13 investigated receptors and is the largest in classes A (7- 13 Å), B1 (14- 19 Å) and F (7 Å) where is combined with a substantial average rotation: 38°, 39° and 38°, respectively. The opposite, extracellular end of TM6 also moves by on average 2.0, 3.4 and 6.2 Å, respectively, in these classes. The movement of TM6 gives it the largest (B1 and F) or second largest (A) number of contacts stabilising an inactive or active state in these classes (Fig. 3). However, our analysis shows that TM6 also displays large mechanistic differences across all the classes. Class B1 has a unique unwinding of the extracellular- facing half of TM6 in all three activated receptors²⁰- ²². This is in agreement with a recent study, comparing the class B1 Glucagon and class A β₂ receptors, linking the weaker ability of the agonist to induce the outward movement of cytoplasmic TM6 to a slower G protein activation²¹. Furthermore, in class C, TM6 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 99, 908, 330]]<|/det|> +uniquely has no extracellular movement and only a small (2.8 Å) movement and rotation (13°). Consequently, TM6 has only one state- specific contacts to other helices whereas TM5 has five and TM7, with two such contacts, is the only switch region with contacts across both states (Fig. 3b). The smaller role of TM6 in the activation throughout the class is supported by a markedly lower conservation of its proline, 55% compared to 93- 100% in other classes A- C while the importance of TM7 is emphasised by a 95% conserved proline (Fig. 2b). The non- canonical small TM6 tilt with a less conserved kink and atypical contacts to the adjacent TM5 and TM7 are associated with a different overall mechanism, as class C GPCRs work as an asymmetric homo- or hetero- dimer in which only one of two subunits couples to a G protein. + +<|ref|>text<|/ref|><|det|>[[88, 372, 908, 510]]<|/det|> +These findings show that the GPCR activation machinery utilises TM6 as universal yet differentiated switches - combining helix toggling, rotation and unwinding. Such commonality across classes in the activation mechanism at the level of transmembrane helices but diversity and nuances at the type of movements provides an important structural rationale for drug discovery and in the future design of experiments to elucidate effects of mutations, ligand efficacy and G protein selectivity23 mechanisms. + +<|ref|>sub_title<|/ref|><|det|>[[90, 538, 905, 584]]<|/det|> +## TM5 is a new universal switch, TM5-7 act in concert and TM3 is a hub for stabilisation of inactive or active receptors + +<|ref|>text<|/ref|><|det|>[[88, 596, 909, 888]]<|/det|> +TM5, like TM6, can be considered a universal switch for GPCR activation, as it moves on the intracellular side in all four classes (with on average A: 2.1, B1: 2.4, C: 2.0 and F: 1.8 Å), including a majority of class A and all class B1, C and F receptors (Fig. 2b and (Extended Data Fig. 2). TM7 also moves in all classes except class C, either on the cytosolic side (all class A receptors), the extracellular side (e.g. 10 Å movement and 100° rotation in GLP- 1, the class B1 receptor with full- length templates) or on both sides (class F). These movements are possible due to a number of conserved proline and glycine residues that induce helix plasticity (Fig. 2b). Class A GPCRs have triple proline kinks in TM5- 7, that allow TM5 and TM7 to close- in on and stabilise TM6. Class B1 TM6 unwinding is facilitated by both a proline kink (P6x47) and a glycine (G6x50). Class B1 GPCRs also feature a proline kink (P5x42) near the extracellular end of TM5, which moves 3.3 Å in GLP- 1, and a glycine kink (G7x50) in TM7. Class F combines the TM6 switch with movements of cytosolic + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 99, 907, 177]]<|/det|> +TM5 (with one Pro and two Gly residues) and TM7 on both sides. This demonstrates that TM6 does not act on its own but is supported by TM5 and TM7 in a concerted movement and that the determinants of this plasticity are conserved throughout the classes. + +<|ref|>text<|/ref|><|det|>[[89, 220, 908, 602]]<|/det|> +We find that TM3 contacts the largest number (3- 5) other receptor segments in each GPCR class (Fig. 3), revealing that TM3 functions as a stabilisation hub – not just to maintain the common transmembrane fold24 – but to stabilise a specific inactive or active state across the GPCR superfamily. Furthermore, whereas an early report based on two class A GPCRs suggested an activation mechanism involving an upward movement of TM325, our analysis of TM helix movements in 15 receptors across classes instead points to rotation as the main mechanism (Extended Data Fig. 1). In 11 out of 13 the investigated receptor pairs (all except A2A and CRF1), TM3 rotates at either the cytosolic (most frequent for class A, average 16°) or extracellular end (classes B1, C and F, average 19°, 12° and 17°, respectively) – while no receptor rotates in both ends or in the mid (Figs. S1- 2). Lateral movement of TM3 contributes to a different extent across classes spanning from both ends (B1), cytosolic end (C), either end for a minority of receptors (A) to no movement (F). In contrast, TM4 – which is peripherally located in the transmembrane helix bundle – has few or no contacts (class A:1, B1:0, C:3 and F:0). This shows that the abundant helix packing has not immobilised TM3 which instead, by an array of mainly local rotation or movement, contributes to GPCR activation in several places. + +<|ref|>text<|/ref|><|det|>[[90, 630, 906, 675]]<|/det|> +Contact state frequency approach uncovers residue ‘microswitches’ in all GPCR classes and expands class A activation model + +<|ref|>text<|/ref|><|det|>[[89, 690, 908, 888]]<|/det|> +To investigate state determinants on the residue ‘microswitch’ level, we indexed topologically corresponding receptor positions with generic residue numbers26 and classified them into ‘inactivators’, ‘activators’ and ‘switches’ based on frequent contacts in inactive, active and both states, respectively (Methods). We uncover a comparable number of state determinants across classes: A: 41, B1: 33, C: 28 and F: 33 (Fig. 4). 86% of determinants (A: 80%, B1: 72%, C: 100% and F: 91%) are inactivators or activators whereas only 16 determinants are switches. Only 9 switches undergo sidechain rotamer shifts (residue text rotation in Fig. 4a) revealing that that the rotamer microswitches – described as major state determinants for class A GPCRs12,13 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 99, 909, 299]]<|/det|> +- play a small role in the GPCR superfamily. Importantly, many determinant positions contain the same highly conserved amino acid (grey scaling in Fig. 4a) and the amino acid pairs observed for each contact are conserved in at least 40% of all receptors (Methods). Notably, this includes the reference positions for generic residue numbers (index x50) in 5 out of 7 transmembrane helices, helix 8 and ICL1 and all major previously known class A state determinants (sequence motifs and microswitches with magenta border in Fig. 4a)11,12. This demonstrates that our new approach (Discussion) identifies both known and new conserved determinants – even where the structural coverage is so far limited including the active state of classes C and F. + +<|ref|>text<|/ref|><|det|>[[88, 338, 909, 870]]<|/det|> +In class A, W6x48, initially described as a rotamer toggle switch12 does not itself undergo a rotamer but a helix rotational shift (of \(10^{\circ}\) ) and is approached by I3x40 which has both types of rotations. This provides an alternative to a reported ‘PIF’ motif27- 29 which is here found to consist most frequently of ‘PIW’ (P5x50 being the third residue). The PIFs motifs last residue, F6x44 is instead here shown to act as a switch in another new triplet ‘LLF’ located on the same helices: TM3, TM5 and TM6. The two residues D2x50 and N7x49 which coordinate a sodium ion30 are here found to have a direct interaction stabilising the inactive state. Notably, N7x49 contacts P7x50 uncovering a concerted stabilisation across the sodium ion site and TM7 helix kink around the ‘NPxxY’ motif. The final residue of the NPxxY motif, a Y7x53 switch11- 13, has three inactivating and two activating state-specific contacts with over 40% frequency difference, including to another switch: I3x46. The TM3 ‘DRY’ motif includes an intra- helical ionic lock from D3x49 to R3x50, which upon activation swings to interact with the G protein as well as the switch Y5x58 on TM5. Of note, the restraining of R3x50 is strengthened by contacts to TM2 and TM6 (A6x34) – however typically not to E6x30, which was part of the first ionic lock reported in Rhodopsin31 but less conserved (E: 25% or D/E: 31% compared to D: 65% or D/E: 86% for position 4x49). On the intracellular side, helix 8 and ICL1 contain three and two determinants, respectively, including the novel switch F8x50 contacting another new switch on TM7, A7x54. These findings corroborate the findings of previous studies on class A10- 14. They also, together with the concerted movement of TM5- 7 and role of TM3 as a state stabilisation hub (above), substantially expand the activation model of class A receptors – the largest class of GPCRs in human. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 115, 905, 161]]<|/det|> +Class B1 shares many class A state determinant residues while having a unique concentration in and around the TM6 helix switch + +<|ref|>text<|/ref|><|det|>[[88, 174, 909, 404]]<|/det|> +The first comparison of unique and common state determinants across the GPCR superfamily shows that nearly half, 16 out of 33, of the B1 determinants maps to equivalent topological positions as in class A compared to 10 for class F and 8 for class C (leftmost in Fig. 4b). Furthermore, the classes B1 and A have four common switches (second to rightmost in Fig. 4b). Notably, this includes the two known microswitches (A/B1 residue number): Y5x58/F5x54 and Y7x53/Y7x57 as well as the pair I3x46/E3x50 and F6x44/L6x49 (magenta in Extended Data Fig. 3). We also find that the class A 'toggle switch' activator W6x48 is also conserved in class B1 as a smaller aromatic residue Y6x53 (Extended Data Fig. 3). Together, these commonalities between classes B1 and A are indicative of an, in part, shared activation mechanism. + +<|ref|>text<|/ref|><|det|>[[88, 444, 909, 744]]<|/det|> +However, there are also markedly unique features in class B1. Strikingly, 14 out of 33 of state determinants in class B1 are clustered on one helix, TM6 which engages 10 additional determinants – of which all except one are located in TM5 and TM7 (Fig. 4) which move together with TM6 (Fig. 2). The high concentration of state determinants in TM6 and packing to adjacent helices, may provide a plausible structural rationale for a recent report demonstrating a higher energy barrier for the formation of the kinked and partially unwound TM6 in the class B1 Glucagon receptor compared to the class A \(\beta_{2}\) - adrenoceptor21. Another unique feature in class B1 are two additional switches on TM7 (Q7x49 and L7x56). Class B1, like class A, has a large structural coverage (10 out of 15 receptors) and contains several major drug targets7. The first map of state determinants in class B1 gives a better understanding of the basic receptor activation mechanisms and presents a foundation for the targeting of determinant networks in structure- based design of new drugs stabilising receptors in desired states. + +<|ref|>sub_title<|/ref|><|det|>[[90, 777, 718, 795]]<|/det|> +## The GPCR superfamily lacks a universal residue-level microswitch mechanism + +<|ref|>text<|/ref|><|det|>[[88, 808, 908, 886]]<|/det|> +Class C is atypical in that it contains as many inactivating and activating contacts and no switches in the transmembrane domain (compared to 3- 8 in the other classes, two- way Venn diagram in Fig. 4a). This is in concordance with its – by far – smallest conformational change (Fig. 2). Another characteristic of class C is + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 98, 909, 450]]<|/det|> +that it has the highest share of determinants, \(36\%\) (compared to A: \(20\%\) , B1: \(24\%\) and F: \(0\%\) ) that form contacts within the same transmembrane helix. As class C is the only GPCR class that binds endogenous ligands entirely in the N- terminus, the transmembrane domain solely transduces signals. On the cytosolic side this involves two determinants in ICL1 (as in class A). Class F uniquely has frequent state stabilising contacts to ECL2, Y45x51 and V45x52 – both of which stabilise the active state by interaction with the same residue, F3x29 in TM3. The two ECL2 positions follow a conserved cysteine C45x50 which forms a covalent disulphide bridge, also to TM3, across the GPCR classes \(^{32}\) . Class F has the highest conservation of determinant consensus amino acids (on average \(90.5\%\) ) showing that the vast majority of contacts identified in the three structural templates: FZD4, FZD5 and Smoothened are likely shared by its remaining eight receptors. Class F has three switches, M3x37, W3x50 and M6x30 of which the former two have a large sidechain rotamer shift ( \(49^{\circ}\) and \(47^{\circ}\) , respectively). Interestingly, like class A, class F has a switch in position 3x50 although occupied by a tryptophan (A has arginine). + +<|ref|>text<|/ref|><|det|>[[88, 490, 909, 840]]<|/det|> +By comparing all GPCR classes, we find that 57 out of 91 determinant positions reside in a distinct topological position and are therefore unique, whereas 25 are found in two classes, 8 in three classes and only 1 throughout the GPCR superfamily (leftmost Venn in Fig. 4b). Of note, the universal determinant position (numbered 3x46 in class A/F and 3x50 in class B1/C) plays a different role as inactivator (C), activator (F) or switch (A/B1). Considering such different roles in all determinant positions reveals an even smaller commonality – presence in a maximum of two GPCR classes for 12 inactivators (max 4 in A-F), 5 activators (max 2 in A-B1 and B1-F) and 5 switches (max 4 A-B1) (second to rightmost Venn in Fig. 4b and Extended Data Fig. 3). These findings demonstrate that although they belong to the same superfamily, use the same structural scaffold and share several macroswitches (helices), the GPCR classes differ in their microswitches (residues). This suggests that there are an ensemble of structural/mechanistic solutions that is available for the GPCR superfamily during evolution that different receptor classes have explored and utilised. It also means that while thinking about developing drugs with different modalities – especially for classes C and F – one should aim to interact with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 99, 905, 145]]<|/det|> +or modulate class specific state determinants and residue- level micro- switches rather than use the same state determinant and switch as in class A. + +<|ref|>text<|/ref|><|det|>[[90, 193, 907, 230]]<|/det|> +State determinant mapping reveals where ligands and G proteins can sense and stabilise receptor states + +<|ref|>text<|/ref|><|det|>[[90, 220, 907, 630]]<|/det|> +To obtain a topological and functional mapping of the state determinants, we mapped their location in relation to 408 ligand and 125 Gα protein interacting positions across the GPCR classes (grey and orange in Fig. 5) extracted from all 488 available GPCR structures. We find that all GPCR classes have the largest share, 60- 68% of state determinants in ligand binding positions (Supplementary Spreadsheet 2). Class B1 is the only GPCR class with an even share across orthosteric and allosteric ligand positions (above/below membrane "Mid" in hexagon residue on TM1- 7). Class A has only three (2%) determinants (W6x48, L7x40 and S7x46) in orthosteric ligand- binding positions and other state stabilisers are located below the membrane mid. Whereas class F has 18% of such determinants, class C lack orthosteric determinants altogether in the 7TM, as it binds its endogenous ligands solely in the N- terminus. When instead considering the proportion of state determinants in G protein binding positions, we find that these account for around one- third in classes A (29%) and B1 (36%) but a much smaller fraction in classes C and F (6- 7%). For class C this may reflect its uniquely shallow G protein binding but the figures in both classes is likely underrepresented due to few G protein complex structures. The remaining determinants (at non- interacting positions) account for 21- 36% on the GPCR class level. + +<|ref|>text<|/ref|><|det|>[[90, 675, 908, 872]]<|/det|> +Given their important role in modulating GPCR activity, we made a specific mapping of switches i.e. determinant residues that alternate contacts across the inactive/active states. In class A, two switches (R3x50 and 5x58) have direct G protein interactions whereas the remaining six are located near or in between ligand and G protein positions facilitating the allosteric signal transduction across the extra- and intra- cellular sides. Class B1 switches are distributed across orthosteric ligand (6x49), G protein (3x50, 7x56 and 7x57) and two other (5x54 and 7x49) positions. Class C uniquely has no switches in the transmembrane domain. Finally, class F shares one class A switch with G protein interaction (3x50) and has two switches in positions without + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 98, 908, 237]]<|/det|> +a known interaction. Together, this functional mapping shows that ligands and G proteins can sense and stabilise receptor states by direct interactions and that their signals are largely transduced via interconnecting pathways while state determinants including switches are spread throughout the 7TM bundle. This may help to explain observed effects on ligand or G protein affinity from remote mutations and presents a residue- level rationale for conformational selection33 and allosteric modulation34. + +<|ref|>sub_title<|/ref|><|det|>[[92, 268, 187, 284]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[88, 297, 909, 828]]<|/det|> +We present the first molecular mechanistic maps for activation across the GPCR superfamily and helix macro- /residue micro- scales while extending beyond the transmembrane region to helix 8 and structurally conserved loop segments. This study has answered the fundamental question of whether the activation of other classes can be modelled based on class A – they cannot. Our findings demonstrate that although they belong to the same superfamily, use the same structural scaffold and share several helix macroswitches, the GPCR classes differ in their microswitch residue positions, contacts and amino acids. This applies also to class B1 which shares about half of the determinant positions in class A, as the similarity is very small if considering the type of determinant – stabilising an inactive, active or both states – and their specific consensus amino acids. This highlights the need to elucidate each GPCR class’ restraints and diverse activation mechanisms separately and in more detail to adequately capture structure- function relationships. Determinants of receptor activity are tightly tied to the molecular mechanisms of conformational selection (Kenakin, 2019) and allosteric modulation34, and influence an array of functions and responses, including ligand affinity, basal activity, efficacy and G protein coupling. Therefore, the contact maps (Fig. 4) presented here provide an actionable foundation for the field to design and interpretate experiments across structural, biophysics (e.g. fluorescence and double electron- electron resonance), molecular dynamics and mutagenesis studies. The maps also inform drug discovery of which inactivating and activating state determinants are located in the orthosteric and allosteric ligand sites and may therefore facilitating effective design of inverse agonists, neutral antagonists or agonists for different receptors. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 98, 909, 448]]<|/det|> +The extensive engineering and limited resolution of some structures is an inherent limitation for all structure- based studies. Therefore, more structures and advances in their determination are of outmost value and will serve to continuously refine structure- function relationships generally. For example, the universal TM5 switch reported herein was not discernible until cryo- electron microscopy allowed for more structures with a native TM5- ICL3- TM6 region (often subjected to deletion and protein fusion in crystallography fusions not in cryo- EM). Our combined contact frequency and residue pair conservation cut- offs uniquely addresses the class- representativeness and allowed for the GPCR superfamily to be described. Class F has a \(90.5\%\) average conservation of determinant consensus amino acids. For these reasons, the common conserved determinants identified herein should still apply as our knowledge expands from new structures, which could also allow additional determinants fulfil these cut- offs. Furthermore, although specific receptors and subsets thereof could have additional activation mechanisms not conserved in the whole class, and such specific and general mechanisms, respectively could act in concert. + +<|ref|>sub_title<|/ref|><|det|>[[93, 491, 262, 508]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[90, 522, 907, 629]]<|/det|> +This work was supported by the Lundbeck Foundation (grants R163- 2013- 16327 and R218- 2016- 1266), the Novo Nordisk Foundation (grant NNF18OC0031226) and Independent Research Fund Denmark | Natural Sciences (grant 8021- 00173B) to D.E.G and by the American Lebanese Syrian Associated Charities (ALSAC). + +<|ref|>sub_title<|/ref|><|det|>[[92, 643, 285, 661]]<|/det|> +## Author Contributions + +<|ref|>text<|/ref|><|det|>[[89, 674, 907, 784]]<|/det|> +Conceptualization, D.E.G.; Methodology, D.E.G., C.M. and A.J.K.; Data Curation, A.J.K.; Investigation, D.E.G.; Validation: D.E.G. and A.J.K.; Writing – Original Draft, D.E.G.; Writing – Review & Editing, A.S.H. and M.M.B.; Visualization, D.E.G., A.S.H. and A.J.K., Funding Acquisition, D.E.G. and M.M.B.; Supervision, D.E.G. + +<|ref|>sub_title<|/ref|><|det|>[[92, 797, 269, 814]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[92, 829, 418, 845]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[92, 100, 213, 115]]<|/det|> +## 1 Online content + +<|ref|>text<|/ref|><|det|>[[90, 129, 905, 177]]<|/det|> +2 Methods, additional references, Nature Research reporting summaries, source data, extended data, peer review3 information; and statements of data and code availability are available at [URL]. + +<|ref|>sub_title<|/ref|><|det|>[[92, 254, 206, 270]]<|/det|> +## 6 Main figures + +<|ref|>image<|/ref|><|det|>[[226, 312, 770, 655]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[90, 683, 905, 755]]<|/det|> +
Fig. 1 | Analysis pipeline for elucidation of GPCR activation mechanisms. Pipeline for analysis of universal and distinct activation macro-/microswitches spanning helix repacking to sidechain rotation and connection to ligand binding, G protein coupling and signal transduction sites (see Methods).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 95, 907, 504]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[90, 536, 907, 744]]<|/det|> +
Fig. 2 | Transmembrane helix movement upon activation and universal TM3 and TM6 helix 'macroswitches'. a, Movements (A) over 1.0 Å at the extracellular end, membrane mid (determined using \(^{35}\) ) and intracellular end of the transmembrane helices, TM1-7 upon comparison of all available receptor inactive- and active-state structure pairs (Extended Data Table 1). Red intensity denotes the number of classes with a consensus movement (for class C, priority is given to the Gi-GABA \(_{B2}\) over the non-coupled mGluR5 structure). b, Movement and conserved hinges of TM6 and the adjacent TM5 and TM7. c, TM3 cytosolic tilt and overall rotation. b-c, GPCR class representative inactive/active receptor structure pairs: A: \(\beta_{2}^{36,37}\) , B1: GLP-1 \(^{18,22}\) , C: GABA \(_{B2}^{38}\) and F: Smoothened \(^{39,40}\) receptors. Proline and glycine residues that increase helix plasticity are shown along with their percent conservation in the GPCR class.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 99, 800, 905]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 99, 909, 500]]<|/det|> +Fig. 3 | GPCR stablise inactive and active states by rerouting contacts between TM1- 7, H8, ICL1 and ECL2. a, State- specific residue- residue contacts in each GPCR class visualised as lines within representative inactive/active receptor structure pairs (same as in Fig. 2b- c). Numbers indicate the total, inactivating and activating contacts in each GPCR class. b, Contact networks between the seven GPCR transmembrane helices, TM1- 7 and the first intracellular (ICL1) and second extracellular (ECL2) loops (intra- segment contacts not shown). Line thickness represents the number of classes (topmost) or contact frequency differences between the inactive and active states. Line colour indicates inactivating (blue) and activating (red) contacts. Receptor segments with a magenta border are “switches”, i.e. have contacts across both states. Contacts are identified based on a higher frequency (%) in one set of 43 inactive and 25 active state receptors. a- b, The frequency difference thresholds was set according to the structural coverage in each GPCR class (no. members and inactive/active state templates): A: 40% (285, 33/14), B1: 67% (15, 3/10), C: 75% (22, 4/1) and F: 100% (11, 3/1) – yielding comparable numbers of determinants and contacts. To ensure that the identified determinants are applicable for throughout each class, we applied a cut- off requiring at least 30% of all its receptors to contain one of the amino acid pairs observed to form the given state specific contact. Contact definitions are explained in the settings menu of the online ‘Comparative structure analysis’ tool (https://review.gpcrdb.org/structure_comparison/comparative_analysis). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[90, 95, 910, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[262, 666, 732, 679]]<|/det|> +
Common and unique state determinants across the human GPCR superfamily
+ +<|ref|>image<|/ref|><|det|>[[90, 682, 909, 820]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 98, 910, 280]]<|/det|> +1 Fig. 4 | State stabilising contact maps and differences at the residue-level 'microswitches'. a, Contact networks 2 visualise the wiring of state determinants from the extracellular (top) to intracellular (bottom) sides. Contact frequency 3 differences between the inactive and active states are shown as varied line thickness and residue rotamers as rotation of 4 the consensus amino acid in analysed structures and its generic residue number. Two-way Venn diagrams depict the 5 number and percentages of inactivator (blue), activator (red) and switch (magenta) state determinant positions. Bar 6 diagrams show their distribution across the TM helices, H8 and loops. b, Comparison of common and unique state 7 determinant positions across all investigated classes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[210, 90, 777, 900]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 99, 909, 333]]<|/det|> +Fig. 5 | Residue positions stabilising an inactive and/or active receptor state. GPCR snakeplots mapping the residue positions that form distinct contacts between state determinants classified as inactivators (blue), activators (red) and switches (magenta). Residues are denoted with the consensus amino acid of the investigated receptor structures (Extended Data Table 1) and their generic residue number26. Filled positions map ligand (grey) and G protein (orange) interaction frequency among all GPCRs in the given class that have such data (Supplementary Spreadsheet 2). Border grayscale denotes the frequency of mutations changing the ligand affinity or activity over 5-fold. The label “Mid” within hexagonal-shaped positions denote the membrane mid, above and below which the ligand positions are considered orthosteric and allosteric, respectively, in class A, B1 and F. In class C, all ligand interacting positions in the seven transmembrane helices are allosteric, as its endogenous ligands bind solely in the N-terminal domain. + +<|ref|>sub_title<|/ref|><|det|>[[93, 403, 190, 420]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[145, 433, 839, 451]]<|/det|> +Kolakowski, L. F., Jr. GCRDb: a G-protein-coupled receptor database. \*Recept Channels\* 2, 1- 71994). + +<|ref|>text<|/ref|><|det|>[[148, 460, 907, 506]]<|/det|> +Attwood, T. K. & Findlay, J. B. Design of a discriminating fingerprint for G-protein-coupled receptors. \*Protein Eng.\* 6, 167- 176, doi:10.1093/protein/6.2.167 (1993). + +<|ref|>text<|/ref|><|det|>[[149, 515, 907, 560]]<|/det|> +Attwood, T. K. & Findlay, J. B. Fingerprinting G-protein-coupled receptors. \*Protein Eng.\* 7, 195- 203, doi:10.1093/protein/7.2.195 (1994). + +<|ref|>text<|/ref|><|det|>[[149, 570, 907, 642]]<|/det|> +Fredriksson, R., Lagerstrom, M. C., Lundin, L. G. & Schioth, H. B. The G-protein-coupled receptors in the human genome form five main families. Phylogenetic analysis, paralogon groups, and fingerprints. \*Molecular pharmacology\* 63, 1256- 1272, doi:10.1124/mol.63.6.1256 (2003). + +<|ref|>text<|/ref|><|det|>[[149, 652, 907, 725]]<|/det|> +Nordstrom, K. J., Sallman Almen, M., Edstam, M. M., Fredriksson, R. & Schioth, H. B. Independent HHsearch, Needleman- Wunsch- based, and motif analyses reveal the overall hierarchy for most of the G protein- coupled receptor families. \*Mol Biol Evol\* 28, 2471- 2480, doi:10.1093/molbev/msr061 (2011). + +<|ref|>text<|/ref|><|det|>[[149, 734, 907, 780]]<|/det|> +Foster, S. R. et al. Discovery of Human Signaling Systems: Pairing Peptides to G Protein- Coupled Receptors. \*Cell\* 179, 895- 908 e821, doi:10.1016/j.cell.2019.10.010 (2019). + +<|ref|>text<|/ref|><|det|>[[149, 789, 907, 861]]<|/det|> +Hauser, A. S., Attwood, M. M., Rask- Andersen, M., Schioth, H. B. & Gloriam, D. E. Trends in GPCR drug discovery: new agents, targets and indications. \*Nat. Rev. Drug Discov.\* 16, 829- 842, doi:10.1038/nrd.2017.178 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 99, 907, 144]]<|/det|> +Sriram, K. & Insel, P. A. G Protein-Coupled Receptors as Targets for Approved Drugs: How Many Targets and How Many Drugs? Molecular pharmacology 93, 251- 258, doi:10.1124/mol.117.111062 (2018). + +<|ref|>text<|/ref|><|det|>[[150, 155, 907, 198]]<|/det|> +Santos, R. et al. A comprehensive map of molecular drug targets. Nat. Rev. Drug Discov. 16, 19-34, doi:10.1038/nrd.2016.230 (2017). + +<|ref|>text<|/ref|><|det|>[[150, 210, 907, 253]]<|/det|> +Flock, T. et al. Selectivity determinants of GPCR-G-protein binding. Nature 545, 317-322, doi:10.1038/nature22070 (2017). + +<|ref|>text<|/ref|><|det|>[[150, 265, 907, 308]]<|/det|> +Zhou, Q. et al. Common activation mechanism of class A GPCRs. Elife 8, e50279, doi:10.7554/eLife.50279 (2019). + +<|ref|>text<|/ref|><|det|>[[150, 320, 907, 390]]<|/det|> +Nygaard, R., Frimurer, T. M., Holst, B., Rosenkilde, M. M. & Schwartz, T. W. Ligand binding and micro-switches in 7TM receptor structures. Trends in pharmacological sciences 30, 249-259, doi:10.1016/j.tips.2009.02.006 (2009). + +<|ref|>text<|/ref|><|det|>[[150, 402, 907, 472]]<|/det|> +Carpenter, B. & Tate, C. G. Active state structures of G protein-coupled receptors highlight the similarities and differences in the G protein and arrestin coupling interfaces. Current opinion in structural biology 45, 124-132, doi:10.1016/j.sbi.2017.04.010 (2017). + +<|ref|>text<|/ref|><|det|>[[150, 484, 905, 528]]<|/det|> +Venkatakrishnan, A. J. et al. Diverse activation pathways in class A GPCRs converge near the G-protein-coupling region. Nature 536, 484-487, doi:10.1038/nature19107 (2016). + +<|ref|>text<|/ref|><|det|>[[150, 539, 905, 583]]<|/det|> +Schonegge, A. M. et al. Evolutionary action and structural basis of the allosteric switch controlling beta2AR functional selectivity. Nat Commun 8, 2169, doi:10.1038/s41467-017-02257-x (2017). + +<|ref|>text<|/ref|><|det|>[[150, 594, 905, 638]]<|/det|> +Hauser, A. S. et al. Pharmacogenomics of GPCR Drug Targets. Cell 172, 41-54 e19, doi:10.1016/j.cell.2017.11.033 (2018). + +<|ref|>text<|/ref|><|det|>[[150, 649, 905, 693]]<|/det|> +Schoneberg, T. & Liebscher, I. Mutations in G Protein-Coupled Receptors: Mechanisms, Pathophysiology and Potential Therapeutic Approaches. Pharmacol. Rev. 73, 89-119, doi:10.1124/pharmrev.120.000011 (2021). + +<|ref|>text<|/ref|><|det|>[[150, 704, 905, 747]]<|/det|> +Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). + +<|ref|>text<|/ref|><|det|>[[150, 759, 905, 802]]<|/det|> +Schoneberg, T., Schultz, G. & Gudermann, T. Structural basis of G protein-coupled receptor function. Molecular and cellular endocrinology 151, 181-193, doi:10.1016/s0303-7207(99)00017-9 (1999). + +<|ref|>text<|/ref|><|det|>[[150, 814, 905, 857]]<|/det|> +Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 99, 910, 884]]<|/det|> +1 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. 2 Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 3 22 Zhang, X. et al. Differential GLP- 1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 4 80, 485- 500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 5 23 Capper, M. J. & Wacker, D. How the ubiquitous GPCR receptor family selectively activates signalling pathways. 6 Nature 558, 529- 530, doi:10.1038/d41586- 018- 05503- 4 (2018). 7 24 Venkatakrishnan, A. J. et al. Molecular signatures of G- protein- coupled receptors. Nature 494, 185- 194, 8 doi:10.1038/nature11896 (2013). 9 25 Tehan, B. G., Bortolato, A., Blaney, F. E., Weir, M. P. & Mason, J. S. Unifying family A GPCR theories of 10 activation. Pharmacology & therapeutics 143, 51- 60, doi:10.1016/j.pharmthera.2014.02.004 (2014). 11 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in 12 pharmacological sciences 36, 22- 31, doi:10.1016/j.tips.2014.11.001 (2015). 13 Rasmussen, S. G. et al. Structure of a nanobody- stabilized active state of the beta(2) adrenoceptor. Nature 469, 14 175- 180, doi:10.1038/nature09648 (2011). 15 Wacker, D. et al. Structural features for functional selectivity at serotonin receptors. Science 340, 615- 619, 16 doi:10.1126/science.1232808 (2013). 17 Huang, W. et al. Structural insights into micro- opioid receptor activation. Nature 524, 315- 321, 18 doi:10.1038/nature14886 (2015). 19 White, K. L. et al. Structural Connection between Activation Microswitch and Allosteric Sodium Site in GPCR 20 Signaling. Structure 26, 259- 269 e255, doi:10.1016/j.str.2017.12.013 (2018). 21 Palczewski, K. et al. Crystal structure of rhodopsin: A G protein- coupled receptor. Science 289, 739- 745, 22 doi:10.1126/science.289.5480.739 (2000). 23 de Graaf, C., Foata, N., Engkvist, O. & Rognan, D. Molecular modeling of the second extracellular loop of G- 24 protein coupled receptors and its implication on structure- based virtual screening. Proteins 71, 599- 620, 25 doi:10.1002/prot.21724 (2008). 26 Kenakin, T. Biased Receptor Signaling in Drug Discovery. Pharmacol. Rev. 71, 267- 315, 27 doi:10.1124/pr.118.016790 (2019). 28 DeVree, B. T. et al. Allosteric coupling from G protein to the agonist- binding pocket in GPCRs. Nature 535, 29 182- 186, doi:10.1038/nature18324 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 99, 910, 465]]<|/det|> +1 35 Lomize, M. A., Pogozheva, I. D., Joo, H., Mosberg, H. I. & Lomize, A. L. OPM database and PPM web server: resources for positioning of proteins in membranes. Nucleic Acids Res. 40, D370-376, doi:10.1093/nar/gkr703 (2012). 4 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 6 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 8 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 9 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 12 40 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[80, 145, 911, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[90, 750, 905, 825]]<|/det|> +
Extended Data Fig. 1 | TM1-7 movement of class A receptor pairs upon activation. Transmembrane helix movement at the extracellular face, membrane mid and intracellular face based on comparison of representative inactive and active state structure pairs of class A GPCRs (Extended Data Table 1).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 92, 910, 560]]<|/det|> + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[175, 95, 820, 720]]<|/det|> +
State determinantsConsensus AARepresentative receptors
AB1CFSumA B1C Fβ2GLP-1GABAB2SMO
TM11x501x472NTN51L159A496T245
1x531x502VTV54A162F499T248
TM22x402x472LF158L166N503F252
2x472x482NNN69N177P517A264
2x502x572ASA76F1841524N271
3x373x332DFD79F187G527F274
TM33x403x442QYV114Q234T561Y322
3x433x432LYI121Y241F568V329
3x493x452WLL124L244M571F332
3x463x503x504IEKV126V246A573V334
TM53x503x543RLWI127E247K574L335
5x505x502PLR131L251V578W339
5x515x475x513LIMP211A316L667V411
5x585x542YFL212I317M668L412
TM65x575x652IGY219F324A675L419
6x386x352LEV222V327T678G422
6x346x302AMA271L349D688T448
6x416x322SRL272A350S689M449
TM76x376x422LTT274S352Y691R451
6x386x432LLL275T3531692L452
6x406x456x363VLG276L354G693G453
6x416x462VII278L356S695F455
TM86x446x496x403FLYM279I357V696G456
6x516x482IIF282L360V699A459
6x476x526x493CHL284T362I701G461
6x486x532WYC285H363M702F462
TM97x427x412LKW286E364C703V463
7x407x447x433LFN312L388V724E518
7x457x442GMI314F390L726I520
7x497x392QLG315T391V727N521
TM107x527x567x553ILWN318Q394C731M525
7x537x572YYI325L401L738M532
7x547x487x573AFWY326Y402F740S533
C327C403V741T534
+ +<|ref|>text<|/ref|><|det|>[[50, 730, 64, 744]]<|/det|> +1 + +<|ref|>text<|/ref|><|det|>[[50, 760, 64, 774]]<|/det|> +2 + +<|ref|>text<|/ref|><|det|>[[50, 787, 906, 880]]<|/det|> +3 **Extended Data Fig. 3 | State determinants conserved across classes.** Heatmap of state determinants shared by at least two GPCR classes along with their consensus amino acid and residues in representative receptors. Each row contains corresponding positions denoted with the generic residue numbers in each class26. Key class A state determinants are shown in bold. + +<|ref|>text<|/ref|><|det|>[[50, 900, 64, 914]]<|/det|> +3 + +<|ref|>text<|/ref|><|det|>[[50, 928, 64, 942]]<|/det|> +4 + +<|ref|>text<|/ref|><|det|>[[50, 956, 64, 970]]<|/det|> +5 + +<|ref|>text<|/ref|><|det|>[[50, 984, 64, 997]]<|/det|> +6 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 100, 905, 115]]<|/det|> +1 Extended Data Table 1 | Representative structures of GPCRs and inactive and active states used as templates. The + +<|ref|>text<|/ref|><|det|>[[52, 127, 870, 141]]<|/det|> +2 structure annotation and criteria-based filtering to select these templates is shown in Supplementary Spreadsheet 1. + +<|ref|>table<|/ref|><|det|>[[201, 150, 795, 900]]<|/det|> +
CIReceptor familyUniProtIUPHARInactive PDB IDActive PDB IDInactive Ref.Active Ref.
AAdrenoceptorsADA2Bα2B6K4141
ACannabinoidCNR1CB16N4B42
AChemokineCCR6CCR66WWZ43
AFormylpeptideFPR2FPR2/ALX6OMM44
ANeurotensinNTR1NTS16OS945
AAdrenoceptorsADRB2β22RH13SN63637
ADopamineDRD3D33PBL46
AHistamineHRH1H13RZE47
ALysophospholipid (S1P)S1PR1S1P13V2Y48
AOpioidOPRKκ4DJH49
AOpioidOPRMμ4DKL6DDE5051
AOpioidOPRDδ4N6H52
AAcetylcholine (muscarinic)ACM3M34U1553
ALysophospholipid (LPA)LPAR1LPA14Z3654
AAngiotensinAGTR1AT14ZUD55
AAcetylcholine (muscarinic)ACM1M15CXV6OIJ5657
AOpioidOPRXNOP5DHH58
AAcetylcholine (muscarinic)ACM4M45DSG56
AAdenosineAA2ARA2A5NM45G535960
AAdenosineAA1RA15UEN6D9H6162
ADopamineDRD4D45WIU63
AOrexinOX2ROX25WQC64
AEndothelinEDNRBETB5X9365
ANeuropeptide YNPY1RY15ZBQ66
AAcetylcholine (muscarinic)ACM2M25ZKC6OIK6757
ACannabinoidCNR2CB25ZTY6KPF6869
A5-Hydroxytryptamine5HT2A5-HT2A6A9470
A5-Hydroxytryptamine5HT2C5-HT2C6BQH71
AComplement peptideC5AR1C5a16C1R72
ADopamineDRD2D26CM473
ATachykininNK1RNK16HLP74
AGhrelinGHSRGhrelin6KO575
AAdrenoceptorsADA2Cα2C6KUW76
AChemokineCXCR2CXCR26LFL6LFO7777
AAcetylcholine (muscarinic)ACM5M56OL978
AMelatoninMTR1AMT16PS879
AOrexinOX1ROX16TOD80
AVasopressin and oxytocinOXYROT6TPK81
B1CalcitoninCALCRCT6NIY82
B1CalcitoninCALRLCalcitonin-like6UVA83
B1Corticotropin-releasing factorCRFR2CRF26PB184
B1GlucagonSCTRSecretin6WZG85
B1VIP and PACAPPACRPAC16M1I86
B1VIP and PACAPVIPR1VPAC16VN787
B1Corticotropin-releasing factorCRFR1CRF14K5Y6P9X8420
B1GlucagonGLRGlucagon5EE76WPW8821
B1Parathyroid hormonePTH1RPTH16NBH89
B1GlucagonGLP1RGLP-16LN26X181822
CMetabotropic glutamateGRM1mGlu14OR290
CMetabotropic glutamateGRM5mGlu56FFI91
CGABABGABR1GABAB17C7S38
CGABABGABR2GABAB27C7S7C7Q3838
FFrizzledSMOSMO4JKV6XBK3940
FFrizzledFZD4FZD46BD492
FFrizzledFZD5FZD56WW293
+ +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[45, 100, 383, 118]]<|/det|> +# Supplemental Excel spreadsheets + +<|ref|>title<|/ref|><|det|>[[44, 131, 800, 179]]<|/det|> +# Supplementary Spreadsheet 1. GPCR structure annotation to define activation states and representative templates + +<|ref|>text<|/ref|><|det|>[[44, 191, 910, 420]]<|/det|> +Excel file containing an annotation of all 488 GPCR structures released in the Protein Data Bank94 before November 1st 2020. Representative structures of each GPCR and inactive or active state were selected by applying comprehensive filter criteria spanning completeness of receptor and G protein (≥86% and ≥43% of generic residue positions, respectively), sequence identity >90% to human, resolution ≤4.0 Å, degree active (≤20% and ≥90% for inactive and active structures, respectively) and consistent ligand modality and state (inverse agonist/antagonist-inactive and agonist-active). A continuous annotation is provided in the GPCRdb structure browser (http://review.gpcrdb.org/structure) and the structure comparison webserver (Extended Data Fig. 1). + +<|ref|>title<|/ref|><|det|>[[44, 492, 900, 510]]<|/det|> +# Supplementary Spreadsheet 2. Topological and functional mapping of the state determinant positions + +<|ref|>text<|/ref|><|det|>[[44, 521, 909, 660]]<|/det|> +Mapping of state determinant 309 ligand and 103 G protein interacting positions (grey and orange fill, respectively) extracted here from 423 GPCR crystal and cryo- EM structures. Each residue position has been assigned one main role defined by ligand and G protein interacting positions located within the intracellular and extracellular TM helix bundle pockets, respectively, and signal transduction pathway residue bridging such positions on the same helix in at least two classes (Methods). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[92, 100, 171, 116]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[88, 130, 909, 365]]<|/det|> +Structure annotation and selection of representative template dataset. We extended the GPCRdb structures (http://review.gpcrdb.org/structure, all 488 GPCR structures released in the Protein Data Bank95 before November 1st 2020) with additional annotation. We selected a representative structure of each GPCR and inactive or active state by applying comprehensive filter criteria spanning completeness of receptor and G protein (≥83% and ≥43% of generic residue positions, respectively), sequence identity >90% to human, resolution ≤3.7 Å, degree active96 (≤20% and ≥90% for inactive and active structures, respectively) and consistent ligand modality and state96 (inverse agonist/antagonist-inactive and agonist-active). For representative templates of the active state, a G protein complex was required. (Supplementary Spreadsheet 1) + +<|ref|>text<|/ref|><|det|>[[88, 401, 908, 540]]<|/det|> +Transmembrane helix rearrangement analysis. 2D plots for TM1- 7 segment movement at the extracellular end, cytosolic end and membrane mid, respectively, were produced using our new webserver for comparative structure analysis (http://review.gpcrdb.org/structure_comparison/comparative_analysis and 96). Class counts of moving TM1- 7 were calculated as the sum of consensus movements above 1.0 Å (for class C, priority is given to the Gi- GABAB2 over the non- coupled mGluR5 structure). + +<|ref|>text<|/ref|><|det|>[[88, 581, 907, 660]]<|/det|> +Segment (TM1- 7, H8 and loops) contact analysis. Activation- dependent changes in segment networks were determined using 2D network plots of segment contacts96. Segments switches (or helix switches) were defined as the segments with distinct contacts (see next section) in both states. + +<|ref|>text<|/ref|><|det|>[[88, 701, 907, 870]]<|/det|> +Generic residue numbering. Corresponding residue positions in each class were indexed with the structure- based GPCRdb generic residue numbering system36. This builds on the sequence- based generic residue numbering systems for class A (Ballesteros- Weinstein), B1 (Wootten), C (Pin) and F (Wang), but preserves gaps from a structural alignment of two receptors, caused by a unique helix bulge or constriction, in the sequence alignment thereby avoiding offset of such and following residues. All schemes assign residue numbers relative to the most conserved amino acid residue, which is given the number 50, and prefixed with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 98, 907, 178]]<|/det|> +the TM helix number (e.g. 3x32 is on TM3 and 18 positions before the reference). This generic residue numbering scheme also uniquely indexes helix 8 and structurally conserved loop segments, which are numbered by the preceding and following TM helix (e.g. 45 is ECL2 located between TM4- 5). + +<|ref|>text<|/ref|><|det|>[[88, 220, 909, 540]]<|/det|> +State stabilising contact identification. State stabilising contacts were identified using the Structure comparison tool96. The most distinct contact in class A (1x49- 7x50 contact) has 80% higher frequency in the inactive than active state, and the specificity of state- specific contacts depends on the number of members and templates in each class. To obtain comparable number of networks and contacts, we therefore adjusted the class frequency difference thresholds accordingly (no. members and inactive/active state templates); A: 40% (285, 33/14), B1: 67% (15, 3/10), C: 75% (22, 4/1) and F: 100% (11, 3/1). To ensure that the identified determinants are have a wide role in each class, we applied a sequence conservation cut- off. This cut- off requires the amino acids pairs that make up a state specific contact to be conserved, and therefore be able to form, in at least 30% of all receptors in the given GPCR class. The remaining residues, referred to as “state determinants” or “state stabilisers” were further classified into ‘inactivators’, ‘activators’ and ‘switches’ based on if their most frequent contact occurs in inactive, active and both states, respectively. + +<|ref|>text<|/ref|><|det|>[[88, 579, 909, 717]]<|/det|> +State determinant topological mapping in relation to ligand and G protein sites. We mapped state stabilisers to functional sites to ligand and G protein sites using GPCRdb’s comparative structure analysis tool96. For positions with both an allosteric ligand and G protein interaction, precedence is given to the latter in the snakeplot data mapping (but both are counted in Fig. 5). Determinants located in positions with no known ligand or G protein interactions were categorised as “Other” in our analysis. + +<|ref|>sub_title<|/ref|><|det|>[[92, 761, 243, 778]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[90, 792, 907, 840]]<|/det|> +All data are available in GPCRdb (https://review.gpcrdb.org), GitHub (https://github.com/protwis/gpcrdb_data) and Supplementary Spreadsheets 1- 2. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[46, 99, 247, 118]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[44, 130, 909, 272]]<|/det|> +Code availabilityNo code was developed for this manuscript which instead used the existing GPCRdb97 resources, including a new comparative structure analysis platform96. All open-source code can be obtained from GitHub (https://github.com/protovis/protovis) under the permissive Apache 2.0 License (https://www.apache.org/licenses/LICENSE- 2.0). A complete list of the software used for data analysis is available from the Nature Research Reporting Summary. + +<|ref|>sub_title<|/ref|><|det|>[[45, 313, 190, 330]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[145, 342, 905, 387]]<|/det|> +Wu, F. et al. Full- length human GLP- 1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467- 020- 14934- 5 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 397, 905, 442]]<|/det|> +Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656- 668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 451, 905, 497]]<|/det|> +Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 506, 905, 552]]<|/det|> +Zhang, X. et al. Differential GLP- 1R Binding and Activation by Peptide and Non- peptide Agonists. Mol. Cell 80, 485- 500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 561, 905, 607]]<|/det|> +Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22- 31, doi:10.1016/j.tips.2014.11.001 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 616, 905, 661]]<|/det|> +Cherezov, V. et al. High- resolution crystal structure of an engineered human beta2- adrenergic G protein- coupled receptor. Science 318, 1258- 1265, doi:10.1126/science.1150577 (2007). + +<|ref|>text<|/ref|><|det|>[[145, 671, 905, 716]]<|/det|> +Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor- Gs protein complex. Nature 477, 549- 555, doi:10.1038/nature10361 (2011). + +<|ref|>text<|/ref|><|det|>[[145, 726, 905, 771]]<|/det|> +Mao, C. et al. Cryo- EM structures of inactive and active GABAB receptor. Cell Res. 30, 564- 573, doi:10.1038/s41422- 020- 0350- 5 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 781, 905, 826]]<|/det|> +Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338- 343, doi:10.1038/nature12167 (2013). + +<|ref|>text<|/ref|><|det|>[[145, 836, 905, 880]]<|/det|> +Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368- 1375, doi:10.1038/s41589- 020- 0646- 2 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 856]]<|/det|> +1 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 2 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 3 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 4 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 5 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). 6 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 7 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, doi:10.1038/nature10236 (2011). 8 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, doi:10.1126/science.1215904 (2012). 9 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, doi:10.1038/nature10939 (2012). 10 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 11 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, doi:10.1038/s41586-018-0219-7 (2018). 12 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, doi:10.1038/nature12944 (2014). 13 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 14 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 99, 905, 145]]<|/det|> +Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. The Journal of biological chemistry 290, 29127- 29139, doi:10.1074/jbc.M115.689000 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 154, 905, 198]]<|/det|> +Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335- 340, doi:10.1038/nature17188 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 207, 905, 252]]<|/det|> +Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic acetylcholine receptor/G-protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 261, 905, 306]]<|/det|> +Miller, R. L. et al. The Importance of Ligand- Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein- Coupled Receptor. Structure 23, 2291- 2299, doi:10.1016/j.str.2015.07.024 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 315, 905, 360]]<|/det|> +Weinert, T. et al. Serial millisecond crystallography for routine room- temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467- 017- 00630- 4 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 370, 905, 415]]<|/det|> +Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 424, 905, 469]]<|/det|> +Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 479, 905, 523]]<|/det|> +Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 533, 905, 578]]<|/det|> +Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 588, 905, 632]]<|/det|> +Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 642, 905, 686]]<|/det|> +Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 697, 905, 741]]<|/det|> +Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). + +<|ref|>text<|/ref|><|det|>[[145, 752, 905, 797]]<|/det|> +Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). + +<|ref|>text<|/ref|><|det|>[[145, 807, 905, 852]]<|/det|> +Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[48, 99, 910, 884]]<|/det|> +1 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor-Gi Complex Structures. Cell 180, 655-665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 3 70 Kimura, K. T. et al. Structures of the 5-HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121-128, doi:10.1038/s41594-018-0180-z (2019). 5 71 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 7 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472-481, doi:10.1038/s41594-018-0067-z (2018). 9 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269-273, doi:10.1038/nature25758 (2018). 10 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467-018-07939-8 (2019). 11 75 Shiimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467-020-17554-1 (2020). 12 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936-2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 13 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135-140, doi:10.1038/s41586-020-2492-5 (2020). 14 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001-26007, doi:10.1073/pnas.1914446116 (2019). 15 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure-based drug design using X-ray lasers. IUCrJ 6, 1106-1119, doi:10.1107/S2052252519013137 (2019). 16 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X-ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528-1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 17 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 18 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31-51, doi:10.1021/acsptsci.8b00056 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 880]]<|/det|> +1 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key 2 Mechanisms in the Control of Receptor Phenotype by Receptor Activity-Modifying Proteins. ACS Pharmacol 3 Transl Sci 3, 263-284, doi:10.1021/acsptsci.9b00080 (2020). 4 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438-443, 5 doi:10.1038/nature12357 (2013). 6 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 7 4137, doi:10.1038/s41467-020-17791-4 (2020). 8 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, 9 doi:10.1038/s41422-020-0280-2 (2020). 10 87 Duan, J. et al. Cryo-EM structure of an activated VIP1 receptor-G protein complex revealed by a NanoBiT 11 tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467-020-17933-8 (2020). 12 88 Jazayeri, A. et al. Extra-helical binding site of a glucagon receptor antagonist. Nature 533, 274-277, 13 doi:10.1038/nature17414 (2016). 14 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor-1. Science 364, 15 148-153, doi:10.1126/science.aav7942 (2019). 16 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. 17 Science 344, 58-64, doi:10.1126/science.1249489 (2014). 18 91 Christopher, J. A. et al. Structure-Based Optimization Strategies for G Protein-Coupled Receptor (GPCR) 19 Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- 20 ray Structures. Journal of medicinal chemistry 62, 207-222, doi:10.1021/acs.jmedchem.7b01722 (2019). 21 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand-free state. Nature 560, 666-670, 22 doi:10.1038/s41586-018-0447-x (2018). 23 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial-assisted cryo-EM supports a heterodimeric 24 mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 25 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and 26 education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464-D474, 27 doi:10.1093/nar/gky1004 (2019). 28 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids 29 Res. 48, D335-D343, doi:10.1093/nar/gkz990 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 99, 910, 856]]<|/det|> +1 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online GPCR structure analysis platform. Submitted. 3 97 Kooistra, A. J. et al. GPCRdb in 2021: integrating GPCR sequence, structure and function. Nucleic Acids Res. 49, D335-D343, doi:10.1093/nar/gkaa1080 (2021). 5 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). 7 20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 9 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 11 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 13 26 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 15 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 17 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549- 555, doi:10.1038/nature10361 (2011). 19 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 21 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338- 343, doi:10.1038/nature12167 (2013). 23 40 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 25 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 27 42 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448- 458 e412, doi:10.1016/j.cell.2018.11.040 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 95, 910, 856]]<|/det|> +43 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 44 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 45 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, doi:10.1038/nature10236 (2011). 48 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, doi:10.1126/science.1215904 (2012). 49 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, doi:10.1038/nature10939 (2012). 50 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 51 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, doi:10.1038/s41586-018-0219-7 (2018). 52 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, doi:10.1038/nature12944 (2014). 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. The Journal of biological chemistry 290, 29127-29139, doi:10.1074/jbc.M115.689000 (2015). 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335-340, doi:10.1038/nature17188 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 99, 910, 856]]<|/det|> +1 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic acetylcholine receptor/G-protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). 2 58 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein-Coupled Receptor. Structure 23, 2291- 2299, doi:10.1016/j.str.2015.07.024 (2015). 3 59 Weinert, T. et al. Serial millisecond crystallography for routine room- temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467- 017- 00630- 4 (2017). 4 60 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). 5 61 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 6 62 Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). 7 63 Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). 8 64 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). 9 65 Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). 10 66 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). 11 67 Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). 12 68 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 13 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor- Gi Complex Structures. Cell 180, 655- 665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 14 70 Kimura, K. T. et al. Structures of the 5- HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121- 128, doi:10.1038/s41594- 018- 0180- z (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 858]]<|/det|> +1 71 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 3 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 5 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). 6 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467- 018- 07939- 8 (2019). 9 75 Shiimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467- 020- 17554- 1 (2020). 10 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936- 2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 11 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135- 140, doi:10.1038/s41586- 020- 2492- 5 (2020). 12 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001- 26007, doi:10.1073/pnas.1914446116 (2019). 13 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure- based drug design using X- ray lasers. IUcrJ 6, 1106- 1119, doi:10.1107/S2052252519013137 (2019). 14 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X- ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528- 1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 15 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 16 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31- 51, doi:10.1021/acsptsci.8b00056 (2019). 17 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity- Modifying Proteins. ACS Pharmacol Transl Sci 3, 263- 284, doi:10.1021/acstpsci.9b00080 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[48, 100, 910, 856]]<|/det|> +84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438- 443, doi:10.1038/nature12357 (2013). 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467- 020- 17791- 4 (2020). 86 Wang, J. et al. Cryo- EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436- 445, doi:10.1038/s41422- 020- 0280- 2 (2020). 87 Duan, J. et al. Cryo- EM structure of an activated VIP1 receptor- G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467- 020- 17933- 8 (2020). 88 Jazayeri, A. et al. Extra- helical binding site of a glucagon receptor antagonist. Nature 533, 274- 277, doi:10.1038/nature17414 (2016). 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor- 1. Science 364, 148- 153, doi:10.1126/science.aav7942 (2019). 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58- 64, doi:10.1126/science.1249489 (2014). 91 Christopher, J. A. et al. Structure- Based Optimization Strategies for G Protein- Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207- 222, doi:10.1021/acs.jmedchem.7b01722 (2019). 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand- free state. Nature 560, 666- 670, doi:10.1038/s41586- 018- 0447- x (2018). 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial- assisted cryo- EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464- D474, doi:10.1093/nar/gky1004 (2019). 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335- D343, doi:10.1093/nar/gkz990 (2020). 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online platform for comparative GPCR structure analysis. Submitted. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 100, 910, 857]]<|/det|> +97 Kooistra, A. J. et al. GPCRdb in 2021: integrating GPCR sequence, structure and function. Nucleic Acids Res. 49, D335-D343, doi:10.1093/nar/gkaa1080 (2021). 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). 20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 26 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 36 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 37 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 38 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 39 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 40 Qi, X., Friedberg, L., De Bose-Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 41 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 42 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 43 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 856]]<|/det|> +1 44 Zhuang, Y. et al. Structure of formylpeptide receptor 2- Gi complex reveals insights into ligand recognition and 2 signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 3 45 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1- Gi1 complex. Nature 572, 80- 85, 4 doi:10.1038/s41586- 019- 1337- 6 (2019). 5 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. 6 Science 330, 1091- 1095, doi:10.1126/science.1197410 (2010). 7 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65- 70, 8 doi:10.1038/nature10236 (2011). 9 48 Hanson, M. A. et al. Crystal structure of a lipid G protein- coupled receptor. Science 335, 851- 855, 10 doi:10.1126/science.1215904 (2012). 11 49 Wu, H. et al. Structure of the human kappa- opioid receptor in complex with JDTic. Nature 485, 327- 332, 12 doi:10.1038/nature10939 (2012). 13 50 Manglik, A. et al. Crystal structure of the micro- opioid receptor bound to a morphinan antagonist. Nature 485, 14 321- 326, doi:10.1038/nature10954 (2012). 15 51 Koehl, A. et al. Structure of the micro- opioid receptor- Gi protein complex. Nature 558, 547- 552, 16 doi:10.1038/s41586- 018- 0219- 7 (2018). 17 52 Fenalti, G. et al. Molecular control of delta- opioid receptor signalling. Nature 506, 191- 196, 18 doi:10.1038/nature12944 (2014). 19 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G 20 Protein- Coupled Receptor Crystallogenesis. Structure 22, 1657- 1664, doi:10.1016/j.str.2014.08.022 (2014). 21 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 22 161, 1633- 1643, doi:10.1016/j.cell.2015.06.002 (2015). 23 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. 24 The Journal of biological chemistry 290, 29127- 29139, doi:10.1074/jbc.M115.689000 (2015). 25 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335- 340, 26 doi:10.1038/nature17188 (2016). 27 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic 28 acetylcholine receptor/G- protein complexes. Science 364, 552- 557, doi:10.1126/science.aaw5188 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 855]]<|/det|> +1 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the N/OFQ G Protein-Coupled Receptor. Structure 23, 2291-2299, doi:10.1016/j.str.2015.07.024 (2015). 2 Weinert, T. et al. Serial millisecond crystallography for routine room-temperature structure determination at synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467-017-00630-4 (2017). 3 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104-107, doi:10.1038/nature18966 (2016). 4 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867-877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 5 Draper-Joyce, C. J. et al. Structure of the adenosine-bound human adenosine A1 receptor-Gi complex. Nature 558, 559-563, doi:10.1038/s41586-018-0236-6 (2018). 6 Wang, S. et al. D4 dopamine receptor high-resolution structures enable the discovery of selective agonists. Science 358, 381-386, doi:10.1126/science.aan5468 (2017). 7 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype-Selective Antagonist EMPA. Structure 26, 7-19 e15, doi:10.1016/j.str.2017.11.005 (2018). 8 Shihoya, W. et al. X-ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758-764, doi:10.1038/nsmb.3450 (2017). 9 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520-524, doi:10.1038/s41586-018-0046-x (2018). 10 Suno, R. et al. Structural insights into the subtype-selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150-1158, doi:10.1038/s41589-018-0152-y (2018). 11 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459-467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 12 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor-Gi Complex Structures. Cell 180, 655-665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 13 Kimura, K. T. et al. Structures of the 5-HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121-128, doi:10.1038/s41594-018-0180-z (2019). 14 Peng, Y. et al. 5-HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719-730 e714, doi:10.1016/j.cell.2018.01.001 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 858]]<|/det|> +72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467- 018- 07939- 8 (2019). 75 Shiimura, Y. et al. Structure of an antagonist- bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467- 020- 17554- 1 (2020). 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936- 2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135- 140, doi:10.1038/s41586- 020- 2492- 5 (2020). 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001- 26007, doi:10.1073/pnas.1914446116 (2019). 79 Ishchenko, A. et al. Toward G protein- coupled receptor structure- based drug design using X- ray lasers. IUCrJ 6, 1106- 1119, doi:10.1107/S2052252519013137 (2019). 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X- ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528- 1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31- 51, doi:10.1021/acsptsci.8b00056 (2019). 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity- Modifying Proteins. ACS Pharmacol Transl Sci 3, 263- 284, doi:10.1021/acsptsci.9b00080 (2020). 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin- releasing factor receptor 1. Nature 499, 438- 443, doi:10.1038/nature12357 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 98, 910, 860]]<|/det|> +1 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467-020-17791-4 (2020). 3 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, doi:10.1038/s41422-020-0280-2 (2020). 5 87 Duan, J. et al. Cryo-EM structure of an activated VIP1 receptor-G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467-020-17933-8 (2020). 7 88 Jazayeri, A. et al. Extra-helical binding site of a glucagon receptor antagonist. Nature 533, 274-277, doi:10.1038/nature17414 (2016). 9 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor-1. Science 364, 148-153, doi:10.1126/science.aav7942 (2019). 10 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58-64, doi:10.1126/science.1249489 (2014). 13 91 Christopher, J. A. et al. Structure-Based Optimization Strategies for G Protein-Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207-222, doi:10.1021/acs.jmedchem.7b01722 (2019). 16 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand-free state. Nature 560, 666-670, doi:10.1038/s41586-018-0447-x (2018). 18 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial-assisted cryo-EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 20 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464-D474, doi:10.1093/nar/gky1004 (2019). 23 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335-D343, doi:10.1093/nar/gkz990 (2020). 25 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online resource for comparative GPCR structure analysis. Submitted. 27 18 Wu, F. et al. Full-length human GLP-1 receptor structure without orthosteric ligands. Nat Commun 11, 1272, doi:10.1038/s41467-020-14934-5 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 856]]<|/det|> +20 Liang, Y. L. et al. Toward a Structural Understanding of Class B GPCR Peptide Binding and Activation. Mol. Cell 77, 656-668 e655, doi:10.1016/j.molcel.2020.01.012 (2020). 21 Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science 369, eaba3373, doi:10.1126/science.aba3373 (2020). 22 Zhang, X. et al. Differential GLP-1R Binding and Activation by Peptide and Non-peptide Agonists. Mol. Cell 80, 485-500 e487, doi:10.1016/j.molcel.2020.09.020 (2020). 23 Isberg, V. et al. Generic GPCR residue numbers - aligning topology maps while minding the gaps. Trends in pharmacological sciences 36, 22-31, doi:10.1016/j.tips.2014.11.001 (2015). 24 Cherezov, V. et al. High-resolution crystal structure of an engineered human beta2-adrenergic G protein-coupled receptor. Science 318, 1258-1265, doi:10.1126/science.1150577 (2007). 25 Rasmussen, S. G. et al. Crystal structure of the beta2 adrenergic receptor-Gs protein complex. Nature 477, 549-555, doi:10.1038/nature10361 (2011). 26 Mao, C. et al. Cryo-EM structures of inactive and active GABAB receptor. Cell Res. 30, 564-573, doi:10.1038/s41422-020-0350-5 (2020). 27 Wang, C. et al. Structure of the human smoothened receptor bound to an antitumour agent. Nature 497, 338-343, doi:10.1038/nature12167 (2013). 28 Qi, X., Friedberg, L., De Bose- Boyd, R., Long, T. & Li, X. Sterols in an intramolecular channel of Smoothened mediate Hedgehog signaling. Nat. Chem. Biol. 16, 1368-1375, doi:10.1038/s41589-020-0646-2 (2020). 29 Yuan, D. et al. Activation of the alpha2B adrenoceptor by the sedative sympatholytic dexmedetomidine. Nat. Chem. Biol. 16, 507-512, doi:10.1038/s41589-020-0492-2 (2020). 30 Krishna Kumar, K. et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448-458 e412, doi:10.1016/j.cell.2018.11.040 (2019). 31 Wasilko, D. J. et al. Structural basis for chemokine receptor CCR6 activation by the endogenous protein ligand CCL20. Nat Commun 11, 3031, doi:10.1038/s41467-020-16820-6 (2020). 32 Zhuang, Y. et al. Structure of formylpeptide receptor 2-Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467-020-14728-9 (2020). 33 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1-Gi1 complex. Nature 572, 80-85, doi:10.1038/s41586-019-1337-6 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[48, 95, 910, 856]]<|/det|> +1 46 Chien, E. Y. et al. Structure of the human dopamine D3 receptor in complex with a D2/D3 selective antagonist. 2 Science 330, 1091-1095, doi:10.1126/science.1197410 (2010). 3 47 Shimamura, T. et al. Structure of the human histamine H1 receptor complex with doxepin. Nature 475, 65-70, 4 doi:10.1038/nature10236 (2011). 5 48 Hanson, M. A. et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851-855, 6 doi:10.1126/science.1215904 (2012). 7 49 Wu, H. et al. Structure of the human kappa-opioid receptor in complex with JDTic. Nature 485, 327-332, 8 doi:10.1038/nature10939 (2012). 9 50 Manglik, A. et al. Crystal structure of the micro-opioid receptor bound to a morphinan antagonist. Nature 485, 321-326, doi:10.1038/nature10954 (2012). 11 51 Koehl, A. et al. Structure of the micro-opioid receptor-Gi protein complex. Nature 558, 547-552, 12 doi:10.1038/s41586-018-0219-7 (2018). 13 52 Fenalti, G. et al. Molecular control of delta-opioid receptor signalling. Nature 506, 191-196, 14 doi:10.1038/nature12944 (2014). 15 53 Thorsen, T. S., Matt, R., Weis, W. I. & Kobilka, B. K. Modified T4 Lysozyme Fusion Proteins Facilitate G 16 Protein-Coupled Receptor Crystallogenesis. Structure 22, 1657-1664, doi:10.1016/j.str.2014.08.022 (2014). 17 54 Chrencik, J. E. et al. Crystal Structure of Antagonist Bound Human Lysophosphatidic Acid Receptor 1. Cell 18 161, 1633-1643, doi:10.1016/j.cell.2015.06.002 (2015). 19 55 Zhang, H. et al. Structural Basis for Ligand Recognition and Functional Selectivity at Angiotensin Receptor. 20 The Journal of biological chemistry 290, 29127-29139, doi:10.1074/jbc.M115.689000 (2015). 21 56 Thal, D. M. et al. Crystal structures of the M1 and M4 muscarinic acetylcholine receptors. Nature 531, 335-340, 22 doi:10.1038/nature17188 (2016). 23 57 Maeda, S., Qu, Q., Robertson, M. J., Skiniotis, G. & Kobilka, B. K. Structures of the M1 and M2 muscarinic 24 acetylcholine receptor/G-protein complexes. Science 364, 552-557, doi:10.1126/science.aaw5188 (2019). 25 58 Miller, R. L. et al. The Importance of Ligand-Receptor Conformational Pairs in Stabilization: Spotlight on the 26 N/OFQ G Protein-Coupled Receptor. Structure 23, 2291-2299, doi:10.1016/j.str.2015.07.024 (2015). 27 59 Weinert, T. et al. Serial millisecond crystallography for routine room-temperature structure determination at 28 synchrotrons. Nat Commun 8, 542, doi:10.1038/s41467-017-00630-4 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 99, 910, 856]]<|/det|> +1 60 Carpenter, B., Nehme, R., Warne, T., Leslie, A. G. & Tate, C. G. Structure of the adenosine A(2A) receptor bound to an engineered G protein. Nature 536, 104- 107, doi:10.1038/nature18966 (2016). 3 61 Glukhova, A. et al. Structure of the Adenosine A1 Receptor Reveals the Basis for Subtype Selectivity. Cell 168, 867- 877 e813, doi:10.1016/j.cell.2017.01.042 (2017). 5 62 Draper- Joyce, C. J. et al. Structure of the adenosine- bound human adenosine A1 receptor- Gi complex. Nature 558, 559- 563, doi:10.1038/s41586- 018- 0236- 6 (2018). 7 63 Wang, S. et al. D4 dopamine receptor high- resolution structures enable the discovery of selective agonists. Science 358, 381- 386, doi:10.1126/science.aan5468 (2017). 9 64 Suno, R. et al. Crystal Structures of Human Orexin 2 Receptor Bound to the Subtype- Selective Antagonist EMPA. Structure 26, 7- 19 e15, doi:10.1016/j.str.2017.11.005 (2018). 11 65 Shihoya, W. et al. X- ray structures of endothelin ETB receptor bound to clinical antagonist bosentan and its analog. Nat. Struct. Mol. Biol. 24, 758- 764, doi:10.1038/nsmb.3450 (2017). 13 66 Yang, Z. et al. Structural basis of ligand binding modes at the neuropeptide Y Y1 receptor. Nature 556, 520- 524, doi:10.1038/s41586- 018- 0046- x (2018). 15 67 Suno, R. et al. Structural insights into the subtype- selective antagonist binding to the M2 muscarinic receptor. Nat. Chem. Biol. 14, 1150- 1158, doi:10.1038/s41589- 018- 0152- y (2018). 17 68 Li, X. et al. Crystal Structure of the Human Cannabinoid Receptor CB2. Cell 176, 459- 467 e413, doi:10.1016/j.cell.2018.12.011 (2019). 19 69 Hua, T. et al. Activation and Signaling Mechanism Revealed by Cannabinoid Receptor- Gi Complex Structures. Cell 180, 655- 665 e618, doi:10.1016/j.cell.2020.01.008 (2020). 21 70 Kimura, K. T. et al. Structures of the 5- HT2A receptor in complex with the antipsychotics risperidone and zotepine. Nat. Struct. Mol. Biol. 26, 121- 128, doi:10.1038/s41594- 018- 0180- z (2019). 23 71 Peng, Y. et al. 5- HT2C Receptor Structures Reveal the Structural Basis of GPCR Polypharmacology. Cell 172, 719- 730 e714, doi:10.1016/j.cell.2018.01.001 (2018). 25 72 Liu, H. et al. Orthosteric and allosteric action of the C5a receptor antagonists. Nat. Struct. Mol. Biol. 25, 472- 481, doi:10.1038/s41594- 018- 0067- z (2018). 27 73 Wang, S. et al. Structure of the D2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature 555, 269- 273, doi:10.1038/nature25758 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 856]]<|/det|> +1 74 Schoppe, J. et al. Crystal structures of the human neurokinin 1 receptor in complex with clinically used antagonists. Nat Commun 10, 17, doi:10.1038/s41467-018-07939-8 (2019). 2 75 Shimura, Y. et al. Structure of an antagonist-bound ghrelin receptor reveals possible ghrelin recognition mode. Nat Commun 11, 4160, doi:10.1038/s41467-020-17554-1 (2020). 3 76 Chen, X. et al. Molecular Mechanism for Ligand Recognition and Subtype Selectivity of alpha2C Adrenergic Receptor. Cell reports 29, 2936-2943 e2934, doi:10.1016/j.celrep.2019.10.112 (2019). 4 77 Liu, K. et al. Structural basis of CXC chemokine receptor 2 activation and signalling. Nature 585, 135-140, doi:10.1038/s41586-020-2492-5 (2020). 5 78 Vuckovic, Z. et al. Crystal structure of the M5 muscarinic acetylcholine receptor. Proceedings of the National Academy of Sciences of the United States of America 116, 26001-26007, doi:10.1073/pnas.1914446116 (2019). 6 79 Ishchenko, A. et al. Toward G protein-coupled receptor structure-based drug design using X-ray lasers. IUCrJ 6, 1106-1119, doi:10.1107/S2052252519013137 (2019). 7 80 Rappas, M. et al. Comparison of Orexin 1 and Orexin 2 Ligand Binding Modes Using X-ray Crystallography and Computational Analysis. Journal of medicinal chemistry 63, 1528-1543, doi:10.1021/acs.jmedchem.9b01787 (2020). 8 81 Waltenspuhl, Y., Schoppe, J., Ehrenmann, J., Kummer, L. & Pluckthun, A. Crystal structure of the human oxytocin receptor. Sci Adv 6, eabb5419, doi:10.1126/sciadv.abb5419 (2020). 9 82 Dal Maso, E. et al. The Molecular Control of Calcitonin Receptor Signaling. ACS Pharmacol Transl Sci 2, 31-51, doi:10.1021/acsptsci.8b00056 (2019). 10 83 Liang, Y. L. et al. Structure and Dynamics of Adrenomedullin Receptors AM1 and AM2 Reveal Key Mechanisms in the Control of Receptor Phenotype by Receptor Activity-Modifying Proteins. ACS Pharmacol Transl Sci 3, 263-284, doi:10.1021/acsptsci.9b00080 (2020). 11 84 Hollenstein, K. et al. Structure of class B GPCR corticotropin-releasing factor receptor 1. Nature 499, 438-443, doi:10.1038/nature12357 (2013). 12 85 Dong, M. et al. Structure and dynamics of the active Gs-coupled human secretin receptor. Nat Commun 11, 4137, doi:10.1038/s41467-020-17791-4 (2020). 13 86 Wang, J. et al. Cryo-EM structures of PAC1 receptor reveal ligand binding mechanism. Cell Res. 30, 436-445, doi:10.1038/s41422-020-0280-2 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 99, 910, 696]]<|/det|> +1 87 Duan, J. et al. Cryo- EM structure of an activated VIP1 receptor- G protein complex revealed by a NanoBiT tethering strategy. Nat Commun 11, 4121, doi:10.1038/s41467- 020- 17933- 8 (2020). 2 88 Jazayeri, A. et al. Extra- helical binding site of a glucagon receptor antagonist. Nature 533, 274- 277, doi:10.1038/nature17414 (2016). 3 89 Zhao, L. H. et al. Structure and dynamics of the active human parathyroid hormone receptor- 1. Science 364, 148- 153, doi:10.1126/science.aav7942 (2019). 4 90 Wu, H. et al. Structure of a class C GPCR metabotropic glutamate receptor 1 bound to an allosteric modulator. Science 344, 58- 64, doi:10.1126/science.1249489 (2014). 5 91 Christopher, J. A. et al. Structure- Based Optimization Strategies for G Protein- Coupled Receptor (GPCR) Allosteric Modulators: A Case Study from Analyses of New Metabotropic Glutamate Receptor 5 (mGlu5) X- ray Structures. Journal of medicinal chemistry 62, 207- 222, doi:10.1021/acs.jmedchem.7b01722 (2019). 6 92 Yang, S. et al. Crystal structure of the Frizzled 4 receptor in a ligand- free state. Nature 560, 666- 670, doi:10.1038/s41586- 018- 0447- x (2018). 7 93 Tsutsumi, N. et al. Structure of human Frizzled5 by fiducial- assisted cryo- EM supports a heterodimeric mechanism of canonical Wnt signaling. Elife 9, doi:10.7554/eLife.58464 (2020). 8 94 Burley, S. K. et al. RCSB Protein Data Bank: biological macromolecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47, D464- D474, doi:10.1093/nar/gky1004 (2019). 9 95 Armstrong, D. R. et al. PDBe: improved findability of macromolecular structure data in the PDB. Nucleic Acids Res. 48, D335- D343, doi:10.1093/nar/gkz990 (2020). 0 96 Kooistra, A. J., Munk, C., Hauser, A. S. & Gloriam, D. E. An online resource for comparative GPCR structure analysis. Submitted. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 44, 143, 69]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[60, 103, 933, 520]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[56, 555, 936, 680]]<|/det|> +Activation engages all transmembrane helices, helix 8 and loops TM6 and TM3 are universal helix switches 1- 7 residue switches are pivotal changing the state More determinants arise by helix movement or rotation than sidechain rotamer shift Web resource for comparative structure analysis allow uncovering of determinants + +<|ref|>sub_title<|/ref|><|det|>[[43, 717, 115, 736]]<|/det|> +## Figure 1 + +<|ref|>text<|/ref|><|det|>[[42, 758, 942, 825]]<|/det|> +Analysis pipeline for elucidation of GPCR activation mechanisms. Pipeline for analysis of universal and distinct activation macro-/microswitches spanning helix repacking to sidechain rotation and connection to ligand binding, G protein coupling and signal transduction sites (see Methods). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 57, 919, 531]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 560, 118, 580]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[41, 601, 953, 806]]<|/det|> +Transmembrane helix movement upon activation and universal TM3 and TM6 helix 'macroswitches'. a, Movements (A) over 1.0 Å at the extracellular end, membrane mid (determined using35) and intracellular end of the transmembrane helices, TM1- 7 upon comparison of all available receptor inactive- and active- state structure pairs (Extended Data Table 1). Red intensity denotes the number of classes with a consensus movement (for class C, priority is given to the Gi- GABAB2 over the non- coupled mGluR5 structure). b, Movement and conserved hinges of TM6 and the adjacent TM5 and TM7. c, TM3 cytosolic tilt and overall rotation. b- c, GPCR class representative inactive/active receptor structure pairs: A: β236,37, B1: GLP- 118,22, C: GABAB238 and F: Smoothened39,40 receptors. Proline and glycine residues that increase helix plasticity are shown along with their percent conservation in the GPCR class. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 42, 652, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 819]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 955, 954]]<|/det|> +GPCR stabilise inactive and active states by rerouting contacts between TM1- 7, H8, ICL1 and ECL2. a, State- specific residue- residue contacts in each GPCR class visualised as lines within representative inactive/active receptor structure pairs (same as in Fig. 2b- c). Numbers indicate the total, inactivating and activating contacts in each GPCR class. b, Contact networks between the seven GPCR transmembrane helices, TM1- 7 and the first intracellular (ICL1) and second extracellular (ECL2) loops (intra- segment + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 45, 955, 315]]<|/det|> +contacts not shown). Line thickness represents the number of classes (topmost) or contact frequency differences between the inactive and active states. Line colour indicates inactivating (blue) and activating (red) contacts. Receptor segments with a magenta border are "switches", i.e. have contacts across both states. Contacts are identified based on a higher frequency (%) in one set of 43 inactive and 25 active state receptors. a- b, The frequency difference thresholds was set according to the structural coverage in each GPCR class' (no. members and inactive/active state templates): A: \(40\%\) (285, 33/14), B1: \(67\%\) (15, 3/10), C: \(75\%\) (22, 4/1) and F: \(100\%\) (11, 3/1) – yielding comparable numbers of determinants and contacts. To ensure that the identified determinants are applicable for throughout each class, we applied a cut- off requiring at least \(30\%\) of all its receptors to contain one of the amino acid pairs observed to form the given state specific contact. Contact definitions are explained in the settings menu of the online 'Comparative structure analysis' tool (https://review.gpcrdb.org/structure_comparison/comparative_analysis). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 50, 800, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 819]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 941, 954]]<|/det|> +State stabilising contact maps and differences at the residue- level 'microswitches'. a, Contact networks visualise the wiring of state determinants from the extracellular (top) to intracellular (bottom) sides. Contact frequency differences between the inactive and active states are shown as varied line thickness and residue rotamers as rotation of the consensus amino acid in analysed structures and its generic residue number. Two- way Venn diagrams depict the number and percentages of inactivator (blue), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 944, 110]]<|/det|> +activator (red) and switch (magenta) state determinant positions. Bar diagrams show their distribution across the TM helices, H8 and loops. b, Comparison of common and unique state determinant positions across all investigated classes. + +<|ref|>image<|/ref|><|det|>[[58, 123, 520, 844]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 869, 118, 888]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 911, 950, 954]]<|/det|> +Residue positions stabilising an inactive and/or active receptor state. GPCR snakeplots mapping the residue positions that form distinct contacts between state determinants classified as inactivators (blue), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 933, 247]]<|/det|> +activators (red) and switches (magenta). Residues are denoted with the consensus amino acid of the investigated receptor structures (Extended Data Table 1) and their generic residue number26. Filled positions map ligand (grey) and G protein (orange) interaction frequency among all GPCRs in the given class that have such data (Supplementary Spreadsheet 2). Border grayscale denotes the frequency of mutations changing the ligand affinity or activity over 5- fold. The label "Mid" within hexagon- shaped positions denote the membrane mid, above and below which the ligand positions are considered orthosteric and allosteric, respectively, in class A, B1 and F. In class C, all ligand interacting positions in the seven transmembrane helices are allosteric, as its endogenous ligands bind solely in the N- terminal domain. + +<--- Page Split ---> diff --git a/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/images_list.json b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..8dea3988692d2c5bdbc62dea0c3016b7ddc37286 --- /dev/null +++ b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Flow chart of the approach to vibrational spectrum assignment. The \"special pair\", consisting of the proton and its two flanking water molecules, is employed as the fundamental moiety for explaining the solvated proton's vibrational spectrum [20]. Utilizing it to directly derive the vibrational vectors for the captured fundamental moiety, the approach comprises four distinct steps. I. Partitioning the MD trajectory into short segments and filtering in those that the interested moiety is well sustained, computing the average structure of the moiety for each collected segment. II. Computing the velocity autocorrelation functions (VACFs) and velocity cross-correlation functions (VCCFs) for each segment. At interested frequencies with prominent spectral signatures, employing the fast Fourier transform (FFT) and its inverse (IFFT), to quantify the amplitudes of all degree of freedom and their mutual phase differences. III. For each segment, casting the calculated amplitudes and phase differences onto the average structure to generate the vibrational vectors. IV. To average the vibrational vectors across segments, a set of vibrational coordinates for each atom are defined for convenience. The segment-specific vibrational vectors are converted from the original Cartesian coordinate system to the vibrational coordinate system, averaged across segments, and casted onto the overall averaged structure. \\(v_{i}(t)\\) and \\(v_{j}(t)\\) denote the velocities evolution of \\(i\\) th and \\(j\\) th degrees of freedom in the moiety.", + "footnote": [], + "bbox": [ + [ + 140, + 66, + 830, + 260 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Simulation and decomposition of the vibrational spectra of the hydrated proton in water. a. The dark blue trace indicates the experimental difference IR spectrum between 0.1 M HCl and water. The red trace indicates the theoretical difference vibrational density of states (VDOS) spectrum between 0.1 M HCl and water. The light blue trace indicates the experimental difference IR spectrum between 4 M HCl and water from ref. [24]. All intensities of the difference spectra have been normalized to the same number of protons. The horizontal dashed line is the zero line for the difference spectra. The simulated difference VDOS spectrum of 0.1 M HCl clearly illustrated most characteristic spectral features of the IR experiment, though the spectral intensity cannot be compared directly. b. The experimental IR spectra of 0.1 M HCl (red trace) and water (dark blue trace). The shaded areas represent the proton stretching bands for Zundel-like and Eigen-like configurations in references [8, 13, 17] with different patterns. c. The average VDOS spectra of the excess proton (red), the flanking water (FW) hydrogens (dark blue), and the bulk water hydrogens (light blue). d. The cyan trace is the same as the red one in (c). The yellow trace is the average VDOS spectrum of proton stretch. Peak 1-6 are the Gaussian peaks obtained by fitting the cyan trace. The blue dashed line is the sum of these Gaussian peaks.", + "footnote": [], + "bbox": [ + [ + 202, + 65, + 767, + 359 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_8.jpg", + "caption": "Fig. 3. Decomposition of the configurational probability density distribution of the \"special pair\" moiety of proton in water. a. The 2D probability density distribution in 0.01 M HCl, which is extracted and mapped onto a coordinate space defined by \\(R_{\\mathrm{O_1H^*}}\\) and \\(R_{\\mathrm{O_2H^*}}\\) , the distances between the excess proton (H\\*) and its two adjacent oxygen atoms (O1 and O2) of flanking water molecules (assuming \\(R_{\\mathrm{O_1H^*}} < R_{\\mathrm{O_2H^*}}\\) ). b. The three 2D Gaussian-type fitting functions are assigned to Zundel-like (blue), Intermediate (yellow) and Eigen-like (green) configurations. The black lines illustrate the contour plot of the sum of the three types. c. The absolute error (AE) distribution of the 2D Gaussian-type fitting, defined as the difference between the probability density distribution and the sum of the three 2D Gaussian-type fitting functions. The unit of root mean square error (RMSE) and AE is \\(\\hat{\\mathrm{A}}^{-2}\\) . Attempts to reduce the number of Gaussian functions lead to a substantial increase in the fitting error, whereas adding more Gaussian functions does not yield significant improvements (see Supplementary Fig. 8).", + "footnote": [], + "bbox": [ + [ + 223, + 62, + 785, + 163 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Schematic illustration of the vibrational vectors of the hydrated proton moiety in water, which are computed for six spectral signatures in the \"proton continuum\".", + "footnote": [], + "bbox": [ + [ + 201, + 61, + 760, + 262 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_10.jpg", + "caption": "Fig. 5. The distributions of the local electric field computed with conditional ensemble average. a-c. The local electric field distributions along collective coordinate \\(q_{1}\\) , for Zundel-like (0-0.13 V/Å), Intermediate (0.13-0.28 V/Å) and Eigen-like (0.28-0.40 V/Å) configurations, respectively. d-f. The local electric field distributions along collective coordinate \\(q_{s}\\) , for Zundel-like, Intermediate and Eigen-like configurations, respectively. The color bar of the scatter points signifies the structural probability densities in units of \\(\\mathrm{A}^{-1}\\) , and it is worth noting that the color bar is uniquely defined for each sub-graph. The local electric field surrounding the excess proton is evaluated using Coulomb's law within the first solvation shell of the \\(\\mathrm{H}_{5}\\mathrm{O}_{2}^{+}\\) moiety, as long-range contributions have minor impact on the internal electric field [31]. The atomic charges utilized in this estimation are derived from the SPC/E water model, which are set as -0.8476 \\(e\\) for oxygen and +0.4238 \\(e\\) for hydrogen [32]. Specifically, \\(E_{1}\\) is defined as the projection of the local electric field onto the \\(\\mathrm{O}_{1} - \\mathrm{H}^{*}\\) bond, where a positive value of \\(E_{1}\\) would stabilize the current configuration by hindering the proton's transfer towards the opposite atom \\(\\mathrm{O}_{2}\\) . The introduction of an additional \"intermediate\" configuration can be necessitated by inspecting the plot of local electric field distribution presented as Supplementary Fig. 10, classifying configurations merely into Zundel-like and Eigen-like cations will result in non-negligible dependence of the field intensity on \\(q_{s}\\) and overlapped electric field between categories of configurations.", + "footnote": [], + "bbox": [ + [ + 223, + 64, + 795, + 263 + ] + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c.mmd b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6c66239ffb8492ac1777cea127474c6f5f20a432 --- /dev/null +++ b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c.mmd @@ -0,0 +1,376 @@ + +# Deciphering the Vibrational Features of Hydrated Proton in Water + +Chungen Liu cgliu@nju.edu.cn + +Institute of Theoretical and Computational Chemistry, Key Laboratory of Mesoscopic Chemistry of the Ministry of Education (MOE), School of Chemistry and Chemical Engineering, Nanjing University https://orcid.org/0000- 0001- 7839- 0799 + +Xuanye Yang Nanjing University https://orcid.org/0009- 0003- 0336- 1354 + +Yu Wu Shanghai Institute of Applied Physics, Chinese Academy of Sciences + +Siyu Deng Computer Science Department, School of Engineering, University of Texas at El Paso + +Ling Liu State Key Laboratory for Artificial Microstructure and Mesoscopic Physics, Frontier Science Center for Nano- optoelectronics and School of Physics, Peking University + +Hongwei Zhao Shanghai Institute of Applied Physics, Chinese Academy of Sciences + +Yi Gao Shanghai Advanced Research Institute https://orcid.org/0000- 0001- 6015- 5694 + +## Article + +Keywords: + +Posted Date: August 16th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4829060/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2025. See the published version at https://doi.org/10.1038/s41467-025-60794-2. + +<--- Page Split ---> + +# 1 Deciphering the Vibrational Features of Hydrated Proton + +## in Water + +3 Xuanye Yang \(^{1}\) , Yu Wu \(^{2,4}\) , Siyu Deng \(^{6}\) , Ling Liu \(^{5*}\) , Hongwei Zhao \(^{2,3*}\) , Yi Gao \(^{3*}\) 4 and Chungen Liu \(^{1*}\) 5 1 Institute of Theoretical and Computational Chemistry, Key Laboratory of 6 Mesoscopic Chemistry of the Ministry of Education (MOE), School of Chemistry 7 and Chemical Engineering, Nanjing University, Nanjing, 210023, China. 8 2 Shanghai Institute of Applied Physics, Chinese Academy of Sciences, Shanghai, 9 201800, China. 10 3 Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai, 11 201210, China. 12 4 University of Chinese Academy of Sciences, Beijing, 100049, China. 13 5 State Key Laboratory for Artificial Microstructure and Mesoscopic Physics, 14 Frontier Science Center for Nano- optoelectronics and School of Physics, Peking 15 University, Beijing, 100871, China. 16 6 Computer Science Department, School of Engineering, University of Texas at El 17 Paso, El Paso, 79968, USA. + +\*Corresponding author(s). E- mail(s): lingliu@pku.edu.cn; zhaohw@sari.ac.cn; gaoyi@sari.ac.cn; cgliu@nju.edu.cn; Contributing authors: 141130122@smail.nju.edu.cn; wuyu@sinap.ac.cn; sdeng@miners.utep.edu; + +## Abstract + +Hydration of proton is the key to understand the acid- base chemistry and biochemical processes, for which the Zundel and Eigen cations have been recognized as the foundation. However, their dominance remains contentious due to the challenge of attributing the infrared signature at \(\sim 1750 \mathrm{cm}^{- 1}\) , stemming from the theoretical dilemma of balancing structural diversity and solvent fluctuations. Herein, we circumvented this obstacle by devising an integrated approach for computing frequency- specific vibrational vectors via inverse Fourier transform of the vibrational density of states. When applied to aqueous acid, it unveiled an additional "Intermediate" configuration, linked to the aforementioned spectral signature, which exhibit a higher population (44%) and longer lifetime (51 fs), compared to Zundel- like (28%, 25 fs) and Eigen- like (28%, 36 fs), benefitting from the local electric field induced by surrounding solvent molecules. This work reshapes the basic understanding of aqueous proton, and offers a universal solution for characterizing transient species in liquids. + +<--- Page Split ---> + +39 that was firstly proposed by Grotthuss more than two centuries ago [1]. According to the modern picture of the Grotthuss mechanism, PT in aqueous solution undergoes ultrafast interconversion between the Eigen and Zundel cations, both of which were suggested as the probable states of the hydrated protons by Wicke [2] and Zundel [3], respectively. While these seminal works settled the foundation of proton dynamics in aqueous solution [4, 5], intensive structural dynamics in the vicinity of the excess proton seemed to blurred the boundary between Zundel- like and Eigen- like configurations, raising sustained dispute on assessing the importance of the two limiting structures in shaping the behavior of hydrated protons [6- 10]. + +Based on the Eigen- Zundel two- structure model, it has been noticed that the experimentally determined relaxation decay of the \(1,750\mathrm{cm}^{- 1}\) signature [11], along with its intricate coupling with other vibrational frequencies [7], cannot be thoroughly interpreted [9, 11]. On the other hand, molecular dynamics (MD) simulations had uncovered a substantial portion of untypical configurations that are intermediate between Zundel and Eigen configurations [4, 5, 9, 12- 16]. By using various specifically designed descriptors to measure the structural deviation from the two limiting structures, many efforts have been made to divide structures into additional classes [8, 9, 15- 17]. However, the unique spectral feature of these intermediating configurations has not been unambiguously captured from the proton- relevant band across \(1,000–3,000\mathrm{cm}^{- 1}\) (as shown in Fig. 2a), raising concerns about the existence of "hidden configurations". + +To decipher the vibrational spectra of hydrated proton in aqueous solution is extremely challenging. Plenty of previous studies simplified the aqueous solution to various static structures of protonated water clusters, and averaging the results to mimic the aqueous spectrum [8- 10, 12, 17- 19]. This approach establishes the explicit structure- spectroscopy relationship, but is typically unable to account for the anharmonic effects caused by the structural dynamics. While the distinctive spectral signatures emerged above \(2,000\mathrm{cm}^{- 1}\) and below \(1,500\mathrm{cm}^{- 1}\) has been unambiguously associated with Eigen- like and Zundel- like configurations, respectively, consensus on interpreting the prominent spectral signature near \(1,750\mathrm{cm}^{- 1}\) is still absent. + +To fully account for the dynamical effect, recent studies simulated the vibrational spectrum through the Fourier transform of the autocorrelation function of dipole moment or velocity of aqueous proton [13, 14]. By simulating the vibrational spectra utilizing short time segments of MD trajectory presenting distinct configurational affiliations, the spectrum corresponds to an ensemble of dynamic structures. Thus, one- to- one assigning the vibrational signatures to specific vibrational modes of distinct structures remains an open problem. To overcome this challenge, we have designed an approach that inversely translates spectral signatures back to frequency- specific vibrational movements and re- establishes the linkage between vibrational characteristics and the underlying structural features by the machine- learning enhanced MD simulations. While the details of the + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Flow chart of the approach to vibrational spectrum assignment. The "special pair", consisting of the proton and its two flanking water molecules, is employed as the fundamental moiety for explaining the solvated proton's vibrational spectrum [20]. Utilizing it to directly derive the vibrational vectors for the captured fundamental moiety, the approach comprises four distinct steps. I. Partitioning the MD trajectory into short segments and filtering in those that the interested moiety is well sustained, computing the average structure of the moiety for each collected segment. II. Computing the velocity autocorrelation functions (VACFs) and velocity cross-correlation functions (VCCFs) for each segment. At interested frequencies with prominent spectral signatures, employing the fast Fourier transform (FFT) and its inverse (IFFT), to quantify the amplitudes of all degree of freedom and their mutual phase differences. III. For each segment, casting the calculated amplitudes and phase differences onto the average structure to generate the vibrational vectors. IV. To average the vibrational vectors across segments, a set of vibrational coordinates for each atom are defined for convenience. The segment-specific vibrational vectors are converted from the original Cartesian coordinate system to the vibrational coordinate system, averaged across segments, and casted onto the overall averaged structure. \(v_{i}(t)\) and \(v_{j}(t)\) denote the velocities evolution of \(i\) th and \(j\) th degrees of freedom in the moiety.
+ +74 integrated numerical scheme are introduced in the Methods section, it is outlined as a flowchart 75 shown in Fig. 1. This approach integrates the merits of Fourier transform- based vibrational spectroscopy, encompassing concurrently anharmonic and dynamical effects [13, 14], 76 and the merit of static cluster- based vibrational analysis methods in generating frequency- specific vibrational vectors [8- 10, 12, 17- 19]. + +79 The theoretical scheme will be combined with the atomic neural network force field (ANNFF) 80 for HCl solutions, which is constructed from the machine learning of the energies and forces com- 81 puted with the density functional theories (DFT) [21]. The constructed ANNFF offers excellent 82 reproduction of the radial distribution functions (RDFs) and the proton diffusion coefficient, pre- 83 sented in Supplementary Figs. 1- 2 and Supplementary Table 1. The computational scheme is highly 84 efficient to allow for implementing nanosecond scale of MD simulations on extremely dilute HCl 85 solutions, with the convenience of eliminating the concentration effect. Based on our vibrational 86 spectrum assignment approach and proton structure decomposition, we have unveiled a distinct 87 "Intermediate" configuration that bridges the Zundel- like and Eigen- like conformations. These 88 structures are characterized by a proton stretching mode frequency of \(1,770 \mathrm{cm}^{- 1}\) and comprise a 89 significant \(44\%\) of the overall distribution, in comparison to the Zundel- like ( \(28\%\) ) and Eigen- like + +<--- Page Split ---> + +90 (28%) configurations. Through further research utilizing local electric field analysis and time correlation function analysis, we have determined that the Intermediate persists for 51 fs, outlasting both the Zundel- like (25 fs) and Eigen- like (36 fs). The enhanced stability of these intermediate configurations is attributed to the stabilizing effects of the local electric field induced by the surrounding hydrogen- bond (HB) network. + +## 95 Results and discussion + +96 Decomposing the vibrational spectrum of hydrated proton. Experimental infrared (IR) spectra of HCl solutions were usually obtained with concentrations above 1 M [6, 7, 22- 24], with inevitable concentration effect which could interfere with the inherent behavior of the excess protons. To diminish this effect, we conduct an attenuated total reflection Fourier- transform infrared (ATR- FTIR) experiment on a series of HCl solutions with concentrations down to 0.01 M. It is found that 0.1 M HCl can still display explicitly the "proton continuum" in the difference spectrum between HCl solution and pure water, which cannot be observed clearly in the case of 0.01 M solution due to the limitation of instrumental signal- to- noise ratio (see Supplementary Fig. 3). As displayed in Fig. 2a, the difference spectrum of 0.1 M HCl solution is essentially similar to that of 4 M solution reported earlier [24]. However, obvious concentration effect can still be witnessed in the difference spectra. The less concentrated solution presents more intensive positive and negative signals around 3,000 cm \(^{- 1}\) , which was interpreted as resulting from the chloride ion's weaker ability to share the proton than water oxygen atoms [14]. Accordingly, the following theoretical exploration of hydrated proton will be based on MD simulations of 0.01 M HCl solution, in order to minimize the impact of the chloride anion (see the brief discussion on the influence of chloride ion in the Supplementary Information, which based on Supplementary Figs. 4- 5). The concentration effect will be discussion briefly in the Supplementary Information, which based on Supplementary Figs. 6- 7, as well as Supplementary Table 2. + +For a better understanding of the spectral features in IR spectrum, we first separate spectral features of the excess proton and hydrogen atoms in its flanking waters, as well as hydrogen atoms in bulk water. Similarity between the spectra of the flanking water hydrogens and bulk water hydrogens is witnessed, though with obvious red- shift of the OH stretching band broadened down to below 2,000 cm \(^{- 1}\) , as shown in Fig. 2c, which was interpreted as the effect of identity exchange between the excess proton and flanking water hydrogen due to "special pair dance" [5, 12], as well as the blue shift of the libration and intermolecular vibration bands below 1,000 cm \(^{- 1}\) [17, 25]. The influence of the excess proton on the spectral feature of its flanking waters is softening the O- H bonds and hindering their libration style of motions, similar to the effects of the external electric field on water [26]. On the other hand, the VDOS spectrum of the excess proton is distinctively + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. Simulation and decomposition of the vibrational spectra of the hydrated proton in water. a. The dark blue trace indicates the experimental difference IR spectrum between 0.1 M HCl and water. The red trace indicates the theoretical difference vibrational density of states (VDOS) spectrum between 0.1 M HCl and water. The light blue trace indicates the experimental difference IR spectrum between 4 M HCl and water from ref. [24]. All intensities of the difference spectra have been normalized to the same number of protons. The horizontal dashed line is the zero line for the difference spectra. The simulated difference VDOS spectrum of 0.1 M HCl clearly illustrated most characteristic spectral features of the IR experiment, though the spectral intensity cannot be compared directly. b. The experimental IR spectra of 0.1 M HCl (red trace) and water (dark blue trace). The shaded areas represent the proton stretching bands for Zundel-like and Eigen-like configurations in references [8, 13, 17] with different patterns. c. The average VDOS spectra of the excess proton (red), the flanking water (FW) hydrogens (dark blue), and the bulk water hydrogens (light blue). d. The cyan trace is the same as the red one in (c). The yellow trace is the average VDOS spectrum of proton stretch. Peak 1-6 are the Gaussian peaks obtained by fitting the cyan trace. The blue dashed line is the sum of these Gaussian peaks.
+ +124 different from the others, presenting clearly the broad continuum across the whole mid- infrared region, the most important feature in the difference spectrum. + +125 region, the most important feature in the difference spectrum. 126 By defining proton stretch as the oscillatory motion of the proton between the two oxygen atoms within the flanking water molecules, it becomes straightforward to isolate the partial VDOS spectrum of proton stretch. Subsequently, this partial spectrum is fitted to three Gaussian- type peaks, as illustrated in Fig. 2d. The residual obtained after subtracting the stretching motion from the total proton VDOS spectrum is then fitted with three additional Gaussian- type peaks. Overall, the VDOS spectrum of the excess proton, within the frequency range of 500- 4,000 cm\(^{- 1}\), is comprehensively decomposed into six Gaussian- type peaks, achieving a high coefficient of determination of \(R^{2} = 0.990\) (see Fig. 2d). + +<--- Page Split ---> +![](images/Supplementary_Figure_8.jpg) + +
Fig. 3. Decomposition of the configurational probability density distribution of the "special pair" moiety of proton in water. a. The 2D probability density distribution in 0.01 M HCl, which is extracted and mapped onto a coordinate space defined by \(R_{\mathrm{O_1H^*}}\) and \(R_{\mathrm{O_2H^*}}\) , the distances between the excess proton (H\*) and its two adjacent oxygen atoms (O1 and O2) of flanking water molecules (assuming \(R_{\mathrm{O_1H^*}} < R_{\mathrm{O_2H^*}}\) ). b. The three 2D Gaussian-type fitting functions are assigned to Zundel-like (blue), Intermediate (yellow) and Eigen-like (green) configurations. The black lines illustrate the contour plot of the sum of the three types. c. The absolute error (AE) distribution of the 2D Gaussian-type fitting, defined as the difference between the probability density distribution and the sum of the three 2D Gaussian-type fitting functions. The unit of root mean square error (RMSE) and AE is \(\hat{\mathrm{A}}^{-2}\) . Attempts to reduce the number of Gaussian functions lead to a substantial increase in the fitting error, whereas adding more Gaussian functions does not yield significant improvements (see Supplementary Fig. 8).
+ +Structure decomposition unveils the "Intermediate". The robust fit of the partial VDOS of proton stretch into three Gaussian functions underscores the necessity of incorporating additional hydrated proton configuration type beyond the conventionally recognized Zundel and Eigen cations. Earlier MD simulations have disclosed that almost half of the configurations in the trajectory possess atypical features, characterized by a moderate degree of asymmetry, thus precluding their classification into either Zundel- like or Eigen- like configurations [4]. The nearly barrierless interconversion between Zundel- like and Eigen- like configurations also indicates the increased probability in finding these atypical configurations, as well as their importance in shaping the vibrational signature in the proton continuum. To explore the hidden linkage between the vibrational spectrum and the strongly fluctuating structure of the excess proton's atomic environment, we take a technical scheme to naturally classify the configurations in the simulated trajectory, which is different from artificially grouping in most previous studies [8–10, 12, 13, 17–19]. + +The distribution presented in Fig. 3a exhibits a significant broadening along the \(R_{\mathrm{O_2H^*}}\) axis, indicating the substantial fluctuations in the HB length associated with proton rattling. A careful error analysis reveals that at least three two- dimensional (2D) Gaussian functions are indispensable for precisely fitting the distribution, achieving an outstanding coefficient of determination of \(R^2 = 0.999\) in Fig. 3c. As shown in Fig. 3b, based on the distinct asymmetries of the proton's position, the decomposed 2D Gaussian distributions can be rationally attributed to Zundel- like, intermediate, and Eigen- like configurations, with weights of 28%, 44%, and 28%, respectively. The populations of the three types are in line with the earlier MD simulation [4], only if the atypical configurations are named as the intermediate class. The existence of the intermediate gains supported from geometrical optimizations of the snapshots from the MD trajectories (details are described in the Supplementary Information). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4. Schematic illustration of the vibrational vectors of the hydrated proton moiety in water, which are computed for six spectral signatures in the "proton continuum".
+ +From a to c, the vibrational vectors for three proton- stretch signatures located in different frequency regions, which are casted onto their corresponding configuration types. From d to f, the vibrational vectors for libration (d) [17], umbrella (e) [8, 17], and bending (f) [13, 18], respectively. The latter three vectors are imposed on the average structure of all configurations, since they are insensitive to the configuration changes. + +The relatively balanced abundance of each configuration type highlights the importance of considering all three types when discussing the properties of the excess proton. To rationally associate these configurations with the three proton stretching signatures in the vibrational spectrum, we calculate the periods of proton stretch for carefully selected segments of MD trajectories belong to distinct types (refer to the Supplementary Information for details). The Zundel- like, intermediate, and Eigen- like configurations exhibited average proton stretching periods of 24.4 fs, 18.8 fs, and 14.0 fs, respectively, equivalent to frequencies of \(1,370 \mathrm{cm}^{- 1}\) , \(1,770 \mathrm{cm}^{- 1}\) , and \(2,380 \mathrm{cm}^{- 1}\) , being consistent with the peaks 1- 3 in Fig. 2d, thereby resolving the controversy regarding the structural assignment of proton stretching frequencies as shown in Fig. 2b. + +Resolving spectral controversy with vibrational vectors. With the established connection between the vibrational signatures and configurations of the excess proton, the vibrational vectors associated with the frequencies at the centers of peaks 1 to 6 are generated and illustrated in Fig. 4. The vectors in Fig. 4a- c clearly indicate the mixing of the proton stretch with other types of vibrational motions, which is evident from the overlap observed among the fitted Gaussian peaks. Additionally, the vectors shown in Fig. 4d- f highlight the non- stretching motions present in the hydrated proton moiety. All of the assignments of these six spectral signatures to specific vibrational modes are listed in Table 1. + +Mixing of the spectral signatures of proton stretch with other motion styles hindered comprehensive understanding of the spectrum and the arrival of a unanimous consensus regarding the assignment of proton dynamics from molecular spectrum. Being one of the major focuses of dispute, a commonly accepted assessment of the spectral features near \(1,750 \mathrm{cm}^{- 1}\) has not been fully + +<--- Page Split ---> + + +Table 1. The assignments of Gaussian vibrational peaks in the average VDOS spectrum of the excess proton. + +
peakfrequency (cm-1)FWHM a (cm-1)vibrational mode
11,250440stretch
21,770510stretch
32,5301,080stretch
4850380libration
51,320240umbrella
61,590190bending
+ +a FWHM stands for "full width at half maxima". + +achieved. Conventionally, both theoretical and experimental studies attributed this spectral signature primarily to the flanking- water bending mode [7, 11, 12, 18, 24]. However, several studies have also indicated the involvement of the proton stretch [8, 13, 17, 19, 27]. The vibrational vectors displayed in Fig. 4b clearly indicate that the signals around \(1,770 \mathrm{cm}^{- 1}\) are mainly contributed from both proton stretch and flanking water bend. Our analysis reveals that both vibrational modes contribute approximately equally to the observed spectral feature. However, since the proton stretch typically exhibits more intensive IR intensity than the flanking- water bend [8], the former mode ought to dominate the IR spectrum at \(1,770 \mathrm{cm}^{- 1}\) . The controversy at \(\sim 1,750 \mathrm{cm}^{- 1}\) might arise due to the accuracy limitation of the semiempirical theories, which sometimes resulted in hundreds of wavenumbers of deviations in predicted vibrational peak positions compared to both ab initio MD (AIMD) trajectory and experimental results [10, 13]. This inconsistency, which undermines the reliability of the vibrational mode assignments, has been essentially avoided in our simulations using machine- leaning neural network force fields from first- principles level of calculations. This finding resolves the long- standing controversy regarding the assignment of this spectral feature and provides a more accurate understanding of the vibrational dynamics in aqueous proton systems. + +The mixed nature of the vibrational signal at \(1,750 \mathrm{cm}^{- 1}\) , particularly, confirming the distinctive contribution from the proton stretching mode of intermediate configurations, offers a new perspective on interpreting state- of- the- art experimental 2D IR spectra for HCl solutions reported by Tokmakoff and colleagues. When assigning the signature solely to the flanking water bending mode of Zundel- like configurations [7, 11, 24], its vibrational relaxation decay should be perfectly fitted with a single exponential similar to water bending [28], contrary to the reported biexponential fitting [11]. By including the proton stretching mode of intermediate configurations, the biexponential fitting can be smoothly interpreted. An additional advantage of introducing this new vibrational feature can also be harvested concerning the interpretation of the cross peaks in 2D IR spectra. The cross peak \((1,200 \mathrm{cm}^{- 1}, 1,750 \mathrm{cm}^{- 1})\) previously attributed to the coupling of proton stretch and flanking water bend in Zundel- like configurations [7, 9] should also involve the contribution of the chemical exchange effect relevant to Zundel- like and intermediate configurations. The ultrafast interconversion between these configurations, which occurs at a timescale of sub- 100 + +<--- Page Split ---> + +fs [5, 6, 23, 29, 30], can be observed when using a waiting time of 150 fs in 2D IR experiments, in contrast to the commonly used picoseconds of waiting time for observing most chemical exchange processes. Notably, incorporating the chemical exchange effect does not contradict the established interpretation based on the experimental observations. For instance, the proton stretching modes of Zundel- like and intermediate configurations can also perfectly rationalize the strong parallel polarization preference observed experimentally [7]. + +Impact of local electric field on proton configurations. Analysis of the local electric field is helpful to exploring how the surrounding water HB network is involved in shaping the behaviour of a hydrated proton. Fig. 5 graphically illustrates the relationship between the local electric field intensity and the collective coordinates, specifically \(q_{1}\) and \(q_{\mathrm{s}}\) , which represent the long (major) and short (minor) axes of 2D Gaussian- type structural distribution functions (see Supplementary Fig. 9 and Supplementary Tables 3- 6), respectively. Fig. 5a- c illustrates, for each category of hydrated proton configuration, how the electric field intensity evolves along \(q_{1}\) . Notably, while all three categories exhibit an enhanced electric field along \(q_{1}\) , the ranges of intensities are distinctly different and essentially non- overlapping. On the other hand, \(q_{\mathrm{s}}\) appears to emphasize the uniformity of configurations within a given configuration, when plotting the electric field intensity along \(q_{\mathrm{s}}\) , essentially horizontal lines are observed for all configurations, as demonstrated in Fig. 5d- f. The two collective variables incline to focusing on either the distinctiveness among categories of configurations or the tendency of gradual change with each category. Their excellent performance highly validates the structure decomposition scheme via 2D Gaussian- type fitting, and more importantly, reveals the feasibility of utilizing \(q_{1}\) as the sole collective coordinate in exploring each category of configurations. + +The current study contributes significantly to the understanding of the mechanisms underlying the remarkable structural diversity of protons in water. Previous research suggested that asymmetries in the HB networks surrounding the \(\mathrm{H_9O_4^+}\) moiety have a substantial influence on \(\delta\) (defined as \(R_{\mathrm{O_2H^+}} - R_{\mathrm{O_1H^+}}\) ) [14], reminding us that the emergence of three distinct types of hydrated proton could potentially stem from a finite set of HB network patterns in the immediate environment. Fluctuations in the HB network can induce local electric fields, which in turn stabilize these critical configurations. Conversely, when the proton cation mismatches with its local environment, the surrounding HB networks tend to promote configurational transformations. As detailed in the Supplementary Information, biexponential fitting of the correlation functions reveals lifetimes of the configurations, which are 25 fs, 51 fs, and 36 fs for the Zundel- like, intermediate, and Eigen- like configurations, respectively (see Supplementary Fig. 11). These ultra- short timescales rationalize the experimentally observed sub- 100 fs process of structural interconversions [5, 6, 23, 29, 30]. + +<--- Page Split ---> +![](images/Supplementary_Figure_10.jpg) + +
Fig. 5. The distributions of the local electric field computed with conditional ensemble average. a-c. The local electric field distributions along collective coordinate \(q_{1}\) , for Zundel-like (0-0.13 V/Å), Intermediate (0.13-0.28 V/Å) and Eigen-like (0.28-0.40 V/Å) configurations, respectively. d-f. The local electric field distributions along collective coordinate \(q_{s}\) , for Zundel-like, Intermediate and Eigen-like configurations, respectively. The color bar of the scatter points signifies the structural probability densities in units of \(\mathrm{A}^{-1}\) , and it is worth noting that the color bar is uniquely defined for each sub-graph. The local electric field surrounding the excess proton is evaluated using Coulomb's law within the first solvation shell of the \(\mathrm{H}_{5}\mathrm{O}_{2}^{+}\) moiety, as long-range contributions have minor impact on the internal electric field [31]. The atomic charges utilized in this estimation are derived from the SPC/E water model, which are set as -0.8476 \(e\) for oxygen and +0.4238 \(e\) for hydrogen [32]. Specifically, \(E_{1}\) is defined as the projection of the local electric field onto the \(\mathrm{O}_{1} - \mathrm{H}^{*}\) bond, where a positive value of \(E_{1}\) would stabilize the current configuration by hindering the proton's transfer towards the opposite atom \(\mathrm{O}_{2}\) . The introduction of an additional "intermediate" configuration can be necessitated by inspecting the plot of local electric field distribution presented as Supplementary Fig. 10, classifying configurations merely into Zundel-like and Eigen-like cations will result in non-negligible dependence of the field intensity on \(q_{s}\) and overlapped electric field between categories of configurations.
+ +## 240 Conclusions + +241 This study has successfully delved into the dynamical behavior of hydrated protons in water 242 through a theoretical exploration that capitalized on the reliability and efficiency of atomic neural 243 network representations of the force field. This approach has allowed us to perform MD simulations 244 on models encompassing thousands of water molecules, effectively minimizing the confounding 245 effects of concentration and enhancing our comprehension of the protons' fundamental character- 246 istics. Furthermore, our proposed IFFT scheme has proven to be an expedient tool for extracting 247 vibrational vectors from MD trajectories, enabling the direct characterization of frequency- specific 248 vibrational features within the broad proton continuum. While the interpretation of these fea- 249 tures remains a subject of ongoing discussions, our findings have noteworthy contributed to this 250 debate. By integrating vibrational spectrum assignment with structural distribution analysis, we 251 have presented compelling evidence for the existence of distinct intermediate configurations of 252 hydrated protons, alongside the established Zundel-like and Eigen-like configurations. Notably, 253 these intermediate configurations exhibit a unique proton stretching peak at 1,770 cm \(^{-1}\) , which is 254 approximately equally mixed with the flanking water bending mode. We hypothesize that the local + +<--- Page Split ---> + +electric field generated by the surrounding HB network in solution plays a pivotal role in stabilizing these intermediate configurations, a phenomenon that is absent in smaller gaseous protonated water clusters. In essence, this study provides a new perspective for understanding the dynamical behavior of hydrated protons in water, laying the groundwork for future investigations into the intricate interactions and structures that govern this seemingly simple but profoundly fascinating system. + +Although the nuclear quantum effects (NQEs) hold recognized importance in the aqueous proton system, we prefer not to consider it when the currently used generalized gradient approximation (GGA) functional satisfactorily replicates the experimental IR spectrum of HCl solution and the proton's diffusion coefficient. Notably, the GGA functional appears to offset errors in electron correlation and NQEs [14, 33]. Previous research also warns that combining the GGA functional with NQEs may lead to notable discrepancies compared to experimental observations [31, 33]. + +## 267 Methods + +## 268 Vibrational spectrum assignment approach + +## 269 The selection of segments + +We began by determining the two hydrogen atoms closest to each oxygen atom in the trajectory, thus identifying the remaining hydrogen atom as the proton. Due to frequent identity switching between an excess proton and a hydrogen atom in flanking water during the simulation, we divided the trajectory into 3 ps segments. We subsequently filtered in 26 segments from the entire 1.2 ns trajectory, each holding a probability greater than 60% of the same hydrogen atom being an excess proton. To highlight the proton's vibrational behaviors, we concentrated on the smallest unit, the proton alongside its two flanking water molecules. + +## 277 The amplitudes and the phase differences for degrees of freedom + +To streamline the mathematical derivation, we abstracted atomic motions into harmonic vibrations represented by sinusoidal functions. However, it's important to note that our approach also contains anharmonic components and other atomic motions. + +## 281 The dimension conversion factor + +For the purpose of maximizing information retention, we picked a reference degree of freedom that exhibits strong relative intensity within the chosen frequency range. The position of this reference degree of freedom, denoted as \(x_{\mathrm{ref}}(t)\) , can be described as + +<--- Page Split ---> + +\[x_{\mathrm{ref}}(t) = x_{\mathrm{ref}}(0) + \sum_{k}^{N_{\mathrm{ref}}}A_{\mathrm{ref}}^{k}\sin (\omega_{\mathrm{ref}}^{k}t + \phi_{\mathrm{ref}}^{k}), \quad (1)\] + +285 where \(x_{\mathrm{ref}}(0)\) is the initial position at time \(t = 0\) , \(N_{\mathrm{ref}}\) represents the total number of vibrational modes encompassed by \(x_{\mathrm{ref}}(t)\) , \(A_{\mathrm{ref}}^{k}\) , \(\omega_{\mathrm{ref}}^{k}\) and \(\phi_{\mathrm{ref}}^{k}\) are the amplitude, frequency and phase of the vibrational mode \(k\) . The mass-weighted velocity \(v_{\mathrm{ref}}(t)\) was calculated as + +\[v_{\mathrm{ref}}(t) = \sqrt{m_{\mathrm{ref}}}\sum_{k}^{N_{\mathrm{ref}}}A_{\mathrm{ref}}^{k}\omega_{\mathrm{ref}}^{k}\cos (\omega_{\mathrm{ref}}^{k}t + \phi_{\mathrm{ref}}^{k}). \quad (2)\] + +288 Therefore, the VACF of the degree of freedom ref can be expressed as + +\[\begin{array}{r l} & {\mathrm{VACF}_{\mathrm{ref}}(t) = \langle v_{\mathrm{ref}}(t)v_{\mathrm{ref}}(0)\rangle}\\ & {\qquad = \lim_{\tau \to \infty}\frac{1}{\tau}\int_{0}^{\tau}v_{\mathrm{ref}}(t_{0} + t)v_{\mathrm{ref}}(t_{0})d t_{0}}\\ & {\qquad = m_{\mathrm{ref}}\sum_{k}^{N_{\mathrm{ref}}}\frac{1}{2} A_{\mathrm{ref}}^{k}{}^{2}\omega_{\mathrm{ref}}^{k}{}^{2}\cos (\omega_{\mathrm{ref}}^{k}t).} \end{array} \quad (3)\] + +289 The \(\mathrm{VACF}_{\mathrm{ref}}(t)\) loses the initial phase \(\phi_{\mathrm{ref}}\) of \(v_{\mathrm{ref}}(t)\) . Note that summation replaced all integrations in our practical computations. By performing the FFT on \(\mathrm{VACF}_{\mathrm{ref}}(t)\) , we obtained the VDOS spectrum \(P_{\mathrm{ref}}(\omega)\) , + +\[\begin{array}{l}{P_{\mathrm{ref}}(\omega) = \int_{-\infty}^{\infty}\mathrm{VACF}_{\mathrm{ref}}(t)e^{-i\omega t}dt}\\ {= m_{\mathrm{ref}}\sum_{k}^{N_{\mathrm{ref}}}\frac{1}{4} A_{\mathrm{ref}}^{k}{}^{2}\omega_{\mathrm{ref}}^{k}{}^{2}[\delta (\omega -\omega_{\mathrm{ref}}^{k}) + \delta (\omega +\omega_{\mathrm{ref}}^{k})].} \end{array} \quad (4)\] + +292 To rescale the VDOS spectrum to the dimension of mass- weighted velocity, we introduced the + +293 dimension conversion factor \(S(\omega)\) , which is defined as + +\[S(\omega) = \left\{ \begin{array}{ll}\frac{1}{2}\sqrt{m_{\mathrm{ref}}} A_{\mathrm{ref}}^{k}\omega_{\mathrm{ref}}^{k}, & \omega = \pm \omega_{\mathrm{ref}}^{k},\\ 0, & \mathrm{else}. \end{array} \right. \quad (5)\] + +294 The motions of all degrees of freedom for selected frequencies + +295 The position \(x_{i}(t)\) of the degree of freedom \(i\) at time \(t\) can be expressed as + +\[x_{i}(t) = x_{i}(0) + \sum_{k}^{N_{i}}A_{i}^{k}\sin (\omega_{i}^{k}t + \phi_{i}^{k}), \quad (6)\] + +296 where \(N_{i}\) is the number of vibrational modes. The mass- weighted velocity \(v_{i}(t)\) was calculated as + +\[v_{i}(t) = \sqrt{m_{i}}\sum_{k}^{N_{i}}A_{i}^{k}\omega_{i}^{k}\cos (\omega_{i}^{k}t + \phi_{i}^{k}). \quad (7)\] + +297 We calculated the velocity correlation function (VCF) between \(v_{i}(t)\) and \(v_{\mathrm{ref}}(t)\) . If \(i\) is ref, it is the + +298 VACF; otherwise, it is the VCCF. + +<--- Page Split ---> + +\[\begin{array}{r l} & {\mathrm{VCF}_{i}(t) = \langle v_{i}(t)v_{\mathrm{ref}}(0)\rangle}\\ & {\qquad = \sqrt{m_{i}m_{\mathrm{ref}}}\sum_{k}^{L}\frac{1}{2} A_{i}^{k}A_{\mathrm{ref}}^{k}\omega^{k^{2}}\cos (\omega^{k}t + \phi_{i}^{k} - \phi_{\mathrm{ref}}^{k}),} \end{array} \quad (8)\] + +299 where \(L\) represents the number of vibrational modes shared by both \(v_{i}(t)\) and \(v_{\mathrm{ref}}(t)\) . The \(\mathrm{VCF}_{i}(t)\) 300 preserves the phase difference between these two velocities. By performing the FFT on \(\mathrm{VCF}_{i}(t)\) , 301 we obtained the \(P_{i}(\omega)\) , + +\[\begin{array}{l}{P_{i}(\omega) = \int_{-\infty}^{\infty}\mathrm{VCF}_{i}(t)e^{-i\omega t}dt}\\ {= \sqrt{m_{i}m_{\mathrm{ref}}}\sum_{k}^{L}\frac{1}{4} A_{i}^{k}A_{\mathrm{ref}}^{k}\omega^{k^{2}}[e^{i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\delta (\omega -\omega^{k}) + e^{-i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\theta (\omega +\omega^{k})].} \end{array} \quad (9)\] + +302 For convenience, the VDOS spectra were rescaled to the dimension of mass- weighted velocity. We 303 divided \(P_{i}(\omega)\) by the dimension conversion factor \(S(\omega)\) whenever \(P_{i}(\omega)\) is non- zero, + +\[P_{i}^{\mathrm{Scale}}(\omega) = \sqrt{m_{i}}\sum_{k}^{L}\frac{1}{2} A_{i}^{k}\omega^{k}[e^{i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\delta (\omega -\omega^{k}) + e^{-i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\theta (\omega +\omega^{k})]. \quad (10)\] + +304 Selecting a specific frequency \(\omega^{s}\) , we obtained \(P_{i}^{s}(\omega)\) by retaining only the value of \(P_{i}^{\mathrm{Scale}}(\omega)\) at 305 \(\pm \omega^{s}\) , and zeroing out all other frequencies. + +\[P_{i}^{s}(\omega) = \frac{1}{2}\sqrt{m_{i}} A_{i}^{s}\omega^{s}[e^{i(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s})}\delta (\omega -\omega^{s}) + e^{-i(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s})}\theta (\omega +\omega^{s})]. \quad (11)\] + +306 By performing the IFFT on \(P_{i}^{s}(\omega)\) , we obtained the velocity \(v_{i}^{s}(t)\) corresponding to the selected 307 frequency \(\omega^{s}\) , + +\[\begin{array}{l}{v_{i}^{s}(t) = \frac{1}{2\pi}\int_{-\infty}^{\infty}P_{i}^{s}(\omega)e^{i\omega t}d\omega}\\ {= \sqrt{m_{i}} A_{i}^{s}\omega^{s}\cos (\omega^{s}t + \phi_{i}^{s} - \phi_{\mathrm{ref}}^{s}).} \end{array} \quad (12)\] + +308 From \(v_{i}^{s}(t)\) , we can straightforwardly determine the amplitude \(A_{i}^{s}\) and the phase difference \(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s}\) 309 for the degree of freedom \(i\) at selected frequency \(\omega^{s}\) . Subsequently, we can associate the amplitudes 310 and phase differences with their corresponding average structure to generate the vibrational vectors. This effective modes analysis based on MD trajectories of small protonated water clusters can 312 provide vibrational vectors for normal modes [34], it cannot fully account for the dynamic effect on 313 the vibrational spectrum of hydrated proton in water. It is noteworthy that the utilization of velocity correlation functions, rather than velocity or position time evolution functions as in previous 315 research [35], significantly enhances the signal- to- noise ratio in the constructed VDOS spectrum. 316 This advancement ensures a more accurate and reliable analysis of the vibrational properties of 317 the system. + +## 318 The vibrational coordinates + +319 To obtain the average vibrational vectors across various segments, we have devised the vibrational 320 coordinates for the \(\mathrm{H}_{5}\mathrm{O}_{2}^{+}\) moiety. For the hydrogen atoms in the flanking water molecules, we + +<--- Page Split ---> + +utilized three unit vectors, one aligned with the O- H axis, another situated within the \(\mathrm{H}_2\mathrm{O}\) plane and perpendicular to the O- H axis, and a third one perpendicular to the \(\mathrm{H}_2\mathrm{O}\) plane. For the oxygen atoms in the flanking water molecules and the proton, our internal coordinates contain three unit vectors as well, one parallel to the O- O axis, one that signifies the mean projection of the two flanking water H- H connections onto the plane orthogonal to the O- O axis, and a third vector perpendicular to the first two. Afterward, we decomposed the vibrational vectors of each segment at the same frequency onto these vibrational coordinates. + +## 328 The average structures for three types of configurations + +For all selected trajectory segments, in cases where \(R_{\mathrm{O_1H^*}}< R_{\mathrm{O_2H^*}}\) , the proton configurations were classified in accordance with the boundary lines shown in Supplementary Fig. 12. Subsequently, the average structure for each configuration type was computed utilizing internal coordinates. + +## 332 MD simulation details + +AIMD simulations were performed utilizing the VASP 5.4.4 package [36, 37]. The valence electron wave functions were expanded in a plane wave basis with an energy cutoff of 400 eV. For the description of core electrons, pseudopotentials were employed using the projector augmented wave (PAW) method [38]. The exchange and correlation energy were computed using the RPBE functional, formulated within the GGA [39, 40]. To incorporate van der Waals (vdW) interactions, the DFT- D3/zero correction method was utilized [41]. The Brillouin zone sampling was simplified to a single \(\Gamma\) point. + +In the case of 0.9 M HCl, our cubic simulation box (measuring 12.42 Å in length) encompassed 63 water molecules, accompanied by one proton and chloride ion pair, resulting in an approximate density of \(1\mathrm{g / cm^3}\) . The convergence criterion for DFT energy and wave function was set to \(10^{- 6}\) eV. By executing three uncorrelated simulations within the canonical ensemble (NVT), maintaining a temperature of 298 K, a pressure of 1 atm, and utilizing a time step of 1 fs, we successfully generated a total of 150 ps trajectories. + +To enhance computational efficiency and extend simulation durations with low ion concentration, we constructed the ANNFFs for HCl solutions [21]. The scalability of ANNFFs is well- established, as they decompose the total energy into the sum of contributions from individual atoms [21]. This allows an ANNFF trained on smaller systems to predict the potential energy surface for larger configurations simply by adding new components to the total energy, provided the atomic environments are adequately represented in the training dataset. The ANNFF based on RPBE- D3 functional has been successfully employed in modeling water dissociation, providing + +<--- Page Split ---> + +RDFs and self- diffusion coefficient for liquid water that show good agreement with experimental results [31]. + +For the training process, we utilized the n2p2 package [42]. Our ANNFFs comprised a series of feedforward neural networks, each featuring two hidden layers with 25 nodes apiece. The local chemical environments of H, O, and Cl atoms were described using a total of 36, 39, and 26 symmetry functions (SFs), respectively, encompassing both radial and angular types, with a cutoff radius of 6 Å. The initial training set contained 1,945 configurations randomly chosen from AIMD trajectories. Following this, preliminary ANNFFs were created and employed in NVT MD simulations using the LAMMPS program equipped with the ANNFF library [21, 43]. Structures exhibiting extrapolation warnings during MD were filtered, recalculated using DFT, and subsequently added to the training set. Several iterations of ANNFF training and configuration selection were conducted until the ANNFFs were ready for ns- scale MD simulations with few extrapolation warnings. The final training set contained 3,259 configurations. For comparison, the test set encompassed all AIMD configurations. Three parallel ANNFFs were trained and tested, exhibiting promising parallelism and accuracy, as detailed in Supplementary Tables 7 and 8. + +By utilizing the first ANNFF listed in Supplementary Table 7, we conducted MD simulations within three cubic simulation boxes of lengths 12.42, 24.84, and 49.67 Å, which contains 63, 511, and 4,095 water molecules, respectively, alongside a proton and chloride ion pair. Following a 0.6 ns equilibration run at 298 K, 1 atm pressure, and with a time step of 0.3 fs, the simulations progressed for an additional 1.2 ns, generating trajectories for further analysis. + +## 373 Experimental details + +## 374 Sample preparation + +The acid solutions were prepared by diluting concentrated HCl aq. (GR, 36.0- 38.0% Sinopharm Chemical Reagent Co., Ltd) with water (Millipore, 18.2 MΩ) to proton concentrations at 0.1 M and 0.01 M. + +## 378 ATR-FTIR measurements + +The IR spectra were recorded on an ATR- FTIR vacuum spectrometer (Bruker VERTEX 80v) equipped with a deuterated triglycine sulfate (DTGS) detector in the range of 650- 4,000 cm \(^{- 1}\) . Each spectrum was averaged over 64 scans with a resolution of 4 cm \(^{- 1}\) . A Platinum ATR unit A 225 with diamond crystal accessory was used. The background spectrum of the bare diamond crystal was recorded and used for the subsequent measurements at room temperature. In all cases a 6 \(\mu\) L of sample solution was placed on the ATR prism. In order to prevent evaporation of the + +<--- Page Split ---> + +385 liquid, the sample was sealed with a lid that was pressed tightly against the platform by a Pressure application device. + +## 387 Data processing + +388 To ensure accurate results, we collected multiple samples of water and HCl solutions and obtained five parallel spectra for each. Afterward, we carefully chose the spectrum sets exhibiting the smallest standard deviation, enabling us to compute the average spectra and the difference spectra between HCl solutions and water. + +## 392 References + +393 [1] de Grothuf, C. J. T. Mémoire sur la décomposition de l'eau et des corps qu'elle tient en dissolution à l'aide de l'électricité galvanique. Ann. Chim. LVIII, 54- 74 (1806). + +395 [2] Wicke, E., Eigen, M. & Ackermann, T. über den Zustand des Protons (Hydroniumions) in wäßriger Lösung. Z. Phys. Chem. 1 (5.6), 340- 364 (1954). + +397 [3] Zundel, G. & Metzger, H. Energiebänder der tunnelnden überschub-protonen in flüssigen säuren. Eine IR-spektroskopische untersuchung der natur der gruppierungen \(\mathrm{H}_5\mathrm{O}_2^+\) . Z. Phys. Chem. 58 (5.6), 225- 245 (1968). + +400 [4] Marx, D., Tuckerman, M. E., Hutter, J. & Parrinello, M. The nature of the hydrated excess proton in water. Nature 397 (6720), 601- 604 (1999). + +402 [5] Markovitch, O. et al. Special pair dance and partner selection: Elementary steps in proton transport in liquid water. J. Phys. Chem. B 112 (31), 9456- 9466 (2008). + +404 [6] Dahms, F., Fingerhut, B. P., Nibbering, E. T., Pines, E. & Elsaesser, T. Large-amplitude transfer motion of hydrated excess protons mapped by ultrafast 2D IR spectroscopy. Science 357 (6350), 491- 495 (2017). + +407 [7] Fournier, J. A., Carpenter, W. B., Lewis, N. H. & Tokmakoff, A. Broadband 2D IR spectroscopy reveals dominant asymmetric \(\mathrm{H}_5\mathrm{O}_2^+\) proton hydration structures in acid solutions. Nat. Chem. 10 (9), 932- 937 (2018). + +410 [8] Yu, Q., Carpenter, W. B., Lewis, N. H., Tokmakoff, A. & Bowman, J. M. High-Level VSCF/VCI Calculations Decode the Vibrational Spectrum of the Aqueous Proton. J. Phys. Chem. B 123 (33), 7214- 7224 (2019). + +<--- Page Split ---> + +[9] Carpenter, W. B. et al. Decoding the 2D IR spectrum of the aqueous proton with high-level VSCF/VCI calculations. J. Chem. Phys. 153 (12), 124506 (2020). + +[10] Calio, P. B., Li, C. & Voth, G. A. Resolving the Structural Debate for the Hydrated Excess Proton in Water. J. Am. Chem. Soc. 143 (44), 18672- 18683 (2021). + +[11] Carpenter, W. B., Fournier, J. A., Lewis, N. H. & Tokmakoff, A. Picosecond proton transfer kinetics in water revealed with ultrafast IR spectroscopy. J. Phys. Chem. B 122 (10), 2792- 2802 (2018). + +[12] Xu, J., Zhang, Y. & Voth, G. A. Infrared spectrum of the hydrated proton in water. J. Phys. Chem. Lett. 2 (2), 81- 86 (2011). + +[13] Kulig, W. & Agmon, N. A 'clusters- in- liquid' method for calculating infrared spectra identifies the proton- transfer mode in acidic aqueous solutions. Nat. Chem. 5 (1), 29- 35 (2013). + +[14] Napoli, J. A., Marsalek, O. & Markland, T. E. Decoding the spectroscopic features and time scales of aqueous proton defects. J. Chem. Phys. 148 (22), 222833 (2018). + +[15] Wang, E., Zhu, B. & Gao, Y. Discontinuous transition between Zundel and Eigen for \(\mathrm{H}_5\mathrm{O}_2^+\) . Chin. Phys. B 29 (8), 083101 (2020). + +[16] Li, C. & Swanson, J. M. Understanding and tracking the excess proton in ab initio simulations; insights from IR spectra. J. Phys. Chem. B 124 (27), 5696- 5708 (2020). + +[17] Biswas, R., Carpenter, W., Fournier, J. A., Voth, G. A. & Tokmakoff, A. IR spectral assignments for the hydrated excess proton in liquid water. J. Chem. Phys. 146 (15), 154507 (2017). + +[18] Kim, J., Schmitt, U. W., Gruetzmacher, J. A., Voth, G. A. & Scherer, N. E. The vibrational spectrum of the hydrated proton: Comparison of experiment, simulation, and normal mode analysis. J. Chem. Phys. 116 (2), 737- 746 (2002). + +[19] Daly Jr, C. A. et al. Decomposition of the experimental Raman and infrared spectra of acidic water into proton, special pair, and counterion contributions. J. Phys. Chem. Lett. 8 (21), 5246- 5252 (2017). + +[20] Schröder, M., Gatti, F., Lauvergnat, D., Meyer, H.- D. & Vendrell, O. The coupling of the hydrated proton to its first solvation shell. Nat. Commun. 13 (1), 6170 (2022). + +<--- Page Split ---> + +[21] Singraber, A., Behler, J. & Dellago, C. Library- based LAMMPS implementation of high- dimensional neural network potentials. J. Chem. Theory Comput. 15 (3), 1827- 1840 (2019) + +[22] Falk, M. & Giguère, P. A. Infrared spectrum of the \(\mathrm{H}_3\mathrm{O}^+\) ion in aqueous solutions. Can. J. Chem. 35 (10), 1195- 1204 (1957). + +[23] Woutersen, S. & Bakker, H. J. Ultrafast vibrational and structural dynamics of the proton in liquid water. Phys. Rev. Lett. 96 (13), 138305 (2006). + +[24] Thamer, M., De Marco, L., Ramasesha, K., Mandal, A. & Tokmakoff, A. Ultrafast 2D IR spectroscopy of the excess proton in liquid water. Science 350 (6256), 78- 82 (2015). + +[25] Liu, J. et al. Towards complete assignment of the infrared spectrum of the protonated water cluster \(\mathrm{H}^{+}(\mathrm{H}_2\mathrm{O})_{21}\) . Nat. Commun. 12 (1), 1- 10 (2021). + +[26] Zhang, Y. & Jiang, B. Universal machine learning for the response of atomistic systems to external fields. Nat. Commun. 14 (1), 6424 (2023). + +[27] Brünig, F. N., Rammler, M., Adams, E. M., Havenith, M. & Netz, R. R. Spectral signatures of excess- proton waiting and transfer- path dynamics in aqueous hydrochloric acid solutions. Nat. Commun. 13 (1), 1- 12 (2022). + +[28] Carpenter, W. B., Fournier, J. A., Biswas, R., Voth, G. A. & Tokmakoff, A. Delocalization and stretch- bend mixing of the HOH bend in liquid water. J. Chem. Phys. 147 (8) (2017). + +[29] Chandra, A., Tuckerman, M. E. & Marx, D. Connecting solvation shell structure to proton transport kinetics in hydrogen- bonded networks via population correlation functions. Phys. Rev. Lett. 99 (14), 145901 (2007). + +[30] Berkelbach, T. C., Lee, H.- S. & Tuckerman, M. E. Concerted hydrogen- bond dynamics in the transport mechanism of the hydrated proton: A first- principles molecular dynamics study. Phys. Rev. Lett. 103 (23), 238302 (2009). + +[31] Liu, L., Tian, Y., Yang, X. & Liu, C. Mechanistic Insights into Water Autoionization through Metadynamics Simulation Enhanced by Machine Learning. Phys. Rev. Lett. 131 (15), 158001 (2023). + +[32] Berendsen, H. J., Grigera, J. R. & Straatsma, T. P. The missing term in effective pair potentials. J. Phys. Chem. 91 (24), 6269- 6271 (1987). + +<--- Page Split ---> + +[33] Atsango, A. O., Morawietz, T., Marsalek, O. & Markland, T. E. Developing machine- learned potentials to simultaneously capture the dynamics of excess protons and hydroxide ions in classical and path integral simulations. J. Chem. Phys. 159 (7) (2023). + +[34] Agostini, F., Vuilleumier, R. & Ciccotti, G. Infrared spectroscopy of small protonated water clusters at room temperature: an effective modes analysis. J. Chem. Phys. 134 (8), 084302 (2011). + +[35] Wang, S. Intrinsic molecular vibration and rigorous vibrational assignment of benzene by first- principles molecular dynamics. Sci. Rep. 10 (1), 17875 (2020). + +[36] Kresse, G. & Hafner, J. Ab initio molecular dynamics for open- shell transition metals. Phys. Rev. B 48 (17), 13115 (1993). + +[37] Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total- energy calculations using a plane- wave basis set. Phys. Rev. B 54 (16), 11169 (1996). + +[38] Blöchl, P. E. Projector augmented- wave method. Phys. Rev. B 50 (24), 17953 (1994). + +[39] Perdew, J. P. et al. Atoms, molecules, solids, and surfaces: Applications of the generalized gradient approximation for exchange and correlation. Phys. Rev. B 46 (11), 6671 (1992). + +[40] Hammer, B., Hansen, L. B. & Nørskov, J. K. Improved adsorption energetics within density- functional theory using revised Perdew- Burke- Ernzerhof functionals. Phys. Rev. B 59 (11), 7413 (1999). + +[41] Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT- D) for the 94 elements H- Pu. J. Chem. Phys. 132 (15), 154104 (2010). + +[42] Singraber, A. et al. CompPhysVienna/n2p2: Version 2.1.4 (2021). + +[43] Thompson, A. P. et al. LAMMPS- a flexible simulation tool for particle- based materials modeling at the atomic, meso, and continuum scales. Comput. Phys. Commun. 108171 (2021). + +## 495 Acknowledgements + +[495 Acknowledgements] This work is supported by the National Key Research and Development Program of China (2023YFA1506902, 2023YFA1506903) and the National Natural Science Foundation of China (22073041, 62075225). We thank the High Performance Computing Center of Nanjing University + +<--- Page Split ---> + +499 for computational resources. CGL and XYY are thankful to Prof. Minghui Yang for constructive suggestions. We would also thank the staff at BL06B beamline of the Shanghai Synchrotron Radiation Facility for their help in data collection. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +structureopt.zip Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c_det.mmd b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..1da4e50c8b7825a1202f1fb78651eba305db8527 --- /dev/null +++ b/preprint/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c/preprint__13c45b352f4324e58c719da7d35b03718b9e3bcac141fff9c42f5e1328c7150c_det.mmd @@ -0,0 +1,514 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 890, 175]]<|/det|> +# Deciphering the Vibrational Features of Hydrated Proton in Water + +<|ref|>text<|/ref|><|det|>[[44, 196, 235, 242]]<|/det|> +Chungen Liu cgliu@nju.edu.cn + +<|ref|>text<|/ref|><|det|>[[44, 268, 941, 333]]<|/det|> +Institute of Theoretical and Computational Chemistry, Key Laboratory of Mesoscopic Chemistry of the Ministry of Education (MOE), School of Chemistry and Chemical Engineering, Nanjing University https://orcid.org/0000- 0001- 7839- 0799 + +<|ref|>text<|/ref|><|det|>[[44, 339, 576, 381]]<|/det|> +Xuanye Yang Nanjing University https://orcid.org/0009- 0003- 0336- 1354 + +<|ref|>text<|/ref|><|det|>[[44, 386, 790, 430]]<|/det|> +Yu Wu Shanghai Institute of Applied Physics, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 433, 790, 475]]<|/det|> +Siyu Deng Computer Science Department, School of Engineering, University of Texas at El Paso + +<|ref|>text<|/ref|><|det|>[[44, 480, 941, 544]]<|/det|> +Ling Liu State Key Laboratory for Artificial Microstructure and Mesoscopic Physics, Frontier Science Center for Nano- optoelectronics and School of Physics, Peking University + +<|ref|>text<|/ref|><|det|>[[44, 548, 704, 590]]<|/det|> +Hongwei Zhao Shanghai Institute of Applied Physics, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 595, 754, 637]]<|/det|> +Yi Gao Shanghai Advanced Research Institute https://orcid.org/0000- 0001- 6015- 5694 + +<|ref|>sub_title<|/ref|><|det|>[[44, 677, 103, 694]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 714, 137, 732]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 752, 321, 771]]<|/det|> +Posted Date: August 16th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 790, 475, 809]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4829060/v1 + +<|ref|>text<|/ref|><|det|>[[44, 828, 914, 870]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 888, 535, 907]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 950, 87]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on July 1st, 2025. See the published version at https://doi.org/10.1038/s41467-025-60794-2. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[100, 142, 822, 167]]<|/det|> +# 1 Deciphering the Vibrational Features of Hydrated Proton + +<|ref|>sub_title<|/ref|><|det|>[[432, 189, 538, 209]]<|/det|> +## in Water + +<|ref|>text<|/ref|><|det|>[[100, 231, 823, 485]]<|/det|> +3 Xuanye Yang \(^{1}\) , Yu Wu \(^{2,4}\) , Siyu Deng \(^{6}\) , Ling Liu \(^{5*}\) , Hongwei Zhao \(^{2,3*}\) , Yi Gao \(^{3*}\) 4 and Chungen Liu \(^{1*}\) 5 1 Institute of Theoretical and Computational Chemistry, Key Laboratory of 6 Mesoscopic Chemistry of the Ministry of Education (MOE), School of Chemistry 7 and Chemical Engineering, Nanjing University, Nanjing, 210023, China. 8 2 Shanghai Institute of Applied Physics, Chinese Academy of Sciences, Shanghai, 9 201800, China. 10 3 Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai, 11 201210, China. 12 4 University of Chinese Academy of Sciences, Beijing, 100049, China. 13 5 State Key Laboratory for Artificial Microstructure and Mesoscopic Physics, 14 Frontier Science Center for Nano- optoelectronics and School of Physics, Peking 15 University, Beijing, 100871, China. 16 6 Computer Science Department, School of Engineering, University of Texas at El 17 Paso, El Paso, 79968, USA. + +<|ref|>text<|/ref|><|det|>[[175, 512, 799, 577]]<|/det|> +\*Corresponding author(s). E- mail(s): lingliu@pku.edu.cn; zhaohw@sari.ac.cn; gaoyi@sari.ac.cn; cgliu@nju.edu.cn; Contributing authors: 141130122@smail.nju.edu.cn; wuyu@sinap.ac.cn; sdeng@miners.utep.edu; + +<|ref|>sub_title<|/ref|><|det|>[[452, 604, 521, 617]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[168, 620, 806, 805]]<|/det|> +Hydration of proton is the key to understand the acid- base chemistry and biochemical processes, for which the Zundel and Eigen cations have been recognized as the foundation. However, their dominance remains contentious due to the challenge of attributing the infrared signature at \(\sim 1750 \mathrm{cm}^{- 1}\) , stemming from the theoretical dilemma of balancing structural diversity and solvent fluctuations. Herein, we circumvented this obstacle by devising an integrated approach for computing frequency- specific vibrational vectors via inverse Fourier transform of the vibrational density of states. When applied to aqueous acid, it unveiled an additional "Intermediate" configuration, linked to the aforementioned spectral signature, which exhibit a higher population (44%) and longer lifetime (51 fs), compared to Zundel- like (28%, 25 fs) and Eigen- like (28%, 36 fs), benefitting from the local electric field induced by surrounding solvent molecules. This work reshapes the basic understanding of aqueous proton, and offers a universal solution for characterizing transient species in liquids. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 61, 872, 250]]<|/det|> +39 that was firstly proposed by Grotthuss more than two centuries ago [1]. According to the modern picture of the Grotthuss mechanism, PT in aqueous solution undergoes ultrafast interconversion between the Eigen and Zundel cations, both of which were suggested as the probable states of the hydrated protons by Wicke [2] and Zundel [3], respectively. While these seminal works settled the foundation of proton dynamics in aqueous solution [4, 5], intensive structural dynamics in the vicinity of the excess proton seemed to blurred the boundary between Zundel- like and Eigen- like configurations, raising sustained dispute on assessing the importance of the two limiting structures in shaping the behavior of hydrated protons [6- 10]. + +<|ref|>text<|/ref|><|det|>[[116, 254, 872, 485]]<|/det|> +Based on the Eigen- Zundel two- structure model, it has been noticed that the experimentally determined relaxation decay of the \(1,750\mathrm{cm}^{- 1}\) signature [11], along with its intricate coupling with other vibrational frequencies [7], cannot be thoroughly interpreted [9, 11]. On the other hand, molecular dynamics (MD) simulations had uncovered a substantial portion of untypical configurations that are intermediate between Zundel and Eigen configurations [4, 5, 9, 12- 16]. By using various specifically designed descriptors to measure the structural deviation from the two limiting structures, many efforts have been made to divide structures into additional classes [8, 9, 15- 17]. However, the unique spectral feature of these intermediating configurations has not been unambiguously captured from the proton- relevant band across \(1,000–3,000\mathrm{cm}^{- 1}\) (as shown in Fig. 2a), raising concerns about the existence of "hidden configurations". + +<|ref|>text<|/ref|><|det|>[[116, 491, 872, 675]]<|/det|> +To decipher the vibrational spectra of hydrated proton in aqueous solution is extremely challenging. Plenty of previous studies simplified the aqueous solution to various static structures of protonated water clusters, and averaging the results to mimic the aqueous spectrum [8- 10, 12, 17- 19]. This approach establishes the explicit structure- spectroscopy relationship, but is typically unable to account for the anharmonic effects caused by the structural dynamics. While the distinctive spectral signatures emerged above \(2,000\mathrm{cm}^{- 1}\) and below \(1,500\mathrm{cm}^{- 1}\) has been unambiguously associated with Eigen- like and Zundel- like configurations, respectively, consensus on interpreting the prominent spectral signature near \(1,750\mathrm{cm}^{- 1}\) is still absent. + +<|ref|>text<|/ref|><|det|>[[116, 681, 872, 890]]<|/det|> +To fully account for the dynamical effect, recent studies simulated the vibrational spectrum through the Fourier transform of the autocorrelation function of dipole moment or velocity of aqueous proton [13, 14]. By simulating the vibrational spectra utilizing short time segments of MD trajectory presenting distinct configurational affiliations, the spectrum corresponds to an ensemble of dynamic structures. Thus, one- to- one assigning the vibrational signatures to specific vibrational modes of distinct structures remains an open problem. To overcome this challenge, we have designed an approach that inversely translates spectral signatures back to frequency- specific vibrational movements and re- establishes the linkage between vibrational characteristics and the underlying structural features by the machine- learning enhanced MD simulations. While the details of the + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 66, 830, 260]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[128, 267, 847, 496]]<|/det|> +
Fig. 1. Flow chart of the approach to vibrational spectrum assignment. The "special pair", consisting of the proton and its two flanking water molecules, is employed as the fundamental moiety for explaining the solvated proton's vibrational spectrum [20]. Utilizing it to directly derive the vibrational vectors for the captured fundamental moiety, the approach comprises four distinct steps. I. Partitioning the MD trajectory into short segments and filtering in those that the interested moiety is well sustained, computing the average structure of the moiety for each collected segment. II. Computing the velocity autocorrelation functions (VACFs) and velocity cross-correlation functions (VCCFs) for each segment. At interested frequencies with prominent spectral signatures, employing the fast Fourier transform (FFT) and its inverse (IFFT), to quantify the amplitudes of all degree of freedom and their mutual phase differences. III. For each segment, casting the calculated amplitudes and phase differences onto the average structure to generate the vibrational vectors. IV. To average the vibrational vectors across segments, a set of vibrational coordinates for each atom are defined for convenience. The segment-specific vibrational vectors are converted from the original Cartesian coordinate system to the vibrational coordinate system, averaged across segments, and casted onto the overall averaged structure. \(v_{i}(t)\) and \(v_{j}(t)\) denote the velocities evolution of \(i\) th and \(j\) th degrees of freedom in the moiety.
+ +<|ref|>text<|/ref|><|det|>[[95, 520, 847, 630]]<|/det|> +74 integrated numerical scheme are introduced in the Methods section, it is outlined as a flowchart 75 shown in Fig. 1. This approach integrates the merits of Fourier transform- based vibrational spectroscopy, encompassing concurrently anharmonic and dynamical effects [13, 14], 76 and the merit of static cluster- based vibrational analysis methods in generating frequency- specific vibrational vectors [8- 10, 12, 17- 19]. + +<|ref|>text<|/ref|><|det|>[[95, 637, 847, 894]]<|/det|> +79 The theoretical scheme will be combined with the atomic neural network force field (ANNFF) 80 for HCl solutions, which is constructed from the machine learning of the energies and forces com- 81 puted with the density functional theories (DFT) [21]. The constructed ANNFF offers excellent 82 reproduction of the radial distribution functions (RDFs) and the proton diffusion coefficient, pre- 83 sented in Supplementary Figs. 1- 2 and Supplementary Table 1. The computational scheme is highly 84 efficient to allow for implementing nanosecond scale of MD simulations on extremely dilute HCl 85 solutions, with the convenience of eliminating the concentration effect. Based on our vibrational 86 spectrum assignment approach and proton structure decomposition, we have unveiled a distinct 87 "Intermediate" configuration that bridges the Zundel- like and Eigen- like conformations. These 88 structures are characterized by a proton stretching mode frequency of \(1,770 \mathrm{cm}^{- 1}\) and comprise a 89 significant \(44\%\) of the overall distribution, in comparison to the Zundel- like ( \(28\%\) ) and Eigen- like + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 64, 870, 177]]<|/det|> +90 (28%) configurations. Through further research utilizing local electric field analysis and time correlation function analysis, we have determined that the Intermediate persists for 51 fs, outlasting both the Zundel- like (25 fs) and Eigen- like (36 fs). The enhanced stability of these intermediate configurations is attributed to the stabilizing effects of the local electric field induced by the surrounding hydrogen- bond (HB) network. + +<|ref|>sub_title<|/ref|><|det|>[[111, 201, 420, 221]]<|/det|> +## 95 Results and discussion + +<|ref|>text<|/ref|><|det|>[[110, 238, 870, 660]]<|/det|> +96 Decomposing the vibrational spectrum of hydrated proton. Experimental infrared (IR) spectra of HCl solutions were usually obtained with concentrations above 1 M [6, 7, 22- 24], with inevitable concentration effect which could interfere with the inherent behavior of the excess protons. To diminish this effect, we conduct an attenuated total reflection Fourier- transform infrared (ATR- FTIR) experiment on a series of HCl solutions with concentrations down to 0.01 M. It is found that 0.1 M HCl can still display explicitly the "proton continuum" in the difference spectrum between HCl solution and pure water, which cannot be observed clearly in the case of 0.01 M solution due to the limitation of instrumental signal- to- noise ratio (see Supplementary Fig. 3). As displayed in Fig. 2a, the difference spectrum of 0.1 M HCl solution is essentially similar to that of 4 M solution reported earlier [24]. However, obvious concentration effect can still be witnessed in the difference spectra. The less concentrated solution presents more intensive positive and negative signals around 3,000 cm \(^{- 1}\) , which was interpreted as resulting from the chloride ion's weaker ability to share the proton than water oxygen atoms [14]. Accordingly, the following theoretical exploration of hydrated proton will be based on MD simulations of 0.01 M HCl solution, in order to minimize the impact of the chloride anion (see the brief discussion on the influence of chloride ion in the Supplementary Information, which based on Supplementary Figs. 4- 5). The concentration effect will be discussion briefly in the Supplementary Information, which based on Supplementary Figs. 6- 7, as well as Supplementary Table 2. + +<|ref|>text<|/ref|><|det|>[[110, 666, 870, 898]]<|/det|> +For a better understanding of the spectral features in IR spectrum, we first separate spectral features of the excess proton and hydrogen atoms in its flanking waters, as well as hydrogen atoms in bulk water. Similarity between the spectra of the flanking water hydrogens and bulk water hydrogens is witnessed, though with obvious red- shift of the OH stretching band broadened down to below 2,000 cm \(^{- 1}\) , as shown in Fig. 2c, which was interpreted as the effect of identity exchange between the excess proton and flanking water hydrogen due to "special pair dance" [5, 12], as well as the blue shift of the libration and intermolecular vibration bands below 1,000 cm \(^{- 1}\) [17, 25]. The influence of the excess proton on the spectral feature of its flanking waters is softening the O- H bonds and hindering their libration style of motions, similar to the effects of the external electric field on water [26]. On the other hand, the VDOS spectrum of the excess proton is distinctively + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[202, 65, 767, 359]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[127, 368, 846, 583]]<|/det|> +
Fig. 2. Simulation and decomposition of the vibrational spectra of the hydrated proton in water. a. The dark blue trace indicates the experimental difference IR spectrum between 0.1 M HCl and water. The red trace indicates the theoretical difference vibrational density of states (VDOS) spectrum between 0.1 M HCl and water. The light blue trace indicates the experimental difference IR spectrum between 4 M HCl and water from ref. [24]. All intensities of the difference spectra have been normalized to the same number of protons. The horizontal dashed line is the zero line for the difference spectra. The simulated difference VDOS spectrum of 0.1 M HCl clearly illustrated most characteristic spectral features of the IR experiment, though the spectral intensity cannot be compared directly. b. The experimental IR spectra of 0.1 M HCl (red trace) and water (dark blue trace). The shaded areas represent the proton stretching bands for Zundel-like and Eigen-like configurations in references [8, 13, 17] with different patterns. c. The average VDOS spectra of the excess proton (red), the flanking water (FW) hydrogens (dark blue), and the bulk water hydrogens (light blue). d. The cyan trace is the same as the red one in (c). The yellow trace is the average VDOS spectrum of proton stretch. Peak 1-6 are the Gaussian peaks obtained by fitting the cyan trace. The blue dashed line is the sum of these Gaussian peaks.
+ +<|ref|>text<|/ref|><|det|>[[88, 607, 845, 643]]<|/det|> +124 different from the others, presenting clearly the broad continuum across the whole mid- infrared region, the most important feature in the difference spectrum. + +<|ref|>text<|/ref|><|det|>[[88, 652, 846, 838]]<|/det|> +125 region, the most important feature in the difference spectrum. 126 By defining proton stretch as the oscillatory motion of the proton between the two oxygen atoms within the flanking water molecules, it becomes straightforward to isolate the partial VDOS spectrum of proton stretch. Subsequently, this partial spectrum is fitted to three Gaussian- type peaks, as illustrated in Fig. 2d. The residual obtained after subtracting the stretching motion from the total proton VDOS spectrum is then fitted with three additional Gaussian- type peaks. Overall, the VDOS spectrum of the excess proton, within the frequency range of 500- 4,000 cm\(^{- 1}\), is comprehensively decomposed into six Gaussian- type peaks, achieving a high coefficient of determination of \(R^{2} = 0.990\) (see Fig. 2d). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[223, 62, 785, 163]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[150, 167, 870, 340]]<|/det|> +
Fig. 3. Decomposition of the configurational probability density distribution of the "special pair" moiety of proton in water. a. The 2D probability density distribution in 0.01 M HCl, which is extracted and mapped onto a coordinate space defined by \(R_{\mathrm{O_1H^*}}\) and \(R_{\mathrm{O_2H^*}}\) , the distances between the excess proton (H\*) and its two adjacent oxygen atoms (O1 and O2) of flanking water molecules (assuming \(R_{\mathrm{O_1H^*}} < R_{\mathrm{O_2H^*}}\) ). b. The three 2D Gaussian-type fitting functions are assigned to Zundel-like (blue), Intermediate (yellow) and Eigen-like (green) configurations. The black lines illustrate the contour plot of the sum of the three types. c. The absolute error (AE) distribution of the 2D Gaussian-type fitting, defined as the difference between the probability density distribution and the sum of the three 2D Gaussian-type fitting functions. The unit of root mean square error (RMSE) and AE is \(\hat{\mathrm{A}}^{-2}\) . Attempts to reduce the number of Gaussian functions lead to a substantial increase in the fitting error, whereas adding more Gaussian functions does not yield significant improvements (see Supplementary Fig. 8).
+ +<|ref|>text<|/ref|><|det|>[[110, 364, 870, 641]]<|/det|> +Structure decomposition unveils the "Intermediate". The robust fit of the partial VDOS of proton stretch into three Gaussian functions underscores the necessity of incorporating additional hydrated proton configuration type beyond the conventionally recognized Zundel and Eigen cations. Earlier MD simulations have disclosed that almost half of the configurations in the trajectory possess atypical features, characterized by a moderate degree of asymmetry, thus precluding their classification into either Zundel- like or Eigen- like configurations [4]. The nearly barrierless interconversion between Zundel- like and Eigen- like configurations also indicates the increased probability in finding these atypical configurations, as well as their importance in shaping the vibrational signature in the proton continuum. To explore the hidden linkage between the vibrational spectrum and the strongly fluctuating structure of the excess proton's atomic environment, we take a technical scheme to naturally classify the configurations in the simulated trajectory, which is different from artificially grouping in most previous studies [8–10, 12, 13, 17–19]. + +<|ref|>text<|/ref|><|det|>[[110, 650, 870, 905]]<|/det|> +The distribution presented in Fig. 3a exhibits a significant broadening along the \(R_{\mathrm{O_2H^*}}\) axis, indicating the substantial fluctuations in the HB length associated with proton rattling. A careful error analysis reveals that at least three two- dimensional (2D) Gaussian functions are indispensable for precisely fitting the distribution, achieving an outstanding coefficient of determination of \(R^2 = 0.999\) in Fig. 3c. As shown in Fig. 3b, based on the distinct asymmetries of the proton's position, the decomposed 2D Gaussian distributions can be rationally attributed to Zundel- like, intermediate, and Eigen- like configurations, with weights of 28%, 44%, and 28%, respectively. The populations of the three types are in line with the earlier MD simulation [4], only if the atypical configurations are named as the intermediate class. The existence of the intermediate gains supported from geometrical optimizations of the snapshots from the MD trajectories (details are described in the Supplementary Information). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[201, 61, 760, 262]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[128, 267, 848, 300]]<|/det|> +
Fig. 4. Schematic illustration of the vibrational vectors of the hydrated proton moiety in water, which are computed for six spectral signatures in the "proton continuum".
+ +<|ref|>text<|/ref|><|det|>[[128, 300, 847, 368]]<|/det|> +From a to c, the vibrational vectors for three proton- stretch signatures located in different frequency regions, which are casted onto their corresponding configuration types. From d to f, the vibrational vectors for libration (d) [17], umbrella (e) [8, 17], and bending (f) [13, 18], respectively. The latter three vectors are imposed on the average structure of all configurations, since they are insensitive to the configuration changes. + +<|ref|>text<|/ref|><|det|>[[88, 393, 848, 600]]<|/det|> +The relatively balanced abundance of each configuration type highlights the importance of considering all three types when discussing the properties of the excess proton. To rationally associate these configurations with the three proton stretching signatures in the vibrational spectrum, we calculate the periods of proton stretch for carefully selected segments of MD trajectories belong to distinct types (refer to the Supplementary Information for details). The Zundel- like, intermediate, and Eigen- like configurations exhibited average proton stretching periods of 24.4 fs, 18.8 fs, and 14.0 fs, respectively, equivalent to frequencies of \(1,370 \mathrm{cm}^{- 1}\) , \(1,770 \mathrm{cm}^{- 1}\) , and \(2,380 \mathrm{cm}^{- 1}\) , being consistent with the peaks 1- 3 in Fig. 2d, thereby resolving the controversy regarding the structural assignment of proton stretching frequencies as shown in Fig. 2b. + +<|ref|>text<|/ref|><|det|>[[88, 607, 848, 789]]<|/det|> +Resolving spectral controversy with vibrational vectors. With the established connection between the vibrational signatures and configurations of the excess proton, the vibrational vectors associated with the frequencies at the centers of peaks 1 to 6 are generated and illustrated in Fig. 4. The vectors in Fig. 4a- c clearly indicate the mixing of the proton stretch with other types of vibrational motions, which is evident from the overlap observed among the fitted Gaussian peaks. Additionally, the vectors shown in Fig. 4d- f highlight the non- stretching motions present in the hydrated proton moiety. All of the assignments of these six spectral signatures to specific vibrational modes are listed in Table 1. + +<|ref|>text<|/ref|><|det|>[[88, 797, 847, 884]]<|/det|> +Mixing of the spectral signatures of proton stretch with other motion styles hindered comprehensive understanding of the spectrum and the arrival of a unanimous consensus regarding the assignment of proton dynamics from molecular spectrum. Being one of the major focuses of dispute, a commonly accepted assessment of the spectral features near \(1,750 \mathrm{cm}^{- 1}\) has not been fully + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[262, 100, 757, 205]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[150, 70, 868, 100]]<|/det|> +Table 1. The assignments of Gaussian vibrational peaks in the average VDOS spectrum of the excess proton. + +
peakfrequency (cm-1)FWHM a (cm-1)vibrational mode
11,250440stretch
21,770510stretch
32,5301,080stretch
4850380libration
51,320240umbrella
61,590190bending
+ +<|ref|>table_footnote<|/ref|><|det|>[[152, 208, 440, 220]]<|/det|> +a FWHM stands for "full width at half maxima". + +<|ref|>text<|/ref|><|det|>[[150, 226, 870, 576]]<|/det|> +achieved. Conventionally, both theoretical and experimental studies attributed this spectral signature primarily to the flanking- water bending mode [7, 11, 12, 18, 24]. However, several studies have also indicated the involvement of the proton stretch [8, 13, 17, 19, 27]. The vibrational vectors displayed in Fig. 4b clearly indicate that the signals around \(1,770 \mathrm{cm}^{- 1}\) are mainly contributed from both proton stretch and flanking water bend. Our analysis reveals that both vibrational modes contribute approximately equally to the observed spectral feature. However, since the proton stretch typically exhibits more intensive IR intensity than the flanking- water bend [8], the former mode ought to dominate the IR spectrum at \(1,770 \mathrm{cm}^{- 1}\) . The controversy at \(\sim 1,750 \mathrm{cm}^{- 1}\) might arise due to the accuracy limitation of the semiempirical theories, which sometimes resulted in hundreds of wavenumbers of deviations in predicted vibrational peak positions compared to both ab initio MD (AIMD) trajectory and experimental results [10, 13]. This inconsistency, which undermines the reliability of the vibrational mode assignments, has been essentially avoided in our simulations using machine- leaning neural network force fields from first- principles level of calculations. This finding resolves the long- standing controversy regarding the assignment of this spectral feature and provides a more accurate understanding of the vibrational dynamics in aqueous proton systems. + +<|ref|>text<|/ref|><|det|>[[150, 583, 870, 885]]<|/det|> +The mixed nature of the vibrational signal at \(1,750 \mathrm{cm}^{- 1}\) , particularly, confirming the distinctive contribution from the proton stretching mode of intermediate configurations, offers a new perspective on interpreting state- of- the- art experimental 2D IR spectra for HCl solutions reported by Tokmakoff and colleagues. When assigning the signature solely to the flanking water bending mode of Zundel- like configurations [7, 11, 24], its vibrational relaxation decay should be perfectly fitted with a single exponential similar to water bending [28], contrary to the reported biexponential fitting [11]. By including the proton stretching mode of intermediate configurations, the biexponential fitting can be smoothly interpreted. An additional advantage of introducing this new vibrational feature can also be harvested concerning the interpretation of the cross peaks in 2D IR spectra. The cross peak \((1,200 \mathrm{cm}^{- 1}, 1,750 \mathrm{cm}^{- 1})\) previously attributed to the coupling of proton stretch and flanking water bend in Zundel- like configurations [7, 9] should also involve the contribution of the chemical exchange effect relevant to Zundel- like and intermediate configurations. The ultrafast interconversion between these configurations, which occurs at a timescale of sub- 100 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[88, 62, 848, 201]]<|/det|> +fs [5, 6, 23, 29, 30], can be observed when using a waiting time of 150 fs in 2D IR experiments, in contrast to the commonly used picoseconds of waiting time for observing most chemical exchange processes. Notably, incorporating the chemical exchange effect does not contradict the established interpretation based on the experimental observations. For instance, the proton stretching modes of Zundel- like and intermediate configurations can also perfectly rationalize the strong parallel polarization preference observed experimentally [7]. + +<|ref|>text<|/ref|><|det|>[[88, 206, 848, 579]]<|/det|> +Impact of local electric field on proton configurations. Analysis of the local electric field is helpful to exploring how the surrounding water HB network is involved in shaping the behaviour of a hydrated proton. Fig. 5 graphically illustrates the relationship between the local electric field intensity and the collective coordinates, specifically \(q_{1}\) and \(q_{\mathrm{s}}\) , which represent the long (major) and short (minor) axes of 2D Gaussian- type structural distribution functions (see Supplementary Fig. 9 and Supplementary Tables 3- 6), respectively. Fig. 5a- c illustrates, for each category of hydrated proton configuration, how the electric field intensity evolves along \(q_{1}\) . Notably, while all three categories exhibit an enhanced electric field along \(q_{1}\) , the ranges of intensities are distinctly different and essentially non- overlapping. On the other hand, \(q_{\mathrm{s}}\) appears to emphasize the uniformity of configurations within a given configuration, when plotting the electric field intensity along \(q_{\mathrm{s}}\) , essentially horizontal lines are observed for all configurations, as demonstrated in Fig. 5d- f. The two collective variables incline to focusing on either the distinctiveness among categories of configurations or the tendency of gradual change with each category. Their excellent performance highly validates the structure decomposition scheme via 2D Gaussian- type fitting, and more importantly, reveals the feasibility of utilizing \(q_{1}\) as the sole collective coordinate in exploring each category of configurations. + +<|ref|>text<|/ref|><|det|>[[88, 586, 848, 865]]<|/det|> +The current study contributes significantly to the understanding of the mechanisms underlying the remarkable structural diversity of protons in water. Previous research suggested that asymmetries in the HB networks surrounding the \(\mathrm{H_9O_4^+}\) moiety have a substantial influence on \(\delta\) (defined as \(R_{\mathrm{O_2H^+}} - R_{\mathrm{O_1H^+}}\) ) [14], reminding us that the emergence of three distinct types of hydrated proton could potentially stem from a finite set of HB network patterns in the immediate environment. Fluctuations in the HB network can induce local electric fields, which in turn stabilize these critical configurations. Conversely, when the proton cation mismatches with its local environment, the surrounding HB networks tend to promote configurational transformations. As detailed in the Supplementary Information, biexponential fitting of the correlation functions reveals lifetimes of the configurations, which are 25 fs, 51 fs, and 36 fs for the Zundel- like, intermediate, and Eigen- like configurations, respectively (see Supplementary Fig. 11). These ultra- short timescales rationalize the experimentally observed sub- 100 fs process of structural interconversions [5, 6, 23, 29, 30]. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[223, 64, 795, 263]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[150, 267, 870, 512]]<|/det|> +
Fig. 5. The distributions of the local electric field computed with conditional ensemble average. a-c. The local electric field distributions along collective coordinate \(q_{1}\) , for Zundel-like (0-0.13 V/Å), Intermediate (0.13-0.28 V/Å) and Eigen-like (0.28-0.40 V/Å) configurations, respectively. d-f. The local electric field distributions along collective coordinate \(q_{s}\) , for Zundel-like, Intermediate and Eigen-like configurations, respectively. The color bar of the scatter points signifies the structural probability densities in units of \(\mathrm{A}^{-1}\) , and it is worth noting that the color bar is uniquely defined for each sub-graph. The local electric field surrounding the excess proton is evaluated using Coulomb's law within the first solvation shell of the \(\mathrm{H}_{5}\mathrm{O}_{2}^{+}\) moiety, as long-range contributions have minor impact on the internal electric field [31]. The atomic charges utilized in this estimation are derived from the SPC/E water model, which are set as -0.8476 \(e\) for oxygen and +0.4238 \(e\) for hydrogen [32]. Specifically, \(E_{1}\) is defined as the projection of the local electric field onto the \(\mathrm{O}_{1} - \mathrm{H}^{*}\) bond, where a positive value of \(E_{1}\) would stabilize the current configuration by hindering the proton's transfer towards the opposite atom \(\mathrm{O}_{2}\) . The introduction of an additional "intermediate" configuration can be necessitated by inspecting the plot of local electric field distribution presented as Supplementary Fig. 10, classifying configurations merely into Zundel-like and Eigen-like cations will result in non-negligible dependence of the field intensity on \(q_{s}\) and overlapped electric field between categories of configurations.
+ +<|ref|>sub_title<|/ref|><|det|>[[100, 534, 293, 554]]<|/det|> +## 240 Conclusions + +<|ref|>text<|/ref|><|det|>[[110, 572, 870, 897]]<|/det|> +241 This study has successfully delved into the dynamical behavior of hydrated protons in water 242 through a theoretical exploration that capitalized on the reliability and efficiency of atomic neural 243 network representations of the force field. This approach has allowed us to perform MD simulations 244 on models encompassing thousands of water molecules, effectively minimizing the confounding 245 effects of concentration and enhancing our comprehension of the protons' fundamental character- 246 istics. Furthermore, our proposed IFFT scheme has proven to be an expedient tool for extracting 247 vibrational vectors from MD trajectories, enabling the direct characterization of frequency- specific 248 vibrational features within the broad proton continuum. While the interpretation of these fea- 249 tures remains a subject of ongoing discussions, our findings have noteworthy contributed to this 250 debate. By integrating vibrational spectrum assignment with structural distribution analysis, we 251 have presented compelling evidence for the existence of distinct intermediate configurations of 252 hydrated protons, alongside the established Zundel-like and Eigen-like configurations. Notably, 253 these intermediate configurations exhibit a unique proton stretching peak at 1,770 cm \(^{-1}\) , which is 254 approximately equally mixed with the flanking water bending mode. We hypothesize that the local + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 63, 848, 201]]<|/det|> +electric field generated by the surrounding HB network in solution plays a pivotal role in stabilizing these intermediate configurations, a phenomenon that is absent in smaller gaseous protonated water clusters. In essence, this study provides a new perspective for understanding the dynamical behavior of hydrated protons in water, laying the groundwork for future investigations into the intricate interactions and structures that govern this seemingly simple but profoundly fascinating system. + +<|ref|>text<|/ref|><|det|>[[85, 207, 848, 343]]<|/det|> +Although the nuclear quantum effects (NQEs) hold recognized importance in the aqueous proton system, we prefer not to consider it when the currently used generalized gradient approximation (GGA) functional satisfactorily replicates the experimental IR spectrum of HCl solution and the proton's diffusion coefficient. Notably, the GGA functional appears to offset errors in electron correlation and NQEs [14, 33]. Previous research also warns that combining the GGA functional with NQEs may lead to notable discrepancies compared to experimental observations [31, 33]. + +<|ref|>sub_title<|/ref|><|det|>[[77, 367, 235, 387]]<|/det|> +## 267 Methods + +<|ref|>sub_title<|/ref|><|det|>[[77, 410, 641, 431]]<|/det|> +## 268 Vibrational spectrum assignment approach + +<|ref|>sub_title<|/ref|><|det|>[[82, 450, 385, 468]]<|/det|> +## 269 The selection of segments + +<|ref|>text<|/ref|><|det|>[[85, 482, 848, 641]]<|/det|> +We began by determining the two hydrogen atoms closest to each oxygen atom in the trajectory, thus identifying the remaining hydrogen atom as the proton. Due to frequent identity switching between an excess proton and a hydrogen atom in flanking water during the simulation, we divided the trajectory into 3 ps segments. We subsequently filtered in 26 segments from the entire 1.2 ns trajectory, each holding a probability greater than 60% of the same hydrogen atom being an excess proton. To highlight the proton's vibrational behaviors, we concentrated on the smallest unit, the proton alongside its two flanking water molecules. + +<|ref|>sub_title<|/ref|><|det|>[[82, 665, 770, 684]]<|/det|> +## 277 The amplitudes and the phase differences for degrees of freedom + +<|ref|>text<|/ref|><|det|>[[85, 698, 848, 761]]<|/det|> +To streamline the mathematical derivation, we abstracted atomic motions into harmonic vibrations represented by sinusoidal functions. However, it's important to note that our approach also contains anharmonic components and other atomic motions. + +<|ref|>sub_title<|/ref|><|det|>[[85, 785, 431, 802]]<|/det|> +## 281 The dimension conversion factor + +<|ref|>text<|/ref|><|det|>[[85, 817, 848, 880]]<|/det|> +For the purpose of maximizing information retention, we picked a reference degree of freedom that exhibits strong relative intensity within the chosen frequency range. The position of this reference degree of freedom, denoted as \(x_{\mathrm{ref}}(t)\) , can be described as + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[338, 85, 866, 106]]<|/det|> +\[x_{\mathrm{ref}}(t) = x_{\mathrm{ref}}(0) + \sum_{k}^{N_{\mathrm{ref}}}A_{\mathrm{ref}}^{k}\sin (\omega_{\mathrm{ref}}^{k}t + \phi_{\mathrm{ref}}^{k}), \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[108, 115, 870, 182]]<|/det|> +285 where \(x_{\mathrm{ref}}(0)\) is the initial position at time \(t = 0\) , \(N_{\mathrm{ref}}\) represents the total number of vibrational modes encompassed by \(x_{\mathrm{ref}}(t)\) , \(A_{\mathrm{ref}}^{k}\) , \(\omega_{\mathrm{ref}}^{k}\) and \(\phi_{\mathrm{ref}}^{k}\) are the amplitude, frequency and phase of the vibrational mode \(k\) . The mass-weighted velocity \(v_{\mathrm{ref}}(t)\) was calculated as + +<|ref|>equation<|/ref|><|det|>[[333, 210, 866, 232]]<|/det|> +\[v_{\mathrm{ref}}(t) = \sqrt{m_{\mathrm{ref}}}\sum_{k}^{N_{\mathrm{ref}}}A_{\mathrm{ref}}^{k}\omega_{\mathrm{ref}}^{k}\cos (\omega_{\mathrm{ref}}^{k}t + \phi_{\mathrm{ref}}^{k}). \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[108, 240, 670, 258]]<|/det|> +288 Therefore, the VACF of the degree of freedom ref can be expressed as + +<|ref|>equation<|/ref|><|det|>[[325, 281, 866, 352]]<|/det|> +\[\begin{array}{r l} & {\mathrm{VACF}_{\mathrm{ref}}(t) = \langle v_{\mathrm{ref}}(t)v_{\mathrm{ref}}(0)\rangle}\\ & {\qquad = \lim_{\tau \to \infty}\frac{1}{\tau}\int_{0}^{\tau}v_{\mathrm{ref}}(t_{0} + t)v_{\mathrm{ref}}(t_{0})d t_{0}}\\ & {\qquad = m_{\mathrm{ref}}\sum_{k}^{N_{\mathrm{ref}}}\frac{1}{2} A_{\mathrm{ref}}^{k}{}^{2}\omega_{\mathrm{ref}}^{k}{}^{2}\cos (\omega_{\mathrm{ref}}^{k}t).} \end{array} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[108, 355, 870, 421]]<|/det|> +289 The \(\mathrm{VACF}_{\mathrm{ref}}(t)\) loses the initial phase \(\phi_{\mathrm{ref}}\) of \(v_{\mathrm{ref}}(t)\) . Note that summation replaced all integrations in our practical computations. By performing the FFT on \(\mathrm{VACF}_{\mathrm{ref}}(t)\) , we obtained the VDOS spectrum \(P_{\mathrm{ref}}(\omega)\) , + +<|ref|>equation<|/ref|><|det|>[[290, 441, 866, 490]]<|/det|> +\[\begin{array}{l}{P_{\mathrm{ref}}(\omega) = \int_{-\infty}^{\infty}\mathrm{VACF}_{\mathrm{ref}}(t)e^{-i\omega t}dt}\\ {= m_{\mathrm{ref}}\sum_{k}^{N_{\mathrm{ref}}}\frac{1}{4} A_{\mathrm{ref}}^{k}{}^{2}\omega_{\mathrm{ref}}^{k}{}^{2}[\delta (\omega -\omega_{\mathrm{ref}}^{k}) + \delta (\omega +\omega_{\mathrm{ref}}^{k})].} \end{array} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[108, 491, 870, 508]]<|/det|> +292 To rescale the VDOS spectrum to the dimension of mass- weighted velocity, we introduced the + +<|ref|>text<|/ref|><|det|>[[108, 515, 547, 531]]<|/det|> +293 dimension conversion factor \(S(\omega)\) , which is defined as + +<|ref|>equation<|/ref|><|det|>[[362, 550, 866, 604]]<|/det|> +\[S(\omega) = \left\{ \begin{array}{ll}\frac{1}{2}\sqrt{m_{\mathrm{ref}}} A_{\mathrm{ref}}^{k}\omega_{\mathrm{ref}}^{k}, & \omega = \pm \omega_{\mathrm{ref}}^{k},\\ 0, & \mathrm{else}. \end{array} \right. \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[108, 615, 725, 632]]<|/det|> +294 The motions of all degrees of freedom for selected frequencies + +<|ref|>text<|/ref|><|det|>[[108, 647, 707, 664]]<|/det|> +295 The position \(x_{i}(t)\) of the degree of freedom \(i\) at time \(t\) can be expressed as + +<|ref|>equation<|/ref|><|det|>[[368, 692, 866, 714]]<|/det|> +\[x_{i}(t) = x_{i}(0) + \sum_{k}^{N_{i}}A_{i}^{k}\sin (\omega_{i}^{k}t + \phi_{i}^{k}), \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[108, 724, 866, 741]]<|/det|> +296 where \(N_{i}\) is the number of vibrational modes. The mass- weighted velocity \(v_{i}(t)\) was calculated as + +<|ref|>equation<|/ref|><|det|>[[368, 770, 866, 791]]<|/det|> +\[v_{i}(t) = \sqrt{m_{i}}\sum_{k}^{N_{i}}A_{i}^{k}\omega_{i}^{k}\cos (\omega_{i}^{k}t + \phi_{i}^{k}). \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[108, 802, 870, 819]]<|/det|> +297 We calculated the velocity correlation function (VCF) between \(v_{i}(t)\) and \(v_{\mathrm{ref}}(t)\) . If \(i\) is ref, it is the + +<|ref|>text<|/ref|><|det|>[[108, 826, 403, 842]]<|/det|> +298 VACF; otherwise, it is the VCCF. + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[270, 80, 844, 125]]<|/det|> +\[\begin{array}{r l} & {\mathrm{VCF}_{i}(t) = \langle v_{i}(t)v_{\mathrm{ref}}(0)\rangle}\\ & {\qquad = \sqrt{m_{i}m_{\mathrm{ref}}}\sum_{k}^{L}\frac{1}{2} A_{i}^{k}A_{\mathrm{ref}}^{k}\omega^{k^{2}}\cos (\omega^{k}t + \phi_{i}^{k} - \phi_{\mathrm{ref}}^{k}),} \end{array} \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[85, 128, 846, 193]]<|/det|> +299 where \(L\) represents the number of vibrational modes shared by both \(v_{i}(t)\) and \(v_{\mathrm{ref}}(t)\) . The \(\mathrm{VCF}_{i}(t)\) 300 preserves the phase difference between these two velocities. By performing the FFT on \(\mathrm{VCF}_{i}(t)\) , 301 we obtained the \(P_{i}(\omega)\) , + +<|ref|>equation<|/ref|><|det|>[[191, 216, 844, 260]]<|/det|> +\[\begin{array}{l}{P_{i}(\omega) = \int_{-\infty}^{\infty}\mathrm{VCF}_{i}(t)e^{-i\omega t}dt}\\ {= \sqrt{m_{i}m_{\mathrm{ref}}}\sum_{k}^{L}\frac{1}{4} A_{i}^{k}A_{\mathrm{ref}}^{k}\omega^{k^{2}}[e^{i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\delta (\omega -\omega^{k}) + e^{-i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\theta (\omega +\omega^{k})].} \end{array} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[85, 263, 846, 304]]<|/det|> +302 For convenience, the VDOS spectra were rescaled to the dimension of mass- weighted velocity. We 303 divided \(P_{i}(\omega)\) by the dimension conversion factor \(S(\omega)\) whenever \(P_{i}(\omega)\) is non- zero, + +<|ref|>equation<|/ref|><|det|>[[214, 333, 844, 355]]<|/det|> +\[P_{i}^{\mathrm{Scale}}(\omega) = \sqrt{m_{i}}\sum_{k}^{L}\frac{1}{2} A_{i}^{k}\omega^{k}[e^{i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\delta (\omega -\omega^{k}) + e^{-i(\phi_{i}^{k} - \phi_{\mathrm{ref}}^{k})}\theta (\omega +\omega^{k})]. \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[85, 364, 846, 405]]<|/det|> +304 Selecting a specific frequency \(\omega^{s}\) , we obtained \(P_{i}^{s}(\omega)\) by retaining only the value of \(P_{i}^{\mathrm{Scale}}(\omega)\) at 305 \(\pm \omega^{s}\) , and zeroing out all other frequencies. + +<|ref|>equation<|/ref|><|det|>[[241, 435, 844, 455]]<|/det|> +\[P_{i}^{s}(\omega) = \frac{1}{2}\sqrt{m_{i}} A_{i}^{s}\omega^{s}[e^{i(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s})}\delta (\omega -\omega^{s}) + e^{-i(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s})}\theta (\omega +\omega^{s})]. \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[85, 465, 846, 506]]<|/det|> +306 By performing the IFFT on \(P_{i}^{s}(\omega)\) , we obtained the velocity \(v_{i}^{s}(t)\) corresponding to the selected 307 frequency \(\omega^{s}\) , + +<|ref|>equation<|/ref|><|det|>[[340, 528, 844, 572]]<|/det|> +\[\begin{array}{l}{v_{i}^{s}(t) = \frac{1}{2\pi}\int_{-\infty}^{\infty}P_{i}^{s}(\omega)e^{i\omega t}d\omega}\\ {= \sqrt{m_{i}} A_{i}^{s}\omega^{s}\cos (\omega^{s}t + \phi_{i}^{s} - \phi_{\mathrm{ref}}^{s}).} \end{array} \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[85, 576, 848, 808]]<|/det|> +308 From \(v_{i}^{s}(t)\) , we can straightforwardly determine the amplitude \(A_{i}^{s}\) and the phase difference \(\phi_{i}^{s} - \phi_{\mathrm{ref}}^{s}\) 309 for the degree of freedom \(i\) at selected frequency \(\omega^{s}\) . Subsequently, we can associate the amplitudes 310 and phase differences with their corresponding average structure to generate the vibrational vectors. This effective modes analysis based on MD trajectories of small protonated water clusters can 312 provide vibrational vectors for normal modes [34], it cannot fully account for the dynamic effect on 313 the vibrational spectrum of hydrated proton in water. It is noteworthy that the utilization of velocity correlation functions, rather than velocity or position time evolution functions as in previous 315 research [35], significantly enhances the signal- to- noise ratio in the constructed VDOS spectrum. 316 This advancement ensures a more accurate and reliable analysis of the vibrational properties of 317 the system. + +<|ref|>sub_title<|/ref|><|det|>[[84, 832, 405, 850]]<|/det|> +## 318 The vibrational coordinates + +<|ref|>text<|/ref|><|det|>[[85, 865, 848, 906]]<|/det|> +319 To obtain the average vibrational vectors across various segments, we have devised the vibrational 320 coordinates for the \(\mathrm{H}_{5}\mathrm{O}_{2}^{+}\) moiety. For the hydrogen atoms in the flanking water molecules, we + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 64, 872, 222]]<|/det|> +utilized three unit vectors, one aligned with the O- H axis, another situated within the \(\mathrm{H}_2\mathrm{O}\) plane and perpendicular to the O- H axis, and a third one perpendicular to the \(\mathrm{H}_2\mathrm{O}\) plane. For the oxygen atoms in the flanking water molecules and the proton, our internal coordinates contain three unit vectors as well, one parallel to the O- O axis, one that signifies the mean projection of the two flanking water H- H connections onto the plane orthogonal to the O- O axis, and a third vector perpendicular to the first two. Afterward, we decomposed the vibrational vectors of each segment at the same frequency onto these vibrational coordinates. + +<|ref|>sub_title<|/ref|><|det|>[[105, 246, 707, 264]]<|/det|> +## 328 The average structures for three types of configurations + +<|ref|>text<|/ref|><|det|>[[110, 279, 870, 343]]<|/det|> +For all selected trajectory segments, in cases where \(R_{\mathrm{O_1H^*}}< R_{\mathrm{O_2H^*}}\) , the proton configurations were classified in accordance with the boundary lines shown in Supplementary Fig. 12. Subsequently, the average structure for each configuration type was computed utilizing internal coordinates. + +<|ref|>sub_title<|/ref|><|det|>[[100, 368, 419, 389]]<|/det|> +## 332 MD simulation details + +<|ref|>text<|/ref|><|det|>[[110, 406, 870, 565]]<|/det|> +AIMD simulations were performed utilizing the VASP 5.4.4 package [36, 37]. The valence electron wave functions were expanded in a plane wave basis with an energy cutoff of 400 eV. For the description of core electrons, pseudopotentials were employed using the projector augmented wave (PAW) method [38]. The exchange and correlation energy were computed using the RPBE functional, formulated within the GGA [39, 40]. To incorporate van der Waals (vdW) interactions, the DFT- D3/zero correction method was utilized [41]. The Brillouin zone sampling was simplified to a single \(\Gamma\) point. + +<|ref|>text<|/ref|><|det|>[[110, 572, 870, 707]]<|/det|> +In the case of 0.9 M HCl, our cubic simulation box (measuring 12.42 Å in length) encompassed 63 water molecules, accompanied by one proton and chloride ion pair, resulting in an approximate density of \(1\mathrm{g / cm^3}\) . The convergence criterion for DFT energy and wave function was set to \(10^{- 6}\) eV. By executing three uncorrelated simulations within the canonical ensemble (NVT), maintaining a temperature of 298 K, a pressure of 1 atm, and utilizing a time step of 1 fs, we successfully generated a total of 150 ps trajectories. + +<|ref|>text<|/ref|><|det|>[[110, 715, 870, 875]]<|/det|> +To enhance computational efficiency and extend simulation durations with low ion concentration, we constructed the ANNFFs for HCl solutions [21]. The scalability of ANNFFs is well- established, as they decompose the total energy into the sum of contributions from individual atoms [21]. This allows an ANNFF trained on smaller systems to predict the potential energy surface for larger configurations simply by adding new components to the total energy, provided the atomic environments are adequately represented in the training dataset. The ANNFF based on RPBE- D3 functional has been successfully employed in modeling water dissociation, providing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[87, 63, 848, 103]]<|/det|> +RDFs and self- diffusion coefficient for liquid water that show good agreement with experimental results [31]. + +<|ref|>text<|/ref|><|det|>[[87, 110, 848, 415]]<|/det|> +For the training process, we utilized the n2p2 package [42]. Our ANNFFs comprised a series of feedforward neural networks, each featuring two hidden layers with 25 nodes apiece. The local chemical environments of H, O, and Cl atoms were described using a total of 36, 39, and 26 symmetry functions (SFs), respectively, encompassing both radial and angular types, with a cutoff radius of 6 Å. The initial training set contained 1,945 configurations randomly chosen from AIMD trajectories. Following this, preliminary ANNFFs were created and employed in NVT MD simulations using the LAMMPS program equipped with the ANNFF library [21, 43]. Structures exhibiting extrapolation warnings during MD were filtered, recalculated using DFT, and subsequently added to the training set. Several iterations of ANNFF training and configuration selection were conducted until the ANNFFs were ready for ns- scale MD simulations with few extrapolation warnings. The final training set contained 3,259 configurations. For comparison, the test set encompassed all AIMD configurations. Three parallel ANNFFs were trained and tested, exhibiting promising parallelism and accuracy, as detailed in Supplementary Tables 7 and 8. + +<|ref|>text<|/ref|><|det|>[[87, 421, 848, 532]]<|/det|> +By utilizing the first ANNFF listed in Supplementary Table 7, we conducted MD simulations within three cubic simulation boxes of lengths 12.42, 24.84, and 49.67 Å, which contains 63, 511, and 4,095 water molecules, respectively, alongside a proton and chloride ion pair. Following a 0.6 ns equilibration run at 298 K, 1 atm pressure, and with a time step of 0.3 fs, the simulations progressed for an additional 1.2 ns, generating trajectories for further analysis. + +<|ref|>sub_title<|/ref|><|det|>[[79, 557, 377, 577]]<|/det|> +## 373 Experimental details + +<|ref|>sub_title<|/ref|><|det|>[[83, 598, 325, 615]]<|/det|> +## 374 Sample preparation + +<|ref|>text<|/ref|><|det|>[[87, 629, 848, 692]]<|/det|> +The acid solutions were prepared by diluting concentrated HCl aq. (GR, 36.0- 38.0% Sinopharm Chemical Reagent Co., Ltd) with water (Millipore, 18.2 MΩ) to proton concentrations at 0.1 M and 0.01 M. + +<|ref|>sub_title<|/ref|><|det|>[[83, 717, 390, 735]]<|/det|> +## 378 ATR-FTIR measurements + +<|ref|>text<|/ref|><|det|>[[87, 749, 848, 885]]<|/det|> +The IR spectra were recorded on an ATR- FTIR vacuum spectrometer (Bruker VERTEX 80v) equipped with a deuterated triglycine sulfate (DTGS) detector in the range of 650- 4,000 cm \(^{- 1}\) . Each spectrum was averaged over 64 scans with a resolution of 4 cm \(^{- 1}\) . A Platinum ATR unit A 225 with diamond crystal accessory was used. The background spectrum of the bare diamond crystal was recorded and used for the subsequent measurements at room temperature. In all cases a 6 \(\mu\) L of sample solution was placed on the ATR prism. In order to prevent evaporation of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 63, 869, 105]]<|/det|> +385 liquid, the sample was sealed with a lid that was pressed tightly against the platform by a Pressure application device. + +<|ref|>sub_title<|/ref|><|det|>[[105, 128, 312, 147]]<|/det|> +## 387 Data processing + +<|ref|>text<|/ref|><|det|>[[108, 160, 870, 249]]<|/det|> +388 To ensure accurate results, we collected multiple samples of water and HCl solutions and obtained five parallel spectra for each. Afterward, we carefully chose the spectrum sets exhibiting the smallest standard deviation, enabling us to compute the average spectra and the difference spectra between HCl solutions and water. + +<|ref|>sub_title<|/ref|><|det|>[[99, 273, 280, 293]]<|/det|> +## 392 References + +<|ref|>text<|/ref|><|det|>[[108, 310, 870, 352]]<|/det|> +393 [1] de Grothuf, C. J. T. Mémoire sur la décomposition de l'eau et des corps qu'elle tient en dissolution à l'aide de l'électricité galvanique. Ann. Chim. LVIII, 54- 74 (1806). + +<|ref|>text<|/ref|><|det|>[[108, 369, 870, 411]]<|/det|> +395 [2] Wicke, E., Eigen, M. & Ackermann, T. über den Zustand des Protons (Hydroniumions) in wäßriger Lösung. Z. Phys. Chem. 1 (5.6), 340- 364 (1954). + +<|ref|>text<|/ref|><|det|>[[108, 428, 870, 494]]<|/det|> +397 [3] Zundel, G. & Metzger, H. Energiebänder der tunnelnden überschub-protonen in flüssigen säuren. Eine IR-spektroskopische untersuchung der natur der gruppierungen \(\mathrm{H}_5\mathrm{O}_2^+\) . Z. Phys. Chem. 58 (5.6), 225- 245 (1968). + +<|ref|>text<|/ref|><|det|>[[108, 511, 870, 554]]<|/det|> +400 [4] Marx, D., Tuckerman, M. E., Hutter, J. & Parrinello, M. The nature of the hydrated excess proton in water. Nature 397 (6720), 601- 604 (1999). + +<|ref|>text<|/ref|><|det|>[[108, 571, 870, 614]]<|/det|> +402 [5] Markovitch, O. et al. Special pair dance and partner selection: Elementary steps in proton transport in liquid water. J. Phys. Chem. B 112 (31), 9456- 9466 (2008). + +<|ref|>text<|/ref|><|det|>[[108, 631, 870, 696]]<|/det|> +404 [6] Dahms, F., Fingerhut, B. P., Nibbering, E. T., Pines, E. & Elsaesser, T. Large-amplitude transfer motion of hydrated excess protons mapped by ultrafast 2D IR spectroscopy. Science 357 (6350), 491- 495 (2017). + +<|ref|>text<|/ref|><|det|>[[108, 713, 870, 780]]<|/det|> +407 [7] Fournier, J. A., Carpenter, W. B., Lewis, N. H. & Tokmakoff, A. Broadband 2D IR spectroscopy reveals dominant asymmetric \(\mathrm{H}_5\mathrm{O}_2^+\) proton hydration structures in acid solutions. Nat. Chem. 10 (9), 932- 937 (2018). + +<|ref|>text<|/ref|><|det|>[[108, 797, 870, 862]]<|/det|> +410 [8] Yu, Q., Carpenter, W. B., Lewis, N. H., Tokmakoff, A. & Bowman, J. M. High-Level VSCF/VCI Calculations Decode the Vibrational Spectrum of the Aqueous Proton. J. Phys. Chem. B 123 (33), 7214- 7224 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 60, 849, 105]]<|/det|> +[9] Carpenter, W. B. et al. Decoding the 2D IR spectrum of the aqueous proton with high-level VSCF/VCI calculations. J. Chem. Phys. 153 (12), 124506 (2020). + +<|ref|>text<|/ref|><|det|>[[85, 122, 847, 164]]<|/det|> +[10] Calio, P. B., Li, C. & Voth, G. A. Resolving the Structural Debate for the Hydrated Excess Proton in Water. J. Am. Chem. Soc. 143 (44), 18672- 18683 (2021). + +<|ref|>text<|/ref|><|det|>[[85, 182, 847, 247]]<|/det|> +[11] Carpenter, W. B., Fournier, J. A., Lewis, N. H. & Tokmakoff, A. Picosecond proton transfer kinetics in water revealed with ultrafast IR spectroscopy. J. Phys. Chem. B 122 (10), 2792- 2802 (2018). + +<|ref|>text<|/ref|><|det|>[[85, 264, 847, 306]]<|/det|> +[12] Xu, J., Zhang, Y. & Voth, G. A. Infrared spectrum of the hydrated proton in water. J. Phys. Chem. Lett. 2 (2), 81- 86 (2011). + +<|ref|>text<|/ref|><|det|>[[85, 323, 847, 366]]<|/det|> +[13] Kulig, W. & Agmon, N. A 'clusters- in- liquid' method for calculating infrared spectra identifies the proton- transfer mode in acidic aqueous solutions. Nat. Chem. 5 (1), 29- 35 (2013). + +<|ref|>text<|/ref|><|det|>[[85, 383, 847, 426]]<|/det|> +[14] Napoli, J. A., Marsalek, O. & Markland, T. E. Decoding the spectroscopic features and time scales of aqueous proton defects. J. Chem. Phys. 148 (22), 222833 (2018). + +<|ref|>text<|/ref|><|det|>[[85, 443, 847, 486]]<|/det|> +[15] Wang, E., Zhu, B. & Gao, Y. Discontinuous transition between Zundel and Eigen for \(\mathrm{H}_5\mathrm{O}_2^+\) . Chin. Phys. B 29 (8), 083101 (2020). + +<|ref|>text<|/ref|><|det|>[[85, 502, 847, 545]]<|/det|> +[16] Li, C. & Swanson, J. M. Understanding and tracking the excess proton in ab initio simulations; insights from IR spectra. J. Phys. Chem. B 124 (27), 5696- 5708 (2020). + +<|ref|>text<|/ref|><|det|>[[85, 561, 847, 625]]<|/det|> +[17] Biswas, R., Carpenter, W., Fournier, J. A., Voth, G. A. & Tokmakoff, A. IR spectral assignments for the hydrated excess proton in liquid water. J. Chem. Phys. 146 (15), 154507 (2017). + +<|ref|>text<|/ref|><|det|>[[85, 643, 847, 710]]<|/det|> +[18] Kim, J., Schmitt, U. W., Gruetzmacher, J. A., Voth, G. A. & Scherer, N. E. The vibrational spectrum of the hydrated proton: Comparison of experiment, simulation, and normal mode analysis. J. Chem. Phys. 116 (2), 737- 746 (2002). + +<|ref|>text<|/ref|><|det|>[[85, 727, 847, 792]]<|/det|> +[19] Daly Jr, C. A. et al. Decomposition of the experimental Raman and infrared spectra of acidic water into proton, special pair, and counterion contributions. J. Phys. Chem. Lett. 8 (21), 5246- 5252 (2017). + +<|ref|>text<|/ref|><|det|>[[85, 810, 847, 853]]<|/det|> +[20] Schröder, M., Gatti, F., Lauvergnat, D., Meyer, H.- D. & Vendrell, O. The coupling of the hydrated proton to its first solvation shell. Nat. Commun. 13 (1), 6170 (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[108, 60, 870, 128]]<|/det|> +[21] Singraber, A., Behler, J. & Dellago, C. Library- based LAMMPS implementation of high- dimensional neural network potentials. J. Chem. Theory Comput. 15 (3), 1827- 1840 (2019) + +<|ref|>text<|/ref|><|det|>[[108, 145, 870, 189]]<|/det|> +[22] Falk, M. & Giguère, P. A. Infrared spectrum of the \(\mathrm{H}_3\mathrm{O}^+\) ion in aqueous solutions. Can. J. Chem. 35 (10), 1195- 1204 (1957). + +<|ref|>text<|/ref|><|det|>[[108, 205, 870, 248]]<|/det|> +[23] Woutersen, S. & Bakker, H. J. Ultrafast vibrational and structural dynamics of the proton in liquid water. Phys. Rev. Lett. 96 (13), 138305 (2006). + +<|ref|>text<|/ref|><|det|>[[108, 264, 870, 308]]<|/det|> +[24] Thamer, M., De Marco, L., Ramasesha, K., Mandal, A. & Tokmakoff, A. Ultrafast 2D IR spectroscopy of the excess proton in liquid water. Science 350 (6256), 78- 82 (2015). + +<|ref|>text<|/ref|><|det|>[[108, 323, 870, 367]]<|/det|> +[25] Liu, J. et al. Towards complete assignment of the infrared spectrum of the protonated water cluster \(\mathrm{H}^{+}(\mathrm{H}_2\mathrm{O})_{21}\) . Nat. Commun. 12 (1), 1- 10 (2021). + +<|ref|>text<|/ref|><|det|>[[108, 382, 870, 427]]<|/det|> +[26] Zhang, Y. & Jiang, B. Universal machine learning for the response of atomistic systems to external fields. Nat. Commun. 14 (1), 6424 (2023). + +<|ref|>text<|/ref|><|det|>[[108, 442, 870, 510]]<|/det|> +[27] Brünig, F. N., Rammler, M., Adams, E. M., Havenith, M. & Netz, R. R. Spectral signatures of excess- proton waiting and transfer- path dynamics in aqueous hydrochloric acid solutions. Nat. Commun. 13 (1), 1- 12 (2022). + +<|ref|>text<|/ref|><|det|>[[108, 525, 870, 570]]<|/det|> +[28] Carpenter, W. B., Fournier, J. A., Biswas, R., Voth, G. A. & Tokmakoff, A. Delocalization and stretch- bend mixing of the HOH bend in liquid water. J. Chem. Phys. 147 (8) (2017). + +<|ref|>text<|/ref|><|det|>[[108, 585, 870, 653]]<|/det|> +[29] Chandra, A., Tuckerman, M. E. & Marx, D. Connecting solvation shell structure to proton transport kinetics in hydrogen- bonded networks via population correlation functions. Phys. Rev. Lett. 99 (14), 145901 (2007). + +<|ref|>text<|/ref|><|det|>[[108, 668, 870, 735]]<|/det|> +[30] Berkelbach, T. C., Lee, H.- S. & Tuckerman, M. E. Concerted hydrogen- bond dynamics in the transport mechanism of the hydrated proton: A first- principles molecular dynamics study. Phys. Rev. Lett. 103 (23), 238302 (2009). + +<|ref|>text<|/ref|><|det|>[[108, 750, 870, 818]]<|/det|> +[31] Liu, L., Tian, Y., Yang, X. & Liu, C. Mechanistic Insights into Water Autoionization through Metadynamics Simulation Enhanced by Machine Learning. Phys. Rev. Lett. 131 (15), 158001 (2023). + +<|ref|>text<|/ref|><|det|>[[108, 834, 870, 878]]<|/det|> +[32] Berendsen, H. J., Grigera, J. R. & Straatsma, T. P. The missing term in effective pair potentials. J. Phys. Chem. 91 (24), 6269- 6271 (1987). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 60, 849, 140]]<|/det|> +[33] Atsango, A. O., Morawietz, T., Marsalek, O. & Markland, T. E. Developing machine- learned potentials to simultaneously capture the dynamics of excess protons and hydroxide ions in classical and path integral simulations. J. Chem. Phys. 159 (7) (2023). + +<|ref|>text<|/ref|><|det|>[[85, 145, 848, 213]]<|/det|> +[34] Agostini, F., Vuilleumier, R. & Ciccotti, G. Infrared spectroscopy of small protonated water clusters at room temperature: an effective modes analysis. J. Chem. Phys. 134 (8), 084302 (2011). + +<|ref|>text<|/ref|><|det|>[[85, 228, 848, 272]]<|/det|> +[35] Wang, S. Intrinsic molecular vibration and rigorous vibrational assignment of benzene by first- principles molecular dynamics. Sci. Rep. 10 (1), 17875 (2020). + +<|ref|>text<|/ref|><|det|>[[85, 287, 847, 330]]<|/det|> +[36] Kresse, G. & Hafner, J. Ab initio molecular dynamics for open- shell transition metals. Phys. Rev. B 48 (17), 13115 (1993). + +<|ref|>text<|/ref|><|det|>[[85, 346, 848, 390]]<|/det|> +[37] Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total- energy calculations using a plane- wave basis set. Phys. Rev. B 54 (16), 11169 (1996). + +<|ref|>text<|/ref|><|det|>[[85, 405, 808, 425]]<|/det|> +[38] Blöchl, P. E. Projector augmented- wave method. Phys. Rev. B 50 (24), 17953 (1994). + +<|ref|>text<|/ref|><|det|>[[85, 440, 848, 485]]<|/det|> +[39] Perdew, J. P. et al. Atoms, molecules, solids, and surfaces: Applications of the generalized gradient approximation for exchange and correlation. Phys. Rev. B 46 (11), 6671 (1992). + +<|ref|>text<|/ref|><|det|>[[85, 500, 848, 567]]<|/det|> +[40] Hammer, B., Hansen, L. B. & Nørskov, J. K. Improved adsorption energetics within density- functional theory using revised Perdew- Burke- Ernzerhof functionals. Phys. Rev. B 59 (11), 7413 (1999). + +<|ref|>text<|/ref|><|det|>[[85, 583, 848, 651]]<|/det|> +[41] Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT- D) for the 94 elements H- Pu. J. Chem. Phys. 132 (15), 154104 (2010). + +<|ref|>text<|/ref|><|det|>[[85, 667, 650, 686]]<|/det|> +[42] Singraber, A. et al. CompPhysVienna/n2p2: Version 2.1.4 (2021). + +<|ref|>text<|/ref|><|det|>[[85, 702, 848, 768]]<|/det|> +[43] Thompson, A. P. et al. LAMMPS- a flexible simulation tool for particle- based materials modeling at the atomic, meso, and continuum scales. Comput. Phys. Commun. 108171 (2021). + +<|ref|>sub_title<|/ref|><|det|>[[77, 793, 356, 814]]<|/det|> +## 495 Acknowledgements + +<|ref|>text<|/ref|><|det|>[[85, 830, 848, 898]]<|/det|> +[495 Acknowledgements] This work is supported by the National Key Research and Development Program of China (2023YFA1506902, 2023YFA1506903) and the National Natural Science Foundation of China (22073041, 62075225). We thank the High Performance Computing Center of Nanjing University + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[103, 65, 870, 128]]<|/det|> +499 for computational resources. CGL and XYY are thankful to Prof. Minghui Yang for constructive suggestions. We would also thank the staff at BL06B beamline of the Shanghai Synchrotron Radiation Facility for their help in data collection. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 131, 223, 176]]<|/det|> +structureopt.zip Sl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/images_list.json b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..d37ae1231691a2e482b1be8554b3b23317ddf596 --- /dev/null +++ b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/images_list.json @@ -0,0 +1,272 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Magnetic reconnection during the confined flare, observed in the center of the \\(\\mathrm{H}\\alpha\\) line. (a) \\(H\\alpha\\) image observed at 08:04:54 UT by NVST superimposed on the line-of-sight magnetogram observed by SDO/HMI. Red (blue) contours indicate positive (negative) polarity of \\(\\pm 200\\) G and \\(\\pm 800\\) G. (b) Line-of-sight magnetogram observed by SDO/HMI at 08:04:55 UT. The two curved red lines indicate the chromospheric fibrils and the active-region filament participating in the reconnection. (c–h) The process of reconnection as seen in the \\(H\\alpha\\) observations. White arrows in c indicate the two reconnecting structures. Black arrows in d and e indicate the new chromospheric fibrils formed by reconnection. Blue arrows in d–f indicate the current sheet. The dotted lines in f and g indicate the cusp-shaped structures (magnetic separatrices) formed between the reconnection in- and outflows.", + "footnote": [], + "bbox": [ + [ + 241, + 155, + 747, + 606 + ] + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Profiles of GOES X-ray flux, inflows, and outflows during reconnection. (a) The profile of GOES X-rays in 1–8 Å (red line) and 0.5–4 Å (blue line). (b) The profile of the RHESSI X-ray counts in the 3–6, 6–12, 12–25, and 25–50 keV bands. (c) Inflow (dotted curve) along the A-B slit inferred from time-slices of the intensity. (d) Outflow along the C-D slit. (e) Outflow along the curved line E-F. (f) Outflow along the curved line G-H. All these lines are marked in Fig. 1h.", + "footnote": [], + "bbox": [ + [ + 177, + 210, + 820, + 650 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Hinode/EIS observation at Fe XVII 255 Å during magnetic reconnection. (a–c) Evolution of the cusp-shaped structure in the intensity images. (d–f) Doppler velocity and (g–i) non-thermal velocity obtained from single Gaussian fitting.", + "footnote": [], + "bbox": [ + [ + 196, + 240, + 803, + 664 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Fine structures in the SDO/AIA 211 Å images revealed by the unsharp masking technique. Red arrows mark the different bright blobs moving from the reconnection region (Fig. 4a–f). The green-dotted line in Fig. 4a indicates the trajectory I-J used to make the time-distance diagram of Fig. 4g. The projected velocities of the blobs lie in the range from \\(79 \\mathrm{km s^{-1}}\\) to \\(208 \\mathrm{km s^{-1}}\\) .", + "footnote": [], + "bbox": [ + [ + 196, + 171, + 783, + 707 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Emission measure and temperature maps and \\(\\mathrm{H}\\alpha\\) images with RHESSI X-ray sources superimposed. (a–d): DEM maps in four temperature ranges. (e) Total emission measure. (f) Temperature of the reconnection region derived from six AIA wavelengths. (g–i): \\(\\mathrm{H}\\alpha\\) images with RHESSI hard X-ray sources superimposed. The levels of the contours are at 50% and 80% of the peak values in the 3–6, 6–12, 12–25, and 25–50 keV bands.", + "footnote": [], + "bbox": [ + [ + 191, + 223, + 800, + 641 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6: RHESSI photon spectra accumulated around 08:12 UT and 08:19 UT during the quoted 1-minute intervals on 2014 February 2. The spectra are fitted by combining the \\(\\mathrm{vth}\\) and thick2 functions (see Methods). The black line shows the photon spectrum with the background subtracted. The pink line shows the background. The green and the yellow lines display the fitting with the single-temperature thermal model and the thick-target nonthermal model, respectively.", + "footnote": [], + "bbox": [ + [ + 200, + 268, + 790, + 592 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7: Magnetic structure of the source region. (a) Selected field lines showing the twisted magnetic structure of the filament (green) and the loop-like structure of the chromospheric fibrils (red) before the reconnection event, obtained by using NLFFF extrapolation from the vector magnetogram. (b)–(c) Map of the magnetic squashing degree \\(\\log Q\\) at heights \\(Z = 0\\) and \\(Z = 2.9\\) Mm above the photosphere. (d) Iso-surface \\(\\log Q = 4\\) in the reconnection region. (e) \\(\\log Q\\) in the vertical cross-section marked by a yellow dashed line in (a).", + "footnote": [], + "bbox": [ + [ + 196, + 250, + 789, + 589 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8: Evolution of magnetic field, current sheet, and plasmoids (blobs) during magnetic reconnection in the data-driven MHD simulation. (a) and (b): Evolution of the magnetic structure involved in the modeled reconnection event. (c) and (d): Zoom showing the magnetic structure of the current sheet at two selected times. The red boxes in (a) and (b) show the field of view of (c)–(f). (e) and (f): Distribution of electric current density ( \\(Jc\\) ) in the current sheet (also overlaid in (c)–(d)) at the height \\(z = 1\\) (11.6 Mm).", + "footnote": [], + "bbox": [ + [ + 270, + 156, + 723, + 690 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9: Evolution of the photospheric vector magnetic field observed by SDO/HMI and flow field derived using the DAVE algorithm. (a–c): Vector magnetograms. The greyscale images show the vertical field component, saturated at 1000 G, and the arrows show the horizontal component. (d–f): Evolution of the flow field. Blue and red arrows indicate the horizontal photospheric velocity in the positive and negative polarities, respectively.", + "footnote": [], + "bbox": [ + [ + 194, + 279, + 795, + 585 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 58, + 102, + 860, + 833 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 48, + 50, + 920, + 644 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 66, + 58, + 904, + 641 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 65, + 45, + 861, + 777 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 68, + 50, + 920, + 650 + ] + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 95, + 55, + 910, + 512 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7", + "footnote": [], + "bbox": [ + [ + 70, + 55, + 928, + 540 + ] + ], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8", + "footnote": [], + "bbox": [ + [ + 60, + 52, + 675, + 777 + ] + ], + "page_idx": 37 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9", + "footnote": [], + "bbox": [ + [ + 65, + 66, + 925, + 510 + ] + ], + "page_idx": 38 + } +] \ No newline at end of file diff --git a/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873.mmd b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873.mmd new file mode 100644 index 0000000000000000000000000000000000000000..75a99f89eb8bb60c43a9c6ad600effaef9e1654e --- /dev/null +++ b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873.mmd @@ -0,0 +1,465 @@ + +# Multi-scale Observation and Data-driven Modeling of Magnetic Reconnection in Flaring Chromospheric Plasmas on the Sun + +Xiaoli Yan (yanxl@ynao.ac.cn) + +Yunnan Observatories, CAS + +Zhike Xue Yunnan Observatories + +Chaowei Jiang Harbin Institute of Technology, Shenzhen https://orcid.org/0000- 0002- 7018- 6862 + +Eric Priest School of Mathematics and Statistics + +Bernhard Kliem University of Potsdam https://orcid.org/0000- 0002- 5740- 8803 + +Liheng Yang Yunnan Observatories + +Jincheng Wang Yunnan Observatories CAS + +Defang Kong Yunnan Observatories + +Yongliang Song Key Laboratory of Solar Activity, National Astronomical Observatories, Chinese Academy of Sciences + +Xueshang Feng Chinese academy of sciences + +Zhong Liu Yunnan Observatories, Chinese Academy of Sciences, China + +## Article + +Keywords: flaring chromospheric plasmas, planetary physics, solar flares + +Posted Date: February 19th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 232633/v1 + +<--- Page Split ---> + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on February 2nd, 2022. See the published version at https://doi.org/10.1038/s41467-022-28269-w. + +<--- Page Split ---> + +# Multi-scale Observation and Data-driven Modeling of Magnetic Reconnection in Flaring Chromospheric Plasmas on the Sun + +Xiaoli Yan \(^{1,2,3}\) , Zhike Xue \(^{1,3}\) , Chaowei Jiang \(^{4}\) , E. R. Priest \(^{5}\) , Bernhard Kliem \(^{6}\) , Liheng Yang \(^{1,3}\) , Jincheng Wang \(^{1,3}\) , Defang Kong \(^{1,3}\) , Yongliang Song \(^{7}\) , Xueshang Feng \(^{2}\) , Zhong Liu \(^{1,8}\) + +February 11, 2021 + +Magnetic reconnection is a multi- faceted process of energy conversion in astrophysical, space and laboratory plasmas that operates at microscopic scales but has macroscopic drivers and consequences. Solar flares present a key laboratory for its study, leaving imprints of the microscopic physics in radiation spectra and allowing the macroscopic evolution to be imaged, yet a full observational characterization remains elusive. Here we combine high resolution imaging and spectral observations of a solar flare at multiple wavelengths with data- driven magnetohydrodynamic modeling to study the dynamics of the involved plasma from the current sheet to the plasmoid scale. The flare resulted from the interaction of a twisted filament and chromospheric fibrils. By inferring the reconnection to be fast and mediated by plasmoids, the relevance of this reconnection mode is found to extend beyond hot flare plasmas to such cool structures in the chromosphere, which have many analogs in astrophysical objects. + +<--- Page Split ---> + +4 Institute of Space Science and Applied Technology, Harbin Institute of Technology, Shenzhen, 5180055, China5 Mathematics Institute, University of St Andrews, St Andrews, KY16 9SS, UK.6 Institute of Physics and Astronomy, University of Potsdam, Potsdam 14476, Germany.7 Key Laboratory of Solar Activity, National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100012, China.8 University of Chinese Academy of Sciences, Yuquan Road, Shijingshan Block, Beijing 100049, Peoples Republic of China. + +## Introduction + +Solar flares result from the most dramatic events of energy release by magnetic reconnection on the Sun and provide a broad range of observational detail about the involved processes[1- 3]. In the standard flare model for eruptive events[4- 7], magnetic reconnection proceeds in a vertical current sheet spawned on the underside of an erupting flux rope which also drives a coronal mass ejection (CME). Rising arcades of reconnected hot, initially cuspshaped magnetic loops ("flare loops"), rooted in a pair of bright "flare ribbons", are formed in the downward reconnection outflow. Confined flares result when the erupting flux is stopped by overlying flux or when neighboring flux system are pressed against each other, and a reconnecting current sheet is formed in the interface[8- 10]. A filament extending along the polarity inversion line of the source region often traces the activated flux. These large- scale geometries of the reconnecting current sheet and its drivers and consequences are well established[11- 20]. + +High resolution imaging observations provide rich information about details of the reconnection process which result from its internal dynamics, especially the formation of plasmoids[21,22] and turbulent structures[23], and from the complexity and three- dimensional (3D) nature of the solar magnetic field[24,25]. Spectral and multi- wavelength data reveal aspects of the microscopic processes, e.g., particle acceleration[26,27], the nonthermal- thermal energy partition[28], the plasmoid instability[29,30], and bursty[31] or turbulent reconnection[32,33], in addition to flows[34]. Finally, besides hot flare plasmas, the Sun also allows the reconnection of cool chromospheric structures to be studied[24,35]. However, despite this broad range of information, comprehensive simultaneous imaging and spectral observations are still limited. + +<--- Page Split ---> + +In particular, while individual components of the reconnection in- and outflows have often been observed[15,18,21,34,36,37], we are aware of only two events revealing all four flows[19,35] but these have no or only limited spectral information about the involved plasma. Such limitation prevents the mode of reconnection, i.e. Petschek- like[12], plasmoid- mediated[29], collisionless[38], or turbulent[39], to be inferred. + +Here we present multi- wavelength imaging and spectral evidence for magnetic reconnection in a strong (M- class) confined solar flare, traced by the cool plasma of the interacting filament and chromospheric fibrils and driven by emerging flux. The flare is unique in that it displayed all four reconnection flows, plasmoids not only in the current sheet but also in the separatrices, was fully covered by spectroscopic observations, and yielded the first hard X- ray images from a current sheet on the solar disk. This allows us to measure both inflows and both outflows and to infer the mode of reconnection from the analysis of the physical conditions in the current sheet. We also perform a state- of- the- art, data- driven numerical simulation of the reconnection event, which supports our interpretation of the observational data. + +## Results + +High- resolution imaging of the reconnection flows. The confined M2.2 flare occurred in Active Region NOAA 11967 at S13E04 on 2014 February 2, starting at 07:17 UT, peaking at 08:20 UT, and ending at 08:29 UT. This was a large active region with a \(\beta \gamma \delta\) sunspot configuration. Several of its main flux components were already decaying, but there was also new flux emerging into the region. Driven by these changes, many C- and M- class flares were produced during its passage across the solar disk[40,41]. + +Figure 1a shows an Hα image from the New Vacuum Solar Telescope (NVST)[42] superimposed with magnetic flux contours from the Solar Dynamics Observatory (SDO)/Helioseismic and Magnetic Imager (HMI)[43] magnetogram in Fig. 1b. The red curved lines in Fig. 1b indicate the positions of the reconnecting structures: chromospheric fibrils on the east (left) side and a filament, which was contained in a twisted flux rope[44], on the west (right) side. The evolution of the flare as seen by the NVST in Hα is displayed in Fig. 1c- h and the related Supplementary Movie 1. Here the white arrows indicate the reconnecting structures at 08:04:54 UT. The structures approached each other and began to reconnect. What may well be a current sheet can be seen at the junction of the two structures marked by the blue arrows in Fig. 1d- f. The newly formed threads are indicated by black arrows + +<--- Page Split ---> + +in Fig. 1d,e. Both sets of new, reconnected structures displayed the cusp characteristic of newly reconnected field lines, as outlined by the dashed blue curved lines in Fig. 1f,g. Simultaneously with the magnetic reconnection, the M- class flare occurred (see the GOES soft X- ray light curve in Fig. 2a and the Reuven Ramaty High- Energy Solar Spectroscopic Imager (RHESSI) [45] X- ray counts in the 3–6, 6–12, 12–25, and 25–50 keV bands in Fig. 2b). + +The projected in- and outflow velocities were measured using time- slices of the intensity along the artificial slits A- B and C- D shown by the red and blue lines, respectively, in Fig. 1h. Up to 08:14 UT, the inflows are found to be \(v_{\mathrm{i}} = 0.06 \mathrm{km s}^{- 1}\) on the east side and \(v_{\mathrm{i}} = 0.05 \mathrm{km s}^{- 1}\) on the west side; see the dashed red lines in Fig. 2c. The outflows are measured from \(v_{\mathrm{o}} = 2.0\) to \(3.7 \mathrm{km s}^{- 1}\) in southward direction, which is presumably downward, and \(v_{\mathrm{o}} = 3.4\) to \(3.8 \mathrm{km s}^{- 1}\) in northward direction, which is presumably upward; see the dashed blue lines in Fig. 2d. The resulting reconnection rate, \(M = v_{\mathrm{i}} / v_{\mathrm{o}} = 0.01–0.03\) , clearly indicates a fast reconnection process. We also measure the outflows along the cusp lines (magnetic separatrices) attached to the ends of the current sheet (see the curved lines E- F and G- H in Fig. 1h). These are found to lie in the ranges \(2.3–5.5 \mathrm{km s}^{- 1}\) and \(0.8–2.6 \mathrm{km s}^{- 1}\) for the south and north cusp lines, respectively (see Fig. 2d,e). As the reconnection continues, an unusual feature of the event is that the chromospheric fibrils and the main body of the filament began to separate from the reconnection region after 08:14 UT with velocities of 0.98 and \(0.39 \mathrm{km s}^{- 1}\) on the east and west sides, respectively (see Fig. 2b), whereas some of the lower chromospheric fibrils and threads of the filament still participated in the reconnection process. This is considered further in the Discussion. + +EUV spectroscopy of reconnection flows and turbulence. The Hinode/EUV Imaging Spectrometer (EIS) [46] observations covered essentially the whole reconnection event. Figure 3 shows the intensity (a–c), Doppler velocity (d–f), and non- thermal velocity (g–i). There is a strong red shift in the southern cusp- shaped structure, with the maximum Doppler velocity reaching about \(80 \mathrm{km s}^{- 1}\) . The blue shift at the northern cusp reaches about \(40 \mathrm{km s}^{- 1}\) . However, the true peak velocity may be higher at the north side (as also indicated by the projected velocities), because the selected spectral line obviously samples only part of the upward reconnection outflow, due to a reduction of density and temperature (adiabatic cooling) in the free upward expansion of the plasma. Nevertheless, both reconnection outflows are spectroscopically detected and prove to be significantly higher near the edges of the current sheet than the outflows inferred from the \(\mathrm{H}\alpha\) structures + +<--- Page Split ---> + +further away from the current sheet (Figs. 1h and 2d). It is also clear that the outflow is mainly downward at the southern edge of the current sheet and mainly upward at the northern edge. The implied reconnection rate, \(M \sim 0.0007 - 0.024\) for the southern outflow, is still fast. The cusp- shaped structures also appear in the non- thermal velocity maps. The highest non- thermal velocities were located in the current sheet, especially in the tip of the cusp- shaped structures, with values in the range \(\sim 180 - 300 \mathrm{km s^{- 1}}\) , which significantly exceed the projected and Doppler velocities of the reconnection outflows. This suggests that the reconnection proceeds in a highly dynamic manner, with plasmoid- mediated reconnection [47,48] being the obvious candidate for the specific mode of reconnection realized in the event. To determine the electron density in the current sheet, we used the theoretical line intensity ratios of the Fe XIV ( \(\lambda 264.92 / \lambda 274.37\) ) density sensitive pair, which were calculated by using the CHIANTI atomic package [49] [50]. Although the line \(\lambda 274.37\) is blended with Si VII \(\lambda 274.18\) , the blend can be safely ignored due to its low intensity compared with \(\lambda 274.37\) in active region conditions. The number density values in the current sheet are found to scatter in the range from \(10^{8}\) to \(10^{12} \mathrm{cm^{- 3}}\) , similar to typical densities in filaments [51]. The highest values are located in the southern (lower) cusp. + +Plasmoids. During the early, slow- rise phase of the flare, from 07:57 to \(\sim 08:06\) UT, many small bright blobs were observed, moving from the current sheet to the legs of the cusp- shaped structures (see Supplementary Movie 1). Figure 4 and and the related Supplementary Movie 2 show the highlighted fine structures of the SDO/AIA 211 Å images revealed by the unsharp masking technique. There are many bright blobs marked by the red arrows moving from the reconnection region (see Fig. 4a–f). The velocities of the blobs along the reconnected field lines are from \(79 \mathrm{km s^{- 1}}\) to \(208 \mathrm{km s^{- 1}}\) (see the time- distance diagram, Fig. 4g, obtained along the green- dotted line in Fig. 4a). To the best of our knowledge, the phenomenon that blobs move from the reconnection region along the separatrices, as found in a 3D particle simulation of reconnection [52], is observed here for the first time. Their number, more than ten, is among the highest reported so far for a single reconnection event [21,53]. They support the interpretation of the high Fe XVII line widths as an indication of plasmoid- mediated reconnection. The increasing brightness and source width of the emissions prevented the detection of blobs by unsharp masking at later times, but their continued formation is suggested by the continued presence of high nonthermal velocities in the EIS spectra, efficient particle acceleration revealed by the hard X- rays, and the + +<--- Page Split ---> + +numerical modeling below. + +Heating. The heating and brightening resulting from the reconnection can also clearly be seen in all extreme ultraviolet (EUV) and UV wavelengths of the SDO/Atmospheric Imaging Assembly (AIA) [54]. The Supplementary Movie 3 shows the evolution of the EUV and UV observations at eight wavelengths and of the line- of- sight magnetogram. The cusp- shaped regions of outflowing hot plasma are obvious, as is the current sheet between them. + +Six EUV wavelengths observed by AIA were used to obtain the differential emission measure (DEM), total EM, and temperature maps shown in Fig. 5a–f and Supplementary Movie 4. The X- shaped geometry of the reconnection region can be seen in the whole temperature range of \(T > 1\) MK, i.e., the cool plasma of the interacting filament and chromospheric fibrils is heated to coronal and flare temperatures. The peak temperature in the reconnection region reaches about 10 MK. + +X- ray spectroscopy of accelerated particles. The X- ray sources from RHESSI in Fig. 5g–i show that X- ray emission \(>3\) keV is emitted from the beginning of the event’s slow- rise phase (Fig. 5g). Higher energies are reached following the development of the flare, and the highest energy release rate into hard X- rays around 08:18 UT coincides with the steepest slope of the soft X- ray light curve (Fig. 2a), as is typically the case in flares. The emission comes mainly from the current sheet, different from eruptive flares, where the foot point sources are usually strongest. Thus, the hard X- rays provide a direct diagnostic of the conditions in the current sheet. These are the first maps of hard X- ray emission from a reconnecting current sheet in a confined flare on disk. The emission originates preferentially from the lower (southern) cusp, where the density in the sheet was found to be highest, similar to limb flares with occulted foot points [55]. + +Figure 6 shows the RHESSI hard X- ray photon spectra taken at about 08:12 UT and 08:19 UT. The temperatures derived from the fitted spectra are 13.1 MK (Fig. 6a) and 33.6 MK (Fig. 6b); this extends the results of the DEM analysis of the AIA EUV data which are limited to \(\approx 11\) MK. Moreover, the spectra show both thermal and non- thermal components. The emission above \(\sim 25\) keV is of non- thermal nature, indicating particle acceleration in the current sheet by the fast reconnection process. Particle acceleration by reconnection is known to be particularly efficient when multiple plasmoids form, implying multiple X- lines [27,56,57]. This corroborates the deduction above that plasmoid- mediated reconnection was at work. + +Magnetic field structure. Using the nonlinear force free field (NLFFF) + +<--- Page Split ---> + +extrapolation of the photospheric vector magnetogram from SDO/HMI at 07:36 UT (see Methods), the magnetic structure of the filament and the involved magnetic loops prior to the onset of reconnection are obtained. Figure 7 shows the magnetic structure of the filament, the magnetic loops, and the map and iso- surface of the magnetic squashing degree \(\mathrm{Q}^{[58]}\) , which measures the gradient in the field line foot point mapping. High values of \(\mathrm{Q}\) , known as quasi- separatrix layers (QSLs), are the preferred locations for the formation of current sheets and the onset of reconnection \({}^{[59]}\) . As can be seen in the NLFFF extrapolation, the magnetic field lines from the main positive sunspot in the east part of the active region connect to the diffuse negative polarity in the north (the pink lines), while the field lines of the filament from the smaller positive spot in the west connect to the main negative spot (the green lines). The latter, weakly twisted magnetic structure (Fig. 7a) is in good agreement with the shape of the filament in the \(\mathrm{H}\alpha\) center observation (Fig. 1a). Figures 7b- c show the map of \(\log Q\) in the reconnection region at heights \(Z = 0\) and \(Z = 2.9\) Mm above the photosphere. The photospheric magnetogram and the iso- surface \(\log Q = 4\) in Fig. 7d reveal the X- shaped structure of high \(\log Q\) values that facilitates the onset of reconnection at the observed position. The vertical cross- section of \(\log Q\) in Fig. 7e demonstrates that the separatrices run between the flux bundles, at the position where the current sheet forms. + +MHD simulation. We further simulated the reconnection process using a data- driven high- resolution magnetohydrodynamic (MHD) model. Since this realistic, data- driven model of the current sheet includes a guide field component in the direction of the current flow, the plasmoids seen in the current density distribution (Fig. 8e- f) are small, three- dimensional flux ropes in the field line plots \({}^{[52]}\) (Fig. 8a- d). As observed in two- dimensional simulations \({}^{[47,48]}\) , as well as in the observations presented above, plasmoids form dynamically and repeatedly, and are accelerated along the current sheet. Two new plasmoids are seen to form in Fig. 8f. Between the plasmoids, the current sheet steepens to the limit allowed by the numerical resolution. The simulation obviously corresponds well to the observations from the global to the plasmoid scale. The plasmoids appeared only after adaptive mesh refinement was applied to permit the current sheet width to shrink to an aspect ratio of \(\sim 0.01\) , in agreement with the theoretical prediction for the occurrence of the plasmoid instability \({}^{[60]}\) . + +Released energy. A rough estimate of the way reconnection releases the magnetic energy required to power an M- class flare ( \(\sim 10^{24}\) J) can be made + +<--- Page Split ---> + +as follows. We expect that part of the energy is released in the reconnecting current sheet and a further part is released by the untwisting of the filament- flux rope after it is connected to the diffuse remote polarity. Suppose first that oppositely directed magnetic field of strength \(B_{\mathrm{i}}\) is brought in from both sides at speed \(v_{\mathrm{i}}\) for a time \(\tau\) towards a reconnection region of surface area \(L_{0}{}^{2}\) , where the current sheet and attached slow- mode shock waves convert the energy. The magnetic energy is brought in from both sides from a volume \(2L_{0}L_{\mathrm{i}}\) , where \(L_{\mathrm{i}} = v_{\mathrm{i}}\tau\) , and so the total energy calculated is + +\[W = \frac{B_{\mathrm{i}}^{2}}{2\mu_{0}} L_{0}{}^{2}v_{\mathrm{i}}\tau . \quad (1)\] + +This may be evaluated for the M- class flare, where \(B_{\mathrm{i}} \sim 500 \mathrm{G}\) from the extrapolation of the magnetogram which is used as the initial condition for the simulation (the average value in the reconnection region), and the values \(L_{0} = 15^{\prime \prime}\) (10.9 Mm), \(v_{\mathrm{i}} = 2 \mathrm{km s}^{- 1}\) , and \(\tau = 70 \mathrm{min}\) are obtained from the \(\mathrm{H}\alpha\) data, to give an energy of \(\sim 10^{24} \mathrm{J}\) , as required. This corresponds well to the energy release of \(\sim 10^{24} \mathrm{J}\) (9.6 \(\times 10^{30}\) erg) in the simulation and agrees with the canonical picture that the flare is powered by magnetic reconnection. + +Additional magnetic energy may be liberated by the untwisting of the flux rope which holds the filament. The energy associated with the twist component of the flux rope field, \(B_{\phi} = (R / a)B_{0\phi}\) , can be written as + +\[W_{\mathrm{FR}} = \frac{B_{0\phi}{}^{2}}{2\mu_{0}}\pi a^{2}L_{\mathrm{R}}, \quad (2)\] + +where \(R\) , \(a\) and \(L_{\mathrm{R}}\) are the rope's major and minor radii and length. Suppose the twist on the edge of the flux rope is \(\Phi\) , then \(\Phi = (L B_{0\phi}) / (a B_{z})\) and \(B_{0\phi}\) is \(\mathrm{a}\Phi B_{z} / L_{\mathrm{R}}\) . In the present case, we estimate \(B_{z} \sim 500 \mathrm{G}\) , \(L_{\mathrm{R}} = 100^{\prime \prime}\) , \(a = 5^{\prime \prime}\) and a twist of \(\Phi \sim 4\pi\) , so \(W_{\mathrm{FR}} = 3 \times 10^{23} \mathrm{J}\) . This is significantly smaller than the energy released in the reconnecting current sheet, unless the flux tube radius is a factor of \(\sim 2\) larger than we have estimated from the \(\mathrm{H}\alpha\) images of the filament. + +In summary, the presented high- resolution imaging and spectral observations of macroscopic flows demonstrate that fast magnetic reconnection of the field in structures originally at chromospheric temperatures released the magnetic energy which powered the solar flare. Plasmoid- mediated reconnection is strongly indicated by direct imaging and nonthermal EUV and hard X- ray spectra. This is further supported by a 3D data- driven MHD simulation + +<--- Page Split ---> + +which reproduces the observed change in connectivity, agrees with estimates of the energy release, and reveals the appearance of twisted plasmoids (mini flux ropes) at the theoretically expected threshold. + +## Discussion + +In order to understand why reconnection occurred so many times in this place, the evolution of the magnetic and flow fields in the photosphere was investigated using HMI vector magnetograms. The vector magnetogram is shown in Fig. 9a–c, where the arrows represent the horizontal magnetic field. The corresponding flow field was calculated from the radial magnetic field using the DAVE algorithm[61] (see Methods); see Fig. 9d–f and the related Supplementary Movie 5. This reveals a flow toward the magnetic reconnection site from the west side, which pushes together oppositely directed flux in the higher atmosphere (as found in the MHD simulation). This persistent flow in the photosphere is likely to be the main reason why so many flares occurred in the same place. Additional distortions of the coronal field configuration, also likely to support the onset of reconnection, are produced by the emergence of flux very near the reconnection site (green ellipses in Fig. 9b,c) and subsequent motion toward the reconnection site. The flows at the east side are more variable, with directions along and away from the current sheet alternating. The latter can explain the separation of the reconnecting flux after \(\sim 08:14\) UT in the present event, as well as the termination of the other homologous events on the same day. + +A neighboring eruption may have had an additional influence. This active region produced three M- class flares and several C- class flares in the first half of 2014 February 2. All these flares were located at the same place and had a similar morphology. The M- class flare studied here was preceded by a weaker, C- class flare above another polarity inversion line in the complex active region, lying to the east of the main positive sunspot, at \((x,y)\approx (- 340, - 50)\) . The C- class flare was caused by the rise of flux in the corona above the filament located at that inversion line. The associated large- scale change of the coronal field, due to the expansion of that flux, may have played an additional role in pushing the chromospheric fibrils east of the reconnection region toward the filament on the west side of the region. With the relaxation of the overlying coronal field during the decay of the C- class flare (07:17–07:58 UT)[62], the fibrils may have experienced a restoring, oppositely directed force, dragging them apart from the formed current sheet and terminating the reconnection inflow. This might have additionally + +<--- Page Split ---> + +contributed to the termination of the reconnection event. + +Once the reconnection in- and outflows are established as inferred from the \(\mathrm{H}\alpha\) data, reconnection would not easily cease by itself, because the process involves a positive feedback: the acceleration of the outflows drives continued inflows which transports new flux with an oppositely directed component toward the sheet. This is consistent with the MHD simulation. The fact that the reconnection event produced an M2.2 flare, more intense by a factor two than the preceding C8.9 flare, shows that the active region had stored free magnetic energy in the area of the reconnection event. This free energy was carried by the electric current associated with the antiparallel field component of the reconnecting structures. Since these structures still existed after the flare, the alternative possibility for the relatively quick ending of the reconnection event by a depletion of the available free magnetic energy appears less likely than the action of an external driver. + +## Methods + +Observational data Images in the center of the \(\mathrm{H}\alpha\) line at 6562.8 A obtained at the 1 m New Vacuum Solar Telescope (NVST) at Fuxian Lake[42], are used in this study. They have a pixel size of \(0.^{\prime \prime}163\) and a cadence of 12 s. The data are calibrated from Level 0 to Level 1 with dark current subtracted and flat field corrected, and then the calibrated images are reconstructed to Level 1+ using a speckle masking method[63]. + +Full- disk UV and EUV images observed by SDO/AIA[54] are also used to clarify the process of magnetic reconnection. These images have a 12- s cadence and a spatial resolution of \(0.^{\prime \prime}6\) per pixel. Vector magnetograms from the Space Weather HMI Active Region Patch (SHARP) series observed by the Helioseismic and Magnetic Imager (HMI)[43,64,65] on board SDO are used. These allow us to infer the photospheric velocity field and to perform a data- driven simulation of the event. The vector magnetograms have a pixel scale of about \(0.^{\prime \prime}5\) and a cadence of 12 minutes. They are derived by using the Very Fast Inversion of the Stokes Vector algorithm[66]. The minimum energy method[67,68] is used to resolve their 180 degree azimuthal ambiguity. The images are remapped using a Lambert (cylindrical equal area) projection centered on the midpoint of the AR, which is tracked at the Carrington rotation rate[69]. + +The spectral data observed by the EUV Imaging Spectrometer (EIS) on board the Hinode mission were used to derive the Doppler velocity and nonthermal velocity in the reconnection region. Hinode/EIS[46] is a scanning slit + +<--- Page Split ---> + +spectrometer that uses two wave bands in the EUV: 170–210 Å and 250–290 Å. The spectral resolution is about 0.0223 Å pixel\(^{- 1}\). It can measure velocities at an accuracy of a few km s\(^{- 1}\). Fortunately, the EIS observation covered the whole duration of the reconnection event and nearly its whole area. It thus provides an excellent opportunity to conduct a spectroscopic study of magnetic reconnection. We use the line Fe XVII 255 Å (log \(T = 6.6\)). The field of view (FOV) is 161'' in the raster direction and 152'' in the slit direction. The spectral profiles are fitted with single Gaussian functions. The reference wavelength used to determine the Doppler and non- thermal velocities is taken from the average of each raster. The sparse raster used the 2'' slit with a step size of 2. ''995 and an exposure time of 4–6 s. + +Data from RHESSI[45] are also employed to study the soft and hard X- ray sources during the M- class flare. The energy ranges of 3–6, 6–12, 12–25, and 25–50 keV were selected to observe the evolution of these sources during the magnetic reconnection. The accumulation time is 30 s for constructing X- ray images by using the Clean imaging algorithm. + +Data- driven simulation. The full set of time- dependent 3D MHD equations were solved, imposing bottom boundary conditions driven continuously[70,71] by the changing photospheric vector magnetic field observed by SDO/HMI (see Observational data). The background plasma is initially in a hydrostatic, isothermal state with \(T = 10^{6}\) K (sound speed \(c_{S} = 128 \mathrm{km s}^{- 1}\) ) in solar gravity. Its density is configured to make the plasma \(\beta\) as small as \(2 \times 10^{- 3}\) (the maximum Alfvén velocity \(v_{\mathrm{A}}\) is \(4 \mathrm{Mm s}^{- 1}\) ) to mimic the coronal low- \(\beta\) and highly tenuous conditions. The plasma thermodynamics are simplified to be adiabatic. The bottom boundary of the model is assumed to be the coronal base, and also the magnetic field measured on the photosphere is used as a reasonable approximation to the field at the coronal base. To obtain the detailed structure in the current sheet, an adaptive refinement mesh is used with highest resolution of \(1 / 16\) arcsecond (i.e., \(45 \mathrm{km}\) ) to capture the fine structures in the reconnecting current sheet. + +Unsharp masking technique. The unsharp masking technique is applied to SDO/AIA 211 Å images to highlight fine structures like blobs. First, a background image by smoothing the original image is obtained as an unsharp masked image. Second, the enhanced image is the residual of the original image subtracting the background one. A smoothing window of \(7 \times 7\) pixels is adopted for optimal effects. + +<--- Page Split ---> + +NLFFF extrapolation. The three- dimensional magnetic structures of the filament and the magnetic loops are reconstructed by using an optimization algorithm[72]. The bottom vector magnetic field data observed by SDO/HMI are modified to remove most of the net force and torque that usually results in an inconsistency between the photospheric magnetic field and the force- free assumption in the NLFFF models by using a preprocessing procedure[73]. The visualization of the 3D magnetic field is realized by the software Paraview. + +RHESSI spectrum. RHESSI hard X- ray photon spectra were created by the Object Spectral Executive (OSPEX)[74] software with the background subtracted. A single- temperature thermal bremsstrahlung radiation function ("vth") and the thick- target non- thermal bremsstrahlung with an isotropic pitch- angle distribution ("thick2") were adopted to fit the spectra. + +Velocity field calculation. The differential affine velocity estimator (DAVE)[61,75] has been applied to the HMI vector magnetograms to derive the photospheric flow field. The DAVE method combines the advection equation and a differential feature tracking technique to detect flows. We used a window size of 19 pixels, which is large enough to include structure information and small enough to have a good spatial resolution. + +## References + +[1] Priest, E. R. & Forbes, T. G. The magnetic nature of solar flares. Astron. Astrophys. Rev. 10, 313- 377 (2002). + +[2] Fletcher, L. et al. An Observational Overview of Solar Flares. Space Sci. Rev. 159, 19- 106 (2011). 1109.5932. + +[3] Benz, A. O. Flare Observations. Living Reviews in Solar Physics 14, 2 (2017). + +[4] Carmichael, H. A Process for Flares. NASA Special Publication 50, 451 (1964). + +[5] Sturrock, P. A. Model of the High- Energy Phase of Solar Flares. Nature 211, 695- 697 (1966). + +<--- Page Split ---> + +[6] Hirayama, T. Theoretical Model of Flares and Prominences. I: Evaporating Flare Model. Solar Phys. 34, 323- 338 (1974). + +[7] Kopp, R. A. & Pneuman, G. W. Magnetic reconnection in the corona and the loop prominence phenomenon. Solar Phys. 50, 85- 98 (1976). + +[8] Ji, H., Wang, H., Schmahl, E. J., Moon, Y. J. & Jiang, Y. Observations of the Failed Eruption of a Filament. Astrophys. J. Lett. 595, L135- L138 (2003). + +[9] Török, T. & Kliem, B. Confined and Ejective Eruptions of Kink-unstable Flux Ropes. Astrophys. J. Lett. 630, L97- L100 (2005). astro- ph/0507662. + +[10] Li, L. et al. Magnetic reconnection between a solar filament and nearby coronal loops. Nature Physics 12, 847- 851 (2016). 1605.03320. + +[11] Tsuneta, S. et al. Observation of a solar flare at the limb with the YOHKOH Soft X-ray Telescope. Pub. Astron. Soc. Japan 44, L63- L69 (1992). + +[12] Tsuneta, S. Structure and Dynamics of Magnetic Reconnection in a Solar Flare. Astrophys. J. 456, 840 (1996). + +[13] Shibata, K. Evidence of Magnetic Reconnection in Solar Flares and a Unified Model of Flares. Astrophys. Space Sci. 264, 129- 144 (1999). + +[14] Lin, J. & Forbes, T. G. Effects of reconnection on the coronal mass ejection process. J. Geophys. Res. 105, 2375- 2392 (2000). + +[15] Yokoyama, T., Akita, K., Morimoto, T., Inoue, K. & Newmark, J. Clear Evidence of Reconnection Inflow of a Solar Flare. Astrophys. J. Lett. 546, L69- L72 (2001). + +[16] Sui, L. & Holman, G. D. Evidence for the Formation of a Large- Scale Current Sheet in a Solar Flare. Astrophys. J. Lett. 596, L251- L254 (2003). + +[17] Lin, J. et al. Direct Observations of the Magnetic Reconnection Site of an Eruption on 2003 November 18. Astrophys. J. 622, 1251- 1264 (2005). + +<--- Page Split ---> + +[18] Savage, S. L. et al. Low- altitude Reconnection Inflow- Outflow Observations during a 2010 November 3 Solar Eruption. Astrophys. J. 754, 13 (2012). 1111.1945. + +[19] Su, Y. et al. Imaging coronal magnetic- field reconnection in a solar flare. Nature Physics 9, 489- 493 (2013). 1307.4527. + +[20] Yan, X. L. et al. Simultaneous Observation of a Flux Rope Eruption and Magnetic Reconnection during an X- class Solar Flare. Astrophys. J. Lett. 853, L18 (2018). 1801.02738. + +[21] Takasao, S., Asai, A., Isobe, H. & Shibata, K. Simultaneous Observation of Reconnection Inflow and Outflow Associated with the 2010 August 18 Solar Flare. Astrophys. J. Lett. 745, L6 (2012). 1112.1398. + +[22] Gou, T., Liu, R., Kliem, B., Wang, Y. & Veronig, A. M. The Birth of A Coronal Mass Ejection. Science Advances 5, 7004 (2019). 1811.04707. + +[23] Cheng, X. et al. Observations of Turbulent Magnetic Reconnection within a Solar Current Sheet. Astrophys. J. 866, 64 (2018). 1808.06071. + +[24] Xue, Z. et al. Observing the release of twist by magnetic reconnection in a solar filament eruption. Nature Communications 7, 11837 (2016). + +[25] Sun, J. Q. et al. Extreme ultraviolet imaging of three- dimensional magnetic reconnection in a solar eruption. Nature Communications 6, 7598 (2015). 1506.08255. + +[26] Masuda, S., Kosugi, T., Hara, H., Tsuneta, S. & Ogawara, Y. A loop- top hard X- ray source in a compact solar flare as evidence for magnetic reconnection. Nature 371, 495- 497 (1994). + +[27] Arnold, H. et al. Electron Acceleration during Macroscale Magnetic Reconnection. arXiv e- prints arXiv:2011.01147 (2020). 2011.01147. + +[28] Warmuth, A. & Mann, G. Thermal- nonthermal energy partition in solar flares derived from X- ray, EUV, and bolometric observations. Discussion of recent studies. Astron. Astrophys. 644, A172 (2020). 2011.04442. + +<--- Page Split ---> + +[29] Innes, D. E., Guo, L.- J., Huang, Y.- M. & Bhattacharjee, A. IRIS Si IV Line Profiles: An Indication for the Plasmoid Instability during Small- scale Magnetic Reconnection on the Sun. Astrophys. J. 813, 86 (2015). 1509.08837. + +[30] Guo, L.- J., De Pontieu, B., Huang, Y.- M., Peter, H. & Bhattacharjee, A. Observations and Modeling of the Onset of Fast Reconnection in the Solar Transition Region. Astrophys. J. 901, 148 (2020). 2009.11475. + +[31] Kliem, B., Karlicky, M. & Benz, A. O. Solar flare radio pulsations as a signature of dynamic magnetic reconnection. Astron. Astrophys. 360, 715- 728 (2000). astro- ph/0006324. + +[32] French, R. J., Judge, P. G., Matthews, S. A. & van Driel- Gesztelyi, L. Spectropolarimetric Insight into Plasma Sheet Dynamics of a Solar Flare. Astrophys. J. Lett. 887, L34 (2019). 1911.12666. + +[33] Chitta, L. P. & Lazarian, A. Onset of Turbulent Fast Magnetic Reconnection Observed in the Solar Atmosphere. Astrophys. J. Lett. 890, L2 (2020). 2001.08595. + +[34] Tian, H. et al. Imaging and Spectroscopic Observations of Magnetic Reconnection and Chromospheric Evaporation in a Solar Flare. Astrophys. J. Lett. 797, L14 (2014). 1411.2301. + +[35] Yang, S., Zhang, J. & Xiang, Y. Magnetic Reconnection between Small- scale Loops Observed with the New Vacuum Solar Telescope. Astrophys. J. Lett. 798, L11 (2015). 1412.1314. + +[36] Wang, T., Sui, L. & Qiu, J. Direct Observation of High- Speed Plasma Outflows Produced by Magnetic Reconnection in Solar Impulsive Events. Astrophys. J. Lett. 661, L207- L210 (2007). 0709.2329. + +[37] Yu, S. et al. Magnetic Reconnection during the Post- impulsive Phase of a Long- duration Solar Flare: Bidirectional Outflows as a Cause of Microwave and X- Ray Bursts. Astrophys. J. 900, 17 (2020). 2007.10443. + +[38] Burch, J. L. et al. Electron- scale measurements of magnetic reconnection in space. Science 352, aaf2939 (2016). + +<--- Page Split ---> + +[39] Phan, T. D. et al. Electron magnetic reconnection without ion coupling in Earth's turbulent magnetosheath. Nature 557, 202- 206 (2018). + +[40] Liu, R., Chen, J., Wang, Y. & Liu, K. Investigating Energetic X- Shaped Flares on the Outskirts of A Solar Active Region. Scientific Reports 6, 34021 (2016). 1609.02713. + +[41] Kawabata, Y., Inoue, S. & Shimizu, T. Non- potential Field Formation in the X- shaped Quadrupole Magnetic Field Configuration. Astrophys. J. 842, 106 (2017). 1705.02560. + +[42] Liu, Z. et al. New vacuum solar telescope and observations with high resolution. Res. Astron. Astrophys. 14, 705- 718 (2014). + +[43] Schou, J. et al. Design and Ground Calibration of the Helioseismic and Magnetic Imager (HMI) Instrument on the Solar Dynamics Observatory (SDO). Solar Phys. 275, 229- 259 (2012). + +[44] Xue, Z., Yan, X., Yang, L., Wang, J. & Zhao, L. Observing Formation of Flux Rope by Tether- cutting Reconnection in the Sun. Astrophys. J. Lett. 840, L23 (2017). + +[45] Lin, R. P. et al. The Reuven Ramaty High- Energy Solar Spectroscopic Imager (RHESSI). Solar Phys. 210, 3- 32 (2002). + +[46] Culhane, J. L. et al. The EUV Imaging Spectrometer for Hinode. Solar Phys. 243, 19- 61 (2007). + +[47] Bhattacharjee, A., Huang, Y.- M., Yang, H. & Rogers, B. Fast reconnection in high- Lundquist- number plasmas due to the plasmoid Instability. Physics of Plasmas 16, 112102 (2009). 0906.5599. + +[48] Ni, L., Kliem, B., Lin, J. & Wu, N. Fast Magnetic Reconnection in the Solar Chromosphere Mediated by the Plasmoid Instability. Astrophys. J. 799, 79 (2015). 1509.06895. + +[49] Dere, K. P., Landi, E., Mason, H. E., Monsignori Fossi, B. C. & Young, P. R. CHIANTI - an atomic database for emission lines. Astron. Astrophys. Suppl. 125, 149- 173 (1997). + +<--- Page Split ---> + +[50] Landi, E. et al. CHIANTI- An Atomic Database for Emission Lines. VII. New Data for X- Rays and Other Improvements. Astrophys. J. Suplement Series 162, 261- 280 (2006). + +[51] Labrosse, N. et al. Physics of Solar Prominences: I—Spectral Diagnostics and Non- LTE Modelling. Space Sci. Rev. 151, 243- 332 (2010). 1001.1620. + +[52] Daughton, W. et al. Role of electron physics in the development of turbulent magnetic reconnection in collisionless plasmas. Nature Physics 7, 539- 542 (2011). + +[53] Longcope, D., Unverferth, J., Klein, C., McCarthy, M. & Priest, E. Evidence for Downflows in the Narrow Plasma Sheet of 2017 September 10 and Their Significance for Flare Reconnection. Astrophys. J. 868, 148 (2018). + +[54] Lemen, J. R. et al. The Atmospheric Imaging Assembly (AIA) on the Solar Dynamics Observatory (SDO). Solar Phys. 275, 17- 40 (2012). + +[55] Krucker, S. & Battaglia, M. Particle Densities within the Acceleration Region of a Solar Flare. Astrophys. J. 780, 107 (2014). + +[56] Kliem, B. Particle orbits, trapping, and acceleration in a filamentary current sheet model. Astrophys. J. Suplement Series 90, 719- 728 (1994). + +[57] Drake, J. F., Swisdak, M., Che, H. & Shay, M. A. Electron acceleration from contracting magnetic islands during reconnection. Nature 443, 553- 556 (2006). + +[58] Titov, V. S., Hornig, G. & Demoulin, P. Theory of magnetic connectivity in the solar corona. Journal of Geophysical Research (Space Physics) 107, 1164 (2002). + +[59] Priest, E. R. & Demoulin, P. Three- dimensional magnetic reconnection without null points. 1. Basic theory of magnetic flipping. J. Geophys. Res. 100, 23443- 23464 (1995). + +[60] Ji, H. & Daughton, W. Phase diagram for magnetic reconnection in heliophysical, astrophysical, and laboratory plasmas. Physics of Plasmas 18, 111207- 111207 (2011). 1109.0756. + +<--- Page Split ---> + +[61] Schuck, P. W. Tracking Magnetic Footpoints with the Magnetic Induction Equation. Astrophys. J. 646, 1358–1391 (2006). + +[62] Li, T. & Zhang, J. Slipping Magnetic Reconnection Triggering a Solar Eruption of a Triangle- shaped Flag Flux Rope. Astrophys. J. Lett. 791, L13 (2014). 1407.4180. + +[63] Xiang, Y.- y., Liu, Z. & Jin, Z.- y. High resolution reconstruction of solar prominence images observed by the New Vacuum Solar Telescope. New Astron 49, 8–12 (2016). + +[64] Bobra, M. G. et al. The Helioseismic and Magnetic Imager (HMI) Vector Magnetic Field Pipeline: SHARPs - Space- Weather HMI Active Region Patches. Solar Phys. 289, 3549–3578 (2014). 1404.1879. + +[65] Centeno, R. et al. The Helioseismic and Magnetic Imager (HMI) Vector Magnetic Field Pipeline: Optimization of the Spectral Line Inversion Code. Solar Phys. 289, 3531–3547 (2014). 1403.3677. + +[66] Borrero, J. M. et al. VFISV: Very Fast Inversion of the Stokes Vector for the Helioseismic and Magnetic Imager. Solar Phys. 273, 267–293 (2011). 0901.2702. + +[67] Metcalf, T. R. Resolving the 180- degree ambiguity in vector magnetic field measurements: The 'minimum' energy solution. Solar Phys. 155, 235–242 (1994). + +[68] Leka, K. D. et al. Resolving the \(180^{\mathrm{deg}}\) Ambiguity in Solar Vector Magnetic Field Data: Evaluating the Effects of Noise, Spatial Resolution, and Method Assumptions. Solar Phys. 260, 83–108 (2009). + +[69] Sun, X. On the Coordinate System of Space- Weather HMI Active Region Patches (SHARPs): A Technical Note. ArXiv e- prints (2013). 1309.2392. + +[70] Jiang, C., Wu, S. T., Feng, X. & Hu, Q. Data- driven magnetohydrodynamic modelling of a flux- emerging active region leading to solar eruption. Nature Communications 7, 11522 (2016). + +[71] Jiang, C. et al. Reconstruction of a Large- scale Pre- flare Coronal Current Sheet Associated with a Homologous X- shaped Flare. Astrophys. J. 850, 8 (2017). 1710.02775. + +<--- Page Split ---> + +[72] Wiegelmann, T. Optimization code with weighting function for the reconstruction of coronal magnetic fields. Solar Phys. 219, 87–108 (2004). 0802.0124. + +[73] Wiegelmann, T., Inhester, B. & Sakurai, T. Preprocessing of Vector Magnetograph Data for a Nonlinear Force- Free Magnetic Field Reconstruction. Solar Phys. 233, 215–232 (2006). astro- ph/0612641. + +[74] Schwartz, R. A. et al. RHESSI Data Analysis Software: Rationale and Methods. Solar Phys. 210, 165–191 (2002). + +[75] Liu, Y., Zhao, J. & Schuck, P. W. Horizontal Flows in the Photosphere and Subphotosphere of Two Active Regions. Solar Phys. 287, 279–291 (2013). + +## Acknowledgements + +We would like to thank the NVST, SDO/AIA, SDO/HMI, GOES, RHESSI, and Hinode teams for high- cadence data support. This work is sponsored by the National Science Foundation of China (NSFC) under the grant numbers (11873087, 11973084, 11803085, 2019FD085, 12003064, 11633008, 11763004, U1831210, 11803002), by the Yunnan Key Science Foundation of China under numbers (2018FA001, 2018FB007) and Yunnan Science Foundation for Distinguished Young Scholars, by Project Supported by the Specialized Research Fund for State Key Laboratories, by Key Research and Development Project of Yunnan Province202003AD150019, and by the joint NSFC/DFG Grant 41761134088/KL817.8- 1. + +## Author contributions + +X.L.Y. developed the ideas, performed analysis of the data, and lead the discussion of the manuscript. Z.K.X., L.H.Y., J.C.W., D.F.K., and Y.L.S. performed analysis of the data. C.W.J. carried out the MHD simulation. E.R.P. evaluated the energy of magnetic reconnection. B.K. contributed to the interpretation of the data and to the discussion. X.S.F. and Z.L. contributed to the discussion. All authors reviewed the manuscript. + +## Additional information + +Correspondence and requests for materials should be addressed to X.L.Y. and C.W.J (email: yanxl@ynao.ac.cn; chaowei@hit.edu.cn). + +Competing financial interests: The authors declare no competing financial interests. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: Magnetic reconnection during the confined flare, observed in the center of the \(\mathrm{H}\alpha\) line. (a) \(H\alpha\) image observed at 08:04:54 UT by NVST superimposed on the line-of-sight magnetogram observed by SDO/HMI. Red (blue) contours indicate positive (negative) polarity of \(\pm 200\) G and \(\pm 800\) G. (b) Line-of-sight magnetogram observed by SDO/HMI at 08:04:55 UT. The two curved red lines indicate the chromospheric fibrils and the active-region filament participating in the reconnection. (c–h) The process of reconnection as seen in the \(H\alpha\) observations. White arrows in c indicate the two reconnecting structures. Black arrows in d and e indicate the new chromospheric fibrils formed by reconnection. Blue arrows in d–f indicate the current sheet. The dotted lines in f and g indicate the cusp-shaped structures (magnetic separatrices) formed between the reconnection in- and outflows.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Profiles of GOES X-ray flux, inflows, and outflows during reconnection. (a) The profile of GOES X-rays in 1–8 Å (red line) and 0.5–4 Å (blue line). (b) The profile of the RHESSI X-ray counts in the 3–6, 6–12, 12–25, and 25–50 keV bands. (c) Inflow (dotted curve) along the A-B slit inferred from time-slices of the intensity. (d) Outflow along the C-D slit. (e) Outflow along the curved line E-F. (f) Outflow along the curved line G-H. All these lines are marked in Fig. 1h.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: Hinode/EIS observation at Fe XVII 255 Å during magnetic reconnection. (a–c) Evolution of the cusp-shaped structure in the intensity images. (d–f) Doppler velocity and (g–i) non-thermal velocity obtained from single Gaussian fitting.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: Fine structures in the SDO/AIA 211 Å images revealed by the unsharp masking technique. Red arrows mark the different bright blobs moving from the reconnection region (Fig. 4a–f). The green-dotted line in Fig. 4a indicates the trajectory I-J used to make the time-distance diagram of Fig. 4g. The projected velocities of the blobs lie in the range from \(79 \mathrm{km s^{-1}}\) to \(208 \mathrm{km s^{-1}}\) .
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Emission measure and temperature maps and \(\mathrm{H}\alpha\) images with RHESSI X-ray sources superimposed. (a–d): DEM maps in four temperature ranges. (e) Total emission measure. (f) Temperature of the reconnection region derived from six AIA wavelengths. (g–i): \(\mathrm{H}\alpha\) images with RHESSI hard X-ray sources superimposed. The levels of the contours are at 50% and 80% of the peak values in the 3–6, 6–12, 12–25, and 25–50 keV bands.
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6: RHESSI photon spectra accumulated around 08:12 UT and 08:19 UT during the quoted 1-minute intervals on 2014 February 2. The spectra are fitted by combining the \(\mathrm{vth}\) and thick2 functions (see Methods). The black line shows the photon spectrum with the background subtracted. The pink line shows the background. The green and the yellow lines display the fitting with the single-temperature thermal model and the thick-target nonthermal model, respectively.
+ +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7: Magnetic structure of the source region. (a) Selected field lines showing the twisted magnetic structure of the filament (green) and the loop-like structure of the chromospheric fibrils (red) before the reconnection event, obtained by using NLFFF extrapolation from the vector magnetogram. (b)–(c) Map of the magnetic squashing degree \(\log Q\) at heights \(Z = 0\) and \(Z = 2.9\) Mm above the photosphere. (d) Iso-surface \(\log Q = 4\) in the reconnection region. (e) \(\log Q\) in the vertical cross-section marked by a yellow dashed line in (a).
+ +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8: Evolution of magnetic field, current sheet, and plasmoids (blobs) during magnetic reconnection in the data-driven MHD simulation. (a) and (b): Evolution of the magnetic structure involved in the modeled reconnection event. (c) and (d): Zoom showing the magnetic structure of the current sheet at two selected times. The red boxes in (a) and (b) show the field of view of (c)–(f). (e) and (f): Distribution of electric current density ( \(Jc\) ) in the current sheet (also overlaid in (c)–(d)) at the height \(z = 1\) (11.6 Mm).
+ +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 9: Evolution of the photospheric vector magnetic field observed by SDO/HMI and flow field derived using the DAVE algorithm. (a–c): Vector magnetograms. The greyscale images show the vertical field component, saturated at 1000 G, and the arrows show the horizontal component. (d–f): Evolution of the flow field. Blue and red arrows indicate the horizontal photospheric velocity in the positive and negative polarities, respectively.
+ +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +Magnetic reconnection during the confined are, observed in the center of the Ha line. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Prols of GOES X- ray flux, in flows, and out flows during reconnection. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Hinode/EIS observation at Fe XVII 255A during magnetic reconnection. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Fine structures in the SDO/AIA 211 A images revealed by the unsharp masking technique. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5
+ +Emission measure and temperature maps and Ha images with RHESSI X- ray sources superimposed. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6
+ +RHESSI photon spectra accumulated around 08:12 UT and 08:19 UT during the quoted 1- minute intervals on 2014 February 2. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7
+ +Magnetic structure of the source region. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8
+ +Evolution of magnetic field, current sheet, and plasmoids (blobs) during magnetic reconnection in the data- driven MHD simulation. (see full caption in Manuscript file) + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 9
+ +Evolution of the photospheric vector magnetic field observed by SDO/HMI and ow field derived using the DAVE algorithm. (see full caption in Manuscript file) + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryMovie1. mp4 SupplementaryMovie2. mp4 SupplementaryMovie3. mp4 SupplementaryMovie4. mp4 SupplementaryMovie5. mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873_det.mmd b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b4557610725798490eea808c8ab4e52b6b8390c4 --- /dev/null +++ b/preprint/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873/preprint__13c57ef8358b42ecc5f0fb6544de8cdd5917504fb7d1bc61d4fa12dfd8db9873_det.mmd @@ -0,0 +1,609 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 955, 207]]<|/det|> +# Multi-scale Observation and Data-driven Modeling of Magnetic Reconnection in Flaring Chromospheric Plasmas on the Sun + +<|ref|>text<|/ref|><|det|>[[44, 229, 350, 250]]<|/det|> +Xiaoli Yan (yanxl@ynao.ac.cn) + +<|ref|>text<|/ref|><|det|>[[55, 253, 290, 270]]<|/det|> +Yunnan Observatories, CAS + +<|ref|>text<|/ref|><|det|>[[44, 277, 250, 315]]<|/det|> +Zhike Xue Yunnan Observatories + +<|ref|>text<|/ref|><|det|>[[44, 323, 771, 363]]<|/det|> +Chaowei Jiang Harbin Institute of Technology, Shenzhen https://orcid.org/0000- 0002- 7018- 6862 + +<|ref|>text<|/ref|><|det|>[[44, 369, 386, 409]]<|/det|> +Eric Priest School of Mathematics and Statistics + +<|ref|>text<|/ref|><|det|>[[44, 415, 608, 455]]<|/det|> +Bernhard Kliem University of Potsdam https://orcid.org/0000- 0002- 5740- 8803 + +<|ref|>text<|/ref|><|det|>[[44, 461, 250, 500]]<|/det|> +Liheng Yang Yunnan Observatories + +<|ref|>text<|/ref|><|det|>[[44, 507, 291, 546]]<|/det|> +Jincheng Wang Yunnan Observatories CAS + +<|ref|>text<|/ref|><|det|>[[44, 553, 250, 592]]<|/det|> +Defang Kong Yunnan Observatories + +<|ref|>text<|/ref|><|det|>[[44, 599, 936, 640]]<|/det|> +Yongliang Song Key Laboratory of Solar Activity, National Astronomical Observatories, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 645, 315, 685]]<|/det|> +Xueshang Feng Chinese academy of sciences + +<|ref|>text<|/ref|><|det|>[[44, 692, 576, 733]]<|/det|> +Zhong Liu Yunnan Observatories, Chinese Academy of Sciences, China + +<|ref|>sub_title<|/ref|><|det|>[[44, 777, 102, 794]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 814, 674, 834]]<|/det|> +Keywords: flaring chromospheric plasmas, planetary physics, solar flares + +<|ref|>text<|/ref|><|det|>[[44, 852, 333, 871]]<|/det|> +Posted Date: February 19th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 890, 463, 909]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 232633/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 911, 87]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 123, 940, 167]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 2nd, 2022. See the published version at https://doi.org/10.1038/s41467-022-28269-w. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[181, 210, 816, 303]]<|/det|> +# Multi-scale Observation and Data-driven Modeling of Magnetic Reconnection in Flaring Chromospheric Plasmas on the Sun + +<|ref|>text<|/ref|><|det|>[[188, 322, 885, 393]]<|/det|> +Xiaoli Yan \(^{1,2,3}\) , Zhike Xue \(^{1,3}\) , Chaowei Jiang \(^{4}\) , E. R. Priest \(^{5}\) , Bernhard Kliem \(^{6}\) , Liheng Yang \(^{1,3}\) , Jincheng Wang \(^{1,3}\) , Defang Kong \(^{1,3}\) , Yongliang Song \(^{7}\) , Xueshang Feng \(^{2}\) , Zhong Liu \(^{1,8}\) + +<|ref|>text<|/ref|><|det|>[[406, 406, 587, 427]]<|/det|> +February 11, 2021 + +<|ref|>text<|/ref|><|det|>[[179, 465, 817, 720]]<|/det|> +Magnetic reconnection is a multi- faceted process of energy conversion in astrophysical, space and laboratory plasmas that operates at microscopic scales but has macroscopic drivers and consequences. Solar flares present a key laboratory for its study, leaving imprints of the microscopic physics in radiation spectra and allowing the macroscopic evolution to be imaged, yet a full observational characterization remains elusive. Here we combine high resolution imaging and spectral observations of a solar flare at multiple wavelengths with data- driven magnetohydrodynamic modeling to study the dynamics of the involved plasma from the current sheet to the plasmoid scale. The flare resulted from the interaction of a twisted filament and chromospheric fibrils. By inferring the reconnection to be fast and mediated by plasmoids, the relevance of this reconnection mode is found to extend beyond hot flare plasmas to such cool structures in the chromosphere, which have many analogs in astrophysical objects. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[177, 157, 818, 343]]<|/det|> +4 Institute of Space Science and Applied Technology, Harbin Institute of Technology, Shenzhen, 5180055, China5 Mathematics Institute, University of St Andrews, St Andrews, KY16 9SS, UK.6 Institute of Physics and Astronomy, University of Potsdam, Potsdam 14476, Germany.7 Key Laboratory of Solar Activity, National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100012, China.8 University of Chinese Academy of Sciences, Yuquan Road, Shijingshan Block, Beijing 100049, Peoples Republic of China. + +<|ref|>sub_title<|/ref|><|det|>[[178, 374, 301, 391]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[179, 393, 816, 647]]<|/det|> +Solar flares result from the most dramatic events of energy release by magnetic reconnection on the Sun and provide a broad range of observational detail about the involved processes[1- 3]. In the standard flare model for eruptive events[4- 7], magnetic reconnection proceeds in a vertical current sheet spawned on the underside of an erupting flux rope which also drives a coronal mass ejection (CME). Rising arcades of reconnected hot, initially cuspshaped magnetic loops ("flare loops"), rooted in a pair of bright "flare ribbons", are formed in the downward reconnection outflow. Confined flares result when the erupting flux is stopped by overlying flux or when neighboring flux system are pressed against each other, and a reconnecting current sheet is formed in the interface[8- 10]. A filament extending along the polarity inversion line of the source region often traces the activated flux. These large- scale geometries of the reconnecting current sheet and its drivers and consequences are well established[11- 20]. + +<|ref|>text<|/ref|><|det|>[[179, 648, 816, 847]]<|/det|> +High resolution imaging observations provide rich information about details of the reconnection process which result from its internal dynamics, especially the formation of plasmoids[21,22] and turbulent structures[23], and from the complexity and three- dimensional (3D) nature of the solar magnetic field[24,25]. Spectral and multi- wavelength data reveal aspects of the microscopic processes, e.g., particle acceleration[26,27], the nonthermal- thermal energy partition[28], the plasmoid instability[29,30], and bursty[31] or turbulent reconnection[32,33], in addition to flows[34]. Finally, besides hot flare plasmas, the Sun also allows the reconnection of cool chromospheric structures to be studied[24,35]. However, despite this broad range of information, comprehensive simultaneous imaging and spectral observations are still limited. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 157, 816, 268]]<|/det|> +In particular, while individual components of the reconnection in- and outflows have often been observed[15,18,21,34,36,37], we are aware of only two events revealing all four flows[19,35] but these have no or only limited spectral information about the involved plasma. Such limitation prevents the mode of reconnection, i.e. Petschek- like[12], plasmoid- mediated[29], collisionless[38], or turbulent[39], to be inferred. + +<|ref|>text<|/ref|><|det|>[[179, 268, 816, 469]]<|/det|> +Here we present multi- wavelength imaging and spectral evidence for magnetic reconnection in a strong (M- class) confined solar flare, traced by the cool plasma of the interacting filament and chromospheric fibrils and driven by emerging flux. The flare is unique in that it displayed all four reconnection flows, plasmoids not only in the current sheet but also in the separatrices, was fully covered by spectroscopic observations, and yielded the first hard X- ray images from a current sheet on the solar disk. This allows us to measure both inflows and both outflows and to infer the mode of reconnection from the analysis of the physical conditions in the current sheet. We also perform a state- of- the- art, data- driven numerical simulation of the reconnection event, which supports our interpretation of the observational data. + +<|ref|>sub_title<|/ref|><|det|>[[179, 485, 245, 500]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[179, 502, 816, 628]]<|/det|> +High- resolution imaging of the reconnection flows. The confined M2.2 flare occurred in Active Region NOAA 11967 at S13E04 on 2014 February 2, starting at 07:17 UT, peaking at 08:20 UT, and ending at 08:29 UT. This was a large active region with a \(\beta \gamma \delta\) sunspot configuration. Several of its main flux components were already decaying, but there was also new flux emerging into the region. Driven by these changes, many C- and M- class flares were produced during its passage across the solar disk[40,41]. + +<|ref|>text<|/ref|><|det|>[[179, 630, 816, 848]]<|/det|> +Figure 1a shows an Hα image from the New Vacuum Solar Telescope (NVST)[42] superimposed with magnetic flux contours from the Solar Dynamics Observatory (SDO)/Helioseismic and Magnetic Imager (HMI)[43] magnetogram in Fig. 1b. The red curved lines in Fig. 1b indicate the positions of the reconnecting structures: chromospheric fibrils on the east (left) side and a filament, which was contained in a twisted flux rope[44], on the west (right) side. The evolution of the flare as seen by the NVST in Hα is displayed in Fig. 1c- h and the related Supplementary Movie 1. Here the white arrows indicate the reconnecting structures at 08:04:54 UT. The structures approached each other and began to reconnect. What may well be a current sheet can be seen at the junction of the two structures marked by the blue arrows in Fig. 1d- f. The newly formed threads are indicated by black arrows + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 159, 816, 268]]<|/det|> +in Fig. 1d,e. Both sets of new, reconnected structures displayed the cusp characteristic of newly reconnected field lines, as outlined by the dashed blue curved lines in Fig. 1f,g. Simultaneously with the magnetic reconnection, the M- class flare occurred (see the GOES soft X- ray light curve in Fig. 2a and the Reuven Ramaty High- Energy Solar Spectroscopic Imager (RHESSI) [45] X- ray counts in the 3–6, 6–12, 12–25, and 25–50 keV bands in Fig. 2b). + +<|ref|>text<|/ref|><|det|>[[179, 269, 817, 613]]<|/det|> +The projected in- and outflow velocities were measured using time- slices of the intensity along the artificial slits A- B and C- D shown by the red and blue lines, respectively, in Fig. 1h. Up to 08:14 UT, the inflows are found to be \(v_{\mathrm{i}} = 0.06 \mathrm{km s}^{- 1}\) on the east side and \(v_{\mathrm{i}} = 0.05 \mathrm{km s}^{- 1}\) on the west side; see the dashed red lines in Fig. 2c. The outflows are measured from \(v_{\mathrm{o}} = 2.0\) to \(3.7 \mathrm{km s}^{- 1}\) in southward direction, which is presumably downward, and \(v_{\mathrm{o}} = 3.4\) to \(3.8 \mathrm{km s}^{- 1}\) in northward direction, which is presumably upward; see the dashed blue lines in Fig. 2d. The resulting reconnection rate, \(M = v_{\mathrm{i}} / v_{\mathrm{o}} = 0.01–0.03\) , clearly indicates a fast reconnection process. We also measure the outflows along the cusp lines (magnetic separatrices) attached to the ends of the current sheet (see the curved lines E- F and G- H in Fig. 1h). These are found to lie in the ranges \(2.3–5.5 \mathrm{km s}^{- 1}\) and \(0.8–2.6 \mathrm{km s}^{- 1}\) for the south and north cusp lines, respectively (see Fig. 2d,e). As the reconnection continues, an unusual feature of the event is that the chromospheric fibrils and the main body of the filament began to separate from the reconnection region after 08:14 UT with velocities of 0.98 and \(0.39 \mathrm{km s}^{- 1}\) on the east and west sides, respectively (see Fig. 2b), whereas some of the lower chromospheric fibrils and threads of the filament still participated in the reconnection process. This is considered further in the Discussion. + +<|ref|>text<|/ref|><|det|>[[179, 614, 816, 850]]<|/det|> +EUV spectroscopy of reconnection flows and turbulence. The Hinode/EUV Imaging Spectrometer (EIS) [46] observations covered essentially the whole reconnection event. Figure 3 shows the intensity (a–c), Doppler velocity (d–f), and non- thermal velocity (g–i). There is a strong red shift in the southern cusp- shaped structure, with the maximum Doppler velocity reaching about \(80 \mathrm{km s}^{- 1}\) . The blue shift at the northern cusp reaches about \(40 \mathrm{km s}^{- 1}\) . However, the true peak velocity may be higher at the north side (as also indicated by the projected velocities), because the selected spectral line obviously samples only part of the upward reconnection outflow, due to a reduction of density and temperature (adiabatic cooling) in the free upward expansion of the plasma. Nevertheless, both reconnection outflows are spectroscopically detected and prove to be significantly higher near the edges of the current sheet than the outflows inferred from the \(\mathrm{H}\alpha\) structures + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 158, 816, 504]]<|/det|> +further away from the current sheet (Figs. 1h and 2d). It is also clear that the outflow is mainly downward at the southern edge of the current sheet and mainly upward at the northern edge. The implied reconnection rate, \(M \sim 0.0007 - 0.024\) for the southern outflow, is still fast. The cusp- shaped structures also appear in the non- thermal velocity maps. The highest non- thermal velocities were located in the current sheet, especially in the tip of the cusp- shaped structures, with values in the range \(\sim 180 - 300 \mathrm{km s^{- 1}}\) , which significantly exceed the projected and Doppler velocities of the reconnection outflows. This suggests that the reconnection proceeds in a highly dynamic manner, with plasmoid- mediated reconnection [47,48] being the obvious candidate for the specific mode of reconnection realized in the event. To determine the electron density in the current sheet, we used the theoretical line intensity ratios of the Fe XIV ( \(\lambda 264.92 / \lambda 274.37\) ) density sensitive pair, which were calculated by using the CHIANTI atomic package [49] [50]. Although the line \(\lambda 274.37\) is blended with Si VII \(\lambda 274.18\) , the blend can be safely ignored due to its low intensity compared with \(\lambda 274.37\) in active region conditions. The number density values in the current sheet are found to scatter in the range from \(10^{8}\) to \(10^{12} \mathrm{cm^{- 3}}\) , similar to typical densities in filaments [51]. The highest values are located in the southern (lower) cusp. + +<|ref|>text<|/ref|><|det|>[[179, 504, 816, 850]]<|/det|> +Plasmoids. During the early, slow- rise phase of the flare, from 07:57 to \(\sim 08:06\) UT, many small bright blobs were observed, moving from the current sheet to the legs of the cusp- shaped structures (see Supplementary Movie 1). Figure 4 and and the related Supplementary Movie 2 show the highlighted fine structures of the SDO/AIA 211 Å images revealed by the unsharp masking technique. There are many bright blobs marked by the red arrows moving from the reconnection region (see Fig. 4a–f). The velocities of the blobs along the reconnected field lines are from \(79 \mathrm{km s^{- 1}}\) to \(208 \mathrm{km s^{- 1}}\) (see the time- distance diagram, Fig. 4g, obtained along the green- dotted line in Fig. 4a). To the best of our knowledge, the phenomenon that blobs move from the reconnection region along the separatrices, as found in a 3D particle simulation of reconnection [52], is observed here for the first time. Their number, more than ten, is among the highest reported so far for a single reconnection event [21,53]. They support the interpretation of the high Fe XVII line widths as an indication of plasmoid- mediated reconnection. The increasing brightness and source width of the emissions prevented the detection of blobs by unsharp masking at later times, but their continued formation is suggested by the continued presence of high nonthermal velocities in the EIS spectra, efficient particle acceleration revealed by the hard X- rays, and the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 159, 402, 176]]<|/det|> +numerical modeling below. + +<|ref|>text<|/ref|><|det|>[[179, 177, 816, 286]]<|/det|> +Heating. The heating and brightening resulting from the reconnection can also clearly be seen in all extreme ultraviolet (EUV) and UV wavelengths of the SDO/Atmospheric Imaging Assembly (AIA) [54]. The Supplementary Movie 3 shows the evolution of the EUV and UV observations at eight wavelengths and of the line- of- sight magnetogram. The cusp- shaped regions of outflowing hot plasma are obvious, as is the current sheet between them. + +<|ref|>text<|/ref|><|det|>[[179, 286, 816, 412]]<|/det|> +Six EUV wavelengths observed by AIA were used to obtain the differential emission measure (DEM), total EM, and temperature maps shown in Fig. 5a–f and Supplementary Movie 4. The X- shaped geometry of the reconnection region can be seen in the whole temperature range of \(T > 1\) MK, i.e., the cool plasma of the interacting filament and chromospheric fibrils is heated to coronal and flare temperatures. The peak temperature in the reconnection region reaches about 10 MK. + +<|ref|>text<|/ref|><|det|>[[179, 413, 816, 650]]<|/det|> +X- ray spectroscopy of accelerated particles. The X- ray sources from RHESSI in Fig. 5g–i show that X- ray emission \(>3\) keV is emitted from the beginning of the event’s slow- rise phase (Fig. 5g). Higher energies are reached following the development of the flare, and the highest energy release rate into hard X- rays around 08:18 UT coincides with the steepest slope of the soft X- ray light curve (Fig. 2a), as is typically the case in flares. The emission comes mainly from the current sheet, different from eruptive flares, where the foot point sources are usually strongest. Thus, the hard X- rays provide a direct diagnostic of the conditions in the current sheet. These are the first maps of hard X- ray emission from a reconnecting current sheet in a confined flare on disk. The emission originates preferentially from the lower (southern) cusp, where the density in the sheet was found to be highest, similar to limb flares with occulted foot points [55]. + +<|ref|>text<|/ref|><|det|>[[179, 652, 816, 833]]<|/det|> +Figure 6 shows the RHESSI hard X- ray photon spectra taken at about 08:12 UT and 08:19 UT. The temperatures derived from the fitted spectra are 13.1 MK (Fig. 6a) and 33.6 MK (Fig. 6b); this extends the results of the DEM analysis of the AIA EUV data which are limited to \(\approx 11\) MK. Moreover, the spectra show both thermal and non- thermal components. The emission above \(\sim 25\) keV is of non- thermal nature, indicating particle acceleration in the current sheet by the fast reconnection process. Particle acceleration by reconnection is known to be particularly efficient when multiple plasmoids form, implying multiple X- lines [27,56,57]. This corroborates the deduction above that plasmoid- mediated reconnection was at work. + +<|ref|>text<|/ref|><|det|>[[205, 833, 814, 851]]<|/det|> +Magnetic field structure. Using the nonlinear force free field (NLFFF) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 158, 816, 540]]<|/det|> +extrapolation of the photospheric vector magnetogram from SDO/HMI at 07:36 UT (see Methods), the magnetic structure of the filament and the involved magnetic loops prior to the onset of reconnection are obtained. Figure 7 shows the magnetic structure of the filament, the magnetic loops, and the map and iso- surface of the magnetic squashing degree \(\mathrm{Q}^{[58]}\) , which measures the gradient in the field line foot point mapping. High values of \(\mathrm{Q}\) , known as quasi- separatrix layers (QSLs), are the preferred locations for the formation of current sheets and the onset of reconnection \({}^{[59]}\) . As can be seen in the NLFFF extrapolation, the magnetic field lines from the main positive sunspot in the east part of the active region connect to the diffuse negative polarity in the north (the pink lines), while the field lines of the filament from the smaller positive spot in the west connect to the main negative spot (the green lines). The latter, weakly twisted magnetic structure (Fig. 7a) is in good agreement with the shape of the filament in the \(\mathrm{H}\alpha\) center observation (Fig. 1a). Figures 7b- c show the map of \(\log Q\) in the reconnection region at heights \(Z = 0\) and \(Z = 2.9\) Mm above the photosphere. The photospheric magnetogram and the iso- surface \(\log Q = 4\) in Fig. 7d reveal the X- shaped structure of high \(\log Q\) values that facilitates the onset of reconnection at the observed position. The vertical cross- section of \(\log Q\) in Fig. 7e demonstrates that the separatrices run between the flux bundles, at the position where the current sheet forms. + +<|ref|>text<|/ref|><|det|>[[179, 541, 816, 814]]<|/det|> +MHD simulation. We further simulated the reconnection process using a data- driven high- resolution magnetohydrodynamic (MHD) model. Since this realistic, data- driven model of the current sheet includes a guide field component in the direction of the current flow, the plasmoids seen in the current density distribution (Fig. 8e- f) are small, three- dimensional flux ropes in the field line plots \({}^{[52]}\) (Fig. 8a- d). As observed in two- dimensional simulations \({}^{[47,48]}\) , as well as in the observations presented above, plasmoids form dynamically and repeatedly, and are accelerated along the current sheet. Two new plasmoids are seen to form in Fig. 8f. Between the plasmoids, the current sheet steepens to the limit allowed by the numerical resolution. The simulation obviously corresponds well to the observations from the global to the plasmoid scale. The plasmoids appeared only after adaptive mesh refinement was applied to permit the current sheet width to shrink to an aspect ratio of \(\sim 0.01\) , in agreement with the theoretical prediction for the occurrence of the plasmoid instability \({}^{[60]}\) . + +<|ref|>text<|/ref|><|det|>[[180, 815, 816, 851]]<|/det|> +Released energy. A rough estimate of the way reconnection releases the magnetic energy required to power an M- class flare ( \(\sim 10^{24}\) J) can be made + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[178, 157, 817, 305]]<|/det|> +as follows. We expect that part of the energy is released in the reconnecting current sheet and a further part is released by the untwisting of the filament- flux rope after it is connected to the diffuse remote polarity. Suppose first that oppositely directed magnetic field of strength \(B_{\mathrm{i}}\) is brought in from both sides at speed \(v_{\mathrm{i}}\) for a time \(\tau\) towards a reconnection region of surface area \(L_{0}{}^{2}\) , where the current sheet and attached slow- mode shock waves convert the energy. The magnetic energy is brought in from both sides from a volume \(2L_{0}L_{\mathrm{i}}\) , where \(L_{\mathrm{i}} = v_{\mathrm{i}}\tau\) , and so the total energy calculated is + +<|ref|>equation<|/ref|><|det|>[[427, 315, 814, 355]]<|/det|> +\[W = \frac{B_{\mathrm{i}}^{2}}{2\mu_{0}} L_{0}{}^{2}v_{\mathrm{i}}\tau . \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[178, 365, 817, 495]]<|/det|> +This may be evaluated for the M- class flare, where \(B_{\mathrm{i}} \sim 500 \mathrm{G}\) from the extrapolation of the magnetogram which is used as the initial condition for the simulation (the average value in the reconnection region), and the values \(L_{0} = 15^{\prime \prime}\) (10.9 Mm), \(v_{\mathrm{i}} = 2 \mathrm{km s}^{- 1}\) , and \(\tau = 70 \mathrm{min}\) are obtained from the \(\mathrm{H}\alpha\) data, to give an energy of \(\sim 10^{24} \mathrm{J}\) , as required. This corresponds well to the energy release of \(\sim 10^{24} \mathrm{J}\) (9.6 \(\times 10^{30}\) erg) in the simulation and agrees with the canonical picture that the flare is powered by magnetic reconnection. + +<|ref|>text<|/ref|><|det|>[[178, 495, 817, 550]]<|/det|> +Additional magnetic energy may be liberated by the untwisting of the flux rope which holds the filament. The energy associated with the twist component of the flux rope field, \(B_{\phi} = (R / a)B_{0\phi}\) , can be written as + +<|ref|>equation<|/ref|><|det|>[[413, 561, 814, 601]]<|/det|> +\[W_{\mathrm{FR}} = \frac{B_{0\phi}{}^{2}}{2\mu_{0}}\pi a^{2}L_{\mathrm{R}}, \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[178, 612, 817, 741]]<|/det|> +where \(R\) , \(a\) and \(L_{\mathrm{R}}\) are the rope's major and minor radii and length. Suppose the twist on the edge of the flux rope is \(\Phi\) , then \(\Phi = (L B_{0\phi}) / (a B_{z})\) and \(B_{0\phi}\) is \(\mathrm{a}\Phi B_{z} / L_{\mathrm{R}}\) . In the present case, we estimate \(B_{z} \sim 500 \mathrm{G}\) , \(L_{\mathrm{R}} = 100^{\prime \prime}\) , \(a = 5^{\prime \prime}\) and a twist of \(\Phi \sim 4\pi\) , so \(W_{\mathrm{FR}} = 3 \times 10^{23} \mathrm{J}\) . This is significantly smaller than the energy released in the reconnecting current sheet, unless the flux tube radius is a factor of \(\sim 2\) larger than we have estimated from the \(\mathrm{H}\alpha\) images of the filament. + +<|ref|>text<|/ref|><|det|>[[178, 742, 816, 851]]<|/det|> +In summary, the presented high- resolution imaging and spectral observations of macroscopic flows demonstrate that fast magnetic reconnection of the field in structures originally at chromospheric temperatures released the magnetic energy which powered the solar flare. Plasmoid- mediated reconnection is strongly indicated by direct imaging and nonthermal EUV and hard X- ray spectra. This is further supported by a 3D data- driven MHD simulation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 159, 816, 213]]<|/det|> +which reproduces the observed change in connectivity, agrees with estimates of the energy release, and reveals the appearance of twisted plasmoids (mini flux ropes) at the theoretically expected threshold. + +<|ref|>sub_title<|/ref|><|det|>[[179, 230, 273, 246]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[179, 247, 816, 575]]<|/det|> +In order to understand why reconnection occurred so many times in this place, the evolution of the magnetic and flow fields in the photosphere was investigated using HMI vector magnetograms. The vector magnetogram is shown in Fig. 9a–c, where the arrows represent the horizontal magnetic field. The corresponding flow field was calculated from the radial magnetic field using the DAVE algorithm[61] (see Methods); see Fig. 9d–f and the related Supplementary Movie 5. This reveals a flow toward the magnetic reconnection site from the west side, which pushes together oppositely directed flux in the higher atmosphere (as found in the MHD simulation). This persistent flow in the photosphere is likely to be the main reason why so many flares occurred in the same place. Additional distortions of the coronal field configuration, also likely to support the onset of reconnection, are produced by the emergence of flux very near the reconnection site (green ellipses in Fig. 9b,c) and subsequent motion toward the reconnection site. The flows at the east side are more variable, with directions along and away from the current sheet alternating. The latter can explain the separation of the reconnecting flux after \(\sim 08:14\) UT in the present event, as well as the termination of the other homologous events on the same day. + +<|ref|>text<|/ref|><|det|>[[179, 575, 816, 849]]<|/det|> +A neighboring eruption may have had an additional influence. This active region produced three M- class flares and several C- class flares in the first half of 2014 February 2. All these flares were located at the same place and had a similar morphology. The M- class flare studied here was preceded by a weaker, C- class flare above another polarity inversion line in the complex active region, lying to the east of the main positive sunspot, at \((x,y)\approx (- 340, - 50)\) . The C- class flare was caused by the rise of flux in the corona above the filament located at that inversion line. The associated large- scale change of the coronal field, due to the expansion of that flux, may have played an additional role in pushing the chromospheric fibrils east of the reconnection region toward the filament on the west side of the region. With the relaxation of the overlying coronal field during the decay of the C- class flare (07:17–07:58 UT)[62], the fibrils may have experienced a restoring, oppositely directed force, dragging them apart from the formed current sheet and terminating the reconnection inflow. This might have additionally + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 159, 658, 177]]<|/det|> +contributed to the termination of the reconnection event. + +<|ref|>text<|/ref|><|det|>[[179, 177, 816, 413]]<|/det|> +Once the reconnection in- and outflows are established as inferred from the \(\mathrm{H}\alpha\) data, reconnection would not easily cease by itself, because the process involves a positive feedback: the acceleration of the outflows drives continued inflows which transports new flux with an oppositely directed component toward the sheet. This is consistent with the MHD simulation. The fact that the reconnection event produced an M2.2 flare, more intense by a factor two than the preceding C8.9 flare, shows that the active region had stored free magnetic energy in the area of the reconnection event. This free energy was carried by the electric current associated with the antiparallel field component of the reconnecting structures. Since these structures still existed after the flare, the alternative possibility for the relatively quick ending of the reconnection event by a depletion of the available free magnetic energy appears less likely than the action of an external driver. + +<|ref|>sub_title<|/ref|><|det|>[[179, 428, 259, 445]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[179, 446, 816, 555]]<|/det|> +Observational data Images in the center of the \(\mathrm{H}\alpha\) line at 6562.8 A obtained at the 1 m New Vacuum Solar Telescope (NVST) at Fuxian Lake[42], are used in this study. They have a pixel size of \(0.^{\prime \prime}163\) and a cadence of 12 s. The data are calibrated from Level 0 to Level 1 with dark current subtracted and flat field corrected, and then the calibrated images are reconstructed to Level 1+ using a speckle masking method[63]. + +<|ref|>text<|/ref|><|det|>[[179, 555, 816, 792]]<|/det|> +Full- disk UV and EUV images observed by SDO/AIA[54] are also used to clarify the process of magnetic reconnection. These images have a 12- s cadence and a spatial resolution of \(0.^{\prime \prime}6\) per pixel. Vector magnetograms from the Space Weather HMI Active Region Patch (SHARP) series observed by the Helioseismic and Magnetic Imager (HMI)[43,64,65] on board SDO are used. These allow us to infer the photospheric velocity field and to perform a data- driven simulation of the event. The vector magnetograms have a pixel scale of about \(0.^{\prime \prime}5\) and a cadence of 12 minutes. They are derived by using the Very Fast Inversion of the Stokes Vector algorithm[66]. The minimum energy method[67,68] is used to resolve their 180 degree azimuthal ambiguity. The images are remapped using a Lambert (cylindrical equal area) projection centered on the midpoint of the AR, which is tracked at the Carrington rotation rate[69]. + +<|ref|>text<|/ref|><|det|>[[180, 793, 815, 848]]<|/det|> +The spectral data observed by the EUV Imaging Spectrometer (EIS) on board the Hinode mission were used to derive the Doppler velocity and nonthermal velocity in the reconnection region. Hinode/EIS[46] is a scanning slit + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 158, 816, 358]]<|/det|> +spectrometer that uses two wave bands in the EUV: 170–210 Å and 250–290 Å. The spectral resolution is about 0.0223 Å pixel\(^{- 1}\). It can measure velocities at an accuracy of a few km s\(^{- 1}\). Fortunately, the EIS observation covered the whole duration of the reconnection event and nearly its whole area. It thus provides an excellent opportunity to conduct a spectroscopic study of magnetic reconnection. We use the line Fe XVII 255 Å (log \(T = 6.6\)). The field of view (FOV) is 161'' in the raster direction and 152'' in the slit direction. The spectral profiles are fitted with single Gaussian functions. The reference wavelength used to determine the Doppler and non- thermal velocities is taken from the average of each raster. The sparse raster used the 2'' slit with a step size of 2. ''995 and an exposure time of 4–6 s. + +<|ref|>text<|/ref|><|det|>[[179, 359, 816, 450]]<|/det|> +Data from RHESSI[45] are also employed to study the soft and hard X- ray sources during the M- class flare. The energy ranges of 3–6, 6–12, 12–25, and 25–50 keV were selected to observe the evolution of these sources during the magnetic reconnection. The accumulation time is 30 s for constructing X- ray images by using the Clean imaging algorithm. + +<|ref|>text<|/ref|><|det|>[[179, 465, 816, 720]]<|/det|> +Data- driven simulation. The full set of time- dependent 3D MHD equations were solved, imposing bottom boundary conditions driven continuously[70,71] by the changing photospheric vector magnetic field observed by SDO/HMI (see Observational data). The background plasma is initially in a hydrostatic, isothermal state with \(T = 10^{6}\) K (sound speed \(c_{S} = 128 \mathrm{km s}^{- 1}\) ) in solar gravity. Its density is configured to make the plasma \(\beta\) as small as \(2 \times 10^{- 3}\) (the maximum Alfvén velocity \(v_{\mathrm{A}}\) is \(4 \mathrm{Mm s}^{- 1}\) ) to mimic the coronal low- \(\beta\) and highly tenuous conditions. The plasma thermodynamics are simplified to be adiabatic. The bottom boundary of the model is assumed to be the coronal base, and also the magnetic field measured on the photosphere is used as a reasonable approximation to the field at the coronal base. To obtain the detailed structure in the current sheet, an adaptive refinement mesh is used with highest resolution of \(1 / 16\) arcsecond (i.e., \(45 \mathrm{km}\) ) to capture the fine structures in the reconnecting current sheet. + +<|ref|>text<|/ref|><|det|>[[179, 735, 816, 844]]<|/det|> +Unsharp masking technique. The unsharp masking technique is applied to SDO/AIA 211 Å images to highlight fine structures like blobs. First, a background image by smoothing the original image is obtained as an unsharp masked image. Second, the enhanced image is the residual of the original image subtracting the background one. A smoothing window of \(7 \times 7\) pixels is adopted for optimal effects. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 158, 816, 304]]<|/det|> +NLFFF extrapolation. The three- dimensional magnetic structures of the filament and the magnetic loops are reconstructed by using an optimization algorithm[72]. The bottom vector magnetic field data observed by SDO/HMI are modified to remove most of the net force and torque that usually results in an inconsistency between the photospheric magnetic field and the force- free assumption in the NLFFF models by using a preprocessing procedure[73]. The visualization of the 3D magnetic field is realized by the software Paraview. + +<|ref|>text<|/ref|><|det|>[[179, 320, 816, 411]]<|/det|> +RHESSI spectrum. RHESSI hard X- ray photon spectra were created by the Object Spectral Executive (OSPEX)[74] software with the background subtracted. A single- temperature thermal bremsstrahlung radiation function ("vth") and the thick- target non- thermal bremsstrahlung with an isotropic pitch- angle distribution ("thick2") were adopted to fit the spectra. + +<|ref|>text<|/ref|><|det|>[[179, 424, 865, 535]]<|/det|> +Velocity field calculation. The differential affine velocity estimator (DAVE)[61,75] has been applied to the HMI vector magnetograms to derive the photospheric flow field. The DAVE method combines the advection equation and a differential feature tracking technique to detect flows. We used a window size of 19 pixels, which is large enough to include structure information and small enough to have a good spatial resolution. + +<|ref|>sub_title<|/ref|><|det|>[[179, 561, 328, 585]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[186, 599, 816, 636]]<|/det|> +[1] Priest, E. R. & Forbes, T. G. The magnetic nature of solar flares. Astron. Astrophys. Rev. 10, 313- 377 (2002). + +<|ref|>text<|/ref|><|det|>[[186, 648, 816, 686]]<|/det|> +[2] Fletcher, L. et al. An Observational Overview of Solar Flares. Space Sci. Rev. 159, 19- 106 (2011). 1109.5932. + +<|ref|>text<|/ref|><|det|>[[186, 697, 816, 735]]<|/det|> +[3] Benz, A. O. Flare Observations. Living Reviews in Solar Physics 14, 2 (2017). + +<|ref|>text<|/ref|><|det|>[[186, 747, 816, 785]]<|/det|> +[4] Carmichael, H. A Process for Flares. NASA Special Publication 50, 451 (1964). + +<|ref|>text<|/ref|><|det|>[[186, 796, 816, 834]]<|/det|> +[5] Sturrock, P. A. Model of the High- Energy Phase of Solar Flares. Nature 211, 695- 697 (1966). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 157, 818, 195]]<|/det|> +[6] Hirayama, T. Theoretical Model of Flares and Prominences. I: Evaporating Flare Model. Solar Phys. 34, 323- 338 (1974). + +<|ref|>text<|/ref|><|det|>[[182, 207, 816, 244]]<|/det|> +[7] Kopp, R. A. & Pneuman, G. W. Magnetic reconnection in the corona and the loop prominence phenomenon. Solar Phys. 50, 85- 98 (1976). + +<|ref|>text<|/ref|><|det|>[[183, 255, 816, 310]]<|/det|> +[8] Ji, H., Wang, H., Schmahl, E. J., Moon, Y. J. & Jiang, Y. Observations of the Failed Eruption of a Filament. Astrophys. J. Lett. 595, L135- L138 (2003). + +<|ref|>text<|/ref|><|det|>[[183, 323, 816, 378]]<|/det|> +[9] Török, T. & Kliem, B. Confined and Ejective Eruptions of Kink-unstable Flux Ropes. Astrophys. J. Lett. 630, L97- L100 (2005). astro- ph/0507662. + +<|ref|>text<|/ref|><|det|>[[178, 390, 816, 427]]<|/det|> +[10] Li, L. et al. Magnetic reconnection between a solar filament and nearby coronal loops. Nature Physics 12, 847- 851 (2016). 1605.03320. + +<|ref|>text<|/ref|><|det|>[[178, 439, 816, 494]]<|/det|> +[11] Tsuneta, S. et al. Observation of a solar flare at the limb with the YOHKOH Soft X-ray Telescope. Pub. Astron. Soc. Japan 44, L63- L69 (1992). + +<|ref|>text<|/ref|><|det|>[[178, 506, 816, 543]]<|/det|> +[12] Tsuneta, S. Structure and Dynamics of Magnetic Reconnection in a Solar Flare. Astrophys. J. 456, 840 (1996). + +<|ref|>text<|/ref|><|det|>[[178, 555, 816, 592]]<|/det|> +[13] Shibata, K. Evidence of Magnetic Reconnection in Solar Flares and a Unified Model of Flares. Astrophys. Space Sci. 264, 129- 144 (1999). + +<|ref|>text<|/ref|><|det|>[[178, 604, 816, 641]]<|/det|> +[14] Lin, J. & Forbes, T. G. Effects of reconnection on the coronal mass ejection process. J. Geophys. Res. 105, 2375- 2392 (2000). + +<|ref|>text<|/ref|><|det|>[[178, 653, 816, 708]]<|/det|> +[15] Yokoyama, T., Akita, K., Morimoto, T., Inoue, K. & Newmark, J. Clear Evidence of Reconnection Inflow of a Solar Flare. Astrophys. J. Lett. 546, L69- L72 (2001). + +<|ref|>text<|/ref|><|det|>[[178, 720, 816, 775]]<|/det|> +[16] Sui, L. & Holman, G. D. Evidence for the Formation of a Large- Scale Current Sheet in a Solar Flare. Astrophys. J. Lett. 596, L251- L254 (2003). + +<|ref|>text<|/ref|><|det|>[[178, 787, 816, 842]]<|/det|> +[17] Lin, J. et al. Direct Observations of the Magnetic Reconnection Site of an Eruption on 2003 November 18. Astrophys. J. 622, 1251- 1264 (2005). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[175, 157, 818, 213]]<|/det|> +[18] Savage, S. L. et al. Low- altitude Reconnection Inflow- Outflow Observations during a 2010 November 3 Solar Eruption. Astrophys. J. 754, 13 (2012). 1111.1945. + +<|ref|>text<|/ref|><|det|>[[178, 224, 816, 262]]<|/det|> +[19] Su, Y. et al. Imaging coronal magnetic- field reconnection in a solar flare. Nature Physics 9, 489- 493 (2013). 1307.4527. + +<|ref|>text<|/ref|><|det|>[[178, 273, 816, 329]]<|/det|> +[20] Yan, X. L. et al. Simultaneous Observation of a Flux Rope Eruption and Magnetic Reconnection during an X- class Solar Flare. Astrophys. J. Lett. 853, L18 (2018). 1801.02738. + +<|ref|>text<|/ref|><|det|>[[178, 340, 816, 396]]<|/det|> +[21] Takasao, S., Asai, A., Isobe, H. & Shibata, K. Simultaneous Observation of Reconnection Inflow and Outflow Associated with the 2010 August 18 Solar Flare. Astrophys. J. Lett. 745, L6 (2012). 1112.1398. + +<|ref|>text<|/ref|><|det|>[[178, 407, 816, 445]]<|/det|> +[22] Gou, T., Liu, R., Kliem, B., Wang, Y. & Veronig, A. M. The Birth of A Coronal Mass Ejection. Science Advances 5, 7004 (2019). 1811.04707. + +<|ref|>text<|/ref|><|det|>[[178, 456, 816, 512]]<|/det|> +[23] Cheng, X. et al. Observations of Turbulent Magnetic Reconnection within a Solar Current Sheet. Astrophys. J. 866, 64 (2018). 1808.06071. + +<|ref|>text<|/ref|><|det|>[[178, 523, 816, 561]]<|/det|> +[24] Xue, Z. et al. Observing the release of twist by magnetic reconnection in a solar filament eruption. Nature Communications 7, 11837 (2016). + +<|ref|>text<|/ref|><|det|>[[178, 572, 816, 628]]<|/det|> +[25] Sun, J. Q. et al. Extreme ultraviolet imaging of three- dimensional magnetic reconnection in a solar eruption. Nature Communications 6, 7598 (2015). 1506.08255. + +<|ref|>text<|/ref|><|det|>[[178, 640, 816, 696]]<|/det|> +[26] Masuda, S., Kosugi, T., Hara, H., Tsuneta, S. & Ogawara, Y. A loop- top hard X- ray source in a compact solar flare as evidence for magnetic reconnection. Nature 371, 495- 497 (1994). + +<|ref|>text<|/ref|><|det|>[[178, 707, 816, 745]]<|/det|> +[27] Arnold, H. et al. Electron Acceleration during Macroscale Magnetic Reconnection. arXiv e- prints arXiv:2011.01147 (2020). 2011.01147. + +<|ref|>text<|/ref|><|det|>[[178, 756, 816, 812]]<|/det|> +[28] Warmuth, A. & Mann, G. Thermal- nonthermal energy partition in solar flares derived from X- ray, EUV, and bolometric observations. Discussion of recent studies. Astron. Astrophys. 644, A172 (2020). 2011.04442. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[175, 157, 819, 231]]<|/det|> +[29] Innes, D. E., Guo, L.- J., Huang, Y.- M. & Bhattacharjee, A. IRIS Si IV Line Profiles: An Indication for the Plasmoid Instability during Small- scale Magnetic Reconnection on the Sun. Astrophys. J. 813, 86 (2015). 1509.08837. + +<|ref|>text<|/ref|><|det|>[[178, 242, 817, 299]]<|/det|> +[30] Guo, L.- J., De Pontieu, B., Huang, Y.- M., Peter, H. & Bhattacharjee, A. Observations and Modeling of the Onset of Fast Reconnection in the Solar Transition Region. Astrophys. J. 901, 148 (2020). 2009.11475. + +<|ref|>text<|/ref|><|det|>[[178, 310, 817, 366]]<|/det|> +[31] Kliem, B., Karlicky, M. & Benz, A. O. Solar flare radio pulsations as a signature of dynamic magnetic reconnection. Astron. Astrophys. 360, 715- 728 (2000). astro- ph/0006324. + +<|ref|>text<|/ref|><|det|>[[178, 377, 816, 433]]<|/det|> +[32] French, R. J., Judge, P. G., Matthews, S. A. & van Driel- Gesztelyi, L. Spectropolarimetric Insight into Plasma Sheet Dynamics of a Solar Flare. Astrophys. J. Lett. 887, L34 (2019). 1911.12666. + +<|ref|>text<|/ref|><|det|>[[178, 444, 816, 500]]<|/det|> +[33] Chitta, L. P. & Lazarian, A. Onset of Turbulent Fast Magnetic Reconnection Observed in the Solar Atmosphere. Astrophys. J. Lett. 890, L2 (2020). 2001.08595. + +<|ref|>text<|/ref|><|det|>[[178, 511, 816, 567]]<|/det|> +[34] Tian, H. et al. Imaging and Spectroscopic Observations of Magnetic Reconnection and Chromospheric Evaporation in a Solar Flare. Astrophys. J. Lett. 797, L14 (2014). 1411.2301. + +<|ref|>text<|/ref|><|det|>[[178, 578, 816, 635]]<|/det|> +[35] Yang, S., Zhang, J. & Xiang, Y. Magnetic Reconnection between Small- scale Loops Observed with the New Vacuum Solar Telescope. Astrophys. J. Lett. 798, L11 (2015). 1412.1314. + +<|ref|>text<|/ref|><|det|>[[178, 646, 816, 703]]<|/det|> +[36] Wang, T., Sui, L. & Qiu, J. Direct Observation of High- Speed Plasma Outflows Produced by Magnetic Reconnection in Solar Impulsive Events. Astrophys. J. Lett. 661, L207- L210 (2007). 0709.2329. + +<|ref|>text<|/ref|><|det|>[[178, 714, 816, 770]]<|/det|> +[37] Yu, S. et al. Magnetic Reconnection during the Post- impulsive Phase of a Long- duration Solar Flare: Bidirectional Outflows as a Cause of Microwave and X- Ray Bursts. Astrophys. J. 900, 17 (2020). 2007.10443. + +<|ref|>text<|/ref|><|det|>[[178, 782, 816, 819]]<|/det|> +[38] Burch, J. L. et al. Electron- scale measurements of magnetic reconnection in space. Science 352, aaf2939 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[175, 157, 819, 195]]<|/det|> +[39] Phan, T. D. et al. Electron magnetic reconnection without ion coupling in Earth's turbulent magnetosheath. Nature 557, 202- 206 (2018). + +<|ref|>text<|/ref|><|det|>[[178, 206, 817, 262]]<|/det|> +[40] Liu, R., Chen, J., Wang, Y. & Liu, K. Investigating Energetic X- Shaped Flares on the Outskirts of A Solar Active Region. Scientific Reports 6, 34021 (2016). 1609.02713. + +<|ref|>text<|/ref|><|det|>[[178, 273, 817, 329]]<|/det|> +[41] Kawabata, Y., Inoue, S. & Shimizu, T. Non- potential Field Formation in the X- shaped Quadrupole Magnetic Field Configuration. Astrophys. J. 842, 106 (2017). 1705.02560. + +<|ref|>text<|/ref|><|det|>[[178, 340, 816, 378]]<|/det|> +[42] Liu, Z. et al. New vacuum solar telescope and observations with high resolution. Res. Astron. Astrophys. 14, 705- 718 (2014). + +<|ref|>text<|/ref|><|det|>[[178, 389, 817, 446]]<|/det|> +[43] Schou, J. et al. Design and Ground Calibration of the Helioseismic and Magnetic Imager (HMI) Instrument on the Solar Dynamics Observatory (SDO). Solar Phys. 275, 229- 259 (2012). + +<|ref|>text<|/ref|><|det|>[[178, 456, 816, 513]]<|/det|> +[44] Xue, Z., Yan, X., Yang, L., Wang, J. & Zhao, L. Observing Formation of Flux Rope by Tether- cutting Reconnection in the Sun. Astrophys. J. Lett. 840, L23 (2017). + +<|ref|>text<|/ref|><|det|>[[178, 523, 816, 562]]<|/det|> +[45] Lin, R. P. et al. The Reuven Ramaty High- Energy Solar Spectroscopic Imager (RHESSI). Solar Phys. 210, 3- 32 (2002). + +<|ref|>text<|/ref|><|det|>[[178, 572, 817, 610]]<|/det|> +[46] Culhane, J. L. et al. The EUV Imaging Spectrometer for Hinode. Solar Phys. 243, 19- 61 (2007). + +<|ref|>text<|/ref|><|det|>[[178, 621, 816, 678]]<|/det|> +[47] Bhattacharjee, A., Huang, Y.- M., Yang, H. & Rogers, B. Fast reconnection in high- Lundquist- number plasmas due to the plasmoid Instability. Physics of Plasmas 16, 112102 (2009). 0906.5599. + +<|ref|>text<|/ref|><|det|>[[178, 688, 816, 744]]<|/det|> +[48] Ni, L., Kliem, B., Lin, J. & Wu, N. Fast Magnetic Reconnection in the Solar Chromosphere Mediated by the Plasmoid Instability. Astrophys. J. 799, 79 (2015). 1509.06895. + +<|ref|>text<|/ref|><|det|>[[178, 756, 816, 812]]<|/det|> +[49] Dere, K. P., Landi, E., Mason, H. E., Monsignori Fossi, B. C. & Young, P. R. CHIANTI - an atomic database for emission lines. Astron. Astrophys. Suppl. 125, 149- 173 (1997). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[177, 157, 816, 213]]<|/det|> +[50] Landi, E. et al. CHIANTI- An Atomic Database for Emission Lines. VII. New Data for X- Rays and Other Improvements. Astrophys. J. Suplement Series 162, 261- 280 (2006). + +<|ref|>text<|/ref|><|det|>[[178, 223, 816, 278]]<|/det|> +[51] Labrosse, N. et al. Physics of Solar Prominences: I—Spectral Diagnostics and Non- LTE Modelling. Space Sci. Rev. 151, 243- 332 (2010). 1001.1620. + +<|ref|>text<|/ref|><|det|>[[178, 289, 817, 344]]<|/det|> +[52] Daughton, W. et al. Role of electron physics in the development of turbulent magnetic reconnection in collisionless plasmas. Nature Physics 7, 539- 542 (2011). + +<|ref|>text<|/ref|><|det|>[[178, 354, 816, 427]]<|/det|> +[53] Longcope, D., Unverferth, J., Klein, C., McCarthy, M. & Priest, E. Evidence for Downflows in the Narrow Plasma Sheet of 2017 September 10 and Their Significance for Flare Reconnection. Astrophys. J. 868, 148 (2018). + +<|ref|>text<|/ref|><|det|>[[178, 437, 816, 475]]<|/det|> +[54] Lemen, J. R. et al. The Atmospheric Imaging Assembly (AIA) on the Solar Dynamics Observatory (SDO). Solar Phys. 275, 17- 40 (2012). + +<|ref|>text<|/ref|><|det|>[[178, 485, 816, 522]]<|/det|> +[55] Krucker, S. & Battaglia, M. Particle Densities within the Acceleration Region of a Solar Flare. Astrophys. J. 780, 107 (2014). + +<|ref|>text<|/ref|><|det|>[[178, 532, 816, 587]]<|/det|> +[56] Kliem, B. Particle orbits, trapping, and acceleration in a filamentary current sheet model. Astrophys. J. Suplement Series 90, 719- 728 (1994). + +<|ref|>text<|/ref|><|det|>[[178, 598, 816, 653]]<|/det|> +[57] Drake, J. F., Swisdak, M., Che, H. & Shay, M. A. Electron acceleration from contracting magnetic islands during reconnection. Nature 443, 553- 556 (2006). + +<|ref|>text<|/ref|><|det|>[[178, 664, 816, 719]]<|/det|> +[58] Titov, V. S., Hornig, G. & Demoulin, P. Theory of magnetic connectivity in the solar corona. Journal of Geophysical Research (Space Physics) 107, 1164 (2002). + +<|ref|>text<|/ref|><|det|>[[178, 730, 816, 784]]<|/det|> +[59] Priest, E. R. & Demoulin, P. Three- dimensional magnetic reconnection without null points. 1. Basic theory of magnetic flipping. J. Geophys. Res. 100, 23443- 23464 (1995). + +<|ref|>text<|/ref|><|det|>[[178, 795, 816, 850]]<|/det|> +[60] Ji, H. & Daughton, W. Phase diagram for magnetic reconnection in heliophysical, astrophysical, and laboratory plasmas. Physics of Plasmas 18, 111207- 111207 (2011). 1109.0756. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[178, 157, 816, 195]]<|/det|> +[61] Schuck, P. W. Tracking Magnetic Footpoints with the Magnetic Induction Equation. Astrophys. J. 646, 1358–1391 (2006). + +<|ref|>text<|/ref|><|det|>[[179, 205, 816, 259]]<|/det|> +[62] Li, T. & Zhang, J. Slipping Magnetic Reconnection Triggering a Solar Eruption of a Triangle- shaped Flag Flux Rope. Astrophys. J. Lett. 791, L13 (2014). 1407.4180. + +<|ref|>text<|/ref|><|det|>[[179, 270, 816, 325]]<|/det|> +[63] Xiang, Y.- y., Liu, Z. & Jin, Z.- y. High resolution reconstruction of solar prominence images observed by the New Vacuum Solar Telescope. New Astron 49, 8–12 (2016). + +<|ref|>text<|/ref|><|det|>[[179, 336, 816, 391]]<|/det|> +[64] Bobra, M. G. et al. The Helioseismic and Magnetic Imager (HMI) Vector Magnetic Field Pipeline: SHARPs - Space- Weather HMI Active Region Patches. Solar Phys. 289, 3549–3578 (2014). 1404.1879. + +<|ref|>text<|/ref|><|det|>[[179, 402, 816, 456]]<|/det|> +[65] Centeno, R. et al. The Helioseismic and Magnetic Imager (HMI) Vector Magnetic Field Pipeline: Optimization of the Spectral Line Inversion Code. Solar Phys. 289, 3531–3547 (2014). 1403.3677. + +<|ref|>text<|/ref|><|det|>[[179, 467, 816, 522]]<|/det|> +[66] Borrero, J. M. et al. VFISV: Very Fast Inversion of the Stokes Vector for the Helioseismic and Magnetic Imager. Solar Phys. 273, 267–293 (2011). 0901.2702. + +<|ref|>text<|/ref|><|det|>[[179, 533, 816, 587]]<|/det|> +[67] Metcalf, T. R. Resolving the 180- degree ambiguity in vector magnetic field measurements: The 'minimum' energy solution. Solar Phys. 155, 235–242 (1994). + +<|ref|>text<|/ref|><|det|>[[179, 598, 816, 653]]<|/det|> +[68] Leka, K. D. et al. Resolving the \(180^{\mathrm{deg}}\) Ambiguity in Solar Vector Magnetic Field Data: Evaluating the Effects of Noise, Spatial Resolution, and Method Assumptions. Solar Phys. 260, 83–108 (2009). + +<|ref|>text<|/ref|><|det|>[[179, 664, 816, 718]]<|/det|> +[69] Sun, X. On the Coordinate System of Space- Weather HMI Active Region Patches (SHARPs): A Technical Note. ArXiv e- prints (2013). 1309.2392. + +<|ref|>text<|/ref|><|det|>[[179, 730, 816, 784]]<|/det|> +[70] Jiang, C., Wu, S. T., Feng, X. & Hu, Q. Data- driven magnetohydrodynamic modelling of a flux- emerging active region leading to solar eruption. Nature Communications 7, 11522 (2016). + +<|ref|>text<|/ref|><|det|>[[179, 796, 816, 850]]<|/det|> +[71] Jiang, C. et al. Reconstruction of a Large- scale Pre- flare Coronal Current Sheet Associated with a Homologous X- shaped Flare. Astrophys. J. 850, 8 (2017). 1710.02775. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[179, 158, 816, 212]]<|/det|> +[72] Wiegelmann, T. Optimization code with weighting function for the reconstruction of coronal magnetic fields. Solar Phys. 219, 87–108 (2004). 0802.0124. + +<|ref|>text<|/ref|><|det|>[[179, 223, 816, 278]]<|/det|> +[73] Wiegelmann, T., Inhester, B. & Sakurai, T. Preprocessing of Vector Magnetograph Data for a Nonlinear Force- Free Magnetic Field Reconstruction. Solar Phys. 233, 215–232 (2006). astro- ph/0612641. + +<|ref|>text<|/ref|><|det|>[[179, 288, 816, 325]]<|/det|> +[74] Schwartz, R. A. et al. RHESSI Data Analysis Software: Rationale and Methods. Solar Phys. 210, 165–191 (2002). + +<|ref|>text<|/ref|><|det|>[[179, 335, 816, 389]]<|/det|> +[75] Liu, Y., Zhao, J. & Schuck, P. W. Horizontal Flows in the Photosphere and Subphotosphere of Two Active Regions. Solar Phys. 287, 279–291 (2013). + +<|ref|>sub_title<|/ref|><|det|>[[179, 401, 352, 418]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[179, 418, 816, 600]]<|/det|> +We would like to thank the NVST, SDO/AIA, SDO/HMI, GOES, RHESSI, and Hinode teams for high- cadence data support. This work is sponsored by the National Science Foundation of China (NSFC) under the grant numbers (11873087, 11973084, 11803085, 2019FD085, 12003064, 11633008, 11763004, U1831210, 11803002), by the Yunnan Key Science Foundation of China under numbers (2018FA001, 2018FB007) and Yunnan Science Foundation for Distinguished Young Scholars, by Project Supported by the Specialized Research Fund for State Key Laboratories, by Key Research and Development Project of Yunnan Province202003AD150019, and by the joint NSFC/DFG Grant 41761134088/KL817.8- 1. + +<|ref|>sub_title<|/ref|><|det|>[[179, 611, 365, 627]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[179, 629, 816, 738]]<|/det|> +X.L.Y. developed the ideas, performed analysis of the data, and lead the discussion of the manuscript. Z.K.X., L.H.Y., J.C.W., D.F.K., and Y.L.S. performed analysis of the data. C.W.J. carried out the MHD simulation. E.R.P. evaluated the energy of magnetic reconnection. B.K. contributed to the interpretation of the data and to the discussion. X.S.F. and Z.L. contributed to the discussion. All authors reviewed the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[179, 750, 380, 766]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[179, 768, 814, 804]]<|/det|> +Correspondence and requests for materials should be addressed to X.L.Y. and C.W.J (email: yanxl@ynao.ac.cn; chaowei@hit.edu.cn). + +<|ref|>text<|/ref|><|det|>[[179, 815, 816, 850]]<|/det|> +Competing financial interests: The authors declare no competing financial interests. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[241, 155, 747, 606]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[178, 621, 816, 860]]<|/det|> +
Figure 1: Magnetic reconnection during the confined flare, observed in the center of the \(\mathrm{H}\alpha\) line. (a) \(H\alpha\) image observed at 08:04:54 UT by NVST superimposed on the line-of-sight magnetogram observed by SDO/HMI. Red (blue) contours indicate positive (negative) polarity of \(\pm 200\) G and \(\pm 800\) G. (b) Line-of-sight magnetogram observed by SDO/HMI at 08:04:55 UT. The two curved red lines indicate the chromospheric fibrils and the active-region filament participating in the reconnection. (c–h) The process of reconnection as seen in the \(H\alpha\) observations. White arrows in c indicate the two reconnecting structures. Black arrows in d and e indicate the new chromospheric fibrils formed by reconnection. Blue arrows in d–f indicate the current sheet. The dotted lines in f and g indicate the cusp-shaped structures (magnetic separatrices) formed between the reconnection in- and outflows.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[177, 210, 820, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 664, 817, 792]]<|/det|> +
Figure 2: Profiles of GOES X-ray flux, inflows, and outflows during reconnection. (a) The profile of GOES X-rays in 1–8 Å (red line) and 0.5–4 Å (blue line). (b) The profile of the RHESSI X-ray counts in the 3–6, 6–12, 12–25, and 25–50 keV bands. (c) Inflow (dotted curve) along the A-B slit inferred from time-slices of the intensity. (d) Outflow along the C-D slit. (e) Outflow along the curved line E-F. (f) Outflow along the curved line G-H. All these lines are marked in Fig. 1h.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[196, 240, 803, 664]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 690, 817, 765]]<|/det|> +
Figure 3: Hinode/EIS observation at Fe XVII 255 Å during magnetic reconnection. (a–c) Evolution of the cusp-shaped structure in the intensity images. (d–f) Doppler velocity and (g–i) non-thermal velocity obtained from single Gaussian fitting.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[196, 171, 783, 707]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 720, 817, 830]]<|/det|> +
Figure 4: Fine structures in the SDO/AIA 211 Å images revealed by the unsharp masking technique. Red arrows mark the different bright blobs moving from the reconnection region (Fig. 4a–f). The green-dotted line in Fig. 4a indicates the trajectory I-J used to make the time-distance diagram of Fig. 4g. The projected velocities of the blobs lie in the range from \(79 \mathrm{km s^{-1}}\) to \(208 \mathrm{km s^{-1}}\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[191, 223, 800, 641]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 655, 818, 784]]<|/det|> +
Figure 5: Emission measure and temperature maps and \(\mathrm{H}\alpha\) images with RHESSI X-ray sources superimposed. (a–d): DEM maps in four temperature ranges. (e) Total emission measure. (f) Temperature of the reconnection region derived from six AIA wavelengths. (g–i): \(\mathrm{H}\alpha\) images with RHESSI hard X-ray sources superimposed. The levels of the contours are at 50% and 80% of the peak values in the 3–6, 6–12, 12–25, and 25–50 keV bands.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[200, 268, 790, 592]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 608, 818, 737]]<|/det|> +
Figure 6: RHESSI photon spectra accumulated around 08:12 UT and 08:19 UT during the quoted 1-minute intervals on 2014 February 2. The spectra are fitted by combining the \(\mathrm{vth}\) and thick2 functions (see Methods). The black line shows the photon spectrum with the background subtracted. The pink line shows the background. The green and the yellow lines display the fitting with the single-temperature thermal model and the thick-target nonthermal model, respectively.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[196, 250, 789, 589]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 604, 818, 753]]<|/det|> +
Figure 7: Magnetic structure of the source region. (a) Selected field lines showing the twisted magnetic structure of the filament (green) and the loop-like structure of the chromospheric fibrils (red) before the reconnection event, obtained by using NLFFF extrapolation from the vector magnetogram. (b)–(c) Map of the magnetic squashing degree \(\log Q\) at heights \(Z = 0\) and \(Z = 2.9\) Mm above the photosphere. (d) Iso-surface \(\log Q = 4\) in the reconnection region. (e) \(\log Q\) in the vertical cross-section marked by a yellow dashed line in (a).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[270, 156, 723, 690]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[177, 710, 816, 858]]<|/det|> +
Figure 8: Evolution of magnetic field, current sheet, and plasmoids (blobs) during magnetic reconnection in the data-driven MHD simulation. (a) and (b): Evolution of the magnetic structure involved in the modeled reconnection event. (c) and (d): Zoom showing the magnetic structure of the current sheet at two selected times. The red boxes in (a) and (b) show the field of view of (c)–(f). (e) and (f): Distribution of electric current density ( \(Jc\) ) in the current sheet (also overlaid in (c)–(d)) at the height \(z = 1\) (11.6 Mm).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[194, 279, 795, 585]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[179, 599, 817, 728]]<|/det|> +
Figure 9: Evolution of the photospheric vector magnetic field observed by SDO/HMI and flow field derived using the DAVE algorithm. (a–c): Vector magnetograms. The greyscale images show the vertical field component, saturated at 1000 G, and the arrows show the horizontal component. (d–f): Evolution of the flow field. Blue and red arrows indicate the horizontal photospheric velocity in the positive and negative polarities, respectively.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 143, 69]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[58, 102, 860, 833]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 850, 115, 869]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 891, 933, 935]]<|/det|> +Magnetic reconnection during the confined are, observed in the center of the Ha line. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[48, 50, 920, 644]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 666, 118, 686]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[45, 707, 950, 729]]<|/det|> +Prols of GOES X- ray flux, in flows, and out flows during reconnection. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 58, 904, 641]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 669, 116, 688]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 711, 925, 754]]<|/det|> +Hinode/EIS observation at Fe XVII 255A during magnetic reconnection. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[65, 45, 861, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[43, 802, 116, 821]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 843, 890, 886]]<|/det|> +Fine structures in the SDO/AIA 211 A images revealed by the unsharp masking technique. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[68, 50, 920, 650]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 672, 117, 692]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 714, 950, 757]]<|/det|> +Emission measure and temperature maps and Ha images with RHESSI X- ray sources superimposed. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 55, 910, 512]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 540, 116, 560]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[44, 583, 951, 627]]<|/det|> +RHESSI photon spectra accumulated around 08:12 UT and 08:19 UT during the quoted 1- minute intervals on 2014 February 2. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 55, 928, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 576, 116, 595]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[42, 618, 693, 639]]<|/det|> +Magnetic structure of the source region. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 52, 675, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 800, 116, 820]]<|/det|> +
Figure 8
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 919, 886]]<|/det|> +Evolution of magnetic field, current sheet, and plasmoids (blobs) during magnetic reconnection in the data- driven MHD simulation. (see full caption in Manuscript file) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[65, 66, 925, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 531, 117, 550]]<|/det|> +
Figure 9
+ +<|ref|>text<|/ref|><|det|>[[42, 572, 945, 615]]<|/det|> +Evolution of the photospheric vector magnetic field observed by SDO/HMI and ow field derived using the DAVE algorithm. (see full caption in Manuscript file) + +<|ref|>sub_title<|/ref|><|det|>[[44, 639, 311, 666]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 689, 765, 709]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 727, 327, 853]]<|/det|> +SupplementaryMovie1. mp4 SupplementaryMovie2. mp4 SupplementaryMovie3. mp4 SupplementaryMovie4. mp4 SupplementaryMovie5. mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/images_list.json b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..8370fd0d16ef51f48c0fc27271d9fb7898117401 --- /dev/null +++ b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/images_list.json @@ -0,0 +1,25 @@ +[ + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "A) Naive (304 cells)", + "footnote": [], + "bbox": [], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "C) Pregated on γδTCR+ T cells:", + "footnote": [], + "bbox": [ + [ + 17, + 28, + 220, + 170 + ] + ], + "page_idx": 39 + } +] \ No newline at end of file diff --git a/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3.mmd b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7266f1bc312d84388a4f04318111ac8c061fdc7b --- /dev/null +++ b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3.mmd @@ -0,0 +1,580 @@ + +# Spatially-resolved single cell transcriptomics reveal a critical role for \(\gamma \delta\) T cells in the control of skin inflammation and subcutaneous adipose wasting during chronic Trypanosoma brucei infection + +Juan Quintana ( \(\square\) juan.quintana@glasgow.ac.uk) University of Glasgow https://orcid.org/0000- 0002- 5092- 5576 + +Matthew Sinton University of Glasgow https://orcid.org/0000- 0003- 1292- 799X + +Praveena Chandrasegaran University of Glasgow + +Agatha Lestari University of Glasgow + +Rhiannon Heslop University of Glasgow https://orcid.org/0000- 0002- 4332- 4393 + +Bachar Cheaib University of Glasgow + +John Ogunsola University of Glasgow + +Dieudonne Mumba Ngoyi National Institute of Biomedical Research & Department of Tropical Medicine, University of Kinshasa https://orcid.org/0000- 0002- 5886- 2004 + +Nono Kuispond Swar National Institute of Biomedical Research + +Anneli Cooper University of Glasgow + +Seth Coffelt Beatson Institute https://orcid.org/0000- 0003- 2257- 2862 + +Annette MacLeod University of Glasgow + +## Article + +Keywords: African trypanosomes, \(\gamma \delta\) T cells, \(\mathrm{V}\gamma 6+\) cells, adipocytes, skin inflammation + +<--- Page Split ---> + +Posted Date: March 14th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2647707/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on August 29th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 40962- y. + +<--- Page Split ---> + +1 Spatially-resolved single cell transcriptomics reveal a critical role for \(\gamma \delta\) T cells in the control of skin inflammation and subcutaneous adipose wasting during chronic Trypanosoma brucei infection 4 Juan F. Quintana \(^{1,2,3\dagger *}\) , Matthew C. Sinton \(^{1*}\) , Praveena Chandrasegaran \(^{1}\) , Agatha Nabila Lestari \(^{1}\) , Rhiannon Heslop \(^{1}\) , Bachar Cheaib \(^{1}\) , John Ogunsola \(^{1}\) , Dieudonne Mumba Ngoyi \(^{4}\) , Nono-Raymond Kuispond Swar \(^{1,4}\) , Anneli Cooper \(^{1}\) , Seth B. Coffelt \(^{5,6}\) , Annette MacLeod \(^{1\dagger}\) + +9 1Wellcome Centre for Integrative Parasitology (WCIP). University of Glasgow, Glasgow, UK. School of Biodiversity, One Health, Veterinary Medicine (SBOHVM), College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow UK. 2Lydia Becker Institute of Immunology and Inflammation, University of Manchester, UK. 3Division of Immunology, Immunity to Infection and Health, Manchester Academic Health Science Centre, University of Manchester, UK. 4Department of Parasitology, National Institute of Biomedical Research, Kinshasa, Democratic Republic of the Congo. 5School of Cancer Sciences, University of Glasgow Glasgow, UK. 6Cancer Research UK Beatson Institute, Glasgow, UK. + +\*These authors contributed equally to this work. †Corresponding authors: juan.quintana@glasgow.ac.uk, annette.macleod@glasgow.ac.uk + +Keywords: African trypanosomes, \(\gamma \delta\) T cells, \(\mathrm{V}\gamma \delta^{+}\) cells, adipocytes, skin inflammation. + +<--- Page Split ---> + +## Abstract + +African trypanosome parasites colonise the skin in a process important for parasite transmission. However, how the skin responses to trypanosome infection remain unresolved. Here, using a combination of spatial and single cell transcriptomics, coupled with in vivo genetic models, we investigated the local immune response of the skin in a murine model of infection. First, we detected a significant expansion of IL- 17A- producing \(\gamma \delta \mathrm{T}\) cells (primarily \(\mathrm{V}\gamma \delta^{+}\) ) in the infected murine skin compared to naive controls that occur mainly in the subcutaneous adipose tissue. Second, interstitial preadipocytes located in the subcutaneous adipose tissue upregulate several genes involved in inflammation and antigen presentation, including T cell activation and survival. In silico cell- cell communication suggests that adipocytes trigger \(\gamma \delta \mathrm{T}\) cell activation locally via Cd40, Il6, Il10, and Tnfsf18 signalling, amongst others. Third, mice deficient in IL- 17A- producing \(\gamma \delta \mathrm{T}\) cells show extensive inflammation, increased frequency of skin- resident IFN \(\gamma\) - producing \(\mathrm{CD8^{+}}\) T cells and limited subcutaneous adipose tissue wasting compared to wild- type infected controls, independent of \(\mathrm{T}_{\mathrm{H}}1\) \(\mathrm{CD4^{+}}\) T cells and parasite burden. Based on these observations, we proposed a model whereby adipocytes as well as \(\mathrm{V}\gamma \delta^{+}\) cells act concertedly in the subcutaneous adipose tissue to limit skin inflammation and tissue wasting. These studies shed light onto the mechanisms of \(\gamma \delta \mathrm{T}\) cell- mediated immunity in the skin in the context of African trypanosome infection, as well as a potential role of immature and mature adipocytes as homeostatic regulators in the skin during chronic infection. + +<--- Page Split ---> + +## Introduction + +The skin represents the ultimate barrier, offering physical and mechanical protection against insults including infections. The skin is a complex organ encompassing several tissue layers, including the dermis and epidermis, which are capable of extensive remodelling and resistance to mechanical and chemical stimuli. Moreover, the skin hosts a complex stromal microenvironment containing a myriad of resident and recruited immune cells that actively survey the tissue. Amongst the dermal immune compartment, \(\gamma \delta\) T cells have emerged as critical regulators of tissue homeostasis \(^{1 - 4}\) . For instance, \(\mathrm{V}\gamma 6^{+}\) cells are enriched in mucosal tissues \(^{3}\) and the dermis \(^{2,4}\) where they are considered resident cells. These skin resident \(\gamma \delta\) T cells express a wide range of effector molecules including interleukin 17A (IL- 17A) and IL- 17F \(^{5}\) , IL- 22 \(^{6}\) , and the epidermal growth factor (EGF) receptor ligand \(^{7}\) , and amphiregulin (AREG) \(^{8}\) , which are critical for promoting tissue repair and limiting inflammation following an insult \(^{9 - 12}\) . Although \(\gamma \delta\) T cells have gathered a lot of attention for their role in cancer \(^{13 - 16}\) , and autoimmune disorders affecting the skin such as psoriasis \(^{17}\) and lupus \(^{18}\) , their function during parasitic colonisation of the skin is still not well understood. + +Despite the robust nature of the skin as a barrier, some pathogens manage to circumvent its defences. Of these Trypanosomatids, including the causative agents of leishmaniasis, chagas disease, and sleeping sickness, are known to colonise the skin in a process that has been postulated as being critical for infection and parasite transmission \(^{19 - 21}\) . Indeed, we and others have recently reported the presence of Trypanosoma brucei in the skin of mice and humans \(^{21 - 23}\) . Such skin infections are often associated with dermatitis and pruritus \(^{22}\) or can occur in the absence of clinical symptoms or a positive result from gold- standard diagnostic methods \(^{22}\) . This strongly indicates that mild or asymptomatic skin infections act as important but overlooked parasite reservoirs hampering ongoing eradication efforts. However, to date, the skin response to \(T\) . brucei infection remains unknown. + +Here, we used a combined single cell and spatial transcriptomics approach to generate a spatially resolved single cell atlas of the murine skin during \(T\) . brucei infection. This combined approach led us to identify that interstitial preadipocytes and Langerhans cells both have central roles in the production of local cytokines and antigenic presentation, taking place largely in the subcutaneous adipose tissue and in proximity to the subcutaneous adipose tissue, respectively. Furthermore, we identified a population of + +<--- Page Split ---> + +skin IL- 17A- producing \(\mathrm{V} / \mathrm{6}^{+}\) cells, located in proximity to adipocytes in the subcutaneous adipose tissue and subcutaneous adipose tissue, that display features of TCR engagement and T cell activation upon infection. Cell- cell communication analyses between subcutaneous adipocytes and \(\mathrm{V} / \mathrm{6}^{+}\) cells predicted several potential interactions mediating not only T cell activation via Tnfsf18, but also promoting adipocyte lipolysis via \(\mathrm{V} / \mathrm{6}^{+}\) cell- derived Clcf1 and Areg. In vivo analyses revealed that \(\mathrm{V} / \mathrm{4} / \mathrm{6} / \mathrm{8} \mathrm{T}\) cell- deficient mice experience severe skin inflammation, likely mediated by an increased frequency of \(\mathrm{IFN} \gamma\) - producing \(\mathrm{CD8}^{+}\) T cells in the skin, but with limited subcutaneous adipose tissue wasting compared to infected wild type controls. We conclude that IL- 17A- producing \(\mathrm{V} / \mathrm{6}^{+}\) cells are critical mediators of skin immunity against African trypanosomes, dampening \(\mathrm{IFN} \gamma\) - mediated \(\mathrm{CD8}^{+}\) T cell responses in the skin and promoting subcutaneous adipose tissue wasting, supporting our recently identified role of IL- 17 signalling as a mediator of adipose tissue wasting during T. brucei infection24. More broadly, our results provide a spatially resolved atlas that can help to dissect further immunological pathways triggered in the skin in response to infection. + +<--- Page Split ---> + +## Materials and methods + +Ethical statement. All animal experiments were approved by the University of Glasgow Ethical Review Committee and performed in accordance with the home office guidelines, UK Animals (Scientific Procedures) Act, 1986 and EU directive 2010/63/EU. All experiments were conducted under SAPO regulations and UK Home Office project licence number PC8C3B25C to Dr. Jean Rodgers. + +Murine infections with Trypanosoma brucei. 6- 8- week- old wild- type female BALB/c (stock 000651) and FVB/NJ mice (stock 001800) were purchased from Jackson Laboratories. \(\mathrm{V}_7 / 4 / 6^{- / - }\) mice (a gift from Rebecca O'Brien, National Jewish Health) were backcrossed to FVB/N. Female mice aged (6- 8 weeks old) were used for infection. Six- to- eight- week- old wild- type female BALB/c (stock 000651) and FVB/NJ mice (stock 001800) were purchased from Jackson Laboratories. For infections, mice were inoculated by intra- peritoneal injection with \(\sim 2 \times 10^{3}\) parasites of strain T. brucei brucei Antat 1.1E25. Click or tap here to enter text. Parasitaemia was monitored by regular sampling from tail venepuncture and blood was examined using phase microscopy and the rapid "matching" method16. Uninfected mice of the same strain, sex and age served as uninfected controls. Mice were fed ad libitum and kept on a 12 h light- dark cycle. All the experiments were conducted between 8h and 12h. + +Skin tissue processing and preparation of single cell suspension for single- cell RNA sequencing. Infected animals and naive controls were anesthetized with isoflurane at 21 days post- infection and perfused transcardially with 25- 30 ml of ice- cold 1X PBS containing \(0.025\%\) (wt/vol) EDTA. Skin biopsies from the abdominal flank area were shaved off completely. Some of these sections were placed in \(4\%\) PFA for 16 hours prior to embedding in paraffin and histological analysis. Single- cell dissociations for scRNAseq experiments were performed using the 2- step protocol published by (Joost et al., 2020). Briefly, excised \(4\mathrm{cm}^2\) skin sections were diced into small pieces with a scalpel blade and enzymatically digested with Collagenase type I (500 U/ml; Gibco) and DNAse I (1 mg/ml; Sigma) in HBSS containing \(0.04\%\) BSA (Invitrogen) for \(\sim 1\) hr at \(37^{\circ}\mathrm{C}\) with shaking at 300rpm. Liberated cells from the partially digested tissue were pushed through a \(70\mu \mathrm{m}\) nylon mesh filter with an equal volume of HBSS \(0.04\%\) BSA, and then kept on ice. The tissue remaining in the \(70\mu \mathrm{m}\) filter was incubated in \(0.05\%\) trypsin EDTA for 15 minutes at \(37^{\circ}\mathrm{C}\) , and then liberated cells pushed through the filter with an equal of HBSS \(0.04\%\) BSA. Flowthrough from the two digestion steps were combined, passed through a \(40\mu \mathrm{M}\) + +<--- Page Split ---> + +filter to remove any cell aggregate and spun at 350g for 10mins at \(4^{\circ}C\) . Finally, cells were passed through a MACS dead cell removal kit (Miltenyi Biotec) and diluted to \(\sim 1,000\) cells/μl in \(200\mu l\) HBSS \(0.04\%\) BSA and kept on ice until single- cell capture using the 10X Chromium platform. The single cell suspensions were loaded onto independent single channels of a Chromium Controller (10X Genomics) single- cell platform. Briefly, \(\sim 25,000\) single cells were loaded for capture using 10X Chromium NextGEM Single cell 3 Reagent kit v3.1 (10X Genomics). Following capture and lysis, complementary DNA was synthesized and amplified (12 cycles) as per the manufacturer's protocol (10X Genomics). The final library preparation was carried out as recommended by the manufacturer with a total of 14 cycles of amplification. The amplified cDNA was used as input to construct an Illumina sequencing library and sequenced on a Novaseq 6000 sequencers by Glasgow Polyomics. + +Read mapping, data processing, and integration. For FASTQ generation and alignments, Illumina basecall files (\\*.bcl) were converted to FASTQs using bcl2fastq. Gene counts were generated using Cellranger v.6.0.0 pipeline against a combined Mus musculus (mm10) and Trypanosoma brucei (TREU927) transcriptome reference. After alignment, reads were grouped based on barcode sequences and demultiplexed using the Unique Molecular Identifiers (UMIs). The mouse- specific digital expression matrices (DEMs) from all six samples were processed using the R (v4.2.1) package Seurat v4.1.0 17. Additional packages used for scRNAseq analysis included dplyr v1.0.7 18, RColorBrewer v1.1.2 (http://colorbrewer.org), ggplot v3.3.5 19, and sctransform v0.3.3 20. We initially captured 65,734 cells mapping specifically against the M. musculus genome across all conditions and biological replicates, with an average of 29,651 reads/cell and a median of 1,762 genes/cell (S1A Table and Supplementary Figure 1). The number of UMIs was then counted for each gene in each cell to generate the digital expression matrix (DEM). Low quality cells were identified according to the following criteria and filtered out: \(i\) ) nFeature \(>100\) or \(< 5,000\) , \(ii\) ) nCounts \(>100\) or \(< 20,000\) , \(iii\) ) \(>30\%\) reads mapping to mitochondrial genes, and \(iv\) ) \(>40\%\) reads mapping to ribosomal genes, v) genes detected \(< 3\) cells. After applying this cut- off, we obtained a total of 56,876 high quality mouse- specific cells with an average of 29,651 reads/cells and a median of 1,683 genes/cell (S1A Table). High- quality cells were then normalised using the SCTransform function, regressing out for total UMI and genes counts, cell cycle genes, and highly variable genes identified by both Seurat and Scater packages, followed by data + +<--- Page Split ---> + +integration using IntegrateData and FindIntegrationAnchors. For this, the number of principal components were chosen using the elbow point in a plot ranking principal components and the percentage of variance explained (15 dimensions) using a total of 5,000 genes, and SCT as normalisation method. + +Cluster analysis, marker gene identification, and subclustering. The integrated dataset was then analysed using RunUMAP (10 dimensions), followed by FindNeighbors (10 dimensions, reduction = "pca") and FindClusters (resolution = 0.4). With this approach, we identified a total of 16 cell clusters The cluster markers were then found using the FindAllMarkers function (logfc.threshold = 0.25, assay = "RNA"). To identify cell identity confidently, we employed a supervised approach. This required the manual inspection of the marker gene list followed by and assignment of cell identity based on the expression of putative marker genes expressed in the unidentified clusters. The resolution used for these analyses was selected using the Clustree package26 (Supplementary Figure 1). A cluster name denoted by a single marker gene indicates that the chosen candidate gene is selectively and robustly expressed by a single cell cluster and is sufficient to define that cluster (e.g., Col1a1, Cd3g, Pparg, Krt14, among others). When manually inspecting the gene markers for the final cell types identified in our dataset, we noted the co-occurrence of genes that could discriminate two or more cell types (e.g., T cells, fibroblasts). To increase the resolution of our clusters to help resolve potential mixed cell populations embedded within a single cluster, we subset fibroblasts, myeloid cells, and T cells and analysed them separately using the same functions described above. In all cases, upon subsetting, the resulting objects were reprocessed using the functions FindVariableFeatures, ScaleData, RunUMAP, FindNeighbors, and FindClusters with default parameters. The number of dimensions used in each case varied depending on the cell type being analysed but ranged between 5 and 10 dimensions. Cell type-level differential expression analysis between experimental conditions was conducted using the FindMarkers function (min.pct = 0.25, test.use = Wilcox) and (DefaultAssay = "SCT"). Cell-cell interaction analysis mediated by ligand-receptor expression level was conducted using NicheNet25 with default parameters using "mouse" as a reference organism, comparing differentially expressed genes between experimental conditions (condition_oi = "Infected", condition_reference = "Uninfected"). + +10X Visium spatial sequencing library preparation and analysis + +<--- Page Split ---> + +Tissue processing and library preparation. Mice were shaved and dissected skin stored in \(4\%\) paraformaldehyde for \(24\mathrm{h}\) , before paraffin embedding. RNA was purified from FFPE sections to measure integrity, using the Absolutely Total RNA FFPE Purification Kit (Agilent Technologies) as per the manufacturer's instructions. RNA integrity was then measured by BioAnalyzer (Agilent Technologies) using the RNA 6000 Pico Kit (Agilent Technologies). Samples selected for sequencing all had a DV200 \(>50\%\) . We then placed \(10\mu \mathrm{m}\) sections within the capture areas of Visium Spatial Slides (10X Genomics) and proceeded to perform H&E staining and image capture, as per the manufacturer's instructions. To deparaffinise, slides were incubated at \(60^{\circ}\mathrm{C}\) for \(2\mathrm{h}\) before incubating twice in xylene at room temperature, for \(10\mathrm{min}\) each, three times in \(100\%\) ethanol for \(3\mathrm{min}\) each, twice in \(96\%\) ethanol for \(3\mathrm{min}\) each, \(85\%\) ethanol for \(3\mathrm{min}\) , and then submerged in Milli- Q water until ready to stain. For H&E staining, slides were incubated with Mayer's haematoxylin Solution (Sigma- Aldrich), Bluing Buffer (Dako) and Alcoholic Eosin solution (Sigma- Aldrich), with thorough washing in ultrapure water between each step. Stained slides were scanned under a microscope (EVOS M5000, Thermo). De- crosslinking was performed by incubating sections twice with \(0.1\mathrm{NHCl}\) for \(1\mathrm{min}\) at room temperature. Sections were then incubated twice with Tris- EDTA (TE) buffer pH 9.0 before incubating at \(70^{\circ}\mathrm{C}\) for \(1\mathrm{h}\) . The Visium Spatial Gene Expression Slide Kit (10X Genomics) was used for reverse transcription and second strand synthesis, followed by denaturation, to allow the transfer of the cDNA from the slide to a collection tube. These cDNA fragments were then used to construct spatially barcoded Illumina- compatible libraries using the Dual Index Kit TS Set A (10x Genomics) was used to add unique i7 and i5 sample indexes, enabling the spatial and UMI barcoding. The final Illumina- compatible sequencing library underwent paired- end sequencing (150 bp) on a NovaSeq 6000 (Illumina) at GenomeScan. After sequencing, the FASTQ files were aligned to a merged reference transcriptome combining the Mus musculus genome (mm10). After alignment, reads were grouped based on spatial barcode sequences and demultiplexed using the UMLs, using the SpaceRanger pipeline version 1.2.2 (10X Genomics). Downstream analyses of the expression matrices were conducted using the Seurat pipeline for spatial RNA integration (Hao et al., 2021b; Stuart et al., 2019) (Table S3A). Specifically, the data was scaled using the SCTransform function with default parameters. We then proceeded with dimensionality reduction and clustering analysis using RunPCA (assay = "SCT"), FindNeighbours and FindClusters functions with default settings and a total of 30 + +<--- Page Split ---> + +dimensions. We then applied the FindSpatiallyVariables function to identify spatially variable genes, using the top 1,000 most variable genes and "markvariogram" as selection method. To integrate our skin scRNAseq with the 10X Visium dataset, we used the FindTransferAnchors function with default parameters, using SCT as normalization method. Then, the TransferData function (weight.reduction = "pca", 30 dimensions) was used to annotate brain regions based on transferred anchors from the scRNAseq reference datasets. To predict the cell-cell communication mediated by ligand-receptor co-expression patterns in the spatial context, we employed NICHES v0.0.2 (Raredon et al., 2022). Upon dimensionality reduction and data normalisation, NICHES was run using fanton5 as ligand-receptor database with default parameters. The resulting object was then scaled using the functions ScaleData, FindVariableFeatures (selection.method = "disp"), RunUMAP with default settings and a total of 15 dimensions. Spatially resolved expression of ligand-receptor pairs was then identified using the FindAllMarkers function (min.pct = 0.25, test.use = "roc"). For visualisation, we used the SpatialFeaturePlot function with default parameters and min.cutoff = "q1". + +Single molecule fluorescence in situ hybridisation (smFISH) using RNAscope. smFISH experiments were conducted as follows. Briefly, to prepare tissue sections for smFISH, infected animals and naive controls were anesthetized with isoflurane, and skin sections from the flank were shaved off completely before placing in 4% PFA overnight prior to embedding in paraffin. Paraffin- embedded skin sections sections (3 μm) were prepared placed on a SuperFrost Plus microscope slides. Sections were then then dehydrated in 50, 70 and 100% ethanol. RNAscope 2.5 Assay (Advanced Cell Diagnostics) was used for all smFISH experiments according to the manufacturer's protocols. All RNAscope smFISH probes were designed and validated by Advanced Cell Diagnostics. For image acquisition, 16- bit laser scanning confocal images were acquired with a 63x/1.4 plan- apochromat objective using an LSM 710 confocal microscope fitted with a 32- channel spectral detector (Carl Zeiss). Lasers of 405nm, 488nm and 633 nm excited all fluorophores simultaneously with corresponding beam splitters of 405nm and 488/561/633nm in the light path. 9.7nm binned images with a pixel size of 0.07um x 0.07um were captured using the 32- channel spectral array in Lambda mode. Single fluorophore reference images were acquired for each fluorophore and the reference spectra were employed to unmix the multiplex images using the Zeiss online + +<--- Page Split ---> + +fingerprinting mode. smFISH images were acquired with minor contrast adjustments as needed, and converted to grayscale, to maintain image consistency. + +
List of RNAscope probes used for smFISH
SupplierCat. NumberSequenceChannelDye used
Trypanosoma brucei probes
Biotechnic1103198-C1Tbr-GapdhChannel 1Opal 520
Biotechnic1103208-C2Tbr-Pyk1Channel 2Opal 650
Biotechnic1103218-C3Tbr-Pad2Channel 3Opal 570
Biotechnic1103221-C4Tbr-Ep1Channel 4Opal 540
+ +## Histological analysis of adipocyte size + +Skin placed into \(4\%\) paraformaldehyde (PFA) and fixed overnight at room temperature. PFA- fixed skin biopsies were then embedded in paraffin, sectioned, and stained by the Veterinary Diagnostic Services facility (University of Glasgow, UK). Sections were Haematoxylin and Eosin (H&E) stained for adipocyte size analysis. Slide imaging was performed by the Veterinary Diagnostic Services facility (University of Glasgow, UK) using an EasyScan Infinity slide scanner (Motic, Hong Kong) at 20X magnification. To determine subcutaneous adipocyte size in skin sections, images were first opened in QuPath (v. 0.3.2)27, before selecting regions and exporting to Fiji28. In Fiji, images were converted to 16- bit format, and we used the Adiposoft plugin29 to quantify adipocyte size within different sections. + +Semi- quantitative evaluation of the parasite burden and inflammation in skin sections. Paraffin- embedded skin samples were cut into \(2.5 \mu \mathrm{m}\) sections and stained for T. brucei parasites using a polyclonal rabbit antibody raised against T. brucei luminal binding protein 1 (BiP) (J. Bangs, SUNY, USA) using a Dako Autostainer Link 48 (Dako, Denmark) and were subsequently counterstained with Gill's Haematoxylin. The extent of inflammatory cell infiltration in skin sections was assessed in haematoxylin and eosin- stained sections. Stained slides were assessed by two independent pathologists blinded to infection status and experimental procedures. Skin parasite burden was assessed for both intravascular (parasites within the lumen of dermal or subcutaneous small to medium- sized vessels) and extravascular locations (parasites outside blood vessels, scattered in the connective tissue of the dermis or in the subcutaneous adipose tissue) and was evaluated in 5 randomly selected fields at 40X magnification for each sample. A + +<--- Page Split ---> + +semi- quantitative ordinal score was used to grade the trypanosomes burden in the skin, as follows: \(0 =\) no parasites; \(1 = 1\) to 19 trypanosomes; \(2 = 20 - 50\) trypanosomes; \(3 = >\) 50 trypanosomes. An average parasite burden score per field of view was calculated for each skin section. The severity of skin inflammation was assessed semi- quantitatively, graded on a 0–3 scale: 0 (leukocytes absent or rarely present); 1, mild (low numbers of mixed inflammatory cells present); 2, moderate (increased numbers of mixed inflammatory cells), and 3, marked (extensive numbers/aggregates of inflammatory cells). The average of 10 randomly selected fields at 20X objective magnification for each skin section determined the inflammatory score. + +DNA extraction and Quantitative Polymerase Chain Reaction (qPCR) of T. brucei in murine skin tissue. Genomic DNA (gDNA) was extracted from 25–30 mg skin tissue preserved at - 80°C. After defrosting on ice, skin was finely chopped with scissors, and disrupted for 8 minutes in 300 μL ATL buffer (Qiagen) using a Qiagen Tissuelyser at 50Hz with a ceramic bead (MPBio). Disrupted tissues was incubated at 56°C with 2mg/ml proteinase K (Invitrogen) overnight and DNA extracted from digested tissue using the Qiagen DNeasy Blood and Tissue Kit (Qiagen, Manchester (UK). The resulting gDNA was quantified using a Qubit Fluorimeter (Thermofisher Scientific). and diluted to 4 ng/μl. Trypanosome load in the skin was determined using Taqman real-time PCR, using primers and probe specifically designed to detect the trypanosome Pfr2 gene, as previously reported30. Reactions were performed in a 25 μL reaction mix comprising 1X Taqman Brilliant III master mix (Agilent, Stockport, UK), 0.2 pmol/μl forward primer (CCAACCGTTGTTTCCTCCT), 0.2 pmol/μl reverse primer (CGGCAGTAGTTTGACACCTTTTC), 0.1 pmol/μl probe (FAM- CTTGTCTTCTCCTTTTTGTCCTTTCCCCCT-TAMRA) (Eurofins Genomics, Germany) and 20 ng template DNA. A standard curve was constructed using a serial 10- fold dilution range: \(1 \times 10^{6}\) to \(1 \times 10^{1}\) copies of PCR 2.1 vector containing the cloned Pfr2 target sequence (Eurofins Genomics, Germany). The amplification was performed on an ARIAMx system (Agilent, USA) with a thermal profile of 95°C for 3 minutes followed by 45 cycles of 95°C for 5 seconds and 60°C for 10 seconds. The Pfr2 copy number within each 20ng DNA skin sample was calculated from the standard curve using the ARIAMx qPCR software (Agilent, USA) as a proxy for estimated trypanosome load. Skin biopsies from naïve controls were included to determine the background signal and detection threshold. + +<--- Page Split ---> + +**Flow cytometry analysis of skin-dwelling T lymphocytes.** Murine skin biopsies were harvested and digested as indicated above. The resulting single cell suspensions were resuspended in ice-cold FACS buffer (2 mM EDTA, 5 U/ml DNAse I, 25 mM HEPES and 2.5% Foetal calf serum (FCS) in 1× PBS), blocked with TruStain FcX (anti-mouse CD16/32) antibody (Biolegend, 1:1,000), and stained for extracellular markers at 1:400 dilution. Macrophages (F4/80), B cells (CD19), and erythrocytes (TER119) were excluded from the analysis by included them in a dump channel. For intracellular staining, single-cell suspensions were stimulated as above in Iscove’s modified Dulbecco’s media (supplemented with 1× non-essential amino acids, 50 U/ml penicillin, 50 μg/ml streptomycin, 50 μM β-mercaptoethanol, 1 mM sodium pyruvate and 10% FBS. Gibco) containing 1X cell Stimulation cocktail containing phorbol 12-myristate 13-acetate (PMA), lonomycin, and Brefeldin A (eBioSciences™) for 5 hours at 37°C and 5% CO₂. Unstimulated controls were also included in these analyses. After surface marker staining, cells were fixed and permeabilized with a Foxp3/Transcription Factor Staining Buffer Set (eBioscience) and stained overnight at 4 °C. The list of flow cytometry antibodies used in this study were obtained from Biolegend and are presented in the table below. For flow cytometry analysis, samples were run on a flow cytometer LSRFortessa (BD Biosciences) and analysed using FlowJo software version 10 (Treestar). + +
SupplierTargetCloneDilution
ThermoFixable viability dye eFluor 780-1:1,000
BiolegendF4/80 PE-Cy7BM81:400
BiolegendCD19 PE-Cy71D3/CD191:400
BiolegendTer119 PE-Cy7TER-1191:400
BiolegendCD45 PEHI301:400
BiolegendCD3e PE-Dazzle 594KT3.1.11:400
BiolegendTCRgd Brilliant Violet 421GL31:400
BiolegendCD27 APCLG.3A101:400
BiolegendCD45 PE-Dazzle 59430-F111:400
BiolegendCD4 APCGK1.51:400
BiolegendCD8a Brilliant Violet 71153-6.71:400
BiolegendIFNγ PEXMG1.21:400
BiolegendCD3e Alexa Fluor 488500A21:400
+ +<--- Page Split ---> + +## Data availability + +The transcriptome data generated in this study have been deposited in the Gene Expression Omnibus (GSE226113). The processed transcript count data and cell metadata generated in this study are available at Zenodo (DOI: 10.5281/zenodo.7677469). Additional data and files can also be sourced via Supplementary Tables. + +## Code availability + +Code used to perform analysis described can be accessed at Zenodo (DOI: 10.5281/zenodo.7677469). + +<--- Page Split ---> + +## Results + +# Spatially resolved single cell atlas of the murine skin chronically infected with African trypanosomes + +To study the immune responses in the murine skin against African trypanosomes, we conducted a combined single cell and spatial transcriptomic analysis using 10X Genomics (Materials and methods). Skin from female BALB/c mice infected with T. brucei Antat 1.1 displayed marked inflammation, as determined by H&E examination (Figure 1A), compared to uninfected skin. Parasites were detected in extravascular spaces (Figure 1B) at 25 days post- infection, as previously reported21. We chose to explore this time point as the T. brucei skin infection is well established, enabling us to explore how the skin adapts to such conditions. Following skin dissociation and scRNAseq analysis, we obtained a total of 56,876 high- quality cells with an average of 1,622 genes and 29,651 reads per cell (Figure 1C, S1 Figure, and S1 Table) from naïve \((n = 2)\) and infected \((n = 2)\) mice. These cells were broadly classified into 16 different clusters (Figure 1C) based on common expression makers putatively associated with these clusters (Figure 1D). The stromal compartment consisted of eight keratinocyte clusters (46,591 cells; KC 1 to 8), characterised by a high expression level for Krt14, Krt15, Krt35, Krt25, and Fabp5, in addition to one cluster with high expression of Col1a1, Ly6a, Thy1, Pdgfra, and Cd34 that we designated as fibroblast- like mesenchymal cells (4,274 cells), one Cd36+ Cldn5+ Pecam1+ endothelial cell cluster (792 cells), one Pparg+ adipocyte (529 cells), and one Mlanat+ melanocyte cluster (755 cells) (Figure 1C and D). In the spatial context, the keratinocyte populations were mainly detected in the dermis and epidermis, whereas the rest of the stromal cells were detected in the muscle layer and subcutaneous adipose tissue in both naïve and infected samples (Figure 1E, S2 Figure, and S2 Table). + +The immune compartment consisted of Cd3g+ Trdc+ T cells (1,103 cells), Cd207+ Langerhans cells (854 cells), Lyz2+ myeloid cells (1,499 cells), and erythrocytes (479 cells) (Figure 1C and 1D). The majority of these immune cells were lowly detected in naïve skin but were readily found in the subcutaneous adipose tissue of infected samples (Figure 1E). A closer examination of differences between experimental conditions allowed us to detect the expansion of individual cell clusters in response to infection. For example, the Lyz2+ myeloid and Cd3g+ T cell clusters have higher frequencies in the infected skin compared to naïve controls (Figure 1F). Similarly, we noted the presence of a small erythrocyte cluster exclusively in the infected sample, which may be indicative + +<--- Page Split ---> + +of infection- induced vascular leakage (Figure 1F). These results are consistent with the increased frequency of inflammatory cells in infected samples that we determined by histopathological examination. + +Skin preadipocytes and keratinocytes provide inflammatory signals and antigen presentation during chronic T. brucei infection + +Stromal cells are increasingly recognised as critical for initiating immunological responses in many tissues, including the skin31. To capture as much stromal cell resolution as possible, and to better understand how this compartment in the murine skin responds to infection, we reclustered stromal cell clusters (keratinocytes, adipocytes, endothelial cells, melanocytes, and fibroblast- like mesenchymal cells) and reanalysed them separately. After re- clustering, we obtained a total of 13 different stromal clusters, seven of which were bona fide keratinocytes expressing Krt25, Krt14, and Krt15 (Figure 2A and 2B). In addition to these clusters, we also detected Mlana+ melanocytes, Plin2+ mature adipocytes, and Cldn5+ endothelial cells (Figure 2A and 2B). Interestingly, the re- clustering analysis led us to identify a clear population of cells expressing bona fide mesenchymal genes including Cd34, Thy1, Ly6a, Pdgfra, Pdgfrb, and Col1a1, which we previously assigned as a fibroblast population (Figure 2A and 2B). We assigned these cells as interstitial preadipocytes (stem- like adipocyte precursor cells) as they also express Dpp4, Pi16, and Dcn (Figure 2A and B), markers previously shown to be upregulated in this cell type32. Some stromal populations were altered during infection. For instance, the frequency of KC2 and KC3 increased from around 15.02% and 10.92% to approximately 19.3% and 14.07%, respectively. In contrast, the adipocyte population decreased from 1.60% to 1.21%, indicating subcutaneous adipose tissue wasting (Figure 2C), as previously reported24,33. We also observed a higher frequency of endothelial cells (ECs; from 6.67% to 8.02%) and interstitial preadipocyte population 2 (IPA2; from 1.98% to 3.81%) in infected compared with naïve skin (Figure 2C). Together this indicates that during T. brucei infection there is a remodelling of the stromal compartment within the skin. Stromal cells are recognised as drivers of inflammatory responses via the production of cytokines and chemokines31, as well as the induction of T cell activation via antigen presentation34. To unbiasedly assess our transcriptomics dataset, to determine whether stromal cells were influencing immune function, we performed module scoring. This was used to determine the global expression levels of cytokines (Figure 2D) and antigen presentation molecules (Figure 2E). This analysis revealed that IPA2 (interstitial preadipocyte 2), and to a lesser extent KC2 (keratinocyte + +<--- Page Split ---> + +2), were the main producers of cytokines and antigen presentation molecules (Figure 2D and 2E). Indeed, almost exclusively in IPA2, we noted increased expression of Il6, Il15, Il18bp and interferon- driven Cxcl9, Cxcl10, as well as the class Ib and II major histocompatibility complex H2- M3 and H2- DMa, respectively (Figure 2F- 2H). In the spatial context, most of these responses were identified as restricted foci in the dermis and epidermis of naive mice (Figure 2I), but significantly localised to the subcutaneous adipose tissue in infected mice (Figure 2I). Taken together, these results indicate that the IPA2 subcluster, located in the subcutaneous adipose tissue, is a key driver of recruitment and activation of immune cells in the skin in response to T. brucei infection. Langerhans cells, dendritic cells, and Cd14+ monocytes contribute to antigen presentation in the murine skin during chronic T. brucei infection + +In addition to the stroma, resident myeloid cells are also involved in the initial responses to infection. Thus, we next examined transcriptional responses within the myeloid compartment, encompassing \(Lyz2^{+}\) myeloid and \(Cd207^{+}\) Langerhans cells (Figure 1C and 1D). To gain as much granularity as possible within this compartment, we reclustered the myeloid subset leading to the identification of six subclusters including \(Cd14^{+}\) monocytes, \(Mrc1^{+}\) macrophages, two clusters of \(Cd207^{+}\) Langerhans cells, \(Hdc^{+}\) Gata2+ mast cells, and \(Clec9a^{+}\) Btla\(^{+}\) Xcr1\(^{+}\) dendritic cells, which we classed as cDC1s (Figure 3A and 3B). Interestingly, \(Mrc1^{+}\) macrophages also expressed Il10, suggesting that these cells may have anti- inflammatory properties (Figure 3B). We found that upon infection, there is an expansion of all the myeloid cells, but notably so within the \(Mrc1^{+}\) macrophage and \(Cd14^{+}\) monocyte and Langerhans cell subclusters (Figure 3A). These subsets, as well as the mast cells and the cDC1s localise to the subcutaneous adipose tissue (Figure 3C), mirroring the localisation of other stromal cells driving immune cell recruitment and activation (Figure 2I). Given their role in initiating an adaptive immune response, we next focussed on characterising myeloid cell function by measuring the global expression levels of pro- inflammatory cytokines and molecules associated with antigen presentation, as we did for the stromal compartment, using module scoring. This analysis revealed that the \(Cd14^{+}\) monocytes and \(Mrc1^{+}\) macrophages express the highest levels of cytokines within the myeloid compartment (Figure 3D). In contrast, antigenic presentation was predominantly driven by Langerhans cells and DCs (Figure 3E). Taken together, these analyses reveal that myeloid cells located in the subcutaneous adipose tissue drive the expression of pro- inflammatory cytokines and antigen presentation concertedly with stromal cells during infection. + +<--- Page Split ---> + +## Both Cd4+ T cells and Vγ6+ cells expand in the skin during chronic T. brucei infection + +We next examined the T cell compartment in our scRNAseq dataset. After re- clustering, we identified a total of 1,043 cells encompassing \(I l5^{+}\) Gata3+ ILC2s (138 cells), Ncr1+ NK cells (172 cells), Icos+ Rora+ CD4+ T cells (426 cells), and two separate cell clusters of \(T r d c^{+}\) \(\gamma \delta \mathrm{T}\) cells, 1 (155 cells) and 2 (152 cells) (Figure 4A and 4B). Both of the \(\gamma \delta \mathrm{T}\) cell clusters express high levels of Tcrg- C1, Cd163/1, but low levels of Cd27 (Figure 4B), and based on recent literature, this indicates that they are likely to be IL- 17- producing \(\mathrm{V}\gamma \delta^{+}\) cells35. In vivo, we observed a significant expansion of CD27- (IL- 17A- producing) \(\gamma \delta \mathrm{T}\) cells and a concomitant reduction in the frequency of CD27+ (IFNγ- producing) \(\gamma \delta \mathrm{T}\) cells in skin biopsies from infected BALB/c mice compared to naïve controls (Figure 4C), following the same trend predicted by the scRNAseq data (Figure 4A and 4B), thus validating our in silico prediction. Intriguingly, we noted that the \(\mathrm{V}\gamma \delta^{+}\) cell cluster 1 was only present in the naïve skin but disappeared upon infection (Figure 4A). In contrast, the \(\mathrm{V}\gamma \delta^{+}\) cell cluster 2 was not present in the naïve skin but appeared during infection (Figure 4A). Based on this finding, we hypothesised that these two \(\gamma \delta \mathrm{T}\) cell clusters represent different activation states, whereby \(\gamma \delta \mathrm{T}\) cells within cluster 1 represent "resting" cells, and \(\gamma \delta \mathrm{T}\) cell within cluster 2 represent "activated" cells. To explore this hypothesis in more detail, we examined the expression level of \(\gamma \delta \mathrm{T}\) cell activation and replication markers, namely, Cd44, Cd69, Nr4a1, and Mki67. We found that, upon infection, the \(\gamma \delta \mathrm{T}\) cells within cluster 2 robustly express Cd69 and Nr4a1, and to a lesser extent Mki67, suggesting local activation and expansion of the \(\gamma \delta \mathrm{T}\) cells (Figure 4D). Spatial module scoring analysis identified that \(\mathrm{V}\gamma \delta^{+}\) cells localise mainly to the dermis and epidermis of naïve mice (Figure 4E- F), but their distribution changes in response to infection, and they localised to the subcutaneous adipose tissue (Figure 4E- F). Together, these results indicate that during infection, there is a population of \(\mathrm{V}\gamma \delta^{+}\) cells within the subcutaneous adipose tissue compartment that exhibits features of local activation. During cold challenge, \(\mathrm{V}\gamma \delta^{+}\) cells are known to drive thermogenesis and lipid mobilisation via lipolysis through crosstalk with brown and beige adipocytes36,37. However, interactions between \(\mathrm{V}\gamma \delta^{+}\) cells and white adipocytes, in particular in the subcutaneous white adipose tissue during infection, has yet to be explored. + +We next examined whether there is a cell- cell communication axis between adipocytes and all T cell clusters found in the skin during infection. NicheNet analysis indicates that + +<--- Page Split ---> + +adipocytes provide several critical cues for T cell recruitment and activation, including the activating factors Cd40, Tnfsf18, and Icam1, chemokines Ccl12, Ccl8, and cytokines Tnf, Il10, and Il6 (Figure 4G). These adipocyte- derived ligands are predicted to be sensed by skin- resident T cells via the expression of Cd40lg, Itgax, Il10ra, Il6st, Tnfrsf1b, and Tnfrsf18. In the spatial context, we noted that the expression of adipocyte derived Tnfsf18 is restricted to the subcutaneous adipose tissue in the infected skin, coinciding with the expression of T cells expressing their cognate receptor, Tnfrsf18 (Figure 4H). Taken together, these results demonstrate that chronic skin infection with T. brucei leads to the local activation of a subpopulation of \(\mathrm{V}\gamma 6^{+}\) cells in a process likely aided by subcutaneous adipocytes. + +\(\mathbf{V}\gamma 6^{+}\) cells are essential for controlling skin inflammation, local \(\mathbf{CD8^{+}}\) T cell activation, and subcutaneous adipose tissue wasting independently of skin- resident \(\mathbf{T}_{\mathrm{H}}1\) T cells + +So far, our data indicates that \(\mathrm{V}\gamma 6^{+}\) cells expand significantly in the chronically infected skin, and we predicted that these populations receive activation signals from adipocytes within the subcutaneous adipose tissue. To better understand if the skin- resident \(\mathrm{V}\gamma 6^{+}\) cells are involved in a communication axis with subcutaneous adipocytes, we re- clustered the T cells and adipocytes together to conduct ligand- receptor mediated cell- cell communication analyses between these cell types. We found that \(\mathrm{V}\gamma 6^{+}\) cells express several key genes involved in driving mesenchymal (including preadipocyte) differentiation and adipocyte lipolysis (Figure 5A). For instance, we detected an upregulation of the Cardiotrophin- like cytokine factor 1 (Clcf1) and Amphiregulin (Areg) in the \(\mathrm{V}\gamma 6^{+}\) cells during infection, which are predicted to engage with their cognate receptors Cntfr and Egfr, respectively (Figure 5A and B). Other \(\mathrm{V}\gamma 6^{+}\) cell- derived factors promoting mesenchymal differentiation are Cd24a, the Placental growth factor (Pgf), Sialophorin (Sgn), and Pleiotrophin (Ptn), which are predicted to engage to Pparg+ adipocytes via Selectin P (Selp), Neuropilin 1/2 (Nrp1/Nrp2), Syndecan 3 (Sdc3), and Sialic Acid Binding Ig Like Lectin 1 (Siglec1), respectively (Figure 5A). Interestingly, some of these \(\mathrm{V}\gamma 6^{+}\) cell- derived factors are known to modulate thermogenesis, leading to mobilisation of lipids via lipolysis36- 38. Together, this suggests that during infection \(\mathrm{V}\gamma 6\) \(\gamma \delta \mathrm{T}\) cells may be involved in promoting lipolysis to mobilise energy storage from adipocytes. Based on these observations, we hypothesised that, during infection, \(\mathrm{V}\gamma 6^{+}\) cells expand in the skin and are important for driving subcutaneous adipose tissue + +<--- Page Split ---> + +wasting, limiting IFNγ- driven skin inflammation, and controlling parasite burden. To test our hypothesis, we infected mice with a double \(\mathrm{V}\gamma 4 / 6\mathrm{T}\) cell knockout (on an FVB/N genetic background; \(\mathrm{V}\gamma 4 / 6^{- / - }\) ) for a period of 25 days, and we monitored parasitaemia and clinical scores throughout infection. We observed that both \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice displayed the same levels of parasitaemia and parasite burden in the skin as measured by qRT- PCR against the trypanosome- specific Prf2 gene (S3 Figure). We also detected both slender and stumpy forms in all dermal layers in both infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice without significant differences between strains (S4 Figure), suggesting that \(\mathrm{V}\gamma 4^{+}\) and \(\mathrm{V}\gamma 6^{+}\) cells are dispensable for controlling parasite burden in the skin. + +We next examined histological sections of skin samples from infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice, as well as their counterpart naïve controls. Compared to infected FVB/N mice, the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice displayed more severe signs of skin inflammation. Specifically, we observed higher follicular atrophy in the dermis and hypodermis, as well as diffuse lymphocyte aggregates containing large number of plasma cells and oedema in the subcutaneous adipose tissue compared to infected FVB/N mice (Figure 5C and S3 Table). Histological analysis indicated that, during infection, FVB/N mice lose a greater proportion of subcutaneous adipocytes than their naïve counterparts, consistent with previous reports in the trypanotolerant C57BL/6 background24 (Figure 5C). In contrast, infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice retain similar adipocyte numbers to their naïve counterparts (Figure 5C and 5D), highlighting a critical role for \(\mathrm{V}\gamma 4 / 6\gamma \delta \mathrm{T}\) cells in modulating subcutaneous adipose tissue wasting. Morphometric analysis revealed that adipocytes in the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice were significantly smaller in area in naïve animals compared to the FVB/N background (Figure 5D), potentially highlighting a role for the \(\mathrm{V}\gamma 4^{+}\) and \(\mathrm{V}\gamma 6^{+}\) cells in maintaining adipocyte function under homeostasis. Moreover, our morphometric analyses revealed that adipocytes within the subcutaneous adipose tissue of infected FVB/N mice were significantly smaller than those in naïve mice, whereas infection did not significantly impact adipocyte size in the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice (Figure 5E). + +Lastly, within the T cell compartment, we failed to detect significant differences in the frequency of skin- resident \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells, or the frequency of IFNγ- producing \(\mathrm{CD4^{+}}\) T cells (S5 and S6 Figures), indicating a dispensable role for \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells in limiting \(\mathrm{T}_{\mathrm{H}}1\) - mediated T cell responses. However, we noted that in the skin of the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice there was a significant increase in the frequency of IFNγ- producing \(\mathrm{CD8^{+}}\) T + +<--- Page Split ---> + +cells compared to FVB/N controls, potentially suggesting that \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells may regulate the activation threshold of \(\mathrm{CD8^{+}}\) T cells under homeostatic conditions (S6 Figure). Taken together, our results demonstrate that \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells are critical for limiting skin pathology, at least in part by controlling the activation state of skin- resident \(\mathrm{IFN}\gamma^{+}\) \(\mathrm{CD8^{+}}\) T cell, as well driving subcutaneous adipose tissue wasting, in a process likely involving IL- 17 signalling. + +## Discussion + +To address key questions about the immune response of the skin to chronic infection with T. brucei, this study aimed to characterise changes in skin cell populations using single cell transcriptomics, as well as to determine how the cell populations detected by single cell transcriptomics are distributed throughout the skin during infection using spatial transcriptomics. With this information, we then modelled cell- cell interactions in the skin during T. brucei infection, to understand immune- stromal crosstalk and how this influences the immune response to infection. Here, using a combination of cutting- edge technologies and genetic murine models, we demonstrated that IL- 17- producing \(\mathrm{V}\gamma 6^{+}\) cells play a critical role in the controlling skin inflammation (Figure 6). Furthermore, our data highlight a previously unappreciated interaction between subcutaneous interstitial preadipocytes and mature adipocytes and skin- dwelling T cells (including \(\gamma \delta\) T cells), which mediates T cell responses and subcutaneous adipose tissue wasting (Figure 6). We first generated a spatially resolved single cell atlas of the murine skin during chronic T. brucei infection. From these analyses, several observations are worth discussing in detail. First, we observed significant changes in the skin stromal and immune compartment without the formation of granulomatous lesions, indicative of subclinical inflammatory processes when compared to naïve controls. Using module scoring of inflammatory cytokines and chemokines, as well as genes associated with antigen presentation, we detected inflammatory signatures predominantly in populations of \(\mathrm{Cd14^{+}}\) monocytes, Langerhans cells, and interstitial preadipocytes located in the subcutaneous adipose tissue. + +Following a reclustering of stromal cells, we identified two populations of Dpp4+ interstitial preadipocytes (IPA1 and IPA2) that upregulate inflammatory cytokines, chemokines, and molecules associated with antigen presentation. The chemokines Cxcl1, Cxcl9 and Cxcl10 were upregulated in both populations of interstitial preadipocytes but were higher in IPA2. These chemokines are secreted to recruit neutrophils39, \(\mathrm{CD4^{+}}\) and \(\gamma \delta\) T cells40,41, + +<--- Page Split ---> + +and natural killer cells42, and their upregulation exclusively by preadipocytes suggests that these cells are critical drivers of immune recruitment to the skin during T. brucei infection. Supporting this, in the skin of infected mice we found expansion of \(\mathrm{CD4^{+}}\) T cells, \(\gamma \delta\) T cells, and NK cells, and interestingly this may represent a feedback loop whereby these immune cells suppress differentiation of preadipocytes to mature adipocytes, as observed using in vitro models43. In addition to preadipocytes, we also found that mature adipocytes upregulate chemokines during infection, including Ccl8 and Ccl12, which are drivers of monocyte recruitment44. Moreover, these cells upregulated Il6 and Il10, which were predicted to communicate with T cells through Il6st and Il10ra, respectively. To our knowledge, although adipose tissue immune populations are known to express IL- 1045, and adipocytes express the IL- 10 receptor46, this is the first time that adipocyte Il10 expression has been reported. In the subcutaneous adipose tissue, IL- 10 signalling limits energy expenditure and lipolysis in mouse models of cold exposure and obesity46, but its effects on adipocytes during infection remain unknown. Our observations may suggest that during T. brucei infection, IL- 10 acts in both an autocrine and paracrine fashion, whereby adipocytes secrete the cytokine and it suppresses adipose tissue lipolysis, whilst simultaneously suppressing \(\mathrm{CD4^{+}}\) T cell activity47. Conversely, IL- 6 is a driver of lipolysis and fatty acid oxidation48 and is associated with weight loss and fat wasting in diseases such as HIV49 and cancer50. It is, therefore, unclear how these two cytokines impact adipocyte activity when both are present. + +Importantly, we also identified a population of skin- resident \(\mathrm{V}\gamma \delta^{+}\) T cells that expand in response to skin infection. These \(\mathrm{V}\gamma \delta^{+}\) cells are primed to produce IL- 17A and IL- 17F during development in the thymus51,52, and upon maturation they migrate to multiple tissues throughout the body, including the skin, where they become resident immune cells, offering a first line of response to infection53. The \(\gamma \delta\) T cells that we identified in our dataset express markers putatively associated with activation, including Cd44, Cd69, and Nr4a1, potentially suggesting the existence of local drivers of \(\gamma \delta\) T cell activation in the infected skin. Both in silico predictions and in vivo analyses indicate that these \(\gamma \delta\) T cells are likely to be IL- 17A- producing \(\mathrm{V}\gamma \delta^{+}\) cells based on the expression levels of Tcrg- C1, Cd163/1, and CD27, as previously reported35. We also observed two populations of \(\mathrm{V}\gamma \delta^{+}\) cells, which we hypothesise represent “resting” and “activated” populations, based on expression of Cd44, Cd69, and Nr4a1. However, unlike other \(\gamma \delta\) T cell populations in the skin, such as dendritic epidermal T cells that reside solely in the dermis, \(\mathrm{V}\gamma \delta^{+}\) cells may + +<--- Page Split ---> + +be able to recirculate between tissues54,55. In this scenario, our data may provide evidence of one \(\mathrm{V}_7\mathrm{6}^+\) population exiting the skin and a separate population entering the skin during infection. However, future studies are required to dissect the migratory dynamics of skin \(\mathrm{V}_7\mathrm{6}^+\) cells in the context of skin infection. Our findings further highlight the importance of IL- 17A signalling in the skin of mice infected with \(T\) . brucei, consistent with our previous work proposing IL- 17A as a critical driver of subcutaneous and inguinal adipose tissue wasting24. Interestingly, in the spatial context, these IL- 17A- producing \(\mathrm{V}_7\mathrm{6}^+\) cells are located in the subcutaneous adipose tissue layer of the skin, and are predicted to establish crosstalk with adipocytes via several molecules, including T cell co- stimulatory signals such as Tnfsf18, which engages with GITR (Tnfsf18) to lower the T cell activation threshold56. These cells also express Cd40, indicating a previously unappreciated crosstalk between stromal adipocytes and \(\mathrm{V}_7\mathrm{6}^+\) cells during \(T\) . brucei infection in the skin. Consistent with this, mice lacking \(\mathrm{V}_7\mathrm{4 / 6}^+\) T cells display a higher number of plasma cells and more severe skin inflammation compared to wild type controls. Mice deficient in \(\mathrm{V}_7\mathrm{4}^+\) and \(\mathrm{V}_7\mathrm{6}^+\) are known to develop increased numbers of plasma cells and spontaneous germinal centre formation57, which may dysregulate the immune response to infection. We found that the increased inflammation in the skin of infected \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice may be in part mediated by controlling the activation threshold of skin- resident \(\mathrm{CD8}^+\) T cells. The increased capacity of \(\mathrm{CD8}^+\) T cells to produce \(\mathrm{IFN}\gamma\) in the skin of naive \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice suggests that these cells play a role in constraining \(\mathrm{CD8}^+\) T cell activity under homeostasis. An alternative possibility, is that the increased severity of skin inflammation in infected \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice is due to exacerbated recruitment of neutrophils, as reported in metastatic breast cancer58. Strikingly, we found that \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice do not lose subcutaneous adipose tissue to the same extent as wild type controls during infection, mirroring our previous studies where we proposed IL- 17A as a driver of infection- associated adipose tissue wasting24. In both wild type and \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice we observed comparable parasite tissue burden, with both slender and stumpy developmental forms of the parasite readily detected in all the dermal layers of these mice. Furthermore, we failed to detect significant differences in the frequency of skin- resident \(\mathrm{T}_{\mathrm{H}}1\) T cells, which strongly suggest that both parasites and \(\mathrm{T}_{\mathrm{H}}1\) T cells are dispensable for driving subcutaneous adipose tissue wasting, which is typically observed in this experimental infection setting24,33. Thus, it is tempting to speculate that IL- 17A- producing \(\mathrm{V}_7\mathrm{6}^+\) cells (and potentially other sources of IL- 17A such + +<--- Page Split ---> + +as \(\mathrm{T_H17T}\) cells) promote subcutaneous adipose tissue lipolysis to fuel an efficient immune response against the parasites, although the molecular mechanisms underlying this process need to be investigated in more detail. However, in the study presented here we were unable to dissect the relative contribution of \(\mathrm{V}\gamma 4^{+}\) or \(\mathrm{V}\gamma 6^{+}\) cells to skin inflammation, or the interactions between different \(\gamma \delta\) T cell subsets that reside in the skin under both homeostatic conditions and inflammation. + +Our data strongly indicate that the subcutaneous adipose tissue is an active site for immune priming and activation, placing the adipocytes at the core of this process. Thus, we propose a model whereby subcutaneous adipocytes (in addition to Langerhans cells and keratinocytes) have a critical role as coordinators of local innate and adaptive immune responses. In the context of trypanosome infection, subcutaneous adipocytes may detect the presence of parasites (e.g., via Toll- like receptor signalling) to trigger the recruitment and activation of innate immune cells such as \(\gamma \delta\) T cells to mobilise energy stores to meet the energetic requirements needed to control infection, as recently proposed59. + +Together, our spatially resolved atlas of the murine skin during T. brucei infection offers a resource to the community interested in understanding how chronic infections affect skin homeostasis and immunity. Future work is required to examine the consequences of infection on adipocyte differentiation and function, and whether these processes are directly controlled by \(\gamma \delta\) T cells, as shown during cold exposure36,37. Furthermore, our dataset provides strong evidence for an engagement of several cell types within the stromal compartment in the skin in response to infection, and may suggest that the efficacy, robustness, and timing of the local immune response may be determined by cell types traditionally associated with non- immunological functions such as mesenchymal cells (including interstitial preadipocytes). We envision that future work dissecting the role of these various cell types and communication axis will address some of the fundamental questions arising from this study. + +<--- Page Split ---> + +## Acknowledgments and funding. + +Acknowledgments and funding.We thank Julie Galbraith and Pawel Herzky (Glasgow Polyomics, University of Glasgow) for their support with library preparation and sequencing. Similarly, we would like to thank the technical staff at the University of Glasgow Biological Services for their assistance in maintaining optimal husbandry conditions and comfort for the animals used in this study. We thank the Maria Kasper Lab, Karolinska Institutet, Sweden for their advice on single cell skin dissociation. We thank Dr Jean Rodgers for the work conducted under her Home Office Animal License (PPL No. PC8C3B25C). This work was funded by a Sir Henry Wellcome postdoctoral fellowship (221640/Z/20/Z to JFQ), a Wellcome Trust FutureScope grant (104111/Z/14/Z Wellcome Centre for Integrative Parasitology to JFQ), and a Wellcome Trust Senior Research fellow (209511/Z/17/Z to AML). SBC is supported by a grant from the Annie McNab Bequest (CRUK Beatson Institute), Breast Cancer Now (2018JulPR1101, 2019DecPhD1349, 2019DecPR1424), Cancer Research Institute (CLIP award), Cancer Research UK (RCCCEA- Nov21100003, EDDPGM- Nov21100001, DRCNPG- Jun22\100007), and Pancreatic Cancer Research and Worldwide Cancer Research (22- 0135). PC and MCS are supported by a Wellcome Trust Senior Research fellowship (209511/Z/17/Z) awarded to AML. RH is a Wellcome Trust PhD student (Wellcome Trust 218518/Z/19/Z). The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. + +## Author contributions + +Conceptualisation: JFQ, MCS. Methodology: JFQ, MCS, PC, RH, JO, BC, AL, AC, NK, SBC, AML. Formal analysis: JFQ, MCS, PC, RH, JO, BC, AL. Writing - original draft: JFQ, MCS. Writing - reviewing and editing: JFQ, MCS, PC, RH, JO, BC, AL, AC, SBC, NK, AML. Funding acquisition: JFQ, AML. The authors declare that they have no competing interests, commercial or otherwise. Correspondence and requests for materials should be addressed to Annette MacLeod (Annette.macleod@glasgow.ac.uk) or Juan F. Quintana (juan.quintana@glasgow.ac.uk) + +<--- Page Split ---> + +## Figures + +Figure 1. Integrative overview of the murine skin infected with T. brucei infection using single cell transcriptomics. A) H&E- stained images from naive and infected mice. Scale bar: \(800 \mu \mathrm{m}\) (top) and \(100 \mu \mathrm{m}\) (bottom). B) Immunohistochemistry against the stumpy- specific marker PAD1 in naive and infected samples, including insets highlighting the presence of stumpy forms in both the epidermis- subcutaneous adipose tissue and the subcutaneous adipose tissue- adipose tissue. Scale bar: \(50 \mu \mathrm{m}\) . C) Uniform manifold approximation and projection (UMAP) of 56,876 high- quality cells from both naive and infected samples, highlighting the major cell types detected in our dataset, including stromal cells (Keratinocytes, fibroblasts, endothelial cells, and adipocytes) and immune cells (myeloid cells, T cells, Langerhans cells, and macrophages). D) Top 10 marker genes defining all the major cell clusters detected in (C). The heatmap is colour- coded based on gene expression intensity. E) Spatial feature plot for selected marker genes defining keratinocytes (Krt15), fibroblast- like mesenchymal cells (Col1a1), adipocytes (Pparg), T cells (Cd3g), myeloid cells (Lyz2), and Langerhans cells (Cd207). The corresponding histological section is included on the left, including an annotation of epidermis (Ep), dermis (D), adipose tissue (Ad), Muscle (M), and subcutaneous adipose tissue (Sc). F) As in (C) but splitting the UMAP plot into naïve samples (25,800 cells; \(n = 2\) mice) and infected (31,076 cells; \(n = 2\) ). KC: keratinocytes, FB: Fibroblasts, EC: endothelial cells, MCs: Macrophages, LC: Langerhans cells, Adipo: Adipocytes, Erythro: Erythrocytes. + +Figure 2. The murine skin stromal cells respond to infection by upregulating genes associated with antigen presentation and chemotaxis. A) Uniform manifold approximation and projection (UMAP) of 52,149 high- quality cells within the stromal subcluster, encompassing seven keratinocyte clusters (KC1- 7), fibroblasts (FB), endothelial cells (ECs), melanocytes (MLCs), and interstitial preadipocytes (IPAs). B) Dot plot representing the expression levels of top marker genes used to catalogue the diversity of skin stromal cells. The side of the dots represent the percentage of cells that express a given marker, and the colour intensity represent the level of expression. C) Frequency plot the different stromal cell types detected in the murine skin in naïve (\(n = 2\)) and infected (\(n = 2\)) samples. Module scoring for the overall expression of inflammatory cytokines (D) and genes associated with antigen presentation (E). Violin plot showing the expression level of several significant adipocyte- specific cytokines (F), chemokines (G), + +<--- Page Split ---> + +and major histocompatibility complex (MHC) I and II molecules (H). I) Spatial module scoring for inflammation and antigen presentation in a naïve (top) and infected (bottom) skin section. Scale bar, \(100 \mu \mathrm{m}\) . + +Figure 3. The murine skin is colonised by a myriad of myeloid cells during chronic T. brucei infection. A) Uniform manifold approximation and projection (UMAP) of 2,353 high- quality cells within the myeloid cluster were re- analysed to identify a total four subclusters, including dendritic cells (DCs), Mast cells, and two populations of macrophages \((Mrc1^{+} \text{mas}\) and \(Cd14^{+}\) mono), and two populations of Langerhans cells (LCs 1 and LCs 2). B) Dot plot representing the expression levels of top marker genes used to catalogue the diversity of myeloid cells. C) Spatial feature plot depicting the expression levels of markers defining \(Cd14^{+}\) monocytes \((Cd14, Ccr2)\) , \(Mrc1^{+}\) macrophages \((Mrc1, Lyve1)\) , mast cells \((Hdc)\) , and DCs \((Xcr1)\) . The corresponding histological section is included on the left, including an annotation of epidermis (Ep), dermis (D), adipose tissue (Ad), Muscle (M), and subcutaneous adipose tissue (Sc). Module scoring for the overall expression of inflammatory cytokines (D) and genes associated with antigen presentation (E). + +Figure 4. Chronic T. brucei infection triggers the activation of skin- dwelling Vγ6 γδT cells localised in the subcutaneous adipose tissue. A) Uniform manifold approximation and projection (UMAP) of 1,043 high- quality T cells from naïve (304 cells) and infected skin samples (739 cells). After reclustering, we detected a total of five subclusters including type 2 innate lymphoid cells (ILC2s), CD4+ T cells, NK cells, and γδT cells. B) Dot Plot depicting the expression level of top marker genes for the skin T cell subcluters. The dot size and the intensity of the colour represents the proportion of cells expressing the genes and the level of expression. C) Top panel: Representative flow cytometry analysis to determine the presence of CD27- (IL-17- producing) and CD27+ (IFNγ- producing) γδT cells in murine skin infected \((n = 5 \text{mice})\) with T. brucei and naïve controls \((n = 4 \text{mice})\) . Bottom panel: Quantification of flow cytometry data. Statistical analysis was conducted using a parametric T test. A \(p\) value \(< 0.05\) was considered significant. D) Dot Plot depicting the expression level of top genes associated with T cell activation and TCR engagement for the skin T cell subcluters. The dot size and the intensity of the colour represents the proportion of cells expressing the genes and the level of expression. E) Spatial feature plot of canonical Vγ6+ cell marker genes. F) Spatial module scoring subclusters in the murine naïve and infected skin tissue for canonical γδ + +<--- Page Split ---> + +T cell marker genes. G) In silico cell- cell interaction analysis between adipocytes ("senders") and T cells ("receivers") based on the upregulation of ligand-receptor pairs. The heatmap is colour coded to represent the strength of the interaction. H) Spatial feature plot depicting the expression of adipose- derived ligand Tnfsf18 and T- cell specific receptorTnfsf18, which is predicted to be one of the most robust interactions between these cell types during infection. + +Figure 5. \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells are essential for controlling skin inflammation and subcutaneous adipose tissue wasting independently of skin- resident \(\mathsf{T}_{\mathsf{H}}1\) T cells. A) In silico cell- cell interaction analysis between \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells ("senders") and \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes ("receivers") based on the upregulation of ligand- receptor pairs. The heatmap is colour coded to represent the strength of the interaction. B) Expression level of Clcf1 and \(A\mathsf{r}\mathsf{e}\mathsf{g}\) , two of the most significant upregulated \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells- derived ligands predicted to interact with subcutaneous adipocytes. C) H&E staining from skin biopsies obtained from FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) naive and infected mice. Scale bar: \(100\mu \mathrm{m}\) . Ep, epidermis; D, dermis; Ad, adipose tissue; M, muscle; Sc, subcutis. D) Analysis of mean adipocyte area \((\mu \mathsf{m}^2)\) in naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. \(N = 5\) biological replicates per group, from two independent experiments. Lipid droplets were measured from 3 distinct areas in each image and then combined for each biological replicate. A non- parametric, one- way ANOVA was used to determine the level of significant. A \(p\) value \(< 0.05\) is considered significant. E) Frequency plot of the adipocyte area represented in (D) for naive and infected FVB/N (left panel) and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice (right panel). \(N = 5\) biological replicates per group, from two independent experiments. Lipid droplets were measured from 3 distinct areas in each image and then combined for each biological replicate. A non- parametric, one- way ANOVA was used to determine the level of significant. A \(p\) value \(< 0.05\) is considered significant. + +Figure 6. Proposed model of stromal- immune interactions in the skin during T. brucei infection. Based on our spatially- resolved single cell atlas, we propose a model whereby \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells act concertedly with \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes (and potentially preadipocytes) to coordinate local immune responses, either via the recruitment of immune cells (e.g., CD8 \(^+\) T cells and NK cells). The \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes in this context provide important cues for T cell activation that we hypothesise might be involved in triggering \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells - mediated responses. Other stromal cells such as Langerhans cells and keratinocytes are also likely to be involved in this process via antigenic presentation. + +<--- Page Split ---> + +## Supplementary Figures + +Supplementary figure 1. Quality control measurements of the murine single cell transcriptomics dataset. A) Number of Unique molecular identifies (UMIs), genes, mitochondrial reads, and library complexity (Log10 UMIs/gene) before (left panel) and after (right panel) applying filtering parameters. B) Clustree output representing the relationship between different cell clusters at various levels of resolution using the function FindClusters. + +Supplementary figure 2. Quality control of 10X Visium datasets from the mouse skin over the course of infection with T. brucei. Spatial clusters and marker genes for each spatial cluster in the naive (A) and infected (B) murine skin using 10X Visium spatial transcriptomics. + +Supplementary figure 3. Characterisation of the \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice during T. brucei infection. A) Measurement of circulating parasitaemia in both female FVB/N \((n = 4)\) and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice \((n = 4)\) for a period of 22 days. ANOVA test with multiple corrections. A \(p\) value \(< 0.05\) is considered significant. B) qRT- PCR of skin- dwelling trypanosomes by measuring the trypanosome- specific gene Prf2 by qRT- PCR. The Prf2 copy numbers were normalised per ng of skin DNA from naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). \(T\) test comparison between infected samples. A p value \(< 0.05\) is considered significant. The detection limit, as measured in skin biopsies from naive controls, is also indicated with a dotted line. C) Representative immunohistochemistry of murine skin biopsies from naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice to detect the presence of T. brucei using an antibody against the trypanosome- specific luminal binding protein 1 (BiP). Scale bar \(= 50\mu \mathrm{m}\) + +Supplementary figure 4. Single molecule fluorescent in situ hybridisation (smFISH) analysis showing the spatial distribution of T. brucei developmental stages in the skin of FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. Top panels: skin sections from naive FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) - mice. Middle panels: skin sections from infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice depicting the expression of the T. brucei- specific transcripts Gapdh and Pyk1 (slender specific markers) and Pad2 (stumpy specific marker). The dotted square indicates an area selected for magnification. Bottom panels: Magnified fields of the infected samples showing the distribution of both slender and stumpy markers in the skin of FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. Scale bar: \(50\mu \mathrm{m}\) + +<--- Page Split ---> + +Supplementary figure 5. Flow cytometry analysis of skin- resident T lymphocytes. Gating strategy for the identification of skin \(\gamma \delta\) T cells (A) and \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells (B). + +Supplementary figure 6. Quantification of skin- resident lymphocytes in the \(\mathrm{V}\gamma 4 / 6^{+}\) mice during T. brucei infection. A) Representative flow cytometry analysis of skin \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). B) Quantification of the frequency of skin \(\mathrm{CD4^{+}}\) (left panel) and \(\mathrm{CD8^{+}}\) T cells (right panel) in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group) as shown in (C). ANOVA test with multiple corrections. A \(p\) value \(< 0.05\) is considered significant. C) Representative flow cytometry analysis of ex vivo recall assay to determine the production of IFN \(\gamma\) in skin- resident \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). D) Quantification of the frequency of skin \(\mathrm{CD4^{+}}\) (top panel) and \(\mathrm{CD8^{+}}\) T cells (bottom panel) in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group) as shown in (C). + +## Table legend + +Supplementary table 1. Overview of the mouse skin single cell transcriptomics during chronic T. brucei infection. S1A) Quality control including mean reads per cell and median genes per cell before and after filtering out low quality cell types. S1B) Overview of the major cell types detected in the single cell dataset at a resolution of 0.4. The marker genes are also included. S1C) Overview of the stromal cells detected in the skin dataset at a resolution of 0.3. The marker genes for these clusters, as well as representative UMAP plots are also included. S1D) As in S1C, but for the myeloid cells at a resolution of 0.3. S1E) As in S1C, but for the T cells at a resolution of 0.3. + +Supplementary table 2. Overview of the spatial transcriptomics of the mouse skin during chronic T. brucei infection. S2A) Overview of the spatial transcriptomics project, including total number of reads sequenced per biological replicate, the median number of genes per spot and the percentage of mappable reads to the mouse genome (mm10). S2B) Mouse marker genes identified in the 10X Visium spatial transcriptomics datasets. + +Supplementary table 3. Histopathological analysis of biopsies taken from naive and infected FVB/NJ and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice. The skin biopsies were harvested at 21 days post- infection \((n = 4\) mice/group), fixed in \(10\%\) PFA and counterstained with the T. brucei- specific antibody TbBiP. Uninfected animals \((n = 4)\) were included as naive controls. Histological examinations were scored using H&E staining and BiP staining and was + +<--- Page Split ---> + +conducted double-blinded. The results reported in this table are in comparison to naïve controls for the corresponding genetic background, and encompass a detailed analysis of the epidermis, dermis, hypodermis, skeletal muscle, and the subcutaneous adipose tissue. The column labelled "BiP" (columns AD to Al) represents the results from the immunohistochemistry analysis against the anti-Trypanosoma anti-BiP antibody. + +<--- Page Split ---> + +## Bibliography + +1. Taylor, K. R., Costanzo, A. E. & Jameson, J. M. Dysfunctional \(\gamma \delta\) T cells contribute to impaired keratinocyte homeostasis in mouse models of obesity. J Invest Dermatol 131, 2409-2418 (2011). + +2. Jameson, J. et al. A role for skin gammadelta T cells in wound repair. Science 296, 747-749 (2002). + +3. Chen, Y., Chou, K., Fuchs, E., Havran, W. L. & Boismenu, R. Protection of the intestinal mucosa by intraepithelial \(\gamma \delta\) T cells. Proceedings of the National Academy of Sciences 99, 14338-14343 (2002). + +4. Sharp, L. L., Jameson, J. M., Cauvi, G. & Havran, W. L. Dendritic epidermal T cells regulate skin homeostasis through local production of insulin-like growth factor 1. Nat Immunol 6, 73-79 (2005). + +5. Roark, C. L., Simonian, P. L., Fontenot, A. P., Born, W. K. & O'Brien, R. L. \(\gamma \delta\) T cells: an important source of IL-17. Curr Opin Immunol 20, 353-357 (2008). + +6. Simonian, P. L. et al. \(\gamma \delta\) T cells protect against lung fibrosis via IL-22. Journal of Experimental Medicine 207, 2239-2253 (2010). + +7. Schilbach, K., Frommer, K., Meier, S., Handgretinger, R. & Eyrich, M. Immune response of human propagated gammadelta-T-cells to neuroblastoma recommend the Vdelta1+ subset for gammadelta-T-cell-based immunotherapy. J Immunol 31, 896-905 (2008). + +8. Tan, L. et al. Single-Cell Transcriptomics Identifies the Adaptation of Scart1+ Vγ6+ T Cells to Skin Residency as Activated Effector Cells. Cell Rep 27, 3657-3671.e4 (2019). + +9. Konieczny, P. et al. Interleukin-17 governs hypoxic adaptation of injured epithelium. Science 377, eabg9302 (2022). + +<--- Page Split ---> + +10. Zaharie, R. D., Popa, C., Schlanger, D., Vălean, D. & Zaharie, F. The Role of IL-22 in Wound Healing. Potential Implications in Clinical Practice. Int J Mol Sci 23, 3693 (2022). + +11. Enhancement of Wound Healing by Topical Treatment with Epidermal Growth Factor | NEJM. https://www.nejm.org/doi/full/10.1056/NEJM198907133210203. + +12. Zaiss, D. M. W. et al. Amphiregulin enhances regulatory T cell-suppressive function via the epidermal growth factor receptor. Immunity 38, 275-284 (2013). +13. Patil, R. S. et al. IL17 producing γδT cells induce angiogenesis and are associated with poor survival in gallbladder cancer patients. Int J Cancer 139, 869-881 (2016). + +14. Wu, X. et al. IL-17 promotes tumor angiogenesis through Stat3 pathway mediated upregulation of VEGF in gastric cancer. Tumor Biol. 37, 5493-5501 (2016). + +15. Yan, J. & Huang, J. Innate γδT17 cells convert cancer-elicited inflammation into immunosuppression through myeloid-derived suppressor cells. Oncoimmunology 3, e953423 (2014). + +16. Pan, B. et al. Interleukin-17 promotes angiogenesis by stimulating VEGF production of cancer cells via the STAT3/GIV signaling pathway in non-small-cell lung cancer. Sci Rep 5, 16053 (2015). + +17. Cai, Y. et al. Differential Developmental Requirement and Peripheral Regulation for Dermal Vγ4 and Vγ6T17 Cells in Health and Inflammation. Nat Commun 5, 3986 (2014). + +18. LUNARDI, C., MARGUERIE, C., BOWNESS, P., WALPORT, M. J. & SO, A. K. Reduction in Tγδ cell numbers and alteration in subset distribution in systemic lupus erythematosus. Clinical and Experimental Immunology 86, 203-206 (1991). + +<--- Page Split ---> + +19. Eiras, D. P., Kirkman, L. A. & Murray, H. W. Cutaneous Leishmaniasis: Current Treatment Practices in the USA for Returning Travelers. Curr Treat Options Infect Dis 7, 52–62 (2015). + +20. Ferraresso, M. G., Torre, A. C., Piva, M. M. M. & Barcan, L. Chagas disease reactivation: cutaneous manifestations in a transplanted patient. An Bras Dermatol 93, 890–892 (2018). + +21. Capewell, P. et al. The skin is a significant but overlooked anatomical reservoir for vector-borne African trypanosomes. eLife 5, e17716 (2016). + +22. Camara, M. et al. Extravascular Dermal Trypanosomes in Suspected and Confirmed Cases of gambiense Human African Trypanosomiasis. Clinical Infectious Diseases 73, 12–20 (2021). + +23. De Niz, M. et al. Intravital imaging of host–parasite interactions in skin and adipose tissues. Cellular Microbiology 21, e13023 (2019). + +24. Sinton, M. C. et al. Interleukin-17 drives sex-dependent weight loss and changes in feeding behaviour during Trypanosoma brucei infection. 2022.09.23.509158 Preprint at https://doi.org/10.1101/2022.09.23.509158 (2022). + +25. Le Ray, D., Barry, J. D., Easton, C. & Vickerman, K. First tseste fly transmission of the 'AnTat' serodeme of Trypanosoma brucei. Ann Soc Belg Med Trop 57, 369–381 (1977). + +26. Zappia, L. & Oshlack, A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. GigaScience 7, (2018). + +27. Bankhead, P. et al. QuPath: Open source software for digital pathology image analysis. Sci Rep 7, 16878 (2017). + +28. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676–682 (2012). + +<--- Page Split ---> + +1051 29. Galarraga, M. et al. Adiposoft: automated software for the analysis of white adipose tissue cellularity in histological sections. Journal of Lipid Research 53, 2791–2796 (2012). + +1053 30. Girard, A. et al. Raman spectroscopic analysis of skin as a diagnostic tool for Human African Trypanosomiasis. PLOS Pathogens 17, e1010060 (2021). + +1056 31. Castro-Manrreza, M. E. et al. Mesenchymal Stromal Cells from the Epidermis and Dermis of Psoriasis Patients: Morphology, Immunophenotype, Differentiation Patterns, and Regulation of T Cell Proliferation. Stem Cells Int 2019, 4541797 (2019). + +1059 32. Stefkovich, M., Traynor, S., Cheng, L., Merrick, D. & Seale, P. Dpp4+ interstitial progenitor cells contribute to basal and high fat diet-induced adipogenesis. Mol Metab 54, 101357 (2021). + +1063 33. Trindade, S. et al. Trypanosoma brucei Parasites Occupy and Functionally Adapt to the Adipose Tissue in Mice. Cell Host & Microbe 19, 837–848 (2016). + +1065 34. Baptista, A. P. et al. Lymph node stromal cells constrain immunity via MHC class II self-antigen presentation. eLife 3, e04433 (2014). + +1067 35. Edwards, S. C. et al. PD-1 and TIM-3 differentially regulate subsets of mouse IL-17A–producing γδ T cells. Journal of Experimental Medicine 220, e20211431 (2022). + +1069 36. Kohlgruber, A. C. et al. γδ T cells producing interleukin-17A regulate adipose regulatory T cell homeostasis and thermogenesis. Nat Immunol 19, 464–474 (2018). + +1071 37. Hu, B. et al. γδ T cells and adipocyte IL-17RC control fat innervation and thermogenesis. Nature 578, 610–614 (2020). + +1073 38. Dumas, S. N., Guo, C., Kim, J. K., Friedline, R. H. & Ntambi, J. M. Interleukin-6 derived from cutaneous deficiency of stearoyl-CoA desaturase- 1 may mediate + +<--- Page Split ---> + +1075 metabolic organ crosstalk among skin, adipose tissue and liver. Biochem Biophys Res Commun 508, 87–91 (2019). + +1077 39. Sawant, K. V. et al. Chemokine CXCL1 mediated neutrophil recruitment: Role of glycosaminoglycan interactions. Sci Rep 6, 33123 (2016). + +1079 40. Ochiai, E. et al. CXCL9 Is Important for Recruiting Immune T Cells into the Brain and Inducing an Accumulation of the T Cells to the Areas of Tachyzoite Proliferation to Prevent Reactivation of Chronic Cerebral Infection with Toxoplasma gondii. Am J Pathol 185, 314–324 (2015). + +1083 41. Ajuebor, M. N. et al. GammadeltaT cells initiate acute inflammation and injury in adenovirus-infected liver via cytokine-chemokine cross talk. J Virol 82, 9564–9576 (2008). + +1086 42. Saudemont, A., Jouy, N., Hetuin, D. & Quesnel, B. NK cells that are activated by CXCL10 can kill dormant tumor cells that resist CTL-mediated lysis and can express B7-H1 that stimulates T cells. Blood 105, 2428–2435 (2005). + +1089 43. Wu, H. et al. T-Cell Accumulation and Regulated on Activation, Normal T Cell Expressed and Secreted Upregulation in Adipose Tissue in Obesity. Circulation 115, 1029–1038 (2007). + +1092 44. Islam, S. A. et al. Mouse CCL8, a CCR8 agonist, promotes atopic dermatitis by recruiting IL-5+ TH2 cells. Nat Immunol 12, 10.1038/ni.1984 (2011). + +1094 45. Acosta, J. R. et al. Human-Specific Function of IL-10 in Adipose Tissue Linked to Insulin Resistance. The Journal of Clinical Endocrinology & Metabolism 104, 4552–4562 (2019). + +1097 46. Rajbhandari, P. et al. IL-10 Signaling Remodels Adipose Chromatin Architecture to Limit Thermogenesis and Energy Expenditure. Cell 172, 218-233.e17 (2018). + +<--- Page Split ---> + +1099 47. Akdis, C. A. & Blaser, K. Mechanisms of interleukin-10-mediated immune suppression. Immunology 103, 131–136 (2001).1101 48. van Hall, G. et al. Interleukin-6 stimulates lipolysis and fat oxidation in humans. J Clin Endocrinol Metab 88, 3005–3010 (2003).1103 49. Adedeji, T. A. et al. Serum Interleukin-6 and Weight Loss in Antiretroviral-naïve and Antiretroviral-treated Patients with HIV/AIDS: Relationships and Predictors. Curr HIV Res 20, 441–456 (2022).1106 50. Han, J. et al. Plasma concentration of interleukin-6 was upregulated in cancer cachexia patients and was positively correlated with plasma free fatty acid in female patients. Nutrition & Metabolism 16, 80 (2019).1109 51. Haas, J. D. et al. Development of Interleukin-17-Producing γδ T Cells Is Restricted to a Functional Embryonic Wave. Immunity 37, 48–59 (2012).1111 52. Spidale, N. A. et al. Interleukin-17-Producing γδ T Cells Originate from SOX13+ Progenitors that Are Independent of γδTCR Signaling. Immunity 49, 857-872.e5 (2018).1113 53. Fink, D. R. et al. Elevated numbers of SCART1+ γδ T cells in skin inflammation and inflammatory bowel disease. Molecular Immunology 47, 1710–1718 (2010).1116 54. Audemard-Verger, A. et al. Macrophages Induce Long-Term Trapping of γδ T Cells with Innate-like Properties within Secondary Lymphoid Organs in the Steady State. The Journal of Immunology 199, 1998–2007 (2017).1119 55. Romagnoli, P. A., Sheridan, B. S., Pham, Q.-M., Lefrançois, L. & Khanna, K. M. IL-17A–producing resident memory γδ T cells orchestrate the innate immune response to secondary oral Listeria monocytogenes infection. Proceedings of the National Academy of Sciences 113, 8502–8507 (2016). + +<--- Page Split ---> + +56. Ronchetti, S. et al. Glucocorticoid-induced TNFR-related protein lowers the threshold of CD28 costimulation in CD8+ T cells. J Immunol 179, 5916–5926 (2007).57. Huang, Y. et al. γδ T cells affect IL-4 production and B-cell tolerance.Proceedings of the National Academy of Sciences 112, E39–E48 (2015).58. Coffelt, S. B. et al. IL-17-producing γδ T cells and neutrophils conspire to promote breast cancer metastasis. Nature 522, 345–348 (2015).59. Machado, H., Hofer, P., Zechner, R. & Figueiredo, L. M. Adipocyte lipolysis protects the host against Trypanosoma brucei infection. 2022.11.05.515274 Preprint at https://doi.org/10.1101/2022.11.05.515274 (2022). + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + + + +![](images/Figure_unknown_1.jpg) + + + +![PLACEHOLDER_39_2] + + + +![PLACEHOLDER_39_3] + + + +![PLACEHOLDER_39_4] + + + +![PLACEHOLDER_39_5] + + + +![PLACEHOLDER_39_6] + + + +![PLACEHOLDER_39_7] + + +<--- Page Split ---> +![PLACEHOLDER_40_0] + + +<--- Page Split ---> +![PLACEHOLDER_41_0] + + +<--- Page Split ---> +![PLACEHOLDER_42_0] + + + +
A) Naive (304 cells)
+ +![PLACEHOLDER_42_1] + + + +
C) Pregated on γδTCR+ T cells:
+ +![PLACEHOLDER_42_2] + + + +![PLACEHOLDER_42_3] + + + +![PLACEHOLDER_42_4] + + + +![PLACEHOLDER_42_5] + + + +![PLACEHOLDER_42_6] + + + +![PLACEHOLDER_42_7] + + + +![PLACEHOLDER_42_8] + + + +![PLACEHOLDER_42_9] + + + +![PLACEHOLDER_42_10] + + + +![PLACEHOLDER_42_11] + + + +![PLACEHOLDER_42_10] + + + +![PLACEHOLDER_42_11] + + +<--- Page Split ---> + +A) + +
Vγ6+ cellsNaiveInfected
Cld1
Cd24a
H2-T23
Ifnk
Pcdh7
Pgf
Spn
Wnt4
Ptin
Cntn1
Ltb
Comp
Hmgb2
Hmgb1
Ttf
Clstn1
Ptprf
Areg
Ptprc
Icam5
+ +C) + +![PLACEHOLDER_43_0] + + + +![PLACEHOLDER_43_1] + + + +E) + +![PLACEHOLDER_43_2] + + + +F) + +B) + +![PLACEHOLDER_43_3] + + + +![PLACEHOLDER_43_4] + + + +![PLACEHOLDER_43_5] + + +<--- Page Split ---> +![PLACEHOLDER_44_0] + + +<--- Page Split ---> +![PLACEHOLDER_45_0] + + +<--- Page Split ---> +![PLACEHOLDER_46_0] + + + +![PLACEHOLDER_46_1] + + + +![PLACEHOLDER_46_2] + + + +![PLACEHOLDER_46_3] + + +<--- Page Split ---> +![PLACEHOLDER_47_0] + + +<--- Page Split ---> +![PLACEHOLDER_48_0] + + +<--- Page Split ---> +![PLACEHOLDER_49_0] + + +<--- Page Split ---> +![PLACEHOLDER_50_0] + + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +NCOMMS2309302RS.pdf Supplementarytable1. xlsx Supplementarytable2. xlsx Supplementarytable3. xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3_det.mmd b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9022d7ba4eca3dd24db93ad877caac6b9195218d --- /dev/null +++ b/preprint/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3/preprint__13d30f85eb5012d593c32ee3ab9bd5a627bfd6782c6574db8cc872c0cd4e69b3_det.mmd @@ -0,0 +1,669 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 940, 243]]<|/det|> +# Spatially-resolved single cell transcriptomics reveal a critical role for \(\gamma \delta\) T cells in the control of skin inflammation and subcutaneous adipose wasting during chronic Trypanosoma brucei infection + +<|ref|>text<|/ref|><|det|>[[44, 262, 608, 303]]<|/det|> +Juan Quintana ( \(\square\) juan.quintana@glasgow.ac.uk) University of Glasgow https://orcid.org/0000- 0002- 5092- 5576 + +<|ref|>text<|/ref|><|det|>[[44, 309, 608, 350]]<|/det|> +Matthew Sinton University of Glasgow https://orcid.org/0000- 0003- 1292- 799X + +<|ref|>text<|/ref|><|det|>[[44, 355, 275, 397]]<|/det|> +Praveena Chandrasegaran University of Glasgow + +<|ref|>text<|/ref|><|det|>[[44, 403, 248, 444]]<|/det|> +Agatha Lestari University of Glasgow + +<|ref|>text<|/ref|><|det|>[[44, 450, 608, 491]]<|/det|> +Rhiannon Heslop University of Glasgow https://orcid.org/0000- 0002- 4332- 4393 + +<|ref|>text<|/ref|><|det|>[[44, 496, 248, 537]]<|/det|> +Bachar Cheaib University of Glasgow + +<|ref|>text<|/ref|><|det|>[[44, 543, 248, 584]]<|/det|> +John Ogunsola University of Glasgow + +<|ref|>text<|/ref|><|det|>[[44, 589, 923, 653]]<|/det|> +Dieudonne Mumba Ngoyi National Institute of Biomedical Research & Department of Tropical Medicine, University of Kinshasa https://orcid.org/0000- 0002- 5886- 2004 + +<|ref|>text<|/ref|><|det|>[[44, 658, 416, 700]]<|/det|> +Nono Kuispond Swar National Institute of Biomedical Research + +<|ref|>text<|/ref|><|det|>[[44, 705, 248, 746]]<|/det|> +Anneli Cooper University of Glasgow + +<|ref|>text<|/ref|><|det|>[[44, 751, 563, 792]]<|/det|> +Seth Coffelt Beatson Institute https://orcid.org/0000- 0003- 2257- 2862 + +<|ref|>text<|/ref|><|det|>[[44, 797, 248, 838]]<|/det|> +Annette MacLeod University of Glasgow + +<|ref|>sub_title<|/ref|><|det|>[[44, 880, 101, 897]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 916, 787, 937]]<|/det|> +Keywords: African trypanosomes, \(\gamma \delta\) T cells, \(\mathrm{V}\gamma 6+\) cells, adipocytes, skin inflammation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 46, 313, 64]]<|/det|> +Posted Date: March 14th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 84, 474, 103]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2647707/v1 + +<|ref|>text<|/ref|><|det|>[[42, 120, 910, 164]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 182, 530, 202]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 237, 930, 280]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on August 29th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 40962- y. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 82, 910, 252]]<|/det|> +1 Spatially-resolved single cell transcriptomics reveal a critical role for \(\gamma \delta\) T cells in the control of skin inflammation and subcutaneous adipose wasting during chronic Trypanosoma brucei infection 4 Juan F. Quintana \(^{1,2,3\dagger *}\) , Matthew C. Sinton \(^{1*}\) , Praveena Chandrasegaran \(^{1}\) , Agatha Nabila Lestari \(^{1}\) , Rhiannon Heslop \(^{1}\) , Bachar Cheaib \(^{1}\) , John Ogunsola \(^{1}\) , Dieudonne Mumba Ngoyi \(^{4}\) , Nono-Raymond Kuispond Swar \(^{1,4}\) , Anneli Cooper \(^{1}\) , Seth B. Coffelt \(^{5,6}\) , Annette MacLeod \(^{1\dagger}\) + +<|ref|>text<|/ref|><|det|>[[66, 280, 910, 450]]<|/det|> +9 1Wellcome Centre for Integrative Parasitology (WCIP). University of Glasgow, Glasgow, UK. School of Biodiversity, One Health, Veterinary Medicine (SBOHVM), College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow UK. 2Lydia Becker Institute of Immunology and Inflammation, University of Manchester, UK. 3Division of Immunology, Immunity to Infection and Health, Manchester Academic Health Science Centre, University of Manchester, UK. 4Department of Parasitology, National Institute of Biomedical Research, Kinshasa, Democratic Republic of the Congo. 5School of Cancer Sciences, University of Glasgow Glasgow, UK. 6Cancer Research UK Beatson Institute, Glasgow, UK. + +<|ref|>text<|/ref|><|det|>[[118, 470, 620, 526]]<|/det|> +\*These authors contributed equally to this work. †Corresponding authors: juan.quintana@glasgow.ac.uk, annette.macleod@glasgow.ac.uk + +<|ref|>text<|/ref|><|det|>[[115, 544, 896, 564]]<|/det|> +Keywords: African trypanosomes, \(\gamma \delta\) T cells, \(\mathrm{V}\gamma \delta^{+}\) cells, adipocytes, skin inflammation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 204, 101]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 103, 908, 604]]<|/det|> +African trypanosome parasites colonise the skin in a process important for parasite transmission. However, how the skin responses to trypanosome infection remain unresolved. Here, using a combination of spatial and single cell transcriptomics, coupled with in vivo genetic models, we investigated the local immune response of the skin in a murine model of infection. First, we detected a significant expansion of IL- 17A- producing \(\gamma \delta \mathrm{T}\) cells (primarily \(\mathrm{V}\gamma \delta^{+}\) ) in the infected murine skin compared to naive controls that occur mainly in the subcutaneous adipose tissue. Second, interstitial preadipocytes located in the subcutaneous adipose tissue upregulate several genes involved in inflammation and antigen presentation, including T cell activation and survival. In silico cell- cell communication suggests that adipocytes trigger \(\gamma \delta \mathrm{T}\) cell activation locally via Cd40, Il6, Il10, and Tnfsf18 signalling, amongst others. Third, mice deficient in IL- 17A- producing \(\gamma \delta \mathrm{T}\) cells show extensive inflammation, increased frequency of skin- resident IFN \(\gamma\) - producing \(\mathrm{CD8^{+}}\) T cells and limited subcutaneous adipose tissue wasting compared to wild- type infected controls, independent of \(\mathrm{T}_{\mathrm{H}}1\) \(\mathrm{CD4^{+}}\) T cells and parasite burden. Based on these observations, we proposed a model whereby adipocytes as well as \(\mathrm{V}\gamma \delta^{+}\) cells act concertedly in the subcutaneous adipose tissue to limit skin inflammation and tissue wasting. These studies shed light onto the mechanisms of \(\gamma \delta \mathrm{T}\) cell- mediated immunity in the skin in the context of African trypanosome infection, as well as a potential role of immature and mature adipocytes as homeostatic regulators in the skin during chronic infection. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 84, 240, 101]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[110, 106, 909, 475]]<|/det|> +The skin represents the ultimate barrier, offering physical and mechanical protection against insults including infections. The skin is a complex organ encompassing several tissue layers, including the dermis and epidermis, which are capable of extensive remodelling and resistance to mechanical and chemical stimuli. Moreover, the skin hosts a complex stromal microenvironment containing a myriad of resident and recruited immune cells that actively survey the tissue. Amongst the dermal immune compartment, \(\gamma \delta\) T cells have emerged as critical regulators of tissue homeostasis \(^{1 - 4}\) . For instance, \(\mathrm{V}\gamma 6^{+}\) cells are enriched in mucosal tissues \(^{3}\) and the dermis \(^{2,4}\) where they are considered resident cells. These skin resident \(\gamma \delta\) T cells express a wide range of effector molecules including interleukin 17A (IL- 17A) and IL- 17F \(^{5}\) , IL- 22 \(^{6}\) , and the epidermal growth factor (EGF) receptor ligand \(^{7}\) , and amphiregulin (AREG) \(^{8}\) , which are critical for promoting tissue repair and limiting inflammation following an insult \(^{9 - 12}\) . Although \(\gamma \delta\) T cells have gathered a lot of attention for their role in cancer \(^{13 - 16}\) , and autoimmune disorders affecting the skin such as psoriasis \(^{17}\) and lupus \(^{18}\) , their function during parasitic colonisation of the skin is still not well understood. + +<|ref|>text<|/ref|><|det|>[[110, 480, 909, 745]]<|/det|> +Despite the robust nature of the skin as a barrier, some pathogens manage to circumvent its defences. Of these Trypanosomatids, including the causative agents of leishmaniasis, chagas disease, and sleeping sickness, are known to colonise the skin in a process that has been postulated as being critical for infection and parasite transmission \(^{19 - 21}\) . Indeed, we and others have recently reported the presence of Trypanosoma brucei in the skin of mice and humans \(^{21 - 23}\) . Such skin infections are often associated with dermatitis and pruritus \(^{22}\) or can occur in the absence of clinical symptoms or a positive result from gold- standard diagnostic methods \(^{22}\) . This strongly indicates that mild or asymptomatic skin infections act as important but overlooked parasite reservoirs hampering ongoing eradication efforts. However, to date, the skin response to \(T\) . brucei infection remains unknown. + +<|ref|>text<|/ref|><|det|>[[110, 750, 909, 895]]<|/det|> +Here, we used a combined single cell and spatial transcriptomics approach to generate a spatially resolved single cell atlas of the murine skin during \(T\) . brucei infection. This combined approach led us to identify that interstitial preadipocytes and Langerhans cells both have central roles in the production of local cytokines and antigenic presentation, taking place largely in the subcutaneous adipose tissue and in proximity to the subcutaneous adipose tissue, respectively. Furthermore, we identified a population of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 84, 909, 457]]<|/det|> +skin IL- 17A- producing \(\mathrm{V} / \mathrm{6}^{+}\) cells, located in proximity to adipocytes in the subcutaneous adipose tissue and subcutaneous adipose tissue, that display features of TCR engagement and T cell activation upon infection. Cell- cell communication analyses between subcutaneous adipocytes and \(\mathrm{V} / \mathrm{6}^{+}\) cells predicted several potential interactions mediating not only T cell activation via Tnfsf18, but also promoting adipocyte lipolysis via \(\mathrm{V} / \mathrm{6}^{+}\) cell- derived Clcf1 and Areg. In vivo analyses revealed that \(\mathrm{V} / \mathrm{4} / \mathrm{6} / \mathrm{8} \mathrm{T}\) cell- deficient mice experience severe skin inflammation, likely mediated by an increased frequency of \(\mathrm{IFN} \gamma\) - producing \(\mathrm{CD8}^{+}\) T cells in the skin, but with limited subcutaneous adipose tissue wasting compared to infected wild type controls. We conclude that IL- 17A- producing \(\mathrm{V} / \mathrm{6}^{+}\) cells are critical mediators of skin immunity against African trypanosomes, dampening \(\mathrm{IFN} \gamma\) - mediated \(\mathrm{CD8}^{+}\) T cell responses in the skin and promoting subcutaneous adipose tissue wasting, supporting our recently identified role of IL- 17 signalling as a mediator of adipose tissue wasting during T. brucei infection24. More broadly, our results provide a spatially resolved atlas that can help to dissect further immunological pathways triggered in the skin in response to infection. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 340, 101]]<|/det|> +## Materials and methods + +<|ref|>text<|/ref|><|det|>[[117, 108, 907, 225]]<|/det|> +Ethical statement. All animal experiments were approved by the University of Glasgow Ethical Review Committee and performed in accordance with the home office guidelines, UK Animals (Scientific Procedures) Act, 1986 and EU directive 2010/63/EU. All experiments were conducted under SAPO regulations and UK Home Office project licence number PC8C3B25C to Dr. Jean Rodgers. + +<|ref|>text<|/ref|><|det|>[[115, 231, 908, 528]]<|/det|> +Murine infections with Trypanosoma brucei. 6- 8- week- old wild- type female BALB/c (stock 000651) and FVB/NJ mice (stock 001800) were purchased from Jackson Laboratories. \(\mathrm{V}_7 / 4 / 6^{- / - }\) mice (a gift from Rebecca O'Brien, National Jewish Health) were backcrossed to FVB/N. Female mice aged (6- 8 weeks old) were used for infection. Six- to- eight- week- old wild- type female BALB/c (stock 000651) and FVB/NJ mice (stock 001800) were purchased from Jackson Laboratories. For infections, mice were inoculated by intra- peritoneal injection with \(\sim 2 \times 10^{3}\) parasites of strain T. brucei brucei Antat 1.1E25. Click or tap here to enter text. Parasitaemia was monitored by regular sampling from tail venepuncture and blood was examined using phase microscopy and the rapid "matching" method16. Uninfected mice of the same strain, sex and age served as uninfected controls. Mice were fed ad libitum and kept on a 12 h light- dark cycle. All the experiments were conducted between 8h and 12h. + +<|ref|>text<|/ref|><|det|>[[115, 531, 908, 900]]<|/det|> +Skin tissue processing and preparation of single cell suspension for single- cell RNA sequencing. Infected animals and naive controls were anesthetized with isoflurane at 21 days post- infection and perfused transcardially with 25- 30 ml of ice- cold 1X PBS containing \(0.025\%\) (wt/vol) EDTA. Skin biopsies from the abdominal flank area were shaved off completely. Some of these sections were placed in \(4\%\) PFA for 16 hours prior to embedding in paraffin and histological analysis. Single- cell dissociations for scRNAseq experiments were performed using the 2- step protocol published by (Joost et al., 2020). Briefly, excised \(4\mathrm{cm}^2\) skin sections were diced into small pieces with a scalpel blade and enzymatically digested with Collagenase type I (500 U/ml; Gibco) and DNAse I (1 mg/ml; Sigma) in HBSS containing \(0.04\%\) BSA (Invitrogen) for \(\sim 1\) hr at \(37^{\circ}\mathrm{C}\) with shaking at 300rpm. Liberated cells from the partially digested tissue were pushed through a \(70\mu \mathrm{m}\) nylon mesh filter with an equal volume of HBSS \(0.04\%\) BSA, and then kept on ice. The tissue remaining in the \(70\mu \mathrm{m}\) filter was incubated in \(0.05\%\) trypsin EDTA for 15 minutes at \(37^{\circ}\mathrm{C}\) , and then liberated cells pushed through the filter with an equal of HBSS \(0.04\%\) BSA. Flowthrough from the two digestion steps were combined, passed through a \(40\mu \mathrm{M}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 908, 375]]<|/det|> +filter to remove any cell aggregate and spun at 350g for 10mins at \(4^{\circ}C\) . Finally, cells were passed through a MACS dead cell removal kit (Miltenyi Biotec) and diluted to \(\sim 1,000\) cells/μl in \(200\mu l\) HBSS \(0.04\%\) BSA and kept on ice until single- cell capture using the 10X Chromium platform. The single cell suspensions were loaded onto independent single channels of a Chromium Controller (10X Genomics) single- cell platform. Briefly, \(\sim 25,000\) single cells were loaded for capture using 10X Chromium NextGEM Single cell 3 Reagent kit v3.1 (10X Genomics). Following capture and lysis, complementary DNA was synthesized and amplified (12 cycles) as per the manufacturer's protocol (10X Genomics). The final library preparation was carried out as recommended by the manufacturer with a total of 14 cycles of amplification. The amplified cDNA was used as input to construct an Illumina sequencing library and sequenced on a Novaseq 6000 sequencers by Glasgow Polyomics. + +<|ref|>text<|/ref|><|det|>[[115, 380, 908, 895]]<|/det|> +Read mapping, data processing, and integration. For FASTQ generation and alignments, Illumina basecall files (\\*.bcl) were converted to FASTQs using bcl2fastq. Gene counts were generated using Cellranger v.6.0.0 pipeline against a combined Mus musculus (mm10) and Trypanosoma brucei (TREU927) transcriptome reference. After alignment, reads were grouped based on barcode sequences and demultiplexed using the Unique Molecular Identifiers (UMIs). The mouse- specific digital expression matrices (DEMs) from all six samples were processed using the R (v4.2.1) package Seurat v4.1.0 17. Additional packages used for scRNAseq analysis included dplyr v1.0.7 18, RColorBrewer v1.1.2 (http://colorbrewer.org), ggplot v3.3.5 19, and sctransform v0.3.3 20. We initially captured 65,734 cells mapping specifically against the M. musculus genome across all conditions and biological replicates, with an average of 29,651 reads/cell and a median of 1,762 genes/cell (S1A Table and Supplementary Figure 1). The number of UMIs was then counted for each gene in each cell to generate the digital expression matrix (DEM). Low quality cells were identified according to the following criteria and filtered out: \(i\) ) nFeature \(>100\) or \(< 5,000\) , \(ii\) ) nCounts \(>100\) or \(< 20,000\) , \(iii\) ) \(>30\%\) reads mapping to mitochondrial genes, and \(iv\) ) \(>40\%\) reads mapping to ribosomal genes, v) genes detected \(< 3\) cells. After applying this cut- off, we obtained a total of 56,876 high quality mouse- specific cells with an average of 29,651 reads/cells and a median of 1,683 genes/cell (S1A Table). High- quality cells were then normalised using the SCTransform function, regressing out for total UMI and genes counts, cell cycle genes, and highly variable genes identified by both Seurat and Scater packages, followed by data + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 908, 175]]<|/det|> +integration using IntegrateData and FindIntegrationAnchors. For this, the number of principal components were chosen using the elbow point in a plot ranking principal components and the percentage of variance explained (15 dimensions) using a total of 5,000 genes, and SCT as normalisation method. + +<|ref|>text<|/ref|><|det|>[[115, 177, 908, 870]]<|/det|> +Cluster analysis, marker gene identification, and subclustering. The integrated dataset was then analysed using RunUMAP (10 dimensions), followed by FindNeighbors (10 dimensions, reduction = "pca") and FindClusters (resolution = 0.4). With this approach, we identified a total of 16 cell clusters The cluster markers were then found using the FindAllMarkers function (logfc.threshold = 0.25, assay = "RNA"). To identify cell identity confidently, we employed a supervised approach. This required the manual inspection of the marker gene list followed by and assignment of cell identity based on the expression of putative marker genes expressed in the unidentified clusters. The resolution used for these analyses was selected using the Clustree package26 (Supplementary Figure 1). A cluster name denoted by a single marker gene indicates that the chosen candidate gene is selectively and robustly expressed by a single cell cluster and is sufficient to define that cluster (e.g., Col1a1, Cd3g, Pparg, Krt14, among others). When manually inspecting the gene markers for the final cell types identified in our dataset, we noted the co-occurrence of genes that could discriminate two or more cell types (e.g., T cells, fibroblasts). To increase the resolution of our clusters to help resolve potential mixed cell populations embedded within a single cluster, we subset fibroblasts, myeloid cells, and T cells and analysed them separately using the same functions described above. In all cases, upon subsetting, the resulting objects were reprocessed using the functions FindVariableFeatures, ScaleData, RunUMAP, FindNeighbors, and FindClusters with default parameters. The number of dimensions used in each case varied depending on the cell type being analysed but ranged between 5 and 10 dimensions. Cell type-level differential expression analysis between experimental conditions was conducted using the FindMarkers function (min.pct = 0.25, test.use = Wilcox) and (DefaultAssay = "SCT"). Cell-cell interaction analysis mediated by ligand-receptor expression level was conducted using NicheNet25 with default parameters using "mouse" as a reference organism, comparing differentially expressed genes between experimental conditions (condition_oi = "Infected", condition_reference = "Uninfected"). + +<|ref|>text<|/ref|><|det|>[[117, 870, 727, 890]]<|/det|> +10X Visium spatial sequencing library preparation and analysis + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 907, 920]]<|/det|> +Tissue processing and library preparation. Mice were shaved and dissected skin stored in \(4\%\) paraformaldehyde for \(24\mathrm{h}\) , before paraffin embedding. RNA was purified from FFPE sections to measure integrity, using the Absolutely Total RNA FFPE Purification Kit (Agilent Technologies) as per the manufacturer's instructions. RNA integrity was then measured by BioAnalyzer (Agilent Technologies) using the RNA 6000 Pico Kit (Agilent Technologies). Samples selected for sequencing all had a DV200 \(>50\%\) . We then placed \(10\mu \mathrm{m}\) sections within the capture areas of Visium Spatial Slides (10X Genomics) and proceeded to perform H&E staining and image capture, as per the manufacturer's instructions. To deparaffinise, slides were incubated at \(60^{\circ}\mathrm{C}\) for \(2\mathrm{h}\) before incubating twice in xylene at room temperature, for \(10\mathrm{min}\) each, three times in \(100\%\) ethanol for \(3\mathrm{min}\) each, twice in \(96\%\) ethanol for \(3\mathrm{min}\) each, \(85\%\) ethanol for \(3\mathrm{min}\) , and then submerged in Milli- Q water until ready to stain. For H&E staining, slides were incubated with Mayer's haematoxylin Solution (Sigma- Aldrich), Bluing Buffer (Dako) and Alcoholic Eosin solution (Sigma- Aldrich), with thorough washing in ultrapure water between each step. Stained slides were scanned under a microscope (EVOS M5000, Thermo). De- crosslinking was performed by incubating sections twice with \(0.1\mathrm{NHCl}\) for \(1\mathrm{min}\) at room temperature. Sections were then incubated twice with Tris- EDTA (TE) buffer pH 9.0 before incubating at \(70^{\circ}\mathrm{C}\) for \(1\mathrm{h}\) . The Visium Spatial Gene Expression Slide Kit (10X Genomics) was used for reverse transcription and second strand synthesis, followed by denaturation, to allow the transfer of the cDNA from the slide to a collection tube. These cDNA fragments were then used to construct spatially barcoded Illumina- compatible libraries using the Dual Index Kit TS Set A (10x Genomics) was used to add unique i7 and i5 sample indexes, enabling the spatial and UMI barcoding. The final Illumina- compatible sequencing library underwent paired- end sequencing (150 bp) on a NovaSeq 6000 (Illumina) at GenomeScan. After sequencing, the FASTQ files were aligned to a merged reference transcriptome combining the Mus musculus genome (mm10). After alignment, reads were grouped based on spatial barcode sequences and demultiplexed using the UMLs, using the SpaceRanger pipeline version 1.2.2 (10X Genomics). Downstream analyses of the expression matrices were conducted using the Seurat pipeline for spatial RNA integration (Hao et al., 2021b; Stuart et al., 2019) (Table S3A). Specifically, the data was scaled using the SCTransform function with default parameters. We then proceeded with dimensionality reduction and clustering analysis using RunPCA (assay = "SCT"), FindNeighbours and FindClusters functions with default settings and a total of 30 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 82, 908, 448]]<|/det|> +dimensions. We then applied the FindSpatiallyVariables function to identify spatially variable genes, using the top 1,000 most variable genes and "markvariogram" as selection method. To integrate our skin scRNAseq with the 10X Visium dataset, we used the FindTransferAnchors function with default parameters, using SCT as normalization method. Then, the TransferData function (weight.reduction = "pca", 30 dimensions) was used to annotate brain regions based on transferred anchors from the scRNAseq reference datasets. To predict the cell-cell communication mediated by ligand-receptor co-expression patterns in the spatial context, we employed NICHES v0.0.2 (Raredon et al., 2022). Upon dimensionality reduction and data normalisation, NICHES was run using fanton5 as ligand-receptor database with default parameters. The resulting object was then scaled using the functions ScaleData, FindVariableFeatures (selection.method = "disp"), RunUMAP with default settings and a total of 15 dimensions. Spatially resolved expression of ligand-receptor pairs was then identified using the FindAllMarkers function (min.pct = 0.25, test.use = "roc"). For visualisation, we used the SpatialFeaturePlot function with default parameters and min.cutoff = "q1". + +<|ref|>text<|/ref|><|det|>[[110, 451, 908, 867]]<|/det|> +Single molecule fluorescence in situ hybridisation (smFISH) using RNAscope. smFISH experiments were conducted as follows. Briefly, to prepare tissue sections for smFISH, infected animals and naive controls were anesthetized with isoflurane, and skin sections from the flank were shaved off completely before placing in 4% PFA overnight prior to embedding in paraffin. Paraffin- embedded skin sections sections (3 μm) were prepared placed on a SuperFrost Plus microscope slides. Sections were then then dehydrated in 50, 70 and 100% ethanol. RNAscope 2.5 Assay (Advanced Cell Diagnostics) was used for all smFISH experiments according to the manufacturer's protocols. All RNAscope smFISH probes were designed and validated by Advanced Cell Diagnostics. For image acquisition, 16- bit laser scanning confocal images were acquired with a 63x/1.4 plan- apochromat objective using an LSM 710 confocal microscope fitted with a 32- channel spectral detector (Carl Zeiss). Lasers of 405nm, 488nm and 633 nm excited all fluorophores simultaneously with corresponding beam splitters of 405nm and 488/561/633nm in the light path. 9.7nm binned images with a pixel size of 0.07um x 0.07um were captured using the 32- channel spectral array in Lambda mode. Single fluorophore reference images were acquired for each fluorophore and the reference spectra were employed to unmix the multiplex images using the Zeiss online + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 85, 905, 128]]<|/det|> +fingerprinting mode. smFISH images were acquired with minor contrast adjustments as needed, and converted to grayscale, to maintain image consistency. + +<|ref|>table<|/ref|><|det|>[[118, 133, 904, 312]]<|/det|> + +
List of RNAscope probes used for smFISH
SupplierCat. NumberSequenceChannelDye used
Trypanosoma brucei probes
Biotechnic1103198-C1Tbr-GapdhChannel 1Opal 520
Biotechnic1103208-C2Tbr-Pyk1Channel 2Opal 650
Biotechnic1103218-C3Tbr-Pad2Channel 3Opal 570
Biotechnic1103221-C4Tbr-Ep1Channel 4Opal 540
+ +<|ref|>sub_title<|/ref|><|det|>[[118, 333, 489, 352]]<|/det|> +## Histological analysis of adipocyte size + +<|ref|>text<|/ref|><|det|>[[115, 358, 907, 597]]<|/det|> +Skin placed into \(4\%\) paraformaldehyde (PFA) and fixed overnight at room temperature. PFA- fixed skin biopsies were then embedded in paraffin, sectioned, and stained by the Veterinary Diagnostic Services facility (University of Glasgow, UK). Sections were Haematoxylin and Eosin (H&E) stained for adipocyte size analysis. Slide imaging was performed by the Veterinary Diagnostic Services facility (University of Glasgow, UK) using an EasyScan Infinity slide scanner (Motic, Hong Kong) at 20X magnification. To determine subcutaneous adipocyte size in skin sections, images were first opened in QuPath (v. 0.3.2)27, before selecting regions and exporting to Fiji28. In Fiji, images were converted to 16- bit format, and we used the Adiposoft plugin29 to quantify adipocyte size within different sections. + +<|ref|>text<|/ref|><|det|>[[115, 604, 907, 896]]<|/det|> +Semi- quantitative evaluation of the parasite burden and inflammation in skin sections. Paraffin- embedded skin samples were cut into \(2.5 \mu \mathrm{m}\) sections and stained for T. brucei parasites using a polyclonal rabbit antibody raised against T. brucei luminal binding protein 1 (BiP) (J. Bangs, SUNY, USA) using a Dako Autostainer Link 48 (Dako, Denmark) and were subsequently counterstained with Gill's Haematoxylin. The extent of inflammatory cell infiltration in skin sections was assessed in haematoxylin and eosin- stained sections. Stained slides were assessed by two independent pathologists blinded to infection status and experimental procedures. Skin parasite burden was assessed for both intravascular (parasites within the lumen of dermal or subcutaneous small to medium- sized vessels) and extravascular locations (parasites outside blood vessels, scattered in the connective tissue of the dermis or in the subcutaneous adipose tissue) and was evaluated in 5 randomly selected fields at 40X magnification for each sample. A + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 83, 908, 300]]<|/det|> +semi- quantitative ordinal score was used to grade the trypanosomes burden in the skin, as follows: \(0 =\) no parasites; \(1 = 1\) to 19 trypanosomes; \(2 = 20 - 50\) trypanosomes; \(3 = >\) 50 trypanosomes. An average parasite burden score per field of view was calculated for each skin section. The severity of skin inflammation was assessed semi- quantitatively, graded on a 0–3 scale: 0 (leukocytes absent or rarely present); 1, mild (low numbers of mixed inflammatory cells present); 2, moderate (increased numbers of mixed inflammatory cells), and 3, marked (extensive numbers/aggregates of inflammatory cells). The average of 10 randomly selected fields at 20X objective magnification for each skin section determined the inflammatory score. + +<|ref|>text<|/ref|><|det|>[[111, 300, 908, 902]]<|/det|> +DNA extraction and Quantitative Polymerase Chain Reaction (qPCR) of T. brucei in murine skin tissue. Genomic DNA (gDNA) was extracted from 25–30 mg skin tissue preserved at - 80°C. After defrosting on ice, skin was finely chopped with scissors, and disrupted for 8 minutes in 300 μL ATL buffer (Qiagen) using a Qiagen Tissuelyser at 50Hz with a ceramic bead (MPBio). Disrupted tissues was incubated at 56°C with 2mg/ml proteinase K (Invitrogen) overnight and DNA extracted from digested tissue using the Qiagen DNeasy Blood and Tissue Kit (Qiagen, Manchester (UK). The resulting gDNA was quantified using a Qubit Fluorimeter (Thermofisher Scientific). and diluted to 4 ng/μl. Trypanosome load in the skin was determined using Taqman real-time PCR, using primers and probe specifically designed to detect the trypanosome Pfr2 gene, as previously reported30. Reactions were performed in a 25 μL reaction mix comprising 1X Taqman Brilliant III master mix (Agilent, Stockport, UK), 0.2 pmol/μl forward primer (CCAACCGTTGTTTCCTCCT), 0.2 pmol/μl reverse primer (CGGCAGTAGTTTGACACCTTTTC), 0.1 pmol/μl probe (FAM- CTTGTCTTCTCCTTTTTGTCCTTTCCCCCT-TAMRA) (Eurofins Genomics, Germany) and 20 ng template DNA. A standard curve was constructed using a serial 10- fold dilution range: \(1 \times 10^{6}\) to \(1 \times 10^{1}\) copies of PCR 2.1 vector containing the cloned Pfr2 target sequence (Eurofins Genomics, Germany). The amplification was performed on an ARIAMx system (Agilent, USA) with a thermal profile of 95°C for 3 minutes followed by 45 cycles of 95°C for 5 seconds and 60°C for 10 seconds. The Pfr2 copy number within each 20ng DNA skin sample was calculated from the standard curve using the ARIAMx qPCR software (Agilent, USA) as a proxy for estimated trypanosome load. Skin biopsies from naïve controls were included to determine the background signal and detection threshold. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 84, 908, 525]]<|/det|> +**Flow cytometry analysis of skin-dwelling T lymphocytes.** Murine skin biopsies were harvested and digested as indicated above. The resulting single cell suspensions were resuspended in ice-cold FACS buffer (2 mM EDTA, 5 U/ml DNAse I, 25 mM HEPES and 2.5% Foetal calf serum (FCS) in 1× PBS), blocked with TruStain FcX (anti-mouse CD16/32) antibody (Biolegend, 1:1,000), and stained for extracellular markers at 1:400 dilution. Macrophages (F4/80), B cells (CD19), and erythrocytes (TER119) were excluded from the analysis by included them in a dump channel. For intracellular staining, single-cell suspensions were stimulated as above in Iscove’s modified Dulbecco’s media (supplemented with 1× non-essential amino acids, 50 U/ml penicillin, 50 μg/ml streptomycin, 50 μM β-mercaptoethanol, 1 mM sodium pyruvate and 10% FBS. Gibco) containing 1X cell Stimulation cocktail containing phorbol 12-myristate 13-acetate (PMA), lonomycin, and Brefeldin A (eBioSciences™) for 5 hours at 37°C and 5% CO₂. Unstimulated controls were also included in these analyses. After surface marker staining, cells were fixed and permeabilized with a Foxp3/Transcription Factor Staining Buffer Set (eBioscience) and stained overnight at 4 °C. The list of flow cytometry antibodies used in this study were obtained from Biolegend and are presented in the table below. For flow cytometry analysis, samples were run on a flow cytometer LSRFortessa (BD Biosciences) and analysed using FlowJo software version 10 (Treestar). + +<|ref|>table<|/ref|><|det|>[[118, 528, 737, 908]]<|/det|> +
SupplierTargetCloneDilution
ThermoFixable viability dye eFluor 780-1:1,000
BiolegendF4/80 PE-Cy7BM81:400
BiolegendCD19 PE-Cy71D3/CD191:400
BiolegendTer119 PE-Cy7TER-1191:400
BiolegendCD45 PEHI301:400
BiolegendCD3e PE-Dazzle 594KT3.1.11:400
BiolegendTCRgd Brilliant Violet 421GL31:400
BiolegendCD27 APCLG.3A101:400
BiolegendCD45 PE-Dazzle 59430-F111:400
BiolegendCD4 APCGK1.51:400
BiolegendCD8a Brilliant Violet 71153-6.71:400
BiolegendIFNγ PEXMG1.21:400
BiolegendCD3e Alexa Fluor 488500A21:400
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 84, 274, 101]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 108, 907, 201]]<|/det|> +The transcriptome data generated in this study have been deposited in the Gene Expression Omnibus (GSE226113). The processed transcript count data and cell metadata generated in this study are available at Zenodo (DOI: 10.5281/zenodo.7677469). Additional data and files can also be sourced via Supplementary Tables. + +<|ref|>sub_title<|/ref|><|det|>[[117, 231, 280, 249]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[115, 255, 905, 300]]<|/det|> +Code used to perform analysis described can be accessed at Zenodo (DOI: 10.5281/zenodo.7677469). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 85, 195, 101]]<|/det|> +## Results + +<|ref|>title<|/ref|><|det|>[[117, 108, 904, 151]]<|/det|> +# Spatially resolved single cell atlas of the murine skin chronically infected with African trypanosomes + +<|ref|>text<|/ref|><|det|>[[115, 155, 907, 694]]<|/det|> +To study the immune responses in the murine skin against African trypanosomes, we conducted a combined single cell and spatial transcriptomic analysis using 10X Genomics (Materials and methods). Skin from female BALB/c mice infected with T. brucei Antat 1.1 displayed marked inflammation, as determined by H&E examination (Figure 1A), compared to uninfected skin. Parasites were detected in extravascular spaces (Figure 1B) at 25 days post- infection, as previously reported21. We chose to explore this time point as the T. brucei skin infection is well established, enabling us to explore how the skin adapts to such conditions. Following skin dissociation and scRNAseq analysis, we obtained a total of 56,876 high- quality cells with an average of 1,622 genes and 29,651 reads per cell (Figure 1C, S1 Figure, and S1 Table) from naïve \((n = 2)\) and infected \((n = 2)\) mice. These cells were broadly classified into 16 different clusters (Figure 1C) based on common expression makers putatively associated with these clusters (Figure 1D). The stromal compartment consisted of eight keratinocyte clusters (46,591 cells; KC 1 to 8), characterised by a high expression level for Krt14, Krt15, Krt35, Krt25, and Fabp5, in addition to one cluster with high expression of Col1a1, Ly6a, Thy1, Pdgfra, and Cd34 that we designated as fibroblast- like mesenchymal cells (4,274 cells), one Cd36+ Cldn5+ Pecam1+ endothelial cell cluster (792 cells), one Pparg+ adipocyte (529 cells), and one Mlanat+ melanocyte cluster (755 cells) (Figure 1C and D). In the spatial context, the keratinocyte populations were mainly detected in the dermis and epidermis, whereas the rest of the stromal cells were detected in the muscle layer and subcutaneous adipose tissue in both naïve and infected samples (Figure 1E, S2 Figure, and S2 Table). + +<|ref|>text<|/ref|><|det|>[[115, 697, 907, 914]]<|/det|> +The immune compartment consisted of Cd3g+ Trdc+ T cells (1,103 cells), Cd207+ Langerhans cells (854 cells), Lyz2+ myeloid cells (1,499 cells), and erythrocytes (479 cells) (Figure 1C and 1D). The majority of these immune cells were lowly detected in naïve skin but were readily found in the subcutaneous adipose tissue of infected samples (Figure 1E). A closer examination of differences between experimental conditions allowed us to detect the expansion of individual cell clusters in response to infection. For example, the Lyz2+ myeloid and Cd3g+ T cell clusters have higher frequencies in the infected skin compared to naïve controls (Figure 1F). Similarly, we noted the presence of a small erythrocyte cluster exclusively in the infected sample, which may be indicative + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 905, 151]]<|/det|> +of infection- induced vascular leakage (Figure 1F). These results are consistent with the increased frequency of inflammatory cells in infected samples that we determined by histopathological examination. + +<|ref|>text<|/ref|><|det|>[[115, 155, 905, 199]]<|/det|> +Skin preadipocytes and keratinocytes provide inflammatory signals and antigen presentation during chronic T. brucei infection + +<|ref|>text<|/ref|><|det|>[[112, 201, 907, 920]]<|/det|> +Stromal cells are increasingly recognised as critical for initiating immunological responses in many tissues, including the skin31. To capture as much stromal cell resolution as possible, and to better understand how this compartment in the murine skin responds to infection, we reclustered stromal cell clusters (keratinocytes, adipocytes, endothelial cells, melanocytes, and fibroblast- like mesenchymal cells) and reanalysed them separately. After re- clustering, we obtained a total of 13 different stromal clusters, seven of which were bona fide keratinocytes expressing Krt25, Krt14, and Krt15 (Figure 2A and 2B). In addition to these clusters, we also detected Mlana+ melanocytes, Plin2+ mature adipocytes, and Cldn5+ endothelial cells (Figure 2A and 2B). Interestingly, the re- clustering analysis led us to identify a clear population of cells expressing bona fide mesenchymal genes including Cd34, Thy1, Ly6a, Pdgfra, Pdgfrb, and Col1a1, which we previously assigned as a fibroblast population (Figure 2A and 2B). We assigned these cells as interstitial preadipocytes (stem- like adipocyte precursor cells) as they also express Dpp4, Pi16, and Dcn (Figure 2A and B), markers previously shown to be upregulated in this cell type32. Some stromal populations were altered during infection. For instance, the frequency of KC2 and KC3 increased from around 15.02% and 10.92% to approximately 19.3% and 14.07%, respectively. In contrast, the adipocyte population decreased from 1.60% to 1.21%, indicating subcutaneous adipose tissue wasting (Figure 2C), as previously reported24,33. We also observed a higher frequency of endothelial cells (ECs; from 6.67% to 8.02%) and interstitial preadipocyte population 2 (IPA2; from 1.98% to 3.81%) in infected compared with naïve skin (Figure 2C). Together this indicates that during T. brucei infection there is a remodelling of the stromal compartment within the skin. Stromal cells are recognised as drivers of inflammatory responses via the production of cytokines and chemokines31, as well as the induction of T cell activation via antigen presentation34. To unbiasedly assess our transcriptomics dataset, to determine whether stromal cells were influencing immune function, we performed module scoring. This was used to determine the global expression levels of cytokines (Figure 2D) and antigen presentation molecules (Figure 2E). This analysis revealed that IPA2 (interstitial preadipocyte 2), and to a lesser extent KC2 (keratinocyte + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 907, 350]]<|/det|> +2), were the main producers of cytokines and antigen presentation molecules (Figure 2D and 2E). Indeed, almost exclusively in IPA2, we noted increased expression of Il6, Il15, Il18bp and interferon- driven Cxcl9, Cxcl10, as well as the class Ib and II major histocompatibility complex H2- M3 and H2- DMa, respectively (Figure 2F- 2H). In the spatial context, most of these responses were identified as restricted foci in the dermis and epidermis of naive mice (Figure 2I), but significantly localised to the subcutaneous adipose tissue in infected mice (Figure 2I). Taken together, these results indicate that the IPA2 subcluster, located in the subcutaneous adipose tissue, is a key driver of recruitment and activation of immune cells in the skin in response to T. brucei infection. Langerhans cells, dendritic cells, and Cd14+ monocytes contribute to antigen presentation in the murine skin during chronic T. brucei infection + +<|ref|>text<|/ref|><|det|>[[112, 350, 907, 917]]<|/det|> +In addition to the stroma, resident myeloid cells are also involved in the initial responses to infection. Thus, we next examined transcriptional responses within the myeloid compartment, encompassing \(Lyz2^{+}\) myeloid and \(Cd207^{+}\) Langerhans cells (Figure 1C and 1D). To gain as much granularity as possible within this compartment, we reclustered the myeloid subset leading to the identification of six subclusters including \(Cd14^{+}\) monocytes, \(Mrc1^{+}\) macrophages, two clusters of \(Cd207^{+}\) Langerhans cells, \(Hdc^{+}\) Gata2+ mast cells, and \(Clec9a^{+}\) Btla\(^{+}\) Xcr1\(^{+}\) dendritic cells, which we classed as cDC1s (Figure 3A and 3B). Interestingly, \(Mrc1^{+}\) macrophages also expressed Il10, suggesting that these cells may have anti- inflammatory properties (Figure 3B). We found that upon infection, there is an expansion of all the myeloid cells, but notably so within the \(Mrc1^{+}\) macrophage and \(Cd14^{+}\) monocyte and Langerhans cell subclusters (Figure 3A). These subsets, as well as the mast cells and the cDC1s localise to the subcutaneous adipose tissue (Figure 3C), mirroring the localisation of other stromal cells driving immune cell recruitment and activation (Figure 2I). Given their role in initiating an adaptive immune response, we next focussed on characterising myeloid cell function by measuring the global expression levels of pro- inflammatory cytokines and molecules associated with antigen presentation, as we did for the stromal compartment, using module scoring. This analysis revealed that the \(Cd14^{+}\) monocytes and \(Mrc1^{+}\) macrophages express the highest levels of cytokines within the myeloid compartment (Figure 3D). In contrast, antigenic presentation was predominantly driven by Langerhans cells and DCs (Figure 3E). Taken together, these analyses reveal that myeloid cells located in the subcutaneous adipose tissue drive the expression of pro- inflammatory cytokines and antigen presentation concertedly with stromal cells during infection. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 905, 128]]<|/det|> +## Both Cd4+ T cells and Vγ6+ cells expand in the skin during chronic T. brucei infection + +<|ref|>text<|/ref|><|det|>[[112, 125, 908, 888]]<|/det|> +We next examined the T cell compartment in our scRNAseq dataset. After re- clustering, we identified a total of 1,043 cells encompassing \(I l5^{+}\) Gata3+ ILC2s (138 cells), Ncr1+ NK cells (172 cells), Icos+ Rora+ CD4+ T cells (426 cells), and two separate cell clusters of \(T r d c^{+}\) \(\gamma \delta \mathrm{T}\) cells, 1 (155 cells) and 2 (152 cells) (Figure 4A and 4B). Both of the \(\gamma \delta \mathrm{T}\) cell clusters express high levels of Tcrg- C1, Cd163/1, but low levels of Cd27 (Figure 4B), and based on recent literature, this indicates that they are likely to be IL- 17- producing \(\mathrm{V}\gamma \delta^{+}\) cells35. In vivo, we observed a significant expansion of CD27- (IL- 17A- producing) \(\gamma \delta \mathrm{T}\) cells and a concomitant reduction in the frequency of CD27+ (IFNγ- producing) \(\gamma \delta \mathrm{T}\) cells in skin biopsies from infected BALB/c mice compared to naïve controls (Figure 4C), following the same trend predicted by the scRNAseq data (Figure 4A and 4B), thus validating our in silico prediction. Intriguingly, we noted that the \(\mathrm{V}\gamma \delta^{+}\) cell cluster 1 was only present in the naïve skin but disappeared upon infection (Figure 4A). In contrast, the \(\mathrm{V}\gamma \delta^{+}\) cell cluster 2 was not present in the naïve skin but appeared during infection (Figure 4A). Based on this finding, we hypothesised that these two \(\gamma \delta \mathrm{T}\) cell clusters represent different activation states, whereby \(\gamma \delta \mathrm{T}\) cells within cluster 1 represent "resting" cells, and \(\gamma \delta \mathrm{T}\) cell within cluster 2 represent "activated" cells. To explore this hypothesis in more detail, we examined the expression level of \(\gamma \delta \mathrm{T}\) cell activation and replication markers, namely, Cd44, Cd69, Nr4a1, and Mki67. We found that, upon infection, the \(\gamma \delta \mathrm{T}\) cells within cluster 2 robustly express Cd69 and Nr4a1, and to a lesser extent Mki67, suggesting local activation and expansion of the \(\gamma \delta \mathrm{T}\) cells (Figure 4D). Spatial module scoring analysis identified that \(\mathrm{V}\gamma \delta^{+}\) cells localise mainly to the dermis and epidermis of naïve mice (Figure 4E- F), but their distribution changes in response to infection, and they localised to the subcutaneous adipose tissue (Figure 4E- F). Together, these results indicate that during infection, there is a population of \(\mathrm{V}\gamma \delta^{+}\) cells within the subcutaneous adipose tissue compartment that exhibits features of local activation. During cold challenge, \(\mathrm{V}\gamma \delta^{+}\) cells are known to drive thermogenesis and lipid mobilisation via lipolysis through crosstalk with brown and beige adipocytes36,37. However, interactions between \(\mathrm{V}\gamma \delta^{+}\) cells and white adipocytes, in particular in the subcutaneous white adipose tissue during infection, has yet to be explored. + +<|ref|>text<|/ref|><|det|>[[115, 870, 905, 916]]<|/det|> +We next examined whether there is a cell- cell communication axis between adipocytes and all T cell clusters found in the skin during infection. NicheNet analysis indicates that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 907, 325]]<|/det|> +adipocytes provide several critical cues for T cell recruitment and activation, including the activating factors Cd40, Tnfsf18, and Icam1, chemokines Ccl12, Ccl8, and cytokines Tnf, Il10, and Il6 (Figure 4G). These adipocyte- derived ligands are predicted to be sensed by skin- resident T cells via the expression of Cd40lg, Itgax, Il10ra, Il6st, Tnfrsf1b, and Tnfrsf18. In the spatial context, we noted that the expression of adipocyte derived Tnfsf18 is restricted to the subcutaneous adipose tissue in the infected skin, coinciding with the expression of T cells expressing their cognate receptor, Tnfrsf18 (Figure 4H). Taken together, these results demonstrate that chronic skin infection with T. brucei leads to the local activation of a subpopulation of \(\mathrm{V}\gamma 6^{+}\) cells in a process likely aided by subcutaneous adipocytes. + +<|ref|>text<|/ref|><|det|>[[115, 331, 905, 400]]<|/det|> +\(\mathbf{V}\gamma 6^{+}\) cells are essential for controlling skin inflammation, local \(\mathbf{CD8^{+}}\) T cell activation, and subcutaneous adipose tissue wasting independently of skin- resident \(\mathbf{T}_{\mathrm{H}}1\) T cells + +<|ref|>text<|/ref|><|det|>[[115, 404, 907, 908]]<|/det|> +So far, our data indicates that \(\mathrm{V}\gamma 6^{+}\) cells expand significantly in the chronically infected skin, and we predicted that these populations receive activation signals from adipocytes within the subcutaneous adipose tissue. To better understand if the skin- resident \(\mathrm{V}\gamma 6^{+}\) cells are involved in a communication axis with subcutaneous adipocytes, we re- clustered the T cells and adipocytes together to conduct ligand- receptor mediated cell- cell communication analyses between these cell types. We found that \(\mathrm{V}\gamma 6^{+}\) cells express several key genes involved in driving mesenchymal (including preadipocyte) differentiation and adipocyte lipolysis (Figure 5A). For instance, we detected an upregulation of the Cardiotrophin- like cytokine factor 1 (Clcf1) and Amphiregulin (Areg) in the \(\mathrm{V}\gamma 6^{+}\) cells during infection, which are predicted to engage with their cognate receptors Cntfr and Egfr, respectively (Figure 5A and B). Other \(\mathrm{V}\gamma 6^{+}\) cell- derived factors promoting mesenchymal differentiation are Cd24a, the Placental growth factor (Pgf), Sialophorin (Sgn), and Pleiotrophin (Ptn), which are predicted to engage to Pparg+ adipocytes via Selectin P (Selp), Neuropilin 1/2 (Nrp1/Nrp2), Syndecan 3 (Sdc3), and Sialic Acid Binding Ig Like Lectin 1 (Siglec1), respectively (Figure 5A). Interestingly, some of these \(\mathrm{V}\gamma 6^{+}\) cell- derived factors are known to modulate thermogenesis, leading to mobilisation of lipids via lipolysis36- 38. Together, this suggests that during infection \(\mathrm{V}\gamma 6\) \(\gamma \delta \mathrm{T}\) cells may be involved in promoting lipolysis to mobilise energy storage from adipocytes. Based on these observations, we hypothesised that, during infection, \(\mathrm{V}\gamma 6^{+}\) cells expand in the skin and are important for driving subcutaneous adipose tissue + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 83, 909, 311]]<|/det|> +wasting, limiting IFNγ- driven skin inflammation, and controlling parasite burden. To test our hypothesis, we infected mice with a double \(\mathrm{V}\gamma 4 / 6\mathrm{T}\) cell knockout (on an FVB/N genetic background; \(\mathrm{V}\gamma 4 / 6^{- / - }\) ) for a period of 25 days, and we monitored parasitaemia and clinical scores throughout infection. We observed that both \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice displayed the same levels of parasitaemia and parasite burden in the skin as measured by qRT- PCR against the trypanosome- specific Prf2 gene (S3 Figure). We also detected both slender and stumpy forms in all dermal layers in both infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice without significant differences between strains (S4 Figure), suggesting that \(\mathrm{V}\gamma 4^{+}\) and \(\mathrm{V}\gamma 6^{+}\) cells are dispensable for controlling parasite burden in the skin. + +<|ref|>text<|/ref|><|det|>[[112, 315, 909, 765]]<|/det|> +We next examined histological sections of skin samples from infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) and FVB/N mice, as well as their counterpart naïve controls. Compared to infected FVB/N mice, the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice displayed more severe signs of skin inflammation. Specifically, we observed higher follicular atrophy in the dermis and hypodermis, as well as diffuse lymphocyte aggregates containing large number of plasma cells and oedema in the subcutaneous adipose tissue compared to infected FVB/N mice (Figure 5C and S3 Table). Histological analysis indicated that, during infection, FVB/N mice lose a greater proportion of subcutaneous adipocytes than their naïve counterparts, consistent with previous reports in the trypanotolerant C57BL/6 background24 (Figure 5C). In contrast, infected \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice retain similar adipocyte numbers to their naïve counterparts (Figure 5C and 5D), highlighting a critical role for \(\mathrm{V}\gamma 4 / 6\gamma \delta \mathrm{T}\) cells in modulating subcutaneous adipose tissue wasting. Morphometric analysis revealed that adipocytes in the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice were significantly smaller in area in naïve animals compared to the FVB/N background (Figure 5D), potentially highlighting a role for the \(\mathrm{V}\gamma 4^{+}\) and \(\mathrm{V}\gamma 6^{+}\) cells in maintaining adipocyte function under homeostasis. Moreover, our morphometric analyses revealed that adipocytes within the subcutaneous adipose tissue of infected FVB/N mice were significantly smaller than those in naïve mice, whereas infection did not significantly impact adipocyte size in the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice (Figure 5E). + +<|ref|>text<|/ref|><|det|>[[115, 769, 907, 891]]<|/det|> +Lastly, within the T cell compartment, we failed to detect significant differences in the frequency of skin- resident \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells, or the frequency of IFNγ- producing \(\mathrm{CD4^{+}}\) T cells (S5 and S6 Figures), indicating a dispensable role for \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells in limiting \(\mathrm{T}_{\mathrm{H}}1\) - mediated T cell responses. However, we noted that in the skin of the \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice there was a significant increase in the frequency of IFNγ- producing \(\mathrm{CD8^{+}}\) T + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 907, 232]]<|/det|> +cells compared to FVB/N controls, potentially suggesting that \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells may regulate the activation threshold of \(\mathrm{CD8^{+}}\) T cells under homeostatic conditions (S6 Figure). Taken together, our results demonstrate that \(\mathrm{V}\gamma 4^{+}\) and/or \(\mathrm{V}\gamma 6^{+}\) cells are critical for limiting skin pathology, at least in part by controlling the activation state of skin- resident \(\mathrm{IFN}\gamma^{+}\) \(\mathrm{CD8^{+}}\) T cell, as well driving subcutaneous adipose tissue wasting, in a process likely involving IL- 17 signalling. + +<|ref|>sub_title<|/ref|><|det|>[[118, 237, 230, 253]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[111, 255, 909, 775]]<|/det|> +To address key questions about the immune response of the skin to chronic infection with T. brucei, this study aimed to characterise changes in skin cell populations using single cell transcriptomics, as well as to determine how the cell populations detected by single cell transcriptomics are distributed throughout the skin during infection using spatial transcriptomics. With this information, we then modelled cell- cell interactions in the skin during T. brucei infection, to understand immune- stromal crosstalk and how this influences the immune response to infection. Here, using a combination of cutting- edge technologies and genetic murine models, we demonstrated that IL- 17- producing \(\mathrm{V}\gamma 6^{+}\) cells play a critical role in the controlling skin inflammation (Figure 6). Furthermore, our data highlight a previously unappreciated interaction between subcutaneous interstitial preadipocytes and mature adipocytes and skin- dwelling T cells (including \(\gamma \delta\) T cells), which mediates T cell responses and subcutaneous adipose tissue wasting (Figure 6). We first generated a spatially resolved single cell atlas of the murine skin during chronic T. brucei infection. From these analyses, several observations are worth discussing in detail. First, we observed significant changes in the skin stromal and immune compartment without the formation of granulomatous lesions, indicative of subclinical inflammatory processes when compared to naïve controls. Using module scoring of inflammatory cytokines and chemokines, as well as genes associated with antigen presentation, we detected inflammatory signatures predominantly in populations of \(\mathrm{Cd14^{+}}\) monocytes, Langerhans cells, and interstitial preadipocytes located in the subcutaneous adipose tissue. + +<|ref|>text<|/ref|><|det|>[[115, 779, 907, 898]]<|/det|> +Following a reclustering of stromal cells, we identified two populations of Dpp4+ interstitial preadipocytes (IPA1 and IPA2) that upregulate inflammatory cytokines, chemokines, and molecules associated with antigen presentation. The chemokines Cxcl1, Cxcl9 and Cxcl10 were upregulated in both populations of interstitial preadipocytes but were higher in IPA2. These chemokines are secreted to recruit neutrophils39, \(\mathrm{CD4^{+}}\) and \(\gamma \delta\) T cells40,41, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 909, 576]]<|/det|> +and natural killer cells42, and their upregulation exclusively by preadipocytes suggests that these cells are critical drivers of immune recruitment to the skin during T. brucei infection. Supporting this, in the skin of infected mice we found expansion of \(\mathrm{CD4^{+}}\) T cells, \(\gamma \delta\) T cells, and NK cells, and interestingly this may represent a feedback loop whereby these immune cells suppress differentiation of preadipocytes to mature adipocytes, as observed using in vitro models43. In addition to preadipocytes, we also found that mature adipocytes upregulate chemokines during infection, including Ccl8 and Ccl12, which are drivers of monocyte recruitment44. Moreover, these cells upregulated Il6 and Il10, which were predicted to communicate with T cells through Il6st and Il10ra, respectively. To our knowledge, although adipose tissue immune populations are known to express IL- 1045, and adipocytes express the IL- 10 receptor46, this is the first time that adipocyte Il10 expression has been reported. In the subcutaneous adipose tissue, IL- 10 signalling limits energy expenditure and lipolysis in mouse models of cold exposure and obesity46, but its effects on adipocytes during infection remain unknown. Our observations may suggest that during T. brucei infection, IL- 10 acts in both an autocrine and paracrine fashion, whereby adipocytes secrete the cytokine and it suppresses adipose tissue lipolysis, whilst simultaneously suppressing \(\mathrm{CD4^{+}}\) T cell activity47. Conversely, IL- 6 is a driver of lipolysis and fatty acid oxidation48 and is associated with weight loss and fat wasting in diseases such as HIV49 and cancer50. It is, therefore, unclear how these two cytokines impact adipocyte activity when both are present. + +<|ref|>text<|/ref|><|det|>[[111, 578, 909, 905]]<|/det|> +Importantly, we also identified a population of skin- resident \(\mathrm{V}\gamma \delta^{+}\) T cells that expand in response to skin infection. These \(\mathrm{V}\gamma \delta^{+}\) cells are primed to produce IL- 17A and IL- 17F during development in the thymus51,52, and upon maturation they migrate to multiple tissues throughout the body, including the skin, where they become resident immune cells, offering a first line of response to infection53. The \(\gamma \delta\) T cells that we identified in our dataset express markers putatively associated with activation, including Cd44, Cd69, and Nr4a1, potentially suggesting the existence of local drivers of \(\gamma \delta\) T cell activation in the infected skin. Both in silico predictions and in vivo analyses indicate that these \(\gamma \delta\) T cells are likely to be IL- 17A- producing \(\mathrm{V}\gamma \delta^{+}\) cells based on the expression levels of Tcrg- C1, Cd163/1, and CD27, as previously reported35. We also observed two populations of \(\mathrm{V}\gamma \delta^{+}\) cells, which we hypothesise represent “resting” and “activated” populations, based on expression of Cd44, Cd69, and Nr4a1. However, unlike other \(\gamma \delta\) T cell populations in the skin, such as dendritic epidermal T cells that reside solely in the dermis, \(\mathrm{V}\gamma \delta^{+}\) cells may + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 909, 911]]<|/det|> +be able to recirculate between tissues54,55. In this scenario, our data may provide evidence of one \(\mathrm{V}_7\mathrm{6}^+\) population exiting the skin and a separate population entering the skin during infection. However, future studies are required to dissect the migratory dynamics of skin \(\mathrm{V}_7\mathrm{6}^+\) cells in the context of skin infection. Our findings further highlight the importance of IL- 17A signalling in the skin of mice infected with \(T\) . brucei, consistent with our previous work proposing IL- 17A as a critical driver of subcutaneous and inguinal adipose tissue wasting24. Interestingly, in the spatial context, these IL- 17A- producing \(\mathrm{V}_7\mathrm{6}^+\) cells are located in the subcutaneous adipose tissue layer of the skin, and are predicted to establish crosstalk with adipocytes via several molecules, including T cell co- stimulatory signals such as Tnfsf18, which engages with GITR (Tnfsf18) to lower the T cell activation threshold56. These cells also express Cd40, indicating a previously unappreciated crosstalk between stromal adipocytes and \(\mathrm{V}_7\mathrm{6}^+\) cells during \(T\) . brucei infection in the skin. Consistent with this, mice lacking \(\mathrm{V}_7\mathrm{4 / 6}^+\) T cells display a higher number of plasma cells and more severe skin inflammation compared to wild type controls. Mice deficient in \(\mathrm{V}_7\mathrm{4}^+\) and \(\mathrm{V}_7\mathrm{6}^+\) are known to develop increased numbers of plasma cells and spontaneous germinal centre formation57, which may dysregulate the immune response to infection. We found that the increased inflammation in the skin of infected \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice may be in part mediated by controlling the activation threshold of skin- resident \(\mathrm{CD8}^+\) T cells. The increased capacity of \(\mathrm{CD8}^+\) T cells to produce \(\mathrm{IFN}\gamma\) in the skin of naive \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice suggests that these cells play a role in constraining \(\mathrm{CD8}^+\) T cell activity under homeostasis. An alternative possibility, is that the increased severity of skin inflammation in infected \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice is due to exacerbated recruitment of neutrophils, as reported in metastatic breast cancer58. Strikingly, we found that \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice do not lose subcutaneous adipose tissue to the same extent as wild type controls during infection, mirroring our previous studies where we proposed IL- 17A as a driver of infection- associated adipose tissue wasting24. In both wild type and \(\mathrm{V}_7\mathrm{4 / 6}^{- / - }\) mice we observed comparable parasite tissue burden, with both slender and stumpy developmental forms of the parasite readily detected in all the dermal layers of these mice. Furthermore, we failed to detect significant differences in the frequency of skin- resident \(\mathrm{T}_{\mathrm{H}}1\) T cells, which strongly suggest that both parasites and \(\mathrm{T}_{\mathrm{H}}1\) T cells are dispensable for driving subcutaneous adipose tissue wasting, which is typically observed in this experimental infection setting24,33. Thus, it is tempting to speculate that IL- 17A- producing \(\mathrm{V}_7\mathrm{6}^+\) cells (and potentially other sources of IL- 17A such + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 907, 229]]<|/det|> +as \(\mathrm{T_H17T}\) cells) promote subcutaneous adipose tissue lipolysis to fuel an efficient immune response against the parasites, although the molecular mechanisms underlying this process need to be investigated in more detail. However, in the study presented here we were unable to dissect the relative contribution of \(\mathrm{V}\gamma 4^{+}\) or \(\mathrm{V}\gamma 6^{+}\) cells to skin inflammation, or the interactions between different \(\gamma \delta\) T cell subsets that reside in the skin under both homeostatic conditions and inflammation. + +<|ref|>text<|/ref|><|det|>[[115, 233, 907, 451]]<|/det|> +Our data strongly indicate that the subcutaneous adipose tissue is an active site for immune priming and activation, placing the adipocytes at the core of this process. Thus, we propose a model whereby subcutaneous adipocytes (in addition to Langerhans cells and keratinocytes) have a critical role as coordinators of local innate and adaptive immune responses. In the context of trypanosome infection, subcutaneous adipocytes may detect the presence of parasites (e.g., via Toll- like receptor signalling) to trigger the recruitment and activation of innate immune cells such as \(\gamma \delta\) T cells to mobilise energy stores to meet the energetic requirements needed to control infection, as recently proposed59. + +<|ref|>text<|/ref|><|det|>[[115, 455, 907, 749]]<|/det|> +Together, our spatially resolved atlas of the murine skin during T. brucei infection offers a resource to the community interested in understanding how chronic infections affect skin homeostasis and immunity. Future work is required to examine the consequences of infection on adipocyte differentiation and function, and whether these processes are directly controlled by \(\gamma \delta\) T cells, as shown during cold exposure36,37. Furthermore, our dataset provides strong evidence for an engagement of several cell types within the stromal compartment in the skin in response to infection, and may suggest that the efficacy, robustness, and timing of the local immune response may be determined by cell types traditionally associated with non- immunological functions such as mesenchymal cells (including interstitial preadipocytes). We envision that future work dissecting the role of these various cell types and communication axis will address some of the fundamental questions arising from this study. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 425, 103]]<|/det|> +## Acknowledgments and funding. + +<|ref|>text<|/ref|><|det|>[[115, 108, 909, 570]]<|/det|> +Acknowledgments and funding.We thank Julie Galbraith and Pawel Herzky (Glasgow Polyomics, University of Glasgow) for their support with library preparation and sequencing. Similarly, we would like to thank the technical staff at the University of Glasgow Biological Services for their assistance in maintaining optimal husbandry conditions and comfort for the animals used in this study. We thank the Maria Kasper Lab, Karolinska Institutet, Sweden for their advice on single cell skin dissociation. We thank Dr Jean Rodgers for the work conducted under her Home Office Animal License (PPL No. PC8C3B25C). This work was funded by a Sir Henry Wellcome postdoctoral fellowship (221640/Z/20/Z to JFQ), a Wellcome Trust FutureScope grant (104111/Z/14/Z Wellcome Centre for Integrative Parasitology to JFQ), and a Wellcome Trust Senior Research fellow (209511/Z/17/Z to AML). SBC is supported by a grant from the Annie McNab Bequest (CRUK Beatson Institute), Breast Cancer Now (2018JulPR1101, 2019DecPhD1349, 2019DecPR1424), Cancer Research Institute (CLIP award), Cancer Research UK (RCCCEA- Nov21100003, EDDPGM- Nov21100001, DRCNPG- Jun22\100007), and Pancreatic Cancer Research and Worldwide Cancer Research (22- 0135). PC and MCS are supported by a Wellcome Trust Senior Research fellowship (209511/Z/17/Z) awarded to AML. RH is a Wellcome Trust PhD student (Wellcome Trust 218518/Z/19/Z). The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. + +<|ref|>sub_title<|/ref|><|det|>[[118, 608, 321, 625]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 638, 908, 806]]<|/det|> +Conceptualisation: JFQ, MCS. Methodology: JFQ, MCS, PC, RH, JO, BC, AL, AC, NK, SBC, AML. Formal analysis: JFQ, MCS, PC, RH, JO, BC, AL. Writing - original draft: JFQ, MCS. Writing - reviewing and editing: JFQ, MCS, PC, RH, JO, BC, AL, AC, SBC, NK, AML. Funding acquisition: JFQ, AML. The authors declare that they have no competing interests, commercial or otherwise. Correspondence and requests for materials should be addressed to Annette MacLeod (Annette.macleod@glasgow.ac.uk) or Juan F. Quintana (juan.quintana@glasgow.ac.uk) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 195, 102]]<|/det|> +## Figures + +<|ref|>text<|/ref|><|det|>[[115, 108, 907, 598]]<|/det|> +Figure 1. Integrative overview of the murine skin infected with T. brucei infection using single cell transcriptomics. A) H&E- stained images from naive and infected mice. Scale bar: \(800 \mu \mathrm{m}\) (top) and \(100 \mu \mathrm{m}\) (bottom). B) Immunohistochemistry against the stumpy- specific marker PAD1 in naive and infected samples, including insets highlighting the presence of stumpy forms in both the epidermis- subcutaneous adipose tissue and the subcutaneous adipose tissue- adipose tissue. Scale bar: \(50 \mu \mathrm{m}\) . C) Uniform manifold approximation and projection (UMAP) of 56,876 high- quality cells from both naive and infected samples, highlighting the major cell types detected in our dataset, including stromal cells (Keratinocytes, fibroblasts, endothelial cells, and adipocytes) and immune cells (myeloid cells, T cells, Langerhans cells, and macrophages). D) Top 10 marker genes defining all the major cell clusters detected in (C). The heatmap is colour- coded based on gene expression intensity. E) Spatial feature plot for selected marker genes defining keratinocytes (Krt15), fibroblast- like mesenchymal cells (Col1a1), adipocytes (Pparg), T cells (Cd3g), myeloid cells (Lyz2), and Langerhans cells (Cd207). The corresponding histological section is included on the left, including an annotation of epidermis (Ep), dermis (D), adipose tissue (Ad), Muscle (M), and subcutaneous adipose tissue (Sc). F) As in (C) but splitting the UMAP plot into naïve samples (25,800 cells; \(n = 2\) mice) and infected (31,076 cells; \(n = 2\) ). KC: keratinocytes, FB: Fibroblasts, EC: endothelial cells, MCs: Macrophages, LC: Langerhans cells, Adipo: Adipocytes, Erythro: Erythrocytes. + +<|ref|>text<|/ref|><|det|>[[115, 602, 907, 893]]<|/det|> +Figure 2. The murine skin stromal cells respond to infection by upregulating genes associated with antigen presentation and chemotaxis. A) Uniform manifold approximation and projection (UMAP) of 52,149 high- quality cells within the stromal subcluster, encompassing seven keratinocyte clusters (KC1- 7), fibroblasts (FB), endothelial cells (ECs), melanocytes (MLCs), and interstitial preadipocytes (IPAs). B) Dot plot representing the expression levels of top marker genes used to catalogue the diversity of skin stromal cells. The side of the dots represent the percentage of cells that express a given marker, and the colour intensity represent the level of expression. C) Frequency plot the different stromal cell types detected in the murine skin in naïve (\(n = 2\)) and infected (\(n = 2\)) samples. Module scoring for the overall expression of inflammatory cytokines (D) and genes associated with antigen presentation (E). Violin plot showing the expression level of several significant adipocyte- specific cytokines (F), chemokines (G), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 905, 152]]<|/det|> +and major histocompatibility complex (MHC) I and II molecules (H). I) Spatial module scoring for inflammation and antigen presentation in a naïve (top) and infected (bottom) skin section. Scale bar, \(100 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[115, 158, 907, 473]]<|/det|> +Figure 3. The murine skin is colonised by a myriad of myeloid cells during chronic T. brucei infection. A) Uniform manifold approximation and projection (UMAP) of 2,353 high- quality cells within the myeloid cluster were re- analysed to identify a total four subclusters, including dendritic cells (DCs), Mast cells, and two populations of macrophages \((Mrc1^{+} \text{mas}\) and \(Cd14^{+}\) mono), and two populations of Langerhans cells (LCs 1 and LCs 2). B) Dot plot representing the expression levels of top marker genes used to catalogue the diversity of myeloid cells. C) Spatial feature plot depicting the expression levels of markers defining \(Cd14^{+}\) monocytes \((Cd14, Ccr2)\) , \(Mrc1^{+}\) macrophages \((Mrc1, Lyve1)\) , mast cells \((Hdc)\) , and DCs \((Xcr1)\) . The corresponding histological section is included on the left, including an annotation of epidermis (Ep), dermis (D), adipose tissue (Ad), Muscle (M), and subcutaneous adipose tissue (Sc). Module scoring for the overall expression of inflammatory cytokines (D) and genes associated with antigen presentation (E). + +<|ref|>text<|/ref|><|det|>[[115, 479, 907, 899]]<|/det|> +Figure 4. Chronic T. brucei infection triggers the activation of skin- dwelling Vγ6 γδT cells localised in the subcutaneous adipose tissue. A) Uniform manifold approximation and projection (UMAP) of 1,043 high- quality T cells from naïve (304 cells) and infected skin samples (739 cells). After reclustering, we detected a total of five subclusters including type 2 innate lymphoid cells (ILC2s), CD4+ T cells, NK cells, and γδT cells. B) Dot Plot depicting the expression level of top marker genes for the skin T cell subcluters. The dot size and the intensity of the colour represents the proportion of cells expressing the genes and the level of expression. C) Top panel: Representative flow cytometry analysis to determine the presence of CD27- (IL-17- producing) and CD27+ (IFNγ- producing) γδT cells in murine skin infected \((n = 5 \text{mice})\) with T. brucei and naïve controls \((n = 4 \text{mice})\) . Bottom panel: Quantification of flow cytometry data. Statistical analysis was conducted using a parametric T test. A \(p\) value \(< 0.05\) was considered significant. D) Dot Plot depicting the expression level of top genes associated with T cell activation and TCR engagement for the skin T cell subcluters. The dot size and the intensity of the colour represents the proportion of cells expressing the genes and the level of expression. E) Spatial feature plot of canonical Vγ6+ cell marker genes. F) Spatial module scoring subclusters in the murine naïve and infected skin tissue for canonical γδ + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 907, 225]]<|/det|> +T cell marker genes. G) In silico cell- cell interaction analysis between adipocytes ("senders") and T cells ("receivers") based on the upregulation of ligand-receptor pairs. The heatmap is colour coded to represent the strength of the interaction. H) Spatial feature plot depicting the expression of adipose- derived ligand Tnfsf18 and T- cell specific receptorTnfsf18, which is predicted to be one of the most robust interactions between these cell types during infection. + +<|ref|>text<|/ref|><|det|>[[115, 231, 907, 700]]<|/det|> +Figure 5. \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells are essential for controlling skin inflammation and subcutaneous adipose tissue wasting independently of skin- resident \(\mathsf{T}_{\mathsf{H}}1\) T cells. A) In silico cell- cell interaction analysis between \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells ("senders") and \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes ("receivers") based on the upregulation of ligand- receptor pairs. The heatmap is colour coded to represent the strength of the interaction. B) Expression level of Clcf1 and \(A\mathsf{r}\mathsf{e}\mathsf{g}\) , two of the most significant upregulated \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells- derived ligands predicted to interact with subcutaneous adipocytes. C) H&E staining from skin biopsies obtained from FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) naive and infected mice. Scale bar: \(100\mu \mathrm{m}\) . Ep, epidermis; D, dermis; Ad, adipose tissue; M, muscle; Sc, subcutis. D) Analysis of mean adipocyte area \((\mu \mathsf{m}^2)\) in naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. \(N = 5\) biological replicates per group, from two independent experiments. Lipid droplets were measured from 3 distinct areas in each image and then combined for each biological replicate. A non- parametric, one- way ANOVA was used to determine the level of significant. A \(p\) value \(< 0.05\) is considered significant. E) Frequency plot of the adipocyte area represented in (D) for naive and infected FVB/N (left panel) and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice (right panel). \(N = 5\) biological replicates per group, from two independent experiments. Lipid droplets were measured from 3 distinct areas in each image and then combined for each biological replicate. A non- parametric, one- way ANOVA was used to determine the level of significant. A \(p\) value \(< 0.05\) is considered significant. + +<|ref|>text<|/ref|><|det|>[[115, 706, 907, 902]]<|/det|> +Figure 6. Proposed model of stromal- immune interactions in the skin during T. brucei infection. Based on our spatially- resolved single cell atlas, we propose a model whereby \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells act concertedly with \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes (and potentially preadipocytes) to coordinate local immune responses, either via the recruitment of immune cells (e.g., CD8 \(^+\) T cells and NK cells). The \(P\mathsf{p}\mathsf{a}\mathsf{r}\mathsf{g}^{+}\) adipocytes in this context provide important cues for T cell activation that we hypothesise might be involved in triggering \(\mathsf{V}\gamma \mathsf{6}^{+}\) cells - mediated responses. Other stromal cells such as Langerhans cells and keratinocytes are also likely to be involved in this process via antigenic presentation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 109, 346, 128]]<|/det|> +## Supplementary Figures + +<|ref|>text<|/ref|><|det|>[[115, 132, 907, 275]]<|/det|> +Supplementary figure 1. Quality control measurements of the murine single cell transcriptomics dataset. A) Number of Unique molecular identifies (UMIs), genes, mitochondrial reads, and library complexity (Log10 UMIs/gene) before (left panel) and after (right panel) applying filtering parameters. B) Clustree output representing the relationship between different cell clusters at various levels of resolution using the function FindClusters. + +<|ref|>text<|/ref|><|det|>[[115, 280, 907, 372]]<|/det|> +Supplementary figure 2. Quality control of 10X Visium datasets from the mouse skin over the course of infection with T. brucei. Spatial clusters and marker genes for each spatial cluster in the naive (A) and infected (B) murine skin using 10X Visium spatial transcriptomics. + +<|ref|>text<|/ref|><|det|>[[115, 378, 908, 673]]<|/det|> +Supplementary figure 3. Characterisation of the \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice during T. brucei infection. A) Measurement of circulating parasitaemia in both female FVB/N \((n = 4)\) and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice \((n = 4)\) for a period of 22 days. ANOVA test with multiple corrections. A \(p\) value \(< 0.05\) is considered significant. B) qRT- PCR of skin- dwelling trypanosomes by measuring the trypanosome- specific gene Prf2 by qRT- PCR. The Prf2 copy numbers were normalised per ng of skin DNA from naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). \(T\) test comparison between infected samples. A p value \(< 0.05\) is considered significant. The detection limit, as measured in skin biopsies from naive controls, is also indicated with a dotted line. C) Representative immunohistochemistry of murine skin biopsies from naive and infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice to detect the presence of T. brucei using an antibody against the trypanosome- specific luminal binding protein 1 (BiP). Scale bar \(= 50\mu \mathrm{m}\) + +<|ref|>text<|/ref|><|det|>[[115, 680, 907, 901]]<|/det|> +Supplementary figure 4. Single molecule fluorescent in situ hybridisation (smFISH) analysis showing the spatial distribution of T. brucei developmental stages in the skin of FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. Top panels: skin sections from naive FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) - mice. Middle panels: skin sections from infected FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice depicting the expression of the T. brucei- specific transcripts Gapdh and Pyk1 (slender specific markers) and Pad2 (stumpy specific marker). The dotted square indicates an area selected for magnification. Bottom panels: Magnified fields of the infected samples showing the distribution of both slender and stumpy markers in the skin of FVB/N and \(\mathsf{V}\gamma 4 / 6^{- / - }\) mice. Scale bar: \(50\mu \mathrm{m}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 905, 123]]<|/det|> +Supplementary figure 5. Flow cytometry analysis of skin- resident T lymphocytes. Gating strategy for the identification of skin \(\gamma \delta\) T cells (A) and \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells (B). + +<|ref|>text<|/ref|><|det|>[[115, 130, 907, 410]]<|/det|> +Supplementary figure 6. Quantification of skin- resident lymphocytes in the \(\mathrm{V}\gamma 4 / 6^{+}\) mice during T. brucei infection. A) Representative flow cytometry analysis of skin \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). B) Quantification of the frequency of skin \(\mathrm{CD4^{+}}\) (left panel) and \(\mathrm{CD8^{+}}\) T cells (right panel) in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group) as shown in (C). ANOVA test with multiple corrections. A \(p\) value \(< 0.05\) is considered significant. C) Representative flow cytometry analysis of ex vivo recall assay to determine the production of IFN \(\gamma\) in skin- resident \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group). D) Quantification of the frequency of skin \(\mathrm{CD4^{+}}\) (top panel) and \(\mathrm{CD8^{+}}\) T cells (bottom panel) in naive and infected FVB/N and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice \((n = 4\) mice/group) as shown in (C). + +<|ref|>sub_title<|/ref|><|det|>[[118, 416, 245, 433]]<|/det|> +## Table legend + +<|ref|>text<|/ref|><|det|>[[115, 439, 907, 629]]<|/det|> +Supplementary table 1. Overview of the mouse skin single cell transcriptomics during chronic T. brucei infection. S1A) Quality control including mean reads per cell and median genes per cell before and after filtering out low quality cell types. S1B) Overview of the major cell types detected in the single cell dataset at a resolution of 0.4. The marker genes are also included. S1C) Overview of the stromal cells detected in the skin dataset at a resolution of 0.3. The marker genes for these clusters, as well as representative UMAP plots are also included. S1D) As in S1C, but for the myeloid cells at a resolution of 0.3. S1E) As in S1C, but for the T cells at a resolution of 0.3. + +<|ref|>text<|/ref|><|det|>[[115, 634, 907, 777]]<|/det|> +Supplementary table 2. Overview of the spatial transcriptomics of the mouse skin during chronic T. brucei infection. S2A) Overview of the spatial transcriptomics project, including total number of reads sequenced per biological replicate, the median number of genes per spot and the percentage of mappable reads to the mouse genome (mm10). S2B) Mouse marker genes identified in the 10X Visium spatial transcriptomics datasets. + +<|ref|>text<|/ref|><|det|>[[115, 782, 907, 902]]<|/det|> +Supplementary table 3. Histopathological analysis of biopsies taken from naive and infected FVB/NJ and \(\mathrm{V}\gamma 4 / 6^{- / - }\) mice. The skin biopsies were harvested at 21 days post- infection \((n = 4\) mice/group), fixed in \(10\%\) PFA and counterstained with the T. brucei- specific antibody TbBiP. Uninfected animals \((n = 4)\) were included as naive controls. Histological examinations were scored using H&E staining and BiP staining and was + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 905, 201]]<|/det|> +conducted double-blinded. The results reported in this table are in comparison to naïve controls for the corresponding genetic background, and encompass a detailed analysis of the epidermis, dermis, hypodermis, skeletal muscle, and the subcutaneous adipose tissue. The column labelled "BiP" (columns AD to Al) represents the results from the immunohistochemistry analysis against the anti-Trypanosoma anti-BiP antibody. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 247, 103]]<|/det|> +## Bibliography + +<|ref|>text<|/ref|><|det|>[[115, 108, 904, 193]]<|/det|> +1. Taylor, K. R., Costanzo, A. E. & Jameson, J. M. Dysfunctional \(\gamma \delta\) T cells contribute to impaired keratinocyte homeostasis in mouse models of obesity. J Invest Dermatol 131, 2409-2418 (2011). + +<|ref|>text<|/ref|><|det|>[[115, 205, 884, 258]]<|/det|> +2. Jameson, J. et al. A role for skin gammadelta T cells in wound repair. Science 296, 747-749 (2002). + +<|ref|>text<|/ref|><|det|>[[115, 270, 893, 355]]<|/det|> +3. Chen, Y., Chou, K., Fuchs, E., Havran, W. L. & Boismenu, R. Protection of the intestinal mucosa by intraepithelial \(\gamma \delta\) T cells. Proceedings of the National Academy of Sciences 99, 14338-14343 (2002). + +<|ref|>text<|/ref|><|det|>[[115, 368, 904, 454]]<|/det|> +4. Sharp, L. L., Jameson, J. M., Cauvi, G. & Havran, W. L. Dendritic epidermal T cells regulate skin homeostasis through local production of insulin-like growth factor 1. Nat Immunol 6, 73-79 (2005). + +<|ref|>text<|/ref|><|det|>[[115, 467, 901, 521]]<|/det|> +5. Roark, C. L., Simonian, P. L., Fontenot, A. P., Born, W. K. & O'Brien, R. L. \(\gamma \delta\) T cells: an important source of IL-17. Curr Opin Immunol 20, 353-357 (2008). + +<|ref|>text<|/ref|><|det|>[[115, 533, 860, 586]]<|/det|> +6. Simonian, P. L. et al. \(\gamma \delta\) T cells protect against lung fibrosis via IL-22. Journal of Experimental Medicine 207, 2239-2253 (2010). + +<|ref|>text<|/ref|><|det|>[[115, 599, 884, 714]]<|/det|> +7. Schilbach, K., Frommer, K., Meier, S., Handgretinger, R. & Eyrich, M. Immune response of human propagated gammadelta-T-cells to neuroblastoma recommend the Vdelta1+ subset for gammadelta-T-cell-based immunotherapy. J Immunol 31, 896-905 (2008). + +<|ref|>text<|/ref|><|det|>[[115, 727, 897, 815]]<|/det|> +8. Tan, L. et al. Single-Cell Transcriptomics Identifies the Adaptation of Scart1+ Vγ6+ T Cells to Skin Residency as Activated Effector Cells. Cell Rep 27, 3657-3671.e4 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 828, 888, 881]]<|/det|> +9. Konieczny, P. et al. Interleukin-17 governs hypoxic adaptation of injured epithelium. Science 377, eabg9302 (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 83, 900, 163]]<|/det|> +10. Zaharie, R. D., Popa, C., Schlanger, D., Vălean, D. & Zaharie, F. The Role of IL-22 in Wound Healing. Potential Implications in Clinical Practice. Int J Mol Sci 23, 3693 (2022). + +<|ref|>text<|/ref|><|det|>[[50, 179, 884, 232]]<|/det|> +11. Enhancement of Wound Healing by Topical Treatment with Epidermal Growth Factor | NEJM. https://www.nejm.org/doi/full/10.1056/NEJM198907133210203. + +<|ref|>text<|/ref|><|det|>[[50, 247, 884, 360]]<|/det|> +12. Zaiss, D. M. W. et al. Amphiregulin enhances regulatory T cell-suppressive function via the epidermal growth factor receptor. Immunity 38, 275-284 (2013). +13. Patil, R. S. et al. IL17 producing γδT cells induce angiogenesis and are associated with poor survival in gallbladder cancer patients. Int J Cancer 139, 869-881 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 360, 900, 414]]<|/det|> +14. Wu, X. et al. IL-17 promotes tumor angiogenesis through Stat3 pathway mediated upregulation of VEGF in gastric cancer. Tumor Biol. 37, 5493-5501 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 429, 900, 483]]<|/det|> +15. Yan, J. & Huang, J. Innate γδT17 cells convert cancer-elicited inflammation into immunosuppression through myeloid-derived suppressor cells. Oncoimmunology 3, e953423 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 478, 895, 561]]<|/det|> +16. Pan, B. et al. Interleukin-17 promotes angiogenesis by stimulating VEGF production of cancer cells via the STAT3/GIV signaling pathway in non-small-cell lung cancer. Sci Rep 5, 16053 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 576, 866, 660]]<|/det|> +17. Cai, Y. et al. Differential Developmental Requirement and Peripheral Regulation for Dermal Vγ4 and Vγ6T17 Cells in Health and Inflammation. Nat Commun 5, 3986 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 674, 895, 757]]<|/det|> +18. LUNARDI, C., MARGUERIE, C., BOWNESS, P., WALPORT, M. J. & SO, A. K. Reduction in Tγδ cell numbers and alteration in subset distribution in systemic lupus erythematosus. Clinical and Experimental Immunology 86, 203-206 (1991). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 84, 888, 145]]<|/det|> +19. Eiras, D. P., Kirkman, L. A. & Murray, H. W. Cutaneous Leishmaniasis: Current Treatment Practices in the USA for Returning Travelers. Curr Treat Options Infect Dis 7, 52–62 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 183, 884, 266]]<|/det|> +20. Ferraresso, M. G., Torre, A. C., Piva, M. M. M. & Barcan, L. Chagas disease reactivation: cutaneous manifestations in a transplanted patient. An Bras Dermatol 93, 890–892 (2018). + +<|ref|>text<|/ref|><|det|>[[50, 281, 888, 333]]<|/det|> +21. Capewell, P. et al. The skin is a significant but overlooked anatomical reservoir for vector-borne African trypanosomes. eLife 5, e17716 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 347, 893, 430]]<|/det|> +22. Camara, M. et al. Extravascular Dermal Trypanosomes in Suspected and Confirmed Cases of gambiense Human African Trypanosomiasis. Clinical Infectious Diseases 73, 12–20 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 445, 845, 496]]<|/det|> +23. De Niz, M. et al. Intravital imaging of host–parasite interactions in skin and adipose tissues. Cellular Microbiology 21, e13023 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 510, 900, 593]]<|/det|> +24. Sinton, M. C. et al. Interleukin-17 drives sex-dependent weight loss and changes in feeding behaviour during Trypanosoma brucei infection. 2022.09.23.509158 Preprint at https://doi.org/10.1101/2022.09.23.509158 (2022). + +<|ref|>text<|/ref|><|det|>[[50, 608, 895, 692]]<|/det|> +25. Le Ray, D., Barry, J. D., Easton, C. & Vickerman, K. First tseste fly transmission of the 'AnTat' serodeme of Trypanosoma brucei. Ann Soc Belg Med Trop 57, 369–381 (1977). + +<|ref|>text<|/ref|><|det|>[[50, 707, 813, 758]]<|/det|> +26. Zappia, L. & Oshlack, A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. GigaScience 7, (2018). + +<|ref|>text<|/ref|><|det|>[[50, 773, 878, 824]]<|/det|> +27. Bankhead, P. et al. QuPath: Open source software for digital pathology image analysis. Sci Rep 7, 16878 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 839, 876, 889]]<|/det|> +28. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676–682 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 83, 905, 144]]<|/det|> +1051 29. Galarraga, M. et al. Adiposoft: automated software for the analysis of white adipose tissue cellularity in histological sections. Journal of Lipid Research 53, 2791–2796 (2012). + +<|ref|>text<|/ref|><|det|>[[50, 150, 900, 240]]<|/det|> +1053 30. Girard, A. et al. Raman spectroscopic analysis of skin as a diagnostic tool for Human African Trypanosomiasis. PLOS Pathogens 17, e1010060 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 247, 884, 366]]<|/det|> +1056 31. Castro-Manrreza, M. E. et al. Mesenchymal Stromal Cells from the Epidermis and Dermis of Psoriasis Patients: Morphology, Immunophenotype, Differentiation Patterns, and Regulation of T Cell Proliferation. Stem Cells Int 2019, 4541797 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 375, 896, 465]]<|/det|> +1059 32. Stefkovich, M., Traynor, S., Cheng, L., Merrick, D. & Seale, P. Dpp4+ interstitial progenitor cells contribute to basal and high fat diet-induced adipogenesis. Mol Metab 54, 101357 (2021). + +<|ref|>text<|/ref|><|det|>[[50, 475, 905, 532]]<|/det|> +1063 33. Trindade, S. et al. Trypanosoma brucei Parasites Occupy and Functionally Adapt to the Adipose Tissue in Mice. Cell Host & Microbe 19, 837–848 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 541, 905, 597]]<|/det|> +1065 34. Baptista, A. P. et al. Lymph node stromal cells constrain immunity via MHC class II self-antigen presentation. eLife 3, e04433 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 606, 904, 662]]<|/det|> +1067 35. Edwards, S. C. et al. PD-1 and TIM-3 differentially regulate subsets of mouse IL-17A–producing γδ T cells. Journal of Experimental Medicine 220, e20211431 (2022). + +<|ref|>text<|/ref|><|det|>[[50, 671, 900, 728]]<|/det|> +1069 36. Kohlgruber, A. C. et al. γδ T cells producing interleukin-17A regulate adipose regulatory T cell homeostasis and thermogenesis. Nat Immunol 19, 464–474 (2018). + +<|ref|>text<|/ref|><|det|>[[50, 737, 832, 793]]<|/det|> +1071 37. Hu, B. et al. γδ T cells and adipocyte IL-17RC control fat innervation and thermogenesis. Nature 578, 610–614 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 803, 888, 858]]<|/det|> +1073 38. Dumas, S. N., Guo, C., Kim, J. K., Friedline, R. H. & Ntambi, J. M. Interleukin-6 derived from cutaneous deficiency of stearoyl-CoA desaturase- 1 may mediate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 84, 884, 135]]<|/det|> +1075 metabolic organ crosstalk among skin, adipose tissue and liver. Biochem Biophys Res Commun 508, 87–91 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 150, 900, 202]]<|/det|> +1077 39. Sawant, K. V. et al. Chemokine CXCL1 mediated neutrophil recruitment: Role of glycosaminoglycan interactions. Sci Rep 6, 33123 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 216, 900, 333]]<|/det|> +1079 40. Ochiai, E. et al. CXCL9 Is Important for Recruiting Immune T Cells into the Brain and Inducing an Accumulation of the T Cells to the Areas of Tachyzoite Proliferation to Prevent Reactivation of Chronic Cerebral Infection with Toxoplasma gondii. Am J Pathol 185, 314–324 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 347, 896, 430]]<|/det|> +1083 41. Ajuebor, M. N. et al. GammadeltaT cells initiate acute inflammation and injury in adenovirus-infected liver via cytokine-chemokine cross talk. J Virol 82, 9564–9576 (2008). + +<|ref|>text<|/ref|><|det|>[[50, 445, 900, 528]]<|/det|> +1086 42. Saudemont, A., Jouy, N., Hetuin, D. & Quesnel, B. NK cells that are activated by CXCL10 can kill dormant tumor cells that resist CTL-mediated lysis and can express B7-H1 that stimulates T cells. Blood 105, 2428–2435 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 542, 896, 626]]<|/det|> +1089 43. Wu, H. et al. T-Cell Accumulation and Regulated on Activation, Normal T Cell Expressed and Secreted Upregulation in Adipose Tissue in Obesity. Circulation 115, 1029–1038 (2007). + +<|ref|>text<|/ref|><|det|>[[50, 640, 888, 692]]<|/det|> +1092 44. Islam, S. A. et al. Mouse CCL8, a CCR8 agonist, promotes atopic dermatitis by recruiting IL-5+ TH2 cells. Nat Immunol 12, 10.1038/ni.1984 (2011). + +<|ref|>text<|/ref|><|det|>[[50, 706, 904, 791]]<|/det|> +1094 45. Acosta, J. R. et al. Human-Specific Function of IL-10 in Adipose Tissue Linked to Insulin Resistance. The Journal of Clinical Endocrinology & Metabolism 104, 4552–4562 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 805, 900, 857]]<|/det|> +1097 46. Rajbhandari, P. et al. IL-10 Signaling Remodels Adipose Chromatin Architecture to Limit Thermogenesis and Energy Expenditure. Cell 172, 218-233.e17 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 80, 900, 860]]<|/det|> +1099 47. Akdis, C. A. & Blaser, K. Mechanisms of interleukin-10-mediated immune suppression. Immunology 103, 131–136 (2001).1101 48. van Hall, G. et al. Interleukin-6 stimulates lipolysis and fat oxidation in humans. J Clin Endocrinol Metab 88, 3005–3010 (2003).1103 49. Adedeji, T. A. et al. Serum Interleukin-6 and Weight Loss in Antiretroviral-naïve and Antiretroviral-treated Patients with HIV/AIDS: Relationships and Predictors. Curr HIV Res 20, 441–456 (2022).1106 50. Han, J. et al. Plasma concentration of interleukin-6 was upregulated in cancer cachexia patients and was positively correlated with plasma free fatty acid in female patients. Nutrition & Metabolism 16, 80 (2019).1109 51. Haas, J. D. et al. Development of Interleukin-17-Producing γδ T Cells Is Restricted to a Functional Embryonic Wave. Immunity 37, 48–59 (2012).1111 52. Spidale, N. A. et al. Interleukin-17-Producing γδ T Cells Originate from SOX13+ Progenitors that Are Independent of γδTCR Signaling. Immunity 49, 857-872.e5 (2018).1113 53. Fink, D. R. et al. Elevated numbers of SCART1+ γδ T cells in skin inflammation and inflammatory bowel disease. Molecular Immunology 47, 1710–1718 (2010).1116 54. Audemard-Verger, A. et al. Macrophages Induce Long-Term Trapping of γδ T Cells with Innate-like Properties within Secondary Lymphoid Organs in the Steady State. The Journal of Immunology 199, 1998–2007 (2017).1119 55. Romagnoli, P. A., Sheridan, B. S., Pham, Q.-M., Lefrançois, L. & Khanna, K. M. IL-17A–producing resident memory γδ T cells orchestrate the innate immune response to secondary oral Listeria monocytogenes infection. Proceedings of the National Academy of Sciences 113, 8502–8507 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[45, 81, 900, 380]]<|/det|> +56. Ronchetti, S. et al. Glucocorticoid-induced TNFR-related protein lowers the threshold of CD28 costimulation in CD8+ T cells. J Immunol 179, 5916–5926 (2007).57. Huang, Y. et al. γδ T cells affect IL-4 production and B-cell tolerance.Proceedings of the National Academy of Sciences 112, E39–E48 (2015).58. Coffelt, S. B. et al. IL-17-producing γδ T cells and neutrophils conspire to promote breast cancer metastasis. Nature 522, 345–348 (2015).59. Machado, H., Hofer, P., Zechner, R. & Figueiredo, L. M. Adipocyte lipolysis protects the host against Trypanosoma brucei infection. 2022.11.05.515274 Preprint at https://doi.org/10.1101/2022.11.05.515274 (2022). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[20, 0, 485, 170]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[20, 210, 485, 440]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[20, 476, 485, 682]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[20, 707, 485, 998]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[500, 0, 970, 170]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[500, 200, 970, 440]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[500, 476, 970, 682]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[500, 707, 970, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[5, 0, 990, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[17, 0, 220, 120]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[17, 12, 45, 25]]<|/det|> +
A) Naive (304 cells)
+ +<|ref|>image<|/ref|><|det|>[[17, 28, 220, 170]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[17, 183, 201, 196]]<|/det|> +
C) Pregated on γδTCR+ T cells:
+ +<|ref|>image<|/ref|><|det|>[[17, 199, 220, 360]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[17, 364, 220, 515]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[17, 519, 220, 620]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[17, 624, 220, 725]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[17, 729, 220, 830]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[17, 834, 220, 985]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 0, 817, 200]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 205, 817, 515]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 519, 817, 725]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 729, 817, 985]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 519, 817, 725]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[250, 729, 817, 985]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[15, 7, 40, 25]]<|/det|> +A) + +<|ref|>table<|/ref|><|det|>[[15, 40, 560, 336]]<|/det|> +
Vγ6+ cellsNaiveInfected
Cld1
Cd24a
H2-T23
Ifnk
Pcdh7
Pgf
Spn
Wnt4
Ptin
Cntn1
Ltb
Comp
Hmgb2
Hmgb1
Ttf
Clstn1
Ptprf
Areg
Ptprc
Icam5
+ +<|ref|>text<|/ref|><|det|>[[15, 384, 40, 401]]<|/det|> +C) + +<|ref|>image<|/ref|><|det|>[[15, 402, 590, 750]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[15, 760, 590, 990]]<|/det|> + + +<|ref|>sub_title<|/ref|><|det|>[[15, 755, 40, 773]]<|/det|> +E) + +<|ref|>image<|/ref|><|det|>[[15, 774, 590, 990]]<|/det|> + + +<|ref|>sub_title<|/ref|><|det|>[[15, 985, 40, 998]]<|/det|> +F) + +<|ref|>sub_title<|/ref|><|det|>[[615, 7, 639, 25]]<|/det|> +B) + +<|ref|>image<|/ref|><|det|>[[615, 7, 997, 336]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[615, 336, 997, 750]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[615, 774, 997, 990]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[15, 0, 978, 978]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[37, 9, 485, 480]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[37, 500, 485, 760]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[37, 780, 485, 998]]<|/det|> + + +<|ref|>image<|/ref|><|det|>[[511, 9, 970, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 310, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 312, 230]]<|/det|> +NCOMMS2309302RS.pdf Supplementarytable1. xlsx Supplementarytable2. xlsx Supplementarytable3. xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/images_list.json b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..b03a930a79190ae14d0fd6391254f1d1ce77c367 --- /dev/null +++ b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. Overall structures of the GHSR-G1 complexes. (a) and (b) Cryo-EM density maps of the complex with ghrelin (orange) and ibutamoren (yellow) and overall structures. GHSR bound to ghrelin and ibutamoren is colored in blue and green, respectively. \\(G_{\\mathrm{ai}}\\) , \\(G_{\\beta}\\) and \\(G_{\\gamma}\\) subunits are colored in cyan, pink and light blue, respectively. ScFv16 is colored in grey. The density maps of the two agonists are shown in light grey. The octanoyl group in ghrelin is circled.", + "footnote": [], + "bbox": [ + [ + 201, + 95, + 780, + 641 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Binding pockets for ghrelin and ibutamoren. (a) Ghrelin binding in the cavity I and II. Ghrelin residues are labeled in orange. (b) Binding of the octanoyl group of ghrelin in the cavity II. (c) Dose-dependent action of ghrelin on the wild type GHSR (wtGHSR) and mutants. (d) Alignment of ghrelin and ibutamoren. (e) Ibutamoren binding pocket. (f) Dose-dependent action of ibutamoren on wtGHSR and mutants. In c and f, agonist-induced GHSR signaling was measured by \\(\\mathrm{Ca}^{2 + }\\) mobilization assay. Each data point represents \\(\\mathrm{Mean} \\pm \\mathrm{S.D.}\\) from 3 independent assays. Polar interactions are shown as dashed lines in a, b and e.", + "footnote": [], + "bbox": [ + [ + 112, + 92, + 879, + 444 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Active conformation of GHSR. Conformational changes of the extracellular regions of TM6 and TM7 (a), the highly conserved transmission switch residues (b), and cytoplasmic regions of TMs (c) in the active GHSR relative to them in the inactive GHSR. Two active GHSR bound to ghrelin and butamoren and the inactive GHSR bound to C12 are colored in blue, green and purple, respectively. Ghrelin, butamoren and C12 are shown as orange, yellow and lemon sticks, respectively. Red arrows indicate changes of residues or regions from inactive to active states.", + "footnote": [], + "bbox": [ + [ + 118, + 88, + 884, + 240 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Conformational changes of critical motifs near the agonist-binding pocket. (a) Different conformations of the E124-R283 salt bridge in the active structures of GHSR with ghrelin and ibutamoren and the inactive structure of GHSR with C12. The salt bridge interactions are shown as dashed lines. (b) Rearrangement of the aromatic cluster residues W276, F279, H280 and F312 in the active and inactive GHSR.", + "footnote": [], + "bbox": [ + [ + 115, + 92, + 875, + 384 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. GHSR and \\(\\mathbf{G}_{\\mathrm{i}}\\) binding interface. (a) and (b) Detailed interactions between GHSR and \\(\\mathbf{G}_{\\mathrm{i}}\\) viewed from two angles. GHSR, \\(\\mathbf{G}_{\\mathrm{ai}}\\) , and \\(\\mathbf{G}_{\\beta}\\) are colored in blue, cyan and pink, respectively. Polar interactions are shown as dashed lines.", + "footnote": [], + "bbox": [ + [ + 121, + 88, + 876, + 317 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6. Structures of the ghrelin-GHSR-Gi complex in Conformer 1 and 2. (a) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 with the neurotensin-NTSR-Gi complex in the canonical state (PDB ID 6OS9). The Gi-coupling mode is highly similar. (b) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 and 2.", + "footnote": [], + "bbox": [ + [ + 120, + 90, + 863, + 363 + ] + ], + "page_idx": 15 + } +] \ No newline at end of file diff --git a/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9.mmd b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9.mmd new file mode 100644 index 0000000000000000000000000000000000000000..174afbf9869560bd7322aa0d479897f39fcd38b9 --- /dev/null +++ b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9.mmd @@ -0,0 +1,222 @@ + +# Structural basis of human ghrelin receptor signaling by ghrelin and the synthetic agonist ibuatomoren + +Cheng Zhang ( \(\circleddash\) chengzh@ptt.edu) University of Pittsburgh https://orcid.org/0000- 0001- 9042- 4007 + +Heng Liu University of Pittsburgh + +Dapeng Sun University of Pittsburgh https://orcid.org/0000- 0003- 0220- 8844 + +Alexander Myasnikov St. Jude Children's Research Hospital + +Marjorie Damian CNRS, Université de Montpellier, ENSCM + +Jean- Louis Baneres IBMM https://orcid.org/0000- 0001- 7078- 1285 + +Ji Sun St. Jude Children's Research Hospital https://orcid.org/0000- 0002- 9302- 3177 + +Article + +Keywords: ghrelin, hunger hormone, cell signaling + +Posted Date: July 12th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 646701/v1 + +License: © \(\circleddash\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on November 4th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26735- 5. + +<--- Page Split ---> + +# Structural basis of human ghrelin receptor signaling by ghrelin and the synthetic agonist ibuatomoren + +Heng Liu \(^{1}\) , Dapeng Sun \(^{1}\) , Alexander Myasnikov \(^{2}\) , Marjorie Damian \(^{3}\) , Jean- Louis Banaeres \(^{3}\) , Ji Sun \(^{2*}\) , Cheng Zhang \(^{1*}\) + +\(^{1}\) Department of Pharmacology and Chemical Biology, School of Medicine, University of Pittsburgh, Pittsburgh, PA15261, USA \(^{2}\) Department of Structural Biology, St. Jude Children's Research Hospital, Memphis, Tennessee, TN38120, USA \(^{3}\) Institut des Biomolécules Max Mousseron, CNRS, Université de Montpellier, ENSCM, Montpellier, France \(^{*}\) Correspondence: Dr. Ji Sun, Ji.Sun@stjude.org or Dr. Cheng Zhang, chengzh@pitt.edu. + +<--- Page Split ---> + +## Abstract + +The 'hunger hormone' ghrelin activates the ghrelin receptor GHSR to stimulate food intake and growth hormone secretion and regulate reward signaling. Acylation of ghrelin at Ser3 is required for its agonistic action on GHSR. Synthetic agonists of GHSR are under clinical evaluation for disorders related to appetite and growth hormone dysregulation. Here, we report high- resolution cryo- EM structures of the GHSR- G \(_i\) signaling complex with ghrelin and the non- peptide agonist butamoren as an investigational new drug. Our structures together with mutagenesis data reveal the molecular basis for the binding of ghrelin and butamoren. Structural comparison suggests a salt bridge and an aromatic cluster near the agonist- binding pocket as important structural motifs in receptor activation. Notable variations of the G \(_i\) binding mode are observed in our cryo- EM analysis, indicating highly dynamic nature of G \(_i\) - coupling to GHSR. Our results provide a framework for understanding GHSR signaling and developing new GHSR agonist drugs. + +<--- Page Split ---> + +## Introduction + +The human ghrelin system is a critical component of the gut- brain axis to regulate energy homeostasis and reward signaling \(^{1 - 3}\) . Ghrelin is a 28- amino acid orexigenic peptide hormone generated in the gut commonly regarded as the 'hunger hormone' or 'survival hormone' \(^{1 - 4}\) . It mainly signals through the growth hormone secretagogue receptor (GHSR, or ghrelin receptor), a class A G protein- coupled receptor (GPCR). Ghrelin signaling through GHSR to stimulate growth hormone secretion and food intake under nutritional or physiological challenges \(^{1,2}\) . In addition, GHSR plays critical roles in the modulation of stress and anxiety \(^{5,6}\) and the regulation of dopamine signaling and thus reward pathways in the CNS \(^{7,8}\) . Therefore, GHSR is a highly pursued drug target \(^{9}\) . A GHSR inverse agonist named PF- 5190457 is under clinical investigation to treat alcohol use disorder \(^{10,11}\) . On the other hand, pharmacological stimulation of GHSR by agonists represents an emerging and exciting avenue to address disorders related to appetite, gastric emptying and growth hormone dysregulation. Relamorelin and anamorelin are two synthetic peptide agonists of GHSR that are in different stages of clinical trials to treat diabetic gastroparesis and cancer- related anorexia- cachexia \(^{12 - 16}\) . Ibutamoren, also known as MK- 0677 or LUM- 201, is a synthetic small- molecule GHSR agonist used as a growth hormone secretagogue in many disease settings \(^{9,17 - 20}\) . + +The ghrelin- GHSR signaling system exhibits several unique properties compared to other GPCR signaling systems. Acylation of ghrelin, usually octanoylation at position- 3 serine, is required for its action on GHSR \(^{3}\) . Such a modification is achieved by the ghrelin O- acyl- transferase (GOAT) in vivo \(^{21 - 23}\) . Acylated ghrelin can activate GHSR to signal through multiple protein partners, including the \(\mathrm{G_{q / 11}}\) and \(\mathrm{G_{i / 0}}\) families of G proteins and \(\beta\) - arrestins \(^{24}\) . Several studies showed that different GHSR signaling pathways regulate distinct physiological functions \(^{25 - 27}\) . The GHSR and \(\mathrm{G_{i}}\) signaling pathway has been linked to attenuated glucose- induced insulin release by ghrelin \(^{25}\) , while the appetite stimulation is mainly mediated by \(\mathrm{G_{q / 11}}\) signaling \(^{26}\) . GHSR also exhibits a remarkably high constitutive activity, which is important for physiological growth hormone regulation \(^{28}\) . Mutations of the GHSR gene that result in loss of GHSR constitutive activity have been associated with the familial short stature syndrome \(^{29}\) . In addition, the constitutive activity of GHSR is negatively regulated by several endogenous mechanisms. Liver expressed antimicrobial peptide 2 (LEAP2) has been characterized as a GHSR inverse agonist or antagonist \(^{30}\) . Melanocortin receptor accessory protein 2 (MRAP2), a single- pass transmembrane protein, can directly associate with GHSR to modulate its activity \(^{3,31}\) . + +Due to the physiological and pathological significance of the ghrelin- GHSR system, intensive research effort has been devoted to the molecular understanding of ghrelin action and GHSR signaling. Biophysical studies and structural modeling were performed to investigate ghrelin recognition by GHSR \(^{32,33}\) . Recently, a crystal structure of highly engineered GHSR at an inactive conformation bound to an antagonist, Compound 12 (C12), has been reported to probe a possible ghrelin recognition mechanism \(^{34}\) . Yet, the molecular mechanism underlying GHSR signaling by diverse peptide and non- peptide agonists still remains elusive. Here, we report two cryo- electron + +<--- Page Split ---> + +microscopy (cryo- EM) structures of human GHSR in complex with \(\mathrm{G_i}\) and two agonists, the endogenous ligand ghrelin and the non- peptide synthetic agonist ibutamoren. Structural analysis together with mutagenesis data revealed mechanisms underlying agonism of different types of GHSR agonists and GHSR- \(\mathrm{G_i}\) coupling. + +## Results and Discussion + +## Structure determination of two GHSR- \(\mathbf{G}_{\mathrm{i}}\) complexes and overall structures + +We used wild type human GHSR and human \(\mathrm{G_{ai}}\) \(\mathrm{G_{b1}}\) and \(\mathrm{G_{v2}}\) to assemble the complexes with two agonists. No modification was introduced to the \(\mathrm{G_i}\) heterotrimer except for the amino- terminal 6xHis tag in \(\mathrm{G_{b1}}\) . We used apyrase to hydrolyze GDP in order to form stable nucleotide- free complexes. We also added an antibody fragment, scFv16, to further stabilize the \(\mathrm{G_i}\) heterotrimer \(^{35}\) . The structures of the GHSR- \(\mathrm{G_i}\) - scFv16 complexes with ghrelin and ibutamoren were both determined to an overall resolution of 2.7- Å (Fig. 1a and b, Extended Data Fig. S1). The cryo- EM maps allowed modeling of most regions of GHSR and \(\mathrm{G_i}\) heterotrimer, ghrelin residues Gly1- Val12 (residues in ghrelin are referred to by three- letter names, and residues in GHSR and other GPCRs are referred to by one- letter names hereafter) together with the octanoyl group, and the entire ibutamoren molecule (Fig. 1a and b, Extended Data Fig. S1). + +We also observed cryo- EM densities likely corresponding to cholesterol molecules bound to GHSR in both structures (Extended Data Fig. S2a). Further experiments showed that cholesterol could positively regulate the activity of GHSR in reconstituted lipid nanodiscs (Extended Data Fig. S2b), suggesting a potentially significant physiological role of cholesterol in ghrelin signaling. + +## Ghrelin and ibutamoren recognition + +In our structure, the N- terminal part of ghrelin adopts an extended conformation and inserts into a deep pocket of GHSR (Fig. 2a); such a binding mode is different from the predicted ghrelin binding mode based on the previous crystal structure \(^{34}\) . It was suggested that the salt bridge between E124 \(^{3,33}\) and R283 \(^{6,55}\) (superscripts represent Ballesteros- Weinstein numbering \(^{36}\) ) of GHSR divides the ghrelin- binding pocket into two cavities, cavity I and II \(^{34}\) . In our structures, we also observed this salt bridge. All the peptide moiety of ghrelin occupies the large cavity I while the octanoyl moiety attached to Ser3 of ghrelin extends into the small cavity II (Fig. 2a). It has been proposed that a crevasse region separating TM6 and TM7 in the inactive GHSR structure served as the recognition site of octanoyl moiety of ghrelin \(^{34}\) . However, our structure didn't support this hypothesis but instead showed that TM7 shifts towards TM6 compared to that in the inactive GHSR, closing the previously observed crevasse in the inactive structure (see discussion below). + +Ghrelin engages in extensive polar and hydrophobic interactions with GHSR in the cavity I. From the bottom region to the extracellular side, the side chains of ghrelin residues Gly1 and Glu8 form salt bridges with the side chains of E124 \(^{3,33}\) and R107 \(^{ECL1}\) of GHSR, respectively; the side chains of Ser2 and Arg11 and the main chain carbonyl of Phe4 of ghrelin are hydrogen- + +<--- Page Split ---> + +bonded by the main chain carbonyl groups of \(\mathrm{F290^{ECL2}}\) and \(\mathrm{S308^{7.38}}\) and the side chain of Y106 of GHSR, respectively (Fig. 2a). In addition to these polar interactions, in the middle region of the cavity II, Phe4 of ghrelin packs against GHSR residues \(\mathrm{L103^{2.64}}\) , \(\mathrm{L306^{7.36}}\) and \(\mathrm{F309^{7.39}}\) to form hydrophobic interactions (Fig. 2a). Leu5 of ghrelin also forms hydrophobic interactions with GHSR residues \(\mathrm{F286^{6.58}}\) and \(\mathrm{F290^{6.62}}\) (Fig. 2a). + +The octanoyl moiety of ghrelin mainly occupies the cavity II with an extended conformation pointing towards the cleft between TM4 and TM5 (Fig. 2b). We observed a strong density corresponding to the proximal half of the octanoyl moiety, while the density for the distal half was not well resolved, suggesting a flexible nature of this part (Fig. 1a). Of note, the cavity II exhibits an amphipathic environment (Fig. 2b). As a result, the \(\mathrm{C_8}\) fatty acid chain of the octanoyl moiety sits on top of an extensive polar interaction network at the bottom of the cavity II including \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) while being capped by hydrophobic GHSR residues from the upper region of the cavity II (Fig. 2b). The acyl group of the octanoyl moiety sticks towards and likely interacts with the side chain of \(\mathrm{Q120^{3.29}}\) of GHSR. Previous mutagenesis studies showed that individual mutations of \(\mathrm{E124^{3.33}}\) , \(\mathrm{R283^{6.55}}\) and \(\mathrm{Q120^{3.29}}\) to non- polar residues led to decreased potency of ghrelin \(^{34,37}\) . We also showed that mutations of hydrophobic residues in the cavity II to larger or polar residues, which potentially disrupt the binding of the octanoyl moiety, resulted in compromised action of ghrelin (Fig. 2c), supporting our structural insights. + +The non- peptide agonist ibuprofen also samples both cavity I and cavity II (Fig. 2d). It mainly occupies the bottom space of the ghrelin- binding pocket similar to the antagonist C12. The structures of ghrelin- and ibuprofen- bound GHSR share similar overall conformation with a root- mean- square deviation (RMSD) of the \(\mathrm{C\alpha}\) atoms at 1.23 Å. Subtle structural differences can be observed at the upper region of the binding pocket including ECL2 and 3 (Extended Data Fig. S3a), likely caused by the binding of Leu5- Val12 of ghrelin in this region. The three arms of ibuprofen point towards three different directions, mimicking the first four residues of ghrelin, Gly1- Phe4, including the octanoyl moiety (Fig. 2d). The phenyl group of ibuprofen sits in the cavity II and forms hydrophobic interactions with surrounding hydrophobic residues \(\mathrm{I178^{4.60}}\) , \(\mathrm{L181^{4.63}}\) and \(\mathrm{L210^{5.36}}\) of GHSR similarly to the octanoyl moiety of ghrelin (Fig. 2e). The other two arms of ibuprofen stick towards the bottom region of cavity II and the extracellular surface, respectively (Fig. 2e). Multiple hydrogen bonds are observed at the bottom of binding pocket between ibuprofen and GHSR residues \(\mathrm{D99^{2.60}}\) , \(\mathrm{Q120^{3.29}}\) , and \(\mathrm{R283^{6.55}}\) (Fig. 2e). Of note, the bottom region of cavity II exhibits a negatively charged environment to accommodate the smallest arm of ibuprofen with two amine groups (Extended Data Fig. S3b). The largest arm of ibuprofen containing an indoline- 3,4'-piperidine ring structure forms hydrophobic and cation- \(\pi\) interactions with the side chains of GHSR residues \(\mathrm{F286^{6.58}}\) and \(\mathrm{R102^{2.63}}\) , respectively. \(\mathrm{R102^{2.63}}\) also forms a hydrogen bond with a hydroxyl group attached to the indoline ring of ibuprofen. Mutations of hydrophobic residues in the cavity II caused similar functional consequences to both ibuprofen and ghrelin (Fig. 2f, Extended Data Fig. S3c). In addition, + +<--- Page Split ---> + +mutation of \(\mathrm{R102^{2.63}}\) , which forms hydrogen bonding and cation- \(\pi\) interactions with butamoren, resulted in compromised action of butamoren as well (Fig. 2f). + +## Distinctive activation mechanism of GHSR + +GHSR adopts an active conformation in our structures, which is stabilized by the \(\mathrm{G_i}\) protein. Both extracellular and cytoplasmic regions of active GHSR showed significant structural differences compared to the inactive GHSR in the crystal structure reported before \(^{34}\) . At the extracellular region, the most prominent difference observed lies in TM7 (Fig. 3a), which is unusual for class A GPCRs \(^{38}\) . Compared to that in the inactive GHSR, the extracellular segment of TM7 in both active GHSR structures with ghrelin and butamoren extends by two more helical turns and moves towards TM6 (Fig. 3a). Such a large conformational change is not likely to be caused directly by agonist- binding since there is little interaction between TM7 and both agonists. We also observed a notable shift of the extracellular half of TM6 in the active GHSR in comparison with the inactive GHSR (Fig. 3a). In the core region of GHSR, we observed remarkable conformational changes of highly conserved residues \(\mathrm{V131^{3.40}}\) , \(\mathrm{P224^{5.50}}\) , \(\mathrm{F272^{6.44}}\) and \(\mathrm{W276^{6.48}}\) (Fig. 3b). Conformational changes of these conserved residues especially the 'transmission switch' residues \(\mathrm{F272^{6.44}}\) and \(\mathrm{W276^{6.48}}\) have been suggested to link the extracellular agonist- binding to the cytoplasmic receptor activation and G protein- coupling for class A GPCRs (Extended Data Fig. 4a) \(^{38 - 40}\) . The cytoplasmic region of GHSR in our structures exhibits typical features of activated GPCRs including a large outward displacement of TM6 and an inward movement of TM7 (Fig. 3c) \(^{38}\) . In addition, the cytoplasmic regions of TMs 1- 4 together with ICL1 and ICL2 all showed notable movements. Surprisingly, the cytoplasmic region of TM5 in the inactive and active structures are almost identical (Fig. 3c). This is in contrast to most of other class A GPCRs, for which TM5 usually undergoes notable conformational changes at the cytoplasmic region during receptor activation \(^{38}\) . Several residues in the middle region of TM5 from \(\mathrm{F221^{5.47}}\) to \(\mathrm{P224^{5.50}}\) do exhibit large conformational changes in the active GHSR compared to those in the inactive GHSR (Extended Data Fig. 4b). However, these changes do not translate to a significant displacement of the cytoplasmic region of TM5. + +Our structural analysis suggested a receptor activation mechanism involving conformational changes of the salt bridge pair \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) and a unique structural motif underneath. Structural alignment of the ligand- binding pockets for the two agonists and the antagonist C12 suggests that agonist- binding causes a subtle change in the salt bridge between \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) (Fig. 4a). Specifically, the side chain of \(\mathrm{R283^{6.55}}\) moves towards \(\mathrm{F279^{6.51}}\) and \(\mathrm{H280^{6.52}}\) due to steric effects with the octanoyl chain of ghrelin or the phenyl group of butamoren (Figs. 4a and b). This conformational change further transduces through \(\mathrm{F279^{6.51}}\) and \(\mathrm{H280^{6.52}}\) to cause rearrangement of an aromatic cluster formed by residues \(\mathrm{W276^{6.48}}\) , \(\mathrm{F279^{6.51}}\) , \(\mathrm{H280^{6.52}}\) and \(\mathrm{F312^{7.42}}\) (Fig. 4b). As a result, \(\mathrm{W276^{6.48}}\) toggles to further induce significant displacement of \(\mathrm{F272^{6.44}}\) and the cytoplasmic segment of TM6, resulting in the activation of GHSR (Fig. 4b, Extended Data Fig. 4a). Such a notion of receptor activation was supported by previous + +<--- Page Split ---> + +mutagenesis studies, which showed that individual mutations of residues \(\mathrm{F279^{6.51}}\) , \(\mathrm{F312^{7.42}}\) and \(\mathrm{W276^{6.48}}\) in the aromatic cluster significantly lowered the potency of ghrelin in activating GHSR \(^{34,37}\) . Our proposed receptor activation mechanism also suggests that the conformation of a GHSR ligand around the \(\mathrm{E124^{3.33}}\) - \(\mathrm{R283^{6.55}}\) salt bridge in the cavity II is an important structural determinant for ligand efficacy, providing a molecular foundation for designing novel ghrelin agonists. + +## Dynamic nature of \(\mathbf{G}_{\mathrm{i}}\) -coupling to GHSR + +The \(\mathrm{G}_{\mathrm{i}}\) - coupling is almost identical in both structures of the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complexes with two agonists. The major interaction site with GHSR is at the C- terminal half of \(\alpha 5\) of \(\mathrm{G}\alpha_{\mathrm{i}}\) (Fig. 5a and b). Hydrophobic residues I344, L348, L353 of \(\mathrm{G}\alpha_{\mathrm{i}}\) line on one side of \(\alpha 5\) to form hydrophobic interactions with GHSR residues \(\mathrm{I145^{3.54}}\) , \(\mathrm{I235^{5.61}}\) , \(\mathrm{I239^{5.65}}\) and \(\mathrm{L265^{6.37}}\) (Fig. 5a). The side chains of N347 and D350 of \(\mathrm{G}\alpha_{\mathrm{i}}\) engage in polar interactions with the main chain carbonyl of A144 and the side chain of K329 of GHSR, respectively (Fig. 5a and b). The side chain of \(\mathrm{R141^{3.50}}\) in the conserved \(\mathrm{DR}^{3.50}\mathrm{Y}\) motif of GHSR also forms a hydrogen bond with the main chain carbonyl of C351 of \(\mathrm{G}\alpha_{\mathrm{i}}\) . In addition, F336, I343 and I344 in \(\alpha 5\) and L194 in \(\beta 2\) - \(\beta 3\) loop of \(\mathrm{G}\alpha_{\mathrm{i}}\) pack against GHSR residues P148 and I149 in ICL2 to form another set of hydrophobic interactions at the receptor and \(\mathrm{G}\alpha_{\mathrm{i}}\) interface (Fig. 5b). Besides \(\mathrm{G}\alpha_{\mathrm{i}}\) , \(\mathrm{G}\beta\) also directly interacts with GHSR through multiple polar interactions (Fig. 5a), similar to those observed in the structure of the formylpeptide 2 (FPR2)- \(\mathrm{G}_{\mathrm{i}}\) complex \(^{41}\) . + +Despite distinctive ligand recognition and receptor activation mechanisms, the \(\mathrm{G}_{\mathrm{i}}\) - coupling mode of GHSR is highly similar to those of other closely related neuropeptide GPCRs such as neurotensin receptors (NTSRs) \(^{42}\) . Previous structural studies on the NTSR1- \(\mathrm{G}_{\mathrm{i}}\) complex revealed two conformational states of the complex, the canonical and non- canonical states \(^{43}\) . Structural comparison indicates that the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex in our structures resembles the canonical NTSR1- \(\mathrm{G}_{\mathrm{i}}\) state with a similar \(\mathrm{G}_{\mathrm{i}}\) interaction profile (Fig. 6a). It is to be noted that our cryo- EM analysis also revealed minor structural classes of particles with different conformations besides the ghrelin- and ibuatamen- GHSR- \(\mathrm{G}_{\mathrm{i}}\) complexes with the 2.7- Å cryo- EM maps (Extended Data Fig. 1c). We assigned the conformation of the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex in the 2.7- Å structures as Conformer 1. 3D reconstruction using particles from one minor class of the ghrelin- GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex yielded a 3.5- Å map of a second conformation, which we assigned as Conformer 2 (Extended Data Figs. 1c and 5). Structural comparison with Conformer 1 showed that there is a small but remarkable shift of the entire \(\mathrm{G}_{\mathrm{i}}\) heterotrimer relative to the receptor with a \(\sim 15^{\circ}\) rotation of the N- terminal \(\alpha\) - helix ( \(\alpha \mathrm{N}\) ) and \(\sim 10^{\circ}\) rotation of the C- terminal \(\alpha\) - helix ( \(\alpha 5\) ) of \(\mathrm{G}\alpha_{\mathrm{i}}\) in Conformer 2 (Fig. 6b). Despite such differences in the orientation of \(\mathrm{G}_{\mathrm{i}}\) , the structure of ghrelin- GHSR and the interactions between \(\mathrm{G}_{\mathrm{i}}\) and GHSR stay largely unchanged in Conformer 2. Similar structural variations were also observed for the NTSR1- \(\mathrm{G}_{\mathrm{i}}\) complex at both canonical and non- canonical states \(^{43}\) . Whether such dynamic nature of \(\mathrm{G}_{\mathrm{i}}\) - coupling is associated with other \(\mathrm{G}_{\mathrm{i}}\) - coupled GPCRs is not clear. Interestingly, similar to GHSR, NTSR1 is also able to couple to multiple G proteins besides \(\mathrm{G}_{\mathrm{i}}\) \(^{44}\) . + +<--- Page Split ---> + +Both receptors lack critical structural elements for selective \(\mathrm{G_i}\) - coupling, which may result in potentially multiple modes of \(\mathrm{G_i}\) recognition. + +Both receptors lack critical structural elements for selective \(\mathrm{G_i}\) - coupling, which may result in potentially multiple modes of \(\mathrm{G_i}\) recognition.In summary, we report two cryo- EM structures of the GHSR- \(\mathrm{G_i}\) complexes with ghrelin and imitamoren. Our structures clearly reveal the ligand- binding pocket for ghrelin, where the peptide moiety of ghrelin mainly occupies cavity I while the octanoyl moiety adopts an extended conformation to occupy cavity II. Ibutamoren mimics the first four N- terminal residues of ghrelin including the octanoyl moiety to bind at the bottom region of the ligand- binding pocket. Both agonists cause conformational changes of the salt bridge pair E124 \(^{3,33}\) and R283 \(^{6,55}\) and the aromatic cluster W276 \(^{6,48}\) , F279 \(^{6,51}\) , H280 \(^{6,52}\) and F312 \(^{7,42}\) , which may in turn lead to GHSR activation. The overall conformation of the GHSR- \(\mathrm{G_i}\) complex in our structures highly resembles the canonical state of the NTSR1- \(\mathrm{G_i}\) complex. The variations of \(\mathrm{G_i}\) - coupling observed for both GHSR and NTSR1 shed light on a potentially highly dynamic nature of \(\mathrm{G_i}\) - coupling to GPCRs with a promiscuous G protein- coupling ability. + +<--- Page Split ---> + +## Acknowledgement + +AcknowledgementWe thank members of the Cryo- electron Microscopy and Tomography Center of St Jude Children's Research Hospital for help with cryo- EM data collection. J. S. is supported by the National Institutes of Health (NIH, HL143037) and American Lebanese Syrian Associated Charities (ALSAC). C.Z. is supported by NIH grants R35GM128641 and R03TR003306. + +## Contributions + +ContributionsH. L. performed protein expression and purification for cryo- EM studies and the calcium signaling assay. A.M. and J.S. performed cryo- EM data collection and processing. H.L., D.S., and C.Z. built the models and performed structure refinement. M.D. and J.B. performed the GTP turnover assay. H.L., S.J. and C.Z. supervised all studies, and C.Z wrote the manuscript with H.L., J.B., and J.S. + +## Competing Interests + +The authors declare no competing financial interests. + +## Data availability + +Data availabilityThe 3D cryo- EM density maps have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD- 24267 for the ghrelin- GHSR- \(\mathrm{G_i}\) complex and EMD- 24268 for the butamoren- GHSR- \(\mathrm{G_i}\) complex. Atomic coordinates for the atomic models of the ghrelin- GHSR- \(\mathrm{G_i}\) and butamoren- GHSR- \(\mathrm{G_i}\) complexes have been deposited in the Protein Data Bank under the accession numbers 7NA7 and 7NA8, respectively. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. Overall structures of the GHSR-G1 complexes. (a) and (b) Cryo-EM density maps of the complex with ghrelin (orange) and ibutamoren (yellow) and overall structures. GHSR bound to ghrelin and ibutamoren is colored in blue and green, respectively. \(G_{\mathrm{ai}}\) , \(G_{\beta}\) and \(G_{\gamma}\) subunits are colored in cyan, pink and light blue, respectively. ScFv16 is colored in grey. The density maps of the two agonists are shown in light grey. The octanoyl group in ghrelin is circled.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Binding pockets for ghrelin and ibutamoren. (a) Ghrelin binding in the cavity I and II. Ghrelin residues are labeled in orange. (b) Binding of the octanoyl group of ghrelin in the cavity II. (c) Dose-dependent action of ghrelin on the wild type GHSR (wtGHSR) and mutants. (d) Alignment of ghrelin and ibutamoren. (e) Ibutamoren binding pocket. (f) Dose-dependent action of ibutamoren on wtGHSR and mutants. In c and f, agonist-induced GHSR signaling was measured by \(\mathrm{Ca}^{2 + }\) mobilization assay. Each data point represents \(\mathrm{Mean} \pm \mathrm{S.D.}\) from 3 independent assays. Polar interactions are shown as dashed lines in a, b and e.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Active conformation of GHSR. Conformational changes of the extracellular regions of TM6 and TM7 (a), the highly conserved transmission switch residues (b), and cytoplasmic regions of TMs (c) in the active GHSR relative to them in the inactive GHSR. Two active GHSR bound to ghrelin and butamoren and the inactive GHSR bound to C12 are colored in blue, green and purple, respectively. Ghrelin, butamoren and C12 are shown as orange, yellow and lemon sticks, respectively. Red arrows indicate changes of residues or regions from inactive to active states.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. Conformational changes of critical motifs near the agonist-binding pocket. (a) Different conformations of the E124-R283 salt bridge in the active structures of GHSR with ghrelin and ibutamoren and the inactive structure of GHSR with C12. The salt bridge interactions are shown as dashed lines. (b) Rearrangement of the aromatic cluster residues W276, F279, H280 and F312 in the active and inactive GHSR.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. GHSR and \(\mathbf{G}_{\mathrm{i}}\) binding interface. (a) and (b) Detailed interactions between GHSR and \(\mathbf{G}_{\mathrm{i}}\) viewed from two angles. GHSR, \(\mathbf{G}_{\mathrm{ai}}\) , and \(\mathbf{G}_{\beta}\) are colored in blue, cyan and pink, respectively. Polar interactions are shown as dashed lines.
+ +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6. Structures of the ghrelin-GHSR-Gi complex in Conformer 1 and 2. (a) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 with the neurotensin-NTSR-Gi complex in the canonical state (PDB ID 6OS9). The Gi-coupling mode is highly similar. (b) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 and 2.
+ +<--- Page Split ---> + +## Materials and Methods + +## GHSR expression and purification + +The coding sequence of wild- type human ghrelin receptor (GHSR) was cloned into pFastbac (ThermoFisher) with an N terminal Flag tag followed by a peptide sequence corresponding to the N- terminal fragment of the human \(\beta_{2}\) - adrenergic receptor to increase its expression. The protein was expressed in Sf9 insect cells (Invitrogen) using the Bac- to- Bac baculovirus expression system (ThermoFisher). 1L cells were lysed by stirring in buffer containing \(20\mathrm{mM}\) Tris- HCl, pH7.5, \(0.2\mu \mathrm{g / ml}\) leupeptin, \(100\mu \mathrm{g / ml}\) benzamidine and \(1\mu \mathrm{M}\) PF- 05190457 (GHSR antagonist) (Tocris). Cell membranes were isolated by centrifugation at \(25,000\mathrm{g}\) for \(40\mathrm{min}\) . Then the pellet was solubilized in buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(750\mathrm{mM}\) NaCl, \(1\%\) (w/v) n- dodecyl- b- D- maltoside (DDM, Anatrace), \(0.2\%\) (w/v) sodium cholate (Sigma), \(0.1\%\) (w/v) cholesterol hemisuccinate (CHS, Anatrace), \(20\%\) (v/v) glycerol, \(0.2\mu \mathrm{g / ml}\) leupeptin, \(100\mu \mathrm{g / ml}\) benzamidine, \(500\mathrm{unit}\) Salt Active Nuclease (Arcticzymes) and \(1\mu \mathrm{M}\) PF- 05190457 at \(4^{\circ}\mathrm{C}\) for \(3\mathrm{h}\) . Insoluble material was separated by centrifugation at \(25,000\mathrm{g}\) for \(40\mathrm{min}\) . The supernatant was incubated with nickel Sepharose resin (GE healthcare) plus \(15\mathrm{mM}\) imidazole at \(4^{\circ}\mathrm{C}\) overnight. The resin was washed with 5 column volume of buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(500\mathrm{mM}\) NaCl, \(0.1\%\) (w/v) DDM, \(0.02\%\) (w/v) CHS, \(25\mathrm{mM}\) imidazole and \(1\mu \mathrm{M}\) PF- 05190457. The protein was eluted with buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(500\mathrm{mM}\) NaCl, \(0.1\%\) (w/v) DDM, \(0.02\%\) (w/v) CHS, \(400\mathrm{mM}\) imidazole and \(1\mu \mathrm{M}\) PF- 05190457 and loaded onto anti- Flag M1 antibody resin (homemade) after adding \(2\mathrm{mM}\) \(\mathrm{CaCl}_2\) . The detergent was slowly exchanged to \(0.01\%\) (w/v) lauryl maltose neopentyl glycol (MNG, Anatrace) containing \(1\mu \mathrm{M}\) ghrelin (Tocris) or MK0677 (Tocris) on the M1 antibody resin. The protein was finally eluted with buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.002\%\) (w/v) MNG, \(0.001\%\) (w/v) CHS, \(200\mathrm{mg / ml}\) Flag peptide, \(5\mathrm{mM}\) EDTA and \(1\mu \mathrm{M}\) ghrelin or MK- 0677, and further purified by size- exclusion chromatography using a Superdex 200 Increase column (Cytiva). The receptor was collected and concentrated to \(5\mathrm{mg / ml}\) using \(100\mathrm{- kDa}\) molecular weight cut- off concentrators (Millipore) for complex assembly. + +## Expression and purification of \(\mathbf{G}_{\mathrm{i}}\) heterotrimer + +The wild- type \(\mathrm{G}_{\mathrm{all}}\) was cloned in pFastBac vector without any tag, and the virus was prepared using Bac- to- Bac system (Invitrogen). N- terminal \(6\times\) His- tagged human \(\mathrm{G}_{\beta 1}\) , and human \(\mathrm{G}_{\gamma 2}\) were cloned into pVL1392 vector, and the virus was prepared using the BestBac system (Expression Systems). The heterotrimeric Gi complex was expressed in Sf9 insect cells by co- expressing all three subunits. Cells at a density of \(4\times 10^{6} / \mathrm{ml}\) were infected with both \(\mathrm{G}_{\mathrm{all}}\) and \(\mathrm{G}_{\mathrm{py}}\) virus at ratios of \(20\mathrm{ml}\) and \(1\mathrm{ml}\) per liter, respectively, at \(27^{\circ}\mathrm{C}\) for 48 hours before harvesting. Cells were lysed in lysis buffer containing \(10\mathrm{mM}\) Tris, pH 7.5, \(100\mu \mathrm{M}\) \(\mathrm{MgCl}_2\) , \(5\mathrm{mM}\) \(\beta\) - mercaptoethanol ( \(\beta\) - ME), \(10\mu \mathrm{M}\) GDP, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The cell membrane was collected by centrifugation at \(25,000\mathrm{g}\) for \(30\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) . Cell membranes were solubilized in solubilization + +<--- Page Split ---> + +buffer containing \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(1\%\) sodium cholate, \(0.05\%\) DDM, \(5\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(2\mu \mathrm{L}\) CIP, \(5\mathrm{mM}\) \(\beta\) - ME, \(10\mu \mathrm{M}\) GDP, \(10\%\) glycerol, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The supernatant was separated by centrifugation at \(25,000\mathrm{g}\) for \(30\mathrm{min}\) and incubated with nickel resin in batch for 1 hour at \(4^{\circ}\mathrm{C}\) . The resin was then washed in batch with solubilization buffer and transferred to a gravity column. The buffer was exchanged on column from solubilization buffer to wash buffer comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(50\mathrm{mM}\) NaCl, \(0.1\%\) DDM, \(1\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(5\mathrm{mM}\) \(\beta\) - ME, \(10\mu \mathrm{M}\) GDP, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The protein was eluted in wash buffer with \(250\mathrm{mM}\) imidazole and treated with Lamda Phosphatase (New England BioLab) and Alkaline Phosphatase (New England BioLab) overnight at \(4^{\circ}\mathrm{C}\) . The protein was further purified with anion exchange chromatography. The low salt buffer was comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(40\mathrm{mM}\) NaCl, \(0.1\%\) DDM, \(1\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(100\mu \mathrm{M}\) TCEP, \(10\mu \mathrm{M}\) GDP. The high salt buffer was prepared by adding \(1\mathrm{M}\) NaCl to the low salt buffer. The pure \(\mathrm{G_i}\) with an appropriate stoichiometry of three subunits was the supplemented by \(10\%\) glycerol, concentrated to \(\sim 20\mathrm{mg / ml}\) , flash- frozen in liquid nitrogen, and stored at \(- 80^{\circ}\mathrm{C}\) . + +## Assembly of GHSR- \(\mathbf{G}_{\mathrm{i}}\) complexes + +Purified GHSR was mixed with \(\mathrm{G_i}\) heterotrimer at a 1:1.2 molar ratio. The coupling reaction was initiated by incubation at \(25^{\circ}\mathrm{C}\) for 1 hour and was followed by addition of Apyrase (New England BioLab) to catalyze the hydrolysis of GDP overnight at \(4^{\circ}\mathrm{C}\) . This was to form the stable nucleotide- free complex. To remove the excess \(\mathrm{G_i}\) protein, the mixture was further purified with anti- Flag M1 antibody resin. The complex was eluted using the buffer comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.01\%\) MNG, \(0.001\%\) CHS, \(1\mu \mathrm{M}\) ghrelin or MK- 0677, \(200\mu \mathrm{g / ml}\) Flag peptide. Finally, a 1.2 molar excess of scFV16 was added to the elution. The GHSR- \(\mathrm{G_i}\) - scFV16 complex was purified and buffer- exchanged by size exclusion chromatography with buffer containing \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.002\%\) MNG, \(0.001\%\) CHS and \(1\mu \mathrm{M}\) ghrelin or MK- 0677. Peak fractions were concentrated to \(\sim 7\mathrm{mg / ml}\) for cryo- EM data collection. + +## Cryo-EM data collection, processing, model building and refinement + +Cryo- EM data collection, processing, model building and refinementCryo- EM grids were prepared with a Vitrobot Mark IV (FEI). Quantifoil R1.2/1.3 holey carbon gold grids (SPI) were glow- discharged for 30s. Then \(3.5\mu \mathrm{L}\) of \(7\mathrm{mg / ml}\) protein sample was pipetted onto the grids, which were blotted for 3s under blot force - 3 at \(100\%\) humidity and frozen in liquid nitrogen cooled liquid ethane. The grids were loaded onto a \(300\mathrm{keV}\) Titan Krios (FEI) with a K3 direct electron detector (Gatan) and an energy filter. + +Images of the ghrelin- GHSR- \(\mathrm{G_i}\) complex dataset were recorded with SerialEM \(^{45}\) in superresolution mode with a pixel size of \(0.649\mathrm{\AA}\) and a defocus range of \(- 0.8\) to \(- 1.6\mu \mathrm{m}\) . Movies were recorded during a 2.1 s exposure with \(30\mathrm{ms}\) subframes (70 total frames) at a dose rate of 1.176 \(\mathrm{e / \AA^2 /}\) frame. The dataset with the MK0677- GHSR- \(\mathrm{G_i}\) complex was collected with a pixel size of \(0.826\mathrm{\AA}\) and a defocus range of \(- 0.8\) to \(- 1.6\mu \mathrm{m}\) . Movies were recorded during a 3 s exposure with \(50\mathrm{ms}\) subframes (60 total frames) at a dose rate of \(1.35\mathrm{e / \AA^2 /}\) frame. + +<--- Page Split ---> + +Both datasets were processed in cryoSPARC \(^{46}\) using a similar strategy (Extended Data Fig. 1c and d). Briefly, super- resolution image stacks were gain- normalized, binned by 2 with Fourier cropping, and corrected for beam- induced motion using MotionCor2 \(^{47}\) . Contrast transfer function (CTF) parameters were estimated from motion- corrected images using GCTF \(^{48}\) . Following multiple rounds of 2D classification, ab initio reconstruction was performed in cryoSPARC asking for four classes, which resulted in one good class and three “trash” classes. Then multiple rounds of heterogenous refinement were performed against the four ab initio models to remove bad particles. Following the CTF refinement and non- uniform (NU) refinement, one last round of heterogenous refinement was carried out using three identical initial models (output of NU refinement). The heterogeneous refinement yielded two slightly different conformations, which were further refined by NU refinement. Resolutions were estimated using the 'gold standard' criterion (FSC=0.143), and local resolution was calculated in cryoSPARC. + +The coordinates of NTSR, \(\mathbf{G}_{\mathrm{i}}\) and scFv16 from the structure of the neurotensin- NTSR- \(\mathbf{G}_{\mathrm{i}}\) - scFv16 complex (PDB ID 6OS9) were used as initial models to dock into the cryo- EM maps using Chimera \(^{49}\) . The structures of GHSR- \(\mathbf{G}_{\mathrm{i}}\) with both agonists were then built by iterative manual building and adjustment in Coot \(^{50}\) and real- space refinement in Phenix \(^{51}\) . The final models were validated by Molprobity \(^{52}\) . Data collection, processing and structure refinement statistics are listed in the Extended Data Table 1. + +## Calcium mobilization assay + +Calcium mobilization assay to measure GHSR signaling was carried out in HEK- 293T cells cultured in DMEM supplemented with \(10\%\) (vol/vol) FBS (Fisher Scientific). The cDNAs encoding human GHSR was cloned into the pcDNA3.1(+) vector (Invitrogen) with a FLAG peptide fused to the N terminus. Mutant variants were then generated by site- directed mutagenesis and all mutants were fully sequenced. Various GHSR constructs were transfected to HEK- 293T cells using FuGENE Transfection Reagent (Promega). The surface expression of wtGHSR and various mutants in HEK- 293T cells was determined and confirmed using FACS with a fluorescent FLAG M1 antibody (homemade). + +To measure the agonist- induced calcium release, \(50~\mu \mathrm{L}\) dye loading buffer made of Hank's Balanced Salt Solution (HBSS) with \(5~\mu \mathrm{M}\) Fluo- 4 (Sigma), \(0.2\%\) pluronicacid and \(1\%\) FBS was added to the GHSR- expressing cells and incubated for \(1\mathrm{h}\) at \(37^{\circ}\mathrm{C}\) . Cells were then washed twice with HBSS buffer and left at room temperature in \(50~\mu \mathrm{L}\) HBBS. Fluo- 4 fluorescence intensity (excitation \(480\mathrm{nm}\) , emission \(520\mathrm{nm}\) ) was measured as an indicator of calcium release in a multimode reader (Spark 20 M, TECAN). For the concentration- dependent responses of ghrelin or MK- 0677, \(50~\mu \mathrm{L}\) HBSS buffer containing different concentrations of ligand was injected to each well, and fluorescence was measured constantly in real time for \(90\mathrm{s}\) . The data were analyzed in GraphPad Prism 8 using the one site dose- response stimulation method. Results were presented as Mean \(\pm\) S.D. from 3 independent experiments. + +<--- Page Split ---> + +## GTP turnover assay + +Human GHSR was expressed in \(E\) . coli inclusion bodies, folded in amphipol A8- 35 and then inserted into lipid nanodiscs formed with MSP1E3D1 and POPC:POPG (3:2 molar ratio) containing or not \(10\%\) cholesterol, as described in Damian et al \(^{53}\) . After insertion into nanodiscs, active receptor fractions were purified using affinity chromatography with the biotinylated version of the JMV2959 antagonist immobilized on a streptavidin column \(^{33}\) . GHSR containing discs were separated from aggregates and possible trace amounts of ligand through a size- exclusion chromatography step on a Superdex 200 increase 10/300 GL column (GE Healthcare) using a 25 mM Na- HEPES, \(150\mathrm{mMNaCl}\) , \(0.5\mathrm{mM}\) EDTA, \(\mathrm{pH}7.5\) buffer as the eluent. GTP turnover was assessed as described previously \(^{54}\) . All experiments were carried out at \(15^{\circ}\mathrm{C}\) . The receptor was first incubated with the isolated G protein and, when applicable, the ligand ( \(10\mu \mathrm{M}\) ) for 30 minutes in a \(25\mathrm{mM}\) HEPES, \(100\mathrm{mMNaCl}\) , \(5\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(\mathrm{pH}7.5\) buffer. GTP turnover was then started by adding GTP ( \(5\mu \mathrm{M}\) ) and GDP ( \(10\mu \mathrm{M}\) ), and the amount of remaining GTP was assessed after 10 minutes incubation at \(15^{\circ}\mathrm{C}\) using the GTPase- Glo assay (Promega). The signal was normalized in each case to that in the absence of receptor ( \(100\%\) ). + +<--- Page Split ---> + +## References + +<--- Page Split ---> + +19 Bright, G. M., Do, M. T., McKew, J. C., Blum, W. F. & Thorner, M. O. Development of a Predictive Enrichment Marker for the Oral GH Secretagogue LUM- 201 in Pediatric Growth Hormone Deficiency. J Endocr Soc 5, bvab030, doi:10.1210/jendso/bvab030 (2021).20 Patchett, A. et al. Design and biological activities of L- 163,191 (MK- 0677): a potent, orally active growth hormone secretagogue. Proceedings of the National Academy of Sciences 92, 7001- 7005 (1995).21 Müller, T. D. et al. Ghrelin. Molecular metabolism 4, 437- 460 (2015).22 Gutierrez, J. A. et al. Ghrelin octanoylation mediated by an orphan lipid transferase. Proceedings of the National Academy of Sciences 105, 6320- 6325 (2008).23 Yang, J., Brown, M. S., Liang, G., Grishin, N. V. & Goldstein, J. L. Identification of the acyltransferase that octanoylates ghrelin, an appetite- stimulating peptide hormone. Cell 132, 387- 396 (2008).24 Hedegaard, M. A. & Holst, B. The Complex Signaling Pathways of the Ghrelin Receptor. Endocrinology 161, doi:10.1210/endocr/bqaa020 (2020).25 Dezaki, K., Kakei, M. & Yada, T. Ghrelin uses Gai2 and activates voltage- dependent K+ channels to attenuate glucose- induced Ca2+ signaling and insulin release in islet β- cells: novel signal transduction of ghrelin. Diabetes 56, 2319- 2327 (2007).26 Mende, F. et al. Translating biased signaling in the ghrelin receptor system into differential in vivo functions. Proc Natl Acad Sci U S A 115, E10255- E10264, doi:10.1073/pnas.1804003115 (2018).27 Nagi, K. & Habib, A. M. Biased signaling: A viable strategy to drug ghrelin receptors for the treatment of obesity. Cell Signal 83, 109976, doi:10.1016/j.cellsig.2021.109976 (2021).28 Mear, Y., Enjalbert, A. & Thirion, S. GHS- R1a constitutive activity and its physiological relevance. Front Neurosci 7, 87, doi:10.3389/fnins.2013.00087 (2013).29 Pantel, J. et al. Loss of constitutive activity of the growth hormone secretagogue receptor in familial short stature. J Clin Invest 116, 760- 768, doi:10.1172/JCI25303 (2006).30 Ge, X. et al. LEAP2 Is an Endogenous Antagonist of the Ghrelin Receptor. Cell Metabolism 27, 461- 469. e466, doi:10.1016/j.cmet.2017.10.016 (2018).31 Srisai, D. et al. MRAP2 regulates ghrelin receptor signaling and hunger sensing. Nature communications 8, 1- 10 (2017).32 Bender, B. J. et al. Structural Model of Ghrelin Bound to its G Protein- Coupled Receptor. Structure 27, 537- 544 e534, doi:10.1016/j.str.2018.12.004 (2019).33 Ferre, G. et al. Structure and dynamics of G protein- coupled receptor- bound ghrelin reveal the critical role of the octanoyl chain. Proc Natl Acad Sci U S A 116, 17525- 17530, doi:10.1073/pnas.1905105116 (2019).34 Shiimura, Y. et al. Structure of an antagonist- bound ghrelin receptor reveals possible ghrelin recognition mode. Nature Communications 11, 4160- 4169, doi:10.1038/s41467- 020- 17554- 1 (2020).35 Koehl, A. et al. Structure of the micro- opioid receptor- Gi protein complex. Nature 558, 547- 552, doi:10.1038/s41586- 018- 0219- 7 (2018).36 Ballesteros, J. A. & Weinstein, H. Integrated methods for the construction of three- dimensional models and computational probing of structure- function relations in G protein- coupled receptors. Methods in Neurosciences 25, 366- 428, doi:https://doi.org/10.1016/S1043- 9471(05)80049- 7 (1995). + +<--- Page Split ---> + +37 Holst, B. et al. Common structural basis for constitutive activity of the ghrelin receptor family. Journal of Biological Chemistry 279, 53806- 53817 (2004).38 Zhou, Q. et al. Common activation mechanism of class A GPCRs. Elife 8, doi:10.7554/eLife.50279 (2019).39 Weis, W. I. & Kobilka, B. K. The Molecular Basis of G Protein- Coupled Receptor Activation. Annu Rev Biochem 87, 897- 919, doi:10.1146/annurev- biochem- 060614- 033910 (2018).40 Manglik, A. & Kruse, A. C. Structural Basis for G Protein- Coupled Receptor Activation. Biochemistry 56, 5628- 5634, doi:10.1021/acs.biochem.7b00747 (2017).41 Zhuang, Y. et al. Structure of formylpeptide receptor 2- Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467- 020- 14728- 9 (2020).42 Fredriksson, R., Lagerstrom, M. C., Lundin, L. G. & Schioth, H. B. The G- protein- coupled receptors in the human genome form five main families. Phylogenetic analysis, paralogon groups, and fingerprints. Mol Pharmacol 63, 1256- 1272, doi:10.1124/mol.63.6.1256 (2003).43 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1- Gi1 complex. Nature 572, 80- 85, doi:10.1038/s41586- 019- 1337- 6 (2019).44 Besserer- Offroy, E. et al. The signaling signature of the neurotensin type 1 receptor with endogenous ligands. Eur J Pharmacol 805, 1- 13, doi:10.1016/j.ejphar.2017.03.046 (2017).45 Mastronarde, D. N. Automated electron microscope tomography using robust prediction of specimen movements. J Struct Biol 152, 36- 51, doi:10.1016/j.jsb.2005.07.007 (2005).46 Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nat Methods 14, 290- 296, doi:10.1038/nmeth.4169 (2017).47 Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nat Methods 14, 331- 332, doi:10.1038/nmeth.4193 (2017).48 Zhang, K. Gctf: Real- time CTF determination and correction. J Struct Biol 193, 1- 12, doi:10.1016/j.jsb.2015.11.003 (2016).49 Pettersen, E. F. et al. UCSF Chimera- a visualization system for exploratory research and analysis. J Comput Chem 25, 1605- 1612, doi:10.1002/jcc.20084 (2004).50 Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr D Biol Crystallogr 66, 486- 501, doi:10.1107/S0907444910007493 (2010).51 Adams, P. D. et al. PHENIX: a comprehensive Python- based system for macromolecular structure solution. Acta Crystallogr D Biol Crystallogr 66, 213- 221, doi:10.1107/S0907444909052925 (2010).52 Chen, V. B. et al. MolProbit: all- atom structure validation for macromolecular crystallography. Acta Crystallogr D Biol Crystallogr 66, 12- 21, doi:10.1107/S0907444909042073 (2010).53 Damian, M. et al. High constitutive activity is an intrinsic feature of ghrelin receptor protein: a study with a functional monomeric GHS- R1a receptor reconstituted in lipid discs. J Biol Chem 287, 3630- 3641, doi:10.1074/jbc.M111.288324 (2012). + +<--- Page Split ---> + +54 Strohman, M. J. et al. Local membrane charge regulates beta2 adrenergic receptor coupling to Gi3. Nat Commun 10, 2234, doi:10.1038/s41467-019-10108-0 (2019). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformationNC2. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9_det.mmd b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..62988dd6928d77e9aa830ec624afbeccf3d0a136 --- /dev/null +++ b/preprint/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9/preprint__13dde86d52ba8d499c9301bf46bf87cbd32a3496b04fb4eaefbf6793d5ecdea9_det.mmd @@ -0,0 +1,280 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 951, 178]]<|/det|> +# Structural basis of human ghrelin receptor signaling by ghrelin and the synthetic agonist ibuatomoren + +<|ref|>text<|/ref|><|det|>[[44, 195, 620, 238]]<|/det|> +Cheng Zhang ( \(\circleddash\) chengzh@ptt.edu) University of Pittsburgh https://orcid.org/0000- 0001- 9042- 4007 + +<|ref|>text<|/ref|><|det|>[[44, 243, 262, 284]]<|/det|> +Heng Liu University of Pittsburgh + +<|ref|>text<|/ref|><|det|>[[44, 290, 620, 333]]<|/det|> +Dapeng Sun University of Pittsburgh https://orcid.org/0000- 0003- 0220- 8844 + +<|ref|>text<|/ref|><|det|>[[44, 337, 380, 377]]<|/det|> +Alexander Myasnikov St. Jude Children's Research Hospital + +<|ref|>text<|/ref|><|det|>[[44, 382, 404, 423]]<|/det|> +Marjorie Damian CNRS, Université de Montpellier, ENSCM + +<|ref|>text<|/ref|><|det|>[[44, 429, 470, 470]]<|/det|> +Jean- Louis Baneres IBMM https://orcid.org/0000- 0001- 7078- 1285 + +<|ref|>text<|/ref|><|det|>[[44, 475, 737, 517]]<|/det|> +Ji Sun St. Jude Children's Research Hospital https://orcid.org/0000- 0002- 9302- 3177 + +<|ref|>text<|/ref|><|det|>[[44, 558, 102, 575]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 594, 474, 614]]<|/det|> +Keywords: ghrelin, hunger hormone, cell signaling + +<|ref|>text<|/ref|><|det|>[[44, 632, 293, 652]]<|/det|> +Posted Date: July 12th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 670, 463, 690]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 646701/v1 + +<|ref|>text<|/ref|><|det|>[[44, 707, 910, 750]]<|/det|> +License: © \(\circleddash\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 787, 945, 830]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 4th, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 26735- 5. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[125, 95, 872, 135]]<|/det|> +# Structural basis of human ghrelin receptor signaling by ghrelin and the synthetic agonist ibuatomoren + +<|ref|>text<|/ref|><|det|>[[115, 140, 852, 179]]<|/det|> +Heng Liu \(^{1}\) , Dapeng Sun \(^{1}\) , Alexander Myasnikov \(^{2}\) , Marjorie Damian \(^{3}\) , Jean- Louis Banaeres \(^{3}\) , Ji Sun \(^{2*}\) , Cheng Zhang \(^{1*}\) + +<|ref|>text<|/ref|><|det|>[[115, 199, 875, 360]]<|/det|> +\(^{1}\) Department of Pharmacology and Chemical Biology, School of Medicine, University of Pittsburgh, Pittsburgh, PA15261, USA \(^{2}\) Department of Structural Biology, St. Jude Children's Research Hospital, Memphis, Tennessee, TN38120, USA \(^{3}\) Institut des Biomolécules Max Mousseron, CNRS, Université de Montpellier, ENSCM, Montpellier, France \(^{*}\) Correspondence: Dr. Ji Sun, Ji.Sun@stjude.org or Dr. Cheng Zhang, chengzh@pitt.edu. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 92, 192, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[115, 115, 879, 344]]<|/det|> +The 'hunger hormone' ghrelin activates the ghrelin receptor GHSR to stimulate food intake and growth hormone secretion and regulate reward signaling. Acylation of ghrelin at Ser3 is required for its agonistic action on GHSR. Synthetic agonists of GHSR are under clinical evaluation for disorders related to appetite and growth hormone dysregulation. Here, we report high- resolution cryo- EM structures of the GHSR- G \(_i\) signaling complex with ghrelin and the non- peptide agonist butamoren as an investigational new drug. Our structures together with mutagenesis data reveal the molecular basis for the binding of ghrelin and butamoren. Structural comparison suggests a salt bridge and an aromatic cluster near the agonist- binding pocket as important structural motifs in receptor activation. Notable variations of the G \(_i\) binding mode are observed in our cryo- EM analysis, indicating highly dynamic nature of G \(_i\) - coupling to GHSR. Our results provide a framework for understanding GHSR signaling and developing new GHSR agonist drugs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 225, 108]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[114, 115, 884, 428]]<|/det|> +The human ghrelin system is a critical component of the gut- brain axis to regulate energy homeostasis and reward signaling \(^{1 - 3}\) . Ghrelin is a 28- amino acid orexigenic peptide hormone generated in the gut commonly regarded as the 'hunger hormone' or 'survival hormone' \(^{1 - 4}\) . It mainly signals through the growth hormone secretagogue receptor (GHSR, or ghrelin receptor), a class A G protein- coupled receptor (GPCR). Ghrelin signaling through GHSR to stimulate growth hormone secretion and food intake under nutritional or physiological challenges \(^{1,2}\) . In addition, GHSR plays critical roles in the modulation of stress and anxiety \(^{5,6}\) and the regulation of dopamine signaling and thus reward pathways in the CNS \(^{7,8}\) . Therefore, GHSR is a highly pursued drug target \(^{9}\) . A GHSR inverse agonist named PF- 5190457 is under clinical investigation to treat alcohol use disorder \(^{10,11}\) . On the other hand, pharmacological stimulation of GHSR by agonists represents an emerging and exciting avenue to address disorders related to appetite, gastric emptying and growth hormone dysregulation. Relamorelin and anamorelin are two synthetic peptide agonists of GHSR that are in different stages of clinical trials to treat diabetic gastroparesis and cancer- related anorexia- cachexia \(^{12 - 16}\) . Ibutamoren, also known as MK- 0677 or LUM- 201, is a synthetic small- molecule GHSR agonist used as a growth hormone secretagogue in many disease settings \(^{9,17 - 20}\) . + +<|ref|>text<|/ref|><|det|>[[114, 432, 884, 746]]<|/det|> +The ghrelin- GHSR signaling system exhibits several unique properties compared to other GPCR signaling systems. Acylation of ghrelin, usually octanoylation at position- 3 serine, is required for its action on GHSR \(^{3}\) . Such a modification is achieved by the ghrelin O- acyl- transferase (GOAT) in vivo \(^{21 - 23}\) . Acylated ghrelin can activate GHSR to signal through multiple protein partners, including the \(\mathrm{G_{q / 11}}\) and \(\mathrm{G_{i / 0}}\) families of G proteins and \(\beta\) - arrestins \(^{24}\) . Several studies showed that different GHSR signaling pathways regulate distinct physiological functions \(^{25 - 27}\) . The GHSR and \(\mathrm{G_{i}}\) signaling pathway has been linked to attenuated glucose- induced insulin release by ghrelin \(^{25}\) , while the appetite stimulation is mainly mediated by \(\mathrm{G_{q / 11}}\) signaling \(^{26}\) . GHSR also exhibits a remarkably high constitutive activity, which is important for physiological growth hormone regulation \(^{28}\) . Mutations of the GHSR gene that result in loss of GHSR constitutive activity have been associated with the familial short stature syndrome \(^{29}\) . In addition, the constitutive activity of GHSR is negatively regulated by several endogenous mechanisms. Liver expressed antimicrobial peptide 2 (LEAP2) has been characterized as a GHSR inverse agonist or antagonist \(^{30}\) . Melanocortin receptor accessory protein 2 (MRAP2), a single- pass transmembrane protein, can directly associate with GHSR to modulate its activity \(^{3,31}\) . + +<|ref|>text<|/ref|><|det|>[[115, 752, 884, 898]]<|/det|> +Due to the physiological and pathological significance of the ghrelin- GHSR system, intensive research effort has been devoted to the molecular understanding of ghrelin action and GHSR signaling. Biophysical studies and structural modeling were performed to investigate ghrelin recognition by GHSR \(^{32,33}\) . Recently, a crystal structure of highly engineered GHSR at an inactive conformation bound to an antagonist, Compound 12 (C12), has been reported to probe a possible ghrelin recognition mechanism \(^{34}\) . Yet, the molecular mechanism underlying GHSR signaling by diverse peptide and non- peptide agonists still remains elusive. Here, we report two cryo- electron + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 884, 172]]<|/det|> +microscopy (cryo- EM) structures of human GHSR in complex with \(\mathrm{G_i}\) and two agonists, the endogenous ligand ghrelin and the non- peptide synthetic agonist ibutamoren. Structural analysis together with mutagenesis data revealed mechanisms underlying agonism of different types of GHSR agonists and GHSR- \(\mathrm{G_i}\) coupling. + +<|ref|>sub_title<|/ref|><|det|>[[116, 180, 309, 197]]<|/det|> +## Results and Discussion + +<|ref|>sub_title<|/ref|><|det|>[[115, 205, 744, 224]]<|/det|> +## Structure determination of two GHSR- \(\mathbf{G}_{\mathrm{i}}\) complexes and overall structures + +<|ref|>text<|/ref|><|det|>[[114, 230, 884, 439]]<|/det|> +We used wild type human GHSR and human \(\mathrm{G_{ai}}\) \(\mathrm{G_{b1}}\) and \(\mathrm{G_{v2}}\) to assemble the complexes with two agonists. No modification was introduced to the \(\mathrm{G_i}\) heterotrimer except for the amino- terminal 6xHis tag in \(\mathrm{G_{b1}}\) . We used apyrase to hydrolyze GDP in order to form stable nucleotide- free complexes. We also added an antibody fragment, scFv16, to further stabilize the \(\mathrm{G_i}\) heterotrimer \(^{35}\) . The structures of the GHSR- \(\mathrm{G_i}\) - scFv16 complexes with ghrelin and ibutamoren were both determined to an overall resolution of 2.7- Å (Fig. 1a and b, Extended Data Fig. S1). The cryo- EM maps allowed modeling of most regions of GHSR and \(\mathrm{G_i}\) heterotrimer, ghrelin residues Gly1- Val12 (residues in ghrelin are referred to by three- letter names, and residues in GHSR and other GPCRs are referred to by one- letter names hereafter) together with the octanoyl group, and the entire ibutamoren molecule (Fig. 1a and b, Extended Data Fig. S1). + +<|ref|>text<|/ref|><|det|>[[115, 444, 884, 527]]<|/det|> +We also observed cryo- EM densities likely corresponding to cholesterol molecules bound to GHSR in both structures (Extended Data Fig. S2a). Further experiments showed that cholesterol could positively regulate the activity of GHSR in reconstituted lipid nanodiscs (Extended Data Fig. S2b), suggesting a potentially significant physiological role of cholesterol in ghrelin signaling. + +<|ref|>sub_title<|/ref|><|det|>[[116, 534, 423, 552]]<|/det|> +## Ghrelin and ibutamoren recognition + +<|ref|>text<|/ref|><|det|>[[114, 559, 880, 807]]<|/det|> +In our structure, the N- terminal part of ghrelin adopts an extended conformation and inserts into a deep pocket of GHSR (Fig. 2a); such a binding mode is different from the predicted ghrelin binding mode based on the previous crystal structure \(^{34}\) . It was suggested that the salt bridge between E124 \(^{3,33}\) and R283 \(^{6,55}\) (superscripts represent Ballesteros- Weinstein numbering \(^{36}\) ) of GHSR divides the ghrelin- binding pocket into two cavities, cavity I and II \(^{34}\) . In our structures, we also observed this salt bridge. All the peptide moiety of ghrelin occupies the large cavity I while the octanoyl moiety attached to Ser3 of ghrelin extends into the small cavity II (Fig. 2a). It has been proposed that a crevasse region separating TM6 and TM7 in the inactive GHSR structure served as the recognition site of octanoyl moiety of ghrelin \(^{34}\) . However, our structure didn't support this hypothesis but instead showed that TM7 shifts towards TM6 compared to that in the inactive GHSR, closing the previously observed crevasse in the inactive structure (see discussion below). + +<|ref|>text<|/ref|><|det|>[[115, 814, 876, 896]]<|/det|> +Ghrelin engages in extensive polar and hydrophobic interactions with GHSR in the cavity I. From the bottom region to the extracellular side, the side chains of ghrelin residues Gly1 and Glu8 form salt bridges with the side chains of E124 \(^{3,33}\) and R107 \(^{ECL1}\) of GHSR, respectively; the side chains of Ser2 and Arg11 and the main chain carbonyl of Phe4 of ghrelin are hydrogen- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 867, 194]]<|/det|> +bonded by the main chain carbonyl groups of \(\mathrm{F290^{ECL2}}\) and \(\mathrm{S308^{7.38}}\) and the side chain of Y106 of GHSR, respectively (Fig. 2a). In addition to these polar interactions, in the middle region of the cavity II, Phe4 of ghrelin packs against GHSR residues \(\mathrm{L103^{2.64}}\) , \(\mathrm{L306^{7.36}}\) and \(\mathrm{F309^{7.39}}\) to form hydrophobic interactions (Fig. 2a). Leu5 of ghrelin also forms hydrophobic interactions with GHSR residues \(\mathrm{F286^{6.58}}\) and \(\mathrm{F290^{6.62}}\) (Fig. 2a). + +<|ref|>text<|/ref|><|det|>[[115, 199, 879, 471]]<|/det|> +The octanoyl moiety of ghrelin mainly occupies the cavity II with an extended conformation pointing towards the cleft between TM4 and TM5 (Fig. 2b). We observed a strong density corresponding to the proximal half of the octanoyl moiety, while the density for the distal half was not well resolved, suggesting a flexible nature of this part (Fig. 1a). Of note, the cavity II exhibits an amphipathic environment (Fig. 2b). As a result, the \(\mathrm{C_8}\) fatty acid chain of the octanoyl moiety sits on top of an extensive polar interaction network at the bottom of the cavity II including \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) while being capped by hydrophobic GHSR residues from the upper region of the cavity II (Fig. 2b). The acyl group of the octanoyl moiety sticks towards and likely interacts with the side chain of \(\mathrm{Q120^{3.29}}\) of GHSR. Previous mutagenesis studies showed that individual mutations of \(\mathrm{E124^{3.33}}\) , \(\mathrm{R283^{6.55}}\) and \(\mathrm{Q120^{3.29}}\) to non- polar residues led to decreased potency of ghrelin \(^{34,37}\) . We also showed that mutations of hydrophobic residues in the cavity II to larger or polar residues, which potentially disrupt the binding of the octanoyl moiety, resulted in compromised action of ghrelin (Fig. 2c), supporting our structural insights. + +<|ref|>text<|/ref|><|det|>[[113, 475, 886, 895]]<|/det|> +The non- peptide agonist ibuprofen also samples both cavity I and cavity II (Fig. 2d). It mainly occupies the bottom space of the ghrelin- binding pocket similar to the antagonist C12. The structures of ghrelin- and ibuprofen- bound GHSR share similar overall conformation with a root- mean- square deviation (RMSD) of the \(\mathrm{C\alpha}\) atoms at 1.23 Å. Subtle structural differences can be observed at the upper region of the binding pocket including ECL2 and 3 (Extended Data Fig. S3a), likely caused by the binding of Leu5- Val12 of ghrelin in this region. The three arms of ibuprofen point towards three different directions, mimicking the first four residues of ghrelin, Gly1- Phe4, including the octanoyl moiety (Fig. 2d). The phenyl group of ibuprofen sits in the cavity II and forms hydrophobic interactions with surrounding hydrophobic residues \(\mathrm{I178^{4.60}}\) , \(\mathrm{L181^{4.63}}\) and \(\mathrm{L210^{5.36}}\) of GHSR similarly to the octanoyl moiety of ghrelin (Fig. 2e). The other two arms of ibuprofen stick towards the bottom region of cavity II and the extracellular surface, respectively (Fig. 2e). Multiple hydrogen bonds are observed at the bottom of binding pocket between ibuprofen and GHSR residues \(\mathrm{D99^{2.60}}\) , \(\mathrm{Q120^{3.29}}\) , and \(\mathrm{R283^{6.55}}\) (Fig. 2e). Of note, the bottom region of cavity II exhibits a negatively charged environment to accommodate the smallest arm of ibuprofen with two amine groups (Extended Data Fig. S3b). The largest arm of ibuprofen containing an indoline- 3,4'-piperidine ring structure forms hydrophobic and cation- \(\pi\) interactions with the side chains of GHSR residues \(\mathrm{F286^{6.58}}\) and \(\mathrm{R102^{2.63}}\) , respectively. \(\mathrm{R102^{2.63}}\) also forms a hydrogen bond with a hydroxyl group attached to the indoline ring of ibuprofen. Mutations of hydrophobic residues in the cavity II caused similar functional consequences to both ibuprofen and ghrelin (Fig. 2f, Extended Data Fig. S3c). In addition, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 875, 130]]<|/det|> +mutation of \(\mathrm{R102^{2.63}}\) , which forms hydrogen bonding and cation- \(\pi\) interactions with butamoren, resulted in compromised action of butamoren as well (Fig. 2f). + +<|ref|>sub_title<|/ref|><|det|>[[115, 139, 477, 157]]<|/det|> +## Distinctive activation mechanism of GHSR + +<|ref|>text<|/ref|><|det|>[[113, 163, 884, 666]]<|/det|> +GHSR adopts an active conformation in our structures, which is stabilized by the \(\mathrm{G_i}\) protein. Both extracellular and cytoplasmic regions of active GHSR showed significant structural differences compared to the inactive GHSR in the crystal structure reported before \(^{34}\) . At the extracellular region, the most prominent difference observed lies in TM7 (Fig. 3a), which is unusual for class A GPCRs \(^{38}\) . Compared to that in the inactive GHSR, the extracellular segment of TM7 in both active GHSR structures with ghrelin and butamoren extends by two more helical turns and moves towards TM6 (Fig. 3a). Such a large conformational change is not likely to be caused directly by agonist- binding since there is little interaction between TM7 and both agonists. We also observed a notable shift of the extracellular half of TM6 in the active GHSR in comparison with the inactive GHSR (Fig. 3a). In the core region of GHSR, we observed remarkable conformational changes of highly conserved residues \(\mathrm{V131^{3.40}}\) , \(\mathrm{P224^{5.50}}\) , \(\mathrm{F272^{6.44}}\) and \(\mathrm{W276^{6.48}}\) (Fig. 3b). Conformational changes of these conserved residues especially the 'transmission switch' residues \(\mathrm{F272^{6.44}}\) and \(\mathrm{W276^{6.48}}\) have been suggested to link the extracellular agonist- binding to the cytoplasmic receptor activation and G protein- coupling for class A GPCRs (Extended Data Fig. 4a) \(^{38 - 40}\) . The cytoplasmic region of GHSR in our structures exhibits typical features of activated GPCRs including a large outward displacement of TM6 and an inward movement of TM7 (Fig. 3c) \(^{38}\) . In addition, the cytoplasmic regions of TMs 1- 4 together with ICL1 and ICL2 all showed notable movements. Surprisingly, the cytoplasmic region of TM5 in the inactive and active structures are almost identical (Fig. 3c). This is in contrast to most of other class A GPCRs, for which TM5 usually undergoes notable conformational changes at the cytoplasmic region during receptor activation \(^{38}\) . Several residues in the middle region of TM5 from \(\mathrm{F221^{5.47}}\) to \(\mathrm{P224^{5.50}}\) do exhibit large conformational changes in the active GHSR compared to those in the inactive GHSR (Extended Data Fig. 4b). However, these changes do not translate to a significant displacement of the cytoplasmic region of TM5. + +<|ref|>text<|/ref|><|det|>[[114, 670, 880, 899]]<|/det|> +Our structural analysis suggested a receptor activation mechanism involving conformational changes of the salt bridge pair \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) and a unique structural motif underneath. Structural alignment of the ligand- binding pockets for the two agonists and the antagonist C12 suggests that agonist- binding causes a subtle change in the salt bridge between \(\mathrm{E124^{3.33}}\) and \(\mathrm{R283^{6.55}}\) (Fig. 4a). Specifically, the side chain of \(\mathrm{R283^{6.55}}\) moves towards \(\mathrm{F279^{6.51}}\) and \(\mathrm{H280^{6.52}}\) due to steric effects with the octanoyl chain of ghrelin or the phenyl group of butamoren (Figs. 4a and b). This conformational change further transduces through \(\mathrm{F279^{6.51}}\) and \(\mathrm{H280^{6.52}}\) to cause rearrangement of an aromatic cluster formed by residues \(\mathrm{W276^{6.48}}\) , \(\mathrm{F279^{6.51}}\) , \(\mathrm{H280^{6.52}}\) and \(\mathrm{F312^{7.42}}\) (Fig. 4b). As a result, \(\mathrm{W276^{6.48}}\) toggles to further induce significant displacement of \(\mathrm{F272^{6.44}}\) and the cytoplasmic segment of TM6, resulting in the activation of GHSR (Fig. 4b, Extended Data Fig. 4a). Such a notion of receptor activation was supported by previous + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 878, 214]]<|/det|> +mutagenesis studies, which showed that individual mutations of residues \(\mathrm{F279^{6.51}}\) , \(\mathrm{F312^{7.42}}\) and \(\mathrm{W276^{6.48}}\) in the aromatic cluster significantly lowered the potency of ghrelin in activating GHSR \(^{34,37}\) . Our proposed receptor activation mechanism also suggests that the conformation of a GHSR ligand around the \(\mathrm{E124^{3.33}}\) - \(\mathrm{R283^{6.55}}\) salt bridge in the cavity II is an important structural determinant for ligand efficacy, providing a molecular foundation for designing novel ghrelin agonists. + +<|ref|>sub_title<|/ref|><|det|>[[115, 221, 460, 240]]<|/det|> +## Dynamic nature of \(\mathbf{G}_{\mathrm{i}}\) -coupling to GHSR + +<|ref|>text<|/ref|><|det|>[[114, 245, 886, 496]]<|/det|> +The \(\mathrm{G}_{\mathrm{i}}\) - coupling is almost identical in both structures of the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complexes with two agonists. The major interaction site with GHSR is at the C- terminal half of \(\alpha 5\) of \(\mathrm{G}\alpha_{\mathrm{i}}\) (Fig. 5a and b). Hydrophobic residues I344, L348, L353 of \(\mathrm{G}\alpha_{\mathrm{i}}\) line on one side of \(\alpha 5\) to form hydrophobic interactions with GHSR residues \(\mathrm{I145^{3.54}}\) , \(\mathrm{I235^{5.61}}\) , \(\mathrm{I239^{5.65}}\) and \(\mathrm{L265^{6.37}}\) (Fig. 5a). The side chains of N347 and D350 of \(\mathrm{G}\alpha_{\mathrm{i}}\) engage in polar interactions with the main chain carbonyl of A144 and the side chain of K329 of GHSR, respectively (Fig. 5a and b). The side chain of \(\mathrm{R141^{3.50}}\) in the conserved \(\mathrm{DR}^{3.50}\mathrm{Y}\) motif of GHSR also forms a hydrogen bond with the main chain carbonyl of C351 of \(\mathrm{G}\alpha_{\mathrm{i}}\) . In addition, F336, I343 and I344 in \(\alpha 5\) and L194 in \(\beta 2\) - \(\beta 3\) loop of \(\mathrm{G}\alpha_{\mathrm{i}}\) pack against GHSR residues P148 and I149 in ICL2 to form another set of hydrophobic interactions at the receptor and \(\mathrm{G}\alpha_{\mathrm{i}}\) interface (Fig. 5b). Besides \(\mathrm{G}\alpha_{\mathrm{i}}\) , \(\mathrm{G}\beta\) also directly interacts with GHSR through multiple polar interactions (Fig. 5a), similar to those observed in the structure of the formylpeptide 2 (FPR2)- \(\mathrm{G}_{\mathrm{i}}\) complex \(^{41}\) . + +<|ref|>text<|/ref|><|det|>[[114, 500, 885, 899]]<|/det|> +Despite distinctive ligand recognition and receptor activation mechanisms, the \(\mathrm{G}_{\mathrm{i}}\) - coupling mode of GHSR is highly similar to those of other closely related neuropeptide GPCRs such as neurotensin receptors (NTSRs) \(^{42}\) . Previous structural studies on the NTSR1- \(\mathrm{G}_{\mathrm{i}}\) complex revealed two conformational states of the complex, the canonical and non- canonical states \(^{43}\) . Structural comparison indicates that the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex in our structures resembles the canonical NTSR1- \(\mathrm{G}_{\mathrm{i}}\) state with a similar \(\mathrm{G}_{\mathrm{i}}\) interaction profile (Fig. 6a). It is to be noted that our cryo- EM analysis also revealed minor structural classes of particles with different conformations besides the ghrelin- and ibuatamen- GHSR- \(\mathrm{G}_{\mathrm{i}}\) complexes with the 2.7- Å cryo- EM maps (Extended Data Fig. 1c). We assigned the conformation of the GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex in the 2.7- Å structures as Conformer 1. 3D reconstruction using particles from one minor class of the ghrelin- GHSR- \(\mathrm{G}_{\mathrm{i}}\) complex yielded a 3.5- Å map of a second conformation, which we assigned as Conformer 2 (Extended Data Figs. 1c and 5). Structural comparison with Conformer 1 showed that there is a small but remarkable shift of the entire \(\mathrm{G}_{\mathrm{i}}\) heterotrimer relative to the receptor with a \(\sim 15^{\circ}\) rotation of the N- terminal \(\alpha\) - helix ( \(\alpha \mathrm{N}\) ) and \(\sim 10^{\circ}\) rotation of the C- terminal \(\alpha\) - helix ( \(\alpha 5\) ) of \(\mathrm{G}\alpha_{\mathrm{i}}\) in Conformer 2 (Fig. 6b). Despite such differences in the orientation of \(\mathrm{G}_{\mathrm{i}}\) , the structure of ghrelin- GHSR and the interactions between \(\mathrm{G}_{\mathrm{i}}\) and GHSR stay largely unchanged in Conformer 2. Similar structural variations were also observed for the NTSR1- \(\mathrm{G}_{\mathrm{i}}\) complex at both canonical and non- canonical states \(^{43}\) . Whether such dynamic nature of \(\mathrm{G}_{\mathrm{i}}\) - coupling is associated with other \(\mathrm{G}_{\mathrm{i}}\) - coupled GPCRs is not clear. Interestingly, similar to GHSR, NTSR1 is also able to couple to multiple G proteins besides \(\mathrm{G}_{\mathrm{i}}\) \(^{44}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 90, 883, 130]]<|/det|> +Both receptors lack critical structural elements for selective \(\mathrm{G_i}\) - coupling, which may result in potentially multiple modes of \(\mathrm{G_i}\) recognition. + +<|ref|>text<|/ref|><|det|>[[115, 141, 884, 371]]<|/det|> +Both receptors lack critical structural elements for selective \(\mathrm{G_i}\) - coupling, which may result in potentially multiple modes of \(\mathrm{G_i}\) recognition.In summary, we report two cryo- EM structures of the GHSR- \(\mathrm{G_i}\) complexes with ghrelin and imitamoren. Our structures clearly reveal the ligand- binding pocket for ghrelin, where the peptide moiety of ghrelin mainly occupies cavity I while the octanoyl moiety adopts an extended conformation to occupy cavity II. Ibutamoren mimics the first four N- terminal residues of ghrelin including the octanoyl moiety to bind at the bottom region of the ligand- binding pocket. Both agonists cause conformational changes of the salt bridge pair E124 \(^{3,33}\) and R283 \(^{6,55}\) and the aromatic cluster W276 \(^{6,48}\) , F279 \(^{6,51}\) , H280 \(^{6,52}\) and F312 \(^{7,42}\) , which may in turn lead to GHSR activation. The overall conformation of the GHSR- \(\mathrm{G_i}\) complex in our structures highly resembles the canonical state of the NTSR1- \(\mathrm{G_i}\) complex. The variations of \(\mathrm{G_i}\) - coupling observed for both GHSR and NTSR1 shed light on a potentially highly dynamic nature of \(\mathrm{G_i}\) - coupling to GPCRs with a promiscuous G protein- coupling ability. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 91, 273, 109]]<|/det|> +## Acknowledgement + +<|ref|>text<|/ref|><|det|>[[115, 115, 884, 199]]<|/det|> +AcknowledgementWe thank members of the Cryo- electron Microscopy and Tomography Center of St Jude Children's Research Hospital for help with cryo- EM data collection. J. S. is supported by the National Institutes of Health (NIH, HL143037) and American Lebanese Syrian Associated Charities (ALSAC). C.Z. is supported by NIH grants R35GM128641 and R03TR003306. + +<|ref|>sub_title<|/ref|><|det|>[[115, 206, 236, 224]]<|/det|> +## Contributions + +<|ref|>text<|/ref|><|det|>[[115, 230, 884, 314]]<|/det|> +ContributionsH. L. performed protein expression and purification for cryo- EM studies and the calcium signaling assay. A.M. and J.S. performed cryo- EM data collection and processing. H.L., D.S., and C.Z. built the models and performed structure refinement. M.D. and J.B. performed the GTP turnover assay. H.L., S.J. and C.Z. supervised all studies, and C.Z wrote the manuscript with H.L., J.B., and J.S. + +<|ref|>sub_title<|/ref|><|det|>[[115, 320, 290, 338]]<|/det|> +## Competing Interests + +<|ref|>text<|/ref|><|det|>[[115, 345, 533, 363]]<|/det|> +The authors declare no competing financial interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 371, 257, 390]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 396, 884, 500]]<|/det|> +Data availabilityThe 3D cryo- EM density maps have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD- 24267 for the ghrelin- GHSR- \(\mathrm{G_i}\) complex and EMD- 24268 for the butamoren- GHSR- \(\mathrm{G_i}\) complex. Atomic coordinates for the atomic models of the ghrelin- GHSR- \(\mathrm{G_i}\) and butamoren- GHSR- \(\mathrm{G_i}\) complexes have been deposited in the Protein Data Bank under the accession numbers 7NA7 and 7NA8, respectively. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[201, 95, 780, 641]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 650, 886, 755]]<|/det|> +
Figure 1. Overall structures of the GHSR-G1 complexes. (a) and (b) Cryo-EM density maps of the complex with ghrelin (orange) and ibutamoren (yellow) and overall structures. GHSR bound to ghrelin and ibutamoren is colored in blue and green, respectively. \(G_{\mathrm{ai}}\) , \(G_{\beta}\) and \(G_{\gamma}\) subunits are colored in cyan, pink and light blue, respectively. ScFv16 is colored in grey. The density maps of the two agonists are shown in light grey. The octanoyl group in ghrelin is circled.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 92, 879, 444]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 446, 884, 590]]<|/det|> +
Figure 2. Binding pockets for ghrelin and ibutamoren. (a) Ghrelin binding in the cavity I and II. Ghrelin residues are labeled in orange. (b) Binding of the octanoyl group of ghrelin in the cavity II. (c) Dose-dependent action of ghrelin on the wild type GHSR (wtGHSR) and mutants. (d) Alignment of ghrelin and ibutamoren. (e) Ibutamoren binding pocket. (f) Dose-dependent action of ibutamoren on wtGHSR and mutants. In c and f, agonist-induced GHSR signaling was measured by \(\mathrm{Ca}^{2 + }\) mobilization assay. Each data point represents \(\mathrm{Mean} \pm \mathrm{S.D.}\) from 3 independent assays. Polar interactions are shown as dashed lines in a, b and e.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 88, 884, 240]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 240, 880, 386]]<|/det|> +
Figure 3. Active conformation of GHSR. Conformational changes of the extracellular regions of TM6 and TM7 (a), the highly conserved transmission switch residues (b), and cytoplasmic regions of TMs (c) in the active GHSR relative to them in the inactive GHSR. Two active GHSR bound to ghrelin and butamoren and the inactive GHSR bound to C12 are colored in blue, green and purple, respectively. Ghrelin, butamoren and C12 are shown as orange, yellow and lemon sticks, respectively. Red arrows indicate changes of residues or regions from inactive to active states.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 92, 875, 384]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 388, 850, 494]]<|/det|> +
Figure 4. Conformational changes of critical motifs near the agonist-binding pocket. (a) Different conformations of the E124-R283 salt bridge in the active structures of GHSR with ghrelin and ibutamoren and the inactive structure of GHSR with C12. The salt bridge interactions are shown as dashed lines. (b) Rearrangement of the aromatic cluster residues W276, F279, H280 and F312 in the active and inactive GHSR.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[121, 88, 876, 317]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 319, 853, 380]]<|/det|> +
Figure 5. GHSR and \(\mathbf{G}_{\mathrm{i}}\) binding interface. (a) and (b) Detailed interactions between GHSR and \(\mathbf{G}_{\mathrm{i}}\) viewed from two angles. GHSR, \(\mathbf{G}_{\mathrm{ai}}\) , and \(\mathbf{G}_{\beta}\) are colored in blue, cyan and pink, respectively. Polar interactions are shown as dashed lines.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 90, 863, 363]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 368, 867, 453]]<|/det|> +
Figure 6. Structures of the ghrelin-GHSR-Gi complex in Conformer 1 and 2. (a) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 with the neurotensin-NTSR-Gi complex in the canonical state (PDB ID 6OS9). The Gi-coupling mode is highly similar. (b) Structural alignment of the ghrelin-GHSR-Gi complex in Conformer 1 and 2.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 91, 313, 108]]<|/det|> +## Materials and Methods + +<|ref|>sub_title<|/ref|><|det|>[[116, 116, 408, 134]]<|/det|> +## GHSR expression and purification + +<|ref|>text<|/ref|><|det|>[[114, 144, 884, 666]]<|/det|> +The coding sequence of wild- type human ghrelin receptor (GHSR) was cloned into pFastbac (ThermoFisher) with an N terminal Flag tag followed by a peptide sequence corresponding to the N- terminal fragment of the human \(\beta_{2}\) - adrenergic receptor to increase its expression. The protein was expressed in Sf9 insect cells (Invitrogen) using the Bac- to- Bac baculovirus expression system (ThermoFisher). 1L cells were lysed by stirring in buffer containing \(20\mathrm{mM}\) Tris- HCl, pH7.5, \(0.2\mu \mathrm{g / ml}\) leupeptin, \(100\mu \mathrm{g / ml}\) benzamidine and \(1\mu \mathrm{M}\) PF- 05190457 (GHSR antagonist) (Tocris). Cell membranes were isolated by centrifugation at \(25,000\mathrm{g}\) for \(40\mathrm{min}\) . Then the pellet was solubilized in buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(750\mathrm{mM}\) NaCl, \(1\%\) (w/v) n- dodecyl- b- D- maltoside (DDM, Anatrace), \(0.2\%\) (w/v) sodium cholate (Sigma), \(0.1\%\) (w/v) cholesterol hemisuccinate (CHS, Anatrace), \(20\%\) (v/v) glycerol, \(0.2\mu \mathrm{g / ml}\) leupeptin, \(100\mu \mathrm{g / ml}\) benzamidine, \(500\mathrm{unit}\) Salt Active Nuclease (Arcticzymes) and \(1\mu \mathrm{M}\) PF- 05190457 at \(4^{\circ}\mathrm{C}\) for \(3\mathrm{h}\) . Insoluble material was separated by centrifugation at \(25,000\mathrm{g}\) for \(40\mathrm{min}\) . The supernatant was incubated with nickel Sepharose resin (GE healthcare) plus \(15\mathrm{mM}\) imidazole at \(4^{\circ}\mathrm{C}\) overnight. The resin was washed with 5 column volume of buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(500\mathrm{mM}\) NaCl, \(0.1\%\) (w/v) DDM, \(0.02\%\) (w/v) CHS, \(25\mathrm{mM}\) imidazole and \(1\mu \mathrm{M}\) PF- 05190457. The protein was eluted with buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(500\mathrm{mM}\) NaCl, \(0.1\%\) (w/v) DDM, \(0.02\%\) (w/v) CHS, \(400\mathrm{mM}\) imidazole and \(1\mu \mathrm{M}\) PF- 05190457 and loaded onto anti- Flag M1 antibody resin (homemade) after adding \(2\mathrm{mM}\) \(\mathrm{CaCl}_2\) . The detergent was slowly exchanged to \(0.01\%\) (w/v) lauryl maltose neopentyl glycol (MNG, Anatrace) containing \(1\mu \mathrm{M}\) ghrelin (Tocris) or MK0677 (Tocris) on the M1 antibody resin. The protein was finally eluted with buffer containing \(20\mathrm{mM}\) HEPES, pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.002\%\) (w/v) MNG, \(0.001\%\) (w/v) CHS, \(200\mathrm{mg / ml}\) Flag peptide, \(5\mathrm{mM}\) EDTA and \(1\mu \mathrm{M}\) ghrelin or MK- 0677, and further purified by size- exclusion chromatography using a Superdex 200 Increase column (Cytiva). The receptor was collected and concentrated to \(5\mathrm{mg / ml}\) using \(100\mathrm{- kDa}\) molecular weight cut- off concentrators (Millipore) for complex assembly. + +<|ref|>sub_title<|/ref|><|det|>[[116, 675, 509, 694]]<|/det|> +## Expression and purification of \(\mathbf{G}_{\mathrm{i}}\) heterotrimer + +<|ref|>text<|/ref|><|det|>[[115, 703, 884, 889]]<|/det|> +The wild- type \(\mathrm{G}_{\mathrm{all}}\) was cloned in pFastBac vector without any tag, and the virus was prepared using Bac- to- Bac system (Invitrogen). N- terminal \(6\times\) His- tagged human \(\mathrm{G}_{\beta 1}\) , and human \(\mathrm{G}_{\gamma 2}\) were cloned into pVL1392 vector, and the virus was prepared using the BestBac system (Expression Systems). The heterotrimeric Gi complex was expressed in Sf9 insect cells by co- expressing all three subunits. Cells at a density of \(4\times 10^{6} / \mathrm{ml}\) were infected with both \(\mathrm{G}_{\mathrm{all}}\) and \(\mathrm{G}_{\mathrm{py}}\) virus at ratios of \(20\mathrm{ml}\) and \(1\mathrm{ml}\) per liter, respectively, at \(27^{\circ}\mathrm{C}\) for 48 hours before harvesting. Cells were lysed in lysis buffer containing \(10\mathrm{mM}\) Tris, pH 7.5, \(100\mu \mathrm{M}\) \(\mathrm{MgCl}_2\) , \(5\mathrm{mM}\) \(\beta\) - mercaptoethanol ( \(\beta\) - ME), \(10\mu \mathrm{M}\) GDP, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The cell membrane was collected by centrifugation at \(25,000\mathrm{g}\) for \(30\mathrm{min}\) at \(4^{\circ}\mathrm{C}\) . Cell membranes were solubilized in solubilization + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 884, 381]]<|/det|> +buffer containing \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(1\%\) sodium cholate, \(0.05\%\) DDM, \(5\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(2\mu \mathrm{L}\) CIP, \(5\mathrm{mM}\) \(\beta\) - ME, \(10\mu \mathrm{M}\) GDP, \(10\%\) glycerol, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The supernatant was separated by centrifugation at \(25,000\mathrm{g}\) for \(30\mathrm{min}\) and incubated with nickel resin in batch for 1 hour at \(4^{\circ}\mathrm{C}\) . The resin was then washed in batch with solubilization buffer and transferred to a gravity column. The buffer was exchanged on column from solubilization buffer to wash buffer comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(50\mathrm{mM}\) NaCl, \(0.1\%\) DDM, \(1\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(5\mathrm{mM}\) \(\beta\) - ME, \(10\mu \mathrm{M}\) GDP, \(0.2\mu \mathrm{g / ml}\) leupeptin and \(150\mu \mathrm{g / ml}\) benzamidine. The protein was eluted in wash buffer with \(250\mathrm{mM}\) imidazole and treated with Lamda Phosphatase (New England BioLab) and Alkaline Phosphatase (New England BioLab) overnight at \(4^{\circ}\mathrm{C}\) . The protein was further purified with anion exchange chromatography. The low salt buffer was comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(40\mathrm{mM}\) NaCl, \(0.1\%\) DDM, \(1\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(100\mu \mathrm{M}\) TCEP, \(10\mu \mathrm{M}\) GDP. The high salt buffer was prepared by adding \(1\mathrm{M}\) NaCl to the low salt buffer. The pure \(\mathrm{G_i}\) with an appropriate stoichiometry of three subunits was the supplemented by \(10\%\) glycerol, concentrated to \(\sim 20\mathrm{mg / ml}\) , flash- frozen in liquid nitrogen, and stored at \(- 80^{\circ}\mathrm{C}\) . + +<|ref|>sub_title<|/ref|><|det|>[[117, 390, 397, 409]]<|/det|> +## Assembly of GHSR- \(\mathbf{G}_{\mathrm{i}}\) complexes + +<|ref|>text<|/ref|><|det|>[[115, 418, 884, 626]]<|/det|> +Purified GHSR was mixed with \(\mathrm{G_i}\) heterotrimer at a 1:1.2 molar ratio. The coupling reaction was initiated by incubation at \(25^{\circ}\mathrm{C}\) for 1 hour and was followed by addition of Apyrase (New England BioLab) to catalyze the hydrolysis of GDP overnight at \(4^{\circ}\mathrm{C}\) . This was to form the stable nucleotide- free complex. To remove the excess \(\mathrm{G_i}\) protein, the mixture was further purified with anti- Flag M1 antibody resin. The complex was eluted using the buffer comprised of \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.01\%\) MNG, \(0.001\%\) CHS, \(1\mu \mathrm{M}\) ghrelin or MK- 0677, \(200\mu \mathrm{g / ml}\) Flag peptide. Finally, a 1.2 molar excess of scFV16 was added to the elution. The GHSR- \(\mathrm{G_i}\) - scFV16 complex was purified and buffer- exchanged by size exclusion chromatography with buffer containing \(20\mathrm{mM}\) HEPES pH 7.5, \(100\mathrm{mM}\) NaCl, \(0.002\%\) MNG, \(0.001\%\) CHS and \(1\mu \mathrm{M}\) ghrelin or MK- 0677. Peak fractions were concentrated to \(\sim 7\mathrm{mg / ml}\) for cryo- EM data collection. + +<|ref|>sub_title<|/ref|><|det|>[[116, 636, 692, 655]]<|/det|> +## Cryo-EM data collection, processing, model building and refinement + +<|ref|>text<|/ref|><|det|>[[115, 663, 883, 766]]<|/det|> +Cryo- EM data collection, processing, model building and refinementCryo- EM grids were prepared with a Vitrobot Mark IV (FEI). Quantifoil R1.2/1.3 holey carbon gold grids (SPI) were glow- discharged for 30s. Then \(3.5\mu \mathrm{L}\) of \(7\mathrm{mg / ml}\) protein sample was pipetted onto the grids, which were blotted for 3s under blot force - 3 at \(100\%\) humidity and frozen in liquid nitrogen cooled liquid ethane. The grids were loaded onto a \(300\mathrm{keV}\) Titan Krios (FEI) with a K3 direct electron detector (Gatan) and an energy filter. + +<|ref|>text<|/ref|><|det|>[[115, 775, 884, 899]]<|/det|> +Images of the ghrelin- GHSR- \(\mathrm{G_i}\) complex dataset were recorded with SerialEM \(^{45}\) in superresolution mode with a pixel size of \(0.649\mathrm{\AA}\) and a defocus range of \(- 0.8\) to \(- 1.6\mu \mathrm{m}\) . Movies were recorded during a 2.1 s exposure with \(30\mathrm{ms}\) subframes (70 total frames) at a dose rate of 1.176 \(\mathrm{e / \AA^2 /}\) frame. The dataset with the MK0677- GHSR- \(\mathrm{G_i}\) complex was collected with a pixel size of \(0.826\mathrm{\AA}\) and a defocus range of \(- 0.8\) to \(- 1.6\mu \mathrm{m}\) . Movies were recorded during a 3 s exposure with \(50\mathrm{ms}\) subframes (60 total frames) at a dose rate of \(1.35\mathrm{e / \AA^2 /}\) frame. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 884, 339]]<|/det|> +Both datasets were processed in cryoSPARC \(^{46}\) using a similar strategy (Extended Data Fig. 1c and d). Briefly, super- resolution image stacks were gain- normalized, binned by 2 with Fourier cropping, and corrected for beam- induced motion using MotionCor2 \(^{47}\) . Contrast transfer function (CTF) parameters were estimated from motion- corrected images using GCTF \(^{48}\) . Following multiple rounds of 2D classification, ab initio reconstruction was performed in cryoSPARC asking for four classes, which resulted in one good class and three “trash” classes. Then multiple rounds of heterogenous refinement were performed against the four ab initio models to remove bad particles. Following the CTF refinement and non- uniform (NU) refinement, one last round of heterogenous refinement was carried out using three identical initial models (output of NU refinement). The heterogeneous refinement yielded two slightly different conformations, which were further refined by NU refinement. Resolutions were estimated using the 'gold standard' criterion (FSC=0.143), and local resolution was calculated in cryoSPARC. + +<|ref|>text<|/ref|><|det|>[[115, 347, 884, 472]]<|/det|> +The coordinates of NTSR, \(\mathbf{G}_{\mathrm{i}}\) and scFv16 from the structure of the neurotensin- NTSR- \(\mathbf{G}_{\mathrm{i}}\) - scFv16 complex (PDB ID 6OS9) were used as initial models to dock into the cryo- EM maps using Chimera \(^{49}\) . The structures of GHSR- \(\mathbf{G}_{\mathrm{i}}\) with both agonists were then built by iterative manual building and adjustment in Coot \(^{50}\) and real- space refinement in Phenix \(^{51}\) . The final models were validated by Molprobity \(^{52}\) . Data collection, processing and structure refinement statistics are listed in the Extended Data Table 1. + +<|ref|>sub_title<|/ref|><|det|>[[117, 482, 348, 500]]<|/det|> +## Calcium mobilization assay + +<|ref|>text<|/ref|><|det|>[[115, 509, 884, 675]]<|/det|> +Calcium mobilization assay to measure GHSR signaling was carried out in HEK- 293T cells cultured in DMEM supplemented with \(10\%\) (vol/vol) FBS (Fisher Scientific). The cDNAs encoding human GHSR was cloned into the pcDNA3.1(+) vector (Invitrogen) with a FLAG peptide fused to the N terminus. Mutant variants were then generated by site- directed mutagenesis and all mutants were fully sequenced. Various GHSR constructs were transfected to HEK- 293T cells using FuGENE Transfection Reagent (Promega). The surface expression of wtGHSR and various mutants in HEK- 293T cells was determined and confirmed using FACS with a fluorescent FLAG M1 antibody (homemade). + +<|ref|>text<|/ref|><|det|>[[115, 683, 884, 893]]<|/det|> +To measure the agonist- induced calcium release, \(50~\mu \mathrm{L}\) dye loading buffer made of Hank's Balanced Salt Solution (HBSS) with \(5~\mu \mathrm{M}\) Fluo- 4 (Sigma), \(0.2\%\) pluronicacid and \(1\%\) FBS was added to the GHSR- expressing cells and incubated for \(1\mathrm{h}\) at \(37^{\circ}\mathrm{C}\) . Cells were then washed twice with HBSS buffer and left at room temperature in \(50~\mu \mathrm{L}\) HBBS. Fluo- 4 fluorescence intensity (excitation \(480\mathrm{nm}\) , emission \(520\mathrm{nm}\) ) was measured as an indicator of calcium release in a multimode reader (Spark 20 M, TECAN). For the concentration- dependent responses of ghrelin or MK- 0677, \(50~\mu \mathrm{L}\) HBSS buffer containing different concentrations of ligand was injected to each well, and fluorescence was measured constantly in real time for \(90\mathrm{s}\) . The data were analyzed in GraphPad Prism 8 using the one site dose- response stimulation method. Results were presented as Mean \(\pm\) S.D. from 3 independent experiments. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[116, 91, 286, 109]]<|/det|> +## GTP turnover assay + +<|ref|>text<|/ref|><|det|>[[114, 118, 884, 409]]<|/det|> +Human GHSR was expressed in \(E\) . coli inclusion bodies, folded in amphipol A8- 35 and then inserted into lipid nanodiscs formed with MSP1E3D1 and POPC:POPG (3:2 molar ratio) containing or not \(10\%\) cholesterol, as described in Damian et al \(^{53}\) . After insertion into nanodiscs, active receptor fractions were purified using affinity chromatography with the biotinylated version of the JMV2959 antagonist immobilized on a streptavidin column \(^{33}\) . GHSR containing discs were separated from aggregates and possible trace amounts of ligand through a size- exclusion chromatography step on a Superdex 200 increase 10/300 GL column (GE Healthcare) using a 25 mM Na- HEPES, \(150\mathrm{mMNaCl}\) , \(0.5\mathrm{mM}\) EDTA, \(\mathrm{pH}7.5\) buffer as the eluent. GTP turnover was assessed as described previously \(^{54}\) . All experiments were carried out at \(15^{\circ}\mathrm{C}\) . The receptor was first incubated with the isolated G protein and, when applicable, the ligand ( \(10\mu \mathrm{M}\) ) for 30 minutes in a \(25\mathrm{mM}\) HEPES, \(100\mathrm{mMNaCl}\) , \(5\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(\mathrm{pH}7.5\) buffer. GTP turnover was then started by adding GTP ( \(5\mu \mathrm{M}\) ) and GDP ( \(10\mu \mathrm{M}\) ), and the amount of remaining GTP was assessed after 10 minutes incubation at \(15^{\circ}\mathrm{C}\) using the GTPase- Glo assay (Promega). The signal was normalized in each case to that in the absence of receptor ( \(100\%\) ). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 210, 108]]<|/det|> +## References + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 880, 895]]<|/det|> +19 Bright, G. M., Do, M. T., McKew, J. C., Blum, W. F. & Thorner, M. O. Development of a Predictive Enrichment Marker for the Oral GH Secretagogue LUM- 201 in Pediatric Growth Hormone Deficiency. J Endocr Soc 5, bvab030, doi:10.1210/jendso/bvab030 (2021).20 Patchett, A. et al. Design and biological activities of L- 163,191 (MK- 0677): a potent, orally active growth hormone secretagogue. Proceedings of the National Academy of Sciences 92, 7001- 7005 (1995).21 Müller, T. D. et al. Ghrelin. Molecular metabolism 4, 437- 460 (2015).22 Gutierrez, J. A. et al. Ghrelin octanoylation mediated by an orphan lipid transferase. Proceedings of the National Academy of Sciences 105, 6320- 6325 (2008).23 Yang, J., Brown, M. S., Liang, G., Grishin, N. V. & Goldstein, J. L. Identification of the acyltransferase that octanoylates ghrelin, an appetite- stimulating peptide hormone. Cell 132, 387- 396 (2008).24 Hedegaard, M. A. & Holst, B. The Complex Signaling Pathways of the Ghrelin Receptor. Endocrinology 161, doi:10.1210/endocr/bqaa020 (2020).25 Dezaki, K., Kakei, M. & Yada, T. Ghrelin uses Gai2 and activates voltage- dependent K+ channels to attenuate glucose- induced Ca2+ signaling and insulin release in islet β- cells: novel signal transduction of ghrelin. Diabetes 56, 2319- 2327 (2007).26 Mende, F. et al. Translating biased signaling in the ghrelin receptor system into differential in vivo functions. Proc Natl Acad Sci U S A 115, E10255- E10264, doi:10.1073/pnas.1804003115 (2018).27 Nagi, K. & Habib, A. M. Biased signaling: A viable strategy to drug ghrelin receptors for the treatment of obesity. Cell Signal 83, 109976, doi:10.1016/j.cellsig.2021.109976 (2021).28 Mear, Y., Enjalbert, A. & Thirion, S. GHS- R1a constitutive activity and its physiological relevance. Front Neurosci 7, 87, doi:10.3389/fnins.2013.00087 (2013).29 Pantel, J. et al. Loss of constitutive activity of the growth hormone secretagogue receptor in familial short stature. J Clin Invest 116, 760- 768, doi:10.1172/JCI25303 (2006).30 Ge, X. et al. LEAP2 Is an Endogenous Antagonist of the Ghrelin Receptor. Cell Metabolism 27, 461- 469. e466, doi:10.1016/j.cmet.2017.10.016 (2018).31 Srisai, D. et al. MRAP2 regulates ghrelin receptor signaling and hunger sensing. Nature communications 8, 1- 10 (2017).32 Bender, B. J. et al. Structural Model of Ghrelin Bound to its G Protein- Coupled Receptor. Structure 27, 537- 544 e534, doi:10.1016/j.str.2018.12.004 (2019).33 Ferre, G. et al. Structure and dynamics of G protein- coupled receptor- bound ghrelin reveal the critical role of the octanoyl chain. Proc Natl Acad Sci U S A 116, 17525- 17530, doi:10.1073/pnas.1905105116 (2019).34 Shiimura, Y. et al. Structure of an antagonist- bound ghrelin receptor reveals possible ghrelin recognition mode. Nature Communications 11, 4160- 4169, doi:10.1038/s41467- 020- 17554- 1 (2020).35 Koehl, A. et al. Structure of the micro- opioid receptor- Gi protein complex. Nature 558, 547- 552, doi:10.1038/s41586- 018- 0219- 7 (2018).36 Ballesteros, J. A. & Weinstein, H. Integrated methods for the construction of three- dimensional models and computational probing of structure- function relations in G protein- coupled receptors. Methods in Neurosciences 25, 366- 428, doi:https://doi.org/10.1016/S1043- 9471(05)80049- 7 (1995). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 879, 877]]<|/det|> +37 Holst, B. et al. Common structural basis for constitutive activity of the ghrelin receptor family. Journal of Biological Chemistry 279, 53806- 53817 (2004).38 Zhou, Q. et al. Common activation mechanism of class A GPCRs. Elife 8, doi:10.7554/eLife.50279 (2019).39 Weis, W. I. & Kobilka, B. K. The Molecular Basis of G Protein- Coupled Receptor Activation. Annu Rev Biochem 87, 897- 919, doi:10.1146/annurev- biochem- 060614- 033910 (2018).40 Manglik, A. & Kruse, A. C. Structural Basis for G Protein- Coupled Receptor Activation. Biochemistry 56, 5628- 5634, doi:10.1021/acs.biochem.7b00747 (2017).41 Zhuang, Y. et al. Structure of formylpeptide receptor 2- Gi complex reveals insights into ligand recognition and signaling. Nat Commun 11, 885, doi:10.1038/s41467- 020- 14728- 9 (2020).42 Fredriksson, R., Lagerstrom, M. C., Lundin, L. G. & Schioth, H. B. The G- protein- coupled receptors in the human genome form five main families. Phylogenetic analysis, paralogon groups, and fingerprints. Mol Pharmacol 63, 1256- 1272, doi:10.1124/mol.63.6.1256 (2003).43 Kato, H. E. et al. Conformational transitions of a neurotensin receptor 1- Gi1 complex. Nature 572, 80- 85, doi:10.1038/s41586- 019- 1337- 6 (2019).44 Besserer- Offroy, E. et al. The signaling signature of the neurotensin type 1 receptor with endogenous ligands. Eur J Pharmacol 805, 1- 13, doi:10.1016/j.ejphar.2017.03.046 (2017).45 Mastronarde, D. N. Automated electron microscope tomography using robust prediction of specimen movements. J Struct Biol 152, 36- 51, doi:10.1016/j.jsb.2005.07.007 (2005).46 Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo- EM structure determination. Nat Methods 14, 290- 296, doi:10.1038/nmeth.4169 (2017).47 Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam- induced motion for improved cryo- electron microscopy. Nat Methods 14, 331- 332, doi:10.1038/nmeth.4193 (2017).48 Zhang, K. Gctf: Real- time CTF determination and correction. J Struct Biol 193, 1- 12, doi:10.1016/j.jsb.2015.11.003 (2016).49 Pettersen, E. F. et al. UCSF Chimera- a visualization system for exploratory research and analysis. J Comput Chem 25, 1605- 1612, doi:10.1002/jcc.20084 (2004).50 Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr D Biol Crystallogr 66, 486- 501, doi:10.1107/S0907444910007493 (2010).51 Adams, P. D. et al. PHENIX: a comprehensive Python- based system for macromolecular structure solution. Acta Crystallogr D Biol Crystallogr 66, 213- 221, doi:10.1107/S0907444909052925 (2010).52 Chen, V. B. et al. MolProbit: all- atom structure validation for macromolecular crystallography. Acta Crystallogr D Biol Crystallogr 66, 12- 21, doi:10.1107/S0907444909042073 (2010).53 Damian, M. et al. High constitutive activity is an intrinsic feature of ghrelin receptor protein: a study with a functional monomeric GHS- R1a receptor reconstituted in lipid discs. J Biol Chem 287, 3630- 3641, doi:10.1074/jbc.M111.288324 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 90, 824, 127]]<|/det|> +54 Strohman, M. J. et al. Local membrane charge regulates beta2 adrenergic receptor coupling to Gi3. Nat Commun 10, 2234, doi:10.1038/s41467-019-10108-0 (2019). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 390, 150]]<|/det|> +SupplementaryInformationNC2. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/images_list.json b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..a1e01e31a2e2b0e3327c88cfde6b1a927e5c7db5 --- /dev/null +++ b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/images_list.json @@ -0,0 +1,122 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 144, + 145, + 839, + 670 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. HTR2A deficiency in mice and hematopoietic stem cells exacerbates psoriasisinflammation. A) Schematic showing the experimental plan for Imiquimod-treatment of mice. B) Line chart showing changes in ear swelling of mice. (n=4 mice for Vas groups, n=10 mice for IMQ groups) C) Pictures showing ears of mice of various groups on Day 7. D) H&E-stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ \\(\\mathsf{V}_7 / 4\\mathsf{T}\\) cells and neutrophils in the respective groups with quantifications. The cells were isolated from the ears of mice. The detailed method is in Material and Methods section. F) qPCR of pro-inflammatory cytokines of respective groups. G) Schematic showing the experimental plan for developing bone marrow chimeric mice. H) Line chart showing changes in ear swelling of mice. (n=5 to 7 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E-stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ \\(\\mathsf{V}_7 / 4\\mathsf{T}\\) cells and neutrophils in the respective groups with quantifications. L) qPCR of pro-inflammatory cytokines of respective groups. Data are representative of three independent experiments (B-F and H-L). p values determined by two-way ANOVA (B and H) followed by Tukey's post-hoc test and one-way ANOVA (D-F and J-L) followed by Tukey's post-hoc test. Mean ± SEM (B, D-F, H, and J-L). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001; ns, non-significant.", + "footnote": [], + "bbox": [ + [ + 144, + 112, + 850, + 590 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 144, + 102, + 850, + 570 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 148, + 115, + 857, + 568 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. IL-23 plays a critical role in exacerbating psoriasisin inflammation by HTR2A deficient monocyte-derived Langerhans cells. A) Heatmap showing mRNA levels of key cytokines in wild type or HTR2A deficient moLCs. (3 biological replicates) The cells were stimulated for 24 hours with Imiquimod (10μg/mL). B) IL-23 ELISA of stimulated and unstimulated 1X105 wild type or HTR2A deficient moLCs. C) Representative histogram of IL-23a MFI among different cell types in cells isolated from ears of HTR2A deficient mice treated with Imiquimod for 7 days. (n=4) The cells were isolated and treated with PMA, Ionomycin, and GolgiStop before antibody staining and flow cytometry analysis. There is quantification of IL-23a MFI and percentage of IL-23a positive cells. D) Schematic showing the experimental design for treating mice with DOI and determining IL-23a expression. E) Line chart showing changes in ear", + "footnote": [], + "bbox": [ + [ + 145, + 103, + 857, + 710 + ] + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 146, + 117, + 806, + 480 + ] + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7", + "footnote": [], + "bbox": [ + [ + 142, + 106, + 842, + 481 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8", + "footnote": [], + "bbox": [ + [ + 147, + 113, + 844, + 501 + ] + ], + "page_idx": 34 + } +] \ No newline at end of file diff --git a/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d.mmd b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d.mmd new file mode 100644 index 0000000000000000000000000000000000000000..177c637090b6ad4f2d542089b9bdac394ec6dddb --- /dev/null +++ b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d.mmd @@ -0,0 +1,498 @@ + +# Serotonin 2A Receptor Attenuates Psoriatic Inflammation by Suppressing IL-23 Secretion in Monocyte-derived Langerhans Cells + +Yungling Lee leolee@ibms.sinica.edu.tw + +Institute of Biomedical Sciences, Academia Sinica https://orcid.org/0000- 0002- 2234- 9479 + +Yeh Fong Tan https://orcid.org/0000- 0003- 3361- 4300 + +Chen- Yun Yeh Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0003- 1049- 8052 + +Sheng- Yun Hsu Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan + +Chun- Hao Lu Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0001- 7084- 9374 + +Ching- Hui Tsai Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0003- 0998- 8230 + +Pei- Chuan Chiang Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0009- 0009- 8057- 4621 + +Hao- Jui Weng Taipei Medical University https://orcid.org/0000- 0002- 7006- 7599 + +Tsen- Fang Tsai Department of Dermatology, National Taiwan University Hospital and National Taiwan University College of Medicine, Taipei, Taiwan + +## Article + +Keywords: + +Posted Date: January 6th, 2025 + +<--- Page Split ---> + +DOI: https://doi.org/10.21203/rs.3.rs- 5628384/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: Yes there is potential Competing Interest. Tsen- Fang Tsai has conducted clinical trials or received honoraria for serving as a consultant for AbbVie, AnaptysBio, Bristol- Myers Squibb, Boehringer Ingelheim, Celgene, Eli Lilly, Galderma, GlaxoSmithKline- Stiefel, Janssen- Cilag, Leo- Pharma, Merck, Novartis, PharmaEssentia, Pfizer, Sanofi, Sun Pharma and UCB. The remaining authors state no conflict of interest. + +Version of Record: A version of this preprint was published at Nature Communications on September 29th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 63971- 5. + +<--- Page Split ---> + +## Title: Serotonin 2A Receptor Attenuates Psoriatic Inflammation by Suppressing IL-23 Secretion in Monocyte-derived Langerhans Cells + +One Sentence Summary: Serotonin 2A receptor expression on monocyte Langerhans cells attenuates psoriatic inflammation by reducing IL- 23 secretion. + +## Authors: + +Yeh Fong Tan1,2, Chen- Yun Yeh2, Sheng- Yun Hsu2, Chun- Hao Lu2, Ching- Hui Tsai2, Pei- Chuan Chiang2, Hao- Jui Weng3,4,5, Tsen- Fang Tsai3, Yungling Leo Lee2,6,7 + +## Affiliations: + +1. Taiwan International Graduate Program in Molecular Medicine, National Yang Ming Chiao Tung University and Academia Sinica; Taipei, Taiwan. +2. Institute of Biomedical Sciences, Academia Sinica; Taipei, Taiwan. +3. Department of Dermatology, National Taiwan University Hospital and National Taiwan University College of Medicine; Taipei, Taiwan. +4. Department of Dermatology, Taipei Medical University-Shuang Ho Hospital; New Taipei City, Taiwan. +5. Department of Dermatology, School of Medicine, College of Medicine, Taipei Medical University; Taipei, Taiwan. +6. College of Public Health, China Medical University, Taichung, Taiwan. +7. Lead contact. + +CORRESPONDENCE: leolee@ibms.sinica.edu.tw + +## Abstract: + +Anecdotal evidence has suggested an association between psychiatric drugs and psoriasis but consensus is absent due to contradicting reports and the mechanism remains poorly defined. Here, we investigated the role played by serotonin 2A receptor (HTR2A), a receptor commonly targeted by psychiatric drugs, in regulating psoriasis. HTR2A antagonistic drugs worsened psoriatic outcome and HTR2A modulation reduced psoriatic inflammation. Using Imiquimod- induced psoriasisform model, HTR2A- deficient mice experienced exacerbated inflammation. Hematopoietic cells, particularly monocyte- derived Langerhans cells (molC), were responsible for this phenotype. Mechanistically, the exacerbated inflammation is due to increased interleukin- 23 (IL- 23) secretion and HTR2A suppresses it by inhibiting activation of the non- canonical NFkB pathway. Serotonin is the putative agonist modulating HTR2A attenuating psoriatic inflammation. Lastly, our findings in mice were also validated clinically. Our data demonstrate serotonin modulates HTR2A, attenuating psoriatic inflammation by suppressing IL- 23 secretion via inhibiting non- canonical NFkB pathway in molCs. + +<--- Page Split ---> + +## INTRODUCTION + +INTRODUCTIONPsoriasis is a chronic immune- mediated inflammatory skin condition that affects millions of people worldwide.(1) Despite the identification of IL- 23/IL- 17 axis playing a key role in psoriasis, much remains unknown about psoriasis, as patients exhibit varied responses despite receiving identical treatment.(2) Of late, the psychological component of psoriasis has been given increased attention, as emerging evidence suggests mental health influences disease outcomes.(3, 4) High rates of psychiatric comorbidities are observed in psoriatic patients leading to the classification of psoriasis as a psychosomatic disease.(3, 5) For many patients, exacerbation of psoriatic episodes is often preceded by stressful life episodes, underscoring the need to unravel the complexities between psychological and physiological factors in psoriasis.(6, 7) + +Interestingly, studies have shown mixed results regarding the impact of mood- altering medications on psoriasis. While some selective serotonin reuptake inhibitors (SSRIs) appear to improve psoriatic conditions (8- 10), others have been linked to worsening psoriasis.(11- 13) This conflicting evidence highlights the need for a closer examination of serotonin's role in psoriasis. Thus, here we focused on second- generation antipsychotics- serotonin and dopamine receptor antagonists, which have strong antagonism for serotonin receptors.(14) In contrast, SSRIs have unpredictable effects on local serotonin levels depending on the duration of treatment.(15) Clozapine and risperidone, both second- generation antipsychotics, with their direct antagonism of serotonin receptors, particularly serotonin 2A receptor (HTR2A), facilitate a clear interpretation of results making them ideal candidates for further studies.(14, 16) + +HTR2A is a G protein- coupled receptor (GPCR) of the Gq subtype and is primarily activated by serotonin.(17) HTR2A can be activated by naturally occurring compounds like psilocybin or synthetic compounds like lysergic acid diethylamide and (R)- 1- (2,5- dimethoxy- 4- iodophenyl)- 2- aminopropane (DOI).(18, 19) In particular, DOI is widely used to study HTR2A as it is a HTR2A- specific agonist.(19) Studies have reported that DOI suppresses inflammation by blocking tumor necrosis factor- alpha (TNF- a)- induced cytokine and chemokine expression both in vivo and in vitro yet we lack a detailed understanding of how HTR2A regulates immune cells and cytokines in the context of psoriasis.(20, 21) + +We previously found HTR2A to be highly expressed in Langerhans cells (LC), indicating HTR2A could impact LC functions affecting psoriatic outcome.(22) LCs are skin macrophages with dendritic cell functions and are known to play a role in the exacerbation of psoriatic symptoms involving the cytokine IL- 23 (23, 24); however, some reports suggest otherwise (25), indicating that LCs might play a role in immune tolerance. This schism could be due to the existence of two subsets of LCs, identified as embryo LC (eLC) and monocyte- derived LC (moLC).(26) eLCs originate from erythro- myeloid progenitors which were formed from the yolk sac, while moLCs originate from monocyte- like precursors which are derived from definitive hematopoiesis. (26) moLCs can be identified by their expression of CD64 and have been known to play a role in exacerbating psoriasisform inflammation.(27- 29) It remains to be investigated how each of the LC subsets is affected by HTR2A. + +While HTR2A's role in regulating cytokine- mediated inflammation is documented, no studies have yet linked it directly to autoimmune diseases like psoriasis. Addressing this gap, our study + +<--- Page Split ---> + +investigates how HTR2A expression influences psoriasisform inflammation. We found that HTR2A expression on molCs attenuates psoriasisform inflammation by reducing IL- 23 secretion through the non- canonical NFκB pathway. + +<--- Page Split ---> + +## RESULTS + +## Modulation and expression levels of HTR2A affect psoriasis + +To determine the association between psoriasis and serotonin receptors, we performed a retrospective cohort study to examine serotonin- related drug use and changes in psoriasis treatment received as a surrogate marker for psoriasis severity. The categories of psoriasis treatment are as defined by Taiwan Dermatological Association with topical treatment for patients with mild psoriasis symptoms, followed by systemic agents, combination therapy, and biologics for patients with most severe symptoms.(30) We defined worsening psoriasis as a change from mild treatment to severe treatment, for example from receiving topical treatment to receiving biologics. We found patients taking anti- psychotics were more likely to have worsening psoriasis as compared to control patients not taking anti- psychotics or SSRIs (Figure 1A and S1C). Correspondingly, patients on anti- psychotics did not show any significant improvement in psoriasis (Figure S1A and S1C). Of note, patients on SSRIs were found to have a significant worsening but no improvement in psoriasis (Figure S1B and S1C). + +We then looked at the hazard ratio of each anti- psychotic and found patients on aripiprazole, chlorpromazine, prochlorperazine, quetiapine, risperidone, sulpiride and anti- psychotics categorized as Others (brexipiprazole, fluphenazine, loxapine, lurasidone, perphenazine, pimozide, thioridazine, and trifluoperazine) were more likely to have worsening psoriasis as compared to controls as they had statistically significant hazard ratios that were greater than one (Figure 1B). We did not find statistically significant hazard ratios in patients taking amisulpride, clozapine, haloperidol, and olanzapine. We then checked the binding affinity of each antipsychotic to each serotonin receptor on the PDSP K, database.(31) We found most of these antipsychotics bound most strongly to HTR2A among the serotonin receptors indicating HTR2A might be responsible for the worsening psoriasis observed (Figure 1C and S1D). + +To determine whether HTR2A plays a role in psoriasis, we modulated HTR2A via the specific agonist of DOI and wondered if it causes transcriptional changes in psoriatic genes. We obtained skin biopsies from lesioned sites of three psoriatic patients who have yet to receive any psoriatic treatment and skin from three healthy subjects as controls. The biopsies were halved, with half of it reserved for RNA sequencing without treatment (NT) and the other half treated with DOI (T) for 16 hours. All samples were subjected to RNA sequencing (Figure 1D) (GSE274449). We determined the differentially expressed genes (DEGs) of lesioned skin T versus lesioned skin NT and identified 653 genes. A total of 1475 psoriatic DEGs were identified when comparing NT healthy skin versus NT lesioned skin. We then compared these two sets of DEGs and found 107 DEGs overlapped (Figure 1E). 94 of these genes were upregulated in psoriasis but were downregulated following DOI treatment and 13 of these genes were downregulated in psoriasis but with DOI treatment, they were upregulated. In healthy skin irrespective of treatment, the identified genes share a similar pattern with the lesioned treated group indicating HTR2A activation on lesioned psoriatic skin reversed psoriasis transcriptomic profile (Figure 1F). We then determined HTR2A expression in psoriatic cases using publicly available datasets on GEO database and found lower HTR2A expression in psoriatic cases (Figure 1G). In all, these data strongly suggest that HTR2A plays a role in affecting psoriatic outcomes. + +<--- Page Split ---> + +## HTR2A deficiency in mice and hematopoietic stem cells exacerbates psoriasisform inflammation + +To investigate HTR2A's role in psoriasis further, we treated HTR2A deficient (KO) and aged- match, gender- matched wild type (WT) mice with Imiquimod (IMQ) for 7 days (Figure 2A). There was a significant increase in ear swelling in IMQ- treated KO mice as compared to WT mice. In contrast, no differences were observed in mice treated with Vaseline irrespective of HTR2A expression (Figure 2B). By visual inspection, we found the ears of IMQ- treated KO mice to be thicker with more scaling (Figure 2C). Histologically, acanthosis along with hyperkeratosis, parakeratosis, Munro's microabscesses, and immune infiltrate are hallmarks of psoriasis and reflect the severity of psoriasis.(32) There was obvious acanthosis (thickening of the epidermis) and immune infiltrate in hematoxylin and eosin (H&E)- stained slides of IMQ- treated KO mice. There was also a significant increase in epidermis thickening in IMQ- treated KO mice as compared to IMQ- treated WT mice (Figure 2D). It has been reported that \(\gamma \delta\) T cells will respond to IL- 23, a key cytokine in the pathogenesis of psoriasis, and differentiate into IL- 17- producing cells.(33) Of note, the \(\mathrm{V}\gamma 4\) subset is one of the major IL- 17- producing \(\gamma \delta\) T cells.(34, 35) Circulating neutrophils are also normally recruited into inflammatory sites and in psoriasis, Munro's microabscesses are filled with neutrophils.(36) Flow cytometry of cells according to our gating strategy revealed an increase in IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4+\) T cells and an increase in neutrophils (Figure 2E and S2). However, there were no statistically significant differences in IL- 17- and IL- 22- producing \(\alpha \beta\) CD4 T cells. Key pro- inflammatory cytokines like Il17a and Il23a were also increased in HTR2A deficient mice (Figure 2F). In all, HTR2A deficiency exacerbates psoriasisform inflammation. + +To investigate which cell was responsible for the exacerbated inflammation, we performed reciprocal bone marrow transplants to generate chimeric mice with WT bone marrow (representing hematopoietic stem cells (HSCs) and KO somatic cells (representing non- HSCs) and vice versa. KO bone marrow was transplanted to KO mice as a control (Figure 2G). We checked the composition of cells 11 weeks post- transplant and approximately \(65\% - 70\%\) of CD45+ cells were replaced with cells from donor bone marrow (Figure S3A). A statistically significant increase in ear swelling was observed in mice with KO bone marrow irrespective of its somatic cells (Figure 2H). Visually, the ears of mice with KO bone marrow appeared thicker with some scaling (Figure 2I). Histologically, the dermis did not vary much between the groups but the epidermis was significantly thicker in mice with KO bone marrow than in mice with WT bone marrow (Figure 2J). Flow cytometry of immune cells revealed an increase in IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4+\) T cells in mice with KO bone marrow. Paradoxically, neutrophil levels were lower in mice with KO bone marrow (Figure 2K). Il17a and Il23a were increased in mice with KO bone marrow (Figure 2L). Our findings suggest that HTR2A deficiency in HSCs is responsible for the exacerbated inflammation. + +Since HTR2A deficiency in HSCs exacerbates inflammation and immune cells make up a large part of HSCs, we wondered are the adaptive immune cells or the innate immune cells causing this exacerbated inflammation. To answer this, we generated mice with conditional deletion of HTR2A in B and T cells (HTR2ARag1) by crossing Rag1- cre mice with HTR2A fixed mice (Figure S4A). (37) We found no increase in ear swelling in IMQ- treated- HTR2ARag1 mice as compared to IMQ- treated- HTR2Afl/fl mice (Figure S4B). Visual inspection of mice ears found minimal scaling and no differences between the groups (Figure S4C). Histologically, there was no increase in the + +<--- Page Split ---> + +thickening of the epidermis or the dermis between the groups. When epidermal thickening was quantified, there were no significant differences (Figure S4D). There were also no differences in pro- inflammatory immune cells and pro- inflammatory cytokines between IMQ- treated- HTR2ARag1 mice and IMQ- treated- HTR2Afl/fl mice (Figure S4E and S4F). The data here indicates that HTR2A deficiency in B and T cells, which are the adaptive immune cells, does not cause exacerbated inflammation. + +## HTR2A deficiency in Langerhans cells recapitulates exacerbated inflammation in HTR2A deficient mice + +As adaptive immune cells were not responsible for the exacerbated inflammation, we turned our attention to innate immune cells. Monocyte- derived inflammatory Langerhans cells have been reported to cause psoriasis- like inflammation.(28) Hence, we decided to generate mice with conditional deletion of HTR2A in Langerhans cells (HTR2ACD207) by crossing CD207- cre mice with HTR2A fixed mice (Figure 3A).(38) We found conditional deletion of HTR2A in Langerhans cells (LC) recapitulated inflammatory levels observed in HTR2A deficient mice with increased ear swelling, increased scaling, increased epidermal thickening, increased IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4\) T cells, and increased mRNA levels of Il17a and Il23a (Figures 3B- 3F). However, the increase in neutrophil recruitment is not statistically significant (Figure 3E). + +To further validate HTR2A's influence on LC exacerbating inflammation, we used a Langerhans cell depletion model, whereby LCs were depleted by administering diphtheria toxins to CD207- diphtheria toxin receptor (CD207- DTR) mice.(28) We validated that this method could lead to the depletion of LCs (Figures S5A- S5D). All the mice were administered with \(0.1\mathrm{mg / kg}\) of DOI intraperitoneally (Figure 3G). In terms of ear swelling, we found DOI- suppressed inflammation in mice with LCs but not in mice with depleted LCs (Figure 3H). A similar trend was observed visually and histologically (Figure 3I and 3J). IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4\) T cells were low in mice with no depletion but high in mice with LCs depletion and the differences were statistically significant (Figure 3K). Neutrophil levels follow a similar trend but there was no statistical significance (Figure 3K). mRNA of Il17a and Il23a is high in mice with LC depletion and low in mice with no depletion, however, only differences in Il17a reach statistical significance (Figure 3L). The data here strongly supports that HTR2A deficiency in LCs causes exacerbated inflammation. + +## HTR2A deficiency in Langerhans cells of monocyte lineage exacerbates psoriasisinflammation + +We have found HTR2A deficiency in both cells of bone marrow origin and LCs lead to exacerbated inflammation, but these two findings contradict each other, as LCs are not replaced during bone marrow transplantation.(28) To reconcile the differences, we suspect HTR2A affects LCs derived from monocytes, which is of bone marrow origin, causing exacerbated inflammation. To test our hypothesis, we differentiated bone marrow cells into monocyte- derived Langerhans cells (molC) as published previously (Figure 4A).(39) These differentiated cells are considered of monocyte origin due to their CD64 expression and Irf8 expression (lineage defining transcription factor for monocytes) (Figure S6A- S6C).(27, 39) They are considered LCs due to their expression of Id2 and Runx3 which are lineage defining transcription factors for LCs (Figure S6C).(39, 40) + +<--- Page Split ---> + +Cells expressing MHC II and CD207 are considered moLcs whereas all the other cells are considered non- moLcs (Figure S6B). These cells were then adoptively transferred into the ear pinna of LC- depleted mice (Figure 4B). Exacerbated inflammation was observed in mice with HTR2A deficient moLcs, whereas a lower- grade inflammation was observed in mice with WT moLcs (Figure 4B- 4D). Mice adoptively transferred with non- moLcs irrespective of HTR2A expression experienced minimal inflammation (Figure 4B- 4D). A similar inflammatory trend was observed in IL- 17- and IL- 22- producing \(\mathrm{V} / \mathrm{Y}4\mathrm{T}\) cells and mRNA levels of Il17a; however, this trend was not observed in neutrophil levels and mRNA levels of Il23a (Figure 4E and 4F). + +To further validate the role of LCs of monocyte origin being responsible for the exacerbated inflammation seen in HTR2A deficient mice, we generated mice with conditional deletion of HTR2A in cells of monocyte lineage (HTR2AMs43) by crossing Ms43- cre mice with HTR2A fixed mice (Figure 4G). (41) Exacerbated inflammation was observed in IMQ- treated- HTR2AMs43 mice in terms of ear thickening, visually, and histologically (Figure 4H- 4I). When quantified, there were significant increases in ear swelling and epidermal thickening (Figure 4H and 4J). IL- 17- and IL- 22- producing \(\mathrm{V} / \mathrm{Y}4\mathrm{T}\) cells but not neutrophils were significantly increased (Figure 4K). mRNA levels of Il17a and Il23a were increased significantly in IMQ- treated- HTR2AMs43 mice (Figure 4L). In all, we conclude that HTR2A deficiency in LCs of monocyte origin exacerbates psoriasisform inflammation. + +## IL-23 secretion by HTR2A deficient monocyte-derived Langerhans cells plays a critical role in exacerbating psoriasisform inflammation + +We turned our focus on the cytokine affected by HTR2A deficiency in moLcs. We determined the mRNA levels of commonly secreted cytokines by Langerhans cells (Tnf, Il1b, Il23a, Il12a, Il6, Il2, Il10 and Tgfb1) after stimulating them with Imiquimod. (22, 39, 42) We found Il23a to be drastically increased in IMQ- stimulated HTR2A deficient moLcs as compared to IMQ- stimulated WT moLcs (Figure 5A). To validate it, we performed IL- 23p19 ELISA on stimulated and unstimulated, WT or HTR2A deficient moLcs and found IL- 23 levels to be increased significantly in stimulated HTR2A deficient moLcs as compared to stimulated WT moLcs (Figure 5B). We then determined IL- 23 secretion among different cell types in IMQ- treated- HTR2A deficient mice and the cells were gated as published (Figure 5Y). (28) We found that moLcs have the highest IL- 23 mean fluorescence intensity (MFI) and moLcs have the most IL- 23- producing cells among all the other cell types (Figure 5C). These data strongly suggest HTR2A deficient moLcs have increased secretion of IL- 23. + +We performed an in vivo experiment where we injected DOI or PBS intraperitoneally into WT mice and looked at inflammatory and IL- 23 levels (Figure 5D). We found mice injected with DOI had less inflammation, as seen in ear swelling, gross evaluation, and epidermal thickness (Figure 5E- 5G). Importantly, we found IL- 23- positive cells to decrease in the DOI- treated group as compared to the control group (Figure 5H). The percentage of IL- 23 positive cells among moLcs is lower in DOI- treated as compared to the control group, but it did not reach statistical significance (Figure 5I). Nevertheless, IL- 23 MFI of DOI- treated moLcs is significantly lower as compared to control moLcs (Figure 5J). + +<--- Page Split ---> + +To further validate IL- 23's role in exacerbating inflammation, we performed an in vivo IL- 23 neutralizing experiment (Figure 5K). In mice injected intradermally with isotype antibody, the inflammation was more severe in HTR2A deficient mice than in WT mice. The inflammation levels were minimal as observed in terms of ear swelling, visual inspection, histological assessment, and quantification of epidermal thickness in \(\alpha\) - IL23r- treated mice irrespective of HTR2A deficiency (Figure 5L- 5N). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells were reduced in \(\alpha\) - IL23r- treated mice in both WT and HTR2A deficient mice; however, neutrophil levels were not significantly different (Figure 5O). mRNA levels of pro- inflammatory cytokines were reduced in both WT and HTR2A deficient mice treated with neutralizing antibodies, with a reduction in IL17a levels reaching statistical significance (Figure 5P). With that, we conclude that IL- 23 plays a pivotal role in exacerbating psoriasisform inflammation in HTR2A- deficient moLCs. + +## HTR2A suppresses IL-23 secretion by inhibiting activation of non-canonical NFκB pathway in monocyte-derived Langerhans cells + +To determine the transcription factor involved in the exacerbated inflammation, we performed bulk RNA- sequencing on HTR2A- deficient moLCs isolated from IMQ- treated ears (GSE274941) and compared it with WT moLCs isolated from IMQ- treated ears as published previously (GSE222197). A total of 140 DEGs were identified, 47 genes were downregulated and 93 genes were upregulated in HTR2A deficient LCs (Figure 6A). The DEGs were analyzed by GAGE pathway analysis on TRRUST Transcription Factor Database and we found the NFkB1 pathway to be the top- ranked statistically significant transcription factor (Figure 6B). The key genes requiring NFkB1 as a transcription factor were listed on the heat map (Figure 6A). + +It has been reported that IL- 23a secretion involves both canonical and non- canonical NFκB pathways.(43) We performed Western blot on key molecules within the non- canonical NFκB pathway and found processing of p100 unit is increased in HTR2A deficient moLCs as p100 levels were low and p52 (the processed subunit of p100) is increased in HTR2A deficient moLCs. The addition of DOI to WT moLCs lead to a reduced level of p52 even though p100 levels were rather similar. There were, however, no changes to RelB levels in HTR2A deficient moLCs as compared to WT moLCs (Figure 6C). + +We then investigated the role of NFκB using the in vitro assays of IL- 23p19 ELISA and coculture of \(\mathrm{V} / 4\) T cell with moLCs, DOI, and antagonist of NFκB (EVP 4593).(44) We found IL- 23 levels to be lower in WT moLCs stimulated with Imiquimod than stimulated HTR2A deficient moLCs. When DOI, the HTR2A- specific agonist, and EVP 4593 were added into WT moLCs, IL- 23 levels were decreased (Figure 6D). Differentiation of \(\mathrm{V} / 4\) T cells into IL- 17- and IL- 22- producing cells was lower when cultured with stimulated WT moLCs than with stimulated HTR2A deficient moLCs. The addition of DOI or EVP 4593 into WT moLCs reduced the differentiation of cocultured \(\mathrm{V} / 4\) T cells when compared to stimulated WT moLCs only. (Figure 6E) In all, we found that HTR2A suppresses IL- 23 secretion involving the non- canonical NFκB pathway. + +## Serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation + +We then wondered what is the putative agonist of HTR2A in physiological conditions. Since serotonin is the known agonist of HTR2A, we wondered if there were any changes to their levels in psoriasis.(45) We performed immunohistochemistry (IHC) staining of serotonin on skin + +<--- Page Split ---> + +samples of healthy volunteers, non- lesioned skin, and lesioned skin of psoriatic patients. We found serotonin levels to be high in healthy volunteers and non- lesioned skin. Serotonin levels were low in lesioned skin (Figure 7A). We quantified the serotonin levels by determining its intensity in terms of H- score and found serotonin levels to be significantly higher in the skin of healthy volunteers than lesioned skin of psoriatic patients irrespective of their psoriasis area and severity index (PASI) score. The difference in serotonin levels of psoriatic skin between low PASI \(< 30\) and high PASI \((\geq 30)\) was not significant (Figure 7B). Serotonin levels were higher in non- lesioned skin than lesioned skin even though both are from the same patient (Figure 7C). + +To further validate serotonin's role as the putative agonist of HTR2A in psoriasis, we injected serotonin intraperitoneally into WT or HTR2A deficient mice (Figure 7D). In terms of ear swelling, visual inspection, and histological assessment, we found WT mice injected with serotonin to have lower inflammatory levels than WT control mice. On the other hand, HTR2A deficient mice injected with serotonin experienced exacerbated inflammation (Figure 7E- 7G). A similar trend was observed during the quantification of epidermal thickness (Figure 7G). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells were highest in HTR2A deficient mice injected with serotonin, followed by WT control mice and WT mice injected with serotonin. The differences were significant (Figure 7H). However, when we looked at neutrophil levels, mRNA levels of Il17a and Il23a, we found a similar trend but the differences were not significant (Figure 7H and 7I). The data here supports serotonin as the putative agonist of HTR2A attenuating psoriatic inflammation. + +We then wondered where does this serotonin originate from. Platelet is known to store peripheral serotonin, hence, we determined plasma and platelet serotonin levels of psoriatic patients and healthy controls.(46) We found no difference in plasma serotonin levels but found a significant difference in platelet serotonin levels of psoriatic patients and healthy controls (Figure S8A and S8B). We then used an established method to deplete platelet serotonin levels by pretreating the mice for 14 days with Fluoxetine (Figure S8C).(15) Mice treated with fluoxetine were found to have very low platelet serotonin levels (Figure S8D). In terms of ear swelling, visual inspection, histological assessment, and quantification of epidermal thickness, we found mice depleted of platelet serotonin have exacerbated inflammation (Figure S8E- S8G). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells, neutrophils, mRNA of Il17a and Il23a were higher in platelet serotonin- depleted mice but only changes in neutrophil levels and changes in mRNA levels of Il17a achieve significance (Figure S8H- S8I). In all, serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation and our preliminary data suggests it originates from platelet. + +## Monocyte-derived Langerhans cells of human psoriatic skin show higher IL23A and lower HTR2A expression + +To ensure our findings in mice are recapitulated in humans, we reanalyzed single cell data from publicly available datasets. We combined two different psoriatic datasets with healthy controls (GSE151177 and GSE162183). We identified 27 cell clusters in the combined dataset (Figure 8A). Importantly, we identified embryo Langerhans cells, monocyte- derived Langerhans cells, and dendritic cell type 3 (DC 3). The embryo Langerhans cell cluster has high Id2 and Runx3 (lineage- defining transcription factors for Langerhans cells) expression but low Irf8 (lineage- defining transcription factor for cells of monocyte lineage) expression.(39) MoLCs have high Irf8 expression and ITGAM (CD11b). Dendritic cell type 3 was identified by its expression of CD163 + +<--- Page Split ---> + +but not CD11b (Figure 8B).(47) We then looked at the expression levels of HTR2A between psoriatic molCs and control molCs and found HTR2A expression on psoriatic molCs to be significantly lower (Figure 8C). Among the cytokines, IL23A and IL1B are highly expressed in molCs (Figure 8D). Expression of NFκB- related subunits were increased in psoriatic molCs (Figure 8E). We then performed immunofluorescence staining of psoriatic lesioned and nonlesioned skin samples for CD1a and HTR2A. CD1a represents Langerhans cells and we found HTR2A expression to be lower in Langerhans cells of psoriatic lesioned skin as compared to nonlesioned skin (Figure 8F). Our findings in mice closely resemble clinical conditions, as HTR2A deficiency in molCs leads to exacerbated inflammation in mice mirrors low HTR2A expression in psoriatic molCs. IL- 23 was found to be a key mediator and NFκB is enriched in psoriatic molCs leading to exacerbated inflammation similar to HTR2A deficient molCs in mice. + +<--- Page Split ---> + +## DISCUSSION + +In our retrospective cohort study, we investigated how anti- psychotics, which antagonize dopamine and serotonin receptors, affect psoriatic outcomes.(48) Patients on anti- psychotics may have worsened psoriatic outcomes and these anti- psychotics have a very strong binding affinity for HTR2A. (16) This indicates that HTR2A antagonism by anti- psychotics worsens psoriatic outcomes. While this association is far from ideal, we confirmed the pivotal role played by HTR2A in regulating psoriatic inflammation by treating human psoriatic skin with an HTR2A- specific agonist ex vivo. Besides, our in vivo murine psoriasisform model further proved HTR2A's pivotal role in suppressing psoriatic inflammation. + +We also found chronic SSRI use led to worsening psoriatic outcomes. In congruent with our findings, there are reports showing patients who had taken fluoxetine (a commonly used SSRI) for extended periods experienced worsened psoriasis.(11, 13) In contrast to our findings, a retrospective study done in Sweden demonstrated an improved outcome with a decreased need for systemic psoriatic treatment following two SSRI dispensations within six months.(9) There is a lack of information on the length of treatment, dosage, and adherence reported in the Swedish study making comparisons between their study and ours impractical. Nevertheless, the conflicting findings could be due to the paradoxical effect SSRI has on serotonin levels. Since SSRI targets serotonin transporter (SERT), it could lead to a temporary increase of serotonin in between synaptic clefts, but chronic use will lead to depletion of platelet serotonin stores.(49- 52) The major source of serotonin in the periphery is in platelets.(53) While platelet serotonin depletion following chronic SSRI use is established in mice, much remains unknown about its impact on in humans. + +We demonstrated that HTR2A deficiency exacerbated psoriasisform inflammation. This is congruent with a previous study that showed hypermethylation of HTR2A, which decreases HTR2A expression, resulted in exacerbated inflammation in rheumatoid arthritis (a disease that involves type 17 immunity).(54, 55) We noticed increased IL- 17- and IL- 22- producing \(\mathrm{V} / 4\mathrm{T}\) cells in our murine psoriasisform model as expected since \(\mathrm{V} / 4\mathrm{T}\) cells are the dominant type 17 immune responders in acute models, particularly psoriatic model.(35, 56) Ear swelling and epidermal thickness which is associated with IL- 22 were also increased reflecting acanthosis commonly seen in psoriasis.(57) Increases in mRNA levels of Il17a and Il23a were observed suggesting the IL- 23/IL- 17 axis plays a key role in exacerbating inflammation. We also observed increased neutrophil levels as predicted, which may be due to neutrophil recruitment at lesioned sites under psoriatic conditions.(36) However, we did not observe any increase in IL- 17- and IL- 22- producing CD4 T cells, which may be due to two reasons: i) conventional CD4 T cells in the periphery will only express IL23r upon activation; and ii) IMQ is an acute model with no antigen limiting the expansion of antigen- specific Th17 cells.(58, 59) + +We delineated the cell type exacerbating inflammation in HTR2A deficient mice as it is crucial in unravelling the complex interplay between HTR2A and type 17 inflammation. HTR2A was found to have minimal impact on B and T cells, as seen in our conditional knockout mice with HTR2A deficient B and T cells. Nevertheless, we are aware our seven- day model will not be able to reveal and reflect the true extent of HTR2A's impact on these cells. Among the innate cells, it was established previously that monocyte- derived inflammatory Langerhans cells play a pivotal + +<--- Page Split ---> + +role in modulating psoriasisform inflammation.(28) In agreement with previous findings, we demonstrated that HTR2A deficiency in monocyte- derived Langerhans cells (molC) played a key role in exacerbating inflammation. We validated this using multiple models including Langerhans cells depletion, adoptive transfer, and two conditional knockout mice models. Our chimeric bone marrow transplant model further supports this, as HTR2A deficiency in hematopoietic cells, which includes molCs but not embryo Langerhans cells, exacerbates inflammation.(28) + +HTR2A affects the function of molCs and embryo Langerhans cells (elC) to varying extents. Under inflammatory conditions, monocytes are recruited in waves to the inflammatory site and they serve as end- type killer cells by secreting chemokines and cytokines, exacerbating the inflammation.(60) This means molCs inherently secrete more IL- 23 than elCs. HTR2A serves as a brake to inhibit cytokine secretion rather than inducing increased cytokine secretion in molCs. This may explain why HTR2A deficiency appears to affect molCs more than elCs. + +We identified molCs by their expression of MHC II, CD207, and CD64 markers. We found it hard to fit our molCs neatly into any of the P1 to P5 dermal macrophages identified by Tamoutounour et al.(61) Rather, our cells closely resemble the murine monocyte- derived macrophage identified by McGovern et al., reflecting heterogeneity in this population.(29) Importantly, molCs identified in our study have high expression of Irf8, a lineage- defining transcription factor for monocytes, and mutations in this gene will lead to genetic deficiency of monocytes and dendritic cells.(39, 62) The molCs also expressed Runx3 and Id2, lineage- defining transcription factors for Langerhans cells, as Langerhans cell development requires TGF- \(\beta\) , and Runx3 and Id2 are critical mediators downstream of TGF- \(\beta\) .(39, 63, 64) MolCs and elCs identified in the single cell RNA- sequencing data reflect this as they have high expression of Runx3 and Id2. The variation in Irf8 expression levels help us differentiate them into molCs and elCs. + +IL- 23 is a key cytokine in the pathogenesis of psoriasis and drugs targeting IL- 23, like guselkumab and risankizumab, are successful in treating psoriatic symptoms.(65) However, to date, no studies have reported any association between HTR2A and IL- 23 levels. We found IL- 23 secretion to be increased drastically in HTR2A deficient molCs and this is observed in both ELISA and intracellular cytokine staining of IL- 23. Stimulation of HTR2A with DOI, a HTR2A- specific agonist, led to a significant decrease in IL- 23 secretion by molCs further supporting the notion that HTR2A expression on molCs modulates psoriatic inflammation via IL- 23. Contradicting our study, Wohn et al. found that IL- 23 was exclusively produced by Langerin- negative DCs in vivo post IMQ- painting.(66) We believe this could be explained by differences in methodology, as we did not inject Brefeldin A into the mice directly; instead, we isolated the cells, treated them with PMA, Ionomycin, and GolgiStop (containing Brefeldin A) and found Langerhans cells, particularly molCs, secrete elevated levels of IL- 23. + +Further downstream, IL- 23 has been reported to induce innate IL- 17 production, especially among \(\gamma \delta\) T cells and IL- 23 has been known to play a cardinal role in mediating Imiquimod- induced psoriasisform inflammation.(67, 68) While IL- 1 \(\beta\) has also been proposed to play a role in inducing innate IL- 17 production, our qPCR of IL- 1 \(\beta\) did not reveal a drastic increase in IL- 1 \(\beta\) levels with HTR2A deficiency.(67) Besides, it has been reported that the stimulation of peritoneal exudate cells with IL- 23 alone but not IL- 1 \(\beta\) is sufficient to induce IL- 17 production by \(\gamma \delta\) T cells.(69) IL- 23 is known to play a central role in T cell differentiation instead of T cell survival and + +<--- Page Split ---> + +expansion.(70) Using our Vγ4 T cell coculture assay, we demonstrated increased Vγ4 T cell differentiation into IL- 17- and IL- 22- producing cells after coculture with molCs. This highlights that IL- 23 is the key cytokine mediating exacerbated inflammation in HTR2A deficient molCs. + +IL- 23 transcription requires NFκB and HTR2A has been suggested to affect NFκB levels.(20, 71, 72) The NFκB pathway is found to be enriched in HTR2A deficient Langerhans cells. It was previously reported both canonical NFκB and non- canonical NFκB pathways are involved in IL- 23 secretion. The canonical NFκB pathway is involved in early- phase induction, and the late- phase induction peaking at 12 hours, requires the non- canonical NFκB p100 subunit.(43) Our findings are congruent with previous findings since immunoblot of our cells post 24- hour treatment showed enrichment in the non- canonical NFκB pathway. Notably, we also found enriched NFκB pathways, both canonical and non- canonical, in psoriatic molCs as compared to control molCs in our reanalysis of single cell RNA- sequencing data. + +We found serotonin levels to be raised in healthy skin and non- lesioned skin samples of psoriatic patients as compared to lesioned skin via immunohistochemical staining. In contrast, there are two reports via immunohistochemical staining indicating serotonin levels were raised in psoriatic lesioned skin.(73, 74) The discrepancy could be due to the difference in controls used, as in their studies, they used the skin of healthy individuals as controls whereas we used both non- lesioned paired samples and healthy skins as controls. Our findings are further supported by our in vivo murine model, where serotonin injected intraperitoneally significantly inhibited type 17 inflammation. Congruent with our findings, a study reported serotonin modulates M2 macrophage polarization and another reported serotonin decreases Th1 and Th17 cytokines in multiple sclerosis patients.(75, 76) We further found platelet serotonin levels but not serum serotonin levels to be decreased in psoriatic patients. Using an in vivo model where we depleted platelet serotonin with chronic fluoxetine pre- treatment, our preliminary data showed platelet serotonin depletion led to exacerbated inflammation even though more experiments are needed to validate this finding.(15) Our data suggest serotonin through its receptor HTR2A plays a key role in modulating psoriatic inflammation. + +Our study provides evidence that drugs antagonizing HTR2A and HTR2A deficiency led to exacerbated psoriatic inflammation. On the other hand, the HTR2A- specific agonist of DOI attenuates inflammation. IL- 23 secretion by HTR2A deficient molCs was increased, as HTR2A functioning as brakes of the immune system via its interference of NFκB pathway, was absent in HTR2A deficient molCs. Serotonin is the putative agonist of HTR2A modulating inflammation and critically, our findings were recapitulated in clinical samples. Overall, our study provides insights into the role of HTR2A in regulating psoriatic inflammation, with the cells, cytokines, and transcription factor involved identified. Thus, our findings could lead to the development of novel therapeutic interventions, providing new avenues for treating psoriasis. + +## Limitations of the study + +Our study has been limited by our Imiquimod- induced psoriasisform murine model as it is an acute inflammatory model. Application of Imiquimod for more than 7 days will lead to resolution of psoriasisform inflammation even without external intervention.(77) Due to the short- term + +<--- Page Split ---> + +nature of our model, we cannot recapitulate the chronic component of psoriatic inflammation like the involvement of CD4 \(\alpha \beta\) T cells. + +<--- Page Split ---> + +## MATERIALS AND METHODS + +## Humans + +A retrospective cohort study was conducted to investigate the relationship between SSRI or antipsychotic use change in treatment received for psoriasis. This cohort study used population- based data from the Taiwan National Health Insurance (NHI) program. This program includes \(>99\%\) of Taiwanese residents and a total of 2 million patients were enrolled in this program. This study only used secondary data from this database for academic purposes. + +SSRI included in this study were citalopram, escitalopram, fluoxetine, paroxetine, sertraline, and fluvoxamine. Anti- psychotic included in this study were aripiprazole, chlorpromazine, prochlorperazine, quetiapine, risperidone, sulpiride, amisulpride, clozapine, haloperidol, olanzapine, and anti- psychotic categorized as Others (brexipiprazole, fluphenazine, loxapine, lurasidone, perphenazine, pimozide, thioridazine, and trifluoperazine). This study used Anatomical Therapeutic Chemical codes to identify SSRI or anti- psychotic exposure. + +Psoriatic cases were identified using the codes 696.1 and 696.8 in the International Classification of Diseases 9 (ICD9). The primary outcome is a change in the categories of psoriatic treatment received within 180 days from the index date (the date starting SSRI or anti- psychotic). Psoriatic treatments were divided into four categories as defined by the Taiwan Dermatological Association namely topical for the mildest symptoms, followed by systemic agents, combination therapy, and biologics for patients with the most severe symptoms. A worsened psoriatic outcome is defined as a change of treatment from treatment of mild symptoms to treatment of severe symptoms, for example from using systemic agents to using biologics. Improved psoriatic outcome is defined as a change of treatment from treatment for severe symptoms to treatment of mild symptoms, for example from biologics to using topical treatment. + +Data from 28,056 psoriatic subjects were included in this study and they were checked for antipsychotic or SSRI exposure. 9,570 subjects were controls, 11,315 subjects were on antipsychotics, and 7,171 subjects were on SSRIs. We then performed Cox proportional hazard model to calculate the hazard ratios and \(95\%\) confidence intervals for worsened psoriatic outcome and improved psoriatic outcomes, between anti- psychotic users versus controls and SSRI users versus controls. + +## Animals + +Htr2a floxed (Htr2aflox/flox) mice in C57BL/6J background were generated by the CRISPR/Cas9 technology. Selection of the sgRNA sequences followed the online resources, the sgRNA Designer: CRISPRko and Cas- OFFinder.(78, 79) The 5'loxP sequence was inserted in the upstream of exon 2, and 3'loxP in the 3'UTR after stop codon 134bp. The two sgRNA target sequences with PAM sites (NGG) were 5'- GTTTGATTGTGAGACATCGG - 3', and 5'- GTCCGGACAGCATTTGAACT - 3', for inserting the 5'-loxP and the 3'-loxP DNA fragments, respectively. sgRNA and Cas9 protein for electroporation were purchased from Synthego Corporation. Electroporation was performed on fertilized eggs from C57BL/6J mice. To minimize off- target effects in CRISPR mice, Htr2aflox/flox mice were first backcrossed to C57BL/6JNarl mice for 2 generations before breeding into cell- specific Htr2a knockout mice. Genotyping of founder mice was performed by PCR. PCR condition for genotyping of offspring was the initial denaturation at \(95^{\circ}C\) 5 min, followed by 45 + +<--- Page Split ---> + +cycles of \(95^{\circ}C\) for 30 sec, \(58^{\circ}C\) for 30 sec and \(72^{\circ}C\) for 30 sec, and the final extension at \(72^{\circ}C\) for 7 min. All techniques for the production of the Htr2aflox/flox mice were provided by the "Transgenic Mouse Models Core Facility of the National Core Facility for Biopharmaceuticals, National Science and Technology Council, Taiwan" and the "Gene Knockout Mouse Core Laboratory of National Taiwan University Center of Genomic and Precision Medicine". + +Others mice used: C57BL/6JNarl, 129S/Sv- Tg(Prm- cre)580g/J (Jax 003328), B6. SJL- Ptprca Pepcb/BoyJ (Jax 002014), C57BL/6- Tg(Rag1- RFP, - cre/ERT2)33Narl/Narl (RMRC 13191), STOCK Tg(CD207- cre/ERT2)1Dhka/J (Jax 028287), B6.12952- Cd207tm3.1(HBEGF/EGFP)Mal/J (Jax 016940), C57BL/6J- Ms4a3em2(cre)Fgnx/J (Jax 036382). + +For the generation of whole- body knockout mice, Prm- cre mice were crossed with Htr2aflox/flox mice. The double gene offspring were then crossed with wild type mice, producing heterozygous offspring. The heterozygous mice were then inbred with a 25 per cent probability of producing homozygous HTR2A deficient mice. For the generation of conditional knockout mice, Rag1- RFP- cre mice were crossed with Htr2aflox/flox mice. The offspring were then backcrossed with Htr2aflox/flox mice. The offspring that are Htr2aflox/flox mice- Rag1- RFP- cre (conditional knockout) will be bred with their littermate that are Htr2aflox/flox mice producing only Htr2aflox/flox mice- Rag1- RFP- cre (HTR2ARag1) or Htr2aflox/flox mice which serve as a littermate control. A similar breeding strategy was adopted for CD207- cre and Ms4a3em2(cre)Fgnx mice. + +Animals were housed in a specific pathogen- free environment in the Experimental Animal Facility at IBMS, Academia Sinica, Taipei, Taiwan. The Institutional Animal Care and Use Committee (IACUC) approved the animal experimentation protocols used in this study. Male and female mice used for this study were aged 8- 12 weeks. + +## RNA-seq of human samples + +Skin- punched biopsies were obtained from the lesioned sites of patients and healthy foreskin were collected from male patients undergoing circumcision. The skins were halved, with half untreated and the other half treated with 100nM of DOI for 16 hours. Total RNA was isolated with Direct- zol RNA Miniprep Kit (Zymo Research) according to the manufacturer's instructions. NGS was performed by NGS High Throughput Genomics Core at Biodiversity Research Centre, Academia Sinica, Taipei, Taiwan. + +## In silico analysis of HTR2A gene expression + +We used the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) to search for HTR2A gene expression in skin tissue between lesioned sites versus non- lesioned sites, as well as psoriasis patients versus healthy controls. The GSE13355 includes Affymetrix HU133 Plus 2.0 microarrays for lesion skin tissues and non- lesion skin tissues from 58 psoriasis patients. (80) GSE34248 includes Affymetrix HU133 Plus 2.0 microarrays for lesion skin tissues and non- lesion skin tissues from 13 psoriasis patients. (81) GSE109248 includes Illumina HumanHT- 12 V4.0 for skin tissues from 17 psoriasis patients and 12 healthy controls. (82) GSE41664 includes Affymetrix Human Genome U133 Plus 2.0 Array of lesioned and non- lesioned skin tissues of 38 psoriasis patients. (81) The differences in gene expression levels between the two groups were calculated by a two- tailed paired Student's T- test. + +<--- Page Split ---> + +Imiquimod psoriasiform model + +For mice to develop psoriasis, we applied 15mg of \(5\%\) Imiquimod (Aldara cream, iNova Pharmaceuticals, Australia) for 7 days on the ears of mice. The ears were harvested the day after. + +## Flow cytometry + +Cells were isolated from the ears by incubating overnight at \(37^{\circ}C\) with Dispase II (Thermo Scientific). The next day, the ear is then cut by scissors into small pieces and then immersed in RPMI (Gibco) containing Collagenase IV (Gibco) and DNase I (Bio Basic) for 90 minutes. The sample is then filtered using a \(70\mu \mathrm{m}\) cell strainer. The isolated single cells were then stained directly with fluorescent antibodies or treated with PMA (Merck), lonomycin (BioGems), and GolgiStop (BD) if intracellular cytokine staining was required. Cells were incubated with LIVE/DEAD Fixable Blue Dead Cell Stain Kit (Thermo Fisher Scientific) for the exclusion of dead cells. The Fc- blocking antibody was then used to prevent non- specific binding followed by incubation with specific antibodies against surface antigens. For intracellular cytokine staining, cells were stimulated with PMA (Merck) and lonomycin (BioGems) in the presence of GolgiStop (BD) at \(37^{\circ}C\) for four hours. The cells were fixed and permeabilized with Cytofix/Cytoperm buffer (BD) according to the manufacturer's instructions. For all samples, acquisition was performed on LSRI (BD). + +RNA purification and real- time quantitative PCR (qPCR) + +Total RNA from tissue was immersed in TRIzol (Invitrogen, 15596026) overnight, homogenized using steel beads together with Bullet Blender Tissue Homogenizer (Next Advance), and total RNA was extracted using RNA MiniPrep Kit (Zymo Research, R2050) according to manufacturer's instructions. The quantity of the total RNA was measured by NanoDrop Spectrophotometer (Thermo Scientific). For qPCR analysis, equivalent amounts of RNA were reverse- transcribed by Maxima H Minus First Strand cDNA Synthesis Kit (Thermo Scientific, K1652). cDNAs were mixed with indicated primers and PowerUp SYBR Green Master Mix (Applied Biosystems, A25742), and RT- qPCR was performed on Quant Studio 5 Real- time PCR System (Applied Biosystems). cDNAs were normalized to equal amounts using primers against Gapdh. The primer sequences can be found in the supplementary table. + +Chimeric bone marrow mice + +CD45.1 mice (Jax 002014) were fed with water added with Trimethoprim and Sulfamethoxazole (Merck, SML3191, with a final concentration \(1.92\mathrm{mg / ml}\) ) for two weeks before they were lethally irradiated (1200 rads) in split dose, with three hours of rest in between, one day before the experiment. Bone marrow from HTR2A deficient mice was harvested on the day of the experiment. 5mL of ACK Lysing Buffer (Gibco, A1049201) was added to the cells for five minutes to lyse red blood cells. The cells were then washed with Phosphate Buffered Saline and quantified. A total of \(5\times 10^{5}\) bone marrow cells were intravenously injected into the lethally irradiated mice. The mice were continuously fed antibiotic- laced water for another 14 days before experiments were carried out on them 12 weeks after transplantation. We also transplanted the bone marrow of CD45.1 mice to HTR2A deficient mice and the bone marrow of HTR2A deficient mice to HTR2A deficient mice. + +Culture of monocyte- derived Langerhans cells + +Bone marrow cells were harvested from the femur and tibia of wild type or HTR2A deficient mice. The cells were lysed with ACK Lysing Buffer (Gibco, A1049201). The cells were then resuspended in RPMI (Gibco) containing \(20\mathrm{ng / ml}\) of recombinant mouse GM- CSF (PeproTech, + +<--- Page Split ---> + +315- 03), 5ng/ml of TGF- \(\beta\) (BioLegend, 763104), and 8ng/ml of IL- 34 (BioLegend, 577604), plated on a 6 well plate, and incubated for 8 days at \(37^{\circ}C\) . The medium was changed every 48 hours. Monocyte- derived Langerhans cells were isolated by a FACSAria IIIu Sorter with \(\alpha\) - CD207, \(\alpha\) - CD64, and \(\alpha\) - MHC II staining. LCs were stimulated with \(10\mu \mathrm{g / mL}\) Imiquimod (Enzo, ALX- 420- 039- M100) and \(10\mu \mathrm{g / ml}\) of DOI (Merck, D101) or \(0.1\mu \mathrm{M}\) of EVP 4593 (MCE, HY- 13812) were added for 24 hours to investigate IL- 23 expression and its ability to induce differentiation of \(\mathrm{V}\gamma 4\mathrm{T}\) cells in a coculture assay. + +## ELISA assay + +ELISA assayIL- 23p19 expression was measured using Mouse IL- 23p19 ELISA Kit (Elabscience, E- EL- M0731) according to the manufacturer's instructions. Mouse and human serotonin levels were measured using a Serotonin ELISA Kit (Enzo, ADI- 900- 175) according to the manufacturer's protocol. The absorbance at \(450\mathrm{nm}\) (Mouse IL- 23 ELISA kit) or \(405\mathrm{nm}\) (Serotonin ELISA Kit) was measured with an ELISA reader (Infinite M1000 PRO, Tecan). + +## Coculture assay + +Spleens of wild type mice were surgically removed and mechanically disrupted by meshing it against a \(70\mu \mathrm{m}\) cell strainer (Jet Biofil, CSS013070). ACK Lysing Buffer (Gibco, A1049201) was added to lyse red blood cells. The cells were then isolated by a FACSAria IIIu Sorter with \(\alpha\) - CD45, \(\alpha\) - CD3, \(\alpha\) - \(\gamma \delta\) , and \(\alpha\) - \(\mathrm{V}\gamma 4\) staining. Monocyte- derived Langerhans cells post- stimulation with DOI or EVP 4593, were incubated with these \(\mathrm{V}\gamma 4\mathrm{T}\) cells for 72 hours in an \(\alpha\) - CD3- coated plate. \(\alpha\) - CD28 was then added to further stimulate \(\mathrm{V}\gamma 4\mathrm{T}\) cells. Post- 72 hours, the cells were then stimulated with PMA, and lonomycin, and incubated with GolgiStop for four hours. The cells were then stained for IL- 17A and IL- 22. The acquisition was performed on LSRII (BD). + +## RNA-seq of mouse samples + +The experiments generating datasets of GSE222197 and GSE274941 were done in parallel. MolCs (defined as \(\mathrm{CD45 + MHCII + CD11c + CD207 + CD64 + }\) ) were pooled from the ears of five mice treated with IMQ for two days, constituting a single biological sample. Cells were sorted and total RNA was isolated with a Direct- zol RNA Microprep Kit (Zymo Research) according to the manufacturer's instructions. NGS was performed by NGS High Throughput Genomics Core at Biodiversity Research Centre, Academia Sinica, Taipei, Taiwan. + +Protein gel electrophoresis and immunoblotting + +Protein gel electrophoresis and immunoblottingTotal cells lysates were prepared using RIPA (20mM Tris HCl, pH8.0, 150mM NaCl, 1mM EDTA, 1mM EGTA, 1% TritonX- 100, 0.5% deoxycholate, 0.1% SDS) supplemented with protease inhibitors (Halt Protease Inhibitor, Thermo Scientific, 78430), or used directly with 1X protein sample loading buffer (50 mM Tris- HCl, pH 6.8, 10% glycerol, 2% SDS, 0.1 M DTT, 0.04% Orange G). The protein concentrations of clarified supernatants were measured by a BCA Protein Assay Kit (Pierce, 23227). Proteins in lysates were separated by Bolt 4- 12% Bis- Tris Plus gels (Invitrogen) and transferred to a PVDF membrane (Thermo Scientific, 88520) by a Mini Trans- Blot cell system (BioRad). Transferred membranes were blocked for 1 h in 5% Bovine Serum Albumin (Merck, A7030) diluted in TBS (20 mM Tris- HCl, pH 7.4, 150 mM NaCl) and then incubated with primary antibodies in TBS with 0.1% Tween- 20 (TBST) overnight at 4°C. Primary antibodies used were as the key resources table. After washing with TBST, membranes were incubated with secondary antibodies in 3% BSA in TBST for one hour at room temperature. A secondary anti- mouse antibody conjugated with HRP (1:10000, Thermo Scientific, 31430) or anti- rabbit antibody conjugated with HRP (1:10000, Thermo Scientific, 31460) was used. Membranes were + +<--- Page Split ---> + +washed in TBST and then incubated in SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Scientific, 34580) for 2 minutes. Membranes were immediately imaged on UVP ChemStudio Western Blot Imaging System (Analytik Jena, 849- 97- 0847- 03). + +Immunostaining of section + +Paraffin- embedded tissue sections were deparaffinized with xylene and rehydrated with serial passage through changes of graded ethanol. Slides were subjected to heat- induced epitope retrieval in EDTA solution (10 mM Tris, 1 mM EDTA, \(0.05\%\) Tween 20, pH 8) followed by blocking with \(5\%\) goat serum or \(3\% \mathrm{H}_2\mathrm{O}_2\) . Tissues were incubated with primary \(\alpha\) - serotonin (YC5/45, Abcam) Abs at \(4^{\circ}C\) overnight. Slides were washed and incubated with HRP- carried Abs (Leadgene) for 1 hour at room temperature. In the immunohistochemistry assay, DAB (Roche, 11718096001) was used as a chromogen to visualize peroxidase activity, and the reaction was stopped by the addition of PBS. The preparations were lightly counterstained with hematoxylin (Sigma, H3136), mounted with Fluoromount Aqueous mounting medium (Sigma, F4680), and examined by Pannoramic 250 FLASH II (3DHistech). + +Quantification and statistical analysis + +FlowJo v10 was used to quantify flow cytometry data, and GraphPad Prism 9 was used for statistical analysis. Student's T- test was used to analyze experiments with two conditions, Student's paired T- test was used to analyze experiments with paired samples, one- way ANOVA was used for experiments with three or more conditions, and two- way ANOVA was used for changes in ear swellings. Statistical details of experiments, including the number of animals, and corresponding analyses are reported in the figure legends. + +## LIST OF SUPPLEMENTARY MATERIALS + +Figure S1. Anti- psychotic and SSRI use worsen psoriatic outcomes. + +Figure S2. Gating strategy for flow cytometry of \(\mathrm{V} / \gamma 4\mathrm{T}\) cells and neutrophils. + +Figure S3. The composition of cells changed after bone marrow transplant. + +Figure S4. HTR2A deficiency in adaptive immune cells does not exacerbate inflammation. + +Figure S5. Depletion of different Langerhans cell subsets using varying doses of diphtheria toxin. + +Figure S6. Characterization of monocyte- derived Langerhans cells developed from bone marrow. + +Figure S7. Gating strategy for identifying various cell clusters after spectral flow cytometry. + +Figure S8. Platelet serotonin is lower in psoriatic patients and its deficiency is associated with exacerbated inflammation. + +<--- Page Split ---> + +## REFERENCES AND NOTES + +1. A. Rendon, K. Schakel, Psoriasis Pathogenesis and Treatment. Int J Mol Sci 20, (2019). +2. A. Di Cesare, P. Di Meglio, F. O. Nestle, The IL-23/Th17 axis in the immunopathogenesis of psoriasis. J Invest Dermatol 129, 1339-1350 (2009). +3. E. Rieder, F. Tausk, Psoriasis, a model of dermatologic psychosomatic disease: psychiatric implications and treatments. Int J Dermatol 51, 12-26 (2012). +4. Eskeland S., Halvorsen J. A., T. L., Antidepressants have Anti-inflammatory Effects that may be Relevant to Dermatology: A Systematic Review. Acta Derm Venereol 97, 897-905 (2017). +5. C. W. Kwon, R. G. Fried, Y. Nousari, C. Ritchlin, F. Tausk, Psoriasis: Psychosomatic, somatopsychic, or both? Clin Dermatol 36, 698-703 (2018). +6. I. M. Michalek, B. Loring, S. M. John, A systematic review of worldwide epidemiology of psoriasis. J Eur Acad Dermatol Venereol 31, 205-212 (2017). +7. L. Rousset, B. Halioua, Stress and psoriasis. Int J Dermatol 57, 1165-1172 (2018). +8. S. Luis Blay, Depression and psoriasis comorbidity. Treatment with paroxetine: two case reports. Ann Clin Psychiatry 18, 271-272 (2006). +9. K. Thorslund, T. Svensson, K. Nordlind, A. Ekbom, C. M. Fored, Use of serotonin reuptake inhibitors in patients with psoriasis is associated with a decreased need for systemic psoriasis treatment: a population-based cohort study. J Intern Med 274, 281-287 (2013). +10. Y. M. Tzeng et al., Protective Effects of Anti-depressants against the Subsequent Development of Psoriasis in Patients with Major Depressive Disorder: a Cohort Study. J Affect Disord 281, 590-596 (2021). +11. R. J. Hemlock C, Winston A, Fluoxetine-induced psoriasis. Annals of Pharmacotherapy 26, 211-212 (1992). +12. S. F. Osborne, L. Stafford, K. G. D. Orr, Paroxetine-Associated Psoriasis. The American Journal of Psychiatry, (2002). +13. L. Tan P.L., S. K. Kwek, Onset of psoriasis during therapy with fluoxetine. Gen Hosp Psychiatry 32, 446 e449-446 e410 (2010). +14. J. Lyne, B. Kelly, W. O. Connor, Schizophrenia: a review of neuropharmacology. Irish Journal of Medical Science 173, 155-159 (2004). +15. Mohamad F. Masri, Chinmay K. Mantri, Abhay P. S. Rathore, A. L. S. John, Peripheral serotonin causes dengue virus-induced thrombocytopenia through 5HT2A receptors. Blood 133, 2325-2337 (2019). +16. J. Masson, M. B. Emerit, M. Hamon, M. Darmon, Serotonergic signaling: multiple effectors and pleiotropic effects. Wiley Interdisciplinary Reviews: Membrane Transport and Signaling 1, 685-713 (2012). +17. K. Kim et al., Structure of a Hallucinogen-Activated Gq-Coupled 5-HT(2A) Serotonin Receptor. Cell 182, 1574-1588 e1519 (2020). +18. J. F. Lopez-Gimenez, J. Gonzalez-Maeso, Hallucinogens and Serotonin 5-HT(2A) Receptor-Mediated Signaling Pathways. Curr Top Behav Neurosci 36, 45-73 (2018). +19. M. Pelletier, R. M. Siegel, Wishing away inflammation? New links between serotonin and TNF signaling. Mol Interv 9, 299-301 (2009). +20. B. Yu et al., Serotonin 5-hydroxytryptamine(2A) receptor activation suppresses tumor necrosis factor-alpha-induced inflammation with extraordinary potency. J Pharmacol Exp Ther 327, 316-323 (2008). +21. F. Nau, Jr., B. Yu, D. Martin, C. D. Nichols, Serotonin 5-HT2A receptor activation blocks TNF-alpha mediated inflammation in vivo. PLoS One 8, e75426 (2013). + +<--- Page Split ---> + +22. C. Y. Yeh et al., PD-L1 Enhanced by cis-Urocanic Acid on Langerhans Cells Inhibits Vgamma4(+) gammadeltaT17 Cells in Imiquimod-Induced Skin Inflammation. J Invest Dermatol 143, 1449-1460 (2023).23. B. Kirby, Langerhans cells in psoriasis: getting to the core of the disease. Br J Dermatol 178, 1240 (2018).24. M. Otsuka, G. Egawa, K. Kabashima, Uncovering the Mysteries of Langerhans Cells, Inflammatory Dendritic Epidermal Cells, and Monocyte-Derived Langerhans Cell-Like Cells in the Epidermis. Front Immunol 9, 1768 (2018).25. L. Zhou, A. Jiang, J. Veenstra, D. M. Ozog, Q. S. Mi, The Roles of Skin Langerhans Cells in Immune Tolerance and Cancer Immunity. Vaccines (Basel) 10, (2022).26. M. Collin, P. Milne, Langerhans cell origin and regulation. Curr Opin Hematol 23, 28-35 (2016).27. C. Langlet et al., CD64 expression distinguishes monocyte-derived and conventional dendritic cells and reveals their distinct role during intramuscular immunization. J Immunol 188, 1751-1760 (2012).28. T. P. Singh et al., Monocyte-derived inflammatory Langerhans cells and dermal dendritic cells mediate psoriasis-like inflammation. Nat Commun 7, 13581 (2016).29. N. McGovern et al., Human dermal CD14(+) cells are a transient population of monocyte-derived macrophages. Immunity 41, 465-477 (2014).30. Consensus Statement on Management of Psoriasis (2017).31. BL Roth, WK Kroeze, S Patel, E. Lopez, The Multiplicity of Serotonin Receptors: Uselessly diverse molecules or an embarrassment of riches? The Neuroscientist 6, 252-262 (2000).32. T. P. Singh, H. H. Zhang, S. T. Hwang, J. M. Farber, IL-23- and Imiquimod-Induced Models of Experimental Psoriasis in Mice. Curr Protoc Immunol 125, e71 (2019).33. Y. Cai et al., Pivotal role of dermal IL-17-producing gammadelta T cells in skin inflammation. Immunity 35, 596-610 (2011).34. R. Castillo-Gonzalez, D. Cibrian, F. Sanchez-Madrid, Dissecting the complexity of gammadelta T-cell subsets in skin homeostasis, inflammation, and malignancy. J Allergy Clin Immunol 147, 2030-2042 (2021).35. S. Pantelyushin et al., Roryt+ innate lymphocytes and gammadelta T cells initiate psoriasisform plaque formation in mice. J Clin Invest 122, 2252-2256 (2012).36. C. C. Chiang, W. J. Cheng, M. Korinek, C. Y. Lin, T. L. Hwang, Neutrophils in Psoriasis. Front Immunol 10, 2376 (2019).37. M. P. McCormack, A. Forster, L. Drynan, R. Pannell, T. H. Rabbitts, The LMO2 T-cell oncogene is activated via chromosomal translocations or retroviral insertion during gene therapy but has no mandatory role in normal T-cell development. Mol Cell Biol 23, 9003-9013 (2003).38. D. H. Kaplan et al., Autocrine/paracrine TGFbeta1 is required for the development of epidermal Langerhans cells. J Exp Med 204, 2545-2552 (2007).39. W. H. C. Ferrer I. R., Henderson S., Ushakov D. S., Santos e Sousa P., Strid J., Chakraverty R., Yates A. J., Bennett C. L., A wave of monocytes is recruited to replenish the long term LC network after immune injury. Science Immunology 4, eaax8704 (2019).40. L. Bellmann et al., Notch-Mediated Generation of Monocyte-Derived Langerhans Cells: Phenotype and Function. J Invest Dermatol 141, 84-94 e86 (2021).41. Z. Liu et al., Fate Mapping via Ms4a3-Expression History Traces Monocyte-Derived Cells. Cell 178, 1509-1525 e1519 (2019). + +<--- Page Split ---> + +42. M. Nakagawa et al., IL-12 secretion by Langerhans cells stimulated with Candida skin test reagent is mediated by dectin-1 in some healthy individuals. Cytokine 65, 202-209 (2014). +43. X. Ma, T. Aoki, S. Narumiya, Prostaglandin E2-EP4 signaling persistently amplifies CD40-mediated induction of IL-23 p19 expression through canonical and non-canonical NF-kappaB pathways. Cell Mol Immunol 13, 240-250 (2016). +44. K. Li et al., Glutamine Metabolism Underlies the Functional Similarity of T Cells between Nile Tilapia and Tetrapod. Adv Sci (Weinh) 10, e2201164 (2023). +45. Bhattacharyya S., Puri S., Miledi R., P. M. M., Internalization and recycling of 5-HT2A receptors activated by serotonin and protein kinase C-mediated mechanisms. Proceedings of the National Academy of Sciences 99, 14470-14475 (2002). +46. Charles P. Mercado, F. Kilic, Molecular Mechanisms of SERT in Platelets: Regulation of Plasma Serotonin Levels. Molecular Interventions 10, 231-241 (2010). +47. C. A. Dutertre et al., Single-Cell Analysis of Human Mononuclear Phagocytes Reveals Subset-Defining Markers and Identifies Circulating Inflammatory Dendritic Cells. Immunity 51, 573-589 e578 (2019). +48. L. S. Chokhawala K. (StatPearls Publishing, Treasure Island (FL), 2024 Jan), vol. 2024. +49. Z. Zhou et al., Antidepressant specificity of serotonin transporter suggested by three LeuT-SSR structures. Nat Struct Mol Biol 16, 652-657 (2009). +50. D. Duerschmied et al., Platelet serotonin promotes the recruitment of neutrophils to sites of acute inflammation in mice. Blood 121, 1008-1015 (2013). +51. M. Rami et al., Chronic Intake of the Selective Serotonin Reuptake Inhibitor Fluoxetine Enhances Atherosclerosis. Arterioscler Thromb Vasc Biol 38, 1007-1019 (2018). +52. Jean-Claude Alvarez, Martin Sanceaume, Charles Advenier, O. Spreux-Varoquaux, Differential changes in brain and platelet 5-HT concentrations after steady-state achievement and repeated administration of antidepressant drugs in mice. European Neuropsychopharmacology 10, 31-36 (1999). +53. D. Keszthelyi, F. J. Troost, A. A. Masclee, Understanding the role of tryptophan and serotonin metabolism in gastrointestinal function. Neurogastroenterol Motil 21, 1239-1249 (2009). +54. B. W. Kirkham, A. Kavanaugh, K. Reich, Interleukin-17A: a unique pathway in immunemediated diseases: psoriasis, psoriatic arthritis and rheumatoid arthritis. Immunology 141, 133-142 (2014). +55. J. Zhao et al., Circulating methylation level of HTR2A is associated with inflammation and disease activity in rheumatoid arthritis. Front Immunol 13, 1054451 (2022). +56. Y. Li et al., Vg4 gd T Cells Provide an Early Source of IL-17A and Accelerate Skin Graft Rejection. J Invest Dermatol 137, 2513-2522 (2017). +57. Y. Zheng et al., Interleukin-22, a T(H)17 cytokine, mediates IL-23-induced dermal inflammation and acanthosis. Nature 445, 648-651 (2007). +58. H. Li et al., IL-23 promotes TCR-mediated negative selection of thymocytes through the upregulation of IL-23 receptor and RORgammat. Nat Commun 5, 4259 (2014). +59. M. Jabeen et al., Advanced Characterization of Imiquimod-Induced Psoriasis-Like Mouse Model. Pharmaceutics 12, (2020). +60. P. Italiani, D. Boraschi, From Monocytes to M1/M2 Macrophages: Phenotypical vs. Functional Differentiation. Front Immunol 5, 514 (2014). +61. S. Tamoutounour et al., Origins and functional specialization of macrophages and of conventional and monocyte-derived dendritic cells in mouse skin. Immunity 39, 925-938 (2013). + +<--- Page Split ---> + +62. S. Hambleton et al., IRF8 Mutations and Human Dendritic-Cell Immunodeficiency. New England Journal of Medicine 365, 127-138 (2011). + +63. L. X. Heinz et al., Differential involvement of PU.1 and Id2 downstream of TGF-beta1 during Langerhans-cell commitment. Blood 107, 1445-1453 (2006). + +64. O. Fainaru et al., Runx3 regulates mouse TGF-beta-mediated dendritic cell function and its absence results in airway inflammation. EMBO J 23, 969-979 (2004). + +65. P. Hu et al., The Role of Helper T Cells in Psoriasis. Front Immunol 12, 788940 (2021). + +66. C. Wohn et al., Langerin(neg) conventional dendritic cells produce IL-23 to drive psoriatic plaque formation in mice. Proc Natl Acad Sci U S A 110, 10723-10728 (2013). + +67. C. E. Sutton et al., Interleukin-1 and IL-23 induce innate IL-17 production from gammadelta T cells, amplifying Th17 responses and autoimmunity. Immunity 31, 331-341 (2009). + +68. W. C. Chen et al., IL-23/IL-17 immune axis mediates the imiquimod-induced psoriatic inflammation by activating ACT1/TRAF6/TAK1/NF-kappaB pathway in macrophages and keratinocytes. Kaohsiung J Med Sci 39, 789-800 (2023). + +69. K. Shibata, H. Yamada, H. Hara, K. Kishihara, Y. Yoshikai, Resident Vdelta1+ gammadelta T cells control early infiltration of neutrophils after Escherichia coli infection via IL-17 production. J Immunol 178, 4466-4472 (2007). + +70. P. Thakker et al., IL-23 is critical in the induction but not in the effector phase of experimental autoimmune encephalomyelitis. J Immunol 178, 2589-2598 (2007). + +71. S. Mise-Omata et al., A proximal kappaB site in the IL-23 p19 promoter is responsible for RelA- and c-Rel-dependent transcription. J Immunol 179, 6596-6603 (2007). + +72. R. J. Carmody, Q. Ruan, H. C. Liou, Y. H. Chen, Essential roles of c-Rel in TLR-induced IL-23 p19 gene expression in dendritic cells. J Immunol 178, 186-191 (2007). + +73. J. Huang, G. Li, J. Xiang, D. Yin, R. Chi, Immunohistochemical study of serotonin in lesions of psoriasis. Int J Dermatol 43, 408-411 (2004). + +74. S. F. Younes, O. A. Bakry, Immunohistochemical Evaluation of Role of Serotonin in Pathogenesis of Psoriasis. J Clin Diagn Res 10, EC05-EC09 (2016). + +75. M. de las Casas-Engel et al., Serotonin skews human macrophage polarization through HTR2B and HTR7. J Immunol 190, 2301-2310 (2013). + +76. P. M. Sacramento et al., Serotonin decreases the production of Th1/Th17 cytokines and elevates the frequency of regulatory CD4(+) T-cell subsets in multiple sclerosis patients. Eur J Immunol 48, 1376-1388 (2018). + +77. X. Wang, J. Sun, J. Hu, IMQ Induced K14-VEGF Mouse: A Stable and Long-Term Mouse Model of Psoriasis-Like Inflammation. PLoS One 10, e0145498 (2015). + +78. J. G. Doench et al., Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. Nat Biotechnol 34, 184-191 (2016). + +79. S. Bae, J. Park, J. S. Kim, Cas-OFFinder: a fast and versatile algorithm that searches for potential off-target sites of Cas9 RNA-guided endonucleases. Bioinformatics 30, 1473-1475 (2014). + +80. R. P. Nair et al., Genome-wide scan reveals association of psoriasis with IL-23 and NF-kappaB pathways. Nat Genet 41, 199-204 (2009). + +81. J. Bigler, H. A. Rand, K. Kerkof, M. Timur, C. B. Russell, Cross-study homogeneity of psoriasis gene expression in skin across a large expression range. PLoS One 8, e52242 (2013). + +82. P. Mande et al., Fas ligand promotes an inducible TLR-dependent model of cutaneous lupus-like inflammation. J Clin Invest 128, 2966-2978 (2018). + +<--- Page Split ---> + +## ACKNOWLEDGEMENTS + +We thank the Academia Sinica Core Facility and Innovative Instrument Project (AS- CFII108- 113) for providing cell sorting services. We also thank the Light Microscopy Core Facility and Pathology Core Facility of the Institute of Biomedical Sciences, Academia Sinica for their technical assistance. We would also like to thank the technical services provided by the Transgenic Mouse Model Core Facility of the National Core Facility for Biopharmaceuticals, The National Science and Technology Council (NSTC), Taiwan and the Animal Resources Laboratory of National Taiwan University Centers of Genomic and Precision Medicine. + +## FUNDING + +This study was supported by grants UN108- 015 and UN110- 032 from National Taiwan University Hospital; and the Institute of Biomedical Sciences in Academia Sinica. + +## AUTHOR CONTRIBUTIONS + +Conceptualization: YFT, YLL; Methodology: YFT; Investigation: YFT, SYH; Visualization: YFT; Funding Acquisition: TFT, YLL; Project Administration: YLL; Resources: CYY, SSH, PCC, HJW, TFT; Software: YFT, CHL, CHT; Supervision: YLL; Writing- Original Draft: YFT; Writing- Review and Editing: YFT, YLL. + +## COMPETING INTERESTS + +TFT has conducted clinical trials or received honoraria for serving as a consultant for AbbVie, AnaptysBio, Bristol- Myers Squibb, Boehringer Ingelheim, Celgene, Eli Lilly, Galderma, GlaxoSmithKline- Stiefel, Janssen- Cilag, Leo- Pharma, Merck, Novartis, PharmaEssentia, Pfizer, Sanofi, Sun Pharma and UCB. The remaining authors state no conflict of interest. + +DATA AND MATERIALS AVAILABILITY: All data, code, and materials used in the analysis are available upon request. Mice are available upon request and subject to materials transfer agreements (MTAs). + +<--- Page Split ---> + +## FIGURES + +![](images/Figure_1.jpg) + +
Figure 1
+ +Figure 1. Modulation and expression levels of HTR2A affect psoriasis. A) Kaplan- Meier plot of anti- psychotic use and worsening psoriatic outcome. B) Forrest plot showing each anti- psychotic drug and its respective hazard ratio on worsening psoriatic outcome. C) Table showing ranking of anti- psychotic drug's affinity for HTR2A among serotonin receptors. D) Schematic depicting the experimental plan of ex vivo experiment from human samples. E) Venn diagram showing the number of genes overlapping between differentially expressed genes of healthy non- treated versus lesioned non- treated and differentially expressed genes of lesioned non- treated versus lesioned DOI- treated. F) Heat map showing 107 overlapping genes and their expression levels from E) (n=3). G) Box plot showing HTR2A expression levels in psoriatic cases as compared to control in publicly available GEO datasets. Data are a summary of two independent experiments (E- F). p values determined by Cox proportional hazard test (A) and unpaired Student's T- test (G). Mean \(\pm\) SEM (B, and G). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.0001. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. HTR2A deficiency in mice and hematopoietic stem cells exacerbates psoriasisinflammation. A) Schematic showing the experimental plan for Imiquimod-treatment of mice. B) Line chart showing changes in ear swelling of mice. (n=4 mice for Vas groups, n=10 mice for IMQ groups) C) Pictures showing ears of mice of various groups on Day 7. D) H&E-stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ \(\mathsf{V}_7 / 4\mathsf{T}\) cells and neutrophils in the respective groups with quantifications. The cells were isolated from the ears of mice. The detailed method is in Material and Methods section. F) qPCR of pro-inflammatory cytokines of respective groups. G) Schematic showing the experimental plan for developing bone marrow chimeric mice. H) Line chart showing changes in ear swelling of mice. (n=5 to 7 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E-stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ \(\mathsf{V}_7 / 4\mathsf{T}\) cells and neutrophils in the respective groups with quantifications. L) qPCR of pro-inflammatory cytokines of respective groups. Data are representative of three independent experiments (B-F and H-L). p values determined by two-way ANOVA (B and H) followed by Tukey's post-hoc test and one-way ANOVA (D-F and J-L) followed by Tukey's post-hoc test. Mean ± SEM (B, D-F, H, and J-L). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001; ns, non-significant.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Figure 3. HTR2A deficiency in Langerhans cells recapitulates exacerbated inflammation observed in HTR2A deficient mice. A) Schematic showing strategy for developing conditional knockout mice with HTR2A knocked out in Langerhans cells. B) Line chart showing changes in ear swelling of mice. (n=3 for Vas, n=6 for IMQ per group) C) Pictures showing ears of mice of various groups on Day 7. D) H&E- stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. F) qPCR of pro-inflammatory cytokines of respective groups. G) Schematic showing treatment schedule on CD207- DTR mice for the depletion of different subsets of Langerhans cells. H) Line chart showing changes in ear swelling of mice. (n=7 to 8 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E- stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. L) qPCR of pro- inflammatory cytokines of respective groups. Data are representative of two independent experiments (B- F and H- L). p values determined by two- way ANOVA (B and H) followed by Tukey's post- hoc test, one- way ANOVA (D- F) followed by Tukey's post- hoc test and Student's T- test (J- L). Mean ± SEM (B- F and H- L). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Figure 4. HTR2A deficiency in Langerhans cells of monocyte lineage results in exacerbated inflammation. A) Schematic showing the experimental design for developing moLCs and adoptively transferring them into Langerhans cells depleted mice. B) Line chart showing changes in ear swelling of mice. (n=12 per group) C) Pictures showing ears of mice of various groups on Day 7. D) H&E- stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. F) qPCR of pro- inflammatory cytokines of respective groups. G) Schematic showing strategy for developing conditional knockout mice with HTR2A knocked out in moLCs. H) Line chart showing changes in ear swelling of mice. (n=6 to 10 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E- stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. L) qPCR of pro- inflammatory cytokines of respective groups. Data are representative of two independent experiments (B- F and H- L). p values determined by two- way ANOVA (B and H) followed by Tukey's post- hoc test and one- way ANOVA (D- F and J- L) followed by Tukey's post- hoc test. Mean ± SEM (B- F and H- L). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001; ns, non- significant. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. IL-23 plays a critical role in exacerbating psoriasisin inflammation by HTR2A deficient monocyte-derived Langerhans cells. A) Heatmap showing mRNA levels of key cytokines in wild type or HTR2A deficient moLCs. (3 biological replicates) The cells were stimulated for 24 hours with Imiquimod (10μg/mL). B) IL-23 ELISA of stimulated and unstimulated 1X105 wild type or HTR2A deficient moLCs. C) Representative histogram of IL-23a MFI among different cell types in cells isolated from ears of HTR2A deficient mice treated with Imiquimod for 7 days. (n=4) The cells were isolated and treated with PMA, Ionomycin, and GolgiStop before antibody staining and flow cytometry analysis. There is quantification of IL-23a MFI and percentage of IL-23a positive cells. D) Schematic showing the experimental design for treating mice with DOI and determining IL-23a expression. E) Line chart showing changes in ear
+ +<--- Page Split ---> + +swelling of mice. (n=8 per group) F) Picture of ears on Day 7 of the respective groups. G) H&E- stained slides of the respective groups with quantification of epidermal thickness. H) Flow cytometry showing the percentage of IL- 23+ cells among CD45+ cells in the respective groups with quantification. I) Flow cytometry showing the percentage of IL- 23+ cells among moLCs in the respective groups with quantification. J) Histogram of moLCs' IL- 23a MFI in the respective groups with quantification. K) Schematic showing the experimental design of IL- 23r blocking experiments. L) Line chart showing changes in ear swelling of mice in the respective groups. (n=6) M) Picture of ears on Day 7 of the respective groups. N) H&E- stained slides of the respective groups with quantification of epidermal thickness. O) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. P) mRNA levels of proinflammatory cytokines of the respective groups. Data are representative of two independent experiments (B, E-J, and L-P). p values determined by two- way ANOVA (E and L) followed by Tukey's post- hoc test, one- way ANOVA (B and N- P) followed by Tukey's post- hoc test, one- way ANOVA (C) followed by Dunnett post- hoc test, and Student's T- test (G- J). Mean ± SEM (B- C, E, G- J, L, and N- P). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 6
+ +Figure 6. HTR2A suppresses IL- 23 secretion by inhibiting activation of non- canonical NFkB pathway in monocyte- derived Langerhans cells. A) Heatmap showing 140 differentially expressed genes (DEGs) between wild type moLCs (GSE222197) treated with IMQ and KO moLCs treated with IMQ (GSE274941) with genes affected by NFkB shown on the right. The experiments generating the datasets of GSE222197 and GSE274941 were done in parallel. MoLCs were pooled from the ears of five mice treated with IMQ for two days, constituting a single biological sample. B) Bar chart showing transcription factors enriched among the DEGs from A) as analyzed using GAGE on the TRRUST Transcription Factor database. C) Western blot of cultured WT and KO moLCs lysate treated with or without DOI and stained with anti- p100/p52, anti- RelB, and anti- \(\beta\) - actin. D) IL- 23 ELISA of 2X105 WT or KO moLCs (5 biological replicates) stimulated for 24 hours with Imiquimod (10μg/mL) and treated with either DOI (10μg/mL) or EVP 4593 (0.1mM). E) Differentiation of Vγ4 T cells into IL- 17+ IL- 22+ cells after coculture with WT or KO moLCs (5 biological replicates) for 3 days. WT or KO moLCs were stimulated for 24 hours with IMQ (10μg/mL) and WT moLCs were treated with either DOI (10μg/mL) or EVP 4593 (0.1mM). Data are a summary of two independent experiments (C- E). p values were determined by one- way ANOVA followed by Dunnett's post- hoc test (D and E). Mean ± or SEM (D and E) *, p<0.05; ***, p<0.001; ****, p<0.0001; ns, non- significant. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 7
+ +Figure 7. Serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation. A) Immunohistochemistry staining of serotonin on the skin of healthy volunteers, non- lesioned skin of psoriatic patients, and lesioned skin of psoriatic patients. B) Quantification of serotonin intensity in terms of H- score using DensitQuant (3DHISTECH LTD) between the skin of healthy volunteers \((n = 13)\) , lesioned skin of psoriatic patients with low PASI score \((< 30)\) \((n = 15)\) and high PASI score \((>30)\) \((n = 12)\) . C) Quantification of serotonin intensity in terms of H- score between lesioned and non- lesioned skin of psoriatic patients \((n = 27)\) . D) Schematic showing strategy for treating wild type or HTR2A deficient mice with serotonin with their respective control. E) Line chart showing changes in ear thickness of mice of the respective groups \((n = 7)\) . F) Pictures showing the ears of mice in different groups. G) H&E- stained slides of respective groups with quantification of epidermal thickness. H) Flow cytometry of IL17+ IL22+ \(\mathrm{V} / \mathrm{4T}\) cells and neutrophils of respective groups with quantifications. I) mRNA levels of pro- inflammatory cytokines of the respective groups. Data are a summary of two independent experiments (D- I and L- R). p values were determined by two- way ANOVA (E), one- way ANOVA followed by Tukey's post- hoc test (B and G- I) or paired Student's T- test (C). Mean \(\pm\) or SEM (B, C, E, G- I) \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Figure 8
+ +Figure 8. Monocyte- derived Langerhans cells of human psoriatic skin show higher IL23A and lower HTR2A expression. A) UMAP of reanalyzed single cell RNA- sequencing data from GSE151177 and GSE162183 showing 27 different clusters. B) Violin map showing lineage defining transcription factors, receptors, and markers of embryo Langerhans cells (eLC), monocyte- derived Langerhans cells (molC), and dendritic cell type 3 (DC3). C) Box plot comparing HTR2A expression levels of control and psoriatic monocyte- derived Langerhans cells. D) Dot plot showing expression of various cytokines by eLC, molC, and DC3. E) Dot plot showing expression of various NFκB pathway- related genes by psoriatic molCs and control molCs. F) Immunofluorescence staining of HTR2A on Langerhans cells (marked by CD1a) in lesioned and non- lesioned psoriatic skin with quantification (n= 27). p values were determined by paired Student's T- test (C, F). \*, p<0.05; \*\*, p<0.01. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- GEODataset.pdf- 20241209Supplementalinformation.pdf- NCOMMS2482399rs.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d_det.mmd b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6280bd6c46b01c0bd8a034c2bcdc387857c44733 --- /dev/null +++ b/preprint/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d/preprint__14079465b678c4c8d53cc20fe71aa99b32265164ac8347493397d2596eede84d_det.mmd @@ -0,0 +1,647 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 867, 207]]<|/det|> +# Serotonin 2A Receptor Attenuates Psoriatic Inflammation by Suppressing IL-23 Secretion in Monocyte-derived Langerhans Cells + +<|ref|>text<|/ref|><|det|>[[44, 229, 335, 275]]<|/det|> +Yungling Lee leolee@ibms.sinica.edu.tw + +<|ref|>text<|/ref|><|det|>[[44, 301, 855, 321]]<|/det|> +Institute of Biomedical Sciences, Academia Sinica https://orcid.org/0000- 0002- 2234- 9479 + +<|ref|>text<|/ref|><|det|>[[44, 327, 400, 368]]<|/det|> +Yeh Fong Tan https://orcid.org/0000- 0003- 3361- 4300 + +<|ref|>text<|/ref|><|det|>[[44, 373, 940, 435]]<|/det|> +Chen- Yun Yeh Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0003- 1049- 8052 + +<|ref|>text<|/ref|><|det|>[[44, 442, 625, 483]]<|/det|> +Sheng- Yun Hsu Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan + +<|ref|>text<|/ref|><|det|>[[44, 488, 940, 550]]<|/det|> +Chun- Hao Lu Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0001- 7084- 9374 + +<|ref|>text<|/ref|><|det|>[[44, 556, 940, 620]]<|/det|> +Ching- Hui Tsai Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0000- 0003- 0998- 8230 + +<|ref|>text<|/ref|><|det|>[[44, 625, 940, 688]]<|/det|> +Pei- Chuan Chiang Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan https://orcid.org/0009- 0009- 8057- 4621 + +<|ref|>text<|/ref|><|det|>[[44, 694, 633, 736]]<|/det|> +Hao- Jui Weng Taipei Medical University https://orcid.org/0000- 0002- 7006- 7599 + +<|ref|>text<|/ref|><|det|>[[44, 740, 890, 805]]<|/det|> +Tsen- Fang Tsai Department of Dermatology, National Taiwan University Hospital and National Taiwan University College of Medicine, Taipei, Taiwan + +<|ref|>sub_title<|/ref|><|det|>[[44, 846, 103, 864]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 885, 135, 903]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 922, 318, 941]]<|/det|> +Posted Date: January 6th, 2025 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 46, 473, 65]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 5628384/v1 + +<|ref|>text<|/ref|><|det|>[[42, 83, 915, 125]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 143, 949, 255]]<|/det|> +Additional Declarations: Yes there is potential Competing Interest. Tsen- Fang Tsai has conducted clinical trials or received honoraria for serving as a consultant for AbbVie, AnaptysBio, Bristol- Myers Squibb, Boehringer Ingelheim, Celgene, Eli Lilly, Galderma, GlaxoSmithKline- Stiefel, Janssen- Cilag, Leo- Pharma, Merck, Novartis, PharmaEssentia, Pfizer, Sanofi, Sun Pharma and UCB. The remaining authors state no conflict of interest. + +<|ref|>text<|/ref|><|det|>[[42, 289, 920, 332]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 29th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 63971- 5. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 97, 848, 133]]<|/det|> +## Title: Serotonin 2A Receptor Attenuates Psoriatic Inflammation by Suppressing IL-23 Secretion in Monocyte-derived Langerhans Cells + +<|ref|>text<|/ref|><|det|>[[144, 140, 828, 175]]<|/det|> +One Sentence Summary: Serotonin 2A receptor expression on monocyte Langerhans cells attenuates psoriatic inflammation by reducing IL- 23 secretion. + +<|ref|>sub_title<|/ref|><|det|>[[145, 184, 221, 200]]<|/det|> +## Authors: + +<|ref|>text<|/ref|><|det|>[[144, 206, 825, 239]]<|/det|> +Yeh Fong Tan1,2, Chen- Yun Yeh2, Sheng- Yun Hsu2, Chun- Hao Lu2, Ching- Hui Tsai2, Pei- Chuan Chiang2, Hao- Jui Weng3,4,5, Tsen- Fang Tsai3, Yungling Leo Lee2,6,7 + +<|ref|>sub_title<|/ref|><|det|>[[145, 247, 247, 263]]<|/det|> +## Affiliations: + +<|ref|>text<|/ref|><|det|>[[144, 263, 841, 444]]<|/det|> +1. Taiwan International Graduate Program in Molecular Medicine, National Yang Ming Chiao Tung University and Academia Sinica; Taipei, Taiwan. +2. Institute of Biomedical Sciences, Academia Sinica; Taipei, Taiwan. +3. Department of Dermatology, National Taiwan University Hospital and National Taiwan University College of Medicine; Taipei, Taiwan. +4. Department of Dermatology, Taipei Medical University-Shuang Ho Hospital; New Taipei City, Taiwan. +5. Department of Dermatology, School of Medicine, College of Medicine, Taipei Medical University; Taipei, Taiwan. +6. College of Public Health, China Medical University, Taichung, Taiwan. +7. Lead contact. + +<|ref|>text<|/ref|><|det|>[[144, 453, 530, 469]]<|/det|> +CORRESPONDENCE: leolee@ibms.sinica.edu.tw + +<|ref|>sub_title<|/ref|><|det|>[[145, 481, 223, 497]]<|/det|> +## Abstract: + +<|ref|>text<|/ref|><|det|>[[143, 505, 840, 725]]<|/det|> +Anecdotal evidence has suggested an association between psychiatric drugs and psoriasis but consensus is absent due to contradicting reports and the mechanism remains poorly defined. Here, we investigated the role played by serotonin 2A receptor (HTR2A), a receptor commonly targeted by psychiatric drugs, in regulating psoriasis. HTR2A antagonistic drugs worsened psoriatic outcome and HTR2A modulation reduced psoriatic inflammation. Using Imiquimod- induced psoriasisform model, HTR2A- deficient mice experienced exacerbated inflammation. Hematopoietic cells, particularly monocyte- derived Langerhans cells (molC), were responsible for this phenotype. Mechanistically, the exacerbated inflammation is due to increased interleukin- 23 (IL- 23) secretion and HTR2A suppresses it by inhibiting activation of the non- canonical NFkB pathway. Serotonin is the putative agonist modulating HTR2A attenuating psoriatic inflammation. Lastly, our findings in mice were also validated clinically. Our data demonstrate serotonin modulates HTR2A, attenuating psoriatic inflammation by suppressing IL- 23 secretion via inhibiting non- canonical NFkB pathway in molCs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 304, 106]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[144, 108, 850, 289]]<|/det|> +INTRODUCTIONPsoriasis is a chronic immune- mediated inflammatory skin condition that affects millions of people worldwide.(1) Despite the identification of IL- 23/IL- 17 axis playing a key role in psoriasis, much remains unknown about psoriasis, as patients exhibit varied responses despite receiving identical treatment.(2) Of late, the psychological component of psoriasis has been given increased attention, as emerging evidence suggests mental health influences disease outcomes.(3, 4) High rates of psychiatric comorbidities are observed in psoriatic patients leading to the classification of psoriasis as a psychosomatic disease.(3, 5) For many patients, exacerbation of psoriatic episodes is often preceded by stressful life episodes, underscoring the need to unravel the complexities between psychological and physiological factors in psoriasis.(6, 7) + +<|ref|>text<|/ref|><|det|>[[144, 300, 845, 483]]<|/det|> +Interestingly, studies have shown mixed results regarding the impact of mood- altering medications on psoriasis. While some selective serotonin reuptake inhibitors (SSRIs) appear to improve psoriatic conditions (8- 10), others have been linked to worsening psoriasis.(11- 13) This conflicting evidence highlights the need for a closer examination of serotonin's role in psoriasis. Thus, here we focused on second- generation antipsychotics- serotonin and dopamine receptor antagonists, which have strong antagonism for serotonin receptors.(14) In contrast, SSRIs have unpredictable effects on local serotonin levels depending on the duration of treatment.(15) Clozapine and risperidone, both second- generation antipsychotics, with their direct antagonism of serotonin receptors, particularly serotonin 2A receptor (HTR2A), facilitate a clear interpretation of results making them ideal candidates for further studies.(14, 16) + +<|ref|>text<|/ref|><|det|>[[144, 493, 844, 639]]<|/det|> +HTR2A is a G protein- coupled receptor (GPCR) of the Gq subtype and is primarily activated by serotonin.(17) HTR2A can be activated by naturally occurring compounds like psilocybin or synthetic compounds like lysergic acid diethylamide and (R)- 1- (2,5- dimethoxy- 4- iodophenyl)- 2- aminopropane (DOI).(18, 19) In particular, DOI is widely used to study HTR2A as it is a HTR2A- specific agonist.(19) Studies have reported that DOI suppresses inflammation by blocking tumor necrosis factor- alpha (TNF- a)- induced cytokine and chemokine expression both in vivo and in vitro yet we lack a detailed understanding of how HTR2A regulates immune cells and cytokines in the context of psoriasis.(20, 21) + +<|ref|>text<|/ref|><|det|>[[144, 650, 850, 850]]<|/det|> +We previously found HTR2A to be highly expressed in Langerhans cells (LC), indicating HTR2A could impact LC functions affecting psoriatic outcome.(22) LCs are skin macrophages with dendritic cell functions and are known to play a role in the exacerbation of psoriatic symptoms involving the cytokine IL- 23 (23, 24); however, some reports suggest otherwise (25), indicating that LCs might play a role in immune tolerance. This schism could be due to the existence of two subsets of LCs, identified as embryo LC (eLC) and monocyte- derived LC (moLC).(26) eLCs originate from erythro- myeloid progenitors which were formed from the yolk sac, while moLCs originate from monocyte- like precursors which are derived from definitive hematopoiesis. (26) moLCs can be identified by their expression of CD64 and have been known to play a role in exacerbating psoriasisform inflammation.(27- 29) It remains to be investigated how each of the LC subsets is affected by HTR2A. + +<|ref|>text<|/ref|><|det|>[[144, 861, 837, 897]]<|/det|> +While HTR2A's role in regulating cytokine- mediated inflammation is documented, no studies have yet linked it directly to autoimmune diseases like psoriasis. Addressing this gap, our study + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 848, 145]]<|/det|> +investigates how HTR2A expression influences psoriasisform inflammation. We found that HTR2A expression on molCs attenuates psoriasisform inflammation by reducing IL- 23 secretion through the non- canonical NFκB pathway. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 90, 237, 107]]<|/det|> +## RESULTS + +<|ref|>sub_title<|/ref|><|det|>[[145, 123, 585, 140]]<|/det|> +## Modulation and expression levels of HTR2A affect psoriasis + +<|ref|>text<|/ref|><|det|>[[144, 150, 841, 369]]<|/det|> +To determine the association between psoriasis and serotonin receptors, we performed a retrospective cohort study to examine serotonin- related drug use and changes in psoriasis treatment received as a surrogate marker for psoriasis severity. The categories of psoriasis treatment are as defined by Taiwan Dermatological Association with topical treatment for patients with mild psoriasis symptoms, followed by systemic agents, combination therapy, and biologics for patients with most severe symptoms.(30) We defined worsening psoriasis as a change from mild treatment to severe treatment, for example from receiving topical treatment to receiving biologics. We found patients taking anti- psychotics were more likely to have worsening psoriasis as compared to control patients not taking anti- psychotics or SSRIs (Figure 1A and S1C). Correspondingly, patients on anti- psychotics did not show any significant improvement in psoriasis (Figure S1A and S1C). Of note, patients on SSRIs were found to have a significant worsening but no improvement in psoriasis (Figure S1B and S1C). + +<|ref|>text<|/ref|><|det|>[[144, 380, 854, 562]]<|/det|> +We then looked at the hazard ratio of each anti- psychotic and found patients on aripiprazole, chlorpromazine, prochlorperazine, quetiapine, risperidone, sulpiride and anti- psychotics categorized as Others (brexipiprazole, fluphenazine, loxapine, lurasidone, perphenazine, pimozide, thioridazine, and trifluoperazine) were more likely to have worsening psoriasis as compared to controls as they had statistically significant hazard ratios that were greater than one (Figure 1B). We did not find statistically significant hazard ratios in patients taking amisulpride, clozapine, haloperidol, and olanzapine. We then checked the binding affinity of each antipsychotic to each serotonin receptor on the PDSP K, database.(31) We found most of these antipsychotics bound most strongly to HTR2A among the serotonin receptors indicating HTR2A might be responsible for the worsening psoriasis observed (Figure 1C and S1D). + +<|ref|>text<|/ref|><|det|>[[143, 572, 850, 884]]<|/det|> +To determine whether HTR2A plays a role in psoriasis, we modulated HTR2A via the specific agonist of DOI and wondered if it causes transcriptional changes in psoriatic genes. We obtained skin biopsies from lesioned sites of three psoriatic patients who have yet to receive any psoriatic treatment and skin from three healthy subjects as controls. The biopsies were halved, with half of it reserved for RNA sequencing without treatment (NT) and the other half treated with DOI (T) for 16 hours. All samples were subjected to RNA sequencing (Figure 1D) (GSE274449). We determined the differentially expressed genes (DEGs) of lesioned skin T versus lesioned skin NT and identified 653 genes. A total of 1475 psoriatic DEGs were identified when comparing NT healthy skin versus NT lesioned skin. We then compared these two sets of DEGs and found 107 DEGs overlapped (Figure 1E). 94 of these genes were upregulated in psoriasis but were downregulated following DOI treatment and 13 of these genes were downregulated in psoriasis but with DOI treatment, they were upregulated. In healthy skin irrespective of treatment, the identified genes share a similar pattern with the lesioned treated group indicating HTR2A activation on lesioned psoriatic skin reversed psoriasis transcriptomic profile (Figure 1F). We then determined HTR2A expression in psoriatic cases using publicly available datasets on GEO database and found lower HTR2A expression in psoriatic cases (Figure 1G). In all, these data strongly suggest that HTR2A plays a role in affecting psoriatic outcomes. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 852, 108]]<|/det|> +## HTR2A deficiency in mice and hematopoietic stem cells exacerbates psoriasisform inflammation + +<|ref|>text<|/ref|><|det|>[[143, 118, 856, 469]]<|/det|> +To investigate HTR2A's role in psoriasis further, we treated HTR2A deficient (KO) and aged- match, gender- matched wild type (WT) mice with Imiquimod (IMQ) for 7 days (Figure 2A). There was a significant increase in ear swelling in IMQ- treated KO mice as compared to WT mice. In contrast, no differences were observed in mice treated with Vaseline irrespective of HTR2A expression (Figure 2B). By visual inspection, we found the ears of IMQ- treated KO mice to be thicker with more scaling (Figure 2C). Histologically, acanthosis along with hyperkeratosis, parakeratosis, Munro's microabscesses, and immune infiltrate are hallmarks of psoriasis and reflect the severity of psoriasis.(32) There was obvious acanthosis (thickening of the epidermis) and immune infiltrate in hematoxylin and eosin (H&E)- stained slides of IMQ- treated KO mice. There was also a significant increase in epidermis thickening in IMQ- treated KO mice as compared to IMQ- treated WT mice (Figure 2D). It has been reported that \(\gamma \delta\) T cells will respond to IL- 23, a key cytokine in the pathogenesis of psoriasis, and differentiate into IL- 17- producing cells.(33) Of note, the \(\mathrm{V}\gamma 4\) subset is one of the major IL- 17- producing \(\gamma \delta\) T cells.(34, 35) Circulating neutrophils are also normally recruited into inflammatory sites and in psoriasis, Munro's microabscesses are filled with neutrophils.(36) Flow cytometry of cells according to our gating strategy revealed an increase in IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4+\) T cells and an increase in neutrophils (Figure 2E and S2). However, there were no statistically significant differences in IL- 17- and IL- 22- producing \(\alpha \beta\) CD4 T cells. Key pro- inflammatory cytokines like Il17a and Il23a were also increased in HTR2A deficient mice (Figure 2F). In all, HTR2A deficiency exacerbates psoriasisform inflammation. + +<|ref|>text<|/ref|><|det|>[[143, 479, 852, 753]]<|/det|> +To investigate which cell was responsible for the exacerbated inflammation, we performed reciprocal bone marrow transplants to generate chimeric mice with WT bone marrow (representing hematopoietic stem cells (HSCs) and KO somatic cells (representing non- HSCs) and vice versa. KO bone marrow was transplanted to KO mice as a control (Figure 2G). We checked the composition of cells 11 weeks post- transplant and approximately \(65\% - 70\%\) of CD45+ cells were replaced with cells from donor bone marrow (Figure S3A). A statistically significant increase in ear swelling was observed in mice with KO bone marrow irrespective of its somatic cells (Figure 2H). Visually, the ears of mice with KO bone marrow appeared thicker with some scaling (Figure 2I). Histologically, the dermis did not vary much between the groups but the epidermis was significantly thicker in mice with KO bone marrow than in mice with WT bone marrow (Figure 2J). Flow cytometry of immune cells revealed an increase in IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4+\) T cells in mice with KO bone marrow. Paradoxically, neutrophil levels were lower in mice with KO bone marrow (Figure 2K). Il17a and Il23a were increased in mice with KO bone marrow (Figure 2L). Our findings suggest that HTR2A deficiency in HSCs is responsible for the exacerbated inflammation. + +<|ref|>text<|/ref|><|det|>[[144, 765, 844, 892]]<|/det|> +Since HTR2A deficiency in HSCs exacerbates inflammation and immune cells make up a large part of HSCs, we wondered are the adaptive immune cells or the innate immune cells causing this exacerbated inflammation. To answer this, we generated mice with conditional deletion of HTR2A in B and T cells (HTR2ARag1) by crossing Rag1- cre mice with HTR2A fixed mice (Figure S4A). (37) We found no increase in ear swelling in IMQ- treated- HTR2ARag1 mice as compared to IMQ- treated- HTR2Afl/fl mice (Figure S4B). Visual inspection of mice ears found minimal scaling and no differences between the groups (Figure S4C). Histologically, there was no increase in the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 842, 199]]<|/det|> +thickening of the epidermis or the dermis between the groups. When epidermal thickening was quantified, there were no significant differences (Figure S4D). There were also no differences in pro- inflammatory immune cells and pro- inflammatory cytokines between IMQ- treated- HTR2ARag1 mice and IMQ- treated- HTR2Afl/fl mice (Figure S4E and S4F). The data here indicates that HTR2A deficiency in B and T cells, which are the adaptive immune cells, does not cause exacerbated inflammation. + +<|ref|>sub_title<|/ref|><|det|>[[144, 210, 795, 245]]<|/det|> +## HTR2A deficiency in Langerhans cells recapitulates exacerbated inflammation in HTR2A deficient mice + +<|ref|>text<|/ref|><|det|>[[144, 256, 853, 420]]<|/det|> +As adaptive immune cells were not responsible for the exacerbated inflammation, we turned our attention to innate immune cells. Monocyte- derived inflammatory Langerhans cells have been reported to cause psoriasis- like inflammation.(28) Hence, we decided to generate mice with conditional deletion of HTR2A in Langerhans cells (HTR2ACD207) by crossing CD207- cre mice with HTR2A fixed mice (Figure 3A).(38) We found conditional deletion of HTR2A in Langerhans cells (LC) recapitulated inflammatory levels observed in HTR2A deficient mice with increased ear swelling, increased scaling, increased epidermal thickening, increased IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4\) T cells, and increased mRNA levels of Il17a and Il23a (Figures 3B- 3F). However, the increase in neutrophil recruitment is not statistically significant (Figure 3E). + +<|ref|>text<|/ref|><|det|>[[144, 431, 853, 668]]<|/det|> +To further validate HTR2A's influence on LC exacerbating inflammation, we used a Langerhans cell depletion model, whereby LCs were depleted by administering diphtheria toxins to CD207- diphtheria toxin receptor (CD207- DTR) mice.(28) We validated that this method could lead to the depletion of LCs (Figures S5A- S5D). All the mice were administered with \(0.1\mathrm{mg / kg}\) of DOI intraperitoneally (Figure 3G). In terms of ear swelling, we found DOI- suppressed inflammation in mice with LCs but not in mice with depleted LCs (Figure 3H). A similar trend was observed visually and histologically (Figure 3I and 3J). IL- 17- and IL- 22- producing \(\mathrm{V}\gamma 4\) T cells were low in mice with no depletion but high in mice with LCs depletion and the differences were statistically significant (Figure 3K). Neutrophil levels follow a similar trend but there was no statistical significance (Figure 3K). mRNA of Il17a and Il23a is high in mice with LC depletion and low in mice with no depletion, however, only differences in Il17a reach statistical significance (Figure 3L). The data here strongly supports that HTR2A deficiency in LCs causes exacerbated inflammation. + +<|ref|>sub_title<|/ref|><|det|>[[144, 680, 767, 714]]<|/det|> +## HTR2A deficiency in Langerhans cells of monocyte lineage exacerbates psoriasisinflammation + +<|ref|>text<|/ref|><|det|>[[144, 726, 852, 890]]<|/det|> +We have found HTR2A deficiency in both cells of bone marrow origin and LCs lead to exacerbated inflammation, but these two findings contradict each other, as LCs are not replaced during bone marrow transplantation.(28) To reconcile the differences, we suspect HTR2A affects LCs derived from monocytes, which is of bone marrow origin, causing exacerbated inflammation. To test our hypothesis, we differentiated bone marrow cells into monocyte- derived Langerhans cells (molC) as published previously (Figure 4A).(39) These differentiated cells are considered of monocyte origin due to their CD64 expression and Irf8 expression (lineage defining transcription factor for monocytes) (Figure S6A- S6C).(27, 39) They are considered LCs due to their expression of Id2 and Runx3 which are lineage defining transcription factors for LCs (Figure S6C).(39, 40) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 847, 236]]<|/det|> +Cells expressing MHC II and CD207 are considered moLcs whereas all the other cells are considered non- moLcs (Figure S6B). These cells were then adoptively transferred into the ear pinna of LC- depleted mice (Figure 4B). Exacerbated inflammation was observed in mice with HTR2A deficient moLcs, whereas a lower- grade inflammation was observed in mice with WT moLcs (Figure 4B- 4D). Mice adoptively transferred with non- moLcs irrespective of HTR2A expression experienced minimal inflammation (Figure 4B- 4D). A similar inflammatory trend was observed in IL- 17- and IL- 22- producing \(\mathrm{V} / \mathrm{Y}4\mathrm{T}\) cells and mRNA levels of Il17a; however, this trend was not observed in neutrophil levels and mRNA levels of Il23a (Figure 4E and 4F). + +<|ref|>text<|/ref|><|det|>[[144, 246, 850, 428]]<|/det|> +To further validate the role of LCs of monocyte origin being responsible for the exacerbated inflammation seen in HTR2A deficient mice, we generated mice with conditional deletion of HTR2A in cells of monocyte lineage (HTR2AMs43) by crossing Ms43- cre mice with HTR2A fixed mice (Figure 4G). (41) Exacerbated inflammation was observed in IMQ- treated- HTR2AMs43 mice in terms of ear thickening, visually, and histologically (Figure 4H- 4I). When quantified, there were significant increases in ear swelling and epidermal thickening (Figure 4H and 4J). IL- 17- and IL- 22- producing \(\mathrm{V} / \mathrm{Y}4\mathrm{T}\) cells but not neutrophils were significantly increased (Figure 4K). mRNA levels of Il17a and Il23a were increased significantly in IMQ- treated- HTR2AMs43 mice (Figure 4L). In all, we conclude that HTR2A deficiency in LCs of monocyte origin exacerbates psoriasisform inflammation. + +<|ref|>sub_title<|/ref|><|det|>[[144, 440, 830, 475]]<|/det|> +## IL-23 secretion by HTR2A deficient monocyte-derived Langerhans cells plays a critical role in exacerbating psoriasisform inflammation + +<|ref|>text<|/ref|><|det|>[[144, 485, 852, 705]]<|/det|> +We turned our focus on the cytokine affected by HTR2A deficiency in moLcs. We determined the mRNA levels of commonly secreted cytokines by Langerhans cells (Tnf, Il1b, Il23a, Il12a, Il6, Il2, Il10 and Tgfb1) after stimulating them with Imiquimod. (22, 39, 42) We found Il23a to be drastically increased in IMQ- stimulated HTR2A deficient moLcs as compared to IMQ- stimulated WT moLcs (Figure 5A). To validate it, we performed IL- 23p19 ELISA on stimulated and unstimulated, WT or HTR2A deficient moLcs and found IL- 23 levels to be increased significantly in stimulated HTR2A deficient moLcs as compared to stimulated WT moLcs (Figure 5B). We then determined IL- 23 secretion among different cell types in IMQ- treated- HTR2A deficient mice and the cells were gated as published (Figure 5Y). (28) We found that moLcs have the highest IL- 23 mean fluorescence intensity (MFI) and moLcs have the most IL- 23- producing cells among all the other cell types (Figure 5C). These data strongly suggest HTR2A deficient moLcs have increased secretion of IL- 23. + +<|ref|>text<|/ref|><|det|>[[144, 716, 848, 861]]<|/det|> +We performed an in vivo experiment where we injected DOI or PBS intraperitoneally into WT mice and looked at inflammatory and IL- 23 levels (Figure 5D). We found mice injected with DOI had less inflammation, as seen in ear swelling, gross evaluation, and epidermal thickness (Figure 5E- 5G). Importantly, we found IL- 23- positive cells to decrease in the DOI- treated group as compared to the control group (Figure 5H). The percentage of IL- 23 positive cells among moLcs is lower in DOI- treated as compared to the control group, but it did not reach statistical significance (Figure 5I). Nevertheless, IL- 23 MFI of DOI- treated moLcs is significantly lower as compared to control moLcs (Figure 5J). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 852, 291]]<|/det|> +To further validate IL- 23's role in exacerbating inflammation, we performed an in vivo IL- 23 neutralizing experiment (Figure 5K). In mice injected intradermally with isotype antibody, the inflammation was more severe in HTR2A deficient mice than in WT mice. The inflammation levels were minimal as observed in terms of ear swelling, visual inspection, histological assessment, and quantification of epidermal thickness in \(\alpha\) - IL23r- treated mice irrespective of HTR2A deficiency (Figure 5L- 5N). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells were reduced in \(\alpha\) - IL23r- treated mice in both WT and HTR2A deficient mice; however, neutrophil levels were not significantly different (Figure 5O). mRNA levels of pro- inflammatory cytokines were reduced in both WT and HTR2A deficient mice treated with neutralizing antibodies, with a reduction in IL17a levels reaching statistical significance (Figure 5P). With that, we conclude that IL- 23 plays a pivotal role in exacerbating psoriasisform inflammation in HTR2A- deficient moLCs. + +<|ref|>sub_title<|/ref|><|det|>[[144, 303, 830, 339]]<|/det|> +## HTR2A suppresses IL-23 secretion by inhibiting activation of non-canonical NFκB pathway in monocyte-derived Langerhans cells + +<|ref|>text<|/ref|><|det|>[[144, 349, 850, 496]]<|/det|> +To determine the transcription factor involved in the exacerbated inflammation, we performed bulk RNA- sequencing on HTR2A- deficient moLCs isolated from IMQ- treated ears (GSE274941) and compared it with WT moLCs isolated from IMQ- treated ears as published previously (GSE222197). A total of 140 DEGs were identified, 47 genes were downregulated and 93 genes were upregulated in HTR2A deficient LCs (Figure 6A). The DEGs were analyzed by GAGE pathway analysis on TRRUST Transcription Factor Database and we found the NFkB1 pathway to be the top- ranked statistically significant transcription factor (Figure 6B). The key genes requiring NFkB1 as a transcription factor were listed on the heat map (Figure 6A). + +<|ref|>text<|/ref|><|det|>[[144, 507, 851, 636]]<|/det|> +It has been reported that IL- 23a secretion involves both canonical and non- canonical NFκB pathways.(43) We performed Western blot on key molecules within the non- canonical NFκB pathway and found processing of p100 unit is increased in HTR2A deficient moLCs as p100 levels were low and p52 (the processed subunit of p100) is increased in HTR2A deficient moLCs. The addition of DOI to WT moLCs lead to a reduced level of p52 even though p100 levels were rather similar. There were, however, no changes to RelB levels in HTR2A deficient moLCs as compared to WT moLCs (Figure 6C). + +<|ref|>text<|/ref|><|det|>[[144, 647, 848, 813]]<|/det|> +We then investigated the role of NFκB using the in vitro assays of IL- 23p19 ELISA and coculture of \(\mathrm{V} / 4\) T cell with moLCs, DOI, and antagonist of NFκB (EVP 4593).(44) We found IL- 23 levels to be lower in WT moLCs stimulated with Imiquimod than stimulated HTR2A deficient moLCs. When DOI, the HTR2A- specific agonist, and EVP 4593 were added into WT moLCs, IL- 23 levels were decreased (Figure 6D). Differentiation of \(\mathrm{V} / 4\) T cells into IL- 17- and IL- 22- producing cells was lower when cultured with stimulated WT moLCs than with stimulated HTR2A deficient moLCs. The addition of DOI or EVP 4593 into WT moLCs reduced the differentiation of cocultured \(\mathrm{V} / 4\) T cells when compared to stimulated WT moLCs only. (Figure 6E) In all, we found that HTR2A suppresses IL- 23 secretion involving the non- canonical NFκB pathway. + +<|ref|>sub_title<|/ref|><|det|>[[144, 824, 725, 841]]<|/det|> +## Serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation + +<|ref|>text<|/ref|><|det|>[[145, 853, 848, 907]]<|/det|> +We then wondered what is the putative agonist of HTR2A in physiological conditions. Since serotonin is the known agonist of HTR2A, we wondered if there were any changes to their levels in psoriasis.(45) We performed immunohistochemistry (IHC) staining of serotonin on skin + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 852, 234]]<|/det|> +samples of healthy volunteers, non- lesioned skin, and lesioned skin of psoriatic patients. We found serotonin levels to be high in healthy volunteers and non- lesioned skin. Serotonin levels were low in lesioned skin (Figure 7A). We quantified the serotonin levels by determining its intensity in terms of H- score and found serotonin levels to be significantly higher in the skin of healthy volunteers than lesioned skin of psoriatic patients irrespective of their psoriasis area and severity index (PASI) score. The difference in serotonin levels of psoriatic skin between low PASI \(< 30\) and high PASI \((\geq 30)\) was not significant (Figure 7B). Serotonin levels were higher in non- lesioned skin than lesioned skin even though both are from the same patient (Figure 7C). + +<|ref|>text<|/ref|><|det|>[[144, 245, 852, 446]]<|/det|> +To further validate serotonin's role as the putative agonist of HTR2A in psoriasis, we injected serotonin intraperitoneally into WT or HTR2A deficient mice (Figure 7D). In terms of ear swelling, visual inspection, and histological assessment, we found WT mice injected with serotonin to have lower inflammatory levels than WT control mice. On the other hand, HTR2A deficient mice injected with serotonin experienced exacerbated inflammation (Figure 7E- 7G). A similar trend was observed during the quantification of epidermal thickness (Figure 7G). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells were highest in HTR2A deficient mice injected with serotonin, followed by WT control mice and WT mice injected with serotonin. The differences were significant (Figure 7H). However, when we looked at neutrophil levels, mRNA levels of Il17a and Il23a, we found a similar trend but the differences were not significant (Figure 7H and 7I). The data here supports serotonin as the putative agonist of HTR2A attenuating psoriatic inflammation. + +<|ref|>text<|/ref|><|det|>[[144, 457, 853, 696]]<|/det|> +We then wondered where does this serotonin originate from. Platelet is known to store peripheral serotonin, hence, we determined plasma and platelet serotonin levels of psoriatic patients and healthy controls.(46) We found no difference in plasma serotonin levels but found a significant difference in platelet serotonin levels of psoriatic patients and healthy controls (Figure S8A and S8B). We then used an established method to deplete platelet serotonin levels by pretreating the mice for 14 days with Fluoxetine (Figure S8C).(15) Mice treated with fluoxetine were found to have very low platelet serotonin levels (Figure S8D). In terms of ear swelling, visual inspection, histological assessment, and quantification of epidermal thickness, we found mice depleted of platelet serotonin have exacerbated inflammation (Figure S8E- S8G). IL- 17- and IL- 22- producing \(\mathrm{V} / 4\) T cells, neutrophils, mRNA of Il17a and Il23a were higher in platelet serotonin- depleted mice but only changes in neutrophil levels and changes in mRNA levels of Il17a achieve significance (Figure S8H- S8I). In all, serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation and our preliminary data suggests it originates from platelet. + +<|ref|>sub_title<|/ref|><|det|>[[144, 707, 808, 741]]<|/det|> +## Monocyte-derived Langerhans cells of human psoriatic skin show higher IL23A and lower HTR2A expression + +<|ref|>text<|/ref|><|det|>[[144, 753, 848, 899]]<|/det|> +To ensure our findings in mice are recapitulated in humans, we reanalyzed single cell data from publicly available datasets. We combined two different psoriatic datasets with healthy controls (GSE151177 and GSE162183). We identified 27 cell clusters in the combined dataset (Figure 8A). Importantly, we identified embryo Langerhans cells, monocyte- derived Langerhans cells, and dendritic cell type 3 (DC 3). The embryo Langerhans cell cluster has high Id2 and Runx3 (lineage- defining transcription factors for Langerhans cells) expression but low Irf8 (lineage- defining transcription factor for cells of monocyte lineage) expression.(39) MoLCs have high Irf8 expression and ITGAM (CD11b). Dendritic cell type 3 was identified by its expression of CD163 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 850, 291]]<|/det|> +but not CD11b (Figure 8B).(47) We then looked at the expression levels of HTR2A between psoriatic molCs and control molCs and found HTR2A expression on psoriatic molCs to be significantly lower (Figure 8C). Among the cytokines, IL23A and IL1B are highly expressed in molCs (Figure 8D). Expression of NFκB- related subunits were increased in psoriatic molCs (Figure 8E). We then performed immunofluorescence staining of psoriatic lesioned and nonlesioned skin samples for CD1a and HTR2A. CD1a represents Langerhans cells and we found HTR2A expression to be lower in Langerhans cells of psoriatic lesioned skin as compared to nonlesioned skin (Figure 8F). Our findings in mice closely resemble clinical conditions, as HTR2A deficiency in molCs leads to exacerbated inflammation in mice mirrors low HTR2A expression in psoriatic molCs. IL- 23 was found to be a key mediator and NFκB is enriched in psoriatic molCs leading to exacerbated inflammation similar to HTR2A deficient molCs in mice. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 266, 108]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[144, 123, 848, 268]]<|/det|> +In our retrospective cohort study, we investigated how anti- psychotics, which antagonize dopamine and serotonin receptors, affect psoriatic outcomes.(48) Patients on anti- psychotics may have worsened psoriatic outcomes and these anti- psychotics have a very strong binding affinity for HTR2A. (16) This indicates that HTR2A antagonism by anti- psychotics worsens psoriatic outcomes. While this association is far from ideal, we confirmed the pivotal role played by HTR2A in regulating psoriatic inflammation by treating human psoriatic skin with an HTR2A- specific agonist ex vivo. Besides, our in vivo murine psoriasisform model further proved HTR2A's pivotal role in suppressing psoriatic inflammation. + +<|ref|>text<|/ref|><|det|>[[143, 279, 855, 515]]<|/det|> +We also found chronic SSRI use led to worsening psoriatic outcomes. In congruent with our findings, there are reports showing patients who had taken fluoxetine (a commonly used SSRI) for extended periods experienced worsened psoriasis.(11, 13) In contrast to our findings, a retrospective study done in Sweden demonstrated an improved outcome with a decreased need for systemic psoriatic treatment following two SSRI dispensations within six months.(9) There is a lack of information on the length of treatment, dosage, and adherence reported in the Swedish study making comparisons between their study and ours impractical. Nevertheless, the conflicting findings could be due to the paradoxical effect SSRI has on serotonin levels. Since SSRI targets serotonin transporter (SERT), it could lead to a temporary increase of serotonin in between synaptic clefts, but chronic use will lead to depletion of platelet serotonin stores.(49- 52) The major source of serotonin in the periphery is in platelets.(53) While platelet serotonin depletion following chronic SSRI use is established in mice, much remains unknown about its impact on in humans. + +<|ref|>text<|/ref|><|det|>[[143, 526, 855, 783]]<|/det|> +We demonstrated that HTR2A deficiency exacerbated psoriasisform inflammation. This is congruent with a previous study that showed hypermethylation of HTR2A, which decreases HTR2A expression, resulted in exacerbated inflammation in rheumatoid arthritis (a disease that involves type 17 immunity).(54, 55) We noticed increased IL- 17- and IL- 22- producing \(\mathrm{V} / 4\mathrm{T}\) cells in our murine psoriasisform model as expected since \(\mathrm{V} / 4\mathrm{T}\) cells are the dominant type 17 immune responders in acute models, particularly psoriatic model.(35, 56) Ear swelling and epidermal thickness which is associated with IL- 22 were also increased reflecting acanthosis commonly seen in psoriasis.(57) Increases in mRNA levels of Il17a and Il23a were observed suggesting the IL- 23/IL- 17 axis plays a key role in exacerbating inflammation. We also observed increased neutrophil levels as predicted, which may be due to neutrophil recruitment at lesioned sites under psoriatic conditions.(36) However, we did not observe any increase in IL- 17- and IL- 22- producing CD4 T cells, which may be due to two reasons: i) conventional CD4 T cells in the periphery will only express IL23r upon activation; and ii) IMQ is an acute model with no antigen limiting the expansion of antigen- specific Th17 cells.(58, 59) + +<|ref|>text<|/ref|><|det|>[[144, 793, 845, 903]]<|/det|> +We delineated the cell type exacerbating inflammation in HTR2A deficient mice as it is crucial in unravelling the complex interplay between HTR2A and type 17 inflammation. HTR2A was found to have minimal impact on B and T cells, as seen in our conditional knockout mice with HTR2A deficient B and T cells. Nevertheless, we are aware our seven- day model will not be able to reveal and reflect the true extent of HTR2A's impact on these cells. Among the innate cells, it was established previously that monocyte- derived inflammatory Langerhans cells play a pivotal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 89, 852, 199]]<|/det|> +role in modulating psoriasisform inflammation.(28) In agreement with previous findings, we demonstrated that HTR2A deficiency in monocyte- derived Langerhans cells (molC) played a key role in exacerbating inflammation. We validated this using multiple models including Langerhans cells depletion, adoptive transfer, and two conditional knockout mice models. Our chimeric bone marrow transplant model further supports this, as HTR2A deficiency in hematopoietic cells, which includes molCs but not embryo Langerhans cells, exacerbates inflammation.(28) + +<|ref|>text<|/ref|><|det|>[[144, 209, 852, 319]]<|/det|> +HTR2A affects the function of molCs and embryo Langerhans cells (elC) to varying extents. Under inflammatory conditions, monocytes are recruited in waves to the inflammatory site and they serve as end- type killer cells by secreting chemokines and cytokines, exacerbating the inflammation.(60) This means molCs inherently secrete more IL- 23 than elCs. HTR2A serves as a brake to inhibit cytokine secretion rather than inducing increased cytokine secretion in molCs. This may explain why HTR2A deficiency appears to affect molCs more than elCs. + +<|ref|>text<|/ref|><|det|>[[144, 329, 855, 530]]<|/det|> +We identified molCs by their expression of MHC II, CD207, and CD64 markers. We found it hard to fit our molCs neatly into any of the P1 to P5 dermal macrophages identified by Tamoutounour et al.(61) Rather, our cells closely resemble the murine monocyte- derived macrophage identified by McGovern et al., reflecting heterogeneity in this population.(29) Importantly, molCs identified in our study have high expression of Irf8, a lineage- defining transcription factor for monocytes, and mutations in this gene will lead to genetic deficiency of monocytes and dendritic cells.(39, 62) The molCs also expressed Runx3 and Id2, lineage- defining transcription factors for Langerhans cells, as Langerhans cell development requires TGF- \(\beta\) , and Runx3 and Id2 are critical mediators downstream of TGF- \(\beta\) .(39, 63, 64) MolCs and elCs identified in the single cell RNA- sequencing data reflect this as they have high expression of Runx3 and Id2. The variation in Irf8 expression levels help us differentiate them into molCs and elCs. + +<|ref|>text<|/ref|><|det|>[[144, 540, 852, 758]]<|/det|> +IL- 23 is a key cytokine in the pathogenesis of psoriasis and drugs targeting IL- 23, like guselkumab and risankizumab, are successful in treating psoriatic symptoms.(65) However, to date, no studies have reported any association between HTR2A and IL- 23 levels. We found IL- 23 secretion to be increased drastically in HTR2A deficient molCs and this is observed in both ELISA and intracellular cytokine staining of IL- 23. Stimulation of HTR2A with DOI, a HTR2A- specific agonist, led to a significant decrease in IL- 23 secretion by molCs further supporting the notion that HTR2A expression on molCs modulates psoriatic inflammation via IL- 23. Contradicting our study, Wohn et al. found that IL- 23 was exclusively produced by Langerin- negative DCs in vivo post IMQ- painting.(66) We believe this could be explained by differences in methodology, as we did not inject Brefeldin A into the mice directly; instead, we isolated the cells, treated them with PMA, Ionomycin, and GolgiStop (containing Brefeldin A) and found Langerhans cells, particularly molCs, secrete elevated levels of IL- 23. + +<|ref|>text<|/ref|><|det|>[[144, 770, 853, 901]]<|/det|> +Further downstream, IL- 23 has been reported to induce innate IL- 17 production, especially among \(\gamma \delta\) T cells and IL- 23 has been known to play a cardinal role in mediating Imiquimod- induced psoriasisform inflammation.(67, 68) While IL- 1 \(\beta\) has also been proposed to play a role in inducing innate IL- 17 production, our qPCR of IL- 1 \(\beta\) did not reveal a drastic increase in IL- 1 \(\beta\) levels with HTR2A deficiency.(67) Besides, it has been reported that the stimulation of peritoneal exudate cells with IL- 23 alone but not IL- 1 \(\beta\) is sufficient to induce IL- 17 production by \(\gamma \delta\) T cells.(69) IL- 23 is known to play a central role in T cell differentiation instead of T cell survival and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 90, 832, 145]]<|/det|> +expansion.(70) Using our Vγ4 T cell coculture assay, we demonstrated increased Vγ4 T cell differentiation into IL- 17- and IL- 22- producing cells after coculture with molCs. This highlights that IL- 23 is the key cytokine mediating exacerbated inflammation in HTR2A deficient molCs. + +<|ref|>text<|/ref|><|det|>[[144, 156, 854, 323]]<|/det|> +IL- 23 transcription requires NFκB and HTR2A has been suggested to affect NFκB levels.(20, 71, 72) The NFκB pathway is found to be enriched in HTR2A deficient Langerhans cells. It was previously reported both canonical NFκB and non- canonical NFκB pathways are involved in IL- 23 secretion. The canonical NFκB pathway is involved in early- phase induction, and the late- phase induction peaking at 12 hours, requires the non- canonical NFκB p100 subunit.(43) Our findings are congruent with previous findings since immunoblot of our cells post 24- hour treatment showed enrichment in the non- canonical NFκB pathway. Notably, we also found enriched NFκB pathways, both canonical and non- canonical, in psoriatic molCs as compared to control molCs in our reanalysis of single cell RNA- sequencing data. + +<|ref|>text<|/ref|><|det|>[[144, 334, 854, 607]]<|/det|> +We found serotonin levels to be raised in healthy skin and non- lesioned skin samples of psoriatic patients as compared to lesioned skin via immunohistochemical staining. In contrast, there are two reports via immunohistochemical staining indicating serotonin levels were raised in psoriatic lesioned skin.(73, 74) The discrepancy could be due to the difference in controls used, as in their studies, they used the skin of healthy individuals as controls whereas we used both non- lesioned paired samples and healthy skins as controls. Our findings are further supported by our in vivo murine model, where serotonin injected intraperitoneally significantly inhibited type 17 inflammation. Congruent with our findings, a study reported serotonin modulates M2 macrophage polarization and another reported serotonin decreases Th1 and Th17 cytokines in multiple sclerosis patients.(75, 76) We further found platelet serotonin levels but not serum serotonin levels to be decreased in psoriatic patients. Using an in vivo model where we depleted platelet serotonin with chronic fluoxetine pre- treatment, our preliminary data showed platelet serotonin depletion led to exacerbated inflammation even though more experiments are needed to validate this finding.(15) Our data suggest serotonin through its receptor HTR2A plays a key role in modulating psoriatic inflammation. + +<|ref|>text<|/ref|><|det|>[[144, 618, 852, 783]]<|/det|> +Our study provides evidence that drugs antagonizing HTR2A and HTR2A deficiency led to exacerbated psoriatic inflammation. On the other hand, the HTR2A- specific agonist of DOI attenuates inflammation. IL- 23 secretion by HTR2A deficient molCs was increased, as HTR2A functioning as brakes of the immune system via its interference of NFκB pathway, was absent in HTR2A deficient molCs. Serotonin is the putative agonist of HTR2A modulating inflammation and critically, our findings were recapitulated in clinical samples. Overall, our study provides insights into the role of HTR2A in regulating psoriatic inflammation, with the cells, cytokines, and transcription factor involved identified. Thus, our findings could lead to the development of novel therapeutic interventions, providing new avenues for treating psoriasis. + +<|ref|>sub_title<|/ref|><|det|>[[145, 794, 323, 811]]<|/det|> +## Limitations of the study + +<|ref|>text<|/ref|><|det|>[[145, 822, 852, 876]]<|/det|> +Our study has been limited by our Imiquimod- induced psoriasisform murine model as it is an acute inflammatory model. Application of Imiquimod for more than 7 days will lead to resolution of psoriasisform inflammation even without external intervention.(77) Due to the short- term + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 832, 127]]<|/det|> +nature of our model, we cannot recapitulate the chronic component of psoriatic inflammation like the involvement of CD4 \(\alpha \beta\) T cells. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[145, 89, 418, 107]]<|/det|> +## MATERIALS AND METHODS + +<|ref|>sub_title<|/ref|><|det|>[[145, 124, 208, 138]]<|/det|> +## Humans + +<|ref|>text<|/ref|><|det|>[[145, 140, 850, 230]]<|/det|> +A retrospective cohort study was conducted to investigate the relationship between SSRI or antipsychotic use change in treatment received for psoriasis. This cohort study used population- based data from the Taiwan National Health Insurance (NHI) program. This program includes \(>99\%\) of Taiwanese residents and a total of 2 million patients were enrolled in this program. This study only used secondary data from this database for academic purposes. + +<|ref|>text<|/ref|><|det|>[[144, 240, 847, 350]]<|/det|> +SSRI included in this study were citalopram, escitalopram, fluoxetine, paroxetine, sertraline, and fluvoxamine. Anti- psychotic included in this study were aripiprazole, chlorpromazine, prochlorperazine, quetiapine, risperidone, sulpiride, amisulpride, clozapine, haloperidol, olanzapine, and anti- psychotic categorized as Others (brexipiprazole, fluphenazine, loxapine, lurasidone, perphenazine, pimozide, thioridazine, and trifluoperazine). This study used Anatomical Therapeutic Chemical codes to identify SSRI or anti- psychotic exposure. + +<|ref|>text<|/ref|><|det|>[[144, 360, 849, 542]]<|/det|> +Psoriatic cases were identified using the codes 696.1 and 696.8 in the International Classification of Diseases 9 (ICD9). The primary outcome is a change in the categories of psoriatic treatment received within 180 days from the index date (the date starting SSRI or anti- psychotic). Psoriatic treatments were divided into four categories as defined by the Taiwan Dermatological Association namely topical for the mildest symptoms, followed by systemic agents, combination therapy, and biologics for patients with the most severe symptoms. A worsened psoriatic outcome is defined as a change of treatment from treatment of mild symptoms to treatment of severe symptoms, for example from using systemic agents to using biologics. Improved psoriatic outcome is defined as a change of treatment from treatment for severe symptoms to treatment of mild symptoms, for example from biologics to using topical treatment. + +<|ref|>text<|/ref|><|det|>[[144, 553, 840, 661]]<|/det|> +Data from 28,056 psoriatic subjects were included in this study and they were checked for antipsychotic or SSRI exposure. 9,570 subjects were controls, 11,315 subjects were on antipsychotics, and 7,171 subjects were on SSRIs. We then performed Cox proportional hazard model to calculate the hazard ratios and \(95\%\) confidence intervals for worsened psoriatic outcome and improved psoriatic outcomes, between anti- psychotic users versus controls and SSRI users versus controls. + +<|ref|>sub_title<|/ref|><|det|>[[145, 680, 206, 694]]<|/det|> +## Animals + +<|ref|>text<|/ref|><|det|>[[144, 696, 856, 896]]<|/det|> +Htr2a floxed (Htr2aflox/flox) mice in C57BL/6J background were generated by the CRISPR/Cas9 technology. Selection of the sgRNA sequences followed the online resources, the sgRNA Designer: CRISPRko and Cas- OFFinder.(78, 79) The 5'loxP sequence was inserted in the upstream of exon 2, and 3'loxP in the 3'UTR after stop codon 134bp. The two sgRNA target sequences with PAM sites (NGG) were 5'- GTTTGATTGTGAGACATCGG - 3', and 5'- GTCCGGACAGCATTTGAACT - 3', for inserting the 5'-loxP and the 3'-loxP DNA fragments, respectively. sgRNA and Cas9 protein for electroporation were purchased from Synthego Corporation. Electroporation was performed on fertilized eggs from C57BL/6J mice. To minimize off- target effects in CRISPR mice, Htr2aflox/flox mice were first backcrossed to C57BL/6JNarl mice for 2 generations before breeding into cell- specific Htr2a knockout mice. Genotyping of founder mice was performed by PCR. PCR condition for genotyping of offspring was the initial denaturation at \(95^{\circ}C\) 5 min, followed by 45 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 88, 842, 180]]<|/det|> +cycles of \(95^{\circ}C\) for 30 sec, \(58^{\circ}C\) for 30 sec and \(72^{\circ}C\) for 30 sec, and the final extension at \(72^{\circ}C\) for 7 min. All techniques for the production of the Htr2aflox/flox mice were provided by the "Transgenic Mouse Models Core Facility of the National Core Facility for Biopharmaceuticals, National Science and Technology Council, Taiwan" and the "Gene Knockout Mouse Core Laboratory of National Taiwan University Center of Genomic and Precision Medicine". + +<|ref|>text<|/ref|><|det|>[[144, 191, 825, 262]]<|/det|> +Others mice used: C57BL/6JNarl, 129S/Sv- Tg(Prm- cre)580g/J (Jax 003328), B6. SJL- Ptprca Pepcb/BoyJ (Jax 002014), C57BL/6- Tg(Rag1- RFP, - cre/ERT2)33Narl/Narl (RMRC 13191), STOCK Tg(CD207- cre/ERT2)1Dhka/J (Jax 028287), B6.12952- Cd207tm3.1(HBEGF/EGFP)Mal/J (Jax 016940), C57BL/6J- Ms4a3em2(cre)Fgnx/J (Jax 036382). + +<|ref|>text<|/ref|><|det|>[[144, 274, 852, 438]]<|/det|> +For the generation of whole- body knockout mice, Prm- cre mice were crossed with Htr2aflox/flox mice. The double gene offspring were then crossed with wild type mice, producing heterozygous offspring. The heterozygous mice were then inbred with a 25 per cent probability of producing homozygous HTR2A deficient mice. For the generation of conditional knockout mice, Rag1- RFP- cre mice were crossed with Htr2aflox/flox mice. The offspring were then backcrossed with Htr2aflox/flox mice. The offspring that are Htr2aflox/flox mice- Rag1- RFP- cre (conditional knockout) will be bred with their littermate that are Htr2aflox/flox mice producing only Htr2aflox/flox mice- Rag1- RFP- cre (HTR2ARag1) or Htr2aflox/flox mice which serve as a littermate control. A similar breeding strategy was adopted for CD207- cre and Ms4a3em2(cre)Fgnx mice. + +<|ref|>text<|/ref|><|det|>[[144, 450, 850, 521]]<|/det|> +Animals were housed in a specific pathogen- free environment in the Experimental Animal Facility at IBMS, Academia Sinica, Taipei, Taiwan. The Institutional Animal Care and Use Committee (IACUC) approved the animal experimentation protocols used in this study. Male and female mice used for this study were aged 8- 12 weeks. + +<|ref|>sub_title<|/ref|><|det|>[[145, 538, 348, 553]]<|/det|> +## RNA-seq of human samples + +<|ref|>text<|/ref|><|det|>[[144, 555, 841, 656]]<|/det|> +Skin- punched biopsies were obtained from the lesioned sites of patients and healthy foreskin were collected from male patients undergoing circumcision. The skins were halved, with half untreated and the other half treated with 100nM of DOI for 16 hours. Total RNA was isolated with Direct- zol RNA Miniprep Kit (Zymo Research) according to the manufacturer's instructions. NGS was performed by NGS High Throughput Genomics Core at Biodiversity Research Centre, Academia Sinica, Taipei, Taiwan. + +<|ref|>sub_title<|/ref|><|det|>[[145, 672, 457, 688]]<|/det|> +## In silico analysis of HTR2A gene expression + +<|ref|>text<|/ref|><|det|>[[144, 689, 850, 857]]<|/det|> +We used the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) to search for HTR2A gene expression in skin tissue between lesioned sites versus non- lesioned sites, as well as psoriasis patients versus healthy controls. The GSE13355 includes Affymetrix HU133 Plus 2.0 microarrays for lesion skin tissues and non- lesion skin tissues from 58 psoriasis patients. (80) GSE34248 includes Affymetrix HU133 Plus 2.0 microarrays for lesion skin tissues and non- lesion skin tissues from 13 psoriasis patients. (81) GSE109248 includes Illumina HumanHT- 12 V4.0 for skin tissues from 17 psoriasis patients and 12 healthy controls. (82) GSE41664 includes Affymetrix Human Genome U133 Plus 2.0 Array of lesioned and non- lesioned skin tissues of 38 psoriasis patients. (81) The differences in gene expression levels between the two groups were calculated by a two- tailed paired Student's T- test. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 91, 373, 106]]<|/det|> +Imiquimod psoriasiform model + +<|ref|>text<|/ref|><|det|>[[145, 108, 850, 141]]<|/det|> +For mice to develop psoriasis, we applied 15mg of \(5\%\) Imiquimod (Aldara cream, iNova Pharmaceuticals, Australia) for 7 days on the ears of mice. The ears were harvested the day after. + +<|ref|>sub_title<|/ref|><|det|>[[145, 157, 261, 172]]<|/det|> +## Flow cytometry + +<|ref|>text<|/ref|><|det|>[[144, 174, 850, 394]]<|/det|> +Cells were isolated from the ears by incubating overnight at \(37^{\circ}C\) with Dispase II (Thermo Scientific). The next day, the ear is then cut by scissors into small pieces and then immersed in RPMI (Gibco) containing Collagenase IV (Gibco) and DNase I (Bio Basic) for 90 minutes. The sample is then filtered using a \(70\mu \mathrm{m}\) cell strainer. The isolated single cells were then stained directly with fluorescent antibodies or treated with PMA (Merck), lonomycin (BioGems), and GolgiStop (BD) if intracellular cytokine staining was required. Cells were incubated with LIVE/DEAD Fixable Blue Dead Cell Stain Kit (Thermo Fisher Scientific) for the exclusion of dead cells. The Fc- blocking antibody was then used to prevent non- specific binding followed by incubation with specific antibodies against surface antigens. For intracellular cytokine staining, cells were stimulated with PMA (Merck) and lonomycin (BioGems) in the presence of GolgiStop (BD) at \(37^{\circ}C\) for four hours. The cells were fixed and permeabilized with Cytofix/Cytoperm buffer (BD) according to the manufacturer's instructions. For all samples, acquisition was performed on LSRI (BD). + +<|ref|>text<|/ref|><|det|>[[144, 409, 545, 425]]<|/det|> +RNA purification and real- time quantitative PCR (qPCR) + +<|ref|>text<|/ref|><|det|>[[144, 427, 849, 595]]<|/det|> +Total RNA from tissue was immersed in TRIzol (Invitrogen, 15596026) overnight, homogenized using steel beads together with Bullet Blender Tissue Homogenizer (Next Advance), and total RNA was extracted using RNA MiniPrep Kit (Zymo Research, R2050) according to manufacturer's instructions. The quantity of the total RNA was measured by NanoDrop Spectrophotometer (Thermo Scientific). For qPCR analysis, equivalent amounts of RNA were reverse- transcribed by Maxima H Minus First Strand cDNA Synthesis Kit (Thermo Scientific, K1652). cDNAs were mixed with indicated primers and PowerUp SYBR Green Master Mix (Applied Biosystems, A25742), and RT- qPCR was performed on Quant Studio 5 Real- time PCR System (Applied Biosystems). cDNAs were normalized to equal amounts using primers against Gapdh. The primer sequences can be found in the supplementary table. + +<|ref|>text<|/ref|><|det|>[[145, 606, 352, 621]]<|/det|> +Chimeric bone marrow mice + +<|ref|>text<|/ref|><|det|>[[144, 622, 848, 807]]<|/det|> +CD45.1 mice (Jax 002014) were fed with water added with Trimethoprim and Sulfamethoxazole (Merck, SML3191, with a final concentration \(1.92\mathrm{mg / ml}\) ) for two weeks before they were lethally irradiated (1200 rads) in split dose, with three hours of rest in between, one day before the experiment. Bone marrow from HTR2A deficient mice was harvested on the day of the experiment. 5mL of ACK Lysing Buffer (Gibco, A1049201) was added to the cells for five minutes to lyse red blood cells. The cells were then washed with Phosphate Buffered Saline and quantified. A total of \(5\times 10^{5}\) bone marrow cells were intravenously injected into the lethally irradiated mice. The mice were continuously fed antibiotic- laced water for another 14 days before experiments were carried out on them 12 weeks after transplantation. We also transplanted the bone marrow of CD45.1 mice to HTR2A deficient mice and the bone marrow of HTR2A deficient mice to HTR2A deficient mice. + +<|ref|>text<|/ref|><|det|>[[145, 824, 481, 839]]<|/det|> +Culture of monocyte- derived Langerhans cells + +<|ref|>text<|/ref|><|det|>[[145, 841, 828, 891]]<|/det|> +Bone marrow cells were harvested from the femur and tibia of wild type or HTR2A deficient mice. The cells were lysed with ACK Lysing Buffer (Gibco, A1049201). The cells were then resuspended in RPMI (Gibco) containing \(20\mathrm{ng / ml}\) of recombinant mouse GM- CSF (PeproTech, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 90, 853, 212]]<|/det|> +315- 03), 5ng/ml of TGF- \(\beta\) (BioLegend, 763104), and 8ng/ml of IL- 34 (BioLegend, 577604), plated on a 6 well plate, and incubated for 8 days at \(37^{\circ}C\) . The medium was changed every 48 hours. Monocyte- derived Langerhans cells were isolated by a FACSAria IIIu Sorter with \(\alpha\) - CD207, \(\alpha\) - CD64, and \(\alpha\) - MHC II staining. LCs were stimulated with \(10\mu \mathrm{g / mL}\) Imiquimod (Enzo, ALX- 420- 039- M100) and \(10\mu \mathrm{g / ml}\) of DOI (Merck, D101) or \(0.1\mu \mathrm{M}\) of EVP 4593 (MCE, HY- 13812) were added for 24 hours to investigate IL- 23 expression and its ability to induce differentiation of \(\mathrm{V}\gamma 4\mathrm{T}\) cells in a coculture assay. + +<|ref|>sub_title<|/ref|><|det|>[[145, 230, 230, 244]]<|/det|> +## ELISA assay + +<|ref|>text<|/ref|><|det|>[[145, 245, 854, 329]]<|/det|> +ELISA assayIL- 23p19 expression was measured using Mouse IL- 23p19 ELISA Kit (Elabscience, E- EL- M0731) according to the manufacturer's instructions. Mouse and human serotonin levels were measured using a Serotonin ELISA Kit (Enzo, ADI- 900- 175) according to the manufacturer's protocol. The absorbance at \(450\mathrm{nm}\) (Mouse IL- 23 ELISA kit) or \(405\mathrm{nm}\) (Serotonin ELISA Kit) was measured with an ELISA reader (Infinite M1000 PRO, Tecan). + +<|ref|>sub_title<|/ref|><|det|>[[145, 345, 261, 360]]<|/det|> +## Coculture assay + +<|ref|>text<|/ref|><|det|>[[144, 361, 848, 500]]<|/det|> +Spleens of wild type mice were surgically removed and mechanically disrupted by meshing it against a \(70\mu \mathrm{m}\) cell strainer (Jet Biofil, CSS013070). ACK Lysing Buffer (Gibco, A1049201) was added to lyse red blood cells. The cells were then isolated by a FACSAria IIIu Sorter with \(\alpha\) - CD45, \(\alpha\) - CD3, \(\alpha\) - \(\gamma \delta\) , and \(\alpha\) - \(\mathrm{V}\gamma 4\) staining. Monocyte- derived Langerhans cells post- stimulation with DOI or EVP 4593, were incubated with these \(\mathrm{V}\gamma 4\mathrm{T}\) cells for 72 hours in an \(\alpha\) - CD3- coated plate. \(\alpha\) - CD28 was then added to further stimulate \(\mathrm{V}\gamma 4\mathrm{T}\) cells. Post- 72 hours, the cells were then stimulated with PMA, and lonomycin, and incubated with GolgiStop for four hours. The cells were then stained for IL- 17A and IL- 22. The acquisition was performed on LSRII (BD). + +<|ref|>sub_title<|/ref|><|det|>[[146, 516, 346, 531]]<|/det|> +## RNA-seq of mouse samples + +<|ref|>text<|/ref|><|det|>[[145, 532, 853, 633]]<|/det|> +The experiments generating datasets of GSE222197 and GSE274941 were done in parallel. MolCs (defined as \(\mathrm{CD45 + MHCII + CD11c + CD207 + CD64 + }\) ) were pooled from the ears of five mice treated with IMQ for two days, constituting a single biological sample. Cells were sorted and total RNA was isolated with a Direct- zol RNA Microprep Kit (Zymo Research) according to the manufacturer's instructions. NGS was performed by NGS High Throughput Genomics Core at Biodiversity Research Centre, Academia Sinica, Taipei, Taiwan. + +<|ref|>text<|/ref|><|det|>[[145, 650, 500, 665]]<|/det|> +Protein gel electrophoresis and immunoblotting + +<|ref|>text<|/ref|><|det|>[[144, 666, 855, 903]]<|/det|> +Protein gel electrophoresis and immunoblottingTotal cells lysates were prepared using RIPA (20mM Tris HCl, pH8.0, 150mM NaCl, 1mM EDTA, 1mM EGTA, 1% TritonX- 100, 0.5% deoxycholate, 0.1% SDS) supplemented with protease inhibitors (Halt Protease Inhibitor, Thermo Scientific, 78430), or used directly with 1X protein sample loading buffer (50 mM Tris- HCl, pH 6.8, 10% glycerol, 2% SDS, 0.1 M DTT, 0.04% Orange G). The protein concentrations of clarified supernatants were measured by a BCA Protein Assay Kit (Pierce, 23227). Proteins in lysates were separated by Bolt 4- 12% Bis- Tris Plus gels (Invitrogen) and transferred to a PVDF membrane (Thermo Scientific, 88520) by a Mini Trans- Blot cell system (BioRad). Transferred membranes were blocked for 1 h in 5% Bovine Serum Albumin (Merck, A7030) diluted in TBS (20 mM Tris- HCl, pH 7.4, 150 mM NaCl) and then incubated with primary antibodies in TBS with 0.1% Tween- 20 (TBST) overnight at 4°C. Primary antibodies used were as the key resources table. After washing with TBST, membranes were incubated with secondary antibodies in 3% BSA in TBST for one hour at room temperature. A secondary anti- mouse antibody conjugated with HRP (1:10000, Thermo Scientific, 31430) or anti- rabbit antibody conjugated with HRP (1:10000, Thermo Scientific, 31460) was used. Membranes were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 90, 843, 141]]<|/det|> +washed in TBST and then incubated in SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Scientific, 34580) for 2 minutes. Membranes were immediately imaged on UVP ChemStudio Western Blot Imaging System (Analytik Jena, 849- 97- 0847- 03). + +<|ref|>text<|/ref|><|det|>[[145, 157, 343, 172]]<|/det|> +Immunostaining of section + +<|ref|>text<|/ref|><|det|>[[144, 174, 845, 343]]<|/det|> +Paraffin- embedded tissue sections were deparaffinized with xylene and rehydrated with serial passage through changes of graded ethanol. Slides were subjected to heat- induced epitope retrieval in EDTA solution (10 mM Tris, 1 mM EDTA, \(0.05\%\) Tween 20, pH 8) followed by blocking with \(5\%\) goat serum or \(3\% \mathrm{H}_2\mathrm{O}_2\) . Tissues were incubated with primary \(\alpha\) - serotonin (YC5/45, Abcam) Abs at \(4^{\circ}C\) overnight. Slides were washed and incubated with HRP- carried Abs (Leadgene) for 1 hour at room temperature. In the immunohistochemistry assay, DAB (Roche, 11718096001) was used as a chromogen to visualize peroxidase activity, and the reaction was stopped by the addition of PBS. The preparations were lightly counterstained with hematoxylin (Sigma, H3136), mounted with Fluoromount Aqueous mounting medium (Sigma, F4680), and examined by Pannoramic 250 FLASH II (3DHistech). + +<|ref|>text<|/ref|><|det|>[[146, 358, 416, 374]]<|/det|> +Quantification and statistical analysis + +<|ref|>text<|/ref|><|det|>[[144, 376, 848, 485]]<|/det|> +FlowJo v10 was used to quantify flow cytometry data, and GraphPad Prism 9 was used for statistical analysis. Student's T- test was used to analyze experiments with two conditions, Student's paired T- test was used to analyze experiments with paired samples, one- way ANOVA was used for experiments with three or more conditions, and two- way ANOVA was used for changes in ear swellings. Statistical details of experiments, including the number of animals, and corresponding analyses are reported in the figure legends. + +<|ref|>sub_title<|/ref|><|det|>[[145, 513, 525, 532]]<|/det|> +## LIST OF SUPPLEMENTARY MATERIALS + +<|ref|>text<|/ref|><|det|>[[144, 543, 627, 560]]<|/det|> +Figure S1. Anti- psychotic and SSRI use worsen psoriatic outcomes. + +<|ref|>text<|/ref|><|det|>[[144, 572, 693, 590]]<|/det|> +Figure S2. Gating strategy for flow cytometry of \(\mathrm{V} / \gamma 4\mathrm{T}\) cells and neutrophils. + +<|ref|>text<|/ref|><|det|>[[144, 601, 690, 618]]<|/det|> +Figure S3. The composition of cells changed after bone marrow transplant. + +<|ref|>text<|/ref|><|det|>[[144, 630, 792, 647]]<|/det|> +Figure S4. HTR2A deficiency in adaptive immune cells does not exacerbate inflammation. + +<|ref|>text<|/ref|><|det|>[[144, 658, 845, 676]]<|/det|> +Figure S5. Depletion of different Langerhans cell subsets using varying doses of diphtheria toxin. + +<|ref|>text<|/ref|><|det|>[[144, 687, 847, 704]]<|/det|> +Figure S6. Characterization of monocyte- derived Langerhans cells developed from bone marrow. + +<|ref|>text<|/ref|><|det|>[[144, 715, 803, 732]]<|/det|> +Figure S7. Gating strategy for identifying various cell clusters after spectral flow cytometry. + +<|ref|>text<|/ref|><|det|>[[144, 743, 825, 778]]<|/det|> +Figure S8. Platelet serotonin is lower in psoriatic patients and its deficiency is associated with exacerbated inflammation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 89, 396, 107]]<|/det|> +## REFERENCES AND NOTES + +<|ref|>text<|/ref|><|det|>[[140, 117, 852, 900]]<|/det|> +1. A. Rendon, K. Schakel, Psoriasis Pathogenesis and Treatment. Int J Mol Sci 20, (2019). +2. A. Di Cesare, P. Di Meglio, F. O. Nestle, The IL-23/Th17 axis in the immunopathogenesis of psoriasis. J Invest Dermatol 129, 1339-1350 (2009). +3. E. Rieder, F. Tausk, Psoriasis, a model of dermatologic psychosomatic disease: psychiatric implications and treatments. Int J Dermatol 51, 12-26 (2012). +4. Eskeland S., Halvorsen J. A., T. L., Antidepressants have Anti-inflammatory Effects that may be Relevant to Dermatology: A Systematic Review. Acta Derm Venereol 97, 897-905 (2017). +5. C. W. Kwon, R. G. Fried, Y. Nousari, C. Ritchlin, F. Tausk, Psoriasis: Psychosomatic, somatopsychic, or both? Clin Dermatol 36, 698-703 (2018). +6. I. M. Michalek, B. Loring, S. M. John, A systematic review of worldwide epidemiology of psoriasis. J Eur Acad Dermatol Venereol 31, 205-212 (2017). +7. L. Rousset, B. Halioua, Stress and psoriasis. Int J Dermatol 57, 1165-1172 (2018). +8. S. Luis Blay, Depression and psoriasis comorbidity. Treatment with paroxetine: two case reports. Ann Clin Psychiatry 18, 271-272 (2006). +9. K. Thorslund, T. Svensson, K. Nordlind, A. Ekbom, C. M. Fored, Use of serotonin reuptake inhibitors in patients with psoriasis is associated with a decreased need for systemic psoriasis treatment: a population-based cohort study. J Intern Med 274, 281-287 (2013). +10. Y. M. Tzeng et al., Protective Effects of Anti-depressants against the Subsequent Development of Psoriasis in Patients with Major Depressive Disorder: a Cohort Study. J Affect Disord 281, 590-596 (2021). +11. R. J. Hemlock C, Winston A, Fluoxetine-induced psoriasis. Annals of Pharmacotherapy 26, 211-212 (1992). +12. S. F. Osborne, L. Stafford, K. G. D. Orr, Paroxetine-Associated Psoriasis. The American Journal of Psychiatry, (2002). +13. L. Tan P.L., S. K. Kwek, Onset of psoriasis during therapy with fluoxetine. Gen Hosp Psychiatry 32, 446 e449-446 e410 (2010). +14. J. Lyne, B. Kelly, W. O. Connor, Schizophrenia: a review of neuropharmacology. Irish Journal of Medical Science 173, 155-159 (2004). +15. Mohamad F. Masri, Chinmay K. Mantri, Abhay P. S. Rathore, A. L. S. John, Peripheral serotonin causes dengue virus-induced thrombocytopenia through 5HT2A receptors. Blood 133, 2325-2337 (2019). +16. J. Masson, M. B. Emerit, M. Hamon, M. Darmon, Serotonergic signaling: multiple effectors and pleiotropic effects. Wiley Interdisciplinary Reviews: Membrane Transport and Signaling 1, 685-713 (2012). +17. K. Kim et al., Structure of a Hallucinogen-Activated Gq-Coupled 5-HT(2A) Serotonin Receptor. Cell 182, 1574-1588 e1519 (2020). +18. J. F. Lopez-Gimenez, J. Gonzalez-Maeso, Hallucinogens and Serotonin 5-HT(2A) Receptor-Mediated Signaling Pathways. Curr Top Behav Neurosci 36, 45-73 (2018). +19. M. Pelletier, R. M. Siegel, Wishing away inflammation? New links between serotonin and TNF signaling. Mol Interv 9, 299-301 (2009). +20. B. Yu et al., Serotonin 5-hydroxytryptamine(2A) receptor activation suppresses tumor necrosis factor-alpha-induced inflammation with extraordinary potency. J Pharmacol Exp Ther 327, 316-323 (2008). +21. F. Nau, Jr., B. Yu, D. Martin, C. D. Nichols, Serotonin 5-HT2A receptor activation blocks TNF-alpha mediated inflammation in vivo. PLoS One 8, e75426 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 850, 870]]<|/det|> +22. C. Y. Yeh et al., PD-L1 Enhanced by cis-Urocanic Acid on Langerhans Cells Inhibits Vgamma4(+) gammadeltaT17 Cells in Imiquimod-Induced Skin Inflammation. J Invest Dermatol 143, 1449-1460 (2023).23. B. Kirby, Langerhans cells in psoriasis: getting to the core of the disease. Br J Dermatol 178, 1240 (2018).24. M. Otsuka, G. Egawa, K. Kabashima, Uncovering the Mysteries of Langerhans Cells, Inflammatory Dendritic Epidermal Cells, and Monocyte-Derived Langerhans Cell-Like Cells in the Epidermis. Front Immunol 9, 1768 (2018).25. L. Zhou, A. Jiang, J. Veenstra, D. M. Ozog, Q. S. Mi, The Roles of Skin Langerhans Cells in Immune Tolerance and Cancer Immunity. Vaccines (Basel) 10, (2022).26. M. Collin, P. Milne, Langerhans cell origin and regulation. Curr Opin Hematol 23, 28-35 (2016).27. C. Langlet et al., CD64 expression distinguishes monocyte-derived and conventional dendritic cells and reveals their distinct role during intramuscular immunization. J Immunol 188, 1751-1760 (2012).28. T. P. Singh et al., Monocyte-derived inflammatory Langerhans cells and dermal dendritic cells mediate psoriasis-like inflammation. Nat Commun 7, 13581 (2016).29. N. McGovern et al., Human dermal CD14(+) cells are a transient population of monocyte-derived macrophages. Immunity 41, 465-477 (2014).30. Consensus Statement on Management of Psoriasis (2017).31. BL Roth, WK Kroeze, S Patel, E. Lopez, The Multiplicity of Serotonin Receptors: Uselessly diverse molecules or an embarrassment of riches? The Neuroscientist 6, 252-262 (2000).32. T. P. Singh, H. H. Zhang, S. T. Hwang, J. M. Farber, IL-23- and Imiquimod-Induced Models of Experimental Psoriasis in Mice. Curr Protoc Immunol 125, e71 (2019).33. Y. Cai et al., Pivotal role of dermal IL-17-producing gammadelta T cells in skin inflammation. Immunity 35, 596-610 (2011).34. R. Castillo-Gonzalez, D. Cibrian, F. Sanchez-Madrid, Dissecting the complexity of gammadelta T-cell subsets in skin homeostasis, inflammation, and malignancy. J Allergy Clin Immunol 147, 2030-2042 (2021).35. S. Pantelyushin et al., Roryt+ innate lymphocytes and gammadelta T cells initiate psoriasisform plaque formation in mice. J Clin Invest 122, 2252-2256 (2012).36. C. C. Chiang, W. J. Cheng, M. Korinek, C. Y. Lin, T. L. Hwang, Neutrophils in Psoriasis. Front Immunol 10, 2376 (2019).37. M. P. McCormack, A. Forster, L. Drynan, R. Pannell, T. H. Rabbitts, The LMO2 T-cell oncogene is activated via chromosomal translocations or retroviral insertion during gene therapy but has no mandatory role in normal T-cell development. Mol Cell Biol 23, 9003-9013 (2003).38. D. H. Kaplan et al., Autocrine/paracrine TGFbeta1 is required for the development of epidermal Langerhans cells. J Exp Med 204, 2545-2552 (2007).39. W. H. C. Ferrer I. R., Henderson S., Ushakov D. S., Santos e Sousa P., Strid J., Chakraverty R., Yates A. J., Bennett C. L., A wave of monocytes is recruited to replenish the long term LC network after immune injury. Science Immunology 4, eaax8704 (2019).40. L. Bellmann et al., Notch-Mediated Generation of Monocyte-Derived Langerhans Cells: Phenotype and Function. J Invest Dermatol 141, 84-94 e86 (2021).41. Z. Liu et al., Fate Mapping via Ms4a3-Expression History Traces Monocyte-Derived Cells. Cell 178, 1509-1525 e1519 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 852, 904]]<|/det|> +42. M. Nakagawa et al., IL-12 secretion by Langerhans cells stimulated with Candida skin test reagent is mediated by dectin-1 in some healthy individuals. Cytokine 65, 202-209 (2014). +43. X. Ma, T. Aoki, S. Narumiya, Prostaglandin E2-EP4 signaling persistently amplifies CD40-mediated induction of IL-23 p19 expression through canonical and non-canonical NF-kappaB pathways. Cell Mol Immunol 13, 240-250 (2016). +44. K. Li et al., Glutamine Metabolism Underlies the Functional Similarity of T Cells between Nile Tilapia and Tetrapod. Adv Sci (Weinh) 10, e2201164 (2023). +45. Bhattacharyya S., Puri S., Miledi R., P. M. M., Internalization and recycling of 5-HT2A receptors activated by serotonin and protein kinase C-mediated mechanisms. Proceedings of the National Academy of Sciences 99, 14470-14475 (2002). +46. Charles P. Mercado, F. Kilic, Molecular Mechanisms of SERT in Platelets: Regulation of Plasma Serotonin Levels. Molecular Interventions 10, 231-241 (2010). +47. C. A. Dutertre et al., Single-Cell Analysis of Human Mononuclear Phagocytes Reveals Subset-Defining Markers and Identifies Circulating Inflammatory Dendritic Cells. Immunity 51, 573-589 e578 (2019). +48. L. S. Chokhawala K. (StatPearls Publishing, Treasure Island (FL), 2024 Jan), vol. 2024. +49. Z. Zhou et al., Antidepressant specificity of serotonin transporter suggested by three LeuT-SSR structures. Nat Struct Mol Biol 16, 652-657 (2009). +50. D. Duerschmied et al., Platelet serotonin promotes the recruitment of neutrophils to sites of acute inflammation in mice. Blood 121, 1008-1015 (2013). +51. M. Rami et al., Chronic Intake of the Selective Serotonin Reuptake Inhibitor Fluoxetine Enhances Atherosclerosis. Arterioscler Thromb Vasc Biol 38, 1007-1019 (2018). +52. Jean-Claude Alvarez, Martin Sanceaume, Charles Advenier, O. Spreux-Varoquaux, Differential changes in brain and platelet 5-HT concentrations after steady-state achievement and repeated administration of antidepressant drugs in mice. European Neuropsychopharmacology 10, 31-36 (1999). +53. D. Keszthelyi, F. J. Troost, A. A. Masclee, Understanding the role of tryptophan and serotonin metabolism in gastrointestinal function. Neurogastroenterol Motil 21, 1239-1249 (2009). +54. B. W. Kirkham, A. Kavanaugh, K. Reich, Interleukin-17A: a unique pathway in immunemediated diseases: psoriasis, psoriatic arthritis and rheumatoid arthritis. Immunology 141, 133-142 (2014). +55. J. Zhao et al., Circulating methylation level of HTR2A is associated with inflammation and disease activity in rheumatoid arthritis. Front Immunol 13, 1054451 (2022). +56. Y. Li et al., Vg4 gd T Cells Provide an Early Source of IL-17A and Accelerate Skin Graft Rejection. J Invest Dermatol 137, 2513-2522 (2017). +57. Y. Zheng et al., Interleukin-22, a T(H)17 cytokine, mediates IL-23-induced dermal inflammation and acanthosis. Nature 445, 648-651 (2007). +58. H. Li et al., IL-23 promotes TCR-mediated negative selection of thymocytes through the upregulation of IL-23 receptor and RORgammat. Nat Commun 5, 4259 (2014). +59. M. Jabeen et al., Advanced Characterization of Imiquimod-Induced Psoriasis-Like Mouse Model. Pharmaceutics 12, (2020). +60. P. Italiani, D. Boraschi, From Monocytes to M1/M2 Macrophages: Phenotypical vs. Functional Differentiation. Front Immunol 5, 514 (2014). +61. S. Tamoutounour et al., Origins and functional specialization of macrophages and of conventional and monocyte-derived dendritic cells in mouse skin. Immunity 39, 925-938 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 850, 124]]<|/det|> +62. S. Hambleton et al., IRF8 Mutations and Human Dendritic-Cell Immunodeficiency. New England Journal of Medicine 365, 127-138 (2011). + +<|ref|>text<|/ref|><|det|>[[140, 123, 850, 158]]<|/det|> +63. L. X. Heinz et al., Differential involvement of PU.1 and Id2 downstream of TGF-beta1 during Langerhans-cell commitment. Blood 107, 1445-1453 (2006). + +<|ref|>text<|/ref|><|det|>[[140, 157, 844, 192]]<|/det|> +64. O. Fainaru et al., Runx3 regulates mouse TGF-beta-mediated dendritic cell function and its absence results in airway inflammation. EMBO J 23, 969-979 (2004). + +<|ref|>text<|/ref|><|det|>[[140, 191, 820, 209]]<|/det|> +65. P. Hu et al., The Role of Helper T Cells in Psoriasis. Front Immunol 12, 788940 (2021). + +<|ref|>text<|/ref|><|det|>[[140, 208, 850, 243]]<|/det|> +66. C. Wohn et al., Langerin(neg) conventional dendritic cells produce IL-23 to drive psoriatic plaque formation in mice. Proc Natl Acad Sci U S A 110, 10723-10728 (2013). + +<|ref|>text<|/ref|><|det|>[[140, 242, 830, 293]]<|/det|> +67. C. E. Sutton et al., Interleukin-1 and IL-23 induce innate IL-17 production from gammadelta T cells, amplifying Th17 responses and autoimmunity. Immunity 31, 331-341 (2009). + +<|ref|>text<|/ref|><|det|>[[140, 293, 846, 345]]<|/det|> +68. W. C. Chen et al., IL-23/IL-17 immune axis mediates the imiquimod-induced psoriatic inflammation by activating ACT1/TRAF6/TAK1/NF-kappaB pathway in macrophages and keratinocytes. Kaohsiung J Med Sci 39, 789-800 (2023). + +<|ref|>text<|/ref|><|det|>[[140, 344, 850, 396]]<|/det|> +69. K. Shibata, H. Yamada, H. Hara, K. Kishihara, Y. Yoshikai, Resident Vdelta1+ gammadelta T cells control early infiltration of neutrophils after Escherichia coli infection via IL-17 production. J Immunol 178, 4466-4472 (2007). + +<|ref|>text<|/ref|><|det|>[[140, 395, 790, 430]]<|/det|> +70. P. Thakker et al., IL-23 is critical in the induction but not in the effector phase of experimental autoimmune encephalomyelitis. J Immunol 178, 2589-2598 (2007). + +<|ref|>text<|/ref|><|det|>[[140, 429, 850, 464]]<|/det|> +71. S. Mise-Omata et al., A proximal kappaB site in the IL-23 p19 promoter is responsible for RelA- and c-Rel-dependent transcription. J Immunol 179, 6596-6603 (2007). + +<|ref|>text<|/ref|><|det|>[[140, 463, 850, 498]]<|/det|> +72. R. J. Carmody, Q. Ruan, H. C. Liou, Y. H. Chen, Essential roles of c-Rel in TLR-induced IL-23 p19 gene expression in dendritic cells. J Immunol 178, 186-191 (2007). + +<|ref|>text<|/ref|><|det|>[[140, 497, 850, 531]]<|/det|> +73. J. Huang, G. Li, J. Xiang, D. Yin, R. Chi, Immunohistochemical study of serotonin in lesions of psoriasis. Int J Dermatol 43, 408-411 (2004). + +<|ref|>text<|/ref|><|det|>[[140, 530, 797, 565]]<|/det|> +74. S. F. Younes, O. A. Bakry, Immunohistochemical Evaluation of Role of Serotonin in Pathogenesis of Psoriasis. J Clin Diagn Res 10, EC05-EC09 (2016). + +<|ref|>text<|/ref|><|det|>[[140, 565, 835, 599]]<|/det|> +75. M. de las Casas-Engel et al., Serotonin skews human macrophage polarization through HTR2B and HTR7. J Immunol 190, 2301-2310 (2013). + +<|ref|>text<|/ref|><|det|>[[140, 599, 844, 650]]<|/det|> +76. P. M. Sacramento et al., Serotonin decreases the production of Th1/Th17 cytokines and elevates the frequency of regulatory CD4(+) T-cell subsets in multiple sclerosis patients. Eur J Immunol 48, 1376-1388 (2018). + +<|ref|>text<|/ref|><|det|>[[140, 650, 840, 685]]<|/det|> +77. X. Wang, J. Sun, J. Hu, IMQ Induced K14-VEGF Mouse: A Stable and Long-Term Mouse Model of Psoriasis-Like Inflammation. PLoS One 10, e0145498 (2015). + +<|ref|>text<|/ref|><|det|>[[140, 684, 847, 719]]<|/det|> +78. J. G. Doench et al., Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. Nat Biotechnol 34, 184-191 (2016). + +<|ref|>text<|/ref|><|det|>[[140, 718, 830, 770]]<|/det|> +79. S. Bae, J. Park, J. S. Kim, Cas-OFFinder: a fast and versatile algorithm that searches for potential off-target sites of Cas9 RNA-guided endonucleases. Bioinformatics 30, 1473-1475 (2014). + +<|ref|>text<|/ref|><|det|>[[140, 769, 825, 804]]<|/det|> +80. R. P. Nair et al., Genome-wide scan reveals association of psoriasis with IL-23 and NF-kappaB pathways. Nat Genet 41, 199-204 (2009). + +<|ref|>text<|/ref|><|det|>[[140, 803, 825, 854]]<|/det|> +81. J. Bigler, H. A. Rand, K. Kerkof, M. Timur, C. B. Russell, Cross-study homogeneity of psoriasis gene expression in skin across a large expression range. PLoS One 8, e52242 (2013). + +<|ref|>text<|/ref|><|det|>[[140, 854, 825, 889]]<|/det|> +82. P. Mande et al., Fas ligand promotes an inducible TLR-dependent model of cutaneous lupus-like inflammation. J Clin Invest 128, 2966-2978 (2018). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[146, 89, 377, 107]]<|/det|> +## ACKNOWLEDGEMENTS + +<|ref|>text<|/ref|><|det|>[[144, 118, 853, 245]]<|/det|> +We thank the Academia Sinica Core Facility and Innovative Instrument Project (AS- CFII108- 113) for providing cell sorting services. We also thank the Light Microscopy Core Facility and Pathology Core Facility of the Institute of Biomedical Sciences, Academia Sinica for their technical assistance. We would also like to thank the technical services provided by the Transgenic Mouse Model Core Facility of the National Core Facility for Biopharmaceuticals, The National Science and Technology Council (NSTC), Taiwan and the Animal Resources Laboratory of National Taiwan University Centers of Genomic and Precision Medicine. + +<|ref|>sub_title<|/ref|><|det|>[[145, 257, 238, 273]]<|/det|> +## FUNDING + +<|ref|>text<|/ref|><|det|>[[145, 285, 850, 320]]<|/det|> +This study was supported by grants UN108- 015 and UN110- 032 from National Taiwan University Hospital; and the Institute of Biomedical Sciences in Academia Sinica. + +<|ref|>sub_title<|/ref|><|det|>[[146, 332, 405, 350]]<|/det|> +## AUTHOR CONTRIBUTIONS + +<|ref|>text<|/ref|><|det|>[[145, 361, 830, 433]]<|/det|> +Conceptualization: YFT, YLL; Methodology: YFT; Investigation: YFT, SYH; Visualization: YFT; Funding Acquisition: TFT, YLL; Project Administration: YLL; Resources: CYY, SSH, PCC, HJW, TFT; Software: YFT, CHL, CHT; Supervision: YLL; Writing- Original Draft: YFT; Writing- Review and Editing: YFT, YLL. + +<|ref|>sub_title<|/ref|><|det|>[[146, 444, 386, 462]]<|/det|> +## COMPETING INTERESTS + +<|ref|>text<|/ref|><|det|>[[145, 473, 822, 545]]<|/det|> +TFT has conducted clinical trials or received honoraria for serving as a consultant for AbbVie, AnaptysBio, Bristol- Myers Squibb, Boehringer Ingelheim, Celgene, Eli Lilly, Galderma, GlaxoSmithKline- Stiefel, Janssen- Cilag, Leo- Pharma, Merck, Novartis, PharmaEssentia, Pfizer, Sanofi, Sun Pharma and UCB. The remaining authors state no conflict of interest. + +<|ref|>text<|/ref|><|det|>[[145, 556, 825, 610]]<|/det|> +DATA AND MATERIALS AVAILABILITY: All data, code, and materials used in the analysis are available upon request. Mice are available upon request and subject to materials transfer agreements (MTAs). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 238, 106]]<|/det|> +## FIGURES + +<|ref|>image<|/ref|><|det|>[[144, 145, 839, 670]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[152, 121, 214, 137]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[143, 688, 852, 908]]<|/det|> +Figure 1. Modulation and expression levels of HTR2A affect psoriasis. A) Kaplan- Meier plot of anti- psychotic use and worsening psoriatic outcome. B) Forrest plot showing each anti- psychotic drug and its respective hazard ratio on worsening psoriatic outcome. C) Table showing ranking of anti- psychotic drug's affinity for HTR2A among serotonin receptors. D) Schematic depicting the experimental plan of ex vivo experiment from human samples. E) Venn diagram showing the number of genes overlapping between differentially expressed genes of healthy non- treated versus lesioned non- treated and differentially expressed genes of lesioned non- treated versus lesioned DOI- treated. F) Heat map showing 107 overlapping genes and their expression levels from E) (n=3). G) Box plot showing HTR2A expression levels in psoriatic cases as compared to control in publicly available GEO datasets. Data are a summary of two independent experiments (E- F). p values determined by Cox proportional hazard test (A) and unpaired Student's T- test (G). Mean \(\pm\) SEM (B, and G). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.0001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 112, 850, 590]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 601, 852, 900]]<|/det|> +
Figure 2. HTR2A deficiency in mice and hematopoietic stem cells exacerbates psoriasisinflammation. A) Schematic showing the experimental plan for Imiquimod-treatment of mice. B) Line chart showing changes in ear swelling of mice. (n=4 mice for Vas groups, n=10 mice for IMQ groups) C) Pictures showing ears of mice of various groups on Day 7. D) H&E-stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ \(\mathsf{V}_7 / 4\mathsf{T}\) cells and neutrophils in the respective groups with quantifications. The cells were isolated from the ears of mice. The detailed method is in Material and Methods section. F) qPCR of pro-inflammatory cytokines of respective groups. G) Schematic showing the experimental plan for developing bone marrow chimeric mice. H) Line chart showing changes in ear swelling of mice. (n=5 to 7 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E-stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ \(\mathsf{V}_7 / 4\mathsf{T}\) cells and neutrophils in the respective groups with quantifications. L) qPCR of pro-inflammatory cytokines of respective groups. Data are representative of three independent experiments (B-F and H-L). p values determined by two-way ANOVA (B and H) followed by Tukey's post-hoc test and one-way ANOVA (D-F and J-L) followed by Tukey's post-hoc test. Mean ± SEM (B, D-F, H, and J-L). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001; ns, non-significant.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[144, 102, 850, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 80, 216, 95]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[142, 576, 853, 871]]<|/det|> +Figure 3. HTR2A deficiency in Langerhans cells recapitulates exacerbated inflammation observed in HTR2A deficient mice. A) Schematic showing strategy for developing conditional knockout mice with HTR2A knocked out in Langerhans cells. B) Line chart showing changes in ear swelling of mice. (n=3 for Vas, n=6 for IMQ per group) C) Pictures showing ears of mice of various groups on Day 7. D) H&E- stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. F) qPCR of pro-inflammatory cytokines of respective groups. G) Schematic showing treatment schedule on CD207- DTR mice for the depletion of different subsets of Langerhans cells. H) Line chart showing changes in ear swelling of mice. (n=7 to 8 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E- stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. L) qPCR of pro- inflammatory cytokines of respective groups. Data are representative of two independent experiments (B- F and H- L). p values determined by two- way ANOVA (B and H) followed by Tukey's post- hoc test, one- way ANOVA (D- F) followed by Tukey's post- hoc test and Student's T- test (J- L). Mean ± SEM (B- F and H- L). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[148, 115, 857, 568]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[150, 90, 217, 105]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[142, 572, 848, 850]]<|/det|> +Figure 4. HTR2A deficiency in Langerhans cells of monocyte lineage results in exacerbated inflammation. A) Schematic showing the experimental design for developing moLCs and adoptively transferring them into Langerhans cells depleted mice. B) Line chart showing changes in ear swelling of mice. (n=12 per group) C) Pictures showing ears of mice of various groups on Day 7. D) H&E- stained slides with quantification of epidermal thickness. E) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. F) qPCR of pro- inflammatory cytokines of respective groups. G) Schematic showing strategy for developing conditional knockout mice with HTR2A knocked out in moLCs. H) Line chart showing changes in ear swelling of mice. (n=6 to 10 per group) I) Pictures showing ears of mice of various groups on Day 7. J) H&E- stained slides with quantification of epidermal thickness. K) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. L) qPCR of pro- inflammatory cytokines of respective groups. Data are representative of two independent experiments (B- F and H- L). p values determined by two- way ANOVA (B and H) followed by Tukey's post- hoc test and one- way ANOVA (D- F and J- L) followed by Tukey's post- hoc test. Mean ± SEM (B- F and H- L). *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001; ns, non- significant. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[145, 103, 857, 710]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[144, 707, 844, 893]]<|/det|> +
Figure 5. IL-23 plays a critical role in exacerbating psoriasisin inflammation by HTR2A deficient monocyte-derived Langerhans cells. A) Heatmap showing mRNA levels of key cytokines in wild type or HTR2A deficient moLCs. (3 biological replicates) The cells were stimulated for 24 hours with Imiquimod (10μg/mL). B) IL-23 ELISA of stimulated and unstimulated 1X105 wild type or HTR2A deficient moLCs. C) Representative histogram of IL-23a MFI among different cell types in cells isolated from ears of HTR2A deficient mice treated with Imiquimod for 7 days. (n=4) The cells were isolated and treated with PMA, Ionomycin, and GolgiStop before antibody staining and flow cytometry analysis. There is quantification of IL-23a MFI and percentage of IL-23a positive cells. D) Schematic showing the experimental design for treating mice with DOI and determining IL-23a expression. E) Line chart showing changes in ear
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 847, 365]]<|/det|> +swelling of mice. (n=8 per group) F) Picture of ears on Day 7 of the respective groups. G) H&E- stained slides of the respective groups with quantification of epidermal thickness. H) Flow cytometry showing the percentage of IL- 23+ cells among CD45+ cells in the respective groups with quantification. I) Flow cytometry showing the percentage of IL- 23+ cells among moLCs in the respective groups with quantification. J) Histogram of moLCs' IL- 23a MFI in the respective groups with quantification. K) Schematic showing the experimental design of IL- 23r blocking experiments. L) Line chart showing changes in ear swelling of mice in the respective groups. (n=6) M) Picture of ears on Day 7 of the respective groups. N) H&E- stained slides of the respective groups with quantification of epidermal thickness. O) Flow cytometry of IL17+ IL22+ Vγ4 T cells and neutrophils in the respective groups with quantifications. P) mRNA levels of proinflammatory cytokines of the respective groups. Data are representative of two independent experiments (B, E-J, and L-P). p values determined by two- way ANOVA (E and L) followed by Tukey's post- hoc test, one- way ANOVA (B and N- P) followed by Tukey's post- hoc test, one- way ANOVA (C) followed by Dunnett post- hoc test, and Student's T- test (G- J). Mean ± SEM (B- C, E, G- J, L, and N- P). \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[146, 117, 806, 480]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[152, 90, 219, 106]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[142, 490, 856, 805]]<|/det|> +Figure 6. HTR2A suppresses IL- 23 secretion by inhibiting activation of non- canonical NFkB pathway in monocyte- derived Langerhans cells. A) Heatmap showing 140 differentially expressed genes (DEGs) between wild type moLCs (GSE222197) treated with IMQ and KO moLCs treated with IMQ (GSE274941) with genes affected by NFkB shown on the right. The experiments generating the datasets of GSE222197 and GSE274941 were done in parallel. MoLCs were pooled from the ears of five mice treated with IMQ for two days, constituting a single biological sample. B) Bar chart showing transcription factors enriched among the DEGs from A) as analyzed using GAGE on the TRRUST Transcription Factor database. C) Western blot of cultured WT and KO moLCs lysate treated with or without DOI and stained with anti- p100/p52, anti- RelB, and anti- \(\beta\) - actin. D) IL- 23 ELISA of 2X105 WT or KO moLCs (5 biological replicates) stimulated for 24 hours with Imiquimod (10μg/mL) and treated with either DOI (10μg/mL) or EVP 4593 (0.1mM). E) Differentiation of Vγ4 T cells into IL- 17+ IL- 22+ cells after coculture with WT or KO moLCs (5 biological replicates) for 3 days. WT or KO moLCs were stimulated for 24 hours with IMQ (10μg/mL) and WT moLCs were treated with either DOI (10μg/mL) or EVP 4593 (0.1mM). Data are a summary of two independent experiments (C- E). p values were determined by one- way ANOVA followed by Dunnett's post- hoc test (D and E). Mean ± or SEM (D and E) *, p<0.05; ***, p<0.001; ****, p<0.0001; ns, non- significant. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[142, 106, 842, 481]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[152, 90, 218, 105]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[142, 490, 852, 785]]<|/det|> +Figure 7. Serotonin is the putative agonist of HTR2A attenuating psoriatic inflammation. A) Immunohistochemistry staining of serotonin on the skin of healthy volunteers, non- lesioned skin of psoriatic patients, and lesioned skin of psoriatic patients. B) Quantification of serotonin intensity in terms of H- score using DensitQuant (3DHISTECH LTD) between the skin of healthy volunteers \((n = 13)\) , lesioned skin of psoriatic patients with low PASI score \((< 30)\) \((n = 15)\) and high PASI score \((>30)\) \((n = 12)\) . C) Quantification of serotonin intensity in terms of H- score between lesioned and non- lesioned skin of psoriatic patients \((n = 27)\) . D) Schematic showing strategy for treating wild type or HTR2A deficient mice with serotonin with their respective control. E) Line chart showing changes in ear thickness of mice of the respective groups \((n = 7)\) . F) Pictures showing the ears of mice in different groups. G) H&E- stained slides of respective groups with quantification of epidermal thickness. H) Flow cytometry of IL17+ IL22+ \(\mathrm{V} / \mathrm{4T}\) cells and neutrophils of respective groups with quantifications. I) mRNA levels of pro- inflammatory cytokines of the respective groups. Data are a summary of two independent experiments (D- I and L- R). p values were determined by two- way ANOVA (E), one- way ANOVA followed by Tukey's post- hoc test (B and G- I) or paired Student's T- test (C). Mean \(\pm\) or SEM (B, C, E, G- I) \*, p<0.05; \*\*, p<0.01; \*\*\*, p<0.001; \*\*\*\*, p<0.0001; ns, non- significant. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[147, 113, 844, 501]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[155, 92, 222, 108]]<|/det|> +
Figure 8
+ +<|ref|>text<|/ref|><|det|>[[143, 512, 849, 714]]<|/det|> +Figure 8. Monocyte- derived Langerhans cells of human psoriatic skin show higher IL23A and lower HTR2A expression. A) UMAP of reanalyzed single cell RNA- sequencing data from GSE151177 and GSE162183 showing 27 different clusters. B) Violin map showing lineage defining transcription factors, receptors, and markers of embryo Langerhans cells (eLC), monocyte- derived Langerhans cells (molC), and dendritic cell type 3 (DC3). C) Box plot comparing HTR2A expression levels of control and psoriatic monocyte- derived Langerhans cells. D) Dot plot showing expression of various cytokines by eLC, molC, and DC3. E) Dot plot showing expression of various NFκB pathway- related genes by psoriatic molCs and control molCs. F) Immunofluorescence staining of HTR2A on Langerhans cells (marked by CD1a) in lesioned and non- lesioned psoriatic skin with quantification (n= 27). p values were determined by paired Student's T- test (C, F). \*, p<0.05; \*\*, p<0.01. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 43, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 430, 203]]<|/det|> +- GEODataset.pdf- 20241209Supplementalinformation.pdf- NCOMMS2482399rs.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/images_list.json b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..27bd2fd70fc335ab9cd6144c29daa0e116a2bdc7 --- /dev/null +++ b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/images_list.json @@ -0,0 +1,10 @@ +[ + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [], + "page_idx": 33 + } +] \ No newline at end of file diff --git a/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5.mmd b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5.mmd new file mode 100644 index 0000000000000000000000000000000000000000..fc502ced7e49cd209041c422d619fff1ec3a179b --- /dev/null +++ b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5.mmd @@ -0,0 +1,289 @@ + +# Cancer coopts differentiation of B-cell precursors into macrophages + +Chen Chen National Institute on Aging + +Bongsoo Park National Institute on Aging + +Emeline Ragonnaud National Institute on Aging + +Monica Bodogai National Institute on Aging + +Le Zong NIA/NIH + +Jung- Min Lee Center for Cancer Research + +Isabel Beerman National Institute on Aging + +Arya Biragyn ( biragyna@mail.nih.gov ) National Institute on Aging https://orcid.org/0000- 0001- 5276- 6102 + +## Article + +Keywords: B- cell precursors, macrophage, transdifferentiation, immune suppression, RNA profiling + +Posted Date: September 13th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 858433/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on September 14th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33117- y. + +<--- Page Split ---> + +## Cancer coopts differentiation of B-cell precursors into macrophages + +Chen Chen \(^{1}\) , Bongsoo Park \(^{2}\) , Emeline Ragonnaud \(^{1}\) , Monica Bodogai \(^{1}\) , Le Zong \(^{2}\) , Jung- Min Lee \(^{3}\) , Isabel Beerman \(^{2*}\) and Arya Biragyn \(^{1,4,5*}\) + +From: \(^{1}\) Immunoregulation Section, Laboratory of Immunology and Molecular Biology; \(^{2}\) Epigenetics and Stem Cell Aging Unit, Translational Gerontology Branch, and National Institute on Aging, Baltimore, MD; \(^{3}\) Women's Malignancies Branch, Center for Cancer Research, National Cancer Institute, Bethesda, MD; \(^{4}\) Lead contact; \(^{5}\) Senior author + +\* Corresponding authors: Arya Biragyn, Ph.D., E- mail: biragyna@mail.nih.gov and Isabel Beerman, Ph.D., E- mail: isabel.beerman@nih.gov + +Running title: B cells transdifferentiation in cancer + +Keywords: B- cell precursors, macrophage, transdifferentiation, immune suppression, RNA profiling + +Conflict- of- interest disclosure: The authors declare no conflicting financial interests + +<--- Page Split ---> + +## ABSTRACT + +We recently reported that some cancers induce accumulation of bone marrow (BM) B- cell precursors in the spleen to convert them into metastasis- promoting, immunosuppressive B cells. Here, using various murine tumor models and samples from humans with breast and ovarian cancers, we provide evidence that cancer cells also coopt differentiation of the extra nodal B- cell precursors to generate macrophages (termed B- MF). We link the transformation to a small subset of CSF1R+ Pax5Low cells within BM pre- B and immature B cells and cancer- secreted M- CSF that downregulates Pax5 via CSF1R signaling. Thus, cancer generates tumor- associated macrophages (TAM) from B- cell precursors in addition to their primary source, monocytes. Based on their differences from monocyte- derived TAM, such as a superior ability to induce FoxP3+ Tregs, suppress proliferation of T cells and more efficiently phagocytize apoptotic cells, we propose that cancer generates B- MF to mediate cancer escape. + +<--- Page Split ---> + +The role of B cells in cancer remains poorly understood despite reports of both positive and negative associations of B cells with the disease outcome. Even in the same murine model, T2- MZ B cells and aging- associated B cells can either increase or mitigate B16- F10 melanoma growth in C57BL/6 mice, respectively (1, 2). B cells can also inhibit or enhance metastasis in BALB/c mice with orthotopic 4T1 breast cancer cells (3, 4). At least some cancer- promoting functions of B cells are attributed to their immunosuppressive, regulatory subsets termed Bregs, whose generation and function appear to be actively controlled by cancer itself. To enforce this regulation, cancer cells secrete various factors that affect survival and differentiation of B cells, including B lymphocyte stimulator (BLyS/BAFF), thymic stromal lymphopoietin (TSLP), colony stimulating factors (CSF such as M- CSF, GM- CSF and G- CSF) and lipid mediators such as 5- lipoxygenase (5- LO) metabolites (5- 10). In BALB/c mice with orthotopic 4T1 breast cancer, a model for human triple negative breast cancer (11), our group linked 5- LO metabolite- induced activation of PPAR \(\alpha\) to the conversion of B cells into TGF \(\beta^{+}\) CD25+ Bregs (tBregs) (7). These converted cells then promote metastasis acting as potent suppressors of antitumor T cell activity and inducers of FoxP3+ Tregs and MDSCs via the TGF \(\beta\) /TGF \(\beta\) RII axis (5, 12). Breast and ovarian cancer cells secret TSLP both to prepare the metastasis “soil” in the lungs and to accumulate BM B- cell precursors in the spleen for conversion to tBregs. In the lungs, TSLP enables infiltration of CCR4+ cancer cells and their protector CCR4+FoxP3+ Tregs by upregulating expression of CCL17 (6, 13), while in the BM, TSLP downregulates expression of CXCR4 and \(\alpha 4\beta 1\) in B- cell precursors, causing their premature emigration and accumulation in the spleen to be converted into TGF \(\beta^{+}\) CD25+ tBregs (10). M- CSF and GM- CSF from cancer cells promote differentiation + +<--- Page Split ---> + +and survival of cancer- promoting myeloid suppressive cells and tumor- associated macrophages (TAM) from BM monocytes (8, 9). Although bifurcation of myeloid and lymphoid lineage from multipotent progenitors occurs before specialization of B- cell progenitors in BM, surprisingly some B- cell precursors appear to retain macrophage differentiation potential. For example, the differentiation of BM B- cell precursors can be diverted towards macrophages upon forced expression or deletion of a single transcription factor (14, 15). Anecdotal reports indicate that this transdifferentiation can also occur in BM of naive mice, which is linked to a small subset of biphenotypic (CD19+B220+ CD16/32++CD11b+) B- cell precursor cells (16). However, the biological consequence of this rare event and whether cancers affect this process remain unknown. + +Here, we report that cancers frequently induce the transdifferentiation of the bona fide BM Csf1R+Pax5Low pre- B and immature B (iB) cells into TAM (termed B- MF). To do this, cancers first drive accumulation of B- cell precursors/iB cells in the spleen and tumors. Unlike “classical” monocyte- derived TAM (9), cancer appears to use B- MF to mediate cancer escape via suppression of antitumor T cells. This does not appear to be a mouse specific phenomenon, as B- MF can be detected in patients with breast and ovarian cancers, and we present evidence of B- MF signatures in published scRNA sequence data of human cancers. + +## RESULTS + +B- cell markers are expressed in TAM. We previously reported that some cancers mobilize BM B- cell precursors in the spleen and tumors (10) to subsequently convert them into \(\mathrm{TGF}\beta^{+}\) tBregs (5, 7, 12). Microarray transcription profiling of B- cell precursors and + +<--- Page Split ---> + +tBregs from mice with BALB/c mice with orthotopic 4T1.2 breast cancer surprisingly revealed significant upregulation of macrophage-associated genes, such as CD68, Csf1r (encodes CSF1R), Cebpb (CCAAT Enhancer binding protein beta), Cebpg (CCAAT Enhancer binding protein gamma), Ccl2 (CCL2), and Csf1 (M-CSF) (Fig. S1A). Given that CEBPB and CSF1R play essential roles in defining macrophage fate (14, 17) and anecdotal reports on retention of macrophage differentiation potential in B-cell precursors (16, 18), we wondered whether the upregulation of macrophage genes indicates that cancer hijacks B-cell differentiation, pushing them to macrophages. To test this possibility, we FACS evaluated tumor-infiltrating B cells (TIB, gated on CD19+) and TAM (gated on F4/80+) in WT BALB/c mice and \(\mu \mathrm{MT}\) BALB/c mice (B-cell differentiation halted at the pro-B-cell stage (19)) bearing orthotopic 4T1.2 breast cancer. While both TIB and TAM in WT mice contained a small but significant proportion of cells co-expressing B-cell and macrophage markers, the double lineage expressing cells were not present in \(\mu \mathrm{MT}\) mice (Fig. 1A, gating strategy is in Fig. S1B). We also evaluated if these cells were present in other tumor models. TIB and TAM co-expressing B-cell and macrophage markers were found in WT BALB/c mice with orthotopic EMT6 breast cancer, and WT C57BL/6 mice with other orthotopic cancers (AT3 breast and MC38 colon cancer) or spontaneous ovarian cancer (Mogp), but were not detected in \(\mu \mathrm{MT}\) or \(\mathrm{J}_{\mathrm{H}}\mathrm{T}\) mice with these cancers (Fig. 1B and Fig. S1C, D). Of note, in \(\mathrm{J}_{\mathrm{H}}\mathrm{T}\) mice, B-cell differentiation is blocked at the pro-B-cell stage (20). A comprehensive FACS evaluation of TAM revealed they do indeed express additional B-cell-specific markers, including Mb1/CD79a, CD20, IgM, and IgD (Fig. S1D-G). To further confirm our finding, we used a B-cell lineage tracer Mb1-EYFP mouse model (21), which expresses enhanced yellow fluorescent protein (EYFP) as a readout for the Ig- \(\alpha\) + +<--- Page Split ---> + +receptor exclusively expressed in B cells (except plasma cells). Upon challenge with MC38 tumor cells, \(\mathrm{EYFP^{+}F4 / 80^{+}CD11b^{inter / low}}\) TAM were significantly increased in the tumor (Fig. S2). In sum, cancers cause a marked increase of macrophage- like cells derived from B cells (hereafter referred to as B- MF). + +Cancer induces B- cell- to- macrophage differentiation. To explain the accumulation of B- MF, we tested whether cancer cells trans- differentiate them from B cells. Sort- purified Lin \(\mathrm{CD19^{+}B220^{+}B}\) cells ( \(>98\%\) purity) from naïve mouse bone marrow (BM), spleen, lymph nodes (LN) and peritoneal cavity (PeC) were cultured for 7 days in conditioned medium (CM) of 4T1.2 cancer cells (4T1.2- CM). Of note, unless specified, from here on this incubation time was chosen because of a noticeable upregulation of macrophage markers (Fig. S3A). 4T1.2- CM, but not control medium, significantly upregulated F4/80 and CD11b in B cells from BM, but not from spleen, LN or PeC (Fig. 1C). The BM B cells acquired features associated with macrophages, such as adherence to plastic, enlarged vacuolar cytoplasm, and ability to phagocytize fluorochrome- labelled \(E\) . coli (Fig. 1D and Fig. S3B). Independent experiments with sort- purified B- cell subsets revealed that only BM B- cell precursors and iB cells, but not splenic transitional, follicular (FOB), or marginal zone (MZB) B cells, can become B- MF (Fig. S3C; gating strategy in Fig. S3D). Consistent with the accumulation of B- MF in mice with various tumors, CM from almost every type of cancer cell (except B16- F10 melanoma cells) induced the generation of B- MF from BM B- cell precursors/iB cells (BMBP) and pre- B- cell line 70z/3 (Fig. 1E- G and Fig. S3E). In single cell Image stream FACS analysis, macrophage and B- cell markers were indeed co- expressed on single B- MF cells derived from BM Mb1- EYFP+ B cells (Fig. 1H). To rule out artifacts of in vitro experiments, we also performed a 7- day + +<--- Page Split ---> + +in vivo conversion assay by i.p. implanting sort-purified BMBP from naïve Mb1- EYFP mice into PeC of C57BL/6 mice with or without ID8 ovarian cancer in PeC. Tumor- bearing mice contained significantly higher frequency of EYFP+ B- MF in PeC than naïve mice (Fig. 1I and Fig. S3F). These cells also markedly upregulated TGFβ/LAP and PD- L1 (Fig. 1I), the two factors utilized in immunoregulation (22, 23). Taken together, we concluded that cancer uses BMBP to generate TAM in addition to their hitherto known source, monocytes (9). + +B- MF and Mo- MF differ functionally. To understand why cancer generates TAM from B cells, we ex vivo generated B- MF and monocyte- derived macrophages (Mo- MF) from sort- purified BMBP and monocytes, respectively, by stimulating with 4T1.2- CM, as described above. While both populations adhered to plastic and comparably upregulated F4/80 and CD11b, only B- MF expressed the B- cell- specific markers (CD79a and IgM, Fig. 2A, B and Fig. S4A). mRNA microarray analysis of B- MF and Mo- MF revealed that regardless of origin, the two populations share expression of numerous macrophage- related genes and do not robustly expresses B- cell signature profiles (Fig. S4B- D and Table S1). Principal component analysis (PCA) and transcription profiling clearly separate B- MF from Mo- MF (Fig. 2C and Table S2). B- MF expressed higher levels of genes involved in fatty acid metabolism, cell cycle, steroid- cholesterol biosynthesis and downregulated expression of pro- inflammatory and IFNγ response genes (Fig. 2D, E). While Mo- MF had more robust M1- like transcription, M2- skewing was more pronounced in B- MF (Fig. 2E and Fig. S4E). The unique transcription profiles were also confirmed in single cell RNA sequencing (scRNA- seq) of in vitro generated B- MF (10,563) and Mo- MF (10,235) cells, with UMAP clustering identifying mostly separate cell- clusters of macrophages derived + +<--- Page Split ---> + +from either B cells or monocytes (Fig. 2F and Table S3). We distinguished 12 cell clusters with the Leiden algorithm, using shared nearest neighbor (SNN) in PCA space and identified the key genes establishing the six clusters accounting for the majority of single cells (Fig. 2G, H and Fig. S4F). B- MF appeared to be more phagocytic than Mo- MF, as they markedly upregulated Mrc1 (encodes CD206, Fig. 2I). Given the unique transcriptional signatures of the B- MF, we next examined scRNA profiles from four independent samples of TAM in mice with 4T1.2 cancer (Fig. 2J). Using signature genes identified from in vitro generated macrophages (Fig. 2H) we noted three clusters (0, 6, and 8) with robust expression of genes identified in B- MF (Fig. S4H). Cluster 8 also strongly overlapped with a mixed macrophage population (Cluster 4, Fig. 2G and Fig. S4F- H), suggesting that only clusters 0 and 6 were most similar to B- MF. We also examined expression of three genes robustly expressed in B- MF and they were also expressed in TAM clusters 0 and 6 (Fig. 2K). In contrast, expression of key Mo- MF genes was mostly found in clusters 1, 3, 4, and 5 (Fig. S4G), suggesting B- MF and Mo- MF retain traceable and different transcription profiles in vivo. + +To confirm B- MF and Mo- MF differences at the functional level, first we examined their proliferation. Only B- MF expressed higher levels of Ki67 and readily incorporated BrdU (pulsed at day 6 and tested at day 7 of the culture) (Fig. 3A, B and Fig. S5A), indicating their proliferative and self- maintaining state. Given the changes seen in fatty acid metabolism (Fig. 2D), we stained the cells with filipin III, a fluorescent polyene antibiotic that binds to free cholesterol (24). Compared with Mo- MF, the cellular cholesterol content was significantly upregulated in both in vitro and in vivo generated B- MF (Fig. 3C, D and Fig. S1E- G). To evaluate phagocytosis, B- MF and Mo- MF were + +<--- Page Split ---> + +cultured with fluorochrome- labeled apoptotic cancer cells. B- MF phagocytized apoptotic cells significantly better than Mo- MF (Fig. 3E, F). Given B- MF markedly upregulated TGFβ/LAP and PD- L1 (Fig. 1I), we tested their function in a 4- day T- cell suppression assay (5). When B- MF or Mo- MF were cultured with naïve mouse splenic T cells stimulated with anti- CD3/CD28 antibodies, only B- MF significantly inhibited proliferation of \(\mathrm{CD4^{+}}\) T cells and, at lesser extent, \(\mathrm{CD8^{+}}\) T cells (Fig. 3G and Fig. S5B- E). Because TGFβ is involved in induction of FoxP3+ Tregs (5), we also performed an in vitro Treg conversion assay by culturing the two macrophages with splenic \(\mathrm{CD25^{+}CD4^{+}}\) T cells in the presence of anti- CD3/CD28 antibodies and IL2 for 4 days. Compared with Mo- MF, B- MF induced higher levels of FoxP3+ Tregs (Fig. S5F), further suggesting that cancer presumably uses B- MF to mediate immunoregulation and thereby tumor progression. We therefore tested whether B- MF can reverse the retarded growth of B16- F10 melanoma in \(\mu \mathrm{MT}\) mice (n=5/group, Fig. 3G), which we previously linked to the lack of B cells (12). Intravenous transfer of B- MF significantly decreased tumor- infiltrated \(\mathrm{CD4^{+}}\) T cells (and at lesser extent \(\mathrm{CD8^{+}}\) T cells) and their IFN- expressing subsets (Fig. 3H, I and Fig. S5H) and markedly increased growth of B16- F10 melanoma in \(\mu \mathrm{MT}\) mice (p<0.05, Fig. 3J and Fig. S5G). In contrast, T cells in the spleen and dLN were not affected by the B- MF transfer (not depicted), suggesting that B- MF primarily regulate tumor- infiltrating T cells. + +Cancer mobilizes BMBP in the spleen to convert them to B- MF. Recently, we reported that breast cancer induces tBregs from BM pre- B cells. To do this, cancer first stimulates accumulation of BMBP in the spleen and tumor by causing their premature emigration from BM (10). We therefore reasoned that cancer targets the same pool of BMBP accumulated in the spleen to generate B- MF. In support of this hypothesis, the total + +<--- Page Split ---> + +number of CD93+ BMBP were markedly increased in the circulation (but decreased in BM) of mice with 4T1.2 and Mogg cancers (Fig. 4A and Fig. S6A, B) as compared with naïve mice (Fig. S6A, B). Stimulation with 4T1.2- CM readily generated B- MF from CD93+ BMBP sort- purified from spleens of mice with 4T1.2 cancer (Fig. S6C) but failed to do so if B cells were isolated from spleens of naïve mice, regardless of CD93 expression (Fig. S6C and Fig. S3C), implying that cancer first drives accumulation of BMBP in the spleen to convert them into B- MF. + +Cancer targets CSF1R+CD93+ BMBP by secreting M- CSF. To understand the mechanism of the B- MF generation, we analyzed CM of cancer cells for secreted factors that could affect differentiation of macrophages. M- CSF, a regulator of macrophage differentiation and survival (8, 17), was among the factors that was highly increased in the cancer cells that induce B- MF (Fig. 4B). Conversely, M- CSF was almost absent in CM from B16- F10 cells (Fig. 4B and Fig. S6D, E), which did not induce the generation of B- MF (Fig. 1E and Fig. S3E). Compared to naïve mice, serum M- CSF was also significantly upregulated in mice with 4T1.2 cancer (Fig. S6F). Importantly, BM CD93+CD19+ BMBP of naïve mice expressed its cognate receptor CSF1R (about 15% iB cells, 6% of pre- B cells and 1.5% pro- B cells, Fig. S6G). In mice with 4T1 cancer, CSF1R+CD93+CD19+ BMBP were markedly reduced in BM but increased in the spleen (Fig. 4A and Fig. S6G), consistent with their cancer- induced emigration from BM, as discussed above. To link these CSF1R+CD93+CD19+ BMBP to the generation of B- MF, we performed the in vitro macrophage conversion assay by stimulating sort- purified CSF1R+ and CSF1R- B cells from BM and spleen of naïve mice with 4T1.2- CM. While CSF1R+ BM B cells readily generated B- MF, the BM CSF1R- subset failed to do so (Fig. 4C). Consistent with an + +<--- Page Split ---> + +inability of splenic CD93+ B cells of naïve mice to generate B- MF (Fig. S6C and Fig. S3C), we failed to convert naïve mouse splenic B cells into B- MF regardless of CSF1R expression (Fig. 4C). We also cultured primary BM BMBP or 70z/3 cells with 4T1.2- CM in the presence or absence of neutralizing M- CSF antibody (Ab) or Ki20227, a specific inhibitor of c- Fms/CSF1R (25). Both cells failed to generate B- MF upon M- CSF neutralization or CSF1R signaling inhibition (Fig. 4D, E and Fig. S6H, I). To rule out artifacts of in vitro assay, we created mice with conditional CSF1R deficiency in B cells (Mb1- CSF1RFlox/Flox mice, gating strategy in Fig. S7). Unlike WT littermates or monocytes from Mb1- CSF1RFlox/Flox mice, the loss of CSF1R in BMBP significantly impaired the cancer CM- induced B- MF differentiation (Fig. 4F, G). Of note, the residual macrophage differentiation seen in Fig. 4F is presumably due to CSF1R expression preceding Mb1 expression and thus only pro- B cells and onward will have Csf1r deletion. + +Given that PAX5 is the key pro- B cell factor that represses Csf1r and other myeloid lineage- specific genes (15, 26), we reasoned that cancer decreases levels of this transcription factor using M- CSF. FACS staining confirmed that Pax5 was markedly decreased in BM CD93+ BMBP, particularly in CSF1R+ but not CSF1R- subsets, from mice with 4T1.2 or Mogp cancer (Fig. 5A and Fig. S8A, C). Importantly, Pax5 was also significantly decreased in BM CSF1R+ BMBP from naïve mice and 70Z/3 cells upon treatment with 4T1- CM or M- CSF (Fig. 5B, C and Fig. S8B). As Pax5 deficiency alone is sufficient to render pro- B cells susceptible to myeloid differentiation (15), we concluded that cancer uses M- CSF to reduce expression of Pax5 in CSF1R+ CD93+ BMBP and thereby promote macrophage differentiation. + +<--- Page Split ---> + +To further understand the B- cell susceptibility towards macrophage conversion, we analyzed chromatin accessibility by performing ATAC- seq on CSF1R+ and CSF1R- BMBP isolated from both BM and spleen. PCA clustering showed the most robust differences in chromatin profiles were driven by the location of the BMBP (BM vs spleen) regardless of CSF1R expression, driving the PC2 axis (blue & purple vs orange & green, Fig. 5D). The chromatin landscapes of the CSF1R+ and CSF1R- BMBP isolated from BM (orange and green, Fig. 5D) also significantly differed from each other, driving the PC3 axis. We then examined the differentially accessible regions (DARs) between CSF1R+ and CSF1R- cells isolated from spleen or BM. Whereas comparisons between CSF1R+ and CSF1R- cells from spleen did not show any differences reaching our threshold for significance, confirming their close clustering on the PC3 axis; the BM CSF1R+ cells contained significantly more open chromatin than the BM CSF1R- cells (Fig. 5D, E). These data suggest the BM CSF1R+ BMBP may have a more permissive chromatin environment, susceptible to macrophage differentiation signals. As the spleen- derived B cells and BM CSF1R- cells were refractory to macrophage conversion (Fig. 4C), we looked at DARs with less accessibility in CSF1R- compared to CSF1R+ BM cells (749 loci) in spleen cells to determine if these regions remain closed and potentially “lock in” the lymphoid lineage potential. Indeed, the overwhelming majority of regions with decreased accessibility in the BM CSF1R- cells remained closed in the cells from spleen (679 of 749). Evaluation of these consensus open regions found in BM CSF1R+ cells for potential transcription factor binding sites permitting macrophage differentiation showed significantly increased accessibility of ERG and RUNX1 sites (Fig. 5F). ERG is known to be expressed both in myeloid and lymphoid progenitor cells (27) and has particular importance in early + +<--- Page Split ---> + +hematopoietic progenitor cells as it binds to coregulators such as RUNX and GATA (28). RUNX1 regulates growth and survival of macrophages via binding to promoter and enhancer regions of Csf1r and upregulating its expression (29). Runx1 is also robustly expressed in early progenitor and myeloid- committed progenitor cells (30). Thus, the increased accessibility to binding sites of both ERG and RUNX1 suggests a potentially more primitive, permissive chromatin state allowing for myeloid lineage transformation of the BM CSF1R+ cells. + +B- MF- generating CSF1R+ BMBP accumulate in humans with cancer. We recently reported that peripheral mobilization of BMBP also occurs in humans with breast cancer (BC) (10), suggesting the generation of B- MF. To test this possibility, we FACS evaluated PB of healthy donors (HD, n=7) and patients with BC (n=8). Compared with HD, PB of BC was markedly increased in CSF1R+ BMBP (Fig. S8D), as we described in mice with cancer. Moreover, microarray transcription profiling of sort- purified B cells from PB of BC patients revealed that they significantly upregulated macrophage- associated genes, such as Cebpa, Marco, and Csf1r, as compared with B cells from HD (Fig. 5G). We also FACS evaluated B- cells from PB of patients with ovarian cancer (OC, n=5). Compared with HD, OC patients significantly increased CSF1R+ BMBP (Fig. 5H and Fig. S8E) with upregulated expression of CD68 and LDLR (Fig. 5H), similar to mice with cancer. Using recently published scRNA- seq data of tumor- infiltrated immune cells from patients with breast cancer (31), we also found a macrophage cluster with overlapping signatures of B- MF- like cells (cluster 3, Fig. 5I) by examining genes with differential expression defined in murine in vitro- generated B- MF (cluster 0, Fig. 2H). In particular, cluster 3 was enriched for expression of EGR1, IER2, IER3, and SLC40A1, which were major drivers of identity + +<--- Page Split ---> + +for murine in vitro- generated B- MF (Fig. 5I). Similarly, in the single- cell transcriptome data from human high- grade serous OC (32), we also detected the B- MF- like signature in macrophages (Cluster 0, Fig. 5J), although with a lesser overlap than in BC, further suggesting that human cancers can promote the B- cell transdifferentiation into macrophages. + +## DISCUSSION + +BMBP undergo a series of subsequent and tightly regulated differentiation steps after their bifurcation from multipotent cells to committed myeloid/lymphoid lineage cells. Despite this, experiments with forced expression or inhibition of a single transcription factor or mutations that drive leukemogenesis (14, 15) reveal that BMBP retain plasticity and myeloid transdifferentiation potential. Unlike these artificial or rare events, here we report that the B- cell- to- macrophage transdifferentiation is commonly used by murine cancers to generate TAM/B- MF. In PB of humans with metastatic/recurrent triple- negative BC and high- grade serous OC, we also detect a significant increase of CSF1R+ CD68+LDLR+ BMBP which also express the macrophage- specific genes Cebpa, Cebpb, and Marco. Importantly, the B- MF signature is also identifiable within unique macrophage clusters using recently published scRNAseq profiles of tumor- infiltrating cells in patients with BC (31) and high- grade serous OC (32). Although the characterization of human B- MF is a topic of a separate study, our results suggest that human and murine cancers transdifferentiate BMBP into macrophages, adding one more feature to the heterogeneity and plasticity of TAM. The inflammatory and antitumor activities of TAM at the early stages of tumor can shift to proangiogenic and tumor- supporting M2- like phenotypes as + +<--- Page Split ---> + +tumor progresses (33), presumably when B- MF would be induced. Interestingly, B- MF resemble both small (S)- TAM and large (L)- TAM (which associate with a poor disease outcome) recently identified in human colorectal liver metastasis (34). The lipid metabolism and phagocytosis genes of B- MF (Fasn, Pltp, Acat1, C1qa, and C1qb) are upregulated in L- TAM, while LDLR, Hmgcr as well S100a8, Vcan, and Thbs1 are increased in S- TAM. + +Although TAM are primarily derived from monocytes (9), our results clearly show that at least some originate from bona fide B cells. The biological relevance of this redundancy in generation of TAM remains poorly understood; however, based on our comparisons of the side- by- side generated B- MF and Mo- MF, we think that the two macrophages may serve different purposes. Transcriptionally, B- MF preferentially upregulate expression of genes involved in cell cycle, fatty acid metabolism, and steroid- cholesterol biosynthesis, implying they utilize unique metabolic and inflammatory functions. Unlike Mo- MF, B- MF proliferate, i.e. self- maintain and thus may persist longer in the tumor. B- MF markedly upregulate surface expression of LDLR, which removes extracellular cholesterol/LDL (35), and this could explain the higher levels of intracellular cholesterol and lipids in B- MF compared to Mo- MF. Consistent with significant upregulation of genes associated with phagocytosis, M2- skewing and immunosuppressive functions (PD- L2, B7- H3, Marco, TGFβ) and downregulation of pro- inflammatory and IFNγ response genes, B- MF express higher levels of surface MRC1 (CD206), PD- L1 (CD274), and TGFβ/LAP and efficiently phagocytize apoptotic cells compared to Mo- MF. This efficient phagocytosis presumably occurs without overt inflammation, as LDLR- mediated cholesterol influx inhibits activation of the inflammasome (36). Our data show + +<--- Page Split ---> + +that cancer generates phenotypically and functionally non- redundant TAM from BMBP and monocytes, where B- MF appear to promote cancer growth presumably by controlling antitumor T cell responses. First, unlike Mo- MF, B- MF efficiently suppress proliferation of T cells or induce the generation FoxP3+ Tregs in vitro. Second, B- MF enhance growth of B16- F10 melanoma in \(\mu \mathrm{M}\) mice. To do this, they primarily decrease \(\mathrm{CD4^{+}}\) T cells and inhibit \(\mathrm{IFN}\gamma^{+}\) \(\mathrm{CD4^{+}}\) T cells in the tumor, presumably by utilizing the B- MF- expressed immunoregulatory factors, \(\mathrm{TGF}\beta /\mathrm{LAP}\) and PD- L1 (22, 23), and LDLR. For example, LDLR may enhance the \(\mathrm{TGF}\beta\) responsiveness of target T cells by removing extracellular LDL/cholesterol that impairs \(\mathrm{TGF}\beta\) binding and thus signaling via \(\mathrm{TGF}\beta \mathrm{RII} / \mathrm{TGF}\beta \mathrm{R1}\) (37). + +We propose that cancer primarily targets a small subset of \(\mathrm{CSF1R^{+}Pax5^{Low}}\) pre- B cells and iB cells recently emigrated from BM. First, B- MF are not found in tumor- bearing mice with the B- cell differentiation blockage at the pro- B cell stage. Second, splenic transitional B cells from naive mice do not generate B- MF regardless of their CSF1R expression state, as cancer first needs to mobilize BMBP into the circulation as the source of B- MF. We and others have reported that cancers use TSLP and G- CSF to mobilize BM pre- B cells and HSPS in the circulation (10, 38). We also find the chromatin accessibility landscape of BM CSF1R+ BMBP to be significantly more open and permissive to macrophage differentiation signals. In contrast, CSF1R+ and CSF1R- splenic B cells and BM CSF1R- cells present a chromatin landscape that is refractory to macrophage conversion. The overwhelming majority of regions with decreased accessibility in the BM CSF1R- cells remained closed in the spleen (679 of 749), presumably “locking in” the state of lymphoid lineage potential. It appears that BM CSF1R+ cells have a potentially more + +<--- Page Split ---> + +primitive, permissive chromatin state allowing for myeloid lineage transformation. The BM CSF1R+ cells have more accessible ERG and RUNX1 binding sites, the two transcription factors expressed in myeloid and lymphoid progenitor cells (27) and early progenitor and myeloid- committed progenitor cells (30), respectively. Given that RUNX1 also upregulates expression of Csf1r by binding to its promoter and enhancer regions (29) and that the PAX5 deletion alone removes the repression of Csf1r and other myeloid lineage- specific genes and induces the BM B- cell precursor transdifferentiation (15, 26), we think that RUNX1 supports CSF1R expression in CSF1R+ BMBP to downregulate Pax5 in response tonic and cancer- secreted M- CSF. Give that the CSF1R+ BMBP in our patients with metastatic/recurrent triple- negative BC and high- grade serous OC co- express macrophage- associated genes, human cancer may also target biphenotypic B- cell precursors reported to have a macrophage- differentiation potential (16). Overall, for the first- time we report the B- cell- to- macrophage transdifferentiation as a physiological and widely utilized phenomenon. Murine and possibly human cancers target the transdifferentiation to generate immunosuppressive TAM. + +<--- Page Split ---> + +## MATERIAL AND METHODS + +Mice, cell lines: The animal protocol was approved by the ACUC committee of the National Institute on Aging (ASP 322- LMBI- 2022) under the Guide for the Care and Use of Laboratory Animals (NIH Publication No. 86- 23, 1985). The study used young (8- 12 weeks old) female mice bred and housed in the same, specific pathogen- free environment at the National Institute on Aging (NIA). C57BL/6J, BALB/CJ, R26R- EYFP (B6.129X1- Gt (ROSA)26Sortm1(EYFP)Cos/J), Csf1rflox mice (B6.Cg- Csf1rtm1.2Jwp/J) and μMT mice (B6.129- Ighm- tm1Cgn/J) and JHT mice (JHT; B6.129P2- Igh- Jtm1Cgn/J) in C57BL/6 background mice were purchased from the Jackson Laboratory (Bar Harbor, ME); μMT mice in BALB/c background were a gift from Dr. Thomas Blankenstein (Max- Delbrück- Center for Molecular Medicine, Berlin, Germany) (39). Mb1- Cre mice in C57BL/6 background (B6.C(Cg)- Cd79atm1(cre) Reth/EhobJ) were a gift from Dr. Richard Maraia (National Institute of Child Health and Human Development, Bethesda, MD) (40). Mogg- tag (Mogp) mice in C57BL/6 background were a gift from professor Dr. I. Miyoshi (Tohoku University Graduate School of Medicine, Miyagi, Japan) (12). To create mice with B- cell specific EYFP reporter (Mb1- EYFP) or CSF1R deletion (Mb1- CSF1RFlox/Flox), Mb1- cre mice were bred with R26R- EYFP and Csf1rflox mice, respectively. + +4T1.2 cells- were a gift from Dr. Robin L. Anderson (Peter McCallum Cancer Center, Melbourne, Australia); MC38 colonic adenocarcinoma cells were a gift from Dr. Jeffrey Schlom (National Cancer Institute, Bethesda, MD) (41); mammary carcinoma AT3 cells were a gift from professor Scott I. Abrams (Roswell Park Comprehensive Cancer Center, Buffalo, NY); ID8- p53-/-- RFP (ID8) cells were gift from professor Sharon Stack (University of Notre Dame, IN); and EMT6 cells and melanoma B16- F10 cells were + +<--- Page Split ---> + +purchased from American Type Culture Collection (Manassas, VA). Cells were tested free of mycoplasma with Mycoplasma Detection Kits (Lonza Basel, Switzerland; and IDEXX BioAnalytics, Columbia, MO). + +Tissues and blood processing: Human samples were collected with written informed consent at the Clinical Core Laboratory, NIA, under Human Subject Protocol # 2003054 and Tissue Procurement Protocol # 2003- 071. Cancer patient PBMC were collected before the clinical trial (protocol # NCT02203513, NCI) from people with metastatic/recurrent triple- negative breast cancer and high- grade serous ovarian cancer (42, 43). The trial was approved by the Institutional Review Board of the Center for Cancer Research, National Cancer Institute, Bethesda, MD, USA. All experiments were performed on PBMC which were cryopreserved after collection. Mouse BM cells were flushed out of femurs and tibias with cold cRPMI. Single cell suspension of BM, spleen, LN were prepared with \(70\mu \mathrm{m}\) strainer (Falcon, Bedford, MA). BM, spleen, and blood cells were treated with ACK buffer to remove red blood cells. Mouse tumor tissues were cut to 3- 5 mm pieces and digested with mouse tumor dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany) following the manufacturer's instruction. + +Flow cytometry (FACS): For immune cell phenotyping, cells were pre- incubated with TruStain FcX™ solution before immunostaining with different combinations of anti- mouse or anti- human Abs (1μg per \(10^{6}\) cells, Suppl. Table 4) and fixable viability dye, then fixed/permeabilized with eBioscience™ intracellular fixation & permeabilization buffer (Thermo Fisher, Waltham, MA). The samples were evaluated on FACSymphony™ (BD, Franklin Lakes, NJ), Amnis ImageStreamX MKII (Millipore, Burlington, MA) or + +<--- Page Split ---> + +CytoFLEX (Beckman Coulter, Brea, CA). The results were analyzed with FlowJo v10(BD), IDEAS (Millipore) or Cytoexpert 2.3 (Beckman). + +Cancer CM media preparation and cytokine quantification: Cells were cultured in RPMI1640 or DMEM (for ID8 cells) supplemented with \(10\%\) FBS, \(1 \times\) HEPES, sodium pyruvate, nonessential amino acids solution, penicillin- streptomycin- glutamine (Gibco, Gaithersburg, MD), and \(55 \mathrm{mmol / L}\) \(\beta\) - mercaptethanol in T75 flask to \(70 - 80\%\) confluency. CM was collected after 5 min centrifugation at \(1500 \mathrm{rpm}\) , filtered with \(0.2 \mu \mathrm{m}\) filter, and stored at \(- 80^{\circ} \mathrm{C}\) as single use aliquots. For cytokine tests, confluent cells were cultured with RPMI without FBS for \(24 \mathrm{~h}\) . Mouse serum was collected using BD Microtainer® Tubes following the manufacturer's instruction. Cytokines and M- CSF in filtered CM or sera were evaluated with Quantikine ELISA kit (R&D, Minneapolis, MN) or with Proteome Profiler Mouse XL Cytokine Array (R&D). Images were captured and analyzed with Fiji software. + +In vitro B- MF conversion: BM Lin+ (TER119, CD11b, Gr- 1, CD3ε, NK1.1 or CD49b, Ly6C, Ly6G, CD11c)− CD19+ B cells were isolated from C57BL/CJ or BALB/CJ mice using FACSAriaTM Fusion sorter and \(10^{6} / \mathrm{ml}\) B cells were cultured in \(50\%\) cancer CM in cRPMI for 7 days in NuncTM Multidishes with UpCellTM Surface (Thermo Fisher) without changing media for 7 days. \(70z / 3\) pre- B cells \((10^{5} / \mathrm{ml})\) were cultured in \(50\%\) cancer CM for up to 30 days with replenishing culture medium every 3- 4 days. Adherent cells (macrophages) were harvested by detaching them at \(4^{\circ} \mathrm{C}\) for 15 mins in PBS. For Giemsa staining, B- MF were fixed with ethanol for 5 mins and Wright- Giemsa stained according to the manufacturer's instruction. CSF1R receptor signaling was blocked with Ki20227 (R&D). + +<--- Page Split ---> + +In vitro assays: For bacterial uptake assay, E. coli (Thermo Fisher) labeled with \(\mathrm{pHrodo^{TM}}\) red (0.1 mg/ml) were cultured with B- MF generated from RAG- GFP for 2h. Cells were washed with PBS, fixed with \(4\%\) formaldehyde and stained with DAPI. For phagocytosis of apoptotic cancer cells, ID8- RFP cells ( \(10^{6} / \mathrm{ml}\) ) were pretreated with 300 nM gemcitabine hydrochloride (Sigma, St. Louis, MO) for 24h, then washed with PBS and cultured with macrophages for 2h. Macrophages were stained with anti- F4/80- FITC Ab and DAPI and phagocytosis was evaluated using Zeiss LSM 710 (Carl Zeiss AG, Jena, Germany) and analyzed with Fiji software. For macrophage proliferation test, BrdU (10 \(\mu \mathrm{M}\) , BD) was added to macrophage cultures at day 5 and the BrdU incorporation was quantified at day 7 using FACSymphonyTM and analyzed by FlowJo. + +T cell suppression assay was described elsewhere (13). Briefly, splenic T cells isolated with \(\mathrm{CD3^{+}}\) T cell enrichment column (R&D) were labeled with \(\mathrm{eFluor^{TM}}\) 450 and cultured with macrophages at 1:10, 1:20 and 1:40 E: T ratios in 96- well flat bottom plates coated with \(5\mu \mathrm{g / ml}\) anti- mouse CD3e antibody (clone 145- 2C11, BD) and free anti- mouse CD28 antibody ( \(2\mu \mathrm{g / ml}\) , clone 37.51, BD) for 4 days. The Treg conversion assay was described elsewhere (5). In brief, FACS- sorted splenic \(\mathrm{CD4^{+}CD25^{- }}\) T cells were cultured with macrophages at 1:5, 1:10 and 1:20 E:T ratios in plates coated with \(5\mu \mathrm{g / ml}\) anti- mouse CD3e antibody and free recombinant murine IL2 ( \(5\mathrm{ng / ml}\) , PeproTech, Rocky Hill, NJ) in for 5 days. Control T cells were cultured with recombinant mouse TGF- \(\beta 1\) ( \(5\mathrm{ng / ml}\) , R&D) in cRPMI without macrophages. + +In vivo experiments: For evaluation of macrophages in vivo, tumor cells were subcutaneously injected into congenic mice, such as 4T1.2 cells and EMT6 cells ( \(1 \times 10^{6}\) ) in BALB/cJ and \(\mu \mathrm{MT}\) mice, and B16- F10, and AT3 and MC- 38 cells ( \(1 \times 10^{6}\) ) in + +<--- Page Split ---> + +C57BL/6J, JHT or Mb1- EYFP mice. ID8- p53-/-- RFP cells \((5 \times 10^{6})\) were intraperitoneally (i.p.) injected into C57BL/6J mice. For in vivo B- MF generation study, C57BL/6J mice with ID8- p53-/-- RFP cells \((5 \times 10^{6})\) in PeC were intraperitoneally injected with BM Lin- CD19+EYFP+ B cells \((5 \times 10^{6})\) sort- purified from Mb1- cre- EYFP mice and then 7 days later, the PeC lavage cells were FACS evaluated. To evaluate tumor- supporting role of B- MF in vivo, \(\mu\) MT mice were intravenously injected with \(3 \times 10^{5}\) in vitro- generated B- MF or PBS 3 and 7 days after subcutaneous challenge with B16- F10 melanoma cells (day 0, n=5). Tumor volume \((\mathrm{V} = \mathrm{a} \times \mathrm{b}, \mathrm{mm}^{2})\) was measured at days 11, 14, 16, 18, and 21, and at day 21, mice were euthanized to evaluate tumor weight and T cells. + +Cellular cholesterol content quantification: Macrophages were fixed with \(4\%\) formaldehyde solution in TBS for 5 mins, then after TBS washes, they were incubated with Filipin III at 1:100 dilution in TBS (5mg/ml stock in \(100\%\) ethanol, Cayman, Ann Arbor, MI) for 60 mins in the dark. Cells were washed with TBS and lipids were quantified with Zeiss LSM 710 and Fiji software, as described above. + +mRNA microarray: For the collected biological samples, the standard RNA extraction protocol was performed by RNeasy Plus Micro kits (QIAGEN, Hilden, Germany), and genome- wide expression was measured using the Agilent platform (Mouse 8X60K v2 and Hs 8X60K v3, Agilent, Santa Clara, CA, USA) according to the manufacture's instruction. Principal Component Analysis (PCA) was performed using the Procomp R function with expression values. Differential expressed genes (DEGs) was assessed using the moderated (empirical Bayesian) t- test implemented in the limma package (version 3.14.4)(44), and correction for multiple hypothesis testing was accomplished by calculating the Benjamini- Hochberg false discovery rate. Enriched + +<--- Page Split ---> + +pathways were discovered by GSEA tool (45) with Molecular Signature Database v7.4. All microarray analyses were performed using the R environment for statistical computing (version 3.6.2). + +scRNA- seq: Sort purified single cell suspensions were loaded into a \(10\times\) Chromium controller (10x Genomics, Pleasanton, CA, USA) and converted to barcoded single cell RNA expression library according to the standard protocol of the Chromium Next GEM Single cell 3' kit (v3.1 chemistry) in Laboratory of Immunology and Molecular Biology, National Institute on Aging, and the single cell 3' gene expression libraries were sequenced on NovaSeq 6000 (Illumina, San Diego, CA, USA) in the Genomics Core facility of the Johns Hopkins School of Medicine. Raw sequencing data was processed using the Cell Ranger version 5.0 (10x Genomics, Pleasanton, CA, USA) pipeline. The raw gene expression matrix was normalized and scaled using the SCTransform method (46) in the Seurat R package (version 4.0) (47). The minimum number of detected genes was set to 1,000, and genes were chosen when they were detected in more than 3 cells. Dimension reduction was performed using principal component analysis (PCA). For visualizing the generated clustering, we used the Uniform Manifold Approximation and Projections (UMAP) plot. We defined clusters with a leiden algorithm using shared nearest neighbor (SNN) in PCA space. From in vitro B- MF and Mo- MF, we generated a total of 12 clusters for in vitro samples (0- 11). Integration of in vivo samples with canonical correlation analysis (CCA) was performed, and we generated 13 clusters for in vivo tumor macrophage samples (0- 12). Finally, we performed a nonparametric Wilcoxon rank- sum test to search for highly expressed genes in the clusters. In addition, human tumor single cell transcriptomes were downloaded from GEO (GSE114725, and GSE146026) and also + +<--- Page Split ---> + +processed with the same pipeline described. We used only macrophage clusters for downstream analysis. All single- cell analyses were performed using the R environment for statistical computing (version 4.0.5). + +ATAC- seq analysis: We utilized a Hi- Seq 2000 machine to sequence the ATAC- seq libraries. We prepared 12 pair- end ATAC- seq libraries including BM CSF1R (+/- ) and Spleen CSF1R (+/- ) samples (n=3 per group). In total, 369M reads were sequenced and average 31M reads were sequenced per sample. We applied NIEHS TaRGETII ATAC- seq pipelines, which are available to the genomics community. All raw reads were trimmed using cutadapt package, and trimmed reads (>36bp minimum alignment length) were mapped against the mm10 reference genome using BWA aligner (48). We used de- duplicated and uniquely mapped reads for peak calling analysis after excluding black- list regions defined by ENCODE (49). The candidate peaks were predicted by MACS peak calling tool (50). In addition, we also applied the DESeq2 (51) to determine differentially accessible regions (DARs); cutoff: Fold change \(>1.5\) , \(\log 2\mathrm{CPM} > 1.2\) , FDR \(< 0.05\) . The differentially accessible regions were submitted for search of potential transcription factor binding sites using HOMER software (52). We used non DARs as background regions in de novo motif analysis. + +Statistical analysis: The results are presented as the mean with each individual data point or in bar graph \(\pm\) SEM. GraphPad Prism (Prism 6; GraphPad Software, Inc) was used to perform statistical analysis. Data were analyzed using Welch t- test or one- way ANOVA. A P- value less than 0.05 was considered significant (**** P<0.0001, ***P<0.001, ** P<0.01, *P<0.05). + +<--- Page Split ---> + +ACKNOWLEDGMENTS: We are grateful to Mrs. Jane Trepel with help with human BC and OC samples; Dr. Xin Wang (NIA) for help with confocal microscopy, Drs. Elin Lehrman, Yongqing Zhang, and Supriyo De (Genomics Core, LGG, NIA) for help with performing microarray experiments and initial data analyses, and data submission to GSE; and Dr. Chandamany Arya (Lilly) for proofreading. + +AUTHOR CONTRIBUTIONS: C. C. performed the research, collected and analyzed data; E. Ra., M.B. and L.Z., performed experiments; B.P. worked on bioinformatics analyses, JMM provided clinical trial samples; A.B. and I.B. supervised the study and wrote the manuscript; and A.B. conceived and designed the study. + +FUNDING: This research was supported by the Intramural Research Program of the National Institute on Aging, NIH. + +COMPETING INTERESTS: The authors are employees of National Institutes of Health, the U.S. government, and declare no competing interests, and there are no patents or patent applications related to this work. + +DATA AND MATERIALS AVAILABILITY: The data associated with this study can be found in the paper or supplementary materials and ATAC- seq data are deposited at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE178716 and =GSE180285 + +SUPLEMENTAL MATERIALS: Figures S1- S8 and Tables S1- 4. + +<--- Page Split ---> + +1. S. N. Ganti, T. C. Albershardt, B. M. Iritani, A. Ruddell, Regulatory B cells preferentially accumulate in tumor-draining lymph nodes and promote tumor growth. Sci Rep 5, 12255 (2015).2. C. Lee-Chang et al., Aging Converts Innate B1a Cells into Potent CD8+ T Cell Inducers. J Immunol 196, 3385-3397 (2016).3. Y. Gu et al., Tumor-educated B cells selectively promote breast cancer lymph node metastasis by HSPA4-targeting IgG. Nat Med 25, 312-322 (2019).4. Q. Li et al., Adoptive transfer of tumor reactive B cells confers host T-cell immunity and tumor regression. Clin Cancer Res 17, 4987-4995 (2011).5. P. B. Olkhanud et al., Tumor-evoked regulatory B cells promote breast cancer metastasis by converting resting CD4+ T cells to T-regulatory cells. Cancer Res 71, 3505-3515 (2011).6. P. B. Olkhanud et al., Thymic stromal lymphopoietin is a key mediator of breast cancer progression. Journal of immunology 186, 5656-5662 (2011).7. K. Wejksza et al., Cancer-produced metabolites of 5-lipoxygenase induce tumor-evoked regulatory B cells via peroxisome proliferator-activated receptor alpha. Journal of immunology 190, 2575-2584 (2013).8. S. Kusmartsev, D. I. Gabrilovich, Effect of tumor-derived cytokines and growth factors on differentiation and immune suppressive features of myeloid cells in cancer. Cancer Metastasis Rev 25, 323-331 (2006).9. K. Movahedi et al., Different tumor microenvironments contain functionally distinct subsets of macrophages derived from Ly6C(high) monocytes. Cancer Res 70, 5728-5739 (2010).10. E. Ragonnaud et al., Tumor-Derived Thymic Stromal Lymphopoietin Expands Bone Marrow B-cell Precursors in Circulation to Support Metastasis. Cancer Res 79, 5826-5838 (2019).11. M. Lelekakis et al., A novel orthotopic model of breast cancer metastasis to bone. Clin.Exp.Metastasis 17, 163-170 (1999).12. M. Bodogai et al., Immunosuppressive and Prometastatic Functions of Myeloid-Derived Suppressive Cells Rely upon Education from Tumor-Associated B Cells. Cancer Res 75, 3456-3465 (2015).13. P. B. Olkhanud et al., Breast cancer lung metastasis requires expression of chemokine receptor CCR4 and regulatory T cells. Cancer Res. 69, 5996-6004 (2009).14. H. Xie, M. Ye, R. Feng, T. Graf, Stepwise reprogramming of B cells into macrophages. Cell 117, 663-676 (2004).15. S. L. Nutt, B. Heavey, A. G. Rolink, M. Busslinger, Commitment to the B-lymphoid lineage depends on the transcription factor Pax5. Nature 401, 556-562 (1999).16. T. Audzevich et al., Pre/pro-B cells generate macrophage populations during homeostasis and inflammation. Proc Natl Acad Sci U S A 114, E3954-E3963 (2017). + +<--- Page Split ---> + +17. E. R. Stanley, M. Cifone, P. M. Heard, V. Defendi, Factors regulating macrophage production and growth: identity of colony-stimulating factor and macrophage growth factor. J Exp Med 143, 631-647 (1976).18. S. R. Almeida et al., Mouse B-1 cell-derived mononuclear phagocyte, a novel cellular component of acute non-specific inflammatory exudate. Int Immunol 13, 1193-1201 (2001).19. D. Kitamura, J. Roes, R. Kuhn, K. Rajewsky, A B cell-deficient mouse by targeted disruption of the membrane exon of the immunoglobulin mu chain gene. Nature 350, 423-426 (1991).20. J. Chen et al., Immunoglobulin gene rearrangement in B cell deficient mice generated by targeted deletion of the JH locus. Int Immunol 5, 647-656 (1993).21. E. Hobeika et al., Testing gene function early in the B cell lineage in mb1-cre mice. Proc Natl Acad Sci U S A 103, 13789-13794 (2006).22. W. Chen, P. Ten Dijke, Immunoregulation by members of the TGFbeta superfamily. Nat Rev Immunol 16, 723-740 (2016).23. L. M. Francisco, P. T. Sage, A. H. Sharpe, The PD-1 pathway in tolerance and autoimmunity. Immunol Rev 236, 219-242 (2010).24. G. Gimpl, Cholesterol-protein interaction: methods and cholesterol reporter molecules. Subcell Biochem 51, 1-45 (2010).25. H. Ohno et al., A c-fms tyrosine kinase inhibitor, Ki20227, suppresses osteoclast differentiation and osteolytic bone destruction in a bone metastasis model. Mol Cancer Ther 5, 2634-2643 (2006).26. A. Delogu et al., Gene repression by Pax5 in B cells is essential for blood cell homeostasis and is reversed in plasma cells. Immunity 24, 269-281 (2006).27. A. P. Ng et al., An Erg-driven transcriptional program controls B cell lymphopoiesis. Nat Commun 11, 3013 (2020).28. S. Taoudi et al., ERG dependence distinguishes developmental control of hematopoietic stem cell maintenance from hematopoietic specification. Genes Dev 25, 251-262 (2011).29. S. R. Himes, S. Cronau, C. Mulford, D. A. Hume, The Runx1 transcription factor controls CSF-1-dependent and -independent growth and survival of macrophages. Oncogene 24, 5278-5286 (2005).30. M. R. Corces et al., Lineage-specific and single-cell chromatin accessibility charts human hematopoiesis and leukemia evolution. Nat Genet 48, 1193-1203 (2016).31. E. Azizi et al., Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment. Cell 174, 1293-1308 e1236 (2018).32. B. Izar et al., A single-cell landscape of high-grade serous ovarian cancer. Nat Med 26, 1271-1279 (2020).33. T. A. Wynn, A. Chawla, J. W. Pollard, Macrophage biology in development, homeostasis and disease. Nature 496, 445-455 (2013).34. M. Donadon et al., Macrophage morphology correlates with single-cell diversity and prognosis in colorectal liver metastasis. J Exp Med 217, (2020).35. H. Jeon, S. C. Blacklow, Structure and physiologic function of the low-density lipoprotein receptor. Annu Rev Biochem 74, 535-562 (2005).36. J. H. Madenspacher et al., Cholesterol 25-hydroxylase promotes efferocytosis and resolution of lung inflammation. JCI Insight 5, (2020). + +<--- Page Split ---> + +37. C. L. Chen, S. S. Huang, J. S. Huang, Cholesterol modulates cellular TGF-beta responsiveness by altering TGF-beta binding to TGF-beta receptors. J Cell Physiol 215, 223-233 (2008). + +38. F. Shojaei et al., G-CSF-initiated myeloid cell mobilization and angiogenesis mediate tumor refractoriness to anti-VEGF therapy in mouse models. Proc Natl Acad Sci U S A 106, 6742-6747 (2009). +39. Z. Qin et al., B cells inhibit induction of T cell-dependent tumor immunity. Nature Medicine 4, 627-630 (1998). +40. S. Gaidamakov et al., Targeted deletion of the gene encoding the La autoantigen (Sjogren's syndrome antigen B) in B cells or the frontal brain causes extensive tissue loss. Mol Cell Biol 34, 123-131 (2014). +41. P. H. Hand, P. F. Robbins, M. L. Salgaller, D. J. Poole, J. Schlom, Evaluation of human carcinoembryonic-antigen (CEA)-transduced and non-transduced murine tumors as potential targets for anti-CEA therapies. Cancer Immunol Immunother 36, 65-75 (1993). +42. M. E. Gatti-Mays et al., A Phase II Single Arm Pilot Study of the CHK1 Inhibitor Prexasertib (LY2606368) in BRCA Wild-Type, Advanced Triple-Negative Breast Cancer. Oncologist 25, 1013-e1824 (2020). +43. J. M. Lee et al., Prexasertib, a cell cycle checkpoint kinase 1 and 2 inhibitor, in BRCA wild-type recurrent high-grade serous ovarian cancer: a first-in-class proof-of-concept phase 2 study. Lancet Oncol 19, 207-215 (2018). +44. M. E. Ritchie et al., limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43, e47 (2015). +45. A. Subramanian et al., Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545-15550 (2005). +46. C. Hafemeister, R. Satija, Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol 20, 296 (2019). +47. Y. Hao et al., Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587 e3529 (2021). +48. H. Li, R. Durbin, Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics 26, 589-595 (2010). +49. H. M. Amemiya, A. Kundaje, A. P. Boyle, The ENCODE Blacklist: Identification of Problematic Regions of the Genome. Sci Rep 9, 9354 (2019). +50. Y. Zhang et al., Model-based analysis of ChIP-Seq (MACS). Genome Biol 9, R137 (2008). +51. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014). +52. S. Heinz et al., Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38, 576-589 (2010). + +<--- Page Split ---> + +Figure 1. Cancer induces differentiation of macrophages from BM B cells. A and B, FACS staining frequency \(\pm\) SEM of co-expression for markers of B cells (CD19) and macrophages (F4/80 \(^+\) ) in A, BALB/CJ and \(\mu \mathrm{MT}\) mice with orthotopic 4T1.2 breast cancer (n=5- 6) and B, C57BL/6 and \(\mathrm{J_{H}T}\) mice with spontaneous ovarian cancer (Mogp- tag and Mogp- \(\mathrm{J_{H}T}\) , respectively, n=4) Frequency of F4/80 \(^+\) CD11b \(^+\) cells within CD19 \(^+\) TIBs (left panels) and CD19 \(^+\) IgM \(^+\) cells within F4/80 \(^+\) TAM (right panels). C, Representative flow plot for F4/80 \(^+\) CD11b \(^+\) macrophage of B cells isolated from BM, spleen, LN and PeC treated with CM from 4T1.2 breast cancer cells. D, Representative Giemsa staining in vitro- generated B- MF. Scale bar represented \(20\mu \mathrm{m}\) . E, Representative FACs plots of F4/80 \(^+\) CD11b \(^+\) macrophage of BM B- cells treated with CM from 4T1.2, EMT6, AT3, B16- F10, and MC- 38 cells. F and G Representative FACs plots for CD68 and CD11b staining of pre- B cell line (70z/3) after 30- day stimulation with (RPMI) or without 4T1.2- CM and composite frequency analysis of F4/80 \(^+\) CD11b \(^+\) and CD68 \(^+\) CD11b \(^+\) cells from four independent experiments. H, Representative Imagestream imaging for surface markers F4/80, CD11b, CD19, and CD20 on in vitro- generated B- MF from naive Mb1- cre- EYFP \(^+\) mice (bottom two panels, H) and WT mice (top two panels, H). I, Frequency \(\pm\) SEM of CD274 \(^+\) and LAP \(^+\) cells within EYFP \(^+\) or EYFP \(^+\) F4/80 \(^+\) CD11b \(^+\) cells quantified 7- day after i.p. injection of BM B cells from naive Mb1- cre- EYFP mice in C57BL/6 mice with ID8 ovarian cancer in PeC (n=3,). \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +Figure 2. Distinct gene expression profiles of B- MF and Mo- MF. A, Representative FACs plots of F4/80 \(^+\) CD11b \(^+\) macrophage of BM B cells or monocytes post 7- day treatment with 4T1.2- CM. Macrophage phenotype gated as F4/80 \(^+\) CD11B \(^+\) from B cell + +<--- Page Split ---> + +(B- MF) and monocytes (Mo- MF) B, Representative expression of B- cell markers CD79a and IgM by FACs analysis on B- MF (Red) and Mo- MF (Orange). C, PCA plot of mRNA expression profiles generated from microarray data of sort purified B- MF (Blue), Mo- MF (Orange) and BM B cells (Green) (n=3). D and E, Bar plots of GSEA predicted pathways enriched in B- MF (D) or Mo- MF cells (E) from the Molecular Signature Database. F and G, UMAP plot of B- MFs (10,563 cells) and Mo- MFs (10,235 cells) analyzed using Seurat with colors depicting clusters by cell type (F) or by gene expression (G). H, Heatmap of top genes with differential expression characterizing the 12 major cell types of in vitro B- MF and Mo- MF single cell clusters. I, Overlay of Mrc1 expression for in vitro B- MF and Mo- MF single cells. J, UMAP plot of sort purified TAMs (10,885 cells) generating 13 unique cell clusters from tumor tissues of 4 mice with 4T1.2 cancers. K, Violin plots of the expression of three in vitro derived B- MF up- regulated DEGs (Egr1, Ier3 and Slc40a1) in in vivo mouse 4T1.2 TAM single cell clusters. + +Figure 3. B- MF and Mo- MF are functionally different. A, Ki67 staining (upper) and BrdU uptake (lower panel) frequency of B- MF and Mo- MF (frequency±SEM, n=2). B and C, Representative Filipin III staining images and quantification (MFI). D and E, Representative images and quantifications of \(\mathrm{RFP}^+\) apoptotic cells in B- MF and Mo- MF cultured with apoptotic ID8- RFP ovarian cancer cells for 2 h. Eight representative fields per sample were quantified and scale bars represent \(20 \mu \mathrm{m}\) (C and E). F, Representative histograms of proliferated \(\mathrm{CD4}^+\) T cells after co- culturing with macrophages at 1:10 (E:T) ratio in the presence of anti- CD3/CD28 Abs for 4 days. Control T cells were cultured alone with (activated) or without (non- activated) anti- CD3/CD28 Abs. Numbers represent + +<--- Page Split ---> + +percentage of proliferated T cells. Macrophages were generated from BM B- cell precursors (B- MF) and monocytes (Mo- MF) with 4T1.2- CM treatment. The results in \(\mathbf{B} - \mathbf{F}\) were independently reproduced at least three times. \(\mathbf{G} - \mathbf{J}\) , B- MF support tumor growth in vivo. \(\mu \mathrm{MT}\) mice were i.v. transferred with \(3\times 10^{5}\) in vitro- generated B- MF or PBS 3 and 7 days after s.c. challenge with B16- F10 melanoma cells (n=5). Each symbol is for a single mouse; Y- axis is for \(\% \pm \mathrm{SEM}\) of tumor- infiltrated \(\mathrm{CD4^{+}}\) T cells in \(\mathrm{CD45^{+}}\) cells (H) and \(\mathrm{IFN}\gamma^{+}\) in \(\mathrm{CD4^{+}}\) T cells (I) and tumor weight \(\pm \mathrm{SEM}\) (J). \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**}\) \(\mathrm{P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +Figure 4. CSF1/CSF1R signaling axis is critical for B- MF differentiation. A, Numbers of \(\mathrm{CD19^{+}}\) (left) and \(\mathrm{CD93^{+}CD19^{+}}\) (middle) B cells and frequency of \(\mathrm{CSF1R^{+}}\) within \(\mathrm{CD93^{+}CD19^{+}}\) B cells (right) in the spleen of naive and 4T1.2 tumor- bearing mice (n=5). B, ELISA measurements of secreted M- CSF in media conditioned by 4T1.2, EMT6, AT3, MC- 38, B16- F10, and ID8 cells (n=6, pg/ml). C, Representative FACS plots of the \(\mathrm{F4 / 80^{+}CD11b^{+}}\) B- MF converted from sort- purified BM CSF1R+ and CSF1R- B- cell precursors and splenic CSF1R+ and CSF1R- B cells after 7- day treatment with 4T1.2- CM. D and E, Representative FACS plots of \(\mathrm{F4 / 80^{+}CD11b^{+}}\) (B- MF) from BM B- cell precursors after 7- day treatment with 4T1.2- CM (D) and pre- B cell line 70z/3 after 30- day treatment with 4T1.2- CM (E) with and without anti- MCSF Ab or Ki- 20227 treatments (a specific MCSF- R inhibitor). F and G, Representative FACS plot (F) and quantification (G) of the generation of B- MF after 7- day 4T1.2- CM treatment on cells isolated from WT or mice with conditional CSF1R deficiency in B cells (Mb1- CSF1RFlox/Flox) (n=6). Results for all panels were independently reproduced minimally three times. \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +<--- Page Split ---> + +Figure 5. Cancer targets BM CSF1R+PAX5Low B-cell precursors. A, FACs analysis of Pax5 expression of freshly isolated BM CSF1R+ or CSF1R+ B cell precursors (n=5,) B, FACs analysis of Pax5 expression of pre- B cell (70z/3 cells) treated with indicated cancer CM (n=3). C, FACs analysis of Pax5 expression of BM Lin- CD19+B220+CD93+IgM- IgD- CSF1R+ B cell precursors treated with M- CSF for 48h (n=3). D, 3D PCA plot of chromatin accessibility generated from ATAC- seq data of BM CSF1R+ and CSF1R- B- cell precursors and spleen CSF1R+ and CSF1R- - follicular B cells E, Heatmap of differentially accessible regions (DARs) between BM CSF1R+ and CSF1R- B- cell precursors (FDR<0.05). F, Top, highly significant de novo motifs are more open in BM CSF1R+ compared to BM CSF1R- as well as Spleen CSF1R+ and CSF1R- - cells (678 sites). G, mRNA microarray heatmap of macrophage related- DEGs in B cells isolated from PB of patients with breast cancer (BC, n=8) compared to healthy donors (HD, n=7). Scale bar is for expression z- score. H, Cell frequency of CSF1R+ (left), CD68+ (middle) and LDLR+ (right) cells within CD19+CD10+ B cells of PB from healthy donors (HD) or patients with ovarian cancer (OC) (n=5-7). I and J, UMAP of scRNA sequencing data of macrophage cells (left) and expression levels of in vitro B-MF enriched genes (right) in published human BC (I) and OC (J) datasets. Error bars represent SEM, **** P<0.0001, ***P<0.001, ** P<0.01, *P<0.05 between indicated comparisons. + +<--- Page Split ---> +![](images/Figure_5.jpg) + + +<--- Page Split ---> +![PLACEHOLDER_34_0] + + +<--- Page Split ---> +![PLACEHOLDER_35_0] + + +<--- Page Split ---> +![PLACEHOLDER_36_0] + + +<--- Page Split ---> +![PLACEHOLDER_37_0] + +
Figure 5
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryTables14. xlsx SupplFigsSl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5_det.mmd b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9a21b3cf1e6215198e758010c6da6ec76de8510c --- /dev/null +++ b/preprint/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5/preprint__140b9f249da4971571ef7250e3e13c2ef2fbd794c681f2056c2cffa6cc43d4f5_det.mmd @@ -0,0 +1,344 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 909, 177]]<|/det|> +# Cancer coopts differentiation of B-cell precursors into macrophages + +<|ref|>text<|/ref|><|det|>[[44, 195, 290, 238]]<|/det|> +Chen Chen National Institute on Aging + +<|ref|>text<|/ref|><|det|>[[44, 244, 290, 285]]<|/det|> +Bongsoo Park National Institute on Aging + +<|ref|>text<|/ref|><|det|>[[44, 290, 290, 332]]<|/det|> +Emeline Ragonnaud National Institute on Aging + +<|ref|>text<|/ref|><|det|>[[44, 337, 290, 378]]<|/det|> +Monica Bodogai National Institute on Aging + +<|ref|>text<|/ref|><|det|>[[44, 383, 140, 421]]<|/det|> +Le Zong NIA/NIH + +<|ref|>text<|/ref|><|det|>[[44, 428, 290, 469]]<|/det|> +Jung- Min Lee Center for Cancer Research + +<|ref|>text<|/ref|><|det|>[[44, 475, 290, 516]]<|/det|> +Isabel Beerman National Institute on Aging + +<|ref|>text<|/ref|><|det|>[[44, 521, 647, 563]]<|/det|> +Arya Biragyn ( biragyna@mail.nih.gov ) National Institute on Aging https://orcid.org/0000- 0001- 5276- 6102 + +<|ref|>sub_title<|/ref|><|det|>[[44, 604, 102, 621]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 641, 890, 661]]<|/det|> +Keywords: B- cell precursors, macrophage, transdifferentiation, immune suppression, RNA profiling + +<|ref|>text<|/ref|><|det|>[[44, 679, 350, 698]]<|/det|> +Posted Date: September 13th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 716, 463, 736]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 858433/v1 + +<|ref|>text<|/ref|><|det|>[[44, 754, 910, 797]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 833, 914, 876]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 14th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33117- y. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 715, 110]]<|/det|> +## Cancer coopts differentiation of B-cell precursors into macrophages + +<|ref|>text<|/ref|><|det|>[[144, 157, 868, 216]]<|/det|> +Chen Chen \(^{1}\) , Bongsoo Park \(^{2}\) , Emeline Ragonnaud \(^{1}\) , Monica Bodogai \(^{1}\) , Le Zong \(^{2}\) , Jung- Min Lee \(^{3}\) , Isabel Beerman \(^{2*}\) and Arya Biragyn \(^{1,4,5*}\) + +<|ref|>text<|/ref|><|det|>[[143, 268, 857, 394]]<|/det|> +From: \(^{1}\) Immunoregulation Section, Laboratory of Immunology and Molecular Biology; \(^{2}\) Epigenetics and Stem Cell Aging Unit, Translational Gerontology Branch, and National Institute on Aging, Baltimore, MD; \(^{3}\) Women's Malignancies Branch, Center for Cancer Research, National Cancer Institute, Bethesda, MD; \(^{4}\) Lead contact; \(^{5}\) Senior author + +<|ref|>text<|/ref|><|det|>[[143, 458, 821, 517]]<|/det|> +\* Corresponding authors: Arya Biragyn, Ph.D., E- mail: biragyna@mail.nih.gov and Isabel Beerman, Ph.D., E- mail: isabel.beerman@nih.gov + +<|ref|>text<|/ref|><|det|>[[143, 583, 555, 602]]<|/det|> +Running title: B cells transdifferentiation in cancer + +<|ref|>text<|/ref|><|det|>[[143, 625, 867, 680]]<|/det|> +Keywords: B- cell precursors, macrophage, transdifferentiation, immune suppression, RNA profiling + +<|ref|>text<|/ref|><|det|>[[142, 701, 820, 722]]<|/det|> +Conflict- of- interest disclosure: The authors declare no conflicting financial interests + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 253, 108]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[142, 123, 857, 529]]<|/det|> +We recently reported that some cancers induce accumulation of bone marrow (BM) B- cell precursors in the spleen to convert them into metastasis- promoting, immunosuppressive B cells. Here, using various murine tumor models and samples from humans with breast and ovarian cancers, we provide evidence that cancer cells also coopt differentiation of the extra nodal B- cell precursors to generate macrophages (termed B- MF). We link the transformation to a small subset of CSF1R+ Pax5Low cells within BM pre- B and immature B cells and cancer- secreted M- CSF that downregulates Pax5 via CSF1R signaling. Thus, cancer generates tumor- associated macrophages (TAM) from B- cell precursors in addition to their primary source, monocytes. Based on their differences from monocyte- derived TAM, such as a superior ability to induce FoxP3+ Tregs, suppress proliferation of T cells and more efficiently phagocytize apoptotic cells, we propose that cancer generates B- MF to mediate cancer escape. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 87, 857, 888]]<|/det|> +The role of B cells in cancer remains poorly understood despite reports of both positive and negative associations of B cells with the disease outcome. Even in the same murine model, T2- MZ B cells and aging- associated B cells can either increase or mitigate B16- F10 melanoma growth in C57BL/6 mice, respectively (1, 2). B cells can also inhibit or enhance metastasis in BALB/c mice with orthotopic 4T1 breast cancer cells (3, 4). At least some cancer- promoting functions of B cells are attributed to their immunosuppressive, regulatory subsets termed Bregs, whose generation and function appear to be actively controlled by cancer itself. To enforce this regulation, cancer cells secrete various factors that affect survival and differentiation of B cells, including B lymphocyte stimulator (BLyS/BAFF), thymic stromal lymphopoietin (TSLP), colony stimulating factors (CSF such as M- CSF, GM- CSF and G- CSF) and lipid mediators such as 5- lipoxygenase (5- LO) metabolites (5- 10). In BALB/c mice with orthotopic 4T1 breast cancer, a model for human triple negative breast cancer (11), our group linked 5- LO metabolite- induced activation of PPAR \(\alpha\) to the conversion of B cells into TGF \(\beta^{+}\) CD25+ Bregs (tBregs) (7). These converted cells then promote metastasis acting as potent suppressors of antitumor T cell activity and inducers of FoxP3+ Tregs and MDSCs via the TGF \(\beta\) /TGF \(\beta\) RII axis (5, 12). Breast and ovarian cancer cells secret TSLP both to prepare the metastasis “soil” in the lungs and to accumulate BM B- cell precursors in the spleen for conversion to tBregs. In the lungs, TSLP enables infiltration of CCR4+ cancer cells and their protector CCR4+FoxP3+ Tregs by upregulating expression of CCL17 (6, 13), while in the BM, TSLP downregulates expression of CXCR4 and \(\alpha 4\beta 1\) in B- cell precursors, causing their premature emigration and accumulation in the spleen to be converted into TGF \(\beta^{+}\) CD25+ tBregs (10). M- CSF and GM- CSF from cancer cells promote differentiation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 87, 857, 423]]<|/det|> +and survival of cancer- promoting myeloid suppressive cells and tumor- associated macrophages (TAM) from BM monocytes (8, 9). Although bifurcation of myeloid and lymphoid lineage from multipotent progenitors occurs before specialization of B- cell progenitors in BM, surprisingly some B- cell precursors appear to retain macrophage differentiation potential. For example, the differentiation of BM B- cell precursors can be diverted towards macrophages upon forced expression or deletion of a single transcription factor (14, 15). Anecdotal reports indicate that this transdifferentiation can also occur in BM of naive mice, which is linked to a small subset of biphenotypic (CD19+B220+ CD16/32++CD11b+) B- cell precursor cells (16). However, the biological consequence of this rare event and whether cancers affect this process remain unknown. + +<|ref|>text<|/ref|><|det|>[[142, 437, 857, 701]]<|/det|> +Here, we report that cancers frequently induce the transdifferentiation of the bona fide BM Csf1R+Pax5Low pre- B and immature B (iB) cells into TAM (termed B- MF). To do this, cancers first drive accumulation of B- cell precursors/iB cells in the spleen and tumors. Unlike “classical” monocyte- derived TAM (9), cancer appears to use B- MF to mediate cancer escape via suppression of antitumor T cells. This does not appear to be a mouse specific phenomenon, as B- MF can be detected in patients with breast and ovarian cancers, and we present evidence of B- MF signatures in published scRNA sequence data of human cancers. + +<|ref|>sub_title<|/ref|><|det|>[[144, 752, 237, 769]]<|/det|> +## RESULTS + +<|ref|>text<|/ref|><|det|>[[143, 786, 856, 876]]<|/det|> +B- cell markers are expressed in TAM. We previously reported that some cancers mobilize BM B- cell precursors in the spleen and tumors (10) to subsequently convert them into \(\mathrm{TGF}\beta^{+}\) tBregs (5, 7, 12). Microarray transcription profiling of B- cell precursors and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 80, 857, 890]]<|/det|> +tBregs from mice with BALB/c mice with orthotopic 4T1.2 breast cancer surprisingly revealed significant upregulation of macrophage-associated genes, such as CD68, Csf1r (encodes CSF1R), Cebpb (CCAAT Enhancer binding protein beta), Cebpg (CCAAT Enhancer binding protein gamma), Ccl2 (CCL2), and Csf1 (M-CSF) (Fig. S1A). Given that CEBPB and CSF1R play essential roles in defining macrophage fate (14, 17) and anecdotal reports on retention of macrophage differentiation potential in B-cell precursors (16, 18), we wondered whether the upregulation of macrophage genes indicates that cancer hijacks B-cell differentiation, pushing them to macrophages. To test this possibility, we FACS evaluated tumor-infiltrating B cells (TIB, gated on CD19+) and TAM (gated on F4/80+) in WT BALB/c mice and \(\mu \mathrm{MT}\) BALB/c mice (B-cell differentiation halted at the pro-B-cell stage (19)) bearing orthotopic 4T1.2 breast cancer. While both TIB and TAM in WT mice contained a small but significant proportion of cells co-expressing B-cell and macrophage markers, the double lineage expressing cells were not present in \(\mu \mathrm{MT}\) mice (Fig. 1A, gating strategy is in Fig. S1B). We also evaluated if these cells were present in other tumor models. TIB and TAM co-expressing B-cell and macrophage markers were found in WT BALB/c mice with orthotopic EMT6 breast cancer, and WT C57BL/6 mice with other orthotopic cancers (AT3 breast and MC38 colon cancer) or spontaneous ovarian cancer (Mogp), but were not detected in \(\mu \mathrm{MT}\) or \(\mathrm{J}_{\mathrm{H}}\mathrm{T}\) mice with these cancers (Fig. 1B and Fig. S1C, D). Of note, in \(\mathrm{J}_{\mathrm{H}}\mathrm{T}\) mice, B-cell differentiation is blocked at the pro-B-cell stage (20). A comprehensive FACS evaluation of TAM revealed they do indeed express additional B-cell-specific markers, including Mb1/CD79a, CD20, IgM, and IgD (Fig. S1D-G). To further confirm our finding, we used a B-cell lineage tracer Mb1-EYFP mouse model (21), which expresses enhanced yellow fluorescent protein (EYFP) as a readout for the Ig- \(\alpha\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 855, 213]]<|/det|> +receptor exclusively expressed in B cells (except plasma cells). Upon challenge with MC38 tumor cells, \(\mathrm{EYFP^{+}F4 / 80^{+}CD11b^{inter / low}}\) TAM were significantly increased in the tumor (Fig. S2). In sum, cancers cause a marked increase of macrophage- like cells derived from B cells (hereafter referred to as B- MF). + +<|ref|>text<|/ref|><|det|>[[142, 222, 857, 880]]<|/det|> +Cancer induces B- cell- to- macrophage differentiation. To explain the accumulation of B- MF, we tested whether cancer cells trans- differentiate them from B cells. Sort- purified Lin \(\mathrm{CD19^{+}B220^{+}B}\) cells ( \(>98\%\) purity) from naïve mouse bone marrow (BM), spleen, lymph nodes (LN) and peritoneal cavity (PeC) were cultured for 7 days in conditioned medium (CM) of 4T1.2 cancer cells (4T1.2- CM). Of note, unless specified, from here on this incubation time was chosen because of a noticeable upregulation of macrophage markers (Fig. S3A). 4T1.2- CM, but not control medium, significantly upregulated F4/80 and CD11b in B cells from BM, but not from spleen, LN or PeC (Fig. 1C). The BM B cells acquired features associated with macrophages, such as adherence to plastic, enlarged vacuolar cytoplasm, and ability to phagocytize fluorochrome- labelled \(E\) . coli (Fig. 1D and Fig. S3B). Independent experiments with sort- purified B- cell subsets revealed that only BM B- cell precursors and iB cells, but not splenic transitional, follicular (FOB), or marginal zone (MZB) B cells, can become B- MF (Fig. S3C; gating strategy in Fig. S3D). Consistent with the accumulation of B- MF in mice with various tumors, CM from almost every type of cancer cell (except B16- F10 melanoma cells) induced the generation of B- MF from BM B- cell precursors/iB cells (BMBP) and pre- B- cell line 70z/3 (Fig. 1E- G and Fig. S3E). In single cell Image stream FACS analysis, macrophage and B- cell markers were indeed co- expressed on single B- MF cells derived from BM Mb1- EYFP+ B cells (Fig. 1H). To rule out artifacts of in vitro experiments, we also performed a 7- day + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 320]]<|/det|> +in vivo conversion assay by i.p. implanting sort-purified BMBP from naïve Mb1- EYFP mice into PeC of C57BL/6 mice with or without ID8 ovarian cancer in PeC. Tumor- bearing mice contained significantly higher frequency of EYFP+ B- MF in PeC than naïve mice (Fig. 1I and Fig. S3F). These cells also markedly upregulated TGFβ/LAP and PD- L1 (Fig. 1I), the two factors utilized in immunoregulation (22, 23). Taken together, we concluded that cancer uses BMBP to generate TAM in addition to their hitherto known source, monocytes (9). + +<|ref|>text<|/ref|><|det|>[[142, 333, 857, 884]]<|/det|> +B- MF and Mo- MF differ functionally. To understand why cancer generates TAM from B cells, we ex vivo generated B- MF and monocyte- derived macrophages (Mo- MF) from sort- purified BMBP and monocytes, respectively, by stimulating with 4T1.2- CM, as described above. While both populations adhered to plastic and comparably upregulated F4/80 and CD11b, only B- MF expressed the B- cell- specific markers (CD79a and IgM, Fig. 2A, B and Fig. S4A). mRNA microarray analysis of B- MF and Mo- MF revealed that regardless of origin, the two populations share expression of numerous macrophage- related genes and do not robustly expresses B- cell signature profiles (Fig. S4B- D and Table S1). Principal component analysis (PCA) and transcription profiling clearly separate B- MF from Mo- MF (Fig. 2C and Table S2). B- MF expressed higher levels of genes involved in fatty acid metabolism, cell cycle, steroid- cholesterol biosynthesis and downregulated expression of pro- inflammatory and IFNγ response genes (Fig. 2D, E). While Mo- MF had more robust M1- like transcription, M2- skewing was more pronounced in B- MF (Fig. 2E and Fig. S4E). The unique transcription profiles were also confirmed in single cell RNA sequencing (scRNA- seq) of in vitro generated B- MF (10,563) and Mo- MF (10,235) cells, with UMAP clustering identifying mostly separate cell- clusters of macrophages derived + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 85, 857, 600]]<|/det|> +from either B cells or monocytes (Fig. 2F and Table S3). We distinguished 12 cell clusters with the Leiden algorithm, using shared nearest neighbor (SNN) in PCA space and identified the key genes establishing the six clusters accounting for the majority of single cells (Fig. 2G, H and Fig. S4F). B- MF appeared to be more phagocytic than Mo- MF, as they markedly upregulated Mrc1 (encodes CD206, Fig. 2I). Given the unique transcriptional signatures of the B- MF, we next examined scRNA profiles from four independent samples of TAM in mice with 4T1.2 cancer (Fig. 2J). Using signature genes identified from in vitro generated macrophages (Fig. 2H) we noted three clusters (0, 6, and 8) with robust expression of genes identified in B- MF (Fig. S4H). Cluster 8 also strongly overlapped with a mixed macrophage population (Cluster 4, Fig. 2G and Fig. S4F- H), suggesting that only clusters 0 and 6 were most similar to B- MF. We also examined expression of three genes robustly expressed in B- MF and they were also expressed in TAM clusters 0 and 6 (Fig. 2K). In contrast, expression of key Mo- MF genes was mostly found in clusters 1, 3, 4, and 5 (Fig. S4G), suggesting B- MF and Mo- MF retain traceable and different transcription profiles in vivo. + +<|ref|>text<|/ref|><|det|>[[142, 611, 857, 876]]<|/det|> +To confirm B- MF and Mo- MF differences at the functional level, first we examined their proliferation. Only B- MF expressed higher levels of Ki67 and readily incorporated BrdU (pulsed at day 6 and tested at day 7 of the culture) (Fig. 3A, B and Fig. S5A), indicating their proliferative and self- maintaining state. Given the changes seen in fatty acid metabolism (Fig. 2D), we stained the cells with filipin III, a fluorescent polyene antibiotic that binds to free cholesterol (24). Compared with Mo- MF, the cellular cholesterol content was significantly upregulated in both in vitro and in vivo generated B- MF (Fig. 3C, D and Fig. S1E- G). To evaluate phagocytosis, B- MF and Mo- MF were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 80, 857, 716]]<|/det|> +cultured with fluorochrome- labeled apoptotic cancer cells. B- MF phagocytized apoptotic cells significantly better than Mo- MF (Fig. 3E, F). Given B- MF markedly upregulated TGFβ/LAP and PD- L1 (Fig. 1I), we tested their function in a 4- day T- cell suppression assay (5). When B- MF or Mo- MF were cultured with naïve mouse splenic T cells stimulated with anti- CD3/CD28 antibodies, only B- MF significantly inhibited proliferation of \(\mathrm{CD4^{+}}\) T cells and, at lesser extent, \(\mathrm{CD8^{+}}\) T cells (Fig. 3G and Fig. S5B- E). Because TGFβ is involved in induction of FoxP3+ Tregs (5), we also performed an in vitro Treg conversion assay by culturing the two macrophages with splenic \(\mathrm{CD25^{+}CD4^{+}}\) T cells in the presence of anti- CD3/CD28 antibodies and IL2 for 4 days. Compared with Mo- MF, B- MF induced higher levels of FoxP3+ Tregs (Fig. S5F), further suggesting that cancer presumably uses B- MF to mediate immunoregulation and thereby tumor progression. We therefore tested whether B- MF can reverse the retarded growth of B16- F10 melanoma in \(\mu \mathrm{MT}\) mice (n=5/group, Fig. 3G), which we previously linked to the lack of B cells (12). Intravenous transfer of B- MF significantly decreased tumor- infiltrated \(\mathrm{CD4^{+}}\) T cells (and at lesser extent \(\mathrm{CD8^{+}}\) T cells) and their IFN- expressing subsets (Fig. 3H, I and Fig. S5H) and markedly increased growth of B16- F10 melanoma in \(\mu \mathrm{MT}\) mice (p<0.05, Fig. 3J and Fig. S5G). In contrast, T cells in the spleen and dLN were not affected by the B- MF transfer (not depicted), suggesting that B- MF primarily regulate tumor- infiltrating T cells. + +<|ref|>text<|/ref|><|det|>[[143, 725, 857, 885]]<|/det|> +Cancer mobilizes BMBP in the spleen to convert them to B- MF. Recently, we reported that breast cancer induces tBregs from BM pre- B cells. To do this, cancer first stimulates accumulation of BMBP in the spleen and tumor by causing their premature emigration from BM (10). We therefore reasoned that cancer targets the same pool of BMBP accumulated in the spleen to generate B- MF. In support of this hypothesis, the total + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 87, 857, 318]]<|/det|> +number of CD93+ BMBP were markedly increased in the circulation (but decreased in BM) of mice with 4T1.2 and Mogg cancers (Fig. 4A and Fig. S6A, B) as compared with naïve mice (Fig. S6A, B). Stimulation with 4T1.2- CM readily generated B- MF from CD93+ BMBP sort- purified from spleens of mice with 4T1.2 cancer (Fig. S6C) but failed to do so if B cells were isolated from spleens of naïve mice, regardless of CD93 expression (Fig. S6C and Fig. S3C), implying that cancer first drives accumulation of BMBP in the spleen to convert them into B- MF. + +<|ref|>text<|/ref|><|det|>[[142, 330, 857, 880]]<|/det|> +Cancer targets CSF1R+CD93+ BMBP by secreting M- CSF. To understand the mechanism of the B- MF generation, we analyzed CM of cancer cells for secreted factors that could affect differentiation of macrophages. M- CSF, a regulator of macrophage differentiation and survival (8, 17), was among the factors that was highly increased in the cancer cells that induce B- MF (Fig. 4B). Conversely, M- CSF was almost absent in CM from B16- F10 cells (Fig. 4B and Fig. S6D, E), which did not induce the generation of B- MF (Fig. 1E and Fig. S3E). Compared to naïve mice, serum M- CSF was also significantly upregulated in mice with 4T1.2 cancer (Fig. S6F). Importantly, BM CD93+CD19+ BMBP of naïve mice expressed its cognate receptor CSF1R (about 15% iB cells, 6% of pre- B cells and 1.5% pro- B cells, Fig. S6G). In mice with 4T1 cancer, CSF1R+CD93+CD19+ BMBP were markedly reduced in BM but increased in the spleen (Fig. 4A and Fig. S6G), consistent with their cancer- induced emigration from BM, as discussed above. To link these CSF1R+CD93+CD19+ BMBP to the generation of B- MF, we performed the in vitro macrophage conversion assay by stimulating sort- purified CSF1R+ and CSF1R- B cells from BM and spleen of naïve mice with 4T1.2- CM. While CSF1R+ BM B cells readily generated B- MF, the BM CSF1R- subset failed to do so (Fig. 4C). Consistent with an + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 87, 857, 494]]<|/det|> +inability of splenic CD93+ B cells of naïve mice to generate B- MF (Fig. S6C and Fig. S3C), we failed to convert naïve mouse splenic B cells into B- MF regardless of CSF1R expression (Fig. 4C). We also cultured primary BM BMBP or 70z/3 cells with 4T1.2- CM in the presence or absence of neutralizing M- CSF antibody (Ab) or Ki20227, a specific inhibitor of c- Fms/CSF1R (25). Both cells failed to generate B- MF upon M- CSF neutralization or CSF1R signaling inhibition (Fig. 4D, E and Fig. S6H, I). To rule out artifacts of in vitro assay, we created mice with conditional CSF1R deficiency in B cells (Mb1- CSF1RFlox/Flox mice, gating strategy in Fig. S7). Unlike WT littermates or monocytes from Mb1- CSF1RFlox/Flox mice, the loss of CSF1R in BMBP significantly impaired the cancer CM- induced B- MF differentiation (Fig. 4F, G). Of note, the residual macrophage differentiation seen in Fig. 4F is presumably due to CSF1R expression preceding Mb1 expression and thus only pro- B cells and onward will have Csf1r deletion. + +<|ref|>text<|/ref|><|det|>[[142, 506, 857, 842]]<|/det|> +Given that PAX5 is the key pro- B cell factor that represses Csf1r and other myeloid lineage- specific genes (15, 26), we reasoned that cancer decreases levels of this transcription factor using M- CSF. FACS staining confirmed that Pax5 was markedly decreased in BM CD93+ BMBP, particularly in CSF1R+ but not CSF1R- subsets, from mice with 4T1.2 or Mogp cancer (Fig. 5A and Fig. S8A, C). Importantly, Pax5 was also significantly decreased in BM CSF1R+ BMBP from naïve mice and 70Z/3 cells upon treatment with 4T1- CM or M- CSF (Fig. 5B, C and Fig. S8B). As Pax5 deficiency alone is sufficient to render pro- B cells susceptible to myeloid differentiation (15), we concluded that cancer uses M- CSF to reduce expression of Pax5 in CSF1R+ CD93+ BMBP and thereby promote macrophage differentiation. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 80, 857, 880]]<|/det|> +To further understand the B- cell susceptibility towards macrophage conversion, we analyzed chromatin accessibility by performing ATAC- seq on CSF1R+ and CSF1R- BMBP isolated from both BM and spleen. PCA clustering showed the most robust differences in chromatin profiles were driven by the location of the BMBP (BM vs spleen) regardless of CSF1R expression, driving the PC2 axis (blue & purple vs orange & green, Fig. 5D). The chromatin landscapes of the CSF1R+ and CSF1R- BMBP isolated from BM (orange and green, Fig. 5D) also significantly differed from each other, driving the PC3 axis. We then examined the differentially accessible regions (DARs) between CSF1R+ and CSF1R- cells isolated from spleen or BM. Whereas comparisons between CSF1R+ and CSF1R- cells from spleen did not show any differences reaching our threshold for significance, confirming their close clustering on the PC3 axis; the BM CSF1R+ cells contained significantly more open chromatin than the BM CSF1R- cells (Fig. 5D, E). These data suggest the BM CSF1R+ BMBP may have a more permissive chromatin environment, susceptible to macrophage differentiation signals. As the spleen- derived B cells and BM CSF1R- cells were refractory to macrophage conversion (Fig. 4C), we looked at DARs with less accessibility in CSF1R- compared to CSF1R+ BM cells (749 loci) in spleen cells to determine if these regions remain closed and potentially “lock in” the lymphoid lineage potential. Indeed, the overwhelming majority of regions with decreased accessibility in the BM CSF1R- cells remained closed in the cells from spleen (679 of 749). Evaluation of these consensus open regions found in BM CSF1R+ cells for potential transcription factor binding sites permitting macrophage differentiation showed significantly increased accessibility of ERG and RUNX1 sites (Fig. 5F). ERG is known to be expressed both in myeloid and lymphoid progenitor cells (27) and has particular importance in early + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 318]]<|/det|> +hematopoietic progenitor cells as it binds to coregulators such as RUNX and GATA (28). RUNX1 regulates growth and survival of macrophages via binding to promoter and enhancer regions of Csf1r and upregulating its expression (29). Runx1 is also robustly expressed in early progenitor and myeloid- committed progenitor cells (30). Thus, the increased accessibility to binding sites of both ERG and RUNX1 suggests a potentially more primitive, permissive chromatin state allowing for myeloid lineage transformation of the BM CSF1R+ cells. + +<|ref|>text<|/ref|><|det|>[[142, 330, 857, 880]]<|/det|> +B- MF- generating CSF1R+ BMBP accumulate in humans with cancer. We recently reported that peripheral mobilization of BMBP also occurs in humans with breast cancer (BC) (10), suggesting the generation of B- MF. To test this possibility, we FACS evaluated PB of healthy donors (HD, n=7) and patients with BC (n=8). Compared with HD, PB of BC was markedly increased in CSF1R+ BMBP (Fig. S8D), as we described in mice with cancer. Moreover, microarray transcription profiling of sort- purified B cells from PB of BC patients revealed that they significantly upregulated macrophage- associated genes, such as Cebpa, Marco, and Csf1r, as compared with B cells from HD (Fig. 5G). We also FACS evaluated B- cells from PB of patients with ovarian cancer (OC, n=5). Compared with HD, OC patients significantly increased CSF1R+ BMBP (Fig. 5H and Fig. S8E) with upregulated expression of CD68 and LDLR (Fig. 5H), similar to mice with cancer. Using recently published scRNA- seq data of tumor- infiltrated immune cells from patients with breast cancer (31), we also found a macrophage cluster with overlapping signatures of B- MF- like cells (cluster 3, Fig. 5I) by examining genes with differential expression defined in murine in vitro- generated B- MF (cluster 0, Fig. 2H). In particular, cluster 3 was enriched for expression of EGR1, IER2, IER3, and SLC40A1, which were major drivers of identity + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 856, 249]]<|/det|> +for murine in vitro- generated B- MF (Fig. 5I). Similarly, in the single- cell transcriptome data from human high- grade serous OC (32), we also detected the B- MF- like signature in macrophages (Cluster 0, Fig. 5J), although with a lesser overlap than in BC, further suggesting that human cancers can promote the B- cell transdifferentiation into macrophages. + +<|ref|>sub_title<|/ref|><|det|>[[144, 299, 266, 317]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[142, 330, 857, 880]]<|/det|> +BMBP undergo a series of subsequent and tightly regulated differentiation steps after their bifurcation from multipotent cells to committed myeloid/lymphoid lineage cells. Despite this, experiments with forced expression or inhibition of a single transcription factor or mutations that drive leukemogenesis (14, 15) reveal that BMBP retain plasticity and myeloid transdifferentiation potential. Unlike these artificial or rare events, here we report that the B- cell- to- macrophage transdifferentiation is commonly used by murine cancers to generate TAM/B- MF. In PB of humans with metastatic/recurrent triple- negative BC and high- grade serous OC, we also detect a significant increase of CSF1R+ CD68+LDLR+ BMBP which also express the macrophage- specific genes Cebpa, Cebpb, and Marco. Importantly, the B- MF signature is also identifiable within unique macrophage clusters using recently published scRNAseq profiles of tumor- infiltrating cells in patients with BC (31) and high- grade serous OC (32). Although the characterization of human B- MF is a topic of a separate study, our results suggest that human and murine cancers transdifferentiate BMBP into macrophages, adding one more feature to the heterogeneity and plasticity of TAM. The inflammatory and antitumor activities of TAM at the early stages of tumor can shift to proangiogenic and tumor- supporting M2- like phenotypes as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 856, 283]]<|/det|> +tumor progresses (33), presumably when B- MF would be induced. Interestingly, B- MF resemble both small (S)- TAM and large (L)- TAM (which associate with a poor disease outcome) recently identified in human colorectal liver metastasis (34). The lipid metabolism and phagocytosis genes of B- MF (Fasn, Pltp, Acat1, C1qa, and C1qb) are upregulated in L- TAM, while LDLR, Hmgcr as well S100a8, Vcan, and Thbs1 are increased in S- TAM. + +<|ref|>text<|/ref|><|det|>[[143, 295, 857, 888]]<|/det|> +Although TAM are primarily derived from monocytes (9), our results clearly show that at least some originate from bona fide B cells. The biological relevance of this redundancy in generation of TAM remains poorly understood; however, based on our comparisons of the side- by- side generated B- MF and Mo- MF, we think that the two macrophages may serve different purposes. Transcriptionally, B- MF preferentially upregulate expression of genes involved in cell cycle, fatty acid metabolism, and steroid- cholesterol biosynthesis, implying they utilize unique metabolic and inflammatory functions. Unlike Mo- MF, B- MF proliferate, i.e. self- maintain and thus may persist longer in the tumor. B- MF markedly upregulate surface expression of LDLR, which removes extracellular cholesterol/LDL (35), and this could explain the higher levels of intracellular cholesterol and lipids in B- MF compared to Mo- MF. Consistent with significant upregulation of genes associated with phagocytosis, M2- skewing and immunosuppressive functions (PD- L2, B7- H3, Marco, TGFβ) and downregulation of pro- inflammatory and IFNγ response genes, B- MF express higher levels of surface MRC1 (CD206), PD- L1 (CD274), and TGFβ/LAP and efficiently phagocytize apoptotic cells compared to Mo- MF. This efficient phagocytosis presumably occurs without overt inflammation, as LDLR- mediated cholesterol influx inhibits activation of the inflammasome (36). Our data show + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 434]]<|/det|> +that cancer generates phenotypically and functionally non- redundant TAM from BMBP and monocytes, where B- MF appear to promote cancer growth presumably by controlling antitumor T cell responses. First, unlike Mo- MF, B- MF efficiently suppress proliferation of T cells or induce the generation FoxP3+ Tregs in vitro. Second, B- MF enhance growth of B16- F10 melanoma in \(\mu \mathrm{M}\) mice. To do this, they primarily decrease \(\mathrm{CD4^{+}}\) T cells and inhibit \(\mathrm{IFN}\gamma^{+}\) \(\mathrm{CD4^{+}}\) T cells in the tumor, presumably by utilizing the B- MF- expressed immunoregulatory factors, \(\mathrm{TGF}\beta /\mathrm{LAP}\) and PD- L1 (22, 23), and LDLR. For example, LDLR may enhance the \(\mathrm{TGF}\beta\) responsiveness of target T cells by removing extracellular LDL/cholesterol that impairs \(\mathrm{TGF}\beta\) binding and thus signaling via \(\mathrm{TGF}\beta \mathrm{RII} / \mathrm{TGF}\beta \mathrm{R1}\) (37). + +<|ref|>text<|/ref|><|det|>[[142, 448, 857, 890]]<|/det|> +We propose that cancer primarily targets a small subset of \(\mathrm{CSF1R^{+}Pax5^{Low}}\) pre- B cells and iB cells recently emigrated from BM. First, B- MF are not found in tumor- bearing mice with the B- cell differentiation blockage at the pro- B cell stage. Second, splenic transitional B cells from naive mice do not generate B- MF regardless of their CSF1R expression state, as cancer first needs to mobilize BMBP into the circulation as the source of B- MF. We and others have reported that cancers use TSLP and G- CSF to mobilize BM pre- B cells and HSPS in the circulation (10, 38). We also find the chromatin accessibility landscape of BM CSF1R+ BMBP to be significantly more open and permissive to macrophage differentiation signals. In contrast, CSF1R+ and CSF1R- splenic B cells and BM CSF1R- cells present a chromatin landscape that is refractory to macrophage conversion. The overwhelming majority of regions with decreased accessibility in the BM CSF1R- cells remained closed in the spleen (679 of 749), presumably “locking in” the state of lymphoid lineage potential. It appears that BM CSF1R+ cells have a potentially more + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 85, 857, 600]]<|/det|> +primitive, permissive chromatin state allowing for myeloid lineage transformation. The BM CSF1R+ cells have more accessible ERG and RUNX1 binding sites, the two transcription factors expressed in myeloid and lymphoid progenitor cells (27) and early progenitor and myeloid- committed progenitor cells (30), respectively. Given that RUNX1 also upregulates expression of Csf1r by binding to its promoter and enhancer regions (29) and that the PAX5 deletion alone removes the repression of Csf1r and other myeloid lineage- specific genes and induces the BM B- cell precursor transdifferentiation (15, 26), we think that RUNX1 supports CSF1R expression in CSF1R+ BMBP to downregulate Pax5 in response tonic and cancer- secreted M- CSF. Give that the CSF1R+ BMBP in our patients with metastatic/recurrent triple- negative BC and high- grade serous OC co- express macrophage- associated genes, human cancer may also target biphenotypic B- cell precursors reported to have a macrophage- differentiation potential (16). Overall, for the first- time we report the B- cell- to- macrophage transdifferentiation as a physiological and widely utilized phenomenon. Murine and possibly human cancers target the transdifferentiation to generate immunosuppressive TAM. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[144, 90, 408, 108]]<|/det|> +## MATERIAL AND METHODS + +<|ref|>text<|/ref|><|det|>[[142, 123, 857, 675]]<|/det|> +Mice, cell lines: The animal protocol was approved by the ACUC committee of the National Institute on Aging (ASP 322- LMBI- 2022) under the Guide for the Care and Use of Laboratory Animals (NIH Publication No. 86- 23, 1985). The study used young (8- 12 weeks old) female mice bred and housed in the same, specific pathogen- free environment at the National Institute on Aging (NIA). C57BL/6J, BALB/CJ, R26R- EYFP (B6.129X1- Gt (ROSA)26Sortm1(EYFP)Cos/J), Csf1rflox mice (B6.Cg- Csf1rtm1.2Jwp/J) and μMT mice (B6.129- Ighm- tm1Cgn/J) and JHT mice (JHT; B6.129P2- Igh- Jtm1Cgn/J) in C57BL/6 background mice were purchased from the Jackson Laboratory (Bar Harbor, ME); μMT mice in BALB/c background were a gift from Dr. Thomas Blankenstein (Max- Delbrück- Center for Molecular Medicine, Berlin, Germany) (39). Mb1- Cre mice in C57BL/6 background (B6.C(Cg)- Cd79atm1(cre) Reth/EhobJ) were a gift from Dr. Richard Maraia (National Institute of Child Health and Human Development, Bethesda, MD) (40). Mogg- tag (Mogp) mice in C57BL/6 background were a gift from professor Dr. I. Miyoshi (Tohoku University Graduate School of Medicine, Miyagi, Japan) (12). To create mice with B- cell specific EYFP reporter (Mb1- EYFP) or CSF1R deletion (Mb1- CSF1RFlox/Flox), Mb1- cre mice were bred with R26R- EYFP and Csf1rflox mice, respectively. + +<|ref|>text<|/ref|><|det|>[[142, 687, 857, 882]]<|/det|> +4T1.2 cells- were a gift from Dr. Robin L. Anderson (Peter McCallum Cancer Center, Melbourne, Australia); MC38 colonic adenocarcinoma cells were a gift from Dr. Jeffrey Schlom (National Cancer Institute, Bethesda, MD) (41); mammary carcinoma AT3 cells were a gift from professor Scott I. Abrams (Roswell Park Comprehensive Cancer Center, Buffalo, NY); ID8- p53-/-- RFP (ID8) cells were gift from professor Sharon Stack (University of Notre Dame, IN); and EMT6 cells and melanoma B16- F10 cells were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 855, 178]]<|/det|> +purchased from American Type Culture Collection (Manassas, VA). Cells were tested free of mycoplasma with Mycoplasma Detection Kits (Lonza Basel, Switzerland; and IDEXX BioAnalytics, Columbia, MO). + +<|ref|>text<|/ref|><|det|>[[142, 193, 857, 632]]<|/det|> +Tissues and blood processing: Human samples were collected with written informed consent at the Clinical Core Laboratory, NIA, under Human Subject Protocol # 2003054 and Tissue Procurement Protocol # 2003- 071. Cancer patient PBMC were collected before the clinical trial (protocol # NCT02203513, NCI) from people with metastatic/recurrent triple- negative breast cancer and high- grade serous ovarian cancer (42, 43). The trial was approved by the Institutional Review Board of the Center for Cancer Research, National Cancer Institute, Bethesda, MD, USA. All experiments were performed on PBMC which were cryopreserved after collection. Mouse BM cells were flushed out of femurs and tibias with cold cRPMI. Single cell suspension of BM, spleen, LN were prepared with \(70\mu \mathrm{m}\) strainer (Falcon, Bedford, MA). BM, spleen, and blood cells were treated with ACK buffer to remove red blood cells. Mouse tumor tissues were cut to 3- 5 mm pieces and digested with mouse tumor dissociation kit (Miltenyi Biotec, Bergisch Gladbach, Germany) following the manufacturer's instruction. + +<|ref|>text<|/ref|><|det|>[[143, 645, 857, 844]]<|/det|> +Flow cytometry (FACS): For immune cell phenotyping, cells were pre- incubated with TruStain FcX™ solution before immunostaining with different combinations of anti- mouse or anti- human Abs (1μg per \(10^{6}\) cells, Suppl. Table 4) and fixable viability dye, then fixed/permeabilized with eBioscience™ intracellular fixation & permeabilization buffer (Thermo Fisher, Waltham, MA). The samples were evaluated on FACSymphony™ (BD, Franklin Lakes, NJ), Amnis ImageStreamX MKII (Millipore, Burlington, MA) or + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 853, 144]]<|/det|> +CytoFLEX (Beckman Coulter, Brea, CA). The results were analyzed with FlowJo v10(BD), IDEAS (Millipore) or Cytoexpert 2.3 (Beckman). + +<|ref|>text<|/ref|><|det|>[[142, 158, 857, 530]]<|/det|> +Cancer CM media preparation and cytokine quantification: Cells were cultured in RPMI1640 or DMEM (for ID8 cells) supplemented with \(10\%\) FBS, \(1 \times\) HEPES, sodium pyruvate, nonessential amino acids solution, penicillin- streptomycin- glutamine (Gibco, Gaithersburg, MD), and \(55 \mathrm{mmol / L}\) \(\beta\) - mercaptethanol in T75 flask to \(70 - 80\%\) confluency. CM was collected after 5 min centrifugation at \(1500 \mathrm{rpm}\) , filtered with \(0.2 \mu \mathrm{m}\) filter, and stored at \(- 80^{\circ} \mathrm{C}\) as single use aliquots. For cytokine tests, confluent cells were cultured with RPMI without FBS for \(24 \mathrm{~h}\) . Mouse serum was collected using BD Microtainer® Tubes following the manufacturer's instruction. Cytokines and M- CSF in filtered CM or sera were evaluated with Quantikine ELISA kit (R&D, Minneapolis, MN) or with Proteome Profiler Mouse XL Cytokine Array (R&D). Images were captured and analyzed with Fiji software. + +<|ref|>text<|/ref|><|det|>[[142, 543, 857, 879]]<|/det|> +In vitro B- MF conversion: BM Lin+ (TER119, CD11b, Gr- 1, CD3ε, NK1.1 or CD49b, Ly6C, Ly6G, CD11c)− CD19+ B cells were isolated from C57BL/CJ or BALB/CJ mice using FACSAriaTM Fusion sorter and \(10^{6} / \mathrm{ml}\) B cells were cultured in \(50\%\) cancer CM in cRPMI for 7 days in NuncTM Multidishes with UpCellTM Surface (Thermo Fisher) without changing media for 7 days. \(70z / 3\) pre- B cells \((10^{5} / \mathrm{ml})\) were cultured in \(50\%\) cancer CM for up to 30 days with replenishing culture medium every 3- 4 days. Adherent cells (macrophages) were harvested by detaching them at \(4^{\circ} \mathrm{C}\) for 15 mins in PBS. For Giemsa staining, B- MF were fixed with ethanol for 5 mins and Wright- Giemsa stained according to the manufacturer's instruction. CSF1R receptor signaling was blocked with Ki20227 (R&D). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 426]]<|/det|> +In vitro assays: For bacterial uptake assay, E. coli (Thermo Fisher) labeled with \(\mathrm{pHrodo^{TM}}\) red (0.1 mg/ml) were cultured with B- MF generated from RAG- GFP for 2h. Cells were washed with PBS, fixed with \(4\%\) formaldehyde and stained with DAPI. For phagocytosis of apoptotic cancer cells, ID8- RFP cells ( \(10^{6} / \mathrm{ml}\) ) were pretreated with 300 nM gemcitabine hydrochloride (Sigma, St. Louis, MO) for 24h, then washed with PBS and cultured with macrophages for 2h. Macrophages were stained with anti- F4/80- FITC Ab and DAPI and phagocytosis was evaluated using Zeiss LSM 710 (Carl Zeiss AG, Jena, Germany) and analyzed with Fiji software. For macrophage proliferation test, BrdU (10 \(\mu \mathrm{M}\) , BD) was added to macrophage cultures at day 5 and the BrdU incorporation was quantified at day 7 using FACSymphonyTM and analyzed by FlowJo. + +<|ref|>text<|/ref|><|det|>[[142, 438, 857, 781]]<|/det|> +T cell suppression assay was described elsewhere (13). Briefly, splenic T cells isolated with \(\mathrm{CD3^{+}}\) T cell enrichment column (R&D) were labeled with \(\mathrm{eFluor^{TM}}\) 450 and cultured with macrophages at 1:10, 1:20 and 1:40 E: T ratios in 96- well flat bottom plates coated with \(5\mu \mathrm{g / ml}\) anti- mouse CD3e antibody (clone 145- 2C11, BD) and free anti- mouse CD28 antibody ( \(2\mu \mathrm{g / ml}\) , clone 37.51, BD) for 4 days. The Treg conversion assay was described elsewhere (5). In brief, FACS- sorted splenic \(\mathrm{CD4^{+}CD25^{- }}\) T cells were cultured with macrophages at 1:5, 1:10 and 1:20 E:T ratios in plates coated with \(5\mu \mathrm{g / ml}\) anti- mouse CD3e antibody and free recombinant murine IL2 ( \(5\mathrm{ng / ml}\) , PeproTech, Rocky Hill, NJ) in for 5 days. Control T cells were cultured with recombinant mouse TGF- \(\beta 1\) ( \(5\mathrm{ng / ml}\) , R&D) in cRPMI without macrophages. + +<|ref|>text<|/ref|><|det|>[[143, 794, 856, 886]]<|/det|> +In vivo experiments: For evaluation of macrophages in vivo, tumor cells were subcutaneously injected into congenic mice, such as 4T1.2 cells and EMT6 cells ( \(1 \times 10^{6}\) ) in BALB/cJ and \(\mu \mathrm{MT}\) mice, and B16- F10, and AT3 and MC- 38 cells ( \(1 \times 10^{6}\) ) in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 391]]<|/det|> +C57BL/6J, JHT or Mb1- EYFP mice. ID8- p53-/-- RFP cells \((5 \times 10^{6})\) were intraperitoneally (i.p.) injected into C57BL/6J mice. For in vivo B- MF generation study, C57BL/6J mice with ID8- p53-/-- RFP cells \((5 \times 10^{6})\) in PeC were intraperitoneally injected with BM Lin- CD19+EYFP+ B cells \((5 \times 10^{6})\) sort- purified from Mb1- cre- EYFP mice and then 7 days later, the PeC lavage cells were FACS evaluated. To evaluate tumor- supporting role of B- MF in vivo, \(\mu\) MT mice were intravenously injected with \(3 \times 10^{5}\) in vitro- generated B- MF or PBS 3 and 7 days after subcutaneous challenge with B16- F10 melanoma cells (day 0, n=5). Tumor volume \((\mathrm{V} = \mathrm{a} \times \mathrm{b}, \mathrm{mm}^{2})\) was measured at days 11, 14, 16, 18, and 21, and at day 21, mice were euthanized to evaluate tumor weight and T cells. + +<|ref|>text<|/ref|><|det|>[[142, 404, 857, 565]]<|/det|> +Cellular cholesterol content quantification: Macrophages were fixed with \(4\%\) formaldehyde solution in TBS for 5 mins, then after TBS washes, they were incubated with Filipin III at 1:100 dilution in TBS (5mg/ml stock in \(100\%\) ethanol, Cayman, Ann Arbor, MI) for 60 mins in the dark. Cells were washed with TBS and lipids were quantified with Zeiss LSM 710 and Fiji software, as described above. + +<|ref|>text<|/ref|><|det|>[[142, 579, 857, 878]]<|/det|> +mRNA microarray: For the collected biological samples, the standard RNA extraction protocol was performed by RNeasy Plus Micro kits (QIAGEN, Hilden, Germany), and genome- wide expression was measured using the Agilent platform (Mouse 8X60K v2 and Hs 8X60K v3, Agilent, Santa Clara, CA, USA) according to the manufacture's instruction. Principal Component Analysis (PCA) was performed using the Procomp R function with expression values. Differential expressed genes (DEGs) was assessed using the moderated (empirical Bayesian) t- test implemented in the limma package (version 3.14.4)(44), and correction for multiple hypothesis testing was accomplished by calculating the Benjamini- Hochberg false discovery rate. Enriched + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 855, 178]]<|/det|> +pathways were discovered by GSEA tool (45) with Molecular Signature Database v7.4. All microarray analyses were performed using the R environment for statistical computing (version 3.6.2). + +<|ref|>text<|/ref|><|det|>[[142, 195, 857, 880]]<|/det|> +scRNA- seq: Sort purified single cell suspensions were loaded into a \(10\times\) Chromium controller (10x Genomics, Pleasanton, CA, USA) and converted to barcoded single cell RNA expression library according to the standard protocol of the Chromium Next GEM Single cell 3' kit (v3.1 chemistry) in Laboratory of Immunology and Molecular Biology, National Institute on Aging, and the single cell 3' gene expression libraries were sequenced on NovaSeq 6000 (Illumina, San Diego, CA, USA) in the Genomics Core facility of the Johns Hopkins School of Medicine. Raw sequencing data was processed using the Cell Ranger version 5.0 (10x Genomics, Pleasanton, CA, USA) pipeline. The raw gene expression matrix was normalized and scaled using the SCTransform method (46) in the Seurat R package (version 4.0) (47). The minimum number of detected genes was set to 1,000, and genes were chosen when they were detected in more than 3 cells. Dimension reduction was performed using principal component analysis (PCA). For visualizing the generated clustering, we used the Uniform Manifold Approximation and Projections (UMAP) plot. We defined clusters with a leiden algorithm using shared nearest neighbor (SNN) in PCA space. From in vitro B- MF and Mo- MF, we generated a total of 12 clusters for in vitro samples (0- 11). Integration of in vivo samples with canonical correlation analysis (CCA) was performed, and we generated 13 clusters for in vivo tumor macrophage samples (0- 12). Finally, we performed a nonparametric Wilcoxon rank- sum test to search for highly expressed genes in the clusters. In addition, human tumor single cell transcriptomes were downloaded from GEO (GSE114725, and GSE146026) and also + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 855, 178]]<|/det|> +processed with the same pipeline described. We used only macrophage clusters for downstream analysis. All single- cell analyses were performed using the R environment for statistical computing (version 4.0.5). + +<|ref|>text<|/ref|><|det|>[[142, 193, 857, 667]]<|/det|> +ATAC- seq analysis: We utilized a Hi- Seq 2000 machine to sequence the ATAC- seq libraries. We prepared 12 pair- end ATAC- seq libraries including BM CSF1R (+/- ) and Spleen CSF1R (+/- ) samples (n=3 per group). In total, 369M reads were sequenced and average 31M reads were sequenced per sample. We applied NIEHS TaRGETII ATAC- seq pipelines, which are available to the genomics community. All raw reads were trimmed using cutadapt package, and trimmed reads (>36bp minimum alignment length) were mapped against the mm10 reference genome using BWA aligner (48). We used de- duplicated and uniquely mapped reads for peak calling analysis after excluding black- list regions defined by ENCODE (49). The candidate peaks were predicted by MACS peak calling tool (50). In addition, we also applied the DESeq2 (51) to determine differentially accessible regions (DARs); cutoff: Fold change \(>1.5\) , \(\log 2\mathrm{CPM} > 1.2\) , FDR \(< 0.05\) . The differentially accessible regions were submitted for search of potential transcription factor binding sites using HOMER software (52). We used non DARs as background regions in de novo motif analysis. + +<|ref|>text<|/ref|><|det|>[[142, 681, 856, 840]]<|/det|> +Statistical analysis: The results are presented as the mean with each individual data point or in bar graph \(\pm\) SEM. GraphPad Prism (Prism 6; GraphPad Software, Inc) was used to perform statistical analysis. Data were analyzed using Welch t- test or one- way ANOVA. A P- value less than 0.05 was considered significant (**** P<0.0001, ***P<0.001, ** P<0.01, *P<0.05). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 840, 250]]<|/det|> +ACKNOWLEDGMENTS: We are grateful to Mrs. Jane Trepel with help with human BC and OC samples; Dr. Xin Wang (NIA) for help with confocal microscopy, Drs. Elin Lehrman, Yongqing Zhang, and Supriyo De (Genomics Core, LGG, NIA) for help with performing microarray experiments and initial data analyses, and data submission to GSE; and Dr. Chandamany Arya (Lilly) for proofreading. + +<|ref|>text<|/ref|><|det|>[[142, 297, 833, 423]]<|/det|> +AUTHOR CONTRIBUTIONS: C. C. performed the research, collected and analyzed data; E. Ra., M.B. and L.Z., performed experiments; B.P. worked on bioinformatics analyses, JMM provided clinical trial samples; A.B. and I.B. supervised the study and wrote the manuscript; and A.B. conceived and designed the study. + +<|ref|>text<|/ref|><|det|>[[143, 477, 815, 533]]<|/det|> +FUNDING: This research was supported by the Intramural Research Program of the National Institute on Aging, NIH. + +<|ref|>text<|/ref|><|det|>[[143, 588, 852, 677]]<|/det|> +COMPETING INTERESTS: The authors are employees of National Institutes of Health, the U.S. government, and declare no competing interests, and there are no patents or patent applications related to this work. + +<|ref|>text<|/ref|><|det|>[[143, 727, 844, 817]]<|/det|> +DATA AND MATERIALS AVAILABILITY: The data associated with this study can be found in the paper or supplementary materials and ATAC- seq data are deposited at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc= GSE178716 and =GSE180285 + +<|ref|>text<|/ref|><|det|>[[143, 867, 680, 887]]<|/det|> +SUPLEMENTAL MATERIALS: Figures S1- S8 and Tables S1- 4. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 140, 850, 875]]<|/det|> +1. S. N. Ganti, T. C. Albershardt, B. M. Iritani, A. Ruddell, Regulatory B cells preferentially accumulate in tumor-draining lymph nodes and promote tumor growth. Sci Rep 5, 12255 (2015).2. C. Lee-Chang et al., Aging Converts Innate B1a Cells into Potent CD8+ T Cell Inducers. J Immunol 196, 3385-3397 (2016).3. Y. Gu et al., Tumor-educated B cells selectively promote breast cancer lymph node metastasis by HSPA4-targeting IgG. Nat Med 25, 312-322 (2019).4. Q. Li et al., Adoptive transfer of tumor reactive B cells confers host T-cell immunity and tumor regression. Clin Cancer Res 17, 4987-4995 (2011).5. P. B. Olkhanud et al., Tumor-evoked regulatory B cells promote breast cancer metastasis by converting resting CD4+ T cells to T-regulatory cells. Cancer Res 71, 3505-3515 (2011).6. P. B. Olkhanud et al., Thymic stromal lymphopoietin is a key mediator of breast cancer progression. Journal of immunology 186, 5656-5662 (2011).7. K. Wejksza et al., Cancer-produced metabolites of 5-lipoxygenase induce tumor-evoked regulatory B cells via peroxisome proliferator-activated receptor alpha. Journal of immunology 190, 2575-2584 (2013).8. S. Kusmartsev, D. I. Gabrilovich, Effect of tumor-derived cytokines and growth factors on differentiation and immune suppressive features of myeloid cells in cancer. Cancer Metastasis Rev 25, 323-331 (2006).9. K. Movahedi et al., Different tumor microenvironments contain functionally distinct subsets of macrophages derived from Ly6C(high) monocytes. Cancer Res 70, 5728-5739 (2010).10. E. Ragonnaud et al., Tumor-Derived Thymic Stromal Lymphopoietin Expands Bone Marrow B-cell Precursors in Circulation to Support Metastasis. Cancer Res 79, 5826-5838 (2019).11. M. Lelekakis et al., A novel orthotopic model of breast cancer metastasis to bone. Clin.Exp.Metastasis 17, 163-170 (1999).12. M. Bodogai et al., Immunosuppressive and Prometastatic Functions of Myeloid-Derived Suppressive Cells Rely upon Education from Tumor-Associated B Cells. Cancer Res 75, 3456-3465 (2015).13. P. B. Olkhanud et al., Breast cancer lung metastasis requires expression of chemokine receptor CCR4 and regulatory T cells. Cancer Res. 69, 5996-6004 (2009).14. H. Xie, M. Ye, R. Feng, T. Graf, Stepwise reprogramming of B cells into macrophages. Cell 117, 663-676 (2004).15. S. L. Nutt, B. Heavey, A. G. Rolink, M. Busslinger, Commitment to the B-lymphoid lineage depends on the transcription factor Pax5. Nature 401, 556-562 (1999).16. T. Audzevich et al., Pre/pro-B cells generate macrophage populations during homeostasis and inflammation. Proc Natl Acad Sci U S A 114, E3954-E3963 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 88, 850, 895]]<|/det|> +17. E. R. Stanley, M. Cifone, P. M. Heard, V. Defendi, Factors regulating macrophage production and growth: identity of colony-stimulating factor and macrophage growth factor. J Exp Med 143, 631-647 (1976).18. S. R. Almeida et al., Mouse B-1 cell-derived mononuclear phagocyte, a novel cellular component of acute non-specific inflammatory exudate. Int Immunol 13, 1193-1201 (2001).19. D. Kitamura, J. Roes, R. Kuhn, K. Rajewsky, A B cell-deficient mouse by targeted disruption of the membrane exon of the immunoglobulin mu chain gene. Nature 350, 423-426 (1991).20. J. Chen et al., Immunoglobulin gene rearrangement in B cell deficient mice generated by targeted deletion of the JH locus. Int Immunol 5, 647-656 (1993).21. E. Hobeika et al., Testing gene function early in the B cell lineage in mb1-cre mice. Proc Natl Acad Sci U S A 103, 13789-13794 (2006).22. W. Chen, P. Ten Dijke, Immunoregulation by members of the TGFbeta superfamily. Nat Rev Immunol 16, 723-740 (2016).23. L. M. Francisco, P. T. Sage, A. H. Sharpe, The PD-1 pathway in tolerance and autoimmunity. Immunol Rev 236, 219-242 (2010).24. G. Gimpl, Cholesterol-protein interaction: methods and cholesterol reporter molecules. Subcell Biochem 51, 1-45 (2010).25. H. Ohno et al., A c-fms tyrosine kinase inhibitor, Ki20227, suppresses osteoclast differentiation and osteolytic bone destruction in a bone metastasis model. Mol Cancer Ther 5, 2634-2643 (2006).26. A. Delogu et al., Gene repression by Pax5 in B cells is essential for blood cell homeostasis and is reversed in plasma cells. Immunity 24, 269-281 (2006).27. A. P. Ng et al., An Erg-driven transcriptional program controls B cell lymphopoiesis. Nat Commun 11, 3013 (2020).28. S. Taoudi et al., ERG dependence distinguishes developmental control of hematopoietic stem cell maintenance from hematopoietic specification. Genes Dev 25, 251-262 (2011).29. S. R. Himes, S. Cronau, C. Mulford, D. A. Hume, The Runx1 transcription factor controls CSF-1-dependent and -independent growth and survival of macrophages. Oncogene 24, 5278-5286 (2005).30. M. R. Corces et al., Lineage-specific and single-cell chromatin accessibility charts human hematopoiesis and leukemia evolution. Nat Genet 48, 1193-1203 (2016).31. E. Azizi et al., Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment. Cell 174, 1293-1308 e1236 (2018).32. B. Izar et al., A single-cell landscape of high-grade serous ovarian cancer. Nat Med 26, 1271-1279 (2020).33. T. A. Wynn, A. Chawla, J. W. Pollard, Macrophage biology in development, homeostasis and disease. Nature 496, 445-455 (2013).34. M. Donadon et al., Macrophage morphology correlates with single-cell diversity and prognosis in colorectal liver metastasis. J Exp Med 217, (2020).35. H. Jeon, S. C. Blacklow, Structure and physiologic function of the low-density lipoprotein receptor. Annu Rev Biochem 74, 535-562 (2005).36. J. H. Madenspacher et al., Cholesterol 25-hydroxylase promotes efferocytosis and resolution of lung inflammation. JCI Insight 5, (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[140, 90, 840, 130]]<|/det|> +37. C. L. Chen, S. S. Huang, J. S. Huang, Cholesterol modulates cellular TGF-beta responsiveness by altering TGF-beta binding to TGF-beta receptors. J Cell Physiol 215, 223-233 (2008). + +<|ref|>text<|/ref|><|det|>[[140, 112, 840, 826]]<|/det|> +38. F. Shojaei et al., G-CSF-initiated myeloid cell mobilization and angiogenesis mediate tumor refractoriness to anti-VEGF therapy in mouse models. Proc Natl Acad Sci U S A 106, 6742-6747 (2009). +39. Z. Qin et al., B cells inhibit induction of T cell-dependent tumor immunity. Nature Medicine 4, 627-630 (1998). +40. S. Gaidamakov et al., Targeted deletion of the gene encoding the La autoantigen (Sjogren's syndrome antigen B) in B cells or the frontal brain causes extensive tissue loss. Mol Cell Biol 34, 123-131 (2014). +41. P. H. Hand, P. F. Robbins, M. L. Salgaller, D. J. Poole, J. Schlom, Evaluation of human carcinoembryonic-antigen (CEA)-transduced and non-transduced murine tumors as potential targets for anti-CEA therapies. Cancer Immunol Immunother 36, 65-75 (1993). +42. M. E. Gatti-Mays et al., A Phase II Single Arm Pilot Study of the CHK1 Inhibitor Prexasertib (LY2606368) in BRCA Wild-Type, Advanced Triple-Negative Breast Cancer. Oncologist 25, 1013-e1824 (2020). +43. J. M. Lee et al., Prexasertib, a cell cycle checkpoint kinase 1 and 2 inhibitor, in BRCA wild-type recurrent high-grade serous ovarian cancer: a first-in-class proof-of-concept phase 2 study. Lancet Oncol 19, 207-215 (2018). +44. M. E. Ritchie et al., limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43, e47 (2015). +45. A. Subramanian et al., Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545-15550 (2005). +46. C. Hafemeister, R. Satija, Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol 20, 296 (2019). +47. Y. Hao et al., Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587 e3529 (2021). +48. H. Li, R. Durbin, Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics 26, 589-595 (2010). +49. H. M. Amemiya, A. Kundaje, A. P. Boyle, The ENCODE Blacklist: Identification of Problematic Regions of the Genome. Sci Rep 9, 9354 (2019). +50. Y. Zhang et al., Model-based analysis of ChIP-Seq (MACS). Genome Biol 9, R137 (2008). +51. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014). +52. S. Heinz et al., Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38, 576-589 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 80, 857, 780]]<|/det|> +Figure 1. Cancer induces differentiation of macrophages from BM B cells. A and B, FACS staining frequency \(\pm\) SEM of co-expression for markers of B cells (CD19) and macrophages (F4/80 \(^+\) ) in A, BALB/CJ and \(\mu \mathrm{MT}\) mice with orthotopic 4T1.2 breast cancer (n=5- 6) and B, C57BL/6 and \(\mathrm{J_{H}T}\) mice with spontaneous ovarian cancer (Mogp- tag and Mogp- \(\mathrm{J_{H}T}\) , respectively, n=4) Frequency of F4/80 \(^+\) CD11b \(^+\) cells within CD19 \(^+\) TIBs (left panels) and CD19 \(^+\) IgM \(^+\) cells within F4/80 \(^+\) TAM (right panels). C, Representative flow plot for F4/80 \(^+\) CD11b \(^+\) macrophage of B cells isolated from BM, spleen, LN and PeC treated with CM from 4T1.2 breast cancer cells. D, Representative Giemsa staining in vitro- generated B- MF. Scale bar represented \(20\mu \mathrm{m}\) . E, Representative FACs plots of F4/80 \(^+\) CD11b \(^+\) macrophage of BM B- cells treated with CM from 4T1.2, EMT6, AT3, B16- F10, and MC- 38 cells. F and G Representative FACs plots for CD68 and CD11b staining of pre- B cell line (70z/3) after 30- day stimulation with (RPMI) or without 4T1.2- CM and composite frequency analysis of F4/80 \(^+\) CD11b \(^+\) and CD68 \(^+\) CD11b \(^+\) cells from four independent experiments. H, Representative Imagestream imaging for surface markers F4/80, CD11b, CD19, and CD20 on in vitro- generated B- MF from naive Mb1- cre- EYFP \(^+\) mice (bottom two panels, H) and WT mice (top two panels, H). I, Frequency \(\pm\) SEM of CD274 \(^+\) and LAP \(^+\) cells within EYFP \(^+\) or EYFP \(^+\) F4/80 \(^+\) CD11b \(^+\) cells quantified 7- day after i.p. injection of BM B cells from naive Mb1- cre- EYFP mice in C57BL/6 mice with ID8 ovarian cancer in PeC (n=3,). \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +<|ref|>text<|/ref|><|det|>[[143, 805, 856, 898]]<|/det|> +Figure 2. Distinct gene expression profiles of B- MF and Mo- MF. A, Representative FACs plots of F4/80 \(^+\) CD11b \(^+\) macrophage of BM B cells or monocytes post 7- day treatment with 4T1.2- CM. Macrophage phenotype gated as F4/80 \(^+\) CD11B \(^+\) from B cell + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 87, 857, 530]]<|/det|> +(B- MF) and monocytes (Mo- MF) B, Representative expression of B- cell markers CD79a and IgM by FACs analysis on B- MF (Red) and Mo- MF (Orange). C, PCA plot of mRNA expression profiles generated from microarray data of sort purified B- MF (Blue), Mo- MF (Orange) and BM B cells (Green) (n=3). D and E, Bar plots of GSEA predicted pathways enriched in B- MF (D) or Mo- MF cells (E) from the Molecular Signature Database. F and G, UMAP plot of B- MFs (10,563 cells) and Mo- MFs (10,235 cells) analyzed using Seurat with colors depicting clusters by cell type (F) or by gene expression (G). H, Heatmap of top genes with differential expression characterizing the 12 major cell types of in vitro B- MF and Mo- MF single cell clusters. I, Overlay of Mrc1 expression for in vitro B- MF and Mo- MF single cells. J, UMAP plot of sort purified TAMs (10,885 cells) generating 13 unique cell clusters from tumor tissues of 4 mice with 4T1.2 cancers. K, Violin plots of the expression of three in vitro derived B- MF up- regulated DEGs (Egr1, Ier3 and Slc40a1) in in vivo mouse 4T1.2 TAM single cell clusters. + +<|ref|>text<|/ref|><|det|>[[142, 575, 857, 878]]<|/det|> +Figure 3. B- MF and Mo- MF are functionally different. A, Ki67 staining (upper) and BrdU uptake (lower panel) frequency of B- MF and Mo- MF (frequency±SEM, n=2). B and C, Representative Filipin III staining images and quantification (MFI). D and E, Representative images and quantifications of \(\mathrm{RFP}^+\) apoptotic cells in B- MF and Mo- MF cultured with apoptotic ID8- RFP ovarian cancer cells for 2 h. Eight representative fields per sample were quantified and scale bars represent \(20 \mu \mathrm{m}\) (C and E). F, Representative histograms of proliferated \(\mathrm{CD4}^+\) T cells after co- culturing with macrophages at 1:10 (E:T) ratio in the presence of anti- CD3/CD28 Abs for 4 days. Control T cells were cultured alone with (activated) or without (non- activated) anti- CD3/CD28 Abs. Numbers represent + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[142, 88, 857, 358]]<|/det|> +percentage of proliferated T cells. Macrophages were generated from BM B- cell precursors (B- MF) and monocytes (Mo- MF) with 4T1.2- CM treatment. The results in \(\mathbf{B} - \mathbf{F}\) were independently reproduced at least three times. \(\mathbf{G} - \mathbf{J}\) , B- MF support tumor growth in vivo. \(\mu \mathrm{MT}\) mice were i.v. transferred with \(3\times 10^{5}\) in vitro- generated B- MF or PBS 3 and 7 days after s.c. challenge with B16- F10 melanoma cells (n=5). Each symbol is for a single mouse; Y- axis is for \(\% \pm \mathrm{SEM}\) of tumor- infiltrated \(\mathrm{CD4^{+}}\) T cells in \(\mathrm{CD45^{+}}\) cells (H) and \(\mathrm{IFN}\gamma^{+}\) in \(\mathrm{CD4^{+}}\) T cells (I) and tumor weight \(\pm \mathrm{SEM}\) (J). \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**}\) \(\mathrm{P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +<|ref|>text<|/ref|><|det|>[[141, 384, 857, 899]]<|/det|> +Figure 4. CSF1/CSF1R signaling axis is critical for B- MF differentiation. A, Numbers of \(\mathrm{CD19^{+}}\) (left) and \(\mathrm{CD93^{+}CD19^{+}}\) (middle) B cells and frequency of \(\mathrm{CSF1R^{+}}\) within \(\mathrm{CD93^{+}CD19^{+}}\) B cells (right) in the spleen of naive and 4T1.2 tumor- bearing mice (n=5). B, ELISA measurements of secreted M- CSF in media conditioned by 4T1.2, EMT6, AT3, MC- 38, B16- F10, and ID8 cells (n=6, pg/ml). C, Representative FACS plots of the \(\mathrm{F4 / 80^{+}CD11b^{+}}\) B- MF converted from sort- purified BM CSF1R+ and CSF1R- B- cell precursors and splenic CSF1R+ and CSF1R- B cells after 7- day treatment with 4T1.2- CM. D and E, Representative FACS plots of \(\mathrm{F4 / 80^{+}CD11b^{+}}\) (B- MF) from BM B- cell precursors after 7- day treatment with 4T1.2- CM (D) and pre- B cell line 70z/3 after 30- day treatment with 4T1.2- CM (E) with and without anti- MCSF Ab or Ki- 20227 treatments (a specific MCSF- R inhibitor). F and G, Representative FACS plot (F) and quantification (G) of the generation of B- MF after 7- day 4T1.2- CM treatment on cells isolated from WT or mice with conditional CSF1R deficiency in B cells (Mb1- CSF1RFlox/Flox) (n=6). Results for all panels were independently reproduced minimally three times. \(\mathrm{***P< 0.0001}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{**P< 0.01}\) , \(\mathrm{*P< 0.05}\) between indicated comparisons. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[141, 120, 861, 740]]<|/det|> +Figure 5. Cancer targets BM CSF1R+PAX5Low B-cell precursors. A, FACs analysis of Pax5 expression of freshly isolated BM CSF1R+ or CSF1R+ B cell precursors (n=5,) B, FACs analysis of Pax5 expression of pre- B cell (70z/3 cells) treated with indicated cancer CM (n=3). C, FACs analysis of Pax5 expression of BM Lin- CD19+B220+CD93+IgM- IgD- CSF1R+ B cell precursors treated with M- CSF for 48h (n=3). D, 3D PCA plot of chromatin accessibility generated from ATAC- seq data of BM CSF1R+ and CSF1R- B- cell precursors and spleen CSF1R+ and CSF1R- - follicular B cells E, Heatmap of differentially accessible regions (DARs) between BM CSF1R+ and CSF1R- B- cell precursors (FDR<0.05). F, Top, highly significant de novo motifs are more open in BM CSF1R+ compared to BM CSF1R- as well as Spleen CSF1R+ and CSF1R- - cells (678 sites). G, mRNA microarray heatmap of macrophage related- DEGs in B cells isolated from PB of patients with breast cancer (BC, n=8) compared to healthy donors (HD, n=7). Scale bar is for expression z- score. H, Cell frequency of CSF1R+ (left), CD68+ (middle) and LDLR+ (right) cells within CD19+CD10+ B cells of PB from healthy donors (HD) or patients with ovarian cancer (OC) (n=5-7). I and J, UMAP of scRNA sequencing data of macrophage cells (left) and expression levels of in vitro B-MF enriched genes (right) in published human BC (I) and OC (J) datasets. Error bars represent SEM, **** P<0.0001, ***P<0.001, ** P<0.01, *P<0.05 between indicated comparisons. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[63, 40, 905, 920]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 35, 930, 936]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 45, 930, 920]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 50, 914, 867]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 28, 960, 920]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[848, 907, 904, 920]]<|/det|> +
Figure 5
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 131, 336, 178]]<|/det|> +SupplementaryTables14. xlsx SupplFigsSl.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/images_list.json b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..762839dca315a1d8031cd54e7e8b78681df3c671 --- /dev/null +++ b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Key organisms in the Southern Ocean and \\(21^{\\mathrm{st}}\\) -century ocean acidification in Marine Protected Areas. Map of established (South Orkney Islands Southern Shelf and Ross Sea region) and proposed (Weddell Sea, East Antarctic, Western Antarctic Peninsula) Marine Protected Areas (MPAs). Key species and groups which may be negatively impacted by ocean acidification (see references in main text) are depicted around the map (not to scale). Bar plots denote the top 2000 m average pH in the five different MPAs in the 1990s (dark grey) and in the 2090s (colors) for the four emission scenarios (sorted from low emission to high emission). For the Weddell Sea, East Antarctic, and Ross Sea MPAs, the hatched bars denote the average pH for the continental shelf south of the 2000 m isobath (grey contour on the map). Using this definition, the majority of the area of the proposed MPA along the western Antarctic Peninsula is continental shelf (see Methods), and bars are therefore hatched for this region. Note the arrows for the SSP5-8.5 scenario, indicating a lower pH on the continental shelves than for the whole-MPA average.", + "footnote": [], + "bbox": [ + [ + 70, + 73, + 925, + 551 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Distribution of anthropogenic carbon across the high-latitude Southern Ocean. Distribution of anthropogenic carbon concentrations in \\(\\mu \\mathrm{mol} \\mathrm{kg}^{-1}\\) a averaged over the top \\(2000 \\mathrm{m}\\) and b at the bottom in the 1990s of the historical simulation. c-d Same as a-b, but for the 2090s of the high-emission scenario SSP5-8.5. The top \\(2000 \\mathrm{m}\\) are equivalent to the whole water column on the continental shelves, defined as the area south of the \\(2000 \\mathrm{m}\\) isobath (grey contour). Note the logarithmic colorbar and the one order of magnitude difference between the 1990s and the 2090s. Outlines of the Marine Protected Areas (MPAs) are denoted in white, and the ice-shelf front is shown as the dark grey contour. e-g Temporal evolution of anthropogenic carbon concentrations with depth and averaged over the continental shelves of e the proposed Weddell Sea MPA, f the proposed East Antarctic MPA, and g the adopted Ross Sea region MPA. i-k Same as e-g, but for the open ocean. h and l show the averages for h the proposed MPA along the western Antarctic Peninsula and l the adopted South Orkney Islands Southern Shelf MPA.", + "footnote": [], + "bbox": [ + [ + 70, + 234, + 928, + 576 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Changes in the vertical distribution of \\(\\Omega_{\\mathrm{arg}}\\) . Vertical profiles of a pH and b the saturation state with respect to aragonite \\(\\left(\\Omega_{\\mathrm{arg}}\\right)\\) in the proposed Weddell Sea Marine Protected Area (MPA). Solid lines show the profiles above 2000 m in the whole MPA, and dashed lines show the profile for the continental shelves south of the 2000 m isobath (see dark grey contour in map). The black lines denote the 1990s in the historical simulation, and the colored lines the 2090s in the four emission scenarios. The profiles on the left in each panel show absolute values, and profiles on the right show the change in each property between the 2090s and the 1990s \\(\\left(\\Delta \\mathrm{pH}\\right.\\) and \\(\\Delta \\Omega_{\\mathrm{arg}}\\) ). In b, undersaturated conditions, i.e., \\(\\Omega_{\\mathrm{arg}}< 1\\) , are highlighted with the colored background, and the depth of \\(\\Omega_{\\mathrm{arg}} = 1\\) for the profile of the whole MPA is highlighted with a circle on the y axis. c-j Same as a-b, but for c-d the proposed East Antarctic MPA, e-f the adopted Ross Sea region MPA, g-h the proposed Antarctic Peninsula MPA, i-j the adopted South Orkney Islands Southern Shelf MPA. Note that only the vertical profiles for the whole MPA are shown in panels g-j (see Methods).", + "footnote": [], + "bbox": [ + [ + 170, + 75, + 844, + 720 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Changes in the volume distribution across different \\(\\Omega_{\\mathrm{arg}}\\) and \\(\\Omega_{\\mathrm{calc}}\\) classes. a-b Distribution of waters across different classes of the saturation state with respect to aragonite \\((\\Omega_{\\mathrm{arg}})\\) a above \\(2000\\mathrm{m}\\) in the whole proposed Weddell Sea Marine Protected Area (MPA) and b on the continental shelves south of the \\(2000\\mathrm{m}\\) isobath (see dark grey contour in map). Results are shown for the 1990s of the historical simulation and for the 2090s in the four emission scenarios. c-d Same as a-b, but for the distribution of waters across different classes of the saturation state with respect to calcite \\((\\Omega_{\\mathrm{calc}})\\) . Note the different classes of saturation states shown for aragonite and calcite. While the 0.65 threshold for \\(\\Omega_{\\mathrm{arg}}\\) was chosen to best highlight the stronger undersaturation on the shelf than in the whole MPA for the highest emission scenario, we assume that it is more costly for an organism to sustain calcification at \\(\\Omega_{\\mathrm{arg}}< 0.65\\) than at \\(0.65\\leq \\Omega_{\\mathrm{arg}}< 1.0\\) (ref. \\(^{88}\\) ) and that this division of the volume of undersaturated waters is therefore meaningful. e-l Same as a-d, but for e-h the proposed East Antarctic MPA and i-l the adopted Ross Sea region MPA. m-p Same as a and c, but for m-n the proposed Antarctic Peninsula MPA and o-p the adopted South Orkney Islands Southern Shelf MPA. Note that only the distribution for the whole MPA is shown in panels m-p (see Methods).", + "footnote": [], + "bbox": [ + [ + 87, + 250, + 927, + 510 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5: Temporal evolution of \\(\\Omega_{\\mathrm{arg}}\\) and sea ice cover. a Temporal evolution of the depth of the \\(\\Omega_{\\mathrm{arg}} = 1\\) isoline in the proposed Weddell Sea Marine Protected Area (MPA) for the SSP5-8.5 scenario (dark blue), the SSP3-7.0 scenario (intermediate blue), and the SSP2-4.5 scenario (light blue). The different line styles represent the annual mean (solid), the June-August mean (JJA; dotted), and the December-March mean (DJF; dashed). The SSP1-2.6 is not shown because the \\(\\Omega_{\\mathrm{arg}} = 1\\) isoline never reaches the top \\(80 \\mathrm{~m}\\) of the water column. b-e Same as a, but for b the proposed East Antarctic MPA, c the adopted Ross Sea region MPA, d the proposed Antarctic Peninsula MPA, and e the adopted South Orkney Islands Southern Shelf MPA. f-j Temporal evolution of the \\(50\\%\\) (thick line) and \\(90\\%\\) (thin line) isolines of the volume with \\(\\Omega_{\\mathrm{arg}} < 1\\) in the upper \\(50 \\mathrm{~m}\\) of the water column in the different MPAs. Colors show the different emission scenarios (see above). k-o Monthly ice-free surface area in the different MPAs for the 1990s (black) and the 2090s in the four emission scenarios (colors).", + "footnote": [], + "bbox": [ + [ + 125, + 93, + 884, + 720 + ] + ], + "page_idx": 26 + } +] \ No newline at end of file diff --git a/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8.mmd b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8.mmd new file mode 100644 index 0000000000000000000000000000000000000000..638c5a21dded77bf340d501dd3c73334d8e00b9a --- /dev/null +++ b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8.mmd @@ -0,0 +1,365 @@ + +# Severe 21st-century ocean acidification demands continuance and expansion of Antarctic Marine Protected Areas + +Cara Nissen (cara.nissen@colorado.edu) University of Colorado, Boulder + +Nicole Lovenduski University of Colorado Boulder https://orcid.org/0000- 0001- 5893- 1009 + +Cassandra Brooks University of Colorado + +Mario Hoppema Alfred Wegener Institute https://orcid.org/0000- 0002- 2326- 619X + +Ralph Timmermann Alfred Wegener Institute for Polar and Marine Research + +Judith Hauck Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research https://orcid.org/0000- 0003- 4723- 9652 + +## Article + +Keywords: + +Posted Date: June 5th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2758143/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on January 4th, 2024. See the published version at https://doi.org/10.1038/s41467- 023- 44438- x. + +<--- Page Split ---> + +# Severe \(21^{\mathrm{st}}\) -century ocean acidification demands continuance and expansion of Antarctic Marine Protected Areas + +Cara Nissen \(^{1,2*}\) , Nicole S. Lovenduski \(^{1}\) , Cassandra M. Brooks \(^{3}\) , Mario Hoppema \(^{2}\) , Ralph Timmermann \(^{2}\) , Judith Hauck \(^{2}\) + +\(^{1}\) Department of Atmospheric and Oceanic Sciences and Institute of Arctic and Alpine Research, University of Colorado Boulder, Boulder, CO, USA. \(^{2}\) Alfred Wegener Institut, Helmholtz- Zentrum für Polar- und Meeresforschung, Bremerhaven, Germany. \(^{3}\) Department of Environmental Studies, University of Colorado Boulder, Boulder, CO, USA. \* corresponding author: cara.nissen@colorado.edu + +1 Antarctic coastal waters are home to several established or proposed Marine Protected Areas (MPAs) supporting exceptional biodiversity, which is threatened by anthropogenic climate change. Despite a particular sensitivity to ocean acidification (OA), little is known about the future carbonate chemistry of high- latitude Southern Ocean waters. Here, we use a high- resolution ocean- sea ice- biogeochemistry model with realistic ice- shelf geometry to investigate \(21^{\mathrm{st}}\) - century OA in Antarctic MPAs under four emission scenarios. By 2100, we project surface pH declines of up to 0.42 (total scale), corresponding to a \(161\%\) increase in hydrogen ion concentration relative to the 1990s. End- of- century aragonite undersaturation is ubiquitous across MPAs under the three highest emission scenarios. Vigorous vertical mixing of anthropogenic carbon on the continental shelves produces severe OA within the Weddell Sea, East Antarctic, and Ross Sea MPAs. Our findings call for continuity and expansion of Antarctic MPAs to reduce pressures on ecosystem integrity. + +Ocean acidification (OA) is an environmental problem \(^{1}\) threatening marine biodiversity across the global oceans \(^{2 - 4}\) . OA is caused by the invasion of anthropogenic carbon (Canth) into the ocean, which results in a decline in pH and in the saturation states of the calcium carbonate minerals aragonite (Ωarag) and calcite (Ωcalc). The Southern Ocean is especially sensitive to OA due to the prevalence of high CO₂ concentrations in cold waters and the upwelling of carbon-rich deep waters with low pH values \(^{1,2}\) . Earth system model projections suggest that surface waters in the open Southern Ocean might become seasonally undersaturated with respect to the more soluble mineral aragonite (Ωarag <1) by 2035 \(^{2,5 - 7}\) . Projections of OA on the Antarctic continental shelves have so far been elusive due to the inability of coarse-resolution models to realistically resolve physical and biogeochemical processes at high resolution and in ice-shelf cavities. Ob + +<--- Page Split ---> + +posed to both OA and warming55. Juvenile Antarctic emerald rockcod and Black rockcod show reduced metabolic capacity to adapt to warming when exposed to \(\mathrm{pH}< 7.75^{56,57}\) . Benthic bivalve and sea urchin larvae are increasingly malformed and damaged a \(\mathrm{pH}< 7.8\) , possibly compromising their survival4,58. Altogether, acknowledging knowledge gaps on direct OA impacts for some keystone species (e.g., Antarctic silverfish and upper-trophic level organisms), even small impacts of OA on individual species can accumulate through the food web, as shifts in development can cause predator- prey mismatches49,54,55. While the overall number of available studies is low and inter- species variability is potentially large3,46,59, OA is among the cumulative threats to marine food webs that among with other direct threats, such as potential overfishing, stand to disrupt marine food webs in these sensitive Antarctic waters. + +Here, we use a global ocean- sea ice- biogeochemistry model with eddy- permitting grid resolution on the Antarctic continental shelves and a representation of ice- shelf cavities to project \(21^{\mathrm{st}}\) - century OA in the regions encompassed by the five adopted and proposed Antarctic MPAs under four emission scenarios60,61. Our results suggest a dramatic increase in the volume of calcium carbonate undersaturated MPA waters by 2100. Further, we show that \(\mathrm{C}_{\mathrm{anth}}\) mixes into shelf waters more readily than open- ocean waters, resulting in more severe OA on the continental shelves. + +More severe ocean acidification on continental shelves than in the open ocean. Simulated top 2000 m average and bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations are at least two times higher on the continental shelves than in the open ocean under both present- day and future conditions throughout the Antarctic (Fig. 2a- d; high- emission scenario SSP5- 8.5). In the open ocean, bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations are elevated in the main dense- water formation regions in the Weddell Sea and the Ross Sea and downstream of these regions (up to \(4\mu \mathrm{mol}\mathrm{kg}^{- 1}\) and \(70\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in some places for the 1990s and 2090s, respectively), whereas \(\mathrm{C}_{\mathrm{anth}}\) concentrations on the continental shelves exceed \(10\mu \mathrm{mol}\mathrm{kg}^{- 1}\) and \(120\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in many places around the Antarctic continent (Fig. 2b & d). This shelf- open ocean gradient is in general agreement with observations, which indicate present- day shelf \(\mathrm{C}_{\mathrm{anth}}\) concentrations of up to \(50\mu \mathrm{mol}\mathrm{kg}^{- 1,15,20,21,23}\) and bottom concentrations of up to \(25\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in the Ross Sea MPA15,20,23 and \(16\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in the Weddell Sea MPA8,16,17. + +The difference in top 2000 m average \(\mathrm{C}_{\mathrm{anth}}\) concentrations between the continental shelves and the open ocean is caused by enhanced vertical mixing of \(\mathrm{C}_{\mathrm{anth}}\) in Antarctic shelf waters62, as + +<--- Page Split ---> + +illustrated by steeper slopes of the \(\mathrm{C}_{\mathrm{anth}}\) isolines in Fig. 2e- h (continental shelves) than in Fig. 2i- 1 (open ocean). This enhanced vertical mixing is a signature of dense water formation on the continental shelves, which readily transports \(\mathrm{C}_{\mathrm{anth}}\) from the surface to depth \(^{10,14,15,19,20,22}\) . Consequently, by 2100, \(\mathrm{C}_{\mathrm{anth}}\) concentrations in the top \(750\mathrm{m}\) exceed \(50\mu \mathrm{mol}\mathrm{kg}^{- 1}\) on all continental shelves in the high- emission scenario, while waters with such high concentrations are restricted to the top \(200\mathrm{m}\) in the open ocean. Given that the South Orkney Southern Shelf MPA only encompasses waters that are deeper than \(2000\mathrm{m}\) , it is unsurprising that \(\mathrm{C}_{\mathrm{anth}}\) penetration in this region is confined to shallower depths than in all other MPAs. The strong vertical mixing on the shelves (where most of the MPA area is located, Fig. 1) is the ultimate cause of severe shelf/MPA OA. + +Over the \(21^{\mathrm{st}}\) century, both the magnitude and sign of the vertical pH gradient in the adopted and proposed MPAs is altered due to OA. Surface waters in all MPAs have a higher pH than waters at depth in the 1990s, with the pH difference between the surface and \(1500\mathrm{m}\) ranging from 0.23 in the Ross Sea MPA to 0.27 in the South Orkney MPA (black lines in Fig. 3). Yet, by the 2090s, surface waters are more acidic than waters at depth for the two highest- emission scenarios, with a reversed vertical pH gradient ranging from - 0.01 in the Ross Sea MPA to - 0.07 in the East Antarctic MPA for the SSP5- 8.5 scenario. This is caused by the up to five times stronger surface decline in pH than at \(1500\mathrm{m}\) across all MPAs for all except the lowest emission scenario, for which the acidification at the surface and at depth are similar (solid lines in Fig. 3; same for \(\Omega_{\mathrm{arg}}\) ; see Supplementary Fig. 1 for \(\Omega_{\mathrm{calc}}\) ). Notably, this change in vertical gradient is less pronounced on the continental shelves in the Ross Sea MPA and Weddell Sea MPA (dashed lines), reflecting the more efficient downward transfer of the OA signal where dense waters are formed. + +In the 1990s, subsurface waters on the continental shelves have a higher pH compared to the average over the regions encompassed by the whole MPA (dashed black lines to the right of the solid black lines in Fig. 3). Averaged over the top \(2000\mathrm{m}\) , the pH in Antarctic coastal waters is up to 0.09 higher than in the regions encompassed by the whole MPAs (with a maximum for the Weddell Sea; Fig. 1). Over the \(21^{\mathrm{st}}\) century, this shelf- open ocean gradient reverses for the two highest- emission scenarios in the regions encompassed by the Weddell Sea, East Antarctic, and Ross Sea MPAs (dashed colored lines to the left of the solid colored lines in Fig. 3). By 2100, + +<--- Page Split ---> + +the top \(2000\mathrm{m}\) average pH on the continental shelves is up to 0.07 lower than in the area of the whole MPA, again demonstrating the overall stronger acidification in the shallow continental shelf seas than in the open ocean (Fig. 1; note that the region of the Antarctic Peninsula MPA is predominantly "continental shelf", while that of the South Orkney MPA is exclusively "open ocean"; see Methods). + +Drastic \(21^{\mathrm{st}}\) - century increase in calcium carbonate undersaturation. By 2100, the saturation state of aragonite and calcite decreases substantially in the Antarctic MPAs. While the depth of the aragonite saturation horizon, i.e., the depth at which \(\Omega_{\mathrm{arag}} = 1\) , ranges from 351- 678 m across the MPAs in the 1990s (deepest in the Weddell Sea), it is projected to rise to the surface by 2100 for the two highest- emission scenarios and to 40- 70 m and 80- 253 m for the lower- emission scenarios SSP2- 4.5 and SSP1- 2.6, respectively (see circles on the y axis of all \(\Omega_{\mathrm{arag}}\) panels in Fig. 3). In the 1990s, 51- 84% of waters in the upper 2000 m across all MPAs are undersaturated with respect to aragonite \((\Omega_{\mathrm{arag}}< 1)\) , whereas only 18- 51% of continental shelf MPA waters experience undersaturation (Fig. 4). Further, all waters in the MPAs are supersaturated with respect to calcite in the top \(2000\mathrm{m}\) . However, by 2100, 17- 63% of waters across all MPAs and 60- 75% of all shelf waters are undersaturated with respect to calcite in the highest- emission scenario. More critically, virtually all (>95%) waters in the MPAs are undersaturated with respect to aragonite for the three highest- emission scenarios, with this ubiquitous undersaturation only being avoidable for the lowest- emission scenario SSP1- 2.6. While 30- 65% remain supersaturated for the shelves in this scenario, undersaturated conditions in the whole MPAs are widespread even under strong climate- change mitigation (>20% supersaturation only for the proposed Weddell Sea and Antarctic Peninsula MPAs but <10% for all others). + +The onset of undersaturated conditions with respect to aragonite in the upper water column varies across seasons, scenarios, and to some extent also regions (Fig. 5). In all MPAs, the saturation state displays seasonal variability in the top \(\sim 50\mathrm{m}\) (dotted and dashed lines in Fig. 5a- e). As expected from the seasonal cycle of biological activity, sea surface temperature, and the entrainment of carbon into the upper ocean, winter is generally projected to be undersaturated earlier than summer. For the highest- emission scenario SSP5- 8.5, the onset of undersaturated conditions at the surface is fairly consistent across all MPAs (Fig. 5a- e), ranging from 2052 to 2060 for winter (June- August), from 2066 to 2069 for the annual mean, and from 2076 to 2085 for summer + +<--- Page Split ---> + +(December- February; latest in the Weddell Sea). Differences in the onset of undersaturation are larger across scenarios than across regions. For the emission scenario SSP3- 7.0, surface waters become undersaturated in the annual mean between 2074 and 2083 across MPAs, with the proposed Antarctic Peninsula MPA and the South Orkney MPA being the only regions for which surface waters become undersaturated in the summer months before 2100. For the emission scenario SSP2- 4.5, surface waters reach undersaturation within the \(21^{\text{st}}\) century during winter for the Ross Sea MPA and the Antarctic Peninsula and East Antarctic MPAs. Under the lowest emission scenario (SSP1- 2.6), the upper ocean remains supersaturated with respect to aragonite across all MPAs. + +Marine organisms are highly sensitive to the rate at which OA progresses, as this determines the time span available for possible adaptation63. In the upper \(50 \text{m}\) of the water column, our model projections suggest several decades of adaptation time between when \(50\%\) and \(90\%\) of all waters become undersaturated (Fig. 5f-j). For a given isoline of undersaturated volume, the number of years passing between the first and last month to reach this threshold varies across MPA regions and especially across emission scenarios (more compressed or elongated shape of isolines in Fig. 5f-j). While 20- 28 years pass in the highest emission scenario between when the first (typically in late fall or winter) and last month (typically in late spring or summer) reach \(50\%\) undersaturation, this is 28- 39 years across MPAs for the second- highest emission scenario SSP3- 7.0. Given that for the SSP2- 4.5 scenario, the threshold of \(50\%\) undersaturated volume is reached as early as 2060 for March and April in the Ross Sea region MPA, but not yet in summer by 2100, the time span between when the first and last month reach a given threshold appears to be longer the lower the emission scenario, providing ecosystems with an opportunity to adapt to a changing environment under higher mitigation scenarios. + +## Discussion and conclusions + +Under intermediate to high emission scenarios, ecosystems in the shallow continental shelf seas of Antarctic MPAs will be exposed to severe OA by 2100. Aragonite undersaturation is ubiquitous across all MPAs by 2100, implying that aragonite- forming organisms such as pteropods4,51,52,54 will be unable to find a refuge with supersaturated conditions. The Weddell Sea is sometimes referred to as a potential climate change refuge, motivating the proposal to establish an MPA in this region30. Based on our model experiments, the volume of supersaturated water is indeed + +<--- Page Split ---> + +173 highest in this region in the 1990s (Fig. 4), and OA is slightly delayed compared to the other MPAs over the \(21^{\mathrm{st}}\) century (Fig. 5). However, by 2100, the severity of OA in the Weddell Sea is barely distinguishable from that in the other MPAs for all except the lowest emission scenario (Fig. 3- 5). Strong local climate- change feedbacks explain why end- of- century OA in the Weddell Sea is comparable to that in other regions despite the delayed onset. While the area of the proposed Weddell Sea MPA has the lowest ice- free area of all MPAs in the 1990s (28% as compared to 48- 61% in the annual mean, 57% as compared to 84- 96% between December- February), all regions are virtually ice free in summer by 2100 in the highest- emission scenario (Fig. 5k- o). In the Weddell Sea, this strong sea- ice retreat facilitates substantially more oceanic \(\mathrm{CO_2}\) uptake than expected from the increase in atmospheric \(\mathrm{CO_2}\) levels alone (Supplementary Fig. 2). Besides the reduced capping effect of sea ice for air- sea exchange following the sea- ice retreat, this can be attributed to the lower buffer capacity resulting in more oceanic \(\mathrm{CO_2}\) uptake per primary production in an acidified ocean64. + +Given the high biodiversity in MPAs, the projected severe OA will likely impact organisms on many trophic levels, ranging from primary producers46 to fish3,57, and calcium carbonate- forming benthic organisms (Fig. 1)4,59. Due to their immobility and often long lifespan65, the impact on the latter will likely be substantial both on the continental shelves and in the open ocean, i.e., downstream of dense- water formation regions, where the OA signal spills from the shelf sea into the deep ocean (Fig. 2). While our knowledge is limited about the possibility for adaptation and physiological acclimatization in each of these organisms over a multi- decadal time span63,66,67, a higher sensitivity of benthic organisms to changing conditions than of mobile pelagic organisms is conceivable due to the smaller variability in pH and \(\Omega\) at depth than in the surface layer (simulated monthly \(\Omega_{\mathrm{arg}}\) in the 1990s averaged over the Antarctic continental shelf spans 0.91- 0.93 and 1.32- 1.67 below 500 m and above 50 m, respectively). In the two highest emission scenarios, the projected pH is \(< 7.8\) for all MPAs (Fig. 1), a level at which many detrimental effects have been reported for primary producers, zooplankton, fish, and benthic organisms3,4,54- 58. Even though it remains unclear to what extent a longer acclimation of adult organisms in laboratory experiments can reduce the severe impacts reported for egg development and juveniles68, strong OA would almost certainly disrupt food webs, thereby even affecting organisms which might not directly be negatively impacted by OA themselves. + +<--- Page Split ---> + +A disruption of Antarctic food webs seems even more likely when considering that OA is not the only marine ecosystem stressor exacerbated by anthropogenic climate change. High- latitude warming has been shown to aggravate OA impacts on Antarctic organisms \(^{53,55,69}\) , resulting in higher mortality and reduced hatching success of Antarctic dragonfish \(^{55}\) and increased resting metabolic rates in Antarctic juvenile pteropods \(^{53}\) than when considering OA alone. Further, warming could push cold- adapted organisms in Antarctic coastal waters beyond their thermal tolerance by 2100 (Supplementary Fig. 3) \(^{70}\) . While our model projections indicate oxygen concentrations above the hypoxic threshold of 61.3 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) in the Antarctic MPAs for all scenarios (Supplementary Fig. 3), the projected decline in oxygen concentrations might nonetheless negatively impact high- latitude ecosystems in conjunction with the projected warming and OA \(^{71}\) . The recent reduction in sea- ice cover due to the warming of waters around the Antarctic Peninsula has already negatively affected larval abundance of the Antarctic silverfish, which uses sea ice as spawning habitat \(^{72}\) , suggesting that the 21 \(^{\mathrm{st}}\) - century sea- ice retreat projected for all MPAs is worrisome (Fig. 5k- o). + +The projected severity of OA is higher for Antarctic shelf waters than open- ocean waters, and \(21^{\mathrm{st}}\) - century OA across all emission scenarios within the MPAs is comparable to OA in continental shelf areas not protected within an MPA (Supplementary Fig. 4). This implies two things: Firstly, regions outside of MPAs could serve as an important reference area for differentiating the ecosystem impacts of fishing from climate change as well as for assessing the effectiveness of MPAs \(^{26,29,30}\) . Secondly and more importantly, a reduction of fishing activities in a larger fraction of Antarctic coastal waters is advisable, given that ecosystems in high- latitude waters outside of the MPAs are not spared from severe OA. Given the pressure on ecosystems from both fishing and environmental change \(^{26}\) , both the mitigation of climate change and the limitation of fishing activities (e.g., via a network of MPAs) are key to minimizing disruptions of Antarctic food webs over the 21 \(^{\mathrm{st}}\) century. Multiple stressors of fishing and climate change can cause cumulative impacts that are more devastating than the impacts of each stressor alone \(^{73}\) . Through maintaining all trophic levels of the ecosystem and increasing species and genetic diversity, MPAs can enhance resilience to environmental change \(^{74}\) . Thus, MPAs and the simultaneous reduction of global emissions are the best policy tools for safeguarding Antarctic marine ecosystems. + +<--- Page Split ---> + +## References + +1. Doney, S. C., Fabry, V. J., Feely, R. A. & Kleypas, J. A. Ocean acidification: the other \(\mathrm{CO_2}\) problem. Annual review of marine science 1, 169-92 (2009). + +2. Orr, J. C. et al. Anthropogenic ocean acidification over the twenty-first century and its impact on calcifying organisms. Nature 437, 681-686 (2005). + +3. Hancock, A. M., King, C. K., Stark, J. S., McMinn, A. & Davidson, A. T. Effects of ocean acidification on Antarctic marine organisms: A meta-analysis. Ecology and Evolution 10, 4495-4514 (2020). + +4. Figuerola, B. et al. A Review and Meta-Analysis of Potential Impacts of Ocean Acidification on Marine Calcifiers From the Southern Ocean. Frontiers in Marine Science 8 (2021). + +5. McNeil, B. I. & Matear, R. J. Southern Ocean acidification: A tipping point at 450-ppm atmospheric \(\mathrm{CO_2}\) . Proceedings of the National Academy of Sciences 105, 18860-18864 (2008). + +6. Hauri, C., Friedrich, T. & Timmermann, A. Abrupt onset and prolongation of aragonite undersaturation events in the Southern Ocean. Nature Climate Change 6, 1-6 (2015). + +7. Negrete-Garcia, G., Lovenduski, N. S., Hauri, C., Krumhardt, K. M. & Lauvset, S. K. Sudden emergence of a shallow aragonite saturation horizon in the Southern Ocean. Nature Climate Change 9, 313-317 (2019). + +8. Anderson, L. G., Holby, O., Lindegren, R. & Ohlson, M. The transport of anthropogenic carbon dioxide into the Weddell Sea. Journal of Geophysical Research 96 (1991). + +9. Chen, C.-T. A. Some indications of excess \(\mathrm{CO_2}\) penetration near Cape Adare off the Ross Sea. La Mer 32, 167-172 (1994). + +10. Hoppema, M., Fahrbach, E., Stoll, M. H. & de Baar, H. J. Increase of carbon dioxide in the bottom water of the Weddell Sea, Antarctica. Marine Chemistry 59, 201-210 (1998). + +<--- Page Split ---> + +11. Hoppema, M., Fahrbach, E., Richter, K.-U., de Baar, H. J. & Kattner, G. Enrichment of silicate and \(\mathrm{CO_2}\) and circulation of the bottom water in the Weddell Sea. Deep Sea Research Part I: Oceanographic Research Papers 45, 1797-1817 (1998). + +12. Hoppema, M., Roether, W., Bellerby, R. G. J. & de Baar, H. J. W. Direct measurements reveal insignificant storage of anthropogenic \(\mathrm{CO_2}\) in the Abyssal Weddell Sea. Geophysical Research Letters 28, 1747-1750 (2001). + +13. McNeil, B. I., Tilbrook, B. & Matear, R. J. Accumulation and uptake of anthropogenic \(\mathrm{CO_2}\) in the Southern Ocean, south of Australia between 1968 and 1996. Journal of Geophysical Research: Oceans 106, 31431-31445 (2001). + +14. Lo Monaco, C., Metzl, N., Poisson, A., Brunet, C. & Schauer, B. Anthropogenic \(\mathrm{CO_2}\) in the Southern Ocean: Distribution and inventory at the Indian-Atlantic boundary (World Ocean Circulation Experiment line 16). Journal of Geophysical Research 110, C06010 (2005). + +15. Sandrini, S. et al. Anthropogenic carbon distribution in the Ross Sea, Antarctica. Antarctic Science 19, 395-407 (2007). + +16. Hauck, J., Hoppema, M., Bellerby, R. G., Völker, C. & Wolf-Gladrow, D. Data-based estimation of anthropogenic carbon and acidification in the Weddell Sea on a decadal timescale. Journal of Geophysical Research: Oceans 115, 1-14 (2010). + +17. van Heuven, S. M., Hoppema, M., Huhn, O., Slagter, H. A. & de Baar, H. J. Direct observation of increasing \(\mathrm{CO_2}\) in the Weddell Gyre along the Prime Meridian during 1973-2008. Deep Sea Research Part II: Topical Studies in Oceanography 58, 2613-2635 (2011). + +18. Huhn, O., Rhein, M., Hoppema, M. & van Heuven, S. Decline of deep and bottom water ventilation and slowing down of anthropogenic carbon storage in the Weddell Sea, 1984-2011. Deep Sea Research Part I: Oceanographic Research Papers 76, 66-84 (2013). + +19. van Heuven, S. M. A. C., Hoppema, M., Jones, E. M. & de Baar, H. J. W. Rapid invasion of anthropogenic \(\mathrm{CO_2}\) into the deep circulation of the Weddell Gyre. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 372, 20130056 (2014). + +<--- Page Split ---> + +20. Shadwick, E. H., Tilbrook, B. & Williams, G. D. Carbonate chemistry in the Mertz Polynya (East Antarctica): Biological and physical modification of dense water outflows and the export of anthropogenic \(\mathrm{CO_2}\) . Journal of Geophysical Research: Oceans 119, 1-14 (2014). + +21. Roden, N. P., Tilbrook, B., Trull, T. W., Virtue, P. & Williams, G. D. Carbon cycling dynamics in the seasonal sea-ice zone of East Antarctica. Journal of Geophysical Research: Oceans 121, 8749-8769 (2016). + +22. Mahieu, L., Lo Monaco, C., Metzl, N., Fin, J. & Mignon, C. Variability and stability of anthropogenic \(\mathrm{CO_2}\) in Antarctic Bottom Water observed in the Indian sector of the Southern Ocean, 1978-2018. Ocean Science 16, 1559-1576 (2020). + +23. Ingrosso, G. et al. Physical and biological controls on anthropogenic \(\mathrm{CO_2}\) sink of the Ross Sea. Frontiers in Marine Science 9, 1-19 (2022). + +24. Arrigo, K. R., van Dijken, G. & Long, M. Coastal Southern Ocean: A strong anthropogenic \(\mathrm{CO_2}\) sink. Geophysical Research Letters 35, 1-6 (2008). + +25. Tagliabue, A. & Arrigo, K. R. Decadal trends in air-sea \(\mathrm{CO_2}\) exchange in the Ross Sea (Antarctica). Geophysical Research Letters 43, 5271-5278 (2016). + +26. Brooks, C. M. et al. Progress towards a representative network of Southern Ocean protected areas. PLOS ONE 15, e0231361 (2020). + +27. Rogers, A. et al. Antarctic Futures: An Assessment of Climate-Driven Changes in Ecosystem Structure, Function, and Service Provisioning in the Southern Ocean. Annual Review of Marine Science 12, 87-120 (2020). + +28. Brooks, C. M. et al. Protect global values of the Southern Ocean ecosystem. Science 378, 477-479 (2022). + +29. Trathan, P. & Grant, S. The South Orkney Islands Southern Shelf Marine Protected Area. In Marine Protected Areas, 67-98 (Elsevier, 2020). + +30. Teschke, K. et al. Planning Marine Protected Areas under the CCAMLR regime – The case of the Weddell Sea (Antarctica). Marine Policy 124, 104370 (2021). + +<--- Page Split ---> + +307. Smith, W. O., Ainley, D. G., Arrigo, K. R. & Dinniman, M. S. The Oceanography and Ecology of the Ross Sea. Annual Review of Marine Science 6, 469–487 (2014). + +309. Nissen, C. & Vogt, M. Factors controlling the competition between Phaeocystis and diatoms in the Southern Ocean and implications for carbon export fluxes. Biogeosciences 18, 251–283 (2021). + +312. Mikhalevich, V. I. The general aspects of the distribution of Antarctic foraminifera. Micropaleontology 50, 179–194 (2004). + +314. Atkinson, A. et al. Oceanic circumpolar habitats of Antarctic krill. Marine Ecology Progress Series 362, 1–23 (2008). + +316. Hunt, B. et al. Pteropods in Southern Ocean ecosystems. Progress in Oceanography 78, 193–221 (2008). + +318. Koubbi, P. et al. Conservation and Management of Antarctic Silverfish Pleuragramma antarctica Populations and Habitats. In Vacchi, M., Pisano, E., Ghigliotti, L. e. (ed.) The Antarctic Silverfish: a Keystone Species in a Changing Ecosystem. Advances in Polar Ecology, vol 3., 287–305 (Springer, Cham., 2017). + +322. Ainley, D. G., Ballard, G. & Weller, J. Part I: Validation of the 2007 CCAMLR Bioregionalization Workshop Marine Protected Areas in the Southern Ocean. Tech. Rep., CCAMLR, WG-EMM-10/11, Hobart, Tasmania (2010). + +325. Koubbi, P. et al. Estimating the biodiversity of the East Antarctic shelf and oceanic zone for ecoregionalisation: Example of the ichthyofauna of the CEAMARC (Collaborative East Antarctic Marine Census) CAML surveys. Polar Science 4, 115–133 (2010). + +328. Ballard, G., Jongsomjit, D., Veloz, S. D. & Ainley, D. G. Coexistence of mesopredators in an intact polar ocean ecosystem: The basis for defining a Ross Sea Marine Protected Area. Biological Conservation 156, 72–82 (2012). + +331. Ropert-Coudert, Y. et al. Two Recent Massive Breeding Failures in an Adélie Penguin Colony Call for the Creation of a Marine Protected Area in D'Urville Sea/Mertz. Frontiers in Marine Science 5, 1–7 (2018). + +<--- Page Split ---> + +41. Brasier, M. J. et al. Benthic biodiversity in the South Orkney Islands Southern Shelf Marine Protected Area. Biodiversity 19, 1-15 (2018). + +42. Mitchell, E. G., Whittle, R. J. & Griffiths, H. J. Benthic ecosystem cascade effects in Antarctica using Bayesian network inference. Communications Biology 3, 582 (2020). + +43. Cummings, V. J., Bowden, D. A., Pinkerton, M. H., Halliday, N. J. & Hewitt, J. E. Ross Sea Benthic Ecosystems: Macro- and Mega-faunal Community Patterns From a Multi-environment Survey. Frontiers in Marine Science 8 (2021). + +44. Griffiths, H. J. et al. Breaking All the Rules: The First Recorded Hard Substrate Sessile Benthic Community Far Beneath an Antarctic Ice Shelf. Frontiers in Marine Science 8 (2021). + +45. Purser, A. et al. A vast iefish breeding colony discovered in the Antarctic. Current Biology 32, 842-850.e4 (2022). + +46. Deppeler, S. L. & Davidson, A. T. Southern Ocean Phytoplankton in a Changing Climate. Frontiers in Marine Science 4 (2017). + +47. Seifert, M., Rost, B., Trimborn, S. & Hauck, J. Meta-analysis of multiple driver effects on marine phytoplankton highlights modulating role of pCO2. Global Change Biology gcb.15341 (2020). + +48. Kawaguchi, S. et al. Risk maps for Antarctic krill under projected Southern Ocean acidification. Nature Climate Change 3, 843-847 (2013). + +49. Ericson, J. A. et al. Adult Antarctic krill proves resilient in a simulated high CO2 ocean. Communications Biology 1, 190 (2018). + +50. Kurihara, H. Effects of CO2-driven ocean acidification on the early developmental stages of invertebrates. Marine Ecology Progress Series 373, 275-284 (2008). + +51. Manno, C., Sandrini, S., Tositti, L. & Accornero, A. First stages of degradation of Limacina helicina shells observed above the aragonite chemical lysocline in Terra Nova Bay (Antarctica). Journal of Marine Systems 68, 91-102 (2007). + +<--- Page Split ---> + +52. Bednaršek, N. et al. Extensive dissolution of live pteropods in the Southern Ocean. Nature Geoscience 5, 881-885 (2012). + +53. Hoshijima, U., Wong, J. M. & Hofmann, G. E. Additive effects of pCO₂ and temperature on respiration rates of the Antarctic pteropod Limacina helicina antarctica. Conservation Physiology 5, 1-14 (2017). + +54. Manno, C., Peck, V. L. & Tarling, G. A. Pteropod eggs released at high pCO₂ lack resilience to ocean acidification. Scientific Reports 6, 25752 (2016). + +55. Flynn, E. E., Bjelde, B. E., Miller, N. A. & Todgham, A. E. Ocean acidification exerts negative effects during warming conditions in a developing Antarctic fish. Conservation Physiology 3, 1-16 (2015). + +56. Davis, B. E. et al. Antarctic emerald rockcod have the capacity to compensate for warming when uncoupled from CO₂-acidification. Global Change Biology 24, e655-e670 (2018). + +57. Todgham, A. E. & Mandic, M. Understanding the Metabolic Capacity of Antarctic Fishes to Acclimate to Future Ocean Conditions. Integrative and Comparative Biology 60, 1425-1437 (2020). + +58. Bylenga, C. H., Cummings, V. J. & Ryan, K. G. High resolution microscopy reveals significant impacts of ocean acidification and warming on larval shell development in Laternula elliptica. PLOS ONE 12, e0175706 (2017). + +59. Ingels, J. et al. Possible effects of global environmental changes on Antarctic benthos: a synthesis across five major taxa. Ecology and Evolution 2, 453-485 (2012). + +60. Nissen, C., Timmermann, R., Hoppema, M., Gurses, Ö. & Hauck, J. Abruptly attenuated carbon sequestration with Weddell Sea dense waters by 2100. Nature Communications 13, 3402 (2022). + +61. Nissen, C., Timmermann, R., Hoppema, M. & Hauck, J. A physical-biogeochemical tipping point cascade initiated on Weddell Sea continental shelves is avoidable in a 2°C scenario. Journal of Climate, in review xx, xx (2023). + +<--- Page Split ---> + +62. Foster, T. D. & Carmack, E. C. Frontal zone mixing and Antarctic Bottom water formation in the southern Weddell Sea. Deep Sea Research and Oceanographic Abstracts 23, 301-317 (1976). + +63. Kelly, M. W. & Hofmann, G. E. Adaptation and the physiology of ocean acidification. Functional Ecology 27, 980-990 (2013). + +64. Hauck, J. & Völker, C. Rising atmospheric \(\mathrm{CO_2}\) leads to large impact of biology on Southern Ocean CO2 uptake via changes of the Revelle factor. Geophysical Research Letters 42, 1459-1464 (2015). + +65. Montero-Serra, I., Linares, C., Doak, D. F., Ledoux, J. B. & Garrabou, J. Strong linkages between depth, longevity and demographic stability across marine sessile species. Proceedings of the Royal Society B: Biological Sciences 285 (2018). + +66. Berteaux, D., Réale, D., McAdam, A. G. & Boutin, S. Keeping Pace with Fast Climate Change: Can Arctic Life Count on Evolution? Integrative and Comparative Biology 44, 140-151 (2004). + +67. Kapsenberg, L. & Cyronak, T. Ocean acidification refugia in variable environments. Global Change Biology 25, 3201-3214 (2019). + +68. Suckling, C. C. et al. Adult acclimation to combined temperature and pH stressors significantly enhances reproductive outcomes compared to short-term exposures. Journal of Animal Ecology 84, 773-784 (2015). + +69. Saba, G., Bockus, A., Shaw, C. & Seibel, B. Combined effects of ocean acidification and elevated temperature on feeding, growth, and physiological processes of Antarctic krill Euphausia superba. Marine Ecology Progress Series 665, 1-18 (2021). + +70. Pörtner, H. O., Peck, L. & Somero, G. Thermal limits and adaptation in marine Antarctic ectotherms: an integrative view. Philosophical Transactions of the Royal Society B: Biological Sciences 362, 2233-2258 (2007). + +71. Pörtner, H.-O. Oxygen- and capacity-limitation of thermal tolerance: a matrix for integrating climate-related stressor effects in marine ecosystems. Journal of Experimental Biology 213, 881-893 (2010). + +<--- Page Split ---> + +72. Corso, A. D., Steinberg, D. K., Stammerjohn, S. E. & Hilton, E. J. Climate drives long-term change in Antarctic Silverfish along the western Antarctic Peninsula. Communications Biology 5, 104 (2022). + +73. Grant, S. M. et al. Local Drivers of Change in Southern Ocean Ecosystems: Human Activities and Policy Implications. Frontiers in Ecology and Evolution 9 (2021). + +74. Roberts, C. M. et al. Marine reserves can mitigate and promote adaptation to climate change. Proceedings of the National Academy of Sciences 114, 6167-6175 (2017). + +75. Wang, Q. et al. The Finite Element Sea Ice-Ocean Model (FESOM) v.1.4: formulation of an ocean general circulation model. Geoscientific Model Development 7, 663-693 (2014). + +76. Danilov, S. et al. Finite-Element Sea Ice Model (FESIM), version 2. Geoscientific Model Development 8, 1747-1761 (2015). + +77. Timmermann, R., Wang, Q. & Hellmer, H. Ice-shelf basal melting in a global finite-element sea-ice/ice-shelf/ocean model. Annals of Glaciology 53, 303-314 (2012). + +78. Schaffer, J. et al. A global, high-resolution data set of ice sheet topography, cavity geometry, and ocean bathymetry. Earth System Science Data 8, 543-557 (2016). + +79. Hauck, J. et al. Seasonally different carbon flux changes in the Southern Ocean in response to the southern annular mode. Global Biogeochemical Cycles 27, 1236-1245 (2013). + +80. Karakus, O. et al. Modeling the Impact of Macrozooplankton on Carbon Export Production in the Southern Ocean. Journal of Geophysical Research: Oceans 126, 1-22 (2021). + +81. Semmler, T. et al. Simulations for CMIP6 with the AWI Climate Model AWI-CM-1-1. Journal of Advances in Modeling Earth Systems 12, 1-34 (2020). + +82. DeVries, T. et al. Magnitude, trends, and variability of the global ocean carbon sink from 1985-2018. Global Biogeochemical Cycles, in review xx, xx (2023). + +83. Meinshausen, M. et al. Historical greenhouse gas concentrations for climate modelling (CMIP6). Geoscientific Model Development 10, 2057-2116 (2017). + +<--- Page Split ---> + +436 O'Neill, B. C. et al. The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6. 437 Geoscientific Model Development 9, 3461- 3482 (2016). 438 85. Brooks, C. M. et al. The Ross Sea, Antarctica: A highly protected MPA in international waters. 439 Marine Policy 134, 104795 (2021). 440 86. Apelgren, N. & Brooks, C. M. Norwegian interests and participation towards the creation of 441 marine protected areas in the Southern Ocean. The Polar Journal 11, 393- 412 (2021). 442 87. Orr, J. C. & Epitalon, J.- M. Improved routines to model the ocean carbonate system: mocsy 443 2.0. Geoscientific Model Development 8, 485- 499 (2015). 444 88. Bednaršek, N., Tarling, G. A., Bakker, D. C. E., Fielding, S. & Feely, R. A. Dissolution Dom- 445 inating Calcification Process in Polar Pteropods Close to the Point of Aragonite Undersatu- 446 ration. PLoS ONE 9, e109183 (2014). + +## Acknowledgements + +C.N. and M.H. acknowledge funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 820989 (project COMFORT). J.H. was supported by the Initiative and Networking Fund of the Helmholtz Association (Helmholtz Young Investigator Group Marine Carbon and Ecosystem Feedbacks in the Earth System [MarESys], grant number VH- NG- 1301) and R.T. by the Helmholtz Climate Initiative REKLIM (Regional Climate Change), a joint research project of the Helmholtz Association of German research centres (HGF). C.N. and N.S.L. are grateful for funding from the U.S. Department of Energy (DE- SC0022243). C.M.B. acknowledges support from the Pew Charitable Trusts, National Science Foundation and NASA. Computing resources were provided by the North- German Supercomputing Alliance (HLRN) project hbk00079. The work reflects only the authors' view; the European Commission and their executive agency are not responsible for any use that may be made of the information the work contains. + +<--- Page Split ---> + +## 460 Author contributions + +461 CN and NSL designed the study. CN ran the model simulations, with support from RT, MH, 462 and JH. CN performed the analysis. CMB provided the MPA masks. All authors contributed to 463 the interpretation of the results. CN wrote the manuscript, with input from all authors. + +## 464 Competing financial interests + +465 The authors declare no competing financial interests. + +<--- Page Split ---> + +## Methods + +Description of FESOM- REcoM and all model experiments. For this study, we use the same model setup as Nissen et al. (2022) \(^{60}\) and Nissen et al. (2023, in review) \(^{61}\) . We will summarize its main features hereafter, but refer the reader to the other publications for further detail. All model experiments are conducted with the global Finite Element Sea ice Ocean Model (FESOM) version 1.4 \(^{75}\) , which includes a sea- ice \(^{76}\) and an ice- shelf component \(^{77}\) with constant ice- shelf geometry derived from RTopo- 2 \(^{78}\) . The biogeochemical cycles of carbon, nitrogen, silicon, iron, and oxygen are resolved by coupling FESOM to the Regulated Ecosystem Model version 2 (RE- coM2) \(^{79,80}\) . REcoM2 includes two phytoplankton groups (silicifying diatoms and a mixed small phytoplankton group) and two zooplankton groups and allows for variable stoichiometry. + +The model equations are solved on a model grid with eddy- permitting resolution on Antarctic continental shelves ( \(< 5\mathrm{km}\) in southernmost parts; Supplementary Fig. 5). The grid resolution is lower in the open ocean ( \(< 100\mathrm{km}\) ). \(79\%\) and \(62\%\) of all surface model grid cells are south of \(60^{\circ}\mathrm{S}\) and \(70^{\circ}\mathrm{S}\) , respectively. The model grid has 99 z- levels in the vertical. At the surface, all simulations are forced with 3- hourly atmospheric output from the model experiments of the AWI Climate Model (AWI- CM) that contributed to the "Coupled Model Intercomparison Project Phase 6 (CMIP6)" \(^{81}\) . The model experiments are started in 1950 and initialized with output from the AWI- CM (physical tracers of FESOM) and with output from a simulation for the "Regional Carbon Cycle Assessment and Processes 2 (RECCAP2)" project (biogeochemical tracers of REcoM2) \(^{82}\) . + +For this study, eight model simulations are analyzed (4x simA, simB, 2x simC, simD) as outlined in the following: Following a historical simulation from 1950- 2014, we simulate four emission scenarios for 2015- 2100: SSP1- 2.6 (lowest emission scenario), SSP2- 4.5, SSP3- 7.0, and SSP5- 8.5 (highest emission scenario). In each case, the ocean model receives fluxes of momentum, heat, and freshwater from the 3- hourly atmospheric output corresponding to the given scenario, and the atmospheric \(\mathrm{CO_2}\) boundary condition for the ocean model increases from 312 ppm in \(1950^{83}\) to 446 ppm, 603 ppm, 867 ppm, and 1135 ppm, respectively, by \(2100^{84}\) (simA). As such, simA represents the net carbon fluxes (the sum of the natural and anthropogenic components). + +To isolate the anthropogenic carbon component, we conduct an additional simulation from 1950- 2100 with varying climate, but with a constant atmospheric \(\mathrm{CO_2}\) concentration bound + +<--- Page Split ---> + +496 ary condition from the year 1950 (simD). The anthropogenic component of the dissolved inorganic carbon (DIC) pool (Canth) can then be computed as the difference between the DIC in 498 simA and that in simD. Due to computational constraints, simD is only available for the high499 emission scenario SSP5- 8.5. For easier comparability with existing observation- based estimates, 500 anthropogenic carbon concentrations are reported in \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) in the manuscript, and we use 501 monthly mean surface- referenced potential density to convert all model fields from the model 502 units (mmol \(\mathrm{m}^{- 3}\) ). We note that since simD was generated using the 1950 atmospheric \(\mathrm{CO}_{2}\) 503 concentration, the anthropogenic carbon quantified in this study is only that which has accumulated since 1950 and not since pre- industrial times. Thus, the resulting model- based values 504 of \(\mathrm{C}_{\mathrm{anth}}\) are likely biased low compared to observation- based estimates which include all an506 throgenic carbon. Yet, since most of the increase in atmospheric \(\mathrm{CO}_{2}\) levels occurred after 507 195083, our model- based \(\mathrm{C}_{\mathrm{anth}}\) estimate can be assumed to capture the vast majority of the sig508 nal. The lower simulated bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations in the East Antarctic open- ocean sector 509 in the 1990s (<1 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) ) than suggested by observations (up to 25 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) 13,14,21,22) may 510 be attributed to the absence of shelf water masses which are dense enough to efficiently trans511 fer anthropogenic carbon to the abyss in this sector (not shown). However, we acknowledge 512 the difficulty a) to compare different observation- based estimates to each other due the uncer513 tainty arising from the use of various back- calculation techniques17 and b) to directly compare 514 the observation- based estimates from various years to the decadal- average model tracer. + +515 To assess the model drift, we run a control simulation simB, in which both climate variability 516 and atmospheric \(\mathrm{CO}_{2}\) concentrations are held constant. For this experiment, the ocean model 517 receives atmospheric \(\mathrm{CO}_{2}\) from the year 1950 and repeated year 1955 momentum, heat, and 518 freshwater fluxes. This year was identified to best represent "normal" atmospheric conditions, 519 as determined from an assessment of the Southern Annular Mode and El Niño Southern Oscil520 lation in the AWI- CM model output (see Nissen et al., 202260 for details). Overall, the surface 521 drift in carbonate chemistry variables is minimal, whereas some subsurface drift exists in simB 522 (see Supplementary Fig. 6 for pH as an example). Any deep- ocean model drift can likely at 523 least partially be attributed to a delayed adjustment of the deep ocean to upper ocean carbonate 524 chemistry conditions. However, we note that since the model drift is generally much smaller 525 than the acidification signal in the different emission scenarios, it only has a small impact on our + +<--- Page Split ---> + +526 results (with the notable exception of the lowest- emission scenario for the proposed Antarctic Peninsula MPA; Supplementary Fig. 6). + +Further, two complementary model experiments for 1950- 2100 are performed in which climate variability is held constant, while atmospheric \(\mathrm{CO_2}\) concentrations increase (simC). By computing \(\sin A\) minus \(\sin C\) of any model tracer or flux, we can isolate the effect of climate change on the projected changes in these. Due to computational limitations, \(\sin C\) is only available for the intermediate scenario SSP2- 4.5 and the high- emission scenario SSP5- 8.5. For this manuscript, we quantify the total climate effect, thus combining the impact of, e.g., warming and changes in circulation, sea- ice cover, freshwater input, and biological productivity on the projected model fields (see Supplementary Fig. 2). + +Marine Protected Areas. Ocean acidification is assessed in five Marine Protected Areas (MPAs), of which two exist already (Ross Sea region and South Orkney Islands Southern Shelf) and three are proposed (Weddell Sea, East Antarctic, Antarctic Peninsula) \(^{85}\) . Masks on the model grid for the different MPAs are derived from previously published coordinates \(^{85,86}\) . For this study, we only consider the MPA area outside of the ice- shelf cavities, and "continental shelf" and "open ocean" are separated at the 2000 m isobath. Based on the model grid, the surface area in each MPA corresponds to 1.44 Mio. km \(^2\) for the Weddell Sea (53% on the continental shelf), 0.97 Mio. km \(^2\) for the East Antarctic (29% on the continental shelf), 1.59 Mio. km \(^2\) for the Ross Sea (45% on the continental shelf), 0.56 Mio. km \(^2\) for the Antarctic Peninsula (73% on the continental shelf), 0.11 Mio. km \(^2\) for the South Orkney Islands (0% on the continental shelf). As the South Orkney Islands Southern Shelf MPA is entirely defined as "open ocean" based on the model topography and as the vast majority of grid points for the Antarctic Peninsula MPA are defined as "continental shelf", we only report values for the whole MPA for these two MPAs in all Figures. In total, 60% of the continental shelf area outside of the ice- shelf cavities is covered by the five MPAs assessed in this study. The area- weighted average grid resolution outside of the ice- shelf cavities corresponds to 35 km in the Weddell Sea MPA (spread 7- 88 km; 7- 55 km for the continental shelf; Supplementary Fig. 5), 46 km in the East Antarctic MPA (spread 6- 93 km; 6- 44 km for the continental shelf), 54 km in the Ross Sea region MPA (spread 4- 138 km; 4- 75 km for the continental shelf), 30 km in the Antarctic Peninsula MPA (spread 2- 84 km; 2- 45 km for the continental shelf), and 72 km in the South Orkney Islands Southern Shelf MPA (spread 57- 85 km). + +<--- Page Split ---> + +We note that while there are between 1250- 3632 model grid cells in the four largest MPAs, only 22 model grid cells lie within the South Orkney MPA. + +Calculation of carbonate chemistry. We calculate monthly mean pH and the saturation states of aragonite ( \(\Omega_{arag}\) ) and calcite ( \(\Omega_{calc}\) ) offline using the python routines to model the ocean carbonate system (mocsy v2.0) \(^{87}\) (code is available at https://github.com/jamesorr/mocsy, last access February, 1, 2023). As input parameters to the functions, we use monthly model output of dissolved inorganic carbon, alkalinity, potential temperature, salinity, silicic acid, phosphate (obtained from model nitrate field using the Redfield ratio 16), and sea- level pressure from the atmospheric output of the AWI- CM. + +## Data Availability + +The model output underlying the findings of this study will be uploaded to Zenodo upon acceptance of the manuscript. The data can be made available to the editor and the reviewers at any time. + +## Code Availability + +The Fortran source code of FESOM1.4- REcoM2 can be obtained via https://fesom.de/models/fesom14/ (last access January, 23, 2023). The analysis of model output was done with the open- source software Python. The routines to compute the carbonate chemistry offline from monthly model output in Python are available via https://github.com/jamesorr/mocsy (downloaded on February, 24, 2022). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: Key organisms in the Southern Ocean and \(21^{\mathrm{st}}\) -century ocean acidification in Marine Protected Areas. Map of established (South Orkney Islands Southern Shelf and Ross Sea region) and proposed (Weddell Sea, East Antarctic, Western Antarctic Peninsula) Marine Protected Areas (MPAs). Key species and groups which may be negatively impacted by ocean acidification (see references in main text) are depicted around the map (not to scale). Bar plots denote the top 2000 m average pH in the five different MPAs in the 1990s (dark grey) and in the 2090s (colors) for the four emission scenarios (sorted from low emission to high emission). For the Weddell Sea, East Antarctic, and Ross Sea MPAs, the hatched bars denote the average pH for the continental shelf south of the 2000 m isobath (grey contour on the map). Using this definition, the majority of the area of the proposed MPA along the western Antarctic Peninsula is continental shelf (see Methods), and bars are therefore hatched for this region. Note the arrows for the SSP5-8.5 scenario, indicating a lower pH on the continental shelves than for the whole-MPA average.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Distribution of anthropogenic carbon across the high-latitude Southern Ocean. Distribution of anthropogenic carbon concentrations in \(\mu \mathrm{mol} \mathrm{kg}^{-1}\) a averaged over the top \(2000 \mathrm{m}\) and b at the bottom in the 1990s of the historical simulation. c-d Same as a-b, but for the 2090s of the high-emission scenario SSP5-8.5. The top \(2000 \mathrm{m}\) are equivalent to the whole water column on the continental shelves, defined as the area south of the \(2000 \mathrm{m}\) isobath (grey contour). Note the logarithmic colorbar and the one order of magnitude difference between the 1990s and the 2090s. Outlines of the Marine Protected Areas (MPAs) are denoted in white, and the ice-shelf front is shown as the dark grey contour. e-g Temporal evolution of anthropogenic carbon concentrations with depth and averaged over the continental shelves of e the proposed Weddell Sea MPA, f the proposed East Antarctic MPA, and g the adopted Ross Sea region MPA. i-k Same as e-g, but for the open ocean. h and l show the averages for h the proposed MPA along the western Antarctic Peninsula and l the adopted South Orkney Islands Southern Shelf MPA.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: Changes in the vertical distribution of \(\Omega_{\mathrm{arg}}\) . Vertical profiles of a pH and b the saturation state with respect to aragonite \(\left(\Omega_{\mathrm{arg}}\right)\) in the proposed Weddell Sea Marine Protected Area (MPA). Solid lines show the profiles above 2000 m in the whole MPA, and dashed lines show the profile for the continental shelves south of the 2000 m isobath (see dark grey contour in map). The black lines denote the 1990s in the historical simulation, and the colored lines the 2090s in the four emission scenarios. The profiles on the left in each panel show absolute values, and profiles on the right show the change in each property between the 2090s and the 1990s \(\left(\Delta \mathrm{pH}\right.\) and \(\Delta \Omega_{\mathrm{arg}}\) ). In b, undersaturated conditions, i.e., \(\Omega_{\mathrm{arg}}< 1\) , are highlighted with the colored background, and the depth of \(\Omega_{\mathrm{arg}} = 1\) for the profile of the whole MPA is highlighted with a circle on the y axis. c-j Same as a-b, but for c-d the proposed East Antarctic MPA, e-f the adopted Ross Sea region MPA, g-h the proposed Antarctic Peninsula MPA, i-j the adopted South Orkney Islands Southern Shelf MPA. Note that only the vertical profiles for the whole MPA are shown in panels g-j (see Methods).
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: Changes in the volume distribution across different \(\Omega_{\mathrm{arg}}\) and \(\Omega_{\mathrm{calc}}\) classes. a-b Distribution of waters across different classes of the saturation state with respect to aragonite \((\Omega_{\mathrm{arg}})\) a above \(2000\mathrm{m}\) in the whole proposed Weddell Sea Marine Protected Area (MPA) and b on the continental shelves south of the \(2000\mathrm{m}\) isobath (see dark grey contour in map). Results are shown for the 1990s of the historical simulation and for the 2090s in the four emission scenarios. c-d Same as a-b, but for the distribution of waters across different classes of the saturation state with respect to calcite \((\Omega_{\mathrm{calc}})\) . Note the different classes of saturation states shown for aragonite and calcite. While the 0.65 threshold for \(\Omega_{\mathrm{arg}}\) was chosen to best highlight the stronger undersaturation on the shelf than in the whole MPA for the highest emission scenario, we assume that it is more costly for an organism to sustain calcification at \(\Omega_{\mathrm{arg}}< 0.65\) than at \(0.65\leq \Omega_{\mathrm{arg}}< 1.0\) (ref. \(^{88}\) ) and that this division of the volume of undersaturated waters is therefore meaningful. e-l Same as a-d, but for e-h the proposed East Antarctic MPA and i-l the adopted Ross Sea region MPA. m-p Same as a and c, but for m-n the proposed Antarctic Peninsula MPA and o-p the adopted South Orkney Islands Southern Shelf MPA. Note that only the distribution for the whole MPA is shown in panels m-p (see Methods).
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5: Temporal evolution of \(\Omega_{\mathrm{arg}}\) and sea ice cover. a Temporal evolution of the depth of the \(\Omega_{\mathrm{arg}} = 1\) isoline in the proposed Weddell Sea Marine Protected Area (MPA) for the SSP5-8.5 scenario (dark blue), the SSP3-7.0 scenario (intermediate blue), and the SSP2-4.5 scenario (light blue). The different line styles represent the annual mean (solid), the June-August mean (JJA; dotted), and the December-March mean (DJF; dashed). The SSP1-2.6 is not shown because the \(\Omega_{\mathrm{arg}} = 1\) isoline never reaches the top \(80 \mathrm{~m}\) of the water column. b-e Same as a, but for b the proposed East Antarctic MPA, c the adopted Ross Sea region MPA, d the proposed Antarctic Peninsula MPA, and e the adopted South Orkney Islands Southern Shelf MPA. f-j Temporal evolution of the \(50\%\) (thick line) and \(90\%\) (thin line) isolines of the volume with \(\Omega_{\mathrm{arg}} < 1\) in the upper \(50 \mathrm{~m}\) of the water column in the different MPAs. Colors show the different emission scenarios (see above). k-o Monthly ice-free surface area in the different MPAs for the 1990s (black) and the 2090s in the four emission scenarios (colors).
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Nissen2023COMFORT3SUPPLEMENTv20230330.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8_det.mmd b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9014351aece8e608463bb52a8d28616ee21e64cd --- /dev/null +++ b/preprint/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8/preprint__140e446d17813813ee1de7b678d523d1deec0f6d422d6904aa4e2ba3721612a8_det.mmd @@ -0,0 +1,489 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 905, 207]]<|/det|> +# Severe 21st-century ocean acidification demands continuance and expansion of Antarctic Marine Protected Areas + +<|ref|>text<|/ref|><|det|>[[44, 230, 437, 270]]<|/det|> +Cara Nissen (cara.nissen@colorado.edu) University of Colorado, Boulder + +<|ref|>text<|/ref|><|det|>[[44, 277, 678, 318]]<|/det|> +Nicole Lovenduski University of Colorado Boulder https://orcid.org/0000- 0001- 5893- 1009 + +<|ref|>text<|/ref|><|det|>[[44, 325, 250, 364]]<|/det|> +Cassandra Brooks University of Colorado + +<|ref|>text<|/ref|><|det|>[[44, 371, 625, 411]]<|/det|> +Mario Hoppema Alfred Wegener Institute https://orcid.org/0000- 0002- 2326- 619X + +<|ref|>text<|/ref|><|det|>[[44, 417, 532, 457]]<|/det|> +Ralph Timmermann Alfred Wegener Institute for Polar and Marine Research + +<|ref|>text<|/ref|><|det|>[[44, 463, 958, 526]]<|/det|> +Judith Hauck Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research https://orcid.org/0000- 0003- 4723- 9652 + +<|ref|>sub_title<|/ref|><|det|>[[44, 568, 101, 585]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 606, 135, 624]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 644, 290, 662]]<|/det|> +Posted Date: June 5th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 682, 474, 700]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2758143/v1 + +<|ref|>text<|/ref|><|det|>[[44, 718, 909, 760]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 780, 530, 799]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 835, 930, 877]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 4th, 2024. See the published version at https://doi.org/10.1038/s41467- 023- 44438- x. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[75, 73, 911, 131]]<|/det|> +# Severe \(21^{\mathrm{st}}\) -century ocean acidification demands continuance and expansion of Antarctic Marine Protected Areas + +<|ref|>text<|/ref|><|det|>[[75, 153, 860, 190]]<|/det|> +Cara Nissen \(^{1,2*}\) , Nicole S. Lovenduski \(^{1}\) , Cassandra M. Brooks \(^{3}\) , Mario Hoppema \(^{2}\) , Ralph Timmermann \(^{2}\) , Judith Hauck \(^{2}\) + +<|ref|>text<|/ref|><|det|>[[75, 192, 925, 281]]<|/det|> +\(^{1}\) Department of Atmospheric and Oceanic Sciences and Institute of Arctic and Alpine Research, University of Colorado Boulder, Boulder, CO, USA. \(^{2}\) Alfred Wegener Institut, Helmholtz- Zentrum für Polar- und Meeresforschung, Bremerhaven, Germany. \(^{3}\) Department of Environmental Studies, University of Colorado Boulder, Boulder, CO, USA. \* corresponding author: cara.nissen@colorado.edu + +<|ref|>text<|/ref|><|det|>[[58, 308, 930, 640]]<|/det|> +1 Antarctic coastal waters are home to several established or proposed Marine Protected Areas (MPAs) supporting exceptional biodiversity, which is threatened by anthropogenic climate change. Despite a particular sensitivity to ocean acidification (OA), little is known about the future carbonate chemistry of high- latitude Southern Ocean waters. Here, we use a high- resolution ocean- sea ice- biogeochemistry model with realistic ice- shelf geometry to investigate \(21^{\mathrm{st}}\) - century OA in Antarctic MPAs under four emission scenarios. By 2100, we project surface pH declines of up to 0.42 (total scale), corresponding to a \(161\%\) increase in hydrogen ion concentration relative to the 1990s. End- of- century aragonite undersaturation is ubiquitous across MPAs under the three highest emission scenarios. Vigorous vertical mixing of anthropogenic carbon on the continental shelves produces severe OA within the Weddell Sea, East Antarctic, and Ross Sea MPAs. Our findings call for continuity and expansion of Antarctic MPAs to reduce pressures on ecosystem integrity. + +<|ref|>text<|/ref|><|det|>[[55, 652, 930, 929]]<|/det|> +Ocean acidification (OA) is an environmental problem \(^{1}\) threatening marine biodiversity across the global oceans \(^{2 - 4}\) . OA is caused by the invasion of anthropogenic carbon (Canth) into the ocean, which results in a decline in pH and in the saturation states of the calcium carbonate minerals aragonite (Ωarag) and calcite (Ωcalc). The Southern Ocean is especially sensitive to OA due to the prevalence of high CO₂ concentrations in cold waters and the upwelling of carbon-rich deep waters with low pH values \(^{1,2}\) . Earth system model projections suggest that surface waters in the open Southern Ocean might become seasonally undersaturated with respect to the more soluble mineral aragonite (Ωarag <1) by 2035 \(^{2,5 - 7}\) . Projections of OA on the Antarctic continental shelves have so far been elusive due to the inability of coarse-resolution models to realistically resolve physical and biogeochemical processes at high resolution and in ice-shelf cavities. Ob + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 73, 928, 352]]<|/det|> +posed to both OA and warming55. Juvenile Antarctic emerald rockcod and Black rockcod show reduced metabolic capacity to adapt to warming when exposed to \(\mathrm{pH}< 7.75^{56,57}\) . Benthic bivalve and sea urchin larvae are increasingly malformed and damaged a \(\mathrm{pH}< 7.8\) , possibly compromising their survival4,58. Altogether, acknowledging knowledge gaps on direct OA impacts for some keystone species (e.g., Antarctic silverfish and upper-trophic level organisms), even small impacts of OA on individual species can accumulate through the food web, as shifts in development can cause predator- prey mismatches49,54,55. While the overall number of available studies is low and inter- species variability is potentially large3,46,59, OA is among the cumulative threats to marine food webs that among with other direct threats, such as potential overfishing, stand to disrupt marine food webs in these sensitive Antarctic waters. + +<|ref|>text<|/ref|><|det|>[[55, 360, 929, 549]]<|/det|> +Here, we use a global ocean- sea ice- biogeochemistry model with eddy- permitting grid resolution on the Antarctic continental shelves and a representation of ice- shelf cavities to project \(21^{\mathrm{st}}\) - century OA in the regions encompassed by the five adopted and proposed Antarctic MPAs under four emission scenarios60,61. Our results suggest a dramatic increase in the volume of calcium carbonate undersaturated MPA waters by 2100. Further, we show that \(\mathrm{C}_{\mathrm{anth}}\) mixes into shelf waters more readily than open- ocean waters, resulting in more severe OA on the continental shelves. + +<|ref|>text<|/ref|><|det|>[[55, 571, 929, 876]]<|/det|> +More severe ocean acidification on continental shelves than in the open ocean. Simulated top 2000 m average and bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations are at least two times higher on the continental shelves than in the open ocean under both present- day and future conditions throughout the Antarctic (Fig. 2a- d; high- emission scenario SSP5- 8.5). In the open ocean, bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations are elevated in the main dense- water formation regions in the Weddell Sea and the Ross Sea and downstream of these regions (up to \(4\mu \mathrm{mol}\mathrm{kg}^{- 1}\) and \(70\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in some places for the 1990s and 2090s, respectively), whereas \(\mathrm{C}_{\mathrm{anth}}\) concentrations on the continental shelves exceed \(10\mu \mathrm{mol}\mathrm{kg}^{- 1}\) and \(120\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in many places around the Antarctic continent (Fig. 2b & d). This shelf- open ocean gradient is in general agreement with observations, which indicate present- day shelf \(\mathrm{C}_{\mathrm{anth}}\) concentrations of up to \(50\mu \mathrm{mol}\mathrm{kg}^{- 1,15,20,21,23}\) and bottom concentrations of up to \(25\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in the Ross Sea MPA15,20,23 and \(16\mu \mathrm{mol}\mathrm{kg}^{- 1}\) in the Weddell Sea MPA8,16,17. + +<|ref|>text<|/ref|><|det|>[[55, 884, 928, 932]]<|/det|> +The difference in top 2000 m average \(\mathrm{C}_{\mathrm{anth}}\) concentrations between the continental shelves and the open ocean is caused by enhanced vertical mixing of \(\mathrm{C}_{\mathrm{anth}}\) in Antarctic shelf waters62, as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 73, 930, 352]]<|/det|> +illustrated by steeper slopes of the \(\mathrm{C}_{\mathrm{anth}}\) isolines in Fig. 2e- h (continental shelves) than in Fig. 2i- 1 (open ocean). This enhanced vertical mixing is a signature of dense water formation on the continental shelves, which readily transports \(\mathrm{C}_{\mathrm{anth}}\) from the surface to depth \(^{10,14,15,19,20,22}\) . Consequently, by 2100, \(\mathrm{C}_{\mathrm{anth}}\) concentrations in the top \(750\mathrm{m}\) exceed \(50\mu \mathrm{mol}\mathrm{kg}^{- 1}\) on all continental shelves in the high- emission scenario, while waters with such high concentrations are restricted to the top \(200\mathrm{m}\) in the open ocean. Given that the South Orkney Southern Shelf MPA only encompasses waters that are deeper than \(2000\mathrm{m}\) , it is unsurprising that \(\mathrm{C}_{\mathrm{anth}}\) penetration in this region is confined to shallower depths than in all other MPAs. The strong vertical mixing on the shelves (where most of the MPA area is located, Fig. 1) is the ultimate cause of severe shelf/MPA OA. + +<|ref|>text<|/ref|><|det|>[[52, 359, 930, 720]]<|/det|> +Over the \(21^{\mathrm{st}}\) century, both the magnitude and sign of the vertical pH gradient in the adopted and proposed MPAs is altered due to OA. Surface waters in all MPAs have a higher pH than waters at depth in the 1990s, with the pH difference between the surface and \(1500\mathrm{m}\) ranging from 0.23 in the Ross Sea MPA to 0.27 in the South Orkney MPA (black lines in Fig. 3). Yet, by the 2090s, surface waters are more acidic than waters at depth for the two highest- emission scenarios, with a reversed vertical pH gradient ranging from - 0.01 in the Ross Sea MPA to - 0.07 in the East Antarctic MPA for the SSP5- 8.5 scenario. This is caused by the up to five times stronger surface decline in pH than at \(1500\mathrm{m}\) across all MPAs for all except the lowest emission scenario, for which the acidification at the surface and at depth are similar (solid lines in Fig. 3; same for \(\Omega_{\mathrm{arg}}\) ; see Supplementary Fig. 1 for \(\Omega_{\mathrm{calc}}\) ). Notably, this change in vertical gradient is less pronounced on the continental shelves in the Ross Sea MPA and Weddell Sea MPA (dashed lines), reflecting the more efficient downward transfer of the OA signal where dense waters are formed. + +<|ref|>text<|/ref|><|det|>[[52, 728, 930, 919]]<|/det|> +In the 1990s, subsurface waters on the continental shelves have a higher pH compared to the average over the regions encompassed by the whole MPA (dashed black lines to the right of the solid black lines in Fig. 3). Averaged over the top \(2000\mathrm{m}\) , the pH in Antarctic coastal waters is up to 0.09 higher than in the regions encompassed by the whole MPAs (with a maximum for the Weddell Sea; Fig. 1). Over the \(21^{\mathrm{st}}\) century, this shelf- open ocean gradient reverses for the two highest- emission scenarios in the regions encompassed by the Weddell Sea, East Antarctic, and Ross Sea MPAs (dashed colored lines to the left of the solid colored lines in Fig. 3). By 2100, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 928, 209]]<|/det|> +the top \(2000\mathrm{m}\) average pH on the continental shelves is up to 0.07 lower than in the area of the whole MPA, again demonstrating the overall stronger acidification in the shallow continental shelf seas than in the open ocean (Fig. 1; note that the region of the Antarctic Peninsula MPA is predominantly "continental shelf", while that of the South Orkney MPA is exclusively "open ocean"; see Methods). + +<|ref|>text<|/ref|><|det|>[[50, 220, 930, 696]]<|/det|> +Drastic \(21^{\mathrm{st}}\) - century increase in calcium carbonate undersaturation. By 2100, the saturation state of aragonite and calcite decreases substantially in the Antarctic MPAs. While the depth of the aragonite saturation horizon, i.e., the depth at which \(\Omega_{\mathrm{arag}} = 1\) , ranges from 351- 678 m across the MPAs in the 1990s (deepest in the Weddell Sea), it is projected to rise to the surface by 2100 for the two highest- emission scenarios and to 40- 70 m and 80- 253 m for the lower- emission scenarios SSP2- 4.5 and SSP1- 2.6, respectively (see circles on the y axis of all \(\Omega_{\mathrm{arag}}\) panels in Fig. 3). In the 1990s, 51- 84% of waters in the upper 2000 m across all MPAs are undersaturated with respect to aragonite \((\Omega_{\mathrm{arag}}< 1)\) , whereas only 18- 51% of continental shelf MPA waters experience undersaturation (Fig. 4). Further, all waters in the MPAs are supersaturated with respect to calcite in the top \(2000\mathrm{m}\) . However, by 2100, 17- 63% of waters across all MPAs and 60- 75% of all shelf waters are undersaturated with respect to calcite in the highest- emission scenario. More critically, virtually all (>95%) waters in the MPAs are undersaturated with respect to aragonite for the three highest- emission scenarios, with this ubiquitous undersaturation only being avoidable for the lowest- emission scenario SSP1- 2.6. While 30- 65% remain supersaturated for the shelves in this scenario, undersaturated conditions in the whole MPAs are widespread even under strong climate- change mitigation (>20% supersaturation only for the proposed Weddell Sea and Antarctic Peninsula MPAs but <10% for all others). + +<|ref|>text<|/ref|><|det|>[[50, 704, 928, 921]]<|/det|> +The onset of undersaturated conditions with respect to aragonite in the upper water column varies across seasons, scenarios, and to some extent also regions (Fig. 5). In all MPAs, the saturation state displays seasonal variability in the top \(\sim 50\mathrm{m}\) (dotted and dashed lines in Fig. 5a- e). As expected from the seasonal cycle of biological activity, sea surface temperature, and the entrainment of carbon into the upper ocean, winter is generally projected to be undersaturated earlier than summer. For the highest- emission scenario SSP5- 8.5, the onset of undersaturated conditions at the surface is fairly consistent across all MPAs (Fig. 5a- e), ranging from 2052 to 2060 for winter (June- August), from 2066 to 2069 for the annual mean, and from 2076 to 2085 for summer + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 74, 930, 322]]<|/det|> +(December- February; latest in the Weddell Sea). Differences in the onset of undersaturation are larger across scenarios than across regions. For the emission scenario SSP3- 7.0, surface waters become undersaturated in the annual mean between 2074 and 2083 across MPAs, with the proposed Antarctic Peninsula MPA and the South Orkney MPA being the only regions for which surface waters become undersaturated in the summer months before 2100. For the emission scenario SSP2- 4.5, surface waters reach undersaturation within the \(21^{\text{st}}\) century during winter for the Ross Sea MPA and the Antarctic Peninsula and East Antarctic MPAs. Under the lowest emission scenario (SSP1- 2.6), the upper ocean remains supersaturated with respect to aragonite across all MPAs. + +<|ref|>text<|/ref|><|det|>[[50, 330, 930, 721]]<|/det|> +Marine organisms are highly sensitive to the rate at which OA progresses, as this determines the time span available for possible adaptation63. In the upper \(50 \text{m}\) of the water column, our model projections suggest several decades of adaptation time between when \(50\%\) and \(90\%\) of all waters become undersaturated (Fig. 5f-j). For a given isoline of undersaturated volume, the number of years passing between the first and last month to reach this threshold varies across MPA regions and especially across emission scenarios (more compressed or elongated shape of isolines in Fig. 5f-j). While 20- 28 years pass in the highest emission scenario between when the first (typically in late fall or winter) and last month (typically in late spring or summer) reach \(50\%\) undersaturation, this is 28- 39 years across MPAs for the second- highest emission scenario SSP3- 7.0. Given that for the SSP2- 4.5 scenario, the threshold of \(50\%\) undersaturated volume is reached as early as 2060 for March and April in the Ross Sea region MPA, but not yet in summer by 2100, the time span between when the first and last month reach a given threshold appears to be longer the lower the emission scenario, providing ecosystems with an opportunity to adapt to a changing environment under higher mitigation scenarios. + +<|ref|>sub_title<|/ref|><|det|>[[75, 741, 384, 761]]<|/det|> +## Discussion and conclusions + +<|ref|>text<|/ref|><|det|>[[50, 770, 940, 936]]<|/det|> +Under intermediate to high emission scenarios, ecosystems in the shallow continental shelf seas of Antarctic MPAs will be exposed to severe OA by 2100. Aragonite undersaturation is ubiquitous across all MPAs by 2100, implying that aragonite- forming organisms such as pteropods4,51,52,54 will be unable to find a refuge with supersaturated conditions. The Weddell Sea is sometimes referred to as a potential climate change refuge, motivating the proposal to establish an MPA in this region30. Based on our model experiments, the volume of supersaturated water is indeed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 74, 930, 437]]<|/det|> +173 highest in this region in the 1990s (Fig. 4), and OA is slightly delayed compared to the other MPAs over the \(21^{\mathrm{st}}\) century (Fig. 5). However, by 2100, the severity of OA in the Weddell Sea is barely distinguishable from that in the other MPAs for all except the lowest emission scenario (Fig. 3- 5). Strong local climate- change feedbacks explain why end- of- century OA in the Weddell Sea is comparable to that in other regions despite the delayed onset. While the area of the proposed Weddell Sea MPA has the lowest ice- free area of all MPAs in the 1990s (28% as compared to 48- 61% in the annual mean, 57% as compared to 84- 96% between December- February), all regions are virtually ice free in summer by 2100 in the highest- emission scenario (Fig. 5k- o). In the Weddell Sea, this strong sea- ice retreat facilitates substantially more oceanic \(\mathrm{CO_2}\) uptake than expected from the increase in atmospheric \(\mathrm{CO_2}\) levels alone (Supplementary Fig. 2). Besides the reduced capping effect of sea ice for air- sea exchange following the sea- ice retreat, this can be attributed to the lower buffer capacity resulting in more oceanic \(\mathrm{CO_2}\) uptake per primary production in an acidified ocean64. + +<|ref|>text<|/ref|><|det|>[[50, 444, 930, 920]]<|/det|> +Given the high biodiversity in MPAs, the projected severe OA will likely impact organisms on many trophic levels, ranging from primary producers46 to fish3,57, and calcium carbonate- forming benthic organisms (Fig. 1)4,59. Due to their immobility and often long lifespan65, the impact on the latter will likely be substantial both on the continental shelves and in the open ocean, i.e., downstream of dense- water formation regions, where the OA signal spills from the shelf sea into the deep ocean (Fig. 2). While our knowledge is limited about the possibility for adaptation and physiological acclimatization in each of these organisms over a multi- decadal time span63,66,67, a higher sensitivity of benthic organisms to changing conditions than of mobile pelagic organisms is conceivable due to the smaller variability in pH and \(\Omega\) at depth than in the surface layer (simulated monthly \(\Omega_{\mathrm{arg}}\) in the 1990s averaged over the Antarctic continental shelf spans 0.91- 0.93 and 1.32- 1.67 below 500 m and above 50 m, respectively). In the two highest emission scenarios, the projected pH is \(< 7.8\) for all MPAs (Fig. 1), a level at which many detrimental effects have been reported for primary producers, zooplankton, fish, and benthic organisms3,4,54- 58. Even though it remains unclear to what extent a longer acclimation of adult organisms in laboratory experiments can reduce the severe impacts reported for egg development and juveniles68, strong OA would almost certainly disrupt food webs, thereby even affecting organisms which might not directly be negatively impacted by OA themselves. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 930, 465]]<|/det|> +A disruption of Antarctic food webs seems even more likely when considering that OA is not the only marine ecosystem stressor exacerbated by anthropogenic climate change. High- latitude warming has been shown to aggravate OA impacts on Antarctic organisms \(^{53,55,69}\) , resulting in higher mortality and reduced hatching success of Antarctic dragonfish \(^{55}\) and increased resting metabolic rates in Antarctic juvenile pteropods \(^{53}\) than when considering OA alone. Further, warming could push cold- adapted organisms in Antarctic coastal waters beyond their thermal tolerance by 2100 (Supplementary Fig. 3) \(^{70}\) . While our model projections indicate oxygen concentrations above the hypoxic threshold of 61.3 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) in the Antarctic MPAs for all scenarios (Supplementary Fig. 3), the projected decline in oxygen concentrations might nonetheless negatively impact high- latitude ecosystems in conjunction with the projected warming and OA \(^{71}\) . The recent reduction in sea- ice cover due to the warming of waters around the Antarctic Peninsula has already negatively affected larval abundance of the Antarctic silverfish, which uses sea ice as spawning habitat \(^{72}\) , suggesting that the 21 \(^{\mathrm{st}}\) - century sea- ice retreat projected for all MPAs is worrisome (Fig. 5k- o). + +<|ref|>text<|/ref|><|det|>[[50, 472, 930, 890]]<|/det|> +The projected severity of OA is higher for Antarctic shelf waters than open- ocean waters, and \(21^{\mathrm{st}}\) - century OA across all emission scenarios within the MPAs is comparable to OA in continental shelf areas not protected within an MPA (Supplementary Fig. 4). This implies two things: Firstly, regions outside of MPAs could serve as an important reference area for differentiating the ecosystem impacts of fishing from climate change as well as for assessing the effectiveness of MPAs \(^{26,29,30}\) . Secondly and more importantly, a reduction of fishing activities in a larger fraction of Antarctic coastal waters is advisable, given that ecosystems in high- latitude waters outside of the MPAs are not spared from severe OA. Given the pressure on ecosystems from both fishing and environmental change \(^{26}\) , both the mitigation of climate change and the limitation of fishing activities (e.g., via a network of MPAs) are key to minimizing disruptions of Antarctic food webs over the 21 \(^{\mathrm{st}}\) century. Multiple stressors of fishing and climate change can cause cumulative impacts that are more devastating than the impacts of each stressor alone \(^{73}\) . Through maintaining all trophic levels of the ecosystem and increasing species and genetic diversity, MPAs can enhance resilience to environmental change \(^{74}\) . Thus, MPAs and the simultaneous reduction of global emissions are the best policy tools for safeguarding Antarctic marine ecosystems. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[58, 138, 224, 161]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[58, 186, 928, 236]]<|/det|> +1. Doney, S. C., Fabry, V. J., Feely, R. A. & Kleypas, J. A. Ocean acidification: the other \(\mathrm{CO_2}\) problem. Annual review of marine science 1, 169-92 (2009). + +<|ref|>text<|/ref|><|det|>[[58, 254, 928, 303]]<|/det|> +2. Orr, J. C. et al. Anthropogenic ocean acidification over the twenty-first century and its impact on calcifying organisms. Nature 437, 681-686 (2005). + +<|ref|>text<|/ref|><|det|>[[58, 322, 928, 400]]<|/det|> +3. Hancock, A. M., King, C. K., Stark, J. S., McMinn, A. & Davidson, A. T. Effects of ocean acidification on Antarctic marine organisms: A meta-analysis. Ecology and Evolution 10, 4495-4514 (2020). + +<|ref|>text<|/ref|><|det|>[[58, 419, 928, 468]]<|/det|> +4. Figuerola, B. et al. A Review and Meta-Analysis of Potential Impacts of Ocean Acidification on Marine Calcifiers From the Southern Ocean. Frontiers in Marine Science 8 (2021). + +<|ref|>text<|/ref|><|det|>[[58, 487, 928, 536]]<|/det|> +5. McNeil, B. I. & Matear, R. J. Southern Ocean acidification: A tipping point at 450-ppm atmospheric \(\mathrm{CO_2}\) . Proceedings of the National Academy of Sciences 105, 18860-18864 (2008). + +<|ref|>text<|/ref|><|det|>[[58, 555, 928, 605]]<|/det|> +6. Hauri, C., Friedrich, T. & Timmermann, A. Abrupt onset and prolongation of aragonite undersaturation events in the Southern Ocean. Nature Climate Change 6, 1-6 (2015). + +<|ref|>text<|/ref|><|det|>[[58, 624, 928, 701]]<|/det|> +7. Negrete-Garcia, G., Lovenduski, N. S., Hauri, C., Krumhardt, K. M. & Lauvset, S. K. Sudden emergence of a shallow aragonite saturation horizon in the Southern Ocean. Nature Climate Change 9, 313-317 (2019). + +<|ref|>text<|/ref|><|det|>[[58, 721, 928, 770]]<|/det|> +8. Anderson, L. G., Holby, O., Lindegren, R. & Ohlson, M. The transport of anthropogenic carbon dioxide into the Weddell Sea. Journal of Geophysical Research 96 (1991). + +<|ref|>text<|/ref|><|det|>[[58, 790, 928, 838]]<|/det|> +9. Chen, C.-T. A. Some indications of excess \(\mathrm{CO_2}\) penetration near Cape Adare off the Ross Sea. La Mer 32, 167-172 (1994). + +<|ref|>text<|/ref|><|det|>[[58, 858, 928, 907]]<|/det|> +10. Hoppema, M., Fahrbach, E., Stoll, M. H. & de Baar, H. J. Increase of carbon dioxide in the bottom water of the Weddell Sea, Antarctica. Marine Chemistry 59, 201-210 (1998). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 928, 152]]<|/det|> +11. Hoppema, M., Fahrbach, E., Richter, K.-U., de Baar, H. J. & Kattner, G. Enrichment of silicate and \(\mathrm{CO_2}\) and circulation of the bottom water in the Weddell Sea. Deep Sea Research Part I: Oceanographic Research Papers 45, 1797-1817 (1998). + +<|ref|>text<|/ref|><|det|>[[50, 173, 928, 250]]<|/det|> +12. Hoppema, M., Roether, W., Bellerby, R. G. J. & de Baar, H. J. W. Direct measurements reveal insignificant storage of anthropogenic \(\mathrm{CO_2}\) in the Abyssal Weddell Sea. Geophysical Research Letters 28, 1747-1750 (2001). + +<|ref|>text<|/ref|><|det|>[[50, 270, 928, 345]]<|/det|> +13. McNeil, B. I., Tilbrook, B. & Matear, R. J. Accumulation and uptake of anthropogenic \(\mathrm{CO_2}\) in the Southern Ocean, south of Australia between 1968 and 1996. Journal of Geophysical Research: Oceans 106, 31431-31445 (2001). + +<|ref|>text<|/ref|><|det|>[[50, 366, 928, 442]]<|/det|> +14. Lo Monaco, C., Metzl, N., Poisson, A., Brunet, C. & Schauer, B. Anthropogenic \(\mathrm{CO_2}\) in the Southern Ocean: Distribution and inventory at the Indian-Atlantic boundary (World Ocean Circulation Experiment line 16). Journal of Geophysical Research 110, C06010 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 463, 928, 511]]<|/det|> +15. Sandrini, S. et al. Anthropogenic carbon distribution in the Ross Sea, Antarctica. Antarctic Science 19, 395-407 (2007). + +<|ref|>text<|/ref|><|det|>[[50, 532, 928, 608]]<|/det|> +16. Hauck, J., Hoppema, M., Bellerby, R. G., Völker, C. & Wolf-Gladrow, D. Data-based estimation of anthropogenic carbon and acidification in the Weddell Sea on a decadal timescale. Journal of Geophysical Research: Oceans 115, 1-14 (2010). + +<|ref|>text<|/ref|><|det|>[[50, 629, 928, 705]]<|/det|> +17. van Heuven, S. M., Hoppema, M., Huhn, O., Slagter, H. A. & de Baar, H. J. Direct observation of increasing \(\mathrm{CO_2}\) in the Weddell Gyre along the Prime Meridian during 1973-2008. Deep Sea Research Part II: Topical Studies in Oceanography 58, 2613-2635 (2011). + +<|ref|>text<|/ref|><|det|>[[50, 726, 928, 802]]<|/det|> +18. Huhn, O., Rhein, M., Hoppema, M. & van Heuven, S. Decline of deep and bottom water ventilation and slowing down of anthropogenic carbon storage in the Weddell Sea, 1984-2011. Deep Sea Research Part I: Oceanographic Research Papers 76, 66-84 (2013). + +<|ref|>text<|/ref|><|det|>[[50, 823, 928, 900]]<|/det|> +19. van Heuven, S. M. A. C., Hoppema, M., Jones, E. M. & de Baar, H. J. W. Rapid invasion of anthropogenic \(\mathrm{CO_2}\) into the deep circulation of the Weddell Gyre. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 372, 20130056 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[47, 74, 928, 152]]<|/det|> +20. Shadwick, E. H., Tilbrook, B. & Williams, G. D. Carbonate chemistry in the Mertz Polynya (East Antarctica): Biological and physical modification of dense water outflows and the export of anthropogenic \(\mathrm{CO_2}\) . Journal of Geophysical Research: Oceans 119, 1-14 (2014). + +<|ref|>text<|/ref|><|det|>[[48, 172, 928, 249]]<|/det|> +21. Roden, N. P., Tilbrook, B., Trull, T. W., Virtue, P. & Williams, G. D. Carbon cycling dynamics in the seasonal sea-ice zone of East Antarctica. Journal of Geophysical Research: Oceans 121, 8749-8769 (2016). + +<|ref|>text<|/ref|><|det|>[[48, 268, 928, 345]]<|/det|> +22. Mahieu, L., Lo Monaco, C., Metzl, N., Fin, J. & Mignon, C. Variability and stability of anthropogenic \(\mathrm{CO_2}\) in Antarctic Bottom Water observed in the Indian sector of the Southern Ocean, 1978-2018. Ocean Science 16, 1559-1576 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 366, 928, 413]]<|/det|> +23. Ingrosso, G. et al. Physical and biological controls on anthropogenic \(\mathrm{CO_2}\) sink of the Ross Sea. Frontiers in Marine Science 9, 1-19 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 433, 928, 481]]<|/det|> +24. Arrigo, K. R., van Dijken, G. & Long, M. Coastal Southern Ocean: A strong anthropogenic \(\mathrm{CO_2}\) sink. Geophysical Research Letters 35, 1-6 (2008). + +<|ref|>text<|/ref|><|det|>[[48, 502, 928, 550]]<|/det|> +25. Tagliabue, A. & Arrigo, K. R. Decadal trends in air-sea \(\mathrm{CO_2}\) exchange in the Ross Sea (Antarctica). Geophysical Research Letters 43, 5271-5278 (2016). + +<|ref|>text<|/ref|><|det|>[[48, 571, 928, 618]]<|/det|> +26. Brooks, C. M. et al. Progress towards a representative network of Southern Ocean protected areas. PLOS ONE 15, e0231361 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 639, 928, 715]]<|/det|> +27. Rogers, A. et al. Antarctic Futures: An Assessment of Climate-Driven Changes in Ecosystem Structure, Function, and Service Provisioning in the Southern Ocean. Annual Review of Marine Science 12, 87-120 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 736, 928, 783]]<|/det|> +28. Brooks, C. M. et al. Protect global values of the Southern Ocean ecosystem. Science 378, 477-479 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 805, 928, 852]]<|/det|> +29. Trathan, P. & Grant, S. The South Orkney Islands Southern Shelf Marine Protected Area. In Marine Protected Areas, 67-98 (Elsevier, 2020). + +<|ref|>text<|/ref|><|det|>[[48, 874, 928, 921]]<|/det|> +30. Teschke, K. et al. Planning Marine Protected Areas under the CCAMLR regime – The case of the Weddell Sea (Antarctica). Marine Policy 124, 104370 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[47, 74, 928, 125]]<|/det|> +307. Smith, W. O., Ainley, D. G., Arrigo, K. R. & Dinniman, M. S. The Oceanography and Ecology of the Ross Sea. Annual Review of Marine Science 6, 469–487 (2014). + +<|ref|>text<|/ref|><|det|>[[48, 144, 928, 220]]<|/det|> +309. Nissen, C. & Vogt, M. Factors controlling the competition between Phaeocystis and diatoms in the Southern Ocean and implications for carbon export fluxes. Biogeosciences 18, 251–283 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 240, 927, 288]]<|/det|> +312. Mikhalevich, V. I. The general aspects of the distribution of Antarctic foraminifera. Micropaleontology 50, 179–194 (2004). + +<|ref|>text<|/ref|><|det|>[[48, 308, 927, 356]]<|/det|> +314. Atkinson, A. et al. Oceanic circumpolar habitats of Antarctic krill. Marine Ecology Progress Series 362, 1–23 (2008). + +<|ref|>text<|/ref|><|det|>[[48, 376, 927, 424]]<|/det|> +316. Hunt, B. et al. Pteropods in Southern Ocean ecosystems. Progress in Oceanography 78, 193–221 (2008). + +<|ref|>text<|/ref|><|det|>[[48, 444, 928, 550]]<|/det|> +318. Koubbi, P. et al. Conservation and Management of Antarctic Silverfish Pleuragramma antarctica Populations and Habitats. In Vacchi, M., Pisano, E., Ghigliotti, L. e. (ed.) The Antarctic Silverfish: a Keystone Species in a Changing Ecosystem. Advances in Polar Ecology, vol 3., 287–305 (Springer, Cham., 2017). + +<|ref|>text<|/ref|><|det|>[[48, 570, 928, 646]]<|/det|> +322. Ainley, D. G., Ballard, G. & Weller, J. Part I: Validation of the 2007 CCAMLR Bioregionalization Workshop Marine Protected Areas in the Southern Ocean. Tech. Rep., CCAMLR, WG-EMM-10/11, Hobart, Tasmania (2010). + +<|ref|>text<|/ref|><|det|>[[48, 666, 928, 743]]<|/det|> +325. Koubbi, P. et al. Estimating the biodiversity of the East Antarctic shelf and oceanic zone for ecoregionalisation: Example of the ichthyofauna of the CEAMARC (Collaborative East Antarctic Marine Census) CAML surveys. Polar Science 4, 115–133 (2010). + +<|ref|>text<|/ref|><|det|>[[48, 763, 928, 839]]<|/det|> +328. Ballard, G., Jongsomjit, D., Veloz, S. D. & Ainley, D. G. Coexistence of mesopredators in an intact polar ocean ecosystem: The basis for defining a Ross Sea Marine Protected Area. Biological Conservation 156, 72–82 (2012). + +<|ref|>text<|/ref|><|det|>[[48, 859, 928, 937]]<|/det|> +331. Ropert-Coudert, Y. et al. Two Recent Massive Breeding Failures in an Adélie Penguin Colony Call for the Creation of a Marine Protected Area in D'Urville Sea/Mertz. Frontiers in Marine Science 5, 1–7 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[47, 75, 928, 124]]<|/det|> +41. Brasier, M. J. et al. Benthic biodiversity in the South Orkney Islands Southern Shelf Marine Protected Area. Biodiversity 19, 1-15 (2018). + +<|ref|>text<|/ref|><|det|>[[48, 144, 928, 193]]<|/det|> +42. Mitchell, E. G., Whittle, R. J. & Griffiths, H. J. Benthic ecosystem cascade effects in Antarctica using Bayesian network inference. Communications Biology 3, 582 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 212, 928, 290]]<|/det|> +43. Cummings, V. J., Bowden, D. A., Pinkerton, M. H., Halliday, N. J. & Hewitt, J. E. Ross Sea Benthic Ecosystems: Macro- and Mega-faunal Community Patterns From a Multi-environment Survey. Frontiers in Marine Science 8 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 310, 928, 359]]<|/det|> +44. Griffiths, H. J. et al. Breaking All the Rules: The First Recorded Hard Substrate Sessile Benthic Community Far Beneath an Antarctic Ice Shelf. Frontiers in Marine Science 8 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 378, 928, 427]]<|/det|> +45. Purser, A. et al. A vast iefish breeding colony discovered in the Antarctic. Current Biology 32, 842-850.e4 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 446, 927, 495]]<|/det|> +46. Deppeler, S. L. & Davidson, A. T. Southern Ocean Phytoplankton in a Changing Climate. Frontiers in Marine Science 4 (2017). + +<|ref|>text<|/ref|><|det|>[[48, 515, 928, 592]]<|/det|> +47. Seifert, M., Rost, B., Trimborn, S. & Hauck, J. Meta-analysis of multiple driver effects on marine phytoplankton highlights modulating role of pCO2. Global Change Biology gcb.15341 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 612, 927, 660]]<|/det|> +48. Kawaguchi, S. et al. Risk maps for Antarctic krill under projected Southern Ocean acidification. Nature Climate Change 3, 843-847 (2013). + +<|ref|>text<|/ref|><|det|>[[48, 680, 927, 728]]<|/det|> +49. Ericson, J. A. et al. Adult Antarctic krill proves resilient in a simulated high CO2 ocean. Communications Biology 1, 190 (2018). + +<|ref|>text<|/ref|><|det|>[[48, 749, 928, 798]]<|/det|> +50. Kurihara, H. Effects of CO2-driven ocean acidification on the early developmental stages of invertebrates. Marine Ecology Progress Series 373, 275-284 (2008). + +<|ref|>text<|/ref|><|det|>[[48, 818, 928, 895]]<|/det|> +51. Manno, C., Sandrini, S., Tositti, L. & Accornero, A. First stages of degradation of Limacina helicina shells observed above the aragonite chemical lysocline in Terra Nova Bay (Antarctica). Journal of Marine Systems 68, 91-102 (2007). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[47, 75, 928, 125]]<|/det|> +52. Bednaršek, N. et al. Extensive dissolution of live pteropods in the Southern Ocean. Nature Geoscience 5, 881-885 (2012). + +<|ref|>text<|/ref|><|det|>[[48, 144, 928, 220]]<|/det|> +53. Hoshijima, U., Wong, J. M. & Hofmann, G. E. Additive effects of pCO₂ and temperature on respiration rates of the Antarctic pteropod Limacina helicina antarctica. Conservation Physiology 5, 1-14 (2017). + +<|ref|>text<|/ref|><|det|>[[48, 240, 928, 289]]<|/det|> +54. Manno, C., Peck, V. L. & Tarling, G. A. Pteropod eggs released at high pCO₂ lack resilience to ocean acidification. Scientific Reports 6, 25752 (2016). + +<|ref|>text<|/ref|><|det|>[[48, 309, 928, 387]]<|/det|> +55. Flynn, E. E., Bjelde, B. E., Miller, N. A. & Todgham, A. E. Ocean acidification exerts negative effects during warming conditions in a developing Antarctic fish. Conservation Physiology 3, 1-16 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 407, 928, 456]]<|/det|> +56. Davis, B. E. et al. Antarctic emerald rockcod have the capacity to compensate for warming when uncoupled from CO₂-acidification. Global Change Biology 24, e655-e670 (2018). + +<|ref|>text<|/ref|><|det|>[[48, 475, 928, 552]]<|/det|> +57. Todgham, A. E. & Mandic, M. Understanding the Metabolic Capacity of Antarctic Fishes to Acclimate to Future Ocean Conditions. Integrative and Comparative Biology 60, 1425-1437 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 572, 928, 650]]<|/det|> +58. Bylenga, C. H., Cummings, V. J. & Ryan, K. G. High resolution microscopy reveals significant impacts of ocean acidification and warming on larval shell development in Laternula elliptica. PLOS ONE 12, e0175706 (2017). + +<|ref|>text<|/ref|><|det|>[[48, 670, 928, 718]]<|/det|> +59. Ingels, J. et al. Possible effects of global environmental changes on Antarctic benthos: a synthesis across five major taxa. Ecology and Evolution 2, 453-485 (2012). + +<|ref|>text<|/ref|><|det|>[[48, 737, 928, 815]]<|/det|> +60. Nissen, C., Timmermann, R., Hoppema, M., Gurses, Ö. & Hauck, J. Abruptly attenuated carbon sequestration with Weddell Sea dense waters by 2100. Nature Communications 13, 3402 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 835, 928, 912]]<|/det|> +61. Nissen, C., Timmermann, R., Hoppema, M. & Hauck, J. A physical-biogeochemical tipping point cascade initiated on Weddell Sea continental shelves is avoidable in a 2°C scenario. Journal of Climate, in review xx, xx (2023). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[48, 74, 928, 125]]<|/det|> +62. Foster, T. D. & Carmack, E. C. Frontal zone mixing and Antarctic Bottom water formation in the southern Weddell Sea. Deep Sea Research and Oceanographic Abstracts 23, 301-317 (1976). + +<|ref|>text<|/ref|><|det|>[[48, 144, 928, 193]]<|/det|> +63. Kelly, M. W. & Hofmann, G. E. Adaptation and the physiology of ocean acidification. Functional Ecology 27, 980-990 (2013). + +<|ref|>text<|/ref|><|det|>[[48, 212, 928, 288]]<|/det|> +64. Hauck, J. & Völker, C. Rising atmospheric \(\mathrm{CO_2}\) leads to large impact of biology on Southern Ocean CO2 uptake via changes of the Revelle factor. Geophysical Research Letters 42, 1459-1464 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 308, 928, 387]]<|/det|> +65. Montero-Serra, I., Linares, C., Doak, D. F., Ledoux, J. B. & Garrabou, J. Strong linkages between depth, longevity and demographic stability across marine sessile species. Proceedings of the Royal Society B: Biological Sciences 285 (2018). + +<|ref|>text<|/ref|><|det|>[[48, 406, 928, 455]]<|/det|> +66. Berteaux, D., Réale, D., McAdam, A. G. & Boutin, S. Keeping Pace with Fast Climate Change: Can Arctic Life Count on Evolution? Integrative and Comparative Biology 44, 140-151 (2004). + +<|ref|>text<|/ref|><|det|>[[48, 474, 928, 524]]<|/det|> +67. Kapsenberg, L. & Cyronak, T. Ocean acidification refugia in variable environments. Global Change Biology 25, 3201-3214 (2019). + +<|ref|>text<|/ref|><|det|>[[48, 543, 928, 620]]<|/det|> +68. Suckling, C. C. et al. Adult acclimation to combined temperature and pH stressors significantly enhances reproductive outcomes compared to short-term exposures. Journal of Animal Ecology 84, 773-784 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 640, 928, 717]]<|/det|> +69. Saba, G., Bockus, A., Shaw, C. & Seibel, B. Combined effects of ocean acidification and elevated temperature on feeding, growth, and physiological processes of Antarctic krill Euphausia superba. Marine Ecology Progress Series 665, 1-18 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 736, 928, 815]]<|/det|> +70. Pörtner, H. O., Peck, L. & Somero, G. Thermal limits and adaptation in marine Antarctic ectotherms: an integrative view. Philosophical Transactions of the Royal Society B: Biological Sciences 362, 2233-2258 (2007). + +<|ref|>text<|/ref|><|det|>[[48, 834, 928, 912]]<|/det|> +71. Pörtner, H.-O. Oxygen- and capacity-limitation of thermal tolerance: a matrix for integrating climate-related stressor effects in marine ecosystems. Journal of Experimental Biology 213, 881-893 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[47, 74, 928, 152]]<|/det|> +72. Corso, A. D., Steinberg, D. K., Stammerjohn, S. E. & Hilton, E. J. Climate drives long-term change in Antarctic Silverfish along the western Antarctic Peninsula. Communications Biology 5, 104 (2022). + +<|ref|>text<|/ref|><|det|>[[48, 171, 928, 222]]<|/det|> +73. Grant, S. M. et al. Local Drivers of Change in Southern Ocean Ecosystems: Human Activities and Policy Implications. Frontiers in Ecology and Evolution 9 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 240, 927, 290]]<|/det|> +74. Roberts, C. M. et al. Marine reserves can mitigate and promote adaptation to climate change. Proceedings of the National Academy of Sciences 114, 6167-6175 (2017). + +<|ref|>text<|/ref|><|det|>[[48, 308, 928, 359]]<|/det|> +75. Wang, Q. et al. The Finite Element Sea Ice-Ocean Model (FESOM) v.1.4: formulation of an ocean general circulation model. Geoscientific Model Development 7, 663-693 (2014). + +<|ref|>text<|/ref|><|det|>[[48, 376, 927, 427]]<|/det|> +76. Danilov, S. et al. Finite-Element Sea Ice Model (FESIM), version 2. Geoscientific Model Development 8, 1747-1761 (2015). + +<|ref|>text<|/ref|><|det|>[[48, 445, 928, 496]]<|/det|> +77. Timmermann, R., Wang, Q. & Hellmer, H. Ice-shelf basal melting in a global finite-element sea-ice/ice-shelf/ocean model. Annals of Glaciology 53, 303-314 (2012). + +<|ref|>text<|/ref|><|det|>[[48, 514, 928, 564]]<|/det|> +78. Schaffer, J. et al. A global, high-resolution data set of ice sheet topography, cavity geometry, and ocean bathymetry. Earth System Science Data 8, 543-557 (2016). + +<|ref|>text<|/ref|><|det|>[[48, 582, 928, 633]]<|/det|> +79. Hauck, J. et al. Seasonally different carbon flux changes in the Southern Ocean in response to the southern annular mode. Global Biogeochemical Cycles 27, 1236-1245 (2013). + +<|ref|>text<|/ref|><|det|>[[48, 651, 928, 702]]<|/det|> +80. Karakus, O. et al. Modeling the Impact of Macrozooplankton on Carbon Export Production in the Southern Ocean. Journal of Geophysical Research: Oceans 126, 1-22 (2021). + +<|ref|>text<|/ref|><|det|>[[48, 720, 928, 771]]<|/det|> +81. Semmler, T. et al. Simulations for CMIP6 with the AWI Climate Model AWI-CM-1-1. Journal of Advances in Modeling Earth Systems 12, 1-34 (2020). + +<|ref|>text<|/ref|><|det|>[[48, 789, 928, 839]]<|/det|> +82. DeVries, T. et al. Magnitude, trends, and variability of the global ocean carbon sink from 1985-2018. Global Biogeochemical Cycles, in review xx, xx (2023). + +<|ref|>text<|/ref|><|det|>[[48, 857, 928, 908]]<|/det|> +83. Meinshausen, M. et al. Historical greenhouse gas concentrations for climate modelling (CMIP6). Geoscientific Model Development 10, 2057-2116 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[48, 73, 928, 430]]<|/det|> +436 O'Neill, B. C. et al. The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6. 437 Geoscientific Model Development 9, 3461- 3482 (2016). 438 85. Brooks, C. M. et al. The Ross Sea, Antarctica: A highly protected MPA in international waters. 439 Marine Policy 134, 104795 (2021). 440 86. Apelgren, N. & Brooks, C. M. Norwegian interests and participation towards the creation of 441 marine protected areas in the Southern Ocean. The Polar Journal 11, 393- 412 (2021). 442 87. Orr, J. C. & Epitalon, J.- M. Improved routines to model the ocean carbonate system: mocsy 443 2.0. Geoscientific Model Development 8, 485- 499 (2015). 444 88. Bednaršek, N., Tarling, G. A., Bakker, D. C. E., Fielding, S. & Feely, R. A. Dissolution Dom- 445 inating Calcification Process in Polar Pteropods Close to the Point of Aragonite Undersatu- 446 ration. PLoS ONE 9, e109183 (2014). + +<|ref|>sub_title<|/ref|><|det|>[[55, 483, 343, 508]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[48, 530, 930, 865]]<|/det|> +C.N. and M.H. acknowledge funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 820989 (project COMFORT). J.H. was supported by the Initiative and Networking Fund of the Helmholtz Association (Helmholtz Young Investigator Group Marine Carbon and Ecosystem Feedbacks in the Earth System [MarESys], grant number VH- NG- 1301) and R.T. by the Helmholtz Climate Initiative REKLIM (Regional Climate Change), a joint research project of the Helmholtz Association of German research centres (HGF). C.N. and N.S.L. are grateful for funding from the U.S. Department of Energy (DE- SC0022243). C.M.B. acknowledges support from the Pew Charitable Trusts, National Science Foundation and NASA. Computing resources were provided by the North- German Supercomputing Alliance (HLRN) project hbk00079. The work reflects only the authors' view; the European Commission and their executive agency are not responsible for any use that may be made of the information the work contains. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[50, 72, 358, 95]]<|/det|> +## 460 Author contributions + +<|ref|>text<|/ref|><|det|>[[50, 119, 928, 196]]<|/det|> +461 CN and NSL designed the study. CN ran the model simulations, with support from RT, MH, 462 and JH. CN performed the analysis. CMB provided the MPA masks. All authors contributed to 463 the interpretation of the results. CN wrote the manuscript, with input from all authors. + +<|ref|>sub_title<|/ref|><|det|>[[50, 238, 466, 262]]<|/det|> +## 464 Competing financial interests + +<|ref|>text<|/ref|><|det|>[[50, 286, 546, 305]]<|/det|> +465 The authors declare no competing financial interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[75, 72, 196, 95]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[50, 118, 930, 370]]<|/det|> +Description of FESOM- REcoM and all model experiments. For this study, we use the same model setup as Nissen et al. (2022) \(^{60}\) and Nissen et al. (2023, in review) \(^{61}\) . We will summarize its main features hereafter, but refer the reader to the other publications for further detail. All model experiments are conducted with the global Finite Element Sea ice Ocean Model (FESOM) version 1.4 \(^{75}\) , which includes a sea- ice \(^{76}\) and an ice- shelf component \(^{77}\) with constant ice- shelf geometry derived from RTopo- 2 \(^{78}\) . The biogeochemical cycles of carbon, nitrogen, silicon, iron, and oxygen are resolved by coupling FESOM to the Regulated Ecosystem Model version 2 (RE- coM2) \(^{79,80}\) . REcoM2 includes two phytoplankton groups (silicifying diatoms and a mixed small phytoplankton group) and two zooplankton groups and allows for variable stoichiometry. + +<|ref|>text<|/ref|><|det|>[[50, 375, 930, 650]]<|/det|> +The model equations are solved on a model grid with eddy- permitting resolution on Antarctic continental shelves ( \(< 5\mathrm{km}\) in southernmost parts; Supplementary Fig. 5). The grid resolution is lower in the open ocean ( \(< 100\mathrm{km}\) ). \(79\%\) and \(62\%\) of all surface model grid cells are south of \(60^{\circ}\mathrm{S}\) and \(70^{\circ}\mathrm{S}\) , respectively. The model grid has 99 z- levels in the vertical. At the surface, all simulations are forced with 3- hourly atmospheric output from the model experiments of the AWI Climate Model (AWI- CM) that contributed to the "Coupled Model Intercomparison Project Phase 6 (CMIP6)" \(^{81}\) . The model experiments are started in 1950 and initialized with output from the AWI- CM (physical tracers of FESOM) and with output from a simulation for the "Regional Carbon Cycle Assessment and Processes 2 (RECCAP2)" project (biogeochemical tracers of REcoM2) \(^{82}\) . + +<|ref|>text<|/ref|><|det|>[[50, 657, 930, 878]]<|/det|> +For this study, eight model simulations are analyzed (4x simA, simB, 2x simC, simD) as outlined in the following: Following a historical simulation from 1950- 2014, we simulate four emission scenarios for 2015- 2100: SSP1- 2.6 (lowest emission scenario), SSP2- 4.5, SSP3- 7.0, and SSP5- 8.5 (highest emission scenario). In each case, the ocean model receives fluxes of momentum, heat, and freshwater from the 3- hourly atmospheric output corresponding to the given scenario, and the atmospheric \(\mathrm{CO_2}\) boundary condition for the ocean model increases from 312 ppm in \(1950^{83}\) to 446 ppm, 603 ppm, 867 ppm, and 1135 ppm, respectively, by \(2100^{84}\) (simA). As such, simA represents the net carbon fluxes (the sum of the natural and anthropogenic components). + +<|ref|>text<|/ref|><|det|>[[50, 885, 928, 934]]<|/det|> +To isolate the anthropogenic carbon component, we conduct an additional simulation from 1950- 2100 with varying climate, but with a constant atmospheric \(\mathrm{CO_2}\) concentration bound + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 70, 930, 610]]<|/det|> +496 ary condition from the year 1950 (simD). The anthropogenic component of the dissolved inorganic carbon (DIC) pool (Canth) can then be computed as the difference between the DIC in 498 simA and that in simD. Due to computational constraints, simD is only available for the high499 emission scenario SSP5- 8.5. For easier comparability with existing observation- based estimates, 500 anthropogenic carbon concentrations are reported in \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) in the manuscript, and we use 501 monthly mean surface- referenced potential density to convert all model fields from the model 502 units (mmol \(\mathrm{m}^{- 3}\) ). We note that since simD was generated using the 1950 atmospheric \(\mathrm{CO}_{2}\) 503 concentration, the anthropogenic carbon quantified in this study is only that which has accumulated since 1950 and not since pre- industrial times. Thus, the resulting model- based values 504 of \(\mathrm{C}_{\mathrm{anth}}\) are likely biased low compared to observation- based estimates which include all an506 throgenic carbon. Yet, since most of the increase in atmospheric \(\mathrm{CO}_{2}\) levels occurred after 507 195083, our model- based \(\mathrm{C}_{\mathrm{anth}}\) estimate can be assumed to capture the vast majority of the sig508 nal. The lower simulated bottom \(\mathrm{C}_{\mathrm{anth}}\) concentrations in the East Antarctic open- ocean sector 509 in the 1990s (<1 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) ) than suggested by observations (up to 25 \(\mu \mathrm{mol} \mathrm{kg}^{- 1}\) 13,14,21,22) may 510 be attributed to the absence of shelf water masses which are dense enough to efficiently trans511 fer anthropogenic carbon to the abyss in this sector (not shown). However, we acknowledge 512 the difficulty a) to compare different observation- based estimates to each other due the uncer513 tainty arising from the use of various back- calculation techniques17 and b) to directly compare 514 the observation- based estimates from various years to the decadal- average model tracer. + +<|ref|>text<|/ref|><|det|>[[50, 614, 930, 920]]<|/det|> +515 To assess the model drift, we run a control simulation simB, in which both climate variability 516 and atmospheric \(\mathrm{CO}_{2}\) concentrations are held constant. For this experiment, the ocean model 517 receives atmospheric \(\mathrm{CO}_{2}\) from the year 1950 and repeated year 1955 momentum, heat, and 518 freshwater fluxes. This year was identified to best represent "normal" atmospheric conditions, 519 as determined from an assessment of the Southern Annular Mode and El Niño Southern Oscil520 lation in the AWI- CM model output (see Nissen et al., 202260 for details). Overall, the surface 521 drift in carbonate chemistry variables is minimal, whereas some subsurface drift exists in simB 522 (see Supplementary Fig. 6 for pH as an example). Any deep- ocean model drift can likely at 523 least partially be attributed to a delayed adjustment of the deep ocean to upper ocean carbonate 524 chemistry conditions. However, we note that since the model drift is generally much smaller 525 than the acidification signal in the different emission scenarios, it only has a small impact on our + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 928, 120]]<|/det|> +526 results (with the notable exception of the lowest- emission scenario for the proposed Antarctic Peninsula MPA; Supplementary Fig. 6). + +<|ref|>text<|/ref|><|det|>[[50, 130, 930, 352]]<|/det|> +Further, two complementary model experiments for 1950- 2100 are performed in which climate variability is held constant, while atmospheric \(\mathrm{CO_2}\) concentrations increase (simC). By computing \(\sin A\) minus \(\sin C\) of any model tracer or flux, we can isolate the effect of climate change on the projected changes in these. Due to computational limitations, \(\sin C\) is only available for the intermediate scenario SSP2- 4.5 and the high- emission scenario SSP5- 8.5. For this manuscript, we quantify the total climate effect, thus combining the impact of, e.g., warming and changes in circulation, sea- ice cover, freshwater input, and biological productivity on the projected model fields (see Supplementary Fig. 2). + +<|ref|>text<|/ref|><|det|>[[50, 365, 930, 928]]<|/det|> +Marine Protected Areas. Ocean acidification is assessed in five Marine Protected Areas (MPAs), of which two exist already (Ross Sea region and South Orkney Islands Southern Shelf) and three are proposed (Weddell Sea, East Antarctic, Antarctic Peninsula) \(^{85}\) . Masks on the model grid for the different MPAs are derived from previously published coordinates \(^{85,86}\) . For this study, we only consider the MPA area outside of the ice- shelf cavities, and "continental shelf" and "open ocean" are separated at the 2000 m isobath. Based on the model grid, the surface area in each MPA corresponds to 1.44 Mio. km \(^2\) for the Weddell Sea (53% on the continental shelf), 0.97 Mio. km \(^2\) for the East Antarctic (29% on the continental shelf), 1.59 Mio. km \(^2\) for the Ross Sea (45% on the continental shelf), 0.56 Mio. km \(^2\) for the Antarctic Peninsula (73% on the continental shelf), 0.11 Mio. km \(^2\) for the South Orkney Islands (0% on the continental shelf). As the South Orkney Islands Southern Shelf MPA is entirely defined as "open ocean" based on the model topography and as the vast majority of grid points for the Antarctic Peninsula MPA are defined as "continental shelf", we only report values for the whole MPA for these two MPAs in all Figures. In total, 60% of the continental shelf area outside of the ice- shelf cavities is covered by the five MPAs assessed in this study. The area- weighted average grid resolution outside of the ice- shelf cavities corresponds to 35 km in the Weddell Sea MPA (spread 7- 88 km; 7- 55 km for the continental shelf; Supplementary Fig. 5), 46 km in the East Antarctic MPA (spread 6- 93 km; 6- 44 km for the continental shelf), 54 km in the Ross Sea region MPA (spread 4- 138 km; 4- 75 km for the continental shelf), 30 km in the Antarctic Peninsula MPA (spread 2- 84 km; 2- 45 km for the continental shelf), and 72 km in the South Orkney Islands Southern Shelf MPA (spread 57- 85 km). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 75, 927, 120]]<|/det|> +We note that while there are between 1250- 3632 model grid cells in the four largest MPAs, only 22 model grid cells lie within the South Orkney MPA. + +<|ref|>text<|/ref|><|det|>[[50, 138, 928, 328]]<|/det|> +Calculation of carbonate chemistry. We calculate monthly mean pH and the saturation states of aragonite ( \(\Omega_{arag}\) ) and calcite ( \(\Omega_{calc}\) ) offline using the python routines to model the ocean carbonate system (mocsy v2.0) \(^{87}\) (code is available at https://github.com/jamesorr/mocsy, last access February, 1, 2023). As input parameters to the functions, we use monthly model output of dissolved inorganic carbon, alkalinity, potential temperature, salinity, silicic acid, phosphate (obtained from model nitrate field using the Redfield ratio 16), and sea- level pressure from the atmospheric output of the AWI- CM. + +<|ref|>sub_title<|/ref|><|det|>[[52, 371, 305, 395]]<|/det|> +## Data Availability + +<|ref|>text<|/ref|><|det|>[[50, 418, 928, 495]]<|/det|> +The model output underlying the findings of this study will be uploaded to Zenodo upon acceptance of the manuscript. The data can be made available to the editor and the reviewers at any time. + +<|ref|>sub_title<|/ref|><|det|>[[52, 537, 312, 561]]<|/det|> +## Code Availability + +<|ref|>text<|/ref|><|det|>[[50, 585, 990, 720]]<|/det|> +The Fortran source code of FESOM1.4- REcoM2 can be obtained via https://fesom.de/models/fesom14/ (last access January, 23, 2023). The analysis of model output was done with the open- source software Python. The routines to compute the carbonate chemistry offline from monthly model output in Python are available via https://github.com/jamesorr/mocsy (downloaded on February, 24, 2022). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 73, 925, 551]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[73, 562, 926, 741]]<|/det|> +
Figure 1: Key organisms in the Southern Ocean and \(21^{\mathrm{st}}\) -century ocean acidification in Marine Protected Areas. Map of established (South Orkney Islands Southern Shelf and Ross Sea region) and proposed (Weddell Sea, East Antarctic, Western Antarctic Peninsula) Marine Protected Areas (MPAs). Key species and groups which may be negatively impacted by ocean acidification (see references in main text) are depicted around the map (not to scale). Bar plots denote the top 2000 m average pH in the five different MPAs in the 1990s (dark grey) and in the 2090s (colors) for the four emission scenarios (sorted from low emission to high emission). For the Weddell Sea, East Antarctic, and Ross Sea MPAs, the hatched bars denote the average pH for the continental shelf south of the 2000 m isobath (grey contour on the map). Using this definition, the majority of the area of the proposed MPA along the western Antarctic Peninsula is continental shelf (see Methods), and bars are therefore hatched for this region. Note the arrows for the SSP5-8.5 scenario, indicating a lower pH on the continental shelves than for the whole-MPA average.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 234, 928, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[73, 590, 928, 768]]<|/det|> +
Figure 2: Distribution of anthropogenic carbon across the high-latitude Southern Ocean. Distribution of anthropogenic carbon concentrations in \(\mu \mathrm{mol} \mathrm{kg}^{-1}\) a averaged over the top \(2000 \mathrm{m}\) and b at the bottom in the 1990s of the historical simulation. c-d Same as a-b, but for the 2090s of the high-emission scenario SSP5-8.5. The top \(2000 \mathrm{m}\) are equivalent to the whole water column on the continental shelves, defined as the area south of the \(2000 \mathrm{m}\) isobath (grey contour). Note the logarithmic colorbar and the one order of magnitude difference between the 1990s and the 2090s. Outlines of the Marine Protected Areas (MPAs) are denoted in white, and the ice-shelf front is shown as the dark grey contour. e-g Temporal evolution of anthropogenic carbon concentrations with depth and averaged over the continental shelves of e the proposed Weddell Sea MPA, f the proposed East Antarctic MPA, and g the adopted Ross Sea region MPA. i-k Same as e-g, but for the open ocean. h and l show the averages for h the proposed MPA along the western Antarctic Peninsula and l the adopted South Orkney Islands Southern Shelf MPA.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[170, 75, 844, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[73, 730, 927, 927]]<|/det|> +
Figure 3: Changes in the vertical distribution of \(\Omega_{\mathrm{arg}}\) . Vertical profiles of a pH and b the saturation state with respect to aragonite \(\left(\Omega_{\mathrm{arg}}\right)\) in the proposed Weddell Sea Marine Protected Area (MPA). Solid lines show the profiles above 2000 m in the whole MPA, and dashed lines show the profile for the continental shelves south of the 2000 m isobath (see dark grey contour in map). The black lines denote the 1990s in the historical simulation, and the colored lines the 2090s in the four emission scenarios. The profiles on the left in each panel show absolute values, and profiles on the right show the change in each property between the 2090s and the 1990s \(\left(\Delta \mathrm{pH}\right.\) and \(\Delta \Omega_{\mathrm{arg}}\) ). In b, undersaturated conditions, i.e., \(\Omega_{\mathrm{arg}}< 1\) , are highlighted with the colored background, and the depth of \(\Omega_{\mathrm{arg}} = 1\) for the profile of the whole MPA is highlighted with a circle on the y axis. c-j Same as a-b, but for c-d the proposed East Antarctic MPA, e-f the adopted Ross Sea region MPA, g-h the proposed Antarctic Peninsula MPA, i-j the adopted South Orkney Islands Southern Shelf MPA. Note that only the vertical profiles for the whole MPA are shown in panels g-j (see Methods).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[87, 250, 927, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[73, 525, 927, 753]]<|/det|> +
Figure 4: Changes in the volume distribution across different \(\Omega_{\mathrm{arg}}\) and \(\Omega_{\mathrm{calc}}\) classes. a-b Distribution of waters across different classes of the saturation state with respect to aragonite \((\Omega_{\mathrm{arg}})\) a above \(2000\mathrm{m}\) in the whole proposed Weddell Sea Marine Protected Area (MPA) and b on the continental shelves south of the \(2000\mathrm{m}\) isobath (see dark grey contour in map). Results are shown for the 1990s of the historical simulation and for the 2090s in the four emission scenarios. c-d Same as a-b, but for the distribution of waters across different classes of the saturation state with respect to calcite \((\Omega_{\mathrm{calc}})\) . Note the different classes of saturation states shown for aragonite and calcite. While the 0.65 threshold for \(\Omega_{\mathrm{arg}}\) was chosen to best highlight the stronger undersaturation on the shelf than in the whole MPA for the highest emission scenario, we assume that it is more costly for an organism to sustain calcification at \(\Omega_{\mathrm{arg}}< 0.65\) than at \(0.65\leq \Omega_{\mathrm{arg}}< 1.0\) (ref. \(^{88}\) ) and that this division of the volume of undersaturated waters is therefore meaningful. e-l Same as a-d, but for e-h the proposed East Antarctic MPA and i-l the adopted Ross Sea region MPA. m-p Same as a and c, but for m-n the proposed Antarctic Peninsula MPA and o-p the adopted South Orkney Islands Southern Shelf MPA. Note that only the distribution for the whole MPA is shown in panels m-p (see Methods).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 93, 884, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[73, 729, 927, 909]]<|/det|> +
Figure 5: Temporal evolution of \(\Omega_{\mathrm{arg}}\) and sea ice cover. a Temporal evolution of the depth of the \(\Omega_{\mathrm{arg}} = 1\) isoline in the proposed Weddell Sea Marine Protected Area (MPA) for the SSP5-8.5 scenario (dark blue), the SSP3-7.0 scenario (intermediate blue), and the SSP2-4.5 scenario (light blue). The different line styles represent the annual mean (solid), the June-August mean (JJA; dotted), and the December-March mean (DJF; dashed). The SSP1-2.6 is not shown because the \(\Omega_{\mathrm{arg}} = 1\) isoline never reaches the top \(80 \mathrm{~m}\) of the water column. b-e Same as a, but for b the proposed East Antarctic MPA, c the adopted Ross Sea region MPA, d the proposed Antarctic Peninsula MPA, and e the adopted South Orkney Islands Southern Shelf MPA. f-j Temporal evolution of the \(50\%\) (thick line) and \(90\%\) (thin line) isolines of the volume with \(\Omega_{\mathrm{arg}} < 1\) in the upper \(50 \mathrm{~m}\) of the water column in the different MPAs. Colors show the different emission scenarios (see above). k-o Monthly ice-free surface area in the different MPAs for the 1990s (black) and the 2090s in the four emission scenarios (colors).
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 92, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 128, 548, 149]]<|/det|> +- Nissen2023COMFORT3SUPPLEMENTv20230330.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/images_list.json b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..153591817c240e54a646662fd2fa496e2d820047 --- /dev/null +++ b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Figures", + "footnote": [], + "bbox": [ + [ + 137, + 87, + 830, + 440 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Device characteristics and mechanism of AFefET neurons. a Planer structure of AFefET device. b The cross-sectional image of AFefET device. c The elemental mapping of the materials in the system for Hf, Zr, Ti, N, and W, respectively. d The line scan EDS of the device cross-section, which corresponds to the red arrow in Fig. 2b. e The HRTEM image of the AFE films. f The crystal structure of the domain, which corresponds to the white square area in Fig. 2e. g The FFT image of the crystalline domain in Fig. 2f. The measured angles and distances between relative crystal faces, and diffraction spots suggest the existence of [0-10] oriented tetragonal \\(P4_{2} / nmc\\) structure. h The integrate-and-fire dynamics of AFefET neuron. At stage 1, the electric field-induced FE domain nucleates, and the electrons begin to accumulate in the channel of AFefET. With the further gate pulse applied (stage 2), the electric field-induced FE domains grow and expand, resulting in more", + "footnote": [], + "bbox": [ + [ + 125, + 135, + 835, + 540 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. AFrFET LIF neuron device characteristics. The dynamic of the strength-modulated integration process of AFrFET neuron under a different gate pulse amplitudes (1.2 V-1.7 V amplitude, fixed \\(100\\mu \\mathrm{s}\\) interval, fixed \\(100\\mu \\mathrm{s}\\) width), b different gate pulse widths (fixed 1.7 V amplitude, fixed \\(100\\mu \\mathrm{s}\\) interval, \\(10\\mu \\mathrm{s} - 150\\mu \\mathrm{s}\\) width), c different gate pulse intervals (fixed 1.7 V amplitude, \\(50\\mu \\mathrm{s} - 600\\mu \\mathrm{s}\\) interval, fixed \\(100\\mu \\mathrm{s}\\) width). When the input pulse intensity is not enough, the AFrFET neuron can not fire. Statistical data of input", + "footnote": [], + "bbox": [ + [ + 123, + 237, + 844, + 702 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Network-level performance of AFoFET neuron. a Schematic of two-layer fully ferroelectric SNN (784x400x10) for classifying MNIST datasets and the proposed hardware implementation of the network based on FeFET synapses and AFoFET neurons. b The inference accuracy as a function of training epochs with ideal LIF neurons and AFoFET", + "footnote": [], + "bbox": [ + [ + 123, + 442, + 848, + 768 + ] + ], + "page_idx": 25 + } +] \ No newline at end of file diff --git a/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41.mmd b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a71fdc598fc62c6c302f37455e718f069e45b0a5 --- /dev/null +++ b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41.mmd @@ -0,0 +1,376 @@ + +# Compact artificial neuron based on anti-ferroelectric transistor + +Rongrong Cao Institute of Microelectronics, Chinese Academy of Sciences + +Xumeng Zhang Frontier Institute of Chip and System, Fudan University + +Sen Liu Chinese Academy of Sciences + +Jikai Lu Chinese Academy of Sciences + +Yongzhou Wang National University of Defense Technology + +Yang Yang Chinese Academy of Sciences + +Yize Sun Chinese Academy of Sciences + +Wei Wei Chinese Academy of Sciences + +Jianlu Wang Fudan University + +Hui Xu National University of Defense Technology + +Qingjiang Li Qi Liu ( liuqi@ime.ac.cn ) Fudan University https://orcid.org/0000- 0001- 7062- 831X + +## Article + +Keywords: spiking neural networks, anti- ferroelectric field- effect transistor + +Posted Date: October 1st, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 927008/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on November 17th, 2022. See the published version at https://doi.org/10.1038/s41467-022-34774-9. + +<--- Page Split ---> + +# Compact artificial neuron based on anti-ferroelectric transistor + +Rongrong Cao \(^{1,4}\) , Xumeng Zhang \(^{2,4}\) , Sen Liu \(^{1,4}\) , Jikai Lu \(^{3}\) , Yongzhou Wang \(^{1}\) , Yang Yang \(^{3}\) , Yize Sun \(^{3}\) , Wei Wei \(^{3}\) , Jianlu Wang \(^{2}\) , Hui Xu \(^{1}\) , Qingjiang Li \(^{1*}\) , Qi Liu \(^{2*}\) + +\(^{1}\) College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China + +\(^{2}\) Frontier Institute of Chip and System, Fudan University, No. 2005 Songhu Road, Yangpu District, Shanghai 200438, China + +\(^{3}\) Key Laboratory of Microelectronic Devices & Integrated Technology Institute of Microelectronics, Chinese Academy of Sciences, No. 3 Beitucheng West Road, Chaoyang District, Beijing 100029, China + +\(^{4}\) These authors contributed equally: Rongrong Cao, Xumeng Zhang and Sen Liu + +\(^{*}\) Corresponding author. E- mail: qingjiangli@nudt.edu.cn, qj_liu@fudan.edu.cn + +## Abstract + +Neuromorphic machines based on spiking neural networks (SNNs) provide a more fascinating platform over traditional computers on building energy- efficient intelligent systems, in which spiking neuron are pivotal components. Recently, memristive neurons, with promising bio- plausibility and density, have been developed, but with limited reliability or bulky capacitors for integration or additional circuits for reset. Here, we propose a novel anti- ferroelectric field- effect transistor (AFefET) neuron based on the inherent polarization and depolarization of \(\mathrm{Hf}_{0.2}\mathrm{Zr}_{0.8}\mathrm{O}_{2}\) anti- ferroelectric film to meet these challenges. In this neuron, the intrinsic polarization accumulation effect in the \(\mathrm{Hf}_{0.2}\mathrm{Zr}_{0.8}\mathrm{O}_{2}\) film increases the channel current of AFefET gradually, which implements the integration feature of the neuronal membrane and avoids using external capacitors. Also, the spontaneous depolarization effect in AFefET emulates the leaky behavior of neurons, saving the hardware overhead of neuron circuits by getting rid of external reset circuits. Moreover, the AFefET neuron exhibits other comprehensive merits, such as low energy + +<--- Page Split ---> + +consumption (37 fJ/spike), excellent endurance ( \(>10^{12}\) ), high uniformity and high stability. Using such an AFeFET neuron, we further construct a two- layer fully ferroelectric (784x400x10) SNN combining established FeFET synapse, achieving \(96.8\%\) recognition accuracy on MNIST datasets. This work opens the way to emulate spiking neurons with antiferroelectric materials and provides a more competitive approach to build high efficient neuromorphic hardware systems. + +## Introduction + +In the past few decades, neuromorphic computing, mimicking the human brain's architecture and operation with electronic devices, has attracted great interest due to its high biomimetic and high energy efficiency \(^{1 - 3}\) . Artificial neurons are the core components of neuromorphic computing implementation, emulating biological neurons functions of potential accumulation and firing \(^{4,5}\) . For the hardware implementation of neurons, hardware overhead, energy efficiency, reliability are the critical evaluation criteria \(^{5,6}\) . Yet, current hardware demonstrations of neurons struggle to satisfy these key metrics simultaneously. + +Generally, the complementary metal- oxide- semiconductor (CMOS) circuit is the most mature and stable scheme for emulating biological neurons. Nevertheless, due to the lack of intrinsic biological resemblance and the complexity of circuits, CMOS neurons face many challenges in density or energy efficiency \(^{7 - 9}\) . Recently, various emerging devices have been extensively explored to emulate biological neurons benefiting from their biological resemblance and scalability. Memristive neurons have trigged the most interests among them, including redox memristors \(^{10 - 13}\) , Mott memristors \(^{14 - 19}\) , phase- change memristors (PCM) \(^{20 - 22}\) , magnetic random access memory (MRAM) \(^{23,24}\) , etc. These neurons utilize the gradual switching of conductance to mimic membrane potential evolution, successfully emulating essential biological neuron functions with low hardware cost. However, high electroforming voltage and limited reproducibility due to temporal and spatial variations are + +<--- Page Split ---> + +still open questions25. In addition, capacitors are usually needed to realize the integration in memristive neurons, which limits their practical applications in large- scale neuromorphic computing systems. In the very recent research, novel ferroelectric polarization- based neurons are proposed and experimentally demonstrated26- 30. They utilize gradual polarization to mimic the integration process of biological neurons without additional capacitors. Moreover, polarization is the intrinsic property of ferroelectric materials, which is recognized to be reproducible, reliable, and energy- efficient27,31. These features are promising to implement excellent neurons. However, ferroelectric devices are nonvolatile, and thus need a feedback path26- 28 or a special design of ferroelectric layer29,30 to achieve spontaneous reset after firing. The feedback path will increase the hardware cost and energy consumption of neuron implementation. In addition, it will increase the complexity of the operation, as each new input must wait for the completion of the previous reset process, especially in the system with a rate coding scheme. Thus, demonstrating an ideal electronic device that processes advanced and balanced neuronal performance without additional capacitors and reset feedback path deserves more attention. + +In this work, for the first time, we report a leaky integrate- and- fire (LIF) neuron based on a CMOS- compatible AFefET. The intrinsic polarization/depolarization processes of the \(\mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFefET successfully emulate the integrate/leaky neuronal functions without any capacitors and reset peripheral circuits. Furthermore, attributing to the plentiful merits of ferroelectric materials, AFefET neuron exhibits many superiorities: electroforming- free, ultra- low energy consumption (37 fJ/spike), high endurance ( \(>10^{12}\) ), small cycle- to- cycle variation (as low as \(3.93\%\) ) and device- to- device variations ( \(7.57\%\) ). Also, we present that the temporal integration speed in such an AFefET neuron depends on the intensity of postsynaptic potential, illustrating the fundamental features for performing classification tasks. Subsequently, we demonstrate a two- layer SNN with full- ferroelectric architecture + +<--- Page Split ---> + +for learning and recognizing the MNIST datasets by simulation, obtaining \(96.8\%\) recognition accuracy comparable to ideal neurons. These results demonstrate that the proposed AFefET neuron is a competitive candidate for constructing neuromorphic systems. + +## Results + +## Volatile AFefET as LIF neurons + +Figure 1a shows the architecture and processing model of biological neurons. In specific, neuronal dendrites receive input spike information from pre- neurons and transmit it to soma. Then soma integrates information and triggers an action potential when the membrane potential reaches a threshold value. The axon transmits the generated action potentials to post- neurons, and the membrane potential depolarizes to a resting state5. The increase/decrease of membrane potential corresponds to the opening or shutting of \(\mathrm{Na}^{+} / \mathrm{K}^{+}\) channels, corresponding to three stages in Fig. 1b. Here, the dynamic process of membrane potential can be mimicked vividly by the intrinsic polarization/depolarization of antiferroelectric (AFE) materials. Under a silent state, AFE materials have spontaneous polarizations, but the orientations of adjacent dipoles are opposite, resulting in zero net macroscopic remanent polarization, as shown in the insert of Fig. 1c. However, the dipoles can be aligned by sufficient electric field, and the phase switches from AFE to ferroelectric (FE)32. Usually, the electric field- induced FE phase is not stable, which will recover to AFE phase when the electric field is released. Thus, AFE materials exhibit volatile characteristics and representative double hysteresis ( \(\mathrm{P}_{\mathrm{r}} \approx 0\) at \(0 \mathrm{MV / cm}\) ) as shown in Fig. 1c. We establish the dynamic relation between the intrinsic volatile characteristic of AFE and integrate/leaky neuronal functions by constructing an AFefET, which is integrated by an AFE capacitor ( \(\mathrm{TiN / Hf}_{0.2} \mathrm{Zr}_{0.8} \mathrm{O}_{2} / \mathrm{TiN}\) ) and a MOSFET (see Methods for the details of fabrication processes), as shown in the inset of Fig. 1d. Fig. 1d shows that the typical transfer characteristic of AFefET exhibits volatility, which differs from that of nonvolatile FeFET + +<--- Page Split ---> + +at \(\mathrm{V_G = 0}\) (Supplementary Fig. 1). + +The volatility of AFeFET emulates the self- recovery of biological neurons, which helps avoid external peripheral reset circuits. Furthermore, the AFeFET device is electroforming- free, which saves an additional high voltage forming circuit. In order to investigate the LIF function of AFeFET, continuous gate pulses (100 \(\mu \mathrm{s}\) width, 100 \(\mu \mathrm{s}\) interval, 1.5 V amplitude) representing postsynaptic potentials are applied to the gate of the AFeFET. The corresponding drain current \((\mathrm{I_d})\) representing the membrane potential demonstrates replicable LIF behavior under gate pulse trains (Fig. 1e). The \(\mathrm{I_d}\) increases under the excitatory spike trains and the neuron fires when \(\mathrm{I_d}\) reaches a threshold (1 \(\mu \mathrm{A}\) ). After firing, \(\mathrm{I_d}\) decreases spontaneously, and the channel switches off in a millisecond of free time eventually, which means the AFeFET neuron recovers and gets prepared for the next firing. Thanks to these merits of the AFeFET, only one AFeFET device is enough to emulate the integration and recovery process of neurons. + +## Device characteristics and mechanism of the AFeFET + +In this work, the volatile characteristics of the AFeFET neuron are dominated by the composition of zirconium in \(\mathrm{Hf_xZr_{1 - x}O_2}\) film. The \(\mathrm{Hf_xZr_{1 - x}O_2}\) exhibits paraelectric- FE- AFE transition with increasing the concentration of zirconium element (0%- 100%) (Supplementary Fig. 2). Actually, \(\mathrm{HfO_2}\) and \(\mathrm{ZrO_2}\) exhibit very similar physical and chemical properties, such as crystal phases, lattice parameters except the dielectric properties. The pure \(\mathrm{HfO_2}\) shows linear dielectric characteristics under electric field due to the centrosymmetric monoclinic structure33. The FE properties occur with the increasing zirconium content in doped \(\mathrm{HfO_2}\) , which is induced by the existence of a non- centrosymmetric \(o\) - phase structure. The hafnium- rich ferroelectric \(\mathrm{Hf_xZr_{1 - x}O_2}\) oxides exhibit nonvolatility34- 36, thus generally serve as memory materials. With further increasing the zirconium content, the volatile AFE properties occur in zirconium- rich \(\mathrm{Hf_xZr_{1 - x}O_2}\) + +<--- Page Split ---> + +oxides. The polarization of AFE can be triggered by an electric field and increases under a higher electric field. But it can still revert to the initial state as the applied electric field is removed (Supplementary Fig. 3). Usually, the polarization of AFE can be ascribed to the phase transition from AFE to FE phase under the influence of electric field. The electric field- induced phase transition is always accompanied by a large- volume change37. When the electric field is released, the induced FE phase will recover to the AFE phase due to the strains resulting from volume expansion35,38,39. As a result, the zirconium- rich \(\mathrm{Hf_xZr_{1-x}O_2}\) oxides exhibit intrinsic volatility. This is the charm of AFE materials used for constructing artificial neurons. + +Figures 2a and 2b show the plane structure and the detailed cross- sectional image of the AFeFET neuron. According to these images, the structure of the AFeFET neuron can be observed clearly, in which an AFE capacitor (yellow square) integrates on the gate of a conventional MOSFET. The energy- dispersive X- ray spectroscopy (EDS) mapping and line scan EDS were performed to further identify the elements and structure of AFeFET (Fig. 2c and 2d). The thickness of \(\mathrm{TiN / Hf_xZr_{1-x}O_2 / TiN}\) is \(40 \mathrm{nm} / 10 \mathrm{nm} / 40 \mathrm{nm}\) , and the interfaces of all layers are clean and flat. In addition, the Hf, Zr, Ti, N, W elements distribute uniform and are free of inter- diffused. Then we focus on investigating the characteristics of AFE layer due to its dominant role in neuronal behavior. The composition of AFE layer is controlled by alternate deposition (one cycle \(\mathrm{HfO_2}\) and four cycles \(\mathrm{ZrO_2}\) ), and is confirmed (hafnium: zirconium \(\approx 1:4\) ) by the peak areas and the relative sensitivity factors in X- ray photoelectron spectroscopy (XPS) results (Supplementary Fig. 4). The high- resolution transmission electron microscopy (HRTEM) image for the details of the AFE films is presented in Fig. 2e. The polycrystalline nature and the lattice fringes of different crystals can be observed clearly. Fig. 2f and 2g depict the crystal structure and corresponding fast Fourier transform (FFT) image of the white square area in Fig. 2e. The relative angle and + +<--- Page Split ---> + +distance between two lattice planes and diffraction spots indicate the existence of [0- 10]- oriented AFE tetragonal \(P4_{2} / nmc\) phase. In addition, the arrangement of zirconium atoms (green dots) is very regular, and the relative angle and lattice constants are measured directly as \(55.6^{\circ}\) , 3.49 Å, and 3.2 Å, respectively. These zirconium atoms parameters match the atomic model of [0- 10]- oriented plane of tetragonal \(P4_{2} / nmc\) phase exactly in Supplementary Fig. 5. These results confirm the existence of \(t\) - phase in AFE films, which is the foundation of the AFeFET neuron. + +In order to demonstrate the dynamic of AFeFET neuron, the mechanism is shown in Fig. 2h, which is related to the transformation between AFE and FE domains. At stage 1, several input pulses as postsynaptic signals are applied to the gate of AFeFET neuron, and the electric field- induced FE orthorhombic phase ( \(o\) - phase) domains nucleate, which transform from AFE tetragonal phase ( \(t\) - phase) domains under the gate pulse stimuli. The polarized charges accumulate in the AFE layer and modulate the channel resistance of MOSFET, resulting in the \(\mathrm{I_d}\) begins stepping up gradually. This process is just as the small portion of \(\mathrm{Na^{+}}\) channels opening. With further applying gate pulses, it comes to stage 2, at which the electric field- induced FE \(o\) - phase domains grow and expand. As a result, more attracted electrons accumulate in the channel of AFeFET, and the \(\mathrm{I_d}\) increases greatly, corresponding more \(\mathrm{Na^{+}}\) channels opening. Once the \(\mathrm{I_d}\) surpasses the threshold, the AFeFET neuron would fire. After firing, the electric field- induced FE \(o\) - phase domains transform back to AFE \(t\) - phase domains due to the release of gate pulse. Consequently, the attracted electrons discharge and the channel of AFeFET switches off, indicating the AFeFET neuron returns to resting potential (stage 3). This process corresponds to the opening of \(\mathrm{K^{+}}\) channels in biological neurons. Then, the AFeFET neuron fires again under another gate pulse stimuli. This repeatable and stable electric field- induced phase transition accounts for the intrinsic neuronal resemblance of AFeFET. The atomic- scale phase transition between \(t\) - phase and + +<--- Page Split ---> + +\(o\) - phase under the influence of electric field has been observed clearly by S. Lombardo et al. via in- situ HRTEM39. + +To present the gradual electric field- induced phase transition of AFE \(t\) - phase, we investigate the tendencies of \(\mathrm{I_d}\) under continuous gate pulses. Figure 2i shows the tendencies of \(\mathrm{I_d}\) under different gate pulse amplitude, while the gate pulse interval and width are fixed to \(100 \mu \mathrm{s}\) , respectively. Under the first 20 continuous gate pulse stimuli, an obvious integration process of \(\mathrm{I_d}\) can be observed. This resulted from the gradual formation of electric field- induced \(o\) - phase, which induces the electrons accumulation in the channel of AFeFET. The gate pulses with larger amplitude result in more \(o\) - phase domains formation and quicker growth of \(\mathrm{I_d}\) . With further gate pulse stimuli, the reversible domains tend to reach a saturation regime. This represents the dynamic balance between the electric field- induced phase transition and the recovery of the AFE \(t\) - phase, and the \(\mathrm{I_d}\) does not increase anymore. Compared between stimuli with different amplitudes, it is clearly that the higher pulse amplitude will lead to faster growth speed and a larger saturation value of \(\mathrm{I_d}\) , which is because that more AFE \(t\) - phase domains can be switched to FE \(o\) - phase domains. Noting that a similar tendency of \(\mathrm{I_d}\) can be observed under different gate pulse intervals and widths, as illustrated in Supplementary Fig. 6. Input stimuli pulses with shorter pulse intervals or wider widths induce faster integration speed and larger saturation value of \(\mathrm{I_d}\) . In all cases, the \(\mathrm{I_d}\) increases gradually and then tends to saturate corresponds to the gradual electric field- induced phase transition and saturation processes of AFE \(t\) - phase. In addition, the tendencies of \(\mathrm{I_d}\) also illustrate the intrinsic plasticity of neurons40, demonstrating that the AFeFET has high potentiality for hardware implementation of artificial LIF neurons. + +## Neuronal characteristics of the AFeFET + +To investigate the strength- modulated integration process of AFeFET neurons, we apply gate pulse stimuli with different amplitudes and widths to implement LIF neuron + +<--- Page Split ---> + +functions, as shown in Fig. 3a and 3b. As the input pulse intensity (amplitude/width) increasing, the AFefET neuron needs fewer input spikes to reach the threshold (1 \(\mu \mathrm{A}\) ), which indicates a higher firing rate under stronger stimuli strength. This is because more electric field- induced \(o\) - phases are formed under stronger pulse intensity, resulting in faster charge integration speed. Correspondingly, a higher \(\mathrm{I_d}\) of the AFefET neuron needs longer time to leak (Supplementary Fig. 7), which could be clarified as adaptive recovery. To further study how the leaky behavior influences the integration process, we measure the \(\mathrm{I_d}\) under gate pulse stimuli with different intervals (50 \(\mu \mathrm{s}\) - 600 \(\mu \mathrm{s}\) ), as shown in Fig. 3c. As the interval increasing, more input pulses are needed to integrate the \(\mathrm{I_d}\) to reach the threshold value. This is because that more charges are released during the free interval time, and more input pulses are required to compensate for that. It is worth noting that, when the interval time is wider enough, the \(\mathrm{I_d}\) cannot reach the threshold anyway. This feature represents the filtering capability of the neuron for weaker input signals, which is important in biological systems and neuromorphic systems41. To further evaluate the stability of the AFefET neuron, we extract the statistical data of input spike numbers for firing as a function of input amplitudes, as shown in Fig. 3d. The firing event needs fewer input spikes and tends to be more stable as the stimuli intensity increasing. This phenomenon exhibits that the AFefET neuron performs high precision computation under enough stimuli intensity, which is favorable for performing high precision tasks. A similar relationship between input spike numbers for firing and pulse widths (or intervals) is observed, as shown in Fig. 3e and 3f, respectively. To directly present the stability of the AFefET neuron, we calculate the standard deviation ( \(\sigma\) ) of integration pulse number under each stimuli condition and label them out in Fig. 3d- 3f. Furthermore, the cycle- to- cycle variation is calculated by dividing standard deviation ( \(\sigma\) ) by mean value ( \(\mu\) ). The lowest variation (3.93%) between cycles is obtained under 1.7 V gate pulse amplitude, 50 \(\mu \mathrm{s}\) interval, and 100 \(\mu \mathrm{s}\) width. It should be + +<--- Page Split ---> + +noted that optimizing the gate pulse parameters may further enhance the uniformity between cycles. These results demonstrate that the AFeFET neurons can successfully emulate the strength- modulated spike frequency characteristics of biological neurons with high stability, making the AFeFET neurons extremely capable of carrying out the classification tasks11,42. + +To further access the compatibility of the AFeFET neuron for implementing unsupervised learning, we investigated the lateral inhibitory property. During the accumulation of AFeFET neuron membrane potential, the excitability will be inhibited immediately when the AFeFET neuron receives inhibitory stimuli from adjacent neurons, as shown in Supplementary Fig. 8a and 8b. Moreover, the AFeFET neuron needs more excitatory inputs for the next firing under stronger lateral inhibition intensity (Supplementary Fig. 8c). This behavior is similar to the suppressive phenomenon in biological neurons between each other, which is valuable for performing competitive learning tasks. + +For artificial neurons, low energy consumption, high endurance, and high reliability are critical merits. To investigate the energy consumption of the AFeFET neuron, we performed a systematic analysis under different input pulse parameters and threshold values, as shown in Fig. 3g (extracted from Supplementary Fig. 9). The lowest energy consumption of \(37 \mathrm{fJ / per}\) spike can be obtained under the \(50 \mathrm{nA}\) threshold, \(1 \mu \mathrm{s}\) pulse width, and \(1 \mu \mathrm{s}\) pulse interval. Furthermore, the energy consumption decreases remarkably as the threshold and pulse width (interval) decrease. Thus, it is reasonable to infer that the energy consumption can be further reduced by decreasing the threshold and pulse width (interval). Moreover, the AFeFET neurons demonstrate considerable repeatability. Supplementary Fig. 10a shows \(5 \times 10^{5}\) stable firing cycles of the AFeFET neuron without any significant deterioration. In order to speed up the measurement, the AFE MIM structure, which is the endurance bottleneck of AFeFET, is measured for higher endurance ( \(10^{12}\) cycles) + +<--- Page Split ---> + +(Supplementary Fig. 10b). Based on the endurance measurements above, it is reasonable to believe the AFefET neuron could support more than \(10^{12}\) firing events (Fig. 3h). Fig. 3i shows the histograms of input spike numbers for firing, which are collected from 100 firing activities of each AFefET neuron. The required pulse number for firing is concentrated nearby 12, and the device- to- device variation is calculated to be as low as \(7.57\%\) . This variation is extracted from 500 firing activities in 5 AFefET neurons. As we claimed before, the uniformity could be further enhanced by optimizing the gate pulse parameters. These results indicate that the AFefET neuron has high uniformity and great potential in large- scale applications. For clearly presenting the comprehensive merits of our AFefET neuron, a benchmark comparison with other typical neurons based on emerging devices is summarized in Table 1. Considering the energy consumption, endurance, and hardware overhead, which are the critical evaluation criteria of the artificial neurons, the AFefET neuron exhibits the most attractive performances. + +## Network-level performance of AFefET neuron + +We have demonstrated that the AFefET neurons can provide better energy efficiency and higher uniformity compared to the other neurons based on emerging devices. It is also essential to evaluate the network- level performance using the AFefET neuron for the hardware implementation of SNNs. Subsequently, we construct a two- layer SNN (784×400×10) for classifying MNIST datasets, as shown in Fig. 4a. In this network, we adopt a time- to- first- spike coding method, in which all input neurons fire exactly one spike per stimulus, but the firing order carries information. A larger input corresponds to an earlier spike of the neuron. In the output layer, the first fired neuron determines the class of stimulus. As soon as one of the output neurons fires, the network assigns the corresponding category to the input, and the inference process stop. Thus, such a coding scheme is much more suitable for hardware implementation. The right panel of Fig. 4a presents the proposed + +<--- Page Split ---> + +hardware implementation of the network based on FeFET synapses and AFoFET neurons. During inference, the input signal is applied to the gates of FeFET synapses (WLs), pulsewidth modulators (PWMs) collect current on source lines (SLs) and convert to pulses with fixed amplitude and various widths. The outputs of PWMs serve as postsynaptic potentials and be applied to the gates of AFoFET neurons for performing the integration process. Then, we train such a network to learn MNIST datasets for illustrating the feasibility. During training, we adopt a supervised temporal backpropagation algorithm proposed by Kheradpisheh43. Figure 4b shows the training results with the AFoFET neurons under 1 \(\mu \mathrm{A}\) threshold, achieving \(\sim 95\%\) recognition accuracy. These results demonstrate that our AFoFET neurons have great potential to be used for fabricating SNNs chips. + +During training, the threshold value determines the number of integrated inputs (the number of membrane states) and thus affects the network performance. Then, the relation between the threshold value and network performance is further investigated as shown in Fig. 4c. It is noting that the recognition accuracy is related to the threshold, with the highest \(96.8\%\) accuracy under the 2 \(\mu \mathrm{A}\) threshold, which is nearly identical to ideal IF neurons. Nonetheless, the inference time increases with increasing the threshold value because more integration number is required to trigger neuron firing at higher threshold cases. Thus, there should be a trade- off between recognition accuracy and inference time, and an appropriate threshold value should be selected to balance the network performance in practical applications. Fortunately, the network can still achieve high accuracy ( \(>86\%\) ) even at a threshold value low to \(62.5 \mathrm{nA}\) , which is favorable for applications that need faster inference time but not rigorous accuracy. As we claimed before, the threshold also affects the energy consumption of the neuron, which is the key parameter for SNNs chip applications. We extract the spike number and energy consumption of the neurons in the system with different training thresholds to evaluate this feature, as shown in Fig. 4d. When the threshold is higher + +<--- Page Split ---> + +than \(0.125 \mu \mathrm{A}\) , the spike number in the hidden layer ( \(2^{\mathrm{nd}}\) layer) decreases with increasing the threshold. This is because the hidden neurons with a higher threshold are hard to fire. On the contrary, we observe that the total energy consumption of neurons decreases as the threshold decreases. When the threshold reduces to \(62.5 \mathrm{nA}\) , the spike number abruptly decreases to be less than 100. This is because under a low threshold value, the winner neuron in the output layer fires earlier and the network could finish the inference process faster. In that case, only a minority of neurons in the hidden layers fire, and thus the total neurons consume less energy. These results demonstrate that decreasing the threshold supports fast inference speed and could greatly decrease the energy consumption of hardware neurons. In addition, device variation is another important parameter in practical applications. Figure 4e presents the inference accuracy under various device variations after training under the \(1 \mu \mathrm{A}\) threshold. As the red dots are shown, the inference accuracy only decreases \(1\%\) even the variation increases to \(\pm 16.7\%\) , illustrating nearly no network performance degradation. These results further demonstrate that the proposed AFefET neuron is suitable for performing SNNs tasks and has great potential for the hardware implementation of SNNs chips. + +## Discussion + +SNNs, inspired by the human brain, are powerful platforms for enabling low- power event- driven neuromorphic hardware. In SNNs, spiking neurons are the key units that enable 'spikes', which exchange information through connected plastic synapses. With rich physical dynamics, memristive devices are considered promising devices to emulate spiking neurons. However, the high energy consumption or limited reliability hinders the applications of memristive neurons in neuromorphic computing. + +In this work, we demonstrated a leaky integrate- and- fire neuron based on an AFefET for the first time. The dynamic relationship between the intrinsic polarization/depolarization + +<--- Page Split ---> + +process of the \(\mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFeFET and integrate/leaky neuronal functions are successfully built. The AFeFET neuron features CMOS- compatible, tunable firing frequency, ultra- low hardware cost (no capacitance and additional reset circuit), ultra- low energy consumption (37 fJ/spike), high endurance ( \(>10^{12}\) ), and high uniformity among different cycles and devices, showing advanced overall performances compared with emerging devices- based neurons in literature. To verify the feasibility of the neuron, we constructed a two- layer SNN combined with FeFET synapses, achieving high recognition accuracy (96.8%), low energy consumption, and high robustness on MNIST datasets. These results demonstrate that the AFeFET neuron is a promising candidate for constructing high- efficient SNN systems and may promote the industrial landing of neuromorphic machines based on anti- ferroelectric materials. + +## Methods + +## Sample fabrication + +The fabrication processes of AFeFET neuron devices are as follows: 1) After ultraviolet lithography and lift- off process, the bottom electrode TiN (40 nm) was deposited on the gate terminal of the NMOS transistor by ion beam sputtering. The NMOS transistor was fabricated by \(0.18 \mu \mathrm{m}\) CMOS technology. 2) And then, \(10 \mathrm{nm} \mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFE thin films were deposited on \(40 \mathrm{nm}\) - thick TiN bottom electrode by atomic layer deposition (ALD) process at \(280^{\circ} \mathrm{C}\) substrate temperature. The \(\mathrm{Hf[N(C_2H_5)CH_3]}_4\) , \(\mathrm{Zr[N(C_2H_5)CH_3]}_4\) and \(\mathrm{H_2O}\) were used as hafnium precursor, Zr precursor and oxygen source, respectively. The hafnium/zirconium ratio was controlled by alternate deposition of one cycle \(\mathrm{HfO_2}\) and four cycles \(\mathrm{ZrO_2}\) . 3) Then after the ultraviolet lithography process, \(40 \mathrm{nm}\) - thick TiN was grown by an ion beam sputtering system and served as the top electrode. After the lift- off process, \(25 \times 25 \mu \mathrm{m}^2\) TiN top electrode was released. The two- terminal metal- insulator- metal + +<--- Page Split ---> + +structure was integrated on the gate of the NMOS transistor. 4) The fabricated device was annealed for 30 seconds at \(500^{\circ}\mathrm{C}\) in nitrogen atmosphere to crystallize. + +## Measurement method + +The element ratio is confirmed by X- ray photoelectron spectroscopy (ESCALAB 250Xi). The cross- section TEM, high- resolution TEM, energy dispersive spectroscopy and crystal structure were analyzed by transmission electron microscopy (FEI Tecnai TF- 20, UK). The DC mode is measured by Agilent B1500 semiconductor parameter analyzer. A B1530A fast measurement unit module was used for generating the voltage pulse and measure the response current at the same time. Capacitance- electric field (C- E) tests are performed with \(10\mathrm{kHz}\) AC probing frequency and \(30\mathrm{mV}\) amplitude by Agilent B1500. + +## SNN simulation method + +In this work, we use time- to- first- spike time coding to encode the input image into a sparse spike train. Each pixel of the input image is encoded into a single spike whose spiking time is inversely proportional to its pixel value. The dense input corresponds to earlier spiking time. And each input pixel will only generate one spike, resulting in sparser spike train than the rate coding method and significantly reducing energy consumption in hardware implementation. We constructed a \(784 \times 400 \times 10\) fully ferroelectric SNN for MNIST recognition based on such a coding method. FeFET synapses were considered during training, whose conductance was between \(5\mu \mathrm{S}\) and \(60\mu \mathrm{S}\) according to the experimental data in Ref.44. The neuron parameters were extracted from Fig. 3d and 3f. We calculated the energy consumption regarding the data in Fig. 3g under \(1\mu \mathrm{s}\) . When the neuron reaches the firing threshold, it will emit a spike to the subsequent layer. After emitting a spike, neurons will remain resting state until the end of the time window. In the output layer, the first spiking neuron determines the network decision. The simulation time steps are set to 256, where the time length of each time step is the read spike time length of + +<--- Page Split ---> + +the device (1 \(\mu \mathrm{s}\) ). The neuron threshold varies from \(62.5 \mathrm{nA}\) to \(2 \mu \mathrm{A}\) , the leakage time constant is \(300 \mu \mathrm{s}\) , and the device variation is \(10\%\) . Before training, the synapse weights are initialized randomly. In the training process, the target firing time of the correct output neuron is the earliest time that all neurons fire, and the target firing time of other neurons is set to be later than the earliest firing time. According to the defined error function, synapses of the fired neuron before the actual firing time will be updated. In order to update the weights of the hidden layer, a backpropagation algorithm is used to calculate the error of the hidden layer \(^{43}\) . During training, all time steps need to calculate the error function. In the inference process, the recognition result is obtained when the first spike is generated in the output layer, so there is no need to perform the later time step. Fewer time steps mean lower recognition latency and less energy consumption. By adjusting the neuron's threshold, the spike generation time can be adjusted, resulting in the adjustable recognition speed and energy consumption with acceptable accuracy loss. + +## Data availability + +The data that support the plots within this article and other findings of this study are available from the corresponding author upon reasonable request. + +## References + +1. Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. \*Nature\* 529, 484-489, (2016). +2. Zhang, W. et al. Neuro-inspired computing chips. \*Nat. Electron.\* 3, 371-382, (2020). +3. Merolla, P. A. et al. A million spiking-neuron integrated circuit with a scalable communication network and interface. \*Science\* 345, 668-673, (2014). +4. Kandel, E. R. et al. \*Principles of Neural Science\*. (McGraw-Hill, 2000). +5. Zhu, J. et al. A comprehensive review on emerging artificial neuromorphic devices. \*Appl. Phys. Rev.\* 7, 011312, (2020). +6. Wang, Z. et al. Resistive switching materials for information processing. \*Nat. Rev. Mater.\* 5, 173-195, (2020). + +<--- Page Split ---> + +7. Indiveri, G. et al. A VLSI array of low-power spiking neurons and bistable synapses with spike-timing dependent plasticity. IEEE Trans. Neural Netw. 17, 211-221, (2006). + +8. Indiveri, G. et al. Neuromorphic silicon neuron circuits. Front. Neurosci. 5, 73, (2011). + +9. Sourikopoulos, I. et al. A 4-fJ/spike artificial neuron in 65 nm CMOS technology. Front. Neurosci. 11, 123, (2017). + +10. Wang, Z. et al. Fully memristive neural networks for pattern classification with unsupervised learning. Nat. Electron. 1, 137-145, (2018). + +11. Zhang, X. et al. An artificial neuron based on a threshold switching memristor. IEEE Electron Device Lett. 39, 308-311, (2018). + +12. Wang, Z. et al. Capacitive neural network with neuro-transistors. Nat. Commun. 9, 3208, (2018). + +13. Lee, D. et al. Various threshold switching devices for integrate and fire neuron applications. Adv. Electron. Mater. 5, 1800866, (2019). + +14. Zhang, X. et al. Experimental demonstration of conversion-based SNNs with 1T1R Mott neurons for neuromorphic Inference. in 2019 IEEE International Electron Devices Meeting (IEDM) 6.7.1-6.7.4, (2019). + +15. Bo, Y. et al. \(\mathrm{NbO_2}\) memristive neurons for burst-based perceptron. Adv. Intell. Syst. 2, 2000066, (2020). + +16. Jerry, M. et al. Ultra-low power probabilistic IMT neurons for stochastic sampling machines. in 2017 Symposium on VLSI Circuits T186-T187, (2017). + +17. Yi, W. et al. Biological plausibility and stochasticity in scalable \(\mathrm{VO_2}\) active memristor neurons. Nat. Commun. 9, 4661, (2018). + +18. Stoliar, P. et al. A leaky-integrate-and-fire neuron analog realized with a Mott insulator. Adv. Funct. Mater. 27, 1604740, (2017). + +19. Pickett, M. D. et al. A scalable neuristor built with Mott memristors. Nat. Mater. 12, 114-117, (2013). + +20. Tuma, T. et al. Stochastic phase-change neurons. Nat. Nanotechnol. 11, 693-699, (2016). + +21. Cobley, R. A. et al. A self-resetting spiking phase-change neuron. Nanotechnology 29, 195202, (2018). + +22. Wright, C. D. et al. Beyond von-neumann computing with nanoscale phase-change memory devices. Adv. Funct. Mater. 23, 2248-2254, (2012). + +23. Sengupta, A. et al. Magnetic tunnel junction mimics stochastic cortical spiking neurons. Sci. Rep. 6, 30039, (2016). + +<--- Page Split ---> + +24. Wu, M.-H. et al. Extremely compact integrate-and-fire STT-MRAM neuron: a pathway toward all-spin artificial deep neural network. in 2019 Symposium on VLSI Technology T34-T35, (2019). + +25. Li, Y. et al. Anomalous resistive switching in memristors based on two-dimensional palladium diselenide using heterophase grain boundaries. Nat. Electron. 4, 348-356, (2021). + +26. Dutta, S. et al. Biologically plausible ferroelectric quasi-leaky integrate and fire neuron. in 2019 Symposium on VLSI Technology T140-T141, (2019). + +27. Dutta, S. et al. Supervised learning in all FeFET-based spiking neural network: opportunities and challenges. Front. Neurosci. 14, 634, (2020). + +28. Wang, Z. et al. Experimental demonstration of ferroelectric spiking neurons for unsupervised clustering. in 2018 IEEE International Electron Devices Meeting (IEDM) 13.13.11-13.13.14, (2018). + +29. Chen, C. et al. Bio-inspired neurons based on novel leaky-FeFET with ultra-low hardware cost and advanced functionality for all-ferroelectric neural network. in 2019 Symposium on VLSI Technology T136-T137, (2019). + +30. Luo, J. et al. Capacitor-less stochastic leaky-FeFET neuron of both excitatory and inhibitory connections for SNN with reduced hardware cost. in 2019 IEEE International Electron Devices Meeting (IEDM) 6.4.1-6.4.4, (2019). + +31. Salahuddin, S. et al. The era of hyper-scaling in electronics. Nat. Electron. 1, 442-450, (2018). + +32. Hao, X. et al. A comprehensive review on the progress of lead zirconate-based anti-ferroelectric materials. Prog. Mater Sci. 63, 1-57, (2014). + +33. Sang, X. et al. On the structural origins of ferroelectricity in \(\mathrm{HfO_2}\) thin films. Appl. Phys. Lett. 106, 162905, (2015). + +34. Muller, J. et al. Ferroelectricity in simple binary \(\mathrm{ZrO_2}\) and \(\mathrm{HfO_2}\) . Nano Lett. 12, 4318-4323, (2012). + +35. Park, M. H. et al. Ferroelectricity and antiferroelectricity of doped thin \(\mathrm{HfO_2}\) -based films. Adv. Mater. 27, 1811-1831, (2015). + +36. Müller, J. et al. Ferroelectric \(\mathrm{Zr_{0.5}Hf_{0.5}O_2}\) thin films for nonvolatile memory applications. Appl. Phys. Lett. 99, 112901, (2011). + +37. Zhang, S.-T. et al. High-strain lead-free anti-ferroelectric electrostriction. Adv. Mater. 21, 4716-4720, (2009). + +38. Pan, W. et al. Field-forced antiferroelectric-to-ferroelectric switching in modified lead zirconate titanate stannate ceramics. J. Am. Ceram. Soc. 72, 571-578, (1989). + +<--- Page Split ---> + +39. Lombardo, S. et al. Atomic-scale imaging of polarization switching in an (anti-)ferroelectric memory material: zirconia (ZrO₂). in 2020 IEEE Symposium on VLSI Technology 1-2, (2020). + +40. Baek, E. et al. Intrinsic plasticity of silicon nanowire neutronistsors for dynamic memory and learning functions. Nat. Electron. 3, 398-408, (2020). + +41. Khan, A. I. et al. The future of ferroelectric field-effect transistor technology. Nat. Electron. 3, 588-597, (2020). + +42. Burkitt, A. N. A review of the integrate-and-fire neuron model: I. Homogeneous synaptic input. Biol. Cybern. 95, 1-19, (2006). + +43. Kheradpisheh, S. R. & Masquelier, T. Temporal backpropagation for spiking neural networks with one spike per neuron. Int. J. Neural Syst. 30, 2050027, (2020). + +44. Jerry, M. et al. Ferroelectric FET analog synapse for acceleration of deep neural network training. in 2017 IEEE International Electron Devices Meeting (IEDM) 6.2.1-6.2.4, (2017). + +45. Li, Y. et al. High-uniformity threshold switching \(\mathrm{HfO_2}\) -based selectors with patterned Ag nanodots. Adv. Sci. 7, 2002251, (2020). + +46. Song, J. et al. Effects of liner thickness on the reliability of AgTe/TiO₂-based threshold switching devices. IEEE Trans. Electron Dev. 64, 4763-4767, (2017). + +47. Jeonghwan, S. et al. Threshold selector with high selectivity and steep slope for cross-point memory array. IEEE Electron Device Lett. 36, 681-683, (2015). + +48. Grisafe, B. et al. Performance enhancement of Ag/HfO₂ metal ion threshold switch cross-point selectors. IEEE Electron Device Lett. 40, 1602-1605, (2019). + +49. Chen, A. et al. Multi-functional controllable memory devices applied for 3D integration based on a single niobium oxide layer. Adv. Electron. Mater. 6, 1900756, (2019). + +50. Lin, C.-Y. et al. A high-speed MIM resistive memory cell with an inherent vanadium selector. Appl. Mater. Today 21, 100848, (2020). + +51. Kang, M. & Son, J. Off-state current reduction in NbO₂-based selector device by using TiO₂ tunneling barrier as an oxygen scavenger. Appl. Phys. Lett. 109, 202101, (2016). + +52. Sebastian, A. et al. Computational phase-change memory: beyond von Neumann computing. J. Phys. D: Appl. Phys. 52, 443002, (2019). + +53. Dorrance, R. et al. Scalability and design-space analysis of a 1T-1MTJ memory cell for STT-RAMs. IEEE Trans. Electron Dev. 59, 878-887, (2012). + +<--- Page Split ---> + +54. Vincent, A. F. et al. Spin-transfer torque magnetic memory as a stochastic memristive synapse for neuromorphic systems. IEEE Trans. Biomed. Circuits Syst. 9, 166-174, (2015). + +55. Wang, K. et al. Reconfigurable codesign of STT-MRAM under process variations in deeply scaled technology. IEEE Trans. Electron Dev. 62, 1769-1777, (2015). + +56. Cao, R. R. et al. Effects of capping electrode on ferroelectric properties of \(\mathrm{Hf}_{0.5}\mathrm{Zr}_{0.5}\mathrm{O}_{2}\) thin films. IEEE Electron Device Lett. 39, 1207-1210, (2018). + +## Acknowledgments + +This work was supported by the National Key R&D Program of China under Grant No. 2018YFA0701500, the National Natural Science Foundation of China under Grant Nos. 62004220, 62004219, 62074166, 61974164, 61804181, 61825404, 61732020, 61821091, and 61851402, the Strategic Priority Research Program of the Chinese Academy of Sciences under Grant XDB44000000 and the China Postdoctoral Science Foundation under Grant No. 2020M681167. The authors thank the valuable discussion from W. Wang, X. Li with National University of Defense Technology, Y. Li and Z.H. Wu with University of Chinese Academy of Sciences. + +## Author contributions + +R.R.C. and X.M.Z. designed the experiments. R.R.C., X.M.Z. and Y.Z.S. designed and fabricated the AFeFET devices. R.R.C. carried out the electrical experiments. X.M.Z., J.K.L. and Y.Z.W. conducted the simulation. R.R.C., Y.Y. and W.W. carried out the TEM and EDS test. S.L., J.L.W and Q.L. assisted with data analysis and interpretation. R.R.C., X.M.Z, and S.L. co- wrote the manuscript. All authors discussed the results and revised the manuscript. Q.J.L and Q.L. supervised the research. + +## Additional information + +Supplementary information accompanies this paper at... + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Figures
+ +Fig. 1. Leaky integrate-and-fire dynamic of biological neuron vs AFoFET artificial neurons. a The architecture and processing model of biological neurons. External information is perceived by dendrites and then transmitted to the soma for processing. When the membrane potential surpasses a threshold value, an action potential can be generated and transmitted to post- neurons through the axon. b The integrate and fire dynamics of a biological neuron. With the signal received by dendrites and transmitted to soma, few \(\mathrm{Na^{+}}\) channels are activated, and the membrane potential goes up gradually (stage 1). As the further signal is received, more \(\mathrm{Na^{+}}\) channels are activated quickly, and a large amount of \(\mathrm{Na^{+}}\) flows inside the membrane, inducing the membrane potential goes up rapidly (stage 2). Once the membrane potential exceeds a certain threshold, it decreases due to the inactivation of \(\mathrm{Na^{+}}\) channels and the opening of \(\mathrm{K^{+}}\) channels (stage 3). c The representative double hysteresis of AFE materials. The zero net macroscopic remanent polarization indicates the volatile characteristics of AFE materials. d The typical transfer curves of AFoFET, exhibiting counterclockwise hysteresis and volatility. The arrows indicate the switching + +<--- Page Split ---> + +sequences. e The continuous firing events of AFefET neuron. After several milliseconds of free time, the AFefET neuron can restart the LIF process and fire again. + +![](images/Figure_2.jpg) + +
Fig. 2. Device characteristics and mechanism of AFefET neurons. a Planer structure of AFefET device. b The cross-sectional image of AFefET device. c The elemental mapping of the materials in the system for Hf, Zr, Ti, N, and W, respectively. d The line scan EDS of the device cross-section, which corresponds to the red arrow in Fig. 2b. e The HRTEM image of the AFE films. f The crystal structure of the domain, which corresponds to the white square area in Fig. 2e. g The FFT image of the crystalline domain in Fig. 2f. The measured angles and distances between relative crystal faces, and diffraction spots suggest the existence of [0-10] oriented tetragonal \(P4_{2} / nmc\) structure. h The integrate-and-fire dynamics of AFefET neuron. At stage 1, the electric field-induced FE domain nucleates, and the electrons begin to accumulate in the channel of AFefET. With the further gate pulse applied (stage 2), the electric field-induced FE domains grow and expand, resulting in more
+ +<--- Page Split ---> + +attracted electrons accumulate in the channel of AFrFET. At stage 3, the electric field- induced FE domains transform back into the AFE domains when the gate pulse electric field is removed. The channel AFrFET switches off, and the neuron backs to resting potential. i The tendencies of \(\mathrm{I_d}\) under continuous gate pulses with different amplitudes (1.3 V- 1.8 V amplitude, fixed \(100\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width). + +![](images/Figure_3.jpg) + +
Fig. 3. AFrFET LIF neuron device characteristics. The dynamic of the strength-modulated integration process of AFrFET neuron under a different gate pulse amplitudes (1.2 V-1.7 V amplitude, fixed \(100\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width), b different gate pulse widths (fixed 1.7 V amplitude, fixed \(100\mu \mathrm{s}\) interval, \(10\mu \mathrm{s} - 150\mu \mathrm{s}\) width), c different gate pulse intervals (fixed 1.7 V amplitude, \(50\mu \mathrm{s} - 600\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width). When the input pulse intensity is not enough, the AFrFET neuron can not fire. Statistical data of input
+ +<--- Page Split ---> + +pulse numbers of AFoFET neuron firing as a function of d amplitudes, e widths, f intervals. The firing event needs fewer input spikes and tends to be more stable as the stimuli intensity increasing. Each data point is collected from 50 firing activities. g The energy consumption dependence on pulse parameters and threshold values of AFoFET neurons. The energy consumption decreases remarkably as the threshold and pulse width (interval) decrease. The experimental lowest energy consumption is 37 fJ per spike and can be further reduced as decreasing pulse parameters (threshold). h The AFoFET neuron can fire more than \(10^{12}\) cycles stably without significant deterioration. The results are extracted from polarization- voltage (PV) hysteresis loops @0 V forward field, 1.5 V forward field, 1.5 V backward field, respectively. i Statistical data of input pulse numbers for AFoFET neuron firing. Data are collected from 100 firing activities for each AFoFET neuron. + +![](images/Figure_4.jpg) + +
Fig. 4. Network-level performance of AFoFET neuron. a Schematic of two-layer fully ferroelectric SNN (784x400x10) for classifying MNIST datasets and the proposed hardware implementation of the network based on FeFET synapses and AFoFET neurons. b The inference accuracy as a function of training epochs with ideal LIF neurons and AFoFET
+ +<--- Page Split ---> + +neurons, \(95\%\) recognition accuracy can be achieved for AFrFET neurons under the \(1\mu \mathrm {A}\) threshold. c The speed- accuracy trade- off. The network is trained under different thresholds for all hidden and output neurons, the recognition accuracy and inference time increases with the threshold increasing. d The average spike number and energy consumption of the network under different threshold values. The spike number in the hidden layer increases with decreasing the threshold while the energy consumption on neurons decreases (threshold \(>62.5\) nA). This indicates the energy- friendly feature under a lower threshold. e The robustness of the network. The network could maintain a high accuracy of \(\sim 94.1\%\) even the device variation increases to \(\pm 16.7\%\) . + +Table 1 Comparison of various hardware implementations of artificial LIF neurons. + +
MechanismVariation#StructureEnduranceEnergy/spikeSelf-resetHardware
Capacitor		Device
Redox\(9.63\%-31.4\%^{10,45-48}\)\(\mathrm {SiO}_{x}\mathrm {N}_{y}\mathrm {:Ag}^{10}\)\(>10^{6}\)\(>60\mathrm {pJ}^{\ast }\)\(\checkmark\)11ED
\(\mathrm {Ag}/\mathrm {SiO}_{2}^{11}\)\(>10^{8}\)\(\sim 500\mathrm {nJ}^{\ast }\)\(\checkmark\)12T+1ED
\(\mathrm {Ag}/\mathrm {HfO}_{2}^{13}\)18pJ\(\checkmark\)11T+1ED
Mott\(6.31\%-11.9\%^{17,49-51}\)\(\mathrm {NbO}_{x}\mathrm {-1TiR}^{14}\)\(>10^{12}\)\(\sim 52\mathrm {fJ}\)\(\checkmark\)11T+1ED
\(\mathrm {VO}_{2}^{17}\)\(>2.6\times 10^{7}\)\(5.6\mathrm {fJ}(\text {simulation})\)\(\checkmark\)22T+2ED
\(\mathrm {GaTa}_{4}\mathrm {Se}_{8}^{18}\)\(\sim 10\mathrm {uJ}^{\ast }\)\(\checkmark\)No1ED
Phase-change\(6.82\%-12.5\%^{20,52}\)\(\mathrm {GST}^{20}\)\(3\times 10^{9}\)\(>50\mathrm {pJ}\)xNo19T+1ED
\(\mathrm {GST}^{22}\)\(10\mathrm {pJ}^{\ast }\)x12T+1ED
Magnetic\(3\%-10\%^{53-55}\)\(\mathrm {MTJ}^{23}\)\(\sim 7\mathrm {fJ}(\text {simulation})\)xNo\(>3\mathrm {T}+2\mathrm {ED}\)
STT-MRAM24xNo \(>21\mathrm {T}+1\mathrm {ED}\)
Ferroelectric\(1\%-4.03\%^{30,56}\)\(\mathrm {FeFET}^{28}\)\(\sim 360\mathrm {pJ}(\text {simulation})\)x38T+1ED
\(\mathrm {FeFET}^{27}\)\(1-10\mathrm {pJ}\)x16T+1ED
Leaky-FeFET30\(\sim 420\mathrm {pJ}^{\ast }\)\(\checkmark\)No2T+1ED
Anti-
ferroelectric		\(<3.93\%^{\ast \ast }\)\(AFeFET\)\(>10^{12}\)37fJ\(\checkmark\)No1ED
+ +**Note:** + +#: The variation results are rarely reported in literatures. In this table, the variation data is obtained from devices with the same mechanism category. + +\(*\) : The energy consumption per spike is calculated approximately from the \(I-t\) and \(V-t\) curves in these reference papers, respectively. + +\(**\) : With further optimizing the stimuli pulse parameters. + +<--- Page Split ---> + +To unify the benchmark of hardware overhead, all the circuit components in these reference papers are equivalent to two categories: emerging device (ED) and transistor (T). In chip manufacturing, the area of a resistor is equivalent to that of a transistor. 1 latch is composed of 4 transistors. 1 XOR gate is composed of 10 transistors. An integrated operational amplifier usually consists of more than 20 transistors. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41_det.mmd b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..686bb0a3f4e8c1c1799891159d0e86dd0f4c18c0 --- /dev/null +++ b/preprint/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41/preprint__1428a550cfbb50c5797e429bc9b8da56c1a20e30ee1bd887cf7303099cb5ca41_det.mmd @@ -0,0 +1,498 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 956, 174]]<|/det|> +# Compact artificial neuron based on anti-ferroelectric transistor + +<|ref|>text<|/ref|><|det|>[[44, 197, 567, 238]]<|/det|> +Rongrong Cao Institute of Microelectronics, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 243, 530, 284]]<|/det|> +Xumeng Zhang Frontier Institute of Chip and System, Fudan University + +<|ref|>text<|/ref|><|det|>[[44, 290, 320, 331]]<|/det|> +Sen Liu Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 337, 320, 377]]<|/det|> +Jikai Lu Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 383, 427, 424]]<|/det|> +Yongzhou Wang National University of Defense Technology + +<|ref|>text<|/ref|><|det|>[[44, 430, 320, 470]]<|/det|> +Yang Yang Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 476, 320, 516]]<|/det|> +Yize Sun Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 522, 320, 562]]<|/det|> +Wei Wei Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 568, 204, 608]]<|/det|> +Jianlu Wang Fudan University + +<|ref|>text<|/ref|><|det|>[[44, 614, 427, 655]]<|/det|> +Hui Xu National University of Defense Technology + +<|ref|>text<|/ref|><|det|>[[44, 660, 562, 722]]<|/det|> +Qingjiang Li Qi Liu ( liuqi@ime.ac.cn ) Fudan University https://orcid.org/0000- 0001- 7062- 831X + +<|ref|>sub_title<|/ref|><|det|>[[44, 765, 102, 782]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 802, 678, 821]]<|/det|> +Keywords: spiking neural networks, anti- ferroelectric field- effect transistor + +<|ref|>text<|/ref|><|det|>[[44, 840, 312, 858]]<|/det|> +Posted Date: October 1st, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 878, 463, 897]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 927008/v1 + +<|ref|>text<|/ref|><|det|>[[44, 915, 910, 957]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 100, 956, 142]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 17th, 2022. See the published version at https://doi.org/10.1038/s41467-022-34774-9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[122, 61, 655, 80]]<|/det|> +# Compact artificial neuron based on anti-ferroelectric transistor + +<|ref|>text<|/ref|><|det|>[[121, 108, 839, 166]]<|/det|> +Rongrong Cao \(^{1,4}\) , Xumeng Zhang \(^{2,4}\) , Sen Liu \(^{1,4}\) , Jikai Lu \(^{3}\) , Yongzhou Wang \(^{1}\) , Yang Yang \(^{3}\) , Yize Sun \(^{3}\) , Wei Wei \(^{3}\) , Jianlu Wang \(^{2}\) , Hui Xu \(^{1}\) , Qingjiang Li \(^{1*}\) , Qi Liu \(^{2*}\) + +<|ref|>text<|/ref|><|det|>[[121, 197, 839, 245]]<|/det|> +\(^{1}\) College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China + +<|ref|>text<|/ref|><|det|>[[121, 255, 839, 303]]<|/det|> +\(^{2}\) Frontier Institute of Chip and System, Fudan University, No. 2005 Songhu Road, Yangpu District, Shanghai 200438, China + +<|ref|>text<|/ref|><|det|>[[121, 314, 839, 361]]<|/det|> +\(^{3}\) Key Laboratory of Microelectronic Devices & Integrated Technology Institute of Microelectronics, Chinese Academy of Sciences, No. 3 Beitucheng West Road, Chaoyang District, Beijing 100029, China + +<|ref|>text<|/ref|><|det|>[[121, 372, 650, 390]]<|/det|> +\(^{4}\) These authors contributed equally: Rongrong Cao, Xumeng Zhang and Sen Liu + +<|ref|>text<|/ref|><|det|>[[121, 402, 644, 420]]<|/det|> +\(^{*}\) Corresponding author. E- mail: qingjiangli@nudt.edu.cn, qj_liu@fudan.edu.cn + +<|ref|>sub_title<|/ref|><|det|>[[63, 455, 139, 472]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[119, 489, 841, 927]]<|/det|> +Neuromorphic machines based on spiking neural networks (SNNs) provide a more fascinating platform over traditional computers on building energy- efficient intelligent systems, in which spiking neuron are pivotal components. Recently, memristive neurons, with promising bio- plausibility and density, have been developed, but with limited reliability or bulky capacitors for integration or additional circuits for reset. Here, we propose a novel anti- ferroelectric field- effect transistor (AFefET) neuron based on the inherent polarization and depolarization of \(\mathrm{Hf}_{0.2}\mathrm{Zr}_{0.8}\mathrm{O}_{2}\) anti- ferroelectric film to meet these challenges. In this neuron, the intrinsic polarization accumulation effect in the \(\mathrm{Hf}_{0.2}\mathrm{Zr}_{0.8}\mathrm{O}_{2}\) film increases the channel current of AFefET gradually, which implements the integration feature of the neuronal membrane and avoids using external capacitors. Also, the spontaneous depolarization effect in AFefET emulates the leaky behavior of neurons, saving the hardware overhead of neuron circuits by getting rid of external reset circuits. Moreover, the AFefET neuron exhibits other comprehensive merits, such as low energy + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 58, 840, 255]]<|/det|> +consumption (37 fJ/spike), excellent endurance ( \(>10^{12}\) ), high uniformity and high stability. Using such an AFeFET neuron, we further construct a two- layer fully ferroelectric (784x400x10) SNN combining established FeFET synapse, achieving \(96.8\%\) recognition accuracy on MNIST datasets. This work opens the way to emulate spiking neurons with antiferroelectric materials and provides a more competitive approach to build high efficient neuromorphic hardware systems. + +<|ref|>sub_title<|/ref|><|det|>[[62, 271, 172, 289]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[120, 304, 841, 533]]<|/det|> +In the past few decades, neuromorphic computing, mimicking the human brain's architecture and operation with electronic devices, has attracted great interest due to its high biomimetic and high energy efficiency \(^{1 - 3}\) . Artificial neurons are the core components of neuromorphic computing implementation, emulating biological neurons functions of potential accumulation and firing \(^{4,5}\) . For the hardware implementation of neurons, hardware overhead, energy efficiency, reliability are the critical evaluation criteria \(^{5,6}\) . Yet, current hardware demonstrations of neurons struggle to satisfy these key metrics simultaneously. + +<|ref|>text<|/ref|><|det|>[[119, 548, 841, 918]]<|/det|> +Generally, the complementary metal- oxide- semiconductor (CMOS) circuit is the most mature and stable scheme for emulating biological neurons. Nevertheless, due to the lack of intrinsic biological resemblance and the complexity of circuits, CMOS neurons face many challenges in density or energy efficiency \(^{7 - 9}\) . Recently, various emerging devices have been extensively explored to emulate biological neurons benefiting from their biological resemblance and scalability. Memristive neurons have trigged the most interests among them, including redox memristors \(^{10 - 13}\) , Mott memristors \(^{14 - 19}\) , phase- change memristors (PCM) \(^{20 - 22}\) , magnetic random access memory (MRAM) \(^{23,24}\) , etc. These neurons utilize the gradual switching of conductance to mimic membrane potential evolution, successfully emulating essential biological neuron functions with low hardware cost. However, high electroforming voltage and limited reproducibility due to temporal and spatial variations are + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 58, 841, 570]]<|/det|> +still open questions25. In addition, capacitors are usually needed to realize the integration in memristive neurons, which limits their practical applications in large- scale neuromorphic computing systems. In the very recent research, novel ferroelectric polarization- based neurons are proposed and experimentally demonstrated26- 30. They utilize gradual polarization to mimic the integration process of biological neurons without additional capacitors. Moreover, polarization is the intrinsic property of ferroelectric materials, which is recognized to be reproducible, reliable, and energy- efficient27,31. These features are promising to implement excellent neurons. However, ferroelectric devices are nonvolatile, and thus need a feedback path26- 28 or a special design of ferroelectric layer29,30 to achieve spontaneous reset after firing. The feedback path will increase the hardware cost and energy consumption of neuron implementation. In addition, it will increase the complexity of the operation, as each new input must wait for the completion of the previous reset process, especially in the system with a rate coding scheme. Thus, demonstrating an ideal electronic device that processes advanced and balanced neuronal performance without additional capacitors and reset feedback path deserves more attention. + +<|ref|>text<|/ref|><|det|>[[119, 583, 841, 917]]<|/det|> +In this work, for the first time, we report a leaky integrate- and- fire (LIF) neuron based on a CMOS- compatible AFefET. The intrinsic polarization/depolarization processes of the \(\mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFefET successfully emulate the integrate/leaky neuronal functions without any capacitors and reset peripheral circuits. Furthermore, attributing to the plentiful merits of ferroelectric materials, AFefET neuron exhibits many superiorities: electroforming- free, ultra- low energy consumption (37 fJ/spike), high endurance ( \(>10^{12}\) ), small cycle- to- cycle variation (as low as \(3.93\%\) ) and device- to- device variations ( \(7.57\%\) ). Also, we present that the temporal integration speed in such an AFefET neuron depends on the intensity of postsynaptic potential, illustrating the fundamental features for performing classification tasks. Subsequently, we demonstrate a two- layer SNN with full- ferroelectric architecture + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 839, 151]]<|/det|> +for learning and recognizing the MNIST datasets by simulation, obtaining \(96.8\%\) recognition accuracy comparable to ideal neurons. These results demonstrate that the proposed AFefET neuron is a competitive candidate for constructing neuromorphic systems. + +<|ref|>sub_title<|/ref|><|det|>[[63, 166, 127, 183]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[63, 200, 342, 219]]<|/det|> +## Volatile AFefET as LIF neurons + +<|ref|>text<|/ref|><|det|>[[118, 230, 841, 923]]<|/det|> +Figure 1a shows the architecture and processing model of biological neurons. In specific, neuronal dendrites receive input spike information from pre- neurons and transmit it to soma. Then soma integrates information and triggers an action potential when the membrane potential reaches a threshold value. The axon transmits the generated action potentials to post- neurons, and the membrane potential depolarizes to a resting state5. The increase/decrease of membrane potential corresponds to the opening or shutting of \(\mathrm{Na}^{+} / \mathrm{K}^{+}\) channels, corresponding to three stages in Fig. 1b. Here, the dynamic process of membrane potential can be mimicked vividly by the intrinsic polarization/depolarization of antiferroelectric (AFE) materials. Under a silent state, AFE materials have spontaneous polarizations, but the orientations of adjacent dipoles are opposite, resulting in zero net macroscopic remanent polarization, as shown in the insert of Fig. 1c. However, the dipoles can be aligned by sufficient electric field, and the phase switches from AFE to ferroelectric (FE)32. Usually, the electric field- induced FE phase is not stable, which will recover to AFE phase when the electric field is released. Thus, AFE materials exhibit volatile characteristics and representative double hysteresis ( \(\mathrm{P}_{\mathrm{r}} \approx 0\) at \(0 \mathrm{MV / cm}\) ) as shown in Fig. 1c. We establish the dynamic relation between the intrinsic volatile characteristic of AFE and integrate/leaky neuronal functions by constructing an AFefET, which is integrated by an AFE capacitor ( \(\mathrm{TiN / Hf}_{0.2} \mathrm{Zr}_{0.8} \mathrm{O}_{2} / \mathrm{TiN}\) ) and a MOSFET (see Methods for the details of fabrication processes), as shown in the inset of Fig. 1d. Fig. 1d shows that the typical transfer characteristic of AFefET exhibits volatility, which differs from that of nonvolatile FeFET + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[121, 62, 380, 81]]<|/det|> +at \(\mathrm{V_G = 0}\) (Supplementary Fig. 1). + +<|ref|>text<|/ref|><|det|>[[119, 95, 842, 502]]<|/det|> +The volatility of AFeFET emulates the self- recovery of biological neurons, which helps avoid external peripheral reset circuits. Furthermore, the AFeFET device is electroforming- free, which saves an additional high voltage forming circuit. In order to investigate the LIF function of AFeFET, continuous gate pulses (100 \(\mu \mathrm{s}\) width, 100 \(\mu \mathrm{s}\) interval, 1.5 V amplitude) representing postsynaptic potentials are applied to the gate of the AFeFET. The corresponding drain current \((\mathrm{I_d})\) representing the membrane potential demonstrates replicable LIF behavior under gate pulse trains (Fig. 1e). The \(\mathrm{I_d}\) increases under the excitatory spike trains and the neuron fires when \(\mathrm{I_d}\) reaches a threshold (1 \(\mu \mathrm{A}\) ). After firing, \(\mathrm{I_d}\) decreases spontaneously, and the channel switches off in a millisecond of free time eventually, which means the AFeFET neuron recovers and gets prepared for the next firing. Thanks to these merits of the AFeFET, only one AFeFET device is enough to emulate the integration and recovery process of neurons. + +<|ref|>sub_title<|/ref|><|det|>[[62, 516, 515, 536]]<|/det|> +## Device characteristics and mechanism of the AFeFET + +<|ref|>text<|/ref|><|det|>[[119, 550, 842, 920]]<|/det|> +In this work, the volatile characteristics of the AFeFET neuron are dominated by the composition of zirconium in \(\mathrm{Hf_xZr_{1 - x}O_2}\) film. The \(\mathrm{Hf_xZr_{1 - x}O_2}\) exhibits paraelectric- FE- AFE transition with increasing the concentration of zirconium element (0%- 100%) (Supplementary Fig. 2). Actually, \(\mathrm{HfO_2}\) and \(\mathrm{ZrO_2}\) exhibit very similar physical and chemical properties, such as crystal phases, lattice parameters except the dielectric properties. The pure \(\mathrm{HfO_2}\) shows linear dielectric characteristics under electric field due to the centrosymmetric monoclinic structure33. The FE properties occur with the increasing zirconium content in doped \(\mathrm{HfO_2}\) , which is induced by the existence of a non- centrosymmetric \(o\) - phase structure. The hafnium- rich ferroelectric \(\mathrm{Hf_xZr_{1 - x}O_2}\) oxides exhibit nonvolatility34- 36, thus generally serve as memory materials. With further increasing the zirconium content, the volatile AFE properties occur in zirconium- rich \(\mathrm{Hf_xZr_{1 - x}O_2}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 58, 842, 359]]<|/det|> +oxides. The polarization of AFE can be triggered by an electric field and increases under a higher electric field. But it can still revert to the initial state as the applied electric field is removed (Supplementary Fig. 3). Usually, the polarization of AFE can be ascribed to the phase transition from AFE to FE phase under the influence of electric field. The electric field- induced phase transition is always accompanied by a large- volume change37. When the electric field is released, the induced FE phase will recover to the AFE phase due to the strains resulting from volume expansion35,38,39. As a result, the zirconium- rich \(\mathrm{Hf_xZr_{1-x}O_2}\) oxides exhibit intrinsic volatility. This is the charm of AFE materials used for constructing artificial neurons. + +<|ref|>text<|/ref|><|det|>[[118, 370, 842, 920]]<|/det|> +Figures 2a and 2b show the plane structure and the detailed cross- sectional image of the AFeFET neuron. According to these images, the structure of the AFeFET neuron can be observed clearly, in which an AFE capacitor (yellow square) integrates on the gate of a conventional MOSFET. The energy- dispersive X- ray spectroscopy (EDS) mapping and line scan EDS were performed to further identify the elements and structure of AFeFET (Fig. 2c and 2d). The thickness of \(\mathrm{TiN / Hf_xZr_{1-x}O_2 / TiN}\) is \(40 \mathrm{nm} / 10 \mathrm{nm} / 40 \mathrm{nm}\) , and the interfaces of all layers are clean and flat. In addition, the Hf, Zr, Ti, N, W elements distribute uniform and are free of inter- diffused. Then we focus on investigating the characteristics of AFE layer due to its dominant role in neuronal behavior. The composition of AFE layer is controlled by alternate deposition (one cycle \(\mathrm{HfO_2}\) and four cycles \(\mathrm{ZrO_2}\) ), and is confirmed (hafnium: zirconium \(\approx 1:4\) ) by the peak areas and the relative sensitivity factors in X- ray photoelectron spectroscopy (XPS) results (Supplementary Fig. 4). The high- resolution transmission electron microscopy (HRTEM) image for the details of the AFE films is presented in Fig. 2e. The polycrystalline nature and the lattice fringes of different crystals can be observed clearly. Fig. 2f and 2g depict the crystal structure and corresponding fast Fourier transform (FFT) image of the white square area in Fig. 2e. The relative angle and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 841, 290]]<|/det|> +distance between two lattice planes and diffraction spots indicate the existence of [0- 10]- oriented AFE tetragonal \(P4_{2} / nmc\) phase. In addition, the arrangement of zirconium atoms (green dots) is very regular, and the relative angle and lattice constants are measured directly as \(55.6^{\circ}\) , 3.49 Å, and 3.2 Å, respectively. These zirconium atoms parameters match the atomic model of [0- 10]- oriented plane of tetragonal \(P4_{2} / nmc\) phase exactly in Supplementary Fig. 5. These results confirm the existence of \(t\) - phase in AFE films, which is the foundation of the AFeFET neuron. + +<|ref|>text<|/ref|><|det|>[[118, 300, 841, 920]]<|/det|> +In order to demonstrate the dynamic of AFeFET neuron, the mechanism is shown in Fig. 2h, which is related to the transformation between AFE and FE domains. At stage 1, several input pulses as postsynaptic signals are applied to the gate of AFeFET neuron, and the electric field- induced FE orthorhombic phase ( \(o\) - phase) domains nucleate, which transform from AFE tetragonal phase ( \(t\) - phase) domains under the gate pulse stimuli. The polarized charges accumulate in the AFE layer and modulate the channel resistance of MOSFET, resulting in the \(\mathrm{I_d}\) begins stepping up gradually. This process is just as the small portion of \(\mathrm{Na^{+}}\) channels opening. With further applying gate pulses, it comes to stage 2, at which the electric field- induced FE \(o\) - phase domains grow and expand. As a result, more attracted electrons accumulate in the channel of AFeFET, and the \(\mathrm{I_d}\) increases greatly, corresponding more \(\mathrm{Na^{+}}\) channels opening. Once the \(\mathrm{I_d}\) surpasses the threshold, the AFeFET neuron would fire. After firing, the electric field- induced FE \(o\) - phase domains transform back to AFE \(t\) - phase domains due to the release of gate pulse. Consequently, the attracted electrons discharge and the channel of AFeFET switches off, indicating the AFeFET neuron returns to resting potential (stage 3). This process corresponds to the opening of \(\mathrm{K^{+}}\) channels in biological neurons. Then, the AFeFET neuron fires again under another gate pulse stimuli. This repeatable and stable electric field- induced phase transition accounts for the intrinsic neuronal resemblance of AFeFET. The atomic- scale phase transition between \(t\) - phase and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[121, 60, 839, 115]]<|/det|> +\(o\) - phase under the influence of electric field has been observed clearly by S. Lombardo et al. via in- situ HRTEM39. + +<|ref|>text<|/ref|><|det|>[[118, 125, 841, 817]]<|/det|> +To present the gradual electric field- induced phase transition of AFE \(t\) - phase, we investigate the tendencies of \(\mathrm{I_d}\) under continuous gate pulses. Figure 2i shows the tendencies of \(\mathrm{I_d}\) under different gate pulse amplitude, while the gate pulse interval and width are fixed to \(100 \mu \mathrm{s}\) , respectively. Under the first 20 continuous gate pulse stimuli, an obvious integration process of \(\mathrm{I_d}\) can be observed. This resulted from the gradual formation of electric field- induced \(o\) - phase, which induces the electrons accumulation in the channel of AFeFET. The gate pulses with larger amplitude result in more \(o\) - phase domains formation and quicker growth of \(\mathrm{I_d}\) . With further gate pulse stimuli, the reversible domains tend to reach a saturation regime. This represents the dynamic balance between the electric field- induced phase transition and the recovery of the AFE \(t\) - phase, and the \(\mathrm{I_d}\) does not increase anymore. Compared between stimuli with different amplitudes, it is clearly that the higher pulse amplitude will lead to faster growth speed and a larger saturation value of \(\mathrm{I_d}\) , which is because that more AFE \(t\) - phase domains can be switched to FE \(o\) - phase domains. Noting that a similar tendency of \(\mathrm{I_d}\) can be observed under different gate pulse intervals and widths, as illustrated in Supplementary Fig. 6. Input stimuli pulses with shorter pulse intervals or wider widths induce faster integration speed and larger saturation value of \(\mathrm{I_d}\) . In all cases, the \(\mathrm{I_d}\) increases gradually and then tends to saturate corresponds to the gradual electric field- induced phase transition and saturation processes of AFE \(t\) - phase. In addition, the tendencies of \(\mathrm{I_d}\) also illustrate the intrinsic plasticity of neurons40, demonstrating that the AFeFET has high potentiality for hardware implementation of artificial LIF neurons. + +<|ref|>sub_title<|/ref|><|det|>[[62, 827, 401, 846]]<|/det|> +## Neuronal characteristics of the AFeFET + +<|ref|>text<|/ref|><|det|>[[120, 862, 839, 917]]<|/det|> +To investigate the strength- modulated integration process of AFeFET neurons, we apply gate pulse stimuli with different amplitudes and widths to implement LIF neuron + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 58, 841, 920]]<|/det|> +functions, as shown in Fig. 3a and 3b. As the input pulse intensity (amplitude/width) increasing, the AFefET neuron needs fewer input spikes to reach the threshold (1 \(\mu \mathrm{A}\) ), which indicates a higher firing rate under stronger stimuli strength. This is because more electric field- induced \(o\) - phases are formed under stronger pulse intensity, resulting in faster charge integration speed. Correspondingly, a higher \(\mathrm{I_d}\) of the AFefET neuron needs longer time to leak (Supplementary Fig. 7), which could be clarified as adaptive recovery. To further study how the leaky behavior influences the integration process, we measure the \(\mathrm{I_d}\) under gate pulse stimuli with different intervals (50 \(\mu \mathrm{s}\) - 600 \(\mu \mathrm{s}\) ), as shown in Fig. 3c. As the interval increasing, more input pulses are needed to integrate the \(\mathrm{I_d}\) to reach the threshold value. This is because that more charges are released during the free interval time, and more input pulses are required to compensate for that. It is worth noting that, when the interval time is wider enough, the \(\mathrm{I_d}\) cannot reach the threshold anyway. This feature represents the filtering capability of the neuron for weaker input signals, which is important in biological systems and neuromorphic systems41. To further evaluate the stability of the AFefET neuron, we extract the statistical data of input spike numbers for firing as a function of input amplitudes, as shown in Fig. 3d. The firing event needs fewer input spikes and tends to be more stable as the stimuli intensity increasing. This phenomenon exhibits that the AFefET neuron performs high precision computation under enough stimuli intensity, which is favorable for performing high precision tasks. A similar relationship between input spike numbers for firing and pulse widths (or intervals) is observed, as shown in Fig. 3e and 3f, respectively. To directly present the stability of the AFefET neuron, we calculate the standard deviation ( \(\sigma\) ) of integration pulse number under each stimuli condition and label them out in Fig. 3d- 3f. Furthermore, the cycle- to- cycle variation is calculated by dividing standard deviation ( \(\sigma\) ) by mean value ( \(\mu\) ). The lowest variation (3.93%) between cycles is obtained under 1.7 V gate pulse amplitude, 50 \(\mu \mathrm{s}\) interval, and 100 \(\mu \mathrm{s}\) width. It should be + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 840, 185]]<|/det|> +noted that optimizing the gate pulse parameters may further enhance the uniformity between cycles. These results demonstrate that the AFeFET neurons can successfully emulate the strength- modulated spike frequency characteristics of biological neurons with high stability, making the AFeFET neurons extremely capable of carrying out the classification tasks11,42. + +<|ref|>text<|/ref|><|det|>[[119, 200, 841, 500]]<|/det|> +To further access the compatibility of the AFeFET neuron for implementing unsupervised learning, we investigated the lateral inhibitory property. During the accumulation of AFeFET neuron membrane potential, the excitability will be inhibited immediately when the AFeFET neuron receives inhibitory stimuli from adjacent neurons, as shown in Supplementary Fig. 8a and 8b. Moreover, the AFeFET neuron needs more excitatory inputs for the next firing under stronger lateral inhibition intensity (Supplementary Fig. 8c). This behavior is similar to the suppressive phenomenon in biological neurons between each other, which is valuable for performing competitive learning tasks. + +<|ref|>text<|/ref|><|det|>[[118, 514, 841, 920]]<|/det|> +For artificial neurons, low energy consumption, high endurance, and high reliability are critical merits. To investigate the energy consumption of the AFeFET neuron, we performed a systematic analysis under different input pulse parameters and threshold values, as shown in Fig. 3g (extracted from Supplementary Fig. 9). The lowest energy consumption of \(37 \mathrm{fJ / per}\) spike can be obtained under the \(50 \mathrm{nA}\) threshold, \(1 \mu \mathrm{s}\) pulse width, and \(1 \mu \mathrm{s}\) pulse interval. Furthermore, the energy consumption decreases remarkably as the threshold and pulse width (interval) decrease. Thus, it is reasonable to infer that the energy consumption can be further reduced by decreasing the threshold and pulse width (interval). Moreover, the AFeFET neurons demonstrate considerable repeatability. Supplementary Fig. 10a shows \(5 \times 10^{5}\) stable firing cycles of the AFeFET neuron without any significant deterioration. In order to speed up the measurement, the AFE MIM structure, which is the endurance bottleneck of AFeFET, is measured for higher endurance ( \(10^{12}\) cycles) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 60, 841, 500]]<|/det|> +(Supplementary Fig. 10b). Based on the endurance measurements above, it is reasonable to believe the AFefET neuron could support more than \(10^{12}\) firing events (Fig. 3h). Fig. 3i shows the histograms of input spike numbers for firing, which are collected from 100 firing activities of each AFefET neuron. The required pulse number for firing is concentrated nearby 12, and the device- to- device variation is calculated to be as low as \(7.57\%\) . This variation is extracted from 500 firing activities in 5 AFefET neurons. As we claimed before, the uniformity could be further enhanced by optimizing the gate pulse parameters. These results indicate that the AFefET neuron has high uniformity and great potential in large- scale applications. For clearly presenting the comprehensive merits of our AFefET neuron, a benchmark comparison with other typical neurons based on emerging devices is summarized in Table 1. Considering the energy consumption, endurance, and hardware overhead, which are the critical evaluation criteria of the artificial neurons, the AFefET neuron exhibits the most attractive performances. + +<|ref|>sub_title<|/ref|><|det|>[[62, 513, 460, 533]]<|/det|> +## Network-level performance of AFefET neuron + +<|ref|>text<|/ref|><|det|>[[118, 548, 841, 917]]<|/det|> +We have demonstrated that the AFefET neurons can provide better energy efficiency and higher uniformity compared to the other neurons based on emerging devices. It is also essential to evaluate the network- level performance using the AFefET neuron for the hardware implementation of SNNs. Subsequently, we construct a two- layer SNN (784×400×10) for classifying MNIST datasets, as shown in Fig. 4a. In this network, we adopt a time- to- first- spike coding method, in which all input neurons fire exactly one spike per stimulus, but the firing order carries information. A larger input corresponds to an earlier spike of the neuron. In the output layer, the first fired neuron determines the class of stimulus. As soon as one of the output neurons fires, the network assigns the corresponding category to the input, and the inference process stop. Thus, such a coding scheme is much more suitable for hardware implementation. The right panel of Fig. 4a presents the proposed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 58, 841, 396]]<|/det|> +hardware implementation of the network based on FeFET synapses and AFoFET neurons. During inference, the input signal is applied to the gates of FeFET synapses (WLs), pulsewidth modulators (PWMs) collect current on source lines (SLs) and convert to pulses with fixed amplitude and various widths. The outputs of PWMs serve as postsynaptic potentials and be applied to the gates of AFoFET neurons for performing the integration process. Then, we train such a network to learn MNIST datasets for illustrating the feasibility. During training, we adopt a supervised temporal backpropagation algorithm proposed by Kheradpisheh43. Figure 4b shows the training results with the AFoFET neurons under 1 \(\mu \mathrm{A}\) threshold, achieving \(\sim 95\%\) recognition accuracy. These results demonstrate that our AFoFET neurons have great potential to be used for fabricating SNNs chips. + +<|ref|>text<|/ref|><|det|>[[118, 408, 841, 920]]<|/det|> +During training, the threshold value determines the number of integrated inputs (the number of membrane states) and thus affects the network performance. Then, the relation between the threshold value and network performance is further investigated as shown in Fig. 4c. It is noting that the recognition accuracy is related to the threshold, with the highest \(96.8\%\) accuracy under the 2 \(\mu \mathrm{A}\) threshold, which is nearly identical to ideal IF neurons. Nonetheless, the inference time increases with increasing the threshold value because more integration number is required to trigger neuron firing at higher threshold cases. Thus, there should be a trade- off between recognition accuracy and inference time, and an appropriate threshold value should be selected to balance the network performance in practical applications. Fortunately, the network can still achieve high accuracy ( \(>86\%\) ) even at a threshold value low to \(62.5 \mathrm{nA}\) , which is favorable for applications that need faster inference time but not rigorous accuracy. As we claimed before, the threshold also affects the energy consumption of the neuron, which is the key parameter for SNNs chip applications. We extract the spike number and energy consumption of the neurons in the system with different training thresholds to evaluate this feature, as shown in Fig. 4d. When the threshold is higher + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 58, 841, 608]]<|/det|> +than \(0.125 \mu \mathrm{A}\) , the spike number in the hidden layer ( \(2^{\mathrm{nd}}\) layer) decreases with increasing the threshold. This is because the hidden neurons with a higher threshold are hard to fire. On the contrary, we observe that the total energy consumption of neurons decreases as the threshold decreases. When the threshold reduces to \(62.5 \mathrm{nA}\) , the spike number abruptly decreases to be less than 100. This is because under a low threshold value, the winner neuron in the output layer fires earlier and the network could finish the inference process faster. In that case, only a minority of neurons in the hidden layers fire, and thus the total neurons consume less energy. These results demonstrate that decreasing the threshold supports fast inference speed and could greatly decrease the energy consumption of hardware neurons. In addition, device variation is another important parameter in practical applications. Figure 4e presents the inference accuracy under various device variations after training under the \(1 \mu \mathrm{A}\) threshold. As the red dots are shown, the inference accuracy only decreases \(1\%\) even the variation increases to \(\pm 16.7\%\) , illustrating nearly no network performance degradation. These results further demonstrate that the proposed AFefET neuron is suitable for performing SNNs tasks and has great potential for the hardware implementation of SNNs chips. + +<|ref|>sub_title<|/ref|><|det|>[[63, 620, 154, 637]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[120, 653, 841, 847]]<|/det|> +SNNs, inspired by the human brain, are powerful platforms for enabling low- power event- driven neuromorphic hardware. In SNNs, spiking neurons are the key units that enable 'spikes', which exchange information through connected plastic synapses. With rich physical dynamics, memristive devices are considered promising devices to emulate spiking neurons. However, the high energy consumption or limited reliability hinders the applications of memristive neurons in neuromorphic computing. + +<|ref|>text<|/ref|><|det|>[[121, 861, 839, 916]]<|/det|> +In this work, we demonstrated a leaky integrate- and- fire neuron based on an AFefET for the first time. The dynamic relationship between the intrinsic polarization/depolarization + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 60, 841, 430]]<|/det|> +process of the \(\mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFeFET and integrate/leaky neuronal functions are successfully built. The AFeFET neuron features CMOS- compatible, tunable firing frequency, ultra- low hardware cost (no capacitance and additional reset circuit), ultra- low energy consumption (37 fJ/spike), high endurance ( \(>10^{12}\) ), and high uniformity among different cycles and devices, showing advanced overall performances compared with emerging devices- based neurons in literature. To verify the feasibility of the neuron, we constructed a two- layer SNN combined with FeFET synapses, achieving high recognition accuracy (96.8%), low energy consumption, and high robustness on MNIST datasets. These results demonstrate that the AFeFET neuron is a promising candidate for constructing high- efficient SNN systems and may promote the industrial landing of neuromorphic machines based on anti- ferroelectric materials. + +<|ref|>sub_title<|/ref|><|det|>[[62, 445, 140, 462]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[62, 480, 225, 499]]<|/det|> +## Sample fabrication + +<|ref|>text<|/ref|><|det|>[[119, 513, 841, 884]]<|/det|> +The fabrication processes of AFeFET neuron devices are as follows: 1) After ultraviolet lithography and lift- off process, the bottom electrode TiN (40 nm) was deposited on the gate terminal of the NMOS transistor by ion beam sputtering. The NMOS transistor was fabricated by \(0.18 \mu \mathrm{m}\) CMOS technology. 2) And then, \(10 \mathrm{nm} \mathrm{Hf_{0.2}Zr_{0.8}O_2}\) AFE thin films were deposited on \(40 \mathrm{nm}\) - thick TiN bottom electrode by atomic layer deposition (ALD) process at \(280^{\circ} \mathrm{C}\) substrate temperature. The \(\mathrm{Hf[N(C_2H_5)CH_3]}_4\) , \(\mathrm{Zr[N(C_2H_5)CH_3]}_4\) and \(\mathrm{H_2O}\) were used as hafnium precursor, Zr precursor and oxygen source, respectively. The hafnium/zirconium ratio was controlled by alternate deposition of one cycle \(\mathrm{HfO_2}\) and four cycles \(\mathrm{ZrO_2}\) . 3) Then after the ultraviolet lithography process, \(40 \mathrm{nm}\) - thick TiN was grown by an ion beam sputtering system and served as the top electrode. After the lift- off process, \(25 \times 25 \mu \mathrm{m}^2\) TiN top electrode was released. The two- terminal metal- insulator- metal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[121, 60, 839, 115]]<|/det|> +structure was integrated on the gate of the NMOS transistor. 4) The fabricated device was annealed for 30 seconds at \(500^{\circ}\mathrm{C}\) in nitrogen atmosphere to crystallize. + +<|ref|>sub_title<|/ref|><|det|>[[63, 131, 249, 149]]<|/det|> +## Measurement method + +<|ref|>text<|/ref|><|det|>[[120, 165, 841, 395]]<|/det|> +The element ratio is confirmed by X- ray photoelectron spectroscopy (ESCALAB 250Xi). The cross- section TEM, high- resolution TEM, energy dispersive spectroscopy and crystal structure were analyzed by transmission electron microscopy (FEI Tecnai TF- 20, UK). The DC mode is measured by Agilent B1500 semiconductor parameter analyzer. A B1530A fast measurement unit module was used for generating the voltage pulse and measure the response current at the same time. Capacitance- electric field (C- E) tests are performed with \(10\mathrm{kHz}\) AC probing frequency and \(30\mathrm{mV}\) amplitude by Agilent B1500. + +<|ref|>sub_title<|/ref|><|det|>[[63, 410, 266, 428]]<|/det|> +## SNN simulation method + +<|ref|>text<|/ref|><|det|>[[119, 444, 841, 918]]<|/det|> +In this work, we use time- to- first- spike time coding to encode the input image into a sparse spike train. Each pixel of the input image is encoded into a single spike whose spiking time is inversely proportional to its pixel value. The dense input corresponds to earlier spiking time. And each input pixel will only generate one spike, resulting in sparser spike train than the rate coding method and significantly reducing energy consumption in hardware implementation. We constructed a \(784 \times 400 \times 10\) fully ferroelectric SNN for MNIST recognition based on such a coding method. FeFET synapses were considered during training, whose conductance was between \(5\mu \mathrm{S}\) and \(60\mu \mathrm{S}\) according to the experimental data in Ref.44. The neuron parameters were extracted from Fig. 3d and 3f. We calculated the energy consumption regarding the data in Fig. 3g under \(1\mu \mathrm{s}\) . When the neuron reaches the firing threshold, it will emit a spike to the subsequent layer. After emitting a spike, neurons will remain resting state until the end of the time window. In the output layer, the first spiking neuron determines the network decision. The simulation time steps are set to 256, where the time length of each time step is the read spike time length of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 60, 842, 501]]<|/det|> +the device (1 \(\mu \mathrm{s}\) ). The neuron threshold varies from \(62.5 \mathrm{nA}\) to \(2 \mu \mathrm{A}\) , the leakage time constant is \(300 \mu \mathrm{s}\) , and the device variation is \(10\%\) . Before training, the synapse weights are initialized randomly. In the training process, the target firing time of the correct output neuron is the earliest time that all neurons fire, and the target firing time of other neurons is set to be later than the earliest firing time. According to the defined error function, synapses of the fired neuron before the actual firing time will be updated. In order to update the weights of the hidden layer, a backpropagation algorithm is used to calculate the error of the hidden layer \(^{43}\) . During training, all time steps need to calculate the error function. In the inference process, the recognition result is obtained when the first spike is generated in the output layer, so there is no need to perform the later time step. Fewer time steps mean lower recognition latency and less energy consumption. By adjusting the neuron's threshold, the spike generation time can be adjusted, resulting in the adjustable recognition speed and energy consumption with acceptable accuracy loss. + +<|ref|>sub_title<|/ref|><|det|>[[62, 515, 204, 533]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[121, 549, 835, 600]]<|/det|> +The data that support the plots within this article and other findings of this study are available from the corresponding author upon reasonable request. + +<|ref|>sub_title<|/ref|><|det|>[[62, 621, 157, 638]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[118, 644, 840, 923]]<|/det|> +1. Silver, D. et al. Mastering the game of Go with deep neural networks and tree search. \*Nature\* 529, 484-489, (2016). +2. Zhang, W. et al. Neuro-inspired computing chips. \*Nat. Electron.\* 3, 371-382, (2020). +3. Merolla, P. A. et al. A million spiking-neuron integrated circuit with a scalable communication network and interface. \*Science\* 345, 668-673, (2014). +4. Kandel, E. R. et al. \*Principles of Neural Science\*. (McGraw-Hill, 2000). +5. Zhu, J. et al. A comprehensive review on emerging artificial neuromorphic devices. \*Appl. Phys. Rev.\* 7, 011312, (2020). +6. Wang, Z. et al. Resistive switching materials for information processing. \*Nat. Rev. Mater.\* 5, 173-195, (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 839, 108]]<|/det|> +7. Indiveri, G. et al. A VLSI array of low-power spiking neurons and bistable synapses with spike-timing dependent plasticity. IEEE Trans. Neural Netw. 17, 211-221, (2006). + +<|ref|>text<|/ref|><|det|>[[120, 118, 727, 136]]<|/det|> +8. Indiveri, G. et al. Neuromorphic silicon neuron circuits. Front. Neurosci. 5, 73, (2011). + +<|ref|>text<|/ref|><|det|>[[120, 147, 839, 194]]<|/det|> +9. Sourikopoulos, I. et al. A 4-fJ/spike artificial neuron in 65 nm CMOS technology. Front. Neurosci. 11, 123, (2017). + +<|ref|>text<|/ref|><|det|>[[120, 205, 839, 251]]<|/det|> +10. Wang, Z. et al. Fully memristive neural networks for pattern classification with unsupervised learning. Nat. Electron. 1, 137-145, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 263, 839, 309]]<|/det|> +11. Zhang, X. et al. An artificial neuron based on a threshold switching memristor. IEEE Electron Device Lett. 39, 308-311, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 321, 790, 339]]<|/det|> +12. Wang, Z. et al. Capacitive neural network with neuro-transistors. Nat. Commun. 9, 3208, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 350, 839, 396]]<|/det|> +13. Lee, D. et al. Various threshold switching devices for integrate and fire neuron applications. Adv. Electron. Mater. 5, 1800866, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 408, 840, 483]]<|/det|> +14. Zhang, X. et al. Experimental demonstration of conversion-based SNNs with 1T1R Mott neurons for neuromorphic Inference. in 2019 IEEE International Electron Devices Meeting (IEDM) 6.7.1-6.7.4, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 495, 837, 512]]<|/det|> +15. Bo, Y. et al. \(\mathrm{NbO_2}\) memristive neurons for burst-based perceptron. Adv. Intell. Syst. 2, 2000066, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 524, 840, 571]]<|/det|> +16. Jerry, M. et al. Ultra-low power probabilistic IMT neurons for stochastic sampling machines. in 2017 Symposium on VLSI Circuits T186-T187, (2017). + +<|ref|>text<|/ref|><|det|>[[120, 582, 839, 628]]<|/det|> +17. Yi, W. et al. Biological plausibility and stochasticity in scalable \(\mathrm{VO_2}\) active memristor neurons. Nat. Commun. 9, 4661, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 640, 839, 686]]<|/det|> +18. Stoliar, P. et al. A leaky-integrate-and-fire neuron analog realized with a Mott insulator. Adv. Funct. Mater. 27, 1604740, (2017). + +<|ref|>text<|/ref|><|det|>[[120, 698, 824, 716]]<|/det|> +19. Pickett, M. D. et al. A scalable neuristor built with Mott memristors. Nat. Mater. 12, 114-117, (2013). + +<|ref|>text<|/ref|><|det|>[[120, 727, 740, 745]]<|/det|> +20. Tuma, T. et al. Stochastic phase-change neurons. Nat. Nanotechnol. 11, 693-699, (2016). + +<|ref|>text<|/ref|><|det|>[[120, 756, 832, 775]]<|/det|> +21. Cobley, R. A. et al. A self-resetting spiking phase-change neuron. Nanotechnology 29, 195202, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 786, 839, 832]]<|/det|> +22. Wright, C. D. et al. Beyond von-neumann computing with nanoscale phase-change memory devices. Adv. Funct. Mater. 23, 2248-2254, (2012). + +<|ref|>text<|/ref|><|det|>[[120, 844, 839, 890]]<|/det|> +23. Sengupta, A. et al. Magnetic tunnel junction mimics stochastic cortical spiking neurons. Sci. Rep. 6, 30039, (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 839, 108]]<|/det|> +24. Wu, M.-H. et al. Extremely compact integrate-and-fire STT-MRAM neuron: a pathway toward all-spin artificial deep neural network. in 2019 Symposium on VLSI Technology T34-T35, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 118, 839, 166]]<|/det|> +25. Li, Y. et al. Anomalous resistive switching in memristors based on two-dimensional palladium diselenide using heterophase grain boundaries. Nat. Electron. 4, 348-356, (2021). + +<|ref|>text<|/ref|><|det|>[[120, 177, 839, 225]]<|/det|> +26. Dutta, S. et al. Biologically plausible ferroelectric quasi-leaky integrate and fire neuron. in 2019 Symposium on VLSI Technology T140-T141, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 235, 839, 282]]<|/det|> +27. Dutta, S. et al. Supervised learning in all FeFET-based spiking neural network: opportunities and challenges. Front. Neurosci. 14, 634, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 292, 839, 340]]<|/det|> +28. Wang, Z. et al. Experimental demonstration of ferroelectric spiking neurons for unsupervised clustering. in 2018 IEEE International Electron Devices Meeting (IEDM) 13.13.11-13.13.14, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 350, 840, 426]]<|/det|> +29. Chen, C. et al. Bio-inspired neurons based on novel leaky-FeFET with ultra-low hardware cost and advanced functionality for all-ferroelectric neural network. in 2019 Symposium on VLSI Technology T136-T137, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 437, 839, 514]]<|/det|> +30. Luo, J. et al. Capacitor-less stochastic leaky-FeFET neuron of both excitatory and inhibitory connections for SNN with reduced hardware cost. in 2019 IEEE International Electron Devices Meeting (IEDM) 6.4.1-6.4.4, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 525, 775, 544]]<|/det|> +31. Salahuddin, S. et al. The era of hyper-scaling in electronics. Nat. Electron. 1, 442-450, (2018). + +<|ref|>text<|/ref|><|det|>[[120, 554, 839, 601]]<|/det|> +32. Hao, X. et al. A comprehensive review on the progress of lead zirconate-based anti-ferroelectric materials. Prog. Mater Sci. 63, 1-57, (2014). + +<|ref|>text<|/ref|><|det|>[[120, 611, 839, 658]]<|/det|> +33. Sang, X. et al. On the structural origins of ferroelectricity in \(\mathrm{HfO_2}\) thin films. Appl. Phys. Lett. 106, 162905, (2015). + +<|ref|>text<|/ref|><|det|>[[120, 669, 805, 688]]<|/det|> +34. Muller, J. et al. Ferroelectricity in simple binary \(\mathrm{ZrO_2}\) and \(\mathrm{HfO_2}\) . Nano Lett. 12, 4318-4323, (2012). + +<|ref|>text<|/ref|><|det|>[[120, 699, 839, 746]]<|/det|> +35. Park, M. H. et al. Ferroelectricity and antiferroelectricity of doped thin \(\mathrm{HfO_2}\) -based films. Adv. Mater. 27, 1811-1831, (2015). + +<|ref|>text<|/ref|><|det|>[[120, 756, 839, 804]]<|/det|> +36. Müller, J. et al. Ferroelectric \(\mathrm{Zr_{0.5}Hf_{0.5}O_2}\) thin films for nonvolatile memory applications. Appl. Phys. Lett. 99, 112901, (2011). + +<|ref|>text<|/ref|><|det|>[[120, 814, 839, 861]]<|/det|> +37. Zhang, S.-T. et al. High-strain lead-free anti-ferroelectric electrostriction. Adv. Mater. 21, 4716-4720, (2009). + +<|ref|>text<|/ref|><|det|>[[120, 872, 839, 920]]<|/det|> +38. Pan, W. et al. Field-forced antiferroelectric-to-ferroelectric switching in modified lead zirconate titanate stannate ceramics. J. Am. Ceram. Soc. 72, 571-578, (1989). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 60, 839, 107]]<|/det|> +39. Lombardo, S. et al. Atomic-scale imaging of polarization switching in an (anti-)ferroelectric memory material: zirconia (ZrO₂). in 2020 IEEE Symposium on VLSI Technology 1-2, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 119, 839, 165]]<|/det|> +40. Baek, E. et al. Intrinsic plasticity of silicon nanowire neutronistsors for dynamic memory and learning functions. Nat. Electron. 3, 398-408, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 177, 839, 222]]<|/det|> +41. Khan, A. I. et al. The future of ferroelectric field-effect transistor technology. Nat. Electron. 3, 588-597, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 235, 839, 280]]<|/det|> +42. Burkitt, A. N. A review of the integrate-and-fire neuron model: I. Homogeneous synaptic input. Biol. Cybern. 95, 1-19, (2006). + +<|ref|>text<|/ref|><|det|>[[120, 293, 839, 339]]<|/det|> +43. Kheradpisheh, S. R. & Masquelier, T. Temporal backpropagation for spiking neural networks with one spike per neuron. Int. J. Neural Syst. 30, 2050027, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 351, 839, 396]]<|/det|> +44. Jerry, M. et al. Ferroelectric FET analog synapse for acceleration of deep neural network training. in 2017 IEEE International Electron Devices Meeting (IEDM) 6.2.1-6.2.4, (2017). + +<|ref|>text<|/ref|><|det|>[[120, 409, 839, 454]]<|/det|> +45. Li, Y. et al. High-uniformity threshold switching \(\mathrm{HfO_2}\) -based selectors with patterned Ag nanodots. Adv. Sci. 7, 2002251, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 467, 839, 512]]<|/det|> +46. Song, J. et al. Effects of liner thickness on the reliability of AgTe/TiO₂-based threshold switching devices. IEEE Trans. Electron Dev. 64, 4763-4767, (2017). + +<|ref|>text<|/ref|><|det|>[[120, 525, 839, 570]]<|/det|> +47. Jeonghwan, S. et al. Threshold selector with high selectivity and steep slope for cross-point memory array. IEEE Electron Device Lett. 36, 681-683, (2015). + +<|ref|>text<|/ref|><|det|>[[120, 583, 839, 628]]<|/det|> +48. Grisafe, B. et al. Performance enhancement of Ag/HfO₂ metal ion threshold switch cross-point selectors. IEEE Electron Device Lett. 40, 1602-1605, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 641, 839, 686]]<|/det|> +49. Chen, A. et al. Multi-functional controllable memory devices applied for 3D integration based on a single niobium oxide layer. Adv. Electron. Mater. 6, 1900756, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 699, 839, 744]]<|/det|> +50. Lin, C.-Y. et al. A high-speed MIM resistive memory cell with an inherent vanadium selector. Appl. Mater. Today 21, 100848, (2020). + +<|ref|>text<|/ref|><|det|>[[120, 757, 839, 802]]<|/det|> +51. Kang, M. & Son, J. Off-state current reduction in NbO₂-based selector device by using TiO₂ tunneling barrier as an oxygen scavenger. Appl. Phys. Lett. 109, 202101, (2016). + +<|ref|>text<|/ref|><|det|>[[120, 815, 839, 860]]<|/det|> +52. Sebastian, A. et al. Computational phase-change memory: beyond von Neumann computing. J. Phys. D: Appl. Phys. 52, 443002, (2019). + +<|ref|>text<|/ref|><|det|>[[120, 873, 839, 918]]<|/det|> +53. Dorrance, R. et al. Scalability and design-space analysis of a 1T-1MTJ memory cell for STT-RAMs. IEEE Trans. Electron Dev. 59, 878-887, (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 840, 107]]<|/det|> +54. Vincent, A. F. et al. Spin-transfer torque magnetic memory as a stochastic memristive synapse for neuromorphic systems. IEEE Trans. Biomed. Circuits Syst. 9, 166-174, (2015). + +<|ref|>text<|/ref|><|det|>[[120, 118, 840, 164]]<|/det|> +55. Wang, K. et al. Reconfigurable codesign of STT-MRAM under process variations in deeply scaled technology. IEEE Trans. Electron Dev. 62, 1769-1777, (2015). + +<|ref|>text<|/ref|><|det|>[[120, 176, 840, 223]]<|/det|> +56. Cao, R. R. et al. Effects of capping electrode on ferroelectric properties of \(\mathrm{Hf}_{0.5}\mathrm{Zr}_{0.5}\mathrm{O}_{2}\) thin films. IEEE Electron Device Lett. 39, 1207-1210, (2018). + +<|ref|>sub_title<|/ref|><|det|>[[63, 253, 220, 271]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[120, 286, 841, 498]]<|/det|> +This work was supported by the National Key R&D Program of China under Grant No. 2018YFA0701500, the National Natural Science Foundation of China under Grant Nos. 62004220, 62004219, 62074166, 61974164, 61804181, 61825404, 61732020, 61821091, and 61851402, the Strategic Priority Research Program of the Chinese Academy of Sciences under Grant XDB44000000 and the China Postdoctoral Science Foundation under Grant No. 2020M681167. The authors thank the valuable discussion from W. Wang, X. Li with National University of Defense Technology, Y. Li and Z.H. Wu with University of Chinese Academy of Sciences. + +<|ref|>sub_title<|/ref|><|det|>[[63, 519, 244, 537]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[120, 552, 841, 731]]<|/det|> +R.R.C. and X.M.Z. designed the experiments. R.R.C., X.M.Z. and Y.Z.S. designed and fabricated the AFeFET devices. R.R.C. carried out the electrical experiments. X.M.Z., J.K.L. and Y.Z.W. conducted the simulation. R.R.C., Y.Y. and W.W. carried out the TEM and EDS test. S.L., J.L.W and Q.L. assisted with data analysis and interpretation. R.R.C., X.M.Z, and S.L. co- wrote the manuscript. All authors discussed the results and revised the manuscript. Q.J.L and Q.L. supervised the research. + +<|ref|>sub_title<|/ref|><|det|>[[63, 744, 260, 763]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[120, 780, 560, 800]]<|/det|> +Supplementary information accompanies this paper at... + +<|ref|>sub_title<|/ref|><|det|>[[63, 821, 234, 840]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[63, 855, 378, 872]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[137, 87, 830, 440]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[72, 61, 131, 78]]<|/det|> +
Figures
+ +<|ref|>text<|/ref|><|det|>[[118, 450, 842, 916]]<|/det|> +Fig. 1. Leaky integrate-and-fire dynamic of biological neuron vs AFoFET artificial neurons. a The architecture and processing model of biological neurons. External information is perceived by dendrites and then transmitted to the soma for processing. When the membrane potential surpasses a threshold value, an action potential can be generated and transmitted to post- neurons through the axon. b The integrate and fire dynamics of a biological neuron. With the signal received by dendrites and transmitted to soma, few \(\mathrm{Na^{+}}\) channels are activated, and the membrane potential goes up gradually (stage 1). As the further signal is received, more \(\mathrm{Na^{+}}\) channels are activated quickly, and a large amount of \(\mathrm{Na^{+}}\) flows inside the membrane, inducing the membrane potential goes up rapidly (stage 2). Once the membrane potential exceeds a certain threshold, it decreases due to the inactivation of \(\mathrm{Na^{+}}\) channels and the opening of \(\mathrm{K^{+}}\) channels (stage 3). c The representative double hysteresis of AFE materials. The zero net macroscopic remanent polarization indicates the volatile characteristics of AFE materials. d The typical transfer curves of AFoFET, exhibiting counterclockwise hysteresis and volatility. The arrows indicate the switching + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[122, 61, 840, 115]]<|/det|> +sequences. e The continuous firing events of AFefET neuron. After several milliseconds of free time, the AFefET neuron can restart the LIF process and fire again. + +<|ref|>image<|/ref|><|det|>[[125, 135, 835, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[120, 559, 841, 928]]<|/det|> +
Fig. 2. Device characteristics and mechanism of AFefET neurons. a Planer structure of AFefET device. b The cross-sectional image of AFefET device. c The elemental mapping of the materials in the system for Hf, Zr, Ti, N, and W, respectively. d The line scan EDS of the device cross-section, which corresponds to the red arrow in Fig. 2b. e The HRTEM image of the AFE films. f The crystal structure of the domain, which corresponds to the white square area in Fig. 2e. g The FFT image of the crystalline domain in Fig. 2f. The measured angles and distances between relative crystal faces, and diffraction spots suggest the existence of [0-10] oriented tetragonal \(P4_{2} / nmc\) structure. h The integrate-and-fire dynamics of AFefET neuron. At stage 1, the electric field-induced FE domain nucleates, and the electrons begin to accumulate in the channel of AFefET. With the further gate pulse applied (stage 2), the electric field-induced FE domains grow and expand, resulting in more
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 58, 840, 220]]<|/det|> +attracted electrons accumulate in the channel of AFrFET. At stage 3, the electric field- induced FE domains transform back into the AFE domains when the gate pulse electric field is removed. The channel AFrFET switches off, and the neuron backs to resting potential. i The tendencies of \(\mathrm{I_d}\) under continuous gate pulses with different amplitudes (1.3 V- 1.8 V amplitude, fixed \(100\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width). + +<|ref|>image<|/ref|><|det|>[[123, 237, 844, 702]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[120, 716, 840, 910]]<|/det|> +
Fig. 3. AFrFET LIF neuron device characteristics. The dynamic of the strength-modulated integration process of AFrFET neuron under a different gate pulse amplitudes (1.2 V-1.7 V amplitude, fixed \(100\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width), b different gate pulse widths (fixed 1.7 V amplitude, fixed \(100\mu \mathrm{s}\) interval, \(10\mu \mathrm{s} - 150\mu \mathrm{s}\) width), c different gate pulse intervals (fixed 1.7 V amplitude, \(50\mu \mathrm{s} - 600\mu \mathrm{s}\) interval, fixed \(100\mu \mathrm{s}\) width). When the input pulse intensity is not enough, the AFrFET neuron can not fire. Statistical data of input
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 58, 842, 430]]<|/det|> +pulse numbers of AFoFET neuron firing as a function of d amplitudes, e widths, f intervals. The firing event needs fewer input spikes and tends to be more stable as the stimuli intensity increasing. Each data point is collected from 50 firing activities. g The energy consumption dependence on pulse parameters and threshold values of AFoFET neurons. The energy consumption decreases remarkably as the threshold and pulse width (interval) decrease. The experimental lowest energy consumption is 37 fJ per spike and can be further reduced as decreasing pulse parameters (threshold). h The AFoFET neuron can fire more than \(10^{12}\) cycles stably without significant deterioration. The results are extracted from polarization- voltage (PV) hysteresis loops @0 V forward field, 1.5 V forward field, 1.5 V backward field, respectively. i Statistical data of input pulse numbers for AFoFET neuron firing. Data are collected from 100 firing activities for each AFoFET neuron. + +<|ref|>image<|/ref|><|det|>[[123, 442, 848, 768]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[120, 782, 840, 907]]<|/det|> +
Fig. 4. Network-level performance of AFoFET neuron. a Schematic of two-layer fully ferroelectric SNN (784x400x10) for classifying MNIST datasets and the proposed hardware implementation of the network based on FeFET synapses and AFoFET neurons. b The inference accuracy as a function of training epochs with ideal LIF neurons and AFoFET
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 60, 841, 358]]<|/det|> +neurons, \(95\%\) recognition accuracy can be achieved for AFrFET neurons under the \(1\mu \mathrm {A}\) threshold. c The speed- accuracy trade- off. The network is trained under different thresholds for all hidden and output neurons, the recognition accuracy and inference time increases with the threshold increasing. d The average spike number and energy consumption of the network under different threshold values. The spike number in the hidden layer increases with decreasing the threshold while the energy consumption on neurons decreases (threshold \(>62.5\) nA). This indicates the energy- friendly feature under a lower threshold. e The robustness of the network. The network could maintain a high accuracy of \(\sim 94.1\%\) even the device variation increases to \(\pm 16.7\%\) . + +<|ref|>table_caption<|/ref|><|det|>[[120, 377, 772, 393]]<|/det|> +Table 1 Comparison of various hardware implementations of artificial LIF neurons. + +<|ref|>table<|/ref|><|det|>[[120, 393, 841, 704]]<|/det|> + +
MechanismVariation#StructureEnduranceEnergy/spikeSelf-resetHardware
Capacitor		Device
Redox\(9.63\%-31.4\%^{10,45-48}\)\(\mathrm {SiO}_{x}\mathrm {N}_{y}\mathrm {:Ag}^{10}\)\(>10^{6}\)\(>60\mathrm {pJ}^{\ast }\)\(\checkmark\)11ED
\(\mathrm {Ag}/\mathrm {SiO}_{2}^{11}\)\(>10^{8}\)\(\sim 500\mathrm {nJ}^{\ast }\)\(\checkmark\)12T+1ED
\(\mathrm {Ag}/\mathrm {HfO}_{2}^{13}\)18pJ\(\checkmark\)11T+1ED
Mott\(6.31\%-11.9\%^{17,49-51}\)\(\mathrm {NbO}_{x}\mathrm {-1TiR}^{14}\)\(>10^{12}\)\(\sim 52\mathrm {fJ}\)\(\checkmark\)11T+1ED
\(\mathrm {VO}_{2}^{17}\)\(>2.6\times 10^{7}\)\(5.6\mathrm {fJ}(\text {simulation})\)\(\checkmark\)22T+2ED
\(\mathrm {GaTa}_{4}\mathrm {Se}_{8}^{18}\)\(\sim 10\mathrm {uJ}^{\ast }\)\(\checkmark\)No1ED
Phase-change\(6.82\%-12.5\%^{20,52}\)\(\mathrm {GST}^{20}\)\(3\times 10^{9}\)\(>50\mathrm {pJ}\)xNo19T+1ED
\(\mathrm {GST}^{22}\)\(10\mathrm {pJ}^{\ast }\)x12T+1ED
Magnetic\(3\%-10\%^{53-55}\)\(\mathrm {MTJ}^{23}\)\(\sim 7\mathrm {fJ}(\text {simulation})\)xNo\(>3\mathrm {T}+2\mathrm {ED}\)
STT-MRAM24xNo \(>21\mathrm {T}+1\mathrm {ED}\)
Ferroelectric\(1\%-4.03\%^{30,56}\)\(\mathrm {FeFET}^{28}\)\(\sim 360\mathrm {pJ}(\text {simulation})\)x38T+1ED
\(\mathrm {FeFET}^{27}\)\(1-10\mathrm {pJ}\)x16T+1ED
Leaky-FeFET30\(\sim 420\mathrm {pJ}^{\ast }\)\(\checkmark\)No2T+1ED
Anti-
ferroelectric		\(<3.93\%^{\ast \ast }\)\(AFeFET\)\(>10^{12}\)37fJ\(\checkmark\)No1ED
+ +<|ref|>text<|/ref|><|det|>[[66, 707, 116, 720]]<|/det|> +**Note:** + +<|ref|>text<|/ref|><|det|>[[120, 740, 839, 789]]<|/det|> +#: The variation results are rarely reported in literatures. In this table, the variation data is obtained from devices with the same mechanism category. + +<|ref|>text<|/ref|><|det|>[[120, 811, 839, 862]]<|/det|> +\(*\) : The energy consumption per spike is calculated approximately from the \(I-t\) and \(V-t\) curves in these reference papers, respectively. + +<|ref|>text<|/ref|><|det|>[[120, 883, 572, 898]]<|/det|> +\(**\) : With further optimizing the stimuli pulse parameters. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[119, 60, 840, 221]]<|/det|> +To unify the benchmark of hardware overhead, all the circuit components in these reference papers are equivalent to two categories: emerging device (ED) and transistor (T). In chip manufacturing, the area of a resistor is equivalent to that of a transistor. 1 latch is composed of 4 transistors. 1 XOR gate is composed of 10 transistors. An integrated operational amplifier usually consists of more than 20 transistors. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 353, 150]]<|/det|> +SupplementaryInformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/images_list.json b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..13420be59c02b716957a738448fc3fc219bbb99a --- /dev/null +++ b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Rotational isomerism of BA molecules during compressive stress of \\(n = 2\\) RPP. a Schematic showing diamond anvil cell and the \\(n = 2\\) RPP crystals with linear tt-BA organic chains. b DFT simulation of lamellar contraction and the step-wise rotational isomerism.", + "footnote": [], + "bbox": [ + [ + 123, + 100, + 878, + 553 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2. Distinctive pressure responses of \\(n = 1\\) and \\(n = 2\\) RPP. a Two generic types of layer-shift mode: \\(\\Gamma_5^+\\) for \\(n = 1\\) and \\(M_5^-\\) (a, b) for \\(n = 2\\) RPP. b, c Pressure-dependent in-plane layer shift factor (LSF) and vertical strain for \\(n = 1\\) and 2 RPPs.", + "footnote": [], + "bbox": [ + [ + 115, + 81, + 884, + 550 + ] + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. Evolution of the inorganic lattice parameters of RPPs under compression and decompression. a Schematic of the 2D Pb-I octahedral cages, highlighting the apical and equatorial angle \\(\\alpha\\) and \\(\\beta\\) , and the in-plane and out-of-plane octahedral tilt angle \\(\\theta\\) and \\(\\gamma\\) , and bond length \\(d_{11 - Pb}\\) and \\(d_{12 - Pb}\\) , respectively. b Evolution of angle \\(\\alpha\\) and \\(\\beta\\) in \\(n = 2\\) under compression and decompression. Evolution of the in-plane and out-of-plane octahedral tiltings in \\(n = 1\\) RPP (c) and \\(n = 2\\) RPP (d) under compression and decompression. Lattice parameters for \\(n = 2\\) under compression is only measured at pressure below 3.16 GPa because severe lattice distortions at high pressures produce large errors.", + "footnote": [], + "bbox": [ + [ + 115, + 88, + 886, + 592 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Raman characterization of the BA tg- and \\(\\mathbf{g}^{*}\\mathbf{g}\\) - conformers. a, b Theoretical simulated Raman spectra for the BA bending and rocking modes in the tt, tg and \\(\\mathbf{g}^{*}\\mathbf{g}\\) structures. c Typical Raman spectral evolution of \\(n = 2\\) RPP with increasing pressure. d, e Raman spectral comparison of RPP sample before and after compression. \\(\\delta\\) and \\(\\rho\\) refers to bending and rocking vibrations. The grey, red and blue lines in (a, b) represent BA vibrations in the tt, tg and \\(\\mathbf{g}^{*}\\mathbf{g}\\) conformations, respectively Insets represent the corresponding vibration patterns of BA isomers, green arrows on the atoms indicate the vibrational directions and relative amplitude of displacements in the given Raman modes.", + "footnote": [], + "bbox": [ + [ + 120, + 91, + 884, + 465 + ] + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 PL characterization of the reversible and irreversible deformation in RPP. In-situ PL spectra of \\(n = 2\\) RPP; individual spectrum in (a) shows compression at specific “X” GPa, and corresponding spectrum in (b) after decompression of “X” GPa indicated as Re “X” GPa. The blue arrows in gradient color represent the proportion of the broadband emission. c Optical photographs and fluorescence images of \\(n = 2\\) RPP exfoliated flakes before and after pressure treatments. Scale bar is \\(50 \\mu \\mathrm{m}\\) . d Plot summarizing the threshold pressure to cross from elastic to plastic regime for \\(n = 1\\) to 4 RPPs, which is concomitant with the transition from a state where PL is reversible after decompression, to one where it is not. Dashed line guides to the eye.", + "footnote": [], + "bbox": [ + [ + 115, + 94, + 880, + 389 + ] + ], + "page_idx": 18 + } +] \ No newline at end of file diff --git a/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7.mmd b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f3c3882cd134e18da94a2a0c12b5966600c925e6 --- /dev/null +++ b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7.mmd @@ -0,0 +1,271 @@ + +# Pressure Driven Rotational Isomerism in 2D Hybrid Perovskites + +TingTing YinNanyang Technological University + +Hejin YanUniversity of Macau + +Ibrahim AbdelwahabNational University of Singapore https://orcid.org/0000- 0002- 0107- 5827 + +Yulia LekinaNanyang Technological University + +Xujie LüCenter for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 8402- 7160 + +Wenge YangCenter for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 8436- 8731 + +Handong SunNanyang Technological University https://orcid.org/0000- 0002- 2261- 7103 + +Kai LengThe Hong Kong Polytechnic University https://orcid.org/0000- 0003- 3408- 5033 + +Yongqing CaiUniversity of Macau + +Ze ShenNanyang Technological UniversityKian Ping Loh ( chmlohkp@nus.edu.sg )National University of Singapore https://orcid.org/0000- 0002- 1491- 743X + +## Article + +Keywords: 2D RPPs, Pressure, Rational isomerism, Tilting, Raman, photoluminescence + +Posted Date: March 10th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1394948/v1 + +<--- Page Split ---> + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Pressure Driven Rotational Isomerism in 2D Hybrid Perovskites + +Tingting Yin \(^{1,\dagger}\) , Hejin Yan \(^{2,\dagger}\) , Ibrahim Abdelwahab \(^{3}\) , Yulia Lekina \(^{1,4}\) , Xujie Lü \(^{5}\) , Wenge Yang \(^{5}\) , Handong Sun \(^{1}\) , Kai Leng \(^{6}\) , Yongqing Cai \(^{2*}\) , Zexiang Shen \(^{1,4*}\) and Kian Ping Loh \(^{3*}\) + +\(^{1}\) Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore 637371, Singapore \(^{2}\) Joint Key Laboratory of Ministry of Education Institute of Applied Physics and Materials Engineering, University of Macau, Macau, China \(^{3}\) Centre for Advanced 2D Materials (CA2DM) and Department of Chemistry, National University of Singapore, Singapore 117543, Singapore \(^{4}\) Centre for Disruptive Photonic Technologies, the Photonics Institute, Nanyang Technological University, 637371, Singapore \(^{5}\) Center for High Pressure Science and Technology Advanced Research (HPSTAR), Shanghai 201203 (P. R. China) \(^{6}\) Department of Applied Physics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, China + +\(^{\dagger}\) These authors contributed equally: Tingting YIN, Hejin YAN. \*Corresponding author. Email: yongqingcai@um.edu.mo; zexiang@ntu.edu.sg and chmlohkp@nus.edu.sg + +<--- Page Split ---> + +## ABSTRACT + +ABSTRACTMultilayers consisting of alternating soft and hard layers offer enhanced toughness compared to all- hard structures. However, shear instability usually exists in physically sputtered multilayers because of deformation incompatibility among hard and soft layers. Here, we demonstrate that 2D hybrid organic- inorganic perovskites (HOIP) provide an interesting platform to study the stress- strain behavior of hard and soft layers undulating with molecular scale periodicity. We investigate the phonon vibrations and photoluminescence properties of Ruddlesden- Popper perovskites (RPPs) under hydrostatic compression using a diamond anvil cell. The C4 alkyl chain in RPP as the organic spacer buffers compressive stress by tilting (for \(n = 1\) RPP) or step- wise rotational isomerism (for \(n > 1\) RPP) during compression. By examining the pressure threshold of the elastic recovery regime across \(n = 1\) to \(n = 4\) RPP perovskites, we obtained molecular insights into the relationship between structure and deformation resistance in hybrid organic- inorganic perovskites. + +Keywords: 2D RPPs, Pressure, Rational isomerism, Tilting, Raman, photoluminescence + +## INTRODUCTION + +Two- dimensional (2D) Ruddlesden- Popper perovskites (RPPs) consist of alternatively stacked layers of soft organic layers and rigid inorganic layers with highly tunable optoelectronic properties \(^{1, 2}\) . The organic layers provide the dielectric and quantum confinement of the inorganic layers, giving rise to large exciton binding energies and high oscillator strength \(^{3 - 5}\) . From a structural mechanic point of view, the inorganic layers serve as mechanical brace to support spring- like molecular layers, forming a "natural" undulating hard- soft system. In the realm of mechanical engineering, such hard- soft multilayer system is highly sought after for its high plastic deformation resistance and increased fracture hardness, but require complex fabrication process. The undulating layers create bands of tensile and compressive stress that is different from that of single- component layers, thus research efforts on inorganic hard- soft multilayers are centered on engineering structure and compositional gradient to enhance the mechanical properties of materials. The general chemical formula for lead halide 2D RPPs is \((\mathrm{RNH}_3)_2\mathrm{Mn} - 1\mathrm{Pb}_n\mathrm{X}_{3n + 1} (n = 1, 2, 3, 4, \ldots)\) , where \(\mathrm{PbX}_4^{2 - }\) is an inorganic perovskite layer of corner- sharing metal halide octahedra, \(\mathrm{RNH}_3^+\) is a long- chain alkylammonium organic molecular layer, such as butyl ammonium (BA), and M is a smaller cation filled into the 12- fold coordinated holes among the \(\mathrm{PbX}_6\) octahedra, such as \(\mathrm{CH}_3\mathrm{NH}_3^+\) (MA \(^+\) ) \(^{6 - 8}\) . The long range structure of the organic array is a result of balancing local strain induced by methyl ammonium ions on the inorganic cage, hydrogen- bonding and electrostatic forces between the cationic head group of the organic molecule and the metal halide framework \(^{9}\) . + +<--- Page Split ---> + +\(^{10}\) . Thanks to the rigid- /soft- layer alternated stacking structure of 2D layered perovskites, the rigid inorganic framework can template the conformation of the soft organic molecule confined in the organic- inorganic atomic interface \(^{11}\) . In the ground state of C4 alkyl chain, each \(\mathrm{CH}_2\) moiety adopts the trans- trans (tt) conformation to avoid steric repulsion. Under compression, it can be expected that the lamellar layers will contract and the strain in the metal halide bonds will affect optical and electronic properties \(^{12 - 21}\) . However, little is known about the conformational change of the organic cations and the plastic deformation or strain hardening process when the organic molecules re- orientate under pressure, this is because X- ray diffraction is mostly sensitive to the heavier inorganic components. + +Here, using a combination of in- situ Raman and Photoluminescence (PL) spectroscopy, we investigate how compressive and tensile strain develops in the organic and inorganic components in (BA) \(_2\) MA \(_{n - 1}\) Pb \(_{n}\) 13 \(_{n + 1}\) ( \(n = 1\) , 2, 3, 4, ...) during compression. Combining experiments and density functional theory (DFT), we observed that \(n = 1\) and \(n > 1\) exhibit markedly different mechanisms of compressive strain under pressure. In \(n = 2\) RPP, under compression, the trans- trans (tt) structure of the butyl ammonium cation gradually transform into the trans- gauche (tg) and gauche- gauche (g \(^+\) g \(^-\) ) conformations via C- C and C- N bond rotations, along with tensile strain of the inorganic slabs in both \(a\) and \(b\) axial directions. Whereas for \(n = 1\) RPP, tilting of the linear organic cation is along with in- plane shift in one direction ( \(b\) axis), resulting in orthorhombic- to- monoclinic phase change under compression. \(n = 1\) RPP shows remarkable elastic recovery ability due to the flexibility of organic cation that acts synergistically with the octahedral tilt of the inorganic cages. Our study shows how differing alternation of soft and hard layers allows compressive and tensile stress to evolve differently across \(n = 1\) to \(n = 4\) RPPs, which provides the essential experimental basis of understanding stress- strain behavior in nanometer scale multilayer hard- soft superlattices. + +## RESULTS + +To lay the groundwork for understanding the complex structural changes in the organic and inorganic parts of the hybrid perovskites upon hydrostatic compression, it is beneficial to organize the paper by discussing the changes revealed by DFT calculations first, followed by comparison with what was observed in the experiments. + +## Theoretical calculations of lamellar contraction + +Figure 1a depicts a unit cell of \(n = 2\) RPP. Two layers of interdigitated linear organic BA molecules alternately stack with two layers of \(\mathrm{PbI}_4^{2 - }\) inorganic slabs. The unit cell height of \(\sim 2.1 \mathrm{nm}\) is determined by atomic force microscopy (AFM) \(^{22}\) . BA organic molecules in \(tt\) conformation dominate the large interlayer spaces, while MA ions occupy the interoctahedral voids. In the case of \(n = 1\) RPP, there are no MA ions and one layer of \(\mathrm{PbI}_4^{2 - }\) inorganic slab is alternatively stacked with two layers of BA molecules (Fig. S2). + +<--- Page Split ---> + +To simulate hydrostatic compression, the bi- layer \(n = 2\) RPP sample is sandwiched between two diamond slabs. The surfaces of the diamond slabs are fully saturated with hydrogen atoms to avoid any dangling bonds. Then, the bottom diamond layer is fixed while the top diamond layer is gradually shifted downward by \(0.4 \text{Å}\) per step, this allows the pressure to increase from 0 to 14 GPa in a stepwise manner (Fig. S1a). Under compression, the lamellar distance contracts. From 0 to 3.16 GPa, compressive stress is accommodated mainly by the deformation of the organic layers. At 0.27 GPa, the lamellar contraction is achieved by rotation of the \(tt\) isomer of BA to tg (green color) and \(\text{g}^+\text{g}^-\) (purple color) conformers on the surface of RPP, which can be accomplished by rotating along the C- C/C- N chain (Fig. 1b). The rotation of the organic cations allows the \(\text{Pbl}_4^{2 - }\) inorganic cage to resist deformation over a wide range of pressure and deforms severely only at relatively high pressures at 5.04 GPa (Fig. S1b). In contrast, the BA molecules are irreversibly transformed to the higher energy conformers, which remain even after decompression (Fig. S1c). The ability to generate new rotational conformers in RPP by compression is subsequently verified in our experiments shown later. + +Compression results in the layer- to- layer shift of the bi- layer inorganic slabs in \(n = 2\) RPP along \(- a\) and \(+b\) axes, as represented by the blue and grey arrows in Fig. 2a (right side), which is related to a \(M_5^-(a,b)\) mode \(^{23}\) . Such an interlayer shift maintains \(n = 2\) RPP in the orthorhombic symmetry, consistent with the conformational change of BA molecules without tilting under pressure. + +Interestingly, we observed pronounced differences in the pressure response of \(n = 1\) RPP compared to its higher homologues, as illustrated in Fig. 2. For \(n = 1\) RPP, the linear BA molecules tilt without undergoing conformational change till 7.35 GPa (Fig. S2). The molecular tilting drives the adjacent inorganic layer to shift in the same direction along \(+b\) axis, as represented by the grey arrows in Fig. 2a (left side), which is related to a \(\Gamma_5^+\) mode \(^{23}\) . The layer shift decreases the initial stacking offset and therefore the crystal undergoes a phase transition to the monoclinic class \(^{24}\) . Such pressure- induced phase transition is however absent in \(n > 1\) RPP. The different responses of \(n = 1\) and \(n > 1\) RPP to compression is related to their different organic- inorganic couplings, refer to Fig. S4 for details. + +## In-plane shift + +To quantitatively compare the degree of in- plane shift between \(n = 1\) and 2 RPPs, we calculate the in- plane layer shift factor (LSF) \(^{25}\) , i.e., \(\Delta\) , in the low- pressure region ( \(< \sim 3\) GPa, where the lamellar contraction is dominated by BA organic layers). Refer to Fig. S3 for the top views of layer shift of \(n = 1\) and 2 RPPs under various pressures. Here, the LSF compares the in- plane coordinates of the vector \(\overline{l_1} - \overline{l_2}\) of each possible pair of Pb atoms \((l_1, l_2)\) in the first and second layers in the \(\sqrt{2} \times \sqrt{2}\) supercell metric \(^{25}\) . The calculated \(\Delta\) values are represented by the green filled circles for \(n = 1\) in Fig. 2b and green /blue filled cubes for \(n = 2\) in Fig. 2c. The strain in \(n = 1\) RPP is uniaxial along \(b\) + +<--- Page Split ---> + +axis, with \(\Delta\) reaching \(\sim 0.33\) only with a pressure of 0.24 GPa. At 0.73 GPa, \(n = 2\) RPP undergoes strain along - a and \(b\) axis, with \(\Delta \sim 0.38\) along the \(b\) axis and 0.19 along the - a axis, respectively. + +According to our calculations, the tilting of BA cations in \(n = 1\) RPP results in a vertical strain of \(\sim 16.9\%\) in response to a pressure of 2.51 GPa (violet filled circles in Fig. 2b), while the rotation isomerism of BA cations in \(n = 2\) RPP results in \(\sim 18.8\%\) vertical strain with a pressure of 2.31 GPa (violet filled cubes in Fig. 2c). + +## Compressibility and Octahedral tilt + +To track the structural deformation and recoverability of the inorganic sublattice during compression- decompression circles, we track several key structural parameters that represent the in- plane and out- of- plane components of the Pb- l octahedral cage (Fig. 3a), namely, anionic cage angles \((\beta\) and \(\alpha\) ), octahedral tilting angles \((\gamma\) and \(\Theta\) ) and Pb- l bond lengths \((\mathrm{d}_{11 - \mathrm{Pb}}\) and \(\mathrm{d}_{12 - \mathrm{Pb}}\) ). + +Under compression, \(n = 2\) RPP undergoes lamellar contraction by the rotational isomerism of the softer cations, this allows the harder Pb- l lattice to remain in the elastic regime when the pressure is less than \(\sim 3\) GPa. The apical angle \(\alpha\) linking the two \(\mathrm{PbI_4^{2 - }}\) octahedral cages serves as a pivot and bends by up to \(10^{\circ}\) during vertical compression, and unbends by almost the same degree after decompression (red symbols in Fig. 3b). The increase in \(\alpha\) beyond a certain bending angle reflects the pressure threshold in the \(n = 2\) system, beyond which irreversible deformation occurs. On the other hand, the equatorial angle \(\beta\) subtends by Pb- I- Pb in the \(ab\) plane (blue symbols in Fig. 3b) reflects the intra- layer compressive strain. Besides, the structural distortion of the inorganic sublattice can also be revealed by octahedral tilt, which occurs under pressures in both the in- plane and out- of- plane directions (solid symbols in Fig. 3d). + +In the case of \(n = 1\) RPP, lamellar contraction occurs by the tilting of the organic cation (Fig. S2). At 7.35 GPa, the out- of- plane octahedral tilt \((8.9^{\circ})\) of the inorganic sublattice is more pronounced than the in- plane tilt \((\sim 0.9^{\circ}\) on average, see the solid symbols in Fig. 3c). These two tilt angles show good recoverability after decompression, as represented by the hollow symbols in Fig. 3c. It suggests that \(n = 1\) RPP resists vertical compression by the combination of the tilting of organic cation as well as out- of- plane octahedral tilt of the inorganic sublattice. An important difference is that the tilted organic cations in \(n = 1\) RPP can be untilted upon decompression, whereas the rotation isomers in \(n = 2\) RPP are locked and cannot be reverted to the low energy isomers upon decompression. The good compressibility of \(n = 1\) system is revealed by the elastically recoverability of tilt angles as well as bond lengths (Fig. S5a). + +## Raman spectroscopic evidence of rotated BA conformers + +<--- Page Split ---> + +Raman spectroscopy was commonly used to probe bond rotation of butyl chain during the pressurization process26, 27. The single crystal of \(n = 2\) RPP was synthesized through a temperature- programmed crystallization method28, 29, and the grounded powder sample was loaded into the diamond anvil cell (DAC) for in- situ high pressure Raman measurement. We used powder sample because it can serve as a pressure transmission medium to avoid the strong Raman signal background induced by using silicon oil. + +As a reference, we have calculated the Raman spectra of \(\mathsf{BA}^+\) gas molecule in the tt, tg and \(\mathsf{g}^+ \mathsf{g}^-\) conformations, as represented by the grey, blue and red solid lines in Fig. 4a and b (see the detailed peak assignment in the Supplementary Discussion). + +Our measured in- situ high- pressure Raman spectra of \(n = 2\) RPP under compression, as shown in Fig. 4c (see also Table S3), reproduces largely the Raman peak evolutions during trans to gauge conformational change of the \(\mathsf{BA}^+\) gas molecules. The measured Raman bands at 0 GPa in the frequency region of \(450 - 890 \text{cm}^{- 1}\) are the skeletal vibrational modes of the BA in the tt conformation, as represented by grey solid line in Fig. 4c. The Raman bands appearing at \(\sim 475.3\) and \(484.7 \text{cm}^{- 1}\) belong to \(\delta (\text{CCC})\) and \(\delta (\text{CCN})\) , and the observed characteristic band at \(\sim 864.7 \text{cm}^{- 1}\) belongs to \(\rho (\text{tt- BA})\) . Besides, there is a very weak band at \(\sim 837.2 \text{cm}^{- 1}\) originating from the rocking of \(\text{CH}_2\) (Table S2). Upon compression at 0.54 GPa, the Raman spectrum shows distinct changes. First, the \(\delta (\text{CCC})\) and \(\delta (\text{CCN})\) band almost merge into one band at \(\sim 487.2 \text{cm}^{- 1}\) , which is the degenerate bending modes of BA in the trans- gauche and gauche- gauche form, i.e., \(\delta (\text{tg- BA})\) and \(\delta (\text{g}^+ \text{g}^- \text{BA})\) , respectively. Besides, the intensity of \(\rho (\text{tt- BA})\) weakens considerably (Raman inactive- like) under compression, while the intensity of the Raman band at \(\sim 840 \text{cm}^{- 1}\) is enhanced due to the formation of Raman- active tg and \(\text{g}^+ \text{g}^-\) BA conformers. Above 3.2 GPa, the distortion of Pb- I sublattice lowers the lattice symmetry19, 20, which splits the degenerate bending and rocking bands of BA in tg and \(\text{g}^+ \text{g}^-\) conformations, as demonstrated by the Raman spectrum measured at 4.7 GPa. However, all Raman peaks becomes weak and disappears beyond 5.5 GPa due to the structural amorphization. The evolution of peak position and intensity of BA bending and rocking modes measured by the in- situ Raman spectra under compression (< 3.2 GPa) evidences the bond- by- bond rotation of the BA molecules to form distinct rotational isomers during lamellar contraction. More importantly, the irreversible changes of BA bending (Fig. 4d) and rocking (Fig. 4e) modes in the Raman spectra measured after decompression further proves the formation of BA tg and \(\text{g}^+ \text{g}^-\) conformers. Interestingly, the characteristic Raman bands of tg and \(\text{g}^+ \text{g}^-\) conformers remain after decompression from high- pressure amorphous state (9.7 GPa), thus the BA conformers are irreversibly strain- hardened, and this agrees very well with our DFT simulation shown in Fig. S1c. + +In the case of \(n = 1\) RPP, previous high- pressure Raman study by us showed that the BA chain remains linear under compression, and a totally reversible Raman spectrum is obtained after decompression21. + +<--- Page Split ---> + +## Broadband photoluminescence spectra in the exfoliated RPP crystals + +The changes in excitonic absorption or emission properties of bulk hybrid perovskite crystals under compression have been attributed to changes in bond lengths and bond angles of the inorganic octahedral cages, which affect the overlap of Pb 6s and I 5p orbitals19, 30. The exfoliated flakes are much thinner than bulk crystal and allow compression or relaxation effects to permeate much more quickly throughout the sample compared to the bulk (Fig. S6), thus upon decompression, the elastic potential energy stored in the inorganic sublattice can be released quickly. In addition, surface effects are strongly amplified in the mechanical exfoliated flakes29, 31, 32, as revealed by the pronounced PL emission located at the low- energy side in addition to the intrinsic excitonic emission as shown in Fig. S722. + +Fig. 5a shows the PL spectra of exfoliated \(n = 2\) RPP flakes with thickness of \(\sim 17 \text{nm}\) ( \(\sim 8\) layers) that are collected under compression, while Fig. 5b shows the one- to- one corresponding spectra after decompression. Before compression, the sample displays a clear intrinsic excitonic emission at \(\sim 2.14 \text{eV}\) arising from the bulk- like state (BS), this corresponds to the state where the majority of its BA molecules adopt the trans configuration typical of ground state alkane molecules. A low- energy PL emission centered at \(\sim 2.06 \text{eV}\) is related to surface state (SS) emission introduced by a small population of relaxed tt- conformer BA molecules33. Upon compression to 2.5 GPa, the BS and SS emissions red shift continuously, then they merge into a single peak at 3.9 GPa, which blue shifts with further increase in pressures until a drastic drop in PL intensity occurs due to amorphization (> 5.8 GPa), as shown in Fig. 5a. Importantly, below the pressure of 2.5 GPa, the PL spectra can recover to the initial state after decompression, as judged from the suppression of broadband emission at the low- energy side in Fig. 5b (highlighted by blue arrow). The ability of the PL to recover is due to the elastic recovery of the Pb- I sublattice, this is despite the fact that the BA cations have been irreversibly transformed to the high energy rotational isomers according to in- situ Raman. In other words, the soft organic cation buffers the compressive strain and allow the inorganic lattice to remain in the elastic regime without crossing the yield point. At pressure higher than 2.5 GPa, e.g., 3.9 GPa, the pressure yield point is exceeded in the stress- strain curves and the PL peak broadened irreversibly. DFT calculations of the state at this point revealed partial fracture of Pb- I bond, as shown in Fig. S1 and the seriously changed bond angles after decompression from higher pressures shown in Fig. 3d. The corresponding optical photographs and fluorescence micrographs before and after compression are shown in Fig. 5c, we can observe that emission is changed from original orange color to red. + +In the case of \(n = 1\) , the linear BA tilting along with the octahedral tilting in the out- of- plane direction works synergistically to accommodate elastic deformation, as demonstrated by + +<--- Page Split ---> + +the robust PL for \(n = 1\) RPP even at 9.85 GPa (Fig. S8a) and the recoverable PL after decompression (Fig. S8b). + +## DISCUSSION + +The role of the organic cations in buffering stress becomes clear when we examine the deformation resistance of the inorganic lattice for \(n = 1\) and 2 RPPs. As \(n\) increases, the ratio of the number of layers of inorganic slab (hard) to organic slab (soft) increases in the RPP homologous series from \(n = 1\) to 4, and varying bands of tensile and compressive stress develop respectively. When the proportion of BA organic fraction decreases, the ability to buffer compressive stress in the hybrid system decreases. As a result, there is increasing strain in the Pb- I inorganic lattice, leading to an increasingly lower pressure threshold from \(n = 1\) to 4 RPPs for permanent deformation, as demonstrated by the in- situ compression- decompression PL spectra shown in Fig. S8. The pressure threshold is reflected by the occurrence of non- recoverable PL after decompression for \(n = 1\) to 4 RPPs, as summarized in Fig. 5d. As \(n\) increases, the critical pressure point, above which irreversible deformation occurs, decreases and i.e., it requires a lower compressive stress to deform the inorganic lattice, as judged by the appearance of the broadband PL. This clearly corroborates the fact that the organic cation helps to buffer the stress during compression. Therefore, when the proportion of organic cations is increasingly diluted by thicker inorganic slab, the deformation resistance decreases. + +In \(n = 1\) RPP, the compressive stress is first accommodated by the tilting of the organic cation, and this acts in concert with the in- plane slip of the inorganic slab, which results in phase change of the crystal upon compression. The absence of methyl ammonium ions occupying the cubo- octahedral site in \(n = 1\) RPP imposes less steric hindrance, and this may allow the organic chain to tilt and un- tilt in a reversible way. By contrast, the compressive stress in \(n > 1\) RPPs is accommodated firstly by the rotational isomerism of the organic cations. The rotation to high energy conformer is accompanied by the in- plane tensile strain along \(a\) and \(b\) axis. Following decompression, elastic recovery of the inorganic octahedral cages in \(n > 1\) can be achieved, but the pressure threshold for irreversible deformation is much lower than the \(n = 1\) case. Our results provide the essential basis for understanding how multilayer organic- inorganic system act synergistically under compressive stress, which may be useful for understanding the isothermal barocaloric effects in perovskites, and for engineering new types of deformation- resistant multilayer coatings. + +## METHODS + +## Sample preparation + +<--- Page Split ---> + +The RPP single crystals were synthesized using three precursors; PbO, \(\mathrm{C_4H_9NH_3I}\) (BAI) and \(\mathrm{CH_3NH_3I}\) (MAI) via a temperature- programmed crystallization method34. For \(\mathrm{(BA)_2MAPb_2I_7}\) \((n = 2)\) , PbO, BAI and MAI (1.9/1.4/1 by molar, MAI: 0.31 M) were dissolved in a concentrated mixture of HI and \(\mathrm{H_3PO_2}\) (9:1, v/v). The solution was then heated at \(110^{\circ}\mathrm{C}\) while stirring for 40 minutes to give clear solutions. Subsequently, the solutions were quickly transferred to an oven set at \(110^{\circ}\mathrm{C}\) and allowed to cool slowly down to room temperature at a rate of \(3^{\circ}\mathrm{C / h}\) , in which the 2D RPP crystals crystalized. The crystals were isolated via vacuum filtration and subsequently dried under vacuum at room temperature28, 29. Thin flakes of RPPs were exfoliated by following the method developed for graphene exfoliation and transferred onto a culet of \(500\mu \mathrm{m}\) of the diamond anvil cell in the glovebox21. + +## High pressure environment realization + +Two identical diamond anvils with a culet of \(500\mu \mathrm{m}\) were employed to generate pressure. A stainless- steel gasket was pre- indented to \(50\mu \mathrm{m}\) with a drilled hole \(300\mu \mathrm{m}\) in diameter serving as the sample chamber. A small ruby ball was loaded together with \(n = 2\) RPP. The pressure was calibrated using the pressure- dependent ruby fluorescent technique30. + +## Raman and PL measurements during compression-decompression cycles + +Raman and PL measurements during compression- decompression cyclesRaman spectra were conducted by using Renishaw Invia Raman microscope. A solid- state \(785\mathrm{nm}\) laser was used to excite Raman scattering to avoid strong PL background as well as any sample degradation. For the high- pressure PL measurements, the bulk RPP crystals were first exfoliated on polydimethylsiloxane (PDMS) following the method developed for graphene exfoliation35, and perovskite flakes with different thicknesses were then selected and transferred onto the pressing surface ( \(500\mu \mathrm{m}\) ) of the DAC using a transfer stage. The exfoliation and transfer processes were carried out in an argon- filled glove box. The RPP microsheets were covered with a silicone oil before taking out of the glove box for in situ PL measurements under compression and decompression. PL spectra were excited by \(532\mathrm{nm}\) laser using WITec alpha 300RAS microscope. + +## First-principles calculations + +The structural evolutions under compression and decompression were simulated by using Vienna ab initio simulation package (VASP)36. To simulate the RPP under a series of different pressures, the RPP layers were sandwiched between two fixed hydrogenated nanodiamond films in with the top diamond layer was gradually shift down. The nanodiamond films were constructed by creating a diamond (111) \(3\times 2\) surface slab to match the lattice constants of \(n = 1\) and 2 RPPs where experimental in- plane values of \(a\) \(= 8.880\) and \(b = 8.697\mathrm{\AA}\) were used for \(n = 1\) and \(a = 8.859\) and \(b = 8.947\mathrm{\AA}\) were used for \(n = 2^{1}\) . The pressure exerted in the sandwiched RPP between the hydrogenated nanodiamond divided by the in- plane area of the cell. An additional vacuum layer of thickness greater than \(15\mathrm{\AA}\) is adopted. The Perdew- Burke- Ernzerhof functional for the exchange- correlation potential was used together with a kinetic energy cutoff of \(400\mathrm{eV}\) + +<--- Page Split ---> + +and a 3x3x1 Monkhorst- Pack grid. All the structures are fully relaxed until the force on each atom is less than \(0.005 \text{eV / \AA}\) . The van der Waal's interactions were treated by using DFT- D2 method. Raman intensities for BA molecules in gas form and in \(n = 2\) RPP were calculated by using the DMol package. + +## REFERENCES + +1. Stoumpos CC, et al. Ruddlesden-popper hybrid lead Iodide perovskite 2D homologous semiconductors. Chem. Mater. 28, 2852-2867 (2016). +2. Mao L, Stoumpos CC, Kanatzidis MG. Two-dimensional hybrid halide perovskites: principles and promises. J. Am. Chem. Soc. 141, 1171-1190 (2019). +3. Blancon J-C, Even J, Stoumpos CC, Kanatzidis MG, Mohite AD. Semiconductor physics of organic-inorganic 2D halide perovskites. Nat. Nanotechnol 15, 969-985 (2020). +4. Guo Z, Wu X, Zhu T, Zhu X, Huang L. Electron-Phonon Scattering in Atomically Thin 2D Perovskites. ACS Nano 10, 9992-9998 (2016). +5. Zhang F, Lu H, Tong J, Berry JJ, Beard MC, Zhu K. Advances in two-dimensional organic-inorganic hybrid perovskites. Energy Environ. Sci. 13, 1154-1186 (2020). +6. Tsai H, et al. Design principles for electronic charge transport in solution-processed vertically stacked 2D perovskite quantum wells. Nat. Commun. 9, 2130 (2018). +7. Grancini G, Nazeeruddin MK. Dimensional tailoring of hybrid perovskites for photovoltaics. Nat. Rev. Mater. 4, 4-22 (2019). +8. Dou L, et al. Atomically thin two-dimensional organic-inorganic hybrid perovskites. Science 349, 1518-1521 (2015). +9. Gan Z, Cheng Y, Chen W, Loh KP, Jia B, Wen X. Photophysics of 2D organic-inorganic hybrid lead halide perovskites: progress, debates, and challenges. Adv. Sci 8, 2001843 (2021). +10. Chen Y, Sun Y, Peng J, Tang J, Zheng K, Liang Z. 2D Ruddlesden-Popper perovskites for optoelectronics. Adv. Mater. 30, 1703487 (2018). +11. Mitzi DB. Templating and structural engineering in organic-inorganic perovskites. J. Chem. Soc., Dalton Trans., 1-12 (2001). +12. Liu S, et al. Manipulating efficient light emission in two-dimensional perovskite crystals by pressure-induced anisotropic deformation. Sci. Adv. 5, eaav9445 (2019). +13. Li H, et al. Unusual pressure-driven phase transformation and band renormalization in 2D vdW hybrid lead halide perovskites. Adv. Mater. 32, 1907364 (2020). +14. Kong L, et al. Highly tunable properties in pressure-treated two-dimensional Dion-Jacobson perovskites. Proc. Natl. Acad. Sci. U.S.A. 117, 16121-16126 (2020). +15. Guo S, et al. Pressure-suppressed carrier trapping leads to enhanced emission in two-dimensional perovskite (HA)2(GA)Pb2I7. Angew. Chem. Int. Ed. Engl. 59, 17533-17539 (2020). +16. Chen Y, et al. Emission enhancement and bandgap retention of a two-dimensional mixed cation lead halide perovskite under high pressure. J. Mater. Chem. A 7, 6357-6362 (2019). +17. Yuan Y, et al. Large band gap narrowing and prolonged carrier lifetime of (C4H9NH3)2PbI4 under high pressure. Adv. Sci 6, 1900240 (2019). +18. Zhang L, Wu L, Wang K, Zou B. Pressure-induced broadband emission of 2D organic-inorganic hybrid perovskite (C6H5C2H4NH3)2PbBr4. Adv. Sci 6, 1801628 (2019). +19. Liu G, et al. Two regimes of bandgap red shift and partial ambient retention in pressure-treated two-dimensional perovskites. ACS Energy Lett. 2, 2518-2524 (2017). +20. Liu G, et al. Isothermal pressure-derived metastable states in 2D hybrid perovskites showing enduring bandgap narrowing. Proc. Natl. Acad. Sci. U.S.A. 115, 8076-8081 (2018). + +<--- Page Split ---> + +21. Yin T, et al. Pressure-engineered structural and optical properties of two-dimensional \((\mathrm{C}_4\mathrm{H}_9\mathrm{NH}_3)_2\mathrm{PbI}_4\) perovskite exfoliated nm-thin flakes. J. Am. Chem. Soc. 141, 1235-1241 (2019). +22. Leng K, et al. Molecularly thin two-dimensional hybrid perovskites with tunable optoelectronic properties due to reversible surface relaxation. Nat. Mater. 17, 908-914 (2018). +23. McNulty JA, Lightfoot P. Structural chemistry of layered lead halide perovskites containing single octahedral layers. IUCrJ 8, 485-513 (2021). +24. Billing DG, Lemmerer A. Synthesis, characterization and phase transitions in the inorganic-organic layered perovskite-type hybrids [(C_nH_2n+1NH_3)_2PbI_4], n = 4, 5 and 6. Acta Crystallogr B 63, 735-747 (2007). +25. Marchenko EI, Korolev VV, Mitrofanov A, Fateev SA, Goodilin EA, Tarasov AB. Layer shift factor in layered hybrid perovskites: univocal quantitative descriptor of composition-structure-property relationships. Chem. Mater. 33, 1213-1217 (2021). +26. Yin T, et al. Hydrogen-bonding evolution during the polymorphic transformations in CH_3NH_3PbBr_3: experiment and theory. Chem. Mater. 29, 5974-5981 (2017). +27. Martins LGP, et al. Raman evidence for pressure-induced formation of diamondene. Nat. Commun. 8, 96 (2017). +28. Abdelwahab I, et al. Highly enhanced third-harmonic generation in 2D perovskites at excitonic resonances. ACS Nano 12, 644-650 (2018). +29. Leng K, Fu W, Liu Y, Chhowalla M, Loh KP. From bulk to molecularly thin hybrid perovskites. Nat. Rev. Mater. 5, 482-500 (2020). +30. Yin T, et al. High-pressure-induced comminution and recrystallization of CH_3NH_3PbBr_3 nanocrystals as large thin nanoplates. Adv. Mater. 30, 1705017 (2018). +31. Shi E, et al. Extrinsic and dynamic edge states of two-dimensional lead halide perovskites. ACS Nano 13, 1635-1644 (2019). +32. Kepenekian M, et al. Concept of lattice mismatch and emergence of surface states in two-dimensional hybrid perovskite quantum wells. Nano Lett. 18, 5603-5609 (2018). +33. Shao Y, et al. Unlocking surface octahedral tilt in two-dimensional Ruddlesden-Popper perovskites. Nat. Commun. 13, 138 (2022). +34. Mitzi DB. Synthesis, Crystal Structure, and Optical and Thermal Properties of (C4H9NH3)2MI4 (M = Ge, Sn, Pb). Chem. Mater. 8, 791-800 (1996). +35. Castellanos-Gomez A, et al. Deterministic transfer of two-dimensional materials by all-dry viscoelastic stamping. 2D Mater. 1, 011002 (2014). +36. Kresse G, Furthmüller J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169-11186 (1996). + +## Acknowledgements + +K. P. L. wishes to acknowledge Singapore's Ministry of Education Tier 2 grant MOE2019-T2-1-037 "Rashba Effect and Ferroelectricity in Two Dimensional Hybrid Perovskites". T. Y. and H. S. would like to acknowledge the support from the Singapore National Research Foundation via Competitive Research Programme (NRF-CRP21-2018-0007 and NRF-CRP23-2019-0007). K. L. would like to acknowledge the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. PolyU15305221 for GRF project funded in 2021/22 Exercise). Y. C. would like to acknowledge the University of Macau (SRG2019-00179-IAPME, MYRG2020-00075-IAPME) and the Science and Technology Development Fund from Macau SAR (FDCT-0163/2019/A3), the Natural + +<--- Page Split ---> + +Science Foundation of China (Grant 22022309) and Natural Science Foundation of Guangdong Province, China (2021A1515010024). This work was performed in part at the High- Performance Computing Cluster (HPCC), which is supported by the Information and Communication Technology Office (ICTO) of the University of Macau. + +## Author contributions + +T. Y. and K. P. L. conceived the idea and designed the experiments. T. Y. performed the high-pressure photoluminescence measurements. Y. L. helped high pressure Raman measurements. I. A. prepared the exfoliated ultrathin layered RPP samples in an argon-filled glovebox via micromechanical cleavage and 2D dry transfer and performed AFM analysis. H. Y. and Y. C. performed DFT calculations. K. L. synthesized the RPP crystals. T. Y., Y. C., Z. S. and K. P. L. analyzed the data. T. Y. and K. P. L. wrote the manuscript. All authors contributed to finalizing the manuscript. + +## Competing interests + +The authors declare no competing interests. + +## Additional information + +## Supplementary information + +Fig. S1. Simulation of the pressuring process of \(n = 2\) RPP (2L). + +Fig. S2. Simulation of the pressuring process of \(n = 1\) RPP (2L). + +Fig. S3. Top view of the layer shift for \(n = 1\) and 2 RPPs. + +Fig. S4. The evolution of the inter molecular dihedral angles of RPP under compression. + +Fig. S5. The evolution of in-plane and out-of-plane bond lengths of RPP under compression and decompression. + +Fig. S6. PL evolution under compression and decompression of \(n = 2\) bulk RPP. + +Fig. S7. Thickness- dependent PL spectra in exfoliated flakes of \(n = 2\) RRP as a function of layer number. + +Fig. S8. In situ PL spectra of \(n = 1\) , 3 and 4 RPP exfoliated thin flakes under compression and after one- to- one decompression. + +Table S1. Calculation by DMol of vibrational modes in the frequency region of 0- 700 cm \(^{- 1}\) of gas BA \(^+\) molecule in tt, tg and g \(^+\) g \(^-\) conformations. + +Table S2. Calculation by DMol of vibrational modes in the frequency region of 700- 1000 cm \(^{- 1}\) of gas BA \(^+\) molecule in tt, tg and g \(^+\) g \(^-\) conformations. + +Table S3. Comparison of frequencies of vibrational modes (700- 1000 cm \(^{- 1}\) ) of tt- and g \(^+\) g \(^-\) BA molecules in gas form and in \(n = 2\) 1L RPP structure. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Rotational isomerism of BA molecules during compressive stress of \(n = 2\) RPP. a Schematic showing diamond anvil cell and the \(n = 2\) RPP crystals with linear tt-BA organic chains. b DFT simulation of lamellar contraction and the step-wise rotational isomerism.
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2. Distinctive pressure responses of \(n = 1\) and \(n = 2\) RPP. a Two generic types of layer-shift mode: \(\Gamma_5^+\) for \(n = 1\) and \(M_5^-\) (a, b) for \(n = 2\) RPP. b, c Pressure-dependent in-plane layer shift factor (LSF) and vertical strain for \(n = 1\) and 2 RPPs.
+ +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3. Evolution of the inorganic lattice parameters of RPPs under compression and decompression. a Schematic of the 2D Pb-I octahedral cages, highlighting the apical and equatorial angle \(\alpha\) and \(\beta\) , and the in-plane and out-of-plane octahedral tilt angle \(\theta\) and \(\gamma\) , and bond length \(d_{11 - Pb}\) and \(d_{12 - Pb}\) , respectively. b Evolution of angle \(\alpha\) and \(\beta\) in \(n = 2\) under compression and decompression. Evolution of the in-plane and out-of-plane octahedral tiltings in \(n = 1\) RPP (c) and \(n = 2\) RPP (d) under compression and decompression. Lattice parameters for \(n = 2\) under compression is only measured at pressure below 3.16 GPa because severe lattice distortions at high pressures produce large errors.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4. Raman characterization of the BA tg- and \(\mathbf{g}^{*}\mathbf{g}\) - conformers. a, b Theoretical simulated Raman spectra for the BA bending and rocking modes in the tt, tg and \(\mathbf{g}^{*}\mathbf{g}\) structures. c Typical Raman spectral evolution of \(n = 2\) RPP with increasing pressure. d, e Raman spectral comparison of RPP sample before and after compression. \(\delta\) and \(\rho\) refers to bending and rocking vibrations. The grey, red and blue lines in (a, b) represent BA vibrations in the tt, tg and \(\mathbf{g}^{*}\mathbf{g}\) conformations, respectively Insets represent the corresponding vibration patterns of BA isomers, green arrows on the atoms indicate the vibrational directions and relative amplitude of displacements in the given Raman modes.
+ +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 PL characterization of the reversible and irreversible deformation in RPP. In-situ PL spectra of \(n = 2\) RPP; individual spectrum in (a) shows compression at specific “X” GPa, and corresponding spectrum in (b) after decompression of “X” GPa indicated as Re “X” GPa. The blue arrows in gradient color represent the proportion of the broadband emission. c Optical photographs and fluorescence images of \(n = 2\) RPP exfoliated flakes before and after pressure treatments. Scale bar is \(50 \mu \mathrm{m}\) . d Plot summarizing the threshold pressure to cross from elastic to plastic regime for \(n = 1\) to 4 RPPs, which is concomitant with the transition from a state where PL is reversible after decompression, to one where it is not. Dashed line guides to the eye.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformation.docx + +<--- Page Split ---> diff --git a/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7_det.mmd b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..3ec1357c6a1ef3966c801655f83a057ef9b7b838 --- /dev/null +++ b/preprint/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7/preprint__143572959b36e15d577c72f710ca232942be9593be26ca0975222b1ef5d7daa7_det.mmd @@ -0,0 +1,347 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 933, 175]]<|/det|> +# Pressure Driven Rotational Isomerism in 2D Hybrid Perovskites + +<|ref|>text<|/ref|><|det|>[[44, 196, 354, 238]]<|/det|> +TingTing YinNanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 243, 231, 284]]<|/det|> +Hejin YanUniversity of Macau + +<|ref|>text<|/ref|><|det|>[[44, 290, 696, 330]]<|/det|> +Ibrahim AbdelwahabNational University of Singapore https://orcid.org/0000- 0002- 0107- 5827 + +<|ref|>text<|/ref|><|det|>[[44, 335, 354, 376]]<|/det|> +Yulia LekinaNanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 381, 925, 445]]<|/det|> +Xujie LüCenter for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 8402- 7160 + +<|ref|>text<|/ref|><|det|>[[44, 451, 925, 515]]<|/det|> +Wenge YangCenter for High Pressure Science and Technology Advanced Research https://orcid.org/0000- 0001- 8436- 8731 + +<|ref|>text<|/ref|><|det|>[[44, 520, 710, 562]]<|/det|> +Handong SunNanyang Technological University https://orcid.org/0000- 0002- 2261- 7103 + +<|ref|>text<|/ref|><|det|>[[44, 567, 747, 608]]<|/det|> +Kai LengThe Hong Kong Polytechnic University https://orcid.org/0000- 0003- 3408- 5033 + +<|ref|>text<|/ref|><|det|>[[44, 613, 231, 653]]<|/det|> +Yongqing CaiUniversity of Macau + +<|ref|>text<|/ref|><|det|>[[44, 659, 700, 746]]<|/det|> +Ze ShenNanyang Technological UniversityKian Ping Loh ( chmlohkp@nus.edu.sg )National University of Singapore https://orcid.org/0000- 0002- 1491- 743X + +<|ref|>sub_title<|/ref|><|det|>[[44, 789, 101, 806]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 825, 790, 846]]<|/det|> +Keywords: 2D RPPs, Pressure, Rational isomerism, Tilting, Raman, photoluminescence + +<|ref|>text<|/ref|><|det|>[[44, 864, 315, 883]]<|/det|> +Posted Date: March 10th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 902, 473, 920]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1394948/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 910, 87]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[220, 89, 775, 108]]<|/det|> +# Pressure Driven Rotational Isomerism in 2D Hybrid Perovskites + +<|ref|>text<|/ref|><|det|>[[115, 127, 884, 169]]<|/det|> +Tingting Yin \(^{1,\dagger}\) , Hejin Yan \(^{2,\dagger}\) , Ibrahim Abdelwahab \(^{3}\) , Yulia Lekina \(^{1,4}\) , Xujie Lü \(^{5}\) , Wenge Yang \(^{5}\) , Handong Sun \(^{1}\) , Kai Leng \(^{6}\) , Yongqing Cai \(^{2*}\) , Zexiang Shen \(^{1,4*}\) and Kian Ping Loh \(^{3*}\) + +<|ref|>text<|/ref|><|det|>[[116, 178, 880, 420]]<|/det|> +\(^{1}\) Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore 637371, Singapore \(^{2}\) Joint Key Laboratory of Ministry of Education Institute of Applied Physics and Materials Engineering, University of Macau, Macau, China \(^{3}\) Centre for Advanced 2D Materials (CA2DM) and Department of Chemistry, National University of Singapore, Singapore 117543, Singapore \(^{4}\) Centre for Disruptive Photonic Technologies, the Photonics Institute, Nanyang Technological University, 637371, Singapore \(^{5}\) Center for High Pressure Science and Technology Advanced Research (HPSTAR), Shanghai 201203 (P. R. China) \(^{6}\) Department of Applied Physics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, China + +<|ref|>text<|/ref|><|det|>[[113, 605, 884, 666]]<|/det|> +\(^{\dagger}\) These authors contributed equally: Tingting YIN, Hejin YAN. \*Corresponding author. Email: yongqingcai@um.edu.mo; zexiang@ntu.edu.sg and chmlohkp@nus.edu.sg + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 89, 216, 106]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[114, 117, 883, 377]]<|/det|> +ABSTRACTMultilayers consisting of alternating soft and hard layers offer enhanced toughness compared to all- hard structures. However, shear instability usually exists in physically sputtered multilayers because of deformation incompatibility among hard and soft layers. Here, we demonstrate that 2D hybrid organic- inorganic perovskites (HOIP) provide an interesting platform to study the stress- strain behavior of hard and soft layers undulating with molecular scale periodicity. We investigate the phonon vibrations and photoluminescence properties of Ruddlesden- Popper perovskites (RPPs) under hydrostatic compression using a diamond anvil cell. The C4 alkyl chain in RPP as the organic spacer buffers compressive stress by tilting (for \(n = 1\) RPP) or step- wise rotational isomerism (for \(n > 1\) RPP) during compression. By examining the pressure threshold of the elastic recovery regime across \(n = 1\) to \(n = 4\) RPP perovskites, we obtained molecular insights into the relationship between structure and deformation resistance in hybrid organic- inorganic perovskites. + +<|ref|>text<|/ref|><|det|>[[115, 416, 880, 437]]<|/det|> +Keywords: 2D RPPs, Pressure, Rational isomerism, Tilting, Raman, photoluminescence + +<|ref|>sub_title<|/ref|><|det|>[[115, 477, 267, 496]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[114, 507, 883, 908]]<|/det|> +Two- dimensional (2D) Ruddlesden- Popper perovskites (RPPs) consist of alternatively stacked layers of soft organic layers and rigid inorganic layers with highly tunable optoelectronic properties \(^{1, 2}\) . The organic layers provide the dielectric and quantum confinement of the inorganic layers, giving rise to large exciton binding energies and high oscillator strength \(^{3 - 5}\) . From a structural mechanic point of view, the inorganic layers serve as mechanical brace to support spring- like molecular layers, forming a "natural" undulating hard- soft system. In the realm of mechanical engineering, such hard- soft multilayer system is highly sought after for its high plastic deformation resistance and increased fracture hardness, but require complex fabrication process. The undulating layers create bands of tensile and compressive stress that is different from that of single- component layers, thus research efforts on inorganic hard- soft multilayers are centered on engineering structure and compositional gradient to enhance the mechanical properties of materials. The general chemical formula for lead halide 2D RPPs is \((\mathrm{RNH}_3)_2\mathrm{Mn} - 1\mathrm{Pb}_n\mathrm{X}_{3n + 1} (n = 1, 2, 3, 4, \ldots)\) , where \(\mathrm{PbX}_4^{2 - }\) is an inorganic perovskite layer of corner- sharing metal halide octahedra, \(\mathrm{RNH}_3^+\) is a long- chain alkylammonium organic molecular layer, such as butyl ammonium (BA), and M is a smaller cation filled into the 12- fold coordinated holes among the \(\mathrm{PbX}_6\) octahedra, such as \(\mathrm{CH}_3\mathrm{NH}_3^+\) (MA \(^+\) ) \(^{6 - 8}\) . The long range structure of the organic array is a result of balancing local strain induced by methyl ammonium ions on the inorganic cage, hydrogen- bonding and electrostatic forces between the cationic head group of the organic molecule and the metal halide framework \(^{9}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 268]]<|/det|> +\(^{10}\) . Thanks to the rigid- /soft- layer alternated stacking structure of 2D layered perovskites, the rigid inorganic framework can template the conformation of the soft organic molecule confined in the organic- inorganic atomic interface \(^{11}\) . In the ground state of C4 alkyl chain, each \(\mathrm{CH}_2\) moiety adopts the trans- trans (tt) conformation to avoid steric repulsion. Under compression, it can be expected that the lamellar layers will contract and the strain in the metal halide bonds will affect optical and electronic properties \(^{12 - 21}\) . However, little is known about the conformational change of the organic cations and the plastic deformation or strain hardening process when the organic molecules re- orientate under pressure, this is because X- ray diffraction is mostly sensitive to the heavier inorganic components. + +<|ref|>text<|/ref|><|det|>[[114, 279, 883, 598]]<|/det|> +Here, using a combination of in- situ Raman and Photoluminescence (PL) spectroscopy, we investigate how compressive and tensile strain develops in the organic and inorganic components in (BA) \(_2\) MA \(_{n - 1}\) Pb \(_{n}\) 13 \(_{n + 1}\) ( \(n = 1\) , 2, 3, 4, ...) during compression. Combining experiments and density functional theory (DFT), we observed that \(n = 1\) and \(n > 1\) exhibit markedly different mechanisms of compressive strain under pressure. In \(n = 2\) RPP, under compression, the trans- trans (tt) structure of the butyl ammonium cation gradually transform into the trans- gauche (tg) and gauche- gauche (g \(^+\) g \(^-\) ) conformations via C- C and C- N bond rotations, along with tensile strain of the inorganic slabs in both \(a\) and \(b\) axial directions. Whereas for \(n = 1\) RPP, tilting of the linear organic cation is along with in- plane shift in one direction ( \(b\) axis), resulting in orthorhombic- to- monoclinic phase change under compression. \(n = 1\) RPP shows remarkable elastic recovery ability due to the flexibility of organic cation that acts synergistically with the octahedral tilt of the inorganic cages. Our study shows how differing alternation of soft and hard layers allows compressive and tensile stress to evolve differently across \(n = 1\) to \(n = 4\) RPPs, which provides the essential experimental basis of understanding stress- strain behavior in nanometer scale multilayer hard- soft superlattices. + +<|ref|>sub_title<|/ref|><|det|>[[115, 609, 209, 627]]<|/det|> +## RESULTS + +<|ref|>text<|/ref|><|det|>[[115, 640, 883, 718]]<|/det|> +To lay the groundwork for understanding the complex structural changes in the organic and inorganic parts of the hybrid perovskites upon hydrostatic compression, it is beneficial to organize the paper by discussing the changes revealed by DFT calculations first, followed by comparison with what was observed in the experiments. + +<|ref|>sub_title<|/ref|><|det|>[[115, 730, 556, 748]]<|/det|> +## Theoretical calculations of lamellar contraction + +<|ref|>text<|/ref|><|det|>[[115, 750, 883, 869]]<|/det|> +Figure 1a depicts a unit cell of \(n = 2\) RPP. Two layers of interdigitated linear organic BA molecules alternately stack with two layers of \(\mathrm{PbI}_4^{2 - }\) inorganic slabs. The unit cell height of \(\sim 2.1 \mathrm{nm}\) is determined by atomic force microscopy (AFM) \(^{22}\) . BA organic molecules in \(tt\) conformation dominate the large interlayer spaces, while MA ions occupy the interoctahedral voids. In the case of \(n = 1\) RPP, there are no MA ions and one layer of \(\mathrm{PbI}_4^{2 - }\) inorganic slab is alternatively stacked with two layers of BA molecules (Fig. S2). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 389]]<|/det|> +To simulate hydrostatic compression, the bi- layer \(n = 2\) RPP sample is sandwiched between two diamond slabs. The surfaces of the diamond slabs are fully saturated with hydrogen atoms to avoid any dangling bonds. Then, the bottom diamond layer is fixed while the top diamond layer is gradually shifted downward by \(0.4 \text{Å}\) per step, this allows the pressure to increase from 0 to 14 GPa in a stepwise manner (Fig. S1a). Under compression, the lamellar distance contracts. From 0 to 3.16 GPa, compressive stress is accommodated mainly by the deformation of the organic layers. At 0.27 GPa, the lamellar contraction is achieved by rotation of the \(tt\) isomer of BA to tg (green color) and \(\text{g}^+\text{g}^-\) (purple color) conformers on the surface of RPP, which can be accomplished by rotating along the C- C/C- N chain (Fig. 1b). The rotation of the organic cations allows the \(\text{Pbl}_4^{2 - }\) inorganic cage to resist deformation over a wide range of pressure and deforms severely only at relatively high pressures at 5.04 GPa (Fig. S1b). In contrast, the BA molecules are irreversibly transformed to the higher energy conformers, which remain even after decompression (Fig. S1c). The ability to generate new rotational conformers in RPP by compression is subsequently verified in our experiments shown later. + +<|ref|>text<|/ref|><|det|>[[115, 399, 883, 499]]<|/det|> +Compression results in the layer- to- layer shift of the bi- layer inorganic slabs in \(n = 2\) RPP along \(- a\) and \(+b\) axes, as represented by the blue and grey arrows in Fig. 2a (right side), which is related to a \(M_5^-(a,b)\) mode \(^{23}\) . Such an interlayer shift maintains \(n = 2\) RPP in the orthorhombic symmetry, consistent with the conformational change of BA molecules without tilting under pressure. + +<|ref|>text<|/ref|><|det|>[[115, 509, 883, 710]]<|/det|> +Interestingly, we observed pronounced differences in the pressure response of \(n = 1\) RPP compared to its higher homologues, as illustrated in Fig. 2. For \(n = 1\) RPP, the linear BA molecules tilt without undergoing conformational change till 7.35 GPa (Fig. S2). The molecular tilting drives the adjacent inorganic layer to shift in the same direction along \(+b\) axis, as represented by the grey arrows in Fig. 2a (left side), which is related to a \(\Gamma_5^+\) mode \(^{23}\) . The layer shift decreases the initial stacking offset and therefore the crystal undergoes a phase transition to the monoclinic class \(^{24}\) . Such pressure- induced phase transition is however absent in \(n > 1\) RPP. The different responses of \(n = 1\) and \(n > 1\) RPP to compression is related to their different organic- inorganic couplings, refer to Fig. S4 for details. + +<|ref|>sub_title<|/ref|><|det|>[[115, 722, 241, 739]]<|/det|> +## In-plane shift + +<|ref|>text<|/ref|><|det|>[[115, 741, 883, 904]]<|/det|> +To quantitatively compare the degree of in- plane shift between \(n = 1\) and 2 RPPs, we calculate the in- plane layer shift factor (LSF) \(^{25}\) , i.e., \(\Delta\) , in the low- pressure region ( \(< \sim 3\) GPa, where the lamellar contraction is dominated by BA organic layers). Refer to Fig. S3 for the top views of layer shift of \(n = 1\) and 2 RPPs under various pressures. Here, the LSF compares the in- plane coordinates of the vector \(\overline{l_1} - \overline{l_2}\) of each possible pair of Pb atoms \((l_1, l_2)\) in the first and second layers in the \(\sqrt{2} \times \sqrt{2}\) supercell metric \(^{25}\) . The calculated \(\Delta\) values are represented by the green filled circles for \(n = 1\) in Fig. 2b and green /blue filled cubes for \(n = 2\) in Fig. 2c. The strain in \(n = 1\) RPP is uniaxial along \(b\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 89, 882, 148]]<|/det|> +axis, with \(\Delta\) reaching \(\sim 0.33\) only with a pressure of 0.24 GPa. At 0.73 GPa, \(n = 2\) RPP undergoes strain along - a and \(b\) axis, with \(\Delta \sim 0.38\) along the \(b\) axis and 0.19 along the - a axis, respectively. + +<|ref|>text<|/ref|><|det|>[[115, 159, 882, 239]]<|/det|> +According to our calculations, the tilting of BA cations in \(n = 1\) RPP results in a vertical strain of \(\sim 16.9\%\) in response to a pressure of 2.51 GPa (violet filled circles in Fig. 2b), while the rotation isomerism of BA cations in \(n = 2\) RPP results in \(\sim 18.8\%\) vertical strain with a pressure of 2.31 GPa (violet filled cubes in Fig. 2c). + +<|ref|>sub_title<|/ref|><|det|>[[115, 250, 444, 268]]<|/det|> +## Compressibility and Octahedral tilt + +<|ref|>text<|/ref|><|det|>[[115, 270, 882, 368]]<|/det|> +To track the structural deformation and recoverability of the inorganic sublattice during compression- decompression circles, we track several key structural parameters that represent the in- plane and out- of- plane components of the Pb- l octahedral cage (Fig. 3a), namely, anionic cage angles \((\beta\) and \(\alpha\) ), octahedral tilting angles \((\gamma\) and \(\Theta\) ) and Pb- l bond lengths \((\mathrm{d}_{11 - \mathrm{Pb}}\) and \(\mathrm{d}_{12 - \mathrm{Pb}}\) ). + +<|ref|>text<|/ref|><|det|>[[114, 386, 883, 607]]<|/det|> +Under compression, \(n = 2\) RPP undergoes lamellar contraction by the rotational isomerism of the softer cations, this allows the harder Pb- l lattice to remain in the elastic regime when the pressure is less than \(\sim 3\) GPa. The apical angle \(\alpha\) linking the two \(\mathrm{PbI_4^{2 - }}\) octahedral cages serves as a pivot and bends by up to \(10^{\circ}\) during vertical compression, and unbends by almost the same degree after decompression (red symbols in Fig. 3b). The increase in \(\alpha\) beyond a certain bending angle reflects the pressure threshold in the \(n = 2\) system, beyond which irreversible deformation occurs. On the other hand, the equatorial angle \(\beta\) subtends by Pb- I- Pb in the \(ab\) plane (blue symbols in Fig. 3b) reflects the intra- layer compressive strain. Besides, the structural distortion of the inorganic sublattice can also be revealed by octahedral tilt, which occurs under pressures in both the in- plane and out- of- plane directions (solid symbols in Fig. 3d). + +<|ref|>text<|/ref|><|det|>[[114, 624, 883, 844]]<|/det|> +In the case of \(n = 1\) RPP, lamellar contraction occurs by the tilting of the organic cation (Fig. S2). At 7.35 GPa, the out- of- plane octahedral tilt \((8.9^{\circ})\) of the inorganic sublattice is more pronounced than the in- plane tilt \((\sim 0.9^{\circ}\) on average, see the solid symbols in Fig. 3c). These two tilt angles show good recoverability after decompression, as represented by the hollow symbols in Fig. 3c. It suggests that \(n = 1\) RPP resists vertical compression by the combination of the tilting of organic cation as well as out- of- plane octahedral tilt of the inorganic sublattice. An important difference is that the tilted organic cations in \(n = 1\) RPP can be untilted upon decompression, whereas the rotation isomers in \(n = 2\) RPP are locked and cannot be reverted to the low energy isomers upon decompression. The good compressibility of \(n = 1\) system is revealed by the elastically recoverability of tilt angles as well as bond lengths (Fig. S5a). + +<|ref|>sub_title<|/ref|><|det|>[[115, 855, 653, 874]]<|/det|> +## Raman spectroscopic evidence of rotated BA conformers + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 883, 208]]<|/det|> +Raman spectroscopy was commonly used to probe bond rotation of butyl chain during the pressurization process26, 27. The single crystal of \(n = 2\) RPP was synthesized through a temperature- programmed crystallization method28, 29, and the grounded powder sample was loaded into the diamond anvil cell (DAC) for in- situ high pressure Raman measurement. We used powder sample because it can serve as a pressure transmission medium to avoid the strong Raman signal background induced by using silicon oil. + +<|ref|>text<|/ref|><|det|>[[115, 219, 883, 279]]<|/det|> +As a reference, we have calculated the Raman spectra of \(\mathsf{BA}^+\) gas molecule in the tt, tg and \(\mathsf{g}^+ \mathsf{g}^-\) conformations, as represented by the grey, blue and red solid lines in Fig. 4a and b (see the detailed peak assignment in the Supplementary Discussion). + +<|ref|>text<|/ref|><|det|>[[113, 288, 884, 830]]<|/det|> +Our measured in- situ high- pressure Raman spectra of \(n = 2\) RPP under compression, as shown in Fig. 4c (see also Table S3), reproduces largely the Raman peak evolutions during trans to gauge conformational change of the \(\mathsf{BA}^+\) gas molecules. The measured Raman bands at 0 GPa in the frequency region of \(450 - 890 \text{cm}^{- 1}\) are the skeletal vibrational modes of the BA in the tt conformation, as represented by grey solid line in Fig. 4c. The Raman bands appearing at \(\sim 475.3\) and \(484.7 \text{cm}^{- 1}\) belong to \(\delta (\text{CCC})\) and \(\delta (\text{CCN})\) , and the observed characteristic band at \(\sim 864.7 \text{cm}^{- 1}\) belongs to \(\rho (\text{tt- BA})\) . Besides, there is a very weak band at \(\sim 837.2 \text{cm}^{- 1}\) originating from the rocking of \(\text{CH}_2\) (Table S2). Upon compression at 0.54 GPa, the Raman spectrum shows distinct changes. First, the \(\delta (\text{CCC})\) and \(\delta (\text{CCN})\) band almost merge into one band at \(\sim 487.2 \text{cm}^{- 1}\) , which is the degenerate bending modes of BA in the trans- gauche and gauche- gauche form, i.e., \(\delta (\text{tg- BA})\) and \(\delta (\text{g}^+ \text{g}^- \text{BA})\) , respectively. Besides, the intensity of \(\rho (\text{tt- BA})\) weakens considerably (Raman inactive- like) under compression, while the intensity of the Raman band at \(\sim 840 \text{cm}^{- 1}\) is enhanced due to the formation of Raman- active tg and \(\text{g}^+ \text{g}^-\) BA conformers. Above 3.2 GPa, the distortion of Pb- I sublattice lowers the lattice symmetry19, 20, which splits the degenerate bending and rocking bands of BA in tg and \(\text{g}^+ \text{g}^-\) conformations, as demonstrated by the Raman spectrum measured at 4.7 GPa. However, all Raman peaks becomes weak and disappears beyond 5.5 GPa due to the structural amorphization. The evolution of peak position and intensity of BA bending and rocking modes measured by the in- situ Raman spectra under compression (< 3.2 GPa) evidences the bond- by- bond rotation of the BA molecules to form distinct rotational isomers during lamellar contraction. More importantly, the irreversible changes of BA bending (Fig. 4d) and rocking (Fig. 4e) modes in the Raman spectra measured after decompression further proves the formation of BA tg and \(\text{g}^+ \text{g}^-\) conformers. Interestingly, the characteristic Raman bands of tg and \(\text{g}^+ \text{g}^-\) conformers remain after decompression from high- pressure amorphous state (9.7 GPa), thus the BA conformers are irreversibly strain- hardened, and this agrees very well with our DFT simulation shown in Fig. S1c. + +<|ref|>text<|/ref|><|det|>[[115, 840, 882, 900]]<|/det|> +In the case of \(n = 1\) RPP, previous high- pressure Raman study by us showed that the BA chain remains linear under compression, and a totally reversible Raman spectrum is obtained after decompression21. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 765, 108]]<|/det|> +## Broadband photoluminescence spectra in the exfoliated RPP crystals + +<|ref|>text<|/ref|><|det|>[[115, 108, 883, 308]]<|/det|> +The changes in excitonic absorption or emission properties of bulk hybrid perovskite crystals under compression have been attributed to changes in bond lengths and bond angles of the inorganic octahedral cages, which affect the overlap of Pb 6s and I 5p orbitals19, 30. The exfoliated flakes are much thinner than bulk crystal and allow compression or relaxation effects to permeate much more quickly throughout the sample compared to the bulk (Fig. S6), thus upon decompression, the elastic potential energy stored in the inorganic sublattice can be released quickly. In addition, surface effects are strongly amplified in the mechanical exfoliated flakes29, 31, 32, as revealed by the pronounced PL emission located at the low- energy side in addition to the intrinsic excitonic emission as shown in Fig. S722. + +<|ref|>text<|/ref|><|det|>[[114, 319, 883, 819]]<|/det|> +Fig. 5a shows the PL spectra of exfoliated \(n = 2\) RPP flakes with thickness of \(\sim 17 \text{nm}\) ( \(\sim 8\) layers) that are collected under compression, while Fig. 5b shows the one- to- one corresponding spectra after decompression. Before compression, the sample displays a clear intrinsic excitonic emission at \(\sim 2.14 \text{eV}\) arising from the bulk- like state (BS), this corresponds to the state where the majority of its BA molecules adopt the trans configuration typical of ground state alkane molecules. A low- energy PL emission centered at \(\sim 2.06 \text{eV}\) is related to surface state (SS) emission introduced by a small population of relaxed tt- conformer BA molecules33. Upon compression to 2.5 GPa, the BS and SS emissions red shift continuously, then they merge into a single peak at 3.9 GPa, which blue shifts with further increase in pressures until a drastic drop in PL intensity occurs due to amorphization (> 5.8 GPa), as shown in Fig. 5a. Importantly, below the pressure of 2.5 GPa, the PL spectra can recover to the initial state after decompression, as judged from the suppression of broadband emission at the low- energy side in Fig. 5b (highlighted by blue arrow). The ability of the PL to recover is due to the elastic recovery of the Pb- I sublattice, this is despite the fact that the BA cations have been irreversibly transformed to the high energy rotational isomers according to in- situ Raman. In other words, the soft organic cation buffers the compressive strain and allow the inorganic lattice to remain in the elastic regime without crossing the yield point. At pressure higher than 2.5 GPa, e.g., 3.9 GPa, the pressure yield point is exceeded in the stress- strain curves and the PL peak broadened irreversibly. DFT calculations of the state at this point revealed partial fracture of Pb- I bond, as shown in Fig. S1 and the seriously changed bond angles after decompression from higher pressures shown in Fig. 3d. The corresponding optical photographs and fluorescence micrographs before and after compression are shown in Fig. 5c, we can observe that emission is changed from original orange color to red. + +<|ref|>text<|/ref|><|det|>[[115, 830, 882, 869]]<|/det|> +In the case of \(n = 1\) , the linear BA tilting along with the octahedral tilting in the out- of- plane direction works synergistically to accommodate elastic deformation, as demonstrated by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 89, 882, 128]]<|/det|> +the robust PL for \(n = 1\) RPP even at 9.85 GPa (Fig. S8a) and the recoverable PL after decompression (Fig. S8b). + +<|ref|>sub_title<|/ref|><|det|>[[115, 140, 240, 158]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[114, 170, 883, 489]]<|/det|> +The role of the organic cations in buffering stress becomes clear when we examine the deformation resistance of the inorganic lattice for \(n = 1\) and 2 RPPs. As \(n\) increases, the ratio of the number of layers of inorganic slab (hard) to organic slab (soft) increases in the RPP homologous series from \(n = 1\) to 4, and varying bands of tensile and compressive stress develop respectively. When the proportion of BA organic fraction decreases, the ability to buffer compressive stress in the hybrid system decreases. As a result, there is increasing strain in the Pb- I inorganic lattice, leading to an increasingly lower pressure threshold from \(n = 1\) to 4 RPPs for permanent deformation, as demonstrated by the in- situ compression- decompression PL spectra shown in Fig. S8. The pressure threshold is reflected by the occurrence of non- recoverable PL after decompression for \(n = 1\) to 4 RPPs, as summarized in Fig. 5d. As \(n\) increases, the critical pressure point, above which irreversible deformation occurs, decreases and i.e., it requires a lower compressive stress to deform the inorganic lattice, as judged by the appearance of the broadband PL. This clearly corroborates the fact that the organic cation helps to buffer the stress during compression. Therefore, when the proportion of organic cations is increasingly diluted by thicker inorganic slab, the deformation resistance decreases. + +<|ref|>text<|/ref|><|det|>[[114, 500, 883, 779]]<|/det|> +In \(n = 1\) RPP, the compressive stress is first accommodated by the tilting of the organic cation, and this acts in concert with the in- plane slip of the inorganic slab, which results in phase change of the crystal upon compression. The absence of methyl ammonium ions occupying the cubo- octahedral site in \(n = 1\) RPP imposes less steric hindrance, and this may allow the organic chain to tilt and un- tilt in a reversible way. By contrast, the compressive stress in \(n > 1\) RPPs is accommodated firstly by the rotational isomerism of the organic cations. The rotation to high energy conformer is accompanied by the in- plane tensile strain along \(a\) and \(b\) axis. Following decompression, elastic recovery of the inorganic octahedral cages in \(n > 1\) can be achieved, but the pressure threshold for irreversible deformation is much lower than the \(n = 1\) case. Our results provide the essential basis for understanding how multilayer organic- inorganic system act synergistically under compressive stress, which may be useful for understanding the isothermal barocaloric effects in perovskites, and for engineering new types of deformation- resistant multilayer coatings. + +<|ref|>sub_title<|/ref|><|det|>[[115, 821, 215, 839]]<|/det|> +## METHODS + +<|ref|>sub_title<|/ref|><|det|>[[115, 851, 300, 869]]<|/det|> +## Sample preparation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 308]]<|/det|> +The RPP single crystals were synthesized using three precursors; PbO, \(\mathrm{C_4H_9NH_3I}\) (BAI) and \(\mathrm{CH_3NH_3I}\) (MAI) via a temperature- programmed crystallization method34. For \(\mathrm{(BA)_2MAPb_2I_7}\) \((n = 2)\) , PbO, BAI and MAI (1.9/1.4/1 by molar, MAI: 0.31 M) were dissolved in a concentrated mixture of HI and \(\mathrm{H_3PO_2}\) (9:1, v/v). The solution was then heated at \(110^{\circ}\mathrm{C}\) while stirring for 40 minutes to give clear solutions. Subsequently, the solutions were quickly transferred to an oven set at \(110^{\circ}\mathrm{C}\) and allowed to cool slowly down to room temperature at a rate of \(3^{\circ}\mathrm{C / h}\) , in which the 2D RPP crystals crystalized. The crystals were isolated via vacuum filtration and subsequently dried under vacuum at room temperature28, 29. Thin flakes of RPPs were exfoliated by following the method developed for graphene exfoliation and transferred onto a culet of \(500\mu \mathrm{m}\) of the diamond anvil cell in the glovebox21. + +<|ref|>sub_title<|/ref|><|det|>[[115, 320, 475, 338]]<|/det|> +## High pressure environment realization + +<|ref|>text<|/ref|><|det|>[[115, 339, 882, 419]]<|/det|> +Two identical diamond anvils with a culet of \(500\mu \mathrm{m}\) were employed to generate pressure. A stainless- steel gasket was pre- indented to \(50\mu \mathrm{m}\) with a drilled hole \(300\mu \mathrm{m}\) in diameter serving as the sample chamber. A small ruby ball was loaded together with \(n = 2\) RPP. The pressure was calibrated using the pressure- dependent ruby fluorescent technique30. + +<|ref|>sub_title<|/ref|><|det|>[[115, 429, 808, 449]]<|/det|> +## Raman and PL measurements during compression-decompression cycles + +<|ref|>text<|/ref|><|det|>[[114, 450, 883, 650]]<|/det|> +Raman and PL measurements during compression- decompression cyclesRaman spectra were conducted by using Renishaw Invia Raman microscope. A solid- state \(785\mathrm{nm}\) laser was used to excite Raman scattering to avoid strong PL background as well as any sample degradation. For the high- pressure PL measurements, the bulk RPP crystals were first exfoliated on polydimethylsiloxane (PDMS) following the method developed for graphene exfoliation35, and perovskite flakes with different thicknesses were then selected and transferred onto the pressing surface ( \(500\mu \mathrm{m}\) ) of the DAC using a transfer stage. The exfoliation and transfer processes were carried out in an argon- filled glove box. The RPP microsheets were covered with a silicone oil before taking out of the glove box for in situ PL measurements under compression and decompression. PL spectra were excited by \(532\mathrm{nm}\) laser using WITec alpha 300RAS microscope. + +<|ref|>sub_title<|/ref|><|det|>[[116, 660, 378, 679]]<|/det|> +## First-principles calculations + +<|ref|>text<|/ref|><|det|>[[114, 681, 883, 899]]<|/det|> +The structural evolutions under compression and decompression were simulated by using Vienna ab initio simulation package (VASP)36. To simulate the RPP under a series of different pressures, the RPP layers were sandwiched between two fixed hydrogenated nanodiamond films in with the top diamond layer was gradually shift down. The nanodiamond films were constructed by creating a diamond (111) \(3\times 2\) surface slab to match the lattice constants of \(n = 1\) and 2 RPPs where experimental in- plane values of \(a\) \(= 8.880\) and \(b = 8.697\mathrm{\AA}\) were used for \(n = 1\) and \(a = 8.859\) and \(b = 8.947\mathrm{\AA}\) were used for \(n = 2^{1}\) . The pressure exerted in the sandwiched RPP between the hydrogenated nanodiamond divided by the in- plane area of the cell. An additional vacuum layer of thickness greater than \(15\mathrm{\AA}\) is adopted. The Perdew- Burke- Ernzerhof functional for the exchange- correlation potential was used together with a kinetic energy cutoff of \(400\mathrm{eV}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 883, 169]]<|/det|> +and a 3x3x1 Monkhorst- Pack grid. All the structures are fully relaxed until the force on each atom is less than \(0.005 \text{eV / \AA}\) . The van der Waal's interactions were treated by using DFT- D2 method. Raman intensities for BA molecules in gas form and in \(n = 2\) RPP were calculated by using the DMol package. + +<|ref|>sub_title<|/ref|><|det|>[[114, 200, 252, 216]]<|/det|> +## REFERENCES + +<|ref|>text<|/ref|><|det|>[[110, 220, 886, 900]]<|/det|> +1. Stoumpos CC, et al. Ruddlesden-popper hybrid lead Iodide perovskite 2D homologous semiconductors. Chem. Mater. 28, 2852-2867 (2016). +2. Mao L, Stoumpos CC, Kanatzidis MG. Two-dimensional hybrid halide perovskites: principles and promises. J. Am. Chem. Soc. 141, 1171-1190 (2019). +3. Blancon J-C, Even J, Stoumpos CC, Kanatzidis MG, Mohite AD. Semiconductor physics of organic-inorganic 2D halide perovskites. Nat. Nanotechnol 15, 969-985 (2020). +4. Guo Z, Wu X, Zhu T, Zhu X, Huang L. Electron-Phonon Scattering in Atomically Thin 2D Perovskites. ACS Nano 10, 9992-9998 (2016). +5. Zhang F, Lu H, Tong J, Berry JJ, Beard MC, Zhu K. Advances in two-dimensional organic-inorganic hybrid perovskites. Energy Environ. Sci. 13, 1154-1186 (2020). +6. Tsai H, et al. Design principles for electronic charge transport in solution-processed vertically stacked 2D perovskite quantum wells. Nat. Commun. 9, 2130 (2018). +7. Grancini G, Nazeeruddin MK. Dimensional tailoring of hybrid perovskites for photovoltaics. Nat. Rev. Mater. 4, 4-22 (2019). +8. Dou L, et al. Atomically thin two-dimensional organic-inorganic hybrid perovskites. Science 349, 1518-1521 (2015). +9. Gan Z, Cheng Y, Chen W, Loh KP, Jia B, Wen X. Photophysics of 2D organic-inorganic hybrid lead halide perovskites: progress, debates, and challenges. Adv. Sci 8, 2001843 (2021). +10. Chen Y, Sun Y, Peng J, Tang J, Zheng K, Liang Z. 2D Ruddlesden-Popper perovskites for optoelectronics. Adv. Mater. 30, 1703487 (2018). +11. Mitzi DB. Templating and structural engineering in organic-inorganic perovskites. J. Chem. Soc., Dalton Trans., 1-12 (2001). +12. Liu S, et al. Manipulating efficient light emission in two-dimensional perovskite crystals by pressure-induced anisotropic deformation. Sci. Adv. 5, eaav9445 (2019). +13. Li H, et al. Unusual pressure-driven phase transformation and band renormalization in 2D vdW hybrid lead halide perovskites. Adv. Mater. 32, 1907364 (2020). +14. Kong L, et al. Highly tunable properties in pressure-treated two-dimensional Dion-Jacobson perovskites. Proc. Natl. Acad. Sci. U.S.A. 117, 16121-16126 (2020). +15. Guo S, et al. Pressure-suppressed carrier trapping leads to enhanced emission in two-dimensional perovskite (HA)2(GA)Pb2I7. Angew. Chem. Int. Ed. Engl. 59, 17533-17539 (2020). +16. Chen Y, et al. Emission enhancement and bandgap retention of a two-dimensional mixed cation lead halide perovskite under high pressure. J. Mater. Chem. A 7, 6357-6362 (2019). +17. Yuan Y, et al. Large band gap narrowing and prolonged carrier lifetime of (C4H9NH3)2PbI4 under high pressure. Adv. Sci 6, 1900240 (2019). +18. Zhang L, Wu L, Wang K, Zou B. Pressure-induced broadband emission of 2D organic-inorganic hybrid perovskite (C6H5C2H4NH3)2PbBr4. Adv. Sci 6, 1801628 (2019). +19. Liu G, et al. Two regimes of bandgap red shift and partial ambient retention in pressure-treated two-dimensional perovskites. ACS Energy Lett. 2, 2518-2524 (2017). +20. Liu G, et al. Isothermal pressure-derived metastable states in 2D hybrid perovskites showing enduring bandgap narrowing. Proc. Natl. Acad. Sci. U.S.A. 115, 8076-8081 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 884, 668]]<|/det|> +21. Yin T, et al. Pressure-engineered structural and optical properties of two-dimensional \((\mathrm{C}_4\mathrm{H}_9\mathrm{NH}_3)_2\mathrm{PbI}_4\) perovskite exfoliated nm-thin flakes. J. Am. Chem. Soc. 141, 1235-1241 (2019). +22. Leng K, et al. Molecularly thin two-dimensional hybrid perovskites with tunable optoelectronic properties due to reversible surface relaxation. Nat. Mater. 17, 908-914 (2018). +23. McNulty JA, Lightfoot P. Structural chemistry of layered lead halide perovskites containing single octahedral layers. IUCrJ 8, 485-513 (2021). +24. Billing DG, Lemmerer A. Synthesis, characterization and phase transitions in the inorganic-organic layered perovskite-type hybrids [(C_nH_2n+1NH_3)_2PbI_4], n = 4, 5 and 6. Acta Crystallogr B 63, 735-747 (2007). +25. Marchenko EI, Korolev VV, Mitrofanov A, Fateev SA, Goodilin EA, Tarasov AB. Layer shift factor in layered hybrid perovskites: univocal quantitative descriptor of composition-structure-property relationships. Chem. Mater. 33, 1213-1217 (2021). +26. Yin T, et al. Hydrogen-bonding evolution during the polymorphic transformations in CH_3NH_3PbBr_3: experiment and theory. Chem. Mater. 29, 5974-5981 (2017). +27. Martins LGP, et al. Raman evidence for pressure-induced formation of diamondene. Nat. Commun. 8, 96 (2017). +28. Abdelwahab I, et al. Highly enhanced third-harmonic generation in 2D perovskites at excitonic resonances. ACS Nano 12, 644-650 (2018). +29. Leng K, Fu W, Liu Y, Chhowalla M, Loh KP. From bulk to molecularly thin hybrid perovskites. Nat. Rev. Mater. 5, 482-500 (2020). +30. Yin T, et al. High-pressure-induced comminution and recrystallization of CH_3NH_3PbBr_3 nanocrystals as large thin nanoplates. Adv. Mater. 30, 1705017 (2018). +31. Shi E, et al. Extrinsic and dynamic edge states of two-dimensional lead halide perovskites. ACS Nano 13, 1635-1644 (2019). +32. Kepenekian M, et al. Concept of lattice mismatch and emergence of surface states in two-dimensional hybrid perovskite quantum wells. Nano Lett. 18, 5603-5609 (2018). +33. Shao Y, et al. Unlocking surface octahedral tilt in two-dimensional Ruddlesden-Popper perovskites. Nat. Commun. 13, 138 (2022). +34. Mitzi DB. Synthesis, Crystal Structure, and Optical and Thermal Properties of (C4H9NH3)2MI4 (M = Ge, Sn, Pb). Chem. Mater. 8, 791-800 (1996). +35. Castellanos-Gomez A, et al. Deterministic transfer of two-dimensional materials by all-dry viscoelastic stamping. 2D Mater. 1, 011002 (2014). +36. Kresse G, Furthmüller J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169-11186 (1996). + +<|ref|>sub_title<|/ref|><|det|>[[116, 706, 301, 723]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[114, 725, 884, 905]]<|/det|> +K. P. L. wishes to acknowledge Singapore's Ministry of Education Tier 2 grant MOE2019-T2-1-037 "Rashba Effect and Ferroelectricity in Two Dimensional Hybrid Perovskites". T. Y. and H. S. would like to acknowledge the support from the Singapore National Research Foundation via Competitive Research Programme (NRF-CRP21-2018-0007 and NRF-CRP23-2019-0007). K. L. would like to acknowledge the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. PolyU15305221 for GRF project funded in 2021/22 Exercise). Y. C. would like to acknowledge the University of Macau (SRG2019-00179-IAPME, MYRG2020-00075-IAPME) and the Science and Technology Development Fund from Macau SAR (FDCT-0163/2019/A3), the Natural + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 88, 883, 170]]<|/det|> +Science Foundation of China (Grant 22022309) and Natural Science Foundation of Guangdong Province, China (2021A1515010024). This work was performed in part at the High- Performance Computing Cluster (HPCC), which is supported by the Information and Communication Technology Office (ICTO) of the University of Macau. + +<|ref|>sub_title<|/ref|><|det|>[[115, 200, 312, 218]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[114, 220, 883, 359]]<|/det|> +T. Y. and K. P. L. conceived the idea and designed the experiments. T. Y. performed the high-pressure photoluminescence measurements. Y. L. helped high pressure Raman measurements. I. A. prepared the exfoliated ultrathin layered RPP samples in an argon-filled glovebox via micromechanical cleavage and 2D dry transfer and performed AFM analysis. H. Y. and Y. C. performed DFT calculations. K. L. synthesized the RPP crystals. T. Y., Y. C., Z. S. and K. P. L. analyzed the data. T. Y. and K. P. L. wrote the manuscript. All authors contributed to finalizing the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[115, 370, 307, 388]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[115, 391, 499, 409]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 420, 327, 439]]<|/det|> +## Additional information + +<|ref|>sub_title<|/ref|><|det|>[[115, 450, 373, 469]]<|/det|> +## Supplementary information + +<|ref|>text<|/ref|><|det|>[[115, 470, 671, 490]]<|/det|> +Fig. S1. Simulation of the pressuring process of \(n = 2\) RPP (2L). + +<|ref|>text<|/ref|><|det|>[[115, 491, 670, 510]]<|/det|> +Fig. S2. Simulation of the pressuring process of \(n = 1\) RPP (2L). + +<|ref|>text<|/ref|><|det|>[[115, 511, 605, 530]]<|/det|> +Fig. S3. Top view of the layer shift for \(n = 1\) and 2 RPPs. + +<|ref|>text<|/ref|><|det|>[[115, 531, 880, 550]]<|/det|> +Fig. S4. The evolution of the inter molecular dihedral angles of RPP under compression. + +<|ref|>text<|/ref|><|det|>[[115, 551, 883, 590]]<|/det|> +Fig. S5. The evolution of in-plane and out-of-plane bond lengths of RPP under compression and decompression. + +<|ref|>text<|/ref|><|det|>[[115, 590, 810, 610]]<|/det|> +Fig. S6. PL evolution under compression and decompression of \(n = 2\) bulk RPP. + +<|ref|>text<|/ref|><|det|>[[115, 611, 883, 650]]<|/det|> +Fig. S7. Thickness- dependent PL spectra in exfoliated flakes of \(n = 2\) RRP as a function of layer number. + +<|ref|>text<|/ref|><|det|>[[115, 650, 883, 689]]<|/det|> +Fig. S8. In situ PL spectra of \(n = 1\) , 3 and 4 RPP exfoliated thin flakes under compression and after one- to- one decompression. + +<|ref|>text<|/ref|><|det|>[[115, 690, 883, 730]]<|/det|> +Table S1. Calculation by DMol of vibrational modes in the frequency region of 0- 700 cm \(^{- 1}\) of gas BA \(^+\) molecule in tt, tg and g \(^+\) g \(^-\) conformations. + +<|ref|>text<|/ref|><|det|>[[115, 730, 883, 770]]<|/det|> +Table S2. Calculation by DMol of vibrational modes in the frequency region of 700- 1000 cm \(^{- 1}\) of gas BA \(^+\) molecule in tt, tg and g \(^+\) g \(^-\) conformations. + +<|ref|>text<|/ref|><|det|>[[115, 770, 883, 810]]<|/det|> +Table S3. Comparison of frequencies of vibrational modes (700- 1000 cm \(^{- 1}\) ) of tt- and g \(^+\) g \(^-\) BA molecules in gas form and in \(n = 2\) 1L RPP structure. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 100, 878, 553]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 574, 883, 654]]<|/det|> +
Fig. 1. Rotational isomerism of BA molecules during compressive stress of \(n = 2\) RPP. a Schematic showing diamond anvil cell and the \(n = 2\) RPP crystals with linear tt-BA organic chains. b DFT simulation of lamellar contraction and the step-wise rotational isomerism.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 81, 884, 550]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[114, 562, 884, 621]]<|/det|> +
Fig. 2. Distinctive pressure responses of \(n = 1\) and \(n = 2\) RPP. a Two generic types of layer-shift mode: \(\Gamma_5^+\) for \(n = 1\) and \(M_5^-\) (a, b) for \(n = 2\) RPP. b, c Pressure-dependent in-plane layer shift factor (LSF) and vertical strain for \(n = 1\) and 2 RPPs.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 88, 886, 592]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 602, 884, 784]]<|/det|> +
Fig. 3. Evolution of the inorganic lattice parameters of RPPs under compression and decompression. a Schematic of the 2D Pb-I octahedral cages, highlighting the apical and equatorial angle \(\alpha\) and \(\beta\) , and the in-plane and out-of-plane octahedral tilt angle \(\theta\) and \(\gamma\) , and bond length \(d_{11 - Pb}\) and \(d_{12 - Pb}\) , respectively. b Evolution of angle \(\alpha\) and \(\beta\) in \(n = 2\) under compression and decompression. Evolution of the in-plane and out-of-plane octahedral tiltings in \(n = 1\) RPP (c) and \(n = 2\) RPP (d) under compression and decompression. Lattice parameters for \(n = 2\) under compression is only measured at pressure below 3.16 GPa because severe lattice distortions at high pressures produce large errors.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 91, 884, 465]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 477, 884, 638]]<|/det|> +
Fig. 4. Raman characterization of the BA tg- and \(\mathbf{g}^{*}\mathbf{g}\) - conformers. a, b Theoretical simulated Raman spectra for the BA bending and rocking modes in the tt, tg and \(\mathbf{g}^{*}\mathbf{g}\) structures. c Typical Raman spectral evolution of \(n = 2\) RPP with increasing pressure. d, e Raman spectral comparison of RPP sample before and after compression. \(\delta\) and \(\rho\) refers to bending and rocking vibrations. The grey, red and blue lines in (a, b) represent BA vibrations in the tt, tg and \(\mathbf{g}^{*}\mathbf{g}\) conformations, respectively Insets represent the corresponding vibration patterns of BA isomers, green arrows on the atoms indicate the vibrational directions and relative amplitude of displacements in the given Raman modes.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 94, 880, 389]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 404, 883, 584]]<|/det|> +
Fig. 5 PL characterization of the reversible and irreversible deformation in RPP. In-situ PL spectra of \(n = 2\) RPP; individual spectrum in (a) shows compression at specific “X” GPa, and corresponding spectrum in (b) after decompression of “X” GPa indicated as Re “X” GPa. The blue arrows in gradient color represent the proportion of the broadband emission. c Optical photographs and fluorescence images of \(n = 2\) RPP exfoliated flakes before and after pressure treatments. Scale bar is \(50 \mu \mathrm{m}\) . d Plot summarizing the threshold pressure to cross from elastic to plastic regime for \(n = 1\) to 4 RPPs, which is concomitant with the transition from a state where PL is reversible after decompression, to one where it is not. Dashed line guides to the eye.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 131, 366, 150]]<|/det|> +SupplementaryInformation.docx + +<--- Page Split ---> diff --git a/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/images_list.json b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..318ba4a8fea339125aa8f902e0e59589aee79164 --- /dev/null +++ b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/images_list.json @@ -0,0 +1,107 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1: A previously unrecognized alternative promoter drives IL-33 receptor expression in antiviral T cells. a, Scheme depicting the curated Il1rl1 gene. b, ST2 surface expression by in vitro differentiated T cell subsets. c, Il1rl1 first exon usage by differentiated T cells (CTL, Th2: \\(n = 4\\) , Th1: \\(n = 3\\) , NIH3T3: \\(n = 2\\) ). d, RNA-Seq coverage and splice junction tracks of ST2+ CTLs, Th1, and Th2 cells ( \\(n = 3\\) per subset) at the Il1rl1 locus. Chr1:40,377,000-40,465,500; GRCm38.p6/mm10 is shown. e-g, Wildtype mice received LCMV-specific P14 or Smarta T cells and were infected with LCMV-WE. P14 CTLs and Smarta Th1 cells were isolated on d7 p.i., and Il1rl1 first exon usage was analyzed by qPCR (e; CTL, Th1: \\(n = 4\\) , Th2 control (in vitro): \\(n = 2\\) ) and RT-PCR (f). g, Chip-Seq tracks indicating T-bet36, STAT438, and GATA-337 binding and ATAC-Seq tracks showing chromatin accessibility in naive or activated", + "footnote": [], + "bbox": [ + [ + 115, + 110, + 880, + 620 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2: Alternative promoter usage at the ll1rl1 locus is conserved between mice", + "footnote": [], + "bbox": [ + [ + 115, + 81, + 876, + 333 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3: Usage of distinct promoters allows T cell subset-specific targeting of ST2 expression. a, Experimental outline and representative FACS plot showing GFP expression by transduced T cells. b,c, Representative FACS plots (b) and quantification (c) of ST2 surface expression by transduced T cells analyzed after 5 days of culture ( \\(n = 6\\) pooled from two (CTL) or three (Th1, Th2) independent experiments). d, Scheme depicting generation of Il1rl1-ExAB \\(^{-/ - }\\) and Il1rl1-ExC \\(^{-/ - }\\) mice using CRISPR/Cas9 in murine zygotes. e,f, Representative histograms (e) and quantification (f) of ST2 surface expression by WT, Il1rl1-ExAB \\(^{-/ - }\\) , or Il1rl1-ExC \\(^{-/ - }\\) T cells ( \\(n = 4\\) ). g, IFN- \\(\\gamma\\) and IL-13 secretion by WT or Il1rl1-ExAB \\(^{-/ - }\\) T cells after IL-33 stimulation ( \\(n = 3\\) ). h-n, Wildtype, Il1rl1- \\(^{-/ - }\\) , Il1rl1-ExAB \\(^{-/ - }\\) , and Il1rl1-ExC \\(^{-/ - }\\) mice were infected with LCMV-WE and ST2 expression by splenic CTLs (i,j), Th1 cells (k,l), and Tregs (m,n) was quantified on d7 p.i. (WT: \\(n = 9\\) , Il1rl1- \\(^{-/ - }\\) : \\(n = 6\\) , Il1rl1-ExAB \\(^{-/ - }\\) : \\(n = 8\\) , Il1rl1-ExC \\(^{-/ - }\\) : \\(n = 7\\) ). Data represent one (g), two (e,f), or three (i-n) independent experiments and are presented as mean \\(\\pm\\) SD with each dot representing one mouse (j,l,n) or one experiment performed with T cells from individual mice (c,f). \\(P\\) was determined using One-way (j,l,n) or Two-way (c,f) ANOVA with Tukey's post-hoc test.", + "footnote": [], + "bbox": [ + [ + 113, + 80, + 884, + 499 + ] + ], + "page_idx": 28 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4: T cell-intrinsic activity of the type-1 Il1rl1 promoter is vital for efficient expansion of antiviral T cells. a-h, Wildtype, Il1rl1-/-; Il1rl1-ExAB-/- and Il1rl1-ExC-/- mice were infected with LCMV-WE, and splenic T cells were analyzed on d7 p.i. (WT: \\(n = 9\\) , Il1rl1-/-: \\(n = 6\\) , Il1rl1-ExAB-/-: \\(n = 8\\) , Il1rl1-ExC-/-: \\(n = 7\\) ). a, Experimental outline. b, c, Frequencies (b) and absolute cell counts (c) of CTLs. d-f, Representative staining (d) and frequencies of activated CD44+ CD62L- CTLs (e) and Ki67+ CTLs (f). g, Absolute cell counts of LCMV GP33-41- or NP396-404-specific CTLs. h, Counts of CTLs expressing effector molecules after ex vivo stimulation with GP33-41. i, Serum IFN-γ levels at indicated timepoints p.i. with LCMV-WE (WT and Il1rl1-/-: \\(n = 4\\) , Il1rl1-ExAB-/-: \\(n = 5\\) , Il1rl1-ExC-/-: \\(n = 3\\) ). j-m, Irradiated WT recipients (CD45.1+) were reconstituted with WT (CD45.1+ CD45.2+) and Il1rl1-ExAB-/- or Il1rl1-ExC-/- (all CD45.2+) bone marrow, infected with LCMV-WE, and analyzed on d10 p.i. (WT + Il1rl1-/-: \\(n = 6\\) ,", + "footnote": [], + "bbox": [ + [ + 113, + 81, + 884, + 615 + ] + ], + "page_idx": 29 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5: Type-1 llrl1 promoter engagement facilitates effector differentiation of CTLs to generate a clonally diverse SLEC population. a-d, n-r, Wildtype and llrl1-ExAB-/- mice were infected with LCMV-WE, and splenic CD44+ CTLs were analyzed by multiplexed scRNA-Seq and scTCR-Seq analysis on d7 p.i. (n = 3 mice pooled per group). a, Experimental outline. b, UMAP plots colored by cluster type. c, UMAP plots showing normalized expression of selected genes in both genotypes. d, Change in cluster composition of llrl1-ExAB-/- CTLs relative to WT CTLs. e, Representative FACS plots showing KLRG1 and CD127 expression by CD44+ CD8+ T cells. f-h, Frequencies (f) and counts (g) of KLRG1+ CD127- SLECs and frequencies (h) of", + "footnote": [], + "bbox": [ + [ + 118, + 75, + 880, + 670 + ] + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6: Transcriptional profiling indicates broad costimulatory and TCR-cooperative functions of IL-33–ST2 signaling in T cells. a-f, Differentiated CTLs,", + "footnote": [], + "bbox": [ + [ + 113, + 80, + 881, + 777 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_1.jpg", + "caption": "837 Extended Data Fig. 1: ST2 expression by CTLs and Th1 cells is regulated by T-", + "footnote": [], + "bbox": [ + [ + 115, + 113, + 880, + 335 + ] + ], + "page_idx": 35 + } +] \ No newline at end of file diff --git a/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12.mmd b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12.mmd new file mode 100644 index 0000000000000000000000000000000000000000..d0b384778ab37b36f14c3a0e9921b0ee4053df5d --- /dev/null +++ b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12.mmd @@ -0,0 +1,596 @@ + +# A type-1 immunity-restricted promoter of the IL-33 receptor gene directs antiviral T cell responses + +Max Löhning ( max.loehning@charite.de ) + +Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0001- 6382- 7281 + +Tobias Brunner German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 5279- 2391 + +Sebastian Serve Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 9839- 7119 + +Anna- Friederike Marx Division of Experimental Virology, Department of Biomedicine, University of Basel + +Maria Dzamukova German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 2389- 3222 + +Nayar Duran Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) + +Philippe Saikali Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 4916- 7376 + +Christoph Kommer Division of Theoretical Systems Biology, German Cancer Research Center (DKFZ) + +Frederik Heinrich German Rheumatism Research Center (DRFZ) Berlin https://orcid.org/0000- 0001- 6097- 5422 + +Pawel Durek German Rheumatism Research Centre + +Gitta Heinz Deutsches Rheuma- Forschungszentrum Berlin https://orcid.org/0000- 0002- 9330- 3689 + +Thomas Höfer German Cancer Research Center https://orcid.org/0000- 0003- 3560- 8780 + +Mir- Farzin Mashreghi Deutsches Rheuma- Forschungszentrum Berlin https://orcid.org/0000- 0002- 8015- 6907 + +Ralf Kuehn Max Delbrueck Center for Molecular Medic https://orcid.org/0000- 0003- 1694- 9803 + +Daniel Pinschewer University of Basel https://orcid.org/0000- 0002- 1594- 3189 + +<--- Page Split ---> + +## Article + +## Keywords: + +Posted Date: February 1st, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 1881814/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: Yes there is potential Competing Interest. D.D.P. is a founder, shareholder, and advisor to Hookiea Pharma Inc. commercializing arenavirus vectors. The remaining authors declare no competing interests. + +Version of Record: A version of this preprint was published at Nature Immunology on January 3rd, 2024. See the published version at https://doi.org/10.1038/s41590-023-01697-6. + +<--- Page Split ---> + +# A type-1 immunity-restricted promoter of the IL-33 receptor gene +directs antiviral T cell responses + +Tobias M. Brunner¹,²,⁹, Sebastian Serve¹,²,³,⁹, Anna-Friederike Marx⁴, Maria +Dzamukova¹,², Nayar Duran¹,², Philippe Saikali¹,², Christoph Kommer⁵,⁶, Frederik +Heinrich⁷, Pawel Durek⁷, Gitta A. Heinz⁷, Thomas Höfer⁵,⁶, Mir-Farzin Mashreghi⁷, +Ralf Kühn⁸, Daniel D. Pinschewer⁴, and Max Löhning¹,²,⁴ + +1 Experimental Immunology and Osteoarthritis Research, Department of +Rheumatology and Clinical Immunology, Charité – Universitätsmedizin Berlin, +corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin; +10117 Berlin, Germany. +2 Pitzer Laboratory of Osteoarthritis Research, German Rheumatism Research +Center (DRFZ), a Leibniz Institute; 10117 Berlin, Germany. +3 Berlin Institute of Health at Charité – Universitätsmedizin Berlin, BIH Biomedical +Innovation Academy; 10117 Berlin, Germany. +4 Division of Experimental Virology, Department of Biomedicine, University of Basel; +4009 Basel, Switzerland. +5 Division of Theoretical Systems Biology, German Cancer Research Center (DKFZ); +69120 Heidelberg, Germany. +6 BioQuant Center, University of Heidelberg; 69120 Heidelberg, Germany. +7 Therapeutic Gene Regulation, Regine von Ramin Lab Molecular Rheumatology, +German Rheumatism Research Center (DRFZ), a Leibniz Institute; 10117 Berlin, +Germany. +8 Max Delbrück Center for Molecular Medicine (MDC); 13125 Berlin, Germany. +9 These authors contributed equally to this work +*Corresponding authors. Email: tobias.brunner@drfz.de; max.loehning@charite.de; +loehning@drfz.de + +<--- Page Split ---> + +## 27 Abstract + +The pleiotropic alarmin interleukin- 33 (IL- 33) drives type- 1, type- 2, and regulatory T- cell responses via its receptor ST2. Subset- specific differences in ST2 expression intensity and dynamics suggest that transcriptional regulation is key in orchestrating the context- dependent activity of IL- 33- ST2 signaling in T- cell immunity. Here, we identify a previously unrecognized alternative promoter in mice and humans that is located far upstream of the curated ST2- coding gene and drives ST2 expression in type- 1 immunity. Mice lacking this promoter exhibit a selective loss of ST2 expression in type- 1- but not type- 2- biased immune cells, resulting in impaired expansion of cytotoxic T cells (CTLs) and T- helper- 1 cells upon viral infection. T- cell- intrinsic IL- 33 signaling via type- 1 promoter- driven ST2 is critical to generate a clonally diverse population of antiviral short- lived effector CTLs. Thus, lineage- specific alternative promoter usage directs alarmin responsiveness in T- cell subsets and offers opportunities for immune cell- specific targeting of the IL- 33- ST2 axis in infections and inflammatory diseases. + +<--- Page Split ---> + +Endogenous danger- associated molecules released upon cellular damage, so- called alarms, act as central orchestrators of inflammation'. Amongst alarms, the IL- 1 cytokine family member IL- 33 stands out as a highly potent cytokine which can trigger both pro- and anti- inflammatory responses by engaging its unique receptor ST2 on adaptive or innate immune cells2,3. ST2, also known as T14,5, was first detected on the surface of T- helper- 2 (Th2) cells and mast cells6- 8, and crosslinking of ST2 or stimulation with IL- 33 was shown to elicit production of type- 2 cytokines, which suggested an important role in type- 2 immune responses9- 11. In line with this, inhibition of IL- 33- ST2 signaling by antibody blockade or by disruption of either the /33 gene or the ST2- coding gene interleukin- 1 receptor- like 1 (Il1rl1) ameliorated type- 2 airway inflammation in mice and impaired immunity against parasitic nematode infections6,12- 14. Additionally, more recent studies established ST2 as a marker for type- 2 innate lymphoid cells (ILC2s) and demonstrated a critical role of IL- 33 for their development and function15,16. By activating ILC2s and a subset of highly suppressive, ST2- expressing regulatory T cells (Tregs), IL- 33 regulates tissue homeostasis, promotes wound healing, and mitigates pathology in acute or chronic inflammation3,17- 22. Beyond its role as a potentiator of Th2 and Treg activity, IL- 33 has also emerged as a key driver of type- 1 immune responses. It is released by fibroblastic reticular cells in lymphoid organs and promotes clonal expansion and activation of antiviral CTLs and T- helper- 1 (Th1) cells to confer protection against replicating viruses at the expense of potential tissue damage23- 27. Moreover, IL- 33- mediated amplification of type- 1 immune cells was shown to exacerbate tissue damage during Graft- versus- Host- Disease (GvHD)28,29 and to contribute to immune dysregulation in systemic inflammatory diseases30. + +Considering its multifaceted mode- of- action, IL- 33 is now recognized as an alarmin which amplifies the activity of distinct T cell subsets with either pro- or antiinflammatory properties in a context- specific manner31,32. To accomplish this versatility, it was suggested that ST2 expression is regulated in a cell- type specific manner, such that certain T cell subsets may become sensitive to IL- 33 signals dependent on the inflammatory environment and/or on costimulatory signals31,32. ST2 is absent from naive T cells but expressed at high levels in a constitutive manner by type- 2 biased immune cells, including memory Th2 cells and a Treg subset6,21. In + +<--- Page Split ---> + +75 these type- 2 biased cells, ST2 expression depends on the master- regulator 76 transcription factor of type- 2 immunity GATA- 321,33. In contrast, antiviral CTLs and Th1 77 cells express ST2 transiently upon infection, and this expression depends on STAT4 78 and T- bet, the key transcription factors of type- 1 immunity24,27. This highly dynamic 79 expression pattern not only contributed to the early assumption that ST2 is 80 preferentially expressed in type- 2 immunity, but it also renders it significantly more 81 difficult to study ST2 on type- 1 immune cells in vivo. Consequently, the molecular 82 mechanism allowing for T cell lineage- specific ST2 expression patterns has remained 83 enigmatic. 84 Here, we report on the identification of a type- 1 immunity- restricted promoter located 85 far upstream of the curated Il1rl1 gene. This previously unrecognized promoter is 86 bound by STAT4 and T- bet, is subject to epigenetic remodeling during type- 1 T cell 87 differentiation, and drives ST2 expression by CTLs and Th1 cells both in mice and 88 humans. Its deletion in mice resulted in a selective impairment of ST2 expression by 89 CTLs and Th1 cells, while ST2 expression on type- 2 immune cells and Tregs was 90 preserved. Furthermore, using lymphocytic choriomeningitis virus (LCMV) infection as 91 a model, we demonstrate that the type- 1 immunity- restricted promoter is essential for 92 the expansion of antiviral T cells and promotes the formation of clonally diverse 93 populations of short- lived effector CTLs. + +<--- Page Split ---> + +## Results + +## Identification of a type-1 immunity-restricted ll1r11 promoter + +Previous studies have demonstrated that the protein- coding exons of the ST2- encoding IL- 1 receptor- like 1 (ll1r11) gene are preceded by two non- coding exons (exon 1a and exon 1b) located in a distal and proximal promoter region, respectively4,34 (Fig. 1a). The proximal promoter drives ST2 expression in fibroblasts, whereas the distal promoter was shown to mediate ST2 expression in Th2 cells and mast cells33,35. To assess which transcriptional start site (TSS) is used by type- 1- polarized T cells, we differentiated naive T cells to CTLs, Th1, or Th2 cells in vitro, which triggers substantial ST2 expression in all subsets (Fig. 1b and Extended Data Fig. 1a- c). By analyzing the incorporation of leader exons into the 5' untranslated regions (UTR) of ST2- coding transcripts, we found that none of the described promoters could possibly account for the expression of ST2 by type- 1- polarized T cells (Fig. 1c). Thus, to map the origin of ll1r11 transcripts in these cells, we next subjected flow- cytometrically sorted ST2- expressing CTLs, Th1, and Th2 cells to RNA- sequencing (RNA- Seq) analysis with high read depth to achieve sufficient read coverage. By in silico alignment, we discovered a TSS located \(\sim 40\) kb upstream of the annotated ll1r11 gene, which was selectively transcribed in CTLs and Th1 cells (type- 1 promoter) but not detectable in Th2 cells (Fig. 1d). The identified TSS gave rise to two ll1r11 transcript isoforms with distinct leader sequences but unaltered protein- coding sequences. We here refer to the exons incorporated in these transcripts as exon A, B, and D. In addition, alternative splicing of exon B to exon C resulted in a transcript that did not contain ST2 protein- coding exons. As expected, the known distal promoter (exon 1a) presented the primary origin of ll1r11 transcripts in Th2 cells (type- 2 promoter). To further confirm that the identified type- 1 promoter is active in CTLs and Th1 cells in vivo, we next transferred naive LCMV- specific T cell receptor (TCR) transgenic \(\mathrm{CD4^{+}}\) (Smarta) or \(\mathrm{CD8^{+}}\) (P14) T cells into wildtype (WT) mice and infected the recipients with LCMV. At day 7 post infection (d7 p.i.), when T cells express high amounts of ST224,27, we reisolated the transferred T cells and quantified the ll1r11 promoter usage. In contrast to Th2 cells, CTLs and Th1 cells had largely incorporated exons A and B but not exon 1a into the 5'UTR of ST2- coding transcripts (Fig. 1e,f). Corroborating previous reports24,27, ST2 expression by CTLs and Th1 cells, but not by Th2 cells or regulatory T cells (Tregs), relied on IL- 12 signals and the transcription + +<--- Page Split ---> + +factors T- bet and STAT4 (Extended Data Fig. 1d- h). Hence, we analyzed T- bet- and STAT4- binding as well as activation- induced changes in chromatin accessibility at the llrl1 locus in type- 1- polarized T cells using publicly available chromatin immunoprecipitation sequencing (ChIP- Seq)36- 38 and assay for transposase- accessible chromatin sequencing (ATAC- Seq)39,40 data. As reported previously, GATA- 3, the key transcription factor of type- 2 immune cells, binds predominantly upstream of the distal promoter (Fig. 1g)33. In contrast, T- bet and STAT4 binding was detected in the vicinity of exons A and B, at sites that were inaccessible in naive \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells but accessible in LCMV- primed Th1 cells and CTLs (Fig. 1g). + +Alternative promoters are abundant41. We were unaware, however, of any other gene with a comparably selective alternative promoter usage in type- 1- and type- 2- polarized immune cells. We thus utilized RNA- Seq data and established computational methods42 to conduct a genome- wide search aimed at identifying additional genes with type- 1 or type- 2 immunity- specific promoters. llrl1 was reliably detected when comparing Th2 cells to CTLs and Th1 cells but, to our surprise, no other gene with lineage- specific TSS usage was found by this approach (Fig. 1h- j). These data suggested that type- 1 and type- 2 T cell lineage- specific utilization of alternative first exons is of rare occurrence, with the llrl1 gene potentially representing a unique case. + +## The type-1 llrl1 promoter is conserved between mice and humans + +To analyze the type- 1 promoter region in more detail we assessed DNA sequence conservation between vertebrate species using a precomputed conservation track 43 and compared it to STAT4- as well as to T- bet- ChIP- Seq and ATAC- Seq data (Fig. 2a). Additionally, we calculated conservation scores using the alignment tool VISTA44. Thereby, we identified a 275nt- spanning, well conserved sequence located \(\sim 5\) kb upstream of exon A (CNS- 5), which is bound by T- bet and STAT4 in Th1 cells and is marked by a sharp ATAC- Seq peak in CTLs (Fig. 2a). Mapping of T- bet and STAT4 binding motifs within CNS- 5 indicated that both transcription factors putatively bind closely to each other at sequences almost identical between mice and humans (Fig. 2b). + +To assess whether T cell lineage- specific promoter usage at the llrl1 locus is indeed conserved between mice and humans, we first examined cap analysis of gene expression (CAGE) and associated transcriptomic data from the FANTOM5 resource41,45. Data from a human NK leukemia- derived cell line (KHYG- 1) provided + +<--- Page Split ---> + +evidence for a putative TSS upstream of the annotated IL1RL1 gene. In human Th1 cells this site is preceded by T- bet binding sites (Fig. 2c)46. Of note, the exon structure closely resembles the exons A and B we identified in mice (cf. Fig. 1d). To determine whether this TSS is utilized in primary human CTLs and Th1 cells, we isolated in vivo- differentiated T cells from peripheral blood and reactivated them ex vivo to quantify IL1RL1 promoter usage (Extended Data Fig. 2a- f). Congruently with mouse T cells, ST2 transcripts of human CTLs and Th1 cells contained exons A and B within their 5'UTRs while Th2 cells had incorporated exon 1a (Fig. 2d). In summary, we have identified a previously unrecognized type- 1 immunity- restricted Il1rl1 promoter, which is instructed by the lineage- associated transcription factors T- bet and STAT4 and orchestrates ST2 expression in CTLs and Th1 cells of humans and mice. + +Alternative Il1rl1 promoters allow lineage- specific targeting of ST2 expression Modulation of the IL- 33- ST2 axis could represent a promising approach in treating inflammatory diseases. However, due to hard- to- predict effects on the balance between IL- 33- mediated inflammation and tissue repair, blockade of IL- 33 or ST2 using therapeutic antibodies has shown conflicting results in preclinical disease models47. Fine- tuned treatment approaches might therefore offer a critical advantage. We thus asked whether lineage- specific alternative promoters can be leveraged to target ST2 expression in a T cell subset- specific manner. To this end, primary T cells were activated and retrovirally transduced to express small- hairpin RNAs (shRNAs) targeting 5' UTRs of lineage- specific transcript isoforms (Fig. 3a). Selective downregulation of ST2 was achieved in CTLs and Th1 cells using shRNAs binding exons A and B, and in Th2 cells with shRNAs directed against exon 1a (Fig. 3b,c). Furthermore, to study the role of the type- 1 immunity- restricted promoter in vivo, we used CRISPR/Cas9 in mouse zygotes to generate a novel knockout mouse strain with a deletion comprising the noncoding exons A and B as well as the surrounding promoter region (Il1rl1- ExAB- ). A second mouse strain lacking exon C (Il1rl1- ExC- ) was generated to control for potential effects of exon C- containing transcripts (Fig. 3d and Extended Data Fig. 3a,b), which we expected to be affected in Il1rl1- ExAB- mice. In steady state both strains of knock- out mice harbored normal numbers of T cells at an activation state comparable to WT mice, both in spleen and lymph nodes (Extended Data Fig. 3c- h). Likewise, the introduced deletions did not affect the composition of splenic innate immune cells (Extended Data Fig. 3i- k). First, to assess ST2 expression + +<--- Page Split ---> + +by //1r/1- ExAB- and //1r/1- ExC- T cells, naive T cells were purified from the respective mice and activated in vitro. Deletion of the regulatory region surrounding exons A and B severely impaired STAT4- dependent transcription of the //1r/1 gene as evident from the absence of ST2 expression by //1r/1- ExAB- CTLs and Th1 cells (Fig. 3e,f). Importantly, //1r/1- ExAB- Th2 cells differentiated from the same pool of naive CD4+ T cells exhibited normal ST2 expression (Fig. 3e,f). As expected, ST2 expression by //1r/1- ExC- T cells was not reduced (Fig. 3e,f). Due to a lack of ST2 expression, CTLs as well as Th1- polarized CD4 T cells of //1r/1- ExAB- T cells were unresponsive to IL- 33 stimulation whereas Th2- polarized cells of the same mice expressed ST2 and responded normally (Fig. 3g, cf. Fig. 3e,f). To verify that ST2 expression is preserved on type- 2- biased immune cells of //1r/1- ExAB- mice in vivo, we analyzed ST2 expression on peritoneal mast cells and on ILC2s. Both cell types utilized the GATA- 3- regulated type- 2 //1r/1 promoter (Extended Data Fig. 4a). Like in Th2 cells, absence of the type- 1 promoter in //1r/1- ExAB- mice had no effect on ST2 expression by mast cells or ILC2s (Extended Data Fig. 4b- g). To investigate the requirement for type- 1 promoter- driven transcription for ST2 expression by CTLs and Th1 cells responding to viral challenge in vivo we infected WT, //1r/1- , //1r/1- ExAB- , and //1r/1- ExC- mice with LCMV (Fig. 3h). Consistent with the above in vitro results, ST2 surface expression was almost absent from CTLs in //1r/1- ExAB- mice and was significantly reduced in T- bet+ FoxP3- Th1 cells at day 7 p.i., while T- bet FoxP3+ Tregs displayed unaltered ST2 surface levels (Fig. 3l- n). Conversely, //1r/1- ExC- CTLs and Th1 cells exhibited slightly enhanced ST2 expression (Fig. 3l- n). This observation was in line with the RNA- Seq data showing that transcripts spliced from exon B to exon C are not further spliced to ST2- coding exons and exon C may thus act as a transcriptional decoy (cf. Fig. 1d). Altogether, we found that the targeting of individual //1r/1 promoters allowed for a selective T cell lineage- specific manipulation of ST2 expression. + +## The type-1 //1r/1 promoter drives antiviral T cell responses + +Next, we studied whether CD8+ T cell responses to LCMV infection required the type- 1 //1r/1 promoter. At the peak of the response (d7 p.i.), //1r/1- ExAB- promoter knock- out mice harbored substantially reduced numbers of CTLs in spleens and livers, resembling in its extent the impairment observed in //1r/1- gene knock- out mice (Fig. 4a- c and Extended Data Fig. 5a- c). Diminished CTL counts were largely accounted for by a reduction in activated CD44+CD62L- effector T cells expressing the + +<--- Page Split ---> + +proliferation marker Ki67 (Fig. 4d- f). Accordingly, CTLs specific for the immunodominant \(\mathsf{GP}_{33 - 41}\) and \(\mathsf{NP}_{396 - 404}\) epitopes of LCMV, respectively, were substantially reduced (Fig. 4g and Extended Data Fig. 5d- f). Ultimately, ExonAB- deficient mice displayed significantly lower numbers of CTLs expressing effector cytokines or cytolytic molecules, and systemic IFN- \(\gamma\) levels were reduced by \(>70\%\) (Fig. 4h,i and Extended Data Fig. 5g- i). In contrast and as expected, CTL responses of \(I11rl1 - ExC^{- / - }\) mice were comparable to WT mice in all of the aforementioned aspects (Fig. 4b- i). + +To determine whether the observed effects were due to a T cell- intrinsic impairment in ST2 expression, mixed bone marrow chimeras were generated by reconstituting irradiated WT mice with bone marrow from \(I11rl1^{- / - }\) , \(I11rl1 - ExAB^{- / - }\) , or \(I11rl1 - ExC^{- / - }\) mice, each of them mixed 1:1 with WT bone marrow. Following LCMV infection, WT: \(I11rl1 - ExC^{- / - }\) chimeras mounted CTL responses that derived at approximately equal parts from both bone marrow compartments. In contrast, WT bone marrow- derived CTLs significantly outnumbered the CTLs derived from \(I11rl1^{- / - }\) or \(I11rl1 - ExAB^{- / - }\) bone marrow in the respective chimeras (Fig. 4j- m and Extended Data Fig. 5j- k). Lastly, to study the impact of \(I11rl1 - ExAB\) - driven ST2 expression on T cell responses in the absence of any potentially confounding irradiation effects on recipient mice, \(I11rl1 - ExAB^{- / - }\) mice were crossed to LCMV TCR- transgenic P14 mice. Congenically marked naive \(I11rl1 - ExAB^{- / - }\) P14 CTLs and WT P14 cells were cotransferred into recipients, which were subsequently infected with LCMV and analyzed at d10 p.i.. Analogously to the data obtained in mixed bone marrow- chimeric mice, \(I11rl1 - ExAB^{- / - }\) and \(I11rl1^{- / - }\) P14 T cells expanded significantly less than their respective cotransferred WT P14 T cell populations (Fig. 4n- r). Similarly, albeit less pronounced, \(I11rl1 - ExAB^{- / - }\) LCMV- TCR transgenic \(\mathrm{CD4^{+}}\) Smarta T cells expanded less than cotransferred WT Smarta cells (Fig. 4s- w and Extended Data Fig. 5l- p). Thus, optimal expansion of antiviral T cells critically depends on cell- intrinsic activity of the type- 1 \(I11rl1\) promoter. + +## The type-1 \(I11rl1\) promoter is critical to establish a clonally diverse population of short-lived effector CTLs + +At the peak of the acute response, the population of antiviral CTLs is heterogeneous and comprises many functionally distinct subsets48. To delineate the impact of type- 1 promoter- driven ST2 expression on CTL differentiation we sorted activated \(\mathrm{CD44^{+}}\) \(\mathrm{CD8^{+}}\) effector T cells from LCMV- infected WT or \(I11rl1 - ExAB^{- / - }\) mice and subjected + +<--- Page Split ---> + +them to combined single- cell gene expression and TCR repertoire analysis. T cells were clustered into separate populations using nearest neighbor modularity optimization, displayed in uniform manifold approximation projections (UMAP), and annotated based on signature gene expression (Fig. 5a,b, Extended Data Fig. 6a,b, and Supplementary Table 1). Analyzing CTLs from both genotypes, six clusters were identified. As expected, the two dominant ones showed expression of genes associated with a short- lived effector cell (SLEC; Klrg1, Gzma, Id2) or memory precursor effector cell phenotype (MPEC; Il7r, Sell, Ccr7) (Fig. 5c). The third cluster was enriched in CTLs expressing Pdcd1 (encoding PD- 1) and Lag3, two markers of exhausted CTLs49, while cells of the fourth cluster exhibited higher expression of Tcf7 (encoding TCF- 1) and Id3, thus likely presenting stem- like precursors of effector CTLs50,51. Lastly, the two remaining clusters were enriched in CTLs expressing higher levels of mitochondrial genes or cell cycle- related markers (Mki67, Top2a). A comparison of CTL cluster composition in the two genotypes revealed that type- 1 Il1rl1 promoter disruption resulted in a particularly pronounced curtailment of SLEC- differentiated cells (Fig. 5d). This translated into a 90- 95% reduction of KLRG1+CD127- CTL numbers in Il1rl1- ExAB-/- mice and mirrored the phenotype of Il1rl1-/- mice (Fig. 5e- g). Further, a lower proportion of Il1rl1- ExAB-/- CTLs exhibited surface expression of the SLEC- associated molecule CXCR3 (Fig. 5h)52. Moreover, Il1rl1- ExAB-/- CD8+ T cells in mixed bone marrow chimeras evidenced a pronounced defect in SLEC generation, similar to the behavior of Il1rl1-/- CD8+ T cells (Fig. 5i). In contrast, Il1rl1- ExC-/- bone marrow- derived CD8+ T cells were as proficient as WT cells in populating the SLEC compartment (Fig. 5i). Lastly, the proportion of ExonAB- deficient P14 T cells differentiating into SLECs and/or CXCR3+ cells was reduced as compared to adaptively cotransferred WT P14 cells (Fig. 5j- l). Altogether these results demonstrated that impaired SLEC formation by ExonAB- deficient CD8+ T cells was due to a cell- intrinsic defect in ST2 signaling. Next, we asked whether type- 1 promoter- driven expression of ST2 drives the selective proliferation of SLEC- differentiated T cell clones or whether it enforces the differentiation of precursors into SLECs. Thus, we first assessed ST2 expression on LCMV- specific CD8+ T cells of WT mice where we found it on both KLRG1+ and KLRG1- cells (Fig. 5m), suggesting IL- 33 signaling can occur prior to SLEC differentiation. Further, we integrated single- cell gene expression data with a TCR repertoire analysis. Interestingly, despite an eight- fold difference in CD44+ CTL cell + +<--- Page Split ---> + +counts per spleen (Fig. 5n), equivalent numbers of clonotypes were identified in Il1rl1- ExAB- and WT mice when equal numbers of CD44+ CTLs were compared (Fig. 5o). This suggested that during the acute phase of infection, IL- 33 expands the pool of activated T cells in a clonotype- unselective manner, which does not significantly alter the TCR diversity, at least not amongst the most abundant clonotypes. Of note, the majority of clonotypes identified were represented less than three times per mouse and no clonotype was found more often than seven times (Fig. 5p). In line with previous reports, this indicated that TCR diversity within the CTL population was high during the acute phase of infection, without any evidence for clearly dominant clones53. By consequence, the severe reduction in SLEC numbers in Il1rl1- ExAB- mice resulted in substantially reduced SLEC clonotype numbers (Fig. 5q,r). Together, these findings demonstrate that in the absence of type- 1 promoter- driven ST2 expression, most CTL clones achieve basal activation, but fail to develop into fully differentiated effector cells. Thus, type- 1 promoter- driven ST2 expression is vital to establish a numerically significant and clonally diverse population of short- lived antiviral effector CTLs. + +## Transcriptional profiling indicates broad costimulatory and TCR-cooperative functions of IL-33- ST2 signaling + +To gain further mechanistic insight on how IL- 33 signaling modulates T cell activation and differentiation, we additionally performed a comprehensive analysis of early IL- 33 target genes in CTLs as well as Th1 and Th2 cells by RNA- Seq profiling. To this end, naive T cells were differentiated, followed by a resting period without antigenic stimulation. Since ST2 signaling is subject to negative feedback mechanisms and also because oxidation of IL- 33 rapidly reduces its activity54,55, gene expression was analyzed before stimulation (0 hours) and 2 hours after treatment with or without IL- 33 (Fig. 6a). In the absence of IL- 33 stimulation the gene expression profiles at 0 hours and at 2 hours were similar. In contrast, short- term stimulation with IL- 33 had a profound effect on the transcriptome of CTLs, Th1, and Th2 cells, and it strongly induced or, in fewer cases, reduced expression of target genes rather than preventing a loss or gain of transcription (Fig. 6b). Whereas a substantial portion of differentially regulated genes was shared between the T cell subsets analyzed, others were regulated in a lineage- specific manner (Fig. 6c). GO- enrichment analysis revealed a broad role of IL- 33- responsive genes in T cell activation, proliferation, and differentiation (Fig. 6d). Importantly, IL- 33 stimulation of CTLs amplified expression of + +<--- Page Split ---> + +Tbx21, Zeb2, and Prdm1, genes of transcription factors known as key drivers of SLEC differentiation56- 58 (Fig. 6b,e). Across the three T cell subsets we observed a prominent upregulation of genes frequently used as indicators of recent TCR activation (Nr4a1, Cd69, Batf) (Fig. 6b,e)59- 61. Coherently, gene set enrichment analysis showed a significant overlap between IL- 33- and TCR- downstream signaling in CTLs (Fig. 6f and Supplementary Table 2). This suggested that IL- 33, which is released by fibroblastic reticular cells early during infection26, may support TCR stimulation to promote potent antiviral CTL responses. + +Thus, to further investigate the interplay between ST2 and TCR signaling strength, we made use of a genetically engineered LCM virus that differs from the wildtype counterpart only by a GP- A39C mutation, rendering its GP33 epitope (GP33- C6: KAVYNFCTC instead of KAVYNFATC) a weak P14 TCR agonist62. P14 and Il1rl1- /- P14 cells were cotransferred into wildtype recipients, which were subsequently infected with LCMV expressing either the high- or the low- affinity GP33 variant (Fig. 6g). We found that ST2- sufficient as well as ST2- deficient P14 cells expanded less when primed with the low- affinity ligand (Fig. 6h). Further, P14 cells depended on ST2 for optimal expansion and effector differentiation, irrespective of the TCR stimulation strength (Fig. 6i,j). Interestingly, the response of WT P14 cells responding to low- affinity virus was comparable to or exceeded the one of high- affinity ligand- primed Il1rl1- /- P14 cells in terms of total and effector cell progeny, respectively (Fig. 6h,j). This suggested that ST2 signals can help reaching effector T cell responses of critical size even when confronted with low- affinity ligands. Of note, the impairment in expansion and effector differentiation between mice infected with high- or low- affinity GP33- expressing LCMV were unlikely due to any potential differences in inflammation or IL- 33 release, as the endogenous NP396- specific T cell responses to the two LCMV variants were indistinguishable (Fig. 6k,l). + +In summary, these data demonstrate that IL- 33- ST2 signaling provides a strong costimulatory signal that can act cooperatively with TCR signaling to promote the expansion and effector cell differentiation of antiviral CTLs. + +<--- Page Split ---> + +## Discussion + +IL- 33 alarm signaling has long been recognized as an important driver of type- 2 immune responses2,3,6,7,12,31,32. Over the past decade its important role in promoting type- 1 immunity has become widely accepted yet remains mechanistically less well understood, particularly due to a lack of understanding how IL- 33 receptor expression is regulated in type- 1 immune cells31,32. Here, we studied the transcriptional regulation of ST2 expression in antiviral T cells and thereby discovered a dedicated type- 1 immunity- restricted promoter located \(\sim 40\) kb upstream of the curated I/1r/1 gene in mice and humans. This type- 1 promoter drives IL- 33 receptor expression by CTLs and Th1 cells in vitro and in viral infections in vivo. As opposed to the previously described "distal" type- 2 promoter, which is regulated by GATA- 333 and is utilized by type- 2 immune cells as well as Tregs21, the identified promoter is controlled by the type- 1 immunity- associated transcription factors T- bet and STAT4 and is subject to epigenetic remodeling during type- 1 T cell differentiation. Thus, we provide evidence for a dedicated regulatory genetic element to control ST2 expression selectively in type- 1 immune cells. Interestingly, this finding explains why transcription of I/1r/1 seems to be coregulated with its neighboring type- 1 immunity- associated gene I/18r63. Although the I/1r/1 gene has been studied intensively8,35,64, the type- 1 promoter has remained unrecognized, likely because it is only transiently active, often resulting in a low abundance of ST2- coding transcripts, depending on the time- point of analysis24,27. The latter renders it difficult to obtain adequate read coverage across splice junctions for a clear definition of exon structures by commonly used RNA- Seq techniques65 – a challenge we approached initially by analyzing T cells that have been differentiated in vitro to express a high amount of I/1r/1 transcripts. In knock- down and knock- out experiments we have subsequently validated the crucial role of this regulatory element in vivo, and by analyzing human T cells have extended the concept to our species. Above all, our finding was surprising, as to the best of our knowledge no other gene has been identified to date, for which type- 1 and type- 2 immune cells exhibit a similarly distinct lineage- specific promoter usage. Consistently, our own attempts at identifying further genes with an analogous dual promoter usage were unsuccessful. We acknowledge that technical limitations might have prevented us from identifying such genes, thus rendering it impossible to firmly conclude that the I/1r/1 gene is the only gene regulated by such highly selective, type- 1 and type- 2 restricted promoters. Still, + +<--- Page Split ---> + +our results suggest that this is not a common feature but represents a fairly unique mechanism to spatiotemporally orchestrate ST2 expression. + +IL- 33 is an exceptionally potent alarmin, which can act as a pro- or anti- inflammatory cytokine, depending on the local composition of immune cells and their responsiveness to IL- 33 at the timing of release32. Likely due to its potential to cause severe inflammation, IL- 33 responsiveness requires stringent regulation. Transcription from the type- 1 immunity- restricted promoter enables transient ST2 expression by CTLs and Th1 cells in response to inflammatory stimuli23,24,27, which may serve to prevent continuous activation of cells with a high tissue- destructive potential. In contrast, constitutive type- 2 promoter- driven ST2 expression on Tregs and ILC2s allows for rapid anti- inflammatory responses to tissue damage19- 21,32. + +Importantly, this dual mode- of- action constitutes a major hurdle for the therapeutic modulation of IL- 33–ST2 signaling32,66. We here demonstrate that the usage of distinct promoters offers opportunities for a T cell subset- specific targeting of ST2 expression at the transcriptional level by shRNA knock- down and at the genomic level by CRISPR/Cas9- mediated deletion of the type- 1 Il1rl1 promoter. ExonAB- deficient mice exhibit a selective, type- 1 immunity- restricted impairment of ST2 expression and display curtailed CD8+ and CD4+ T cell responses against LCMV, while ST2 expression by Tregs and type- 2 immune cells was fully preserved. Of note, the effects of IL- 33 on CTL expansion and differentiation were almost exclusively dependent on the type- 1 promoter, while its deletion did not fully abrogate ST2 expression by Th1 cells. Nevertheless, the type- 1 promoter was critical for optimal expansion of antiviral Th1 cells. T cell subset- specific targeting approaches could be of interest to modulate IL- 33 responses in inflammatory diseases. For instance, IL- 33 administration was shown to drive Treg expansion in the context of GvHD, promoting tolerance induction and disease amelioration67- 69. However, IL- 33 also augments type- 1 alloimmunity by acting as a costimulatory molecule for donor CTLs and Th1 cells28,29. A targeted disruption of ST2 selectively on type- 1 immune cells might minimize the pathological response during GvHD while the protective effects of IL- 33 should remain preserved. Our study provides mechanistic insight into the role of type- 1 promoter- driven ST2 expression and IL- 33 signaling in CD8+ T cell differentiation. Single- cell gene expression analysis of antiviral CTLs revealed that in the absence of the type- 1 ST2 promoter, mice display a particularly pronounced reduction in antiviral SLECs. So far, it remained unaddressed whether IL- 33 merely enforces proliferation of effector CTLs, + +<--- Page Split ---> + +or if IL- 33 sensing actively promotes the differentiation of SLECs. Clonotype diversity in the SLEC population was high in wildtype mice and diminished proportionally to the cell counts in ExonAB- deficient mice. This suggests that IL- 33 can indeed foster the transition of an activated CTL into a cell with potent effector functions rather than expanding a pool of pre- differentiated SLECs. Terminal differentiation has been associated with STAT4 signaling and with high levels of T- bet, Blimp- 1, and Zeb256- 58. Likewise, the activity of the type- 1 ST2 promoter is positively regulated by STAT4 and T- bet. Interestingly, in- depth RNA- Seq profiling of IL- 33 target genes in CTLs demonstrated an induction of Tbx21 (T- bet), Prdm1 (Blimp- 1), and Zeb2 transcript expression shortly upon stimulation, thus inferring a positive feedback loop that further reinforces both type- 1 promoter- driven ST2 expression and effector differentiation via T- bet. + +Besides these cell- intrinsic factors, extrinsic cues via the TCR determine the fate of CTLs. Above- threshold TCR signaling strength is directly linked to asymmetric cell division and acquisition of effector properties70. Our data show that IL- 33 stimulation of CTLs strongly induces the transcription of several TCR- dependent genes. Moreover, IL- 33 signals can restore the otherwise suboptimal expansion and effector differentiation of CTLs in response to a low- affinity antigenic peptide. In addition, recent studies demonstrated a loss of IL- 33 in lymphoid organs early after LCMV infection, suggesting substantial alarmin release during T cell priming26. Together, this implies that ST2- signaling might act in conjunction with TCR- signaling to achieve above- threshold activation required for fully functional effector differentiation. + +In summary, we here uncover lineage- specific promoter usage as the molecular mechanism governing disparate expression patterns of the IL- 33 receptor ST2 in distinctly polarized T cell subsets. By generating a type- 1 promoter- deficient mouse strain, in which ST2 expression by Tregs and type- 2 immune cells is preserved, we offer a tool to better assess the role of IL- 33- ST2 signaling in type- 1 immunity. Utilizing these mice, we demonstrate that the type- 1 immunity- restricted Il1rl1 promoter is essential for fully functional antiviral CD8+ and CD4+ T cell responses and critical for the formation of a clonally diverse population of effector CTLs. These findings open new avenues for the modulation and exploitation of IL- 33 signaling in type- 1 immunity- mediated inflammatory diseases and T cell- based cancer immunotherapy, respectively. + +<--- Page Split ---> + +455 References456 1 Yang, D., Han, Z. & Oppenheim, J. J. Alarms and immunity. Immunol Rev457 280, 41- 56, doi:10.1111/imr.12577 (2017).458 2 Dinarello, C. A. Overview of the IL- 1 family in innate inflammation and459 acquired immunity. Immunol Rev 281, 8- 27, doi:10.1111/imr.12621 (2018).460 3 Liew, F. Y., Girard, J. P. & Turnquist, H. R. Interleukin- 33 in health and461 disease. Nat Rev Immunol 16, 676- 689, doi:10.1038/nri.2016.95 (2016).462 4 Klemenz, R., Hoffmann, S. & Werenskiold, A. K. Serum- and oncoprotein-463 mediated induction of a gene with sequence similarity to the gene encoding464 carcinoembryonic antigen. Proc Natl Acad Sci U S A 86, 5708- 5712,465 doi:10.1073/pnas.86.15.5708 (1989).466 5 Tominaga, S. A putative protein of a growth specific cDNA from BALB/c- 3T3467 cells is highly similar to the extracellular portion of mouse interleukin 1468 receptor. FEBS Lett 258, 301- 304, doi:10.1016/0014- 5793(89)81679- 5(1989).469 (470 6 Löhning, M. et al. T1/ST2 is preferentially expressed on murine Th2 cells,471 independent of interleukin 4, interleukin 5, and interleukin 10, and important472 for Th2 effector function. Proc Natl Acad Sci U S A 95, 6930- 6935,473 doi:10.1073/pnas.95.12.6930 (1998).474 7 Xu, D. et al. Selective expression of a stable cell surface molecule on type 2475 but not type 1 helper T cells. J Exp Med 187, 787- 794,476 doi:10.1084/jem.187.5.787 (1998).477 8 Gachter, T., Werenskiold, A. K. & Klemenz, R. Transcription of the interleukin-478 1 receptor- related T1 gene is initiated at different promoters in mast cells and479 fibroblasts. J Biol Chem 271, 124- 129, doi:10.1074/jbc.271.1.124 (1996).480 9 Schmitz, J. et al. IL- 33, an interleukin- 1- like cytokine that signals via the IL- 1481 receptor- related protein ST2 and induces T helper type 2- associated482 cytokines. Immunity 23, 479- 490, doi:10.1016/j.immuni.2005.09.015 (2005).483 10 Meisel, C. et al. Regulation and function of T1/ST2 expression on CD4+ T484 cells: induction of type 2 cytokine production by T1/ST2 cross- linking. J485 Immunol 166, 3143- 3150, doi:10.4049/jimmunol.166.5.3143 (2001).486 11 Ho, L. H. et al. IL- 33 induces IL- 13 production by mouse mast cells487 independently of IgE- FcepsilonRI signals. J Leukoc Biol 82, 1481- 1490,488 doi:10.1189/jlb.0407200 (2007).489 12 Townsend, M. J., Fallon, P. G., Matthews, D. J., Jolin, H. E. & McKenzie, A.490 N. T1/ST2- deficient mice demonstrate the importance of T1/ST2 in developing491 primary T helper cell type 2 responses. J Exp Med 191, 1069- 1076,492 doi:10.1084/jem.191.6.1069 (2000). + +<--- Page Split ---> + +493 13 Oboki, K. et al. IL- 33 is a crucial amplifier of innate rather than acquired 494 immunity. Proc Natl Acad Sci U S A 107, 18581- 18586, 495 doi:10.1073/pnas.1003059107 (2010). 496 14 Coyle, A. J. et al. Crucial role of the interleukin 1 receptor family member 497 T1/ST2 in T helper cell type 2- mediated lung mucosal immune responses. J 498 Exp Med 190, 895- 902, doi:10.1084/jem.190.7.895 (1999). 499 15 Hoyler, T. et al. The transcription factor GATA- 3 controls cell fate and 500 maintenance of type 2 innate lymphoid cells. Immunity 37, 634- 648, 501 doi:10.1016/j.immuni.2012.06.020 (2012). 502 16 Moro, K. et al. Innate production of T(H)2 cytokines by adipose tissue- 503 associated c- Kit(+)Sca- 1(+) lymphoid cells. Nature 463, 540- 544, 504 doi:10.1038/nature08636 (2010). 505 17 Mahlakoir, T. et al. Stromal cells maintain immune cell homeostasis in 506 adipose tissue via production of interleukin- 33. Sci Immunol 4, 507 doi:10.1126/sciimmunol.aax0416 (2019). 508 18 Siede, J. et al. IL- 33 Receptor- Expressing Regulatory T Cells Are Highly 509 Activated, Th2 Biased and Suppress CD4 T Cell Proliferation through IL- 10 510 and TGFbeta Release. PLoS One 11, e0161507, 511 doi:10.1371/journal.pone.0161507 (2016). 512 19 Kuswanto, W. et al. Poor Repair of Skeletal Muscle in Aging Mice Reflects a 513 Defect in Local, Interleukin- 33- Dependent Accumulation of Regulatory T 514 Cells. Immunity 44, 355- 367, doi:10.1016/j.immuni.2016.01.009 (2016). 515 20 Monticelli, L. A. et al. IL- 33 promotes an innate immune pathway of intestinal 516 tissue protection dependent on amphi-regulin- EGFR interactions. Proc Natl 517 Acad Sci U S A 112, 10762- 10767, doi:10.1073/pnas.1509070112 (2015). 518 21 Schiering, C. et al. The alarmin IL- 33 promotes regulatory T- cell function in 519 the intestine. Nature 513, 564- 568, doi:10.1038/nature13577 (2014). 520 22 Hemmers, S., Schizas, M. & Rudensky, A. Y. T reg cell- intrinsic requirements 521 for ST2 signaling in health and neuroinflammation. J Exp Med 218, 522 doi:10.1084/jem.20201234 (2021). 523 23 Bonilla, W. V. et al. The alarmin interleukin- 33 drives protective antiviral 524 CD8(+) T cell responses. Science 335, 984- 989, 525 doi:10.1126/science.1215418 (2012). 526 24 Baumann, C. et al. T- bet- and STAT4- dependent IL- 33 receptor expression 527 directly promotes antiviral Th1 cell responses. Proc Natl Acad Sci U S A 112, 528 4056- 4061, doi:10.1073/pnas.1418549112 (2015). 529 25 McLaren, J. E. et al. IL- 33 Augments Virus- Specific Memory T Cell Inflation 530 and Potentiates the Efficacy of an Attenuated Cytomegalovirus- Based 531 Vaccine. J Immunol 202, 943- 955, doi:10.4049/jimmunol.1701757 (2019). + +<--- Page Split ---> + +532 26 Aparicio-Domingo, P. et al. Fibroblast-derived IL-33 is dispensable for lymph node homeostasis but critical for CD8 T-cell responses to acute and chronic viral infection. Eur J Immunol 51, 76-90, doi:10.1002/ejj.201948413 (2021).535 27 Baumann, C. et al. Memory CD8(+) T Cell Protection From Viral Reinfection Depends on Interleukin-33 Alarmin Signals. Front Immunol 10, 1833, doi:10.3389/fimmu.2019.01833 (2019).538 28 Reichenbach, D. K. et al. The IL-33/ST2 axis augments effector T-cell responses during acute GVHD. Blood 125, 3183-3192, doi:10.1182/blood-2014-10-606830 (2015).541 29 Dwyer, G. K. et al. IL-33 acts as a costimulatory signal to generate alloreactive Th1 cells in graft-versus-host disease. J Clin Invest 132, doi:10.1172/JCI150927 (2022).544 30 Rood, J. E. et al. ST2 contributes to T-cell hyperactivation and fatal hemophagocytic lymphohistiocytosis in mice. Blood 127, 426-435, doi:10.1182/blood-2015-07-659813 (2016).547 31 Peine, M., Marek, R. M. & Löhning, M. IL-33 in T Cell Differentiation, Function, and Immune Homeostasis. Trends Immunol 37, 321-333, doi:10.1016/j.it.2016.03.007 (2016).550 32 Molofsky, A. B., Savage, A. K. & Locksley, R. M. Interleukin-33 in Tissue Homeostasis, Injury, and Inflammation. Immunity 42, 1005-1019, doi:10.1016/j.immuni.2015.06.006 (2015).553 33 Hayakawa, M. et al. T-helper type 2 cell-specific expression of the ST2 gene is regulated by transcription factor GATA-3. Biochim Biophys Acta 1728, 53-64, doi:10.1016/j.bbaexp.2005.01.012 (2005).556 34 Iwahana, H. et al. Different promoter usage and multiple transcription initiation sites of the interleukin-1 receptor-related human ST2 gene in UT-7 and TM12 cells. Eur J Biochem 264, 397-406, doi:10.1046/j.1432-1327.1999.00615.x (1999).557 35 Lipsky, B. P., Toy, D. Y., Swart, D. A., Smithgall, M. D. & Smith, D. Deletion of the ST2 proximal promoter disrupts fibroblast-specific expression but does not reduce the amount of soluble ST2 in circulation. Eur J Immunol 42, 1863-1869, doi:10.1002/ejj.201142274 (2012).556 36 Gökmen, M. R. et al. Genome-wide regulatory analysis reveals that T-bet controls Th17 lineage differentiation through direct suppression of IRF4. J Immunol 191, 5925-5932, doi:10.4049/jimmunol.1202254 (2013).557 37 Wei, G. et al. Genome-wide analyses of transcription factor GATA3-mediated gene regulation in distinct T cell types. Immunity 35, 299-311, doi:10.1016/j.immuni.2011.08.007 (2011).570 38 Wei, L. et al. Discrete roles of STAT4 and STAT6 transcription factors in tuning epigenetic modifications and transcription during T helper cell + +<--- Page Split ---> + +572 differentiation. Immunity 32, 840- 851, doi:10.1016/j.immuni.2010.06.003 (2010). + +574 39 Chen, X. et al. The histone methyltransferase EZH2 primes the early differentiation of follicular helper T cells during acute viral infection. Cell Mol Immunol 17, 247-260, doi:10.1038/s41423-019-0219-z (2020). + +577 40 Yu, B. et al. Epigenetic landscapes reveal transcription factors that regulate CD8(+) T cell differentiation. Nat Immunol 18, 573-582, doi:10.1038/ni.3706 (2017). + +580 41 Fantom-Consortium et al. A promoter-level mammalian expression atlas. Nature 507, 462-470, doi:10.1038/nature13182 (2014). + +582 42 Demircioğlu, D. et al. A Pan-cancer Transcriptome Analysis Reveals Pervasive Regulation through Alternative Promoters. Cell 178, 1465-1477 e1417, doi:10.1016/j.cell.2019.08.018 (2019). + +585 43 Siepel, A. et al. Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res 15, 1034-1050, doi:10.1101/gr.3715005 (2005). + +588 44 Frazer, K. A., Pachter, L., Poliakov, A., Rubin, E. M. & Dubchak, I. VISTA: computational tools for comparative genomics. Nucleic Acids Res 32, W273-279, doi:10.1093/nar/gkh458 (2004). + +591 45 Hon, C. C. et al. An atlas of human long non-coding RNAs with accurate 5' ends. Nature 543, 199-204, doi:10.1038/nature21374 (2017). + +593 46 Kanhere, A. et al. T-bet and GATA3 orchestrate Th1 and Th2 differentiation through lineage-specific targeting of distal regulatory elements. Nat Commun 3, 1268, doi:10.1038/ncomms2260 (2012). + +596 47 Griesenauer, B. & Paczesny, S. The ST2/IL-33 Axis in Immune Cells during Inflammatory Diseases. Front Immunol 8, 475, doi:10.3389/fimmu.2017.00475 (2017). + +599 48 McLane, L. M., Abdel-Hakeem, M. S. & Wherry, E. J. CD8 T Cell Exhaustion During Chronic Viral Infection and Cancer. Annu Rev Immunol 37, 457-495, doi:10.1146/annurev-immunol-041015-055318 (2019). + +602 49 Yao, C. et al. Single-cell RNA-seq reveals TOX as a key regulator of CD8(+) T cell persistence in chronic infection. Nat Immunol 20, 890-901, doi:10.1038/s41590-019-0403-4 (2019). + +605 50 Pais Ferreira, D. et al. Central memory CD8(+) T cells derive from stem-like Tcf7(hi) effector cells in the absence of cytotoxic differentiation. Immunity 53, 985-1000 e1011, doi:10.1016/j.immuni.2020.09.005 (2020). + +608 51 Zehn, D., Thimme, R., Lugli, E., de Almeida, G. P. & Oxenius, A. 'Stem-like' precursors are the fount to sustain persistent CD8(+) T cell responses. Nat Immunol 23, 836-847, doi:10.1038/s41590-022-01219-w (2022). + +<--- Page Split ---> + +611 52 Kurachi, M. et al. Chemokine receptor CXCR3 facilitates CD8(+) T cell differentiation into short-lived effector cells leading to memory degeneration. J Exp Med 208, 1605- 1620, doi:10.1084/jem.20102101 (2011). + +614 53 Chang, Y. M. et al. T Cell Receptor Diversity and Lineage Relationship between Virus-Specific CD8 T Cell Subsets during Chronic Lymphocytic Choriomeningitis Virus Infection. J Virol 94, doi:10.1128/JVI.00935-20 (2020). + +617 54 Yamamoto, T. et al. DUSP10 constrains innate IL-33-mediated cytokine production in ST2(hi) memory-type pathogenic Th2 cells. Nat Commun 9, 4231, doi:10.1038/s41467-018-06468-8 (2018). + +620 55 Cohen, E. S. et al. Oxidation of the alarmin IL-33 regulates ST2-dependent inflammation. Nat Commun 6, 8327, doi:10.1038/ncomms9327 (2015). + +622 56 Joshi, N. S. et al. Inflammation directs memory precursor and short-lived effector CD8(+) T cell fates via the graded expression of T-bet transcription factor. Immunity 27, 281-295, doi:10.1016/j.immuni.2007.07.010 (2007). + +625 57 Dominguez, C. X. et al. The transcription factors ZEB2 and T-bet cooperate to program cytotoxic T cell terminal differentiation in response to LCMV viral infection. J Exp Med 212, 2041-2056, doi:10.1084/jem.20150186 (2015). + +628 58 Rutishauser, R. L. et al. Transcriptional repressor Blimp-1 promotes CD8(+) T cell terminal differentiation and represses the acquisition of central memory T cell properties. Immunity 31, 296-308, doi:10.1016/j.immuni.2009.05.014 (2009). + +632 59 Ashouri, J. F. et al. Reporters of TCR signaling identify arthritogenic T cells in murine and human autoimmune arthritis. Proc Natl Acad Sci U S A 116, 18517-18527, doi:10.1073/pnas.1904271116 (2019). + +635 60 Cibrian, D. & Sanchez-Madrid, F. CD69: from activation marker to metabolic gatekeeper. Eur J Immunol 47, 946-953, doi:10.1002/eji.201646837 (2017). + +637 61 Iwata, A. et al. Quality of TCR signaling determined by differential affinities of enhancers for the composite BATT-IRF4 transcription factor complex. Nat Immunol 18, 563-572, doi:10.1038/ni.3714 (2017). + +640 62 Utzschneider, D. T. et al. High antigen levels induce an exhausted phenotype in a chronic infection without impairing T cell expansion and survival. J Exp Med 213, 1819-1834, doi:10.1084/jem.20150598 (2016). + +643 63 Xu, D. et al. Selective expression and functions of interleukin 18 receptor on T helper (Th) type 1 but not Th2 cells. J Exp Med 188, 1485-1492, doi:10.1084/jem.188.8.1485 (1998). + +646 64 Bergers, G., Reikerstorfer, A., Braselmann, S., Graninger, P. & Busslinger, M. Alternative promoter usage of the Fos-responsive gene Fit-1 generates mRNA isoforms coding for either secreted or membrane-bound proteins related to the IL-1 receptor. EMBO J 13, 1176-1188 (1994). + +<--- Page Split ---> + +650 65 SEQC-MAQC-III-Consortium. A comprehensive assessment of RNA-seq accuracy, reproducibility and information content by the Sequencing Quality Control Consortium. Nat Biotechnol 32, 903- 914, doi:10.1038/nbt.2957 (2014).652 653 (2014).654 66 Braun, H., Afonina, I. S., Mueller, C. & Beyaert, R. Dichotomous function of IL- 655 33 in health and disease: From biology to clinical implications. Biochem656 Pharmacol 148, 238- 252, doi:10.1016/j.bcp.2018.01.010 (2018).657 67 Matta, B. M. et al. Peri- alloHCT IL- 33 administration expands recipient T- 658 regulatory cells that protect mice against acute GVHD. Blood 128, 427- 439,659 doi:10.1182/blood- 2015- 12- 684142 (2016).660 68 Yang, J. et al. Rorc restrains the potency of ST2+ regulatory T cells in661 ameliorating intestinal graft- versus- host disease. JCI Insight 4,662 doi:10.1172/jci.insight.122014 (2019).663 69 Zeiser, R. & Blazar, B. R. Acute Graft- versus- Host Disease - Biologic664 Process, Prevention, and Therapy. N Engl J Med 377, 2167- 2179,665 doi:10.1056/NEJMra1609337 (2017).666 70 King, C. G. et al. T cell affinity regulates asymmetric division, effector cell667 differentiation, and tissue pathology. Immunity 37, 709- 720,668 doi:10.1016/j.immuni.2012.06.021 (2012). + +<--- Page Split ---> + +## Acknowledgments + +We thank J. Kirsch, T. Kaiser (Flow Cytometry Core Facility, DRFZ, Berlin), G. Guerra, K. Lehmann, V. Holecska, I. Panse, C.L. Tran (DRFZ, Berlin), C. Scholl, A. Leschke (Transgenics Core facility, MDC, Berlin), T. Abreu Mota, M. Ji- Lu (University of Basel), A. Greco (DKFZ, Heidelberg), and A.N. Hegazy (Charité, Berlin) for technical assistance and advice. This work was supported by the German Research Foundation (DFG grants LO 1542/5- 1, 1542/4- 1), the Swiss National Science Foundation (project grants 310030_185318/1 and Sinergia grant CRSII3_160772/1), the Willy Robert Pitzer Foundation (Pitzer Laboratory of Osteoarthritis Research), the Dr. Rolf M. Schwiete Foundation (Osteoarthritis Research Program), the state of Berlin, and the European Regional Development Fund (ERDF 2014- 2020, EFRE 1.8/11). S.S. is a member of the Berlin Institute of Health at Charité – Universitätsmedizin Berlin. T.M.B., M.D., and N.D. were fellows of the International Max Planck Research School for Infectious Diseases and Immunology. + +## Author contributions + +T.M.B., S.S., and M.L. conceptualized the study. T.M.B., S.S., A.F.M., D.D.P., and M.L. interpreted the results. T.M.B., S.S., and M.L. wrote the manuscript. T.M.B., S.S., A.F.M., M.D., N.D., and P.S. conducted in vivo experiments. S.S., F.H., P.D., G.A.H., and M.F.M. performed and analyzed RNA-Seq and scRNA-Seq experiments. T.M.B. and R.K. designed and generated ll1rl1- ExAB- and ll1rl1- ExC- mice. C.K., T.H., and D.D.P. provided expertise and advice. All authors reviewed and edited the manuscript. + +## Declaration of interest + +D.D.P. is a founder, shareholder, and advisor to Hookipa Pharma Inc. commercializing arenavirus vectors. The remaining authors declare no competing interests. + +## Data availability + +RNA sequencing data generated in this study are available from the Gene Expression Omnibus (GEO, National Center for Biotechnology Information) under the accession codes GSE204695 and GSE204693. All other data supporting the findings of this study are available from the corresponding authors upon reasonable request. + +<--- Page Split ---> + +## 698 Code availability + +699 Code related to the data analysis of this study has been deposited to GitLab (https://agloehninggitlab.gitlab.io/BrunnerServe- Type1- ST2/). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1: A previously unrecognized alternative promoter drives IL-33 receptor expression in antiviral T cells. a, Scheme depicting the curated Il1rl1 gene. b, ST2 surface expression by in vitro differentiated T cell subsets. c, Il1rl1 first exon usage by differentiated T cells (CTL, Th2: \(n = 4\) , Th1: \(n = 3\) , NIH3T3: \(n = 2\) ). d, RNA-Seq coverage and splice junction tracks of ST2+ CTLs, Th1, and Th2 cells ( \(n = 3\) per subset) at the Il1rl1 locus. Chr1:40,377,000-40,465,500; GRCm38.p6/mm10 is shown. e-g, Wildtype mice received LCMV-specific P14 or Smarta T cells and were infected with LCMV-WE. P14 CTLs and Smarta Th1 cells were isolated on d7 p.i., and Il1rl1 first exon usage was analyzed by qPCR (e; CTL, Th1: \(n = 4\) , Th2 control (in vitro): \(n = 2\) ) and RT-PCR (f). g, Chip-Seq tracks indicating T-bet36, STAT438, and GATA-337 binding and ATAC-Seq tracks showing chromatin accessibility in naive or activated
+ +<--- Page Split ---> + +714 LCMV- specific T cells \(^{39,40}\) . h, Computational pipeline to identify alternative transcription start sites (TSSs) between type- 1 (CTL, Th1) and type- 2 (Th2) polarized T cells using cufflinks \(^{71}\) and ProActiv \(^{42}\) . i, Manhattan plot showing identified hits (parameters minAbs=0.25 and promoter FC=2, dashed line: \(P = 0.01\) ). j, Heatmap of identified TSSs with \(P < 0.01\) and their respective ProActiv- normalized expression across all replicates. Data in c, e, and f are representative of two independent experiments. Data are presented as mean ± SD with each dot representing T cells isolated from individual mice. \(P\) was determined using One- way ANOVA with Tukey's post- hoc test. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2: Alternative promoter usage at the ll1rl1 locus is conserved between mice
+ +and humans. a, IGV browser display of the ll1rl1 type- 1 promoter region (chr1:40,386,428 - 40,386,458; GRCm38. p6/mm10) showing a CTL RNA- Seq track, T- bet- and STAT4- binding36,38 chromatin accessibility in naive and activated CTLs39,40 and a PhyloP track indicating evolutionary conservation across 60 vertebrate species43,72. CNS- 5, as identified using Vista44, is highlighted (chr1:40,386,282- 40,386,541; GRCm38. p6/mm10). b, T- bet and STAT4 binding motifs and their respective prediction score within CNS- 5 as determined using JASPAR motif analysis73 (chr1:40,386,429- 40,386,459; GRCm38. p6/mm10). c, IGV browser display of cap analysis of gene expression sequencing (CAGE- Seq) data and corresponding CAGE- associated transcripts (CAT) as provided and published by the FANTOM5 consortium41,45. ChIP- data track showing T- bet binding in human Th1 cells46 (chr2:102,862,400- 102,970,100; GRCh37/hg19). d, IL1RL1 first exon expression in human T cell subsets (CTL: \(n = 6\) , Th1 \(n = 5\) , Th2: \(n = 4\) ). Data are presented as mean \(\pm\) SD with each dot representing T cells isolated from individual donors. \(P\) was determined using One- way ANOVA with Tukey's post- hoc test. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3: Usage of distinct promoters allows T cell subset-specific targeting of ST2 expression. a, Experimental outline and representative FACS plot showing GFP expression by transduced T cells. b,c, Representative FACS plots (b) and quantification (c) of ST2 surface expression by transduced T cells analyzed after 5 days of culture ( \(n = 6\) pooled from two (CTL) or three (Th1, Th2) independent experiments). d, Scheme depicting generation of Il1rl1-ExAB \(^{-/ - }\) and Il1rl1-ExC \(^{-/ - }\) mice using CRISPR/Cas9 in murine zygotes. e,f, Representative histograms (e) and quantification (f) of ST2 surface expression by WT, Il1rl1-ExAB \(^{-/ - }\) , or Il1rl1-ExC \(^{-/ - }\) T cells ( \(n = 4\) ). g, IFN- \(\gamma\) and IL-13 secretion by WT or Il1rl1-ExAB \(^{-/ - }\) T cells after IL-33 stimulation ( \(n = 3\) ). h-n, Wildtype, Il1rl1- \(^{-/ - }\) , Il1rl1-ExAB \(^{-/ - }\) , and Il1rl1-ExC \(^{-/ - }\) mice were infected with LCMV-WE and ST2 expression by splenic CTLs (i,j), Th1 cells (k,l), and Tregs (m,n) was quantified on d7 p.i. (WT: \(n = 9\) , Il1rl1- \(^{-/ - }\) : \(n = 6\) , Il1rl1-ExAB \(^{-/ - }\) : \(n = 8\) , Il1rl1-ExC \(^{-/ - }\) : \(n = 7\) ). Data represent one (g), two (e,f), or three (i-n) independent experiments and are presented as mean \(\pm\) SD with each dot representing one mouse (j,l,n) or one experiment performed with T cells from individual mice (c,f). \(P\) was determined using One-way (j,l,n) or Two-way (c,f) ANOVA with Tukey's post-hoc test.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4: T cell-intrinsic activity of the type-1 Il1rl1 promoter is vital for efficient expansion of antiviral T cells. a-h, Wildtype, Il1rl1-/-; Il1rl1-ExAB-/- and Il1rl1-ExC-/- mice were infected with LCMV-WE, and splenic T cells were analyzed on d7 p.i. (WT: \(n = 9\) , Il1rl1-/-: \(n = 6\) , Il1rl1-ExAB-/-: \(n = 8\) , Il1rl1-ExC-/-: \(n = 7\) ). a, Experimental outline. b, c, Frequencies (b) and absolute cell counts (c) of CTLs. d-f, Representative staining (d) and frequencies of activated CD44+ CD62L- CTLs (e) and Ki67+ CTLs (f). g, Absolute cell counts of LCMV GP33-41- or NP396-404-specific CTLs. h, Counts of CTLs expressing effector molecules after ex vivo stimulation with GP33-41. i, Serum IFN-γ levels at indicated timepoints p.i. with LCMV-WE (WT and Il1rl1-/-: \(n = 4\) , Il1rl1-ExAB-/-: \(n = 5\) , Il1rl1-ExC-/-: \(n = 3\) ). j-m, Irradiated WT recipients (CD45.1+) were reconstituted with WT (CD45.1+ CD45.2+) and Il1rl1-ExAB-/- or Il1rl1-ExC-/- (all CD45.2+) bone marrow, infected with LCMV-WE, and analyzed on d10 p.i. (WT + Il1rl1-/-: \(n = 6\) ,
+ +<--- Page Split ---> + +770 WT + Il1rl1- ExAB- and WT + Il1rl1- ExC-: \(n = 7\) ). j, Experimental outline. k, Representative FACS plots showing CTL populations. I,m, Cell counts of splenic CTLs (I) and CTLs expressing IFN- \(\gamma\) after ex vivo stimulation with \(\mathsf{GP}_{33 - 41}(\mathsf{m})\) . n- r, P14 T cells (CD45.1+ CD45.2+) were adoptively cotransferred with Il1rl1- P14 or Il1rl1- ExAB- P14 T cells (CD45.1+) into WT mice (CD45.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. ( \(n = 7\) ). n, Experimental outline. o, Representative FACS plots showing CTL populations. p,q, Frequencies (p) and cell counts (q) of splenic P14 T cells. r, Counts of IFN- \(\gamma\) - expressing P14 cells after ex vivo stimulation with \(\mathsf{GP}_{33 - 41}\) . s- w, Smarta T cells (CD90.1+) were adoptively cotransferred with Il1rl1- Smarta or Il1rl1- ExAB- Smarta T cells (CD90.1+ CD90.2+) into WT mice (CD90.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. (Il1rl1- Smarta: n=6, Il1rl1- ExAB- Smarta: n=7). s, Experimental outline. t, Representative FACS plots showing CD4+ T cell populations. u,v, Frequencies (u) and cell counts (v) of splenic Smarta T cells. w, Counts of IFN- \(\gamma\) - expressing Smarta cells after ex vivo stimulation with \(\mathsf{GP}_{64 - 79}\) . Data represent one (i- m), two (n- w), or three (a- h) independent experiments and are presented as mean \(\pm\) SD with each dot representing one mouse or the arithmetic mean of all replicates (h,i). P was determined using Oneway ANOVA (b,c,e,f), Two- way (g,i) ANOVA with Tukey's post- hoc test, or Two- way RM ANOVA with Sidak's post- hoc test (I,m,p- r,u- w). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5: Type-1 llrl1 promoter engagement facilitates effector differentiation of CTLs to generate a clonally diverse SLEC population. a-d, n-r, Wildtype and llrl1-ExAB-/- mice were infected with LCMV-WE, and splenic CD44+ CTLs were analyzed by multiplexed scRNA-Seq and scTCR-Seq analysis on d7 p.i. (n = 3 mice pooled per group). a, Experimental outline. b, UMAP plots colored by cluster type. c, UMAP plots showing normalized expression of selected genes in both genotypes. d, Change in cluster composition of llrl1-ExAB-/- CTLs relative to WT CTLs. e, Representative FACS plots showing KLRG1 and CD127 expression by CD44+ CD8+ T cells. f-h, Frequencies (f) and counts (g) of KLRG1+ CD127- SLECs and frequencies (h) of
+ +<--- Page Split ---> + +799 CXCR3+ CTLs in LCMV- WE infected WT, Il1rl1- /-, Il1rl1- ExAB- /-, and Il1rl1- ExC- /- mice (e- h, WT: \(n = 9\) , Il1rl1- /- : \(n = 6\) , Il1rl1- ExAB- /- : \(n = 8\) , Il1rl1- ExC- /- : \(n = 7\) ). i, Frequencies of KLRG1+ CD127- SLECs in infected mixed bone marrow chimeras (WT + Il1rl1- /- : \(n = 6\) , WT + Il1rl1- ExAB- /- and WT + Il1rl1- ExC- /- : \(n = 7\) ). j, k, Frequencies of KLRG1+ CD127- SLECs (j) and CXCR3+ P14 cells (k) in adoptive cotransfer experiments on d10 p.i. \((n = 7)\) . m, ST2 expression by GP33- 41- specific SLECs and MPECs in WT mice at d7 p.i. \((n = 9)\) . n, Counts of CD44+ CTLs per spleen. o, TCR clonotypes in sequenced CD44+ CTLs. p, Graph displaying number of TCR clonotypes and their abundancy among all analyzed CD44+ CTLs. q, TCR clonotypes in SLEC and MPEC clusters. r, TCR repertoire occupation in individual WT and Il1rl1- ExAB- /- mice. Data represent one (b- d, i, n- r) or two (e- h, j- m) independent experiments and are presented as mean \(\pm\) SD with each dot or line representing one mouse (f- j, l, n- q). \(P\) was determined using two- tailed t- tests (n, q), One- way ANOVA with Tukey's post- hoc test (f, h), or Two- way RM ANOVA with Sidák's post- hoc test (i, j, l). + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6: Transcriptional profiling indicates broad costimulatory and TCR-cooperative functions of IL-33–ST2 signaling in T cells. a-f, Differentiated CTLs,
+ +Th1 cells, and Th2 cells were stimulated with IL- 33 or left untreated for 2h and subjected to RNA- Seq analysis \((n = 3\) independent cultures per subset). a, Experimental outline. b, Heatmaps depicting differentially expressed genes in each T + +<--- Page Split ---> + +819 cell subset (log2 fold change \(>1.0\) ; \(P\) adjusted \(< 0.01\) ). c- f, Comparison of IL- 33- 820 stimulated and untreated T cells (2h timepoint). c, Venn- diagram illustrating the 821 overlap in differentially regulated genes between CTLs, Th1, and Th2 cells. d, Gene 822 ontology biological process overrepresentation analysis of IL- 33- induced genes 823 shared among all subsets. e, Expression of selected transcription factors, cytokines 824 and chemokines, and TCR- regulated genes in each T cell subset. f, Gene set 825 enrichment analysis of TCR downstream genes in IL- 33- stimulated versus 826 unstimulated CTLs. g- I, P14 T cells ( \(\mathrm{CD45.1^{+}CD45.2^{+}}\) ) were adoptively cotransferred 827 with \(11r1r1^{- / - }\) P14 T cells ( \(\mathrm{CD45.1^{+}}\) ) into WT mice ( \(\mathrm{CD45.2^{+}}\) ). Recipients were infected 828 with high- or low- affinity GP33- expressing LCMV- CI13 and analyzed at d10 p.i. (High- 829 affinity group: \(n = 7\) , Low- affinity group: \(n = 6\) ). g, Experimental outline. h, Counts of 830 recovered P14 and \(11r1r1^{- / - }\) P14 T cells. i,j, Frequencies (i) and counts (j) of KLRG1+ 831 CD127- P14 cells. k,l, Counts of endogenous \(\mathrm{NP}_{396 - 404}\) - specific CTLs (k) and KLRG1+ 832 CD127- \(\mathrm{NP}_{396 - 404}\) - specific CTLs (l). Data represent one (a- f) or two (g- l) independent 833 experiments. Data are presented as mean ± SD with each dot representing one mouse 834 (h- l). \(P\) was determined using One- way ANOVA with Tukey's post- hoc test (h- j) or 835 two- tailed t- tests (k,l). + +<--- Page Split ---> +![](images/Extended_Data_Figure_1.jpg) + +
837 Extended Data Fig. 1: ST2 expression by CTLs and Th1 cells is regulated by T-
+ +837 Extended Data Fig. 1: ST2 expression by CTLs and Th1 cells is regulated by T- bet and STAT4. a, Experimental scheme of in vitro T cell differentiation. b, Representative T- bet and GATA- 3 stainings of differentiated T cells with expression intensity depicted as geometric mean index (GMI), normalized to isotype control stainings (grey) \((n = 4)\) . c, Cytokine expression of polarized T cells upon restimulation \((n = 4)\) . d- g, Representative histograms (d,f) and quantification (e,g) of ST2 surface expression by in vitro differentiated wildtype, STAT4-, or T- bet- deficient T cells, or T cells activated in the absence of IL- 12. Stainings with secondary reagents without primary ST2 antibody served as staining controls (ctrl) (dotted line, bottom). h, ST2 expression by splenic Tregs in wildtype, Stat4- , or Tbx21- mice. Data represent two independent experiments and are presented as mean ± SD (b- h) with each dot representing one mouse (h) or one culture performed with T cells from individual mice (c,e,g). P was determined using One- way ANOVA with Tukey's post- hoc test (e,g,h). + +<--- Page Split ---> +![PLACEHOLDER_36_0] + + +Extended Data Fig. 2: Flow- cytometric sorting and analysis of human T cells. a, Gating strategy for the sorting of human CXCR3+ CRTH2- Th1 cells and CXCR3+ CRTH2+ Th2 cells. b, Gating strategy for the sorting of human CD8+ T cells. c,d, Representative histograms and quantification of T- bet (c) and GATA- 3 (d) expression by activated T cells at day 5 of culture \((n = 3)\) . e,f, Representative histograms and quantification of IFN- \(\gamma\) (e) and IL- 13 (f) expression by T cells stimulated with PMA/ionomycin at day 5 of culture \((n = 3)\) . Data represent two independent experiments and are presented as mean \(\pm\) SD with each dot representing one culture performed with T cells from individual donors (c- f). \(P\) was determined using One- way ANOVA with Tukey's post- hoc test (c- f). + +<--- Page Split ---> +![PLACEHOLDER_37_0] + + +Extended Data Fig. 3: Generation of Il1rl1- ExAB- and Il1rl1- ExC- mice. a, Schematic depiction of the gene- targeting approach for the generation of Il1rl1- ExAB- and Il1rl1- ExC- mice. b, Representative genotyping PCRs to identify heterozygous and homozygous mutant mice. Chromatograms depicting the sequence of joined DNA + +<--- Page Split ---> + +segments as analyzed by Sanger- Sequencing. c- k, Analysis of adaptive and innate immune cells in spleens and lymph nodes (LN) of llrl1- ExAB- and llrl1- ExC- mice. c, Representative staining of CD4 and CD8 on splenic T cells. d, e, Frequencies (d) and absolute cell counts (e) of \(\mathsf{CD8^{+}}\) T cells, \(\mathsf{CD4^{+}}\) T cells, and B cells. f, Representative staining of CD62L and CD44 on splenic \(\mathsf{CD8^{+}}\) T cells. g, Frequency of effector and central memory \(\mathsf{CD8^{+}}\) T cells. h, Frequencies and absolute cell counts of splenic Tregs. i, Gating strategy for the analysis of innate immune cells. j, k, Frequencies (j) and absolute cell counts (k) of splenic neutrophils, eosinophils, macrophages (MΦ), conventional dendritic cells (cDCs), and natural killer (NK) cells. Data represent two independent experiments and are presented as mean ± SD with each dot representing one mouse (Wildtype, llrl1- ExAB- , and llrl1- ExC- : \(n = 4\) , llrl1- ExAB- : \(n = 5\) ; Treg analysis: Wildtype, llrl1- ExAB- : \(n = 8\) and llrl1- ExC- : \(n = 6\) ; NK cell analysis: Wildtype, llrl1- ExAB- , and llrl1- ExC- : \(n = 4\) ). P was determined using One- way ANOVA (h, j, k) or Two- way ANOVA (d, e, g) with with Tukey's post- hoc test. + +<--- Page Split ---> +![PLACEHOLDER_39_0] + + +Extended Data Fig. 4: Type- 1 llrl1 promoter deficiency does not impair ST2 expression by type- 2 innate immune cells. a, RNA- Seq coverage tracks and detected splice junctions at the llrl1 locus of ILC2s74 and mast cells75. chr1:40,377,000- 40,465,500; GRCm38. p6/mm10 is shown. b, Gating strategy for the analysis of Lin- CD45+ CD3- CD127+ ILCs. c, Representative histograms showing ST2 expression by KLRG1+ IL- 18R- ILC2s isolated from lungs of wildtype or llrl1- ExAB- mice. d, Frequencies of ST2+ ILC2s and ST2 expression intensities (MFI). e, Gating strategy for the analysis of c- kit+ FcεRla+ peritoneal mast cells. f, Representative histograms showing ST2 expression by peritoneal mast cells isolated from wildtype or llrl1- ExAB- mice. g, Frequencies of ST2+ mast cells and ST2 expression intensities (MFI). Results are presented as mean ± SD with each dot representing one mouse (n + +<--- Page Split ---> + +894 = 4 mice per genotype). Data are representative of two experiments. \(P\) was 895 determined using using two- tailed t- tests (d,g). + +<--- Page Split ---> +![PLACEHOLDER_41_0] + + +Extended Data Fig. 5: The type- 1 llr1l promoter drives expansion and activation of antiviral T cells. a- i, Wildtype, llr1l- ExAB- , and llr1l- ExC- mice were infected with LCMV- WE and the T cell response was analyzed on d7 p.i. (wildtype: \(n = 9\) , llr1l- : \(n = 6\) , llr1l- ExAB- : \(n = 8\) , llr1l- ExC- : \(n = 7\) ). a, Experimental outline. b, c, Representative FACS plots (b) and quantification (c) of CTLs in livers of infected animals. d- f, Representative FACS plots (d) and quantification of ST2 expression (e) and GP33- 41- Tetramer+ cells in livers (f). g- i, Representative FACS plots (g) and quantification (h, i) of effector molecule expression by splenic CTLs stimulated with LCMV GP33- 41 peptide. j, k, Irradiated wildtype recipients (CD45.1+) were reconstituted + +<--- Page Split ---> + +with wildtype (CD45.1+ CD45.2+) and Il1rl1- /-, Il1rl1- ExAB- /-, or Il1rl1- ExC- (all CD45.2+) bone marrow, infected with LCMV- WE, and analyzed on d10 p.i. (WT + Il1rl1- /-: \(n = 6\) , WT + Il1rl1- ExAB- /- and WT + Il1rl1- ExC- /-: \(n = 7\) ). j, Experimental outline. k, Absolute cell counts of LCMV GP33- 41- specific CTLs. I- p, Smarta T cells (CD90.1+) were adoptively cotransferred with Il1rl1- /- Smarta or Il1rl1- ExAB- /- Smarta T cells (CD90.1+ CD90.2+) into WT mice (CD90.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. (Il1rl1- /- Smarta: \(n = 6\) , Il1rl1- ExAB- /- Smarta: \(n = 7\) ). I, Experimental outline. m, n, Representative FACS plots showing CD4+ T cell populations before transfer (m) and after infection (n). o, p, Frequencies (o) and cell counts (p) of Smarta T cells in livers of infected mice. Data represent one (j- p), two (a- f), or three (g- i) independent experiments and are presented as mean ± SD with each dot representing one mouse. P was determined using One- way ANOVA (c, e, h, i) with Tukey's post- hoc test, Two- way ANOVA (f) with Tukey's post- hoc test, or with Two- way RM ANOVA with Sidak's post- hoc test (k, o, p). + +<--- Page Split ---> +![PLACEHOLDER_43_0] + + +Extended Data Fig. 6: scRNA-Seq profiling of antiviral T cells in wildtype and Il1rl1- ExAB- mice. a, Gating strategy for the flow- cytometric sorting of CD44+ CTLs from spleens of LCMV- infected wildtype and Il1rl1- ExAB- mice. b, Heatmap displaying all analyzed CTLs and the top ten marker genes per cluster. + +<--- Page Split ---> + +## Methods + +## Mice + +C57BL/6J mice (wildtype), LCMV- TCRtg P1476 and Smarta77 mice expressing the congenic markers CD45.1 (Ly5.1) or CD90.1 (Thy1.1), respectively, I1/r1- /- 12, I1/r1- 929 ExAB- /-, I1/r1- ExC- /-, Stat4- /- 78, Tbx21- /- 79, I1/r1- ExAB- /- Smarta, Stat4- /- Smarta, Tbx21- /- Smarta, I1/r1- /- P14, I1/r1- ExAB- /- P14, and TCRβδ- /- 80 mice were bred under specific- pathogen- free conditions in approved animal- care facilities at the Research Institute for Experimental Medicine (FEM) of the Charité – Universitätsmedizin Berlin (Berlin, Germany) or at the Laboratory Animal Facility of the ETH Zürich (ETH Phenomics Center, Zürich, Switzerland). Mice were housed in individually ventilated cages with a 12 h light/dark cycle and had ad libitum access to drinking water and chow. Both, male and female mice between 8 and 26 weeks of age were used for experiments. For LCMV infections, experimental groups were age- and sex- matched. Mice used for scRNA- Seq analyses were cohoused for 4 weeks prior to LCMV infection. Animal experiments were performed in accordance with the German or Swiss law for animal protection and approved by the respective governmental authority (Landesamt für Gesundheit und Soziales Berlin and the Cantonal Veterinary Office of the Canton of Basel; T0058/08, G0111/17, G0206/17, G0245/19). + +## Generation of I1/r1-ExAB- and I1/r1-ExC- mice + +I1/r1- ExAB- and I1/r1- ExC- mice were generated in the Transgenics Core Facility of the Max Delbrück Centrum Berlin using established protocols81. In brief, gRNA sequences with minimal predicted off- target effects (gRNA- ExAB- 5'.1: TAATGCATAGGACCGACCAAAGG, gRNA- ExAB 3'.1: ATTTGGACTGTCATCGTC GTGGG, gRNA- ExAB- 5'.2: GCACCCTATGAATCTACACAGGG, gRNA- ExAB- 3'.2: GACTTGCATCGTCTGGGCCGGG, gRNA- EXC- 5': CCAACGTGGCAGTATGTC CAGG, gRNA- EXC- 3': ATGAAACGGTGACTCAGTTAGGG) were identified using the web- based tool CRISPR82. Zygotes were collected from C57BL/6J mice (Charles River), microinjected with synthetic gRNAs (Integrated DNA Technologies) and recombinant Cas9 protein (Integrated DNA Technologies), and subsequently transferred into pseudo- pregnant C57BL/6J mice. Resulting F0 offspring mice were screened for successful deletion by PCR amplification of wildtype or knockout alleles using the following primers: I1/r1- ExAB knockout (forward: CATAGGCCCAGGTTA + +<--- Page Split ---> + +GGCCTGAAg,reverse: GAATGTGAACCCTCGTCGACCTG), Il1rl1- ExAB wildtype (forward: CCATTAGCGTGTCTCTACTACGACAg, reverse: GAATGTGAACCCTCG TGTGACGTG), Il1rl1- ExC knockout (forward: GAAGGCTAGTGGCTTCTGTGAGCA T, reverse: CACAGTCTCAGGCAAGCTGCCTG), Il1rl1- ExC wildtype (forward: GAA GGCTAGTGGCCTTCTGTGAGCAT, reverse: TCAGCAATAACCAGCCATCGAGG). + +## Viruses and LCMV infection + +LCMV- WE and LCMV- CI13 strains were propagated on L929 or BHK- 21 cells, respectively. Viral titers in stock solutions were determined by immunofocus assay on MC57G cells as described before83. In brief, MC57G cells were plated with virus stock dilutions and subsequently overlaid with \(2\%\) methylcellulose. After \(48h\) of incubation at \(37^{\circ}C\) , the confluent monolayer of cells was fixed with \(4\%\) formaldehyde, permeabilized with Triton X- 100 ( \(1\%\) , v/v), and stained with antibodies against LCMV nucleoprotein. After a secondary staining step with peroxidase conjugated anti- rat IgG antibody, foci were developed by 20 min incubation with OPD substrate. Mice were infected intravenously (i.v.) with either 200 plaque forming units (PFU) of LCMV- WE (mixed bone- marrow chimera experiments), 200 PFU LCMV- CI13 (adoptive cotransfer experiments, LCMV- CI13 wt or C6 variant where indicated), or \(2\times 10^{6}\) PFU of LCMV- WE in minimal essential medium (MEM, Thermo Scientific). + +## Adoptive T cell transfers + +For adoptive transfer experiments, P14, Il1rl1- P14, Il1rl1- ExAB- P14, Smarta, Il1rl1- Smarta, or Il1rl1- ExAB- Smarta T cells expressing the congenital markers CD45.1 or CD90.1 were enriched in a negative selection approach. To this end, splenocytes of donor mice were stained with biotinylated antibodies against mouse CD11b (M1/70, eBioscience), CD11c (N418, eBioscience), CD19 (1D3, eBioscience), CD25 (7D4, eBioscience), Gr- 1 (RB6- 8C5, eBioscience), NK1.1 (PK136, eBioscience), CXCR3 (CXCR3- 173, eBioscience), and CD8a (53- 6.7, eBioscience, for isolation of Smarta T cells) or CD4 (RM4- 5, for isolation of P14 T cells) followed by incubation with anti- biotin microbeads (Miltenyi Biotec). Subsequently, labeled cells were depleted by magnetic activated cell sorting (MACS) using LS columns (Miltenyi Biotec). \(5\times 10^{4}\) T cells (single transfers experiments), \(1\times 10^{3}\) P14 T cells or \(2\times 10^{4}\) Smarta T cells (cotransfer experiments) were transferred i.v. into C57BL/6J mice. Recipients were infected 1- 2 days post transfer and analyzed at indicated timepoints post infection. + +<--- Page Split ---> + +## Mixed bone marrow chimeras + +To generate mixed bone marrow chimeras, CD45.1+/+ wildtype recipients were lethally irradiated (two doses of 5.5 gray given in a 6- hour interval). One day later, recipients were reconstituted with a 1:1 mixture of CD45.1+/+ wildtype and CD45.2+/+ knockout bone- marrow cells and splenocytes. After 8 weeks of hematopoietic reconstitution, CTL frequencies of respective donor populations were determined in blood, and mice were infected with LCMV- WE (200 PFU i.v.). Data were analyzed on d10 p.i. and normalized to CTL frequencies before infection. + +## Lymphocyte isolation + +To isolate lymphocytes, spleens were mechanically disrupted and filtered through 70 \(\mu \mathrm{m}\) strainers. Erythrocytes were lysed by incubation in erythrocyte lysis buffer (10 mM KHCO3, 155 mM NH4Cl, 0.1 mM EDTA, pH 7.5). Livers were collected in PBS/BSA, meshed and centrifugated at 30 g for 2 min to remove debris. Supernatants were subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich) and lymphocytes were collected at the gradient interphase. Lungs were cut into small pieces and digested with Collagenase D (0.1 U/ml) in RPMI1640 (supplemented with 10% FCS and 15mM HEPES) for 1 h at 37°C. Afterwards, lymphocytes were isolated by Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich). To isolate peritoneal cavity cells, 5 ml of cold PBS was injected into the peritoneal cavity of euthanized mice. After a brief massage of the peritoneum, cell- containing liquid was collected and subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich). + +## Flow cytometry + +Surface stainings of purified lymphocytes were performed using different combinations of following anti- mouse antibodies diluted in PBS: CD4 PerCP- Cyanine5.5 (RM4- 5, eBioscience), FITC (RM4- 5, eBioscience), Pacific Blue (YTS19.1, DRFZ inhouse production) or APC- Cy7 (GK1.5, BD); CD8a PE (53- 6.7, BD), BV500 (53- 6.7, BD), BV605 (53- 6.7, Biolegend), BV786 (53- 6.7, Biolegend) PerCP- Cyanine5.5 (53- 6.7, BD) or eFluor450 (53- 6.7, eBioscience); Va2 FITC (B20.1, BD); CD19 Cy5 (ID3, DRFZ inhouse production), APC- Fire750 (6D5, Biolegend) or PerCP- Cyanine5.5 (eBio1D3, eBioscience); B220 APC- Fire750 (RA3- 6B2, Biolegend); CD90.1 Pacific Blue (Ox- 7, DRFZ inhouse production); CD90.2 FITC (53- 2.1, Biolegend); CD45.1 Pacific Blue (A20, Biolegend), FITC (A20, Biolegend) or PE (A20, BD); CD45.2 APC (104, BD), + +<--- Page Split ---> + +1021 V500 (104, BD), BV510 (30- F11, Biolegend) or APC- Fire750 (104, Biolegend); CD44 1022 APC- Cy7 (IM7, Biolegend) or BV450 (IM7, Biolegend); CD62L PE- Cy7 (MEL- 14, 1023 eBioscience); KLRG1 FITC (2F1, eBioscience) or PerCP- Cy5.5 (2F1/KLRG1, 1024 Biolegend); CD127 PE- Cy7 (A7R34, eBioscience) or BV421 (A7R34, Biolegend); 1025 NK1.1 PE (PK136, BD); GR- 1 A488 (RB6- 8C5, DRFZ inhouse- production) or APC- 1026 Cy7 (RB6- 8C5, Biolegend); Siglec- F PE (E50- 2440, BD); c- kit (2B8, eBioscience); 1027 FcεRla FITC (MAR- 1, eBioscience) or APC- Cy7 (MAR- 1, Biolegend); MHC- II Cy5 1028 (M5/114, DRFZ inhouse production); CD11c APC- Fire750 (N418, Biolegend); CD11b 1029 PE- Cy7 (M1/70, BD) or APC- Fire750 (M1/70, Biolegend); Ly6G PerCP- Cy5.5 (1A8, 1030 Biolegend); F4/80 PE- Cy7 (BM8, Biolegend) or APC- Fire750 (BM8, Biolegend); ST2 1031 BV421 (DIH9, Biolegend). Unspecific staining was minimized by blocking with rat 1032 immunoglobulin G (IgG, Jackson ImmunoResearch) and anti- mouse CD16/32 (2.4G2, 1033 DRFZ inhouse production) prior to staining. Dead cells were labelled using Zombie 1034 Aqua or Zombie NIR fixable live/dead staining reagents (Biolegend) or by adding 1035 propidium iodide (PI) prior to acquisition. For detection of ST2 on T cells, lymphocytes 1036 were first stained with digoxigenin- conjugated antibody against ST2 (DJ8, 1037 mdBioscience), followed by a secondary staining with PE- or APC- conjugated anti- 1038 digoxigenin Fab fragments (Roche). Further, stainings were enhanced by two- rounds 1039 of PE- or APC- FASER amplification (Miltenyi Biotec). To identify LCMV- specific T 1040 cells, lymphocytes were stained with LCMV GP33- 41 or NP396- 404 peptide- loaded MHC 1041 class I (H2- Db) tetramers (PE- or APC- conjugated, MBL or NIH Tetramer Core 1042 Facility) for 30 min at 37°C. For detection of transcription factors or Ki67 expression, 1043 surface- stained cells were fixed and stained using the FoxP3 staining buffer set 1044 (Thermo Scientific). Briefly, cells were fixed with 1x fixation/permeabilization reagent 1045 for 30 min at 4°C and washed with permeabilization buffer. Subsequently, cells were 1046 stained with antibodies against mouse T- bet BV421 (4B10, Biolegend) or PE (4B10, 1047 eBioscience); GATA- 3 eFluor660 (TWAJ, eBioscience), FoxP3 eFluor450 (FJK- 16s, 1048 eBioscience), Ki67 FITC (SolA15, eBioscience), or with respective isotype control 1049 antibodies: mouse IgG1 kappa BV421 (MOPC- 21, Biolegend, T- bet isotype) or mouse 1050 IgG1 kappa PE (P3.6.2.8.1, eBioscience, T- bet isotype) or rat IgG2b kappa eFluor660 1051 (10H5, eBioscience, GATA- 3 isotype) diluted in permeabilization buffer for 30 min at 1052 4°C. + +<--- Page Split ---> + +For flow- cytometric detection of cytokines, lymphocytes were restimulated with phorbol myristate acetate (PMA) (5 ng/ml, Sigma- Aldrich) and ionomycin (5 μg/ml, Sigma- Aldrich), recombinant LCMV GP33- 41 (Charité Berlin, 1 μg/ml) or LCMV GP64- 79 (Charité Berlin, 1 μg/ml) for 4 h at 37°C. After 35 min, brefeldin A (5 μg/ml, Sigma- Aldrich) was added to prevent secretion of produced cytokines. After the stimulation period, cells were labelled with surface- antibodies and fixable live/dead staining reagents, followed by fixation in 2% paraformaldehyde for 10 min at RT. Intracellular cytokines were stained with antibodies against IFN- γ eFluor450 (XMG1.2, eBioscience) or APC (XMG1.2, Biolegend); IL- 2 APC (JES6- 5H4, BD); TNF eFluor450 (MP6- XT22, eBioscience); Granzyme A PE (3G8.5, Biolegend); Granzyme B APC (QA16A02, Biolegend); Perforin PE (S16009B, Biolegend); IL- 4 PE (11B11, eBioscience); IL- 13 Alexa Fluor488 (eBio13A, eBioscience); IL- 10 APC (JES5- 16E3) in PBS containing 0.05% Saponin for 30 min at 4°C and washed before acquisition. Cells were acquired on a Canto II or LSRFortessa flow- cytometer (BD). Flowcytometric sorting was performed on Aria and Aria II devices (BD). Cell numbers were determined using MACSQuant (Miltenyi Biotec) or ImmunoSpot (CTL) analyzers. + +## Legendplex cytometric bead assay + +To assess cytokine production by T cells in response to IL- 33, 5x105 T cells were stimulated in 48- well plates with IL- 33 (R&D, 10 ng/ml) for 24 h at 37°C. Afterwards, individual wells were harvested and centrifuged for 5 min at 350 g. To obtain serum, blood of individual mice was collected using yellow microtainers (BD). Serum and cell- free supernatant were frozen and stored at - 80°C until analysis. Cytokine content was measured using LEGENDplex bead- based immunoassays (Biolegend) according to manufacturer's instructions. In brief, samples and serial dilutions of the standard were incubated for 2 h on a plate shaker set to 900 rpm with premixed, antibody- coated beads to capture cytokines. Afterwards, beads were washed twice with washing buffer and stained with biotinylated detection antibodies. After 1 h incubation on a plate shaker at 900 rpm, streptavidin- PE was added for additional 30 min. Beads were washed twice and acquired at a Canto II flow- cytometer (BD). Cytokine concentration was extrapolated from standard titrations. + +<--- Page Split ---> + +## Mouse T cell cultures + +Mouse T cell culturesNaive T cells from spleens of indicated mice were pre- enriched by staining with biotinylated antibodies against CD8a (53- 6.7, eBioscience) or CD4 (RM4- 5, eBioscience), followed by incubation with anti- biotin microbeads (Miltenyi Biotec) and subsequent separation by MACS using LS columns (Miltenyi Biotec). Following enrichment, naive (CD62L+ CD44- CD25- CXCR3-) CD8+ or CD4+ T cells were flow- cytometrically sorted and differentiated in the presence of irradiated TCRβδ- splenocytes and antibodies against CD3ε (145- 2C11, eBioscience) and CD28 (37.51, both 2.5 μg/ml). When naive T cells were isolated from LCMV- TCRtg mice, cognate LCMV GP33- 41 (P14 mice) or GP64- 79 peptide (Smarta mice, both 1μg/ml) where added instead. T cells were cultivated in RPMI1640 + GlutaMax I (Thermo Scientific) medium supplemented with fetal calf serum (FCS) (10% v/v, Thermo Scientific), penicillin (100 U/ml, Thermo Scientific), streptomycin (100 μg/ml, Thermo Scientific), gentamycin (10 μg/ml, Thermo Scientific), and β- mercaptoethanol (50 ng/ml, Sigma- Aldrich). For CTL and Th1 differentiation, IL- 12 (5 ng/ml), IL- 2 (5 ng/ml, all Miltenyi Biotec), and anti- IL- 4 (11B11, 10 μg/ml, DRFZ inhouse production) were added. For Th2 differentiation, IL- 4 (5 ng/ml), IL- 2 (5 ng/ml, all Miltenyi Biotec), anti- IL- 12 (C18.2, 10 μg/ml), and anti- IFN- γ (XMG1.2, 10 μg/ml, all DRFZ inhouse production) were added. T cells were split after 2- 3 days of culture in a 1:3 ratio with fresh medium containing IL- 2 (5 ng/ml), harvested at day 5 of culture using Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich), and cultivated for additional 5 days in identical culture conditions. + +## Retroviral transduction of T cells + +Retroviral transduction of T cellsFor cloning of shRNA expression vectors, the following sense- and antisense- shRNA sequences were ordered as phosphorylated oligos with a 5' Sal- I restriction overhang (Eurofins Genomics) and annealed by subjecting equimolar amounts of oligos diluted in oligo annealing buffer (100 mM Tris- HCl, 1 M NaCl, 10 mM EDTA, pH 7.5) to a decreasing temperature gradient (95°C to 25°C with 1°C/min): shRNA- ExAB: (Sense: TGGCCTAGAATGAGAAAGGAAATTCAAGAGATTTCCTTCCA TTCTGAGCCTTTTTTC, antisense: TCGAGAAAAAAGGCTCAGAATGGAAAGGAA TTCTCTGAATTCCTTTCCTATTCTGAGCCA); shRNA- Ex1a (Sense: TGTTGGAAGC TGAGGAATAATTCAAGAGATTATTCCTCAGCTTGCAAACTTTTTTC, antisense: TC GAGAAAAAAGTTTCGAAGCTGAGGAATAATCTCTTGAATTATTCCTCAGCTTGCAA CA). PQCXIX- GFP target vector84 was digested by Sall and Hpal restriction enzymes + +<--- Page Split ---> + +(Thermo Scientific) and simultaneously dephosphorylated with FastAP alkaline phosphatase (Thermo Scientific). Annealed oligos were ligated using T4 Ligase according to standard protocols (NEB). Heat inactivated ligation reactions were directly used for heat-shock transformation into Oneshot TOP10 chemically competent E.coli (Thermo Scientific). Single transformed bacterial clones were selected on LB- agar plates (MP Biomedicals) containing ampicillin (100 \(\mu \mathrm{g / ml}\) , Sigma-Aldrich), and plasmid DNA was prepared using QIAprep Spin Plasmid Maxi or Midi kits (Qiagen. Correct plasmid sequences were verified by Sanger- sequencing (Eurofins Genomics). Virus particles were generated by co- transfection of HEK293T cells with shRNA- containing vectors and packaging plasmids pCGP and pECO85 using T5 Transporter reagent. For retroviral transduction, mouse T cells were activated in the presence of irradiated APCs and antibodies against CD3ε and CD28 described above. 36- 48 h after plating, culture medium was temporarily replaced with virus- containing supernatant, polybrene (8 \(\mu \mathrm{g / ml}\) , Sigma-Aldrich) was added and plates were centrifuged for 90 min at 450 g at RT. T cells were incubated at 37°C for 6- 8 h. Afterwards, viral supernatant was replaced with conditioned cell culture medium and cells were split in a 1:3 ratio with fresh IL- 2- containing medium. Transduced T cells were analyzed between day 5 and day 7 of culture. + +## Human T cell cultures + +Human peripheral blood which was obtained from the German Red Cross (DRK Berlin, Germany; Ethics approval EA1/149/12). For isolation of T cells, blood was first subjected to Ficoll- Paque PLUS density centrifugation (1.077 g/ml, Cytiva). Interphases were collected, stained with anti- CD4 microbeads (Miltenyi Biotec), and separated using LS columns (Miltenyi Biotec). CD4+ T cell- depleted fractions were used for a MACS enrichment of CD8+ T cells using anti- CD8a microbeads (Miltenyi Biotec). CD4- enriched fractions were stained with antibodies against human CD4 Pacific Blue (TT1, DRFZ inhouse production), CXCR3 FITC (G025H7, Biolegend), and CRTH2 biotin (BM16, Miltenyi Biotec) for 15 min at 4°C followed by a secondary staining with Streptavidin PE (BD). CD8- enriched fractions were stained with antibodies against human CD8 (GN11/134.7, DRFZ inhouse production), CD56 APC (REA196, Miltenyi Biotec), CD62L PE- Cy7 (DREG- 56, Biolegend), and CD45RA PE (HI100, DRFZ inhouse production). In vivo differentiated Th1 cells were sorted as CD4+ CXCR3+ CRTH2+, and Th2 cells were sorted as CD4+ CXCR3+ CRTH2+. CD8+ + +<--- Page Split ---> + +1149 effector/effector memory T cells were sorted as CD8+ CD56- CD45RA- CD62L- . For 1150 activation of human T cells, suspension culture plates were coated with antibodies 1151 against human CD3ε (OKT3, 2.5 μg/ml, Miltenyi Biotec) and CD28 (15E3, 2.5 μg/ml, 1152 Miltenyi Biotec), and sorted T cells were plated in RPMI1640 + GlutaMax I (Thermo 1153 Scientific) medium supplemented with FCS (10% v/v, Thermo Scientific), penicillin 1154 (100 U/ml, Thermo Scientific), streptomycin (100 μg/ml, Thermo Scientific), 1155 gentamycin (10 μg/ml, Thermo Scientific), and \(\beta\) - mercaptoethanol (50 ng/ml, Sigma1156 Aldrich). To CTL and Th1 cultures, IL- 12 (10 ng/ml, R&D Systems), IL- 2 (10 ng/ml, 1157 R&D Systems), and anti- IL- 4 (7A3- 3, 10 μg/ml, Miltenyi Biotec) were added, while Th2 1158 were cultured in the presence of IL- 4 (10 ng/ml, R&D Systems), IL- 2 (10 ng/ml, R&D 1159 Systems), anti- IL- 12 (C8.6, 10 μg/ml, Miltenyi Biotec) and anti- IFN- γ (45- 15, 10 μg/ml, 1160 Miltenyi Biotec). Cells were withdrawn from coated plates after 24 h of activation, split 1161 after 3 days in a 1:3 ratio with fresh medium containing IL- 2 (10 ng/ml, R&D Systems), 1162 and analyzed on day 5 of culture. + +## RNA isolation and qRT-PCR + +To isolate RNA for qRT- PCR analysis or bulk RNA sequencing, \(10^{5} - 10^{6} \mathrm{T}\) cells were harvested, lysed in RA- 1 buffer (Macherey & Nagel). Total RNA was purified using the Nucleospin RNA XS Micro kit (Macherey & Nagel) according to manufacturer's instructions, without addition of carrier RNA. For qRT- PCR analysis, RNA was transcribed into cDNA utilizing Taqman Reverse Transcription Reagents (Applied Biosystems). cDNA was then subjected to qRT- PCR analysis using PowerUp SYBR Green or Taqman Fast Advanced Mastermix reagents (Applied Biosystems) and the following primers or probes: ExAB (forward: TGGTGAACAGTTCAGTTCTTCACAG, reverse: ACGATTTACTGCCCTCCGTAAC), Ex1a (forward: ATTCTGGTGGTTCGA AGCTGAGGA, reverse: ACGATTTACTGCCCTCCGTAAC), Ex1b (forward: ATGAG ACGAAGGAGCGCCAAG, reverse: ACGATTTACTGCCCTCCGTAAC), ST2 (forward d: AAGGCACACCATAAGGCTGA, reverse: TCGTAGAGCTTGGCCATCGTT), Hprt: (forward: CATAACCTGGTTCATCATCGc, reverse: TCCTCCTCAGACCGCTTTT), hIL1RL1- ExAB (forward: ATACAGACGTCAGAACTGAGAC, reverse: GCATACGAC AGTGAAGGTCA), hIL1RL1- Ex1a (forward: GAACTCAAGTACAACCCAATGA, reverse: GCATACGACAGTGAAGGTCa), IL1RL1: Hs00249384_m1 (Thermo Scientific) GAPDH: Hs02758991_g1 (Thermo Scientific). Amplifications were performed in triplicates by using a QuantStudio 7 device (Applied Biosystems) and + +<--- Page Split ---> + +expression levels were quantified with the \(\Delta \Delta \mathrm{Ct}\) - method by normalizing target gene expression to levels of Hprt (mouse) or GAPDH (human). + +## Single cell RNA-library preparation, sequencing, and analysis + +Single cell suspensions of \(\mathrm{CD90^{+}}\) \(\mathrm{CD8^{+}}\) \(\mathrm{CD44^{+}}\) T cells were obtained by flow- cytometrical sorting of CD19- depleted splenocytes from LCMV- infected wildtype and \(IIIrI1 - ExAB^{- / - }\) mice at day 7 p.i. with LCMV- WE (2x10 \(^6\) PFU). Sorted T cells of individual mice were barcoded using TotalSeq- C anti- mouse Hashtags (anti- mouse Hashtag 1, 2 and 3, all Biolegend). T cells of each genotype were then pooled and applied to the 10x Genomics workflow for cell capturing. For the preparation of scRNA gene expression (GEX), TCR and CiteSeq libraries the Chromium Next GEM Single Cell 5' Library & Gel Bead Kit v1.1 as well as the Chromium Single Cell 5' Feature Barcode Library Kit (10x Genomics) were used. After cDNA amplification the CiteSeq libraries were prepared separately using the Single Index Kit N Set A (10x Genomics). TCR target enrichment was performed using the Chromium Single Cell V(D)J Enrichment Kit for mouse T cells (10x Genomics). Final GEX and TCR libraries were obtained after fragmentation, adapter ligation and final Index PCR using the Single Index Kit T Set A. The Qubit dsDNA HS assay kit (Life technologies) was used for library quantification and fragment sizes were determined using the Fragment Analyzer with the NGS Fragment Kit (1- 6000bp) (Agilent). Sequencing was performed on a NextSeq2000 device (Illumina) using P2 Reagents v3 (200 cycles) with the recommended sequencing conditions for 5' GEX and barcode libraries (read1: 26nt, read2: 98nt, index1: 8nt, index2: n.a.) and on a NextSeq500 device (Illumina) using a Mid Output v2 Kit (300 cycles) for TCR libraries (read1: 150nt, read2: 150nt, index1: 8nt, index2: n.a., 20% PhiX spike- in). Raw data were processed using cellranger- 3.1.0 with refdata- gex- mm10- 2020- A and refdata- cellranger- vdj_GRCm38_alts_ensembl- mouse- 2.2.0 as reference. Mkfastq, count and vdj were used in default parameter settings with 3000 expected cells for demultiplexing, detection of intact cells, quantification of gene expression, antibody capture as well as assembly and quantification of T cell receptor sequences. + +The cellranger output was further analyzed in R using the Seurat package (version 4.0.0) \(^{86}\) . In particular, the CiteSeq data, hashtag sequences, detxramer counts of three individual \(IIIrI1 - ExAB^{- / - }\) and wildtype mice were imported and combined. Centered log ratio transformation was used for normalization. Seurat's default method was used for + +<--- Page Split ---> + +scaling. Features with correlation coefficients \(>0.85\) to Gm42418, Malat1, AY036118, and Lars2 were removed from the count matrix. Hashtag demultiplexing (representing the 3 biological replicates per genotype) was performed based on Seurats HTODemux with the parameter positive- quantile at 0.99. Doublets and untagged cells were filtered out. Cells with expression values for Cd8a or Cd8b1 and Cd3g, Cd3d, or Cd3e, with \(>200\) and \(< 4500\) features, and \(< 10\%\) UMI for mitochondrial genes were kept for further analysis. + +After ranking by residual variance, 3000 variable genes were determined. The genes encoding TCR variable regions (Trav, Trbv, Trdv, Trgv) were removed. 30 principal components (PC's) were computed and stored. Uniform Manifold Approximation and Projection (UMAP) and t- distributed stochastic neighbor embedding (TSNE) were run using the first 15 PC's. Transcriptionally similar clusters were identified using shared nearest neighbor modularity optimization, with a resolution of 0.35. For visualization, cells of the llrl1- ExAB- condition were down- sampled to match the number of cells in the WT condition. Signature genes were identified using the FindAllMarkers function in default parameter settings (only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25). Heatmaps and Dotplots for the single cell data were plotted with Seurat's DoHeatmap and DotPlot function, respectively, using default settings. Identified clusters were annotated based on the expression of key markers for CD8+ T cell subsets and cell functions. Two KIng1 expressing clusters were merged to the SLEC cluster. Further, two clusters were merged to form the "HighMito"- Cluster based on their expression of mitochondrial genes. After combining the stated clusters, signature genes were identified again using the same method as described above. Cluster size was defined as the number of cells in one cluster. Relative cluster sizes were calculated by analyzing the number of cells in one cluster per genotype divided by the total number of cells. For feature plots, the expression of single features was plotted on the UMAP by using Seurat's default function FeaturePlot with the option "keep.scale = 'all'". + +For the analysis of the TCR the immune profiles were integrated using identical cellular barcodes. For cells with more than one contig for the heavy or light TCR chain the most abundant, productive contig was chosen. Cell numbers were equalized by subsampling the larger condition. Cells without TCR annotation were excluded from the analysis. The R package immunarch (version 0.6.6) \(^{87}\) was used for clonality analysis after downsampling to the wildtype condition. + +<--- Page Split ---> + +## RNA sequencing + +Smarta and P14 T cells used for bulk RNA- Seq experiments were activated and differentiated as described above. To enrich for ST2- expressing T cells, CTLs, Th1 and Th2 cells were stained for ST2 at day 10 of culture and \(\mathsf{ST2^{+}}\) T cells were flowcytometrically sorted. For global analysis of IL- 33- responsive genes, T cells were harvested at day 10 of culture and rested for 3 days in the presence of IL- 2 (5 ng/ml) and IL- 7 (5 ng/ml, both Miltenyi Biotec), without irradiated splenocytes or cognate peptide. At day 13, IL- 12 (5 ng/ml) was added to CTLs and Th1 conditions to trigger ST2 expression. At day 14 of culture, T cells were subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich) and stimulated in conditioned medium with IL- 33 (10 ng/ml, R&D Systems) for 2 h. When used for RNA sequencing, quality of the isolated RNA was assessed with a Fragment Analyzer System (Agilent). All processed samples showed high RNA integrity (RQN \(>8\) ). cDNA libraries were prepared using the Smart- Seq v4 mRNA Ultra Low Input RNA Kit (Clontech) with up to 10 ng of RNA (IL- 33 stimulated T cells) or TrueSeq stranded total RNA library kit (Illumina) with up to 1 \(\mu \mathrm{g}\) of RNA (ST2- enriched T cells) according to manufacturer's instructions. Paired- end sequencing (2x75 bp) of cDNA libraries was performed on an Illumina NextSeq500 device using the NextSeq 500/550 High output Kit v2. Obtained reads were mapped to the mm10 or hg19 genome (annotation releases: GRCh37.p13 and GRCm38.p6) using Tophat288 and Bowtie289 with very- sensitive settings. Read counts were determined with featureCounts90. DESeq291 was used in RStudio for differential gene expression analysis. The DESeq2 count matrix was pre- filtered for genes with \(\geq\) 100 summarized read counts across the analyzed samples. A gene was considered as differentially expressed when \(\log 2\mathrm{FC} > 1.0\) and P adjusted \(< 0.01\) . AnnotationDbi92, EnhancedVolcano93, ComplexHeatmap94, heatmap95 and ggplot296 were used in RStudio for data visualization. + +For PCA and sample distance calculation, a blind variance stabilizing transformation (VST) was performed on the unnormalized counts across all samples. Sample distance plots are based on pairwise calculation of the Pearson correlation. + +## GSEA and Overrepresentation Analysis + +For gene set enrichment analysis (GSEA), the R package clusterProfiler97 was used. A gene list ranked by log2FoldChange containing all expressed genes served as input + +<--- Page Split ---> + +and was tested for enrichment of biological- process gene sets from the gene ontology resource98,99. + +For overrepresentation analysis, differentially expressed genes were split into up- and downregulated genes. Over- representation analysis was performed against biological- process gene sets from the gene ontology resource using a one- sided version of the Fisher's exact test. All expressed genes in the respective conditions were used as a background gene list. The results were simplified in order to reduce overlaps between ontology terms by using the clusterProfiler::simplifyGO function with a cutoff of 0.7. + +## Analysis of alternative transcription start sites + +Raw RNA- Seq reads of indicated T cell subsets were aligned using hisat2 (version 2.2.1)100 and assembled using Cufflinks (version 2.2.1)71 in RABT mode with EnsEMBL annotation release 67. Assemblies were then merged into a new reference annotation with the public reference using the cuffmerge function. The resulting annotation was used for an analysis with the R package ProActiv42. The getAlternativePromoters function was used with standard parameters except for minAbs = 5. Only results with FDR<0.01 were considered. + +## Processing of published Chip-Seq, ATAC-Seq data, and conservation data + +Fastq files of published RNA- Seq datasets (ILCS2: Run ID SRR7549295, mast cells: Run ID SRR6155875) were obtained from the NCBI Sequence read archive and aligned using hisat2 with default settings100. ChIP- and ATAC- Seq data sets were downloaded from the NCBI GEO Database, if required crossmapped to the mm10 genome using CrossMap101 (ChIP- Seq: GSM550303, GSM998272, GSM523226; ATAC- Seq: GSE120532, GSE111902). PhyloP conservation tracks were downloaded from the UCSC Sequence and Annotation database43,102. JASPAR motif analysis was performed on indicated sequences using the web- based JASPAR Scan tool73. FANTOM5 CAGE- Seq data and CAGE- associated transcript data were downloaded from the FANTOM5 data resource45,103,104. All NGS data tracks were visualized in the integrated genome viewer (IGV)105. + +## Quantification and statistical analysis + +Statistical analysis was performed using GraphPad Prism (v9.0.0). Normal distribution was tested using Shapiro- Wilk and Kolmogorov–Smirnov tests. Unpaired or paired + +<--- Page Split ---> + +1313 two- tailed Student's t- tests were used when two groups were compared with respect 1314 to one parameter. More than two groups were analyzed by One- way ANOVA with 1315 Tukey's post- hoc test for multiple comparison. For comparisons of more than one 1316 parameter between two or more groups, Two- way ANOVA with Tukey's post- hoc tests 1317 (unpaired samples) or Two- way repeated measurement ANOVA with Sidak's post- hoc 1318 tests (paired samples) were performed as indicated in the figure legends. + +<--- Page Split ---> + +# References Figure legends and Methods + +Trapnell, C. et al. Transcript assembly and quantification by RNA- Seq reveals unannotated transcripts and isoform switching during cell differentiation. Nat Biotechnol 28, 511- 515, doi:10.1038/nbt.1621 (2010). + +Pollard, K. S., Hubisz, M. J., Rosenbloom, K. R. & Siepel, A. Detection of nonneutral substitution rates on mammalian phylogenies. Genome Res 20, 110- 121, doi:10.1101/gr.097857.109 (2010). + +Fornes, O. et al. JASPAR 2020: update of the open- access database of transcription factor binding profiles. Nucleic Acids Res 48, D87- D92, doi:10.1093/nar/gkz1001 (2020). + +Ricardo- Gonzalez, R. R. et al. Tissue signals imprint ILC2 identity with anticipatory function. Nat Immunol 19, 1093- 1099, doi:10.1038/s41590- 018- 0201- 4 (2018). + +Gentek, R. et al. Hemogenic Endothelial Fate Mapping Reveals Dual Developmental Origin of Mast Cells. Immunity 48, 1160- 1171 e1165, doi:10.1016/j.immuni.2018.04.025 (2018). + +Pircher, H., Burki, K., Lang, R., Hengartner, H. & Zinkernagel, R. M. Tolerance induction in double specific T- cell receptor transgenic mice varies with antigen. Nature 342, 559- 561, doi:10.1038/342559a0 (1989). + +Oxenius, A., Bachmann, M. F., Zinkernagel, R. M. & Hengartner, H. Virus- specific MHC- class II- restricted TCR- transgenic mice: effects on humoral and cellular immune responses after viral infection. Eur J Immunol 28, 390- 400, doi:10.1002/(SICI)1521- 4141(199801)28:01<390::AID- IMMU390>3.0. CO;2- O (1998). + +Kaplan, M. H., Sun, Y. L., Hoey, T. & Grusby, M. J. Impaired IL- 12 responses and enhanced development of Th2 cells in Stat4- deficient mice. Nature 382, 174- 177, doi:10.1038/382174a0 (1996). + +Szabo, S. J. et al. Distinct effects of T- bet in TH1 lineage commitment and IFN- gamma production in CD4 and CD8 T cells. Science 295, 338- 342, doi:10.1126/science.1065543 (2002). + +Mombaerts, P. et al. Mutations in T- cell antigen receptor genes alpha and beta block thymocyte development at different stages. Nature 360, 225- 231, doi:10.1038/360225a0 (1992). + +Wefers, B., Bashir, S., Rossius, J., Wurst, W. & Kuhn, R. Gene editing in mouse zygotes using the CRISPR/Cas9 system. Methods 121- 122, 55- 67, doi:10.1016/j.ymeth.2017.02.008 (2017). + +Concordet, J. P. & Haeussler, M. CRISPR: intuitive guide selection for CRISPR/Cas9 genome editing experiments and screens. Nucleic Acids Res 46, W242- W245, doi:10.1093/nar/gky354 (2018). + +<--- Page Split ---> + +1358 83 Battegay, M. et al. Quantification of lymphocytic choriomeningitis virus with an immunological focus assay in 24- or 96- well plates. J Virol Methods 33, 191- 198, doi:10.1016/0166- 0934(91)90018- u (1991). + +1361 84 Stittrich, A. B. et al. The microRNA miR- 182 is induced by IL- 2 and promotes clonal expansion of activated helper T lymphocytes. Nat Immunol 11, 1057- 1062, doi:10.1038/ni.1945 (2010). + +1364 85 Haftmann, C. et al. miR- 148a is upregulated by Twist1 and T- bet and promotes Th1- cell survival by regulating the proapoptotic gene Bim. Eur J Immunol 45, 1192- 1205, doi:10.1002/ejj.201444633 (2015). + +1367 86 Hao, Y. et al. Integrated analysis of multimodal single- cell data. Cell 184, 3573- 3587 e3529, doi:10.1016/j.cell.2021.04.048 (2021). + +1369 87 Nazarov, V. I., Tsvetkov, V. O., Rumynskiy, E., Popov, A. A. & Balashov, I. immunarch: Bioinformatics Analysis of T- Cell and B- Cell Immune Repertoires. (2021). + +1372 88 Kim, D. et al. TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol 14, R36, doi:10.1186/gb- 2013- 14- 4- r36 (2013). + +1375 89 Langmead, B. & Salzberg, S. L. Fast gapped- read alignment with Bowtie 2. Nat Methods 9, 357- 359, doi:10.1038/nmeth.1923 (2012). + +1377 90 Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923- 930, doi:10.1093/bioinformatics/btf656 (2014). + +1380 91 Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 15, 550, doi:10.1186/s13059- 014- 0550- 8 (2014). + +1383 92 Pagès H, C. M., Falcon S, Li N AnnotationDbi: Manipulation of SQLite- based annotations in Bioconductor. (2020). + +1385 93 Blighe, K., Rana, S. & Lewis, M. EnhancedVolcano: Publication- ready volcano plots with enhanced colouring and labeling. (2021). + +1387 94 Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847- 2849, doi:10.1093/bioinformatics/btw313 (2016). + +1390 95 Kolde, R. Pretty Heatmaps. (2019). + +1391 96 Wickham, H. ggplot2: Elegant Graphics for Data Analysis. Springer- Verlag New York (2016). + +1393 97 Yu, G., Wang, L. G., Han, Y. & He, Q. Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16, 284- 287, doi:10.1089/omi.2011.0118 (2012). + +<--- Page Split ---> + +1396 98 Gene Ontology, C. The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 49, D325- D334, doi:10.1093/nar/gkaa1113 (2021). + +1398 99 Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25-29, doi:10.1038/75556 (2000). + +1400 100 Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol 37, 907-915, doi:10.1038/s41587-019-0201-4 (2019). + +1403 101 Zhao, H. et al. CrossMap: a versatile tool for coordinate conversion between genome assemblies. Bioinformatics 30, 1006-1007, doi:10.1093/bioinformatics/bt730 (2014). + +1406 102 Kent, W. J. et al. The human genome browser at UCSC. Genome Res 12, 996-1006, doi:10.1101/gr.229102 (2002). + +1408 103 Lizio, M. et al. Gateways to the FANTOM5 promoter level mammalian expression atlas. Genome Biol 16, 22, doi:10.1186/s13059-014-0560-6 (2015). + +1411 104 Lizio, M. et al. Update of the FANTOM web resource: expansion to provide additional transcriptome atlases. Nucleic Acids Res 47, D752-D758, doi:10.1093/nar/gky1099 (2019). + +1414 105 Robinson, J. T. et al. Integrative genomics viewer. Nat Biotechnol 29, 24-26, doi:10.1038/nbt.1754 (2011). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +20230124SupplementaryTable1scRNASeq.xlsx20230124SupplementaryTable2RNASeqIL33Stimulation.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12_det.mmd b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a5b3c0c144834a543b2393d78806c53f117a0332 --- /dev/null +++ b/preprint/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12/preprint__1445e32614f54f9a7288176ca18ac26ee6579e1e6ad6957074eada650ebe3b12_det.mmd @@ -0,0 +1,752 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 916, 175]]<|/det|> +# A type-1 immunity-restricted promoter of the IL-33 receptor gene directs antiviral T cell responses + +<|ref|>text<|/ref|><|det|>[[44, 195, 440, 215]]<|/det|> +Max Löhning ( max.loehning@charite.de ) + +<|ref|>text<|/ref|><|det|>[[55, 218, 800, 259]]<|/det|> +Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0001- 6382- 7281 + +<|ref|>text<|/ref|><|det|>[[44, 265, 808, 307]]<|/det|> +Tobias Brunner German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 5279- 2391 + +<|ref|>text<|/ref|><|det|>[[44, 312, 800, 375]]<|/det|> +Sebastian Serve Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 9839- 7119 + +<|ref|>text<|/ref|><|det|>[[44, 380, 760, 422]]<|/det|> +Anna- Friederike Marx Division of Experimental Virology, Department of Biomedicine, University of Basel + +<|ref|>text<|/ref|><|det|>[[44, 427, 812, 469]]<|/det|> +Maria Dzamukova German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 2389- 3222 + +<|ref|>text<|/ref|><|det|>[[44, 474, 800, 516]]<|/det|> +Nayar Duran Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) + +<|ref|>text<|/ref|><|det|>[[44, 520, 800, 584]]<|/det|> +Philippe Saikali Charité - Universitätsmedizin Berlin and German Rheumatism Research Center (DRFZ) https://orcid.org/0000- 0002- 4916- 7376 + +<|ref|>text<|/ref|><|det|>[[44, 589, 757, 631]]<|/det|> +Christoph Kommer Division of Theoretical Systems Biology, German Cancer Research Center (DKFZ) + +<|ref|>text<|/ref|><|det|>[[44, 635, 866, 678]]<|/det|> +Frederik Heinrich German Rheumatism Research Center (DRFZ) Berlin https://orcid.org/0000- 0001- 6097- 5422 + +<|ref|>text<|/ref|><|det|>[[44, 683, 389, 724]]<|/det|> +Pawel Durek German Rheumatism Research Centre + +<|ref|>text<|/ref|><|det|>[[44, 730, 816, 772]]<|/det|> +Gitta Heinz Deutsches Rheuma- Forschungszentrum Berlin https://orcid.org/0000- 0002- 9330- 3689 + +<|ref|>text<|/ref|><|det|>[[44, 777, 694, 818]]<|/det|> +Thomas Höfer German Cancer Research Center https://orcid.org/0000- 0003- 3560- 8780 + +<|ref|>text<|/ref|><|det|>[[44, 823, 816, 864]]<|/det|> +Mir- Farzin Mashreghi Deutsches Rheuma- Forschungszentrum Berlin https://orcid.org/0000- 0002- 8015- 6907 + +<|ref|>text<|/ref|><|det|>[[44, 869, 784, 911]]<|/det|> +Ralf Kuehn Max Delbrueck Center for Molecular Medic https://orcid.org/0000- 0003- 1694- 9803 + +<|ref|>text<|/ref|><|det|>[[44, 916, 576, 956]]<|/det|> +Daniel Pinschewer University of Basel https://orcid.org/0000- 0002- 1594- 3189 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 79, 101, 96]]<|/det|> +## Article + +<|ref|>sub_title<|/ref|><|det|>[[44, 116, 135, 135]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 154, 323, 174]]<|/det|> +Posted Date: February 1st, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 192, 473, 212]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1881814/v1 + +<|ref|>text<|/ref|><|det|>[[44, 229, 910, 272]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 290, 936, 356]]<|/det|> +Additional Declarations: Yes there is potential Competing Interest. D.D.P. is a founder, shareholder, and advisor to Hookiea Pharma Inc. commercializing arenavirus vectors. The remaining authors declare no competing interests. + +<|ref|>text<|/ref|><|det|>[[42, 390, 941, 434]]<|/det|> +Version of Record: A version of this preprint was published at Nature Immunology on January 3rd, 2024. See the published version at https://doi.org/10.1038/s41590-023-01697-6. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[135, 90, 861, 140]]<|/det|> +# A type-1 immunity-restricted promoter of the IL-33 receptor gene +directs antiviral T cell responses + +<|ref|>text<|/ref|><|det|>[[135, 170, 864, 262]]<|/det|> +Tobias M. Brunner¹,²,⁹, Sebastian Serve¹,²,³,⁹, Anna-Friederike Marx⁴, Maria +Dzamukova¹,², Nayar Duran¹,², Philippe Saikali¹,², Christoph Kommer⁵,⁶, Frederik +Heinrich⁷, Pawel Durek⁷, Gitta A. Heinz⁷, Thomas Höfer⁵,⁶, Mir-Farzin Mashreghi⁷, +Ralf Kühn⁸, Daniel D. Pinschewer⁴, and Max Löhning¹,²,⁴ + +<|ref|>text<|/ref|><|det|>[[67, 290, 864, 840]]<|/det|> +1 Experimental Immunology and Osteoarthritis Research, Department of +Rheumatology and Clinical Immunology, Charité – Universitätsmedizin Berlin, +corporate member of Freie Universität Berlin and Humboldt-Universität zu Berlin; +10117 Berlin, Germany. +2 Pitzer Laboratory of Osteoarthritis Research, German Rheumatism Research +Center (DRFZ), a Leibniz Institute; 10117 Berlin, Germany. +3 Berlin Institute of Health at Charité – Universitätsmedizin Berlin, BIH Biomedical +Innovation Academy; 10117 Berlin, Germany. +4 Division of Experimental Virology, Department of Biomedicine, University of Basel; +4009 Basel, Switzerland. +5 Division of Theoretical Systems Biology, German Cancer Research Center (DKFZ); +69120 Heidelberg, Germany. +6 BioQuant Center, University of Heidelberg; 69120 Heidelberg, Germany. +7 Therapeutic Gene Regulation, Regine von Ramin Lab Molecular Rheumatology, +German Rheumatism Research Center (DRFZ), a Leibniz Institute; 10117 Berlin, +Germany. +8 Max Delbrück Center for Molecular Medicine (MDC); 13125 Berlin, Germany. +9 These authors contributed equally to this work +*Corresponding authors. Email: tobias.brunner@drfz.de; max.loehning@charite.de; +loehning@drfz.de + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[67, 84, 204, 101]]<|/det|> +## 27 Abstract + +<|ref|>text<|/ref|><|det|>[[66, 113, 884, 455]]<|/det|> +The pleiotropic alarmin interleukin- 33 (IL- 33) drives type- 1, type- 2, and regulatory T- cell responses via its receptor ST2. Subset- specific differences in ST2 expression intensity and dynamics suggest that transcriptional regulation is key in orchestrating the context- dependent activity of IL- 33- ST2 signaling in T- cell immunity. Here, we identify a previously unrecognized alternative promoter in mice and humans that is located far upstream of the curated ST2- coding gene and drives ST2 expression in type- 1 immunity. Mice lacking this promoter exhibit a selective loss of ST2 expression in type- 1- but not type- 2- biased immune cells, resulting in impaired expansion of cytotoxic T cells (CTLs) and T- helper- 1 cells upon viral infection. T- cell- intrinsic IL- 33 signaling via type- 1 promoter- driven ST2 is critical to generate a clonally diverse population of antiviral short- lived effector CTLs. Thus, lineage- specific alternative promoter usage directs alarmin responsiveness in T- cell subsets and offers opportunities for immune cell- specific targeting of the IL- 33- ST2 axis in infections and inflammatory diseases. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 110, 883, 700]]<|/det|> +Endogenous danger- associated molecules released upon cellular damage, so- called alarms, act as central orchestrators of inflammation'. Amongst alarms, the IL- 1 cytokine family member IL- 33 stands out as a highly potent cytokine which can trigger both pro- and anti- inflammatory responses by engaging its unique receptor ST2 on adaptive or innate immune cells2,3. ST2, also known as T14,5, was first detected on the surface of T- helper- 2 (Th2) cells and mast cells6- 8, and crosslinking of ST2 or stimulation with IL- 33 was shown to elicit production of type- 2 cytokines, which suggested an important role in type- 2 immune responses9- 11. In line with this, inhibition of IL- 33- ST2 signaling by antibody blockade or by disruption of either the /33 gene or the ST2- coding gene interleukin- 1 receptor- like 1 (Il1rl1) ameliorated type- 2 airway inflammation in mice and impaired immunity against parasitic nematode infections6,12- 14. Additionally, more recent studies established ST2 as a marker for type- 2 innate lymphoid cells (ILC2s) and demonstrated a critical role of IL- 33 for their development and function15,16. By activating ILC2s and a subset of highly suppressive, ST2- expressing regulatory T cells (Tregs), IL- 33 regulates tissue homeostasis, promotes wound healing, and mitigates pathology in acute or chronic inflammation3,17- 22. Beyond its role as a potentiator of Th2 and Treg activity, IL- 33 has also emerged as a key driver of type- 1 immune responses. It is released by fibroblastic reticular cells in lymphoid organs and promotes clonal expansion and activation of antiviral CTLs and T- helper- 1 (Th1) cells to confer protection against replicating viruses at the expense of potential tissue damage23- 27. Moreover, IL- 33- mediated amplification of type- 1 immune cells was shown to exacerbate tissue damage during Graft- versus- Host- Disease (GvHD)28,29 and to contribute to immune dysregulation in systemic inflammatory diseases30. + +<|ref|>text<|/ref|><|det|>[[115, 704, 883, 896]]<|/det|> +Considering its multifaceted mode- of- action, IL- 33 is now recognized as an alarmin which amplifies the activity of distinct T cell subsets with either pro- or antiinflammatory properties in a context- specific manner31,32. To accomplish this versatility, it was suggested that ST2 expression is regulated in a cell- type specific manner, such that certain T cell subsets may become sensitive to IL- 33 signals dependent on the inflammatory environment and/or on costimulatory signals31,32. ST2 is absent from naive T cells but expressed at high levels in a constitutive manner by type- 2 biased immune cells, including memory Th2 cells and a Treg subset6,21. In + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 81, 884, 547]]<|/det|> +75 these type- 2 biased cells, ST2 expression depends on the master- regulator 76 transcription factor of type- 2 immunity GATA- 321,33. In contrast, antiviral CTLs and Th1 77 cells express ST2 transiently upon infection, and this expression depends on STAT4 78 and T- bet, the key transcription factors of type- 1 immunity24,27. This highly dynamic 79 expression pattern not only contributed to the early assumption that ST2 is 80 preferentially expressed in type- 2 immunity, but it also renders it significantly more 81 difficult to study ST2 on type- 1 immune cells in vivo. Consequently, the molecular 82 mechanism allowing for T cell lineage- specific ST2 expression patterns has remained 83 enigmatic. 84 Here, we report on the identification of a type- 1 immunity- restricted promoter located 85 far upstream of the curated Il1rl1 gene. This previously unrecognized promoter is 86 bound by STAT4 and T- bet, is subject to epigenetic remodeling during type- 1 T cell 87 differentiation, and drives ST2 expression by CTLs and Th1 cells both in mice and 88 humans. Its deletion in mice resulted in a selective impairment of ST2 expression by 89 CTLs and Th1 cells, while ST2 expression on type- 2 immune cells and Tregs was 90 preserved. Furthermore, using lymphocytic choriomeningitis virus (LCMV) infection as 91 a model, we demonstrate that the type- 1 immunity- restricted promoter is essential for 92 the expansion of antiviral T cells and promotes the formation of clonally diverse 93 populations of short- lived effector CTLs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 90, 194, 107]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[117, 123, 694, 143]]<|/det|> +## Identification of a type-1 immunity-restricted ll1r11 promoter + +<|ref|>text<|/ref|><|det|>[[110, 144, 883, 900]]<|/det|> +Previous studies have demonstrated that the protein- coding exons of the ST2- encoding IL- 1 receptor- like 1 (ll1r11) gene are preceded by two non- coding exons (exon 1a and exon 1b) located in a distal and proximal promoter region, respectively4,34 (Fig. 1a). The proximal promoter drives ST2 expression in fibroblasts, whereas the distal promoter was shown to mediate ST2 expression in Th2 cells and mast cells33,35. To assess which transcriptional start site (TSS) is used by type- 1- polarized T cells, we differentiated naive T cells to CTLs, Th1, or Th2 cells in vitro, which triggers substantial ST2 expression in all subsets (Fig. 1b and Extended Data Fig. 1a- c). By analyzing the incorporation of leader exons into the 5' untranslated regions (UTR) of ST2- coding transcripts, we found that none of the described promoters could possibly account for the expression of ST2 by type- 1- polarized T cells (Fig. 1c). Thus, to map the origin of ll1r11 transcripts in these cells, we next subjected flow- cytometrically sorted ST2- expressing CTLs, Th1, and Th2 cells to RNA- sequencing (RNA- Seq) analysis with high read depth to achieve sufficient read coverage. By in silico alignment, we discovered a TSS located \(\sim 40\) kb upstream of the annotated ll1r11 gene, which was selectively transcribed in CTLs and Th1 cells (type- 1 promoter) but not detectable in Th2 cells (Fig. 1d). The identified TSS gave rise to two ll1r11 transcript isoforms with distinct leader sequences but unaltered protein- coding sequences. We here refer to the exons incorporated in these transcripts as exon A, B, and D. In addition, alternative splicing of exon B to exon C resulted in a transcript that did not contain ST2 protein- coding exons. As expected, the known distal promoter (exon 1a) presented the primary origin of ll1r11 transcripts in Th2 cells (type- 2 promoter). To further confirm that the identified type- 1 promoter is active in CTLs and Th1 cells in vivo, we next transferred naive LCMV- specific T cell receptor (TCR) transgenic \(\mathrm{CD4^{+}}\) (Smarta) or \(\mathrm{CD8^{+}}\) (P14) T cells into wildtype (WT) mice and infected the recipients with LCMV. At day 7 post infection (d7 p.i.), when T cells express high amounts of ST224,27, we reisolated the transferred T cells and quantified the ll1r11 promoter usage. In contrast to Th2 cells, CTLs and Th1 cells had largely incorporated exons A and B but not exon 1a into the 5'UTR of ST2- coding transcripts (Fig. 1e,f). Corroborating previous reports24,27, ST2 expression by CTLs and Th1 cells, but not by Th2 cells or regulatory T cells (Tregs), relied on IL- 12 signals and the transcription + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 883, 304]]<|/det|> +factors T- bet and STAT4 (Extended Data Fig. 1d- h). Hence, we analyzed T- bet- and STAT4- binding as well as activation- induced changes in chromatin accessibility at the llrl1 locus in type- 1- polarized T cells using publicly available chromatin immunoprecipitation sequencing (ChIP- Seq)36- 38 and assay for transposase- accessible chromatin sequencing (ATAC- Seq)39,40 data. As reported previously, GATA- 3, the key transcription factor of type- 2 immune cells, binds predominantly upstream of the distal promoter (Fig. 1g)33. In contrast, T- bet and STAT4 binding was detected in the vicinity of exons A and B, at sites that were inaccessible in naive \(\mathrm{CD4^{+}}\) and \(\mathrm{CD8^{+}}\) T cells but accessible in LCMV- primed Th1 cells and CTLs (Fig. 1g). + +<|ref|>text<|/ref|><|det|>[[115, 306, 883, 520]]<|/det|> +Alternative promoters are abundant41. We were unaware, however, of any other gene with a comparably selective alternative promoter usage in type- 1- and type- 2- polarized immune cells. We thus utilized RNA- Seq data and established computational methods42 to conduct a genome- wide search aimed at identifying additional genes with type- 1 or type- 2 immunity- specific promoters. llrl1 was reliably detected when comparing Th2 cells to CTLs and Th1 cells but, to our surprise, no other gene with lineage- specific TSS usage was found by this approach (Fig. 1h- j). These data suggested that type- 1 and type- 2 T cell lineage- specific utilization of alternative first exons is of rare occurrence, with the llrl1 gene potentially representing a unique case. + +<|ref|>sub_title<|/ref|><|det|>[[117, 540, 760, 559]]<|/det|> +## The type-1 llrl1 promoter is conserved between mice and humans + +<|ref|>text<|/ref|><|det|>[[115, 564, 883, 805]]<|/det|> +To analyze the type- 1 promoter region in more detail we assessed DNA sequence conservation between vertebrate species using a precomputed conservation track 43 and compared it to STAT4- as well as to T- bet- ChIP- Seq and ATAC- Seq data (Fig. 2a). Additionally, we calculated conservation scores using the alignment tool VISTA44. Thereby, we identified a 275nt- spanning, well conserved sequence located \(\sim 5\) kb upstream of exon A (CNS- 5), which is bound by T- bet and STAT4 in Th1 cells and is marked by a sharp ATAC- Seq peak in CTLs (Fig. 2a). Mapping of T- bet and STAT4 binding motifs within CNS- 5 indicated that both transcription factors putatively bind closely to each other at sequences almost identical between mice and humans (Fig. 2b). + +<|ref|>text<|/ref|><|det|>[[115, 810, 881, 903]]<|/det|> +To assess whether T cell lineage- specific promoter usage at the llrl1 locus is indeed conserved between mice and humans, we first examined cap analysis of gene expression (CAGE) and associated transcriptomic data from the FANTOM5 resource41,45. Data from a human NK leukemia- derived cell line (KHYG- 1) provided + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 348]]<|/det|> +evidence for a putative TSS upstream of the annotated IL1RL1 gene. In human Th1 cells this site is preceded by T- bet binding sites (Fig. 2c)46. Of note, the exon structure closely resembles the exons A and B we identified in mice (cf. Fig. 1d). To determine whether this TSS is utilized in primary human CTLs and Th1 cells, we isolated in vivo- differentiated T cells from peripheral blood and reactivated them ex vivo to quantify IL1RL1 promoter usage (Extended Data Fig. 2a- f). Congruently with mouse T cells, ST2 transcripts of human CTLs and Th1 cells contained exons A and B within their 5'UTRs while Th2 cells had incorporated exon 1a (Fig. 2d). In summary, we have identified a previously unrecognized type- 1 immunity- restricted Il1rl1 promoter, which is instructed by the lineage- associated transcription factors T- bet and STAT4 and orchestrates ST2 expression in CTLs and Th1 cells of humans and mice. + +<|ref|>text<|/ref|><|det|>[[112, 365, 883, 905]]<|/det|> +Alternative Il1rl1 promoters allow lineage- specific targeting of ST2 expression Modulation of the IL- 33- ST2 axis could represent a promising approach in treating inflammatory diseases. However, due to hard- to- predict effects on the balance between IL- 33- mediated inflammation and tissue repair, blockade of IL- 33 or ST2 using therapeutic antibodies has shown conflicting results in preclinical disease models47. Fine- tuned treatment approaches might therefore offer a critical advantage. We thus asked whether lineage- specific alternative promoters can be leveraged to target ST2 expression in a T cell subset- specific manner. To this end, primary T cells were activated and retrovirally transduced to express small- hairpin RNAs (shRNAs) targeting 5' UTRs of lineage- specific transcript isoforms (Fig. 3a). Selective downregulation of ST2 was achieved in CTLs and Th1 cells using shRNAs binding exons A and B, and in Th2 cells with shRNAs directed against exon 1a (Fig. 3b,c). Furthermore, to study the role of the type- 1 immunity- restricted promoter in vivo, we used CRISPR/Cas9 in mouse zygotes to generate a novel knockout mouse strain with a deletion comprising the noncoding exons A and B as well as the surrounding promoter region (Il1rl1- ExAB- ). A second mouse strain lacking exon C (Il1rl1- ExC- ) was generated to control for potential effects of exon C- containing transcripts (Fig. 3d and Extended Data Fig. 3a,b), which we expected to be affected in Il1rl1- ExAB- mice. In steady state both strains of knock- out mice harbored normal numbers of T cells at an activation state comparable to WT mice, both in spleen and lymph nodes (Extended Data Fig. 3c- h). Likewise, the introduced deletions did not affect the composition of splenic innate immune cells (Extended Data Fig. 3i- k). First, to assess ST2 expression + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 883, 722]]<|/det|> +by //1r/1- ExAB- and //1r/1- ExC- T cells, naive T cells were purified from the respective mice and activated in vitro. Deletion of the regulatory region surrounding exons A and B severely impaired STAT4- dependent transcription of the //1r/1 gene as evident from the absence of ST2 expression by //1r/1- ExAB- CTLs and Th1 cells (Fig. 3e,f). Importantly, //1r/1- ExAB- Th2 cells differentiated from the same pool of naive CD4+ T cells exhibited normal ST2 expression (Fig. 3e,f). As expected, ST2 expression by //1r/1- ExC- T cells was not reduced (Fig. 3e,f). Due to a lack of ST2 expression, CTLs as well as Th1- polarized CD4 T cells of //1r/1- ExAB- T cells were unresponsive to IL- 33 stimulation whereas Th2- polarized cells of the same mice expressed ST2 and responded normally (Fig. 3g, cf. Fig. 3e,f). To verify that ST2 expression is preserved on type- 2- biased immune cells of //1r/1- ExAB- mice in vivo, we analyzed ST2 expression on peritoneal mast cells and on ILC2s. Both cell types utilized the GATA- 3- regulated type- 2 //1r/1 promoter (Extended Data Fig. 4a). Like in Th2 cells, absence of the type- 1 promoter in //1r/1- ExAB- mice had no effect on ST2 expression by mast cells or ILC2s (Extended Data Fig. 4b- g). To investigate the requirement for type- 1 promoter- driven transcription for ST2 expression by CTLs and Th1 cells responding to viral challenge in vivo we infected WT, //1r/1- , //1r/1- ExAB- , and //1r/1- ExC- mice with LCMV (Fig. 3h). Consistent with the above in vitro results, ST2 surface expression was almost absent from CTLs in //1r/1- ExAB- mice and was significantly reduced in T- bet+ FoxP3- Th1 cells at day 7 p.i., while T- bet FoxP3+ Tregs displayed unaltered ST2 surface levels (Fig. 3l- n). Conversely, //1r/1- ExC- CTLs and Th1 cells exhibited slightly enhanced ST2 expression (Fig. 3l- n). This observation was in line with the RNA- Seq data showing that transcripts spliced from exon B to exon C are not further spliced to ST2- coding exons and exon C may thus act as a transcriptional decoy (cf. Fig. 1d). Altogether, we found that the targeting of individual //1r/1 promoters allowed for a selective T cell lineage- specific manipulation of ST2 expression. + +<|ref|>sub_title<|/ref|><|det|>[[115, 737, 680, 756]]<|/det|> +## The type-1 //1r/1 promoter drives antiviral T cell responses + +<|ref|>text<|/ref|><|det|>[[115, 760, 883, 903]]<|/det|> +Next, we studied whether CD8+ T cell responses to LCMV infection required the type- 1 //1r/1 promoter. At the peak of the response (d7 p.i.), //1r/1- ExAB- promoter knock- out mice harbored substantially reduced numbers of CTLs in spleens and livers, resembling in its extent the impairment observed in //1r/1- gene knock- out mice (Fig. 4a- c and Extended Data Fig. 5a- c). Diminished CTL counts were largely accounted for by a reduction in activated CD44+CD62L- effector T cells expressing the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 882, 277]]<|/det|> +proliferation marker Ki67 (Fig. 4d- f). Accordingly, CTLs specific for the immunodominant \(\mathsf{GP}_{33 - 41}\) and \(\mathsf{NP}_{396 - 404}\) epitopes of LCMV, respectively, were substantially reduced (Fig. 4g and Extended Data Fig. 5d- f). Ultimately, ExonAB- deficient mice displayed significantly lower numbers of CTLs expressing effector cytokines or cytolytic molecules, and systemic IFN- \(\gamma\) levels were reduced by \(>70\%\) (Fig. 4h,i and Extended Data Fig. 5g- i). In contrast and as expected, CTL responses of \(I11rl1 - ExC^{- / - }\) mice were comparable to WT mice in all of the aforementioned aspects (Fig. 4b- i). + +<|ref|>text<|/ref|><|det|>[[112, 280, 883, 744]]<|/det|> +To determine whether the observed effects were due to a T cell- intrinsic impairment in ST2 expression, mixed bone marrow chimeras were generated by reconstituting irradiated WT mice with bone marrow from \(I11rl1^{- / - }\) , \(I11rl1 - ExAB^{- / - }\) , or \(I11rl1 - ExC^{- / - }\) mice, each of them mixed 1:1 with WT bone marrow. Following LCMV infection, WT: \(I11rl1 - ExC^{- / - }\) chimeras mounted CTL responses that derived at approximately equal parts from both bone marrow compartments. In contrast, WT bone marrow- derived CTLs significantly outnumbered the CTLs derived from \(I11rl1^{- / - }\) or \(I11rl1 - ExAB^{- / - }\) bone marrow in the respective chimeras (Fig. 4j- m and Extended Data Fig. 5j- k). Lastly, to study the impact of \(I11rl1 - ExAB\) - driven ST2 expression on T cell responses in the absence of any potentially confounding irradiation effects on recipient mice, \(I11rl1 - ExAB^{- / - }\) mice were crossed to LCMV TCR- transgenic P14 mice. Congenically marked naive \(I11rl1 - ExAB^{- / - }\) P14 CTLs and WT P14 cells were cotransferred into recipients, which were subsequently infected with LCMV and analyzed at d10 p.i.. Analogously to the data obtained in mixed bone marrow- chimeric mice, \(I11rl1 - ExAB^{- / - }\) and \(I11rl1^{- / - }\) P14 T cells expanded significantly less than their respective cotransferred WT P14 T cell populations (Fig. 4n- r). Similarly, albeit less pronounced, \(I11rl1 - ExAB^{- / - }\) LCMV- TCR transgenic \(\mathrm{CD4^{+}}\) Smarta T cells expanded less than cotransferred WT Smarta cells (Fig. 4s- w and Extended Data Fig. 5l- p). Thus, optimal expansion of antiviral T cells critically depends on cell- intrinsic activity of the type- 1 \(I11rl1\) promoter. + +<|ref|>sub_title<|/ref|><|det|>[[115, 763, 880, 806]]<|/det|> +## The type-1 \(I11rl1\) promoter is critical to establish a clonally diverse population of short-lived effector CTLs + +<|ref|>text<|/ref|><|det|>[[115, 812, 881, 905]]<|/det|> +At the peak of the acute response, the population of antiviral CTLs is heterogeneous and comprises many functionally distinct subsets48. To delineate the impact of type- 1 promoter- driven ST2 expression on CTL differentiation we sorted activated \(\mathrm{CD44^{+}}\) \(\mathrm{CD8^{+}}\) effector T cells from LCMV- infected WT or \(I11rl1 - ExAB^{- / - }\) mice and subjected + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 80, 883, 920]]<|/det|> +them to combined single- cell gene expression and TCR repertoire analysis. T cells were clustered into separate populations using nearest neighbor modularity optimization, displayed in uniform manifold approximation projections (UMAP), and annotated based on signature gene expression (Fig. 5a,b, Extended Data Fig. 6a,b, and Supplementary Table 1). Analyzing CTLs from both genotypes, six clusters were identified. As expected, the two dominant ones showed expression of genes associated with a short- lived effector cell (SLEC; Klrg1, Gzma, Id2) or memory precursor effector cell phenotype (MPEC; Il7r, Sell, Ccr7) (Fig. 5c). The third cluster was enriched in CTLs expressing Pdcd1 (encoding PD- 1) and Lag3, two markers of exhausted CTLs49, while cells of the fourth cluster exhibited higher expression of Tcf7 (encoding TCF- 1) and Id3, thus likely presenting stem- like precursors of effector CTLs50,51. Lastly, the two remaining clusters were enriched in CTLs expressing higher levels of mitochondrial genes or cell cycle- related markers (Mki67, Top2a). A comparison of CTL cluster composition in the two genotypes revealed that type- 1 Il1rl1 promoter disruption resulted in a particularly pronounced curtailment of SLEC- differentiated cells (Fig. 5d). This translated into a 90- 95% reduction of KLRG1+CD127- CTL numbers in Il1rl1- ExAB-/- mice and mirrored the phenotype of Il1rl1-/- mice (Fig. 5e- g). Further, a lower proportion of Il1rl1- ExAB-/- CTLs exhibited surface expression of the SLEC- associated molecule CXCR3 (Fig. 5h)52. Moreover, Il1rl1- ExAB-/- CD8+ T cells in mixed bone marrow chimeras evidenced a pronounced defect in SLEC generation, similar to the behavior of Il1rl1-/- CD8+ T cells (Fig. 5i). In contrast, Il1rl1- ExC-/- bone marrow- derived CD8+ T cells were as proficient as WT cells in populating the SLEC compartment (Fig. 5i). Lastly, the proportion of ExonAB- deficient P14 T cells differentiating into SLECs and/or CXCR3+ cells was reduced as compared to adaptively cotransferred WT P14 cells (Fig. 5j- l). Altogether these results demonstrated that impaired SLEC formation by ExonAB- deficient CD8+ T cells was due to a cell- intrinsic defect in ST2 signaling. Next, we asked whether type- 1 promoter- driven expression of ST2 drives the selective proliferation of SLEC- differentiated T cell clones or whether it enforces the differentiation of precursors into SLECs. Thus, we first assessed ST2 expression on LCMV- specific CD8+ T cells of WT mice where we found it on both KLRG1+ and KLRG1- cells (Fig. 5m), suggesting IL- 33 signaling can occur prior to SLEC differentiation. Further, we integrated single- cell gene expression data with a TCR repertoire analysis. Interestingly, despite an eight- fold difference in CD44+ CTL cell + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 448]]<|/det|> +counts per spleen (Fig. 5n), equivalent numbers of clonotypes were identified in Il1rl1- ExAB- and WT mice when equal numbers of CD44+ CTLs were compared (Fig. 5o). This suggested that during the acute phase of infection, IL- 33 expands the pool of activated T cells in a clonotype- unselective manner, which does not significantly alter the TCR diversity, at least not amongst the most abundant clonotypes. Of note, the majority of clonotypes identified were represented less than three times per mouse and no clonotype was found more often than seven times (Fig. 5p). In line with previous reports, this indicated that TCR diversity within the CTL population was high during the acute phase of infection, without any evidence for clearly dominant clones53. By consequence, the severe reduction in SLEC numbers in Il1rl1- ExAB- mice resulted in substantially reduced SLEC clonotype numbers (Fig. 5q,r). Together, these findings demonstrate that in the absence of type- 1 promoter- driven ST2 expression, most CTL clones achieve basal activation, but fail to develop into fully differentiated effector cells. Thus, type- 1 promoter- driven ST2 expression is vital to establish a numerically significant and clonally diverse population of short- lived antiviral effector CTLs. + +<|ref|>sub_title<|/ref|><|det|>[[117, 466, 880, 510]]<|/det|> +## Transcriptional profiling indicates broad costimulatory and TCR-cooperative functions of IL-33- ST2 signaling + +<|ref|>text<|/ref|><|det|>[[115, 515, 882, 904]]<|/det|> +To gain further mechanistic insight on how IL- 33 signaling modulates T cell activation and differentiation, we additionally performed a comprehensive analysis of early IL- 33 target genes in CTLs as well as Th1 and Th2 cells by RNA- Seq profiling. To this end, naive T cells were differentiated, followed by a resting period without antigenic stimulation. Since ST2 signaling is subject to negative feedback mechanisms and also because oxidation of IL- 33 rapidly reduces its activity54,55, gene expression was analyzed before stimulation (0 hours) and 2 hours after treatment with or without IL- 33 (Fig. 6a). In the absence of IL- 33 stimulation the gene expression profiles at 0 hours and at 2 hours were similar. In contrast, short- term stimulation with IL- 33 had a profound effect on the transcriptome of CTLs, Th1, and Th2 cells, and it strongly induced or, in fewer cases, reduced expression of target genes rather than preventing a loss or gain of transcription (Fig. 6b). Whereas a substantial portion of differentially regulated genes was shared between the T cell subsets analyzed, others were regulated in a lineage- specific manner (Fig. 6c). GO- enrichment analysis revealed a broad role of IL- 33- responsive genes in T cell activation, proliferation, and differentiation (Fig. 6d). Importantly, IL- 33 stimulation of CTLs amplified expression of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 883, 275]]<|/det|> +Tbx21, Zeb2, and Prdm1, genes of transcription factors known as key drivers of SLEC differentiation56- 58 (Fig. 6b,e). Across the three T cell subsets we observed a prominent upregulation of genes frequently used as indicators of recent TCR activation (Nr4a1, Cd69, Batf) (Fig. 6b,e)59- 61. Coherently, gene set enrichment analysis showed a significant overlap between IL- 33- and TCR- downstream signaling in CTLs (Fig. 6f and Supplementary Table 2). This suggested that IL- 33, which is released by fibroblastic reticular cells early during infection26, may support TCR stimulation to promote potent antiviral CTL responses. + +<|ref|>text<|/ref|><|det|>[[113, 279, 883, 728]]<|/det|> +Thus, to further investigate the interplay between ST2 and TCR signaling strength, we made use of a genetically engineered LCM virus that differs from the wildtype counterpart only by a GP- A39C mutation, rendering its GP33 epitope (GP33- C6: KAVYNFCTC instead of KAVYNFATC) a weak P14 TCR agonist62. P14 and Il1rl1- /- P14 cells were cotransferred into wildtype recipients, which were subsequently infected with LCMV expressing either the high- or the low- affinity GP33 variant (Fig. 6g). We found that ST2- sufficient as well as ST2- deficient P14 cells expanded less when primed with the low- affinity ligand (Fig. 6h). Further, P14 cells depended on ST2 for optimal expansion and effector differentiation, irrespective of the TCR stimulation strength (Fig. 6i,j). Interestingly, the response of WT P14 cells responding to low- affinity virus was comparable to or exceeded the one of high- affinity ligand- primed Il1rl1- /- P14 cells in terms of total and effector cell progeny, respectively (Fig. 6h,j). This suggested that ST2 signals can help reaching effector T cell responses of critical size even when confronted with low- affinity ligands. Of note, the impairment in expansion and effector differentiation between mice infected with high- or low- affinity GP33- expressing LCMV were unlikely due to any potential differences in inflammation or IL- 33 release, as the endogenous NP396- specific T cell responses to the two LCMV variants were indistinguishable (Fig. 6k,l). + +<|ref|>text<|/ref|><|det|>[[115, 722, 881, 791]]<|/det|> +In summary, these data demonstrate that IL- 33- ST2 signaling provides a strong costimulatory signal that can act cooperatively with TCR signaling to promote the expansion and effector cell differentiation of antiviral CTLs. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 230, 101]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[111, 100, 883, 895]]<|/det|> +IL- 33 alarm signaling has long been recognized as an important driver of type- 2 immune responses2,3,6,7,12,31,32. Over the past decade its important role in promoting type- 1 immunity has become widely accepted yet remains mechanistically less well understood, particularly due to a lack of understanding how IL- 33 receptor expression is regulated in type- 1 immune cells31,32. Here, we studied the transcriptional regulation of ST2 expression in antiviral T cells and thereby discovered a dedicated type- 1 immunity- restricted promoter located \(\sim 40\) kb upstream of the curated I/1r/1 gene in mice and humans. This type- 1 promoter drives IL- 33 receptor expression by CTLs and Th1 cells in vitro and in viral infections in vivo. As opposed to the previously described "distal" type- 2 promoter, which is regulated by GATA- 333 and is utilized by type- 2 immune cells as well as Tregs21, the identified promoter is controlled by the type- 1 immunity- associated transcription factors T- bet and STAT4 and is subject to epigenetic remodeling during type- 1 T cell differentiation. Thus, we provide evidence for a dedicated regulatory genetic element to control ST2 expression selectively in type- 1 immune cells. Interestingly, this finding explains why transcription of I/1r/1 seems to be coregulated with its neighboring type- 1 immunity- associated gene I/18r63. Although the I/1r/1 gene has been studied intensively8,35,64, the type- 1 promoter has remained unrecognized, likely because it is only transiently active, often resulting in a low abundance of ST2- coding transcripts, depending on the time- point of analysis24,27. The latter renders it difficult to obtain adequate read coverage across splice junctions for a clear definition of exon structures by commonly used RNA- Seq techniques65 – a challenge we approached initially by analyzing T cells that have been differentiated in vitro to express a high amount of I/1r/1 transcripts. In knock- down and knock- out experiments we have subsequently validated the crucial role of this regulatory element in vivo, and by analyzing human T cells have extended the concept to our species. Above all, our finding was surprising, as to the best of our knowledge no other gene has been identified to date, for which type- 1 and type- 2 immune cells exhibit a similarly distinct lineage- specific promoter usage. Consistently, our own attempts at identifying further genes with an analogous dual promoter usage were unsuccessful. We acknowledge that technical limitations might have prevented us from identifying such genes, thus rendering it impossible to firmly conclude that the I/1r/1 gene is the only gene regulated by such highly selective, type- 1 and type- 2 restricted promoters. Still, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 127]]<|/det|> +our results suggest that this is not a common feature but represents a fairly unique mechanism to spatiotemporally orchestrate ST2 expression. + +<|ref|>text<|/ref|><|det|>[[115, 130, 883, 350]]<|/det|> +IL- 33 is an exceptionally potent alarmin, which can act as a pro- or anti- inflammatory cytokine, depending on the local composition of immune cells and their responsiveness to IL- 33 at the timing of release32. Likely due to its potential to cause severe inflammation, IL- 33 responsiveness requires stringent regulation. Transcription from the type- 1 immunity- restricted promoter enables transient ST2 expression by CTLs and Th1 cells in response to inflammatory stimuli23,24,27, which may serve to prevent continuous activation of cells with a high tissue- destructive potential. In contrast, constitutive type- 2 promoter- driven ST2 expression on Tregs and ILC2s allows for rapid anti- inflammatory responses to tissue damage19- 21,32. + +<|ref|>text<|/ref|><|det|>[[115, 352, 883, 917]]<|/det|> +Importantly, this dual mode- of- action constitutes a major hurdle for the therapeutic modulation of IL- 33–ST2 signaling32,66. We here demonstrate that the usage of distinct promoters offers opportunities for a T cell subset- specific targeting of ST2 expression at the transcriptional level by shRNA knock- down and at the genomic level by CRISPR/Cas9- mediated deletion of the type- 1 Il1rl1 promoter. ExonAB- deficient mice exhibit a selective, type- 1 immunity- restricted impairment of ST2 expression and display curtailed CD8+ and CD4+ T cell responses against LCMV, while ST2 expression by Tregs and type- 2 immune cells was fully preserved. Of note, the effects of IL- 33 on CTL expansion and differentiation were almost exclusively dependent on the type- 1 promoter, while its deletion did not fully abrogate ST2 expression by Th1 cells. Nevertheless, the type- 1 promoter was critical for optimal expansion of antiviral Th1 cells. T cell subset- specific targeting approaches could be of interest to modulate IL- 33 responses in inflammatory diseases. For instance, IL- 33 administration was shown to drive Treg expansion in the context of GvHD, promoting tolerance induction and disease amelioration67- 69. However, IL- 33 also augments type- 1 alloimmunity by acting as a costimulatory molecule for donor CTLs and Th1 cells28,29. A targeted disruption of ST2 selectively on type- 1 immune cells might minimize the pathological response during GvHD while the protective effects of IL- 33 should remain preserved. Our study provides mechanistic insight into the role of type- 1 promoter- driven ST2 expression and IL- 33 signaling in CD8+ T cell differentiation. Single- cell gene expression analysis of antiviral CTLs revealed that in the absence of the type- 1 ST2 promoter, mice display a particularly pronounced reduction in antiviral SLECs. So far, it remained unaddressed whether IL- 33 merely enforces proliferation of effector CTLs, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 882, 373]]<|/det|> +or if IL- 33 sensing actively promotes the differentiation of SLECs. Clonotype diversity in the SLEC population was high in wildtype mice and diminished proportionally to the cell counts in ExonAB- deficient mice. This suggests that IL- 33 can indeed foster the transition of an activated CTL into a cell with potent effector functions rather than expanding a pool of pre- differentiated SLECs. Terminal differentiation has been associated with STAT4 signaling and with high levels of T- bet, Blimp- 1, and Zeb256- 58. Likewise, the activity of the type- 1 ST2 promoter is positively regulated by STAT4 and T- bet. Interestingly, in- depth RNA- Seq profiling of IL- 33 target genes in CTLs demonstrated an induction of Tbx21 (T- bet), Prdm1 (Blimp- 1), and Zeb2 transcript expression shortly upon stimulation, thus inferring a positive feedback loop that further reinforces both type- 1 promoter- driven ST2 expression and effector differentiation via T- bet. + +<|ref|>text<|/ref|><|det|>[[115, 378, 882, 618]]<|/det|> +Besides these cell- intrinsic factors, extrinsic cues via the TCR determine the fate of CTLs. Above- threshold TCR signaling strength is directly linked to asymmetric cell division and acquisition of effector properties70. Our data show that IL- 33 stimulation of CTLs strongly induces the transcription of several TCR- dependent genes. Moreover, IL- 33 signals can restore the otherwise suboptimal expansion and effector differentiation of CTLs in response to a low- affinity antigenic peptide. In addition, recent studies demonstrated a loss of IL- 33 in lymphoid organs early after LCMV infection, suggesting substantial alarmin release during T cell priming26. Together, this implies that ST2- signaling might act in conjunction with TCR- signaling to achieve above- threshold activation required for fully functional effector differentiation. + +<|ref|>text<|/ref|><|det|>[[115, 637, 882, 903]]<|/det|> +In summary, we here uncover lineage- specific promoter usage as the molecular mechanism governing disparate expression patterns of the IL- 33 receptor ST2 in distinctly polarized T cell subsets. By generating a type- 1 promoter- deficient mouse strain, in which ST2 expression by Tregs and type- 2 immune cells is preserved, we offer a tool to better assess the role of IL- 33- ST2 signaling in type- 1 immunity. Utilizing these mice, we demonstrate that the type- 1 immunity- restricted Il1rl1 promoter is essential for fully functional antiviral CD8+ and CD4+ T cell responses and critical for the formation of a clonally diverse population of effector CTLs. These findings open new avenues for the modulation and exploitation of IL- 33 signaling in type- 1 immunity- mediated inflammatory diseases and T cell- based cancer immunotherapy, respectively. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 63, 870, 880]]<|/det|> +455 References456 1 Yang, D., Han, Z. & Oppenheim, J. J. Alarms and immunity. Immunol Rev457 280, 41- 56, doi:10.1111/imr.12577 (2017).458 2 Dinarello, C. A. Overview of the IL- 1 family in innate inflammation and459 acquired immunity. Immunol Rev 281, 8- 27, doi:10.1111/imr.12621 (2018).460 3 Liew, F. Y., Girard, J. P. & Turnquist, H. R. Interleukin- 33 in health and461 disease. Nat Rev Immunol 16, 676- 689, doi:10.1038/nri.2016.95 (2016).462 4 Klemenz, R., Hoffmann, S. & Werenskiold, A. K. Serum- and oncoprotein-463 mediated induction of a gene with sequence similarity to the gene encoding464 carcinoembryonic antigen. Proc Natl Acad Sci U S A 86, 5708- 5712,465 doi:10.1073/pnas.86.15.5708 (1989).466 5 Tominaga, S. A putative protein of a growth specific cDNA from BALB/c- 3T3467 cells is highly similar to the extracellular portion of mouse interleukin 1468 receptor. FEBS Lett 258, 301- 304, doi:10.1016/0014- 5793(89)81679- 5(1989).469 (470 6 Löhning, M. et al. T1/ST2 is preferentially expressed on murine Th2 cells,471 independent of interleukin 4, interleukin 5, and interleukin 10, and important472 for Th2 effector function. Proc Natl Acad Sci U S A 95, 6930- 6935,473 doi:10.1073/pnas.95.12.6930 (1998).474 7 Xu, D. et al. Selective expression of a stable cell surface molecule on type 2475 but not type 1 helper T cells. J Exp Med 187, 787- 794,476 doi:10.1084/jem.187.5.787 (1998).477 8 Gachter, T., Werenskiold, A. K. & Klemenz, R. Transcription of the interleukin-478 1 receptor- related T1 gene is initiated at different promoters in mast cells and479 fibroblasts. J Biol Chem 271, 124- 129, doi:10.1074/jbc.271.1.124 (1996).480 9 Schmitz, J. et al. IL- 33, an interleukin- 1- like cytokine that signals via the IL- 1481 receptor- related protein ST2 and induces T helper type 2- associated482 cytokines. Immunity 23, 479- 490, doi:10.1016/j.immuni.2005.09.015 (2005).483 10 Meisel, C. et al. Regulation and function of T1/ST2 expression on CD4+ T484 cells: induction of type 2 cytokine production by T1/ST2 cross- linking. J485 Immunol 166, 3143- 3150, doi:10.4049/jimmunol.166.5.3143 (2001).486 11 Ho, L. H. et al. IL- 33 induces IL- 13 production by mouse mast cells487 independently of IgE- FcepsilonRI signals. J Leukoc Biol 82, 1481- 1490,488 doi:10.1189/jlb.0407200 (2007).489 12 Townsend, M. J., Fallon, P. G., Matthews, D. J., Jolin, H. E. & McKenzie, A.490 N. T1/ST2- deficient mice demonstrate the importance of T1/ST2 in developing491 primary T helper cell type 2 responses. J Exp Med 191, 1069- 1076,492 doi:10.1084/jem.191.6.1069 (2000). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 85, 864, 900]]<|/det|> +493 13 Oboki, K. et al. IL- 33 is a crucial amplifier of innate rather than acquired 494 immunity. Proc Natl Acad Sci U S A 107, 18581- 18586, 495 doi:10.1073/pnas.1003059107 (2010). 496 14 Coyle, A. J. et al. Crucial role of the interleukin 1 receptor family member 497 T1/ST2 in T helper cell type 2- mediated lung mucosal immune responses. J 498 Exp Med 190, 895- 902, doi:10.1084/jem.190.7.895 (1999). 499 15 Hoyler, T. et al. The transcription factor GATA- 3 controls cell fate and 500 maintenance of type 2 innate lymphoid cells. Immunity 37, 634- 648, 501 doi:10.1016/j.immuni.2012.06.020 (2012). 502 16 Moro, K. et al. Innate production of T(H)2 cytokines by adipose tissue- 503 associated c- Kit(+)Sca- 1(+) lymphoid cells. Nature 463, 540- 544, 504 doi:10.1038/nature08636 (2010). 505 17 Mahlakoir, T. et al. Stromal cells maintain immune cell homeostasis in 506 adipose tissue via production of interleukin- 33. Sci Immunol 4, 507 doi:10.1126/sciimmunol.aax0416 (2019). 508 18 Siede, J. et al. IL- 33 Receptor- Expressing Regulatory T Cells Are Highly 509 Activated, Th2 Biased and Suppress CD4 T Cell Proliferation through IL- 10 510 and TGFbeta Release. PLoS One 11, e0161507, 511 doi:10.1371/journal.pone.0161507 (2016). 512 19 Kuswanto, W. et al. Poor Repair of Skeletal Muscle in Aging Mice Reflects a 513 Defect in Local, Interleukin- 33- Dependent Accumulation of Regulatory T 514 Cells. Immunity 44, 355- 367, doi:10.1016/j.immuni.2016.01.009 (2016). 515 20 Monticelli, L. A. et al. IL- 33 promotes an innate immune pathway of intestinal 516 tissue protection dependent on amphi-regulin- EGFR interactions. Proc Natl 517 Acad Sci U S A 112, 10762- 10767, doi:10.1073/pnas.1509070112 (2015). 518 21 Schiering, C. et al. The alarmin IL- 33 promotes regulatory T- cell function in 519 the intestine. Nature 513, 564- 568, doi:10.1038/nature13577 (2014). 520 22 Hemmers, S., Schizas, M. & Rudensky, A. Y. T reg cell- intrinsic requirements 521 for ST2 signaling in health and neuroinflammation. J Exp Med 218, 522 doi:10.1084/jem.20201234 (2021). 523 23 Bonilla, W. V. et al. The alarmin interleukin- 33 drives protective antiviral 524 CD8(+) T cell responses. Science 335, 984- 989, 525 doi:10.1126/science.1215418 (2012). 526 24 Baumann, C. et al. T- bet- and STAT4- dependent IL- 33 receptor expression 527 directly promotes antiviral Th1 cell responses. Proc Natl Acad Sci U S A 112, 528 4056- 4061, doi:10.1073/pnas.1418549112 (2015). 529 25 McLaren, J. E. et al. IL- 33 Augments Virus- Specific Memory T Cell Inflation 530 and Potentiates the Efficacy of an Attenuated Cytomegalovirus- Based 531 Vaccine. J Immunol 202, 943- 955, doi:10.4049/jimmunol.1701757 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 85, 880, 920]]<|/det|> +532 26 Aparicio-Domingo, P. et al. Fibroblast-derived IL-33 is dispensable for lymph node homeostasis but critical for CD8 T-cell responses to acute and chronic viral infection. Eur J Immunol 51, 76-90, doi:10.1002/ejj.201948413 (2021).535 27 Baumann, C. et al. Memory CD8(+) T Cell Protection From Viral Reinfection Depends on Interleukin-33 Alarmin Signals. Front Immunol 10, 1833, doi:10.3389/fimmu.2019.01833 (2019).538 28 Reichenbach, D. K. et al. The IL-33/ST2 axis augments effector T-cell responses during acute GVHD. Blood 125, 3183-3192, doi:10.1182/blood-2014-10-606830 (2015).541 29 Dwyer, G. K. et al. IL-33 acts as a costimulatory signal to generate alloreactive Th1 cells in graft-versus-host disease. J Clin Invest 132, doi:10.1172/JCI150927 (2022).544 30 Rood, J. E. et al. ST2 contributes to T-cell hyperactivation and fatal hemophagocytic lymphohistiocytosis in mice. Blood 127, 426-435, doi:10.1182/blood-2015-07-659813 (2016).547 31 Peine, M., Marek, R. M. & Löhning, M. IL-33 in T Cell Differentiation, Function, and Immune Homeostasis. Trends Immunol 37, 321-333, doi:10.1016/j.it.2016.03.007 (2016).550 32 Molofsky, A. B., Savage, A. K. & Locksley, R. M. Interleukin-33 in Tissue Homeostasis, Injury, and Inflammation. Immunity 42, 1005-1019, doi:10.1016/j.immuni.2015.06.006 (2015).553 33 Hayakawa, M. et al. T-helper type 2 cell-specific expression of the ST2 gene is regulated by transcription factor GATA-3. Biochim Biophys Acta 1728, 53-64, doi:10.1016/j.bbaexp.2005.01.012 (2005).556 34 Iwahana, H. et al. Different promoter usage and multiple transcription initiation sites of the interleukin-1 receptor-related human ST2 gene in UT-7 and TM12 cells. Eur J Biochem 264, 397-406, doi:10.1046/j.1432-1327.1999.00615.x (1999).557 35 Lipsky, B. P., Toy, D. Y., Swart, D. A., Smithgall, M. D. & Smith, D. Deletion of the ST2 proximal promoter disrupts fibroblast-specific expression but does not reduce the amount of soluble ST2 in circulation. Eur J Immunol 42, 1863-1869, doi:10.1002/ejj.201142274 (2012).556 36 Gökmen, M. R. et al. Genome-wide regulatory analysis reveals that T-bet controls Th17 lineage differentiation through direct suppression of IRF4. J Immunol 191, 5925-5932, doi:10.4049/jimmunol.1202254 (2013).557 37 Wei, G. et al. Genome-wide analyses of transcription factor GATA3-mediated gene regulation in distinct T cell types. Immunity 35, 299-311, doi:10.1016/j.immuni.2011.08.007 (2011).570 38 Wei, L. et al. Discrete roles of STAT4 and STAT6 transcription factors in tuning epigenetic modifications and transcription during T helper cell + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 85, 870, 120]]<|/det|> +572 differentiation. Immunity 32, 840- 851, doi:10.1016/j.immuni.2010.06.003 (2010). + +<|ref|>text<|/ref|><|det|>[[50, 131, 858, 183]]<|/det|> +574 39 Chen, X. et al. The histone methyltransferase EZH2 primes the early differentiation of follicular helper T cells during acute viral infection. Cell Mol Immunol 17, 247-260, doi:10.1038/s41423-019-0219-z (2020). + +<|ref|>text<|/ref|><|det|>[[50, 195, 857, 246]]<|/det|> +577 40 Yu, B. et al. Epigenetic landscapes reveal transcription factors that regulate CD8(+) T cell differentiation. Nat Immunol 18, 573-582, doi:10.1038/ni.3706 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 258, 830, 292]]<|/det|> +580 41 Fantom-Consortium et al. A promoter-level mammalian expression atlas. Nature 507, 462-470, doi:10.1038/nature13182 (2014). + +<|ref|>text<|/ref|><|det|>[[50, 305, 840, 356]]<|/det|> +582 42 Demircioğlu, D. et al. A Pan-cancer Transcriptome Analysis Reveals Pervasive Regulation through Alternative Promoters. Cell 178, 1465-1477 e1417, doi:10.1016/j.cell.2019.08.018 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 368, 875, 419]]<|/det|> +585 43 Siepel, A. et al. Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res 15, 1034-1050, doi:10.1101/gr.3715005 (2005). + +<|ref|>text<|/ref|><|det|>[[50, 431, 864, 482]]<|/det|> +588 44 Frazer, K. A., Pachter, L., Poliakov, A., Rubin, E. M. & Dubchak, I. VISTA: computational tools for comparative genomics. Nucleic Acids Res 32, W273-279, doi:10.1093/nar/gkh458 (2004). + +<|ref|>text<|/ref|><|det|>[[50, 495, 848, 528]]<|/det|> +591 45 Hon, C. C. et al. An atlas of human long non-coding RNAs with accurate 5' ends. Nature 543, 199-204, doi:10.1038/nature21374 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 541, 870, 592]]<|/det|> +593 46 Kanhere, A. et al. T-bet and GATA3 orchestrate Th1 and Th2 differentiation through lineage-specific targeting of distal regulatory elements. Nat Commun 3, 1268, doi:10.1038/ncomms2260 (2012). + +<|ref|>text<|/ref|><|det|>[[50, 605, 880, 656]]<|/det|> +596 47 Griesenauer, B. & Paczesny, S. The ST2/IL-33 Axis in Immune Cells during Inflammatory Diseases. Front Immunol 8, 475, doi:10.3389/fimmu.2017.00475 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 669, 864, 720]]<|/det|> +599 48 McLane, L. M., Abdel-Hakeem, M. S. & Wherry, E. J. CD8 T Cell Exhaustion During Chronic Viral Infection and Cancer. Annu Rev Immunol 37, 457-495, doi:10.1146/annurev-immunol-041015-055318 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 733, 864, 783]]<|/det|> +602 49 Yao, C. et al. Single-cell RNA-seq reveals TOX as a key regulator of CD8(+) T cell persistence in chronic infection. Nat Immunol 20, 890-901, doi:10.1038/s41590-019-0403-4 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 796, 865, 846]]<|/det|> +605 50 Pais Ferreira, D. et al. Central memory CD8(+) T cells derive from stem-like Tcf7(hi) effector cells in the absence of cytotoxic differentiation. Immunity 53, 985-1000 e1011, doi:10.1016/j.immuni.2020.09.005 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 860, 855, 910]]<|/det|> +608 51 Zehn, D., Thimme, R., Lugli, E., de Almeida, G. P. & Oxenius, A. 'Stem-like' precursors are the fount to sustain persistent CD8(+) T cell responses. Nat Immunol 23, 836-847, doi:10.1038/s41590-022-01219-w (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 85, 878, 135]]<|/det|> +611 52 Kurachi, M. et al. Chemokine receptor CXCR3 facilitates CD8(+) T cell differentiation into short-lived effector cells leading to memory degeneration. J Exp Med 208, 1605- 1620, doi:10.1084/jem.20102101 (2011). + +<|ref|>text<|/ref|><|det|>[[50, 148, 875, 199]]<|/det|> +614 53 Chang, Y. M. et al. T Cell Receptor Diversity and Lineage Relationship between Virus-Specific CD8 T Cell Subsets during Chronic Lymphocytic Choriomeningitis Virus Infection. J Virol 94, doi:10.1128/JVI.00935-20 (2020). + +<|ref|>text<|/ref|><|det|>[[50, 212, 835, 262]]<|/det|> +617 54 Yamamoto, T. et al. DUSP10 constrains innate IL-33-mediated cytokine production in ST2(hi) memory-type pathogenic Th2 cells. Nat Commun 9, 4231, doi:10.1038/s41467-018-06468-8 (2018). + +<|ref|>text<|/ref|><|det|>[[50, 276, 850, 310]]<|/det|> +620 55 Cohen, E. S. et al. Oxidation of the alarmin IL-33 regulates ST2-dependent inflammation. Nat Commun 6, 8327, doi:10.1038/ncomms9327 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 323, 850, 373]]<|/det|> +622 56 Joshi, N. S. et al. Inflammation directs memory precursor and short-lived effector CD8(+) T cell fates via the graded expression of T-bet transcription factor. Immunity 27, 281-295, doi:10.1016/j.immuni.2007.07.010 (2007). + +<|ref|>text<|/ref|><|det|>[[50, 386, 878, 436]]<|/det|> +625 57 Dominguez, C. X. et al. The transcription factors ZEB2 and T-bet cooperate to program cytotoxic T cell terminal differentiation in response to LCMV viral infection. J Exp Med 212, 2041-2056, doi:10.1084/jem.20150186 (2015). + +<|ref|>text<|/ref|><|det|>[[50, 450, 874, 516]]<|/det|> +628 58 Rutishauser, R. L. et al. Transcriptional repressor Blimp-1 promotes CD8(+) T cell terminal differentiation and represses the acquisition of central memory T cell properties. Immunity 31, 296-308, doi:10.1016/j.immuni.2009.05.014 (2009). + +<|ref|>text<|/ref|><|det|>[[50, 530, 870, 579]]<|/det|> +632 59 Ashouri, J. F. et al. Reporters of TCR signaling identify arthritogenic T cells in murine and human autoimmune arthritis. Proc Natl Acad Sci U S A 116, 18517-18527, doi:10.1073/pnas.1904271116 (2019). + +<|ref|>text<|/ref|><|det|>[[50, 593, 861, 628]]<|/det|> +635 60 Cibrian, D. & Sanchez-Madrid, F. CD69: from activation marker to metabolic gatekeeper. Eur J Immunol 47, 946-953, doi:10.1002/eji.201646837 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 641, 870, 690]]<|/det|> +637 61 Iwata, A. et al. Quality of TCR signaling determined by differential affinities of enhancers for the composite BATT-IRF4 transcription factor complex. Nat Immunol 18, 563-572, doi:10.1038/ni.3714 (2017). + +<|ref|>text<|/ref|><|det|>[[50, 704, 872, 753]]<|/det|> +640 62 Utzschneider, D. T. et al. High antigen levels induce an exhausted phenotype in a chronic infection without impairing T cell expansion and survival. J Exp Med 213, 1819-1834, doi:10.1084/jem.20150598 (2016). + +<|ref|>text<|/ref|><|det|>[[50, 767, 878, 816]]<|/det|> +643 63 Xu, D. et al. Selective expression and functions of interleukin 18 receptor on T helper (Th) type 1 but not Th2 cells. J Exp Med 188, 1485-1492, doi:10.1084/jem.188.8.1485 (1998). + +<|ref|>text<|/ref|><|det|>[[50, 830, 872, 896]]<|/det|> +646 64 Bergers, G., Reikerstorfer, A., Braselmann, S., Graninger, P. & Busslinger, M. Alternative promoter usage of the Fos-responsive gene Fit-1 generates mRNA isoforms coding for either secreted or membrane-bound proteins related to the IL-1 receptor. EMBO J 13, 1176-1188 (1994). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 84, 866, 470]]<|/det|> +650 65 SEQC-MAQC-III-Consortium. A comprehensive assessment of RNA-seq accuracy, reproducibility and information content by the Sequencing Quality Control Consortium. Nat Biotechnol 32, 903- 914, doi:10.1038/nbt.2957 (2014).652 653 (2014).654 66 Braun, H., Afonina, I. S., Mueller, C. & Beyaert, R. Dichotomous function of IL- 655 33 in health and disease: From biology to clinical implications. Biochem656 Pharmacol 148, 238- 252, doi:10.1016/j.bcp.2018.01.010 (2018).657 67 Matta, B. M. et al. Peri- alloHCT IL- 33 administration expands recipient T- 658 regulatory cells that protect mice against acute GVHD. Blood 128, 427- 439,659 doi:10.1182/blood- 2015- 12- 684142 (2016).660 68 Yang, J. et al. Rorc restrains the potency of ST2+ regulatory T cells in661 ameliorating intestinal graft- versus- host disease. JCI Insight 4,662 doi:10.1172/jci.insight.122014 (2019).663 69 Zeiser, R. & Blazar, B. R. Acute Graft- versus- Host Disease - Biologic664 Process, Prevention, and Therapy. N Engl J Med 377, 2167- 2179,665 doi:10.1056/NEJMra1609337 (2017).666 70 King, C. G. et al. T cell affinity regulates asymmetric division, effector cell667 differentiation, and tissue pathology. Immunity 37, 709- 720,668 doi:10.1016/j.immuni.2012.06.021 (2012). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 300, 101]]<|/det|> +## Acknowledgments + +<|ref|>text<|/ref|><|det|>[[115, 106, 883, 421]]<|/det|> +We thank J. Kirsch, T. Kaiser (Flow Cytometry Core Facility, DRFZ, Berlin), G. Guerra, K. Lehmann, V. Holecska, I. Panse, C.L. Tran (DRFZ, Berlin), C. Scholl, A. Leschke (Transgenics Core facility, MDC, Berlin), T. Abreu Mota, M. Ji- Lu (University of Basel), A. Greco (DKFZ, Heidelberg), and A.N. Hegazy (Charité, Berlin) for technical assistance and advice. This work was supported by the German Research Foundation (DFG grants LO 1542/5- 1, 1542/4- 1), the Swiss National Science Foundation (project grants 310030_185318/1 and Sinergia grant CRSII3_160772/1), the Willy Robert Pitzer Foundation (Pitzer Laboratory of Osteoarthritis Research), the Dr. Rolf M. Schwiete Foundation (Osteoarthritis Research Program), the state of Berlin, and the European Regional Development Fund (ERDF 2014- 2020, EFRE 1.8/11). S.S. is a member of the Berlin Institute of Health at Charité – Universitätsmedizin Berlin. T.M.B., M.D., and N.D. were fellows of the International Max Planck Research School for Infectious Diseases and Immunology. + +<|ref|>sub_title<|/ref|><|det|>[[118, 441, 321, 459]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 464, 883, 605]]<|/det|> +T.M.B., S.S., and M.L. conceptualized the study. T.M.B., S.S., A.F.M., D.D.P., and M.L. interpreted the results. T.M.B., S.S., and M.L. wrote the manuscript. T.M.B., S.S., A.F.M., M.D., N.D., and P.S. conducted in vivo experiments. S.S., F.H., P.D., G.A.H., and M.F.M. performed and analyzed RNA-Seq and scRNA-Seq experiments. T.M.B. and R.K. designed and generated ll1rl1- ExAB- and ll1rl1- ExC- mice. C.K., T.H., and D.D.P. provided expertise and advice. All authors reviewed and edited the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[118, 626, 335, 644]]<|/det|> +## Declaration of interest + +<|ref|>text<|/ref|><|det|>[[115, 650, 881, 694]]<|/det|> +D.D.P. is a founder, shareholder, and advisor to Hookipa Pharma Inc. commercializing arenavirus vectors. The remaining authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[118, 715, 273, 732]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 738, 881, 831]]<|/det|> +RNA sequencing data generated in this study are available from the Gene Expression Omnibus (GEO, National Center for Biotechnology Information) under the accession codes GSE204695 and GSE204693. All other data supporting the findings of this study are available from the corresponding authors upon reasonable request. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[68, 84, 280, 102]]<|/det|> +## 698 Code availability + +<|ref|>text<|/ref|><|det|>[[67, 108, 880, 152]]<|/det|> +699 Code related to the data analysis of this study has been deposited to GitLab (https://agloehninggitlab.gitlab.io/BrunnerServe- Type1- ST2/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 110, 880, 620]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 630, 881, 897]]<|/det|> +
Fig. 1: A previously unrecognized alternative promoter drives IL-33 receptor expression in antiviral T cells. a, Scheme depicting the curated Il1rl1 gene. b, ST2 surface expression by in vitro differentiated T cell subsets. c, Il1rl1 first exon usage by differentiated T cells (CTL, Th2: \(n = 4\) , Th1: \(n = 3\) , NIH3T3: \(n = 2\) ). d, RNA-Seq coverage and splice junction tracks of ST2+ CTLs, Th1, and Th2 cells ( \(n = 3\) per subset) at the Il1rl1 locus. Chr1:40,377,000-40,465,500; GRCm38.p6/mm10 is shown. e-g, Wildtype mice received LCMV-specific P14 or Smarta T cells and were infected with LCMV-WE. P14 CTLs and Smarta Th1 cells were isolated on d7 p.i., and Il1rl1 first exon usage was analyzed by qPCR (e; CTL, Th1: \(n = 4\) , Th2 control (in vitro): \(n = 2\) ) and RT-PCR (f). g, Chip-Seq tracks indicating T-bet36, STAT438, and GATA-337 binding and ATAC-Seq tracks showing chromatin accessibility in naive or activated
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 81, 883, 300]]<|/det|> +714 LCMV- specific T cells \(^{39,40}\) . h, Computational pipeline to identify alternative transcription start sites (TSSs) between type- 1 (CTL, Th1) and type- 2 (Th2) polarized T cells using cufflinks \(^{71}\) and ProActiv \(^{42}\) . i, Manhattan plot showing identified hits (parameters minAbs=0.25 and promoter FC=2, dashed line: \(P = 0.01\) ). j, Heatmap of identified TSSs with \(P < 0.01\) and their respective ProActiv- normalized expression across all replicates. Data in c, e, and f are representative of two independent experiments. Data are presented as mean ± SD with each dot representing T cells isolated from individual mice. \(P\) was determined using One- way ANOVA with Tukey's post- hoc test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 81, 876, 333]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 346, 881, 368]]<|/det|> +
Fig. 2: Alternative promoter usage at the ll1rl1 locus is conserved between mice
+ +<|ref|>text<|/ref|><|det|>[[115, 371, 883, 734]]<|/det|> +and humans. a, IGV browser display of the ll1rl1 type- 1 promoter region (chr1:40,386,428 - 40,386,458; GRCm38. p6/mm10) showing a CTL RNA- Seq track, T- bet- and STAT4- binding36,38 chromatin accessibility in naive and activated CTLs39,40 and a PhyloP track indicating evolutionary conservation across 60 vertebrate species43,72. CNS- 5, as identified using Vista44, is highlighted (chr1:40,386,282- 40,386,541; GRCm38. p6/mm10). b, T- bet and STAT4 binding motifs and their respective prediction score within CNS- 5 as determined using JASPAR motif analysis73 (chr1:40,386,429- 40,386,459; GRCm38. p6/mm10). c, IGV browser display of cap analysis of gene expression sequencing (CAGE- Seq) data and corresponding CAGE- associated transcripts (CAT) as provided and published by the FANTOM5 consortium41,45. ChIP- data track showing T- bet binding in human Th1 cells46 (chr2:102,862,400- 102,970,100; GRCh37/hg19). d, IL1RL1 first exon expression in human T cell subsets (CTL: \(n = 6\) , Th1 \(n = 5\) , Th2: \(n = 4\) ). Data are presented as mean \(\pm\) SD with each dot representing T cells isolated from individual donors. \(P\) was determined using One- way ANOVA with Tukey's post- hoc test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 80, 884, 499]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 508, 881, 570]]<|/det|> +
Fig. 3: Usage of distinct promoters allows T cell subset-specific targeting of ST2 expression. a, Experimental outline and representative FACS plot showing GFP expression by transduced T cells. b,c, Representative FACS plots (b) and quantification (c) of ST2 surface expression by transduced T cells analyzed after 5 days of culture ( \(n = 6\) pooled from two (CTL) or three (Th1, Th2) independent experiments). d, Scheme depicting generation of Il1rl1-ExAB \(^{-/ - }\) and Il1rl1-ExC \(^{-/ - }\) mice using CRISPR/Cas9 in murine zygotes. e,f, Representative histograms (e) and quantification (f) of ST2 surface expression by WT, Il1rl1-ExAB \(^{-/ - }\) , or Il1rl1-ExC \(^{-/ - }\) T cells ( \(n = 4\) ). g, IFN- \(\gamma\) and IL-13 secretion by WT or Il1rl1-ExAB \(^{-/ - }\) T cells after IL-33 stimulation ( \(n = 3\) ). h-n, Wildtype, Il1rl1- \(^{-/ - }\) , Il1rl1-ExAB \(^{-/ - }\) , and Il1rl1-ExC \(^{-/ - }\) mice were infected with LCMV-WE and ST2 expression by splenic CTLs (i,j), Th1 cells (k,l), and Tregs (m,n) was quantified on d7 p.i. (WT: \(n = 9\) , Il1rl1- \(^{-/ - }\) : \(n = 6\) , Il1rl1-ExAB \(^{-/ - }\) : \(n = 8\) , Il1rl1-ExC \(^{-/ - }\) : \(n = 7\) ). Data represent one (g), two (e,f), or three (i-n) independent experiments and are presented as mean \(\pm\) SD with each dot representing one mouse (j,l,n) or one experiment performed with T cells from individual mice (c,f). \(P\) was determined using One-way (j,l,n) or Two-way (c,f) ANOVA with Tukey's post-hoc test.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 81, 884, 615]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 622, 881, 916]]<|/det|> +
Fig. 4: T cell-intrinsic activity of the type-1 Il1rl1 promoter is vital for efficient expansion of antiviral T cells. a-h, Wildtype, Il1rl1-/-; Il1rl1-ExAB-/- and Il1rl1-ExC-/- mice were infected with LCMV-WE, and splenic T cells were analyzed on d7 p.i. (WT: \(n = 9\) , Il1rl1-/-: \(n = 6\) , Il1rl1-ExAB-/-: \(n = 8\) , Il1rl1-ExC-/-: \(n = 7\) ). a, Experimental outline. b, c, Frequencies (b) and absolute cell counts (c) of CTLs. d-f, Representative staining (d) and frequencies of activated CD44+ CD62L- CTLs (e) and Ki67+ CTLs (f). g, Absolute cell counts of LCMV GP33-41- or NP396-404-specific CTLs. h, Counts of CTLs expressing effector molecules after ex vivo stimulation with GP33-41. i, Serum IFN-γ levels at indicated timepoints p.i. with LCMV-WE (WT and Il1rl1-/-: \(n = 4\) , Il1rl1-ExAB-/-: \(n = 5\) , Il1rl1-ExC-/-: \(n = 3\) ). j-m, Irradiated WT recipients (CD45.1+) were reconstituted with WT (CD45.1+ CD45.2+) and Il1rl1-ExAB-/- or Il1rl1-ExC-/- (all CD45.2+) bone marrow, infected with LCMV-WE, and analyzed on d10 p.i. (WT + Il1rl1-/-: \(n = 6\) ,
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 80, 883, 555]]<|/det|> +770 WT + Il1rl1- ExAB- and WT + Il1rl1- ExC-: \(n = 7\) ). j, Experimental outline. k, Representative FACS plots showing CTL populations. I,m, Cell counts of splenic CTLs (I) and CTLs expressing IFN- \(\gamma\) after ex vivo stimulation with \(\mathsf{GP}_{33 - 41}(\mathsf{m})\) . n- r, P14 T cells (CD45.1+ CD45.2+) were adoptively cotransferred with Il1rl1- P14 or Il1rl1- ExAB- P14 T cells (CD45.1+) into WT mice (CD45.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. ( \(n = 7\) ). n, Experimental outline. o, Representative FACS plots showing CTL populations. p,q, Frequencies (p) and cell counts (q) of splenic P14 T cells. r, Counts of IFN- \(\gamma\) - expressing P14 cells after ex vivo stimulation with \(\mathsf{GP}_{33 - 41}\) . s- w, Smarta T cells (CD90.1+) were adoptively cotransferred with Il1rl1- Smarta or Il1rl1- ExAB- Smarta T cells (CD90.1+ CD90.2+) into WT mice (CD90.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. (Il1rl1- Smarta: n=6, Il1rl1- ExAB- Smarta: n=7). s, Experimental outline. t, Representative FACS plots showing CD4+ T cell populations. u,v, Frequencies (u) and cell counts (v) of splenic Smarta T cells. w, Counts of IFN- \(\gamma\) - expressing Smarta cells after ex vivo stimulation with \(\mathsf{GP}_{64 - 79}\) . Data represent one (i- m), two (n- w), or three (a- h) independent experiments and are presented as mean \(\pm\) SD with each dot representing one mouse or the arithmetic mean of all replicates (h,i). P was determined using Oneway ANOVA (b,c,e,f), Two- way (g,i) ANOVA with Tukey's post- hoc test, or Two- way RM ANOVA with Sidak's post- hoc test (I,m,p- r,u- w). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 75, 880, 670]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 680, 881, 897]]<|/det|> +
Fig. 5: Type-1 llrl1 promoter engagement facilitates effector differentiation of CTLs to generate a clonally diverse SLEC population. a-d, n-r, Wildtype and llrl1-ExAB-/- mice were infected with LCMV-WE, and splenic CD44+ CTLs were analyzed by multiplexed scRNA-Seq and scTCR-Seq analysis on d7 p.i. (n = 3 mice pooled per group). a, Experimental outline. b, UMAP plots colored by cluster type. c, UMAP plots showing normalized expression of selected genes in both genotypes. d, Change in cluster composition of llrl1-ExAB-/- CTLs relative to WT CTLs. e, Representative FACS plots showing KLRG1 and CD127 expression by CD44+ CD8+ T cells. f-h, Frequencies (f) and counts (g) of KLRG1+ CD127- SLECs and frequencies (h) of
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 81, 883, 425]]<|/det|> +799 CXCR3+ CTLs in LCMV- WE infected WT, Il1rl1- /-, Il1rl1- ExAB- /-, and Il1rl1- ExC- /- mice (e- h, WT: \(n = 9\) , Il1rl1- /- : \(n = 6\) , Il1rl1- ExAB- /- : \(n = 8\) , Il1rl1- ExC- /- : \(n = 7\) ). i, Frequencies of KLRG1+ CD127- SLECs in infected mixed bone marrow chimeras (WT + Il1rl1- /- : \(n = 6\) , WT + Il1rl1- ExAB- /- and WT + Il1rl1- ExC- /- : \(n = 7\) ). j, k, Frequencies of KLRG1+ CD127- SLECs (j) and CXCR3+ P14 cells (k) in adoptive cotransfer experiments on d10 p.i. \((n = 7)\) . m, ST2 expression by GP33- 41- specific SLECs and MPECs in WT mice at d7 p.i. \((n = 9)\) . n, Counts of CD44+ CTLs per spleen. o, TCR clonotypes in sequenced CD44+ CTLs. p, Graph displaying number of TCR clonotypes and their abundancy among all analyzed CD44+ CTLs. q, TCR clonotypes in SLEC and MPEC clusters. r, TCR repertoire occupation in individual WT and Il1rl1- ExAB- /- mice. Data represent one (b- d, i, n- r) or two (e- h, j- m) independent experiments and are presented as mean \(\pm\) SD with each dot or line representing one mouse (f- j, l, n- q). \(P\) was determined using two- tailed t- tests (n, q), One- way ANOVA with Tukey's post- hoc test (f, h), or Two- way RM ANOVA with Sidák's post- hoc test (i, j, l). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 80, 881, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 785, 880, 828]]<|/det|> +
Fig. 6: Transcriptional profiling indicates broad costimulatory and TCR-cooperative functions of IL-33–ST2 signaling in T cells. a-f, Differentiated CTLs,
+ +<|ref|>text<|/ref|><|det|>[[115, 833, 880, 902]]<|/det|> +Th1 cells, and Th2 cells were stimulated with IL- 33 or left untreated for 2h and subjected to RNA- Seq analysis \((n = 3\) independent cultures per subset). a, Experimental outline. b, Heatmaps depicting differentially expressed genes in each T + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 82, 883, 496]]<|/det|> +819 cell subset (log2 fold change \(>1.0\) ; \(P\) adjusted \(< 0.01\) ). c- f, Comparison of IL- 33- 820 stimulated and untreated T cells (2h timepoint). c, Venn- diagram illustrating the 821 overlap in differentially regulated genes between CTLs, Th1, and Th2 cells. d, Gene 822 ontology biological process overrepresentation analysis of IL- 33- induced genes 823 shared among all subsets. e, Expression of selected transcription factors, cytokines 824 and chemokines, and TCR- regulated genes in each T cell subset. f, Gene set 825 enrichment analysis of TCR downstream genes in IL- 33- stimulated versus 826 unstimulated CTLs. g- I, P14 T cells ( \(\mathrm{CD45.1^{+}CD45.2^{+}}\) ) were adoptively cotransferred 827 with \(11r1r1^{- / - }\) P14 T cells ( \(\mathrm{CD45.1^{+}}\) ) into WT mice ( \(\mathrm{CD45.2^{+}}\) ). Recipients were infected 828 with high- or low- affinity GP33- expressing LCMV- CI13 and analyzed at d10 p.i. (High- 829 affinity group: \(n = 7\) , Low- affinity group: \(n = 6\) ). g, Experimental outline. h, Counts of 830 recovered P14 and \(11r1r1^{- / - }\) P14 T cells. i,j, Frequencies (i) and counts (j) of KLRG1+ 831 CD127- P14 cells. k,l, Counts of endogenous \(\mathrm{NP}_{396 - 404}\) - specific CTLs (k) and KLRG1+ 832 CD127- \(\mathrm{NP}_{396 - 404}\) - specific CTLs (l). Data represent one (a- f) or two (g- l) independent 833 experiments. Data are presented as mean ± SD with each dot representing one mouse 834 (h- l). \(P\) was determined using One- way ANOVA with Tukey's post- hoc test (h- j) or 835 two- tailed t- tests (k,l). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 113, 880, 335]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 350, 880, 418]]<|/det|> +
837 Extended Data Fig. 1: ST2 expression by CTLs and Th1 cells is regulated by T-
+ +<|ref|>text<|/ref|><|det|>[[115, 421, 882, 664]]<|/det|> +837 Extended Data Fig. 1: ST2 expression by CTLs and Th1 cells is regulated by T- bet and STAT4. a, Experimental scheme of in vitro T cell differentiation. b, Representative T- bet and GATA- 3 stainings of differentiated T cells with expression intensity depicted as geometric mean index (GMI), normalized to isotype control stainings (grey) \((n = 4)\) . c, Cytokine expression of polarized T cells upon restimulation \((n = 4)\) . d- g, Representative histograms (d,f) and quantification (e,g) of ST2 surface expression by in vitro differentiated wildtype, STAT4-, or T- bet- deficient T cells, or T cells activated in the absence of IL- 12. Stainings with secondary reagents without primary ST2 antibody served as staining controls (ctrl) (dotted line, bottom). h, ST2 expression by splenic Tregs in wildtype, Stat4- , or Tbx21- mice. Data represent two independent experiments and are presented as mean ± SD (b- h) with each dot representing one mouse (h) or one culture performed with T cells from individual mice (c,e,g). P was determined using One- way ANOVA with Tukey's post- hoc test (e,g,h). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 85, 839, 530]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[55, 536, 884, 780]]<|/det|> +Extended Data Fig. 2: Flow- cytometric sorting and analysis of human T cells. a, Gating strategy for the sorting of human CXCR3+ CRTH2- Th1 cells and CXCR3+ CRTH2+ Th2 cells. b, Gating strategy for the sorting of human CD8+ T cells. c,d, Representative histograms and quantification of T- bet (c) and GATA- 3 (d) expression by activated T cells at day 5 of culture \((n = 3)\) . e,f, Representative histograms and quantification of IFN- \(\gamma\) (e) and IL- 13 (f) expression by T cells stimulated with PMA/ionomycin at day 5 of culture \((n = 3)\) . Data represent two independent experiments and are presented as mean \(\pm\) SD with each dot representing one culture performed with T cells from individual donors (c- f). \(P\) was determined using One- way ANOVA with Tukey's post- hoc test (c- f). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 78, 880, 800]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 799, 881, 894]]<|/det|> +Extended Data Fig. 3: Generation of Il1rl1- ExAB- and Il1rl1- ExC- mice. a, Schematic depiction of the gene- targeting approach for the generation of Il1rl1- ExAB- and Il1rl1- ExC- mice. b, Representative genotyping PCRs to identify heterozygous and homozygous mutant mice. Chromatograms depicting the sequence of joined DNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 82, 884, 448]]<|/det|> +segments as analyzed by Sanger- Sequencing. c- k, Analysis of adaptive and innate immune cells in spleens and lymph nodes (LN) of llrl1- ExAB- and llrl1- ExC- mice. c, Representative staining of CD4 and CD8 on splenic T cells. d, e, Frequencies (d) and absolute cell counts (e) of \(\mathsf{CD8^{+}}\) T cells, \(\mathsf{CD4^{+}}\) T cells, and B cells. f, Representative staining of CD62L and CD44 on splenic \(\mathsf{CD8^{+}}\) T cells. g, Frequency of effector and central memory \(\mathsf{CD8^{+}}\) T cells. h, Frequencies and absolute cell counts of splenic Tregs. i, Gating strategy for the analysis of innate immune cells. j, k, Frequencies (j) and absolute cell counts (k) of splenic neutrophils, eosinophils, macrophages (MΦ), conventional dendritic cells (cDCs), and natural killer (NK) cells. Data represent two independent experiments and are presented as mean ± SD with each dot representing one mouse (Wildtype, llrl1- ExAB- , and llrl1- ExC- : \(n = 4\) , llrl1- ExAB- : \(n = 5\) ; Treg analysis: Wildtype, llrl1- ExAB- : \(n = 8\) and llrl1- ExC- : \(n = 6\) ; NK cell analysis: Wildtype, llrl1- ExAB- , and llrl1- ExC- : \(n = 4\) ). P was determined using One- way ANOVA (h, j, k) or Two- way ANOVA (d, e, g) with with Tukey's post- hoc test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 880, 595]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 603, 882, 871]]<|/det|> +Extended Data Fig. 4: Type- 1 llrl1 promoter deficiency does not impair ST2 expression by type- 2 innate immune cells. a, RNA- Seq coverage tracks and detected splice junctions at the llrl1 locus of ILC2s74 and mast cells75. chr1:40,377,000- 40,465,500; GRCm38. p6/mm10 is shown. b, Gating strategy for the analysis of Lin- CD45+ CD3- CD127+ ILCs. c, Representative histograms showing ST2 expression by KLRG1+ IL- 18R- ILC2s isolated from lungs of wildtype or llrl1- ExAB- mice. d, Frequencies of ST2+ ILC2s and ST2 expression intensities (MFI). e, Gating strategy for the analysis of c- kit+ FcεRla+ peritoneal mast cells. f, Representative histograms showing ST2 expression by peritoneal mast cells isolated from wildtype or llrl1- ExAB- mice. g, Frequencies of ST2+ mast cells and ST2 expression intensities (MFI). Results are presented as mean ± SD with each dot representing one mouse (n + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 83, 880, 129]]<|/det|> +894 = 4 mice per genotype). Data are representative of two experiments. \(P\) was 895 determined using using two- tailed t- tests (d,g). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 75, 881, 690]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[113, 694, 881, 911]]<|/det|> +Extended Data Fig. 5: The type- 1 llr1l promoter drives expansion and activation of antiviral T cells. a- i, Wildtype, llr1l- ExAB- , and llr1l- ExC- mice were infected with LCMV- WE and the T cell response was analyzed on d7 p.i. (wildtype: \(n = 9\) , llr1l- : \(n = 6\) , llr1l- ExAB- : \(n = 8\) , llr1l- ExC- : \(n = 7\) ). a, Experimental outline. b, c, Representative FACS plots (b) and quantification (c) of CTLs in livers of infected animals. d- f, Representative FACS plots (d) and quantification of ST2 expression (e) and GP33- 41- Tetramer+ cells in livers (f). g- i, Representative FACS plots (g) and quantification (h, i) of effector molecule expression by splenic CTLs stimulated with LCMV GP33- 41 peptide. j, k, Irradiated wildtype recipients (CD45.1+) were reconstituted + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 81, 884, 425]]<|/det|> +with wildtype (CD45.1+ CD45.2+) and Il1rl1- /-, Il1rl1- ExAB- /-, or Il1rl1- ExC- (all CD45.2+) bone marrow, infected with LCMV- WE, and analyzed on d10 p.i. (WT + Il1rl1- /-: \(n = 6\) , WT + Il1rl1- ExAB- /- and WT + Il1rl1- ExC- /-: \(n = 7\) ). j, Experimental outline. k, Absolute cell counts of LCMV GP33- 41- specific CTLs. I- p, Smarta T cells (CD90.1+) were adoptively cotransferred with Il1rl1- /- Smarta or Il1rl1- ExAB- /- Smarta T cells (CD90.1+ CD90.2+) into WT mice (CD90.2+). Recipients were infected with LCMV- CI13 and analyzed at d10 p.i. (Il1rl1- /- Smarta: \(n = 6\) , Il1rl1- ExAB- /- Smarta: \(n = 7\) ). I, Experimental outline. m, n, Representative FACS plots showing CD4+ T cell populations before transfer (m) and after infection (n). o, p, Frequencies (o) and cell counts (p) of Smarta T cells in livers of infected mice. Data represent one (j- p), two (a- f), or three (g- i) independent experiments and are presented as mean ± SD with each dot representing one mouse. P was determined using One- way ANOVA (c, e, h, i) with Tukey's post- hoc test, Two- way ANOVA (f) with Tukey's post- hoc test, or with Two- way RM ANOVA with Sidak's post- hoc test (k, o, p). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 884, 710]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[55, 722, 884, 818]]<|/det|> +Extended Data Fig. 6: scRNA-Seq profiling of antiviral T cells in wildtype and Il1rl1- ExAB- mice. a, Gating strategy for the flow- cytometric sorting of CD44+ CTLs from spleens of LCMV- infected wildtype and Il1rl1- ExAB- mice. b, Heatmap displaying all analyzed CTLs and the top ten marker genes per cluster. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 84, 204, 101]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[117, 123, 167, 139]]<|/det|> +## Mice + +<|ref|>text<|/ref|><|det|>[[115, 145, 881, 537]]<|/det|> +C57BL/6J mice (wildtype), LCMV- TCRtg P1476 and Smarta77 mice expressing the congenic markers CD45.1 (Ly5.1) or CD90.1 (Thy1.1), respectively, I1/r1- /- 12, I1/r1- 929 ExAB- /-, I1/r1- ExC- /-, Stat4- /- 78, Tbx21- /- 79, I1/r1- ExAB- /- Smarta, Stat4- /- Smarta, Tbx21- /- Smarta, I1/r1- /- P14, I1/r1- ExAB- /- P14, and TCRβδ- /- 80 mice were bred under specific- pathogen- free conditions in approved animal- care facilities at the Research Institute for Experimental Medicine (FEM) of the Charité – Universitätsmedizin Berlin (Berlin, Germany) or at the Laboratory Animal Facility of the ETH Zürich (ETH Phenomics Center, Zürich, Switzerland). Mice were housed in individually ventilated cages with a 12 h light/dark cycle and had ad libitum access to drinking water and chow. Both, male and female mice between 8 and 26 weeks of age were used for experiments. For LCMV infections, experimental groups were age- and sex- matched. Mice used for scRNA- Seq analyses were cohoused for 4 weeks prior to LCMV infection. Animal experiments were performed in accordance with the German or Swiss law for animal protection and approved by the respective governmental authority (Landesamt für Gesundheit und Soziales Berlin and the Cantonal Veterinary Office of the Canton of Basel; T0058/08, G0111/17, G0206/17, G0245/19). + +<|ref|>sub_title<|/ref|><|det|>[[117, 555, 583, 574]]<|/det|> +## Generation of I1/r1-ExAB- and I1/r1-ExC- mice + +<|ref|>text<|/ref|><|det|>[[115, 580, 881, 896]]<|/det|> +I1/r1- ExAB- and I1/r1- ExC- mice were generated in the Transgenics Core Facility of the Max Delbrück Centrum Berlin using established protocols81. In brief, gRNA sequences with minimal predicted off- target effects (gRNA- ExAB- 5'.1: TAATGCATAGGACCGACCAAAGG, gRNA- ExAB 3'.1: ATTTGGACTGTCATCGTC GTGGG, gRNA- ExAB- 5'.2: GCACCCTATGAATCTACACAGGG, gRNA- ExAB- 3'.2: GACTTGCATCGTCTGGGCCGGG, gRNA- EXC- 5': CCAACGTGGCAGTATGTC CAGG, gRNA- EXC- 3': ATGAAACGGTGACTCAGTTAGGG) were identified using the web- based tool CRISPR82. Zygotes were collected from C57BL/6J mice (Charles River), microinjected with synthetic gRNAs (Integrated DNA Technologies) and recombinant Cas9 protein (Integrated DNA Technologies), and subsequently transferred into pseudo- pregnant C57BL/6J mice. Resulting F0 offspring mice were screened for successful deletion by PCR amplification of wildtype or knockout alleles using the following primers: I1/r1- ExAB knockout (forward: CATAGGCCCAGGTTA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 875, 200]]<|/det|> +GGCCTGAAg,reverse: GAATGTGAACCCTCGTCGACCTG), Il1rl1- ExAB wildtype (forward: CCATTAGCGTGTCTCTACTACGACAg, reverse: GAATGTGAACCCTCG TGTGACGTG), Il1rl1- ExC knockout (forward: GAAGGCTAGTGGCTTCTGTGAGCA T, reverse: CACAGTCTCAGGCAAGCTGCCTG), Il1rl1- ExC wildtype (forward: GAA GGCTAGTGGCCTTCTGTGAGCAT, reverse: TCAGCAATAACCAGCCATCGAGG). + +<|ref|>sub_title<|/ref|><|det|>[[118, 221, 388, 239]]<|/det|> +## Viruses and LCMV infection + +<|ref|>text<|/ref|><|det|>[[115, 244, 882, 535]]<|/det|> +LCMV- WE and LCMV- CI13 strains were propagated on L929 or BHK- 21 cells, respectively. Viral titers in stock solutions were determined by immunofocus assay on MC57G cells as described before83. In brief, MC57G cells were plated with virus stock dilutions and subsequently overlaid with \(2\%\) methylcellulose. After \(48h\) of incubation at \(37^{\circ}C\) , the confluent monolayer of cells was fixed with \(4\%\) formaldehyde, permeabilized with Triton X- 100 ( \(1\%\) , v/v), and stained with antibodies against LCMV nucleoprotein. After a secondary staining step with peroxidase conjugated anti- rat IgG antibody, foci were developed by 20 min incubation with OPD substrate. Mice were infected intravenously (i.v.) with either 200 plaque forming units (PFU) of LCMV- WE (mixed bone- marrow chimera experiments), 200 PFU LCMV- CI13 (adoptive cotransfer experiments, LCMV- CI13 wt or C6 variant where indicated), or \(2\times 10^{6}\) PFU of LCMV- WE in minimal essential medium (MEM, Thermo Scientific). + +<|ref|>sub_title<|/ref|><|det|>[[118, 555, 356, 573]]<|/det|> +## Adoptive T cell transfers + +<|ref|>text<|/ref|><|det|>[[115, 578, 882, 893]]<|/det|> +For adoptive transfer experiments, P14, Il1rl1- P14, Il1rl1- ExAB- P14, Smarta, Il1rl1- Smarta, or Il1rl1- ExAB- Smarta T cells expressing the congenital markers CD45.1 or CD90.1 were enriched in a negative selection approach. To this end, splenocytes of donor mice were stained with biotinylated antibodies against mouse CD11b (M1/70, eBioscience), CD11c (N418, eBioscience), CD19 (1D3, eBioscience), CD25 (7D4, eBioscience), Gr- 1 (RB6- 8C5, eBioscience), NK1.1 (PK136, eBioscience), CXCR3 (CXCR3- 173, eBioscience), and CD8a (53- 6.7, eBioscience, for isolation of Smarta T cells) or CD4 (RM4- 5, for isolation of P14 T cells) followed by incubation with anti- biotin microbeads (Miltenyi Biotec). Subsequently, labeled cells were depleted by magnetic activated cell sorting (MACS) using LS columns (Miltenyi Biotec). \(5\times 10^{4}\) T cells (single transfers experiments), \(1\times 10^{3}\) P14 T cells or \(2\times 10^{4}\) Smarta T cells (cotransfer experiments) were transferred i.v. into C57BL/6J mice. Recipients were infected 1- 2 days post transfer and analyzed at indicated timepoints post infection. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[119, 85, 404, 102]]<|/det|> +## Mixed bone marrow chimeras + +<|ref|>text<|/ref|><|det|>[[117, 108, 882, 275]]<|/det|> +To generate mixed bone marrow chimeras, CD45.1+/+ wildtype recipients were lethally irradiated (two doses of 5.5 gray given in a 6- hour interval). One day later, recipients were reconstituted with a 1:1 mixture of CD45.1+/+ wildtype and CD45.2+/+ knockout bone- marrow cells and splenocytes. After 8 weeks of hematopoietic reconstitution, CTL frequencies of respective donor populations were determined in blood, and mice were infected with LCMV- WE (200 PFU i.v.). Data were analyzed on d10 p.i. and normalized to CTL frequencies before infection. + +<|ref|>sub_title<|/ref|><|det|>[[119, 295, 328, 313]]<|/det|> +## Lymphocyte isolation + +<|ref|>text<|/ref|><|det|>[[115, 319, 882, 608]]<|/det|> +To isolate lymphocytes, spleens were mechanically disrupted and filtered through 70 \(\mu \mathrm{m}\) strainers. Erythrocytes were lysed by incubation in erythrocyte lysis buffer (10 mM KHCO3, 155 mM NH4Cl, 0.1 mM EDTA, pH 7.5). Livers were collected in PBS/BSA, meshed and centrifugated at 30 g for 2 min to remove debris. Supernatants were subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich) and lymphocytes were collected at the gradient interphase. Lungs were cut into small pieces and digested with Collagenase D (0.1 U/ml) in RPMI1640 (supplemented with 10% FCS and 15mM HEPES) for 1 h at 37°C. Afterwards, lymphocytes were isolated by Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich). To isolate peritoneal cavity cells, 5 ml of cold PBS was injected into the peritoneal cavity of euthanized mice. After a brief massage of the peritoneum, cell- containing liquid was collected and subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich). + +<|ref|>sub_title<|/ref|><|det|>[[118, 629, 269, 646]]<|/det|> +## Flow cytometry + +<|ref|>text<|/ref|><|det|>[[113, 652, 882, 894]]<|/det|> +Surface stainings of purified lymphocytes were performed using different combinations of following anti- mouse antibodies diluted in PBS: CD4 PerCP- Cyanine5.5 (RM4- 5, eBioscience), FITC (RM4- 5, eBioscience), Pacific Blue (YTS19.1, DRFZ inhouse production) or APC- Cy7 (GK1.5, BD); CD8a PE (53- 6.7, BD), BV500 (53- 6.7, BD), BV605 (53- 6.7, Biolegend), BV786 (53- 6.7, Biolegend) PerCP- Cyanine5.5 (53- 6.7, BD) or eFluor450 (53- 6.7, eBioscience); Va2 FITC (B20.1, BD); CD19 Cy5 (ID3, DRFZ inhouse production), APC- Fire750 (6D5, Biolegend) or PerCP- Cyanine5.5 (eBio1D3, eBioscience); B220 APC- Fire750 (RA3- 6B2, Biolegend); CD90.1 Pacific Blue (Ox- 7, DRFZ inhouse production); CD90.2 FITC (53- 2.1, Biolegend); CD45.1 Pacific Blue (A20, Biolegend), FITC (A20, Biolegend) or PE (A20, BD); CD45.2 APC (104, BD), + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[75, 80, 884, 870]]<|/det|> +1021 V500 (104, BD), BV510 (30- F11, Biolegend) or APC- Fire750 (104, Biolegend); CD44 1022 APC- Cy7 (IM7, Biolegend) or BV450 (IM7, Biolegend); CD62L PE- Cy7 (MEL- 14, 1023 eBioscience); KLRG1 FITC (2F1, eBioscience) or PerCP- Cy5.5 (2F1/KLRG1, 1024 Biolegend); CD127 PE- Cy7 (A7R34, eBioscience) or BV421 (A7R34, Biolegend); 1025 NK1.1 PE (PK136, BD); GR- 1 A488 (RB6- 8C5, DRFZ inhouse- production) or APC- 1026 Cy7 (RB6- 8C5, Biolegend); Siglec- F PE (E50- 2440, BD); c- kit (2B8, eBioscience); 1027 FcεRla FITC (MAR- 1, eBioscience) or APC- Cy7 (MAR- 1, Biolegend); MHC- II Cy5 1028 (M5/114, DRFZ inhouse production); CD11c APC- Fire750 (N418, Biolegend); CD11b 1029 PE- Cy7 (M1/70, BD) or APC- Fire750 (M1/70, Biolegend); Ly6G PerCP- Cy5.5 (1A8, 1030 Biolegend); F4/80 PE- Cy7 (BM8, Biolegend) or APC- Fire750 (BM8, Biolegend); ST2 1031 BV421 (DIH9, Biolegend). Unspecific staining was minimized by blocking with rat 1032 immunoglobulin G (IgG, Jackson ImmunoResearch) and anti- mouse CD16/32 (2.4G2, 1033 DRFZ inhouse production) prior to staining. Dead cells were labelled using Zombie 1034 Aqua or Zombie NIR fixable live/dead staining reagents (Biolegend) or by adding 1035 propidium iodide (PI) prior to acquisition. For detection of ST2 on T cells, lymphocytes 1036 were first stained with digoxigenin- conjugated antibody against ST2 (DJ8, 1037 mdBioscience), followed by a secondary staining with PE- or APC- conjugated anti- 1038 digoxigenin Fab fragments (Roche). Further, stainings were enhanced by two- rounds 1039 of PE- or APC- FASER amplification (Miltenyi Biotec). To identify LCMV- specific T 1040 cells, lymphocytes were stained with LCMV GP33- 41 or NP396- 404 peptide- loaded MHC 1041 class I (H2- Db) tetramers (PE- or APC- conjugated, MBL or NIH Tetramer Core 1042 Facility) for 30 min at 37°C. For detection of transcription factors or Ki67 expression, 1043 surface- stained cells were fixed and stained using the FoxP3 staining buffer set 1044 (Thermo Scientific). Briefly, cells were fixed with 1x fixation/permeabilization reagent 1045 for 30 min at 4°C and washed with permeabilization buffer. Subsequently, cells were 1046 stained with antibodies against mouse T- bet BV421 (4B10, Biolegend) or PE (4B10, 1047 eBioscience); GATA- 3 eFluor660 (TWAJ, eBioscience), FoxP3 eFluor450 (FJK- 16s, 1048 eBioscience), Ki67 FITC (SolA15, eBioscience), or with respective isotype control 1049 antibodies: mouse IgG1 kappa BV421 (MOPC- 21, Biolegend, T- bet isotype) or mouse 1050 IgG1 kappa PE (P3.6.2.8.1, eBioscience, T- bet isotype) or rat IgG2b kappa eFluor660 1051 (10H5, eBioscience, GATA- 3 isotype) diluted in permeabilization buffer for 30 min at 1052 4°C. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 883, 475]]<|/det|> +For flow- cytometric detection of cytokines, lymphocytes were restimulated with phorbol myristate acetate (PMA) (5 ng/ml, Sigma- Aldrich) and ionomycin (5 μg/ml, Sigma- Aldrich), recombinant LCMV GP33- 41 (Charité Berlin, 1 μg/ml) or LCMV GP64- 79 (Charité Berlin, 1 μg/ml) for 4 h at 37°C. After 35 min, brefeldin A (5 μg/ml, Sigma- Aldrich) was added to prevent secretion of produced cytokines. After the stimulation period, cells were labelled with surface- antibodies and fixable live/dead staining reagents, followed by fixation in 2% paraformaldehyde for 10 min at RT. Intracellular cytokines were stained with antibodies against IFN- γ eFluor450 (XMG1.2, eBioscience) or APC (XMG1.2, Biolegend); IL- 2 APC (JES6- 5H4, BD); TNF eFluor450 (MP6- XT22, eBioscience); Granzyme A PE (3G8.5, Biolegend); Granzyme B APC (QA16A02, Biolegend); Perforin PE (S16009B, Biolegend); IL- 4 PE (11B11, eBioscience); IL- 13 Alexa Fluor488 (eBio13A, eBioscience); IL- 10 APC (JES5- 16E3) in PBS containing 0.05% Saponin for 30 min at 4°C and washed before acquisition. Cells were acquired on a Canto II or LSRFortessa flow- cytometer (BD). Flowcytometric sorting was performed on Aria and Aria II devices (BD). Cell numbers were determined using MACSQuant (Miltenyi Biotec) or ImmunoSpot (CTL) analyzers. + +<|ref|>sub_title<|/ref|><|det|>[[118, 494, 456, 512]]<|/det|> +## Legendplex cytometric bead assay + +<|ref|>text<|/ref|><|det|>[[115, 515, 883, 832]]<|/det|> +To assess cytokine production by T cells in response to IL- 33, 5x105 T cells were stimulated in 48- well plates with IL- 33 (R&D, 10 ng/ml) for 24 h at 37°C. Afterwards, individual wells were harvested and centrifuged for 5 min at 350 g. To obtain serum, blood of individual mice was collected using yellow microtainers (BD). Serum and cell- free supernatant were frozen and stored at - 80°C until analysis. Cytokine content was measured using LEGENDplex bead- based immunoassays (Biolegend) according to manufacturer's instructions. In brief, samples and serial dilutions of the standard were incubated for 2 h on a plate shaker set to 900 rpm with premixed, antibody- coated beads to capture cytokines. Afterwards, beads were washed twice with washing buffer and stained with biotinylated detection antibodies. After 1 h incubation on a plate shaker at 900 rpm, streptavidin- PE was added for additional 30 min. Beads were washed twice and acquired at a Canto II flow- cytometer (BD). Cytokine concentration was extrapolated from standard titrations. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 84, 327, 101]]<|/det|> +## Mouse T cell cultures + +<|ref|>text<|/ref|><|det|>[[115, 106, 883, 600]]<|/det|> +Mouse T cell culturesNaive T cells from spleens of indicated mice were pre- enriched by staining with biotinylated antibodies against CD8a (53- 6.7, eBioscience) or CD4 (RM4- 5, eBioscience), followed by incubation with anti- biotin microbeads (Miltenyi Biotec) and subsequent separation by MACS using LS columns (Miltenyi Biotec). Following enrichment, naive (CD62L+ CD44- CD25- CXCR3-) CD8+ or CD4+ T cells were flow- cytometrically sorted and differentiated in the presence of irradiated TCRβδ- splenocytes and antibodies against CD3ε (145- 2C11, eBioscience) and CD28 (37.51, both 2.5 μg/ml). When naive T cells were isolated from LCMV- TCRtg mice, cognate LCMV GP33- 41 (P14 mice) or GP64- 79 peptide (Smarta mice, both 1μg/ml) where added instead. T cells were cultivated in RPMI1640 + GlutaMax I (Thermo Scientific) medium supplemented with fetal calf serum (FCS) (10% v/v, Thermo Scientific), penicillin (100 U/ml, Thermo Scientific), streptomycin (100 μg/ml, Thermo Scientific), gentamycin (10 μg/ml, Thermo Scientific), and β- mercaptoethanol (50 ng/ml, Sigma- Aldrich). For CTL and Th1 differentiation, IL- 12 (5 ng/ml), IL- 2 (5 ng/ml, all Miltenyi Biotec), and anti- IL- 4 (11B11, 10 μg/ml, DRFZ inhouse production) were added. For Th2 differentiation, IL- 4 (5 ng/ml), IL- 2 (5 ng/ml, all Miltenyi Biotec), anti- IL- 12 (C18.2, 10 μg/ml), and anti- IFN- γ (XMG1.2, 10 μg/ml, all DRFZ inhouse production) were added. T cells were split after 2- 3 days of culture in a 1:3 ratio with fresh medium containing IL- 2 (5 ng/ml), harvested at day 5 of culture using Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich), and cultivated for additional 5 days in identical culture conditions. + +<|ref|>sub_title<|/ref|><|det|>[[118, 620, 437, 638]]<|/det|> +## Retroviral transduction of T cells + +<|ref|>text<|/ref|><|det|>[[115, 644, 883, 910]]<|/det|> +Retroviral transduction of T cellsFor cloning of shRNA expression vectors, the following sense- and antisense- shRNA sequences were ordered as phosphorylated oligos with a 5' Sal- I restriction overhang (Eurofins Genomics) and annealed by subjecting equimolar amounts of oligos diluted in oligo annealing buffer (100 mM Tris- HCl, 1 M NaCl, 10 mM EDTA, pH 7.5) to a decreasing temperature gradient (95°C to 25°C with 1°C/min): shRNA- ExAB: (Sense: TGGCCTAGAATGAGAAAGGAAATTCAAGAGATTTCCTTCCA TTCTGAGCCTTTTTTC, antisense: TCGAGAAAAAAGGCTCAGAATGGAAAGGAA TTCTCTGAATTCCTTTCCTATTCTGAGCCA); shRNA- Ex1a (Sense: TGTTGGAAGC TGAGGAATAATTCAAGAGATTATTCCTCAGCTTGCAAACTTTTTTC, antisense: TC GAGAAAAAAGTTTCGAAGCTGAGGAATAATCTCTTGAATTATTCCTCAGCTTGCAA CA). PQCXIX- GFP target vector84 was digested by Sall and Hpal restriction enzymes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 881, 523]]<|/det|> +(Thermo Scientific) and simultaneously dephosphorylated with FastAP alkaline phosphatase (Thermo Scientific). Annealed oligos were ligated using T4 Ligase according to standard protocols (NEB). Heat inactivated ligation reactions were directly used for heat-shock transformation into Oneshot TOP10 chemically competent E.coli (Thermo Scientific). Single transformed bacterial clones were selected on LB- agar plates (MP Biomedicals) containing ampicillin (100 \(\mu \mathrm{g / ml}\) , Sigma-Aldrich), and plasmid DNA was prepared using QIAprep Spin Plasmid Maxi or Midi kits (Qiagen. Correct plasmid sequences were verified by Sanger- sequencing (Eurofins Genomics). Virus particles were generated by co- transfection of HEK293T cells with shRNA- containing vectors and packaging plasmids pCGP and pECO85 using T5 Transporter reagent. For retroviral transduction, mouse T cells were activated in the presence of irradiated APCs and antibodies against CD3ε and CD28 described above. 36- 48 h after plating, culture medium was temporarily replaced with virus- containing supernatant, polybrene (8 \(\mu \mathrm{g / ml}\) , Sigma-Aldrich) was added and plates were centrifuged for 90 min at 450 g at RT. T cells were incubated at 37°C for 6- 8 h. Afterwards, viral supernatant was replaced with conditioned cell culture medium and cells were split in a 1:3 ratio with fresh IL- 2- containing medium. Transduced T cells were analyzed between day 5 and day 7 of culture. + +<|ref|>sub_title<|/ref|><|det|>[[118, 542, 331, 560]]<|/det|> +## Human T cell cultures + +<|ref|>text<|/ref|><|det|>[[115, 565, 882, 906]]<|/det|> +Human peripheral blood which was obtained from the German Red Cross (DRK Berlin, Germany; Ethics approval EA1/149/12). For isolation of T cells, blood was first subjected to Ficoll- Paque PLUS density centrifugation (1.077 g/ml, Cytiva). Interphases were collected, stained with anti- CD4 microbeads (Miltenyi Biotec), and separated using LS columns (Miltenyi Biotec). CD4+ T cell- depleted fractions were used for a MACS enrichment of CD8+ T cells using anti- CD8a microbeads (Miltenyi Biotec). CD4- enriched fractions were stained with antibodies against human CD4 Pacific Blue (TT1, DRFZ inhouse production), CXCR3 FITC (G025H7, Biolegend), and CRTH2 biotin (BM16, Miltenyi Biotec) for 15 min at 4°C followed by a secondary staining with Streptavidin PE (BD). CD8- enriched fractions were stained with antibodies against human CD8 (GN11/134.7, DRFZ inhouse production), CD56 APC (REA196, Miltenyi Biotec), CD62L PE- Cy7 (DREG- 56, Biolegend), and CD45RA PE (HI100, DRFZ inhouse production). In vivo differentiated Th1 cells were sorted as CD4+ CXCR3+ CRTH2+, and Th2 cells were sorted as CD4+ CXCR3+ CRTH2+. CD8+ + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[66, 82, 883, 428]]<|/det|> +1149 effector/effector memory T cells were sorted as CD8+ CD56- CD45RA- CD62L- . For 1150 activation of human T cells, suspension culture plates were coated with antibodies 1151 against human CD3ε (OKT3, 2.5 μg/ml, Miltenyi Biotec) and CD28 (15E3, 2.5 μg/ml, 1152 Miltenyi Biotec), and sorted T cells were plated in RPMI1640 + GlutaMax I (Thermo 1153 Scientific) medium supplemented with FCS (10% v/v, Thermo Scientific), penicillin 1154 (100 U/ml, Thermo Scientific), streptomycin (100 μg/ml, Thermo Scientific), 1155 gentamycin (10 μg/ml, Thermo Scientific), and \(\beta\) - mercaptoethanol (50 ng/ml, Sigma1156 Aldrich). To CTL and Th1 cultures, IL- 12 (10 ng/ml, R&D Systems), IL- 2 (10 ng/ml, 1157 R&D Systems), and anti- IL- 4 (7A3- 3, 10 μg/ml, Miltenyi Biotec) were added, while Th2 1158 were cultured in the presence of IL- 4 (10 ng/ml, R&D Systems), IL- 2 (10 ng/ml, R&D 1159 Systems), anti- IL- 12 (C8.6, 10 μg/ml, Miltenyi Biotec) and anti- IFN- γ (45- 15, 10 μg/ml, 1160 Miltenyi Biotec). Cells were withdrawn from coated plates after 24 h of activation, split 1161 after 3 days in a 1:3 ratio with fresh medium containing IL- 2 (10 ng/ml, R&D Systems), 1162 and analyzed on day 5 of culture. + +<|ref|>sub_title<|/ref|><|det|>[[118, 446, 390, 465]]<|/det|> +## RNA isolation and qRT-PCR + +<|ref|>text<|/ref|><|det|>[[66, 470, 883, 910]]<|/det|> +To isolate RNA for qRT- PCR analysis or bulk RNA sequencing, \(10^{5} - 10^{6} \mathrm{T}\) cells were harvested, lysed in RA- 1 buffer (Macherey & Nagel). Total RNA was purified using the Nucleospin RNA XS Micro kit (Macherey & Nagel) according to manufacturer's instructions, without addition of carrier RNA. For qRT- PCR analysis, RNA was transcribed into cDNA utilizing Taqman Reverse Transcription Reagents (Applied Biosystems). cDNA was then subjected to qRT- PCR analysis using PowerUp SYBR Green or Taqman Fast Advanced Mastermix reagents (Applied Biosystems) and the following primers or probes: ExAB (forward: TGGTGAACAGTTCAGTTCTTCACAG, reverse: ACGATTTACTGCCCTCCGTAAC), Ex1a (forward: ATTCTGGTGGTTCGA AGCTGAGGA, reverse: ACGATTTACTGCCCTCCGTAAC), Ex1b (forward: ATGAG ACGAAGGAGCGCCAAG, reverse: ACGATTTACTGCCCTCCGTAAC), ST2 (forward d: AAGGCACACCATAAGGCTGA, reverse: TCGTAGAGCTTGGCCATCGTT), Hprt: (forward: CATAACCTGGTTCATCATCGc, reverse: TCCTCCTCAGACCGCTTTT), hIL1RL1- ExAB (forward: ATACAGACGTCAGAACTGAGAC, reverse: GCATACGAC AGTGAAGGTCA), hIL1RL1- Ex1a (forward: GAACTCAAGTACAACCCAATGA, reverse: GCATACGACAGTGAAGGTCa), IL1RL1: Hs00249384_m1 (Thermo Scientific) GAPDH: Hs02758991_g1 (Thermo Scientific). Amplifications were performed in triplicates by using a QuantStudio 7 device (Applied Biosystems) and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 85, 880, 130]]<|/det|> +expression levels were quantified with the \(\Delta \Delta \mathrm{Ct}\) - method by normalizing target gene expression to levels of Hprt (mouse) or GAPDH (human). + +<|ref|>sub_title<|/ref|><|det|>[[117, 149, 710, 169]]<|/det|> +## Single cell RNA-library preparation, sequencing, and analysis + +<|ref|>text<|/ref|><|det|>[[115, 171, 883, 810]]<|/det|> +Single cell suspensions of \(\mathrm{CD90^{+}}\) \(\mathrm{CD8^{+}}\) \(\mathrm{CD44^{+}}\) T cells were obtained by flow- cytometrical sorting of CD19- depleted splenocytes from LCMV- infected wildtype and \(IIIrI1 - ExAB^{- / - }\) mice at day 7 p.i. with LCMV- WE (2x10 \(^6\) PFU). Sorted T cells of individual mice were barcoded using TotalSeq- C anti- mouse Hashtags (anti- mouse Hashtag 1, 2 and 3, all Biolegend). T cells of each genotype were then pooled and applied to the 10x Genomics workflow for cell capturing. For the preparation of scRNA gene expression (GEX), TCR and CiteSeq libraries the Chromium Next GEM Single Cell 5' Library & Gel Bead Kit v1.1 as well as the Chromium Single Cell 5' Feature Barcode Library Kit (10x Genomics) were used. After cDNA amplification the CiteSeq libraries were prepared separately using the Single Index Kit N Set A (10x Genomics). TCR target enrichment was performed using the Chromium Single Cell V(D)J Enrichment Kit for mouse T cells (10x Genomics). Final GEX and TCR libraries were obtained after fragmentation, adapter ligation and final Index PCR using the Single Index Kit T Set A. The Qubit dsDNA HS assay kit (Life technologies) was used for library quantification and fragment sizes were determined using the Fragment Analyzer with the NGS Fragment Kit (1- 6000bp) (Agilent). Sequencing was performed on a NextSeq2000 device (Illumina) using P2 Reagents v3 (200 cycles) with the recommended sequencing conditions for 5' GEX and barcode libraries (read1: 26nt, read2: 98nt, index1: 8nt, index2: n.a.) and on a NextSeq500 device (Illumina) using a Mid Output v2 Kit (300 cycles) for TCR libraries (read1: 150nt, read2: 150nt, index1: 8nt, index2: n.a., 20% PhiX spike- in). Raw data were processed using cellranger- 3.1.0 with refdata- gex- mm10- 2020- A and refdata- cellranger- vdj_GRCm38_alts_ensembl- mouse- 2.2.0 as reference. Mkfastq, count and vdj were used in default parameter settings with 3000 expected cells for demultiplexing, detection of intact cells, quantification of gene expression, antibody capture as well as assembly and quantification of T cell receptor sequences. + +<|ref|>text<|/ref|><|det|>[[117, 813, 881, 905]]<|/det|> +The cellranger output was further analyzed in R using the Seurat package (version 4.0.0) \(^{86}\) . In particular, the CiteSeq data, hashtag sequences, detxramer counts of three individual \(IIIrI1 - ExAB^{- / - }\) and wildtype mice were imported and combined. Centered log ratio transformation was used for normalization. Seurat's default method was used for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 882, 250]]<|/det|> +scaling. Features with correlation coefficients \(>0.85\) to Gm42418, Malat1, AY036118, and Lars2 were removed from the count matrix. Hashtag demultiplexing (representing the 3 biological replicates per genotype) was performed based on Seurats HTODemux with the parameter positive- quantile at 0.99. Doublets and untagged cells were filtered out. Cells with expression values for Cd8a or Cd8b1 and Cd3g, Cd3d, or Cd3e, with \(>200\) and \(< 4500\) features, and \(< 10\%\) UMI for mitochondrial genes were kept for further analysis. + +<|ref|>text<|/ref|><|det|>[[115, 253, 882, 767]]<|/det|> +After ranking by residual variance, 3000 variable genes were determined. The genes encoding TCR variable regions (Trav, Trbv, Trdv, Trgv) were removed. 30 principal components (PC's) were computed and stored. Uniform Manifold Approximation and Projection (UMAP) and t- distributed stochastic neighbor embedding (TSNE) were run using the first 15 PC's. Transcriptionally similar clusters were identified using shared nearest neighbor modularity optimization, with a resolution of 0.35. For visualization, cells of the llrl1- ExAB- condition were down- sampled to match the number of cells in the WT condition. Signature genes were identified using the FindAllMarkers function in default parameter settings (only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25). Heatmaps and Dotplots for the single cell data were plotted with Seurat's DoHeatmap and DotPlot function, respectively, using default settings. Identified clusters were annotated based on the expression of key markers for CD8+ T cell subsets and cell functions. Two KIng1 expressing clusters were merged to the SLEC cluster. Further, two clusters were merged to form the "HighMito"- Cluster based on their expression of mitochondrial genes. After combining the stated clusters, signature genes were identified again using the same method as described above. Cluster size was defined as the number of cells in one cluster. Relative cluster sizes were calculated by analyzing the number of cells in one cluster per genotype divided by the total number of cells. For feature plots, the expression of single features was plotted on the UMAP by using Seurat's default function FeaturePlot with the option "keep.scale = 'all'". + +<|ref|>text<|/ref|><|det|>[[115, 771, 882, 915]]<|/det|> +For the analysis of the TCR the immune profiles were integrated using identical cellular barcodes. For cells with more than one contig for the heavy or light TCR chain the most abundant, productive contig was chosen. Cell numbers were equalized by subsampling the larger condition. Cells without TCR annotation were excluded from the analysis. The R package immunarch (version 0.6.6) \(^{87}\) was used for clonality analysis after downsampling to the wildtype condition. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 283, 103]]<|/det|> +## RNA sequencing + +<|ref|>text<|/ref|><|det|>[[110, 102, 883, 725]]<|/det|> +Smarta and P14 T cells used for bulk RNA- Seq experiments were activated and differentiated as described above. To enrich for ST2- expressing T cells, CTLs, Th1 and Th2 cells were stained for ST2 at day 10 of culture and \(\mathsf{ST2^{+}}\) T cells were flowcytometrically sorted. For global analysis of IL- 33- responsive genes, T cells were harvested at day 10 of culture and rested for 3 days in the presence of IL- 2 (5 ng/ml) and IL- 7 (5 ng/ml, both Miltenyi Biotec), without irradiated splenocytes or cognate peptide. At day 13, IL- 12 (5 ng/ml) was added to CTLs and Th1 conditions to trigger ST2 expression. At day 14 of culture, T cells were subjected to Histopaque density centrifugation (1.083 g/ml, Sigma- Aldrich) and stimulated in conditioned medium with IL- 33 (10 ng/ml, R&D Systems) for 2 h. When used for RNA sequencing, quality of the isolated RNA was assessed with a Fragment Analyzer System (Agilent). All processed samples showed high RNA integrity (RQN \(>8\) ). cDNA libraries were prepared using the Smart- Seq v4 mRNA Ultra Low Input RNA Kit (Clontech) with up to 10 ng of RNA (IL- 33 stimulated T cells) or TrueSeq stranded total RNA library kit (Illumina) with up to 1 \(\mu \mathrm{g}\) of RNA (ST2- enriched T cells) according to manufacturer's instructions. Paired- end sequencing (2x75 bp) of cDNA libraries was performed on an Illumina NextSeq500 device using the NextSeq 500/550 High output Kit v2. Obtained reads were mapped to the mm10 or hg19 genome (annotation releases: GRCh37.p13 and GRCm38.p6) using Tophat288 and Bowtie289 with very- sensitive settings. Read counts were determined with featureCounts90. DESeq291 was used in RStudio for differential gene expression analysis. The DESeq2 count matrix was pre- filtered for genes with \(\geq\) 100 summarized read counts across the analyzed samples. A gene was considered as differentially expressed when \(\log 2\mathrm{FC} > 1.0\) and P adjusted \(< 0.01\) . AnnotationDbi92, EnhancedVolcano93, ComplexHeatmap94, heatmap95 and ggplot296 were used in RStudio for data visualization. + +<|ref|>text<|/ref|><|det|>[[118, 720, 881, 790]]<|/det|> +For PCA and sample distance calculation, a blind variance stabilizing transformation (VST) was performed on the unnormalized counts across all samples. Sample distance plots are based on pairwise calculation of the Pearson correlation. + +<|ref|>sub_title<|/ref|><|det|>[[118, 810, 500, 829]]<|/det|> +## GSEA and Overrepresentation Analysis + +<|ref|>text<|/ref|><|det|>[[118, 833, 881, 878]]<|/det|> +For gene set enrichment analysis (GSEA), the R package clusterProfiler97 was used. A gene list ranked by log2FoldChange containing all expressed genes served as input + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 125]]<|/det|> +and was tested for enrichment of biological- process gene sets from the gene ontology resource98,99. + +<|ref|>text<|/ref|><|det|>[[115, 131, 881, 300]]<|/det|> +For overrepresentation analysis, differentially expressed genes were split into up- and downregulated genes. Over- representation analysis was performed against biological- process gene sets from the gene ontology resource using a one- sided version of the Fisher's exact test. All expressed genes in the respective conditions were used as a background gene list. The results were simplified in order to reduce overlaps between ontology terms by using the clusterProfiler::simplifyGO function with a cutoff of 0.7. + +<|ref|>sub_title<|/ref|><|det|>[[116, 319, 565, 338]]<|/det|> +## Analysis of alternative transcription start sites + +<|ref|>text<|/ref|><|det|>[[115, 343, 881, 510]]<|/det|> +Raw RNA- Seq reads of indicated T cell subsets were aligned using hisat2 (version 2.2.1)100 and assembled using Cufflinks (version 2.2.1)71 in RABT mode with EnsEMBL annotation release 67. Assemblies were then merged into a new reference annotation with the public reference using the cuffmerge function. The resulting annotation was used for an analysis with the R package ProActiv42. The getAlternativePromoters function was used with standard parameters except for minAbs = 5. Only results with FDR<0.01 were considered. + +<|ref|>sub_title<|/ref|><|det|>[[115, 529, 833, 549]]<|/det|> +## Processing of published Chip-Seq, ATAC-Seq data, and conservation data + +<|ref|>text<|/ref|><|det|>[[115, 553, 881, 820]]<|/det|> +Fastq files of published RNA- Seq datasets (ILCS2: Run ID SRR7549295, mast cells: Run ID SRR6155875) were obtained from the NCBI Sequence read archive and aligned using hisat2 with default settings100. ChIP- and ATAC- Seq data sets were downloaded from the NCBI GEO Database, if required crossmapped to the mm10 genome using CrossMap101 (ChIP- Seq: GSM550303, GSM998272, GSM523226; ATAC- Seq: GSE120532, GSE111902). PhyloP conservation tracks were downloaded from the UCSC Sequence and Annotation database43,102. JASPAR motif analysis was performed on indicated sequences using the web- based JASPAR Scan tool73. FANTOM5 CAGE- Seq data and CAGE- associated transcript data were downloaded from the FANTOM5 data resource45,103,104. All NGS data tracks were visualized in the integrated genome viewer (IGV)105. + +<|ref|>sub_title<|/ref|><|det|>[[116, 840, 481, 858]]<|/det|> +## Quantification and statistical analysis + +<|ref|>text<|/ref|><|det|>[[115, 863, 880, 907]]<|/det|> +Statistical analysis was performed using GraphPad Prism (v9.0.0). Normal distribution was tested using Shapiro- Wilk and Kolmogorov–Smirnov tests. Unpaired or paired + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 83, 883, 226]]<|/det|> +1313 two- tailed Student's t- tests were used when two groups were compared with respect 1314 to one parameter. More than two groups were analyzed by One- way ANOVA with 1315 Tukey's post- hoc test for multiple comparison. For comparisons of more than one 1316 parameter between two or more groups, Two- way ANOVA with Tukey's post- hoc tests 1317 (unpaired samples) or Two- way repeated measurement ANOVA with Sidak's post- hoc 1318 tests (paired samples) were performed as indicated in the figure legends. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[118, 84, 508, 101]]<|/det|> +# References Figure legends and Methods + +<|ref|>text<|/ref|><|det|>[[118, 108, 875, 159]]<|/det|> +Trapnell, C. et al. Transcript assembly and quantification by RNA- Seq reveals unannotated transcripts and isoform switching during cell differentiation. Nat Biotechnol 28, 511- 515, doi:10.1038/nbt.1621 (2010). + +<|ref|>text<|/ref|><|det|>[[118, 172, 851, 222]]<|/det|> +Pollard, K. S., Hubisz, M. J., Rosenbloom, K. R. & Siepel, A. Detection of nonneutral substitution rates on mammalian phylogenies. Genome Res 20, 110- 121, doi:10.1101/gr.097857.109 (2010). + +<|ref|>text<|/ref|><|det|>[[118, 235, 822, 285]]<|/det|> +Fornes, O. et al. JASPAR 2020: update of the open- access database of transcription factor binding profiles. Nucleic Acids Res 48, D87- D92, doi:10.1093/nar/gkz1001 (2020). + +<|ref|>text<|/ref|><|det|>[[118, 298, 858, 348]]<|/det|> +Ricardo- Gonzalez, R. R. et al. Tissue signals imprint ILC2 identity with anticipatory function. Nat Immunol 19, 1093- 1099, doi:10.1038/s41590- 018- 0201- 4 (2018). + +<|ref|>text<|/ref|><|det|>[[118, 361, 795, 411]]<|/det|> +Gentek, R. et al. Hemogenic Endothelial Fate Mapping Reveals Dual Developmental Origin of Mast Cells. Immunity 48, 1160- 1171 e1165, doi:10.1016/j.immuni.2018.04.025 (2018). + +<|ref|>text<|/ref|><|det|>[[118, 424, 860, 475]]<|/det|> +Pircher, H., Burki, K., Lang, R., Hengartner, H. & Zinkernagel, R. M. Tolerance induction in double specific T- cell receptor transgenic mice varies with antigen. Nature 342, 559- 561, doi:10.1038/342559a0 (1989). + +<|ref|>text<|/ref|><|det|>[[118, 488, 874, 570]]<|/det|> +Oxenius, A., Bachmann, M. F., Zinkernagel, R. M. & Hengartner, H. Virus- specific MHC- class II- restricted TCR- transgenic mice: effects on humoral and cellular immune responses after viral infection. Eur J Immunol 28, 390- 400, doi:10.1002/(SICI)1521- 4141(199801)28:01<390::AID- IMMU390>3.0. CO;2- O (1998). + +<|ref|>text<|/ref|><|det|>[[118, 583, 870, 634]]<|/det|> +Kaplan, M. H., Sun, Y. L., Hoey, T. & Grusby, M. J. Impaired IL- 12 responses and enhanced development of Th2 cells in Stat4- deficient mice. Nature 382, 174- 177, doi:10.1038/382174a0 (1996). + +<|ref|>text<|/ref|><|det|>[[118, 647, 840, 697]]<|/det|> +Szabo, S. J. et al. Distinct effects of T- bet in TH1 lineage commitment and IFN- gamma production in CD4 and CD8 T cells. Science 295, 338- 342, doi:10.1126/science.1065543 (2002). + +<|ref|>text<|/ref|><|det|>[[118, 711, 860, 760]]<|/det|> +Mombaerts, P. et al. Mutations in T- cell antigen receptor genes alpha and beta block thymocyte development at different stages. Nature 360, 225- 231, doi:10.1038/360225a0 (1992). + +<|ref|>text<|/ref|><|det|>[[118, 774, 848, 824]]<|/det|> +Wefers, B., Bashir, S., Rossius, J., Wurst, W. & Kuhn, R. Gene editing in mouse zygotes using the CRISPR/Cas9 system. Methods 121- 122, 55- 67, doi:10.1016/j.ymeth.2017.02.008 (2017). + +<|ref|>text<|/ref|><|det|>[[118, 838, 860, 888]]<|/det|> +Concordet, J. P. & Haeussler, M. CRISPR: intuitive guide selection for CRISPR/Cas9 genome editing experiments and screens. Nucleic Acids Res 46, W242- W245, doi:10.1093/nar/gky354 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 87, 875, 138]]<|/det|> +1358 83 Battegay, M. et al. Quantification of lymphocytic choriomeningitis virus with an immunological focus assay in 24- or 96- well plates. J Virol Methods 33, 191- 198, doi:10.1016/0166- 0934(91)90018- u (1991). + +<|ref|>text<|/ref|><|det|>[[42, 147, 870, 199]]<|/det|> +1361 84 Stittrich, A. B. et al. The microRNA miR- 182 is induced by IL- 2 and promotes clonal expansion of activated helper T lymphocytes. Nat Immunol 11, 1057- 1062, doi:10.1038/ni.1945 (2010). + +<|ref|>text<|/ref|><|det|>[[42, 209, 840, 261]]<|/det|> +1364 85 Haftmann, C. et al. miR- 148a is upregulated by Twist1 and T- bet and promotes Th1- cell survival by regulating the proapoptotic gene Bim. Eur J Immunol 45, 1192- 1205, doi:10.1002/ejj.201444633 (2015). + +<|ref|>text<|/ref|><|det|>[[42, 272, 830, 307]]<|/det|> +1367 86 Hao, Y. et al. Integrated analysis of multimodal single- cell data. Cell 184, 3573- 3587 e3529, doi:10.1016/j.cell.2021.04.048 (2021). + +<|ref|>text<|/ref|><|det|>[[42, 319, 875, 370]]<|/det|> +1369 87 Nazarov, V. I., Tsvetkov, V. O., Rumynskiy, E., Popov, A. A. & Balashov, I. immunarch: Bioinformatics Analysis of T- Cell and B- Cell Immune Repertoires. (2021). + +<|ref|>text<|/ref|><|det|>[[42, 382, 866, 434]]<|/det|> +1372 88 Kim, D. et al. TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol 14, R36, doi:10.1186/gb- 2013- 14- 4- r36 (2013). + +<|ref|>text<|/ref|><|det|>[[42, 445, 852, 480]]<|/det|> +1375 89 Langmead, B. & Salzberg, S. L. Fast gapped- read alignment with Bowtie 2. Nat Methods 9, 357- 359, doi:10.1038/nmeth.1923 (2012). + +<|ref|>text<|/ref|><|det|>[[42, 492, 860, 544]]<|/det|> +1377 90 Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923- 930, doi:10.1093/bioinformatics/btf656 (2014). + +<|ref|>text<|/ref|><|det|>[[42, 555, 870, 606]]<|/det|> +1380 91 Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA- seq data with DESeq2. Genome Biol 15, 550, doi:10.1186/s13059- 014- 0550- 8 (2014). + +<|ref|>text<|/ref|><|det|>[[42, 618, 870, 653]]<|/det|> +1383 92 Pagès H, C. M., Falcon S, Li N AnnotationDbi: Manipulation of SQLite- based annotations in Bioconductor. (2020). + +<|ref|>text<|/ref|><|det|>[[42, 665, 880, 700]]<|/det|> +1385 93 Blighe, K., Rana, S. & Lewis, M. EnhancedVolcano: Publication- ready volcano plots with enhanced colouring and labeling. (2021). + +<|ref|>text<|/ref|><|det|>[[42, 712, 870, 763]]<|/det|> +1387 94 Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847- 2849, doi:10.1093/bioinformatics/btw313 (2016). + +<|ref|>text<|/ref|><|det|>[[42, 776, 494, 795]]<|/det|> +1390 95 Kolde, R. Pretty Heatmaps. (2019). + +<|ref|>text<|/ref|><|det|>[[42, 806, 845, 841]]<|/det|> +1391 96 Wickham, H. ggplot2: Elegant Graphics for Data Analysis. Springer- Verlag New York (2016). + +<|ref|>text<|/ref|><|det|>[[42, 853, 840, 905]]<|/det|> +1393 97 Yu, G., Wang, L. G., Han, Y. & He, Q. Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS 16, 284- 287, doi:10.1089/omi.2011.0118 (2012). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 84, 864, 120]]<|/det|> +1396 98 Gene Ontology, C. The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 49, D325- D334, doi:10.1093/nar/gkaa1113 (2021). + +<|ref|>text<|/ref|><|det|>[[42, 133, 868, 167]]<|/det|> +1398 99 Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25-29, doi:10.1038/75556 (2000). + +<|ref|>text<|/ref|><|det|>[[42, 180, 848, 230]]<|/det|> +1400 100 Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol 37, 907-915, doi:10.1038/s41587-019-0201-4 (2019). + +<|ref|>text<|/ref|><|det|>[[42, 243, 862, 293]]<|/det|> +1403 101 Zhao, H. et al. CrossMap: a versatile tool for coordinate conversion between genome assemblies. Bioinformatics 30, 1006-1007, doi:10.1093/bioinformatics/bt730 (2014). + +<|ref|>text<|/ref|><|det|>[[42, 306, 842, 340]]<|/det|> +1406 102 Kent, W. J. et al. The human genome browser at UCSC. Genome Res 12, 996-1006, doi:10.1101/gr.229102 (2002). + +<|ref|>text<|/ref|><|det|>[[42, 354, 816, 404]]<|/det|> +1408 103 Lizio, M. et al. Gateways to the FANTOM5 promoter level mammalian expression atlas. Genome Biol 16, 22, doi:10.1186/s13059-014-0560-6 (2015). + +<|ref|>text<|/ref|><|det|>[[42, 418, 848, 468]]<|/det|> +1411 104 Lizio, M. et al. Update of the FANTOM web resource: expansion to provide additional transcriptome atlases. Nucleic Acids Res 47, D752-D758, doi:10.1093/nar/gky1099 (2019). + +<|ref|>text<|/ref|><|det|>[[42, 481, 860, 515]]<|/det|> +1414 105 Robinson, J. T. et al. Integrative genomics viewer. Nat Biotechnol 29, 24-26, doi:10.1038/nbt.1754 (2011). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 614, 178]]<|/det|> +20230124SupplementaryTable1scRNASeq.xlsx20230124SupplementaryTable2RNASeqIL33Stimulation.xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/images_list.json b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c255df5bdbbf8007ff6277c827ce25c28d416617 --- /dev/null +++ b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 | Population, land use and agricultural change with urbanization. These figures demonstrate changes in population, land use and agricultural system with urbanization from 2017 to 2050. a, Geo-distribution of population change. b, Change of population. c, Geo-distribution of land use change, with the same colors as in d. d, Quantity change of land use. e, Geo-distribution of urbanization change. f, Geo-distribution of cropland change. g, Geo-distribution of livestock change. The base map is applied without endorsement from GADM data (https://gadm.org/).", + "footnote": [], + "bbox": [ + [ + 135, + 90, + 765, + 537 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 | Spatial variation of nitrogen loss with urbanization.", + "footnote": [], + "bbox": [ + [ + 230, + 95, + 771, + 441 + ] + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 | Nitrogen budgets within key subsystems and their contribution to N losses reduction with urbanization.", + "footnote": [], + "bbox": [ + [ + 166, + 81, + 848, + 368 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 | Spatial variation of \\(\\mathrm{Nr}\\) species loss with urbanization.", + "footnote": [], + "bbox": [ + [ + 123, + 90, + 795, + 707 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 | Spatial variation of atmospheric PM2.5 concentration and \\(\\mathbf{N}_{\\mathrm{r}}\\) to ocean with urbanization.", + "footnote": [], + "bbox": [ + [ + 142, + 100, + 789, + 388 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 | Costs and benefits of halving \\(\\mathbf{N}_{\\mathrm{r}}\\) pollution with urbanization. a-c, Geographic distribution of implementation cost on waste treatment (a), agricultural optimization (b), and industrial pollution system upgrades (c) to achieve the above mentioned reduction potential of \\(\\mathbf{N}_{\\mathrm{r}}\\) pollution with urbanization. d-f, Geographic distribution of monetized benefits on climate (d), human health (e) and ecosystem (f). g, Geographic distribution of economic benefit from fertilizer saving and yield increase. h, National costs and benefits with urbanization. Blue indicates implementation costs, while red indicates benefits. The base map is applied without endorsement from GADM data (https://gadm.org/).", + "footnote": [], + "bbox": [ + [ + 120, + 90, + 857, + 608 + ] + ], + "page_idx": 27 + } +] \ No newline at end of file diff --git a/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5.mmd b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5.mmd new file mode 100644 index 0000000000000000000000000000000000000000..88025640bd724502e5f4dad3c42ecfb2cf6863f2 --- /dev/null +++ b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5.mmd @@ -0,0 +1,616 @@ + +# Well-managed urbanization could halve nitrogen pollution in China + +Baojing Gu ( \(\boxed{\pm}\) bjgu@zju.edu.cn) Zhejiang University https://orcid.org/0000- 0003- 3986- 3519 + +Ouping Deng Zhejiang University + +Sitong Wang Zhejiang University + +Jiangyou Ran Sichuan Agricultural University + +Shuai Huang Sichuan Agricultural University + +Xiuming Zhang Zhejiang University + +Jiakun Duan Zhejiang University + +Lin Zhang Peking University https://orcid.org/0000- 0003- 2383- 8431 + +Yongqiu Xia Stefan Reis DLR Projekttraeger + +Jiayu Xu Peking University + +Jianming Xu Zhejiang University https://orcid.org/0000- 0002- 2954- 9764 + +Wim de Vries Wageningen University and Research https://orcid.org/0000- 0001- 9974- 0612 + +Mark Sutton NERC Centre for Ecology and Hydrology + +Article + +Keywords: + +Posted Date: July 13th, 2023 + +<--- Page Split ---> + +DOI: https://doi.org/10.21203/rs.3.rs- 3148472/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on January 9th, 2024. See the published version at https://doi.org/10.1038/s41467- 023- 44685- y. + +<--- Page Split ---> + +## Well-managed urbanization could halve nitrogen pollution in China + +Ouping Deng \(^{1,2,\#}\) , Sitong Wang \(^{1,3,\#}\) , Jiangyou Ran \(^{2}\) , Shuai Huang \(^{2}\) , Xiuming Zhang \(^{1}\) , Jiakun Duan \(^{1}\) , Lin Zhang \(^{4}\) , Yongqiu Xia \(^{5}\) , Stefan Reis \(^{6}\) , Jiayu Xu \(^{4}\) , Jianming Xu \(^{1,7}\) , Wim de Vries \(^{8}\) , Mark A. Sutton \(^{9}\) , Baojing Gu \(^{1,3,10*}\) + +\(^{1}\) College of Environmental and Resource Sciences, Zhejiang University, Hangzhou, China. + +\(^{2}\) College of Resources, Sichuan Agricultural University, Chengdu, China. + +\(^{3}\) Policy Simulation Laboratory, Zhejiang University, Hangzhou 310058, China + +\(^{4}\) Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing, China. + +\(^{5}\) Key Laboratory of Soil and Sustainable Agriculture, Changshu National Agr- Ecosystem Observation and Research Station, Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China. + +\(^{6}\) Unit for Environment and Sustainability at the German Aerospace Centre's Project Funding Agency, DLR Projektraeger, Bonn, Germany. + +\(^{7}\) Zhejiang Provincial Key Laboratory of Agricultural Resources and Environment, Zhejiang University, Hangzhou 310058, China + +\(^{8}\) Environmental Systems Analysis Group, Wageningen University & Research, Wageningen, The Netherlands. + +\(^{9}\) UK Centre for Ecology & Hydrology, Bush Estate, Penicuik, UK. + +\(^{10}\) Ministry of Education Key Laboratory of Environment Remediation and Ecological Health, Zhejiang University, Hangzhou 310058, China + +# Co- first author, \* bjgu@zju.edu.cn (B.G.) + +Halving nitrogen pollution is crucial for achieving sustainable development goals (SDGs). However, how to reduce nitrogen pollution from multiple sources remains a grand challenge. Here we show that reactive nitrogen (Nr) pollution could be roughly halved by well- managed urbanization in China by 2050, with emissions of NH3, NOx and N2O to air declining by 44%, 30% and 33%, respectively, and Nr to water bodies by 53%. Urbanization shifts population from rural to urban areas and promotes large- scale and crop- livestock coupled farming, which reduces non- point source pollution from rural sewage and agriculture. Although rural- to- urban migration increases point- source nitrogen emissions in metropolitan areas, regional air and water quality can be improved by reducing upstream and regional total Nr losses, with increased opportunities to control of point source emissions vs. diffuse emissions. Approximate US\(61 billion would be required for additional urban waste treatment, agricultural land consolidation and livestock relocation, as well as upgrading industrial facilities. However, the overall benefits are calculated at US\)245 billion due to increases in agricultural productivity and improvements in environmental quality. Such a large benefit- to- cost ratio suggests the feasibility and cost- effectiveness of halving Nr pollution through urbanization, and this would make significant contributions to achieving several SDG targets. + +The invention of the Haber- Bosch process for nitrogen fixation (HBNF) has doubled global grain production that has supported rapid growth of the global population. However, reactive nitrogen (Nr) losses during food and energy production and consumption have led to multiple unintended consequences, including air and water pollution, soil acidification, biodiversity + +<--- Page Split ---> + +loss and global warming2-4. The United Nations Environment Programme (UNEP) has formulated the objective to halve nitrogen waste globally to achieve global sustainable development goals (SDGs)5, which has now been embraced by the UN Convention on Biological Diversity6. Nitrogen use and losses are directly linked to about half of SDGs such as No poverty (SDG1), Zero hunger (SDG2), Clean water (SDG6) and Responsible consumption and production (SDG12), Climate Action (SDG14), and Life and Land (SDG15)7,8. Therefore, halving nitrogen waste is key to achieving SDGs in coming decades. + +China is a hotspot of global \(\mathrm{N}_{\mathrm{r}}\) pollution, using about \(30\%\) of global nitrogen fertilizers and feeding \(22\%\) of global livestock9. Wastage of \(\mathrm{N}_{\mathrm{r}}\) from agriculture, industrial production and human activities has substantially contributed to China's air and water quality challenges, leading to millions of premature deaths annually10-12. However, it is difficult to reduce nitrogen waste given the multitude of \(\mathrm{N}_{\mathrm{r}}\) sources (e.g. cropland, livestock, sewage) and \(\mathrm{N}_{\mathrm{r}}\) forms (e.g., ammonia ( \(\mathrm{NH}_3\) ), nitrogen oxides ( \(\mathrm{NO}_x\) ), nitrous oxide ( \(\mathrm{N}_2\mathrm{O}\) ), nitrate ( \(\mathrm{NO}_3\) ). This is especially the case for non- point source \(\mathrm{N}_{\mathrm{r}}\) pollution from agriculture and rural areas13. Advanced technologies and management options reduce \(\mathrm{N}_{\mathrm{r}}\) losses from some forms, but may increase \(\mathrm{N}_{\mathrm{r}}\) losses of other forms14,15. Moreover, some abatement measures are rarely implemented to their full potential due to socioeconomic barriers, such as small farm size, which especially limits control of agricultural non- point source pollution in China16,17. It is therefore crucial to identify and realize synergies of \(\mathrm{N}_{\mathrm{r}}\) abatement across different sources and different nitrogen forms. + +Urbanization is often associated with food security challenges as cropland conversion to urban areas reduces agricultural outputs and increases the concentration of pollutant emissions in population centres18,19. Unmanaged urbanization also holds many risks, such as inadequate water treatment infrastructure, leading to major pollution and health damages. However, research indicates that well- managed urbanization can benefit large- scale crop production since urbanization increased the total cropland area and the farm size which allows a faster introduction of modern agricultural practices, which potentially benefits crop- livestock integration for sustainable agriculture20,21. Meanwhile, urbanization moves rural populations to urban areas, which benefits the control of diffuse, non- point source pollution such as rural sewage, waste and livestock production, considering well managed urbanization. + +Currently, approximately 14 million people move from rural to urban areas annually in China22,23, which has a substantial impact on the spatial distribution of resource use and environmental pollution along a rural- urban gradient. However, it has remained little understood how such a substantial socioeconomic change affects \(\mathrm{N}_{\mathrm{r}}\) use and losses. In this paper, we estimated the potential to halve nitrogen pollution in China through well- managed urbanization in all of the \(>2800\) counties of China through analysis in four stages: 1) assessing changes of population, land use and agriculture with urbanization; 2) quantifying how these changes reduce \(\mathrm{N}_{\mathrm{r}}\) pollution from different sources and \(\mathrm{N}_{\mathrm{r}}\) forms; 3) estimating how these reductions in wasteful \(\mathrm{N}_{\mathrm{r}}\) losses on air and water quality; and 4) examining the feasibility and policy implications, including conducting a cost- benefit analysis. + +## Results and Discussion + +## Changes with well-managed urbanization + +Urbanization in China has been increasing by about \(1\%\) annually since the 1990s22,23. The United Nations' World Urbanization Prospects projected that China's urban population would increase from 0.8 billion in 2017 to 1.1 billion in 2050 with a population urbanization level of + +<--- Page Split ---> + +around \(80\%\) if this trend continues (Fig. 1 a, b and Extended Data Fig. 1, 2). We predict that rural- to- urban migration also leads to a relocation of population from inland to coastal areas. Overall, population growth is projected to occur in about \(20\%\) of Chinese cities and counties, mainly in the three largest coastal urban agglomerations, the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, while the remaining \(80\%\) of regions would experience rural depopulation both through migration and natural aging (Fig. 1a). Rural depopulation and continued urban prosperity would lead to an increased level of urbanization in \(82\%\) of the counties, mainly in south of the Hu Line (Fig. 1e). + +The expansion of urban area in response to urban population growth will take up natural and agricultural lands (Fig. 1c and Table S1). During the period from 2017 to 2050, China's total urban area is expected to increase from 5.5 to 7.8 Mha (+2.2 Mha, million hectares, an increase of about \(40\%\) ), while rural built- up areas would decrease from 13.4 to 6.2 Mha (- 7.2 Mha, \(\sim - 54\%\) ) (Fig. 1c, d and Table S1). Although 1.3 Mha of cropland area would be occupied due to urban expansion, a total area of 6.9 Mha could be reclaimed from rural homesteads and converted to cropland. As a consequence, net cropland area can potentially increase by 5.6 Mha, mainly located in North China Plain and Northeast Plain (Fig. 1f). A decrease of cropland is estimated to mainly occur in the three largest metropolitan areas, Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta. + +The reclamation of rural land associated with urbanization could benefit the ongoing increase of large- scale farming practices, with opportunities to reduce nitrogen pollution and increase efficiencies. We estimate that the proportion of large- scale cropland farms (size over 10 ha) could increase from \(9\%\) (2017) to \(90\%\) in 2050 (Extended Data Fig. 3). On the one hand, rural land reclamation increases the total area of croplands available; on the other hand, it contributes to increasing the spatial connectivity of cropland, since rural homesteads are normally next to croplands, which are naturally suitable to be reclaimed for better agricultural production. + +Urbanization also increases the potential for coupled crop- livestock production. Smallholders normally quit livestock production due to having to keep part- time jobs in non- agricultural sectors to ensure their economic viability, which is not comparable with the time needed for livestock feeding as they tend to work partly in urban areas24. As part of the ongoing increase of large- scale farming, a higher share of full- time professional farmers is emerging, who have the capacity to manage both crop and livestock production at the same time25. If this transition is accompanied by appropriate training and equipment, coupling of crop and livestock production at household scale would be enabled, which has already been observed in areas where new farming models (e.g., family farm, cooperative farm, industrial farm) with larger farm size and full- time farmers have been introduced26. These new farming models would gradually replace smallholder farming as the dominant farming method with less rural laborers and larger farm size by 2050, as has happened historically in other world regions. To ensure a well- managed transition, however, it is important that the nitrogen management opportunities are utilized, such as maximizing effective use of manures and other organic residues, while minimizing losses to the environment. + +A further consequence of the distribution of large- scale farms in 2050 would be the relocation of livestock, taking into account the constraint that livestock manure production does not exceed the carrying capacity of crop requirement at county scale. We found that 325 million pig units (all livestock species are converted to pig units, a more detailed description of the method can be found in the Method section) would ultimately need to be relocated. For + +<--- Page Split ---> + +instance, approximately 178 million pigs need to be relocated from the southern region (Fujian, Yunnan, Guangdong, Hunan, etc.) to the central region (Sichuan, Chongqing, Henan, etc.) (Fig. 1g and Extended Data Fig. 4). + +In 2050, it is assuming that nitrogen management will adhere to the latest Five- year plan of the central government. The government's plan involves collecting excreta from urban areas through a septic- sewer network, which will undergo treatment using nitrogen removal or recycling technologies (Extended Data Fig. 5). It aims to reduce \(\mathrm{N}_{\mathrm{f}}\) emissions by upgrading industries and restructuring energy consumption for transportation and heating in urban areas. Managing non- point source pollution in rural areas presents a significant challenge, particularly with regard to wastewater and municipal solid waste. While controlling point source pollution in urban areas is comparatively more manageable, addressing non- point source pollution in rural areas is more complex. Therefore, the government has established precise indicators for wastewater and waste treatment efficiency in urban areas, and we assume that the status quo will persist in rural areas. + +## Reduction of total \(\mathbf{N}_{\mathrm{f}}\) loss + +According to our well- managed urbanization scenario, at national scale, \(\mathrm{N}_{\mathrm{f}}\) losses would be reduced from 33.8 to \(18.3\mathrm{Tg}\) \((- 46\%)\) , with reductions of 10.9, 3.4 and \(1.2\mathrm{Tg}\) in agricultural, human and natural system losses, respectively (Fig. 2). Agricultural systems dominate \(\mathrm{N}_{\mathrm{f}}\) losses at national scale, and large- scale farming, manure recycling fully and emission abatement measures using substantially reduce losses from agricultural systems (e.g., \(\mathrm{NH}_3\) emissions) (Fig. 3a, b). In agricultural systems, \(\mathrm{N}_{\mathrm{f}}\) inputs including fertilizer, deposition and irrigation would substantially decrease as professional farmers look to optimize costs, especially in a world where nitrogen prices may be higher than present, while manure and straw recycling increase. In the human system, migration would increase the share of population connected to centralized sewage treatment (including nitrogen recovery opportunities) and foster household energy transformation, hence leading to both reductions of \(\mathrm{N}_{\mathrm{f}}\) emissions to water bodies and atmosphere in rural areas. The reduction of overall \(\mathrm{N}_{\mathrm{f}}\) losses to the environment from agricultural and human systems would result in lower \(\mathrm{N}_{\mathrm{f}}\) deposition and related \(\mathrm{N}_{\mathrm{f}}\) inputs, and thus also reduce \(\mathrm{N}_{\mathrm{f}}\) losses from natural ecosystems, such as forests. + +Although a few counties in megalopolises with population over 10 million such as Beijing, Shanghai, Shenzhen, are estimated to have more \(\mathrm{N}_{\mathrm{f}}\) loss to environment due to urbanization by 2050, more than \(96\%\) of counties are projected to experience reductions of \(\mathrm{N}_{\mathrm{f}}\) losses (Fig. 2). Key areas where reductions occur are mainly located in the North China Plain, Middle- lower Yangtze River region, Sichuan Plain and Pearl River Delta, consistent with current high rural \(\mathrm{N}_{\mathrm{f}}\) losses in those regions in 2017 (Fig. 2). Below we summarize the changes in nitrogen pollution according to the main forms of \(\mathrm{N}_{\mathrm{f}}\) loss to the environment: + +\(\mathrm{NH}_3\) emissions. \(\mathrm{NH}_3\) emissions are calculated to decrease from \(11.9\mathrm{Tg}\mathrm{N}\) to \(6.7\mathrm{Tg}\mathrm{N}\) \((- 5.2\) \(\mathrm{Tg}\mathrm{N}\) \(- 44\%\) ), with 2.8 and \(1.8\mathrm{Tg}\mathrm{N}\) reductions from cropland and livestock systems, respectively (Fig. 3c). Such large reductions are mainly due to the enhanced \(\mathrm{N}_{\mathrm{f}}\) recycling ratio (NRR, which equaling manure nitrogen returned to field divided by the amount produced) and the substantial increase in nitrogen use efficiency (NUE) of crop and livestock systems, which are the dominant source of \(\mathrm{NH}_3\) . We estimate an increase of NUE in croplands from \(43\%\) to \(54\%\) as a result of large- scale farming, which would ultimately lead to a reduction of total fertilizer use from 28.9 to \(18.5\mathrm{Tg}\mathrm{N}\) by 2050 (Extended Data Fig. 6). According to our estimates, re- coupling of crop and livestock production would increase manure NRR from + +<--- Page Split ---> + +\(23\%\) to \(63\%\) , which would further reduce fertilizer use to \(12.9\mathrm{TgN}\) and the NUE of croplands would thus increase to \(59\%\) , close to current values found in developed countries \(^{29}\) . Reduction of \(\mathrm{NH_3}\) emissions from human subsystems is calculated at approximately \(0.4\mathrm{TgN}\) , mainly owing to reduced emissions from rural sewage and from energy consumption due to upgraded industrial facilities in urban areas. Compared with the free drainage systems in rural areas, better management on sewage and industrial energy generation systems in urban areas are key aspects leading to a reduction of \(\mathrm{NH_3}\) emission in non- agricultural sectors. + +\(\mathrm{NH_3}\) emissions are calculated to be reduced in about \(97\%\) of the counties, mainly located in North China Plain, Hubei- Hunan Plain, Sichuan Basin and southwest of Guangdong and Guangxi (Fig. 4c and Extended Data Fig. 7). This change is mainly due to the increased farm size and livestock relocation in these hotspot regions. While increases in \(\mathrm{NH_3}\) emissions is only modeled for \(3\%\) of counties and cities including the Yangtze River Delta and the Pearl River Delta, which would receive much of the incoming urban populations, leading to more \(\mathrm{NH_3}\) emissions from human excretion and energy consumption. + +\(\mathrm{NO_x}\) emissions. We estimate that well- managed urbanization would reduce \(\mathrm{NO_x}\) emissions from \(4.8\) to \(3.4\mathrm{TgN}\) (- 1.4 Tg N, \(- 30\%\) ) (Fig. 3d), with the human system contributing most to \(\mathrm{NO_x}\) abatement (- 1.4 Tg N). Fossil fuel consumption is the sole largest emission source of \(\mathrm{NO_x}\) , with over \(80\%\) of total emissions. Well- managed urbanization would allow an increase in energy use efficiency, here expressed as \(\mathrm{NO_x}\) emissions per unit of energy supply, which would be reduced both due to increased urban populations, and the increase of alternative clean energy generation methods in 2050, which are more feasible in urban context. The Clean Air Act has reduced \(\mathrm{NO_x}\) emissions by \(10\%\) between 2015 and 2020 mainly through reductions per unit of energy supply \(^{30}\) . This would reduce total \(\mathrm{NO_x}\) emissions from energy sectors, including bioenergy use in rural areas \(^{31}\) . To reduce the land occupied by municipal solid waste in urban area, incineration rate is predicted increase, resulting in \(\mathrm{NO_x}\) emissions increasing by \(0.1\mathrm{TgN}\) (+ \(80\%\) ). However, incineration of waste accounts for only \(7\%\) of the total \(\mathrm{NO_x}\) emissions and is accompanied by a substantial decrease in \(\mathrm{N_r}\) runoff and leaching loss. In addition to \(\mathrm{NO_x}\) emissions from combustion processes, a small amount is estimated to be emitted from soil nitrogen transformation such as denitrification. With the reduction of \(\mathrm{N_r}\) input to agricultural and natural land by professional farmers, including from \(\mathrm{N_r}\) deposition, \(\mathrm{NO_x}\) emissions from soil are estimated to be would be reduced by \(10 - 30\%\) . + +We estimate that \(\mathrm{NO_x}\) emissions would be reduced in \(82\%\) of the counties. The hotspots of \(\mathrm{NO_x}\) emissions reduction are mainly found in the North China Plain where a large number of people are projected to move to metropolitan areas, while and industry upgrades would further reduce \(\mathrm{NO_x}\) emission there (Fig. 4f and Extended Data Fig. 7). However, some significant increases of \(\mathrm{NO_x}\) emissions are projected in the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta. Despite anticipated industrial upgrades as part of well- managed urbanization reducing the \(\mathrm{NO_x}\) emissions per unit of energy supply, total energy supply would need to substantially increase as population would grow by over 200 million in particular in existing metropolitan areas. This would lead to a regional concentration of \(\mathrm{NO_x}\) emissions at in these regions in China. + +\(\mathrm{N_2O}\) and \(\mathrm{N_2}\) emissions. \(\mathrm{N_2O}\) emission reductions is calculated to be comparatively modest in absolute terms (- 0.47 Tg N), yet accounting for \(33\%\) of the total \(\mathrm{N_2O}\) emissions, thus contributing significantly to climate change mitigation (Fig. 3e). The largest contribution is from agricultural subsystem (- 0.36 Tg N), where \(\mathrm{N_2O}\) mainly originates as direct emissions from cropland soil due to fertilization, nitrogen deposition and BNF, as well as manure + +<--- Page Split ---> + +recycling from livestock farming4,32. Even if cropland and livestock N2O emission factors remain constant, N2O emissions would be reduced due to the reduction of nitrogen input and manure recycling. Meanwhile, the reduction of nitrogen inputs in agricultural and human subsystems is calculated to reduce indirect N2O emissions by 0.11 Tg, which occur downstream or down- wind to forest and grassland via water bodies or the atmosphere33. + +We predict that about \(85\%\) of the counties would experience N2O emission reductions, mainly in the North China Plain and the Sichuan Basin (Fig. 4i). The emission reduction in the North China Plain is mainly due to population migration, which leads to a reduction in wastewater and waste disposal in rural areas (Extended Data Fig. 7). In contrast, emission reductions in the Sichuan Basin are driven by a combination of migration and agricultural management changes. However, in the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, emissions of N2O would be substantially increased. + +Based on current knowledge, denitrification emissions of N2 are highly uncertain but are still quantified in our approach for mass consistency as shown in Extended Data Fig. 8. Although N2 emissions are environmentally benign, they are usually ten times greater than N2O emissions and result in a significant loss of valuable Nr resources31. Therefore, reducing N2 emissions and recovering associated Nr resources presents an opportunity to minimize fresh Nr inputs, promoting a cleaner and more circular system. Overall, we observed a \(- 10\%\) reduction impact of urbanization on N2 emissions by 2050 (- 2.3 Tg N). By application of nitrogen removal technologies, N2 generation are predicted increase by 0.9 and 2.3 Tg N in wastewater and garbage subsystems, respectively. Agricultural systems produced \(40\%\) less N2 from denitrification (- 3.8 Tg N) due to lower input Nr, such as fertilizer. Furthermore, the reduction in nitrogen deposition from natural subsystems resulted in a \(16\%\) reduction in N2 emissions (- 1.0 Tg). + +\(N_{\mathrm{r}}\) loss to water. Nr runoff and leaching to water bodies (dominated by NO3-, and labeled as NO3- in Fig. 3, although also including other Nr forms including organic nitrogen) would be reduced from 15.6 to 7.3 Tg (- 8.3 Tg N, - 53%), with the largest contribution from livestock subsystem (- 3.2 Tg N) (Fig. 3f). Livestock- cropland coupled and sustainable livestock management practices will significantly increase manure nitrogen recycling and help to reduce Nr loss to water. The second largest reduction in Nr loss to water (- 2.7 Tg N) is calculated for the cropland subsystem due to reduced Nr inputs and increased NUE. In addition, Nr loss to water is predicted increase 0.5 Tg N in wastewater treatment plants, but decrease 2.1 Tg in wastewater directly discharge and garbage leaching. In 2017, the ratio of population connected to domestic wastewater treatment was \(22\%\) in rural areas compared to \(92\%\) in urban areas35. Urban wastewater treatment ratio will further increase according to China's \(14^{\text{th}}\) Five- Year Plan and 2035 Vision Plan27,36. Higher wastewater treatment capacity and environmental investment in urban areas is key to reduce Nr loss to water37. + +Nationwide, Nr loss to water is calculated to be reduced in \(94\%\) counties (Fig. 4l). The North China Plain, the Two Lakes Plain and the Sichuan Basin would show significant decreases, which are the combined result of population migration, improved sewage treatment capacity and increased agricultural NUE (Extended Data Fig. 7). Industrial upgrading is predicted to further reduce industrial loss in the North China. The Beijing- Tianjin- Hebei region would experience a significant increase in Nr loss to water due to the increase in population. + +## Reduction of \(\mathbf{PM}_{2.5}\) and \(\mathbf{N}_{\mathrm{r}}\) output to water bodies + +Reduction of \(\mathrm{NH_3}\) and \(\mathrm{NO_x}\) emissions are predicted to decrease national average surface + +<--- Page Split ---> + +concentrations of \(\mathrm{PM}_{2.5}\) from 28.0 to \(25.6 \mu \mathrm{g} \mathrm{m}^{- 3}\) (- 2.4 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , - 9%). Regions with the largest reductions are projected to be in the North China Plain (Fig. 5a- c), and the largest reduction of \(\mathrm{PM}_{2.5}\) concentration could reach - 7.7 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , accounting for 18% of \(\mathrm{PM}_{2.5}\) concentrations in 2017. We predict that the three major coastal metropolitan areas, i.e., Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, would also experience reductions of \(\mathrm{PM}_{2.5}\) concentrations, although urbanization is likely to increase \(\mathrm{NH}_{3}\) and \(\mathrm{NO}_{x}\) emissions there. The meteorological conditions in these coastal areas, however, are more favorable for the dispersion of air pollutants, and more importantly, the large reduction of \(\mathrm{NH}_{3}\) and \(\mathrm{NO}_{x}\) emissions in surrounding regions will contribute to an overall reduction of \(\mathrm{PM}_{2.5}\) concentrations at regional scale. In line with the nationwide decrease of \(\mathrm{PM}_{2.5}\) concentrations, the national population- weighted \(\mathrm{PM}_{2.5}\) concentrations will drop from 41.4 to \(32.8 \mu \mathrm{g} \mathrm{m}^{- 3}\) (- 8.7 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , - 21%), and consequently reduce population exposed to the \(\mathrm{PM}_{2.5}\) levels above the national clean air standard (35 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) ) from 609 million to 476 million (- 22%)38, leading to a marked public health benefit and reducing premature deaths. + +Urbanization has also a strong reducing effect on \(\mathrm{N}_{t}\) loading to water bodies and the related \(\mathrm{N}_{t}\) export to the ocean from 6.8 to \(3.5 \mathrm{Tg}\) (- 49%), especially in the Bohai Sea region, which has the potential to improve the eutrophication status of the coastal waters. Simulations based on the total \(\mathrm{N}_{t}\) to surface water and purification capacity of lake, reservoirs, and rivers through denitrification and \(\mathrm{N}_{t}\) accumulation in sediments. On the one hand, the rural- to- urban migration by 2050 is predicted to lead to a shift of hotspots of \(\mathrm{N}_{t}\) losses from inland to coastal urban agglomerations, which weakens the purification potential of inland water bodies (- 2%), due to the shortened retention time of \(\mathrm{N}_{t}\) pollutants in the rivers. On the other hand, the substantial reduction in domestic and agricultural \(\mathrm{N}_{t}\) discharge is predicted to decrease the overall \(\mathrm{N}_{t}\) loading to the lower reaches, with the most significant reduction occurring in the North China Plain (Fig. 5d- f). However, water \(\mathrm{N}_{t}\) loading is predicted to increase in a few scattered reaches of rivers in the Pearl River Delta and the Yangtze River Delta. + +## Cost-benefit assessment of \(\mathrm{N}_{t}\) pollution abatement + +The above- mentioned \(\mathrm{N}_{t}\) pollution reduction potentials implies substantial benefits to the whole society, but it also requires the implementation of abatement measures with related costs. We calculate that approximate US\$ 61 billion \(\mathrm{yr}^{- 1}\) is be required to achieve the abovementioned \(\mathrm{N}_{t}\) pollution reduction potential (Fig. 6h). The one- time implementation cost of US\$ 1404 billion are required for agricultural management optimization over the period of 2017- 2050, including rural homestead reclamation, cropland consolidation and livestock relocation (Extended data Fig. 9). Assuming that these changes would be completed by 2050, the annual implementation cost would be US\$ 43 billion. The cost of cropland consolidation is estimated around US\$ 15 billion \(\mathrm{yr}^{- 1}\) , mainly in the North China Plain and the Middle- lower Yangtze River regions (Fig. 6b), which have the largest potential for the implementation of large- scale farming. As a result of the increase of large- scale farming, livestock production would relocate to these regions, with an annual cost of US\$ 14 billion. The implementation cost of rural reclamation is estimated at US\$ 13 billion \(\mathrm{yr}^{- 1}\) , more than half of which is calculated for the Beijing- Tianjin- Hebei region, Yangtze River Delta and Pearl River Delta. Although the rural reclamation area is small in these three regions, the unit cost per area is one order of magnitude higher than that in other regions due to their relatively higher economic levels and land prices (Table S10). + +In contrast to agricultural measures, the upgrading industrial processes and domestic waste treatment have lower implementation cost, requiring only US\$ 11 and US\$ 8 billion \(\mathrm{yr}^{- 1}\) , respectively (Fig. 6h). Industrial process upgrading is mainly utilized for energy sectors, such + +<--- Page Split ---> + +as \(\mathrm{NO_x}\) emission reduction measures in power plants and vehicles. The implementation cost includes both fixed input and operation cost, with the largest abatement cost in the North China Plain and the northeast regions (Fig. 6c). Relocating population from rural to urban areas would require more facilities to treat domestic sewage and solid waste. We found that urbanization would lead to an additional 40 million tons of wastewater and 0.3 million tons of solid waste daily in urban areas by 2050, requiring roughly a US\(106 billion one- time investment in facilities construction. Assuming these facilities would be completed by 2050, the annual cost would are calculated at US\)3 billion. Beyond the construction of treatment facilities, the unit cost per ton to treat wastewater and solid waste is estimated at US\(0.16 and US\)2.8, respectively, leading to around US\(4 billion yr^{- 1}\) operating costs. The Beijing- Tianjin- Hebei region, Yangtze River Delta and Pearl River Delta are calculate to account for more than 60% of the total implementation cost for domestic waste treatment (Fig. 6a). + +In contrast to these implementation costs, estimated benefits due to reduced regional \(\mathrm{N_r}\) pollution, due to reduced impacts on human and ecosystem health, climate benefits and increased economic returns through fertilizer saving and higher yields, and are estimated near US\(245 billion yr^{- 1}\). Ecosystem benefits are estimated at US\(156 billion yr^{- 1}\), mainly from the mitigation of eutrophication, improvement of drinking water quality and aquaculture production due to reduction of \(\mathrm{NO_3}\) loading of water bodies, as well as preserving forest biodiversity through reduction \(\mathrm{N_r}\) emission and deposition. These benefits mainly occur in the Yangtze River Delta and Pearl River Delta regions where the largest reductions of \(\mathrm{NO_3}\) discharge are estimated and which have advanced levels of economic development (Fig. 6f). Reduction of \(\mathrm{NH_3}\) and \(\mathrm{NO_x}\) emissions will improve air quality due to the reduction of \(\mathrm{PM_{2.5}}\) and near- surface \(\mathrm{O_3}\) concentrations, with benefits to human health estimated at US\(75 billion yr^{- 1}\), with the North China Plain being experiencing the most propound air quality improvements (Fig. 6e). Reduction of agricultural inputs from fertilizers is calculated to lead to savings of US\(12 billion yr^{- 1}\), which would increase farmers' income. Similarly, increases in crop yields are predicted to generate US\(12 billion yr^{- 1}\) economic returns to farmers. The reduction of \(\mathrm{N_2O}\) emissions would results in climate benefits valued at US\(6 billion yr^{- 1}\), but being more than offset by the reduction of carbon sequestration (US\(- 15 billion yr^{- 1}\)) due to reduced \(\mathrm{N_r}\) deposition, resulting in a net climate change effect valued at US\(- 9 billion yr^{- 1}\). In total, urbanization could thus result in a societal benefit of approximately US\(245 billion yr^{- 1}\) in China, approximately five times the total implementation cost, suggesting it would be both feasible and cost- beneficial to adopt ambitious targets for mitigating \(\mathrm{N_r}\). The North China Plain and the Middle- lower Yangtze River regions have the largest total benefit share and benefit- to- cost ratio, highlighting those as preferential areas for \(\mathrm{N_r}\) management. + +## Feasibility of measures in view of policy implications + +To achieve the above- mentioned \(\mathrm{N_r}\) pollution reduction potentials, being near the goal of halving \(\mathrm{N_r}\) waste, the measures introduced here have to be implemented which implies considerable policy implications. Land consolidation is an important pathway for large- scale farming, which requires not only the reclamation of idle rural homesteads, but also cropland consolidation into large- scale farms from small patches. With urbanization and rural aging, hollow villages have become widespread in central and west China, as younger generations have already moved to urban areas. Once the aging people pass away, the younger generation is not likely to return villages since they do not have the ability to engage in farming anymore. In such a context, rural homestead reclamation appears to be feasible. The total area for potential rural reclamation has reached over 40,000 ha during the past decade \(^{39}\) , and this trend is likely to accelerate and further rural aging. Household farm size increase as a result of continuing depopulation of rural areas further promotes cropland consolidation, which + +<--- Page Split ---> + +currently is considered the preferred agricultural land management approach in China. Approximately US\$ 21 billion yr- 1 has so far been invested to consolidate small patches of croplands into large- scale and high- standard farms. The Chinese government has already embarked on cropland consolidation and aims to achieve 80 Mha high- standard croplands with modern management and facilities in the coming decades40. Given experiences so far, investment and governmental future plans, the aspect of land consolidation for large- scale farming appear to be well in line with our projections regarding urbanization. Further policies to address the land tenure system and public services supporting migration to urban areas are considered to provide additional benefits. + +For livestock relocation, the Chinese central government requires newly- built livestock farms to be surrounded by a certain area of croplands for manure recycling. This ensures livestock relocation is not required. Where livestock farms have already been built, livestock relocation costs could be subsidized by government. It should be noted that cross- province relocation of livestock only accounts for \(8\%\) of livestock needed to be relocated (Extended data Fig. 4), suggesting the bulk of the relocation is estimated to occur within provinces, increasing the feasibility of livestock relocation due to farmers may not want to move far away from home. With the increase of large- scale croplands, having both crop and livestock production located close to each other becomes more viable for rural households, further aided by a reduction of implementation barriers by governments. In any case, subsidies supporting the relocation of livestock are vital, and central government has recently provided subsidies worth about US\$ 750 million per county to improve the coupling of livestock and crop production as a measure to reduce environmental pollution originating from manure management in 100 demonstration counties41. + +Upgrading industrial and waste treatment facilities are essential measures responding to an increase in urban population. A continued focus on the implementation of the Clean Air Act and the Clean Water Act would further aid the reduction of \(\mathrm{N_r}\) losses from the human system. Meanwhile, the newly implemented policy "Co- reduction of pollution and carbon emission" would link efforts for pollution control with the target of carbon neutrality (NetZero) while pursuing urbanization objectives. This will place a focus on measures which can achieve both goals in a cost- effective and overall cost- beneficial manner. These policy goals have already put \(\mathrm{N_r}\) reductions on a good pathway, enhancing the feasibility of halving \(\mathrm{N_r}\) pollution in the context of urbanization. Further actions to include measures and policies with relevance for \(\mathrm{N_r}\) pollution reduction into wider considerations, including linking to climate change mitigation and NetZero targets, would contribute to achieving a range of SDGs in China and beyond. + +## Limitation + +Urbanization can affect dietary and production structures, which in turn can affect the use and loss of nitrogen. On the one hand, the increasing proportion of meat consumption among urban residents can increase the amount of livestock farming and the cultivation of animal feed such as soybeans. China has launched forage/soybean revitalization plans to address this issue. The increase in livestock farming can lead to more nitrogen loss, but the replacement of food crops with forage/soybean cultivation can increase nitrogen use efficiency (NUE) and reduce nitrogen loss12. In addition, the demand for aquatic products, which accounted for about \(4\%\) of China's food intake in 2021, is increasing, leading to an increase in aquaculture production. + +Generally, economic growth increases the demand for food and the use of nitrogen fertilizers, + +<--- Page Split ---> + +leading to more nitrogen pollution. However, at the same level of economic development, urban residents consume less food per capita than rural residents due to factors such as less physical activity, higher education levels, a greater emphasis on a balanced diet, and reduced waste43. This complex process makes it difficult to predict how dietary changes will affect nitrogen cycling in urbanization in China, and this study did not take into account these complex factors. + +## Methods + +Data sources. High Resolution Remote Sensing Monitoring of Chinese Land Cover 2018 (HRRSM- CLC2018, \(30\mathrm{m}\) ) was derived from the Resource and Environment Data Cloud Platform (REDCP, https://www.resdc.cn/), which was used to model land use change with urbanization, as well as accompanying potential of large- scale cropland. The digital elevation model (DEM) data used for slope and elevation extraction was derived from Geospatial cloud (http://www.gscloud.cn/). Records of urban and rural population were derived from county- level census data set in China County Yearbook 2018, and provincial level statistical data from National Bureau of Statistics of China 2018 (http://www.stats.gov.cn/tjsj/ndsj/2018/indexch.htm). Records of livestock were derived from the Agriculture Organization's online statistical database (FAOSTAT, https://www.fao.org/faostat/en/#data/EMN), National Bureau of Statistics of China 2018 (NBSC, https://data.stats.gov.cn/easyquery.htm?cn=C01&zb=A0D00&sj=2018) and Agricultural Pollution Source Census of China in 2017 (APSCC2017)25. The FAOSTAT and NBSC provides comprehensive survey data for livestock species in 2017. The APSCC2017 provides grided data of livestock numbers, production and excretion to cropland. + +## Quantification of changes in urban and rural areas, farm size and crop-livestock + +recoupling. urban expansion and land reclamation. Changes in urban and rural area by migration with related changes in urban areas and crop land areas were simulated from 2017 onwards using the High Resolution Remote Sensing Monitoring of Chinese Land Cover 2018 (HRRSM- CLC2018) at a \(30\mathrm{m}\times 30\mathrm{m}\) resolution as the base line. Migration and land use change modeling were based on following assumptions: (1) urban expansion has the highest priority, and can occupy surrounding grids, including cropland; (2) rural land reclamation is used to supply cropland; (3) no limits on migration distance. Spatial land use data and county level urban/rural population data were transformed to gridded data. The change of urban and rural grids at a resolution of \(30\times 30\mathrm{m}\) from 2017- 2050 was predicted by the ArcGIS platform (version 10.6), aligning the number of people on urban and rural grids with the urban and rural populations in 2050 as projected by the United Nations World Urbanization Prospects (UNWUP, https://population. un.org/wup/Download/), respectively. More details of the prediction of urban expansion and land reclamation are in Supplementary Methods. Urbanization level of the county (UL) in 2050 was defined as the ratio of urban residents to the total residents in the county: + +\[U L_{t}^{2050} = \frac{U P_{t}^{2017} + \Delta U P_{t}}{U P_{t}^{2050} + \Delta U P_{t} + R P_{t}^{2017} - \Delta R P_{t}} \quad (1)\] + +where \(U P_{t}\) and \(R P_{t}\) are the urban and rural residents in the \(t\) th county, respectively. \(\Delta U P_{t}\) and \(\Delta U P_{t}\) are the changes on urban and rural residents during 2017- 2050, respectively. + +Potential for Large- scale farming. Potential large- scale cropland is predicted and mapped on county scale, assuming it could be connected to the maximum extent, as long as not divided by other land uses, such as rivers, roads, villages, etc. The simulation method for + +<--- Page Split ---> + +potential larger- scale farming by ArcGIS are described in detail in the Supplementary Methods. + +Crop- livestock coupling. Livestock was relocated based on the distribution of large- scale farming by spatial analysis and linear optimization. FAOSTAT and NBSC were used to estimate the livestock numbers. Details of conversion of pig units are shown in the Extended Data Fig. 4 and Supplementary Methods. Livestock relocation potential \((LRP i)\) is used to determine if the livestock in the \(i\) th grid exceeds the demand of the crop or the carrying capacity of the arable land, and to quantify the potential for moving out or in of livestock. + +\[L R P_{i} = L C P_{\mathrm{max}} - \frac{L I_{i}}{L C A_{i}},i f.L R P_{\mathrm{max}}\leq D N_{i} \quad (2)\] + +\[L R P_{i} = D N_{i} - \frac{L I_{i}}{L C A_{i}},i f.L R P_{\mathrm{max}} > D N_{i} \quad (3)\] + +\[D N_{i} = F_{i}^{2050} + M_{i} \quad (4)\] + +where \(LCP_{max}\) represents the maximum manure Nr carrying capacity of cropland (around 144 kg N ha \(^{- 1}\) ) \(44\) , \(LI_{i}\) represents the total manure Nr (kg) in the county where the \(i\) th grid is located (kg), \(LCA_{i}\) represents the areas of large- scale cropland where the \(i\) th grid is located (km \(^2\) ), \(DNi\) represents the crop demand for Nr, defined as the sum of actual synthetic fertilizer inputs ( \(F_{i}\) , kg ha \(^{- 1}\) ) and manure Nr returned to cropland ( \(M_{i}\) , kg ha \(^{- 1}\) ) in 2017. + +Given reducing the cost of demolition and construction, the livestock would be prioritized to move from the N overload grids into the surrounding grids. Statistics Tool in ArcGIS was used to calculate the priority of all non- overload grid \((LP_{i})\) with a value of 1 to the overload grid \((LRPi\leq 0)\) and 0 to the other grid \((LRPi > 0)\) , respectively. + +\[L P_{i} = \frac{N_{i}}{20\times 20} \quad (5)\] + +where \(N_{i}\) is the number of overloads grids within surrounding 400 grids and \(i\) th grids in the central of 400 grids. Based on the following linear programming, grids are selected for livestock moved in. + +\[m a x m i z e n\] \[s.t\sum_{j = l}^{n}L R P_{j,r}\leq P L R P_{r}\] + +where \(j\) is the serial number for non- overload grids in descending order according to \(L R P_{j,r}\) , \(r\) represents province, \(PLRP_{r}\) is the livestock relocation capacity in the \(r\) th province. If the provincial livestock volume exceeds the provincial carrying capacity, it is assumed that the maximum livestock volume is matched on all the large- scale cropland. The excess livestock volume is moved to the nearest province, and the method of spatial allocation is the same as the intra- province relocation. Livestock cannot be relocated to provinces with restricted development based on the vulnerability level of the surface water, as determined by the Ministry of Agriculture and Rural Affairs (Table S11). + +According to the National Key Research and Development Program in China, we assume that + +<--- Page Split ---> + +crop- livestock coupled would reduce \(50\% \mathrm{NH}_3\) in manure management45. This reduction can be achieved through various measures, including slurry acidification, covering slurry storages, and using closed manure composting technologies. Additionally, when manure is recycled to replace synthetic fertilizer, the \(\mathrm{NH}_3\) emission factor is assumed the same. Furthermore, due to strict water pollution regulations, direct discharge of \(\mathrm{NO}_3^-\) is assumed not allowed in \(2050^{15}\) . Table S4 provides further details on the optimization of various parameters. + +## Quantification of changes in urbanization on Nr pollution + +Ammonia emission. Large- scale farming can increase the adoption of new technologies such as 4R nutrient stewardship (i.e., the right rate, type, time, and place of fertilizer application), optimal irrigation techniques, and organic amendments, resulting in reduced fertilizer application and improved nitrogen use efficiency. This, in turn, can reduce fertilizer \(\mathrm{NH}_3\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{N}\) loss to water. The estimation of influence of farm size on fertilizer use and \(\mathrm{NH}_3\) emission followed Duan et al., \(2021^{46}\) and Wang et al., \(2022^{47}\) . The equation (7) and (8) are used to calculate fertilizer use and \(\mathrm{NH}_3\) emission on county scale in 2050. + +\[F_{t}^{2050} = \exp \left(\alpha \times Ln X_{t}^{2017} - \alpha \times Ln X_{t}^{2050}\right)\times F_{t}^{2017} \quad (7)\] + +\[E_{t}^{2050} = \exp \left(\beta \times Ln X_{t}^{2017} - \beta \times Ln X_{t}^{2017}\right)\times E F_{t}^{2017}\times F_{t}^{2050} \quad (8)\] + +where \(F_{t}\) represents the synthetic fertilizer input in \(t\) th county (kg). \(X_{t}\) stands for the farm size in \(t\) th county (ha). \(E_{t}\) represents the total \(\mathrm{NH}_3\) emission in \(t\) th county (kg), \(EF_{t}\) represents \(\mathrm{NH}_3\) emission factor in \(t\) th county (%), \(\alpha , \beta\) refers to estimated coefficients of large- scale farming from Duan et al46 and Wang et al., \(2022^{47}\) . + +Meanwhile, the management of livestock would be optimized through mitigation measures, such as surface coverage, frequent removal of manure from animal houses, and low crude protein feeding to reduce ammonia volatilization. We cited the research findings of Zhang et al., \(2021^{45}\) , which showed that a combination of different measures in three stages of feeding (including low- protein feed and phased feeding), housing (including ventilation systems, floor mats, dry manure removal, and solid- liquid separation), and manure management (including anaerobic digestion, and high- temperature composting) can achieve a \(60\%\) reduction in \(\mathrm{NH}_3\) emissions from the livestock farming process in China. + +\(N_{r}\) loss to water. Since there is no direct report on the relationship between farm size and other Nr loss to water, the study assumed constant loss efficiency runoff he leaching in the field. The reduction of agricultural nitrogen losses may be underestimated here. In addition, the adoption of a crop- livestock coupled system can ensure that all manure is collected and applied to cropland45. + +Industrial \(\mathrm{NO}_x\) emission. An inverted U- shaped relationship between urbanization and industrial pollutants in China was identified based on Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model and panel data48,49. The changes of industrial \(\mathrm{NO}_x\) emission (NE) and \(\mathrm{NO}_3^-\) discharge (ND) with urbanization were estimated as follow: + +\[NE_{g}^{2050} = \exp \left(\alpha \times Ln NE_{g}^{2017} - \alpha \times Ln NE_{g}^{2050}\right)\times NE_{g}^{2017} \quad (9)\] + +\(NE_{g}\) represent industrial \(\mathrm{NO}_x\) emission in the gth province (kg), \(\alpha\) refers to estimated coefficients from Xu et al48,49,48,49. + +<--- Page Split ---> + +Nitrogen budget. The CHANS model34,50 was used in this study to analyze the impact of urbanization on the national nitrogen budget (Extended Data Fig. 1). The CHANS model has 14 subsystems (e.g., croplands, livestock, human, industry, forest, grassland, atmosphere, water bodies, etc.) that covers all the nitrogen budget calculations in this paper. The baseline of nitrogen budget in 2017 was built first, then urbanization scenarios with corresponding parameters adjusting and urban/rural subsystem developing integrated into CHANS model to forecast the nitrogen cycle and loss in 2050. Urban/rural population change and accompanying difference of domestic pollution treatment efficiency are the key parameters which affect nitrogen budgets through urbanization. Details about the data sources, parameters and modeling methods can be found in Table S3 and Table S4. + +## Quantification of reductions in Nr losses on air \(\mathbf{PM}_{2.5}\) + +Weather Research and Forecasting model coupled with Chemistry (WRF- Chem)51 was used to estimate the changes on surface concentrations of \(\mathrm{PM}_{2.5}\) based on \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emission reduction through well- managed urbanization. The baseline simulation in 2017 included all anthropogenic and natural emissions. Anthropogenic emissions data of air pollution was taken from the MEIC 2017 for mainland China52,53 and the MIX Asia emission inventory for the rest of the WRF- Chem domain54(www.meicmodel.org). The \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emission inventories were replaced with the results of this study, thus enabling the assessment of the impact of urbanization- reduced \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emissions on \(\mathrm{PM}_{2.5}\) . More details for model simulation and validation are given in the Supplementary Methods. + +The population- weighted \(\mathrm{PM}_{2.5}\) concentration \((PWC)\) , which considers population as weights at different exposure to \(\mathrm{PM}_{2.5}\) , is used to reflect the impact of \(\mathrm{PM}_{2.5}\) concentrations on the national population. It is defined as: + +\[PWC = \frac{\sum_{i = 1}^{n}(P_i\times C_i)}{P} \quad (10)\] + +where \(PWE\) is population- weighted \(\mathrm{PM}_{2.5}\) concentration in China, \(P_i\) is the population in the ith grid, \(n\) is the total number of grids in the China, \(C_i\) is the \(\mathrm{PM}_{2.5}\) concentration in the ith grid, and \(P\) is the total population of China. + +## Quantification the reductions of Nr output to sea + +A water network- based framework (WNF) was used to estimate the changes on Nr output to sea based on Nr loss to surface water through well- managed urbanization. WNF is parsimonious solution for surface water modeling that incorporates topology structure, hydrological and biogeochemical processes. And it ensured that Nr transport processes were simulated in complete agreement and the loss of Nr to ocean is only influenced by the changes in Nr loss to surface water. Reactive nitrogen loss to surface runoff (SR) depend on the Nr input to the grid and the loss coefficient determined by the type of land use. + +\[SL = \sum_{k = 1}^{m}Input_{k}\times \alpha_{k} \quad (11)\] + +Where \(input_k\) represent Nr input from the \(k\) th source, covering all of anthropogenic and natural sources, i.g. cropland grid input includes fertilizer, deposition, irrigation, livestock manure, human excretion, cropland BNF and straw recycling. \(\alpha_k\) is the surface runoff loss coefficient. Details about Nr transfer process estimation during surface water based on WNF + +<--- Page Split ---> + +are given in the Supplementary Methods. + +## Quantification of costs and benefits of urbanization. + +\(\mathbf{N}_{\mathrm{f}}\) related implementation cost with urbanization includes the costs of construction and operation of waste treatment facilities, as well as the costs of agricultural management optimization and industrial upgrading. + +Cost of waste treatment. To assess unit cost of sewage/garbage facilities construction (CSCost and \(CGCost\) ), over 200 actual investment projects for waste treatment were collected from the Public- Private Partnership Service Platform under the National Development and Reform Commission of China (www.chinappp.cn). All cost data from the literature or web were measured in constant 2017 US\( (i.e., 100 CNY = US\) 16.16). Taking into account both population migration and economic development level, China was divided into three regions for investment assessment: a population inflow region (PIR), a developed outflow region (DOR) and a less developed outflow region (LOR), as detailed in Table S8. The numbers of projects and unit investment costs for each region are given in Table S9. In addition, the referenced daily unit costs of operating wastewater treatment plants, landfill plants and waste incineration plants (US\( 0.2, 6.5 and 23.0 t^{-1} \) were used to calculate the operating costs\(^{56- 58}\). Annual construction and operation costs of waste treatment facilities (CFC and OFC) were estimated on county scale as follows: + +\[CFC = \sum_{t,n}[(SD\times CSCost_{t} + GD\times CGCost_{t,n}\times R_{n})\times MP_{t}] / YS \quad (12)\] + +\[OFC = \sum_{t,n}[(SD\times OSCost_{t} + GD\times OGCost_{t,n}\times R_{n})\times MP_{t}]\times 365 \quad (13)\] + +where \(t\) represents \(>2800\) counties, \(SD\) and \(GD\) represent the daily production of sewage and garbage per capita (0.15 t capita \(^{- 1}\) and 1.17 kg capita \(^{- 1}\) , respectively \(^{59,60}\) ), \(MP_{t}\) represents the numbers of migrated people in the \(t\) th county (million capita), \(YS\) represents the time span from 2017 to 2050, equal to 33 years, \(CSCost\) and \(OSCost\) represent the unit cost for the construction and annual operation of sewage treatment facilities, respectively (US\( t^{- 1} \), \(CGCost\) and \(OGCost\) represent the unit cost for the construction and annual operation of garbage treatment facilities, respectively (US\( t^{- 1} \), \(n\) represents two types of garbage disposal, i.e. incineration and landfill and \(R_{n}\) is the predicted ratio of incineration and landfill (65% and 35%), referring to the national 14th Five year plan \(^{27}\) , + +Cost of agricultural optimization. Agricultural management would be changed with urbanization in China, with the increasing on the size and scale of farmland and coupling crop- livestock production. Implementation cost of this agricultural optimization are calculated including homestead reclamation, cropland consolidation and livestock relocation. Cost data of homestead reclamation and cropland consolidation were collected from more than 200 projects in China Land Consolidation and Rehabilitation (CLCR, http://www.lerc.org.cn), and were divided into three regions according to migration and economic development levels (Table S10). Annual construction cost of rural reclamation and cropland consolidation (RRC and CCC) were estimated on county scale. + +\[\begin{array}{l}{RRC = \sum_{t}RA_{t}\times RRCost_{t} / YS}\\ {CCC = \sum_{t}CA_{t}\times CCCost_{t} / YS} \end{array} \quad (14)\] + +<--- Page Split ---> + +where \(RA_{t}\) and \(CA_{t}\) are the areas of rural reclamation and cropland consolidation in the \(t\) th county (ha), respectively. \(YS\) represents the time span from 2017 to 2050 (33 years). \(RRCost\) and \(CCCost\) are the unit cost of rural reclamation and cropland consolidation (US\$ ha\(^{- 1}\)), respectively. + +The cost of livestock relocation \((LC)\) mainly includes the costs of dismantling livestock farms and reconstruction new livestock farms. The cost of dismantling a livestock farm includes the cost of the farming facilities and the livestock in stock. Cost data on facilities and livestock \((FLCost\) and \(LLCost)\) were obtained from around 35 local compensation programs by local governments for the removal of livestock farming (Table S10). Reconstruction cost \((RCCost)\) was derived from National Compilation of Agricultural Cost- benefit information \(2018^{61}\) . + +\[LC = \sum_{t}[PR_{t}\times (\epsilon \times FLCost_{t} + LLCost_{t}) / YS] + \sum_{t}PI_{t}\times RCCost \quad (16)\] + +where \(PR_{t}\) represents the livestock that need to be moved out of \(t\) th county (unit pig), \(PI_{t}\) represents the livestock that need to be resettled in the \(t\) th county (unit pig), \(\epsilon\) represents the area of land occupied per unit pig \((2.3 \mathrm{m}^{2}\) per pig) \(^{62}\) , \(FLCost\) and \(LLCost\) represents the unit cost for removal facilities and livestock (US\$, respectively. \(RCCost\) represents the unit product cost of livestock (US\$)\$^{61}. + +Cost of industrial upgrading. The implementation cost of industrial upgrading is defined as direct expenditure for the abatement measures with urbanization to reduce \(\mathrm{N}_{\mathrm{T}}\) emission. Here we mainly refer to the database and methodology of cost assessment from the online Greenhouse Gas and Air Pollution Interactions and Synergies (GAINS) model (https://gains.iiaa.ac.at/models/index.html) to calculate the abatement costs of industrial mitigating \(\mathrm{N}_{\mathrm{T}}\) emissions. International technology cost data have been taken into account and adjusted in GAINS for province- specific conditions in China, which take into account Consideration of local labour costs, energy prices, costs of by- products, etc. The costing module of GAINS has been described in detail by Klimont and Winiwarter \(^{63}\) and Zhang \(^{64}\) . The annual cost of implementation \((COST_{ind})\) is calculated as follows: + +\[IC = \sum_{t,q}\Delta NR_{t,q}\times ICost_{t,q} \quad (17)\] + +where \(q\) represents the forms of \(\mathrm{N}_{\mathrm{T}}\) losses including \(\mathrm{NH}_{3}\) , \(\mathrm{N}_{2}\mathrm{O}\) , \(\mathrm{NO}_{x}\) and \(\mathrm{NO}_{3}\) , \(\Delta NR\) is the reduction of \(\mathrm{N}_{\mathrm{T}}\) through industrial upgrading in county scale, \(ICost\) is the integrated mitigation unit cost of industrial upgrading to reduce \(\mathrm{N}_{\mathrm{T}}\) emission (US\$ per kg N) (Table S6), which is derived from the online GAINS model database. + +Benefit assessment. The total benefit \((TB)\) of reducing \(\mathrm{N}_{\mathrm{T}}\) loss with urbanization include human health benefit \((HBq)\) , ecosystem benefit \((EBq)\) and climate benefit \((CBq)\) and economic benefit \((EcB)\) by reducing agricultural inputs and increasing total yield as follow: + +\[TB = \sum_{q}(HB_{q} + EB_{q} + CB_{q}) + EcB \quad (18)\] + +where \(q\) represents the forms of \(\mathrm{N}_{\mathrm{T}}\) losses, i.e. \(\mathrm{NH}_{3}\) , \(\mathrm{NO}_{x}\) , \(\mathrm{N}_{2}\mathrm{O}\) and \(\mathrm{NO}_{3}\) . \(HB\) , \(EB\) , and \(CB\) denote social benefits, which need to be defined in conjunction to \(\mathrm{N}_{\mathrm{T}}\) loss and converted for monetization. The impact of \(\mathrm{N}_{\mathrm{T}}\) loss on human health is focused on \(\mathrm{PM}_{2.5}\) caused various disease, such as ischemic heart disease (IHD), lung cancer (LC) and so on. The impact on environment mainly refers to water pollution, i.e. eutrophication and biodiversity loss of aquatic ecosystems, due to \(\mathrm{NO}_{3}\) - loss to water, as well as soil acidification and biodiversity + +<--- Page Split ---> + +726 loss of terrestrial ecosystems due to Nr emission and deposition. The impact of \(\mathrm{N_r}\) loss on climate change is complex, with benefits from \(\mathrm{N}_2\mathrm{O}\) reduction and damages from \(\mathrm{NH}_3\) and \(\mathrm{NO_x}\) reduction that affecting carbon sequestration in forests. The monetary benefit of HB, EB and CB were estimated by multiplying social benefits per unit Nr by reduction in Nr with the following equations: + +\[\begin{array}{r}HB_{q} = \Delta NR_{q}\times HCost_{q}\\ EB_{q} = \Delta NR_{q}\times ECost_{q}\\ CB_{q} = \Delta NR_{q}\times CCost_{q} \end{array} \quad (19)\] + +where \(\Delta NR_{q}\) is the reduction of \(q\) th Nr with urbanization in China (kg). \(HCost_{q}\) , \(ECost_{q}\) and \(Ccost_{q}\) are the unit damage cost of the three societal benefit due to the \(q\) th Nr loss (US\$ kg \(\mathrm{N}^*\) ), respectively. \(HCost_{q}\) was calculated by the cause- specific integrated exposure- response functions which can estimate the relative mortality risk from exposure to \(\mathrm{PM}_{2.5}^{65}\) . \(ECost_{q}\) was derived from the nitrogen assessment in Europe after correction for difference in the willingness to pay (WTP) and the purchasing power parity (PPP) between EU and China \(^{66,67}\) . \(Ccost_{q}\) was calculated based on the marginal abatement cost (the carbon price) to reduce or increase emission of CO2- eq \(^{68}\) . A detailed method description of the unit damage cost of varying forms of N loss can be found in Table S7. HB, EB, and CB are assigned to counties based on relevant parameters, for example, changes in air and water pollution, population density, and VSL were used to correct for HB. More details are given in the Supplementary Methods. + +Agricultural management optimization during urbanization increases economic benefit ( \(EcB\) ) by reducing agricultural inputs such as fertilizers, and increasing total yield due to cropland area increase. + +\[EcB = \sum_{i}(A_{i}\times Y_{i}^{2017}\times PC + F_{i}\times FC) \quad (22)\] + +where \(t\) represents \(>2800\) counties; \(\mathrm{A_i}\) is the increasing potential of cropland due to homestead reclamation in the \(t\) th county (ha); \(Y_{i}\) is crop yield in the \(t\) th county in 2017 (kg ha \(^{- 1}\) ), which is derived from the National Bureau of Statistics (www. stats.gov.cn); \(PC\) is the average market price of the wheat in China in 2017 (US\$ 0.38 kg \(^{- 1}\) ), which was from National Development and Reform Commission (https://www.ndrc.gov.cn/xxgk/zcfb/tz/201610/t20161021_963246. html); \(F_{i}\) represents the reduction of synthetic fertilizer input of \(t\) th county related to an increased NUE, associated with an increased farm size, as given in Eq.7 (kg); \(FC\) is the unit cost of Nr fertilizer (US\$ 0.66 kg \(\mathrm{N}^{- 1}\) ), estimated with reference to the 2017 urea market price from the price monitoring network of the Ministry of Agriculture and Rural Affairs, PRC (http://zdscxx.moa.gov.cn:8080/nyb/pc/search.jsp). + +## Reference: + +1 Erisman, J. W., Sutton, M. A., Galloway, J., Klimont, Z. & Winiwarter, W. How a century of ammonia synthesis changed the world. Nat. Geosci. 1 636- 639 (2008). 2 Erisman, J. W. et al.. Consequences of human modification of the global nitrogen cycle. Philos Trans R Soc Lond B Biol Sci 368 20130116 (2013). 3 Galloway, J. N. et al.. Transformation of the Nitrogen Cycle: Recent Trends, Questions, and Potential Solutions. Science 320 889- 892 (2008). 4 Harris, E. et al.. Warming and redistribution of nitrogen inputs drive an increase in + +<--- Page Split ---> + +terrestrial nitrous oxide emission factor. Nat. Commun. 13 (2022). Nitrogen Challenges in Agri- food systems: Halve Nitrogen Waste by 2030. (UNEP, 2020) https://www.unep.org/events/webinar/nitrogen- challenges- agri- food- systems- halve- nitrogen- waste- 2030 UNEP. Kunming Montreal Global Biodiversity Framework, CBD/COP/15/L.25, Target 7. (UNEP, Montreal, 2022). Morseletto, P. Confronting the nitrogen challenge: Options for governance and target setting. Global Environmental Change 54 40- 49 (2019). Sutton, M. A. et al.. The nitrogen decade: mobilizing global actionnonnitrogento2030and beyond. One Earth 4 10- 14 (2021). FAOSTAT: FAO Statistical Databases. (FAO, 2022) https://www.fao.org/statistics/en/ Yu, C. et al.. Managing nitrogen to restore water quality in China. Nature 567 516- 520 (2019). Gu, B. et al.. Abating ammonia is more cost- effective than nitrogen oxides for mitigating PM2.5 air pollution. Science 374 758- 762 (2021). Yun, X. et al.. Residential solid fuel emissions contribute significantly to air pollution and associated health impacts in China. Sci Adv 6 a7621 (2020). Gu, B. et al.. A Credit System to Solve Agricultural Nitrogen Pollution. The Innovation 2 100079 (2021). Shen, G. et al.. Impacts of air pollutants from rural Chinese households under the rapid residential energy transition. Nat. Commun. 10 (2019). Rose, T. J., Wood, R. H., Rose, M. T. & Van Zwieten, L. A re- evaluation of the agronomic effectiveness of the nitrification inhibitors DCD and DMPP and the urease inhibitor NBPT. Agriculture, Ecosystems & Environment 252 69- 73 (2018). Wu, Y. et al.. Policy distortions, farm size, and the overuse of agricultural chemicals in China. Proc Natl Acad Sci 115 7010- 7015 (2018). Ren, C., Zhang, X., Reis, S. & Gu, B. Socioeconomic barriers of nitrogen management for agricultural and environmental sustainability. Agriculture, Ecosystems & Environment 333 107950 (2022). Bren D Amour, C. et al.. Future urban land expansion and implications for global croplands. Proc Natl Acad Sci 114 8939- 8944 (2017). Shi, K. et al.. Urban Expansion and Agricultural Land Loss in China: A Multiscale Perspective. Sustainability- Basel 8 790 (2016). Wang, S. et al.. Urbanization can benefit agricultural production with large- scale farming in China. Nature Food 2 183- 191 (2021). Gu, B., Zhang, X., Bai, X., Fu, B. & Chen, D. Four steps to food security for swelling cities. Nature 566 31- 33 (2019). Lu, Y. et al.. Forty years of reform and opening up: China's progress toward a sustainable path. Sci Adv 5 9413 (2019). Bai, X., Shi, P. & Liu, Y. Society: Realizing China's urban dream. Nature 509 158- 160 (2014). Jin, S. et al.. Decoupling livestock and crop production at the household level in China. Nature Sustainability 4 48- 55 (2021). Zhu, Z. et al.. Integrated livestock sector nitrogen pollution abatement measures could + +<--- Page Split ---> + +generate net benefits for human and ecosystem health in China. Nature Food 3 161- 168 (2022). + +Gu, B. Recoupling livestock and crops. Nature Food 3 102- 103 (2022). + +The Outline of the 14th Five- Year Plan (2021- 2025) for National Economic and Social Development and Vision 2035 of the People's Republic of China. (The State Council, PRC, 2021) http://www.gov.cn/xinwen/2021- 03/13/content_5592681. htm + +Beckinghausen, A., Odlare, M., Thorin, E. & Schwede, S. From removal to recovery: An evaluation of nitrogen recovery techniques from wastewater. Appl Energ 263 114616 (2020). + +Zhang, X. et al. Managing nitrogen for sustainable development. Nature 528 51- 59 (2015). + +Liu, H., Wang, C., Zhang, M. & Wang, S. Evaluating the effects of air pollution control policies in China using a difference- in- differences approach. Sci Total Environ 845 157333 (2022). + +Shen, H. et al. Urbanization- induced population migration has reduced ambient PM2.5 concentrations in China. Sci Adv 3 e1700300 (2017). + +Tian, H. et al. A comprehensive quantification of global nitrous oxide sources and sinks. Nature 586 248- 256 (2020). + +IPCC. 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. (IPCC, Genève, Japan., 2019). + +Gu, B., Ju, X., Chang, J., Ge, Y. & Vitousek, P. M. Integrated reactive nitrogen budgets and future trends in China. Proc. Natl. Acad. Sci. U.S.A. 112 8792- 8797 (2015). + +MHC: Sewage Treatment Rate in Rural Areas was 22% in 2016. (China Daily, 2016) http://www.aswater.com.cn/page196?article_id=102 + +Urban domestic waste classification and treatment facilities development plan in 14th five year. (National Development and Reform Commission, PRC, 2021) https://www.ndrc.gov.cn/xxgk/zcfb/tz/202105/t20210513_1279763_ext.html + +Ma, T. et al. China's improving inland surface water quality since 2003. Sci Adv 6 3798 (2020). + +Ambient air quality standards GB30952012. (Ministry of Ecology and Environment, PRC, 2012) https://www.mee.gov.cn/ywgz/fgbz/bz/bwdb/dqhjbdb/dqhjzlbz/ + +Resource and Environment Data Cloud Platform. (Resource and Environment Science and Data Center, 2018) http://www.resdc.cn/Default.aspx + +The National High- standard Farmland Construction Plan (2021- 2030). (Ministry of Agriculture and Rural Affairs, PRC, 2021) http://www.ntjss.moa.gov.cn/zcfb/202109/t20210915_6376511. htm + +Notice on the implementation of livestock and poultry manure resource utilization projects in 2019. (Ministry of Agriculture and Rural Affairs, PRC, 2019) http://www.moa.gov.cn/govpublic/SYJ/201904/t20190426_6212935. htm + +Cheng, L. et al. A 12% switch from monogastric to ruminant livestock production can reduce emissions and boost crop production for 525 million people. Nature Food 3 1040- 1051 (2022). + +Wu, H. et al. A new dietary guideline balancing sustainability and nutrition for China's rural and urban residents. iScience 25 105048 (2022). + +Technical Guide for Measuring the Land Carrying Capacity of Livestock and Poultry + +<--- Page Split ---> + +Manure. (Ministry of Agriculture and Rural Affairs, PRC, 2018) http://www.xmsyj.moa.gov.cn/zcjd/201904/t20190429_6287183. htm + +Bai, Z. et al.. Relocate 10 billion livestock to reduce harmful nitrogen pollution exposure for \(90\%\) of China' s population. Nature Food 3 152- 160 (2022). + +Duan, J. et al.. Consolidation of agricultural land can contribute to agricultural sustainability in China. Nature Food 2 1014- 1022 (2021). + +Wang, C. et al.. Ammonia Emissions from Croplands Decrease with Farm Size in China. Environ Sci Technol (2022). + +Xu, F. et al.. Reexamining the relationship between urbanization and pollutant emissions in China based on the STIRPAT model. J. Environ Manage 273 111134 (2020). + +Xu, S. et al.. Regional differences in nonlinear impacts of economic growth, export and FDI on air pollutants in China based on provincial panel data. J. Clean Prod 228 455- 466 (2019). + +Gu, B. et al.. Toward a Generic Analytical Framework for Sustainable Nitrogen Management: Application for China. Environ Sci Technol 53 1109- 1118 (2019). + +Grell, G. A. et al.. Fully coupled "online" chemistry within the WRF model. Atmos Environ 39 6957- 6975 (2005). + +Li, M. et al.. Persistent growth of anthropogenic non- methane volatile organic compound (NMVOC) emissions in China during 1990 - 2017: drivers, speciation and ozone formation potential. Atmos Chem Phys 19 8897- 8913 (2019). + +Zheng, B. et al.. Trends in China's anthropogenic emissions since 2010 as the consequence of clean air actions. Atmos Chem Phys 18 14095- 14111 (2018). + +Li, M. et al.. MIX: a mosaic Asian anthropogenic emission inventory under the international collaboration framework of the MICS- Asia and HTAP. Atmos Chem Phys 17 935- 963 (2017). + +Xia, Y. et al.. A new framework to model the distributed transfer and retention of nutrients by incorporating topology structure of small water bodies. Water Res 238 119991 (2023). + +Zhou, C., Gong, Z., Hu, J., Cao, A. & Liang, H. A cost- benefit analysis of landfill mining and material recycling in China. Waste Manage 35 191- 198 (2015). + +Zhao, X., Jiang, G., Li, A. & Wang, L. Economic analysis of waste- to- energy industry in China. Waste Manage 48 604- 618 (2016). + +Tan, X., Shi, L., Ma, Z., Zhang, X. & Lu, G. Institutional analysis of sewage treatment charge based on operating cost of sewage treatment plant—an empirical research of 227 samples in China. China Environmental Science 12 3833- 3840 (2015). + +Wei, X. et al.. Temporal and spatial characteristics of municipal solid waste generation and treatment in China from 1979 to 2016. China Environmental Science 38 3833- 3843 (2018). + +Handbook of accounting methods and coefficients for the statistical investigation of emission sources. (Ministry of Ecology and Environment, PRC, 2021) https://www.mee.gov.cn/xxgk2018/xxgk/xxgk01/202106/t20210618_839512. html (2021). + +China Agricultural Products Cost- Benefit Yearbook 2018. (NDRC, 2018) https://data.cnki.net/trade/Yearbook/Single/N2019010190?zcode=Z009 + +Zhou, D. Improving the old- fashioned traditional pig houses in mountainous rural areas to achieve both environmental protection and comprehensive production benefits. Chinese Journal of Animal Husbandry and Veterinary Medicine 73- 74 (2019). + +<--- Page Split ---> + +903 Klimont, Z. & Winiwarter, W. Integrated ammonia abatement - Modelling of emission 905 control potentials and costs in GAINS. (IIASA, Laxenburg, Austria, 2011). 906 Zhang, X. et al.. Societal benefits of halving agricultural ammonia emissions in China far 907 exceed the abatement costs. Nat. Commun. 11 (2020). 908 Gu, B. et al.. Abating ammonia is more cost- effective than nitrogen oxides for mitigating 909 \(\mathrm{PM}_{2.5}\) air pollution. Science 374 758- 762 (2021). 910 Van Grinsven, H. J. M. et al.. Costs and benefits of nitrogen for Europe and implications 911 for mitigation. Environ Sci Technol 47 3571- 3579 (2013). 912 Sobota, D., Compton, J., McCrackin, M. & Singh, S. Cost of reactive nitrogen release from 913 human activities to the environment in the United States. Environ. Res. Lett. 10 25006 914 (2015). 915 West, J. J. et al.. Co- benefits of mitigating global greenhouse gas emissions for future air 916 quality and human health. Nat. Clim. Change 3 885- 889 (2013). 917 Lesiv, M. et al.. Estimating the global distribution of field size using crowdsourcing. Global 918 Change Biol 25 174- 186 (2018). + +## Data availability + +Data supporting the findings of this study are available within the article and its supplementary information files. + +## Acknowledgements + +This study was supported by the National Natural Science Foundation of China (42261144001 and 42061124001), National Key Research and Development Project of China (2022YFD1700014 and 2022YFE010118). This work is a contribution from Activity 1.4 to the 'Towards the International Nitrogen Management System' project (INMS, http://www.inms.international/) funded by the Global Environment Facility (GEF) through the United Nations Environment Programme (UNEP). + +## Author contributions + +B.G. conceived the study. O.D., J.R. and S.W. performed the urbanization modeling. J.D. and S.H. performed the agricultural optimization modeling. O.D. and X.Z. performed the CHANS modeling and cost- benefit. L.Z., Y.X. and Z.X. performed the air and water quality estimation. O.D. and B.G. wrote the first draft, while S.R., J.X. W.V and M.A.S revised the paper. + +## Competing interests + +All authors have no conflicts of interest to report. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1 | Population, land use and agricultural change with urbanization. These figures demonstrate changes in population, land use and agricultural system with urbanization from 2017 to 2050. a, Geo-distribution of population change. b, Change of population. c, Geo-distribution of land use change, with the same colors as in d. d, Quantity change of land use. e, Geo-distribution of urbanization change. f, Geo-distribution of cropland change. g, Geo-distribution of livestock change. The base map is applied without endorsement from GADM data (https://gadm.org/).
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 | Spatial variation of nitrogen loss with urbanization.
+ +a, Distribution of total reactive nitrogen (TN) losses in 2017. b, Distribution of predicted TN losses in 2050 assuming an urbanization level of \(80\%\) . c, Distribution of the change in TN losses during the 2017- 2050 period. d, National \(\mathrm{NH}_3\) , \(\mathrm{NO_x}\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{NO}_3\) - loss in 2017 and 2050. TN losses include \(\mathrm{NH}_3\) , \(\mathrm{NO_x}\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{NO}_3\) - loss. We use \(\mathrm{NO}_3\) - to represent the \(\mathrm{N}_r\) loss via runoff and leaching, given that most of the other dissolve forms of \(\mathrm{N}_r\) would be converted to \(\mathrm{NO}_3\) - during moving with water. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 | Nitrogen budgets within key subsystems and their contribution to N losses reduction with urbanization.
+ +a, b, Total N budgets for four subsystems and the environment in 2017 (a) and 2050 (b). These four subsystems, i.e. cropland, livestock, human and nature, are defined by the CHANS model (Fig. S1). Human subsystem set includes urban human, rural human, wastewater treatment, garbage treatment and industry subsystems. Nature subsystem set includes grassland and forest subsystem. The colored arrows represent N flows to varying subsystems as follows: yellow, cropland subsystem; green, livestock subsystem; blue, human subsystem; brown, nature subsystem. The black arrows indicate the N flows that eventually enter the atmosphere and water bodies through gaseous emissions and runoff losses. c-f, National nitrogen losses in 2017 and 2050, and the sources of the difference. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 | Spatial variation of \(\mathrm{Nr}\) species loss with urbanization.
+ +(a-c), Distribution of \(\mathrm{NH_3}\) in 2017 (a), 2050 (b) and the difference (c). (d-f), Distribution of \(\mathrm{NO_x}\) in 2017 (d), 2050 (e) and the difference (f). (g-i), Distribution of \(\mathrm{N_2O}\) in 2017 (g), 2050 (h) and the difference (i). (j-l), Distribution of runoff \(\mathrm{N}\) loss in 2017 (j), 2050 (k) and the difference (l). The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5 | Spatial variation of atmospheric PM2.5 concentration and \(\mathbf{N}_{\mathrm{r}}\) to ocean with urbanization.
+ +a- c, Geographic distribution of surface air PM2.5 concentration in 2017 (a), 2050 (b), and the difference (c). d- f, Geographic distribution of \(\mathbf{N}_{\mathrm{r}}\) loading to downstream reach in 2017 (d), 2050 (e), and the difference (f). The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig. 6 | Costs and benefits of halving \(\mathbf{N}_{\mathrm{r}}\) pollution with urbanization. a-c, Geographic distribution of implementation cost on waste treatment (a), agricultural optimization (b), and industrial pollution system upgrades (c) to achieve the above mentioned reduction potential of \(\mathbf{N}_{\mathrm{r}}\) pollution with urbanization. d-f, Geographic distribution of monetized benefits on climate (d), human health (e) and ecosystem (f). g, Geographic distribution of economic benefit from fertilizer saving and yield increase. h, National costs and benefits with urbanization. Blue indicates implementation costs, while red indicates benefits. The base map is applied without endorsement from GADM data (https://gadm.org/).
+ +<--- Page Split ---> +![PLACEHOLDER_28_0] + + +Extended Data Fig. 1 | Mechanism of N loss reduction with urbanization. Rural- urban migration benefits the management of domestic sewage and waste due to centralized waste management and pollution control measures. Meanwhile, quantity and scale of croplands and the coupling of crop and livestock production would be increased according to rural reclamation and rural population reduction. These processes are beneficial for improving N use efficiency (NUE) and reducing N losses. + +<--- Page Split ---> +![PLACEHOLDER_29_0] + + +Extended Data Fig. 2 | Changes on population with urbanization. a- c, Geo- distribution of population in 2017, 2050 and the difference. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![PLACEHOLDER_30_0] + + +## Extended Data Fig. 3 | Field size distribution through cropland consolidation and urbanization. + +a- c, The geographic distribution of cropland field size in 2017 (a), and integration of consolidation and urbanization in 2050 (b). c, Field size share with cropland consolidation and urbanization. Data of current field size in 2017 and potential field size after cropland consolidation were derived from Lesiv et al., 201869 and Duan et al., 202146, respectively. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![PLACEHOLDER_31_0] + + +# Extended Data Fig. 4 | Livestock change through crop-livestock coupled. + +Followed by the distribution of large- scale farms in 2050, livestock would be relocated with constrains that livestock manure production does not exceed the carrying capacity of crop requirement on county scale. a- c, Geographic distribution of livestock in 2017 (a), 2050 (b) and difference (c). d, The number of livestock moving into the province, out of the province, and relocation within province in 31 provinces. The full names of 31 provinces are shown in table S6. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![PLACEHOLDER_32_0] + + +Extended Data Fig. 5 | Nitrogen emission in urban and rural waste treatment system + +The diagram includes traditional and well- managed urban and rural waste treatment systems. The red arrows represent traditional treatment path; the blue arrows represent harmless treatment methods. The yellow box represents the final discharge form of nitrogen; the gray box represents ancillary facilities in the harmless treatment process, including toilets, waste transfer systems, wastewater treatment plants (WTP), landfills, and incineration plants; the white boxes represent types of waste from urban or rural areas. + +<--- Page Split ---> +![PLACEHOLDER_33_0] + + +1039 + +1040 + +1041 + +1042 + +1043 + +1044 + +1045 + +1046 + +1047 + +1048 + +1049 + +1050 + +1051 + +1052 + +1053 + +1054 + +1055 + +1056 + +1057 + +1058 + +1059 + +1060 + +1061 + +1062 + +1063 + +1064 + +1065 + +1066 + +1067 + +1068 + +1069 + +1070 + +1080 + + + +<--- Page Split ---> +![PLACEHOLDER_34_0] + + +Extended Data Fig. 7 | Geographic distribution changes of \(\mathbf{N}_{\mathrm{r}}\) emissions through three urbanization processes. + +a- c, Geographic distribution changes of \(\mathrm{NH}_3\) through population migration (PM), agricultural optimization(AO) and industrial upgrading(IU), respectively. d- f, Geographic distribution changes of \(\mathrm{NO_x}\) through PM, AO and IU, respectively. g- i, Geographic distribution changes of \(\mathrm{NO_3}\) through PM, AO and IU, respectively. j- l, Geographic distribution changes of \(\mathrm{NO_3}\) through PM, AO and IU,respectively. Since industrial upgrading primarily focuses on reducing \(\mathrm{NO_x}\) emissions and has little impact on \(\mathrm{NH}_3\) and \(\mathrm{N}_2\mathrm{O}\) emissions, there is minimal change in c and i. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +![PLACEHOLDER_35_0] + + +Extended Data Fig. 8 | The Geographic distribution of N₂ emission with urbanization. a-c, Spatial distribution of N₂ emissions in 2017 (a) and in 2050 (b), and the difference (c). d, National N₂ emissions in 2017 and 2050, and the sources of the difference. + +<--- Page Split ---> +![PLACEHOLDER_36_0] + + +## Extended Data Fig. 9| Methods for quantifying effects of urbanization on nitrogen pollution. + +Graphical Model for urbanization process represents the whole simulation process, including migration, urban expansion, industrial upgrading, cropland consolidation, livestock relocation, and rural reclamation. It changes the whole nitrogen cycling on national scale. The urbanization process covering the period from 2017 to 2050 would eventually reduce \(\mathrm{N}_{\mathrm{r}}\) losses, which enhance air and water quality. The red boxes represent the estimation of costs incurred through urbanization. Abbreviations: NTE, Nitrogen Treatment Efficiencies; NTE, Nitrogen Use Efficiency; NRR, Nitrogen Recycle Ratio; WNRR, Water Nitrogen Retention and Removing model; WRF-Chem, Weather Research and Forecasting model coupled with Chemistry. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SIUrbanizationXNCv.0707. docx + +<--- Page Split ---> diff --git a/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5_det.mmd b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..eb597570b3ada18b80b1c0d7affd8e5a7a8fdaca --- /dev/null +++ b/preprint/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5/preprint__144634e3b50cbbfd3f517c5e0cf8adc9da2922848226a43f00f0e2c307d0a8d5_det.mmd @@ -0,0 +1,835 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 886, 175]]<|/det|> +# Well-managed urbanization could halve nitrogen pollution in China + +<|ref|>text<|/ref|><|det|>[[44, 195, 582, 238]]<|/det|> +Baojing Gu ( \(\boxed{\pm}\) bjgu@zju.edu.cn) Zhejiang University https://orcid.org/0000- 0003- 3986- 3519 + +<|ref|>text<|/ref|><|det|>[[44, 243, 225, 283]]<|/det|> +Ouping Deng Zhejiang University + +<|ref|>text<|/ref|><|det|>[[44, 290, 225, 330]]<|/det|> +Sitong Wang Zhejiang University + +<|ref|>text<|/ref|><|det|>[[44, 336, 323, 376]]<|/det|> +Jiangyou Ran Sichuan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 382, 323, 422]]<|/det|> +Shuai Huang Sichuan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 428, 225, 468]]<|/det|> +Xiuming Zhang Zhejiang University + +<|ref|>text<|/ref|><|det|>[[44, 474, 225, 514]]<|/det|> +Jiakun Duan Zhejiang University + +<|ref|>text<|/ref|><|det|>[[44, 520, 562, 561]]<|/det|> +Lin Zhang Peking University https://orcid.org/0000- 0003- 2383- 8431 + +<|ref|>text<|/ref|><|det|>[[44, 567, 220, 627]]<|/det|> +Yongqiu Xia Stefan Reis DLR Projekttraeger + +<|ref|>text<|/ref|><|det|>[[44, 635, 210, 675]]<|/det|> +Jiayu Xu Peking University + +<|ref|>text<|/ref|><|det|>[[44, 682, 582, 722]]<|/det|> +Jianming Xu Zhejiang University https://orcid.org/0000- 0002- 2954- 9764 + +<|ref|>text<|/ref|><|det|>[[44, 728, 737, 769]]<|/det|> +Wim de Vries Wageningen University and Research https://orcid.org/0000- 0001- 9974- 0612 + +<|ref|>text<|/ref|><|det|>[[44, 775, 402, 815]]<|/det|> +Mark Sutton NERC Centre for Ecology and Hydrology + +<|ref|>text<|/ref|><|det|>[[44, 857, 101, 874]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 895, 137, 912]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 932, 296, 951]]<|/det|> +Posted Date: July 13th, 2023 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 474, 64]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3148472/v1 + +<|ref|>text<|/ref|><|det|>[[42, 82, 910, 125]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 143, 531, 163]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 199, 930, 242]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 9th, 2024. See the published version at https://doi.org/10.1038/s41467- 023- 44685- y. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 701, 100]]<|/det|> +## Well-managed urbanization could halve nitrogen pollution in China + +<|ref|>text<|/ref|><|det|>[[115, 113, 881, 166]]<|/det|> +Ouping Deng \(^{1,2,\#}\) , Sitong Wang \(^{1,3,\#}\) , Jiangyou Ran \(^{2}\) , Shuai Huang \(^{2}\) , Xiuming Zhang \(^{1}\) , Jiakun Duan \(^{1}\) , Lin Zhang \(^{4}\) , Yongqiu Xia \(^{5}\) , Stefan Reis \(^{6}\) , Jiayu Xu \(^{4}\) , Jianming Xu \(^{1,7}\) , Wim de Vries \(^{8}\) , Mark A. Sutton \(^{9}\) , Baojing Gu \(^{1,3,10*}\) + +<|ref|>text<|/ref|><|det|>[[115, 181, 850, 199]]<|/det|> +\(^{1}\) College of Environmental and Resource Sciences, Zhejiang University, Hangzhou, China. + +<|ref|>text<|/ref|><|det|>[[115, 199, 710, 215]]<|/det|> +\(^{2}\) College of Resources, Sichuan Agricultural University, Chengdu, China. + +<|ref|>text<|/ref|><|det|>[[115, 216, 752, 232]]<|/det|> +\(^{3}\) Policy Simulation Laboratory, Zhejiang University, Hangzhou 310058, China + +<|ref|>text<|/ref|><|det|>[[115, 232, 880, 265]]<|/det|> +\(^{4}\) Department of Atmospheric and Oceanic Sciences, School of Physics, Peking University, Beijing, China. + +<|ref|>text<|/ref|><|det|>[[115, 266, 880, 313]]<|/det|> +\(^{5}\) Key Laboratory of Soil and Sustainable Agriculture, Changshu National Agr- Ecosystem Observation and Research Station, Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China. + +<|ref|>text<|/ref|><|det|>[[115, 314, 880, 348]]<|/det|> +\(^{6}\) Unit for Environment and Sustainability at the German Aerospace Centre's Project Funding Agency, DLR Projektraeger, Bonn, Germany. + +<|ref|>text<|/ref|><|det|>[[115, 349, 880, 381]]<|/det|> +\(^{7}\) Zhejiang Provincial Key Laboratory of Agricultural Resources and Environment, Zhejiang University, Hangzhou 310058, China + +<|ref|>text<|/ref|><|det|>[[115, 382, 880, 415]]<|/det|> +\(^{8}\) Environmental Systems Analysis Group, Wageningen University & Research, Wageningen, The Netherlands. + +<|ref|>text<|/ref|><|det|>[[115, 415, 655, 432]]<|/det|> +\(^{9}\) UK Centre for Ecology & Hydrology, Bush Estate, Penicuik, UK. + +<|ref|>text<|/ref|><|det|>[[115, 432, 880, 466]]<|/det|> +\(^{10}\) Ministry of Education Key Laboratory of Environment Remediation and Ecological Health, Zhejiang University, Hangzhou 310058, China + +<|ref|>text<|/ref|><|det|>[[115, 480, 472, 497]]<|/det|> +# Co- first author, \* bjgu@zju.edu.cn (B.G.) + +<|ref|>text<|/ref|><|det|>[[115, 530, 880, 825]]<|/det|> +Halving nitrogen pollution is crucial for achieving sustainable development goals (SDGs). However, how to reduce nitrogen pollution from multiple sources remains a grand challenge. Here we show that reactive nitrogen (Nr) pollution could be roughly halved by well- managed urbanization in China by 2050, with emissions of NH3, NOx and N2O to air declining by 44%, 30% and 33%, respectively, and Nr to water bodies by 53%. Urbanization shifts population from rural to urban areas and promotes large- scale and crop- livestock coupled farming, which reduces non- point source pollution from rural sewage and agriculture. Although rural- to- urban migration increases point- source nitrogen emissions in metropolitan areas, regional air and water quality can be improved by reducing upstream and regional total Nr losses, with increased opportunities to control of point source emissions vs. diffuse emissions. Approximate US\(61 billion would be required for additional urban waste treatment, agricultural land consolidation and livestock relocation, as well as upgrading industrial facilities. However, the overall benefits are calculated at US\)245 billion due to increases in agricultural productivity and improvements in environmental quality. Such a large benefit- to- cost ratio suggests the feasibility and cost- effectiveness of halving Nr pollution through urbanization, and this would make significant contributions to achieving several SDG targets. + +<|ref|>text<|/ref|><|det|>[[115, 841, 875, 908]]<|/det|> +The invention of the Haber- Bosch process for nitrogen fixation (HBNF) has doubled global grain production that has supported rapid growth of the global population. However, reactive nitrogen (Nr) losses during food and energy production and consumption have led to multiple unintended consequences, including air and water pollution, soil acidification, biodiversity + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 866, 201]]<|/det|> +loss and global warming2-4. The United Nations Environment Programme (UNEP) has formulated the objective to halve nitrogen waste globally to achieve global sustainable development goals (SDGs)5, which has now been embraced by the UN Convention on Biological Diversity6. Nitrogen use and losses are directly linked to about half of SDGs such as No poverty (SDG1), Zero hunger (SDG2), Clean water (SDG6) and Responsible consumption and production (SDG12), Climate Action (SDG14), and Life and Land (SDG15)7,8. Therefore, halving nitrogen waste is key to achieving SDGs in coming decades. + +<|ref|>text<|/ref|><|det|>[[113, 214, 880, 428]]<|/det|> +China is a hotspot of global \(\mathrm{N}_{\mathrm{r}}\) pollution, using about \(30\%\) of global nitrogen fertilizers and feeding \(22\%\) of global livestock9. Wastage of \(\mathrm{N}_{\mathrm{r}}\) from agriculture, industrial production and human activities has substantially contributed to China's air and water quality challenges, leading to millions of premature deaths annually10-12. However, it is difficult to reduce nitrogen waste given the multitude of \(\mathrm{N}_{\mathrm{r}}\) sources (e.g. cropland, livestock, sewage) and \(\mathrm{N}_{\mathrm{r}}\) forms (e.g., ammonia ( \(\mathrm{NH}_3\) ), nitrogen oxides ( \(\mathrm{NO}_x\) ), nitrous oxide ( \(\mathrm{N}_2\mathrm{O}\) ), nitrate ( \(\mathrm{NO}_3\) ). This is especially the case for non- point source \(\mathrm{N}_{\mathrm{r}}\) pollution from agriculture and rural areas13. Advanced technologies and management options reduce \(\mathrm{N}_{\mathrm{r}}\) losses from some forms, but may increase \(\mathrm{N}_{\mathrm{r}}\) losses of other forms14,15. Moreover, some abatement measures are rarely implemented to their full potential due to socioeconomic barriers, such as small farm size, which especially limits control of agricultural non- point source pollution in China16,17. It is therefore crucial to identify and realize synergies of \(\mathrm{N}_{\mathrm{r}}\) abatement across different sources and different nitrogen forms. + +<|ref|>text<|/ref|><|det|>[[113, 445, 880, 630]]<|/det|> +Urbanization is often associated with food security challenges as cropland conversion to urban areas reduces agricultural outputs and increases the concentration of pollutant emissions in population centres18,19. Unmanaged urbanization also holds many risks, such as inadequate water treatment infrastructure, leading to major pollution and health damages. However, research indicates that well- managed urbanization can benefit large- scale crop production since urbanization increased the total cropland area and the farm size which allows a faster introduction of modern agricultural practices, which potentially benefits crop- livestock integration for sustainable agriculture20,21. Meanwhile, urbanization moves rural populations to urban areas, which benefits the control of diffuse, non- point source pollution such as rural sewage, waste and livestock production, considering well managed urbanization. + +<|ref|>text<|/ref|><|det|>[[113, 646, 880, 810]]<|/det|> +Currently, approximately 14 million people move from rural to urban areas annually in China22,23, which has a substantial impact on the spatial distribution of resource use and environmental pollution along a rural- urban gradient. However, it has remained little understood how such a substantial socioeconomic change affects \(\mathrm{N}_{\mathrm{r}}\) use and losses. In this paper, we estimated the potential to halve nitrogen pollution in China through well- managed urbanization in all of the \(>2800\) counties of China through analysis in four stages: 1) assessing changes of population, land use and agriculture with urbanization; 2) quantifying how these changes reduce \(\mathrm{N}_{\mathrm{r}}\) pollution from different sources and \(\mathrm{N}_{\mathrm{r}}\) forms; 3) estimating how these reductions in wasteful \(\mathrm{N}_{\mathrm{r}}\) losses on air and water quality; and 4) examining the feasibility and policy implications, including conducting a cost- benefit analysis. + +<|ref|>sub_title<|/ref|><|det|>[[115, 827, 314, 842]]<|/det|> +## Results and Discussion + +<|ref|>sub_title<|/ref|><|det|>[[115, 844, 476, 860]]<|/det|> +## Changes with well-managed urbanization + +<|ref|>text<|/ref|><|det|>[[115, 860, 878, 910]]<|/det|> +Urbanization in China has been increasing by about \(1\%\) annually since the 1990s22,23. The United Nations' World Urbanization Prospects projected that China's urban population would increase from 0.8 billion in 2017 to 1.1 billion in 2050 with a population urbanization level of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 872, 216]]<|/det|> +around \(80\%\) if this trend continues (Fig. 1 a, b and Extended Data Fig. 1, 2). We predict that rural- to- urban migration also leads to a relocation of population from inland to coastal areas. Overall, population growth is projected to occur in about \(20\%\) of Chinese cities and counties, mainly in the three largest coastal urban agglomerations, the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, while the remaining \(80\%\) of regions would experience rural depopulation both through migration and natural aging (Fig. 1a). Rural depopulation and continued urban prosperity would lead to an increased level of urbanization in \(82\%\) of the counties, mainly in south of the Hu Line (Fig. 1e). + +<|ref|>text<|/ref|><|det|>[[115, 230, 872, 396]]<|/det|> +The expansion of urban area in response to urban population growth will take up natural and agricultural lands (Fig. 1c and Table S1). During the period from 2017 to 2050, China's total urban area is expected to increase from 5.5 to 7.8 Mha (+2.2 Mha, million hectares, an increase of about \(40\%\) ), while rural built- up areas would decrease from 13.4 to 6.2 Mha (- 7.2 Mha, \(\sim - 54\%\) ) (Fig. 1c, d and Table S1). Although 1.3 Mha of cropland area would be occupied due to urban expansion, a total area of 6.9 Mha could be reclaimed from rural homesteads and converted to cropland. As a consequence, net cropland area can potentially increase by 5.6 Mha, mainly located in North China Plain and Northeast Plain (Fig. 1f). A decrease of cropland is estimated to mainly occur in the three largest metropolitan areas, Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta. + +<|ref|>text<|/ref|><|det|>[[115, 410, 880, 544]]<|/det|> +The reclamation of rural land associated with urbanization could benefit the ongoing increase of large- scale farming practices, with opportunities to reduce nitrogen pollution and increase efficiencies. We estimate that the proportion of large- scale cropland farms (size over 10 ha) could increase from \(9\%\) (2017) to \(90\%\) in 2050 (Extended Data Fig. 3). On the one hand, rural land reclamation increases the total area of croplands available; on the other hand, it contributes to increasing the spatial connectivity of cropland, since rural homesteads are normally next to croplands, which are naturally suitable to be reclaimed for better agricultural production. + +<|ref|>text<|/ref|><|det|>[[115, 558, 875, 805]]<|/det|> +Urbanization also increases the potential for coupled crop- livestock production. Smallholders normally quit livestock production due to having to keep part- time jobs in non- agricultural sectors to ensure their economic viability, which is not comparable with the time needed for livestock feeding as they tend to work partly in urban areas24. As part of the ongoing increase of large- scale farming, a higher share of full- time professional farmers is emerging, who have the capacity to manage both crop and livestock production at the same time25. If this transition is accompanied by appropriate training and equipment, coupling of crop and livestock production at household scale would be enabled, which has already been observed in areas where new farming models (e.g., family farm, cooperative farm, industrial farm) with larger farm size and full- time farmers have been introduced26. These new farming models would gradually replace smallholder farming as the dominant farming method with less rural laborers and larger farm size by 2050, as has happened historically in other world regions. To ensure a well- managed transition, however, it is important that the nitrogen management opportunities are utilized, such as maximizing effective use of manures and other organic residues, while minimizing losses to the environment. + +<|ref|>text<|/ref|><|det|>[[115, 820, 875, 904]]<|/det|> +A further consequence of the distribution of large- scale farms in 2050 would be the relocation of livestock, taking into account the constraint that livestock manure production does not exceed the carrying capacity of crop requirement at county scale. We found that 325 million pig units (all livestock species are converted to pig units, a more detailed description of the method can be found in the Method section) would ultimately need to be relocated. For + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 873, 135]]<|/det|> +instance, approximately 178 million pigs need to be relocated from the southern region (Fujian, Yunnan, Guangdong, Hunan, etc.) to the central region (Sichuan, Chongqing, Henan, etc.) (Fig. 1g and Extended Data Fig. 4). + +<|ref|>text<|/ref|><|det|>[[115, 149, 875, 330]]<|/det|> +In 2050, it is assuming that nitrogen management will adhere to the latest Five- year plan of the central government. The government's plan involves collecting excreta from urban areas through a septic- sewer network, which will undergo treatment using nitrogen removal or recycling technologies (Extended Data Fig. 5). It aims to reduce \(\mathrm{N}_{\mathrm{f}}\) emissions by upgrading industries and restructuring energy consumption for transportation and heating in urban areas. Managing non- point source pollution in rural areas presents a significant challenge, particularly with regard to wastewater and municipal solid waste. While controlling point source pollution in urban areas is comparatively more manageable, addressing non- point source pollution in rural areas is more complex. Therefore, the government has established precise indicators for wastewater and waste treatment efficiency in urban areas, and we assume that the status quo will persist in rural areas. + +<|ref|>sub_title<|/ref|><|det|>[[117, 346, 336, 362]]<|/det|> +## Reduction of total \(\mathbf{N}_{\mathrm{f}}\) loss + +<|ref|>text<|/ref|><|det|>[[115, 363, 872, 609]]<|/det|> +According to our well- managed urbanization scenario, at national scale, \(\mathrm{N}_{\mathrm{f}}\) losses would be reduced from 33.8 to \(18.3\mathrm{Tg}\) \((- 46\%)\) , with reductions of 10.9, 3.4 and \(1.2\mathrm{Tg}\) in agricultural, human and natural system losses, respectively (Fig. 2). Agricultural systems dominate \(\mathrm{N}_{\mathrm{f}}\) losses at national scale, and large- scale farming, manure recycling fully and emission abatement measures using substantially reduce losses from agricultural systems (e.g., \(\mathrm{NH}_3\) emissions) (Fig. 3a, b). In agricultural systems, \(\mathrm{N}_{\mathrm{f}}\) inputs including fertilizer, deposition and irrigation would substantially decrease as professional farmers look to optimize costs, especially in a world where nitrogen prices may be higher than present, while manure and straw recycling increase. In the human system, migration would increase the share of population connected to centralized sewage treatment (including nitrogen recovery opportunities) and foster household energy transformation, hence leading to both reductions of \(\mathrm{N}_{\mathrm{f}}\) emissions to water bodies and atmosphere in rural areas. The reduction of overall \(\mathrm{N}_{\mathrm{f}}\) losses to the environment from agricultural and human systems would result in lower \(\mathrm{N}_{\mathrm{f}}\) deposition and related \(\mathrm{N}_{\mathrm{f}}\) inputs, and thus also reduce \(\mathrm{N}_{\mathrm{f}}\) losses from natural ecosystems, such as forests. + +<|ref|>text<|/ref|><|det|>[[115, 624, 877, 740]]<|/det|> +Although a few counties in megalopolises with population over 10 million such as Beijing, Shanghai, Shenzhen, are estimated to have more \(\mathrm{N}_{\mathrm{f}}\) loss to environment due to urbanization by 2050, more than \(96\%\) of counties are projected to experience reductions of \(\mathrm{N}_{\mathrm{f}}\) losses (Fig. 2). Key areas where reductions occur are mainly located in the North China Plain, Middle- lower Yangtze River region, Sichuan Plain and Pearl River Delta, consistent with current high rural \(\mathrm{N}_{\mathrm{f}}\) losses in those regions in 2017 (Fig. 2). Below we summarize the changes in nitrogen pollution according to the main forms of \(\mathrm{N}_{\mathrm{f}}\) loss to the environment: + +<|ref|>text<|/ref|><|det|>[[115, 755, 880, 904]]<|/det|> +\(\mathrm{NH}_3\) emissions. \(\mathrm{NH}_3\) emissions are calculated to decrease from \(11.9\mathrm{Tg}\mathrm{N}\) to \(6.7\mathrm{Tg}\mathrm{N}\) \((- 5.2\) \(\mathrm{Tg}\mathrm{N}\) \(- 44\%\) ), with 2.8 and \(1.8\mathrm{Tg}\mathrm{N}\) reductions from cropland and livestock systems, respectively (Fig. 3c). Such large reductions are mainly due to the enhanced \(\mathrm{N}_{\mathrm{f}}\) recycling ratio (NRR, which equaling manure nitrogen returned to field divided by the amount produced) and the substantial increase in nitrogen use efficiency (NUE) of crop and livestock systems, which are the dominant source of \(\mathrm{NH}_3\) . We estimate an increase of NUE in croplands from \(43\%\) to \(54\%\) as a result of large- scale farming, which would ultimately lead to a reduction of total fertilizer use from 28.9 to \(18.5\mathrm{Tg}\mathrm{N}\) by 2050 (Extended Data Fig. 6). According to our estimates, re- coupling of crop and livestock production would increase manure NRR from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 879, 199]]<|/det|> +\(23\%\) to \(63\%\) , which would further reduce fertilizer use to \(12.9\mathrm{TgN}\) and the NUE of croplands would thus increase to \(59\%\) , close to current values found in developed countries \(^{29}\) . Reduction of \(\mathrm{NH_3}\) emissions from human subsystems is calculated at approximately \(0.4\mathrm{TgN}\) , mainly owing to reduced emissions from rural sewage and from energy consumption due to upgraded industrial facilities in urban areas. Compared with the free drainage systems in rural areas, better management on sewage and industrial energy generation systems in urban areas are key aspects leading to a reduction of \(\mathrm{NH_3}\) emission in non- agricultural sectors. + +<|ref|>text<|/ref|><|det|>[[115, 214, 872, 330]]<|/det|> +\(\mathrm{NH_3}\) emissions are calculated to be reduced in about \(97\%\) of the counties, mainly located in North China Plain, Hubei- Hunan Plain, Sichuan Basin and southwest of Guangdong and Guangxi (Fig. 4c and Extended Data Fig. 7). This change is mainly due to the increased farm size and livestock relocation in these hotspot regions. While increases in \(\mathrm{NH_3}\) emissions is only modeled for \(3\%\) of counties and cities including the Yangtze River Delta and the Pearl River Delta, which would receive much of the incoming urban populations, leading to more \(\mathrm{NH_3}\) emissions from human excretion and energy consumption. + +<|ref|>text<|/ref|><|det|>[[115, 345, 878, 624]]<|/det|> +\(\mathrm{NO_x}\) emissions. We estimate that well- managed urbanization would reduce \(\mathrm{NO_x}\) emissions from \(4.8\) to \(3.4\mathrm{TgN}\) (- 1.4 Tg N, \(- 30\%\) ) (Fig. 3d), with the human system contributing most to \(\mathrm{NO_x}\) abatement (- 1.4 Tg N). Fossil fuel consumption is the sole largest emission source of \(\mathrm{NO_x}\) , with over \(80\%\) of total emissions. Well- managed urbanization would allow an increase in energy use efficiency, here expressed as \(\mathrm{NO_x}\) emissions per unit of energy supply, which would be reduced both due to increased urban populations, and the increase of alternative clean energy generation methods in 2050, which are more feasible in urban context. The Clean Air Act has reduced \(\mathrm{NO_x}\) emissions by \(10\%\) between 2015 and 2020 mainly through reductions per unit of energy supply \(^{30}\) . This would reduce total \(\mathrm{NO_x}\) emissions from energy sectors, including bioenergy use in rural areas \(^{31}\) . To reduce the land occupied by municipal solid waste in urban area, incineration rate is predicted increase, resulting in \(\mathrm{NO_x}\) emissions increasing by \(0.1\mathrm{TgN}\) (+ \(80\%\) ). However, incineration of waste accounts for only \(7\%\) of the total \(\mathrm{NO_x}\) emissions and is accompanied by a substantial decrease in \(\mathrm{N_r}\) runoff and leaching loss. In addition to \(\mathrm{NO_x}\) emissions from combustion processes, a small amount is estimated to be emitted from soil nitrogen transformation such as denitrification. With the reduction of \(\mathrm{N_r}\) input to agricultural and natural land by professional farmers, including from \(\mathrm{N_r}\) deposition, \(\mathrm{NO_x}\) emissions from soil are estimated to be would be reduced by \(10 - 30\%\) . + +<|ref|>text<|/ref|><|det|>[[115, 639, 879, 805]]<|/det|> +We estimate that \(\mathrm{NO_x}\) emissions would be reduced in \(82\%\) of the counties. The hotspots of \(\mathrm{NO_x}\) emissions reduction are mainly found in the North China Plain where a large number of people are projected to move to metropolitan areas, while and industry upgrades would further reduce \(\mathrm{NO_x}\) emission there (Fig. 4f and Extended Data Fig. 7). However, some significant increases of \(\mathrm{NO_x}\) emissions are projected in the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta. Despite anticipated industrial upgrades as part of well- managed urbanization reducing the \(\mathrm{NO_x}\) emissions per unit of energy supply, total energy supply would need to substantially increase as population would grow by over 200 million in particular in existing metropolitan areas. This would lead to a regional concentration of \(\mathrm{NO_x}\) emissions at in these regions in China. + +<|ref|>text<|/ref|><|det|>[[115, 820, 870, 904]]<|/det|> +\(\mathrm{N_2O}\) and \(\mathrm{N_2}\) emissions. \(\mathrm{N_2O}\) emission reductions is calculated to be comparatively modest in absolute terms (- 0.47 Tg N), yet accounting for \(33\%\) of the total \(\mathrm{N_2O}\) emissions, thus contributing significantly to climate change mitigation (Fig. 3e). The largest contribution is from agricultural subsystem (- 0.36 Tg N), where \(\mathrm{N_2O}\) mainly originates as direct emissions from cropland soil due to fertilization, nitrogen deposition and BNF, as well as manure + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 869, 168]]<|/det|> +recycling from livestock farming4,32. Even if cropland and livestock N2O emission factors remain constant, N2O emissions would be reduced due to the reduction of nitrogen input and manure recycling. Meanwhile, the reduction of nitrogen inputs in agricultural and human subsystems is calculated to reduce indirect N2O emissions by 0.11 Tg, which occur downstream or down- wind to forest and grassland via water bodies or the atmosphere33. + +<|ref|>text<|/ref|><|det|>[[115, 182, 878, 298]]<|/det|> +We predict that about \(85\%\) of the counties would experience N2O emission reductions, mainly in the North China Plain and the Sichuan Basin (Fig. 4i). The emission reduction in the North China Plain is mainly due to population migration, which leads to a reduction in wastewater and waste disposal in rural areas (Extended Data Fig. 7). In contrast, emission reductions in the Sichuan Basin are driven by a combination of migration and agricultural management changes. However, in the Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, emissions of N2O would be substantially increased. + +<|ref|>text<|/ref|><|det|>[[115, 312, 883, 510]]<|/det|> +Based on current knowledge, denitrification emissions of N2 are highly uncertain but are still quantified in our approach for mass consistency as shown in Extended Data Fig. 8. Although N2 emissions are environmentally benign, they are usually ten times greater than N2O emissions and result in a significant loss of valuable Nr resources31. Therefore, reducing N2 emissions and recovering associated Nr resources presents an opportunity to minimize fresh Nr inputs, promoting a cleaner and more circular system. Overall, we observed a \(- 10\%\) reduction impact of urbanization on N2 emissions by 2050 (- 2.3 Tg N). By application of nitrogen removal technologies, N2 generation are predicted increase by 0.9 and 2.3 Tg N in wastewater and garbage subsystems, respectively. Agricultural systems produced \(40\%\) less N2 from denitrification (- 3.8 Tg N) due to lower input Nr, such as fertilizer. Furthermore, the reduction in nitrogen deposition from natural subsystems resulted in a \(16\%\) reduction in N2 emissions (- 1.0 Tg). + +<|ref|>text<|/ref|><|det|>[[115, 525, 861, 740]]<|/det|> +\(N_{\mathrm{r}}\) loss to water. Nr runoff and leaching to water bodies (dominated by NO3-, and labeled as NO3- in Fig. 3, although also including other Nr forms including organic nitrogen) would be reduced from 15.6 to 7.3 Tg (- 8.3 Tg N, - 53%), with the largest contribution from livestock subsystem (- 3.2 Tg N) (Fig. 3f). Livestock- cropland coupled and sustainable livestock management practices will significantly increase manure nitrogen recycling and help to reduce Nr loss to water. The second largest reduction in Nr loss to water (- 2.7 Tg N) is calculated for the cropland subsystem due to reduced Nr inputs and increased NUE. In addition, Nr loss to water is predicted increase 0.5 Tg N in wastewater treatment plants, but decrease 2.1 Tg in wastewater directly discharge and garbage leaching. In 2017, the ratio of population connected to domestic wastewater treatment was \(22\%\) in rural areas compared to \(92\%\) in urban areas35. Urban wastewater treatment ratio will further increase according to China's \(14^{\text{th}}\) Five- Year Plan and 2035 Vision Plan27,36. Higher wastewater treatment capacity and environmental investment in urban areas is key to reduce Nr loss to water37. + +<|ref|>text<|/ref|><|det|>[[115, 755, 866, 855]]<|/det|> +Nationwide, Nr loss to water is calculated to be reduced in \(94\%\) counties (Fig. 4l). The North China Plain, the Two Lakes Plain and the Sichuan Basin would show significant decreases, which are the combined result of population migration, improved sewage treatment capacity and increased agricultural NUE (Extended Data Fig. 7). Industrial upgrading is predicted to further reduce industrial loss in the North China. The Beijing- Tianjin- Hebei region would experience a significant increase in Nr loss to water due to the increase in population. + +<|ref|>sub_title<|/ref|><|det|>[[115, 870, 544, 886]]<|/det|> +## Reduction of \(\mathbf{PM}_{2.5}\) and \(\mathbf{N}_{\mathrm{r}}\) output to water bodies + +<|ref|>text<|/ref|><|det|>[[115, 887, 830, 903]]<|/det|> +Reduction of \(\mathrm{NH_3}\) and \(\mathrm{NO_x}\) emissions are predicted to decrease national average surface + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 878, 313]]<|/det|> +concentrations of \(\mathrm{PM}_{2.5}\) from 28.0 to \(25.6 \mu \mathrm{g} \mathrm{m}^{- 3}\) (- 2.4 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , - 9%). Regions with the largest reductions are projected to be in the North China Plain (Fig. 5a- c), and the largest reduction of \(\mathrm{PM}_{2.5}\) concentration could reach - 7.7 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , accounting for 18% of \(\mathrm{PM}_{2.5}\) concentrations in 2017. We predict that the three major coastal metropolitan areas, i.e., Beijing- Tianjin region, the Yangtze River Delta and the Pearl River Delta, would also experience reductions of \(\mathrm{PM}_{2.5}\) concentrations, although urbanization is likely to increase \(\mathrm{NH}_{3}\) and \(\mathrm{NO}_{x}\) emissions there. The meteorological conditions in these coastal areas, however, are more favorable for the dispersion of air pollutants, and more importantly, the large reduction of \(\mathrm{NH}_{3}\) and \(\mathrm{NO}_{x}\) emissions in surrounding regions will contribute to an overall reduction of \(\mathrm{PM}_{2.5}\) concentrations at regional scale. In line with the nationwide decrease of \(\mathrm{PM}_{2.5}\) concentrations, the national population- weighted \(\mathrm{PM}_{2.5}\) concentrations will drop from 41.4 to \(32.8 \mu \mathrm{g} \mathrm{m}^{- 3}\) (- 8.7 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) , - 21%), and consequently reduce population exposed to the \(\mathrm{PM}_{2.5}\) levels above the national clean air standard (35 \(\mu \mathrm{g} \mathrm{m}^{- 3}\) ) from 609 million to 476 million (- 22%)38, leading to a marked public health benefit and reducing premature deaths. + +<|ref|>text<|/ref|><|det|>[[115, 328, 878, 526]]<|/det|> +Urbanization has also a strong reducing effect on \(\mathrm{N}_{t}\) loading to water bodies and the related \(\mathrm{N}_{t}\) export to the ocean from 6.8 to \(3.5 \mathrm{Tg}\) (- 49%), especially in the Bohai Sea region, which has the potential to improve the eutrophication status of the coastal waters. Simulations based on the total \(\mathrm{N}_{t}\) to surface water and purification capacity of lake, reservoirs, and rivers through denitrification and \(\mathrm{N}_{t}\) accumulation in sediments. On the one hand, the rural- to- urban migration by 2050 is predicted to lead to a shift of hotspots of \(\mathrm{N}_{t}\) losses from inland to coastal urban agglomerations, which weakens the purification potential of inland water bodies (- 2%), due to the shortened retention time of \(\mathrm{N}_{t}\) pollutants in the rivers. On the other hand, the substantial reduction in domestic and agricultural \(\mathrm{N}_{t}\) discharge is predicted to decrease the overall \(\mathrm{N}_{t}\) loading to the lower reaches, with the most significant reduction occurring in the North China Plain (Fig. 5d- f). However, water \(\mathrm{N}_{t}\) loading is predicted to increase in a few scattered reaches of rivers in the Pearl River Delta and the Yangtze River Delta. + +<|ref|>sub_title<|/ref|><|det|>[[115, 542, 550, 558]]<|/det|> +## Cost-benefit assessment of \(\mathrm{N}_{t}\) pollution abatement + +<|ref|>text<|/ref|><|det|>[[115, 559, 880, 837]]<|/det|> +The above- mentioned \(\mathrm{N}_{t}\) pollution reduction potentials implies substantial benefits to the whole society, but it also requires the implementation of abatement measures with related costs. We calculate that approximate US\$ 61 billion \(\mathrm{yr}^{- 1}\) is be required to achieve the abovementioned \(\mathrm{N}_{t}\) pollution reduction potential (Fig. 6h). The one- time implementation cost of US\$ 1404 billion are required for agricultural management optimization over the period of 2017- 2050, including rural homestead reclamation, cropland consolidation and livestock relocation (Extended data Fig. 9). Assuming that these changes would be completed by 2050, the annual implementation cost would be US\$ 43 billion. The cost of cropland consolidation is estimated around US\$ 15 billion \(\mathrm{yr}^{- 1}\) , mainly in the North China Plain and the Middle- lower Yangtze River regions (Fig. 6b), which have the largest potential for the implementation of large- scale farming. As a result of the increase of large- scale farming, livestock production would relocate to these regions, with an annual cost of US\$ 14 billion. The implementation cost of rural reclamation is estimated at US\$ 13 billion \(\mathrm{yr}^{- 1}\) , more than half of which is calculated for the Beijing- Tianjin- Hebei region, Yangtze River Delta and Pearl River Delta. Although the rural reclamation area is small in these three regions, the unit cost per area is one order of magnitude higher than that in other regions due to their relatively higher economic levels and land prices (Table S10). + +<|ref|>text<|/ref|><|det|>[[115, 853, 872, 904]]<|/det|> +In contrast to agricultural measures, the upgrading industrial processes and domestic waste treatment have lower implementation cost, requiring only US\$ 11 and US\$ 8 billion \(\mathrm{yr}^{- 1}\) , respectively (Fig. 6h). Industrial process upgrading is mainly utilized for energy sectors, such + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 83, 880, 281]]<|/det|> +as \(\mathrm{NO_x}\) emission reduction measures in power plants and vehicles. The implementation cost includes both fixed input and operation cost, with the largest abatement cost in the North China Plain and the northeast regions (Fig. 6c). Relocating population from rural to urban areas would require more facilities to treat domestic sewage and solid waste. We found that urbanization would lead to an additional 40 million tons of wastewater and 0.3 million tons of solid waste daily in urban areas by 2050, requiring roughly a US\(106 billion one- time investment in facilities construction. Assuming these facilities would be completed by 2050, the annual cost would are calculated at US\)3 billion. Beyond the construction of treatment facilities, the unit cost per ton to treat wastewater and solid waste is estimated at US\(0.16 and US\)2.8, respectively, leading to around US\(4 billion yr^{- 1}\) operating costs. The Beijing- Tianjin- Hebei region, Yangtze River Delta and Pearl River Delta are calculate to account for more than 60% of the total implementation cost for domestic waste treatment (Fig. 6a). + +<|ref|>text<|/ref|><|det|>[[112, 295, 880, 675]]<|/det|> +In contrast to these implementation costs, estimated benefits due to reduced regional \(\mathrm{N_r}\) pollution, due to reduced impacts on human and ecosystem health, climate benefits and increased economic returns through fertilizer saving and higher yields, and are estimated near US\(245 billion yr^{- 1}\). Ecosystem benefits are estimated at US\(156 billion yr^{- 1}\), mainly from the mitigation of eutrophication, improvement of drinking water quality and aquaculture production due to reduction of \(\mathrm{NO_3}\) loading of water bodies, as well as preserving forest biodiversity through reduction \(\mathrm{N_r}\) emission and deposition. These benefits mainly occur in the Yangtze River Delta and Pearl River Delta regions where the largest reductions of \(\mathrm{NO_3}\) discharge are estimated and which have advanced levels of economic development (Fig. 6f). Reduction of \(\mathrm{NH_3}\) and \(\mathrm{NO_x}\) emissions will improve air quality due to the reduction of \(\mathrm{PM_{2.5}}\) and near- surface \(\mathrm{O_3}\) concentrations, with benefits to human health estimated at US\(75 billion yr^{- 1}\), with the North China Plain being experiencing the most propound air quality improvements (Fig. 6e). Reduction of agricultural inputs from fertilizers is calculated to lead to savings of US\(12 billion yr^{- 1}\), which would increase farmers' income. Similarly, increases in crop yields are predicted to generate US\(12 billion yr^{- 1}\) economic returns to farmers. The reduction of \(\mathrm{N_2O}\) emissions would results in climate benefits valued at US\(6 billion yr^{- 1}\), but being more than offset by the reduction of carbon sequestration (US\(- 15 billion yr^{- 1}\)) due to reduced \(\mathrm{N_r}\) deposition, resulting in a net climate change effect valued at US\(- 9 billion yr^{- 1}\). In total, urbanization could thus result in a societal benefit of approximately US\(245 billion yr^{- 1}\) in China, approximately five times the total implementation cost, suggesting it would be both feasible and cost- beneficial to adopt ambitious targets for mitigating \(\mathrm{N_r}\). The North China Plain and the Middle- lower Yangtze River regions have the largest total benefit share and benefit- to- cost ratio, highlighting those as preferential areas for \(\mathrm{N_r}\) management. + +<|ref|>sub_title<|/ref|><|det|>[[115, 689, 570, 706]]<|/det|> +## Feasibility of measures in view of policy implications + +<|ref|>text<|/ref|><|det|>[[113, 706, 880, 904]]<|/det|> +To achieve the above- mentioned \(\mathrm{N_r}\) pollution reduction potentials, being near the goal of halving \(\mathrm{N_r}\) waste, the measures introduced here have to be implemented which implies considerable policy implications. Land consolidation is an important pathway for large- scale farming, which requires not only the reclamation of idle rural homesteads, but also cropland consolidation into large- scale farms from small patches. With urbanization and rural aging, hollow villages have become widespread in central and west China, as younger generations have already moved to urban areas. Once the aging people pass away, the younger generation is not likely to return villages since they do not have the ability to engage in farming anymore. In such a context, rural homestead reclamation appears to be feasible. The total area for potential rural reclamation has reached over 40,000 ha during the past decade \(^{39}\) , and this trend is likely to accelerate and further rural aging. Household farm size increase as a result of continuing depopulation of rural areas further promotes cropland consolidation, which + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 83, 875, 232]]<|/det|> +currently is considered the preferred agricultural land management approach in China. Approximately US\$ 21 billion yr- 1 has so far been invested to consolidate small patches of croplands into large- scale and high- standard farms. The Chinese government has already embarked on cropland consolidation and aims to achieve 80 Mha high- standard croplands with modern management and facilities in the coming decades40. Given experiences so far, investment and governmental future plans, the aspect of land consolidation for large- scale farming appear to be well in line with our projections regarding urbanization. Further policies to address the land tenure system and public services supporting migration to urban areas are considered to provide additional benefits. + +<|ref|>text<|/ref|><|det|>[[113, 247, 875, 477]]<|/det|> +For livestock relocation, the Chinese central government requires newly- built livestock farms to be surrounded by a certain area of croplands for manure recycling. This ensures livestock relocation is not required. Where livestock farms have already been built, livestock relocation costs could be subsidized by government. It should be noted that cross- province relocation of livestock only accounts for \(8\%\) of livestock needed to be relocated (Extended data Fig. 4), suggesting the bulk of the relocation is estimated to occur within provinces, increasing the feasibility of livestock relocation due to farmers may not want to move far away from home. With the increase of large- scale croplands, having both crop and livestock production located close to each other becomes more viable for rural households, further aided by a reduction of implementation barriers by governments. In any case, subsidies supporting the relocation of livestock are vital, and central government has recently provided subsidies worth about US\$ 750 million per county to improve the coupling of livestock and crop production as a measure to reduce environmental pollution originating from manure management in 100 demonstration counties41. + +<|ref|>text<|/ref|><|det|>[[113, 492, 880, 674]]<|/det|> +Upgrading industrial and waste treatment facilities are essential measures responding to an increase in urban population. A continued focus on the implementation of the Clean Air Act and the Clean Water Act would further aid the reduction of \(\mathrm{N_r}\) losses from the human system. Meanwhile, the newly implemented policy "Co- reduction of pollution and carbon emission" would link efforts for pollution control with the target of carbon neutrality (NetZero) while pursuing urbanization objectives. This will place a focus on measures which can achieve both goals in a cost- effective and overall cost- beneficial manner. These policy goals have already put \(\mathrm{N_r}\) reductions on a good pathway, enhancing the feasibility of halving \(\mathrm{N_r}\) pollution in the context of urbanization. Further actions to include measures and policies with relevance for \(\mathrm{N_r}\) pollution reduction into wider considerations, including linking to climate change mitigation and NetZero targets, would contribute to achieving a range of SDGs in China and beyond. + +<|ref|>sub_title<|/ref|><|det|>[[115, 689, 210, 705]]<|/det|> +## Limitation + +<|ref|>text<|/ref|><|det|>[[115, 722, 880, 872]]<|/det|> +Urbanization can affect dietary and production structures, which in turn can affect the use and loss of nitrogen. On the one hand, the increasing proportion of meat consumption among urban residents can increase the amount of livestock farming and the cultivation of animal feed such as soybeans. China has launched forage/soybean revitalization plans to address this issue. The increase in livestock farming can lead to more nitrogen loss, but the replacement of food crops with forage/soybean cultivation can increase nitrogen use efficiency (NUE) and reduce nitrogen loss12. In addition, the demand for aquatic products, which accounted for about \(4\%\) of China's food intake in 2021, is increasing, leading to an increase in aquaculture production. + +<|ref|>text<|/ref|><|det|>[[112, 887, 870, 904]]<|/det|> +Generally, economic growth increases the demand for food and the use of nitrogen fertilizers, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 870, 183]]<|/det|> +leading to more nitrogen pollution. However, at the same level of economic development, urban residents consume less food per capita than rural residents due to factors such as less physical activity, higher education levels, a greater emphasis on a balanced diet, and reduced waste43. This complex process makes it difficult to predict how dietary changes will affect nitrogen cycling in urbanization in China, and this study did not take into account these complex factors. + +<|ref|>sub_title<|/ref|><|det|>[[116, 200, 195, 214]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[115, 215, 881, 461]]<|/det|> +Data sources. High Resolution Remote Sensing Monitoring of Chinese Land Cover 2018 (HRRSM- CLC2018, \(30\mathrm{m}\) ) was derived from the Resource and Environment Data Cloud Platform (REDCP, https://www.resdc.cn/), which was used to model land use change with urbanization, as well as accompanying potential of large- scale cropland. The digital elevation model (DEM) data used for slope and elevation extraction was derived from Geospatial cloud (http://www.gscloud.cn/). Records of urban and rural population were derived from county- level census data set in China County Yearbook 2018, and provincial level statistical data from National Bureau of Statistics of China 2018 (http://www.stats.gov.cn/tjsj/ndsj/2018/indexch.htm). Records of livestock were derived from the Agriculture Organization's online statistical database (FAOSTAT, https://www.fao.org/faostat/en/#data/EMN), National Bureau of Statistics of China 2018 (NBSC, https://data.stats.gov.cn/easyquery.htm?cn=C01&zb=A0D00&sj=2018) and Agricultural Pollution Source Census of China in 2017 (APSCC2017)25. The FAOSTAT and NBSC provides comprehensive survey data for livestock species in 2017. The APSCC2017 provides grided data of livestock numbers, production and excretion to cropland. + +<|ref|>sub_title<|/ref|><|det|>[[115, 478, 808, 495]]<|/det|> +## Quantification of changes in urban and rural areas, farm size and crop-livestock + +<|ref|>text<|/ref|><|det|>[[115, 495, 877, 740]]<|/det|> +recoupling. urban expansion and land reclamation. Changes in urban and rural area by migration with related changes in urban areas and crop land areas were simulated from 2017 onwards using the High Resolution Remote Sensing Monitoring of Chinese Land Cover 2018 (HRRSM- CLC2018) at a \(30\mathrm{m}\times 30\mathrm{m}\) resolution as the base line. Migration and land use change modeling were based on following assumptions: (1) urban expansion has the highest priority, and can occupy surrounding grids, including cropland; (2) rural land reclamation is used to supply cropland; (3) no limits on migration distance. Spatial land use data and county level urban/rural population data were transformed to gridded data. The change of urban and rural grids at a resolution of \(30\times 30\mathrm{m}\) from 2017- 2050 was predicted by the ArcGIS platform (version 10.6), aligning the number of people on urban and rural grids with the urban and rural populations in 2050 as projected by the United Nations World Urbanization Prospects (UNWUP, https://population. un.org/wup/Download/), respectively. More details of the prediction of urban expansion and land reclamation are in Supplementary Methods. Urbanization level of the county (UL) in 2050 was defined as the ratio of urban residents to the total residents in the county: + +<|ref|>equation<|/ref|><|det|>[[343, 755, 879, 797]]<|/det|> +\[U L_{t}^{2050} = \frac{U P_{t}^{2017} + \Delta U P_{t}}{U P_{t}^{2050} + \Delta U P_{t} + R P_{t}^{2017} - \Delta R P_{t}} \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[115, 800, 864, 840]]<|/det|> +where \(U P_{t}\) and \(R P_{t}\) are the urban and rural residents in the \(t\) th county, respectively. \(\Delta U P_{t}\) and \(\Delta U P_{t}\) are the changes on urban and rural residents during 2017- 2050, respectively. + +<|ref|>text<|/ref|><|det|>[[115, 856, 860, 908]]<|/det|> +Potential for Large- scale farming. Potential large- scale cropland is predicted and mapped on county scale, assuming it could be connected to the maximum extent, as long as not divided by other land uses, such as rivers, roads, villages, etc. The simulation method for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 812, 117]]<|/det|> +potential larger- scale farming by ArcGIS are described in detail in the Supplementary Methods. + +<|ref|>text<|/ref|><|det|>[[115, 131, 870, 232]]<|/det|> +Crop- livestock coupling. Livestock was relocated based on the distribution of large- scale farming by spatial analysis and linear optimization. FAOSTAT and NBSC were used to estimate the livestock numbers. Details of conversion of pig units are shown in the Extended Data Fig. 4 and Supplementary Methods. Livestock relocation potential \((LRP i)\) is used to determine if the livestock in the \(i\) th grid exceeds the demand of the crop or the carrying capacity of the arable land, and to quantify the potential for moving out or in of livestock. + +<|ref|>equation<|/ref|><|det|>[[360, 245, 880, 283]]<|/det|> +\[L R P_{i} = L C P_{\mathrm{max}} - \frac{L I_{i}}{L C A_{i}},i f.L R P_{\mathrm{max}}\leq D N_{i} \quad (2)\] + +<|ref|>equation<|/ref|><|det|>[[360, 303, 880, 340]]<|/det|> +\[L R P_{i} = D N_{i} - \frac{L I_{i}}{L C A_{i}},i f.L R P_{\mathrm{max}} > D N_{i} \quad (3)\] + +<|ref|>equation<|/ref|><|det|>[[457, 358, 880, 380]]<|/det|> +\[D N_{i} = F_{i}^{2050} + M_{i} \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[115, 399, 880, 483]]<|/det|> +where \(LCP_{max}\) represents the maximum manure Nr carrying capacity of cropland (around 144 kg N ha \(^{- 1}\) ) \(44\) , \(LI_{i}\) represents the total manure Nr (kg) in the county where the \(i\) th grid is located (kg), \(LCA_{i}\) represents the areas of large- scale cropland where the \(i\) th grid is located (km \(^2\) ), \(DNi\) represents the crop demand for Nr, defined as the sum of actual synthetic fertilizer inputs ( \(F_{i}\) , kg ha \(^{- 1}\) ) and manure Nr returned to cropland ( \(M_{i}\) , kg ha \(^{- 1}\) ) in 2017. + +<|ref|>text<|/ref|><|det|>[[115, 496, 881, 564]]<|/det|> +Given reducing the cost of demolition and construction, the livestock would be prioritized to move from the N overload grids into the surrounding grids. Statistics Tool in ArcGIS was used to calculate the priority of all non- overload grid \((LP_{i})\) with a value of 1 to the overload grid \((LRPi\leq 0)\) and 0 to the other grid \((LRPi > 0)\) , respectively. + +<|ref|>equation<|/ref|><|det|>[[440, 577, 880, 612]]<|/det|> +\[L P_{i} = \frac{N_{i}}{20\times 20} \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[115, 630, 853, 681]]<|/det|> +where \(N_{i}\) is the number of overloads grids within surrounding 400 grids and \(i\) th grids in the central of 400 grids. Based on the following linear programming, grids are selected for livestock moved in. + +<|ref|>equation<|/ref|><|det|>[[440, 682, 550, 739]]<|/det|> +\[m a x m i z e n\] \[s.t\sum_{j = l}^{n}L R P_{j,r}\leq P L R P_{r}\] + +<|ref|>text<|/ref|><|det|>[[115, 743, 870, 877]]<|/det|> +where \(j\) is the serial number for non- overload grids in descending order according to \(L R P_{j,r}\) , \(r\) represents province, \(PLRP_{r}\) is the livestock relocation capacity in the \(r\) th province. If the provincial livestock volume exceeds the provincial carrying capacity, it is assumed that the maximum livestock volume is matched on all the large- scale cropland. The excess livestock volume is moved to the nearest province, and the method of spatial allocation is the same as the intra- province relocation. Livestock cannot be relocated to provinces with restricted development based on the vulnerability level of the surface water, as determined by the Ministry of Agriculture and Rural Affairs (Table S11). + +<|ref|>text<|/ref|><|det|>[[115, 892, 875, 909]]<|/det|> +According to the National Key Research and Development Program in China, we assume that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 82, 880, 183]]<|/det|> +crop- livestock coupled would reduce \(50\% \mathrm{NH}_3\) in manure management45. This reduction can be achieved through various measures, including slurry acidification, covering slurry storages, and using closed manure composting technologies. Additionally, when manure is recycled to replace synthetic fertilizer, the \(\mathrm{NH}_3\) emission factor is assumed the same. Furthermore, due to strict water pollution regulations, direct discharge of \(\mathrm{NO}_3^-\) is assumed not allowed in \(2050^{15}\) . Table S4 provides further details on the optimization of various parameters. + +<|ref|>sub_title<|/ref|><|det|>[[115, 215, 614, 232]]<|/det|> +## Quantification of changes in urbanization on Nr pollution + +<|ref|>text<|/ref|><|det|>[[115, 231, 864, 348]]<|/det|> +Ammonia emission. Large- scale farming can increase the adoption of new technologies such as 4R nutrient stewardship (i.e., the right rate, type, time, and place of fertilizer application), optimal irrigation techniques, and organic amendments, resulting in reduced fertilizer application and improved nitrogen use efficiency. This, in turn, can reduce fertilizer \(\mathrm{NH}_3\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{N}\) loss to water. The estimation of influence of farm size on fertilizer use and \(\mathrm{NH}_3\) emission followed Duan et al., \(2021^{46}\) and Wang et al., \(2022^{47}\) . The equation (7) and (8) are used to calculate fertilizer use and \(\mathrm{NH}_3\) emission on county scale in 2050. + +<|ref|>equation<|/ref|><|det|>[[343, 360, 880, 383]]<|/det|> +\[F_{t}^{2050} = \exp \left(\alpha \times Ln X_{t}^{2017} - \alpha \times Ln X_{t}^{2050}\right)\times F_{t}^{2017} \quad (7)\] + +<|ref|>equation<|/ref|><|det|>[[310, 396, 880, 421]]<|/det|> +\[E_{t}^{2050} = \exp \left(\beta \times Ln X_{t}^{2017} - \beta \times Ln X_{t}^{2017}\right)\times E F_{t}^{2017}\times F_{t}^{2050} \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[115, 440, 872, 508]]<|/det|> +where \(F_{t}\) represents the synthetic fertilizer input in \(t\) th county (kg). \(X_{t}\) stands for the farm size in \(t\) th county (ha). \(E_{t}\) represents the total \(\mathrm{NH}_3\) emission in \(t\) th county (kg), \(EF_{t}\) represents \(\mathrm{NH}_3\) emission factor in \(t\) th county (%), \(\alpha , \beta\) refers to estimated coefficients of large- scale farming from Duan et al46 and Wang et al., \(2022^{47}\) . + +<|ref|>text<|/ref|><|det|>[[115, 522, 880, 655]]<|/det|> +Meanwhile, the management of livestock would be optimized through mitigation measures, such as surface coverage, frequent removal of manure from animal houses, and low crude protein feeding to reduce ammonia volatilization. We cited the research findings of Zhang et al., \(2021^{45}\) , which showed that a combination of different measures in three stages of feeding (including low- protein feed and phased feeding), housing (including ventilation systems, floor mats, dry manure removal, and solid- liquid separation), and manure management (including anaerobic digestion, and high- temperature composting) can achieve a \(60\%\) reduction in \(\mathrm{NH}_3\) emissions from the livestock farming process in China. + +<|ref|>text<|/ref|><|det|>[[115, 670, 880, 753]]<|/det|> +\(N_{r}\) loss to water. Since there is no direct report on the relationship between farm size and other Nr loss to water, the study assumed constant loss efficiency runoff he leaching in the field. The reduction of agricultural nitrogen losses may be underestimated here. In addition, the adoption of a crop- livestock coupled system can ensure that all manure is collected and applied to cropland45. + +<|ref|>text<|/ref|><|det|>[[115, 768, 870, 850]]<|/det|> +Industrial \(\mathrm{NO}_x\) emission. An inverted U- shaped relationship between urbanization and industrial pollutants in China was identified based on Stochastic Impacts by Regression on Population, Affluence, and Technology (STIRPAT) model and panel data48,49. The changes of industrial \(\mathrm{NO}_x\) emission (NE) and \(\mathrm{NO}_3^-\) discharge (ND) with urbanization were estimated as follow: + +<|ref|>equation<|/ref|><|det|>[[318, 848, 880, 870]]<|/det|> +\[NE_{g}^{2050} = \exp \left(\alpha \times Ln NE_{g}^{2017} - \alpha \times Ln NE_{g}^{2050}\right)\times NE_{g}^{2017} \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[115, 874, 880, 908]]<|/det|> +\(NE_{g}\) represent industrial \(\mathrm{NO}_x\) emission in the gth province (kg), \(\alpha\) refers to estimated coefficients from Xu et al48,49,48,49. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 98, 873, 265]]<|/det|> +Nitrogen budget. The CHANS model34,50 was used in this study to analyze the impact of urbanization on the national nitrogen budget (Extended Data Fig. 1). The CHANS model has 14 subsystems (e.g., croplands, livestock, human, industry, forest, grassland, atmosphere, water bodies, etc.) that covers all the nitrogen budget calculations in this paper. The baseline of nitrogen budget in 2017 was built first, then urbanization scenarios with corresponding parameters adjusting and urban/rural subsystem developing integrated into CHANS model to forecast the nitrogen cycle and loss in 2050. Urban/rural population change and accompanying difference of domestic pollution treatment efficiency are the key parameters which affect nitrogen budgets through urbanization. Details about the data sources, parameters and modeling methods can be found in Table S3 and Table S4. + +<|ref|>sub_title<|/ref|><|det|>[[115, 280, 574, 297]]<|/det|> +## Quantification of reductions in Nr losses on air \(\mathbf{PM}_{2.5}\) + +<|ref|>text<|/ref|><|det|>[[115, 297, 880, 447]]<|/det|> +Weather Research and Forecasting model coupled with Chemistry (WRF- Chem)51 was used to estimate the changes on surface concentrations of \(\mathrm{PM}_{2.5}\) based on \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emission reduction through well- managed urbanization. The baseline simulation in 2017 included all anthropogenic and natural emissions. Anthropogenic emissions data of air pollution was taken from the MEIC 2017 for mainland China52,53 and the MIX Asia emission inventory for the rest of the WRF- Chem domain54(www.meicmodel.org). The \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emission inventories were replaced with the results of this study, thus enabling the assessment of the impact of urbanization- reduced \(\mathrm{NH}_3\) and \(\mathrm{NO}_x\) emissions on \(\mathrm{PM}_{2.5}\) . More details for model simulation and validation are given in the Supplementary Methods. + +<|ref|>text<|/ref|><|det|>[[115, 460, 872, 511]]<|/det|> +The population- weighted \(\mathrm{PM}_{2.5}\) concentration \((PWC)\) , which considers population as weights at different exposure to \(\mathrm{PM}_{2.5}\) , is used to reflect the impact of \(\mathrm{PM}_{2.5}\) concentrations on the national population. It is defined as: + +<|ref|>equation<|/ref|><|det|>[[410, 512, 880, 574]]<|/det|> +\[PWC = \frac{\sum_{i = 1}^{n}(P_i\times C_i)}{P} \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[115, 574, 864, 626]]<|/det|> +where \(PWE\) is population- weighted \(\mathrm{PM}_{2.5}\) concentration in China, \(P_i\) is the population in the ith grid, \(n\) is the total number of grids in the China, \(C_i\) is the \(\mathrm{PM}_{2.5}\) concentration in the ith grid, and \(P\) is the total population of China. + +<|ref|>sub_title<|/ref|><|det|>[[115, 640, 537, 657]]<|/det|> +## Quantification the reductions of Nr output to sea + +<|ref|>text<|/ref|><|det|>[[115, 657, 872, 787]]<|/det|> +A water network- based framework (WNF) was used to estimate the changes on Nr output to sea based on Nr loss to surface water through well- managed urbanization. WNF is parsimonious solution for surface water modeling that incorporates topology structure, hydrological and biogeochemical processes. And it ensured that Nr transport processes were simulated in complete agreement and the loss of Nr to ocean is only influenced by the changes in Nr loss to surface water. Reactive nitrogen loss to surface runoff (SR) depend on the Nr input to the grid and the loss coefficient determined by the type of land use. + +<|ref|>equation<|/ref|><|det|>[[352, 794, 880, 833]]<|/det|> +\[SL = \sum_{k = 1}^{m}Input_{k}\times \alpha_{k} \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[115, 841, 875, 909]]<|/det|> +Where \(input_k\) represent Nr input from the \(k\) th source, covering all of anthropogenic and natural sources, i.g. cropland grid input includes fertilizer, deposition, irrigation, livestock manure, human excretion, cropland BNF and straw recycling. \(\alpha_k\) is the surface runoff loss coefficient. Details about Nr transfer process estimation during surface water based on WNF + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 450, 100]]<|/det|> +are given in the Supplementary Methods. + +<|ref|>sub_title<|/ref|><|det|>[[115, 116, 565, 133]]<|/det|> +## Quantification of costs and benefits of urbanization. + +<|ref|>text<|/ref|><|det|>[[115, 133, 825, 183]]<|/det|> +\(\mathbf{N}_{\mathrm{f}}\) related implementation cost with urbanization includes the costs of construction and operation of waste treatment facilities, as well as the costs of agricultural management optimization and industrial upgrading. + +<|ref|>text<|/ref|><|det|>[[115, 198, 877, 410]]<|/det|> +Cost of waste treatment. To assess unit cost of sewage/garbage facilities construction (CSCost and \(CGCost\) ), over 200 actual investment projects for waste treatment were collected from the Public- Private Partnership Service Platform under the National Development and Reform Commission of China (www.chinappp.cn). All cost data from the literature or web were measured in constant 2017 US\( (i.e., 100 CNY = US\) 16.16). Taking into account both population migration and economic development level, China was divided into three regions for investment assessment: a population inflow region (PIR), a developed outflow region (DOR) and a less developed outflow region (LOR), as detailed in Table S8. The numbers of projects and unit investment costs for each region are given in Table S9. In addition, the referenced daily unit costs of operating wastewater treatment plants, landfill plants and waste incineration plants (US\( 0.2, 6.5 and 23.0 t^{-1} \) were used to calculate the operating costs\(^{56- 58}\). Annual construction and operation costs of waste treatment facilities (CFC and OFC) were estimated on county scale as follows: + +<|ref|>equation<|/ref|><|det|>[[234, 425, 880, 464]]<|/det|> +\[CFC = \sum_{t,n}[(SD\times CSCost_{t} + GD\times CGCost_{t,n}\times R_{n})\times MP_{t}] / YS \quad (12)\] + +<|ref|>equation<|/ref|><|det|>[[203, 465, 880, 500]]<|/det|> +\[OFC = \sum_{t,n}[(SD\times OSCost_{t} + GD\times OGCost_{t,n}\times R_{n})\times MP_{t}]\times 365 \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[115, 502, 870, 653]]<|/det|> +where \(t\) represents \(>2800\) counties, \(SD\) and \(GD\) represent the daily production of sewage and garbage per capita (0.15 t capita \(^{- 1}\) and 1.17 kg capita \(^{- 1}\) , respectively \(^{59,60}\) ), \(MP_{t}\) represents the numbers of migrated people in the \(t\) th county (million capita), \(YS\) represents the time span from 2017 to 2050, equal to 33 years, \(CSCost\) and \(OSCost\) represent the unit cost for the construction and annual operation of sewage treatment facilities, respectively (US\( t^{- 1} \), \(CGCost\) and \(OGCost\) represent the unit cost for the construction and annual operation of garbage treatment facilities, respectively (US\( t^{- 1} \), \(n\) represents two types of garbage disposal, i.e. incineration and landfill and \(R_{n}\) is the predicted ratio of incineration and landfill (65% and 35%), referring to the national 14th Five year plan \(^{27}\) , + +<|ref|>text<|/ref|><|det|>[[115, 667, 880, 815]]<|/det|> +Cost of agricultural optimization. Agricultural management would be changed with urbanization in China, with the increasing on the size and scale of farmland and coupling crop- livestock production. Implementation cost of this agricultural optimization are calculated including homestead reclamation, cropland consolidation and livestock relocation. Cost data of homestead reclamation and cropland consolidation were collected from more than 200 projects in China Land Consolidation and Rehabilitation (CLCR, http://www.lerc.org.cn), and were divided into three regions according to migration and economic development levels (Table S10). Annual construction cost of rural reclamation and cropland consolidation (RRC and CCC) were estimated on county scale. + +<|ref|>equation<|/ref|><|det|>[[325, 828, 880, 905]]<|/det|> +\[\begin{array}{l}{RRC = \sum_{t}RA_{t}\times RRCost_{t} / YS}\\ {CCC = \sum_{t}CA_{t}\times CCCost_{t} / YS} \end{array} \quad (14)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 864, 150]]<|/det|> +where \(RA_{t}\) and \(CA_{t}\) are the areas of rural reclamation and cropland consolidation in the \(t\) th county (ha), respectively. \(YS\) represents the time span from 2017 to 2050 (33 years). \(RRCost\) and \(CCCost\) are the unit cost of rural reclamation and cropland consolidation (US\$ ha\(^{- 1}\)), respectively. + +<|ref|>text<|/ref|><|det|>[[115, 164, 876, 264]]<|/det|> +The cost of livestock relocation \((LC)\) mainly includes the costs of dismantling livestock farms and reconstruction new livestock farms. The cost of dismantling a livestock farm includes the cost of the farming facilities and the livestock in stock. Cost data on facilities and livestock \((FLCost\) and \(LLCost)\) were obtained from around 35 local compensation programs by local governments for the removal of livestock farming (Table S10). Reconstruction cost \((RCCost)\) was derived from National Compilation of Agricultural Cost- benefit information \(2018^{61}\) . + +<|ref|>equation<|/ref|><|det|>[[199, 277, 880, 313]]<|/det|> +\[LC = \sum_{t}[PR_{t}\times (\epsilon \times FLCost_{t} + LLCost_{t}) / YS] + \sum_{t}PI_{t}\times RCCost \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[115, 316, 860, 399]]<|/det|> +where \(PR_{t}\) represents the livestock that need to be moved out of \(t\) th county (unit pig), \(PI_{t}\) represents the livestock that need to be resettled in the \(t\) th county (unit pig), \(\epsilon\) represents the area of land occupied per unit pig \((2.3 \mathrm{m}^{2}\) per pig) \(^{62}\) , \(FLCost\) and \(LLCost\) represents the unit cost for removal facilities and livestock (US\$, respectively. \(RCCost\) represents the unit product cost of livestock (US\$)\$^{61}. + +<|ref|>text<|/ref|><|det|>[[115, 414, 870, 577]]<|/det|> +Cost of industrial upgrading. The implementation cost of industrial upgrading is defined as direct expenditure for the abatement measures with urbanization to reduce \(\mathrm{N}_{\mathrm{T}}\) emission. Here we mainly refer to the database and methodology of cost assessment from the online Greenhouse Gas and Air Pollution Interactions and Synergies (GAINS) model (https://gains.iiaa.ac.at/models/index.html) to calculate the abatement costs of industrial mitigating \(\mathrm{N}_{\mathrm{T}}\) emissions. International technology cost data have been taken into account and adjusted in GAINS for province- specific conditions in China, which take into account Consideration of local labour costs, energy prices, costs of by- products, etc. The costing module of GAINS has been described in detail by Klimont and Winiwarter \(^{63}\) and Zhang \(^{64}\) . The annual cost of implementation \((COST_{ind})\) is calculated as follows: + +<|ref|>equation<|/ref|><|det|>[[437, 576, 878, 613]]<|/det|> +\[IC = \sum_{t,q}\Delta NR_{t,q}\times ICost_{t,q} \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[115, 617, 880, 683]]<|/det|> +where \(q\) represents the forms of \(\mathrm{N}_{\mathrm{T}}\) losses including \(\mathrm{NH}_{3}\) , \(\mathrm{N}_{2}\mathrm{O}\) , \(\mathrm{NO}_{x}\) and \(\mathrm{NO}_{3}\) , \(\Delta NR\) is the reduction of \(\mathrm{N}_{\mathrm{T}}\) through industrial upgrading in county scale, \(ICost\) is the integrated mitigation unit cost of industrial upgrading to reduce \(\mathrm{N}_{\mathrm{T}}\) emission (US\$ per kg N) (Table S6), which is derived from the online GAINS model database. + +<|ref|>text<|/ref|><|det|>[[115, 697, 858, 749]]<|/det|> +Benefit assessment. The total benefit \((TB)\) of reducing \(\mathrm{N}_{\mathrm{T}}\) loss with urbanization include human health benefit \((HBq)\) , ecosystem benefit \((EBq)\) and climate benefit \((CBq)\) and economic benefit \((EcB)\) by reducing agricultural inputs and increasing total yield as follow: + +<|ref|>equation<|/ref|><|det|>[[354, 761, 878, 798]]<|/det|> +\[TB = \sum_{q}(HB_{q} + EB_{q} + CB_{q}) + EcB \quad (18)\] + +<|ref|>text<|/ref|><|det|>[[115, 803, 860, 902]]<|/det|> +where \(q\) represents the forms of \(\mathrm{N}_{\mathrm{T}}\) losses, i.e. \(\mathrm{NH}_{3}\) , \(\mathrm{NO}_{x}\) , \(\mathrm{N}_{2}\mathrm{O}\) and \(\mathrm{NO}_{3}\) . \(HB\) , \(EB\) , and \(CB\) denote social benefits, which need to be defined in conjunction to \(\mathrm{N}_{\mathrm{T}}\) loss and converted for monetization. The impact of \(\mathrm{N}_{\mathrm{T}}\) loss on human health is focused on \(\mathrm{PM}_{2.5}\) caused various disease, such as ischemic heart disease (IHD), lung cancer (LC) and so on. The impact on environment mainly refers to water pollution, i.e. eutrophication and biodiversity loss of aquatic ecosystems, due to \(\mathrm{NO}_{3}\) - loss to water, as well as soil acidification and biodiversity + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 875, 168]]<|/det|> +726 loss of terrestrial ecosystems due to Nr emission and deposition. The impact of \(\mathrm{N_r}\) loss on climate change is complex, with benefits from \(\mathrm{N}_2\mathrm{O}\) reduction and damages from \(\mathrm{NH}_3\) and \(\mathrm{NO_x}\) reduction that affecting carbon sequestration in forests. The monetary benefit of HB, EB and CB were estimated by multiplying social benefits per unit Nr by reduction in Nr with the following equations: + +<|ref|>equation<|/ref|><|det|>[[437, 164, 879, 240]]<|/det|> +\[\begin{array}{r}HB_{q} = \Delta NR_{q}\times HCost_{q}\\ EB_{q} = \Delta NR_{q}\times ECost_{q}\\ CB_{q} = \Delta NR_{q}\times CCost_{q} \end{array} \quad (19)\] + +<|ref|>text<|/ref|><|det|>[[115, 242, 876, 440]]<|/det|> +where \(\Delta NR_{q}\) is the reduction of \(q\) th Nr with urbanization in China (kg). \(HCost_{q}\) , \(ECost_{q}\) and \(Ccost_{q}\) are the unit damage cost of the three societal benefit due to the \(q\) th Nr loss (US\$ kg \(\mathrm{N}^*\) ), respectively. \(HCost_{q}\) was calculated by the cause- specific integrated exposure- response functions which can estimate the relative mortality risk from exposure to \(\mathrm{PM}_{2.5}^{65}\) . \(ECost_{q}\) was derived from the nitrogen assessment in Europe after correction for difference in the willingness to pay (WTP) and the purchasing power parity (PPP) between EU and China \(^{66,67}\) . \(Ccost_{q}\) was calculated based on the marginal abatement cost (the carbon price) to reduce or increase emission of CO2- eq \(^{68}\) . A detailed method description of the unit damage cost of varying forms of N loss can be found in Table S7. HB, EB, and CB are assigned to counties based on relevant parameters, for example, changes in air and water pollution, population density, and VSL were used to correct for HB. More details are given in the Supplementary Methods. + +<|ref|>text<|/ref|><|det|>[[115, 454, 880, 504]]<|/det|> +Agricultural management optimization during urbanization increases economic benefit ( \(EcB\) ) by reducing agricultural inputs such as fertilizers, and increasing total yield due to cropland area increase. + +<|ref|>equation<|/ref|><|det|>[[318, 518, 879, 555]]<|/det|> +\[EcB = \sum_{i}(A_{i}\times Y_{i}^{2017}\times PC + F_{i}\times FC) \quad (22)\] + +<|ref|>text<|/ref|><|det|>[[115, 556, 881, 738]]<|/det|> +where \(t\) represents \(>2800\) counties; \(\mathrm{A_i}\) is the increasing potential of cropland due to homestead reclamation in the \(t\) th county (ha); \(Y_{i}\) is crop yield in the \(t\) th county in 2017 (kg ha \(^{- 1}\) ), which is derived from the National Bureau of Statistics (www. stats.gov.cn); \(PC\) is the average market price of the wheat in China in 2017 (US\$ 0.38 kg \(^{- 1}\) ), which was from National Development and Reform Commission (https://www.ndrc.gov.cn/xxgk/zcfb/tz/201610/t20161021_963246. html); \(F_{i}\) represents the reduction of synthetic fertilizer input of \(t\) th county related to an increased NUE, associated with an increased farm size, as given in Eq.7 (kg); \(FC\) is the unit cost of Nr fertilizer (US\$ 0.66 kg \(\mathrm{N}^{- 1}\) ), estimated with reference to the 2017 urea market price from the price monitoring network of the Ministry of Agriculture and Rural Affairs, PRC (http://zdscxx.moa.gov.cn:8080/nyb/pc/search.jsp). + +<|ref|>sub_title<|/ref|><|det|>[[115, 753, 210, 768]]<|/det|> +## Reference: + +<|ref|>text<|/ref|><|det|>[[115, 769, 881, 900]]<|/det|> +1 Erisman, J. W., Sutton, M. A., Galloway, J., Klimont, Z. & Winiwarter, W. How a century of ammonia synthesis changed the world. Nat. Geosci. 1 636- 639 (2008). 2 Erisman, J. W. et al.. Consequences of human modification of the global nitrogen cycle. Philos Trans R Soc Lond B Biol Sci 368 20130116 (2013). 3 Galloway, J. N. et al.. Transformation of the Nitrogen Cycle: Recent Trends, Questions, and Potential Solutions. Science 320 889- 892 (2008). 4 Harris, E. et al.. Warming and redistribution of nitrogen inputs drive an increase in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 78, 884, 901]]<|/det|> +terrestrial nitrous oxide emission factor. Nat. Commun. 13 (2022). Nitrogen Challenges in Agri- food systems: Halve Nitrogen Waste by 2030. (UNEP, 2020) https://www.unep.org/events/webinar/nitrogen- challenges- agri- food- systems- halve- nitrogen- waste- 2030 UNEP. Kunming Montreal Global Biodiversity Framework, CBD/COP/15/L.25, Target 7. (UNEP, Montreal, 2022). Morseletto, P. Confronting the nitrogen challenge: Options for governance and target setting. Global Environmental Change 54 40- 49 (2019). Sutton, M. A. et al.. The nitrogen decade: mobilizing global actionnonnitrogento2030and beyond. One Earth 4 10- 14 (2021). FAOSTAT: FAO Statistical Databases. (FAO, 2022) https://www.fao.org/statistics/en/ Yu, C. et al.. Managing nitrogen to restore water quality in China. Nature 567 516- 520 (2019). Gu, B. et al.. Abating ammonia is more cost- effective than nitrogen oxides for mitigating PM2.5 air pollution. Science 374 758- 762 (2021). Yun, X. et al.. Residential solid fuel emissions contribute significantly to air pollution and associated health impacts in China. Sci Adv 6 a7621 (2020). Gu, B. et al.. A Credit System to Solve Agricultural Nitrogen Pollution. The Innovation 2 100079 (2021). Shen, G. et al.. Impacts of air pollutants from rural Chinese households under the rapid residential energy transition. Nat. Commun. 10 (2019). Rose, T. J., Wood, R. H., Rose, M. T. & Van Zwieten, L. A re- evaluation of the agronomic effectiveness of the nitrification inhibitors DCD and DMPP and the urease inhibitor NBPT. Agriculture, Ecosystems & Environment 252 69- 73 (2018). Wu, Y. et al.. Policy distortions, farm size, and the overuse of agricultural chemicals in China. Proc Natl Acad Sci 115 7010- 7015 (2018). Ren, C., Zhang, X., Reis, S. & Gu, B. Socioeconomic barriers of nitrogen management for agricultural and environmental sustainability. Agriculture, Ecosystems & Environment 333 107950 (2022). Bren D Amour, C. et al.. Future urban land expansion and implications for global croplands. Proc Natl Acad Sci 114 8939- 8944 (2017). Shi, K. et al.. Urban Expansion and Agricultural Land Loss in China: A Multiscale Perspective. Sustainability- Basel 8 790 (2016). Wang, S. et al.. Urbanization can benefit agricultural production with large- scale farming in China. Nature Food 2 183- 191 (2021). Gu, B., Zhang, X., Bai, X., Fu, B. & Chen, D. Four steps to food security for swelling cities. Nature 566 31- 33 (2019). Lu, Y. et al.. Forty years of reform and opening up: China's progress toward a sustainable path. Sci Adv 5 9413 (2019). Bai, X., Shi, P. & Liu, Y. Society: Realizing China's urban dream. Nature 509 158- 160 (2014). Jin, S. et al.. Decoupling livestock and crop production at the household level in China. Nature Sustainability 4 48- 55 (2021). Zhu, Z. et al.. Integrated livestock sector nitrogen pollution abatement measures could + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 881, 120]]<|/det|> +generate net benefits for human and ecosystem health in China. Nature Food 3 161- 168 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 121, 730, 139]]<|/det|> +Gu, B. Recoupling livestock and crops. Nature Food 3 102- 103 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 140, 881, 194]]<|/det|> +The Outline of the 14th Five- Year Plan (2021- 2025) for National Economic and Social Development and Vision 2035 of the People's Republic of China. (The State Council, PRC, 2021) http://www.gov.cn/xinwen/2021- 03/13/content_5592681. htm + +<|ref|>text<|/ref|><|det|>[[145, 195, 881, 248]]<|/det|> +Beckinghausen, A., Odlare, M., Thorin, E. & Schwede, S. From removal to recovery: An evaluation of nitrogen recovery techniques from wastewater. Appl Energ 263 114616 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 250, 881, 285]]<|/det|> +Zhang, X. et al. Managing nitrogen for sustainable development. Nature 528 51- 59 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 286, 881, 320]]<|/det|> +Liu, H., Wang, C., Zhang, M. & Wang, S. Evaluating the effects of air pollution control policies in China using a difference- in- differences approach. Sci Total Environ 845 157333 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 323, 881, 359]]<|/det|> +Shen, H. et al. Urbanization- induced population migration has reduced ambient PM2.5 concentrations in China. Sci Adv 3 e1700300 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 360, 881, 395]]<|/det|> +Tian, H. et al. A comprehensive quantification of global nitrous oxide sources and sinks. Nature 586 248- 256 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 397, 881, 433]]<|/det|> +IPCC. 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. (IPCC, Genève, Japan., 2019). + +<|ref|>text<|/ref|><|det|>[[145, 435, 881, 470]]<|/det|> +Gu, B., Ju, X., Chang, J., Ge, Y. & Vitousek, P. M. Integrated reactive nitrogen budgets and future trends in China. Proc. Natl. Acad. Sci. U.S.A. 112 8792- 8797 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 472, 881, 507]]<|/det|> +MHC: Sewage Treatment Rate in Rural Areas was 22% in 2016. (China Daily, 2016) http://www.aswater.com.cn/page196?article_id=102 + +<|ref|>text<|/ref|><|det|>[[145, 508, 881, 561]]<|/det|> +Urban domestic waste classification and treatment facilities development plan in 14th five year. (National Development and Reform Commission, PRC, 2021) https://www.ndrc.gov.cn/xxgk/zcfb/tz/202105/t20210513_1279763_ext.html + +<|ref|>text<|/ref|><|det|>[[145, 563, 881, 598]]<|/det|> +Ma, T. et al. China's improving inland surface water quality since 2003. Sci Adv 6 3798 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 600, 881, 635]]<|/det|> +Ambient air quality standards GB30952012. (Ministry of Ecology and Environment, PRC, 2012) https://www.mee.gov.cn/ywgz/fgbz/bz/bwdb/dqhjbdb/dqhjzlbz/ + +<|ref|>text<|/ref|><|det|>[[145, 636, 881, 671]]<|/det|> +Resource and Environment Data Cloud Platform. (Resource and Environment Science and Data Center, 2018) http://www.resdc.cn/Default.aspx + +<|ref|>text<|/ref|><|det|>[[145, 673, 881, 727]]<|/det|> +The National High- standard Farmland Construction Plan (2021- 2030). (Ministry of Agriculture and Rural Affairs, PRC, 2021) http://www.ntjss.moa.gov.cn/zcfb/202109/t20210915_6376511. htm + +<|ref|>text<|/ref|><|det|>[[145, 728, 881, 781]]<|/det|> +Notice on the implementation of livestock and poultry manure resource utilization projects in 2019. (Ministry of Agriculture and Rural Affairs, PRC, 2019) http://www.moa.gov.cn/govpublic/SYJ/201904/t20190426_6212935. htm + +<|ref|>text<|/ref|><|det|>[[145, 783, 881, 836]]<|/det|> +Cheng, L. et al. A 12% switch from monogastric to ruminant livestock production can reduce emissions and boost crop production for 525 million people. Nature Food 3 1040- 1051 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 839, 881, 874]]<|/det|> +Wu, H. et al. A new dietary guideline balancing sustainability and nutrition for China's rural and urban residents. iScience 25 105048 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 876, 881, 894]]<|/det|> +Technical Guide for Measuring the Land Carrying Capacity of Livestock and Poultry + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 85, 881, 120]]<|/det|> +Manure. (Ministry of Agriculture and Rural Affairs, PRC, 2018) http://www.xmsyj.moa.gov.cn/zcjd/201904/t20190429_6287183. htm + +<|ref|>text<|/ref|><|det|>[[145, 122, 881, 157]]<|/det|> +Bai, Z. et al.. Relocate 10 billion livestock to reduce harmful nitrogen pollution exposure for \(90\%\) of China' s population. Nature Food 3 152- 160 (2022). + +<|ref|>text<|/ref|><|det|>[[145, 158, 881, 193]]<|/det|> +Duan, J. et al.. Consolidation of agricultural land can contribute to agricultural sustainability in China. Nature Food 2 1014- 1022 (2021). + +<|ref|>text<|/ref|><|det|>[[145, 195, 881, 230]]<|/det|> +Wang, C. et al.. Ammonia Emissions from Croplands Decrease with Farm Size in China. Environ Sci Technol (2022). + +<|ref|>text<|/ref|><|det|>[[145, 232, 881, 267]]<|/det|> +Xu, F. et al.. Reexamining the relationship between urbanization and pollutant emissions in China based on the STIRPAT model. J. Environ Manage 273 111134 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 269, 881, 322]]<|/det|> +Xu, S. et al.. Regional differences in nonlinear impacts of economic growth, export and FDI on air pollutants in China based on provincial panel data. J. Clean Prod 228 455- 466 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 325, 881, 360]]<|/det|> +Gu, B. et al.. Toward a Generic Analytical Framework for Sustainable Nitrogen Management: Application for China. Environ Sci Technol 53 1109- 1118 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 361, 881, 396]]<|/det|> +Grell, G. A. et al.. Fully coupled "online" chemistry within the WRF model. Atmos Environ 39 6957- 6975 (2005). + +<|ref|>text<|/ref|><|det|>[[145, 398, 881, 452]]<|/det|> +Li, M. et al.. Persistent growth of anthropogenic non- methane volatile organic compound (NMVOC) emissions in China during 1990 - 2017: drivers, speciation and ozone formation potential. Atmos Chem Phys 19 8897- 8913 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 454, 881, 489]]<|/det|> +Zheng, B. et al.. Trends in China's anthropogenic emissions since 2010 as the consequence of clean air actions. Atmos Chem Phys 18 14095- 14111 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 491, 881, 544]]<|/det|> +Li, M. et al.. MIX: a mosaic Asian anthropogenic emission inventory under the international collaboration framework of the MICS- Asia and HTAP. Atmos Chem Phys 17 935- 963 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 546, 881, 581]]<|/det|> +Xia, Y. et al.. A new framework to model the distributed transfer and retention of nutrients by incorporating topology structure of small water bodies. Water Res 238 119991 (2023). + +<|ref|>text<|/ref|><|det|>[[145, 583, 881, 618]]<|/det|> +Zhou, C., Gong, Z., Hu, J., Cao, A. & Liang, H. A cost- benefit analysis of landfill mining and material recycling in China. Waste Manage 35 191- 198 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 620, 881, 655]]<|/det|> +Zhao, X., Jiang, G., Li, A. & Wang, L. Economic analysis of waste- to- energy industry in China. Waste Manage 48 604- 618 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 657, 881, 710]]<|/det|> +Tan, X., Shi, L., Ma, Z., Zhang, X. & Lu, G. Institutional analysis of sewage treatment charge based on operating cost of sewage treatment plant—an empirical research of 227 samples in China. China Environmental Science 12 3833- 3840 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 712, 881, 747]]<|/det|> +Wei, X. et al.. Temporal and spatial characteristics of municipal solid waste generation and treatment in China from 1979 to 2016. China Environmental Science 38 3833- 3843 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 749, 881, 802]]<|/det|> +Handbook of accounting methods and coefficients for the statistical investigation of emission sources. (Ministry of Ecology and Environment, PRC, 2021) https://www.mee.gov.cn/xxgk2018/xxgk/xxgk01/202106/t20210618_839512. html (2021). + +<|ref|>text<|/ref|><|det|>[[145, 804, 881, 839]]<|/det|> +China Agricultural Products Cost- Benefit Yearbook 2018. (NDRC, 2018) https://data.cnki.net/trade/Yearbook/Single/N2019010190?zcode=Z009 + +<|ref|>text<|/ref|><|det|>[[145, 841, 881, 896]]<|/det|> +Zhou, D. Improving the old- fashioned traditional pig houses in mountainous rural areas to achieve both environmental protection and comprehensive production benefits. Chinese Journal of Animal Husbandry and Veterinary Medicine 73- 74 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 81, 884, 362]]<|/det|> +903 Klimont, Z. & Winiwarter, W. Integrated ammonia abatement - Modelling of emission 905 control potentials and costs in GAINS. (IIASA, Laxenburg, Austria, 2011). 906 Zhang, X. et al.. Societal benefits of halving agricultural ammonia emissions in China far 907 exceed the abatement costs. Nat. Commun. 11 (2020). 908 Gu, B. et al.. Abating ammonia is more cost- effective than nitrogen oxides for mitigating 909 \(\mathrm{PM}_{2.5}\) air pollution. Science 374 758- 762 (2021). 910 Van Grinsven, H. J. M. et al.. Costs and benefits of nitrogen for Europe and implications 911 for mitigation. Environ Sci Technol 47 3571- 3579 (2013). 912 Sobota, D., Compton, J., McCrackin, M. & Singh, S. Cost of reactive nitrogen release from 913 human activities to the environment in the United States. Environ. Res. Lett. 10 25006 914 (2015). 915 West, J. J. et al.. Co- benefits of mitigating global greenhouse gas emissions for future air 916 quality and human health. Nat. Clim. Change 3 885- 889 (2013). 917 Lesiv, M. et al.. Estimating the global distribution of field size using crowdsourcing. Global 918 Change Biol 25 174- 186 (2018). + +<|ref|>sub_title<|/ref|><|det|>[[117, 377, 262, 393]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 393, 761, 428]]<|/det|> +Data supporting the findings of this study are available within the article and its supplementary information files. + +<|ref|>sub_title<|/ref|><|det|>[[117, 443, 286, 459]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[115, 459, 881, 558]]<|/det|> +This study was supported by the National Natural Science Foundation of China (42261144001 and 42061124001), National Key Research and Development Project of China (2022YFD1700014 and 2022YFE010118). This work is a contribution from Activity 1.4 to the 'Towards the International Nitrogen Management System' project (INMS, http://www.inms.international/) funded by the Global Environment Facility (GEF) through the United Nations Environment Programme (UNEP). + +<|ref|>sub_title<|/ref|><|det|>[[117, 574, 302, 589]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[115, 590, 880, 657]]<|/det|> +B.G. conceived the study. O.D., J.R. and S.W. performed the urbanization modeling. J.D. and S.H. performed the agricultural optimization modeling. O.D. and X.Z. performed the CHANS modeling and cost- benefit. L.Z., Y.X. and Z.X. performed the air and water quality estimation. O.D. and B.G. wrote the first draft, while S.R., J.X. W.V and M.A.S revised the paper. + +<|ref|>sub_title<|/ref|><|det|>[[117, 672, 291, 688]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[115, 688, 515, 705]]<|/det|> +All authors have no conflicts of interest to report. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[135, 90, 765, 537]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 546, 881, 660]]<|/det|> +
Fig. 1 | Population, land use and agricultural change with urbanization. These figures demonstrate changes in population, land use and agricultural system with urbanization from 2017 to 2050. a, Geo-distribution of population change. b, Change of population. c, Geo-distribution of land use change, with the same colors as in d. d, Quantity change of land use. e, Geo-distribution of urbanization change. f, Geo-distribution of cropland change. g, Geo-distribution of livestock change. The base map is applied without endorsement from GADM data (https://gadm.org/).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[230, 95, 771, 441]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 456, 625, 473]]<|/det|> +
Fig. 2 | Spatial variation of nitrogen loss with urbanization.
+ +<|ref|>text<|/ref|><|det|>[[115, 473, 883, 592]]<|/det|> +a, Distribution of total reactive nitrogen (TN) losses in 2017. b, Distribution of predicted TN losses in 2050 assuming an urbanization level of \(80\%\) . c, Distribution of the change in TN losses during the 2017- 2050 period. d, National \(\mathrm{NH}_3\) , \(\mathrm{NO_x}\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{NO}_3\) - loss in 2017 and 2050. TN losses include \(\mathrm{NH}_3\) , \(\mathrm{NO_x}\) , \(\mathrm{N}_2\mathrm{O}\) and \(\mathrm{NO}_3\) - loss. We use \(\mathrm{NO}_3\) - to represent the \(\mathrm{N}_r\) loss via runoff and leaching, given that most of the other dissolve forms of \(\mathrm{N}_r\) would be converted to \(\mathrm{NO}_3\) - during moving with water. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[166, 81, 848, 368]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 373, 816, 405]]<|/det|> +
Fig. 3 | Nitrogen budgets within key subsystems and their contribution to N losses reduction with urbanization.
+ +<|ref|>text<|/ref|><|det|>[[115, 405, 882, 555]]<|/det|> +a, b, Total N budgets for four subsystems and the environment in 2017 (a) and 2050 (b). These four subsystems, i.e. cropland, livestock, human and nature, are defined by the CHANS model (Fig. S1). Human subsystem set includes urban human, rural human, wastewater treatment, garbage treatment and industry subsystems. Nature subsystem set includes grassland and forest subsystem. The colored arrows represent N flows to varying subsystems as follows: yellow, cropland subsystem; green, livestock subsystem; blue, human subsystem; brown, nature subsystem. The black arrows indicate the N flows that eventually enter the atmosphere and water bodies through gaseous emissions and runoff losses. c-f, National nitrogen losses in 2017 and 2050, and the sources of the difference. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 90, 795, 707]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 710, 638, 727]]<|/det|> +
Fig. 4 | Spatial variation of \(\mathrm{Nr}\) species loss with urbanization.
+ +<|ref|>text<|/ref|><|det|>[[115, 727, 883, 794]]<|/det|> +(a-c), Distribution of \(\mathrm{NH_3}\) in 2017 (a), 2050 (b) and the difference (c). (d-f), Distribution of \(\mathrm{NO_x}\) in 2017 (d), 2050 (e) and the difference (f). (g-i), Distribution of \(\mathrm{N_2O}\) in 2017 (g), 2050 (h) and the difference (i). (j-l), Distribution of runoff \(\mathrm{N}\) loss in 2017 (j), 2050 (k) and the difference (l). The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[142, 100, 789, 388]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 392, 819, 425]]<|/det|> +
Fig. 5 | Spatial variation of atmospheric PM2.5 concentration and \(\mathbf{N}_{\mathrm{r}}\) to ocean with urbanization.
+ +<|ref|>text<|/ref|><|det|>[[115, 425, 881, 494]]<|/det|> +a- c, Geographic distribution of surface air PM2.5 concentration in 2017 (a), 2050 (b), and the difference (c). d- f, Geographic distribution of \(\mathbf{N}_{\mathrm{r}}\) loading to downstream reach in 2017 (d), 2050 (e), and the difference (f). The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 90, 857, 608]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 616, 881, 755]]<|/det|> +
Fig. 6 | Costs and benefits of halving \(\mathbf{N}_{\mathrm{r}}\) pollution with urbanization. a-c, Geographic distribution of implementation cost on waste treatment (a), agricultural optimization (b), and industrial pollution system upgrades (c) to achieve the above mentioned reduction potential of \(\mathbf{N}_{\mathrm{r}}\) pollution with urbanization. d-f, Geographic distribution of monetized benefits on climate (d), human health (e) and ecosystem (f). g, Geographic distribution of economic benefit from fertilizer saving and yield increase. h, National costs and benefits with urbanization. Blue indicates implementation costs, while red indicates benefits. The base map is applied without endorsement from GADM data (https://gadm.org/).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[196, 95, 800, 293]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[113, 310, 875, 412]]<|/det|> +Extended Data Fig. 1 | Mechanism of N loss reduction with urbanization. Rural- urban migration benefits the management of domestic sewage and waste due to centralized waste management and pollution control measures. Meanwhile, quantity and scale of croplands and the coupling of crop and livestock production would be increased according to rural reclamation and rural population reduction. These processes are beneficial for improving N use efficiency (NUE) and reducing N losses. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[140, 101, 860, 262]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 280, 875, 333]]<|/det|> +Extended Data Fig. 2 | Changes on population with urbanization. a- c, Geo- distribution of population in 2017, 2050 and the difference. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[173, 88, 816, 230]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[115, 245, 816, 279]]<|/det|> +## Extended Data Fig. 3 | Field size distribution through cropland consolidation and urbanization. + +<|ref|>text<|/ref|><|det|>[[115, 280, 880, 365]]<|/det|> +a- c, The geographic distribution of cropland field size in 2017 (a), and integration of consolidation and urbanization in 2050 (b). c, Field size share with cropland consolidation and urbanization. Data of current field size in 2017 and potential field size after cropland consolidation were derived from Lesiv et al., 201869 and Duan et al., 202146, respectively. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[175, 106, 840, 425]]<|/det|> + +<|ref|>title<|/ref|><|det|>[[115, 432, 740, 449]]<|/det|> +# Extended Data Fig. 4 | Livestock change through crop-livestock coupled. + +<|ref|>text<|/ref|><|det|>[[115, 449, 875, 551]]<|/det|> +Followed by the distribution of large- scale farms in 2050, livestock would be relocated with constrains that livestock manure production does not exceed the carrying capacity of crop requirement on county scale. a- c, Geographic distribution of livestock in 2017 (a), 2050 (b) and difference (c). d, The number of livestock moving into the province, out of the province, and relocation within province in 31 provinces. The full names of 31 provinces are shown in table S6. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 105, 770, 430]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[117, 444, 866, 462]]<|/det|> +Extended Data Fig. 5 | Nitrogen emission in urban and rural waste treatment system + +<|ref|>text<|/ref|><|det|>[[117, 462, 866, 555]]<|/det|> +The diagram includes traditional and well- managed urban and rural waste treatment systems. The red arrows represent traditional treatment path; the blue arrows represent harmless treatment methods. The yellow box represents the final discharge form of nitrogen; the gray box represents ancillary facilities in the harmless treatment process, including toilets, waste transfer systems, wastewater treatment plants (WTP), landfills, and incineration plants; the white boxes represent types of waste from urban or rural areas. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 90, 777, 535]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[50, 540, 88, 552]]<|/det|> +1039 + +<|ref|>text<|/ref|><|det|>[[50, 555, 88, 567]]<|/det|> +1040 + +<|ref|>text<|/ref|><|det|>[[50, 570, 85, 582]]<|/det|> +1041 + +<|ref|>text<|/ref|><|det|>[[50, 585, 88, 598]]<|/det|> +1042 + +<|ref|>text<|/ref|><|det|>[[50, 601, 88, 614]]<|/det|> +1043 + +<|ref|>text<|/ref|><|det|>[[50, 616, 88, 629]]<|/det|> +1044 + +<|ref|>text<|/ref|><|det|>[[50, 632, 88, 644]]<|/det|> +1045 + +<|ref|>text<|/ref|><|det|>[[50, 647, 88, 660]]<|/det|> +1046 + +<|ref|>text<|/ref|><|det|>[[50, 663, 88, 675]]<|/det|> +1047 + +<|ref|>text<|/ref|><|det|>[[50, 678, 88, 691]]<|/det|> +1048 + +<|ref|>text<|/ref|><|det|>[[50, 694, 88, 707]]<|/det|> +1049 + +<|ref|>text<|/ref|><|det|>[[50, 710, 88, 722]]<|/det|> +1050 + +<|ref|>text<|/ref|><|det|>[[50, 725, 88, 737]]<|/det|> +1051 + +<|ref|>text<|/ref|><|det|>[[50, 740, 88, 752]]<|/det|> +1052 + +<|ref|>text<|/ref|><|det|>[[50, 755, 88, 767]]<|/det|> +1053 + +<|ref|>text<|/ref|><|det|>[[50, 770, 88, 782]]<|/det|> +1054 + +<|ref|>text<|/ref|><|det|>[[50, 785, 88, 797]]<|/det|> +1055 + +<|ref|>text<|/ref|><|det|>[[50, 800, 88, 812]]<|/det|> +1056 + +<|ref|>text<|/ref|><|det|>[[50, 815, 88, 827]]<|/det|> +1057 + +<|ref|>text<|/ref|><|det|>[[50, 830, 88, 842]]<|/det|> +1058 + +<|ref|>text<|/ref|><|det|>[[50, 845, 88, 857]]<|/det|> +1059 + +<|ref|>text<|/ref|><|det|>[[50, 860, 88, 872]]<|/det|> +1060 + +<|ref|>text<|/ref|><|det|>[[50, 875, 88, 887]]<|/det|> +1061 + +<|ref|>text<|/ref|><|det|>[[50, 890, 88, 902]]<|/det|> +1062 + +<|ref|>text<|/ref|><|det|>[[50, 905, 88, 917]]<|/det|> +1063 + +<|ref|>text<|/ref|><|det|>[[50, 920, 88, 932]]<|/det|> +1064 + +<|ref|>text<|/ref|><|det|>[[50, 935, 88, 947]]<|/det|> +1065 + +<|ref|>text<|/ref|><|det|>[[50, 950, 88, 961]]<|/det|> +1066 + +<|ref|>text<|/ref|><|det|>[[50, 964, 88, 976]]<|/det|> +1067 + +<|ref|>text<|/ref|><|det|>[[50, 979, 88, 991]]<|/det|> +1068 + +<|ref|>text<|/ref|><|det|>[[50, 984, 88, 996]]<|/det|> +1069 + +<|ref|>text<|/ref|><|det|>[[50, 989, 88, 997]]<|/det|> +1070 + +<|ref|>text<|/ref|><|det|>[[50, 989, 88, 997]]<|/det|> +1080 + +<|ref|>text<|/ref|><|det|>[[50, 989, 88, 997]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[133, 90, 770, 675]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[115, 710, 858, 744]]<|/det|> +Extended Data Fig. 7 | Geographic distribution changes of \(\mathbf{N}_{\mathrm{r}}\) emissions through three urbanization processes. + +<|ref|>text<|/ref|><|det|>[[115, 744, 880, 880]]<|/det|> +a- c, Geographic distribution changes of \(\mathrm{NH}_3\) through population migration (PM), agricultural optimization(AO) and industrial upgrading(IU), respectively. d- f, Geographic distribution changes of \(\mathrm{NO_x}\) through PM, AO and IU, respectively. g- i, Geographic distribution changes of \(\mathrm{NO_3}\) through PM, AO and IU, respectively. j- l, Geographic distribution changes of \(\mathrm{NO_3}\) through PM, AO and IU,respectively. Since industrial upgrading primarily focuses on reducing \(\mathrm{NO_x}\) emissions and has little impact on \(\mathrm{NH}_3\) and \(\mathrm{N}_2\mathrm{O}\) emissions, there is minimal change in c and i. The base map is applied without endorsement from GADM data (https://gadm.org/). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[178, 92, 800, 496]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[113, 512, 884, 568]]<|/det|> +Extended Data Fig. 8 | The Geographic distribution of N₂ emission with urbanization. a-c, Spatial distribution of N₂ emissions in 2017 (a) and in 2050 (b), and the difference (c). d, National N₂ emissions in 2017 and 2050, and the sources of the difference. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 95, 860, 476]]<|/det|> + +<|ref|>sub_title<|/ref|><|det|>[[115, 480, 822, 515]]<|/det|> +## Extended Data Fig. 9| Methods for quantifying effects of urbanization on nitrogen pollution. + +<|ref|>text<|/ref|><|det|>[[115, 515, 877, 666]]<|/det|> +Graphical Model for urbanization process represents the whole simulation process, including migration, urban expansion, industrial upgrading, cropland consolidation, livestock relocation, and rural reclamation. It changes the whole nitrogen cycling on national scale. The urbanization process covering the period from 2017 to 2050 would eventually reduce \(\mathrm{N}_{\mathrm{r}}\) losses, which enhance air and water quality. The red boxes represent the estimation of costs incurred through urbanization. Abbreviations: NTE, Nitrogen Treatment Efficiencies; NTE, Nitrogen Use Efficiency; NRR, Nitrogen Recycle Ratio; WNRR, Water Nitrogen Retention and Removing model; WRF-Chem, Weather Research and Forecasting model coupled with Chemistry. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 356, 149]]<|/det|> +SIUrbanizationXNCv.0707. docx + +<--- Page Split ---> diff --git a/preprint/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb_det.mmd b/preprint/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..c80b44ed9ffa403fec90442923e76d5ff8861a97 --- /dev/null +++ b/preprint/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb/preprint__1447393825288b856af8401fb14f38d4fcedf8dddf51cce59d4fb8548d07b4fb_det.mmd @@ -0,0 +1,639 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 914, 210]]<|/det|> +# ASPP2/PP1 complexes maintain the integrity of pseudostratified epithelia undergoing remodelling during morphogenesis + +<|ref|>text<|/ref|><|det|>[[44, 228, 530, 250]]<|/det|> +Christophe Royer ( \(\boxed{ \begin{array}{r l} \end{array} }\) christophe.royer@dpag.ox.ac.uk) + +<|ref|>text<|/ref|><|det|>[[44, 252, 231, 270]]<|/det|> +University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 277, 231, 316]]<|/det|> +Elizabeth Sandham University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 323, 231, 363]]<|/det|> +Elizabeth Slee University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 369, 231, 408]]<|/det|> +Jonathan Godwin University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 415, 231, 454]]<|/det|> +Nisha Veits University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 460, 231, 500]]<|/det|> +Holly Hathrell University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 506, 231, 546]]<|/det|> +Felix zhou University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 553, 231, 592]]<|/det|> +karolis Leonavicius University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 599, 231, 639]]<|/det|> +Jemma Garratt University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 645, 231, 685]]<|/det|> +Tanaya Narendra University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 691, 488, 732]]<|/det|> +Anna Vincent Oxford Fertility, Institute of Reproductive Sciences + +<|ref|>text<|/ref|><|det|>[[44, 738, 231, 778]]<|/det|> +Celine Jones University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 784, 231, 824]]<|/det|> +Tim Child University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 830, 231, 870]]<|/det|> +Kevin Coward University of Oxford + +<|ref|>text<|/ref|><|det|>[[44, 876, 208, 916]]<|/det|> +Chris Graham Oxford University + +<|ref|>text<|/ref|><|det|>[[44, 922, 102, 940]]<|/det|> +Xin Lu + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 45, 588, 111]]<|/det|> +University of Oxford https://orcid.org/0000- 0002- 6587- 1152Shankar SrinivasUniversity of Oxford https://orcid.org/0000- 0001- 5726- 7791 + +<|ref|>sub_title<|/ref|><|det|>[[44, 152, 102, 170]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 189, 692, 210]]<|/det|> +Keywords: ASPP2/PP1 complex, pseudostratified epithelia, morphogenesis + +<|ref|>text<|/ref|><|det|>[[44, 228, 336, 247]]<|/det|> +Posted Date: December 9th, 2020 + +<|ref|>text<|/ref|><|det|>[[44, 266, 463, 286]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 107253/v1 + +<|ref|>text<|/ref|><|det|>[[44, 303, 909, 346]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 380, 945, 424]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 17th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28590- 4. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[45, 87, 874, 190]]<|/det|> +# ASPP2/PP1 complexes maintain the integrity of pseudostratified epithelia undergoing remodelling during morphogenesis + +<|ref|>text<|/ref|><|det|>[[85, 210, 911, 270]]<|/det|> +Christophe Royer \(^{1*}\) , Elizabeth Sandham \(^{1}\) , Elizabeth Slee \(^{2}\) , Jonathan Godwin \(^{1,3}\) , Nisha Veits \(^{1}\) , Holly Hathrell \(^{1}\) , Felix Zhou \(^{2}\) , Karolis Leonavicius \(^{1,4}\) , Jemma Garratt \(^{1,6}\) , Tanaya Narendra \(^{1,6}\) , Anna Vincent \(^{5}\) , Celine Jones \(^{6}\) , Tim Child \(^{5,6}\) , Kevin Coward \(^{6}\) , Chris Graham \(^{6}\) , Xin Lu \(^{2}\) and Shankar Srinivas \(^{1*}\) + +<|ref|>text<|/ref|><|det|>[[85, 316, 911, 375]]<|/det|> +\(^{1}\) Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford OX1 3QX, United Kingdom. \(^{2}\) Ludwig Institute for Cancer Research, Nuffield Department of Medicine, University of Oxford, Oxford OX3 7DQ, United Kingdom. + +<|ref|>text<|/ref|><|det|>[[85, 379, 880, 397]]<|/det|> +\(^{3}\) Department of Biochemistry, University of Oxford, South Parks Road, Oxford OX1 3QU, United Kingdom. + +<|ref|>text<|/ref|><|det|>[[85, 400, 692, 417]]<|/det|> +\(^{4}\) Current address: Institute of Biotechnology, Vilnius University, Vilnius, Lithuania. + +<|ref|>text<|/ref|><|det|>[[85, 421, 911, 459]]<|/det|> +\(^{5}\) Oxford Fertility, Institute of Reproductive Sciences, Oxford Business Park North, Oxford, OX4 2HW, United Kingdom. + +<|ref|>text<|/ref|><|det|>[[85, 462, 911, 501]]<|/det|> +\(^{6}\) Nuffield Department of Women's and Reproductive Health, University of Oxford, Level 3, Women's Centre, John Radcliffe Hospital, Headington, Oxford, OX3 9DU, United Kingdom. + +<|ref|>text<|/ref|><|det|>[[85, 525, 653, 543]]<|/det|> +\* e- mail: christophe.royer@dpag.ox.ac.uk; shankar.srinivas@dpag.ox.ac.uk + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 912, 345]]<|/det|> +During development, pseudostratified epithelia undergo large scale morphogenetic events associated with increased mechanical stress. The molecular mechanisms that maintain tissue integrity in this context are poorly understood. Using a variety of genetic and imaging approaches, we uncover that the ASPP2/PP1 complex ensures proper epiblast and proamniotic cavity architecture via a mechanism that specifically prevents the most apical daughter cells from delaminating apically following cell division events. The ASPP2/PP1 complex achieves this by maintaining the integrity and organisation of the F-actin cytoskeleton at the apical surface of dividing cells. ASPP2/PP1 is also essential during gastrulation in the primitive streak, in somites and in the head fold region, suggesting that this complex is required across a wide range of pseudostratified epithelia during morphogenetic events that are accompanied by intense tissue remodelling and high cell proliferation. Finally, our study also suggests that the interaction between ASPP2 and PP1 is essential to the tumour suppressor function of ASPP2 which may be particularly relevant in the context of tissues that are subject to increased mechanical stress. + +<|ref|>text<|/ref|><|det|>[[85, 371, 912, 632]]<|/det|> +Pseudostratified epithelia are common building blocks and organ precursors throughout embryonic development in a wide array of organisms'. As in other epithelia, their cells establish and maintain apicalbasal polarity. However, their high nuclear density, high proliferation rate and nuclei movement during interkinetic nuclear migration (IKNM) make them unique. As IKNM proceeds, mitotic cells round up at the apical surface of the epithelium before dividing. During this process, mitotic cells generate enough force to locally distort the shape of the epithelium or accelerate invagination?. During development, pseudostratified epithelia are also subject to large scale morphogenetic events that dramatically affect their shape and organisation. This is particularly true during gastrulation when cells apically constrict in the primitive streak as they push their cell body basally to eventually delaminate into the underlying mesoderm cell layer4- 7 or when the ectoderm is reshaped to form the head folds. The combined mechanical strains due to IKNM and morphogenetic events poses an incredible challenge for pseudostratified epithelia to maintain tissue integrity during development. However, the molecular mechanisms that allow them to cope with increased mechanical stress are poorly defined. + +<|ref|>text<|/ref|><|det|>[[85, 637, 912, 797]]<|/det|> +During these morphogenetic events, epithelial cells continually rely on apical constrictions involving specific F- actin cytoskeleton organisation and actomyosin contractility to modify tissue shape and organisation8,9. As cells apically constrict, the coupling of apical junctions to the actomyosin network is essential in transmitting forces across tissues10. Reciprocally, as apical constrictions reshape the apical domain of epithelial cells, apical junctions must be able to withstand the forces generated to maintain tissue integrity. Apical- basal polarity components, such as Par3, are vital for apical constrictions11 and the integrity of apical junctions12. However, it remains unknown how components of the apical- basal polarity machinery maintain tissue integrity in conditions of increased mechanical stress as morphogenetic events occur. + +<|ref|>text<|/ref|><|det|>[[85, 802, 912, 900]]<|/det|> +ASPP2 is a Par3 interactor and component of the apical junctions13,14. Here, using a variety of genetic and imaging approaches, we show that during morphogenetic events crucial for the normal development of the early post- implantation embryo, ASPP2 maintains epithelial integrity in pseudostratified epithelia under increased mechanical stress. ASPP2 is required for proamniotic cavity formation, the maintenance of primitive streak architecture, somite structure and head fold formation. In the proamniotic cavity, ASPP2 maintains + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 82, 912, 180]]<|/det|> +78 epithelial architecture by preventing apical daughter cells from escaping the epiblast. Mechanistically, we 79 show that this is achieved via the ability of ASPP2 to directly recruit protein phosphatase 1 and through its key 80 role in maintaining F- actin cytoskeleton organisation at the apical junctions. Our results show that ASPP2 is 81 an essential component of a system that maintains tissue integrity under conditions of increased mechanical 82 stress in a broad range of tissues. + +<|ref|>sub_title<|/ref|><|det|>[[86, 201, 152, 214]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[86, 222, 777, 238]]<|/det|> +## The phosphatase and polarity function of ASPP2 are not required for TE development. + +<|ref|>text<|/ref|><|det|>[[85, 243, 912, 401]]<|/det|> +ASPp2 can regulate both apical- basal cell polarity and the phosphorylation status of YAP/TAZ through its interaction with \(\mathrm{Par}3^{13,14}\) and \(\mathrm{PP1^{15,16}}\) respectively. Both cell polarity and the phosphorylation of YAP and TAZ are crucial to TE development \(^{17 - 21}\) . We therefore started with the hypothesis that ASPP2 may be important for outside cell polarisation and TE fate determination during preimplantation development. In support of this, we found that ASPP2 could start to be detected as early as E2.5 at cell- cell junctions (Extended Data Fig. 1a). As seen in other examples of polarised epithelia \(^{13,14,16,22}\) , ASPP2 was strongly localised to the apical junction in the trophectoderm from the 32- cell stage onwards (Fig. 1a). This localisation pattern was similar in human blastocysts, suggesting that ASPP2 behaves in a similar way across mammals (Extended Data Fig. 1b). + +<|ref|>text<|/ref|><|det|>[[82, 405, 912, 814]]<|/det|> +A Previous study revealed that ASPP2 may play a role during early embryogenesis, as ASPP2- mutant embryos in which exons 10- 17 were deleted could not be recovered at E6.5 \(^{23}\) . However, the phenotype of these embryos was not described, and earlier stages were not examined. To investigate the role of ASPP2 during preimplantation development we generated ASPP2- null embryos in which exon 4 was deleted (ASPp2ΔE4/ΔE4) resulting in a frameshift and early stop codons. To be able to distinguish between phenotypes relating to ASPP2's PP1 regulatory function and other functions, we also generated embryos homozygous for a mutant form of ASPP2 that was specifically unable to associate with PP1 (ASPp2RAKA/RAKA) \(^{24}\) . ASPp2ΔE4/ΔE4 and ASPp2RAKA/RAKA blastocysts appeared to be morphologically normal with properly formed blastocyst cavities (Fig. 1b and Extended Data Fig. 1c). YAP was clearly nuclear in the TE and cytoplasmic in the ICM suggesting that the ICM and TE lineage were properly allocated (Fig. 1b). The polarity protein Par3 (Fig. 1b) and F- actin cables (Extended Data Fig. 1c) were strongly localised at the apical junctions, suggesting that polarity and overall cell architecture were normal. It was sometimes possible to see some residual ASPP2 protein at the apical junction in early ASPp2ΔE4/ΔE4 blastocysts, potentially due to residual maternal ASPP2 expression. To eliminate the possibility that perdurance of maternally encoded ASPP2 compensated for the zygotic mutations, we microinjected 1- cell embryo with siRNA against ASPP2 and cultured them to the blastocyst stage. Control and ASPP2- depleted embryos were morphologically indistinguishable. The localisation of YAP was similar between control and ASPP2- depleted embryos (Fig. 1c). YAP phosphorylation at Serine 127 was stronger in the cytoplasm of ICM cells in comparison to the TE in both controls and ASPP2- depleted embryos (Extended Data Fig. 1d). Taken together, these results show that neither ASPP2's polarity function nor its PP1 regulatory function are required during preimplantation development. + +<|ref|>sub_title<|/ref|><|det|>[[85, 840, 648, 856]]<|/det|> +## ASPP2/PP1 interaction is required for proamniotic cavity architecture. + +<|ref|>text<|/ref|><|det|>[[85, 860, 910, 898]]<|/det|> +Since it has previously been shown that deletion of ASPP2 may be embryonic lethal around E6.5 \(^{23}\) , we next investigated whether ASPP2 was required at early post- implantation stages. To test this, we generated + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[88, 90, 904, 510]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[80, 529, 917, 796]]<|/det|> +
Fig. 1 | The ASPP2/PP1 complex is not required during preimplantation development. a, ASPP2 was detected by indirect immunofluorescence in E3.5 embryos to analyse its localisation pattern. A cross section through the equatorial plane of a representative embryo is shown (top row), as well as a 3D opacity rendering of the same embryo (bottom row). The F-actin cytoskeleton and nuclei were visualised using Phalloidin and DAPI, respectively. A magnified image of the dashed area is shown on the right. Note how ASPP2 colocalises with F-actin at the apical junctions in cells of the trophectoderm (white arrowheads). This is quantified in the juxtaposed graph showing ASPP2 and F-actin signal intensity along the apical-basal axis of cell-cell junctions. AJ: apical junction; B: base of the trophectoderm. Scale bars: 20 μm and 5 μm (for the magnification). b, The localisation pattern of YAP and Par3 was analysed in wild type and ASPP2RAKAembryos by indirect immunofluorescence. A cross section of representative embryos through the equatorial plane shows the localisation of YAP in the nuclei of the trophectoderm in both wild type and ASPP2RAKAembryos. Maximum intensity projections of these embryos show the localisation of Par3 at the level of apical junctions in the trophectoderm. Scale bar: 20 μm. c, ASPP2 knockdown in E3.5 embryos using siRNA against ASPP2 mRNA. ASPP2 knockdown was confirmed by indirect immunofluorescence. Note how signal at the apical junctions is specific to ASPP2 and how YAP is normally localised to the nuclei of TE cells in ASPP2-depleted embryos. Scale bar: 20 μm.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 83, 912, 426]]<|/det|> +ASPP2ΔE4/ΔE4 embryos at different stages and examined the localisation of Par6 and F- actin to assess apical- basal polarity and overall tissue organisation, respectively. At E5.5, ASPP2ΔE4/ΔE4 embryos did not exhibit obvious morphological defects in comparison to wild type and heterozygous litter mates. Polarised Par6 could be detected at the apical membrane in the visceral endoderm (VE) and the epiblast, suggesting that both cell layers properly polarised (Extended Data Fig. 2a). In contrast, E6.5 ASPP2ΔE4/ΔE4 embryos exhibited strong morphological defects in comparison to wild type and heterozygous litter mates. The proamniotic cavity either entirely failed to form (6 embryos out of 10 mutants) or was greatly reduced in size at E6.5 (4 embryos out of 10 mutants) (Fig. 2a) and always absent at E7.5 (Extended Data Fig. 2b). In embryos lacking a proamniotic cavity, the epiblast was disorganised, and instead of being a pseudostratified epithelium, appeared multilayered. This seemed to be the result of an ectopic accumulation of cells from the epiblast in place of the proamniotic cavity. The ectopic cells in the centre of the embryo exhibited a complete lack of polarised Par6 and in embryos with reduced cavity size, the accumulated cells showed reduced apical Par6 (Fig. 2b). This suggested that the ectopic accumulation of cells where the proamniotic cavity ought to be was accompanied by a progressive loss of cell polarity in the epiblast. F- actin localisation was also profoundly abnormal in these cells (Fig. 2b). In wild type embryos, F- actin was enriched at the apical junctions of epiblast cells, whereas in ASPP2ΔE4/ΔE4 embryos it was distributed more uniformly across the apical surface (Fig. 2c). This suggests that ASPP2 is required for organising the F- actin cytoskeleton at the apical junctions. + +<|ref|>text<|/ref|><|det|>[[85, 430, 912, 650]]<|/det|> +Because signals from the basement membrane are believed to be essential for proamniotic cavity formation25, we examined the localisation of laminin and found it unaltered in ASPP2ΔE4/ΔE4 embryos. This suggested that the loss of cell polarity and ectopic accumulation of cells was not accompanied by, or due to, breakage in the basement membrane (Fig. 2d,e). Finally, when we examined the basolateral membrane marker SCRIB, we observed that its basolateral localisation was unaffected in ASPP2ΔE4/ΔE4 embryos. However, SCRIB was also strongly expressed at the apical junctions in the epiblast of wild type embryos. This particular localisation pattern was intermittently disrupted in ASPP2ΔE4/ΔE4 embryos, specifically at the interface between cells of the epiblast and cells ectopically accumulating in the proamniotic cavity (Fig. 2f,g). Together, these results suggest that the apparent loss of apical cell polarity seen is ASPP2ΔE4/ΔE4 embryos originates from defects specific to the apical junctions rather than at the level of the basolateral or basement membranes. + +<|ref|>text<|/ref|><|det|>[[85, 655, 912, 797]]<|/det|> +The outside VE monolayer epithelium also normally expresses ASPP2 (Fig. 6a and Extended Data Fig. 5a,b) but was intact and exhibited apical Par6 in ASPP2ΔE4/ΔE4 mutants (Fig. 2a), suggesting that its epithelial architecture and integrity was maintained and that ASPP2 is required specifically in the epiblast at this stage. To verify that an epiblast specific requirement for ASPP2 led to the failure of proamniotic cavity formation in ASPP2ΔE4/ΔE4 embryos, we conditionally ablated ASPP2 expression in just the epiblast (ASPP2EpiΔE4/ΔE4 embryos) (Extended Data Fig. 2c). These embryos phenocopied ASPP2ΔE4/ΔE4 embryos, demonstrating that at this stage, ASPP2 is required only in the epiblast. + +<|ref|>text<|/ref|><|det|>[[85, 801, 912, 900]]<|/det|> +To test whether this requirement for ASPP2 is rooted in its ability to recruit and regulate PP1, we analysed ASPP2RAKA/RAKA embryos at E6.5. We found that ASPP2RAKA/RAKA embryos, similarly to ASPP2ΔE4/ΔE4 embryos, exhibit either reduced proamniotic cavity size or no cavity at all. This was again accompanied by a reduced apical Par6 in the epiblast when the proamniotic cavity was of reduced size and absence of apical Par6 when no cavity was present (Fig. 2h,i). The basolateral localisation of SCRIB once again was not affected, whereas + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[45, 44, 944, 775]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[78, 781, 916, 955]]<|/det|> +
Fig.2|TheASPP2/PP1 complexis requiredfortheformationofthe proamniotic cavity.a,Immunofluorescence of wild type and \(\mathrm{ASPP2}^{\Delta \mathrm{E4} / \Delta \mathrm{E4}}\) E6.5 embryos using an anti-Par6 antibody. The phenotypic variability of \(\mathrm{ASPP2}^{\Delta \mathrm{E4} / }\) \(\Delta \mathrm{E4}\) embryos is illustrated, with embryos either lacking cavities (middle row) or exhibiting smaller cavities (bottom row). The green dashed line highlights the ectopic accumulation of cells in the epiblast of \(\mathrm{ASPP2}^{\Delta \mathrm{E4} / \Delta \mathrm{E4}}\) embryos. b, Magnification of the corresponding regions shown in panel a. Blue arrowheads highlight the enrichment of F-actin at the apical junctions in the epiblast. Note how F-actin is not enriched at the apical junctions but is instead more homogenously distributed across the apical surface of epiblast cells in \(\mathrm{ASPP2}^{\Delta \mathrm{E4} / \Delta \mathrm{E4}}\) embryos (orange arrowhead). The insets within images are 2x magnifications of the corresponding dashed areas. c, Quantification of F-actin signal intensity along the apical surface of epiblast cells of wild type (n=3 embryos, 5 measurements per embryo) and \(\mathrm{ASPP2}^{\Delta \mathrm{E4} / \Delta \mathrm{E4}}\) embryos (n=3 embryos, 5 measurements per embryo). Measurements were made
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[81, 38, 916, 252]]<|/det|> +on cross sections along the apical domain of individual epiblast cells from apical junction to apical junction (represented with a blue background in the graph). See material and methods for details. d, Immunofluorescence of wild type and \(ASPP2^{AE4 / AE4}\) E5.5 embryos using an anti- Laminin antibody. e, Magnification of the corresponding dashed areas in panel d. f, Immunofluorescence of wild type and \(ASPP2^{AE4 / AE4}\) E6.5 embryos using an anti- SCRIB antibody. g, Magnification of the corresponding dashed areas in panel f. Green arrowheads highlight basolateral SCRIB. Note the enrichment of SCRIB at the apical junctions in the epiblast of wild type embryos (blue arrowhead) and its absence in the corresponding localisation in \(ASPP2^{AE4 / AE4}\) embryos (orange arrowhead). h, Immunofluorescence of wild type and \(ASPP2^{RAKA / RAKA}\) E6.5 embryos using an anti- Par6 antibody. The green dashed line highlights the ectopic accumulation of cells in the epiblast of \(ASPP2^{RAKA / RAKA}\) embryos. i, Magnification of the corresponding dashed regions in h. Note the reduced amount of Par6 along the apical domain of epiblast cells in \(ASPP2^{RAKA / RAKA}\) embryos (orange arrowhead). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: \(20 \mu m\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 911, 161]]<|/det|> +its localisation at the apical junctions was severely disrupted in ASPP2RAKA/RAKA embryos (Extended Data Fig. 2d,e). This shows that at E6.5, ASPP2RAKA/RAKA mutant embryos have an identical phenotype to ASPP2ΔE4/ΔE4 mutant embryos, demonstrating the key role of the ASPP2/PP1 interaction in regulating epiblast and proamniotic cavity architecture. + +<|ref|>sub_title<|/ref|><|det|>[[85, 191, 644, 207]]<|/det|> +## ASPP2 controls apical daughter cell reincorporation into the epiblast. + +<|ref|>text<|/ref|><|det|>[[85, 211, 912, 452]]<|/det|> +Our results so far show that ASPP2 is essential for the architecture of the epiblast and the formation of the proamniotic cavity. When ASPP2's function is impaired, apolar cells accumulate ectopically in place of the proamniotic cavity. However, it remained unclear how this occurred and what biological process ASPP2 actually controls in the epiblast. Amongst possible explanations is that the phenotype was the consequence of epiblast cells delaminating apically into the proamniotic cavity because of a drastic shift in the proportion of orthogonal cells divisions, a breakdown of the apical junction domain, a failure of daughter cells reincorporating basally following cell divisions or a combination of these. To answer this question, we generated ASPP2ΔE4/ΔE4 embryos with fluorescently labelled membranes, which enabled us to follow the movement of epiblast cells in these embryos by time- lapse confocal microscopy (Fig. 3a and Movie 1). In wild type and heterozygous embryos, we could observe the movement of cell bodies along the apical- basal axis during interkinetic nuclear migration (INM), with mitotic cells rounding up at the apical surface of the epiblast before dividing as previously described26. + +<|ref|>text<|/ref|><|det|>[[85, 456, 912, 596]]<|/det|> +We first analysed the orientation of cell divisions but could not detect differences in overall cell division angle in ASPP2ΔE4/ΔE4 embryos in comparison to controls (Extended Data Fig. 3a), even when division events were binned into categories as "orthogonal", "parallel" or "oblique". To determine if there was a defect in interkinetic nuclear migration, we also analysed the distance at which cell divisions occurred from the basement membrane of the epiblast. Again, there was no notable difference between ASPP2ΔE4/ΔE4 and wild type embryos, suggesting that even in the absence of apical- basal polarity and the proamniotic cavity, cells of the epiblast were able to proceed with INM (Extended Data Fig. 3b). + +<|ref|>text<|/ref|><|det|>[[85, 600, 912, 883]]<|/det|> +We next looked at the behaviour of daughter cells after cytokinesis. In wild type embryos, following cell divisions, the cell body of both daughters moved basally so that they came to span the entire height of the epithelium along the apical- basal extent of the epiblast. In contrast, in ASPP2ΔE4/ΔE4 embryos, dividing cells moved towards the embryonic centre as normal, but upon division, daughter cells delaminated apically towards the centre of mass of the embryonic region (Fig. 3a). This suggested that ASPP2 may be specifically required for the retention of daughter cells within the epiblast. To further characterise this failure of dividing cells to reintegrate into the epiblast epithelium, we quantified the movement of daughter cells from the initial point of cell division (Fig. 3b- d). We found that in wild type and heterozygous embryos, this movement was almost always basal for both daughters (51/56, 91.1%). In contrast, in mutant embryos, for half the daughter cells (33/66, 50%) the movement was apical (Fig. 3c, d). We found that in the majority of cases (29/33, 87.9%), it was the daughter that was relatively more apically positioned with respect to its sister that abnormally moved apically following cell divisions (Fig. 3d). This suggests that ASPP2 is involved in a novel mechanism specifically required for apical daughter cell reintegration into the pseudostratified epiblast following cell division, which is crucial in maintaining the architecture of the epiblast and proamniotic cavity. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 58, 950, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[80, 580, 917, 955]]<|/det|> +
Fig. 3 | ASPP2 is required for apical daughter cell reincorporation into the epiblast following cell division events. a, Time lapse imaging of wild type and \(ASP^{2\Delta E4 / \Delta E4}\) embryos. mT/mG-labelled cell membranes were used to manually track cell movement. Yellow dots highlight mother cells at the apical surface of the epiblast immediately prior to a cell division event. Green and magenta dots identify the resulting daughter cells. Note how both daughters reintegrate the epiblast in the wild type whereas one of the two daughters fails to do so in the absence of ASPP2 even after a prolonged period of time (t=82.5'). b, Diagram illustrating the method used to quantify daughter cell movement following cell divisions. Daughter cell movement was characterised by both the distance travelled (d) and the direction of travel (θ) expressed as the angle between the reference vector (the green vector starting from the initial position of the mother cell prior to the division event to the centre of the embryonic region) and the vector characterising absolute daughter cell movement (the red vector starting from the initial position of the mother cell prior to the division event to the final position of the daughter cell). The left panel illustrates the case of a daughter moving basally to reincorporate the epiblast and the right panel describes abnormal daughter cell movement towards the centre of the embryonic region such as seen in \(ASP^{2\Delta E4 / \Delta E4}\) embryos. c, Graph quantifying cell movement in wild type (n=3 embryos, 56 cells) and \(ASP^{2\Delta E4 / \Delta E4}\) embryos (n=3 embryos, 66 cells). For a given pair of daughter cells, each daughter was defined as "apical" or "basal" depending on their respective position relative to the centre of the embryonic region immediately after a cell division event. d, Proportion of daughter cells with an overall apical or basal movement in wild type and \(ASP^{2\Delta E4 / \Delta E4}\) embryos. Left panel: Quantification of the proportion of daughter cells with an overall apical (θ from 0° to 90°) or basal movement (θ from 90° to 180°) in wild type and \(ASP^{2\Delta E4 / \Delta E4}\) embryos. Right panel: quantification of the proportion of apical and basal daughters with an overall apical (θ from 0° to 90°) or basal movement (θ from 90° to 180°) in wild type and \(ASP^{2\Delta E4 / \Delta E4}\) embryos. \*\*\* p<0.0001, NS: non-significant (Fisher's exact test of independence).
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[85, 84, 795, 100]]<|/det|> +## ASPP2/PP1 complexes maintain epithelial integrity in regions of high mechanical stress. + +<|ref|>text<|/ref|><|det|>[[85, 105, 912, 346]]<|/det|> +Our results all point to an important role for ASPP2 in regulating tissue architecture, possibly via the regulation of F- actin organisation at the apical junction. However, this was difficult to study in ASPP2ΔE4/ΔE4 and ASPP2RAKA/RAKA embryos on a C57BL/6 background because of the relative severity of the defect. We therefore bred ASPP2RAKA/RAKA mutation into a BALB/c background, to take advantage of the fact that ASPP2 phenotypes are often not as dramatic in this background14. Consistent with this, BALB/c ASPP2RAKA/RAKA homozygous embryos completely bypassed the phenotype at E6.5 observed in C57BL/6 ASPP2RAKA/RAKA embryos. Instead, the phenotype of these embryos was milder, and they were only grossly different from wild type and heterozygous embryos one day later, at E7.5. ASPP2RAKA/RAKA embryos exhibited two distinct phenotypes. The majority (34/41, 82.9%), that we termed type I embryos, exhibited a strong accumulation of cells in their posterior, suggestive of a defect in the primitive streak (Fig. 4a and Extended Data Fig. 4a). A minority (7/41, 17.1%), that we termed type II embryos, were developmentally delayed but did not exhibit any structural defects. + +<|ref|>text<|/ref|><|det|>[[85, 350, 912, 632]]<|/det|> +Given that ASPP2ΔE4/ΔE4 and ASPP2RAKA/RAKA in a Bl/6 background showed striking abnormalities in the localisation of F- actin, we examined the localisation of F- actin in the posterior of type I ASPP2RAKA/RAKA embryos at E7.5 (Fig. 4b and Extended Data Fig. 4c). In wild type embryos, we were able to clearly identify cells apically constricting and pushing their cell body basally towards the nascent mesodermal cell layer. This is characteristic of cells in the primitive streak in the process of delaminating basally4. It was also evident that F- actin was enriched at the apical junctions in these cells (Fig. 4c). In contrast, ASPP2RAKA/RAKA embryos exhibited a clear ectopic accumulation of cells apical to the primitive streak as visualised by T expression (Fig. 4b). In these cells, F- actin was abnormally uniformly disturbed along the apical surface, with no clear apical- junction enrichment (Fig. 4c). To investigate this further, we performed Airyscan super- resolution imaging of these embryos (Fig. 4d- f). This revealed that the mesh- like structure normally formed by F- actin at the apical junctions in cells of the epiblast was severely disrupted in the posterior of ASPP2RAKA/RAKA embryos. Instead, F- actin appeared to form spike- like structures at the surface of cells in this region (Movie 2) (Fig. 4e, f). This profound disruption of F- actin localisation indicates that the PP1 regulatory function of ASPP2 is required for F- actin organisation in the cells of the primitive streak. + +<|ref|>text<|/ref|><|det|>[[85, 636, 912, 901]]<|/det|> +The epiblast specific requirement for ASPP2 despite its broad expression (Fig. 1a and Fig. 6a- c) and the localisation of the phenotype primarily to the posterior region of the epiblast in the BALB/c background suggest that specific epithelia or epithelial regions are more sensitive to ASPP2 deficiency than others. We hypothesised therefore that ASPP2 may be important particularly for dividing cells to reintegrate within epithelia subject to increased mechanical stress at the apical junction, for example during the apical curving required to form a cavity. In support of this hypothesis, phospho- myosin levels were higher in dividing cells in the epiblast, in particular at the apical junctions. This shows that actomyosin contractility increases at the apical junctions of dividing cells, which might result in increased mechanical stress (Fig. 5a). One prediction of this hypothesis is that increasing the mechanical stress in Type I ASPP2RAKA/RAKA embryos in a BALB/c background might induce an earlier or more severe phenotype reminiscent to that seen in the C57BL/6 background. To test this prediction, we cultured E6.5 BALB/c wild type and mutant embryos (a day before any phenotype is evident in mutants) within the confines of cylindrical cavities made of biocompatible hydrogels, in order to alter their shape27 and subject the epiblast epithelium to higher levels of mechanical stress (Fig. 5b). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[48, 42, 960, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[80, 584, 916, 958]]<|/det|> +
Fig. 4 | ASPP2 is required for epithelial integrity in the primitive streak. a, Posterior thickening in E7.5 \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos in a BALB/C background. Left panel: the anteroposterior axis was defined using AMOT localisation pattern. Right panel: comparison of tissue thickness in the anterior (3 measurements per embryo) and the posterior (3 measurements per embryo) of wild type (n=5 embryos) and \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos (n=5 embryos). \(^\ast \mathrm{p}< 0.05\) , \(^{***}\mathrm{p}< 0.0001\) (nested ANOVA). b, Cells accumulate in the primitive streak region of \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) \(\mathrm{RAA}\) embryos. Immunofluorescence of E7.5 wild type and \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos using a T (Brachyury) antibody. c, Cells ectopically accumulating in the primitive streak region are unable to apically constrict and do not have enriched F-actin at the apical junctions (orange arrowhead) in comparison to wild type (blue arrowheads and magenta dotted lines). Right panel: quantification of F-actin signal intensity along the apical surface of epiblast cells in the primitive streak region of wild type (n=3 embryos, 5 cells per embryo) and \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos (n=3 embryos, 5 cells per embryo). d-f, Airyscan imaging reveals the extent of F-actin disorganisation at the surface of cells accumulating ectopically in the primitive streak region of \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos. d, 3D opacity rendering of embryo optical halves, enabling visualisation of the apical surface of epiblast cells in the proamniotic cavity. Note the absence of the typical F-actin mesh pattern at the apical surface of cells in the posterior of \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos (green dotted line). e, Cross section through the primitive streak region, showing enriched F-actin at the apical junctions of wild type embryos (blue arrowheads) and the formation of F-actin spike-like structures at the contact-free surface of \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryos. f, En face view of the epiblast's apical surface in the posterior of an \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryo. Green dotted lines demarcate the disorganised apical region of the posterior and the more organised lateral regions of the epiblast. Right panel: magnification of the epiblast's apical surface in the posterior of an \(\mathrm{ASP}2^{\mathrm{RAA / RAA}}\) embryo showing F-actin forming spike-like structures. Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: \(20\mu \mathrm{m}\) .
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[37, 70, 928, 380]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[80, 389, 917, 532]]<|/det|> +
Fig. 5 | ASPP2RAKA/RAKA embryo are more susceptible to mechanical stress. a, Phospho-myosin levels are higher at the apical junctions of dividing cells in the epiblast (white arrowheads). b, wild type (n=4) and ASPP2RAKA/RAKA (n=2) embryos were grown for 30' in cylindrical cavities made of biocompatible hydrogels. The localisation pattern of GATA6 and Myosin was then analysed by immunofluorescence. c, Magnification of the embryos shown in b. The green dotted line highlights the ectopic accumulation of cells seen in ASPP2RAKA/RAKA embryos. Note how Myosin is enriched at the apical junctions of wild type epiblast cells (blue arrowheads). The orange arrowhead points to the abnormal distribution of Myosin at the apical surface of these cells. Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 912, 265]]<|/det|> +Wild type embryos elongated without showing any sign of disrupted tissue integrity (0 out of 4 embryos). Conversely, \(ASPP2^{RAKA/RAKA}\) embryos showed reduced cavity size with a clear accumulation of cells (2 out of 2 embryos), reminiscent of \(ASPP2^{\Delta E4 / \Delta E4}\) and \(ASPP2^{RAKA/RAKA}\) mutant embryos' phenotype in a C57BL/6 background (Fig. 5c). Interestingly, F-actin and Myosin were abnormally distributed at the apical surface of cells accumulating ectopically in \(ASPP2^{RAKA/RAKA}\) mutant embryos, suggesting that actomyosin contractility was disrupted. Together, This indicates that although \(ASPP2^{RAKA/RAKA}\) mutants in a BALB/c background can bypass the proamniotic cavity phenotype, increasing mechanical stress is sufficient to make them again susceptible to it and suggests that ASPP2 may be required in response to increased mechanical stress to maintain epithelial tissue integrity. + +<|ref|>sub_title<|/ref|><|det|>[[85, 291, 880, 327]]<|/det|> +## ASPP2 maintains the architecture of the F-actin cytoskeleton during cell division events via its PP1 regulatory function. + +<|ref|>text<|/ref|><|det|>[[85, 332, 912, 534]]<|/det|> +To understand how the absence of ASPP2 specifically regulates apical daughter cell reintegration into the epiblast, we analysed in detail its localisation pattern. ASPP2 was localised at the apical junctions in the VE (Fig. 6a and Extended Data Fig. 5a,b) and the epiblast (Fig. 6b and see Extended Data Fig. 2c for antibody specificity). In the former, ASPP2 was uniformly distributed along the apical junctions forming a regular mesh at the surface of the embryo (Fig. 6a and Extended Data Fig. 5a,b). In the epiblast however, ASPP2 appeared enriched at specific locations along the apical junctions (Fig. 6b). The high curvature of the inner apical surface of the epiblast makes it difficult to examine from standard confocal volumes. We therefore computationally 'unwrapped'28 the apical surfaces of the VE and epiblast so that we could more directly compare them (Fig. 6c). This revealed that, although ASPP2 was uniform in its distribution along all junctions in the VE, in the epiblast, it was enriched specifically at F-actin- rich tricellular junctions. + +<|ref|>text<|/ref|><|det|>[[85, 538, 912, 904]]<|/det|> +The enrichment of ASPP2 in regions of high F- actin and the disruption of F- actin localisation in mutants suggests that ASPP2 may somehow be linked to the F- actin cytoskeleton. Because ASPP2 does not possess any known F- actin binding domain, we looked if previously identified ASPP2 binding partners could provide this link. Interestingly, ASPP2 has been found to interact with Afadin in a number of proteomic studies29,30. Afadin is an F- actin- binding protein that has previously been shown to not only be enriched at tricellular junctions but to also regulate their architecture31. Moreover, at E7.5, Afadin- null embryos display a phenotype reminiscent of the phenotype observed in E7.5 \(ASP2^{\Delta exon4}\) embryos (Extended Data Fig. 2b) with cells accumulating in the proamniotic cavity32, suggesting that Afadin and ASPP2 have overlapping functions. To confirm that ASPP2 and Afadin can be found within the same protein complex, we immunoprecipitated endogenous Afadin in Caco- 2 cells, a colorectal cancer cell line with strong epithelial characteristics that retains the ability to polarise. ASPP2 co- immunoprecipitated with Afadin, indicating that they are indeed found in the same protein complex (Fig. 6d). To further investigate where this complex might form, we analysed the localisation of endogenous ASPP2 and Afadin in Caco- 2 and MDCK cells using super- resolution Airyscan microscopy. We found the proteins colocalised primarily at tricellular junctions, where F- actin was also enriched, including in dividing cells in metaphase (Fig. 6e and Extended Data Fig. 5c). Their expression pattern also partially overlapped at bicellular junctions (Extended Data Fig. 5c,d). Interestingly, Afadin was also found at the mitotic spindles (Fig. 6e) and cleavage furrow (Extended Data Fig. 5d), where ASPP2 was juxtaposed with Afadin. In E6.5 embryos, Afadin showed a similar localisation pattern to ASPP2, at the apical + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[45, 44, 936, 280]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[42, 272, 933, 830]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[75, 837, 916, 942]]<|/det|> +
Fig. 6 | the ASPP2/PP1 complex is required for F-actin organisation during cell division events. a, 3D opacity rendering showing the localisation of ASPP2 in E5.5 wild type embryos at the apical junctions of the visceral endoderm where it colocalises with F-actin. b, Cross section (top row) and 3D opacity rendering (bottom row) of the proamniotic cavity showing the localisation pattern of ASPP2 and F-actin at the apical junctions (white arrowhead). c, The outer surface of the VE and apical surface of the epiblast were computationally 'unwrapped', revealing the enrichment of ASPP2 and F-actin at tricellular junctions in the epiblast (green arrowheads). d, The
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[81, 60, 916, 275]]<|/det|> +interaction between endogenous ASPP2 and the F- actin- binding protein Afadin was examined in Caco- 2 cells by co- immunoprecipitation. e, The localisation pattern of endogenous ASPP2 and Afadin in Caco- 2 cells was examined by immunofluorescence. The bottom row represents the magnified region highlighted by a dotted box and shows the enrichment of ASPP2 and Afadin at tricellular junctions. ASPP2, Afadin and F- actin signal intensity was quantified across tricellular junctions (graph on the right). f, The localisation pattern of Afadin in the proamniotic cavity was analysed by immunofluorescence in E6.5 wild type embryos. Afadin colocalised strongly with F- actin at tricellular junctions (blue arrowhead). g, The localisation pattern of F- actin was analysed by timelapse microscopy in wild type and ASPP2RAKARAKA LifeAct- GFP positive embryos. Note how apical F- actin is disrupted in ASPP2RAKARAKA LifeAct- GFP positive embryos following a cell division event (orange arrowhead). h, At later time points, the ectopic accumulation of cells in the epiblast of ASPP2RAKARAKA LifeAct- GFP positive embryos was evident (dotted line). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin respectively. Scale bars: \(20 \mu \mathrm{m}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 83, 910, 120]]<|/det|> +junction of cells in the epiblast and VE (Fig. 6f and Extended Data Fig. 5e). In the epiblast, similarly to ASPP2, it was more abundant at F- actin- rich tricellular junctions. + +<|ref|>text<|/ref|><|det|>[[85, 124, 912, 326]]<|/det|> +Together, these results highlight the importance of the localisation pattern of ASPP2 in the epiblast, suggesting that it may be able to interact with F- actin at the apical junctions via its interaction with Afadin. They also suggest that this interaction might be important in dividing cells. To test this possibility, we generated ASPP2RAKA/RAKA embryos in a C57BL/6 background carrying a LifeAct- GFP transgene33 that allowed us to visualise F- actin in living embryos with time- lapse confocal microscopy (Fig. 6g,h and Extended Data Fig. 5f). During cell division events, following mitotic rounding, apical F- actin localisation was disrupted in ASPP2RAKA/RAKA embryos whereas it was maintained in wild type embryos (Fig. 6g). This was followed by an ectopic accumulation of cells at the apical surface of the epiblast, reducing the size of the proamniotic cavity (Fig. 6h). These results suggest that the PP1 regulatory function of ASPP2 is required to maintain the architecture of apical F- actin, particularly during cell division events in the epiblast. + +<|ref|>sub_title<|/ref|><|det|>[[86, 352, 707, 369]]<|/det|> +## ASPP2 supports tissue integrity across a variety of pseudostratified epithelia + +<|ref|>text<|/ref|><|det|>[[85, 373, 912, 555]]<|/det|> +Next, we investigated whether ASPP2 was only required in the epiblast or whether it also functioned in other tissues undergoing morphogenesis. To this end, we examined ASPP2ΔE4/ΔE4 embryos at later stages of development around the time of gastrulation. At E7.5, during late primitive streak stages, mesoderm formation (marked by expression of T) and migration were broadly comparable between ASPP2ΔE4/ΔE4 embryos and wild type littermates, despite the absence of a proamniotic cavity and the dramatic accumulation of cells now filling the entirety of the space inside the embryos (Fig. 7a, Movie 3). Furthermore, there was no difference in the velocity, directionality and distance travelled by mesoderm cells migrating from wild type and ASPP2ΔE4/ΔE4 mesoderm explants (Extended Data Fig. 6a and Movie 4). This suggested that ASPP2 was not required for mesoderm specification or migration. + +<|ref|>text<|/ref|><|det|>[[85, 558, 912, 905]]<|/det|> +To test whether further patterning of the mesoderm occurred in the absence of ASPP2, we examined ASPP2ΔE4/ΔE4 embryos at E8.5. Morphologically, these embryos were severely disrupted, shorter along the anterior- posterior axis and without head folds (Fig. 7b). However, using the cardiac progenitor marker NKX2.5, we found that this population of cells was able to migrate rostrally despite the dramatic morphological defects present in ASPP2ΔE4/ΔE4 embryos (Fig. 7b). At E9.5, some cells in the anterior of these embryos were also positive for alpha sarcomeric \(\alpha\) - actinin, suggesting that the NKX2- 5 positive cells could differentiate into cardiomyocytes (Fig. 7c and Extended Data Fig. 6b). Next, we analysed whether the mesoderm could go on to form structurally normal somites. Using FOXC2 as a marker of somatic mesoderm, we were able to identify distinct somite- like structures in ASPP2ΔE4/ΔE4 embryos (Fig. 7d and Extended Data Fig. 6b). However, we found that these somites were smaller than normal (Fig. 7e). Compared to wild- type control somites, a large proportion of mutant somites exhibited disrupted epithelial organisation (Fig. 7d,f) and a reduced proportion formed cavities (Fig. 7d,g). The relative size of the somitocoel in ASPP2ΔE4/ΔE4 embryos was also significantly reduced in comparison to wild type controls (Fig. 7h). Importantly, somites with disrupted epithelial organisation and lacking cavities had features reminiscent of ASPP2ΔE4/ΔE4 epiblasts: cells could be seen accumulating in the centre resulting in the obliteration of the somitocoel (Fig. 7d). These cells also displayed a lack of apical Par6, suggesting that as in the epiblast, in the forming somites, apical- basal polarity was disrupted (Fig. 7i). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[39, 35, 930, 860]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[81, 872, 916, 941]]<|/det|> +
Fig. 7 | The ASPP2/PP1 complex is required for tissue integrity across a variety of highly proliferative pseudostratified epithelia. a, The primitive streak expands comparatively in E7.5 wild type and ASPP2ΔE4/ΔE4 embryos. Mesoderm cells were labelled by immunofluorescence using an antibody against Brachyury (T). b, Patterning proceeds normally in the absence of ASPP2. The ectoderm and cardiac progenitors were labelled
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 78, 916, 382]]<|/det|> +in E8.5 wild type and \(ASPP2^{\Delta E4 / \Delta E4}\) embryos with antibodies against SOX2 and NKX2.5, respectively. ys: yolk sack, al: allantois, s: somites, hf: head fold, am: amnion, pc: proamniotic cavity. c, Cardiac progenitors can differentiate into cardiomyocytes in E9.5 \(ASPP2^{\Delta E4 / \Delta E4}\) embryos. The presence of the contractile machinery (magenta arrowheads) was assessed in wild type and \(ASPP2^{\Delta E4 / \Delta E4}\) embryos using an antibody against sarcomeric \(\alpha\) - actinin. d, Somite architecture is disrupted in \(ASPP2^{\Delta E4 / \Delta E4}\) embryos. The dotted line highlights the contour of a somite in an \(ASPP2^{\Delta E4 / \Delta E4}\) embryo. The star indicates the ectopic accumulation of cells in the centre of this somite. Arrowheads point to mitotic Fig.s. e- h, Quantification of somite characteristics in wild type (n=10 embryos, 58 somites) and \(ASPP2^{\Delta E4 / \Delta E4}\) (n=6 embryos, 35 somites) embryos at E8.5. \* p<0.05, \*\*\*\*p<0.0001 (Student's T-test). i, Apical- basal polarity is defective in the somites of \(ASPP2^{\Delta E4 / \Delta E4}\) embryos. Par6 localised apically in wild type somites (arrowhead) whereas it was absent in \(ASPP2^{\Delta E4 / \Delta E4}\) embryos (star). de: definitive endoderm. j, Head fold formation is defective in \(ASPP2^{\text{RAAK / RAAK}}\) embryos. The organisation of apical F- actin was disorganised locally in the anterior ectoderm of \(ASPP2^{\text{RAAK / RAAK}}\) embryos (orange dotted line). F- actin signal intensity along the apical surface of ectoderm cells in disrupted areas in \(ASPP2^{\text{RAAK / RAAK}}\) embryos (n=3 embryos, 5 cells per embryo) was compared to wild type cells (n=3 embryos, 5 cells per embryo). Measurements were made on cross sections along the apical domain of individual ectoderm cells from apical junction to apical junction (represented with a blue background in the graph). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm (d, i), 50 μm (a, c), 100 μm (b, j). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 83, 912, 244]]<|/det|> +These results suggest that ASPP2 is required not only in the epiblast, but more generally in pseudostratified epithelia14. To investigate this further, we tested the requirement for ASPP2 in lumen formation during cystogenesis. We derived embryonic stem cells (ESC) from embryos with exon 4 of ASPP2 flanked by two LoxP sites. To generate \(ASP^{2\Delta E4/\Delta E4}\) ESC, they were infected with a CRE-recombinase-expressing adenovirus. When grown in Matrigel, we found that the majority of \(ASP^{2\Delta E4/\Delta E4}\) ESC-derived cysts failed to form lumens in comparison to control cysts (Extended Data Fig. 6c). Similarly, ESC derived from \(ASP^{2RAA/RAKA}\) embryos failed to form lumens in comparison to wild type ESC, suggesting that the formation of lumens during cystogenesis requires ASPP2/PP1 interaction (Extended Data Fig. 6d). + +<|ref|>text<|/ref|><|det|>[[85, 248, 912, 530]]<|/det|> +Since our data suggests that ASPP2 is required in regions undergoing increased mechanical stress (Fig. 5), we wanted to examine further the potential importance of ASPP2 during head fold formation. We took advantage of the \(ASP^{2RAA/RAKA}\) embryos in a BALB/c background as their phenotype is milder and they develop a proamniotic cavity (Fig. 4a), reducing the likelihood that phenotypes observed in the head fold region are secondary defects due to overall tissue disorganisation. At E8.5, wild type embryos exhibited fully formed head folds (Fig. 7j). In contrast, \(ASP^{2RAA/RAKA}\) embryos failed to form head folds in the rostral region of the ectoderm. In this region, the epithelium buckled locally, without being able to fully complete head fold morphogenesis. Interestingly, this was accompanied by a loss of organisation of F-actin at the apical junction, similarly to what was observed in the proamniotic cavity of \(ASP^{2\Delta E4/\Delta E4}\) embryos and in the primitive streak of \(ASP^{2RAA/RAKA}\) embryos in a BALB/C background (Fig. 7j). These results strongly suggest that ASPP2 is required to maintain tissue integrity by regulating F-actin organisation at the apical junctions as tensions increase in the rostral region of the ectoderm during head fold formation. Together, these results also reinforce the idea that ASPP2, and its interaction with PP1, are required during morphogenetic events that result in increased tensions at the level of apical junctions in epithelial tissues. + +<|ref|>sub_title<|/ref|><|det|>[[88, 570, 180, 584]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[85, 590, 911, 648]]<|/det|> +Our study unveils a central role for ASPP2 in maintaining pseudostratified epithelial integrity under increased mechanical stress during major morphogenetic events: in the forming proamniotic cavity, in the primitive streak during gastrulation, in somites and in the head fold region. + +<|ref|>text<|/ref|><|det|>[[85, 653, 912, 874]]<|/det|> +Strikingly, ASPP2 controls proamniotic cavity formation via a mechanism that, following cell divisions, specifically prevents the most apical daughter cells from delaminating apically. ASPP2 achieves this by maintaining the integrity and organisation of the F- actin cytoskeleton at the apical surface of dividing cells. This mechanism is consistent with ASPP2 playing a role in maintaining epithelial integrity under increased mechanical stress considering that mitotic rounding results in forces sufficient to contract the epithelium in the apical- basal axis and mechanically contribute to expansion of the lumen2. However, it remains unclear why it is only the more apically localised daughter cells that are affected in ASPP2 mutant embryos. One possibility is that these daughters do not inherit the basal process that tethers cells to the basement membrane in pseudostratified epithelia. However, the idea that the basal process is inherited asymmetrically is controversial34. Other mechanisms, such as the extent of cell- cell interactions between neighbouring cells and apical daughter cells in comparison to the more basal daughters may be involved. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 83, 912, 201]]<|/det|> +It has been suggested that the purpose of interkinetic nuclear migration is to ensure that cells divide apically to safeguard the integrity of pseudostratified epithelia35. Here we show that this is not sufficient to maintain tissue integrity as, in the absence of ASPP2, the apical-basal movement of nuclei and the position of cell divisions during IKNM proceeds unhindered. Our observations also suggest that the organisation of the F-actin cytoskeleton at the apical junctions is not required for nuclear movement during INKM and that intact basolateral domain and attachment to the basement membrane are sufficient instead. + +<|ref|>text<|/ref|><|det|>[[85, 206, 912, 428]]<|/det|> +Given that the ASPP2 interactors Par36,37 and Afadin38- 40, tricellular junctions41, as well as tissue tension42- 45 have all been shown to determine cell division orientation to some extent, it was important to explore whether ASPP2 could play a role in this process. However, our results indicate that ASPP2 does not control the orientation of cell divisions in the epiblast. Similarly to previous work26, we also find no bias towards cell divisions orientated in the plane of the epithelium, suggesting that, at these early stages of development, planar cell polarity may not play a role in directing cell division orientation. We therefore cannot rule out that, in a different context, ASPP2 might control cell division orientation in conjunction with Afadin. In fact, later in development in E8.5 ASPP2RAKA/RAKA embryos, cells sometimes delaminated basally in the anterior regions of the ectoderm, reminiscent of the phenotype observed in SCRIB- and DLG- depleted Drosophila wing discs, where cell divisions are normally orientated in the plane of the epithelium by cell- cell junctions to maintain epithelial integrity46. + +<|ref|>text<|/ref|><|det|>[[85, 432, 912, 653]]<|/det|> +Our results highlight the previously underappreciated localisation of ASPP2 at tricellular junctions in epithelial cells, in particular of the epiblast, and reveal the unknown biological function of ASPP2/PP1 complexes in the regulation of F- actin organisation at the apical junction. Considering that Afadin regulates the architecture of tricellular junctions in response to tensions31,47, the interaction between Afadin and ASPP2 strongly suggests that ASPP2 may exert its F- actin function at tricellular junctions via Afadin. The role of Afadin in regulating the linkage between F- actin and junctions during apical constrictions48 suggests that ASPP2 may also be important in this process, which may be particularly relevant in the primitive streak. Tricellular junction are emerging as a particularly important aspect of tissue homeostasis, at the intersection between actomyosin contractility and apical- basal organisation in the context of tissue tensions49. We therefore suggest that ASPP2 may be directly involved in the response to tissue tension by interacting with Afadin at the level of tricellular junctions to maintain F- actin organisation. + +<|ref|>text<|/ref|><|det|>[[85, 658, 912, 901]]<|/det|> +Our study also suggests that the interaction between ASPP2 and PP1 might be essential to the well documented tumour suppressor function of ASPP250- 52. Simply abrogating the ability of ASPP2 to recruit PP1 is enough to induce the formation of abnormal discrete clusters of cells in the epiblast reminiscent of tumours. This suggests that mutations in ASPP2 that interfere with its interaction with PP1 might, in conjunction with mechanical stress, lead to tumour development. These mutations could be in the canonical PP1- binding domain of ASPP2, but also in other key domains which have been shown to contribute to the interaction53. Recent findings support the idea that ASPP2 mutations could lead to tumorigenesis in the presence of mechanical stress. Using insertional mutagenesis in mice with mammary- specific inactivation of Cdh1, ASPP2 was identified as part of a mutually exclusive group containing three other potential tumour suppressor genes (Myh9, Ppp1r12a and Ppp1r12b), suggesting that these genes target the same process54. With our finding that ASPP2 controls the organisation of the F- actin cytoskeleton, it now becomes apparent that, in addition to three of these genes being PP1- regulatory subunits, all four are in fact F- actin regulators. Biological studies to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 83, 911, 141]]<|/det|> +test specific mutations found in ASPP2 in cancer and elucidating the substrates and specific phosphor residues targeted by the ASPP2/PP1 complex will therefore provide new insights into the tumour suppressor role of ASPP2 and might help developing new approaches to cancer treatment. + +<|ref|>sub_title<|/ref|><|det|>[[88, 180, 271, 195]]<|/det|> +## Materials and Methods + +<|ref|>text<|/ref|><|det|>[[85, 200, 912, 256]]<|/det|> +Mouse strains and embryo generation. All animal experiments complied with the UK Animals (Scientific Procedures) Act 1986, were approved by the local Biological Services Ethical Review Process, and were performed under UK Home Office project licenses PPL 30/3420 and PCB8EF1B4. + +<|ref|>text<|/ref|><|det|>[[85, 260, 912, 339]]<|/det|> +All mice were maintained on a 12- hour light, 12- hour dark cycle. Noon on the day of finding a vaginal plug was designated 0.5 dpc. For preimplantation stages, embryos were flushed using M2 medium (Sigma M7167) at the indicated stages. For post- implantation stages, embryos of the appropriate stage were dissected in M2 medium with fine forceps and tungsten needles. + +<|ref|>text<|/ref|><|det|>[[85, 344, 912, 655]]<|/det|> +We originally obtained ASPP2 mutant mice in which exons 10- 17 were replaced with a neo- r gene23 from Jackson Laboratory. After careful characterisation of this mouse line, we found that the Neo cassette was not inserted in the ASPP2 locus. As a consequence, we used a different strategy to generate ASPP2 mutant mice. C57BL/6N- Trp53bp2text<|/ref|><|det|>[[85, 658, 911, 714]]<|/det|> +ASPP2WT/ΔE4 mice were used to generate ASPP2ΔE4/ΔE4 embryos. To produce epiblast- specific ASPP2- null embryos (ASPP2EpiΔE4/ΔE4 embryos), ASPP2WT/ΔE4 mice homozygous for the Sox2Cre transgene were crossed with ASPP2IE4/IE4 mice. To generate ASPP2ΔE4/ΔE4 embryos with fluorescently labelled membranes, we established ASPP2WT/ΔE4 mice homozygous for the mT/mG transgene56 and crossed them with ASPP2WT/ΔE4 mice. + +<|ref|>text<|/ref|><|det|>[[85, 717, 912, 896]]<|/det|> +The ASPP2WT/RAKA mice were made by inGenious Targeting Labs (Ronkonkoma, NY). A BAC clone containing exon 14 of the trp53bp2 gene was subcloned into a \(\sim 2.4\mathrm{kb}\) backbone vector (pSP72, Promega) containing an ampicillin selection cassette for retransformation of the construct prior to electroporation. A pGK- gb2 FRT Neo cassette was inserted into the gene. In the targeting vector, the wild type GTG AAA TTC was mutated to GCG AAA GCC by overlap extension PCR and introduced into C57BL/6 x 129/SvEv ES cells by electroporation. Inclusion of the mutations in positive ES cell clones was confirmed by PCR, sequencing and Southern blotting. ES cells were microinjected into C57BL/6 blastocysts and resulting chimeras mated with C57BL/6 FLP mice to remove the Neo cassette. The presence of the mutation was confirmed by sequencing. Mice were then back- crossed with BALB/cOlaHsd or C57BL/6J mice for at least eight generations to obtain the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 911, 142]]<|/det|> +RAKA mutation in the respective pure background. \(ASP^{2R A K A / R A K A}\) embryos were generated from heterozygous crosses. To generate LifeAct- GFP- positive \(ASP^{2R A K A / R A K A}\) embryos, we generated \(ASP^{2W T / R A K A}\) mice heterozygous for the LifeAct- GFP transgene33. + +<|ref|>text<|/ref|><|det|>[[85, 165, 912, 408]]<|/det|> +siRNA microinjections. siGENOME RISC- Free Control siRNA (Dharmacon) and Silencer Select Pre- designed siRNAs against mouse ASPP2 (Ambion) were resuspended in nuclease- free sterile water and used at \(20 \mu \mathrm{M}\) . For zygotes, 3 to 4 week old CD- 1 females (Charles River UK) were injected intraperitoneally with 5 IU of PMSG (Intervet) and 48 h later with 5 IU of hCG (Intervet), and were paired with C57Bl/6J male mice (in house). Zygotes were retrieved from oviductal ampullae at 20 hours post- hCG. Cumulus- enclosed zygotes were denuded by exposure to \(1 \mathrm{mg / mL}\) hyaluronidase (Sigma) in modified mHTF (Life Global) containing 3 mg/ml BSA for 3- 6 min and cultured in LGGG- 020 (life Global) containing \(3 \mathrm{mg / ml}\) BSA in the presence of \(5\%\) CO2 at \(37^{\circ} \mathrm{C}\) . Microinjection of zygotes commenced 2 hours after release from cumulus mass. Zygotes with a normal morphology were microinjected into the cytoplasm in \(30 \mu \mathrm{l}\) drops of modified HTF media containing 4 mg/ml BSA using a PMM- 150FU Piezo impact drive (Primetech) using homemade glass capillaries with \(\sim 5-\) 10 pl of siRNA. Zygotes were returned to LGGG- 020 containing \(3 \mathrm{mg / ml}\) BSA in the presence of \(5\%\) CO2 at \(37^{\circ} \mathrm{C}\) until analysis. + +<|ref|>text<|/ref|><|det|>[[85, 432, 911, 532]]<|/det|> +Human embryo collection. Human embryos were donated from patients attending the Oxford Fertility with approval from the Human Fertilization and Embryology Authority (centre 0035, project RO198) and the Oxfordshire Research Ethics Committee (Reference number 14/SC/0011). Informed consent was attained from all patients. Embryos were fixed in \(4\%\) paraformaldehyde, washed twice and kept in PBS containing \(2\%\) bovine serum albumin (PBS- BSA) at \(4^{\circ} \mathrm{C}\) until they were used for immunohistochemistry. + +<|ref|>text<|/ref|><|det|>[[85, 557, 912, 819]]<|/det|> +Wholemount immunohistochemistry. Post- implantation embryos were fixed in \(4\%\) paraformaldehyde in phosphate- buffered saline (PBS) at room temperature for 20 to 45 minutes depending on embryo stages. Embryos were washed twice for 10 minutes in \(0.1\%\) PBS- Tween (PBS containing \(0.1\%\) Tween 20). Embryos were then permeabilized with \(0.25\%\) PBS- Triton (PBS containing 0.25 Triton X- 100) for 25 minutes to 1 hour depending on embryo stages and then washed twice for 10 minutes in \(0.1\%\) PBS- Tween. Embryos were incubated overnight in a blocking solution ( \(3\%\) Bovine serum albumin, \(2.5\%\) donkey serum in \(0.1\%\) PBS- Tween). The next day, primary antibodies were diluted in blocking solution and added to the embryos overnight. The following day, embryos were washed three times for 15 minutes in \(0.1\%\) PBS- Tween and then incubated with secondary antibodies and Phalloidin diluted in blocking solution overnight. Finally, embryos were washed four times in \(0.1\%\) PBS- Tween and kept in DAPI- containing VECTASHIELD Antifade Mounting Medium (Vector Laboratories) at \(4^{\circ} \mathrm{C}\) until used for imaging. Short incubation steps were carried out in wells of a 12- well plate on a rocker at room temperature and overnight steps were carried out in \(1.5 \mathrm{ml}\) Eppendorf tubes at \(4^{\circ} \mathrm{C}\) . + +<|ref|>text<|/ref|><|det|>[[85, 824, 911, 903]]<|/det|> +For preimplantation embryos, fixation and permeabilization times were reduced to 15 minutes and \(2\%\) PBS- BSA (PBS containing \(2\%\) Bovine serum albumin) was used for washing steps. Blocking and secondary antibody incubation steps were reduced to one hour. Embryos were transferred between solutions by mouth- pipetting. The embryos were mounted in 8- well chambers in droplets consisting of \(0.5 \mu \mathrm{l}\) DAPI- containing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 83, 910, 120]]<|/det|> +VECTASHIELD and \(0.5 \mu \mathrm{l} 2\%\) PBS- BSA. After mounting the embryos were kept in the dark at \(4^{\circ} \mathrm{C}\) until they were imaged. + +<|ref|>text<|/ref|><|det|>[[85, 144, 912, 388]]<|/det|> +Immunocytochemistry. Caco- 2 and MDCK cells were maintained in Dulbecco's modified Eagle's medium containing \(10\%\) fetal bovine serum, penicillin, and streptomycin at \(37^{\circ} \mathrm{C}\) in a \(5\%\) CO2 atmosphere incubator. In preparation for immunocytochemistry, Caco- 2 cells were seeded onto coverslips in 24- well plates with fresh medium. 48 hours later, cells were fixed with \(4\%\) paraformaldehyde (in PBS) for 10 minutes, washed twice in PBS, and then permeabilized with \(0.1\%\) Triton X- 100 in PBS for 4 minutes. Cells were washed twice in PBS and \(2\%\) PBS- BSA was then used as a blocking solution for 30 minutes prior to incubation with primary antibodies. Primary antibodies were diluted in \(2\%\) PBS- BSA and applied to cells for 40 minutes. Cells were then washed three times with PBS. Secondary antibodies (1:400), DAPI (1:2000, Invitrogen) and Phalloidin (1:400) were diluted in \(2\%\) PBS- BSA and applied to cells for 20 minutes. Coverslips were then washed three times with PBS and mounted onto glass slides with a small drop of Fluoromount- G (SouthernBiotech). They were air- dried before being sealed with nail varnish. All incubation steps were carried out at room temperature on a rocker. Samples were kept in the dark at \(4^{\circ} \mathrm{C}\) until they were imaged. + +<|ref|>text<|/ref|><|det|>[[85, 412, 912, 737]]<|/det|> +Antibodies and phalloidin conjugates. The following antibodies were used at the stated dilutions: rabbit anti- ASPP2 (Sigma, HPA021603), 1:100- 1:200 (IHC); mouse anti- ASPP2 (Santa Cruz Biotechnologies, sc135818), 1:100 (ICC), 1:1000 (IB); mouse anti- YAP (Santa Cruz Biotechnology, sc- 101199), 1:100 (IHC); rabbit anti- pYAP S127 (Cell Signaling, 4911), 1:100 (IHC); rabbit anti- Par3 (Millipore, 07- 330), 1:100 (IHC); rabbit anti- Pard6b (Santa Cruz Biotechnology, sc- 67393), 1:100 (IHC); rabbit anti- SCRIB (Santa Cruz Biotechnology, sc28737), 1:100 (IHC); goat anti- Brachyury (Santa Cruz Biotechnology, sc17745), 1:100 (IHC); rabbit anti- Sarcomeric \(\alpha\) - actinin (Abcam, ab68167), 1:100 (IHC); mouse anti- FOXC2 (Santa Cruz Biotechnology, sc515234), 1:100 (IHC); rabbit anti- SOX- 2 (Millipore, AB5603), \(2 \mu \mathrm{l}\) per mg of cell lysate (co- IP), 1:100 (IHC); goat anti- NKX2.5 (Santa Cruz Biotechnology, sc8697), 1:100 (IHC); rabbit anti- Afadin (Sigma, A0224), \(2 \mu \mathrm{l}\) per mg of cell lysate (co- IP), 1:100 (IHC, ICC), 1:1000 (IB); rabbit anti- Laminin (Sigma, L9393), 1:200 (IHC); goat anti- AMOT (Santa Cruz Biotechnologies, sc82491), 1:200 (IHC); goat anti- GATA- 6 (R&D Systems, AF1700), 1:100 (IHC); rabbit anti- Myosin IIa (Cell Signaling, #3403), 1:100; rabbit anti- phospho- Myosin light chain 2 (Cell Signaling, #3674), 1:100. The following were used at 1:100 for IHC and 1:400 for ICC: Alexa fluor 555 donkey- anti- mouse (Invitrogen, A- 31570), Alexa fluor 647 goat- anti- rat (Invitrogen, A- 21247), Alexa fluor 488 donkey- anti- rabbit (Invitrogen, A21206), Phalloidin- Atto 488 (Sigma, 49409), Phalloidin- Atto 647N (Sigma, 65906). + +<|ref|>text<|/ref|><|det|>[[85, 760, 912, 902]]<|/det|> +Confocal microscopy, image analysis and quantification. Samples were imaged on a Zeiss Airyscan LSM 880 confocal microscope with a C- Apochromat 40x/1.2 W Korr M27 water immersion objective or a Plan- Apochromat 63x/1.4 OIL DIC M27 objective. For super- resolution imaging, an Airyscan detector was used. Velocity (version 6.3.1, PerkinElmer) and Zen (Zeiss) software were used to produce maximum intensity projections and 3D opacity renderings. Image analysis was performed on optical sections. For signal intensity profiles along the apical- basal axis and across tricellular junctions, the arrow tool in the Zen software was used. Anterior and posterior embryo widths measurements were made using the line tool in Velocity. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 912, 222]]<|/det|> +For F- actin signal intensity profiles across the apical surface of epiblast or ectoderm cells, Fiji's freehand line tool with a width of "3" was used. Because the size of the apical domain was different for each cell measured, distances were expressed as percentages, with 100% representing the total distance across the apical domain. To account for depth- dependent signal attenuation, F- actin signal intensity at the apical domain was normalized by mean F- actin intensity in the nucleus of the cell measured. In each experiment, for each genotype, three embryos were used for measurements and 5 cells were analysed per embryo. The LOWES method was used to fit a line to the data. + +<|ref|>text<|/ref|><|det|>[[85, 248, 912, 550]]<|/det|> +Mouse embryo culture for live imaging and image analysis. To restrain embryo movement during imaging, lanes were constructed inside 8- well Lab- Tek II chamber slide (Nunc), using glass rods made from hand- drawn glass capillaries. Shorter pieces were used as spaces between two rods to create a space slightly wider than an embryo. Silicone grease was used to maintain the rods together. Each well was filled with medium containing 50% phenol red- free CMRL (PAN- Biotech, Germany) supplemented with 10 mM L/glutamine (Sigma- Aldrich) and 50% Knockout Serum Replacement (Life Technologies, England). The chamber was equilibrated at 37°C and an atmosphere of 5% CO₂ for at least 2 h prior to use. Freshly dissected embryos were placed in the lanes between two rods and allowed to settle prior to imaging on a Zeiss LSM 880 confocal microscope equipped with an environmental chamber to maintain conditions of 37°C and 5% CO₂. Embryos were imaged with a C- Apochromat 40x/1.2 W Korr M27 water immersion objective. Using a laser excitation wavelength of 561 nm, embryos labelled with mT/mG were imaged every 7.5 minutes and for each time point, nine z- sections were acquired every 3 μm around the midsagittal plane for up to 10 hours. For LifeAct- positive embryos, a laser excitation wavelength of 488 nm was used, and embryos were imaged every 15 minutes for 6 hours. For each time point, 12 z- sections every 1.5 μm were collected around the midsagittal plane. + +<|ref|>text<|/ref|><|det|>[[85, 555, 912, 839]]<|/det|> +Daughter cell movement was quantified using the Fiji plugin TrackMate (v5.2.0). Timepoints were registered using Fiji. Jittering was accounted for by correcting cell coordinates relatively to the centre of the embryonic region. The distance travelled by daughter cells (d) was analysed by calculating the distance between the coordinates of their final position and the coordinates of their respective mother cell immediately prior cell division. The direction of daughter cell movement (θ) was analysed by calculating the angle between the vector describing cell movement (that is the vector originating from the coordinates of the mother cell immediately prior cell division to the coordinates of the daughter cell at its final position) and the vector from the coordinates of the mother cell prior cell division to the coordinates of the embryonic region's centre. To establish the angle of cell division, we first defined a vector starting at the coordinates of one daughter and ending at the coordinates of the other immediately after cell division. We then defined a second vector originating halfway between the two daughters and terminating at the centre of the embryonic region. The angle of cell division was defined as the angle between those two vectors. The relative position of cell divisions was defined as the distance between the position of the mother cell immediately prior to cell division and the base of the epiblast. + +<|ref|>text<|/ref|><|det|>[[85, 864, 910, 901]]<|/det|> +Embryo culture in channels. Channels were formed by casting a 5% (which corresponded to approximately 4.2kPa stiffness58) acrylamide hydrogel (containing 39:1 bisacrylamide) around 60 μm wires within the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 912, 265]]<|/det|> +confinement of a two- part mould (10x10x1 mm). Ammonium persulphate \((0.1\%)\) and TEMED \((1\%)\) were added to polymerize polyacrylamide. The wires were then removed to form cylindrical cavities within hydrogel pieces. The hydrogels were carefully washed and equilibrated in embryo culture media at \(37^{\circ}C\) and \(5\% \text{CO2}\) . The embryos were then inserted into the channels using a glass capillary with a diameter slightly larger than the embryo itself. It was used to stretch the hydrogel channel before injecting the embryos and letting the channels relax and deform the embryos. Cell viability in channels had previously been assessed without any noticeable difference with control embryos27. After 30 minutes, embryos were fixed inside the hydrogels with \(4\%\) PFA for 35 minutes. Once fixed, embryos were removed from the hydrogel channels and wholemount immunohistochemistry was performed. + +<|ref|>text<|/ref|><|det|>[[85, 288, 912, 512]]<|/det|> +Co- immunoprecipitation and SDS- PAGE/Immunoblotting. For immunoprecipitation experiments, Caco- 2 cells from confluent 10 cm diameter dishes were washed twice with PBS and then lysed in \(500 \mu \mathrm{l}\) of a buffer containing \(50 \mathrm{mM}\) Tris- HCl at pH 8, \(150 \mathrm{mM}\) NaCl, \(1 \mathrm{mM}\) EDTA, Complete Protease Inhibitor Cocktail (Roche) and \(1\%\) Triton X- 100. Lysates were left on ice for 30 minutes, briefly sonicated and spun down at \(21,000 \times \mathrm{g}\) for 30 minutes at \(4^{\circ}\mathrm{C}\) . The supernatant was transferred to another tube and protein concentration was measured (Bradford, Bio- Rad). \(1 \mathrm{mg}\) of protein lysate was used per condition. Lysates were precleared using \(20 \mu \mathrm{l}\) protein G Sepharose 4 fast flow (1:1 in PBS, GE Healthcare) for 30 minutes at \(4^{\circ}\mathrm{C}\) on a shaker. The supernatant was incubated for 30 minutes at \(4^{\circ}\mathrm{C}\) on a shaker with \(2 \mu \mathrm{l}\) of the indicated antibody. \(30 \mu \mathrm{l}\) protein G Sepharose 4 Fast Flow (1:1 in PBS) was added to each condition and samples were incubated overnight at \(4^{\circ}\mathrm{C}\) on a shaker. Samples were washed 5 times with ice cold lysis buffer. \(25 \mu \mathrm{l}\) sample buffer was added and samples were incubated at \(95^{\circ}\mathrm{C}\) for 5 minutes before being subjected to SDS- PAGE/Immunoblotting. + +<|ref|>text<|/ref|><|det|>[[85, 533, 912, 715]]<|/det|> +Mesoderm explants and mesoderm cell migration. \(\text{ASPP}^{2\text{WTRAKA}}\) mice heterozygous for the LifeAct- GFP transgene were crossed and E7.5 embryos were dissected in M2. Embryos were then incubated in a \(2.5\%\) pancreatic mixture on ice for 20 minutes. Using tungsten needles, the visceral endoderm layer was removed and then the mesodermal wings were separated from the underlying epiblast. Mesodermal tissue was grown in fibronectin- coated 8- well Lab- Tek II chamber slides and cultured in DMEM containing \(10\%\) fetal bovine serum, penicillin, and streptomycin at \(37^{\circ}\mathrm{C}\) and \(5\% \mathrm{CO}_2^{59}\) . Samples were imaged on a Zeiss LSM 880 confocal microscope equipped with an environmental chamber to maintain conditions of \(37^{\circ}\mathrm{C}\) and \(5\% \mathrm{CO}_2\) . A laser excitation wavelength of \(488 \mathrm{nm}\) was used, and explants were imaged every 5 minutes for 5 hours. For each time point, 9 z- section with \(1 \mu \mathrm{m}\) step were collected. + +<|ref|>text<|/ref|><|det|>[[85, 720, 912, 839]]<|/det|> +Individual cells migrating away from the explants were tracked using the manual tracking plugin in Fiji. The movement, velocity and directionality of individual cells was analysed. Movement represented the total distance travelled in \(\mu \mathrm{m}\) by an individual cell. Velocity represented the average speed in \(\mu \mathrm{m / min}\) of a given cell. Directionality was used as a measure of how direct or convoluted a cell's path was and was calculated as the ratio between the total distance travelled and the distance in a straight line between a cell's start and end position60. + +<|ref|>text<|/ref|><|det|>[[85, 864, 910, 901]]<|/det|> +Embryonic stem cell- derived cysts. Using small- molecule inhibitors of Erk and Gsk3 signalling61, \(\text{ASPP}^{2\text{IE4/IE4}}\) and \(\text{ASPP}^{2\text{RAKA/RAKA}}\) (and \(\text{ASPP}^{2\text{WT/WT}}\) controls) ESC were generated from flushed E2.5 embryos + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 82, 911, 180]]<|/det|> +obtained from crosses between \(ASPP2^{\text{IE4/IIE4}}\) and \(ASPP2^{\text{WT/RAKA}}\) mice, respectively. Briefly, embryos were grown for two days in organ culture dishes, containing pre- equilibrated preimplantation embryo culture media supplemented with 1 \(\mu \text{M}\) PDO325901 and 3 \(\mu \text{M}\) CHIR99021 (Sigma- Aldrich). Embryos were grown one more day in NDiff 227 media (Takara) supplemented with 1 \(\mu \text{M}\) PDO325901 and 3 \(\mu \text{M}\) CHIR99021 (NDiff + 2i). The trophectoderm was removed by immunosurgery and "epiblasts" were grown in gelatinised dishes in the presence of NDiff + 2i and ESGRO (recombinant mouse LIF Protein, Millipore) to establish ESC lines. + +<|ref|>text<|/ref|><|det|>[[85, 184, 911, 201]]<|/det|> +presence of NDiff + 2i and ESGRO (recombinant mouse LIF Protein, Millipore) to establish ESC lines. + +<|ref|>text<|/ref|><|det|>[[85, 205, 911, 365]]<|/det|> +\(ASPP2^{\text{IE4/IIE4}}\) ESC were infected with an Ad- CMV- iCre adenovirus (Vector Biolabs) to delete exon 4 of \(ASPP2\) . Deletion of exon 4 was assessed by PCR. Non- infected \(ASPP2^{\text{IE4/IIE4}}\) ESC were used as controls. Wild type ESC derived from litter mates were used as controls for \(ASPP2^{\text{RAKA/RAKA}}\) ESC. To form cysts, 4500 ESC were resuspended in 150 \(\mu \text{l}\) Matrigel (354230, Corning) and plated into a well of an 8- well Lab- Tek II chamber slide. The gel was left to set for 10 minutes at \(37^{\circ}\text{C}\) before 300 \(\mu \text{l}\) differentiation medium (DMEM supplemented with 15% FCS, 1% Penicillin/Streptomycin, 1% Glutamine, 1% MEM non- essential amino acids, 0.1 mM 2- mercaptoethanol and 1mM sodium pyruvate) was added. ESC were grown for 72 hours at \(37^{\circ}\text{C}\) and 5% \(\text{CO}_2\) before immunostaining was performed. + +<|ref|>sub_title<|/ref|><|det|>[[88, 385, 285, 400]]<|/det|> +## ACKNOWLEDGEMENTS + +<|ref|>text<|/ref|><|det|>[[88, 405, 911, 442]]<|/det|> +This work was funded by Wellcome Senior Investigator Award 103788/Z/14/Z (SS). We thank Jenny Nichols for advice and protocols for deriving Embryonic Stem Cells. + +<|ref|>sub_title<|/ref|><|det|>[[88, 461, 308, 477]]<|/det|> +## AUTHOR CONTRIBUTIONS + +<|ref|>text<|/ref|><|det|>[[85, 482, 912, 580]]<|/det|> +C.R., and S.S. led the project, conceived, and designed the experiments. C.R., E.Sa., E.Sl., J.Go., N. V., K.L., J.Ga., and T.N. conducted the experiments. C.R. and S.S. analysed the data. C.R. performed the statistical analyses. E.Sl. and X.L. designed and established the \(ASPP2^{\text{WT/RAKA}}\) mouse line. H.H. and F.Z. performed the unwrapping of ASPP2 and F- actin immunostaining in E5.5 embryos. A.V., C.J., T.C., K.C. and C.G. organised the collection of human embryos. C.R. and S.S. wrote the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[88, 619, 206, 634]]<|/det|> +## REFERENCES + +<|ref|>text<|/ref|><|det|>[[85, 639, 911, 901]]<|/det|> +1. Norden, C. Pseudostratified epithelia - cell biology, diversity and roles in organ formation at a glance. J. Cell Sci. 130, 1859-1863 (2017). +2. Hojman, E., Rubbini, D., Colombelli, J. & Alsina, B. Mitotic cell rounding and epithelial thinning regulate lumen growth and shape. Nat. Commun. 6, (2015). +3. Kondo, T. & Hayashi, S. Mitotic cell rounding accelerates epithelial invagination. Nature 494, 125-129 (2013). +4. Williams, M., Bursdal, C., Periasamy, A., Lewandoski, M. & Sutherland, A. Mouse primitive streak forms in situ by initiation of epithelial to mesenchymal transition without migration of a cell population. Dev. Dyn. 241, 270-283 (2012). +5. Voiculescu, O., Bodenstein, L., Jun, I. L. & Stern, C. D. Local cell interactions and self-amplifying individual cell ingression drive amniote gastrulation. Elife 2014, 1-26 (2014). +6. Shook, D. & Keller, R. Mechanisms, mechanics and function of epithelial-mesenchymal transitions in early development. Mech. Dev. 120, 1351-1383 (2003). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 80, 914, 900]]<|/det|> +7. Ramkumar, N. et al. Crumbs2 promotes cell ingression during the epithelial-to-mesenchymal transition at gastrulation. Nat. Cell Biol. 18, 1281-1291 (2016). +8. Martin, A. C. & Goldstein, B. Apical constriction: themes and variations on a cellular mechanism driving morphogenesis. Development 141, 1987-98 (2014). +9. Heisenberg, C.-P. & Bellaiche, Y. Forces in Tissue Morphogenesis and Patterning. Cell 153, 948-962 (2013). +10. Martin, A. C., Gelbart, M., Fernandez-Gonzalez, R., Kaschube, M. & Wieschaus, E. F. Integration of contractile forces during tissue invagination. J. Cell Biol. 188, 735-749 (2010). +11. Ishiuchi, T. & Takeichi, M. Willin and Par3 cooperatively regulate epithelial apical constriction through aPKC-mediated ROCK phosphorylation. Nat. Cell Biol. 13, 860-6 (2011). +12. Chen, X. & Macara, I. G. Par-3 controls tight junction assembly through the Rac exchange factor Tiam1. Nat. Cell Biol. 7, 262-9 (2005). +13. Cong, W. et al. ASPP2 regulates epithelial cell polarity through the PAR complex. Curr. Biol. 20, 1408-14 (2010). +14. Sottocornola, R. et al. ASPP2 binds Par-3 and controls the polarity and proliferation of neural progenitors during CNS development. Dev. Cell 19, 126-37 (2010). +15. Liu, C.-Y. et al. PP1 cooperates with ASPP2 to dephosphorylate and activate TAZ. J. Biol. Chem. 286, 5558-66 (2011). +16. Royer, C. et al. ASPP2 Links the Apical Lateral Polarity Complex to the Regulation of YAP Activity in Epithelial Cells. PLoS One 9, e111384 (2014). +17. Plusa, B. et al. Downregulation of Par3 and aPKC function directs cells towards the ICM in the preimplantation mouse embryo. J. Cell Sci. 118, 505-15 (2005). +18. Nishioka, N. et al. The Hippo Signaling Pathway Components Lats and Yap Pattern Tead4 Activity to Distinguish Mouse Trophectoderm from Inner Cell Mass. Dev. Cell 16, 398-410 (2009). +19. Hirate, Y. et al. Polarity-dependent distribution of angiomotin localizes hippo signaling in preimplantation embryos. Curr. Biol. 23, 1181-1194 (2013). +20. Leung, C. Y. & Zernicka-Goetz, M. Angiomotin prevents pluripotent lineage differentiation in mouse embryos via Hippo pathway-dependent and -independent mechanisms. Nat. Commun. 4, 2251 (2013). +21. Mamada, H., Sato, T., Ota, M. & Sasaki, H. Cell competition in mouse NIH3T3 embryonic fibroblasts is controlled by the activity of Tead family proteins and Myc. J. Cell Sci. 128, 790-803 (2015). +22. Buti, L. et al. CagA-ASPP2 complex mediates loss of cell polarity and favors H. Pylori colonization of human gastric organoids. Proc. Natl. Acad. Sci. 117, 2645-2655 (2020). +23. Kampa, K. M. et al. heterozygous mice are tumor-prone and have attenuated cellular damage - response thresholds. Proc. Natl. Acad. Sci. 53, (2009). +24. Llanos, S. et al. Inhibitory member of the apoptosis-stimulating proteins of the p53 family (iASPP) interacts with protein phosphatase 1 via a noncanonical binding motif. J. Biol. Chem. 286, 43039-44 (2011). +25. Bedzhov, I. & Zernicka-Goetz, M. Self-organizing properties of mouse pluripotent cells initiate morphogenesis upon implantation. Cell 156, 1032-44 (2014). +26. Ichikawa, T. et al. Live Imaging of Whole Mouse Embryos during Gastrulation: Migration Analyses of Epiblast and Mesodermal Cells. PLoS One 8, e64506 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 78, 912, 900]]<|/det|> +27. Leonavicius, K. et al. Mechanics of mouse blastocyst hatching revealed by a hydrogel-based microdeformation assay. Proc. Natl. Acad. Sci. 115, 10375–10380 (2018). +28. Rauzi, M. et al. Embryo-scale tissue mechanics during Drosophila gastrulation movements. Nat. Commun. 6, 8677 (2015). +29. Hennigan, R. F., Fletcher, J. S., Guard, S. & Ratner, N. Proximity biotinylation identifies a set of conformation-specific interactions between Merlin and cell junction proteins. Sci. Signal. 12, (2019). +30. Zhang, P. et al. ASPP1/2-PP1 complexes are required for chromosome segregation and kinetochore-microtubule attachments. Oncotarget 6, 41550–41565 (2015). +31. Choi, W. et al. Remodeling the zonula adherens in response to tension and the role of afadin in this response. J. Cell Biol. 213, 243–260 (2016). +32. Ikeda, W. et al. Afadin: A key molecule essential for structural organization of cell-cell junctions of polarized epithelia during embryogenesis. J. Cell Biol. 146, 1117–1131 (1999). +33. Riedl, J. et al. Lifecut mice for studying F-actin dynamics. Nat. Methods 7, 168–169 (2010). +34. Kosodo, Y. & Huttner, W. B. Basal process and cell divisions of neural progenitors in the developing brain. Dev. Growth Differ. 51, 251–261 (2009). +35. Strzyz, P. J. et al. Interkinetic Nuclear Migration Is Centrosome Independent and Ensures Apical Cell Division to Maintain Tissue Integrity. Dev. Cell 32, 203–219 (2015). +36. Hao, Y. et al. Par3 controls epithelial spindle orientation by aPKC-mediated phosphorylation of apical Pins. Curr. Biol. 20, 1809–18 (2010). +37. Williams, S. E., Ratliff, L. a, Postiglione, M. P., Knoblich, J. a & Fuchs, E. Par3-mlnsc and Gai3 cooperate to promote oriented epidermal cell divisions through LGN. Nat. Cell Biol. 16, (2014). +38. Keder, A. et al. The Hippo pathway core cassette regulates asymmetric cell division. Curr. Biol. 25, 2739–2750 (2015). +39. Gao, L. et al. Afadin orients cell division to position the tubule lumen in developing renal tubules. Development 144, 3511–3520 (2017). +40. Speicher, S., Fischer, A., Knoblich, J. & Carmena, A. The PDZ Protein Canoe Regulates the Asymmetric Division of Drosophila Neuroblasts and Muscle Progenitors. Curr. Biol. 18, 831–837 (2008). +41. Bosveld, F. et al. Epithelial tricellular junctions act as interphase cell shape sensors to orient mitosis. Nature 530, 495–498 (2016). +42. Hart, K. C. et al. E-cadherin and LGN align epithelial cell divisions with tissue tension independently of cell shape. Proc. Natl. Acad. Sci. U. S. A. 114, E5845–E5853 (2017). +43. Tang, Z. et al. Mechanical Forces Program the Orientation of Cell Division during Airway Tube Morphogenesis. Dev. Cell 44, 313-325.e5 (2018). +44. Scarpa, E., Finet, C., Blanchard, G. B. & Sanson, B. Actomyosin-Driven Tension at Compartmental Boundaries Orients Cell Division Independently of Cell Geometry In Vivo. Dev. Cell 47, 727-740.e6 (2018). +45. Finegan, T. M. et al. Tissue tension and not interphase cell shape determines cell division orientation in the Drosophila follicular epithelium. EMBO J. 38, 1–18 (2019). +46. Nakajima, Y.-I., Meyer, E. J., Kroesen, A., McKinney, S. a. & Gibson, M. C. Epithelial junctions maintain tissue architecture by directing planar spindle orientation. Nature 500, 1–5 (2013). +47. Manning, L. A., Perez-Vale, K. Z., Schaefer, K. N., Sewell, M. T. & Peifer, M. The Drosophila Afadin and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[82, 80, 912, 120]]<|/det|> +ZO- 1 homologues Canoe and Polychaetoid act in parallel to maintain epithelial integrity when challenged by adherens junction remodeling. Mol. Biol. Cell 30, 1938–1960 (2019). + +<|ref|>text<|/ref|><|det|>[[82, 123, 911, 181]]<|/det|> +48. Sawyer, J. K., Harris, N. J., Slep, K. C., Gaul, U. & Peifer, M. The Drosophila afadin homologue Canoe regulates linkage of the actin cytoskeleton to adherens junctions during apical constriction. J. Cell Biol. 186, 57–73 (2009). + +<|ref|>text<|/ref|><|det|>[[82, 185, 911, 223]]<|/det|> +49. Salomon, J. et al. Contractile forces at tricellular contacts modulate epithelial organization and monolayer integrity. Nat. Commun. 8, (2017). + +<|ref|>text<|/ref|><|det|>[[82, 226, 911, 264]]<|/det|> +50. Vives, V. et al. ASPP2 is a haploinsufficient tumor suppressor that cooperates with p53 to suppress tumor growth. Genes Dev. 20, 1262–7 (2006). + +<|ref|>text<|/ref|><|det|>[[82, 267, 911, 305]]<|/det|> +51. Wang, Y. et al. ASPP2 controls epithelial plasticity and inhibits metastasis through \(\beta\) -catenin-dependent regulation of ZEB1. Nat. Cell Biol. 16, (2014). + +<|ref|>text<|/ref|><|det|>[[82, 308, 911, 346]]<|/det|> +52. Tordella, L. et al. ASPP2 suppresses squamous cell carcinoma via RelA/p65-mediated repression of p63. Proc. Natl. Acad. Sci. U. S. A. 110, 17969–74 (2013). + +<|ref|>text<|/ref|><|det|>[[82, 349, 911, 387]]<|/det|> +53. Bertran, M. T. et al. ASPP proteins discriminate between PP1 catalytic subunits through their SH3 domain and the PP1 C-tail. Nat. Commun. 10, (2019). + +<|ref|>text<|/ref|><|det|>[[82, 390, 911, 428]]<|/det|> +54. Kas, S. M. et al. Insertional mutagenesis identifies drivers of a novel oncogenic pathway in invasive lobular breast carcinoma. Nat. Genet. 49, 1219–1230 (2017). + +<|ref|>text<|/ref|><|det|>[[82, 431, 911, 470]]<|/det|> +55. Hayashi, S., Lewis, P., Pevny, L. & McMahon, A. P. Efficient gene modulation in mouse epiblast using a Sox2Cre transgenic mouse strain. Mech. Dev. 119, S97–S101 (2002). + +<|ref|>text<|/ref|><|det|>[[82, 473, 911, 510]]<|/det|> +56. Muzumdar, M. D., Tasic, B., Miyamichi, K., Li, N. & Luo, L. A global double-fluorescent cre reporter mouse. Genesis 45, 593–605 (2007). + +<|ref|>text<|/ref|><|det|>[[82, 513, 911, 551]]<|/det|> +57. Huff, J. The Airyscan detector from ZEISS: confocal imaging with improved signal-to-noise ratio and super-resolution. Nat. Methods 12, i–ii (2015). + +<|ref|>text<|/ref|><|det|>[[82, 554, 911, 592]]<|/det|> +58. Tse, J. R. & Engler, A. J. Preparation of hydrogel substrates with tunable mechanical properties. Current Protocols in Cell Biology vol. 47 10.16.1-10.16.16 (2010). + +<|ref|>text<|/ref|><|det|>[[82, 595, 911, 653]]<|/det|> +59. Bazzi, H., Soroka, E., Alcorn, H. L., Anderson, K. V & Hogan, B. L. M. STRIP1, a core component of STRIPAK complexes, is essential for normal mesoderm migration in the mouse embryo. Proc. Natl. Acad. Sci. U. S. A. 114, E10928–E10936 (2017). + +<|ref|>text<|/ref|><|det|>[[82, 656, 911, 694]]<|/det|> +60. Petrie, R. J., Doyle, A. D. & Yamada, K. M. Random versus directionally persistent cell migration. Nat. Rev. Mol. Cell Biol. 10, 538–549 (2009). + +<|ref|>text<|/ref|><|det|>[[82, 697, 911, 735]]<|/det|> +61. Nichols, J. & Jones, K. Derivation of mouse embryonic stem (ES) cell lines using small-molecule inhibitors of Erk and Gsk3 signaling (2i). Cold Spring Harb. Protoc. 2017, 379–386 (2017). + +<|ref|>sub_title<|/ref|><|det|>[[85, 774, 379, 789]]<|/det|> +## EXTENDED DATA MOVIE LEGENDS + +<|ref|>text<|/ref|><|det|>[[82, 795, 911, 813]]<|/det|> +Movie 1 | Time lapse imaging of wild type and ASPP2AEA4 embryos with mT/mG-labelled cell membranes + +<|ref|>text<|/ref|><|det|>[[82, 820, 911, 837]]<|/det|> +Movie 2 | Airyscan imaging and 3D rendering of F-actin in the primitive streak region of representative + +<|ref|>text<|/ref|><|det|>[[85, 844, 380, 861]]<|/det|> +wild type and ASPP2RAKA/RAKA embryos + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[30, 84, 911, 199]]<|/det|> +749 Movie 3 | 3D opacity rendering showing that the primitive streak expands comparatively in E7.5 wild type and ASPP2ΔE4/ΔE4 embryos. Mesoderm cells were labelled by immunofluorescence using an antibody against Brachyury (T). Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. 752 Movie 4 | Time lapse imaging of wild type and ASPP2RAKA/RAKA mesoderm explants positive for the LifeAct- 753 GFP transgene + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 0, 900, 732]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[50, 736, 917, 965]]<|/det|> +Extended Data Fig. 1 | ASPP2 is not required during preimplantation development. a, Localisation pattern of ASPP2 in 16- cell embryos. A cross section through the equatorial plane of a representative embryo is shown. The F- actin cytoskeleton and nuclei were visualised using Phalloidin and DAPI, respectively. The dashed area is magnified on the right. The white arrowhead shows ASPP2 and F- actin colocalising at an apical junction between two outside cells. Scale bars: \(20 \mu \mathrm{m}\) (left panel) and \(5 \mu \mathrm{m}\) (right panel). b, localisation pattern of ASPP2 in human blastocysts. The top panel shows a cross section through the equatorial plane of a representative embryo. The dashed area is magnified on the right to highlight the colocalisation between ASPP2 and F- actin at the level of apical junctions in the trophectoderm (white arrowheads). The bottom row shows a 3D opacity rendering of the same embryo in its totality (left panel) and a focus on its inner cell mass (right panel). TE: trophectoderm; ICM: Inner cell mass. Scale bar: \(20 \mu \mathrm{m}\) . c, F- actin is normally distributed at the apical junctions in the trophectoderm of ASPP2ΔE4/ΔE4 embryos. Maximum intensity projections of representative wild type and ASPP2ΔE4/ΔE4 embryos stained with Phalloidin and DAPI. d, ASPP2 knockdown in E3.5 embryos using siRNA targeting ASPP2 mRNA. Note that the localisation pattern of pYAP S127 was similar in control and ASPP2- depleted embryos. Scale bar: \(20 \mu \mathrm{m}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[68, 75, 928, 644]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[80, 666, 917, 914]]<|/det|> +Extended Data Fig. 2 | ASPP2 is required specifically in the epiblast during proamniotic cavity formation. a, Immunofluorescence of wild type and \(ASP^{2\Delta E4 / \Delta E4}\) E5.5 embryos using an anti- Par6 antibody. b, Phenotype of \(ASP^{2\Delta E4 / \Delta E4}\) embryos at E7.5. The localisation of Brachyury (T) was analysed by indirect immunofluorescence. The ectopic accumulation of cells in the proamniotic cavity of \(ASP^{2\Delta E4 / \Delta E4}\) embryos is indicated by a green star. c, The expression of ASPP2 was conditionally prevented in the epiblast to test for its epiblast- specific requirement ( \(ASP^{2Epi\Delta E4 / \Delta E4}\) embryos). ASPP2 expression pattern was analysed by indirect immunofluorescence. ASPP2 proteins were completely absent at the apical junction of epiblast cells in \(ASP^{2Epi\Delta E4 / \Delta E4}\) embryos. Note that the ASPP2 antibody results in non- specific nuclear signal (also seen in Fig. 1c when depleting ASPP2 by siRNA). The dashed area highlights the ectopic accumulation of cells in the epiblast. d, SCRIB expression pattern was analysed by indirect immunofluorescence in wild type and \(ASP^{2RKA / RAKA}\) embryos. The ectopic accumulations of cells in the epiblast of \(ASP^{2RKA / RAKA}\) embryos was highlighted by a dashed line. e, Magnification of the corresponding regions in d. The blue arrowhead points to the enrichment of SCRIB at the apical junctions. Red arrowheads point to basolateral SCRIB. The F- actin cytoskeleton and nuclei were visualised using Phalloidin and DAPI, respectively. Scale bars: \(20 \mu m\). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 70, 940, 228]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[80, 255, 917, 397]]<|/det|> +Extended Data Fig. 3 | The angle of cell division and their relative position from the base of the epiblast is unaffected in the absence of ASPP2. a, Quantification of cell division angles in the epiblasts of wild type \((n = 3\) embryos, 28 cells) and \(ASPp2^{\Delta E4 / \Delta E4}\) embryos \((n = 3\) embryos, 33 cells). Left panel: Comparison of all cell division angles in wild type and \(ASPp2^{\Delta E4 / \Delta E4}\) embryos. NS: non- significant (nested ANOVA). right panel: Cell division angles were defined as either parallel \((0^{\circ}\) to \(30^{\circ}\) ), oblique \((30^{\circ}\) to \(60^{\circ}\) ) or orthogonal \((60^{\circ}\) to \(90^{\circ}\) ). NS: nonsignificant (Fisher's exact test of independence). b, Position of cell division events in wild type and \(ASPp2^{\Delta E4 / \Delta E4}\) embryos. The relative position of cell division events was expressed as the distance between mother cell position immediately prior to a division event and the base of the epithelium. NS: non- significant (nested ANOVA). + +<|ref|>image<|/ref|><|det|>[[60, 550, 936, 732]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[81, 738, 917, 844]]<|/det|> +Extended Data Fig. 4 | Gross defects in \(ASPp2^{\text{RAKA/RAKA}}\) embryos in a BALB/c background at E7.5. a, Bright field images of wild type and type I (34/41, \(82.9\%\) ) and type II (7/41, \(17.1\%\) ) phenotypes observed in \(ASPp2^{\text{RAKA/RAKA}}\) embryos. Type I embryos exhibited a strong accumulation of cells in their posterior. b, Cells ectopically accumulating in the primitive streak region are unable to apically constrict and do not have enriched F- actin at the apical junctions (orange arrowhead) in comparison to wild type (blue arrow heads). The F- actin cytoskeleton was visualised using Phalloidin. Scale bars: \(20 \mu m\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 45, 933, 202]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[67, 245, 933, 440]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[67, 465, 904, 688]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[80, 703, 917, 915]]<|/det|> +Extended Data Fig. 5 | ASPP2 is required for the integrity of the F- actin cytoskeleton in the epiblast as cells divide. a, The localisation pattern of ASPP2 was analysed in wild type E5.5 embryos by immunofluorescence. b, Magnification of the boxed region in a, showing the localisation pattern of ASPP2 in the VE at the apical junctions (blue arrowhead). Note that ASPP2's nuclear signal is non- specific. c, Super- resolution Airyscan imaging of MDCK cells immunostained for ASPP2 and Afadin. Orange arrowheads highlight the colocalisation of Afadin and ASPP2 at tricellular junctions. d, The localisation pattern of ASPP2 and Afadin was analysed in Caco-2 cells. Afadin and ASPP2, in addition to being enriched at tricellular junctions, could also be found colocalising at bicellular junctions (blue arrowhead) and were in close proximity at the cleavage furrow (orange arrowhead). e, 3D opacity rendering showing the localisation of Afadin at the apical junctions of the VE in an E6.5 wild type embryo. f, Time- lapse imaging of wild type and ASPP2RAKA/RAKA LifeAct- GFP positive embryos. Note how apical F- actin is disrupted in ASPP2RAKA/RAKA LifeAct- GFP positive embryos following a cell division event (black arrowhead). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[68, 50, 927, 465]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[80, 510, 919, 705]]<|/det|> +Extended Data Fig. 6 | General required for ASPP2 across pseudostratified epithelia. a, Mesoderm cell migration is unaffected in the absence of ASPP2. Cell movement, velocity, and directionality from wild type (n=4 explants, 3 cells per explant) and ASPP2RAKA/RAKA (n=4 explants, 3 cells per explant) mesoderm explants were quantified. NS: non- significant (nested ANOVA). b, E9.5 wild type and ASPP2ΔE4/ΔE4 embryos labelled by immunofluorescence with antibodies against FOXC2 (somatic mesoderm) and sarcomeric α-actinin (sarcomeres in cardiomyocytes). Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin respectively. Scale bars: 200 μm. c, Quantification of the proportion of ASPP2ΔE4/ΔE4 and ASPP2ΔE4/ΔE4 ESC-derived cysts with lumens after three days in culture in Matrigel. d, Representative images of wild type and ASPP2RAKA/RAKA ESC-derived cysts after three days in culture in Matrigel. e, Cells from the ectoderm occasionally delaminate basally into the underlying mesoderm in E8.5 ASPP2RAKA/RAKA embryos. The F-actin cytoskeleton was visualised with Phalloidin, respectively. Scale bars: 20 μm. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 143, 70]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[55, 104, 925, 597]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 616, 116, 636]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[40, 655, 952, 953]]<|/det|> +The ASPP2/PP1 complex is not required during preimplantation development. a, ASPP2 was detected by indirect immunofluorescence in E3.5 embryos to analyse its localisation pattern. A cross section through the equatorial plane of a representative embryo is shown (top row), as well as a 3D opacity rendering of the same embryo (bottom row). The F- actin cytoskeleton and nuclei were visualised using Phalloidin and DAPI, respectively. A magnified image of the dashed area is shown on the right. Note how ASPP2 colocalises with F- actin at the apical junctions in cells of the trophectoderm (white arrowheads). This is quantified in the juxtaposed graph showing ASPP2 and F- actin signal intensity along the apical- basal axis of cell- cell junctions. AJ: apical junction; B: base of the trophectoderm. Scale bars: \(20 \mu \mathrm{m}\) and \(5 \mu \mathrm{m}\) (for the magnification). b, The localisation pattern of YAP and Par3 was analysed in wild type and ASPP2RAKA/RAKA embryos by indirect immunofluorescence. A cross section of representative embryos through the equatorial plane shows the localisation of YAP in the nuclei of the trophectoderm in both wild type and ASPP2RAKA/RAKA embryos. Maximum intensity projections of these embryos show the localisation of Par3 at the level of apical junctions in the trophectoderm. Scale bar: \(20 \mu \mathrm{m}\) . c, ASPP2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 940, 111]]<|/det|> +knockdown in E3.5 embryos using siRNA against ASPP2 mRNA. ASPP2 knockdown was confirmed by indirect immunofluorescence. Note how signal at the apical junctions is specific to ASPP2 and how YAP is normally localised to the nuclei of TE cells in ASPP2-depleted embryos. Scale bar: 20 μm. + +<|ref|>image<|/ref|><|det|>[[55, 123, 875, 848]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[43, 871, 121, 890]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[43, 914, 951, 956]]<|/det|> +The ASPP2/PP1 complex is required for the formation of the proamniotic cavity. a, Immunofluorescence of wild type and ASPP2ΔE4/ΔE4 E6.5 embryos using an anti-Par6 antibody. The phenotypic variability of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[38, 42, 960, 545]]<|/det|> +ASPP2ΔE4/ ΔE4 embryos is illustrated, with embryos either lacking cavities (middle row) or exhibiting smaller cavities (bottom row). The green dashed line highlights the ectopic accumulation of cells in the epiblast of ASPP2ΔE4/ΔE4 embryos. b, Magnification of the corresponding regions shown in panel a. Blue arrowheads highlight the enrichment of F- actin at the apical junctions in the epiblast. Note how F- actin is not enriched at the apical junctions but is instead more homogenously distributed across the apical surface of epiblast cells in ASPP2ΔE4/ΔE4 embryos (orange arrowhead). The insets within images are 2x magnifications of the corresponding dashed areas. c, Quantification of F- actin signal intensity along the apical surface of epiblast cells of wild type (n=3 embryos, 5 measurements per embryo) and ASPP2ΔE4/ΔE4 embryos (n=3 embryos, 5 measurements per embryo). Measurements were made on cross sections along the apical domain of individual epiblast cells from apical junction to apical junction (represented with a blue background in the graph). See material and methods for details. d, Immunofluorescence of wild type and ASPP2ΔE4/ΔE4 E5.5 embryos using an anti- Laminin antibody. e, Magnification of the corresponding dashed areas in panel d. f, Immunofluorescence of wild type and ASPP2ΔE4/ΔE4 E6.5 embryos using an anti- SCRIB antibody. g, Magnification of the corresponding dashed areas in panel f. Green arrowheads highlight basolateral SCRIB. Note the enrichment of SCRIB at the apical junctions in the epiblast of wild type embryos (blue arrowhead) and its absence in the corresponding localisation in ASPP2ΔE4/ΔE4 embryos (orange arrowhead). h, Immunofluorescence of wild type and ASPP2RAKA/RAKA E6.5 embryos using an anti- Par6 antibody. The green dashed line highlights the ectopic accumulation of cells in the epiblast of ASPP2RAKA/RAKA embryos. i, Magnification of the corresponding dashed regions in h. Note the reduced amount of Par6 along the apical domain of epiblast cells in ASPP2RAKA/RAKA embryos (orange arrowhead). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 50, 560, 610]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 630, 120, 650]]<|/det|> +
Figure 3
+ +<|ref|>image<|/ref|><|det|>[[600, 55, 930, 547]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[40, 670, 955, 944]]<|/det|> +ASPP2 is required for apical daughter cell reincorporation into the epiblast following cell division events. a, Time lapse imaging of wild type and ASPP2ΔE4/ΔE4 embryos. mT/mG- labelled cell membranes were used to manually track cell movement. Yellow dots highlight mother cells at the apical surface of the epiblast immediately prior to a cell division event. Green and magenta dots identify the resulting daughter cells. Note how both daughters reintegrate the epiblast in the wild type whereas one of the two daughters fails to do so in the absence of ASPP2 even after a prolonged period of time (t=82.5'). b, Diagram illustrating the method used to quantify daughter cell movement following cell divisions. Daughter cell movement was characterised by both the distance travelled (d) and the direction of travel (θ) expressed as the angle between the reference vector (the green vector starting from the initial position of the mother cell prior to the division event to the centre of the embryonic region) and the vector characterising absolute daughter cell movement (the red vector starting from the initial position of the mother cell prior to the division event to the final position of the daughter cell). The left panel illustrates the case of a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 951, 294]]<|/det|> +daughter moving basally to reincorporate the epiblast and the right panel describes abnormal daughter cell movement towards the centre of the embryonic region such as seen in ASPP2ΔE4/ΔE4 embryos. c, Graph quantifying cell movement in wild type (n=3 embryos, 56 cells) and ASPP2ΔE4/ΔE4 embryos (n=3 embryos, 66 cells). For a given pair of daughter cells, each daughter was defined as “apical” or “basal” depending on their respective position relative to the centre of the embryonic region immediately after a cell division event. d, Proportion of daughter cells with an overall apical or basal movement in wild type and ASPP2ΔE4/ΔE4 embryos. Left panel: Quantification of the proportion of daughter cells with an overall apical (0 from 0° to 90°) or basal movement (0 from 90° to 180°) in wild type and ASPP2ΔE4/ΔE4 embryos. Right panel: quantification of the proportion of apical and basal daughters with an overall apical (0 from 0° to 90°) or basal movement (0 from 90° to 180°) in wild type and ASPP2ΔE4/ΔE4 embryos. \(\yen 123,456,7\) p<0.0001, NS: non-significant (Fisher's exact test of independence). + +<|ref|>image<|/ref|><|det|>[[55, 310, 930, 875]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 900, 118, 919]]<|/det|> +
Figure 4
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 43, 950, 590]]<|/det|> +ASPP2 is required for epithelial integrity in the primitive streak. a, Posterior thickening in E7.5ASPP2RAKA/RAKA embryos in a BALB/C background. Left panel: the anteroposterior axis was defined using AMOT localisation pattern. Right panel: comparison of tissue thickness in the anterior (3 measurements per embryo) and the posterior (3 measurements per embryo) of wild type (n=5 embryos) and ASPP2RAKA/RAKA embryos (n=5 embryos). \* p<0.05, \*\*\*\*p<0.0001 (nested ANOVA). b, Cells accumulate in the primitive streak region of ASPP2RAKA/RAKA embryos. Immunofluorescence of E7.5 wild type and ASPP2RAKA/RAKA embryos using a T (Brachyury) antibody. c, Cells ectopically accumulating in the primitive streak region are unable to apically constrict and do not have enriched F- actin at the apical junctions (orange arrowhead) in comparison to wild type (blue arrowheads and magenta dotted lines). Right panel: quantification of F-actin signal intensity along the apical surface of epiblast cells in the primitive streak region of wild type (n=3 embryos, 5 cells per embryo) and ASPP2RAKA/RAKA embryos (n=3 embryos, 5 cells per embryo). d-f, Airyscan imaging reveals the extent of F-actin disorganisation at the surface of cells accumulating ectopically in the primitive streak region of ASPP2RAKA/RAKA embryos. d, 3D opacity rendering of embryo optical halves, enabling visualisation of the apical surface of epiblast cells in the proamniotic cavity. Note the absence of the typical F-actin mesh pattern at the apical surface of cells in the posterior of ASPP2RAKA/RAKA embryos (green dotted line). e, Cross section through the primitive streak region, showing enriched F-actin at the apical junctions of wild type embryos (blue arrowheads) and the formation of F-actin spike-like structures at the contact-free surface of ASPP2RAKA/RAKA embryos. f, En face view of the epiblast's apical surface in the posterior of an ASPP2RAKA/RAKA embryo. Green dotted lines demarcate the disorganised apical region of the posterior and the more organised lateral regions of the epiblast. Right panel: magnification of the epiblast's apical surface in the posterior of an ASPP2RAKA/RAKA embryo showing F-actin forming spike-like structures. Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm. + +<|ref|>image<|/ref|><|det|>[[55, 603, 936, 944]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 117, 61]]<|/det|> +## Figure 5 + +<|ref|>text<|/ref|><|det|>[[41, 84, 949, 286]]<|/det|> +ASPP2RAKA/RAKA embryo are more susceptible to mechanical stress. a, Phospho- myosin levels are higher at the apical junctions of dividing cells in the epiblast (white arrowheads). b, wild type \((n = 4)\) and ASPP2RAKA/RAKA \((n = 2)\) embryos were grown for \(30'\) in cylindrical cavities made of biocompatible hydrogels. The localisation pattern of GATA6 and Myosin was then analysed by immunofluorescence. c, Magnification of the embryos shown in b. The green dotted line highlights the ectopic accumulation of cells seen in ASPP2RAKA/RAKA embryos. Note how Myosin is enriched at the apical junctions of wild type epiblast cells (blue arrowheads). The orange arrowhead points to the abnormal distribution of Myosin at the apical surface of these cells. Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: \(20 \mu \mathrm{m}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 50, 800, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 802, 116, 820]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 949, 955]]<|/det|> +the ASPP2/PP1 complex is required for F- actin organisation during cell division events. a, 3D opacity rendering showing the localisation of ASPP2 in E5.5 wild type embryos at the apical junctions of the visceral endoderm where it colocalises with F- actin. b, Cross section (top row) and 3D opacity rendering (bottom row) of the proamniotic cavity showing the localisation pattern of ASPP2 and F- actin at the apical junctions (white arrowhead). c, The outer surface of the VE and apical surface of the epiblast were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[40, 44, 960, 361]]<|/det|> +computationally ‘unwrapped’, revealing the enrichment of ASPP2 and F- actin at tricellular junctions in the epiblast (green arrowheads). d, The interaction between endogenous ASPP2 and the F- actin- binding protein Afadin was examined in Caco- 2 cells by co- immunoprecipitation. e, The localisation pattern of endogenous ASPP2 and Afadin in Caco- 2 cells was examined by immunofluorescence. The bottom row represents the magnified region highlighted by a dotted box and shows the enrichment of ASPP2 and Afadin at tricellular junctions. ASPP2, Afadin and F- actin signal intensity was quantified across tricellular junctions (graph on the right). f, The localisation pattern of Afadin in the proamniotic cavity was analysed by immunofluorescence in E6.5 wild type embryos. Afadin colocalised strongly with F- actin at tricellular junctions (blue arrowhead). g, The localisation pattern of F- actin was analysed by timelapse microscopy in wild type and ASPP2RAKA/RAKA LifeAct- GFP positive embryos. Note how apical F- actin is disrupted in ASPP2RAKA/RAKA LifeAct- GFP positive embryos following a cell division event (orange arrowhead). h, At later time points, the ectopic accumulation of cells in the epiblast of ASPP2RAKA/RAKA LifeAct- GFP positive embryos was evident (dotted line). Nuclei and the F- actin cytoskeleton were visualised with DAPI and Phalloidin respectively. Scale bars: 20 μm. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 50, 770, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 801, 116, 820]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 949, 955]]<|/det|> +The ASPP2/PP1 complex is required for tissue integrity across a variety of highly proliferative pseudostratified epithelia. a, The primitive streak expands comparatively in E7.5 wild type and ASPP2ΔE4/ΔE4 embryos. Mesoderm cells were labelled by immunofluorescence using an antibody against Brachyury (T). b, Patterning proceeds normally in the absence of ASPP2. The ectoderm and cardiac progenitors were labelled in E8.5 wild type and ASPP2ΔE4/ΔE4 embryos with antibodies against + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 44, 949, 451]]<|/det|> +SOX2 and NKX2.5, respectively. ys: yolk sack, al: allantois, s: somites, hf: head fold, am: amnion, pc: proamniotic cavity. c, Cardiac progenitors can differentiate into cardiomyocytes in E9.5 ASPP2ΔE4/ΔE4 embryos. The presence of the contractile machinery (magenta arrowheads) was assessed in wild type and ASPP2ΔE4/ΔE4 embryos using an antibody against sarcomeric α-actinin. d, Somite architecture is disrupted in ASPP2ΔE4/ΔE4 embryos. The dotted line highlights the contour of a somite in an ASPP2ΔE4/ΔE4 embryo. The star indicates the ectopic accumulation of cells in the centre of this somite. Arrowheads point to mitotic Fig.s. e-h, Quantification of somite characteristics in wild type (n=10 embryos, 58 somites) and ASPP2ΔE4/ΔE4 (n=6 embryos, 35 somites) embryos at E8.5. \* p<0.05, \*\*\*\*p<0.0001 (Student's T-test). i, Apical-basal polarity is defective in the somites of ASPP2ΔE4/ΔE4 embryos. Par6 localised apically in wild type somites (arrowhead) whereas it was absent in ASPP2ΔE4/ΔE4 embryos (star). de: definitive endoderm. j, Head fold formation is defective in ASPP2RAKA/RAKA embryos. The organisation of apical F-actin was disorganised locally in the anterior ectoderm of ASPP2RAKA/RAKA embryos (orange dotted line). F-actin signal intensity along the apical surface of ectoderm cells in disrupted areas in ASPP2RAKA/RAKA embryos (n=3 embryos, 5 cells per embryo) was compared to wild type cells (n=3 embryos, 5 cells per embryo). Measurements were made on cross sections along the apical domain of individual ectoderm cells from apical junction to apical junction (represented with a blue background in the graph). Nuclei and the F-actin cytoskeleton were visualised with DAPI and Phalloidin, respectively. Scale bars: 20 μm (d, i), 50 μm (a, c), 100 μm (b, j). + +<|ref|>sub_title<|/ref|><|det|>[[44, 473, 311, 501]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 523, 765, 545]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 562, 181, 662]]<|/det|> +- Movie1.avi- Movie2.avi- Movie3.avi- Movie4.avi + +<--- Page Split ---> diff --git a/preprint/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9_det.mmd b/preprint/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..ced72e1801f892a98007d9b54c4ef0a7ab957c4c --- /dev/null +++ b/preprint/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9/preprint__1604363e270ed24b44d61a94bde0781537ced8189eb8577dd0ab9120cd9f21c9_det.mmd @@ -0,0 +1,244 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 875, 175]]<|/det|> +# Relative, local and global dimension in complex networks + +<|ref|>text<|/ref|><|det|>[[44, 196, 700, 238]]<|/det|> +Robert Peach University Hospital of Wuerzburg https://orcid.org/0000- 0002- 8738- 5825 + +<|ref|>text<|/ref|><|det|>[[44, 243, 640, 330]]<|/det|> +Alexis Amaudon Imperial College London Mauricio Barahona ( \(\square\) m.barahona@imperial.ac.uk) Imperial College London https://orcid.org/0000- 0002- 1089- 5675 + +<|ref|>sub_title<|/ref|><|det|>[[44, 371, 102, 389]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 407, 904, 451]]<|/det|> +Keywords: Ideal Notions of Dimension, Boundary Constraints, Inhomogeneities Distortions, Discrete Systems, Diffusive Process, Allosteric Communication, Scale- dependent Definition + +<|ref|>text<|/ref|><|det|>[[44, 469, 283, 489]]<|/det|> +Posted Date: July 6th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 507, 463, 527]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 637874/v1 + +<|ref|>text<|/ref|><|det|>[[42, 544, 907, 587]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 623, 907, 667]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on June 2nd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 30705-w. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[223, 63, 778, 81]]<|/det|> +# Relative, local and global dimension in complex networks + +<|ref|>text<|/ref|><|det|>[[115, 95, 890, 170]]<|/det|> +Robert Peach, \(^{1,2,*}\) Alexis Arnaudon, \(^{1,3,*}\) and Mauricio Barahona \(^{1}\) \(^{1}\) Department of Mathematics, Imperial College, London SW7 2AZ, UK \(^{2}\) Department of Neurology, University Hospital Würzburg, Würzburg, Germany \(^{3}\) Blue Brain Project, École Polytechnique Fédérale de Lausanne (EPFL), CampusBiotech, 1202 Geneva, Switzerland (Dated: June 9, 2021) + +<|ref|>text<|/ref|><|det|>[[85, 201, 488, 578]]<|/det|> +Dimension is a fundamental property of objects and the space in which they are embedded. Yet ideal notions of dimension, as in Euclidean spaces, do not always translate to physical spaces, which can be constrained by boundaries and distorted by inhomogeneities, or to intrinsically discrete systems such as networks. To take into account locality, finiteness and discreteness, dynamical processes can be used to probe the space geometry and define its dimension. Here we show that each point in space can be assigned a relative dimension with respect to the source of a diffusive process, a concept that provides a scale- dependent definition for local and global dimension also applicable to networks. To showcase its application to physical systems, we demonstrate that the local dimension of structural protein graphs correlates with structural flexibility, and the relative dimension with respect to the active site uncovers regions involved in allosteric communication. In simple models of epidemics on networks, the relative dimension is predictive of the spreading capability of nodes, and identifies scales at which the graph structure is predictive of infectivity. + +<|ref|>text<|/ref|><|det|>[[86, 581, 488, 882]]<|/det|> +One of the first forays into graph dimensionality originated with Erdős, when he explored the embedding of graphs into a minimum finite dimensional Euclidean space [1]. This line of study helped realise the algorithmic importance of geometric interpretations of graphs [2] but was unfortunately no more than a by- product of the graph embedding process, yielding little actionable information [3]. Later, by characterising the fractal properties of complex networks, a measure of network dimension was defined in terms of the scaling property of a networks topological volume [4- 6]. Whilst the fractal approach showed that dimension plays an important role in characterising network topology and governing dynamical processes such as percolation [7], it was initially limited to global descriptions of network dimension. Extensions that considered the local scaling properties of the volume at different topological distances from a node were introduced [8] and have been used to define a node- centric dimension that can identify influential/central nodes [9, 10] or vital spreaders in infection models [11]. + +<|ref|>text<|/ref|><|det|>[[86, 886, 488, 914], [515, 202, 916, 396]]<|/det|> +However, methodologies based on fractal approaches assume that the topological volume follows a power- law distribution, a strong assumption, not necessarily accurate in real world networks exhibiting heterogeneities [5]. Similarly, in classic papers such as [12] where the dimension of a node is defined using the decay rate of diffusion, the same assumptions of homogeneity are required. Take for example a diffusive source located at the joining of a 1- d and a 2- d space, by measuring the decay rate we immediately ignore the heterogeneity of the space and simply find a dimension somewhere between 1 and 2. In this paper, we posit that the dimension at a node can, and should be, defined as relative to another node. Using the solution of diffusion at other nodes relative to the source we are able to define our relative dimension. + +<|ref|>text<|/ref|><|det|>[[515, 398, 916, 427]]<|/det|> +We start with the Green's function of the diffusion equation + +<|ref|>equation<|/ref|><|det|>[[586, 435, 915, 470]]<|/det|> +\[G_{t}(\mathbf{x}) = (4\pi \sigma t)^{-d / 2}\exp \left(-\frac{\|\mathbf{x}\|^{2}}{4\sigma t}\right), \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[515, 479, 916, 569]]<|/det|> +which, together with an initial condition as a delta function at some position \(\mathbf{x}_0\) , provides a solution of diffusion equation as \(p(\mathbf{x}, t) = G_t(\mathbf{x} - \mathbf{x}_0)\) . As already considered in [13], these solutions have a maxima in there transient response at any other location \(\mathbf{x}\) , at time \(\hat{t}\) and amplitude \(\hat{p}\) given as + +<|ref|>equation<|/ref|><|det|>[[584, 578, 915, 609]]<|/det|> +\[\widehat{t} (\mathbf{x}) = \frac{\|\mathbf{x}\|^2}{2d\sigma},\qquad \widehat{p} (\widehat{t}) = (4e\pi \sigma \widehat{t})^{-\frac{d}{2}}, \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[515, 619, 916, 662]]<|/det|> +where, without loss of generality, \(\mathbf{x}_0 = 0\) . Then, the dimension at any point \(\mathbf{x}\) relative to \(\mathbf{x}_0\) can be evaluated to yield the definition of the relative dimension + +<|ref|>equation<|/ref|><|det|>[[635, 670, 915, 705]]<|/det|> +\[d(\mathbf{x}|\mathbf{x}_0) = \frac{-2\ln\widehat{p}}{\ln\left(4e\pi\sigma\widehat{t}\right)}. \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[515, 717, 916, 808]]<|/det|> +Notice that on the Euclidean space \(\mathbb{R}^d\) , the relative dimension is always equal to \(d\) , independently of \(\mathbf{x}\) and \(\mathbf{x}_0\) . However, if we instead consider a compact subspace \(\Omega \subset \mathbb{R}^d\) , the diffusion dynamics will deviate from those prescribed in Equation 1 due to the presence of boundaries relative to \(\mathbf{x}\) and \(\mathbf{x}_0\) . + +<|ref|>text<|/ref|><|det|>[[515, 810, 916, 915]]<|/det|> +The key property of this Equation (3) that allows us to generalise to graphs is that the positions \(\mathbf{x}_0\) and \(\mathbf{x}\) are not explicit in the right hand side but only used as labels to initialise the diffusion dynamics and measure the transient response. Consequently, the relative dimension is intrinsic as it does not rely on any Euclidean embedding, but only on the existence of diffusion dynamic on + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[84, 60, 920, 500]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[84, 510, 920, 725]]<|/det|> +
FIG. 1. a-c Line graph example with \(n = 500\) nodes representing the interval \([0,1]\) . a The relative dimension of nodes given a source located at \(x = 1 / 3\) . The grey lines are the transient responses of the (non-source) nodes and the position of the peaks in the transient responses are highlighted by dots, coloured by their relative dimension. Dashed lines are isolines of integer relative dimensions. Top inset, a histogram of transient response peaks where the far right bin corresponds to nodes where no peak was observed. b The relative dimension as a function of position in \([0,1]\) shows a plateau near \(d_{rel} = 1\) for nodes near the source. The grey region indicates the set of nodes for which no peak was observed. c The local dimension of each node as a function of scale, where above \(\tau = 1\) , the stationary state is attained and the local dimension is stable. Below, we also observe the increasing effect of the boundaries. d The local dimension of the grid graph \((n = 500)\) at scale \(\tau = 0.1\) , showing inhomogeneities due to the boundaries similar to the line graph. e The evolution of the global dimension as a function of scale for the same line and grid graph as well as their periodic equivalent graphs, illustrating differing behaviors emerging from the influence of the boundaries or the topology. f For the same graphs as in e, we increase the number of nodes in each dimension to measure the convergence rate to the underlying Euclidean dimension \(d_{eucl}\) of \(d_{global} = \max_{j}\mathcal{D}(\tau)\) , showing a faster convergence for lower dimensional spaces and periodic grids. g The relative dimension from a point source (pink dot) to other nodes in a Delaunay grid graph (edges not shown). Grey nodes indicate unreachable nodes for which no transient response peak was detected. The regular grid is shown in i, and an additional mass is added in ii-iv, with varying mass and position, showing a effect similar to gravitational lensing.
+ +<|ref|>text<|/ref|><|det|>[[85, 751, 488, 780]]<|/det|> +the original space. In particular, on graphs we can use the standard diffusion process + +<|ref|>equation<|/ref|><|det|>[[232, 800, 486, 817]]<|/det|> +\[\partial_{t}\mathbf{p}(t) = -L\mathbf{p}, \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[85, 837, 488, 911]]<|/det|> +for a time- dependent node vector \(\mathbf{p}(t)\) with \(L\) the normalised graph Laplacian \(L = K^{- 1}(K - A)\) (corresponding to Euclidean diffusion in the continuous limit [14]). With delta function at node \(i\) with mass \(m_{i}\) as our initial condition, the \(j\) - th coordinate of the solution of Equa + +<|ref|>text<|/ref|><|det|>[[515, 750, 766, 765]]<|/det|> +tion (4) is given by the heat kernel + +<|ref|>equation<|/ref|><|det|>[[635, 769, 916, 792]]<|/det|> +\[p_{j}(t|i) = m_{i}\left(e^{-tL}\right)_{ij}. \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[515, 796, 917, 870]]<|/det|> +Following the previous derivation, we measure the time \(\widehat{t}_{ij}\) and amplitude \(\widehat{p}_{ij}\) of transient response peaks for a node \(j\) relative to the node \(i\) by numerically solving Equation (5), and compute the \(N \times N\) matrix or relative dimensions + +<|ref|>equation<|/ref|><|det|>[[645, 873, 916, 908]]<|/det|> +\[d_{ij} = \frac{-2\ln\widehat{p}_{ij}}{\ln\left(4e\pi\sigma\widehat{t}_{ij}\right)}. \quad (6)\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 66, 489, 216]]<|/det|> +To illustrate our definition of relative dimension, we used a line graph (Figure 1a- b) as a discrete representation of the continuous 1- D interval. We observe that due to the boundaries, a large fraction of nodes do not have a peak in transient response, however for nodes near the source, where the boundary has no influence, the relative dimension is close to the expected \(d = 1\) . It is then natural to define the local dimension of the source node by averaging the relative dimension of the nodes displaying a peak before a given time \(\tau\) as + +<|ref|>equation<|/ref|><|det|>[[171, 227, 487, 268]]<|/det|> +\[\mathcal{D}_i(\tau) = \frac{\sum_{j = 1,j\neq i}^{n}d_{ij}(\tau)\mathbb{1}_{\hat{t}_{ij}< \tau}}{\sum_{j = 1,j\neq i}^{n}\mathbb{1}_{\hat{t}_{ij}< \tau}}, \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[86, 277, 488, 473]]<|/det|> +where \(\mathbb{1}_{\hat{t}_{ij}< \tau}\) is the indicator function for nodes that display a transient response peak before time \(\tau\) . Whilst the local dimension can be likened to a measure of centrality, it is directly capturing the dimension of the local embedding space. In Fig. 1c we observe the increasing effect of the boundaries on local dimension as we increase the scale. Near the center of the line, and when considering nearby nodes (at short scales), one can expect to estimate a dimension near 1, or equivalently 2 for the grid shown in Fig. 1d. The 'boundary insensitive central region' collapses at \(\tau = 1\) (corresponding to the spectral gap of the graph) when all nodes have aggregated information about the boundaries of the line graph. + +<|ref|>text<|/ref|><|det|>[[85, 474, 488, 519]]<|/det|> +Finally, we can define a graph measure of dimension by averaging the local dimensions at each scale to obtain the global dimension + +<|ref|>equation<|/ref|><|det|>[[211, 530, 486, 570]]<|/det|> +\[\mathcal{D}(\tau) = \frac{1}{n}\sum_{i = 1}^{n}\mathcal{D}_i(\tau), \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[86, 581, 488, 912]]<|/det|> +still dependent on \(\tau\) . In Fig. 1e we display the global dimension (as a ratio to the expected Euclidean dimension) for the line and grid graphs and their periodic equivalents. The non- periodic graphs display a maximum in global dimension when the effect of the boundaries is lowest. In contrast, the periodic graphs do not exhibit a peak of the same magnitude suggesting that the topological effect of a compact space has less impact on the global dimension than the presence of a boundary. In the context of graphs as discrete Euclidean spaces, the maximum of this curve can be seen as an approximation of the Euclidean dimension, whereas the global dimension at largest scale characterise the effect of the boundary or topology of the graph. It should be noted that for a non grid- like graph, what is a boundary or a topological effect is not clear, and can be considered together. In addition, by increasing the graph size, and thus reducing the effects of the boundaries, the global dimension converges towards the expected Euclidean dimension (Fig. 1f). For the grid, the surface of the boundary increases with respect to the volume of the space and results in a slower convergence, whereas the global dimension of the periodic + +<|ref|>text<|/ref|><|det|>[[516, 67, 916, 96]]<|/det|> +grids is only affected by the topology, and thus converges faster. + +<|ref|>text<|/ref|><|det|>[[516, 99, 917, 504]]<|/det|> +To develop some intuition for our measure of relative dimension, we consider a simple constructive example using Delaunay meshes (Figure 1g). Given a source- node located at the left boundary or an homogeneous delaunay mesh, relative dimension displays an inhomogeneous distribution radially from the source until nodes are unreachable (Figure 1g(i)). Adding nodes near the centre of the Delaunay grid graph creates local inhomogeneities modifying the underlying space, with a clear analogy to the theory of gravitation. In particular, the added mass acts as a gravitational lens for the diffusion process, whereby nodes directly behind the point mass that were previously unreachable can be reached if the mass is sufficiently large. Small masses are reminiscent of weak lensing (Figure 1g(ii)), whereas larger masses are closer to strong lensing (Figure 1g(iii)) [15]. The behaviour of relative dimension in the presence of inhomogeneities suggests that diffusion effectively occurs on a curved geometry induced by the presence of the mass. Moving the mass towards one boundary (Figure 1g(iv)) shows a coupling between the lensing effect and the presence of the boundary. All three possible effects, boundaries, topology and inhomogeneities are thus important in the notion of dimensions, but may not be distinguishable in more complex networks. Nevertheless, our notion of relative dimension is able to capture them all in one intrinsic graph theoretical measure. + +<|ref|>text<|/ref|><|det|>[[516, 506, 917, 912]]<|/det|> +To further illustrate the benefits of relative dimension we examine allostery in proteins, a phenomena whereby a subset of a protein (active site) can be modulated (activated or inactivated) through binding of a ligand at another subset of the protein (allosteric site). We examine three well- studied allosteric proteins: HRas GTPas, Lac repressor and PDK1 in Figure 2 (for more details on these proteins, see Methods). In HRas, we find a low relative dimension at the active site given the allosteric site as the source (Fig. 2a(i)), whilst in reverse the allosteric site is unreachable to the active site (Fig. 2a(ii)). Whilst an exact statement of allosteric mechanism is not our purpose here, its interesting to note that a low relative dimension suggests a more 'direct' or 'funneled' communication from the allosteric site to the active site. Moreover, the asymmetry of the communication suggests that each half of the protein is purposed for diffusion in opposing directions. The lac repressor protein is constructed from two separate monomers and it is generally understood that binding of both NPF molecules (one on each monomer) is required to activate the lac repressor via a cooperative allosteric effect acting on the hinge region [16]. Given that the allosteric mechanism is cooperative, we don't expect direct communication to the active site from the allosteric site, and instead we examined the change in relative dimension upon using a single allosteric site as a source (Fig. 2b(i)) vs. both allosteric sites as sources + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[88, 67, 480, 600]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 618, 488, 727]]<|/det|> +
FIG. 2. Relative dimensionality in allosteric proteins. a The relative dimensionality of all atoms in HRas GTPase (PDB ID: 3K8Y) given (i) the allosteric site and (ii) the active site as the source of diffusion. b Relative dimension in the multi-allosteric site Lac Repressor protein (PDB ID: 1EFA) given a (i) single allosteric site source and (ii) for both allosteric sites simultaneously. c The relative dimension give the allosteric site as the source in PDK1 (PDB ID: 3ORX).
+ +<|ref|>text<|/ref|><|det|>[[86, 761, 488, 911], [516, 66, 917, 140]]<|/det|> +simultaneously (Fig. 2b(ii)). We find that when binding NFP to just one monomer the relative dimension across the entire protein is lower when compared to using both allosteric sites as sources of diffusion. Finally, binding at the PDK1 interacting fragment (PIF) on PDK1 triggers a signal to start the phosphorylation of the activation loop of the substrates at the ATP pocket, or active site [17], and thus we would expect direct communication between the active and allosteric site. Using the allosteric site as the source of our diffusion, we find that the ac tivation loop is the only protein region that is reachable and returns a measure of relative dimension (Figure 2c). Similarly to HRas, we observe a lower relative dimension at the active site suggesting that communication is more focused from the allosteric to active site. + +<|ref|>text<|/ref|><|det|>[[515, 142, 917, 474]]<|/det|> +Whilst the relative dimension provides insights into allostery, we can look to the local and global dimension to examine protein dynamics. In Figure 3(a), we show a strong correlation between the local dimension and \(\log_{10}(1 / \mathrm{RMSF})\) of residues for Figure 3a(i) an unglycosylated antibody CH2 domain and Figure 3a(ii) an Estrogen Related Receptor g protein, and thus a larger local dimension reduces the flexibility. To examine this further, we plotted the Pearson correlation between local dimension and \(\log_{10}(1 / \mathrm{RMSF})\) for 12 randomly chosen proteins in Figure 3(b). We see that at middling to long timescales of diffusion the correlation plateaus with an average at about \(\sigma = 0.55\) suggesting that the relationship between local dimension and protein flexibility is robust. Calculating the global dimension for the same set of proteins in Figure 3(c), we find a strong correlation (Pearson \(\sigma = 0.73\) ) between global dimension and the \(\log_{10}(1 / \mathrm{RMSF})\) of a protein. The global values of dimension sit between 1.36 and 1.5 for the 12 proteins. These results agree with studies that show spectral dimensionality is generally \(< 2\) and decreases with an increase in flexibility [12, 18]. + +<|ref|>text<|/ref|><|det|>[[515, 475, 917, 701]]<|/det|> +We now take a deeper look at Aquifex Adenylate Kinase (ADK), a dynamical protein with three subdomains: the lid, AMP and core domains. We find that the closed conformation displays a higher local dimension due to the presence of stabilising interactions, not present in the open conformation, creating a more compact structure (Figure 3d). The AMP and lid domains are known to open and close around substrate, agreeably we find both have a lower local dimension relative to the core domain (Figure 3e). Furthermore, we find the AMP domain to have a lower average local dimension than the lid domain in both conformations, a result that we validated using experimental fluorescence correlation spectroscopy that shows the AMP domain to open and close at a faster rate \((16.2\mu s)\) than the lid domain \((46.6\mu s)\) [19, 20]. + +<|ref|>text<|/ref|><|det|>[[515, 703, 917, 911]]<|/det|> +Until now we have considered systems with spatially embedded networks and the relationships between dimension and structure. But what about dynamical processes on networks? Using an SIR model on Watts Strogatz small- world networks [21] and by scanning the infection probability \(\beta\) , we show that the local dimension of a node strongly predicts its infectiousness (Fig. 4a). Below the critical regime of large infectiousness, we find that \(\beta\) is positively correlated with the scale, i.e. the size of the local neighbourhood that should be considered grows with the infection probability. However, near criticality we observe a behavior similar to a phase transition, whereby the best time scale diverges towards values near unity, corresponding to the largest scale of the local + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 61, 884, 320]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 331, 919, 414]]<|/det|> +
FIG. 3. The relationship between root-mean-square fluctuations (RMSF) of protein residues and their local and global dimension. a The log-inverse RMSF vs local dimensionality for each residue in i unglycosylated antibody CH2 domain, i Estrogen Related Receptor g. b The positive correlation (0.539) between local dimension and log-inverse RMSF at the residue level across 12 different proteins as a function of Markov time. c A strong positive correlation between global dimensionality of 12 proteins against log-inverse RMSF. d The local dimension of each residue (i) mapped onto Aquifex Adenylate Kinase in the open (PDB ID: 2RH5) and closed (PDB ID: 2RGX) conformations and (ii) plotted by residue id.
+ +<|ref|>image<|/ref|><|det|>[[85, 427, 919, 611]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 624, 919, 732]]<|/det|> +
FIG. 4. a Heatmap of Pearson correlation between local dimension and node infectiousness for small-world graph \((n = 100\) , average degree \(k = 10\) , probability of rewiring \(p = 0.015\) ). The black line is the average proportion of infected nodes given a single-seed node for a given infection probability \(\beta\) . The transition from low to high proportion of infected nodes indicates the critical point. The dashed line is the maximum correlation for each \(\beta\) . b We vary the probability of rewiring edges \(p\) of small world graphs and display (i) the diffusion time that maximises the correlation between local dimension and infectiousness for varying \(\beta\) , and (ii) the associated correlation coefficient. The correlation is near one close to criticality and above 0.8 for a large range of \(\beta\) . We repeat the analysis of the small-world graph for c a Delaunay grid graph \((n = 400)\) and d the European powergrid network, and show similar linear relationships between scale and infection probability \(\beta\) prior to criticality.
+ +<|ref|>text<|/ref|><|det|>[[86, 760, 161, 774]]<|/det|> +dimension. + +<|ref|>text<|/ref|><|det|>[[86, 777, 488, 912]]<|/det|> +We further computed the local dimension and SIR dynamics for small- world graphs whilst varying the probability of rewiring \(p\) parameter, to interpolate between near regular graphs to random graphs. We find that the relationship between the optimal scale to determine local dimension and infectiousness of a node disappears with the randomness of the network (Fig. 4). At low \(\beta\) , node infectiousness is determined by the distance from hubs in a small- world graph and as \(\beta\) increases, the spreading dy + +<|ref|>text<|/ref|><|det|>[[515, 760, 917, 911]]<|/det|> +namics are faster and nodes further away can be infected. A local dimension with larger time horizons is therefore necessary to obtain a better prediction on node infectiousness. However, in random networks all nodes are on average at equal distance from hubs and no meaningful scale exists. We find similar linear relationships between \(\beta\) and scale in a Delaunay grid graph (Fig. 4c) and the European power grid (Fig. 4d). The decrease in scale for the local dimension to be a good predictor beyond \(\beta\) for both graphs echoed the results of high probability + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 66, 488, 111]]<|/det|> +re- wiring in small- world graphs, suggesting that global graph structure becomes less important if the infection probability is sufficiently high. + +<|ref|>text<|/ref|><|det|>[[86, 115, 488, 310]]<|/det|> +In this paper we have introduced a new framework to define intrinsic notions of dimensions not only on graphs, but on any spaces where a diffusion process can be defined, or even more generally, where a dynamical process from which we can infer the Euclidean dimension can be defined. We have shown the relevance of this approach to examine real world systems such as protein dynamics or epidemic spreading by exploiting the underlying graph structure, but many other applications are within the scope of these three measures, through detailed studies with the relative dimension, probing local dimensions at various scales, or characterising entire graphs with the global dimension. + +<|ref|>text<|/ref|><|det|>[[86, 313, 488, 433]]<|/det|> +We thank David Infield, Thomas Higginson, Francesca Vianello, Florian Song, Paul Expert, Asher Mullokandov and Sophia Yaliraki for valuable discussions. We acknowledge funding through EPSRC award EP/N014529/1 supporting the EPSRC Centre for Mathematics of Precision Healthcare at Imperial and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project- ID 424778381- TRR 295. + +<|ref|>text<|/ref|><|det|>[[89, 490, 489, 911]]<|/det|> +\* These authors contributed equally. [1] P. Erdős, F. Harary, and W. T. Tutte, Mathematika 12, 118 (1965). [2] L. Lovász, Graphs and geometry, Vol. 65 (American Mathematical Soc., 2019). [3] N. Linial, E. London, and Y. Rabinovich, Combinatorica 15, 215 (1995). [4] G. Csányi and B. Szendrői, Physical Review E 70, 016122 (2004). [5] M. T. Gastner and M. E. Newman, The European Physical Journal B- Condensed Matter and Complex Systems 49, 247 (2006). [6] O. Shanker, Modern Physics Letters B 21, 321 (2007). [7] L. Daqing, K. Kosmidis, A. Bunde, and S. Havlin, Nature Physics 7, 481 (2011). [8] F. N. Silva and L. d. F. Costa, arXiv preprint arXiv:1209.2476 (2012). [9] J. Pu, X. Chen, D. Wei, Q. Liu, and Y. Deng, EPL (Europhysics Letters) 107, 10010 (2014). [10] T. Bian and Y. Deng, Chaos: An Interdisciplinary Journal of Nonlinear Science 28, 043109 (2018). [11] T. Wen, D. Pelusi, and Y. Deng, Knowledge- Based Systems, 105717 (2020). [12] S. Reuveni, R. Granek, and J. Klafter, Proceedings of the National Academy of Sciences 107, 13696 (2010). [13] A. Arnaudon, R. L. Peach, and M. Barahona, Physical Review Research 2, 033104 (2020). [14] A. Singer, Applied and Computational Harmonic Analysis 21, 128 (2006). [15] C. W. Misner, K. S. Thorne, J. A. Wheeler, et al., Gravitation (Macmillan, 1973). [16] B. Müller- Hill and S. Oehler, The lac operon (Walter de + +<|ref|>text<|/ref|><|det|>[[515, 66, 919, 636]]<|/det|> +Gryuter New York:, 1996). [17] R. M. Biondi, A. Kieloch, R. A. Currie, M. Deak, and D. R. Alessi, The EMBO journal 20, 4380 (2001). [18] S. Reuveni, R. Granek, and J. Klafter, Physical review letters 100, 208101 (2008). [19] R. Peach, Exploring protein dynamics using graph theory and single- molecule spectroscopy, Ph.D. thesis, Imperial College London (2017). [20] R. L. Peach, D. Saman, S. N. Yaliraki, D. R. Klug, L. Ying, K. R. Willison, and M. Barahona, bioRxiv, 847426 (2019). [21] D. J. Watts and S. H. Strogatz, nature 393, 440 (1998). [22] N. Masuda, M. A. Porter, and R. Lambiotte, Physics reports 716, 1 (2017). [23] F. Song, S. N. Yaliraki, and M. Barahona, (2021), https://doi.org/10.6084/m9.figshare.14039723. v1. [24] B. R. Amor, M. T. Schaub, S. N. Yaliraki, and M. Barahona, Nature communications 7, 12477 (2016). [25] S. Mersmann, L. Strömich, F. J. Song, N. Wu, F. Vianello, M. Barahona, and S. Yaliraki, Nucleic Acids Research (2021), 10.1093/nar/gkab350. [26] A. Kuriata, A. M. Gierut, T. Oleniecki, M. P. Ciemny, A. Kolinski, M. Kurcinski, and S. Kmiecik, Nucleic acids research 46, W338 (2018). [27] F. McCormick, Molecular reproduction and development 42, 500 (1995). [28] G. Buhrman, G. Holzapfel, S. Fetics, and C. Mattos, Proceedings of the National Academy of Sciences 107, 4931 (2010). [29] N. A. Becker, A. M. Greiner, J. P. Peters, and L. J. Maher III, Nucleic acids research 42, 5495 (2014). [30] C. Wilson, H. Zhan, L. Swint- Kruse, and K. Matthews, Cellular and molecular life sciences 64, 3 (2007). [31] J. D. Sadowsky, M. A. Burlingame, D. W. Wolan, C. L. McClendon, M. P. Jacobson, and J. A. Wells, Proceedings of the National Academy of Sciences 108, 6056 (2011). [32] K. a. Henzler- Wildman, V. Thai, M. Lei, M. Ott, M. Wolf- Watz, T. Fenn, E. Pozharski, M. a. Wilson, G. a. Petsko, M. Karplus, C. G. Hübner, and D. Kern, Nature 450, 838 (2007). [33] I. Z. Kiss, J. C. Miller, P. L. Simon, et al., Cham: Springer 598 (2017). + +<|ref|>sub_title<|/ref|><|det|>[[675, 670, 756, 684]]<|/det|> +## METHOD + +<|ref|>text<|/ref|><|det|>[[515, 701, 917, 911]]<|/det|> +Graph diffusion. A network (or a graph) \(G\) is a tuple \(G = (\mathcal{V}, \mathcal{E})\) , consisting of the set of nodes \(N = |\mathcal{V}|\) vertices and \(M = |\mathcal{E}|\) edges connecting them. The network can be described by its \(N \times N\) adjacency matrix which indicates the existence and the weight of a connection (edge) between each pair of nodes. On a graph, there are several non- equivalent definitions of diffusion, which are defined by different forms of the graph Laplacian. However, only one forms corresponds to the Euclidean diffusion, described by the normalised Laplacian \(L = K^{- 1}(K - A)\) where \(K\) is the diagonal matrix of weighted degrees and \(A\) the weighted adjacency matrix [14]. Using the definition of the Laplacian, we can state the diffusion equation for a \(N \times 1\) time- dependent node vector \(\mathbf{p}(t)\) as (4), which + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 66, 488, 110]]<|/det|> +is also known as consensus dynamics [22]. For an initial condition with a delta function of mass \(m\) at node \(i\) , the \(j\) - th coordinate of the solution of (4) is given by (5). + +<|ref|>text<|/ref|><|det|>[[86, 111, 488, 185]]<|/det|> +For comparability across different graphs, we normalise the Markov time of diffusion by the second smallest eigenvalue of the graph Laplacian, \(\lambda_{2}\) (the spectral gap), thus \(\tau = 1\) is the timescale for the diffusion to reach stationarity. + +<|ref|>text<|/ref|><|det|>[[86, 187, 488, 336]]<|/det|> +From our choice of Laplacian, the relative dimension matrix \(d\) (that we introduce in the next section) is symmetric if the initial masses \(m\) are chosen inversely proportional to the weighted node degrees. In addition, to ensure that the stationary state of the diffusion sums to unity, we take \(m_{i} = \overline{k} /(nk_{i})\) where \(\overline{k}\) is the mean weighted degree and \(n\) is the number of nodes in the source. This is used in the protein example, where the initial mass are distributed on all the atoms of the allosteric or active site. + +<|ref|>text<|/ref|><|det|>[[86, 337, 488, 442]]<|/det|> +Comparison with fractal dimension Looking more closely at our definition of relative dimension of Equation 6, it is proportional to the ratio of natural logarithms of peak amplitude and time, which displays similarities to the fractal based approaches where an approximate dimension can be derived from the ratio of natural logarithms of mass at a radius \(r\) , + +<|ref|>equation<|/ref|><|det|>[[238, 448, 487, 483]]<|/det|> +\[d\sim \frac{\log(M)}{\log(r)}, \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[85, 491, 488, 520]]<|/det|> +where the mass \(M\) is simply the number of nodes within some link distance \(r\) [7]. + +<|ref|>text<|/ref|><|det|>[[86, 521, 488, 596]]<|/det|> +Computational aspects. Python code to compute the relative, local and global dimensions is available at https://github.com/barahona- research- group/DynGDim, based on the package NetworkX and numpy/scipy standard libraries. + +<|ref|>text<|/ref|><|det|>[[86, 597, 488, 761]]<|/det|> +Delaunay mesh with mass. We apply Delaunay triangulation to a 40 by 40 grid to return a weighted planar graph for which no point is inside the circumcircle of any triangle. The size of the grid is one unit of the code distance units. We define the weights of each edge as the inverse Euclidean lengths between points and thus obtain a discretisation of the plane. To simulate the gravitational lensing effect, we added additional nodes sampled from a gaussian distribution with parameters with variance 0.05 in the unit square with various positions and number of nodes. + +<|ref|>text<|/ref|><|det|>[[86, 763, 488, 882]]<|/det|> +Protein Graph Construction. The graph representation of the proteins used in this work are computed using [23], an extension of [24]. In short, from a pdb file, each atom is represented by a node, and bonds between atoms by an edge weighted by the energy of the bond. The choice of bonds is key to create a meaningful graph representation, and is explained in [23, 24], see [25] to access the code. + +<|ref|>text<|/ref|><|det|>[[85, 884, 488, 912]]<|/det|> +Root- mean square fluctuation calculations. Enzymatic proteins are inherently flexible and known to ex + +<|ref|>text<|/ref|><|det|>[[515, 66, 917, 156]]<|/det|> +hibit motions across a wide range of temporal and spatial scales. Using simulations, each atom can be assigned a root- mean square fluctuation (RMSF). We calculate the RMSF using the CABS- flex 2.0 webserver which simulates protein dynamics using a coarse- grained protein model [26]. + +<|ref|>text<|/ref|><|det|>[[515, 157, 916, 186]]<|/det|> +Protein dataset. We present here more details on the main set of proteins we used in this work. + +<|ref|>text<|/ref|><|det|>[[515, 188, 916, 383]]<|/det|> +HRas. HRas plays an important role in signal transduction during cell- cycle regulation [27]. Previous studies have shown that calcium acetate acts as an allosteric activator and its mechanism of allostery is mediated by a network of hydrogen bonds, involving structural water molecules, that link the allosteric site to the catalytic residue Q61 [28]. Here, we treat the allosteric and active sites, that are located at opposite ends of the protein (PDB ID: 3K8Y), as the source or target nodes in our relative dimension (since multiple atoms compose the allosteric and active sites, we use all nodes as the source of the diffusive process with a uniform distribution on them). + +<|ref|>text<|/ref|><|det|>[[515, 384, 917, 564]]<|/det|> +Lactose repressor (lac). As a second example, we examine the well- studied lactose repressor (lac) (PDB ID: 1EFA) in Figure 2b, present in E. coli and which binds to the lac operon, a section of DNA, to inhibit the expression of proteins for the metabolism of lactose when no lactose is present [29, 30]. In its complete form, it consists of 4 monomers, with two binding sites to a single DNA strand, inhibiting the genes located between them. The combination of two monomers co- operate to form one of the two binding sites (orange region in Figure 2b). On each monomer there is an allosteric site for the binding of NPF molecules that activate the lac repressor. + +<|ref|>text<|/ref|><|det|>[[515, 566, 916, 761]]<|/det|> +PDK1. Our final allosteric protein is a well- known protein Kinase called PDK1 (PDB ID: 3ORX) that is implicated in the progression of Melanoma's [31]. The allosteric site of PDK1 is a sequence of amino acids, called the PDK1 interacting fragment (PIF), that binds to a phosphate on the catalytic domain. This binding triggers a signal to start the phosphorylation of the activation loop of the substrates at the ATP pocket, or active site [17]. The crystallographic structure (PDB ID: 3ORX) used for our analysis has the molecule BI4 bound at the active site [31] via three hydrogen bounds to a region of high relative dimension, and interacts through hydrophobic forces on a region of low relative dimension. + +<|ref|>text<|/ref|><|det|>[[515, 763, 916, 912]]<|/det|> +Fluorescence correlation microscopy experiments. Protein plasmids of Aquifex Adenylate Kinase (ID:18092 Plasmid:peT3a- AqAdk/MVGDH) were purchased from AddGene as deposited by 'Dorothee Kern Lab Plasmids'. The plasmids were already encoded with two cysteine mutations for maleimide conjugation. ADK was expressed in a 1 litre culture BL21 (DE3) cells via inoculation with 1 mM IPTG. BugBuster was used for cell lysis and TCEP and protease inhibitor was added to the lysate. ADK was purified via HIS- tag with a gravi + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 66, 488, 502]]<|/det|> +trap (GE- healthcare), and a PD- 10 column was used to remove imidazole and exchange into protein buffer (20 mM TRIS, 5 0mM NaCl). TCEP and protease inhibitor were added throughout the purification process. Alexa 488- labelled ADK was prepared overnight using 20 \(\mu \mathrm{M}\) protein with molar ratio 1:10 of protein:Alexa 488. Excess dye was removed using HIS- tag purification and a PD- 10 column. A Typhoon was used to examine the gel of the purified- labelled ADK product and showed no excess fluorophore. The label sites for the FRET experiment were Tyr 52 (AMP \(\mathrm{bd}\) domain) changed to Cys and Val 145 changed to Cys (lid domain) [32]. Samples were diluted to 200 pM in pH 7.5 FRET buffer (20 mM TRIS, 50mM NaCl) with 0.3 mg/ml BSA to prevent surface adsorption. Measurements were taken at thermal equilibrium such that all processes under analysis are statistical fluctuations around the equilibrium. Freely diffusing single- molecules were detected using a home- built dual- channel confocal fluorescence microscope. A tunable wavelength argon ion laser (model 35LAP321- 230, Melles Griot, Carlsbad, CA) was set to 514.5 nm to excite Alexa 488. The beam was focused into the sample solution to a diffraction- limited spot with a high numerical aperture oil- immersion objective (Nikon Plan Apo TIRF 60x, NA 1.45). The closer refractive indexes of oil and glass relative to water and glass make oil immersion preferable due to reduced light reflection. Type FF immersion oil (Cargille, USA) was used due to its negligible fluorescent properties. The obtained fluctuat + +<|ref|>text<|/ref|><|det|>[[515, 66, 917, 141]]<|/det|> +tions of fluorescence intensity are autocorrelated. We fit the autocorrelation curves with a global model that includes components for triplet excitation, conformational dynamics and diffusion, with the assumption that they differed by a factor of 1.6 to distinguish the components, + +<|ref|>equation<|/ref|><|det|>[[566, 156, 866, 230]]<|/det|> +\[G(\tau) = G(0)\left(\frac{1}{1 + \frac{\tau}{\tau_D}}\right)\left(1 - F + Fe^{\frac{\tau}{\tau_m}}\right)\] \[\qquad \left(1 - F_2 + F_2e^{\frac{\tau}{\tau_{conf}}}\right),\] + +<|ref|>text<|/ref|><|det|>[[515, 243, 917, 318]]<|/det|> +where \(\tau_{c}\) , \(\tau_{m}\) and \(\tau_{D}\) are the dynamical timescales of the protein conformational dynamics, mean triplet relaxation and the protein diffusion respectively. \(F_{1}\) is the fraction of molecules entering the triplet state and \(F_{2}\) is the fraction of molecules conformationally fluctuating. + +<|ref|>text<|/ref|><|det|>[[515, 322, 917, 503]]<|/det|> +SIR. For the example with SIR dynamics, we simulated the standard SIR model on networks, using the fast approximation of [33], with open sourced code available at https://github.com/springer- math/Mathematics- of- Epidemics- on- Networks and estimated the infectiousness of each node as the averaged number of removed nodes when the spread started from this node over 500 realisation of the dynamics. To estimate the critical value for the infectiousness \(\beta\) , we computed the average infectability across all nodes for each \(\beta\) and estimated \(\beta_{\mathrm{crit}}\) as the value for which half of the nodes are infected. + +<--- Page Split ---> diff --git a/preprint/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374_det.mmd b/preprint/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6cc2495988cce07cbc54626fd6682559f7c2b0c8 --- /dev/null +++ b/preprint/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374/preprint__161c4cca5fcf8b0ea80ab78fe13b70de6f8ef2f7dad9afa9e6d39d4cb6610374_det.mmd @@ -0,0 +1,707 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 933, 208]]<|/det|> +# Complex Trait Analysis of Human Gut Microbiome-Active Traits in Sorghum bicolor: a new category of human health traits in food crops + +<|ref|>text<|/ref|><|det|>[[44, 230, 682, 270]]<|/det|> +Qinnan Yang University of Nebraska Lincoln https://orcid.org/0000- 0001- 6444- 6931 + +<|ref|>text<|/ref|><|det|>[[44, 276, 328, 316]]<|/det|> +Mallory Van Haute University of Nebraska Lincoln + +<|ref|>text<|/ref|><|det|>[[44, 323, 328, 363]]<|/det|> +Nate Korth University of Nebraska Lincoln + +<|ref|>text<|/ref|><|det|>[[44, 370, 153, 408]]<|/det|> +Scott Sattler USDA- ARS + +<|ref|>text<|/ref|><|det|>[[44, 415, 152, 454]]<|/det|> +John Toy USDA- ARS + +<|ref|>text<|/ref|><|det|>[[44, 461, 328, 501]]<|/det|> +Devin Rose University of Nebraska Lincoln + +<|ref|>text<|/ref|><|det|>[[44, 508, 688, 549]]<|/det|> +James Schnable University of Nebraska- Lincoln https://orcid.org/0000- 0001- 6739- 5527 + +<|ref|>text<|/ref|><|det|>[[44, 554, 407, 594]]<|/det|> +Andrew Benson ( abenson1@unl.edu ) University of Nebraska- Lincoln + +<|ref|>sub_title<|/ref|><|det|>[[44, 638, 102, 655]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 675, 135, 693]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 712, 314, 732]]<|/det|> +Posted Date: March 30th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 750, 473, 770]]<|/det|> +DOI: https://doi.org/10.21203/rs.3. rs- 1490527/v1 + +<|ref|>text<|/ref|><|det|>[[44, 787, 910, 830]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 867, 914, 910]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 26th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33419- 1. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[80, 90, 870, 144]]<|/det|> +Complex Trait Analysis of Human Gut Microbiome-Active Traits in Sorghum bicolor: a new category of human health traits in food crops + +<|ref|>text<|/ref|><|det|>[[80, 199, 856, 250]]<|/det|> +Qinnan Yang \(^{1,2}\) , Mallory Van Haute \(^{1,2}\) , Nate Korth \(^{2,3}\) , Scott E. Sattler \(^{4,5}\) , John Toy \(^{4,5}\) , Devin Rose \(^{1,2,5}\) , James C. Schnable \(^{2,5,6}\) , and Andrew K. Benson \(^{1,2*}\) + +<|ref|>text<|/ref|><|det|>[[80, 312, 671, 330]]<|/det|> +\(^{1}\) Department of Food Science and Technology, University of Nebraska + +<|ref|>text<|/ref|><|det|>[[80, 350, 577, 368]]<|/det|> +\(^{2}\) Nebraska Food for Health Center, University of Nebraska + +<|ref|>text<|/ref|><|det|>[[80, 388, 631, 406]]<|/det|> +\(^{3}\) Complex Biosystems Graduate Program, University of Nebraska + +<|ref|>text<|/ref|><|det|>[[80, 425, 656, 443]]<|/det|> +\(^{4}\) Wheat, Sorghum and Forage Research Unit, USDA- ARS, Lincoln, NE + +<|ref|>text<|/ref|><|det|>[[80, 462, 655, 480]]<|/det|> +\(^{5}\) Department of Agronomy and Horticulture, University of Nebraska + +<|ref|>text<|/ref|><|det|>[[80, 499, 595, 517]]<|/det|> +\(^{6}\) Center for Plant Science Innovation, University of Nebraska + +<|ref|>text<|/ref|><|det|>[[80, 536, 305, 553]]<|/det|> +\(^{*}\) Corresponding Author + +<|ref|>text<|/ref|><|det|>[[80, 574, 264, 590]]<|/det|> +Andrew K. Benson + +<|ref|>text<|/ref|><|det|>[[80, 610, 633, 627]]<|/det|> +Nebraska Food for Health Center, University of Nebraska- Lincoln + +<|ref|>text<|/ref|><|det|>[[80, 647, 727, 664]]<|/det|> +Department of Food Science and Technology, University of Nebraska- Lincoln + +<|ref|>text<|/ref|><|det|>[[80, 684, 393, 700]]<|/det|> +Email address: abenson1@unl.edu + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 92, 188, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[111, 123, 885, 707]]<|/det|> +AbstractSeveral bioactive components of the human diet have major effects on composition and function of the gut microbiome, but no systematic framework exists for understanding variation in microbiome- active components amid the vast amount of genotypic and phenotypic variation within a given species of food crop. Here we present a powerful new approach for complex trait analysis of Microbiome- Active Traits (MATs) in food crops. Capitalizing on a novel automated in vitro microbiome screening (AiMS) methodology to quantify human gut microbiome phenotypes after fermentation of grain from genetically diverse lines, we show how microbiome phenotypes can be used as quantitative traits for genetic analysis. Quantitative Trait Locus (QTL) analysis of AiMS- based phenotypes across grain samples from 294 sorghum (Sorghum bicolor) recombinant inbred lines identified significant QTLs at 10 different genomic regions that collectively control MATs affecting 16 different microbial taxa. Segregation analysis and validation in Near- Isogenic Lines (NILs) confirmed that overlapping QTL peaks for microbiome phenotypes, seed color, and tannin concentration are driven by variation in the Tan2 (chromosome 2) and Tan1 (chromosome 4) regulators of the tannin biosynthetic pathway. Candidate genes at other QTLs suggest that variation in a diverse array of plant molecules can drive MATs. + +<|ref|>sub_title<|/ref|><|det|>[[113, 755, 214, 773]]<|/det|> +## Significance + +<|ref|>text<|/ref|><|det|>[[111, 790, 875, 885]]<|/det|> +SignificanceThe ability to define the architecture of genetic and phenotypic variation controlling MATs in a crop species provides an entirely new approach for studying the complex interactions between bioactive dietary components, the human gut microbiome, and microbiome- associated human + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 872, 258]]<|/det|> +health outcomes. Our study highlights how existing genetic resource populations of food crops can be exploited for co- analysis of MATs and seed traits to efficiently pinpoint candidate loci and pathways controlling MATs. Ultimately, this new approach will enable discovery of novel MATs and paves the way for strategies to incorporate MATs into crop improvement programs to improve human health traits of food crops. + +<|ref|>sub_title<|/ref|><|det|>[[115, 276, 220, 293]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[110, 309, 876, 777]]<|/det|> +Crop breeding and improvement programs have led to significant gains in agronomic and yield traits of major food crops, and current efforts to improve sustainability traits such as carbon footprints and water use hold great promise for feeding the growing world population (1, 2). Tools for genetic improvement including whole genome assemblies, novel methods for phenotyping complex traits, and high- density genotyping of unique genetic resource populations are becoming more widely available across different crop species (1–6). While trait improvement to feed a growing world population is important, efforts to improve nutritional and health- promoting traits in food crops are disproportionate, with most such efforts being focused on traits that can improve nutritional deficiencies in underdeveloped countries (7–9). The disproportionate attention to health and nutrition versus yield and sustainability was recently highlighted in the Plant Science Decadal Vision (9), which lists increasing emphasis on research to improve nutritional/ health characteristics in crop plants among the major goals of the vision. + +<|ref|>text<|/ref|><|det|>[[111, 791, 848, 886]]<|/det|> +Over the last six decades the incidence of complex lifestyle diseases such as obesity, diabetes, metabolic disease, inflammatory bowel diseases have grown at alarming rates in countries with westernized diets (10–13). In addition to genetic and environmental + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 881, 320]]<|/det|> +components, many diseases in these categories are also associated with disconfiguration of the human gastrointestinal microbiome (14–16). In instances such as obesity and metabolic liver diseases, disconfigured microbiomes have been shown to be causal to disease processes (17–20). Dietary factors are also associated with these same diseases and diet has a major effect on taxonomic configuration and function of the gut microbiome (21, 22). Consequently, there is tremendous interest in developing novel foods and novel food ingredients that could be used to manipulate the gut microbiome in predictable ways to reduce susceptibility to diseases (23). + +<|ref|>text<|/ref|><|det|>[[110, 348, 883, 626]]<|/det|> +Several bioactive molecules in human diets are known to affect taxonomic configuration and function of the gut microbiome (24, 25), but major knowledge and methodological gaps currently preclude the systematic examination of the interactions between dietary components and the human gut microbiome. In addition to gaps in our understanding of mechanisms and lack of data defining effects across microbiomes of diverse human populations, there is currently no context for understanding how the vast degree of genetic variation within food crop species can affect concentrations and bioavailability of microbiome- active dietary components. + +<|ref|>text<|/ref|><|det|>[[110, 644, 870, 888]]<|/det|> +We have developed a novel approach for complex trait analysis of human gut Microbiome- Active Traits (MATs) to systematically study variation in microbiome- active components within a food crop species. Our approach uses Automated in vitro Microbiome Screening (AiMS), a new high- throughput method for quantitative phenotyping of microbiome activity across genetically diverse panels of crop species. In this proof- of- concept study, we used quantitative measurements of microbiome phenotypes from AiMS reactions—taxonomic abundances and concentrations of major end products of microbial fermentation (Short Chain + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 370]]<|/det|> +Fatty Acids—SCFA) that are important to human health—along with seed color and tannin concentration phenotypes for genetic analysis of MATs and seed traits across a well- characterized population of Recombinant Inbred Lines (RILs) of Sorghum bicolor. We use Quantitative Trait Locus (QTL) mapping to characterize the genetic architecture of loci controlling MATs defined by taxonomic and metabolic phenotypes and segregation analysis to demonstrate how MATs and grain traits (seed color and tannin production) can be combined to identify candidate loci. We also demonstrate how Near- Isogenic Lines (NILs) differing only in alleles at a candidate locus to confirm effects of causal variants. + +<|ref|>sub_title<|/ref|><|det|>[[113, 387, 177, 404]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[112, 421, 880, 479]]<|/det|> +Automated in vitro Microbiome Screening (AiMS) of grain from BTx623 and IS3620C identifies distinct effects of the parental lines on human gut microbiomes. + +<|ref|>text<|/ref|><|det|>[[111, 495, 880, 889]]<|/det|> +For this proof- of- concept study, we focused on a well- characterized set of sorghum RILs derived from two genetically diverse parents (IS3620C from the Guinea sorghum subpopulation and BTx623 a blend of the Kafir and Caudatum sorghum subpopulations) (26, 27). Variation in human gut microbiome- active traits (MATs) was initially evaluated in fermentations of grain from the BTx623 and IS3620C parental lines across microbiomes from 12 different human donors (Materials and Methods). Baseline (before fermentation) composition of the microbiomes from each of the 12 human donors are shown in Figure S1. PERMANOVA analysis based on \(\beta\) - diversity using Bray- Curtiss distances of the microbiome after fermentation illustrates significant differences in overall microbiome composition after fermentation of grain from either of the parental lines (Figure S2), with microbiomes from 10/12 of the donors showing significant effect of parental line ( \(p < 0.05\) ). In the ten cases where microbiomes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 873, 180]]<|/det|> +displayed significant line- specific effects, sorghum genotype explained \(27 - 53\%\) of the observed variation, suggesting differences in grain composition between the BTx623 and IS3620C parental lines may drive distinct fermentation patterns of the gut microbiome. + +<|ref|>text<|/ref|><|det|>[[111, 199, 872, 590]]<|/det|> +The individual taxonomic responses of the microbiomes to the parental lines were donor microbiome- specific, with 27 microbial clades showing significant differences in abundances in fermentations of the two parental lines ( \(p < 0.05\) , unadjusted repeated measures ANOVA, Figure 1). Nine clades showed consistent, differential responses to the parental grain lines in microbiomes of multiple (three or more) donors ( \(p < 0.05\) , Wilcoxon test). For example, the clades Coprococcus 1, Coprococcus 3 and Bifidobacterium were more abundant in fermentations of IS3620 grain across 4, 7 and 5 donors, respectively, than that of BTx623, and didn't change significantly in the remaining microbiomes. In contrast, Escherichia, Parasutterella and Bacteroides were more abundant across 7, 5 and 7 donor microbiomes in fermentation of BTx623 than in IS3620C, respectively, and did not respond significantly in the remaining microbiomes. + +<|ref|>sub_title<|/ref|><|det|>[[112, 607, 604, 627]]<|/det|> +## Mapping QTL for MATs in a recombinant inbred population. + +<|ref|>text<|/ref|><|det|>[[111, 643, 870, 816]]<|/det|> +The microbiome of donor subject S765 showed the most significant difference (Figure S2A) between parental lines and many of the individual taxonomic responses observed in this microbiome were consistent with the population- wide average responses observed across all twelve donors (Figure S2B). This microbiome was subsequently used for AiMS- based phenotyping of 294 individual RILs drawn from the BTx623 x IS3620C RIL population. + +<|ref|>text<|/ref|><|det|>[[111, 830, 870, 886]]<|/det|> +Across the microbial profiles of AiMS reactions from all 294 RILs, a total of 84 genera were represented by at least 100 reads in the resulting 16S microbiome data after rarefaction to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 879, 448]]<|/det|> +15,452 reads per sample and these taxa were used for downstream analyses. After log transformation, observed abundances for many of these taxa across fermentation reactions of the 294 RILs were non- normally distributed (Table S1; Shapiro- Wilk) and consequently, we used a non- parametric model for QTL mapping with the log transformed abundances of individual taxa (Dataset S1) serving as phenotypes and a previously published set of genetic markers scored for genotypes across the BTx623 x IS3620C RIL population (27). A total of 26 significant QTLs (p < 0.05, 1,000 permutations) and several suggestive QTLs (p < 0.1, 1,000 permutations) were identified that influence the abundance of 19 different microbial clades (Table S2). These QTLs were positioned on nine of the ten sorghum chromosomes, illustrating a polygenic architecture for genetic control of variation in human microbiome phenotypes in sorghum. + +<|ref|>text<|/ref|><|det|>[[111, 460, 872, 888]]<|/det|> +The microbial taxa for which significant QTLs were detected in the sorghum genome included representatives of three different phyla which are commonly dominant members of human gut microbiome communities (Bacteroidetes, Firmicutes, and Proteobacteria). The genetic associations were most frequently detected for members of the phylum Firmicutes, with 23 of the 26 significant QTLs corresponding to taxa from this phylum. Among these 23 significant QTLs, 14 correspond to genera in the families Ruminococcaceae or Lachnospiraceae. Genera from these families, particularly Faecalibacterium and Roseburia, are increasingly being recognized as beneficial organisms in the microbiome that reduce susceptibility to inflammatory diseases (28–35). Importantly, while Faecalibacterium and Roseburia showed highly significant genetic associations with variation in the RIL population, they did not exhibit significant differences between the parental lines, suggesting these microbial phenotypes and genetic associations in the RILs could be due to transgressive segregation. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 126, 858, 184]]<|/det|> +## Overlapping QTLs for multiple microbial taxa and their metabolic products of fermentation (SCFAs) define major effect loci (MEL) in the sorghum genome + +<|ref|>text<|/ref|><|det|>[[111, 198, 875, 667]]<|/det|> +Among the 26 significant QTLs, overlapping QTLs for two or more microbial taxa were present on chromosome 2, chromosome 3, chromosome 4 and chromosome 5 (Table 1 and Figure 2C), suggesting they behave as MELs with pleiotropic effects on multiple microbial taxa. The putative MEL influencing the greatest number of traits was located on chromosome 4, where QTLs in a genomic interval from 59- 62Mb affected the relative abundance of a diverse group of eight different microbial taxa ( \(p< 0.05\) ; 1000 permutations). Specifically, the physical peaks for Catenibacterium, Roseburia, Coprococcus 1, and Faecalibacterium (61,878,324), were immediately adjacent to peaks for RuminococcaceaeUCG.002 (61,555,802), Christensenellaceae_R7 group (61,304,986), Paraprevotella (61,161,519), Eubacterium brachy group (60,114,467) and Paeniclostridium (59,462,696). These QTL peaks also exhibited some of the highest LOD scores, (LOD 8.64 for Paeniclostridium, LOD 7.37 for Christensenellaceae_R7 group, LOD 7.01 for Faecalibacterium, LOD 5.41 for Roseburia, LOD 5.26 for Catenibacterium, and LOD 4.73 for Ruminococcaceae UCg.002). + +<|ref|>text<|/ref|><|det|>[[112, 680, 840, 886]]<|/det|> +Two MELs were identified on chromosome 2 with a peak in the 7.9- 9.6 Mb region that affected three genera (Christensenellaceae_R7 group, Ruminococcaceae_UCG002 group, Paeniclostridium, and Paraprevotella). The chromosome 2 signal came from a low recombination region of the genome, resulting in broad confidence intervals for QTL peaks spanning the range from 4.83 Mb – 77.22 Mb. Two MELs were identified on chromosome 3, one with a peak at 4.04 Mb affecting two members of the Lachnospiraceae (Dorea and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 881, 295]]<|/det|> +Coproccoccus 3) and a second with a peak at 71.5Mb affecting a member of the family Lachnospiraceae (Roseburia) and a member of the Christensenellaceae_R7 group. Both chromosome 3 peaks were within regions of low recombination and shared broad confidence intervals from 3.81 Mb to 73.24 Mb. Similarly, a fifth MEL within a low- recombination region on chromosome 5 from 7.45Mb to 55.7Mb affected three genera from the family Lachnospiraceae (Coproccoccus 3, Blautia, and Dorea). + +<|ref|>text<|/ref|><|det|>[[111, 310, 874, 850]]<|/det|> +In addition to testing for association with abundances of microbial taxa, we also used metabolic end products of microbial fermentation as traits, specifically focusing on concentrations of major SCFA from the AiMS fermentations as functional phenotypes of microbiome. Significant QTLs were detected for propionate, butyrate, and valerate production (Table 1 and Figure 2A). The overlapping QTL peaks on chromosome 2 at 65,688,971 bp for butyrate and valerate production both shared confidence intervals from 65.54 Mb - 67.24 Mb, which overlapped with the QTL for Paraprevotella. QTLs on chromosome 5 within the confidence interval of 7.45 Mb - 55.7 Mb were associated with propionate, butyrate, and valerate and overlapped with QTLs for abundances of Coprococcus3, Blautia, and Dorea. Overlapping of QTLs for microbial taxa and SCFAs implies variation at these loci may drive significant effects on taxonomic abundances and metabolites from their fermentation activities. Interestingly, the most significant MEL on chromosome 4 (based on number of microbial taxa affected) did not show significant effects on SCFA, suggesting that the broad effect on microbial taxa may confound the SCFA phenotypes or that the QTLs manifest in ways that do not significantly influence metabolic end products of fermentation. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[111, 89, 820, 144]]<|/det|> +# Major QTLs for seed color overlap with MELs for MATs and QTLs for tannin content on chromosome 2 and chromosome 4. + +<|ref|>text<|/ref|><|det|>[[110, 161, 870, 667]]<|/det|> +While both parental sorghum lines produce light colored seeds, many of the 294 RILs produce dark colored seeds (Dataset S2), suggesting transgressive segregation of seed color phenotypes. QTL analysis using seed color identified major QTL for seed color on chromosome 2 (7.93 Mb) and chromosome 4 (61.5 Mb) (Figure 3A). The chromosome 4 QTL peak for seed color (61.5 Mb) overlapped with the physical location (61.1- 61.8Mb) of the MEL on chromosome 4 for eight microbial traits and the QTL peak on chromosome 2 for seed color overlapped with the QTL peaks for the MEL on chromosome 2 at 4.83- 12.79Mb for three microbial traits (Figure 3A). The putative overlapping QTL peaks for seed color and MATs implied that allelic variation at the 61.5Mb region of chromosome 4 and 4.83- 12.79Mb region of chromosome 2 affected molecular components of seed color, and these molecules may be driving the observed MATs in the human gut microbiome. Importantly, both the seed color phenotypes and microbial phenotypes affected by these QTLs displayed transgressive phenotypes in the RILs since seed color and the microbes affected by the QTL were not significantly different between the parental lines. + +<|ref|>text<|/ref|><|det|>[[110, 680, 884, 885]]<|/det|> +The overlapping QTLs for seed color and MATs are very close to the physical locations of the Sorghum bicolor Tan2 (chromosome 2, 7.97Mb) and Tan1 (chromosome 4 62.3Mb) genes, which encode transcription factors that are necessary for expression of genes in the polyphenol/flavonoid pathways, including proanthocyanins (condensed tannins) that contribute to seed color in sorghum (36- 39). The Tan1 gene (Sobic004G280800) is located physically at 62,315,396- 62,318,779 and encodes a WD40- like protein while the Tan2 gene on chromosome + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 860, 146]]<|/det|> +2 (Sobic002G076600) is located physically at 7,975,937- 7,985,221 and encodes an b- HLH- like transcription factor (37). + +<|ref|>text<|/ref|><|det|>[[111, 163, 881, 406]]<|/det|> +We subsequently quantified tannin production from each of the 294 RIL lines and from two parental lines (Dataset S3) and like the trait values for seed color and MATs, tannin production was also transgressive with 57 lines producing significant levels of tannins (10.34- 118.08 mg/g seed) whereas the remaining 237 lines only produced low levels (0.09- 9.60 mg/g), similar to the parental lines (BTx623: 1.47 mg/g and IS3620C: 1.12mg/g). Highly significant QTLs for tannin content were also identified on chromosome 2 and chromosome 4 overlapping with the peaks for RGB and specific microbial taxa (Figure 3A). + +<|ref|>sub_title<|/ref|><|det|>[[112, 458, 882, 515]]<|/det|> +## Segregation of parental alleles at Tan1 and Tan2 in the RILs explains transgressive segregation of seed color, tannin production, and microbiome phenotypes + +<|ref|>text<|/ref|><|det|>[[111, 531, 880, 888]]<|/det|> +Functional products from both the Tan1 and Tan2 regulatory genes are required for tannin synthesis and high- impact mutations in either gene can mask effects of a dominant allele at the other locus (duplicate recessive epistasis), blocking tannin synthesis and yielding light colored seed (36). Neither BTx623 nor IS3620C produce significant quantities of tannins as a result of recessive loss of function alleles in the Tan1 (BTx623) or the Tan2 (IS3620C) genes. More specifically, BTx623 carries dominant (wildtype) alleles of Tan2, but at the Tan1 locus is homozygous for the tan1- b allele, which has a 10- base insertion in the C- terminal exon that truncates 35 amino acids from the C- terminus (37). In contrast, the IS3620C parent carries dominant alleles at Tan1, but is homozygous for the tan2- c allele with a 95- base deletion that removes the entire intron between exons 7 and 8 (36). Thus, inheritance of either or both of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 875, 360]]<|/det|> +tan1- b (BTx623 parent) and tan2- c (IS3620C parent) alleles in the RILs would yield non- tannin phenotypes with light colored seeds. In contrast, RILs inheriting dominant Tan1 +/+ (IS3620C parent) and Tan2+/+ (BTx623) alleles from each parent would be expected to result in a tannin phenotype with dark seed color (transgressive phenotypes). If tannin production or function of Tan1 and Tan2 are indeed drivers of the QTLs for microbial taxa associated with the chromosome 2 and chromosome 4, then we would expect the transgressive phenotypes of the microbes to co- segregate with tannin production phenotype, dark seed color, and inheritance of dominant parental haplotypes linked to Tan1 +/+ (IS3620C parent) and Tan2+/+ (BTx623). + +<|ref|>text<|/ref|><|det|>[[111, 385, 881, 888]]<|/det|> +When the RILs are grouped based on parental haplotypes of markers linked to Tan1 and Tan2, we were able to predict parental haplotypes at Tan 1 and Tan 2 in 254 of the lines. Tannin production and dark seeds were almost exclusively found among the 45 RILs with haplotypes linked to wildtype Tan1 alleles from the IS3620C parent and wildtype Tan2 alleles from BTx623 (Figure 3B, Figure 3C and Dataset S4). Like the seed phenotypes, analysis of microbiome phenotypes across RILs in the four different genotypic classes at Tan1 and Tan2 showed that microbial genera associated with the chromosome 2 and chromosome 4 QTL peaks displayed the expected distribution across the four genotypic classes of RILs. Christensenellaceae R7 group, Catenibacterium, Coprococcus 1, Roseburia, Paeniclostridium, Faecalibacterium, and Ruminococcaceae_UCG.002 group all showed significantly higher abundances in tannin- producing RILs carrying wildtype haplotypes at the chromosome 4 Tan 1 (IS3620C) and chromosome 2 Tan2 (BTx623) regions versus the other three categories carrying either tan1- b from BTx623, tan2- c from IS3620C, or both tan1- b/tan2- c from each parent (Figure 3C and Dataset S4) and there were no significant differences between the other three categories. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 876, 257]]<|/det|> +Paraprevotella also showed a phenotypic association but was significantly lower abundance in the RILs carrying wildtype alleles at Tan1 and Tan2. Thus, the microbiome phenotypes, seed color, and tannin production show the expected pattern of co- segregation with allelic variation at the Tan1 and Tan2 loci, which can explain the transgressive seed color, tannin, and microbial phenotypes. + +<|ref|>text<|/ref|><|det|>[[111, 310, 875, 404]]<|/det|> +RILs grouped by haplotype of markers linked to Tan1 and Tan2 and Near- Isogenic Lines (NILs) differing at the Tan1 locus confirm the effects of mutation at Tan1 across microbiomes from multiple human subjects. + +<|ref|>text<|/ref|><|det|>[[111, 420, 880, 884]]<|/det|> +To evaluate the effects of Tan1 and Tan2 on gut microbes across microbiome from multiple human subjects, seeds from RILs were grouped by haplotype of markers linked to Tan1 and Tan2 and tested in fermentations with the subject 765 microbiome as well as the microbiomes from 11 other subjects used in the pilot study. Thirty genera were detected with significant different abundance between Tan1 Tan2 haplotype and vs all other haplotypes (rANOVA analysis followed by FDR correction, Figure 4A). For example, Tan1, Tan2 lines had significantly higher abundances of Faecalibacterium, Christensenellaceae R7 group, Roseburia and Coprococcus1 than lines carrying tan1- b, tan2- c or both alleles. On the other hand, the abundance of Odoribacter was significantly lower in Tan1 Tan2 lines (Figure 4B). Thus, the effects of variation at Tan1 and Tan2 appear to be similar across diverse human microbiomes. To further confirm causal effects of variation at the tannin regulatory loci on gut microbes, we used three pairs of NILs developed in different genetic backgrounds ("Wheatland", BTx631, KS5) with each pair effectively differing only in their allelic content at + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 881, 593]]<|/det|> +the Tan1 locus but are fixed for the wildtype allele at the Tan2 locus. Seed from the pairs of NILs was used as substrate in AiMS reactions with microbiomes from the same 12 subjects. As illustrated in Figure S3, the null mutation of Tan1 locus had significant effects on \(\beta\) - diversity of microbiome from subject 765 as well as the microbiomes from each of the other 11 donors. Several taxa across each of the 12 microbiomes showed significant responses to tannin (Figure 4A and 4B). For example, the NILs homozygous for wildtype alleles at Tan1 and RILs homozygous for wildtype haplotypes at Tan1 and Tan2 had significantly higher abundances of Faecalibacterium, Christensenellaceae R7 group and Roseburia with consistent responses of these taxa in magnitude and directionality. Correlation analysis of abundant taxa in AiMS reactions across all 12 subjects also showed significant correlation of multiple taxa from 14 different families that aggregate to 5 different phyla (Figure 4C, \(\mathrm{R} = 0.74\) , \(\mathrm{p}< 6.6\mathrm{e} - 07\) ). Thus, despite the differences in genetic background of the Tan1 alleles in the NILs versus the RILs, and despite the unique microbiome context of each subject, microorganisms across broad ranges of phylogenetic space show very similar responses. + +<|ref|>text<|/ref|><|det|>[[111, 608, 876, 888]]<|/det|> +Candidate genes associated with QTLs on chromosome 2, chromosome 3 and chromosome 5 QTLs defining putative MELs with pleiotropic effects on the microbiome were also identified on chromosome 2, chromosome 3, and chromosome 5 (Fig. 2). To identify potential candidate variation that may be driving the microbiome phenotypes, we first examined tissue- specific expression patterns from publicly available expression datasets (38) for all annotated gene models between the flanking genetic markers for each MEL, specifically focusing on genes that are moderately- highly expressed in seed. Genes expressed in seed were further filtered to identify gene models with sequence variation in one of the two parental lines classified as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 866, 360]]<|/det|> +having moderate or high impact on gene function using public resequencing data (39). This approach identified several candidates, and their predicted functions suggest quite different effects on seed composition. Candidates include a putative cytochrome P450 associated with synthesis of sesquiterpenes (Sobic.002G273600), putative gibberellin 3-beta-dioxygenase (Sobic.003G045900) and abscisic acid insensitive3 (ABI3) transcription factor (Sobic.003G398200) that modulate GA and ABA signaling in seed development, and a putative xylanase inhibitor protein precursor (Sobic.005G098700) (Fig. 2) that could influence microbial xylanases in the AiMS reactions. + +<|ref|>sub_title<|/ref|><|det|>[[115, 423, 203, 440]]<|/det|> +## Discussion + +<|ref|>sub_title<|/ref|><|det|>[[115, 459, 678, 479]]<|/det|> +## Genetic architecture of MATs in the BTx623 X IS3620C RIL population + +<|ref|>text<|/ref|><|det|>[[111, 495, 884, 850]]<|/det|> +In this study, we describe a powerful, innovative approach for genetic analysis of MATs in food crops that capitalizes on high- throughput AiMS methodology to quantify human gut microbiome phenotypes. Our QTL analysis of MATs in the well- characterized S. bicolor BTx623 X IS3620C RIL population revealed a complex polygenic genetic architecture with ten significant QTLs affecting different combinations of MATs defined by their effects on abundances of specific microbial taxa, concentrations of microbial fermentation products (SCFAs), and variation in seed composition (color and tannin concentration). Several of the QTLs had pleiotropic effects on two or more microbial taxa (Figure 2), with the QTL on chromosome 4 alone affecting the abundances of 8 different microbial taxa in the fermentations, illustrating how genetic variation at a given QTL can drive MATs with broad effects on the gut microbiome. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 876, 900]]<|/det|> +MATs and the genetic architecture driving them in sorghum are likely to be far more complex than the ten QTL identified in our study because genetic diversity of the BTx623 and IS3620C parents represents only a fraction of the total genetic diversity of the Sorghum bicolor species (40, 41). Indeed, genetic analysis of MATs in populations such as the Sorghum Association Panel (41–43), which captures genetic diversity of sorghum from all four of the major subpopulations that appear to represent two independent domestication events, will help provide a more comprehensive picture of MATs in sorghum and the underlying genetics. In addition to greater genetic diversity in sorghum, use of additional human gut microbiomes for AiMS phenotyping in future genetic analyses of MATs will also provide a more comprehensive catalogue of MATs because the ability to detect distinct types of MATs will depend on species composition of the human microbiome used for AiMS phenotypes. However, it is important to point out that even with use of a single human microbiome for AiMS phenotyping, emphasis on MELs with broad effects on multiple microbial taxa allowed us to identify loci where variation can affect diverse human microbiomes. For example, the MEL on chromosome 4 for MATs from the microbiome of subject 765 was linked to variation at the Tan 1 locus. Comparison of AiMS reactions of NILs differing only in alleles at Tan1 and RILs with haplotypes differing at Tan1/Tan2 detected significant effects of Tan 1 variation on the microbiomes of 12 different human subjects. While the overall effects of the Tan1/tan-1-b alleles were unique to each microbiome, a subset of microbial taxa shared across microbiomes from multiple subjects showed similar response patterns to Tan1 variation in both the RILs and the NILs (Figure 4). Thus, while MATs can manifest as effects on microbiome-specific combinations of organisms, they can also display shared effects on the same or similar + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 872, 220]]<|/det|> +organisms across diverse human microbiomes. This principle is informative because it suggests a hierarchical approach using small numbers of human donor microbiomes to identify candidate loci affecting multiple microbes (MELs) followed by validation using larger numbers of donor microbiomes may be an efficient strategy for complex trait analysis of MATs. + +<|ref|>sub_title<|/ref|><|det|>[[111, 274, 787, 295]]<|/det|> +## The Tan1 and Tan2 regulators affect a wide range of traits, including human MATs + +<|ref|>text<|/ref|><|det|>[[111, 310, 870, 850]]<|/det|> +The major regulatory roles of the Tan1 and Tan2 genes in production of tannins, along with the strong associations of allelic variation at Tan1 and Tan2 with MATs in the RILs and Tan1 with MATs in the NILs clearly implicate tannin production as a major MAT. The mechanism through which variation in Tan1 and Tan2 affect the microbiome is still unclear as the regulatory functions of Tan1 and Tan2 are quite pleiotropic. Both regulators required for expression of the entire phenylpropanoid pathway, which is highly conserved among many plant species, and allows plant species to convert phenylalanine and tyrosine to a wide variety of flavones, flavanols, flavonoids, anthocyanins, and proanthocyanins (44–52). Expression of most genes in the phenylpropanoid pathway in S. bicolor are blocked by mutations in either Tan1 or Tan 2, including upstream enzymes of the pathway (Chalcone Synthase and Chalcone isomerase) (37). Thus, it remains unclear if Tan1- and Tan2- driven effects on MATs manifest through effects on synthesis of condensed tannins (end products) or one or more of the intermediates. Moreover, mutations in Tan1 also have pleiotropic effects on the concentration of certain fatty acids (53), further implicating variation in non- tannin molecules as potential drivers of the MAT. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 667]]<|/det|> +In support of tannins themselves as mediators of the MATs driven by variation at Tan1 and Tan2, we note striking parallels between the tannin- associated microbiome phenotypes in our study and tannin- mediated effects on the microbiome in other studies. For example, recent studies using purified tannins in in vitro fermentation assays with human stool samples showed that condensed tannins from tree barks stimulated abundances of members of the Lachnospiraceae, Ruminococcaceae, Christensenellaceae, and Peptostreptococcaceae, and decreased the abundance of Parabacteriodes (54), remarkably like the microbiome signature we observed in the Tan1 +/+ versus Tan1- /- RILs and NILs. Equally striking is the responsiveness of the gut microbiome in swine and poultry, where supplementing feed with high- tannin vs low- tannin sorghum also shows that high- tannin content stimulates members of the Lachnospiraceae, Ruminococcaceae (including Faecalibacterium), and Peptostreptococcaceae in fecal and cecal microbiota (55–57). What’s more, in a crossover randomized human trial, sorghum containing high tannin induced increased abundances of Faecalibacterium prausnitzii (58). Clearly, detailed mechanistic studies will be necessary to determine if the MAT phenotypes driven by alleles at Tan1 and Tan2 are due to effects on tannin production, one or more intermediates in the phenylpropanoid pathway, or other biosynthetic pathways. + +<|ref|>text<|/ref|><|det|>[[111, 681, 884, 888]]<|/det|> +In sorghum, condensed tannins are only produced in the testa layer of the seeds, where along with other polyphenolics, they contribute to the diverse array of seed color phenotypes that can be found across sorghum populations (59–61). Tannin production is uncommon among domesticated grains, being observed in sorghum and finger millet and it is believed that the trait was lost in other major grains during domestication due to selection against tannin- mediated phenotypes such as bitter taste (62, 63). The trait may have been maintained in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 881, 558]]<|/det|> +sorghum, however, because tannin content contributes to agronomically- important traits such resistance to bird predation (36, 53). In Africa, where sorghum is a significant component of the human diet, cultivation of tannin versus non- tannin lines correlates geographically with the intensity of bird feeding pressure and allelic variation at Tan1 and Tan2 are the major drivers of tannin and non- tannin phenotypes in locally cultivated sorghum varieties (36, 64). Remarkably, this same study also demonstrated geographic patterns of allelic variation in human bitter taste receptor genes with locally- cultivated sorghum varieties, where enrichment of alleles that decrease bitter taste perception were found in human populations where high- tannin lines were cultivated due to high- levels of bird predation (36). Our work showing significant effects of tannin on beneficial microorganisms in the gut microbiome further adds to these complex allelochemical interactions and begs the question of whether disproportionate health outcomes associated with effects on beneficial microbes in the human gut microbiome may occur in populations regularly consuming tannin versus non- tannin sorghum. + +<|ref|>text<|/ref|><|det|>[[110, 570, 877, 888]]<|/det|> +Although tannin production can be agronomically- desirable, tannin content of sorghum is known to reduce weight gain in food animals consuming sorghum grain (65, 66). However, the goal of improved weight gain in food animals is dramatically different than the goal of promoting human health. Indeed, in countries with westernized diets, enhanced weight gain in humans would be viewed as undesirable. Moreover, the microorganisms that were stimulated by tannins in our studies and animal studies are members of butyrate- producing Lachnospiraceae and Ruminococcaceae, and some organisms in these families such as Faecalibacterium and Roseburia have well- documented associations with human health benefits (28–35). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 876, 707]]<|/det|> +In addition to tannins, we also detected QTLs on chromosome 2, chromosome 3, and chromosome 5 for distinct MATs and identified potential candidate genes. Among these regions, we observed several annotated candidate genes exhibiting strong seed- specific expression activity and for which one of the parents of the RIL population carries potentially- impactful variation. These candidates included putative cytochrome P450 (Sobic.002G273600), gibberellin 3- beta- dioxygenase (Sobic.003G045900), abscisic acid insensitive3 (ABI3) transcription factor (Sobic.003G398200), and a putative xylanase inhibitor protein precursor (Sobic.005G098700). Cytochrome P450 proteins participate in a wide variety of biosynthesis of lignin, defense compounds, fatty acids, hormones, and signaling molecules (67–69). Even more strikingly, Sobic.003G045900 and Sobic.003G398200 collectively influence synthesis of major plant signaling molecules (GA and ABA) that regulate seed composition, development, and maturation. Gibberellin 3- beta- dioxygenase can alter grain composition, including starch content by its effects on α- amylase (70) and the abscisic acid insensitive3 (ABI3) transcription factor also plays important roles in seed lipid and protein content as well as development (71). Xylanase inhibitor proteins have also been shown to inhibit microbial xylanases, which would influence the ability of xylanase- producing microbes in to degrade arabinoxylans in AiMS reactions. + +<|ref|>text<|/ref|><|det|>[[112, 718, 870, 885]]<|/det|> +Clearly, much future work remains to define the causal variants driving these MATs. However, the unique set candidate genes suggests that genetic variation in sorghum can influence seed composition and fermentation patterns by human gut microbes through a diverse array of pathways and mechanisms. The broad spectrum of potential mechanisms also illustrates the potential of AiMS- based phenotyping and complex trait analysis of MATs to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 877, 257]]<|/det|> +define variation in a wide array of plant molecules with human gut microbiome activity. Thus, complex trait analysis of MATs in food crops provides a new approach for developing a comprehensive catalogue of MATs affecting the human gut microbiome and it paves the way for use of MATs with major effects on beneficial gut microbes as novel traits crop improvement strategies that can have profound outcomes with respect to human health. + +<|ref|>sub_title<|/ref|><|det|>[[115, 312, 309, 330]]<|/det|> +## Materials and Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 350, 212, 367]]<|/det|> +## Germplasm + +<|ref|>text<|/ref|><|det|>[[111, 384, 880, 889]]<|/det|> +A total of 417 F7- 8 RILs derived from a S. bicolor BTx623 by IS3620C cross (27, 72) were grown in the Greenhouse Innovation Center at the University of Nebraska- Lincoln, Lincoln, NE, USA. The greenhouse growout was planted on September \(18^{\text{th}}\) , 2017 and harvested on February \(6^{\text{th}}\) , 2018. The temperature was maintained between \(26.6^{\circ}\text{C}\) and \(27.8^{\circ}\text{C}\) during day light hours and between 21.1 and \(23.3^{\circ}\text{C}\) during night hours with a target relative humidity of \(30\%\) . Supplemental LED lighting was employed to maintain total photosynthesis actively radiation (PAR) at or above \(230 \mu \text{mol} \text{m}^{- 2} \text{s}^{- 1}\) . All plants were watered to field capacity. Heads were bagged to ensure self- pollination of individual RILs. The B Wheatland sorghum line containing tannins was developed through cross- pollination of 'B Wheatland' by 'B SD106', the source of the tannin trait. The resulting F1 progeny were allowed to self- pollinate, and F2 progeny containing tannins were identified and backcrossed to the recurrent parent B Wheatland. The B Tx631 sorghum line was cross- pollinated with two different tannin producing line, 'Waconia' and 'B KS5'. The resulting F1 progenies were allowed to self- pollinate, and F2 progenies were backcrossed to the recurrent parent B Tx631 and B KS5, respectively. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 882, 145]]<|/det|> +tannin trait was identified in homozygous lines in the generation following second round of self- pollination. + +<|ref|>sub_title<|/ref|><|det|>[[113, 164, 256, 182]]<|/det|> +## In vitro digestion + +<|ref|>text<|/ref|><|det|>[[110, 199, 880, 560]]<|/det|> +Two grams of seeds from each line were milled using a high- throughput ball mill (2025 GenoGrinder; SPEX SamplePrep, Metuchen, NJ, USA). Twenty milligrams (+/- 0.5 mg) of each flour per replicate was dispensed into 1 mL- deep wells in 96- well plates using an automated powder dispenser (Flex PowderDose; Chemspeed Technologies AG, Füllinsdorf, Switzerland). The dispensed flour was mixed with 425 ul of water for 15 minutes for complete dispersion and steamed for 20 minutes. Forty- five microliters of 0.5 M HCl + 10% (w/v) pepsin (P7000; Sigma, St. Louis, MO) were added to the samples and incubated at 37 °C for one hour. Then 25 μl of 0.5 M sodium malate buffer (pH = 6, containing 1 mM CaCl2), 40 ul of 0.5 M NaHCO3, 40 μl of 12.5% (w/v) pancreatin (P7545; Sigma, St. Louis, MO) + 4 % (w/v) amyloglucosidase (E- AMGDF, 3,260 U/mL, Megazyme) was added into samples before incubating at 37 °C for six hours. + +<|ref|>text<|/ref|><|det|>[[110, 570, 882, 774]]<|/det|> +After digestion, samples were transferred to 96- well dialysis plates (MWCO 1,000; DispoDialyzer; Harvard Apparatus, Holliston, MA, USA) and dialyzed against 5 gallons of distilled water for 72 hours at 4 °C with freshwater changes at 12 hour intervals. During dialysis, each well was stirred individually with tumble stir bars in each well using a tumble stirrer (VP 710L V&P Scientific, San Diego, CA, USA). Following dialysis, the retentate was transferred into 1 mL- deep wells in 96- well plates and stored in - 80°C until fermentation. + +<|ref|>text<|/ref|><|det|>[[110, 791, 881, 848]]<|/det|> +RILs were grouped by haplotype of markers linked to Tan1 and Tan2 and were randomly selected and pooled within each haplotype group (Tan1Tan2: 41 lines, Tan1tan2- c: 15 lines, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 871, 145]]<|/det|> +tan1- bTan2: 15 lines, tan1- btan2- c: 15 lines) (supplementary Dataset S5). Selected RILS groups and NILs were digested and dialyzed following established procedures (73). + +<|ref|>sub_title<|/ref|><|det|>[[115, 163, 467, 182]]<|/det|> +## Fecal donor and in vitro fecal fermentation + +<|ref|>text<|/ref|><|det|>[[111, 200, 884, 368]]<|/det|> +Fermentation media and stool microbiomes were added to the retentate in an anaerobic chamber. Fermentations were incubated in an anaerobic chamber at \(37^{\circ}C\) for 16 hours. After fermentation, samples were centrifuged to separate supernatants from bacterial pellets to determine microbiome composition and SCFA production. Samples were stored at - 80 \(^\circ \mathrm{C}\) until further processing. + +<|ref|>sub_title<|/ref|><|det|>[[115, 386, 500, 405]]<|/det|> +## DNA extraction and 16S rRNA gene sequencing + +<|ref|>text<|/ref|><|det|>[[111, 421, 856, 552]]<|/det|> +DNA was extracted from the fecal pellets using the BioSprint 96 workstation (Qiagen, Germantown, MD) and the BioSprint 96 one- for- all Vet kit with the addition of buffer ASL (Qiagen, Germantown, MD) and bead beating (74). The V4 region of the bacterial 16S rRNA gene was amplified from each sample using the dual- indexing sequencing strategy (75). + +<|ref|>sub_title<|/ref|><|det|>[[115, 570, 427, 589]]<|/det|> +## 16S rRNA gene sequencing processing + +<|ref|>text<|/ref|><|det|>[[111, 606, 881, 888]]<|/det|> +Paired- end sequences were analyzed using Quantitative Insights Into Microbial Ecology (QIIME) program (version 2) (76). Sequences were truncated (220 bases for forward reads and 160 bases for reverse reads) and denoised into amplicon sequence variants (ASVs) using DADA2 (77). All ASVs were assigned with taxonomic information using pre- fitted sklearn- based taxonomy classifier SILVA database (release 132) (78) and were then binned at genus level and transformed to relative abundance by dividing each value in a sample by the total reads in that sample. A neighbor- joining tree of representative sequence was generated using MUltiple Sequence Comparison by Log- Expectation (MUSCLE). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 91, 277, 108]]<|/det|> +## Seed color analyses + +<|ref|>text<|/ref|><|det|>[[112, 125, 872, 293]]<|/det|> +Seed color analysesSix individual grains from each line of the RIL population were scanned using an EPSON Perfection V600 scanner. Image analysis was conducted using a set of scripts for automatic seed image analysis (https://github.com/alejandropages/SLHTP). The average red, green, blue value from each line was extracted and the first principal component of RGB value was used as trait for QTL analysis. + +<|ref|>sub_title<|/ref|><|det|>[[113, 312, 316, 330]]<|/det|> +## Tannin content analyses + +<|ref|>text<|/ref|><|det|>[[112, 348, 833, 404]]<|/det|> +Tannin content of each line of the RIL population was measured using Vanillin/HCl method (79). + +<|ref|>sub_title<|/ref|><|det|>[[113, 423, 232, 441]]<|/det|> +## SCFA analyses + +<|ref|>text<|/ref|><|det|>[[112, 459, 878, 551]]<|/det|> +SCFA analysesSCFA (acetate, propionate, butyrate and valerate) and branched chain fatty acids (BCFA; iso- butyrate and iso- valerate) from fermentation samples were analyzed by gas chromatography as described previously (80). + +<|ref|>sub_title<|/ref|><|det|>[[113, 570, 476, 589]]<|/det|> +## Genetic map construction and QTL mapping + +<|ref|>text<|/ref|><|det|>[[112, 606, 879, 850]]<|/det|> +Genetic map construction and QTL mappingA genetic map was constructed using data from 616 informative SNP markers from the BTx623 × IS3620C RILs previously reported by Kong et al. (27) and the ASMap package version 1.0- 4 in R (81). A non- parametric interval QTL mapping procedure was employed to identify a single QTL model for each phenotypic trait previously described using the R/qtl package version 1.47- 9 (82). LOD score significance level thresholds were also calculated based on 1,000 permutations of the data with a single QTL genome scan per permutation. Significant QTL and suggestive QTL were then identified using the threshold of \(p < 0.05\) and \(p < 0.1\) , respectively. + +<|ref|>sub_title<|/ref|><|det|>[[113, 867, 266, 884]]<|/det|> +## Statistical analysis + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 875, 295]]<|/det|> +All analyses were performed using R and Rstudio (83, 84). Bacterial community \(\beta\) - diversity in Bray- Curtis distance was calculated using rarefied amplicon sequence variant (ASV) data with the phyloseq and vegan packages (85, 86). Differences in the microbiome communities were compared by PERMANOVA using the Adonis function in vegan. Bacterial genera abundances and SCFA production were compared by Wilcoxon test. Data was visualized using different programs and R packages (87–90). + +<|ref|>sub_title<|/ref|><|det|>[[115, 350, 279, 367]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[111, 384, 884, 775]]<|/det|> +This research was supported by funds from the Jeff and Tricia Raikes Foundation, the Bill and Melinda Gates Foundation, the Don Dillon Foundation to A.K.B., funds from the Hogemeyer Family Foundation and the McConnell fund to J.C.S, and USDA- ARS project 3042- 21220- 033- 00D. Q.Y. was supported in part by scholarship from the China Scholarship Council and N.K. was supported in part by Foundation for Food and Agriculture Research (FFAR) and the FFAR Fellows Program. This work was completed utilizing the Holland Computing Center of the University of Nebraska which receives support from the Nebraska Research Initiative. We thank Alejandro Pages for assistance and methods development in sorghum seed phenotyping, William McQueeney for assistance with AiMS phenotyping, Bryce Askey for assistance in QTL mapping, and Christine Smith, Vicent Stoeger, and Troy Pabst for their assistance in plant care and harvesting. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 875, 179]]<|/det|> +1. J. G. Wallace, E. Rodgers-Melnick, E. S. Buckler, On the road to breeding 4.0: Unraveling the good, the bad, and the boring of crop quantitative genomics. Annu. Rev. Genet. 52, 421–444 (2018). + +<|ref|>text<|/ref|><|det|>[[55, 198, 876, 220]]<|/det|> +2. L. T. Hickey, et al., Breeding crops to feed 10 billion. Nat. Biotechnol. 37, 744–754 (2019). + +<|ref|>text<|/ref|><|det|>[[55, 238, 856, 260]]<|/det|> +3. M. Schreiber, N. Stein, M. Mascher, Genomic approaches for studying crop evolution. + +<|ref|>text<|/ref|><|det|>[[165, 276, 362, 295]]<|/det|> +Genome Biol. 19 (2018). + +<|ref|>text<|/ref|><|det|>[[55, 312, 825, 334]]<|/det|> +4. P. S. Schnable, et al., The B73 maize genome: Complexity, diversity, and dynamics. + +<|ref|>text<|/ref|><|det|>[[165, 350, 479, 370]]<|/det|> +Science (80-. ). 326, 1112–1115 (2009). + +<|ref|>text<|/ref|><|det|>[[55, 387, 852, 409]]<|/det|> +5. C. Feuillet, J. E. Leach, J. Rogers, P. S. Schnable, K. Eversole, Crop genome sequencing: + +<|ref|>text<|/ref|><|det|>[[165, 425, 633, 444]]<|/det|> +Lessons and rationales. Trends Plant Sci. 16, 77–88 (2011). + +<|ref|>text<|/ref|><|det|>[[55, 461, 825, 483]]<|/det|> +6. W. Yang, et al., Crop Phenomics and High-Throughput Phenotyping: Past Decades, + +<|ref|>text<|/ref|><|det|>[[165, 499, 778, 519]]<|/det|> +Current Challenges, and Future Perspectives. Mol. Plant 13, 187–214 (2020). + +<|ref|>text<|/ref|><|det|>[[55, 536, 856, 598]]<|/det|> +7. M. Garg, et al., Biofortified Crops Generated by Breeding, Agronomy, and Transgenic Approaches Are Improving Lives of Millions of People around the World. Front. Nutr. 0, 12 (2018). + +<|ref|>text<|/ref|><|det|>[[165, 614, 847, 635]]<|/det|> +8. A. Saltzman, et al., Availability, production, and consumption of crops biofortified by + +<|ref|>text<|/ref|><|det|>[[165, 652, 859, 673]]<|/det|> +plant breeding: current evidence and future potential. Ann. N. Y. Acad. Sci. 1390, 104– + +<|ref|>text<|/ref|><|det|>[[165, 689, 260, 708]]<|/det|> +114 (2017). + +<|ref|>text<|/ref|><|det|>[[55, 726, 880, 748]]<|/det|> +9. N. Henkhaus, et al., Plant science decadal vision 2020–2030: Reimagining the potential of + +<|ref|>text<|/ref|><|det|>[[165, 764, 755, 784]]<|/det|> +plants for a healthy and sustainable future. Plant Direct 4, e00252 (2020). + +<|ref|>text<|/ref|><|det|>[[55, 800, 592, 821]]<|/det|> +10. WHO, Obesity and overweight (2020) (July 27, 2021). + +<|ref|>text<|/ref|><|det|>[[55, 836, 811, 858]]<|/det|> +11. A. D. Deshpande, M. Harris-Hayes, M. Schootman, Epidemiology of Diabetes and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 88, 707, 110]]<|/det|> +570 Diabetes- Related Complications. Phys. Ther. 88, 1254–1264 (2008). + +<|ref|>text<|/ref|><|det|>[[56, 125, 825, 221]]<|/det|> +571 12. C. Estes, H. Razavi, R. Loomba, Z. Younossi, A. J. Sanyal, Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease. Hepatology 67, 123–133 (2018). + +<|ref|>text<|/ref|><|det|>[[56, 235, 857, 333]]<|/det|> +574 13. S. C. Ng, et al., Worldwide incidence and prevalence of inflammatory bowel disease in the 21st century: a systematic review of population-based studies. Lancet 390, 2769–2778 (2017). + +<|ref|>text<|/ref|><|det|>[[56, 348, 864, 405]]<|/det|> +577 14. A. B. Shreiner, J. Y. Kao, V. B. Young, The gut microbiome in health and in disease. Curr. Opin. Gastroenterol. 31, 69–75 (2015). + +<|ref|>text<|/ref|><|det|>[[56, 420, 877, 477]]<|/det|> +579 15. A. M. Valdes, J. Walter, E. Segal, T. D. Spector, Role of the gut microbiota in nutrition and health. BMJ 361, k2179 (2018). + +<|ref|>text<|/ref|><|det|>[[56, 492, 852, 550]]<|/det|> +581 16. Y. Fan, O. Pedersen, Gut microbiota in human metabolic health and disease. Nat. Rev. Microbiol. 19, 55–71 (2021). + +<|ref|>text<|/ref|><|det|>[[56, 566, 840, 624]]<|/det|> +583 17. R. E. Ley, P. J. Turnbaugh, S. Klein, J. I. Gordon, Human gut microbes associated with obesity. Nature 444, 1022–1023 (2006). + +<|ref|>text<|/ref|><|det|>[[56, 639, 803, 697]]<|/det|> +585 18. V. K. Ridaura, et al., Gut microbiota from twins discordant for obesity modulate metabolism in mice. Science (80-. ). 341 (2013). + +<|ref|>text<|/ref|><|det|>[[56, 712, 880, 770]]<|/det|> +587 19. J. Henao-Mejia, et al., Inflammasome-mediated dysbiosis regulates progression of NAFLD and obesity. Nature 482, 179–185 (2012). + +<|ref|>text<|/ref|><|det|>[[56, 786, 850, 844]]<|/det|> +589 20. T. Le Roy, et al., Intestinal microbiota determines development of non-alcoholic fatty liver disease in mice. Gut 62, 1787–1794 (2013). + +<|ref|>text<|/ref|><|det|>[[56, 860, 852, 881]]<|/det|> +591 21. D. Rothschild, et al., Environment dominates over host genetics in shaping human gut + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[56, 90, 490, 110]]<|/det|> +592 microbiota. Nature 555, 210–215 (2018). + +<|ref|>text<|/ref|><|det|>[[56, 127, 807, 184]]<|/det|> +593 22. R. N. Carmody, et al., Diet Dominates Host Genotype in Shaping the Murine Gut +594 Microbiota. Cell Host Microbe 17, 72–84 (2015). + +<|ref|>text<|/ref|><|det|>[[56, 202, 855, 257]]<|/det|> +595 23. D. Ercolini, V. Fogliano, Food Design to Feed the Human Gut Microbiota. J. Agric. Food +596 Chem. 66, 3754–3758 (2018). + +<|ref|>text<|/ref|><|det|>[[56, 276, 855, 331]]<|/det|> +597 24. R. K. Singh, et al., Influence of diet on the gut microbiome and implications for human +598 health. J. Transl. Med. 15, 1–17 (2017). + +<|ref|>text<|/ref|><|det|>[[56, 350, 870, 406]]<|/det|> +599 25. A. Beam, E. Clinger, L. Hao, Effect of diet and dietary components on the composition of +600 the gut microbiota. Nutrients 13 (2021). + +<|ref|>text<|/ref|><|det|>[[56, 425, 866, 480]]<|/det|> +601 26. J. Evans, et al., Extensive variation in the density and distribution of DNA polymorphism +602 in sorghum genomes. PLoS One 8, e79192 (2013). + +<|ref|>text<|/ref|><|det|>[[56, 499, 845, 590]]<|/det|> +603 27. W. Kong, et al., Genotyping by Sequencing of 393 Sorghum bicolor BTx623 × IS3620C +604 Recombinant Inbred Lines Improves Sensitivity and Resolution of QTL Detection. G3 +605 Genes | Genomes | Genetics 8, 2563–2572 (2018). + +<|ref|>text<|/ref|><|det|>[[56, 609, 818, 664]]<|/det|> +606 28. M. E. Sanders, D. J. Merenstein, G. Reid, G. R. Gibson, R. A. Rastall, Probiotics and +607 prebiotics in intestinal health and disease: from biology to the clinic. Nat. Rev. + +<|ref|>text<|/ref|><|det|>[[56, 683, 523, 700]]<|/det|> +608 Gastroenterol. Hepatol. 16, 605–616 (2019). + +<|ref|>text<|/ref|><|det|>[[56, 720, 844, 775]]<|/det|> +609 29. Z. Tamanai-Shacoori, et al., Roseburia spp.: a marker of health? Future Microbiol. 12, +610 157–170 (2017). + +<|ref|>text<|/ref|><|det|>[[56, 794, 880, 850]]<|/det|> +611 30. V. Sundaram, R. Jalan, Roseburia Species: Prime Candidates for Microbial Therapeutics in +612 Inflammatory Bowel Disease. Gastroenterology 157, 1163–1164 (2019). + +<|ref|>text<|/ref|><|det|>[[56, 869, 880, 885]]<|/det|> +613 31. B. Seo, et al., Roseburia spp. Abundance Associates with Alcohol Consumption in Humans + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 88, 855, 144]]<|/det|> +and Its Administration Ameliorates Alcoholic Fatty Liver in Mice. Cell Host Microbe 27, 25- 40. e6 (2020). + +<|ref|>text<|/ref|><|det|>[[163, 161, 861, 256]]<|/det|> +C. V. Ferreira-Halder, A. V. de S. Faria, S. S. Andrade, Action and function of Faecalibacterium prausnitzii in health and disease. Best Pract. Res. Clin. Gastroenterol. 31, 643-648 (2017). + +<|ref|>text<|/ref|><|det|>[[163, 273, 880, 368]]<|/det|> +R. Martín, et al., Functional Characterization of Novel Faecalibacterium prausnitzii Strains Isolated from Healthy Volunteers: A Step Forward in the Use of F. prausnitzii as a Next-Generation Probiotic. Front. Microbiol. 0, 1226 (2017). + +<|ref|>text<|/ref|><|det|>[[163, 384, 845, 480]]<|/det|> +H. Sokol, et al., Faecalibacterium prausnitzii is an anti-inflammatory commensal bacterium identified by gut microbiota analysis of Crohn disease patients. Proc. Natl. Acad. Sci. U. S. A. 105, 16731-16736 (2008). + +<|ref|>text<|/ref|><|det|>[[163, 495, 850, 551]]<|/det|> +S. Miquel, et al., Faecalibacterium prausnitzii and human intestinal health. Curr. Opin. Microbiol. 16, 255-261 (2013). + +<|ref|>text<|/ref|><|det|>[[163, 568, 815, 625]]<|/det|> +Y. Wu, et al., Allelochemicals targeted to balance competing selections in African agroecosystems. Nat. Plants 5, 1229-1236 (2019). + +<|ref|>text<|/ref|><|det|>[[163, 641, 857, 698]]<|/det|> +Y. Wu, et al., Presence of tannins in sorghum grains is conditioned by different natural alleles of Tannin1. Proc. Natl. Acad. Sci. U. S. A. 109, 10281-10286 (2012). + +<|ref|>text<|/ref|><|det|>[[163, 714, 855, 771]]<|/det|> +D. M. Goodstein, et al., Phytozome: a comparative platform for green plant genomics. Nucleic Acids Res. 40, D1178-D1186 (2012). + +<|ref|>text<|/ref|><|det|>[[163, 788, 872, 880]]<|/det|> +R. F. McCormick, et al., The Sorghum bicolor reference genome: improved assembly, gene annotations, a transcriptome atlas, and signatures of genome organization. Plant J. 93, 338-354 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 88, 883, 149]]<|/det|> +40. M. L. Wang, et al., Genetic diversity and population structure analysis of accessions in the US historic sweet sorghum collection. Theor. Appl. Genet. 120, 13–23 (2009). + +<|ref|>text<|/ref|><|det|>[[60, 165, 816, 222]]<|/det|> +41. G. P. Morris, et al., Population genomic and genome-wide association studies of agroclimatic traits in sorghum. Proc. Natl. Acad. Sci. U. S. A. 110, 453–458 (2013). + +<|ref|>text<|/ref|><|det|>[[60, 239, 820, 296]]<|/det|> +42. A. M. Casa, et al., Community resources and strategies for association mapping in Sorghum. Crop Sci. 48, 30–40 (2008). + +<|ref|>text<|/ref|><|det|>[[60, 312, 797, 405]]<|/det|> +43. J. L. Boatwright, et al., Sorghum Association Panel Whole-Genome Sequencing Establishes Pivotal Resource for Dissecting Genomic Diversity. bioRxiv, 2021.12.22.473950 (2021). + +<|ref|>text<|/ref|><|det|>[[60, 422, 692, 443]]<|/det|> +44. T. Vogt, Phenylpropanoid biosynthesis. Mol. Plant 3, 2–20 (2010). + +<|ref|>text<|/ref|><|det|>[[60, 460, 866, 517]]<|/det|> +45. V. Yadav, et al., Phenylpropanoid pathway engineering: An emerging approach towards plant defense. Pathogens 9, 312 (2020). + +<|ref|>text<|/ref|><|det|>[[60, 534, 866, 666]]<|/det|> +46. E. Habyarimana, M. Dall’Agata, P. De Franceschi, F. S. Baloch, Genome-wide association mapping of total antioxidant capacity, phenols, tannins, and flavonoids in a panel of Sorghum bicolor and S. Bicolor × S. Halepense populations using multi-locus models. PLoS One 14, e0225979 (2019). + +<|ref|>text<|/ref|><|det|>[[60, 682, 864, 740]]<|/det|> +47. A. Baudry, et al., TT2, TT8, and TTG1 synergistically specify the expression of BANYULS and proanthocyanidin biosynthesis in Arabidopsis thaliana. Plant J. 39, 366–380 (2004). + +<|ref|>text<|/ref|><|det|>[[60, 756, 814, 814]]<|/det|> +48. I. Hichri, et al., Recent advances in the transcriptional regulation of the flavonoid biosynthetic pathway. J. Exp. Bot. 62, 2465–2483 (2011). + +<|ref|>text<|/ref|><|det|>[[60, 831, 875, 888]]<|/det|> +49. C. C. Carey, J. T. Strahle, D. A. Selinger, V. L. Chandler, Mutations in the pale aleurone color1 Regulatory Gene of the Zea mays Anthocyanin Pathway Have Distinct Phenotypes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 88, 828, 144]]<|/det|> +Relative to the Functionally Similar Transparent Testa Glabra1 Gene in Arabidopsis thaliana. Plant Cell 16, 450–464 (2004). + +<|ref|>text<|/ref|><|det|>[[163, 162, 755, 220]]<|/det|> +F. He, Q. H. Pan, Y. Shi, C. Q. Duan, Biosynthesis and genetic regulation of proanthocyanidins in plants. Molecules 13, 2674–2703 (2008). + +<|ref|>text<|/ref|><|det|>[[163, 237, 875, 333]]<|/det|> +M. T. Sweeney, M. J. Thomson, B. E. Pfeil, S. McCouch, Caught red-handed: Rc encodes a basic helix-loop-helix protein conditioning red pericarp in rice. Plant Cell 18, 283–294 (2006). + +<|ref|>text<|/ref|><|det|>[[163, 349, 872, 406]]<|/det|> +E. Cavallini, et al., The phenylpropanoid pathway is controlled at different branches by a set of R2R3-MYB C2 repressors in grapevine. Plant Physiol. 167, 1448–1470 (2015). + +<|ref|>text<|/ref|><|det|>[[163, 422, 869, 515]]<|/det|> +P. Xie, et al., Control of Bird Feeding Behavior by Tannin1 through Modulating the Biosynthesis of Polyphenols and Fatty Acid-Derived Volatiles in Sorghum. Mol. Plant 12, 1315–1324 (2019). + +<|ref|>text<|/ref|><|det|>[[163, 532, 870, 589]]<|/det|> +S. Molino, et al., Enrichment of Food With Tannin Extracts Promotes Healthy Changes in the Human Gut Microbiota. Front. Microbiol. 12, 570 (2021). + +<|ref|>text<|/ref|><|det|>[[163, 606, 866, 663]]<|/det|> +Y. Y. Choy, et al., Phenolic metabolites and substantial microbiome changes in pig feces by ingesting grape seed proanthocyanidins. Food Funct. 5, 2298–2308 (2014). + +<|ref|>text<|/ref|><|det|>[[163, 680, 875, 737]]<|/det|> +J. M. Díaz Carrasco, et al., Tannins and Bacitracin Differentially Modulate Gut Microbiota of Broiler Chickens. Biomed Res. Int. 2018, 1–11 (2018). + +<|ref|>text<|/ref|><|det|>[[163, 754, 856, 848]]<|/det|> +G. Brugaletta, et al., Insights into the mode of action of tannin-based feed additives in broiler chickens: looking for connections with the plasma metabolome and caecal microbiota. Ital. J. Anim. Sci. 19, 1349–1362 (2020). + +<|ref|>text<|/ref|><|det|>[[163, 866, 821, 886]]<|/det|> +D. V. Seidel, et al., Sorghum-based dietary intervention enriches Faecalibacterium + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 884, 904]]<|/det|> +prausnitzii in fecal samples of overweight individuals. FASEB J. 27, 1056.12- 1056.12 (2013). 59. J. M. Awika, L. W. Rooney, Sorghum phytochemicals and their potential impact on human health. Phytochemistry 65, 1199- 1221 (2004). 60. Y. Xiong, P. Zhang, R. D. Warner, Z. Fang, Sorghum Grain: From Genotype, Nutrition, and Phenolic Profile to Its Health Benefits and Food Applications. Compr. Rev. Food Sci. Food Saf. 18, 2025- 2046 (2019). 61. L. Dykes, G. C. Peterson, W. L. Rooney, L. W. Rooney, Flavonoid composition of lemon- yellow sorghum genotypes. Food Chem. 128, 173- 179 (2011). 62. L. Dykes, L. W. Rooney, Phenolic compounds in cereal grains and their health benefits. Cereal Foods World 52, 105- 111 (2007). 63. A. Drewnowski, C. Gomez- Carneros, Bitter taste, phytonutrients, and the consumer: A review. Am. J. Clin. Nutr. 72, 1424- 1435 (2000). 64. P. Civaň, When bitter is better. Nat. Plants 5, 1205- 1206 (2019). 65. S. I. Chang, H. L. Fuller, Effect of Tannin Content of Grain Sorghums on Their Feeding Value for Growing Chicks. Poult. Sci. 43, 30- 36 (1964). 66. B. W. Cousins, T. D. Tanksley, D. A. Knabe, T. Zebrowska, Nutrient digestibility and performance of pigs fed sorghums varying in tannin concentration. J. Anim. Sci. 53, 1524- 1537 (1981). 67. M. A. Schuler, D. Werck- Reichhart, Functional Genomics of P450s. Annu. Rev. Plant Biol. 54, 629- 667 (2003). 68. B. Zhang, et al., Structure and function of the cytochrome p450 monooxygenase + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[163, 88, 839, 110]]<|/det|> +cinnamate 4- hydroxylase from sorghum bicolor. Plant Physiol. 183, 957–973 (2020). + +<|ref|>text<|/ref|><|det|>[[163, 127, 874, 220]]<|/det|> +Y. Fang, et al., Cytochrome P450 Superfamily: Evolutionary and Functional Divergence in Sorghum (Sorghum bicolor) Stress Resistance. J. Agric. Food Chem. 69, 10952–10961 (2021). + +<|ref|>text<|/ref|><|det|>[[163, 239, 867, 333]]<|/det|> +N. E. J. Appleford, et al., Decreased shoot stature and grain α-amylase activity following ectopic expression of a gibberellin 2-oxidase gene in transgenic wheat. J. Exp. Bot. 58, 3213–3226 (2007). + +<|ref|>text<|/ref|><|det|>[[163, 351, 840, 406]]<|/det|> +R. Tian, et al., Direct and indirect targets of the arabidopsis seed transcription factor ABSCISIC ACID INSENSITIVE3. Plant J. 103, 1679–1694 (2020). + +<|ref|>text<|/ref|><|det|>[[163, 424, 850, 479]]<|/det|> +G. B. Burow, et al., Registration of the BTx623/IS3620C Recombinant Inbred Mapping Population of Sorghum. J. Plant Regist. 5, 141–145 (2011). + +<|ref|>text<|/ref|><|det|>[[163, 497, 866, 620]]<|/det|> +Q. Yang, et al., Near isogenic lines (NIL) of sorghum carrying wild type or waxy alleles of the granule-bound starch synthase (GBSS) gene have distinct effects on human gut microbiome phenotypes and host physiological characteristics (2022) https://doi.org/10.21203/RS.3.RS-1405055/V1 (March 18, 2022). + +<|ref|>text<|/ref|><|det|>[[163, 645, 856, 737]]<|/det|> +A. K. Benson, et al., Microbial Successions Are Associated with Changes in Chemical Profiles of a Model Refrigerated Fresh Pork Sausage during an 80-Day Shelf Life Study. Appl. Environ. Microbiol. 80, 5178–5194 (2014). + +<|ref|>text<|/ref|><|det|>[[163, 757, 870, 888]]<|/det|> +J. J. Kozich, S. L. Westcott, N. T. Baxter, S. K. Highlander, P. D. Schloss, Development of a Dual-Index Sequencing Strategy and Curation Pipeline for Analyzing Amplicon Sequence Data on the MiSeq Illumina Sequencing Platform. Appl. Environ. Microbiol. 79, 5112–5120 (2013). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 884, 889]]<|/det|> +76. E. Bolyen, et al., Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat. Biotechnol. 37, 852–857 (2019). +77. B. J. Callahan, et al., DADA2: High-resolution sample inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016). +78. F. Pedregosa FABIANPEDREGOSA, et al., “Scikit-learn: Machine Learning in Python” (2011). +79. M. L. Price, S. Van Scoyoc, L. G. Butler, A Critical Evaluation of the Vanillin Reaction as an Assay for Tannin in Sorghum Grain. J. Agric. Food Chem. 26, 1214–1218 (1978). +80. J. Yang, et al., Disparate Metabolic Responses in Mice Fed a High-Fat Diet Supplemented with Maize-Derived Non-Digestible Feruloylated Oligo- and Polysaccharides Are Linked to Changes in the Gut Microbiota. PLoS One 11, e0146144 (2016). +81. J. Taylor, ASMap: Linkage Map Construction using the MSTmap Algorithm version 1.0-4 from CRAN (September 15, 2021). +82. K. W. Broman, H. Wu, R/qtl software for mapping quantitative trait loci (2019) (September 15, 2021). +83. R Core Team, R: A language and environment for statistical computing (2021) (May 17, 2021). +84. RStudio Team, RStudio: Integrated Development Environment for R (May 17, 2021). +85. J. Oksanen, et al., vegan: Community Ecology Package (2020) (May 17, 2021). +86. P. J. McMurdie, S. Holmes, phyloseq: An R Package for Reproducible Interactive Analysis and Graphics of Microbiome Census Data. PLoS One 8, e61217 (2013). +87. Hadley Wickham, ggplot2: Elegant Graphics for Data Analysis (2016) (May 17, 2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 870, 110]]<|/det|> +88. A. Kassambara, ggpubr: "ggplot2" Based Publication Ready Plots (2020) (May 17, 2021). + +<|ref|>text<|/ref|><|det|>[[56, 126, 825, 184]]<|/det|> +89. Z. Gu, R. Eils, M. Schlesner, Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847-2849 (2016). + +<|ref|>text<|/ref|><|det|>[[56, 200, 870, 258]]<|/det|> +90. I. Letunic, P. Bork, Interactive Tree Of Life (iTOL) v5: an online tool for phylogenetic tree display and annotation. Nucleic Acids Res. 49, W293-W296 (2021). + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[120, 102, 875, 640]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[55, 91, 404, 107]]<|/det|> +753 Table 1. Major effect loci for MATs. + +
MATsChromoso meMarker Position (Mb)Confidence Interval (Mb)
Ruminococcusaceae_RuminococcusaceaeUCG.0 0227.937.01-12.79
Peptostreptococcaceae_Paeniclostridium29.694.83-77.22
Christensenellaceae_ChristensenellaceaeR. 7group29.697.93-9.69
Prevotellaceae_Paraprevotella265.5463.91-69.84
Butyrate265.6965.54-67.24
Valeric265.6965865.54-67.24
Lachnospiraceae_Dorea34.043.81-7.46
Lachnospiraceae_Coproccous334.043.81-72.46
Lachnospiraceae_Roseburia371.514.04-73.24
Christensenellaceae_ChristensenellaceaeR. 7group371.5171.51-72.69
Peptostreptococcaceae_Paeniclostridium459.4658.94-60.16
Prevotellaceae_Paraprevotella461.1660.49-61.88
Christensenellaceae_ChristensenellaceaeR. 7group461.360.84-61.88
Ruminococcusaceae_RuminococcusaceaeUCG.0 02461.5660.84-62.37
Lachnospiraceae_Coproccous1461.8861.16-62.4
Lachnospiraceae_Roseburia461.8848.57-62.4
Ruminococcusaceae_Faecalibacterium461.8861.16-62.4
Erysipelotrichaceae_Catenibacterium461.8859.46-62.37
Valeric5137.45-49.41
Lachnospiraceae_Coproccous3547.697.45-54.78
Lachnospiraceae_Blauria547.6913-55.7
Lachnospiraceae_Dorea547.6913-55.7
Butyrate547.6913-55.7
Propionate554.7852.23-55.7
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 128, 861, 300]]<|/det|> +Figure 1. Shifts in the abundances of bacterial taxa between IS3620C and BTx623 sorghum. The neighbor- joining tree was developed from representative sequences of each bacterial genus using MUSCLE. Bacterial genus from same family were shaded in same color. A heatmap of the mean log2- transformed fold change of genera that showed significant effects of IS3620C sorghum relative to BTx623 sorghum in each subject is shown to the right of each taxon. Statistical significance of changes between IS3620C and BTx623 sorghum were determined by Wilcoxon test; \(q < 0.05\) considered significant and denoted by asterisk. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 128, 880, 352]]<|/det|> +Figure 2. QTL analysis of the RIL population using the S765 microbiome. (A) LOD score profiles for microbial taxa showing significant QTLs in the RIL population. The circular plots illustrate each chromosome and positions are marked around the outer track in Centimorgans. The individual tracks (A- l) depict the LOD plots of individual genera belonging to the same microbial family or SCFA that are color coded with the corresponding key shown at the bottom of the figure. LOD values are indicated on the Y- axis by chromosome for each LOD plot. The MELs on Chr 2, Chr 3, Chr 4, and Chr 5 are marked by red triangles corresponding to the QTL peaks on the outside of the diagram. (B) Tissue- specific gene expression data for sorghum genes within the MELs on Chr 2, Chr 3, Chr 4, and Chr 5 that are highly expressed in seed (normalized FPKM > 5) and with sequence variation between the two parental lines classified as having moderate or high impact on gene function. Gene structure of Tan1 and Tan2 and the causal polymorphisms in two parental lines. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 117, 877, 470]]<|/det|> +Figure 3. Tan1 and Tan2 determine the seed color, tannin, and microbiome phenotypes in the RILs. (A) LOD score profiles for RGBPC1, tannin content and the abundance of Faecalibacterium and Christensenellaceae R7 group in the RIL population. The significance level was determined by 1000 times permutation test. (B) Each RIL was assigned to one of four genotypic categories based on haplotypes of markers closely linked to the Tan 1 and Tan 2 loci. The four genotypic categories correspond to tannin-negative categories (tan 1- b/Tan2; Tan 1/tan- 2- c; tan- 1- b/tan- 2- c) and the tannin positive category (Tan1/Tan2). Pictures of whole plant, whole grain and color of powders from seed of a single line in each category are illustrated. (C) Box and whisker plots are shown for phenotypic values of tannin concentration, seed color, and individual microbial taxa (Faecalibacterium, Christensenellaceae_R7 group, Roseburia, and Coprococcus 1) corresponding to RILs in each of the genotypic categories. Different letters (a, b and c) indicate significant difference among different genotypic categories. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 135, 881, 405]]<|/det|> +Figure4. Fermentation profile of RILs and NILs differing in tannin production across microbiomes from multiple human subjects. (A) Heatmap of the mean log2- transformed fold change of genera that showed significant effects of tannin positive category (Tan1/Tan2) relative to tannin- negative categories (tan 1- b/Tan2; Tan 1/tan- 2- c; tan- 1- b/tan- 2- c) in RILs. (B) Heatmap of the mean log2- transformed fold change of genera that showed significant effects of tannin sorghum relative to non- tannin sorghum in NILs. (C) Correlation of the average log2 fold change in each genus with significant change between tannin positive and tannin negative genotype from RILs versus average log2 fold change for each genus between tannin positive and tannin negative NILs across 12 subjects. (Pearson correlation coefficient and p values were showed on the figure). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[187, 127, 646, 380]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 437, 115, 456]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[44, 479, 758, 500]]<|/det|> +Shifts in the abundances of bacterial taxa between IS3620C and BTx623 sorghum. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 50, 778, 451]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 819]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[44, 842, 593, 862]]<|/det|> +QTL analysis of the RIL population using the S765 microbiome. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 60, 711, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 116, 819]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[44, 841, 809, 862]]<|/det|> +Tan1 and Tan2 determine the seed color, tannin, and microbiome phenotypes in the RILs. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[80, 50, 700, 630]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 117, 818]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 930, 884]]<|/det|> +Fermentation profile of RILs and NILs differing in tannin production across microbiomes from multiple human subjects. + +<|ref|>sub_title<|/ref|><|det|>[[44, 907, 311, 933]]<|/det|> +## Supplementary Files + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 765, 65]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 83, 325, 210]]<|/det|> +SupplementaryDataset.xlsx FigureS1. pdf FigureS2. pdf FigureS3. pdf SupportingInformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14_det.mmd b/preprint/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..cfca29671b9aeecec06b8a62661032be58dbd6e3 --- /dev/null +++ b/preprint/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14/preprint__16297f08d4bee87d91d82e52bb78067c2240415f69e31ec84422d74b9d45fb14_det.mmd @@ -0,0 +1,683 @@ +<|ref|>title<|/ref|><|det|>[[42, 108, 890, 209]]<|/det|> +# A 1150-year-long AMV reconstruction suggests early warning for a North Atlantic climate tipping point + +<|ref|>text<|/ref|><|det|>[[44, 230, 660, 272]]<|/det|> +Simon Michel ( \(\square\) simon.michel@hotmail.fr) Environnements et Paleoenvironnements Océaniques et Continentaux + +<|ref|>text<|/ref|><|det|>[[44, 277, 645, 319]]<|/det|> +Didier Swingedouw EPOC, Universite Bordeaux https://orcid.org/0000- 0002- 0583- 0850 + +<|ref|>text<|/ref|><|det|>[[44, 324, 680, 365]]<|/det|> +Juliette Mignot Sorbonne université (LOCEAN) https://orcid.org/0000- 0002- 4894- 898X + +<|ref|>text<|/ref|><|det|>[[44, 370, 321, 410]]<|/det|> +Guillaume Gastineau Sorbonne université (LOCEAN) + +<|ref|>text<|/ref|><|det|>[[44, 416, 426, 457]]<|/det|> +Pablo Ortega Universitat Politecnica de Catalunya (BSC) + +<|ref|>text<|/ref|><|det|>[[44, 463, 321, 503]]<|/det|> +Myriam Khodri Sorbonne université (LOCEAN) + +<|ref|>text<|/ref|><|det|>[[44, 509, 592, 550]]<|/det|> +Gerard McCarthy Maynooth University https://orcid.org/0000- 0002- 2363- 0561 + +<|ref|>sub_title<|/ref|><|det|>[[44, 591, 101, 608]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 628, 728, 648]]<|/det|> +Keywords: Atlantic Multidecadal Variability (AMV), climate change, ecosystems + +<|ref|>text<|/ref|><|det|>[[44, 667, 288, 686]]<|/det|> +Posted Date: June 7th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 705, 462, 724]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 588180/v1 + +<|ref|>text<|/ref|><|det|>[[42, 742, 909, 784]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 820, 955, 863]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on September 2nd, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32704- 3. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[78, 85, 880, 152]]<|/det|> +# A 1150-year-long AMV reconstruction suggests early warning for a North Atlantic climate tipping point + +<|ref|>text<|/ref|><|det|>[[70, 170, 885, 640]]<|/det|> +3 Simon Michel \(^{1,5*}\) , Didier Swingedouw \(^{1}\) , Juliette Mignot \(^{2}\) , Guillaume Gastineau \(^{2}\) , Pablo Ortega \(^{3}\) , Myriam Khodri \(^{2}\) , Gerard McCarthy \(^{4}\) 6 7 1 EPOC, Université de Bordeaux, Allée Geoffroy Saint- Hilaire, Pessac 33615, France. 8 2 LOCEAN, Sorbonne université (UPMC, Univ. Paris 06)- CNRS- IRD- MNHN, 4 place Jussieu, 9 75005 Paris, France 10 3 Barcelona Supercomputing Center (BSC- CNS), Edificio NEXUS I, Campus Nord UPC, Grand 11 Capitan, 2- 4, 08034, Barcelona, Spain. 12 4 Irish Climate Analysis and Research UnitS (ICARUS), Department of Geography, Maynooth 13 University, Maynooth, Ireland. 14 5 Institute of Marine and Atmospheric research Utrecht (IMAU), Department of Physics, 15 Utrecht University, Utrecht, Netherlands. 16 \*s.l.l.michel@uu.nl + +<|ref|>text<|/ref|><|det|>[[113, 694, 883, 885]]<|/det|> +The Atlantic Multidecadal Variability (AMV) is a large- scale climate phenomenon with crucial impacts on human societies and ecosystems. Its periodicity and drivers are controversial due to the short temporal extent of instrumental observations and competing impacts of external forcing and internal variability. Here, we use a well- verified set of paleoclimate proxy records and compare four regression methods to perform different reconstructions of the AMV since 850 C.E., built to only reflect North Atlantic internal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 83, 883, 312]]<|/det|> +variability. The best performing reconstruction, when verified both against climate model outputs and independent proxy records is obtained using the non- linear random forest method. It exhibits large multi- decadal variations in the range of 20- 90 years, a broader range than the 50- 70 years identified in instrumental records. The reconstruction shows that AMV autocorrelation properties have experienced significant changes in the recent decades, suggesting an early warning signal for the proximity of a tipping point in the Atlantic. + +<|ref|>sub_title<|/ref|><|det|>[[118, 364, 266, 384]]<|/det|> +## Introduction: + +<|ref|>text<|/ref|><|det|>[[113, 399, 883, 910]]<|/det|> +Since the beginning of the 20th century, the North Atlantic region has exhibited successive decades of anomalously warm and cold sea surface temperatures1 (SST) relative to the global average, subsequently contributing to amplify or damp the global warming effects in the Atlantic sector2,3. The underlying variability mode, the Atlantic Multidecadal Variability (AMV), is related to diverse climatic effects in the North Atlantic neighboring regions4,5 (Fig. 1). At the global scale, it also influences drought and rainfall in the Sahel6,7, Northeastern Brazil7 and Central Asia7 (Fig. 1); Atlantic hurricane frequency and intensity8,9; sea ice thickness and extent over the Arctic10; and is linked to Pacific climate variability11. The processes driving the AMV and its resulting spectral properties remain a source of controversy. Disagreements partly come from the relatively short period over which the AMV is directly observed, which mainly encompasses climate changes significantly affected by human activity12. Internal ocean variability has been identified in climate models as a driver of the AMV pattern through variations in the Atlantic Meridional Overturning Circulation13,14 (AMOC), a large- scale circulation that is known to be potentially unstable in the long- term projections of future climate15,16. These results, based on coupled ocean- atmosphere general circulation models, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 80, 883, 905]]<|/det|> +have been further supported by a tide gauge- based reconstruction of the ocean circulation intensity in the North Atlantic intergyre region that exhibits strong correlations with the AMV index over the last 60 years17. Several long paleoclimate records also indicate the existence of such multidecadal variability on longer time scales18,19. Nevertheless, it has also been shown that many features of the observed AMV can be reproduced in slab-ocean models by the integration in the ocean of the stochastic forcing of the North Atlantic Oscillation, thus without any influence of ocean dynamics20. It has also been highlighted that periods of cold AMV are dominated by both strong anthropogenic aerosol emissions from Europe and North America and strong volcanic activity21. Notably, a recent study22 has shown that no consistent multidecadal variability in the global mean surface temperature (GMST) is found in control simulations of 16 models contributing to the Climate Model Intercomparison Project phase 6 (CMIP6), as compared to last millennium experiments including estimations of volcanic and solar forcing as boundary conditions. This study argues that the incoming solar radiation has varied in a 50 to 70 years band due to volcanic aerosols during the last millennium. Their conclusion is that these spectral peaks in last millennium natural forcings can solely explain the simulated basin-wide multidecadal variability of North Atlantic SSTs. However, multidecadal variability is believed to be inherent to the North Atlantic, while this former study focused on the GMST, which can also include important contributions from interannual variability modes, like El-Niño Southern Oscillation. In addition, GCMs might be overly sensitive to volcanic forcing, as in proxy-based climate reconstructions of the largest eruptions are generally associated with weaker temperature responses than in the models23. Moreover, a study of the AMV over the last 8000 years, investigating a set of ice core and marine records, has shown that its 55- 70 years timescale of variability were largely driven by internal ocean-atmosphere variability and poorly influenced by natural forcings for a large + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 242]]<|/det|> +part of the Holocene24. The coming years of AMV observations could be very informative in this respect, as recent observations indicate the likely onset of a negative phase25. The study of this potential transition to a new negative phase of the AMV could be very fruitful in understanding the relative roles of natural forcings and internal variability behind its fluctuations. + +<|ref|>text<|/ref|><|det|>[[115, 258, 883, 520]]<|/det|> +In terms of spectral properties, there are notable discrepancies between the observed periodicity in AMV (50- 70 years) and the periodicity found in most of the preindustrial control experiments contributing to the CMIP5 (10- 30 years)26. Disentangling the relative contributions of the AMV driving factors to its spatial and spectral properties in observations could therefore be useful in assessing models and eventually developing emergent constraint approaches to reduce uncertainty in decadal climate predictions27. To reach this long- term goal, an improved knowledge of internal variability of North Atlantic Sea Surface Temperature (NASST) is crucial. + +<|ref|>text<|/ref|><|det|>[[115, 536, 883, 904]]<|/det|> +The AMV index definition is of paramount importance to tackle correctly the lingering questions concerning its drivers. Since climate is currently in a period with a dominant anthropogenically- induced warming signal, several methods are used to isolate the internal variability inherent to the North Atlantic basin. In this study, we use three yearly AMV indices in which all the externally forced signals (including anthropogenic forcing, solar variations, and volcanic eruptions) in annual NASST have been removed with different techniques, to thus explore the sensitivity to the method. We denote these indices as \(AMV_{TS}^8\) , \(AMV_T^2\) and \(AMV_F^{28}\) (see Methods and Fig. 1a), where the subscripts TS, T, and F refer to the associated references2,8,28. They all have the advantage of removing a large part of the externally forced variability from greenhouse gases and aerosols (see Methods) as opposed to detrending methods9,29, which only remove the effect of anthropogenic greenhouse gases emissions28. A + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 207]]<|/det|> +regression analysis between the mean of the three AMV indices with surface temperatures and precipitation observations from the instrumental dataset CRUTS430, shows significant and widespread links of the AMV with both variables, in particular over the continents (Fig. 1b- k), consistently with early studies4- 7. + +<|ref|>text<|/ref|><|det|>[[113, 222, 883, 905]]<|/det|> +The last millennium is well- covered in space by high- resolution proxy records describing local or regional precipitation or temperature variations, making it very suitable for producing annually- resolved paleoclimate reconstructions. It is of particular interest since it encompasses two contrasting periods in terms of global temperature: the warm Medieval Climate Anomaly and the cold Little Ice Age. The relative contribution of external forcing and internal climate variability to the transition between both periods is unclear, as well as their temporal evolution and spatial extent31- 33. Taking advantage of paleoclimate data covering the last millennium, there have been some attempts to reconstruct AMV indices. Previous studies18,34 usually reconstructed AMV over the last millennium using a Principal Components Regression (PCR) method and proxy records from continental borders of the North Atlantic. However, their statistical models were calibrated using NASST anomalies18,34, which include a strong forced signal over the historical period. Indeed, the NASST variability during the last millennium might have different links with global climate, as internal variability could be more dominant, which could affect the resulting reconstruction. In addition, it is worth noting that building a statistical model in which the predictand has a strong trend is not recommended in our case because trends in the predictors can easily match by chance the one in the predictand, leading to the selection of spurious predictors that compromise the quality of the reconstruction. These former reconstructions of the AMV are furthermore based on a single regression method18,34 (PCR, cf. Methods), and do not include any objective criterion for the selection of proxy records, meaning that some of them may have potentially little correlation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 346]]<|/det|> +with the target AMV index over the instrumental period. They also do not include any pseudoproxy validation of the reconstruction as a means to verify that the assumed climatic relationships between the AMV and the proxies are sound. To circumvent most of the previous limitations the present study combines multiple proxies with different regression methods to reconstruct the AMV for the past 1150 years, using also two validation approaches to evaluate the robustness of the reconstruction, the first based on the comparison with independent ocean proxy records and the second on the use of pseudoproxy experiments. + +<|ref|>sub_title<|/ref|><|det|>[[118, 397, 214, 417]]<|/det|> +## Results: + +<|ref|>sub_title<|/ref|><|det|>[[118, 435, 700, 454]]<|/det|> +## An AMV reconstruction based on optimal regression model approach + +<|ref|>text<|/ref|><|det|>[[115, 468, 883, 907]]<|/det|> +The availability of large numbers of annually- resolved proxy records sensitive to temperature and precipitation provides a key opportunity to test and develop new objective reconstruction methodologies using advanced techniques. The different reconstructions produced in this study use proxy records significantly correlated at the 95% confidence level with the target AMV indices (cf. Fig. 1). The selection of these proxies is made from a large set of Northern Hemisphere and annually- resolved proxy records that includes the PAGES 2k database35 and 41 other records published elsewhere (Extended Data Table 1). This database (hereafter P2k+) has been constructed using different quality criteria (see Methods) and comprises a total of 457 records. An important preprocessing step before selecting the proxies is to remove in each proxy record from P2k+ (see Methods) the forced variability of NASST, using linear regression starting in 1870. This is a novel step that allows us to produce an AMV reconstruction that exclusively reflects the internal variability signals. The forced component is estimated using a signal- to- noise maximizing empirical orthogonal function + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 136]]<|/det|> +technique2 (Methods, Extended Data Fig. 1) from historical simulations of 37 CMIP5 models (Extended Data Table 2). + +<|ref|>text<|/ref|><|det|>[[115, 152, 882, 348]]<|/det|> +Reconstruction methods applied in this study require that the predictors (i.e., the proxy records) have no temporal gaps. To use the maximum number of proxies available to reconstruct each temporal window we follow a nested reconstruction approach34 with moving windows of one year. We set the reconstructions to cover the period 850 C.E. to present. This will allow us to estimate the accuracy and reliability of the method through the use of pseudo- proxy experiments37 (PPE). + +<|ref|>text<|/ref|><|det|>[[115, 362, 882, 905]]<|/det|> +The specific methodology to be used for the final nested reconstruction will be determined in a preliminary phase in which a total of 312 reconstructions covering the whole temporal extension (back to 850 C.E.) will be produced, by combining the three AMV definitions, 26 temporal windows (only differing in the last year covered by the proxies) and four regression methods37: PCR, partial least squares, elastic- net and random forest (detailed in Supplementary Information). We use the ClimIndRec reconstruction toolbox38 to generate these reconstructions (84 for each AMV index, cf. Methods). This tool is dedicated to quickly produce reconstructions with up to four regression techniques, where each of their specific control parameters are automatically optimized with cross- validation, and the reliability of the final reconstruction is directly estimated using the Coefficient of Efficiency (CE) metric39 (Methods) over training and testing samples. The CE metric is defined between \(- \infty\) and 1. As the CE is positive for a given testing sample, it indicates that the statistical model gives better estimations than the empirical average of the corresponding training sample, and its use for reconstruction purposes can be considered39. In this study we set that the 312 statistical models are preliminarily evaluated with the CE metric for 30 pairs of training and testing samples. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 882, 313]]<|/det|> +CE scores of the 312 reconstructions for each level of inputs (Methods, Extended Data Fig. 2) show that the random forest40 (RF), despite not necessarily being the best method for all the reconstructions produced, does provide the highest averaged score of all reconstructions. The latter is obtained for the \(AMV_F\) index, for the reconstruction period 850- 1987. For this specific reconstruction, the average of CE scores over the 30 training/testing splits is positive at the 99% confidence level, which validates its use for reconstruction purposes ( \(CE \in [- 0.07, 0.52]\) , \(med(CE) = 0.25\) , \(avg(CE) = 0.23\) , cf. Methods). + +<|ref|>text<|/ref|><|det|>[[115, 328, 883, 696]]<|/det|> +Using the previous setup yielding the best CE, we perform 1020 nested reconstructions of the \(AMV_F\) index based on RF models, applying one- year increments to the initial year of the reconstruction, thus going from the longest time window (i.e., 850- 1987) to the shortest (i.e., 1869- 1987). For each of them, only proxy records covering the whole window and significantly correlated with the observed index at the 95% confidence level are used. The final nested reconstruction is thus obtained by averaging the 1020 interlocking reconstructions over their common timesteps (Fig. 2a). Evaluating each of these reconstructions with CE scores for 30 training/testing splits, we find that the reconstruction has varying validation scores over time, although they overall slightly increase with generally higher scores for the shortest (most recent) reconstructions which are based on higher numbers of proxy records (Fig. 2b). + +<|ref|>text<|/ref|><|det|>[[115, 711, 882, 902]]<|/det|> +The nested reconstruction uses a total of 55 Northern Hemisphere proxy records. Their a posteriori weights, given by random forest importance (cf. Methods), and their temporal availability, are presented in Fig. 2c- d. We identify three main clusters of records with fairly distributed weights: central Asia, Europe and western North America. It is worth noting that proxies from Asia and western North America are highly represented in the PAGES 2k database35, which basically explains their relatively large number used for the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 78, 883, 905]]<|/det|> +reconstruction. Interestingly, there are a number of highly weighted proxy records of annual and boreal summer (June- July- August) temperatures over the Eastern Pakistan/Tibetan Plateau. This link has been supported in a couple of recent studies which have highlighted the role of AMV variations for spring and summer temperatures in this region41,42, notably by affecting large scale pressure gradients in the Eurasian sector. We also use a large number of proxy records from eastern Asia/northern China, for which climate conditions have also been shown to be significantly affected by the AMV variations through atmospheric Rossby wave propagation and altered heat advection in the western Pacific43. Given the large- scale AMV teleconnections shown by these studies41- 43 and the large weights we find for Asian proxy records, we justify their inclusion in our new AMV reconstruction, the first one to date to include them18,34. Western North American proxy records are mostly sensitive to summer and yearly variations of temperature and precipitation (Fig. 2c- d, Extended Data Table 3). It appears that Fig. 1 does not show a consistent relationship between the AMV and instrumental summer and annual temperatures in this region, although it is observed for some precipitation time series over the historical period (Fig. 1). Finally, the fact that only five proxy records from Europe are used is mostly due to their relatively reduced presence in the proxy record database we use (<10%). However, we find that one of them has a relatively large weight for the reconstruction (>7%) and covers the entire reconstruction period. It corresponds to a time series of tree rings growth measurements from European Alps44, which is strongly correlated with summer temperature over the historical period (r=0.7, p<0.01, Extended Data Table 3). The four other European proxy records are related to either summer or annual temperature and precipitation. The selection of these records for our reconstruction is thus highly consistent with the well- documented fingerprint of the AMV on European summer temperatures4,5, as also shown in Fig. 1. A detailed description of each + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 207]]<|/det|> +proxy record used for our reconstruction and their correlations with instrumental data and AMV is given in Extended Data Table 3. Additionally, the same maps of the AMV fingerprints from Fig. 1 including the corresponding selected proxy records for various seasons and climate variables are shown in Extended Data Fig. 3. + +<|ref|>sub_title<|/ref|><|det|>[[118, 225, 408, 242]]<|/det|> +## Data and model-based validations + +<|ref|>text<|/ref|><|det|>[[115, 255, 883, 875]]<|/det|> +As for the previous reconstructions18,34, most of the proxy records used in this study are terrestrial, which might be surprising given that the AMV is an oceanic mode. Our method indeed selected a poor amount of ocean proxies from P2k+ (1 out of 55), essentially because those with a resolution lower than annual have been excluded (Methods). To evaluate if our AMV reconstruction is in line with existing ocean proxy records, we consider a posteriori those not used in the reconstruction. Since some of these proxies have very low temporal resolution, which might lead to spurious significant correlations with the AMV, we use a significance test that takes the time series autocorrelations into account, similarly to previous studies17,38 (see Methods). We find that 37 ocean records (23 from the North Atlantic including the Mediterranean Sea) from the Ocean 2k database35 are significantly correlated at least at the 90% confidence level with our AMV reconstruction (Fig. 3a). As an additional validation, we compute two composite time series of these coral and sediment- based proxy data (see methods). The first one is an average- based aggregate of the 37 ocean proxies from Fig. 3a, and the second one is only based on the 23 North Atlantic proxies from Fig. 3a (Fig. 3b). Since these composites are mostly based on low resolution sediment data, they do not capture well the annual or decadal climate variations as our reconstruction does. However, in terms of low- frequency, we find strong and significant correlations between the 30- year filtered AMV reconstruction (mostly based on terrestrial records as explained above, Fig. 2a- + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 138]]<|/det|> +d), and these composite time series based on all ocean proxies from Fig. 3a (r=0.75, p<0.01, Fig. 3b) and only those from the North Atlantic (r=0.76 p<0.01, Fig. 3b). + +<|ref|>text<|/ref|><|det|>[[115, 188, 881, 310]]<|/det|> +A complementary validation of the physical consistency of the reconstruction is performed with PPE37, using 12 members from the Last Millennium Large Ensemble of the Community Earth System Model 1 (CESM1- LME, see Methods). Here, we conduct two types of PPE experiments for each CESM1- LME member (see Methods for detailed explanations). + +<|ref|>text<|/ref|><|det|>[[115, 327, 882, 870]]<|/det|> +The first PPE experiment consists in exactly reproducing the reconstruction from the real- world experiment (RWE) in the model simulations by only training statistical models over the same period (1870- 1987), with the timeseries of the nearest grid points to the real- world proxy record for each time frame of the nested reconstruction. The validation then consists in comparing the reconstructed last millennium AMV in the model using the RWE methodological setup (see previous section) and the \(AMV_F\) effectively simulated within the model (hereafter, the model AMV). For each member, we calculate CE scores, as well as the correlation between the model AMV and the AMV reconstructed from pseudo- proxies (Fig. 4a). All correlations between the reconstructed AMV and the corresponding model AMV are significant at least at the 90% confidence level for the 10- year smoothed time series (r∈[0.41,0.57] for the 12 members). In terms of skill scores we find that the real- world median skill score (med(CE)=0.29 for the whole nested reconstruction) falls within the range of those from the PPE, which are significantly positive at the 95% confidence for all the members except number 7 (med(CE)∈[0,0.6]). This first PPE validation, based on an ensemble of 12 last millennium simulations, therefore provides further confidence in our reconstruction and constitutes the first model- based validation for an AMV reconstruction. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 883, 736]]<|/det|> +The second experiment is adding constraints from the model for the reconstruction of the AMV using real proxy records. It consists in first subselecting pseudo-proxies according to the correlation between the model AMV and the 55 pseudo-proxies. The RF models are trained within the model simulations over the longest period covered by each proxy in the RWE, and then applied to the real-world proxies (see methods). This method is similar to the PPE performed in a previous NAO reconstruction study based on the PCR method45. The produced AMV reconstructions, based on AMV/pseudo-proxies relationships over long time frames within simulation members, are then compared to the AMV reconstruction from the RWE. The RF models trained over LME simulations applied to the values of the real proxy records lead to very similar reconstructions than those given by the RWE with a significance of at least 95% for the 12 members (r∈[0.53,0.88], Fig. 5a). CE skill scores are also significantly positive in the 12 members for this PPE (med(CE)∈[0.2,0.35], p<0.01 for the 12 members, Fig. 5a), which indicates a satisfactory level of robustness for the RF models trained in the CESM1- LME members. The correlation of the ensemble mean of model-based AMV reconstructions with the RWE reconstruction is also highly significant (r=0.88, p<0.01). This second PPE indicates that training RF models within the millennial-long simulations of the CESM1-LME, with respect to proxies temporal availability in the RWE (Fig 2d, see Methods), reproduces a very similar reconstruction as the RWE when they are applied to the real values of proxy records. + +<|ref|>text<|/ref|><|det|>[[115, 745, 881, 904]]<|/det|> +Another important aspect to highlight, that supports the validity of the reconstruction, is the fact that the network of proxy records used in the RWE reconstruction has similar weights in both PPE experiments: large weights are not restricted to North Atlantic bordering regions, they also occur in central to eastern Asia and western North America, in agreement with the teleconnections highlighted in numerous studies5,41- 43 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 119, 606, 137]]<|/det|> +## Unforced multidecadal variability and ongoing bifurcation + +<|ref|>text<|/ref|><|det|>[[115, 150, 883, 770]]<|/det|> +As a complement to the AMV reconstruction, which has been used as an index definition that focuses on the internal variability, we have also performed a reconstruction of NASST, which does include the influence of the external forcings, as done in previous reconstruction studies18,34. Such an additional reconstruction might help to disentangle the role of internal variability from total variability recorded in the proxy records and in the reconstruction of variations in the North Atlantic. For doing so, we use the same statistical model selection and the same nested approach than for the AMV to reconstruct the NASST back to 850 C.E. as well, using proxy records in which the externally forced signal has not been removed. The best reconstruction of NASST is obtained with the PCR approach. It exhibits a significant correlation with the AMV reconstruction (r=0.64; p<0.01, Fig. 6a). This suggests that more than 40% of the variability of NASST over the last millennium can be explained by internal variability only, the rest being related to external forcing. Strikingly, validation scores obtained by this optimal regression approach are much higher than for the AMV reconstruction (med(CE)=0.44 and med(CE)=0.25, respectively). A plausible explanation for these discrepancies in skill scores between NASST and AMV reconstructions is that the way the AMV is constructed accompanied with external forcing removal from proxy records might in turn partly decorrelate NASST and proxies by removing their common responses to the same forcings. + +<|ref|>text<|/ref|><|det|>[[117, 780, 881, 904]]<|/det|> +Using a recent reconstruction of volcanic activity46, we further perform a superposed epoch analysis47 (Methods) on both the NASST and the AMV reconstructions, to characterize the response to the 10 largest eruptions of the last millennium (Extended Data Table 4). While the reconstructed NASST has a similar response a decade after the eruption than a previous + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 416]]<|/det|> +reconstruction also based on NASST34, no significant response is found for the AMV reconstruction (Fig. 6). This is partly expected given that we have reconstructed an index of the internal climate variability. When looking at individual responses, there are eruptions where slightly significant negative AMV responses are found, although they are not the strongest of the last millennium (1171, 1601, Extended Data Fig. 4). These responses might thus not be a direct signal forced by the eruption. Furthermore, no significant cooling is observed for any of the second and third strongest eruptions, respectively in 1815 (Tambora) and 1453 (Kuwait). There is even a large annual positive peak of temperature that is found for the strongest eruption in 1257 (Samalas), while no difference is found in the AMV state before and after it (Extended Data Fig. 4). + +<|ref|>text<|/ref|><|det|>[[115, 431, 882, 728]]<|/det|> +Regarding solar forcing, neither the 10- year or 30- year filtered time series from the PMIP3 TSI reconstruction48 is significantly correlated with our 10- year filtered AMV reconstruction, even when solar forcing leads by a few years (r=0.23, p>0.2, lag=12; r=0.32, p>0.2, lag=13; respectively, Extended Data Fig. 5). Both the 10- year and the 30- year filtered time series of the TSI reconstruction a modestly significantly correlated with the NASST reconstruction (r=0.5, 0.1text<|/ref|><|det|>[[115, 780, 882, 903]]<|/det|> +The wavelet analysis in Fig. 7 shows that the AMV reconstruction exhibits important multidecadal variations (Fig. 7), contrary to what has been recently suggested for an ensemble of control simulations from 16 CMIP6 models22. Our reconstructed AMV primarily varies in the 20- 90 years band except for the 1400- 1800 period, which is more dominated by + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 883, 347]]<|/det|> +shorter (20- 40 years) cycles (Fig. 7). Thus, the 50- 70 years periodicity suggested from observations since 1850 may not be systematic, as suggested by the variability produced by climate models, in control simulation with fixed external forcings49. Our reconstruction of internal variability of the North Atlantic also contradicts a recent study suggesting that North Atlantic multi-decadal variability in the 50 to 70- year frequency band can be entirely explained through the pulses of volcanic activity during the last millennium22. These eruptions certainly contributed to the variability in the 50- 70 years spectrum band but cannot explain the whole range of variations according to this AMV reconstruction. + +<|ref|>text<|/ref|><|det|>[[115, 362, 883, 907]]<|/det|> +The wavelet analysis hints at a recent increase in the overall spectral power of the AMV. The changes in spectral characteristic of a time series can be used as an early warning of regime shift in dynamical systems, as shown in numerous studies50- 52. In particular, the AMOC is well- known as a potential tipping element of the climate system53 and it has been shown that an AMOC regime shift in climate models might necessitate the knowledge of hundreds of years of time series of AMOC variations50,52, which is prohibitive with direct observations of it at 26°N that only last for less than 20 years54. Here, the reconstruction of our AMV as an internal mode of variability can be related to the internal dynamics of the AMOC through its impact on heat transport and the AMV13,14,55. Thus, the change in spectral characteristics of the AMV might be seen as a potential early warning of a regime shift50- 52 in the ocean circulation. It is, however, difficult to tell if the overall increase in spectral power is robust as the values for the largest periodicities are outside of the cone of influence and therefore subject to edge effects. To test the hypothesis of the imminent occurrence of a potential tipping point, we use a similar approach as a former model- based study evaluating early warning signals for an AMOC critical slowdown51. Dynamically, this approach assumes that a given system is likely to be slowing down if its memory increases over time, i.e., if the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 882, 488]]<|/det|> +system state at time \(t + 1\) gets more and more dependent on the system state at time \(t\) when approaching the bifurcation point. In terms of time series, the memory of the system can be measured using autoregressive AR1 coefficients, which are then assumed to be increasing when approaching a critical slowdown \(^{50}\) . Thus, Kendall \(\tau\) statistics are computed for the AR1 coefficients for different sliding window lengths \(^{50 - 52}\) (from 200- year to 400- year with an increment of 50 years). Kendall \(\tau\) quantifies the time- evolution of autocorrelations as a ranked correlation between the AMV sliding AR1 coefficients and time (see Methods). It indicates a highly significant increase in AMV memory over the recent period for the different window lengths tested ( \(p< 0.01\) for all, Fig 7b, Methods). According to the tipping points detection theory \(^{50}\) , this constitutes the first observation- based estimate that the AMV may now be approaching a tipping point, after which the Atlantic current system might change its mean state. + +<|ref|>text<|/ref|><|det|>[[115, 501, 882, 905]]<|/det|> +These recent changes in autocorrelation properties of the AMV towards higher values can be related to a possible approach of a tipping point in the AMV. Such an AMV tipping might reflect changes in the AMOC, subpolar gyre or the Arctic circulation, which were previously reported to have tipping points in models \(^{15,16,53}\) . In this respect, the significant critical slowdown test of the AMV reconstruction could be interpreted as a long- term relative cooling of the North Atlantic in the near- term future. This assumption based on our real- data reconstruction has previously been proposed using CMIP5 models, among which such an abrupt change happens in projections of nearly half of the best ones in representing ocean convection in the northern North Atlantic \(^{15}\) , even under scenarios with low anthropogenic emissions of greenhouse gases. It thus raises serious concerns, while more and more evidence seem to indicate an on- going long- term slowdown in the Atlantic current \(^{56}\) and that the impacts of such a change are numerous \(^{16,57,58}\) . This further highlights the need for an + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 85, 880, 137]]<|/det|> +382 appropriate account of the potential implications in terms of climate adaptation plan in case of rapid changes in the Atlantic59. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 118, 881, 171]]<|/det|> +1. Kushnir, Y. Interdecadal variations in North Atlantic Sea Surface Temperatures and associated atmospheric conditions. J. Clim. 7, 1-20 (1994). + +<|ref|>text<|/ref|><|det|>[[144, 189, 880, 241]]<|/det|> +2. Ting, M., Kushnir, Y., Saeger, R. & Cuihua, L. Forced and internal twentieth-century SST trends in the North Atlantic. J. Clim. 22, 1469-1481 (2009). + +<|ref|>text<|/ref|><|det|>[[144, 258, 880, 311]]<|/det|> +3. Mann, M. E., Steinman, B. A & Miller, S. K. On forced temperature changes, internal variability, and the AMO. Geophys. Res. Lett. 41, 3211-3219 (2014) + +<|ref|>text<|/ref|><|det|>[[144, 328, 880, 381]]<|/det|> +4. Sutton, R. T. & Dong, B. Atlantic Ocean influence on a shift in European climate in the 1990s. Nat. Geosci. 5, 788-792 (2012). + +<|ref|>text<|/ref|><|det|>[[144, 398, 880, 451]]<|/det|> +5. Sutton, R. T. & Hodson, D. L. R. Atlantic Ocean forcing of North American and European summer climate. Science. 309(5731), 115-118 (2005). + +<|ref|>text<|/ref|><|det|>[[144, 468, 880, 588]]<|/det|> +6. Defrance, D., Ramstein, G., Charbit, S., Vrac, M., Famine, A. M., Sultan, B., Swingedouw, D., Dumas, C., Gemenne, F., Alvarez-Solas, J. & Vanderlinden, J.-P. Consequences of rapid ice sheet melting on the Sahelian population vulnerability. PNAS. 114(25), 6533-6538 (2017). + +<|ref|>text<|/ref|><|det|>[[144, 606, 880, 658]]<|/det|> +7. Knight, J., Folland, C & Scaife, A. Climate impacts of the Atlantic Multidecadal Oscillation. Geophys. Res. Lett. 33, L17706 (2006). + +<|ref|>text<|/ref|><|det|>[[144, 676, 880, 728]]<|/det|> +8. Trenberth, K. & Shea, D. Atlantic Hurricanes and natural variability in 2005. Geophys. Res. Lett. 33, L12704 (2006). + +<|ref|>text<|/ref|><|det|>[[144, 745, 881, 833]]<|/det|> +9. Enfield, D. & Cid-Serrano, L. Secular and multidecadal warmings in the North Atlantic and their relationships with major hurricane activity. Int. J. Climatol. 30, 174-184 (2010). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 84, 881, 171]]<|/det|> +10. Miles, M., Divine, D., Furevik, T., Jansen, E., Moros, M. & Ogilvie, A. E. J. A signal of persistent Atlantic Multidecadal variability in Arctic sea ice. Geophys. Res. Lett. 41, 463-469 (2014). + +<|ref|>text<|/ref|><|det|>[[147, 187, 881, 312]]<|/det|> +11. Ruprich-Robert, Y., Msadek, R., Castruccio, F., Yeager, S., Delworth, T. & Danabasoglu, G. Assessing the climate impacts of the Observed Atlantic Multidecadal Variability Using the GFDL CM2.1 and NCAR CESM1 Global Coupled Models. J. Clim. 30(8), 2785-2810 (2017). + +<|ref|>text<|/ref|><|det|>[[147, 327, 881, 487]]<|/det|> +12. Cook, J., Oreskes, N., Doran, P. T., Anderegg, W. R. L., Verheggen, B., Maibach, E. W., Carlon, J. S., Lewandowsky, S., Skuce, A. G., Green, S. A., Nuccitelli, D., Jacobs, P., Richardson, M., Winkler, B., Painting, R. & Rice, K. Consensus on consensus: a synthesis of consensus estimates on human-caused global warming. Environ. Res. Lett. 11, 048002 (2016). + +<|ref|>text<|/ref|><|det|>[[147, 502, 880, 591]]<|/det|> +13. Muir, L. C. & Fedorov, A. V. How the AMOC affects ocean temperatures on decadal to centennial timescales: the North Atlantic versus an interhemispheric seesaw. Clim. Dynam. 45(1-2), 151-160 (2015). + +<|ref|>text<|/ref|><|det|>[[147, 606, 880, 660]]<|/det|> +14. Yan., X., Zhang., R. & Knutson, T. Underestimated AMOC variability and implications for AMV and predictability in CMIP models. Geophys. Res. Lett. 45, 4319-4328 (2018). + +<|ref|>text<|/ref|><|det|>[[147, 675, 880, 730]]<|/det|> +15. Sgubin, G., Swingedouw, D., Drijfhout, S., Mary, Y. & Bennabi, A. Abrupt cooling over the North Atlantic in modern climate models. Nat. Commun. 8, 14375 (2017). + +<|ref|>text<|/ref|><|det|>[[147, 745, 881, 869]]<|/det|> +16. Collins, M., Sutherland, M., Bouwer, L., Cheong, S.-M., Frölicher, T., Jacot Des Combes, H., Koll Roxy, M., Losada, I., McInnes, K., Ratter, B., Rivera-Arriaga, E., Susanto, R. D., Swingedouw, D. & Tibig, L. [Pörtner, H.-O., Roberts, D. C., Masson-Delmotte, V., Zhai, P., Tignor, M., Poloczanska, E., Mintenbeck, K., Alegría, A., Nicolai, M., Okem, A., + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[175, 85, 880, 137]]<|/det|> +Petzold, J., Rama, B. & Weyer, N. M. (eds.)]. Extremes, Abrupt Changes and Managing Risk. IPCC SROCC. Chapter 6 (2019). + +<|ref|>text<|/ref|><|det|>[[147, 154, 880, 241]]<|/det|> +17. McCarthy, G. D., Haigh, I. D., Hirshi, J. J.-M., Grist, J. P., Smeed, D. A. Ocean impact on decadal Atlantic climate variability revealed by sea-level observations. Nature. 521, 508-512 (2015). + +<|ref|>text<|/ref|><|det|>[[147, 258, 880, 345]]<|/det|> +18. Gray, S., Graumlich, L., Betancourt, J. & Pederson, G. D. A tree-ring based reconstruction of the Atlantic Multidecadal Oscillation since 1567 A.D. Geophys. Res. Lett. 31, L12205 (2004). + +<|ref|>text<|/ref|><|det|>[[147, 362, 880, 451]]<|/det|> +19. Mjell, T. R., Ninnemann, U. S., Eldevik, T. & Kleiven, H. K. F. Holocene multidecadal-to millennial-scale variations in Iceland-Scotland overflow and their relationship to climate. Paleoc. and Paleoclimatol. 30(5), 558-569 (2015). + +<|ref|>text<|/ref|><|det|>[[147, 467, 880, 555]]<|/det|> +20. Clement, A., Bellomo, K., Murphy, L. N., Cane, M. A., Mauritsen, T., Rädel, G. & Stevens, B. The Atlantic Multidecadal Oscillation without a role for ocean circulation. Science. 350(6250), 320-324 (2015). + +<|ref|>text<|/ref|><|det|>[[147, 571, 880, 658]]<|/det|> +21. Booth, B. B. B., Dunstone, N. J., Halloran, P. R., Andrews, T. & Bellouin, N. Aerosols implicated as a prime driver of twentieth-century North Atlantic climate variability. Nature, 484, 228-232 (2012). + +<|ref|>text<|/ref|><|det|>[[147, 675, 880, 763]]<|/det|> +22. Mann, M. M., Steinman, B. A., Brouillette, D. J. & Miller, S. K. Multidecadal climate oscillations during the past millennium driven by volcanic forcing. Science. 371, 1014-1019 (2021). + +<|ref|>text<|/ref|><|det|>[[115, 779, 880, 890]]<|/det|> +Schneider, L., Smerdson, J. E., Pretis, F., Hartl-Meier & C. Esper, J. A new archive of large volcanic events over the past millennium derived from reconstructed summer temperatures. Environ. Res. Lett. 12, 094005 (2017).23. Knudsen, M. F., Seidenkrantz, M.-S., Jacobsen, B. H., Kuijpers, A. Tracking the Atlantic Multidecadal Oscillation through the last 8,000 years. Nat. Commun. 2, 178 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 85, 881, 138]]<|/det|> +24. Frajka-Williams, E., Beaulieu, C. & Duchez A. Emerging negative Atlantic Multidecadal Oscillation index in spite of warm subtropics. Sci. Rep. 7, 11224 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 153, 881, 207]]<|/det|> +25. Lin, P., Yu, Z., Lü, J., Ding, M., Hu, A. & Liu, H. Two regimes of Atlantic Multidecadal Oscillation: cross-basin dependent or Atlantic-intrinsic. Sci. Bull. 64(3), 198-204 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 222, 881, 276]]<|/det|> +26. Hall, A., Cox, P. M., Huntingford, C. & Klein, S. Progressing emergent constraints on future climate change. Nat. Clim. Chang. 9(4), 269-278 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 291, 881, 380]]<|/det|> +27. Frankignoul, C., Gastineau, G. & Kwon, Y.-O. Estimation of the SST Response to Anthropogenic and External Forcing and Its Impact on the Atlantic Multidecadal Oscillation and the Pacific Decadal Oscillation. J. Clim. 30(24), 9871-9895 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 396, 881, 487]]<|/det|> +28. Enfield, D. & Cid-Serrano, L. Secular and multidecadal warmings in the North Atlantic and their relationships with major hurricane activity. Int. J. Climatol. 30, 174-184 (2010). + +<|ref|>text<|/ref|><|det|>[[145, 502, 880, 556]]<|/det|> +29. Harris, I., Osborn, T. J., Jones, P. & Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data. 7(1), 109 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 571, 880, 625]]<|/det|> +30. Shurer, A., Tett, S. & Hegeril, G. Small influence of solar variability on climate over the past millennium. Nat. Geosci. 7, 104-108 (2014). + +<|ref|>text<|/ref|><|det|>[[145, 640, 881, 730]]<|/det|> +31. Schleussner, C., Divine, D, Donges, J., Miettinen, A., Donner, R. V. Indications for a North Atlantic Ocean regime shift at the onset of the Little Ice Age. Clim. Dynam. 45, 3623-3633 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 745, 881, 833]]<|/det|> +32. Moreno-Chamarro, E., Zanchettin, D., Lohmann, K. & Jungclaus, J. H. An abrupt weakening of the subpolar gyre as trigger of LIA-type episodes. Clim. Dynam. 48(3-4), 727-744 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 881, 172]]<|/det|> +33. Wang, J., Yang, B., Charpentier Ljungqvist, F., Luterbacher, J., Osborn, T. J., Briffa, K. R. & Zorita, E. Internal and external forcing of multidecadal Atlantic variability over the past 1,200 years. Nat. Geosci. 10, 512-518 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 189, 881, 241]]<|/det|> +34. PAGES 2k Consortium. A global multiproxy database for temperature reconstructions of the Common Era. Sci. Data. 4, 170088 (2017). + +<|ref|>text<|/ref|><|det|>[[145, 258, 881, 383]]<|/det|> +35. Rayner, N. A., Parker, D. E., Horton, E. B., Folland, C. K., Alexander, L. V., Rowell, D. P., Kent, E. C. & Kaplan, A. Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth. Journ. Geophys. Res. 108(D14) (2003). + +<|ref|>text<|/ref|><|det|>[[145, 398, 881, 451]]<|/det|> +36. Neukom, R., Shurer, A. P., Steiger, N. J. & Hegerl, G. C. Possible causes of data model discrepancy in the temperature history of the last Millennium. Sci. Rep. 8, 7572 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 467, 881, 555]]<|/det|> +37. Michel, S., Swingedouw, D., Ortega, P., Khodri, M., Mignot, J. & Chavent, M. Reconstructing climatic modes of variability from proxy records using ClimIndRec version 1.0. Geosci. Mod. Dev. 13, 841-858 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 572, 881, 625]]<|/det|> +38. Nash, J. E. & Sutcliffe, J. V. River flow forecasting through conceptual models part I: A discussion of principles. J. Clim. 10, 282-290 (1970). + +<|ref|>text<|/ref|><|det|>[[145, 641, 661, 660]]<|/det|> +39. Breiman, L. Random Forests. Mach. Learn. 45, 5-32 (2001). + +<|ref|>text<|/ref|><|det|>[[145, 676, 881, 800]]<|/det|> +40. Shi, C., Sun, C., Wu, G., Wu, X., Chen, D., Masson-Delmotte, V., Li, J., Xue, J., Li, Z., Ji, D., Zhang, J., Fan, Z., Shen, M., Shu, L. & Ciais, P. Summer temperature over the Tibetan Plateau modulated by Atlantic Multidecadal Variability. Journal of Climate. 32(13), 4055-4067 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 816, 881, 905]]<|/det|> +41. Li, J., Li, F., He, S., Wang, H. & Orsolini, Y. J. The Atlantic Multidecadal Variability phase dependence of teleconnection between the North Atlantic Oscillation in February and the Tibetan Plateau in March. J. Clim. 34(11), 4227-4242 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[144, 84, 880, 137]]<|/det|> +42. Monerie, P.-A., Robson, J., Dong, B. & Hodson, D. Role of the Atlantic Multidecadal Variability in modulating East Asian Climate. Clim Dynam. 56, 381-398 (2021). + +<|ref|>text<|/ref|><|det|>[[145, 153, 880, 207]]<|/det|> +43. Buntgen, U., Frank, D. C., Nievergelt, D & Esper, J. Summer temperature variations in the European Alps, A.D. 755-2004. J. Clim. 19(21), 5606-5623 (2006). + +<|ref|>text<|/ref|><|det|>[[145, 223, 881, 312]]<|/det|> +44. Ortega, P., Lehner, F., Swingedouw, D., Masson-Delmotte, V., Raible, C. C., Casado, M., and Yiou, P. A model-testet North Atlantic Oscillation reconstruction for the past millennium. Nature. 523, 71-74 (2015) + +<|ref|>text<|/ref|><|det|>[[145, 328, 881, 520]]<|/det|> +45. Sigl, M., Winstrup, M., McConnell, J. R., Welten, K. C., Plunkett, G., Ludlow, F., Buntgen, U., Caffee, M., Chellman, N., Dahl-Jensen, D., Fisher, H., Kipfstuhl, S., Kostick, C., Maselli, J., Mekhaldi, F., Mulvaney, R., Muscheler, R., Pasteris, D. R., Pilcher, J. R., Salzer, M., Schüpbach, S., Steffensen, J. P., Vinther, B. M. & Woodruff, T. E. Timing and climate forcing of volcanic eruptions for the past 2,500 years. Nature. 523, 543-549 (2015). + +<|ref|>text<|/ref|><|det|>[[145, 537, 880, 625]]<|/det|> +46. Rao, M. P., Cook, E. R., Cook B. I., Anchukaitis, K. J., D'Arrigo, R. D., Krusic, P. J. & LeGrande, A. N. A double bootstrap approach to Superposed Epoch Analysis to evaluate response uncertainty. Dendrochronologia. 55, 119-124 (2019). + +<|ref|>text<|/ref|><|det|>[[145, 641, 880, 694]]<|/det|> +47. Vieira, L. E. A., Solanki, S. K., Krikova, N. A. & Usoskin, I. Evolution of the solar irradiance during the Holocene. Astronom., Astrophys. 531, A6 (2011). + +<|ref|>text<|/ref|><|det|>[[145, 711, 880, 800]]<|/det|> +48. Frankcombe, L. M., von der Heydt, A. & Dijkstra, H. North Atlantic multidecadal climate variability: An investigation of dominant time scales and processes. J. Clim. 23, 3626-3638 (2010). + +<|ref|>text<|/ref|><|det|>[[145, 816, 880, 869]]<|/det|> +49. Lenton, T. M. Early warning of climate tipping point. Nat. Clim. Change. 1, 201-208 (2011). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[145, 84, 881, 172]]<|/det|> +50. Lenton, T. M., Livina, V. N., Dakos, V., van Nes, E. H. and Scheffer, M. Early warning of climate tipping points from critical slowing down: comparing methods to improve robustness. Phil. Trans. R. Soc. A. 370, 1185-1204 (2012). + +<|ref|>text<|/ref|><|det|>[[145, 188, 881, 277]]<|/det|> +51. Boulton, C. A., Allison, L. C. & Lenton, T. M. Early warning signals of Atlantic Meridional Overturning Circulation collapse in a fully coupled climate model. Nat. Commun. 5, 5752 (2014). + +<|ref|>text<|/ref|><|det|>[[145, 292, 881, 383]]<|/det|> +52. Swingedouw, D., Speranza, C. I., Bartsch, A., Durand, G., Jamet, C., Beaugrand, G., & Conversi, A. Early warning from space for a few kipping points in physical, biological and social-ecological systems. Surv. Geophys. 41, 1237-1284 (2020). + +<|ref|>text<|/ref|><|det|>[[145, 397, 881, 520]]<|/det|> +53. Smeed, D. A., Josey, S. A., Beaulieu, C., Johns, W. E., Moat, B. I., Frajka-Williams, E., Rayner, D., Meinen, C. S., Baringer, M. O., Bryden, H. L. & McCarthy, G. D. The North Atlantic Ocean is in a state of reduced overturning. Geophys. Res. Lett. 45(3), 1527-1533 (2018). + +<|ref|>text<|/ref|><|det|>[[145, 536, 881, 623]]<|/det|> +54. Knight, J. R., Allan, R. J., Folland, C. K. Vellinga, M. & Mann, M. E. A signature of persistent natural thermohaline circulation cycles in observed climate. Geophys. Res. Lett. 32(20), L20708 (2005). + +<|ref|>text<|/ref|><|det|>[[145, 639, 881, 730]]<|/det|> +55. Caesar, L., McCarthy, G. D., Thornalley, D. J. R., Cahill, N. & Rahmstorf, S. Current Atlantic Meridional Overturning Circulation weakest in the last millennium. Nat. Geosci. 14, 118-120 (2021). + +<|ref|>text<|/ref|><|det|>[[145, 745, 881, 869]]<|/det|> +56. Defrance, D., Carty, T., Rajaud, A., Dessay, N. & Sultan, B. Impacts of Greenland and Antarctic ice sheet melt on future Köppen climate zone changes simulated by an atmospheric and oceanic general circulation model. Applied Geography. 119, 102216 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[174, 84, 880, 172]]<|/det|> +57. Valesco, J. A., Estrada, F., Calderón-Bustamante, O., Swingedouw, D., Ureta, C., Gay, C. & Defrance, D. Synergistic impacts of global warming and thermohaline circulation collapse on amphibians. Commun. Biol. 4, 141 (2021). + +<|ref|>text<|/ref|><|det|>[[175, 189, 880, 242]]<|/det|> +58. Sutton, R. T. Climate science needs to take risk assesment much more seriously. Bull. Am. Meteor. Soc. 100(9), 1637-1642 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 291, 882, 800]]<|/det|> +Acknowledgements: This research was partly funded by the Universite de Bordeaux. It is also funded by the LEFE- IMAGO project. This study benefited from the IPSL Prodiguer- Ciclad and Camelot facilities, supported by CNRS, UPMC Labex L- IPSL. Part of the code has also been run on the SurfSara Cartesius supercomputer (Amsterdam). Simulations for the LME experiments have been downloaded on the NCAR Climate Data gateway website (https://www.earthsystemgrid.org/). CRUTS4 data have been downloaded from the CRU data download webpage (https://crudata.uea.ac.uk/cru/data/hrg/#current). HadISST data have been downloaded from the Met Office Hadley Center website (https://www.metoffice.gov.uk/hadobs/hadisst/). For the historical experiments of CMIP5 models, data can be downloaded from on the ESGF website (https://esgf- node.llnl.gov/projects/esgf- llnl/). Codes for reconstructions, statistical analysis, statistical tests, and figures have been integrally implemented in R and bash UNIX languages. Authors were also funded by EU- H2020 Blue Action (Grant Agreement no. 727852, D.S., G.G, and J.M.), EUCP (Grant Agreement no 776613, D.S. and J.M.), ROADMAP (J.M. and G.G.) ARCHANGE (ANR- 18- MPGA- 0001, J.M. and G.G.) and TiPES (S. M.) research programmes. + +<|ref|>text<|/ref|><|det|>[[115, 850, 880, 904]]<|/det|> +Author contributions: S.M. have performed the different reconstructions, statistical analysis, figures, and the pseudo- proxy experiment of this study. S. M. has mainly written the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 174]]<|/det|> +manuscript with important participation of D. S. D. S., J. M, M. K., P. O., G. G. and G. M have contributed to the results assessments, the manuscript writing, and have made suggestions to set up the manuscript's guiding thread. + +<|ref|>text<|/ref|><|det|>[[115, 225, 661, 243]]<|/det|> +Competing interests: The authors declare no competing interests. + +<|ref|>text<|/ref|><|det|>[[115, 290, 880, 339]]<|/det|> +Codes and data availability: All codes and data needed to reproduce this study are publicly available on the following Zenodo link: https://zenodo.org/record/4896670#.YljdOS2w3dc. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[117, 86, 201, 101]]<|/det|> +## Methods: + +<|ref|>sub_title<|/ref|><|det|>[[118, 119, 344, 136]]<|/det|> +## Instrumental AMV indices: + +<|ref|>text<|/ref|><|det|>[[116, 153, 881, 241]]<|/det|> +Historical AMV indices have been calculated using annually- resolved values. The AMV reconstructions compared and presented in this study are then smoothed with a ten year kernel filter. + +<|ref|>text<|/ref|><|det|>[[116, 258, 882, 415]]<|/det|> +We have computed, over the period 1870- 2019, the three instrumental AMV indices \(^{2,8,28}\) of this study using the Hadley Center Global Sea Ice and Sea Surface Temperatures (HadISST) dataset \(^{36}\) . The three indices are based on the spatially averaged SST over the North Atlantic (between latitudes \(0^{\circ}\) and \(60^{\circ}\mathrm{N}\) ) and differ in the way the externally forced signal is removed. + +<|ref|>text<|/ref|><|det|>[[115, 431, 882, 905]]<|/det|> +The different approaches used assume that external forcings factors have different implications for temperatures and precipitation over time. The first index uses the global averaged SST anomalies as a proxy for the externally forced signal that is subtracted from NASST, resulting in the \(AMV_{TS}\) index \(^{8}\) (Fig. 1a). This approach has limitations since it does not account for regional variations in external forcings, such as the distribution of anthropogenic aerosols, which was denser in North America- Europe in the middle of last century and became more prominent in Asia in recent decades \(^{60}\) . To circumvent this problem, a different index, the \(AMV_{T}\) (Fig. 1a), is built using climate model historical simulations to isolate the forced component in NASST, which is calculated with a signal- to- noise maximizing empirical orthogonal function (EOF) that is then removed by an estimate of its 10- year smoothed effect \(^{2}\) at each grid point of the North Atlantic. \(AMV_{T}\) is then obtained as the spatial average of the previously regressed time series. Finally, the \(AMV_{F}\) \(^{28}\) is obtained as the spatial average of the North Atlantic time series regressed onto the ten years moving average of the global mean SST between \(60^{\circ}\mathrm{S}\) and \(60^{\circ}\mathrm{N}\) . + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 84, 778, 103]]<|/det|> +## Estimation of NASSTs forced component using signal-to-noise maximizing EOF: + +<|ref|>text<|/ref|><|det|>[[115, 117, 881, 278]]<|/det|> +The forced component of NASST is estimated from historical simulations of 37 climate models (Extended Data Table 2). For each, the NASST anomalies are extracted and merged as columns of the same matrix. Using a Principal Component Analysis of the latter matrix, the first Principal Component is retained as the estimated forced component of NASST (Extended Data Fig. 1). For additional details, the reader is referred to ref. 2. + +<|ref|>sub_title<|/ref|><|det|>[[118, 329, 322, 346]]<|/det|> +## Proxy records database: + +<|ref|>text<|/ref|><|det|>[[115, 362, 882, 556]]<|/det|> +To select the proxy records of this study, a first large dataset is made by merging the PAGES 2k database (686) with other proxy records in neighboring continents of the North Atlantic used for previous AMV and NAO reconstructions34,45. Duplicates from the 3 sources are removed leading to a final database of 727 proxy records (P2k- ALL). Since the focus is on reconstructing the annual variations of the AMV, the proxy records from P2k- ALL finally used in the reconstruction have been selected to fulfill the following conditions: + +<|ref|>text<|/ref|><|det|>[[145, 571, 881, 728]]<|/det|> +1) They are annually-resolved +2) They are located in the Northern hemisphere (latitude \(>0^{\circ}\) ). +3) They are significantly correlated at the 95% confidence level with at least one historical time series of either annual or seasonal precipitation or surface temperatures from the nearest grid point within the CRUTS4 dataset30. + +<|ref|>text<|/ref|><|det|>[[115, 744, 882, 903]]<|/det|> +The first analyses when preparing this study singled out a proxy record from Asia (named "Asia.MOR1JU" in P2k- ALL), which had abnormally large RF weights (more than 5 times higher than the second) as compared to all the other proxy records. To prevent highly biasing the reconstruction towards this single proxy we have decided to remove it from the database used in this study. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 881, 276]]<|/det|> +The reconstruction procedure of this study is then based on the resulting database (P2k+). In the next method sections, we describe how different reconstructions are compared and how a final nested reconstruction of the AMV is obtained. These reconstructions will also use correlation tests to select proxies from P2k+ that are significantly correlated with a given AMV index, for a given learning and reconstruction period. This means that only a subset of the most relevant proxy records from the P2k+ database is finally used in each reconstruction. + +<|ref|>sub_title<|/ref|><|det|>[[117, 292, 690, 310]]<|/det|> +## Generation of the 312 initial reconstructions compared in the study: + +<|ref|>text<|/ref|><|det|>[[115, 327, 882, 694]]<|/det|> +The 312 reconstructions compared in this study, are performed for 26 time windows \(\Gamma\) : from 850- 1975 to 850- 2000, thus subsequently incrementing the superior boundary by one year. We use this approach as a way to sample the sensitivity of the reconstructions to the calibration period, as for the shortest windows more proxies are available but the regression models are built with shorter time series and therefore fewer degrees of freedom. These 26 temporal windows are used in combination with three AMV indices \(^{2,8,28}\) and the four regression methods \(^{40,61 - 63}\) . Detailed description of these regression methods is given in ref. 38 and Supplementary Information. All these setups are tested by only using proxies available and significantly correlated at the 95% confidence level with the respective AMV index. We thus end up with \(26 \times 3 \times 4 = 312\) final reconstructions that are compared in this study using the CE metric. + +<|ref|>text<|/ref|><|det|>[[117, 710, 880, 762]]<|/det|> +For the NASST reconstruction, 84 setups are compared, by shuffling the same 26 temporal windows and the same 4 regression methods as the AMV indices. + +<|ref|>sub_title<|/ref|><|det|>[[117, 780, 530, 798]]<|/det|> +## Computation of a reconstruction and evaluation: + +<|ref|>text<|/ref|><|det|>[[117, 814, 881, 903]]<|/det|> +We define the reconstruction period as \(\Gamma\) , defined by \(N\) annual time steps, and the common period of the proxy records and the AMV index as \(T\) , in this case defined by \(n < N\) annual time steps such that \(T \subset \Gamma\) . We then define the AMV index as \(Y \in R^n\) and the matrix of the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 179]]<|/det|> +available proxy records as \(X \in R^{N \times p}\) . We finally denote \(\text{as} x \in R^{n \times p}\) the sub- matrix of \(X\) that contains the proxy records values over the time frame \(T\) . \(X\) can then be denoted as \(X = [(X_{t}^{j})_{t \in T}]_{1 \leq k \leq p}\) and \(x = [x^{j}]_{1 \leq j \leq p} = [(X_{t}^{j})_{t \in T}]_{1 \leq j \leq p}\) . + +<|ref|>text<|/ref|><|det|>[[115, 192, 881, 321]]<|/det|> +We then randomly split \(T\) in \(R = 30\) pairs of training/testing samples respectively denoted, \(\forall 1 \leq r \leq R\) , by \(\{x_{(train)}^{(r)}; Y_{(train)}^{(r)}\}\) and \(\{x_{(test)}^{(r)}; Y_{(test)}^{(r)}\}\) . Here, the training sample size is set to be \(80\%\) of the length of \(T\) and, by extension, the testing sample size is \(20\%\) of the length of \(T\) . + +<|ref|>text<|/ref|><|det|>[[115, 336, 881, 475]]<|/det|> +For statistical modelling, we use \(Y_{(train)}^{(r)}\) as predictand and \(x_{(train)}^{(r)}\) as predictors. For a given regression method denoted \(M\) , we apply KFCV (cf. Methods, section "K- Fold cross- validation (KFCV)") to each training set \(\left\{x_{(train)}^{(r)}; Y_{(train)}^{(r)}\right\}\) as a metric to find the optimal set of parameters associated to the training sample and \(M\) . + +<|ref|>text<|/ref|><|det|>[[115, 490, 881, 660]]<|/det|> +\(M\) and the associated optimal set of control parameters are then applied to \(X^{(r)}\) in order to reconstruct \(Y^{(r)}\) on both the testing period, giving \(\hat{Y}_{(test)}^{(r)}\) , and the reconstruction period, giving \(\hat{Y}_{(rec)}^{(r)}\) . This involves that \(\hat{Y}_{(test)}^{(r)} = (\hat{Y}_{(rec)}^{(r)})_{t \in T}\) . The validation score associated to the \(r^{th}\) training sample is then calculated using the Coefficient of Efficiency (CE) \(^{39}\) over the \(r^{th}\) testing sample: + +<|ref|>equation<|/ref|><|det|>[[115, 675, 805, 720]]<|/det|> +\[s^{r} = CE\left(\hat{Y}_{(test)}^{(r)}, Y_{(test)}^{(r)}\right) = 1 - \frac{\sum_{i = 1}^{m} (Y_{i(test)}^{(r)} - \hat{Y}_{i(test)}^{(r)})^{2}}{\sum_{i = 1}^{m} (Y_{i(test)}^{(r)} - Y_{(test)}^{(r)})^{2}}, \text{with} \frac{Y_{(test)}^{(r)}}{Y_{(test)}} = \frac{1}{m} \sum_{i = 1}^{m} Y_{i(test)}^{(r)}\] + +<|ref|>text<|/ref|><|det|>[[115, 737, 485, 756]]<|/det|> +Where \(m\) is the length of the testing sample. + +<|ref|>text<|/ref|><|det|>[[115, 771, 881, 894]]<|/det|> +This validation score gives an estimation of the accuracy of the statistical model when reconstructing the observed variability not included in the reconstruction period. \(CE < 0\) means that the sample average of the testing period is more reliable than the output given by the statistical model \(^{39}\) . Contrarily, \(CE > 0\) , means that the statistical model gives a more + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 880, 138]]<|/det|> +reliable reconstruction than the empirical average of the testing sample39, the associated reconstruction is thereby considered as reliable in this study. + +<|ref|>text<|/ref|><|det|>[[115, 152, 882, 315]]<|/det|> +The reconstruction for a given AMV index \(Y\) performed on a given time frame \(I\) using a given statistical regression method \(M\) is obtained by applying it with KFCV38 (Methods, next section) over the whole learning sample. This reconstruction is thus associated to a global validation score, calculated as the mean of the individual validation scores obtained over the random splits: \(s = avg(\{s_r\}_{1 \leq r \leq R})\) . + +<|ref|>sub_title<|/ref|><|det|>[[118, 328, 377, 346]]<|/det|> +## K-Fold cross-validation (KFCV): + +<|ref|>text<|/ref|><|det|>[[115, 361, 882, 485]]<|/det|> +Each method requires an optimization of its own set of control parameters \(\theta\) . To estimate the optimal set of control parameters \(\theta_{opt}\) on a given training set \(\{X_{train}, Y_{train}\}\) , we use the KFCV approach64,65. Cross validation methods are in general widely used as model validation and selection techniques. + +<|ref|>text<|/ref|><|det|>[[115, 500, 882, 771]]<|/det|> +The KFCV splits the observations into a partition of \(K\) groups of the same sizes (or with approximately the same size if the length of the training set is not divisible by \(K\) ). \(\forall 1 \leq k \leq K\) , we denote \(\{X_{(k)}, Y_{(k)}\}\) , which contain only observations for the \(k^{th}\) drawn sample. We denote \(\{X_{(-k)}, Y_{(-k)}\}\) the \(K - 1\) other sets. For all possible values of \(\theta \in \theta\) , we scan the \(K\) models based on the sets \(\{X_{(-k)}, Y_{(-k)}\}\) . The empirical optimal set of control parameters is obtained by minimizing the averaged Root Mean Squared Errors (RMSE) on the K splits by considering all possible values of \(\theta\) . The optimal KFCV set of control parameters \(\theta_{KF}\) is determined by: + +<|ref|>equation<|/ref|><|det|>[[175, 783, 550, 814]]<|/det|> +\[\hat{\theta}_{opt} = \theta_{KF} = arg\frac{1}{K}\sum_{k = 1}^{K}R M S E(Y_{(k)},\hat{Y}_{(k),\theta})\] + +<|ref|>sub_title<|/ref|><|det|>[[118, 828, 313, 845]]<|/det|> +## Nested reconstruction: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 208]]<|/det|> +In this study, the best reconstruction found (defined as the one yielding the best CE metric) is the reconstruction of the \(AMV_F\) index with the random forest method over the period 850- 1987, using only the proxy records significantly correlated at the 95% confidence level with the \(AMV_F\) index over the training periods (see Extended Data Fig. 2). + +<|ref|>text<|/ref|><|det|>[[115, 223, 882, 415]]<|/det|> +Using these methodological choices (calibration period, AMV definition and reconstruction method), we have performed a set of 1021 nested reconstructions for the periods 850- 1987 to 1869- 1987, subsequently using increments of one year for the inferior boundary, which allow us to use an increasing number of proxy records to reconstruct the most recent years. The nested reconstruction (e.g., the reconstruction presented in this study), is obtained by concatenating the first year in each of the 1021 reconstructions34. + +<|ref|>sub_title<|/ref|><|det|>[[118, 432, 424, 450]]<|/det|> +## Random forest variable importance: + +<|ref|>text<|/ref|><|det|>[[115, 465, 882, 904]]<|/det|> +The weights of the proxy records used for the nested reconstruction are presented in Fig. 2c. Those weights have been calculated using the random forest variable importance40. Different importance metrics exist, and for this study we have selected the commonly used Mean Decrease in Impurity (MDI), also known as Gini importance. The MDI of a given proxy record is calculated as the sum of the number of splits where it is used across the \(K\) trees (see Supplementary Information for details on the regression methods), proportionally to the numbers of split samples in all trees40 (cf. Supplementary Information). For Fig. 2c, the MDI for each proxy is aggregated over the 1021 reconstructions using a weight of \(n/N\), where \(n\) is the number of available proxies for a given time step, and \(N\) the total number of proxies used at the end for the reconstruction (i.e., \(N = 55\)). Finally, Fig. 2c is computed by calculating the importance of each proxy as a fraction of the previously calculated importance and the total importance over the 1001 reconstructions. The same is done in Fig. 3b, but each importance is also averaged over the PPE performed on the 12 members of CESM1-LME used. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 85, 308, 102]]<|/det|> +## NASST reconstruction: + +<|ref|>text<|/ref|><|det|>[[115, 117, 882, 312]]<|/det|> +To reconstruct the NASST, the same procedure as for the AMV is applied to select the optimal reconstruction approach, with the only difference that only one index definition is considered, as no method to remove the forced variability is applied. This means that the optimal model selection is made for \(312 / 3 = 104\) compared reconstructions. The final NASST reconstruction is obtained using nested PCR reconstructions for the 1021 periods 850- 1977 to 1869- 1977, with an increment of one year for the inferior boundary. + +<|ref|>sub_title<|/ref|><|det|>[[118, 328, 516, 346]]<|/det|> +## Composites of ocean proxy records time series: + +<|ref|>text<|/ref|><|det|>[[115, 360, 882, 836]]<|/det|> +Since corals often are very short records and ocean sediment cores have too low temporal resolution (preventing them to meet the requirement of being annually resolved), there is a shortage of ocean records contributing to the reconstructions, which is almost exclusively based on terrestrial records. Interestingly, the low- frequency part in the annually resolved reconstructions can be verified against the ocean records from the Ocean 2K database in Fig. 3 that have not been used in the reconstruction. To avoid overfitting, correlations significance shown in Fig. 3a are only calculated for the preindustrial period (before 1870) with an AR1 correction for the correlation tests to avoid falsely detected significance due to the low resolution of some proxies (See "Statistical information" method section). For the same overfitting reasons, the composite average time series are performed by multiplying by - 1 ocean proxies which have negative correlations with the AMV over the historical period only. There is an exception for some ocean records (<10 from Fig 3a) that do not overlap with our reconstruction over the historical period. These time series are therefore multiplied by - 1 if their correlations are negative over their overlapping period with the AMV reconstruction. + +<|ref|>sub_title<|/ref|><|det|>[[118, 885, 551, 903]]<|/det|> +## Two-way multimember pseudo-proxy experiments: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 82, 882, 348]]<|/det|> +For the pseudo proxy experiment, we use 12 last millennium large ensemble members of the National Center of Atmospheric Research (NCAR) Community Earth System Model 1 (CESM1). Since the calculation of a trend for a given NASST time series is time- dependent, we distinguish the calculation of the model AMV over the preindustrial (PI) period (pre- 1870, \(AMV_{PI}\) ) and the historical one ( \(AMV_{H}\) ), notably because anthropogenic forcings were small during the PI period as compared to the recent one (historical). For \(AMV_{H}\) , we calculate the \(AMV_{F}\) , similarly to the real experiment. For calculating \(AMV_{PI}\) , the same is applied but by estimating the NASSTs relationships with the global SST over the PI period. + +<|ref|>text<|/ref|><|det|>[[115, 362, 882, 556]]<|/det|> +The "Proxy records database" method section emphasizes the fact that the proxy records do not target the same climate variables and seasonalities. For this reason, pseudo- proxies mimic the real- world proxies by taking, in the model, the variable and season (or annual values) with which the real proxies exhibit the largest absolute correlations (see Extended Data Table 3) with the closest grid points from the CRUTS4 dataset. Gaps and missing values in real proxy records are also reproduced in the pseudo- proxy time series. + +<|ref|>text<|/ref|><|det|>[[115, 570, 882, 660]]<|/det|> +For both PPE cases presented below, and for the sake of reducing computational costs, nested reconstructions have been made with a 20 years time step for the inferior boundary (from 850- 1987 to 1850- 1987) instead of the 1 year one which is used for the real experiment. + +<|ref|>sub_title<|/ref|><|det|>[[118, 712, 249, 728]]<|/det|> +## Proxy to model: + +<|ref|>text<|/ref|><|det|>[[115, 745, 882, 903]]<|/det|> +The pseudo- proxy experiment is first used to reconstruct a model AMV reconstruction using the same proxy records and RF method as for the real experiment, and for each time step of the nested reconstruction. Therefore, the reconstruction scores presented in Fig. 4.1 for each CESM1 member are those obtained over the 1020 nested reconstruction timeframes. The correlations are those calculated between the RF- based reconstruction of the model AMV + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 84, 880, 137]]<|/det|> +and \(AMV_{PI}\) . For each member, the proxy records' weights are calculated in the same way as for Fig 2c, and an ensemble average is presented in Fig. 4b. + +<|ref|>sub_title<|/ref|><|det|>[[117, 190, 250, 206]]<|/det|> +## Model to proxy: + +<|ref|>text<|/ref|><|det|>[[115, 222, 883, 591]]<|/det|> +The second step of the PPE consists in training RF models directly within the CESM1 members, in which the pseudo- proxies are selected using \(95\%\) confidence level correlation tests with the model AMV. These trained RF models tailored to the model simulations are then applied to the historical simulated \(AMV_F\) index and compared to the reconstruction using real- world derived weights in Fig. 5. Since real- world proxy records have been measured with specific units (tree ring MXD, ice core \(\delta^{18}O\) ,...), the model pseudo- proxies are rescaled to the mean and the variance of the corresponding real- world proxy. For the same reason, the pseudo- proxy is multiplied by - 1 if its correlation with the model AMV has an opposite sign to that of the real- world proxy with the real world AMV. This PPE approach is similar to the model- constrained one from a published NAO reconstruction based on the PCR method45, which has been adapted to the RF one in this study. + +<|ref|>sub_title<|/ref|><|det|>[[118, 644, 330, 661]]<|/det|> +## Early warning signal test: + +<|ref|>text<|/ref|><|det|>[[115, 675, 883, 904]]<|/det|> +We base our approach on methods for the detection of incoming climate tipping points50,51, recently applied to detect an AMOC slowdown in a general circulation model52. The AMV reconstruction is firstly smoothed using a Kernel Gaussian filtering with a bandwidth of 100 years. The annually- resolved AMV is then regressed onto its long term filtered version. AR(1) coefficients of the residuals from this regression are calculated for different sliding window lengths WL=200, 250, 300, 350, 400 years (Fig. 4b). Kendall \(\tau\) is calculated for each of the AR(1) coefficient series. Contrary to a former study focusing on early warning signal applied to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 172]]<|/det|> +model- based investigation of an AMOC collapse52, we cannot use a model- based estimate of the significance of Kendall \(\tau\) , which is rather calculated using a gaussian approximation as detailed in the "Statistical information" Methods section. + +<|ref|>sub_title<|/ref|><|det|>[[115, 225, 200, 241]]<|/det|> +## Boxplots: + +<|ref|>text<|/ref|><|det|>[[115, 258, 881, 415]]<|/det|> +For all boxplots of the study, the median is shown as a heavy darkline. Boxplots edges give first and third quartiles. Boxplot "whiskers" gives the full range without including outliers, which are not shown here for better graphical representations. A point from a boxplot is here considered as an outlier when it is outside 1.5 times the interquartile range above the upper quartile and below the lower quartile. + +<|ref|>sub_title<|/ref|><|det|>[[115, 468, 313, 485]]<|/det|> +## Statistical information: + +<|ref|>text<|/ref|><|det|>[[115, 502, 644, 520]]<|/det|> +This section describes the different statistical tests of this study. + +<|ref|>text<|/ref|><|det|>[[175, 536, 881, 658]]<|/det|> +Fig 1b,c: For each grid point, a two- tailed Student test is applied to the regression coefficients between the corresponding climate variable and the AMV indices. The degrees of freedom are corrected using time series autocorrelations as in ref. 17 and 38. + +<|ref|>text<|/ref|><|det|>[[175, 675, 881, 764]]<|/det|> +Fig. 4b: The Kendall rank correlation coefficient, or Kendall \(\tau\) coefficient, measures the ordinal association between two quantities, here AR(1) coefficients denoted \((x_{i})_{1 \leq i \leq n}\) here, and time denoted \((y_{i})_{1 \leq i \leq n}\) . The statistic is given by: + +<|ref|>equation<|/ref|><|det|>[[444, 778, 552, 813]]<|/det|> +\[\tau = \frac{n_{c} - n_{d}}{n_{0}}\] + +<|ref|>text<|/ref|><|det|>[[204, 815, 833, 835]]<|/det|> +Where, considering \((x_{1}, y_{1}), (x_{2}, y_{2}), \ldots , (x_{n}, y_{n})\) the ensemble of joint pairs: + +<|ref|>equation<|/ref|><|det|>[[206, 849, 832, 911]]<|/det|> +\[n_{c} = card_{i\neq j}\{[x_{i} > x_{j}\cap y_{i} > y_{j}\} \cup \{x_{i}< x_{j}\cap y_{i}< y_{j}\} ,(i,j)\in [[1,n]]^{2\] \[n_{d} = card_{i\neq j}\{[x_{i} > x_{j}\cap y_{i}< y_{j}\} \cup \{x_{i}< x_{j}\cap y_{i} > y_{j}\} ,(i,j)\in [[1,n]]^{2\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[432, 83, 565, 118]]<|/det|> +\[n_0 = \frac{n(n - 1)}{2}\] + +<|ref|>text<|/ref|><|det|>[[206, 119, 881, 260]]<|/det|> +For large sample \((n > 50)\) , as in this study, the distribution is approximated with a Gaussian distribution of mean 0 and variance \(\frac{2(2n + 5)}{9n(n - 1)}\) , under the null hypothesis \(H_0: \tau = 0\) "which is tested against the alternative hypothesis \(H_1: \tau \neq 0\) ". The p- value (shown in Fig 4b) of the test is deduced from the quantile of this distribution. + +<|ref|>text<|/ref|><|det|>[[175, 302, 882, 460]]<|/det|> +- Correlation tests: The same bilateral Student test for correlation than ref. 17 and 38 is applied for the whole study, with corrected degrees of freedom using time series autocorrelation. The p-values of all correlations presented in this study are also based on this test, including Fig 3a that includes tests with low-resolution proxy records. + +<|ref|>text<|/ref|><|det|>[[145, 510, 881, 599]]<|/det|> +60. Smith, D. M., Booth, B. B. B., Dunstone, N. J., Eade, R., Hermanson, L., Jones, G. S., Scaife, A. A., Sheen, K. L., Thompson, V. Role of volcanic and anthropogenic aerosols in the recent global surface warming slowdown. Nat. Clim. Change. 6, 936-940 (2016). + +<|ref|>text<|/ref|><|det|>[[145, 615, 880, 670]]<|/det|> +61. Hotelling, H. The relations of the newer multivariate statistical methods to factor analysis. British Journal of Statistical Psychology. 10, 69-76 (1957). + +<|ref|>text<|/ref|><|det|>[[145, 685, 880, 738]]<|/det|> +62. Wold, H. System analysis by Partial Least Squares. IIASA Collaborative paper. IIASA Laxenburg. (1983) + +<|ref|>text<|/ref|><|det|>[[145, 755, 879, 808]]<|/det|> +63. Zou, H., Hastie, T. Regularization and variable selection via the elastic net. J. Roy. Stat. Soc. 67, 301-32 (2005). + +<|ref|>text<|/ref|><|det|>[[145, 824, 881, 878]]<|/det|> +64. Stone, M. Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society. 36, 111-147 (1974). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 84, 883, 103]]<|/det|> +877 65. Geisser, S. The predictive sample reuse method with applications. Journal of the Royal + +<|ref|>text<|/ref|><|det|>[[50, 119, 496, 137]]<|/det|> +878 Statistical Society. 70, 320- 328 (1975). + +<|ref|>text<|/ref|><|det|>[[50, 153, 95, 520]]<|/det|> +879 880 881 882 883 884 885 886 887 888 889 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 141, 70]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[40, 90, 861, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 504, 115, 523]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[40, 545, 952, 750]]<|/det|> +Climatic impacts of the AMV over the historical era. a) Historical evolution of AMV indices investigated in this study for the period 1870- 2017 calculated using the HadISST dataset36 (cf. Methods). b,d,f,h,j) Map of averaged regression coefficients between the 10- years smoothed composite of the three AMV indices from a) and CRUTS430 precipitation data for the period 1901- 2017. Maps are respectively relating Annual, DJF, MAM, JJA and SON regression coefficients. c,e,g,i,k) Map of regression coefficients between the composite of the three AMV indices from a) and CRUTS430 surface temperature data for the period 1901- 2017. Maps are respectively relating Annual, DJF, MAM, JJA and SON regression coefficients. For c- e), white grid points indicate that regression coefficients are not significantly different than 0 at the \(90\%\) confidence level, using a two- tailed student test with corrected degrees of freedom17,38 (Methods). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[40, 48, 861, 437]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 456, 117, 475]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[39, 498, 951, 701]]<|/det|> +Nested reconstruction of the AMV and related proxies. a) Black line: Annually- resolved nested reconstruction of AMV_F ( \(^\circ \mathrm{C}\) ) using random forest40 (cf. Methods). Red line: 10 years kernel smooth of the annually- resolved nested reconstruction (black line). The regression uncertainties of the annually- resolved nested reconstruction (black line) are defined for each time step of the nested reconstruction as \(\pm 2\) standard error of the regression. Green line is the time series of the instrumental AMV_F calculated from historical SST data30 b) Validation metrics (CE in yellow and correlation in orange) obtained for 30 training- testing splits, and proxy records types availability for the nested AMV_F reconstruction (bottom). c) Proxies weights from the random forest method, relative to the proxy records temporal availability (see Methods) d) Temporal coverage of the availability of the proxy records. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[42, 42, 633, 494]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 520, 116, 538]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 560, 955, 627]]<|/det|> +Comparison with independent ocean proxy records. a) 37 Ocean 2k proxy records35 significantly correlated at least at the \(90\%\) confidence with the AMV over the pre- industrial period (i.e., prior to 1870). b) 10- years (black) and 30- years (blue) kernel smooth of the AMV reconstruction. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[62, 45, 780, 472]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 501, 116, 520]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 542, 951, 632]]<|/det|> +Proxy to model PPE validation. a) CE scores (yellow boxplots), correlation scores (orange boxplots) and correlation between the model AMV and the reconstructed AMV within the model simulations (Methods) (purple line) for 12 members of CESM1- LME. b) Weights of the proxy records from the model simulations from 12 members of CESM1- LME (Methods). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 52, 644, 592]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 620, 120, 639]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[41, 660, 950, 842]]<|/det|> +Model to proxy PPE validation. a) CE scores (yellow boxplots), correlation scores (orange boxplots) and correlation between the real- world AMV_F reconstruction and model- constrained44 AMV_F reconstructions (Methods) (purple line) for the 12 members of CESM1- LME. Green line indicates the fraction of proxy records from the real experiments used in the PPE experiment (see Methods) b) Weights of the proxy records from the real- world experiment for RF trained within the model simulations (Methods). c) Grey lines: 10- years kernel smooth of the 12 model- based experiment, based on each CESM1- LME member. Black: 10- years kernel smooth of the ensemble average of the 12 model- based reconstructions. Red: 10- years kernel smooth of the reconstruction from the real experiment. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[40, 52, 760, 545]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 572, 116, 591]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[40, 613, 949, 749]]<|/det|> +Comparison of the AMV with NASST and volcanic forcing. a) Final reconstructions of AMV and NASST. b) Superposed epoch analysis47 for responses of the AMV and NASST reconstructions to the ten largest eruptions46 of the last millennium (see Extended Data Table 4). Composite series are performed for 31 years, for which the 11th are the actual years of the eruptions. Each individual response is centered to its values 10 years before the eruption (from N- 10 to N- 1) before computing the composite time series. 95% confidence levels have been calculated using a Monte- Carlo approach47. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 48, 780, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 588, 117, 608]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[40, 629, 955, 833]]<|/det|> +Spectral analysis and early warning signal. a) Discrete wavelet transform of the nested AMV reconstruction from this study. Contours provide a \(90\%\) confidence level. The white line and the light white- shaded area below indicate the cone of influence. The cone of influence gives the spectrum borders where the edge effect (i.e., the time boundary effect) becomes too important, which cannot be robustly interpreted. b) Early warning signal test50- 52 of the nested AMV_F reconstruction (Methods) based on AR1 coefficients, for different window lengths (WL). For each WL, sliding AR1 coefficient are computed and a Kendall \(\tau\) statistics between time and the sliding AR1 time series are calculated. Significance is approximated using Gaussian distributions because of the large length (>50) of the AR1 coefficients (see Methods). + +<|ref|>sub_title<|/ref|><|det|>[[44, 856, 311, 882]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 906, 764, 925]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 45, 543, 333]]<|/det|> +ExtendedDataFig1. docxExtendedDataFig2. docxExtendedDataFig3. docxExtendedDataFig4. docxExtendedDataFig5. docxExtendedDataTable1. docxExtendedDataTable2. docxExtendedDataTable3. docxExtendedDataTable4. docxSupplementaryInformationREFSForExtDataTab1. pdfSupplementaryInformationRegressionMethods. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/images_list.json b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..391e3c67af15e2356973adc64649b4de50ac8189 --- /dev/null +++ b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/images_list.json @@ -0,0 +1,244 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 57, + 66, + 920, + 393 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 60, + 115, + 420, + 570 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 70, + 78, + 430, + 732 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 57, + 68, + 784, + 455 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 58, + 70, + 420, + 484 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Extended Data Figure 2", + "footnote": [], + "bbox": [], + "page_idx": 46 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Extended Figure 3", + "footnote": [], + "bbox": [ + [ + 60, + 81, + 320, + 500 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "C", + "footnote": [], + "bbox": [ + [ + 333, + 87, + 602, + 201 + ] + ], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Extended Data Figure 4", + "footnote": [], + "bbox": [], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Extended Data Figure 5", + "footnote": [], + "bbox": [], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Extended Data Figure 6", + "footnote": [], + "bbox": [ + [ + 60, + 65, + 350, + 216 + ] + ], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Extended Data Figure 7", + "footnote": [], + "bbox": [], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Extended Data Figure 8. Locations and spread of bilateral ChR2 injections to the LPO in rats used for in vivo optogenetic experiments.", + "footnote": [], + "bbox": [ + [ + 60, + 75, + 571, + 440 + ] + ], + "page_idx": 50 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 53, + 92, + 920, + 423 + ] + ], + "page_idx": 51 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 60, + 92, + 420, + 551 + ] + ], + "page_idx": 51 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 44, + 45, + 333, + 540 + ] + ], + "page_idx": 52 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 57, + 92, + 714, + 440 + ] + ], + "page_idx": 53 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 57, + 46, + 380, + 415 + ] + ], + "page_idx": 56 + } +] \ No newline at end of file diff --git a/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e.mmd b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e.mmd new file mode 100644 index 0000000000000000000000000000000000000000..dc30d62f70f4285291072bb3e63f451286f35b2d --- /dev/null +++ b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e.mmd @@ -0,0 +1,697 @@ + +# A diencephalic circuit for opioid analgesia but not positive reinforcement + +Maggie Wang University of California, San Francisco + +Kayla Maanum University of California, San Francisco + +Joseph Driscoll University of California, San Francisco + +Chris O'Brien Rutgers University + +Svetlana Bryant Rutgers University + +Kasra Mansourian University of California, San Francisco + +Marisela Morales National Institute on Drug Abuse https://orcid.org/0000- 0002- 3845- 9402 + +David Barker Rutgers University + +Elyssa Margolis ( Elyssa.Margolis@ucsf.edu ) University of California, San Francisco https://orcid.org/0000- 0001- 8777- 302X + +## Article + +Keywords: Mu opioid receptor agonists, epithelialic lateral habenula, opioid analgesia + +Posted Date: January 6th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 125555/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on February 9th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28332- 6. + +<--- Page Split ---> + +1 A diencephalic circuit for opioid analgesia but not positive reinforcement + +2 + +3 Maggie W. Waung \(^{1}\) , Kayla A. Maanum \(^{1}\) , Joseph R. Driscoll \(^{1}\) , Chris O'Brien \(^{2}\) , Svetlana Bryant \(^{2}\) , Kasra + +4 Mansourian \(^{1}\) , Marisela Morales \(^{3}\) , David J. Barker \(^{2,3}\) , and Elyssa B. Margolis \(^{1*}\) + +5 + +6 'UCSF Weill Institute for Neuroscience, Department of Neurology, University of California, + +7 San Francisco, CA, United States + +8 'Department of Psychology, Rutgers University, New Brunswick, NJ, United States + +9 'National Institute on Drug Abuse, Neuronal Networks Section, National Institutes of Health, Baltimore, + +10 MD, United States + +11 + +12 \*Corresponding Author + +13 †Email: Elyssa.Margolis@ucsf.edu + +14 + +15 Title- 10; Abstract- 149, Introduction- 316, Results- 2574, Discussion- 1326; Figures 5, 8 extended data + +16 figures; Tables- 1 Supplementary Table + +<--- Page Split ---> + +## 21 Abstract + +21 AbstractMu opioid receptor (MOR) agonists are the most effective analgesics, but their use risks respiratory depression and addiction. The epithelialal lateral habenula (LHb) is a critical site that signals aversive states, often via indirect inhibition of reward circuitry, and MORs are highly expressed in the LHb. We found that the LHb is a potent site for both MOR- agonist analgesia. Strikingly, LHb MOR activation generates negative reinforcement but is not rewarding in the absence of noxious input. While the LHb receives inputs from multiple sites, we found that inputs from the lateral preoptic area of the hypothalamus (LPO) are excited by noxious stimulation, express MOR mRNA, and are preferentially targeted by MOR selective agonists. Critically, optogenetic stimulation of LHb- projecting LPO neurons produces an aversive state relieved by LHb MOR activation. Therefore targeting this MOR sensitive forebrain circuit can relieve pain yet lower the risk of misuse by pain free individuals. + +## 33 Introduction + +33 IntroductionOpioids are the most effective pain medications, but the risk of overdose and opioid use disorder limits their clinical utility. Uncoupling the analgesic actions of opioids from those that underlie positive reinforcement is a longstanding goal for pharmacotherapeutic development. Identifying circuits that can drive relief of ongoing pain but not reward in the absence of pain is a critical step towards this goal. The lateral habenula (LHb) may participate in such a circuit, which is not only activated in a pain setting1–5 but also by other aversive states including reward omission6, and animal models of depression7. Several CNS sites involved in pain signaling with reported strong inputs to the LHb include the lateral hypothalamus (LH)8,9 and anterior cingulate cortex (ACC)10–12. Furthermore, efferents from the LHb target pain-responsive regions including the lateral periaqueductal gray, dorsal raphe, and parabrachial nucleus13,14. While morphine injections that covered a combination of the LHb, medial habenula, and posteromedial + +<--- Page Split ---> + +thalamus reduce pain- related behavior in an acute pain model15, whether these effects are due to MOR activation specifically in the LHb is an open question. + +Since increased activity in LHb neurons encodes aversive states including ongoing pain, inhibition of this activity should relieve pain and generate negative reinforcement. We previously found that MOR activation can decrease neural activity in the LHb via both postsynaptic hyperpolarization and inhibition of glutamate release onto subsets of LHb neurons in naïve animals16. Here we investigated the specific LHb input circuit and synaptic mechanism by which MOR activation in the LHb produces pain relief. Among six potential inputs to the LHb, we determined that the glutamatergic innervation from the lateral preoptic area of the hypothalamus (LPO) is both pain- responsive and most strongly inhibited by MOR activation. Importantly, we show that activating MORs in this circuit in the absence of pain does not produce reinforcement, suggesting that targeting this circuit could be a significant advance in pain therapeutics. + +## Results + +## MOR activation in the LHb produces pain relief but not positive reinforcement + +To examine the behavioral impact of selective MOR activation in the LHb on ongoing pain, we used the spared nerve injury (SNI) model of persistent neuropathic pain and implanted bilateral cannulae above the LHb in Sprague Dawley rats (Fig. 1a). After recovery from surgery, we evaluated allodynia by measuring mechanical stimulation thresholds with graded von Frey filaments. Bilateral microinjections of the MOR- selective agonist DAMGO (10 \(\mu \mathrm{M}\) ; 300 nL/hemisphere) into the LHb increased the average hindpaw withdrawal threshold compared to saline microinjections in the same animals, indicating that DAMGO reduced the mechanical allodynia generated by SNI in male rats (Fig. 1b). In contrast, intra- LHb DAMGO microinjections in sham- injured male rats had no effect on mechanical withdrawal thresholds compared to saline (Fig. 1b). Consistent with the rat literature17, we did not observe a significant decrease + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +Figure 1. MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500 \mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10 \mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . SNI males: Wilcoxon signed rank test, \(\mathrm{V} = 2\) , \(\mathrm{p} = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with \(\mathrm{SNI} (n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.53\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 31\) , \(\mathrm{p} = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra- LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 15.93\) , \(\mathrm{p} = 0.002\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.01\) ; paired t-tests, sham adjusted \(\mathrm{p} = 1\) ; SNI adjusted \(\mathrm{p} = 0.016\) . Female SNI animals trended towards a preference for the DAMGO- paired chamber: Paired t-test \(\mathrm{p} = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two- way mixed ANOVA, \(\mathrm{F}(1,19) = 2.239\) , \(\mathrm{p} = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100 \mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 5\) , \(\mathrm{p} = 1\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 6\) , \(\mathrm{p} = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100 \mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two- way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 6.234\) , \(\mathrm{p} = 0.027\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.07\) ; paired t-tests, sham adjusted \(\mathrm{p} = 0.48\) ; SNI adjusted \(\mathrm{p} = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} < 0.01\) + +<--- Page Split ---> + +in withdrawal latency to heat in the Hargreaves test after SNI compared to sham- injured controls, and DAMGO microinjections into the LHb did not alter heat withdrawal latency compared to saline in SNI or in sham animals (Extended Data Fig. 1a). Intra- LHb DAMGO also reversed mechanical allodynia induced by inflammatory pain in the CFA model (Extended Data Fig. 1b- g). + +To evaluate whether MOR activation in the LHb influences the affective experience of pain, we used the place conditioning paradigm in the same group of rats. In a three- chamber apparatus, we paired intra- LHb DAMGO microinjections with one chamber and saline microinjections with the opposite side chamber. Rats with SNI developed a significant conditioned place preference (CPP) for the LHb- DAMGO- paired chamber, while sham- injured rats did not prefer either chamber following conditioning (Fig. 1c). + +To rule out the potential confound of off- target effects due to DAMGO entering the CSF space via the nearby third ventricle, we microinjected the same solutions intracerebroventricularly (i.e.v.) in male rats with SNI or sham injury. This manipulation did not influence mechanical withdrawal thresholds compared to saline microinjection in either group (Fig. 1b). Furthermore, i.e.v. DAMGO microinjections did not generate a CPP in either SNI or sham animals (Fig. 1c). Therefore, we conclude that the behavioral effects of our DAMGO microinjections were due to actions specifically in the LHb. + +We also investigated whether LHb MOR activation had the same effects on allodynia and affective pain in female rats. Using the same microinjection parameters as in male rats, female rats with SNI showed a trend towards reduced mechanical allodynia following DAMGO microinjections into the LHb compared to saline (Fig. 1b). Female rats with SNI also showed a trend towards a preference for the DAMGO- paired chamber (Fig. 1c). As in males, females with SNI displayed no difference in heat withdrawal latency between DAMGO and saline microinjections (Extended Data Fig. 1a). Because female rats may be less sensitive to the analgesic effect of opioids18,19, we tested a 10- fold higher concentration of intra- LHb DAMGO (100 μM) in a separate cohort of females. This dose was chosen based on studies demonstrating a three- fold reduction in i.e.v. DAMGO efficacy for females in the tail flick test20. The + +<--- Page Split ---> + +higher DAMGO dose induced a significant CPP, though still did not reverse the mechanical allodynia in female rats (Fig. 1d, e). Thus, we conclude that LHb MOR activation can reverse the affective experience of pain without raising the baseline threshold for reflex withdrawal in both male and female rats. + +## MOR synaptic function persists in LHb neurons in animals with chronic pain + +In various CNS regions, chronic pain induces changes in MOR expression and function, including downregulation21- 23. We previously reported that in naïve male rats MOR activation inhibits glutamate release onto a subset of LHb neurons and also hyperpolarizes approximately \(30\%\) of LHb neurons16. Here we examined whether these MOR effects are altered in a persistent pain state. To evaluate postsynaptic MOR function we performed whole cell voltage clamp recordings of LHb neurons from acute brain slices from male rats with SNI and measured DAMGO induced changes to holding current (Fig. 2a, b). These responses did not differ from our observations in naïve rats. We also tested MOR inhibition of glutamatergic electrically- evoked excitatory postsynaptic currents (EPSCs) in LHb neurons from animals with SNI. These responses were also consistent with observations from naïve rats (Fig. 2c, d). We conclude that these actions of MOR on LHb cell bodies and glutamatergic terminals do not change in animals with ongoing pain. + +Increased activity in LHb neurons as well as increased glutamatergic synaptic strength onto LHb neurons are associated with aversive behavioral states7,24,25. To evaluate this in LHb neurons from animals with SNI, we first measured the paired pulse ratio in the evoked EPSCs, a measure of probability of release. There was no difference in paired pulse ratio between groups (Fig. 2e). We next compared the frequency and magnitude of spontaneous glutamatergic EPSCs (sEPSCs) in animals with SNI to those in naïve animals. Mean sEPSC frequency and amplitude were also similar in LHb neurons from SNI and naïve animals (Fig. 2f, g). Together, these observations suggest a lack of synaptic plasticity induced on glutamatergic inputs to LHb neurons by painful injury. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Figure 2. Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(\mathrm{V_m} = - 60\) \(\mathrm{mV}\) . a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 32\) , \(\mathrm{t} = 0.892\) , \(\mathrm{p} = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 21\) , \(\mathrm{t} = 0.137\) , \(\mathrm{p} = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t-test, \(\mathrm{df} = 24\) , \(\mathrm{t} = 0.102\) , \(\mathrm{p} = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t-test unequal variances, \(\mathrm{df} = 37\) , \(\mathrm{t} = -0.17\) , \(\mathrm{p} = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t-test unequal variances, \(\mathrm{df} = 38\) , \(\mathrm{t} = -0.09\) , \(\mathrm{p} = 0.93\) . Data from naïve rats previously published in \(^{16}\) . + +<--- Page Split ---> + +The LHb receives functional synaptic input from the LPO, LH, VTA, VP, and EPN, but not the ACC. + +The LHb receives functional synaptic input from the LPO, LH, VTA, VP, and EPN, but not the ACC.LHb neuron firing activity increases with acute noxious stimulation, and an increase in ongoing firing frequency is during aversive behavioral states2,3,7. In ongoing pain, since we did not observe evidence for changes in glutamatergic synaptic strength, such increases in firing may be driven by greater activity in the glutamatergic axons innervating LHb neurons. Therefore, we hypothesized that the intra- LHb DAMGO- induced behavioral effects that we observed in injured animals were due to MOR inhibition of glutamatergic axon terminals, thus decreasing the aversive excitatory drive onto the LHb neurons. Since MOR activation only inhibits glutamatergic inputs onto a subset of LHb neurons16, MORs might be preferentially expressed on specific afferent inputs. Prior work characterizing direct functional synaptic connections to the LHb is limited to mice26, therefore first we sought to confirm these functional connections in the rat. We investigated inputs from the entopeduncular nucleus (EPN), lateral preoptic area of the hypothalamus (LPO), and ventral tegmental area (VTA) because stimulating glutamatergic LHb inputs from these sources has been shown to be aversive24,27,28. We also investigated inputs from the LH and ACC that are strongly implicated in pain processing29 and the VP because stimulating glutamatergic VP neurons increases the firing rate of LHb neurons30. We injected AAV2-hSyn-hChR2(H134R)-mCherry into one of these six regions in order to express channelrhodopsin (ChR2) in these different input populations (Fig. 3a). We then made whole cell recordings in LHb neurons and measured light-evoked synaptic inputs while blind to injection site. ChR2 was activated by an LED (\(\lambda = 473\)) coupled to an optic fiber placed approximately \(100 \mu \mathrm{m}\) from the recorded cell. Post synaptic currents (PSCs) were measured in response to paired light pulses (1 or 5 ms, 50 ms inter stimulus interval) at holding potentials of -60 mV and -40 mV to probe for EPSCs and GABAAR mediated inhibitory PSCs (IPSCs) in each cell, respectively. Roughly similar proportions of LHb neurons received synaptic input from each of these targets, with the exception of the ACC, where we did not detect any fast PSC connections (Fig. 3b, topographical distribution of connected neurons in Extended Data Fig. 2). The absence of a functional synaptic input from the ACC to the LHb was surprising, as both anterograde10,12 + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Figure 3. Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2-hSyn-hChR2(H134R)-mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venom diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically-evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One-way ANOVA, \(\mathrm{df} = 4\) , \(\mathrm{F} = 4.11\) , \(\mathrm{p} = 0.0057\) followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically-evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light-evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was \(< 2\) ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically-evoked LPO-LHb EPSCs ( \(\mathrm{n} = 12\) ). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more GABAergic connections than glutamate connections. (Right) DAMGO inhibited light-evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +<--- Page Split ---> + +and retrograde tracers11 have previously demonstrated modest inputs. Moreover, the ACC is extensively implicated in behavioral responses to pain and MOR- agonist induced pain relief31,32. As a secondary measure of the strength of the innervation, we performed a systematic evaluation of the potential connection using ChR2 as an anterograde tracer, making large injections of AAV2-hSyn-hChR2(H134R)- mCherry throughout the anteroposterior range of the ACC (Extended Data Fig. 3a). This tracing revealed extensive innervation of the nearby mediodorsal thalamus (MDL), but minimal stereologically- quantified labeling in the LHb (Extended Data Fig. 3b, c). Therefore, while we cannot completely rule out a functional input from the ACC to the LHb, the innervation is extremely small compared to the other sources of input to the LHb investigated here. + +For each brain region from which fast synaptic PSCs were detected in the LHb, both glutamate and GABA inputs were observed in varying proportions (Fig. 3c). Interestingly, while for each input there were individual LHb neurons that received both glutamate and GABA synaptic connections, for each input more than half of the connected LHb neurons received just one type of fast PSC, in varying proportions. The VTA was the only input where more LHb neurons received GABAergic synaptic connections than glutamatergic synaptic connections. Among observed synaptic connections, a wide range of EPSC and IPSC amplitudes were observed for most of the inputs, except for the glutamatergic inputs from the VTA that were consistently small (Extended Data Fig. 4a,c). The delay to light evoked EPSC onset also varied across input source, with LH inputs having the shortest mean latency (Extended Data Fig. 4b). + +The nature of local LHb neural connections will also impact the circuit's response to MOR activation. There are strong local glutamatergic connections within the LHb16,33, but there is recent evidence both for34,35 and against36,37 the existence of local GABA interneurons. As we only observed somatodendritic MOR responses in a subset of LHb neurons, evidence for GABAergic interneurons in the LHb would impact our model of how MOR activation modulates LHb neural activity, if such interneurons preferentially express the MOR. In the rat, a small number of GAD1 positive neurons are present in the + +<--- Page Split ---> + +lateral LHb, though these neurons do not co- express vesicular GABA transporter36, the protein required for loading GABA into synaptic vesicles. In order to detect functional local GABAergic connections within the rat LHb, we injected AAV2- hSyn- hChR2(H134R)- mCherry into the LHb and recorded from LHb neurons. Because some recorded neurons expressed ChR2, we measured light responses before and after application of receptor antagonists in order to isolate the synaptically driven response from the ChR2 mediated currents. Under these conditions we did not observe any light activated local IPSCs in LHb neurons (Extended Data Fig. 5). As expected, many neurons received local glutamatergic inputs (Extended Data Fig. 5). We conclude that there is very limited or no local GABAergic interneuron connectivity in the LHb of adult rats. + +## MOR activation most strongly inhibits LPO inputs to the LHb + +Next, we tested for functional MOR modulation of the light evoked glutamatergic inputs to LHb neurons from each of the regions characterized above. DAMGO induced the strongest and most consistent inhibitions in the terminals arising from LPO neurons (Fig. 3d, e). On average the inhibition was greater in these LPO inputs than the MOR impact on ESPCs observed from the LH, VTA, VP, or EPN (Fig. 3d). We also tested whether MOR inhibits LPO glutamatergic inputs to the LHb independent of sex; the mean inhibition of glutamate release from LPO terminals to LHb neurons in female rats was equivalent to that observed in males (Fig. 3f). Because of the prevalence of local glutamatergic connections in the LHb16,33 (Extended Data Fig. 5) and postsynaptic MOR inhibition of a subset of LHb neurons16, we sought to rule out a polysynaptic connection. First, a polysynaptic contribution seems unlikely for all of the glutamatergic inputs reported here because the delay from light pulse onset to EPSC onset was consistently less than 3 ms (Extended Data Fig. 4b). Second, to directly test isolated monosynaptic connections, we expressed ChR2 in LPO neurons and recorded in the LHb; in neurons with light evoked EPSC responses, we applied tetrodotoxin (TTX, 500 nM) and 4- aminopyridine (4 AP; 10 μM). In 8 of 8 tested neurons the light evoked EPSCs persisted in this monosynaptic signal sparing preparation, and this monosynaptic response was inhibited by DAMGO in all 5 tested neurons (Extended Data Fig. 6). + +<--- Page Split ---> + +Therefore, we conclude that MORs are functionally expressed on LPO terminals that monosynaptically contact LHb neurons, and when these glutamatergic inputs are activated in vivo, DAMGO application should inhibit them. + +## MOR mRNA is enriched in LHb-projecting LPO neurons + +Multiple basal forebrain structures express high levels of MOR including the VP, medial preoptic area (MPO), horizontal diagonal band (HDB), ventral bed nucleus of the stria terminalis (vBNST), and other regions of the extended amygdala complex (EAC) \(^{38}\) . Some of these not only express MOR to a greater extent than the LPO, but they also project to the LHb. To further evaluate the specificity of MOR expression in LHb-projecting neurons in the LPO compared to nearby brain regions, we performed in situ hybridization for MOR mRNA (OPRM1) in brain slices from Sprague Dawley rats where the retrograde tracer Fluoro- Gold had been iontophoresed into the LHb (Fig. 4a - c). With this independent approach, the LHb-projecting LPO neurons showed the strongest OPRM1 expression and contained the greatest number of retrogradely labeled FG(+) neurons co- labeled for OPRM1 ( \(112 \pm 9\) cells), corresponding to \(57.1\% \pm 4.3\%\) of all MOR(+)FG(+) neurons in the basal forebrain (336/602 total cells). Inputs were also observed from the HDB ( \(23.6\% \pm 5.9\%\) of MOR(+) cells; 153/602 total cells), VP ( \(9.5\% \pm 3.1\%\) of MOR(+) cells; 52/602 total cells), MPO ( \(3.0\% \pm 1.1\%\) of MOR(+) cells; 19/602 total cells), vBNST ( \(3.7\% \pm 1.2\%\) of MOR(+) cells; 23/602 total cells) and EAC ( \(3.0\% \pm 2.1\%\) MOR(+) cells; 19/602 total cells; Fig. 4d - f). Overall, these anatomical data are highly consistent with our electrophysiology results, supporting the conclusion that the LPO projection to the LHb is strongly regulated by MORs. + +## Noxious stimulation activates glutamatergic LHb-projecting LPO neurons + +We next tested whether glutamatergic LHb- projecting LPO neurons are activated by noxious stimulation and whether ongoing pain alters this response. We expressed the calcium indicator GCaMP6m in LHb- projecting LPO neurons using a Cre- dependent, retrograde viral construct HSV- hEF1α- LS1L- GCaMP6m injected to the LHb of VGluT2::Cre mice (Fig. 5a). We implanted optic fibers above the LPO in control + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Figure 4. LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb (n = 3 rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar = 20 μm. (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5
+ +Figure 5. LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV-hEF1α- LSI1-GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb-projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI \((n = 5)\) exhibited significantly larger changes in GCaMP6m fluorescence time-locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls \((n = 9)\) : Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(\mathrm{p} = 0.038\) ; Holm-Sidak post-hoc test, \(\mathrm{p} = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2-mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intra- LHb saline or DAMGO. d, Representative images of Cre-dependent ChR2-mCherry fluorescence in LPO cell bodies. (Left) Scale bar = \(250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber (“o.f.”) implant amidst ChR2-mCherry-expressing cell bodies. Scale bar = \(50 \mu \mathrm{m}\) . e, Only animals with active ChR2 \((n = 8)\) , but not mCherry controls \((n = 8)\) , developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2-mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t-tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2-mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t- tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2- mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +<--- Page Split ---> + +and SNI mice and performed fiber photometry prior to and during the Hargreaves task. On average, GCaMP6m signal increased as paw withdrawal commenced, and this response was greatly potentiated in mice with SNI (Fig. 5b), who also displayed shorter withdrawal latencies to heat (Extended Data Fig. 7). Thus, glutamatergic LPO neurons that project to the LHb are activated in response to noxious peripheral stimulation, and the magnitude of activation is higher during ongoing pain. + +## LHb MOR activation blocks the aversiveness of LPO-LHb stimulation + +Because LHb- projecting LPO glutamatergic neurons are activated by noxious stimulation, stimulating these terminals in the LHb is aversive28, and MOR activation inhibits these terminals in the LHb, we hypothesized that the aversiveness produced by stimulating the LHb- projecting LPO neurons should be reduced by MOR agonist injection into the LHb. To selectively express ChR2 in LHb- projecting LPO neurons, we used an intersectional viral approach in rats, injecting the retrograde CAV- Cre into the LHb and AAV2- EF1α- DIO- hChR2(H134R)- mCherry (or AAV2- EF1α- DIO- mCherry for controls) into the LPO bilaterally (Fig. 5c,d). Bilateral optic fibers were subsequently implanted above the LPO and cannulae were implanted above the LHb (Fig. 5c). During place conditioning sessions, rats received blue light activation (473 nm, 20 Hz, 5 ms, 10- 12 mW) of the LHb- projecting LPO neurons in both environments of the place conditioning apparatus; chambers were paired with either intra- LHb DAMGO or saline (Fig. 5c). Rats with ChR2 expression developed a CPP for the chamber associated with DAMGO, while control mCherry expressing animals did not (Fig. 5e). Therefore, MOR activation in the LHb blocked the aversiveness of stimulating the LPO input to the LHb without producing positive reinforcement in the absence of input stimulation. + +## Discussion + +Here we identified a circuit that can be targeted by MOR agonists to relieve the aversiveness of ongoing pain but does not produce reward in pain free rodents. MOR agonist action in the LHb was sufficient to + +<--- Page Split ---> + +reverse injury- induced allodynia and to produce a CPP, decreasing both sensory and affective pain responses, respectively. Importantly, sham- injured animals did not develop a CPP, indicating that MOR activation in the LHb does not produce positive reinforcement in the absence of pain. Unexpectedly, rather than inputs from brain regions previously established as mediating pain and opioid- induced pain relief, it is the glutamatergic inputs from the LPO that we demonstrate here are preferentially controlled by LHb MOR activation. We found that the aversiveness produced by optogenetic activation of LHb- projecting LPO neurons is blocked by MOR agonist microinjection into the LHb, showing that a MOR inhibitory action on this circuit is sufficient to relieve pain. Together, these experiments show that MOR activation in the LHb can generate negative reinforcement via pain relief, but not positive reinforcement in the absence of noxious stimuli. + +## The LHb in pain and relief + +The LHb plays a role in the perception of noxious stimuli in injury and depression models39,40, and most LHb neurons fire more in aversive behavioral states and in response to noxious stimuli2,3 Bilateral lesions of the LHb decrease allodynia in the chronic constriction ischemia model of neuropathic pain in rats41, supporting the notion that signaling through the LHb contributes to injury- induced mechanical allodynia. Chemogenetic inhibition of the LHb relieves thermal hyperalgesia in animals undergoing alcohol withdrawal39. We found that MOR activation in the LHb reverses mechanical allodynia in a model of neuropathic pain in both males and females. Together these observations support our conclusion that MOR activation impacts the behavioral state of the animal by decreasing LHb neural activity in vivo. We previously found two potential mechanisms by which MOR activation could inhibit LHb neural activity in animals with ongoing pain: presynaptically inhibiting glutamate release and postsynaptically driving an outward current in a subset of LHb neurons16. Here we found that intra- LHb MOR activation induces negative reinforcement in both male and female rats with ongoing pain. Yet sham controls (male or female) did not develop a CPP to intra- LHb DAMGO. Because the behavioral effect of MOR activation + +<--- Page Split ---> + +was specific to injured rats we hypothesized that increased glutamatergic drive to the LHb, in an input that is MOR sensitive, is a key element in the LHb circuit dynamics involved in pain and relief. + +The vast majority of LHb neurons are glutamatergic, and there is anatomical evidence for dense boutonlike structures arising from local LHb neurons42. This local feed forward connectivity may distribute a specific excitatory LHb input across the various LHb projections that include the dorsal raphe, ventrolateral PAG, MDL, centromedian thalamus, LH, RMTg, and VTA9,13,14,43. Local connectivity has also been observed with functional assays, including that TTX application decreases the frequency of sEPSC and sEPSP events in a subset of LHb neurons16,33 and that optogenetic activation of LHb neurons induces glutamatergic EPSCs (Extended Data Fig. 5). This feed forward circuit enables the distribution of an afferent excitatory signal across LHb neurons, and it raises the possibility that inhibition of such an input by a MOR agonist will decrease activity in LHb projection neurons, even those that are not directly innervated by the excited pathway. Therefore, it is possible that MOR inhibition of one specific input, such as the LPO, can decrease the excitatory drive onto many efferent LHb projections. + +We characterized the functional glutamatergic connections from a variety of brain regions to the LHb in rats; our findings were largely consistent with prior reports utilizing similar techniques in mice. Still, we note that outcomes of these experiments are dependent on the types and number of neurons that express ChR2 following the virus injections and are limited to the geometry of an injection site. We detected the strongest glutamatergic inputs from the LH and EPN, while glutamatergic synaptic responses from the VTA were quite small. We found no functional inputs from the ACC, and in a systematic anatomical analysis we detected very few afferent fibers from the ACC in the LHb. Of the detected glutamatergic inputs, we expected brain regions associated with pain perception and pain relief, such as the LH and VTA, to be more strongly modulated by MOR activation. However, we instead found MOR mRNA and function were clearly enriched in LPO inputs to the LHb. + +<--- Page Split ---> + +## The LPO: a brain region contributing to pain perception + +Because LHb neurons generally fire more in response to noxious stimuli, and optogenetic stimulation of various glutamatergic inputs to the LHb is uniformly aversive \(^{25,27,28,44,45}\) , we hypothesized that a glutamatergic input to the LHb transmits the pain signal. Further, aversive stressors lead to an increase in the ratio of excitatory glutamatergic to inhibitory GABAergic in synaptic input to LHb neurons \(^{44,45}\) , and restoration of this ratio is associated with relief of aversive states such as foot shock- induced learned helplessness \(^{45}\) and cocaine withdrawal \(^{44}\) . Therefore, a pharmacological manipulation that decreases the excitatory drive onto LHb neurons should also relieve aversive states. Because we found that DAMGO in the LHb generates CPP only in animals with ongoing pain, our data suggest a MOR- sensitive glutamatergic input to the LHb is active during pain and relatively inactive in the absence of pain. + +The LPO projection to the LHb is composed of neurons releasing either glutamate or GABA; these neurons do not co- release glutamate and GABA \(^{28}\) , unlike other LHb inputs \(^{46}\) . Optogenetic activation of LPO glutamate projections to the LHb is aversive \(^{28}\) , and here we posited that this activation mimics an ongoing pain signal. We were able to relieve the aversiveness of activation of this connection with a MOR agonist in the LHb. While a causal role for the LPO in pain perception is unexplored to date, anterograde \(^{47}\) and retrograde \(^{48}\) tracing has shown a direct input to the LPO from the spinal cord. Also, injections of the pro- inflammatory cytokine IL- 1β into the LPO induces hyperalgesia, indicating LPO participation in a nociception circuit \(^{49}\) . Painful stimuli such as subcutaneous formalin injections, mild electric shock and tail pinch also increase firing in some LPO neurons \(^{50,51}\) . Here we show that a major output for LPO pain signals is to the LHb. + +## The LPO to LHb circuit: A unique target for analgesia + +Opioids remain the best available clinical analgesics, yet ongoing systemically administered opioids can result in the serious adverse consequences including opioid use disorder and respiratory depression \(^{52}\) . Opioid- induced positive reinforcement and euphoria, combined with the development of dependence, + +<--- Page Split ---> + +underlie opioid abuse liability. To date, there is little clinical evidence that the analgesic and euphoric effects of opioids can be decoupled in humans, yet this possibility is a potential pathway to improve therapies for pain. Here we have identified a key circuit whose modulation relieves pain but does not generate reward in the absence of pain. This dissociation between negative and positive reinforcement provides a neural target to achieve pain relief without promoting substance use disorder. Is it possible to activate MORs in the LHb but not in a reward circuit? + +There is some preclinical evidence that this dissociation is possible by developing MOR ligands with the appropriate opioid pharmacology. For example, a novel, cyclized, stabilized MOR selective agonist based on endomorphin I produces pain relief but not reward53. An alternative approach would be to target a different receptor with high expression levels in this circuit in a way that would decrease LHb neural activity. Relevant to this approach, mRNA expression for a selection of orphan G- protein coupled receptors (GPCRs) is enriched in the habenula37,54,55. One such receptor is GPR151, and LHb neurons containing GPR151 receive input from the LPO56. Such directed strategies present a range of new anatomic and molecular targets for pain therapy. By identifying the LPO- LHb connection as a site able to provide relief from neuropathic pain and injury- induced allodynia, we have discovered a unique circuit that may achieve effective opioid- mediated pain relief independent of the drug's addictive properties. + +## Online Methods + +## Animals + +All experiments were performed in accordance to the guidelines of the National Institutes of Health Guide for the Care and Use of Laboratory Animals, and the Institutional Animal Care and Use Committees (IACUC) at the University of California, San Francisco, the National Institute on Drug Abuse (NIDA), and Rutgers University. + +<--- Page Split ---> + +Male and female Sprague Dawley rats were obtained from Charles River Laboratories. Rats were allowed access to food and water ad libitum and maintained on a 12h:12h light/dark cycle. Rats used in behavioral and in situ hybridization studies were housed under reverse light/dark cycle conditions. Rats were group housed until they underwent surgery, after which they were singly housed. + +Male and female VGluT2::Cre mice were bred from mice obtained from the Jackson Laboratory (Jax # 016963). Mice were allowed access to food and water ad libitum and maintained on a 12h:12h light/dark cycle with lights on at 7 AM. Mice were always housed in groups of 2- 5. + +## Viral constructs + +AAV2- hSyn- hChR2(H134R)- mCherry (titer: 2.9e+12), AAV2- hSyn- mCherry (titer: 4.7e+12), and AAV2- EF1α- DIO- ChR2- mCherry (titer: 5.1e+12) were purchased from the University of North Carolina Vector Core with available stock constructs from the laboratory of K. Deisseroth at Stanford University. CAV- Cre (titer: 2.5e+12) was purchased from Montpelier University, France. HSV- hEF1α- GCaMP6m (titer: 5e+9) was purchased from the Gene Delivery Technology Core at Massachusetts General Hospital. + +## Stereotaxic injections + +Rats weighing 275- 300 g were anesthetized with 3- 5% isoflurane (Henry Schein) via inhalation and secured in a stereotaxic frame. Bilateral craniotomies were created with a dental drill above the injection site. For electrophysiology experiments, injections of AAV2- hSyn- hChR2(H134R)- mCherry were made bilaterally into the LPO (- 0.3 mm anteroposterior (AP), \(\pm 1.4 \mathrm{mm}\) mediolateral (ML), - 8.4 mm dorsoventral (DV)), VP (- 0.24 mm AP, \(\pm 2.6 \mathrm{mm}\) ML, - 7.8 mm DV), EPN (- 2.4 mm AP, \(\pm 3.0 \mathrm{mm}\) ML, - 7.0 mm DV), LH (- 2.6 mm AP, \(\pm 1.7 \mathrm{mm}\) ML, - 8.2 mm DV), VTA (- 5.8 mm AP, \(\pm 0.5 \mathrm{mm}\) ML, - 8.5 mm DV), anterior ACC (+2.2 mm AP, \(\pm 0.6 \mathrm{mm}\) ML, - 2.6 mm DV), or posterior ACC (+1.7 AP, \(\pm 0.6 \mathrm{mm}\) ML, - 2.0 mm DV) using a Nanoject II (Drummond Scientific, Broomall, PA). A volume of \(\sim 500 - 830 \mathrm{nL}\) was injected per hemisphere over a period of 4.5 min. The glass injector tip was left in place for at least 2 + +<--- Page Split ---> + +additional minutes before slow withdrawal to prevent backflow and infection of tissue dorsal to the injection target. + +For ChR2 behavior experiments, AAV2- EF1α- DIO- hChR2(H134R)- mCherry or AAV2- EF1α- DIO- + +mCherry were injected into the LPO as above and CAV- Cre was injected bilaterally into the LHb (- 3.7 + +mm AP, \(\pm 0.65 \mathrm{mm} \mathrm{ML}\) , - 5.4 mm DV) with a microinjector connected via polypropylene tubing to + +Hamilton syringes controlled by a dual syringe pump (KD pump) guided by a bilateral 33G stainless steel + +guide cannula (Plastics One). + +Rats were treated with subcutaneous Carprofen (5 mg/kg, Zoetis) and topical \(2\%\) Lidocaine (Phoenix + +Pharmaceutical, Inc.) during the surgery for pain control. After surgery, animals had access to liquid + +Tylenol ( \(\sim 1:40\) ) in their drinking water for 3- 5 days or were administered Meloxicam (s.c. 2mg/kg, + +Pivetal) once per day for two days. + +## Cannulation and optic fiber implantation + +Two to four weeks after virus injection, rats slated for behavioral testing underwent a second cranial surgery to implant custom- made \(200 \mu \mathrm{m}\) optic fibers at a \(5^{\circ}\) angle of rotation in the coronal plane into the bilateral LPO (- 0.6 mm AP, \(\pm 3.05 \mathrm{mm} \mathrm{ML}\) , - 7.5 mm DV). For microinjections into the LHb, bilateral guide cannulae were implanted 1 mm above the LHb (- 3.7 mm AP, \(\pm 0.65 \mathrm{mm} \mathrm{ML}\) , - 4.4 mm DV). A dummy stylet was inserted to maintain patency of the cannulae. Optic fibers and cannulae were anchored with flat point screws and dental cement. For i.c.v. microinjections, unilateral cannulae were implanted into the right lateral ventricle (- 1.0 mm AP, \(+1.5 \mathrm{mm} \mathrm{ML}\) , - 3.5 mm DV). + +Analgesia during surgery and recovery was administered as described above. Animals were allowed to recover for 1- 2 weeks prior to behavioral testing. All virus injections, optic fiber and cannulae placements were histologically verified postmortem based on the standard rat brain atlas57. + +<--- Page Split ---> + +## Spared nerve injury + +Spared nerve injury (SNI) of the sciatic nerve branch was performed to model chronic neuropathic pain17. Under isoflurane anesthesia, a 2- cm skin incision was made over the left hindlimb. The biceps femoris muscle was blunt dissected to expose the branches of the sciatic nerve. The common peroneal and tibial nerves were ligated with 5.0 silk surgical suture and transected distally, with sparing of the sural nerve branch. For sham procedures, a skin incision was made and biceps femoris muscle was exposed without dissection. The overlying skin was closed with a monocryl suture. Animals were allowed seven days to recover from surgery prior to behavioral testing. A majority of animals demonstrated decreased mechanical withdrawal thresholds after SNI; those that did not demonstrate allodynia were not used in further experiments. + +## Inflammatory pain model + +Peripheral inflammation was induced using Complete Freund's Adjuvant (CFA; Sigma Life Science). Under isoflurane anesthesia, a 1:1 emulsion of CFA and sterile saline (150 μL) was injected into the footpad of the rat's left hindpaw with a 27 G needle. Sham- injured controls were injected with sterile saline (150 μL). + +## Microinjections + +Rats were lightly restrained in a cloth wrap for intracranial microinjections. A bilateral 33G microinjector (PlasticsOne) that extended 1 mm ventrally beyond the guide cannula was inserted to target drug delivery into the LHb or i.c.v. Hamilton syringes were driven by a dual syringe pump to infuse either vehicle (phosphate buffered saline, PBS) or DAMGO (10 μM, 300 nL/hemisphere administered over 2 min). A separate cohort of female rats was microinjected with 100 μM DAMGO. + +<--- Page Split ---> + +Standard von Frey sensory assessments were performed as described58. Briefly, rats were habituated to sensory testing chambers (Plexiglass boxes with mesh- like flooring) for at least 2 days prior to testing. Behavioral assessments did not begin until exploratory behavior subsided. Testing was completed with eight Touch Test ® fibers (North Coast Medical & Rehabilitation Products, Gilroy, CA, USA) ranging from 0.4 to 15 g. Fibers were pressed perpendicularly to the mid- plantar left hindpaw with sufficient force to cause bending in the fiber and held for 3- 4 s. A positive response was noted if the paw was sharply withdrawn. Ambulation was not considered a positive response. When responding was ambiguous, testing was repeated. 50% withdrawal thresholds were calculated using fiber application in ascending stiffness order or using the up down method. + +In behavioral pharmacology experiments rats underwent sensory testing 5 min after counterbalanced saline or DAMGO microinjections on the same day, with at least 4 h between infusions. Experimenters were blinded to the solution composition during administration and testing. + +## Hargreaves sensory assessment + +Hargreaves tests were completed in sensory testing chambers with glass flooring. Assessment commenced after rats acclimated to the chamber. Testing was completed using a plantar test analgesia meter (Series 8 Model 390, IITC Life Science, Woodhills, CA, USA). Radiant light was directed toward the mid- plantar left hindpaw until a sharp paw withdrawal response was observed with a 30 s maximum cutoff. Ambulation was not considered a positive response. At baseline testing, measurements were repeated at different intensity levels until the average withdrawal latency of eight trials was \(15 \pm 2\) s. Subsequent measurements were performed using this individualized intensity level. + +## Place conditioning + +Conditioned place preference (CPP) pairings occurred twice daily for four consecutive days following the post- surgery sensory tests with DAMGO and saline. The conditioning apparatus (Med. Associates, + +<--- Page Split ---> + +Georgia, VT, USA) was divided into two chambers (25 cm x 21 cm x 21 cm) with distinct visual (horizontal vs. vertical stripes) and textural (thick vs. thin mesh flooring) cues, separated by a third, smaller gray chamber (12 cm x 21 cm x 21 cm). Before conditioning commenced, animals were allowed up to three opportunities to show neutrality across the chambers during 30- minute baseline sessions. Rats that displayed a consistent baseline preference (>65% of time spent in one chamber) were excluded from the study. Rats were pseudorandomly assigned to receive DAMGO in one of the larger chambers, and assignments were counterbalanced for each cohort. + +During conditioning sessions, microinjections were performed as described above through the intra- LHb cannulae just before the rat was confined to the designated chamber for 30 min. One saline or one DAMGO microinfusion was administered per conditioning session, morning and afternoon pairing sessions were at least 4 h apart, and the order of administration was alternated on each day of conditioning. On test day, rats were allowed to freely explore the chambers for 30 min and time spent (s) in each partition was recorded. Difference score was defined as (Time spent in DAMGO- paired chamber) - (Time spent in saline- paired chamber). + +For ChR2 activation studies, rats were acclimated to handling and attachment of fiber cables to fiber implants in a neutral environment. On procedure days, fiber implants were connected to optic fiber cables attached to a 1x 2 fiber optic rotary joint (Doric Lenses, Quebec, Canada). A laser light source (MBL 473, OEM Laser Systems, East Lansing, MI) was used with light intensity at the end of the output fiber adjusted to 80- 120 mW/mm². Light stimulation (5 ms pulses at 20 Hz) commenced upon placement of rats into the designated chamber of a custom built apparatus on conditioning days. During baseline and testing sessions, the time spent in each chamber was recorded using a webcam and analyzed using Viewer software (Biobserve, Bonn, Germany). + +<--- Page Split ---> + +## Brain removal and immunohistochemistry + +Rats were deeply anesthetized with an intraperitoneal injection of Euthasol (0.1 mg/kg, Virbac Animal Health, Fort Worth, TX) after 0.3 μl of Chicago Sky Blue (in 2% PBS) was injected through the cannulae to mark injection locations. After becoming unresponsive to noxious stimuli, the rats were transcardially perfused with 400 mL of saline, followed by 400 mL of 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer. The brains were extracted and immersion- fixed in PFA for 2 h at room temperature (RT), washed two times with PBS to remove excess PFA, and stored in 1X PBS at 4°C until they were sectioned (50 μm) using a vibratome (Leica VT 1000 S). Sections were mounted on slides using VECTASHIELD® mounting medium (Vector Laboratories, Burlingame, CA, USA). Images were taken under a Zeiss Stemi 2000- C (Pleasanton, CA) using an Amscope MD800E running AmScope x64 3.0 Imaging Software. + +buffer. The brains were extracted and immersion- fixed in PFA for 2 h at room temperature (RT), washed two times with PBS to remove excess PFA, and stored in 1X PBS at \(4^{\circ}\mathrm{C}\) until they were sectioned (50 μm) using a vibratome (Leica VT 1000 S). Sections were mounted on slides using VECTASHIELD® mounting medium (Vector Laboratories, Burlingame, CA, USA). Images were taken under a Zeiss Stemi 2000- C (Pleasanton, CA) using an Amscope MD800E running AmScope x64 3.0 Imaging Software. + +To label biocytin filled cells after slice electrophysiology recordings, slices were washed three times for 5 min each with PBS (Gibco, Waltham, MA), then blocked with a solution containing: bovine serum albumin \((0.2\%)\) , normal goat serum \((5\%)\) and Tween20 \((0.3\%)\) ; Sigma- Aldrich, St. Louis, MO) for \(2\mathrm{h}\) at RT. Slices were incubated in DTAF- streptavidin (1:200; Jackson Immuno Research) diluted in PBS + \(0.3\%\) Tween20 for \(48\mathrm{h}\) at \(4^{\circ}\mathrm{C}\) . After five, 10- min rinses, brain slices were mounted onto glass slides as above and imaged using a Zeiss Axioskop upright microscope \((2.5\mathrm{X}\) , \(\mathrm{NA} = 0.075\) or Plan Apochromat \(20\mathrm{X}\) , \(\mathrm{NA} = 0.75\) ). + +## Electrophysiology + +Rats were deeply anesthetized with isoflurane, decapitated, and brains were quickly removed into ice- cold artificial cerebrospinal fluid (aCSF) consisting of (in mM): 119 NaCl, 2.5 KCl, 1.0 NaH₂PO₄, 26.2 NaHCO₃, 11 glucose, 1.3 MgSO₄, 2.5 CaCl₂, saturated with 95% O₂- 5% CO₂, with a measured osmolarity 310–320 mOsm/L. Two hundred μm coronal sections through the LHb were cut with a Leica VT 1000 S vibratome. Slices were incubated in oxygenated aCSF at 33 °C and allowed to recover for at least one hour. A single slice was placed in the recording chamber and continuously superfused at a rate of + +<--- Page Split ---> + +2 mL/min with oxygenated aCSF. Neurons were visualized with an upright microscope (Zeiss AxioExaminer.D1) equipped with infrared-differential interference contrast, Dodt optics, and fluorescent illumination. Whole cell recordings were made at \(34^{\circ}\mathrm{C}\) using borosilicate glass microelectrodes (3- 5 MΩ) filled with K- gluconate internal solution containing (in mM): 123 K- gluconate, 10 HEPES, 8 NaCl, 0.2 EGTA, 2 MgATP, 0.3 \(\mathrm{Na_3GTP}\) , and \(0.1\%\) biocytin (pH 7.2 adjusted with KOH; 275 mOsm/L). Liquid junction potentials were not corrected during recordings. Input and series resistance were monitored throughout voltage clamp experiments with a \(- 4\mathrm{mV}\) step every 30 seconds. Series resistance was required to be 5- 30 MΩ and cells with series resistance changes \(>25\%\) were excluded. + +Signals were recorded using a patch clamp amplifier (Axopatch 1D, Molecular Devices, San Jose, CA or IPA, Sutter Instruments, Novato, CA). Signals were filtered at \(5\mathrm{kHz}\) and collected at \(20\mathrm{kHz}\) using IGOR Pro (Wavemetrics) or collected at \(10\mathrm{kHz}\) using SutterPatch software (Sutter Instruments). Light evoked EPSCs and IPSCs were evoked by two blue light pulses (473 nm, 1- 10 ms) administered 50 ms apart, once every 30 s. LHb recordings were generally made in LHb subregions enriched in ChR2- expressing fibers. Recordings were made in voltage- clamp mode, with membrane potential clamped at \(\mathrm{V_m} = - 60\mathrm{mV}\) and \(- 40\mathrm{mV}\) , for EPSCs and IPSCs respectively. Light was delivered by an LED coupled to an optic fiber aimed at the recorded cell (7- 10 mW). To calculate connectivity rates, only the first neuron patched per slice was included in order to avoid over sampling from slices or animals with lower infection rates. + +Data analysis for electrophysiology. Light pulses were considered to reveal synaptic connections when three conditions were met: (1) the average of 8 traces showed a deviation from baseline \(I_{\mathrm{holding}}\) such that the mean trace exceeded 4 SD of 10 ms baseline period within the 10 ms window after initiation of light pulse, (2) the putative response was observed in at least 3 independent trials, and (3) the delays from the light stimulation onset of putative responses were time locked (<1 ms jitter) across trials. Latency was calculated as time from start of light pulse to when the rate of rise exceeded \(- 40,000\mathrm{V / s}\) . In some cases DNQX (10 \(\mu \mathrm{M}\) ) or gabazine (10 \(\mu \mathrm{M}\) ) was bath applied to confirm inward and outward currents as AMPA or \(\mathrm{GABA_A}\) receptor- mediated, respectively. All measurements of DAMGO effects on EPSCs were + +<--- Page Split ---> + +completed in the presence of gabazine. After recordings, slices were drop fixed in \(4\%\) PFA for at least 2 h at \(4^{\circ}\mathrm{C}\) and processed for biocytin labeling. + +## Combined retrograde tracing and in situ hybridization + +Tracer injections. Male Sprague Dawley rats (300- 500 g) were anesthetized with \(2 - 5\%\) isoflurane. \(1\%\) Fluoro- Gold (FG; FluoroChrome LLC) solution in a \(0.1\mathrm{M}\) caccodylate buffer (pH 7.5) was delivered unilaterally into the LHb \((- 3.4\mathrm{mmAP},\pm 0.9\mathrm{mmML}\) , and \(- 5.4\mathrm{mmDV}\) ) iontophoretically through a stereotaxically positioned glass micropipette (18- 25 \(\mu \mathrm{m}\) inner diameter) by applying \(1\mu \mathrm{A}\) , 7 s pulses at 14 s intervals for 20 min. The micropipette was then left in place for an additional 10 min to prevent backflow. Following surgery, rats were singly housed and perfused 3 weeks later. + +Tissue Preparation. Rats were anesthetized with chloral hydrate (0.5 ml/kg) and perfused transcardially with \(4\%\) (w/v) PFA in \(0.1\mathrm{M}\) phosphate buffer treated with diethylpyrocarbonate (DEPC), pH 7.3. Brains were post- fixed in \(4\%\) PFA for 2 h before being transferred to an \(18\%\) sucrose solution (w/v in \(0.1\mathrm{M}\) PBS) and stored overnight at \(4^{\circ}\mathrm{C}\) . Coronal sections of the LHb (30 \(\mu \mathrm{m}\) ) and LPO (16 \(\mu \mathrm{m}\) ) were prepared. Phenotyping of retrogradely labeled cells by immunocytochemistry and in situ hybridization. Sections in the LPO were incubated for 2 h at \(30^{\circ}\mathrm{C}\) with rabbit anti- FG antibody (1:500; AB153; Millipore) supplemented with RNAsin. Sections were then incubated in biotinylated goat anti- rabbit antibody (1:200; BA1000; Vector Laboratories) for 1 h at \(30^{\circ}\mathrm{C}\) . Sections were then rinsed and treated with \(0.2\mathrm{N}\) HCl, rinsed, and then acetylated in \(0.25\%\) acetic anhydride in \(0.1\mathrm{M}\) triethanolamine. Subsequently, sections were rinsed and post- fixed with \(4\%\) PFA, rinsed, and then incubated in a hybridization buffer for 2 h at \(55^{\circ}\mathrm{C}\) . + +Hybridization was then performed for radioactive detection of MOR mRNA by hybridizing sections for 16 h at \(55^{\circ}\mathrm{C}\) with [35S]- and [33P]- labeled (107 c.p.m./mL) single- stranded antisense probes. Following hybridization, sections were treated with \(4\mu \mathrm{g / mL}\) of RNAse A at \(37^{\circ}\mathrm{C}\) for 1 h, washed with 1X saline- sodium citrate and \(50\%\) formamide for 1 h at \(55^{\circ}\mathrm{C}\) , and then with \(0.1\mathrm{X}\) saline- sodium citrate at \(68^{\circ}\mathrm{C}\) for 1 + +<--- Page Split ---> + +h. To visualize \(\mathrm{FG(+)}\) cells, sections were rinsed with PBS and incubated for 1 h at RT in avidinbiotinylated horseradish peroxidase (1:100, ABC kit; Vector Laboratories). Sections were then rinsed, and the peroxidase reaction was developed with \(0.05\% 3,3'\) -diaminobenzidine tetrahydrochloride (DAB) and \(0.003\% \mathrm{H}_2\mathrm{O}_2\) . Sections were then photographed under bright field illumination and mounted on coated slides. Finally, slides were dipped in Ilford K.5 nuclear tract emulsion (Polysciences; 1:1 dilution in double-distilled water) and exposed in the dark at \(4^{\circ}\mathrm{C}\) for 3-4 weeks before development and photographs of silver-grain epiluminescence. + +Data analysis of in situ hybridization studies. Methods for analysis of in situ hybridization material have been described previously46. Briefly, pictures were adjusted to match contrast and brightness by using Adobe Photoshop (Adobe Systems). Cell counting was completed independently by three scorers blind to the hypothesis of the study. Radioactive in situ material was analyzed using epiluminescence to increase the contrast of silver grains as described previously59. \(\mathrm{FG(+)}\) cells (detected by fluorescence and brown DAB-label) were evaluated for the presence of MOR mRNA: a cell was considered to express MOR mRNA when its soma contained concentric aggregates of silver grains that exceeded background levels. + +## Fiber photometry + +Surgery. Male and female VGlut2::Cre mice (20- 30 g; 6- 12 weeks) were anesthetized with 1- 5% isoflurane and secured to a stereotaxic frame. Using a Micro4 controller and UltraMicroPump, 0.2 \(\mu \mathrm{L}\) of a retrograde, Cre- dependent HSV encoding GCaMP6m (HSV- hEF1α- LSI1- GCaMP6m) was injected into the LHb (- 1.5 mm AP, +0.45 mm ML, - 3.0 mm DV). Syringes were left in place for 7- 10 min following injections to minimize diffusion. For fiber photometry calcium imaging experiments, a 400 \(\mu \mathrm{m}\) core optic fiber (Doric Lenses) embedded in a 2.5 mm ferrule was implanted over the LPO (+0.5 mm AP, +0.8 mm ML, - 5.05 mm DV) and secured to the skull using #000 screws (Fasteners and Metal products Corp; #000- 120 X 1/16) and dental cement. Following surgery, mice recovered on a warm heating pad before + +<--- Page Split ---> + +being transferred back to the vivarium home cage. Three weeks after the virus and fiber surgery, mice were given either SNI or sham control surgery as described above. + +Recording. Signals from GCaMP6 were recorded across 10 trials of stimulation in the Hargreaves test using a Plantar Test Instrument. The onset and offset time for each trial was digitized and sent to an RZ5D (Tucker Davis Technologies). For the acquisition of LPO \(\rightarrow\) LHB activity, GCaMP6 was excited at two wavelengths (490nm, calcium- dependent signal and 405 nm isosbestic control) by amplitude modulated signals from two light- emitting diodes reflected off dichroic mirrors and coupled into a 400μm 0.48NA optic fiber. Signals emitted from GCaMP6m and its isosbestic control channel then returned through the same optic fiber and were acquired using photoreceiver (Doric Lenses), digitized at 1kHz, and then recorded by a real- time signal processor (RZ5D; Tucker Davis Technologies) running the Synapse software suite. Analysis of the resulting signal was then performed using custom- written MATLAB scripts available in a general release form at + +https://github.com/djamesbarker/FiberPhotometry. Briefly, changes in fluorescence across the experimental session ( \(\Delta \mathrm{F} / \mathrm{F}\) ) were calculated by smoothing signals from the isosbestic control channel, scaling the isosbestic control signal by regressing it on the smoothed GCaMP signal, and then generating a predicted 405 nm signal using the linear model generated during the regression. Calcium independent signals on the predicted 405 nm channel were then subtracted from the raw GCaMP signal to remove movement, photo- bleaching, and fiber bending artifacts. Signals from the GCaMP channel were then divided by the control signal to generate the \(\Delta \mathrm{F} / \mathrm{F}\) . Peri- event histograms were then created by averaging changes in fluorescence ( \(\Delta \mathrm{F} / \mathrm{F}\) ) across repeated trials during windows encompassing behavioral events of interest. The area under the curve (AUC) was calculated for a pre- stimulation 5 s baseline commencing - 10 s before paw withdrawal and for the 5 s period initiated with paw withdrawal. + +<--- Page Split ---> + +## Anterograde tracing from the ACC + +In rats, unilateral injections of AAV2- hSyn- hChR2(H134R)- mCherry (759 nL) were made throughout anteroposterior range of the ACC (+2.6 to - 0.4 mm AP, - 0.4 to - 0.5 mm ML, - 2.6 to - 2.8 mm DV). Rats were perfused and brains fixed five weeks later, as described above. Coronal sections (50 \(\mu \mathrm{m}\) ) containing the ACC and the LHb were collected. After verification of the injection site, every sixth slice containing the LHb was rinsed twice with PBS. Tissue was pre- permeabilized in 1:1 EtOH:PBS for 30 min at \(4^{\circ}\mathrm{C}\) , rinsed briefly in PBS before blocked in \(3\% \mathrm{H}_2\mathrm{O}_2\) for 10 min. Following PBS washes (3x5 min), tissue was blocked in solution containing normal goat serum (NGS, \(10\%\) ) for 1 h at RT. Slices were incubated in rabbit anti- mCherry antibody (1:5000 in PBS + 0.3% Triton X100 + NGS \(10\%\) ; Abcam) overnight at \(4^{\circ}\mathrm{C}\) . After PBS washes (4x10 min), slices were incubated in biotinylated goat anti- rabbit secondary antibody (1:200 in PBS; Vector Laboratories) for 2 h at \(4^{\circ}\mathrm{C}\) . Following PBS wash (4x10 min), slices were incubated in VECTASTAIN® ABC Reagent (VECTASTAIN® ABC Kit, Vector Laboratories) for 30 min followed by peroxidase substrate (DAB Substrate Kit, Vector Laboratories) for 10 min. Once dark brown DAB precipitate formed, slices were rinsed in PBS (5x5 min), mounted on glass slides, and cover slipped with DEPEX (Electron Microscopy Sciences). DAB- stained fibers were visualized under brightfield illumination and quantified in Stereo Investigator software (MBF Bioscience) using the virtual isotropic space balls probe60. + +## General experimental design + +For behavioral experiments, subject numbers were determined by pilot studies and power analyses (power \(= 0.80\) , significance level \(= 0.05\) , effect size \(= 15 - 30\%\) ). All behavioral experiments were performed blinded to experimental condition. For immunohistological experiments, three animals with injections targeted at the anterior, middle, and posterior ACC were used to obtain a comprehensive estimation of fibers projecting to the LHb. Electrophysiology experiments were conducted blind to injection site. + +<--- Page Split ---> + +## Quantification and statistical analysis + +Data are expressed as mean \(\pm\) SEM or mean with \(25^{\mathrm{th}}\) and \(75^{\mathrm{th}}\) percentiles as indicated in figure legends and text. Significance was set at \(\mathrm{p}< 0.05\) . Datasets were evaluated to determine whether parametric or non- parametric statistical approaches were most appropriate as indicated in Supplemental Table 1. All tests were two tailed, and statistical analyses were performed in GraphPad Prism or R. The bandwidth for electrophysiology violin plots was determined by Silverman's rule of thumb in Plotly for Python; violin plots for behavior experiments were constructed with heavy kernel density estimations in Prism. Sample sizes are reported in figure panels, legends, and Supplementary Table 1. Outliers, including extreme outliers, are reported in Supplementary Table 1 and were not removed from datasets. + +## Reporting summary + +Further information on research design is available in the Nature Research Reporting Summary linked to this paper. + +## Data availability + +All data described in the main text or extended data are available from the corresponding author upon request. + +## Acknowledgements + +The authors would like to thank Ryan Carothers, Gabrielle Mintz, Lucy He, Venkateswaran Ganesh, and Benjamin Snyder for their technical assistance with histology, stereotaxic surgeries, and behavioral + +studies. This work was supported by National Institutes of Health grants R01DA042025 (to E.B.M.), K08 + +<--- Page Split ---> + +NS097632 (to M.W.W.), and the Intramural Research Program (IRP) of the National Institute on Drug Abuse (IRP/NIDA/NIH) and a NIDA K99/R00 pathway to independence award (DA043572) to D.J.B. + +## Competing Interests Statement + +The authors have no competing interests. + +## Figure Legends + +Figure 1. MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500 \mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10 \mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . SNI males: Wilcoxon signed rank test, \(\mathrm{V} = 2\) , \(\mathrm{p} = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with SNI \((n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.53\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 31\) , \(\mathrm{p} = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra-LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 15.93\) , \(\mathrm{p} = 0.002\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.01\) ; paired t-tests, sham adjusted \(\mathrm{p} = 1\) ; SNI adjusted \(\mathrm{p} = 0.016\) . Female SNI animals trended towards a preference for the DAMGO-paired chamber: Paired t-test \(\mathrm{p} = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two-way mixed ANOVA, \(\mathrm{F}(1,19) = 2.239\) , \(\mathrm{p} = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100 \mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 5\) , \(\mathrm{p} = 1\) . SNI: Wilcoxon signed + +<--- Page Split ---> + +rank test, \(\mathrm{V} = 6\) , \(\mathrm{p} = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100~\mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two- way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 6.234\) , \(\mathrm{p} = 0.027\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.07\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.48\) ; SNI adjusted \(\mathrm{p} = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(^{*}\mathrm{p}\) \(\leq 0.05\) , \(^{**}p< 0.01\) + +Figure 2. Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) . a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 32\) , \(\mathrm{t} = 0.892\) , \(\mathrm{p} = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 21\) , \(\mathrm{t} = 0.137\) , \(\mathrm{p} = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t-test, \(\mathrm{df} = 24\) , \(\mathrm{t} = 0.102\) , \(\mathrm{p} = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t-test unequal variances, \(\mathrm{df} = 37\) , \(\mathrm{t} = -0.17\) , \(\mathrm{p} = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t-test unequal variances, \(\mathrm{df} = 38\) , \(\mathrm{t} = -0.09\) , \(\mathrm{p} = 0.93\) . Data from naïve rats previously published in \(^{16}\) . + +<--- Page Split ---> + +Figure 3. Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2- hSyn- hChR2(H134R)- mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venn diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically- evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One- way ANOVA, \(\mathrm{df} =\) 4, \(\mathrm{F} = 4.11\) , \(\mathrm{p} = 0.0057\) followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically- evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light- evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was \(< 2\) ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically- evoked LPO- LHb EPSCs ( \(\mathrm{n} = 12\) ). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more GABAergic connections than glutamate connections. (Right) DAMGO inhibited light- evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +Figure 4. LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb ( \(\mathrm{n} = 3\) rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar \(= 20 \mu \mathrm{m}\) . (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. + +<--- Page Split ---> + +f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +Figure 5. LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV- hEF1α- LS1L- GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb- projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI (green, \(\mathrm{n} = 5\) ) exhibited significantly larger changes in GCaMP6m fluorescence time- locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls (grey, \(\mathrm{n} = 9\) ): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2- mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intra- LHb saline or DAMGO. d, Representative images of Cre- dependent ChR2- mCherry fluorescence in LPO cell bodies. (Left) Scale bar \(= 250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber (“o.f.”) implant amidst ChR2- mCherry- expressing cell bodies. Scale bar \(= 50 \mu \mathrm{m}\) . e, Only animals with active ChR2 ( \(\mathrm{n} = 8\) ), but not mCherry controls ( \(\mathrm{n} = 8\) ), developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t- tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2- mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +# Extended Data Figure Legends + +<--- Page Split ---> + +Extended Data Figure 1. Further characterization of neuropathic pain and inflammatory pain models and the impact of intra- LHb MOR activation in rat. a, (Left) Violin plots of withdrawal latency to heat in sham \((\mathrm{n} = 9)\) , male \((\mathrm{n} = 8)\) and female \((\mathrm{n} = 9)\) rats with SNI after saline vs DAMGO ( \(10~\mu \mathrm{M}\) ) microinjections into the LHb. Sham males: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) ; SNI males: Paired t- test, \(\mathrm{df} = 7\) , \(\mathrm{t} = - 1.92\) , \(\mathrm{p} = 0.096\) ; SNI females: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 1.4\) , \(\mathrm{p} = 0.199\) . (Right) i.c.v. sham male: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 0.08\) , \(\mathrm{p} = 0.938\) . i.c.v. SNI male: Paired t- test, \(\mathrm{df} = 11\) , \(\mathrm{t} = 3.2\) , \(\mathrm{p} = 0.00846\) . To test if the effects of MOR activation in the LHb generalize to other forms of pain, we repeated the manipulations above with an inflammatory model of pain. Male rats received an intradermal injection of CFA or sterile saline into the plantar aspect of the hindpaw, b, Schematic diagram of inflammatory injury preparations and cannulation targeting the LHb. c, Mechanical withdrawal thresholds in sham and CFA- injured rats. In animals with CFA, we observed an increase in the average withdrawal threshold following intra- LHb DAMGO ( \(10~\mu \mathrm{M}\) ) compared to saline, indicating a reduction in mechanical allodynia: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 2.89\) , \(\mathrm{p} = 0.0202\) ; but not shams Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . d, Withdrawal latency to heat. Similar to results in rats with SNI, DAMGO in the LHb did not reverse the CFA- induced decrease in withdrawal latency to heat. Sham: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) . CFA: Wilcoxon signed rank test, \(\mathrm{V} = 12\) , \(\mathrm{p} = 0.25\) . e, Rats with CFA injury also did not develop a preference for the DAMGO- paired chamber in the place conditioning paradigm: Two- way mixed ANOVA, significant interaction between CFA/sham and baseline/test \(\mathrm{F}(1,14) = 0.474\) , \(\mathrm{p} = 0.502\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.286\) ; SNI adjusted \(\mathrm{p} = 0.761\) . The discrepancy between place conditioning results of our neuropathic and inflammatory pain animals may be due to the natural history of the injury caused in the CFA model: whereas SNI animals underwent permanent nerve ligation, animals injected with CFA presented with transient swelling and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were + +<--- Page Split ---> + +tested daily; each line represents a male rat. \(\mathbf{g}\) , Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(^{*}\mathrm{p}\leq 0.05\) , \(^{**}p\leq 0.01\) + +Extended Data Figure 2. Neurons receiving functional synaptic inputs from the five brain regions showing connectivity were distributed throughout the LHb. Locations of neurons found connected using whole- cell recordings from optogenetic stimulation of terminals arising from each input tested. Locations are based on biocytin immunohistochemistry and low magnification images taken on the recording microscope where the recorded cell is centered within the field of view. Color indicates source of innervation. + +Extended Data Figure 3. ACC minimally innervates the LHb. a, Diagram of the extent of unilateral AAV2- hSyn- hChR2(H134R)- mCherry injection sites ( \(\mathrm{n} = 9\) male rats) throughout anteroposterior range of the ACC for anterograde tracing study. b, Example ipsilateral DAB- positive fibers (black) visualized under brightfield illumination. Fibers heavily innervate the MDL, which abuts the lateral edge of the LHb, while sparse to no fibers innervate the LHb. Contralateral innervation of LHb and MDL was negligible compared to the ipsilateral side, and therefore were omitted from our analysis. (Left) Scale bar \(= 250 \mu \mathrm{m}\) . (Right) Scale bar \(= 50 \mu \mathrm{m}\) . c, Average stereologically- quantified DAB- positive fiber lengths throughout anteroposterior range of the LHb, compared to the MDL innervation in the same coronal slice. Mann- Whitney test, two- tailed: \(\mathrm{U} = 0\) , \(\mathrm{p} = 0.0006\) . + +<--- Page Split ---> + +Extended Data Figure 4. Sources of glutamatergic inputs to the rat LHb vary in strength and delay in synaptic transmission. a, We observed some variations between inputs in the mean light evoked EPSCs \(\mathrm{(V_{m} = - 60 mV)}\) . In particular, excitatory inputs from the VTA were consistently small: Kruskal- Wallis \(\chi 2\) \(= 25.5\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.00004\) , followed by Dunn Test for pairwise comparisons. b, Differences were also detected in the delay to the onset of light evoked EPSCs to LHb neurons from these different sources, with inputs from LH showing the fastest response. Although these data deviate from a normal distribution (Shapiro's test, \(\mathrm{p} = 0.0000001\) ), KDEs (violins) are consistent with continuous distributions, suggesting reliable polysynaptic events were rarely detected. c, We observed some variations between inputs in the mean light evoked IPSCs \(\mathrm{(V_{m} = - 40 mV)}\) . While there were no statistically significant differences in amplitudes detected, the mean inhibitory input from the VP was particularly small: Kruskal- Wallis \(\chi 2 = 2.3\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.7\) . \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.0005\) + +Extended Data Figure 5. Optogenetic experiments detect local glutamate, but not GABA, functional connections in rat LHb. AAV2-hSyn-hChR2(H134R)- mCherry was stereotaxically injected into the LHb at least 4 weeks prior to ex vivo whole cell recordings in the LHb to detect local synaptic connections. Neurons were recorded blind to ChR2 expression, therefore in some cases the patched neuron expressed ChR2. Therefore, in addition to connectivity criteria used for other afferent inputs, in these experiments only light evoked inward currents that were blocked by \(10 \mu \mathrm{M}\) DNQX were considered glutamatergic connections, and direct ChR2 induced inward currents were subtracted out for the quantification illustrated here. Each cell was probed for both glutamate and GABA inputs in voltage clamp by holding neurons at \(\mathrm{V_{m} = - 60 mV}\) and \(- 40 \mathrm{mV}\) , light pulse durations 1, 5, and \(10 \mathrm{ms}\) durations. a, Example recording at \(\mathrm{V_{m} = - 60 mV}\) showing a light evoked response that was blocked by DNQX. b, Example recording at \(\mathrm{V_{m} = - 40 mV}\) with minimal outward current response within \(7 \mathrm{ms}\) of light pulse. c, Summary of all LHb recordings tested in this experiment. Filled circles represent cells where responses could be classified as local ChR2 induced synaptic transmission. When no clear response was detected, the + +<--- Page Split ---> + +measure indicated is the difference between the mean \(I_{\text{holding}}\) of the baseline 100 ms period just prior to the light pulse and the mean \(I_{\text{holding}}\) 2 ms period starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V_m} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +Extended Data Figure 6. Isolated monosynaptic glutamatergic inputs from LPO to LHb neurons are inhibited by MOR activation. a, Example light evoked EPSC responses in an LHb neuron from a rat with ChR2 expression in LPO neurons. This response persisted in monosynaptic isolation by \(500 \mathrm{nM}\) TTX and \(100 \mu \mathrm{M} 4 \mathrm{AP}\) (green), and this isolated response was inhibited by \(500 \mathrm{nM}\) DAMGO (magenta). b, Summary of DAMGO effects on isolated monosynaptic EPSC inputs to LHb neurons expressed as \(\%\) of baseline monosynaptic response (left) and as raw EPSC magnitudes (right). Paired t-test, \(\mathrm{df} = 4\) , \(\mathrm{t} = - 5.1\) , \(\mathrm{p} = 0.007\) . \(^{**} \mathrm{p} < 0.01\) + +Extended Data Figure 7. Mice with SNI show hypersensitivity to heat and increased activity in LHb- projecting LPO neurons during paw withdrawal from thermal stimulation. a, Mice with SNI show a reduced latency to withdraw their paw following thermal stimulation in the Hargreaves task. Unpaired t- test, \(\mathrm{t}(12) = 3.007\) , \(\mathrm{p} = 0.011\) . b, VGluT2- expressing LPO neurons that project to the LHb expressed GCaMP6m and showed a greater calcium response during paw withdrawal to Hargreaves thermal stimulation (area under the curve, deviation from baseline fluorescence) in SNI animals ( \(\mathrm{n} = 5\) ) compared to sham controls ( \(\mathrm{n} = 9\) ): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.4\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . \(^{**} \mathrm{p} < 0.01\) + +Extended Data Figure 8. Locations and spread of bilateral ChR2 injections to the LPO in rats used for in vivo optogenetic experiments. + +<--- Page Split ---> + +References1. Matsumoto, M. & Hikosaka, O. Representation of negative motivational value in the primate lateral habenula. Nat Neurosci 12, 77–84 (2009).2. Benabid, A. L. & Jeaugey, L. Cells of the rat lateral habenula respond to high-threshold somatosensory inputs. Neurosci Lett 96, 289–94 (1989).3. Dafny, N. & Qiao, J. T. Habenular neuron responses to noxious input are modified by dorsal raphe stimulation. Neurol. Res. 12, 117–121 (1990).4. Nagao, M., Kamo, H., Akiguchi, I. & Kimura, J. Induction of c-Fos-like protein in the lateral habenular nucleus by persistent noxious peripheral stimulation. Neurosci. Lett. 151, 37–40 (1993).5. Huang, T. et al. Identifying the pathways required for coping behaviours associated with sustained pain. Nature 565, 86–90 (2019).6. Matsumoto, M. & Hikosaka, O. Lateral habenula as a source of negative reward signals in dopamine neurons. Nature 447, 1111–1115 (2007).7. Li, B. et al. Synaptic potentiation onto habenula neurons in the learned helplessness model of depression. Nature 470, 535–539 (2011).8. Poller, W. C., Madai, V. I., Bernard, R., Laube, G. & Veh, R. W. A glutamatergic projection from the lateral hypothalamus targets VTA-projecting neurons in the lateral habenula of the rat. Brain Res 1507, 45–60 (2013).9. Herkenham, M. & Nauta, W. J. Afferent connections of the habenular nuclei in the rat. A horseradish peroxidase study, with a note on the fiber-of-passage problem. J Comp Neurol 173, 123–46 (1977).10. Vertes, R. P. Analysis of projections from the medial prefrontal cortex to the thalamus in the rat, with emphasis on nucleus reuniens. J. Comp. Neurol. 442, 163–187 (2002).11. Yetnikoff, L., Cheng, A. Y., Lavezzi, H. N., Parsley, K. P. & Zahm, D. S. Sources of input to the rostromedial tegmental nucleus, ventral tegmental area, and lateral habenula compared: A study in rat. J Comp Neurol 523, 2426–56 (2015). + +<--- Page Split ---> + +12. Kim, U. & Lee, T. Topography of descending projections from anterior insular and medial prefrontal regions to the lateral habenula of the epithalamus in the rat. Eur J Neurosci 35, 1253–69 (2012). + +13. Araki, M., McGeer, P. L. & Kimura, H. The efferent projections of the rat lateral habenular nucleus revealed by the PHA-L anterograde tracing method. Brain Res. 441, 319–330 (1988). + +14. Quina, L. A. et al. Efferent pathways of the mouse lateral habenula. J Comp Neurol 523, 32–60 (2015). + +15. Cohen, S. R. & Melzack, R. Morphine injected into the habenula and dorsal posteromedial thalamus produces analgesia in the formalin test. Brain Res. 359, 131–139 (1985). + +16. Margolis, E. B. & Fields, H. L. Mu Opioid Receptor Actions in the Lateral Habenula. PLoS One 11, e0159097 (2016). + +17. Decosterd, I. & Woolf, C. J. Spared nerve injury: an animal model of persistent peripheral neuropathic pain. PAIN 87, 149–158 (2000). + +18. Miaskowski, C. et al. Does opioid analgesia show a gender preference for females? Pain Forum 8, 34–44 (1999). + +19. Kest, B., Sarton, E. & Dahan, A. Gender differences in opioid-mediated analgesia: animal and human studies. Anesthesiology 93, 539–547 (2000). + +20. Kepler, K. L. et al. Gender effects and central opioid analgesia. Pain 45, 87–94 (1991). + +21. Thompson, S. J. et al. Chronic neuropathic pain reduces opioid receptor availability with associated anhedonia in rat. Pain 159, 1856–1866 (2018). + +22. Porreca, F. et al. Spinal opioid mu receptor expression in lumbar spinal cord of rats following nerve injury. Brain Res. 795, 197–203 (1998). + +23. Zhang, X. et al. Down-regulation of mu-opioid receptors in rat and monkey dorsal root ganglion neurons and spinal cord after peripheral axotomy. Neuroscience 82, 223–240 (1998). + +<--- Page Split ---> + +24. Shabel, S. J., Proulx, C. D., Trias, A., Murphy, R. T. & Malinow, R. Input to the lateral habenula from the basal ganglia is excitatory, aversive, and suppressed by serotonin. Neuron 74, 475–481 (2012). + +25. Lecca, S. et al. Aversive stimuli drive hypothalamus-to-habenula excitation to promote escape behavior. eLife 6, (2017). + +26. Hu, H., Cui, Y. & Yang, Y. Circuits and functions of the lateral habenula in health and in disease. Nat. Rev. Neurosci. 21, 277–295 (2020). + +27. Root, D. H., Mejias-Aponte, C. A., Qi, J. & Morales, M. Role of glutamatergic projections from ventral tegmental area to lateral habenula in aversive conditioning. J Neurosci 34, 13906–10 (2014). + +28. Barker, D. J. et al. Lateral Preoptic Control of the Lateral Habenula through Convergent Glutamate and GABA Transmission. Cell Rep. 21, 1757–1769 (2017). + +29. Fields, H. State-dependent opioid control of pain. Nat Rev Neurosci 5, 565–75 (2004). + +30. Tooley, J. et al. Glutamatergic Ventral Pallidal Neurons Modulate Activity of the Habenula-Tegmental Circuitry and Constrain Reward Seeking. Biol. Psychiatry 83, 1012–1023 (2018). + +31. Johansen, J. P., Fields, H. L. & Manning, B. H. The affective component of pain in rodents: direct evidence for a contribution of the anterior cingulate cortex. Proc. Natl. Acad. Sci. U. S. A. 98, 8077–8082 (2001). + +32. LaGraize, S. C., Borzan, J., Peng, Y. B. & Fuchs, P. N. Selective regulation of pain affect following activation of the opioid anterior cingulate cortex system. Exp. Neurol. 197, 22–30 (2006). + +33. Kim, U. & Chang, S. Y. Dendritic morphology, local circuitry, and intrinsic electrophysiology of neurons in the rat medial and lateral habenular nuclei of the epithalamus. J Comp Neurol 483, 236–50 (2005). + +34. Flanigan, M. E. et al. Orexin signaling in GABAergic lateral habenula neurons modulates aggressive behavior in male mice. Nat. Neurosci. 23, 638–650 (2020). + +<--- Page Split ---> + +35. Zhang, L. et al. A GABAergic cell type in the lateral habenula links hypothalamic homeostatic and midbrain motivation circuits with sex steroid signaling. Transl. Psychiatry 8, 50 (2018). + +36. Quina, L. A., Walker, A., Morton, G., Han, V. & Turner, E. E. GAD2 Expression Defines a Class of Excitatory Lateral Habenula Neurons in Mice that Project to the Raphe and Pontine Tegmentum. eNeuro 7, (2020). + +37. Wallace, M. L. et al. Anatomical and single-cell transcriptional profiling of the murine habenular complex. eLife 9, (2020). + +38. Mansour, A., Khachaturian, H., Lewis, M. E., Akil, H. & Watson, S. J. Autoradiographic differentiation of mu, delta, and kappa opioid receptors in the rat forebrain and midbrain. J. Neurosci. Off. J. Soc. Neurosci. 7, 2445–2464 (1987). + +39. Kang, S. et al. Downregulation of M-channels in lateral habenula mediates hyperalgesia during alcohol withdrawal in rats. Sci. Rep. 9, 2714 (2019). + +40. Li, J., Li, Y., Zhang, B., Shen, X. & Zhao, H. Why depression and pain often coexist and mutually reinforce: Role of the lateral habenula. Exp. Neurol. 284, 106–113 (2016). + +41. Li, Y. et al. Role of the Lateral Habanula in Pain-Associated Depression. Front. Behav. Neurosci. 11, 31 (2017). + +42. Weiss, T. & Veh, R. W. Morphological and electrophysiological characteristics of neurons within identified subnuclei of the lateral habenula in rat brain slices. Neuroscience 172, 74–93 (2011). + +43. Metzger, M. et al. Habanular connections with the dopaminergic and serotonergic system and their role in stress-related psychiatric disorders. Eur. J. Neurosci. (2019) doi:10.1111/ejn.14647. + +44. Meye, F. J. et al. Shifted pallidal co-release of GABA and glutamate in habenula drives cocaine withdrawal and relapse. Nat. Neurosci. 19, 1019–1024 (2016). + +45. Shabel, S. J., Proulx, C. D., Piriz, J. & Malinow, R. Mood regulation. GABA/glutamate co-release controls habenula output and is modified by antidepressant treatment. Science 345, 1494–8 (2014). + +<--- Page Split ---> + +Root, D. H. et al. Selective Brain Distribution and Distinctive Synaptic Architecture of Dual Glutamatergic- GABAergic Neurons. Cell Rep. 23, 3465–3479 (2018). + +Newman, H. M., Stevens, R. T. & Apkarian, A. V. Direct spinal projections to limbic and striatal areas: anterograde transport studies from the upper cervical spinal cord and the cervical enlargement in squirrel monkey and rat. J. Comp. Neurol. 365, 640–658 (1996). + +Burstein, R., Cliffer, K. D. & Giesler, G. J. Direct somatosensory projections from the spinal cord to the hypothalamus and telencephalon. J. Neurosci. Off. J. Soc. Neurosci. 7, 4159–4164 (1987). + +Hori, T., Oka, T., Hosoi, M. & Aou, S. Pain modulatory actions of cytokines and prostaglandin E2 in the brain. Ann. N. Y. Acad. Sci. 840, 269–281 (1998). + +Ono, T. & Nakamura, K. Learning and integration of rewarding and aversive stimuli in the rat lateral hypothalamus. Brain Res. 346, 368–373 (1985). + +Almli, C. R. & McMullen, N. T. Ontogeny of lateral preoptic unit activity in rats. Brain Res. Bull. 4, 773–781 (1979). + +Vowles, K. E. et al. Rates of opioid misuse, abuse, and addiction in chronic pain: a systematic review and data synthesis. Pain 156, 569–576 (2015). + +Zadina, J. E. et al. Endomorphin analog analgesics with reduced abuse liability, respiratory depression, motor impairment, tolerance, and glial activation relative to morphine. + +Neuropharmacology 105, 215–227 (2016). + +Ehrlich, A. T. et al. Expression map of 78 brain-expressed mouse orphan GPCRs provides a translational resource for neuropsychiatric research. Commun. Biol. 1, 102 (2018). + +Wagner, F., Bernard, R., Derst, C., French, L. & Veh, R. W. Microarray analysis of transcripts with elevated expressions in the rat medial or lateral habenula suggest fast GABAergic excitation in the medial habenula and habenular involvement in the regulation of feeding and energy balance. Brain Struct. Funct. 221, 4663–4689 (2016). + +<--- Page Split ---> + +56. Broms, J. et al. Monosynaptic retrograde tracing of neurons expressing the G-protein coupled + +receptor Gpr151 in the mouse brain. J. Comp. Neurol. 525, 3227- 3250 (2017). + +57. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates, Compact. (Academic Press, 1997). + +58. Chaplan, S. R., Bach, F. W., Pogrel, J. W., Chung, J. M. & Yaksh, T. L. Quantitative assessment of tactile allodynia in the rat paw. J. Neurosci. Methods 53, 55-63 (1994). + +59. Yamaguchi, T., Wang, H.-L., Li, X., Ng, T. H. & Morales, M. Mesocorticolimbic glutamatergic pathway. J. Neurosci. Off. J. Soc. Neurosci. 31, 8476-8490 (2011). + +60. Mouton, P. R., Gokhale, A. M., Ward, N. L. & West, M. J. Stereological length estimation using spherical probes. J. Microsc. 206, 54-64 (2002). + +<--- Page Split ---> +![](images/Figure_2.jpg) + + +Extended Data Figure 1. Further characterization of neuropathic pain and inflammatory pain models and the impact of intra- LHb MOR activation in rat. a, (Left) Violin plots of withdrawal latency to heat in sham \((\mathrm{n} = 9)\) , male \((\mathrm{n} = 8)\) and female \((\mathrm{n} = 9)\) rats with SNI after saline vs DAMGO \((10~\mu \mathrm{M})\) microinjections into the LHb. Sham males: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) ; SNI males: Paired t- test, \(\mathrm{df} = 7\) , \(\mathrm{t} = - 1.92\) , \(\mathrm{p} = 0.096\) ; SNI females: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 1.4\) , \(\mathrm{p} = 0.199\) . (Right) i.c.v. sham male: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 0.08\) , \(\mathrm{p} = 0.938\) . i.c.v. SNI male: Paired t- test, \(\mathrm{df} = 11\) , \(\mathrm{t} = 3.2\) , \(\mathrm{p} = 0.00846\) . To test if the effects of MOR activation in the LHb generalize to other forms of pain, we repeated the manipulations above with an inflammatory model of pain. Male rats received an intradermal injection of CFA or sterile saline into the plantar aspect of the hindpaw, b, Schematic diagram of inflammatory injury preparations and cannulation targeting the LHb. c, Mechanical withdrawal thresholds in sham and CFA- injured rats. In animals with CFA, we observed an increase in the average withdrawal threshold following intra- LHb DAMGO \((10~\mu \mathrm{M})\) compared to saline, indicating a reduction in mechanical allodynia: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 2.89\) , \(\mathrm{p} = 0.0202\) ; but not shams Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . d, Withdrawal latency to heat. Similar to results in rats with SNI, DAMGO in the LHb did not reverse the CFA- induced decrease in withdrawal latency to heat. Sham: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) . CFA: Wilcoxon signed rank test, \(\mathrm{V} = 12\) , \(\mathrm{p} = 0.25\) . e, Rats with CFA injury also did not develop a preference for the DAMGO- paired chamber in the place conditioning paradigm: Two- way mixed ANOVA, significant interaction between CFA/sham and baseline/test \(\mathrm{F}(1,14) = 0.474\) , \(\mathrm{p} = 0.502\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.286\) ; SNI adjusted \(\mathrm{p} = 0.761\) . The discrepancy between place conditioning results of our neuropathic and inflammatory pain animals may be due to the natural history of the injury caused in the CFA model: whereas SNI animals underwent permanent nerve ligation, animals injected with CFA presented with transient swelling and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were tested daily; each line represents a male rat. g, Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} \leq 0.01\) + +and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were tested daily; each line represents a male rat. g, Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} \leq 0.01\) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Extended Data Figure 2
+ +Extended Data Figure 2. Neurons receiving functional synaptic inputs from the five brain regions showing connectivity were distributed throughout the LHb. Locations of neurons found connected using whole- cell recordings from optogenetic stimulation of terminals arising from each input tested. Locations are based on biocytin immunohistochemistry and low magnification images taken on the recording microscope where the recorded cell is centered within the field of view. Color indicates source of innervation. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Extended Figure 3
+ +![](images/Figure_4.jpg) + +
C
+ +![](images/Figure_5.jpg) + + +Extended Data Figure 3. ACC minimally innervates the LHb. a, Diagram of the extent of unilateral AAV2- hSyn- hChR2(H134R)- mCherry injection sites (n = 9 male rats) throughout anteroposterior range of the ACC for anterograde tracing study. b, Example ipsilateral DAB- positive fibers (black) visualized under brightfield illumination. Fibers heavily innervate the MDL, which abuts the lateral edge of the LHb, while sparse to no fibers innervate the LHb. Contralateral innervation of LHb and MDL was negligible compared to the ipsilateral side, and therefore were omitted from our analysis. (Left) Scale bar = 250 \(\mu \mathrm{m}\) . (Right) Scale bar = 50 \(\mu \mathrm{m}\) . c, Average stereologically- quantified DAB- positive fiber lengths throughout anteroposterior range of the LHb, compared to the MDL innervation in the same coronal slice. Mann- Whitney test, two- tailed: \(\mathrm{U} = 0\) , \(\mathrm{p} = 0.0006\) . + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Extended Data Figure 4
+ +Extended Data Figure 4. Sources of glutamatergic inputs to the rat LHb vary in strength and delay in synaptic transmission. a, We observed some variations between inputs in the mean light evoked EPSCs ( \(\mathrm{V_m} = - 60 \mathrm{mV}\) ). In particular, excitatory inputs from the VTA were consistently small: Kruskal- Wallis \(\chi_2 = 25.5\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.00004\) , followed by Dunn Test for pairwise comparisons. b, Differences were also detected in the delay to the onset of light evoked EPSCs to LHb neurons from these different sources, with inputs from LH showing the fastest response. Although these data deviate from a normal distribution (Shapiro's test, \(\mathrm{p} = 0.0000001\) ), KDEs (violins) are consistent with continuous distributions, suggesting reliable polysynaptic events were rarely detected. c, We observed some variations between inputs in the mean light evoked IPSCs ( \(\mathrm{V_m} = - 40 \mathrm{mV}\) ). While there were no statistically significant differences in amplitudes detected, the mean inhibitory input from the VP was particularly small: Kruskal- Wallis \(\chi_2 = 2.3\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.7\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) , \(^{****} \mathrm{p} < 0.0005\) + +![](images/Figure_7.jpg) + + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Extended Data Figure 5
+ +Extended Data Figure 5. Optogenetic experiments detect local glutamate, but not GABA, functional connections in rat LHb. AAV2-hSyn-hChR2(H134R)- mCherry was stereotaxically injected into the LHb at least 4 weeks prior to ex vivo whole cell recordings in the LHb to detect local synaptic connections. Neurons were recorded blind to ChR2 expression, therefore in some cases the patched neuron expressed ChR2. Therefore, in addition to connectivity criteria used for other afferent inputs, in these experiments only light evoked inward currents that were blocked by \(10 \mu \mathrm{M}\) DNQX were considered glutamatergic connections, and direct ChR2 induced inward currents were subtracted out for the quantification illustrated here. Each cell was probed for both glutamate and GABA inputs in voltage clamp by holding neurons at \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) and \(- 40 \mathrm{mV}\) , light pulse durations 1, 5, and 10 ms durations. a, Example recording at \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) showing a light evoked response that was blocked by DNQX. b, Example recording at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) with minimal outward current response within 7 ms of light pulse. c, Summary of all LHb recordings tested in this experiment. Filled circles represent cells where responses could be classified as local ChR2 induced synaptic transmission. When no clear response was detected, the measure indicated is the difference between the mean \(I_{\mathrm{holding}}\) of the baseline \(100 \mathrm{ms}\) period just prior to the light pulse and the mean \(I_{\mathrm{holding}}\) 2 ms period starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Extended Data Figure 6
+ +![](images/Figure_2.jpg) + + +Extended Data Figure 6. Isolated monosynaptic glutamatergic inputs from LPO to LHb neurons are inhibited by MOR activation. a, Example light evoked EPSC responses in an LHb neuron from a rat with ChR2 expression in LPO neurons. This response persisted in monosynaptic isolation by \(500~\mathrm{nM}\) TTX and \(100~\mu \mathrm{M}4\mathrm{AP}\) (green), and this isolated response was inhibited by \(500~\mathrm{nM}\) DAMGO (magenta). b, Summary of + +DAMGO effects on isolated monosynaptic EPSC inputs to LHb neurons expressed as \(\%\) of baseline monosynaptic response (left) and as raw EPSC magnitudes (right). Paired t- test, \(\mathrm{df} = 4\) , \(\mathrm{t} = - 5.1\) , \(\mathrm{p} = 0.007\) . \(^{**} \mathrm{p} < 0.01\) + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Extended Data Figure 7
+ +Extended Data Figure 7. Mice with SNI show hypersensitivity to heat and increased activity in LHb- projecting LPO neurons during paw withdrawal from thermal stimulation. a, Mice with SNI show a reduced latency to withdraw their paw following thermal stimulation in the Hargreaves task. Unpaired t- test, \(\mathrm{t}(12) = 3.007\) , \(\mathrm{p} = 0.011\) . b, VGluT2- expressing LPO neurons that project to the LHb expressed GCaMP6m and showed a greater calcium response during paw withdrawal to Hargreaves + +thermal stimulation (area under the curve, deviation from baseline fluorescence) in SNI animals \(\mathrm{(n = 5)}\) compared to sham controls \(\mathrm{(n = 9)}\) : Two- way ANOVA, \(\mathrm{F}(1,12) = 5.4\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . \*\* \(\mathrm{p} < 0.01\) + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Extended Data Figure 8. Locations and spread of bilateral ChR2 injections to the LPO in rats used for in vivo optogenetic experiments.
+ +<--- Page Split ---> + +Supplementary Table 1: Assumption testing on behavioral data + +
ExperimentFigure# animals;
# Outliers;
# extreme
outliers		Shapiro-Wilk
Test of
Normality
(across all
groups)
Statistic; p val		Test for
Homogeneity of
Variances
Statistic; p val
(Test used)		Parametric testNon-parametric test:
Repeated measures
Wilcoxon signed rank
exact test
V; p val
von Frey: Male; Sham;
10 μM DAMGO in
LHb		1b9; 0; 00.773; 0.0006380.0760; 0.786
(Levene's Test)		n/a1.5; 1
von Frey: Male; SNI;
10 μM DAMGO in
LHb		1b8; 0; 00.866; 0.02401.49; 0.242
(Levene's Test)		n/a2; 0.05024
von Frey: Female;
SNI; 10 μM DAMGO
in LHb		1b9; 2; 10.805; 0.001811.82; 0.196
(Levene's Test)		n/a7; 0.07422
von Frey: Male; Sham;
10 μM DAMGO in
i.c.v.		1b9; 0; 00.768; 0.000550.0897; 0.768
(Levene's Test)		n/a7; 0.5294
von Frey: Male; SNI;
10 μM DAMGO in
i.c.v.		1b12; 2; 20.756;
0.0000618		0.168; 0.686
(Levene's Test)		n/a31; 0.08006
Place Conditioning:
Male Sham/SNI; 10
μM DAMGO in LHb		1c15; 4; 0Sham x
Baseline: 0.95;
0.726
SNI x Baseline:
0.94; 0.686
Sham x Test:
0.936; 0.542
SNI x Test:
0.840; 0.129		4.68; 0.0305
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,13)=15.932;p\) \(=0.002\)
effect of group on baseline: \(\mathrm {F}=0.293;\) adjusted \(\mathrm {p}=1\)
effect of group on test day: \(\mathrm {F}=11.3;\) adjusted \(\mathrm {p}=0.01\)
Paired t-tests, adjusted
Sham: adjusted \(\mathrm {p}=1\)
SNI: adjusted \(\mathrm {p}=0.012\)		n/a
Place Conditioning:
Sham vs. Male SNI;
10 μM DAMGO in
i.c.v.		1c21; 2; 0Sham x
Baseline: 0.901;
0.257
SNI x Baseline:
0.967; 0.872
Sham x Test:
0.971; 0.905
SNI x Test:
0.985; 0.997		23.9;
0.000000997
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,19)=2.239;p=\) 0.151
Paired t-tests for non-significant two-way interaction:
Sham: adjusted \(\mathrm {p}=0.286\)
SNI: adjusted \(\mathrm {p}=0.433\)		n/a
von Frey: Female;
Sham; 100 μM
DAMGO in LHb		1d6; 0; 00.838; 0.0260.493; 0.499
(Levene's Test)		n/a5; 1
von Frey: Female;
SNI; 100 μM
DAMGO in LHb		1d9; 1; 10.839; 0.009281.13; 0.306
(Levene's Test)		n/a6; 0.4017
Place Conditioning:
Female Sham/SNI;
100 μM DAMGO in
LHb		1e15; 3; 0Sham x
Baseline: 0.904
0.66
SNI x Baseline:
0.919; 0.384
Sham x Test:
0.87; 0.226
SNI x Test:
0.900; 0.250		8.59; 0.00339
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,13)=6.234;p=\) 0.027
effect of group on baseline: \(\mathrm {F}=0.006;\) adjusted \(\mathrm {p}=1\)
effect of group on test day: \(\mathrm {F}=5.56;\) adjusted \(\mathrm {p}=0.07\)
Paired t-tests, adjusted
Sham: adjusted \(\mathrm {p}=0.476\)
SNI: adjusted \(\mathrm {p}=0.0349\)		n/a
Fiber photometry;
Sham vs. SNI mice		5b14; not
determined		not determinednot determinedTwo-way mixed design ANOVA
\(\mathrm {F}(1,12)=5.439;\) \(p=0.038;\)
Holm-Sidak post-hoc test, \(\mathrm {p}=0.0074\)		n/a
Place Conditioning;
Male; mCherry vs.
ChR2; 10 μM
DAMGO in LHb		5e16; 0; 0mCherry x
Baseline: 0.941;
0.618
ChR2 x
Baseline: 0.913;
0.374		18.3; 0.0000186
(Box M-test)		Two-way mixed design ANOVA
two-way interaction, \(\mathrm {F}(1,14)=9.982;p\) \(=0.007\)
effect of group on baseline: \(\mathrm {F}=0.01;\) adjusted \(\mathrm {p}=1\)		n/a
+ +<--- Page Split ---> + +
mCherry x Test: 0.914; 0.385 ChR2 x Test: 0.847; 0.0891effect of group on test day: F = 12.5; adjusted p = 0.006
Paired t-tests, adjusted mCherry: adjusted p = 0.427 ChR2: adjusted p = 0.0027
Hargreaves: Male; Sham; 10 μM DAMGO in LHbED 1a9; 2; 00.966; 0.726Bartlett's K-squared = 0.035658, df = 1, p-value = 0.8502 (Bartlett Test)Paired t-test
Df = 8; t = 1.06; p = 0.32		n/a
Hargreaves; Male; SNI; 10 μM DAMGO in LHbED 1a8; 0; 00.941; 0.365Bartlett's K-squared = 3.704 df = 1, p-value = 0.096 (Bartlett Test)Paired t-test
Df = 7; t = -1.92; p = 0.096		n/a
Hargreaves; Female; SNI; 10 μM DAMGO in LHbED 1a9; 1; 10.944; 0.333Bartlett's K-squared = 1.35 df = 1, p-value = 0.245 (Bartlett Test)Paired t-test
Df = 8; t = -1.4; p = 0.199		n/a
Hargreaves: Male; Sham; 10 μM DAMGO in i.c.v.ED 1a9; 3; 00.927; 0.1740.840; 0.373 (Levene's Test)Paired t-test
Df = 8; t = -0.08; p = 0.938		n/a
Hargreaves; Male; SNI; 10 μM DAMGO in i.c.v.ED 1a8; 0; 00.941; 0.3650.013; 0.911 (Levene's Test)Paired t-test
Df = 11; t = 3.2; p = 0.00846		n/a
von Frey; Male; Sham; 10 μM DAMGO in LHbED 1c9; 0; 00.773; 0.0006380.0760; 0.786 (Levene's Test)n/a1.5; 1
von Frey; Male; CFA; 10 μM DAMGO in LHbED 1c9; 0; 00.903; 0.06393.80 0.0691 (Levene's Test)Paired t-test
Df = 8; t = -2.89; p = 0.0202		n/a
Hargreaves; Male; Sham; 10 μM DAMGO in LHbED 1d9; 2; 00.966; 0.726Bartlett's K-squared = 0.035658, df = 1, p-value = 0.8502 (Bartlett Test)Paired t-test
Df = 8; t = 1.06; p = 0.32		n/a
Hargreaves; Male; CFA; 10 μM DAMGO in LHbED 1d9; 2; 20.858; 0.01130.0461 0.833 (Levene's Test)n/a12; 0.25
Place Conditioning; Male; Sham vs. CFA; 10 μM DAMGO in LHbED 1e16; 3; 0Sham x Baseline: 0.901; 0.257 CFA x Baseline: 0.94; 0.635 Sham x Test: 0.971; 0.905 CFA x Test: 0.828; 0.077413.4; 0.000253 (Box M-test)Two-way mixed design ANOVA two-way interaction F(1,14) = 0.474; p = 0.502
Paired t-tests for non-significant two-way interaction:
Sham: adjusted p = 0.286
CFA: adjusted p = 0.761		n/a
ACC fiber innervation; Ipsi LHb vs. Ipsi MDLED 3c9; 0; 0not determinednot determinedn/aMann-Whitney test, two-tailed, U=0; p = 0.0006
Hargreaves; Sham vs. SNI miceED 7a14; 3; 20.896; 0.09940.0012; 0.973 (Levene's Test)Unpaired t-test
Df = 12, t = 3.007, p = 0.0109		n/a
AUC summary: Hargreaves/Fiber photometry; Sham vs. SNI miceED 7b14; 0; 0Sham x Baseline: 0.979; 0.257 Sham x paw withdrawal: 0.932; 0.496 SNI x Baseline: 0.813; 0.103 SNI x paw withdrawal: 0.921; 0.53853.4; 2.71 e-13 (Box M-test)Two-way mixed design ANOVA F(1,12) = 5.439; p = 0.038; Holm-Sidak post-hoc test, p=0.0074n/a
+ +<--- Page Split ---> + +## Figures + +![](images/Figure_5.jpg) + +
Figure 1
+ +MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500\mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10\mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(V = 1.5\) , \(p = 1\) . SNI males: Wilcoxon signed rank test, \(V = 2\) , \(p = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(V = 7\) , \(p = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with SNI \((n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(V = 7\) , \(p = 0.53\) . SNI: Wilcoxon signed rank test, \(V = 31\) , \(p = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra- LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(F(1,13) = 15.93\) , \(p = 0.002\) ; post hoc effect group on test day adjusted \(p = 0.01\) ; paired t-tests, sham adjusted \(p = 1\) ; SNI adjusted \(p = 0.016\) . Female SNI animals trended towards a preference for the DAMGO-paired chamber: Paired t-test \(p = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two-way mixed ANOVA, \(F(1,19) = 2.239\) , \(p = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100\mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(V = 5\) , \(p = 1\) . SNI: Wilcoxon signed rank test, \(V = 6\) , \(p = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100\mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(F(1,13) = 6.234\) , \(p = 0.027\) ; post hoc effect group on test day adjusted \(p = 0.07\) ; paired t-tests, sham adjusted \(p = 0.48\) ; SNI + +<--- Page Split ---> + +adjusted \(p = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(*p \leq 0.05\) , \(**p < 0.01\) + +![PLACEHOLDER_57_0] + +
Figure 2
+ +Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(V_{m} = - 60 \text{mV}\) , a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t- test df = 32, \(t = 0.892\) , \(p = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t- test df = 21, \(t = 0.137\) , \(p = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t- test, df = 24, \(t = 0.102\) , \(p = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t- test unequal variances, df = 37, \(t = -0.17\) , \(p = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t- test unequal variances, df = 38, \(t = -0.09\) , \(p = 0.93\) . Data from naïve rats previously published in 16. + +<--- Page Split ---> +![PLACEHOLDER_58_0] + +
Figure 3
+ +Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2- hSynhChR2(H134R)- mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venn diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically- evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One- way ANOVA, df = 4, F = 4.11, p = 0.0057 followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically- evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light- evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was < 2 ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically- evoked LPOLHb EPSCs (n = 12). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more + +<--- Page Split ---> + +GABAergic connections than glutamate connections. (Right) DAMGO inhibited light-evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +![PLACEHOLDER_59_0] + +
Figure 4
+ +LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb ( \(n = 3\) rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar \(= 20 \mu \mathrm{m}\) . (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +<--- Page Split ---> +![PLACEHOLDER_60_0] + +
Figure 5
+ +LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV- hEF1aLS1L- GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb- projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI (n = 5) exhibited significantly larger changes in GCaMP6m fluorescence time- locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls (n = 9): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(p = 0.038\) ; Holm- Sidak post- hoc test, \(p = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2- mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intraLHb saline or DAMGO. d, Representative images of Cre- dependent ChR2- mCherry fluorescence in LPO cell bodies. (Left) Scale bar = \(250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber ("o.f.") implant amidst ChR2- mCherry- expressing cell bodies. Scale bar = \(50 \mu \mathrm{m}\) . e, Only animals with active ChR2 (n = 8), but not mCherry controls (n = 8), developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(p = 0.007\) ; post hoc effect group on test day adjusted \(p = 0.006\) ; paired t- tests, mCherry adjusted \(p = 0.427\) ; ChR2- mCherry adjusted \(p = 0.0027\) . \(^{**}p < 0.01\) , \(^{***}p < 0.005\) + +<--- Page Split ---> diff --git a/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e_det.mmd b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..e546e144b79d964059bb8abf245721d7601fee50 --- /dev/null +++ b/preprint/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e/preprint__16550fdb4530efdecd9dd7dcd06ff8e8f70320f2d568f1b44f9fbd87d94c800e_det.mmd @@ -0,0 +1,901 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 912, 175]]<|/det|> +# A diencephalic circuit for opioid analgesia but not positive reinforcement + +<|ref|>text<|/ref|><|det|>[[44, 196, 390, 238]]<|/det|> +Maggie Wang University of California, San Francisco + +<|ref|>text<|/ref|><|det|>[[44, 244, 390, 285]]<|/det|> +Kayla Maanum University of California, San Francisco + +<|ref|>text<|/ref|><|det|>[[44, 291, 390, 331]]<|/det|> +Joseph Driscoll University of California, San Francisco + +<|ref|>text<|/ref|><|det|>[[44, 337, 215, 377]]<|/det|> +Chris O'Brien Rutgers University + +<|ref|>text<|/ref|><|det|>[[44, 383, 215, 423]]<|/det|> +Svetlana Bryant Rutgers University + +<|ref|>text<|/ref|><|det|>[[44, 429, 390, 470]]<|/det|> +Kasra Mansourian University of California, San Francisco + +<|ref|>text<|/ref|><|det|>[[44, 475, 697, 516]]<|/det|> +Marisela Morales National Institute on Drug Abuse https://orcid.org/0000- 0002- 3845- 9402 + +<|ref|>text<|/ref|><|det|>[[44, 521, 215, 561]]<|/det|> +David Barker Rutgers University + +<|ref|>text<|/ref|><|det|>[[44, 567, 749, 608]]<|/det|> +Elyssa Margolis ( Elyssa.Margolis@ucsf.edu ) University of California, San Francisco https://orcid.org/0000- 0001- 8777- 302X + +<|ref|>sub_title<|/ref|><|det|>[[44, 650, 102, 667]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 687, 789, 707]]<|/det|> +Keywords: Mu opioid receptor agonists, epithelialic lateral habenula, opioid analgesia + +<|ref|>text<|/ref|><|det|>[[44, 725, 317, 744]]<|/det|> +Posted Date: January 6th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 763, 462, 782]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 125555/v1 + +<|ref|>text<|/ref|><|det|>[[44, 800, 910, 844]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 879, 933, 921]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 9th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 28332- 6. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[75, 95, 620, 113]]<|/det|> +1 A diencephalic circuit for opioid analgesia but not positive reinforcement + +<|ref|>text<|/ref|><|det|>[[75, 129, 88, 142]]<|/det|> +2 + +<|ref|>text<|/ref|><|det|>[[75, 168, 857, 188]]<|/det|> +3 Maggie W. Waung \(^{1}\) , Kayla A. Maanum \(^{1}\) , Joseph R. Driscoll \(^{1}\) , Chris O'Brien \(^{2}\) , Svetlana Bryant \(^{2}\) , Kasra + +<|ref|>text<|/ref|><|det|>[[75, 201, 682, 220]]<|/det|> +4 Mansourian \(^{1}\) , Marisela Morales \(^{3}\) , David J. Barker \(^{2,3}\) , and Elyssa B. Margolis \(^{1*}\) + +<|ref|>text<|/ref|><|det|>[[75, 245, 88, 259]]<|/det|> +5 + +<|ref|>text<|/ref|><|det|>[[75, 285, 787, 304]]<|/det|> +6 'UCSF Weill Institute for Neuroscience, Department of Neurology, University of California, + +<|ref|>text<|/ref|><|det|>[[75, 328, 365, 346]]<|/det|> +7 San Francisco, CA, United States + +<|ref|>text<|/ref|><|det|>[[75, 370, 723, 389]]<|/det|> +8 'Department of Psychology, Rutgers University, New Brunswick, NJ, United States + +<|ref|>text<|/ref|><|det|>[[75, 411, 872, 431]]<|/det|> +9 'National Institute on Drug Abuse, Neuronal Networks Section, National Institutes of Health, Baltimore, + +<|ref|>text<|/ref|><|det|>[[75, 445, 255, 462]]<|/det|> +10 MD, United States + +<|ref|>text<|/ref|><|det|>[[75, 488, 88, 501]]<|/det|> +11 + +<|ref|>text<|/ref|><|det|>[[75, 528, 291, 546]]<|/det|> +12 \*Corresponding Author + +<|ref|>text<|/ref|><|det|>[[75, 570, 373, 588]]<|/det|> +13 †Email: Elyssa.Margolis@ucsf.edu + +<|ref|>text<|/ref|><|det|>[[75, 613, 88, 626]]<|/det|> +14 + +<|ref|>text<|/ref|><|det|>[[75, 653, 867, 672]]<|/det|> +15 Title- 10; Abstract- 149, Introduction- 316, Results- 2574, Discussion- 1326; Figures 5, 8 extended data + +<|ref|>text<|/ref|><|det|>[[75, 686, 405, 704]]<|/det|> +16 figures; Tables- 1 Supplementary Table + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[68, 92, 186, 108]]<|/det|> +## 21 Abstract + +<|ref|>text<|/ref|><|det|>[[110, 130, 864, 442]]<|/det|> +21 AbstractMu opioid receptor (MOR) agonists are the most effective analgesics, but their use risks respiratory depression and addiction. The epithelialal lateral habenula (LHb) is a critical site that signals aversive states, often via indirect inhibition of reward circuitry, and MORs are highly expressed in the LHb. We found that the LHb is a potent site for both MOR- agonist analgesia. Strikingly, LHb MOR activation generates negative reinforcement but is not rewarding in the absence of noxious input. While the LHb receives inputs from multiple sites, we found that inputs from the lateral preoptic area of the hypothalamus (LPO) are excited by noxious stimulation, express MOR mRNA, and are preferentially targeted by MOR selective agonists. Critically, optogenetic stimulation of LHb- projecting LPO neurons produces an aversive state relieved by LHb MOR activation. Therefore targeting this MOR sensitive forebrain circuit can relieve pain yet lower the risk of misuse by pain free individuals. + +<|ref|>sub_title<|/ref|><|det|>[[68, 498, 217, 514]]<|/det|> +## 33 Introduction + +<|ref|>text<|/ref|><|det|>[[110, 536, 884, 845]]<|/det|> +33 IntroductionOpioids are the most effective pain medications, but the risk of overdose and opioid use disorder limits their clinical utility. Uncoupling the analgesic actions of opioids from those that underlie positive reinforcement is a longstanding goal for pharmacotherapeutic development. Identifying circuits that can drive relief of ongoing pain but not reward in the absence of pain is a critical step towards this goal. The lateral habenula (LHb) may participate in such a circuit, which is not only activated in a pain setting1–5 but also by other aversive states including reward omission6, and animal models of depression7. Several CNS sites involved in pain signaling with reported strong inputs to the LHb include the lateral hypothalamus (LH)8,9 and anterior cingulate cortex (ACC)10–12. Furthermore, efferents from the LHb target pain-responsive regions including the lateral periaqueductal gray, dorsal raphe, and parabrachial nucleus13,14. While morphine injections that covered a combination of the LHb, medial habenula, and posteromedial + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 857, 141]]<|/det|> +thalamus reduce pain- related behavior in an acute pain model15, whether these effects are due to MOR activation specifically in the LHb is an open question. + +<|ref|>text<|/ref|><|det|>[[112, 163, 879, 470]]<|/det|> +Since increased activity in LHb neurons encodes aversive states including ongoing pain, inhibition of this activity should relieve pain and generate negative reinforcement. We previously found that MOR activation can decrease neural activity in the LHb via both postsynaptic hyperpolarization and inhibition of glutamate release onto subsets of LHb neurons in naïve animals16. Here we investigated the specific LHb input circuit and synaptic mechanism by which MOR activation in the LHb produces pain relief. Among six potential inputs to the LHb, we determined that the glutamatergic innervation from the lateral preoptic area of the hypothalamus (LPO) is both pain- responsive and most strongly inhibited by MOR activation. Importantly, we show that activating MORs in this circuit in the absence of pain does not produce reinforcement, suggesting that targeting this circuit could be a significant advance in pain therapeutics. + +<|ref|>sub_title<|/ref|><|det|>[[115, 528, 175, 544]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 569, 703, 587]]<|/det|> +## MOR activation in the LHb produces pain relief but not positive reinforcement + +<|ref|>text<|/ref|><|det|>[[112, 608, 886, 888]]<|/det|> +To examine the behavioral impact of selective MOR activation in the LHb on ongoing pain, we used the spared nerve injury (SNI) model of persistent neuropathic pain and implanted bilateral cannulae above the LHb in Sprague Dawley rats (Fig. 1a). After recovery from surgery, we evaluated allodynia by measuring mechanical stimulation thresholds with graded von Frey filaments. Bilateral microinjections of the MOR- selective agonist DAMGO (10 \(\mu \mathrm{M}\) ; 300 nL/hemisphere) into the LHb increased the average hindpaw withdrawal threshold compared to saline microinjections in the same animals, indicating that DAMGO reduced the mechanical allodynia generated by SNI in male rats (Fig. 1b). In contrast, intra- LHb DAMGO microinjections in sham- injured male rats had no effect on mechanical withdrawal thresholds compared to saline (Fig. 1b). Consistent with the rat literature17, we did not observe a significant decrease + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[57, 66, 920, 393]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[57, 57, 130, 72]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[54, 410, 940, 781]]<|/det|> +Figure 1. MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500 \mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10 \mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . SNI males: Wilcoxon signed rank test, \(\mathrm{V} = 2\) , \(\mathrm{p} = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with \(\mathrm{SNI} (n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.53\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 31\) , \(\mathrm{p} = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra- LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 15.93\) , \(\mathrm{p} = 0.002\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.01\) ; paired t-tests, sham adjusted \(\mathrm{p} = 1\) ; SNI adjusted \(\mathrm{p} = 0.016\) . Female SNI animals trended towards a preference for the DAMGO- paired chamber: Paired t-test \(\mathrm{p} = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two- way mixed ANOVA, \(\mathrm{F}(1,19) = 2.239\) , \(\mathrm{p} = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100 \mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 5\) , \(\mathrm{p} = 1\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 6\) , \(\mathrm{p} = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100 \mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two- way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 6.234\) , \(\mathrm{p} = 0.027\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.07\) ; paired t-tests, sham adjusted \(\mathrm{p} = 0.48\) ; SNI adjusted \(\mathrm{p} = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} < 0.01\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 880, 207]]<|/det|> +in withdrawal latency to heat in the Hargreaves test after SNI compared to sham- injured controls, and DAMGO microinjections into the LHb did not alter heat withdrawal latency compared to saline in SNI or in sham animals (Extended Data Fig. 1a). Intra- LHb DAMGO also reversed mechanical allodynia induced by inflammatory pain in the CFA model (Extended Data Fig. 1b- g). + +<|ref|>text<|/ref|><|det|>[[111, 225, 880, 377]]<|/det|> +To evaluate whether MOR activation in the LHb influences the affective experience of pain, we used the place conditioning paradigm in the same group of rats. In a three- chamber apparatus, we paired intra- LHb DAMGO microinjections with one chamber and saline microinjections with the opposite side chamber. Rats with SNI developed a significant conditioned place preference (CPP) for the LHb- DAMGO- paired chamber, while sham- injured rats did not prefer either chamber following conditioning (Fig. 1c). + +<|ref|>text<|/ref|><|det|>[[111, 395, 875, 580]]<|/det|> +To rule out the potential confound of off- target effects due to DAMGO entering the CSF space via the nearby third ventricle, we microinjected the same solutions intracerebroventricularly (i.e.v.) in male rats with SNI or sham injury. This manipulation did not influence mechanical withdrawal thresholds compared to saline microinjection in either group (Fig. 1b). Furthermore, i.e.v. DAMGO microinjections did not generate a CPP in either SNI or sham animals (Fig. 1c). Therefore, we conclude that the behavioral effects of our DAMGO microinjections were due to actions specifically in the LHb. + +<|ref|>text<|/ref|><|det|>[[111, 598, 874, 876]]<|/det|> +We also investigated whether LHb MOR activation had the same effects on allodynia and affective pain in female rats. Using the same microinjection parameters as in male rats, female rats with SNI showed a trend towards reduced mechanical allodynia following DAMGO microinjections into the LHb compared to saline (Fig. 1b). Female rats with SNI also showed a trend towards a preference for the DAMGO- paired chamber (Fig. 1c). As in males, females with SNI displayed no difference in heat withdrawal latency between DAMGO and saline microinjections (Extended Data Fig. 1a). Because female rats may be less sensitive to the analgesic effect of opioids18,19, we tested a 10- fold higher concentration of intra- LHb DAMGO (100 μM) in a separate cohort of females. This dose was chosen based on studies demonstrating a three- fold reduction in i.e.v. DAMGO efficacy for females in the tail flick test20. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 89, 880, 174]]<|/det|> +higher DAMGO dose induced a significant CPP, though still did not reverse the mechanical allodynia in female rats (Fig. 1d, e). Thus, we conclude that LHb MOR activation can reverse the affective experience of pain without raising the baseline threshold for reflex withdrawal in both male and female rats. + +<|ref|>sub_title<|/ref|><|det|>[[115, 197, 694, 216]]<|/det|> +## MOR synaptic function persists in LHb neurons in animals with chronic pain + +<|ref|>text<|/ref|><|det|>[[111, 235, 884, 576]]<|/det|> +In various CNS regions, chronic pain induces changes in MOR expression and function, including downregulation21- 23. We previously reported that in naïve male rats MOR activation inhibits glutamate release onto a subset of LHb neurons and also hyperpolarizes approximately \(30\%\) of LHb neurons16. Here we examined whether these MOR effects are altered in a persistent pain state. To evaluate postsynaptic MOR function we performed whole cell voltage clamp recordings of LHb neurons from acute brain slices from male rats with SNI and measured DAMGO induced changes to holding current (Fig. 2a, b). These responses did not differ from our observations in naïve rats. We also tested MOR inhibition of glutamatergic electrically- evoked excitatory postsynaptic currents (EPSCs) in LHb neurons from animals with SNI. These responses were also consistent with observations from naïve rats (Fig. 2c, d). We conclude that these actions of MOR on LHb cell bodies and glutamatergic terminals do not change in animals with ongoing pain. + +<|ref|>text<|/ref|><|det|>[[111, 598, 877, 844]]<|/det|> +Increased activity in LHb neurons as well as increased glutamatergic synaptic strength onto LHb neurons are associated with aversive behavioral states7,24,25. To evaluate this in LHb neurons from animals with SNI, we first measured the paired pulse ratio in the evoked EPSCs, a measure of probability of release. There was no difference in paired pulse ratio between groups (Fig. 2e). We next compared the frequency and magnitude of spontaneous glutamatergic EPSCs (sEPSCs) in animals with SNI to those in naïve animals. Mean sEPSC frequency and amplitude were also similar in LHb neurons from SNI and naïve animals (Fig. 2f, g). Together, these observations suggest a lack of synaptic plasticity induced on glutamatergic inputs to LHb neurons by painful injury. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 115, 420, 570]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[64, 92, 144, 110]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[427, 78, 940, 480]]<|/det|> +Figure 2. Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(\mathrm{V_m} = - 60\) \(\mathrm{mV}\) . a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 32\) , \(\mathrm{t} = 0.892\) , \(\mathrm{p} = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 21\) , \(\mathrm{t} = 0.137\) , \(\mathrm{p} = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t-test, \(\mathrm{df} = 24\) , \(\mathrm{t} = 0.102\) , \(\mathrm{p} = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t-test unequal variances, \(\mathrm{df} = 37\) , \(\mathrm{t} = -0.17\) , \(\mathrm{p} = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t-test unequal variances, \(\mathrm{df} = 38\) , \(\mathrm{t} = -0.09\) , \(\mathrm{p} = 0.93\) . Data from naïve rats previously published in \(^{16}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 875, 110]]<|/det|> +The LHb receives functional synaptic input from the LPO, LH, VTA, VP, and EPN, but not the ACC. + +<|ref|>text<|/ref|><|det|>[[111, 124, 875, 895]]<|/det|> +The LHb receives functional synaptic input from the LPO, LH, VTA, VP, and EPN, but not the ACC.LHb neuron firing activity increases with acute noxious stimulation, and an increase in ongoing firing frequency is during aversive behavioral states2,3,7. In ongoing pain, since we did not observe evidence for changes in glutamatergic synaptic strength, such increases in firing may be driven by greater activity in the glutamatergic axons innervating LHb neurons. Therefore, we hypothesized that the intra- LHb DAMGO- induced behavioral effects that we observed in injured animals were due to MOR inhibition of glutamatergic axon terminals, thus decreasing the aversive excitatory drive onto the LHb neurons. Since MOR activation only inhibits glutamatergic inputs onto a subset of LHb neurons16, MORs might be preferentially expressed on specific afferent inputs. Prior work characterizing direct functional synaptic connections to the LHb is limited to mice26, therefore first we sought to confirm these functional connections in the rat. We investigated inputs from the entopeduncular nucleus (EPN), lateral preoptic area of the hypothalamus (LPO), and ventral tegmental area (VTA) because stimulating glutamatergic LHb inputs from these sources has been shown to be aversive24,27,28. We also investigated inputs from the LH and ACC that are strongly implicated in pain processing29 and the VP because stimulating glutamatergic VP neurons increases the firing rate of LHb neurons30. We injected AAV2-hSyn-hChR2(H134R)-mCherry into one of these six regions in order to express channelrhodopsin (ChR2) in these different input populations (Fig. 3a). We then made whole cell recordings in LHb neurons and measured light-evoked synaptic inputs while blind to injection site. ChR2 was activated by an LED (\(\lambda = 473\)) coupled to an optic fiber placed approximately \(100 \mu \mathrm{m}\) from the recorded cell. Post synaptic currents (PSCs) were measured in response to paired light pulses (1 or 5 ms, 50 ms inter stimulus interval) at holding potentials of -60 mV and -40 mV to probe for EPSCs and GABAAR mediated inhibitory PSCs (IPSCs) in each cell, respectively. Roughly similar proportions of LHb neurons received synaptic input from each of these targets, with the exception of the ACC, where we did not detect any fast PSC connections (Fig. 3b, topographical distribution of connected neurons in Extended Data Fig. 2). The absence of a functional synaptic input from the ACC to the LHb was surprising, as both anterograde10,12 + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[70, 78, 430, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[70, 55, 150, 73]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[443, 80, 941, 576]]<|/det|> +Figure 3. Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2-hSyn-hChR2(H134R)-mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venom diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically-evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One-way ANOVA, \(\mathrm{df} = 4\) , \(\mathrm{F} = 4.11\) , \(\mathrm{p} = 0.0057\) followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically-evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light-evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was \(< 2\) ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically-evoked LPO-LHb EPSCs ( \(\mathrm{n} = 12\) ). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more GABAergic connections than glutamate connections. (Right) DAMGO inhibited light-evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 883, 365]]<|/det|> +and retrograde tracers11 have previously demonstrated modest inputs. Moreover, the ACC is extensively implicated in behavioral responses to pain and MOR- agonist induced pain relief31,32. As a secondary measure of the strength of the innervation, we performed a systematic evaluation of the potential connection using ChR2 as an anterograde tracer, making large injections of AAV2-hSyn-hChR2(H134R)- mCherry throughout the anteroposterior range of the ACC (Extended Data Fig. 3a). This tracing revealed extensive innervation of the nearby mediodorsal thalamus (MDL), but minimal stereologically- quantified labeling in the LHb (Extended Data Fig. 3b, c). Therefore, while we cannot completely rule out a functional input from the ACC to the LHb, the innervation is extremely small compared to the other sources of input to the LHb investigated here. + +<|ref|>text<|/ref|><|det|>[[112, 387, 861, 695]]<|/det|> +For each brain region from which fast synaptic PSCs were detected in the LHb, both glutamate and GABA inputs were observed in varying proportions (Fig. 3c). Interestingly, while for each input there were individual LHb neurons that received both glutamate and GABA synaptic connections, for each input more than half of the connected LHb neurons received just one type of fast PSC, in varying proportions. The VTA was the only input where more LHb neurons received GABAergic synaptic connections than glutamatergic synaptic connections. Among observed synaptic connections, a wide range of EPSC and IPSC amplitudes were observed for most of the inputs, except for the glutamatergic inputs from the VTA that were consistently small (Extended Data Fig. 4a,c). The delay to light evoked EPSC onset also varied across input source, with LH inputs having the shortest mean latency (Extended Data Fig. 4b). + +<|ref|>text<|/ref|><|det|>[[112, 717, 866, 896]]<|/det|> +The nature of local LHb neural connections will also impact the circuit's response to MOR activation. There are strong local glutamatergic connections within the LHb16,33, but there is recent evidence both for34,35 and against36,37 the existence of local GABA interneurons. As we only observed somatodendritic MOR responses in a subset of LHb neurons, evidence for GABAergic interneurons in the LHb would impact our model of how MOR activation modulates LHb neural activity, if such interneurons preferentially express the MOR. In the rat, a small number of GAD1 positive neurons are present in the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 365]]<|/det|> +lateral LHb, though these neurons do not co- express vesicular GABA transporter36, the protein required for loading GABA into synaptic vesicles. In order to detect functional local GABAergic connections within the rat LHb, we injected AAV2- hSyn- hChR2(H134R)- mCherry into the LHb and recorded from LHb neurons. Because some recorded neurons expressed ChR2, we measured light responses before and after application of receptor antagonists in order to isolate the synaptically driven response from the ChR2 mediated currents. Under these conditions we did not observe any light activated local IPSCs in LHb neurons (Extended Data Fig. 5). As expected, many neurons received local glutamatergic inputs (Extended Data Fig. 5). We conclude that there is very limited or no local GABAergic interneuron connectivity in the LHb of adult rats. + +<|ref|>sub_title<|/ref|><|det|>[[115, 388, 576, 408]]<|/det|> +## MOR activation most strongly inhibits LPO inputs to the LHb + +<|ref|>text<|/ref|><|det|>[[111, 427, 880, 899]]<|/det|> +Next, we tested for functional MOR modulation of the light evoked glutamatergic inputs to LHb neurons from each of the regions characterized above. DAMGO induced the strongest and most consistent inhibitions in the terminals arising from LPO neurons (Fig. 3d, e). On average the inhibition was greater in these LPO inputs than the MOR impact on ESPCs observed from the LH, VTA, VP, or EPN (Fig. 3d). We also tested whether MOR inhibits LPO glutamatergic inputs to the LHb independent of sex; the mean inhibition of glutamate release from LPO terminals to LHb neurons in female rats was equivalent to that observed in males (Fig. 3f). Because of the prevalence of local glutamatergic connections in the LHb16,33 (Extended Data Fig. 5) and postsynaptic MOR inhibition of a subset of LHb neurons16, we sought to rule out a polysynaptic connection. First, a polysynaptic contribution seems unlikely for all of the glutamatergic inputs reported here because the delay from light pulse onset to EPSC onset was consistently less than 3 ms (Extended Data Fig. 4b). Second, to directly test isolated monosynaptic connections, we expressed ChR2 in LPO neurons and recorded in the LHb; in neurons with light evoked EPSC responses, we applied tetrodotoxin (TTX, 500 nM) and 4- aminopyridine (4 AP; 10 μM). In 8 of 8 tested neurons the light evoked EPSCs persisted in this monosynaptic signal sparing preparation, and this monosynaptic response was inhibited by DAMGO in all 5 tested neurons (Extended Data Fig. 6). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 864, 172]]<|/det|> +Therefore, we conclude that MORs are functionally expressed on LPO terminals that monosynaptically contact LHb neurons, and when these glutamatergic inputs are activated in vivo, DAMGO application should inhibit them. + +<|ref|>sub_title<|/ref|><|det|>[[115, 197, 545, 215]]<|/det|> +## MOR mRNA is enriched in LHb-projecting LPO neurons + +<|ref|>text<|/ref|><|det|>[[112, 234, 880, 707]]<|/det|> +Multiple basal forebrain structures express high levels of MOR including the VP, medial preoptic area (MPO), horizontal diagonal band (HDB), ventral bed nucleus of the stria terminalis (vBNST), and other regions of the extended amygdala complex (EAC) \(^{38}\) . Some of these not only express MOR to a greater extent than the LPO, but they also project to the LHb. To further evaluate the specificity of MOR expression in LHb-projecting neurons in the LPO compared to nearby brain regions, we performed in situ hybridization for MOR mRNA (OPRM1) in brain slices from Sprague Dawley rats where the retrograde tracer Fluoro- Gold had been iontophoresed into the LHb (Fig. 4a - c). With this independent approach, the LHb-projecting LPO neurons showed the strongest OPRM1 expression and contained the greatest number of retrogradely labeled FG(+) neurons co- labeled for OPRM1 ( \(112 \pm 9\) cells), corresponding to \(57.1\% \pm 4.3\%\) of all MOR(+)FG(+) neurons in the basal forebrain (336/602 total cells). Inputs were also observed from the HDB ( \(23.6\% \pm 5.9\%\) of MOR(+) cells; 153/602 total cells), VP ( \(9.5\% \pm 3.1\%\) of MOR(+) cells; 52/602 total cells), MPO ( \(3.0\% \pm 1.1\%\) of MOR(+) cells; 19/602 total cells), vBNST ( \(3.7\% \pm 1.2\%\) of MOR(+) cells; 23/602 total cells) and EAC ( \(3.0\% \pm 2.1\%\) MOR(+) cells; 19/602 total cells; Fig. 4d - f). Overall, these anatomical data are highly consistent with our electrophysiology results, supporting the conclusion that the LPO projection to the LHb is strongly regulated by MORs. + +<|ref|>sub_title<|/ref|><|det|>[[115, 727, 670, 746]]<|/det|> +## Noxious stimulation activates glutamatergic LHb-projecting LPO neurons + +<|ref|>text<|/ref|><|det|>[[113, 767, 880, 883]]<|/det|> +We next tested whether glutamatergic LHb- projecting LPO neurons are activated by noxious stimulation and whether ongoing pain alters this response. We expressed the calcium indicator GCaMP6m in LHb- projecting LPO neurons using a Cre- dependent, retrograde viral construct HSV- hEF1α- LS1L- GCaMP6m injected to the LHb of VGluT2::Cre mice (Fig. 5a). We implanted optic fibers above the LPO in control + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[57, 68, 784, 455]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 45, 140, 62]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[56, 473, 944, 622]]<|/det|> +Figure 4. LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb (n = 3 rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar = 20 μm. (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 70, 420, 484]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 45, 141, 62]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[430, 77, 941, 485]]<|/det|> +Figure 5. LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV-hEF1α- LSI1-GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb-projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI \((n = 5)\) exhibited significantly larger changes in GCaMP6m fluorescence time-locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls \((n = 9)\) : Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(\mathrm{p} = 0.038\) ; Holm-Sidak post-hoc test, \(\mathrm{p} = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2-mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intra- LHb saline or DAMGO. d, Representative images of Cre-dependent ChR2-mCherry fluorescence in LPO cell bodies. (Left) Scale bar = \(250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber (“o.f.”) implant amidst ChR2-mCherry-expressing cell bodies. Scale bar = \(50 \mu \mathrm{m}\) . e, Only animals with active ChR2 \((n = 8)\) , but not mCherry controls \((n = 8)\) , developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2-mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t-tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2-mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +<|ref|>text<|/ref|><|det|>[[58, 487, 930, 560]]<|/det|> +developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t- tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2- mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 875, 238]]<|/det|> +and SNI mice and performed fiber photometry prior to and during the Hargreaves task. On average, GCaMP6m signal increased as paw withdrawal commenced, and this response was greatly potentiated in mice with SNI (Fig. 5b), who also displayed shorter withdrawal latencies to heat (Extended Data Fig. 7). Thus, glutamatergic LPO neurons that project to the LHb are activated in response to noxious peripheral stimulation, and the magnitude of activation is higher during ongoing pain. + +<|ref|>sub_title<|/ref|><|det|>[[115, 259, 639, 278]]<|/det|> +## LHb MOR activation blocks the aversiveness of LPO-LHb stimulation + +<|ref|>text<|/ref|><|det|>[[111, 298, 875, 740]]<|/det|> +Because LHb- projecting LPO glutamatergic neurons are activated by noxious stimulation, stimulating these terminals in the LHb is aversive28, and MOR activation inhibits these terminals in the LHb, we hypothesized that the aversiveness produced by stimulating the LHb- projecting LPO neurons should be reduced by MOR agonist injection into the LHb. To selectively express ChR2 in LHb- projecting LPO neurons, we used an intersectional viral approach in rats, injecting the retrograde CAV- Cre into the LHb and AAV2- EF1α- DIO- hChR2(H134R)- mCherry (or AAV2- EF1α- DIO- mCherry for controls) into the LPO bilaterally (Fig. 5c,d). Bilateral optic fibers were subsequently implanted above the LPO and cannulae were implanted above the LHb (Fig. 5c). During place conditioning sessions, rats received blue light activation (473 nm, 20 Hz, 5 ms, 10- 12 mW) of the LHb- projecting LPO neurons in both environments of the place conditioning apparatus; chambers were paired with either intra- LHb DAMGO or saline (Fig. 5c). Rats with ChR2 expression developed a CPP for the chamber associated with DAMGO, while control mCherry expressing animals did not (Fig. 5e). Therefore, MOR activation in the LHb blocked the aversiveness of stimulating the LPO input to the LHb without producing positive reinforcement in the absence of input stimulation. + +<|ref|>sub_title<|/ref|><|det|>[[115, 797, 200, 813]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[115, 835, 870, 886]]<|/det|> +Here we identified a circuit that can be targeted by MOR agonists to relieve the aversiveness of ongoing pain but does not produce reward in pain free rodents. MOR agonist action in the LHb was sufficient to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 870, 397]]<|/det|> +reverse injury- induced allodynia and to produce a CPP, decreasing both sensory and affective pain responses, respectively. Importantly, sham- injured animals did not develop a CPP, indicating that MOR activation in the LHb does not produce positive reinforcement in the absence of pain. Unexpectedly, rather than inputs from brain regions previously established as mediating pain and opioid- induced pain relief, it is the glutamatergic inputs from the LPO that we demonstrate here are preferentially controlled by LHb MOR activation. We found that the aversiveness produced by optogenetic activation of LHb- projecting LPO neurons is blocked by MOR agonist microinjection into the LHb, showing that a MOR inhibitory action on this circuit is sufficient to relieve pain. Together, these experiments show that MOR activation in the LHb can generate negative reinforcement via pain relief, but not positive reinforcement in the absence of noxious stimuli. + +<|ref|>sub_title<|/ref|><|det|>[[116, 420, 322, 438]]<|/det|> +## The LHb in pain and relief + +<|ref|>text<|/ref|><|det|>[[111, 460, 880, 867]]<|/det|> +The LHb plays a role in the perception of noxious stimuli in injury and depression models39,40, and most LHb neurons fire more in aversive behavioral states and in response to noxious stimuli2,3 Bilateral lesions of the LHb decrease allodynia in the chronic constriction ischemia model of neuropathic pain in rats41, supporting the notion that signaling through the LHb contributes to injury- induced mechanical allodynia. Chemogenetic inhibition of the LHb relieves thermal hyperalgesia in animals undergoing alcohol withdrawal39. We found that MOR activation in the LHb reverses mechanical allodynia in a model of neuropathic pain in both males and females. Together these observations support our conclusion that MOR activation impacts the behavioral state of the animal by decreasing LHb neural activity in vivo. We previously found two potential mechanisms by which MOR activation could inhibit LHb neural activity in animals with ongoing pain: presynaptically inhibiting glutamate release and postsynaptically driving an outward current in a subset of LHb neurons16. Here we found that intra- LHb MOR activation induces negative reinforcement in both male and female rats with ongoing pain. Yet sham controls (male or female) did not develop a CPP to intra- LHb DAMGO. Because the behavioral effect of MOR activation + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 860, 141]]<|/det|> +was specific to injured rats we hypothesized that increased glutamatergic drive to the LHb, in an input that is MOR sensitive, is a key element in the LHb circuit dynamics involved in pain and relief. + +<|ref|>text<|/ref|><|det|>[[111, 161, 880, 504]]<|/det|> +The vast majority of LHb neurons are glutamatergic, and there is anatomical evidence for dense boutonlike structures arising from local LHb neurons42. This local feed forward connectivity may distribute a specific excitatory LHb input across the various LHb projections that include the dorsal raphe, ventrolateral PAG, MDL, centromedian thalamus, LH, RMTg, and VTA9,13,14,43. Local connectivity has also been observed with functional assays, including that TTX application decreases the frequency of sEPSC and sEPSP events in a subset of LHb neurons16,33 and that optogenetic activation of LHb neurons induces glutamatergic EPSCs (Extended Data Fig. 5). This feed forward circuit enables the distribution of an afferent excitatory signal across LHb neurons, and it raises the possibility that inhibition of such an input by a MOR agonist will decrease activity in LHb projection neurons, even those that are not directly innervated by the excited pathway. Therefore, it is possible that MOR inhibition of one specific input, such as the LPO, can decrease the excitatory drive onto many efferent LHb projections. + +<|ref|>text<|/ref|><|det|>[[111, 524, 874, 833]]<|/det|> +We characterized the functional glutamatergic connections from a variety of brain regions to the LHb in rats; our findings were largely consistent with prior reports utilizing similar techniques in mice. Still, we note that outcomes of these experiments are dependent on the types and number of neurons that express ChR2 following the virus injections and are limited to the geometry of an injection site. We detected the strongest glutamatergic inputs from the LH and EPN, while glutamatergic synaptic responses from the VTA were quite small. We found no functional inputs from the ACC, and in a systematic anatomical analysis we detected very few afferent fibers from the ACC in the LHb. Of the detected glutamatergic inputs, we expected brain regions associated with pain perception and pain relief, such as the LH and VTA, to be more strongly modulated by MOR activation. However, we instead found MOR mRNA and function were clearly enriched in LPO inputs to the LHb. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 539, 110]]<|/det|> +## The LPO: a brain region contributing to pain perception + +<|ref|>text<|/ref|><|det|>[[111, 130, 880, 410]]<|/det|> +Because LHb neurons generally fire more in response to noxious stimuli, and optogenetic stimulation of various glutamatergic inputs to the LHb is uniformly aversive \(^{25,27,28,44,45}\) , we hypothesized that a glutamatergic input to the LHb transmits the pain signal. Further, aversive stressors lead to an increase in the ratio of excitatory glutamatergic to inhibitory GABAergic in synaptic input to LHb neurons \(^{44,45}\) , and restoration of this ratio is associated with relief of aversive states such as foot shock- induced learned helplessness \(^{45}\) and cocaine withdrawal \(^{44}\) . Therefore, a pharmacological manipulation that decreases the excitatory drive onto LHb neurons should also relieve aversive states. Because we found that DAMGO in the LHb generates CPP only in animals with ongoing pain, our data suggest a MOR- sensitive glutamatergic input to the LHb is active during pain and relatively inactive in the absence of pain. + +<|ref|>text<|/ref|><|det|>[[111, 428, 860, 736]]<|/det|> +The LPO projection to the LHb is composed of neurons releasing either glutamate or GABA; these neurons do not co- release glutamate and GABA \(^{28}\) , unlike other LHb inputs \(^{46}\) . Optogenetic activation of LPO glutamate projections to the LHb is aversive \(^{28}\) , and here we posited that this activation mimics an ongoing pain signal. We were able to relieve the aversiveness of activation of this connection with a MOR agonist in the LHb. While a causal role for the LPO in pain perception is unexplored to date, anterograde \(^{47}\) and retrograde \(^{48}\) tracing has shown a direct input to the LPO from the spinal cord. Also, injections of the pro- inflammatory cytokine IL- 1β into the LPO induces hyperalgesia, indicating LPO participation in a nociception circuit \(^{49}\) . Painful stimuli such as subcutaneous formalin injections, mild electric shock and tail pinch also increase firing in some LPO neurons \(^{50,51}\) . Here we show that a major output for LPO pain signals is to the LHb. + +<|ref|>sub_title<|/ref|><|det|>[[115, 758, 526, 778]]<|/det|> +## The LPO to LHb circuit: A unique target for analgesia + +<|ref|>text<|/ref|><|det|>[[112, 799, 870, 884]]<|/det|> +Opioids remain the best available clinical analgesics, yet ongoing systemically administered opioids can result in the serious adverse consequences including opioid use disorder and respiratory depression \(^{52}\) . Opioid- induced positive reinforcement and euphoria, combined with the development of dependence, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 869, 270]]<|/det|> +underlie opioid abuse liability. To date, there is little clinical evidence that the analgesic and euphoric effects of opioids can be decoupled in humans, yet this possibility is a potential pathway to improve therapies for pain. Here we have identified a key circuit whose modulation relieves pain but does not generate reward in the absence of pain. This dissociation between negative and positive reinforcement provides a neural target to achieve pain relief without promoting substance use disorder. Is it possible to activate MORs in the LHb but not in a reward circuit? + +<|ref|>text<|/ref|><|det|>[[111, 290, 884, 600]]<|/det|> +There is some preclinical evidence that this dissociation is possible by developing MOR ligands with the appropriate opioid pharmacology. For example, a novel, cyclized, stabilized MOR selective agonist based on endomorphin I produces pain relief but not reward53. An alternative approach would be to target a different receptor with high expression levels in this circuit in a way that would decrease LHb neural activity. Relevant to this approach, mRNA expression for a selection of orphan G- protein coupled receptors (GPCRs) is enriched in the habenula37,54,55. One such receptor is GPR151, and LHb neurons containing GPR151 receive input from the LPO56. Such directed strategies present a range of new anatomic and molecular targets for pain therapy. By identifying the LPO- LHb connection as a site able to provide relief from neuropathic pain and injury- induced allodynia, we have discovered a unique circuit that may achieve effective opioid- mediated pain relief independent of the drug's addictive properties. + +<|ref|>sub_title<|/ref|><|det|>[[115, 656, 243, 673]]<|/det|> +## Online Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 698, 182, 714]]<|/det|> +## Animals + +<|ref|>text<|/ref|><|det|>[[112, 737, 884, 855]]<|/det|> +All experiments were performed in accordance to the guidelines of the National Institutes of Health Guide for the Care and Use of Laboratory Animals, and the Institutional Animal Care and Use Committees (IACUC) at the University of California, San Francisco, the National Institute on Drug Abuse (NIDA), and Rutgers University. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 882, 205]]<|/det|> +Male and female Sprague Dawley rats were obtained from Charles River Laboratories. Rats were allowed access to food and water ad libitum and maintained on a 12h:12h light/dark cycle. Rats used in behavioral and in situ hybridization studies were housed under reverse light/dark cycle conditions. Rats were group housed until they underwent surgery, after which they were singly housed. + +<|ref|>text<|/ref|><|det|>[[113, 226, 875, 311]]<|/det|> +Male and female VGluT2::Cre mice were bred from mice obtained from the Jackson Laboratory (Jax # 016963). Mice were allowed access to food and water ad libitum and maintained on a 12h:12h light/dark cycle with lights on at 7 AM. Mice were always housed in groups of 2- 5. + +<|ref|>sub_title<|/ref|><|det|>[[115, 335, 239, 352]]<|/det|> +## Viral constructs + +<|ref|>text<|/ref|><|det|>[[112, 374, 880, 528]]<|/det|> +AAV2- hSyn- hChR2(H134R)- mCherry (titer: 2.9e+12), AAV2- hSyn- mCherry (titer: 4.7e+12), and AAV2- EF1α- DIO- ChR2- mCherry (titer: 5.1e+12) were purchased from the University of North Carolina Vector Core with available stock constructs from the laboratory of K. Deisseroth at Stanford University. CAV- Cre (titer: 2.5e+12) was purchased from Montpelier University, France. HSV- hEF1α- GCaMP6m (titer: 5e+9) was purchased from the Gene Delivery Technology Core at Massachusetts General Hospital. + +<|ref|>sub_title<|/ref|><|det|>[[115, 550, 279, 567]]<|/det|> +## Stereotaxic injections + +<|ref|>text<|/ref|><|det|>[[112, 588, 876, 867]]<|/det|> +Rats weighing 275- 300 g were anesthetized with 3- 5% isoflurane (Henry Schein) via inhalation and secured in a stereotaxic frame. Bilateral craniotomies were created with a dental drill above the injection site. For electrophysiology experiments, injections of AAV2- hSyn- hChR2(H134R)- mCherry were made bilaterally into the LPO (- 0.3 mm anteroposterior (AP), \(\pm 1.4 \mathrm{mm}\) mediolateral (ML), - 8.4 mm dorsoventral (DV)), VP (- 0.24 mm AP, \(\pm 2.6 \mathrm{mm}\) ML, - 7.8 mm DV), EPN (- 2.4 mm AP, \(\pm 3.0 \mathrm{mm}\) ML, - 7.0 mm DV), LH (- 2.6 mm AP, \(\pm 1.7 \mathrm{mm}\) ML, - 8.2 mm DV), VTA (- 5.8 mm AP, \(\pm 0.5 \mathrm{mm}\) ML, - 8.5 mm DV), anterior ACC (+2.2 mm AP, \(\pm 0.6 \mathrm{mm}\) ML, - 2.6 mm DV), or posterior ACC (+1.7 AP, \(\pm 0.6 \mathrm{mm}\) ML, - 2.0 mm DV) using a Nanoject II (Drummond Scientific, Broomall, PA). A volume of \(\sim 500 - 830 \mathrm{nL}\) was injected per hemisphere over a period of 4.5 min. The glass injector tip was left in place for at least 2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 835, 140]]<|/det|> +additional minutes before slow withdrawal to prevent backflow and infection of tissue dorsal to the injection target. + +<|ref|>text<|/ref|><|det|>[[112, 164, 857, 188]]<|/det|> +For ChR2 behavior experiments, AAV2- EF1α- DIO- hChR2(H134R)- mCherry or AAV2- EF1α- DIO- + +<|ref|>text<|/ref|><|det|>[[112, 199, 863, 220]]<|/det|> +mCherry were injected into the LPO as above and CAV- Cre was injected bilaterally into the LHb (- 3.7 + +<|ref|>text<|/ref|><|det|>[[112, 231, 830, 252]]<|/det|> +mm AP, \(\pm 0.65 \mathrm{mm} \mathrm{ML}\) , - 5.4 mm DV) with a microinjector connected via polypropylene tubing to + +<|ref|>text<|/ref|><|det|>[[112, 263, 880, 283]]<|/det|> +Hamilton syringes controlled by a dual syringe pump (KD pump) guided by a bilateral 33G stainless steel + +<|ref|>text<|/ref|><|det|>[[112, 295, 330, 313]]<|/det|> +guide cannula (Plastics One). + +<|ref|>text<|/ref|><|det|>[[112, 335, 848, 355]]<|/det|> +Rats were treated with subcutaneous Carprofen (5 mg/kg, Zoetis) and topical \(2\%\) Lidocaine (Phoenix + +<|ref|>text<|/ref|><|det|>[[112, 367, 840, 387]]<|/det|> +Pharmaceutical, Inc.) during the surgery for pain control. After surgery, animals had access to liquid + +<|ref|>text<|/ref|><|det|>[[112, 398, 833, 419]]<|/det|> +Tylenol ( \(\sim 1:40\) ) in their drinking water for 3- 5 days or were administered Meloxicam (s.c. 2mg/kg, + +<|ref|>text<|/ref|><|det|>[[112, 431, 370, 450]]<|/det|> +Pivetal) once per day for two days. + +<|ref|>sub_title<|/ref|><|det|>[[115, 473, 425, 491]]<|/det|> +## Cannulation and optic fiber implantation + +<|ref|>text<|/ref|><|det|>[[112, 515, 880, 732]]<|/det|> +Two to four weeks after virus injection, rats slated for behavioral testing underwent a second cranial surgery to implant custom- made \(200 \mu \mathrm{m}\) optic fibers at a \(5^{\circ}\) angle of rotation in the coronal plane into the bilateral LPO (- 0.6 mm AP, \(\pm 3.05 \mathrm{mm} \mathrm{ML}\) , - 7.5 mm DV). For microinjections into the LHb, bilateral guide cannulae were implanted 1 mm above the LHb (- 3.7 mm AP, \(\pm 0.65 \mathrm{mm} \mathrm{ML}\) , - 4.4 mm DV). A dummy stylet was inserted to maintain patency of the cannulae. Optic fibers and cannulae were anchored with flat point screws and dental cement. For i.c.v. microinjections, unilateral cannulae were implanted into the right lateral ventricle (- 1.0 mm AP, \(+1.5 \mathrm{mm} \mathrm{ML}\) , - 3.5 mm DV). + +<|ref|>text<|/ref|><|det|>[[112, 752, 881, 833]]<|/det|> +Analgesia during surgery and recovery was administered as described above. Animals were allowed to recover for 1- 2 weeks prior to behavioral testing. All virus injections, optic fiber and cannulae placements were histologically verified postmortem based on the standard rat brain atlas57. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 92, 266, 109]]<|/det|> +## Spared nerve injury + +<|ref|>text<|/ref|><|det|>[[112, 130, 875, 410]]<|/det|> +Spared nerve injury (SNI) of the sciatic nerve branch was performed to model chronic neuropathic pain17. Under isoflurane anesthesia, a 2- cm skin incision was made over the left hindlimb. The biceps femoris muscle was blunt dissected to expose the branches of the sciatic nerve. The common peroneal and tibial nerves were ligated with 5.0 silk surgical suture and transected distally, with sparing of the sural nerve branch. For sham procedures, a skin incision was made and biceps femoris muscle was exposed without dissection. The overlying skin was closed with a monocryl suture. Animals were allowed seven days to recover from surgery prior to behavioral testing. A majority of animals demonstrated decreased mechanical withdrawal thresholds after SNI; those that did not demonstrate allodynia were not used in further experiments. + +<|ref|>sub_title<|/ref|><|det|>[[115, 430, 310, 448]]<|/det|> +## Inflammatory pain model + +<|ref|>text<|/ref|><|det|>[[115, 470, 852, 590]]<|/det|> +Peripheral inflammation was induced using Complete Freund's Adjuvant (CFA; Sigma Life Science). Under isoflurane anesthesia, a 1:1 emulsion of CFA and sterile saline (150 μL) was injected into the footpad of the rat's left hindpaw with a 27 G needle. Sham- injured controls were injected with sterile saline (150 μL). + +<|ref|>sub_title<|/ref|><|det|>[[115, 614, 236, 631]]<|/det|> +## Microinjections + +<|ref|>text<|/ref|><|det|>[[113, 653, 880, 806]]<|/det|> +Rats were lightly restrained in a cloth wrap for intracranial microinjections. A bilateral 33G microinjector (PlasticsOne) that extended 1 mm ventrally beyond the guide cannula was inserted to target drug delivery into the LHb or i.c.v. Hamilton syringes were driven by a dual syringe pump to infuse either vehicle (phosphate buffered saline, PBS) or DAMGO (10 μM, 300 nL/hemisphere administered over 2 min). A separate cohort of female rats was microinjected with 100 μM DAMGO. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 130, 884, 405]]<|/det|> +Standard von Frey sensory assessments were performed as described58. Briefly, rats were habituated to sensory testing chambers (Plexiglass boxes with mesh- like flooring) for at least 2 days prior to testing. Behavioral assessments did not begin until exploratory behavior subsided. Testing was completed with eight Touch Test ® fibers (North Coast Medical & Rehabilitation Products, Gilroy, CA, USA) ranging from 0.4 to 15 g. Fibers were pressed perpendicularly to the mid- plantar left hindpaw with sufficient force to cause bending in the fiber and held for 3- 4 s. A positive response was noted if the paw was sharply withdrawn. Ambulation was not considered a positive response. When responding was ambiguous, testing was repeated. 50% withdrawal thresholds were calculated using fiber application in ascending stiffness order or using the up down method. + +<|ref|>text<|/ref|><|det|>[[115, 428, 865, 512]]<|/det|> +In behavioral pharmacology experiments rats underwent sensory testing 5 min after counterbalanced saline or DAMGO microinjections on the same day, with at least 4 h between infusions. Experimenters were blinded to the solution composition during administration and testing. + +<|ref|>sub_title<|/ref|><|det|>[[115, 536, 353, 553]]<|/det|> +## Hargreaves sensory assessment + +<|ref|>text<|/ref|><|det|>[[113, 577, 872, 787]]<|/det|> +Hargreaves tests were completed in sensory testing chambers with glass flooring. Assessment commenced after rats acclimated to the chamber. Testing was completed using a plantar test analgesia meter (Series 8 Model 390, IITC Life Science, Woodhills, CA, USA). Radiant light was directed toward the mid- plantar left hindpaw until a sharp paw withdrawal response was observed with a 30 s maximum cutoff. Ambulation was not considered a positive response. At baseline testing, measurements were repeated at different intensity levels until the average withdrawal latency of eight trials was \(15 \pm 2\) s. Subsequent measurements were performed using this individualized intensity level. + +<|ref|>sub_title<|/ref|><|det|>[[115, 812, 259, 828]]<|/det|> +## Place conditioning + +<|ref|>text<|/ref|><|det|>[[113, 852, 882, 903]]<|/det|> +Conditioned place preference (CPP) pairings occurred twice daily for four consecutive days following the post- surgery sensory tests with DAMGO and saline. The conditioning apparatus (Med. Associates, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 877, 301]]<|/det|> +Georgia, VT, USA) was divided into two chambers (25 cm x 21 cm x 21 cm) with distinct visual (horizontal vs. vertical stripes) and textural (thick vs. thin mesh flooring) cues, separated by a third, smaller gray chamber (12 cm x 21 cm x 21 cm). Before conditioning commenced, animals were allowed up to three opportunities to show neutrality across the chambers during 30- minute baseline sessions. Rats that displayed a consistent baseline preference (>65% of time spent in one chamber) were excluded from the study. Rats were pseudorandomly assigned to receive DAMGO in one of the larger chambers, and assignments were counterbalanced for each cohort. + +<|ref|>text<|/ref|><|det|>[[111, 322, 880, 536]]<|/det|> +During conditioning sessions, microinjections were performed as described above through the intra- LHb cannulae just before the rat was confined to the designated chamber for 30 min. One saline or one DAMGO microinfusion was administered per conditioning session, morning and afternoon pairing sessions were at least 4 h apart, and the order of administration was alternated on each day of conditioning. On test day, rats were allowed to freely explore the chambers for 30 min and time spent (s) in each partition was recorded. Difference score was defined as (Time spent in DAMGO- paired chamber) - (Time spent in saline- paired chamber). + +<|ref|>text<|/ref|><|det|>[[111, 556, 886, 802]]<|/det|> +For ChR2 activation studies, rats were acclimated to handling and attachment of fiber cables to fiber implants in a neutral environment. On procedure days, fiber implants were connected to optic fiber cables attached to a 1x 2 fiber optic rotary joint (Doric Lenses, Quebec, Canada). A laser light source (MBL 473, OEM Laser Systems, East Lansing, MI) was used with light intensity at the end of the output fiber adjusted to 80- 120 mW/mm². Light stimulation (5 ms pulses at 20 Hz) commenced upon placement of rats into the designated chamber of a custom built apparatus on conditioning days. During baseline and testing sessions, the time spent in each chamber was recorded using a webcam and analyzed using Viewer software (Biobserve, Bonn, Germany). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 434, 108]]<|/det|> +## Brain removal and immunohistochemistry + +<|ref|>text<|/ref|><|det|>[[111, 130, 880, 240]]<|/det|> +Rats were deeply anesthetized with an intraperitoneal injection of Euthasol (0.1 mg/kg, Virbac Animal Health, Fort Worth, TX) after 0.3 μl of Chicago Sky Blue (in 2% PBS) was injected through the cannulae to mark injection locations. After becoming unresponsive to noxious stimuli, the rats were transcardially perfused with 400 mL of saline, followed by 400 mL of 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer. The brains were extracted and immersion- fixed in PFA for 2 h at room temperature (RT), washed two times with PBS to remove excess PFA, and stored in 1X PBS at 4°C until they were sectioned (50 μm) using a vibratome (Leica VT 1000 S). Sections were mounted on slides using VECTASHIELD® mounting medium (Vector Laboratories, Burlingame, CA, USA). Images were taken under a Zeiss Stemi 2000- C (Pleasanton, CA) using an Amscope MD800E running AmScope x64 3.0 Imaging Software. + +<|ref|>text<|/ref|><|det|>[[111, 250, 879, 405]]<|/det|> +buffer. The brains were extracted and immersion- fixed in PFA for 2 h at room temperature (RT), washed two times with PBS to remove excess PFA, and stored in 1X PBS at \(4^{\circ}\mathrm{C}\) until they were sectioned (50 μm) using a vibratome (Leica VT 1000 S). Sections were mounted on slides using VECTASHIELD® mounting medium (Vector Laboratories, Burlingame, CA, USA). Images were taken under a Zeiss Stemi 2000- C (Pleasanton, CA) using an Amscope MD800E running AmScope x64 3.0 Imaging Software. + +<|ref|>text<|/ref|><|det|>[[111, 427, 879, 641]]<|/det|> +To label biocytin filled cells after slice electrophysiology recordings, slices were washed three times for 5 min each with PBS (Gibco, Waltham, MA), then blocked with a solution containing: bovine serum albumin \((0.2\%)\) , normal goat serum \((5\%)\) and Tween20 \((0.3\%)\) ; Sigma- Aldrich, St. Louis, MO) for \(2\mathrm{h}\) at RT. Slices were incubated in DTAF- streptavidin (1:200; Jackson Immuno Research) diluted in PBS + \(0.3\%\) Tween20 for \(48\mathrm{h}\) at \(4^{\circ}\mathrm{C}\) . After five, 10- min rinses, brain slices were mounted onto glass slides as above and imaged using a Zeiss Axioskop upright microscope \((2.5\mathrm{X}\) , \(\mathrm{NA} = 0.075\) or Plan Apochromat \(20\mathrm{X}\) , \(\mathrm{NA} = 0.75\) ). + +<|ref|>sub_title<|/ref|><|det|>[[115, 666, 252, 683]]<|/det|> +## Electrophysiology + +<|ref|>text<|/ref|><|det|>[[111, 705, 886, 885]]<|/det|> +Rats were deeply anesthetized with isoflurane, decapitated, and brains were quickly removed into ice- cold artificial cerebrospinal fluid (aCSF) consisting of (in mM): 119 NaCl, 2.5 KCl, 1.0 NaH₂PO₄, 26.2 NaHCO₃, 11 glucose, 1.3 MgSO₄, 2.5 CaCl₂, saturated with 95% O₂- 5% CO₂, with a measured osmolarity 310–320 mOsm/L. Two hundred μm coronal sections through the LHb were cut with a Leica VT 1000 S vibratome. Slices were incubated in oxygenated aCSF at 33 °C and allowed to recover for at least one hour. A single slice was placed in the recording chamber and continuously superfused at a rate of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 883, 336]]<|/det|> +2 mL/min with oxygenated aCSF. Neurons were visualized with an upright microscope (Zeiss AxioExaminer.D1) equipped with infrared-differential interference contrast, Dodt optics, and fluorescent illumination. Whole cell recordings were made at \(34^{\circ}\mathrm{C}\) using borosilicate glass microelectrodes (3- 5 MΩ) filled with K- gluconate internal solution containing (in mM): 123 K- gluconate, 10 HEPES, 8 NaCl, 0.2 EGTA, 2 MgATP, 0.3 \(\mathrm{Na_3GTP}\) , and \(0.1\%\) biocytin (pH 7.2 adjusted with KOH; 275 mOsm/L). Liquid junction potentials were not corrected during recordings. Input and series resistance were monitored throughout voltage clamp experiments with a \(- 4\mathrm{mV}\) step every 30 seconds. Series resistance was required to be 5- 30 MΩ and cells with series resistance changes \(>25\%\) were excluded. + +<|ref|>text<|/ref|><|det|>[[111, 355, 883, 633]]<|/det|> +Signals were recorded using a patch clamp amplifier (Axopatch 1D, Molecular Devices, San Jose, CA or IPA, Sutter Instruments, Novato, CA). Signals were filtered at \(5\mathrm{kHz}\) and collected at \(20\mathrm{kHz}\) using IGOR Pro (Wavemetrics) or collected at \(10\mathrm{kHz}\) using SutterPatch software (Sutter Instruments). Light evoked EPSCs and IPSCs were evoked by two blue light pulses (473 nm, 1- 10 ms) administered 50 ms apart, once every 30 s. LHb recordings were generally made in LHb subregions enriched in ChR2- expressing fibers. Recordings were made in voltage- clamp mode, with membrane potential clamped at \(\mathrm{V_m} = - 60\mathrm{mV}\) and \(- 40\mathrm{mV}\) , for EPSCs and IPSCs respectively. Light was delivered by an LED coupled to an optic fiber aimed at the recorded cell (7- 10 mW). To calculate connectivity rates, only the first neuron patched per slice was included in order to avoid over sampling from slices or animals with lower infection rates. + +<|ref|>text<|/ref|><|det|>[[111, 654, 880, 900]]<|/det|> +Data analysis for electrophysiology. Light pulses were considered to reveal synaptic connections when three conditions were met: (1) the average of 8 traces showed a deviation from baseline \(I_{\mathrm{holding}}\) such that the mean trace exceeded 4 SD of 10 ms baseline period within the 10 ms window after initiation of light pulse, (2) the putative response was observed in at least 3 independent trials, and (3) the delays from the light stimulation onset of putative responses were time locked (<1 ms jitter) across trials. Latency was calculated as time from start of light pulse to when the rate of rise exceeded \(- 40,000\mathrm{V / s}\) . In some cases DNQX (10 \(\mu \mathrm{M}\) ) or gabazine (10 \(\mu \mathrm{M}\) ) was bath applied to confirm inward and outward currents as AMPA or \(\mathrm{GABA_A}\) receptor- mediated, respectively. All measurements of DAMGO effects on EPSCs were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 872, 141]]<|/det|> +completed in the presence of gabazine. After recordings, slices were drop fixed in \(4\%\) PFA for at least 2 h at \(4^{\circ}\mathrm{C}\) and processed for biocytin labeling. + +<|ref|>sub_title<|/ref|><|det|>[[115, 171, 522, 190]]<|/det|> +## Combined retrograde tracing and in situ hybridization + +<|ref|>text<|/ref|><|det|>[[113, 211, 864, 393]]<|/det|> +Tracer injections. Male Sprague Dawley rats (300- 500 g) were anesthetized with \(2 - 5\%\) isoflurane. \(1\%\) Fluoro- Gold (FG; FluoroChrome LLC) solution in a \(0.1\mathrm{M}\) caccodylate buffer (pH 7.5) was delivered unilaterally into the LHb \((- 3.4\mathrm{mmAP},\pm 0.9\mathrm{mmML}\) , and \(- 5.4\mathrm{mmDV}\) ) iontophoretically through a stereotaxically positioned glass micropipette (18- 25 \(\mu \mathrm{m}\) inner diameter) by applying \(1\mu \mathrm{A}\) , 7 s pulses at 14 s intervals for 20 min. The micropipette was then left in place for an additional 10 min to prevent backflow. Following surgery, rats were singly housed and perfused 3 weeks later. + +<|ref|>text<|/ref|><|det|>[[112, 414, 880, 760]]<|/det|> +Tissue Preparation. Rats were anesthetized with chloral hydrate (0.5 ml/kg) and perfused transcardially with \(4\%\) (w/v) PFA in \(0.1\mathrm{M}\) phosphate buffer treated with diethylpyrocarbonate (DEPC), pH 7.3. Brains were post- fixed in \(4\%\) PFA for 2 h before being transferred to an \(18\%\) sucrose solution (w/v in \(0.1\mathrm{M}\) PBS) and stored overnight at \(4^{\circ}\mathrm{C}\) . Coronal sections of the LHb (30 \(\mu \mathrm{m}\) ) and LPO (16 \(\mu \mathrm{m}\) ) were prepared. Phenotyping of retrogradely labeled cells by immunocytochemistry and in situ hybridization. Sections in the LPO were incubated for 2 h at \(30^{\circ}\mathrm{C}\) with rabbit anti- FG antibody (1:500; AB153; Millipore) supplemented with RNAsin. Sections were then incubated in biotinylated goat anti- rabbit antibody (1:200; BA1000; Vector Laboratories) for 1 h at \(30^{\circ}\mathrm{C}\) . Sections were then rinsed and treated with \(0.2\mathrm{N}\) HCl, rinsed, and then acetylated in \(0.25\%\) acetic anhydride in \(0.1\mathrm{M}\) triethanolamine. Subsequently, sections were rinsed and post- fixed with \(4\%\) PFA, rinsed, and then incubated in a hybridization buffer for 2 h at \(55^{\circ}\mathrm{C}\) . + +<|ref|>text<|/ref|><|det|>[[112, 786, 880, 900]]<|/det|> +Hybridization was then performed for radioactive detection of MOR mRNA by hybridizing sections for 16 h at \(55^{\circ}\mathrm{C}\) with [35S]- and [33P]- labeled (107 c.p.m./mL) single- stranded antisense probes. Following hybridization, sections were treated with \(4\mu \mathrm{g / mL}\) of RNAse A at \(37^{\circ}\mathrm{C}\) for 1 h, washed with 1X saline- sodium citrate and \(50\%\) formamide for 1 h at \(55^{\circ}\mathrm{C}\) , and then with \(0.1\mathrm{X}\) saline- sodium citrate at \(68^{\circ}\mathrm{C}\) for 1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 884, 301]]<|/det|> +h. To visualize \(\mathrm{FG(+)}\) cells, sections were rinsed with PBS and incubated for 1 h at RT in avidinbiotinylated horseradish peroxidase (1:100, ABC kit; Vector Laboratories). Sections were then rinsed, and the peroxidase reaction was developed with \(0.05\% 3,3'\) -diaminobenzidine tetrahydrochloride (DAB) and \(0.003\% \mathrm{H}_2\mathrm{O}_2\) . Sections were then photographed under bright field illumination and mounted on coated slides. Finally, slides were dipped in Ilford K.5 nuclear tract emulsion (Polysciences; 1:1 dilution in double-distilled water) and exposed in the dark at \(4^{\circ}\mathrm{C}\) for 3-4 weeks before development and photographs of silver-grain epiluminescence. + +<|ref|>text<|/ref|><|det|>[[111, 323, 881, 534]]<|/det|> +Data analysis of in situ hybridization studies. Methods for analysis of in situ hybridization material have been described previously46. Briefly, pictures were adjusted to match contrast and brightness by using Adobe Photoshop (Adobe Systems). Cell counting was completed independently by three scorers blind to the hypothesis of the study. Radioactive in situ material was analyzed using epiluminescence to increase the contrast of silver grains as described previously59. \(\mathrm{FG(+)}\) cells (detected by fluorescence and brown DAB-label) were evaluated for the presence of MOR mRNA: a cell was considered to express MOR mRNA when its soma contained concentric aggregates of silver grains that exceeded background levels. + +<|ref|>sub_title<|/ref|><|det|>[[115, 558, 250, 575]]<|/det|> +## Fiber photometry + +<|ref|>text<|/ref|><|det|>[[111, 597, 884, 842]]<|/det|> +Surgery. Male and female VGlut2::Cre mice (20- 30 g; 6- 12 weeks) were anesthetized with 1- 5% isoflurane and secured to a stereotaxic frame. Using a Micro4 controller and UltraMicroPump, 0.2 \(\mu \mathrm{L}\) of a retrograde, Cre- dependent HSV encoding GCaMP6m (HSV- hEF1α- LSI1- GCaMP6m) was injected into the LHb (- 1.5 mm AP, +0.45 mm ML, - 3.0 mm DV). Syringes were left in place for 7- 10 min following injections to minimize diffusion. For fiber photometry calcium imaging experiments, a 400 \(\mu \mathrm{m}\) core optic fiber (Doric Lenses) embedded in a 2.5 mm ferrule was implanted over the LPO (+0.5 mm AP, +0.8 mm ML, - 5.05 mm DV) and secured to the skull using #000 screws (Fasteners and Metal products Corp; #000- 120 X 1/16) and dental cement. Following surgery, mice recovered on a warm heating pad before + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 89, 856, 141]]<|/det|> +being transferred back to the vivarium home cage. Three weeks after the virus and fiber surgery, mice were given either SNI or sham control surgery as described above. + +<|ref|>text<|/ref|><|det|>[[111, 165, 883, 480]]<|/det|> +Recording. Signals from GCaMP6 were recorded across 10 trials of stimulation in the Hargreaves test using a Plantar Test Instrument. The onset and offset time for each trial was digitized and sent to an RZ5D (Tucker Davis Technologies). For the acquisition of LPO \(\rightarrow\) LHB activity, GCaMP6 was excited at two wavelengths (490nm, calcium- dependent signal and 405 nm isosbestic control) by amplitude modulated signals from two light- emitting diodes reflected off dichroic mirrors and coupled into a 400μm 0.48NA optic fiber. Signals emitted from GCaMP6m and its isosbestic control channel then returned through the same optic fiber and were acquired using photoreceiver (Doric Lenses), digitized at 1kHz, and then recorded by a real- time signal processor (RZ5D; Tucker Davis Technologies) running the Synapse software suite. Analysis of the resulting signal was then performed using custom- written MATLAB scripts available in a general release form at + +<|ref|>text<|/ref|><|det|>[[112, 488, 876, 796]]<|/det|> +https://github.com/djamesbarker/FiberPhotometry. Briefly, changes in fluorescence across the experimental session ( \(\Delta \mathrm{F} / \mathrm{F}\) ) were calculated by smoothing signals from the isosbestic control channel, scaling the isosbestic control signal by regressing it on the smoothed GCaMP signal, and then generating a predicted 405 nm signal using the linear model generated during the regression. Calcium independent signals on the predicted 405 nm channel were then subtracted from the raw GCaMP signal to remove movement, photo- bleaching, and fiber bending artifacts. Signals from the GCaMP channel were then divided by the control signal to generate the \(\Delta \mathrm{F} / \mathrm{F}\) . Peri- event histograms were then created by averaging changes in fluorescence ( \(\Delta \mathrm{F} / \mathrm{F}\) ) across repeated trials during windows encompassing behavioral events of interest. The area under the curve (AUC) was calculated for a pre- stimulation 5 s baseline commencing - 10 s before paw withdrawal and for the 5 s period initiated with paw withdrawal. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 379, 109]]<|/det|> +## Anterograde tracing from the ACC + +<|ref|>text<|/ref|><|det|>[[111, 128, 880, 630]]<|/det|> +In rats, unilateral injections of AAV2- hSyn- hChR2(H134R)- mCherry (759 nL) were made throughout anteroposterior range of the ACC (+2.6 to - 0.4 mm AP, - 0.4 to - 0.5 mm ML, - 2.6 to - 2.8 mm DV). Rats were perfused and brains fixed five weeks later, as described above. Coronal sections (50 \(\mu \mathrm{m}\) ) containing the ACC and the LHb were collected. After verification of the injection site, every sixth slice containing the LHb was rinsed twice with PBS. Tissue was pre- permeabilized in 1:1 EtOH:PBS for 30 min at \(4^{\circ}\mathrm{C}\) , rinsed briefly in PBS before blocked in \(3\% \mathrm{H}_2\mathrm{O}_2\) for 10 min. Following PBS washes (3x5 min), tissue was blocked in solution containing normal goat serum (NGS, \(10\%\) ) for 1 h at RT. Slices were incubated in rabbit anti- mCherry antibody (1:5000 in PBS + 0.3% Triton X100 + NGS \(10\%\) ; Abcam) overnight at \(4^{\circ}\mathrm{C}\) . After PBS washes (4x10 min), slices were incubated in biotinylated goat anti- rabbit secondary antibody (1:200 in PBS; Vector Laboratories) for 2 h at \(4^{\circ}\mathrm{C}\) . Following PBS wash (4x10 min), slices were incubated in VECTASTAIN® ABC Reagent (VECTASTAIN® ABC Kit, Vector Laboratories) for 30 min followed by peroxidase substrate (DAB Substrate Kit, Vector Laboratories) for 10 min. Once dark brown DAB precipitate formed, slices were rinsed in PBS (5x5 min), mounted on glass slides, and cover slipped with DEPEX (Electron Microscopy Sciences). DAB- stained fibers were visualized under brightfield illumination and quantified in Stereo Investigator software (MBF Bioscience) using the virtual isotropic space balls probe60. + +<|ref|>sub_title<|/ref|><|det|>[[115, 654, 333, 672]]<|/det|> +## General experimental design + +<|ref|>text<|/ref|><|det|>[[112, 692, 884, 842]]<|/det|> +For behavioral experiments, subject numbers were determined by pilot studies and power analyses (power \(= 0.80\) , significance level \(= 0.05\) , effect size \(= 15 - 30\%\) ). All behavioral experiments were performed blinded to experimental condition. For immunohistological experiments, three animals with injections targeted at the anterior, middle, and posterior ACC were used to obtain a comprehensive estimation of fibers projecting to the LHb. Electrophysiology experiments were conducted blind to injection site. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 400, 109]]<|/det|> +## Quantification and statistical analysis + +<|ref|>text<|/ref|><|det|>[[112, 130, 877, 375]]<|/det|> +Data are expressed as mean \(\pm\) SEM or mean with \(25^{\mathrm{th}}\) and \(75^{\mathrm{th}}\) percentiles as indicated in figure legends and text. Significance was set at \(\mathrm{p}< 0.05\) . Datasets were evaluated to determine whether parametric or non- parametric statistical approaches were most appropriate as indicated in Supplemental Table 1. All tests were two tailed, and statistical analyses were performed in GraphPad Prism or R. The bandwidth for electrophysiology violin plots was determined by Silverman's rule of thumb in Plotly for Python; violin plots for behavior experiments were constructed with heavy kernel density estimations in Prism. Sample sizes are reported in figure panels, legends, and Supplementary Table 1. Outliers, including extreme outliers, are reported in Supplementary Table 1 and were not removed from datasets. + +<|ref|>sub_title<|/ref|><|det|>[[115, 446, 274, 464]]<|/det|> +## Reporting summary + +<|ref|>text<|/ref|><|det|>[[115, 486, 874, 536]]<|/det|> +Further information on research design is available in the Nature Research Reporting Summary linked to this paper. + +<|ref|>sub_title<|/ref|><|det|>[[115, 608, 247, 625]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 650, 852, 700]]<|/det|> +All data described in the main text or extended data are available from the corresponding author upon request. + +<|ref|>sub_title<|/ref|><|det|>[[115, 773, 268, 789]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[115, 813, 879, 864]]<|/det|> +The authors would like to thank Ryan Carothers, Gabrielle Mintz, Lucy He, Venkateswaran Ganesh, and Benjamin Snyder for their technical assistance with histology, stereotaxic surgeries, and behavioral + +<|ref|>text<|/ref|><|det|>[[115, 876, 880, 896]]<|/det|> +studies. This work was supported by National Institutes of Health grants R01DA042025 (to E.B.M.), K08 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 860, 141]]<|/det|> +NS097632 (to M.W.W.), and the Intramural Research Program (IRP) of the National Institute on Drug Abuse (IRP/NIDA/NIH) and a NIDA K99/R00 pathway to independence award (DA043572) to D.J.B. + +<|ref|>sub_title<|/ref|><|det|>[[115, 199, 358, 216]]<|/det|> +## Competing Interests Statement + +<|ref|>text<|/ref|><|det|>[[115, 240, 415, 258]]<|/det|> +The authors have no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[115, 303, 239, 320]]<|/det|> +## Figure Legends + +<|ref|>text<|/ref|><|det|>[[112, 340, 881, 877]]<|/det|> +Figure 1. MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500 \mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10 \mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . SNI males: Wilcoxon signed rank test, \(\mathrm{V} = 2\) , \(\mathrm{p} = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with SNI \((n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 7\) , \(\mathrm{p} = 0.53\) . SNI: Wilcoxon signed rank test, \(\mathrm{V} = 31\) , \(\mathrm{p} = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra-LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 15.93\) , \(\mathrm{p} = 0.002\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.01\) ; paired t-tests, sham adjusted \(\mathrm{p} = 1\) ; SNI adjusted \(\mathrm{p} = 0.016\) . Female SNI animals trended towards a preference for the DAMGO-paired chamber: Paired t-test \(\mathrm{p} = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two-way mixed ANOVA, \(\mathrm{F}(1,19) = 2.239\) , \(\mathrm{p} = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100 \mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(\mathrm{V} = 5\) , \(\mathrm{p} = 1\) . SNI: Wilcoxon signed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 876, 270]]<|/det|> +rank test, \(\mathrm{V} = 6\) , \(\mathrm{p} = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100~\mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two- way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(\mathrm{F}(1,13) = 6.234\) , \(\mathrm{p} = 0.027\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.07\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.48\) ; SNI adjusted \(\mathrm{p} = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(^{*}\mathrm{p}\) \(\leq 0.05\) , \(^{**}p< 0.01\) + +<|ref|>text<|/ref|><|det|>[[111, 336, 884, 802]]<|/det|> +Figure 2. Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) . a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 32\) , \(\mathrm{t} = 0.892\) , \(\mathrm{p} = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t-test \(\mathrm{df} = 21\) , \(\mathrm{t} = 0.137\) , \(\mathrm{p} = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t-test, \(\mathrm{df} = 24\) , \(\mathrm{t} = 0.102\) , \(\mathrm{p} = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t-test unequal variances, \(\mathrm{df} = 37\) , \(\mathrm{t} = -0.17\) , \(\mathrm{p} = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t-test unequal variances, \(\mathrm{df} = 38\) , \(\mathrm{t} = -0.09\) , \(\mathrm{p} = 0.93\) . Data from naïve rats previously published in \(^{16}\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 880, 620]]<|/det|> +Figure 3. Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2- hSyn- hChR2(H134R)- mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venn diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically- evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One- way ANOVA, \(\mathrm{df} =\) 4, \(\mathrm{F} = 4.11\) , \(\mathrm{p} = 0.0057\) followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically- evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light- evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was \(< 2\) ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically- evoked LPO- LHb EPSCs ( \(\mathrm{n} = 12\) ). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more GABAergic connections than glutamate connections. (Right) DAMGO inhibited light- evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +<|ref|>text<|/ref|><|det|>[[111, 686, 879, 896]]<|/det|> +Figure 4. LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb ( \(\mathrm{n} = 3\) rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar \(= 20 \mu \mathrm{m}\) . (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 885, 141]]<|/det|> +f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +<|ref|>text<|/ref|><|det|>[[111, 201, 864, 800]]<|/det|> +Figure 5. LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV- hEF1α- LS1L- GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb- projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI (green, \(\mathrm{n} = 5\) ) exhibited significantly larger changes in GCaMP6m fluorescence time- locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls (grey, \(\mathrm{n} = 9\) ): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2- mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intra- LHb saline or DAMGO. d, Representative images of Cre- dependent ChR2- mCherry fluorescence in LPO cell bodies. (Left) Scale bar \(= 250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber (“o.f.”) implant amidst ChR2- mCherry- expressing cell bodies. Scale bar \(= 50 \mu \mathrm{m}\) . e, Only animals with active ChR2 ( \(\mathrm{n} = 8\) ), but not mCherry controls ( \(\mathrm{n} = 8\) ), developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(\mathrm{p} = 0.007\) ; post hoc effect group on test day adjusted \(\mathrm{p} = 0.006\) ; paired t- tests, mCherry adjusted \(\mathrm{p} = 0.427\) ; ChR2- mCherry adjusted \(\mathrm{p} = 0.0027\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) + +<|ref|>title<|/ref|><|det|>[[115, 867, 357, 884]]<|/det|> +# Extended Data Figure Legends + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 883, 880]]<|/det|> +Extended Data Figure 1. Further characterization of neuropathic pain and inflammatory pain models and the impact of intra- LHb MOR activation in rat. a, (Left) Violin plots of withdrawal latency to heat in sham \((\mathrm{n} = 9)\) , male \((\mathrm{n} = 8)\) and female \((\mathrm{n} = 9)\) rats with SNI after saline vs DAMGO ( \(10~\mu \mathrm{M}\) ) microinjections into the LHb. Sham males: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) ; SNI males: Paired t- test, \(\mathrm{df} = 7\) , \(\mathrm{t} = - 1.92\) , \(\mathrm{p} = 0.096\) ; SNI females: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 1.4\) , \(\mathrm{p} = 0.199\) . (Right) i.c.v. sham male: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 0.08\) , \(\mathrm{p} = 0.938\) . i.c.v. SNI male: Paired t- test, \(\mathrm{df} = 11\) , \(\mathrm{t} = 3.2\) , \(\mathrm{p} = 0.00846\) . To test if the effects of MOR activation in the LHb generalize to other forms of pain, we repeated the manipulations above with an inflammatory model of pain. Male rats received an intradermal injection of CFA or sterile saline into the plantar aspect of the hindpaw, b, Schematic diagram of inflammatory injury preparations and cannulation targeting the LHb. c, Mechanical withdrawal thresholds in sham and CFA- injured rats. In animals with CFA, we observed an increase in the average withdrawal threshold following intra- LHb DAMGO ( \(10~\mu \mathrm{M}\) ) compared to saline, indicating a reduction in mechanical allodynia: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 2.89\) , \(\mathrm{p} = 0.0202\) ; but not shams Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . d, Withdrawal latency to heat. Similar to results in rats with SNI, DAMGO in the LHb did not reverse the CFA- induced decrease in withdrawal latency to heat. Sham: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) . CFA: Wilcoxon signed rank test, \(\mathrm{V} = 12\) , \(\mathrm{p} = 0.25\) . e, Rats with CFA injury also did not develop a preference for the DAMGO- paired chamber in the place conditioning paradigm: Two- way mixed ANOVA, significant interaction between CFA/sham and baseline/test \(\mathrm{F}(1,14) = 0.474\) , \(\mathrm{p} = 0.502\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.286\) ; SNI adjusted \(\mathrm{p} = 0.761\) . The discrepancy between place conditioning results of our neuropathic and inflammatory pain animals may be due to the natural history of the injury caused in the CFA model: whereas SNI animals underwent permanent nerve ligation, animals injected with CFA presented with transient swelling and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 877, 237]]<|/det|> +tested daily; each line represents a male rat. \(\mathbf{g}\) , Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(^{*}\mathrm{p}\leq 0.05\) , \(^{**}p\leq 0.01\) + +<|ref|>text<|/ref|><|det|>[[112, 301, 884, 480]]<|/det|> +Extended Data Figure 2. Neurons receiving functional synaptic inputs from the five brain regions showing connectivity were distributed throughout the LHb. Locations of neurons found connected using whole- cell recordings from optogenetic stimulation of terminals arising from each input tested. Locations are based on biocytin immunohistochemistry and low magnification images taken on the recording microscope where the recorded cell is centered within the field of view. Color indicates source of innervation. + +<|ref|>text<|/ref|><|det|>[[112, 545, 884, 821]]<|/det|> +Extended Data Figure 3. ACC minimally innervates the LHb. a, Diagram of the extent of unilateral AAV2- hSyn- hChR2(H134R)- mCherry injection sites ( \(\mathrm{n} = 9\) male rats) throughout anteroposterior range of the ACC for anterograde tracing study. b, Example ipsilateral DAB- positive fibers (black) visualized under brightfield illumination. Fibers heavily innervate the MDL, which abuts the lateral edge of the LHb, while sparse to no fibers innervate the LHb. Contralateral innervation of LHb and MDL was negligible compared to the ipsilateral side, and therefore were omitted from our analysis. (Left) Scale bar \(= 250 \mu \mathrm{m}\) . (Right) Scale bar \(= 50 \mu \mathrm{m}\) . c, Average stereologically- quantified DAB- positive fiber lengths throughout anteroposterior range of the LHb, compared to the MDL innervation in the same coronal slice. Mann- Whitney test, two- tailed: \(\mathrm{U} = 0\) , \(\mathrm{p} = 0.0006\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 875, 430]]<|/det|> +Extended Data Figure 4. Sources of glutamatergic inputs to the rat LHb vary in strength and delay in synaptic transmission. a, We observed some variations between inputs in the mean light evoked EPSCs \(\mathrm{(V_{m} = - 60 mV)}\) . In particular, excitatory inputs from the VTA were consistently small: Kruskal- Wallis \(\chi 2\) \(= 25.5\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.00004\) , followed by Dunn Test for pairwise comparisons. b, Differences were also detected in the delay to the onset of light evoked EPSCs to LHb neurons from these different sources, with inputs from LH showing the fastest response. Although these data deviate from a normal distribution (Shapiro's test, \(\mathrm{p} = 0.0000001\) ), KDEs (violins) are consistent with continuous distributions, suggesting reliable polysynaptic events were rarely detected. c, We observed some variations between inputs in the mean light evoked IPSCs \(\mathrm{(V_{m} = - 40 mV)}\) . While there were no statistically significant differences in amplitudes detected, the mean inhibitory input from the VP was particularly small: Kruskal- Wallis \(\chi 2 = 2.3\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.7\) . \(^{**}p< 0.01\) , \(^{***}p< 0.005\) , \(^{****}p< 0.0005\) + +<|ref|>text<|/ref|><|det|>[[112, 494, 881, 896]]<|/det|> +Extended Data Figure 5. Optogenetic experiments detect local glutamate, but not GABA, functional connections in rat LHb. AAV2-hSyn-hChR2(H134R)- mCherry was stereotaxically injected into the LHb at least 4 weeks prior to ex vivo whole cell recordings in the LHb to detect local synaptic connections. Neurons were recorded blind to ChR2 expression, therefore in some cases the patched neuron expressed ChR2. Therefore, in addition to connectivity criteria used for other afferent inputs, in these experiments only light evoked inward currents that were blocked by \(10 \mu \mathrm{M}\) DNQX were considered glutamatergic connections, and direct ChR2 induced inward currents were subtracted out for the quantification illustrated here. Each cell was probed for both glutamate and GABA inputs in voltage clamp by holding neurons at \(\mathrm{V_{m} = - 60 mV}\) and \(- 40 \mathrm{mV}\) , light pulse durations 1, 5, and \(10 \mathrm{ms}\) durations. a, Example recording at \(\mathrm{V_{m} = - 60 mV}\) showing a light evoked response that was blocked by DNQX. b, Example recording at \(\mathrm{V_{m} = - 40 mV}\) with minimal outward current response within \(7 \mathrm{ms}\) of light pulse. c, Summary of all LHb recordings tested in this experiment. Filled circles represent cells where responses could be classified as local ChR2 induced synaptic transmission. When no clear response was detected, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 881, 204]]<|/det|> +measure indicated is the difference between the mean \(I_{\text{holding}}\) of the baseline 100 ms period just prior to the light pulse and the mean \(I_{\text{holding}}\) 2 ms period starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V_m} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +<|ref|>text<|/ref|><|det|>[[113, 270, 883, 483]]<|/det|> +Extended Data Figure 6. Isolated monosynaptic glutamatergic inputs from LPO to LHb neurons are inhibited by MOR activation. a, Example light evoked EPSC responses in an LHb neuron from a rat with ChR2 expression in LPO neurons. This response persisted in monosynaptic isolation by \(500 \mathrm{nM}\) TTX and \(100 \mu \mathrm{M} 4 \mathrm{AP}\) (green), and this isolated response was inhibited by \(500 \mathrm{nM}\) DAMGO (magenta). b, Summary of DAMGO effects on isolated monosynaptic EPSC inputs to LHb neurons expressed as \(\%\) of baseline monosynaptic response (left) and as raw EPSC magnitudes (right). Paired t-test, \(\mathrm{df} = 4\) , \(\mathrm{t} = - 5.1\) , \(\mathrm{p} = 0.007\) . \(^{**} \mathrm{p} < 0.01\) + +<|ref|>text<|/ref|><|det|>[[113, 545, 876, 789]]<|/det|> +Extended Data Figure 7. Mice with SNI show hypersensitivity to heat and increased activity in LHb- projecting LPO neurons during paw withdrawal from thermal stimulation. a, Mice with SNI show a reduced latency to withdraw their paw following thermal stimulation in the Hargreaves task. Unpaired t- test, \(\mathrm{t}(12) = 3.007\) , \(\mathrm{p} = 0.011\) . b, VGluT2- expressing LPO neurons that project to the LHb expressed GCaMP6m and showed a greater calcium response during paw withdrawal to Hargreaves thermal stimulation (area under the curve, deviation from baseline fluorescence) in SNI animals ( \(\mathrm{n} = 5\) ) compared to sham controls ( \(\mathrm{n} = 9\) ): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.4\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . \(^{**} \mathrm{p} < 0.01\) + +<|ref|>text<|/ref|><|det|>[[113, 852, 870, 905]]<|/det|> +Extended Data Figure 8. Locations and spread of bilateral ChR2 injections to the LPO in rats used for in vivo optogenetic experiments. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 88, 870, 888]]<|/det|> +References1. Matsumoto, M. & Hikosaka, O. Representation of negative motivational value in the primate lateral habenula. Nat Neurosci 12, 77–84 (2009).2. Benabid, A. L. & Jeaugey, L. Cells of the rat lateral habenula respond to high-threshold somatosensory inputs. Neurosci Lett 96, 289–94 (1989).3. Dafny, N. & Qiao, J. T. Habenular neuron responses to noxious input are modified by dorsal raphe stimulation. Neurol. Res. 12, 117–121 (1990).4. Nagao, M., Kamo, H., Akiguchi, I. & Kimura, J. Induction of c-Fos-like protein in the lateral habenular nucleus by persistent noxious peripheral stimulation. Neurosci. Lett. 151, 37–40 (1993).5. Huang, T. et al. Identifying the pathways required for coping behaviours associated with sustained pain. Nature 565, 86–90 (2019).6. Matsumoto, M. & Hikosaka, O. Lateral habenula as a source of negative reward signals in dopamine neurons. Nature 447, 1111–1115 (2007).7. Li, B. et al. Synaptic potentiation onto habenula neurons in the learned helplessness model of depression. Nature 470, 535–539 (2011).8. Poller, W. C., Madai, V. I., Bernard, R., Laube, G. & Veh, R. W. A glutamatergic projection from the lateral hypothalamus targets VTA-projecting neurons in the lateral habenula of the rat. Brain Res 1507, 45–60 (2013).9. Herkenham, M. & Nauta, W. J. Afferent connections of the habenular nuclei in the rat. A horseradish peroxidase study, with a note on the fiber-of-passage problem. J Comp Neurol 173, 123–46 (1977).10. Vertes, R. P. Analysis of projections from the medial prefrontal cortex to the thalamus in the rat, with emphasis on nucleus reuniens. J. Comp. Neurol. 442, 163–187 (2002).11. Yetnikoff, L., Cheng, A. Y., Lavezzi, H. N., Parsley, K. P. & Zahm, D. S. Sources of input to the rostromedial tegmental nucleus, ventral tegmental area, and lateral habenula compared: A study in rat. J Comp Neurol 523, 2426–56 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 879, 140]]<|/det|> +12. Kim, U. & Lee, T. Topography of descending projections from anterior insular and medial prefrontal regions to the lateral habenula of the epithalamus in the rat. Eur J Neurosci 35, 1253–69 (2012). + +<|ref|>text<|/ref|><|det|>[[110, 154, 870, 204]]<|/det|> +13. Araki, M., McGeer, P. L. & Kimura, H. The efferent projections of the rat lateral habenular nucleus revealed by the PHA-L anterograde tracing method. Brain Res. 441, 319–330 (1988). + +<|ref|>text<|/ref|><|det|>[[110, 219, 840, 269]]<|/det|> +14. Quina, L. A. et al. Efferent pathways of the mouse lateral habenula. J Comp Neurol 523, 32–60 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 283, 872, 334]]<|/det|> +15. Cohen, S. R. & Melzack, R. Morphine injected into the habenula and dorsal posteromedial thalamus produces analgesia in the formalin test. Brain Res. 359, 131–139 (1985). + +<|ref|>text<|/ref|><|det|>[[110, 348, 870, 398]]<|/det|> +16. Margolis, E. B. & Fields, H. L. Mu Opioid Receptor Actions in the Lateral Habenula. PLoS One 11, e0159097 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 412, 800, 462]]<|/det|> +17. Decosterd, I. & Woolf, C. J. Spared nerve injury: an animal model of persistent peripheral neuropathic pain. PAIN 87, 149–158 (2000). + +<|ref|>text<|/ref|><|det|>[[110, 476, 860, 526]]<|/det|> +18. Miaskowski, C. et al. Does opioid analgesia show a gender preference for females? Pain Forum 8, 34–44 (1999). + +<|ref|>text<|/ref|><|det|>[[110, 540, 880, 590]]<|/det|> +19. Kest, B., Sarton, E. & Dahan, A. Gender differences in opioid-mediated analgesia: animal and human studies. Anesthesiology 93, 539–547 (2000). + +<|ref|>text<|/ref|><|det|>[[110, 604, 777, 623]]<|/det|> +20. Kepler, K. L. et al. Gender effects and central opioid analgesia. Pain 45, 87–94 (1991). + +<|ref|>text<|/ref|><|det|>[[110, 637, 870, 686]]<|/det|> +21. Thompson, S. J. et al. Chronic neuropathic pain reduces opioid receptor availability with associated anhedonia in rat. Pain 159, 1856–1866 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 700, 870, 750]]<|/det|> +22. Porreca, F. et al. Spinal opioid mu receptor expression in lumbar spinal cord of rats following nerve injury. Brain Res. 795, 197–203 (1998). + +<|ref|>text<|/ref|><|det|>[[110, 764, 844, 814]]<|/det|> +23. Zhang, X. et al. Down-regulation of mu-opioid receptors in rat and monkey dorsal root ganglion neurons and spinal cord after peripheral axotomy. Neuroscience 82, 223–240 (1998). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 850, 171]]<|/det|> +24. Shabel, S. J., Proulx, C. D., Trias, A., Murphy, R. T. & Malinow, R. Input to the lateral habenula from the basal ganglia is excitatory, aversive, and suppressed by serotonin. Neuron 74, 475–481 (2012). + +<|ref|>text<|/ref|><|det|>[[110, 185, 833, 234]]<|/det|> +25. Lecca, S. et al. Aversive stimuli drive hypothalamus-to-habenula excitation to promote escape behavior. eLife 6, (2017). + +<|ref|>text<|/ref|><|det|>[[110, 248, 855, 298]]<|/det|> +26. Hu, H., Cui, Y. & Yang, Y. Circuits and functions of the lateral habenula in health and in disease. Nat. Rev. Neurosci. 21, 277–295 (2020). + +<|ref|>text<|/ref|><|det|>[[110, 312, 870, 363]]<|/det|> +27. Root, D. H., Mejias-Aponte, C. A., Qi, J. & Morales, M. Role of glutamatergic projections from ventral tegmental area to lateral habenula in aversive conditioning. J Neurosci 34, 13906–10 (2014). + +<|ref|>text<|/ref|><|det|>[[110, 377, 866, 427]]<|/det|> +28. Barker, D. J. et al. Lateral Preoptic Control of the Lateral Habenula through Convergent Glutamate and GABA Transmission. Cell Rep. 21, 1757–1769 (2017). + +<|ref|>text<|/ref|><|det|>[[110, 440, 770, 460]]<|/det|> +29. Fields, H. State-dependent opioid control of pain. Nat Rev Neurosci 5, 565–75 (2004). + +<|ref|>text<|/ref|><|det|>[[110, 473, 820, 523]]<|/det|> +30. Tooley, J. et al. Glutamatergic Ventral Pallidal Neurons Modulate Activity of the Habenula-Tegmental Circuitry and Constrain Reward Seeking. Biol. Psychiatry 83, 1012–1023 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 536, 872, 617]]<|/det|> +31. Johansen, J. P., Fields, H. L. & Manning, B. H. The affective component of pain in rodents: direct evidence for a contribution of the anterior cingulate cortex. Proc. Natl. Acad. Sci. U. S. A. 98, 8077–8082 (2001). + +<|ref|>text<|/ref|><|det|>[[110, 631, 872, 681]]<|/det|> +32. LaGraize, S. C., Borzan, J., Peng, Y. B. & Fuchs, P. N. Selective regulation of pain affect following activation of the opioid anterior cingulate cortex system. Exp. Neurol. 197, 22–30 (2006). + +<|ref|>text<|/ref|><|det|>[[110, 695, 884, 777]]<|/det|> +33. Kim, U. & Chang, S. Y. Dendritic morphology, local circuitry, and intrinsic electrophysiology of neurons in the rat medial and lateral habenular nuclei of the epithalamus. J Comp Neurol 483, 236–50 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 792, 877, 842]]<|/det|> +34. Flanigan, M. E. et al. Orexin signaling in GABAergic lateral habenula neurons modulates aggressive behavior in male mice. Nat. Neurosci. 23, 638–650 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 870, 144]]<|/det|> +35. Zhang, L. et al. A GABAergic cell type in the lateral habenula links hypothalamic homeostatic and midbrain motivation circuits with sex steroid signaling. Transl. Psychiatry 8, 50 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 153, 875, 237]]<|/det|> +36. Quina, L. A., Walker, A., Morton, G., Han, V. & Turner, E. E. GAD2 Expression Defines a Class of Excitatory Lateral Habenula Neurons in Mice that Project to the Raphe and Pontine Tegmentum. eNeuro 7, (2020). + +<|ref|>text<|/ref|><|det|>[[110, 248, 875, 302]]<|/det|> +37. Wallace, M. L. et al. Anatomical and single-cell transcriptional profiling of the murine habenular complex. eLife 9, (2020). + +<|ref|>text<|/ref|><|det|>[[110, 312, 880, 396]]<|/det|> +38. Mansour, A., Khachaturian, H., Lewis, M. E., Akil, H. & Watson, S. J. Autoradiographic differentiation of mu, delta, and kappa opioid receptors in the rat forebrain and midbrain. J. Neurosci. Off. J. Soc. Neurosci. 7, 2445–2464 (1987). + +<|ref|>text<|/ref|><|det|>[[110, 407, 839, 461]]<|/det|> +39. Kang, S. et al. Downregulation of M-channels in lateral habenula mediates hyperalgesia during alcohol withdrawal in rats. Sci. Rep. 9, 2714 (2019). + +<|ref|>text<|/ref|><|det|>[[110, 471, 861, 525]]<|/det|> +40. Li, J., Li, Y., Zhang, B., Shen, X. & Zhao, H. Why depression and pain often coexist and mutually reinforce: Role of the lateral habenula. Exp. Neurol. 284, 106–113 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 535, 880, 589]]<|/det|> +41. Li, Y. et al. Role of the Lateral Habanula in Pain-Associated Depression. Front. Behav. Neurosci. 11, 31 (2017). + +<|ref|>text<|/ref|><|det|>[[110, 599, 857, 653]]<|/det|> +42. Weiss, T. & Veh, R. W. Morphological and electrophysiological characteristics of neurons within identified subnuclei of the lateral habenula in rat brain slices. Neuroscience 172, 74–93 (2011). + +<|ref|>text<|/ref|><|det|>[[110, 663, 864, 717]]<|/det|> +43. Metzger, M. et al. Habanular connections with the dopaminergic and serotonergic system and their role in stress-related psychiatric disorders. Eur. J. Neurosci. (2019) doi:10.1111/ejn.14647. + +<|ref|>text<|/ref|><|det|>[[110, 727, 842, 781]]<|/det|> +44. Meye, F. J. et al. Shifted pallidal co-release of GABA and glutamate in habenula drives cocaine withdrawal and relapse. Nat. Neurosci. 19, 1019–1024 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 791, 861, 845]]<|/det|> +45. Shabel, S. J., Proulx, C. D., Piriz, J. & Malinow, R. Mood regulation. GABA/glutamate co-release controls habenula output and is modified by antidepressant treatment. Science 345, 1494–8 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 816, 140]]<|/det|> +Root, D. H. et al. Selective Brain Distribution and Distinctive Synaptic Architecture of Dual Glutamatergic- GABAergic Neurons. Cell Rep. 23, 3465–3479 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 153, 875, 237]]<|/det|> +Newman, H. M., Stevens, R. T. & Apkarian, A. V. Direct spinal projections to limbic and striatal areas: anterograde transport studies from the upper cervical spinal cord and the cervical enlargement in squirrel monkey and rat. J. Comp. Neurol. 365, 640–658 (1996). + +<|ref|>text<|/ref|><|det|>[[110, 249, 876, 301]]<|/det|> +Burstein, R., Cliffer, K. D. & Giesler, G. J. Direct somatosensory projections from the spinal cord to the hypothalamus and telencephalon. J. Neurosci. Off. J. Soc. Neurosci. 7, 4159–4164 (1987). + +<|ref|>text<|/ref|><|det|>[[110, 312, 880, 364]]<|/det|> +Hori, T., Oka, T., Hosoi, M. & Aou, S. Pain modulatory actions of cytokines and prostaglandin E2 in the brain. Ann. N. Y. Acad. Sci. 840, 269–281 (1998). + +<|ref|>text<|/ref|><|det|>[[110, 376, 876, 428]]<|/det|> +Ono, T. & Nakamura, K. Learning and integration of rewarding and aversive stimuli in the rat lateral hypothalamus. Brain Res. 346, 368–373 (1985). + +<|ref|>text<|/ref|><|det|>[[110, 440, 875, 491]]<|/det|> +Almli, C. R. & McMullen, N. T. Ontogeny of lateral preoptic unit activity in rats. Brain Res. Bull. 4, 773–781 (1979). + +<|ref|>text<|/ref|><|det|>[[110, 504, 880, 556]]<|/det|> +Vowles, K. E. et al. Rates of opioid misuse, abuse, and addiction in chronic pain: a systematic review and data synthesis. Pain 156, 569–576 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 568, 806, 620]]<|/det|> +Zadina, J. E. et al. Endomorphin analog analgesics with reduced abuse liability, respiratory depression, motor impairment, tolerance, and glial activation relative to morphine. + +<|ref|>text<|/ref|><|det|>[[110, 633, 460, 652]]<|/det|> +Neuropharmacology 105, 215–227 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 664, 816, 715]]<|/det|> +Ehrlich, A. T. et al. Expression map of 78 brain-expressed mouse orphan GPCRs provides a translational resource for neuropsychiatric research. Commun. Biol. 1, 102 (2018). + +<|ref|>text<|/ref|><|det|>[[110, 728, 870, 844]]<|/det|> +Wagner, F., Bernard, R., Derst, C., French, L. & Veh, R. W. Microarray analysis of transcripts with elevated expressions in the rat medial or lateral habenula suggest fast GABAergic excitation in the medial habenula and habenular involvement in the regulation of feeding and energy balance. Brain Struct. Funct. 221, 4663–4689 (2016). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 89, 825, 108]]<|/det|> +56. Broms, J. et al. Monosynaptic retrograde tracing of neurons expressing the G-protein coupled + +<|ref|>text<|/ref|><|det|>[[144, 120, 717, 140]]<|/det|> +receptor Gpr151 in the mouse brain. J. Comp. Neurol. 525, 3227- 3250 (2017). + +<|ref|>text<|/ref|><|det|>[[111, 152, 850, 204]]<|/det|> +57. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates, Compact. (Academic Press, 1997). + +<|ref|>text<|/ref|><|det|>[[111, 216, 867, 268]]<|/det|> +58. Chaplan, S. R., Bach, F. W., Pogrel, J. W., Chung, J. M. & Yaksh, T. L. Quantitative assessment of tactile allodynia in the rat paw. J. Neurosci. Methods 53, 55-63 (1994). + +<|ref|>text<|/ref|><|det|>[[111, 280, 843, 333]]<|/det|> +59. Yamaguchi, T., Wang, H.-L., Li, X., Ng, T. H. & Morales, M. Mesocorticolimbic glutamatergic pathway. J. Neurosci. Off. J. Soc. Neurosci. 31, 8476-8490 (2011). + +<|ref|>text<|/ref|><|det|>[[111, 344, 847, 397]]<|/det|> +60. Mouton, P. R., Gokhale, A. M., Ward, N. L. & West, M. J. Stereological length estimation using spherical probes. J. Microsc. 206, 54-64 (2002). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[59, 75, 424, 650]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[437, 45, 941, 650]]<|/det|> +Extended Data Figure 1. Further characterization of neuropathic pain and inflammatory pain models and the impact of intra- LHb MOR activation in rat. a, (Left) Violin plots of withdrawal latency to heat in sham \((\mathrm{n} = 9)\) , male \((\mathrm{n} = 8)\) and female \((\mathrm{n} = 9)\) rats with SNI after saline vs DAMGO \((10~\mu \mathrm{M})\) microinjections into the LHb. Sham males: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) ; SNI males: Paired t- test, \(\mathrm{df} = 7\) , \(\mathrm{t} = - 1.92\) , \(\mathrm{p} = 0.096\) ; SNI females: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 1.4\) , \(\mathrm{p} = 0.199\) . (Right) i.c.v. sham male: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 0.08\) , \(\mathrm{p} = 0.938\) . i.c.v. SNI male: Paired t- test, \(\mathrm{df} = 11\) , \(\mathrm{t} = 3.2\) , \(\mathrm{p} = 0.00846\) . To test if the effects of MOR activation in the LHb generalize to other forms of pain, we repeated the manipulations above with an inflammatory model of pain. Male rats received an intradermal injection of CFA or sterile saline into the plantar aspect of the hindpaw, b, Schematic diagram of inflammatory injury preparations and cannulation targeting the LHb. c, Mechanical withdrawal thresholds in sham and CFA- injured rats. In animals with CFA, we observed an increase in the average withdrawal threshold following intra- LHb DAMGO \((10~\mu \mathrm{M})\) compared to saline, indicating a reduction in mechanical allodynia: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = - 2.89\) , \(\mathrm{p} = 0.0202\) ; but not shams Wilcoxon signed rank test, \(\mathrm{V} = 1.5\) , \(\mathrm{p} = 1\) . d, Withdrawal latency to heat. Similar to results in rats with SNI, DAMGO in the LHb did not reverse the CFA- induced decrease in withdrawal latency to heat. Sham: Paired t- test, \(\mathrm{df} = 8\) , \(\mathrm{t} = 1.06\) , \(\mathrm{p} = 0.32\) . CFA: Wilcoxon signed rank test, \(\mathrm{V} = 12\) , \(\mathrm{p} = 0.25\) . e, Rats with CFA injury also did not develop a preference for the DAMGO- paired chamber in the place conditioning paradigm: Two- way mixed ANOVA, significant interaction between CFA/sham and baseline/test \(\mathrm{F}(1,14) = 0.474\) , \(\mathrm{p} = 0.502\) ; paired t- tests, sham adjusted \(\mathrm{p} = 0.286\) ; SNI adjusted \(\mathrm{p} = 0.761\) . The discrepancy between place conditioning results of our neuropathic and inflammatory pain animals may be due to the natural history of the injury caused in the CFA model: whereas SNI animals underwent permanent nerve ligation, animals injected with CFA presented with transient swelling and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were tested daily; each line represents a male rat. g, Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} \leq 0.01\) + +<|ref|>text<|/ref|><|det|>[[56, 650, 941, 797]]<|/det|> +and erythema of the hindpaw. Within a few days of CFA injection, this phenotype had resolved in our rats. f, Mechanical allodynia also reversed quickly: in a separate cohort of rats, all subjects displayed a sharp decline in withdrawal threshold from baseline on day 1 following CFA injection but the injury- induced allodynia returned to control levels by day 5 in many animals. Individual animals were tested daily; each line represents a male rat. g, Average across rats shows that our CFA- induced reduction in mechanical allodynia reversed within a few days. Since our place conditioning training exceeds this time window, resolution of the injury probably occurred before training was completed, thus preventing the formation of a robust association between pain relief and place conditioning apparatus context to produce a CPP. \(* \mathrm{p} \leq 0.05\) , \(** \mathrm{p} \leq 0.01\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 72, 423, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 45, 280, 62]]<|/det|> +
Extended Data Figure 2
+ +<|ref|>text<|/ref|><|det|>[[488, 61, 936, 243]]<|/det|> +Extended Data Figure 2. Neurons receiving functional synaptic inputs from the five brain regions showing connectivity were distributed throughout the LHb. Locations of neurons found connected using whole- cell recordings from optogenetic stimulation of terminals arising from each input tested. Locations are based on biocytin immunohistochemistry and low magnification images taken on the recording microscope where the recorded cell is centered within the field of view. Color indicates source of innervation. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 81, 320, 500]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 53, 234, 70]]<|/det|> +
Extended Figure 3
+ +<|ref|>image<|/ref|><|det|>[[333, 87, 602, 201]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[333, 228, 345, 240]]<|/det|> +
C
+ +<|ref|>image<|/ref|><|det|>[[333, 230, 600, 401]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[620, 47, 937, 396]]<|/det|> +Extended Data Figure 3. ACC minimally innervates the LHb. a, Diagram of the extent of unilateral AAV2- hSyn- hChR2(H134R)- mCherry injection sites (n = 9 male rats) throughout anteroposterior range of the ACC for anterograde tracing study. b, Example ipsilateral DAB- positive fibers (black) visualized under brightfield illumination. Fibers heavily innervate the MDL, which abuts the lateral edge of the LHb, while sparse to no fibers innervate the LHb. Contralateral innervation of LHb and MDL was negligible compared to the ipsilateral side, and therefore were omitted from our analysis. (Left) Scale bar = 250 \(\mu \mathrm{m}\) . (Right) Scale bar = 50 \(\mu \mathrm{m}\) . c, Average stereologically- quantified DAB- positive fiber lengths throughout anteroposterior range of the LHb, compared to the MDL innervation in the same coronal slice. Mann- Whitney test, two- tailed: \(\mathrm{U} = 0\) , \(\mathrm{p} = 0.0006\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[58, 80, 333, 280]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[59, 50, 280, 67]]<|/det|> +
Extended Data Figure 4
+ +<|ref|>text<|/ref|><|det|>[[343, 46, 931, 270]]<|/det|> +Extended Data Figure 4. Sources of glutamatergic inputs to the rat LHb vary in strength and delay in synaptic transmission. a, We observed some variations between inputs in the mean light evoked EPSCs ( \(\mathrm{V_m} = - 60 \mathrm{mV}\) ). In particular, excitatory inputs from the VTA were consistently small: Kruskal- Wallis \(\chi_2 = 25.5\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.00004\) , followed by Dunn Test for pairwise comparisons. b, Differences were also detected in the delay to the onset of light evoked EPSCs to LHb neurons from these different sources, with inputs from LH showing the fastest response. Although these data deviate from a normal distribution (Shapiro's test, \(\mathrm{p} = 0.0000001\) ), KDEs (violins) are consistent with continuous distributions, suggesting reliable polysynaptic events were rarely detected. c, We observed some variations between inputs in the mean light evoked IPSCs ( \(\mathrm{V_m} = - 40 \mathrm{mV}\) ). While there were no statistically significant differences in amplitudes detected, the mean inhibitory input from the VP was particularly small: Kruskal- Wallis \(\chi_2 = 2.3\) , \(\mathrm{df} = 4\) , \(\mathrm{p} = 0.7\) . \(^{**} \mathrm{p} < 0.01\) , \(^{***} \mathrm{p} < 0.005\) , \(^{****} \mathrm{p} < 0.0005\) + +<|ref|>image<|/ref|><|det|>[[58, 291, 333, 680]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[63, 66, 310, 360]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[63, 46, 283, 63]]<|/det|> +
Extended Data Figure 5
+ +<|ref|>text<|/ref|><|det|>[[325, 46, 941, 375]]<|/det|> +Extended Data Figure 5. Optogenetic experiments detect local glutamate, but not GABA, functional connections in rat LHb. AAV2-hSyn-hChR2(H134R)- mCherry was stereotaxically injected into the LHb at least 4 weeks prior to ex vivo whole cell recordings in the LHb to detect local synaptic connections. Neurons were recorded blind to ChR2 expression, therefore in some cases the patched neuron expressed ChR2. Therefore, in addition to connectivity criteria used for other afferent inputs, in these experiments only light evoked inward currents that were blocked by \(10 \mu \mathrm{M}\) DNQX were considered glutamatergic connections, and direct ChR2 induced inward currents were subtracted out for the quantification illustrated here. Each cell was probed for both glutamate and GABA inputs in voltage clamp by holding neurons at \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) and \(- 40 \mathrm{mV}\) , light pulse durations 1, 5, and 10 ms durations. a, Example recording at \(\mathrm{V}_{\mathrm{m}} = - 60 \mathrm{mV}\) showing a light evoked response that was blocked by DNQX. b, Example recording at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) with minimal outward current response within 7 ms of light pulse. c, Summary of all LHb recordings tested in this experiment. Filled circles represent cells where responses could be classified as local ChR2 induced synaptic transmission. When no clear response was detected, the measure indicated is the difference between the mean \(I_{\mathrm{holding}}\) of the baseline \(100 \mathrm{ms}\) period just prior to the light pulse and the mean \(I_{\mathrm{holding}}\) 2 ms period starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +<|ref|>text<|/ref|><|det|>[[58, 376, 927, 412]]<|/det|> +starting 2 ms after initiation of the light pulse, consistent with monosynaptic delay timing. None of the recordings made at \(\mathrm{V}_{\mathrm{m}} = - 40 \mathrm{mV}\) met the criteria for a response to light stimulation with an outward current. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 65, 350, 216]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[61, 46, 282, 62]]<|/det|> +
Extended Data Figure 6
+ +<|ref|>image<|/ref|><|det|>[[320, 63, 601, 230]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[616, 64, 941, 245]]<|/det|> +Extended Data Figure 6. Isolated monosynaptic glutamatergic inputs from LPO to LHb neurons are inhibited by MOR activation. a, Example light evoked EPSC responses in an LHb neuron from a rat with ChR2 expression in LPO neurons. This response persisted in monosynaptic isolation by \(500~\mathrm{nM}\) TTX and \(100~\mu \mathrm{M}4\mathrm{AP}\) (green), and this isolated response was inhibited by \(500~\mathrm{nM}\) DAMGO (magenta). b, Summary of + +<|ref|>text<|/ref|><|det|>[[58, 248, 870, 284]]<|/det|> +DAMGO effects on isolated monosynaptic EPSC inputs to LHb neurons expressed as \(\%\) of baseline monosynaptic response (left) and as raw EPSC magnitudes (right). Paired t- test, \(\mathrm{df} = 4\) , \(\mathrm{t} = - 5.1\) , \(\mathrm{p} = 0.007\) . \(^{**} \mathrm{p} < 0.01\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 72, 440, 204]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[60, 45, 280, 62]]<|/det|> +
Extended Data Figure 7
+ +<|ref|>text<|/ref|><|det|>[[478, 64, 937, 208]]<|/det|> +Extended Data Figure 7. Mice with SNI show hypersensitivity to heat and increased activity in LHb- projecting LPO neurons during paw withdrawal from thermal stimulation. a, Mice with SNI show a reduced latency to withdraw their paw following thermal stimulation in the Hargreaves task. Unpaired t- test, \(\mathrm{t}(12) = 3.007\) , \(\mathrm{p} = 0.011\) . b, VGluT2- expressing LPO neurons that project to the LHb expressed GCaMP6m and showed a greater calcium response during paw withdrawal to Hargreaves + +<|ref|>text<|/ref|><|det|>[[57, 209, 937, 245]]<|/det|> +thermal stimulation (area under the curve, deviation from baseline fluorescence) in SNI animals \(\mathrm{(n = 5)}\) compared to sham controls \(\mathrm{(n = 9)}\) : Two- way ANOVA, \(\mathrm{F}(1,12) = 5.4\) , \(\mathrm{p} = 0.038\) ; Holm- Sidak post- hoc test, \(\mathrm{p} = 0.0074\) . \*\* \(\mathrm{p} < 0.01\) + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 75, 571, 440]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[585, 64, 940, 117]]<|/det|> +
Extended Data Figure 8. Locations and spread of bilateral ChR2 injections to the LPO in rats used for in vivo optogenetic experiments.
+ +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[57, 46, 650, 63]]<|/det|> +Supplementary Table 1: Assumption testing on behavioral data + +<|ref|>table<|/ref|><|det|>[[58, 78, 936, 940]]<|/det|> + +
ExperimentFigure# animals;
# Outliers;
# extreme
outliers		Shapiro-Wilk
Test of
Normality
(across all
groups)
Statistic; p val		Test for
Homogeneity of
Variances
Statistic; p val
(Test used)		Parametric testNon-parametric test:
Repeated measures
Wilcoxon signed rank
exact test
V; p val
von Frey: Male; Sham;
10 μM DAMGO in
LHb		1b9; 0; 00.773; 0.0006380.0760; 0.786
(Levene's Test)		n/a1.5; 1
von Frey: Male; SNI;
10 μM DAMGO in
LHb		1b8; 0; 00.866; 0.02401.49; 0.242
(Levene's Test)		n/a2; 0.05024
von Frey: Female;
SNI; 10 μM DAMGO
in LHb		1b9; 2; 10.805; 0.001811.82; 0.196
(Levene's Test)		n/a7; 0.07422
von Frey: Male; Sham;
10 μM DAMGO in
i.c.v.		1b9; 0; 00.768; 0.000550.0897; 0.768
(Levene's Test)		n/a7; 0.5294
von Frey: Male; SNI;
10 μM DAMGO in
i.c.v.		1b12; 2; 20.756;
0.0000618		0.168; 0.686
(Levene's Test)		n/a31; 0.08006
Place Conditioning:
Male Sham/SNI; 10
μM DAMGO in LHb		1c15; 4; 0Sham x
Baseline: 0.95;
0.726
SNI x Baseline:
0.94; 0.686
Sham x Test:
0.936; 0.542
SNI x Test:
0.840; 0.129		4.68; 0.0305
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,13)=15.932;p\) \(=0.002\)
effect of group on baseline: \(\mathrm {F}=0.293;\) adjusted \(\mathrm {p}=1\)
effect of group on test day: \(\mathrm {F}=11.3;\) adjusted \(\mathrm {p}=0.01\)
Paired t-tests, adjusted
Sham: adjusted \(\mathrm {p}=1\)
SNI: adjusted \(\mathrm {p}=0.012\)		n/a
Place Conditioning:
Sham vs. Male SNI;
10 μM DAMGO in
i.c.v.		1c21; 2; 0Sham x
Baseline: 0.901;
0.257
SNI x Baseline:
0.967; 0.872
Sham x Test:
0.971; 0.905
SNI x Test:
0.985; 0.997		23.9;
0.000000997
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,19)=2.239;p=\) 0.151
Paired t-tests for non-significant two-way interaction:
Sham: adjusted \(\mathrm {p}=0.286\)
SNI: adjusted \(\mathrm {p}=0.433\)		n/a
von Frey: Female;
Sham; 100 μM
DAMGO in LHb		1d6; 0; 00.838; 0.0260.493; 0.499
(Levene's Test)		n/a5; 1
von Frey: Female;
SNI; 100 μM
DAMGO in LHb		1d9; 1; 10.839; 0.009281.13; 0.306
(Levene's Test)		n/a6; 0.4017
Place Conditioning:
Female Sham/SNI;
100 μM DAMGO in
LHb		1e15; 3; 0Sham x
Baseline: 0.904
0.66
SNI x Baseline:
0.919; 0.384
Sham x Test:
0.87; 0.226
SNI x Test:
0.900; 0.250		8.59; 0.00339
(Box M-test)		Two-way mixed design ANOVA
two-way interaction \(\mathrm {F}(1,13)=6.234;p=\) 0.027
effect of group on baseline: \(\mathrm {F}=0.006;\) adjusted \(\mathrm {p}=1\)
effect of group on test day: \(\mathrm {F}=5.56;\) adjusted \(\mathrm {p}=0.07\)
Paired t-tests, adjusted
Sham: adjusted \(\mathrm {p}=0.476\)
SNI: adjusted \(\mathrm {p}=0.0349\)		n/a
Fiber photometry;
Sham vs. SNI mice		5b14; not
determined		not determinednot determinedTwo-way mixed design ANOVA
\(\mathrm {F}(1,12)=5.439;\) \(p=0.038;\)
Holm-Sidak post-hoc test, \(\mathrm {p}=0.0074\)		n/a
Place Conditioning;
Male; mCherry vs.
ChR2; 10 μM
DAMGO in LHb		5e16; 0; 0mCherry x
Baseline: 0.941;
0.618
ChR2 x
Baseline: 0.913;
0.374		18.3; 0.0000186
(Box M-test)		Two-way mixed design ANOVA
two-way interaction, \(\mathrm {F}(1,14)=9.982;p\) \(=0.007\)
effect of group on baseline: \(\mathrm {F}=0.01;\) adjusted \(\mathrm {p}=1\)		n/a
+ +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[56, 42, 936, 832]]<|/det|> +
mCherry x Test: 0.914; 0.385 ChR2 x Test: 0.847; 0.0891effect of group on test day: F = 12.5; adjusted p = 0.006
Paired t-tests, adjusted mCherry: adjusted p = 0.427 ChR2: adjusted p = 0.0027
Hargreaves: Male; Sham; 10 μM DAMGO in LHbED 1a9; 2; 00.966; 0.726Bartlett's K-squared = 0.035658, df = 1, p-value = 0.8502 (Bartlett Test)Paired t-test
Df = 8; t = 1.06; p = 0.32		n/a
Hargreaves; Male; SNI; 10 μM DAMGO in LHbED 1a8; 0; 00.941; 0.365Bartlett's K-squared = 3.704 df = 1, p-value = 0.096 (Bartlett Test)Paired t-test
Df = 7; t = -1.92; p = 0.096		n/a
Hargreaves; Female; SNI; 10 μM DAMGO in LHbED 1a9; 1; 10.944; 0.333Bartlett's K-squared = 1.35 df = 1, p-value = 0.245 (Bartlett Test)Paired t-test
Df = 8; t = -1.4; p = 0.199		n/a
Hargreaves: Male; Sham; 10 μM DAMGO in i.c.v.ED 1a9; 3; 00.927; 0.1740.840; 0.373 (Levene's Test)Paired t-test
Df = 8; t = -0.08; p = 0.938		n/a
Hargreaves; Male; SNI; 10 μM DAMGO in i.c.v.ED 1a8; 0; 00.941; 0.3650.013; 0.911 (Levene's Test)Paired t-test
Df = 11; t = 3.2; p = 0.00846		n/a
von Frey; Male; Sham; 10 μM DAMGO in LHbED 1c9; 0; 00.773; 0.0006380.0760; 0.786 (Levene's Test)n/a1.5; 1
von Frey; Male; CFA; 10 μM DAMGO in LHbED 1c9; 0; 00.903; 0.06393.80 0.0691 (Levene's Test)Paired t-test
Df = 8; t = -2.89; p = 0.0202		n/a
Hargreaves; Male; Sham; 10 μM DAMGO in LHbED 1d9; 2; 00.966; 0.726Bartlett's K-squared = 0.035658, df = 1, p-value = 0.8502 (Bartlett Test)Paired t-test
Df = 8; t = 1.06; p = 0.32		n/a
Hargreaves; Male; CFA; 10 μM DAMGO in LHbED 1d9; 2; 20.858; 0.01130.0461 0.833 (Levene's Test)n/a12; 0.25
Place Conditioning; Male; Sham vs. CFA; 10 μM DAMGO in LHbED 1e16; 3; 0Sham x Baseline: 0.901; 0.257 CFA x Baseline: 0.94; 0.635 Sham x Test: 0.971; 0.905 CFA x Test: 0.828; 0.077413.4; 0.000253 (Box M-test)Two-way mixed design ANOVA two-way interaction F(1,14) = 0.474; p = 0.502
Paired t-tests for non-significant two-way interaction:
Sham: adjusted p = 0.286
CFA: adjusted p = 0.761		n/a
ACC fiber innervation; Ipsi LHb vs. Ipsi MDLED 3c9; 0; 0not determinednot determinedn/aMann-Whitney test, two-tailed, U=0; p = 0.0006
Hargreaves; Sham vs. SNI miceED 7a14; 3; 20.896; 0.09940.0012; 0.973 (Levene's Test)Unpaired t-test
Df = 12, t = 3.007, p = 0.0109		n/a
AUC summary: Hargreaves/Fiber photometry; Sham vs. SNI miceED 7b14; 0; 0Sham x Baseline: 0.979; 0.257 Sham x paw withdrawal: 0.932; 0.496 SNI x Baseline: 0.813; 0.103 SNI x paw withdrawal: 0.921; 0.53853.4; 2.71 e-13 (Box M-test)Two-way mixed design ANOVA F(1,12) = 5.439; p = 0.038; Holm-Sidak post-hoc test, p=0.0074n/a
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 43, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[53, 92, 920, 423]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 448, 115, 468]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[38, 490, 955, 944]]<|/det|> +MOR activation in the LHb relieves allodynia and generates negative reinforcement in rats with neuropathic pain. a, Timeline of behavioral experiments. Lower inset, example injection site centered in LHb. Scale bar \(= 500\mu \mathrm{m}\) . b, (Left) Violin plots of \(50\%\) mechanical withdrawal thresholds following saline or \(10\mu \mathrm{M}\) DAMGO microinjections into the LHb in sham males \((n = 9)\) , SNI males \((n = 8)\) , and SNI females \((n = 9)\) . Sham males: Wilcoxon signed rank test, \(V = 1.5\) , \(p = 1\) . SNI males: Wilcoxon signed rank test, \(V = 2\) , \(p = 0.050\) . Females trended towards an increase in threshold following DAMGO infusion: Wilcoxon signed rank test, \(V = 7\) , \(p = 0.074\) . (Right) Mechanical withdrawal thresholds in male rats with i.c.v. cannulae, with SNI \((n = 12)\) or sham \((n = 9)\) injuries. Sham: Wilcoxon signed rank test, \(V = 7\) , \(p = 0.53\) . SNI: Wilcoxon signed rank test, \(V = 31\) , \(p = 0.080\) . c, Difference scores before and after conditioning with intracranial microinjections. (Left) Males with SNI developed a preference for the chamber paired with intra- LHb DAMGO. Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(F(1,13) = 15.93\) , \(p = 0.002\) ; post hoc effect group on test day adjusted \(p = 0.01\) ; paired t-tests, sham adjusted \(p = 1\) ; SNI adjusted \(p = 0.016\) . Female SNI animals trended towards a preference for the DAMGO-paired chamber: Paired t-test \(p = 0.054\) . (Right) i.c.v. DAMGO did not produce a preference in sham or SNI male rats: Two-way mixed ANOVA, \(F(1,19) = 2.239\) , \(p = 0.15\) . d, Mechanical withdrawal thresholds in sham \((n = 6)\) and SNI \((n = 9)\) female rats following \(100\mu \mathrm{M}\) DAMGO infusions into the LHb. Sham: Wilcoxon signed rank test, \(V = 5\) , \(p = 1\) . SNI: Wilcoxon signed rank test, \(V = 6\) , \(p = 0.40\) . e, Place conditioning of sham and SNI females to intra- LHb \(100\mu \mathrm{M}\) DAMGO vs saline. Females with SNI developed a significant CPP: Two-way mixed ANOVA, significant interaction between SNI/sham and baseline/test \(F(1,13) = 6.234\) , \(p = 0.027\) ; post hoc effect group on test day adjusted \(p = 0.07\) ; paired t-tests, sham adjusted \(p = 0.48\) ; SNI + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 936, 88]]<|/det|> +adjusted \(p = 0.035\) . Means represented by horizontal lines; SEM and quartiles indicated by vertical lines with and without caps, respectively. \(*p \leq 0.05\) , \(**p < 0.01\) + +<|ref|>image<|/ref|><|det|>[[60, 92, 420, 551]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 575, 117, 594]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[38, 612, 951, 958]]<|/det|> +Ongoing pain does not alter MOR agonist synaptic effects or probability of glutamate release in the LHb. Responses to the MOR agonist DAMGO were recorded in voltage clamp, \(V_{m} = - 60 \text{mV}\) , a, In animals with SNI, a subset of neurons responded to bath application of the MOR agonist DAMGO (500 nM) with a time locked outward current that was reversed with the MOR selective antagonist CTAP (500 nM). b, Postsynaptic holding current changes in response to bath application of DAMGO were not different between neurons from naïve and SNI rats: Unpaired t- test df = 32, \(t = 0.892\) , \(p = 0.379\) . c, In rats with SNI, DAMGO (500 nM) induced a time locked inhibition of electrically evoked glutamatergic EPSCs, and CTAP (500 nM) reversed the effect. d, This DAMGO inhibition of electrically evoked EPSC amplitude was not different between LHb neurons from naïve and SNI rats: Unpaired t- test df = 21, \(t = 0.137\) , \(p = 0.89\) . Baseline probability of release at glutamatergic synapses was similar between LHb neurons in naïve and SNI rats as measured by (e) paired pulse ratio (P2/P1) of electrically stimulated EPSCs (50 ms inter stimulus interval): Unpaired t- test, df = 24, \(t = 0.102\) , \(p = 0.92\) ; (f) baseline spontaneous EPSC frequency: Unpaired t- test unequal variances, df = 37, \(t = -0.17\) , \(p = 0.87\) ; and (g) spontaneous EPSC amplitude: Unpaired t- test unequal variances, df = 38, \(t = -0.09\) , \(p = 0.93\) . Data from naïve rats previously published in 16. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[44, 45, 333, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 570, 117, 589]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[39, 607, 951, 955]]<|/det|> +Glutamatergic inputs to the LHb from the LPO are inhibited by MOR activation. a, Schematic diagram of AAV2- hSynhChR2(H134R)- mCherry injection targets which project to the LHb. b, Proportions of LHb neurons receiving functional synaptic connections from six input regions measured with optogenetic stimulation of terminals expressing ChR2. c, Among connected neurons, venn diagrams depict the numbers of LHb cells receiving optically stimulated glutamatergic and/or GABAergic synaptic input. d, (Left) Experiment design: DAMGO (500 nM) was applied during optical synaptic stimulation experiments to probe for functional MOR presynaptic control of glutamate release from specific inputs. (Right) Change from baseline amplitude of optically- evoked EPSCs during DAMGO application. EPSCs from LPO terminal stimulation were most strongly inhibited by MOR activation: One- way ANOVA, df = 4, F = 4.11, p = 0.0057 followed by Tukey HSD pairwise comparisons. e, (Left) Representative traces of optically- evoked EPSCs in an LHb neuron from a rat with ChR2 expressed in inputs from the LPO. DAMGO decreased the amplitude of light- evoked EPSC, which was partially reversed by the selective MOR antagonist CTAP (500 nM). Rise time was < 2 ms, indicative of a monosynaptic response. (Right) Summary time course of the DAMGO inhibition across all optically- evoked LPOLHb EPSCs (n = 12). f, LPO inputs to the LHb in female rats (left) had a similar overall connectivity rate but (middle) more + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 955, 88]]<|/det|> +GABAergic connections than glutamate connections. (Right) DAMGO inhibited light-evoked glutamatergic synaptic inputs from the LPO to the LHb in female rats as well. + +<|ref|>image<|/ref|><|det|>[[57, 92, 714, 440]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 470, 117, 488]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[40, 510, 953, 714]]<|/det|> +LPO inputs to the LHb highly express MOR mRNA. a, Schematic of unilateral iontophoretic deposition of the retrograde tracer FG into the LHb. b, FG injection sites in LHb ( \(n = 3\) rats). c, (Left and middle) Representative brightfield images of FG- positive/MOR- positive somas (yellow arrowhead) and FG- positive/MOR- negative somas (blue arrowhead) in the LPO. Scale bar \(= 20 \mu \mathrm{m}\) . (Right) Darkfield image of in situ hybridization localization for MOR mRNA (OPRM1). d, Number of MOR- positive/FG- positive (yellow) and MOR- negative/FG- positive (blue) cells in various basal forebrain regions. e, Donut plot depicting proportions of total MOR- positive inputs to the LHb from various basal forebrain regions. f, Locations of FG- positive/MOR- positive (red) and FG- positive/MOR- negative (blue) somata throughout basal forebrain. Co- labeled cells are separated from the singly labelled population for display purposes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[57, 46, 380, 415]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 444, 117, 462]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[38, 485, 960, 895]]<|/det|> +LHb MOR activation relieves aversiveness of glutamatergic LPO input to the LHb. a, (Top) Schematic diagram of HSV- hEF1aLS1L- GCaMP6m injection into the LHb of VGluT2::Cre mice, to induce expression of GCaMP6m in glutamatergic LPO neurons innervating the LHb, and optic fiber implant for fiber photometry imaging in the LHb- projecting LPO neurons. (Bottom) Experimental timeline. b, Mice with SNI (n = 5) exhibited significantly larger changes in GCaMP6m fluorescence time- locked to paw withdrawal in response to thermal stimulation in the Hargreaves apparatus compared to sham controls (n = 9): Two- way ANOVA, \(\mathrm{F}(1,12) = 5.439\) , \(p = 0.038\) ; Holm- Sidak post- hoc test, \(p = 0.0074\) . c, (Left) Schematic diagram of dual virus injections to express ChR2- mCherry or mCherry selectively in LPO neurons that project to the LHb, optic fiber implantation targeting the LPO cell bodies, and cannula implantation targeting the LHb for experiments in male rats. (Right) Rats received bilateral blue light stimulation of the LPO during all conditioning sessions and in both chambers, sides paired with intraLHb saline or DAMGO. d, Representative images of Cre- dependent ChR2- mCherry fluorescence in LPO cell bodies. (Left) Scale bar = \(250 \mu \mathrm{m}\) . (Right) Lesion from optic fiber ("o.f.") implant amidst ChR2- mCherry- expressing cell bodies. Scale bar = \(50 \mu \mathrm{m}\) . e, Only animals with active ChR2 (n = 8), but not mCherry controls (n = 8), developed a preference for the intra- LHb DAMGO paired side of the place conditioning apparatus. Two- way mixed ANOVA, significant interaction between ChR2- mCherry/mCherry and baseline/test \(\mathrm{F}(1,14) = 9.982\) , \(p = 0.007\) ; post hoc effect group on test day adjusted \(p = 0.006\) ; paired t- tests, mCherry adjusted \(p = 0.427\) ; ChR2- mCherry adjusted \(p = 0.0027\) . \(^{**}p < 0.01\) , \(^{***}p < 0.005\) + +<--- Page Split ---> diff --git a/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/images_list.json b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..edc5d2c7b13738f97c97e42318d6a8a4fef58757 --- /dev/null +++ b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/images_list.json @@ -0,0 +1,167 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig.1: O-GlcNAc level correlates with endometrial cancer grading", + "footnote": [], + "bbox": [ + [ + 120, + 150, + 888, + 846 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "Fig.S1: Construction of a gene signature for O-GlcNAc level", + "footnote": [], + "bbox": [ + [ + 123, + 128, + 888, + 776 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig.2: Validation with TCGA endometrial cancer dataset", + "footnote": [], + "bbox": [ + [ + 125, + 125, + 875, + 856 + ] + ], + "page_idx": 37 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig.3: Increase of O-GlcNAc level promotes proliferation and stemness", + "footnote": [], + "bbox": [ + [ + 160, + 120, + 842, + 796 + ] + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "Fig.S2: Generation of EE-O and EC-O, and TMG treatment on EE-O", + "footnote": [], + "bbox": [ + [ + 130, + 115, + 880, + 860 + ] + ], + "page_idx": 41 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig.4: Decrease of O-GlcNAc level induces differentiation and cell death", + "footnote": [], + "bbox": [ + [ + 123, + 115, + 875, + 855 + ] + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig.5: Screen for TSGs that regulate O-GlcNAc homeostasis", + "footnote": [], + "bbox": [ + [ + 152, + 110, + 878, + 840 + ] + ], + "page_idx": 45 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_2.jpg", + "caption": "Fig.S3: Survival analysis of potential O-GlcNAc regulators in UCEC", + "footnote": [], + "bbox": [ + [ + 128, + 110, + 833, + 870 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig.6: FBXO31 interacts with and ubiquitinates OGT", + "footnote": [], + "bbox": [ + [ + 123, + 110, + 888, + 878 + ] + ], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Fig.7: Loss of FBXO31 increases O-GlcNAc level in clinical samples", + "footnote": [], + "bbox": [ + [ + 123, + 120, + 889, + 875 + ] + ], + "page_idx": 51 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Fig.8: Working model", + "footnote": [], + "bbox": [ + [ + 122, + 125, + 884, + 400 + ] + ], + "page_idx": 53 + } +] \ No newline at end of file diff --git a/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9.mmd b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9.mmd new file mode 100644 index 0000000000000000000000000000000000000000..dacfde4b4cd175f0839c7974e1e53c4e49508cd1 --- /dev/null +++ b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9.mmd @@ -0,0 +1,592 @@ + +# Impairment of FBXO31-mediated Ubiquitination of OGT Upregulates O-GlcNAcylation to Advance Endometrial Malignancy + +Kai Yuan + +yuankai@csu.edu.cn + +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. https://orcid.org/0000- 0001- 7002- 5703 + +Na Zhang + +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. + +Yang Meng + +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University Song Mao + +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University + +Huiling Ni + +Center for Medical Genetics, School of Life Sciences, Central South University Canhua Huang + +Xiangya Hospital, Central South University + +Licong Shen + +Department of Gynecology, Xiangya Hospital, Central South University + +Kun Fu + +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University + +Lu Lv + +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University + +Chunhong Yu + +Institute of Molecular Precision Medicine, Xiangya Hospital, Central South University + +Fang Chen + +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University + +Yu Zhang + +<--- Page Split ---> + +## Article + +## Keywords: + +Posted Date: April 12th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4019799/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on February 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 56633- z. + +<--- Page Split ---> + +Impairment of FBXO31- mediated Ubiquitination of OGT Upregulates O- GlcNAcylation to Advance Endometrial Malignancy Na Zhang1, Yang Meng2,3, Song Mao1, Huiling Ni1,2, Canhua Huang1, Licong Shen1, Kun Fu1, Lu Lv1, Chunhong Yu1, Fang Chen1, Yu Zhang1, Kai Yuan1,2,4,5,6,# 1. Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. 2. Center for Medical Genetics, School of Life Sciences, Central South University, Changsha, Hunan, China. 3. School of Pharmaceutical Sciences, Tsinghua University, Beijing, China. 4. Furong Laboratory, Hunan, China. 5. National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, Hunan, China. 6. The Biobank of Xiangya Hospital, Central South University, Changsha, Hunan, China. # Correspondence: yuankai@csu.edu.cn (K.Y.) + +## Abstract + +Aberrant O- GlcNAc cycling of the cancer proteome is a manifestation of its metabolic plasticity. As one of the most common cancer of the female genital tract associated with metabolic syndrome, endometrial cancer (EC) tissues often bear altered O- GlcNAcylation patterns. However, integration of O- GlcNAc status with existing histomorphologic and molecular subtypes of EC in large cohorts and identification of molecular modules controlling the O- GlcNAc homeostasis remain to be accomplished. Here we establish a positive correlation of O- GlcNAcylation with histologic grade of EC in a Chinese cohort containing 219 tumors and consolidate it in The Cancer Genome Atlas (TCGA) EC dataset. Higher O- GlcNAc level is associated with less pathological differentiation and poorer prognosis. Functionally, increasing O- GlcNAcylation promotes proliferation and stem- like + +<--- Page Split ---> + +cell properties in normal endometrial epithelial organoids (EE- Os), whereas decreasing O- GlcNAcylation limits the growth of endometrial cancer organoids (EC- Os). Using genome- wide CRISPR screen, we further identify that the F- box only protein 31 (FBXO31), whose loss of heterozygosity is frequently observed in cancer, regulates O- GlcNAc homeostasis. FBXO31 acts as a substrate receptor of the SCF ubiquitin ligase complex to ubiquitinate the O- GlcNAc transferase OGT. Loss of FBXO31 results in accumulation of OGT and upregulation of O- GlcNAcylation in EC. Our study highlights the O- GlcNAcylation as a useful stratification marker and potential therapeutic target for the advanced, poorly differentiated EC cases. + +## Introduction + +Post- translational modifications (PTMs) endow the proteome with functional plasticity to cope with intrinsic and extrinsic perturbations under various developmental and disease conditions. Protein O- GlcNAcylation, catalyzed by a pair of evolutionarily conserved enzymes O- GlcNAc transferase (OGT) and O- GlcNAcase (OGA), is a PTM involving the covalent addition of single O- linked N- acetylglucosamine (O- GlcNAc) modifications to serine and threonine residues of intracellular proteins1. The GlcNAc moieties are supplied by a metabolite uridine diphosphate N- acetylglucosamine (UDP- GlcNAc), whose synthesis via the Hexosamine biosynthesis pathway (HBP) requires fructose- 6- P, acetyl- CoA, glutamine, and UTP, substrates from all major cellular metabolic pathways. As a result, O- GlcNAcylation is sensitive to nutrient availability and intrinsic metabolic reprogramming. Meanwhile, O- GlcNAc cycling is highly responsive to a wide range of extrinsic stimuli, including osmotic, oxidative, hyperthermic, and genotoxic stresses1-3, making it an important cellular stress sensing mechanism. O- GlcNAcylation is required for the maintenance of pluripotency of embryonic stem cells (ESCs), and its level declines as ESCs differentiate, alongside the cellular metabolic switch from glycolysis to oxidative phosphorylation4. Cancer cells often hijack embryonic programs to support their uncontrolled proliferation and cell fate transition, adopting a metabolic lifestyle relying on aerobic glycolysis (Warburg effect). Elevated O- GlcNAcylation has been observed in many + +<--- Page Split ---> + +cancer cell lines \(^{5 - 8}\) , probably as a result of its increased nutrient consumption, or imbalanced enzymatic activity of OGT and OGA due to somatic mutations or altered protein stability \(^{9 - 13}\) . To date, systemic assessment of \(O\) - GlcNAc status in major cancer cohorts and functional dissection of \(O\) - GlcNAc homeostasis in patient- derived organoids haven't been conducted, hindering the utilization of \(O\) - GlcNAcylation level to guide patient stratification and subsequent development of personalized interventions. + +Endometrial cancer (EC), the incidence of which has increased over \(50\%\) during the past two decades, is the most common cancer within the female reproductive system in developed countries \(^{14}\) . In China, as of 2022, there were approximately 77,700 newly diagnosed EC cases and 13,500 estimated EC cancer deaths \(^{15}\) . EC comprises a panel of tumors that are clinically and biologically heterogeneous. It can be grouped into type I or type II tumors according to the clinical and endocrine features \(^{16}\) , or classified as endometrioid carcinoma, serous carcinoma, carcinosarcoma, or clear- cell carcinoma based on its histopathological characteristics \(^{17}\) . The Cancer Genome Atlas Research Network (TCGA) study of EC has established a more precise genomic classification including four molecular subtypes: \(POLE\) - mutated, microsatellite- instable (MSI), copy- number low, and copy- number high tumors \(^{18}\) . More recently, integration of proteomic analysis to the genomic classification has accelerated the identification of clinically actionable molecular targets in EC \(^{19,20}\) . However, PTMs, which add tremendous functional complexity to the proteome, remain to be comprehensively characterized in EC tumor samples and complemented into the current classification system. + +\(O\) - GlcNAcylation as an important PTM responsive to cellular metabolism and stress has been linked to molecular etiology of EC. Both \(OGT\) and \(OGA\) manifested highest alterations, mainly gene mutation and amplification, in EC among major female cancer types \(^{21}\) . The mRNA levels of \(OGT\) and \(OGA\) were increased in EC tumor samples of higher histologic grade \(^{22}\) . Despite that \(O\) - GlcNAc homeostasis is maintained by delicate and complex feedback loops, elevated \(O\) - GlcNAcylation level in EC tissues was observed using a small + +<--- Page Split ---> + +tissue microarray \(^{23}\) . O- GlcNAcylation was reported to support proliferation and migration and promote epithelial- mesenchymal transition in cultured EC cell lines by regulating Wnt/ \(\beta\) - catenin and Hippo- YAP signaling pathways \(^{21,23 - 25}\) . These observations suggest that altered O- GlcNAcylation may contribute to EC progression, and it is worthy of thorough interrogation in large EC cohorts to determine whether O- GlcNAcylation can be utilized both as a new stratification factor and potential drugable target. + +In this study, utilizing a Chinese EC cohort containing 219 tumors and the TCGA EC dataset, we uncovered that O- GlcNAcylation level correlates with histologic grade, International Federation of Gynecology and Obstetrics (FIGO) stage, and patients' prognosis. Moreover, we experimentally demonstrated that upregulation of O- GlcNAcylation promotes proliferation and stem- like cell properties in normal endometrial epithelial organoids (EE- Os), whereas downregulation of O- GlcNAcylation impedes the proliferation of endometrial cancer organoids (EC- Os). Furthermore, we identified FBXO31 as a key regulator of O- GlcNAcylation homeostasis, by controlling the ubiquitin- dependent protein degradation of OGT. Our findings highlight that O- GlcNAcylation is a useful factor complementary to the current classification system to better stratify EC patients, and targeting O- GlcNAcylation signaling is a promising differentiation therapeutic strategy worthy of clinical exploitation for high grade EC patients. + +## Results + +Elevated O- GlcNAc level is positively correlated with histologic grade and poor prognosis in endometrial cancer + +To get a glimpse of global O- GlcNAcylation level in EC tissues, we first obtained an EC tissue array from Xinchao Biotech (Shanghai, China) containing 23 normal and 31 tumorous endometrial specimens, and performed IHC analyses to examine O- GlcNAcylation as well as OGT and OGA levels (fig. 1a). The expression of OGT and the amount of O- GlcNAcylation were significantly higher in EC tissues relative to normal tissues (fig. 1b- 1c), consistent with a previous report \(^{23}\) . The expression of OGA however showed no + +<--- Page Split ---> + +significant difference between normal and tumorous endometrial tissues (fig. sla- slb). + +To elaborate the relationship between the O- GlcNAcylation level and clinical characteristics of EC, we expanded the analyses to an EC cohort containing 219 tumor patients who received surgery in the Department of Gynecology, Xiangya Hospital, Central South University (fig. 1a). The paraffin- embedded EC tissue sections were subjected to IHC staining, and the O- GlcNAcylation level revealed by RL2 antibody staining for each specimen was semi- quantified to categorize the patients into O- GlcNAc high and O- GlcNAc low groups. This O- GlcNAc status exhibited significant association with histologic grade, FIGO stage, and distant metastasis of EC (table 1). Accordingly, the O- GlcNAcylation level manifested a significant increase in EC tissues from patients' groups with more advanced histologic grade (fig. 1f- 1g). Further statistical analysis established a positive correlation between the O- GlcNAcylation level and tumor histologic grade (Goodman- Kruskal gamma statistic \(\mathrm{P} \leq 0.0001\) ; 2- sided gamma- knife gamma \(= 0.473\) ), as well as distant metastasis (Goodman- Kruskal gamma statistic \(\mathrm{P} = 0.003\) ; 2- sided gamma- knife gamma \(= 1\) ). Kaplan- Meier analysis indicated that patients in the O- GlcNAc high group exhibited significantly shorter progression- free survival (PFS) and overall survival (OS) than that in the O- GlcNAc low group (fig. 1h- 1i). Univariate analysis revealed that O- GlcNAcylation level, alongside age, FIGO stage, and myometrial invasion, was significantly associated with PFS. Subsequent multivariate Cox regression analysis using all the statistically significant variables \((\mathrm{P} < 0.05)\) pinpointed O- GlcNAcylation level and age as independent predictors of clinical outcome of EC patients (table 2). + +Calculated virtual O- GlcNAc index is correlated with tumor histologic grade and survival in TCGA endometrial cancer dataset + +We wanted to validate the correlations observed in our EC cohort using the TCGA EC dataset. Kaplan- Meier analysis of the OS based on either \(OGT\) or \(OGA\) expression level showed no statistical difference (fig. slc- sld), suggesting that the mRNA abundance of \(OGT\) or \(OGA\) alone is insufficient to reflect the \(O\) - GlcNAcylation level. To better estimate + +<--- Page Split ---> + +the O- GlcNAcylation level using transcriptomic data, we sent 40 O- GlcNAc high- and 15 O- GlcNAc low- frozen EC samples according to their corresponding RL2 staining index (SI) for RNA- seq (fig. s1e). We calculated the Pearson's correlation coefficient (r) of the transcript level with the SI for each gene, and included the top 1000 genes with \(r > 0.3\) in the O- GlcNAcylation correlated geneset (fig. s1f, supplementary table 1). Gene ontology (GO) analysis revealed that they were enriched in biological processes including cilium organization, cilium assembly, and microtubule- based movement (fig. s1g, supplementary table 2). + +We subsequently constructed a mathematical model based on the expression matrix of the O- GlcNAcylation correlated geneset using machine learning algorithms in R to calculate a virtual O- GlcNAc index for each sample in the TCGA EC cohort (fig. s1e). The calculated O- GlcNAc index in the TCGA dataset exhibited significant association with histologic grade and FIGO stage (fig. 2a, supplementary table 3), consolidating the observations made in our EC cohort. Patients in the advanced histologic grade (WHO grade 3) group had higher O- GlcNAc index in comparison to that in the grade 1 or grade 2 group (fig. 2b). Similarly, EC patients at FIGO stages II, III, or IV demonstrated an increased O- GlcNAc index than that at stage I (fig. 2c). Of note, patients in the copy- number high molecular subtype group, which had the worst clinical outcome among all EC cases26, bore significantly higher O- GlcNAc index than that in other groups (fig. 2d). The O- GlcNAc index increased with age (fig. 2e), but showed no difference between diabetic and non- diabetic groups (fig. 2f). We further stratified the EC patients in the TCGA cohort into O- GlcNAc high and O- GlcNAc low groups using the median O- GlcNAc index as the cutoff. Patients from the O- GlcNAc high group experienced significantly shorter progression- free interval (PFI) and OS than that from the O- GlcNAc low group (fig. 2g- 2h, supplementary table 4). + +In summary, the O- GlcNAc status of EC tissues manifests a significant correlation with histologic grade, both in our EC cohort and the TCGA EC dataset, with elevated + +<--- Page Split ---> + +O- GlcNAcylation associating with advanced tumor grade and poor clinical outcome. + +Increase of O- GlcNAcylation by inhibition of OGA promotes proliferation and stemness in normal endometrial organoids + +To dissect the functional impact of altered O- GlcNAcylation level on endometrial tissues, we generated endometrial organoids from surgical samples, including normal endometrial epithelial organoids (EE- Os) and endometrial cancer organoids (EC- Os). The EE- Os retained many characteristics of endometrial epithelium, including production of mucins, and expressions of estrogen receptor \(\alpha\) (ER) and progesterone receptor (PR) (fig. 3a). The EC- Os however manifested more irregular cell organizations and elevated O- GlcNAcylation level than the EE- Os, in accordance with their primary tissues (fig. 2a- 2b). + +We treated the EE- Os with OGA small molecular inhibitor Thiamet- G (TMG) to increase the cellular O- GlcNAcylation level (fig. 3b). The addition of TMG resulted in enhanced colony formation and organoids growth of EE- Os (fig. 3c- 3e), along with a rise in the number of mitotic cells within each EE- O (fig. 3f- 3g). Acetylated alpha- tubulin (Ac- tubulin) and PAEP are differentiation markers for multiciliated epithelial cells and secretory cells respectively in the endometrium27. TMG treatment reduced the number of both PAEP positive cells and Ac- tubulin labeled multiciliated cells (fig. 3h- 3i), suggesting that the elevated O- GlcNAcylation level caused de- differentiation of the endometrial cells in the EE- Os. We further examined the expression levels of a panel of stemness markers of the endometrium, including SSEA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2. In contrast to PAEP whose mRNA level was decreased upon TMG treatment, all the examined stemness markers showed upregulated expressions (fig. 3c). + +To further characterize the influence of TMG treatment on different cell subtypes in the EE- Os, the control and TMG treated EE- Os were subject to single- cell RNA- seq analysis (fig. 3j). The cells were clustered and classified into six major subtypes according to the specific expressions of known markers28,29: pre- ciliated, ciliated, stem, proliferative, + +<--- Page Split ---> + +O- GlcNAc- related stem- like, and inflammatory (fig. s2d- s2g, supplementary table 5). Of note, we identified an O- GlcNAc- related stem- like subtype in which the cells displayed activated signaling pathways regulating pluripotency of stem cells, as well as the O- glycan biosynthesis (fig. s2c). The TMG treatment of EE- Os resulted in a substantial decrease of cells in the ciliated and pre- ciliated subtypes, and a concurrent increase of cells in the proliferative and O- GlcNAc- related stem- like subtypes (fig. 3k). Together, these results suggest that upregulation of O- GlcNAcylation level promotes proliferation and stemness of endometrial epithelial cells. + +Inhibition of OGT decreases cell proliferation and induces differentiation and cell death in endometrial cancer organoids + +The EC- Os faithfully reflected the molecular characteristics of their primary EC tissues, and bore higher O- GlcNAcylation level relative to EE- Os (fig. 4a and s2b). We treated the EC- Os with OSMI- 1, a chemical inhibitor of OGT, to reduce the O- GlcNAcylation level (fig. 4b). Addition of OSMI- 1 impeded the formation and growth of EC- Os (fig. 4c). A significant fraction of the EC- Os displayed darkening and cell lysing in the presence of OSMI- 1, resulting in reductions of both the number and size of the EC- Os compared to time- matched control (fig. 4d- 4c). TUNEL stainings revealed that many cells in the OSMI- 1 treated EC- Os underwent apoptosis (fig. 4f). We performed immunofluorescence on the remaining EC- Os with relatively normal size and morphology. Mitotic cells as visualized by phospho- histone H3 (PH3) staining became barely detectable in EC- Os after OSMI- 1 treatment (fig. 4g- 4h). Meanwhile, the population of both the PAEP positive secretory cells and Ac- tubulin labeled multiciliated cells increased in these EC- Os (fig. 4i- j), suggesting that OSMI- 1 treatment promoted differentiation. Consistently, the expression of stemness markers, including SSEA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2, were significantly downregulated in the OSMI- 1 treated EC- Os, accompanying the upregulation of the differentiation marker PAEP (fig. 4k). + +In summary, these results indicate that a balanced O- GlcNAcylation level is crucial for + +<--- Page Split ---> + +maintaining the cell fate in endometrial organoids. Elevated O- GlcNAcylation promotes proliferation and stemness of EE- Os, whereas downregulated O- GlcNAcylation induces differentiation and cell death, limiting the uncontrolled proliferation of cells in the EC- Os. + +Genome- wide screen for tumor suppressors that maintain O- GlcNAcylation homeostasis To identify crucial factors regulating O- GlcNAcylation homeostasis in EC, we conducted a comprehensive genome- wide CRISPR- Cas9 knockout screen. A lentiviral single guide RNA (sgRNA) library targeting 19,050 genes (6 sgRNAs/gene) was transduced into 293T cells, along with 1000 nontargeting control sgRNAs, at a multiplicity of infection (MOI) of 0.3 to ensure each cell expressed only one sgRNA. Following cell staining with an anti- O- GlcNAc antibody (RL2), we isolated the top \(5\%\) RL2- positive cells via fluorescence- activated cell sorting (FACS) and conducted deep sequencing of the sgRNAs from this cell population (fig. 5a- 5b). The sgRNA abundance was then used to calculate a robust rank aggregation (RRA) score for each gene using MAGeCK30, and the genes were ranked accordingly, with a smaller RRA score indicated greater essentiality (supplementary table 6). We reviewed the literatures and collected known regulators whose inactivation could impact cellular O- GlcNAcylation homeostasis31- 46. Genes that negatively regulate O- GlcNAcylation, such as TSC2, SIRT1, and TP53, had smaller RRA scores and were enriched in the first half of the gene list, comparing to the known positive regulators of O- GlcNAcylation (fig. 5c). We performed Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis on the 1038 high- confidence genes ( \(P < 0.05\) ) from the genome- wide screen. These genes were enriched in pathways including ECM- receptor interaction, thermogenesis, histidine metabolism, proteoglycan in cancer, and maturity onset diabetes of the young (fig. 5d, supplementary table 7). + +To further pinpoint key regulators that impact O- GlcNAcylation level in EC tissues, we cross- referenced the 1038 positive hits in the screen with 526 putative tumor suppressor genes of EC47,48. As a result, 18 overlapping genes were identified, including ACVR1C, AGTR1, CADM2, PRKAA1, CDKN1C, CMTM3, DIRAS3, SIK1, EPHB4, GATA5, ITGAV, + +<--- Page Split ---> + +KLF10, MAP3K8, PLA2G2A, PTPN11, RNASEL, SPARCL1, and FBXO31 (fig. 5e). We conducted Kaplan- Meier analysis using the TCGA EC dataset for each gene, and found that only FBXO31 showed downregulated expression in EC that was associated with poor survival (fig. s3a and 5g). Therefore, we generated FBXO31 knockout (FBXO31- KO) 293T cells using CRISPR (fig. s3b). Immunostaining with RL2 antibody confirmed that O- GlcNAcylation level was significantly increased in the FBXO31- KO cells (fig. 5f). + +FBXO31 interacts with and ubiquitinates OGT to limit the O- GlcNAc level + +FBXO31 functions as a substrate recognition component in the SCF ubiquitin E3 ligase complex to control the homeostasis of many proteins49- 54. Accordingly, FBXO31 might regulate O- GlcNAcylation level by directly binding and ubiquitinating the O- GlcNAc transferase OGT. To confirm the interaction between FBXO31 and OGT, we performed pull- down assay using bacterially purified GST- OGT to incubate with lysates of 293T cells expressing GFP- FBXO31. Western blot showed that GST- OGT pulled down significant amount of GFP- FBXO31 relative to GST control (fig. 6a). We further validated the interaction using co- immunoprecipitation in 293T cells overexpressing Flag- OGT and GFP- FBXO31. GFP- FBXO31 was co- immunoprecipitated with Flag- OGT, and both the amounts of Flag- OGT and GFP- FBXO31 in the immunoprecipitant were increased in the presence of the proteasome inhibitor MG132 (fig. 6b). + +To assess whether the interaction with FBXO31 controlled the protein homeostasis of OGT, we transfected 293T cells with increasing amounts of GFP- FBXO31 and detected the levels of OGT as well as O- GlcNAcylation by western blot. Both the OGT protein and cellular O- GlcNAcylation levels demonstrated a negative correlation with the amount of GFP- FBXO31 (fig. 6c). Additionally, the downregulation of OGT induced by GFP- FBXO31 overexpression was significantly reversed by MG132, suggesting that FBXO31 controlled the OGT level via the ubiquitin- dependent proteasome degradation process (fig. 6d). To ascertain that FBXO31 could induce ubiquitination of OGT, we immunoprecipitated OGT from 293T cell lysates overexpressing GFP- FBXO31 and + +<--- Page Split ---> + +HA- ubiquitin. Western blot detected strong polyubiquitination of OGT in the presence of GFP- FBXO31 (fig. 6e). We further tested whether FBXO31 could ubiquitinate OGT in vitro. The SCF complex was affinity- purified with anti- HA magnetic beads from 293T cells expressing HA- tagged Skp1, Cul1, and Roc1 with or without FBXO31, and then incubated with bacterially purified E1, E2, ubiquitin, and His- OGT. Polyubiquitination signals of His- OGT were detected, suggesting that the FBXO31- containing SCF complex could directly ubiquitinate OGT (fig. 6f). Skp1 in the SCF complex recruits F- box proteins via their F- box motif. We mutated the F- box of FBXO31 (FBXO31 \(\Delta \mathrm{F}\) ) and assessed its ability to induce polyubiquitination of OGT in 293T cells. Overexpression of HA- ubiquitin and GFP- FBXO31 resulted in strong polyubiquitination of the immunoprecipitated Flag- OGT, which was significantly reduced when GFP- FBXO31 was replaced with the GFP- FBXO31 \(\Delta \mathrm{F}\) mutant (fig. 6g). These results confirmed that FBXO31, together with other components of SCF complex, possessed a new ubiquitin E3 ligase activity toward OGT. We further evaluated the impact of FBXO31 in controlling the cellular OGT homeostasis using FBXO31- KO 293T cells. Both the OGT and O- GlcNAcylation levels were increased in FBXO31- KO cells (fig. 6h). Cycloheximide (CHX) treatment, which blocked new protein synthesis, uncovered that the half- life of OGT was significantly extended in FBXO31- KO cells relative to control (fig. 6i), indicating that FBXO31 is indispensable for limiting the cellular OGT level. + +Loss of FBXO31 increases O- GlcNAcylation and promotes endometrial organoids growthWe investigated the clinical relevance of the FBXO31- mediated ubiquitination of OGT using the endometrial specimens in our EC cohort. IHC staining revealed that the protein level of FBXO31 was significantly downregulated in EC relative to normal endometrial tissues, often manifesting an anti- correlation pattern to that of O- GlcNAcylation (fig. 7a- 7b). We semi- quantified the expression level of FBXO31 based on the IHC signals, and found that the FBXO31 protein level in the O- GlcNAc low EC group was markedly higher than that in the O- GlcNAc high group (fig. 7c). Western blot uncovered that the FBXO31 protein level was decreased, accompanying the increase of OGT level, in EC- Os comparing to the + +<--- Page Split ---> + +EE- Os (fig. 7d), suggesting that the elevated O- GlcNAcylation in EC tissues was due to stabilization and upregulation of OGT. Moreover, we categorized the EC cases into FBXO31- low and FBXO31- high groups. The FBXO31 expression exhibited significant association with histologic grade, diabetes, as well as the O- GlcNAc status (table 3). + +To elucidate the functional impact of FBXO31 alterations in endometrial tissues, we knocked down the expression of FBXO31 using lentivirus- mediated expression of shRNAs in EE- Os. Downregulation of FBXO31 resulted in increased amount of O- GlcNAcylation in EE- Os (fig. 7e). Particularly, the growth of EE- Os was significantly enhanced by FBXO31 knockdown, in alignment with upregulated expression of the stemness markers SSA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2 (fig. 7f- 7g). This enhanced growth of EE- Os after FBXO31 knockdown could be inhibited by OSMI- 1 treatment, indicating that it was a result of elevated O- GlcNAcylation (fig. 7h). Reciprocally, given that FBXO31 was downregulated in EC- Os, we supplemented the EC- Os with GFP- FBXO31 or GFP control using lentivirus- mediated transduction. Overexpression of GFP- FBXO31 significantly impeded the formation of EC- Os (fig. 7i). + +In summary, our results identify FBXO31 as one of the key rheostats that controls the O- GlcNAc homeostasis by ubiquitinating OGT. FBXO31 is frequently downregulated in EC, resulting in stabilization of OGT and elevation of cellular O- GlcNAcylation level that advance endometrial malignancy. + +## Discussion + +This study delves deeply into the intricate relationship between O- GlcNAc homeostasis and the progression of EC, elucidating the clinical significance of abnormal O- GlcNAcylation and unveiling an important regulatory module controlling the O- GlcNAc homeostasis in endometrial tissues. First, by examining the O- GlcNAc status in a Chinese EC cohort containing 219 tumors, we found that O- GlcNAcylation level correlates with histologic grade, FIGO stage, and distant metastasis. Patients with higher O- GlcNAcylation levels + +<--- Page Split ---> + +experienced shorter OS and PFS, indicating that it is an independent risk factor for poor prognosis in EC. Second, we performed RNA- seq on 55 frozen EC samples (40 O- GlcNAc high- and 15 O- GlcNAc low- specimens according to their corresponding immunohistochemistry (IHC) scores) and constructed a predicting model to evaluate the O- GlcNAc level based on transcriptomic data. We calculated the virtual O- GlcNAc index for each sample in the TCGA EC cohort and validated that O- GlcNAc index is associated with histologic grade, FIGO stage, and poor prognosis. Third, we assessed the functional significance of O- GlcNAcylation in EC progression by treatment of patient- derived organoids with clinically relevant chemical inhibitors. Inhibition of OGA by TMG increased O- GlcNAcylation and promoted proliferation and stem- like cell properties in normal EE- Os, whereas inhibition of OGT by OSMI- 1 decreased O- GlcNAcylation, downregulating stemness and limiting the proliferation of EC- Os. Last, we performed genome- wide CRISPR screen for known tumor suppressors whose inactivation could increase cellular O- GlcNAc level. We identified that FBXO31 can function as a substrate receptor of the SCF ubiquitin ligase complex to ubiquitinate the O- GlcNAc transferase OGT in vitro. FBXO31 was downregulated in EC tissues, anticorrelating with O- GlcNAc levels. Knockdown of FBXO31 in EE- Os promoted expression of stemness markers and accelerated organoid growth, suggesting that inactivation of FBXO31 advances endometrial malignancy by stabilizing OGT and elevating cellular O- GlcNAcylation level. + +Integrative analysis of O- GlcNAcylation in large clinical EC cohort to assess its relationships with current histomorphologic and molecular subtypes of EC had not been conducted till this study. Altered O- GlcNAc cycling has been linked to many characteristics of EC cells, impacting their survival and proliferation signaling, epithelial- mesenchymal transition (EMT) and metastatic behaviors, drug- resistance, as well as metabolic and cell fate plasticity1,5,8. A previous report using 76 EC samples revealed that the two executing enzymes of O- GlcNAcylation, OGT and OGA, manifested increased mRNA levels in ECs of higher histologic grade relative to the well- differentiated tumors22. A more recent IHC analysis on a tissue microarray containing 28 EC specimens showed that both the OGT and + +<--- Page Split ---> + +O- GlcNAcylation levels were increased in EC tissues than the adjacent normal endometrial tissues23. This pilot study indicated that increased O- GlcNAcylation was associated with histologic grade, clinical stage, and lymph node metastasis. However, when repeating the IHC analysis using the same tissue microarray, we only observed increased OGT and O- GlcNAcylation levels in ECs but failed to associate O- GlcNAcylation levels with any of the clinical parameters, probably due to the limited sample size and differences in inclusion and exclusion criteria. Nonetheless, when we expanded the analysis to our EC cohort containing 219 patients as well as the TCGA EC dataset, the clinical significance of O- GlcNAcylation became invariable. The O- GlcNAcylation level shows strong association with histologic grade, FIGO stage, and poor prognosis. Particularly, the O- GlcNAc high EC group enriches more serous carcinoma as well as high grade endometrioid carcinoma patients. This is perhaps because serous carcinoma often carries TP53 mutations and p53 is a known negative regulator of O- GlcNAcylation level not only identified in our genome- wide screen but also reported in a previous study31. Moreover, EC patients belonging to the copy- number high molecular subtype group showed significantly higher O- GlcNAcylation level than that in the other groups. These results suggest that O- GlcNAcylation is a useful factor complementary to the current classification system to better identify EC patients with poor clinical outcome. + +Our understanding on the molecular circuitry controlling the cellular O- GlcNAcylation homeostasis is far from complete. Given that O- GlcNAcylation is dependent on nutrient availability, metabolic factors such as GFPT1, POLDIP2, and PPM1K have been reported to influence O- GlcNAc level by modulating the metabolic flux of the HBP pathway31,34,39,45,46. However, emerging evidence indicates that O- GlcNAcylation may also be regulated by non- nutrient dependent mechanisms, particularly at the protein level of OGT32,33,37,41- 43. OGT has been reported to be regulated by the balance of ubiquitination and deubiquitination9,10,12,13. The E3 ligases XIAP and E6AP could promote the ubiquitin- dependent proteasome degradation of OGT10,11. The histone demethylase LSD2 displayed an atypical ubiquitin E3 ligase activity toward OGT in the A549 cells9. However, + +<--- Page Split ---> + +none of these reported E3 activities showed clinical relevance in ECs. Our results uncovered that FBXO31, together with other components in the SCF complex, functions as a new E3 ligase for OGT, impacting the progression of endometrial malignancy. FBXO31 is a tumor suppressor gene located in the 16q24.3 region, with frequently observed loss of heterozygosity in several cancers, including breast, ovarian, hepatocellular, and prostate cancers55,56. The SCFFBXO31 complex can ubiquitinate many protein substrates that are cell cycle regulators, such as cyclin D149, Cdt150, MDM252, and cyclin A57; signaling molecules, such as c- Myc58, β- catenin59, and MKK651; epithelial- mesenchymal transition (EMT) factors, Snail160 and Slug61; as well as ferroptosis inhibitor GPX462. How it recognizes these substrates remains unclear. Our results indicate that FBXO31 can directly bind to OGT, triggering its polyubiquitination. Future molecular and structural characterization of this interaction between FBXO31 and OGT may help elucidate the substrate recognition mechanism of SCFFBXO31, paving the way for the development of new intervention strategies. + +Last but not least, how the aberrant O- GlcNAcome in ECs influences the tumor cells' proliferation, survival, and cell fate plasticity is not fully understood. Low- throughput, individual characterizations of potential O- GlcNAcylation substrates have revealed that many EC- related oncogenes and tumor suppressors, such as PI3K, PTEN, ARID1A63, p5364, Myc65, and β- catenin66, possess O- GlcNAcylation sites. Particularly, in alignment of our finding that increased O- GlcNAcylation promotes stemness of EC cells, master regulators controlling the stem cells' self- renewal and pluripotency, including Oct47, Sox26, and Sox99, are able to be modified by O- GlcNAcylation. It is worthy of functional interrogation of these putative O- GlcNAcylated substrates in patients- derived endometrial organoids. Ultimately, future development of high- throughput, tissue- specific proteomic profiling methods is needed for fully capture the spatiotemporal dynamics of the O- GlcNAcome during the progression of ECs and other pathophysiological processes, consolidating the foundation of targeting O- GlcNAc cycling to develop new therapeutic strategies in clinical settings. + +<--- Page Split ---> + +## Methods + +Human tissues + +All fresh tissues and paraffin- embedded (FFPE) tissues were prospectively obtained from patients with endometrial diseases at Xiangya Hospital, Central South University. Clinical data and histopathological characteristics were retrieved from patient records and routine pathology reports. The study was approved by the Medical Ethics Committee of Central South University (No. 202103076), and all participating patients provided informed written consent. The study was registered with and approved by the Human Genetics Resource (HGR) office of the Minister of Science and Technology of China. + +Organoids culturing from endometrial surgical samples + +The endometrial organoids were generated as previously described70. Tumor tissues and normal tissues were isolated and stored in ice- cold serum- free DMEM medium supplemented with 1% penicillin- streptomycin. The tissues were then washed in ice- cold DPBS (Biological Industries) supplemented with penicillin- streptomycin and minced into small pieces. The tissues were digested by collagenase IV (1- 2 mg/mL; 17104019, Thermo Fisher Scientific) in the presence of Rock inhibitor (10 μM; SCM075, Merck Millipore) and penicillin- streptomycin for 1 hour on a shaker at 37°C, then incubated for 15 minutes in TrypLE (1×; 12604013, Thermo Fisher Scientific) supplemented with Rock inhibitor and penicillin- streptomycin. Subsequently, the tissue digests were stopped by ice- cold serum- free DMEM/F12 and after centrifugation, a 100- μm cell strainer was used to obtain cell pellets. Finally, the cell pellets were resuspended in 70% matrigel/30% DMEM/F12 (356231, Corning and 11039021, Gibco, respectively) and seeded in 50 μL droplets in non- treated 24- well plates. After incubation at 37°C and 5% CO2 in a cell culture incubator for 20- 30 minutes, the pre- warmed organoid complete medium (DMEM/F12 supplemented with 1% penicillin- streptomycin, 2% B27 supplement minus vitamin A (12587010, Gibco), 5% R- spondin conditioned medium, 1% chemically defined lipid concentrate (11905031, Gibco), recombinant human Noggin 100 ng/mL (HY- P7051A, MCE), 1% N2 (17502048, + +<--- Page Split ---> + +Gibco), N-acetyl- L- cysteine 1.25 mM (A7250, Sigma Aldrich), Nicotinamide 10 \(\mu \mathrm{M}\) (73240, Sigma Aldrich), recombinant human EGF 50 ng/mL (236- EG- 01M, R&D Systems), Y- 27632 10 \(\mu \mathrm{M}\) (SCM075, Sigma Aldrich), 17- \(\beta\) estradiol 10 nM (E8872, Sigma Aldrich), SB202190 0.1 \(\mu \mathrm{M}\) (S7067, Sigma Aldrich), A83- 01 0.25 \(\mu \mathrm{M}\) (SML0788, Sigma Aldrich), recombinant human IGF 40 ng/mL (100- 11, Peprotec), recombinant human HGF 20 ng/mL (100- 39, Peprotec), IL- 6 5 ng/mL (200- 06, Peprotec)) was added. The organoid medium was changed every 2 days, and the organoids were passaged after 7- 10 days of culture. + +## Immunohistochemistry (IHC) + +Endometrial cancer tissue array was purchased from Xinchao Biotechnology Company (HUteA060CS01, Shanghai, China), consisting of 26 pairs of cancerous and paracancerous tissue specimens, along with an additional 8 cases of cancerous tissue without paired paracancerous tissue. After removing the incomplete tissue spots, 31 cases of cancer tissue and 23 cases of para- cancerous tissue were included in IHC analysis. IHC was performed as previously described71, with primary antibody incubation overnight after antigen retrieval and endogenous peroxidase activity blocking on paraffin sections. The IHC staining signal levels were blindly scored by two independent assessors without knowledge of clinical parameters. The staining index (SI) was calculated by multiplying the staining intensity score (0- 3) and the proportion of positively stained tumor cells score (0- 4), resulting in a SI ranging from 0 to 12. High and low expression were defined as SI 0- 6 and SI 8- 12, respectively72. + +## Survival analysis + +Progression- free survival (PFS) was calculated as the time between the surgery that procured the sample and the date of disease progression or of a new metastatic event in a different location. Overall survival (OS) was defined as the interval between the date of surgery and the date of death or last follow- up. Progression- free interval (PFI) was defined as the duration from surgery to the first occurrence of disease progression or death after treatment. The curves were stratified based on the O- GlcNAcylation (RL2) level. Log- rank + +<--- Page Split ---> + +test was used to compare the two groups over a follow- up time of 61 months. Kaplan- Meier survival curves were generated and compared using GraphPad Prism (version 8.0.2). + +Generation of O- GlcNAc index prediction model + +The RNA- seq data of 15 low O- GlcNAcylation level (RL2 by IHC) tumor tissues and 40 high O- GlcNAcylation level tumor tissues were processed to identify the O- GlcNAc correlated genes. The gene expression matrix of these 55 EC samples was correlated with the O- GlcNAc IHC staining index using the Pearson correlation method in the mlr3. filters package within the mlr3 framework in R. The top 1000 genes with a correlation coefficient greater than 0.3 were included in the O- GlcNAc correlated geneset. Subsequently, mlr3 learners including six regression model- based approaches (regr.lm, regr.glmnet, regr.kknn, regr.ranger, regr.rpart, regr.svm) was applied to the expression matrix of the 1000 O- GlcNAc correlated genes. The O- GlcNAc indices for the 55 EC tissues were calculated, subjecting to 5- fold cross- validations of training and ranking based on predefined performance metrics. The reliability of the prediction model was assessed by comparing the calculated O- GlcNAc indices with actual IHC SI scores. The regr.glmnet demonstrated the lowest mean squared error (MSE) and was selected for establishment of the final prediction model. The O- GlcNAc indices were then calculated using the prediction model for the 589 EC samples in TCGA. The patients were categorized into high and low O- GlcNAc groups using the median of the calculated O- GlcNAc indices. Wilcoxon Mann- Whitney tests were used to assess differences between the two groups in terms of histologic grade, FIGO stage, molecular subtype, age, and diabetes. Log- rank tests were employed to compare the OS and PFI differences between the high and low O- GlcNAc groups, and Kaplan- Meier survival curves were generated and compared using R (version 4.03). + +Immunofluorescence of organoids + +Immunofluorescence staining experiments were performed on organoids as previously described73. When the organoids reached a size of approximately \(100 \mu \mathrm{m}\) , they were selected for staining. After washing twice with pre- cooled DPBS, \(500 \mu \mathrm{L}\) of cell recovery + +<--- Page Split ---> + +solution (354253, Corning) was added to each well, and the matrigel was dissolved on ice to ensure that the morphology of the organoids was not disrupted. After 30 minutes, all the organoids were collected into a \(15~\mathrm{mL}\) centrifuge tube, fixed with \(4\%\) paraformaldehyde for 30 minutes, and then centrifuged to remove the supernatant. Next, \(10~\mathrm{mL}\) of \(1\%\) PBST was added to stop the tissue fixation. After blocking with Organoid Washing Buffer (OWB, \(0.1\%\) Triton X- 100, \(0.2\%\) BSA in DPBS), the primary antibody was added and incubated overnight at \(4^{\circ}\mathrm{C}\) with shaking at \(60~\mathrm{rpm}\) . On the following day, the organoids were washed three times with OWB for 2 hours each time, and then the corresponding fluorescent secondary antibody was added. The organoids were incubated overnight on a shaker in the dark. On the third day, \(4^{\prime},6\) - Diamidino- 2- phenylindole dihydrochloride (DAPI, D9542, Sigma) at \(10~\mu \mathrm{g / mL}\) was added for 30 minutes. After washing with OWB, the samples were spun down at \(70\times \mathrm{g}\) for 5 minutes at \(4^{\circ}\mathrm{C}\) . Finally, the organoids were resuspended with fructose- glycerol clearing solution ( \(60\%\) glycerol and \(2.5\mathrm{M}\) fructose in \(\mathrm{ddH_2O}\) ) and imaged using an LSM880 confocal microscope (Zeiss). A cell death detection (TUNEL) kit (Roche) was used to identify dead cells in accordance with the company's description. All the antibodies used in this study were listed in supplementary table 9. + +## Lentiviral transduction of organoids + +For organoid lentiviral transduction, pLKO.1- puro vectors and TK- PCDH- copGFP- T2A- Puro vectors were used. The organoids were washed twice with pre- cooled DPBS, and \(500~\mu \mathrm{L}\) of TrypLE (12604013, Thermo Fisher Scientific) was added to each well for 10 minutes at \(37^{\circ}\mathrm{C}\) . The matrigel was disrupted by pipetting the mixture up and down repeatedly during digestion. TrypLE was inactivated by adding \(10~\mathrm{mL}\) of ice- cold serum- free DMEM/F12, and the mixture was centrifuged for 5 minutes at \(200\times \mathrm{g}\) . After digestion, the organoids were made into single cells or cell mass and resuspended in virus infection solution containing Rock inhibitor, polybrene, and concentrated lentivirus in organoid culture media. The cell suspension was added to a 6- well plate, spun at \(2000~\mathrm{rpm}\) for 1 hour, and then incubated at \(37^{\circ}\mathrm{C}\) for 5- 6 hours. The cells were then transferred to a 15 mL centrifuge tube, washed twice with serum- free DMEM/F12, and seeded in a prewarmed + +<--- Page Split ---> + +24- well plate with \(70\%\) matrigel. Then, \(500~\mu \mathrm{L}\) of organoid medium was added to each well, followed by incubation at \(37^{\circ}\mathrm{C}\) with \(5\%\) \(\mathrm{CO_2}\) for 20 minutes. The medium was changed every 2 days. Puromycin selection (1 \(\mu \mathrm{g / mL}\) ) in organoid culture was conducted for 3- 4 days to establish stably infected organoids. The stable organoids were validated by western blot or quantitative RT- PCR. + +## Quantitative RT-PCR + +RNA extraction was performed using TRIzol (87804, Life Technologies) according to the manufacturer's protocol for all samples, including cells, organoids, and primary tissues. The extracted RNA was then reverse transcribed to cDNA using the PrimeScript RT Reagent Kit (RR037A, Takara). The cDNA was used as a template for qPCR, which was performed using the SYBR Green qPCR Master Mix (QST- 100, SolomonBio) on the QuantStudio 3 Real- Time PCR system (Applied Biosystems). All the primers were listed in supplementary table 8. + +## Western blot and immunoprecipitation + +Cells were lysed in sample buffer (2% SDS, \(10\%\) glycerol, and \(62.5\mathrm{mM}\) Tris- HCl, pH 6.8) supplemented with \(1\times\) protease inhibitor cocktail (P8340, Sigma). The protein concentration was measured using a BCA kit (P0009, Beyotime). Cell lysates were separated by SDS- PAGE and transferred onto a nitrocellulose membrane. The membrane was then blocked with \(5\%\) non- fat dry milk for 1 hour at room temperature and probed with the indicated primary antibodies overnight at \(4^{\circ}\mathrm{C}\) Antigen- antibody complexes were detected by incubating with horseradish peroxidase secondary antibodies followed by ECL substrates (WBKLS0500, Millipore). For immunoprecipitation experiments, cells were washed twice with ice- cold PBS and then lysed in lysis buffer (20 mM Tris- HCl (pH 8.0), 137 mM NaCl, \(1\%\) NP- 40, \(2\mathrm{mM}\) EDTA) on ice for 30 minutes. Cell lysates were gently mixed with specific antibodies overnight at \(4^{\circ}\mathrm{C}\) under gentle rotation, then incubated with protein A/G beads (SC- 2003, Santa cruz) for 1- 2 hours at \(4^{\circ}\mathrm{C}\) Immunoprecipitants were washed three times with lysis buffer. After the final wash, the supernatant was aspirated and discarded, + +<--- Page Split ---> + +and the pellet was resuspended in \(2 \times\) SDS sample buffer (0.125 M Tris HCl (pH 6.8), 4% SDS, 20% glycerol, 2% \(\beta\) - mercaptoethanol, 0.02% bromophenol blue). The sample was then subjected to reducing SDS- PAGE and western blot. All the antibodies used in this study were listed in supplementary table 9. + +## Cell culture and generation of cell lines + +HA- R- Spondin1- Fc 293T cell line (3710- 001- 01, R&D Systems) was used to produce R- spondin conditional media. HEK293T and HeLa cells were maintained in DMEM (06- 1055- 57- 1 ACS, Vivocell) supplemented with 10% FBS. All cells were cultured at \(37^{\circ}\mathrm{C}\) in a humidified incubator with 5% \(\mathrm{CO_2}\) and periodically screened for Mycoplasma contamination. To generate 293T FBXO31 KO cell lines, the cells were transfected with LentiCRISPR- V2 plasmid carrying sgFBXO31 (supplementary table 8) and further selected with \(1 \mu \mathrm{g / mL}\) puromycin (s7417, Selleck) for 3 days. The cells were then plated at single- cell density in 100 mm petri dishes, and the individual clones that emerged were picked and replated into 24- well plates. The loss of FBXO31 expression was confirmed by western blot and Sanger sequencing. + +## In vivo and in vitro ubiquitination assay + +For detection of ubiquitinated proteins in vivo, 293T cells were co- transfected with expression vectors for HA- ubiquitin and the indicated proteins. Polyubiquitinated OGT was detected by immunoprecipitation of OGT with ANTI- FLAG® M2 Affinity Gel (A2220, Merck Millipore) under denaturing conditions followed by Western blot with an anti- HA antibody. In vitro ubiquitination was performed as previously described74. The SCF- FBXO31 (E3) complexes were immunopurified from the cell lysate using Pierce™ Anti- HA Magnetic Beads (88836, Thermo Fisher Scientific) and incubated with His- OGT fusion protein expressed and purified from E. coli as previously reported75 in the presence of recombinant purified E1 (UBA1; 11990- H20B, sinobiological), E2 (UBE2D1; 11432- H07E, sinobiological), recombinant human ubiquitin protein (U- 100H, Boston Biochem), and + +<--- Page Split ---> + +ubiquitination buffer (20 mM Tris- HCl, pH 7.5, 5 mM MgCl2, 0.5 mM DTT, 2 mM ATP). The reaction was stopped by adding \(2 \times\) SDS sample buffer and boiling for 10 minutes. + +CRISPR- Cas9 screen and data analysis + +The human genome- scale CRISPR knockout library (GeCKO v2, Addgene #1000000048) in the lentiCRISPR v2 vector (Addgene #52961) consists of 123,411 sgRNAs that target 19,050 protein- coding genes (6 sgRNAs per gene) and 1,000 nontargeting control sgRNAs was used \(^{76,77}\) . The human GeCKO v2 library was transduced into 293T cells by lentivirus at a multiplicity of infection of 0.3. Cells were selected with puromycin for 7 days followed by fluorescence- activated cell sorting (FACS) based on their O- GlcNAc staining intensities. An unsorted sample was used to assess sgRNA library coverage, and the sorted RL2 high population was subjected to genomic DNA extraction. The inserted sgRNA library was amplified by two steps of PCR for next- generation sequencing. Each screen was performed twice. For data analysis, reads from the fastq files generated by sequencing were tallied for each guide by taking the first 20 bp from each read and mapping to the identical short gRNA sequence. For each screen, a table of reads per guide that includes the counts from the RL2 high population of both replicates was generated and loaded into MAGeCK \(^{30}\) . Top genes were determined based on their mean log2 fold change, FDR, and robust ranking aggregation (RRA) score. + +## Data download + +The TCGA UCEC dataset used in this study, including the gene raw count data (htseq- count files), and the annotated somatic simple nucleotide variation files (MuTect2 VCF), were downloaded using the gdc- client v1.6.0. The clinical OS and PFI information were obtained from Liu.et al \(^{78}\) . + +## RNA-seq and bioinformatic analysis + +Total RNA was isolated from EC tissues, and libraries were generated using the NEBNext UltraTM RNA Library Prep Kit (New England Biolabs) for the Illumina system. Sequencing + +<--- Page Split ---> + +was conducted using the Illumina Novaseq 6000 platform (Novogene). Trim Galore v0.6.4 was employed to eliminate adapter sequences and remove reads of poor quality. Subsequently, the reads from each RNA- seq data were aligned to the hg38 genome assembly using STAR v2.7.2d. The key alignment parameters were set as follows: '- - outFilterMismatchNoverLmax 0.04 - - outSAMtype BAM SortedByCoordinate - - outFilterMultimapNmax 500 - - outMultimapperOrder Random - - outSAMmultNmax 1'. Gene expression was quantified using featureCounts v2.0.0. Heatmaps were created using R package pheatmap v1.0.12. The GO enrichment analysis was performed using the function "enrichGO" from the R package clusterProfiler v3.10.18879. + +## Organoid single-cell analysis + +EE- Os were treated with \(10\mu \mathrm{M}\) TMG or vehicle control (0.1% DMSO). Following treatment, the organoids were dissociated into single cells using TrypLE digestion, and the mixture was passed through a \(40\mu \mathrm{m}\) cell strainer. The cells were then counted and viability assessed. Single- cell suspensions ( \(2\times 10^{5}\) cells/mL) in PBS (HyClone) were loaded onto microwell chip using the Singleron Matrix® Single Cell Processing System. Barcoding beads were subsequently collected from the microwell chip, followed by reverse transcription of the mRNA captured to obtain the cDNA. After PCR amplification, the amplified cDNA was then fragmented and ligated with sequencing adapters. The scRNA- seq libraries were constructed according to the protocol of the GEXSCOPE® Single Cell RNA Library Kits (Singleron)80. Individual libraries were diluted to \(4\mathrm{nM}\) , pooled, and sequenced on Illumina Novaseq 6000 with 150 bp paired end reads. Raw reads were processed to generate gene expression profiles using CeleScope v2.0.7 (Singleron) with default parameters. Briefly, barcodes and UMIs were extracted from R1 reads and corrected. Adapter sequences and polyA tails were trimmed from R2 reads and the trimmed R2 reads were aligned to the hg38 transcriptome using STAR (v2.6.1b). Uniquely mapped reads were then assigned to exons with featureCounts (v2.0.1). Successfully assigned reads with the same cell barcode, UMI and gene were grouped together to generate the gene expression matrix. Omicverse V1.5.4 was used for quality control, dimensionality reduction and + +<--- Page Split ---> + +clustering under Python 3.8. The following criteria were used to filter the expression matrix: 1) cells with gene count less than 500 were excluded; 2) cells detected genes less than 250 were excluded; 3) cells with mitochondrial content more than \(15\%\) were excluded; 4) genes expressed in less than 3 cells were excluded. After filtering, 20736 cells were retained for the downstream analyses. The raw count matrix was normalized by total counts per cell and logarithmically transformed into normalized data matrix. The top 3000 highly variable genes were selected by setting flavor = 'seurat'. Principal Component Analysis (PCA) was performed on the scaled variable gene matrix, and 50 principal components were used for clustering and dimensional reduction. 'Harmony' was employed to integrate samples. Cells were separated into 6 clusters using Leiden algorithm with the resolution parameter at 0.25. Subsequently, omicverse was used to calculate the ranking of highly differential genes in each cluster to identify marker genes. Cell clusters were visualized using Minimum-Distortion Embedding (mde). Cell types were annotated based on the cell type auto- annotation tool SCSA, and the known cellular markers from the literature28,29,81- 83; epithelial cells (EPCAM, KRT8, KRT18), stem cells (LGR5, SOX9, POU5F1, PROM1, AXIN2), proliferative cells (MMP7, TOP2A, MK167), ciliated cells (PIFO, FOXJ1, TPPP3), pre- ciliated cells (CDC20B, DYDC2, CCNO), and inflammatory cells (IL4I1, IL32, S100A9, CD14, IL1RN). The O- GlcNAc related stem like cells annotation was mainly based on the results of KEGG pathway enrichment. The expression of markers used to identify each cell type was visualized using violin plot. + +AUCell geneset enrichment analysis + +Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were collected and used as functional genesets for AUCell scoring. AUCell scores of genesets were visualized using sc.pl.embedding. P- values from t tests were used for estimating the statistical significance between cell types and groups. + +Statistical analysis + +The experiments were conducted in at least three independent biological replicates, and the + +<--- Page Split ---> + +data were presented as mean \(\pm\) SD. If not specified, the Student's t test was used to perform a statistical significance test between different groups, and \(\mathrm{P}< 0.05\) was considered significant. Overall survival curves were estimated by the Kaplan- Meier method and Cox proportional hazards model. All statistical and correlation analyses were performed using the GraphPad Prism 8.0 software (GraphPad Software) and SPSS 26.0 (SPSS Software). + +## Data and materials availability + +All data and materials will be made available after acceptance of the manuscript. + +## Acknowledgements + +We gratefully acknowledge Drs. Daan van Aalten, Kum Kum Khanna, Xiaowei Yang, Timothy Mitchison, Xuebiao Yao, Chao Xu, Cuiting Yong, Lisha Wu, Wenqing Yang, and Hongqiang Qin for reagents or inspiring discussions. This project has been supported by the National Natural Science Foundation of China (grants 92153301, 32170821, and 32370821 to K.Y, 32101034 to F.C), National Key Research and Development Program of China (2021YFC2701200), Department of Science & Technology of Hunan Province (grants 2023RC1028, 2023SK2091, and 2021JJ10054 to K.Y). + +## Contributions + +Conceptualization: K.Y.; Methodology: N.Z., Y.M., H.N., C.H., L.S., L.L., C.Y., S.M., F.C., Y.Z., K.Y.; Validation: H.N., C.H., L.S., K.F.; Software: N.Z., Y.M., S.M.; Formal Analysis: N.Z., Y.M., K.Y.; Investigation: N.Z., Y.M., H.N., S.M., K.Y.; Resources: Y.Z., K.Y.; Data Curation: N.Z., Y.M., H.N.; Writing- Original Draft: N.Z., K.Y.; Writing- Review & Editing: K.Y.; Visualization: N.Z., Y.M., S.M., K.Y.; Supervision: K.Y.; Project Administration: L.L., K.Y.; Funding Acquisition: Y.Z., K.Y. + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> + +Yang, X. & Qian, K. Protein O-GlcNAcylation: emerging mechanisms and functions. Nat Rev Mol Cell Biol 18, 452- 465, doi:10.1038/nrm.2017.22 (2017).Zachara, N. E. & Hart, G. W. O-GlcNAc a sensor of cellular state: the role of nucleocytoplasmic glycosylation in modulating cellular function in response to nutrition and stress. Biochim Biophys Acta 1673, 13- 28, doi:10.1016/j.bbagen.2004.03.016 (2004).Chatham, J. C., Zhang, J. & Wende, A. R. Role of O- Linked N- Acetylglucosamine Protein Modification in Cellular (Patho)Physiology. Physiol Rev 101, 427- 493, doi:10.1152/physrev.00043.2019 (2021).Sheikh, M. A., Emerald, B. S. & Ansari, S. A. Stem cell fate determination through protein O-GlcNAcylation. J Biol Chem 296, 100035, doi:10.1074/jbc.REV120.014915 (2021).Slawson, C. & Hart, G. W. O- GlcNAc signalling: implications for cancer cell biology. Nat Rev Cancer 11, 678- 684, doi:10.1038/nrc3114 (2011).de Queiroz, R. M., Carvalho, E. & Dias, W. B. O- GlcNAcylation: The Sweet Side of the Cancer. Front Oncol 4, 132, doi:10.3389/fonc.2014.00132 (2014).Trinca, G. M. & Hagan, C. R. O- GlcNAcylation in women's cancers: breast, endometrial and ovarian. J Bioenerg Biomembr 50, 199- 204, doi:10.1007/s10863- 017- 9730- z (2018).Le Minh, G., Esquea, E. M., Young, R. G., Huang, J. & Reginato, M. J. On a sugar high: Role of O- GlcNAcylation in cancer. J Biol Chem 299, 105344, doi:10.1016/j.jbc.2023.105344 (2023).Yang, Y., Yin, X., Yang, H. & Xu, Y. Histone demethylase LSD2 acts as an E3 ubiquitin ligase and inhibits cancer cell growth through promoting proteasomal degradation of OGT. Mol Cell 58, 47- 59, doi:10.1016/j.molcel.2015.01.038 (2015).Seo, H. G. et al. Mutual regulation between OGT and XIAP to control colon cancer cell growth and invasion. Cell Death Dis 11, 815, doi:10.1038/s41419- 020- 02999- 5 (2020).Peng, K. et al. Regulation of O- Linked N- Acetyl Glucosamine Transferase (OGT) through E6 Stimulation of the Ubiquitin Ligase Activity of E6AP. Int J Mol Sci 22, doi:10.3390/ijms221910286 (2021).Tang, J. et al. Targeting USP8 Inhibits O- GlcNAcylation of SLC7A11 to Promote Ferroptosis of Hepatocellular Carcinoma via Stabilization of OGT. Adv Sci (Weinh) 10, e2302953, doi:10.1002/advs.202302953 (2023).Tang, J. et al. The deubiquitinase EIF3H promotes hepatocellular carcinoma progression by stabilizing OGT and inhibiting ferroptosis. Cell Commun Signal 21, 198, doi:10.1186/s12964- 023- 01220- 2 (2023).Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71, 209- 249, doi:10.3322/caac.21660 (2021).Han, B. et al. Cancer incidence and mortality in China, 2022. Journal of the National Cancer Center, doi:10.1016/j.jncc.2024.01.006 (2024). + +<--- Page Split ---> + +768 16 Bokhman, J. V. Two pathogenetic types of endometrial carcinoma. Gynecol Oncol 15, 10- 17, doi:10.1016/0090- 8258(83)90111- 7 (1983). 769 17 Murali, R., Soslow, R. A. & Weigelt, B. Classification of endometrial carcinoma: more than two types. The Lancet Oncology 15, e268- e278, doi:10.1016/s1470- 2045(13)70591- 6 (2014). 772 18 Cancer Genome Atlas Research, N. et al. Integrated genomic characterization of endometrial carcinoma. Nature 497, 67- 73, doi:10.1038/nature12113 (2013). 774 19 Jamaluddin, M. F. B. et al. Proteomic and functional characterization of intra- tumor heterogeneity in human endometrial cancer. Cell Rep Med 3, 100738, doi:10.1016/j.xcrm.2022.100738 (2022). 778 20 Dou, Y. et al. Proteogenomic insights suggest druggable pathways in endometrial carcinoma. Cancer Cell 41, 1586- 1605 e1515, doi:10.1016/j.ccell.2023.07.007 (2023). 781 21 Jaskiewicz, N. M. & Townson, D. H. Hyper- O- GlcNAcylation promotes epithelial- mesenchymal transition in endometrial cancer cells. Oncotarget 10, 2899- 2910, doi:10.18632/oncotarget.26884 (2019). 784 22 Krzeslak, A., Wojcik- Krowiranda, K., Forma, E., Bienkiewicz, A. & Brys, M. Expression of genes encoding for enzymes associated with O- GlcNAcylation in endometrial carcinomas: clinicopathologic correlations. Ginekol Pol 83, 22- 26 (2012). 788 23 Zhai, L. et al. O- GlcNAcylation mediates endometrial cancer progression by regulating the Hippo- YAP pathway. Int J Oncol 63, doi:10.3892/ijo.2023.5538 (2023). 791 24 Zhou, F. et al. Elevated glucose levels impair the WNT/beta- catenin pathway via the activation of the hexosamine biosynthesis pathway in endometrial cancer. J Steroid Biochem Mol Biol 159, 19- 25, doi:10.1016/j.jsbmb.2016.02.015 (2016). 793 25 Ciesielski, P., Jozwiak, P., Forma, E. & Krzeslak, A. TET3- and OGT- Dependent Expression of Genes Involved in Epithelial- Mesenchymal Transition in Endometrial Cancer. Int J Mol Sci 22, doi:10.3390/ijms222413239 (2021). 796 26 Brooks, R. A. et al. Current recommendations and recent progress in endometrial cancer. CA Cancer J Clin 69, 258- 279, doi:10.3322/caac.21561 (2019). 798 27 Turco, M. Y. et al. Long- term, hormone- responsive organoid cultures of human endometrium in a chemically defined medium. Nat Cell Biol 19, 568- 577, doi:10.1038/ncb3516 (2017). 802 28 Fitzgerald, H. C., Dhakal, P., Behura, S. K., Schust, D. J. & Spencer, T. E. Self- renewing endometrial epithelial organoids of the human uterus. Proc Natl Acad Sci U S A 116, 23132- 23142, doi:10.1073/pnas.1915389116 (2019). 805 29 Garcia- Alonso, L. et al. Mapping the temporal and spatial dynamics of the human endometrium in vivo and in vitro. Nat Genet 53, 1698- 1711, doi:10.1038/s41588- 021- 00972- 2 (2021). 808 30 Li, W. et al. MAGeCK enables robust identification of essential genes from genome- scale CRISPR/Cas9 knockout screens. Genome Biol 15, 554, doi:10.1186/s13059- 014- 0554- 4 (2014). + +<--- Page Split ---> + +811 31 Kawauchi, K., Araki, K., Tobiume, K. & Tanaka, N. Loss of p53 enhances catalytic activity of IKKbeta through O-linked beta-N-acetyl glucosamine modification. Proc Natl Acad Sci U S A 106, 3431- 3436, doi:10.1073/pnas.0813210106 (2009). 814 32 Zhang, F., Snead, C. M. & Catravas, J. D. Hsp90 regulates O-linked beta-N-acetylglucosamine transferase: a novel mechanism of modulation of protein O-linked beta-N-acetylglucosamine modification in endothelial cells. Am J Physiol Cell Physiol 302, C1786- 1796, doi:10.1152/ajpcell.00004.2012 (2012). 818 33 Deplus, R. et al. TET2 and TET3 regulate GlcNAcylation and H3K4 methylation through OGT and SET1/COMPASS. EMBO J 32, 645- 655, doi:10.1038/emboj.2012.357 (2013). 821 34 Ledee, D. et al. c- Myc Alters Substrate Utilization and O-GlcNAc Protein Posttranslational Modifications without Altering Cardiac Function during Early Aortic Constriction. PLoS One 10, e0135262, doi:10.1371/journal.pone.0135262 (2015). 825 35 Li, Y. N., Hu, J. A. & Wang, H. M. Inhibition of HIF- 1alpha Affects Autophagy Mediated Glycosylation in Oral Squamous Cell Carcinoma Cells. Dis Markers 2015, 239479, doi:10.1155/2015/239479 (2015). 828 36 Muthusamy, S., Hong, K. U., Dassanayaka, S., Hamid, T. & Jones, S. P. E2F1 Transcription Factor Regulates O-linked N-acetylglucosamine (O-GlcNAc) Transferase and O-GlcNAcase Expression. J Biol Chem 290, 31013- 31024, doi:10.1074/jbc.M115.677534 (2015). 832 37 Sodi, V. L. et al. mTOR/MYC Axis Regulates O-GlcNAc Transferase Expression and O-GlcNAcylation in Breast Cancer. Mol Cancer Res 13, 923- 933, doi:10.1158/1541- 7786.MCR- 14- 0536 (2015). 835 38 Zhang, X. et al. MAPK/ERK signaling pathway- induced hyper- O- GlcNAcylation enhances cancer malignancy. Mol Cell Biochem 410, 101- 110, doi:10.1007/s11010- 015- 2542- 8 (2015). 838 39 Li, T. et al. Defective Branched- Chain Amino Acid Catabolism Disrupts Glucose Metabolism and Sensitizes the Heart to Ischemia- Reperfusion Injury. Cell Metab 25, 374- 385, doi:10.1016/j.cmet.2016.11.005 (2017). 841 40 Zibrova, D. et al. GFAT1 phosphorylation by AMPK promotes VEGF- induced angiogenesis. Biochem J 474, 983- 1001, doi:10.1042/BCJ20160980 (2017). 843 41 Berthier, A. et al. Combinatorial regulation of hepatic cytoplasmic signaling and nuclear transcriptional events by the OGT/REV- ERBalpha complex. Proc Natl Acad Sci U S A 115, E11033- E11042, doi:10.1073/pnas.1805397115 (2018). 846 42 Lai, C. Y. et al. Identification of UAP1L1 as a critical factor for protein O- GlcNAcylation and cell proliferation in human hepatoma cells. Oncogene 38, 317- 331, doi:10.1038/s41388- 018- 0442- 6 (2019). 849 43 Deng, X. et al. ROCK2 mediates osteosarcoma progression and TRAIL resistance by modulating O- GlcNAc transferase degradation. Am J Cancer Res 10, 781- 798 (2020). 852 44 Lu, S. et al. SIRT1 regulates O- GlcNAcylation of tau through OGT. Aging (Albany NY) 12, 7042- 7055, doi:10.18632/aging.103062 (2020). + +<--- Page Split ---> + +854 45 Paredes, F., Williams, H. C., Quintana, R. A. & San Martin, A. Mitochondrial Protein 855 Poldip2 (Polymerase Delta Interacting Protein 2) Controls Vascular Smooth Muscle 856 Differentiated Phenotype by O-Linked GlcNAc (N-Acetylglucosamine) 857 Transferase-Dependent Inhibition of a Ubiquitin Proteasome System. Circ Res 126, 858 41- 56, doi:10.1161/CIRCRESAHA.119.315932 (2020). 859 46 Walter, L. A. et al. Inhibiting the Hexosamine Biosynthetic Pathway Lowers 860 O-GlcNAcylation Levels and Sensitizes Cancer to Environmental Stress. 861 Biochemistry 59, 3169- 3179, doi:10.1021/acs.biochem.9b00560 (2020). 862 47 Zhao, M., Sun, J. & Zhao, Z. TSGene: a web resource for tumor suppressor genes. 863 Nucleic Acids Res 41, D970- 976, doi:10.1093/nar/gks937 (2013). 864 48 Zhao, M., Kim, P., Mitra, R., Zhao, J. & Zhao, Z. TSGene 2.0: an updated 865 literature- based knowledgebase for tumor suppressor genes. Nucleic Acids Res 44, 866 D1023- 1031, doi:10.1093/nar/gkv1268 (2016). 867 49 Santra, M. K., Wajapeyee, N. & Green, M. R. F- box protein FBXO31 mediates 868 cyclin D1 degradation to induce G1 arrest after DNA damage. Nature 459, 722- 725, 869 doi:10.1038/nature08011 (2009). 870 50 Johansson, P. et al. SCF- FBXO31 E3 ligase targets DNA replication factor Cdt1 for 871 proteolysis in the G2 phase of cell cycle to prevent re- replication. J Biol Chem 289, 872 18514- 18525, doi:10.1074/jbc.M114.559930 (2014). 873 51 Liu, J. et al. F- box only protein 31 (FBXO31) negatively regulates p38 874 mitogen- activated protein kinase (MAPK) signaling by mediating lysine 48- linked 875 ubiquitination and degradation of mitogen- activated protein kinase kinase 6 (MKK6). 876 J Biol Chem 289, 21508- 21518, doi:10.1074/jbc.M114.560342 (2014). 877 52 Malonia, S. K., Dutta, P., Santra, M. K. & Green, M. R. F- box protein FBXO31 878 directs degradation of MDM2 to facilitate p53- mediated growth arrest following 879 genotoxic stress. Proc Natl Acad Sci U S A 112, 8632- 8637, 880 doi:10.1073/pnas.1510929112 (2015). 881 53 Jeffery, J. M. et al. FBXO31 protects against genomic instability by capping FOXM1 882 levels at the G2/M transition. Oncogene 36, 1012- 1022, doi:10.1038/onc.2016.268 883 (2017). 884 54 Duan, S. et al. Loss of FBXO31- mediated degradation of DUSP6 dysregulates ERK 885 and PI3K- AKT signaling and promotes prostate tumorigenesis. Cell Rep 37, 109870, 886 doi:10.1016/j.celrep.2021.109870 (2021). 887 55 Tan, Y., Liu, D., Gong, J., Liu, J. & Huo, J. The role of F- box only protein 31 in 888 cancer. Oncol Lett 15, 4047- 4052, doi:10.3892/ol.2018.7816 (2018). 889 56 Tekcham, D. S. et al. F- box proteins and cancer: an update from functional and 890 regulatory mechanism to therapeutic clinical prospects. Theranostics 10, 4150- 4167, 891 doi:10.7150/thno.42735 (2020). 892 57 Dutta, P. et al. The tumor suppressor FBXO31 preserves genomic integrity by 893 regulating DNA replication and segregation through precise control of cyclin A 894 levels. J Biol Chem 294, 14879- 14895, doi:10.1074/jbc.RA118.007055 (2019). 895 58 Islam, S. et al. Feedback- regulated transcriptional repression of FBXO31 by c- Myc 896 triggers ovarian cancer tumorigenesis. Int J Cancer 150, 1512- 1524, 897 doi:10.1002/ijc.33854 (2022). + +<--- Page Split ---> + +Baek, D. et al. Ubiquitin-specific protease 53 promotes osteogenic differentiation of human bone marrow-derived mesenchymal stem cells. Cell Death Dis 12, 238, doi:10.1038/s41419-021-03517-x (2021). +Zou, S. et al. FBXO31 Suppresses Gastric Cancer EMT by Targeting Snail1 for Proteasomal Degradation. Mol Cancer Res 16, 286-295, doi:10.1158/1541-7786.MCR-17-0432 (2018). +Manne, R. K. et al. A MicroRNA/Ubiquitin Ligase Feedback Loop Regulates Slug-Mediated Invasion in Breast Cancer. Neoplasia 19, 483-495, doi:10.1016/j.neo.2017.02.013 (2017). +Zhu, Z. et al. FBXO31 sensitizes cancer stem cells-like cells to cisplatin by promoting ferroptosis and facilitating proteasomal degradation of GPX4 in cholangiocarcinoma. Liver Int 42, 2871-2888, doi:10.1111/liv.15462 (2022). +Wulff-Fuentes, E. et al. The human O-GlcNAcome database and meta-analysis. Sci Data 8, 25, doi:10.1038/s41597-021-00810-4 (2021). +Yang, W. H. et al. Modification of p53 with O-linked N-acetylglucosamine regulates p53 activity and stability. Nat Cell Biol 8, 1074-1083, doi:10.1038/ncb1470 (2006). +Chou, T. Y., Hart, G. W. & Dang, C. V. c-Myc is glycosylated at threonine 58, a known phosphorylation site and a mutational hot spot in lymphomas. J Biol Chem 270, 18961-18965, doi:10.1074/jbc.270.32.18961 (1995). +Zhu, Y. & Hart, G. W. Dual-specificity RNA aptamers enable manipulation of target-specific O-GlcNAcylation and unveil functions of O-GlcNAc on beta-catenin. Cell 186, 428-445 e427, doi:10.1016/j.cell.2022.12.016 (2023). +Jang, H. et al. O-GlcNAc Regulates Pluripotency and Reprogramming by Directly Acting on Core Components of the Pluripotency Network. Cell Stem Cell 11, 62-74, doi:10.1016/j.stem.2012.03.001 (2012). +Kim, D. K. et al. O-GlcNAcylation of Sox2 at threonine 258 regulates the self-renewal and early cell fate of embryonic stem cells. Experimental & Molecular Medicine 53, 1759-1768, doi:10.1038/s12276-021-00707-7 (2021). +Sun, C. et al. Glucose regulates tissue-specific chondro-osteogenic differentiation of human cartilage endplate stem cells via O-GlcNAcylation of Sox9 and Runx2. Stem Cell Res Ther 10, 357, doi:10.1186/s13287-019-1440-5 (2019). +Boretto, M. et al. Patient-derived organoids from endometrial disease capture clinical heterogeneity and are amenable to drug screening. Nat Cell Biol 21, 1041-1051, doi:10.1038/s41556-019-0360-z (2019). +Liu, H. et al. Transcriptional pausing induced by ionizing radiation enables the acquisition of radioresistance in nasopharyngeal carcinoma. J Mol Cell Biol, doi:10.1093/jmcb/mjad044 (2023). +Wu, N. et al. O-GlcNAcylation promotes colorectal cancer progression by regulating protein stability and potential carcinogenic function of DDX5. J Cell Mol Med 23, 1354-1362, doi:10.1111/jcmm.14038 (2019). +Dekkers, J. F. et al. High-resolution 3D imaging of fixed and cleared organoids. Nat Protoc 14, 1756-1771, doi:10.1038/s41596-019-0160-8 (2019). +Choo, Y. S. & Zhang, Z. Detection of protein ubiquitination. J Vis Exp, doi:10.3791/1293 (2009). + +<--- Page Split ---> + +942 75 Gundogdu, M. et al. The O-GlcNAc Transferase Intellectual Disability Mutation 943 L254F Distorts the TPR Helix. Cell Chem Biol 25, 513-518 e514, 944 doi:10.1016/j.chembiol.2018.03.004 (2018). 945 76 Sanjana, N. E., Shalem, O. & Zhang, F. Improved vectors and genome-wide libraries 946 for CRISPR screening. Nat Methods 11, 783-784, doi:10.1038/nmeth.3047 (2014). 947 77 Shalem, O. et al. Genome-scale CRISPR-Cas9 knockout screening in human cells. 948 Science 343, 84-87, doi:10.1126/science.1247005 (2014). 949 78 Liu, J. et al. An Integrated TCGA Pan-Cancer Clinical Data Resource to Drive 950 High-Quality Survival Outcome Analytics. Cell 173, 400-416 e411, 951 doi:10.1016/j.cell.2018.02.052 (2018). 952 79 Yu, C. et al. ARID1A loss derepresses a group of human endogenous retrovirus-H 953 loci to modulate BRD4-dependent transcription. Nat Commun 13, 3501, 954 doi:10.1038/s41467-022-31197-4 (2022). 955 80 Dura, B. et al. scFTD-seq: freeze-thaw lysis based, portable approach toward highly 956 distributed single-cell 3' mRNA profiling. Nucleic Acids Res 47, e16, 957 doi:10.1093/nar/gky1173 (2019). 958 81 Lai, Z. Z. et al. Single-cell transcriptome profiling of the human endometrium of 959 patients with recurrent implantation failure. Theranostics 12, 6527-6547, 960 doi:10.7150/thno.74053 (2022). 961 82 Tan, Y. et al. Single-cell analysis of endometriosis reveals a coordinated 962 transcriptional programme driving immunotolerance and angiogenesis across eutopic 963 and ectopic tissues. Nat Cell Biol 24, 1306-1318, doi:10.1038/s41556-022-00961-5 964 (2022). 965 83 Fonseca, M. A. S. et al. Single-cell transcriptomic analysis of endometriosis. Nat 966 Genet 55, 255-267, doi:10.1038/s41588-022-01254-1 (2023). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig.1: O-GlcNAc level correlates with endometrial cancer grading
+ +
Fig. 1 Correlative analysis of O-GlcNAc level with clinical parameters.
+ +(a) A flowchart illustrating the process of clinical sample selection, data collection, and + +<--- Page Split ---> + +analysis. All samples were derived from patients receiving their initial treatment, and none of the patients had concurrent or previous tumors. + +(b- c) Representative images depicting IHC staining of \(O\) - GlcNAcylation (RL2) and OGT in EC and adjacent normal tissues on the FFPE tissue array. Scale bars: \(50 \mu \mathrm{m}\) . + +(d- e) Quantitative analysis of the levels of \(O\) - GlcNAcylation (RL2) and OGT in the EC tissue arrays. The levels of \(O\) - GlcNAcylation and OGT were assessed semi- quantitatively based on both the intensity and area of the stainings. Statistical significance was calculated using unpaired two- tailed Student's t- test, \(* \mathrm{P} < 0.05\) , \(** \mathrm{P} < 0.01\) . + +(f) Representative images of IHC staining showing varying levels of \(O\) - GlcNAcylation in serial sections of EC tissues with different histologic grades (well differentiated G1, moderately differentiated G2, and poorly differentiated G3). Scale bar: \(50 \mu \mathrm{m}\) . + +(g) Percentage of samples with high or low levels of \(O\) - GlcNAcylation in different histologic grade groups. High and low categories were determined using a scoring system (high score: 8-12; low score: 0-6). Statistical significance between groups was calculated using Fisher's exact test, \(*** \mathrm{P} < 0.0001\) . + +(h- i) Kaplan- Meier survival curves of PFS and OS of the EC patients stratified by the levels of \(O\) - GlcNAcylation derived from IHC scores. Statistical significance was determined by the log- rank test. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
Fig.S1: Construction of a gene signature for O-GlcNAc level
+ +Fig. S1 Construction of a gene signature for O-GlcNAc level. + +(a) Representative images depicting IHC staining of OGA in EC and adjacent normal tissues on the FFPE tissue array. Scale bar: \(50 \mu \mathrm{m}\). + +(b) Quantitative analysis of the level of OGA in EC tissue array. The expression level of + +OGA was assessed semi-quantitatively based on both the staining intensity and area. + +<--- Page Split ---> + +996 Statistical significance was calculated using unpaired two- tailed Student's t- test, ns: not 997 significant. 998 (c- d) Kaplan- Meier analysis of the OS of EC patients based on OGA or OGT expression 999 levels in TCGA (http://kmplot.com/analysis/). EC cases were stratified using the median 1000 cut- off, and statistical significance was determined using the log- rank test. 1001 (e) Flow chart of sample selection and data process for constructing an O- GlcNAc level 1002 prediction model based on transcriptomic data. 1003 (f) Heatmap showing the expression levels of the 1000 O- GlcNAc correlated genes derived 1004 from RNA- seq of the selected EC tissues (n = 55). 1005 (g) Gene Ontology (GO) enrichment analysis of the 1000 O- GlcNAc correlated genes. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig.2: Validation with TCGA endometrial cancer dataset
+ +
Fig. 2 Validation with TCGA endometrial cancer dataset.
+ +(a) Heatmap displaying the expression profiles of the 1000 O-GlcNAc correlated genes in + +<--- Page Split ---> + +1009 the TCGA UCEC RNA- seq dataset (n = 589). The EC samples are annotated by clinical parameters, including BMI, menopause status, diabetes, histologic grades, molecular subtypes (integrative cluster), FIGO stage, age, and primary diagnosis. Patients were categorized into O- GlcNAc high or O- GlcNAc low group using the median of the calculated O- GlcNAc index as the threshold. The '*' symbol indicates a statistically significant difference of the calculated O- GlcNAc index among the patients' groups according to the indicated clinical parameter. Wilcoxon test, \(**P < 0.01\) , \(****P < 0.0001\) .(b- f) The O- GlcNAc index in different EC groups stratified by histologic grade, FIGO stage, integrative cluster, age, or diabetes in the TCGA UCEC dataset. Wilcoxon test, \(*P < 0.05\) , \(**P < 0.01\) , \(***P < 0.001\) , \(****P < 0.0001\) .(g- h) Kaplan- Meier survival curves for Progression- free interval (PFI) and OS of EC groups with high or low O- GlcNAc index in the TCGA UCEC dataset. Statistical significance was determined by the log- rank test. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig.3: Increase of O-GlcNAc level promotes proliferation and stemness
+ +(a) Mucin detection (Alcian blue staining) and IHC examination of endometrial markers (ER and PR) in primary endometrial tissue and corresponding EE-Os. Scale bars: \(50 \mu \mathrm{m}\). +(b) Immunoblot with RL2 antibody assessing O-GlcNAc levels in the EE-Os treated with + +<--- Page Split ---> + +DMSO, \(5 \mu \mathrm{M}\) , or \(10 \mu \mathrm{M}\) Thiamet- G (TMG) for 48 hours. Tubulin was used as the loading control. + +(c) EE-Os bright-field images depicting responses to TMG at two different time points (day 1 and day 3). Representative images from control (DMSO) and \(10 \mu \mathrm{M}\) TMG treated EE-Os groups are presented. Scale bar: \(50 \mu \mathrm{m}\). + +(d) Comparison of the numbers of EE-Os at day 3 of culture after treatment with \(10 \mu \mathrm{M}\) TMG versus control (DMSO). The results are presented as the mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +(e) Measurement of cross-sectional area of EE-Os at day 3 of culture after treatment with \(10 \mu \mathrm{M}\) TMG compared with control (DMSO). The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +(f) Representative immunofluorescence images of control and TMG-treated EE-Os. EE-O cells are stained with PH3 (red) and Tubulin (green) antibodies. Nuclei are visualized with DAPI (blue), and F-actin is labeled by Phalloidine (magenta). Scale bar: \(5 \mu \mathrm{m}\) . + +(g) Quantification of the number of PH3 positive (PH3+) cells in each EE-Os. The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{*P < 0.05}\) . + +(h) Representative immunofluorescence images of control and TMG-treated EE-Os. Ciliated epithelium is labeled by acetylated alpha-tubulin (Ac-tubulin, green), and secretory cells labeled by PAEP (red). Nuclei are visualized with DAPI (blue), and F-actin is labeled by Phalloidine (magenta). Scale bar: \(50 \mu \mathrm{m}\) . The right insets display a magnification of the area in the white box, scale bar: \(5 \mu \mathrm{m}\) . + +(i) Quantification of the number of ciliated cells (Ac-tubulin+) in each EE-Os. The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +(j) MDE projections of scRNA-seq data of control and TMG treated EE-Os. + +(k) Subclustered epithelial populations of EE-Os (left), and the proportion of each cell types in control and TMG-treated groups (right). + +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
Fig.S2: Generation of EE-O and EC-O, and TMG treatment on EE-O
+ +<--- Page Split ---> + +(a) Bright-field and fluorescent images of 3D EE-Os and EC-Os. F-actin is labeled by + +Phalloidine (golden yellow). Nuclei are labeled using DAPI (cyan). Scale bars: \(50 \mu \mathrm{m}\) . + +(b) Representative pictures of IHC staining for \(O\) -GlcNAcylation levels in primary tissues and their corresponding organoids. Scale bar: \(50 \mu \mathrm{m}\) . + +(c) qPCR data of stemness markers expression in EE-O treated with TMG or DMSO (control), normalized to actin mRNA levels. The results are presented as mean \(\pm\) SD from three independent experiments. Statistical significance was determined by unpaired two-tailed Student's t-test, \(^{*}\mathrm{P}< 0.05\) , \(^{**} \mathrm{P}< 0.01\) . + +(d) Feature plots of representative gene expressions in different epithelial subclusters resolved by scRNA-seq analysis. + +(e) Visualization of the indicated KEGG pathway on the MDE plot. + +(f) MDE plot revealing 6 different subclusters of epithelial cells and their cell numbers in the EE-Os subject to scRNA-seq analysis. + +(g) Expression levels of differentially expressed genes in the 6 subclusters of epithelial cells. + +The size and color of each dot represent the expression level and cell fraction of the marker + +genes. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig.4: Decrease of O-GlcNAc level induces differentiation and cell death
+ +(a) Mucin detection (Alcian blue staining) and IHC examination of EC markers in primary + +<--- Page Split ---> + +EC tissues and the corresponding EC- Os. Scale bar: \(50\mu \mathrm{m}\) .(b) Immunoblot assessing \(O\) - GlcNAc levels in the EC- Os treated with 25 or \(50\mu \mathrm{M}\) OSMI- 1 for 48 hours. Actin was used as the loading control.(c) Representative 3D EC- O bright- field images depicting the responses to \(50\mu \mathrm{M}\) OSMI- 1 treatment at two different time points (day 1 and day 3). Scale bar: \(50\mu \mathrm{m}\) .(d) Comparison of the numbers of EC- Os at day 3 of culture after treatment with OSMI- 1 versus the control (DMSO). The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.001}\) .(e) Analysis of the EC- Os cross- sectional area at day 3 of culture after treatment with OSMI- 1 compared with control (DMSO). The results are presented as mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.0001}\) .(f) TUNEL staining demonstrating the existence of apoptotic cells in EC- Os after treated with \(50\mu \mathrm{M}\) OSMI- 1. Nuclei are visualized with DAPI (blue). Scale bar: \(50\mu \mathrm{m}\) .(g) Representative immunofluorescence images of control and OSMI- 1 treated EC- Os. Mitotic cells are marked by PH3 (red). Tubulin staining is shown in green, DAPI labeled nuclei in blue, and Phalloidine labeled F- actin in magenta. Scale bar: \(5\mu \mathrm{m}\) .(h) Quantification of the number of PH3+ cells. Each dot represents one 3D EC- O. The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.0001}\) .(i) Representative immunofluorescence images of control and OSMI- 1 treated EC- Os. Ciliated epithelial cells are labeled by acetylated alpha- tubulin (Ac- tubulin, green), and secretory cells are labeled by PAEP (red). Nuclei are visualized with DAPI (blue), and F- actin with Phalloidine (magenta). Scale bar: \(50\mu \mathrm{m}\) . The right insets display a magnification of the area in the white box, scale bar: \(5\mu \mathrm{m}\) .(j) Quantification of the number of Ac- tubulin+ cells. Each dot represents one 3D EC- O. The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.01}\) .(k) qPCR data of stemness markers expression in EC- Os treated with OSMI- 1 or DMSO (control), normalized to actin mRNA levels. The results are presented as mean \(\pm\) SD from three independent experiments. Statistical significance was determined by unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.01}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{***P< 0.0001}\) . + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig.5: Screen for TSGs that regulate O-GlcNAc homeostasis
+ +(a) Schematic representation of the FACS-based genome-wide CRISPR-Cas9 screen for putative regulators of O-GlcNAc homeostasis. + +<--- Page Split ---> + +1110 (b) Validation the sensitivity of RL2 staining (red) with 293T cells transfected with OGT (green). Nuclei are labeled with DAPI (blue). Scale bar: \(5 \mu \mathrm{m}\) .1112 (c) Genes plotted according to their relative ranking analysis (RRA) enrichment scores, with known O- GlcNAc regulators highlighted in red and blue.1114 (d) KEGG analysis showing enrichment of putative O- GlcNAc regulators in the indicated pathways. Analysis was performed on the 1038 top scoring genes ( \(P < 0.05\) ).1116 (e) Venn diagram showing the overlap between the 526 Human UCEC TSGs and the 1038 high- confidence genes from the O- GlcNAc screen.1118 (f) Immunofluorescent detection of O- GlcNAc level by RL2 (red) in WT and FBXO31 KO 293T cells. Nuclei were stained with DAPI (blue). Scale bar: \(5 \mu \mathrm{m}\) .1120 (g) Kaplan- Meier analysis of the OS of the EC patients stratified by the expression levels of FBXO31 (http://kmlpot.com/analysis/). EC cases were stratified using the median cut- off, and statistical significance was determined using the log- rank test. + +<--- Page Split ---> +![](images/Figure_unknown_2.jpg) + +
Fig.S3: Survival analysis of potential O-GlcNAc regulators in UCEC
+ +<--- Page Split ---> + +1125 (a) Kaplan-Meier analyses were conducted on EC patients in the TCGA UCEC dataset to 1126 elucidate the relationship between the expression levels of the 17 overlapping genes and 1127 patients' OS over time. EC cases were stratified using the median expression level as cut- off, 1128 and statistical significance was determined using the log- rank test. 1129 (b) Validation the FBXO31 KO 293T cells using sanger sequencing. The KO cells harbor a 1130 17 bp deletion, which causes a frameshift mutation starting from the \(48^{\text{th}}\) amino acid of 1131 FBXO31, terminating prematurely at the \(106^{\text{th}}\) amino acid. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Fig.6: FBXO31 interacts with and ubiquitinates OGT
+ +<--- Page Split ---> + +(a) Immobilized recombinant GST-OGT protein but not GST control absorbed GFP-FBX031 from 293T cell lysates. GST and GST-OGT were detected by Coomassie brilliant blue (CBB) staining, and FBXO31 was detected by western blotting with FBXO31 antibody. + +(b) Co-immunoprecipitation of GFP-FBX031 with Flag-OGT in 293T cell lysates. The presence of MG132 enhanced the interaction between Flag-OGT and GFP-FBX031. + +(c) Western blotting assessing the protein level of OGT as well as the global O-GlcNAc (RL2) levels in 293T cells transfected with increasing amount of GFP-FBX031. + +(d) Western blotting detecting the protein level of endogenous OGT and its ubiquitination in 293T cells transfected with different amount of HA-Ub and GFP-FBX031. + +(e) In vitro ubiquitination of His-OGT by the SCF complex together with FBXO31. HA-tagged SCF components (Skp1, Cul1, and Roc1) and HA-FBX031 were affinity-purified using anti-HA-conjugated magnetic beads from HEK293T cell lysates. The purified protein complex was incubated with E1 (UBA1), E2 (UBE2D1), Ub, and His-OGT in ubiquitination buffer. The reaction was halted by the addition of SDS sample buffer, and the samples were subjected to western blotting using the indicated antibodies. + +(f) In vivo ubiquitination assay was performed to evaluate the ubiquitination levels of exogenous Flag-OGT in 293T cells transfected with HA-tagged Ub and GFP-FBX031 or its F-box domain deletion mutant GFP-FBX031ΔF. + +(g) Western blotting quantification of the protein level of endogenous OGT in 293T cells transfected with GFP-FBX031. MG132 was added to inhibit the ubiquitination-mediated proteasome degradation. Statistical significance was determined by unpaired two-tailed Student's t-test, n = 3, ***P < 0.001. + +(h) Western blotting detecting the O-GlcNAc (RL2) and OGT levels in WT and FBXO31 KO 293T cells. + +(i) Western blotting quantitation of OGT protein level following cycloheximide (CHX) treatment in WT and FBXO31 KO 293T cells. Statistical significance was determined by unpaired two-tailed Student's t-test, n = 3, *P < 0.05. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Fig.7: Loss of FBXO31 increases O-GlcNAc level in clinical samples
+ +<--- Page Split ---> + +(a) Representative images depicting IHC staining of FBXO31 in EC and adjacent normal tissues on an FFPE tissue array. Scale bar: \(50 \mu \mathrm{m}\) . + +(b) Quantitative analysis of the level of FBXO31 in the EC tissue array. The expression level of FBXO31 was assessed semi-quantitatively based on both staining intensity and area. Statistical significance was calculated using unpaired two-tailed Student's t-test, \(**\mathrm{P}< 0.01\) . +(c) Percentage of samples with high or low FBXO31 levels by IHC in the two different \(O\) -GlcNAcylation level groups (high-RL2 or low-RL2). High and low expression categories of FBXO31 were determined using a scoring system (high score: 8-12; low score: 0-6). + +(d) Protein levels of OGT and FBXO31 were assessed by Western blotting in EC-Os and EE-Os derived from different patients. + +(e) Immunofluorescence detection of \(O\) -GlcNAcylation (RL2, green) and FBXO31 (red) in control EE-Os and shFBXO31 infected EE-Os. The nuclei were stained with DAPI (blue), and F-actin was labeled by Phalloidine (magenta). Scale bar: \(50 \mu \mathrm{m}\) . + +(f) qPCR data of stemness markers expression in control shNT and shFBXO31 infected EE-Os, normalized to actin mRNA levels. Statistical significance was calculated using unpaired two-tailed Student's t-test, \(n = 3\) , ns: not significant, \(**\mathrm{P}< 0.05\) , \(**\mathrm{P}< 0.01\) . + +(g) Quantification of organoid colony-forming numbers of the control and shFBXO31 infected EE-Os in 3D culture. Representative bright-field images are provided on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(***\mathrm{P}< 0.001\) . + +(h) Quantification of organoid colony-forming numbers of shFBXO31 treated EE-Os at day 3 of culture after treatment with OSMI-1 compared with control (DMSO). Representative bright-field images are shown on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(***\mathrm{P}< 0.0001\) . + +(i) Quantification of organoid colony-forming numbers of EC-Os overexpressing GFP or GFP-FBXO31. Bright-field and fluorescent images of the treated EC-Os are shown on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(**\mathrm{P}< 0.01\) . + +<--- Page Split ---> +![](images/Figure_8.jpg) + +
Fig.8: Working model
+ +1192 + +1193 Fig. 8 Working model.1194 FBXO31- mediated ubiquitination of OGT maintains a relatively low level of1195 \(O\) - GlcNAcylation in the normal endometrium. Inactivation of FBXO31 in endometrial1196 cancer tissues results in accumulation of OGT and concurrent increase of \(O\) - GlcNAcylation. + +<--- Page Split ---> + + +Table 1. Association of O-GlcNAcylation expression with clinicopathological parameters of patients with EC + +
ParameterDescribeNO-GlcNAcylation levelP value
LowHigh
Age&lt; 6016589760.4305
≥ 60543321
Histologic gradeG1714922&lt; 0.0001
G21066343
G3421032
FIGO stageI13778590.0080
II332211
III412219
IV808
BMI&lt; 2816895730.8721
≥ 28512724
DiabetesNo196107890.3808
Yes23158
HypertensionNo15990690.6494
Yes603228
Distant metastasisNegative211122890.0013
Positive808
Lymph node metastasisNegative195112830.1910
Positive241014
Myometrial invasion&lt; 1/215393600.0654
≥ 1/2582632
Serosa835
+ +<--- Page Split ---> + +Table 2. Univariate and multivariate Cox regression analysis for PFS in EC patients + +
CharacteristicsNUnivariate analysisMultivariate analysis
HR(95% CI)P valueHR(95% CI)P value
Age
\(<60\)154
\(\geq 60\)503.437(1.104-10.694)0.0333.980(1.220-12.982)0.022
BMI
\(<28\)155
\(\geq 28\)490.635(0.139-2.902)0.558
Diabetes
No184
Yes200.922(0.119-7.164)0.938
FIGO stage
I+II157
III+IV473.327(1.073-10.319)0.0372.552(0.755-8.630)0.132
Histologic grade
Grade 167
Grade \(2+3\)1372.233(0.489-10.200)0.300
Lymph node
metastasis
Negative182
Positive223.067(0.828-11.353)0.093
Myometrial
invasion
\(<1/2\)142
\(\geq 1/2+S\)erosa624.848(1.459-16.108)0.0102.591(0.727-9.236)0.142
O-GlcNAcylation
level
Low115
High894.823(1.297-17.937)0.0194.611(1.212-17.536)0.025
+ +1199 HR: hazard ratio; CI: confidence interval. + +<--- Page Split ---> + + +Table 3. Association of FBXO31 expression with clinicopathological parameters of patients with EC + +
ParameterDescribeNFBXO31 level LowHighP value
Age&lt; 608848400.1022
≥ 60331221
Histologic gradeG140931&lt; 0.0001
G2512427
G330273
FIGO stageI8939500.0862
II1073
III1587
IV761
BMI&lt; 289044460.8372
≥ 28311615
DiabetesNo10858500.016
Yes13211
HypertensionNo8143380.2732
Yes401723
Distant metastasisNegative11454600.0614
Positive761
Lymph node metastasisNegative10750570.0956
Positive14104
O-GlcNAcylation levelLow38731&lt; 0.0001
High835330
Myometrial invasion&lt; 1/28039410.3559
≥ 1/2412120
Serosa220
+ +<--- Page Split ---> + +# Descriptions of additional Supplementary Files + +1201 Supplementary Table 1 1203 TPM of 1000 O- GlcNAc correlated genes from 55 EC tissues 1204 Supplementary Table 2 1206 GO analysis for 1000 O- GlcNAc correlated genes 1207 Supplementary Table 3 1209 TPM of 1000 O- GlcNAc correlated genes from TCGA UCEC dataset 1210 Supplementary Table 4 1212 Clinical data and predicted O- GlcNAc index in TCGA UCEC dataset 1213 Supplementary Table 5 1215 Cluster markers 1216 Supplementary Table 6 1218 RL2_pos_vs_RL2_ctrl genes summarized with MAGeCK- RRA 1219 Supplementary Table 7 1221 KEGG Pathway analysis on the 1038 top scoring genes 1222 Supplementary Table 8 1224 Primers for plasmid constructs, and RT- qPCR 1225 Supplementary Table 9 1227 Antibodies for immunostaining and WB + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +Supplementarytable1. xlsx Supplementarytable2. xlsx Supplementarytable3. xlsx Supplementarytable4. xlsx Supplementarytable5. xlsx Supplementarytable6. xlsx Supplementarytable7. xlsx Supplementarytable8. xlsx Supplementarytable9. xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9_det.mmd b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..41a6330132272a402c92c061d53bfcc6c182909c --- /dev/null +++ b/preprint/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9/preprint__1662d8c0ea64e5beea710276f2c5a587fea381216a0688b9e0dfc09265e509b9_det.mmd @@ -0,0 +1,752 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 920, 210]]<|/det|> +# Impairment of FBXO31-mediated Ubiquitination of OGT Upregulates O-GlcNAcylation to Advance Endometrial Malignancy + +<|ref|>text<|/ref|><|det|>[[44, 230, 123, 248]]<|/det|> +Kai Yuan + +<|ref|>text<|/ref|><|det|>[[52, 257, 256, 275]]<|/det|> +yuankai@csu.edu.cn + +<|ref|>text<|/ref|><|det|>[[44, 302, 907, 368]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. https://orcid.org/0000- 0001- 7002- 5703 + +<|ref|>text<|/ref|><|det|>[[44, 373, 135, 392]]<|/det|> +Na Zhang + +<|ref|>text<|/ref|><|det|>[[44, 395, 907, 438]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. + +<|ref|>text<|/ref|><|det|>[[44, 443, 145, 461]]<|/det|> +Yang Meng + +<|ref|>text<|/ref|><|det|>[[44, 464, 907, 506]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University Song Mao + +<|ref|>text<|/ref|><|det|>[[44, 509, 907, 553]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 557, 130, 575]]<|/det|> +Huiling Ni + +<|ref|>text<|/ref|><|det|>[[44, 579, 730, 620]]<|/det|> +Center for Medical Genetics, School of Life Sciences, Central South University Canhua Huang + +<|ref|>text<|/ref|><|det|>[[52, 624, 422, 644]]<|/det|> +Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 649, 156, 667]]<|/det|> +Licong Shen + +<|ref|>text<|/ref|><|det|>[[52, 671, 662, 691]]<|/det|> +Department of Gynecology, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 696, 106, 713]]<|/det|> +Kun Fu + +<|ref|>text<|/ref|><|det|>[[44, 717, 907, 760]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 765, 95, 782]]<|/det|> +Lu Lv + +<|ref|>text<|/ref|><|det|>[[44, 787, 907, 808]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 812, 163, 830]]<|/det|> +Chunhong Yu + +<|ref|>text<|/ref|><|det|>[[44, 833, 789, 854]]<|/det|> +Institute of Molecular Precision Medicine, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 859, 140, 876]]<|/det|> +Fang Chen + +<|ref|>text<|/ref|><|det|>[[44, 880, 907, 902]]<|/det|> +Hunan Key Laboratory of Molecular Precision Medicine, Xiangya Hospital, Central South University + +<|ref|>text<|/ref|><|det|>[[44, 906, 128, 924]]<|/det|> +Yu Zhang + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 106, 104, 124]]<|/det|> +## Article + +<|ref|>sub_title<|/ref|><|det|>[[44, 144, 136, 163]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 181, 300, 201]]<|/det|> +Posted Date: April 12th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 220, 474, 239]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4019799/v1 + +<|ref|>text<|/ref|><|det|>[[42, 257, 916, 300]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 317, 535, 338]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 372, 944, 416]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on February 2nd, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 56633- z. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[80, 95, 884, 570]]<|/det|> +Impairment of FBXO31- mediated Ubiquitination of OGT Upregulates O- GlcNAcylation to Advance Endometrial Malignancy Na Zhang1, Yang Meng2,3, Song Mao1, Huiling Ni1,2, Canhua Huang1, Licong Shen1, Kun Fu1, Lu Lv1, Chunhong Yu1, Fang Chen1, Yu Zhang1, Kai Yuan1,2,4,5,6,# 1. Hunan Key Laboratory of Molecular Precision Medicine, Department of Oncology & Department of Gynecology, Xiangya Hospital, Central South University, Changsha, Hunan, China. 2. Center for Medical Genetics, School of Life Sciences, Central South University, Changsha, Hunan, China. 3. School of Pharmaceutical Sciences, Tsinghua University, Beijing, China. 4. Furong Laboratory, Hunan, China. 5. National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, Hunan, China. 6. The Biobank of Xiangya Hospital, Central South University, Changsha, Hunan, China. # Correspondence: yuankai@csu.edu.cn (K.Y.) + +<|ref|>sub_title<|/ref|><|det|>[[128, 598, 207, 614]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[124, 622, 883, 894]]<|/det|> +Aberrant O- GlcNAc cycling of the cancer proteome is a manifestation of its metabolic plasticity. As one of the most common cancer of the female genital tract associated with metabolic syndrome, endometrial cancer (EC) tissues often bear altered O- GlcNAcylation patterns. However, integration of O- GlcNAc status with existing histomorphologic and molecular subtypes of EC in large cohorts and identification of molecular modules controlling the O- GlcNAc homeostasis remain to be accomplished. Here we establish a positive correlation of O- GlcNAcylation with histologic grade of EC in a Chinese cohort containing 219 tumors and consolidate it in The Cancer Genome Atlas (TCGA) EC dataset. Higher O- GlcNAc level is associated with less pathological differentiation and poorer prognosis. Functionally, increasing O- GlcNAcylation promotes proliferation and stem- like + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 882, 338]]<|/det|> +cell properties in normal endometrial epithelial organoids (EE- Os), whereas decreasing O- GlcNAcylation limits the growth of endometrial cancer organoids (EC- Os). Using genome- wide CRISPR screen, we further identify that the F- box only protein 31 (FBXO31), whose loss of heterozygosity is frequently observed in cancer, regulates O- GlcNAc homeostasis. FBXO31 acts as a substrate receptor of the SCF ubiquitin ligase complex to ubiquitinate the O- GlcNAc transferase OGT. Loss of FBXO31 results in accumulation of OGT and upregulation of O- GlcNAcylation in EC. Our study highlights the O- GlcNAcylation as a useful stratification marker and potential therapeutic target for the advanced, poorly differentiated EC cases. + +<|ref|>sub_title<|/ref|><|det|>[[127, 375, 240, 392]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[125, 400, 882, 896]]<|/det|> +Post- translational modifications (PTMs) endow the proteome with functional plasticity to cope with intrinsic and extrinsic perturbations under various developmental and disease conditions. Protein O- GlcNAcylation, catalyzed by a pair of evolutionarily conserved enzymes O- GlcNAc transferase (OGT) and O- GlcNAcase (OGA), is a PTM involving the covalent addition of single O- linked N- acetylglucosamine (O- GlcNAc) modifications to serine and threonine residues of intracellular proteins1. The GlcNAc moieties are supplied by a metabolite uridine diphosphate N- acetylglucosamine (UDP- GlcNAc), whose synthesis via the Hexosamine biosynthesis pathway (HBP) requires fructose- 6- P, acetyl- CoA, glutamine, and UTP, substrates from all major cellular metabolic pathways. As a result, O- GlcNAcylation is sensitive to nutrient availability and intrinsic metabolic reprogramming. Meanwhile, O- GlcNAc cycling is highly responsive to a wide range of extrinsic stimuli, including osmotic, oxidative, hyperthermic, and genotoxic stresses1-3, making it an important cellular stress sensing mechanism. O- GlcNAcylation is required for the maintenance of pluripotency of embryonic stem cells (ESCs), and its level declines as ESCs differentiate, alongside the cellular metabolic switch from glycolysis to oxidative phosphorylation4. Cancer cells often hijack embryonic programs to support their uncontrolled proliferation and cell fate transition, adopting a metabolic lifestyle relying on aerobic glycolysis (Warburg effect). Elevated O- GlcNAcylation has been observed in many + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 883, 255]]<|/det|> +cancer cell lines \(^{5 - 8}\) , probably as a result of its increased nutrient consumption, or imbalanced enzymatic activity of OGT and OGA due to somatic mutations or altered protein stability \(^{9 - 13}\) . To date, systemic assessment of \(O\) - GlcNAc status in major cancer cohorts and functional dissection of \(O\) - GlcNAc homeostasis in patient- derived organoids haven't been conducted, hindering the utilization of \(O\) - GlcNAcylation level to guide patient stratification and subsequent development of personalized interventions. + +<|ref|>text<|/ref|><|det|>[[125, 290, 883, 699]]<|/det|> +Endometrial cancer (EC), the incidence of which has increased over \(50\%\) during the past two decades, is the most common cancer within the female reproductive system in developed countries \(^{14}\) . In China, as of 2022, there were approximately 77,700 newly diagnosed EC cases and 13,500 estimated EC cancer deaths \(^{15}\) . EC comprises a panel of tumors that are clinically and biologically heterogeneous. It can be grouped into type I or type II tumors according to the clinical and endocrine features \(^{16}\) , or classified as endometrioid carcinoma, serous carcinoma, carcinosarcoma, or clear- cell carcinoma based on its histopathological characteristics \(^{17}\) . The Cancer Genome Atlas Research Network (TCGA) study of EC has established a more precise genomic classification including four molecular subtypes: \(POLE\) - mutated, microsatellite- instable (MSI), copy- number low, and copy- number high tumors \(^{18}\) . More recently, integration of proteomic analysis to the genomic classification has accelerated the identification of clinically actionable molecular targets in EC \(^{19,20}\) . However, PTMs, which add tremendous functional complexity to the proteome, remain to be comprehensively characterized in EC tumor samples and complemented into the current classification system. + +<|ref|>text<|/ref|><|det|>[[125, 736, 884, 893]]<|/det|> +\(O\) - GlcNAcylation as an important PTM responsive to cellular metabolism and stress has been linked to molecular etiology of EC. Both \(OGT\) and \(OGA\) manifested highest alterations, mainly gene mutation and amplification, in EC among major female cancer types \(^{21}\) . The mRNA levels of \(OGT\) and \(OGA\) were increased in EC tumor samples of higher histologic grade \(^{22}\) . Despite that \(O\) - GlcNAc homeostasis is maintained by delicate and complex feedback loops, elevated \(O\) - GlcNAcylation level in EC tissues was observed using a small + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 882, 255]]<|/det|> +tissue microarray \(^{23}\) . O- GlcNAcylation was reported to support proliferation and migration and promote epithelial- mesenchymal transition in cultured EC cell lines by regulating Wnt/ \(\beta\) - catenin and Hippo- YAP signaling pathways \(^{21,23 - 25}\) . These observations suggest that altered O- GlcNAcylation may contribute to EC progression, and it is worthy of thorough interrogation in large EC cohorts to determine whether O- GlcNAcylation can be utilized both as a new stratification factor and potential drugable target. + +<|ref|>text<|/ref|><|det|>[[125, 290, 882, 615]]<|/det|> +In this study, utilizing a Chinese EC cohort containing 219 tumors and the TCGA EC dataset, we uncovered that O- GlcNAcylation level correlates with histologic grade, International Federation of Gynecology and Obstetrics (FIGO) stage, and patients' prognosis. Moreover, we experimentally demonstrated that upregulation of O- GlcNAcylation promotes proliferation and stem- like cell properties in normal endometrial epithelial organoids (EE- Os), whereas downregulation of O- GlcNAcylation impedes the proliferation of endometrial cancer organoids (EC- Os). Furthermore, we identified FBXO31 as a key regulator of O- GlcNAcylation homeostasis, by controlling the ubiquitin- dependent protein degradation of OGT. Our findings highlight that O- GlcNAcylation is a useful factor complementary to the current classification system to better stratify EC patients, and targeting O- GlcNAcylation signaling is a promising differentiation therapeutic strategy worthy of clinical exploitation for high grade EC patients. + +<|ref|>sub_title<|/ref|><|det|>[[127, 652, 194, 668]]<|/det|> +## Results + +<|ref|>text<|/ref|><|det|>[[125, 678, 881, 723]]<|/det|> +Elevated O- GlcNAc level is positively correlated with histologic grade and poor prognosis in endometrial cancer + +<|ref|>text<|/ref|><|det|>[[125, 732, 882, 892]]<|/det|> +To get a glimpse of global O- GlcNAcylation level in EC tissues, we first obtained an EC tissue array from Xinchao Biotech (Shanghai, China) containing 23 normal and 31 tumorous endometrial specimens, and performed IHC analyses to examine O- GlcNAcylation as well as OGT and OGA levels (fig. 1a). The expression of OGT and the amount of O- GlcNAcylation were significantly higher in EC tissues relative to normal tissues (fig. 1b- 1c), consistent with a previous report \(^{23}\) . The expression of OGA however showed no + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 830, 115]]<|/det|> +significant difference between normal and tumorous endometrial tissues (fig. sla- slb). + +<|ref|>text<|/ref|><|det|>[[125, 150, 883, 702]]<|/det|> +To elaborate the relationship between the O- GlcNAcylation level and clinical characteristics of EC, we expanded the analyses to an EC cohort containing 219 tumor patients who received surgery in the Department of Gynecology, Xiangya Hospital, Central South University (fig. 1a). The paraffin- embedded EC tissue sections were subjected to IHC staining, and the O- GlcNAcylation level revealed by RL2 antibody staining for each specimen was semi- quantified to categorize the patients into O- GlcNAc high and O- GlcNAc low groups. This O- GlcNAc status exhibited significant association with histologic grade, FIGO stage, and distant metastasis of EC (table 1). Accordingly, the O- GlcNAcylation level manifested a significant increase in EC tissues from patients' groups with more advanced histologic grade (fig. 1f- 1g). Further statistical analysis established a positive correlation between the O- GlcNAcylation level and tumor histologic grade (Goodman- Kruskal gamma statistic \(\mathrm{P} \leq 0.0001\) ; 2- sided gamma- knife gamma \(= 0.473\) ), as well as distant metastasis (Goodman- Kruskal gamma statistic \(\mathrm{P} = 0.003\) ; 2- sided gamma- knife gamma \(= 1\) ). Kaplan- Meier analysis indicated that patients in the O- GlcNAc high group exhibited significantly shorter progression- free survival (PFS) and overall survival (OS) than that in the O- GlcNAc low group (fig. 1h- 1i). Univariate analysis revealed that O- GlcNAcylation level, alongside age, FIGO stage, and myometrial invasion, was significantly associated with PFS. Subsequent multivariate Cox regression analysis using all the statistically significant variables \((\mathrm{P} < 0.05)\) pinpointed O- GlcNAcylation level and age as independent predictors of clinical outcome of EC patients (table 2). + +<|ref|>text<|/ref|><|det|>[[127, 735, 881, 781]]<|/det|> +Calculated virtual O- GlcNAc index is correlated with tumor histologic grade and survival in TCGA endometrial cancer dataset + +<|ref|>text<|/ref|><|det|>[[127, 790, 881, 892]]<|/det|> +We wanted to validate the correlations observed in our EC cohort using the TCGA EC dataset. Kaplan- Meier analysis of the OS based on either \(OGT\) or \(OGA\) expression level showed no statistical difference (fig. slc- sld), suggesting that the mRNA abundance of \(OGT\) or \(OGA\) alone is insufficient to reflect the \(O\) - GlcNAcylation level. To better estimate + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 882, 310]]<|/det|> +the O- GlcNAcylation level using transcriptomic data, we sent 40 O- GlcNAc high- and 15 O- GlcNAc low- frozen EC samples according to their corresponding RL2 staining index (SI) for RNA- seq (fig. s1e). We calculated the Pearson's correlation coefficient (r) of the transcript level with the SI for each gene, and included the top 1000 genes with \(r > 0.3\) in the O- GlcNAcylation correlated geneset (fig. s1f, supplementary table 1). Gene ontology (GO) analysis revealed that they were enriched in biological processes including cilium organization, cilium assembly, and microtubule- based movement (fig. s1g, supplementary table 2). + +<|ref|>text<|/ref|><|det|>[[125, 346, 882, 811]]<|/det|> +We subsequently constructed a mathematical model based on the expression matrix of the O- GlcNAcylation correlated geneset using machine learning algorithms in R to calculate a virtual O- GlcNAc index for each sample in the TCGA EC cohort (fig. s1e). The calculated O- GlcNAc index in the TCGA dataset exhibited significant association with histologic grade and FIGO stage (fig. 2a, supplementary table 3), consolidating the observations made in our EC cohort. Patients in the advanced histologic grade (WHO grade 3) group had higher O- GlcNAc index in comparison to that in the grade 1 or grade 2 group (fig. 2b). Similarly, EC patients at FIGO stages II, III, or IV demonstrated an increased O- GlcNAc index than that at stage I (fig. 2c). Of note, patients in the copy- number high molecular subtype group, which had the worst clinical outcome among all EC cases26, bore significantly higher O- GlcNAc index than that in other groups (fig. 2d). The O- GlcNAc index increased with age (fig. 2e), but showed no difference between diabetic and non- diabetic groups (fig. 2f). We further stratified the EC patients in the TCGA cohort into O- GlcNAc high and O- GlcNAc low groups using the median O- GlcNAc index as the cutoff. Patients from the O- GlcNAc high group experienced significantly shorter progression- free interval (PFI) and OS than that from the O- GlcNAc low group (fig. 2g- 2h, supplementary table 4). + +<|ref|>text<|/ref|><|det|>[[125, 846, 880, 893]]<|/det|> +In summary, the O- GlcNAc status of EC tissues manifests a significant correlation with histologic grade, both in our EC cohort and the TCGA EC dataset, with elevated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 815, 115]]<|/det|> +O- GlcNAcylation associating with advanced tumor grade and poor clinical outcome. + +<|ref|>text<|/ref|><|det|>[[125, 150, 880, 197]]<|/det|> +Increase of O- GlcNAcylation by inhibition of OGA promotes proliferation and stemness in normal endometrial organoids + +<|ref|>text<|/ref|><|det|>[[125, 205, 881, 394]]<|/det|> +To dissect the functional impact of altered O- GlcNAcylation level on endometrial tissues, we generated endometrial organoids from surgical samples, including normal endometrial epithelial organoids (EE- Os) and endometrial cancer organoids (EC- Os). The EE- Os retained many characteristics of endometrial epithelium, including production of mucins, and expressions of estrogen receptor \(\alpha\) (ER) and progesterone receptor (PR) (fig. 3a). The EC- Os however manifested more irregular cell organizations and elevated O- GlcNAcylation level than the EE- Os, in accordance with their primary tissues (fig. 2a- 2b). + +<|ref|>text<|/ref|><|det|>[[125, 428, 881, 754]]<|/det|> +We treated the EE- Os with OGA small molecular inhibitor Thiamet- G (TMG) to increase the cellular O- GlcNAcylation level (fig. 3b). The addition of TMG resulted in enhanced colony formation and organoids growth of EE- Os (fig. 3c- 3e), along with a rise in the number of mitotic cells within each EE- O (fig. 3f- 3g). Acetylated alpha- tubulin (Ac- tubulin) and PAEP are differentiation markers for multiciliated epithelial cells and secretory cells respectively in the endometrium27. TMG treatment reduced the number of both PAEP positive cells and Ac- tubulin labeled multiciliated cells (fig. 3h- 3i), suggesting that the elevated O- GlcNAcylation level caused de- differentiation of the endometrial cells in the EE- Os. We further examined the expression levels of a panel of stemness markers of the endometrium, including SSEA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2. In contrast to PAEP whose mRNA level was decreased upon TMG treatment, all the examined stemness markers showed upregulated expressions (fig. 3c). + +<|ref|>text<|/ref|><|det|>[[125, 790, 881, 893]]<|/det|> +To further characterize the influence of TMG treatment on different cell subtypes in the EE- Os, the control and TMG treated EE- Os were subject to single- cell RNA- seq analysis (fig. 3j). The cells were clustered and classified into six major subtypes according to the specific expressions of known markers28,29: pre- ciliated, ciliated, stem, proliferative, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 883, 310]]<|/det|> +O- GlcNAc- related stem- like, and inflammatory (fig. s2d- s2g, supplementary table 5). Of note, we identified an O- GlcNAc- related stem- like subtype in which the cells displayed activated signaling pathways regulating pluripotency of stem cells, as well as the O- glycan biosynthesis (fig. s2c). The TMG treatment of EE- Os resulted in a substantial decrease of cells in the ciliated and pre- ciliated subtypes, and a concurrent increase of cells in the proliferative and O- GlcNAc- related stem- like subtypes (fig. 3k). Together, these results suggest that upregulation of O- GlcNAcylation level promotes proliferation and stemness of endometrial epithelial cells. + +<|ref|>text<|/ref|><|det|>[[124, 346, 881, 392]]<|/det|> +Inhibition of OGT decreases cell proliferation and induces differentiation and cell death in endometrial cancer organoids + +<|ref|>text<|/ref|><|det|>[[123, 400, 883, 840]]<|/det|> +The EC- Os faithfully reflected the molecular characteristics of their primary EC tissues, and bore higher O- GlcNAcylation level relative to EE- Os (fig. 4a and s2b). We treated the EC- Os with OSMI- 1, a chemical inhibitor of OGT, to reduce the O- GlcNAcylation level (fig. 4b). Addition of OSMI- 1 impeded the formation and growth of EC- Os (fig. 4c). A significant fraction of the EC- Os displayed darkening and cell lysing in the presence of OSMI- 1, resulting in reductions of both the number and size of the EC- Os compared to time- matched control (fig. 4d- 4c). TUNEL stainings revealed that many cells in the OSMI- 1 treated EC- Os underwent apoptosis (fig. 4f). We performed immunofluorescence on the remaining EC- Os with relatively normal size and morphology. Mitotic cells as visualized by phospho- histone H3 (PH3) staining became barely detectable in EC- Os after OSMI- 1 treatment (fig. 4g- 4h). Meanwhile, the population of both the PAEP positive secretory cells and Ac- tubulin labeled multiciliated cells increased in these EC- Os (fig. 4i- j), suggesting that OSMI- 1 treatment promoted differentiation. Consistently, the expression of stemness markers, including SSEA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2, were significantly downregulated in the OSMI- 1 treated EC- Os, accompanying the upregulation of the differentiation marker PAEP (fig. 4k). + +<|ref|>text<|/ref|><|det|>[[124, 873, 880, 893]]<|/det|> +In summary, these results indicate that a balanced O- GlcNAcylation level is crucial for + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 882, 171]]<|/det|> +maintaining the cell fate in endometrial organoids. Elevated O- GlcNAcylation promotes proliferation and stemness of EE- Os, whereas downregulated O- GlcNAcylation induces differentiation and cell death, limiting the uncontrolled proliferation of cells in the EC- Os. + +<|ref|>text<|/ref|><|det|>[[123, 207, 883, 755]]<|/det|> +Genome- wide screen for tumor suppressors that maintain O- GlcNAcylation homeostasis To identify crucial factors regulating O- GlcNAcylation homeostasis in EC, we conducted a comprehensive genome- wide CRISPR- Cas9 knockout screen. A lentiviral single guide RNA (sgRNA) library targeting 19,050 genes (6 sgRNAs/gene) was transduced into 293T cells, along with 1000 nontargeting control sgRNAs, at a multiplicity of infection (MOI) of 0.3 to ensure each cell expressed only one sgRNA. Following cell staining with an anti- O- GlcNAc antibody (RL2), we isolated the top \(5\%\) RL2- positive cells via fluorescence- activated cell sorting (FACS) and conducted deep sequencing of the sgRNAs from this cell population (fig. 5a- 5b). The sgRNA abundance was then used to calculate a robust rank aggregation (RRA) score for each gene using MAGeCK30, and the genes were ranked accordingly, with a smaller RRA score indicated greater essentiality (supplementary table 6). We reviewed the literatures and collected known regulators whose inactivation could impact cellular O- GlcNAcylation homeostasis31- 46. Genes that negatively regulate O- GlcNAcylation, such as TSC2, SIRT1, and TP53, had smaller RRA scores and were enriched in the first half of the gene list, comparing to the known positive regulators of O- GlcNAcylation (fig. 5c). We performed Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis on the 1038 high- confidence genes ( \(P < 0.05\) ) from the genome- wide screen. These genes were enriched in pathways including ECM- receptor interaction, thermogenesis, histidine metabolism, proteoglycan in cancer, and maturity onset diabetes of the young (fig. 5d, supplementary table 7). + +<|ref|>text<|/ref|><|det|>[[125, 791, 882, 893]]<|/det|> +To further pinpoint key regulators that impact O- GlcNAcylation level in EC tissues, we cross- referenced the 1038 positive hits in the screen with 526 putative tumor suppressor genes of EC47,48. As a result, 18 overlapping genes were identified, including ACVR1C, AGTR1, CADM2, PRKAA1, CDKN1C, CMTM3, DIRAS3, SIK1, EPHB4, GATA5, ITGAV, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 882, 255]]<|/det|> +KLF10, MAP3K8, PLA2G2A, PTPN11, RNASEL, SPARCL1, and FBXO31 (fig. 5e). We conducted Kaplan- Meier analysis using the TCGA EC dataset for each gene, and found that only FBXO31 showed downregulated expression in EC that was associated with poor survival (fig. s3a and 5g). Therefore, we generated FBXO31 knockout (FBXO31- KO) 293T cells using CRISPR (fig. s3b). Immunostaining with RL2 antibody confirmed that O- GlcNAcylation level was significantly increased in the FBXO31- KO cells (fig. 5f). + +<|ref|>text<|/ref|><|det|>[[125, 290, 736, 309]]<|/det|> +FBXO31 interacts with and ubiquitinates OGT to limit the O- GlcNAc level + +<|ref|>text<|/ref|><|det|>[[125, 317, 882, 615]]<|/det|> +FBXO31 functions as a substrate recognition component in the SCF ubiquitin E3 ligase complex to control the homeostasis of many proteins49- 54. Accordingly, FBXO31 might regulate O- GlcNAcylation level by directly binding and ubiquitinating the O- GlcNAc transferase OGT. To confirm the interaction between FBXO31 and OGT, we performed pull- down assay using bacterially purified GST- OGT to incubate with lysates of 293T cells expressing GFP- FBXO31. Western blot showed that GST- OGT pulled down significant amount of GFP- FBXO31 relative to GST control (fig. 6a). We further validated the interaction using co- immunoprecipitation in 293T cells overexpressing Flag- OGT and GFP- FBXO31. GFP- FBXO31 was co- immunoprecipitated with Flag- OGT, and both the amounts of Flag- OGT and GFP- FBXO31 in the immunoprecipitant were increased in the presence of the proteasome inhibitor MG132 (fig. 6b). + +<|ref|>text<|/ref|><|det|>[[125, 650, 882, 893]]<|/det|> +To assess whether the interaction with FBXO31 controlled the protein homeostasis of OGT, we transfected 293T cells with increasing amounts of GFP- FBXO31 and detected the levels of OGT as well as O- GlcNAcylation by western blot. Both the OGT protein and cellular O- GlcNAcylation levels demonstrated a negative correlation with the amount of GFP- FBXO31 (fig. 6c). Additionally, the downregulation of OGT induced by GFP- FBXO31 overexpression was significantly reversed by MG132, suggesting that FBXO31 controlled the OGT level via the ubiquitin- dependent proteasome degradation process (fig. 6d). To ascertain that FBXO31 could induce ubiquitination of OGT, we immunoprecipitated OGT from 293T cell lysates overexpressing GFP- FBXO31 and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 883, 616]]<|/det|> +HA- ubiquitin. Western blot detected strong polyubiquitination of OGT in the presence of GFP- FBXO31 (fig. 6e). We further tested whether FBXO31 could ubiquitinate OGT in vitro. The SCF complex was affinity- purified with anti- HA magnetic beads from 293T cells expressing HA- tagged Skp1, Cul1, and Roc1 with or without FBXO31, and then incubated with bacterially purified E1, E2, ubiquitin, and His- OGT. Polyubiquitination signals of His- OGT were detected, suggesting that the FBXO31- containing SCF complex could directly ubiquitinate OGT (fig. 6f). Skp1 in the SCF complex recruits F- box proteins via their F- box motif. We mutated the F- box of FBXO31 (FBXO31 \(\Delta \mathrm{F}\) ) and assessed its ability to induce polyubiquitination of OGT in 293T cells. Overexpression of HA- ubiquitin and GFP- FBXO31 resulted in strong polyubiquitination of the immunoprecipitated Flag- OGT, which was significantly reduced when GFP- FBXO31 was replaced with the GFP- FBXO31 \(\Delta \mathrm{F}\) mutant (fig. 6g). These results confirmed that FBXO31, together with other components of SCF complex, possessed a new ubiquitin E3 ligase activity toward OGT. We further evaluated the impact of FBXO31 in controlling the cellular OGT homeostasis using FBXO31- KO 293T cells. Both the OGT and O- GlcNAcylation levels were increased in FBXO31- KO cells (fig. 6h). Cycloheximide (CHX) treatment, which blocked new protein synthesis, uncovered that the half- life of OGT was significantly extended in FBXO31- KO cells relative to control (fig. 6i), indicating that FBXO31 is indispensable for limiting the cellular OGT level. + +<|ref|>text<|/ref|><|det|>[[125, 652, 883, 893]]<|/det|> +Loss of FBXO31 increases O- GlcNAcylation and promotes endometrial organoids growthWe investigated the clinical relevance of the FBXO31- mediated ubiquitination of OGT using the endometrial specimens in our EC cohort. IHC staining revealed that the protein level of FBXO31 was significantly downregulated in EC relative to normal endometrial tissues, often manifesting an anti- correlation pattern to that of O- GlcNAcylation (fig. 7a- 7b). We semi- quantified the expression level of FBXO31 based on the IHC signals, and found that the FBXO31 protein level in the O- GlcNAc low EC group was markedly higher than that in the O- GlcNAc high group (fig. 7c). Western blot uncovered that the FBXO31 protein level was decreased, accompanying the increase of OGT level, in EC- Os comparing to the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 200]]<|/det|> +EE- Os (fig. 7d), suggesting that the elevated O- GlcNAcylation in EC tissues was due to stabilization and upregulation of OGT. Moreover, we categorized the EC cases into FBXO31- low and FBXO31- high groups. The FBXO31 expression exhibited significant association with histologic grade, diabetes, as well as the O- GlcNAc status (table 3). + +<|ref|>text<|/ref|><|det|>[[125, 234, 883, 533]]<|/det|> +To elucidate the functional impact of FBXO31 alterations in endometrial tissues, we knocked down the expression of FBXO31 using lentivirus- mediated expression of shRNAs in EE- Os. Downregulation of FBXO31 resulted in increased amount of O- GlcNAcylation in EE- Os (fig. 7e). Particularly, the growth of EE- Os was significantly enhanced by FBXO31 knockdown, in alignment with upregulated expression of the stemness markers SSA- 1, SOX9, ALDH1, OCT4, CD133, and SOX2 (fig. 7f- 7g). This enhanced growth of EE- Os after FBXO31 knockdown could be inhibited by OSMI- 1 treatment, indicating that it was a result of elevated O- GlcNAcylation (fig. 7h). Reciprocally, given that FBXO31 was downregulated in EC- Os, we supplemented the EC- Os with GFP- FBXO31 or GFP control using lentivirus- mediated transduction. Overexpression of GFP- FBXO31 significantly impeded the formation of EC- Os (fig. 7i). + +<|ref|>text<|/ref|><|det|>[[126, 569, 884, 669]]<|/det|> +In summary, our results identify FBXO31 as one of the key rheostats that controls the O- GlcNAc homeostasis by ubiquitinating OGT. FBXO31 is frequently downregulated in EC, resulting in stabilization of OGT and elevation of cellular O- GlcNAcylation level that advance endometrial malignancy. + +<|ref|>sub_title<|/ref|><|det|>[[128, 707, 221, 723]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[126, 734, 883, 893]]<|/det|> +This study delves deeply into the intricate relationship between O- GlcNAc homeostasis and the progression of EC, elucidating the clinical significance of abnormal O- GlcNAcylation and unveiling an important regulatory module controlling the O- GlcNAc homeostasis in endometrial tissues. First, by examining the O- GlcNAc status in a Chinese EC cohort containing 219 tumors, we found that O- GlcNAcylation level correlates with histologic grade, FIGO stage, and distant metastasis. Patients with higher O- GlcNAcylation levels + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[122, 95, 883, 616]]<|/det|> +experienced shorter OS and PFS, indicating that it is an independent risk factor for poor prognosis in EC. Second, we performed RNA- seq on 55 frozen EC samples (40 O- GlcNAc high- and 15 O- GlcNAc low- specimens according to their corresponding immunohistochemistry (IHC) scores) and constructed a predicting model to evaluate the O- GlcNAc level based on transcriptomic data. We calculated the virtual O- GlcNAc index for each sample in the TCGA EC cohort and validated that O- GlcNAc index is associated with histologic grade, FIGO stage, and poor prognosis. Third, we assessed the functional significance of O- GlcNAcylation in EC progression by treatment of patient- derived organoids with clinically relevant chemical inhibitors. Inhibition of OGA by TMG increased O- GlcNAcylation and promoted proliferation and stem- like cell properties in normal EE- Os, whereas inhibition of OGT by OSMI- 1 decreased O- GlcNAcylation, downregulating stemness and limiting the proliferation of EC- Os. Last, we performed genome- wide CRISPR screen for known tumor suppressors whose inactivation could increase cellular O- GlcNAc level. We identified that FBXO31 can function as a substrate receptor of the SCF ubiquitin ligase complex to ubiquitinate the O- GlcNAc transferase OGT in vitro. FBXO31 was downregulated in EC tissues, anticorrelating with O- GlcNAc levels. Knockdown of FBXO31 in EE- Os promoted expression of stemness markers and accelerated organoid growth, suggesting that inactivation of FBXO31 advances endometrial malignancy by stabilizing OGT and elevating cellular O- GlcNAcylation level. + +<|ref|>text<|/ref|><|det|>[[123, 650, 883, 893]]<|/det|> +Integrative analysis of O- GlcNAcylation in large clinical EC cohort to assess its relationships with current histomorphologic and molecular subtypes of EC had not been conducted till this study. Altered O- GlcNAc cycling has been linked to many characteristics of EC cells, impacting their survival and proliferation signaling, epithelial- mesenchymal transition (EMT) and metastatic behaviors, drug- resistance, as well as metabolic and cell fate plasticity1,5,8. A previous report using 76 EC samples revealed that the two executing enzymes of O- GlcNAcylation, OGT and OGA, manifested increased mRNA levels in ECs of higher histologic grade relative to the well- differentiated tumors22. A more recent IHC analysis on a tissue microarray containing 28 EC specimens showed that both the OGT and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 883, 590]]<|/det|> +O- GlcNAcylation levels were increased in EC tissues than the adjacent normal endometrial tissues23. This pilot study indicated that increased O- GlcNAcylation was associated with histologic grade, clinical stage, and lymph node metastasis. However, when repeating the IHC analysis using the same tissue microarray, we only observed increased OGT and O- GlcNAcylation levels in ECs but failed to associate O- GlcNAcylation levels with any of the clinical parameters, probably due to the limited sample size and differences in inclusion and exclusion criteria. Nonetheless, when we expanded the analysis to our EC cohort containing 219 patients as well as the TCGA EC dataset, the clinical significance of O- GlcNAcylation became invariable. The O- GlcNAcylation level shows strong association with histologic grade, FIGO stage, and poor prognosis. Particularly, the O- GlcNAc high EC group enriches more serous carcinoma as well as high grade endometrioid carcinoma patients. This is perhaps because serous carcinoma often carries TP53 mutations and p53 is a known negative regulator of O- GlcNAcylation level not only identified in our genome- wide screen but also reported in a previous study31. Moreover, EC patients belonging to the copy- number high molecular subtype group showed significantly higher O- GlcNAcylation level than that in the other groups. These results suggest that O- GlcNAcylation is a useful factor complementary to the current classification system to better identify EC patients with poor clinical outcome. + +<|ref|>text<|/ref|><|det|>[[125, 624, 883, 894]]<|/det|> +Our understanding on the molecular circuitry controlling the cellular O- GlcNAcylation homeostasis is far from complete. Given that O- GlcNAcylation is dependent on nutrient availability, metabolic factors such as GFPT1, POLDIP2, and PPM1K have been reported to influence O- GlcNAc level by modulating the metabolic flux of the HBP pathway31,34,39,45,46. However, emerging evidence indicates that O- GlcNAcylation may also be regulated by non- nutrient dependent mechanisms, particularly at the protein level of OGT32,33,37,41- 43. OGT has been reported to be regulated by the balance of ubiquitination and deubiquitination9,10,12,13. The E3 ligases XIAP and E6AP could promote the ubiquitin- dependent proteasome degradation of OGT10,11. The histone demethylase LSD2 displayed an atypical ubiquitin E3 ligase activity toward OGT in the A549 cells9. However, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[124, 95, 884, 476]]<|/det|> +none of these reported E3 activities showed clinical relevance in ECs. Our results uncovered that FBXO31, together with other components in the SCF complex, functions as a new E3 ligase for OGT, impacting the progression of endometrial malignancy. FBXO31 is a tumor suppressor gene located in the 16q24.3 region, with frequently observed loss of heterozygosity in several cancers, including breast, ovarian, hepatocellular, and prostate cancers55,56. The SCFFBXO31 complex can ubiquitinate many protein substrates that are cell cycle regulators, such as cyclin D149, Cdt150, MDM252, and cyclin A57; signaling molecules, such as c- Myc58, β- catenin59, and MKK651; epithelial- mesenchymal transition (EMT) factors, Snail160 and Slug61; as well as ferroptosis inhibitor GPX462. How it recognizes these substrates remains unclear. Our results indicate that FBXO31 can directly bind to OGT, triggering its polyubiquitination. Future molecular and structural characterization of this interaction between FBXO31 and OGT may help elucidate the substrate recognition mechanism of SCFFBXO31, paving the way for the development of new intervention strategies. + +<|ref|>text<|/ref|><|det|>[[125, 512, 883, 893]]<|/det|> +Last but not least, how the aberrant O- GlcNAcome in ECs influences the tumor cells' proliferation, survival, and cell fate plasticity is not fully understood. Low- throughput, individual characterizations of potential O- GlcNAcylation substrates have revealed that many EC- related oncogenes and tumor suppressors, such as PI3K, PTEN, ARID1A63, p5364, Myc65, and β- catenin66, possess O- GlcNAcylation sites. Particularly, in alignment of our finding that increased O- GlcNAcylation promotes stemness of EC cells, master regulators controlling the stem cells' self- renewal and pluripotency, including Oct47, Sox26, and Sox99, are able to be modified by O- GlcNAcylation. It is worthy of functional interrogation of these putative O- GlcNAcylated substrates in patients- derived endometrial organoids. Ultimately, future development of high- throughput, tissue- specific proteomic profiling methods is needed for fully capture the spatiotemporal dynamics of the O- GlcNAcome during the progression of ECs and other pathophysiological processes, consolidating the foundation of targeting O- GlcNAc cycling to develop new therapeutic strategies in clinical settings. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[128, 125, 207, 141]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[128, 153, 250, 168]]<|/det|> +Human tissues + +<|ref|>text<|/ref|><|det|>[[126, 179, 881, 365]]<|/det|> +All fresh tissues and paraffin- embedded (FFPE) tissues were prospectively obtained from patients with endometrial diseases at Xiangya Hospital, Central South University. Clinical data and histopathological characteristics were retrieved from patient records and routine pathology reports. The study was approved by the Medical Ethics Committee of Central South University (No. 202103076), and all participating patients provided informed written consent. The study was registered with and approved by the Human Genetics Resource (HGR) office of the Minister of Science and Technology of China. + +<|ref|>text<|/ref|><|det|>[[128, 401, 580, 419]]<|/det|> +Organoids culturing from endometrial surgical samples + +<|ref|>text<|/ref|><|det|>[[126, 427, 881, 894]]<|/det|> +The endometrial organoids were generated as previously described70. Tumor tissues and normal tissues were isolated and stored in ice- cold serum- free DMEM medium supplemented with 1% penicillin- streptomycin. The tissues were then washed in ice- cold DPBS (Biological Industries) supplemented with penicillin- streptomycin and minced into small pieces. The tissues were digested by collagenase IV (1- 2 mg/mL; 17104019, Thermo Fisher Scientific) in the presence of Rock inhibitor (10 μM; SCM075, Merck Millipore) and penicillin- streptomycin for 1 hour on a shaker at 37°C, then incubated for 15 minutes in TrypLE (1×; 12604013, Thermo Fisher Scientific) supplemented with Rock inhibitor and penicillin- streptomycin. Subsequently, the tissue digests were stopped by ice- cold serum- free DMEM/F12 and after centrifugation, a 100- μm cell strainer was used to obtain cell pellets. Finally, the cell pellets were resuspended in 70% matrigel/30% DMEM/F12 (356231, Corning and 11039021, Gibco, respectively) and seeded in 50 μL droplets in non- treated 24- well plates. After incubation at 37°C and 5% CO2 in a cell culture incubator for 20- 30 minutes, the pre- warmed organoid complete medium (DMEM/F12 supplemented with 1% penicillin- streptomycin, 2% B27 supplement minus vitamin A (12587010, Gibco), 5% R- spondin conditioned medium, 1% chemically defined lipid concentrate (11905031, Gibco), recombinant human Noggin 100 ng/mL (HY- P7051A, MCE), 1% N2 (17502048, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 283]]<|/det|> +Gibco), N-acetyl- L- cysteine 1.25 mM (A7250, Sigma Aldrich), Nicotinamide 10 \(\mu \mathrm{M}\) (73240, Sigma Aldrich), recombinant human EGF 50 ng/mL (236- EG- 01M, R&D Systems), Y- 27632 10 \(\mu \mathrm{M}\) (SCM075, Sigma Aldrich), 17- \(\beta\) estradiol 10 nM (E8872, Sigma Aldrich), SB202190 0.1 \(\mu \mathrm{M}\) (S7067, Sigma Aldrich), A83- 01 0.25 \(\mu \mathrm{M}\) (SML0788, Sigma Aldrich), recombinant human IGF 40 ng/mL (100- 11, Peprotec), recombinant human HGF 20 ng/mL (100- 39, Peprotec), IL- 6 5 ng/mL (200- 06, Peprotec)) was added. The organoid medium was changed every 2 days, and the organoids were passaged after 7- 10 days of culture. + +<|ref|>sub_title<|/ref|><|det|>[[128, 319, 369, 336]]<|/det|> +## Immunohistochemistry (IHC) + +<|ref|>text<|/ref|><|det|>[[125, 345, 882, 670]]<|/det|> +Endometrial cancer tissue array was purchased from Xinchao Biotechnology Company (HUteA060CS01, Shanghai, China), consisting of 26 pairs of cancerous and paracancerous tissue specimens, along with an additional 8 cases of cancerous tissue without paired paracancerous tissue. After removing the incomplete tissue spots, 31 cases of cancer tissue and 23 cases of para- cancerous tissue were included in IHC analysis. IHC was performed as previously described71, with primary antibody incubation overnight after antigen retrieval and endogenous peroxidase activity blocking on paraffin sections. The IHC staining signal levels were blindly scored by two independent assessors without knowledge of clinical parameters. The staining index (SI) was calculated by multiplying the staining intensity score (0- 3) and the proportion of positively stained tumor cells score (0- 4), resulting in a SI ranging from 0 to 12. High and low expression were defined as SI 0- 6 and SI 8- 12, respectively72. + +<|ref|>sub_title<|/ref|><|det|>[[128, 708, 269, 725]]<|/det|> +## Survival analysis + +<|ref|>text<|/ref|><|det|>[[125, 734, 882, 893]]<|/det|> +Progression- free survival (PFS) was calculated as the time between the surgery that procured the sample and the date of disease progression or of a new metastatic event in a different location. Overall survival (OS) was defined as the interval between the date of surgery and the date of death or last follow- up. Progression- free interval (PFI) was defined as the duration from surgery to the first occurrence of disease progression or death after treatment. The curves were stratified based on the O- GlcNAcylation (RL2) level. Log- rank + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 880, 142]]<|/det|> +test was used to compare the two groups over a follow- up time of 61 months. Kaplan- Meier survival curves were generated and compared using GraphPad Prism (version 8.0.2). + +<|ref|>text<|/ref|><|det|>[[125, 180, 525, 198]]<|/det|> +Generation of O- GlcNAc index prediction model + +<|ref|>text<|/ref|><|det|>[[123, 204, 881, 755]]<|/det|> +The RNA- seq data of 15 low O- GlcNAcylation level (RL2 by IHC) tumor tissues and 40 high O- GlcNAcylation level tumor tissues were processed to identify the O- GlcNAc correlated genes. The gene expression matrix of these 55 EC samples was correlated with the O- GlcNAc IHC staining index using the Pearson correlation method in the mlr3. filters package within the mlr3 framework in R. The top 1000 genes with a correlation coefficient greater than 0.3 were included in the O- GlcNAc correlated geneset. Subsequently, mlr3 learners including six regression model- based approaches (regr.lm, regr.glmnet, regr.kknn, regr.ranger, regr.rpart, regr.svm) was applied to the expression matrix of the 1000 O- GlcNAc correlated genes. The O- GlcNAc indices for the 55 EC tissues were calculated, subjecting to 5- fold cross- validations of training and ranking based on predefined performance metrics. The reliability of the prediction model was assessed by comparing the calculated O- GlcNAc indices with actual IHC SI scores. The regr.glmnet demonstrated the lowest mean squared error (MSE) and was selected for establishment of the final prediction model. The O- GlcNAc indices were then calculated using the prediction model for the 589 EC samples in TCGA. The patients were categorized into high and low O- GlcNAc groups using the median of the calculated O- GlcNAc indices. Wilcoxon Mann- Whitney tests were used to assess differences between the two groups in terms of histologic grade, FIGO stage, molecular subtype, age, and diabetes. Log- rank tests were employed to compare the OS and PFI differences between the high and low O- GlcNAc groups, and Kaplan- Meier survival curves were generated and compared using R (version 4.03). + +<|ref|>text<|/ref|><|det|>[[125, 792, 405, 808]]<|/det|> +Immunofluorescence of organoids + +<|ref|>text<|/ref|><|det|>[[125, 818, 881, 892]]<|/det|> +Immunofluorescence staining experiments were performed on organoids as previously described73. When the organoids reached a size of approximately \(100 \mu \mathrm{m}\) , they were selected for staining. After washing twice with pre- cooled DPBS, \(500 \mu \mathrm{L}\) of cell recovery + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 883, 533]]<|/det|> +solution (354253, Corning) was added to each well, and the matrigel was dissolved on ice to ensure that the morphology of the organoids was not disrupted. After 30 minutes, all the organoids were collected into a \(15~\mathrm{mL}\) centrifuge tube, fixed with \(4\%\) paraformaldehyde for 30 minutes, and then centrifuged to remove the supernatant. Next, \(10~\mathrm{mL}\) of \(1\%\) PBST was added to stop the tissue fixation. After blocking with Organoid Washing Buffer (OWB, \(0.1\%\) Triton X- 100, \(0.2\%\) BSA in DPBS), the primary antibody was added and incubated overnight at \(4^{\circ}\mathrm{C}\) with shaking at \(60~\mathrm{rpm}\) . On the following day, the organoids were washed three times with OWB for 2 hours each time, and then the corresponding fluorescent secondary antibody was added. The organoids were incubated overnight on a shaker in the dark. On the third day, \(4^{\prime},6\) - Diamidino- 2- phenylindole dihydrochloride (DAPI, D9542, Sigma) at \(10~\mu \mathrm{g / mL}\) was added for 30 minutes. After washing with OWB, the samples were spun down at \(70\times \mathrm{g}\) for 5 minutes at \(4^{\circ}\mathrm{C}\) . Finally, the organoids were resuspended with fructose- glycerol clearing solution ( \(60\%\) glycerol and \(2.5\mathrm{M}\) fructose in \(\mathrm{ddH_2O}\) ) and imaged using an LSM880 confocal microscope (Zeiss). A cell death detection (TUNEL) kit (Roche) was used to identify dead cells in accordance with the company's description. All the antibodies used in this study were listed in supplementary table 9. + +<|ref|>sub_title<|/ref|><|det|>[[128, 569, 425, 586]]<|/det|> +## Lentiviral transduction of organoids + +<|ref|>text<|/ref|><|det|>[[125, 595, 883, 893]]<|/det|> +For organoid lentiviral transduction, pLKO.1- puro vectors and TK- PCDH- copGFP- T2A- Puro vectors were used. The organoids were washed twice with pre- cooled DPBS, and \(500~\mu \mathrm{L}\) of TrypLE (12604013, Thermo Fisher Scientific) was added to each well for 10 minutes at \(37^{\circ}\mathrm{C}\) . The matrigel was disrupted by pipetting the mixture up and down repeatedly during digestion. TrypLE was inactivated by adding \(10~\mathrm{mL}\) of ice- cold serum- free DMEM/F12, and the mixture was centrifuged for 5 minutes at \(200\times \mathrm{g}\) . After digestion, the organoids were made into single cells or cell mass and resuspended in virus infection solution containing Rock inhibitor, polybrene, and concentrated lentivirus in organoid culture media. The cell suspension was added to a 6- well plate, spun at \(2000~\mathrm{rpm}\) for 1 hour, and then incubated at \(37^{\circ}\mathrm{C}\) for 5- 6 hours. The cells were then transferred to a 15 mL centrifuge tube, washed twice with serum- free DMEM/F12, and seeded in a prewarmed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 225]]<|/det|> +24- well plate with \(70\%\) matrigel. Then, \(500~\mu \mathrm{L}\) of organoid medium was added to each well, followed by incubation at \(37^{\circ}\mathrm{C}\) with \(5\%\) \(\mathrm{CO_2}\) for 20 minutes. The medium was changed every 2 days. Puromycin selection (1 \(\mu \mathrm{g / mL}\) ) in organoid culture was conducted for 3- 4 days to establish stably infected organoids. The stable organoids were validated by western blot or quantitative RT- PCR. + +<|ref|>sub_title<|/ref|><|det|>[[128, 263, 304, 280]]<|/det|> +## Quantitative RT-PCR + +<|ref|>text<|/ref|><|det|>[[125, 290, 881, 475]]<|/det|> +RNA extraction was performed using TRIzol (87804, Life Technologies) according to the manufacturer's protocol for all samples, including cells, organoids, and primary tissues. The extracted RNA was then reverse transcribed to cDNA using the PrimeScript RT Reagent Kit (RR037A, Takara). The cDNA was used as a template for qPCR, which was performed using the SYBR Green qPCR Master Mix (QST- 100, SolomonBio) on the QuantStudio 3 Real- Time PCR system (Applied Biosystems). All the primers were listed in supplementary table 8. + +<|ref|>sub_title<|/ref|><|det|>[[128, 514, 440, 530]]<|/det|> +## Western blot and immunoprecipitation + +<|ref|>text<|/ref|><|det|>[[125, 540, 881, 893]]<|/det|> +Cells were lysed in sample buffer (2% SDS, \(10\%\) glycerol, and \(62.5\mathrm{mM}\) Tris- HCl, pH 6.8) supplemented with \(1\times\) protease inhibitor cocktail (P8340, Sigma). The protein concentration was measured using a BCA kit (P0009, Beyotime). Cell lysates were separated by SDS- PAGE and transferred onto a nitrocellulose membrane. The membrane was then blocked with \(5\%\) non- fat dry milk for 1 hour at room temperature and probed with the indicated primary antibodies overnight at \(4^{\circ}\mathrm{C}\) Antigen- antibody complexes were detected by incubating with horseradish peroxidase secondary antibodies followed by ECL substrates (WBKLS0500, Millipore). For immunoprecipitation experiments, cells were washed twice with ice- cold PBS and then lysed in lysis buffer (20 mM Tris- HCl (pH 8.0), 137 mM NaCl, \(1\%\) NP- 40, \(2\mathrm{mM}\) EDTA) on ice for 30 minutes. Cell lysates were gently mixed with specific antibodies overnight at \(4^{\circ}\mathrm{C}\) under gentle rotation, then incubated with protein A/G beads (SC- 2003, Santa cruz) for 1- 2 hours at \(4^{\circ}\mathrm{C}\) Immunoprecipitants were washed three times with lysis buffer. After the final wash, the supernatant was aspirated and discarded, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 199]]<|/det|> +and the pellet was resuspended in \(2 \times\) SDS sample buffer (0.125 M Tris HCl (pH 6.8), 4% SDS, 20% glycerol, 2% \(\beta\) - mercaptoethanol, 0.02% bromophenol blue). The sample was then subjected to reducing SDS- PAGE and western blot. All the antibodies used in this study were listed in supplementary table 9. + +<|ref|>sub_title<|/ref|><|det|>[[128, 235, 451, 253]]<|/det|> +## Cell culture and generation of cell lines + +<|ref|>text<|/ref|><|det|>[[125, 262, 882, 533]]<|/det|> +HA- R- Spondin1- Fc 293T cell line (3710- 001- 01, R&D Systems) was used to produce R- spondin conditional media. HEK293T and HeLa cells were maintained in DMEM (06- 1055- 57- 1 ACS, Vivocell) supplemented with 10% FBS. All cells were cultured at \(37^{\circ}\mathrm{C}\) in a humidified incubator with 5% \(\mathrm{CO_2}\) and periodically screened for Mycoplasma contamination. To generate 293T FBXO31 KO cell lines, the cells were transfected with LentiCRISPR- V2 plasmid carrying sgFBXO31 (supplementary table 8) and further selected with \(1 \mu \mathrm{g / mL}\) puromycin (s7417, Selleck) for 3 days. The cells were then plated at single- cell density in 100 mm petri dishes, and the individual clones that emerged were picked and replated into 24- well plates. The loss of FBXO31 expression was confirmed by western blot and Sanger sequencing. + +<|ref|>sub_title<|/ref|><|det|>[[128, 569, 454, 586]]<|/det|> +## In vivo and in vitro ubiquitination assay + +<|ref|>text<|/ref|><|det|>[[125, 595, 882, 865]]<|/det|> +For detection of ubiquitinated proteins in vivo, 293T cells were co- transfected with expression vectors for HA- ubiquitin and the indicated proteins. Polyubiquitinated OGT was detected by immunoprecipitation of OGT with ANTI- FLAG® M2 Affinity Gel (A2220, Merck Millipore) under denaturing conditions followed by Western blot with an anti- HA antibody. In vitro ubiquitination was performed as previously described74. The SCF- FBXO31 (E3) complexes were immunopurified from the cell lysate using Pierce™ Anti- HA Magnetic Beads (88836, Thermo Fisher Scientific) and incubated with His- OGT fusion protein expressed and purified from E. coli as previously reported75 in the presence of recombinant purified E1 (UBA1; 11990- H20B, sinobiological), E2 (UBE2D1; 11432- H07E, sinobiological), recombinant human ubiquitin protein (U- 100H, Boston Biochem), and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 880, 144]]<|/det|> +ubiquitination buffer (20 mM Tris- HCl, pH 7.5, 5 mM MgCl2, 0.5 mM DTT, 2 mM ATP). The reaction was stopped by adding \(2 \times\) SDS sample buffer and boiling for 10 minutes. + +<|ref|>text<|/ref|><|det|>[[128, 180, 451, 198]]<|/det|> +CRISPR- Cas9 screen and data analysis + +<|ref|>text<|/ref|><|det|>[[125, 207, 883, 616]]<|/det|> +The human genome- scale CRISPR knockout library (GeCKO v2, Addgene #1000000048) in the lentiCRISPR v2 vector (Addgene #52961) consists of 123,411 sgRNAs that target 19,050 protein- coding genes (6 sgRNAs per gene) and 1,000 nontargeting control sgRNAs was used \(^{76,77}\) . The human GeCKO v2 library was transduced into 293T cells by lentivirus at a multiplicity of infection of 0.3. Cells were selected with puromycin for 7 days followed by fluorescence- activated cell sorting (FACS) based on their O- GlcNAc staining intensities. An unsorted sample was used to assess sgRNA library coverage, and the sorted RL2 high population was subjected to genomic DNA extraction. The inserted sgRNA library was amplified by two steps of PCR for next- generation sequencing. Each screen was performed twice. For data analysis, reads from the fastq files generated by sequencing were tallied for each guide by taking the first 20 bp from each read and mapping to the identical short gRNA sequence. For each screen, a table of reads per guide that includes the counts from the RL2 high population of both replicates was generated and loaded into MAGeCK \(^{30}\) . Top genes were determined based on their mean log2 fold change, FDR, and robust ranking aggregation (RRA) score. + +<|ref|>sub_title<|/ref|><|det|>[[128, 652, 258, 669]]<|/det|> +## Data download + +<|ref|>text<|/ref|><|det|>[[127, 679, 882, 782]]<|/det|> +The TCGA UCEC dataset used in this study, including the gene raw count data (htseq- count files), and the annotated somatic simple nucleotide variation files (MuTect2 VCF), were downloaded using the gdc- client v1.6.0. The clinical OS and PFI information were obtained from Liu.et al \(^{78}\) . + +<|ref|>sub_title<|/ref|><|det|>[[128, 820, 424, 837]]<|/det|> +## RNA-seq and bioinformatic analysis + +<|ref|>text<|/ref|><|det|>[[127, 846, 882, 893]]<|/det|> +Total RNA was isolated from EC tissues, and libraries were generated using the NEBNext UltraTM RNA Library Prep Kit (New England Biolabs) for the Illumina system. Sequencing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 338]]<|/det|> +was conducted using the Illumina Novaseq 6000 platform (Novogene). Trim Galore v0.6.4 was employed to eliminate adapter sequences and remove reads of poor quality. Subsequently, the reads from each RNA- seq data were aligned to the hg38 genome assembly using STAR v2.7.2d. The key alignment parameters were set as follows: '- - outFilterMismatchNoverLmax 0.04 - - outSAMtype BAM SortedByCoordinate - - outFilterMultimapNmax 500 - - outMultimapperOrder Random - - outSAMmultNmax 1'. Gene expression was quantified using featureCounts v2.0.0. Heatmaps were created using R package pheatmap v1.0.12. The GO enrichment analysis was performed using the function "enrichGO" from the R package clusterProfiler v3.10.18879. + +<|ref|>sub_title<|/ref|><|det|>[[128, 374, 368, 392]]<|/det|> +## Organoid single-cell analysis + +<|ref|>text<|/ref|><|det|>[[125, 400, 881, 895]]<|/det|> +EE- Os were treated with \(10\mu \mathrm{M}\) TMG or vehicle control (0.1% DMSO). Following treatment, the organoids were dissociated into single cells using TrypLE digestion, and the mixture was passed through a \(40\mu \mathrm{m}\) cell strainer. The cells were then counted and viability assessed. Single- cell suspensions ( \(2\times 10^{5}\) cells/mL) in PBS (HyClone) were loaded onto microwell chip using the Singleron Matrix® Single Cell Processing System. Barcoding beads were subsequently collected from the microwell chip, followed by reverse transcription of the mRNA captured to obtain the cDNA. After PCR amplification, the amplified cDNA was then fragmented and ligated with sequencing adapters. The scRNA- seq libraries were constructed according to the protocol of the GEXSCOPE® Single Cell RNA Library Kits (Singleron)80. Individual libraries were diluted to \(4\mathrm{nM}\) , pooled, and sequenced on Illumina Novaseq 6000 with 150 bp paired end reads. Raw reads were processed to generate gene expression profiles using CeleScope v2.0.7 (Singleron) with default parameters. Briefly, barcodes and UMIs were extracted from R1 reads and corrected. Adapter sequences and polyA tails were trimmed from R2 reads and the trimmed R2 reads were aligned to the hg38 transcriptome using STAR (v2.6.1b). Uniquely mapped reads were then assigned to exons with featureCounts (v2.0.1). Successfully assigned reads with the same cell barcode, UMI and gene were grouped together to generate the gene expression matrix. Omicverse V1.5.4 was used for quality control, dimensionality reduction and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 883, 646]]<|/det|> +clustering under Python 3.8. The following criteria were used to filter the expression matrix: 1) cells with gene count less than 500 were excluded; 2) cells detected genes less than 250 were excluded; 3) cells with mitochondrial content more than \(15\%\) were excluded; 4) genes expressed in less than 3 cells were excluded. After filtering, 20736 cells were retained for the downstream analyses. The raw count matrix was normalized by total counts per cell and logarithmically transformed into normalized data matrix. The top 3000 highly variable genes were selected by setting flavor = 'seurat'. Principal Component Analysis (PCA) was performed on the scaled variable gene matrix, and 50 principal components were used for clustering and dimensional reduction. 'Harmony' was employed to integrate samples. Cells were separated into 6 clusters using Leiden algorithm with the resolution parameter at 0.25. Subsequently, omicverse was used to calculate the ranking of highly differential genes in each cluster to identify marker genes. Cell clusters were visualized using Minimum-Distortion Embedding (mde). Cell types were annotated based on the cell type auto- annotation tool SCSA, and the known cellular markers from the literature28,29,81- 83; epithelial cells (EPCAM, KRT8, KRT18), stem cells (LGR5, SOX9, POU5F1, PROM1, AXIN2), proliferative cells (MMP7, TOP2A, MK167), ciliated cells (PIFO, FOXJ1, TPPP3), pre- ciliated cells (CDC20B, DYDC2, CCNO), and inflammatory cells (IL4I1, IL32, S100A9, CD14, IL1RN). The O- GlcNAc related stem like cells annotation was mainly based on the results of KEGG pathway enrichment. The expression of markers used to identify each cell type was visualized using violin plot. + +<|ref|>text<|/ref|><|det|>[[127, 680, 425, 697]]<|/det|> +AUCell geneset enrichment analysis + +<|ref|>text<|/ref|><|det|>[[126, 707, 881, 810]]<|/det|> +Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were collected and used as functional genesets for AUCell scoring. AUCell scores of genesets were visualized using sc.pl.embedding. P- values from t tests were used for estimating the statistical significance between cell types and groups. + +<|ref|>text<|/ref|><|det|>[[127, 848, 282, 864]]<|/det|> +Statistical analysis + +<|ref|>text<|/ref|><|det|>[[126, 875, 880, 893]]<|/det|> +The experiments were conducted in at least three independent biological replicates, and the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 881, 226]]<|/det|> +data were presented as mean \(\pm\) SD. If not specified, the Student's t test was used to perform a statistical significance test between different groups, and \(\mathrm{P}< 0.05\) was considered significant. Overall survival curves were estimated by the Kaplan- Meier method and Cox proportional hazards model. All statistical and correlation analyses were performed using the GraphPad Prism 8.0 software (GraphPad Software) and SPSS 26.0 (SPSS Software). + +<|ref|>sub_title<|/ref|><|det|>[[128, 253, 396, 271]]<|/det|> +## Data and materials availability + +<|ref|>text<|/ref|><|det|>[[128, 281, 780, 300]]<|/det|> +All data and materials will be made available after acceptance of the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[128, 328, 297, 345]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[125, 355, 881, 540]]<|/det|> +We gratefully acknowledge Drs. Daan van Aalten, Kum Kum Khanna, Xiaowei Yang, Timothy Mitchison, Xuebiao Yao, Chao Xu, Cuiting Yong, Lisha Wu, Wenqing Yang, and Hongqiang Qin for reagents or inspiring discussions. This project has been supported by the National Natural Science Foundation of China (grants 92153301, 32170821, and 32370821 to K.Y, 32101034 to F.C), National Key Research and Development Program of China (2021YFC2701200), Department of Science & Technology of Hunan Province (grants 2023RC1028, 2023SK2091, and 2021JJ10054 to K.Y). + +<|ref|>sub_title<|/ref|><|det|>[[127, 580, 252, 596]]<|/det|> +## Contributions + +<|ref|>text<|/ref|><|det|>[[125, 606, 881, 764]]<|/det|> +Conceptualization: K.Y.; Methodology: N.Z., Y.M., H.N., C.H., L.S., L.L., C.Y., S.M., F.C., Y.Z., K.Y.; Validation: H.N., C.H., L.S., K.F.; Software: N.Z., Y.M., S.M.; Formal Analysis: N.Z., Y.M., K.Y.; Investigation: N.Z., Y.M., H.N., S.M., K.Y.; Resources: Y.Z., K.Y.; Data Curation: N.Z., Y.M., H.N.; Writing- Original Draft: N.Z., K.Y.; Writing- Review & Editing: K.Y.; Visualization: N.Z., Y.M., S.M., K.Y.; Supervision: K.Y.; Project Administration: L.L., K.Y.; Funding Acquisition: Y.Z., K.Y. + +<|ref|>sub_title<|/ref|><|det|>[[127, 801, 303, 818]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[127, 829, 480, 847]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 103, 884, 900]]<|/det|> +Yang, X. & Qian, K. Protein O-GlcNAcylation: emerging mechanisms and functions. Nat Rev Mol Cell Biol 18, 452- 465, doi:10.1038/nrm.2017.22 (2017).Zachara, N. E. & Hart, G. W. O-GlcNAc a sensor of cellular state: the role of nucleocytoplasmic glycosylation in modulating cellular function in response to nutrition and stress. Biochim Biophys Acta 1673, 13- 28, doi:10.1016/j.bbagen.2004.03.016 (2004).Chatham, J. C., Zhang, J. & Wende, A. R. Role of O- Linked N- Acetylglucosamine Protein Modification in Cellular (Patho)Physiology. Physiol Rev 101, 427- 493, doi:10.1152/physrev.00043.2019 (2021).Sheikh, M. A., Emerald, B. S. & Ansari, S. A. Stem cell fate determination through protein O-GlcNAcylation. J Biol Chem 296, 100035, doi:10.1074/jbc.REV120.014915 (2021).Slawson, C. & Hart, G. W. O- GlcNAc signalling: implications for cancer cell biology. Nat Rev Cancer 11, 678- 684, doi:10.1038/nrc3114 (2011).de Queiroz, R. M., Carvalho, E. & Dias, W. B. O- GlcNAcylation: The Sweet Side of the Cancer. Front Oncol 4, 132, doi:10.3389/fonc.2014.00132 (2014).Trinca, G. M. & Hagan, C. R. O- GlcNAcylation in women's cancers: breast, endometrial and ovarian. J Bioenerg Biomembr 50, 199- 204, doi:10.1007/s10863- 017- 9730- z (2018).Le Minh, G., Esquea, E. M., Young, R. G., Huang, J. & Reginato, M. J. On a sugar high: Role of O- GlcNAcylation in cancer. J Biol Chem 299, 105344, doi:10.1016/j.jbc.2023.105344 (2023).Yang, Y., Yin, X., Yang, H. & Xu, Y. Histone demethylase LSD2 acts as an E3 ubiquitin ligase and inhibits cancer cell growth through promoting proteasomal degradation of OGT. Mol Cell 58, 47- 59, doi:10.1016/j.molcel.2015.01.038 (2015).Seo, H. G. et al. Mutual regulation between OGT and XIAP to control colon cancer cell growth and invasion. Cell Death Dis 11, 815, doi:10.1038/s41419- 020- 02999- 5 (2020).Peng, K. et al. Regulation of O- Linked N- Acetyl Glucosamine Transferase (OGT) through E6 Stimulation of the Ubiquitin Ligase Activity of E6AP. Int J Mol Sci 22, doi:10.3390/ijms221910286 (2021).Tang, J. et al. Targeting USP8 Inhibits O- GlcNAcylation of SLC7A11 to Promote Ferroptosis of Hepatocellular Carcinoma via Stabilization of OGT. Adv Sci (Weinh) 10, e2302953, doi:10.1002/advs.202302953 (2023).Tang, J. et al. The deubiquitinase EIF3H promotes hepatocellular carcinoma progression by stabilizing OGT and inhibiting ferroptosis. Cell Commun Signal 21, 198, doi:10.1186/s12964- 023- 01220- 2 (2023).Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 71, 209- 249, doi:10.3322/caac.21660 (2021).Han, B. et al. Cancer incidence and mortality in China, 2022. Journal of the National Cancer Center, doi:10.1016/j.jncc.2024.01.006 (2024). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 92, 884, 888]]<|/det|> +768 16 Bokhman, J. V. Two pathogenetic types of endometrial carcinoma. Gynecol Oncol 15, 10- 17, doi:10.1016/0090- 8258(83)90111- 7 (1983). 769 17 Murali, R., Soslow, R. A. & Weigelt, B. Classification of endometrial carcinoma: more than two types. The Lancet Oncology 15, e268- e278, doi:10.1016/s1470- 2045(13)70591- 6 (2014). 772 18 Cancer Genome Atlas Research, N. et al. Integrated genomic characterization of endometrial carcinoma. Nature 497, 67- 73, doi:10.1038/nature12113 (2013). 774 19 Jamaluddin, M. F. B. et al. Proteomic and functional characterization of intra- tumor heterogeneity in human endometrial cancer. Cell Rep Med 3, 100738, doi:10.1016/j.xcrm.2022.100738 (2022). 778 20 Dou, Y. et al. Proteogenomic insights suggest druggable pathways in endometrial carcinoma. Cancer Cell 41, 1586- 1605 e1515, doi:10.1016/j.ccell.2023.07.007 (2023). 781 21 Jaskiewicz, N. M. & Townson, D. H. Hyper- O- GlcNAcylation promotes epithelial- mesenchymal transition in endometrial cancer cells. Oncotarget 10, 2899- 2910, doi:10.18632/oncotarget.26884 (2019). 784 22 Krzeslak, A., Wojcik- Krowiranda, K., Forma, E., Bienkiewicz, A. & Brys, M. Expression of genes encoding for enzymes associated with O- GlcNAcylation in endometrial carcinomas: clinicopathologic correlations. Ginekol Pol 83, 22- 26 (2012). 788 23 Zhai, L. et al. O- GlcNAcylation mediates endometrial cancer progression by regulating the Hippo- YAP pathway. Int J Oncol 63, doi:10.3892/ijo.2023.5538 (2023). 791 24 Zhou, F. et al. Elevated glucose levels impair the WNT/beta- catenin pathway via the activation of the hexosamine biosynthesis pathway in endometrial cancer. J Steroid Biochem Mol Biol 159, 19- 25, doi:10.1016/j.jsbmb.2016.02.015 (2016). 793 25 Ciesielski, P., Jozwiak, P., Forma, E. & Krzeslak, A. TET3- and OGT- Dependent Expression of Genes Involved in Epithelial- Mesenchymal Transition in Endometrial Cancer. Int J Mol Sci 22, doi:10.3390/ijms222413239 (2021). 796 26 Brooks, R. A. et al. Current recommendations and recent progress in endometrial cancer. CA Cancer J Clin 69, 258- 279, doi:10.3322/caac.21561 (2019). 798 27 Turco, M. Y. et al. Long- term, hormone- responsive organoid cultures of human endometrium in a chemically defined medium. Nat Cell Biol 19, 568- 577, doi:10.1038/ncb3516 (2017). 802 28 Fitzgerald, H. C., Dhakal, P., Behura, S. K., Schust, D. J. & Spencer, T. E. Self- renewing endometrial epithelial organoids of the human uterus. Proc Natl Acad Sci U S A 116, 23132- 23142, doi:10.1073/pnas.1915389116 (2019). 805 29 Garcia- Alonso, L. et al. Mapping the temporal and spatial dynamics of the human endometrium in vivo and in vitro. Nat Genet 53, 1698- 1711, doi:10.1038/s41588- 021- 00972- 2 (2021). 808 30 Li, W. et al. MAGeCK enables robust identification of essential genes from genome- scale CRISPR/Cas9 knockout screens. Genome Biol 15, 554, doi:10.1186/s13059- 014- 0554- 4 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 90, 884, 888]]<|/det|> +811 31 Kawauchi, K., Araki, K., Tobiume, K. & Tanaka, N. Loss of p53 enhances catalytic activity of IKKbeta through O-linked beta-N-acetyl glucosamine modification. Proc Natl Acad Sci U S A 106, 3431- 3436, doi:10.1073/pnas.0813210106 (2009). 814 32 Zhang, F., Snead, C. M. & Catravas, J. D. Hsp90 regulates O-linked beta-N-acetylglucosamine transferase: a novel mechanism of modulation of protein O-linked beta-N-acetylglucosamine modification in endothelial cells. Am J Physiol Cell Physiol 302, C1786- 1796, doi:10.1152/ajpcell.00004.2012 (2012). 818 33 Deplus, R. et al. TET2 and TET3 regulate GlcNAcylation and H3K4 methylation through OGT and SET1/COMPASS. EMBO J 32, 645- 655, doi:10.1038/emboj.2012.357 (2013). 821 34 Ledee, D. et al. c- Myc Alters Substrate Utilization and O-GlcNAc Protein Posttranslational Modifications without Altering Cardiac Function during Early Aortic Constriction. PLoS One 10, e0135262, doi:10.1371/journal.pone.0135262 (2015). 825 35 Li, Y. N., Hu, J. A. & Wang, H. M. Inhibition of HIF- 1alpha Affects Autophagy Mediated Glycosylation in Oral Squamous Cell Carcinoma Cells. Dis Markers 2015, 239479, doi:10.1155/2015/239479 (2015). 828 36 Muthusamy, S., Hong, K. U., Dassanayaka, S., Hamid, T. & Jones, S. P. E2F1 Transcription Factor Regulates O-linked N-acetylglucosamine (O-GlcNAc) Transferase and O-GlcNAcase Expression. J Biol Chem 290, 31013- 31024, doi:10.1074/jbc.M115.677534 (2015). 832 37 Sodi, V. L. et al. mTOR/MYC Axis Regulates O-GlcNAc Transferase Expression and O-GlcNAcylation in Breast Cancer. Mol Cancer Res 13, 923- 933, doi:10.1158/1541- 7786.MCR- 14- 0536 (2015). 835 38 Zhang, X. et al. MAPK/ERK signaling pathway- induced hyper- O- GlcNAcylation enhances cancer malignancy. Mol Cell Biochem 410, 101- 110, doi:10.1007/s11010- 015- 2542- 8 (2015). 838 39 Li, T. et al. Defective Branched- Chain Amino Acid Catabolism Disrupts Glucose Metabolism and Sensitizes the Heart to Ischemia- Reperfusion Injury. Cell Metab 25, 374- 385, doi:10.1016/j.cmet.2016.11.005 (2017). 841 40 Zibrova, D. et al. GFAT1 phosphorylation by AMPK promotes VEGF- induced angiogenesis. Biochem J 474, 983- 1001, doi:10.1042/BCJ20160980 (2017). 843 41 Berthier, A. et al. Combinatorial regulation of hepatic cytoplasmic signaling and nuclear transcriptional events by the OGT/REV- ERBalpha complex. Proc Natl Acad Sci U S A 115, E11033- E11042, doi:10.1073/pnas.1805397115 (2018). 846 42 Lai, C. Y. et al. Identification of UAP1L1 as a critical factor for protein O- GlcNAcylation and cell proliferation in human hepatoma cells. Oncogene 38, 317- 331, doi:10.1038/s41388- 018- 0442- 6 (2019). 849 43 Deng, X. et al. ROCK2 mediates osteosarcoma progression and TRAIL resistance by modulating O- GlcNAc transferase degradation. Am J Cancer Res 10, 781- 798 (2020). 852 44 Lu, S. et al. SIRT1 regulates O- GlcNAcylation of tau through OGT. Aging (Albany NY) 12, 7042- 7055, doi:10.18632/aging.103062 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 90, 884, 905]]<|/det|> +854 45 Paredes, F., Williams, H. C., Quintana, R. A. & San Martin, A. Mitochondrial Protein 855 Poldip2 (Polymerase Delta Interacting Protein 2) Controls Vascular Smooth Muscle 856 Differentiated Phenotype by O-Linked GlcNAc (N-Acetylglucosamine) 857 Transferase-Dependent Inhibition of a Ubiquitin Proteasome System. Circ Res 126, 858 41- 56, doi:10.1161/CIRCRESAHA.119.315932 (2020). 859 46 Walter, L. A. et al. Inhibiting the Hexosamine Biosynthetic Pathway Lowers 860 O-GlcNAcylation Levels and Sensitizes Cancer to Environmental Stress. 861 Biochemistry 59, 3169- 3179, doi:10.1021/acs.biochem.9b00560 (2020). 862 47 Zhao, M., Sun, J. & Zhao, Z. TSGene: a web resource for tumor suppressor genes. 863 Nucleic Acids Res 41, D970- 976, doi:10.1093/nar/gks937 (2013). 864 48 Zhao, M., Kim, P., Mitra, R., Zhao, J. & Zhao, Z. TSGene 2.0: an updated 865 literature- based knowledgebase for tumor suppressor genes. Nucleic Acids Res 44, 866 D1023- 1031, doi:10.1093/nar/gkv1268 (2016). 867 49 Santra, M. K., Wajapeyee, N. & Green, M. R. F- box protein FBXO31 mediates 868 cyclin D1 degradation to induce G1 arrest after DNA damage. Nature 459, 722- 725, 869 doi:10.1038/nature08011 (2009). 870 50 Johansson, P. et al. SCF- FBXO31 E3 ligase targets DNA replication factor Cdt1 for 871 proteolysis in the G2 phase of cell cycle to prevent re- replication. J Biol Chem 289, 872 18514- 18525, doi:10.1074/jbc.M114.559930 (2014). 873 51 Liu, J. et al. F- box only protein 31 (FBXO31) negatively regulates p38 874 mitogen- activated protein kinase (MAPK) signaling by mediating lysine 48- linked 875 ubiquitination and degradation of mitogen- activated protein kinase kinase 6 (MKK6). 876 J Biol Chem 289, 21508- 21518, doi:10.1074/jbc.M114.560342 (2014). 877 52 Malonia, S. K., Dutta, P., Santra, M. K. & Green, M. R. F- box protein FBXO31 878 directs degradation of MDM2 to facilitate p53- mediated growth arrest following 879 genotoxic stress. Proc Natl Acad Sci U S A 112, 8632- 8637, 880 doi:10.1073/pnas.1510929112 (2015). 881 53 Jeffery, J. M. et al. FBXO31 protects against genomic instability by capping FOXM1 882 levels at the G2/M transition. Oncogene 36, 1012- 1022, doi:10.1038/onc.2016.268 883 (2017). 884 54 Duan, S. et al. Loss of FBXO31- mediated degradation of DUSP6 dysregulates ERK 885 and PI3K- AKT signaling and promotes prostate tumorigenesis. Cell Rep 37, 109870, 886 doi:10.1016/j.celrep.2021.109870 (2021). 887 55 Tan, Y., Liu, D., Gong, J., Liu, J. & Huo, J. The role of F- box only protein 31 in 888 cancer. Oncol Lett 15, 4047- 4052, doi:10.3892/ol.2018.7816 (2018). 889 56 Tekcham, D. S. et al. F- box proteins and cancer: an update from functional and 890 regulatory mechanism to therapeutic clinical prospects. Theranostics 10, 4150- 4167, 891 doi:10.7150/thno.42735 (2020). 892 57 Dutta, P. et al. The tumor suppressor FBXO31 preserves genomic integrity by 893 regulating DNA replication and segregation through precise control of cyclin A 894 levels. J Biol Chem 294, 14879- 14895, doi:10.1074/jbc.RA118.007055 (2019). 895 58 Islam, S. et al. Feedback- regulated transcriptional repression of FBXO31 by c- Myc 896 triggers ovarian cancer tumorigenesis. Int J Cancer 150, 1512- 1524, 897 doi:10.1002/ijc.33854 (2022). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 90, 884, 907]]<|/det|> +Baek, D. et al. Ubiquitin-specific protease 53 promotes osteogenic differentiation of human bone marrow-derived mesenchymal stem cells. Cell Death Dis 12, 238, doi:10.1038/s41419-021-03517-x (2021). +Zou, S. et al. FBXO31 Suppresses Gastric Cancer EMT by Targeting Snail1 for Proteasomal Degradation. Mol Cancer Res 16, 286-295, doi:10.1158/1541-7786.MCR-17-0432 (2018). +Manne, R. K. et al. A MicroRNA/Ubiquitin Ligase Feedback Loop Regulates Slug-Mediated Invasion in Breast Cancer. Neoplasia 19, 483-495, doi:10.1016/j.neo.2017.02.013 (2017). +Zhu, Z. et al. FBXO31 sensitizes cancer stem cells-like cells to cisplatin by promoting ferroptosis and facilitating proteasomal degradation of GPX4 in cholangiocarcinoma. Liver Int 42, 2871-2888, doi:10.1111/liv.15462 (2022). +Wulff-Fuentes, E. et al. The human O-GlcNAcome database and meta-analysis. Sci Data 8, 25, doi:10.1038/s41597-021-00810-4 (2021). +Yang, W. H. et al. Modification of p53 with O-linked N-acetylglucosamine regulates p53 activity and stability. Nat Cell Biol 8, 1074-1083, doi:10.1038/ncb1470 (2006). +Chou, T. Y., Hart, G. W. & Dang, C. V. c-Myc is glycosylated at threonine 58, a known phosphorylation site and a mutational hot spot in lymphomas. J Biol Chem 270, 18961-18965, doi:10.1074/jbc.270.32.18961 (1995). +Zhu, Y. & Hart, G. W. Dual-specificity RNA aptamers enable manipulation of target-specific O-GlcNAcylation and unveil functions of O-GlcNAc on beta-catenin. Cell 186, 428-445 e427, doi:10.1016/j.cell.2022.12.016 (2023). +Jang, H. et al. O-GlcNAc Regulates Pluripotency and Reprogramming by Directly Acting on Core Components of the Pluripotency Network. Cell Stem Cell 11, 62-74, doi:10.1016/j.stem.2012.03.001 (2012). +Kim, D. K. et al. O-GlcNAcylation of Sox2 at threonine 258 regulates the self-renewal and early cell fate of embryonic stem cells. Experimental & Molecular Medicine 53, 1759-1768, doi:10.1038/s12276-021-00707-7 (2021). +Sun, C. et al. Glucose regulates tissue-specific chondro-osteogenic differentiation of human cartilage endplate stem cells via O-GlcNAcylation of Sox9 and Runx2. Stem Cell Res Ther 10, 357, doi:10.1186/s13287-019-1440-5 (2019). +Boretto, M. et al. Patient-derived organoids from endometrial disease capture clinical heterogeneity and are amenable to drug screening. Nat Cell Biol 21, 1041-1051, doi:10.1038/s41556-019-0360-z (2019). +Liu, H. et al. Transcriptional pausing induced by ionizing radiation enables the acquisition of radioresistance in nasopharyngeal carcinoma. J Mol Cell Biol, doi:10.1093/jmcb/mjad044 (2023). +Wu, N. et al. O-GlcNAcylation promotes colorectal cancer progression by regulating protein stability and potential carcinogenic function of DDX5. J Cell Mol Med 23, 1354-1362, doi:10.1111/jcmm.14038 (2019). +Dekkers, J. F. et al. High-resolution 3D imaging of fixed and cleared organoids. Nat Protoc 14, 1756-1771, doi:10.1038/s41596-019-0160-8 (2019). +Choo, Y. S. & Zhang, Z. Detection of protein ubiquitination. J Vis Exp, doi:10.3791/1293 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 90, 884, 568]]<|/det|> +942 75 Gundogdu, M. et al. The O-GlcNAc Transferase Intellectual Disability Mutation 943 L254F Distorts the TPR Helix. Cell Chem Biol 25, 513-518 e514, 944 doi:10.1016/j.chembiol.2018.03.004 (2018). 945 76 Sanjana, N. E., Shalem, O. & Zhang, F. Improved vectors and genome-wide libraries 946 for CRISPR screening. Nat Methods 11, 783-784, doi:10.1038/nmeth.3047 (2014). 947 77 Shalem, O. et al. Genome-scale CRISPR-Cas9 knockout screening in human cells. 948 Science 343, 84-87, doi:10.1126/science.1247005 (2014). 949 78 Liu, J. et al. An Integrated TCGA Pan-Cancer Clinical Data Resource to Drive 950 High-Quality Survival Outcome Analytics. Cell 173, 400-416 e411, 951 doi:10.1016/j.cell.2018.02.052 (2018). 952 79 Yu, C. et al. ARID1A loss derepresses a group of human endogenous retrovirus-H 953 loci to modulate BRD4-dependent transcription. Nat Commun 13, 3501, 954 doi:10.1038/s41467-022-31197-4 (2022). 955 80 Dura, B. et al. scFTD-seq: freeze-thaw lysis based, portable approach toward highly 956 distributed single-cell 3' mRNA profiling. Nucleic Acids Res 47, e16, 957 doi:10.1093/nar/gky1173 (2019). 958 81 Lai, Z. Z. et al. Single-cell transcriptome profiling of the human endometrium of 959 patients with recurrent implantation failure. Theranostics 12, 6527-6547, 960 doi:10.7150/thno.74053 (2022). 961 82 Tan, Y. et al. Single-cell analysis of endometriosis reveals a coordinated 962 transcriptional programme driving immunotolerance and angiogenesis across eutopic 963 and ectopic tissues. Nat Cell Biol 24, 1306-1318, doi:10.1038/s41556-022-00961-5 964 (2022). 965 83 Fonseca, M. A. S. et al. Single-cell transcriptomic analysis of endometriosis. Nat 966 Genet 55, 255-267, doi:10.1038/s41588-022-01254-1 (2023). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 150, 888, 846]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[147, 118, 864, 140]]<|/det|> +
Fig.1: O-GlcNAc level correlates with endometrial cancer grading
+<|ref|>image_caption<|/ref|><|det|>[[125, 846, 735, 864]]<|/det|> +
Fig. 1 Correlative analysis of O-GlcNAc level with clinical parameters.
+ +<|ref|>text<|/ref|><|det|>[[125, 874, 880, 893]]<|/det|> +(a) A flowchart illustrating the process of clinical sample selection, data collection, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[125, 95, 880, 142]]<|/det|> +analysis. All samples were derived from patients receiving their initial treatment, and none of the patients had concurrent or previous tumors. + +<|ref|>text<|/ref|><|det|>[[125, 151, 880, 199]]<|/det|> +(b- c) Representative images depicting IHC staining of \(O\) - GlcNAcylation (RL2) and OGT in EC and adjacent normal tissues on the FFPE tissue array. Scale bars: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[125, 207, 881, 308]]<|/det|> +(d- e) Quantitative analysis of the levels of \(O\) - GlcNAcylation (RL2) and OGT in the EC tissue arrays. The levels of \(O\) - GlcNAcylation and OGT were assessed semi- quantitatively based on both the intensity and area of the stainings. Statistical significance was calculated using unpaired two- tailed Student's t- test, \(* \mathrm{P} < 0.05\) , \(** \mathrm{P} < 0.01\) . + +<|ref|>text<|/ref|><|det|>[[125, 317, 881, 393]]<|/det|> +(f) Representative images of IHC staining showing varying levels of \(O\) - GlcNAcylation in serial sections of EC tissues with different histologic grades (well differentiated G1, moderately differentiated G2, and poorly differentiated G3). Scale bar: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[125, 401, 881, 504]]<|/det|> +(g) Percentage of samples with high or low levels of \(O\) - GlcNAcylation in different histologic grade groups. High and low categories were determined using a scoring system (high score: 8-12; low score: 0-6). Statistical significance between groups was calculated using Fisher's exact test, \(*** \mathrm{P} < 0.0001\) . + +<|ref|>text<|/ref|><|det|>[[125, 512, 880, 586]]<|/det|> +(h- i) Kaplan- Meier survival curves of PFS and OS of the EC patients stratified by the levels of \(O\) - GlcNAcylation derived from IHC scores. Statistical significance was determined by the log- rank test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 128, 888, 776]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[180, 92, 808, 113]]<|/det|> +
Fig.S1: Construction of a gene signature for O-GlcNAc level
+ +<|ref|>text<|/ref|><|det|>[[125, 789, 650, 808]]<|/det|> +Fig. S1 Construction of a gene signature for O-GlcNAc level. + +<|ref|>text<|/ref|><|det|>[[125, 810, 880, 858]]<|/det|> +(a) Representative images depicting IHC staining of OGA in EC and adjacent normal tissues on the FFPE tissue array. Scale bar: \(50 \mu \mathrm{m}\). + +<|ref|>text<|/ref|><|det|>[[125, 865, 848, 884]]<|/det|> +(b) Quantitative analysis of the level of OGA in EC tissue array. The expression level of + +<|ref|>text<|/ref|><|det|>[[125, 892, 812, 911]]<|/det|> +OGA was assessed semi-quantitatively based on both the staining intensity and area. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 95, 870, 368]]<|/det|> +996 Statistical significance was calculated using unpaired two- tailed Student's t- test, ns: not 997 significant. 998 (c- d) Kaplan- Meier analysis of the OS of EC patients based on OGA or OGT expression 999 levels in TCGA (http://kmplot.com/analysis/). EC cases were stratified using the median 1000 cut- off, and statistical significance was determined using the log- rank test. 1001 (e) Flow chart of sample selection and data process for constructing an O- GlcNAc level 1002 prediction model based on transcriptomic data. 1003 (f) Heatmap showing the expression levels of the 1000 O- GlcNAc correlated genes derived 1004 from RNA- seq of the selected EC tissues (n = 55). 1005 (g) Gene Ontology (GO) enrichment analysis of the 1000 O- GlcNAc correlated genes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 125, 875, 856]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[208, 92, 784, 113]]<|/det|> +
Fig.2: Validation with TCGA endometrial cancer dataset
+<|ref|>image_caption<|/ref|><|det|>[[125, 860, 620, 877]]<|/det|> +
Fig. 2 Validation with TCGA endometrial cancer dataset.
+ +<|ref|>text<|/ref|><|det|>[[125, 885, 880, 904]]<|/det|> +(a) Heatmap displaying the expression profiles of the 1000 O-GlcNAc correlated genes in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[70, 93, 883, 420]]<|/det|> +1009 the TCGA UCEC RNA- seq dataset (n = 589). The EC samples are annotated by clinical parameters, including BMI, menopause status, diabetes, histologic grades, molecular subtypes (integrative cluster), FIGO stage, age, and primary diagnosis. Patients were categorized into O- GlcNAc high or O- GlcNAc low group using the median of the calculated O- GlcNAc index as the threshold. The '*' symbol indicates a statistically significant difference of the calculated O- GlcNAc index among the patients' groups according to the indicated clinical parameter. Wilcoxon test, \(**P < 0.01\) , \(****P < 0.0001\) .(b- f) The O- GlcNAc index in different EC groups stratified by histologic grade, FIGO stage, integrative cluster, age, or diabetes in the TCGA UCEC dataset. Wilcoxon test, \(*P < 0.05\) , \(**P < 0.01\) , \(***P < 0.001\) , \(****P < 0.0001\) .(g- h) Kaplan- Meier survival curves for Progression- free interval (PFI) and OS of EC groups with high or low O- GlcNAc index in the TCGA UCEC dataset. Statistical significance was determined by the log- rank test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[160, 120, 842, 796]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[163, 95, 835, 115]]<|/det|> +
Fig.3: Increase of O-GlcNAc level promotes proliferation and stemness
+ +<|ref|>text<|/ref|><|det|>[[125, 828, 880, 900]]<|/det|> +(a) Mucin detection (Alcian blue staining) and IHC examination of endometrial markers (ER and PR) in primary endometrial tissue and corresponding EE-Os. Scale bars: \(50 \mu \mathrm{m}\). +(b) Immunoblot with RL2 antibody assessing O-GlcNAc levels in the EE-Os treated with + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[77, 95, 881, 141]]<|/det|> +DMSO, \(5 \mu \mathrm{M}\) , or \(10 \mu \mathrm{M}\) Thiamet- G (TMG) for 48 hours. Tubulin was used as the loading control. + +<|ref|>text<|/ref|><|det|>[[125, 152, 881, 226]]<|/det|> +(c) EE-Os bright-field images depicting responses to TMG at two different time points (day 1 and day 3). Representative images from control (DMSO) and \(10 \mu \mathrm{M}\) TMG treated EE-Os groups are presented. Scale bar: \(50 \mu \mathrm{m}\). + +<|ref|>text<|/ref|><|det|>[[125, 234, 881, 310]]<|/det|> +(d) Comparison of the numbers of EE-Os at day 3 of culture after treatment with \(10 \mu \mathrm{M}\) TMG versus control (DMSO). The results are presented as the mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +<|ref|>text<|/ref|><|det|>[[125, 318, 881, 418]]<|/det|> +(e) Measurement of cross-sectional area of EE-Os at day 3 of culture after treatment with \(10 \mu \mathrm{M}\) TMG compared with control (DMSO). The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +<|ref|>text<|/ref|><|det|>[[125, 428, 881, 504]]<|/det|> +(f) Representative immunofluorescence images of control and TMG-treated EE-Os. EE-O cells are stained with PH3 (red) and Tubulin (green) antibodies. Nuclei are visualized with DAPI (blue), and F-actin is labeled by Phalloidine (magenta). Scale bar: \(5 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[125, 512, 881, 586]]<|/det|> +(g) Quantification of the number of PH3 positive (PH3+) cells in each EE-Os. The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{*P < 0.05}\) . + +<|ref|>text<|/ref|><|det|>[[125, 595, 881, 726]]<|/det|> +(h) Representative immunofluorescence images of control and TMG-treated EE-Os. Ciliated epithelium is labeled by acetylated alpha-tubulin (Ac-tubulin, green), and secretory cells labeled by PAEP (red). Nuclei are visualized with DAPI (blue), and F-actin is labeled by Phalloidine (magenta). Scale bar: \(50 \mu \mathrm{m}\) . The right insets display a magnification of the area in the white box, scale bar: \(5 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[125, 735, 881, 808]]<|/det|> +(i) Quantification of the number of ciliated cells (Ac-tubulin+) in each EE-Os. The results are presented as mean \(\pm\) SD. Statistical significance was determined by unpaired two-tailed Student's t-test, \(\mathrm{***P < 0.0001}\) . + +<|ref|>text<|/ref|><|det|>[[125, 817, 744, 837]]<|/det|> +(j) MDE projections of scRNA-seq data of control and TMG treated EE-Os. + +<|ref|>text<|/ref|><|det|>[[125, 845, 880, 892]]<|/det|> +(k) Subclustered epithelial populations of EE-Os (left), and the proportion of each cell types in control and TMG-treated groups (right). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[130, 115, 880, 860]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[125, 87, 844, 110]]<|/det|> +
Fig.S2: Generation of EE-O and EC-O, and TMG treatment on EE-O
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 95, 825, 115]]<|/det|> +(a) Bright-field and fluorescent images of 3D EE-Os and EC-Os. F-actin is labeled by + +<|ref|>text<|/ref|><|det|>[[120, 123, 830, 143]]<|/det|> +Phalloidine (golden yellow). Nuclei are labeled using DAPI (cyan). Scale bars: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[120, 151, 860, 199]]<|/det|> +(b) Representative pictures of IHC staining for \(O\) -GlcNAcylation levels in primary tissues and their corresponding organoids. Scale bar: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[120, 207, 858, 305]]<|/det|> +(c) qPCR data of stemness markers expression in EE-O treated with TMG or DMSO (control), normalized to actin mRNA levels. The results are presented as mean \(\pm\) SD from three independent experiments. Statistical significance was determined by unpaired two-tailed Student's t-test, \(^{*}\mathrm{P}< 0.05\) , \(^{**} \mathrm{P}< 0.01\) . + +<|ref|>text<|/ref|><|det|>[[120, 316, 820, 363]]<|/det|> +(d) Feature plots of representative gene expressions in different epithelial subclusters resolved by scRNA-seq analysis. + +<|ref|>text<|/ref|><|det|>[[120, 372, 675, 392]]<|/det|> +(e) Visualization of the indicated KEGG pathway on the MDE plot. + +<|ref|>text<|/ref|><|det|>[[120, 400, 857, 448]]<|/det|> +(f) MDE plot revealing 6 different subclusters of epithelial cells and their cell numbers in the EE-Os subject to scRNA-seq analysis. + +<|ref|>text<|/ref|><|det|>[[120, 456, 875, 475]]<|/det|> +(g) Expression levels of differentially expressed genes in the 6 subclusters of epithelial cells. + +<|ref|>text<|/ref|><|det|>[[120, 483, 870, 501]]<|/det|> +The size and color of each dot represent the expression level and cell fraction of the marker + +<|ref|>text<|/ref|><|det|>[[120, 511, 181, 528]]<|/det|> +genes. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 115, 875, 855]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[130, 92, 866, 113]]<|/det|> +
Fig.4: Decrease of O-GlcNAc level induces differentiation and cell death
+ +<|ref|>text<|/ref|><|det|>[[125, 884, 878, 903]]<|/det|> +(a) Mucin detection (Alcian blue staining) and IHC examination of EC markers in primary + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[78, 95, 881, 895]]<|/det|> +EC tissues and the corresponding EC- Os. Scale bar: \(50\mu \mathrm{m}\) .(b) Immunoblot assessing \(O\) - GlcNAc levels in the EC- Os treated with 25 or \(50\mu \mathrm{M}\) OSMI- 1 for 48 hours. Actin was used as the loading control.(c) Representative 3D EC- O bright- field images depicting the responses to \(50\mu \mathrm{M}\) OSMI- 1 treatment at two different time points (day 1 and day 3). Scale bar: \(50\mu \mathrm{m}\) .(d) Comparison of the numbers of EC- Os at day 3 of culture after treatment with OSMI- 1 versus the control (DMSO). The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.001}\) .(e) Analysis of the EC- Os cross- sectional area at day 3 of culture after treatment with OSMI- 1 compared with control (DMSO). The results are presented as mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.0001}\) .(f) TUNEL staining demonstrating the existence of apoptotic cells in EC- Os after treated with \(50\mu \mathrm{M}\) OSMI- 1. Nuclei are visualized with DAPI (blue). Scale bar: \(50\mu \mathrm{m}\) .(g) Representative immunofluorescence images of control and OSMI- 1 treated EC- Os. Mitotic cells are marked by PH3 (red). Tubulin staining is shown in green, DAPI labeled nuclei in blue, and Phalloidine labeled F- actin in magenta. Scale bar: \(5\mu \mathrm{m}\) .(h) Quantification of the number of PH3+ cells. Each dot represents one 3D EC- O. The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.0001}\) .(i) Representative immunofluorescence images of control and OSMI- 1 treated EC- Os. Ciliated epithelial cells are labeled by acetylated alpha- tubulin (Ac- tubulin, green), and secretory cells are labeled by PAEP (red). Nuclei are visualized with DAPI (blue), and F- actin with Phalloidine (magenta). Scale bar: \(50\mu \mathrm{m}\) . The right insets display a magnification of the area in the white box, scale bar: \(5\mu \mathrm{m}\) .(j) Quantification of the number of Ac- tubulin+ cells. Each dot represents one 3D EC- O. The results are reported as the mean \(\pm\) SD. Unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.01}\) .(k) qPCR data of stemness markers expression in EC- Os treated with OSMI- 1 or DMSO (control), normalized to actin mRNA levels. The results are presented as mean \(\pm\) SD from three independent experiments. Statistical significance was determined by unpaired two- tailed Student's t- test, \(\mathrm{***P< 0.01}\) , \(\mathrm{***P< 0.001}\) , \(\mathrm{***P< 0.0001}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[152, 110, 878, 840]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[196, 88, 815, 109]]<|/det|> +
Fig.5: Screen for TSGs that regulate O-GlcNAc homeostasis
+ +<|ref|>text<|/ref|><|det|>[[125, 866, 880, 911]]<|/det|> +(a) Schematic representation of the FACS-based genome-wide CRISPR-Cas9 screen for putative regulators of O-GlcNAc homeostasis. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 95, 884, 450]]<|/det|> +1110 (b) Validation the sensitivity of RL2 staining (red) with 293T cells transfected with OGT (green). Nuclei are labeled with DAPI (blue). Scale bar: \(5 \mu \mathrm{m}\) .1112 (c) Genes plotted according to their relative ranking analysis (RRA) enrichment scores, with known O- GlcNAc regulators highlighted in red and blue.1114 (d) KEGG analysis showing enrichment of putative O- GlcNAc regulators in the indicated pathways. Analysis was performed on the 1038 top scoring genes ( \(P < 0.05\) ).1116 (e) Venn diagram showing the overlap between the 526 Human UCEC TSGs and the 1038 high- confidence genes from the O- GlcNAc screen.1118 (f) Immunofluorescent detection of O- GlcNAc level by RL2 (red) in WT and FBXO31 KO 293T cells. Nuclei were stained with DAPI (blue). Scale bar: \(5 \mu \mathrm{m}\) .1120 (g) Kaplan- Meier analysis of the OS of the EC patients stratified by the expression levels of FBXO31 (http://kmlpot.com/analysis/). EC cases were stratified using the median cut- off, and statistical significance was determined using the log- rank test. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[128, 110, 833, 870]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[135, 88, 816, 109]]<|/det|> +
Fig.S3: Survival analysis of potential O-GlcNAc regulators in UCEC
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[67, 95, 884, 283]]<|/det|> +1125 (a) Kaplan-Meier analyses were conducted on EC patients in the TCGA UCEC dataset to 1126 elucidate the relationship between the expression levels of the 17 overlapping genes and 1127 patients' OS over time. EC cases were stratified using the median expression level as cut- off, 1128 and statistical significance was determined using the log- rank test. 1129 (b) Validation the FBXO31 KO 293T cells using sanger sequencing. The KO cells harbor a 1130 17 bp deletion, which causes a frameshift mutation starting from the \(48^{\text{th}}\) amino acid of 1131 FBXO31, terminating prematurely at the \(106^{\text{th}}\) amino acid. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 110, 888, 878]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[226, 92, 772, 112]]<|/det|> +
Fig.6: FBXO31 interacts with and ubiquitinates OGT
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[123, 95, 881, 199]]<|/det|> +(a) Immobilized recombinant GST-OGT protein but not GST control absorbed GFP-FBX031 from 293T cell lysates. GST and GST-OGT were detected by Coomassie brilliant blue (CBB) staining, and FBXO31 was detected by western blotting with FBXO31 antibody. + +<|ref|>text<|/ref|><|det|>[[123, 207, 881, 253]]<|/det|> +(b) Co-immunoprecipitation of GFP-FBX031 with Flag-OGT in 293T cell lysates. The presence of MG132 enhanced the interaction between Flag-OGT and GFP-FBX031. + +<|ref|>text<|/ref|><|det|>[[123, 261, 881, 308]]<|/det|> +(c) Western blotting assessing the protein level of OGT as well as the global O-GlcNAc (RL2) levels in 293T cells transfected with increasing amount of GFP-FBX031. + +<|ref|>text<|/ref|><|det|>[[123, 316, 881, 362]]<|/det|> +(d) Western blotting detecting the protein level of endogenous OGT and its ubiquitination in 293T cells transfected with different amount of HA-Ub and GFP-FBX031. + +<|ref|>text<|/ref|><|det|>[[123, 371, 881, 530]]<|/det|> +(e) In vitro ubiquitination of His-OGT by the SCF complex together with FBXO31. HA-tagged SCF components (Skp1, Cul1, and Roc1) and HA-FBX031 were affinity-purified using anti-HA-conjugated magnetic beads from HEK293T cell lysates. The purified protein complex was incubated with E1 (UBA1), E2 (UBE2D1), Ub, and His-OGT in ubiquitination buffer. The reaction was halted by the addition of SDS sample buffer, and the samples were subjected to western blotting using the indicated antibodies. + +<|ref|>text<|/ref|><|det|>[[123, 538, 881, 613]]<|/det|> +(f) In vivo ubiquitination assay was performed to evaluate the ubiquitination levels of exogenous Flag-OGT in 293T cells transfected with HA-tagged Ub and GFP-FBX031 or its F-box domain deletion mutant GFP-FBX031ΔF. + +<|ref|>text<|/ref|><|det|>[[123, 622, 881, 723]]<|/det|> +(g) Western blotting quantification of the protein level of endogenous OGT in 293T cells transfected with GFP-FBX031. MG132 was added to inhibit the ubiquitination-mediated proteasome degradation. Statistical significance was determined by unpaired two-tailed Student's t-test, n = 3, ***P < 0.001. + +<|ref|>text<|/ref|><|det|>[[123, 733, 880, 779]]<|/det|> +(h) Western blotting detecting the O-GlcNAc (RL2) and OGT levels in WT and FBXO31 KO 293T cells. + +<|ref|>text<|/ref|><|det|>[[123, 789, 881, 864]]<|/det|> +(i) Western blotting quantitation of OGT protein level following cycloheximide (CHX) treatment in WT and FBXO31 KO 293T cells. Statistical significance was determined by unpaired two-tailed Student's t-test, n = 3, *P < 0.05. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 120, 889, 875]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[140, 88, 870, 112]]<|/det|> +
Fig.7: Loss of FBXO31 increases O-GlcNAc level in clinical samples
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[120, 95, 881, 142]]<|/det|> +(a) Representative images depicting IHC staining of FBXO31 in EC and adjacent normal tissues on an FFPE tissue array. Scale bar: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[120, 151, 881, 299]]<|/det|> +(b) Quantitative analysis of the level of FBXO31 in the EC tissue array. The expression level of FBXO31 was assessed semi-quantitatively based on both staining intensity and area. Statistical significance was calculated using unpaired two-tailed Student's t-test, \(**\mathrm{P}< 0.01\) . +(c) Percentage of samples with high or low FBXO31 levels by IHC in the two different \(O\) -GlcNAcylation level groups (high-RL2 or low-RL2). High and low expression categories of FBXO31 were determined using a scoring system (high score: 8-12; low score: 0-6). + +<|ref|>text<|/ref|><|det|>[[120, 317, 881, 364]]<|/det|> +(d) Protein levels of OGT and FBXO31 were assessed by Western blotting in EC-Os and EE-Os derived from different patients. + +<|ref|>text<|/ref|><|det|>[[120, 372, 881, 447]]<|/det|> +(e) Immunofluorescence detection of \(O\) -GlcNAcylation (RL2, green) and FBXO31 (red) in control EE-Os and shFBXO31 infected EE-Os. The nuclei were stained with DAPI (blue), and F-actin was labeled by Phalloidine (magenta). Scale bar: \(50 \mu \mathrm{m}\) . + +<|ref|>text<|/ref|><|det|>[[120, 456, 881, 530]]<|/det|> +(f) qPCR data of stemness markers expression in control shNT and shFBXO31 infected EE-Os, normalized to actin mRNA levels. Statistical significance was calculated using unpaired two-tailed Student's t-test, \(n = 3\) , ns: not significant, \(**\mathrm{P}< 0.05\) , \(**\mathrm{P}< 0.01\) . + +<|ref|>text<|/ref|><|det|>[[120, 539, 881, 640]]<|/det|> +(g) Quantification of organoid colony-forming numbers of the control and shFBXO31 infected EE-Os in 3D culture. Representative bright-field images are provided on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(***\mathrm{P}< 0.001\) . + +<|ref|>text<|/ref|><|det|>[[120, 650, 881, 752]]<|/det|> +(h) Quantification of organoid colony-forming numbers of shFBXO31 treated EE-Os at day 3 of culture after treatment with OSMI-1 compared with control (DMSO). Representative bright-field images are shown on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(***\mathrm{P}< 0.0001\) . + +<|ref|>text<|/ref|><|det|>[[120, 761, 881, 863]]<|/det|> +(i) Quantification of organoid colony-forming numbers of EC-Os overexpressing GFP or GFP-FBXO31. Bright-field and fluorescent images of the treated EC-Os are shown on the left. Scale bar: \(50 \mu \mathrm{m}\) . Statistical significance was calculated using unpaired two-tailed Student's t-test, \(**\mathrm{P}< 0.01\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[122, 125, 884, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[388, 98, 639, 120]]<|/det|> +
Fig.8: Working model
+ +<|ref|>text<|/ref|><|det|>[[63, 410, 88, 424]]<|/det|> +1192 + +<|ref|>text<|/ref|><|det|>[[60, 435, 884, 515]]<|/det|> +1193 Fig. 8 Working model.1194 FBXO31- mediated ubiquitination of OGT maintains a relatively low level of1195 \(O\) - GlcNAcylation in the normal endometrium. Inactivation of FBXO31 in endometrial1196 cancer tissues results in accumulation of OGT and concurrent increase of \(O\) - GlcNAcylation. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[125, 150, 826, 610]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[147, 121, 810, 156]]<|/det|> +Table 1. Association of O-GlcNAcylation expression with clinicopathological parameters of patients with EC + +
ParameterDescribeNO-GlcNAcylation levelP value
LowHigh
Age&lt; 6016589760.4305
≥ 60543321
Histologic gradeG1714922&lt; 0.0001
G21066343
G3421032
FIGO stageI13778590.0080
II332211
III412219
IV808
BMI&lt; 2816895730.8721
≥ 28512724
DiabetesNo196107890.3808
Yes23158
HypertensionNo15990690.6494
Yes603228
Distant metastasisNegative211122890.0013
Positive808
Lymph node metastasisNegative195112830.1910
Positive241014
Myometrial invasion&lt; 1/215393600.0654
≥ 1/2582632
Serosa835
+ +<--- Page Split ---> +<|ref|>table_caption<|/ref|><|det|>[[135, 102, 860, 118]]<|/det|> +Table 2. Univariate and multivariate Cox regression analysis for PFS in EC patients + +<|ref|>table<|/ref|><|det|>[[120, 120, 875, 671]]<|/det|> + +
CharacteristicsNUnivariate analysisMultivariate analysis
HR(95% CI)P valueHR(95% CI)P value
Age
\(<60\)154
\(\geq 60\)503.437(1.104-10.694)0.0333.980(1.220-12.982)0.022
BMI
\(<28\)155
\(\geq 28\)490.635(0.139-2.902)0.558
Diabetes
No184
Yes200.922(0.119-7.164)0.938
FIGO stage
I+II157
III+IV473.327(1.073-10.319)0.0372.552(0.755-8.630)0.132
Histologic grade
Grade 167
Grade \(2+3\)1372.233(0.489-10.200)0.300
Lymph node
metastasis
Negative182
Positive223.067(0.828-11.353)0.093
Myometrial
invasion
\(<1/2\)142
\(\geq 1/2+S\)erosa624.848(1.459-16.108)0.0102.591(0.727-9.236)0.142
O-GlcNAcylation
level
Low115
High894.823(1.297-17.937)0.0194.611(1.212-17.536)0.025
+ +<|ref|>text<|/ref|><|det|>[[65, 675, 466, 689]]<|/det|> +1199 HR: hazard ratio; CI: confidence interval. + +<--- Page Split ---> +<|ref|>table<|/ref|><|det|>[[125, 115, 848, 620]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[140, 94, 831, 128]]<|/det|> +Table 3. Association of FBXO31 expression with clinicopathological parameters of patients with EC + +
ParameterDescribeNFBXO31 level LowHighP value
Age&lt; 608848400.1022
≥ 60331221
Histologic gradeG140931&lt; 0.0001
G2512427
G330273
FIGO stageI8939500.0862
II1073
III1587
IV761
BMI&lt; 289044460.8372
≥ 28311615
DiabetesNo10858500.016
Yes13211
HypertensionNo8143380.2732
Yes401723
Distant metastasisNegative11454600.0614
Positive761
Lymph node metastasisNegative10750570.0956
Positive14104
O-GlcNAcylation levelLow38731&lt; 0.0001
High835330
Myometrial invasion&lt; 1/28039410.3559
≥ 1/2412120
Serosa220
+ +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[298, 92, 707, 109]]<|/det|> +# Descriptions of additional Supplementary Files + +<|ref|>text<|/ref|><|det|>[[70, 111, 692, 592]]<|/det|> +1201 Supplementary Table 1 1203 TPM of 1000 O- GlcNAc correlated genes from 55 EC tissues 1204 Supplementary Table 2 1206 GO analysis for 1000 O- GlcNAc correlated genes 1207 Supplementary Table 3 1209 TPM of 1000 O- GlcNAc correlated genes from TCGA UCEC dataset 1210 Supplementary Table 4 1212 Clinical data and predicted O- GlcNAc index in TCGA UCEC dataset 1213 Supplementary Table 5 1215 Cluster markers 1216 Supplementary Table 6 1218 RL2_pos_vs_RL2_ctrl genes summarized with MAGeCK- RRA 1219 Supplementary Table 7 1221 KEGG Pathway analysis on the 1038 top scoring genes 1222 Supplementary Table 8 1224 Primers for plasmid constructs, and RT- qPCR 1225 Supplementary Table 9 1227 Antibodies for immunostaining and WB + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 92, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[58, 130, 312, 365]]<|/det|> +Supplementarytable1. xlsx Supplementarytable2. xlsx Supplementarytable3. xlsx Supplementarytable4. xlsx Supplementarytable5. xlsx Supplementarytable6. xlsx Supplementarytable7. xlsx Supplementarytable8. xlsx Supplementarytable9. xlsx + +<--- Page Split ---> diff --git a/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/images_list.json b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c8102cae0e286254c1abdade575145dc81c94562 --- /dev/null +++ b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/images_list.json @@ -0,0 +1,55 @@ +[ + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 115, + 105, + 880, + 576 + ] + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 113, + 95, + 880, + 460 + ] + ], + "page_idx": 35 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 115, + 110, + 876, + 590 + ] + ], + "page_idx": 37 + } +] \ No newline at end of file diff --git a/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2.mmd b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2.mmd new file mode 100644 index 0000000000000000000000000000000000000000..f7e6f399f595d6b57136ad1d64d7d4004e528164 --- /dev/null +++ b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2.mmd @@ -0,0 +1,472 @@ + +# Dynamin-dependent entry of Chlamydia trachomatis is sequentially regulated by the effectors TarP and TmeA + +Rey Carabeo + +rey.carabeo@unmc.edu + +University of Nebraska Medical Center https://orcid.org/0000- 0002- 5708- 5493 Matthew Romero University of Nebraska Medical Center https://orcid.org/0000- 0002- 3459- 1019 + +## Article + +# Keywords: + +Posted Date: September 27th, 2023 + +DOI: https://doi.org/10.21203/rs.3. rs- 3376558/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on June 10th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49350- 6. + +<--- Page Split ---> + +# Dynamin-dependent entry of Chlamydia trachomatis is sequentially regulated by + +# the effectors TarP and TmeA + +3 + +Matthew D. Romero and Rey A. Carabeo1 + +5 + +Department of Pathology and Microbiology, College of Medicine, University of Nebraska Medical Center, + +Omaha, NE + +8 + +9 + +1Corresponding Author: + +Department of Pathology and Microbiology + +University of Nebraska Medical Center + +985900 Nebraska Medical Center, Omaha, NE + +Email: rey.carabeo@unmc.edu + +<--- Page Split ---> + +## 16 Abstract + +16 AbstractChlamydia invasion of epithelial cells is a pathogen- driven process involving two functionally distinct effectors – TarP and TmeA. They collaborate to promote robust actin dynamics at sites of entry. Here, we extend studies on the molecular mechanism of invasion by implicating the host GTPase dynamin 2 (Dyn2) in the completion of pathogen uptake. Importantly, Dyn2 function is modulated by TarP and TmeA at the levels of recruitment and activation through oligomerization, respectively. TarP- dependent recruitment requires phosphatidylinositol 3- kinase and the small GTPase Rac1, while TmeA has a post- recruitment role related to Dyn2 oligomerization. This is based on the rescue of invasion duration and efficiency in the absence of TmeA by the Dyn2 oligomer- stabilizing small molecule activator Ryngo 1- 23. Notably, Dyn2 also regulated turnover of TarP- and TmeA- associated actin networks, with disrupted Dyn2 function resulting in aberrant turnover dynamics, thus establishing the interdependent functional relationship between Dyn2 and the effectors TarP and TmeA. + +## 28 Introduction + +28 IntroductionChlamydia trachomatis is an obligate intracellular bacterium which infects ocular and genital epithelial cells, causing pelvic inflammatory disease, tubal factor infertility, ectopic pregnancy, and preventable blindness1. Chlamydia features a biphasic developmental cycle divided between metabolically quiescent elementary bodies (EBs) which invade host cells and vegetative reticulate bodies (RBs) which replicate inside membrane vacuoles termed inclusions2. Given its obligate intracellular nature, entry into host cells is essential for pathogen survival; consequently, Chlamydia possesses a robust suite of resources that regulate its uptake. Invasion also underpins pathogenicity, as it promotes access to the intracellular niche where it hijacks several host cell processes. Initial interaction with host epithelial cells is mediated by a reversible electrostatic interaction between a Chlamydia adhesin and host heparin sulfate proteoglycans3. Subsequently, Chlamydia engages multiple host receptors and delivers a variety of protein effectors via a + +<--- Page Split ---> + +type III secretion system4-6. Signaling from the effectors TarP and TmeA establishes a robust actin modulatory network that induces the assembly of actin- rich structures that engulf invading bacteria7-9. The resultant actin recruitment is characteristically highly localized to invading EBs and exhibits rapid kinetics of actin recruitment and turnover, such that actin network assembly and disassembly occurs within 200 seconds7,10. The majority of studies regarding chlamydial invasion focus on the mechanism of actin recruitment, while the process of disassembly at the end of invasion remains understudied, despite evidence pointing to its importance to elementary body uptake. We recently reported that altering the dynamics of actin turnover correlated with decreased invasion efficiency7. + +Although multiple uptake mechanisms have been implicated as potential pathways for C. trachomatis invasion11- 13, the role of host dynamins during this process has been controversial. Dynamins are large GTPases that form oligomeric structures in a helical configuration around membrane lipids during clathrin- and caveolin- mediated endocytosis, mediating scission of vesiculated cargoes following GTP hydrolysis14. They are comprised of a catalytic G domain, a lipid- binding pleckstrin homology (PH) domain, and a proline- rich domain (PRD) that interacts with Src homology 3 (SH3) domain- containing proteins15. Absent activation, dynamins possess low intrinsic GTPase activity and assemble into dimers or tetramers16. These are utilized to generate higher- order oligomers such as half- rings, rings, and helices, the latter forming at the collar of invaginating vesicles17. GTP hydrolysis induces a conformational change along the oligomer that promotes constriction followed by vesicle scission, prompting rapid turnover of the dynamin superstructure18. Dynamin oligomerization is promoted by several effectors, including SH3 domain- containing proteins19, actin filaments20, and membrane lipids21. Many known effectors of dynamin oligomerization are present at C. trachomatis invasion sites, raising the possibility that that dynamin- dependent scission is utilized during terminal stages of this process. Several host proteins present during invasion are also directly or indirectly targeted by chlamydial effectors4,22- 24, highlighting the level of control the pathogen exerts on the invasion process. + +<--- Page Split ---> + +RNA interference of dynamin 2 (Dyn2) restricted C. trachomatis uptake12, bolstering support for a dynamin- dependent uptake mechanism. In contrast, pretreatment with the dynamin inhibitor MiTMAB, which targets the PH domain of dynamin, did not alter C. trachomatis invasion efficiency11. However, this study also identified that SNX9, a BAR domain protein which promotes dynamin oligomerization, is recruited during invasion, and that its depletion attenuated Chlamydia entry. Furthermore, overexpression of dominant negative GTPase- inactive Dyn1 K44A did not prevent C. trachomatis infection of HeLa cells25. Notably, this study did not investigate C. trachomatis uptake frequency and did not target Dyn2, the predominant dynamin species expressed in epithelial cells. In this study, we aim to reconcile the controversial involvement of host dynamins during C. trachomatis entry, monitoring its involvement using a series of high- resolution tools previously employed to characterize the regulation of actin remodeling during invasion7. + +Given that dynamin interacts both with actin itself and with several proteins that regulate actin polymerization20,26- 29, it has become increasingly apparent that the dynamin GTPase cycle and actin polymerization are co- regulated. On this basis, the secreted effectors TarP and TmeA, which are themselves regulators of actin dynamics, likely also regulate host Dyn2 during invasion. Once secreted, TmeA associates with the plasma membrane and activates N- WASP, followed by Arp2/3 complex activation and nucleation of actin polymerization9,30. Likewise, TarP signaling activates host signaling proteins such as Rac1, PI3K, and the WAVE2 complex, in addition to recruiting the actin effectors formin and Arp2/37,31. Many host proteins associated with TarP and TmeA signaling are known to regulate Dyn2 oligomerization, such as cortactin32, EPS8333, profilin32,34 and the Arp2/3 complex35. Thus, in addition to the previously established role of TarP and TmeA signaling as synergistic effectors of rapid actin kinetics7,8, it is likely that they have a role in Dyn2 localization dynamics during Chlamydia entry. + +Here, we demonstrate that Dyn2 is co- recruited alongside actin during Chlamydia invasion and coordinates efficient engulfment of the pathogen. This phenomenon is contingent upon signaling from + +<--- Page Split ---> + +both TarP and TmeA, such that TarP signaling is necessary for local recruitment of Dyn2, whereas TmeA signaling activates Dyn2 by promoting oligomerization. The application of the Dyn2 activator Ryngo 1- 23, which promotes oligomerization and stabilizes Dyn2 polymers rescues invasion defects associated with TmeA deletion, enhancing its entry efficiency, and restoring kinetics of Dyn2 and actin recruitment and turnover to near wild- type levels. Further, we discovered that actin disassembly is dependent on Dyn2 function, thus ensuring the completion of invasion. Altogether, these findings resolve the long- standing controversy within the field, providing a novel regulatory function which accounts for both rapid assembly and disassembly of Chlamydia engulfment machinery in addition to a comprehensive model for the utilization and regulation of host Dyn2 during C. trachomatis invasion. They also highlight cooperation between TarP and TmeA and illustrate the broader impact of establishing their respective actin networks beyond the formation of engulfment structures. + +## Results + +## Dynamin 2 and actin are co-recruited during Chlamydia entry + +Since conflicting reports persist regarding host dynamin- 2 (Dyn2) involvement during C. trachomatis invasion, we revisited the question and evaluated its recruitment in greater detail using quantitative imaging approaches. We first determined whether Dyn2 was present within entry sites by co- transfecting Cos7 cells with GFP- Dyn2 and iRFP670- LifeAct prior to infection with wild- type C. trachomatis (MOI=20) stained with the red fluorescent dye CMTPX. Using live- cell confocal microscopy, we monitored Dyn2 and actin recruitment during entry, acquiring images at 20 second intervals (Fig. 1A). As previously reported7, we observed rapid actin recruitment, which was concomitant with arrival of Dyn2 and resulted in rapid uptake of Chlamydia, characterized by loss of CMTPX- CTL2 signal within 200- 300 sec. In contrast, expression of mutant Dyn2 K44A (Dyn2 DN), which is defective in GTPase binding and hydrolysis and + +<--- Page Split ---> + +cannot mediate vesicle scission (Fig. 1B), prolonged internalization to 400- 700 sec. Delayed pathogen uptake following Dyn2 DN expression could arise from several potential sources, such as inefficient Dyn2 recruitment, impaired actin dynamics, or disruptions within the Dyn2 GTPase cycle that prevent vesicle scission. To address each of these possibilities, we employed a previously established protocol for quantitatively assessing host protein recruitment dynamics during Chlamydia invasion, starting by characterizing Dyn2 WT and Dyn2 DN recruitment dynamics (Fig. 1C). While both Dyn2 WT and Dyn2 DN were recruited during entry, we noted that Dyn2 DN achieved peak mean fluorescence intensity (MFI) roughly 80 seconds later than Dyn2 WT and persisted within entry sites for a longer duration, indicating that rapid recruitment of Dyn2 and subsequent rapid entry of Chlamydia is contingent upon Dyn2 GTPase activity. To further substantiate this claim, we converted time- lapse images of actin, Dyn2, and Chlamydia into kymographs, upon which we indicated the start (i.e. initiation of actin/Dyn2 recruitment) and end (i.e. loss of EB fluorescence) of invasion (Fig. 1D, S1). The duration between initiation of actin/Dyn2 recruitment and pathogen entry was prolonged by expression of Dyn2 DN (Fig. 1D,E), such that Chlamydia uptake in cells expressing Dyn2 WT occurred within 180 sec, which was delayed by over two- fold (380 sec) when Dyn2 DN was expressed. Moreover, slow pathogen uptake following Dyn2 DN expression coincided with slower Dyn2 recruitment and turnover (Fig. 1F,G), reducing the rate of Dyn2 recruitment by 40 percent and turnover by 60 percent compared to Dyn2 WT. Altogether, these data indicate that Dyn2 is co- recruited alongside actin during Chlamydia entry, and that Dyn2 GTPase activity is necessary for efficient recruitment dynamics and rapid pathogen entry. + +## Dynamin 2 inhibition restricts Chlamydia entry and actin turnover + +The recruitment of Dyn2 alongside its role in facilitating rapid pathogen entry suggests that dynamin- dependent uptake is an important component of Chlamydia invasion. Previous reports indicate that Dyn2 self- assembly and actin polymerization are co- regulated26,29,35, such that delayed Chlamydia entry following Dyn2 disruption may be due to defective actin polymerization. To test this, we disrupted Dyn2 + +<--- Page Split ---> + +activity via pharmacological inhibition or by RNA interference prior to monitoring actin recruitment during Chlamydia invasion. Since co- overexpression of Dyn2 DN and actin may artificially influence actin dynamics, we instead inhibited endogenous Dyn2 using the dynamin inhibitor Dynasore, which mimics Dyn2 DN by restricting Dyn2 GTPase activity and subsequent scission (Fig. 2A). Furthermore, we were limited to \(\sim 50\%\) Dyn2 knockdown via RNA interference (Fig. 2B), as excessive Dyn2 depletion prevented cell adherence and cell proliferation, rendering these cells unsuitable for further analysis. Nonetheless, we noted that both \(25 \mu M\) Dynasore treatment and partial siRNA depletion of Dyn2 attenuated actin dynamics during CTL2 WT invasion (Fig 2B), resulting in prolonged actin retention within entry sites. Interestingly, actin recruitment kinetics were largely unchanged by Dyn2 disruption, yielding comparable rates across all conditions (Fig. 2C). In contrast, actin turnover was significantly attenuated by both Dynasore treatment and Dyn2 siRNA knockdown, with Dynasore treatment halving the actin turnover rate, while Dyn2 siRNA treatment slowed actin turnover by 25 percent (Fig. 2D). Given the importance of rapid actin turnover kinetics toward efficient invasion7, it is possible that Dyn2 inhibition (or absence) prolongs Chlamydia entry through defects in actin turnover. In support of this notion, we observed that both inhibition and depletion of Dyn2 delayed Chlamydia entry by roughly two- fold (Fig. 2E,F), comparable to the delay observed following Dyn2 DN overexpression (Fig. 1E), indicating that active Dyn2 is required for efficient actin turnover and rapid Chlamydia entry. Moreover, we observed a comparable attenuation in wild- type Chlamydia entry efficiency following Dyn2 inhibition (Fig. 2G) or siRNA depletion (Fig. 2H), reducing Chlamydia uptake by roughly 20 percent. Therefore, Dyn2 activity regulates actin turnover during invasion such that disruption of Dyn2 impedes actin depolymerization within entry sites. + +## Signaling from both TarP and TmeA is required for dynamin-dependent entry + +Several studies have indicated that mutant Chlamydia strains harboring TarP and TmeA deletion or loss- of- function mutations exhibit substantially dysregulated pathogen entry7,36,37. As such, we monitored invasion of Chlamydia mutant strains lacking either TarP or TmeA (ΔTmeA, ΔTarP) or both (DKO) to + +<--- Page Split ---> + +determine if their respective routes of entry were affected by Dyn2 inhibition or depletion. Loss of either TarP or TmeA rendered their respective invasion processes resistant to Dyn2 inhibition (Fig. 2G), likely indicating the utilization of an alternative entry mechanism, i.e. fluid- phase uptake, which is dynamin- independent (Fig. S2). Entry efficiency of these strains were similarly insensitive to Dyn2 depletion via RNA interference (Fig. 2H), confirming that Dyn2 does not contribute to pathogen invasion following TarP or TmeA deletion. Finally, we noted that cis- complementation of the \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) and \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) mutants (cis- TmeA, cis- TarP) restored Dynasore sensitivity (Fig. 2G). + +For the \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) mutant, Dyn2 dispensability was unsurprising given the spatial profiles of actin exhibited by this mutant, which assembles structures typically associated with fluid- phase uptake, such as large blooms and mini- ruffles38 (Fig. S2A,E). Indeed, \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) EBs frequently colocalized with the fluid- phase marker Dextran- Alexa Fluor 647; 40 percent of EBs were dextran positive within 20 minutes post- entry (Fig. S2F,G). In contrast, the \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) mutant retained punctate recruitment of actin characteristic of wild- type EBs (Video S2,3) and exhibited lower incidence of dextran colocalization (Fig. S2F). Thus, invasion of \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) EBs is mechanistically distinct from \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) , adopting a spatial configuration that may benefit from Dyn2 activity. As such, the apparent insensitivity of \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) EBs toward Dyn2 inhibition might reflect that Dyn2 is required for entry, but present in a non- functional state that rendered inhibition by Dynasore moot, which will be addressed in detail later in this study. Altogether, our data unequivocally reveal that dynamin- dependent uptake is an important component of C. trachomatis invasion which is contingent upon both TarP and TmeA signaling, wherein each effector likely regulates different invasion- associated aspects of Dyn2. + +## TarP and TmeA mediate recruitment and post-recruitment activation of Dyn2, respectively + +TarP and Dyn2 function. Strikingly, TarP deletion prevented localized recruitment of Dyn2 at sites of pathogen entry (Fig. S2, Video S1), indicating that TarP signaling regulates early aspects of Dyn2 recruitment. We hypothesize that the actin network induced by TarP, rather than TarP itself, is responsible + +<--- Page Split ---> + +for Dyn2 recruitment, given actin and Dyn2 colocalization (Fig. 1) and the reported functional relationship between these proteins20,29. To provide a mechanistic basis for TarP- dependent regulation of Dyn2, we investigated how Dyn2 localization dynamics are affected by ablation of PI3K/Rac1 signaling, which contributes to TarP- mediated actin remodeling31 (Fig. 3B). We monitored Dyn2 recruitment following administration of the Rac- specific inhibitor EHop- 016 (10μM) at entry sites of wild- type and ΔTmeA EBs, since both strains retain TarP signaling (Fig. 3). Rac inhibition did not affect the rate of Dyn2 recruitment (Fig. 3G) but substantially attenuated its turnover (Fig. 3H), resulting in prolonged retention of Dyn2 within CTL2 WT entry sites (Mock = 260 sec, EHop = 520 sec) (Fig. 3A). Thus, TarP- mediated actin remodeling not only coordinates local recruitment of Dyn2 within entry sites, but also regulates its retention via Rac1 signaling. Interestingly, Dyn2 localization dynamics of ΔTmeA mutants were unaffected by Rac inhibition (Fig. 3C), exhibiting similar recruitment and turnover rates between mock- and EHop- treated samples (Fig. 3G,H). TmeA- dependent sensitivity of Dyn2 localization dynamics toward Rac signaling hints at a significant role for this effector in Dyn2 function, which likely manifests at later (i.e., post- recruitment) stages. + +We next tested the role of PI3K/Vav2 signaling, which is one of the Rac- activating pathways linked to TarP, the other being Abi1/Eps8/Sos1 signaling31 (Fig. 4D). To determine the functional outcome of PI3K signaling toward Dyn2 regulation, we monitored the invasion of wild- type and ΔTmeA EBs in the presence of the PI3K inhibitor Wortmannin (100 nM). Pretreatment with Wortmannin yielded intense and long- lasting Dyn2 localization relative to mock at wild- type entry sites (Fig. 4A,B) and attenuated the rate of Dyn2 turnover (Fig. 4H), consistent with PI3K signaling through Rac (Fig. 3C,H). Interestingly, PI3K inhibition did not alter Dyn2 recruitment during ΔTmeA invasion (Fig. 4A,B), indicating that in absence of TmeA, Dyn2 is not in its proper context to be affected further by wortmannin treatment. Moreover, wortmannin pretreatment did not alter the invasion efficiency of any strain tested (Fig. 4C) yet induced a significant delay in CTL2 WT uptake (Mock = 180 sec, Wort = 320 sec) (Fig. 4E,F). This disparity may arise due to the enhanced sensitivity of our kymograph- based internalization assay (Fig. 4E,F), which employs quantitative + +<--- Page Split ---> + +fluorescence- based live- cell imaging to identify invasion defects. The former internalization assay (Fig. 4C) relies on antibody accessibility to measure invasion efficiency, a low- resolution approach with elevated likelihood of missing regulatory interactions between host and pathogen. In summary, our data indicate that TarP signaling is essential for dynamin- dependent entry of Chlamydia and is required for local recruitment of Dyn2 within entry sites, while also regulating its retention as a consequence of the actin network generated via the PI3K/Rac1 signaling axis. + +TmeA and Dyn2 function. Although \(\Delta\) TmeA EBs recruit Dyn2 in a highly localized and punctate manner, similar to CTL2 WT (Fig 3A, 4A), inhibition of function via ectopic expression of dominant negative Dyn2 or \(25 \mu M\) Dynasore treatment did not alter uptake duration or Dyn2 dynamics associated with \(\Delta\) TmeA (Fig. S3). The apparent insensitivity toward Dyn2 disruption following TmeA deletion may reflect a lack of Dyn2 involvement during \(\Delta\) TmeA entry, or that TmeA deletion induces Dyn2 loss of function. To distinguish between these two possibilities, we employed the Dyn2 activator Ryngo 1- 23, a small molecule compound that stimulates Dyn2 oligomerization in a manner comparable to short actin filaments39. As such, we quantified Chlamydia entry after 30 minute preincubation with \(40 \mu M\) Ryngo 1- 23, wherein \(\Delta\) TmeA invasion efficiency was improved to near wild- type levels (Mock CTL2 WT = 79.8%, Ryngo \(\Delta\) TmeA = 71.0%) (Fig. 5A). Moreover, this compound restored normal Dyn2 recruitment dynamics during \(\Delta\) TmeA entry (Fig. 5B), generating a Dyn2 recruitment profile comparable to mock- treated CTL2 WT (Fig. 5D,G,H). Likewise, both mock CTL2 WT and Ryngo \(\Delta\) TmeA were internalized within 180 seconds on average, which was prolonged to 240 seconds for \(\Delta\) TmeA in absence of Ryngo (Fig. 5E,F), and that compound- assisted entry reduced the incidence of fluid- phase uptake (Fig. S2F). Taken together, these data suggest that Dyn2 oligomerization is defective when TmeA signaling is absent, and that Ryngo bypasses the requirement for TmeA signaling, enabling dynamin- dependent entry of \(\Delta\) TmeA EBs. In contrast, invasion efficiency, Dyn2 localization dynamics, and duration of internalization associated with wild type CTL2 were all negatively affected by Ryngo (Fig. 5A- F). A possible explanation may be that joint activation of Dyn2 by both TmeA + +<--- Page Split ---> + +signaling and Ryngo administration results in Dyn2 hyperactivation that prevents normal completion of the Dyn2 GTPase cycle. Indeed, Gu et. al found that Ryngo 1- 23 abrogated Dyn1 helical collar assembly, instead promoting stacked ring assembly (Fig. 5C), exhibiting reduced GTPase activity and attenuated vesicle scission compared to helices39. Additionally, CTL2 WT entry was comparably attenuated by either Dynasore- mediated inhibition of Dyn2 (Fig. 2G) or Ryngo- mediated Dyn2 activation (Fig. 5A), implying that dynamin- dependent entry of Chlamydia is sensitive to both hypo- and hyperactivation of Dyn2. Finally, whereas Ryngo administration prior to infection with \(\Delta\) TarP EBs restored localized recruitment of Dyn2 (Figs. S2C, S4B, Video S1), its recruitment was vastly dysregulated relative to wild- type (Fig. S4C,F,G) and failed to elicit rapid internalization of the pathogen (Fig. S4E). Together, this implies that Dyn2 is not organized in a proper context within entry sites when TarP is absent despite restoration of recruitment by Ryngo. In contrast, Dyn2 dynamics and function were restored by Ryngo treatment in \(\Delta\) TmeA EB invasion because Dyn2 proteins were in a context that favors oligomerization. In summary, these data indicate that TmeA signaling activates Dyn2, promoting its oligomerization in support of rapid dynamin- dependent entry of Chlamydia. In addition, the ordered roles of TarP and TmeA regarding Dyn2 function highlights the previously reported collaboration between these two effectors. + +## Actin turnover is correlated with Dynamin 2 activation status and Chlamydia uptake + +Previous studies have identified that TmeA deletion dysregulates the actin network generated by Chlamydia during invasion, causing poor actin retention and abnormally fast actin turnover7- 9. Moreover, in this study, we have noted a functional link between Dyn2 activity and actin turnover, wherein actin recruitment was abnormally persistent upon pharmacological inhibition of Dyn2 or upon expression of Dyn2 K44A (Figs. 2B, S1F), resulting in delayed pathogen uptake. In light of these observations, we opted to evaluate the influence of the dynamin activator Ryngo 1- 23 on actin kinetics to determine whether compound- mediated restoration of Dyn2 activity within \(\Delta\) TmeA entry sites also restores normal actin dynamics. While administration of Ryngo prior to infection strongly increased the persistence of actin + +<--- Page Split ---> + +recruitment at entry sites of both wild- type and \(\Delta \mathsf{TmeA}\) EBs (Fig. 6A), relative to the respective mock- treated controls, the slowed turnover associated with the \(\Delta \mathsf{TmeA}\) mutant was indistinguishable from mock- treated wild type control (Fig. 6A- C, Video S2, S3). As expected, we observed that Ryngo treatment restored the duration of internalization of \(\Delta \mathsf{TmeA}\) mutants to levels of mock- treated CTL2 WT (Fig. 6E,F). However, when invasion signaling was intact, i.e. when TarP and TmeA are both present, the additional Dyn2 activation by Ryngo had a negative effect on actin turnover and pathogen uptake (Fig. 6A,F, Video S3). This paralleled the effects of Ryngo on Dyn2 recruitment (Fig. 5), underscoring a possible relationship between actin disassembly and Dyn2 turnover (Fig. 6D). Indeed, either insufficient Dyn2 activity (i.e., Dynasore treatment, Dyn2 DN, Mock/ \(\Delta \mathsf{TmeA}\) ; Fig. 2B- D) or Dyn2 hyperactivation (i.e., Ryngo/CTL2 WT; Fig. 6A- F) results in similar dysregulated actin turnover and delayed pathogen uptake. Collectively, our data is consistent with a model whereby actin remodeling by TarP and TmeA, in addition to forming engulfment structures, also ensures Dyn2 recruitment and activation. With Dyn2 regulating actin turnover, this self- contained invasion mechanism ensures that disassembly of the invasion structures is properly coordinated with a successful scission event indicated by Dyn2 turnover. + +## Discussion + +In this study, we conclusively demonstrated that C. trachomatis utilizes host Dyn2 to complete invasion. Dyn2 function is modulated by the effectors TarP and TmeA, which respectively mediate recruitment to invasion sites and activation by promoting oligomerization. Neither TarP nor TmeA possesses domains that mediate direct interaction with Dyn2 to facilitate recruitment and oligomerization; instead, TarP and TmeA modulate Dyn2 via their respective actin networks. Interestingly, Dyn2 influences actin turnover, wherein perturbation of Dyn2 function induces persistent actin retention. This functional interdependence constitutes a self- regulating system, such that Dyn2 function and pathogen engulfment are regulated by the actin network assembled via TarP and TmeA signaling. Reciprocally, Dyn2 function and subsequent membrane fission promotes actin disassembly and mediates resolution of engulfment structures. + +<--- Page Split ---> + +Moreover, TarP and TmeA signaling are sequentially coordinated such that the essential steps of invasion are initiated and completed. Specifically, we found that TarP signaling via PI3K/Rac1 coordinated initial recruitment and retention of Dyn2 within entry sites. Once recruited, Dyn2 is activated by TmeA signaling on the basis that defects associated with TmeA deletion were rescued by administration of the small molecular activator Ryngo 1- 23, which promotes Dyn2 oligomerization. Moreover, these data are consistent with previous observations suggesting that TmeA regulates latter stages of invasion. Finally, our study provides several high- resolution methods for tracking pathogen uptake, enabling detailed analysis of host- pathogen interactions underpinning Chlamydia entry, exceeding the limitations of previously employed techniques. In summary, we report that Dyn2 activation is an important component of Chlamydia invasion, which is regulated synergistically by TarP and TmeA to mediate scission of Chlamydia- containing vesicles and initiate turnover of host proteins following invasion. Altogether, findings underscore the high degree of control Chlamydia has over its invasion process. + +TarP- deficient strains were incapable of localized and punctate Dyn2 recruitment, indicating that TarP signaling is required to prompt Dyn2 recruitment into a scission- competent configuration. Given that Dyn2 directly interacts with several TarP- associated actin regulators, including cortactin32, EPS833, and profilin32,34, we propose that the actin network generated by TarP signaling regulates Dyn2 function. Whether this interaction is mediated by direct interaction with actin, which has been reported previously20, or by various signaling molecules recruited by TarP is not known. One possibility is that TarP- mediated actin remodeling induces changes to the local environment that enrich and retain Dyn2 at sufficient quantities to achieve functionality. For example, robust actin polymerization can promote membrane curvature to support binding of Bin/amphiphysin/Rvs (BAR) domain proteins, some of which (e.g., SNX9) are known Dyn2 interactors40. This would also account for temporal regulation of Dyn2, wherein the timing of host protein recruitment influences both the concentration and orientation of Dyn2. Although our study demonstrates that Dyn2 and actin dynamics are functionally linked, a comprehensive + +<--- Page Split ---> + +model of Dyn2 involvement during invasion will require further characterization of its recruitment and activation. + +Recently, we reported that TarP signaling uniquely recruited host forms, which utilize profilin/actin complexes to acquire monomeric actin, and are important regulators of actin polymerization during Chlamydia entry. Moreover, the Arp2/3 complex is extensively associated with Dyn2 activity and collaborates with host forms to enhance actin remodeling during invasion. Robust actin remodeling provides a mechanism to ensure Dyn2 recruitment at sufficient levels; consequently, the pathways employed by Chlamydia to mediate actin nucleator activation are highly relevant points of Dyn2 regulation. For instance, we observed that TarP signaling via the PI3K/Rac1 axis, which regulates actin polymerization during invasion, also governed Dyn2 retention within entry sites. There is also precedence for Dyn2 modulation of actin remodeling, specifically insofar as disruption of Dyn2 dysregulates Rac localization and impairs actin dynamics within lamellipodia, highlighting that regulation of Dyn2 and Rac1 are functionally linked. Furthermore, actin stability and Dyn2 oligomerization are co-regulated, such that inhibition of Arp2/3 was sufficient to shift the balance of actin dynamics toward net disassembly, preventing scission of phagocytized particles and increasing Dyn2 persistence. Thus, destabilization of invasion-associated actin networks following Rac inhibition likely interferes with Dyn2 scission and subsequent turnover, yielding abnormally persistent signal. Conversely, Rac activation would promote Dyn2 function, a role demonstrably fulfilled by TarP. + +We also found that TmeA signaling promoted Dyn2 activation, wherein strains lacking TmeA exhibited defective uptake that could be rescued by Ryngo 1- 23 administration. Several lines of evidence suggest that TmeA regulates Dyn2 via its previously established role in actin remodeling. In- vitro assays identified that short actin filaments stimulate Dyn2 ring assembly, and that Ryngo 1- 23 promotes Dyn2 ring formation via a comparable mechanism. Thus, one possibility is that membrane localized TmeA generates actin filaments which scaffold the initial activation of Dyn2 at the plasma membrane. Signaling + +<--- Page Split ---> + +via the TmeA/N- WASP axis drives Arp2/3 activation, which synergizes with TarP signaling to promote actin polymerization and pathogen engulfment \(^{30,47}\) . Collaboration between TarP and TmeA may additionally regulate Dyn2, wherein TmeA- mediated actin polymerization functions after Dyn2 recruitment to promote oligomerization. Indeed, both TarP and TmeA were necessary for dynamin- dependent entry, as strains lacking either effector were insensitive to Dyn2 disruption, although the basis for their insensitivity differed. How might the regulatory contributions of TarP and TmeA be distinguished, given the shared importance of their respective actin remodeling functions? For TmeA, the involvement of N- WASP might offer some clues. This nucleation promoting factor harbors a proline- rich domain (PRD) that binds proteins with Src- homology 3 (SH3) domains. The SH3 domain- containing protein SNX9 interacts with dynamin and stimulates Dyn2 oligomerization \(^{40}\) and is important for C. trachomatis invasion \(^{51}\) . As such, interaction between N- WASP and SNX9 might account for Dyn2 dependence toward TmeA signaling. Intriguingly, TmeA also bears similarity with the C. pneumoniae secreted effector SemD \(^{52,53}\) , which recruits the BAR- domain proteins PACSIN and SNX9 to induce membrane curvature and promote pathogen engulfment. On this basis, TmeA- mediated Dyn2 regulation could manifest via the formation of SNX9/Dyn2 heterodimers, providing a mechanism of Dyn2 modulation distinct from its actin remodeling function. Therefore, there are at least two molecular interactions that uniquely link TmeA signaling with Dyn2 function. + +While the precise nature of how TmeA signaling modulates the Dyn2 GTPase cycle remains unknown, analysis of Dyn2 mutants may provide insight toward TmeA/Dyn2 regulation, and perhaps the mechanism of Ryngo- mediated rescue. Studies regarding the formation of progressive higher- order dynamin oligomers have benefited from various mutations that affect protein- protein interactions, GTPase activity, conformational changes during constriction, etc. Determining the exact mechanism of compound- mediated rescue following TmeA deletion will require elucidating which oligomeric species of Dyn2 is induced by either Ryngo or TmeA signaling. Mutations which prevent dynamin self- assembly (i.e. Dyn1 I670K \(^{54}\) ) or those which ablate membrane association (i.e. Dyn2 K562E \(^{55}\) ) could be informative toward this + +<--- Page Split ---> + +end, as these mutants are membrane scission- deficient and are not rescued by Ryngo50,56. Our working model predicts that these mutants should likewise be unaffected by TmeA signaling. Dyn1 K/E exhibits reduced affinity for actin filaments and is partially rescued by Ryngo in- vitro50, whereas Dyn2ΔPRD cannot bind SH3 domain- containing proteins and is dominant- negative for endocytosis57. Studies incorporating these mutants could disambiguate whether TmeA signaling operates by mediating Dyn2/actin interactions, or by promoting interaction with SH3 domain- containing proteins like SNX9. Using this report as a foundation, future studies could interrogate the effects of each Dyn2 mutant during Chlamydia invasion and determine the precise nature of effector signaling toward dynamin- dependent entry. + +Interestingly, unlike ΔTmeA, Ryngo treatment impaired wild- type Chlamydia invasion, restricting pathogen entry and yielding obvious defects in Dyn2 and actin recruitment. One explanation may be that in certain contexts, Ryngo stimulates Dyn2 oligomerization into a scission- incompetent configuration. FRET analysis of dynamin oligomerization found that Ryngo prompted the assembly of stacked Dyn2 rings around membrane tubules50, representing a lower- order oligomerization state that achieved insufficient GTPase activity to induce membrane scission. As such, co- stimulation of Dyn2 activation by both Ryngo and Chlamydia/TmeA signaling may interfere with the relative abundance of Dyn2 oligomeric species. Specifically, stimulation with Ryngo is expected to generate a disproportionate quantity of Dyn2 rings which interfere with further oligomerization steps. Elimination of Chlamydia- specific Dyn2 activation (i.e., ΔTmeA) may prevent overstimulation, encouraging proper assembly of higher- order, scission- competent Dyn2 oligomers. Meanwhile, whereas Ryngo pretreatment restored local Dyn2 recruitment at ΔTarP entry sites, it failed to prompt rapid engulfment of ΔTarP EBs and had no rescuing effect on its entry efficiency, suggesting that post- recruitment, Dyn2 needs to be primed for activation by Ryngo. + +Finally, our study identified that both Dyn2 and actin turnover were co- regulated. Mechanistically, Dyn2 turnover is intuitive, occurring either during or shortly after membrane scission as a function of GTP hydrolysis58,59. As such, Dyn2- mediated scission of Chlamydia- containing vacuoles may intrinsically prompt + +<--- Page Split ---> + +Dyn2 turnover while also providing a signal to initiate actin turnover. Interventions which prevent dynamin- mediated membrane fission also accumulate F- actin around tubulated membranes \(^{60,61}\) , whereas scission is consistently associated with actin turnover and sensitizes actin filaments toward cofilin- mediated severing \(^{29,62,63}\) . Furthermore, given that dynamin extensively interacts with actin- associated proteins \(^{28,63 - 65}\) , post- scission turnover of actin regulatory machinery alongside Dyn2 may shift actin regulation toward turnover. Importantly, actin polymerization during Chlamydia invasion is both intricately regulated and pathogen- directed \(^{66}\) ; consequently, turnover of actin and other invasion- associated host proteins could be regulated distinctly from turnover associated with routine engulfment of cellular cargoes (i.e., growth factors, transferrin). This could require additional factors that fine- tune their function and/or dynamics to accommodate pathogen- mediated uptake mechanisms. As such, further study is required to gain a more comprehensive perspective on host protein turnover post- invasion. + +Overall, our findings of Dyn2 modulation by TarP and TmeA fit well with the proposed pathogen- directed invasion model proposed by Byrne and Moulder \(^{67}\) . While the majority of molecular studies of chlamydial invasion focus on actin recruitment, we demonstrate here that latter stages are also targeted by TarP and TmeA, highlighting their central function in invasion, comprising a self- contained signaling module capable of mediating the initial, middle, and end stages of invasion. + +## Materials and Methods: + +## Cell and Bacterial Culture + +Green monkey kidney fibroblast- like (Cos7) cells and cervical adenocarcinoma epithelial (HeLa) cells were cultured at \(37^{\circ}C\) with \(5\%\) atmospheric CO2 in Dulbecco's Modified Eagle Medium (DMEM; Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with \(10\mu \mathrm{g / mL}\) gentamicin, \(2\mathrm{mM}\) L- glutamine, and \(10\%\) (v/v) filter- sterilized fetal bovine serum (FBS). HeLa and Cos7 cells were cultured for a maximum of + +<--- Page Split ---> + +15 passages for all experiments. McCoy B mouse fibroblasts (originally from Dr. Harlan Caldwell, NIH/NIAID) were cultured under comparable conditions. Chlamydia trachomatis serovar L2 (434/Bu) was propagated in McCoy cells and EBs were purified using a Gastrografin density gradient as described previously \(^{68}\) . + +## Reagents + +Wortmannin (Selleck, Houston, TX, USA) was diluted upon receipt to 40mM stock concentration in DMSO, Dynasore (Cayman Chemical, Ann Arbor, MI, USA) and EHop- 016 (Cayman) were diluted to 25mM stock concentration in DMSO, and Ryngo 1- 23 (Abcam, Cambridge, MA, USA) was diluted to 20mM stock concentration in DMSO. All inhibitors were dispensed into single use aliquots and stored at \(- 20^{\circ}C\) for no longer than 1 year after receipt. Wortmannin was diluted to a working concentration of 40nM (1:10000), Dynasore was diluted to a working concentration of 25 μM (1:1000), EHop- 016 was diluted to a working concentration of 10μM (1:2500), and Ryngo 1- 23 was diluted to a working concentration of 40μM (1:500), each using supplemented DMEM as diluent. + +## Invasion Assay + +C. trachomatis internalization efficiency was conducted using HeLa cells and was performed as described previously \(^{10}\) . Briefly, HeLa cells were seeded in 24-well plates containing acid-etched glass coverslips and allowed to adhere overnight. Cells were pretreated with Wortmannin (40nM), Dynasore (25μM), EHop-016 (10μM), or Ryngo (40μM) for 30 minutes prior to infection. Dyn2 siRNA or scramble RNA were transfected and allowed to incubate 24 hours prior to infection. Following inhibitor treatment or RNA interference, cells were infected with EBs derived from wild-type C. trachomatis L2 (434/Bu), C. trachomatis in which TarP, TmeA, or both were deleted by FRAEM (ΔTarP, ΔTmeA, ΔTmeA/ΔTarP), or C. trachomatis in which TarP or TmeA expression was restored by cis-complementation (cis-TarP, cis-TmeA) + +<--- Page Split ---> + +at MOI=50. EBs were allowed to attach onto HeLa cells for 30 min at \(4^{\circ}C\) before rinsing coverslips with cold HBSS, followed by addition of supplemented DMEM prewarmed to \(37^{\circ}C\) , before incubating cells at \(37^{\circ}C\) for 10 min. After incubation, cells were stringently washed with cold HBSS containing \(100\mu g / mL\) heparin to remove any transiently adherent EBs before fixation in \(4\%\) paraformaldehyde at room temperature for 15 min. Fixed cells were labeled with a mouse monoclonal anti- MOMP antibody (Novus Biologicals, Centennial, CO, USA #NB10066403), rinsed with \(1x\) PBS, and fixed once more in \(4\%\) paraformaldehyde for 10 min. Next, cells were permeabilized using \(0.1\%\) (w/v) Triton X- 100 for 10 minutes at room temperature, rinsed with HBSS and labeled with rabbit polyclonal anti- Chlamydia trachomatis antibody (Abcam ab252762). Cells were then rinsed in \(1x\) PBS and labeled with Alexa Fluor 594 anti- mouse (ThermoFisher #A11032, Waltham, MA, USA) and Alexa Fluor 488 anti- rabbit (ThermoFisher #A11034) IgG secondary antibodies. Coverslips were mounted and observed on a Nikon CSU- W1 confocal microscope (Nikon, Melville, NY, USA), obtaining Z- stacks using a 0.3 micron step size across the height of the cell monolayer. Monolayer Z- stacks were transformed via Z- projection according to maximal fluorescence intensity in ImageJ prior to quantifying percent invasion efficiency as follows: total EBs (green) – extracellular EBs (red)/total EBs (green) x \(100\%\) . + +## Quantitative live cell imaging of Chlamydia invasion + +Cos7 cells were seeded onto libdi \(\mu\) - Slide 8- well glass- bottomed chambers (Ibidii, Fitchburg, WI, USA) and allowed to adhere overnight prior to transfection. Cells were transfected with fluorescent proteins as indicated, using Lipofectamine 3000 (Thermo Fisher, Waltham, MA, USA) according to manufacturer directions. Transfection was allowed to proceed overnight before replacing media with fresh DMEM + \(10\%\) FBS/2 mM L- glutamine and allowing protein expression to continue for a total of 24 hours post- transfection. Transfection efficiency was verified on a Nikon CSU- W1 spinning disk confocal microscope prior to application of DMEM containing Wortmannin (40nM), Dynasore (25μM), EHop- 016 (10μM), or + +<--- Page Split ---> + +Ryngo (40μM). For RNA interference, Dyn2 siRNA or scramble RNA was co- transfected alongside GFP- actin or mRuby- LifeAct and allowed to incubate for 24 hours prior to imaging. Wells were individually infected with CMTPX- labeled wild- type C. trachomatis L2 (434/Bu), unless otherwise indicated, at MOI=20 and promptly imaged using a 60x objective (NA 1.40) in a heated and humidified enclosure. Images were collected once every 20 seconds for 30 minutes, with focal plane maintained using an infrared auto- focusing system. Upon completion of the imaging protocol, the next well was infected and imaging repeated; mock- treated wells were imaged first to allow inhibitor treatment sufficient time to achieve inhibition. Images were compiled into videos using NIH ImageJ and analyzed to identify protein recruitment events. The mean fluorescence intensity (MFI) of recruitment events was measured for each timepoint alongside the local background MFI of a concentric region immediately outside the recruitment event. Background MFI was subtracted from recruitment MFI for each timepoint and normalized as percent maximal fluorescence intensity for each timepoint, repeating this normalization process for each recruitment event. + +## RNA interference + +Cos- 7 or HeLa cells were seeded onto Ibidi \(\mu\) - Slide 8- well glass- bottomed chambers (live- cell imaging) or in 24- well plates containing acid- etched glass coverslips (invasion assay) and allowed to adhere overnight. Mission esiRNAs were custom- ordered to target Cos7 Dyn2 mRNA, ensuring that the resultant esiRNA targeted a shared sequence found in all recorded mRNA transcript variants. Cells were transfected with either 100 nM Mission anti- Dyn2 esiRNA (Eupheria Biotech, Dresden, Germany) or 100 nM Trilencer- 27 Universal scrambled negative control (Origene SR30004, Rockville, MD, USA) using Lipofectamine RNAiMAX reagent (Thermo Fisher) according to manufacturer directions. Incubation was allowed to proceed for 24 hours before conducting live- cell imaging or invasion assays using methods described + +<--- Page Split ---> + +earlier. Lysates for Western blotting were obtained from Cos7 cells by applying 2x Laemmli buffer to cells after the completion of live- cell imaging. + +## Western Blotting + +Lysates generated as described above were resolved via SDS- PAGE in \(10\%\) polyacrylamide gels at 120 volts for 1.5 hours or until the dye front has begun to evacuate the bottom of the gel cassette. Gels were transferred onto \(0.45\mu M\) pore size nitrocellulose in 1x Towbin buffer \(+10\%\) methanol at \(90~\mathrm{mA}\) for 16 hours. Western blots were blocked in \(5\%\) bovine serum albumin for 1 hour, briefly rinsed in Tris buffered saline \(+0.1\%\) Tween- 20 (TBST) and incubated with appropriate primary antibody for 1 hour. Blots were then washed three times for 5 minutes in TBST and probed with appropriate HRP- conjugated secondary antibodies for 1 hour. Protein bands were resolved by chemiluminescence using Immobilon Western HRP Substrate (Millipore Sigma, St Louis, MO, USA). Dynamin 2 knockdown efficiency was calculated by densitometry analysis, comparing the ratio of Dyn2 antibody signal (Thermo PA1- 661) against \(\beta\) - actin loading control (Abcam ab49900). + +## Dextran Uptake Assay + +Cos7 cells were seeded onto 24- well plates containing acid- etched glass coverslips and allowed to adhere overnight. Cells were treated with \(40~\mu M\) Ryngo 1- 23 or DMSO in media containing \(100~\mu g / mL\) Dextran Alexa Fluor 647; 10,000 MW (Thermo D22914) and incubated at \(37^{\circ}C\) for 30 minutes. Cells were infected with wild- type or mutant Chlamydia strains at MOI=50, synchronizing infection by sedimentation at \(4^{\circ}C\) on a rocking incubator for 30 minutes. Infection was initiated by addition of prewarmed media, followed by incubation in a \(37^{\circ}C\) incubator for 20 minutes prior to fixation in \(4\%\) paraformaldehyde for 10 minutes at room temperature. Fluorescent dextran and inhibitor were maintained in media for each indicated stage. Cells were then permeabilized using \(0.1\%\) (w/v) Triton X- 100 for 10 minutes at room temperature, + +<--- Page Split ---> + +rinsed with HBSS and labeled with a mouse monoclonal anti- MOMP antibody (Novus #NB10066403). Cells were rinsed in 1x PBS and labeled with Alexa Fluor 488 anti- mouse (ThermoFisher # A- 11001) IgG secondary antibody. Coverslips were mounted and observed on a Nikon CSU- W1 confocal microscope (Nikon), obtaining Z- stacks using a 0.3 micron step size across the height of the cell monolayer. Monolayer Z- stacks were transformed via Z- projection according to maximal fluorescence intensity in ImageJ prior to quantifying the percentage of elementary bodies which colocalize with fluorescent dextran, according to the following equation: [ Dextran\* EBs (magenta/green) / Total EBs (green) ] x 100%. + +## Plasmids and DNA preparation + +pEGFP- Actin- C169 was a gift from Dr Scott Grieshaber (University of Idaho), and mRuby- LifeAct- 7 (Addgene plasmid #54560) was a gift from Michael Davidson. Dyn2- pmCherryN1 was a gift from Christien Merrifield (Addgene plasmid #27689), RFP Dynamin2 K44A was a gift from Jennifer Lippincott- Schwartz (Addgene plasmid #128153), WT Dyn2 pEGFP was a gift from Sandra Schmid (Addgene plasmid #34686), and GFP- Dynamin 2 K44A was a gift from Pietro De Camilli (Addgene plasmid #22301). Upon receipt, bacterial stab cultures were streak- plated onto LB agar containing appropriate antibiotic (Kanamycin, carbenicillin) for each plasmid. Resultant antibiotic- resistant colonies were selected and propagated in LB broth + antibiotic for plasmid isolation prior to sequence verification. All plasmids were isolated using MiniPrep DNA isolation kits (Qiagen, Valencia, CA, USA) following a variant protocol for DNA isolation termed MiraPrep70. Following plasmid isolation, the eluate was precipitated by addition of 3M sodium acetate (Invitrogen, Waltham, MA, USA) at 10% (v/v) of eluate volume followed by addition of 250% (v/v) absolute ethanol calculated after addition of sodium acetate. The mixture was incubated at 4°C overnight and centrifuged at 14,000×g for 15 minutes at 4°C. Supernatant was removed and 70% ethanol was added, followed by centrifugation at 14,000×g for 10 minutes at 4°C. Supernatant was removed once more, and precipitated + +<--- Page Split ---> + +DNA was resuspended in nuclease- free \(\mathsf{H}_2\mathsf{O}\) . Sequencing was conducted by Eurofins Genomics (Louisville, KY, USA), using standard sequencing primers provided by the company. + +## Graphs and statistical analysis + +Violin plots were made using the ggplot2 base package (version 3.1.0) as a component of the Tidyverse package (https://cran.r- project.org/web/packages/tidyverse/index.html) in rStudio (version 4.0.3). Wilcoxon ranked- sum tests to determine statistical significance between violin plots were conducted using base R statistics in rStudio. Recruitment plots, invasion assays, and all statistics associated with these data (pairwise T- test followed by Bonferroni post- analysis, SEM) were performed in Excel (Microsoft, Redmond, WA, USA). All graphs were assembled using the free and open- source software GNU Image Manipulation Program (GIMP, https://www.gimp.org/) and Inkscape (https://inkscape.org/). Proposed model for Dyn2 oligomerization (Figs. 1- 6, S4) was assembled using BioRender (https://app.biorender.com/). + +## Acknowledgements + +The authors would like to acknowledge the following investigators for their generosity - Dr. Ken Fields (University of Kentucky) for providing the mutant strains of C. trachomatis described in this study; pEGFP- Actin- C169 from Dr. Scott Grieshaber (University of Idaho), mRuby- LifeAct- 7 (Addgene plasmid #54560) from Dr. Michael Davidson; Dyn2- pmCherryN1 from Dr. Christien Merrifield (Addgene plasmid #27689); RFP Dynamin2 K44A from Dr. Jennifer Lippincott- Schwartz (Addgene plasmid #128153), WT Dyn2 pEGFP from Dr. Sandra Schmid (Addgene plasmid #34686), and GFP- Dynamin 2 K44A was from Dr. Pietro De Camilli (Addgene plasmid #22301). The authors would also like to thank members of the Carabeo laboratory for helpful suggestions. This work is funded by NIH AI065545 to R.C. + +<--- Page Split ---> + +## Sources Cited: + +1. Malhotra, M., Sood, S., Mukherjee, A., Muralidhar, S. & Bala, M. Genital Chlamydia trachomatis: An update. Indian J Med Res 138, 303-316 (2013). + +2. Elwell, C., Mirrashidi, K. & Engel, J. Chlamydia cell biology and pathogenesis. Nature Reviews Microbiology 14, 385-400 (2016). + +3. Stephens, R. S., Koshiyama, K., Lewis, E. & Kubo, A. Heparin-binding outer membrane protein of chlamydiae. Mol Microbiol 40, 691-699 (2001). + +4. Gitsels, A., Sanders, N. & Vanrompay, D. Chlamydial Infection From Outside to Inside. Front Microbiol 10, (2019). + +5. Chen, Y.-S. et al. The Chlamydia trachomatis type III secretion chaperone Slc1 engages multiple early effectors, including TepP, a tyrosine-phosphorylated protein required for the recruitment of CrkII-II to nascent inclusions and innate immune signaling. PLoS Pathog 10, e1003954 (2014). + +6. Fields, K. A., Mead, D. J., Dooley, C. A. & Hackstadt, T. Chlamydia trachomatis type III secretion: evidence for a functional apparatus during early-cycle development. Mol Microbiol 48, 671-683 (2003). + +7. Romero, M. D. & Carabeo, R. A. Distinct roles of the Chlamydia trachomatis effectors TarP and TmeA in the regulation of formin and Arp2/3 during entry. Journal of Cell Science 135, jcs260185 (2022). + +8. Keb, G., Ferrell, J., Scanlon, K. R., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeA Directly Activates N-WASP To Promote Actin Polymerization and Functions Synergistically with TarP during Invasion. mBio 12, (2021). + +9. Faris, R., McCullough, A., Andersen, S. E., Moninger, T. O. & Weber, M. M. The Chlamydia trachomatis secreted effector TmeA hijacks the N-WASP-ARP2/3 actin remodeling axis to facilitate cellular invasion. PLOS Pathogens 16, e1008878 (2020). + +<--- Page Split ---> + +10. Carabeo, R. A., Grieshaber, S. S., Fischer, E. & Hackstadt, T. Chlamydia trachomatis Induces Remodeling of the Actin Cytoskeleton during Attachment and Entry into HeLa Cells. Infect Immun 70, 3793-3803 (2002). + +11. Ford, C., Nans, A., Boucrot, E. & Hayward, R. D. Chlamydia exploits filopodia capture and a macropinocytosis-like pathway for host cell entry. PLOS Pathogens 14, e1007051 (2018). + +12. Hybiske, K. & Stephens, R. S. Mechanisms of Chlamydia trachomatis Entry into Nonphagocytic Cells. Infection and Immunity 75, 3925-3934 (2007). + +13. Nans, A., Saibil, H. R. & Hayward, R. D. Pathogen-host reorganization during Chlamydia invasion revealed by cryo-electron tomography. Cellular Microbiology 16, 1457-1472 (2014). + +14. Ferguson, S. M. & De Camilli, P. Dynamin, a membrane remodelling GTPase. Nat Rev Mol Cell Biol 13, 75-88 (2012). + +15. Ford, M. G. J., Jenni, S. & Nunnari, J. The crystal structure of dynamin. Nature 477, 561-566 (2011). + +16. Muhlberg, A. B., Warnock, D. E. & Schmid, S. L. Domain structure and intramolecular regulation of dynamin GTPase. EMBO J 16, 6676-6683 (1997). + +17. Warnock, D. E., Hinshaw, J. E. & Schmid, S. L. Dynamin self-assembly stimulates its GTPase activity. J Biol Chem 271, 22310-22314 (1996). + +18. Antonny, B. et al. Membrane fission by dynamin: what we know and what we need to know. EMBO J 35, 2270-2284 (2016). + +19. Ross, J. A. et al. Dimeric Endophilin A2 Stimulates Assembly and GTPase Activity of Dynamin 2. Biophys J 100, 729-737 (2011). + +20. Gu, C. et al. Direct dynamin-actin interactions regulate the actin cytoskeleton. EMBO J 29, 3593-3606 (2010). + +21. Praefcke, G. J. K. & McMahon, H. T. The dynamin superfamily: universal membrane tubulation and fission molecules? Nat Rev Mol Cell Biol 5, 133-147 (2004). + +<--- Page Split ---> + +22. Carabeo, R. A., Dooley, C. A., Grieshaber, S. S. & Hackstadt, T. Rac interacts with Abi-1 and WAVE2 to promote an Arp2/3-dependent actin recruitment during chlamydial invasion. Cellular Microbiology 9, 2278–2288 (2007). + +23. Caven, L. & Carabeo, R. A. Pathogenic Puppetry: Manipulation of the Host Actin Cytoskeleton by Chlamydia Trachomatis. (2019). + +24. Thwaites, T. R., Pedrosa, A. T., Peacock, T. P. & Carabeo, R. A. Vinculin Interacts with the Chlamydia Effector TarP Via a Tripartite Vinculin Binding Domain to Mediate Actin Recruitment and Assembly at the Plasma Membrane. Front. Cell. Infect. Microbiol. 5, (2015). + +25. Boleti, H., Benmerah, A., Ojcius, D. M., Cerf-Bensussan, N. & Dautry-Varsat, A. Chlamydia infection of epithelial cells expressing dynamin and Eps15 mutants: clathrin-independent entry into cells and dynamin-dependent productive growth. Journal of Cell Science 112, 1487–1496 (1999). + +26. Yarar, D., Waterman-Storer, C. M. & Schmid, S. L. SNX9 Couples Actin Assembly to Phosphoinositide Signals and Is Required for Membrane Remodeling during Endocytosis. Developmental Cell 13, 43–56 (2007). + +27. Schafer, D. A. et al. Dynamin2 and Cortactin Regulate Actin Assembly and Filament Organization. Curr Biol 12, 1852–1857 (2002). + +28. Taylor, M. J., Perrais, D. & Merrifield, C. J. A high precision survey of the molecular dynamics of mammalian clathrin-mediated endocytosis. PLoS Biol 9, e1000604 (2011). + +29. Taylor, M. J., Lampe, M. & Merrifield, C. J. A feedback loop between dynamin and actin recruitment during clathrin-mediated endocytosis. PLoS Biol 10, e1001302 (2012). + +30. Keb, G., Ferrell, J., Scanlon, K. R., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeA Directly Activates N-WASP To Promote Actin Polymerization and Functions Synergistically with TarP during Invasion. mBio 12, e02861-20 (2021). + +<--- Page Split ---> + +31. Lane, B. J., Mutchler, C., Khodor, S. A., Grieshaber, S. S. & Carabeo, R. A. Chlamydial Entry Involves TARP Binding of Guanine Nucleotide Exchange Factors. PLOS Pathogens 4, e1000014 (2008). + +32. Orth, J. D. & McNiven, M. A. Dynamin at the actin-membrane interface. Current Opinion in Cell Biology 15, 31-39 (2003). + +33. Auciello, G., Cunningham, D. L., Tatar, T., Heath, J. K. & Rappoport, J. Z. Regulation of fibroblast growth factor receptor signalling and trafficking by Src and Eps8. Journal of Cell Science 126, 613-624 (2013). + +34. Witke, W. et al. In mouse brain profilin I and profilin II associate with regulators of the endocytic pathway and actin assembly. EMBO J 17, 967-976 (1998). + +35. Krueger, E. W., Orth, J. D., Cao, H. & McNiven, M. A. A dynamin-cortactin-Arp2/3 complex mediates actin reorganization in growth factor-stimulated cells. Mol Biol Cell 14, 1085-1096 (2003). + +36. McKuen, M. J., Mueller, K. E., Bae, Y. S. & Fields, K. A. Fluorescence-Reported Allelic Exchange Mutagenesis Reveals a Role for Chlamydia trachomatis TmeA in Invasion That Is Independent of Host AHNAK. Infect Immun 85, (2017). + +37. Ghosh, S. et al. Fluorescence-Reported Allelic Exchange Mutagenesis-Mediated Gene Deletion Indicates a Requirement for Chlamydia trachomatis Tarp during In Vivo Infectivity and Reveals a Specific Role for the C Terminus during Cellular Invasion. Infection and Immunity 88, (2020). + +38. Swanson, J. A. & Watts, C. Macropinocytosis. Trends in Cell Biology 5, 424-428 (1995). + +39. Gu, C. et al. Regulation of Dynamin Oligomerization in Cells: The Role of Dynamin-Actin Interactions and Its GTPase Activity. Traffic 15, 819-838 (2014). + +40. Soulet, F., Yarar, D., Leonard, M. & Schmid, S. L. SNX9 Regulates Dynamin Assembly and Is Required for Efficient Clathrin-mediated Endocytosis. Mol Biol Cell 16, 2058-2067 (2005). + +41. Schlüter, K., Jockusch, B. M. & Rothkegel, M. Profilins as regulators of actin dynamics. Biochim Biophys Acta 1359, 97-109 (1997). + +<--- Page Split ---> + +42. Itoh, T. et al. Dynamin and the Actin Cytoskeleton Cooperatively Regulate Plasma Membrane Invagination by BAR and F-BAR Proteins. Developmental Cell 9, 791-804 (2005). + +43. Merrifield, C. J., Qualmann, B., Kessels, M. M. & Almers, W. Neural Wiskott Aldrich Syndrome Protein (N-WASP) and the Arp2/3 complex are recruited to sites of clathrin-mediated endocytosis in cultured fibroblasts. European Journal of Cell Biology 83, 13-18 (2004). + +44. Benesch, S. et al. N-WASP deficiency impairs EGF internalization and actin assembly at clathrin-coated pits. Journal of Cell Science 118, 3103-3115 (2005). + +45. Schlunck, G. et al. Modulation of Rac Localization and Function by Dynamin. MBoC 15, 256-267 (2004). + +46. Marie-Anais, F., Mazzolini, J., Herit, F. & Niedergang, F. Dynamin-Actin Cross Talk Contributes to Phagosome Formation and Closure. Traffic 17, 487-499 (2016). + +47. Romero, M. D. & Carabeo, R. A. Distinct roles of the Chlamydia trachomatis effectors TarP and TmeA in the regulation of formin and Arp2/3 during entry. J Cell Sci 135, jcs260185 (2022). + +48. Faris, R., McCullough, A., Andersen, S. E., Moninger, T. O. & Weber, M. M. The Chlamydia trachomatis secreted effector TmeA hijacks the N-WASP-ARP2/3 actin remodeling axis to facilitate cellular invasion. PLOS Pathogens 16, e1008878 (2020). + +49. Gu, C. et al. Direct dynamin-actin interactions regulate the actin cytoskeleton. EMBO J 29, 3593-3606 (2010). + +50. Gu, C. et al. Regulation of dynamin oligomerization in cells: the role of dynamin-actin interactions and its GTPase activity. Traffic 15, 819-838 (2014). + +51. Ford, C., Nans, A., Boucrot, E. & Hayward, R. D. Chlamydia exploits filopodia capture and a macropinocytosis-like pathway for host cell entry. PLOS Pathogens 14, e1007051 (2018). + +<--- Page Split ---> + +52. Spona, D., Hanisch, P. T., Hegemann, J. H. & Mölleken, K. A single chlamydial protein reshapes the plasma membrane and serves as recruiting platform for central endocytic effector proteins. Commun Biol 6, 520 (2023). + +53. Scanlon, K. R., Keb, G., Wolf, K., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeB antagonizes actin polymerization via direct interference with Arp2/3 activity. Frontiers in Cellular and Infection Microbiology 13, (2023). + +54. Song, B. D., Yarar, D. & Schmid, S. L. An Assembly-incompetent Mutant Establishes a Requirement for Dynamin Self-assembly in Clathrin-mediated Endocytosis In Vivo. Mol Biol Cell 15, 2243-2252 (2004). + +55. Chin, Y.-H. et al. Dynamin-2 mutations associated with centronuclear myopathy are hypermorphic and lead to T-tubule fragmentation. Human Molecular Genetics 24, 5542-5554 (2015). + +56. Hamasaki, E. et al. The Lipid-Binding Defective Dynamin 2 Mutant in Charcot-Marie-Tooth Disease Impairs Proper Actin Bundling and Actin Organization in Glomerular Podocytes. Front Cell Dev Biol 10, 884509 (2022). + +57. Szaszák, M. et al. Role of the Proline-rich Domain of Dynamin-2 and Its Interactions with Src Homology 3 Domains during Endocytosis of the AT1 Angiotensin Receptor. Journal of Biological Chemistry 277, 21650-21656 (2002). + +58. Antonny, B. et al. Membrane fission by dynamin: what we know and what we need to know. The EMBO Journal 35, 2270-2284 (2016). + +59. Warnock, D. E., Hinshaw, J. E. & Schmid, S. L. Dynamin self-assembly stimulates its GTPase activity. J Biol Chem 271, 22310-22314 (1996). + +60. Ferguson, S. M. et al. Coordinated actions of actin and BAR proteins upstream of dynamin at endocytic clathrin-coated pits. Dev Cell 17, 811-822 (2009). + +<--- Page Split ---> + +676 61. Kruchten, A. E. & McNiven, M. A. Dynamin as a mover and pincher during cell migration and invasion. J Cell Sci 119, 1683–1690 (2006). + +678 62. Grassart, A. et al. Actin and dynamin2 dynamics and interplay during clathrin-mediated endocytosis. Journal of Cell Biology 205, 721–735 (2014). + +680 63. Mooren, O. L., Kotova, T. I., Moore, A. J. & Schafer, D. A. Dynamin2 GTPase and Cortactin Remodel Actin Filaments. J Biol Chem 284, 23995–24005 (2009). + +682 64. Shin, N. et al. Sorting Nexin 9 Interacts with Dynamin 1 and N-WASP and Coordinates Synaptic Vesicle Endocytosis. Journal of Biological Chemistry 282, 28939–28950 (2007). + +684 65. Kessels, M. M., Engqvist-Goldstein, Å. E. Y., Drubin, D. G. & Qualmann, B. Mammalian Abp1, a Signal-Responsive F-Actin–Binding Protein, Links the Actin Cytoskeleton to Endocytosis via the Gtpase Dynamin. Journal of Cell Biology 153, 351–366 (2001). + +687 66. Caven, L. & Carabeo, R. A. Pathogenic Puppetry: Manipulation of the Host Actin Cytoskeleton by Chlamydia trachomatis. Int J Mol Sci 21, 90 (2019). + +689 67. Byrne, G. I. & Moulder, J. W. Parasite-specified phagocytosis of Chlamydia psittaci and Chlamydia trachomatis by L and HeLa cells. Infect Immun 19, 598–606 (1978). + +691 68. Caldwell, H. D., Kromhout, J. & Schachter, J. Purification and partial characterization of the major outer membrane protein of Chlamydia trachomatis. Infect Immun 31, 1161–1176 (1981). + +693 69. Heinzen, R. A., Grieshaber, S. S., Van Kirk, L. S. & Devin, C. J. Dynamics of Actin-Based Movement by Rickettsia rickettsii in Vero Cells. Infect Immun 67, 4201–4207 (1999). + +695 70. Pronobis, M. I., Deuitch, N. & Peifer, M. The Miraprep: A Protocol that Uses a Miniprep Kit and Provides Maxiprep Yields. PLOS ONE 11, e0160509 (2016). + +<--- Page Split ---> +![](images/Figure_3.jpg) + + +<--- Page Split ---> + +## Figure 1: Dynamin 2 and actin are co-recruited during Chlamydia entry + +(A) Cos7 cells were transfected with GFP-Dyn2 WT or K44A (DN) and miRFP-670 LifeAct for 24 hours prior to infection with wild-type Chlamydia at MOI=20. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes to identify sites exhibiting actin and Dyn2 co-recruitment. Scale bar = 1 micron. (B) Schematic depicting Dyn2 oligomerization, highlighting disruption of Dyn2 scission by K44A mutation. (C) Mean fluorescence intensity (MFI) of Dyn2 recruitment at Chlamydia entry sites was quantified, normalized as percent maximal MFI, and plotted onto a line graph depicting %max Dyn2 MFI +/- SEM for each timepoint. Background Dyn2 fluorescence was subtracted prior to normalization, which was performed independently for each Dyn2 WT and DN recruitment event. (D) Kymographs depicting RFP-Dyn2, GFP-Chlamydia, and far red actin signal over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (E-G) Detailed analysis of each recruitment event obtained via live cell imaging, plotting the (E) internalization duration, (F) rate of Dyn2 recruitment and (G) Dyn2 turnover of each event on a violin plot with inset boxplot reporting the median value +/- SD for each condition. (E) Internalization duration was quantified by calculating the elapsed time between initiation of protein recruitment and termination of pathogen entry, as detailed in Fig. S1. Individual rates of Dyn2 recruitment (F) and turnover (G) were calculated by measuring the slope derived from basal Dyn2 MFI to peak MFI for recruitment, and peak Dyn2 MFI to basal MFI for turnover, as detailed in Fig. S1. Data was obtained from a minimum N=23 individual rates. Statistical significance was determined by Wilcoxon ranked-sum. All data are representative of 3 independent experiments, * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001. + +<--- Page Split ---> +![](images/Figure_4.jpg) + + +<--- Page Split ---> + +## Figure 2: Disruption of Dynamin 2 restricts actin turnover and Chlamydia entry + +(A) Schematic depicting Dyn2 oligomerization, highlighting disruption of Dyn2 scission by Dynasore treatment. (B) Cos7 cells were transfected with GFP actin for 24 hours prior to infection with RFP CMTPX-stained wild-type Chlamydia (MOI=20). Infection was monitored by live-cell confocal microscopy, obtaining images every 20 seconds for 30 minutes to identify sites of actin recruitment proximal to invading bacteria. Actin recruitment at pathogen entry sites was quantified as described earlier (Fig. 1C) and plotted as %max actin MFI for each timepoint +/- SEM compiled from a minimum N=36 recruitment events. Upon completion of imaging, cells which received either scramble RNA or Dyn2 siRNA were lysed in 2x Laemmli buffer, resolving protein expression via Western blot to determine the knockdown efficiency of Dyn2 siRNA compared to actin loading control. Kinetics of (C) actin recruitment and (D) actin turnover, and (F) internalization duration were obtained using the same methodology described in Fig. 1E-G. Violin plots contain a minimum N=34 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank-sum. (E) Kymographs depicting RFP-Dyn2 and GFP-Chlamydia fluorescence over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (G,H) HeLa cells were infected with the indicated Chlamydia strain at MOI=50 and stained using the "in-and-out" method which distinguishes non-internalized EBs from total cell-associated EBs, as described in Materials and Methods. (G) Cells were pre-treated with 25 μM Dynasore for 30 minutes prior to infection, or (H) transfected with either scramble or Dyn2-specific siRNA for 24 hours prior to infection. Invasion efficiency of each Chlamydia strain was plotted as mean +/- SEM. Data was collected from 15 fields, with each field containing an average of 50 Chlamydia. Statistical significance was determined by pairwise T-test with Bonferroni post-correction. All data are representative of at least 3 independent experiments, * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 3
+ +(A) Cos7 cells were transfected with GFP- or RFP-Dyn2 WT for 24 hours prior to infection with wild-type or \(\Delta \mathsf{Tm eA}\) EBs at MOI=20 in the presence or absence of \(10~\mu \mathrm{M}\) EHop-016. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes, identifying sites exhibiting Dyn2 recruitment during Chlamydia entry. Scale bar = 1 micron. (B) Schematic depicting TarP signaling via PI3K/Rac1, subsequent recruitment of actin and Dyn2, and Dyn2 oligomerization, highlighting EHop-016 inhibition of Rac1 and promotion of Dyn2 recruitment by TarP (C) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=20 recruitment events. (D) HeLa cells were + +<--- Page Split ---> + +treated with 10μM EHop- 016 for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the "in- and- out" method described earlier to quantify pathogen entry efficiency. Statistical significance was determined by pairwise T- test with Bonferroni post- correction. (E) Kymographs depicting RFP- Dyn2 and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=20 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank- sum. All data are representative of at least 3 independent experiments, \* P ≤ 0.05, \*\* P ≤ 0.01, \*\*\* P ≤ 0.001. + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
Figure 4
+ +(A) Cos7 cells were transfected with GFP- or RFP-Dyn2 WT for 24 hours prior to infection with wild-type or \(\Delta \mathsf{Tm eA}\) EBs at MOI=20 in the presence or absence of 40nM Wortmannin. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes identifying sites exhibiting Dyn2 recruitment during Chlamydia entry. Scale bar = 1 micron. (B) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=18 recruitment events. (C) HeLa cells were treated with 40nM Wortmannin for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the "in-and-out" method described earlier to quantify pathogen entry + +<--- Page Split ---> + +efficiency. Statistical significance was determined by pairwise T- test with Bonferroni post- correction. (D) 776 Schematic depicting TarP signaling via PI3K/Rac1, subsequent recruitment of actin and Dyn2, and Dyn2 777 oligomerization, highlighting Wortmannin inhibition of PI3K and promotion of Dyn2 recruitment by TarP 778 (E) Kymographs depicting RFP- Dyn2 and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top 779 arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen 780 entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were 781 obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=20 782 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon 783 Rank- sum. All data are representative of at least 3 independent experiments, \* P ≤ 0.05, \*\* P ≤ 0.01, \*\*\* 784 P ≤ 0.001. + +<--- Page Split ---> +![PLACEHOLDER_39_0] + +
Figure 5
+ +(A) HeLa cells were treated with \(40\mu M\) Ryngo 1-23 for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the “in-and-out” method described earlier to quantify pathogen entry efficiency. Statistical significance was determined by pairwise T-test with Bonferroni post-correction. (B) Cos7 cells were transfected with RFP-Dyn2 WT for 24 hours prior to infection with \(\Delta\) TmeA EBs at MOI=20 in the presence or absence of \(40\mu M\) Ryngo 1-23. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes, highlighting Dyn2 recruitment at \(\Delta\) TmeA entry sites. Scale bar = 1 micron. (C) Schematic depicting Dyn2 oligomerization, promotion of Dyn2 self-assembly by TmeA signaling and enhancement of Dyn2 ring assembly via Ryngo 1-23 treatment. (D) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=19 recruitment events. (E) Kymographs depicting RFP-Dyn2 and GFP-Chlamydia fluorescence over a 30 + +<--- Page Split ---> + +minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=19 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank- sum. All data are representative of at least 3 independent experiments, \* P \(\leq 0.05\) , \*\* P \(\leq 0.01\) , \*\*\* P \(\leq 0.001\) . + +<--- Page Split ---> +![PLACEHOLDER_41_0] + +
Figure 6
+ +Figure 6: Actin turnover is correlated with Dynamin 2 activation status and Chlamydia uptake + +(A) Cos7 cells were transfected with GFP-Actin or mRuby-LifeAct for 24 hours prior to infection with wild-type or \(\Delta\) TmeA EBs at MOI=20 in the presence or absence of \(40\mu \mathrm{M}\) Ryngo 1-23, monitoring pathogen invasion via live-cell confocal microscopy. Actin recruitment was quantified as described earlier (Fig. 1C) and plotted as %max actin MFI for each timepoint +/- SEM compiled from a minimum N=21 recruitment events. (B,C) Kinetics of Dyn2 recruitment (B) and turnover (C) were obtained using the same methodology described in Fig. 1F-G. Violin plots contain a minimum N=21 individual events, reporting the median rate +/- SD. Statistical significance was determined by Wilcoxon Rank-sum. (D) Schematic depicting Dyn2 + +<--- Page Split ---> + +813 oligomerization, promotion of Dyn2 self-assembly by TmeA signaling, enhancement of Dyn2 ring assembly 814 via Ryngo 1- 23 treatment, and proposed initiation of actin turnover following Dyn2 scission. (E) 815 Kymographs depicting RFP-actin and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top arrow 816 indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F) 817 Internalization duration was quantified by calculating the elapsed time between initiation of actin 818 recruitment and termination of pathogen entry, as detailed in Fig. S1. Violin plots contain a minimum N=21 819 individual events, reporting the median internalization duration +/- SD. Statistical significance was 820 determined by Wilcoxon Rank-sum. All data are representative of at least 3 independent experiments, \* P 821 \(\leq 0.05\) , \*\* P \(\leq 0.01\) , \*\*\* P \(\leq 0.001\) . + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- VideoS1.mp4- VideoS2.mp4- VideoS3.mp4- SupplementalFiguresNatComm.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2_det.mmd b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..9db3e5b881b619ed57b1615309eddf5eab3f44f0 --- /dev/null +++ b/preprint/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2/preprint__166b44bc2d53ecba0245c2b5a60c3fbc59ca3e4edfaad14eff73dea280d95de2_det.mmd @@ -0,0 +1,616 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 950, 207]]<|/det|> +# Dynamin-dependent entry of Chlamydia trachomatis is sequentially regulated by the effectors TarP and TmeA + +<|ref|>text<|/ref|><|det|>[[44, 230, 140, 248]]<|/det|> +Rey Carabeo + +<|ref|>text<|/ref|><|det|>[[52, 257, 279, 273]]<|/det|> +rey.carabeo@unmc.edu + +<|ref|>text<|/ref|><|det|>[[44, 302, 750, 369]]<|/det|> +University of Nebraska Medical Center https://orcid.org/0000- 0002- 5708- 5493 Matthew Romero University of Nebraska Medical Center https://orcid.org/0000- 0002- 3459- 1019 + +<|ref|>sub_title<|/ref|><|det|>[[44, 409, 103, 427]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 447, 135, 465]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 485, 355, 504]]<|/det|> +Posted Date: September 27th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 523, 474, 542]]<|/det|> +DOI: https://doi.org/10.21203/rs.3. rs- 3376558/v1 + +<|ref|>text<|/ref|><|det|>[[42, 560, 914, 601]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 621, 533, 640]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 677, 916, 720]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on June 10th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49350- 6. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[75, 90, 884, 113]]<|/det|> +# Dynamin-dependent entry of Chlamydia trachomatis is sequentially regulated by + +<|ref|>title<|/ref|><|det|>[[115, 135, 386, 155]]<|/det|> +# the effectors TarP and TmeA + +<|ref|>text<|/ref|><|det|>[[115, 180, 122, 193]]<|/det|> +3 + +<|ref|>text<|/ref|><|det|>[[115, 211, 417, 228]]<|/det|> +Matthew D. Romero and Rey A. Carabeo1 + +<|ref|>text<|/ref|><|det|>[[115, 250, 122, 263]]<|/det|> +5 + +<|ref|>text<|/ref|><|det|>[[115, 279, 883, 298]]<|/det|> +Department of Pathology and Microbiology, College of Medicine, University of Nebraska Medical Center, + +<|ref|>text<|/ref|><|det|>[[115, 313, 201, 329]]<|/det|> +Omaha, NE + +<|ref|>text<|/ref|><|det|>[[115, 348, 122, 361]]<|/det|> +8 + +<|ref|>text<|/ref|><|det|>[[115, 380, 122, 394]]<|/det|> +9 + +<|ref|>text<|/ref|><|det|>[[115, 413, 291, 430]]<|/det|> +1Corresponding Author: + +<|ref|>text<|/ref|><|det|>[[115, 448, 435, 465]]<|/det|> +Department of Pathology and Microbiology + +<|ref|>text<|/ref|><|det|>[[115, 481, 397, 498]]<|/det|> +University of Nebraska Medical Center + +<|ref|>text<|/ref|><|det|>[[115, 515, 448, 532]]<|/det|> +985900 Nebraska Medical Center, Omaha, NE + +<|ref|>text<|/ref|><|det|>[[115, 550, 342, 566]]<|/det|> +Email: rey.carabeo@unmc.edu + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[66, 91, 181, 107]]<|/det|> +## 16 Abstract + +<|ref|>text<|/ref|><|det|>[[111, 133, 886, 494]]<|/det|> +16 AbstractChlamydia invasion of epithelial cells is a pathogen- driven process involving two functionally distinct effectors – TarP and TmeA. They collaborate to promote robust actin dynamics at sites of entry. Here, we extend studies on the molecular mechanism of invasion by implicating the host GTPase dynamin 2 (Dyn2) in the completion of pathogen uptake. Importantly, Dyn2 function is modulated by TarP and TmeA at the levels of recruitment and activation through oligomerization, respectively. TarP- dependent recruitment requires phosphatidylinositol 3- kinase and the small GTPase Rac1, while TmeA has a post- recruitment role related to Dyn2 oligomerization. This is based on the rescue of invasion duration and efficiency in the absence of TmeA by the Dyn2 oligomer- stabilizing small molecule activator Ryngo 1- 23. Notably, Dyn2 also regulated turnover of TarP- and TmeA- associated actin networks, with disrupted Dyn2 function resulting in aberrant turnover dynamics, thus establishing the interdependent functional relationship between Dyn2 and the effectors TarP and TmeA. + +<|ref|>sub_title<|/ref|><|det|>[[115, 519, 211, 535]]<|/det|> +## 28 Introduction + +<|ref|>text<|/ref|><|det|>[[111, 560, 886, 885]]<|/det|> +28 IntroductionChlamydia trachomatis is an obligate intracellular bacterium which infects ocular and genital epithelial cells, causing pelvic inflammatory disease, tubal factor infertility, ectopic pregnancy, and preventable blindness1. Chlamydia features a biphasic developmental cycle divided between metabolically quiescent elementary bodies (EBs) which invade host cells and vegetative reticulate bodies (RBs) which replicate inside membrane vacuoles termed inclusions2. Given its obligate intracellular nature, entry into host cells is essential for pathogen survival; consequently, Chlamydia possesses a robust suite of resources that regulate its uptake. Invasion also underpins pathogenicity, as it promotes access to the intracellular niche where it hijacks several host cell processes. Initial interaction with host epithelial cells is mediated by a reversible electrostatic interaction between a Chlamydia adhesin and host heparin sulfate proteoglycans3. Subsequently, Chlamydia engages multiple host receptors and delivers a variety of protein effectors via a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 346]]<|/det|> +type III secretion system4-6. Signaling from the effectors TarP and TmeA establishes a robust actin modulatory network that induces the assembly of actin- rich structures that engulf invading bacteria7-9. The resultant actin recruitment is characteristically highly localized to invading EBs and exhibits rapid kinetics of actin recruitment and turnover, such that actin network assembly and disassembly occurs within 200 seconds7,10. The majority of studies regarding chlamydial invasion focus on the mechanism of actin recruitment, while the process of disassembly at the end of invasion remains understudied, despite evidence pointing to its importance to elementary body uptake. We recently reported that altering the dynamics of actin turnover correlated with decreased invasion efficiency7. + +<|ref|>text<|/ref|><|det|>[[111, 370, 886, 900]]<|/det|> +Although multiple uptake mechanisms have been implicated as potential pathways for C. trachomatis invasion11- 13, the role of host dynamins during this process has been controversial. Dynamins are large GTPases that form oligomeric structures in a helical configuration around membrane lipids during clathrin- and caveolin- mediated endocytosis, mediating scission of vesiculated cargoes following GTP hydrolysis14. They are comprised of a catalytic G domain, a lipid- binding pleckstrin homology (PH) domain, and a proline- rich domain (PRD) that interacts with Src homology 3 (SH3) domain- containing proteins15. Absent activation, dynamins possess low intrinsic GTPase activity and assemble into dimers or tetramers16. These are utilized to generate higher- order oligomers such as half- rings, rings, and helices, the latter forming at the collar of invaginating vesicles17. GTP hydrolysis induces a conformational change along the oligomer that promotes constriction followed by vesicle scission, prompting rapid turnover of the dynamin superstructure18. Dynamin oligomerization is promoted by several effectors, including SH3 domain- containing proteins19, actin filaments20, and membrane lipids21. Many known effectors of dynamin oligomerization are present at C. trachomatis invasion sites, raising the possibility that that dynamin- dependent scission is utilized during terminal stages of this process. Several host proteins present during invasion are also directly or indirectly targeted by chlamydial effectors4,22- 24, highlighting the level of control the pathogen exerts on the invasion process. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 450]]<|/det|> +RNA interference of dynamin 2 (Dyn2) restricted C. trachomatis uptake12, bolstering support for a dynamin- dependent uptake mechanism. In contrast, pretreatment with the dynamin inhibitor MiTMAB, which targets the PH domain of dynamin, did not alter C. trachomatis invasion efficiency11. However, this study also identified that SNX9, a BAR domain protein which promotes dynamin oligomerization, is recruited during invasion, and that its depletion attenuated Chlamydia entry. Furthermore, overexpression of dominant negative GTPase- inactive Dyn1 K44A did not prevent C. trachomatis infection of HeLa cells25. Notably, this study did not investigate C. trachomatis uptake frequency and did not target Dyn2, the predominant dynamin species expressed in epithelial cells. In this study, we aim to reconcile the controversial involvement of host dynamins during C. trachomatis entry, monitoring its involvement using a series of high- resolution tools previously employed to characterize the regulation of actin remodeling during invasion7. + +<|ref|>text<|/ref|><|det|>[[112, 472, 886, 830]]<|/det|> +Given that dynamin interacts both with actin itself and with several proteins that regulate actin polymerization20,26- 29, it has become increasingly apparent that the dynamin GTPase cycle and actin polymerization are co- regulated. On this basis, the secreted effectors TarP and TmeA, which are themselves regulators of actin dynamics, likely also regulate host Dyn2 during invasion. Once secreted, TmeA associates with the plasma membrane and activates N- WASP, followed by Arp2/3 complex activation and nucleation of actin polymerization9,30. Likewise, TarP signaling activates host signaling proteins such as Rac1, PI3K, and the WAVE2 complex, in addition to recruiting the actin effectors formin and Arp2/37,31. Many host proteins associated with TarP and TmeA signaling are known to regulate Dyn2 oligomerization, such as cortactin32, EPS8333, profilin32,34 and the Arp2/3 complex35. Thus, in addition to the previously established role of TarP and TmeA signaling as synergistic effectors of rapid actin kinetics7,8, it is likely that they have a role in Dyn2 localization dynamics during Chlamydia entry. + +<|ref|>text<|/ref|><|det|>[[112, 855, 886, 910]]<|/det|> +Here, we demonstrate that Dyn2 is co- recruited alongside actin during Chlamydia invasion and coordinates efficient engulfment of the pathogen. This phenomenon is contingent upon signaling from + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 450]]<|/det|> +both TarP and TmeA, such that TarP signaling is necessary for local recruitment of Dyn2, whereas TmeA signaling activates Dyn2 by promoting oligomerization. The application of the Dyn2 activator Ryngo 1- 23, which promotes oligomerization and stabilizes Dyn2 polymers rescues invasion defects associated with TmeA deletion, enhancing its entry efficiency, and restoring kinetics of Dyn2 and actin recruitment and turnover to near wild- type levels. Further, we discovered that actin disassembly is dependent on Dyn2 function, thus ensuring the completion of invasion. Altogether, these findings resolve the long- standing controversy within the field, providing a novel regulatory function which accounts for both rapid assembly and disassembly of Chlamydia engulfment machinery in addition to a comprehensive model for the utilization and regulation of host Dyn2 during C. trachomatis invasion. They also highlight cooperation between TarP and TmeA and illustrate the broader impact of establishing their respective actin networks beyond the formation of engulfment structures. + +<|ref|>sub_title<|/ref|><|det|>[[115, 517, 172, 533]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 560, 574, 579]]<|/det|> +## Dynamin 2 and actin are co-recruited during Chlamydia entry + +<|ref|>text<|/ref|><|det|>[[111, 601, 886, 896]]<|/det|> +Since conflicting reports persist regarding host dynamin- 2 (Dyn2) involvement during C. trachomatis invasion, we revisited the question and evaluated its recruitment in greater detail using quantitative imaging approaches. We first determined whether Dyn2 was present within entry sites by co- transfecting Cos7 cells with GFP- Dyn2 and iRFP670- LifeAct prior to infection with wild- type C. trachomatis (MOI=20) stained with the red fluorescent dye CMTPX. Using live- cell confocal microscopy, we monitored Dyn2 and actin recruitment during entry, acquiring images at 20 second intervals (Fig. 1A). As previously reported7, we observed rapid actin recruitment, which was concomitant with arrival of Dyn2 and resulted in rapid uptake of Chlamydia, characterized by loss of CMTPX- CTL2 signal within 200- 300 sec. In contrast, expression of mutant Dyn2 K44A (Dyn2 DN), which is defective in GTPase binding and hydrolysis and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 725]]<|/det|> +cannot mediate vesicle scission (Fig. 1B), prolonged internalization to 400- 700 sec. Delayed pathogen uptake following Dyn2 DN expression could arise from several potential sources, such as inefficient Dyn2 recruitment, impaired actin dynamics, or disruptions within the Dyn2 GTPase cycle that prevent vesicle scission. To address each of these possibilities, we employed a previously established protocol for quantitatively assessing host protein recruitment dynamics during Chlamydia invasion, starting by characterizing Dyn2 WT and Dyn2 DN recruitment dynamics (Fig. 1C). While both Dyn2 WT and Dyn2 DN were recruited during entry, we noted that Dyn2 DN achieved peak mean fluorescence intensity (MFI) roughly 80 seconds later than Dyn2 WT and persisted within entry sites for a longer duration, indicating that rapid recruitment of Dyn2 and subsequent rapid entry of Chlamydia is contingent upon Dyn2 GTPase activity. To further substantiate this claim, we converted time- lapse images of actin, Dyn2, and Chlamydia into kymographs, upon which we indicated the start (i.e. initiation of actin/Dyn2 recruitment) and end (i.e. loss of EB fluorescence) of invasion (Fig. 1D, S1). The duration between initiation of actin/Dyn2 recruitment and pathogen entry was prolonged by expression of Dyn2 DN (Fig. 1D,E), such that Chlamydia uptake in cells expressing Dyn2 WT occurred within 180 sec, which was delayed by over two- fold (380 sec) when Dyn2 DN was expressed. Moreover, slow pathogen uptake following Dyn2 DN expression coincided with slower Dyn2 recruitment and turnover (Fig. 1F,G), reducing the rate of Dyn2 recruitment by 40 percent and turnover by 60 percent compared to Dyn2 WT. Altogether, these data indicate that Dyn2 is co- recruited alongside actin during Chlamydia entry, and that Dyn2 GTPase activity is necessary for efficient recruitment dynamics and rapid pathogen entry. + +<|ref|>sub_title<|/ref|><|det|>[[115, 745, 608, 764]]<|/det|> +## Dynamin 2 inhibition restricts Chlamydia entry and actin turnover + +<|ref|>text<|/ref|><|det|>[[111, 787, 886, 910]]<|/det|> +The recruitment of Dyn2 alongside its role in facilitating rapid pathogen entry suggests that dynamin- dependent uptake is an important component of Chlamydia invasion. Previous reports indicate that Dyn2 self- assembly and actin polymerization are co- regulated26,29,35, such that delayed Chlamydia entry following Dyn2 disruption may be due to defective actin polymerization. To test this, we disrupted Dyn2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 757]]<|/det|> +activity via pharmacological inhibition or by RNA interference prior to monitoring actin recruitment during Chlamydia invasion. Since co- overexpression of Dyn2 DN and actin may artificially influence actin dynamics, we instead inhibited endogenous Dyn2 using the dynamin inhibitor Dynasore, which mimics Dyn2 DN by restricting Dyn2 GTPase activity and subsequent scission (Fig. 2A). Furthermore, we were limited to \(\sim 50\%\) Dyn2 knockdown via RNA interference (Fig. 2B), as excessive Dyn2 depletion prevented cell adherence and cell proliferation, rendering these cells unsuitable for further analysis. Nonetheless, we noted that both \(25 \mu M\) Dynasore treatment and partial siRNA depletion of Dyn2 attenuated actin dynamics during CTL2 WT invasion (Fig 2B), resulting in prolonged actin retention within entry sites. Interestingly, actin recruitment kinetics were largely unchanged by Dyn2 disruption, yielding comparable rates across all conditions (Fig. 2C). In contrast, actin turnover was significantly attenuated by both Dynasore treatment and Dyn2 siRNA knockdown, with Dynasore treatment halving the actin turnover rate, while Dyn2 siRNA treatment slowed actin turnover by 25 percent (Fig. 2D). Given the importance of rapid actin turnover kinetics toward efficient invasion7, it is possible that Dyn2 inhibition (or absence) prolongs Chlamydia entry through defects in actin turnover. In support of this notion, we observed that both inhibition and depletion of Dyn2 delayed Chlamydia entry by roughly two- fold (Fig. 2E,F), comparable to the delay observed following Dyn2 DN overexpression (Fig. 1E), indicating that active Dyn2 is required for efficient actin turnover and rapid Chlamydia entry. Moreover, we observed a comparable attenuation in wild- type Chlamydia entry efficiency following Dyn2 inhibition (Fig. 2G) or siRNA depletion (Fig. 2H), reducing Chlamydia uptake by roughly 20 percent. Therefore, Dyn2 activity regulates actin turnover during invasion such that disruption of Dyn2 impedes actin depolymerization within entry sites. + +<|ref|>sub_title<|/ref|><|det|>[[115, 778, 690, 797]]<|/det|> +## Signaling from both TarP and TmeA is required for dynamin-dependent entry + +<|ref|>text<|/ref|><|det|>[[112, 821, 886, 921]]<|/det|> +Several studies have indicated that mutant Chlamydia strains harboring TarP and TmeA deletion or loss- of- function mutations exhibit substantially dysregulated pathogen entry7,36,37. As such, we monitored invasion of Chlamydia mutant strains lacking either TarP or TmeA (ΔTmeA, ΔTarP) or both (DKO) to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 314]]<|/det|> +determine if their respective routes of entry were affected by Dyn2 inhibition or depletion. Loss of either TarP or TmeA rendered their respective invasion processes resistant to Dyn2 inhibition (Fig. 2G), likely indicating the utilization of an alternative entry mechanism, i.e. fluid- phase uptake, which is dynamin- independent (Fig. S2). Entry efficiency of these strains were similarly insensitive to Dyn2 depletion via RNA interference (Fig. 2H), confirming that Dyn2 does not contribute to pathogen invasion following TarP or TmeA deletion. Finally, we noted that cis- complementation of the \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) and \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) mutants (cis- TmeA, cis- TarP) restored Dynasore sensitivity (Fig. 2G). + +<|ref|>text<|/ref|><|det|>[[111, 336, 886, 767]]<|/det|> +For the \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) mutant, Dyn2 dispensability was unsurprising given the spatial profiles of actin exhibited by this mutant, which assembles structures typically associated with fluid- phase uptake, such as large blooms and mini- ruffles38 (Fig. S2A,E). Indeed, \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) EBs frequently colocalized with the fluid- phase marker Dextran- Alexa Fluor 647; 40 percent of EBs were dextran positive within 20 minutes post- entry (Fig. S2F,G). In contrast, the \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) mutant retained punctate recruitment of actin characteristic of wild- type EBs (Video S2,3) and exhibited lower incidence of dextran colocalization (Fig. S2F). Thus, invasion of \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) EBs is mechanistically distinct from \(\Delta \mathsf{T}\mathsf{a}\mathsf{r}\mathsf{P}\) , adopting a spatial configuration that may benefit from Dyn2 activity. As such, the apparent insensitivity of \(\Delta \mathsf{T}\mathsf{m}\mathsf{e}\mathsf{A}\) EBs toward Dyn2 inhibition might reflect that Dyn2 is required for entry, but present in a non- functional state that rendered inhibition by Dynasore moot, which will be addressed in detail later in this study. Altogether, our data unequivocally reveal that dynamin- dependent uptake is an important component of C. trachomatis invasion which is contingent upon both TarP and TmeA signaling, wherein each effector likely regulates different invasion- associated aspects of Dyn2. + +<|ref|>sub_title<|/ref|><|det|>[[112, 792, 789, 811]]<|/det|> +## TarP and TmeA mediate recruitment and post-recruitment activation of Dyn2, respectively + +<|ref|>text<|/ref|><|det|>[[112, 820, 886, 901]]<|/det|> +TarP and Dyn2 function. Strikingly, TarP deletion prevented localized recruitment of Dyn2 at sites of pathogen entry (Fig. S2, Video S1), indicating that TarP signaling regulates early aspects of Dyn2 recruitment. We hypothesize that the actin network induced by TarP, rather than TarP itself, is responsible + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 520]]<|/det|> +for Dyn2 recruitment, given actin and Dyn2 colocalization (Fig. 1) and the reported functional relationship between these proteins20,29. To provide a mechanistic basis for TarP- dependent regulation of Dyn2, we investigated how Dyn2 localization dynamics are affected by ablation of PI3K/Rac1 signaling, which contributes to TarP- mediated actin remodeling31 (Fig. 3B). We monitored Dyn2 recruitment following administration of the Rac- specific inhibitor EHop- 016 (10μM) at entry sites of wild- type and ΔTmeA EBs, since both strains retain TarP signaling (Fig. 3). Rac inhibition did not affect the rate of Dyn2 recruitment (Fig. 3G) but substantially attenuated its turnover (Fig. 3H), resulting in prolonged retention of Dyn2 within CTL2 WT entry sites (Mock = 260 sec, EHop = 520 sec) (Fig. 3A). Thus, TarP- mediated actin remodeling not only coordinates local recruitment of Dyn2 within entry sites, but also regulates its retention via Rac1 signaling. Interestingly, Dyn2 localization dynamics of ΔTmeA mutants were unaffected by Rac inhibition (Fig. 3C), exhibiting similar recruitment and turnover rates between mock- and EHop- treated samples (Fig. 3G,H). TmeA- dependent sensitivity of Dyn2 localization dynamics toward Rac signaling hints at a significant role for this effector in Dyn2 function, which likely manifests at later (i.e., post- recruitment) stages. + +<|ref|>text<|/ref|><|det|>[[111, 541, 886, 901]]<|/det|> +We next tested the role of PI3K/Vav2 signaling, which is one of the Rac- activating pathways linked to TarP, the other being Abi1/Eps8/Sos1 signaling31 (Fig. 4D). To determine the functional outcome of PI3K signaling toward Dyn2 regulation, we monitored the invasion of wild- type and ΔTmeA EBs in the presence of the PI3K inhibitor Wortmannin (100 nM). Pretreatment with Wortmannin yielded intense and long- lasting Dyn2 localization relative to mock at wild- type entry sites (Fig. 4A,B) and attenuated the rate of Dyn2 turnover (Fig. 4H), consistent with PI3K signaling through Rac (Fig. 3C,H). Interestingly, PI3K inhibition did not alter Dyn2 recruitment during ΔTmeA invasion (Fig. 4A,B), indicating that in absence of TmeA, Dyn2 is not in its proper context to be affected further by wortmannin treatment. Moreover, wortmannin pretreatment did not alter the invasion efficiency of any strain tested (Fig. 4C) yet induced a significant delay in CTL2 WT uptake (Mock = 180 sec, Wort = 320 sec) (Fig. 4E,F). This disparity may arise due to the enhanced sensitivity of our kymograph- based internalization assay (Fig. 4E,F), which employs quantitative + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 280]]<|/det|> +fluorescence- based live- cell imaging to identify invasion defects. The former internalization assay (Fig. 4C) relies on antibody accessibility to measure invasion efficiency, a low- resolution approach with elevated likelihood of missing regulatory interactions between host and pathogen. In summary, our data indicate that TarP signaling is essential for dynamin- dependent entry of Chlamydia and is required for local recruitment of Dyn2 within entry sites, while also regulating its retention as a consequence of the actin network generated via the PI3K/Rac1 signaling axis. + +<|ref|>text<|/ref|><|det|>[[110, 300, 888, 904]]<|/det|> +TmeA and Dyn2 function. Although \(\Delta\) TmeA EBs recruit Dyn2 in a highly localized and punctate manner, similar to CTL2 WT (Fig 3A, 4A), inhibition of function via ectopic expression of dominant negative Dyn2 or \(25 \mu M\) Dynasore treatment did not alter uptake duration or Dyn2 dynamics associated with \(\Delta\) TmeA (Fig. S3). The apparent insensitivity toward Dyn2 disruption following TmeA deletion may reflect a lack of Dyn2 involvement during \(\Delta\) TmeA entry, or that TmeA deletion induces Dyn2 loss of function. To distinguish between these two possibilities, we employed the Dyn2 activator Ryngo 1- 23, a small molecule compound that stimulates Dyn2 oligomerization in a manner comparable to short actin filaments39. As such, we quantified Chlamydia entry after 30 minute preincubation with \(40 \mu M\) Ryngo 1- 23, wherein \(\Delta\) TmeA invasion efficiency was improved to near wild- type levels (Mock CTL2 WT = 79.8%, Ryngo \(\Delta\) TmeA = 71.0%) (Fig. 5A). Moreover, this compound restored normal Dyn2 recruitment dynamics during \(\Delta\) TmeA entry (Fig. 5B), generating a Dyn2 recruitment profile comparable to mock- treated CTL2 WT (Fig. 5D,G,H). Likewise, both mock CTL2 WT and Ryngo \(\Delta\) TmeA were internalized within 180 seconds on average, which was prolonged to 240 seconds for \(\Delta\) TmeA in absence of Ryngo (Fig. 5E,F), and that compound- assisted entry reduced the incidence of fluid- phase uptake (Fig. S2F). Taken together, these data suggest that Dyn2 oligomerization is defective when TmeA signaling is absent, and that Ryngo bypasses the requirement for TmeA signaling, enabling dynamin- dependent entry of \(\Delta\) TmeA EBs. In contrast, invasion efficiency, Dyn2 localization dynamics, and duration of internalization associated with wild type CTL2 were all negatively affected by Ryngo (Fig. 5A- F). A possible explanation may be that joint activation of Dyn2 by both TmeA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 585]]<|/det|> +signaling and Ryngo administration results in Dyn2 hyperactivation that prevents normal completion of the Dyn2 GTPase cycle. Indeed, Gu et. al found that Ryngo 1- 23 abrogated Dyn1 helical collar assembly, instead promoting stacked ring assembly (Fig. 5C), exhibiting reduced GTPase activity and attenuated vesicle scission compared to helices39. Additionally, CTL2 WT entry was comparably attenuated by either Dynasore- mediated inhibition of Dyn2 (Fig. 2G) or Ryngo- mediated Dyn2 activation (Fig. 5A), implying that dynamin- dependent entry of Chlamydia is sensitive to both hypo- and hyperactivation of Dyn2. Finally, whereas Ryngo administration prior to infection with \(\Delta\) TarP EBs restored localized recruitment of Dyn2 (Figs. S2C, S4B, Video S1), its recruitment was vastly dysregulated relative to wild- type (Fig. S4C,F,G) and failed to elicit rapid internalization of the pathogen (Fig. S4E). Together, this implies that Dyn2 is not organized in a proper context within entry sites when TarP is absent despite restoration of recruitment by Ryngo. In contrast, Dyn2 dynamics and function were restored by Ryngo treatment in \(\Delta\) TmeA EB invasion because Dyn2 proteins were in a context that favors oligomerization. In summary, these data indicate that TmeA signaling activates Dyn2, promoting its oligomerization in support of rapid dynamin- dependent entry of Chlamydia. In addition, the ordered roles of TarP and TmeA regarding Dyn2 function highlights the previously reported collaboration between these two effectors. + +<|ref|>sub_title<|/ref|><|det|>[[115, 608, 741, 627]]<|/det|> +## Actin turnover is correlated with Dynamin 2 activation status and Chlamydia uptake + +<|ref|>text<|/ref|><|det|>[[110, 650, 886, 912]]<|/det|> +Previous studies have identified that TmeA deletion dysregulates the actin network generated by Chlamydia during invasion, causing poor actin retention and abnormally fast actin turnover7- 9. Moreover, in this study, we have noted a functional link between Dyn2 activity and actin turnover, wherein actin recruitment was abnormally persistent upon pharmacological inhibition of Dyn2 or upon expression of Dyn2 K44A (Figs. 2B, S1F), resulting in delayed pathogen uptake. In light of these observations, we opted to evaluate the influence of the dynamin activator Ryngo 1- 23 on actin kinetics to determine whether compound- mediated restoration of Dyn2 activity within \(\Delta\) TmeA entry sites also restores normal actin dynamics. While administration of Ryngo prior to infection strongly increased the persistence of actin + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 551]]<|/det|> +recruitment at entry sites of both wild- type and \(\Delta \mathsf{TmeA}\) EBs (Fig. 6A), relative to the respective mock- treated controls, the slowed turnover associated with the \(\Delta \mathsf{TmeA}\) mutant was indistinguishable from mock- treated wild type control (Fig. 6A- C, Video S2, S3). As expected, we observed that Ryngo treatment restored the duration of internalization of \(\Delta \mathsf{TmeA}\) mutants to levels of mock- treated CTL2 WT (Fig. 6E,F). However, when invasion signaling was intact, i.e. when TarP and TmeA are both present, the additional Dyn2 activation by Ryngo had a negative effect on actin turnover and pathogen uptake (Fig. 6A,F, Video S3). This paralleled the effects of Ryngo on Dyn2 recruitment (Fig. 5), underscoring a possible relationship between actin disassembly and Dyn2 turnover (Fig. 6D). Indeed, either insufficient Dyn2 activity (i.e., Dynasore treatment, Dyn2 DN, Mock/ \(\Delta \mathsf{TmeA}\) ; Fig. 2B- D) or Dyn2 hyperactivation (i.e., Ryngo/CTL2 WT; Fig. 6A- F) results in similar dysregulated actin turnover and delayed pathogen uptake. Collectively, our data is consistent with a model whereby actin remodeling by TarP and TmeA, in addition to forming engulfment structures, also ensures Dyn2 recruitment and activation. With Dyn2 regulating actin turnover, this self- contained invasion mechanism ensures that disassembly of the invasion structures is properly coordinated with a successful scission event indicated by Dyn2 turnover. + +<|ref|>sub_title<|/ref|><|det|>[[115, 575, 197, 591]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 617, 886, 909]]<|/det|> +In this study, we conclusively demonstrated that C. trachomatis utilizes host Dyn2 to complete invasion. Dyn2 function is modulated by the effectors TarP and TmeA, which respectively mediate recruitment to invasion sites and activation by promoting oligomerization. Neither TarP nor TmeA possesses domains that mediate direct interaction with Dyn2 to facilitate recruitment and oligomerization; instead, TarP and TmeA modulate Dyn2 via their respective actin networks. Interestingly, Dyn2 influences actin turnover, wherein perturbation of Dyn2 function induces persistent actin retention. This functional interdependence constitutes a self- regulating system, such that Dyn2 function and pathogen engulfment are regulated by the actin network assembled via TarP and TmeA signaling. Reciprocally, Dyn2 function and subsequent membrane fission promotes actin disassembly and mediates resolution of engulfment structures. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 483]]<|/det|> +Moreover, TarP and TmeA signaling are sequentially coordinated such that the essential steps of invasion are initiated and completed. Specifically, we found that TarP signaling via PI3K/Rac1 coordinated initial recruitment and retention of Dyn2 within entry sites. Once recruited, Dyn2 is activated by TmeA signaling on the basis that defects associated with TmeA deletion were rescued by administration of the small molecular activator Ryngo 1- 23, which promotes Dyn2 oligomerization. Moreover, these data are consistent with previous observations suggesting that TmeA regulates latter stages of invasion. Finally, our study provides several high- resolution methods for tracking pathogen uptake, enabling detailed analysis of host- pathogen interactions underpinning Chlamydia entry, exceeding the limitations of previously employed techniques. In summary, we report that Dyn2 activation is an important component of Chlamydia invasion, which is regulated synergistically by TarP and TmeA to mediate scission of Chlamydia- containing vesicles and initiate turnover of host proteins following invasion. Altogether, findings underscore the high degree of control Chlamydia has over its invasion process. + +<|ref|>text<|/ref|><|det|>[[111, 506, 886, 899]]<|/det|> +TarP- deficient strains were incapable of localized and punctate Dyn2 recruitment, indicating that TarP signaling is required to prompt Dyn2 recruitment into a scission- competent configuration. Given that Dyn2 directly interacts with several TarP- associated actin regulators, including cortactin32, EPS833, and profilin32,34, we propose that the actin network generated by TarP signaling regulates Dyn2 function. Whether this interaction is mediated by direct interaction with actin, which has been reported previously20, or by various signaling molecules recruited by TarP is not known. One possibility is that TarP- mediated actin remodeling induces changes to the local environment that enrich and retain Dyn2 at sufficient quantities to achieve functionality. For example, robust actin polymerization can promote membrane curvature to support binding of Bin/amphiphysin/Rvs (BAR) domain proteins, some of which (e.g., SNX9) are known Dyn2 interactors40. This would also account for temporal regulation of Dyn2, wherein the timing of host protein recruitment influences both the concentration and orientation of Dyn2. Although our study demonstrates that Dyn2 and actin dynamics are functionally linked, a comprehensive + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 140]]<|/det|> +model of Dyn2 involvement during invasion will require further characterization of its recruitment and activation. + +<|ref|>text<|/ref|><|det|>[[112, 165, 886, 696]]<|/det|> +Recently, we reported that TarP signaling uniquely recruited host forms, which utilize profilin/actin complexes to acquire monomeric actin, and are important regulators of actin polymerization during Chlamydia entry. Moreover, the Arp2/3 complex is extensively associated with Dyn2 activity and collaborates with host forms to enhance actin remodeling during invasion. Robust actin remodeling provides a mechanism to ensure Dyn2 recruitment at sufficient levels; consequently, the pathways employed by Chlamydia to mediate actin nucleator activation are highly relevant points of Dyn2 regulation. For instance, we observed that TarP signaling via the PI3K/Rac1 axis, which regulates actin polymerization during invasion, also governed Dyn2 retention within entry sites. There is also precedence for Dyn2 modulation of actin remodeling, specifically insofar as disruption of Dyn2 dysregulates Rac localization and impairs actin dynamics within lamellipodia, highlighting that regulation of Dyn2 and Rac1 are functionally linked. Furthermore, actin stability and Dyn2 oligomerization are co-regulated, such that inhibition of Arp2/3 was sufficient to shift the balance of actin dynamics toward net disassembly, preventing scission of phagocytized particles and increasing Dyn2 persistence. Thus, destabilization of invasion-associated actin networks following Rac inhibition likely interferes with Dyn2 scission and subsequent turnover, yielding abnormally persistent signal. Conversely, Rac activation would promote Dyn2 function, a role demonstrably fulfilled by TarP. + +<|ref|>text<|/ref|><|det|>[[112, 720, 886, 910]]<|/det|> +We also found that TmeA signaling promoted Dyn2 activation, wherein strains lacking TmeA exhibited defective uptake that could be rescued by Ryngo 1- 23 administration. Several lines of evidence suggest that TmeA regulates Dyn2 via its previously established role in actin remodeling. In- vitro assays identified that short actin filaments stimulate Dyn2 ring assembly, and that Ryngo 1- 23 promotes Dyn2 ring formation via a comparable mechanism. Thus, one possibility is that membrane localized TmeA generates actin filaments which scaffold the initial activation of Dyn2 at the plasma membrane. Signaling + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 886, 620]]<|/det|> +via the TmeA/N- WASP axis drives Arp2/3 activation, which synergizes with TarP signaling to promote actin polymerization and pathogen engulfment \(^{30,47}\) . Collaboration between TarP and TmeA may additionally regulate Dyn2, wherein TmeA- mediated actin polymerization functions after Dyn2 recruitment to promote oligomerization. Indeed, both TarP and TmeA were necessary for dynamin- dependent entry, as strains lacking either effector were insensitive to Dyn2 disruption, although the basis for their insensitivity differed. How might the regulatory contributions of TarP and TmeA be distinguished, given the shared importance of their respective actin remodeling functions? For TmeA, the involvement of N- WASP might offer some clues. This nucleation promoting factor harbors a proline- rich domain (PRD) that binds proteins with Src- homology 3 (SH3) domains. The SH3 domain- containing protein SNX9 interacts with dynamin and stimulates Dyn2 oligomerization \(^{40}\) and is important for C. trachomatis invasion \(^{51}\) . As such, interaction between N- WASP and SNX9 might account for Dyn2 dependence toward TmeA signaling. Intriguingly, TmeA also bears similarity with the C. pneumoniae secreted effector SemD \(^{52,53}\) , which recruits the BAR- domain proteins PACSIN and SNX9 to induce membrane curvature and promote pathogen engulfment. On this basis, TmeA- mediated Dyn2 regulation could manifest via the formation of SNX9/Dyn2 heterodimers, providing a mechanism of Dyn2 modulation distinct from its actin remodeling function. Therefore, there are at least two molecular interactions that uniquely link TmeA signaling with Dyn2 function. + +<|ref|>text<|/ref|><|det|>[[112, 643, 886, 899]]<|/det|> +While the precise nature of how TmeA signaling modulates the Dyn2 GTPase cycle remains unknown, analysis of Dyn2 mutants may provide insight toward TmeA/Dyn2 regulation, and perhaps the mechanism of Ryngo- mediated rescue. Studies regarding the formation of progressive higher- order dynamin oligomers have benefited from various mutations that affect protein- protein interactions, GTPase activity, conformational changes during constriction, etc. Determining the exact mechanism of compound- mediated rescue following TmeA deletion will require elucidating which oligomeric species of Dyn2 is induced by either Ryngo or TmeA signaling. Mutations which prevent dynamin self- assembly (i.e. Dyn1 I670K \(^{54}\) ) or those which ablate membrane association (i.e. Dyn2 K562E \(^{55}\) ) could be informative toward this + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 348]]<|/det|> +end, as these mutants are membrane scission- deficient and are not rescued by Ryngo50,56. Our working model predicts that these mutants should likewise be unaffected by TmeA signaling. Dyn1 K/E exhibits reduced affinity for actin filaments and is partially rescued by Ryngo in- vitro50, whereas Dyn2ΔPRD cannot bind SH3 domain- containing proteins and is dominant- negative for endocytosis57. Studies incorporating these mutants could disambiguate whether TmeA signaling operates by mediating Dyn2/actin interactions, or by promoting interaction with SH3 domain- containing proteins like SNX9. Using this report as a foundation, future studies could interrogate the effects of each Dyn2 mutant during Chlamydia invasion and determine the precise nature of effector signaling toward dynamin- dependent entry. + +<|ref|>text<|/ref|><|det|>[[111, 370, 886, 797]]<|/det|> +Interestingly, unlike ΔTmeA, Ryngo treatment impaired wild- type Chlamydia invasion, restricting pathogen entry and yielding obvious defects in Dyn2 and actin recruitment. One explanation may be that in certain contexts, Ryngo stimulates Dyn2 oligomerization into a scission- incompetent configuration. FRET analysis of dynamin oligomerization found that Ryngo prompted the assembly of stacked Dyn2 rings around membrane tubules50, representing a lower- order oligomerization state that achieved insufficient GTPase activity to induce membrane scission. As such, co- stimulation of Dyn2 activation by both Ryngo and Chlamydia/TmeA signaling may interfere with the relative abundance of Dyn2 oligomeric species. Specifically, stimulation with Ryngo is expected to generate a disproportionate quantity of Dyn2 rings which interfere with further oligomerization steps. Elimination of Chlamydia- specific Dyn2 activation (i.e., ΔTmeA) may prevent overstimulation, encouraging proper assembly of higher- order, scission- competent Dyn2 oligomers. Meanwhile, whereas Ryngo pretreatment restored local Dyn2 recruitment at ΔTarP entry sites, it failed to prompt rapid engulfment of ΔTarP EBs and had no rescuing effect on its entry efficiency, suggesting that post- recruitment, Dyn2 needs to be primed for activation by Ryngo. + +<|ref|>text<|/ref|><|det|>[[112, 821, 886, 921]]<|/det|> +Finally, our study identified that both Dyn2 and actin turnover were co- regulated. Mechanistically, Dyn2 turnover is intuitive, occurring either during or shortly after membrane scission as a function of GTP hydrolysis58,59. As such, Dyn2- mediated scission of Chlamydia- containing vacuoles may intrinsically prompt + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 450]]<|/det|> +Dyn2 turnover while also providing a signal to initiate actin turnover. Interventions which prevent dynamin- mediated membrane fission also accumulate F- actin around tubulated membranes \(^{60,61}\) , whereas scission is consistently associated with actin turnover and sensitizes actin filaments toward cofilin- mediated severing \(^{29,62,63}\) . Furthermore, given that dynamin extensively interacts with actin- associated proteins \(^{28,63 - 65}\) , post- scission turnover of actin regulatory machinery alongside Dyn2 may shift actin regulation toward turnover. Importantly, actin polymerization during Chlamydia invasion is both intricately regulated and pathogen- directed \(^{66}\) ; consequently, turnover of actin and other invasion- associated host proteins could be regulated distinctly from turnover associated with routine engulfment of cellular cargoes (i.e., growth factors, transferrin). This could require additional factors that fine- tune their function and/or dynamics to accommodate pathogen- mediated uptake mechanisms. As such, further study is required to gain a more comprehensive perspective on host protein turnover post- invasion. + +<|ref|>text<|/ref|><|det|>[[113, 472, 886, 628]]<|/det|> +Overall, our findings of Dyn2 modulation by TarP and TmeA fit well with the proposed pathogen- directed invasion model proposed by Byrne and Moulder \(^{67}\) . While the majority of molecular studies of chlamydial invasion focus on actin recruitment, we demonstrate here that latter stages are also targeted by TarP and TmeA, highlighting their central function in invasion, comprising a self- contained signaling module capable of mediating the initial, middle, and end stages of invasion. + +<|ref|>sub_title<|/ref|><|det|>[[115, 697, 297, 713]]<|/det|> +## Materials and Methods: + +<|ref|>sub_title<|/ref|><|det|>[[115, 730, 306, 747]]<|/det|> +## Cell and Bacterial Culture + +<|ref|>text<|/ref|><|det|>[[113, 763, 886, 885]]<|/det|> +Green monkey kidney fibroblast- like (Cos7) cells and cervical adenocarcinoma epithelial (HeLa) cells were cultured at \(37^{\circ}C\) with \(5\%\) atmospheric CO2 in Dulbecco's Modified Eagle Medium (DMEM; Gibco, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with \(10\mu \mathrm{g / mL}\) gentamicin, \(2\mathrm{mM}\) L- glutamine, and \(10\%\) (v/v) filter- sterilized fetal bovine serum (FBS). HeLa and Cos7 cells were cultured for a maximum of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 884, 210]]<|/det|> +15 passages for all experiments. McCoy B mouse fibroblasts (originally from Dr. Harlan Caldwell, NIH/NIAID) were cultured under comparable conditions. Chlamydia trachomatis serovar L2 (434/Bu) was propagated in McCoy cells and EBs were purified using a Gastrografin density gradient as described previously \(^{68}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 261, 187, 277]]<|/det|> +## Reagents + +<|ref|>text<|/ref|><|det|>[[111, 291, 886, 550]]<|/det|> +Wortmannin (Selleck, Houston, TX, USA) was diluted upon receipt to 40mM stock concentration in DMSO, Dynasore (Cayman Chemical, Ann Arbor, MI, USA) and EHop- 016 (Cayman) were diluted to 25mM stock concentration in DMSO, and Ryngo 1- 23 (Abcam, Cambridge, MA, USA) was diluted to 20mM stock concentration in DMSO. All inhibitors were dispensed into single use aliquots and stored at \(- 20^{\circ}C\) for no longer than 1 year after receipt. Wortmannin was diluted to a working concentration of 40nM (1:10000), Dynasore was diluted to a working concentration of 25 μM (1:1000), EHop- 016 was diluted to a working concentration of 10μM (1:2500), and Ryngo 1- 23 was diluted to a working concentration of 40μM (1:500), each using supplemented DMEM as diluent. + +<|ref|>sub_title<|/ref|><|det|>[[115, 599, 228, 616]]<|/det|> +## Invasion Assay + +<|ref|>text<|/ref|><|det|>[[111, 631, 886, 890]]<|/det|> +C. trachomatis internalization efficiency was conducted using HeLa cells and was performed as described previously \(^{10}\) . Briefly, HeLa cells were seeded in 24-well plates containing acid-etched glass coverslips and allowed to adhere overnight. Cells were pretreated with Wortmannin (40nM), Dynasore (25μM), EHop-016 (10μM), or Ryngo (40μM) for 30 minutes prior to infection. Dyn2 siRNA or scramble RNA were transfected and allowed to incubate 24 hours prior to infection. Following inhibitor treatment or RNA interference, cells were infected with EBs derived from wild-type C. trachomatis L2 (434/Bu), C. trachomatis in which TarP, TmeA, or both were deleted by FRAEM (ΔTarP, ΔTmeA, ΔTmeA/ΔTarP), or C. trachomatis in which TarP or TmeA expression was restored by cis-complementation (cis-TarP, cis-TmeA) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 585]]<|/det|> +at MOI=50. EBs were allowed to attach onto HeLa cells for 30 min at \(4^{\circ}C\) before rinsing coverslips with cold HBSS, followed by addition of supplemented DMEM prewarmed to \(37^{\circ}C\) , before incubating cells at \(37^{\circ}C\) for 10 min. After incubation, cells were stringently washed with cold HBSS containing \(100\mu g / mL\) heparin to remove any transiently adherent EBs before fixation in \(4\%\) paraformaldehyde at room temperature for 15 min. Fixed cells were labeled with a mouse monoclonal anti- MOMP antibody (Novus Biologicals, Centennial, CO, USA #NB10066403), rinsed with \(1x\) PBS, and fixed once more in \(4\%\) paraformaldehyde for 10 min. Next, cells were permeabilized using \(0.1\%\) (w/v) Triton X- 100 for 10 minutes at room temperature, rinsed with HBSS and labeled with rabbit polyclonal anti- Chlamydia trachomatis antibody (Abcam ab252762). Cells were then rinsed in \(1x\) PBS and labeled with Alexa Fluor 594 anti- mouse (ThermoFisher #A11032, Waltham, MA, USA) and Alexa Fluor 488 anti- rabbit (ThermoFisher #A11034) IgG secondary antibodies. Coverslips were mounted and observed on a Nikon CSU- W1 confocal microscope (Nikon, Melville, NY, USA), obtaining Z- stacks using a 0.3 micron step size across the height of the cell monolayer. Monolayer Z- stacks were transformed via Z- projection according to maximal fluorescence intensity in ImageJ prior to quantifying percent invasion efficiency as follows: total EBs (green) – extracellular EBs (red)/total EBs (green) x \(100\%\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 633, 503, 651]]<|/det|> +## Quantitative live cell imaging of Chlamydia invasion + +<|ref|>text<|/ref|><|det|>[[112, 664, 886, 890]]<|/det|> +Cos7 cells were seeded onto libdi \(\mu\) - Slide 8- well glass- bottomed chambers (Ibidii, Fitchburg, WI, USA) and allowed to adhere overnight prior to transfection. Cells were transfected with fluorescent proteins as indicated, using Lipofectamine 3000 (Thermo Fisher, Waltham, MA, USA) according to manufacturer directions. Transfection was allowed to proceed overnight before replacing media with fresh DMEM + \(10\%\) FBS/2 mM L- glutamine and allowing protein expression to continue for a total of 24 hours post- transfection. Transfection efficiency was verified on a Nikon CSU- W1 spinning disk confocal microscope prior to application of DMEM containing Wortmannin (40nM), Dynasore (25μM), EHop- 016 (10μM), or + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 516]]<|/det|> +Ryngo (40μM). For RNA interference, Dyn2 siRNA or scramble RNA was co- transfected alongside GFP- actin or mRuby- LifeAct and allowed to incubate for 24 hours prior to imaging. Wells were individually infected with CMTPX- labeled wild- type C. trachomatis L2 (434/Bu), unless otherwise indicated, at MOI=20 and promptly imaged using a 60x objective (NA 1.40) in a heated and humidified enclosure. Images were collected once every 20 seconds for 30 minutes, with focal plane maintained using an infrared auto- focusing system. Upon completion of the imaging protocol, the next well was infected and imaging repeated; mock- treated wells were imaged first to allow inhibitor treatment sufficient time to achieve inhibition. Images were compiled into videos using NIH ImageJ and analyzed to identify protein recruitment events. The mean fluorescence intensity (MFI) of recruitment events was measured for each timepoint alongside the local background MFI of a concentric region immediately outside the recruitment event. Background MFI was subtracted from recruitment MFI for each timepoint and normalized as percent maximal fluorescence intensity for each timepoint, repeating this normalization process for each recruitment event. + +<|ref|>sub_title<|/ref|><|det|>[[115, 565, 247, 581]]<|/det|> +## RNA interference + +<|ref|>text<|/ref|><|det|>[[112, 597, 886, 855]]<|/det|> +Cos- 7 or HeLa cells were seeded onto Ibidi \(\mu\) - Slide 8- well glass- bottomed chambers (live- cell imaging) or in 24- well plates containing acid- etched glass coverslips (invasion assay) and allowed to adhere overnight. Mission esiRNAs were custom- ordered to target Cos7 Dyn2 mRNA, ensuring that the resultant esiRNA targeted a shared sequence found in all recorded mRNA transcript variants. Cells were transfected with either 100 nM Mission anti- Dyn2 esiRNA (Eupheria Biotech, Dresden, Germany) or 100 nM Trilencer- 27 Universal scrambled negative control (Origene SR30004, Rockville, MD, USA) using Lipofectamine RNAiMAX reagent (Thermo Fisher) according to manufacturer directions. Incubation was allowed to proceed for 24 hours before conducting live- cell imaging or invasion assays using methods described + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 141]]<|/det|> +earlier. Lysates for Western blotting were obtained from Cos7 cells by applying 2x Laemmli buffer to cells after the completion of live- cell imaging. + +<|ref|>sub_title<|/ref|><|det|>[[115, 194, 245, 210]]<|/det|> +## Western Blotting + +<|ref|>text<|/ref|><|det|>[[111, 225, 886, 550]]<|/det|> +Lysates generated as described above were resolved via SDS- PAGE in \(10\%\) polyacrylamide gels at 120 volts for 1.5 hours or until the dye front has begun to evacuate the bottom of the gel cassette. Gels were transferred onto \(0.45\mu M\) pore size nitrocellulose in 1x Towbin buffer \(+10\%\) methanol at \(90~\mathrm{mA}\) for 16 hours. Western blots were blocked in \(5\%\) bovine serum albumin for 1 hour, briefly rinsed in Tris buffered saline \(+0.1\%\) Tween- 20 (TBST) and incubated with appropriate primary antibody for 1 hour. Blots were then washed three times for 5 minutes in TBST and probed with appropriate HRP- conjugated secondary antibodies for 1 hour. Protein bands were resolved by chemiluminescence using Immobilon Western HRP Substrate (Millipore Sigma, St Louis, MO, USA). Dynamin 2 knockdown efficiency was calculated by densitometry analysis, comparing the ratio of Dyn2 antibody signal (Thermo PA1- 661) against \(\beta\) - actin loading control (Abcam ab49900). + +<|ref|>sub_title<|/ref|><|det|>[[115, 599, 282, 616]]<|/det|> +## Dextran Uptake Assay + +<|ref|>text<|/ref|><|det|>[[111, 631, 886, 888]]<|/det|> +Cos7 cells were seeded onto 24- well plates containing acid- etched glass coverslips and allowed to adhere overnight. Cells were treated with \(40~\mu M\) Ryngo 1- 23 or DMSO in media containing \(100~\mu g / mL\) Dextran Alexa Fluor 647; 10,000 MW (Thermo D22914) and incubated at \(37^{\circ}C\) for 30 minutes. Cells were infected with wild- type or mutant Chlamydia strains at MOI=50, synchronizing infection by sedimentation at \(4^{\circ}C\) on a rocking incubator for 30 minutes. Infection was initiated by addition of prewarmed media, followed by incubation in a \(37^{\circ}C\) incubator for 20 minutes prior to fixation in \(4\%\) paraformaldehyde for 10 minutes at room temperature. Fluorescent dextran and inhibitor were maintained in media for each indicated stage. Cells were then permeabilized using \(0.1\%\) (w/v) Triton X- 100 for 10 minutes at room temperature, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 886, 313]]<|/det|> +rinsed with HBSS and labeled with a mouse monoclonal anti- MOMP antibody (Novus #NB10066403). Cells were rinsed in 1x PBS and labeled with Alexa Fluor 488 anti- mouse (ThermoFisher # A- 11001) IgG secondary antibody. Coverslips were mounted and observed on a Nikon CSU- W1 confocal microscope (Nikon), obtaining Z- stacks using a 0.3 micron step size across the height of the cell monolayer. Monolayer Z- stacks were transformed via Z- projection according to maximal fluorescence intensity in ImageJ prior to quantifying the percentage of elementary bodies which colocalize with fluorescent dextran, according to the following equation: [ Dextran\* EBs (magenta/green) / Total EBs (green) ] x 100%. + +<|ref|>sub_title<|/ref|><|det|>[[115, 361, 348, 378]]<|/det|> +## Plasmids and DNA preparation + +<|ref|>text<|/ref|><|det|>[[110, 393, 886, 858]]<|/det|> +pEGFP- Actin- C169 was a gift from Dr Scott Grieshaber (University of Idaho), and mRuby- LifeAct- 7 (Addgene plasmid #54560) was a gift from Michael Davidson. Dyn2- pmCherryN1 was a gift from Christien Merrifield (Addgene plasmid #27689), RFP Dynamin2 K44A was a gift from Jennifer Lippincott- Schwartz (Addgene plasmid #128153), WT Dyn2 pEGFP was a gift from Sandra Schmid (Addgene plasmid #34686), and GFP- Dynamin 2 K44A was a gift from Pietro De Camilli (Addgene plasmid #22301). Upon receipt, bacterial stab cultures were streak- plated onto LB agar containing appropriate antibiotic (Kanamycin, carbenicillin) for each plasmid. Resultant antibiotic- resistant colonies were selected and propagated in LB broth + antibiotic for plasmid isolation prior to sequence verification. All plasmids were isolated using MiniPrep DNA isolation kits (Qiagen, Valencia, CA, USA) following a variant protocol for DNA isolation termed MiraPrep70. Following plasmid isolation, the eluate was precipitated by addition of 3M sodium acetate (Invitrogen, Waltham, MA, USA) at 10% (v/v) of eluate volume followed by addition of 250% (v/v) absolute ethanol calculated after addition of sodium acetate. The mixture was incubated at 4°C overnight and centrifuged at 14,000×g for 15 minutes at 4°C. Supernatant was removed and 70% ethanol was added, followed by centrifugation at 14,000×g for 10 minutes at 4°C. Supernatant was removed once more, and precipitated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 90, 884, 143]]<|/det|> +DNA was resuspended in nuclease- free \(\mathsf{H}_2\mathsf{O}\) . Sequencing was conducted by Eurofins Genomics (Louisville, KY, USA), using standard sequencing primers provided by the company. + +<|ref|>sub_title<|/ref|><|det|>[[115, 194, 342, 210]]<|/det|> +## Graphs and statistical analysis + +<|ref|>text<|/ref|><|det|>[[112, 225, 886, 483]]<|/det|> +Violin plots were made using the ggplot2 base package (version 3.1.0) as a component of the Tidyverse package (https://cran.r- project.org/web/packages/tidyverse/index.html) in rStudio (version 4.0.3). Wilcoxon ranked- sum tests to determine statistical significance between violin plots were conducted using base R statistics in rStudio. Recruitment plots, invasion assays, and all statistics associated with these data (pairwise T- test followed by Bonferroni post- analysis, SEM) were performed in Excel (Microsoft, Redmond, WA, USA). All graphs were assembled using the free and open- source software GNU Image Manipulation Program (GIMP, https://www.gimp.org/) and Inkscape (https://inkscape.org/). Proposed model for Dyn2 oligomerization (Figs. 1- 6, S4) was assembled using BioRender (https://app.biorender.com/). + +<|ref|>sub_title<|/ref|><|det|>[[115, 527, 266, 543]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[112, 558, 864, 817]]<|/det|> +The authors would like to acknowledge the following investigators for their generosity - Dr. Ken Fields (University of Kentucky) for providing the mutant strains of C. trachomatis described in this study; pEGFP- Actin- C169 from Dr. Scott Grieshaber (University of Idaho), mRuby- LifeAct- 7 (Addgene plasmid #54560) from Dr. Michael Davidson; Dyn2- pmCherryN1 from Dr. Christien Merrifield (Addgene plasmid #27689); RFP Dynamin2 K44A from Dr. Jennifer Lippincott- Schwartz (Addgene plasmid #128153), WT Dyn2 pEGFP from Dr. Sandra Schmid (Addgene plasmid #34686), and GFP- Dynamin 2 K44A was from Dr. Pietro De Camilli (Addgene plasmid #22301). The authors would also like to thank members of the Carabeo laboratory for helpful suggestions. This work is funded by NIH AI065545 to R.C. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 225, 107]]<|/det|> +## Sources Cited: + +<|ref|>text<|/ref|><|det|>[[111, 123, 880, 170]]<|/det|> +1. Malhotra, M., Sood, S., Mukherjee, A., Muralidhar, S. & Bala, M. Genital Chlamydia trachomatis: An update. Indian J Med Res 138, 303-316 (2013). + +<|ref|>text<|/ref|><|det|>[[111, 189, 822, 240]]<|/det|> +2. Elwell, C., Mirrashidi, K. & Engel, J. Chlamydia cell biology and pathogenesis. Nature Reviews Microbiology 14, 385-400 (2016). + +<|ref|>text<|/ref|><|det|>[[111, 258, 848, 310]]<|/det|> +3. Stephens, R. S., Koshiyama, K., Lewis, E. & Kubo, A. Heparin-binding outer membrane protein of chlamydiae. Mol Microbiol 40, 691-699 (2001). + +<|ref|>text<|/ref|><|det|>[[111, 327, 884, 378]]<|/det|> +4. Gitsels, A., Sanders, N. & Vanrompay, D. Chlamydial Infection From Outside to Inside. Front Microbiol 10, (2019). + +<|ref|>text<|/ref|><|det|>[[111, 395, 882, 481]]<|/det|> +5. Chen, Y.-S. et al. The Chlamydia trachomatis type III secretion chaperone Slc1 engages multiple early effectors, including TepP, a tyrosine-phosphorylated protein required for the recruitment of CrkII-II to nascent inclusions and innate immune signaling. PLoS Pathog 10, e1003954 (2014). + +<|ref|>text<|/ref|><|det|>[[111, 497, 842, 581]]<|/det|> +6. Fields, K. A., Mead, D. J., Dooley, C. A. & Hackstadt, T. Chlamydia trachomatis type III secretion: evidence for a functional apparatus during early-cycle development. Mol Microbiol 48, 671-683 (2003). + +<|ref|>text<|/ref|><|det|>[[111, 598, 875, 650]]<|/det|> +7. Romero, M. D. & Carabeo, R. A. Distinct roles of the Chlamydia trachomatis effectors TarP and TmeA in the regulation of formin and Arp2/3 during entry. Journal of Cell Science 135, jcs260185 (2022). + +<|ref|>text<|/ref|><|det|>[[111, 666, 861, 753]]<|/det|> +8. Keb, G., Ferrell, J., Scanlon, K. R., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeA Directly Activates N-WASP To Promote Actin Polymerization and Functions Synergistically with TarP during Invasion. mBio 12, (2021). + +<|ref|>text<|/ref|><|det|>[[111, 769, 861, 853]]<|/det|> +9. Faris, R., McCullough, A., Andersen, S. E., Moninger, T. O. & Weber, M. M. The Chlamydia trachomatis secreted effector TmeA hijacks the N-WASP-ARP2/3 actin remodeling axis to facilitate cellular invasion. PLOS Pathogens 16, e1008878 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 884, 175]]<|/det|> +10. Carabeo, R. A., Grieshaber, S. S., Fischer, E. & Hackstadt, T. Chlamydia trachomatis Induces Remodeling of the Actin Cytoskeleton during Attachment and Entry into HeLa Cells. Infect Immun 70, 3793-3803 (2002). + +<|ref|>text<|/ref|><|det|>[[111, 191, 809, 244]]<|/det|> +11. Ford, C., Nans, A., Boucrot, E. & Hayward, R. D. Chlamydia exploits filopodia capture and a macropinocytosis-like pathway for host cell entry. PLOS Pathogens 14, e1007051 (2018). + +<|ref|>text<|/ref|><|det|>[[111, 260, 875, 312]]<|/det|> +12. Hybiske, K. & Stephens, R. S. Mechanisms of Chlamydia trachomatis Entry into Nonphagocytic Cells. Infection and Immunity 75, 3925-3934 (2007). + +<|ref|>text<|/ref|><|det|>[[111, 327, 848, 380]]<|/det|> +13. Nans, A., Saibil, H. R. & Hayward, R. D. Pathogen-host reorganization during Chlamydia invasion revealed by cryo-electron tomography. Cellular Microbiology 16, 1457-1472 (2014). + +<|ref|>text<|/ref|><|det|>[[111, 395, 881, 448]]<|/det|> +14. Ferguson, S. M. & De Camilli, P. Dynamin, a membrane remodelling GTPase. Nat Rev Mol Cell Biol 13, 75-88 (2012). + +<|ref|>text<|/ref|><|det|>[[111, 463, 870, 485]]<|/det|> +15. Ford, M. G. J., Jenni, S. & Nunnari, J. The crystal structure of dynamin. Nature 477, 561-566 (2011). + +<|ref|>text<|/ref|><|det|>[[111, 499, 866, 551]]<|/det|> +16. Muhlberg, A. B., Warnock, D. E. & Schmid, S. L. Domain structure and intramolecular regulation of dynamin GTPase. EMBO J 16, 6676-6683 (1997). + +<|ref|>text<|/ref|><|det|>[[111, 565, 875, 617]]<|/det|> +17. Warnock, D. E., Hinshaw, J. E. & Schmid, S. L. Dynamin self-assembly stimulates its GTPase activity. J Biol Chem 271, 22310-22314 (1996). + +<|ref|>text<|/ref|><|det|>[[111, 633, 880, 684]]<|/det|> +18. Antonny, B. et al. Membrane fission by dynamin: what we know and what we need to know. EMBO J 35, 2270-2284 (2016). + +<|ref|>text<|/ref|><|det|>[[111, 700, 835, 752]]<|/det|> +19. Ross, J. A. et al. Dimeric Endophilin A2 Stimulates Assembly and GTPase Activity of Dynamin 2. Biophys J 100, 729-737 (2011). + +<|ref|>text<|/ref|><|det|>[[111, 768, 857, 819]]<|/det|> +20. Gu, C. et al. Direct dynamin-actin interactions regulate the actin cytoskeleton. EMBO J 29, 3593-3606 (2010). + +<|ref|>text<|/ref|><|det|>[[111, 835, 864, 888]]<|/det|> +21. Praefcke, G. J. K. & McMahon, H. T. The dynamin superfamily: universal membrane tubulation and fission molecules? Nat Rev Mol Cell Biol 5, 133-147 (2004). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 877, 175]]<|/det|> +22. Carabeo, R. A., Dooley, C. A., Grieshaber, S. S. & Hackstadt, T. Rac interacts with Abi-1 and WAVE2 to promote an Arp2/3-dependent actin recruitment during chlamydial invasion. Cellular Microbiology 9, 2278–2288 (2007). + +<|ref|>text<|/ref|><|det|>[[110, 191, 850, 244]]<|/det|> +23. Caven, L. & Carabeo, R. A. Pathogenic Puppetry: Manipulation of the Host Actin Cytoskeleton by Chlamydia Trachomatis. (2019). + +<|ref|>text<|/ref|><|det|>[[110, 258, 881, 345]]<|/det|> +24. Thwaites, T. R., Pedrosa, A. T., Peacock, T. P. & Carabeo, R. A. Vinculin Interacts with the Chlamydia Effector TarP Via a Tripartite Vinculin Binding Domain to Mediate Actin Recruitment and Assembly at the Plasma Membrane. Front. Cell. Infect. Microbiol. 5, (2015). + +<|ref|>text<|/ref|><|det|>[[110, 360, 883, 447]]<|/det|> +25. Boleti, H., Benmerah, A., Ojcius, D. M., Cerf-Bensussan, N. & Dautry-Varsat, A. Chlamydia infection of epithelial cells expressing dynamin and Eps15 mutants: clathrin-independent entry into cells and dynamin-dependent productive growth. Journal of Cell Science 112, 1487–1496 (1999). + +<|ref|>text<|/ref|><|det|>[[110, 462, 872, 550]]<|/det|> +26. Yarar, D., Waterman-Storer, C. M. & Schmid, S. L. SNX9 Couples Actin Assembly to Phosphoinositide Signals and Is Required for Membrane Remodeling during Endocytosis. Developmental Cell 13, 43–56 (2007). + +<|ref|>text<|/ref|><|det|>[[110, 565, 850, 617]]<|/det|> +27. Schafer, D. A. et al. Dynamin2 and Cortactin Regulate Actin Assembly and Filament Organization. Curr Biol 12, 1852–1857 (2002). + +<|ref|>text<|/ref|><|det|>[[110, 632, 840, 684]]<|/det|> +28. Taylor, M. J., Perrais, D. & Merrifield, C. J. A high precision survey of the molecular dynamics of mammalian clathrin-mediated endocytosis. PLoS Biol 9, e1000604 (2011). + +<|ref|>text<|/ref|><|det|>[[110, 699, 872, 752]]<|/det|> +29. Taylor, M. J., Lampe, M. & Merrifield, C. J. A feedback loop between dynamin and actin recruitment during clathrin-mediated endocytosis. PLoS Biol 10, e1001302 (2012). + +<|ref|>text<|/ref|><|det|>[[110, 767, 860, 854]]<|/det|> +30. Keb, G., Ferrell, J., Scanlon, K. R., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeA Directly Activates N-WASP To Promote Actin Polymerization and Functions Synergistically with TarP during Invasion. mBio 12, e02861-20 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 857, 147]]<|/det|> +31. Lane, B. J., Mutchler, C., Khodor, S. A., Grieshaber, S. S. & Carabeo, R. A. Chlamydial Entry Involves TARP Binding of Guanine Nucleotide Exchange Factors. PLOS Pathogens 4, e1000014 (2008). + +<|ref|>text<|/ref|><|det|>[[110, 159, 844, 212]]<|/det|> +32. Orth, J. D. & McNiven, M. A. Dynamin at the actin-membrane interface. Current Opinion in Cell Biology 15, 31-39 (2003). + +<|ref|>text<|/ref|><|det|>[[110, 225, 860, 312]]<|/det|> +33. Auciello, G., Cunningham, D. L., Tatar, T., Heath, J. K. & Rappoport, J. Z. Regulation of fibroblast growth factor receptor signalling and trafficking by Src and Eps8. Journal of Cell Science 126, 613-624 (2013). + +<|ref|>text<|/ref|><|det|>[[110, 326, 850, 381]]<|/det|> +34. Witke, W. et al. In mouse brain profilin I and profilin II associate with regulators of the endocytic pathway and actin assembly. EMBO J 17, 967-976 (1998). + +<|ref|>text<|/ref|><|det|>[[110, 394, 875, 449]]<|/det|> +35. Krueger, E. W., Orth, J. D., Cao, H. & McNiven, M. A. A dynamin-cortactin-Arp2/3 complex mediates actin reorganization in growth factor-stimulated cells. Mol Biol Cell 14, 1085-1096 (2003). + +<|ref|>text<|/ref|><|det|>[[110, 462, 881, 550]]<|/det|> +36. McKuen, M. J., Mueller, K. E., Bae, Y. S. & Fields, K. A. Fluorescence-Reported Allelic Exchange Mutagenesis Reveals a Role for Chlamydia trachomatis TmeA in Invasion That Is Independent of Host AHNAK. Infect Immun 85, (2017). + +<|ref|>text<|/ref|><|det|>[[110, 563, 845, 650]]<|/det|> +37. Ghosh, S. et al. Fluorescence-Reported Allelic Exchange Mutagenesis-Mediated Gene Deletion Indicates a Requirement for Chlamydia trachomatis Tarp during In Vivo Infectivity and Reveals a Specific Role for the C Terminus during Cellular Invasion. Infection and Immunity 88, (2020). + +<|ref|>text<|/ref|><|det|>[[110, 665, 774, 685]]<|/det|> +38. Swanson, J. A. & Watts, C. Macropinocytosis. Trends in Cell Biology 5, 424-428 (1995). + +<|ref|>text<|/ref|><|det|>[[110, 699, 877, 753]]<|/det|> +39. Gu, C. et al. Regulation of Dynamin Oligomerization in Cells: The Role of Dynamin-Actin Interactions and Its GTPase Activity. Traffic 15, 819-838 (2014). + +<|ref|>text<|/ref|><|det|>[[110, 767, 870, 822]]<|/det|> +40. Soulet, F., Yarar, D., Leonard, M. & Schmid, S. L. SNX9 Regulates Dynamin Assembly and Is Required for Efficient Clathrin-mediated Endocytosis. Mol Biol Cell 16, 2058-2067 (2005). + +<|ref|>text<|/ref|><|det|>[[110, 836, 832, 890]]<|/det|> +41. Schlüter, K., Jockusch, B. M. & Rothkegel, M. Profilins as regulators of actin dynamics. Biochim Biophys Acta 1359, 97-109 (1997). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 825, 144]]<|/det|> +42. Itoh, T. et al. Dynamin and the Actin Cytoskeleton Cooperatively Regulate Plasma Membrane Invagination by BAR and F-BAR Proteins. Developmental Cell 9, 791-804 (2005). + +<|ref|>text<|/ref|><|det|>[[111, 156, 884, 245]]<|/det|> +43. Merrifield, C. J., Qualmann, B., Kessels, M. M. & Almers, W. Neural Wiskott Aldrich Syndrome Protein (N-WASP) and the Arp2/3 complex are recruited to sites of clathrin-mediated endocytosis in cultured fibroblasts. European Journal of Cell Biology 83, 13-18 (2004). + +<|ref|>text<|/ref|><|det|>[[111, 258, 843, 312]]<|/det|> +44. Benesch, S. et al. N-WASP deficiency impairs EGF internalization and actin assembly at clathrin-coated pits. Journal of Cell Science 118, 3103-3115 (2005). + +<|ref|>text<|/ref|><|det|>[[111, 325, 845, 380]]<|/det|> +45. Schlunck, G. et al. Modulation of Rac Localization and Function by Dynamin. MBoC 15, 256-267 (2004). + +<|ref|>text<|/ref|><|det|>[[111, 393, 844, 449]]<|/det|> +46. Marie-Anais, F., Mazzolini, J., Herit, F. & Niedergang, F. Dynamin-Actin Cross Talk Contributes to Phagosome Formation and Closure. Traffic 17, 487-499 (2016). + +<|ref|>text<|/ref|><|det|>[[111, 461, 876, 516]]<|/det|> +47. Romero, M. D. & Carabeo, R. A. Distinct roles of the Chlamydia trachomatis effectors TarP and TmeA in the regulation of formin and Arp2/3 during entry. J Cell Sci 135, jcs260185 (2022). + +<|ref|>text<|/ref|><|det|>[[111, 529, 863, 617]]<|/det|> +48. Faris, R., McCullough, A., Andersen, S. E., Moninger, T. O. & Weber, M. M. The Chlamydia trachomatis secreted effector TmeA hijacks the N-WASP-ARP2/3 actin remodeling axis to facilitate cellular invasion. PLOS Pathogens 16, e1008878 (2020). + +<|ref|>text<|/ref|><|det|>[[111, 630, 852, 685]]<|/det|> +49. Gu, C. et al. Direct dynamin-actin interactions regulate the actin cytoskeleton. EMBO J 29, 3593-3606 (2010). + +<|ref|>text<|/ref|><|det|>[[111, 698, 860, 753]]<|/det|> +50. Gu, C. et al. Regulation of dynamin oligomerization in cells: the role of dynamin-actin interactions and its GTPase activity. Traffic 15, 819-838 (2014). + +<|ref|>text<|/ref|><|det|>[[111, 766, 808, 820]]<|/det|> +51. Ford, C., Nans, A., Boucrot, E. & Hayward, R. D. Chlamydia exploits filopodia capture and a macropinocytosis-like pathway for host cell entry. PLOS Pathogens 14, e1007051 (2018). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 884, 175]]<|/det|> +52. Spona, D., Hanisch, P. T., Hegemann, J. H. & Mölleken, K. A single chlamydial protein reshapes the plasma membrane and serves as recruiting platform for central endocytic effector proteins. Commun Biol 6, 520 (2023). + +<|ref|>text<|/ref|><|det|>[[110, 191, 877, 279]]<|/det|> +53. Scanlon, K. R., Keb, G., Wolf, K., Jewett, T. J. & Fields, K. A. Chlamydia trachomatis TmeB antagonizes actin polymerization via direct interference with Arp2/3 activity. Frontiers in Cellular and Infection Microbiology 13, (2023). + +<|ref|>text<|/ref|><|det|>[[110, 292, 884, 380]]<|/det|> +54. Song, B. D., Yarar, D. & Schmid, S. L. An Assembly-incompetent Mutant Establishes a Requirement for Dynamin Self-assembly in Clathrin-mediated Endocytosis In Vivo. Mol Biol Cell 15, 2243-2252 (2004). + +<|ref|>text<|/ref|><|det|>[[110, 395, 864, 448]]<|/det|> +55. Chin, Y.-H. et al. Dynamin-2 mutations associated with centronuclear myopathy are hypermorphic and lead to T-tubule fragmentation. Human Molecular Genetics 24, 5542-5554 (2015). + +<|ref|>text<|/ref|><|det|>[[110, 462, 864, 550]]<|/det|> +56. Hamasaki, E. et al. The Lipid-Binding Defective Dynamin 2 Mutant in Charcot-Marie-Tooth Disease Impairs Proper Actin Bundling and Actin Organization in Glomerular Podocytes. Front Cell Dev Biol 10, 884509 (2022). + +<|ref|>text<|/ref|><|det|>[[110, 564, 844, 650]]<|/det|> +57. Szaszák, M. et al. Role of the Proline-rich Domain of Dynamin-2 and Its Interactions with Src Homology 3 Domains during Endocytosis of the AT1 Angiotensin Receptor. Journal of Biological Chemistry 277, 21650-21656 (2002). + +<|ref|>text<|/ref|><|det|>[[110, 666, 853, 718]]<|/det|> +58. Antonny, B. et al. Membrane fission by dynamin: what we know and what we need to know. The EMBO Journal 35, 2270-2284 (2016). + +<|ref|>text<|/ref|><|det|>[[110, 733, 872, 785]]<|/det|> +59. Warnock, D. E., Hinshaw, J. E. & Schmid, S. L. Dynamin self-assembly stimulates its GTPase activity. J Biol Chem 271, 22310-22314 (1996). + +<|ref|>text<|/ref|><|det|>[[110, 801, 822, 854]]<|/det|> +60. Ferguson, S. M. et al. Coordinated actions of actin and BAR proteins upstream of dynamin at endocytic clathrin-coated pits. Dev Cell 17, 811-822 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 88, 875, 135]]<|/det|> +676 61. Kruchten, A. E. & McNiven, M. A. Dynamin as a mover and pincher during cell migration and invasion. J Cell Sci 119, 1683–1690 (2006). + +<|ref|>text<|/ref|><|det|>[[55, 156, 880, 208]]<|/det|> +678 62. Grassart, A. et al. Actin and dynamin2 dynamics and interplay during clathrin-mediated endocytosis. Journal of Cell Biology 205, 721–735 (2014). + +<|ref|>text<|/ref|><|det|>[[55, 227, 861, 276]]<|/det|> +680 63. Mooren, O. L., Kotova, T. I., Moore, A. J. & Schafer, D. A. Dynamin2 GTPase and Cortactin Remodel Actin Filaments. J Biol Chem 284, 23995–24005 (2009). + +<|ref|>text<|/ref|><|det|>[[55, 295, 835, 345]]<|/det|> +682 64. Shin, N. et al. Sorting Nexin 9 Interacts with Dynamin 1 and N-WASP and Coordinates Synaptic Vesicle Endocytosis. Journal of Biological Chemistry 282, 28939–28950 (2007). + +<|ref|>text<|/ref|><|det|>[[55, 364, 880, 447]]<|/det|> +684 65. Kessels, M. M., Engqvist-Goldstein, Å. E. Y., Drubin, D. G. & Qualmann, B. Mammalian Abp1, a Signal-Responsive F-Actin–Binding Protein, Links the Actin Cytoskeleton to Endocytosis via the Gtpase Dynamin. Journal of Cell Biology 153, 351–366 (2001). + +<|ref|>text<|/ref|><|det|>[[55, 465, 850, 517]]<|/det|> +687 66. Caven, L. & Carabeo, R. A. Pathogenic Puppetry: Manipulation of the Host Actin Cytoskeleton by Chlamydia trachomatis. Int J Mol Sci 21, 90 (2019). + +<|ref|>text<|/ref|><|det|>[[55, 535, 852, 586]]<|/det|> +689 67. Byrne, G. I. & Moulder, J. W. Parasite-specified phagocytosis of Chlamydia psittaci and Chlamydia trachomatis by L and HeLa cells. Infect Immun 19, 598–606 (1978). + +<|ref|>text<|/ref|><|det|>[[55, 604, 852, 654]]<|/det|> +691 68. Caldwell, H. D., Kromhout, J. & Schachter, J. Purification and partial characterization of the major outer membrane protein of Chlamydia trachomatis. Infect Immun 31, 1161–1176 (1981). + +<|ref|>text<|/ref|><|det|>[[55, 672, 872, 722]]<|/det|> +693 69. Heinzen, R. A., Grieshaber, S. S., Van Kirk, L. S. & Devin, C. J. Dynamics of Actin-Based Movement by Rickettsia rickettsii in Vero Cells. Infect Immun 67, 4201–4207 (1999). + +<|ref|>text<|/ref|><|det|>[[55, 740, 837, 790]]<|/det|> +695 70. Pronobis, M. I., Deuitch, N. & Peifer, M. The Miraprep: A Protocol that Uses a Miniprep Kit and Provides Maxiprep Yields. PLOS ONE 11, e0160509 (2016). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 120, 800, 800]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[113, 90, 642, 109]]<|/det|> +## Figure 1: Dynamin 2 and actin are co-recruited during Chlamydia entry + +<|ref|>text<|/ref|><|det|>[[111, 125, 888, 800]]<|/det|> +(A) Cos7 cells were transfected with GFP-Dyn2 WT or K44A (DN) and miRFP-670 LifeAct for 24 hours prior to infection with wild-type Chlamydia at MOI=20. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes to identify sites exhibiting actin and Dyn2 co-recruitment. Scale bar = 1 micron. (B) Schematic depicting Dyn2 oligomerization, highlighting disruption of Dyn2 scission by K44A mutation. (C) Mean fluorescence intensity (MFI) of Dyn2 recruitment at Chlamydia entry sites was quantified, normalized as percent maximal MFI, and plotted onto a line graph depicting %max Dyn2 MFI +/- SEM for each timepoint. Background Dyn2 fluorescence was subtracted prior to normalization, which was performed independently for each Dyn2 WT and DN recruitment event. (D) Kymographs depicting RFP-Dyn2, GFP-Chlamydia, and far red actin signal over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (E-G) Detailed analysis of each recruitment event obtained via live cell imaging, plotting the (E) internalization duration, (F) rate of Dyn2 recruitment and (G) Dyn2 turnover of each event on a violin plot with inset boxplot reporting the median value +/- SD for each condition. (E) Internalization duration was quantified by calculating the elapsed time between initiation of protein recruitment and termination of pathogen entry, as detailed in Fig. S1. Individual rates of Dyn2 recruitment (F) and turnover (G) were calculated by measuring the slope derived from basal Dyn2 MFI to peak MFI for recruitment, and peak Dyn2 MFI to basal MFI for turnover, as detailed in Fig. S1. Data was obtained from a minimum N=23 individual rates. Statistical significance was determined by Wilcoxon ranked-sum. All data are representative of 3 independent experiments, * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[118, 130, 872, 860]]<|/det|> + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 702, 109]]<|/det|> +## Figure 2: Disruption of Dynamin 2 restricts actin turnover and Chlamydia entry + +<|ref|>text<|/ref|><|det|>[[111, 125, 888, 840]]<|/det|> +(A) Schematic depicting Dyn2 oligomerization, highlighting disruption of Dyn2 scission by Dynasore treatment. (B) Cos7 cells were transfected with GFP actin for 24 hours prior to infection with RFP CMTPX-stained wild-type Chlamydia (MOI=20). Infection was monitored by live-cell confocal microscopy, obtaining images every 20 seconds for 30 minutes to identify sites of actin recruitment proximal to invading bacteria. Actin recruitment at pathogen entry sites was quantified as described earlier (Fig. 1C) and plotted as %max actin MFI for each timepoint +/- SEM compiled from a minimum N=36 recruitment events. Upon completion of imaging, cells which received either scramble RNA or Dyn2 siRNA were lysed in 2x Laemmli buffer, resolving protein expression via Western blot to determine the knockdown efficiency of Dyn2 siRNA compared to actin loading control. Kinetics of (C) actin recruitment and (D) actin turnover, and (F) internalization duration were obtained using the same methodology described in Fig. 1E-G. Violin plots contain a minimum N=34 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank-sum. (E) Kymographs depicting RFP-Dyn2 and GFP-Chlamydia fluorescence over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (G,H) HeLa cells were infected with the indicated Chlamydia strain at MOI=50 and stained using the "in-and-out" method which distinguishes non-internalized EBs from total cell-associated EBs, as described in Materials and Methods. (G) Cells were pre-treated with 25 μM Dynasore for 30 minutes prior to infection, or (H) transfected with either scramble or Dyn2-specific siRNA for 24 hours prior to infection. Invasion efficiency of each Chlamydia strain was plotted as mean +/- SEM. Data was collected from 15 fields, with each field containing an average of 50 Chlamydia. Statistical significance was determined by pairwise T-test with Bonferroni post-correction. All data are representative of at least 3 independent experiments, * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 105, 880, 565]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[121, 88, 213, 107]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[112, 630, 886, 890]]<|/det|> +(A) Cos7 cells were transfected with GFP- or RFP-Dyn2 WT for 24 hours prior to infection with wild-type or \(\Delta \mathsf{Tm eA}\) EBs at MOI=20 in the presence or absence of \(10~\mu \mathrm{M}\) EHop-016. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes, identifying sites exhibiting Dyn2 recruitment during Chlamydia entry. Scale bar = 1 micron. (B) Schematic depicting TarP signaling via PI3K/Rac1, subsequent recruitment of actin and Dyn2, and Dyn2 oligomerization, highlighting EHop-016 inhibition of Rac1 and promotion of Dyn2 recruitment by TarP (C) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=20 recruitment events. (D) HeLa cells were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 383]]<|/det|> +treated with 10μM EHop- 016 for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the "in- and- out" method described earlier to quantify pathogen entry efficiency. Statistical significance was determined by pairwise T- test with Bonferroni post- correction. (E) Kymographs depicting RFP- Dyn2 and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=20 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank- sum. All data are representative of at least 3 independent experiments, \* P ≤ 0.05, \*\* P ≤ 0.01, \*\*\* P ≤ 0.001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 105, 880, 576]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[116, 90, 210, 108]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[112, 639, 886, 897]]<|/det|> +(A) Cos7 cells were transfected with GFP- or RFP-Dyn2 WT for 24 hours prior to infection with wild-type or \(\Delta \mathsf{Tm eA}\) EBs at MOI=20 in the presence or absence of 40nM Wortmannin. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes identifying sites exhibiting Dyn2 recruitment during Chlamydia entry. Scale bar = 1 micron. (B) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=18 recruitment events. (C) HeLa cells were treated with 40nM Wortmannin for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the "in-and-out" method described earlier to quantify pathogen entry + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 411]]<|/det|> +efficiency. Statistical significance was determined by pairwise T- test with Bonferroni post- correction. (D) 776 Schematic depicting TarP signaling via PI3K/Rac1, subsequent recruitment of actin and Dyn2, and Dyn2 777 oligomerization, highlighting Wortmannin inhibition of PI3K and promotion of Dyn2 recruitment by TarP 778 (E) Kymographs depicting RFP- Dyn2 and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top 779 arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen 780 entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were 781 obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=20 782 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon 783 Rank- sum. All data are representative of at least 3 independent experiments, \* P ≤ 0.05, \*\* P ≤ 0.01, \*\*\* 784 P ≤ 0.001. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 95, 880, 460]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 88, 201, 106]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[112, 525, 886, 885]]<|/det|> +(A) HeLa cells were treated with \(40\mu M\) Ryngo 1-23 for 30 minutes before infection with the indicated Chlamydia strains at MOI=50 and stained using the “in-and-out” method described earlier to quantify pathogen entry efficiency. Statistical significance was determined by pairwise T-test with Bonferroni post-correction. (B) Cos7 cells were transfected with RFP-Dyn2 WT for 24 hours prior to infection with \(\Delta\) TmeA EBs at MOI=20 in the presence or absence of \(40\mu M\) Ryngo 1-23. Infection was monitored by live-cell confocal microscopy using a Nikon CSU-W1 spinning disk microscope, obtaining images every 20 seconds for 30 minutes, highlighting Dyn2 recruitment at \(\Delta\) TmeA entry sites. Scale bar = 1 micron. (C) Schematic depicting Dyn2 oligomerization, promotion of Dyn2 self-assembly by TmeA signaling and enhancement of Dyn2 ring assembly via Ryngo 1-23 treatment. (D) Dyn2 recruitment was quantified as described earlier (Fig. 1C) and plotted as %max Dyn2 MFI for each timepoint +/- SEM compiled from a minimum N=19 recruitment events. (E) Kymographs depicting RFP-Dyn2 and GFP-Chlamydia fluorescence over a 30 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 886, 280]]<|/det|> +minute timelapse. Top arrow indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F- H) Internalization duration (F) and kinetics of Dyn2 recruitment (G) and turnover (H) were obtained using the same methodology described in Fig. 1E- G. Violin plots contain a minimum N=19 individual events, reporting the median value +/- SD. Statistical significance was determined by Wilcoxon Rank- sum. All data are representative of at least 3 independent experiments, \* P \(\leq 0.05\) , \*\* P \(\leq 0.01\) , \*\*\* P \(\leq 0.001\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 110, 876, 590]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[124, 90, 235, 112]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[112, 626, 812, 647]]<|/det|> +Figure 6: Actin turnover is correlated with Dynamin 2 activation status and Chlamydia uptake + +<|ref|>text<|/ref|><|det|>[[111, 653, 886, 878]]<|/det|> +(A) Cos7 cells were transfected with GFP-Actin or mRuby-LifeAct for 24 hours prior to infection with wild-type or \(\Delta\) TmeA EBs at MOI=20 in the presence or absence of \(40\mu \mathrm{M}\) Ryngo 1-23, monitoring pathogen invasion via live-cell confocal microscopy. Actin recruitment was quantified as described earlier (Fig. 1C) and plotted as %max actin MFI for each timepoint +/- SEM compiled from a minimum N=21 recruitment events. (B,C) Kinetics of Dyn2 recruitment (B) and turnover (C) were obtained using the same methodology described in Fig. 1F-G. Violin plots contain a minimum N=21 individual events, reporting the median rate +/- SD. Statistical significance was determined by Wilcoxon Rank-sum. (D) Schematic depicting Dyn2 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 88, 886, 383]]<|/det|> +813 oligomerization, promotion of Dyn2 self-assembly by TmeA signaling, enhancement of Dyn2 ring assembly 814 via Ryngo 1- 23 treatment, and proposed initiation of actin turnover following Dyn2 scission. (E) 815 Kymographs depicting RFP-actin and GFP- Chlamydia fluorescence over a 30 minute timelapse. Top arrow 816 indicates initiation of protein recruitment and bottom arrow indicates completion of pathogen entry. (F) 817 Internalization duration was quantified by calculating the elapsed time between initiation of actin 818 recruitment and termination of pathogen entry, as detailed in Fig. S1. Violin plots contain a minimum N=21 819 individual events, reporting the median internalization duration +/- SD. Statistical significance was 820 determined by Wilcoxon Rank-sum. All data are representative of at least 3 independent experiments, \* P 821 \(\leq 0.05\) , \*\* P \(\leq 0.01\) , \*\*\* P \(\leq 0.001\) . + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[42, 42, 312, 71]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[42, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 130, 393, 230]]<|/det|> +- VideoS1.mp4- VideoS2.mp4- VideoS3.mp4- SupplementalFiguresNatComm.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/images_list.json b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..c60c11b8c0029bb61c0fdd920d404b5848a6e241 --- /dev/null +++ b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/images_list.json @@ -0,0 +1,56 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1", + "footnote": [], + "bbox": [ + [ + 100, + 45, + 888, + 901 + ] + ], + "page_idx": 47 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2", + "footnote": [], + "bbox": [ + [ + 115, + 60, + 880, + 915 + ] + ], + "page_idx": 48 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_2.jpg", + "caption": "Extended Data Fig. 2", + "footnote": [], + "bbox": [], + "page_idx": 49 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_6.jpg", + "caption": "Extended Data Fig. 6", + "footnote": [], + "bbox": [], + "page_idx": 50 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_9.jpg", + "caption": "Extended Data Fig. 9", + "footnote": [], + "bbox": [], + "page_idx": 50 + } +] \ No newline at end of file diff --git a/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224.mmd b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224.mmd new file mode 100644 index 0000000000000000000000000000000000000000..29d55d8a0899d44dd40b3f95114fa2a692d9dc14 --- /dev/null +++ b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224.mmd @@ -0,0 +1,630 @@ + +# CoCo-ST: Comparing and Contrasting Spatial Transcriptomics data sets using graph contrastive learning + +Jia Wu Jku11@mdanderson.org + +The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0001- 8392- 8338 + +Muhammad Aminu The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 9903- 8812 + +Bo Zhu The University of Texas MD Anderson Cancer Center + +Natalie Vokes The University of Texas MD Anderson Cancer Center + +Hong Chen The University of Texas MD Anderson Cancer Center + +Lingzhi Hong The University of Texas MD Anderson Cancer Center + +Jianrong Li Baylor College Medicine + +Junya Fujimoto Hiroshima University + +Alissa Poteete The University of Texas MD Anderson Cancer Center + +Monique Nilsson The University of Texas MD Anderson Cancer Center + +Xiuning Li The University of Texas MD Anderson Cancer Center + +Tina Cascone UT M.D. Anderson Cancer Center + +David Jaffray The University of Texas MD Anderson Cancer Center + +Nicholas Navin The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 2106- 8624 + +Lauren Byers + +<--- Page Split ---> + +The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 0780- 2677 + +Don Gibbons The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0003- 2362- 3094 + +John Heymach MD Anderson Cancer Center https://orcid.org/0000- 0001- 9068- 8942 + +Ken Chen The University of Texas MD Anderson Cancer Center + +Chao Cheng Baylor College of Medicine https://orcid.org/0000- 0002- 5002- 3417 + +Jianjun Zhang The University of Texas MD Anderson Cancer Center + +Yuqui Yang UT Southwestern University + +Tao Wang The University of Texas Southwestern Medical Center https://orcid.org/0000- 0002- 4355- 149X + +Bo Wang University of Toronto + +## Letter + +Keywords: + +Posted Date: May 20th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 4359834/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Cell Biology on October 13th, 2025. See the published version at https://doi.org/10.1038/s41556- 025- 01781- z. + +<--- Page Split ---> + +# CoCo-ST: Comparing and Contrasting Spatial Transcriptomics data sets using graph contrastive learning + +Muhammad Aminu1,10, Bo Zhu2,10, Natalie Vokes2,10, Hong Chen2, Lingzhi Hong2, Jianrong Li9, Junya Fujimoto8, Yuqiu Yang12, Tao Wang12, Bo Wang13, Alissa Poteete2, Monique B. Nilsson2, Xiuning Le2, Cascone Tina2, David Jaffray3,7, Nick Navin5, Lauren A. Byers2, Don Gibbons2, John Heymach2, Ken Chen6, Chao Cheng9, Jianjun Zhang2,11 & Jia Wu1,2,7,11 + +1Department of Imaging Physics, 2Department of Thoracic/Head and Neck Medical Oncology, 3Office of the Chief Technology and Digital Officer, 5Department of Systems Biology, 6Department of Bioinformatics and Computational Biology, 7Institute for Data Science in Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA. 8Clinical Research Center, Hiroshima University, Hiroshima, Japan. 9Department of Medicine, Institution of Clinical and Translational Research, Baylor College of Medicine, Houston, TX, USA. Department of Public Health, UT Southwestern Medical Center, Dallas, TX, USA. 13Department of Medical Biophysics, University of Toronto, Ontario, Canada. 10These authors contributed equally: Muhammad Aminu, Bo Zhu, Natalie Vokes. 11Co- senior authors: Jianjun Zhang, Jia Wu. + +## Corresponding Author + +Jia Wu, PhDDepartment of Imaging PhysicsDepartment of Thoracic/Head and Neck Medical Oncology + +<--- Page Split ---> + +The University of Texas MD Anderson Cancer Center + +1515 Holcombe Blvd + +Houston, TX 77030, USA + +Telephone: 713- 563- 2719 + +e- mail: jwu11@mdanderson.org + +<--- Page Split ---> + +## Abstract + +Traditional feature dimension reduction methods have been widely used to uncover biological patterns or structures within individual spatial transcriptomics data. However, these methods are designed to yield feature representations that emphasize patterns or structures with dominant high variance, such as the normal tissue spatial pattern in a precancer setting. Consequently, they may inadvertently overlook patterns of interest that are potentially masked by these high- variance structures. Herein we present our graph contrastive feature representation method called CoCo- ST (Comparing and Contrasting Spatial Transcriptomics) to overcome this limitation. By incorporating a background data set representing normal tissue, this approach enhances the identification of interesting patterns in a target data set representing precancerous tissue. Simultaneously, it mitigates the influence of dominant common patterns shared by the background and target data sets. This enables discerning biologically relevant features crucial for capturing tissue- specific patterns, a capability we showcased through the analysis of serial mouse precancerous lung tissue samples. + +<--- Page Split ---> + +Analyzing spatial transcriptomics (ST) data requires robust feature representation methods to effectively capture the intricate biological information or patterns enriched in these high- dimensional data sets. Although traditional dimension reduction techniques like principal component analysis (PCA \(^{1}\) ) and nonnegative matrix factorization (NMF \(^{2}\) ) have been widely adopted as off- the- shelf approaches for ST data dimension reduction, they primarily aimed at capturing global patterns and variations in the original high- dimensional ST data sets. More recently, the integration of spatial constraints into dimension reduction algorithms has led to the emergence of robust feature representation approaches such as nonnegative spatial factorization \(^{3}\) , spatial PCA \(^{4}\) , and MEFISTO \(^{5}\) . However, these methods tend to prioritize the identification of prominent global patterns with high variability, potentially missing finer localized intrinsic structures marked by lower variability. Furthermore, they are designed to explore one data set at a time and are not tailored to studying the evolutionary dynamics of a tumor microenvironment across multiple data sets. These constraints can result in overlooked information, particularly when studying carcinogenesis, in which tumors progress from a few isolated precancerous sites to invasive cancer across various tissue samples. The majority of these samples exhibit common global patterns (representing normal tissue biology) that may not be of primary interest. Conversely, a small portion of samples contain unique, crucial precancerous structures that require specific attention. + +To address these constraints, we proposed a graph contrastive learning framework that we called CoCo- ST (Compare and Contrast Spatial Transcriptomics). CoCo- ST operates by taking two ST data sets as inputs: one serving as the reference (background) and another as the target. These ST data sets typically have certain common structures that are usually not the primary foci. + +<--- Page Split ---> + +The goal is to extract feature representations that emphasize the new and unique structures enriched in the target ST data set. + +In the present study, we used CoCo- ST to thoroughly investigate carcinogenesis using ST data sets from an in- house curated carcinogenesis mouse model. This approach yielded feature representations that enhanced our ability to discern distinctive and noteworthy structures within the target ST data, leading to improvements in downstream analysis. + +CoCo- ST was inspired by the recent successes of contrastive learning approaches6- 8, which learn discriminative feature representations by contrasting positive pairs (similar samples) with negative pairs (dissimilar samples). In our CoCo- ST design workflow (Fig. 1a), we began by collecting tissue samples from mouse lung and processing them using the Visium technology (10x Genomics) to obtain the ST data. We then organized the resulting gene expression data into a gene- spot matrix and further normalized the data to eliminate technical artifacts. CoCo- ST proceeded to construct two weighted graphs, one each for the background and target ST data sets—allowing us to capture the local structures within the data sets. We derived contrastive feature representations by comparing and contrasting the local variances of the background and target graphs. We achieved this by assessing the difference between their respective local total scatter matrices. In the case of a new target ST data set, CoCo- ST simply uses the learned transformation to generate feature representations for the new data (Fig. 1a). These contrastive feature representations can serve as inputs for various other ST analysis tools, for enhanced downstream analysis. We have illustrated the effectiveness of these contrastive feature representations across multiple downstream analysis tasks, including ST data visualization, spatial domain identification, tissue- specific spatial trajectory inference, trajectory inference across multiple tissues, and + +<--- Page Split ---> + +examination of cell- cell interaction. It is worth mentioning here that CoCo- ST is generically applicable to any ST data types that can be represented in form a gene- spot matrix. + +We first applied CoCo- ST to learn transformation by using a mouse normal lung tissue sample (MLP- 1) as the background and an abnormal lung tissue sample (MLP- 6) containing structures other than the normal spatial domain (Extended Data Fig. 1) as the target. We designated MLP- 1 as the background ST data because its spatial structures belong to the normal lung spatial domain, which was also present in all the rest of the tissue samples. We then applied the learned transformation to the remaining tissue samples, resulting in contrastive feature representations that we subsequently used for spatial domain identification (Extended Data Fig. 1) and further downstream analysis. Note, CoCo- ST does not require much data to determine a good transformation compared to the conventional machine learning approaches. Additionally, it has the potential to capture more specific structures within individual samples. These properties make CoCo- ST a valuable complement to large foundation model- based approaches. + +Uniform manifold approximation and projection (UMAP) embedding of the learned contrastive features in the target ST data (Extended Data Fig. 2a) illustrated CoCo- ST's effectiveness in determining feature representations that provide robust discrimination of various spatial structures in the target tissue (Fig. 1b). Clustering the ST data based on the learned contrastive components led to the identification of six clusters, each corresponding to a unique spatial structure. These spatial structures detected using CoCo- ST's contrastive components agree well with pathologist- annotated regions (Fig. 1b). Spatial clustering of spots based on components determined using the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods failed to effectively detect the hotspot region annotated as hyperplasia by the pathologist (Fig. 1b). Inability to detect spatial structures of low variability affects the performance of the compared + +<--- Page Split ---> + +methods in detecting the early adenoma (hotspot) region. However, Seurat (PCA) detected the hotspot region but annotated it as belonging to spatial domain 2. + +We further annotated the detected spatial structures detected using CoCo- ST based on their differentially expressed marker genes (Extended Data Fig. 2b) and spatial locations. The distribution of these marker genes, including \*Epas1\* for normal lung tissue (endothelial PAS domain), Slc26a4 for fibrotic/scarred tissue, Cybb for adjacent normal tissue, Hp for the bronchus/alveoli, Ctsh for the adenoma, and Msln for the membrane, showed the expected high expression patterns (Extended Data Fig. 2c). To further validate the adenoma region (hotspot) detected using CoCo- ST, we investigated the most differentially expressed marker genes for the detected adenoma regions and found 3498 marker genes at a false- discovery rate of 5% (Fig. 1c). The most differentially expressed marker genes were domain- specific metagenes for the adenoma region (including the hotspot region). For example, a metagene consisting of \*Ctsh\*, \*Cxcl15\*, and \*Slc34a2\* marked the hotspot region clearly, as these genes exhibited high expression patterns in both the larger adenoma region and smaller hotspot region (Fig. 1e). The \*Cxcl15\*, and \*Slc34a2\* genes are uniquely identified by CoCo- ST. The high expression of these genes at both the large and hotspot adenoma regions indicates that these two spatial domains are anatomically similar. Seurat's inability to identify these important marker genes results to categorizing the hotspot region as belonging to the fibrotic/scarred tissue (Fig. 1b). Also, \*Ctsh\* gene was reported to be differentially expressed in adenoma region of patients with colorectal cancer9. Gene set enrichment analysis of the 10 most differentially expressed marker genes in our study identified biological processes related to lung fibrosis, apoptotic processes, and cell polarity (Extended Data Fig. 2d). For comparison, we also investigated the most differentially expressed marker genes for the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods (Fig. 1d, Extended Data + +<--- Page Split ---> + +Fig. 2e) based on the learned embedding of these methods and found several genes, most of which marked the larger adenoma region but not the smaller hotspot region. For example, the Trf gene was the top marker gene for all of the compared methods (Extended Data Fig. 2e); however, this gene had a high expression pattern in the larger adenoma region but not in the hotspot region (Fig. 1f). These results demonstrated that the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods focus on identifying the main adenoma region with the largest variance, lacking the ability to identify domain- specific metagenes that capture the smaller adenoma structure (hotspot) with relatively low variance. + +Examining the weights of the first five contrastive components revealed that CoCo- ST effectively identified major spatial domains (Fig. 2a), indicating that it captured local variations associated with the interesting spatial structures in the target data. For example, component 1 explained variation in multiple spatial domains, which was characterized by large positive weights around the adenoma and alveoli/bronchus and negative weights around the normal lung. Comparing to Seurat (PCA), STUtility (NMF), NSF and MEFISTO, the top components of these methods predominantly focus on the normal lung structure with the largest variance (Fig. 2d). For example, the first components of both Seurat PCA and NSF exhibited larger weights on normal lung structures. Because the first few components of these methods are expected to capture most of the information in the original data and are subsequently used as inputs for downstream analysis, relying solely on these components may result in overlooking crucial biological insights. To gain deeper insight into the underlying biological processes associated with these components, we further investigated the top 20 genes with the largest weights on each of the CoCo- ST's contrastive components (Fig. 2b). This highlighted individual genes encoding domain- specific signatures such as Retnla, Cyp2f2, Ctsh, Ccl6, and Acta2 (Fig. 2c) as well as gene sets linked with broader + +<--- Page Split ---> + +biological processes and pathways. Gene set enrichment analysis with the top 20 marker genes for each component revealed enriched gene ontology terms and KEGG pathways specific to each spatial domain. These included heme binding on component 1, retinol metabolism on component 2, IgA immunoglobulin complex on component 3, lysosome on component 4, and extracellular matrix on component 5 (Extended Data Fig. 3). + +To investigate the impact of different graph construction methods (molecular vs. spatial) on CoCo- ST's performance, we constructed a similarity graph based on spatial coordinates rather than gene expression data as done in our prior experiments. This approach has proven highly effective10, as it assumes that neighboring spots in the tissue have similar gene expression patterns and likely belong to the same spatial domain. Our findings demonstrated robust CoCo- ST performance when using the similarity graph constructed from the spatial coordinates, effectively identifying the major spatial domains across all target tissue samples (Extended Data Fig. 4). In summary, CoCo- ST demonstrates robust performance with similarity graphs constructed from both spatial coordinates and gene expression data. + +Next, we performed deconvolution analysis to infer the cell type composition at each of the spatial domains detected using CoCo- ST. For this analysis, we used matched single- cell RNA sequencing (scRNA- seq) data (Extended Data Fig. 5a) obtained from the same MLP tissue samples as a reference. As expected for the MLP- 6 tissue sample (Extended Data Fig. 5b), we observed a concentration of endothelial cells in the normal lung spatial domain (endothelial PAS domain) (Extended Data Figs. 5c,d, 6, 7). The fibrotic/scarred and bronchus/alveoli spatial domains were enriched with fibroblasts. In the adjacent normal spatial domain was an abundance of endothelial cells, whereas the adenoma spatial domain had enrichment of macrophages and proliferating macrophages (Extended Data Figs. 5c,d, 6, 7). Notably, we observed tumor- associated + +<--- Page Split ---> + +macrophages (TAMs) in the adjacent normal spatial domain (Extended Data Figs. 5c,d, 6, 7), which exhibited significantly upregulated Ccl6. This gene was the top gene with the highest weight on component 4 (Fig. 2c). Of note, component 4 exhibited large weights in spatial regions corresponding to the regions with the highest Ccl6 gene expression. Also, high expression of the Ccl6 gene in a mouse model of lung cancer was reported to be associated with tumor growth and increased metastasis11. This evidence underscores the intricate cellular compositions within specific spatial domains, shedding light on potential implications for the progression of lung cancer. + +After determining the composition of cell types in the various tissue samples through our deconvolution analysis, we next inferred their communication patterns. Initially, we identified cell- cell interactions by examining ligand- receptor patterns within the individual MLP tissue samples. Our analysis of the MLP- 6 tissue sample revealed a strong pattern of communication between endothelial and epithelial cells as well as between endothelial cells and fibroblasts (Extended Data Fig. 5e). Also, we observed strong communication initiating from both proliferating macrophages and B cells within the adenoma spatial domain, indicating an active immune response. + +We observed that multiple signaling pathways, including programmed death- ligand 1, GRN, inducible co- stimulator, NECTIN, interleukin- 6, WNT, and CXCL, played pivotal roles in cell interactions across different spatial domains. Notably, we predominantly observed WNT ligand- receptor interactions in endothelial cells, epithelial cells, fibroblasts, and macrophages (Extended Data Fig. 5f). Additionally, we observed WNT signaling interactions between proliferating macrophages and B cells, which are enriched in the adenoma spatial domain. Meanwhile, we found self- interaction (among cells of the same group) to be the strongest in proliferating T cells, proliferating macrophages, and endothelial cells (Extended Data Fig. 5g). + +<--- Page Split ---> + +Network centrality analysis of the inferred WNT signaling network identified TAMs (macrophages and proliferating macrophages) as prominent mediators (gatekeepers) as well as influencers controlling the communication (Extended Data Fig. 5h). Prior studies demonstrated that WNT signaling supports TAMs as drivers of tumor growth and that TAM- derived WNT ligands support tumorigenesis12. + +We delved deeper into the cell- cell interactions across groups of tissue samples associated with the adenoma and adenocarcinoma spatial domains as determined using CoCo- ST. Specifically, we aggregated the communication weights of multiple tissue samples containing the adenoma (MLP- 3, - 4, - 5, - 6, - 7, and - 9) and adenocarcinoma (MLP- 8 and - 10) spatial domains to investigate the cell- cell interactions on a broader scale. Of note, we observed a bidirectional interaction between epithelial cells and proliferating macrophages in the adenocarcinoma group (Extended Data Fig. 8a,b) but did not see a similar interaction pattern in the adenoma group (Extended Data Fig. 8c,d). This is consistent with the established role of TAMs in promoting tumor growth and metastasis by engaging in an autocrine loop with cancer cells, thereby stimulating cancer cell progression13- 16. + +Next, we investigated how the normal endothelial, adjacent normal, and tumor spatial domains are connected to each other during tumorigenesis. Specifically, we performed spatial trajectory inference with MLP- 6 tissue using the contrastive components derived from CoCo- ST. This analysis revealed a trajectory starting from the normal endothelial domain and moving toward the adjacent normal domain and further into the adenoma spatial domain (Extended Data Fig. 9a,b). To gain a comprehensive view of the trajectory of precancer evolution across the entire population, we combined spots belonging to the adenoma and adenocarcinoma spatial domains as identified by our contrastive components. We then determine a UMAP embedding of the spots + +<--- Page Split ---> + +(Extended Data Fig. 9c) with which the trajectories were reconstructed (Extended Data Fig. 9d). As seen in Extended Data Fig. 9c, the contrastive components effectively discriminated the three spatial domains and identified a trajectory starting from the normal lung, passing toward the adenoma, and ending at the adenocarcinoma cluster (Extended Data Fig. 9d). These findings align with the well- known biology of mouse tumorigenesis, consisting of a transition from normal tissue to hyperplasia, adenoma, and finally adenocarcinoma. Furthermore, we identified modules of differentially expressed genes that were co- expressed across spots in the different spatial domains as determined using CoCo- ST (Extended Data Fig. 9e,f). Notably, these modules demonstrated high specificity for the different spatial domains, further indicating the effectiveness of CoCo- ST in determining feature representations that captured both the shared and unique spatial structures across the different tissues. + +Lastly, we employed CoCo- ST to analyze a publicly available Visium data set generated from mouse brain (anterior and posterior). This data set shows tissue structures that are considerably more complex than the mouse lung precancer data set described above. First, we examined the spatial domain identification performance of CoCo- ST when considering the anterior slice as the reference and the posterior slice as the target and vice versa. The spatial domains detected using CoCo- ST's contrastive components agree well with the Allen Institute for Brain Science reference atlas diagram (Extended Data Fig. 10a) \(^{17}\) . We further investigated the top five contrastive components as determined using CoCo- ST for both the anterior and posterior slices. All of these components captured spatial patterns highlighting specific major anatomical regions in the brain (Extended Data Fig. 10b,e). Similar to the mouse precancer model, these components exhibited high component values on specific anatomical regions, such as the cerebral cortex (for anterior component 1) and choroid plexus (for posterior component 2). The top genes for each + +<--- Page Split ---> + +component (Extended Data Fig. 10c,f) had distinct spatial patterns and exhibited spatial localization to specific brain regions (Extended Data Fig. 10d,g). + +To summarize, we introduced an ST feature representation method that opens up the application of graph contrastive learning to ST data analysis. This approach offers significant advantages, particularly in scenarios involving the analysis of multiple ST data sets. It effectively identifies interesting, unique spatial structures in a target ST data set while mitigating the influence of dominant high- variance spatial structures that are common to both target and background ST data sets. Whereas we focused on the ST and Visium platforms, adaptation of CoCo- ST to other platforms such as Xenium, CosMX SMI and MERFISH on which the data can be represented in the form of a gene spot matrix is plausible. + +<--- Page Split ---> + +## Methods + +## Problem definition and notation + +We represented a spatially resolved ST slice from a spatial genomics technology as the set of pairs \(\{x_{i},y_{i}\}_{i = 1}^{n}\) , with \(y_{i}\in R^{2}\) denoting a vector of spatial coordinates and \(x_{i}\in R^{d}\) denoting a vector of measured gene expression at a corresponding spatial location. We referred to a single spatial location \(x_{i}\) as a spot and \(s\in \{1,2,\dots,S_{i}\}\) as a slice containing \(n_{s}\) spots. Let \(X_{s} = \left[x_{1}^{s},x_{2}^{s},\dots,x_{n_{s}}^{s}\right]^{T}\) denote the matrix containing the spot gene expression measurements and \(Y_{s} = \left[y_{1}^{s},y_{2}^{s},\dots,y_{n_{s}}^{s}\right]^{T}\) denote the corresponding spatial location matrix from slice \(s\) . Worth noting is that the number of spots can differ across different slices and that the slices may be from the same tissue sample or from two different tissue sample. + +Our goal is to analyze these \(S_{l}\) slices by finding discriminative feature representations that capture the interesting spatial patterns within the different slices. To do this, we identified a background ST data set containing dominant high- variance spatial structures that were present across all slices. + +The background ST data play a crucial role in effectively contrasting dominant high- variance spatial structures, which was not the primary focus of this analysis, and in turn assists in detecting the intriguing unique spatial structures enriched in individual target slices. Three key advancements underlie the robust performance of our graph contrastive learning approach. First, we used paired slices to mitigate the impact of spatial structures that are not of primary interest, which subsequently aided the detection of unique spatial structures of particular interest in individual target slices. Second, we constructed local similarity graphs to capture the nuanced local structures in both the background and target ST data sets, thereby ensuring that important spatial structures are not lost. Third, we applied the concept of contrastive learning to compare and + +<--- Page Split ---> + +contrast the graph embedding of the background and target ST data sets, ensuring that similar spots are positioned close to each other and that dissimilar ones are distanced in the latent space. This collective methodology ensures the accurate identification and representation of distinctive spatial structures. + +## Graph representation learning + +Recent advances in spatial molecular profiling made graph learning a focus of attention because of the innate resemblance of spatial information to spatial graphs. Graph embedding techniques have great potential for various applications across spatially resolved transcriptomics. Because ST data sets can be represented in a matrix format, we can identify spots as entities of interest and interrogate their interaction. This is equivalent to constructing gene or spot graphs based on suitable similarity measures. Herein we describe the construction of such molecular similarity graphs. An essential task in ST data analysis is to find a lower dimensional manifold space that captures local neighborhood information. Given an ST datum (slice), we can construct a weighted graph \(G = (V, E)\) representing complex, non- Euclidean structures, with edges \(e_{ij} \in E\) connecting nearby nodes \(i\) and \(j\) ( \(i, j \in V\) ) to each other if spots \(x_{i}^{s}\) and \(x_{j}^{s}\) are molecularly similar. A natural variation of this graph is to construct a graph of \(k\) - nearest neighbors in which similarity of nodes is usually quantified using the Euclidean metric (i.e., nodes \(i\) and \(j\) are connected by an edge \(e_{ij}\) if \(x_{i}^{s}\) is among the \(k\) - nearest neighbors of \(x_{j}^{s}\) or \(x_{j}^{s}\) is among the \(k\) - nearest neighbors of \(x_{i}^{s}\) ). The graph structure \(G = (V, E)\) is commonly encoded in an \(n_{s} \times n_{s}\) affinity matrix \(S\) with entries in \([0, 1]\) and takes large values if \(x_{i}^{s}\) and \(x_{j}^{s}\) are close (or similar). Several approaches to computing the affinity matrix \(S\) are available, one of which is the heat kernel weighting technique depicted by the equation + +<--- Page Split ---> + +\[S_{ij}^{s} = \left\{ \begin{array}{l l}{\frac{\left\|x_{i}^{s} - x_{j}^{s}\right\|^{2}}{t},} & {\mathrm{if~}x_{i}^{s}\in N\left(x_{j}^{s}\right)\mathrm{~or~}x_{j}^{s}\in N\left(x_{i}^{s}\right)}\\ 0, & {\mathrm{Otherwise}} \end{array} \right. \quad (1),\] + +where \(N(x_{j}^{s})\) denotes the set of \(k\) - nearest neighbors of \(x_{j}^{s}\) and \(t\) is a user- specified parameter. + +Based on the graph construction approach described above, the similarity among spots is quantified based on gene expression measurements at the corresponding spots. However, because gene expression measurements are captured alongside its spatial information in ST, these spatial locations can be used to construct similarity graphs. The spatial graphs constructed in this way are similar to molecular similarity graphs in the sense that nodes correspond to spots. However, edges capture proximity of spots in the \(R^{2}\) coordinate space. The affinity matrix with the spatial locations can now be constructed as + +\[S_{ij}^{s} = \left\{ \begin{array}{l l}{\frac{\left\|y_{i}^{s} - y_{j}^{s}\right\|^{2}}{t},} & {\mathrm{if~}y_{i}^{s}\in N\left(y_{j}^{s}\right)\mathrm{~or~}x_{j}^{s}\in N\left(x_{i}^{s}\right)}\\ 0, & {\mathrm{Otherwise}} \end{array} \right. \quad (2).\] + +Also, the spatial graph can be constructed using both the spatial locations and the molecular profiles treated as node features. Graph representation learning approaches are considered to determine biologically meaningful representations of these graphs by finding meaningful lower dimensional representations of nodes present in a complex graph, where local structures in the data are well captured. A widely used criterion for determining such a representation is to solve the objective function + +\[\min_{W}\sum_{i,j = 1}^{n_{s}}\left\| z_{i}^{s} - z_{j}^{s}\right\|^{2}S_{ij}^{s} \quad (3),\] + +where \(z_{i}^{s} = W^{T}x_{i}^{s}\) denotes the lower dimensional representation of \(x_{i}^{s}\) . Solving Eq. (3) under appropriate constraints ensures that if \(x_{i}^{s}\) and \(x_{j}^{s}\) are similar (or nodes \(i\) and \(j\) are connected in the graph), then \(z_{i}^{s}\) and \(z_{j}^{s}\) are similar (close), as well. + +<--- Page Split ---> + +## Contrastive representation learning + +Contrastive learning has recently emerged as a successful method of unsupervised graph representation learning. Contrastive learning methods first perform augmentation of the input data and enforce via a suitable objective function mapping of augmentation of the same data (positive pairs) close to each other in the representation (latent) space and augmentation of different data (negative pairs) far apart from each other. Arguably, a low- dimensional representation that is near optimal in the contrastive objective function is guaranteed to linearly separate similar data from dissimilar data. Such representations provide competitive performance in a host of downstream tasks. In early visual representation learning studies, researchers leveraged a pixel as local view to conduct local- to- local \(^{18}\) or local- to- global \(^{19}\) contrastive learning, whereas researchers recently found that randomly cropped image snippets help contrastive models better capture the relationships between image elements \(^{6}\) . This motivated us to perform contrastive representation learning at the global image level. + +Like several other machine learning approaches, contrastive representation learning can be performed in an unsupervised (self- supervised) or supervised learning strategy. In self- supervised settings, contrastive learning methods learn discriminative feature representations based on some similarity measure defined according to the data. Consider the objective function defined by \(^{20}\) + +\[L_{1} = (1 - Y)*\frac{1}{2}\left\| x_{i} - x_{j}\right\|^{2} + \frac{Y}{2}*\left\{\max \left(0,m - \left\| x_{i} - x_{j}\right\|^{2}\right)\right\}^{2}\] + +(4), + +where \(m > 0\) is a hyperparameter defining the lower bound distance between dissimilar samples, \(Y\) is a binary label with \(Y = 0\) if \(x_{i}\) and \(x_{j}\) are similar, and \(Y = 1\) if \(x_{i}\) and \(x_{j}\) are dissimilar. Minimizing the objective function is an attempt to determine a lower dimensional manifold + +<--- Page Split ---> + +subspace where similar input samples are mapped nearby and dissimilar samples are far apart. When sample labels are available, they can be integrated into the definition of similarity and dissimilarity to better guide the contrastive model to mapped samples belonging to the same class (same label) close to each other and samples of different classes farther apart. This approach is referred to as supervised contrastive representation learning. Both the self- supervised and fully supervised contrastive learning approaches are powerful methods of learning discriminative feature representations. + +## Graph contrastive feature representation using CoCo-ST + +Most of the traditional feature representation approaches are designed to determine feature representations through maximization of data variance. These approaches can perform poorly if the ST data structures with maximal variances are not the structures of interest, as the local structures of interest are masked by the dominant high- variance structures. The feature representations determined using these approaches capture little to no useful information reflecting the unique low- variance local structures present in the ST data, which are usually treated as noise. Also, these traditional approaches are designed to explore one ST data set at a time, which can hinder their performance in cases where there are multiple interconnected data sets that need to be explored. + +To overcome these limitations, we propose CoCo- ST, which compares and contrasts the global and local variances in ST data sets to better capture discriminant and structural information. More generally, we use two ST data sets (background and target) and subsequently construct two similarity graph views: one for the background ST data set and the other for the target ST data set. We then design a contrastive objective function to learn feature representations that capture high + +<--- Page Split ---> + +global (and/or local) variances enriched in the target ST data while simultaneously attaining small global (and/or local) variances in the background ST data. Given a background ST data set \(X_{b} = \left[x_{1}^{b}, x_{2}^{b}, \ldots , x_{n_{b}}^{b}\right]^{T}\) containing spatial structures of no primary interest, such as a normal lung region, we can use the following two terms to measure the smoothness of the lower dimensional representation: + +\[\begin{array}{l}{{\mathcal{R}_{1}=\min _{W}\sum_{i=1}^{n_{b}}\left\|x_{i}^{b}-W W^{T}x_{i}^{b}\right\|^{2}}}\\ {{\quad=\max _{W}t r(W^{T}X_{b}X_{b}^{T}W)}}\end{array} \quad (5)\] + +and + +\[\begin{array}{r l} & {\mathcal{R}_{2} = \min_{W}\sum_{i,j = 1}^{n_{b}}\left\| W^{T}x_{i}^{b} - W^{T}x_{j}^{b}\right\|^{2}S_{i j}^{b}}\\ & {= \min_{W}\left(\sum_{i,j = 1}^{n_{b}}W^{T}x_{i}^{b}D_{i i}^{b}\big(x_{i}^{b}\big)^{T}W^{T} - W^{T}x_{i}^{b}S_{i j}^{b}\big(x_{j}^{b}\big)^{T}W^{T}\right)}\\ & {\qquad = \min_{W}t r(W^{T}X_{b}L^{b}X_{b}^{T}W)} \end{array} \quad (6),\] + +where \(t r(\cdot)\) is the trace operator, \(D\) is a diagonal matrix whose entries are the column (or row) sums of \(S\) , \(D_{i i} = \sum_{k} S_{i k}\) , and \(L = D - S\) is the graph Laplacian matrix. We consider the symmetric normalized graph Laplacian matrix \(\bar{L} = D^{- 1 / 2} L D^{- 1 / 2}\) in our later derivations. By minimizing \(\mathcal{R}_{1}\) , we aim to minimize the reconstruction error, whereas minimizing \(\mathcal{R}_{2}\) is an attempt to preserve the local structure (i.e., if two spots \(x_{i}^{s}\) and \(x_{j}^{s}\) are molecularly similar, their low- dimensional representations \(W^{T} x_{i}^{b}\) and \(W^{T} x_{j}^{b}\) are also similar). Combining Eqs. (5) and (6), we can have the equivalent formulation + +<--- Page Split ---> + +\[\begin{array}{l}{{O_{1}=\max _{W^{T}W=I}t r(W^{T}X_{b}X_{b}^{T}W)-\mu_{1}t r(W^{T}X_{b}\bar{L}^{b}X_{b}^{T}W)}}\\ {{\quad=\max _{W^{T}W=I}t r(W^{T}X_{b}H_{b}X_{b}^{T}W)}}\end{array} \quad (7), \quad (7)\] + +where \(H_{b} = I - \mu_{1}\bar{L}^{b}\) , \(I\) is an identity matrix, \(\bar{L}^{b}\) is the normalized graph Laplacian for the background ST data, and \(0 \leq \mu_{1} \leq 1\) is a hyperparameter that controls the smoothness of the new representation. The matrix \(H_{b} = I - \mu_{1}\bar{L}^{b}\) can be considered a graph Laplacian filter \(^{21}\) that helps smooth the data while preserving underlying spatial structures in an ST slice. + +Similarly, for a target ST data set \(X_{t} = \left[x_{1}^{t}, x_{2}^{t}, \ldots , x_{n_{t}}^{t}\right]^{T}\) containing unique, interesting spatial structures, we can write the formulation + +\[O_{2} = \max_{W^{T}W = I}t r(W^{T}X_{t}H_{t}X_{t}^{T}W) \quad (8),\] + +where \(H_{t} = I - \mu_{2}\bar{L}^{t}\) , \(\bar{L}^{t}\) is the normalized graph Laplacian for the target ST data and \(0 \leq \mu_{2} \leq 1\) is a hyperparameter. + +Combining Eqs. (7) and (8), CoCo- ST solves the following objective function + +\[O_{3} = \max_{W^{T}W = I}t r(W^{T}X_{t}H_{t}X_{t}^{T}W) - \eta t r(W^{T}X_{b}H_{b}X_{b}^{T}W) \quad (9),\] + +where \(\eta \geq 0\) is the contrastive parameter that determines the tradeoff between high target global (and/or local) variance and low background global (and/or local) variance. We will first describe how to maximize the objective function \(O_{3}\) . Let \(\Lambda\) be the Lagrange multiplier for the constraint \(W^{T}W = I\) . The Lagrange \(\mathcal{L}\) is + +<--- Page Split ---> + +\[\mathcal{L} = t r(W^{T}X_{t}H_{t}X_{t}^{T}W) - \eta t r(W^{T}X_{b}H_{b}X_{b}^{T}W) - \Lambda t r(W^{T}W - I) \quad (10).\] + +The partial derivative of \(\mathcal{L}\) with respect to \(W\) is + +\[\frac{\partial\mathcal{L}}{\partial W} = X_{t}H_{t}X_{t}^{T}W - \eta X_{b}H_{b}X_{b}^{T}W - \Lambda W\] + +(11). + +The optimum solution to Eq. (10) satisfies \(\frac{\partial\mathcal{L}}{\partial W} = 0\) . We therefore have + +\[\begin{array}{r l} & {X_{t}H_{t}X_{t}^{T}W - \eta X_{b}H_{b}X_{b}^{T}W - \Lambda W = 0}\\ & {(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})W = \Lambda W} \end{array} \quad (12).\] + +Thus, the transformation matrix that maximizes the objective function \(\mathcal{O}_{3}\) can be obtained by solving the eigenvalue problem (Eq. 12). Let \(w_{1},w_{2},\dots,w_{p}\) be the eigenvectors from Eq. (12) corresponding to the top \(p\) largest eigenvalues \(\lambda_{1}\geq \lambda_{2}\geq \dots ,\geq \lambda_{p}\) . The lower dimensional manifold representation can then be obtained as follows: + +\[x_{t}^{t}\rightarrow z_{t}^{t} = W^{T}x_{t}^{t} \quad (13),\] + +\[W^{T} = [w_{1},w_{2},\dots,w_{p}]\] + +<--- Page Split ---> + +where \(z_{t}^{t}\) is a \(p\) - dimensional representation of \(x_{t}^{t}\) , and \(W\) is a \(d \times p\) matrix. This feature representation preserves the local structure of the ST data sets. A step- by- step description of the proposed CoCo- ST method is summarized in Algorithm 1. + +## Algorithm 1. CoCo-ST. + +Input: Background \(X_{b} = \left[x_{1}^{b},x_{2}^{b},\dots,x_{n_{b}}^{b}\right]^{T}\) and target \(X_{t} = \left[x_{1}^{t},x_{2}^{t},\dots,x_{n_{t}}^{t}\right]^{T}\) ST data sets, together with corresponding spatial locations \(Y_{b} = \left[y_{1}^{b},y_{2}^{b},\dots,y_{n_{b}}^{b}\right]^{T}\) and \(Y_{t} =\) \(\left[y_{1}^{t},y_{2}^{t},\dots,y_{n_{t}}^{t}\right]^{T}\) , the number of nearest neighbors \((k)\) , and the hyperparameters \(\mu_{1},\mu_{2}\) and \(\eta\) . + +Output: The low- dimensional contrastive feature representations for the target ST data \(Z_{t} =\) \(W^{T}X_{t}\) + +1. Construct the adjacency matrix for both the background and target ST data sets according to Eq. (1) or (2). + +2. Construct the normalized graph Laplacian matrices \(\bar{L}^{b}\) and \(\bar{L}^{t}\) together with the graph Laplacian filters \(H_{b} = I - \mu_{1}\bar{L}^{b}\) and \(H_{t} = I - \mu_{2}\bar{L}^{t}\) . + +3. Compute the matrices \(X_{b}H_{b}X_{b}^{T}\) and \(X_{t}H_{t}X_{t}^{T}\) . + +4. Solve the eigenvalue problem in Eq. (12). + +5. Compute the low-dimensional contrastive feature representations for the target ST data as \(Z_{t} = W^{T}X_{t}\) . + +We next investigate the computational complexity of the proposed CoCo- ST algorithm. Its complexity is dominated mainly by three parts: local similarity graph construction, matrix + +<--- Page Split ---> + +multiplication, and solving an eigenvalue problem. Assuming we have \(n_{b}\) and \(n_{t}\) spots in \(d\) - dimensional spaces ( \(d\) gene expression measurements) for the background and target ST data sets, to construct the similarity graphs, we first perform a \(k\) - nearest neighbor search for both data sets. The distance between any two spots in the background ST data can be computed in \(O(dn_{b}^{2})\) , and the \(k\) - nearest neighbors can be found with \(O(kn_{b}^{2})\) . Thus, the \(k\) - nearest neighbor search for the background and target ST data sets has complexities \(O\big((d + k)n_{b}^{2}\big)\) and \(O\big((d + k)n_{t}^{2}\big)\) , respectively. The complexities for computing the matrices \(X_{b}H_{b}X_{b}^{T}\) and \(X_{t}H_{t}X_{t}^{T}\) are \(O\big((n_{b}^{2} + n_{b}d)d\big)\) and \(O\big((n_{t}^{2} + n_{t}d)d\big)\) , respectively. The last part is computing the eigenvectors corresponding to the top \(p\) eigenvalues of the eigenproblem in Eq. (12), whose complexity is \(O(pd^{2})\) . Therefore, the time complexity of the CoCo- ST algorithm is \(O\big((d + k)(n_{b}^{2} + n_{t}^{2}) + \big((n_{b} + d)n_{b} + (n_{t} + d)n_{t} + pd\big)d\big)\) . Because \(k\ll n_{b}(\mathrm{or}n_{t})\) and \(p\ll d\) , the overall complexity of CoCo- ST is determined by the number of spots \(n_{b}(\mathrm{or}n_{t})\) and the number of genes \((d)\) . + +Several aspects of the proposed CoCo- ST approach are worth highlighting. Specifically: + +1. If \(\mu_{1} = \mu_{2} = 0\) , the matrices \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) reduce to \(X_{t}X_{t}^{T}\) and \(X_{b}X_{b}^{T}\) , respectively, so the objective function \((O_{3})\) reduces to that of contrastive PCA (cPCA) \(^{22}\) . Therefore, cPCA can be regarded as a variant of CoCo- ST. + +2. Whereas cPCA and the majority of the traditional feature representation approaches focus on global geometrical structures, CoCo-ST can exploit the intrinsic geometric structures of ST data sets and incorporate them as additional regularization terms. Through construction of a graph to model local geometric structures, CoCo-ST can have more discriminating power than cPCA and the traditional feature representation approaches. + +<--- Page Split ---> + +3. CoCo-ST simultaneously learns both global and local-level representations to complement tissue-wide representations, enabling it to distinguish different spatial areas in an ST tissue slice. + +4. The graphs in our proposed CoCo-ST approach are solely unsupervised and constructed from molecular data or spatial location information. Other information, such as label information, can also be used to guide graph construction, leading to other extensions of CoCo-ST such as supervised or semisupervised CoCo-ST. + +5. The proposed CoCo-ST approach differs from existing graph contrastive learning approaches that focus on graph neural network architectures for graph structured data. CoCo-ST considers the gene expression data and tries to learn local representations to better capture ST data structural information. As such, the objective functions of CoCo-ST and the conventional graph neural networks are different. + +## Why is CoCo-ST good for ST data analysis? + +CoCo-ST imposes molecularly or spatially similar spots to have similar feature representations, by which the intrinsic geometric structure of the ST data tends to be preserved. This is a useful property in ST data analysis because interesting spatial structures will not be lost owing to feature representation. In addition, CoCo-ST determines its discriminant (contrastive) feature representations from both the background and target ST data sets and thus can provide even more discriminative feature representations than the traditional approaches that focus only on a single ST data set. To explain this, we provided the following remarks and theorem. + +## Remark 1 + +<--- Page Split ---> + +When \(\eta = 0\) , CoCo- ST degenerates to a feature representation method that determines its discriminant vectors from the range space of the matrix \(X_{t}H_{t}X_{t}^{T}\) associated with the target data alone. When \(\eta >0\) , the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) is not guaranteed to be positive semidefinite even though \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are both symmetric and positive semidefinite. Let \(w\) be the eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda < 0\) . We then have + +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] + +\[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w + \lambda\] + +\[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w} = \eta +\frac{\lambda}{w X_{b}H_{b}X_{b}^{T}w}\] + +Because both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are positive semidefinite, we can conclude that + +\[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w} = \eta +\frac{\lambda}{w X_{b}H_{b}X_{b}^{T}w}\geq 0\] + +Thus, the eigenvectors corresponding to the negative eigenvalues are derived from the range space of \(X_{b}H_{b}X_{b}^{T}\) and contain some discriminant information. + +## Theorem 1 + +Suppose the matrix \(X_{b}H_{b}X_{b}^{T}\) is singular and that \(w\) is an eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda >0\) . The eigenvector \(w\) is then in the null space of \(X_{b}H_{b}X_{b}^{T}\) when \(\eta \rightarrow \infty\) . + +Proof. Because \(w\) is the eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda >0\) , we have + +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] + +<--- Page Split ---> + +\[w X_{b}H_{b}X_{b}^{T}w = \frac{1}{\eta} (w X_{t}H_{t}X_{t}^{T}w - \lambda)\] + +Since \(\lambda > 0\) , we have the following: + +\[w X_{b}H_{b}X_{b}^{T}w< \frac{1}{\eta} w X_{t}H_{t}X_{t}^{T}w\] + +Of note is that both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are positive semidefinite (i.e., \(w X_{t}H_{t}X_{t}^{T}w\geq 0\) and \(w X_{b}H_{b}X_{b}^{T}w\geq 0\) ). As a result, we have + +\[\lim_{\eta \to \infty}w X_{b}H_{b}X_{b}^{T}w = 0\] + +Thus, as \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues belong to the null space of \(X_{b}H_{b}X_{b}^{T}\) . + +## Remark 2 + +As \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues of the eigenproblem (Eq. [12]) contain the most discriminant information. We can rewrite the eigenvalue problem (Eq. [12]) as + +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] \[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w + \lambda\] \[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w}\to \infty\] + +Thus, as \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues contain the most discriminant information. + +## Remark 3 + +<--- Page Split ---> + +As \(\eta \to \infty\) , the eigenvectors corresponding to the zero eigenvalues of the eigenproblem (Eq. [12]) contain no discriminant information. When \(\lambda = 0\) , the eigenvalue problem reduces to + +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w = 0\] + +\[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w\] + +Since \(w X_{t}H_{t}X_{t}^{T}w\) and \(w X_{b}H_{b}X_{b}^{T}w\) are finite and \(\eta \to \infty\) , we have + +\[w X_{t}H_{t}X_{t}^{T}w = 0,\qquad w X_{b}H_{b}X_{b}^{T}w = 0\] + +Thus, the eigenvectors corresponding to the zero eigenvalues contain no discriminant information, as \(\eta \to \infty\) . In general, we can conclude that CoCo- ST derives its discriminant feature vectors from the range spaces of both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) . The parameter \(\eta\) can be used to balance the contribution from the two spaces. Moreover, by extracting the eigenvectors of the eigenvalue problem in Eq. (12) corresponding to the largest positive eigenvalues, CoCo- ST can capture the most discriminant information in both the background and target ST data sets, enabling effective identification of the interesting spatial structures enriched in the target ST data set. + +## Nonlinear extension of CoCo-ST + +Thus far, we have focused on linear feature representation. However, biological data are well known to be complex and highly nonlinear. Therefore, we extended CoCo- ST to perform nonlinear feature representation in a reproducing kernel Hilbert space \(\mathcal{H}\) , which gives rise to nonlinear CoCo- ST. We considered nonlinear mapping \(\phi (\cdot)\) of both the background \(X_{b}\) and target \(X_{t}\) ST data sets from the original input spaces to \(\mathcal{H}\) . Let \(\Phi_{b}\) and \(\Phi_{t}\) denote the background and target ST data sets in \(\mathcal{H}\) : + +\[\Phi_{b} = \left[\phi (x_{1}^{b}),\phi (x_{2}^{b}),\dots ,\phi (x_{n_{b}}^{b})\right]^{T}\] + +<--- Page Split ---> + +\[\Phi_{t} = \left[\phi (x_{1}^{t}),\phi (x_{2}^{t}),\dots ,\phi (x_{n_{t}}^{t})\right]^{T}\] + +Denote by \(V\) the projection matrix in \(\mathcal{H}\) . The corresponding objective function ( \(\mathcal{O}_{3}\) ) of CoCo- ST in \(\mathcal{H}\) is + +\[\mathcal{O}_{4} = \max_{V^{T}V = I}tr(V^{T}\Phi_{t}H_{t}\Phi_{t}^{T}V) - \eta tr(V^{T}\Phi_{b}H_{b}\Phi_{b}^{T}V)\] + +(14). + +Let \(N = n_{b} + n_{t}\) , and define the data \(q_{1},q_{2},\dots,q_{N}\) by + +\[q_{i} = \left\{ \begin{array}{l l}{x_{i}^{t},} & {i f 1\leq i\leq n_{t}}\] \[x_{i - n_{t}}^{b},} & {o t h e r w i s e} \end{array} \right.\] + +Since the projection vectors \(\nu_{1},\nu_{2},\dots,\nu_{p}\) (column vectors in \(V\) ) are linear combinations of \(\phi (q_{1}),\phi (q_{2}),\dots,\phi (q_{N})\) , coefficients \(\alpha_{i},i = 1,2,\dots,N\) exist such that + +\[\nu_{k} = \sum_{i = 1}^{N}\alpha_{i}\phi (q_{i}) = \Phi_{c}\alpha\] \[\Rightarrow V = \Phi_{c}A\] + +where \(\alpha = (\alpha_{1},\alpha_{2},\dots,\alpha_{N})^{T}\in R^{N}\) , \(\mathrm{A} = [\alpha^{1},\alpha^{2},\dots,\alpha^{p}]\) . Following some algebraic formulations, we can rewrite the objective function ( \(\mathcal{O}_{4}\) ) in the following equivalent form: + +\[\mathcal{O}_{4} = \max_{\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{c}\mathrm{A} = I}tr(\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{t}H_{t}\Phi_{t}^{T}\Phi_{c}\mathrm{A}) - \eta tr(\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{b}H_{t}\Phi_{b}^{T}\Phi_{c}\mathrm{A})\] \[\qquad = \max_{\mathrm{A}^{T}K_{c c}\mathrm{A} = I}tr(\mathrm{A}^{T}K_{c t}H_{t}K_{t c}\mathrm{A}) - \eta tr(\mathrm{A}^{T}K_{c b}H_{b}K_{b c}\mathrm{A})\] + +(15), + +where \(K_{c c} = \Phi_{c}^{T}\Phi_{c}\) , \(K_{c t} = \Phi_{c}^{T}\Phi_{t}\) , \(K_{t c} = \Phi_{t}^{T}\Phi_{c}\) , \(K_{c b} = \Phi_{c}^{T}\Phi_{b}\) , and \(K_{b c} = \Phi_{b}^{T}\Phi_{c}\) are the kernel matrices. Several choices of the kernel functions are available, including the polynomial kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \big((x_{t}^{t})^{T}x_{t}^{b} + 1\big)^{d}\) ; Gaussian kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \exp \left(-\frac{\left\|x_{t}^{t} - x_{t}^{b}\right\|^{2}}{\sigma^{2}}\right)\) ; and sigmoid kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \big((x_{t}^{t})^{T}x_{t}^{b} + \gamma \big)\) . + +<--- Page Split ---> + +Following approach similar to that in linear CoCo- ST, the projection vectors in Eq. (15) can be obtained as the eigenvectors corresponding to the top \(p\) largest eigenvalues of the generalized eigenvalue problem + +\[(K_{c t}H_{t}K_{t c} - \eta K_{c b}H_{b}K_{b c})\mathrm{A} = \Lambda K_{c c}\mathrm{A}\] + +To obtain a stable solution of the eigenvalue problem in Eq. (16), the kernel matrix \(K_{cc}\) must be nonsingular. When \(K_{cc}\) is singular, we can adopt the idea of regularization by adding a small constant value \(\rho\) to the diagonal of \(K_{cc}\) as \(K_{cc} + \rho I\) for any \(\rho > 0\) . The matrix \(K_{cc} + \rho I\) is nonsingular, and the projection vectors can be computed as the generalized eigenvectors of + +\[(K_{c t}H_{t}K_{t c} - \eta K_{c b}H_{b}K_{b c})\mathrm{A} = \Lambda (K_{c c} + \rho I)\mathrm{A}\] \[(17).\] + +## Animal model + +Wild- type mice (strain #009104; \(n = 12,9\mathrm{S4}\) ) were purchased from The Jackson Laboratory and housed in colony cages under pathogen- free conditions at The University of Texas MD Anderson Cancer Center Research Animal Support Facility. The mice were housed at an ambient temperature of \(20 - 26^{\circ}\mathrm{C}\) and humidity range of \(30 - 70\%\) with a 12- h light- dark cycle. All animal experiments were conducted following MD Anderson Institutional Animal Care and Use Committee- approved protocols (approval number 00001217- RN03). For carcinogen- induced mouse models, a urethane- induced mouse model was used. Specifically, the 12,9S4 wild- type mice described above received intraperitoneal injections of \(1\mathrm{mg / g}\) (body weight) urethane three times over 8 days when they were 6 weeks old. The mice were killed 7, 14, 20, 30, and 40 weeks after + +<--- Page Split ---> + +urethane administration, with a 0- week time point for mice that received no treatment. Both normal lung and lung tumor tissue samples were collected from the mice for downstream analysis. + +## Single-cell sequencing and analysis + +Fresh normal lung and lung tumor tissue samples collected from mice were immediately cut into pieces and placed in RPMI 1640 medium (Thermo Fisher Scientific) with \(10\%\) fetal bovine serum (FBS; Gibco). The tissue samples were enzymatically digested using a tumor dissociation mixture composed of \(1\mathrm{mg / ml}\) collagenase A (Sigma), \(0.4\mathrm{mg / ml}\) hyaluronidase (Sigma), and 1:5 bovine serum albumin fraction V (Thermo Fisher Scientific) according to the manufacturers' instructions. Dissociation of tissue was carried out for \(2\mathrm{h}\) on a rotary shaker at \(37^{\circ}\mathrm{C}\) until all large tissue fragments were digested. Next, the dissociated tissues were transferred to conical tube and centrifuged at \(350\mathrm{g}\) for \(5\mathrm{min}\) . The supernatant was removed, and 1- 5 ml of prewarmed trypsin- EDTA was added to the collagenase/hyaluronidase- dissociated cells, resuspending them. Subsequently, \(10\mathrm{ml}\) of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS was added and centrifuged at \(350\mathrm{g}\) for \(5\mathrm{min}\) . As much of the supernatant as possible was collected, and \(5\mathrm{ml}\) of prewarmed \(5\mathrm{U / ml}\) dispase (STEMCELL Technologies) and \(50\mu \mathrm{l}\) of DNase I solution (10 \(\mathrm{mg / ml}\) in \(0.15\mathrm{M}\) NaCl; STEMCELL Technologies) were added. The samples were pipetted for 1 min using a 1- ml micropipettor to further dissociate cell clumps. The cell suspension was diluted with an additional \(10\mathrm{ml}\) of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS, and the cell suspension was filtered through a \(40\mathrm{- }\mu \mathrm{m}\) Falcon cell strainer (Thermo Fisher Scientific) into a \(50\mathrm{- }\mathrm{ml}\) tube. The cell suspension was further centrifuged at \(450\mathrm{g}\) for \(5\mathrm{min}\) , and the supernatant was discarded. The pellet was resuspended in a 1:4 mixture of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS and an ammonium chloride solution (STEMCELL + +<--- Page Split ---> + +Technologies), which was followed by centrifugation at \(450g\) for 5 min and discarding of the supernatant. Ten microliters of the cell suspension for each sample was analyzed using an automated cell counter (Thermo Fisher Scientific) to determine the number of live cells. Throughout the dissociation procedure, cells were kept on ice when possible. The cells were then loaded onto a Chromium single- cell controller (10x Genomics) to create single- cell gel beads in an emulsion according to the manufacturer's protocol. ScRNA- seq libraries were constructed using a Single Cell 5' Library and Gel Bead Kit v3.1 (10x Genomics) and sequenced using a NovaSeq 6000 sequencer (Illumina) at the Genomic and RNA Profiling Core at Baylor College of Medicine. + +## Tissue preparation and ST + +Normal and tumor tissue samples from mouse lungs were fixed in \(10\%\) formalin at room temperature for 24- 48 h using a fixative volume 5- 10 times greater than that of the tissue volume. Fixed tissues were transferred to \(70\%\) ethanol for temporary storage at \(4^{\circ}\mathrm{C}\) . Paraffin embedding was conducted by the MD Anderson Research Histology Core Laboratory. Formalin- fixed, paraffin- embedded blocks were cut into \(10 - \mu \mathrm{m}\) - thick sections using a precooled RNase- free microtome. These sections were then transferred onto Visium Spatial Gene Expression slides (10x Genomics), which were pretreated via floating in a water bath at \(43^{\circ}\mathrm{C}\) . Following sectioning, the slides were dried at \(42^{\circ}\mathrm{C}\) in a SimpliAmp Thermal Cycler (Thermo Fisher Scientific) for \(3\mathrm{~h}\) according to the manufacturer's instructions. The slides were placed in a slide mailer, sealed with thermoplastic (Parafilm: Thermo Fisher Scientific), and stored overnight in a refrigerator at \(4^{\circ}\mathrm{C}\) . The slides were then deparaffinized, fixed, stained with hematoxylin and eosin, and imaged at \(5\mathrm{x}\) magnification using a DM5500 B microscope (Leica Microsystems). Tile scans of the entire array + +<--- Page Split ---> + +were acquired using Leica Application Suite X software and merged. Spatial gene expression libraries (Visium ST; 10x Genomics) were processed according to the manufacturer's instructions and sequenced using a NovaSeq 6000 sequencer (Illumina). All hematoxylin and eosin staining, imaging, library preparation, and sequencing processes were carried out at the Genomic and RNA Profiling Core at Baylor College of Medicine. + +## Data processing + +ScRNA- seq data. Raw base call files were analyzed using Cell Ranger v.3.0.2 software (10x Genomics). The mkfastq command was used to generate FASTQ files, and the count command was used to generate raw gene- barcode matrices aligned to the GRCh38 Ensembl 93 genome. The data were aggregated using the cellranger aggr command, and further downstream analysis was conducted in R version 4.1.0 using the Seurat package (v.4.1.1). To ensure our analysis was performed using high- quality cells, filtering of cells was conducted by retaining cells that had unique feature counts greater than 200 or less than 5000 and had mitochondrial content less than \(15\%\) . After removing doublets, the total cell number was 70,698. + +ST data. The ST data sets were processed using Space Ranger software (v.2.0.1; 10x Genomics). The spatial sequencing data were aligned to mouse pre- mRNA genome reference version mm10 (downloaded from the 10x Genomics website) using Space Ranger, and mRNA count matrices were generated by adding intronic and exonic reads for each gene in each location. Paired histological hematoxylin and eosin stained images of tissues were processed using Space Ranger to select locations covered by tissue by aligning prerecorded spot locations with fiducial border spots in the images. + +<--- Page Split ---> + +## Data analysis + +ScRNA- seq analysis. The scRNA- seq data were first normalized, and the 2000 most highly variable genes in the data were identified using variance- stabilizing transformation implemented in the Seurat package. Data were then scaled, and the first 30 principal components were extracted. The principal components were further transformed into the UMAP embedding space for which clustering analysis was conducted. The original Louvain algorithm was used for modularity optimization. The resulting 14 clusters were visualized in a 2D UMAP representation and annotated to known biological cell types using canonical marker genes. The following cell types were annotated (selected markers are listed in parentheses): endothelial cells (Pecam1, Vwf, Ets1, Ace, Eng, Cldn5, and Mcam), epithelial cells (Epcam, Muc1, Cdh1, Krt7, and Krt8), fibroblasts (Pdpn, Dcn, Col3a1, Mgp, Col1a1, and Col6a1), macrophages (Apoe, C1qa, C1qb, C1qc, Marco, Mrc1, Fabp4, Inhba, Ccl4, Cxcl10, Rsad2, and Herc6), conventional dendritic cells (cDC; H2-Aa, Ccr7, Flt3, Fscn1, and Cdec9a), proliferating macrophages (Mki67, Tubb5, and Tuba1b), B cells (Cd19, Ms4a1, Cd79a, Cd79b, and Blnk), T cells (Trbc2, Cd2, Cd3d, Cd3e, Cd3g, Cd4, Cd8a, Cd8b1, Il2ra, and Foxp3), proliferating T cells (Mki67, Tubb5, and Tuba1b), plasmacytoid dendritic cells (pDC; Siglech, Ly6c2, and Cd209d), neutrophils (S100a8, S100a9, and Csf3r), plasma cells (Sdc1, Mzb1, Xbp1, and Jchain), monocytes (Cd14, Fcgr4, Lst1, and Vcan), and natural killer cells (Nkg7, Klrg1, and Ncr1). + +ST analysis. The raw expression count matrices for both the background and target ST data sets were normalized using variance- stabilizing transformation implemented in the Seurat package. The normalized data were then standardized to have zero mean and unit standard deviation. The + +<--- Page Split ---> + +standardized expression data matrices with 3000 genes were then used as inputs to our CoCo- ST method for low- dimensional feature representation. Clustering on the UMAP- embedded learned contrastive feature representations was then performed. Further differential gene expression analysis was conducted, and spatial domains were annotated based on the differentially expressed marker genes. + +## Pathway analysis + +The most important genes (the 20 genes with the largest weights) on the top five contrastive components were identified, and the biological processes associated with these contrastive components were examined. Specifically, gene set enrichment analysis was performed with these 20 genes with the largest weights in the loading matrix using the g:GOSf function in the gprofiler2 package. In this analysis, all of the input 3000 genes were used as the background, and the default options in the g:SCS method in gprofiler2 were used for multiple testing correction. The gene sets were downloaded from the Molecular Signatures Database, including the KEGG, Gene Ontology biological processes, Gene Ontology cellular components, and Gene Ontology molecular functions. + +## Cell type deconvolution + +Cell type deconvolution in ST enables estimation of cell type composition on each spatial location by leveraging a reference scRNA- seq data set. Cell type deconvolution was performed using the RCTD \(^{23}\) method implemented in the spacexr R package. ScRNA- seq data for the same mouse lung tumor samples (MLP samples) served as the reference data for deconvolution. The reference data contained 70,698 cells of multiple immune and malignant types as described in the scRNA- seq + +<--- Page Split ---> + +analysis section. The RCTD method was run in doublet mode to estimate the reference cell type composition on each spatial location. Other parameters were set to the default settings. + +## Cell-cell interaction + +Cell- cell interaction for the ST data sets was performed using CellChat24. The CellChatDB.mouse database of ligand- receptor interactions specifically curated for mice was used to identify overexpressed ligand- receptor interactions. The group- level communication probability or interaction weights were then computed using the truncated mean method with a \(10\%\) truncated mean. Subsequently, the communication probability at the signaling pathway level was computed by summarizing the communication probabilities of all ligand- receptor interactions associated with each signaling pathway. Finally, the cell- cell communication network was aggregated by summarizing the overall communication probabilities. + +## Trajectory inference analysis + +For spatial trajectory analysis of individual tissue samples, the low- dimensional contrastive feature representations were used as inputs to the Slingshot algorithm25. Slingshot was applied to the contrastive feature representations so that nearby tissue spatial locations with similar gene expression would have similar pseudotimes. Because Slingshot requires predefined cluster labels, the spatial domain labels from the spatial domain identification analysis were used for Slingshot. The normal lung spatial domain was set as the start cluster (beginning of the trajectory or pseudotime) with a focus on trajectory inference on tumor and tumor- adjacent spatial domains to determine how these locations are connected to one another during tumorigenesis. + +<--- Page Split ---> + +For the trajectory analysis with combined tissue samples, spots belonging to normal lung, adenoma, and adenocarcinoma spatial domains as determined using the contrastive feature representations were collected, and Monocle \(^{36}\) was used to infer the trajectory. First, the combined data (spots) were processed using the standard Seurat approach, including total count normalization, scaling, and PCA analysis. Next, UMAP embedding was determined, which was used to learn the trajectory that fits the spots' UMAP coordinates. A principal graph was then fit on the UMAP embedding, and the spots were ordered according to their progress along the learned trajectory. To identify genes that varies among spot clusters in the UMAP embedding space, spatial autocorrelation analysis (Moran's I) was performed, and the obtained variable genes were grouped into modules by determining UMAP embedding of the genes followed by gene clustering based on Louvain community detection analysis. + +<--- Page Split ---> + +## Data availability + +Data availabilityThe scRNA- seq and ST data sets analyzed in this study will be made available upon reasonable request through a data access agreement with the corresponding authors. + +## Code availability + +Code availabilityInstallation instructions and tutorials, together with the code used for data analysis and generating figures, can be found at https://github.com/WuLabMDA/CoCo- ST. + +## Acknowledgements + +AcknowledgementsThis work was supported by generous philanthropic contributions to the MD Anderson Lung Cancer Moon Shot program as well as by the NIH/NCI under award number P30CA016672. This research was partially supported by NIH grants R01CA262425 and R01CA276178. Furthermore, this work was supported by generous philanthropic contributions from Andrea Mugnaini and Edward L. C. Smith. Finally, this work was supported by Rexanna's Foundation for Fighting Lung Cancer. We thank Don Norwood in the Research Medical Library at The University of Texas MD Anderson Cancer Center for editing this article. + +## Author contributions + +Author contributionsM.A. and J.W. formulated and applied the method. B.Z. and J.Z. acquired the data. M.A. developed the software. M.A., B.Z., N.V., C.C., K.C., J.Z. and J.W. design the experiments. M.A., B.Z., H.C, N.V. and L.H. analyzed the data. All authors contributed to the interpretation of the data. M.A., B.Z. and H.C. prepared the first draft of the manuscript. L.H., N.V., C.C., K.C., J.Z. and J.W. revised the manuscript. J.Z. and J.W. supervised the project. All authors read and approved the + +<--- Page Split ---> + +final version of the manuscript. All authors were responsible for the final decision to submit the manuscript for publication. + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> + +## References + +1 Zhao, E. et al. Spatial transcriptomics at subspot resolution with BayesSpace. Nature biotechnology 39, 1375- 1384 (2021). 2 Bergensträhle, J., Larsson, L. & Lundeberg, J. Seamless integration of image and molecular analysis for spatial transcriptomics workflows. BMC Genomics 21, 482, doi:10.1186/s12864- 020- 06832- 3 (2020). 3 Townes, F. W. & Engelhardt, B. E. Nonnegative spatial factorization applied to spatial genomics. Nature Methods 20, 229- 238, doi:10.1038/s41592- 022- 01687- w (2023). 4 Shang, L. & Zhou, X. Spatially aware dimension reduction for spatial transcriptomics. Nature Communications 13, 7203, doi:10.1038/s41467- 022- 34879- 1 (2022). 5 Velten, B. et al. Identifying temporal and spatial patterns of variation from multimodal data using MEFISTO. Nature methods 19, 179- 186 (2022). 6 Chen, T., Kornblith, S., Norouzi, M. & Hinton, G. in International conference on machine learning. 1597- 1607 (PMLR). 7 You, Y. et al. Graph contrastive learning with augmentations. Advances in neural information processing systems 33, 5812- 5823 (2020). 8 Wang, Y., Wang, J., Cao, Z. & Barati Farimani, A. Molecular contrastive learning of representations via graph neural networks. Nature Machine Intelligence 4, 279- 287 (2022). 9 Dang, H. et al. Cancer- associated fibroblasts are key determinants of cancer cell invasion in the earliest stage of colorectal cancer. Cellular and Molecular Gastroenterology and Hepatology 16, 107- 131 (2023). 10 Hu, J. et al. SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nature methods 18, 1342- 1351 (2021). 11 Yi, F., Jaffe, R. & Prochownik, E. V. The CCL6 chemokine is differentially regulated by c- Myc and L- Myc, and promotes tumorigenesis and metastasis. Cancer research 63, 2923- 2932 (2003). 12 Tigue, M. L. et al. Wnt signaling in the phenotype and function of tumor- associated macrophages. Cancer Research 83, 3- 11 (2023). 13 Schmoll, A. et al. Macrophage and cancer cell cross- talk via CCR2 and CX3CR1 is a fundamental mechanism driving lung cancer. American journal of respiratory and critical care medicine 191, 437- 447 (2015). 14 Garrido- Navas, C. et al. Cooperative and escaping mechanisms between circulating tumor cells and blood constituents. Cells 8, 1382 (2019). 15 Sarode, P., Schaefer, M. B., Grimminger, F., Seeger, W. & Savai, R. Macrophage and tumor cell cross- talk is fundamental for lung tumor progression: we need to talk. Frontiers in Oncology 10, 324 (2020). 16 Ge, Z. & Ding, S. The crosstalk between tumor- associated macrophages (TAMs) and tumor cells and the corresponding targeted therapy. Frontiers in oncology 10, 590941 (2020). 17 Allen Reference Atlas - Mouse Brain [brain atlas] 18 Wang, W. et al. in Proceedings of the IEEE/CVF International Conference on Computer Vision. 7303- 7313. 19 Miao, J., Yang, Z., Fan, L. & Yang, Y. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8042- 8052. 20 Hadsell, R., Chopra, S. & LeCun, Y. in 2006 IEEE computer society conference on computer vision and pattern recognition (CVPR'06). 1735- 1742 (IEEE). 21 Liu, Y. et al. Simple contrastive graph clustering. IEEE Transactions on Neural Networks and Learning Systems (2023). + +<--- Page Split ---> + +22 Abid, A., Zhang, M. J., Bagaria, V. K. & Zou, J. Exploring patterns enriched in a dataset with contrastive principal component analysis. \*Nature communications\* 9, 2134 (2018).23 Cable, D. M. et al. Robust decomposition of cell type mixtures in spatial transcriptomics. \*Nature biotechnology\* 40, 517- 526 (2022).24 Jin, S. et al. Inference and analysis of cell- cell communication using CellChat. \*Nature communications\* 12, 1088 (2021).25 Street, K. et al. Slingshot: cell lineage and pseudotime inference for single- cell transcriptomics. \*BMC genomics\* 19, 1- 16 (2018).26 Cao, J. et al. The single- cell transcriptional landscape of mammalian organogenesis. \*Nature\* 566, 496- 502 (2019). + +<--- Page Split ---> + +## Figure Legends + +Fig. 1 | CoCo- ST identifies unique, interesting spatial structures enriched in ST data sets. a, Overview of the CoCo- ST workflow. b, A target ST tissue sample containing unique, interesting spatial structures annotated by a pathologist and spatial domains/regions identified using the different feature representation methods. c, Volcano plot of the most differentially expressed genes for the adenoma spatial domain identified by CoCo- ST. d, Volcano plot of the most differentially expressed genes for the adenoma spatial domain identified using the compared approaches. e, Spatial expression patterns for the most differentially expressed genes (Ctsh, Cxcl15, and Slc34a2) for the adenoma spatial domain identified using CoCo- ST. These genes had high expression patterns in both the larger and smaller (hotspot) adenoma spatial domains. f, Spatial expression pattern for the most differentially expressed gene (Trf) for the adenoma spatial domain identified using the compared approaches. This gene had high expression pattern only within the larger adenoma spatial domain, with no such pattern observed in the smaller (hotspot) region. + +Fig. 2 | CoCo- ST's contrastive components marked interesting spatial structures enriched in ST data sets. a, Spatial patterns captured by the first five contrastive components of CoCo- ST. b, The top 20 genes with the largest weights on the corresponding first five contrastive components. Symbols to the right of the bars indicate the signs of the weights. c, Expression patterns for the top representative genes for each of the first five contrastive components. d, Spatial patterns captured by the first five components of the compared approaches. + +<--- Page Split ---> + +## Supplementary Information + +Supplementary InformationExtended Data Fig. 1 | Spatial domains identified on all MLP tissue samples using CoCo- ST's contrastive components. The similarity graphs for both the background and target ST data sets were constructed based on the molecular data sets. + +Extended Data Fig. 2 | Differential gene expression analysis of detected spatial domains. a, UMAP embedding of the contrastive components determined using CoCo- ST on the target ST tissue sample. B, UMAP embedding of spotsshowing the expression of some of the most differentially expressed genes in different clusters identified using the contrastive feature representations from CoCo- ST. c, Violin plots of the expression levels for the most differentially expressed genes for the different spatial domains identified using CoCo- ST. d, Biological processes and pathways associated with the 10 most differentially expressed genes for the adenoma spatial domain detected using CoCo- ST. e, Violin plots of the expression levels for the most differentially expressed genes for the different spatial domains identified using the compared feature representation approaches. + +Extended Data Fig. 3 | Biological processes and pathways associated with CoCo- ST's contrastive components. + +Extended Data Fig. 4 | Spatial domains identified on all MLP tissue samples using CoCo- ST's contrastive components. The similarity graphs for both the background and target ST data sets were constructed based on spatial locations. + +<--- Page Split ---> + +Extended Data Fig. 5 | Application of CoCo- ST's contrastive components to studying cell- cell interaction at different cancer stages. a, UMAP embedding of the scRNA- seq data set used as a reference for cell type deconvolution. b, Spatial domains identified in the MLP- 6 tissue sample using CoCo- ST's contrastive components. c, Cell type annotation on each of the spatial locations in MLP- 6 tissue sample as inferred by the RCTD deconvolution algorithm. d, Percentage of different cell types (y- axis) in the different spatial domains (x- axis) detected using CoCo- ST. e, Cell- cell interaction weight plot for MLP- 6 tissue sample. The thicker the line, the stronger the interaction between the cell types. f, Chord plot of the cell- cell interactions via canonical WNT signaling. g, Heat map of the communication probabilities for WNT signaling from senders (sources) to receivers (targets). h, Heat map of network centrality scores for WNT signaling highlighting the major signaling roles of the different cell groups. + +Extended Data Fig. 6 | Predicted spatial distributions of major cell types in the MLP- 6 tissue sample. + +Extended Data Fig. 7 | Distribution of different cell types in each spatial domain on the MLP- 6 tissue sample determined using CoCo- ST. The cell type percentages in each spatial domain add up to \(100\%\) . + +Extended Data Fig. 8 | Aggregated cell- cell interaction plots on the combined MLP tissue samples containing the adenoma and adenocarcinoma spatial domains. a, Cell- cell interaction weight plot for the adenocarcinoma- related MLP tissue samples. b, Simplified cell- cell interaction plots for a showing signaling sent from each cell group. The thicker the line, the + +<--- Page Split ---> + +stronger the communication. c, Cell- cell interaction weight plot for the adenoma- related MLP tissue samples. d, Simplified cell- cell interaction plots for c showing signaling sent from each cell group. The thicker the line, the stronger the communication. + +Extended Data Fig. 9 | Application of CoCo- ST's contrastive components to trajectory inference (cancer evolution). a, Spatial trajectory inference based on CoCo- ST's determined contrastive components. The arrows indicate the direction of the trajectory, which points from the normal lung spatial domain to the adenoma spatial domain. b, Learned trajectory pseudotime, with red- to green- colored regions indicating tissue locations with low and high pseudotime. c, UMAP embedding of spots belonging to the combined normal, adenoma, and adenocarcinoma spatial domains as determined using CoCo- ST. d, Trajectory inference of the cancer evolution from normal tissue to adenoma to adenocarcinoma colored according to their corresponding pseudotimes. e, Heat map of gene modules containing differentially co- expressed genes that vary across the different stages of cancer as determined from the learned trajectory in d. f, Bar plot of the number of differentially co- expressed genes in each module in e. + +Extended Data Fig. 10 | Application of CoCo- ST to a mouse brain ST data set. a, Spatial domains/regions identified on anterior and posterior mouse brain tissue samples based on CoCo- ST's contrastive components. b, Spatial patterns on the anterior mouse brain tissue sample captured by the first five contrastive components of CoCo- ST. c, The 20 genes with the largest weights on the first five contrastive components in b. Symbols to the right of the bars indicate the signs of the weights. d, Expression patterns for some representative genes in c. e, Spatial patterns on the posterior mouse brain tissue sample captured by the first five contrastive + +<--- Page Split ---> + +components of CoCo- ST. f, The 20 genes with the largest weights on the first five contrastive components in e. g, Expression patterns for some representative genes in f. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1
+ +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2
+ +<--- Page Split ---> +![](images/Extended_Data_Figure_2.jpg) + + +<--- Page Split ---> +![](images/Extended_Data_Figure_6.jpg) + + +![](images/Extended_Data_Figure_9.jpg) + +
Extended Data Fig. 2
+ +<--- Page Split ---> +![PLACEHOLDER_51_0] + + +<--- Page Split ---> +![PLACEHOLDER_52_0] + + +<--- Page Split ---> +![PLACEHOLDER_53_0] + + +<--- Page Split ---> +![PLACEHOLDER_54_0] + +
Extended Data Fig. 6
+ +<--- Page Split ---> +![PLACEHOLDER_55_0] + + +<--- Page Split ---> +![PLACEHOLDER_56_0] + + +<--- Page Split ---> +![PLACEHOLDER_57_0] + +
Extended Data Fig. 9
+ +<--- Page Split ---> +![PLACEHOLDER_58_0] + + +<--- Page Split ---> diff --git a/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224_det.mmd b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7aff2097ede6d4b82da186c7c5aa39172fc8bd06 --- /dev/null +++ b/preprint/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224/preprint__168a56549a3c8b925a9c954a2854adb571d828f70d83eca67ba3e041ee941224_det.mmd @@ -0,0 +1,815 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 919, 210]]<|/det|> +# CoCo-ST: Comparing and Contrasting Spatial Transcriptomics data sets using graph contrastive learning + +<|ref|>text<|/ref|><|det|>[[44, 230, 279, 276]]<|/det|> +Jia Wu Jku11@mdanderson.org + +<|ref|>text<|/ref|><|det|>[[44, 300, 875, 323]]<|/det|> +The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0001- 8392- 8338 + +<|ref|>text<|/ref|><|det|>[[44, 328, 875, 370]]<|/det|> +Muhammad Aminu The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 9903- 8812 + +<|ref|>text<|/ref|><|det|>[[44, 374, 512, 416]]<|/det|> +Bo Zhu The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 420, 512, 462]]<|/det|> +Natalie Vokes The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 466, 512, 508]]<|/det|> +Hong Chen The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 512, 512, 555]]<|/det|> +Lingzhi Hong The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 559, 270, 600]]<|/det|> +Jianrong Li Baylor College Medicine + +<|ref|>text<|/ref|><|det|>[[44, 605, 240, 645]]<|/det|> +Junya Fujimoto Hiroshima University + +<|ref|>text<|/ref|><|det|>[[44, 650, 512, 692]]<|/det|> +Alissa Poteete The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 697, 512, 739]]<|/det|> +Monique Nilsson The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 744, 512, 786]]<|/det|> +Xiuning Li The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 791, 344, 832]]<|/det|> +Tina Cascone UT M.D. Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 837, 512, 879]]<|/det|> +David Jaffray The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 884, 875, 926]]<|/det|> +Nicholas Navin The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 2106- 8624 + +<|ref|>text<|/ref|><|det|>[[44, 931, 160, 949]]<|/det|> +Lauren Byers + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[52, 45, 872, 67]]<|/det|> +The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0002- 0780- 2677 + +<|ref|>text<|/ref|><|det|>[[44, 70, 875, 113]]<|/det|> +Don Gibbons The University of Texas MD Anderson Cancer Center https://orcid.org/0000- 0003- 2362- 3094 + +<|ref|>text<|/ref|><|det|>[[44, 116, 666, 159]]<|/det|> +John Heymach MD Anderson Cancer Center https://orcid.org/0000- 0001- 9068- 8942 + +<|ref|>text<|/ref|><|det|>[[44, 163, 514, 206]]<|/det|> +Ken Chen The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 210, 650, 253]]<|/det|> +Chao Cheng Baylor College of Medicine https://orcid.org/0000- 0002- 5002- 3417 + +<|ref|>text<|/ref|><|det|>[[44, 256, 514, 299]]<|/det|> +Jianjun Zhang The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[44, 303, 300, 345]]<|/det|> +Yuqui Yang UT Southwestern University + +<|ref|>text<|/ref|><|det|>[[44, 350, 884, 393]]<|/det|> +Tao Wang The University of Texas Southwestern Medical Center https://orcid.org/0000- 0002- 4355- 149X + +<|ref|>text<|/ref|><|det|>[[44, 396, 240, 437]]<|/det|> +Bo Wang University of Toronto + +<|ref|>sub_title<|/ref|><|det|>[[44, 476, 98, 494]]<|/det|> +## Letter + +<|ref|>text<|/ref|><|det|>[[44, 514, 137, 533]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 552, 297, 572]]<|/det|> +Posted Date: May 20th, 2024 + +<|ref|>text<|/ref|><|det|>[[44, 590, 475, 610]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 4359834/v1 + +<|ref|>text<|/ref|><|det|>[[42, 627, 914, 671]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 688, 535, 709]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 743, 949, 788]]<|/det|> +Version of Record: A version of this preprint was published at Nature Cell Biology on October 13th, 2025. See the published version at https://doi.org/10.1038/s41556- 025- 01781- z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[114, 88, 883, 144]]<|/det|> +# CoCo-ST: Comparing and Contrasting Spatial Transcriptomics data sets using graph contrastive learning + +<|ref|>text<|/ref|><|det|>[[113, 191, 885, 320]]<|/det|> +Muhammad Aminu1,10, Bo Zhu2,10, Natalie Vokes2,10, Hong Chen2, Lingzhi Hong2, Jianrong Li9, Junya Fujimoto8, Yuqiu Yang12, Tao Wang12, Bo Wang13, Alissa Poteete2, Monique B. Nilsson2, Xiuning Le2, Cascone Tina2, David Jaffray3,7, Nick Navin5, Lauren A. Byers2, Don Gibbons2, John Heymach2, Ken Chen6, Chao Cheng9, Jianjun Zhang2,11 & Jia Wu1,2,7,11 + +<|ref|>text<|/ref|><|det|>[[113, 366, 886, 701]]<|/det|> +1Department of Imaging Physics, 2Department of Thoracic/Head and Neck Medical Oncology, 3Office of the Chief Technology and Digital Officer, 5Department of Systems Biology, 6Department of Bioinformatics and Computational Biology, 7Institute for Data Science in Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA. 8Clinical Research Center, Hiroshima University, Hiroshima, Japan. 9Department of Medicine, Institution of Clinical and Translational Research, Baylor College of Medicine, Houston, TX, USA. Department of Public Health, UT Southwestern Medical Center, Dallas, TX, USA. 13Department of Medical Biophysics, University of Toronto, Ontario, Canada. 10These authors contributed equally: Muhammad Aminu, Bo Zhu, Natalie Vokes. 11Co- senior authors: Jianjun Zhang, Jia Wu. + +<|ref|>sub_title<|/ref|><|det|>[[115, 751, 310, 770]]<|/det|> +## Corresponding Author + +<|ref|>text<|/ref|><|det|>[[114, 786, 585, 875]]<|/det|> +Jia Wu, PhDDepartment of Imaging PhysicsDepartment of Thoracic/Head and Neck Medical Oncology + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 90, 545, 108]]<|/det|> +The University of Texas MD Anderson Cancer Center + +<|ref|>text<|/ref|><|det|>[[116, 125, 288, 142]]<|/det|> +1515 Holcombe Blvd + +<|ref|>text<|/ref|><|det|>[[115, 159, 323, 177]]<|/det|> +Houston, TX 77030, USA + +<|ref|>text<|/ref|><|det|>[[115, 195, 323, 212]]<|/det|> +Telephone: 713- 563- 2719 + +<|ref|>text<|/ref|><|det|>[[115, 230, 373, 248]]<|/det|> +e- mail: jwu11@mdanderson.org + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 91, 190, 108]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[112, 121, 886, 565]]<|/det|> +Traditional feature dimension reduction methods have been widely used to uncover biological patterns or structures within individual spatial transcriptomics data. However, these methods are designed to yield feature representations that emphasize patterns or structures with dominant high variance, such as the normal tissue spatial pattern in a precancer setting. Consequently, they may inadvertently overlook patterns of interest that are potentially masked by these high- variance structures. Herein we present our graph contrastive feature representation method called CoCo- ST (Comparing and Contrasting Spatial Transcriptomics) to overcome this limitation. By incorporating a background data set representing normal tissue, this approach enhances the identification of interesting patterns in a target data set representing precancerous tissue. Simultaneously, it mitigates the influence of dominant common patterns shared by the background and target data sets. This enables discerning biologically relevant features crucial for capturing tissue- specific patterns, a capability we showcased through the analysis of serial mouse precancerous lung tissue samples. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 80, 886, 707]]<|/det|> +Analyzing spatial transcriptomics (ST) data requires robust feature representation methods to effectively capture the intricate biological information or patterns enriched in these high- dimensional data sets. Although traditional dimension reduction techniques like principal component analysis (PCA \(^{1}\) ) and nonnegative matrix factorization (NMF \(^{2}\) ) have been widely adopted as off- the- shelf approaches for ST data dimension reduction, they primarily aimed at capturing global patterns and variations in the original high- dimensional ST data sets. More recently, the integration of spatial constraints into dimension reduction algorithms has led to the emergence of robust feature representation approaches such as nonnegative spatial factorization \(^{3}\) , spatial PCA \(^{4}\) , and MEFISTO \(^{5}\) . However, these methods tend to prioritize the identification of prominent global patterns with high variability, potentially missing finer localized intrinsic structures marked by lower variability. Furthermore, they are designed to explore one data set at a time and are not tailored to studying the evolutionary dynamics of a tumor microenvironment across multiple data sets. These constraints can result in overlooked information, particularly when studying carcinogenesis, in which tumors progress from a few isolated precancerous sites to invasive cancer across various tissue samples. The majority of these samples exhibit common global patterns (representing normal tissue biology) that may not be of primary interest. Conversely, a small portion of samples contain unique, crucial precancerous structures that require specific attention. + +<|ref|>text<|/ref|><|det|>[[113, 715, 884, 842]]<|/det|> +To address these constraints, we proposed a graph contrastive learning framework that we called CoCo- ST (Compare and Contrast Spatial Transcriptomics). CoCo- ST operates by taking two ST data sets as inputs: one serving as the reference (background) and another as the target. These ST data sets typically have certain common structures that are usually not the primary foci. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 883, 144]]<|/det|> +The goal is to extract feature representations that emphasize the new and unique structures enriched in the target ST data set. + +<|ref|>text<|/ref|><|det|>[[114, 158, 884, 283]]<|/det|> +In the present study, we used CoCo- ST to thoroughly investigate carcinogenesis using ST data sets from an in- house curated carcinogenesis mouse model. This approach yielded feature representations that enhanced our ability to discern distinctive and noteworthy structures within the target ST data, leading to improvements in downstream analysis. + +<|ref|>text<|/ref|><|det|>[[112, 295, 885, 847]]<|/det|> +CoCo- ST was inspired by the recent successes of contrastive learning approaches6- 8, which learn discriminative feature representations by contrasting positive pairs (similar samples) with negative pairs (dissimilar samples). In our CoCo- ST design workflow (Fig. 1a), we began by collecting tissue samples from mouse lung and processing them using the Visium technology (10x Genomics) to obtain the ST data. We then organized the resulting gene expression data into a gene- spot matrix and further normalized the data to eliminate technical artifacts. CoCo- ST proceeded to construct two weighted graphs, one each for the background and target ST data sets—allowing us to capture the local structures within the data sets. We derived contrastive feature representations by comparing and contrasting the local variances of the background and target graphs. We achieved this by assessing the difference between their respective local total scatter matrices. In the case of a new target ST data set, CoCo- ST simply uses the learned transformation to generate feature representations for the new data (Fig. 1a). These contrastive feature representations can serve as inputs for various other ST analysis tools, for enhanced downstream analysis. We have illustrated the effectiveness of these contrastive feature representations across multiple downstream analysis tasks, including ST data visualization, spatial domain identification, tissue- specific spatial trajectory inference, trajectory inference across multiple tissues, and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 144]]<|/det|> +examination of cell- cell interaction. It is worth mentioning here that CoCo- ST is generically applicable to any ST data types that can be represented in form a gene- spot matrix. + +<|ref|>text<|/ref|><|det|>[[113, 158, 885, 529]]<|/det|> +We first applied CoCo- ST to learn transformation by using a mouse normal lung tissue sample (MLP- 1) as the background and an abnormal lung tissue sample (MLP- 6) containing structures other than the normal spatial domain (Extended Data Fig. 1) as the target. We designated MLP- 1 as the background ST data because its spatial structures belong to the normal lung spatial domain, which was also present in all the rest of the tissue samples. We then applied the learned transformation to the remaining tissue samples, resulting in contrastive feature representations that we subsequently used for spatial domain identification (Extended Data Fig. 1) and further downstream analysis. Note, CoCo- ST does not require much data to determine a good transformation compared to the conventional machine learning approaches. Additionally, it has the potential to capture more specific structures within individual samples. These properties make CoCo- ST a valuable complement to large foundation model- based approaches. + +<|ref|>text<|/ref|><|det|>[[113, 541, 885, 877]]<|/det|> +Uniform manifold approximation and projection (UMAP) embedding of the learned contrastive features in the target ST data (Extended Data Fig. 2a) illustrated CoCo- ST's effectiveness in determining feature representations that provide robust discrimination of various spatial structures in the target tissue (Fig. 1b). Clustering the ST data based on the learned contrastive components led to the identification of six clusters, each corresponding to a unique spatial structure. These spatial structures detected using CoCo- ST's contrastive components agree well with pathologist- annotated regions (Fig. 1b). Spatial clustering of spots based on components determined using the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods failed to effectively detect the hotspot region annotated as hyperplasia by the pathologist (Fig. 1b). Inability to detect spatial structures of low variability affects the performance of the compared + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 883, 144]]<|/det|> +methods in detecting the early adenoma (hotspot) region. However, Seurat (PCA) detected the hotspot region but annotated it as belonging to spatial domain 2. + +<|ref|>text<|/ref|><|det|>[[112, 155, 886, 880]]<|/det|> +We further annotated the detected spatial structures detected using CoCo- ST based on their differentially expressed marker genes (Extended Data Fig. 2b) and spatial locations. The distribution of these marker genes, including \*Epas1\* for normal lung tissue (endothelial PAS domain), Slc26a4 for fibrotic/scarred tissue, Cybb for adjacent normal tissue, Hp for the bronchus/alveoli, Ctsh for the adenoma, and Msln for the membrane, showed the expected high expression patterns (Extended Data Fig. 2c). To further validate the adenoma region (hotspot) detected using CoCo- ST, we investigated the most differentially expressed marker genes for the detected adenoma regions and found 3498 marker genes at a false- discovery rate of 5% (Fig. 1c). The most differentially expressed marker genes were domain- specific metagenes for the adenoma region (including the hotspot region). For example, a metagene consisting of \*Ctsh\*, \*Cxcl15\*, and \*Slc34a2\* marked the hotspot region clearly, as these genes exhibited high expression patterns in both the larger adenoma region and smaller hotspot region (Fig. 1e). The \*Cxcl15\*, and \*Slc34a2\* genes are uniquely identified by CoCo- ST. The high expression of these genes at both the large and hotspot adenoma regions indicates that these two spatial domains are anatomically similar. Seurat's inability to identify these important marker genes results to categorizing the hotspot region as belonging to the fibrotic/scarred tissue (Fig. 1b). Also, \*Ctsh\* gene was reported to be differentially expressed in adenoma region of patients with colorectal cancer9. Gene set enrichment analysis of the 10 most differentially expressed marker genes in our study identified biological processes related to lung fibrosis, apoptotic processes, and cell polarity (Extended Data Fig. 2d). For comparison, we also investigated the most differentially expressed marker genes for the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods (Fig. 1d, Extended Data + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 353]]<|/det|> +Fig. 2e) based on the learned embedding of these methods and found several genes, most of which marked the larger adenoma region but not the smaller hotspot region. For example, the Trf gene was the top marker gene for all of the compared methods (Extended Data Fig. 2e); however, this gene had a high expression pattern in the larger adenoma region but not in the hotspot region (Fig. 1f). These results demonstrated that the compared Seurat (PCA), STUtility (NMF), NSF and MEFISTO methods focus on identifying the main adenoma region with the largest variance, lacking the ability to identify domain- specific metagenes that capture the smaller adenoma structure (hotspot) with relatively low variance. + +<|ref|>text<|/ref|><|det|>[[112, 366, 886, 880]]<|/det|> +Examining the weights of the first five contrastive components revealed that CoCo- ST effectively identified major spatial domains (Fig. 2a), indicating that it captured local variations associated with the interesting spatial structures in the target data. For example, component 1 explained variation in multiple spatial domains, which was characterized by large positive weights around the adenoma and alveoli/bronchus and negative weights around the normal lung. Comparing to Seurat (PCA), STUtility (NMF), NSF and MEFISTO, the top components of these methods predominantly focus on the normal lung structure with the largest variance (Fig. 2d). For example, the first components of both Seurat PCA and NSF exhibited larger weights on normal lung structures. Because the first few components of these methods are expected to capture most of the information in the original data and are subsequently used as inputs for downstream analysis, relying solely on these components may result in overlooking crucial biological insights. To gain deeper insight into the underlying biological processes associated with these components, we further investigated the top 20 genes with the largest weights on each of the CoCo- ST's contrastive components (Fig. 2b). This highlighted individual genes encoding domain- specific signatures such as Retnla, Cyp2f2, Ctsh, Ccl6, and Acta2 (Fig. 2c) as well as gene sets linked with broader + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 248]]<|/det|> +biological processes and pathways. Gene set enrichment analysis with the top 20 marker genes for each component revealed enriched gene ontology terms and KEGG pathways specific to each spatial domain. These included heme binding on component 1, retinol metabolism on component 2, IgA immunoglobulin complex on component 3, lysosome on component 4, and extracellular matrix on component 5 (Extended Data Fig. 3). + +<|ref|>text<|/ref|><|det|>[[113, 262, 884, 562]]<|/det|> +To investigate the impact of different graph construction methods (molecular vs. spatial) on CoCo- ST's performance, we constructed a similarity graph based on spatial coordinates rather than gene expression data as done in our prior experiments. This approach has proven highly effective10, as it assumes that neighboring spots in the tissue have similar gene expression patterns and likely belong to the same spatial domain. Our findings demonstrated robust CoCo- ST performance when using the similarity graph constructed from the spatial coordinates, effectively identifying the major spatial domains across all target tissue samples (Extended Data Fig. 4). In summary, CoCo- ST demonstrates robust performance with similarity graphs constructed from both spatial coordinates and gene expression data. + +<|ref|>text<|/ref|><|det|>[[113, 577, 884, 876]]<|/det|> +Next, we performed deconvolution analysis to infer the cell type composition at each of the spatial domains detected using CoCo- ST. For this analysis, we used matched single- cell RNA sequencing (scRNA- seq) data (Extended Data Fig. 5a) obtained from the same MLP tissue samples as a reference. As expected for the MLP- 6 tissue sample (Extended Data Fig. 5b), we observed a concentration of endothelial cells in the normal lung spatial domain (endothelial PAS domain) (Extended Data Figs. 5c,d, 6, 7). The fibrotic/scarred and bronchus/alveoli spatial domains were enriched with fibroblasts. In the adjacent normal spatial domain was an abundance of endothelial cells, whereas the adenoma spatial domain had enrichment of macrophages and proliferating macrophages (Extended Data Figs. 5c,d, 6, 7). Notably, we observed tumor- associated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 352]]<|/det|> +macrophages (TAMs) in the adjacent normal spatial domain (Extended Data Figs. 5c,d, 6, 7), which exhibited significantly upregulated Ccl6. This gene was the top gene with the highest weight on component 4 (Fig. 2c). Of note, component 4 exhibited large weights in spatial regions corresponding to the regions with the highest Ccl6 gene expression. Also, high expression of the Ccl6 gene in a mouse model of lung cancer was reported to be associated with tumor growth and increased metastasis11. This evidence underscores the intricate cellular compositions within specific spatial domains, shedding light on potential implications for the progression of lung cancer. + +<|ref|>text<|/ref|><|det|>[[113, 367, 884, 596]]<|/det|> +After determining the composition of cell types in the various tissue samples through our deconvolution analysis, we next inferred their communication patterns. Initially, we identified cell- cell interactions by examining ligand- receptor patterns within the individual MLP tissue samples. Our analysis of the MLP- 6 tissue sample revealed a strong pattern of communication between endothelial and epithelial cells as well as between endothelial cells and fibroblasts (Extended Data Fig. 5e). Also, we observed strong communication initiating from both proliferating macrophages and B cells within the adenoma spatial domain, indicating an active immune response. + +<|ref|>text<|/ref|><|det|>[[113, 610, 884, 876]]<|/det|> +We observed that multiple signaling pathways, including programmed death- ligand 1, GRN, inducible co- stimulator, NECTIN, interleukin- 6, WNT, and CXCL, played pivotal roles in cell interactions across different spatial domains. Notably, we predominantly observed WNT ligand- receptor interactions in endothelial cells, epithelial cells, fibroblasts, and macrophages (Extended Data Fig. 5f). Additionally, we observed WNT signaling interactions between proliferating macrophages and B cells, which are enriched in the adenoma spatial domain. Meanwhile, we found self- interaction (among cells of the same group) to be the strongest in proliferating T cells, proliferating macrophages, and endothelial cells (Extended Data Fig. 5g). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 249]]<|/det|> +Network centrality analysis of the inferred WNT signaling network identified TAMs (macrophages and proliferating macrophages) as prominent mediators (gatekeepers) as well as influencers controlling the communication (Extended Data Fig. 5h). Prior studies demonstrated that WNT signaling supports TAMs as drivers of tumor growth and that TAM- derived WNT ligands support tumorigenesis12. + +<|ref|>text<|/ref|><|det|>[[113, 262, 885, 596]]<|/det|> +We delved deeper into the cell- cell interactions across groups of tissue samples associated with the adenoma and adenocarcinoma spatial domains as determined using CoCo- ST. Specifically, we aggregated the communication weights of multiple tissue samples containing the adenoma (MLP- 3, - 4, - 5, - 6, - 7, and - 9) and adenocarcinoma (MLP- 8 and - 10) spatial domains to investigate the cell- cell interactions on a broader scale. Of note, we observed a bidirectional interaction between epithelial cells and proliferating macrophages in the adenocarcinoma group (Extended Data Fig. 8a,b) but did not see a similar interaction pattern in the adenoma group (Extended Data Fig. 8c,d). This is consistent with the established role of TAMs in promoting tumor growth and metastasis by engaging in an autocrine loop with cancer cells, thereby stimulating cancer cell progression13- 16. + +<|ref|>text<|/ref|><|det|>[[113, 610, 885, 876]]<|/det|> +Next, we investigated how the normal endothelial, adjacent normal, and tumor spatial domains are connected to each other during tumorigenesis. Specifically, we performed spatial trajectory inference with MLP- 6 tissue using the contrastive components derived from CoCo- ST. This analysis revealed a trajectory starting from the normal endothelial domain and moving toward the adjacent normal domain and further into the adenoma spatial domain (Extended Data Fig. 9a,b). To gain a comprehensive view of the trajectory of precancer evolution across the entire population, we combined spots belonging to the adenoma and adenocarcinoma spatial domains as identified by our contrastive components. We then determine a UMAP embedding of the spots + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 886, 458]]<|/det|> +(Extended Data Fig. 9c) with which the trajectories were reconstructed (Extended Data Fig. 9d). As seen in Extended Data Fig. 9c, the contrastive components effectively discriminated the three spatial domains and identified a trajectory starting from the normal lung, passing toward the adenoma, and ending at the adenocarcinoma cluster (Extended Data Fig. 9d). These findings align with the well- known biology of mouse tumorigenesis, consisting of a transition from normal tissue to hyperplasia, adenoma, and finally adenocarcinoma. Furthermore, we identified modules of differentially expressed genes that were co- expressed across spots in the different spatial domains as determined using CoCo- ST (Extended Data Fig. 9e,f). Notably, these modules demonstrated high specificity for the different spatial domains, further indicating the effectiveness of CoCo- ST in determining feature representations that captured both the shared and unique spatial structures across the different tissues. + +<|ref|>text<|/ref|><|det|>[[113, 473, 886, 879]]<|/det|> +Lastly, we employed CoCo- ST to analyze a publicly available Visium data set generated from mouse brain (anterior and posterior). This data set shows tissue structures that are considerably more complex than the mouse lung precancer data set described above. First, we examined the spatial domain identification performance of CoCo- ST when considering the anterior slice as the reference and the posterior slice as the target and vice versa. The spatial domains detected using CoCo- ST's contrastive components agree well with the Allen Institute for Brain Science reference atlas diagram (Extended Data Fig. 10a) \(^{17}\) . We further investigated the top five contrastive components as determined using CoCo- ST for both the anterior and posterior slices. All of these components captured spatial patterns highlighting specific major anatomical regions in the brain (Extended Data Fig. 10b,e). Similar to the mouse precancer model, these components exhibited high component values on specific anatomical regions, such as the cerebral cortex (for anterior component 1) and choroid plexus (for posterior component 2). The top genes for each + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 144]]<|/det|> +component (Extended Data Fig. 10c,f) had distinct spatial patterns and exhibited spatial localization to specific brain regions (Extended Data Fig. 10d,g). + +<|ref|>text<|/ref|><|det|>[[113, 158, 885, 423]]<|/det|> +To summarize, we introduced an ST feature representation method that opens up the application of graph contrastive learning to ST data analysis. This approach offers significant advantages, particularly in scenarios involving the analysis of multiple ST data sets. It effectively identifies interesting, unique spatial structures in a target ST data set while mitigating the influence of dominant high- variance spatial structures that are common to both target and background ST data sets. Whereas we focused on the ST and Visium platforms, adaptation of CoCo- ST to other platforms such as Xenium, CosMX SMI and MERFISH on which the data can be represented in the form of a gene spot matrix is plausible. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 191, 107]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 124, 387, 143]]<|/det|> +## Problem definition and notation + +<|ref|>text<|/ref|><|det|>[[113, 157, 886, 440]]<|/det|> +We represented a spatially resolved ST slice from a spatial genomics technology as the set of pairs \(\{x_{i},y_{i}\}_{i = 1}^{n}\) , with \(y_{i}\in R^{2}\) denoting a vector of spatial coordinates and \(x_{i}\in R^{d}\) denoting a vector of measured gene expression at a corresponding spatial location. We referred to a single spatial location \(x_{i}\) as a spot and \(s\in \{1,2,\dots,S_{i}\}\) as a slice containing \(n_{s}\) spots. Let \(X_{s} = \left[x_{1}^{s},x_{2}^{s},\dots,x_{n_{s}}^{s}\right]^{T}\) denote the matrix containing the spot gene expression measurements and \(Y_{s} = \left[y_{1}^{s},y_{2}^{s},\dots,y_{n_{s}}^{s}\right]^{T}\) denote the corresponding spatial location matrix from slice \(s\) . Worth noting is that the number of spots can differ across different slices and that the slices may be from the same tissue sample or from two different tissue sample. + +<|ref|>text<|/ref|><|det|>[[113, 454, 885, 579]]<|/det|> +Our goal is to analyze these \(S_{l}\) slices by finding discriminative feature representations that capture the interesting spatial patterns within the different slices. To do this, we identified a background ST data set containing dominant high- variance spatial structures that were present across all slices. + +<|ref|>text<|/ref|><|det|>[[113, 592, 885, 895]]<|/det|> +The background ST data play a crucial role in effectively contrasting dominant high- variance spatial structures, which was not the primary focus of this analysis, and in turn assists in detecting the intriguing unique spatial structures enriched in individual target slices. Three key advancements underlie the robust performance of our graph contrastive learning approach. First, we used paired slices to mitigate the impact of spatial structures that are not of primary interest, which subsequently aided the detection of unique spatial structures of particular interest in individual target slices. Second, we constructed local similarity graphs to capture the nuanced local structures in both the background and target ST data sets, thereby ensuring that important spatial structures are not lost. Third, we applied the concept of contrastive learning to compare and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 214]]<|/det|> +contrast the graph embedding of the background and target ST data sets, ensuring that similar spots are positioned close to each other and that dissimilar ones are distanced in the latent space. This collective methodology ensures the accurate identification and representation of distinctive spatial structures. + +<|ref|>sub_title<|/ref|><|det|>[[115, 262, 375, 283]]<|/det|> +## Graph representation learning + +<|ref|>text<|/ref|><|det|>[[112, 293, 886, 896]]<|/det|> +Recent advances in spatial molecular profiling made graph learning a focus of attention because of the innate resemblance of spatial information to spatial graphs. Graph embedding techniques have great potential for various applications across spatially resolved transcriptomics. Because ST data sets can be represented in a matrix format, we can identify spots as entities of interest and interrogate their interaction. This is equivalent to constructing gene or spot graphs based on suitable similarity measures. Herein we describe the construction of such molecular similarity graphs. An essential task in ST data analysis is to find a lower dimensional manifold space that captures local neighborhood information. Given an ST datum (slice), we can construct a weighted graph \(G = (V, E)\) representing complex, non- Euclidean structures, with edges \(e_{ij} \in E\) connecting nearby nodes \(i\) and \(j\) ( \(i, j \in V\) ) to each other if spots \(x_{i}^{s}\) and \(x_{j}^{s}\) are molecularly similar. A natural variation of this graph is to construct a graph of \(k\) - nearest neighbors in which similarity of nodes is usually quantified using the Euclidean metric (i.e., nodes \(i\) and \(j\) are connected by an edge \(e_{ij}\) if \(x_{i}^{s}\) is among the \(k\) - nearest neighbors of \(x_{j}^{s}\) or \(x_{j}^{s}\) is among the \(k\) - nearest neighbors of \(x_{i}^{s}\) ). The graph structure \(G = (V, E)\) is commonly encoded in an \(n_{s} \times n_{s}\) affinity matrix \(S\) with entries in \([0, 1]\) and takes large values if \(x_{i}^{s}\) and \(x_{j}^{s}\) are close (or similar). Several approaches to computing the affinity matrix \(S\) are available, one of which is the heat kernel weighting technique depicted by the equation + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[288, 88, 850, 150]]<|/det|> +\[S_{ij}^{s} = \left\{ \begin{array}{l l}{\frac{\left\|x_{i}^{s} - x_{j}^{s}\right\|^{2}}{t},} & {\mathrm{if~}x_{i}^{s}\in N\left(x_{j}^{s}\right)\mathrm{~or~}x_{j}^{s}\in N\left(x_{i}^{s}\right)}\\ 0, & {\mathrm{Otherwise}} \end{array} \right. \quad (1),\] + +<|ref|>text<|/ref|><|det|>[[113, 164, 850, 186]]<|/det|> +where \(N(x_{j}^{s})\) denotes the set of \(k\) - nearest neighbors of \(x_{j}^{s}\) and \(t\) is a user- specified parameter. + +<|ref|>text<|/ref|><|det|>[[112, 201, 886, 432]]<|/det|> +Based on the graph construction approach described above, the similarity among spots is quantified based on gene expression measurements at the corresponding spots. However, because gene expression measurements are captured alongside its spatial information in ST, these spatial locations can be used to construct similarity graphs. The spatial graphs constructed in this way are similar to molecular similarity graphs in the sense that nodes correspond to spots. However, edges capture proximity of spots in the \(R^{2}\) coordinate space. The affinity matrix with the spatial locations can now be constructed as + +<|ref|>equation<|/ref|><|det|>[[288, 446, 850, 508]]<|/det|> +\[S_{ij}^{s} = \left\{ \begin{array}{l l}{\frac{\left\|y_{i}^{s} - y_{j}^{s}\right\|^{2}}{t},} & {\mathrm{if~}y_{i}^{s}\in N\left(y_{j}^{s}\right)\mathrm{~or~}x_{j}^{s}\in N\left(x_{i}^{s}\right)}\\ 0, & {\mathrm{Otherwise}} \end{array} \right. \quad (2).\] + +<|ref|>text<|/ref|><|det|>[[112, 521, 886, 715]]<|/det|> +Also, the spatial graph can be constructed using both the spatial locations and the molecular profiles treated as node features. Graph representation learning approaches are considered to determine biologically meaningful representations of these graphs by finding meaningful lower dimensional representations of nodes present in a complex graph, where local structures in the data are well captured. A widely used criterion for determining such a representation is to solve the objective function + +<|ref|>equation<|/ref|><|det|>[[406, 730, 850, 764]]<|/det|> +\[\min_{W}\sum_{i,j = 1}^{n_{s}}\left\| z_{i}^{s} - z_{j}^{s}\right\|^{2}S_{ij}^{s} \quad (3),\] + +<|ref|>text<|/ref|><|det|>[[112, 777, 884, 874]]<|/det|> +where \(z_{i}^{s} = W^{T}x_{i}^{s}\) denotes the lower dimensional representation of \(x_{i}^{s}\) . Solving Eq. (3) under appropriate constraints ensures that if \(x_{i}^{s}\) and \(x_{j}^{s}\) are similar (or nodes \(i\) and \(j\) are connected in the graph), then \(z_{i}^{s}\) and \(z_{j}^{s}\) are similar (close), as well. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 125, 419, 144]]<|/det|> +## Contrastive representation learning + +<|ref|>text<|/ref|><|det|>[[112, 157, 886, 565]]<|/det|> +Contrastive learning has recently emerged as a successful method of unsupervised graph representation learning. Contrastive learning methods first perform augmentation of the input data and enforce via a suitable objective function mapping of augmentation of the same data (positive pairs) close to each other in the representation (latent) space and augmentation of different data (negative pairs) far apart from each other. Arguably, a low- dimensional representation that is near optimal in the contrastive objective function is guaranteed to linearly separate similar data from dissimilar data. Such representations provide competitive performance in a host of downstream tasks. In early visual representation learning studies, researchers leveraged a pixel as local view to conduct local- to- local \(^{18}\) or local- to- global \(^{19}\) contrastive learning, whereas researchers recently found that randomly cropped image snippets help contrastive models better capture the relationships between image elements \(^{6}\) . This motivated us to perform contrastive representation learning at the global image level. + +<|ref|>text<|/ref|><|det|>[[113, 576, 885, 701]]<|/det|> +Like several other machine learning approaches, contrastive representation learning can be performed in an unsupervised (self- supervised) or supervised learning strategy. In self- supervised settings, contrastive learning methods learn discriminative feature representations based on some similarity measure defined according to the data. Consider the objective function defined by \(^{20}\) + +<|ref|>equation<|/ref|><|det|>[[230, 715, 735, 750]]<|/det|> +\[L_{1} = (1 - Y)*\frac{1}{2}\left\| x_{i} - x_{j}\right\|^{2} + \frac{Y}{2}*\left\{\max \left(0,m - \left\| x_{i} - x_{j}\right\|^{2}\right)\right\}^{2}\] + +<|ref|>text<|/ref|><|det|>[[291, 765, 321, 783]]<|/det|> +(4), + +<|ref|>text<|/ref|><|det|>[[113, 797, 884, 893]]<|/det|> +where \(m > 0\) is a hyperparameter defining the lower bound distance between dissimilar samples, \(Y\) is a binary label with \(Y = 0\) if \(x_{i}\) and \(x_{j}\) are similar, and \(Y = 1\) if \(x_{i}\) and \(x_{j}\) are dissimilar. Minimizing the objective function is an attempt to determine a lower dimensional manifold + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 319]]<|/det|> +subspace where similar input samples are mapped nearby and dissimilar samples are far apart. When sample labels are available, they can be integrated into the definition of similarity and dissimilarity to better guide the contrastive model to mapped samples belonging to the same class (same label) close to each other and samples of different classes farther apart. This approach is referred to as supervised contrastive representation learning. Both the self- supervised and fully supervised contrastive learning approaches are powerful methods of learning discriminative feature representations. + +<|ref|>sub_title<|/ref|><|det|>[[115, 367, 595, 388]]<|/det|> +## Graph contrastive feature representation using CoCo-ST + +<|ref|>text<|/ref|><|det|>[[113, 401, 885, 701]]<|/det|> +Most of the traditional feature representation approaches are designed to determine feature representations through maximization of data variance. These approaches can perform poorly if the ST data structures with maximal variances are not the structures of interest, as the local structures of interest are masked by the dominant high- variance structures. The feature representations determined using these approaches capture little to no useful information reflecting the unique low- variance local structures present in the ST data, which are usually treated as noise. Also, these traditional approaches are designed to explore one ST data set at a time, which can hinder their performance in cases where there are multiple interconnected data sets that need to be explored. + +<|ref|>text<|/ref|><|det|>[[113, 715, 884, 876]]<|/det|> +To overcome these limitations, we propose CoCo- ST, which compares and contrasts the global and local variances in ST data sets to better capture discriminant and structural information. More generally, we use two ST data sets (background and target) and subsequently construct two similarity graph views: one for the background ST data set and the other for the target ST data set. We then design a contrastive objective function to learn feature representations that capture high + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 884, 256]]<|/det|> +global (and/or local) variances enriched in the target ST data while simultaneously attaining small global (and/or local) variances in the background ST data. Given a background ST data set \(X_{b} = \left[x_{1}^{b}, x_{2}^{b}, \ldots , x_{n_{b}}^{b}\right]^{T}\) containing spatial structures of no primary interest, such as a normal lung region, we can use the following two terms to measure the smoothness of the lower dimensional representation: + +<|ref|>equation<|/ref|><|det|>[[365, 268, 846, 370]]<|/det|> +\[\begin{array}{l}{{\mathcal{R}_{1}=\min _{W}\sum_{i=1}^{n_{b}}\left\|x_{i}^{b}-W W^{T}x_{i}^{b}\right\|^{2}}}\\ {{\quad=\max _{W}t r(W^{T}X_{b}X_{b}^{T}W)}}\end{array} \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[113, 384, 148, 401]]<|/det|> +and + +<|ref|>equation<|/ref|><|det|>[[270, 415, 852, 595]]<|/det|> +\[\begin{array}{r l} & {\mathcal{R}_{2} = \min_{W}\sum_{i,j = 1}^{n_{b}}\left\| W^{T}x_{i}^{b} - W^{T}x_{j}^{b}\right\|^{2}S_{i j}^{b}}\\ & {= \min_{W}\left(\sum_{i,j = 1}^{n_{b}}W^{T}x_{i}^{b}D_{i i}^{b}\big(x_{i}^{b}\big)^{T}W^{T} - W^{T}x_{i}^{b}S_{i j}^{b}\big(x_{j}^{b}\big)^{T}W^{T}\right)}\\ & {\qquad = \min_{W}t r(W^{T}X_{b}L^{b}X_{b}^{T}W)} \end{array} \quad (6),\] + +<|ref|>text<|/ref|><|det|>[[112, 606, 884, 846]]<|/det|> +where \(t r(\cdot)\) is the trace operator, \(D\) is a diagonal matrix whose entries are the column (or row) sums of \(S\) , \(D_{i i} = \sum_{k} S_{i k}\) , and \(L = D - S\) is the graph Laplacian matrix. We consider the symmetric normalized graph Laplacian matrix \(\bar{L} = D^{- 1 / 2} L D^{- 1 / 2}\) in our later derivations. By minimizing \(\mathcal{R}_{1}\) , we aim to minimize the reconstruction error, whereas minimizing \(\mathcal{R}_{2}\) is an attempt to preserve the local structure (i.e., if two spots \(x_{i}^{s}\) and \(x_{j}^{s}\) are molecularly similar, their low- dimensional representations \(W^{T} x_{i}^{b}\) and \(W^{T} x_{j}^{b}\) are also similar). Combining Eqs. (5) and (6), we can have the equivalent formulation + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[285, 87, 850, 163]]<|/det|> +\[\begin{array}{l}{{O_{1}=\max _{W^{T}W=I}t r(W^{T}X_{b}X_{b}^{T}W)-\mu_{1}t r(W^{T}X_{b}\bar{L}^{b}X_{b}^{T}W)}}\\ {{\quad=\max _{W^{T}W=I}t r(W^{T}X_{b}H_{b}X_{b}^{T}W)}}\end{array} \quad (7), \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[113, 208, 884, 339]]<|/det|> +where \(H_{b} = I - \mu_{1}\bar{L}^{b}\) , \(I\) is an identity matrix, \(\bar{L}^{b}\) is the normalized graph Laplacian for the background ST data, and \(0 \leq \mu_{1} \leq 1\) is a hyperparameter that controls the smoothness of the new representation. The matrix \(H_{b} = I - \mu_{1}\bar{L}^{b}\) can be considered a graph Laplacian filter \(^{21}\) that helps smooth the data while preserving underlying spatial structures in an ST slice. + +<|ref|>text<|/ref|><|det|>[[113, 353, 883, 414]]<|/det|> +Similarly, for a target ST data set \(X_{t} = \left[x_{1}^{t}, x_{2}^{t}, \ldots , x_{n_{t}}^{t}\right]^{T}\) containing unique, interesting spatial structures, we can write the formulation + +<|ref|>equation<|/ref|><|det|>[[382, 462, 850, 494]]<|/det|> +\[O_{2} = \max_{W^{T}W = I}t r(W^{T}X_{t}H_{t}X_{t}^{T}W) \quad (8),\] + +<|ref|>text<|/ref|><|det|>[[113, 539, 882, 598]]<|/det|> +where \(H_{t} = I - \mu_{2}\bar{L}^{t}\) , \(\bar{L}^{t}\) is the normalized graph Laplacian for the target ST data and \(0 \leq \mu_{2} \leq 1\) is a hyperparameter. + +<|ref|>text<|/ref|><|det|>[[171, 612, 789, 634]]<|/det|> +Combining Eqs. (7) and (8), CoCo- ST solves the following objective function + +<|ref|>equation<|/ref|><|det|>[[323, 680, 848, 712]]<|/det|> +\[O_{3} = \max_{W^{T}W = I}t r(W^{T}X_{t}H_{t}X_{t}^{T}W) - \eta t r(W^{T}X_{b}H_{b}X_{b}^{T}W) \quad (9),\] + +<|ref|>text<|/ref|><|det|>[[113, 758, 884, 888]]<|/det|> +where \(\eta \geq 0\) is the contrastive parameter that determines the tradeoff between high target global (and/or local) variance and low background global (and/or local) variance. We will first describe how to maximize the objective function \(O_{3}\) . Let \(\Lambda\) be the Lagrange multiplier for the constraint \(W^{T}W = I\) . The Lagrange \(\mathcal{L}\) is + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[260, 122, 860, 145]]<|/det|> +\[\mathcal{L} = t r(W^{T}X_{t}H_{t}X_{t}^{T}W) - \eta t r(W^{T}X_{b}H_{b}X_{b}^{T}W) - \Lambda t r(W^{T}W - I) \quad (10).\] + +<|ref|>text<|/ref|><|det|>[[113, 195, 488, 214]]<|/det|> +The partial derivative of \(\mathcal{L}\) with respect to \(W\) is + +<|ref|>equation<|/ref|><|det|>[[346, 264, 659, 296]]<|/det|> +\[\frac{\partial\mathcal{L}}{\partial W} = X_{t}H_{t}X_{t}^{T}W - \eta X_{b}H_{b}X_{b}^{T}W - \Lambda W\] + +<|ref|>text<|/ref|><|det|>[[346, 312, 390, 329]]<|/det|> +(11). + +<|ref|>text<|/ref|><|det|>[[113, 381, 669, 411]]<|/det|> +The optimum solution to Eq. (10) satisfies \(\frac{\partial\mathcal{L}}{\partial W} = 0\) . We therefore have + +<|ref|>equation<|/ref|><|det|>[[348, 456, 860, 517]]<|/det|> +\[\begin{array}{r l} & {X_{t}H_{t}X_{t}^{T}W - \eta X_{b}H_{b}X_{b}^{T}W - \Lambda W = 0}\\ & {(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})W = \Lambda W} \end{array} \quad (12).\] + +<|ref|>text<|/ref|><|det|>[[113, 565, 884, 696]]<|/det|> +Thus, the transformation matrix that maximizes the objective function \(\mathcal{O}_{3}\) can be obtained by solving the eigenvalue problem (Eq. 12). Let \(w_{1},w_{2},\dots,w_{p}\) be the eigenvectors from Eq. (12) corresponding to the top \(p\) largest eigenvalues \(\lambda_{1}\geq \lambda_{2}\geq \dots ,\geq \lambda_{p}\) . The lower dimensional manifold representation can then be obtained as follows: + +<|ref|>equation<|/ref|><|det|>[[435, 708, 860, 730]]<|/det|> +\[x_{t}^{t}\rightarrow z_{t}^{t} = W^{T}x_{t}^{t} \quad (13),\] + +<|ref|>equation<|/ref|><|det|>[[408, 746, 589, 767]]<|/det|> +\[W^{T} = [w_{1},w_{2},\dots,w_{p}]\] + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 180]]<|/det|> +where \(z_{t}^{t}\) is a \(p\) - dimensional representation of \(x_{t}^{t}\) , and \(W\) is a \(d \times p\) matrix. This feature representation preserves the local structure of the ST data sets. A step- by- step description of the proposed CoCo- ST method is summarized in Algorithm 1. + +<|ref|>sub_title<|/ref|><|det|>[[123, 230, 321, 249]]<|/det|> +## Algorithm 1. CoCo-ST. + +<|ref|>text<|/ref|><|det|>[[123, 264, 877, 413]]<|/det|> +Input: Background \(X_{b} = \left[x_{1}^{b},x_{2}^{b},\dots,x_{n_{b}}^{b}\right]^{T}\) and target \(X_{t} = \left[x_{1}^{t},x_{2}^{t},\dots,x_{n_{t}}^{t}\right]^{T}\) ST data sets, together with corresponding spatial locations \(Y_{b} = \left[y_{1}^{b},y_{2}^{b},\dots,y_{n_{b}}^{b}\right]^{T}\) and \(Y_{t} =\) \(\left[y_{1}^{t},y_{2}^{t},\dots,y_{n_{t}}^{t}\right]^{T}\) , the number of nearest neighbors \((k)\) , and the hyperparameters \(\mu_{1},\mu_{2}\) and \(\eta\) . + +<|ref|>text<|/ref|><|det|>[[123, 428, 875, 485]]<|/det|> +Output: The low- dimensional contrastive feature representations for the target ST data \(Z_{t} =\) \(W^{T}X_{t}\) + +<|ref|>text<|/ref|><|det|>[[123, 498, 875, 556]]<|/det|> +1. Construct the adjacency matrix for both the background and target ST data sets according to Eq. (1) or (2). + +<|ref|>text<|/ref|><|det|>[[123, 568, 875, 628]]<|/det|> +2. Construct the normalized graph Laplacian matrices \(\bar{L}^{b}\) and \(\bar{L}^{t}\) together with the graph Laplacian filters \(H_{b} = I - \mu_{1}\bar{L}^{b}\) and \(H_{t} = I - \mu_{2}\bar{L}^{t}\) . + +<|ref|>text<|/ref|><|det|>[[123, 642, 512, 664]]<|/det|> +3. Compute the matrices \(X_{b}H_{b}X_{b}^{T}\) and \(X_{t}H_{t}X_{t}^{T}\) . + +<|ref|>text<|/ref|><|det|>[[123, 678, 483, 699]]<|/det|> +4. Solve the eigenvalue problem in Eq. (12). + +<|ref|>text<|/ref|><|det|>[[123, 712, 875, 770]]<|/det|> +5. Compute the low-dimensional contrastive feature representations for the target ST data as \(Z_{t} = W^{T}X_{t}\) . + +<|ref|>text<|/ref|><|det|>[[113, 825, 884, 884]]<|/det|> +We next investigate the computational complexity of the proposed CoCo- ST algorithm. Its complexity is dominated mainly by three parts: local similarity graph construction, matrix + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 545]]<|/det|> +multiplication, and solving an eigenvalue problem. Assuming we have \(n_{b}\) and \(n_{t}\) spots in \(d\) - dimensional spaces ( \(d\) gene expression measurements) for the background and target ST data sets, to construct the similarity graphs, we first perform a \(k\) - nearest neighbor search for both data sets. The distance between any two spots in the background ST data can be computed in \(O(dn_{b}^{2})\) , and the \(k\) - nearest neighbors can be found with \(O(kn_{b}^{2})\) . Thus, the \(k\) - nearest neighbor search for the background and target ST data sets has complexities \(O\big((d + k)n_{b}^{2}\big)\) and \(O\big((d + k)n_{t}^{2}\big)\) , respectively. The complexities for computing the matrices \(X_{b}H_{b}X_{b}^{T}\) and \(X_{t}H_{t}X_{t}^{T}\) are \(O\big((n_{b}^{2} + n_{b}d)d\big)\) and \(O\big((n_{t}^{2} + n_{t}d)d\big)\) , respectively. The last part is computing the eigenvectors corresponding to the top \(p\) eigenvalues of the eigenproblem in Eq. (12), whose complexity is \(O(pd^{2})\) . Therefore, the time complexity of the CoCo- ST algorithm is \(O\big((d + k)(n_{b}^{2} + n_{t}^{2}) + \big((n_{b} + d)n_{b} + (n_{t} + d)n_{t} + pd\big)d\big)\) . Because \(k\ll n_{b}(\mathrm{or}n_{t})\) and \(p\ll d\) , the overall complexity of CoCo- ST is determined by the number of spots \(n_{b}(\mathrm{or}n_{t})\) and the number of genes \((d)\) . + +<|ref|>text<|/ref|><|det|>[[170, 555, 872, 576]]<|/det|> +Several aspects of the proposed CoCo- ST approach are worth highlighting. Specifically: + +<|ref|>text<|/ref|><|det|>[[144, 590, 884, 682]]<|/det|> +1. If \(\mu_{1} = \mu_{2} = 0\) , the matrices \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) reduce to \(X_{t}X_{t}^{T}\) and \(X_{b}X_{b}^{T}\) , respectively, so the objective function \((O_{3})\) reduces to that of contrastive PCA (cPCA) \(^{22}\) . Therefore, cPCA can be regarded as a variant of CoCo- ST. + +<|ref|>text<|/ref|><|det|>[[143, 696, 886, 856]]<|/det|> +2. Whereas cPCA and the majority of the traditional feature representation approaches focus on global geometrical structures, CoCo-ST can exploit the intrinsic geometric structures of ST data sets and incorporate them as additional regularization terms. Through construction of a graph to model local geometric structures, CoCo-ST can have more discriminating power than cPCA and the traditional feature representation approaches. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[143, 88, 884, 179]]<|/det|> +3. CoCo-ST simultaneously learns both global and local-level representations to complement tissue-wide representations, enabling it to distinguish different spatial areas in an ST tissue slice. + +<|ref|>text<|/ref|><|det|>[[143, 193, 884, 319]]<|/det|> +4. The graphs in our proposed CoCo-ST approach are solely unsupervised and constructed from molecular data or spatial location information. Other information, such as label information, can also be used to guide graph construction, leading to other extensions of CoCo-ST such as supervised or semisupervised CoCo-ST. + +<|ref|>text<|/ref|><|det|>[[143, 333, 884, 494]]<|/det|> +5. The proposed CoCo-ST approach differs from existing graph contrastive learning approaches that focus on graph neural network architectures for graph structured data. CoCo-ST considers the gene expression data and tries to learn local representations to better capture ST data structural information. As such, the objective functions of CoCo-ST and the conventional graph neural networks are different. + +<|ref|>sub_title<|/ref|><|det|>[[114, 542, 486, 562]]<|/det|> +## Why is CoCo-ST good for ST data analysis? + +<|ref|>text<|/ref|><|det|>[[113, 576, 884, 806]]<|/det|> +CoCo-ST imposes molecularly or spatially similar spots to have similar feature representations, by which the intrinsic geometric structure of the ST data tends to be preserved. This is a useful property in ST data analysis because interesting spatial structures will not be lost owing to feature representation. In addition, CoCo-ST determines its discriminant (contrastive) feature representations from both the background and target ST data sets and thus can provide even more discriminative feature representations than the traditional approaches that focus only on a single ST data set. To explain this, we provided the following remarks and theorem. + +<|ref|>sub_title<|/ref|><|det|>[[115, 857, 200, 874]]<|/det|> +## Remark 1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 885, 288]]<|/det|> +When \(\eta = 0\) , CoCo- ST degenerates to a feature representation method that determines its discriminant vectors from the range space of the matrix \(X_{t}H_{t}X_{t}^{T}\) associated with the target data alone. When \(\eta >0\) , the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) is not guaranteed to be positive semidefinite even though \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are both symmetric and positive semidefinite. Let \(w\) be the eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda < 0\) . We then have + +<|ref|>equation<|/ref|><|det|>[[370, 303, 628, 325]]<|/det|> +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] + +<|ref|>equation<|/ref|><|det|>[[365, 339, 631, 361]]<|/det|> +\[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w + \lambda\] + +<|ref|>equation<|/ref|><|det|>[[368, 375, 630, 419]]<|/det|> +\[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w} = \eta +\frac{\lambda}{w X_{b}H_{b}X_{b}^{T}w}\] + +<|ref|>text<|/ref|><|det|>[[113, 434, 769, 456]]<|/det|> +Because both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are positive semidefinite, we can conclude that + +<|ref|>equation<|/ref|><|det|>[[348, 470, 647, 514]]<|/det|> +\[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w} = \eta +\frac{\lambda}{w X_{b}H_{b}X_{b}^{T}w}\geq 0\] + +<|ref|>text<|/ref|><|det|>[[113, 528, 883, 585]]<|/det|> +Thus, the eigenvectors corresponding to the negative eigenvalues are derived from the range space of \(X_{b}H_{b}X_{b}^{T}\) and contain some discriminant information. + +<|ref|>sub_title<|/ref|><|det|>[[114, 636, 208, 653]]<|/det|> +## Theorem 1 + +<|ref|>text<|/ref|><|det|>[[113, 669, 884, 764]]<|/det|> +Suppose the matrix \(X_{b}H_{b}X_{b}^{T}\) is singular and that \(w\) is an eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda >0\) . The eigenvector \(w\) is then in the null space of \(X_{b}H_{b}X_{b}^{T}\) when \(\eta \rightarrow \infty\) . + +<|ref|>text<|/ref|><|det|>[[113, 779, 883, 836]]<|/det|> +Proof. Because \(w\) is the eigenvector of the matrix \(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T}\) corresponding to the eigenvalue \(\lambda >0\) , we have + +<|ref|>equation<|/ref|><|det|>[[372, 852, 624, 873]]<|/det|> +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[355, 88, 641, 128]]<|/det|> +\[w X_{b}H_{b}X_{b}^{T}w = \frac{1}{\eta} (w X_{t}H_{t}X_{t}^{T}w - \lambda)\] + +<|ref|>text<|/ref|><|det|>[[113, 143, 404, 163]]<|/det|> +Since \(\lambda > 0\) , we have the following: + +<|ref|>equation<|/ref|><|det|>[[380, 179, 617, 218]]<|/det|> +\[w X_{b}H_{b}X_{b}^{T}w< \frac{1}{\eta} w X_{t}H_{t}X_{t}^{T}w\] + +<|ref|>text<|/ref|><|det|>[[113, 232, 884, 293]]<|/det|> +Of note is that both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) are positive semidefinite (i.e., \(w X_{t}H_{t}X_{t}^{T}w\geq 0\) and \(w X_{b}H_{b}X_{b}^{T}w\geq 0\) ). As a result, we have + +<|ref|>equation<|/ref|><|det|>[[398, 306, 652, 337]]<|/det|> +\[\lim_{\eta \to \infty}w X_{b}H_{b}X_{b}^{T}w = 0\] + +<|ref|>text<|/ref|><|det|>[[113, 351, 883, 409]]<|/det|> +Thus, as \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues belong to the null space of \(X_{b}H_{b}X_{b}^{T}\) . + +<|ref|>sub_title<|/ref|><|det|>[[114, 457, 202, 475]]<|/det|> +## Remark 2 + +<|ref|>text<|/ref|><|det|>[[113, 491, 883, 580]]<|/det|> +As \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues of the eigenproblem (Eq. [12]) contain the most discriminant information. We can rewrite the eigenvalue problem (Eq. [12]) as + +<|ref|>equation<|/ref|><|det|>[[366, 597, 630, 714]]<|/det|> +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w\] \[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w + \lambda\] \[\frac{w X_{t}H_{t}X_{t}^{T}w}{w X_{b}H_{b}X_{b}^{T}w}\to \infty\] + +<|ref|>text<|/ref|><|det|>[[113, 729, 883, 784]]<|/det|> +Thus, as \(\eta \to \infty\) , the eigenvectors corresponding to the positive eigenvalues contain the most discriminant information. + +<|ref|>sub_title<|/ref|><|det|>[[114, 834, 202, 852]]<|/det|> +## Remark 3 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 144]]<|/det|> +As \(\eta \to \infty\) , the eigenvectors corresponding to the zero eigenvalues of the eigenproblem (Eq. [12]) contain no discriminant information. When \(\lambda = 0\) , the eigenvalue problem reduces to + +<|ref|>equation<|/ref|><|det|>[[354, 159, 642, 180]]<|/det|> +\[(X_{t}H_{t}X_{t}^{T} - \eta X_{b}H_{b}X_{b}^{T})w = \lambda w = 0\] + +<|ref|>equation<|/ref|><|det|>[[381, 196, 614, 216]]<|/det|> +\[w X_{t}H_{t}X_{t}^{T}w = \eta w X_{b}H_{b}X_{b}^{T}w\] + +<|ref|>text<|/ref|><|det|>[[113, 231, 639, 253]]<|/det|> +Since \(w X_{t}H_{t}X_{t}^{T}w\) and \(w X_{b}H_{b}X_{b}^{T}w\) are finite and \(\eta \to \infty\) , we have + +<|ref|>equation<|/ref|><|det|>[[342, 268, 654, 290]]<|/det|> +\[w X_{t}H_{t}X_{t}^{T}w = 0,\qquad w X_{b}H_{b}X_{b}^{T}w = 0\] + +<|ref|>text<|/ref|><|det|>[[112, 339, 885, 572]]<|/det|> +Thus, the eigenvectors corresponding to the zero eigenvalues contain no discriminant information, as \(\eta \to \infty\) . In general, we can conclude that CoCo- ST derives its discriminant feature vectors from the range spaces of both \(X_{t}H_{t}X_{t}^{T}\) and \(X_{b}H_{b}X_{b}^{T}\) . The parameter \(\eta\) can be used to balance the contribution from the two spaces. Moreover, by extracting the eigenvectors of the eigenvalue problem in Eq. (12) corresponding to the largest positive eigenvalues, CoCo- ST can capture the most discriminant information in both the background and target ST data sets, enabling effective identification of the interesting spatial structures enriched in the target ST data set. + +<|ref|>sub_title<|/ref|><|det|>[[114, 620, 390, 639]]<|/det|> +## Nonlinear extension of CoCo-ST + +<|ref|>text<|/ref|><|det|>[[113, 654, 885, 852]]<|/det|> +Thus far, we have focused on linear feature representation. However, biological data are well known to be complex and highly nonlinear. Therefore, we extended CoCo- ST to perform nonlinear feature representation in a reproducing kernel Hilbert space \(\mathcal{H}\) , which gives rise to nonlinear CoCo- ST. We considered nonlinear mapping \(\phi (\cdot)\) of both the background \(X_{b}\) and target \(X_{t}\) ST data sets from the original input spaces to \(\mathcal{H}\) . Let \(\Phi_{b}\) and \(\Phi_{t}\) denote the background and target ST data sets in \(\mathcal{H}\) : + +<|ref|>equation<|/ref|><|det|>[[358, 867, 637, 893]]<|/det|> +\[\Phi_{b} = \left[\phi (x_{1}^{b}),\phi (x_{2}^{b}),\dots ,\phi (x_{n_{b}}^{b})\right]^{T}\] + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[363, 88, 633, 118]]<|/det|> +\[\Phi_{t} = \left[\phi (x_{1}^{t}),\phi (x_{2}^{t}),\dots ,\phi (x_{n_{t}}^{t})\right]^{T}\] + +<|ref|>text<|/ref|><|det|>[[112, 131, 884, 186]]<|/det|> +Denote by \(V\) the projection matrix in \(\mathcal{H}\) . The corresponding objective function ( \(\mathcal{O}_{3}\) ) of CoCo- ST in \(\mathcal{H}\) is + +<|ref|>equation<|/ref|><|det|>[[303, 202, 717, 232]]<|/det|> +\[\mathcal{O}_{4} = \max_{V^{T}V = I}tr(V^{T}\Phi_{t}H_{t}\Phi_{t}^{T}V) - \eta tr(V^{T}\Phi_{b}H_{b}\Phi_{b}^{T}V)\] + +<|ref|>text<|/ref|><|det|>[[290, 248, 331, 265]]<|/det|> +(14). + +<|ref|>text<|/ref|><|det|>[[112, 281, 540, 301]]<|/det|> +Let \(N = n_{b} + n_{t}\) , and define the data \(q_{1},q_{2},\dots,q_{N}\) by + +<|ref|>equation<|/ref|><|det|>[[390, 315, 608, 362]]<|/det|> +\[q_{i} = \left\{ \begin{array}{l l}{x_{i}^{t},} & {i f 1\leq i\leq n_{t}}\] \[x_{i - n_{t}}^{b},} & {o t h e r w i s e} \end{array} \right.\] + +<|ref|>text<|/ref|><|det|>[[112, 375, 884, 435]]<|/det|> +Since the projection vectors \(\nu_{1},\nu_{2},\dots,\nu_{p}\) (column vectors in \(V\) ) are linear combinations of \(\phi (q_{1}),\phi (q_{2}),\dots,\phi (q_{N})\) , coefficients \(\alpha_{i},i = 1,2,\dots,N\) exist such that + +<|ref|>equation<|/ref|><|det|>[[393, 448, 604, 540]]<|/det|> +\[\nu_{k} = \sum_{i = 1}^{N}\alpha_{i}\phi (q_{i}) = \Phi_{c}\alpha\] \[\Rightarrow V = \Phi_{c}A\] + +<|ref|>text<|/ref|><|det|>[[112, 551, 884, 607]]<|/det|> +where \(\alpha = (\alpha_{1},\alpha_{2},\dots,\alpha_{N})^{T}\in R^{N}\) , \(\mathrm{A} = [\alpha^{1},\alpha^{2},\dots,\alpha^{p}]\) . Following some algebraic formulations, we can rewrite the objective function ( \(\mathcal{O}_{4}\) ) in the following equivalent form: + +<|ref|>equation<|/ref|><|det|>[[226, 620, 770, 700]]<|/det|> +\[\mathcal{O}_{4} = \max_{\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{c}\mathrm{A} = I}tr(\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{t}H_{t}\Phi_{t}^{T}\Phi_{c}\mathrm{A}) - \eta tr(\mathrm{A}^{T}\Phi_{c}^{T}\Phi_{b}H_{t}\Phi_{b}^{T}\Phi_{c}\mathrm{A})\] \[\qquad = \max_{\mathrm{A}^{T}K_{c c}\mathrm{A} = I}tr(\mathrm{A}^{T}K_{c t}H_{t}K_{t c}\mathrm{A}) - \eta tr(\mathrm{A}^{T}K_{c b}H_{b}K_{b c}\mathrm{A})\] + +<|ref|>text<|/ref|><|det|>[[290, 717, 331, 735]]<|/det|> +(15), + +<|ref|>text<|/ref|><|det|>[[112, 749, 884, 907]]<|/det|> +where \(K_{c c} = \Phi_{c}^{T}\Phi_{c}\) , \(K_{c t} = \Phi_{c}^{T}\Phi_{t}\) , \(K_{t c} = \Phi_{t}^{T}\Phi_{c}\) , \(K_{c b} = \Phi_{c}^{T}\Phi_{b}\) , and \(K_{b c} = \Phi_{b}^{T}\Phi_{c}\) are the kernel matrices. Several choices of the kernel functions are available, including the polynomial kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \big((x_{t}^{t})^{T}x_{t}^{b} + 1\big)^{d}\) ; Gaussian kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \exp \left(-\frac{\left\|x_{t}^{t} - x_{t}^{b}\right\|^{2}}{\sigma^{2}}\right)\) ; and sigmoid kernel \(\mathrm{K}\big(x_{t}^{t},x_{t}^{b}\big) = \big((x_{t}^{t})^{T}x_{t}^{b} + \gamma \big)\) . + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 883, 179]]<|/det|> +Following approach similar to that in linear CoCo- ST, the projection vectors in Eq. (15) can be obtained as the eigenvectors corresponding to the top \(p\) largest eigenvalues of the generalized eigenvalue problem + +<|ref|>equation<|/ref|><|det|>[[348, 193, 648, 214]]<|/det|> +\[(K_{c t}H_{t}K_{t c} - \eta K_{c b}H_{b}K_{b c})\mathrm{A} = \Lambda K_{c c}\mathrm{A}\] + +<|ref|>text<|/ref|><|det|>[[113, 263, 884, 392]]<|/det|> +To obtain a stable solution of the eigenvalue problem in Eq. (16), the kernel matrix \(K_{cc}\) must be nonsingular. When \(K_{cc}\) is singular, we can adopt the idea of regularization by adding a small constant value \(\rho\) to the diagonal of \(K_{cc}\) as \(K_{cc} + \rho I\) for any \(\rho > 0\) . The matrix \(K_{cc} + \rho I\) is nonsingular, and the projection vectors can be computed as the generalized eigenvectors of + +<|ref|>equation<|/ref|><|det|>[[348, 405, 711, 459]]<|/det|> +\[(K_{c t}H_{t}K_{t c} - \eta K_{c b}H_{b}K_{b c})\mathrm{A} = \Lambda (K_{c c} + \rho I)\mathrm{A}\] \[(17).\] + +<|ref|>sub_title<|/ref|><|det|>[[115, 511, 236, 529]]<|/det|> +## Animal model + +<|ref|>text<|/ref|><|det|>[[113, 544, 885, 846]]<|/det|> +Wild- type mice (strain #009104; \(n = 12,9\mathrm{S4}\) ) were purchased from The Jackson Laboratory and housed in colony cages under pathogen- free conditions at The University of Texas MD Anderson Cancer Center Research Animal Support Facility. The mice were housed at an ambient temperature of \(20 - 26^{\circ}\mathrm{C}\) and humidity range of \(30 - 70\%\) with a 12- h light- dark cycle. All animal experiments were conducted following MD Anderson Institutional Animal Care and Use Committee- approved protocols (approval number 00001217- RN03). For carcinogen- induced mouse models, a urethane- induced mouse model was used. Specifically, the 12,9S4 wild- type mice described above received intraperitoneal injections of \(1\mathrm{mg / g}\) (body weight) urethane three times over 8 days when they were 6 weeks old. The mice were killed 7, 14, 20, 30, and 40 weeks after + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 883, 144]]<|/det|> +urethane administration, with a 0- week time point for mice that received no treatment. Both normal lung and lung tumor tissue samples were collected from the mice for downstream analysis. + +<|ref|>sub_title<|/ref|><|det|>[[115, 194, 410, 213]]<|/det|> +## Single-cell sequencing and analysis + +<|ref|>text<|/ref|><|det|>[[112, 222, 886, 888]]<|/det|> +Fresh normal lung and lung tumor tissue samples collected from mice were immediately cut into pieces and placed in RPMI 1640 medium (Thermo Fisher Scientific) with \(10\%\) fetal bovine serum (FBS; Gibco). The tissue samples were enzymatically digested using a tumor dissociation mixture composed of \(1\mathrm{mg / ml}\) collagenase A (Sigma), \(0.4\mathrm{mg / ml}\) hyaluronidase (Sigma), and 1:5 bovine serum albumin fraction V (Thermo Fisher Scientific) according to the manufacturers' instructions. Dissociation of tissue was carried out for \(2\mathrm{h}\) on a rotary shaker at \(37^{\circ}\mathrm{C}\) until all large tissue fragments were digested. Next, the dissociated tissues were transferred to conical tube and centrifuged at \(350\mathrm{g}\) for \(5\mathrm{min}\) . The supernatant was removed, and 1- 5 ml of prewarmed trypsin- EDTA was added to the collagenase/hyaluronidase- dissociated cells, resuspending them. Subsequently, \(10\mathrm{ml}\) of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS was added and centrifuged at \(350\mathrm{g}\) for \(5\mathrm{min}\) . As much of the supernatant as possible was collected, and \(5\mathrm{ml}\) of prewarmed \(5\mathrm{U / ml}\) dispase (STEMCELL Technologies) and \(50\mu \mathrm{l}\) of DNase I solution (10 \(\mathrm{mg / ml}\) in \(0.15\mathrm{M}\) NaCl; STEMCELL Technologies) were added. The samples were pipetted for 1 min using a 1- ml micropipettor to further dissociate cell clumps. The cell suspension was diluted with an additional \(10\mathrm{ml}\) of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS, and the cell suspension was filtered through a \(40\mathrm{- }\mu \mathrm{m}\) Falcon cell strainer (Thermo Fisher Scientific) into a \(50\mathrm{- }\mathrm{ml}\) tube. The cell suspension was further centrifuged at \(450\mathrm{g}\) for \(5\mathrm{min}\) , and the supernatant was discarded. The pellet was resuspended in a 1:4 mixture of cold RPMI 1640 without phenol red supplemented with \(2\%\) FBS and an ammonium chloride solution (STEMCELL + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 886, 388]]<|/det|> +Technologies), which was followed by centrifugation at \(450g\) for 5 min and discarding of the supernatant. Ten microliters of the cell suspension for each sample was analyzed using an automated cell counter (Thermo Fisher Scientific) to determine the number of live cells. Throughout the dissociation procedure, cells were kept on ice when possible. The cells were then loaded onto a Chromium single- cell controller (10x Genomics) to create single- cell gel beads in an emulsion according to the manufacturer's protocol. ScRNA- seq libraries were constructed using a Single Cell 5' Library and Gel Bead Kit v3.1 (10x Genomics) and sequenced using a NovaSeq 6000 sequencer (Illumina) at the Genomic and RNA Profiling Core at Baylor College of Medicine. + +<|ref|>sub_title<|/ref|><|det|>[[116, 437, 342, 456]]<|/det|> +## Tissue preparation and ST + +<|ref|>text<|/ref|><|det|>[[112, 472, 886, 880]]<|/det|> +Normal and tumor tissue samples from mouse lungs were fixed in \(10\%\) formalin at room temperature for 24- 48 h using a fixative volume 5- 10 times greater than that of the tissue volume. Fixed tissues were transferred to \(70\%\) ethanol for temporary storage at \(4^{\circ}\mathrm{C}\) . Paraffin embedding was conducted by the MD Anderson Research Histology Core Laboratory. Formalin- fixed, paraffin- embedded blocks were cut into \(10 - \mu \mathrm{m}\) - thick sections using a precooled RNase- free microtome. These sections were then transferred onto Visium Spatial Gene Expression slides (10x Genomics), which were pretreated via floating in a water bath at \(43^{\circ}\mathrm{C}\) . Following sectioning, the slides were dried at \(42^{\circ}\mathrm{C}\) in a SimpliAmp Thermal Cycler (Thermo Fisher Scientific) for \(3\mathrm{~h}\) according to the manufacturer's instructions. The slides were placed in a slide mailer, sealed with thermoplastic (Parafilm: Thermo Fisher Scientific), and stored overnight in a refrigerator at \(4^{\circ}\mathrm{C}\) . The slides were then deparaffinized, fixed, stained with hematoxylin and eosin, and imaged at \(5\mathrm{x}\) magnification using a DM5500 B microscope (Leica Microsystems). Tile scans of the entire array + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 249]]<|/det|> +were acquired using Leica Application Suite X software and merged. Spatial gene expression libraries (Visium ST; 10x Genomics) were processed according to the manufacturer's instructions and sequenced using a NovaSeq 6000 sequencer (Illumina). All hematoxylin and eosin staining, imaging, library preparation, and sequencing processes were carried out at the Genomic and RNA Profiling Core at Baylor College of Medicine. + +<|ref|>sub_title<|/ref|><|det|>[[114, 299, 251, 317]]<|/det|> +## Data processing + +<|ref|>text<|/ref|><|det|>[[113, 332, 885, 596]]<|/det|> +ScRNA- seq data. Raw base call files were analyzed using Cell Ranger v.3.0.2 software (10x Genomics). The mkfastq command was used to generate FASTQ files, and the count command was used to generate raw gene- barcode matrices aligned to the GRCh38 Ensembl 93 genome. The data were aggregated using the cellranger aggr command, and further downstream analysis was conducted in R version 4.1.0 using the Seurat package (v.4.1.1). To ensure our analysis was performed using high- quality cells, filtering of cells was conducted by retaining cells that had unique feature counts greater than 200 or less than 5000 and had mitochondrial content less than \(15\%\) . After removing doublets, the total cell number was 70,698. + +<|ref|>text<|/ref|><|det|>[[113, 647, 885, 875]]<|/det|> +ST data. The ST data sets were processed using Space Ranger software (v.2.0.1; 10x Genomics). The spatial sequencing data were aligned to mouse pre- mRNA genome reference version mm10 (downloaded from the 10x Genomics website) using Space Ranger, and mRNA count matrices were generated by adding intronic and exonic reads for each gene in each location. Paired histological hematoxylin and eosin stained images of tissues were processed using Space Ranger to select locations covered by tissue by aligning prerecorded spot locations with fiducial border spots in the images. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[114, 125, 230, 143]]<|/det|> +## Data analysis + +<|ref|>text<|/ref|><|det|>[[112, 155, 886, 741]]<|/det|> +ScRNA- seq analysis. The scRNA- seq data were first normalized, and the 2000 most highly variable genes in the data were identified using variance- stabilizing transformation implemented in the Seurat package. Data were then scaled, and the first 30 principal components were extracted. The principal components were further transformed into the UMAP embedding space for which clustering analysis was conducted. The original Louvain algorithm was used for modularity optimization. The resulting 14 clusters were visualized in a 2D UMAP representation and annotated to known biological cell types using canonical marker genes. The following cell types were annotated (selected markers are listed in parentheses): endothelial cells (Pecam1, Vwf, Ets1, Ace, Eng, Cldn5, and Mcam), epithelial cells (Epcam, Muc1, Cdh1, Krt7, and Krt8), fibroblasts (Pdpn, Dcn, Col3a1, Mgp, Col1a1, and Col6a1), macrophages (Apoe, C1qa, C1qb, C1qc, Marco, Mrc1, Fabp4, Inhba, Ccl4, Cxcl10, Rsad2, and Herc6), conventional dendritic cells (cDC; H2-Aa, Ccr7, Flt3, Fscn1, and Cdec9a), proliferating macrophages (Mki67, Tubb5, and Tuba1b), B cells (Cd19, Ms4a1, Cd79a, Cd79b, and Blnk), T cells (Trbc2, Cd2, Cd3d, Cd3e, Cd3g, Cd4, Cd8a, Cd8b1, Il2ra, and Foxp3), proliferating T cells (Mki67, Tubb5, and Tuba1b), plasmacytoid dendritic cells (pDC; Siglech, Ly6c2, and Cd209d), neutrophils (S100a8, S100a9, and Csf3r), plasma cells (Sdc1, Mzb1, Xbp1, and Jchain), monocytes (Cd14, Fcgr4, Lst1, and Vcan), and natural killer cells (Nkg7, Klrg1, and Ncr1). + +<|ref|>text<|/ref|><|det|>[[114, 785, 884, 876]]<|/det|> +ST analysis. The raw expression count matrices for both the background and target ST data sets were normalized using variance- stabilizing transformation implemented in the Seurat package. The normalized data were then standardized to have zero mean and unit standard deviation. The + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 249]]<|/det|> +standardized expression data matrices with 3000 genes were then used as inputs to our CoCo- ST method for low- dimensional feature representation. Clustering on the UMAP- embedded learned contrastive feature representations was then performed. Further differential gene expression analysis was conducted, and spatial domains were annotated based on the differentially expressed marker genes. + +<|ref|>sub_title<|/ref|><|det|>[[115, 299, 262, 317]]<|/det|> +## Pathway analysis + +<|ref|>text<|/ref|><|det|>[[112, 333, 886, 631]]<|/det|> +The most important genes (the 20 genes with the largest weights) on the top five contrastive components were identified, and the biological processes associated with these contrastive components were examined. Specifically, gene set enrichment analysis was performed with these 20 genes with the largest weights in the loading matrix using the g:GOSf function in the gprofiler2 package. In this analysis, all of the input 3000 genes were used as the background, and the default options in the g:SCS method in gprofiler2 were used for multiple testing correction. The gene sets were downloaded from the Molecular Signatures Database, including the KEGG, Gene Ontology biological processes, Gene Ontology cellular components, and Gene Ontology molecular functions. + +<|ref|>sub_title<|/ref|><|det|>[[115, 682, 315, 700]]<|/det|> +## Cell type deconvolution + +<|ref|>text<|/ref|><|det|>[[113, 715, 885, 876]]<|/det|> +Cell type deconvolution in ST enables estimation of cell type composition on each spatial location by leveraging a reference scRNA- seq data set. Cell type deconvolution was performed using the RCTD \(^{23}\) method implemented in the spacexr R package. ScRNA- seq data for the same mouse lung tumor samples (MLP samples) served as the reference data for deconvolution. The reference data contained 70,698 cells of multiple immune and malignant types as described in the scRNA- seq + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 144]]<|/det|> +analysis section. The RCTD method was run in doublet mode to estimate the reference cell type composition on each spatial location. Other parameters were set to the default settings. + +<|ref|>sub_title<|/ref|><|det|>[[115, 194, 283, 212]]<|/det|> +## Cell-cell interaction + +<|ref|>text<|/ref|><|det|>[[113, 227, 885, 494]]<|/det|> +Cell- cell interaction for the ST data sets was performed using CellChat24. The CellChatDB.mouse database of ligand- receptor interactions specifically curated for mice was used to identify overexpressed ligand- receptor interactions. The group- level communication probability or interaction weights were then computed using the truncated mean method with a \(10\%\) truncated mean. Subsequently, the communication probability at the signaling pathway level was computed by summarizing the communication probabilities of all ligand- receptor interactions associated with each signaling pathway. Finally, the cell- cell communication network was aggregated by summarizing the overall communication probabilities. + +<|ref|>sub_title<|/ref|><|det|>[[115, 543, 362, 562]]<|/det|> +## Trajectory inference analysis + +<|ref|>text<|/ref|><|det|>[[113, 576, 885, 841]]<|/det|> +For spatial trajectory analysis of individual tissue samples, the low- dimensional contrastive feature representations were used as inputs to the Slingshot algorithm25. Slingshot was applied to the contrastive feature representations so that nearby tissue spatial locations with similar gene expression would have similar pseudotimes. Because Slingshot requires predefined cluster labels, the spatial domain labels from the spatial domain identification analysis were used for Slingshot. The normal lung spatial domain was set as the start cluster (beginning of the trajectory or pseudotime) with a focus on trajectory inference on tumor and tumor- adjacent spatial domains to determine how these locations are connected to one another during tumorigenesis. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 459]]<|/det|> +For the trajectory analysis with combined tissue samples, spots belonging to normal lung, adenoma, and adenocarcinoma spatial domains as determined using the contrastive feature representations were collected, and Monocle \(^{36}\) was used to infer the trajectory. First, the combined data (spots) were processed using the standard Seurat approach, including total count normalization, scaling, and PCA analysis. Next, UMAP embedding was determined, which was used to learn the trajectory that fits the spots' UMAP coordinates. A principal graph was then fit on the UMAP embedding, and the spots were ordered according to their progress along the learned trajectory. To identify genes that varies among spot clusters in the UMAP embedding space, spatial autocorrelation analysis (Moran's I) was performed, and the obtained variable genes were grouped into modules by determining UMAP embedding of the genes followed by gene clustering based on Louvain community detection analysis. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 256, 108]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[115, 124, 883, 179]]<|/det|> +Data availabilityThe scRNA- seq and ST data sets analyzed in this study will be made available upon reasonable request through a data access agreement with the corresponding authors. + +<|ref|>sub_title<|/ref|><|det|>[[115, 228, 260, 247]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[115, 262, 883, 318]]<|/det|> +Code availabilityInstallation instructions and tutorials, together with the code used for data analysis and generating figures, can be found at https://github.com/WuLabMDA/CoCo- ST. + +<|ref|>sub_title<|/ref|><|det|>[[115, 368, 280, 386]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[113, 401, 885, 631]]<|/det|> +AcknowledgementsThis work was supported by generous philanthropic contributions to the MD Anderson Lung Cancer Moon Shot program as well as by the NIH/NCI under award number P30CA016672. This research was partially supported by NIH grants R01CA262425 and R01CA276178. Furthermore, this work was supported by generous philanthropic contributions from Andrea Mugnaini and Edward L. C. Smith. Finally, this work was supported by Rexanna's Foundation for Fighting Lung Cancer. We thank Don Norwood in the Research Medical Library at The University of Texas MD Anderson Cancer Center for editing this article. + +<|ref|>sub_title<|/ref|><|det|>[[115, 682, 296, 700]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[113, 715, 885, 876]]<|/det|> +Author contributionsM.A. and J.W. formulated and applied the method. B.Z. and J.Z. acquired the data. M.A. developed the software. M.A., B.Z., N.V., C.C., K.C., J.Z. and J.W. design the experiments. M.A., B.Z., H.C, N.V. and L.H. analyzed the data. All authors contributed to the interpretation of the data. M.A., B.Z. and H.C. prepared the first draft of the manuscript. L.H., N.V., C.C., K.C., J.Z. and J.W. revised the manuscript. J.Z. and J.W. supervised the project. All authors read and approved the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 89, 882, 143]]<|/det|> +final version of the manuscript. All authors were responsible for the final decision to submit the manuscript for publication. + +<|ref|>sub_title<|/ref|><|det|>[[115, 195, 285, 212]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[115, 230, 459, 247]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 209, 107]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[110, 120, 888, 905]]<|/det|> +1 Zhao, E. et al. Spatial transcriptomics at subspot resolution with BayesSpace. Nature biotechnology 39, 1375- 1384 (2021). 2 Bergensträhle, J., Larsson, L. & Lundeberg, J. Seamless integration of image and molecular analysis for spatial transcriptomics workflows. BMC Genomics 21, 482, doi:10.1186/s12864- 020- 06832- 3 (2020). 3 Townes, F. W. & Engelhardt, B. E. Nonnegative spatial factorization applied to spatial genomics. Nature Methods 20, 229- 238, doi:10.1038/s41592- 022- 01687- w (2023). 4 Shang, L. & Zhou, X. Spatially aware dimension reduction for spatial transcriptomics. Nature Communications 13, 7203, doi:10.1038/s41467- 022- 34879- 1 (2022). 5 Velten, B. et al. Identifying temporal and spatial patterns of variation from multimodal data using MEFISTO. Nature methods 19, 179- 186 (2022). 6 Chen, T., Kornblith, S., Norouzi, M. & Hinton, G. in International conference on machine learning. 1597- 1607 (PMLR). 7 You, Y. et al. Graph contrastive learning with augmentations. Advances in neural information processing systems 33, 5812- 5823 (2020). 8 Wang, Y., Wang, J., Cao, Z. & Barati Farimani, A. Molecular contrastive learning of representations via graph neural networks. Nature Machine Intelligence 4, 279- 287 (2022). 9 Dang, H. et al. Cancer- associated fibroblasts are key determinants of cancer cell invasion in the earliest stage of colorectal cancer. Cellular and Molecular Gastroenterology and Hepatology 16, 107- 131 (2023). 10 Hu, J. et al. SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nature methods 18, 1342- 1351 (2021). 11 Yi, F., Jaffe, R. & Prochownik, E. V. The CCL6 chemokine is differentially regulated by c- Myc and L- Myc, and promotes tumorigenesis and metastasis. Cancer research 63, 2923- 2932 (2003). 12 Tigue, M. L. et al. Wnt signaling in the phenotype and function of tumor- associated macrophages. Cancer Research 83, 3- 11 (2023). 13 Schmoll, A. et al. Macrophage and cancer cell cross- talk via CCR2 and CX3CR1 is a fundamental mechanism driving lung cancer. American journal of respiratory and critical care medicine 191, 437- 447 (2015). 14 Garrido- Navas, C. et al. Cooperative and escaping mechanisms between circulating tumor cells and blood constituents. Cells 8, 1382 (2019). 15 Sarode, P., Schaefer, M. B., Grimminger, F., Seeger, W. & Savai, R. Macrophage and tumor cell cross- talk is fundamental for lung tumor progression: we need to talk. Frontiers in Oncology 10, 324 (2020). 16 Ge, Z. & Ding, S. The crosstalk between tumor- associated macrophages (TAMs) and tumor cells and the corresponding targeted therapy. Frontiers in oncology 10, 590941 (2020). 17 Allen Reference Atlas - Mouse Brain [brain atlas] 18 Wang, W. et al. in Proceedings of the IEEE/CVF International Conference on Computer Vision. 7303- 7313. 19 Miao, J., Yang, Z., Fan, L. & Yang, Y. in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8042- 8052. 20 Hadsell, R., Chopra, S. & LeCun, Y. in 2006 IEEE computer society conference on computer vision and pattern recognition (CVPR'06). 1735- 1742 (IEEE). 21 Liu, Y. et al. Simple contrastive graph clustering. IEEE Transactions on Neural Networks and Learning Systems (2023). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 260]]<|/det|> +22 Abid, A., Zhang, M. J., Bagaria, V. K. & Zou, J. Exploring patterns enriched in a dataset with contrastive principal component analysis. \*Nature communications\* 9, 2134 (2018).23 Cable, D. M. et al. Robust decomposition of cell type mixtures in spatial transcriptomics. \*Nature biotechnology\* 40, 517- 526 (2022).24 Jin, S. et al. Inference and analysis of cell- cell communication using CellChat. \*Nature communications\* 12, 1088 (2021).25 Street, K. et al. Slingshot: cell lineage and pseudotime inference for single- cell transcriptomics. \*BMC genomics\* 19, 1- 16 (2018).26 Cao, J. et al. The single- cell transcriptional landscape of mammalian organogenesis. \*Nature\* 566, 496- 502 (2019). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 249, 109]]<|/det|> +## Figure Legends + +<|ref|>text<|/ref|><|det|>[[112, 123, 880, 565]]<|/det|> +Fig. 1 | CoCo- ST identifies unique, interesting spatial structures enriched in ST data sets. a, Overview of the CoCo- ST workflow. b, A target ST tissue sample containing unique, interesting spatial structures annotated by a pathologist and spatial domains/regions identified using the different feature representation methods. c, Volcano plot of the most differentially expressed genes for the adenoma spatial domain identified by CoCo- ST. d, Volcano plot of the most differentially expressed genes for the adenoma spatial domain identified using the compared approaches. e, Spatial expression patterns for the most differentially expressed genes (Ctsh, Cxcl15, and Slc34a2) for the adenoma spatial domain identified using CoCo- ST. These genes had high expression patterns in both the larger and smaller (hotspot) adenoma spatial domains. f, Spatial expression pattern for the most differentially expressed gene (Trf) for the adenoma spatial domain identified using the compared approaches. This gene had high expression pattern only within the larger adenoma spatial domain, with no such pattern observed in the smaller (hotspot) region. + +<|ref|>text<|/ref|><|det|>[[112, 610, 870, 807]]<|/det|> +Fig. 2 | CoCo- ST's contrastive components marked interesting spatial structures enriched in ST data sets. a, Spatial patterns captured by the first five contrastive components of CoCo- ST. b, The top 20 genes with the largest weights on the corresponding first five contrastive components. Symbols to the right of the bars indicate the signs of the weights. c, Expression patterns for the top representative genes for each of the first five contrastive components. d, Spatial patterns captured by the first five components of the compared approaches. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 354, 109]]<|/det|> +## Supplementary Information + +<|ref|>text<|/ref|><|det|>[[113, 123, 879, 214]]<|/det|> +Supplementary InformationExtended Data Fig. 1 | Spatial domains identified on all MLP tissue samples using CoCo- ST's contrastive components. The similarity graphs for both the background and target ST data sets were constructed based on the molecular data sets. + +<|ref|>text<|/ref|><|det|>[[112, 262, 879, 598]]<|/det|> +Extended Data Fig. 2 | Differential gene expression analysis of detected spatial domains. a, UMAP embedding of the contrastive components determined using CoCo- ST on the target ST tissue sample. B, UMAP embedding of spotsshowing the expression of some of the most differentially expressed genes in different clusters identified using the contrastive feature representations from CoCo- ST. c, Violin plots of the expression levels for the most differentially expressed genes for the different spatial domains identified using CoCo- ST. d, Biological processes and pathways associated with the 10 most differentially expressed genes for the adenoma spatial domain detected using CoCo- ST. e, Violin plots of the expression levels for the most differentially expressed genes for the different spatial domains identified using the compared feature representation approaches. + +<|ref|>text<|/ref|><|det|>[[113, 647, 825, 702]]<|/det|> +Extended Data Fig. 3 | Biological processes and pathways associated with CoCo- ST's contrastive components. + +<|ref|>text<|/ref|><|det|>[[113, 750, 879, 840]]<|/det|> +Extended Data Fig. 4 | Spatial domains identified on all MLP tissue samples using CoCo- ST's contrastive components. The similarity graphs for both the background and target ST data sets were constructed based on spatial locations. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 875, 459]]<|/det|> +Extended Data Fig. 5 | Application of CoCo- ST's contrastive components to studying cell- cell interaction at different cancer stages. a, UMAP embedding of the scRNA- seq data set used as a reference for cell type deconvolution. b, Spatial domains identified in the MLP- 6 tissue sample using CoCo- ST's contrastive components. c, Cell type annotation on each of the spatial locations in MLP- 6 tissue sample as inferred by the RCTD deconvolution algorithm. d, Percentage of different cell types (y- axis) in the different spatial domains (x- axis) detected using CoCo- ST. e, Cell- cell interaction weight plot for MLP- 6 tissue sample. The thicker the line, the stronger the interaction between the cell types. f, Chord plot of the cell- cell interactions via canonical WNT signaling. g, Heat map of the communication probabilities for WNT signaling from senders (sources) to receivers (targets). h, Heat map of network centrality scores for WNT signaling highlighting the major signaling roles of the different cell groups. + +<|ref|>text<|/ref|><|det|>[[114, 506, 839, 561]]<|/det|> +Extended Data Fig. 6 | Predicted spatial distributions of major cell types in the MLP- 6 tissue sample. + +<|ref|>text<|/ref|><|det|>[[114, 609, 855, 701]]<|/det|> +Extended Data Fig. 7 | Distribution of different cell types in each spatial domain on the MLP- 6 tissue sample determined using CoCo- ST. The cell type percentages in each spatial domain add up to \(100\%\) . + +<|ref|>text<|/ref|><|det|>[[113, 749, 870, 876]]<|/det|> +Extended Data Fig. 8 | Aggregated cell- cell interaction plots on the combined MLP tissue samples containing the adenoma and adenocarcinoma spatial domains. a, Cell- cell interaction weight plot for the adenocarcinoma- related MLP tissue samples. b, Simplified cell- cell interaction plots for a showing signaling sent from each cell group. The thicker the line, the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 856, 178]]<|/det|> +stronger the communication. c, Cell- cell interaction weight plot for the adenoma- related MLP tissue samples. d, Simplified cell- cell interaction plots for c showing signaling sent from each cell group. The thicker the line, the stronger the communication. + +<|ref|>text<|/ref|><|det|>[[112, 226, 883, 597]]<|/det|> +Extended Data Fig. 9 | Application of CoCo- ST's contrastive components to trajectory inference (cancer evolution). a, Spatial trajectory inference based on CoCo- ST's determined contrastive components. The arrows indicate the direction of the trajectory, which points from the normal lung spatial domain to the adenoma spatial domain. b, Learned trajectory pseudotime, with red- to green- colored regions indicating tissue locations with low and high pseudotime. c, UMAP embedding of spots belonging to the combined normal, adenoma, and adenocarcinoma spatial domains as determined using CoCo- ST. d, Trajectory inference of the cancer evolution from normal tissue to adenoma to adenocarcinoma colored according to their corresponding pseudotimes. e, Heat map of gene modules containing differentially co- expressed genes that vary across the different stages of cancer as determined from the learned trajectory in d. f, Bar plot of the number of differentially co- expressed genes in each module in e. + +<|ref|>text<|/ref|><|det|>[[113, 645, 876, 877]]<|/det|> +Extended Data Fig. 10 | Application of CoCo- ST to a mouse brain ST data set. a, Spatial domains/regions identified on anterior and posterior mouse brain tissue samples based on CoCo- ST's contrastive components. b, Spatial patterns on the anterior mouse brain tissue sample captured by the first five contrastive components of CoCo- ST. c, The 20 genes with the largest weights on the first five contrastive components in b. Symbols to the right of the bars indicate the signs of the weights. d, Expression patterns for some representative genes in c. e, Spatial patterns on the posterior mouse brain tissue sample captured by the first five contrastive + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 853, 144]]<|/det|> +components of CoCo- ST. f, The 20 genes with the largest weights on the first five contrastive components in e. g, Expression patterns for some representative genes in f. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[100, 45, 888, 901]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 907, 150, 920]]<|/det|> +
Fig. 1
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 60, 880, 915]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[116, 919, 154, 931]]<|/det|> +
Fig. 2
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 90, 850, 954]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[163, 95, 899, 440]]<|/det|> + +<|ref|>image<|/ref|><|det|>[[163, 460, 866, 800]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[127, 813, 248, 826]]<|/det|> +
Extended Data Fig. 2
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[124, 87, 861, 415]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[92, 45, 821, 911]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[99, 88, 900, 560]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[121, 80, 896, 800]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[131, 815, 252, 828]]<|/det|> +
Extended Data Fig. 6
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[320, 170, 911, 860]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[95, 100, 920, 850]]<|/det|> + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[133, 92, 925, 587]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[125, 610, 248, 622]]<|/det|> +
Extended Data Fig. 9
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 70, 844, 930]]<|/det|> + +<--- Page Split ---> diff --git a/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/images_list.json b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..a5459fe2d6058514525a72d6c4c60143297e201e --- /dev/null +++ b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/images_list.json @@ -0,0 +1,32 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 | PiggyBac bioprospecting.", + "footnote": [], + "bbox": [ + [ + 125, + 80, + 936, + 640 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2 | Synthetic mega-active PiggyBac generation via protein language model finetuning", + "footnote": [], + "bbox": [ + [ + 132, + 167, + 850, + 476 + ] + ], + "page_idx": 6 + } +] \ No newline at end of file diff --git a/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8.mmd b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b6ae41f52737198666904be5872e5ff192a3c062 --- /dev/null +++ b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8.mmd @@ -0,0 +1,296 @@ + +# Discovery and language model-guided design of hyperactive transposase + +Marc Güell + +marc.gue11@upf.edu + +Pompeu Fabra University https://orcid.org/0000- 0003- 4000- 7912 + +Dimitrije Ivančić Pompeu Fabra University + +Alejandro Agudelo Integra Therapeutics https://orcid.org/0009- 0008- 5515- 3975 + +Jonathan Lindstrom- Vautri Integra Therapeutics + +Jessica Jaraba- Wallace Integra Therapeutics + +Maria Gallo Pompeu Fabra University + +Alejandro Ragel Integra Therapeutics + +Irene Higueras Pompeu Fabra University + +Federico Billcei Integra Therapeutics + +Marta Sanvicente Integra Therapeutics + +Paolo Petazzi Integra Therapeutics + +Noelia Ferruz Centre for Genomic Regulation https://orcid.org/0000- 0003- 4172- 8201 + +Avencia Sánchez- Mejías Integra Therapeutics + +Ravi Das Integra Therapeutics + +<--- Page Split ---> + +## Keywords: + +Posted Date: December 9th, 2024 + +DOI: https://doi.org/10.21203/rs.3.rs- 5536951/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: Yes there is potential Competing Interest. AA, JLV, JJW, Maria G, MS, RD, MG, ASM, NF and DI are employed or have consulted for Integra Therapeutics. MG and ASM are shareholders of Integra therapeutics. DI, MG, ASM, AA and RD have filed patents related to this work. + +Version of Record: A version of this preprint was published at Nature Biotechnology on October 2nd, 2025. See the published version at https://doi.org/10.1038/s41587-025-02816-4. + +<--- Page Split ---> + +# Discovery and language model-guided design of hyperactive transposases + +Dimitrije Ivaničić2,4,*, Alejandro Agudelo2,4,*, Jonathan Lindstrom-Vautrin4, Jessica Jaraba-Wallace4, Maria Gallo4, Ravi Das4, Alejandro Ragel4, Irene Higueras4, Federico Billeci4, Marta Sanvicente4, Paolo Petazzi4, Noelia Ferruz1, Avencia Sánchez-Mejías4, Marc Güell2,3,4,* + +1 Center for Genomic Regulation, Barcelona, Spain + +2 Department of Medicine and Life Sciences, Universitat Pompeu Fabra, Barcelona, Spain + +3 ICREA, Institució Catalana de Recerca i Estudis Avançats, Barcelona, Spain + +4 Integra Therapeutics, Barcelona, Spain + +*These authors contributed equally + ++Correspondence: marc.guell@upf.edu (Marc Güell), dimitrie.ivancic@upf.edu (Dimitrije Ivaničić) + +<--- Page Split ---> + +## Abstract + +AbstractThe PiggyBac transposase gene writing system has been efficiently used across biotechnological applications, however its diversity and biochemical potential remain largely unexplored. By developing a eukaryotic transposon mining pipeline, we expand the known diversity by two orders of magnitude and experimentally validate a subset of highly divergent PiggyBacs. We then fine tune a protein language model to further expand PiggyBac sequence space and discover transposons with improved activity, compatible with T- cell engineering and Cas9- guided programmable transposition. Our work illustrates how combining bioprospecting and Al- driven sequence exploration can accelerate the discovery of novel eukaryotic gene writing tools. + +## Main + +MainThe advancement of genome engineering technologies has transformed biological engineering and opened new avenues for therapeutic and biotechnological applications1. Central to these developments are tools that enable efficient insertion of large DNA sequences into target genomes, an essential capability to unlock the full potential of synthetic biology2,3. DNA transposons have been widely adapted for genome modification across numerous organisms4,5. Among these tools, the PiggyBac transposase has gained significant attention due to its ability to integrate substantial DNA cargo across diverse cellular environments, making it a highly versatile platform for gene writing6, with the possibility of precise gene integration via a Cas9 fusion7. + +Synthetic biology has been traditionally restricted to bioprospecting in Nature followed by refactoring and optimization. With the development of generative AI methods applied to protein design, sampled natural diversity can be augmented to generate functional sequences not seen in nature8- 10. For instance, a combination of RFdiffusion11 and novel methodologies to design catalytic sites has created active synthetic serine hydrolases with new folds12. Interestingly, a protein large language model has been used recently to generate a CRISPR- cas9, a remarkably complex multidomain protein which binds to DNA and RNA, that does not exist in nature but that functions well for gene editing applications8. The development of such models has opened very exciting opportunities to expand biodiversity beyond the natural catalog, and leverage generative AI to create gene writers for biotechnology with higher activities. + +Previous studies have experimentally identified and characterized active PiggyBac transposases in insect and bat genomes14,15. While few piggyBacs have been phylogenetically identified across multiple Eukaryotic families16, much of their evolutionary diversity and biochemical utility remain untapped. To address this knowledge gap, our study undertakes an extensive exploration of PiggyBac transposon diversity through phylogenetic mining across all publicly available eukaryotic genomes, discovering over 13,000 previously unknown PiggyBac elements and report novel domain acquisition in multiple PiggyBac clusters. We validated a selection of these elements experimentally, identifying 10 active transposases, with up to 30% sequence identity to one another, thereby expanding the functional repertoire of known PiggyBac elements. We further augmented natural diversity and efficiency by fine- tuning a protein language model to generate "mega- active" synthetic variants of the widely used laboratory- evolved HyPB transposase, demonstrating the applicability of these novel PiggyBac orthologs in critical gene editing contexts, such as primary T- cell engineering and programmable transposition guided by CRISPR- Cas9. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1 | PiggyBac bioprospecting.
+ +a) PiggyBac identification and testing pipeline overview. (See Supplementary Figure 1 and methods for detailed pipeline) b) PiggyBac phylogenetic tree from the 2.5k identified clusters at 0.6 identity. Cluster size is represented by circle radius on top of tree leaves, and number of unique taxonomic species present in the cluster is shown by circle color. Tree rings represent 1) 5 identified PB classes, 2) major cluster taxonomic groups, 3) clusters with more than 1 broad taxonomic group, 4) CRD classification and 5) clusters with fusion domains. Tested piggyBac clusters are marked in orange triangles and green triangles indicate functional transposons found. These sequences were chosen to encompass all five major PB groups, both primary CRD types (HC6H and C5HC2), and a representative range of taxonomic groups. c) Experimental validation of PiggyBac orthologs, by RFP transposon payload integration in HEK293T cells two weeks after transfection, in presence (PB+TP, pink) or absence (TP, green) of transposase plasmid. d) Sequence identity heatmap between active orthologs from Figure 1c. e) Effect of N-term phosphorylation mutations on excision f) Targeted DNA integration with orthologs measured with junction qPCR. g) Species containing top two PB hits. + +A total of 273,643 piggyBac (PB) transposon ORFs together with their DNA sequences were retrieved by searching and annotating all available eukaryotic genome assemblies in + +<--- Page Split ---> + +the NCBI database (31,565) and appending this dataset with PiggyBac sequences from the DFAM database (20,638) \(^{17}\) (Figure 1a, Supplementary Figure 1). Domestication of transposon domains is a source of host genetic innovation \(^{18}\) with multiple PB- derived proteins reported to encode for host functions beyond transposition \(^{19}\) . To generate a dataset of transposition competent elements, we filtered genome retrieved sequences by: presence of DDE domain, conserved cysteine- rich domain (CRD), terminal inverted repeats (TIR), and a target site duplication (TSD) with the TTAA motif (Supplementary Figure 1). These motifs are reported to be crucial for DNA excision and integration \(^{15}\) . Filtering yielded 116,216 piggyBac sequences that resulted in 13,693 PiggyBac subfamilies after clustering at 80% sequence identity. + +The eukaryotic distribution of piggyBac transposons is notably diverse, encompassing taxa from fungi and plants to mammals (Figure 1b, Supplementary Figure 2b); it is predominantly represented in insects, \((\sim 60\%)\) , followed by fish and molluscs \((5\%)\) . We identified five main piggyBac Groups (Figure 1b, Supplementary Figure 3) based on main tree phylogenetic branches, taxonomic distribution, and the conserved cysteine- rich domain (CRD) types. More than 200 clusters are represented by more than 1 broad taxonomic group (Figure 1b, ring 3) indicating widespread horizontal gene transfer (HGT) across groups, as previously reported in other transposable elements \(^{20}\) . Group 4 has a unique unexpected taxonomic distribution with presence in fungi, land plants and algae (Figure 1b, ring 1, purple). We also observed "superhost" species, characterized by containing numerous piggyBacs. Just the top 3 superhosts captured \(9.5\%\) of all PiggyBac diversity (Supplementary Figure 4). Additionally, we found multiple domain acquisition events at both N and C- term with \(4.6\%\) of all the reported clusters containing a fusion domain, and N- term fusions being more predominant (Figure 1b, ring 5). DNA binding domains and fusogens were the most abundant acquired domains (Supplementary Figure 5), suggesting multiple novel transposition mechanisms for DNA recognition and cell entry. + +We used structural clustering to further understand the diversity of the CRD domain. We identified two main CRD cross brace zinc finger folds, HC6H and C5HC2 (Supplementary Figure 3c). In contrast to C5HC2, the HC6H group is longer and retains two unique beta sheets in its integration domain, possibly contributing to specific interactions with TIR. Analysis of the catalytic DDE domain, in contrast, indicates high structural conservation (RMSD of the catalytic region near 2 Å and a TM score of 0.915) despite high sequence divergence (Supplementary Figure 3a). + +To explore the potential of bioprospected transposon diversity for gene writing, we selected 23 representative PB sequences across the phylogenetic tree for experimental testing (Figure 1b, colored triangles). These sequences were chosen to encompass all five major PB groups, both primary CRD types and a representative range of taxonomic groups. Transposition activity was validated through detecting excision of the donor plasmid (Supplementary Figure 6) and long term integration of an RFP- containing transposon payload in HEK293T cells (Figure 1c). Of the tested sequences, 9 (\~40%) had detectable activity, with two sequences equivalent to laboratory evolved Hyperactive PiggyBac \(^{5}\) (HyPB). Active sequences were spread across phylogeny, and had low sequence identity to HyPB (Figure 1d). This broad distribution of active elements across taxonomic and CRD diversity underscores the potential of PB transposons as versatile tools in genetic engineering and gene transfer applications. With the aim to build a model that could predict active transposons, we used the tested dataset to build a classifier of active transposon sequences using an Xtreme gradient boosting method lightGBM \(^{50}\) and the tested PB + +<--- Page Split ---> + +orthologs (average AUC 0.85 over 5 Crossfold Validation) (Supplementary Figure 7b). To further improve transposon activity, we identified and removed CKII phosphorylation motifs in the N- terminal of piggyBac previously reported to inhibit its transposition activity in HyPB23 (Supplementary Figure 7c). CKII site removal increased transposition activity in both orthologs (Figure 1e). + +![](images/Figure_2.jpg) + +
Figure 2 | Synthetic mega-active PiggyBac generation via protein language model finetuning
+ +a) Overview of the fine-tuning and sequence generation pipeline: Progen2-base model was fine-tuned on a set of over 10,000 piggyBac orthologs identified through our bioprospecting efforts. Over 100,000 sequences were generated with a sequence identity between 35-99% to the hyperactive piggyBac. Sequences were then filtered using a set of basic (grey) and piggyBac specific (green) metrics, and scored using a set of structural (orange) and deep learning based scores (blue) to select a final subset of 22 sequences for experimental validation. b) Distribution of four key metrics (sequence identity, plddt, ProteinMPNN score and ESM1v score) for natural sequences from the HyPB cluster at 60% identity (orange) and sequences generated from our progen-ft model (blue) post-filtering. c) Relative excision for progen-ft generated variants normalized to HyPB WT activity (highlighted in green). Bars reflect the mean relative excision over the four trials and points represent the mean relative excision of replicates in each trial. d) Correlations between calculated and measured features to relative excision of the progen-ft generated variants. Significant correlations are highlighted in dark blue. e) Precise integration with top LLM generated mutants, measured with a fluorescence reporter cell line containing 2/2 GFP. + +Next, we sought to explore how the generated corpus of natural sequences could be used to improve the activity of existing transposases. We fine- tuned the ProGen2- base language model12 using over 10,000 bioprospected sequences, similarly to previously described for Cas9 nucleases8. We spiked in HyPB sequences to bias the model towards improvement of this particular target sequence. We then generated over 100,000 sequences from these two models prompted with the first 50 (N- C) or last 50 (C- N) amino acids. Sequences were first filtered based on a set of basic protein properties in addition to PB specific properties (Figure 2a, Supplementary Figure 10a). We further filtered sequences by structural (pLDDT, RMSD to experimental structure, SURFMAP44 and TM scores) and deep learning scores (Progen perplexity, ProteinMPNN43 and ESM1v41). Generated sequences had better plddt, ESM1v and ProteinMPNN scores when compared to a matched subset of natural + +<--- Page Split ---> + +sequences (Figure 3b). Interestingly, these metrics have previously been used for computational scoring of enzymes \(^{13}\) . + +We experimentally tested 11 sequences from each model (22 total), between 15 and 54 mutations apart from the original sequence (Supplementary Figure 11). All of the generated sequences displayed excision activity with an average \(\% \mathrm{RFP}\) ranging from \(15 - 48\%\) (Supplementary Figure 10b). Of the tested sequences, 7/22 were significantly more active than laboratory evolved HyPB (Figure 2c) (Mann- Whitney U test with a p- value cutoff of 0.05) + +To evaluate the relevance of the proposed sequence improvement approach, we both tested ESM1v selected "zero- shot" single mutants and bioprosected sequences near Poetur sequence space (Supplementary Figure 8). None of these approaches lead to mutants with increased activity. + +We gathered multiple metrics to both inform our selection and post- hoc learning of properties associated with transposase activity. We found that net charge of the protein, charged fraction of amino acids and ProteinMPNN \(^{43}\) score seem to be positively correlated to protein activity. In contrast, perplexity score from the N- C fine- tuned model, used model (N- C or C- N), and Wimley white \(^{42}\) surface structural similarity score to be negatively correlated (Figure 2d). + +We then tested top hits for programmable integration (Figure 2e, Supplementary Figure 9). We found that synthetic sequence 657 improved targeted integration three fold, demonstrating that improved LLM generated sequences are compatible with programmable gene writing. To illustrate the potential impact of bioprospecting and LLM guided sequence discovery for therapeutic applications, we stably delivered a GFP transposon cargo with Poeturb and Angra transposases in T- cells (Figure 2f). + +Our work significantly expands the phylogenetic tree PiggyBac transposons by two orders of magnitude, unveiling a previously unexplored diversity within this family of mobile genetic elements. This expansion has led to the discovery and characterization of 9 new active PiggyBac orthologs, broadening the range of transposase variants available for research and biotechnological applications. Among these newly identified orthologs, two stand out for their exceptional performance, demonstrating activity levels comparable to those of evolved hyperactive PiggyBac variants. and robust activity in primary T- cells, an essential target for many therapeutic applications in gene and cell therapy. Importantly, the newly discovered orthologs are compatible with the FiCAT programmable gene writing system. FiCAT integrates the site- specific DNA targeting precision of CRISPR- Cas systems with the transposition capabilities of PiggyBac transposons \(^{7}\) . This compatibility paves the way for innovative approaches to gene writing, enhancing the system's versatility in applications ranging from gene therapy to synthetic biology. Furthermore, we exemplified how transposases orthologs can be enhanced through advanced computational tools. Large Language Model (LLM) de novo sequence generation offers a powerful approach to improving transposase activities. This method not only accelerates the optimization process but also ensures that the modifications are informed by a comprehensive understanding of sequence- function relationships. By leveraging the predictive capabilities of LLMs, researchers could use the described method to systematically identify novel variants with enhanced properties. + +Our findings underscore the power of combining bioprospecting with AI- driven sequence optimization to accelerate the discovery and enhancement of next- generation gene writing + +<--- Page Split ---> + +tools. This approach not only expands the PiggyBac toolkit but also provides a valuable framework for the development of new gene modification tools for precise and efficient genome manipulation applicable across biotechnology and therapeutic fields. + +## Methods + +Retrieval of piggyBac Transposons. Complete PB transposon sequences were gathered from all available eukaryotic genomes in the NCBI database28 (15.3 terabytes, 31,565 genomes) and all PB in the DFAM database29 (20,638). DFAM sequences were directly downloaded by selecting entries labeled as piggyBac, as the transposons in this database are fully annotated with both the transposase and its associated DNA sequence. NCBI- derived transposase sequences were identified using Bath25,26, with a custom HMM model constructed from all active PB sequences reported in the literature. Unlike DFAM, NCBI results contain only the transposase coding sequence; thus, flanking regions of 4kbp upstream and downstream were included to capture the complete transposon sequence. An initial filtering was performed in which only PB with a transposase longer than 250 AA were selected. After this filtering, a total of 273,643 PB were recovered, with a mean transposase length of 500 residues and mean transposon length of 3298 bp. + +To refine the boundaries of each transposon in the NCBI dataset, we clustered the DDE domains of the PB hits at a 0.9 similarity threshold with MMseqs234, followed by multiple sequence alignment (MSA) of the complete DNA sequences (including flanking regions) within clusters using ClustalW27. Transposon boundaries were then delimited based on the MSA results. + +Filtering for Active piggyBac Elements. To identify active PB transposons, we applied the following sequential filters: + +1. DDE Domain Identification: The presence of a DDE domain was confirmed using RPS-BLAST30, with the Conserved Domain Database (CDD)31 as the reference database and selecting only sequences with a DDE domain longer than 250 amino acids. + +2. CRD Identification: Initially, candidate CRDs were identified based on the presence of at least five cysteines and one histidine, key residues for the zinc finger configuration, in the last 70 residues of the transposase. However, this initial approach produced a high false-positive rate. To improve specificity, 50 representative CRDs were manually curated and structurally modeled using AlphaFold32 to identify residues directly involved in zinc ion coordination. Based on this curated set, we derived a set of sequence motifs (Supplementary Table 2), revealing major CRD groups and their variants. CRDs were then identified using regular expressions matching these curated motifs. + +3. TIR Identification: Terminal inverted repeats (TIRs) were identified in the flanking DNA regions using the EMBOSS tool Palindrome33, focusing on pairs of palindromic sequences located on opposite flanks of the transposon in the first and last 200bp. We retained only TIRs with at least 2 palindromic sequences of 10 bp or longer and allowing up to two mismatches, consistent with the fact that TIRs in the piggyBac transposons are imperfect. As an additional quality control step, only palindromes in which the two most common nucleotides account for less than 80% of the palindrome were selected. + +<--- Page Split ---> + +4. TSD Identification: Target site duplications (TSDs) were searched for with regular expression within the first and last 50 bp of each transposon, using the motif TTAACC, with up to two allowed mismatches. + +A total of 116,216 were recovered after applying the filtering process. The code for the refinement of transposon boundaries and active PB filtering can be found in the github repository. + +Dataset clustering. The filtered dataset was then clustered in order to reduce redundancy with the DDE domain of the transposase, as it is the most conserved domain. We performed two clusterings with MMseqs2, one at 0.8 id and one at 0.6 id. The 0.8 clustering was done following transposon annotation 80- 80- 8024, as it is considered two TE elements belong to the same family if they share 80% (or more) sequence identity in at least 80% of their coding or internal domain. And the clustering at 0.6 was done in order to make a broader classification of PB families. The clustering at 0.8 produced 13,693 clusters, while the 0.6 produced 2,572. + +Analysis of bioprosected sequences. The phylogenetic tree was built with IQ- TREE v.1.6.1235, based on an MSA generated with the 2,572 centroids from the 0.6 clustering with MUSCLE36. Model finder37 was employed to select the optimal model for accurate phylogenetic estimation (LG+R10), and UFBoot38 was used for bootstrap approximation for 1000 replicates. The resulting tree was visualized using iTOL39. Additional PB domains were identified in the same way as the DDE domain. Molecular graphics were generated using UCSF Chimera40. + +Blast identification of Poetur orthologs. A blast search with blastn on the core nucleotide database was done using the highest performing PB, Poetur, on the NCBI blast website (https://blast.ncbi.nlm.nih.gov/Blast.cgi). The whole transposon, including the TIR and TSD were included in order to find hits that also possessed these motifs. A total of 4 hits from 4 different species were manually selected based on them having a coverage higher than 88%, sequence identity higher than 83% and the presence of all the necessary functional domains for transposition activity (DDE,CRD,TIR,TSD). + +Model Fine- tuning. The ProGen2- base language model12 of 764 million parameters was fine- tuned on over 10,000 sequences from the PB orthologs clustered at 0.8. The models were fine- tuned using the full sequences excluding the N- terminal domain, which is an extremely variable domain. The sequences were split using a 80:20 train:test split. In addition to the set of orthologous sequences used in the training, additional wild- type HyPB sequences (5- 10) were added to the training set to bias the model towards HyPB. This allowed us to generate sequences in a closer sequence identity range to HyPB than we were able to otherwise. Fine- tuning was performed using the Trainer module from Hugging Face over 2 epochs with a training batch size of 4 and evaluation batch size of 8. A constant learning rate of 5.0E- 5 was used and the model was evaluated after every 2000 steps with the highest scoring model on the evaluation set being used for sequence generation. The remaining parameters were kept to the default values. A full exploration of fine- tuning parameters was not done, as with these fairly standard parameters, we were able to generate convincing sequences with our desired properties. + +Plasmid assembly. Transposase ORF amino acid sequences were codon optimized for Homo Sapiens and ordered and synthesized as gene fragments to TWIST biosciences. Gene fragments were cloned into a CMV based expression vector by Golden Gate assembly + +<--- Page Split ---> + +using Esp3l restriction enzyme. Transposon (cargo vector) plasmid sequences were defined as the first 150bp from the transposon ends from both 5' and 3' ITR sequences and synthesized as gene fragments by TWIST biosciences with added overhangs for golden gate assembly. EF1a RFP polyA expression cassette was included between the ITRs. + +Transposition activity assays. Hek293T cells, obtained from Thermo Fisher Scientific, were cultured in Dulbecco's modified eagle medium (DMEM) supplemented with high glucose (Gibco, Thermo Fisher), \(10\%\) Fetal Bovine Serum (FBS), \(2 \text{mM}\) glutamine, and 100 U penicillin/0.1 mg/mL streptomycin at \(37^{\circ} \text{C}\) in a \(5\%\) CO2 incubator. For transfection experiments, cells were treated with Polyethyleneimine (PEI, Thermo Fisher Scientific) at a 1:3 DNA- PEI ratio in OptiMem. Prior to transfection, cells were seeded to achieve \(70\%\) confluency the following day, typically using 120,000 cells per adherent p24 well plate. Plasmid DNA was mixed at a ratio of 1 transposase:3 gRNA:5 transposon, with 0.035 pmols of transposase used per p24 well plate. + +The expression of the ITR vector RFP was assessed two days and twenty days post- transfection using cell cytometry with the Cytek Aurora™ CS System. The RFP signal at day twenty was considered indicative of stable transgene integration and was utilized to determine the integration efficiency for each of the tested systems. + +Fluorescent- Excision activity assays. HEK293T cells (Thermo Fisher Scientific) were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco, Thermo Fisher) supplemented with \(10\%\) fetal bovine serum (FBS), \(2 \text{mM}\) L- glutamine, 100 U/mL penicillin, and \(0.1 \text{mg/mL}\) streptomycin. Cells were maintained in a \(37^{\circ} \text{C}\) incubator with \(5\%\) CO2. For each experiment, cells were seeded in 24- well plates at a density of 120,000 cells per well 24 hours prior to transfection to ensure approximately \(70\%\) confluency on the day of transfection. Transfections were performed in 24- well plate using Polyethyleneimine (PEI, Thermo Fisher Scientific) at a 1:3 DNA- PEI ratio in Opti- MEM (Thermo Fisher). A dual- component transfection system was employed, consisting of a plasmid encoding the transposase and a second plasmid containing a disrupted mCherry reporter sequence flanked by transposase recognition sites. Upon excision by the transposase, the mCherry sequence was restored, allowing fluorescence to correlate with excision efficiency (Supplementary Figure 12). For each well, two solutions were prepared separately in Opti- MEM: one containing the plasmid DNA mix (transposase and transposon plasmids at a 1:3 ratio, with a total of 0.035 pmol of transposase) and another containing PEI. The two solutions were then mixed, incubated for 20 minutes at room temperature, and added dropwise to the cells. After 72 hours post- transfection, cells were harvested, and mCherry reporter expression was assessed by flow cytometry using the Cytek Aurora™ CS System. + +Al Sequence Generation. In both models, 50 wild- type amino acids from HyPB were used to prompt sequence generation. For the N- C model, the first 50 amino acids after the N- terminal domain were used and in the C- N model, the final 50 amino acids of the CRD were used to prompt sequence generation. For the C- N model sequences were generated 'backwards' and then reversed to have the standard directionality. The maximum sequence length for both models was set to 500 amino acids and a temperature of \(T = 0.5\) and nucleus probability \(P = 0.95\) were used. + +Al Sequence Filtering. The generated sequences first went through a set of three basic filters. First, duplicated sequences were removed. Second, sequences with non- canonical amino acids were removed. And third, sequences were filtered using a k- mer repetition filter so that no amino acid motif of 6/4/3/2 residues was repeated 2/3/6/8 times consecutively. The next set of filters were HyPB specific and included testing for a PB CRD (based on the presence of at least 7 cysteine amino acids in the final 50 amino acids), sequence identity to WT (between 80- 95% to the DDE+CRD domains), and specific key residues including catalytic site, alpha bridge residues, hyperactive residues and another extensive set of key residues including DNA interacting residues. + +<--- Page Split ---> + +For all of these sequences we calculated perplexity using the ProGen2- base model as well as the fine- tuned model responsible for generating a given sequence. For a subset of sequences that passed our filters, structures were predicted using ESMFold46. Structures were then compared to the experimentally available PB structure47 (https://www.rcsb.org/structure/6x67) to extract RMSD and tm- scores using PyMOL48 and TMAIign49 respectively. Finally, structures were aligned to the experimental PB structure and several surface properties were calculated using SURFMAP44,45: a tool that projects surface residues from a protein structure into a 2D space and can calculate different amino acid residue properties. The 5 metrics we calculated using SURFMAP were stickiness, circular variance, wimley white, kyte doolittle and electrostatics. We then computed cosine similarities between each surface feature in the generated structures and the experimental structure. Finally ProteinMPNN43 and ESM1v41 scores were calculated. ProteinMPNN is a deep learning- based sequence design method that can decode amino acid sequences from structural representations of proteins. ProteinMPNN can also be used to generate a log- likelihood score for any given sequence. Wimley white is a measure of residue hydrophobicity42 which in this case has been applied to surface residues using SURFMAP: + +An additional set of filters was created to narrow down the 1,000 sequences to a more manageable number. Sequences were required to be in the top 75th percentile for both ProteinMPNN and ESM1v, sequences were filtered on length to exclude sequences that were too short, a conservative plddt filter of 90 was used, and an acceptable range for net charge of the proteins was established. After this, sequences were selected manually in an attempt to cover a range of sequence identities \(90 - 97\%\) to the entire wild- type hyPB sequence with high quality sequences. During this manual selection process, sequences with a higher proportion of the key residues were selected for and any sequences that had particularly bad scores in any of the calculated metrics were avoided. A final selection of 22 sequences was made. + +In silico deep mutational scan. Esm- 1v was used in a zero shot version where the Poecilopsis amino acid sequence was given as an input. Esm1- v creates a fitness score for all possible aminoacids for residue position by calculating a log odds ratio, assuming an additive model when multiple mutations exist. Then, the sum is made over the mutated positions and the sequence is masked at every mutated position41: + +\[\sum_{t\in T}\log p\left(x_{t} = x_{t}^{mt}|x\backslash_{T}\right) - \log p\left(x_{t} = x_{t}^{wt}|x\backslash_{T}\right)\] + +Variant prediction was run in google colab pro with one A- 100 80gb ram gpu. The script used to run the variant prediction can be found in https://github.com/Alejo945/ISHyPB. The output is a tsv with all possible variants and their scores. + +PCR- Excision activity assays. 48h post- transfection, cells were harvested and a miniprep protocol was performed using the NZYMiniprep kit from NZYtech to extract the remaining plasmids. Primers X and Y, which were flanking the ITRs in the transposon plasmids, were utilized to detect transposon excision. Expecting a 2900bps band indicated not- excised transposon, whereas a 1200bps band indicated excision of both ITR and transposon. + +Targeted Transposition activity assays. Triple mutant residue selection was performed by aligning the ortholog piggyBacs to Trichoplusia Ni PiggyBac mutated sequence. Plasmids encoding the triple mutant variants (PBx3) were co- transfected with Cas9, gRNA and transposon plasmids in a 1:1:3:5 molar ratio into 0.5M Hek23T cells seeded in a p6 plate the day before transfection. Cells were analyzed for RFP expression two days after transfection using cell cytometry with the Cytek Aurora™ CS System. Subsequently, two rounds of enrichment via RFP sorting were conducted with BD FACSAria (Biosciences), one week + +<--- Page Split ---> + +and two weeks after transfection. Non- enriched cells were parallelly maintained in culture to measure overall integration levels after 3 weeks. Genomic DNA was extracted using Quiagen DNeasy Blood & Tissue Kit column's four days after the second sorting. A 3' Junction PCR was performed and gel purification using the QIAquick Gel Extraction Kit was performed when necessary before Sanger sequencing the amplified bands. + +Transposition in T- cells. Peripheral Blood Mononuclear Cells (PBMCs) from two different donors, isolated from buffy coats and cryopreserved, were thawed and seeded on p24- coated plates containing anti- CD3/CD28 (1:1,000; BD Sciences) at a density of \(1 \times 10^{6}\) cells/ml in 3 ml of CTS™ OpTimizer™ T Cell Expansion SFM medium (Thermo Fisher), supplemented with IL- 7 and IL- 15 (10 ng/ml each; Miltenyi Biotec). + +On the third day of culture, cells were prepared for electroporation and divided into two groups based on the DNA plasmids: transposon only or transposon + transposase. Electroporation was conducted using the P3 Primary Cell 4D- Nucleofector™ X Kit (Lonza). Cells were washed with PBS (Capricorn) and adjusted to a concentration of \(7.5 \times 10^{6}\) cells per condition. The cell suspension was prepared in 20 μl of nucleofection buffer, consisting of 16.4 μl P3 Primary Cell Nucleofector Solution and 3.6 μl Supplement 1 (Lonza). Subsequently, 1 μg of each DNA plasmid was added to the suspension, and electroporation was carried out using the EO- 115 nucleofection program. + +Following electroporation, 80 μl of complete medium was added, and cells were incubated at \(37^{\circ}C\) for 20 minutes. The cells were then carefully resuspended and transferred to a fresh p24 plate containing 500 μl of medium for recovery and expansion. Approximately one- third of the well volume was used for flow cytometric analysis using the Aurora system (Cytek) to assess RFP expression across three time points, continuing until episomal decay was observed in the transposon- only condition. + +## Data availability + +Bioprospected and synthetically generated transposon sequence files are available in supplementary Table 1. Top active transposon and transposase plasmids have been deposited at addgene. + +Model finetuning and PiggyBac generation code is available at Github https://github.com/Integra- tx/Piggybac bioprospecting_pipeline. + +## Acknowledgements + +We thank Cedric Feschotte for feedback on the bioprospecting pipeline, and George M. Church for advice on approaches for ML/AI and directed evolution. + +## Funding + +Integra Therapeutics S.L. received funding from NEOTEC (CDTI, SNEO- 20222363). UPF received funding from UPGRADE (European Union Horizon 2020, grant agreement No 825825); Ministerio de Economia, Industria y Competitividad de España (Plan Estatal 2013- 2016 (Grant agreement No) + +## Author contributions + +DI, ASM and MG conceived the study. AA and DI designed the bioprospecting pipeline. AA implemented the bioprospecting pipeline. JLV implemented the LLM and fine tuning work with help from NF, AA and DI. DI and JJW designed the experiments with help from RD and Maria G. + +<--- Page Split ---> + +Maria G, RD and JJW performed cell experiments. IH assisted in sequence assembly. AR and PP performed T- cell work. FB contributed to genome data accession and zeroshot modeling. MS analysed targeted integration data. DI and MG supervised the study. AA, DI and JVL plotted the data. AA, DI, MG and JVL wrote the manuscript with contribution from all authors, specially NF. + +## Conflict of interest + +AA, JLV, JJW, Maria G MS, RD, MG, ASM, NF and DI are employed or have consulted for Integra Therapeutics. MG and ASM are shareholders of Integra therapeutics. DI, MG, ASM, AA and RD have filed patents related to this work. + +## References + +1. Wang, J. Y. & Doudna, J. A. CRISPR technology: A decade of genome editing is only the beginning. Science 379, eadd8643 (2023). +2. Yarnall, M. T. N. et al. Drag-and-drop genome insertion of large sequences without double-strand DNA cleavage using CRISPR-directed integrases. Nat. Biotechnol. 41, 500–512 (2023). +3. Mukhametzyanova, L. et al. Activation of recombinases at specific DNA loci by zinc-finger domain insertions. Nat. Biotechnol. (2024) doi:10.1038/s41587-023-02121-y. +4. Li, X. et al. piggyBac transposase tools for genome engineering. Proc. Natl. Acad. Sci. U. S. A. 110, E2279–87 (2013). +5. Yusa, K., Zhou, L., Li, M. A., Bradley, A. & Craig, N. L. A hyperactive piggyBac transposase for mammalian applications. Proc. Natl. Acad. Sci. U. S. A. 108, 1531–1536 (2011). +6. Yusa, K. PiggyBac transposon. Microbiol. Spectr. 3, MDNA3–0028–2014 (2015). +7. Pallarès-Masmitjà, M. et al. Find and cut-and-transfer (FiCAT) mammalian genome engineering. Nat. Commun. 12, 7071 (2021). +8. Ruffolo, J. A. et al. Design of highly functional genome editors by modeling the universe of CRISPR-Cas sequences. bioRxiv (2024) doi:10.1101/2024.04.22.590591. +9. Hayes, T. et al. Simulating 500 million years of evolution with a language model. bioRxiv (2024) doi:10.1101/2024.07.01.600583. +10. Alley, E. C., Khimulya, G., Biswas, S., AlQuraishi, M. & Church, G. M. Unified rational protein engineering with sequence-based deep representation learning. Nat. Methods 16, 1315–1322 (2019). +11. Watson, J. L. et al. De novo design of protein structure and function with RFdiffusion. Nature 620, 1089–1100 (2023). +12. Lauko, A. et al. Computational design of serine hydrolases. bioRxiv (2024) doi:10.1101/2024.08.29.610411. +13. Johnson, S. R. et al. Computational scoring and experimental evaluation of enzymes generated by neural networks. Nat. Biotechnol. (2024) doi:10.1038/s41587-024-02214-2. +14. Mitra, R. et al. Functional characterization of piggyBat from the bat Myotis lucifugus unveils an active mammalian DNA transposon. Proc. Natl. Acad. Sci. U. S. A. 110, 234–239 (2013). +15. Chen, Q. et al. Structural basis of seamless excision and specific targeting by piggyBac transposase. Nat. Commun. 11, 3446 (2020). +16. Yuan, Y.-W. & Wessler, S. R. The catalytic domain of all eukaryotic cut-and-paste transposase superfamilies. Proc. Natl. Acad. Sci. U. S. A. 108, 7884–7889 (2011). +17. Hubley, R. et al. The Dfam database of repetitive DNA families. Nucleic Acids Res. 44, D81–9 (2016). +18. Cosby, R. L. et al. Recurrent evolution of vertebrate transcription factors by transposase capture. Science 371, eabc6405 (2021). + +<--- Page Split ---> + +19. Bouallègue, M., Rouault, J.-D., Hua-Van, A., Makni, M. & Capy, P. Molecular evolution of piggyBac superfamily: From selfishness to domestication. Genome Biol. Evol. 9, 323-339 (2017).20. Zhang, H.-H., Peccoud, J., Xu, M.-R.-X., Zhang, X.-G. & Gilbert, C. Horizontal transfer and evolution of transposable elements in vertebrates. Nat. Commun. 11, 1362 (2020).22. Jangam, D., Feschotte, C., & Betrán, E. (2017). Transposable Element Domestication As an Adaptation to Evolutionary Conflicts. Trends in genetics: TIG, 33(11), 817-831.23. Luo, W., Hickman, A.B., Genzor, P., Ghirlando, R., Furman, C.M., Menshikh, A., 832 Haase, A., Dyda, F., & Wilson, M.H. Transposase N-terminal phosphorylation and 833 asymmetric transposon ends inhibit piggyBac transposition in mammalian cells. 83424. Wicker T, Sabot F, Hua-Van A, Bennetzen JL, Capy P, et al. 2007. A unified classification system for eukaryotic transposable elements. Nat. Rev. Genet 8(12):973-82.25. Sean R Eddy. Accelerated profile hmm searches. PLoS computational biology, 7(10):e1002195, 2011.26. Krause, G. R., Shands, W., & Wheeler, T. J. (2024). Sensitive and error-tolerant annotation of protein-coding DNA with BATH. Bioinformatics Advances, 4(1).27. Thompson, J. D., Higgins, D. G., & Gibson, T. J. (1994). CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic acids research, 22(22), 4673-4680.28. Paul A Kitts, Deanna M Church, Françoise Thibaud-Nissen, Jinna Choi, Vichet Hem, Victor Sapojnikov, Robert G Smith, Tatiana Tatusova, Charlie Xiang, Andrey Zherikov, et al. Assembly: a resource for assembled genomes at ncbi. Nucleic acids research, 44(D1):D73-D80, 2016.29. Storer, J., Hubley, R., Rosen, J., Wheeler, T. J., & Smit, A. F. (2021). The Dfam community resource of transposable element families, sequence models, and genome annotations. Mobile DNA, 12(1), 2.30. Camacho, C., Coulouris, G., Avagyan, V., Ma, N., Papadopoulos, J., Bealer, K., & Madden, T. L. (2009). BLAST+: architecture and applications. BMC bioinformatics, 10, 421.31. Lu, S., Wang, J., Chitsaz, F., Derbyshire, M. K., Geer, R. C., Gonzales, N. R., Gwadz, M., Hurwitz, D. I., Marchler, G. H., Song, J. S., Thanki, N., Yamashita, R. A., Yang, M., Zhang, D., Zheng, C., Lanczycki, C. J., & Marchler-Bauer, A. (2020). CDD/SPARCLE: the conserved domain database in 2020. Nucleic acids research, 48(D1), D265-D268.32. Abramson, J., Adler, J., Dunger, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493-500 (2024).33. Rice P., Longden I. and Bleasby A. EMBOSS: The European Molecular Biology Open Software Suite. Trends in Genetics. 2000 16(6):276-27734. Steinegger, M., Söding, J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nat Biotechnol 35, 1026-1028 (2017)35. B.Q. Minh, H.A. Schmidt, O. Chernomor, D. Schrempf, M.D. Woodhams, A. von + +<--- Page Split ---> + +Haeseler, R. Lanfear (2020) IQ- TREE 2: New models and efficient methods for phylogenetic inference in the genomic era. Mol. Biol. Evol., 37:1530- 1534 + +36. Edgar, R.C. MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics 5, 113 (2004) + +37. S. Kalyaanamoorthy, B.Q. Minh, T.K.F. Wong, A. von Haeseler, L.S. Jermiin (2017) ModelFinder: Fast model selection for accurate phylogenetic estimates. Nat. Methods, 14:587-589 + +38. D.T. Hoang, O. Chernomor, A. von Haeseler, B.Q. Minh, and L.S. Vinh (2018) UFBoot2: Improving the ultrafast bootstrap approximation. Mol. Biol. Evol., 35:518-522 + +39. Ivica Letunic and Peer Bork. Interactive tree of life (itol) v5: an online tool for phylogenetic tree display and annotation. Nucleic acids research, 49(W1):W293-W296, 2021 + +40. UCSF Chimera--a visualization system for exploratory research and analysis. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE. J Comput Chem. 2004 Oct;25(13):1605-12 + +41. Meier, J., Rao, R., Verkuil, R., Liu, J., Sercu, T., & Rives, A. (2021). Language models enable zero-shot prediction of the effects of mutations on protein function. bioRxiv. + +42. Wimley, W. C.; White, S. H. Experimentally Determined Hydrophobicity Scale for Proteins at Membrane Interfaces. Nat. Struct. Biol. 1996, 3 (10), 842-848. + +43. J. Dauparas et al., Robust deep learning-based protein sequence design using ProteinMPNN. Science. 2022, 378, 49-56 DOI:10.1126/science.add2187 + +44. Hugo Schweke, Marie-Hélène Mucchielli, Nicolas Chevrollier, Simon Gosset, Anne Lopes. SURFMAP: a software for mapping in two dimensions protein surface features. J. Chem. Inf. Model. 2022. + +45. Sanner, M. F., Olson A.J. & Spehner, J.-C. (1996). Reduced Surface: An Efficient Way to Compute Molecular Surfaces. Biopolymers 38:305-320. + +46. Zeming Lin et al, Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 2023, 379, 1123-1130. DOI:10.1126/science.ade2574 + +47. Chen, Q., Luo, W., Veach, R. A., Hickman, A. B., Wilson, M. H., & Dyda, F. (2020). Structural basis of seamless excision and specific targeting by piggyBac transposase. Nature Communications, 11, 3446. https://doi.org/10.1038/s41467-020-17128-1 + +48. PyMOL: The PyMOL Molecular Graphics System, Version 3.1 Schrödinger, LLC. + +49. Zhang, Y., Skolnick, J. (2005). TM-align: A protein structure alignment algorithm based on TM-score. Nucleic Acids Research, 33(7), 2302-2309. + +50. Ke, G. et al, (2017). Lightgbm: A highly efficient gradient boosting decision tree. Advances in Neural Information Processing Systems, 30, 3146-3154. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- supplementaryTable1.csv- supplementaryTable2.txt- supplementary.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8_det.mmd b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..6a9605bc87dc285f476b5a4b8dea9a2f8b2d7bdd --- /dev/null +++ b/preprint/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8/preprint__16a2c0672be713996d897b673cb59a9d347e678206aa0b9ad47f18b9af9896b8_det.mmd @@ -0,0 +1,399 @@ +<|ref|>title<|/ref|><|det|>[[44, 106, 883, 175]]<|/det|> +# Discovery and language model-guided design of hyperactive transposase + +<|ref|>text<|/ref|><|det|>[[44, 196, 155, 214]]<|/det|> +Marc Güell + +<|ref|>text<|/ref|><|det|>[[54, 223, 256, 240]]<|/det|> +marc.gue11@upf.edu + +<|ref|>text<|/ref|><|det|>[[44, 268, 636, 288]]<|/det|> +Pompeu Fabra University https://orcid.org/0000- 0003- 4000- 7912 + +<|ref|>text<|/ref|><|det|>[[44, 293, 275, 333]]<|/det|> +Dimitrije Ivančić Pompeu Fabra University + +<|ref|>text<|/ref|><|det|>[[44, 339, 598, 380]]<|/det|> +Alejandro Agudelo Integra Therapeutics https://orcid.org/0009- 0008- 5515- 3975 + +<|ref|>text<|/ref|><|det|>[[44, 385, 280, 425]]<|/det|> +Jonathan Lindstrom- Vautri Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 431, 253, 471]]<|/det|> +Jessica Jaraba- Wallace Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 477, 275, 517]]<|/det|> +Maria Gallo Pompeu Fabra University + +<|ref|>text<|/ref|><|det|>[[44, 523, 275, 563]]<|/det|> +Alejandro Ragel Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 570, 275, 610]]<|/det|> +Irene Higueras Pompeu Fabra University + +<|ref|>text<|/ref|><|det|>[[44, 616, 275, 656]]<|/det|> +Federico Billcei Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 662, 275, 702]]<|/det|> +Marta Sanvicente Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 708, 275, 748]]<|/det|> +Paolo Petazzi Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 754, 684, 795]]<|/det|> +Noelia Ferruz Centre for Genomic Regulation https://orcid.org/0000- 0003- 4172- 8201 + +<|ref|>text<|/ref|><|det|>[[44, 800, 258, 840]]<|/det|> +Avencia Sánchez- Mejías Integra Therapeutics + +<|ref|>text<|/ref|><|det|>[[44, 847, 235, 887]]<|/det|> +Ravi Das Integra Therapeutics + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 46, 137, 64]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 83, 338, 102]]<|/det|> +Posted Date: December 9th, 2024 + +<|ref|>text<|/ref|><|det|>[[43, 121, 474, 141]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 5536951/v1 + +<|ref|>text<|/ref|><|det|>[[43, 158, 916, 202]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 219, 951, 285]]<|/det|> +Additional Declarations: Yes there is potential Competing Interest. AA, JLV, JJW, Maria G, MS, RD, MG, ASM, NF and DI are employed or have consulted for Integra Therapeutics. MG and ASM are shareholders of Integra therapeutics. DI, MG, ASM, AA and RD have filed patents related to this work. + +<|ref|>text<|/ref|><|det|>[[42, 319, 914, 363]]<|/det|> +Version of Record: A version of this preprint was published at Nature Biotechnology on October 2nd, 2025. See the published version at https://doi.org/10.1038/s41587-025-02816-4. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 113, 875, 169]]<|/det|> +# Discovery and language model-guided design of hyperactive transposases + +<|ref|>text<|/ref|><|det|>[[118, 174, 880, 238]]<|/det|> +Dimitrije Ivaničić2,4,*, Alejandro Agudelo2,4,*, Jonathan Lindstrom-Vautrin4, Jessica Jaraba-Wallace4, Maria Gallo4, Ravi Das4, Alejandro Ragel4, Irene Higueras4, Federico Billeci4, Marta Sanvicente4, Paolo Petazzi4, Noelia Ferruz1, Avencia Sánchez-Mejías4, Marc Güell2,3,4,* + +<|ref|>text<|/ref|><|det|>[[118, 250, 544, 267]]<|/det|> +1 Center for Genomic Regulation, Barcelona, Spain + +<|ref|>text<|/ref|><|det|>[[118, 280, 860, 296]]<|/det|> +2 Department of Medicine and Life Sciences, Universitat Pompeu Fabra, Barcelona, Spain + +<|ref|>text<|/ref|><|det|>[[118, 309, 763, 325]]<|/det|> +3 ICREA, Institució Catalana de Recerca i Estudis Avançats, Barcelona, Spain + +<|ref|>text<|/ref|><|det|>[[118, 339, 460, 355]]<|/det|> +4 Integra Therapeutics, Barcelona, Spain + +<|ref|>text<|/ref|><|det|>[[118, 368, 408, 384]]<|/det|> +*These authors contributed equally + +<|ref|>text<|/ref|><|det|>[[118, 396, 880, 430]]<|/det|> ++Correspondence: marc.guell@upf.edu (Marc Güell), dimitrie.ivancic@upf.edu (Dimitrije Ivaničić) + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[118, 86, 243, 109]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[118, 112, 878, 251]]<|/det|> +AbstractThe PiggyBac transposase gene writing system has been efficiently used across biotechnological applications, however its diversity and biochemical potential remain largely unexplored. By developing a eukaryotic transposon mining pipeline, we expand the known diversity by two orders of magnitude and experimentally validate a subset of highly divergent PiggyBacs. We then fine tune a protein language model to further expand PiggyBac sequence space and discover transposons with improved activity, compatible with T- cell engineering and Cas9- guided programmable transposition. Our work illustrates how combining bioprospecting and Al- driven sequence exploration can accelerate the discovery of novel eukaryotic gene writing tools. + +<|ref|>sub_title<|/ref|><|det|>[[118, 272, 189, 294]]<|/det|> +## Main + +<|ref|>text<|/ref|><|det|>[[118, 298, 878, 436]]<|/det|> +MainThe advancement of genome engineering technologies has transformed biological engineering and opened new avenues for therapeutic and biotechnological applications1. Central to these developments are tools that enable efficient insertion of large DNA sequences into target genomes, an essential capability to unlock the full potential of synthetic biology2,3. DNA transposons have been widely adapted for genome modification across numerous organisms4,5. Among these tools, the PiggyBac transposase has gained significant attention due to its ability to integrate substantial DNA cargo across diverse cellular environments, making it a highly versatile platform for gene writing6, with the possibility of precise gene integration via a Cas9 fusion7. + +<|ref|>text<|/ref|><|det|>[[118, 445, 878, 612]]<|/det|> +Synthetic biology has been traditionally restricted to bioprospecting in Nature followed by refactoring and optimization. With the development of generative AI methods applied to protein design, sampled natural diversity can be augmented to generate functional sequences not seen in nature8- 10. For instance, a combination of RFdiffusion11 and novel methodologies to design catalytic sites has created active synthetic serine hydrolases with new folds12. Interestingly, a protein large language model has been used recently to generate a CRISPR- cas9, a remarkably complex multidomain protein which binds to DNA and RNA, that does not exist in nature but that functions well for gene editing applications8. The development of such models has opened very exciting opportunities to expand biodiversity beyond the natural catalog, and leverage generative AI to create gene writers for biotechnology with higher activities. + +<|ref|>text<|/ref|><|det|>[[118, 621, 878, 835]]<|/det|> +Previous studies have experimentally identified and characterized active PiggyBac transposases in insect and bat genomes14,15. While few piggyBacs have been phylogenetically identified across multiple Eukaryotic families16, much of their evolutionary diversity and biochemical utility remain untapped. To address this knowledge gap, our study undertakes an extensive exploration of PiggyBac transposon diversity through phylogenetic mining across all publicly available eukaryotic genomes, discovering over 13,000 previously unknown PiggyBac elements and report novel domain acquisition in multiple PiggyBac clusters. We validated a selection of these elements experimentally, identifying 10 active transposases, with up to 30% sequence identity to one another, thereby expanding the functional repertoire of known PiggyBac elements. We further augmented natural diversity and efficiency by fine- tuning a protein language model to generate "mega- active" synthetic variants of the widely used laboratory- evolved HyPB transposase, demonstrating the applicability of these novel PiggyBac orthologs in critical gene editing contexts, such as primary T- cell engineering and programmable transposition guided by CRISPR- Cas9. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[125, 80, 936, 640]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 664, 407, 680]]<|/det|> +
Figure 1 | PiggyBac bioprospecting.
+ +<|ref|>text<|/ref|><|det|>[[116, 678, 879, 857]]<|/det|> +a) PiggyBac identification and testing pipeline overview. (See Supplementary Figure 1 and methods for detailed pipeline) b) PiggyBac phylogenetic tree from the 2.5k identified clusters at 0.6 identity. Cluster size is represented by circle radius on top of tree leaves, and number of unique taxonomic species present in the cluster is shown by circle color. Tree rings represent 1) 5 identified PB classes, 2) major cluster taxonomic groups, 3) clusters with more than 1 broad taxonomic group, 4) CRD classification and 5) clusters with fusion domains. Tested piggyBac clusters are marked in orange triangles and green triangles indicate functional transposons found. These sequences were chosen to encompass all five major PB groups, both primary CRD types (HC6H and C5HC2), and a representative range of taxonomic groups. c) Experimental validation of PiggyBac orthologs, by RFP transposon payload integration in HEK293T cells two weeks after transfection, in presence (PB+TP, pink) or absence (TP, green) of transposase plasmid. d) Sequence identity heatmap between active orthologs from Figure 1c. e) Effect of N-term phosphorylation mutations on excision f) Targeted DNA integration with orthologs measured with junction qPCR. g) Species containing top two PB hits. + +<|ref|>text<|/ref|><|det|>[[118, 869, 877, 904]]<|/det|> +A total of 273,643 piggyBac (PB) transposon ORFs together with their DNA sequences were retrieved by searching and annotating all available eukaryotic genome assemblies in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 82, 879, 256]]<|/det|> +the NCBI database (31,565) and appending this dataset with PiggyBac sequences from the DFAM database (20,638) \(^{17}\) (Figure 1a, Supplementary Figure 1). Domestication of transposon domains is a source of host genetic innovation \(^{18}\) with multiple PB- derived proteins reported to encode for host functions beyond transposition \(^{19}\) . To generate a dataset of transposition competent elements, we filtered genome retrieved sequences by: presence of DDE domain, conserved cysteine- rich domain (CRD), terminal inverted repeats (TIR), and a target site duplication (TSD) with the TTAA motif (Supplementary Figure 1). These motifs are reported to be crucial for DNA excision and integration \(^{15}\) . Filtering yielded 116,216 piggyBac sequences that resulted in 13,693 PiggyBac subfamilies after clustering at 80% sequence identity. + +<|ref|>text<|/ref|><|det|>[[117, 270, 879, 547]]<|/det|> +The eukaryotic distribution of piggyBac transposons is notably diverse, encompassing taxa from fungi and plants to mammals (Figure 1b, Supplementary Figure 2b); it is predominantly represented in insects, \((\sim 60\%)\) , followed by fish and molluscs \((5\%)\) . We identified five main piggyBac Groups (Figure 1b, Supplementary Figure 3) based on main tree phylogenetic branches, taxonomic distribution, and the conserved cysteine- rich domain (CRD) types. More than 200 clusters are represented by more than 1 broad taxonomic group (Figure 1b, ring 3) indicating widespread horizontal gene transfer (HGT) across groups, as previously reported in other transposable elements \(^{20}\) . Group 4 has a unique unexpected taxonomic distribution with presence in fungi, land plants and algae (Figure 1b, ring 1, purple). We also observed "superhost" species, characterized by containing numerous piggyBacs. Just the top 3 superhosts captured \(9.5\%\) of all PiggyBac diversity (Supplementary Figure 4). Additionally, we found multiple domain acquisition events at both N and C- term with \(4.6\%\) of all the reported clusters containing a fusion domain, and N- term fusions being more predominant (Figure 1b, ring 5). DNA binding domains and fusogens were the most abundant acquired domains (Supplementary Figure 5), suggesting multiple novel transposition mechanisms for DNA recognition and cell entry. + +<|ref|>text<|/ref|><|det|>[[118, 560, 879, 682]]<|/det|> +We used structural clustering to further understand the diversity of the CRD domain. We identified two main CRD cross brace zinc finger folds, HC6H and C5HC2 (Supplementary Figure 3c). In contrast to C5HC2, the HC6H group is longer and retains two unique beta sheets in its integration domain, possibly contributing to specific interactions with TIR. Analysis of the catalytic DDE domain, in contrast, indicates high structural conservation (RMSD of the catalytic region near 2 Å and a TM score of 0.915) despite high sequence divergence (Supplementary Figure 3a). + +<|ref|>text<|/ref|><|det|>[[118, 696, 879, 908]]<|/det|> +To explore the potential of bioprospected transposon diversity for gene writing, we selected 23 representative PB sequences across the phylogenetic tree for experimental testing (Figure 1b, colored triangles). These sequences were chosen to encompass all five major PB groups, both primary CRD types and a representative range of taxonomic groups. Transposition activity was validated through detecting excision of the donor plasmid (Supplementary Figure 6) and long term integration of an RFP- containing transposon payload in HEK293T cells (Figure 1c). Of the tested sequences, 9 (\~40%) had detectable activity, with two sequences equivalent to laboratory evolved Hyperactive PiggyBac \(^{5}\) (HyPB). Active sequences were spread across phylogeny, and had low sequence identity to HyPB (Figure 1d). This broad distribution of active elements across taxonomic and CRD diversity underscores the potential of PB transposons as versatile tools in genetic engineering and gene transfer applications. With the aim to build a model that could predict active transposons, we used the tested dataset to build a classifier of active transposon sequences using an Xtreme gradient boosting method lightGBM \(^{50}\) and the tested PB + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 878, 161]]<|/det|> +orthologs (average AUC 0.85 over 5 Crossfold Validation) (Supplementary Figure 7b). To further improve transposon activity, we identified and removed CKII phosphorylation motifs in the N- terminal of piggyBac previously reported to inhibit its transposition activity in HyPB23 (Supplementary Figure 7c). CKII site removal increased transposition activity in both orthologs (Figure 1e). + +<|ref|>image<|/ref|><|det|>[[132, 167, 850, 476]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[118, 483, 857, 500]]<|/det|> +
Figure 2 | Synthetic mega-active PiggyBac generation via protein language model finetuning
+ +<|ref|>text<|/ref|><|det|>[[117, 510, 879, 705]]<|/det|> +a) Overview of the fine-tuning and sequence generation pipeline: Progen2-base model was fine-tuned on a set of over 10,000 piggyBac orthologs identified through our bioprospecting efforts. Over 100,000 sequences were generated with a sequence identity between 35-99% to the hyperactive piggyBac. Sequences were then filtered using a set of basic (grey) and piggyBac specific (green) metrics, and scored using a set of structural (orange) and deep learning based scores (blue) to select a final subset of 22 sequences for experimental validation. b) Distribution of four key metrics (sequence identity, plddt, ProteinMPNN score and ESM1v score) for natural sequences from the HyPB cluster at 60% identity (orange) and sequences generated from our progen-ft model (blue) post-filtering. c) Relative excision for progen-ft generated variants normalized to HyPB WT activity (highlighted in green). Bars reflect the mean relative excision over the four trials and points represent the mean relative excision of replicates in each trial. d) Correlations between calculated and measured features to relative excision of the progen-ft generated variants. Significant correlations are highlighted in dark blue. e) Precise integration with top LLM generated mutants, measured with a fluorescence reporter cell line containing 2/2 GFP. + +<|ref|>text<|/ref|><|det|>[[117, 717, 879, 907]]<|/det|> +Next, we sought to explore how the generated corpus of natural sequences could be used to improve the activity of existing transposases. We fine- tuned the ProGen2- base language model12 using over 10,000 bioprospected sequences, similarly to previously described for Cas9 nucleases8. We spiked in HyPB sequences to bias the model towards improvement of this particular target sequence. We then generated over 100,000 sequences from these two models prompted with the first 50 (N- C) or last 50 (C- N) amino acids. Sequences were first filtered based on a set of basic protein properties in addition to PB specific properties (Figure 2a, Supplementary Figure 10a). We further filtered sequences by structural (pLDDT, RMSD to experimental structure, SURFMAP44 and TM scores) and deep learning scores (Progen perplexity, ProteinMPNN43 and ESM1v41). Generated sequences had better plddt, ESM1v and ProteinMPNN scores when compared to a matched subset of natural + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 877, 118]]<|/det|> +sequences (Figure 3b). Interestingly, these metrics have previously been used for computational scoring of enzymes \(^{13}\) . + +<|ref|>text<|/ref|><|det|>[[118, 132, 879, 236]]<|/det|> +We experimentally tested 11 sequences from each model (22 total), between 15 and 54 mutations apart from the original sequence (Supplementary Figure 11). All of the generated sequences displayed excision activity with an average \(\% \mathrm{RFP}\) ranging from \(15 - 48\%\) (Supplementary Figure 10b). Of the tested sequences, 7/22 were significantly more active than laboratory evolved HyPB (Figure 2c) (Mann- Whitney U test with a p- value cutoff of 0.05) + +<|ref|>text<|/ref|><|det|>[[118, 250, 878, 319]]<|/det|> +To evaluate the relevance of the proposed sequence improvement approach, we both tested ESM1v selected "zero- shot" single mutants and bioprosected sequences near Poetur sequence space (Supplementary Figure 8). None of these approaches lead to mutants with increased activity. + +<|ref|>text<|/ref|><|det|>[[118, 333, 878, 436]]<|/det|> +We gathered multiple metrics to both inform our selection and post- hoc learning of properties associated with transposase activity. We found that net charge of the protein, charged fraction of amino acids and ProteinMPNN \(^{43}\) score seem to be positively correlated to protein activity. In contrast, perplexity score from the N- C fine- tuned model, used model (N- C or C- N), and Wimley white \(^{42}\) surface structural similarity score to be negatively correlated (Figure 2d). + +<|ref|>text<|/ref|><|det|>[[118, 450, 878, 555]]<|/det|> +We then tested top hits for programmable integration (Figure 2e, Supplementary Figure 9). We found that synthetic sequence 657 improved targeted integration three fold, demonstrating that improved LLM generated sequences are compatible with programmable gene writing. To illustrate the potential impact of bioprospecting and LLM guided sequence discovery for therapeutic applications, we stably delivered a GFP transposon cargo with Poeturb and Angra transposases in T- cells (Figure 2f). + +<|ref|>text<|/ref|><|det|>[[117, 567, 878, 869]]<|/det|> +Our work significantly expands the phylogenetic tree PiggyBac transposons by two orders of magnitude, unveiling a previously unexplored diversity within this family of mobile genetic elements. This expansion has led to the discovery and characterization of 9 new active PiggyBac orthologs, broadening the range of transposase variants available for research and biotechnological applications. Among these newly identified orthologs, two stand out for their exceptional performance, demonstrating activity levels comparable to those of evolved hyperactive PiggyBac variants. and robust activity in primary T- cells, an essential target for many therapeutic applications in gene and cell therapy. Importantly, the newly discovered orthologs are compatible with the FiCAT programmable gene writing system. FiCAT integrates the site- specific DNA targeting precision of CRISPR- Cas systems with the transposition capabilities of PiggyBac transposons \(^{7}\) . This compatibility paves the way for innovative approaches to gene writing, enhancing the system's versatility in applications ranging from gene therapy to synthetic biology. Furthermore, we exemplified how transposases orthologs can be enhanced through advanced computational tools. Large Language Model (LLM) de novo sequence generation offers a powerful approach to improving transposase activities. This method not only accelerates the optimization process but also ensures that the modifications are informed by a comprehensive understanding of sequence- function relationships. By leveraging the predictive capabilities of LLMs, researchers could use the described method to systematically identify novel variants with enhanced properties. + +<|ref|>text<|/ref|><|det|>[[118, 879, 877, 911]]<|/det|> +Our findings underscore the power of combining bioprospecting with AI- driven sequence optimization to accelerate the discovery and enhancement of next- generation gene writing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 877, 130]]<|/det|> +tools. This approach not only expands the PiggyBac toolkit but also provides a valuable framework for the development of new gene modification tools for precise and efficient genome manipulation applicable across biotechnology and therapeutic fields. + +<|ref|>sub_title<|/ref|><|det|>[[118, 153, 243, 175]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[117, 180, 878, 361]]<|/det|> +Retrieval of piggyBac Transposons. Complete PB transposon sequences were gathered from all available eukaryotic genomes in the NCBI database28 (15.3 terabytes, 31,565 genomes) and all PB in the DFAM database29 (20,638). DFAM sequences were directly downloaded by selecting entries labeled as piggyBac, as the transposons in this database are fully annotated with both the transposase and its associated DNA sequence. NCBI- derived transposase sequences were identified using Bath25,26, with a custom HMM model constructed from all active PB sequences reported in the literature. Unlike DFAM, NCBI results contain only the transposase coding sequence; thus, flanking regions of 4kbp upstream and downstream were included to capture the complete transposon sequence. An initial filtering was performed in which only PB with a transposase longer than 250 AA were selected. After this filtering, a total of 273,643 PB were recovered, with a mean transposase length of 500 residues and mean transposon length of 3298 bp. + +<|ref|>text<|/ref|><|det|>[[118, 370, 878, 448]]<|/det|> +To refine the boundaries of each transposon in the NCBI dataset, we clustered the DDE domains of the PB hits at a 0.9 similarity threshold with MMseqs234, followed by multiple sequence alignment (MSA) of the complete DNA sequences (including flanking regions) within clusters using ClustalW27. Transposon boundaries were then delimited based on the MSA results. + +<|ref|>text<|/ref|><|det|>[[118, 460, 877, 495]]<|/det|> +Filtering for Active piggyBac Elements. To identify active PB transposons, we applied the following sequential filters: + +<|ref|>text<|/ref|><|det|>[[147, 508, 878, 576]]<|/det|> +1. DDE Domain Identification: The presence of a DDE domain was confirmed using RPS-BLAST30, with the Conserved Domain Database (CDD)31 as the reference database and selecting only sequences with a DDE domain longer than 250 amino acids. + +<|ref|>text<|/ref|><|det|>[[147, 589, 878, 744]]<|/det|> +2. CRD Identification: Initially, candidate CRDs were identified based on the presence of at least five cysteines and one histidine, key residues for the zinc finger configuration, in the last 70 residues of the transposase. However, this initial approach produced a high false-positive rate. To improve specificity, 50 representative CRDs were manually curated and structurally modeled using AlphaFold32 to identify residues directly involved in zinc ion coordination. Based on this curated set, we derived a set of sequence motifs (Supplementary Table 2), revealing major CRD groups and their variants. CRDs were then identified using regular expressions matching these curated motifs. + +<|ref|>text<|/ref|><|det|>[[148, 757, 879, 894]]<|/det|> +3. TIR Identification: Terminal inverted repeats (TIRs) were identified in the flanking DNA regions using the EMBOSS tool Palindrome33, focusing on pairs of palindromic sequences located on opposite flanks of the transposon in the first and last 200bp. We retained only TIRs with at least 2 palindromic sequences of 10 bp or longer and allowing up to two mismatches, consistent with the fact that TIRs in the piggyBac transposons are imperfect. As an additional quality control step, only palindromes in which the two most common nucleotides account for less than 80% of the palindrome were selected. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[147, 83, 878, 135]]<|/det|> +4. TSD Identification: Target site duplications (TSDs) were searched for with regular expression within the first and last 50 bp of each transposon, using the motif TTAACC, with up to two allowed mismatches. + +<|ref|>text<|/ref|><|det|>[[118, 149, 878, 202]]<|/det|> +A total of 116,216 were recovered after applying the filtering process. The code for the refinement of transposon boundaries and active PB filtering can be found in the github repository. + +<|ref|>text<|/ref|><|det|>[[117, 215, 878, 354]]<|/det|> +Dataset clustering. The filtered dataset was then clustered in order to reduce redundancy with the DDE domain of the transposase, as it is the most conserved domain. We performed two clusterings with MMseqs2, one at 0.8 id and one at 0.6 id. The 0.8 clustering was done following transposon annotation 80- 80- 8024, as it is considered two TE elements belong to the same family if they share 80% (or more) sequence identity in at least 80% of their coding or internal domain. And the clustering at 0.6 was done in order to make a broader classification of PB families. The clustering at 0.8 produced 13,693 clusters, while the 0.6 produced 2,572. + +<|ref|>text<|/ref|><|det|>[[117, 368, 878, 488]]<|/det|> +Analysis of bioprosected sequences. The phylogenetic tree was built with IQ- TREE v.1.6.1235, based on an MSA generated with the 2,572 centroids from the 0.6 clustering with MUSCLE36. Model finder37 was employed to select the optimal model for accurate phylogenetic estimation (LG+R10), and UFBoot38 was used for bootstrap approximation for 1000 replicates. The resulting tree was visualized using iTOL39. Additional PB domains were identified in the same way as the DDE domain. Molecular graphics were generated using UCSF Chimera40. + +<|ref|>text<|/ref|><|det|>[[117, 503, 878, 623]]<|/det|> +Blast identification of Poetur orthologs. A blast search with blastn on the core nucleotide database was done using the highest performing PB, Poetur, on the NCBI blast website (https://blast.ncbi.nlm.nih.gov/Blast.cgi). The whole transposon, including the TIR and TSD were included in order to find hits that also possessed these motifs. A total of 4 hits from 4 different species were manually selected based on them having a coverage higher than 88%, sequence identity higher than 83% and the presence of all the necessary functional domains for transposition activity (DDE,CRD,TIR,TSD). + +<|ref|>text<|/ref|><|det|>[[117, 635, 878, 847]]<|/det|> +Model Fine- tuning. The ProGen2- base language model12 of 764 million parameters was fine- tuned on over 10,000 sequences from the PB orthologs clustered at 0.8. The models were fine- tuned using the full sequences excluding the N- terminal domain, which is an extremely variable domain. The sequences were split using a 80:20 train:test split. In addition to the set of orthologous sequences used in the training, additional wild- type HyPB sequences (5- 10) were added to the training set to bias the model towards HyPB. This allowed us to generate sequences in a closer sequence identity range to HyPB than we were able to otherwise. Fine- tuning was performed using the Trainer module from Hugging Face over 2 epochs with a training batch size of 4 and evaluation batch size of 8. A constant learning rate of 5.0E- 5 was used and the model was evaluated after every 2000 steps with the highest scoring model on the evaluation set being used for sequence generation. The remaining parameters were kept to the default values. A full exploration of fine- tuning parameters was not done, as with these fairly standard parameters, we were able to generate convincing sequences with our desired properties. + +<|ref|>text<|/ref|><|det|>[[118, 857, 877, 905]]<|/det|> +Plasmid assembly. Transposase ORF amino acid sequences were codon optimized for Homo Sapiens and ordered and synthesized as gene fragments to TWIST biosciences. Gene fragments were cloned into a CMV based expression vector by Golden Gate assembly + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 877, 146]]<|/det|> +using Esp3l restriction enzyme. Transposon (cargo vector) plasmid sequences were defined as the first 150bp from the transposon ends from both 5' and 3' ITR sequences and synthesized as gene fragments by TWIST biosciences with added overhangs for golden gate assembly. EF1a RFP polyA expression cassette was included between the ITRs. + +<|ref|>text<|/ref|><|det|>[[118, 155, 878, 292]]<|/det|> +Transposition activity assays. Hek293T cells, obtained from Thermo Fisher Scientific, were cultured in Dulbecco's modified eagle medium (DMEM) supplemented with high glucose (Gibco, Thermo Fisher), \(10\%\) Fetal Bovine Serum (FBS), \(2 \text{mM}\) glutamine, and 100 U penicillin/0.1 mg/mL streptomycin at \(37^{\circ} \text{C}\) in a \(5\%\) CO2 incubator. For transfection experiments, cells were treated with Polyethyleneimine (PEI, Thermo Fisher Scientific) at a 1:3 DNA- PEI ratio in OptiMem. Prior to transfection, cells were seeded to achieve \(70\%\) confluency the following day, typically using 120,000 cells per adherent p24 well plate. Plasmid DNA was mixed at a ratio of 1 transposase:3 gRNA:5 transposon, with 0.035 pmols of transposase used per p24 well plate. + +<|ref|>text<|/ref|><|det|>[[118, 303, 877, 364]]<|/det|> +The expression of the ITR vector RFP was assessed two days and twenty days post- transfection using cell cytometry with the Cytek Aurora™ CS System. The RFP signal at day twenty was considered indicative of stable transgene integration and was utilized to determine the integration efficiency for each of the tested systems. + +<|ref|>text<|/ref|><|det|>[[117, 373, 878, 647]]<|/det|> +Fluorescent- Excision activity assays. HEK293T cells (Thermo Fisher Scientific) were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco, Thermo Fisher) supplemented with \(10\%\) fetal bovine serum (FBS), \(2 \text{mM}\) L- glutamine, 100 U/mL penicillin, and \(0.1 \text{mg/mL}\) streptomycin. Cells were maintained in a \(37^{\circ} \text{C}\) incubator with \(5\%\) CO2. For each experiment, cells were seeded in 24- well plates at a density of 120,000 cells per well 24 hours prior to transfection to ensure approximately \(70\%\) confluency on the day of transfection. Transfections were performed in 24- well plate using Polyethyleneimine (PEI, Thermo Fisher Scientific) at a 1:3 DNA- PEI ratio in Opti- MEM (Thermo Fisher). A dual- component transfection system was employed, consisting of a plasmid encoding the transposase and a second plasmid containing a disrupted mCherry reporter sequence flanked by transposase recognition sites. Upon excision by the transposase, the mCherry sequence was restored, allowing fluorescence to correlate with excision efficiency (Supplementary Figure 12). For each well, two solutions were prepared separately in Opti- MEM: one containing the plasmid DNA mix (transposase and transposon plasmids at a 1:3 ratio, with a total of 0.035 pmol of transposase) and another containing PEI. The two solutions were then mixed, incubated for 20 minutes at room temperature, and added dropwise to the cells. After 72 hours post- transfection, cells were harvested, and mCherry reporter expression was assessed by flow cytometry using the Cytek Aurora™ CS System. + +<|ref|>text<|/ref|><|det|>[[118, 656, 877, 764]]<|/det|> +Al Sequence Generation. In both models, 50 wild- type amino acids from HyPB were used to prompt sequence generation. For the N- C model, the first 50 amino acids after the N- terminal domain were used and in the C- N model, the final 50 amino acids of the CRD were used to prompt sequence generation. For the C- N model sequences were generated 'backwards' and then reversed to have the standard directionality. The maximum sequence length for both models was set to 500 amino acids and a temperature of \(T = 0.5\) and nucleus probability \(P = 0.95\) were used. + +<|ref|>text<|/ref|><|det|>[[118, 773, 877, 912]]<|/det|> +Al Sequence Filtering. The generated sequences first went through a set of three basic filters. First, duplicated sequences were removed. Second, sequences with non- canonical amino acids were removed. And third, sequences were filtered using a k- mer repetition filter so that no amino acid motif of 6/4/3/2 residues was repeated 2/3/6/8 times consecutively. The next set of filters were HyPB specific and included testing for a PB CRD (based on the presence of at least 7 cysteine amino acids in the final 50 amino acids), sequence identity to WT (between 80- 95% to the DDE+CRD domains), and specific key residues including catalytic site, alpha bridge residues, hyperactive residues and another extensive set of key residues including DNA interacting residues. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 878, 325]]<|/det|> +For all of these sequences we calculated perplexity using the ProGen2- base model as well as the fine- tuned model responsible for generating a given sequence. For a subset of sequences that passed our filters, structures were predicted using ESMFold46. Structures were then compared to the experimentally available PB structure47 (https://www.rcsb.org/structure/6x67) to extract RMSD and tm- scores using PyMOL48 and TMAIign49 respectively. Finally, structures were aligned to the experimental PB structure and several surface properties were calculated using SURFMAP44,45: a tool that projects surface residues from a protein structure into a 2D space and can calculate different amino acid residue properties. The 5 metrics we calculated using SURFMAP were stickiness, circular variance, wimley white, kyte doolittle and electrostatics. We then computed cosine similarities between each surface feature in the generated structures and the experimental structure. Finally ProteinMPNN43 and ESM1v41 scores were calculated. ProteinMPNN is a deep learning- based sequence design method that can decode amino acid sequences from structural representations of proteins. ProteinMPNN can also be used to generate a log- likelihood score for any given sequence. Wimley white is a measure of residue hydrophobicity42 which in this case has been applied to surface residues using SURFMAP: + +<|ref|>text<|/ref|><|det|>[[117, 334, 878, 488]]<|/det|> +An additional set of filters was created to narrow down the 1,000 sequences to a more manageable number. Sequences were required to be in the top 75th percentile for both ProteinMPNN and ESM1v, sequences were filtered on length to exclude sequences that were too short, a conservative plddt filter of 90 was used, and an acceptable range for net charge of the proteins was established. After this, sequences were selected manually in an attempt to cover a range of sequence identities \(90 - 97\%\) to the entire wild- type hyPB sequence with high quality sequences. During this manual selection process, sequences with a higher proportion of the key residues were selected for and any sequences that had particularly bad scores in any of the calculated metrics were avoided. A final selection of 22 sequences was made. + +<|ref|>text<|/ref|><|det|>[[118, 497, 878, 575]]<|/det|> +In silico deep mutational scan. Esm- 1v was used in a zero shot version where the Poecilopsis amino acid sequence was given as an input. Esm1- v creates a fitness score for all possible aminoacids for residue position by calculating a log odds ratio, assuming an additive model when multiple mutations exist. Then, the sum is made over the mutated positions and the sequence is masked at every mutated position41: + +<|ref|>equation<|/ref|><|det|>[[125, 590, 655, 640]]<|/det|> +\[\sum_{t\in T}\log p\left(x_{t} = x_{t}^{mt}|x\backslash_{T}\right) - \log p\left(x_{t} = x_{t}^{wt}|x\backslash_{T}\right)\] + +<|ref|>text<|/ref|><|det|>[[118, 656, 878, 704]]<|/det|> +Variant prediction was run in google colab pro with one A- 100 80gb ram gpu. The script used to run the variant prediction can be found in https://github.com/Alejo945/ISHyPB. The output is a tsv with all possible variants and their scores. + +<|ref|>text<|/ref|><|det|>[[118, 713, 878, 790]]<|/det|> +PCR- Excision activity assays. 48h post- transfection, cells were harvested and a miniprep protocol was performed using the NZYMiniprep kit from NZYtech to extract the remaining plasmids. Primers X and Y, which were flanking the ITRs in the transposon plasmids, were utilized to detect transposon excision. Expecting a 2900bps band indicated not- excised transposon, whereas a 1200bps band indicated excision of both ITR and transposon. + +<|ref|>text<|/ref|><|det|>[[118, 800, 878, 908]]<|/det|> +Targeted Transposition activity assays. Triple mutant residue selection was performed by aligning the ortholog piggyBacs to Trichoplusia Ni PiggyBac mutated sequence. Plasmids encoding the triple mutant variants (PBx3) were co- transfected with Cas9, gRNA and transposon plasmids in a 1:1:3:5 molar ratio into 0.5M Hek23T cells seeded in a p6 plate the day before transfection. Cells were analyzed for RFP expression two days after transfection using cell cytometry with the Cytek Aurora™ CS System. Subsequently, two rounds of enrichment via RFP sorting were conducted with BD FACSAria (Biosciences), one week + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 878, 161]]<|/det|> +and two weeks after transfection. Non- enriched cells were parallelly maintained in culture to measure overall integration levels after 3 weeks. Genomic DNA was extracted using Quiagen DNeasy Blood & Tissue Kit column's four days after the second sorting. A 3' Junction PCR was performed and gel purification using the QIAquick Gel Extraction Kit was performed when necessary before Sanger sequencing the amplified bands. + +<|ref|>text<|/ref|><|det|>[[118, 170, 878, 248]]<|/det|> +Transposition in T- cells. Peripheral Blood Mononuclear Cells (PBMCs) from two different donors, isolated from buffy coats and cryopreserved, were thawed and seeded on p24- coated plates containing anti- CD3/CD28 (1:1,000; BD Sciences) at a density of \(1 \times 10^{6}\) cells/ml in 3 ml of CTS™ OpTimizer™ T Cell Expansion SFM medium (Thermo Fisher), supplemented with IL- 7 and IL- 15 (10 ng/ml each; Miltenyi Biotec). + +<|ref|>text<|/ref|><|det|>[[118, 258, 878, 380]]<|/det|> +On the third day of culture, cells were prepared for electroporation and divided into two groups based on the DNA plasmids: transposon only or transposon + transposase. Electroporation was conducted using the P3 Primary Cell 4D- Nucleofector™ X Kit (Lonza). Cells were washed with PBS (Capricorn) and adjusted to a concentration of \(7.5 \times 10^{6}\) cells per condition. The cell suspension was prepared in 20 μl of nucleofection buffer, consisting of 16.4 μl P3 Primary Cell Nucleofector Solution and 3.6 μl Supplement 1 (Lonza). Subsequently, 1 μg of each DNA plasmid was added to the suspension, and electroporation was carried out using the EO- 115 nucleofection program. + +<|ref|>text<|/ref|><|det|>[[118, 389, 878, 481]]<|/det|> +Following electroporation, 80 μl of complete medium was added, and cells were incubated at \(37^{\circ}C\) for 20 minutes. The cells were then carefully resuspended and transferred to a fresh p24 plate containing 500 μl of medium for recovery and expansion. Approximately one- third of the well volume was used for flow cytometric analysis using the Aurora system (Cytek) to assess RFP expression across three time points, continuing until episomal decay was observed in the transposon- only condition. + +<|ref|>sub_title<|/ref|><|det|>[[119, 503, 348, 526]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[118, 530, 878, 576]]<|/det|> +Bioprospected and synthetically generated transposon sequence files are available in supplementary Table 1. Top active transposon and transposase plasmids have been deposited at addgene. + +<|ref|>text<|/ref|><|det|>[[118, 575, 875, 608]]<|/det|> +Model finetuning and PiggyBac generation code is available at Github https://github.com/Integra- tx/Piggybac bioprospecting_pipeline. + +<|ref|>sub_title<|/ref|><|det|>[[119, 644, 403, 668]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[118, 671, 875, 704]]<|/det|> +We thank Cedric Feschotte for feedback on the bioprospecting pipeline, and George M. Church for advice on approaches for ML/AI and directed evolution. + +<|ref|>sub_title<|/ref|><|det|>[[118, 725, 240, 749]]<|/det|> +## Funding + +<|ref|>text<|/ref|><|det|>[[118, 752, 856, 814]]<|/det|> +Integra Therapeutics S.L. received funding from NEOTEC (CDTI, SNEO- 20222363). UPF received funding from UPGRADE (European Union Horizon 2020, grant agreement No 825825); Ministerio de Economia, Industria y Competitividad de España (Plan Estatal 2013- 2016 (Grant agreement No) + +<|ref|>sub_title<|/ref|><|det|>[[118, 835, 420, 858]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[118, 862, 878, 909]]<|/det|> +DI, ASM and MG conceived the study. AA and DI designed the bioprospecting pipeline. AA implemented the bioprospecting pipeline. JLV implemented the LLM and fine tuning work with help from NF, AA and DI. DI and JJW designed the experiments with help from RD and Maria G. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 879, 146]]<|/det|> +Maria G, RD and JJW performed cell experiments. IH assisted in sequence assembly. AR and PP performed T- cell work. FB contributed to genome data accession and zeroshot modeling. MS analysed targeted integration data. DI and MG supervised the study. AA, DI and JVL plotted the data. AA, DI, MG and JVL wrote the manuscript with contribution from all authors, specially NF. + +<|ref|>sub_title<|/ref|><|det|>[[119, 167, 388, 190]]<|/det|> +## Conflict of interest + +<|ref|>text<|/ref|><|det|>[[118, 195, 878, 241]]<|/det|> +AA, JLV, JJW, Maria G MS, RD, MG, ASM, NF and DI are employed or have consulted for Integra Therapeutics. MG and ASM are shareholders of Integra therapeutics. DI, MG, ASM, AA and RD have filed patents related to this work. + +<|ref|>sub_title<|/ref|><|det|>[[119, 263, 282, 286]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[115, 293, 876, 905]]<|/det|> +1. Wang, J. Y. & Doudna, J. A. CRISPR technology: A decade of genome editing is only the beginning. Science 379, eadd8643 (2023). +2. Yarnall, M. T. N. et al. Drag-and-drop genome insertion of large sequences without double-strand DNA cleavage using CRISPR-directed integrases. Nat. Biotechnol. 41, 500–512 (2023). +3. Mukhametzyanova, L. et al. Activation of recombinases at specific DNA loci by zinc-finger domain insertions. Nat. Biotechnol. (2024) doi:10.1038/s41587-023-02121-y. +4. Li, X. et al. piggyBac transposase tools for genome engineering. Proc. Natl. Acad. Sci. U. S. A. 110, E2279–87 (2013). +5. Yusa, K., Zhou, L., Li, M. A., Bradley, A. & Craig, N. L. A hyperactive piggyBac transposase for mammalian applications. Proc. Natl. Acad. Sci. U. S. A. 108, 1531–1536 (2011). +6. Yusa, K. PiggyBac transposon. Microbiol. Spectr. 3, MDNA3–0028–2014 (2015). +7. Pallarès-Masmitjà, M. et al. Find and cut-and-transfer (FiCAT) mammalian genome engineering. Nat. Commun. 12, 7071 (2021). +8. Ruffolo, J. A. et al. Design of highly functional genome editors by modeling the universe of CRISPR-Cas sequences. bioRxiv (2024) doi:10.1101/2024.04.22.590591. +9. Hayes, T. et al. Simulating 500 million years of evolution with a language model. bioRxiv (2024) doi:10.1101/2024.07.01.600583. +10. Alley, E. C., Khimulya, G., Biswas, S., AlQuraishi, M. & Church, G. M. Unified rational protein engineering with sequence-based deep representation learning. Nat. Methods 16, 1315–1322 (2019). +11. Watson, J. L. et al. De novo design of protein structure and function with RFdiffusion. Nature 620, 1089–1100 (2023). +12. Lauko, A. et al. Computational design of serine hydrolases. bioRxiv (2024) doi:10.1101/2024.08.29.610411. +13. Johnson, S. R. et al. Computational scoring and experimental evaluation of enzymes generated by neural networks. Nat. Biotechnol. (2024) doi:10.1038/s41587-024-02214-2. +14. Mitra, R. et al. Functional characterization of piggyBat from the bat Myotis lucifugus unveils an active mammalian DNA transposon. Proc. Natl. Acad. Sci. U. S. A. 110, 234–239 (2013). +15. Chen, Q. et al. Structural basis of seamless excision and specific targeting by piggyBac transposase. Nat. Commun. 11, 3446 (2020). +16. Yuan, Y.-W. & Wessler, S. R. The catalytic domain of all eukaryotic cut-and-paste transposase superfamilies. Proc. Natl. Acad. Sci. U. S. A. 108, 7884–7889 (2011). +17. Hubley, R. et al. The Dfam database of repetitive DNA families. Nucleic Acids Res. 44, D81–9 (2016). +18. Cosby, R. L. et al. Recurrent evolution of vertebrate transcription factors by transposase capture. Science 371, eabc6405 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 84, 881, 905]]<|/det|> +19. Bouallègue, M., Rouault, J.-D., Hua-Van, A., Makni, M. & Capy, P. Molecular evolution of piggyBac superfamily: From selfishness to domestication. Genome Biol. Evol. 9, 323-339 (2017).20. Zhang, H.-H., Peccoud, J., Xu, M.-R.-X., Zhang, X.-G. & Gilbert, C. Horizontal transfer and evolution of transposable elements in vertebrates. Nat. Commun. 11, 1362 (2020).22. Jangam, D., Feschotte, C., & Betrán, E. (2017). Transposable Element Domestication As an Adaptation to Evolutionary Conflicts. Trends in genetics: TIG, 33(11), 817-831.23. Luo, W., Hickman, A.B., Genzor, P., Ghirlando, R., Furman, C.M., Menshikh, A., 832 Haase, A., Dyda, F., & Wilson, M.H. Transposase N-terminal phosphorylation and 833 asymmetric transposon ends inhibit piggyBac transposition in mammalian cells. 83424. Wicker T, Sabot F, Hua-Van A, Bennetzen JL, Capy P, et al. 2007. A unified classification system for eukaryotic transposable elements. Nat. Rev. Genet 8(12):973-82.25. Sean R Eddy. Accelerated profile hmm searches. PLoS computational biology, 7(10):e1002195, 2011.26. Krause, G. R., Shands, W., & Wheeler, T. J. (2024). Sensitive and error-tolerant annotation of protein-coding DNA with BATH. Bioinformatics Advances, 4(1).27. Thompson, J. D., Higgins, D. G., & Gibson, T. J. (1994). CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic acids research, 22(22), 4673-4680.28. Paul A Kitts, Deanna M Church, Françoise Thibaud-Nissen, Jinna Choi, Vichet Hem, Victor Sapojnikov, Robert G Smith, Tatiana Tatusova, Charlie Xiang, Andrey Zherikov, et al. Assembly: a resource for assembled genomes at ncbi. Nucleic acids research, 44(D1):D73-D80, 2016.29. Storer, J., Hubley, R., Rosen, J., Wheeler, T. J., & Smit, A. F. (2021). The Dfam community resource of transposable element families, sequence models, and genome annotations. Mobile DNA, 12(1), 2.30. Camacho, C., Coulouris, G., Avagyan, V., Ma, N., Papadopoulos, J., Bealer, K., & Madden, T. L. (2009). BLAST+: architecture and applications. BMC bioinformatics, 10, 421.31. Lu, S., Wang, J., Chitsaz, F., Derbyshire, M. K., Geer, R. C., Gonzales, N. R., Gwadz, M., Hurwitz, D. I., Marchler, G. H., Song, J. S., Thanki, N., Yamashita, R. A., Yang, M., Zhang, D., Zheng, C., Lanczycki, C. J., & Marchler-Bauer, A. (2020). CDD/SPARCLE: the conserved domain database in 2020. Nucleic acids research, 48(D1), D265-D268.32. Abramson, J., Adler, J., Dunger, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493-500 (2024).33. Rice P., Longden I. and Bleasby A. EMBOSS: The European Molecular Biology Open Software Suite. Trends in Genetics. 2000 16(6):276-27734. Steinegger, M., Söding, J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nat Biotechnol 35, 1026-1028 (2017)35. B.Q. Minh, H.A. Schmidt, O. Chernomor, D. Schrempf, M.D. Woodhams, A. von + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[150, 84, 878, 117]]<|/det|> +Haeseler, R. Lanfear (2020) IQ- TREE 2: New models and efficient methods for phylogenetic inference in the genomic era. Mol. Biol. Evol., 37:1530- 1534 + +<|ref|>text<|/ref|><|det|>[[115, 125, 880, 160]]<|/det|> +36. Edgar, R.C. MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics 5, 113 (2004) + +<|ref|>text<|/ref|><|det|>[[117, 169, 878, 217]]<|/det|> +37. S. Kalyaanamoorthy, B.Q. Minh, T.K.F. Wong, A. von Haeseler, L.S. Jermiin (2017) ModelFinder: Fast model selection for accurate phylogenetic estimates. Nat. Methods, 14:587-589 + +<|ref|>text<|/ref|><|det|>[[117, 227, 877, 260]]<|/det|> +38. D.T. Hoang, O. Chernomor, A. von Haeseler, B.Q. Minh, and L.S. Vinh (2018) UFBoot2: Improving the ultrafast bootstrap approximation. Mol. Biol. Evol., 35:518-522 + +<|ref|>text<|/ref|><|det|>[[117, 270, 878, 318]]<|/det|> +39. Ivica Letunic and Peer Bork. Interactive tree of life (itol) v5: an online tool for phylogenetic tree display and annotation. Nucleic acids research, 49(W1):W293-W296, 2021 + +<|ref|>text<|/ref|><|det|>[[117, 327, 878, 376]]<|/det|> +40. UCSF Chimera--a visualization system for exploratory research and analysis. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE. J Comput Chem. 2004 Oct;25(13):1605-12 + +<|ref|>text<|/ref|><|det|>[[117, 386, 878, 433]]<|/det|> +41. Meier, J., Rao, R., Verkuil, R., Liu, J., Sercu, T., & Rives, A. (2021). Language models enable zero-shot prediction of the effects of mutations on protein function. bioRxiv. + +<|ref|>text<|/ref|><|det|>[[117, 444, 877, 477]]<|/det|> +42. Wimley, W. C.; White, S. H. Experimentally Determined Hydrophobicity Scale for Proteins at Membrane Interfaces. Nat. Struct. Biol. 1996, 3 (10), 842-848. + +<|ref|>text<|/ref|><|det|>[[117, 488, 877, 521]]<|/det|> +43. J. Dauparas et al., Robust deep learning-based protein sequence design using ProteinMPNN. Science. 2022, 378, 49-56 DOI:10.1126/science.add2187 + +<|ref|>text<|/ref|><|det|>[[117, 531, 878, 580]]<|/det|> +44. Hugo Schweke, Marie-Hélène Mucchielli, Nicolas Chevrollier, Simon Gosset, Anne Lopes. SURFMAP: a software for mapping in two dimensions protein surface features. J. Chem. Inf. Model. 2022. + +<|ref|>text<|/ref|><|det|>[[117, 589, 877, 622]]<|/det|> +45. Sanner, M. F., Olson A.J. & Spehner, J.-C. (1996). Reduced Surface: An Efficient Way to Compute Molecular Surfaces. Biopolymers 38:305-320. + +<|ref|>text<|/ref|><|det|>[[117, 632, 877, 665]]<|/det|> +46. Zeming Lin et al, Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 2023, 379, 1123-1130. DOI:10.1126/science.ade2574 + +<|ref|>text<|/ref|><|det|>[[117, 675, 877, 723]]<|/det|> +47. Chen, Q., Luo, W., Veach, R. A., Hickman, A. B., Wilson, M. H., & Dyda, F. (2020). Structural basis of seamless excision and specific targeting by piggyBac transposase. Nature Communications, 11, 3446. https://doi.org/10.1038/s41467-020-17128-1 + +<|ref|>text<|/ref|><|det|>[[117, 733, 805, 751]]<|/det|> +48. PyMOL: The PyMOL Molecular Graphics System, Version 3.1 Schrödinger, LLC. + +<|ref|>text<|/ref|><|det|>[[117, 761, 877, 794]]<|/det|> +49. Zhang, Y., Skolnick, J. (2005). TM-align: A protein structure alignment algorithm based on TM-score. Nucleic Acids Research, 33(7), 2302-2309. + +<|ref|>text<|/ref|><|det|>[[117, 804, 875, 836]]<|/det|> +50. Ke, G. et al, (2017). Lightgbm: A highly efficient gradient boosting decision tree. Advances in Neural Information Processing Systems, 30, 3146-3154. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[59, 131, 309, 203]]<|/det|> +- supplementaryTable1.csv- supplementaryTable2.txt- supplementary.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/images_list.json b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..607e1f166f8bd25ddc7ba39ac953e8a73fb0bad5 --- /dev/null +++ b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/images_list.json @@ -0,0 +1,212 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "FIG. 1. Snapshots illustrating steady-state configurations of the vesicle and enclosed active filaments as a function of filament aspect ratio \\(a\\) and initial volume fraction \\(\\phi\\) . See SI Movie 1 for animations of the corresponding simulations [51]. The marked regions of parameter space indicate the typical vesicle conformation: (I) spherical, (II) oblate, (III) polar-prolate, (IV) apolar-prolate, and (V) polyhedral. The symbols associate the conformation with the internal filament organization: homogeneous throughout the bulk or on the surface, with no vesicle deformation \\((\\bullet)\\) ; transient clusters and/or bands, with oblate vesicle shapes \\((\\star)\\) ; stable polar rings \\((\\circ)\\) ; stable caps \\((\\circ\\) , with a number of intersecting lines equal to the median number of caps); and dynamic caps \\((\\diamondsuit)\\) . The dashed line shows the transition to aligned states predicted from the competition between the characteristic collision and reorientation timescales \\((\\phi = (\\pi /4)^2 /a)\\) described in the text, and the horizontal dotted line indicates the approximate threshold aspect ratio for the filaments to be in the strong confinement limit. Other parameters are filament bending modulus \\(\\kappa_{\\mathrm{fil}} = 10^4\\) and vesicle radius \\(R_{\\mathrm{ves}} = 25\\) .", + "footnote": [], + "bbox": [ + [ + 543, + 60, + 890, + 297 + ] + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "FIG. 2. a) The onset of ring and cap formation is determined by a competition of timescales: the timescale associated with rotations parallel to the vesicle (left) and the timescale associated with collisions that tend to orient the filaments perpendicular to the vesicle (right). b) Schematic of the theory for the number of caps (Eqs. (1) and (2)). We assume an activity-induced effective attractive interaction that is quadratic in the rod-rod contact length \\(\\Delta l\\) (left). The cap is assumed circular, with size parameterized by the angle \\(\\theta\\) between the cap center and edge. Vesicle curvature leads to a shearing of rods within the cap (right).", + "footnote": [], + "bbox": [ + [ + 123, + 60, + 450, + 191 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "FIG. 3. Steady-state configurations as a function of \\(\\phi\\) and active force \\(f_{\\mathrm{a}}\\) . The marked regions are defined as in Fig. 1. The dashed line shows the transition to aligned states predicted by the timescale competition, which is independent of \\(f_{\\mathrm{a}}\\) . Other parameters are \\(a = 10.5\\) and \\(\\kappa_{\\mathrm{fl}} = 10^{4}\\) . See SI Movie 2 for corresponding animations.", + "footnote": [], + "bbox": [ + [ + 110, + 60, + 460, + 255 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "FIG. 4. The number of caps measured in simulations (symbols) compared to the theory (Eqs. (1) and (2), dashed line), with \\(A = 4R_{\\mathrm{ves}} / \\sqrt{3\\pi \\sigma}\\) and \\(a^{*} \\approx 130\\) (chosen by eye). Diamonds indicate dynamic cap states. Note that the number of caps in the simulation results is likely under counted for the dynamic states due to the caps' motility. The simulation data is the same as in Fig. 1. Active filaments are colored by which cap they belong to for visual clarity.", + "footnote": [], + "bbox": [ + [ + 543, + 61, + 890, + 257 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "FIG. 5. a) Vesicle conformations and filament organizations as a function of filament rigidity \\(\\kappa_{\\mathrm{fil}}\\) and active force strength \\(f_{\\mathrm{a}}\\) , for volume fraction \\(\\phi = 0.2\\) and filament aspect ratio \\(a = 10.5\\) . For a given filament stiffness, increasing activity reduces the number of caps until an upper-threshold activity value \\(f_{\\mathrm{a}}^{\\mathrm{SC}}\\) , beyond which the system transitions into an undeformed state. As described in SI section E, this transition occurs because activity renormalizes the filament bending modulus to smaller values [41], thus reducing filament alignment interactions and causing the system to leave the strong confinement limit. The dashed line shows the prediction for \\(f_{\\mathrm{a}}^{\\mathrm{SC}}\\) given by Eq. S35. Note that there is no adjustable parameter. In the rigid rod limit \\((\\kappa_{\\mathrm{fil}} > 10^{3})\\) all non-zero active force values that we simulated led to cap formation. b) Selected snapshots of states shown in (a). Animations of these states can be found in SI Movie 3.", + "footnote": [], + "bbox": [ + [ + 536, + 63, + 896, + 442 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "FIG. 6. a) Simulation snapshots illustrating vesicle conformations and filament organizations as a function of filament volume fraction and vesicle rigidity. As the vesicle rigidity is reduced below a critical value (corresponding to the critical Foppl-von Karman number \\(\\mathrm{FvK} \\approx 154\\) [67]), the vesicle undergoes a buckling transition leading to the formation of facets. While we observe most of the same classes of filament self-organization in faceted and round vesicles, polar bands trace a dynamic path between vertices in faceted vesicles, while they trace a stable geodesic in round vesicles. Animations of these states can be found in SI movie 5. b) Comparisons between flexible and rigid vesicles as a function of filament aspect ratio, with other parameters set to \\(f_{\\mathrm{a}} = 8\\) , \\(\\phi = 0.1\\) , and \\(\\kappa_{\\mathrm{fil}} = 10^{4}\\) . In contrast to flexible vesicles, rigid vesicles do not allow for the formation of stable caps or rings. When polar rings do form in rigid vesicles, they are transient—continuously breaking and reforming over the course of the trajectory. Animations of this comparison can be found in SI movie 6.", + "footnote": [], + "bbox": [ + [ + 85, + 65, + 480, + 432 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 50, + 110, + 933, + 720 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 60, + 66, + 938, + 408 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 55, + 60, + 930, + 550 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 55, + 55, + 933, + 560 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_0.jpg", + "caption": "b)", + "footnote": [], + "bbox": [ + [ + 55, + 60, + 680, + 400 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 188, + 444, + 559, + 732 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_unknown_1.jpg", + "caption": "b)", + "footnote": [], + "bbox": [ + [ + 66, + 60, + 825, + 410 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 230, + 447, + 768, + 772 + ] + ], + "page_idx": 14 + } +] \ No newline at end of file diff --git a/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947.mmd b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947.mmd new file mode 100644 index 0000000000000000000000000000000000000000..3e2cd2b0170a80b8c6e26820e9e482b41cee7b08 --- /dev/null +++ b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947.mmd @@ -0,0 +1,239 @@ + +# Vesicle shape transformations driven by confined active filaments + +Matthew Peterson Brandeis University https://orcid.org/0000- 0002- 5749- 7770 + +Apama Baskaran Brandeis University + +Michael Hagan (hagan@brandeis.edu) Brandeis University https://orcid.org/0000- 0002- 9211- 2434 + +## Article + +Keywords: Active Matter Systems, Deformable Boundaries, Particle- based Simulations, Active Stress Organization, Emergent Behaviors, Targeted Shape Dynamics + +Posted Date: May 6th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 448564/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on December 1st, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 27310- 8. + +<--- Page Split ---> + +# Vesicle shape transformations driven by confined active filaments + +Matthew S. E. Peterson, Aparna Baskaran,\* and Michael F. Hagan† Martin A. Fisher School of Physics, Brandeis University, Waltham, MA, 02453 (Dated: April 21, 2021) + +In active matter systems, deformable boundaries provide a mechanism to organize internal active stresses and perform work on the external environment. To study a minimal model of such a system, we perform particle- based simulations of an elastic vesicle containing a collection of polar active filaments. The interplay between the active stress organization due to interparticle interactions and that due to the deformability of the confinement leads to a variety of filament spatiotemporal organizations that have not been observed in bulk systems or under rigid confinement, including highly- aligned rings and caps. In turn, these filament assemblies drive dramatic and tunable transformations of the vesicle shape and its dynamics. We present simple scaling models that reveal the mechanisms underlying these emergent behaviors and yield design principles for engineering active materials with targeted shape dynamics. + +## I. INTRODUCTION + +Active matter encompasses systems whose microscopic constituents consume energy at the particle scale to produce forces and motion. Novel macroscale phenomena emerge in these systems when these forces collectively organize into mesoscale 'active stresses'. Harnessing these active stresses to drive particular emergent behaviors could enable a new class of materials with life- like properties that would be impossible in traditional equilibrium materials. However, rationally designing active constituents to generate a desired emergent behavior requires identifying the principles that govern organization of mesoscopic active stresses. Similarly, many biological functions, such as cytoplasmic streaming, morphogenesis, and cell migration, are driven by active stresses that emerge from active components confined within a cell [1- 8]. Understanding physical mechanisms that underlie these functions is a key goal of cellular biophysics. + +As a step toward these goals, this article describes a computational and theoretical study of a minimal model of active filaments enclosed within a deformable vesicle. We thereby identify a generic route to control selforganizing active stresses by enclosing active components with anisotropic shapes and/or internal degrees of freedom within deformable confining boundaries. Our results demonstrate that coupling between boundary deformations and the assembly of internal active components leads to a positive feedback capable of driving diverse nonlinear transformations of morphology and dynamics. + +The field of active matter has identified two key mechanisms that control self- organization of active stresses: (1) anisotropic interactions between active components that realign forces, and (2) confining boundaries. For example, interactions between self- propelled particles that drive interparticle alignment result in bands or flocks [9, 10], changing the length and stiffness of active polymers leads to dramatic reorganization of active stresses [11, 12], and confining active particles leads to system- spanning effects such as spontaneous flow [13- 20]. Furthermore, deformable confining boundaries enable non- equilibrium boundary fluctuations [21- 26], including elongated tendrils and bolas [22]. The latter results suggest that flexibility is a key characteristic of a confining boundary, as it allows shape transformations, sensing and response to environmental cues, and performing work on the surroundings. Achieving such capabilities is critical to leverage minimal bio- derived experimental systems (e.g. [23, 27, 28]) to engineer controlled shape transformations. However, little is known about the behaviors that may arise when these two active stress organization modes are combined. In particular, most existing theoretical and computational studies have focused on rigid boundaries [29- 32], isotropic active particles [21, 22, 24, 33- 36], or have been in 2D [37- 39]. + +In this work, we use Langevin dynamics simulations of polar self- propelled semiflexible filaments confined within 3D flexible vesicles to study the combination of active interparticle alignment interactions and 3D deformable confinement. Our simulations show that interplay between these two organization modes leads to a positive feedback, in which active forces drive boundary deformation while passive stresses from the boundary guide and reinforce self- organization of the internal active stresses. This leads to a rich variety of steady- state behaviors that have not been observed in bulk systems or under rigid confinement, including highly- aligned rings, and caps that have tunable self- limited sizes, number, and symmetry. Each filament organization drives a characteristic large- scale vesicle shape transformation that can be selected by varying parameters such as filament length, density, and flexibility. Asymmetric states lead to net vesicle motion, consistent with the recent experiments which find that enclosing self- propelled particles, such as bacteria, in droplets can lead to collective motility [40]. We present simple scaling analyses that reveal how the feedback between vesicle geometry and filament organization drives and stabilizes these emergent behaviors. The applicability of these scaling arguments suggests that these behaviors + +<--- Page Split ---> + +arise generically due to feedback between vesicle elasticity and active filament organization, independent of the specific model. + +Understanding the fundamental mechanisms that govern this coupling between self- organization of active stress and deformations of a flexible boundary will establish design principles for soft robotics, artificial cells, or other advanced materials that mimic the capabilities of living organisms. From a biological perspective, our minimal model is not intended to directly describe the cytoskeleton—in this model, active filament propulsion enters in the overdamped limit without long- ranged hydrodynamic coupling. However, the generic feedback between vesicle elasticity and active filament organization identified in our simulations and scaling arguments may elucidate how mesoscopic active stresses in the cellular cytoskeleton drive large scale cellular shape transformations that underlie essential biological functions such as motility [1, 2], division [3- 5], and endo-/exocytosis [6, 7]. + +## II. METHODS + +We simulate a system of \(N_{\mathrm{fil}}\) active filaments confined within an elastic vesicle, which has radius \(R_{\mathrm{ves}}\) in its undeformed state. We represent active filaments using the model in Joshi, et al. [41]—modified so that the active forcing is polar rather than nematic—in which each filament is a nearly- inextensible, semiflexible chain of \(M\) beads of diameter \(\sigma\) [42]. Bonded beads interact through an expanded FENE potential [43], while non- bonded beads interact through a purely repulsive expanded Weeks- Chandler- Andersen (eWCA) potential [44] with strength \(\epsilon\) . The equilibrium bond length is set to \(b_{\mathrm{fil}} = \sigma /2\) to minimize surface roughness between interacting filaments, thereby preventing filaments from interlocking at high density [45- 49]. The filaments are made semiflexible with bending rigidity \(\kappa_{\mathrm{fil}}\) through a harmonic angle potential applied to each set of three consecutive beads along the chain. Since our model is not intended to describe any specific biofilament system, we incorporate activity in a minimal manner—a polar active force of magnitude \(f_{\mathrm{a}}\) acts on each bead, in a direction tangent to the filament and toward the filament head. The filament volume fraction in the undeformed state is given by \(\phi = N_{\mathrm{fil}}V_{\mathrm{fil}} / V_{\mathrm{ves}}\) , where \(V_{\mathrm{fil}} = \pi \sigma^3 /6 + (M - 1)\pi b\sigma^2 /4\) is the approximate volume of a single filament—accounting for the overlap of bonded monomers—and \(V_{\mathrm{ves}} = 4\pi R_{\mathrm{ves}}^3 /3\) is the nominal volume of the vesicle. Since the mesh topology is conserved in our simulations, we model an elastic vesicle. In subsequent work we plan to consider the effects of fluidizing the vesicle and imposing area or volume constraints. + +We simulate the coupled Langevin equations for the filament and vesicle bead dynamics using LAMMPS [50], modified to include the active force. We neglect long- ranged hydrodynamic interactions for this system of high filament density; we will investigate their effect in a future study. We have set units such that the mass of all beads is + +![](images/Figure_1.jpg) + +
FIG. 1. Snapshots illustrating steady-state configurations of the vesicle and enclosed active filaments as a function of filament aspect ratio \(a\) and initial volume fraction \(\phi\) . See SI Movie 1 for animations of the corresponding simulations [51]. The marked regions of parameter space indicate the typical vesicle conformation: (I) spherical, (II) oblate, (III) polar-prolate, (IV) apolar-prolate, and (V) polyhedral. The symbols associate the conformation with the internal filament organization: homogeneous throughout the bulk or on the surface, with no vesicle deformation \((\bullet)\) ; transient clusters and/or bands, with oblate vesicle shapes \((\star)\) ; stable polar rings \((\circ)\) ; stable caps \((\circ\) , with a number of intersecting lines equal to the median number of caps); and dynamic caps \((\diamondsuit)\) . The dashed line shows the transition to aligned states predicted from the competition between the characteristic collision and reorientation timescales \((\phi = (\pi /4)^2 /a)\) described in the text, and the horizontal dotted line indicates the approximate threshold aspect ratio for the filaments to be in the strong confinement limit. Other parameters are filament bending modulus \(\kappa_{\mathrm{fil}} = 10^4\) and vesicle radius \(R_{\mathrm{ves}} = 25\) .
+ +\(m = 1\) , and energies, lengths, and time are respectively in units of \(k_{\mathrm{B}}T\) , \(\sigma\) , and \(\tau = \sqrt{m\sigma^2 / \epsilon}\) . The friction constant is set to \(\gamma = 1 / \tau\) . For additional model details, see the Supplemental Materials [51]. + +## III. RESULTS AND DISCUSSION + +### A. Simulation results + +To discover the steady- state conformations that arise due to coupling between active propulsion and elasticity, we have performed simulations over a wide range of control parameters (Fig. 1, Fig. 3, Fig. 5, and Fig. 6): the volume fraction of filaments in the vesicle \(\phi \in [0.01, 0.4]\) , filament aspect ratio \(a = 1 + (M - 1)(b_{\mathrm{fil}} / \sigma) \in [3, 25.5]\) , active propulsion strength \(f_{\mathrm{a}} \in [0, 10]\) , filament stiffness \(\kappa_{\mathrm{fil}} \in [10^2, 10^4]\) , and vesicle rigidity \(\kappa_{\mathrm{ves}} \in [10^2, 10^4]\) . + +Fig. 1 shows the steady- states as a function of filament volume fraction and aspect ratio for moderate activity \(f_{\mathrm{a}} = 8\) . At this activity and vesicle size, for aspect + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
FIG. 2. a) The onset of ring and cap formation is determined by a competition of timescales: the timescale associated with rotations parallel to the vesicle (left) and the timescale associated with collisions that tend to orient the filaments perpendicular to the vesicle (right). b) Schematic of the theory for the number of caps (Eqs. (1) and (2)). We assume an activity-induced effective attractive interaction that is quadratic in the rod-rod contact length \(\Delta l\) (left). The cap is assumed circular, with size parameterized by the angle \(\theta\) between the cap center and edge. Vesicle curvature leads to a shearing of rods within the cap (right).
+ +ratios \(a\gtrsim (8\pi R_{\mathrm{ves}} / f_{\mathrm{a}})^{1 / 3}\approx 4.3\) the system is in the strong confinement limit: because the persistence length \(l_{\mathrm{p}}^{\mathrm{COM}}\propto f_{\mathrm{a}}a^{3}\) of the filament center- of- mass motion is larger than the vesicle size \(l_{\mathrm{p}}^{\mathrm{COM}} > 2R_{\mathrm{ves}}\) , most filaments are found on the vesicle surface at all times [52] (see SI Sec. A [51]). + +Under these conditions we can classify the steady- state vesicle conformations into several categories: (I) spherical, (II) oblate, (III) polar- prolate, (IV) apolar- prolate, and (V) polyhedral. These vesicle configurations are tightly coupled to the spatiotemporal organization of the filaments within, as follows. + +(I): Spherical vesicle shapes arise at low filament volume fractions and aspect ratios. Under these conditions, filament- filament collisions are rare and inter- filament aligning forces are weak [53- 59]. Thus, filament positions and orientations are homogeneous (throughout the vesicle interior below strong confinement, or on the vesicle surface above strong confinement), leading to little deformation of the vesicle. + +(II): For low volume fraction but high aspect ratios, such that the filament length \(L = a\sigma\) is comparable to the unperturbed vesicle radius, \(L\sim R_{\mathrm{ves}}\) , the vesicle deforms into oblate spheroid conformations. This transition is driven by the filaments organizing into a stable polar band, which deforms the vesicle along a geodesic. This filament arrangement closely resembles the polar bands observed on the surface of rigid spheres for active particles with polar propulsion and polar interparticle alignment interactions [35], which arise due to topological requirements for a surface- constrained polarization field [31]. However, note that such polar bands would be unstable in our system if the confining geometry was a rigid sphere because the filament- filament interactions in our system are nematic (head- tail symmetric) [60, 61]. The finite deformability of the vesicle is essential to stabilize this + +configuration—active forces due to the polar band force the vesicle into an oblate shape, which in turn provides a restoring force to stabilize filament alignment within the band. In support of this conclusion, simulations on infinitely rigid vesicles did not exhibit stable polar bands (see Fig. 6b and SI movie 6). Thus, this configuration provides a concrete example of how feedback between passive stresses and self- organization of active stresses can generate steady states that would be otherwise disallowed by symmetry. + +(III- IV): For intermediate volume fractions and aspect ratios, the vesicle deforms into a prolate spheroid. These prolate vesicle conformations can be further classified by their motion, either polar (III) or apolar (IV). Further increasing the volume fraction or decreasing the aspect ratio leads to polyhedral conformations, (V). States (III- V) all result from filaments assembling into crystalline caps in which the rods are highly aligned and perpendicular to the vesicle surface. Interestingly, the caps are 'self- limited' in that their typical size decreases with decreasing aspect ratio, but is roughly independent of the total number of filaments \(N_{\mathrm{fil}}\) in the vesicle. Increasing \(N_{\mathrm{fil}}\) at fixed aspect ratio increases the number of caps; we observe up to 12 caps for the finite vesicle size that we consider (Fig. 4). Further, caps drive local curvature of the vesicle, leading to elasticity- mediated cap- cap repulsions which favor symmetric arrangements of caps. Thus, the vesicle morphology can be sensitively tuned by controlling filament aspect ratio and density to achieve a specific number of caps. The polar- prolate (III), apolar- prolate (IV), and polyhedral states (V) respectively have 1, 2, and \(\geq 3\) caps. Generally, states with two or more caps do not exhibit directed motion. However, for enough caps in the vesicle (typically more than 3), the caps themselves can become motile, and collide with, merge with, and split from other caps (see below). + +### B. Mechanisms underlying stress organization and deformation + +To understand how these conformations are governed by the interplay between propulsion- induced aligning forces, vesicle deformability, and vesicle curvature, we develop simple scaling estimates for the timescales and forces that govern filament alignment and interactions with the vesicle. First, we consider the transition between undeformed spherical vesicle states characterized by unaligned or weakly aligned filaments as in state (I), to the highly deformed oblate, prolate, and polyhedral vesicle shapes of states (II- V). Our simulations demonstrate that such significant vesicle shape deformations occur when filament- filament interactions mediate the organization of ordered structures either in the plane of the vesicle or orthogonal to it. + +a. The onset of filament assembly: The onset of this transition can be understood by considering a competition between two characteristic timescales that respec + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
FIG. 3. Steady-state configurations as a function of \(\phi\) and active force \(f_{\mathrm{a}}\) . The marked regions are defined as in Fig. 1. The dashed line shows the transition to aligned states predicted by the timescale competition, which is independent of \(f_{\mathrm{a}}\) . Other parameters are \(a = 10.5\) and \(\kappa_{\mathrm{fl}} = 10^{4}\) . See SI Movie 2 for corresponding animations.
+ +tively govern collision- induced filament- vesicle alignment and filament- filament alignment (see Fig. 2). Filament- vesicle collisions, which tend to reorient filaments parallel to the surface [62, 63], have a characteristic timescale \(\tau_{\mathrm{rot}} \sim L / v_{0}\) [64], with \(v_{0} = f_{\mathrm{a}} / \gamma\) the filament self- propulsion velocity. We can estimate the timescale for filament interactions by considering filament- filament pairwise collisions whose timescale is given by \(\tau_{\mathrm{coll}} \sim \sigma / v_{0} \phi\) (see SI [51]). Thus, deformed vesicle states will arise when \(\tau_{\mathrm{coll}} < \tau_{\mathrm{rot}}\) or equivalently \(a \phi > c\) , where \(c \cong (\pi / 4)^{2}\) is independent of activity and filament length (see [51]). This defines a boundary separating highly deformed states of the vesicle from the undeformed spherical states (the dashed line in Fig. 1). + +Notably, the active force drops out of this argument because both collision and reorientation times are \(\propto f_{\mathrm{a}}\) . Thus, the theory predicts that the emergence of deformed vesicle states is independent of activity of the enclosed filaments (above a threshold activity). As a test of this prediction, Fig. 3 shows the steady- states as a function of \(\phi\) and \(f_{\mathrm{a}}\) for fixed aspect ratio \(a = 10.5\) . Indeed, formation of large deformations does not depend on activity, with non- spherical shapes forming for \(\phi \geq c / a \approx 0.06\) (as predicted by the above timescale argument) for all \(f_{\mathrm{a}} > 0\) that we considered. + +This simple theoretical picture gives a predictive principle, in terms of properties of the active filaments, for when vesicle shape transformations occur. However, the theory assumes the strong activity, long filament limit and thus neglects thermal noise. Below a threshold activity ( \(f_{\mathrm{a}} \lesssim 1\) in our units) the vesicle will not deform because filament organization is destroyed by thermal fluctuations. Also, cap formation (and thus vesicle shape transitions) do not occur when the filaments are below the strong confinement limit discussed above ( \(a \lesssim 4.3\) for the parameters of Fig. 1, shown as a dotted line). + +![](images/Figure_4.jpg) + +
FIG. 4. The number of caps measured in simulations (symbols) compared to the theory (Eqs. (1) and (2), dashed line), with \(A = 4R_{\mathrm{ves}} / \sqrt{3\pi \sigma}\) and \(a^{*} \approx 130\) (chosen by eye). Diamonds indicate dynamic cap states. Note that the number of caps in the simulation results is likely under counted for the dynamic states due to the caps' motility. The simulation data is the same as in Fig. 1. Active filaments are colored by which cap they belong to for visual clarity.
+ +b. Cap morphologies: We can derive further insight into shape transformations by considering the system in the strongly deformed regime with polyhedral shapes. The defining characteristic underlying these states is filament assembly into well-ordered caps. Most cap states are relatively static, with occasional association/dissociation of individual rods (See SI Movie 4), except for the parameters that lead to the highly dynamic, reconfiguring caps discussed below. In a static steady state, the active and elastic forces must balance. In particular, the dense crystalline nature of caps arises because the active force and the presence of the vesicle surface leads to an effective attractive interaction between nearby filaments. This attraction drives radial growth of a cap, since filaments on the cap periphery have fewer neighbors, leading to an effective interfacial tension. This effect is both reinforced by and competes with vesicle elasticity. The active force of small caps drives vesicle deformations whose local curvature enhances effective filament-filament attractions. However, as the cap grows in radius, vesicle curvature drives an effective shear of filaments (see Fig. 2b) that reduces rod-rod overlaps and thus opposes the active force. + +We describe this competition by constructing an effective 'free energy' whose gradients correspond to the active and passive forces (Fig. 2b). Since the active force favors rods to align in a smectic layer, the shear due to vesicle curvature imposes an 'energy' cost of \(U_{\mathrm{shear}}(\theta) = n_{\mathrm{cap}} 2\pi R_{\mathrm{ves}}^{2} \frac{G}{2} [\cos \theta + \sec \theta - 2]\) , with \(\theta\) the angle subtended by the cap on the vesicle surface, \(n_{\mathrm{cap}}\) the number of caps, and a 'shear modulus' \(G \sim f_{\mathrm{a}}\) (but independent of \(L_{\mathrm{rod}}\) ) [51]. In the strongly deformed region the caps are roughly circular, so the interfacial energy is given by \(U_{\mathrm{int}}(\theta) = n_{\mathrm{cap}} 2\pi R_{\mathrm{ves}} \gamma \sin \theta\) , with the 'interfacial tension' \(\lambda \sim L_{\mathrm{rod}} f_{\mathrm{a}}\) accounting for the diminished inter + +<--- Page Split ---> + +actions at the cap boundary. This results in a free energy as a function of cap size [51]: + +\[f(\theta) = \frac{1}{1 - \cos\theta}\left[\frac{1}{2} (\cos \theta +\sec \theta -2) + \zeta \sin \theta\right] \quad (1)\] + +where \(\zeta = G / \gamma R_{\mathrm{ves}}\sim L / R_{\mathrm{ves}}\) is given by the balance between the effective interfacial tension and shear modulus, and should be linear in filament length but roughly independent of \(f_{\mathrm{a}}\) since both of these effects are driven by activity. + +Minimizing this per- filament free energy yields an optimal \(\theta\) [65, 66] corresponding to the self- limited cap size. Assuming that we are well above the onset of cap formation so that essentially all filaments are in caps, + +\[n_{\mathrm{cap}}\propto \phi a^{-1}[1 + (a / a^{*})^{-2 / 3}] \quad (2)\] + +where \(a^{*}\propto R_{\mathrm{ves}} / \sigma\) is an adjustable parameter that may depend on activity. This expression holds provided \(a\ll\) \(a^{*}\) . For the data in Fig. 1, we obtain \(a^{*}\approx 130\) , leading to the dashed line shown in Fig. 4. + +Except for states with many \((n_{\mathrm{cap}}\gtrsim 7)\) motile caps, there is close agreement between the observed and predicted \(n_{\mathrm{cap}}\) . Above this threshold our cap- counting algorithm likely under counts \(n_{\mathrm{cap}}\) , since different caps are often adjacent and interacting. Further, the prediction of Eqs. (1) and (2) that the self- limited cap size is independent of activity is consistent with observations at different \(f_{\mathrm{a}}\) (see Fig. 3). The motile cap states appear to arise when the curved vesicle geometry forces interactions between the inward- facing ends of adjacent caps. Such interactions occur above a threshold number and aspect ratio of filaments, given by \(N_{\mathrm{fil}}\gtrsim C(1 - a\sigma /R_{\mathrm{ves}})^2\) , where \(C\) is a constant (see SI Sec. D [51]). + +We note that the geometric factors governing the selflimited cap size parallel those in a recently studied equilibrium system of rigid filaments end- adsorbed onto a rigid spherical nanoparticle, which self- assemble due to direct pairwise inter- filament attractions [66]. However, in the present system, the effective filament- filament interactions and vesicle geometry are many- body and emergent in that they arise due to feedback between non- equilibrium active forces and vesicle deformations. + +### C. Effect of filament and vesicle rigidity + +Thus far, we have focused on the interplay between activity and vesicle deformability by performing simulations in the limit of rigid rods, \(\kappa_{\mathrm{fil}} = 10^{4}\) , and high (but finite) vesicle rigidity \(\kappa_{\mathrm{ves}} = 5\times 10^{3}\) . We now briefly discuss the effect of allowing for finite filament and vesicle flexibility. + +Fig. 5 shows the vesicle conformation and filament organization states as a function of filament bending modulus and activity, for fixed filament volume fraction \(\phi = 0.2\) . We see that for finite filament flexibility, the transition to aligned ring and cap states is suppressed + +![](images/Figure_5.jpg) + +
FIG. 5. a) Vesicle conformations and filament organizations as a function of filament rigidity \(\kappa_{\mathrm{fil}}\) and active force strength \(f_{\mathrm{a}}\) , for volume fraction \(\phi = 0.2\) and filament aspect ratio \(a = 10.5\) . For a given filament stiffness, increasing activity reduces the number of caps until an upper-threshold activity value \(f_{\mathrm{a}}^{\mathrm{SC}}\) , beyond which the system transitions into an undeformed state. As described in SI section E, this transition occurs because activity renormalizes the filament bending modulus to smaller values [41], thus reducing filament alignment interactions and causing the system to leave the strong confinement limit. The dashed line shows the prediction for \(f_{\mathrm{a}}^{\mathrm{SC}}\) given by Eq. S35. Note that there is no adjustable parameter. In the rigid rod limit \((\kappa_{\mathrm{fil}} > 10^{3})\) all non-zero active force values that we simulated led to cap formation. b) Selected snapshots of states shown in (a). Animations of these states can be found in SI Movie 3.
+ +above a threshold activity, which decreases with decreasing \(\kappa_{\mathrm{fil}}\) . + +This result can be understood as follows. On generic grounds, decreasing the filament rigidity will reduce the tendency for filaments to align and thus impede the formation of aligned rings and caps. For filament stiffness values well below the rigid rod limit, the process by which caps and rings form is more complicated than considered previously. The upper- threshold activity for filament organization can be, at least in part, explained by the observation that activity renormalizes filament rigidity to smaller values according to \(\kappa_{\mathrm{fil}}^{\mathrm{eff}} \cong \kappa_{\mathrm{fil}} / \left(1 + f_{\mathrm{a}}^{2}\right)\) [41]. Interactions between flexible active agents is such that the active energy preferentially dissipates into bend modes, + +<--- Page Split ---> +![](images/Figure_6.jpg) + +
FIG. 6. a) Simulation snapshots illustrating vesicle conformations and filament organizations as a function of filament volume fraction and vesicle rigidity. As the vesicle rigidity is reduced below a critical value (corresponding to the critical Foppl-von Karman number \(\mathrm{FvK} \approx 154\) [67]), the vesicle undergoes a buckling transition leading to the formation of facets. While we observe most of the same classes of filament self-organization in faceted and round vesicles, polar bands trace a dynamic path between vertices in faceted vesicles, while they trace a stable geodesic in round vesicles. Animations of these states can be found in SI movie 5. b) Comparisons between flexible and rigid vesicles as a function of filament aspect ratio, with other parameters set to \(f_{\mathrm{a}} = 8\) , \(\phi = 0.1\) , and \(\kappa_{\mathrm{fil}} = 10^{4}\) . In contrast to flexible vesicles, rigid vesicles do not allow for the formation of stable caps or rings. When polar rings do form in rigid vesicles, they are transient—continuously breaking and reforming over the course of the trajectory. Animations of this comparison can be found in SI movie 6.
+ +effectively increasing filament flexibility and therefore suppressing filament alignment. In particular, the upper- threshold activity corresponds to the point when the activity- renormalized flexibility of filaments causes the system to leave the strong confinement limit. This occurs for \(f_{\mathrm{a}} \gtrsim C \kappa_{\mathrm{fil}}^{3 / 5}\) , where \(C = (8 \pi R_{\mathrm{ves}})^{- 1 / 5}\) (see SI Sec. D for details), which is shown as the dashed line in Fig. 5. + +Fig. 6a shows the conformations obtained by varying the vesicle rigidity \(\kappa_{\mathrm{ves}}\) and filament volume fraction \(\phi\) , while fixing the filament rigidity \(\kappa_{\mathrm{fil}} = 10^{4}\) and active force \(f_{\mathrm{a}} = 8\) . The most striking effect of reducing the vesicle rigidity is that it drives a faceting transition when the the Foppl- von Karman number, \(\mathrm{FvK} = Y R_{\mathrm{ves}}^{2} / \kappa\) where \(Y\) is + +the Young's modulus of the vesicle and \(\kappa = \sqrt{3} \kappa_{\mathrm{ves}}\) is the bending modulus [68], is increased above a critical value, \(\mathrm{FvK} \gtrsim 154\) . This is an equilibrium property of an elastic vesicle, independent of the active filaments [67]. Our results indicate that faceting does not qualitatively change the formation of caps, but that caps form at slightly lower filament volume fraction for reduced vesicle bending modulus. This could be anticipated from the theoretical arguments described above, since reducing the bending modulus allows filaments' active forces to further deform the vesicle, leading to a smaller local radius of curvature in the vicinity of a cap. More interestingly, the facets appear to destabilize the polar bands and rings. For round vesicles (with bending modulus such that \(\mathrm{FvK} < 154\) ), a stable ring forms along a geodesic. In contrast, in faceted vesicles at the same activity and filament volume fraction, rings or bands tend to form paths that connect facet vertices. The bending of the ring path imposed by the facet connectivity destabilizes the ring, causing it to transiently break and reform (similar to the transient band state described above). This behavior suggests that it will be interesting to explore the possibility of coupling between vesicle faceting and filament organization in a future work. + +Fig. 6b compares configurations observed with a flexible vesicle ( \(\kappa_{\mathrm{ves}} = 5 \times 10^{3}\) ) and a rigid vesicle ( \(\kappa_{\mathrm{ves}} \to \infty\) ) for \(a \in [15.5, 25.5]\) , \(\phi = 0.10\) , \(f_{\mathrm{a}} = 8\) , and \(\kappa_{\mathrm{fil}} = 10^{4}\) . While the flexible vesicle exhibits stable polar rings and single caps at these parameters (Fig. 1), the rigid vesicle system is unable to form the single- cap state, and only exhibits transient polar rings, which continuously break apart and reform as the simulation progresses. These results emphasize the importance of the feedback between active stress organization and vesicle deformation, which allows for stable states that are otherwise inaccessible under rigid confinement. + +## IV. CONCLUSIONS + +This work demonstrates that confining active filaments within a deformable vesicle leads to multiple transformations of the vesicle shape and motility, which can be precisely tuned by control parameters. The feedback enabled by coupling deformable boundaries with anisotropic particles significantly enriches the available modes of self- organization. While the self- limited caps are the most striking class of such behaviors, the stable polar bands for particles with nematic interactions provides a clear example of how boundary deformations can stabilize novel states. Notably, both of these classes of behaviors arise due to a spontaneous symmetry breaking of the initially spherical boundary. + +These results have implications for future experiments on active materials constructed from anisotropic particles confined within deformable boundaries. In particular, the transitions can be controlled by tuning parameters that are readily accessible in experiments—filament length, + +<--- Page Split ---> + +flexibility, and volume fraction. In contrast, activity is a complicated function of motor properties and ATP in bio- derived systems [69, 70]. Thus, our computational results suggest strategies to engineer active vesicles with designable shapes and dynamics, and other capabilities resembling those of living cells. Furthermore, our theoretical analysis identifies the mechanisms that underlie these emergent morphologies by revealing how filament- filament interactions and vesicle deformations couple to spatiotemporally organize stress. This provides a model- independent roadmap for exploring additional classes of emergent functionalities in parameter regimes beyond the scope of the present work, including highly deformable fluidized vesicles and other symmetries of activity. + +## ACKNOWLEDGMENTS + +We acknowledge support from NSF DMR- 1855914 and the Brandeis Center for Bioinspired Soft Materials, an NSF MRSEC (DMR- 2011846). We also acknowledge computational support from NSF XSEDE computing resources allocation TG- MCB090163 (Stampede and Comet) and the Brandeis HPCC which is partially supported by DMR- MRSEC 2011486. We also acknowledge the KITP Active20 program, during which some of these ideas were developed, which is supported in part by the National Science Foundation under Grant No. NSF PHY- 1748958. + +Author contributions: MSEP, AB, and MFH designed the research; MSEP performed the computational modeling; MSEP, AB, and MFH performed the theoretical modeling; MSEP analyzed the data; and MSEP, AB, and MFH wrote the paper. + +Competing interests: The authors declare that they have no competing interests. + +[1] P. Sens, Proc. Natl. Acad. Sci. U. S. A., 202011785 (2020). [2] R. Ananthakrishnan and A. Ehrlicher, Int. J. Biol. Sci., 303 (2007). [3] M. Leptin and B. Grunewald, Development 110, 73 (1990). [4] A. L. Miller, Curr. Biol. 21, R976 (2011). [5] O. Polyakov, B. He, M. Swan, J. W. Shaevitz, M. Kaschube, and E. Wieschaus, Biophys. J. 107, 998 (2014). [6] J. M. Besterman and R. B. Low, Biochem. J. 210, 1 (1983). [7] G. J. Doherty and H. T. McMahon, Annu. Rev. Biochem. 78, 857 (2009). [8] Y. Maroudas- Sacks, L. Garion, L. Shani- Zerbib, A. Livshits, E. Braun, and K. Keren, Nat. Phys. (2020), 10.1038/s41567- 020- 01083- 1. [9] V. Schaller, C. Weber, C. Semmrich, E. Frey, and A. R. Bausch, Nature 467, 73 (2010). [10] A. Bricard, J.- B. Caussin, N. Desreumaux, O. Dauchot, and D. Bartolo, Nature 503, 95 (2013). [11] A. Senoussi, S. Kashida, R. Voituriez, J.- C. Galas, A. Maitra, and A. Estevez- Torres, Proc. Natl. Acad. Sci. 116, 22464 (2019). [12] T. Strubing, A. Khosravanizadeh, A. Vilfan, E. Bodenschatz, R. Golestanian, and I. Guido, Nano Letters 20, 6281 (2020). [13] S. A. Edwards and J. M. Yeomans, Europhys. Lett. 85, 18008 (2009). [14] F. G. Woodhouse and R. E. Goldstein, Phys. Rev. Lett. 109, 168105 (2012). [15] H. Wioland, F. G. Woodhouse, J. Dunkel, J. O. Kessler, and R. E. Goldstein, Phys. Rev. Lett. 110, 268102 (2013). [16] M. Ravnik and J. M. Yeomans, Phys. Rev. Lett. 110, 026001 (2013). [17] E. Lushi, H. Wioland, and R. E. Goldstein, Proc. Natl. Acad. Sci. U. S. A. 111, 9733 (2014). [18] H. Wioland, E. Lushi, and R. E. Goldstein, New J. Phys. + +18, 075002 (2016). [19] K.- T. Wu, J. B. Hishamunda, D. T. N. Chen, S. J. DeCamp, Y.- W. Chang, A. Fernández- Nieves, S. Fraden, and Z. Dogic, Science 355 (2017). [20] M. Varghese, A. Baskaran, M. F. Hagan, and A. Baskaran, Phys. Rev. Lett. 125, 268003 (2020). [21] Y. Li and P. R. ten Wolde, Phys. Rev. Lett. 123, 148003 (2019). [22] H. R. Vutukuri, M. Hoore, C. Abaurrea- Velasco, L. van Buren, A. Dutto, T. Auth, D. A. Fedosov, G. Gompper, and J. Vermant, Nature 586, 52 (2020). [23] F. C. Keber, E. Loiseau, T. Sanchez, S. J. DeCamp, L. Giomi, M. J. Bowick, M. C. Marchetti, Z. Dogic, and A. R. Bausch, Science 345, 1135 (2014). [24] S. C. Takatori and A. Sahu, Phys. Rev. Lett. 124, 158102 (2020). [25] R. Hughes and J. M. Yeomans, Phys. Rev. E 102, 020601 (2020). [26] P. W. Miller, N. Stoop, and J. Dunkel, Phys. Rev. Lett. 120, 268001 (2018). [27] M. Elbaum, D. K. Fygenson, and A. Libchaber, Phys. Rev. Lett. 76, 4078 (1996). [28] D. K. Fygenson, J. F. Marko, and A. Libchaber, Phys. Rev. Lett. 79, 4497 (1997). [29] M. M. Norton, A. Baskaran, A. Opathalage, B. Langeslay, S. Fraden, A. Baskaran, and M. F. Hagan, Phys. Rev. E 97, 012702 (2018). [30] S. Henkes, M. C. Marchetti, and R. Sknepnek, Phys. Rev. E 97, 042605 (2018). [31] S. Shankar, M. J. Bowick, and M. C. Marchetti, Phys. Rev. X 7, 031039 (2017). [32] F. Alaimo, C. Köhler, and A. Voigt, Sci. Rep. 7, 5211 (2017). [33] M. Paoluzzi, R. Di Leonardo, M. C. Marchetti, and L. Angelani, Sci. Rep. 6, 34146 (2016). [34] C. Wang, Y.- k. Guo, W.- d. Tian, and K. Chen, J. Chem. Phys. 150, 044907 (2019). + +<--- Page Split ---> + +[35] R. Sknepnek and S. Henkes, Phys. Rev. E 91, 022306 (2015). [36] I. R. Bruss and S. C. Glotzer, Soft matter 13, 5117 (2017). [37] C. Abaurrea- Velasco, T. Auth, and G. Gompper, New J. Phys. 21, 123024 (2019). [38] A. C. Quillen, J. P. Smucker, and A. Peshkov, Phys. Rev. E 101, 052618 (2020). [39] G. A. Vliegenthart, A. Ravichandran, M. Ripoll, T. Auth, and G. Gompper, Sci. Adv. 6 (2020), 10.1126/sci- adv.aaw9975. [40] M. Rajabi, H. Baza, T. Turiv, and O. D. Lavrentovich, Nat. Phys. 17, 260 (2020). [41] A. Joshi, E. Putzig, A. Baskaran, and M. F. Hagan, Soft Matter 15, 94 (2019). [42] R. G. Winkler and G. Gompper, J. Chem. Phys. 153, 040901 (2020). [43] K. Kremer and G. S. Grest, J. Chem. Phys. 92, 5057 (1990). [44] J. D. Weeks, D. Chandler, and H. C. Andersen, J. Chem. Phys. 54, 5237 (1971). [45] J. Elgeti and G. Gompper, Europhys. Lett. 101, 48003 (2013). [46] R. E. Isele- Holder, J. Elgeti, and G. Gompper, Soft Matter 11, 7181 (2015). [47] R. E. Isele- Holder, J. Jager, G. Saggiorato, J. Elgeti, and G. Gompper, Soft Matter 12, 8495 (2016). [48] O. Duman, R. E. Isele- Holder, J. Elgeti, and G. Gompper, Soft Matter 14, 4483 (2018). [49] R. Chelakkot, M. F. Hagan, L. Mahadevan, and A. Gopinath, bioRxiv (2020), 10.1101/2020.06.08.140731. [50] S. Plimpton, J. Comp. Phys. 117, 1 (1995). [51] See Supplemental Material at the end of this document for model details and additional figures. [52] Y. Fily, A. Baskaran, and M. F. Hagan, Soft Matter 10, 5609 (2014). [53] F. Peruani, A. Deutsch, and M. Bar, Phys. Rev. E 74, 030904 (2006). [54] F. Peruani, J. Starrub, V. Jakovljevic, L. Sogaard- Andersen, A. Deutsch, and M. Bar, Phys. Rev. Lett. 108, 098102 (2012). [55] A. Baskaran and M. C. Marchetti, Phys. Rev. Lett. 101, 268101 (2008). [56] A. Baskaran and M. C. Marchetti, Phys. Rev. E 77, 011920 (2008). [57] S. R. McCandlish, A. Baskaran, and M. F. Hagan, Soft Matter 8, 2527 (2012). [58] F. Ginelli, F. Peruani, M. Bar, and H. Chate, Phys. Rev. Lett. 104, 184502 (2010). + +[59] M. Bar, R. Grossmann, S. Heidenreich, and F. Peruani, Annu. Rev. Condens. Matter Phys. 11, 441 (2020). [60] E. Bertin, H. Chate, F. Ginelli, S. Mishra, A. Peshkov, and S. Ramaswamy, New J. Phys. 15, 085032 (2013). [61] S. Ngo, A. Peshkov, I. S. Aranson, E. Bertin, F. Ginelli, and H. Chate, Phys. Rev. Lett. 113, 038302 (2014). [62] H. H. Wensink and H. Löwen, Phys. Rev. E 78, 031409 (2008). [63] C. Bechinger, R. Di Leonardo, H. Löwen, C. Reichhardt, G. Volpe, and G. Volpe, Rev. Mod. Phys. 88, 045006 (2016). [64] The ability of collisions of self-propelled particles on a surface to drive formation of smectic layers is supported by a recent observation in bacterial colonies growing on flat surfaces, in which bacteria form 'rosettes' with the rod- like bacteria oriented perpendicular to the surface [71]. However, in contrast to the self- limited caps in our system, the bacterial rosettes do not exhibit a preferred size because they are on a flat boundary. [65] M. F. Hagan and G. M. Grason, "Equilibrium mechanisms of self- limiting assembly," (2021), arXiv:2007.01927 [cond- mat.soft]. [66] N. Yu, A. Ghosh, and M. F. Hagan, Soft Matter 12, 8990 (2016). [67] J. Lidmar, L. Mirny, and D. R. Nelson, Phys. Rev. E 68, 051910 (2003). [68] G. Gompper and D. M. Kroll, J. Phys. I France 6, 1305 (1996). [69] L. M. Lemma, M. M. Norton, S. J. DeCamp, S. A. Aghvami, S. Fraden, M. F. Hagan, and Z. Dogic, arXiv:2006.15184 [cond- mat] (2020), arXiv:2006.15184 [cond- mat]. [70] P. Chandrakar, J. Berezney, B. Lemma, B. Hishamunda, A. Berry, K.- T. Wu, R. Subramanian, J. Chung, D. Needleman, J. Gelles, and Z. Dogic, "Microtubule- based active fluids with improved lifetime, temporal stability and miscibility with passive soft materials," (2018), arXiv:1811.05026 [cond- mat.soft]. [71] O. J. Meacock, A. Doostmohammadi, K. R. Foster, J. M. Yeomans, and W. M. Durham, Nat. Phys. (2020), 10.1038/s41567- 020- 01070- 6. [72] Y. Fily, A. Baskaran, and M. F. Hagan, "Active particles on curved surfaces," (2016), arXiv:1601.00324 [cond- mat.soft]. [73] M. Doi and S. F. Edwards, The Theory of Polymer Dynamics, International Series of Monographs on Physics No. 73 (Clarendon Press, Oxford, 2007). + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1
+ +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2
+ +Please see manuscript .pdf for full caption. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3
+ +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +![](images/Figure_unknown_0.jpg) + +
b)
+ +![](images/Figure_5.jpg) + +
Figure 5
+ +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +![](images/Figure_unknown_1.jpg) + +
b)
+ +![](images/Figure_6.jpg) + +
Figure 6
+ +Please see manuscript.pdf for full caption. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +<--- Page Split ---> + +movie1. mp4 movie2. mp4 movie3. mp4 movie4. mp4 movie5. mp4 movie6. mp4 supplemental.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947_det.mmd b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..57ee7c31f9a19240ad6faebd9388905359fe8f98 --- /dev/null +++ b/preprint/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947/preprint__16af086965d05a012f6b3b613effcc33fa6d655d72b0fb82f2743cd30453f947_det.mmd @@ -0,0 +1,312 @@ +<|ref|>title<|/ref|><|det|>[[42, 107, 910, 175]]<|/det|> +# Vesicle shape transformations driven by confined active filaments + +<|ref|>text<|/ref|><|det|>[[44, 196, 582, 238]]<|/det|> +Matthew Peterson Brandeis University https://orcid.org/0000- 0002- 5749- 7770 + +<|ref|>text<|/ref|><|det|>[[44, 243, 225, 282]]<|/det|> +Apama Baskaran Brandeis University + +<|ref|>text<|/ref|><|det|>[[44, 288, 582, 330]]<|/det|> +Michael Hagan (hagan@brandeis.edu) Brandeis University https://orcid.org/0000- 0002- 9211- 2434 + +<|ref|>sub_title<|/ref|><|det|>[[44, 371, 102, 389]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 408, 910, 451]]<|/det|> +Keywords: Active Matter Systems, Deformable Boundaries, Particle- based Simulations, Active Stress Organization, Emergent Behaviors, Targeted Shape Dynamics + +<|ref|>text<|/ref|><|det|>[[44, 470, 284, 489]]<|/det|> +Posted Date: May 6th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 508, 463, 527]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 448564/v1 + +<|ref|>text<|/ref|><|det|>[[44, 545, 910, 587]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 624, 944, 667]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on December 1st, 2021. See the published version at https://doi.org/10.1038/s41467- 021- 27310- 8. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[186, 63, 816, 81]]<|/det|> +# Vesicle shape transformations driven by confined active filaments + +<|ref|>text<|/ref|><|det|>[[222, 95, 775, 137]]<|/det|> +Matthew S. E. Peterson, Aparna Baskaran,\* and Michael F. Hagan† Martin A. Fisher School of Physics, Brandeis University, Waltham, MA, 02453 (Dated: April 21, 2021) + +<|ref|>text<|/ref|><|det|>[[175, 145, 830, 277]]<|/det|> +In active matter systems, deformable boundaries provide a mechanism to organize internal active stresses and perform work on the external environment. To study a minimal model of such a system, we perform particle- based simulations of an elastic vesicle containing a collection of polar active filaments. The interplay between the active stress organization due to interparticle interactions and that due to the deformability of the confinement leads to a variety of filament spatiotemporal organizations that have not been observed in bulk systems or under rigid confinement, including highly- aligned rings and caps. In turn, these filament assemblies drive dramatic and tunable transformations of the vesicle shape and its dynamics. We present simple scaling models that reveal the mechanisms underlying these emergent behaviors and yield design principles for engineering active materials with targeted shape dynamics. + +<|ref|>sub_title<|/ref|><|det|>[[201, 302, 371, 316]]<|/det|> +## I. INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[86, 335, 488, 580]]<|/det|> +Active matter encompasses systems whose microscopic constituents consume energy at the particle scale to produce forces and motion. Novel macroscale phenomena emerge in these systems when these forces collectively organize into mesoscale 'active stresses'. Harnessing these active stresses to drive particular emergent behaviors could enable a new class of materials with life- like properties that would be impossible in traditional equilibrium materials. However, rationally designing active constituents to generate a desired emergent behavior requires identifying the principles that govern organization of mesoscopic active stresses. Similarly, many biological functions, such as cytoplasmic streaming, morphogenesis, and cell migration, are driven by active stresses that emerge from active components confined within a cell [1- 8]. Understanding physical mechanisms that underlie these functions is a key goal of cellular biophysics. + +<|ref|>text<|/ref|><|det|>[[87, 581, 488, 740]]<|/det|> +As a step toward these goals, this article describes a computational and theoretical study of a minimal model of active filaments enclosed within a deformable vesicle. We thereby identify a generic route to control selforganizing active stresses by enclosing active components with anisotropic shapes and/or internal degrees of freedom within deformable confining boundaries. Our results demonstrate that coupling between boundary deformations and the assembly of internal active components leads to a positive feedback capable of driving diverse nonlinear transformations of morphology and dynamics. + +<|ref|>text<|/ref|><|det|>[[87, 740, 488, 841], [515, 302, 916, 560]]<|/det|> +The field of active matter has identified two key mechanisms that control self- organization of active stresses: (1) anisotropic interactions between active components that realign forces, and (2) confining boundaries. For example, interactions between self- propelled particles that drive interparticle alignment result in bands or flocks [9, 10], changing the length and stiffness of active polymers leads to dramatic reorganization of active stresses [11, 12], and confining active particles leads to system- spanning effects such as spontaneous flow [13- 20]. Furthermore, deformable confining boundaries enable non- equilibrium boundary fluctuations [21- 26], including elongated tendrils and bolas [22]. The latter results suggest that flexibility is a key characteristic of a confining boundary, as it allows shape transformations, sensing and response to environmental cues, and performing work on the surroundings. Achieving such capabilities is critical to leverage minimal bio- derived experimental systems (e.g. [23, 27, 28]) to engineer controlled shape transformations. However, little is known about the behaviors that may arise when these two active stress organization modes are combined. In particular, most existing theoretical and computational studies have focused on rigid boundaries [29- 32], isotropic active particles [21, 22, 24, 33- 36], or have been in 2D [37- 39]. + +<|ref|>text<|/ref|><|det|>[[515, 565, 916, 913]]<|/det|> +In this work, we use Langevin dynamics simulations of polar self- propelled semiflexible filaments confined within 3D flexible vesicles to study the combination of active interparticle alignment interactions and 3D deformable confinement. Our simulations show that interplay between these two organization modes leads to a positive feedback, in which active forces drive boundary deformation while passive stresses from the boundary guide and reinforce self- organization of the internal active stresses. This leads to a rich variety of steady- state behaviors that have not been observed in bulk systems or under rigid confinement, including highly- aligned rings, and caps that have tunable self- limited sizes, number, and symmetry. Each filament organization drives a characteristic large- scale vesicle shape transformation that can be selected by varying parameters such as filament length, density, and flexibility. Asymmetric states lead to net vesicle motion, consistent with the recent experiments which find that enclosing self- propelled particles, such as bacteria, in droplets can lead to collective motility [40]. We present simple scaling analyses that reveal how the feedback between vesicle geometry and filament organization drives and stabilizes these emergent behaviors. The applicability of these scaling arguments suggests that these behaviors + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 66, 488, 109]]<|/det|> +arise generically due to feedback between vesicle elasticity and active filament organization, independent of the specific model. + +<|ref|>text<|/ref|><|det|>[[86, 110, 488, 343]]<|/det|> +Understanding the fundamental mechanisms that govern this coupling between self- organization of active stress and deformations of a flexible boundary will establish design principles for soft robotics, artificial cells, or other advanced materials that mimic the capabilities of living organisms. From a biological perspective, our minimal model is not intended to directly describe the cytoskeleton—in this model, active filament propulsion enters in the overdamped limit without long- ranged hydrodynamic coupling. However, the generic feedback between vesicle elasticity and active filament organization identified in our simulations and scaling arguments may elucidate how mesoscopic active stresses in the cellular cytoskeleton drive large scale cellular shape transformations that underlie essential biological functions such as motility [1, 2], division [3- 5], and endo-/exocytosis [6, 7]. + +<|ref|>sub_title<|/ref|><|det|>[[223, 373, 350, 387]]<|/det|> +## II. METHODS + +<|ref|>text<|/ref|><|det|>[[86, 405, 488, 825]]<|/det|> +We simulate a system of \(N_{\mathrm{fil}}\) active filaments confined within an elastic vesicle, which has radius \(R_{\mathrm{ves}}\) in its undeformed state. We represent active filaments using the model in Joshi, et al. [41]—modified so that the active forcing is polar rather than nematic—in which each filament is a nearly- inextensible, semiflexible chain of \(M\) beads of diameter \(\sigma\) [42]. Bonded beads interact through an expanded FENE potential [43], while non- bonded beads interact through a purely repulsive expanded Weeks- Chandler- Andersen (eWCA) potential [44] with strength \(\epsilon\) . The equilibrium bond length is set to \(b_{\mathrm{fil}} = \sigma /2\) to minimize surface roughness between interacting filaments, thereby preventing filaments from interlocking at high density [45- 49]. The filaments are made semiflexible with bending rigidity \(\kappa_{\mathrm{fil}}\) through a harmonic angle potential applied to each set of three consecutive beads along the chain. Since our model is not intended to describe any specific biofilament system, we incorporate activity in a minimal manner—a polar active force of magnitude \(f_{\mathrm{a}}\) acts on each bead, in a direction tangent to the filament and toward the filament head. The filament volume fraction in the undeformed state is given by \(\phi = N_{\mathrm{fil}}V_{\mathrm{fil}} / V_{\mathrm{ves}}\) , where \(V_{\mathrm{fil}} = \pi \sigma^3 /6 + (M - 1)\pi b\sigma^2 /4\) is the approximate volume of a single filament—accounting for the overlap of bonded monomers—and \(V_{\mathrm{ves}} = 4\pi R_{\mathrm{ves}}^3 /3\) is the nominal volume of the vesicle. Since the mesh topology is conserved in our simulations, we model an elastic vesicle. In subsequent work we plan to consider the effects of fluidizing the vesicle and imposing area or volume constraints. + +<|ref|>text<|/ref|><|det|>[[86, 826, 488, 912]]<|/det|> +We simulate the coupled Langevin equations for the filament and vesicle bead dynamics using LAMMPS [50], modified to include the active force. We neglect long- ranged hydrodynamic interactions for this system of high filament density; we will investigate their effect in a future study. We have set units such that the mass of all beads is + +<|ref|>image<|/ref|><|det|>[[543, 60, 890, 297]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[515, 310, 919, 576]]<|/det|> +
FIG. 1. Snapshots illustrating steady-state configurations of the vesicle and enclosed active filaments as a function of filament aspect ratio \(a\) and initial volume fraction \(\phi\) . See SI Movie 1 for animations of the corresponding simulations [51]. The marked regions of parameter space indicate the typical vesicle conformation: (I) spherical, (II) oblate, (III) polar-prolate, (IV) apolar-prolate, and (V) polyhedral. The symbols associate the conformation with the internal filament organization: homogeneous throughout the bulk or on the surface, with no vesicle deformation \((\bullet)\) ; transient clusters and/or bands, with oblate vesicle shapes \((\star)\) ; stable polar rings \((\circ)\) ; stable caps \((\circ\) , with a number of intersecting lines equal to the median number of caps); and dynamic caps \((\diamondsuit)\) . The dashed line shows the transition to aligned states predicted from the competition between the characteristic collision and reorientation timescales \((\phi = (\pi /4)^2 /a)\) described in the text, and the horizontal dotted line indicates the approximate threshold aspect ratio for the filaments to be in the strong confinement limit. Other parameters are filament bending modulus \(\kappa_{\mathrm{fil}} = 10^4\) and vesicle radius \(R_{\mathrm{ves}} = 25\) .
+ +<|ref|>text<|/ref|><|det|>[[515, 602, 917, 660]]<|/det|> +\(m = 1\) , and energies, lengths, and time are respectively in units of \(k_{\mathrm{B}}T\) , \(\sigma\) , and \(\tau = \sqrt{m\sigma^2 / \epsilon}\) . The friction constant is set to \(\gamma = 1 / \tau\) . For additional model details, see the Supplemental Materials [51]. + +<|ref|>sub_title<|/ref|><|det|>[[576, 688, 856, 703]]<|/det|> +## III. RESULTS AND DISCUSSION + +<|ref|>sub_title<|/ref|><|det|>[[630, 721, 803, 735]]<|/det|> +### A. Simulation results + +<|ref|>text<|/ref|><|det|>[[515, 753, 917, 869]]<|/det|> +To discover the steady- state conformations that arise due to coupling between active propulsion and elasticity, we have performed simulations over a wide range of control parameters (Fig. 1, Fig. 3, Fig. 5, and Fig. 6): the volume fraction of filaments in the vesicle \(\phi \in [0.01, 0.4]\) , filament aspect ratio \(a = 1 + (M - 1)(b_{\mathrm{fil}} / \sigma) \in [3, 25.5]\) , active propulsion strength \(f_{\mathrm{a}} \in [0, 10]\) , filament stiffness \(\kappa_{\mathrm{fil}} \in [10^2, 10^4]\) , and vesicle rigidity \(\kappa_{\mathrm{ves}} \in [10^2, 10^4]\) . + +<|ref|>text<|/ref|><|det|>[[515, 869, 917, 912]]<|/det|> +Fig. 1 shows the steady- states as a function of filament volume fraction and aspect ratio for moderate activity \(f_{\mathrm{a}} = 8\) . At this activity and vesicle size, for aspect + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[123, 60, 450, 191]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 203, 488, 350]]<|/det|> +
FIG. 2. a) The onset of ring and cap formation is determined by a competition of timescales: the timescale associated with rotations parallel to the vesicle (left) and the timescale associated with collisions that tend to orient the filaments perpendicular to the vesicle (right). b) Schematic of the theory for the number of caps (Eqs. (1) and (2)). We assume an activity-induced effective attractive interaction that is quadratic in the rod-rod contact length \(\Delta l\) (left). The cap is assumed circular, with size parameterized by the angle \(\theta\) between the cap center and edge. Vesicle curvature leads to a shearing of rods within the cap (right).
+ +<|ref|>text<|/ref|><|det|>[[86, 380, 488, 471]]<|/det|> +ratios \(a\gtrsim (8\pi R_{\mathrm{ves}} / f_{\mathrm{a}})^{1 / 3}\approx 4.3\) the system is in the strong confinement limit: because the persistence length \(l_{\mathrm{p}}^{\mathrm{COM}}\propto f_{\mathrm{a}}a^{3}\) of the filament center- of- mass motion is larger than the vesicle size \(l_{\mathrm{p}}^{\mathrm{COM}} > 2R_{\mathrm{ves}}\) , most filaments are found on the vesicle surface at all times [52] (see SI Sec. A [51]). + +<|ref|>text<|/ref|><|det|>[[86, 473, 488, 560]]<|/det|> +Under these conditions we can classify the steady- state vesicle conformations into several categories: (I) spherical, (II) oblate, (III) polar- prolate, (IV) apolar- prolate, and (V) polyhedral. These vesicle configurations are tightly coupled to the spatiotemporal organization of the filaments within, as follows. + +<|ref|>text<|/ref|><|det|>[[86, 562, 488, 679]]<|/det|> +(I): Spherical vesicle shapes arise at low filament volume fractions and aspect ratios. Under these conditions, filament- filament collisions are rare and inter- filament aligning forces are weak [53- 59]. Thus, filament positions and orientations are homogeneous (throughout the vesicle interior below strong confinement, or on the vesicle surface above strong confinement), leading to little deformation of the vesicle. + +<|ref|>text<|/ref|><|det|>[[86, 681, 488, 911]]<|/det|> +(II): For low volume fraction but high aspect ratios, such that the filament length \(L = a\sigma\) is comparable to the unperturbed vesicle radius, \(L\sim R_{\mathrm{ves}}\) , the vesicle deforms into oblate spheroid conformations. This transition is driven by the filaments organizing into a stable polar band, which deforms the vesicle along a geodesic. This filament arrangement closely resembles the polar bands observed on the surface of rigid spheres for active particles with polar propulsion and polar interparticle alignment interactions [35], which arise due to topological requirements for a surface- constrained polarization field [31]. However, note that such polar bands would be unstable in our system if the confining geometry was a rigid sphere because the filament- filament interactions in our system are nematic (head- tail symmetric) [60, 61]. The finite deformability of the vesicle is essential to stabilize this + +<|ref|>text<|/ref|><|det|>[[516, 66, 917, 211]]<|/det|> +configuration—active forces due to the polar band force the vesicle into an oblate shape, which in turn provides a restoring force to stabilize filament alignment within the band. In support of this conclusion, simulations on infinitely rigid vesicles did not exhibit stable polar bands (see Fig. 6b and SI movie 6). Thus, this configuration provides a concrete example of how feedback between passive stresses and self- organization of active stresses can generate steady states that would be otherwise disallowed by symmetry. + +<|ref|>text<|/ref|><|det|>[[516, 213, 917, 587]]<|/det|> +(III- IV): For intermediate volume fractions and aspect ratios, the vesicle deforms into a prolate spheroid. These prolate vesicle conformations can be further classified by their motion, either polar (III) or apolar (IV). Further increasing the volume fraction or decreasing the aspect ratio leads to polyhedral conformations, (V). States (III- V) all result from filaments assembling into crystalline caps in which the rods are highly aligned and perpendicular to the vesicle surface. Interestingly, the caps are 'self- limited' in that their typical size decreases with decreasing aspect ratio, but is roughly independent of the total number of filaments \(N_{\mathrm{fil}}\) in the vesicle. Increasing \(N_{\mathrm{fil}}\) at fixed aspect ratio increases the number of caps; we observe up to 12 caps for the finite vesicle size that we consider (Fig. 4). Further, caps drive local curvature of the vesicle, leading to elasticity- mediated cap- cap repulsions which favor symmetric arrangements of caps. Thus, the vesicle morphology can be sensitively tuned by controlling filament aspect ratio and density to achieve a specific number of caps. The polar- prolate (III), apolar- prolate (IV), and polyhedral states (V) respectively have 1, 2, and \(\geq 3\) caps. Generally, states with two or more caps do not exhibit directed motion. However, for enough caps in the vesicle (typically more than 3), the caps themselves can become motile, and collide with, merge with, and split from other caps (see below). + +<|ref|>sub_title<|/ref|><|det|>[[516, 619, 914, 647]]<|/det|> +### B. Mechanisms underlying stress organization and deformation + +<|ref|>text<|/ref|><|det|>[[516, 665, 917, 867]]<|/det|> +To understand how these conformations are governed by the interplay between propulsion- induced aligning forces, vesicle deformability, and vesicle curvature, we develop simple scaling estimates for the timescales and forces that govern filament alignment and interactions with the vesicle. First, we consider the transition between undeformed spherical vesicle states characterized by unaligned or weakly aligned filaments as in state (I), to the highly deformed oblate, prolate, and polyhedral vesicle shapes of states (II- V). Our simulations demonstrate that such significant vesicle shape deformations occur when filament- filament interactions mediate the organization of ordered structures either in the plane of the vesicle or orthogonal to it. + +<|ref|>text<|/ref|><|det|>[[516, 870, 917, 911]]<|/det|> +a. The onset of filament assembly: The onset of this transition can be understood by considering a competition between two characteristic timescales that respec + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[110, 60, 460, 255]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 270, 488, 351]]<|/det|> +
FIG. 3. Steady-state configurations as a function of \(\phi\) and active force \(f_{\mathrm{a}}\) . The marked regions are defined as in Fig. 1. The dashed line shows the transition to aligned states predicted by the timescale competition, which is independent of \(f_{\mathrm{a}}\) . Other parameters are \(a = 10.5\) and \(\kappa_{\mathrm{fl}} = 10^{4}\) . See SI Movie 2 for corresponding animations.
+ +<|ref|>text<|/ref|><|det|>[[86, 384, 488, 588]]<|/det|> +tively govern collision- induced filament- vesicle alignment and filament- filament alignment (see Fig. 2). Filament- vesicle collisions, which tend to reorient filaments parallel to the surface [62, 63], have a characteristic timescale \(\tau_{\mathrm{rot}} \sim L / v_{0}\) [64], with \(v_{0} = f_{\mathrm{a}} / \gamma\) the filament self- propulsion velocity. We can estimate the timescale for filament interactions by considering filament- filament pairwise collisions whose timescale is given by \(\tau_{\mathrm{coll}} \sim \sigma / v_{0} \phi\) (see SI [51]). Thus, deformed vesicle states will arise when \(\tau_{\mathrm{coll}} < \tau_{\mathrm{rot}}\) or equivalently \(a \phi > c\) , where \(c \cong (\pi / 4)^{2}\) is independent of activity and filament length (see [51]). This defines a boundary separating highly deformed states of the vesicle from the undeformed spherical states (the dashed line in Fig. 1). + +<|ref|>text<|/ref|><|det|>[[86, 590, 488, 750]]<|/det|> +Notably, the active force drops out of this argument because both collision and reorientation times are \(\propto f_{\mathrm{a}}\) . Thus, the theory predicts that the emergence of deformed vesicle states is independent of activity of the enclosed filaments (above a threshold activity). As a test of this prediction, Fig. 3 shows the steady- states as a function of \(\phi\) and \(f_{\mathrm{a}}\) for fixed aspect ratio \(a = 10.5\) . Indeed, formation of large deformations does not depend on activity, with non- spherical shapes forming for \(\phi \geq c / a \approx 0.06\) (as predicted by the above timescale argument) for all \(f_{\mathrm{a}} > 0\) that we considered. + +<|ref|>text<|/ref|><|det|>[[86, 754, 488, 913]]<|/det|> +This simple theoretical picture gives a predictive principle, in terms of properties of the active filaments, for when vesicle shape transformations occur. However, the theory assumes the strong activity, long filament limit and thus neglects thermal noise. Below a threshold activity ( \(f_{\mathrm{a}} \lesssim 1\) in our units) the vesicle will not deform because filament organization is destroyed by thermal fluctuations. Also, cap formation (and thus vesicle shape transitions) do not occur when the filaments are below the strong confinement limit discussed above ( \(a \lesssim 4.3\) for the parameters of Fig. 1, shown as a dotted line). + +<|ref|>image<|/ref|><|det|>[[543, 61, 890, 257]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[515, 270, 918, 377]]<|/det|> +
FIG. 4. The number of caps measured in simulations (symbols) compared to the theory (Eqs. (1) and (2), dashed line), with \(A = 4R_{\mathrm{ves}} / \sqrt{3\pi \sigma}\) and \(a^{*} \approx 130\) (chosen by eye). Diamonds indicate dynamic cap states. Note that the number of caps in the simulation results is likely under counted for the dynamic states due to the caps' motility. The simulation data is the same as in Fig. 1. Active filaments are colored by which cap they belong to for visual clarity.
+ +<|ref|>text<|/ref|><|det|>[[515, 404, 917, 737]]<|/det|> +b. Cap morphologies: We can derive further insight into shape transformations by considering the system in the strongly deformed regime with polyhedral shapes. The defining characteristic underlying these states is filament assembly into well-ordered caps. Most cap states are relatively static, with occasional association/dissociation of individual rods (See SI Movie 4), except for the parameters that lead to the highly dynamic, reconfiguring caps discussed below. In a static steady state, the active and elastic forces must balance. In particular, the dense crystalline nature of caps arises because the active force and the presence of the vesicle surface leads to an effective attractive interaction between nearby filaments. This attraction drives radial growth of a cap, since filaments on the cap periphery have fewer neighbors, leading to an effective interfacial tension. This effect is both reinforced by and competes with vesicle elasticity. The active force of small caps drives vesicle deformations whose local curvature enhances effective filament-filament attractions. However, as the cap grows in radius, vesicle curvature drives an effective shear of filaments (see Fig. 2b) that reduces rod-rod overlaps and thus opposes the active force. + +<|ref|>text<|/ref|><|det|>[[515, 739, 917, 913]]<|/det|> +We describe this competition by constructing an effective 'free energy' whose gradients correspond to the active and passive forces (Fig. 2b). Since the active force favors rods to align in a smectic layer, the shear due to vesicle curvature imposes an 'energy' cost of \(U_{\mathrm{shear}}(\theta) = n_{\mathrm{cap}} 2\pi R_{\mathrm{ves}}^{2} \frac{G}{2} [\cos \theta + \sec \theta - 2]\) , with \(\theta\) the angle subtended by the cap on the vesicle surface, \(n_{\mathrm{cap}}\) the number of caps, and a 'shear modulus' \(G \sim f_{\mathrm{a}}\) (but independent of \(L_{\mathrm{rod}}\) ) [51]. In the strongly deformed region the caps are roughly circular, so the interfacial energy is given by \(U_{\mathrm{int}}(\theta) = n_{\mathrm{cap}} 2\pi R_{\mathrm{ves}} \gamma \sin \theta\) , with the 'interfacial tension' \(\lambda \sim L_{\mathrm{rod}} f_{\mathrm{a}}\) accounting for the diminished inter + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 64, 488, 95]]<|/det|> +actions at the cap boundary. This results in a free energy as a function of cap size [51]: + +<|ref|>equation<|/ref|><|det|>[[106, 105, 487, 140]]<|/det|> +\[f(\theta) = \frac{1}{1 - \cos\theta}\left[\frac{1}{2} (\cos \theta +\sec \theta -2) + \zeta \sin \theta\right] \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[85, 151, 488, 225]]<|/det|> +where \(\zeta = G / \gamma R_{\mathrm{ves}}\sim L / R_{\mathrm{ves}}\) is given by the balance between the effective interfacial tension and shear modulus, and should be linear in filament length but roughly independent of \(f_{\mathrm{a}}\) since both of these effects are driven by activity. + +<|ref|>text<|/ref|><|det|>[[85, 226, 488, 283]]<|/det|> +Minimizing this per- filament free energy yields an optimal \(\theta\) [65, 66] corresponding to the self- limited cap size. Assuming that we are well above the onset of cap formation so that essentially all filaments are in caps, + +<|ref|>equation<|/ref|><|det|>[[185, 293, 487, 312]]<|/det|> +\[n_{\mathrm{cap}}\propto \phi a^{-1}[1 + (a / a^{*})^{-2 / 3}] \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[85, 325, 488, 383]]<|/det|> +where \(a^{*}\propto R_{\mathrm{ves}} / \sigma\) is an adjustable parameter that may depend on activity. This expression holds provided \(a\ll\) \(a^{*}\) . For the data in Fig. 1, we obtain \(a^{*}\approx 130\) , leading to the dashed line shown in Fig. 4. + +<|ref|>text<|/ref|><|det|>[[86, 384, 488, 573]]<|/det|> +Except for states with many \((n_{\mathrm{cap}}\gtrsim 7)\) motile caps, there is close agreement between the observed and predicted \(n_{\mathrm{cap}}\) . Above this threshold our cap- counting algorithm likely under counts \(n_{\mathrm{cap}}\) , since different caps are often adjacent and interacting. Further, the prediction of Eqs. (1) and (2) that the self- limited cap size is independent of activity is consistent with observations at different \(f_{\mathrm{a}}\) (see Fig. 3). The motile cap states appear to arise when the curved vesicle geometry forces interactions between the inward- facing ends of adjacent caps. Such interactions occur above a threshold number and aspect ratio of filaments, given by \(N_{\mathrm{fil}}\gtrsim C(1 - a\sigma /R_{\mathrm{ves}})^2\) , where \(C\) is a constant (see SI Sec. D [51]). + +<|ref|>text<|/ref|><|det|>[[86, 574, 488, 704]]<|/det|> +We note that the geometric factors governing the selflimited cap size parallel those in a recently studied equilibrium system of rigid filaments end- adsorbed onto a rigid spherical nanoparticle, which self- assemble due to direct pairwise inter- filament attractions [66]. However, in the present system, the effective filament- filament interactions and vesicle geometry are many- body and emergent in that they arise due to feedback between non- equilibrium active forces and vesicle deformations. + +<|ref|>sub_title<|/ref|><|det|>[[129, 734, 444, 749]]<|/det|> +### C. Effect of filament and vesicle rigidity + +<|ref|>text<|/ref|><|det|>[[86, 767, 488, 840]]<|/det|> +Thus far, we have focused on the interplay between activity and vesicle deformability by performing simulations in the limit of rigid rods, \(\kappa_{\mathrm{fil}} = 10^{4}\) , and high (but finite) vesicle rigidity \(\kappa_{\mathrm{ves}} = 5\times 10^{3}\) . We now briefly discuss the effect of allowing for finite filament and vesicle flexibility. + +<|ref|>text<|/ref|><|det|>[[86, 840, 488, 912]]<|/det|> +Fig. 5 shows the vesicle conformation and filament organization states as a function of filament bending modulus and activity, for fixed filament volume fraction \(\phi = 0.2\) . We see that for finite filament flexibility, the transition to aligned ring and cap states is suppressed + +<|ref|>image<|/ref|><|det|>[[536, 63, 896, 442]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[515, 464, 919, 675]]<|/det|> +
FIG. 5. a) Vesicle conformations and filament organizations as a function of filament rigidity \(\kappa_{\mathrm{fil}}\) and active force strength \(f_{\mathrm{a}}\) , for volume fraction \(\phi = 0.2\) and filament aspect ratio \(a = 10.5\) . For a given filament stiffness, increasing activity reduces the number of caps until an upper-threshold activity value \(f_{\mathrm{a}}^{\mathrm{SC}}\) , beyond which the system transitions into an undeformed state. As described in SI section E, this transition occurs because activity renormalizes the filament bending modulus to smaller values [41], thus reducing filament alignment interactions and causing the system to leave the strong confinement limit. The dashed line shows the prediction for \(f_{\mathrm{a}}^{\mathrm{SC}}\) given by Eq. S35. Note that there is no adjustable parameter. In the rigid rod limit \((\kappa_{\mathrm{fil}} > 10^{3})\) all non-zero active force values that we simulated led to cap formation. b) Selected snapshots of states shown in (a). Animations of these states can be found in SI Movie 3.
+ +<|ref|>text<|/ref|><|det|>[[515, 707, 917, 736]]<|/det|> +above a threshold activity, which decreases with decreasing \(\kappa_{\mathrm{fil}}\) . + +<|ref|>text<|/ref|><|det|>[[515, 737, 919, 912]]<|/det|> +This result can be understood as follows. On generic grounds, decreasing the filament rigidity will reduce the tendency for filaments to align and thus impede the formation of aligned rings and caps. For filament stiffness values well below the rigid rod limit, the process by which caps and rings form is more complicated than considered previously. The upper- threshold activity for filament organization can be, at least in part, explained by the observation that activity renormalizes filament rigidity to smaller values according to \(\kappa_{\mathrm{fil}}^{\mathrm{eff}} \cong \kappa_{\mathrm{fil}} / \left(1 + f_{\mathrm{a}}^{2}\right)\) [41]. Interactions between flexible active agents is such that the active energy preferentially dissipates into bend modes, + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[85, 65, 480, 432]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[85, 443, 490, 682]]<|/det|> +
FIG. 6. a) Simulation snapshots illustrating vesicle conformations and filament organizations as a function of filament volume fraction and vesicle rigidity. As the vesicle rigidity is reduced below a critical value (corresponding to the critical Foppl-von Karman number \(\mathrm{FvK} \approx 154\) [67]), the vesicle undergoes a buckling transition leading to the formation of facets. While we observe most of the same classes of filament self-organization in faceted and round vesicles, polar bands trace a dynamic path between vertices in faceted vesicles, while they trace a stable geodesic in round vesicles. Animations of these states can be found in SI movie 5. b) Comparisons between flexible and rigid vesicles as a function of filament aspect ratio, with other parameters set to \(f_{\mathrm{a}} = 8\) , \(\phi = 0.1\) , and \(\kappa_{\mathrm{fil}} = 10^{4}\) . In contrast to flexible vesicles, rigid vesicles do not allow for the formation of stable caps or rings. When polar rings do form in rigid vesicles, they are transient—continuously breaking and reforming over the course of the trajectory. Animations of this comparison can be found in SI movie 6.
+ +<|ref|>text<|/ref|><|det|>[[86, 717, 488, 823]]<|/det|> +effectively increasing filament flexibility and therefore suppressing filament alignment. In particular, the upper- threshold activity corresponds to the point when the activity- renormalized flexibility of filaments causes the system to leave the strong confinement limit. This occurs for \(f_{\mathrm{a}} \gtrsim C \kappa_{\mathrm{fil}}^{3 / 5}\) , where \(C = (8 \pi R_{\mathrm{ves}})^{- 1 / 5}\) (see SI Sec. D for details), which is shown as the dashed line in Fig. 5. + +<|ref|>text<|/ref|><|det|>[[86, 827, 488, 912]]<|/det|> +Fig. 6a shows the conformations obtained by varying the vesicle rigidity \(\kappa_{\mathrm{ves}}\) and filament volume fraction \(\phi\) , while fixing the filament rigidity \(\kappa_{\mathrm{fil}} = 10^{4}\) and active force \(f_{\mathrm{a}} = 8\) . The most striking effect of reducing the vesicle rigidity is that it drives a faceting transition when the the Foppl- von Karman number, \(\mathrm{FvK} = Y R_{\mathrm{ves}}^{2} / \kappa\) where \(Y\) is + +<|ref|>text<|/ref|><|det|>[[516, 65, 919, 412]]<|/det|> +the Young's modulus of the vesicle and \(\kappa = \sqrt{3} \kappa_{\mathrm{ves}}\) is the bending modulus [68], is increased above a critical value, \(\mathrm{FvK} \gtrsim 154\) . This is an equilibrium property of an elastic vesicle, independent of the active filaments [67]. Our results indicate that faceting does not qualitatively change the formation of caps, but that caps form at slightly lower filament volume fraction for reduced vesicle bending modulus. This could be anticipated from the theoretical arguments described above, since reducing the bending modulus allows filaments' active forces to further deform the vesicle, leading to a smaller local radius of curvature in the vicinity of a cap. More interestingly, the facets appear to destabilize the polar bands and rings. For round vesicles (with bending modulus such that \(\mathrm{FvK} < 154\) ), a stable ring forms along a geodesic. In contrast, in faceted vesicles at the same activity and filament volume fraction, rings or bands tend to form paths that connect facet vertices. The bending of the ring path imposed by the facet connectivity destabilizes the ring, causing it to transiently break and reform (similar to the transient band state described above). This behavior suggests that it will be interesting to explore the possibility of coupling between vesicle faceting and filament organization in a future work. + +<|ref|>text<|/ref|><|det|>[[516, 413, 919, 585]]<|/det|> +Fig. 6b compares configurations observed with a flexible vesicle ( \(\kappa_{\mathrm{ves}} = 5 \times 10^{3}\) ) and a rigid vesicle ( \(\kappa_{\mathrm{ves}} \to \infty\) ) for \(a \in [15.5, 25.5]\) , \(\phi = 0.10\) , \(f_{\mathrm{a}} = 8\) , and \(\kappa_{\mathrm{fil}} = 10^{4}\) . While the flexible vesicle exhibits stable polar rings and single caps at these parameters (Fig. 1), the rigid vesicle system is unable to form the single- cap state, and only exhibits transient polar rings, which continuously break apart and reform as the simulation progresses. These results emphasize the importance of the feedback between active stress organization and vesicle deformation, which allows for stable states that are otherwise inaccessible under rigid confinement. + +<|ref|>sub_title<|/ref|><|det|>[[631, 617, 801, 631]]<|/det|> +## IV. CONCLUSIONS + +<|ref|>text<|/ref|><|det|>[[516, 650, 919, 839]]<|/det|> +This work demonstrates that confining active filaments within a deformable vesicle leads to multiple transformations of the vesicle shape and motility, which can be precisely tuned by control parameters. The feedback enabled by coupling deformable boundaries with anisotropic particles significantly enriches the available modes of self- organization. While the self- limited caps are the most striking class of such behaviors, the stable polar bands for particles with nematic interactions provides a clear example of how boundary deformations can stabilize novel states. Notably, both of these classes of behaviors arise due to a spontaneous symmetry breaking of the initially spherical boundary. + +<|ref|>text<|/ref|><|det|>[[516, 840, 919, 912]]<|/det|> +These results have implications for future experiments on active materials constructed from anisotropic particles confined within deformable boundaries. In particular, the transitions can be controlled by tuning parameters that are readily accessible in experiments—filament length, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[86, 65, 488, 268]]<|/det|> +flexibility, and volume fraction. In contrast, activity is a complicated function of motor properties and ATP in bio- derived systems [69, 70]. Thus, our computational results suggest strategies to engineer active vesicles with designable shapes and dynamics, and other capabilities resembling those of living cells. Furthermore, our theoretical analysis identifies the mechanisms that underlie these emergent morphologies by revealing how filament- filament interactions and vesicle deformations couple to spatiotemporally organize stress. This provides a model- independent roadmap for exploring additional classes of emergent functionalities in parameter regimes beyond the scope of the present work, including highly deformable fluidized vesicles and other symmetries of activity. + +<|ref|>sub_title<|/ref|><|det|>[[620, 66, 814, 80]]<|/det|> +## ACKNOWLEDGMENTS + +<|ref|>text<|/ref|><|det|>[[515, 98, 919, 256]]<|/det|> +We acknowledge support from NSF DMR- 1855914 and the Brandeis Center for Bioinspired Soft Materials, an NSF MRSEC (DMR- 2011846). We also acknowledge computational support from NSF XSEDE computing resources allocation TG- MCB090163 (Stampede and Comet) and the Brandeis HPCC which is partially supported by DMR- MRSEC 2011486. We also acknowledge the KITP Active20 program, during which some of these ideas were developed, which is supported in part by the National Science Foundation under Grant No. NSF PHY- 1748958. + +<|ref|>text<|/ref|><|det|>[[515, 257, 919, 330]]<|/det|> +Author contributions: MSEP, AB, and MFH designed the research; MSEP performed the computational modeling; MSEP, AB, and MFH performed the theoretical modeling; MSEP analyzed the data; and MSEP, AB, and MFH wrote the paper. + +<|ref|>text<|/ref|><|det|>[[515, 330, 919, 359]]<|/det|> +Competing interests: The authors declare that they have no competing interests. + +<|ref|>text<|/ref|><|det|>[[86, 410, 490, 904]]<|/det|> +[1] P. Sens, Proc. Natl. Acad. Sci. U. S. A., 202011785 (2020). [2] R. Ananthakrishnan and A. Ehrlicher, Int. J. Biol. Sci., 303 (2007). [3] M. Leptin and B. Grunewald, Development 110, 73 (1990). [4] A. L. Miller, Curr. Biol. 21, R976 (2011). [5] O. Polyakov, B. He, M. Swan, J. W. Shaevitz, M. Kaschube, and E. Wieschaus, Biophys. J. 107, 998 (2014). [6] J. M. Besterman and R. B. Low, Biochem. J. 210, 1 (1983). [7] G. J. Doherty and H. T. McMahon, Annu. Rev. Biochem. 78, 857 (2009). [8] Y. Maroudas- Sacks, L. Garion, L. Shani- Zerbib, A. Livshits, E. Braun, and K. Keren, Nat. Phys. (2020), 10.1038/s41567- 020- 01083- 1. [9] V. Schaller, C. Weber, C. Semmrich, E. Frey, and A. R. Bausch, Nature 467, 73 (2010). [10] A. Bricard, J.- B. Caussin, N. Desreumaux, O. Dauchot, and D. Bartolo, Nature 503, 95 (2013). [11] A. Senoussi, S. Kashida, R. Voituriez, J.- C. Galas, A. Maitra, and A. Estevez- Torres, Proc. Natl. Acad. Sci. 116, 22464 (2019). [12] T. Strubing, A. Khosravanizadeh, A. Vilfan, E. Bodenschatz, R. Golestanian, and I. Guido, Nano Letters 20, 6281 (2020). [13] S. A. Edwards and J. M. Yeomans, Europhys. Lett. 85, 18008 (2009). [14] F. G. Woodhouse and R. E. Goldstein, Phys. Rev. Lett. 109, 168105 (2012). [15] H. Wioland, F. G. Woodhouse, J. Dunkel, J. O. Kessler, and R. E. Goldstein, Phys. Rev. Lett. 110, 268102 (2013). [16] M. Ravnik and J. M. Yeomans, Phys. Rev. Lett. 110, 026001 (2013). [17] E. Lushi, H. Wioland, and R. E. Goldstein, Proc. Natl. Acad. Sci. U. S. A. 111, 9733 (2014). [18] H. Wioland, E. Lushi, and R. E. Goldstein, New J. Phys. + +<|ref|>text<|/ref|><|det|>[[515, 411, 920, 902]]<|/det|> +18, 075002 (2016). [19] K.- T. Wu, J. B. Hishamunda, D. T. N. Chen, S. J. DeCamp, Y.- W. Chang, A. Fernández- Nieves, S. Fraden, and Z. Dogic, Science 355 (2017). [20] M. Varghese, A. Baskaran, M. F. Hagan, and A. Baskaran, Phys. Rev. Lett. 125, 268003 (2020). [21] Y. Li and P. R. ten Wolde, Phys. Rev. Lett. 123, 148003 (2019). [22] H. R. Vutukuri, M. Hoore, C. Abaurrea- Velasco, L. van Buren, A. Dutto, T. Auth, D. A. Fedosov, G. Gompper, and J. Vermant, Nature 586, 52 (2020). [23] F. C. Keber, E. Loiseau, T. Sanchez, S. J. DeCamp, L. Giomi, M. J. Bowick, M. C. Marchetti, Z. Dogic, and A. R. Bausch, Science 345, 1135 (2014). [24] S. C. Takatori and A. Sahu, Phys. Rev. Lett. 124, 158102 (2020). [25] R. Hughes and J. M. Yeomans, Phys. Rev. E 102, 020601 (2020). [26] P. W. Miller, N. Stoop, and J. Dunkel, Phys. Rev. Lett. 120, 268001 (2018). [27] M. Elbaum, D. K. Fygenson, and A. Libchaber, Phys. Rev. Lett. 76, 4078 (1996). [28] D. K. Fygenson, J. F. Marko, and A. Libchaber, Phys. Rev. Lett. 79, 4497 (1997). [29] M. M. Norton, A. Baskaran, A. Opathalage, B. Langeslay, S. Fraden, A. Baskaran, and M. F. Hagan, Phys. Rev. E 97, 012702 (2018). [30] S. Henkes, M. C. Marchetti, and R. Sknepnek, Phys. Rev. E 97, 042605 (2018). [31] S. Shankar, M. J. Bowick, and M. C. Marchetti, Phys. Rev. X 7, 031039 (2017). [32] F. Alaimo, C. Köhler, and A. Voigt, Sci. Rep. 7, 5211 (2017). [33] M. Paoluzzi, R. Di Leonardo, M. C. Marchetti, and L. Angelani, Sci. Rep. 6, 34146 (2016). [34] C. Wang, Y.- k. Guo, W.- d. Tian, and K. Chen, J. Chem. Phys. 150, 044907 (2019). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[84, 66, 492, 707]]<|/det|> +[35] R. Sknepnek and S. Henkes, Phys. Rev. E 91, 022306 (2015). [36] I. R. Bruss and S. C. Glotzer, Soft matter 13, 5117 (2017). [37] C. Abaurrea- Velasco, T. Auth, and G. Gompper, New J. Phys. 21, 123024 (2019). [38] A. C. Quillen, J. P. Smucker, and A. Peshkov, Phys. Rev. E 101, 052618 (2020). [39] G. A. Vliegenthart, A. Ravichandran, M. Ripoll, T. Auth, and G. Gompper, Sci. Adv. 6 (2020), 10.1126/sci- adv.aaw9975. [40] M. Rajabi, H. Baza, T. Turiv, and O. D. Lavrentovich, Nat. Phys. 17, 260 (2020). [41] A. Joshi, E. Putzig, A. Baskaran, and M. F. Hagan, Soft Matter 15, 94 (2019). [42] R. G. Winkler and G. Gompper, J. Chem. Phys. 153, 040901 (2020). [43] K. Kremer and G. S. Grest, J. Chem. Phys. 92, 5057 (1990). [44] J. D. Weeks, D. Chandler, and H. C. Andersen, J. Chem. Phys. 54, 5237 (1971). [45] J. Elgeti and G. Gompper, Europhys. Lett. 101, 48003 (2013). [46] R. E. Isele- Holder, J. Elgeti, and G. Gompper, Soft Matter 11, 7181 (2015). [47] R. E. Isele- Holder, J. Jager, G. Saggiorato, J. Elgeti, and G. Gompper, Soft Matter 12, 8495 (2016). [48] O. Duman, R. E. Isele- Holder, J. Elgeti, and G. Gompper, Soft Matter 14, 4483 (2018). [49] R. Chelakkot, M. F. Hagan, L. Mahadevan, and A. Gopinath, bioRxiv (2020), 10.1101/2020.06.08.140731. [50] S. Plimpton, J. Comp. Phys. 117, 1 (1995). [51] See Supplemental Material at the end of this document for model details and additional figures. [52] Y. Fily, A. Baskaran, and M. F. Hagan, Soft Matter 10, 5609 (2014). [53] F. Peruani, A. Deutsch, and M. Bar, Phys. Rev. E 74, 030904 (2006). [54] F. Peruani, J. Starrub, V. Jakovljevic, L. Sogaard- Andersen, A. Deutsch, and M. Bar, Phys. Rev. Lett. 108, 098102 (2012). [55] A. Baskaran and M. C. Marchetti, Phys. Rev. Lett. 101, 268101 (2008). [56] A. Baskaran and M. C. Marchetti, Phys. Rev. E 77, 011920 (2008). [57] S. R. McCandlish, A. Baskaran, and M. F. Hagan, Soft Matter 8, 2527 (2012). [58] F. Ginelli, F. Peruani, M. Bar, and H. Chate, Phys. Rev. Lett. 104, 184502 (2010). + +<|ref|>text<|/ref|><|det|>[[515, 66, 920, 707]]<|/det|> +[59] M. Bar, R. Grossmann, S. Heidenreich, and F. Peruani, Annu. Rev. Condens. Matter Phys. 11, 441 (2020). [60] E. Bertin, H. Chate, F. Ginelli, S. Mishra, A. Peshkov, and S. Ramaswamy, New J. Phys. 15, 085032 (2013). [61] S. Ngo, A. Peshkov, I. S. Aranson, E. Bertin, F. Ginelli, and H. Chate, Phys. Rev. Lett. 113, 038302 (2014). [62] H. H. Wensink and H. Löwen, Phys. Rev. E 78, 031409 (2008). [63] C. Bechinger, R. Di Leonardo, H. Löwen, C. Reichhardt, G. Volpe, and G. Volpe, Rev. Mod. Phys. 88, 045006 (2016). [64] The ability of collisions of self-propelled particles on a surface to drive formation of smectic layers is supported by a recent observation in bacterial colonies growing on flat surfaces, in which bacteria form 'rosettes' with the rod- like bacteria oriented perpendicular to the surface [71]. However, in contrast to the self- limited caps in our system, the bacterial rosettes do not exhibit a preferred size because they are on a flat boundary. [65] M. F. Hagan and G. M. Grason, "Equilibrium mechanisms of self- limiting assembly," (2021), arXiv:2007.01927 [cond- mat.soft]. [66] N. Yu, A. Ghosh, and M. F. Hagan, Soft Matter 12, 8990 (2016). [67] J. Lidmar, L. Mirny, and D. R. Nelson, Phys. Rev. E 68, 051910 (2003). [68] G. Gompper and D. M. Kroll, J. Phys. I France 6, 1305 (1996). [69] L. M. Lemma, M. M. Norton, S. J. DeCamp, S. A. Aghvami, S. Fraden, M. F. Hagan, and Z. Dogic, arXiv:2006.15184 [cond- mat] (2020), arXiv:2006.15184 [cond- mat]. [70] P. Chandrakar, J. Berezney, B. Lemma, B. Hishamunda, A. Berry, K.- T. Wu, R. Subramanian, J. Chung, D. Needleman, J. Gelles, and Z. Dogic, "Microtubule- based active fluids with improved lifetime, temporal stability and miscibility with passive soft materials," (2018), arXiv:1811.05026 [cond- mat.soft]. [71] O. J. Meacock, A. Doostmohammadi, K. R. Foster, J. M. Yeomans, and W. M. Durham, Nat. Phys. (2020), 10.1038/s41567- 020- 01070- 6. [72] Y. Fily, A. Baskaran, and M. F. Hagan, "Active particles on curved surfaces," (2016), arXiv:1601.00324 [cond- mat.soft]. [73] M. Doi and S. F. Edwards, The Theory of Polymer Dynamics, International Series of Monographs on Physics No. 73 (Clarendon Press, Oxford, 2007). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 110, 933, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 747, 115, 766]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 789, 417, 808]]<|/det|> +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[60, 66, 938, 408]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 448, 118, 468]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[42, 490, 418, 510]]<|/det|> +Please see manuscript .pdf for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 60, 930, 550]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 578, 116, 597]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[42, 620, 417, 640]]<|/det|> +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 55, 933, 560]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 590, 117, 609]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 632, 417, 652]]<|/det|> +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[55, 60, 680, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[52, 430, 86, 454]]<|/det|> +
b)
+ +<|ref|>image<|/ref|><|det|>[[188, 444, 559, 732]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 767, 116, 785]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 809, 416, 828]]<|/det|> +Please see manuscript.pdf for full caption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[66, 60, 825, 410]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[65, 420, 105, 450]]<|/det|> +
b)
+ +<|ref|>image<|/ref|><|det|>[[230, 447, 768, 772]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 116, 819]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[44, 843, 417, 862]]<|/det|> +Please see manuscript.pdf for full caption. + +<|ref|>sub_title<|/ref|><|det|>[[44, 885, 310, 912]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 936, 765, 956]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 45, 238, 227]]<|/det|> +movie1. mp4 movie2. mp4 movie3. mp4 movie4. mp4 movie5. mp4 movie6. mp4 supplemental.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/images_list.json b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..2bb2fc5dcb63836d5eadf529a3922280065f34ca --- /dev/null +++ b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/images_list.json @@ -0,0 +1,92 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1 High-resolution physical mapping of \\(P m57\\) on chromosome 2S's. (a) Initial physical-mapping by comparing the chromosome recombination breakpoints of the resistant and susceptible recombinants positioned \\(P m57\\) within an", + "footnote": [], + "bbox": [ + [ + 163, + 490, + 882, + 875 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2 Identification of Pm57 candidate gene using ethyl methane sulfonate (EMS)-induced mutants. (a)", + "footnote": [], + "bbox": [ + [ + 163, + 99, + 870, + 610 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3 Functional validation of WTK7-vWA by transgenics. (a) Responses of 36 independent (L1 to L36) transgenic lines to \\(Bgt\\) isolate E09. The susceptible positive \\(\\mathrm{T_0}\\) transgenic plants are indicated in red. (b) \\(\\mathrm{T_1}\\) transgenic lines (L1 to L4) of WTK7-vWA show high resistance to \\(Bgt\\) isolate E09. Fielder and the negative line L5 were used as susceptible controls, respectively. Three individuals of each independent line are shown. The ‘+’ and ‘–’ signs on each leaf designate the presence or absence of WTK7-vWA gene.", + "footnote": [], + "bbox": [ + [ + 150, + 99, + 895, + 520 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4 Subcellular localization and protein structure prediction of Pm57. (a) Subcellular localization of Pm57 in wheat protoplasts. 35S::GFP or 35S::Pm57-GFP plasmids were co-transformed into wheat protoplasts with nucleus marker plasmid AtPIF4-mCherry. The GFP, mCherry and Chlorophyll fluorescence were visualized under a confocal laser-scanning microscope. 35S::GFP was served as a control. Scale bars, \\(10 \\mu \\mathrm{m}\\) . (b) Protein structure prediction of Pm57. Three-dimensional model of Pm57 is predicted by AlphaFold. Orange, kinase domain; blue, pseudokinase domain; Green, vWA domain; Yellow, putative Vwaint domain. The red spheres indicate amino acid substitutions resulting from the EMS mutagenesis.", + "footnote": [], + "bbox": [ + [ + 190, + 95, + 844, + 485 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5 Evaluation of the resistance mechanism of \\(Pm57\\) . (a) Detection of \\(\\mathrm{H}_2\\mathrm{O}_2\\) accumulation by DAB staining of \\(Bgt\\) -infected leaves. Seven-day-old plants were inoculated with \\(Bgt\\) isolate E09. Staining was performed on the \\(Bgt\\) -infected leaves at 2 d post inoculation (dpi). Brown staining shows the accumulation of \\(\\mathrm{H}_2\\mathrm{O}_2\\) . Bar, \\(200 \\mu \\mathrm{m}\\) . (b) The pathogenic symptoms and trypan blue staining of the \\(Bgt\\) -infected leaves at 7 dpi to visualize plant cell death. Scale bar, \\(200 \\mu \\mathrm{m}\\) . Representative leaves were photographed at 7 dpi. Bar, \\(10 \\mathrm{mm}\\) . (c) Expression analyses of \\(Pm57\\) and pathogenesis-related (PR) genes in \\(89(5)69\\) . The transcript levels were examined by qRT-PCR. TaActin1 was used as an endogenous control. Error bars represent mean \\(\\pm \\mathrm{SD}\\) from three biological samples. hpi, hours post inoculation. \\(^{*}\\mathrm{p} < 0.05\\) , \\(^{**} \\mathrm{p} < 0.01\\) (Student's \\(t\\) test).", + "footnote": [], + "bbox": [ + [ + 147, + 230, + 857, + 690 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Fig. 6 Statistical analysis of agronomic traits in WT and Pm57 transgenic lines. (a) Representative wheat whole plants, spikes and seeds of Fielder (-) and Pm57 transgenic plant (+). (b-h) Comparison of agronomic traits in Fielder and Pm57 transgenic plants: (b) heading date (n = 3), (c) thousand-grain weight (n = 3), (d) seeds per spike (n = 3), (e) plant height (n = 15), (f) spike length (n = 15), (g) spikelets per spike (n = 15), and (h) tiller numbers (n = 15). Data in (b-d) are displayed as bar graphs. Data are represented as the mean ± SD from three replicates. Data in (e-g) are displayed as box and whisker plots with individual data point. The error bars represent maximum and minimum values. Center line, median; box limits, 25th and 75th percentiles. ns, no statistically significant difference by two-tailed Student's \\(t\\) test.", + "footnote": [], + "bbox": [ + [ + 171, + 156, + 857, + 579 + ] + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf.mmd b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf.mmd new file mode 100644 index 0000000000000000000000000000000000000000..df75f57659404e9276eb728e5609b4ad6d8459fb --- /dev/null +++ b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf.mmd @@ -0,0 +1,377 @@ + +# Pm57 from Aegilops searsii encodes a novel tandem kinase protein conferring powdery mildew resistance in bread wheat + +Wenxuan Liu + +wx1iu2003@hotmail.com + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Yue Zhao + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Zhenjie Dong + +College of Agronomy, Nanjing Agricultural University + +Jingnan Miao + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Qianwen Liu + +State Key Laboratory of Agrobiotechnology, College of Biological Sciences, China Agricultural University + +Chao Ma + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Xiubin Tian + +Institute of Genetics and Developmental Biology, Chinese Academy of Sciences + +Jinqiu He + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Huihui Bi + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +Wen Yao + +Henan Agricultural University https://orcid.org/0000- 0002- 0643- 506X + +Tao Li + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<--- Page Split ---> + +## Harsimardeep Gill + +Department of Agronomy, Horticulture and Plant Science, South Dakota State University + +## Zhibin Zhang + +Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China + +## Aizhong Cao + +State Key Lab of Crop Genetics and Germplasm Enhancement, Cytogenetics Institute, Nanjing Agricultural University/JCIC- MCP + +## Bao Liu + +Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China https://orcid.org/0000- 0001- 5481- 1675 + +## Huanhuan Li + +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +## Sunish Sehgal + +Department of Agronomy, Horticulture and Plant Science, South Dakota State University + +## Article + +## Keywords: + +Posted Date: April 28th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2844708/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on June 5th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49257- 2. + +<--- Page Split ---> + +# Pm57 from Aegilops searsii encodes a novel tandem kinase protein + +# conferring powdery mildew resistance in bread wheat + +Yue Zhao1,5, Zhenjie Dong2,5, Jingnan Miao1, Qianwen Liu1, Chao Ma1, Xiubin Tian1, Jinqiu He1, Huihui Bi1, Wen Yao1, Tao Li1, Harsimardeep S Gill3, Zhibin Zhang4, Aizhong Cao2, Bao Liu4, Huanhuan Li1\*, Sunish K Sehgal3\* & Wenxuan Liu1\*1 State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University, Zhengzhou 450002, China2 College of Agronomy, Nanjing Agricultural University, Nanjing 210000, China3 Department of Agronomy, Horticulture and Plant Science, South Dakota State University, Brookings, SD 57007, U.S.A.4 Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China5 These authors contributed equally*e-mail: lihuanhuanhappy@henau.edu.cn; sunish.sehgal@sdtate.edu; wxliu2003@hotmail.com + +<--- Page Split ---> + +## Abstract + +Powdery mildew is a devastating disease that affects wheat yield and quality. Despite wheat wild relatives being a valuable source of resistance genes, their incorporation into wheat improvement is constrained by their adverse effects on agronomic traits, difficulty in isolating genes and poor understanding of the resistance mechanisms. Here, we report cloning of \(Pm57\) , the first gene isolated from Aegilops searsii. It encodes an unusual wheat tandem kinase (WTK) protein with putative kinase- pseudokinase domains followed by a von Willebrand factor A (vWA) domain, designated WTK7- vWA. The resistance function of \(Pm57\) was validated by independent mutants, gene silencing, and transgenic assays. Stable \(Pm57\) transgenic wheat lines showed high levels and all- stage resistance against diverse isolates of \(Bgt\) fungus with no adverse effects on agronomic traits. Our findings highlight the emerging role of kinase fusion proteins in plant disease resistance and provide a powerful gene for wheat breeding. + +## 29 Introduction + +Bread wheat (Triticum aestivum L.) is an important staple food crop that plays a vital role in global food security'. However, sustainable wheat production across the globe is threatened due to its susceptibility to various pests and diseases. Among various diseases, powdery mildew caused by Blumeria graminis f. sp. tritici (Bgt) is a major wheat disease worldwide affecting grain yield and processing quality. This disease typically leads to yield losses ranging from 10 to \(15\%\) which may reach up to \(62\%\) in severe cases?. Development and deployment of resistant wheat varieties have been regarded as the most economical, effective, and sustainable way to mitigate the losses caused by powdery mildew. + +To date, over a hundred \(Pm\) resistance genes/alleles at approximately 60 loci in wheat and its wild relatives have been documented?, however only a few \(Pm\) genes including \(Pm2\) , \(Pm4\) , \(Pm5\) , \(Pm6\) , \(Pm8\) , \(Pm21\) , and \(Pm52\) have been widely used in developing disease- resistant wheat varieties. The majority of \(Pm\) genes could not be utilized due to associated linkage drags causing adverse pleiotropism or narrow- spectrum resistance that often becomes ineffective with evolution of new \(Bgt\) races?. This necessitates the identification and cloning of new genes that offer broad- spectrum powdery mildew resistance with no adverse effects on the wheat agronomic characteristics. + +Recently, several strategies have emerged for cloning resistant genes in plants, however, the classical method of map- based cloning still remains the most effective gene cloning approach in absence of an annotated reference genome. Owing to its huge genome size, cloning a gene in wheat by map- based cloning is time- consuming and challenging?. However, advances in genomics and availability of genome sequences of more than ten hexaploid wheat cultivars have largely facilitated gene cloning in + +<--- Page Split ---> + +wheat. Since the first Pm gene Pm3b was isolated from wheat in \(2004^{7}\) , about \(16Pm\) resistance genes have been cloned. Eleven of these 16 genes i.e. Pm1a \(^{8}\) , Pm2 \(^{9}\) , Pm3/Pm8/Pm17 \(^{7,10,11}\) , Pm5e \(^{12}\) , Pm12 \(^{13}\) , Pm21 \(^{14,15}\) , Pm41 \(^{16}\) , Pm60 \(^{17}\) , and Pm69 \(^{18}\) , encode nucleotide-binding leucine-rich repeat (NLR) immune receptors. Of the remaining five Pm genes, Pm24 and WTK4 encode tandem kinases \(^{3}\) . The race- specific resistance gene Pm4 was characterized to be a putative serine/threonine kinase \(^{3}\) . Another two broad- spectrum resistance genes, Pm38 and Pm46, encode an ABC transporter and hexose transporter, respectively \(^{20,21}\) . + +Further, wild relatives of hexaploid bread wheat serve as reservoirs of genetic diversity for important agronomic traits including disease resistance genes \(^{14}\) and more than half of the currently designated Pm genes have been derived from secondary and tertiary gene pool of wheat \(^{22,23}\) . Though Pm genes from wild relatives play an important role in breeding for disease resistance, it is very difficult to fine- map and clone the alien genes from secondary or tertiary gene pool in wheat background compared to genes from common wheat due to homoeologous recombination suppression, lack of alien chromosomes specific markers and unavailability annotated reference genomes. To date only four Pm genes have been cloned from wild relatives in secondary and tertiary gene pools of wheat, among which Pm12 (Ae. speltoides) and Pm21 (Dasypyrum villosum L.) were orthologous and Pm8 and Pm17 (Secale cereale L.) were cloned by Pm3 homology- based cloning, and all four genes encode NLR proteins. This hinders the better understanding of molecular basis of these genes, thus limiting their deployment in wheat breeding. + +Powdery mildew resistance gene Pm57 \(^{24}\) was derived from Aegilops searsii Feldman & Kislev ex Hammer \((2\mathrm{n} = 2\mathrm{x} = 14\) , S'sS), an S- genome species from section Sitopsis (Jaub. & Spach) Zhuk in the secondary gene pool of wheat. Pm57 is the first Pm gene and second disease resistance gene following Sr51 that we transferred into bread wheat (Chinese Spring- Ae. searsii disomic addition line, TA3581) from Ae. searsii \(^{25}\) and mapped it to long arm of 2S\*#1 \(^{24,26}\) . In the present study, we report map- based cloning of Pm57, first gene to be isolated from Ae. searsii. Interestingly, Pm57 gene encodes a novel and unusual tandem kinase protein with putative kinase- pseudokinase domains followed by a von Willebrand factor A (vWA) domain and is designated WTK7- vWA. The stable Pm57 transgenic wheat lines showed high level and all- stage resistance against multiple Bgt isolates with no adverse effect on other agronomic traits. Our results offer significant insight into the molecular mechanisms underlying wheat powdery mildew resistance and will enable development of wheat varieties with broad- spectrum powdery mildew resistance. + +## 81 Results + +## 82 High-resolution mapping and map-based cloning of Pm57 + +<--- Page Split ---> + +\(P m57\) was previously mapped to a 5.13 Mb region flanked by markers \(X67593\) and \(X62492\) in the terminal region of chromosome 2S's long arm using wheat- Ae. searsii 2S's recombinants26 (Fig. 1a). To fine- map \(P m57\) , progenies of single heterozygous wheat- Ae. searsii 2S's recombinant were used developed a large \(P m57\) segregation population. Two flanking markers \((X67593\) and \(X62492)\) were used for screening of 3,380 individuals and a total of 104 CS- Ae. searsii 2S's recombinants were identified (Fig. 1b). Subsequently, we performed \(B g t\) response assays on \(\mathrm{F}_{2:3}\) families of the 104 recombinants by inoculating \(B g t\) isolate E09. + +Next, 16 2S's- specific markers were designed from the mapping interval of \(P m57\) flanked by markers \(X67593\) and \(X62492\) using the recently released genome sequence of Ae. searsii (TE01)1 (Table S1). Marker analysis of those 2S's recombinants grouped them into six different types (Types I to VI). Integrating \(B g t\) response of the 104 recombinants with the marker analysis, we mapped \(P m57\) to the region between markers \(X10\) and \(X13\) (Fig. 1b). The region corresponds to a physical interval of 710 kb on the long arm of Ae. searsii chromosome 2S's, which harbors 12 genes (referred to as G1- G12) based on genome sequence of Ae. searsii1. Two of these 12 genes (G4 and G5) contain putative tandem kinase domains and were considered potential candidate genes for \(P m57\) , while none of the other 10 genes were annotated to resemble any previously identified disease- resistant genes (Fig. 1c). + +![](images/Figure_1.jpg) + +
Fig. 1 High-resolution physical mapping of \(P m57\) on chromosome 2S's. (a) Initial physical-mapping by comparing the chromosome recombination breakpoints of the resistant and susceptible recombinants positioned \(P m57\) within an
+ +<--- Page Split ---> + +interval flanked by markers X67593 and X62492 on the long arm of chromosome 2Ss. (b) A high-resolution genetic map delimited Pm57 to a 710 kb region between the markers X10 and X13. Markers X67593 and X62492 were derived from RNA-seq of CS- Ae. searsii 2Ss disomic addition line, and X1- X16 were developed using the recently released genome sequence of Ae. searsii (TE01)1. (c) Schematic representation of the twelve annotated genes G1- G12 in the mapping interval of Ae. searsii reference genome (TE01). Two of these 12 genes (G4 and G5) contain putative tandem kinase domains and were considered potential candidate genes for Pm57. + +## Identification of Pm57 candidate genes by MutRNA-Seq + +To unravel Pm57 candidates, we performed EMS mutagenesis of Pm57 donor wheat- Ae. searsii recombinant line 89(5)69. About 10,000 \(\mathrm{M_0}\) seeds were treated with \(0.6\%\) EMS and developed 1,598 \(\mathrm{M_2}\) lines. Fifteen seeds from each of the randomly selected \(300\mathrm{M}_2\) families were screened for susceptible mutants using the Bgt isolate E09. Finally, 15 independent susceptible mutants were identified by Pm57- specific markers and a 55K SNP chip and further validated in the \(\mathrm{M}_3\) generation. + +Further, we performed mutant RNA- seq (MutRNA- seq) using five susceptible mutants (Mut51, Mut60, Mut141, Mut209 and Mut216) along with resistant wild type 89(5)69 (Fig. 2a). Alignment of the 12 genes from Pm57 mapping interval to the transcriptome sequences revealed that three (Mut51, Mut141, and Mut209) of the five mutants had missense mutations of gene G4, while no mutations found in genes G2, G5, G6, G7, G8, G9, G10 and G11. The expression levels of the remaining three genes (G1, G3 and G12) were too low to reliably call mutations in transcriptome sequences (Fig. 2a and Table S2). Therefore, gene G4 encoding a tandem kinase - vWA domains protein emerged as the most likely candidate of Pm57 among the 12 genes, and was designated as WTK7- vWA since six WTKs (WTK1- 6) have been previously reported27. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2 Identification of Pm57 candidate gene using ethyl methane sulfonate (EMS)-induced mutants. (a)
+ +Candidate gene identification by MutRNA- Seq. RNA- Seq reads from 89(5)69 and 5 susceptible EMS mutants (Mut51, Mut60, Mut141, Mut209 and Mut216) are mapped in Ae. searsii reference genome sequence (TE01). Only G4 of the 12 candidate genes in the mapping interval carry missense mutations (red dots) in three mutants (Mut51, Mut141, and Mut209). (b) Schematic representation of WTK7- vWA and identification of WTK7- vWA using ethyl methane sulfonate (EMS)-induced mutants. The position of three conserved domains and the predicted amino-acid changes caused by the EMS mutations were indicated. Exons and introns are represented by rectangles and lines, respectively. The two protein kinase domains and the vWA domain are shown in orange, blue, and green colors, respectively. The proportion of each isoform in the expression data of 89(5)69 at 0 and 24 h post inoculation (hpi) with Bgt isolate E09 is presented on the right side. IF1 is the main variant with 14 exons encoding for a complete protein. IF2- 7 transcripts were derived from the mis- splicing of exon or intron and generated premature stop codon (marked in red star). + +## Validation of WTK7- vWA by EMS mutants, gene silencing, and transgenic assays + +To further validate the candidacy of WTK7- vWA for Pm57, we cloned the genomic DNA and cDNA + +<--- Page Split ---> + +sequences of WTK7- vWA gene (G4) and another WTK gene (G5) from 89(5)69 as well as the 15 susceptible mutants for sequence comparison. In the resistant line 89(5)69, the WTK7- vWA gene was 9,473 bp and contained 14 exons with a coding sequence of 3,489 bp (Fig. 2b). Intriguingly, WTK7- vWA had at least seven alternative splicing variants, designated IF1 - IF7, of which, IF1 was the main isoform, with a proportion of \(50.6\%\) (39 out of 77 tested WTK7- vWA cDNA clones) at \(0\mathrm{h}\) post- inoculation (hpi) with \(Bgt\) isolate E09 increasing to \(80.9\%\) at \(24\mathrm{hpi}\) , and isoforms IF2 - IF7 were much less abundant ( \(1.3\%\) - \(27.3\%\) ). IF1 encodes a full- length intact WTK7- vWA protein with Kin I, Kin II and vWA domains, while IF2 and IF3 encode proteins with truncated Kin I and Kin II domains, and IF4- IF7 encode proteins with only truncated Kin I domain (Fig. 2b). Gene sequence comparison revealed that ten of the 15 susceptible mutants had SNPs in WTK7- vWA that resulted in amino acid substitutions, premature stop codons, or relocation of the intron/exon splice sites (Fig. 2b). Specifically, a frameshift mutation was detected in Mut216 with a G/A point mutation in the splice acceptor site of intron 9. Mut351 had a nonsense mutation that gave rise to a premature stop codon at the amino acid position of 1,081. The other eight mutants (G78D in Mut223, G177E in Mut141, G193R in Mut51, D209N in Mut210, G424D in Mut92, P747L in Mut121, R829W in Mut22, and G903D in Mut209) harbored missense mutations that occurred in the kinase I (Kin I), kinase II (Kin II) or vWA domains (Fig. 2b). In addition, no sequence variations in gene G5 (WTK) were found among all susceptible mutants. + +Next, virus- induced gene silencing (VIGS) was performed for validation of candidate genes G4 (WTK7- vWA) and G5 (WTK). The resistant introgression line 89(5)69 inoculated with G4- VIGS constructs lost resistance to powdery mildew, whereas the plants inoculated with G5- VIGS constructs and empty vector control constructs remained resistant (Fig. S1). These results were consistent with the conclusions derived from MutRNA- Seq analyses and suggested G4 (WTK7- vWA) was the prime candidate for Pm57. + +Transgenic complementation of susceptible cv. Fielder was subsequently performed to confirm the role of WTK7- vWA in conferring resistance against powdery mildew. A total of 36 \(\mathrm{T_0}\) plants (L1 to L36) were generated, and 33 were identified as positive transgenic plants (Fig. 3a). With the exception of L20 and L29, all the other positive \(\mathrm{T_0}\) transgenic plants (+) were highly resistant to the Bgt isolate E09, whereas all the negative plants (- , L5, L23 and L27) were as susceptible as the WT Fielder (Fig. 3a). This difference in powdery mildew disease resistance was also observed for \(\mathrm{T_1}\) transgenic lines and WT (Table S3), and four representative lines (L1, L2, L4 and L5) were shown in Fig. 3b. In addition, qRT- PCR analyses showed that there was no expression of WTK7- vWA gene in the positive transgenic lines L20 and L29, explaining why these two positive lines behaved susceptible to powdery mildew (Fig. S2). Taken together, the genetic mapping, mutant analysis, gene silencing, and transgenic assays confirmed that WTK7- vWA is Pm57. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3 Functional validation of WTK7-vWA by transgenics. (a) Responses of 36 independent (L1 to L36) transgenic lines to \(Bgt\) isolate E09. The susceptible positive \(\mathrm{T_0}\) transgenic plants are indicated in red. (b) \(\mathrm{T_1}\) transgenic lines (L1 to L4) of WTK7-vWA show high resistance to \(Bgt\) isolate E09. Fielder and the negative line L5 were used as susceptible controls, respectively. Three individuals of each independent line are shown. The ‘+’ and ‘–’ signs on each leaf designate the presence or absence of WTK7-vWA gene.
+ +## Subcellular localization and structural analysis of Pm57 protein + +Subcellular localization of Pm57 was investigated through transient transformation in wheat leaf protoplasts. We constructed the fusion protein of Pm57 with green florescent protein (GFP), and the resulting Pm57- GFP construct as well as the GFP control were introduced to wheat protoplasts, respectively. As shown in Fig. 4a, green fluorescence was ubiquitously detected in the cells transformed with the GFP control. Similarly, fluorescence of the Pm57- GFP fusion protein was observed in both nucleus and cytosol (Fig. 4a). To explore the structure of Pm57 in detail, the AlphaFold \(^{28}\) was used to generate a 3D model of Pm57. The tertiary structure revealed that Pm57 possesses a modular structure, with kinase- pseudokinase domains, a vWA domain, and a putative Vwaint domain as described by Wang et al. \(^{27}\) and two kinase domains in Pm57 were highly symmetrical (Fig. 4b). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4 Subcellular localization and protein structure prediction of Pm57. (a) Subcellular localization of Pm57 in wheat protoplasts. 35S::GFP or 35S::Pm57-GFP plasmids were co-transformed into wheat protoplasts with nucleus marker plasmid AtPIF4-mCherry. The GFP, mCherry and Chlorophyll fluorescence were visualized under a confocal laser-scanning microscope. 35S::GFP was served as a control. Scale bars, \(10 \mu \mathrm{m}\) . (b) Protein structure prediction of Pm57. Three-dimensional model of Pm57 is predicted by AlphaFold. Orange, kinase domain; blue, pseudokinase domain; Green, vWA domain; Yellow, putative Vwaint domain. The red spheres indicate amino acid substitutions resulting from the EMS mutagenesis.
+ +## Evaluation of the resistance mechanism of Pm57 + +In order to evaluate the potential mechanism involved in Pm57- mediated resistance, the process of Bgt development was observed in the susceptible CS, resistant 89(5)69 and the transgenic lines. It was found that haustorium formation was inhibited by Pm57 in the early infection stage, and no secondary hypha branches or conidiophore could be observed on the epidermal cells (Fig. 5a). Bgt- infected leaf staining by diaminobenzidine (DAB) and trypan blue (TPN) further displayed robust accumulation of \(\mathrm{H}_2\mathrm{O}_2\) and cell death in the Bgt- interacting host cells of resistant 89(5)69 and the transgenic lines (Fig. 5a and 5b). These results suggest Pm57 inhibits haustorium formation of Bgt, and this process was correlated with \(\mathrm{H}_2\mathrm{O}_2\) accumulation and induced cell death. + +The qRT- PCR analysis using seedling leaves of 89(5)69 demonstrated that Pm57 was highly expressed + +<--- Page Split ---> + +in wheat leaf tissues and up-regulated by \(Bgt\) infection (Fig. S3, Fig. 5c). In addition, pathogenesis-related (PR) genes, including \(PR1\) , \(PR2\) , \(PR3\) , \(PR4\) , and \(PR9\) , were also up-regulated upon \(Bgt\) inoculation following \(Pm57\) up-regulation (Fig. 5c). These expression patterns of \(Pm57\) and \(PR\) genes imply that \(Pm57\) would be an upstream regulator of \(PR\) genes in the pathogen defense response pathway. + +![](images/Figure_5.jpg) + +
Fig. 5 Evaluation of the resistance mechanism of \(Pm57\) . (a) Detection of \(\mathrm{H}_2\mathrm{O}_2\) accumulation by DAB staining of \(Bgt\) -infected leaves. Seven-day-old plants were inoculated with \(Bgt\) isolate E09. Staining was performed on the \(Bgt\) -infected leaves at 2 d post inoculation (dpi). Brown staining shows the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) . Bar, \(200 \mu \mathrm{m}\) . (b) The pathogenic symptoms and trypan blue staining of the \(Bgt\) -infected leaves at 7 dpi to visualize plant cell death. Scale bar, \(200 \mu \mathrm{m}\) . Representative leaves were photographed at 7 dpi. Bar, \(10 \mathrm{mm}\) . (c) Expression analyses of \(Pm57\) and pathogenesis-related (PR) genes in \(89(5)69\) . The transcript levels were examined by qRT-PCR. TaActin1 was used as an endogenous control. Error bars represent mean \(\pm \mathrm{SD}\) from three biological samples. hpi, hours post inoculation. \(^{*}\mathrm{p} < 0.05\) , \(^{**} \mathrm{p} < 0.01\) (Student's \(t\) test).
+ +## Evolutionary analysis of \(Pm57\) locus based on the collinearity comparison + +<--- Page Split ---> + +A comparative analysis of published wheat genome sequences showed synteny is conserved to large extent in tribe Triticaeae for \(Pm57\) mapping genomic regions (Fig. S4), however, \(Pm57\) orthologs were present only in Aegilops bicornis (TB01) (SbSb), Ae. longissima (TL05) (SlSl), Ae. speltoides (TS01) (SS), and B sub genome of T. dicoccoides (WEWSeq v1) (AABB), and the encoding proteins of these orthologous genes had \(81.8 - 87.8\%\) sequence similarities with \(Pm57\) (Fig. S4 and Fig. S5). Notably, \(Pm57\) is absent in the wheat reference genome sequence of cv. Chinese Spring (CS), but highly similar sequences of \(Pm57\) were present in the syntenic region of several bread wheat cultivars (Table S4). These highly similar sequences were manually annotated and shared more than \(80\%\) amino acid sequence identity to \(Pm57\) . + +To test whether \(Pm57\) exists in other wheat germplasms, a pair of gene- specific primers were designed to amplify a \(530\mathrm{bp}\) genomic sequence of the \(Pm57\) gene from 71 wheat accessions (including T. urartu, T. boeoticum, Ae. tauschii, T. monococcum, T. dicoccoides, T. durum, T. aestivum ssp. yunnanese, T. aestivum ssp. macha, T. aestivum ssp. spelta, T. aestivum ssp. tibetanum accessions and Chinese common wheat landraces and modern cultivars). The amplified fragment of Ae. searsii (TE01) was identical to that of the CS- Ae. searsii chromosome \(2\mathrm{S}^{\mathrm{s}}\) introgression line 89(5)69, but no fragment was obtained in the remaining 69 accessions (Fig. S6). These results indicate that \(Pm57\) probably exists only in Ae. searsii and has not been used in modern wheat breeding programs. The \(Pm57\) gene- specific primer pairs can be used as a diagnostic marker for effective detection of \(Pm57\) in molecular marker- assisted selection (MAS) breeding. + +## Evaluation of \(Pm57\) application in wheat breeding + +Our previous studies showed that \(Pm57\) conferred high resistance to mixed Bgt isolates collected in Henan Province26. To further study the spectrum resistance offered by \(Pm57\) against genetically divergent Bgt isolates, we collected 29 Bgt isolates from major wheat growing regions of China, and used them for Bgt resistance assays of both wheat- Ae. searsii introgression lines 89(5)69 and homozygous transgenic line L1. As a result, plants of both 89(5)69 and transgenic line L1 showed high resistance, with Infection type 0- 1, to all of the 29 divergent Bgt isolates (Table S5). In addition, we found 89(5)69 and positive transgenic plants showed high resistance from the seedling stage to adult stage (Fig. S7). + +To determine the practical value of \(Pm57\) in wheat breeding, \(Pm57\) transgenic wheat lines were evaluated agronomic traits. No significant differences were observed for heading date, plant height, tiller number, thousand grain weight and spike morphology between three transgenic lines and control lines, including both negative lines and WT (Fig. 6), showing no obvious adverse effect of \(Pm57\) on important agronomic traits. In addition, all of the \(\mathrm{F_1}\) plants derived from crosses of CS- Ae. searsii recombinants having small segments harboring \(Pm57\) with 22 wheat varieties were high resistance to + +<--- Page Split ---> + +powdery mildew, indicating that Pm57 could play resistance role in diverse genetic backgrounds (Fig. S8). + +![](images/Figure_6.jpg) + +
Fig. 6 Statistical analysis of agronomic traits in WT and Pm57 transgenic lines. (a) Representative wheat whole plants, spikes and seeds of Fielder (-) and Pm57 transgenic plant (+). (b-h) Comparison of agronomic traits in Fielder and Pm57 transgenic plants: (b) heading date (n = 3), (c) thousand-grain weight (n = 3), (d) seeds per spike (n = 3), (e) plant height (n = 15), (f) spike length (n = 15), (g) spikelets per spike (n = 15), and (h) tiller numbers (n = 15). Data in (b-d) are displayed as bar graphs. Data are represented as the mean ± SD from three replicates. Data in (e-g) are displayed as box and whisker plots with individual data point. The error bars represent maximum and minimum values. Center line, median; box limits, 25th and 75th percentiles. ns, no statistically significant difference by two-tailed Student's \(t\) test.
+ +## Discussion + +Development of wheat- alien recombinants is not merely a tool to clone alien genes, but also is one of the best approaches to transfer favorable genes from wild relatives to increase the genetic diversity of cultivated wheat. However, homoeologous recombination between wheat and alien chromosomes is suppressed by the pairing homoeologous (Ph) genes hampering the development of recombinants with smaller alien segments which is essential for mapping and deployment of exotic wheat genes in bread + +<--- Page Split ---> + +wheat cultivars. In this study, we used a Ph1 locus deletion mutant (TA3809, CS- ph1b) to induce homoeologous recombination29 between wheat and Ae. searsii to fine map and clone a broad- spectrum powdery mildew resistance gene Pm57 which is the first gene to be isolated from Ae. searsii. In absence of reference genome sequence of Ae. searsii1, we performed transcriptome sequence (RNA- Seq) analysis of CS- Ae. searsii 2Ss disomic addition line to develop chromosome 2Ss specific markers to physically map Pm57 into a 5.13 Mb interval on the long arm of chromosome2Ss26. Recent release of Ae. searsii reference genome sequence1 further accelerated the development of specific molecular markers for high- resolution mapping of Pm57. Combined with phenotypic analyses, Pm57 was fine- mapped to a genomic region of 710 kb. Meanwhile, we identified various translocation lines with small alien segments carrying Pm57 (Type IV, fragment size between 5.57 Mb and 9.12 Mb, \(\sim 1\%\) of 2Ss chromosome) which will greatly facilitate the deployment of Pm57 in elite wheat varieties through markers- assisted selection (Fig. 1b). + +Pm57 candidate regions harbored 12 genes in Ae. searsii reference genome assembly (TE01)1. We further performed MutRNA- Seq on 89(5)69 (wild type) and five powdery mildew susceptible mutants to narrow down Pm57 candidate genes in target mapping interval and identified WTK7- vWA as the most likely candidate for further transgenic validation. Unexpectedly, the CDS sequence of WTK7- vWA was incompletely annotated as Asea[EVM0016946 in Ae. searsii reference genome, which encodes a protein with a single kinase domain followed by a vWA domain. We obtained the full- length sequence of WTK7- vWA in Pm57- direct donor RNA- seq data26 and aligned the MutRNA- Seq from the susceptible mutants against the WTK7- vWA- corresponding unigenes. Three of five loss- of- function mutants had SNPs in WTK7- vWA- corresponding unigenes leading to amino acid changes in the kinase and vWA domains (Fig. 2). Recently, two other genes have been cloned Sr6230 and Lr9/Lr5827 by exploiting similar MutRNA- Seq approach. Transgenic complementation of susceptible cv. Fielder confirmed WTK7- vWA was Pm57. + +Recently, tandem kinase proteins (TKP) have emerged as a new prominent player involved in disease resistance in Triticeae31. To date eight TKPs, including Rpg132, Yr15 (WTK1)33, Sr60 (WTK2)34, Sr62 (WTK5)30, Pm24 (WTK3)3 and WTK435, Lr9 (WTK6)27, and Rwt436, have been identified and confer resistance against various fungal pathogens. Pm57 (WTK7- vWA) from Ae. searsii is a new member of the TKP family conferring wheat powdery mildew resistance. Based on the sequence conservation of the key amino acid residues in the two kinase domains3, Pm57 was classified as a tandem kinase- pseudokinase protein, similar to Yr15, Pm24, Lr9, and Sr62 (Fig. S9). However, Pm57 contains tandem kinase domains and a Willebrand factor A (vWA) domain, the second such unusual WTK- vWA structure to be identified following Lr9 which provides resistance to leaf rust whereas Pm57 confers resistance to powdery mildew. The vWA/Vwaint domains are presents in bacteria, archaea and Eukaryota organisms are considered to participate in protein- protein interactions37,38. For example, the vWA domain of human copine are capable of interacting with a wide variety of signaling molecules + +<--- Page Split ---> + +with coiled- coil domains39. In plants, the vWA- containing copine proteins were shown to be regulators of basal and \(R\) - mediated disease resistance, suggesting that vWA- containing proteins may play an important role in plant disease resistance40. In this study, eight susceptible mutants harbored missense mutations in Kin I, Kin II or vWA domains (Fig. 2), indicating that each domain of Pm57 is essential for resistance to the \(Bgt\) pathogen. A similar phenomenon was also observed for \(Rpg1\) , \(Pm24\) , \(Sr62\) and \(Lr9\) genes3,27,30,32. No sequence variation of \(Pm57\) was detected in another five susceptible mutants, indicating possible mutations in genes or elements involved in the \(Pm57\) regulation pathway. + +To explore the structure of Pm57 in more detail, we generated its 3D model using AlphaFold28. We observed that kinase domains in Pm57 were highly symmetrical like WTK4 and Lr9 but not similar with the other five reported TPKs (Fig. 4b and Fig. S10), indicating that kinase domains of Pm57 arose from gene duplication41. Although Pm57 had the same kind of domains and a high similarity of \(88.3\%\) in amino acid sequences with Lr9, the amino acid sequences in the pseudokinase and vWA domains and protein structures are clearly different in Pm57 (Fig. S10 and Fig. S11). Wang et al. hypothesized that the pseudokinase and vWA domains of Lr9 might serve as integrated decoys for the detection of pathogen effectors27. It is reasonable to expect the differences of pseudokinase and vWA domains from Pm57 and Lr9 may lead to differently detect the effectors of \(Bgt\) and \(Pt\) pathogens, and thus confer resistance to powdery mildew and leaf rust, respectively. + +Comparative analysis Pm57 orthologs among plant kingdoms revealed that its orthologs are only present in \(Ae\) . bicornis (TB01.2S01G0903000.1), \(Ae\) . longissima (TL05.2S01G0920200.1), \(Ae\) . speltoides (TS01.2B01G0896900.1), \(T\) . dicoccoides (TRIDC2BG085800.1) and several sequenced bread wheat accessions (Fig. S4 and Table S4), suggesting a likely recent origin of \(Pm57\) after the divergence of Triticaea species, and the reticulate evolutionary nature of wheat42,43. However, several genes encoding proteins with a single kinase domain followed by a vWA domain were detected not only in Triticaea species but also in other species (Table S4). Moreover, the sequences of single kinase- vWA proteins from \(Ae\) . bicornis. TB01. Un01G0378900.1 and Thint.05G0470400.1. p are very similar with Pm57 ( \(>90\%\) , Fig. S12). These results suggest that tandem kinase- vWA proteins are probably derived from single kinase- vWA proteins. We further found that the Kin I and Kin II domains of Pm57 are located in the same Clade of a phylogenetic tree generated using each kinase domain of tandem kinase proteins (Fig. S13), indicating that the tandem kinase of Pm57 could have resulted from a duplication event31. The relatively high similarity (58.04%) of amino acid sequences between the Kin I and Kin II of Pm57 confirms a relatively recent duplication event of the two- kinase domains, being consistent with the comparative analysis of Pm57 orthologs. + +In summary, we cloned \(Ae\) . searsii- derived \(Pm57\) that confers broad- spectrum and all- stage resistance against \(Bgt\) . Pm57 encodes an unusual protein consisting of tandem kinase domains and a vWA domain (WTK7- vWA), the first of this kind of structure conferring powdery mildew among cloned plant + +<--- Page Split ---> + +resistance genes. The isolation of \(Pm57\) lays a solid foundation for further understanding of the molecular mechanism behind WTK- vWA- mediated resistance to various plant diseases. The new introgression lines carrying \(Pm57\) in a small alien segment and with diagnostic markers will facilitate development of elite wheat varieties with durable powdery mildew resistance. + +## Materials and methods + +## Plant materials + +Common wheat (T. aestivum L.) cultivar Chinese Spring (CS, TA3808), CS \(ph1b\) mutant stock (TA3809), and heterozygous CS- Ae. searsii chromosome \(2\mathrm{S}^{\mathrm{s}}\) introgression line 89(5)69 (T2BS- 2BL- 2S#1L, TA5109) and 89(6)88 (T2S#1S- 2S#1L- 2AL), both carrying \(Pm57\) , were used to develop the wheat- Ae. searsii recombinant population \(^{26}\) . 89(5)69 and the susceptible control CS were used for \(Pm57\) gene cloning and expression analyses. The wheat cultivar Fielder was used for wheat protoplast preparation and transformation. A total of 71 wheat accessions including diploid, tetraploid and hexaploid wheat accessions were employed to check the presence of the \(Pm57\) gene (Table S6). All materials were grown in a greenhouse that was maintained at \(18 - 24^{\circ}\mathrm{C}\) with \(16\mathrm{~h~}\) light/8 h dark and approximately \(70\%\) relative humidity. + +## Phenotypic response to Bgt (Powdery mildew) + +Bgt isolate E09 and other 28 genetically divergent isolates (Table S5) collected from different regions of China were used for powdery mildew evaluation. Wheat seedlings were inoculated with Bgt isolates as previously described \(^{24}\) . Disease symptoms were recorded 7 days post- inoculation (dpi) using a scale from infection type 0 to 4 (IT 0 for no visible symptoms, IT 0; for hypersensitive necrotic flecks, IT 1- 4 for highly resistant, moderately resistant, moderately susceptible and highly susceptible) \(^{44}\) . Based on IT scores, tested plants were classified into two groups, resistant (R, IT 0- 2) and susceptible (S, IT 3- 4). CS was used as a susceptible control and for propagating Bgt isolates. The primary transgenic plants (T₀) were tested for powdery mildew resistance using detached leaves. Briefly, detached leaves from plants were placed on phytagar media (0.5% phytagar; 30 ppm Benzimidazole), inoculated with Bgt isolate E09 and cultured in a greenhouse. ITs were recorded 7 days after inoculation. + +## Physical mapping of Pm57 + +The wheat- Ae. searsii \(2\mathrm{S}^{\mathrm{s}}\) recombinants population were developed from the cross of CS- \(ph1b\) mutant stock TA3809 and 89(6)88 as previously described \(^{26}\) . Heterozygous resistant wheat- Ae. searsii \(2\mathrm{S}^{\mathrm{s}}\) recombinants were self- pollinated to produce a secondary mapping population for further mapping of \(Pm57\) . The Ae. searsii genome sequence was used for the development of STS- PCR markers \(^{1}\) . All markers (Table S1) were designed using DNAMAN 7 software (Lynnon Biosoft, San Ramon, CA, + +<--- Page Split ---> + +USA). DNA was extracted using the CTAB method, and genotyping of the recombinants was performed in \(15 \mu \mathrm{L}\) volumes with parameters as previously described24. + +## Mutant development and MutRNA-Seq + +Seeds of 89(5)69 were soaked in distilled water for \(6 \mathrm{~h}\) and treated for \(16 \mathrm{~h}\) with a \(0.6\%\) (v/v) EMS solution with shaking at \(150 \mathrm{rpm}\) at room temperature. The solution was then removed, and the treated seeds were rinsed with running water for \(6 \mathrm{~h}\) . The mutagenized \(\mathrm{M}_{0}\) seeds were planted in the field and the \(\mathrm{M}_{2}\) seeds from each \(\mathrm{M}_{1}\) plant were harvested at maturity. The \(\mathrm{M}_{2}\) seedlings (10- 15 plants for each family) were phenotyped with \(Bgt\) isolate E09 in a greenhouse. Susceptible \(\mathrm{M}_{2}\) plants were advanced to \(\mathrm{M}_{3}\) generation, and \(\mathrm{M}_{3}\) seedlings were tested to confirm their susceptibility. In order to eliminate the susceptibility caused by seed contamination or missing \(2 \mathrm{~S}^{3}\) , a subset of mutants was verified using \(2 \mathrm{~S}^{3}\) - specific molecular markers \(X67593\) and \(X62492\) flanking \(\mathrm{Pm}57\) . + +MutRNA- Seq was performed as described by Yu et al.30. Five susceptible mutants derived from independent \(\mathrm{M}_{2}\) families were selected for RNA- seq. Total RNA from the \(Bgt\) - inoculated seedlings of the susceptible mutants and the wild- type line 89(5)69 was extracted using the TRIzol reagent (TransGen, Beijing, China). RNA- seq was performed as a service at Annoroad Gene Technology Co., Ltd (Beijing, China). Illumina HiSeq X Ten platform (Illumina, USA) was used to generate 150 bp pair- ended reads. Clean reads from the five mutants and 89(5)69 were mapped to the CDS of the twelve genes in the \(\mathrm{Pm}57\) mapping interval using BWA (version 0.7.17) and SAMtools (version 1.9) pipeline. Gene expression levels were quantified using the featureCounts tool in subread software (version 1.4.4). Picard- tools (version 2.27.1) and GATK4 were used for single nucleotide polymorphism (SNP) calling. + +## Virus-induced gene silencing + +Virus- induced gene silencing (VIGS) was performed as previously described45. To develop specific VIGS targets, we first blasted the sequences of \(G4\) and \(G5\) against CS and \(Ae\) . \(searsii\) genome sequence. \(G4\) and \(G5\) fragments of 200- 250 bp with very low similarities with other genes were selected as targets; they were separately cloned into BSMV- \(\gamma\) (BSMV, barley stripe mosaic virus) vector, resulting in constructs \(\gamma\) - G4 and \(\gamma\) - G5. Equimolar amounts of in vitro transcripts of BSMV- \(\alpha\) , BSMV- \(\beta\) and \(\gamma\) - G4 or \(\gamma\) - G5 were mixed to inoculate the full- expanded second leaves of 89(5)69 seedlings, and the leaves infected with BSMV- TaPDS and BSMV- \(\gamma\) (empty vector) were used as controls as previously described45. About 14 days after virus infection, the \(3^{\mathrm{rd}}\) and \(4^{\mathrm{th}}\) leaves were detached and placed on \(1\%\) agar plates supplemented with \(20 \mathrm{mg / mL}\) 6- phenyl- adenine (6- BA), and then inoculated with \(Bgt\) . Seven days later, the phenotype of powdery mildew resistance was evaluated. + +## Gene cloning and sequence analysis + +<--- Page Split ---> + +The full- length genomic DNA (gDNA) sequences and cDNAs of WTK7- vWA from 89(5)69 and each of the 15 susceptible mutants were amplified using the primers listed in Table S1. PCRs were performed in \(30~\mu \mathrm{L}\) volumes using high- fidelity Primestar polymerase (TaKaRa, Dalian, China). The PCR conditions were as previously described46. The PCR products were sequenced by Sanger dideoxy DNA sequencing method. The sequences of WTK7- vWA in each mutant and in the wild- type 89(5)69 were compared using DNAMAN 7 software. + +## Gene expression analysis + +TRIzol reagent (TransGen, Beijing, China) was used for RNA extraction and \(2\mu \mathrm{g}\) of total RNA was used for cDNA synthesis using HiScript II 1st Strand cDNA Synthesis Kit (+gDNA wiper) (Vazyme, Nanjing, China). Quantitative RT- PCR (qRT- PCR) analysis was carried out using SYBR Mix (TaKaRa, Dalian, China) on a CFX96 real- time PCR detection System (Bio- Rad, Hercules, CA, USA), with three biological replicates for each sample or treatment. The conditions for qRT- PCR were the same as previously described47. The transcript levels were calculated using the comparative CT method48. The primers used for qRT- PCR are listed in Table S1. + +## Wheat transformation + +The full- length coding sequence of WTK7- vWA was inserted into pWMB110 vectors using the restriction enzymes BamH I under the control of the maize ubiquitous (Ubi) promoter. Wheat transformation was performed using the Agrobacterium- mediated method with strain EHA105 and calluses induced from cv. Fielder immature embryos49. To determine positive transgenic events, DNA was extracted from leaves of independent \(\mathrm{T_0}\) plants, and specific PCR primers were designed to amplify a 344 bp fragment of the WTK7- vWA gene. qRT- PCR analysis was performed to evaluate the expression levels of WTK7- vWA in the leaves of transgenic wheat plants in the \(\mathrm{T_0}\) generation. The disease responses of the transgenic plants to powdery mildew were tested as described above. + +## Agronomic evaluation of Pm57 transgenic lines + +Three \(\mathrm{T_2}\) transgenic lines of Pm57 and control lines (including both negative lines and non- transformed cv. Fielder) were planted in the experimental fields of Henan Agricultural University (Zhengzhou, China) using a randomized block design with three replications. Each plot consisted of two 1.5- m rows spaced 25 cm apart, and 20 seeds were sown in each row. Regular field management, including irrigation and fertilization, were applied. In each plot, 5 plants were chosen to measure various agronomic traits, including heading date, plant height, tiller numbers, spike length, spikelets per spike, seeds per spike, and thousand- grain weight. The significance of differences among means of agronomic traits was determined using Student's \(t\) - test. + +<--- Page Split ---> + +## Subcellular localization analysis + +To determine the subcellular location of \(Pm57\) , the coding sequence of \(Pm57\) was cloned into pJIT163- GFP vector, in which the expression of \(Pm57 - GFP\) was driven by the \(CaMV35S\) promoter. An empty vector, pJIT163- GFP, was used as the negative control. Under an induction of \(40\%\) PEG- 4000, control or recombinant plasmids were co- transformed into wheat protoplasts with nucleus marker plasmid AtPIF4- mCherry. The transformed protoplasts were cultured at \(25^{\circ}\mathrm{C}\) for \(16\mathrm{h}\) under dark conditions, and observed using a laser confocal microscope (A1F, Nikon, Tokyo, Japan). + +## Pm57 protein 3D modeling prediction + +To predict the 3D structure of Pm57 and other WTKs, we used the open- source code of AlphaFold v2.128. We input the amino acid sequence of each WTKs into AlphaFold v2.1, and obtained five unrelaxed, five relaxed and five ranked models in .pdb format. Among the output models, the ranked_1. pdb model had the highest confidence with the best Local Distance Difference Test (IDDT) score were utilized. The structural graphics and the positions of amino acid substitutions were visualized using PyMOL (v.2.6.0). + +## Detection of \(\mathrm{H}_2\mathrm{O}_2\) accumulation and plant cell death + +To detect the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) , the first leaves cut from 89(5)69, transgenic line L1 and CS at 2 d post inoculation (dpi) were immediately incubated in a 3, \(3^{\prime}\) - diaminobenzidine (DAB) solution (1 mg/mL, pH 5.8) for \(12\mathrm{h}\) , and then bleached in absolute ethanol. Before assessing the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) , the bleached leaves were incubated in a \(0.6\%\) (w/v) Coomassie blue solution for \(10\mathrm{s}\) and then washed with water. To detect plant cell death, the primary leaves from the 89(5)69, L1 and CS at 7 dpi were incubated in a \(0.4\%\) Trypan blue solution for \(1\mathrm{min}\) in boiling water, washed with sterile water, bleached for \(16\mathrm{h}\) in chloral hydrate solution (2.5 g/mL), fixed for \(20\mathrm{min}\) in ethanol- acetic acid 3:1 (v/v), and stained in a \(0.6\%\) (w/v) Coomassie blue solution for \(30\mathrm{s}\) . The treated leaves were viewed under a microscope (Olympus BX53). + +## Collinearity analysis, homology searching and phylogenic analysis + +Collinearity analysis among different species or subgenomes was performed using the online tool Triticeae- GeneTribe with default parameters50. The WheatOmics 1.0 (http://202.194.139.32/) and Phytozome v13 database (https://phytozome- next.jgi.doe.gov/) were used to find proteins similar to Pm57 in plant genomes51,52. Pm57 were used as queries for BLAST analysis and the retrieved proteins with kinase domain and vWA domain were selected. For kinase domain analysis, the 182 putative kinase or pseudokinase domains used for phylogenetic analysis of WTK3 ( \(Pm24^{3}\) ) were also used in this study. In addition, the proteins homologous to Pm57 with a tandem kinase - vWA structure were + +<--- Page Split ---> + +included in the phylogenetic analysis of kinase domains. Multiple sequence alignments were carried out with ClustalW software with default settings. The conserved motifs in the Kin I and Kin II domains were annotated as described previously3. Phylogenetic analysis was conducted with Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/) and drawn with iTOL (https://itol.embl.de/). + +## Data availability + +All data generated or analyzed in this study are included in this article and Supplementary Information files as well as the public databases. The WTK7- vWA sequence have been deposited in NCBI Genbank under accession number OQ675542. The MutRNA-Seq data derived from the WT and five mutant plants have been deposited in NCBI's Sequence Read Archive (SRA) under accession number PRJNA947672. The other data that support the findings of this study are available from the corresponding author upon request. + +## Acknowledgements + +We are grateful to Prof. Zhongfu Ni and Huiru Peng from China Agricultural University, Beijing, China, for their advice and supports during this research. We are also grateful to Prof. Pengtao Ma of Yantai University, Yantai, Shandong, China, for providing Bgt isolates and powdery mildew resistance assays. This research was financially supported by the National Natural Science Foundation of China (31801363, 31971887 and 32272070), the Scientific and Technological Research Project of Henan Province of China (222103810004) and the Key Scientific Research Projects of Higher Education Institutions in Henan Province (23A210020) and South Dakota Wheat Commission (3x2030). + +## Author contributions + +W.L., S.S. and H.L. designed the study. Y.Z., Z.D., J.M., C.M., X.T., and J.H. performed the research. Q.L., H.B., W.Y., T.L., A.C., H.L. H.G. and S.S. analyzed the data. W.L. and Y.Z. wrote the manuscript and S.S. contributed to revising the draft. All authors have read and approved the final manuscript. + +## Competing interests + +The authors declare no competing interests. + +## Additional information + +Supplementary information is available for this paper. + +## References: + +1. Li, L. et al. Genome sequences of five Sitopsis species of Aegilops and the origin of polyploid wheat B subgenome. + +<--- Page Split ---> + +Molecular Plant 15, 488- 503 (2022).2. Wu, L. et al. Genetic dissection of the powdery mildew resistance in wheat breeding line LS5082 using BSR-Seq. The Crop Journal 10, 1120- 1130 (2022).3. Lu, P. et al. A rare gain of function mutation in a wheat tandem kinase confers resistance to powdery mildew. Nature Communications 11, 680 (2020).4. Wang, W. et al. Characterization of the powdery mildew resistance gene in wheat breeding line KN0816 and its evaluation in marker- assisted selection. Plant Dis 105, 4042- 4050 (2021).5. Li, H. et al. A spontaneous wheat- Aegilops longissima translocation carrying Pm66 confers resistance to powdery mildew. Theoretical and Applied Genetics 133, 1149- 1159 (2020).6. Wang, Y. et al. Mapping stripe rust resistance gene YrZH22 in Chinese wheat cultivar Zhoumai 22 by bulked segregant RNA- Seq (BSR- Seq) and comparative genomics analyses. Theor Appl Genet 130, 2191- 2201 (2017).7. Yahiaoui, N., Srichumpa, P., Dudler, R. & Keller, B. Genome analysis at different ploidy levels allows cloning of the powdery mildew resistance gene Pm3b from hexaploid wheat. Plant J 37, 528- 38 (2004).8. Hewitt, T. et al. A highly differentiated region of wheat chromosome 7AL encodes a Pm1a immune receptor that recognizes its corresponding AvrPm1a effector from Blumeria graminis. New Phytologist 229, 2812- 2826 (2021).9. Sánchez- Martín, J. et al. Rapid gene isolation in barley and wheat by mutant chromosome sequencing. Genome Biology 17, 221 (2016).10. Singh, S.P. et al. Evolutionary divergence of the rye Pm17 and Pm8 resistance genes reveals ancient diversity. Plant Molecular Biology 98, 249- 260 (2018).11. Hurni, S. et al. Rye Pm8 and wheat Pm3 are orthologous genes and show evolutionary conservation of resistance function against powdery mildew. Plant J 76, 957- 969 (2013).12. Xie, J. et al. A rare single nucleotide variant in Pm5e confers powdery mildew resistance in common wheat. New Phytologist 228, 1011- 1026 (2020).13. Zhu, S. et al. Orthologous genes Pm12 and Pm21 from two wild relatives of wheat show evolutionary conservation but divergent powdery mildew resistance. Plant Commun 4, 100472 (2023).14. Xing, L. et al. Pm21 from Haynaldia villosa encodes a CC- NBS- LRR protein conferring powdery mildew resistance in wheat. Molecular Plant 11, 874- 878 (2018).15. He, H. et al. Pm21, encoding a typical CC- NBS- LRR protein, confers broad- spectrum resistance to wheat powdery mildew disease. Molecular Plant 11, 879- 882 (2018).16. Li, M. et al. A CNL protein in wild emmer wheat confers powdery mildew resistance. New Phytologist 228, 1027- 1037 (2020).17. Zou, S., Wang, H., Li, Y., Kong, Z. & Tang, D. The NB- LRR gene Pm60 confers powdery mildew resistance in wheat. New Phytologist 218, 298- 309 (2018).18. Li, Y. et al. Long- read genome sequencing accelerated the cloning of by resolving the complexity of a rapidly evolving resistance gene cluster in wheat. bioRxiv, 2022.10.14.512294 (2022).19. Sánchez- Martín, J. et al. Wheat Pm4 resistance to powdery mildew is controlled by alternative splice variants encoding chimeric proteins. Nature Plants 7, 327- 341 (2021). + +<--- Page Split ---> + +20. Moore, J.W. et al. A recently evolved hexose transporter variant confers resistance to multiple pathogens in wheat. Nature Genetics 47, 1494-1498 (2015).21. Krattinger, S.G. et al. A putative ABC transporter confers durable resistance to multiple fungal pathogens in wheat. Science 323, 1360-1363 (2009).22. Zhu, K. et al. Fine mapping of powdery mildew resistance gene M1WE74 derived from wild emmer wheat (Triticum turgidum ssp. dicoccoides) in an NBS-LRR gene cluster. Theor Appl Genet 135, 1235-1245 (2022).23. Friebe, B., Jiang, J., Raupp, W.J., Mcintosh, R.A. & Gill, B.S. Characterization of wheat-alien translocations conferring resistance to diseases and pests: current status. Euphytica (1996).24. Liu, W. et al. Homoecologous recombination-based transfer and molecular cytogenetic mapping of powdery mildew-resistant gene Pm57 from Aegilops searsii into wheat. Theoretical and Applied Genetics 130, 841-848 (2017).25. Liu, W. et al. Development and characterization of wheat-Ae. searsii Robertsonian translocations and a recombinant chromosome conferring resistance to stem rust. Theoretical and applied genetics 122, 1537-1545 (2011).26. Dong, Z. et al. Physical mapping of Pm57, a powdery mildew resistance gene derived from Aegilops searsii. Int J Mol Sci 21, 322 (2020).27. Wang, Y. et al. An unusual tandem kinase fusion protein confers leaf rust resistance in wheat. PREPRINT (Version 1) available at Research Square (2022).28. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583-589 (2021).29. Gyawali, Y., Zhang, W., Chao, S., Xu, S. & Cai, X. Delimitation of wheat ph1b deletion and development of ph1b-specific DNA markers. Theor Appl Genet 132, 195-204 (2019).30. Yu, G. et al. Aegilops sharonensis genome-assisted identification of stem rust resistance gene Sr62. Nature Communications 13, 1607 (2022).31. Klymiuk, V., Coaker, G., Fahima, T. & Pozniak, C.J. Tandem protein kinases emerge as new regulators of plant immunity. Molecular Plant-Microbe Interactions 34, 1094-1102 (2021).32. Brueggeman, R. et al. The barley stem rust-resistance gene Rpg1 is a novel disease-resistance gene with homology to receptor kinases. Proc Natl Acad Sci U S A 99, 9328-33 (2002).33. Klymiuk, V. et al. Cloning of the wheat Yr15 resistance gene sheds light on the plant tandem kinase-pseudokinase family. Nature Communications 9, 3735 (2018).34. Chen, S. et al. Wheat gene Sr60 encodes a protein with two putative kinase domains that confers resistance to stem rust. New Phytologist 225, 948-959 (2020).35. Gaurav, K. et al. Population genomic analysis of Aegilops tauschii identifies targets for bread wheat improvement. Nature Biotechnology 40, 422-431 (2022).36. Arora, S. et al. A wheat kinase and immune receptor form host-specificity barriers against the blast fungus. Nat Plants (2023).37. Li, Y., Gou, M., Sun, Q. & Hua, J. Requirement of calcium binding, myristoylation, and protein-protein interaction for the Copine BON1 function in Arabidopsis. J Biol Chem 285, 29884-91 (2010).38. Whittaker, C.A. & Hynes, R.O. Distribution and evolution of von Willebrand/integrin A domains: widely dispersed domains with roles in cell adhesion and elsewhere. Mol Biol Cell 13, 3369-87 (2002). + +<--- Page Split ---> + +39. Tomsig, J.L., Snyder, S.L. & Creutz, C.E. Identification of targets for calcium signaling through the copine family of proteins. Characterization of a coiled-coil copine-binding motif. J Biol Chem 278, 10048-54 (2003). + +40. Zou, B. et al. Identification and analysis of copine/BONZAI proteins among evolutionarily diverse plant species. Genome 59, 565-73 (2016). + +41. Pich, I.R.O. & Kondrashov, F.A. Long-term asymmetrical acceleration of protein evolution after gene duplication. Genome Biol Evol 6, 1949-55 (2014). + +42. Zhao, X., Fu, X., Yin, C. & Lu, F. Wheat speciation and adaptation: perspectives from reticulate evolution. Abiotech 2, 386-402 (2021). + +43. Wang, Z. et al. Dispersed emergence and protracted domestication of polyploid wheat uncovered by mosaic ancestral haploblock inference. Nat Commun 13, 3891 (2022). + +44. Wang, Z.L. et al. Seedling and adult plant resistance to powdery mildew in Chinese bread wheat cultivars and lines. Plant Disease 89, 457-463 (2005). + +45. Zhou, W. et al. TaNAC6s are involved in the basal and broad-spectrum resistance to powdery mildew in wheat. Plant Sci 277, 218-228 (2018). + +46. Zhao, Y. et al. Wheat heat shock factor TaHsfA2d contributes to plant responses to phosphate deficiency. Plant Physiol Biochem 185, 178-187 (2022). + +47. Zhao, Y. et al. The wheat MYB transcription factor TaMYB31 is involved in drought stress responses in Arabidopsis. Front Plant Sci 9, 1426 (2018). + +48. Schmittgen, T.D. & Livak, K.J. Analyzing real-time PCR data by the comparative C(T) method. Nat Protoc 3, 1101-8 (2008). + +49. Wang, K., Liu, H., Du, L. & Ye, X. Generation of marker-free transgenic hexaploid wheat via an Agrobacterium-mediated co-transformation strategy in commercial Chinese wheat varieties. Plant Biotechnology Journal 15, 614-623 (2017). + +50. Chen, Y. et al. A collinearity-incorporating homology inference strategy for connecting emerging assemblies in the Triticeae tribe as a pilot practice in the plant pangenomic era. Mol Plant 13, 1694-1708 (2020). + +51. Ma, S. et al. WheatOmics: A platform combining multiple omics data to accelerate functional genomics studies in wheat. Mol Plant 14, 1965-1968 (2021). + +52. Goodstein, D.M. et al. Phytozome: a comparative platform for green plant genomics. Nucleic Acids Res 40, D1178-86 (2012). + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +SupplementaryInformation4.19. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf_det.mmd b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..555f3eba4cffa67007dcc40fb4106337ef2ab381 --- /dev/null +++ b/preprint/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf/preprint__16bcf7f6cd89ab5cdb2a8579b6ff3a34550342621e51e3d6f35162100676c5bf_det.mmd @@ -0,0 +1,508 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 917, 207]]<|/det|> +# Pm57 from Aegilops searsii encodes a novel tandem kinase protein conferring powdery mildew resistance in bread wheat + +<|ref|>text<|/ref|><|det|>[[44, 230, 156, 247]]<|/det|> +Wenxuan Liu + +<|ref|>text<|/ref|><|det|>[[52, 258, 289, 273]]<|/det|> +wx1iu2003@hotmail.com + +<|ref|>text<|/ref|><|det|>[[44, 303, 858, 345]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 351, 130, 368]]<|/det|> +Yue Zhao + +<|ref|>text<|/ref|><|det|>[[44, 372, 858, 412]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 418, 160, 437]]<|/det|> +Zhenjie Dong + +<|ref|>text<|/ref|><|det|>[[52, 441, 510, 460]]<|/det|> +College of Agronomy, Nanjing Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 465, 164, 483]]<|/det|> +Jingnan Miao + +<|ref|>text<|/ref|><|det|>[[44, 487, 858, 528]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 534, 152, 551]]<|/det|> +Qianwen Liu + +<|ref|>text<|/ref|><|det|>[[44, 555, 951, 575]]<|/det|> +State Key Laboratory of Agrobiotechnology, College of Biological Sciences, China Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 581, 125, 598]]<|/det|> +Chao Ma + +<|ref|>text<|/ref|><|det|>[[44, 602, 858, 644]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 650, 150, 667]]<|/det|> +Xiubin Tian + +<|ref|>text<|/ref|><|det|>[[52, 671, 750, 691]]<|/det|> +Institute of Genetics and Developmental Biology, Chinese Academy of Sciences + +<|ref|>text<|/ref|><|det|>[[44, 696, 130, 714]]<|/det|> +Jinqiu He + +<|ref|>text<|/ref|><|det|>[[44, 718, 858, 758]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 764, 125, 782]]<|/det|> +Huihui Bi + +<|ref|>text<|/ref|><|det|>[[44, 787, 858, 827]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>text<|/ref|><|det|>[[44, 833, 125, 850]]<|/det|> +Wen Yao + +<|ref|>text<|/ref|><|det|>[[52, 855, 670, 874]]<|/det|> +Henan Agricultural University https://orcid.org/0000- 0002- 0643- 506X + +<|ref|>text<|/ref|><|det|>[[44, 880, 100, 897]]<|/det|> +Tao Li + +<|ref|>text<|/ref|><|det|>[[44, 902, 858, 942]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 202, 60]]<|/det|> +## Harsimardeep Gill + +<|ref|>text<|/ref|><|det|>[[45, 64, 820, 85]]<|/det|> +Department of Agronomy, Horticulture and Plant Science, South Dakota State University + +<|ref|>sub_title<|/ref|><|det|>[[44, 90, 160, 108]]<|/det|> +## Zhibin Zhang + +<|ref|>text<|/ref|><|det|>[[44, 111, 880, 153]]<|/det|> +Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China + +<|ref|>sub_title<|/ref|><|det|>[[44, 159, 157, 177]]<|/det|> +## Aizhong Cao + +<|ref|>text<|/ref|><|det|>[[44, 181, 872, 222]]<|/det|> +State Key Lab of Crop Genetics and Germplasm Enhancement, Cytogenetics Institute, Nanjing Agricultural University/JCIC- MCP + +<|ref|>sub_title<|/ref|><|det|>[[44, 228, 112, 245]]<|/det|> +## Bao Liu + +<|ref|>text<|/ref|><|det|>[[44, 249, 880, 291]]<|/det|> +Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China https://orcid.org/0000- 0001- 5481- 1675 + +<|ref|>sub_title<|/ref|><|det|>[[44, 296, 155, 314]]<|/det|> +## Huanhuan Li + +<|ref|>text<|/ref|><|det|>[[44, 318, 860, 360]]<|/det|> +The State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University + +<|ref|>sub_title<|/ref|><|det|>[[44, 365, 170, 383]]<|/det|> +## Sunish Sehgal + +<|ref|>text<|/ref|><|det|>[[50, 387, 818, 407]]<|/det|> +Department of Agronomy, Horticulture and Plant Science, South Dakota State University + +<|ref|>sub_title<|/ref|><|det|>[[44, 447, 103, 465]]<|/det|> +## Article + +<|ref|>sub_title<|/ref|><|det|>[[44, 485, 135, 504]]<|/det|> +## Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 522, 300, 542]]<|/det|> +Posted Date: April 28th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 560, 475, 580]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2844708/v1 + +<|ref|>text<|/ref|><|det|>[[44, 598, 914, 641]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 658, 535, 679]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 714, 905, 757]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on June 5th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 49257- 2. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[75, 100, 931, 123]]<|/det|> +# Pm57 from Aegilops searsii encodes a novel tandem kinase protein + +<|ref|>title<|/ref|><|det|>[[75, 145, 715, 169]]<|/det|> +# conferring powdery mildew resistance in bread wheat + +<|ref|>text<|/ref|><|det|>[[70, 184, 933, 625]]<|/det|> +Yue Zhao1,5, Zhenjie Dong2,5, Jingnan Miao1, Qianwen Liu1, Chao Ma1, Xiubin Tian1, Jinqiu He1, Huihui Bi1, Wen Yao1, Tao Li1, Harsimardeep S Gill3, Zhibin Zhang4, Aizhong Cao2, Bao Liu4, Huanhuan Li1\*, Sunish K Sehgal3\* & Wenxuan Liu1\*1 State Key Laboratory of Wheat and Maize Crop Science, College of Life Sciences, Henan Agricultural University, Zhengzhou 450002, China2 College of Agronomy, Nanjing Agricultural University, Nanjing 210000, China3 Department of Agronomy, Horticulture and Plant Science, South Dakota State University, Brookings, SD 57007, U.S.A.4 Key Laboratory of Molecular Epigenetics of the Ministry of Education (MOE), Northeast Normal University, Changchun 130024, China5 These authors contributed equally*e-mail: lihuanhuanhappy@henau.edu.cn; sunish.sehgal@sdtate.edu; wxliu2003@hotmail.com + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[63, 91, 202, 110]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[110, 130, 932, 378]]<|/det|> +Powdery mildew is a devastating disease that affects wheat yield and quality. Despite wheat wild relatives being a valuable source of resistance genes, their incorporation into wheat improvement is constrained by their adverse effects on agronomic traits, difficulty in isolating genes and poor understanding of the resistance mechanisms. Here, we report cloning of \(Pm57\) , the first gene isolated from Aegilops searsii. It encodes an unusual wheat tandem kinase (WTK) protein with putative kinase- pseudokinase domains followed by a von Willebrand factor A (vWA) domain, designated WTK7- vWA. The resistance function of \(Pm57\) was validated by independent mutants, gene silencing, and transgenic assays. Stable \(Pm57\) transgenic wheat lines showed high levels and all- stage resistance against diverse isolates of \(Bgt\) fungus with no adverse effects on agronomic traits. Our findings highlight the emerging role of kinase fusion proteins in plant disease resistance and provide a powerful gene for wheat breeding. + +<|ref|>sub_title<|/ref|><|det|>[[64, 398, 240, 418]]<|/det|> +## 29 Introduction + +<|ref|>text<|/ref|><|det|>[[110, 437, 932, 616]]<|/det|> +Bread wheat (Triticum aestivum L.) is an important staple food crop that plays a vital role in global food security'. However, sustainable wheat production across the globe is threatened due to its susceptibility to various pests and diseases. Among various diseases, powdery mildew caused by Blumeria graminis f. sp. tritici (Bgt) is a major wheat disease worldwide affecting grain yield and processing quality. This disease typically leads to yield losses ranging from 10 to \(15\%\) which may reach up to \(62\%\) in severe cases?. Development and deployment of resistant wheat varieties have been regarded as the most economical, effective, and sustainable way to mitigate the losses caused by powdery mildew. + +<|ref|>text<|/ref|><|det|>[[110, 637, 932, 793]]<|/det|> +To date, over a hundred \(Pm\) resistance genes/alleles at approximately 60 loci in wheat and its wild relatives have been documented?, however only a few \(Pm\) genes including \(Pm2\) , \(Pm4\) , \(Pm5\) , \(Pm6\) , \(Pm8\) , \(Pm21\) , and \(Pm52\) have been widely used in developing disease- resistant wheat varieties. The majority of \(Pm\) genes could not be utilized due to associated linkage drags causing adverse pleiotropism or narrow- spectrum resistance that often becomes ineffective with evolution of new \(Bgt\) races?. This necessitates the identification and cloning of new genes that offer broad- spectrum powdery mildew resistance with no adverse effects on the wheat agronomic characteristics. + +<|ref|>text<|/ref|><|det|>[[110, 814, 932, 925]]<|/det|> +Recently, several strategies have emerged for cloning resistant genes in plants, however, the classical method of map- based cloning still remains the most effective gene cloning approach in absence of an annotated reference genome. Owing to its huge genome size, cloning a gene in wheat by map- based cloning is time- consuming and challenging?. However, advances in genomics and availability of genome sequences of more than ten hexaploid wheat cultivars have largely facilitated gene cloning in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[62, 88, 933, 247]]<|/det|> +wheat. Since the first Pm gene Pm3b was isolated from wheat in \(2004^{7}\) , about \(16Pm\) resistance genes have been cloned. Eleven of these 16 genes i.e. Pm1a \(^{8}\) , Pm2 \(^{9}\) , Pm3/Pm8/Pm17 \(^{7,10,11}\) , Pm5e \(^{12}\) , Pm12 \(^{13}\) , Pm21 \(^{14,15}\) , Pm41 \(^{16}\) , Pm60 \(^{17}\) , and Pm69 \(^{18}\) , encode nucleotide-binding leucine-rich repeat (NLR) immune receptors. Of the remaining five Pm genes, Pm24 and WTK4 encode tandem kinases \(^{3}\) . The race- specific resistance gene Pm4 was characterized to be a putative serine/threonine kinase \(^{3}\) . Another two broad- spectrum resistance genes, Pm38 and Pm46, encode an ABC transporter and hexose transporter, respectively \(^{20,21}\) . + +<|ref|>text<|/ref|><|det|>[[61, 266, 933, 537]]<|/det|> +Further, wild relatives of hexaploid bread wheat serve as reservoirs of genetic diversity for important agronomic traits including disease resistance genes \(^{14}\) and more than half of the currently designated Pm genes have been derived from secondary and tertiary gene pool of wheat \(^{22,23}\) . Though Pm genes from wild relatives play an important role in breeding for disease resistance, it is very difficult to fine- map and clone the alien genes from secondary or tertiary gene pool in wheat background compared to genes from common wheat due to homoeologous recombination suppression, lack of alien chromosomes specific markers and unavailability annotated reference genomes. To date only four Pm genes have been cloned from wild relatives in secondary and tertiary gene pools of wheat, among which Pm12 (Ae. speltoides) and Pm21 (Dasypyrum villosum L.) were orthologous and Pm8 and Pm17 (Secale cereale L.) were cloned by Pm3 homology- based cloning, and all four genes encode NLR proteins. This hinders the better understanding of molecular basis of these genes, thus limiting their deployment in wheat breeding. + +<|ref|>text<|/ref|><|det|>[[61, 555, 933, 825]]<|/det|> +Powdery mildew resistance gene Pm57 \(^{24}\) was derived from Aegilops searsii Feldman & Kislev ex Hammer \((2\mathrm{n} = 2\mathrm{x} = 14\) , S'sS), an S- genome species from section Sitopsis (Jaub. & Spach) Zhuk in the secondary gene pool of wheat. Pm57 is the first Pm gene and second disease resistance gene following Sr51 that we transferred into bread wheat (Chinese Spring- Ae. searsii disomic addition line, TA3581) from Ae. searsii \(^{25}\) and mapped it to long arm of 2S\*#1 \(^{24,26}\) . In the present study, we report map- based cloning of Pm57, first gene to be isolated from Ae. searsii. Interestingly, Pm57 gene encodes a novel and unusual tandem kinase protein with putative kinase- pseudokinase domains followed by a von Willebrand factor A (vWA) domain and is designated WTK7- vWA. The stable Pm57 transgenic wheat lines showed high level and all- stage resistance against multiple Bgt isolates with no adverse effect on other agronomic traits. Our results offer significant insight into the molecular mechanisms underlying wheat powdery mildew resistance and will enable development of wheat varieties with broad- spectrum powdery mildew resistance. + +<|ref|>sub_title<|/ref|><|det|>[[63, 847, 188, 867]]<|/det|> +## 81 Results + +<|ref|>sub_title<|/ref|><|det|>[[63, 888, 599, 908]]<|/det|> +## 82 High-resolution mapping and map-based cloning of Pm57 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 933, 247]]<|/det|> +\(P m57\) was previously mapped to a 5.13 Mb region flanked by markers \(X67593\) and \(X62492\) in the terminal region of chromosome 2S's long arm using wheat- Ae. searsii 2S's recombinants26 (Fig. 1a). To fine- map \(P m57\) , progenies of single heterozygous wheat- Ae. searsii 2S's recombinant were used developed a large \(P m57\) segregation population. Two flanking markers \((X67593\) and \(X62492)\) were used for screening of 3,380 individuals and a total of 104 CS- Ae. searsii 2S's recombinants were identified (Fig. 1b). Subsequently, we performed \(B g t\) response assays on \(\mathrm{F}_{2:3}\) families of the 104 recombinants by inoculating \(B g t\) isolate E09. + +<|ref|>text<|/ref|><|det|>[[111, 265, 933, 469]]<|/det|> +Next, 16 2S's- specific markers were designed from the mapping interval of \(P m57\) flanked by markers \(X67593\) and \(X62492\) using the recently released genome sequence of Ae. searsii (TE01)1 (Table S1). Marker analysis of those 2S's recombinants grouped them into six different types (Types I to VI). Integrating \(B g t\) response of the 104 recombinants with the marker analysis, we mapped \(P m57\) to the region between markers \(X10\) and \(X13\) (Fig. 1b). The region corresponds to a physical interval of 710 kb on the long arm of Ae. searsii chromosome 2S's, which harbors 12 genes (referred to as G1- G12) based on genome sequence of Ae. searsii1. Two of these 12 genes (G4 and G5) contain putative tandem kinase domains and were considered potential candidate genes for \(P m57\) , while none of the other 10 genes were annotated to resemble any previously identified disease- resistant genes (Fig. 1c). + +<|ref|>image<|/ref|><|det|>[[163, 490, 882, 875]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 899, 931, 940]]<|/det|> +
Fig. 1 High-resolution physical mapping of \(P m57\) on chromosome 2S's. (a) Initial physical-mapping by comparing the chromosome recombination breakpoints of the resistant and susceptible recombinants positioned \(P m57\) within an
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 933, 222]]<|/det|> +interval flanked by markers X67593 and X62492 on the long arm of chromosome 2Ss. (b) A high-resolution genetic map delimited Pm57 to a 710 kb region between the markers X10 and X13. Markers X67593 and X62492 were derived from RNA-seq of CS- Ae. searsii 2Ss disomic addition line, and X1- X16 were developed using the recently released genome sequence of Ae. searsii (TE01)1. (c) Schematic representation of the twelve annotated genes G1- G12 in the mapping interval of Ae. searsii reference genome (TE01). Two of these 12 genes (G4 and G5) contain putative tandem kinase domains and were considered potential candidate genes for Pm57. + +<|ref|>sub_title<|/ref|><|det|>[[110, 243, 585, 264]]<|/det|> +## Identification of Pm57 candidate genes by MutRNA-Seq + +<|ref|>text<|/ref|><|det|>[[110, 283, 933, 394]]<|/det|> +To unravel Pm57 candidates, we performed EMS mutagenesis of Pm57 donor wheat- Ae. searsii recombinant line 89(5)69. About 10,000 \(\mathrm{M_0}\) seeds were treated with \(0.6\%\) EMS and developed 1,598 \(\mathrm{M_2}\) lines. Fifteen seeds from each of the randomly selected \(300\mathrm{M}_2\) families were screened for susceptible mutants using the Bgt isolate E09. Finally, 15 independent susceptible mutants were identified by Pm57- specific markers and a 55K SNP chip and further validated in the \(\mathrm{M}_3\) generation. + +<|ref|>text<|/ref|><|det|>[[110, 414, 936, 618]]<|/det|> +Further, we performed mutant RNA- seq (MutRNA- seq) using five susceptible mutants (Mut51, Mut60, Mut141, Mut209 and Mut216) along with resistant wild type 89(5)69 (Fig. 2a). Alignment of the 12 genes from Pm57 mapping interval to the transcriptome sequences revealed that three (Mut51, Mut141, and Mut209) of the five mutants had missense mutations of gene G4, while no mutations found in genes G2, G5, G6, G7, G8, G9, G10 and G11. The expression levels of the remaining three genes (G1, G3 and G12) were too low to reliably call mutations in transcriptome sequences (Fig. 2a and Table S2). Therefore, gene G4 encoding a tandem kinase - vWA domains protein emerged as the most likely candidate of Pm57 among the 12 genes, and was designated as WTK7- vWA since six WTKs (WTK1- 6) have been previously reported27. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[163, 99, 870, 610]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 614, 935, 631]]<|/det|> +
Fig. 2 Identification of Pm57 candidate gene using ethyl methane sulfonate (EMS)-induced mutants. (a)
+ +<|ref|>text<|/ref|><|det|>[[110, 636, 935, 860]]<|/det|> +Candidate gene identification by MutRNA- Seq. RNA- Seq reads from 89(5)69 and 5 susceptible EMS mutants (Mut51, Mut60, Mut141, Mut209 and Mut216) are mapped in Ae. searsii reference genome sequence (TE01). Only G4 of the 12 candidate genes in the mapping interval carry missense mutations (red dots) in three mutants (Mut51, Mut141, and Mut209). (b) Schematic representation of WTK7- vWA and identification of WTK7- vWA using ethyl methane sulfonate (EMS)-induced mutants. The position of three conserved domains and the predicted amino-acid changes caused by the EMS mutations were indicated. Exons and introns are represented by rectangles and lines, respectively. The two protein kinase domains and the vWA domain are shown in orange, blue, and green colors, respectively. The proportion of each isoform in the expression data of 89(5)69 at 0 and 24 h post inoculation (hpi) with Bgt isolate E09 is presented on the right side. IF1 is the main variant with 14 exons encoding for a complete protein. IF2- 7 transcripts were derived from the mis- splicing of exon or intron and generated premature stop codon (marked in red star). + +<|ref|>sub_title<|/ref|><|det|>[[110, 880, 789, 900]]<|/det|> +## Validation of WTK7- vWA by EMS mutants, gene silencing, and transgenic assays + +<|ref|>text<|/ref|><|det|>[[110, 920, 933, 940]]<|/det|> +To further validate the candidacy of WTK7- vWA for Pm57, we cloned the genomic DNA and cDNA + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 933, 498]]<|/det|> +sequences of WTK7- vWA gene (G4) and another WTK gene (G5) from 89(5)69 as well as the 15 susceptible mutants for sequence comparison. In the resistant line 89(5)69, the WTK7- vWA gene was 9,473 bp and contained 14 exons with a coding sequence of 3,489 bp (Fig. 2b). Intriguingly, WTK7- vWA had at least seven alternative splicing variants, designated IF1 - IF7, of which, IF1 was the main isoform, with a proportion of \(50.6\%\) (39 out of 77 tested WTK7- vWA cDNA clones) at \(0\mathrm{h}\) post- inoculation (hpi) with \(Bgt\) isolate E09 increasing to \(80.9\%\) at \(24\mathrm{hpi}\) , and isoforms IF2 - IF7 were much less abundant ( \(1.3\%\) - \(27.3\%\) ). IF1 encodes a full- length intact WTK7- vWA protein with Kin I, Kin II and vWA domains, while IF2 and IF3 encode proteins with truncated Kin I and Kin II domains, and IF4- IF7 encode proteins with only truncated Kin I domain (Fig. 2b). Gene sequence comparison revealed that ten of the 15 susceptible mutants had SNPs in WTK7- vWA that resulted in amino acid substitutions, premature stop codons, or relocation of the intron/exon splice sites (Fig. 2b). Specifically, a frameshift mutation was detected in Mut216 with a G/A point mutation in the splice acceptor site of intron 9. Mut351 had a nonsense mutation that gave rise to a premature stop codon at the amino acid position of 1,081. The other eight mutants (G78D in Mut223, G177E in Mut141, G193R in Mut51, D209N in Mut210, G424D in Mut92, P747L in Mut121, R829W in Mut22, and G903D in Mut209) harbored missense mutations that occurred in the kinase I (Kin I), kinase II (Kin II) or vWA domains (Fig. 2b). In addition, no sequence variations in gene G5 (WTK) were found among all susceptible mutants. + +<|ref|>text<|/ref|><|det|>[[111, 516, 932, 650]]<|/det|> +Next, virus- induced gene silencing (VIGS) was performed for validation of candidate genes G4 (WTK7- vWA) and G5 (WTK). The resistant introgression line 89(5)69 inoculated with G4- VIGS constructs lost resistance to powdery mildew, whereas the plants inoculated with G5- VIGS constructs and empty vector control constructs remained resistant (Fig. S1). These results were consistent with the conclusions derived from MutRNA- Seq analyses and suggested G4 (WTK7- vWA) was the prime candidate for Pm57. + +<|ref|>text<|/ref|><|det|>[[110, 670, 933, 918]]<|/det|> +Transgenic complementation of susceptible cv. Fielder was subsequently performed to confirm the role of WTK7- vWA in conferring resistance against powdery mildew. A total of 36 \(\mathrm{T_0}\) plants (L1 to L36) were generated, and 33 were identified as positive transgenic plants (Fig. 3a). With the exception of L20 and L29, all the other positive \(\mathrm{T_0}\) transgenic plants (+) were highly resistant to the Bgt isolate E09, whereas all the negative plants (- , L5, L23 and L27) were as susceptible as the WT Fielder (Fig. 3a). This difference in powdery mildew disease resistance was also observed for \(\mathrm{T_1}\) transgenic lines and WT (Table S3), and four representative lines (L1, L2, L4 and L5) were shown in Fig. 3b. In addition, qRT- PCR analyses showed that there was no expression of WTK7- vWA gene in the positive transgenic lines L20 and L29, explaining why these two positive lines behaved susceptible to powdery mildew (Fig. S2). Taken together, the genetic mapping, mutant analysis, gene silencing, and transgenic assays confirmed that WTK7- vWA is Pm57. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[150, 99, 895, 520]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 523, 932, 630]]<|/det|> +
Fig. 3 Functional validation of WTK7-vWA by transgenics. (a) Responses of 36 independent (L1 to L36) transgenic lines to \(Bgt\) isolate E09. The susceptible positive \(\mathrm{T_0}\) transgenic plants are indicated in red. (b) \(\mathrm{T_1}\) transgenic lines (L1 to L4) of WTK7-vWA show high resistance to \(Bgt\) isolate E09. Fielder and the negative line L5 were used as susceptible controls, respectively. Three individuals of each independent line are shown. The ‘+’ and ‘–’ signs on each leaf designate the presence or absence of WTK7-vWA gene.
+ +<|ref|>sub_title<|/ref|><|det|>[[112, 653, 648, 672]]<|/det|> +## Subcellular localization and structural analysis of Pm57 protein + +<|ref|>text<|/ref|><|det|>[[111, 693, 933, 917]]<|/det|> +Subcellular localization of Pm57 was investigated through transient transformation in wheat leaf protoplasts. We constructed the fusion protein of Pm57 with green florescent protein (GFP), and the resulting Pm57- GFP construct as well as the GFP control were introduced to wheat protoplasts, respectively. As shown in Fig. 4a, green fluorescence was ubiquitously detected in the cells transformed with the GFP control. Similarly, fluorescence of the Pm57- GFP fusion protein was observed in both nucleus and cytosol (Fig. 4a). To explore the structure of Pm57 in detail, the AlphaFold \(^{28}\) was used to generate a 3D model of Pm57. The tertiary structure revealed that Pm57 possesses a modular structure, with kinase- pseudokinase domains, a vWA domain, and a putative Vwaint domain as described by Wang et al. \(^{27}\) and two kinase domains in Pm57 were highly symmetrical (Fig. 4b). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[190, 95, 844, 485]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 500, 933, 655]]<|/det|> +
Fig. 4 Subcellular localization and protein structure prediction of Pm57. (a) Subcellular localization of Pm57 in wheat protoplasts. 35S::GFP or 35S::Pm57-GFP plasmids were co-transformed into wheat protoplasts with nucleus marker plasmid AtPIF4-mCherry. The GFP, mCherry and Chlorophyll fluorescence were visualized under a confocal laser-scanning microscope. 35S::GFP was served as a control. Scale bars, \(10 \mu \mathrm{m}\) . (b) Protein structure prediction of Pm57. Three-dimensional model of Pm57 is predicted by AlphaFold. Orange, kinase domain; blue, pseudokinase domain; Green, vWA domain; Yellow, putative Vwaint domain. The red spheres indicate amino acid substitutions resulting from the EMS mutagenesis.
+ +<|ref|>sub_title<|/ref|><|det|>[[113, 676, 520, 695]]<|/det|> +## Evaluation of the resistance mechanism of Pm57 + +<|ref|>text<|/ref|><|det|>[[111, 715, 933, 894]]<|/det|> +In order to evaluate the potential mechanism involved in Pm57- mediated resistance, the process of Bgt development was observed in the susceptible CS, resistant 89(5)69 and the transgenic lines. It was found that haustorium formation was inhibited by Pm57 in the early infection stage, and no secondary hypha branches or conidiophore could be observed on the epidermal cells (Fig. 5a). Bgt- infected leaf staining by diaminobenzidine (DAB) and trypan blue (TPN) further displayed robust accumulation of \(\mathrm{H}_2\mathrm{O}_2\) and cell death in the Bgt- interacting host cells of resistant 89(5)69 and the transgenic lines (Fig. 5a and 5b). These results suggest Pm57 inhibits haustorium formation of Bgt, and this process was correlated with \(\mathrm{H}_2\mathrm{O}_2\) accumulation and induced cell death. + +<|ref|>text<|/ref|><|det|>[[110, 914, 930, 935]]<|/det|> +The qRT- PCR analysis using seedling leaves of 89(5)69 demonstrated that Pm57 was highly expressed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 933, 201]]<|/det|> +in wheat leaf tissues and up-regulated by \(Bgt\) infection (Fig. S3, Fig. 5c). In addition, pathogenesis-related (PR) genes, including \(PR1\) , \(PR2\) , \(PR3\) , \(PR4\) , and \(PR9\) , were also up-regulated upon \(Bgt\) inoculation following \(Pm57\) up-regulation (Fig. 5c). These expression patterns of \(Pm57\) and \(PR\) genes imply that \(Pm57\) would be an upstream regulator of \(PR\) genes in the pathogen defense response pathway. + +<|ref|>image<|/ref|><|det|>[[147, 230, 857, 690]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[110, 700, 933, 876]]<|/det|> +
Fig. 5 Evaluation of the resistance mechanism of \(Pm57\) . (a) Detection of \(\mathrm{H}_2\mathrm{O}_2\) accumulation by DAB staining of \(Bgt\) -infected leaves. Seven-day-old plants were inoculated with \(Bgt\) isolate E09. Staining was performed on the \(Bgt\) -infected leaves at 2 d post inoculation (dpi). Brown staining shows the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) . Bar, \(200 \mu \mathrm{m}\) . (b) The pathogenic symptoms and trypan blue staining of the \(Bgt\) -infected leaves at 7 dpi to visualize plant cell death. Scale bar, \(200 \mu \mathrm{m}\) . Representative leaves were photographed at 7 dpi. Bar, \(10 \mathrm{mm}\) . (c) Expression analyses of \(Pm57\) and pathogenesis-related (PR) genes in \(89(5)69\) . The transcript levels were examined by qRT-PCR. TaActin1 was used as an endogenous control. Error bars represent mean \(\pm \mathrm{SD}\) from three biological samples. hpi, hours post inoculation. \(^{*}\mathrm{p} < 0.05\) , \(^{**} \mathrm{p} < 0.01\) (Student's \(t\) test).
+ +<|ref|>sub_title<|/ref|><|det|>[[110, 898, 730, 917]]<|/det|> +## Evolutionary analysis of \(Pm57\) locus based on the collinearity comparison + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 933, 292]]<|/det|> +A comparative analysis of published wheat genome sequences showed synteny is conserved to large extent in tribe Triticaeae for \(Pm57\) mapping genomic regions (Fig. S4), however, \(Pm57\) orthologs were present only in Aegilops bicornis (TB01) (SbSb), Ae. longissima (TL05) (SlSl), Ae. speltoides (TS01) (SS), and B sub genome of T. dicoccoides (WEWSeq v1) (AABB), and the encoding proteins of these orthologous genes had \(81.8 - 87.8\%\) sequence similarities with \(Pm57\) (Fig. S4 and Fig. S5). Notably, \(Pm57\) is absent in the wheat reference genome sequence of cv. Chinese Spring (CS), but highly similar sequences of \(Pm57\) were present in the syntenic region of several bread wheat cultivars (Table S4). These highly similar sequences were manually annotated and shared more than \(80\%\) amino acid sequence identity to \(Pm57\) . + +<|ref|>text<|/ref|><|det|>[[111, 312, 935, 537]]<|/det|> +To test whether \(Pm57\) exists in other wheat germplasms, a pair of gene- specific primers were designed to amplify a \(530\mathrm{bp}\) genomic sequence of the \(Pm57\) gene from 71 wheat accessions (including T. urartu, T. boeoticum, Ae. tauschii, T. monococcum, T. dicoccoides, T. durum, T. aestivum ssp. yunnanese, T. aestivum ssp. macha, T. aestivum ssp. spelta, T. aestivum ssp. tibetanum accessions and Chinese common wheat landraces and modern cultivars). The amplified fragment of Ae. searsii (TE01) was identical to that of the CS- Ae. searsii chromosome \(2\mathrm{S}^{\mathrm{s}}\) introgression line 89(5)69, but no fragment was obtained in the remaining 69 accessions (Fig. S6). These results indicate that \(Pm57\) probably exists only in Ae. searsii and has not been used in modern wheat breeding programs. The \(Pm57\) gene- specific primer pairs can be used as a diagnostic marker for effective detection of \(Pm57\) in molecular marker- assisted selection (MAS) breeding. + +<|ref|>sub_title<|/ref|><|det|>[[112, 558, 535, 577]]<|/det|> +## Evaluation of \(Pm57\) application in wheat breeding + +<|ref|>text<|/ref|><|det|>[[111, 597, 933, 777]]<|/det|> +Our previous studies showed that \(Pm57\) conferred high resistance to mixed Bgt isolates collected in Henan Province26. To further study the spectrum resistance offered by \(Pm57\) against genetically divergent Bgt isolates, we collected 29 Bgt isolates from major wheat growing regions of China, and used them for Bgt resistance assays of both wheat- Ae. searsii introgression lines 89(5)69 and homozygous transgenic line L1. As a result, plants of both 89(5)69 and transgenic line L1 showed high resistance, with Infection type 0- 1, to all of the 29 divergent Bgt isolates (Table S5). In addition, we found 89(5)69 and positive transgenic plants showed high resistance from the seedling stage to adult stage (Fig. S7). + +<|ref|>text<|/ref|><|det|>[[111, 797, 933, 930]]<|/det|> +To determine the practical value of \(Pm57\) in wheat breeding, \(Pm57\) transgenic wheat lines were evaluated agronomic traits. No significant differences were observed for heading date, plant height, tiller number, thousand grain weight and spike morphology between three transgenic lines and control lines, including both negative lines and WT (Fig. 6), showing no obvious adverse effect of \(Pm57\) on important agronomic traits. In addition, all of the \(\mathrm{F_1}\) plants derived from crosses of CS- Ae. searsii recombinants having small segments harboring \(Pm57\) with 22 wheat varieties were high resistance to + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 90, 930, 133]]<|/det|> +powdery mildew, indicating that Pm57 could play resistance role in diverse genetic backgrounds (Fig. S8). + +<|ref|>image<|/ref|><|det|>[[171, 156, 857, 579]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[111, 585, 933, 761]]<|/det|> +
Fig. 6 Statistical analysis of agronomic traits in WT and Pm57 transgenic lines. (a) Representative wheat whole plants, spikes and seeds of Fielder (-) and Pm57 transgenic plant (+). (b-h) Comparison of agronomic traits in Fielder and Pm57 transgenic plants: (b) heading date (n = 3), (c) thousand-grain weight (n = 3), (d) seeds per spike (n = 3), (e) plant height (n = 15), (f) spike length (n = 15), (g) spikelets per spike (n = 15), and (h) tiller numbers (n = 15). Data in (b-d) are displayed as bar graphs. Data are represented as the mean ± SD from three replicates. Data in (e-g) are displayed as box and whisker plots with individual data point. The error bars represent maximum and minimum values. Center line, median; box limits, 25th and 75th percentiles. ns, no statistically significant difference by two-tailed Student's \(t\) test.
+ +<|ref|>sub_title<|/ref|><|det|>[[113, 783, 219, 802]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[112, 822, 933, 933]]<|/det|> +Development of wheat- alien recombinants is not merely a tool to clone alien genes, but also is one of the best approaches to transfer favorable genes from wild relatives to increase the genetic diversity of cultivated wheat. However, homoeologous recombination between wheat and alien chromosomes is suppressed by the pairing homoeologous (Ph) genes hampering the development of recombinants with smaller alien segments which is essential for mapping and deployment of exotic wheat genes in bread + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 90, 933, 360]]<|/det|> +wheat cultivars. In this study, we used a Ph1 locus deletion mutant (TA3809, CS- ph1b) to induce homoeologous recombination29 between wheat and Ae. searsii to fine map and clone a broad- spectrum powdery mildew resistance gene Pm57 which is the first gene to be isolated from Ae. searsii. In absence of reference genome sequence of Ae. searsii1, we performed transcriptome sequence (RNA- Seq) analysis of CS- Ae. searsii 2Ss disomic addition line to develop chromosome 2Ss specific markers to physically map Pm57 into a 5.13 Mb interval on the long arm of chromosome2Ss26. Recent release of Ae. searsii reference genome sequence1 further accelerated the development of specific molecular markers for high- resolution mapping of Pm57. Combined with phenotypic analyses, Pm57 was fine- mapped to a genomic region of 710 kb. Meanwhile, we identified various translocation lines with small alien segments carrying Pm57 (Type IV, fragment size between 5.57 Mb and 9.12 Mb, \(\sim 1\%\) of 2Ss chromosome) which will greatly facilitate the deployment of Pm57 in elite wheat varieties through markers- assisted selection (Fig. 1b). + +<|ref|>text<|/ref|><|det|>[[110, 380, 933, 650]]<|/det|> +Pm57 candidate regions harbored 12 genes in Ae. searsii reference genome assembly (TE01)1. We further performed MutRNA- Seq on 89(5)69 (wild type) and five powdery mildew susceptible mutants to narrow down Pm57 candidate genes in target mapping interval and identified WTK7- vWA as the most likely candidate for further transgenic validation. Unexpectedly, the CDS sequence of WTK7- vWA was incompletely annotated as Asea[EVM0016946 in Ae. searsii reference genome, which encodes a protein with a single kinase domain followed by a vWA domain. We obtained the full- length sequence of WTK7- vWA in Pm57- direct donor RNA- seq data26 and aligned the MutRNA- Seq from the susceptible mutants against the WTK7- vWA- corresponding unigenes. Three of five loss- of- function mutants had SNPs in WTK7- vWA- corresponding unigenes leading to amino acid changes in the kinase and vWA domains (Fig. 2). Recently, two other genes have been cloned Sr6230 and Lr9/Lr5827 by exploiting similar MutRNA- Seq approach. Transgenic complementation of susceptible cv. Fielder confirmed WTK7- vWA was Pm57. + +<|ref|>text<|/ref|><|det|>[[110, 670, 933, 940]]<|/det|> +Recently, tandem kinase proteins (TKP) have emerged as a new prominent player involved in disease resistance in Triticeae31. To date eight TKPs, including Rpg132, Yr15 (WTK1)33, Sr60 (WTK2)34, Sr62 (WTK5)30, Pm24 (WTK3)3 and WTK435, Lr9 (WTK6)27, and Rwt436, have been identified and confer resistance against various fungal pathogens. Pm57 (WTK7- vWA) from Ae. searsii is a new member of the TKP family conferring wheat powdery mildew resistance. Based on the sequence conservation of the key amino acid residues in the two kinase domains3, Pm57 was classified as a tandem kinase- pseudokinase protein, similar to Yr15, Pm24, Lr9, and Sr62 (Fig. S9). However, Pm57 contains tandem kinase domains and a Willebrand factor A (vWA) domain, the second such unusual WTK- vWA structure to be identified following Lr9 which provides resistance to leaf rust whereas Pm57 confers resistance to powdery mildew. The vWA/Vwaint domains are presents in bacteria, archaea and Eukaryota organisms are considered to participate in protein- protein interactions37,38. For example, the vWA domain of human copine are capable of interacting with a wide variety of signaling molecules + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 933, 247]]<|/det|> +with coiled- coil domains39. In plants, the vWA- containing copine proteins were shown to be regulators of basal and \(R\) - mediated disease resistance, suggesting that vWA- containing proteins may play an important role in plant disease resistance40. In this study, eight susceptible mutants harbored missense mutations in Kin I, Kin II or vWA domains (Fig. 2), indicating that each domain of Pm57 is essential for resistance to the \(Bgt\) pathogen. A similar phenomenon was also observed for \(Rpg1\) , \(Pm24\) , \(Sr62\) and \(Lr9\) genes3,27,30,32. No sequence variation of \(Pm57\) was detected in another five susceptible mutants, indicating possible mutations in genes or elements involved in the \(Pm57\) regulation pathway. + +<|ref|>text<|/ref|><|det|>[[110, 266, 933, 492]]<|/det|> +To explore the structure of Pm57 in more detail, we generated its 3D model using AlphaFold28. We observed that kinase domains in Pm57 were highly symmetrical like WTK4 and Lr9 but not similar with the other five reported TPKs (Fig. 4b and Fig. S10), indicating that kinase domains of Pm57 arose from gene duplication41. Although Pm57 had the same kind of domains and a high similarity of \(88.3\%\) in amino acid sequences with Lr9, the amino acid sequences in the pseudokinase and vWA domains and protein structures are clearly different in Pm57 (Fig. S10 and Fig. S11). Wang et al. hypothesized that the pseudokinase and vWA domains of Lr9 might serve as integrated decoys for the detection of pathogen effectors27. It is reasonable to expect the differences of pseudokinase and vWA domains from Pm57 and Lr9 may lead to differently detect the effectors of \(Bgt\) and \(Pt\) pathogens, and thus confer resistance to powdery mildew and leaf rust, respectively. + +<|ref|>text<|/ref|><|det|>[[110, 510, 933, 850]]<|/det|> +Comparative analysis Pm57 orthologs among plant kingdoms revealed that its orthologs are only present in \(Ae\) . bicornis (TB01.2S01G0903000.1), \(Ae\) . longissima (TL05.2S01G0920200.1), \(Ae\) . speltoides (TS01.2B01G0896900.1), \(T\) . dicoccoides (TRIDC2BG085800.1) and several sequenced bread wheat accessions (Fig. S4 and Table S4), suggesting a likely recent origin of \(Pm57\) after the divergence of Triticaea species, and the reticulate evolutionary nature of wheat42,43. However, several genes encoding proteins with a single kinase domain followed by a vWA domain were detected not only in Triticaea species but also in other species (Table S4). Moreover, the sequences of single kinase- vWA proteins from \(Ae\) . bicornis. TB01. Un01G0378900.1 and Thint.05G0470400.1. p are very similar with Pm57 ( \(>90\%\) , Fig. S12). These results suggest that tandem kinase- vWA proteins are probably derived from single kinase- vWA proteins. We further found that the Kin I and Kin II domains of Pm57 are located in the same Clade of a phylogenetic tree generated using each kinase domain of tandem kinase proteins (Fig. S13), indicating that the tandem kinase of Pm57 could have resulted from a duplication event31. The relatively high similarity (58.04%) of amino acid sequences between the Kin I and Kin II of Pm57 confirms a relatively recent duplication event of the two- kinase domains, being consistent with the comparative analysis of Pm57 orthologs. + +<|ref|>text<|/ref|><|det|>[[110, 870, 932, 936]]<|/det|> +In summary, we cloned \(Ae\) . searsii- derived \(Pm57\) that confers broad- spectrum and all- stage resistance against \(Bgt\) . Pm57 encodes an unusual protein consisting of tandem kinase domains and a vWA domain (WTK7- vWA), the first of this kind of structure conferring powdery mildew among cloned plant + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 90, 932, 178]]<|/det|> +resistance genes. The isolation of \(Pm57\) lays a solid foundation for further understanding of the molecular mechanism behind WTK- vWA- mediated resistance to various plant diseases. The new introgression lines carrying \(Pm57\) in a small alien segment and with diagnostic markers will facilitate development of elite wheat varieties with durable powdery mildew resistance. + +<|ref|>sub_title<|/ref|><|det|>[[112, 199, 342, 219]]<|/det|> +## Materials and methods + +<|ref|>sub_title<|/ref|><|det|>[[112, 240, 245, 258]]<|/det|> +## Plant materials + +<|ref|>text<|/ref|><|det|>[[111, 279, 933, 481]]<|/det|> +Common wheat (T. aestivum L.) cultivar Chinese Spring (CS, TA3808), CS \(ph1b\) mutant stock (TA3809), and heterozygous CS- Ae. searsii chromosome \(2\mathrm{S}^{\mathrm{s}}\) introgression line 89(5)69 (T2BS- 2BL- 2S#1L, TA5109) and 89(6)88 (T2S#1S- 2S#1L- 2AL), both carrying \(Pm57\) , were used to develop the wheat- Ae. searsii recombinant population \(^{26}\) . 89(5)69 and the susceptible control CS were used for \(Pm57\) gene cloning and expression analyses. The wheat cultivar Fielder was used for wheat protoplast preparation and transformation. A total of 71 wheat accessions including diploid, tetraploid and hexaploid wheat accessions were employed to check the presence of the \(Pm57\) gene (Table S6). All materials were grown in a greenhouse that was maintained at \(18 - 24^{\circ}\mathrm{C}\) with \(16\mathrm{~h~}\) light/8 h dark and approximately \(70\%\) relative humidity. + +<|ref|>sub_title<|/ref|><|det|>[[112, 502, 500, 521]]<|/det|> +## Phenotypic response to Bgt (Powdery mildew) + +<|ref|>text<|/ref|><|det|>[[111, 541, 933, 766]]<|/det|> +Bgt isolate E09 and other 28 genetically divergent isolates (Table S5) collected from different regions of China were used for powdery mildew evaluation. Wheat seedlings were inoculated with Bgt isolates as previously described \(^{24}\) . Disease symptoms were recorded 7 days post- inoculation (dpi) using a scale from infection type 0 to 4 (IT 0 for no visible symptoms, IT 0; for hypersensitive necrotic flecks, IT 1- 4 for highly resistant, moderately resistant, moderately susceptible and highly susceptible) \(^{44}\) . Based on IT scores, tested plants were classified into two groups, resistant (R, IT 0- 2) and susceptible (S, IT 3- 4). CS was used as a susceptible control and for propagating Bgt isolates. The primary transgenic plants (T₀) were tested for powdery mildew resistance using detached leaves. Briefly, detached leaves from plants were placed on phytagar media (0.5% phytagar; 30 ppm Benzimidazole), inoculated with Bgt isolate E09 and cultured in a greenhouse. ITs were recorded 7 days after inoculation. + +<|ref|>sub_title<|/ref|><|det|>[[112, 787, 338, 806]]<|/det|> +## Physical mapping of Pm57 + +<|ref|>text<|/ref|><|det|>[[111, 827, 933, 938]]<|/det|> +The wheat- Ae. searsii \(2\mathrm{S}^{\mathrm{s}}\) recombinants population were developed from the cross of CS- \(ph1b\) mutant stock TA3809 and 89(6)88 as previously described \(^{26}\) . Heterozygous resistant wheat- Ae. searsii \(2\mathrm{S}^{\mathrm{s}}\) recombinants were self- pollinated to produce a secondary mapping population for further mapping of \(Pm57\) . The Ae. searsii genome sequence was used for the development of STS- PCR markers \(^{1}\) . All markers (Table S1) were designed using DNAMAN 7 software (Lynnon Biosoft, San Ramon, CA, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 89, 931, 133]]<|/det|> +USA). DNA was extracted using the CTAB method, and genotyping of the recombinants was performed in \(15 \mu \mathrm{L}\) volumes with parameters as previously described24. + +<|ref|>sub_title<|/ref|><|det|>[[112, 154, 448, 173]]<|/det|> +## Mutant development and MutRNA-Seq + +<|ref|>text<|/ref|><|det|>[[111, 193, 933, 373]]<|/det|> +Seeds of 89(5)69 were soaked in distilled water for \(6 \mathrm{~h}\) and treated for \(16 \mathrm{~h}\) with a \(0.6\%\) (v/v) EMS solution with shaking at \(150 \mathrm{rpm}\) at room temperature. The solution was then removed, and the treated seeds were rinsed with running water for \(6 \mathrm{~h}\) . The mutagenized \(\mathrm{M}_{0}\) seeds were planted in the field and the \(\mathrm{M}_{2}\) seeds from each \(\mathrm{M}_{1}\) plant were harvested at maturity. The \(\mathrm{M}_{2}\) seedlings (10- 15 plants for each family) were phenotyped with \(Bgt\) isolate E09 in a greenhouse. Susceptible \(\mathrm{M}_{2}\) plants were advanced to \(\mathrm{M}_{3}\) generation, and \(\mathrm{M}_{3}\) seedlings were tested to confirm their susceptibility. In order to eliminate the susceptibility caused by seed contamination or missing \(2 \mathrm{~S}^{3}\) , a subset of mutants was verified using \(2 \mathrm{~S}^{3}\) - specific molecular markers \(X67593\) and \(X62492\) flanking \(\mathrm{Pm}57\) . + +<|ref|>text<|/ref|><|det|>[[111, 392, 933, 617]]<|/det|> +MutRNA- Seq was performed as described by Yu et al.30. Five susceptible mutants derived from independent \(\mathrm{M}_{2}\) families were selected for RNA- seq. Total RNA from the \(Bgt\) - inoculated seedlings of the susceptible mutants and the wild- type line 89(5)69 was extracted using the TRIzol reagent (TransGen, Beijing, China). RNA- seq was performed as a service at Annoroad Gene Technology Co., Ltd (Beijing, China). Illumina HiSeq X Ten platform (Illumina, USA) was used to generate 150 bp pair- ended reads. Clean reads from the five mutants and 89(5)69 were mapped to the CDS of the twelve genes in the \(\mathrm{Pm}57\) mapping interval using BWA (version 0.7.17) and SAMtools (version 1.9) pipeline. Gene expression levels were quantified using the featureCounts tool in subread software (version 1.4.4). Picard- tools (version 2.27.1) and GATK4 were used for single nucleotide polymorphism (SNP) calling. + +<|ref|>sub_title<|/ref|><|det|>[[112, 644, 358, 663]]<|/det|> +## Virus-induced gene silencing + +<|ref|>text<|/ref|><|det|>[[111, 670, 934, 896]]<|/det|> +Virus- induced gene silencing (VIGS) was performed as previously described45. To develop specific VIGS targets, we first blasted the sequences of \(G4\) and \(G5\) against CS and \(Ae\) . \(searsii\) genome sequence. \(G4\) and \(G5\) fragments of 200- 250 bp with very low similarities with other genes were selected as targets; they were separately cloned into BSMV- \(\gamma\) (BSMV, barley stripe mosaic virus) vector, resulting in constructs \(\gamma\) - G4 and \(\gamma\) - G5. Equimolar amounts of in vitro transcripts of BSMV- \(\alpha\) , BSMV- \(\beta\) and \(\gamma\) - G4 or \(\gamma\) - G5 were mixed to inoculate the full- expanded second leaves of 89(5)69 seedlings, and the leaves infected with BSMV- TaPDS and BSMV- \(\gamma\) (empty vector) were used as controls as previously described45. About 14 days after virus infection, the \(3^{\mathrm{rd}}\) and \(4^{\mathrm{th}}\) leaves were detached and placed on \(1\%\) agar plates supplemented with \(20 \mathrm{mg / mL}\) 6- phenyl- adenine (6- BA), and then inoculated with \(Bgt\) . Seven days later, the phenotype of powdery mildew resistance was evaluated. + +<|ref|>sub_title<|/ref|><|det|>[[112, 917, 414, 935]]<|/det|> +## Gene cloning and sequence analysis + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 89, 933, 222]]<|/det|> +The full- length genomic DNA (gDNA) sequences and cDNAs of WTK7- vWA from 89(5)69 and each of the 15 susceptible mutants were amplified using the primers listed in Table S1. PCRs were performed in \(30~\mu \mathrm{L}\) volumes using high- fidelity Primestar polymerase (TaKaRa, Dalian, China). The PCR conditions were as previously described46. The PCR products were sequenced by Sanger dideoxy DNA sequencing method. The sequences of WTK7- vWA in each mutant and in the wild- type 89(5)69 were compared using DNAMAN 7 software. + +<|ref|>sub_title<|/ref|><|det|>[[113, 244, 324, 262]]<|/det|> +## Gene expression analysis + +<|ref|>text<|/ref|><|det|>[[110, 284, 933, 440]]<|/det|> +TRIzol reagent (TransGen, Beijing, China) was used for RNA extraction and \(2\mu \mathrm{g}\) of total RNA was used for cDNA synthesis using HiScript II 1st Strand cDNA Synthesis Kit (+gDNA wiper) (Vazyme, Nanjing, China). Quantitative RT- PCR (qRT- PCR) analysis was carried out using SYBR Mix (TaKaRa, Dalian, China) on a CFX96 real- time PCR detection System (Bio- Rad, Hercules, CA, USA), with three biological replicates for each sample or treatment. The conditions for qRT- PCR were the same as previously described47. The transcript levels were calculated using the comparative CT method48. The primers used for qRT- PCR are listed in Table S1. + +<|ref|>sub_title<|/ref|><|det|>[[113, 461, 303, 479]]<|/det|> +## Wheat transformation + +<|ref|>text<|/ref|><|det|>[[110, 501, 933, 680]]<|/det|> +The full- length coding sequence of WTK7- vWA was inserted into pWMB110 vectors using the restriction enzymes BamH I under the control of the maize ubiquitous (Ubi) promoter. Wheat transformation was performed using the Agrobacterium- mediated method with strain EHA105 and calluses induced from cv. Fielder immature embryos49. To determine positive transgenic events, DNA was extracted from leaves of independent \(\mathrm{T_0}\) plants, and specific PCR primers were designed to amplify a 344 bp fragment of the WTK7- vWA gene. qRT- PCR analysis was performed to evaluate the expression levels of WTK7- vWA in the leaves of transgenic wheat plants in the \(\mathrm{T_0}\) generation. The disease responses of the transgenic plants to powdery mildew were tested as described above. + +<|ref|>sub_title<|/ref|><|det|>[[112, 701, 510, 720]]<|/det|> +## Agronomic evaluation of Pm57 transgenic lines + +<|ref|>text<|/ref|><|det|>[[110, 741, 933, 920]]<|/det|> +Three \(\mathrm{T_2}\) transgenic lines of Pm57 and control lines (including both negative lines and non- transformed cv. Fielder) were planted in the experimental fields of Henan Agricultural University (Zhengzhou, China) using a randomized block design with three replications. Each plot consisted of two 1.5- m rows spaced 25 cm apart, and 20 seeds were sown in each row. Regular field management, including irrigation and fertilization, were applied. In each plot, 5 plants were chosen to measure various agronomic traits, including heading date, plant height, tiller numbers, spike length, spikelets per spike, seeds per spike, and thousand- grain weight. The significance of differences among means of agronomic traits was determined using Student's \(t\) - test. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[111, 91, 384, 110]]<|/det|> +## Subcellular localization analysis + +<|ref|>text<|/ref|><|det|>[[111, 130, 932, 265]]<|/det|> +To determine the subcellular location of \(Pm57\) , the coding sequence of \(Pm57\) was cloned into pJIT163- GFP vector, in which the expression of \(Pm57 - GFP\) was driven by the \(CaMV35S\) promoter. An empty vector, pJIT163- GFP, was used as the negative control. Under an induction of \(40\%\) PEG- 4000, control or recombinant plasmids were co- transformed into wheat protoplasts with nucleus marker plasmid AtPIF4- mCherry. The transformed protoplasts were cultured at \(25^{\circ}\mathrm{C}\) for \(16\mathrm{h}\) under dark conditions, and observed using a laser confocal microscope (A1F, Nikon, Tokyo, Japan). + +<|ref|>sub_title<|/ref|><|det|>[[112, 285, 432, 305]]<|/det|> +## Pm57 protein 3D modeling prediction + +<|ref|>text<|/ref|><|det|>[[111, 325, 932, 459]]<|/det|> +To predict the 3D structure of Pm57 and other WTKs, we used the open- source code of AlphaFold v2.128. We input the amino acid sequence of each WTKs into AlphaFold v2.1, and obtained five unrelaxed, five relaxed and five ranked models in .pdb format. Among the output models, the ranked_1. pdb model had the highest confidence with the best Local Distance Difference Test (IDDT) score were utilized. The structural graphics and the positions of amino acid substitutions were visualized using PyMOL (v.2.6.0). + +<|ref|>sub_title<|/ref|><|det|>[[111, 479, 553, 499]]<|/det|> +## Detection of \(\mathrm{H}_2\mathrm{O}_2\) accumulation and plant cell death + +<|ref|>text<|/ref|><|det|>[[111, 519, 932, 722]]<|/det|> +To detect the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) , the first leaves cut from 89(5)69, transgenic line L1 and CS at 2 d post inoculation (dpi) were immediately incubated in a 3, \(3^{\prime}\) - diaminobenzidine (DAB) solution (1 mg/mL, pH 5.8) for \(12\mathrm{h}\) , and then bleached in absolute ethanol. Before assessing the accumulation of \(\mathrm{H}_2\mathrm{O}_2\) , the bleached leaves were incubated in a \(0.6\%\) (w/v) Coomassie blue solution for \(10\mathrm{s}\) and then washed with water. To detect plant cell death, the primary leaves from the 89(5)69, L1 and CS at 7 dpi were incubated in a \(0.4\%\) Trypan blue solution for \(1\mathrm{min}\) in boiling water, washed with sterile water, bleached for \(16\mathrm{h}\) in chloral hydrate solution (2.5 g/mL), fixed for \(20\mathrm{min}\) in ethanol- acetic acid 3:1 (v/v), and stained in a \(0.6\%\) (w/v) Coomassie blue solution for \(30\mathrm{s}\) . The treated leaves were viewed under a microscope (Olympus BX53). + +<|ref|>sub_title<|/ref|><|det|>[[111, 741, 668, 762]]<|/det|> +## Collinearity analysis, homology searching and phylogenic analysis + +<|ref|>text<|/ref|><|det|>[[111, 782, 932, 938]]<|/det|> +Collinearity analysis among different species or subgenomes was performed using the online tool Triticeae- GeneTribe with default parameters50. The WheatOmics 1.0 (http://202.194.139.32/) and Phytozome v13 database (https://phytozome- next.jgi.doe.gov/) were used to find proteins similar to Pm57 in plant genomes51,52. Pm57 were used as queries for BLAST analysis and the retrieved proteins with kinase domain and vWA domain were selected. For kinase domain analysis, the 182 putative kinase or pseudokinase domains used for phylogenetic analysis of WTK3 ( \(Pm24^{3}\) ) were also used in this study. In addition, the proteins homologous to Pm57 with a tandem kinase - vWA structure were + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 88, 933, 179]]<|/det|> +included in the phylogenetic analysis of kinase domains. Multiple sequence alignments were carried out with ClustalW software with default settings. The conserved motifs in the Kin I and Kin II domains were annotated as described previously3. Phylogenetic analysis was conducted with Clustal Omega (https://www.ebi.ac.uk/Tools/msa/clustalo/) and drawn with iTOL (https://itol.embl.de/). + +<|ref|>sub_title<|/ref|><|det|>[[112, 199, 278, 220]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[110, 239, 933, 373]]<|/det|> +All data generated or analyzed in this study are included in this article and Supplementary Information files as well as the public databases. The WTK7- vWA sequence have been deposited in NCBI Genbank under accession number OQ675542. The MutRNA-Seq data derived from the WT and five mutant plants have been deposited in NCBI's Sequence Read Archive (SRA) under accession number PRJNA947672. The other data that support the findings of this study are available from the corresponding author upon request. + +<|ref|>sub_title<|/ref|><|det|>[[112, 393, 305, 414]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[110, 432, 935, 589]]<|/det|> +We are grateful to Prof. Zhongfu Ni and Huiru Peng from China Agricultural University, Beijing, China, for their advice and supports during this research. We are also grateful to Prof. Pengtao Ma of Yantai University, Yantai, Shandong, China, for providing Bgt isolates and powdery mildew resistance assays. This research was financially supported by the National Natural Science Foundation of China (31801363, 31971887 and 32272070), the Scientific and Technological Research Project of Henan Province of China (222103810004) and the Key Scientific Research Projects of Higher Education Institutions in Henan Province (23A210020) and South Dakota Wheat Commission (3x2030). + +<|ref|>sub_title<|/ref|><|det|>[[112, 610, 323, 630]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[110, 650, 933, 715]]<|/det|> +W.L., S.S. and H.L. designed the study. Y.Z., Z.D., J.M., C.M., X.T., and J.H. performed the research. Q.L., H.B., W.Y., T.L., A.C., H.L. H.G. and S.S. analyzed the data. W.L. and Y.Z. wrote the manuscript and S.S. contributed to revising the draft. All authors have read and approved the final manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[112, 735, 312, 757]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[112, 777, 457, 796]]<|/det|> +The authors declare no competing interests. + +<|ref|>sub_title<|/ref|><|det|>[[112, 817, 310, 835]]<|/det|> +## Additional information + +<|ref|>text<|/ref|><|det|>[[112, 857, 538, 876]]<|/det|> +Supplementary information is available for this paper. + +<|ref|>sub_title<|/ref|><|det|>[[112, 902, 228, 920]]<|/det|> +## References: + +<|ref|>text<|/ref|><|det|>[[112, 925, 930, 944]]<|/det|> +1. Li, L. et al. Genome sequences of five Sitopsis species of Aegilops and the origin of polyploid wheat B subgenome. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 90, 936, 930]]<|/det|> +Molecular Plant 15, 488- 503 (2022).2. Wu, L. et al. Genetic dissection of the powdery mildew resistance in wheat breeding line LS5082 using BSR-Seq. The Crop Journal 10, 1120- 1130 (2022).3. Lu, P. et al. A rare gain of function mutation in a wheat tandem kinase confers resistance to powdery mildew. Nature Communications 11, 680 (2020).4. Wang, W. et al. Characterization of the powdery mildew resistance gene in wheat breeding line KN0816 and its evaluation in marker- assisted selection. Plant Dis 105, 4042- 4050 (2021).5. Li, H. et al. A spontaneous wheat- Aegilops longissima translocation carrying Pm66 confers resistance to powdery mildew. Theoretical and Applied Genetics 133, 1149- 1159 (2020).6. Wang, Y. et al. Mapping stripe rust resistance gene YrZH22 in Chinese wheat cultivar Zhoumai 22 by bulked segregant RNA- Seq (BSR- Seq) and comparative genomics analyses. Theor Appl Genet 130, 2191- 2201 (2017).7. Yahiaoui, N., Srichumpa, P., Dudler, R. & Keller, B. Genome analysis at different ploidy levels allows cloning of the powdery mildew resistance gene Pm3b from hexaploid wheat. Plant J 37, 528- 38 (2004).8. Hewitt, T. et al. A highly differentiated region of wheat chromosome 7AL encodes a Pm1a immune receptor that recognizes its corresponding AvrPm1a effector from Blumeria graminis. New Phytologist 229, 2812- 2826 (2021).9. Sánchez- Martín, J. et al. Rapid gene isolation in barley and wheat by mutant chromosome sequencing. Genome Biology 17, 221 (2016).10. Singh, S.P. et al. Evolutionary divergence of the rye Pm17 and Pm8 resistance genes reveals ancient diversity. Plant Molecular Biology 98, 249- 260 (2018).11. Hurni, S. et al. Rye Pm8 and wheat Pm3 are orthologous genes and show evolutionary conservation of resistance function against powdery mildew. Plant J 76, 957- 969 (2013).12. Xie, J. et al. A rare single nucleotide variant in Pm5e confers powdery mildew resistance in common wheat. New Phytologist 228, 1011- 1026 (2020).13. Zhu, S. et al. Orthologous genes Pm12 and Pm21 from two wild relatives of wheat show evolutionary conservation but divergent powdery mildew resistance. Plant Commun 4, 100472 (2023).14. Xing, L. et al. Pm21 from Haynaldia villosa encodes a CC- NBS- LRR protein conferring powdery mildew resistance in wheat. Molecular Plant 11, 874- 878 (2018).15. He, H. et al. Pm21, encoding a typical CC- NBS- LRR protein, confers broad- spectrum resistance to wheat powdery mildew disease. Molecular Plant 11, 879- 882 (2018).16. Li, M. et al. A CNL protein in wild emmer wheat confers powdery mildew resistance. New Phytologist 228, 1027- 1037 (2020).17. Zou, S., Wang, H., Li, Y., Kong, Z. & Tang, D. The NB- LRR gene Pm60 confers powdery mildew resistance in wheat. New Phytologist 218, 298- 309 (2018).18. Li, Y. et al. Long- read genome sequencing accelerated the cloning of by resolving the complexity of a rapidly evolving resistance gene cluster in wheat. bioRxiv, 2022.10.14.512294 (2022).19. Sánchez- Martín, J. et al. Wheat Pm4 resistance to powdery mildew is controlled by alternative splice variants encoding chimeric proteins. Nature Plants 7, 327- 341 (2021). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 90, 936, 930]]<|/det|> +20. Moore, J.W. et al. A recently evolved hexose transporter variant confers resistance to multiple pathogens in wheat. Nature Genetics 47, 1494-1498 (2015).21. Krattinger, S.G. et al. A putative ABC transporter confers durable resistance to multiple fungal pathogens in wheat. Science 323, 1360-1363 (2009).22. Zhu, K. et al. Fine mapping of powdery mildew resistance gene M1WE74 derived from wild emmer wheat (Triticum turgidum ssp. dicoccoides) in an NBS-LRR gene cluster. Theor Appl Genet 135, 1235-1245 (2022).23. Friebe, B., Jiang, J., Raupp, W.J., Mcintosh, R.A. & Gill, B.S. Characterization of wheat-alien translocations conferring resistance to diseases and pests: current status. Euphytica (1996).24. Liu, W. et al. Homoecologous recombination-based transfer and molecular cytogenetic mapping of powdery mildew-resistant gene Pm57 from Aegilops searsii into wheat. Theoretical and Applied Genetics 130, 841-848 (2017).25. Liu, W. et al. Development and characterization of wheat-Ae. searsii Robertsonian translocations and a recombinant chromosome conferring resistance to stem rust. Theoretical and applied genetics 122, 1537-1545 (2011).26. Dong, Z. et al. Physical mapping of Pm57, a powdery mildew resistance gene derived from Aegilops searsii. Int J Mol Sci 21, 322 (2020).27. Wang, Y. et al. An unusual tandem kinase fusion protein confers leaf rust resistance in wheat. PREPRINT (Version 1) available at Research Square (2022).28. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583-589 (2021).29. Gyawali, Y., Zhang, W., Chao, S., Xu, S. & Cai, X. Delimitation of wheat ph1b deletion and development of ph1b-specific DNA markers. Theor Appl Genet 132, 195-204 (2019).30. Yu, G. et al. Aegilops sharonensis genome-assisted identification of stem rust resistance gene Sr62. Nature Communications 13, 1607 (2022).31. Klymiuk, V., Coaker, G., Fahima, T. & Pozniak, C.J. Tandem protein kinases emerge as new regulators of plant immunity. Molecular Plant-Microbe Interactions 34, 1094-1102 (2021).32. Brueggeman, R. et al. The barley stem rust-resistance gene Rpg1 is a novel disease-resistance gene with homology to receptor kinases. Proc Natl Acad Sci U S A 99, 9328-33 (2002).33. Klymiuk, V. et al. Cloning of the wheat Yr15 resistance gene sheds light on the plant tandem kinase-pseudokinase family. Nature Communications 9, 3735 (2018).34. Chen, S. et al. Wheat gene Sr60 encodes a protein with two putative kinase domains that confers resistance to stem rust. New Phytologist 225, 948-959 (2020).35. Gaurav, K. et al. Population genomic analysis of Aegilops tauschii identifies targets for bread wheat improvement. Nature Biotechnology 40, 422-431 (2022).36. Arora, S. et al. A wheat kinase and immune receptor form host-specificity barriers against the blast fungus. Nat Plants (2023).37. Li, Y., Gou, M., Sun, Q. & Hua, J. Requirement of calcium binding, myristoylation, and protein-protein interaction for the Copine BON1 function in Arabidopsis. J Biol Chem 285, 29884-91 (2010).38. Whittaker, C.A. & Hynes, R.O. Distribution and evolution of von Willebrand/integrin A domains: widely dispersed domains with roles in cell adhesion and elsewhere. Mol Biol Cell 13, 3369-87 (2002). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[58, 90, 936, 128]]<|/det|> +39. Tomsig, J.L., Snyder, S.L. & Creutz, C.E. Identification of targets for calcium signaling through the copine family of proteins. Characterization of a coiled-coil copine-binding motif. J Biol Chem 278, 10048-54 (2003). + +<|ref|>text<|/ref|><|det|>[[58, 133, 936, 172]]<|/det|> +40. Zou, B. et al. Identification and analysis of copine/BONZAI proteins among evolutionarily diverse plant species. Genome 59, 565-73 (2016). + +<|ref|>text<|/ref|><|det|>[[58, 180, 933, 219]]<|/det|> +41. Pich, I.R.O. & Kondrashov, F.A. Long-term asymmetrical acceleration of protein evolution after gene duplication. Genome Biol Evol 6, 1949-55 (2014). + +<|ref|>text<|/ref|><|det|>[[58, 225, 933, 264]]<|/det|> +42. Zhao, X., Fu, X., Yin, C. & Lu, F. Wheat speciation and adaptation: perspectives from reticulate evolution. Abiotech 2, 386-402 (2021). + +<|ref|>text<|/ref|><|det|>[[58, 272, 933, 311]]<|/det|> +43. Wang, Z. et al. Dispersed emergence and protracted domestication of polyploid wheat uncovered by mosaic ancestral haploblock inference. Nat Commun 13, 3891 (2022). + +<|ref|>text<|/ref|><|det|>[[58, 317, 933, 356]]<|/det|> +44. Wang, Z.L. et al. Seedling and adult plant resistance to powdery mildew in Chinese bread wheat cultivars and lines. Plant Disease 89, 457-463 (2005). + +<|ref|>text<|/ref|><|det|>[[58, 362, 933, 401]]<|/det|> +45. Zhou, W. et al. TaNAC6s are involved in the basal and broad-spectrum resistance to powdery mildew in wheat. Plant Sci 277, 218-228 (2018). + +<|ref|>text<|/ref|><|det|>[[58, 408, 933, 447]]<|/det|> +46. Zhao, Y. et al. Wheat heat shock factor TaHsfA2d contributes to plant responses to phosphate deficiency. Plant Physiol Biochem 185, 178-187 (2022). + +<|ref|>text<|/ref|><|det|>[[58, 453, 933, 492]]<|/det|> +47. Zhao, Y. et al. The wheat MYB transcription factor TaMYB31 is involved in drought stress responses in Arabidopsis. Front Plant Sci 9, 1426 (2018). + +<|ref|>text<|/ref|><|det|>[[58, 498, 933, 538]]<|/det|> +48. Schmittgen, T.D. & Livak, K.J. Analyzing real-time PCR data by the comparative C(T) method. Nat Protoc 3, 1101-8 (2008). + +<|ref|>text<|/ref|><|det|>[[58, 544, 933, 604]]<|/det|> +49. Wang, K., Liu, H., Du, L. & Ye, X. Generation of marker-free transgenic hexaploid wheat via an Agrobacterium-mediated co-transformation strategy in commercial Chinese wheat varieties. Plant Biotechnology Journal 15, 614-623 (2017). + +<|ref|>text<|/ref|><|det|>[[58, 610, 933, 650]]<|/det|> +50. Chen, Y. et al. A collinearity-incorporating homology inference strategy for connecting emerging assemblies in the Triticeae tribe as a pilot practice in the plant pangenomic era. Mol Plant 13, 1694-1708 (2020). + +<|ref|>text<|/ref|><|det|>[[58, 656, 933, 695]]<|/det|> +51. Ma, S. et al. WheatOmics: A platform combining multiple omics data to accelerate functional genomics studies in wheat. Mol Plant 14, 1965-1968 (2021). + +<|ref|>text<|/ref|><|det|>[[58, 701, 933, 741]]<|/det|> +52. Goodstein, D.M. et al. Phytozome: a comparative platform for green plant genomics. Nucleic Acids Res 40, D1178-86 (2012). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[43, 93, 768, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 391, 150]]<|/det|> +SupplementaryInformation4.19. pdf + +<--- Page Split ---> diff --git a/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/images_list.json b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..ffab6731a3c61a7def8e9bcbb3cef870a86f43b3 --- /dev/null +++ b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/images_list.json @@ -0,0 +1,250 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1 | Non-Telecentric beam optics in 2-photon microscopy. a, Schematics of larval Drosophila (left), larval zebrafish (centre) and adult mouse (right) with central nervous system highlighted (green) to illustrate size differences. Insets next to the mouse for direct size-comparison between these species on the same scale. b, Optical configurations of standard diffraction limited (DL, left) 2P setup with parallel laser beam entering objective's back aperture. Right, non-telecentric (nTC, middle, right) configurations use a still diverging laser beam instead. As a result, the field of view and focal distance are expanded, and the point spread function (PSF) elongates. These effects scale with the angle of divergence (compare nTC₁ and nTC₂). c, Schematic representations of typical neuronal soma in species shown in (a), as interrogated by 2P setups shown in (b), respectively. d, In vivo 7 dpf larval zebrafish (HuC::GCaMP6f) imaged with an out-of-the-box Sutter-MOM DL setup at full field of view (top) and when zoomed in to reveal individual neuronal soma (bottom) as indicated. e, same zebrafish as shown in (d), as well as two further zebrafish imaged using nTC₂ configuration at maximal field of view (top). Zooming in to the same area as in (d, bottom) nonetheless reveals cellular", + "footnote": [], + "bbox": [ + [ + 120, + 75, + 884, + 720 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2 | Spatial resolution. a, Schematic (top) and photograph (bottom) of the setup used to directly film excitation volumes. b, effective scan-planes directly visualised as indicated in (a) for all optical configurations as indicated, in each case with scan-points spaced to facilitate inspection of individual PSFs. c, The same set of neurons of the 7 dpf larval zebrafish upper spinal cord (HuC:GCaMP6f, random sparse expression, see overview scan and schematic on the left) was imaged in all optical configurations as indicated at 512x512 px (1 Hz). Arrowheads highlight the same", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 884, + 825 + ] + ], + "page_idx": 10 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3 | Rapid remote focussing. a,b, Scan-profiles with the electrically tunable lens (ETL) \"flat\" (zero input current, lowest profile) and engaged to achieved axially elevated scan planes at \\(+300\\) and \\(+600 \\mu m\\) (middle, upper profiles, respectively) in nTC1 (a) and nTC2 (b) configuration, as indicated. Associated size-changes in the effective full field of view were generally \\(< 5\\%\\) (compare top and bottom planes). In each case, axial-shifts required \\(< 25\\%\\) unidirectional peak current on the ETL which in turn facilitated rapid ETL-settling times: c,d, Schematic (c) and measured (d) axial jumps and settling time: the ETL was programmed to iteratively focus up and down by \\(150 \\mu m\\) at each end of two long (5 ms) scan lines, as indicated. This enabled a direct read-out of ETL settling at each line-onset (oscillations in d). For the \\(150 \\mu m\\) jumps shown, oscillations decayed below detectability within 2-3 ms. For corresponding readouts of the ETL-position signal, see Fig. S3.", + "footnote": [], + "bbox": [ + [ + 115, + 81, + 877, + 315 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4 | Mesoscale imaging of zebrafish larvae. a, Photograph of two 9 dpf zebrafish larvae mounted head-to-head in a microscope chamber with mm-scale ruler in background. B, The same 2 fish (HuC:GCaMP6f) as in (a) imaged under 2-photon with nTC2 3.5 mm FOV configuration, at 512x128 px (3.91 Hz). c,d, Activity-correlation (cf. Fig. 2e) of the scan in (b) during presentation of", + "footnote": [], + "bbox": [ + [ + 117, + 78, + 841, + 850 + ] + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5 | 3D random access scanning of the zebrafish eye and brain. a,b, Schematic of zebrafish larva from top (a) and front (b) with scan configurations indicated. c, direct x-z visualisation of the scan-profile used in the below. d, nTC1, 1,024x1,024 px scan across an Islet2b:mGCaMP6f 6 dpf larval zebrafish eye and brain. At the centre of the scan, the axial focus is shifted upwards such that the axonal processes of retinal ganglion cells (RGCs) in the tectum (top) and their somata and dendritic processes in the eye (bottom) can be quasi-simultaneously captured. e,f, 1024x1024 px split-plane random access jump between tectum (e) and eye (f) and g-j, 2 times 64x128 px (15.6 Hz)", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 880, + 802 + ] + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6 | 2P plane-bending to image the in vivo larval zebrafish brain. a-c, Schematic of HuC:GCaMP6f larval zebrafish brain viewed from top (a) and front (b) with scan planes indicated, and (c) example-scan-profiles. d, nTC₁ 512x1024 scans of a 6 dpf zebrafish brain with different plane", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 880, + 870 + ] + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8 | Mesoscale random-access imaging of mouse cortex in vivo. a,b, Schematic of Thy1:GcAmP6f mouse brain in vivo (a) with cranial window over the somatosensory cortex (b). c,d, 1024x1024 px nTC2 (c) and nTC1 (d) images as indicated. Red arrows indicate rapid transitions between scan regions, with the inset indicating the scan-profile. e-i, 2 times 128x256 px (3.91 Hz) random access scan as indicated in (d) with mean-projection (e,f), activity-correlation (g,h, cf. Fig. 2e) and fluorescence traces (i), taken from the ROIs as indicated in (g,h). j-l, nTC1 128x128 px xyz-tilted plane (7.82 Hz) traversing through cortical layers 1-4 at \\(\\sim 45^{\\circ}\\) relative to vertical with mean image (k) and activity-correlation (l, cf. Fig. 2e).", + "footnote": [], + "bbox": [], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9 | Multi-plane imaging and optogenetics for functional circuit mapping. a,b, DL (a) and nTC₁ (b) 1,024x1,024 px scans of the ventral nerve cord of a 3rd instar VGlut:GCamP6f Drosophila", + "footnote": [], + "bbox": [ + [ + 115, + 80, + 880, + 884 + ] + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 44, + 88, + 860, + 831 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 50, + 40, + 750, + 780 + ] + ], + "page_idx": 36 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 55, + 303, + 940, + 608 + ] + ], + "page_idx": 38 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 50, + 45, + 692, + 780 + ] + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5", + "footnote": [], + "bbox": [ + [ + 55, + 115, + 760, + 852 + ] + ], + "page_idx": 40 + }, + { + "type": "image", + "img_path": "images/Figure_6.jpg", + "caption": "Figure 6", + "footnote": [], + "bbox": [ + [ + 52, + 40, + 707, + 784 + ] + ], + "page_idx": 41 + }, + { + "type": "image", + "img_path": "images/Figure_7.jpg", + "caption": "Figure 7", + "footnote": [], + "bbox": [ + [ + 55, + 111, + 550, + 850 + ] + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/Figure_8.jpg", + "caption": "Figure 8", + "footnote": [], + "bbox": [ + [ + 50, + 55, + 785, + 777 + ] + ], + "page_idx": 44 + }, + { + "type": "image", + "img_path": "images/Figure_9.jpg", + "caption": "Figure 9", + "footnote": [], + "bbox": [ + [ + 44, + 90, + 688, + 825 + ] + ], + "page_idx": 46 + } +] \ No newline at end of file diff --git a/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22.mmd b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22.mmd new file mode 100644 index 0000000000000000000000000000000000000000..432be21b8b2cd84552acc52ebd42edf3ddd196dd --- /dev/null +++ b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22.mmd @@ -0,0 +1,456 @@ + +# Non-Telecentric 2P microscopy for 3D random access mesoscale imaging at single cell resolution + +Filip Janiak ( F.K.Janiak@sussex.ac.uk) University of Sussex https://orcid.org/0000- 0002- 9295- 2740 + +Philipp Bartel University of Sussex + +Michael Bale University of Sussex https://orcid.org/0000- 0002- 5325- 1992 + +Takeshi Yoshimatsu University of Sussex https://orcid.org/0000- 0002- 4939- 2020 + +Emilia Komulainen University of Sussex + +Mingyi Zhou University of Sussex + +Kevin Staras University of Sussex https://orcid.org/0000- 0003- 4141- 339X + +Lucia Prieto- Godino The Francis Crick Institute https://orcid.org/0000- 0002- 2980- 362X + +Thomas Euler University of Tubingen https://orcid.org/0000- 0002- 4567- 6966 + +Miguel Maravall University of Sussex https://orcid.org/0000- 0002- 8869- 7206 + +Tom Baden ( T.Baden@sussex.ac.uk) University of Sussex https://orcid.org/0000- 0003- 2808- 4210 + +## Article + +Keywords: two- photon (2P) microscopy, limitations, non- telecentric (nTC) optical design, three- dimensional field, single- cell resolution, imaging neuronal activity + +Posted Date: December 15th, 2020 + +DOI: https://doi.org/10.21203/rs.3.rs- 121292/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +Version of Record: A version of this preprint was published at Nature Communications on January 27th, 2022. See the published version at https://doi.org/10.1038/s41467-022-28192-0. + +<--- Page Split ---> + +# Non-Telecentric 2P microscopy for 3D random access mesoscale imaging at single cell resolution + +Janiak FK \(^{1\S}\) , Bartel P \(^{1}\) , Bale MR \(^{1}\) , Yoshimatsu T \(^{1}\) , Komulainen E \(^{1}\) , Zhou M \(^{1}\) , Staras K \(^{1}\) , Prieto- Godino LL \(^{2}\) , Euler T \(^{3,4}\) , Maravall M \(^{1}\) , Baden T \(^{1,3\S}\) + +1: Sussex Neuroscience, School of Life Sciences, University of Sussex, UK; 2: The Francis Crick Institute, London, UK; 3: Institute of Ophthalmic Research, University of Tübingen, Germany; 4: Centre for Integrative Neuroscience, University of Tübingen, Germany. + +\(\) \\(\) Correspondence at f.k.janiak@sussex.ac.uk and t.baden@sussex.ac.uk + +8 Figures, 4 Supplementary Figures + +7 Supplementary Videos + +1 User Manual (available at https://github.com/BadenLab/nTCscope) + +Summary. In neuroscience, diffraction limited two- photon (2P) microscopy is a cornerstone technique that permits minimally invasive optical monitoring of neuronal activity. However, most conventional 2P microscopes impose significant constraints on the size of the imaging field- of- view and the specific shape of the effective excitation volume, thus limiting the scope of biological questions that can be addressed and the information obtainable. Here, employing a non- telecentric (nTC) optical design, we present an ultra- low- cost, easily implemented and flexible solution to address these limitations, offering a several- fold expanded three- dimensional field of view that also maintains single- cell resolution. We show that this implementation also allows for straight- forward tailoring of the point- spread- function, increases effective excitation power, and achievable image brightness. Moreover, rapid laser- focus control via an electrically tunable lens allows near- simultaneous imaging of remote regions separated in three dimensions and permits the bending of imaging planes to follow natural curvatures in biological structures. Crucially, our core design is readily implemented (and reversed) within a matter of hours, and compatible with a wide range of existing 2P customizations, making it highly suitable as a base platform for further development. We demonstrate the application of our system for imaging neuronal activity in a variety of examples in zebrafish, mice and fruit flies. + +Author contributions, FKJ and TB designed the study, with inputs from TE and all authors; FKJ implemented and tested hardware and software modifications, with input from PB, TY, TB and TE. FKJ and TB analysed the data, with inputs from all authors. PB assisted with hardware and software testing and troubleshooting and built the visual stimulator. MRB and MM provided mice for in vivo imaging and assisted with their handling and imaging. TY and MZ generated Islet2b:mGCaMP line and assisted with zebrafish sample preparation and testing. EK and KS provided mouse brain acute slice samples and assisted with handling and imaging. LLPG provided Drosophila sample and assisted with handing and imaging. TB built the optogenetics stimulator. FKJ and TB wrote the manuscript with inputs from all authors. + +Acknowledgements. We thank Sabi Abdul- Raouf Issa for providing the VGlut:GCaMP6f Drosophila sample, and John Bear for helping with the generation of the Islet2b:mGCaMP6f line. The authors would also like to acknowledge support from the FENS- Kavli Network of Excellence and the EMBO YIP. + +Funding. Funding was provided by the European Research Council (ERC- StG "NeuroVisEco" 677687 to TB, ERC- StG "EvolutioNeuroCircuit" 802531 to LLPG), The UKRI (BBSRC, BB/R014817/1 to TB, BB/S00310X/1 to KS, and MRC, MC_PC_15071 to TB and MM, MR/P006639/1 to MM and MR/P010121/1 to KS), the Leverhulme Trust (PLP- 2017- 005 to TB), the Lister Institute for Preventive Medicine (to TB), the Marie Curie Sklodowska Actions individual fellowship ("ColourFish" 748716 to TY) from the European Union's Horizon 2020 research and innovation programme, and the German Research Foundation (DFG) through Collaborative Research Center CRC 1233 (project number 276693517, to TE). LLPG's research was supported by the The Francis Crick Institute. + +<--- Page Split ---> + +## INTRODUCTION + +Laser scanning two photon (2P) microscopy allows the imaging of live cellular processes deep inside intact tissue with high signal- to- noise, temporal fidelity and spatial resolution1. Nonetheless, standard diffraction- limited 2P setups with a collimated laser excitation beam have several key characteristics that constrain their broad applicability; namely, a typically small field of view (FOV), a fixed- size excitation spot and restricted options for rapid random access 3- dimensional scans. These are significant limitations because the biological samples that are interrogated with 2P microscopy can exhibit substantial variations in size and spatial structure. For example, the volume of an adult mouse brain is approximately four orders of magnitude larger than that of a larval zebrafish, and seven orders of magnitude larger than a first instar larval fruit fly (Fig. 1a). Similarly, neuronal sub- structures are also highly variable in density and size, ranging from sub- micron levels for some synapses up to \(20 \mu \mathrm{m}\) or more for some somata. Additionally, neural densities vary by more than an order of magnitude across different animal brains2. As such, 2P microscopy tends to reveal very different levels of detail and organization across its diverse experimental applications. To maximize biological information, upgrades for 2P microscopy should enable the imaging of neuronal activity from many neural structures of a given size and density across a sufficiently large 3D volume of tissue at sufficiently high frame rates for the chosen neuronal process and biosensor. + +In response to this demand, a profusion of custom modifications to 2P microscopes have been developed to expand the spatial and temporal boundaries over which neural structures can be optically interrogated. For example, the maximal planar field of view (FOV) has been increased from typically \(0.5 \mathrm{mm}\) to between \(3.1 - 10 \mathrm{mm}\) diameter by the exchange ( \(3 \mathrm{mm}\) : Ref3, \(7 \mathrm{mm}\) : Ref4) or size- increase of optical components ( \(10 \mathrm{mm}\) ), custom built objectives ( \(3.1 \mathrm{mm}\) ), enhanced scan engines ( \(5 \mathrm{mm}\) )7 and a mesoscope configuration ( \(5 \mathrm{mm}\) )8 to allow 'mesoscale' interrogation of neural circuits. In parallel, customizations using multiple beams have allowed simultaneous scanning of distant brain regions6,9,10. Likewise, higher temporal resolutions have been achieved by tailoring the point spread function (PSF) to the geometry and distribution of the neuronal structures of interest, thus increasing signal- to- noise ratio (SNR) and, in turn, decreasing the minimally- required dwell time per pixel11,12. Moreover, the imaging plane has been axially expanded by engineering an excitation spot with Bessel focus13,14 or by elongated Gaussian foci stereoscopy15. These customizations provide efficient ways to merge image structures that are located at different depths into a single volumetric plane. Furthermore, in recent years, systems integrating acousto- optic deflectors16,17, electrical tunable lenses9,18- 21 and remote focusing units8,14 have enabled quasi- simultaneous multiplane volumetric scans. + +These types of extensions have been essential in driving the field forward, yet many are expensive, require custom- produced optical elements, complex optical alignment and/or introduce new limitations. The latter can include limitations in both excitation (e.g. power loss8, wavefront + +<--- Page Split ---> + +dispersion \(^{17}\) ) and collection \(^{4,5}\) . Here we introduce an alternative design for 2P microscopy that overcomes many of these limitations while simultaneously approaching the capabilities of a wide range of state- of- the- art performance customisations, while being ultra- low- cost, simple and flexible. + +Our non- telecentric (nTC) design, implemented for \(\sim \xi 1,000\) on an existing 2P setup equipped with a standard 20X objective, allows the expansion of the planar FOV from typically \(\sim 0.5 \mathrm{mm}\) in diameter to anywhere up to 3.5 mm to flexibly suit experimental needs (Fig. 1). This expansion is accompanied by a moderate and adjustable ("PSF tailoring" \(^{12}\) ) increase in the system's 3D PSF while maintaining single cell resolution over a wide range of biological applications. For example, unlike a standard diffraction limited (DL) setup (left in Fig. 1b,c), our nTC setup (right in Fig. 1b,c) allows simultaneous imaging of three entire zebrafish brains (Fig. 1e), or about a third of the width of a mouse's brain, while in each case maintaining single cell resolution (Fig. 1f, Supplementary Video S1). The addition of an electrically tunable lens (ETL) then allows near- simultaneous sampling in distant brain regions separated in 3 dimensions. Crucially, our solution is both comparatively low- cost and easy to implement on any existing 2P setup without the need for complex optical calibration, thus facilitating its widespread adoption in the community. We anticipate that others will be able to build on our core optical design using existing and new modifications to further increase its capability in the future. We demonstrate the current performance of our system with a range of examples from zebrafish, mice and fruit flies. + +## RESULTS + +Non- Telecentric optics for field of view expansion. In traditional laser scanning 2P microscopy (left in Fig. 1b,c,g), a diffraction limited (DL) PSF is generated to excite fluorophores in a typically sub- micron volume of tissue. Here, xy- scanning mirrors reflect the laser beam into a collimation system comprised of a scan and a tube lens. The collimated beam then enters the back aperture of a high numerical aperture (N.A.) objective \(^{22,23}\) to converge at parallel rays into a DL spot at focal distance \(^{24}\) . The Gaussian shape of the excitation beam dictates that it is not possible to perfectly match beam width to the objective's back aperture. Instead, the back aperture is typically overfilled with a factor of \(1 / \mathrm{e}^2\) as a compromise between maximising spatial resolution (i.e. small PSF size) and power transmission \(^{25}\) . + +In contrast, our nTC design (middle and right in Fig. 1b,c,g) illuminates the objective's back- aperture with a decollimated and divergent beam. This leads to an increased angle of view as the light exits the objective's front aperture, such that the same angular scan- mirror movement leads to a larger absolute shift in the image plane – thereby greatly increasing the FOV. In parallel, this also alters the effective excitation numerical aperture (N.A.) to yield a larger- than- DL excitation spot (i.e. an elongated PSF) at greater focal distance. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1 | Non-Telecentric beam optics in 2-photon microscopy. a, Schematics of larval Drosophila (left), larval zebrafish (centre) and adult mouse (right) with central nervous system highlighted (green) to illustrate size differences. Insets next to the mouse for direct size-comparison between these species on the same scale. b, Optical configurations of standard diffraction limited (DL, left) 2P setup with parallel laser beam entering objective's back aperture. Right, non-telecentric (nTC, middle, right) configurations use a still diverging laser beam instead. As a result, the field of view and focal distance are expanded, and the point spread function (PSF) elongates. These effects scale with the angle of divergence (compare nTC₁ and nTC₂). c, Schematic representations of typical neuronal soma in species shown in (a), as interrogated by 2P setups shown in (b), respectively. d, In vivo 7 dpf larval zebrafish (HuC::GCaMP6f) imaged with an out-of-the-box Sutter-MOM DL setup at full field of view (top) and when zoomed in to reveal individual neuronal soma (bottom) as indicated. e, same zebrafish as shown in (d), as well as two further zebrafish imaged using nTC₂ configuration at maximal field of view (top). Zooming in to the same area as in (d, bottom) nonetheless reveals cellular
+ +<--- Page Split ---> + +detail (e, bottom). f, In vivo adult mouse cranial window over somatosensory cortex imaged with nTC2 maximal field of view (top) and when zoomed in as indicated (bottom). g, Left: Optical configuration of a standard DL setup with collimation system consisting of a scan lens and a tube lens to set- up an infinity collimated laser beam at the level of the objective's back aperture. Effective refractive power and relative distances of lenses indicated. The intermediary focal point (IFP) is immediately behind the scan lens (arrowhead). Middle: nTC1 configuration replaces the scan lens with a pair of plano- convex lenses (L1,2). The relative position of L2 to the tube lens defines the position of the new IFP, which is now further along the laser path. As a result, the field of view can be expanded to between 1.2 and 1.8 mm. Right: nTC2 configuration using a single plano- convex lens (L3) allows FOV expansion to 2.5 - 3.5 mm. h, complete nTC setup, including also an ETL positioned in front of the scan mirrors for rapid axial- scanning. PMTs, Photomultipliers. i, FOV expansion under nTC combines two effects: Increased focal distance (left) and reduced numerical aperture (N.A., right), which together give rise to a larger effective focal plane and enlarged PSF. j, Power at sample measured for all configurations, expressed as a percentage of the power that reaches the scanning mirrors. m, point spread functions (PSFs) measured for all optical configurations, with size of typical neuronal somata of different species indicated. All scale- bars 10 μm. n.o, lateral (n) and axial (o) spread of the PSFs quantified. Errors in s.d.. The specified numbers are for a Sutter MOM 2P microscope with Zeiss Objective W "Plan- Apochromat" 20x/1.0. Power at the sample plane was 0.35 mW. + +The magnitudes of each of these effects scale with the angle of divergence as the beam enters the back aperture of the objective. Accordingly, simply shifting the plano- convex lenses up or down the laser path, or switching between different refractive power lenses, provides for easy control over the system's optical properties to flexibly suit the user's needs. + +In the following we show that the use of nTC in 2P microscopy brings about important advantages over the traditional, collimated and diffraction limited (DL) design: + +1. The total field of view (FOV) can be expanded several-fold to suit the user's needs. + +2. Scan-mirror movements translate into correspondingly larger xy-shifts in the image plane, meaning that even multi-millimetre random access jumps can be achieved with millisecond precision. + +3. The addition of an electrically tunable lens (ETL) in front of the scan mirrors allows for similarly extensive expansion in the axial dimension. + +4. The simplified optical path and under-filling of the objective's back aperture means more laser power is available at the sample plane. + +5. It allows flexible and partially FOV-independent PSF-shape adjustment for imaging neurons of different size to individually optimise detection sensitivity for different biological samples2,11. + +6. The increased working distance provides additional space for access to the preparation, for example with electrodes or stimulation equipment. + +<--- Page Split ---> + +7. Combining points 4-6, nTC can yield an overall signal boost of up to \(\sim 18\) -fold with an above-stage detector (or \(\sim 4 - 5\) -fold at equal laser power on the sample), and more if a substage-collector is added. + +We first discuss the required optical modifications and their impact on key excitation parameters (Figs. 1- 3, Figs. S1,2), before presenting a series of key use cases of different configurations for the interrogation of neural structure and function across diverse models (Figs. 4- 9, Figs. S3- 5). + +## PART I. OPTICS, EXCITATION AND OPTICAL SAMPLING + +A simple scan- lens modification yields up to 7- fold FOV expansion. An off- the- shelf infinity- corrected galvo- galvo Sutter- MOM setup equipped with a 20x objective (Zeiss Objective W "Plan- Apochromat" 20x/1.0) offers a FOV diameter of \(\sim 0.5 \mathrm{mm}\) (left in Fig. 1g). However, when underfilling the back aperture of the objective with a diverging laser (middle and right in Fig. 1g), the beam exits the objective front aperture at increasingly obtuse angles at an effectively decreased N.A. (Fig 1i, Fig. S1a) and comes into focus at a greater distance (Fig. 1g, Fig. S1b). Together, this expands the effective excitation FOV in both xy (increased angle and decreased N.A.) and z (elongated PSF). To achieve this effect, it is necessary to bring the collimated laser beam, having passed the scan mirrors, to an "early" intermediary focal point (IFP) prior to reaching the objective, thus setting up the diverging beam thereafter (Fig. 1g, arrowheads). The specific divergence angle as the beam enters the back- aperture of the objective, which depends on IFP, defines the magnitude of the above- mentioned effects. We present two simple optical solutions (nTC1 and nTC2) to set- up an early IFP and thus expand the effective FOV to varying degrees. + +In the standard DL configuration, the scan- lens (SL) and tube lens (TL) are separated from each other at a distance that is equal to their combined focal lengths \((50_{\mathrm{SL}} + 200_{\mathrm{TL}} \mathrm{mm} = 250 \mathrm{mm})\) to collimate the beam (left, Fig. 1g). In nTC1, we removed SL and instead inserted two off- the- shelf plano- convex lenses (L1, modified VISIR 1534SPR136, Leica; L2, LA1229 Thorlabs) with focal lengths 190 and 175 mm, respectively (middle, Fig. 1g, Methods). L1 was fixed 190 mm in front of TL to set up an IFP exactly at the TL. Next, L2 was positioned between L1 and TL to further increase laser convergence and thus shift the exact position of IFP away from the TL. Accordingly, IFP is always in front of the TL, with L2 determining its exact position: Simply shifting L2 along the laser path between 100 and 5 mm distance from the TL expanded the effective FOV diameter to anywhere between 1.2 and 1.8 mm, respectively (compare Fig. 1g, middle). In nTC2 (Fig. 1g, right), we replaced SL with a single lens (L3) of 200 mm focal length (LA1708, Thorlabs). L3 operated in much the same way as L2 in the previous modification M1, however now the IFP was behind rather than in front of TL. Depending on the position of L3, this yielded effective FOV diameters anywhere between 2.5 and 3.5 mm. Importantly, in each case effective image brightness remained approximately constant across the full FOV (Fig. S1c- f, Methods). Here, + +<--- Page Split ---> + +the marginal brightness increase towards the edges is related to the slight upwards bend in the imaging plane as commonly seen for large FOV 2P microscopes \(^{5,8}\) – see also below. The axial difference between the edge and centre of the imaging plane was 20, 45, 87 and 170 μm for 1.2, 1.8, 2.5, 3.5 mm FOV, respectively. + +While these specific lens configurations readily work for the commercially available Sutter MOM, the fundamental concept of setting up an IFP to yield a diverging beam is readily applicable to any standard 2P microscope, provided the optical path between the scanning mirrors and tube lens is accessible. In fact, 2P scan- lenses tend to consist of multiple custom- designed optical elements which by themselves easily exceed the cost of our solution. Accordingly, if provided directly by the microscope's manufacturer, our simplified optics should decrease the cost of such an off- the- shelf system. + +Our design's full optical path and control logic are shown in Fig. 1h. All functions are executed from the scan software, which directly controls the xy- scan path as usual. To synchronize an electrically tunable lens (ETL, see below) and/or a Pockels cell to this xy- scan, a copy of the fast- mirror command is sent to two microcontrollers. Each of these then executes preloaded line- synchronized commands that are defined using a standalone graphical user interface (GUI). In this way, this standalone z- control- system only requires a copy of the scan mirror command, meaning that it can be directly added to any 2P microscope setup without the need for software modifications. + +Increased effective laser power. Because our nTC design avoids overfilling of the objective's back aperture and uses fewer optical elements in the laser path, total laser power at the sample was increased approximately 4- fold compared to all configurations of the DL setup (Fig. 1j). This additional power could, for example, be used to facilitate imaging deep in the brain, or alternatively to drive additional setups from the same laser source. For instance, when imaging the small brains of larval zebrafish or fruit flies, there is rarely a need to exceed 50 mW, meaning that it is theoretically possible to drive ten such nTC setups from a single standard laser (e.g. Coherent Chameleon Vision- S Laser, average power \(\sim 1.5\) W at 930- 960 nm, assuming 50% loss through the setup). + +Spatial resolution under nTC. To establish how our nTC approach affected the excitation PSF, we first imaged 175 nm fluorescent beads across all configurations at 927 nm wavelength and constant laser power at the sample (Methods, Supplementary discussion). Starting from a DL spot- volume of 0.56 and 3.15 μm (xy and z, respectively), our different modifications elongated and laterally expanded the PSF to varying degrees, from 0.77 (xy) and 9.94 (z) μm for the 1.2 mm FOV configuration to 2.21 (xy) and 41.49 (z) μm at 3.5 mm FOV (Fig. 1m- o, cf. Fig. S1g- i). Accordingly, increasing the FOV using nTC mainly elongated the PSF, while restricting its lateral expansion to remain principally suitable for providing single cell resolution even for the largest 3.5 mm expansion. + +<--- Page Split ---> + +However, PSF expansions were generally stronger than in other large FOV 2P- approaches which, for example, reported \(\sim 15 \mu \mathrm{m}\) at the edge of a 10 mm FOV \(^{5}\) or \(< 10 \mu \mathrm{m}\) at the edge of a \(\sim 5 \mathrm{mm}\) FOV \(^{8}\) , see also \(^{5,6}\) . These approaches achieve their optical results through custom made, large- diameter optics, which are generally more expensive and more difficult to retrofit to existing setups. Notably, beyond the microscope's optics itself, PSF- dimensions are affected by a myriad of additional factors such as laser wavelength and power (SFig. 1g- i) as well as the specific measurement method (e.g. bead types). Accordingly, directly comparing their dimensions across studies remains difficult. Notwithstanding, the possibility of optically merging adjacent image structures strongly depends on the size and spatial distribution of labelled biological structures - a general limitation in optical microscopy, rather than a specific limitation to our nTC approach (discussed e.g. in Ref \(^{26}\) ). + +To further assess how the different optical configurations impacted PSF- shapes across the whole FOV, we next visualised excitation volumes using a camera (Fig. 2a,b) \(^{27}\) . Specifically, we positioned a fluorescein- solution- filled cuvette below the objective and filmed it from the side (Fig. 2a). Compared to imaging beads (Fig. 1m- o) this approach had the advantage that excitation volumes could be visualised much more directly, as well as across different positions in space in rapid succession (Supplementary Video 2). Fig. 2b shows a direct, scale- matched visualisation of effective scan profiles for all optical configurations. This confirmed that the DL configuration had the smallest PSFs, followed by increasing- FOV variations of nTC \(_{1,2}\) . Moreover, scan- profiles were curved to different degrees, with correspondingly tilted PSFs towards the edge \(^{5,8}\) . If required, this can be part- corrected via the ETL. However, biological structures are rarely perfectly flat either. As described further below, often a more useful solution might be to instead fit the scan- plane curvature to the 3D curvature of the interrogated sample. + +Next, we directly compared the resultant effective spatial resolutions by imaging the same sample in each configuration. For this, we sparsely expressed GCaMP6f under the pan- neuronal promotor HuC in larval zebrafish \(^{28}\) and imaged one animal that randomly exhibited sparse and easily recognisable expression in neurons of the upper spinal cord, including one cell body ( \(\sim 7 \mu \mathrm{m}\) diameter) and several individual synapses ( \(\sim 1 \mu \mathrm{m}\) diameter, arrowheads indicate matching position of 2 such synapses across scans) (Fig. 2c). These image structures were consistently recognisable across all optical configurations, demonstrating that that even with nTC \(_{2}\) , single cell resolution could be readily achieved. This notion was further confirmed in functional scans during visual stimulation that was time- interweaved with the scanner retrace to avoid crosstalk (Fig. 2d- e, Methods). For example, full- field UV- flashes elicited different responses in different image structures on a trial- to- trial basis (e.g. see highlighted traces of nTC \(_{2}\) condition). Taken together, single cell resolution was readily preserved across all optical configurations. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2 | Spatial resolution. a, Schematic (top) and photograph (bottom) of the setup used to directly film excitation volumes. b, effective scan-planes directly visualised as indicated in (a) for all optical configurations as indicated, in each case with scan-points spaced to facilitate inspection of individual PSFs. c, The same set of neurons of the 7 dpf larval zebrafish upper spinal cord (HuC:GCaMP6f, random sparse expression, see overview scan and schematic on the left) was imaged in all optical configurations as indicated at 512x512 px (1 Hz). Arrowheads highlight the same
+ +<--- Page Split ---> + +synaptic structures in each scan. d- f, 64x64 px (7.81 Hz) activity scan from fields of view shown in (c) for all five configurations during presentation of full- field flashes of UV- light which stochastic elicited activity in these imaged neural structures. In each case the average scan projection (d) and neighbour- correlation based activity projection (e) are shown (hereafter referred to as "activity- correlation"). Darker shadings, equalised for visibility, denotes increased local activity (for details, see Ref). Black traces in (f) show time- traces for the same structure in all cases. For the nTC2 2.5 mm FOV condition, time- traces from different neural structures are extracted to illustrate different responses in different structures. All activity traces in this and the following figures are shown in z- scores relative to their own baseline. We choose this metric over dF/F as it emphasises detectability of events rather than the relative change from the indicator's baseline fluorescence, which differs between biosensors. + +PSF- tailoring. The systematic effects on PSF shape across configurations also meant that our nTC approach could be used to flexibly tailor PSF dimensions to specific experimental needs. This can be achieved by varying the degree of underfilling of the objective's back aperture while simultaneously keeping the laser's divergence angle approximately constant. We demonstrate this principal capability by setting up a "high- resolution" (small PSF) variant of nTC1 (Fig. S2). In general, such PSF- tailoring is useful for balancing the spatial resolution with the SNR. For example, the sub- micron DL PSF offered by typical collimated 2P- setups maximises spatial resolution which is invaluable for resolving small synaptic processes or the somata of larval fruit flies (typically <5 μm). However, many species' cell bodies are much larger. For example, in the brain of larval zebrafish a very small DL PSF spatially typically oversamples the "mid- sized" ~5- 10 μm somata at the expense of a potentially substantial loss in SNR. This limitation can be avoided by nTC- mediated tailoring of the PSF (cf. Fig. 2). Similarly, for picking up somatic signals from cortical neurons in the mouse, a "10- fold expanded" ~5 μm PSF yields the best SNR12. + +An increased image brightness. Beside shaping the FOV and PSF dimensions, our nTC approach also generally boosted image brightness and thus signal- to- noise ratio (e.g. Fig. 2c- f). Here, three main factors contribute: (i) total effective laser power at the sample plane, (ii) the spatial relationship of PSF shape to structure(s) in the sample, and (iii) the fraction of emitted photons that can be detected (i.e. collection). As discussed above, the ~4- fold increased effective laser power (Fig. 1j) and larger excitation volumes under nTC (Figs. 1m- o, 2a- c) generally served to boost the number of emitted photons available for collection in the first place. In contrast, the increased working distance of nTC (Fig. S1b) meant that correspondingly smaller fractions of these photons could collected by the above- stage objective used for excitation. Notwithstanding, signal- collection from below the sample, being independent of above- stage working distance, is approximately unaffected. To explore the balance between all these factors, we imaged a piece of fluorescein- soaked tissue paper under all optical configurations, in each case finding the same field of view, and used two independent detector systems: one above the stage, and another below (Fig. S2g- i). In this way, the sub- stage signal in isolation could be used to measure brightness approximately independent of + +<--- Page Split ---> + +excitation working distance (Fig. S2i, blue), while comparison of this sub- stage reading with the corresponding above- stage reading could then be used to estimate the relative above- stage collection loss (Fig. S2i, green vs. blue.). This revealed that the relative signal loss above the stage was very small indeed for nTC1 (1.2 mm: \(\sim 3\%\) ; 1.8 mm: \(\sim 9\%\) ), but started to noticeably affect collection under nTC2 (nTC2 2.5mm: \(33\%\) , nTC2 3.5 mm: \(64\%\) ). However, in all cases, this signal loss was greatly outweighed by the increased overall availability of photons for collection in the first place (relative to DL: \(\sim 10 - 20\) - fold for nTC1 and 16- 18 for nTC2 and 2). Taking all effects together, the highest overall signal- boost if using only an above- stage detector was achieved under nTC1 1.8 mm configuration ( \(\sim 18\) - fold), followed by nTC2 2.5 ( \(\sim 10\) - fold), nTC1 1.2 mm ( \(\sim 9\) - fold) and finally nTC2 3.5 mm ( \(\sim 6\) fold). Correspondingly larger signal boosts incur if a substage- detector is added. Accordingly, even if keeping effective laser power on the sample constant by correspondingly tuning down laser power as it enters the microscope (factor \(\sim 4\) , cf. Fig. 1j), all nTC configurations serve to boost effective image brightness relative to a DL configuration. + +Rapid axial scans. In addition to expanding the FOV, our de- collimated design also shifts the excitation point beyond the objective's nominal focal distance (Fig. S1b). The same optical effect can be exploited to drive rapid axial shifts in the excitation plane by introduction of an electrically tunable lens (ETL) early in the laser path (Fig. 3, Fig. S3, cf. Fig. 1h) \(^{18,21}\) . Specifically, we positioned an off- the- shelf ETL (EL- 16- 40- TC- 20D, Optotune) 200 mm in front of the first scan mirror and controlled it with a custom driver board (see user manual). In this position, already a minor deviation from the perfectly flat curvature at zero input current slightly converged the laser which, in turn, strongly shifted the effective z- focus below the objective. For example, in both nTC1 and nTC2, stepping the input current from zero to \(25\%\) (50 mA) gave rise to a \(\sim 600 \mu \mathrm{m}\) z- shift of the excitation plane (Fig. 3a,b). The use of only a small fraction of the ETL's full dynamic range enabled short turnaround times (1- 10 ms, depending on distance jumped, Fig. 3c,d, SFig. 3) and prevented overheating \(^{18,29}\) . If required, rapid synchronization of the ETL curvature with a Pockels cell for controlling effective laser power at the sample plane can compensate for any systematic variations in image brightness associated with increased penetration depth. + +Taken together, our design therefore presents a low- cost ( \(\sim \mathbb{E}1,000\) , cf. user manual) and easily implemented solution to expand the FOV of any 2P microscope in three dimensions while maintaining image quality suitable for single cell resolution. In the following, we demonstrate how these capabilities can be exploited in a range of neurophysiological applications in larval zebrafish, as well as the mouse cortex and fruit fly brain. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3 | Rapid remote focussing. a,b, Scan-profiles with the electrically tunable lens (ETL) "flat" (zero input current, lowest profile) and engaged to achieved axially elevated scan planes at \(+300\) and \(+600 \mu m\) (middle, upper profiles, respectively) in nTC1 (a) and nTC2 (b) configuration, as indicated. Associated size-changes in the effective full field of view were generally \(< 5\%\) (compare top and bottom planes). In each case, axial-shifts required \(< 25\%\) unidirectional peak current on the ETL which in turn facilitated rapid ETL-settling times: c,d, Schematic (c) and measured (d) axial jumps and settling time: the ETL was programmed to iteratively focus up and down by \(150 \mu m\) at each end of two long (5 ms) scan lines, as indicated. This enabled a direct read-out of ETL settling at each line-onset (oscillations in d). For the \(150 \mu m\) jumps shown, oscillations decayed below detectability within 2-3 ms. For corresponding readouts of the ETL-position signal, see Fig. S3.
+ +## Part II. IMAGING THE STRUCTURE AND FUNCTION OF NEURONS + +Imaging zebrafish under 2P. Owing to their small size and transparent larval stage, zebrafish have become a valuable model for interrogating brain- wide neural circuit function \(^{30,31}\) . However, from tip to tail, the brain and spinal cord of a 7- 9 dpf larval zebrafish reaches about 3.5 - 4.5 mm, with the central brain occupying approximately 1.2 mm in length and 0.7 mm in width. This is too large to fit into the FOV of a typical DL 2P setup. As a consequence, studies routinely "tile- scan" the brain in sequential stages to provide brain- wide data \(^{32,33}\) . + +On the other hand, the transparent body wall of larval zebrafish makes them well- suited for 1- photon selective- plane- illumination microscopy (1p- SPIM / "lightsheet microscopy"), which is not FOV- limited in the same way as 2P microscopy \(^{29,30,34}\) . However, 1p- SPIM and related techniques \(^{35}\) have a number of drawbacks, including constraints on achieving a homogenous image due to scattering and divergence of the excitation light with increasing lateral depth \(^{2}\) , limited access to tissues that are shadowed by strongly- scattering tissue such as the eyes \(^{36,37}\) and, critically, a direct and bidirectional interference between the imaging system itself and any light stimuli applied for studying zebrafish vision \(^{38}\) . These specific challenges could be readily addressed by our nTC 2P setup. To demonstrate this, we imaged a multiple larval zebrafish in a range of optical configurations. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4 | Mesoscale imaging of zebrafish larvae. a, Photograph of two 9 dpf zebrafish larvae mounted head-to-head in a microscope chamber with mm-scale ruler in background. B, The same 2 fish (HuC:GCaMP6f) as in (a) imaged under 2-photon with nTC2 3.5 mm FOV configuration, at 512x128 px (3.91 Hz). c,d, Activity-correlation (cf. Fig. 2e) of the scan in (b) during presentation of
+ +<--- Page Split ---> + +full- field flashes of UV- light, with hand- selected exemplary ROIs, extracted time- traces (d) and light- stimulus- aligned averages (e). f- i, the same fish as shown on the left in (b, fish 1), now shown at full 3.5 mm field of view (f, 512x128 px, 3.91 Hz) and increased spatial resolution scans of regions as indicated to reveal cellular detail (g- l, 1,024x1,024 px, 0.49 Hz). + +Mesoscale whole- zebrafish 2P imaging. First, we used the 3.5 mm configuration of nTC2 to capture the largest- possible FOV of two larval zebrafish facing each other. This configuration comfortably allowed simultaneous mesoscale imaging of two entire zebrafish brains, here responding to full- field flashes of UV- light (Fig. 4a- e, Supplementary Video 3). Alternatively, the same configuration could be used to capture the entire central nervous system of one fish in a single frame, including the brain and nearly up to the tip of the spinal cord (Fig. 2f). Zooming in throughout the sample enabled resolving cellular details (Fig. 2g- i). + +Next, beyond mesoscale imaging, many studies of zebrafish neuronal function focus on either the brain or the spinal cord (rather than both). In this case, using the more highly resolved nTC1 configuration with 1.2 mm FOV may be preferable; this just about fits one full zebrafish brain at a time while comfortably providing single cell resolution, as demonstrated below. + +3D random access scanning across the zebrafish eye and brain. In the nervous system, key functionally linked circuits are often separated in 3D space, representing a general problem for systems neuroscience. For example, the retinal ganglion cells of the zebrafish eye project to the contralateral tectum and pretectum, which are both axially and laterally displaced by several 100s of microns. Accordingly, it has been difficult to simultaneously record at both sites, for example to study how the output of the eye is linked to the visual input to the brain. To address this problem, we used our nTC1 configuration in synchronisation with the ETL to establish quasi- simultaneous 3D random access scanning of the zebrafish's retinal ganglion cells across both the eye and brain (Fig. 5a- c). For this we used an Islet2b:mGCaMP6f line which labels the majority of retinal ganglion cells in larval zebrafish. We first defined a slow, high- spatial resolution scan (512x512 px, 0.98 Hz) that captured the entire front of the head, however with a single z- jump at the centre of the frame to set- up a "staircase- shaped" scan- path (Fig. 5b,c). Here, empirical adjustment of the magnitude of the z- jump allowed us to identify the axonal processes of retinal ganglion cells in the brain, and their dendritic processes in the contralateral eye in the top and bottom of the same imaging frame, respectively. Based on this image, we next defined two scan regions for 3D random access scanning, one capturing a single plane across the tectum, while the other captured a smaller area of a subset of RGC dendrites and somata in the eye (Fig. 5d- f). Finally, we decreased the spatial resolution to 64x64 px to quasi- simultaneously image both regions at 7.81 Hz. This configuration allowed reliable recording of light- driven signals from individual RGC neurites across the eye and brain (Fig. 5g- j). Next, we repeated this experiment, however this time in zebrafish larvae that were transiently injected with Islet2b:mGCaMP6f plasmid. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5 | 3D random access scanning of the zebrafish eye and brain. a,b, Schematic of zebrafish larva from top (a) and front (b) with scan configurations indicated. c, direct x-z visualisation of the scan-profile used in the below. d, nTC1, 1,024x1,024 px scan across an Islet2b:mGCaMP6f 6 dpf larval zebrafish eye and brain. At the centre of the scan, the axial focus is shifted upwards such that the axonal processes of retinal ganglion cells (RGCs) in the tectum (top) and their somata and dendritic processes in the eye (bottom) can be quasi-simultaneously captured. e,f, 1024x1024 px split-plane random access jump between tectum (e) and eye (f) and g-j, 2 times 64x128 px (15.6 Hz)
+ +<--- Page Split ---> + +random access scan of the same scan regions with raw (g) and event- averaged (h) fluorescence traces, mean image (i) and activity- correlation (j, cf. Fig. 2e). The stimulus was a series of full- field broadband flashes of light as indicated. k- o, as (d- j), with individual RGCs transiently expressing GCaMP6f under the same promoter. + +These animals stochastically express mGCaMP6f in only a very small number of RGCs, making it possible in principle to identify the processes belonging to the same RGC in both the eye and brain. As a proof of principle, we present one such experiment where we could clearly image the processes of single RGCs at both sites (Fig. 5k- o). For this type of application, it will be important to optimise the genetic protocol to improve expression levels and thereby facilitate the identification of the same RGC's processes at both sites. + +## 3D plane-bending for imaging activity across the zebrafish brain. + +During standard planar scans of the larval zebrafish brain, the powerful optical sectioning afforded by the 2P approach highlights the 3D curvature of distinct brain regions by cutting right across them (Fig. 6a- d). While it was possible to quasi- simultaneously image anywhere within the brain at high spatial resolution using nTC1, a planar scan grossly misrepresented the real 3D structure of the zebrafish brain (Fig. 6d, top panel). For example, the tectum in larval zebrafish is tilted upwards \(\sim 30^{\circ}\) , meaning that rather than either cleanly sampling across its retinotopically organized surface, or perpendicularly across its stacked functional layers, the planar image instead cut the tectum at an effective \(30^{\circ}\) angle to yield a mixture of both, thus confounding interpretation. To ameliorate these issues, we used a 3D curved scan plane by driving the ETL as a sqrt(cosine) function of the slow y- mirror command (Methods). This enabled z- curvature "halfpipe" scans that could be empirically fitted to follow the natural curvature of the brain, thereby closely capturing the functional anatomical organisation of the zebrafish brain (Fig. 6b- d, Supplementary Video S4). From here, we chose a single halfpipe plane that best followed the curvature of the two tecta and imaged this plane at 7.81 Hz (256x128 px, 1ms/line, Fig. 6e). We then presented spectrally broad full- field light stimulation. This allowed us to interrogate brain- wide visual function in response to arbitrary wavelength light (Fig. 6f- h). As required, the halfpipe scans could also be staggered for multiplane imaging at correspondingly lower image rates, including negative bends that surveyed the difficult- to- reach bottom of the brain between the eyes (Fig. S4, Supplementary Video S5). + +<--- Page Split ---> +![](images/Figure_6.jpg) + + + +
Figure 6 | 2P plane-bending to image the in vivo larval zebrafish brain. a-c, Schematic of HuC:GCaMP6f larval zebrafish brain viewed from top (a) and front (b) with scan planes indicated, and (c) example-scan-profiles. d, nTC₁ 512x1024 scans of a 6 dpf zebrafish brain with different plane
+ +<--- Page Split ---> + +curvatures, with peak axial displacement at scan centre as indicated. At curvatures \(\sim 100 - 150 \mu m\) peak displacement the scan approximately traverses the surface of the tectum. e- h, mean (e), activity- correlation (f, cf. Fig. 2e) and fluorescence traces (g, raw and h, event- triggered mean) from a \(170 \times 340 px\) scan (5.88 Hz) of the \(100 \mu m\) peak displacement configuration (image 3 in (d)). The fish was presented with full- field and spectrally broad ( \(\sim 360 - 650 nm\) ) series of light- flashes. + +Mesoscale and 3D random access imaging of the mouse brain. The width of the adult mouse's brain is \(\sim 10 \text{mm}^{40}\) which makes it too large to be comprehensively captured by conventional 2P microscopy. Here, an experimental goal might be to reliably resolve the \(\sim 20+ \mu m\) somata of major cortical or subcortical neurons across a \(10 \text{mm FOV}\) . At the Nyquist detection limit, this would "only" require \(\sim 1,000\) pixels across, which is well within the range of standard high- resolution scan- configurations. Accordingly, currently the main limitation in achieving this goal is the microscope's maximal FOV. Our nTC design makes important steps to address this limitation. + +When configured for a \(3.5 \text{mm FOV}\) (nTC2), our setup captures about a third of the width of a mouse's brain. In this configuration, a scan of a transverse section from a Thy1:GCaMP6f mouse (Fig. 7a,b) illustrates how the objective's back aperture casts a shadow at the image edge, thus limiting the spatial extent of the scan (Fig. 7c). Within this maximal window, a high- resolution \(1,024 \times 1,024 px\) scan allowed us to resolve the somata of major cortical and hippocampal neurons (Fig. 7d, Supplementary Video S6). Accordingly, at this largest FOV configuration, effective signal detection largely sufficed to capture the mouse brain's major neuron populations. However, with our galvo- galvo setup, scan rates at this level of spatial detail were slow (0.49 Hz, 2 ms/line). Accordingly, we used a mesoscale imaging approach with reduced spatial sampling (256x256 px, 1 ms/line) to capture the entire image at 3.91 Hz. This permitted simultaneous population- level "brain- wide" recording of seizure- like activity across the cortex and underlying hippocampus following bath application of an epileptogenic (high K+, zero Mg2+) solution (Fig. 7e- g). To demonstrate the value of the system for more detailed readout of neuronal activity, we also used random access scans to simultaneously capture distant smaller scan- fields at high resolution, both spatially and temporally (two times 256x128 px at 3.91 Hz, Fig. 7d, h- l, Supplementary Video S6). In the example provided, the laser travelled between the two scan fields separated by \(\sim 1 \text{mm}\) within two 1 ms scan lines. This allowed us to record quasi- simultaneous neural activity across both the cortex and hippocampus at single cell resolution. The generally high SNR in these recordings also suggested that additional temporal or spatial resolution could be gained by the use of resonance scanners in place of our galvos41. The large FOV nTC2 configuration also lends itself to imaging mouse cortical neurons activity in vivo (Fig. 8), an increasingly common demand in neuroscience. Here, the maximal 3.5 mm FOV captured an entire cranial window of a Thy1- GCaMP6f mouse prepared for optical interrogation of the somatosensory cortex, comprising an estimated \(10,000+\) neurons in a given image plane (Fig. 8a- c, Supplementary Video S7). + +<--- Page Split ---> +![](images/Figure_8.jpg) + + +<--- Page Split ---> + +Figure 7 | Mesoscale and random-access imaging in mouse brain slice. a,b, Schematic of brain (a) and transverse section (b) of a Thy1:GCalMP6f mouse. c,d, \(1024 \times 1024\) px \(nTC_2\) example scan of slice through cortex and hippocampus at maximal FOV (c) and \(nTC_2\) zoom in (d) as indicated. Red arrows indicate rapid transitions between scan regions, with the inset showing scan- profiles. The slice was bathed in an epileptogenic (high \(K^+\) , zero \(Mg^{2 + }\) ) solution to elicit seizures. e-g, Mean of 256x256 px scan (3.91 Hz) of (d) with regions of interest (ROIs) indicated (e), activity- correlation projection (Methods) indicating regions within the scan showing regions of activity computed as mean correlation of each pixel's activity over time to all its neighbours (for details, see Ref \(^{70}\) ) (f) and z- normalised fluorescence traces (g). h-l, 2 times 128x256 px (3.91 Hz) random access scan of two regions as indicated in (d) allows quasi- simultaneous imaging of the cortex (h) and hippocampus (i) at increased spatial resolution, with activity- correlation (j,k, cf. Fig. 2e) and fluorescence traces (l) extracted as in (j,k). + +Even in an intermediate \(nTC_1\) configuration (in this case a 1.5 mm FOV) the full image still comprised several 1,000s of neurons (Fig. 8d), many more than could be simultaneously captured at scan- rates suitable for functional circuit interrogation with a galvo- galvo setup. In an example scan we again used a random- access approach to quasi- simultaneously record two \(330 \times 210 \mu m\) regions separated by \(\sim 1.2\) mm (two times 128x64 px at 7.81 Hz). As in the brain slice preparation (Fig. 7), this reliably resolved individual neurons in spatially distinct regions of the mouse brain (Fig. 8d- i). Finally, we also recruited the ETL to set up an axially tilted scan plane. This allowed quasi- simultaneous recording from neurons separated several hundreds of \(\mu m\) in depth across layers 1- 4 of the mouse cortex in vivo (Fig. 8j- l). + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 8 | Mesoscale random-access imaging of mouse cortex in vivo. a,b, Schematic of Thy1:GcAmP6f mouse brain in vivo (a) with cranial window over the somatosensory cortex (b). c,d, 1024x1024 px nTC2 (c) and nTC1 (d) images as indicated. Red arrows indicate rapid transitions between scan regions, with the inset indicating the scan-profile. e-i, 2 times 128x256 px (3.91 Hz) random access scan as indicated in (d) with mean-projection (e,f), activity-correlation (g,h, cf. Fig. 2e) and fluorescence traces (i), taken from the ROIs as indicated in (g,h). j-l, nTC1 128x128 px xyz-tilted plane (7.82 Hz) traversing through cortical layers 1-4 at \(\sim 45^{\circ}\) relative to vertical with mean image (k) and activity-correlation (l, cf. Fig. 2e).
+ +<--- Page Split ---> + +Multi- plane circuit mapping with optogenetics in Drosophila. Despite the generally enlarged FOV and concomitant increase in the PSF, our setup was still capable of resolving details of small neural processes in the <0.1 mm diameter nervous system of a first instar larval fruit fly. To assess the difference in image- resolution between our nTC setup and a DL- configuration, we first obtained anatomical scans from a third instar VGlut:GCaMP6f larva which expressed GCaMP6f in structurally well- defined neurons of the ventral nerve cord (Fig 9a,b). This revealed that while the DL image was clearly sharper (Fig. 9a), the nTC1 system nevertheless comfortably delineated individual somata (Fig. 9b). + +Drosophila was an ideal preparation to demonstrate our system's capacity for multi- plane imaging for optogenetic functional circuit mapping (Fig. 9c- k). At the first larval stage, the height of the brain excluding the ventral nerve cord is in the order of \(\sim 60 - 70 \mu m\) . Assuming an axial capture of \(\sim 3 \mu m\) per plane in a DL configuration (cf. Fig. 10), comprehensively sampling from the whole brain would therefore require upwards of 20 planes (Fig. 9c). Here, the slightly elongated PSF of the nTC 1.2 mm configuration served as a useful compromise between spatial resolution and sampling density (Fig. 9d). To demonstrate the sampling that can be achieved under these conditions, we used a transgenic first instar larva that expressed the red- shifted optogenetic effector CsChrimson in all olfactory- sensory neurons (OSNs) on a background of pan- neuronal GCaMP6s (elav:GCaMP6s) (Fig. 9e,f). To reveal any potential bilateral crosstalk of olfactory signal processing across the brain's two hemispheres, one of the olfactory nerves was cut. We set up six image planes (six times 340x170 px), each separated by \(\sim 15 \mu m\) which together captured the entire brain across both hemispheres at \(\sim 1 \text{Hz}\) (Fig. 9d,f). In this configuration, presentation of 2 s flashes of red light from a scanline- synchronized 590 nm LED activated olfactory sensory neurons (OSNs). These in turn propagated the signal to higher processing centres, which we visualised as regionally restricted GCaMP6s responses in the brain (Fig. 9g- k, Supplementary Video S8). The most strongly activated region was the ipsilateral antennal lobe (AL) (see also SFig. 5) which is directly innervated by the still- intact OSNs. Similarly, the olfactory second order processing centres, the mushroom body and the lateral horn, showed clear ipsilateral activation. In addition to these three major olfactory centres and their connecting tracts (e.g. plane 3), further processes and somata across both the ipsi- and contralateral lobe were also activated. Taken together, despite the slight expansion of the DL excitation spot, our nTC setup nevertheless allowed us to delineate key structural and functional information in this small insect brain. + +<--- Page Split ---> +![](images/Figure_1.jpg) + + + +
Figure 9 | Multi-plane imaging and optogenetics for functional circuit mapping. a,b, DL (a) and nTC₁ (b) 1,024x1,024 px scans of the ventral nerve cord of a 3rd instar VGlut:GCamP6f Drosophila
+ +<--- Page Split ---> + +lava. c- f, Scan- profiles taken in DL (c) and nTC; (d) across 6 planes spaced \(\sim 15 \mu m\) apart. e,f, Schematic of first instar elav:GCamP6s; Ocro:CsChrimson Drosophila larva from top (e) and side (f), with CsChrimson (red) and GCaMP6s (green) expression pattern and scan- planes indicated. g- k optogenetic circuit mapping of olfactory processing centres across the larval brain. Six scan planes (170x340 px each) were taken at 0.98 Hz/plane (i.e. volume rate) during presentation of 587 nm light flashes (2 s) to activate CsChrimson in olfactory sensory neurons (OSNs). Brain anatomy (g) and false- colour coded fluorescence difference image (h, 1- 2 s after flash onset minus 1- 2 s prior to flash onset), with fluorescence activity traces (i, raw and j, event triggered average). For a zoom- in on the antennal lobe in a different specimen, see also SFig. 5. k, data from (h) summarised: top right: max- projection through the brain, with left and bottom showing transverse max- projections across the same data- stack. + +## DISCUSSION + +The ongoing development of sophisticated optical probes to report on key biophysical events has increasingly raised the demand in neuroscience for high SNR and large FOV 2P microscopes. To date, however, these characteristics are almost exclusively limited to high- end and, inevitably, high- cost platforms. Here, we exploit the fact that in 2P microscopy there is no "traditional" collection plane, allowing us to deviate from the diffraction limited regime that is typically used in systems where the planes of excitation and collection must superimpose to avoid image blur. Instead, we propose a simple core modification of the laser path that allows upgrading an out- of- the- box DL 2P microscope into a system capable of performing high SNR and large- FOV volumetric scans while at the same time preserving single cell resolution. We demonstrate the capabilities of this system for interrogating dynamic events in the brains of a range of key model species that are already widely used in neuroscience research. Since the core modification only requires the user to swap the scan lens for one or two off- the- shelf lenses, it can be tested (and fully reversed) within a matter of hours without the need for optical re- alignment or calibration. We anticipate that the simplicity and cost- effectiveness of this solution and the significant enhancement in 2P imaging capabilities that it permits, will lead to its wide adoption by the neuroscience community. + +Combining an nTC approach with existing custom 2P designs. The estimation of metrics that meaningfully compare the capability of our nTC design with other custom solutions is difficult, as these can depend strongly on the specific objective (N.A., back aperture size, working distance (focus)), its distance from the tube lens, and indeed the nature of the interrogated sample and the biological question itself. Rather, because our nTC approach fundamentally differs from traditional DL optics, it opens the possibility to further enhance the capabilities of existing custom 2P microscope designs. + +A key benefit of our nTC approach is the flexibility that it offers. It can be seamlessly implemented on setups with galvanometric or resonant mirrors to work with a wide range of scan- strategies. Here, the "extra" optical magnification afforded by the FOV expansion means that scan- mirror and ETL movements translate to relatively larger xy or z- translations, + +<--- Page Split ---> + +respectively, making it easy to rapidly execute complex and large- scale 3D scan- paths. Our approach can also be combined with existing setups that use rapid piezo- positioning of the objective for axial scans, although in this case the objective movement relative to the tube lens will generate small but systematic variations in FOV and PSF shape. Accordingly, the use of remote focussing before the scan- mirrors is likely to be preferable in most applications. + +Like in most 2P designs, our use of a Gaussian beam does not permit the generation of a truly arbitrary PSF shape. Nevertheless, if used in combination with temporal focussing12,42 it would, in principle, be possible to modulate axial PSF expansion without strongly affecting lateral expansion, thus facilitating a greater range of PSF shapes. Similarly, an optimized design of the objective lens43 and other optical elements4 including the use of large diameter lenses to minimize aberrations5, could all be combined with our optical design to further enhance the quality of 2P excitation. + +nTC and optical aberrations. In general, beyond the PSF expansion that results from bypassing the objective's infinity correction (Figs. 1,2, Fi. S1), the change from a standard 2P DL- setup to an nTC configuration does not bring about new types of aberrations. In short, chromatic aberrations (which necessitate complex optical corrections in 1P microscopy) do not apply in 2P microscopy, because the excitation laser is essentially monochromatic and collection is spatially invariant. Instead, spherical aberrations tend to be dominant in 2P microscopy, i.e. when peripheral and axial rays do not converge to a point44- 49. The optical element that has the largest impact here is the objective, which is not changed under nTC. Further monochromatic aberrations are mainly related to the sample structure and surrounding (immersion) medium itself. In the future, it will be useful to explore how adaptive optics can address many of the above points, including spherical aberrations as well as coma and astigmatism46 - see also Supplementary Discussion. + +Axial signal integration. As well as permitting the tailoring of the PSF to a given biological application, the use of a non- DL excitation spot can also bring about additional benefits. First, the lower effective excitation N.A. produces a narrower light cone which is less likely to be scattered by tissue inhomogeneities50. Second, objects that are smaller than the focal excitation volume become dimmer, while objects that are similar in size or larger remain bright8,51. Third, PSF expansion also reduces photobleaching and photodamage which can have a more- than- quadratic intensity dependence52,53. For example, when using the large PSF of the 3.5 mm FOV configuration, it was possible to use up to 250 mW laser power without causing notable damage when imaging deep in the mouse cortex54. Here, calculations and experimental experience suggest that in general, our strategy of underfilling the objective's back aperture will greatly ameliorate photodamage48,52,53. Notwithstanding, any axial expansion in the + +<--- Page Split ---> + +PSF must be balanced with potentially undesirable merging of distinct image structures separated in depth. + +Conclusion. Taken together, our nTC approach offers key advantages over traditional DL 2P microscopy, including the capacity for an increased FOV, PSF- tailoring, rapid z- travel through minimal ETL commands, overall increased laser power at the sample plane, and reduced spherical aberrations48,55,56. Moreover, it can principally be combined with a wide range of existing customisations to further push the capabilities of 2P microscopy in general. At the same time, our nTC approach is cost effective and can be readily implemented on an existing DL setup with minimal need for optical alignments and calibration. + +## METHODS + +User manual. A complete user manual for the nTC design, as well as a bill of materials (BOM), 3D printable lens holders and printed circuit board (PCB) designs are available online at https://github.com/BadenLab/nTCscope. + +DL 2P microscope. Our setup was based on a Sutter MOM- type two- photon microscope (designed by W. Denk, MPI, Martinsried; purchased through Sutter Instruments) as described previously57. + +Excitation path. The excitation beam was generated by a tuneable femtosecond Ti:Sapphire laser (Coherent Vision- S, 75 fs, 80 MHz, \(>2.5\) W). The laser passed an achromatic half- wave plate (AHWP05M- 980, Thorlabs) and was subsequently equally split to supply two independent 2P setups using a beam- splitter for ultrashort pulses (10RQ00UB.4, Newport). Next, the beam passed a Pockels cell (350- 80 with model 302 driver, Conoptics), a telescope (AC254- 075- B and AC254- 150- B, Thorlabs), and was finally reflected into the head part of Sutter MOM stage by a set of three silver mirrors (PF10- 03- P01). We used a pair of single- axis galvanometric scan mirrors (6215H, Cambridge Technology) which directed the beam into a 50 mm focal length scan lens (VISIR 1534SPR136, Leica) at a distance of 56.6 mm. A 200 mm focal length tube lens (MXA22018, Nikon) was positioned 250 mm further along the optical path. From here, the now collimated excitation beam was directed onto the xyz- movable head of the Eyecup scope57 which was controlled by a motorized micromanipulator (MP285- 3Z, Sutter Instruments). Here, the beam was reflected by two silver parabolic mirrors to pass the collection path dichroic mirror (T470/640rpc, Chroma) to finally slightly overfill the back aperture of the objective (Zeiss Objective W "Plan- Apochromat" 20x/1.0), thus creating a diffraction- limited excitation spot at the objective's nominal working distance of 1.8 mm. The distance between the tube lens and the objective's back aperture was 95 mm at the centre position of the xyz displacement mechanism, and the parabolic mirrors ensured that the + +<--- Page Split ---> + +optical excitation axes stayed aligned during movements of the microscope head. + +Collection path. Collection was exclusively through the objective (except for Fig. S2g- i). For this, a dichroic mirror (T470/640rpc, Chroma) was positioned \(18\mathrm{mm}\) above the objective's back aperture to reflect fluorescence light into the collection arm. Here, a \(140\mathrm{mm}\) focal length collecting lens was followed by a \(580\mathrm{- nm}\) dichroic mirror (H 568 LPXR, superflat) to split the signal into two wavebands. The "green" and "red" channels each used a single- band bandpass filter (ET525/50 and ET 605/50, respectively, Chroma) and an aspheric condenser lens (G317703000, Linos) to focus light on a PMT detector chip (H10770PA- 40, Hamamatsu). + +For collection efficiency measurements (Fig. S2g- i) an additional sub- stage collection path was installed. To facilitate comparison, all optical components were identical to the above- stage excitation as collection patch (with the exception of the lack of the above- objective dichroic in the sub- stage setup). For this, a second objective (Zeiss W "Plan- Apochromat" \(20\mathrm{x} / 1.0\) ) was focused on the sample plane, and the collimated fluorescence light was subsequently focused through aspheric condenser lens (G317703000, Linos) and single- band bandpass filter (ET525/50, Chroma) on the PMT detector chip (H10770PA- 40, Hamamatsu). + +Image acquisition. We used custom- written software (ScanM, by M. Mueller, MPI, Martinsried and T. Euler, CIN, Tuebingen) running under IGOR pro 6.3 for Windows (Wavemetrics) to control the setup. For hardware- software communication we use two multifunction I/O devices (PCle- 6363 and PCI- 6110, National Instrument). Within ScanM, we defined custom scan- configurations: \(1,024\mathrm{x}1,024\) and \(512\mathrm{x}512\) pixel images with 2 ms per line were used for high- resolution morphology scans, while faster, 1 ms or 2 ms linespeed image sequences with \(256\times 256\) (3.91Hz), \(128\times 128\) (7.81 Hz), \(340\times 170\) (5.88 Hz) or \(128\times 64\) (15.6 Hz) pixels were used for activity scans. All scans were unidirectional, and the laser was blanked via the Pockels cell during the turnarounds and retrace. This period was also used for light stimulation (zebrafish visual system and Drosophila optogenetics, see below). + +Non- collimated 2P microscope modifications. We used two sets of modifications (nTC1 and nTC2) to de- collimate the excitation path to different degrees. For nTC1 (FOV \(1.2 - 1.8\mathrm{mm}\) ) we modified the original Sutter- MOM scan lens (VISIR 1534SPR136, Leica) by removing the second lens (i.e. the one closer to the tube lens) from the compound mount which changed the focal length from 50 to \(190\mathrm{mm}\) . Alternatively, the entire de- constructed scan lens could also be replaced by a similar power off- the- shelf plano- convex lens. Our \(190\mathrm{mm}\) lens (L1) was placed exactly \(190\mathrm{mm}\) in front of the tube lens (so shifted \(60\mathrm{mm}\) forward from its original position). Next, we introduced an additional plano- convex \(175\mathrm{mm}\) focal + +<--- Page Split ---> + +distance lens (L2) (LA1229, Thorlabs). L2 was held in place by custom 3D printed mount (cf. user manual) inside the MOM's tube-lens holder and positioned anywhere between 0 and \(10\mathrm{mm}\) in front of the tube lens. Depending on the exact position of L2 within this range, the effective FOV at the image plane could be adjusted between \(1.2\mathrm{mm}\) ( \(10\mathrm{mm}\) distance) to \(1.8\mathrm{mm}\) (L2 and tube lens almost touching). + +For \(n\mathrm{TC}_2\) (FOV \(2.5 - 3.5\mathrm{mm}\) ), we replaced the original scan lens with a single, \(200\mathrm{mm}\) focal length plano-convex lens L3 (LA1708, Thorlabs). Like L2 in \(n\mathrm{TC}_1\) , L3 was mounted on the same custom 3D printed holder and positioned anywhere within a distance of \(0 - 10\mathrm{mm}\) in front of the tube lens. In this case the FOV at the image plane could be adjusted between 2.5 mm ( \(10\mathrm{mm}\) distance) to \(3.5\mathrm{mm}\) (L3 and tube lens almost touching). For detailed instructions including photos of the optical path, consult the user manual. + +We selected lens types and positions based on the available space within the Sutter MOM head such that for \(n\mathrm{TC}_1\) and \(n\mathrm{TC}_2\) , the IFP was always located in front of or behind the TL, respectively. However, depending on the design of a given 2P setup's excitation path, numerous alternative configurations are possible. Here, a straight- forward means to rapidly estimate the nature and scale of a given modification is to use a fluorescence test- slide and observe the change in working distance and FOV as the scan path is modified. + +Electrically tunable lens (ETL) for rapid axial focussing. For rapid z- focussing we added a horizontal ETL (EL- 16- 40- TC- 20D, Optotune) into the vertical beam path after the silver mirror that reflected the excitation beam up into the MOM head, \(200\mathrm{mm}\) in front of the scan- mirrors. To drive the ETL we used a custom current driver controlled by an Arduino Duo microcontroller (see user manual), capable of generating positive currents between 0- 300 mA. The Arduino Duo received a copy of the scan- line command and in turn output commands to the current driver to effect line- synchronised changes in ETL curvature. Prior to initiating a scan, the specific to- be- executed Arduino programme was uploaded to the Arduino via serial from a PC running a custom Matlab- script (Mathworks). This Matlab script launched a simple graphical user interface (GUI) that allowed the user to configure the exact lens- path during a custom scan (see user manual). Accordingly, ETL control remained flexible and fully independent of the scan software. In this way, our solution can be readily integrated with any 2P system without need to change the software or acquisition/driver hardware. Notably, this ETL implementation can also be used by itself, without need for implementing any of the other optical adjustments described in this work. However, depending on the system's optics, the effective range of z- travel would likely be smaller. A detailed step- by- step guide to implement the ETL, including the control software and hardware is provided in the user manual. + +<--- Page Split ---> + +Pockels cell. To control excitation laser intensity, we use a Pockels cell (Model 350- 80, Conoptics; driver model 302, Conoptics). A line- synchronised blanking signal was sent from the DAQ to the drive to minimise laser power during the retrace. In addition, a custom circuit allowed controlling effective laser brightness during each scan line via a potentiometer (see user manual, designed by Ruediger Bernd, HIH, University of Tübingen). As required, this amplitude- modulated signal could then be further modulated by a second Arduino Due controlled by a standalone Matlab GUI to automatically vary effective laser power as a function of scanline index. In this way, laser power could be arbitrarily modulated on a line by line basis, for example to compensate for possible power loss when imaging at increased depth. + +Light stimulation. For visual stimulation of zebrafish larvae (Figs. 5, S2, 6) we used a full- field, broadband spot of light projected directly onto the eyes of the fish from the front via a liquid light guide (77555, Newport) connected to a custom collimated LED bank (Roithner LaserTechnik) with emission peak wavelengths between 650 and 390 nm to yield an approximately equal power spectrum over the zebrafish's visual sensitivity range (described in detail in Ref58). LEDs were line- synchronised to the scanner retrace by an Arduino Due. For CsChrimson activation (Fig. 7) we used a custom 2P line synchronised LED stimulator (https://github.com/BadenLab/Tetra- Chromatic- Stimulator) equipped with four 587 nm peak emission LEDs embedded in a custom 3D printed recording chamber. + +Image brightness measurements. We imaged a uniform florescent sample consisting of two microscopy slides (S8902, Sigma- Aldrich) encapsulating a drop of low melting point agarose (Fisher Scientific, BP1360- 100) mixed with low concentrated Acid Yellow 73 fluorescein solution (F6377 Sigma- Aldrich). Show is the average brightness over the radius from the centre to the edge of the FOV (Matlab, custom scripts). + +PSF measurements. We used \(0.175 \pm 0.005 \mu m\) yellow- green (505/515) fluorescent beads (P7220, Invitrogen) embedded in a 1 mm depth block of \(1\%\) low melting point agarose (Fisher Scientific, BP1360- 100). Image stacks were acquired across \(30 \times 30 \mu m\) lateral field of view with 256x256 pixels resolution (0.12 \(\mu m\) /pixel) and \(0.5 \mu m\) axial steps/frame. For xy and z- dimensions, we calculated the full width at half maximum (FWHM) from Gaussian fits to the respective intensity profiles. Measurements were taken from set of the beads distributed across the entire FOV, and presented results are averages of at least 10 measurements of different beads, with error bars given in s.d.. + +To film the PSF and effective scan- plane(s) we focussed an air- objective (Plan Apo \(4 \times /0.20\) , Nikon) onto the excitation spot elicited in a plastic cuvette with fluorescein (F2456 Sigma- Aldrich) dissolved in water which was positioned beneath the excitation objective. The camera path was fitted with a single- band bandpass filter (ET525/50, Chroma) and a colour + +<--- Page Split ---> + +CCD camera (Manta G- 031C, Allied Vision). The camera was controlled with its dedicated software (VIMBA, Allied Vision). + +Animal experiments. All animal experiments presented in this work were carried out in accordance with the UK Animal (Scientific Procedures) Act 1986 and institutional regulations at the University of Sussex. All procedures were carried out in accordance with institutional, national (UK Home Office PPL70/8400 (mice), PPL/PE08A2AD2 (zebrafish)) and international (EU directive 2010/63/EU) regulations for the care and use of animals in research. + +Zebrafish larvae preparation and in- vivo imaging. Zebrafish were housed under a standard 14:10 day/night rhythm and fed 3 times a day. Animals were grown in 200 mM 1- phenyl- 2- thiourea (Sigma) from 1 day post fertilization (dpf) to prevent melanogenesis. Preparation and mounting of zebrafish larvae was carried out as described previously59. In brief, we used 6- 7 dpf zebrafish (Danio rerio) larvae that were immobilised in 2% low melting point agarose (Fisher Scientific, Cat: BP1360- 100), placed on the side on a glass coverslip and submerged in fish water. For eye- brain imaging, eye movements were prevented by injection of a- bungarotoxin (1 nL of 2 mg/ml; Tocris, Cat: 2133) into the ocular muscles behind the eye. Transgenic lines used were Islet2b:mGCaMP6f (eye- brain imaging) and HuC:GCaMP6f28 (image of 3 zebrafish in same FOV). Zebrafish were imaged at 930 nm and 30- 60 or 50- 100 mW for brain and eye imaging, respectively. + +Creation of Islet2b:mGCaMP6f transgenic line. Tg(is12b:nlsTrpR, tUAS:memGCaMP6f) was generated by co- injecting pTol2- isl2b- hlsTrpR- pA and pBH- tUAS- memGaMP6f- pA plasmids into single- cell stage eggs. Injected fish were out- crossed with wild- type fish to screen for founders. Positive progenies were raised to establish transgenic lines. All plasmids were made using the Gateway system (ThermoFisher, 12538120) with combinations of entry and destination plasmids as follows: pTol2- isl2b- nlsTrpR- pA; pTol2pA60, p5E- isl2b61, pME- nlsTrpR62, p3E- pA60; pBH- tUAS- memGaMP6f- pA; pBH63, p5E- tUAS62, pME- memGCaMP6f, p3E- pA. Plasmid pME- memGCaMP6f was generated by inserting a polymerase chain reaction (PCR)- amplified membrane targeting sequence from GAP- 4364 into pME plasmid and subsequently inserting a PCR amplified GCaMP6f65 at the 3' end of the membrane targeting sequence. + +Acute brain slices. 1- 2 month old male Thy1- GCaMP6f- GP5.1766 mice were used. Acute transverse brain slices (300 μm) were prepared using a vibroslicer (VT1200S, Leica Microsystems, Germany) in ice- cold artificial cerebrospinal fluid (ACSF) containing (in mM): 125 NaCl, 2.5 KCl, 25 glucose, 1.25 NaH2PO4, 26 NaHCO3, 1 MgCl2, 2 CaCl2 (bubbled with 95% O2 and 5% CO2, pH 7.3), and allowed to recover in the same buffer at 37°C for 60 minutes67. During imaging, slices were constantly perfused with 37°C modified (epileptogenic) saline (37°C) containing 125 NaCl, 5 + +<--- Page Split ---> + +KCl, 25 glucose, 1.25 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂. Brain slices were imaged at 930 nm and 100- 150 mW. + +Mouse surgical procedures for in vivo- imaging of the barrel cortex. Head bar implantation surgery has been described elsewhere68. Briefly, under aseptic conditions, a male mouse expressing a calcium indicator in pyramidal neurons (GCaMP6f; GP5.17⁶⁶) was anaesthetised with isoflurane and implanted with a custom- made head bar. A circular 3 mm diameter craniotomy centred at 3.0 mm lateral and 1.0 mm posterior to bregma was made to expose the cranial surface. A cranial window, consisting of a 3 mm circular coverslip and a 5 mm circular coverslip (Harvard Apparatus), was placed over the craniotomy and secured in place with cyanoacrylate tissue sealant (Vetbond, 3M). Following 7 days of recovery, the mouse was handled daily and acclimated to a head fixation apparatus over a treadmill for a further 9 days. During 2P imaging, the head- fixed mouse could locomote freely on a custom- made treadmill. The mouse was awake and received fluid rewards between imaging batches. Cortical neurons were imaged at 960 nm and 100- 150 mW. + +Drosophila larval preparation and in- vivo imaging. Flies were maintained at 25°C in 12 h light:12 h dark conditions. Fly stocks were generated using standard procedures. The genotypes of the D. melanogaster flies used were: elav- Gal4; LexAOp- CsChrimson and w; UAS- GCaMP6s; Orco- LexA. These two strains were crossed to each other (collecting virgins from the first one and males from the second one) and placed on laying- pots at 25°C for larval collection. The laying- pots had a grape juice agar plate with an added drop of yeast paste supplemented with all- trans retinal (Sigma- Aldrich) to a final concentration of 0.2 mM. Yeast supplemented agar plates were changed every day and first instar larvae were picked off the new changed plate. First instar larvae were collected from yeast supplemented agar plates and dissected on physiological saline as in Ref⁶⁹ (in mM): 135 NaCl, 5 KCl, 5 CaCl₂- 2H₂O, 4 MgCl₂- 6H₂O, 5 TES (2- [[1,3- dihydroxy- 2- (hydroxymethyl)propan- 2- yl]amino]ethanesulfonic acid), 36 Sucrose, adjusted to pH 7.15 with NaOH. Larvae were dissected to expose the brain while maintaining intact the anterior part of the animal and the connection between OSN cell bodies and the brain, subsequently one of the olfactory nerves was cut with the forceps. The preparation was then positioned on top of a coverslip coated with poly- lysine (Sigma- Aldrich, P1524- 100MG), and covered in 2% low melting point agarose (Fisher Scientific, Cat: BP1360- 100) diluted in physiological saline, to prevent movement associated with mouth- hook contractions. The sample was then submerged in physiological saline. Larval brains were imaged at 930 nm and 30- 60 mW. + +<--- Page Split ---> + +1130 1. Denk, W., Strickler, J. H. H., Webb, W. W. 1131 W., Series, N. & Apr, N. Two- Photon Laser 1132 Scanning Fluorescence Microscopy. 1133 Science (80- ). 248, 73- 76 (1990). 1134 2. Weisenburger, S. & Vaziri, A. A guide to 1135 emerging technologies for large- scale and 1136 whole brain optical imaging of neuronal 1137 activity. (2018). doi:10.1146/annurev- neuro- 072116- 031458 1139 3. Ota, K. et al. Fast scanning high optical 1140 invariant two- photon microscopy for 1141 monitoring a large neural network activity 1142 with cellular resolution. bioRxiv 1143 2020.07.14.201699 (2020). 1144 doi:10.1101/2020.07.14.201699 1145 4. Bumstead, J. R. Designing a large field- of- 1146 view two- photon microscope using optical 1147 invariant analysis. Neurophotonics 5, 1 1148 (2018). 1149 5. Tsai, P. S. et al. Ultra- large field- of- view 1150 two- photon microscopy. Opt. Express 1151 (2015). doi:10.1364/oe.23.013833 1152 6. Stirman, J. N., Smith, I. T., Kudenov, M. W. 1153 & Smith, S. L. Wide field- of- view, multi- 1154 region, two- photon imaging of neuronal 1155 activity in the mammalian brain. Nat. 1156 Biotechnol. 34, 857- 862 (2016). 1157 7. Yu #, C.- H., Stirman #, J. N., Yu, Y., Hira, 1158 R. & Smith, S. L. Diesel2p mesoscope with 1159 dual independent scan engines for flexible 1160 capture of dynamics in distributed neural 1161 circuitry. bioRxiv 2020.09.20.305508 1162 (2020). doi:10.1101/2020.09.20.305508 1163 8. Sofroniew, N. J., Flickinger, D., King, J. & 1164 Svoboda, K. A large field of view two- 1165 photon mesoscope with subcellular 1166 resolution for in vivo imaging. Elife (2016). 1167 doi:10.7554/elife.14472 1168 9. Han, S., Yang, W. & Yuste, R. Two- Color 1169 Volumetric Imaging of Neuronal Activity of 1170 Cortical Columns. Cell Rep. (2019). 1171 doi:10.1016/j.celrep.2019.04.075 1172 10. Cheng, A., Goncalves, J. T., Golshani, P., 1173 Arisaka, K. & Portera- Cailliau, C. 1174 Simultaneous two- photon calcium imaging 1175 at different depths with spatiotemporal 1176 multiplexing. Nat. Methods (2011). 1177 doi:10.1038/nmeth.1552 1178 11. Prevedel, R. et al. Fast volumetric calcium 1179 imaging across multiple cortical layers using 1180 sculpted light. Nat. Methods (2016). 1181 doi:10.1038/nmeth.4040 1182 12. Weisenburger, S. et al. Volumetric Ca 2+ 1183 Imaging in the Mouse Brain Using Hybrid 1184 Multiplexed Sculpted Light Microscopy. Cell 1185 (2019). doi:10.1016/j.cell.2019.03.011 1186 13. Lu, R. et al. Video- rate volumetric functional 1187 imaging of the brain at synaptic resolution. 1188 Nat. Neurosci. (2017). doi:10.1038/nn.4516 1189 14. Botcherby, E. J., Juškaitis, R. & Wilson, T. 1190 Scanning two photon fluorescence 1191 microscopy with extended depth of field. 1192 (2006). doi:10.1016/j.optcom.2006.07.026 1193 15. Song, A. et al. Volumetric two- photon 1194 + +1195 imaging of neurons using stereoscopy (vtwins). Nat. Methods (2017). doi:10.1038/nmeth.4226 16. Grewe, B. F., Langer, D., Kasper, H., Kampa, B. M. & Helmchen, F. high- speed in vivo calcium imaging reveals neuronal network activity with near- millisecond precision. (2010). doi:10.1038/nmeth.1453 17. Chong, E. Z., Panniello, M., Barreiros, I., Kohl, M. M. & Booth, M. J. Quasi- simultaneous multiplane calcium imaging of neuronal circuits. Biomed. Opt. Express (2019). doi:10.1364/boe.10.000267 18. Grewe, B. F., Voigt, F. F., van 't Hoff, M. & Helmchen, F. Fast two- layer two- photon imaging of neuronal cell populations using an electrically tunable lens. Biomed. Opt. Express (2011). doi:10.1364/boe.2.002035 19. Yang, W., Carrillo- Reid, L., Bando, Y., Peterka, D. S. & Yuste, R. Simultaneous two- photon imaging and two- photon optogenetics of cortical circuits in three dimensions. Elife (2018). doi:10.7554/elife.32671 20. Sheffield, M. E. J. & Dombeck, D. A. Calcium transient prevalence across the dendritic arbor predicts place field properties. Nature (2015). doi:10.1038/nature13871 21. Zhao, Z. et al. The temporal structure of the inner retina at a single glance. bioRxiv 743047 (2019). doi:10.1101/743047 22. Denk, W. & Svoboda, K. Why multiphoton is more than a gimmick. Neuron (1997). doi:10.1016/S0896- 6273(00)81237- 4 23. Svoboda, K. & Yasuda, R. Principles of Two- Photon Excitation Microscopy and Its Applications to Neuroscience. Neuron (2006). doi:10.1016/j.neuron.2006.05.019 24. Born, M. & Wolf, E. Principles of Optics Electromagnetic Theory of Propagation, Interference and Diffraction of Light. Princ. Opt. Electromagn. Theory Propagation, Interf. Diffr. Light by Max Born, Emil Wolf Oxford, GB Pergamon Press. 1980 (1980). 25. Philibert S. Tsai and David Kleinfeld. In Vivo Optical Imaging of Brain Function, Second Edition. Methods (2009). doi:10.1201/9781420076851 26. Charles, A., Song, A., Gauthier, J., Pillow, J. & Tank, D. Neural Anatomy and Optical Microscopy (NAOMi) Simulation for evaluating calcium imaging methods. bioRxiv 726174 (2019). doi:10.1101/726174 27. Zipfel, W. R., Williams, R. M. & Webb, W. W. Nonlinear magic: multiphoton microscopy in the biosciences. Nat. Biotechnol. (2003). doi:10.1038/nbt899 28. Quirin, S. et al. Calcium imaging of neural circuits with extended depth- of- field light- sheet microscopy. Opt. Lett. (2016). doi:10.1364/ol.41.000855 29. Fahrbach, F. O. et al. Rapid 3D light- sheet microscopy with a tunable lens "light" + +<--- Page Split ---> + +1258 Accuracy 3D Quantum Dot Tracking with 1325 1259 Multifocal Plane Microscopy for the Study of 1326 1260 Fast Intracellular Dynamics in Live Cells. 1327 Biophys. J (2008). 1326 doi:10.1364/OE.21.021010 1329 30. Ahrens, M. B., Orger, M. B., Robson, D. N., 1330 1264 Li, J. M. & Keller, P. J. Whole-brain 1331 functional imaging at cellular resolution 1332 using light-sheet microscopy. Nat. Methods 1333 10, 413- 420 (2013). 1334 1268 31. Leung, L. C., Wang, G. X. & Mourrain, P. 1335 Imaging zebrafish neural circuitry from 1336 whole brain to synapse. Front. Neural 1337 Circuits (2013). 1338 doi:10.3389/fncir.2013.00076 1339 32. Kermen, F., Lal, P., Faturos, N. G. & Yaksi, 1340 E. Interhemispheric connections between 1341 olfactory bulbs improve odor detection. 1342 PLoS Biol. 18, e3000701 (2020). 1343 33. Wu, Y., dal Maschio, M., Kubo, F. & Baier, 1344 H. An Optical Illusion Pinpoints an Essential 1345 Circuit Node for Global Motion Processing. 1346 Neuron (2020). 1347 doi:10.1016/j.neuron.2020.08.027 1348 34. Sancataldo, G. et al. Flexible multi-beam 1349 light-sheet fluorescence microscope for live 1350 imaging without striping artifacts. Front. 1351 Neuroanat. 13, (2019). 1352 35. Huisken, J. & Stainier, D. Y. R. Selective 1353 plane illumination microscopy techniques in 1354 developmental biology. Development 136, 1963- 1975 (2009). 1356 Lavagnino, Z. et al. Two-photon excitation 1357 selective plane illumination microscopy 1358 (2PE- SPIM) of highly scattering samples: 1359 characterization and application References 1360 and links "Optical sectioning deep inside 1361 live embryos by selective plane illumination 1362 microscopy "High- reso. Nat. Methods 305, 1363 (2004). 1364 Hillman, E. M. C., Voleti, V., Li, W. & Yu, H. 1365 Light- Sheet Microscopy in Neuroscience. 1366 (2019). doi:10.1146/annurev-neuro-070918 1367 Vladimirov, N. et al. Light- sheet functional 1368 imaging in fictively behaving zebrafish. Nat. 1369 Methods 11, 883- 4 (2014). 1370 39. Wulliman, M. F., Rupp, B. & Reichert., H. 1371 Neuroanatomy of the Zebrafish Brain: A 1372 Topological Atlas. (Springer Birkhaeuser, 1373 1996). 1374 40. Kovačević, N. et al. A three- dimensional 1375 MRI atlas of the mouse brain with estimates 1376 of the average and variability. Cereb. 1377 Cortex (2005). doi:10.1093/cercor/bhh165 1378 41. Fan, G. Y. et al. Video- rate scanning two- 1379 photon excitation fluorescence microscopy 1380 and ratio imaging with cameleons. Biophys. 1381 J. (1999). doi:10.1016/S0006- 3495(99)77396- 0 1382 42. Schrödel, T., Prevedel, R., Aumayr, K., 1383 Zimmer, M. & Vaziri, A. Brain- wide 3d 1384 imaging of neuronal activity in 1385 Caenorhabditis elegans with sculpted light. 1386 Artic. Nat. methods 10, (2013). 1387 43. Negrean, A. & Mansvelder, H. D. Optimal 1388 lens design and use in laser- scanning 1389 microscopy. Biomed. Opt. Express (2014). 1390 + +doi:10.1364/boe.5.001588 44. Gerritsen, H. C. & Grauw, C. J. De. Imaging of optically thick specimen using two- photon excitation microscopy. Microsc. Res. Tech. 47, 206- 209 (1999). 45. Egner, A. & Hell, S. W. Aberrations in confocal and multi- photon fluorescence microscopy induced by refractive index mismatch. in Handbook of Biological Confocal Microscopy: Third Edition 404- 413 (Springer US, 2006). doi:10.1007/978- 0- 387- 45524- 2_20 46. Matsumoto, N., Konno, A., Inoue, T. & Okazaki, S. Aberration correction considering curved sample surface shape for non- contact two- photon excitation microscopy with spatial light modulator. Sci. Rep. 8, 1- 13 (2018). 47. Booth, M. J. & Wilson, T. Refractive- index- mismatch induced aberrations in single- photon and two- photon microscopy and the use of aberration correction. J. Biomed. Opt. 6, 266 (2001). 48. Young, P. A., Clendenon, S. G., Byars, J. M., Decca, R. S. & Dunn, K. W. The effects of spherical aberration on multiphoton fluorescence excitation microscopy. J. Microsc. 242, 157- 165 (2011). 49. Matsumoto, N., Inoue, T., Matsumoto, A. & Okazaki, S. Correction of depth- induced spherical aberration for deep observation using two- photon excitation fluorescence microscopy with spatial light modulator. Biomed. Opt. Express 6, 2575 (2015). 50. Helmchen, F. & Denk, W. Deep tissue two- photon microscopy (also about filling of objective). Nat. Methods (2005). doi:10.1038/NMETH818 51. Birge, R. R. Two- Photon Spectroscopy of Protein- Bound Chromophores. Acc. Chem. Res 19, (1986). 52. Patterson, G. H. & Piston, D. W. Photobleaching in two- photon excitation microscopy. Biophys. J. (2000). doi:10.1016/S0006- 3495(00)76762- 2 53. Hopt, A. & Neher, E. Highly nonlinear photodamage in two- photon fluorescence microscopy. Biophys. J. (2001). doi:10.1016/S0006- 3495(01)76173- 5 54. Podgorski, K. & Ranganathan, G. Brain heating induced by near- infrared lasers during multiphoton microscopy. J. Neurophysiol 116, 1012- 1023 (2016). 55. Tung, C.- K. et al. Effects of objective numerical apertures on achievable imaging depths in multiphoton microscopy. Microsc. Res. Tech. 65, 308- 314 (2004). 56. Sieracki, C. K., Levey, C. G. & Hansen, E. W. Simple binary optical elements for aberration correction in confocal microscopy. Opt. Lett. 20, 1213 (1995). 57. Euler, T. et al. Eyecup scope- optical recordings of light stimulus- evoked fluorescence signals in the retina. Pflugers Arch. 457, 1393- 414 (2009). 58. Yoshimatsu, T., Bartel, P., Janiak, F. & Baden, T. Optimal rotation of colour space + +<--- Page Split ---> + +1392 by zebrafish cones in vivo. F1000Research 145 74. 1393 8, (2019). 145 1394 59. Zimmermann, M. J. Y. et al. Zebrafish 145 1395 Differentially Process Color across Visual 145 1396 Space to Match Natural Scenes. Curr. Biol. 145 1397 28, 2018- 2032. e5 (2018). 145 1398 60. Kwan, K. M. et al. The Tol2kit: A multisite 146 1399 gateway- based construction kit forTol2 146 1400 transposon transgenesis constructs. Dev. 146 1401 Dyn. 236, 3088- 3099 (2007). 146 1402 61. Pittman, A. J., Law, M.- Y. & Chien, C.- B. 146 1403 Pathfinding in a largevertebrate axon tract: 146 1404 isotypic interactions guide retinotectal 146 1405 axons at multiple choice points. 146 1406 Development 129, 617- 624 (2008). 146 1407 62. Suli, A., Guler, A. D., Raible, D. W. & 146 1408 Kimelman, D. A targeted gene expression 147 1409 system using the tryptophan repressor in 147 1410 zebrafish shows no silencing in subsequent 147 1411 generations. Development 141, 1167- 74 147 1412 (2014). 147 1413 Yoshimatsu, T. et al. Presynaptic partner 147 1414 selection during retinal circuit reassembly 147 1415 varies with timing of neuronal regeneration 147 1416 in vivo. Nat. Commun. 7, 10590 (2016). 147 1417 64. Kay, J. N. et al. Development. Development 147 1418 128, 2497- 2508 (2004). 148 1419 Chen, T.- W. et al. Ultrassensitive fluorescent 148 1420 proteins for imaging neuronal activity. 148 1421 Nature 499, 295- 300 (2013). 148 1422 66. Dana, H. et al. Thy1- GCaMP6 transgenic 148 1423 mice for neuronal population imaging in 148 1424 vivo. PLoS One (2014). 148 1425 doi:10.1371/journal.pone.0108697 148 1426 67. Rey, S., Marra, V., Smith, C. & Staras, K. 148 1427 Nanoscale Remodeling of Functional 148 1428 Synaptic Vesicle Pools in Hebbian 149 1429 Plasticity. Cell Rep. (2020). 149 1430 doi:10.1016/j.celrep.2020.01.051 149 1431 68. Bale, M. R. et al. Learning and recognition 149 1432 of tactile temporal sequences by mice and 149 1433 humans. Elife (2017). 149 1434 doi:10.7554/eLife.27333 149 1435 Prieto- Godino, L. L., Diegelmann, S. & 149 1436 Bate, M. Embryonic Origin of Olfactory 149 1437 Circuitry in Drosophila: Contact and Activity- 149 1438 Mediated Interactions Pattern Connectivity 150 1439 in the Antennal Lobe. PLoS Biol. 10, (2012). 150 1440 70. Franke, K. et al. Inhibition decorrelates 150 1441 visual feature representations in the inner 150 1442 retina. Nature 542, 439- 444 (2017). 150 1443 71. Goppert- Mayer, M. Über Elementarakte mit 150 1444 zwei Quantensprüngen. Ann. Phys. (1931). 150 1445 doi:10.1002/andp.19314010303 150 1446 72. Larson, D. R. et al. Water- soluble quantum 150 1447 dots for multiphoton fluorescence imaging 150 1448 in vivo. Science (80- ). (2003). 150 1449 doi:10.1126/science.1083780 151 1450 73. Ricard, C. et al. Two- photon probes for in 151 1451 vivo multicolor microscopy of the structure 151 1452 and signals of brain cells. 223, 3011- 3043 151 1453 (2018). + +1454 74. Colon, J. & Lim, H. Shaping field for 3D 1455 laser scanning microscopy. Opt. Lett. 40, 3300- 3 (2015). 1456 Hiraoka, Y., Sedat, J. W. & Agard, D. A. 1457 Determination of three- dimensional imaging 1458 properties of a light microscope system. 1459 Partial confocal behavior in epifluorescence 1460 microscopy. Biophys. J. 57, 325- 333 1461 (1990). 1462 Hell, S., Reiner, G., Cremer, C. & Stelzer, 1463 E. H. K. Aberrations in confocal 1464 fluorescence microscopy induced by 1465 mismatches in refractive index. J. Microsc. 1466 169, 391- 405 (1993). 1467 Costantini, I., Cicchi, R., Silvestri, L., Vanzi, 1468 F. & Pavone, F. S. In- vivo and ex- vivo 1469 optical clearing methods for biological 1470 tissues: review. Biomed. Opt. Express 10, 5251 (2019). 1472 Azucena, O. et al. Wavefront aberration 1473 measurements and corrections through 1474 thick tissue using fluorescent microsphere 1475 reference beacons. Opt. Express 18, 17521 1476 (2010). 1477 Booth, M. J. Wavefront sensorless adaptive 1478 optics for large aberrations. Opt. Lett. 32, 5 1479 (2007). 1480 Ji, N., Milkie, D. E. & Betzig, E. Adaptive 1481 optics via pupil segmentation for high- 1482 resolution imaging in biological tissues. Nat. 1483 Methods 7, 141- 147 (2010). 1484 Tao, X. et al. Live imaging using adaptive 1485 optics with fluorescent protein guide- stars. 1486 Opt. Express 20, 15969 (2012). 1487 Wang, K. et al. Rapid adaptive optical 1488 recovery of optimal resolution over large 1489 volumes. Nature Methods 11, 625- 628 1490 (2014). 1491 Neil, M. A. A. et al. Adaptive aberration 1492 correction in a two- photon microscope. J. 1493 Microsc. 200, 105- 108 (2000). 1494 Theofanidou, E., Wilson, L., Hossack, W. J. 1495 & Arlt, J. Spherical aberration correction for 1496 optical tweezers. Opt. Commun. 236, 145- 150 (2004). 1497 Park, J. H., Kong, L., Zhou, Y. & Cui, M. 1498 Large- field- of- view imaging by multi- pupil 1499 adaptive optics. Nat. Methods 14, 581- 583 1500 (2017). 1501 Booth, M. J., Marsh, ; P, Burns, D. & Girkin, 1502 J. Fluorescence microscopy; (180.6900) 1503 Three- dimensional microscopy; (170.3880) 1504 Medical and biological imaging. Phil. Trans. 1505 R. Soc. A- Math. Phys. Eng. Sci 365, (Wiley, 1506 2007). 1507 Aoyagi, Y., Kawakami, R., Osanai, H., Hibi, 1508 T. & Nemoto, T. A rapid optical clearing 1509 protocol using 2,2'- thiodiethanol for 1510 microscopic observation of fixed mouse 1511 brain. PLoS One (2015). 1512 doi:10.1371/journal.pone.0116280 + +<--- Page Split ---> + +## Figures + +![](images/Figure_2.jpg) + +
Figure 1
+ +Non- Telecentric beam optics in 2- photon microscopy. a, Schematics of larval Drosophila (left). larval zebrafish (centre) and adult mouse (right) with central nervous system highlighted (green) to illustrate size differences. Insets next to the mouse for direct size- comparison between these species on the same + +<--- Page Split ---> + +scale. b, Optical configurations of standard diffraction limited (DL, left) 2P setup with parallel laser beam entering objective's back aperture. Right, non- telecentric (nTC, middle, right) configurations use a still diverging laser beam instead. As a result, the field of view and focal distance are expanded, and the point spread function (PSF) elongates. These effects scale with the angle of divergence (compare n TC1 and nTC2). C, Schematic representations of typical neuronal somata in species shown in (a), as interrogated by 2P setups shown in (b), respectively. d, In vivo 7 dpf larval zebrafish (Huc::GCAMP6f) imaged with an out- of- the- box Sutter- MOM DL setup at full field of view (top) and when zoomed in to reveal individual neuronal somata (bottom) as indicated. e, same zebrafish as shown in (d), as well as two further zebrafish imaged using n TC2 configuration at maximal field of view (top). Zooming in to the same area as in (d, bottom) nonetheless reveals cellular detail (e, bottom). f, In vivo adult mouse cranial window over somatosensory cortex imaged with nTC2 maximal field of view (top) and when zoomed in as indicated (bottom) g, Left: Optical configuration of a standard DL setup with collimation system consisting of a scan lens and a tube lens to set- up an infinity collimated laser beam at the level of the objective's back aperture. Effective refractive power and relative distances of lenses indicated. The intermediary focal point (IFP) is immediately behind the scan lens (arrowhead). Middle: n TC1 configuration replaces the scan lens with a pair of planoconvex lenses (L1,2). The relative position of L2 to the tube lens defines the position of the new IFP, which is now further along the laser path. As a result, the field of view can be expanded to between 1.2 and 1.8 mm. Right: n TC2 configuration using a single plano- convex lens (L3) allows FOV expansion to 2.5 - 3.5 mm. h, complete nTC setup, including also an ETL positioned in front of the scan mirrors for rapid axial- scanning. PMTS, Photomultipliers. I, FOV expansion under nTC combines two effects: Increased focal distance (left) and reduced numerical aperture (N.A., right), which together give rise to a larger effective focal plane and enlarged PSFj, Power at sample measured for all configurations, expressed as a percentage of the power that reaches the scanning mirrors. m, point spread functions (PSFs) measured for all optical configurations, with size of typical neuronal somata of different species indicated. All scale- bars 10 um. n.o, lateral (n) and axial (o) spread of the PSFs quantified. Errors in s.d.. The specified numbers are for a Sutter MOM 2P microscope with Zeiss Objective W "Plan- Apochromat" 20x/1.0. Power at the sample plane was 0.35 mW. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 2
+ +Spatial resolution. a, Schematic (top) and photograph (bottom) of the setup used to directly film excitation volumes. b, effective scan- planes directly visualised as indicated in (a) for all optical configurations as indicated, in each case with scan- points spaced to facilitate inspection of individual PSFs. c. The same set of neurons of the 7 dpf larval zebrafish upper spinal cord (HuC:GCAMP6f, random sparse expression, see overview scan and schematic on the left) was imaged in all optical configurations + +<--- Page Split ---> + +as indicated at \(512 \times 512\) px (1 Hz). Arrowheads highlight the same synaptic structures in each scan. d-f, \(64 \times 64\) px (7.81 Hz) activity scan from fields of view shown in (c) for all five configurations during presentation of full-field flashes of UV-light which stochastic elicited activity in these imaged neural structures. In each case the average scan projection (d) and neighbour- correlation based activity projection (e) are shown (hereafter referred to as "activity- correlation"). Darker shadings, equalised for visibility, denotes increased local activity (for details, see Refo). Black traces in (f) show time-traces for the same structure in all cases. For the nTC2 2.5 mm FOV condition, time-traces from different neural structures are extracted to illustrate different responses in different structures. All activity traces in this and the following figures are shown in z-scores relative to their own baseline. We choose this metric over dF/F as it emphasises detectability of events rather than the relative change from the indicator's baseline fluorescence, which differs between biosensors. + +![](images/Figure_4.jpg) + +
Figure 3
+ +Rapid remote focussing. a,b, Scan- profiles with the electrically tunable lens (ETL) "flat" (zero input current, lowest profile) and engaged to achieved axially elevated scan planes at \(+300\) and \(+600\) pm (middle, upper profiles, respectively) in n TC1 (a) and n TC2 (b) configuration, as indicated. Associated size- changes in the effective full field of view were generally \(< 5\%\) (compare top and bottom planes). In each case, axial- shifts required \(< 25\%\) unidirectional peak current on the ETL which in turn facilitated rapid ETL- settling times: c,d, Schematic (c) and measured (d) axial jumps and settling time: the ETL was programmed to iteratively focus up and down by 150 um at each end of two long (5 ms) scan lines, as indicated. This enabled a direct read- out of ETL settling at each line- onset (oscillations in d). For the 150 um jumps shown, oscillations decayed below detectability within 2- 3 ms. For corresponding readouts of the ETL- position signal, see Fig. S3. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 4
+ +Mesoscale imaging of zebrafish larvae. a, Photograph of two 9 dpf zebrafish larvae mounted head- to- head in a microscope chamber with mm- scale ruler in background. B. The same 2 fish (Huc:GCAMP6f) as in (a) imaged under 2- photon with nTC2 3.5 mm FOV configuration, at \(512 \times 128\) px (3.91 Hz). c, d, Activity- correlation (cf. Fig. 2e) of the scan in (b) during presentation of full- field flashes of UV- light, with hand- selected exemplary ROIs, extracted time- traces (d) and light- stimulus- aligned averages (e). f- i, the same + +<--- Page Split ---> + +fish as shown on the left in (b, fish 1), now shown at full 3.5 mm field of view (f, 512x128 px, 3.91 Hz) and increased spatial resolution scans of regions as indicated to reveal cellular detail (9-1, 1,024x1,024 px, 0.49 Hz). + +![](images/Figure_6.jpg) + +
Figure 5
+ +3D random access scanning of the zebrafish eye and brain. a, b, Schematic of zebrafish larva from top (a) and front (b) with scan configurations indicated. c, direct x-z visualisation of the scan-profile used in + +<--- Page Split ---> + +the below. d, nTC1 1,024x1,024 px scan across an Islet2b:mGCAMP6f 6 dpf larval zebrafish eye and brain. At the centre of the scan, the axial focus is shifted upwards such that the axonal processes of retinal ganglion cells (RGCs) in the tectum (top) and their somata and dendritic processes in the eye (bottom) can be quasi- simultaneously captured, e, f, 1024x1024 px split- plane random access jump between tectum (e) and eye (f) and g-j, 2 times 64x128 px (15.6 Hz) random access scan of the same scan regions with raw (g) and event- averaged (h) fluorescence traces, mean image (i) and activity- correlation (cf. Fig. 2e). The stimulus was a series of full- field broadband flashes of light as indicated. k- o, as (d-j), with individual RGCs transiently expressing GCAMP6f under the same promoter. + +<--- Page Split ---> +![](images/Figure_7.jpg) + +
Figure 6
+ +2P plane- bending to image the in vivo larval zebrafish brain. a- c, Schematic of HuC:GCAMP6f larval zebrafish brain viewed from top (a) and front (b) with scan planes indicated, and (c) example- scan- profiles. d, nTC1 512x1024 scans of a 6 dpf zebrafish brain with different plane curvatures, with peak axial displacement at scan centre as indicated. At curvatures \(\sim 100 - 150 \mu m\) peak displacement the scan approximately traverses the surface of the tectum. e- h, mean (e), activity- correlation (f. cf. Fig. 2e) and + +<--- Page Split ---> + +fluorescence traces (g, raw and h, event- triggered mean) from a 170x340 px scan (5.88 Hz) of the 100 μm peak displacement configuration (image 3 in (d)). The fish was presented with full- field and spectrally broad (-360-650 nm) series of light- flashes. + +![](images/Figure_8.jpg) + +
Figure 7
+ +Mesoscale and random- access imaging in mouse brain slice. a,b, Schematic of brain (a) and transverse section (b) of a Thy 1:GCAMP6f mouse. c, d, 1024x1024 px n TC2 example scan of slice through cortex + +<--- Page Split ---> + +and hippocampus at maximal FOV (c) and nTC2 zoom in (d) as indicated. Red arrows indicate rapid transitions between scan regions, with the inset showing scan- profiles. The slice was bathed in an epileptogenic (high K+, zero Mg2+) solution to elicit seizures. e- g, Mean of 256x256 px scan (3.91 Hz) of (d) with regions of interest (ROIs) indicated (e), activity- correlation projection (Methods) indicating regions within the scan showing regions of activity computed as mean correlation of each pixel's activity over time to all its neighbours (for details, see Ref70) (f) and z- normalised fluorescence traces (g). h- l, 2 times 128x256 px (3.91 Hz) random access scan of two regions as indicated in (d) allows quasi- simultaneous imaging of the cortex (h) and hippocampus () at increased spatial resolution, with activity- correlation (j,k, cf. Fig. 2e) and fluorescence traces (1) extracted as in (k). + +<--- Page Split ---> +![](images/Figure_9.jpg) + +
Figure 8
+ +Mesoscale random- access imaging of mouse cortex in vivo. a, b, Schematic of Thy 1: GCamP6f mouse brain in vivo (a) with cranial window over the somatosensory cortex (b). c,d, \(1024 \times 1024\) px nTC2 (c) and nTC1 (d) images as indicated. Red arrows indicate rapid transitions between scan regions, with the inset indicating the scan- profile. e- i, 2 times \(128 \times 256\) px (3.91 Hz) random access scan as indicated in (d) with mean- projection (e,f), activity- correlation (g, h, cf. Fig. 2e) and fluorescence traces (i), taken from the ROIs + +<--- Page Split ---> + +as indicated in (g,h). j-l, NTC1 128x128 px xyz-tilted plane (7.82 Hz) traversing through cortical layers 1-4 at \(\sim 45^{\circ}\) relative to vertical with mean image (k) and activity-correlation (l, cf. Fig. 2e). + +![PLACEHOLDER_47_0] + +
Figure 9
+ +Multi- plane imaging and optogenetics for functional circuit mapping. a,b, DL (a) and NTC1 (b) 1,024x1,024 px scans of the ventral nerve cord of a 3rd instar VGlut:GCamP6f Drosophila larva. c-f, Scan- profiles taken in DL (C) and nTC1 (d) across 6 planes spaced \(\sim 15 \mu m\) apart. e, f, Schematic of first instar + +<--- Page Split ---> + +elav:GCamP6s; Ocro: CsChrimson Drosophila larva from top (e) and side (1), with CsChrimson (red) and GCAMP6s (green) expression pattern and scan- planes indicated. g- k optogenetic circuit mapping of olfactory processing centres across the larval brain. Six scan planes (170x340 px each) were taken at 0.98 Hz/plane (i.e. volume rate) during presentation of 587 nm light flashes (2 s) to activate CsChrimson in olfactory sensory neurons (OSNs). Brain anatomy (g) and false- colour coded fluorescence difference image (h, 1- 2 s after flash onset minus 1- 2 s prior to flash onset), with fluorescence activity traces (i, raw and j, event triggered average). For a zoom- in on the antennal lobe in a different specimen, see also SFig. 5. k, data from (h) summarised: top right; max- projection through the brain, with left and bottom showing transverse max- projections across the same data- stack. + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- JaniaketalSupplementaryRevision.pdf- VideoS13fish.mp4- VideoS2PSFexpansions.mp4- VideoS3ETL.mp4- VideoS42Fish.mp4- VideoS5Planebending.mp4- VideoS6Planebending2.mp4- VideoS7Mousebrainslice.mp4- VideoS8Mousecortex.mp4- VideoS9Drosophilaopticalcircuitmapping.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22_det.mmd b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..8cf1905e5038ed47f9e1482dfce1bccede3b3a1f --- /dev/null +++ b/preprint/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22/preprint__16e7144ba251ce7f335759692630e3727e0667fdf4717aef62639321ec367d22_det.mmd @@ -0,0 +1,567 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 939, 175]]<|/det|> +# Non-Telecentric 2P microscopy for 3D random access mesoscale imaging at single cell resolution + +<|ref|>text<|/ref|><|det|>[[44, 195, 592, 238]]<|/det|> +Filip Janiak ( F.K.Janiak@sussex.ac.uk) University of Sussex https://orcid.org/0000- 0002- 9295- 2740 + +<|ref|>text<|/ref|><|det|>[[44, 243, 234, 283]]<|/det|> +Philipp Bartel University of Sussex + +<|ref|>text<|/ref|><|det|>[[44, 290, 592, 333]]<|/det|> +Michael Bale University of Sussex https://orcid.org/0000- 0002- 5325- 1992 + +<|ref|>text<|/ref|><|det|>[[44, 338, 592, 380]]<|/det|> +Takeshi Yoshimatsu University of Sussex https://orcid.org/0000- 0002- 4939- 2020 + +<|ref|>text<|/ref|><|det|>[[44, 384, 234, 424]]<|/det|> +Emilia Komulainen University of Sussex + +<|ref|>text<|/ref|><|det|>[[44, 430, 234, 470]]<|/det|> +Mingyi Zhou University of Sussex + +<|ref|>text<|/ref|><|det|>[[44, 476, 592, 517]]<|/det|> +Kevin Staras University of Sussex https://orcid.org/0000- 0003- 4141- 339X + +<|ref|>text<|/ref|><|det|>[[44, 522, 630, 563]]<|/det|> +Lucia Prieto- Godino The Francis Crick Institute https://orcid.org/0000- 0002- 2980- 362X + +<|ref|>text<|/ref|><|det|>[[44, 568, 610, 610]]<|/det|> +Thomas Euler University of Tubingen https://orcid.org/0000- 0002- 4567- 6966 + +<|ref|>text<|/ref|><|det|>[[44, 615, 592, 656]]<|/det|> +Miguel Maravall University of Sussex https://orcid.org/0000- 0002- 8869- 7206 + +<|ref|>text<|/ref|><|det|>[[44, 661, 592, 703]]<|/det|> +Tom Baden ( T.Baden@sussex.ac.uk) University of Sussex https://orcid.org/0000- 0003- 2808- 4210 + +<|ref|>sub_title<|/ref|><|det|>[[44, 744, 101, 761]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 780, 857, 824]]<|/det|> +Keywords: two- photon (2P) microscopy, limitations, non- telecentric (nTC) optical design, three- dimensional field, single- cell resolution, imaging neuronal activity + +<|ref|>text<|/ref|><|det|>[[44, 841, 346, 860]]<|/det|> +Posted Date: December 15th, 2020 + +<|ref|>text<|/ref|><|det|>[[44, 879, 463, 899]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 121292/v1 + +<|ref|>text<|/ref|><|det|>[[44, 916, 909, 958]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 75, 940, 120]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on January 27th, 2022. See the published version at https://doi.org/10.1038/s41467-022-28192-0. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[118, 84, 852, 125]]<|/det|> +# Non-Telecentric 2P microscopy for 3D random access mesoscale imaging at single cell resolution + +<|ref|>text<|/ref|><|det|>[[118, 140, 866, 170]]<|/det|> +Janiak FK \(^{1\S}\) , Bartel P \(^{1}\) , Bale MR \(^{1}\) , Yoshimatsu T \(^{1}\) , Komulainen E \(^{1}\) , Zhou M \(^{1}\) , Staras K \(^{1}\) , Prieto- Godino LL \(^{2}\) , Euler T \(^{3,4}\) , Maravall M \(^{1}\) , Baden T \(^{1,3\S}\) + +<|ref|>text<|/ref|><|det|>[[118, 179, 840, 222]]<|/det|> +1: Sussex Neuroscience, School of Life Sciences, University of Sussex, UK; 2: The Francis Crick Institute, London, UK; 3: Institute of Ophthalmic Research, University of Tübingen, Germany; 4: Centre for Integrative Neuroscience, University of Tübingen, Germany. + +<|ref|>text<|/ref|><|det|>[[118, 234, 670, 249]]<|/det|> +\(\) \\(\) Correspondence at f.k.janiak@sussex.ac.uk and t.baden@sussex.ac.uk + +<|ref|>text<|/ref|><|det|>[[118, 261, 388, 277]]<|/det|> +8 Figures, 4 Supplementary Figures + +<|ref|>text<|/ref|><|det|>[[118, 284, 304, 299]]<|/det|> +7 Supplementary Videos + +<|ref|>text<|/ref|><|det|>[[118, 307, 630, 323]]<|/det|> +1 User Manual (available at https://github.com/BadenLab/nTCscope) + +<|ref|>text<|/ref|><|det|>[[118, 356, 880, 612]]<|/det|> +Summary. In neuroscience, diffraction limited two- photon (2P) microscopy is a cornerstone technique that permits minimally invasive optical monitoring of neuronal activity. However, most conventional 2P microscopes impose significant constraints on the size of the imaging field- of- view and the specific shape of the effective excitation volume, thus limiting the scope of biological questions that can be addressed and the information obtainable. Here, employing a non- telecentric (nTC) optical design, we present an ultra- low- cost, easily implemented and flexible solution to address these limitations, offering a several- fold expanded three- dimensional field of view that also maintains single- cell resolution. We show that this implementation also allows for straight- forward tailoring of the point- spread- function, increases effective excitation power, and achievable image brightness. Moreover, rapid laser- focus control via an electrically tunable lens allows near- simultaneous imaging of remote regions separated in three dimensions and permits the bending of imaging planes to follow natural curvatures in biological structures. Crucially, our core design is readily implemented (and reversed) within a matter of hours, and compatible with a wide range of existing 2P customizations, making it highly suitable as a base platform for further development. We demonstrate the application of our system for imaging neuronal activity in a variety of examples in zebrafish, mice and fruit flies. + +<|ref|>text<|/ref|><|det|>[[118, 648, 880, 746]]<|/det|> +Author contributions, FKJ and TB designed the study, with inputs from TE and all authors; FKJ implemented and tested hardware and software modifications, with input from PB, TY, TB and TE. FKJ and TB analysed the data, with inputs from all authors. PB assisted with hardware and software testing and troubleshooting and built the visual stimulator. MRB and MM provided mice for in vivo imaging and assisted with their handling and imaging. TY and MZ generated Islet2b:mGCaMP line and assisted with zebrafish sample preparation and testing. EK and KS provided mouse brain acute slice samples and assisted with handling and imaging. LLPG provided Drosophila sample and assisted with handing and imaging. TB built the optogenetics stimulator. FKJ and TB wrote the manuscript with inputs from all authors. + +<|ref|>text<|/ref|><|det|>[[118, 757, 880, 796]]<|/det|> +Acknowledgements. We thank Sabi Abdul- Raouf Issa for providing the VGlut:GCaMP6f Drosophila sample, and John Bear for helping with the generation of the Islet2b:mGCaMP6f line. The authors would also like to acknowledge support from the FENS- Kavli Network of Excellence and the EMBO YIP. + +<|ref|>text<|/ref|><|det|>[[118, 807, 880, 907]]<|/det|> +Funding. Funding was provided by the European Research Council (ERC- StG "NeuroVisEco" 677687 to TB, ERC- StG "EvolutioNeuroCircuit" 802531 to LLPG), The UKRI (BBSRC, BB/R014817/1 to TB, BB/S00310X/1 to KS, and MRC, MC_PC_15071 to TB and MM, MR/P006639/1 to MM and MR/P010121/1 to KS), the Leverhulme Trust (PLP- 2017- 005 to TB), the Lister Institute for Preventive Medicine (to TB), the Marie Curie Sklodowska Actions individual fellowship ("ColourFish" 748716 to TY) from the European Union's Horizon 2020 research and innovation programme, and the German Research Foundation (DFG) through Collaborative Research Center CRC 1233 (project number 276693517, to TE). LLPG's research was supported by the The Francis Crick Institute. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[260, 84, 405, 99]]<|/det|> +## INTRODUCTION + +<|ref|>text<|/ref|><|det|>[[260, 101, 880, 480]]<|/det|> +Laser scanning two photon (2P) microscopy allows the imaging of live cellular processes deep inside intact tissue with high signal- to- noise, temporal fidelity and spatial resolution1. Nonetheless, standard diffraction- limited 2P setups with a collimated laser excitation beam have several key characteristics that constrain their broad applicability; namely, a typically small field of view (FOV), a fixed- size excitation spot and restricted options for rapid random access 3- dimensional scans. These are significant limitations because the biological samples that are interrogated with 2P microscopy can exhibit substantial variations in size and spatial structure. For example, the volume of an adult mouse brain is approximately four orders of magnitude larger than that of a larval zebrafish, and seven orders of magnitude larger than a first instar larval fruit fly (Fig. 1a). Similarly, neuronal sub- structures are also highly variable in density and size, ranging from sub- micron levels for some synapses up to \(20 \mu \mathrm{m}\) or more for some somata. Additionally, neural densities vary by more than an order of magnitude across different animal brains2. As such, 2P microscopy tends to reveal very different levels of detail and organization across its diverse experimental applications. To maximize biological information, upgrades for 2P microscopy should enable the imaging of neuronal activity from many neural structures of a given size and density across a sufficiently large 3D volume of tissue at sufficiently high frame rates for the chosen neuronal process and biosensor. + +<|ref|>text<|/ref|><|det|>[[260, 481, 880, 844]]<|/det|> +In response to this demand, a profusion of custom modifications to 2P microscopes have been developed to expand the spatial and temporal boundaries over which neural structures can be optically interrogated. For example, the maximal planar field of view (FOV) has been increased from typically \(0.5 \mathrm{mm}\) to between \(3.1 - 10 \mathrm{mm}\) diameter by the exchange ( \(3 \mathrm{mm}\) : Ref3, \(7 \mathrm{mm}\) : Ref4) or size- increase of optical components ( \(10 \mathrm{mm}\) ), custom built objectives ( \(3.1 \mathrm{mm}\) ), enhanced scan engines ( \(5 \mathrm{mm}\) )7 and a mesoscope configuration ( \(5 \mathrm{mm}\) )8 to allow 'mesoscale' interrogation of neural circuits. In parallel, customizations using multiple beams have allowed simultaneous scanning of distant brain regions6,9,10. Likewise, higher temporal resolutions have been achieved by tailoring the point spread function (PSF) to the geometry and distribution of the neuronal structures of interest, thus increasing signal- to- noise ratio (SNR) and, in turn, decreasing the minimally- required dwell time per pixel11,12. Moreover, the imaging plane has been axially expanded by engineering an excitation spot with Bessel focus13,14 or by elongated Gaussian foci stereoscopy15. These customizations provide efficient ways to merge image structures that are located at different depths into a single volumetric plane. Furthermore, in recent years, systems integrating acousto- optic deflectors16,17, electrical tunable lenses9,18- 21 and remote focusing units8,14 have enabled quasi- simultaneous multiplane volumetric scans. + +<|ref|>text<|/ref|><|det|>[[260, 845, 880, 912]]<|/det|> +These types of extensions have been essential in driving the field forward, yet many are expensive, require custom- produced optical elements, complex optical alignment and/or introduce new limitations. The latter can include limitations in both excitation (e.g. power loss8, wavefront + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 83, 880, 170]]<|/det|> +dispersion \(^{17}\) ) and collection \(^{4,5}\) . Here we introduce an alternative design for 2P microscopy that overcomes many of these limitations while simultaneously approaching the capabilities of a wide range of state- of- the- art performance customisations, while being ultra- low- cost, simple and flexible. + +<|ref|>text<|/ref|><|det|>[[260, 187, 880, 532]]<|/det|> +Our non- telecentric (nTC) design, implemented for \(\sim \xi 1,000\) on an existing 2P setup equipped with a standard 20X objective, allows the expansion of the planar FOV from typically \(\sim 0.5 \mathrm{mm}\) in diameter to anywhere up to 3.5 mm to flexibly suit experimental needs (Fig. 1). This expansion is accompanied by a moderate and adjustable ("PSF tailoring" \(^{12}\) ) increase in the system's 3D PSF while maintaining single cell resolution over a wide range of biological applications. For example, unlike a standard diffraction limited (DL) setup (left in Fig. 1b,c), our nTC setup (right in Fig. 1b,c) allows simultaneous imaging of three entire zebrafish brains (Fig. 1e), or about a third of the width of a mouse's brain, while in each case maintaining single cell resolution (Fig. 1f, Supplementary Video S1). The addition of an electrically tunable lens (ETL) then allows near- simultaneous sampling in distant brain regions separated in 3 dimensions. Crucially, our solution is both comparatively low- cost and easy to implement on any existing 2P setup without the need for complex optical calibration, thus facilitating its widespread adoption in the community. We anticipate that others will be able to build on our core optical design using existing and new modifications to further increase its capability in the future. We demonstrate the current performance of our system with a range of examples from zebrafish, mice and fruit flies. + +<|ref|>sub_title<|/ref|><|det|>[[261, 550, 350, 565]]<|/det|> +## RESULTS + +<|ref|>text<|/ref|><|det|>[[260, 568, 880, 774]]<|/det|> +Non- Telecentric optics for field of view expansion. In traditional laser scanning 2P microscopy (left in Fig. 1b,c,g), a diffraction limited (DL) PSF is generated to excite fluorophores in a typically sub- micron volume of tissue. Here, xy- scanning mirrors reflect the laser beam into a collimation system comprised of a scan and a tube lens. The collimated beam then enters the back aperture of a high numerical aperture (N.A.) objective \(^{22,23}\) to converge at parallel rays into a DL spot at focal distance \(^{24}\) . The Gaussian shape of the excitation beam dictates that it is not possible to perfectly match beam width to the objective's back aperture. Instead, the back aperture is typically overfilled with a factor of \(1 / \mathrm{e}^2\) as a compromise between maximising spatial resolution (i.e. small PSF size) and power transmission \(^{25}\) . + +<|ref|>text<|/ref|><|det|>[[260, 776, 880, 912]]<|/det|> +In contrast, our nTC design (middle and right in Fig. 1b,c,g) illuminates the objective's back- aperture with a decollimated and divergent beam. This leads to an increased angle of view as the light exits the objective's front aperture, such that the same angular scan- mirror movement leads to a larger absolute shift in the image plane – thereby greatly increasing the FOV. In parallel, this also alters the effective excitation numerical aperture (N.A.) to yield a larger- than- DL excitation spot (i.e. an elongated PSF) at greater focal distance. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[120, 75, 884, 720]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 725, 881, 904]]<|/det|> +
Figure 1 | Non-Telecentric beam optics in 2-photon microscopy. a, Schematics of larval Drosophila (left), larval zebrafish (centre) and adult mouse (right) with central nervous system highlighted (green) to illustrate size differences. Insets next to the mouse for direct size-comparison between these species on the same scale. b, Optical configurations of standard diffraction limited (DL, left) 2P setup with parallel laser beam entering objective's back aperture. Right, non-telecentric (nTC, middle, right) configurations use a still diverging laser beam instead. As a result, the field of view and focal distance are expanded, and the point spread function (PSF) elongates. These effects scale with the angle of divergence (compare nTC₁ and nTC₂). c, Schematic representations of typical neuronal soma in species shown in (a), as interrogated by 2P setups shown in (b), respectively. d, In vivo 7 dpf larval zebrafish (HuC::GCaMP6f) imaged with an out-of-the-box Sutter-MOM DL setup at full field of view (top) and when zoomed in to reveal individual neuronal soma (bottom) as indicated. e, same zebrafish as shown in (d), as well as two further zebrafish imaged using nTC₂ configuration at maximal field of view (top). Zooming in to the same area as in (d, bottom) nonetheless reveals cellular
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 83, 881, 333]]<|/det|> +detail (e, bottom). f, In vivo adult mouse cranial window over somatosensory cortex imaged with nTC2 maximal field of view (top) and when zoomed in as indicated (bottom). g, Left: Optical configuration of a standard DL setup with collimation system consisting of a scan lens and a tube lens to set- up an infinity collimated laser beam at the level of the objective's back aperture. Effective refractive power and relative distances of lenses indicated. The intermediary focal point (IFP) is immediately behind the scan lens (arrowhead). Middle: nTC1 configuration replaces the scan lens with a pair of plano- convex lenses (L1,2). The relative position of L2 to the tube lens defines the position of the new IFP, which is now further along the laser path. As a result, the field of view can be expanded to between 1.2 and 1.8 mm. Right: nTC2 configuration using a single plano- convex lens (L3) allows FOV expansion to 2.5 - 3.5 mm. h, complete nTC setup, including also an ETL positioned in front of the scan mirrors for rapid axial- scanning. PMTs, Photomultipliers. i, FOV expansion under nTC combines two effects: Increased focal distance (left) and reduced numerical aperture (N.A., right), which together give rise to a larger effective focal plane and enlarged PSF. j, Power at sample measured for all configurations, expressed as a percentage of the power that reaches the scanning mirrors. m, point spread functions (PSFs) measured for all optical configurations, with size of typical neuronal somata of different species indicated. All scale- bars 10 μm. n.o, lateral (n) and axial (o) spread of the PSFs quantified. Errors in s.d.. The specified numbers are for a Sutter MOM 2P microscope with Zeiss Objective W "Plan- Apochromat" 20x/1.0. Power at the sample plane was 0.35 mW. + +<|ref|>text<|/ref|><|det|>[[261, 346, 880, 432]]<|/det|> +The magnitudes of each of these effects scale with the angle of divergence as the beam enters the back aperture of the objective. Accordingly, simply shifting the plano- convex lenses up or down the laser path, or switching between different refractive power lenses, provides for easy control over the system's optical properties to flexibly suit the user's needs. + +<|ref|>text<|/ref|><|det|>[[261, 450, 880, 501]]<|/det|> +In the following we show that the use of nTC in 2P microscopy brings about important advantages over the traditional, collimated and diffraction limited (DL) design: + +<|ref|>text<|/ref|><|det|>[[260, 518, 880, 551]]<|/det|> +1. The total field of view (FOV) can be expanded several-fold to suit the user's needs. + +<|ref|>text<|/ref|><|det|>[[260, 570, 880, 621]]<|/det|> +2. Scan-mirror movements translate into correspondingly larger xy-shifts in the image plane, meaning that even multi-millimetre random access jumps can be achieved with millisecond precision. + +<|ref|>text<|/ref|><|det|>[[260, 639, 880, 673]]<|/det|> +3. The addition of an electrically tunable lens (ETL) in front of the scan mirrors allows for similarly extensive expansion in the axial dimension. + +<|ref|>text<|/ref|><|det|>[[260, 691, 880, 725]]<|/det|> +4. The simplified optical path and under-filling of the objective's back aperture means more laser power is available at the sample plane. + +<|ref|>text<|/ref|><|det|>[[260, 743, 880, 794]]<|/det|> +5. It allows flexible and partially FOV-independent PSF-shape adjustment for imaging neurons of different size to individually optimise detection sensitivity for different biological samples2,11. + +<|ref|>text<|/ref|><|det|>[[260, 812, 880, 846]]<|/det|> +6. The increased working distance provides additional space for access to the preparation, for example with electrodes or stimulation equipment. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 879, 135]]<|/det|> +7. Combining points 4-6, nTC can yield an overall signal boost of up to \(\sim 18\) -fold with an above-stage detector (or \(\sim 4 - 5\) -fold at equal laser power on the sample), and more if a substage-collector is added. + +<|ref|>text<|/ref|><|det|>[[260, 145, 880, 213]]<|/det|> +We first discuss the required optical modifications and their impact on key excitation parameters (Figs. 1- 3, Figs. S1,2), before presenting a series of key use cases of different configurations for the interrogation of neural structure and function across diverse models (Figs. 4- 9, Figs. S3- 5). + +<|ref|>sub_title<|/ref|><|det|>[[260, 230, 759, 247]]<|/det|> +## PART I. OPTICS, EXCITATION AND OPTICAL SAMPLING + +<|ref|>text<|/ref|><|det|>[[259, 264, 880, 559]]<|/det|> +A simple scan- lens modification yields up to 7- fold FOV expansion. An off- the- shelf infinity- corrected galvo- galvo Sutter- MOM setup equipped with a 20x objective (Zeiss Objective W "Plan- Apochromat" 20x/1.0) offers a FOV diameter of \(\sim 0.5 \mathrm{mm}\) (left in Fig. 1g). However, when underfilling the back aperture of the objective with a diverging laser (middle and right in Fig. 1g), the beam exits the objective front aperture at increasingly obtuse angles at an effectively decreased N.A. (Fig 1i, Fig. S1a) and comes into focus at a greater distance (Fig. 1g, Fig. S1b). Together, this expands the effective excitation FOV in both xy (increased angle and decreased N.A.) and z (elongated PSF). To achieve this effect, it is necessary to bring the collimated laser beam, having passed the scan mirrors, to an "early" intermediary focal point (IFP) prior to reaching the objective, thus setting up the diverging beam thereafter (Fig. 1g, arrowheads). The specific divergence angle as the beam enters the back- aperture of the objective, which depends on IFP, defines the magnitude of the above- mentioned effects. We present two simple optical solutions (nTC1 and nTC2) to set- up an early IFP and thus expand the effective FOV to varying degrees. + +<|ref|>text<|/ref|><|det|>[[259, 560, 880, 905]]<|/det|> +In the standard DL configuration, the scan- lens (SL) and tube lens (TL) are separated from each other at a distance that is equal to their combined focal lengths \((50_{\mathrm{SL}} + 200_{\mathrm{TL}} \mathrm{mm} = 250 \mathrm{mm})\) to collimate the beam (left, Fig. 1g). In nTC1, we removed SL and instead inserted two off- the- shelf plano- convex lenses (L1, modified VISIR 1534SPR136, Leica; L2, LA1229 Thorlabs) with focal lengths 190 and 175 mm, respectively (middle, Fig. 1g, Methods). L1 was fixed 190 mm in front of TL to set up an IFP exactly at the TL. Next, L2 was positioned between L1 and TL to further increase laser convergence and thus shift the exact position of IFP away from the TL. Accordingly, IFP is always in front of the TL, with L2 determining its exact position: Simply shifting L2 along the laser path between 100 and 5 mm distance from the TL expanded the effective FOV diameter to anywhere between 1.2 and 1.8 mm, respectively (compare Fig. 1g, middle). In nTC2 (Fig. 1g, right), we replaced SL with a single lens (L3) of 200 mm focal length (LA1708, Thorlabs). L3 operated in much the same way as L2 in the previous modification M1, however now the IFP was behind rather than in front of TL. Depending on the position of L3, this yielded effective FOV diameters anywhere between 2.5 and 3.5 mm. Importantly, in each case effective image brightness remained approximately constant across the full FOV (Fig. S1c- f, Methods). Here, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 879, 170]]<|/det|> +the marginal brightness increase towards the edges is related to the slight upwards bend in the imaging plane as commonly seen for large FOV 2P microscopes \(^{5,8}\) – see also below. The axial difference between the edge and centre of the imaging plane was 20, 45, 87 and 170 μm for 1.2, 1.8, 2.5, 3.5 mm FOV, respectively. + +<|ref|>text<|/ref|><|det|>[[260, 170, 879, 325]]<|/det|> +While these specific lens configurations readily work for the commercially available Sutter MOM, the fundamental concept of setting up an IFP to yield a diverging beam is readily applicable to any standard 2P microscope, provided the optical path between the scanning mirrors and tube lens is accessible. In fact, 2P scan- lenses tend to consist of multiple custom- designed optical elements which by themselves easily exceed the cost of our solution. Accordingly, if provided directly by the microscope's manufacturer, our simplified optics should decrease the cost of such an off- the- shelf system. + +<|ref|>text<|/ref|><|det|>[[260, 326, 879, 499]]<|/det|> +Our design's full optical path and control logic are shown in Fig. 1h. All functions are executed from the scan software, which directly controls the xy- scan path as usual. To synchronize an electrically tunable lens (ETL, see below) and/or a Pockels cell to this xy- scan, a copy of the fast- mirror command is sent to two microcontrollers. Each of these then executes preloaded line- synchronized commands that are defined using a standalone graphical user interface (GUI). In this way, this standalone z- control- system only requires a copy of the scan mirror command, meaning that it can be directly added to any 2P microscope setup without the need for software modifications. + +<|ref|>text<|/ref|><|det|>[[260, 515, 879, 704]]<|/det|> +Increased effective laser power. Because our nTC design avoids overfilling of the objective's back aperture and uses fewer optical elements in the laser path, total laser power at the sample was increased approximately 4- fold compared to all configurations of the DL setup (Fig. 1j). This additional power could, for example, be used to facilitate imaging deep in the brain, or alternatively to drive additional setups from the same laser source. For instance, when imaging the small brains of larval zebrafish or fruit flies, there is rarely a need to exceed 50 mW, meaning that it is theoretically possible to drive ten such nTC setups from a single standard laser (e.g. Coherent Chameleon Vision- S Laser, average power \(\sim 1.5\) W at 930- 960 nm, assuming 50% loss through the setup). + +<|ref|>text<|/ref|><|det|>[[260, 721, 879, 911]]<|/det|> +Spatial resolution under nTC. To establish how our nTC approach affected the excitation PSF, we first imaged 175 nm fluorescent beads across all configurations at 927 nm wavelength and constant laser power at the sample (Methods, Supplementary discussion). Starting from a DL spot- volume of 0.56 and 3.15 μm (xy and z, respectively), our different modifications elongated and laterally expanded the PSF to varying degrees, from 0.77 (xy) and 9.94 (z) μm for the 1.2 mm FOV configuration to 2.21 (xy) and 41.49 (z) μm at 3.5 mm FOV (Fig. 1m- o, cf. Fig. S1g- i). Accordingly, increasing the FOV using nTC mainly elongated the PSF, while restricting its lateral expansion to remain principally suitable for providing single cell resolution even for the largest 3.5 mm expansion. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 83, 880, 326]]<|/det|> +However, PSF expansions were generally stronger than in other large FOV 2P- approaches which, for example, reported \(\sim 15 \mu \mathrm{m}\) at the edge of a 10 mm FOV \(^{5}\) or \(< 10 \mu \mathrm{m}\) at the edge of a \(\sim 5 \mathrm{mm}\) FOV \(^{8}\) , see also \(^{5,6}\) . These approaches achieve their optical results through custom made, large- diameter optics, which are generally more expensive and more difficult to retrofit to existing setups. Notably, beyond the microscope's optics itself, PSF- dimensions are affected by a myriad of additional factors such as laser wavelength and power (SFig. 1g- i) as well as the specific measurement method (e.g. bead types). Accordingly, directly comparing their dimensions across studies remains difficult. Notwithstanding, the possibility of optically merging adjacent image structures strongly depends on the size and spatial distribution of labelled biological structures - a general limitation in optical microscopy, rather than a specific limitation to our nTC approach (discussed e.g. in Ref \(^{26}\) ). + +<|ref|>text<|/ref|><|det|>[[260, 343, 880, 619]]<|/det|> +To further assess how the different optical configurations impacted PSF- shapes across the whole FOV, we next visualised excitation volumes using a camera (Fig. 2a,b) \(^{27}\) . Specifically, we positioned a fluorescein- solution- filled cuvette below the objective and filmed it from the side (Fig. 2a). Compared to imaging beads (Fig. 1m- o) this approach had the advantage that excitation volumes could be visualised much more directly, as well as across different positions in space in rapid succession (Supplementary Video 2). Fig. 2b shows a direct, scale- matched visualisation of effective scan profiles for all optical configurations. This confirmed that the DL configuration had the smallest PSFs, followed by increasing- FOV variations of nTC \(_{1,2}\) . Moreover, scan- profiles were curved to different degrees, with correspondingly tilted PSFs towards the edge \(^{5,8}\) . If required, this can be part- corrected via the ETL. However, biological structures are rarely perfectly flat either. As described further below, often a more useful solution might be to instead fit the scan- plane curvature to the 3D curvature of the interrogated sample. + +<|ref|>text<|/ref|><|det|>[[260, 636, 880, 911]]<|/det|> +Next, we directly compared the resultant effective spatial resolutions by imaging the same sample in each configuration. For this, we sparsely expressed GCaMP6f under the pan- neuronal promotor HuC in larval zebrafish \(^{28}\) and imaged one animal that randomly exhibited sparse and easily recognisable expression in neurons of the upper spinal cord, including one cell body ( \(\sim 7 \mu \mathrm{m}\) diameter) and several individual synapses ( \(\sim 1 \mu \mathrm{m}\) diameter, arrowheads indicate matching position of 2 such synapses across scans) (Fig. 2c). These image structures were consistently recognisable across all optical configurations, demonstrating that that even with nTC \(_{2}\) , single cell resolution could be readily achieved. This notion was further confirmed in functional scans during visual stimulation that was time- interweaved with the scanner retrace to avoid crosstalk (Fig. 2d- e, Methods). For example, full- field UV- flashes elicited different responses in different image structures on a trial- to- trial basis (e.g. see highlighted traces of nTC \(_{2}\) condition). Taken together, single cell resolution was readily preserved across all optical configurations. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 884, 825]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 832, 881, 913]]<|/det|> +
Figure 2 | Spatial resolution. a, Schematic (top) and photograph (bottom) of the setup used to directly film excitation volumes. b, effective scan-planes directly visualised as indicated in (a) for all optical configurations as indicated, in each case with scan-points spaced to facilitate inspection of individual PSFs. c, The same set of neurons of the 7 dpf larval zebrafish upper spinal cord (HuC:GCaMP6f, random sparse expression, see overview scan and schematic on the left) was imaged in all optical configurations as indicated at 512x512 px (1 Hz). Arrowheads highlight the same
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 880, 237]]<|/det|> +synaptic structures in each scan. d- f, 64x64 px (7.81 Hz) activity scan from fields of view shown in (c) for all five configurations during presentation of full- field flashes of UV- light which stochastic elicited activity in these imaged neural structures. In each case the average scan projection (d) and neighbour- correlation based activity projection (e) are shown (hereafter referred to as "activity- correlation"). Darker shadings, equalised for visibility, denotes increased local activity (for details, see Ref). Black traces in (f) show time- traces for the same structure in all cases. For the nTC2 2.5 mm FOV condition, time- traces from different neural structures are extracted to illustrate different responses in different structures. All activity traces in this and the following figures are shown in z- scores relative to their own baseline. We choose this metric over dF/F as it emphasises detectability of events rather than the relative change from the indicator's baseline fluorescence, which differs between biosensors. + +<|ref|>text<|/ref|><|det|>[[260, 250, 880, 561]]<|/det|> +PSF- tailoring. The systematic effects on PSF shape across configurations also meant that our nTC approach could be used to flexibly tailor PSF dimensions to specific experimental needs. This can be achieved by varying the degree of underfilling of the objective's back aperture while simultaneously keeping the laser's divergence angle approximately constant. We demonstrate this principal capability by setting up a "high- resolution" (small PSF) variant of nTC1 (Fig. S2). In general, such PSF- tailoring is useful for balancing the spatial resolution with the SNR. For example, the sub- micron DL PSF offered by typical collimated 2P- setups maximises spatial resolution which is invaluable for resolving small synaptic processes or the somata of larval fruit flies (typically <5 μm). However, many species' cell bodies are much larger. For example, in the brain of larval zebrafish a very small DL PSF spatially typically oversamples the "mid- sized" ~5- 10 μm somata at the expense of a potentially substantial loss in SNR. This limitation can be avoided by nTC- mediated tailoring of the PSF (cf. Fig. 2). Similarly, for picking up somatic signals from cortical neurons in the mouse, a "10- fold expanded" ~5 μm PSF yields the best SNR12. + +<|ref|>text<|/ref|><|det|>[[260, 579, 880, 907]]<|/det|> +An increased image brightness. Beside shaping the FOV and PSF dimensions, our nTC approach also generally boosted image brightness and thus signal- to- noise ratio (e.g. Fig. 2c- f). Here, three main factors contribute: (i) total effective laser power at the sample plane, (ii) the spatial relationship of PSF shape to structure(s) in the sample, and (iii) the fraction of emitted photons that can be detected (i.e. collection). As discussed above, the ~4- fold increased effective laser power (Fig. 1j) and larger excitation volumes under nTC (Figs. 1m- o, 2a- c) generally served to boost the number of emitted photons available for collection in the first place. In contrast, the increased working distance of nTC (Fig. S1b) meant that correspondingly smaller fractions of these photons could collected by the above- stage objective used for excitation. Notwithstanding, signal- collection from below the sample, being independent of above- stage working distance, is approximately unaffected. To explore the balance between all these factors, we imaged a piece of fluorescein- soaked tissue paper under all optical configurations, in each case finding the same field of view, and used two independent detector systems: one above the stage, and another below (Fig. S2g- i). In this way, the sub- stage signal in isolation could be used to measure brightness approximately independent of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 880, 377]]<|/det|> +excitation working distance (Fig. S2i, blue), while comparison of this sub- stage reading with the corresponding above- stage reading could then be used to estimate the relative above- stage collection loss (Fig. S2i, green vs. blue.). This revealed that the relative signal loss above the stage was very small indeed for nTC1 (1.2 mm: \(\sim 3\%\) ; 1.8 mm: \(\sim 9\%\) ), but started to noticeably affect collection under nTC2 (nTC2 2.5mm: \(33\%\) , nTC2 3.5 mm: \(64\%\) ). However, in all cases, this signal loss was greatly outweighed by the increased overall availability of photons for collection in the first place (relative to DL: \(\sim 10 - 20\) - fold for nTC1 and 16- 18 for nTC2 and 2). Taking all effects together, the highest overall signal- boost if using only an above- stage detector was achieved under nTC1 1.8 mm configuration ( \(\sim 18\) - fold), followed by nTC2 2.5 ( \(\sim 10\) - fold), nTC1 1.2 mm ( \(\sim 9\) - fold) and finally nTC2 3.5 mm ( \(\sim 6\) fold). Correspondingly larger signal boosts incur if a substage- detector is added. Accordingly, even if keeping effective laser power on the sample constant by correspondingly tuning down laser power as it enters the microscope (factor \(\sim 4\) , cf. Fig. 1j), all nTC configurations serve to boost effective image brightness relative to a DL configuration. + +<|ref|>text<|/ref|><|det|>[[260, 394, 880, 722]]<|/det|> +Rapid axial scans. In addition to expanding the FOV, our de- collimated design also shifts the excitation point beyond the objective's nominal focal distance (Fig. S1b). The same optical effect can be exploited to drive rapid axial shifts in the excitation plane by introduction of an electrically tunable lens (ETL) early in the laser path (Fig. 3, Fig. S3, cf. Fig. 1h) \(^{18,21}\) . Specifically, we positioned an off- the- shelf ETL (EL- 16- 40- TC- 20D, Optotune) 200 mm in front of the first scan mirror and controlled it with a custom driver board (see user manual). In this position, already a minor deviation from the perfectly flat curvature at zero input current slightly converged the laser which, in turn, strongly shifted the effective z- focus below the objective. For example, in both nTC1 and nTC2, stepping the input current from zero to \(25\%\) (50 mA) gave rise to a \(\sim 600 \mu \mathrm{m}\) z- shift of the excitation plane (Fig. 3a,b). The use of only a small fraction of the ETL's full dynamic range enabled short turnaround times (1- 10 ms, depending on distance jumped, Fig. 3c,d, SFig. 3) and prevented overheating \(^{18,29}\) . If required, rapid synchronization of the ETL curvature with a Pockels cell for controlling effective laser power at the sample plane can compensate for any systematic variations in image brightness associated with increased penetration depth. + +<|ref|>text<|/ref|><|det|>[[260, 739, 880, 843]]<|/det|> +Taken together, our design therefore presents a low- cost ( \(\sim \mathbb{E}1,000\) , cf. user manual) and easily implemented solution to expand the FOV of any 2P microscope in three dimensions while maintaining image quality suitable for single cell resolution. In the following, we demonstrate how these capabilities can be exploited in a range of neurophysiological applications in larval zebrafish, as well as the mouse cortex and fruit fly brain. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 81, 877, 315]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[115, 323, 881, 460]]<|/det|> +
Figure 3 | Rapid remote focussing. a,b, Scan-profiles with the electrically tunable lens (ETL) "flat" (zero input current, lowest profile) and engaged to achieved axially elevated scan planes at \(+300\) and \(+600 \mu m\) (middle, upper profiles, respectively) in nTC1 (a) and nTC2 (b) configuration, as indicated. Associated size-changes in the effective full field of view were generally \(< 5\%\) (compare top and bottom planes). In each case, axial-shifts required \(< 25\%\) unidirectional peak current on the ETL which in turn facilitated rapid ETL-settling times: c,d, Schematic (c) and measured (d) axial jumps and settling time: the ETL was programmed to iteratively focus up and down by \(150 \mu m\) at each end of two long (5 ms) scan lines, as indicated. This enabled a direct read-out of ETL settling at each line-onset (oscillations in d). For the \(150 \mu m\) jumps shown, oscillations decayed below detectability within 2-3 ms. For corresponding readouts of the ETL-position signal, see Fig. S3.
+ +<|ref|>sub_title<|/ref|><|det|>[[260, 493, 840, 510]]<|/det|> +## Part II. IMAGING THE STRUCTURE AND FUNCTION OF NEURONS + +<|ref|>text<|/ref|><|det|>[[260, 528, 880, 666]]<|/det|> +Imaging zebrafish under 2P. Owing to their small size and transparent larval stage, zebrafish have become a valuable model for interrogating brain- wide neural circuit function \(^{30,31}\) . However, from tip to tail, the brain and spinal cord of a 7- 9 dpf larval zebrafish reaches about 3.5 - 4.5 mm, with the central brain occupying approximately 1.2 mm in length and 0.7 mm in width. This is too large to fit into the FOV of a typical DL 2P setup. As a consequence, studies routinely "tile- scan" the brain in sequential stages to provide brain- wide data \(^{32,33}\) . + +<|ref|>text<|/ref|><|det|>[[260, 667, 880, 874]]<|/det|> +On the other hand, the transparent body wall of larval zebrafish makes them well- suited for 1- photon selective- plane- illumination microscopy (1p- SPIM / "lightsheet microscopy"), which is not FOV- limited in the same way as 2P microscopy \(^{29,30,34}\) . However, 1p- SPIM and related techniques \(^{35}\) have a number of drawbacks, including constraints on achieving a homogenous image due to scattering and divergence of the excitation light with increasing lateral depth \(^{2}\) , limited access to tissues that are shadowed by strongly- scattering tissue such as the eyes \(^{36,37}\) and, critically, a direct and bidirectional interference between the imaging system itself and any light stimuli applied for studying zebrafish vision \(^{38}\) . These specific challenges could be readily addressed by our nTC 2P setup. To demonstrate this, we imaged a multiple larval zebrafish in a range of optical configurations. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 78, 841, 850]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 853, 881, 910]]<|/det|> +
Figure 4 | Mesoscale imaging of zebrafish larvae. a, Photograph of two 9 dpf zebrafish larvae mounted head-to-head in a microscope chamber with mm-scale ruler in background. B, The same 2 fish (HuC:GCaMP6f) as in (a) imaged under 2-photon with nTC2 3.5 mm FOV configuration, at 512x128 px (3.91 Hz). c,d, Activity-correlation (cf. Fig. 2e) of the scan in (b) during presentation of
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 84, 880, 141]]<|/det|> +full- field flashes of UV- light, with hand- selected exemplary ROIs, extracted time- traces (d) and light- stimulus- aligned averages (e). f- i, the same fish as shown on the left in (b, fish 1), now shown at full 3.5 mm field of view (f, 512x128 px, 3.91 Hz) and increased spatial resolution scans of regions as indicated to reveal cellular detail (g- l, 1,024x1,024 px, 0.49 Hz). + +<|ref|>text<|/ref|><|det|>[[260, 156, 880, 310]]<|/det|> +Mesoscale whole- zebrafish 2P imaging. First, we used the 3.5 mm configuration of nTC2 to capture the largest- possible FOV of two larval zebrafish facing each other. This configuration comfortably allowed simultaneous mesoscale imaging of two entire zebrafish brains, here responding to full- field flashes of UV- light (Fig. 4a- e, Supplementary Video 3). Alternatively, the same configuration could be used to capture the entire central nervous system of one fish in a single frame, including the brain and nearly up to the tip of the spinal cord (Fig. 2f). Zooming in throughout the sample enabled resolving cellular details (Fig. 2g- i). + +<|ref|>text<|/ref|><|det|>[[260, 312, 880, 398]]<|/det|> +Next, beyond mesoscale imaging, many studies of zebrafish neuronal function focus on either the brain or the spinal cord (rather than both). In this case, using the more highly resolved nTC1 configuration with 1.2 mm FOV may be preferable; this just about fits one full zebrafish brain at a time while comfortably providing single cell resolution, as demonstrated below. + +<|ref|>text<|/ref|><|det|>[[260, 414, 880, 899]]<|/det|> +3D random access scanning across the zebrafish eye and brain. In the nervous system, key functionally linked circuits are often separated in 3D space, representing a general problem for systems neuroscience. For example, the retinal ganglion cells of the zebrafish eye project to the contralateral tectum and pretectum, which are both axially and laterally displaced by several 100s of microns. Accordingly, it has been difficult to simultaneously record at both sites, for example to study how the output of the eye is linked to the visual input to the brain. To address this problem, we used our nTC1 configuration in synchronisation with the ETL to establish quasi- simultaneous 3D random access scanning of the zebrafish's retinal ganglion cells across both the eye and brain (Fig. 5a- c). For this we used an Islet2b:mGCaMP6f line which labels the majority of retinal ganglion cells in larval zebrafish. We first defined a slow, high- spatial resolution scan (512x512 px, 0.98 Hz) that captured the entire front of the head, however with a single z- jump at the centre of the frame to set- up a "staircase- shaped" scan- path (Fig. 5b,c). Here, empirical adjustment of the magnitude of the z- jump allowed us to identify the axonal processes of retinal ganglion cells in the brain, and their dendritic processes in the contralateral eye in the top and bottom of the same imaging frame, respectively. Based on this image, we next defined two scan regions for 3D random access scanning, one capturing a single plane across the tectum, while the other captured a smaller area of a subset of RGC dendrites and somata in the eye (Fig. 5d- f). Finally, we decreased the spatial resolution to 64x64 px to quasi- simultaneously image both regions at 7.81 Hz. This configuration allowed reliable recording of light- driven signals from individual RGC neurites across the eye and brain (Fig. 5g- j). Next, we repeated this experiment, however this time in zebrafish larvae that were transiently injected with Islet2b:mGCaMP6f plasmid. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 880, 802]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 806, 880, 901]]<|/det|> +
Figure 5 | 3D random access scanning of the zebrafish eye and brain. a,b, Schematic of zebrafish larva from top (a) and front (b) with scan configurations indicated. c, direct x-z visualisation of the scan-profile used in the below. d, nTC1, 1,024x1,024 px scan across an Islet2b:mGCaMP6f 6 dpf larval zebrafish eye and brain. At the centre of the scan, the axial focus is shifted upwards such that the axonal processes of retinal ganglion cells (RGCs) in the tectum (top) and their somata and dendritic processes in the eye (bottom) can be quasi-simultaneously captured. e,f, 1024x1024 px split-plane random access jump between tectum (e) and eye (f) and g-j, 2 times 64x128 px (15.6 Hz)
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[118, 83, 880, 141]]<|/det|> +random access scan of the same scan regions with raw (g) and event- averaged (h) fluorescence traces, mean image (i) and activity- correlation (j, cf. Fig. 2e). The stimulus was a series of full- field broadband flashes of light as indicated. k- o, as (d- j), with individual RGCs transiently expressing GCaMP6f under the same promoter. + +<|ref|>text<|/ref|><|det|>[[260, 155, 880, 293]]<|/det|> +These animals stochastically express mGCaMP6f in only a very small number of RGCs, making it possible in principle to identify the processes belonging to the same RGC in both the eye and brain. As a proof of principle, we present one such experiment where we could clearly image the processes of single RGCs at both sites (Fig. 5k- o). For this type of application, it will be important to optimise the genetic protocol to improve expression levels and thereby facilitate the identification of the same RGC's processes at both sites. + +<|ref|>sub_title<|/ref|><|det|>[[260, 311, 878, 328]]<|/det|> +## 3D plane-bending for imaging activity across the zebrafish brain. + +<|ref|>text<|/ref|><|det|>[[260, 328, 880, 741]]<|/det|> +During standard planar scans of the larval zebrafish brain, the powerful optical sectioning afforded by the 2P approach highlights the 3D curvature of distinct brain regions by cutting right across them (Fig. 6a- d). While it was possible to quasi- simultaneously image anywhere within the brain at high spatial resolution using nTC1, a planar scan grossly misrepresented the real 3D structure of the zebrafish brain (Fig. 6d, top panel). For example, the tectum in larval zebrafish is tilted upwards \(\sim 30^{\circ}\) , meaning that rather than either cleanly sampling across its retinotopically organized surface, or perpendicularly across its stacked functional layers, the planar image instead cut the tectum at an effective \(30^{\circ}\) angle to yield a mixture of both, thus confounding interpretation. To ameliorate these issues, we used a 3D curved scan plane by driving the ETL as a sqrt(cosine) function of the slow y- mirror command (Methods). This enabled z- curvature "halfpipe" scans that could be empirically fitted to follow the natural curvature of the brain, thereby closely capturing the functional anatomical organisation of the zebrafish brain (Fig. 6b- d, Supplementary Video S4). From here, we chose a single halfpipe plane that best followed the curvature of the two tecta and imaged this plane at 7.81 Hz (256x128 px, 1ms/line, Fig. 6e). We then presented spectrally broad full- field light stimulation. This allowed us to interrogate brain- wide visual function in response to arbitrary wavelength light (Fig. 6f- h). As required, the halfpipe scans could also be staggered for multiplane imaging at correspondingly lower image rates, including negative bends that surveyed the difficult- to- reach bottom of the brain between the eyes (Fig. S4, Supplementary Video S5). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 880, 870]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[117, 871, 884, 914]]<|/det|> +
Figure 6 | 2P plane-bending to image the in vivo larval zebrafish brain. a-c, Schematic of HuC:GCaMP6f larval zebrafish brain viewed from top (a) and front (b) with scan planes indicated, and (c) example-scan-profiles. d, nTC₁ 512x1024 scans of a 6 dpf zebrafish brain with different plane
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 880, 155]]<|/det|> +curvatures, with peak axial displacement at scan centre as indicated. At curvatures \(\sim 100 - 150 \mu m\) peak displacement the scan approximately traverses the surface of the tectum. e- h, mean (e), activity- correlation (f, cf. Fig. 2e) and fluorescence traces (g, raw and h, event- triggered mean) from a \(170 \times 340 px\) scan (5.88 Hz) of the \(100 \mu m\) peak displacement configuration (image 3 in (d)). The fish was presented with full- field and spectrally broad ( \(\sim 360 - 650 nm\) ) series of light- flashes. + +<|ref|>text<|/ref|><|det|>[[260, 170, 880, 342]]<|/det|> +Mesoscale and 3D random access imaging of the mouse brain. The width of the adult mouse's brain is \(\sim 10 \text{mm}^{40}\) which makes it too large to be comprehensively captured by conventional 2P microscopy. Here, an experimental goal might be to reliably resolve the \(\sim 20+ \mu m\) somata of major cortical or subcortical neurons across a \(10 \text{mm FOV}\) . At the Nyquist detection limit, this would "only" require \(\sim 1,000\) pixels across, which is well within the range of standard high- resolution scan- configurations. Accordingly, currently the main limitation in achieving this goal is the microscope's maximal FOV. Our nTC design makes important steps to address this limitation. + +<|ref|>text<|/ref|><|det|>[[260, 357, 880, 914]]<|/det|> +When configured for a \(3.5 \text{mm FOV}\) (nTC2), our setup captures about a third of the width of a mouse's brain. In this configuration, a scan of a transverse section from a Thy1:GCaMP6f mouse (Fig. 7a,b) illustrates how the objective's back aperture casts a shadow at the image edge, thus limiting the spatial extent of the scan (Fig. 7c). Within this maximal window, a high- resolution \(1,024 \times 1,024 px\) scan allowed us to resolve the somata of major cortical and hippocampal neurons (Fig. 7d, Supplementary Video S6). Accordingly, at this largest FOV configuration, effective signal detection largely sufficed to capture the mouse brain's major neuron populations. However, with our galvo- galvo setup, scan rates at this level of spatial detail were slow (0.49 Hz, 2 ms/line). Accordingly, we used a mesoscale imaging approach with reduced spatial sampling (256x256 px, 1 ms/line) to capture the entire image at 3.91 Hz. This permitted simultaneous population- level "brain- wide" recording of seizure- like activity across the cortex and underlying hippocampus following bath application of an epileptogenic (high K+, zero Mg2+) solution (Fig. 7e- g). To demonstrate the value of the system for more detailed readout of neuronal activity, we also used random access scans to simultaneously capture distant smaller scan- fields at high resolution, both spatially and temporally (two times 256x128 px at 3.91 Hz, Fig. 7d, h- l, Supplementary Video S6). In the example provided, the laser travelled between the two scan fields separated by \(\sim 1 \text{mm}\) within two 1 ms scan lines. This allowed us to record quasi- simultaneous neural activity across both the cortex and hippocampus at single cell resolution. The generally high SNR in these recordings also suggested that additional temporal or spatial resolution could be gained by the use of resonance scanners in place of our galvos41. The large FOV nTC2 configuration also lends itself to imaging mouse cortical neurons activity in vivo (Fig. 8), an increasingly common demand in neuroscience. Here, the maximal 3.5 mm FOV captured an entire cranial window of a Thy1- GCaMP6f mouse prepared for optical interrogation of the somatosensory cortex, comprising an estimated \(10,000+\) neurons in a given image plane (Fig. 8a- c, Supplementary Video S7). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 732, 915]]<|/det|> + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[166, 82, 880, 249]]<|/det|> +Figure 7 | Mesoscale and random-access imaging in mouse brain slice. a,b, Schematic of brain (a) and transverse section (b) of a Thy1:GCalMP6f mouse. c,d, \(1024 \times 1024\) px \(nTC_2\) example scan of slice through cortex and hippocampus at maximal FOV (c) and \(nTC_2\) zoom in (d) as indicated. Red arrows indicate rapid transitions between scan regions, with the inset showing scan- profiles. The slice was bathed in an epileptogenic (high \(K^+\) , zero \(Mg^{2 + }\) ) solution to elicit seizures. e-g, Mean of 256x256 px scan (3.91 Hz) of (d) with regions of interest (ROIs) indicated (e), activity- correlation projection (Methods) indicating regions within the scan showing regions of activity computed as mean correlation of each pixel's activity over time to all its neighbours (for details, see Ref \(^{70}\) ) (f) and z- normalised fluorescence traces (g). h-l, 2 times 128x256 px (3.91 Hz) random access scan of two regions as indicated in (d) allows quasi- simultaneous imaging of the cortex (h) and hippocampus (i) at increased spatial resolution, with activity- correlation (j,k, cf. Fig. 2e) and fluorescence traces (l) extracted as in (j,k). + +<|ref|>text<|/ref|><|det|>[[260, 281, 880, 490]]<|/det|> +Even in an intermediate \(nTC_1\) configuration (in this case a 1.5 mm FOV) the full image still comprised several 1,000s of neurons (Fig. 8d), many more than could be simultaneously captured at scan- rates suitable for functional circuit interrogation with a galvo- galvo setup. In an example scan we again used a random- access approach to quasi- simultaneously record two \(330 \times 210 \mu m\) regions separated by \(\sim 1.2\) mm (two times 128x64 px at 7.81 Hz). As in the brain slice preparation (Fig. 7), this reliably resolved individual neurons in spatially distinct regions of the mouse brain (Fig. 8d- i). Finally, we also recruited the ETL to set up an axially tilted scan plane. This allowed quasi- simultaneous recording from neurons separated several hundreds of \(\mu m\) in depth across layers 1- 4 of the mouse cortex in vivo (Fig. 8j- l). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 884, 768]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[117, 777, 881, 890]]<|/det|> +
Figure 8 | Mesoscale random-access imaging of mouse cortex in vivo. a,b, Schematic of Thy1:GcAmP6f mouse brain in vivo (a) with cranial window over the somatosensory cortex (b). c,d, 1024x1024 px nTC2 (c) and nTC1 (d) images as indicated. Red arrows indicate rapid transitions between scan regions, with the inset indicating the scan-profile. e-i, 2 times 128x256 px (3.91 Hz) random access scan as indicated in (d) with mean-projection (e,f), activity-correlation (g,h, cf. Fig. 2e) and fluorescence traces (i), taken from the ROIs as indicated in (g,h). j-l, nTC1 128x128 px xyz-tilted plane (7.82 Hz) traversing through cortical layers 1-4 at \(\sim 45^{\circ}\) relative to vertical with mean image (k) and activity-correlation (l, cf. Fig. 2e).
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 83, 880, 256]]<|/det|> +Multi- plane circuit mapping with optogenetics in Drosophila. Despite the generally enlarged FOV and concomitant increase in the PSF, our setup was still capable of resolving details of small neural processes in the <0.1 mm diameter nervous system of a first instar larval fruit fly. To assess the difference in image- resolution between our nTC setup and a DL- configuration, we first obtained anatomical scans from a third instar VGlut:GCaMP6f larva which expressed GCaMP6f in structurally well- defined neurons of the ventral nerve cord (Fig 9a,b). This revealed that while the DL image was clearly sharper (Fig. 9a), the nTC1 system nevertheless comfortably delineated individual somata (Fig. 9b). + +<|ref|>text<|/ref|><|det|>[[260, 255, 880, 792]]<|/det|> +Drosophila was an ideal preparation to demonstrate our system's capacity for multi- plane imaging for optogenetic functional circuit mapping (Fig. 9c- k). At the first larval stage, the height of the brain excluding the ventral nerve cord is in the order of \(\sim 60 - 70 \mu m\) . Assuming an axial capture of \(\sim 3 \mu m\) per plane in a DL configuration (cf. Fig. 10), comprehensively sampling from the whole brain would therefore require upwards of 20 planes (Fig. 9c). Here, the slightly elongated PSF of the nTC 1.2 mm configuration served as a useful compromise between spatial resolution and sampling density (Fig. 9d). To demonstrate the sampling that can be achieved under these conditions, we used a transgenic first instar larva that expressed the red- shifted optogenetic effector CsChrimson in all olfactory- sensory neurons (OSNs) on a background of pan- neuronal GCaMP6s (elav:GCaMP6s) (Fig. 9e,f). To reveal any potential bilateral crosstalk of olfactory signal processing across the brain's two hemispheres, one of the olfactory nerves was cut. We set up six image planes (six times 340x170 px), each separated by \(\sim 15 \mu m\) which together captured the entire brain across both hemispheres at \(\sim 1 \text{Hz}\) (Fig. 9d,f). In this configuration, presentation of 2 s flashes of red light from a scanline- synchronized 590 nm LED activated olfactory sensory neurons (OSNs). These in turn propagated the signal to higher processing centres, which we visualised as regionally restricted GCaMP6s responses in the brain (Fig. 9g- k, Supplementary Video S8). The most strongly activated region was the ipsilateral antennal lobe (AL) (see also SFig. 5) which is directly innervated by the still- intact OSNs. Similarly, the olfactory second order processing centres, the mushroom body and the lateral horn, showed clear ipsilateral activation. In addition to these three major olfactory centres and their connecting tracts (e.g. plane 3), further processes and somata across both the ipsi- and contralateral lobe were also activated. Taken together, despite the slight expansion of the DL excitation spot, our nTC setup nevertheless allowed us to delineate key structural and functional information in this small insect brain. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[115, 80, 880, 884]]<|/det|> + + +<|ref|>image_caption<|/ref|><|det|>[[115, 884, 880, 912]]<|/det|> +
Figure 9 | Multi-plane imaging and optogenetics for functional circuit mapping. a,b, DL (a) and nTC₁ (b) 1,024x1,024 px scans of the ventral nerve cord of a 3rd instar VGlut:GCamP6f Drosophila
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[117, 83, 881, 237]]<|/det|> +lava. c- f, Scan- profiles taken in DL (c) and nTC; (d) across 6 planes spaced \(\sim 15 \mu m\) apart. e,f, Schematic of first instar elav:GCamP6s; Ocro:CsChrimson Drosophila larva from top (e) and side (f), with CsChrimson (red) and GCaMP6s (green) expression pattern and scan- planes indicated. g- k optogenetic circuit mapping of olfactory processing centres across the larval brain. Six scan planes (170x340 px each) were taken at 0.98 Hz/plane (i.e. volume rate) during presentation of 587 nm light flashes (2 s) to activate CsChrimson in olfactory sensory neurons (OSNs). Brain anatomy (g) and false- colour coded fluorescence difference image (h, 1- 2 s after flash onset minus 1- 2 s prior to flash onset), with fluorescence activity traces (i, raw and j, event triggered average). For a zoom- in on the antennal lobe in a different specimen, see also SFig. 5. k, data from (h) summarised: top right: max- projection through the brain, with left and bottom showing transverse max- projections across the same data- stack. + +<|ref|>sub_title<|/ref|><|det|>[[261, 265, 378, 280]]<|/det|> +## DISCUSSION + +<|ref|>text<|/ref|><|det|>[[260, 283, 880, 627]]<|/det|> +The ongoing development of sophisticated optical probes to report on key biophysical events has increasingly raised the demand in neuroscience for high SNR and large FOV 2P microscopes. To date, however, these characteristics are almost exclusively limited to high- end and, inevitably, high- cost platforms. Here, we exploit the fact that in 2P microscopy there is no "traditional" collection plane, allowing us to deviate from the diffraction limited regime that is typically used in systems where the planes of excitation and collection must superimpose to avoid image blur. Instead, we propose a simple core modification of the laser path that allows upgrading an out- of- the- box DL 2P microscope into a system capable of performing high SNR and large- FOV volumetric scans while at the same time preserving single cell resolution. We demonstrate the capabilities of this system for interrogating dynamic events in the brains of a range of key model species that are already widely used in neuroscience research. Since the core modification only requires the user to swap the scan lens for one or two off- the- shelf lenses, it can be tested (and fully reversed) within a matter of hours without the need for optical re- alignment or calibration. We anticipate that the simplicity and cost- effectiveness of this solution and the significant enhancement in 2P imaging capabilities that it permits, will lead to its wide adoption by the neuroscience community. + +<|ref|>text<|/ref|><|det|>[[260, 644, 880, 799]]<|/det|> +Combining an nTC approach with existing custom 2P designs. The estimation of metrics that meaningfully compare the capability of our nTC design with other custom solutions is difficult, as these can depend strongly on the specific objective (N.A., back aperture size, working distance (focus)), its distance from the tube lens, and indeed the nature of the interrogated sample and the biological question itself. Rather, because our nTC approach fundamentally differs from traditional DL optics, it opens the possibility to further enhance the capabilities of existing custom 2P microscope designs. + +<|ref|>text<|/ref|><|det|>[[260, 817, 879, 903]]<|/det|> +A key benefit of our nTC approach is the flexibility that it offers. It can be seamlessly implemented on setups with galvanometric or resonant mirrors to work with a wide range of scan- strategies. Here, the "extra" optical magnification afforded by the FOV expansion means that scan- mirror and ETL movements translate to relatively larger xy or z- translations, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 880, 204]]<|/det|> +respectively, making it easy to rapidly execute complex and large- scale 3D scan- paths. Our approach can also be combined with existing setups that use rapid piezo- positioning of the objective for axial scans, although in this case the objective movement relative to the tube lens will generate small but systematic variations in FOV and PSF shape. Accordingly, the use of remote focussing before the scan- mirrors is likely to be preferable in most applications. + +<|ref|>text<|/ref|><|det|>[[260, 205, 880, 360]]<|/det|> +Like in most 2P designs, our use of a Gaussian beam does not permit the generation of a truly arbitrary PSF shape. Nevertheless, if used in combination with temporal focussing12,42 it would, in principle, be possible to modulate axial PSF expansion without strongly affecting lateral expansion, thus facilitating a greater range of PSF shapes. Similarly, an optimized design of the objective lens43 and other optical elements4 including the use of large diameter lenses to minimize aberrations5, could all be combined with our optical design to further enhance the quality of 2P excitation. + +<|ref|>text<|/ref|><|det|>[[260, 377, 880, 636]]<|/det|> +nTC and optical aberrations. In general, beyond the PSF expansion that results from bypassing the objective's infinity correction (Figs. 1,2, Fi. S1), the change from a standard 2P DL- setup to an nTC configuration does not bring about new types of aberrations. In short, chromatic aberrations (which necessitate complex optical corrections in 1P microscopy) do not apply in 2P microscopy, because the excitation laser is essentially monochromatic and collection is spatially invariant. Instead, spherical aberrations tend to be dominant in 2P microscopy, i.e. when peripheral and axial rays do not converge to a point44- 49. The optical element that has the largest impact here is the objective, which is not changed under nTC. Further monochromatic aberrations are mainly related to the sample structure and surrounding (immersion) medium itself. In the future, it will be useful to explore how adaptive optics can address many of the above points, including spherical aberrations as well as coma and astigmatism46 - see also Supplementary Discussion. + +<|ref|>text<|/ref|><|det|>[[260, 653, 880, 896]]<|/det|> +Axial signal integration. As well as permitting the tailoring of the PSF to a given biological application, the use of a non- DL excitation spot can also bring about additional benefits. First, the lower effective excitation N.A. produces a narrower light cone which is less likely to be scattered by tissue inhomogeneities50. Second, objects that are smaller than the focal excitation volume become dimmer, while objects that are similar in size or larger remain bright8,51. Third, PSF expansion also reduces photobleaching and photodamage which can have a more- than- quadratic intensity dependence52,53. For example, when using the large PSF of the 3.5 mm FOV configuration, it was possible to use up to 250 mW laser power without causing notable damage when imaging deep in the mouse cortex54. Here, calculations and experimental experience suggest that in general, our strategy of underfilling the objective's back aperture will greatly ameliorate photodamage48,52,53. Notwithstanding, any axial expansion in the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 878, 118]]<|/det|> +PSF must be balanced with potentially undesirable merging of distinct image structures separated in depth. + +<|ref|>text<|/ref|><|det|>[[260, 135, 880, 290]]<|/det|> +Conclusion. Taken together, our nTC approach offers key advantages over traditional DL 2P microscopy, including the capacity for an increased FOV, PSF- tailoring, rapid z- travel through minimal ETL commands, overall increased laser power at the sample plane, and reduced spherical aberrations48,55,56. Moreover, it can principally be combined with a wide range of existing customisations to further push the capabilities of 2P microscopy in general. At the same time, our nTC approach is cost effective and can be readily implemented on an existing DL setup with minimal need for optical alignments and calibration. + +<|ref|>sub_title<|/ref|><|det|>[[260, 326, 356, 341]]<|/det|> +## METHODS + +<|ref|>text<|/ref|><|det|>[[260, 344, 880, 411]]<|/det|> +User manual. A complete user manual for the nTC design, as well as a bill of materials (BOM), 3D printable lens holders and printed circuit board (PCB) designs are available online at https://github.com/BadenLab/nTCscope. + +<|ref|>text<|/ref|><|det|>[[260, 429, 880, 480]]<|/det|> +DL 2P microscope. Our setup was based on a Sutter MOM- type two- photon microscope (designed by W. Denk, MPI, Martinsried; purchased through Sutter Instruments) as described previously57. + +<|ref|>text<|/ref|><|det|>[[260, 498, 880, 895]]<|/det|> +Excitation path. The excitation beam was generated by a tuneable femtosecond Ti:Sapphire laser (Coherent Vision- S, 75 fs, 80 MHz, \(>2.5\) W). The laser passed an achromatic half- wave plate (AHWP05M- 980, Thorlabs) and was subsequently equally split to supply two independent 2P setups using a beam- splitter for ultrashort pulses (10RQ00UB.4, Newport). Next, the beam passed a Pockels cell (350- 80 with model 302 driver, Conoptics), a telescope (AC254- 075- B and AC254- 150- B, Thorlabs), and was finally reflected into the head part of Sutter MOM stage by a set of three silver mirrors (PF10- 03- P01). We used a pair of single- axis galvanometric scan mirrors (6215H, Cambridge Technology) which directed the beam into a 50 mm focal length scan lens (VISIR 1534SPR136, Leica) at a distance of 56.6 mm. A 200 mm focal length tube lens (MXA22018, Nikon) was positioned 250 mm further along the optical path. From here, the now collimated excitation beam was directed onto the xyz- movable head of the Eyecup scope57 which was controlled by a motorized micromanipulator (MP285- 3Z, Sutter Instruments). Here, the beam was reflected by two silver parabolic mirrors to pass the collection path dichroic mirror (T470/640rpc, Chroma) to finally slightly overfill the back aperture of the objective (Zeiss Objective W "Plan- Apochromat" 20x/1.0), thus creating a diffraction- limited excitation spot at the objective's nominal working distance of 1.8 mm. The distance between the tube lens and the objective's back aperture was 95 mm at the centre position of the xyz displacement mechanism, and the parabolic mirrors ensured that the + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 878, 117]]<|/det|> +optical excitation axes stayed aligned during movements of the microscope head. + +<|ref|>text<|/ref|><|det|>[[260, 135, 880, 306]]<|/det|> +Collection path. Collection was exclusively through the objective (except for Fig. S2g- i). For this, a dichroic mirror (T470/640rpc, Chroma) was positioned \(18\mathrm{mm}\) above the objective's back aperture to reflect fluorescence light into the collection arm. Here, a \(140\mathrm{mm}\) focal length collecting lens was followed by a \(580\mathrm{- nm}\) dichroic mirror (H 568 LPXR, superflat) to split the signal into two wavebands. The "green" and "red" channels each used a single- band bandpass filter (ET525/50 and ET 605/50, respectively, Chroma) and an aspheric condenser lens (G317703000, Linos) to focus light on a PMT detector chip (H10770PA- 40, Hamamatsu). + +<|ref|>text<|/ref|><|det|>[[260, 308, 880, 480]]<|/det|> +For collection efficiency measurements (Fig. S2g- i) an additional sub- stage collection path was installed. To facilitate comparison, all optical components were identical to the above- stage excitation as collection patch (with the exception of the lack of the above- objective dichroic in the sub- stage setup). For this, a second objective (Zeiss W "Plan- Apochromat" \(20\mathrm{x} / 1.0\) ) was focused on the sample plane, and the collimated fluorescence light was subsequently focused through aspheric condenser lens (G317703000, Linos) and single- band bandpass filter (ET525/50, Chroma) on the PMT detector chip (H10770PA- 40, Hamamatsu). + +<|ref|>text<|/ref|><|det|>[[260, 498, 880, 722]]<|/det|> +Image acquisition. We used custom- written software (ScanM, by M. Mueller, MPI, Martinsried and T. Euler, CIN, Tuebingen) running under IGOR pro 6.3 for Windows (Wavemetrics) to control the setup. For hardware- software communication we use two multifunction I/O devices (PCle- 6363 and PCI- 6110, National Instrument). Within ScanM, we defined custom scan- configurations: \(1,024\mathrm{x}1,024\) and \(512\mathrm{x}512\) pixel images with 2 ms per line were used for high- resolution morphology scans, while faster, 1 ms or 2 ms linespeed image sequences with \(256\times 256\) (3.91Hz), \(128\times 128\) (7.81 Hz), \(340\times 170\) (5.88 Hz) or \(128\times 64\) (15.6 Hz) pixels were used for activity scans. All scans were unidirectional, and the laser was blanked via the Pockels cell during the turnarounds and retrace. This period was also used for light stimulation (zebrafish visual system and Drosophila optogenetics, see below). + +<|ref|>text<|/ref|><|det|>[[260, 740, 880, 912]]<|/det|> +Non- collimated 2P microscope modifications. We used two sets of modifications (nTC1 and nTC2) to de- collimate the excitation path to different degrees. For nTC1 (FOV \(1.2 - 1.8\mathrm{mm}\) ) we modified the original Sutter- MOM scan lens (VISIR 1534SPR136, Leica) by removing the second lens (i.e. the one closer to the tube lens) from the compound mount which changed the focal length from 50 to \(190\mathrm{mm}\) . Alternatively, the entire de- constructed scan lens could also be replaced by a similar power off- the- shelf plano- convex lens. Our \(190\mathrm{mm}\) lens (L1) was placed exactly \(190\mathrm{mm}\) in front of the tube lens (so shifted \(60\mathrm{mm}\) forward from its original position). Next, we introduced an additional plano- convex \(175\mathrm{mm}\) focal + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 880, 187]]<|/det|> +distance lens (L2) (LA1229, Thorlabs). L2 was held in place by custom 3D printed mount (cf. user manual) inside the MOM's tube-lens holder and positioned anywhere between 0 and \(10\mathrm{mm}\) in front of the tube lens. Depending on the exact position of L2 within this range, the effective FOV at the image plane could be adjusted between \(1.2\mathrm{mm}\) ( \(10\mathrm{mm}\) distance) to \(1.8\mathrm{mm}\) (L2 and tube lens almost touching). + +<|ref|>text<|/ref|><|det|>[[260, 204, 880, 343]]<|/det|> +For \(n\mathrm{TC}_2\) (FOV \(2.5 - 3.5\mathrm{mm}\) ), we replaced the original scan lens with a single, \(200\mathrm{mm}\) focal length plano-convex lens L3 (LA1708, Thorlabs). Like L2 in \(n\mathrm{TC}_1\) , L3 was mounted on the same custom 3D printed holder and positioned anywhere within a distance of \(0 - 10\mathrm{mm}\) in front of the tube lens. In this case the FOV at the image plane could be adjusted between 2.5 mm ( \(10\mathrm{mm}\) distance) to \(3.5\mathrm{mm}\) (L3 and tube lens almost touching). For detailed instructions including photos of the optical path, consult the user manual. + +<|ref|>text<|/ref|><|det|>[[260, 344, 880, 482]]<|/det|> +We selected lens types and positions based on the available space within the Sutter MOM head such that for \(n\mathrm{TC}_1\) and \(n\mathrm{TC}_2\) , the IFP was always located in front of or behind the TL, respectively. However, depending on the design of a given 2P setup's excitation path, numerous alternative configurations are possible. Here, a straight- forward means to rapidly estimate the nature and scale of a given modification is to use a fluorescence test- slide and observe the change in working distance and FOV as the scan path is modified. + +<|ref|>text<|/ref|><|det|>[[260, 499, 880, 878]]<|/det|> +Electrically tunable lens (ETL) for rapid axial focussing. For rapid z- focussing we added a horizontal ETL (EL- 16- 40- TC- 20D, Optotune) into the vertical beam path after the silver mirror that reflected the excitation beam up into the MOM head, \(200\mathrm{mm}\) in front of the scan- mirrors. To drive the ETL we used a custom current driver controlled by an Arduino Duo microcontroller (see user manual), capable of generating positive currents between 0- 300 mA. The Arduino Duo received a copy of the scan- line command and in turn output commands to the current driver to effect line- synchronised changes in ETL curvature. Prior to initiating a scan, the specific to- be- executed Arduino programme was uploaded to the Arduino via serial from a PC running a custom Matlab- script (Mathworks). This Matlab script launched a simple graphical user interface (GUI) that allowed the user to configure the exact lens- path during a custom scan (see user manual). Accordingly, ETL control remained flexible and fully independent of the scan software. In this way, our solution can be readily integrated with any 2P system without need to change the software or acquisition/driver hardware. Notably, this ETL implementation can also be used by itself, without need for implementing any of the other optical adjustments described in this work. However, depending on the system's optics, the effective range of z- travel would likely be smaller. A detailed step- by- step guide to implement the ETL, including the control software and hardware is provided in the user manual. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 880, 291]]<|/det|> +Pockels cell. To control excitation laser intensity, we use a Pockels cell (Model 350- 80, Conoptics; driver model 302, Conoptics). A line- synchronised blanking signal was sent from the DAQ to the drive to minimise laser power during the retrace. In addition, a custom circuit allowed controlling effective laser brightness during each scan line via a potentiometer (see user manual, designed by Ruediger Bernd, HIH, University of Tübingen). As required, this amplitude- modulated signal could then be further modulated by a second Arduino Due controlled by a standalone Matlab GUI to automatically vary effective laser power as a function of scanline index. In this way, laser power could be arbitrarily modulated on a line by line basis, for example to compensate for possible power loss when imaging at increased depth. + +<|ref|>text<|/ref|><|det|>[[260, 308, 880, 515]]<|/det|> +Light stimulation. For visual stimulation of zebrafish larvae (Figs. 5, S2, 6) we used a full- field, broadband spot of light projected directly onto the eyes of the fish from the front via a liquid light guide (77555, Newport) connected to a custom collimated LED bank (Roithner LaserTechnik) with emission peak wavelengths between 650 and 390 nm to yield an approximately equal power spectrum over the zebrafish's visual sensitivity range (described in detail in Ref58). LEDs were line- synchronised to the scanner retrace by an Arduino Due. For CsChrimson activation (Fig. 7) we used a custom 2P line synchronised LED stimulator (https://github.com/BadenLab/Tetra- Chromatic- Stimulator) equipped with four 587 nm peak emission LEDs embedded in a custom 3D printed recording chamber. + +<|ref|>text<|/ref|><|det|>[[260, 532, 880, 636]]<|/det|> +Image brightness measurements. We imaged a uniform florescent sample consisting of two microscopy slides (S8902, Sigma- Aldrich) encapsulating a drop of low melting point agarose (Fisher Scientific, BP1360- 100) mixed with low concentrated Acid Yellow 73 fluorescein solution (F6377 Sigma- Aldrich). Show is the average brightness over the radius from the centre to the edge of the FOV (Matlab, custom scripts). + +<|ref|>text<|/ref|><|det|>[[260, 653, 880, 825]]<|/det|> +PSF measurements. We used \(0.175 \pm 0.005 \mu m\) yellow- green (505/515) fluorescent beads (P7220, Invitrogen) embedded in a 1 mm depth block of \(1\%\) low melting point agarose (Fisher Scientific, BP1360- 100). Image stacks were acquired across \(30 \times 30 \mu m\) lateral field of view with 256x256 pixels resolution (0.12 \(\mu m\) /pixel) and \(0.5 \mu m\) axial steps/frame. For xy and z- dimensions, we calculated the full width at half maximum (FWHM) from Gaussian fits to the respective intensity profiles. Measurements were taken from set of the beads distributed across the entire FOV, and presented results are averages of at least 10 measurements of different beads, with error bars given in s.d.. + +<|ref|>text<|/ref|><|det|>[[260, 827, 880, 912]]<|/det|> +To film the PSF and effective scan- plane(s) we focussed an air- objective (Plan Apo \(4 \times /0.20\) , Nikon) onto the excitation spot elicited in a plastic cuvette with fluorescein (F2456 Sigma- Aldrich) dissolved in water which was positioned beneath the excitation objective. The camera path was fitted with a single- band bandpass filter (ET525/50, Chroma) and a colour + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 84, 878, 118]]<|/det|> +CCD camera (Manta G- 031C, Allied Vision). The camera was controlled with its dedicated software (VIMBA, Allied Vision). + +<|ref|>text<|/ref|><|det|>[[260, 135, 880, 255]]<|/det|> +Animal experiments. All animal experiments presented in this work were carried out in accordance with the UK Animal (Scientific Procedures) Act 1986 and institutional regulations at the University of Sussex. All procedures were carried out in accordance with institutional, national (UK Home Office PPL70/8400 (mice), PPL/PE08A2AD2 (zebrafish)) and international (EU directive 2010/63/EU) regulations for the care and use of animals in research. + +<|ref|>text<|/ref|><|det|>[[260, 273, 880, 515]]<|/det|> +Zebrafish larvae preparation and in- vivo imaging. Zebrafish were housed under a standard 14:10 day/night rhythm and fed 3 times a day. Animals were grown in 200 mM 1- phenyl- 2- thiourea (Sigma) from 1 day post fertilization (dpf) to prevent melanogenesis. Preparation and mounting of zebrafish larvae was carried out as described previously59. In brief, we used 6- 7 dpf zebrafish (Danio rerio) larvae that were immobilised in 2% low melting point agarose (Fisher Scientific, Cat: BP1360- 100), placed on the side on a glass coverslip and submerged in fish water. For eye- brain imaging, eye movements were prevented by injection of a- bungarotoxin (1 nL of 2 mg/ml; Tocris, Cat: 2133) into the ocular muscles behind the eye. Transgenic lines used were Islet2b:mGCaMP6f (eye- brain imaging) and HuC:GCaMP6f28 (image of 3 zebrafish in same FOV). Zebrafish were imaged at 930 nm and 30- 60 or 50- 100 mW for brain and eye imaging, respectively. + +<|ref|>text<|/ref|><|det|>[[260, 532, 880, 755]]<|/det|> +Creation of Islet2b:mGCaMP6f transgenic line. Tg(is12b:nlsTrpR, tUAS:memGCaMP6f) was generated by co- injecting pTol2- isl2b- hlsTrpR- pA and pBH- tUAS- memGaMP6f- pA plasmids into single- cell stage eggs. Injected fish were out- crossed with wild- type fish to screen for founders. Positive progenies were raised to establish transgenic lines. All plasmids were made using the Gateway system (ThermoFisher, 12538120) with combinations of entry and destination plasmids as follows: pTol2- isl2b- nlsTrpR- pA; pTol2pA60, p5E- isl2b61, pME- nlsTrpR62, p3E- pA60; pBH- tUAS- memGaMP6f- pA; pBH63, p5E- tUAS62, pME- memGCaMP6f, p3E- pA. Plasmid pME- memGCaMP6f was generated by inserting a polymerase chain reaction (PCR)- amplified membrane targeting sequence from GAP- 4364 into pME plasmid and subsequently inserting a PCR amplified GCaMP6f65 at the 3' end of the membrane targeting sequence. + +<|ref|>text<|/ref|><|det|>[[260, 774, 880, 911]]<|/det|> +Acute brain slices. 1- 2 month old male Thy1- GCaMP6f- GP5.1766 mice were used. Acute transverse brain slices (300 μm) were prepared using a vibroslicer (VT1200S, Leica Microsystems, Germany) in ice- cold artificial cerebrospinal fluid (ACSF) containing (in mM): 125 NaCl, 2.5 KCl, 25 glucose, 1.25 NaH2PO4, 26 NaHCO3, 1 MgCl2, 2 CaCl2 (bubbled with 95% O2 and 5% CO2, pH 7.3), and allowed to recover in the same buffer at 37°C for 60 minutes67. During imaging, slices were constantly perfused with 37°C modified (epileptogenic) saline (37°C) containing 125 NaCl, 5 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[260, 85, 878, 119]]<|/det|> +KCl, 25 glucose, 1.25 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂. Brain slices were imaged at 930 nm and 100- 150 mW. + +<|ref|>text<|/ref|><|det|>[[260, 135, 880, 394]]<|/det|> +Mouse surgical procedures for in vivo- imaging of the barrel cortex. Head bar implantation surgery has been described elsewhere68. Briefly, under aseptic conditions, a male mouse expressing a calcium indicator in pyramidal neurons (GCaMP6f; GP5.17⁶⁶) was anaesthetised with isoflurane and implanted with a custom- made head bar. A circular 3 mm diameter craniotomy centred at 3.0 mm lateral and 1.0 mm posterior to bregma was made to expose the cranial surface. A cranial window, consisting of a 3 mm circular coverslip and a 5 mm circular coverslip (Harvard Apparatus), was placed over the craniotomy and secured in place with cyanoacrylate tissue sealant (Vetbond, 3M). Following 7 days of recovery, the mouse was handled daily and acclimated to a head fixation apparatus over a treadmill for a further 9 days. During 2P imaging, the head- fixed mouse could locomote freely on a custom- made treadmill. The mouse was awake and received fluid rewards between imaging batches. Cortical neurons were imaged at 960 nm and 100- 150 mW. + +<|ref|>text<|/ref|><|det|>[[260, 410, 880, 825]]<|/det|> +Drosophila larval preparation and in- vivo imaging. Flies were maintained at 25°C in 12 h light:12 h dark conditions. Fly stocks were generated using standard procedures. The genotypes of the D. melanogaster flies used were: elav- Gal4; LexAOp- CsChrimson and w; UAS- GCaMP6s; Orco- LexA. These two strains were crossed to each other (collecting virgins from the first one and males from the second one) and placed on laying- pots at 25°C for larval collection. The laying- pots had a grape juice agar plate with an added drop of yeast paste supplemented with all- trans retinal (Sigma- Aldrich) to a final concentration of 0.2 mM. Yeast supplemented agar plates were changed every day and first instar larvae were picked off the new changed plate. First instar larvae were collected from yeast supplemented agar plates and dissected on physiological saline as in Ref⁶⁹ (in mM): 135 NaCl, 5 KCl, 5 CaCl₂- 2H₂O, 4 MgCl₂- 6H₂O, 5 TES (2- [[1,3- dihydroxy- 2- (hydroxymethyl)propan- 2- yl]amino]ethanesulfonic acid), 36 Sucrose, adjusted to pH 7.15 with NaOH. Larvae were dissected to expose the brain while maintaining intact the anterior part of the animal and the connection between OSN cell bodies and the brain, subsequently one of the olfactory nerves was cut with the forceps. The preparation was then positioned on top of a coverslip coated with poly- lysine (Sigma- Aldrich, P1524- 100MG), and covered in 2% low melting point agarose (Fisher Scientific, Cat: BP1360- 100) diluted in physiological saline, to prevent movement associated with mouth- hook contractions. The sample was then submerged in physiological saline. Larval brains were imaged at 930 nm and 30- 60 mW. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 110, 468, 904]]<|/det|> +1130 1. Denk, W., Strickler, J. H. H., Webb, W. W. 1131 W., Series, N. & Apr, N. Two- Photon Laser 1132 Scanning Fluorescence Microscopy. 1133 Science (80- ). 248, 73- 76 (1990). 1134 2. Weisenburger, S. & Vaziri, A. A guide to 1135 emerging technologies for large- scale and 1136 whole brain optical imaging of neuronal 1137 activity. (2018). doi:10.1146/annurev- neuro- 072116- 031458 1139 3. Ota, K. et al. Fast scanning high optical 1140 invariant two- photon microscopy for 1141 monitoring a large neural network activity 1142 with cellular resolution. bioRxiv 1143 2020.07.14.201699 (2020). 1144 doi:10.1101/2020.07.14.201699 1145 4. Bumstead, J. R. Designing a large field- of- 1146 view two- photon microscope using optical 1147 invariant analysis. Neurophotonics 5, 1 1148 (2018). 1149 5. Tsai, P. S. et al. Ultra- large field- of- view 1150 two- photon microscopy. Opt. Express 1151 (2015). doi:10.1364/oe.23.013833 1152 6. Stirman, J. N., Smith, I. T., Kudenov, M. W. 1153 & Smith, S. L. Wide field- of- view, multi- 1154 region, two- photon imaging of neuronal 1155 activity in the mammalian brain. Nat. 1156 Biotechnol. 34, 857- 862 (2016). 1157 7. Yu #, C.- H., Stirman #, J. N., Yu, Y., Hira, 1158 R. & Smith, S. L. Diesel2p mesoscope with 1159 dual independent scan engines for flexible 1160 capture of dynamics in distributed neural 1161 circuitry. bioRxiv 2020.09.20.305508 1162 (2020). doi:10.1101/2020.09.20.305508 1163 8. Sofroniew, N. J., Flickinger, D., King, J. & 1164 Svoboda, K. A large field of view two- 1165 photon mesoscope with subcellular 1166 resolution for in vivo imaging. Elife (2016). 1167 doi:10.7554/elife.14472 1168 9. Han, S., Yang, W. & Yuste, R. Two- Color 1169 Volumetric Imaging of Neuronal Activity of 1170 Cortical Columns. Cell Rep. (2019). 1171 doi:10.1016/j.celrep.2019.04.075 1172 10. Cheng, A., Goncalves, J. T., Golshani, P., 1173 Arisaka, K. & Portera- Cailliau, C. 1174 Simultaneous two- photon calcium imaging 1175 at different depths with spatiotemporal 1176 multiplexing. Nat. Methods (2011). 1177 doi:10.1038/nmeth.1552 1178 11. Prevedel, R. et al. Fast volumetric calcium 1179 imaging across multiple cortical layers using 1180 sculpted light. Nat. Methods (2016). 1181 doi:10.1038/nmeth.4040 1182 12. Weisenburger, S. et al. Volumetric Ca 2+ 1183 Imaging in the Mouse Brain Using Hybrid 1184 Multiplexed Sculpted Light Microscopy. Cell 1185 (2019). doi:10.1016/j.cell.2019.03.011 1186 13. Lu, R. et al. Video- rate volumetric functional 1187 imaging of the brain at synaptic resolution. 1188 Nat. Neurosci. (2017). doi:10.1038/nn.4516 1189 14. Botcherby, E. J., Juškaitis, R. & Wilson, T. 1190 Scanning two photon fluorescence 1191 microscopy with extended depth of field. 1192 (2006). doi:10.1016/j.optcom.2006.07.026 1193 15. Song, A. et al. Volumetric two- photon 1194 + +<|ref|>text<|/ref|><|det|>[[476, 110, 880, 904]]<|/det|> +1195 imaging of neurons using stereoscopy (vtwins). Nat. Methods (2017). doi:10.1038/nmeth.4226 16. Grewe, B. F., Langer, D., Kasper, H., Kampa, B. M. & Helmchen, F. high- speed in vivo calcium imaging reveals neuronal network activity with near- millisecond precision. (2010). doi:10.1038/nmeth.1453 17. Chong, E. Z., Panniello, M., Barreiros, I., Kohl, M. M. & Booth, M. J. Quasi- simultaneous multiplane calcium imaging of neuronal circuits. Biomed. Opt. Express (2019). doi:10.1364/boe.10.000267 18. Grewe, B. F., Voigt, F. F., van 't Hoff, M. & Helmchen, F. Fast two- layer two- photon imaging of neuronal cell populations using an electrically tunable lens. Biomed. Opt. Express (2011). doi:10.1364/boe.2.002035 19. Yang, W., Carrillo- Reid, L., Bando, Y., Peterka, D. S. & Yuste, R. Simultaneous two- photon imaging and two- photon optogenetics of cortical circuits in three dimensions. Elife (2018). doi:10.7554/elife.32671 20. Sheffield, M. E. J. & Dombeck, D. A. Calcium transient prevalence across the dendritic arbor predicts place field properties. Nature (2015). doi:10.1038/nature13871 21. Zhao, Z. et al. The temporal structure of the inner retina at a single glance. bioRxiv 743047 (2019). doi:10.1101/743047 22. Denk, W. & Svoboda, K. Why multiphoton is more than a gimmick. Neuron (1997). doi:10.1016/S0896- 6273(00)81237- 4 23. Svoboda, K. & Yasuda, R. Principles of Two- Photon Excitation Microscopy and Its Applications to Neuroscience. Neuron (2006). doi:10.1016/j.neuron.2006.05.019 24. Born, M. & Wolf, E. Principles of Optics Electromagnetic Theory of Propagation, Interference and Diffraction of Light. Princ. Opt. Electromagn. Theory Propagation, Interf. Diffr. Light by Max Born, Emil Wolf Oxford, GB Pergamon Press. 1980 (1980). 25. Philibert S. Tsai and David Kleinfeld. In Vivo Optical Imaging of Brain Function, Second Edition. Methods (2009). doi:10.1201/9781420076851 26. Charles, A., Song, A., Gauthier, J., Pillow, J. & Tank, D. Neural Anatomy and Optical Microscopy (NAOMi) Simulation for evaluating calcium imaging methods. bioRxiv 726174 (2019). doi:10.1101/726174 27. Zipfel, W. R., Williams, R. M. & Webb, W. W. Nonlinear magic: multiphoton microscopy in the biosciences. Nat. Biotechnol. (2003). doi:10.1038/nbt899 28. Quirin, S. et al. Calcium imaging of neural circuits with extended depth- of- field light- sheet microscopy. Opt. Lett. (2016). doi:10.1364/ol.41.000855 29. Fahrbach, F. O. et al. Rapid 3D light- sheet microscopy with a tunable lens "light" + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 80, 470, 905]]<|/det|> +1258 Accuracy 3D Quantum Dot Tracking with 1325 1259 Multifocal Plane Microscopy for the Study of 1326 1260 Fast Intracellular Dynamics in Live Cells. 1327 Biophys. J (2008). 1326 doi:10.1364/OE.21.021010 1329 30. Ahrens, M. B., Orger, M. B., Robson, D. N., 1330 1264 Li, J. M. & Keller, P. J. Whole-brain 1331 functional imaging at cellular resolution 1332 using light-sheet microscopy. Nat. Methods 1333 10, 413- 420 (2013). 1334 1268 31. Leung, L. C., Wang, G. X. & Mourrain, P. 1335 Imaging zebrafish neural circuitry from 1336 whole brain to synapse. Front. Neural 1337 Circuits (2013). 1338 doi:10.3389/fncir.2013.00076 1339 32. Kermen, F., Lal, P., Faturos, N. G. & Yaksi, 1340 E. Interhemispheric connections between 1341 olfactory bulbs improve odor detection. 1342 PLoS Biol. 18, e3000701 (2020). 1343 33. Wu, Y., dal Maschio, M., Kubo, F. & Baier, 1344 H. An Optical Illusion Pinpoints an Essential 1345 Circuit Node for Global Motion Processing. 1346 Neuron (2020). 1347 doi:10.1016/j.neuron.2020.08.027 1348 34. Sancataldo, G. et al. Flexible multi-beam 1349 light-sheet fluorescence microscope for live 1350 imaging without striping artifacts. Front. 1351 Neuroanat. 13, (2019). 1352 35. Huisken, J. & Stainier, D. Y. R. Selective 1353 plane illumination microscopy techniques in 1354 developmental biology. Development 136, 1963- 1975 (2009). 1356 Lavagnino, Z. et al. Two-photon excitation 1357 selective plane illumination microscopy 1358 (2PE- SPIM) of highly scattering samples: 1359 characterization and application References 1360 and links "Optical sectioning deep inside 1361 live embryos by selective plane illumination 1362 microscopy "High- reso. Nat. Methods 305, 1363 (2004). 1364 Hillman, E. M. C., Voleti, V., Li, W. & Yu, H. 1365 Light- Sheet Microscopy in Neuroscience. 1366 (2019). doi:10.1146/annurev-neuro-070918 1367 Vladimirov, N. et al. Light- sheet functional 1368 imaging in fictively behaving zebrafish. Nat. 1369 Methods 11, 883- 4 (2014). 1370 39. Wulliman, M. F., Rupp, B. & Reichert., H. 1371 Neuroanatomy of the Zebrafish Brain: A 1372 Topological Atlas. (Springer Birkhaeuser, 1373 1996). 1374 40. Kovačević, N. et al. A three- dimensional 1375 MRI atlas of the mouse brain with estimates 1376 of the average and variability. Cereb. 1377 Cortex (2005). doi:10.1093/cercor/bhh165 1378 41. Fan, G. Y. et al. Video- rate scanning two- 1379 photon excitation fluorescence microscopy 1380 and ratio imaging with cameleons. Biophys. 1381 J. (1999). doi:10.1016/S0006- 3495(99)77396- 0 1382 42. Schrödel, T., Prevedel, R., Aumayr, K., 1383 Zimmer, M. & Vaziri, A. Brain- wide 3d 1384 imaging of neuronal activity in 1385 Caenorhabditis elegans with sculpted light. 1386 Artic. Nat. methods 10, (2013). 1387 43. Negrean, A. & Mansvelder, H. D. Optimal 1388 lens design and use in laser- scanning 1389 microscopy. Biomed. Opt. Express (2014). 1390 + +<|ref|>text<|/ref|><|det|>[[489, 80, 876, 905]]<|/det|> +doi:10.1364/boe.5.001588 44. Gerritsen, H. C. & Grauw, C. J. De. Imaging of optically thick specimen using two- photon excitation microscopy. Microsc. Res. Tech. 47, 206- 209 (1999). 45. Egner, A. & Hell, S. W. Aberrations in confocal and multi- photon fluorescence microscopy induced by refractive index mismatch. in Handbook of Biological Confocal Microscopy: Third Edition 404- 413 (Springer US, 2006). doi:10.1007/978- 0- 387- 45524- 2_20 46. Matsumoto, N., Konno, A., Inoue, T. & Okazaki, S. Aberration correction considering curved sample surface shape for non- contact two- photon excitation microscopy with spatial light modulator. Sci. Rep. 8, 1- 13 (2018). 47. Booth, M. J. & Wilson, T. Refractive- index- mismatch induced aberrations in single- photon and two- photon microscopy and the use of aberration correction. J. Biomed. Opt. 6, 266 (2001). 48. Young, P. A., Clendenon, S. G., Byars, J. M., Decca, R. S. & Dunn, K. W. The effects of spherical aberration on multiphoton fluorescence excitation microscopy. J. Microsc. 242, 157- 165 (2011). 49. Matsumoto, N., Inoue, T., Matsumoto, A. & Okazaki, S. Correction of depth- induced spherical aberration for deep observation using two- photon excitation fluorescence microscopy with spatial light modulator. Biomed. Opt. Express 6, 2575 (2015). 50. Helmchen, F. & Denk, W. Deep tissue two- photon microscopy (also about filling of objective). Nat. Methods (2005). doi:10.1038/NMETH818 51. Birge, R. R. Two- Photon Spectroscopy of Protein- Bound Chromophores. Acc. Chem. Res 19, (1986). 52. Patterson, G. H. & Piston, D. W. Photobleaching in two- photon excitation microscopy. Biophys. J. (2000). doi:10.1016/S0006- 3495(00)76762- 2 53. Hopt, A. & Neher, E. Highly nonlinear photodamage in two- photon fluorescence microscopy. Biophys. J. (2001). doi:10.1016/S0006- 3495(01)76173- 5 54. Podgorski, K. & Ranganathan, G. Brain heating induced by near- infrared lasers during multiphoton microscopy. J. Neurophysiol 116, 1012- 1023 (2016). 55. Tung, C.- K. et al. Effects of objective numerical apertures on achievable imaging depths in multiphoton microscopy. Microsc. Res. Tech. 65, 308- 314 (2004). 56. Sieracki, C. K., Levey, C. G. & Hansen, E. W. Simple binary optical elements for aberration correction in confocal microscopy. Opt. Lett. 20, 1213 (1995). 57. Euler, T. et al. Eyecup scope- optical recordings of light stimulus- evoked fluorescence signals in the retina. Pflugers Arch. 457, 1393- 414 (2009). 58. Yoshimatsu, T., Bartel, P., Janiak, F. & Baden, T. Optimal rotation of colour space + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[60, 85, 468, 850]]<|/det|> +1392 by zebrafish cones in vivo. F1000Research 145 74. 1393 8, (2019). 145 1394 59. Zimmermann, M. J. Y. et al. Zebrafish 145 1395 Differentially Process Color across Visual 145 1396 Space to Match Natural Scenes. Curr. Biol. 145 1397 28, 2018- 2032. e5 (2018). 145 1398 60. Kwan, K. M. et al. The Tol2kit: A multisite 146 1399 gateway- based construction kit forTol2 146 1400 transposon transgenesis constructs. Dev. 146 1401 Dyn. 236, 3088- 3099 (2007). 146 1402 61. Pittman, A. J., Law, M.- Y. & Chien, C.- B. 146 1403 Pathfinding in a largevertebrate axon tract: 146 1404 isotypic interactions guide retinotectal 146 1405 axons at multiple choice points. 146 1406 Development 129, 617- 624 (2008). 146 1407 62. Suli, A., Guler, A. D., Raible, D. W. & 146 1408 Kimelman, D. A targeted gene expression 147 1409 system using the tryptophan repressor in 147 1410 zebrafish shows no silencing in subsequent 147 1411 generations. Development 141, 1167- 74 147 1412 (2014). 147 1413 Yoshimatsu, T. et al. Presynaptic partner 147 1414 selection during retinal circuit reassembly 147 1415 varies with timing of neuronal regeneration 147 1416 in vivo. Nat. Commun. 7, 10590 (2016). 147 1417 64. Kay, J. N. et al. Development. Development 147 1418 128, 2497- 2508 (2004). 148 1419 Chen, T.- W. et al. Ultrassensitive fluorescent 148 1420 proteins for imaging neuronal activity. 148 1421 Nature 499, 295- 300 (2013). 148 1422 66. Dana, H. et al. Thy1- GCaMP6 transgenic 148 1423 mice for neuronal population imaging in 148 1424 vivo. PLoS One (2014). 148 1425 doi:10.1371/journal.pone.0108697 148 1426 67. Rey, S., Marra, V., Smith, C. & Staras, K. 148 1427 Nanoscale Remodeling of Functional 148 1428 Synaptic Vesicle Pools in Hebbian 149 1429 Plasticity. Cell Rep. (2020). 149 1430 doi:10.1016/j.celrep.2020.01.051 149 1431 68. Bale, M. R. et al. Learning and recognition 149 1432 of tactile temporal sequences by mice and 149 1433 humans. Elife (2017). 149 1434 doi:10.7554/eLife.27333 149 1435 Prieto- Godino, L. L., Diegelmann, S. & 149 1436 Bate, M. Embryonic Origin of Olfactory 149 1437 Circuitry in Drosophila: Contact and Activity- 149 1438 Mediated Interactions Pattern Connectivity 150 1439 in the Antennal Lobe. PLoS Biol. 10, (2012). 150 1440 70. Franke, K. et al. Inhibition decorrelates 150 1441 visual feature representations in the inner 150 1442 retina. Nature 542, 439- 444 (2017). 150 1443 71. Goppert- Mayer, M. Über Elementarakte mit 150 1444 zwei Quantensprüngen. Ann. Phys. (1931). 150 1445 doi:10.1002/andp.19314010303 150 1446 72. Larson, D. R. et al. Water- soluble quantum 150 1447 dots for multiphoton fluorescence imaging 150 1448 in vivo. Science (80- ). (2003). 150 1449 doi:10.1126/science.1083780 151 1450 73. Ricard, C. et al. Two- photon probes for in 151 1451 vivo multicolor microscopy of the structure 151 1452 and signals of brain cells. 223, 3011- 3043 151 1453 (2018). + +<|ref|>text<|/ref|><|det|>[[485, 87, 875, 840]]<|/det|> +1454 74. Colon, J. & Lim, H. Shaping field for 3D 1455 laser scanning microscopy. Opt. Lett. 40, 3300- 3 (2015). 1456 Hiraoka, Y., Sedat, J. W. & Agard, D. A. 1457 Determination of three- dimensional imaging 1458 properties of a light microscope system. 1459 Partial confocal behavior in epifluorescence 1460 microscopy. Biophys. J. 57, 325- 333 1461 (1990). 1462 Hell, S., Reiner, G., Cremer, C. & Stelzer, 1463 E. H. K. Aberrations in confocal 1464 fluorescence microscopy induced by 1465 mismatches in refractive index. J. Microsc. 1466 169, 391- 405 (1993). 1467 Costantini, I., Cicchi, R., Silvestri, L., Vanzi, 1468 F. & Pavone, F. S. In- vivo and ex- vivo 1469 optical clearing methods for biological 1470 tissues: review. Biomed. Opt. Express 10, 5251 (2019). 1472 Azucena, O. et al. Wavefront aberration 1473 measurements and corrections through 1474 thick tissue using fluorescent microsphere 1475 reference beacons. Opt. Express 18, 17521 1476 (2010). 1477 Booth, M. J. Wavefront sensorless adaptive 1478 optics for large aberrations. Opt. Lett. 32, 5 1479 (2007). 1480 Ji, N., Milkie, D. E. & Betzig, E. Adaptive 1481 optics via pupil segmentation for high- 1482 resolution imaging in biological tissues. Nat. 1483 Methods 7, 141- 147 (2010). 1484 Tao, X. et al. Live imaging using adaptive 1485 optics with fluorescent protein guide- stars. 1486 Opt. Express 20, 15969 (2012). 1487 Wang, K. et al. Rapid adaptive optical 1488 recovery of optimal resolution over large 1489 volumes. Nature Methods 11, 625- 628 1490 (2014). 1491 Neil, M. A. A. et al. Adaptive aberration 1492 correction in a two- photon microscope. J. 1493 Microsc. 200, 105- 108 (2000). 1494 Theofanidou, E., Wilson, L., Hossack, W. J. 1495 & Arlt, J. Spherical aberration correction for 1496 optical tweezers. Opt. Commun. 236, 145- 150 (2004). 1497 Park, J. H., Kong, L., Zhou, Y. & Cui, M. 1498 Large- field- of- view imaging by multi- pupil 1499 adaptive optics. Nat. Methods 14, 581- 583 1500 (2017). 1501 Booth, M. J., Marsh, ; P, Burns, D. & Girkin, 1502 J. Fluorescence microscopy; (180.6900) 1503 Three- dimensional microscopy; (170.3880) 1504 Medical and biological imaging. Phil. Trans. 1505 R. Soc. A- Math. Phys. Eng. Sci 365, (Wiley, 1506 2007). 1507 Aoyagi, Y., Kawakami, R., Osanai, H., Hibi, 1508 T. & Nemoto, T. A rapid optical clearing 1509 protocol using 2,2'- thiodiethanol for 1510 microscopic observation of fixed mouse 1511 brain. PLoS One (2015). 1512 doi:10.1371/journal.pone.0116280 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 144, 70]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[44, 88, 860, 831]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 850, 115, 870]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 891, 944, 956]]<|/det|> +Non- Telecentric beam optics in 2- photon microscopy. a, Schematics of larval Drosophila (left). larval zebrafish (centre) and adult mouse (right) with central nervous system highlighted (green) to illustrate size differences. Insets next to the mouse for direct size- comparison between these species on the same + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 40, 960, 660]]<|/det|> +scale. b, Optical configurations of standard diffraction limited (DL, left) 2P setup with parallel laser beam entering objective's back aperture. Right, non- telecentric (nTC, middle, right) configurations use a still diverging laser beam instead. As a result, the field of view and focal distance are expanded, and the point spread function (PSF) elongates. These effects scale with the angle of divergence (compare n TC1 and nTC2). C, Schematic representations of typical neuronal somata in species shown in (a), as interrogated by 2P setups shown in (b), respectively. d, In vivo 7 dpf larval zebrafish (Huc::GCAMP6f) imaged with an out- of- the- box Sutter- MOM DL setup at full field of view (top) and when zoomed in to reveal individual neuronal somata (bottom) as indicated. e, same zebrafish as shown in (d), as well as two further zebrafish imaged using n TC2 configuration at maximal field of view (top). Zooming in to the same area as in (d, bottom) nonetheless reveals cellular detail (e, bottom). f, In vivo adult mouse cranial window over somatosensory cortex imaged with nTC2 maximal field of view (top) and when zoomed in as indicated (bottom) g, Left: Optical configuration of a standard DL setup with collimation system consisting of a scan lens and a tube lens to set- up an infinity collimated laser beam at the level of the objective's back aperture. Effective refractive power and relative distances of lenses indicated. The intermediary focal point (IFP) is immediately behind the scan lens (arrowhead). Middle: n TC1 configuration replaces the scan lens with a pair of planoconvex lenses (L1,2). The relative position of L2 to the tube lens defines the position of the new IFP, which is now further along the laser path. As a result, the field of view can be expanded to between 1.2 and 1.8 mm. Right: n TC2 configuration using a single plano- convex lens (L3) allows FOV expansion to 2.5 - 3.5 mm. h, complete nTC setup, including also an ETL positioned in front of the scan mirrors for rapid axial- scanning. PMTS, Photomultipliers. I, FOV expansion under nTC combines two effects: Increased focal distance (left) and reduced numerical aperture (N.A., right), which together give rise to a larger effective focal plane and enlarged PSFj, Power at sample measured for all configurations, expressed as a percentage of the power that reaches the scanning mirrors. m, point spread functions (PSFs) measured for all optical configurations, with size of typical neuronal somata of different species indicated. All scale- bars 10 um. n.o, lateral (n) and axial (o) spread of the PSFs quantified. Errors in s.d.. The specified numbers are for a Sutter MOM 2P microscope with Zeiss Objective W "Plan- Apochromat" 20x/1.0. Power at the sample plane was 0.35 mW. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 40, 750, 780]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 820]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 951, 955]]<|/det|> +Spatial resolution. a, Schematic (top) and photograph (bottom) of the setup used to directly film excitation volumes. b, effective scan- planes directly visualised as indicated in (a) for all optical configurations as indicated, in each case with scan- points spaced to facilitate inspection of individual PSFs. c. The same set of neurons of the 7 dpf larval zebrafish upper spinal cord (HuC:GCAMP6f, random sparse expression, see overview scan and schematic on the left) was imaged in all optical configurations + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[39, 45, 949, 293]]<|/det|> +as indicated at \(512 \times 512\) px (1 Hz). Arrowheads highlight the same synaptic structures in each scan. d-f, \(64 \times 64\) px (7.81 Hz) activity scan from fields of view shown in (c) for all five configurations during presentation of full-field flashes of UV-light which stochastic elicited activity in these imaged neural structures. In each case the average scan projection (d) and neighbour- correlation based activity projection (e) are shown (hereafter referred to as "activity- correlation"). Darker shadings, equalised for visibility, denotes increased local activity (for details, see Refo). Black traces in (f) show time-traces for the same structure in all cases. For the nTC2 2.5 mm FOV condition, time-traces from different neural structures are extracted to illustrate different responses in different structures. All activity traces in this and the following figures are shown in z-scores relative to their own baseline. We choose this metric over dF/F as it emphasises detectability of events rather than the relative change from the indicator's baseline fluorescence, which differs between biosensors. + +<|ref|>image<|/ref|><|det|>[[55, 303, 940, 608]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 634, 118, 653]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[39, 675, 953, 902]]<|/det|> +Rapid remote focussing. a,b, Scan- profiles with the electrically tunable lens (ETL) "flat" (zero input current, lowest profile) and engaged to achieved axially elevated scan planes at \(+300\) and \(+600\) pm (middle, upper profiles, respectively) in n TC1 (a) and n TC2 (b) configuration, as indicated. Associated size- changes in the effective full field of view were generally \(< 5\%\) (compare top and bottom planes). In each case, axial- shifts required \(< 25\%\) unidirectional peak current on the ETL which in turn facilitated rapid ETL- settling times: c,d, Schematic (c) and measured (d) axial jumps and settling time: the ETL was programmed to iteratively focus up and down by 150 um at each end of two long (5 ms) scan lines, as indicated. This enabled a direct read- out of ETL settling at each line- onset (oscillations in d). For the 150 um jumps shown, oscillations decayed below detectability within 2- 3 ms. For corresponding readouts of the ETL- position signal, see Fig. S3. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 45, 692, 780]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 820]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 955, 955]]<|/det|> +Mesoscale imaging of zebrafish larvae. a, Photograph of two 9 dpf zebrafish larvae mounted head- to- head in a microscope chamber with mm- scale ruler in background. B. The same 2 fish (Huc:GCAMP6f) as in (a) imaged under 2- photon with nTC2 3.5 mm FOV configuration, at \(512 \times 128\) px (3.91 Hz). c, d, Activity- correlation (cf. Fig. 2e) of the scan in (b) during presentation of full- field flashes of UV- light, with hand- selected exemplary ROIs, extracted time- traces (d) and light- stimulus- aligned averages (e). f- i, the same + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 933, 111]]<|/det|> +fish as shown on the left in (b, fish 1), now shown at full 3.5 mm field of view (f, 512x128 px, 3.91 Hz) and increased spatial resolution scans of regions as indicated to reveal cellular detail (9-1, 1,024x1,024 px, 0.49 Hz). + +<|ref|>image<|/ref|><|det|>[[55, 115, 760, 852]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 870, 117, 889]]<|/det|> +
Figure 5
+ +<|ref|>text<|/ref|><|det|>[[42, 911, 936, 954]]<|/det|> +3D random access scanning of the zebrafish eye and brain. a, b, Schematic of zebrafish larva from top (a) and front (b) with scan configurations indicated. c, direct x-z visualisation of the scan-profile used in + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 960, 224]]<|/det|> +the below. d, nTC1 1,024x1,024 px scan across an Islet2b:mGCAMP6f 6 dpf larval zebrafish eye and brain. At the centre of the scan, the axial focus is shifted upwards such that the axonal processes of retinal ganglion cells (RGCs) in the tectum (top) and their somata and dendritic processes in the eye (bottom) can be quasi- simultaneously captured, e, f, 1024x1024 px split- plane random access jump between tectum (e) and eye (f) and g-j, 2 times 64x128 px (15.6 Hz) random access scan of the same scan regions with raw (g) and event- averaged (h) fluorescence traces, mean image (i) and activity- correlation (cf. Fig. 2e). The stimulus was a series of full- field broadband flashes of light as indicated. k- o, as (d-j), with individual RGCs transiently expressing GCAMP6f under the same promoter. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[52, 40, 707, 784]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 800, 117, 820]]<|/det|> +
Figure 6
+ +<|ref|>text<|/ref|><|det|>[[42, 841, 940, 955]]<|/det|> +2P plane- bending to image the in vivo larval zebrafish brain. a- c, Schematic of HuC:GCAMP6f larval zebrafish brain viewed from top (a) and front (b) with scan planes indicated, and (c) example- scan- profiles. d, nTC1 512x1024 scans of a 6 dpf zebrafish brain with different plane curvatures, with peak axial displacement at scan centre as indicated. At curvatures \(\sim 100 - 150 \mu m\) peak displacement the scan approximately traverses the surface of the tectum. e- h, mean (e), activity- correlation (f. cf. Fig. 2e) and + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 953, 111]]<|/det|> +fluorescence traces (g, raw and h, event- triggered mean) from a 170x340 px scan (5.88 Hz) of the 100 μm peak displacement configuration (image 3 in (d)). The fish was presented with full- field and spectrally broad (-360-650 nm) series of light- flashes. + +<|ref|>image<|/ref|><|det|>[[55, 111, 550, 850]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 870, 116, 890]]<|/det|> +
Figure 7
+ +<|ref|>text<|/ref|><|det|>[[42, 911, 940, 954]]<|/det|> +Mesoscale and random- access imaging in mouse brain slice. a,b, Schematic of brain (a) and transverse section (b) of a Thy 1:GCAMP6f mouse. c, d, 1024x1024 px n TC2 example scan of slice through cortex + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 44, 949, 249]]<|/det|> +and hippocampus at maximal FOV (c) and nTC2 zoom in (d) as indicated. Red arrows indicate rapid transitions between scan regions, with the inset showing scan- profiles. The slice was bathed in an epileptogenic (high K+, zero Mg2+) solution to elicit seizures. e- g, Mean of 256x256 px scan (3.91 Hz) of (d) with regions of interest (ROIs) indicated (e), activity- correlation projection (Methods) indicating regions within the scan showing regions of activity computed as mean correlation of each pixel's activity over time to all its neighbours (for details, see Ref70) (f) and z- normalised fluorescence traces (g). h- l, 2 times 128x256 px (3.91 Hz) random access scan of two regions as indicated in (d) allows quasi- simultaneous imaging of the cortex (h) and hippocampus () at increased spatial resolution, with activity- correlation (j,k, cf. Fig. 2e) and fluorescence traces (1) extracted as in (k). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 55, 785, 777]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 802, 117, 821]]<|/det|> +
Figure 8
+ +<|ref|>text<|/ref|><|det|>[[42, 842, 949, 955]]<|/det|> +Mesoscale random- access imaging of mouse cortex in vivo. a, b, Schematic of Thy 1: GCamP6f mouse brain in vivo (a) with cranial window over the somatosensory cortex (b). c,d, \(1024 \times 1024\) px nTC2 (c) and nTC1 (d) images as indicated. Red arrows indicate rapid transitions between scan regions, with the inset indicating the scan- profile. e- i, 2 times \(128 \times 256\) px (3.91 Hz) random access scan as indicated in (d) with mean- projection (e,f), activity- correlation (g, h, cf. Fig. 2e) and fluorescence traces (i), taken from the ROIs + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 45, 944, 88]]<|/det|> +as indicated in (g,h). j-l, NTC1 128x128 px xyz-tilted plane (7.82 Hz) traversing through cortical layers 1-4 at \(\sim 45^{\circ}\) relative to vertical with mean image (k) and activity-correlation (l, cf. Fig. 2e). + +<|ref|>image<|/ref|><|det|>[[44, 90, 688, 825]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 845, 117, 864]]<|/det|> +
Figure 9
+ +<|ref|>text<|/ref|><|det|>[[44, 887, 952, 952]]<|/det|> +Multi- plane imaging and optogenetics for functional circuit mapping. a,b, DL (a) and NTC1 (b) 1,024x1,024 px scans of the ventral nerve cord of a 3rd instar VGlut:GCamP6f Drosophila larva. c-f, Scan- profiles taken in DL (C) and nTC1 (d) across 6 planes spaced \(\sim 15 \mu m\) apart. e, f, Schematic of first instar + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[41, 45, 950, 248]]<|/det|> +elav:GCamP6s; Ocro: CsChrimson Drosophila larva from top (e) and side (1), with CsChrimson (red) and GCAMP6s (green) expression pattern and scan- planes indicated. g- k optogenetic circuit mapping of olfactory processing centres across the larval brain. Six scan planes (170x340 px each) were taken at 0.98 Hz/plane (i.e. volume rate) during presentation of 587 nm light flashes (2 s) to activate CsChrimson in olfactory sensory neurons (OSNs). Brain anatomy (g) and false- colour coded fluorescence difference image (h, 1- 2 s after flash onset minus 1- 2 s prior to flash onset), with fluorescence activity traces (i, raw and j, event triggered average). For a zoom- in on the antennal lobe in a different specimen, see also SFig. 5. k, data from (h) summarised: top right; max- projection through the brain, with left and bottom showing transverse max- projections across the same data- stack. + +<|ref|>sub_title<|/ref|><|det|>[[44, 270, 311, 298]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 320, 765, 341]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 357, 486, 618]]<|/det|> +- JaniaketalSupplementaryRevision.pdf- VideoS13fish.mp4- VideoS2PSFexpansions.mp4- VideoS3ETL.mp4- VideoS42Fish.mp4- VideoS5Planebending.mp4- VideoS6Planebending2.mp4- VideoS7Mousebrainslice.mp4- VideoS8Mousecortex.mp4- VideoS9Drosophilaopticalcircuitmapping.mp4 + +<--- Page Split ---> diff --git a/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/images_list.json b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..06d6d4dccac5dfa43d7065e248bd85398c593a56 --- /dev/null +++ b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/images_list.json @@ -0,0 +1,108 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1. Seismic network and icequake data at Rutford Ice Stream, Antarctica. (a) Location of Rutford Ice Stream (RIS) relative to the Antarctic continent. Topography is from Bedmap232. (b) Map of network with respect to RIS shear-margins. (c) Map of the experiment and icequake data at Rutford Ice Stream, from November 2018 to February 2019. Red scatter points show icequake locations. All icequakes are approximately at ice stream bed28. Green inverted triangles show geophone locations. Bed topography data are from the literature33. Pink dashed line indicates a bed-character boundary from the literature33.", + "footnote": [], + "bbox": [ + [ + 217, + 88, + 784, + 380 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_5.jpg", + "caption": "Fig. 2. All icequake-derived basal sliding parameters through time. Data are a subset of icequake clusters over the period of \\(5^{\\text{th}}\\) to \\(15^{\\text{th}}\\) January 2019. Each colored line represents an individual icequake cluster. Uncertainties are shown by shaded regions. (a) Effective-normal-stress. Red dashed-dotted line indicates the maximum possible ice overburden pressure. (b) Total frictional basal shear-stress. (c) Bed shear-modulus. Previous estimates from literature are indicated by the dashed lines \\(^{14,15}\\) . (d) Slip associated with individual icequakes. (e) Inter-event time between icequakes in a cluster. (f) Equivalent daily slip-rate calculated from the slip and inter-event times in (d) and (e). All data is smoothed by applying a 100-event moving-average window. All uncertainties are estimated using calculus-derived uncertainty propagation methods. Sensitivity analysis of the rate-and-state model is shown in Extended Data Fig. 5.", + "footnote": [], + "bbox": [ + [ + 230, + 152, + 771, + 628 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3. Basal effective-normal-stress, shear-stress and slip-rate for the entire experiment. Colored lines represent individual icequake clusters. (a) Effective-normal-stress on the fault. (b) Shear-stress through time. (c) Slip-rate through time. (d) to (f) Histograms of the respective time-series data in (a) to (c). Uncertainties in (a) to (c) are given by the shaded regions. Other labels as in Fig. 2. Uncertainties are as defined in Fig. 2.", + "footnote": [], + "bbox": [ + [ + 230, + 137, + 763, + 540 + ] + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4. Spatial variability in average basal shear-stress, slip-rate and fault radius for the clusters. (a) Average shear-stress for the clusters. (b) Average slip-rate for the clusters. (c) Average fault radius for the clusters. Residual topography data is from ground-penetrating radar33. Size of scatter points indicates fault radius. Green inverted triangles indicate the locations of the network of receivers used in this study.", + "footnote": [], + "bbox": [ + [ + 155, + 275, + 842, + 488 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Fig. 5. Schematic diagram summarizing the findings of this study in relation to basal friction and slip with bed characteristics. Bed properties are labelled in the legend. Numbered points are referred to in the text. Note that features not to scale, but arranged approximately according to spatial trends in Fig. 4. Regime I and regime II are shown schematically, with regime I being clast-on-rock icequake slip behavior and regime II being till-on-till slip behavior.", + "footnote": [], + "bbox": [ + [ + 252, + 85, + 744, + 483 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_2.jpg", + "caption": "Extended Data Fig. 2. Quality factor (Q) and corner frequency \\((f_{c})\\) distributions for the icequakes in this experiment. (a) Histogram of Q. (b) Histogram of \\(f_{c}\\) . Values for each icequake are averaged for all individual station observations.", + "footnote": [], + "bbox": [], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_4.jpg", + "caption": "Extended Data Fig. 4. Results of the least squares inversion of Equation 17. Blue scatter points are the data and red scatter points show the least-squares inversion result.", + "footnote": [], + "bbox": [], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/Extended_Data_Figure_5.jpg", + "caption": "Extended Data Fig. 5. Rate-and-state friction model sensitivity analysis. Plot of the sensitivity in frictional shear-stress at the bed, \\(\\tau_{bed}\\) , and slip-rate at the bed, \\(\\nu_{slip}\\) , with perturbation of the key observational parameters. The reference values used to normalize the variations are the average values of \\(\\tau_{bed}\\) and \\(\\nu_{slip}\\) observed at all the clusters. The magnitude of variation in each parameter are summarized in Table S1. See supplementary text for further details.", + "footnote": [], + "bbox": [ + [ + 272, + 82, + 728, + 417 + ] + ], + "page_idx": 23 + } +] \ No newline at end of file diff --git a/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98.mmd b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98.mmd new file mode 100644 index 0000000000000000000000000000000000000000..7f82a2f48a2c8a786a57e7571d81d028058c8c9e --- /dev/null +++ b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98.mmd @@ -0,0 +1,512 @@ + +# Friction and slip measured at the bed of an Antarctic ice stream + +Thomas Hudson ( \(\boxed{\bullet}\) thomas.hudson@earth.ox.ac.uk) University of Oxford https://orcid.org/0000- 0003- 2944- 883X + +Sofia- Katerina Kufner German Research Centre for Geosciences Potsdam https://orcid.org/0000- 0002- 9687- 5455 + +Alex Brisbourne British Antarctic Survey, Natural Environment Research Council https://orcid.org/0000- 0002- 9887- 7120 + +Michael Kendall University of Oxford https://orcid.org/0000- 0002- 1486- 3945 + +Andrew Smith British Antarctic Survey https://orcid.org/0000- 0001- 8577- 482X + +Richard Alley Pennsylvania State University + +Robert Arthern N.E.R.C. British Antarctic Survey + +Tavi Murray Swansea University https://orcid.org/0000- 0001- 6714- 6512 + +Article + +Keywords: + +Posted Date: January 13th, 2022 + +DOI: https://doi.org/10.21203/rs.3.rs- 1214097/v1 + +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> + +# Friction and slip measured at the bed of an Antarctic ice stream + +T.S. Hudson \(^{1*}\) , SK Kufner \(^{2}\) , A.M. Brisbourne \(^{2}\) , JM Kendall \(^{1}\) , A.M. Smith \(^{2}\) , R.B. Alley \(^{3}\) , R.J. Arthern \(^{2}\) , T. Murray \(^{4}\) + +## Affiliations: + +\(^{1}\) Department of Earth Sciences, University of Oxford; 3 South Parks Rd, Oxford, OX1 3AN, UK + +\(^{2}\) UKRI British Antarctic Survey; High Cross, Madingley Rd, Cambridge, CB3 0ET, UK + +\(^{3}\) Department of Geosciences, and Earth and Environmental Systems Institute, Pennsylvania State University, University Park, PA 16802, USA + +\(^{4}\) Department of Geography, Swansea University, Swansea, Singleton Park, Swansea, SA2 8PP, UK + +\*Corresponding author email address: thomas.hudson@earth.ox.ac.uk + +## Abstract: + +The slip of glaciers over the underlying bed is the dominant mechanism governing the migration of ice from land into the oceans, contributing to sea- level rise. Yet glacier slip remains poorly understood or constrained by observations. Here we observe both frictional shear- stress and slip at the bed of an ice stream, using 100,000 repetitive stick- slip icequakes from Rutford Ice Stream, Antarctica. Basal shear- stresses and slip- rates vary from \(10^{4}\) to \(10^{7}Pa\) and 0.2 to \(1.5mday^{- 1}\) , respectively. Friction and slip vary temporally over the order of hours and spatially over 10s of meters, caused by corresponding variations in ice- bed interface material and effective- normal- stress. Our findings also suggest that the bed is substantially more complex than currently assumed in ice stream models and that basal effective- normal- stresses may be significantly higher than previously thought. The observations also provide previously unresolved constraint of the basal boundary conditions of ice dynamics models. This is critical for constraining the primary contribution of ice mass loss in Antarctica, and hence the endeavor to reduce uncertainty in sea- level rise projections. + +<--- Page Split ---> + +## Main Text: + +Glacier slip is the primary mechanism governing the migration of ice from land into the oceans, providing a major contribution to sea- level rise \(^{1,2}\) . Friction at the bed of a glacier fundamentally limits the speed at which the ice can slip. This friction is controlled by a number of factors, including bed material, the presence of debris in basal ice, and hydrological systems that modulate effective- normal- stresses. However, basal friction and slip remain poorly understood or constrained by observations \(^{1,3,5}\) . Such observational constraint of friction and slip is critical for the verification of ice- bed boundary condition assumptions in ice dynamics models, which are required to reduce uncertainty in corresponding sea- level rise projections \(^{4,6,7}\) . + +Previous contributions to address this critical observational void come from laboratory- based experiments \(^{8 - 13}\) geophysical studies \(^{14 - 23}\) , and borehole measurements \(^{24,25}\) . However, to date there have been challenges with such approaches. Laboratory experiments provide insight into fundamental physical properties of the bed material (till) \(^{10}\) and ice- bed interface interactions \(^{12}\) but are limited by scale and the diversity of natural glacier beds. Geophysical studies have measured the in- situ bed strength, but with sparse spatial and temporal resolution \(^{14}\) . Borehole measurements of slip are not only sparse, but have not been accompanied by measurements of shear- stress, making quantitative interpretations difficult. The ice streams and outlet glaciers that contribute the majority of ice flux into the oceans likely have active, spatially- and temporally- varying hydrological systems \(^{26,27}\) , perturbing basal friction and slip over short time- and length- scales. An observational void therefore remains. + +Here we address this observational void by using icequakes to provide the first spatially- mapped, in- situ observations of both frictional drag and slip- rate at the bed of an ice stream. These icequakes are generated by the sudden release of strain at or near the ice- bed interface. The dataset analyzed comprises 100,000 icequakes \(^{28}\) from Rutford Ice Stream (RIS), Antarctica (see Fig. 1). The icequakes originate approximately at the center of the ice stream, where the dominant source of drag is postulated to originate from the bed rather than from the shear margins. These icequakes nucleate in clusters that are highly repetitive (see Extended Data Fig. 1), with near- constant inter- event times of the order of 100s of seconds and icequakes clusters active for hours to days \(^{28}\) . These icequakes are inferred to be at the bed from: their hypocentral depths; the consistent flow- and bed- parallel orientation of their double- couple focal mechanism slip- vectors; and full- waveform modelling a typical RIS icequake source \(^{20,28 - 30}\) . The tight spatial clustering and repetitive nature motivate our use of a rate- and- state friction law in combination with icequake observations to investigate the glacier sliding process. This rate- and- state friction law \(^{31}\) also enables the calculation of other basal parameters including bed shear moduli and insight into the modulation of glaciological effective- normal- pressures. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Fig. 1. Seismic network and icequake data at Rutford Ice Stream, Antarctica. (a) Location of Rutford Ice Stream (RIS) relative to the Antarctic continent. Topography is from Bedmap232. (b) Map of network with respect to RIS shear-margins. (c) Map of the experiment and icequake data at Rutford Ice Stream, from November 2018 to February 2019. Red scatter points show icequake locations. All icequakes are approximately at ice stream bed28. Green inverted triangles show geophone locations. Bed topography data are from the literature33. Pink dashed line indicates a bed-character boundary from the literature33.
+ +## Results + +## Observed ice-bed friction and slip-rate + +The icequake source properties and inter- event times are used in combination with a rate- and- state friction law to calculate: fault effective- normal- stress \((\bar{\sigma})\) ; total frictional shear- stress, or drag per unit area \((\tau)\) ; shear- modulus \((G_{bed})\) ; slip \((d)\) ; and slip- rate \((\nu_{slip})\) at the bed of RIS. Fig. 2 shows these results for a representative subset of icequake clusters. Fault effective- normal- stress, shear- stress and shear- modulus (Fig. 2a- c) vary by orders of magnitude between clusters, even after accounting for uncertainty. However, these parameters are all confined within expected physical limits. Effective- normal- stresses remain below the maximum ice overburden pressure, which is the upper possible limit for the average effective- normal- stress over the entire fault. The observed shear- stress ranges from \(\sim 10^{4}\) to \(10^{7}Pa\) . If the icequake cluster locations, or sticky- spots, contribute more drag than the surrounding bed, then sticky- spot shear- stresses could theoretically have a much higher limit than the average bed shear- stress. Although bed shear moduli vary significantly between clusters, the majority of the clusters' shear moduli agree with one of the only previous seismically- derived in- situ measurements, \(70MPa\) , from Whillans Ice Stream14. Additionally, measurements do not exceed the shear- modulus of ice. + +<--- Page Split ---> + +Slip-rates show smaller variations in amplitude, from \(\sim 0.2\) to \(1.5 \text{m day}^{- 1}\) , but have higher associated uncertainties due to their dependence on both shear-modulus and fault-area. While a number of clusters exhibit time-averaged slip-rates approximately equal to the steady-state surface velocity of RIS (dashed line, Fig. 2f) \(^{34}\) , other clusters have significantly lower slip-rates. + +![](images/Extended_Data_Figure_5.jpg) + +
Fig. 2. All icequake-derived basal sliding parameters through time. Data are a subset of icequake clusters over the period of \(5^{\text{th}}\) to \(15^{\text{th}}\) January 2019. Each colored line represents an individual icequake cluster. Uncertainties are shown by shaded regions. (a) Effective-normal-stress. Red dashed-dotted line indicates the maximum possible ice overburden pressure. (b) Total frictional basal shear-stress. (c) Bed shear-modulus. Previous estimates from literature are indicated by the dashed lines \(^{14,15}\) . (d) Slip associated with individual icequakes. (e) Inter-event time between icequakes in a cluster. (f) Equivalent daily slip-rate calculated from the slip and inter-event times in (d) and (e). All data is smoothed by applying a 100-event moving-average window. All uncertainties are estimated using calculus-derived uncertainty propagation methods. Sensitivity analysis of the rate-and-state model is shown in Extended Data Fig. 5.
+ +Fig. 3a-c show the variation in effective-normal-stress, shear-stress and slip-rate for the entire experiment duration. Histograms of the stress and slip-rate distributions are shown in Fig. 3d-f. The normal and shear-stress histograms show a bimodal distribution, with more than two thirds of the icequakes having effective-normal-stresses lower than \(\sim 5 \times 10^{5} \text{Pa}\) and shear-stresses + +<--- Page Split ---> + +lower than \(2 \times 10^{5}\) Pa. Conversely, the slip-rates exhibit a unimodal distribution, tailing off below \(0.2 m \text{day}^{- 1}\) and above \(1.5 m \text{day}^{- 1}\) . + +![](images/Figure_3.jpg) + +
Fig. 3. Basal effective-normal-stress, shear-stress and slip-rate for the entire experiment. Colored lines represent individual icequake clusters. (a) Effective-normal-stress on the fault. (b) Shear-stress through time. (c) Slip-rate through time. (d) to (f) Histograms of the respective time-series data in (a) to (c). Uncertainties in (a) to (c) are given by the shaded regions. Other labels as in Fig. 2. Uncertainties are as defined in Fig. 2.
+ +Individual icequake clusters switch on and off, being active for the order of hours to days (see Fig. 2). Within single clusters, bed friction and slip are modulated by signals with dominant periods of \(\sim 6\) to 12 hours (see Fig. 2). However, although this alludes to tidal modulation of basal friction, and indeed surface velocities are known to be modulated by tidal frequencies \(^{34,35}\) , we cannot decipher a clear relationship between tidal signals propagated 40 km upstream from RIS's grounding line and our signals \(^{28}\) . We therefore do not discuss any link with tidal signals further. + +The spatial distribution of average basal shear- stress, slip- rate and fault radius for each cluster over a \(7 \times 6\) km region are shown in Fig. 4. Shear- stresses are largest at the clusters farthest upstream, approximately where the bed properties are inferred to transition from unconsolidated to consolidated till \(^{33}\) (pink dashed- line, Fig. 1) and where the bed has shorter wavelength + +<--- Page Split ---> + +topography than upstream that likely inhibits ice flow. Average slip- rate is spatially consistent across all clusters. This is expected, as our study site is located near the center of the ice stream, with no spatial variation in surface slip- rate34. Fault radius, defining the area of an icequake cluster sticky- spot is also measured (see Fig. 4c). Fault radii indicate that individual seismically active sticky- spots have areas \(< 2800 m^2\) . Only a small number of sticky- spots are active at any instant. This suggests that regions of sufficiently high basal friction to generate seismicity are confined to the minority of the bed at a given point in time, yet invoke significant basal drag. Aseismic regions between icequake clusters likely also contribute to the basal drag, presumably providing the dominant source of aseismic drag upstream of the unconsolidated- consolidated sediment boundary (pink dashed line, Fig. 1). + +![](images/Figure_4.jpg) + +
Fig. 4. Spatial variability in average basal shear-stress, slip-rate and fault radius for the clusters. (a) Average shear-stress for the clusters. (b) Average slip-rate for the clusters. (c) Average fault radius for the clusters. Residual topography data is from ground-penetrating radar33. Size of scatter points indicates fault radius. Green inverted triangles indicate the locations of the network of receivers used in this study.
+ +## Discussion + +## Frictional shear-stress and slip-rate + +The most important, immediate finding of this work is the ability to observe in- situ frictional shear- stress and slip- rate, the two critical parameters for constraining the basal drag boundary conditions of ice dynamics models. Our approach could be applied to any glacier that generates icequakes. Most fast- moving glaciers likely generate such icequakes, with the majority of glaciers on which seismometers have been deployed exhibiting at least some basal seismicity16,28,36- 44. Seismic tremor associated with sliding can also occur19,45, thought to initiate at the boundary between the conditionally- stable and unstable regimes of the rate- and- state friction model19,31. Indeed, the premise of this study was inspired by such observations19. However, due to the inability to extract both corner frequency and inter- event time information from tremor, it cannot be used to measure shear- stress and slip using our approach. + +<--- Page Split ---> + +Our confidence in the frictional shear- stress and slip- rate measurements is founded partially on the uncertainty amplitudes, but also fundamentally on the agreement between the observed basal slip- rates and GNSS- derived surface displacement34. This agreement validates assumptions of slip- dominant rather than deformation- dominant flow at RIS and the use of a rate- and- state model and assumptions of the icequake source properties. The small discrepancy between the surface and basal slip- rates is primarily due to uncertainty, except for a minority of particularly sticky- spots. These sticky- spots exhibit particularly strong frictional drag that significantly inhibits local ice flow, albeit for short durations of the order of hours to days. + +Observed basal shear- stresses are of the order of \(10^{4}\) to \(10^{7}\) Pa, acting at sticky- spots with diameters of the order of 10 to \(60\mathrm{m}\) (see Fig. 4). Basal shear- stresses of the order \(10^{5}\) Pa are typical values used in ice dynamics models46 and laboratory experiments8 for RIS's surface slip- rate of \(\sim 400\mathrm{m / yr^{34}}\) . Basal shear- stresses of \(10^{6}\) to \(10^{7}\) Pa might initially appear inconsistently high compared to models and experiments47. However, these high friction sticky- spots are spatially small compared to the total bed area. Our results therefore imply that certain icequake clusters accommodate a considerable proportion of the total basal drag. + +## Icequake generation mechanisms + +We propose that the icequakes are generated by at least one of two mechanisms, or sliding regimes. The presence of two sliding regimes is motivated by the physical system and the bimodal distributions observed in Fig. 3d,e. The regimes (see Fig. 5) are: regime I, rock- on- rock friction between ice- entrained clasts and bedrock at the fault interface; and regime II, where clasts plough through till, with failure accommodated by a till- on- till fault interface. Clasts are pieces of rock partially entrained into the ice (see Fig. 5). The presence of such clasts is discussed in the literature8,12,13,48,49. The motivations for these clast- based icequake models are that they can explain the rate- weakening friction required to generate icequakes13, that clasts are required to generate icequakes in laboratory environments12, and that such icequakes likely originate within one seismic wavelength of the ice- bed interface20. We suggest that the highest effective- normal- stress icequakes exhibit regime I sliding, since this regime allows for the average effective- normal- stress over the entire fault- area to be concentrated over much smaller clast- bedrock contact areas. Similarly, we postulate that the lower effective- normal- stress icequakes are associated with regime II sliding, although we cannot rule out that all icequakes are generated via regime I. + +## Effective normal-stress vs. effective fluid pressure + +Our results imply significant temporal variation in basal effective- normal- stress. Such increases and decreases in effective- normal- stress are inferred to be caused by corresponding decreases and increases in basal water pressure16,50,51. However, while the icequake- derived effective- normal- stresses, \(\bar{\sigma}\) , averaged over the entire fault are equivalent to the average glaciological effective pressure, \(P_{eff} = P_{ice overburden} - P_{water}\) , within that same fault- area, asperities and bed heterogeneity on length- scales shorter than the fault diameter could significantly perturb local effective- normal- pressures. Although all our measured effective- normal- stresses remain below the ice overburden pressure, current hydrological models cannot reconcile glaciological effective pressures greater than \(\sim 0.5\mathrm{MPa}\) for expected till porosities. Sparse observations of + +<--- Page Split ---> + +effective- normal- pressures at RIS from borehole measurements find \(P_{eff} \approx 0.2 MPa^{49}\) , although till acoustic impedance measurements at RIS suggest that dewatering is possible \(^{23}\) . Dewatered till would imply \(P_{eff} = P_{ice overburden}\) . Our highest observed effective- normal- stresses therefore suggest either: that our understanding of bed characteristics and associated physical models may have to be revisited, at the very least for RIS; or that the rate- and- state model does not adequately describe icequake physics. + +We suggest three possible explanations for resolving the discrepancy between observed and theoretical maximum effective- normal- stresses. Firstly, prominent bedrock outcrops could significantly inhibit ice flow, allowing stoss- side effective- normal- stresses of the order of at least \(1 MPa\) to develop \(^{52}\) . A second explanation again lies with bedrock outcrops, whereby impermeable bedrock might inhibit the transport of fluids, facilitating dewatered regions. A third explanation is that till porosities are far lower than conceived in current models of bed properties, possibly supported by till impedance measurements \(^{23}\) . These suggestions are not exhaustive. We only suggest here that our results motivate new models to explain such observations. + +If effective- normal- stresses are related to \(P_{eff}\) at sticky- spots, then they provide an observational foundation for calibrating basal- fluid- pressures assumed in: laboratory- experiments \(^{8}\) ; ice dynamics models \(^{46}\) ; and glacier basal hydrology and tidal forcing \(^{34,35}\) . + +## Enhanced knowledge of Rutford Ice Stream bed conditions + +The considerable variation in bed properties observed at RIS are presented as an example of the enhanced knowledge of the bed properties that our approach provides. Firstly, for the unconsolidated till (label 1, Fig. 5) and much of the consolidated till (label 2, Fig. 5), the effective- normal- stresses are too low to generate the unstable stick- slip conditions required for icequake nucleation. Within consolidated till regions (label 2, Fig. 5), there are small zones that become seismically active if the effective- normal- stress is sufficiently high (label 3, Fig. 5). These sticky- spots turn on and off, modulated by changes in effective- normal- stress, bed strength and bed material. + +Frictional shear- stress at an individual sticky- spot can vary temporally by up to an order of magnitude and spatially by several orders of magnitude. This variation occurs over the order of hours and 100s meters. This implies that there are both till and bedrock outcrops, combined with an active hydrological system or permeable bed capable of such variable spatial and temporal variations in basal fluid pressure. The regions exhibiting the highest shear- stresses are at the upstream edges of local topographic highs near the unconsolidated- consolidated bed boundary (label 6, Fig. 5). This is likely because resistive stresses of these materially stronger, topographic highs can accommodate more basal drag. Some regions of consolidated till might contain pockets of melt water (label 5, Fig. 5). However, we cannot observe such a phenomenon seismically as these patches would have an approximately zero shear- modulus. + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Fig. 5. Schematic diagram summarizing the findings of this study in relation to basal friction and slip with bed characteristics. Bed properties are labelled in the legend. Numbered points are referred to in the text. Note that features not to scale, but arranged approximately according to spatial trends in Fig. 4. Regime I and regime II are shown schematically, with regime I being clast-on-rock icequake slip behavior and regime II being till-on-till slip behavior.
+ +## Wider implications + +Our results show that much of the basal drag at an ice stream can be accommodated within small zones of significantly higher- than- average friction. This could have a profound impact on how sliding is formulated in ice dynamics models. However, although friction varies significantly, average basal slip- rates remain predominantly stable at RIS. This is encouraging for current modelling efforts since if the temporally- and spatially- averaged slip- rates are approximately constant, then perhaps such models are not required to be sensitive to small- scale, rapid variations in bed friction. Our observations quantify the highly variable bed properties over a sufficient duration required to test such a hypothesis. + +Another important question to address is how our approach could be implemented at ice sheet scale. One could deploy temporary seismic arrays on important ice streams and outlet glaciers for short durations. A number of targeted deployments would allow verification of the link between surface- and basal- velocity at ice sheet scales2. + +<--- Page Split ---> + +A further question that this study raises is could a rate-and-state friction model used for the icequake sliding analysis also be used as a mathematical basis for informing sliding laws used in ice dynamics models more generally. Such a model was recently proposed to describe surging glacier behavior \(^{53}\) and has been validated at laboratory scale \(^{12,13}\) . The rate-and-state model meets the conditional stability requirement, not allowing runaway acceleration of a glacier. A more comprehensive comparison to sliding laws for deformable beds, showing agreement for surface velocities \(>100 \mathrm{m / yr}\) , is provided in the supplementary text. + +Finally, these icequakes observations can aid the understanding of earthquake mechanics more generally. Even the smallest magnitude icequakes \((M_{w} \approx - 1.5)\) have high signal- to- noise- ratios, and so could elucidate any lower limits on the fundamental size of earthquake nucleation for given fault properties \(^{54 - 56}\) . Additionally, icequakes in this study have stress- drops that vary with magnitude, contrary to magnitude- invariant stress- drops observed for larger earthquakes \(^{57}\) . + +Our findings show that icequakes can provide the critical observations required to constrain the highly variable friction at the bed of an Antarctic ice stream. Applying such observational constraint to ice dynamics models would reduce uncertainty in corresponding sea- level rise projections. + +## References + +1. Morlighem, M. et al. Spatial patterns of basal drag inferred using control methods from a full-Stokes and simpler models for Pine Island Glacier, West Antarctica. Geophys. Res. Lett. 37, 1–6 (2010). +2. Rignot, E., Mouginot, J. & Scheuchl, B. Ice Flow of the Antarctic Ice Sheet. Science (80-. ). 333, 1427–1430 (2011). +3. Ritz, C. et al. Potential sea-level rise from Antarctic ice-sheet instability constrained by observations. Nature 528, 115–118 (2015). +4. Tsai, V. C., Stewart, A. L. & Thompson, A. F. Marine ice-sheet profiles and stability under Coulomb basal conditions. J. Glaciol. 61, 205–215 (2015). +5. Joughin, I., Smith, B. E. & Schoof, C. G. Regularized Coulomb Friction Laws for Ice Sheet Sliding: Application to Pine Island Glacier, Antarctica. Geophys. Res. Lett. 46, 4764–4771 (2019). +6. Oppenheimer, M. et al. Sea Level Rise and Implications for Low-Lying Islands, Coasts and Communities. IPCC Spec. Rep. Ocean Cryosph. a Chang. Clim. 321–445 (2019). +7. Schlegel, N. J. et al. Exploration of Antarctic Ice Sheet 100-year contribution to sea level rise and associated model uncertainties using the ISSM framework. Cryosphere 12, 3511–3534 (2018). +8. Zoet, L. K. & Iverson, N. R. A slip law for glaciers on deformable beds. Science (80-. ). 368, 76–78 (2020). +9. Iverson, N., Baker, R., Hooke, R., Hanson, B. & Jansson, P. Coupling between a glacier and a soft bed: I. A relation between effective pressure and local shear stress determined from till elasticity. J. Glaciol. 45, 31–40 (1999). +10. Tulaczyk, S., Kamb, W. B. & Engelhart, H. F. Basal mechanics of Ice Stream B, west Antarctica: 1. Till mechanics. J. Geophys. Reasearch 105, 463–481 (2000). +11. Leeman, J. R., Valdez, R. D., Alley, R. B., Anandakrishnan, S. & Saffer, D. M. Mechanical and hydrologic properties of Whillans Ice Stream till: Implications for basal strength and stick-slip failure. J. Geophys. Res. Earth Surf. 121, 1–17 (2016). +12. Zoet, L. K. et al. Application of Constitutive Friction Laws to Glacier Seismicity. Geophys. Res. Lett. 47, 1–9 (2020). + +<--- Page Split ---> + +13. Lipovsky, B. P. et al. Glacier sliding, seismicity and sediment entrainment. Ann. Glaciol. 60, 182–192 (2019). +14. Blankenship, D. D., Bentley, C. R., Rooney, S. T. & Alley, R. B. Till beneath Ice Stream B 1. Properties derived from seismic travel times. J. Geophys. Res. 92, 8903–8911 (1987). +15. Zoet, L. K., Anandakrishnan, S., Alley, R. B., Nyblade, A. A. & Wiens, D. A. Motion of an Antarctic glacier by repeated tidally modulated earthquakes. Nat. Geosci. 5, 623–626 (2012). +16. Gräff, D. & Walter, F. Changing friction at the base of an Alpine glacier. Sci. Rep. 11, 1–10 (2021). +17. Winberry, J. P., Anandakrishnan, S., Alley, R. B., Bindschadler, R. A. & King, M. A. Basal mechanics of ice streams: Insights from the stick-slip motion of Whillans Ice Stream, West Antarctica. J. Geophys. Res. 114, 1–11 (2009). +18. Anandakrishnan, S. & Bentley, C. R. Micro-earthquakes beneath ice streams B and C, West Antarctica: observations and implications. J. Glaciol. 39, 455–462 (1993). +19. Lipovsky, B. P. & Dunham, E. M. Tremor during ice-stream stick slip. Cryosphere 10, 385–399 (2016). +20. Hudson, T. S. et al. Icequake Source Mechanisms for Studying Glacial Sliding. J. Geophys. Res. Earth Surf. 125, (2020). +21. Wiens, D. A., Anandakrishnan, S., Winberry, J. P. & King, M. A. Simultaneous teleseismic and geodetic observations of the stick-slip motion of an Antarctic ice stream. Nature 453, 770–774 (2008). +22. Barcheck, C. G., Tulaczyk, S., Schwartz, S. Y., Walter, J. I. & Winberry, J. P. Implications of basal micro-earthquakes and tremor for ice stream mechanics: Stick-slip basal sliding and till erosion. Earth Planet. Sci. Lett. 486, 54–60 (2018). +23. Smith, A. M. & Murray, T. Bedform topography and basal conditions beneath a fast-flowing West Antarctic ice stream. Quat. Sci. Rev. 28, 584–596 (2009). +24. Engelhardt, H. Basal sliding of Ice Stream B, West Antarctica. J. Glaciol. 44, 223–230 (1998). +25. Truffer, M., Harrison, W. D. & Echelmeyer, K. A. Glacier motion dominated by processes deep in underlying till. J. Glaciol. 46, 213–221 (2000). +26. Rosier, S. H. R., Gudmundsson, G. H. & Green, J. A. M. Temporal variations in the flow of a large Antarctic ice stream controlled by tidally induced changes in the subglacial water system. Cryosph. 9, 1649–1661 (2015). +27. Damsgaard, A. et al. Ice flow dynamics forced by water pressure variations in subglacial granular beds. Geophys. Res. Lett. 43, 12,165-12,173 (2016). +28. Kufner, S. et al. Not all Icequakes are Created Equal: Basal Icequakes Suggest Diverse Bed Deformation Mechanisms at Rutford Ice Stream, West Antarctica. J. Geophys. Res. Earth Surf. 126, (2021). +29. Smith, E. C., Smith, A. M., White, R. S., Brisbourne, A. M. & Pritchard, H. D. Mapping the ice-bed interface characteristics of Rutford Ice Stream, West Antarctica, using microseismicity. J. Geophys. Res. Earth Surf. 120, 1881–1894 (2015). +30. Smith, A. M. Microearthquakes and subglacial conditions. Geophys. Res. Lett. 33, 1–5 (2006). +31. Scholz, C. H. Earthquakes and friction laws. Nature 391, 37–42 (1998). +32. Fretwell, P. et al. Bedmap2: Improved ice bed, surface and thickness datasets for Antarctica. Cryosphere 7, 375–393 (2013). +33. King, E. C., Pritchard, H. D. & Smith, A. M. Subglacial landforms beneath Rutford Ice Stream, Antarctica: detailed bed topography from ice-penetrating radar. Earth Syst. Sci. Data 8, 151–158 (2016). +34. Minchew, B. M., Simons, M., Riel, B. & Milillo, P. Tidally induced variations in vertical and horizontal motion on Rutford Ice Stream, West Antarctica, inferred from remotely sensed observations. J. Geophys. Res. Earth Surf. 122, 167–190 (2017). +35. Rosier, S. H. R., Gudmundsson, G. H. & Green, J. A. M. Temporal variations in the flow of a large Antarctic ice stream controlled by tidally induced changes in the subglacial water system. Cryosphere 9, 1649–1661 (2015). + +<--- Page Split ---> + +36. Stuart, G., Murray, T., Brisbourne, A., Styles, P. & Toon, S. Seismic emissions from a surging glacier: Bakaninbreen, Svalbard. Ann. Glaciol. 42, 151–157 (2005). +37. Walter, F., Deichmann, N. & Funk, M. Basal icequakes during changing subglacial water pressures beneath Gornergletscher, Switzerland. Mitteilungen der Versuchsanstalt fur Wasserbau, Hydrol. und Glaziologie an der Eidgenoss. Tech. Hochschule Zurich 54, 511–521 (2008). +38. Hudson, T. S., Smith, J., Brisbourne, A. & White, R. Automated detection of basal icequakes and discrimination from surface crevassing. Ann. Glaciol. 60, 1–11 (2019). +39. Roeoesli, C., Helmstetter, A., Walter, F. & Kissling, E. Meltwater influences on deep stick-slip icequakes near the base of the Greenland Ice Sheet. J. Geophys. Res. Earth Surf. 1–18 (2016) doi:10.1002/2015JF003601. +40. McBrearty, I. W., Zoet, L. K. & Anandakrishnan, S. Basal seismicity of the Northeast Greenland Ice Stream. J. Glaciol. 1–17 (2020) doi:10.1017/jog.2020.17. +41. Bindschadler, R. A., King, M. A., Alley, R. B., Anandakrishnan, S. & Padman, L. Tidally controlled stick-slip discharge of a West Antarctic ice stream. Science (80-. ). 301, 1087–1089 (2003). +42. Anandakrishnan, S. & Alley, R. B. Ice Stream C, Antarctica, sticky-spots detected by microearthquake monitoring. Ann. Glaciol. 20, 183–186 (1994). +43. Deichmann, N. et al. Evidence for deep icequakes in an Alpine glacier. Ann. Glaciol. 31, 85–90 (2000). +44. Köhler, A., Maupin, V., Nuth, C. & Pelt, W. V. A. N. Characterization of seasonal glacial seismicity from a single-station on-ice record at Holtedahlfonna, Svalbard. 1–14 (2019) doi:10.1017/aog.2019.15. +45. Winberry, J. P., Anandakrishnan, S., Wiens, D. A. & Alley, R. B. Nucleation and seismic tremor associated with the glacial earthquakes of Whillans Ice Stream, Antarctica. Geophys. Res. Lett. 40, 312–315 (2013). +46. Cornford, S. L. et al. Results of the third Marine Ice Sheet Model Intercomparison Project (MISMIP+). Cryosphere 14, 2283–2301 (2020). +47. Ranganathan, M., Minchew, B., Meyer, C. R. & Gudmundsson, G. H. A new approach to inferring basal drag and ice rheology in ice streams, with applications to West Antarctic Ice Streams. J. Glaciol. 67, 229–242 (2021). +48. Hallet, B. A theoretical model of glacier abrasion. J. Glaciol. 23, 39–50 (1979). +49. Smith, A. M. et al. Ice stream subglacial access for ice-sheet history and fast ice flow: The BEAMISH Project on Rutford Ice Stream, West Antarctica and initial results on basal conditions. Ann. Glaciol. (2020) doi:10.1017/aog.2020.82. +50. Sugiyama, S. & Gudmundsson, G. H. Short-term variations in glacier flow controlled by subglacial water pressure at Lauteraargletscher, Bernese Alps, Switzerland. J. Glaciol. 50, 353–362 (2004). +51. Bindschadler, R. The importance of pressurized subglacial water in separation and sliding at the glacier bed. J. Glaciol. 29, 3–19 (1983). +52. Alley, R. B. et al. Bedforms of Thwaites Glacier, West Antarctica: Character and Origin. J. Geophys. Res. Earth Surf. (in Rev. (2021) doi:10.1029/2021JF006339. +53. Thøgersen, K., Gilbert, A., Schuler, T. V. & Malthe-Sørenssen, A. Rate-and-state friction explains glacier surge propagation. Nat. Commun. 10, 1–8 (2019). +54. Dieterich, J. H. Earthquake nucleation on faults with rate-and state-dependent strength. Tectonophysics 211, 115–134 (1992). +55. Cattania, C. & Segall, P. Crack Models of Repeating Earthquakes Predict Observed Moment-Recurrence Scaling. J. Geophys. Res. Solid Earth 124, 476–503 (2019). +56. Chen, T. & Lapusta, N. Scaling of small repeating earthquakes explained by interaction of seismic and aseismic slip in a rate and state fault model. J. Geophys. Res. Solid Earth 114, 1–12 (2009). +57. Abercrombie, R. E. Resolution and uncertainties in estimates of earthquake stress drop and energy release. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences vol. 379 (2021). + +<--- Page Split ---> + +## Acknowledgments: + +We thank NERC British Antarctic Survey for logistics and field support, and specifically the BEAMISH field team (2018/2019). We also thank J. Hawthorne, R. Katz and S. Anandakrishnan for valuable discussions and feedback. AMB, AMS, and TM were funded by Natural Environment Research Council grants NE/G014159/1 and NE/G013187/1, Seismic instruments were provided by NERC SEIS- UK (Loans 1017 and 1111), by BAS and by the Incorporated Research Institutions for Seismology (IRIS) through the PASSCAL Instrument Center at New Mexico Tech. The facilities of the IRIS Consortium are supported by the National Science Foundation's Seismological Facilities for the Advancement of Geoscience (SAGE) Award under Cooperative Support Agreement EAR- 1851048. All the seismic data used in this analysis will be deposited on the IRIS seismological data repository. The icequake catalogue used for this analysis is available from the UK Polar Data Centre58, with details on how this catalogue was constructed given in the peer- reviewed publication28. All the fundamental code used in this study is available open source. QuakeMigrate59 was used for icequake detection, NonLinLoc for icequake relocation60 and SeisSrcMoment for the moment magnitude and other source parameter analysis61. + +<--- Page Split ---> + +## Methods + +## The icequake dataset + +This study uses 100,000 icequakes at the bed of Rutford Ice Stream (RIS), Antarctica. An example of such an icequake arrival can be found in Extended Data Fig. 1. These data were collected over the period of November 2018 to February 2019. The icequakes are detected using QuakeMigrate \(^{38,59}\) and relocated using NonLinLoc \(^{60}\) . A full description of the detection, location and clustering analysis of this seismicity can be found in \(^{28}\) . The hypocentral depths, orientation of focal mechanisms, and full waveform modelling provide us with confidence that these icequakes are associated with sliding within one seismic wavelength ( \(\sim 10 m\) ) of the bed \(^{20,28,29}\) . + +## Observable parameters from stick-slip icequakes + +Earthquake source models can be used to calculate the size of the earthquake, its duration, the fault radius and the shear- stress- drop associated with the release of seismic energy. These observable parameters are required for any analysis of frictional behavior at the bed of glaciers using icequakes. The methods we use to obtain these parameters from the icequake signals are described below. + +## Seismic moment + +The seismic moment, \(M_{0}\) , of an earthquake describes the energy released and is defined as \(^{62}\) , + +\[M_{0} = \frac{4\pi\rho\nu_{1}^{2}r\Omega_{0}}{A_{rad,i}C_{free - surface}} (1),\] + +where \(\rho\) is the density of the medium at the earthquake source, \(\nu_{i}\) is the velocity of the seismic phase \(i\) (P or S), \(r\) is the hypocentre- receiver distance, \(\Omega_{0}\) is the long- period spectral amplitude, \(A_{rad,i}\) is the amplitude of radiation of seismic phase \(i\) for the particular source- receiver azimuth and take- off angle, and \(C_{free - surface}\) is the free surface correction term, which depends upon the angle of inclination of the seismic phase arrival at the surface. For this study, we assume typical ice values of \(\rho = 917 kg m^{- 3}\) , \(\nu_{p,ice} = 3841 m s^{- 1}\) , \(\nu_{s,ice} = 1970 m s^{- 1}\) . \(A_{rad,i}\) is calculated as described in \(^{20}\) , based on the assumption that all the icequakes in this study are double- couple (DC) sources with strikes aligned with the ice flow direction. This assumption is based upon previous observations at Rutford Ice Stream \(^{20,28,29}\) . \(\Omega_{0}\) is calculated by fitting a Brune source model to the noise- removed spectrum of the icequake \(^{63}\) . + +## Corner frequency + +The spectrum of an earthquake contains more information than just the long- period spectral amplitude. If one assumes that an earthquake's spectrum can be described by a Brune model \(^{63}\) then one can also measure the corner frequency, \(f_{c}\) , of the earthquake. However, an earthquake's spectrum is also particularly sensitive to seismic attenuation. + +<--- Page Split ---> + +Seismic attenuation, often described by the quality factor, \(Q\) , reduces the amplitude of an earthquake spectrum non- linearly across all frequencies. If path attenuation is poorly constrained then it can lead to detrimental uncertainty in the measured corner frequency, as evidenced by the trade- off between \(Q\) and \(f_{c}\) in the Brune model \(^{63}\) , + +\[\Omega (\mathrm{f}) = \frac{\Omega_{0} e^{-\pi f \frac{t}{Q}}}{1 + \left(\frac{f}{f_{c}}\right)^{2}} \quad (2),\] + +where \(\Omega (f)\) is the amplitude of the spectrum for a certain frequency \(f\) and \(t\) is the traveltime. + +To obtain an accurate measurement of corner frequency, we therefore use a linearized spectral ratios method to constrain \(Q\) . This spectral ratios method isolates the path effects from the source effects. An example of the linearized Brune model fit and the observed spectrum for an example icequake is shown in Extended Data Fig. 1c. We obtain estimates of \(Q\) from this method of the order of 200 to 800 (see Extended Data Fig. 2b), which are in agreement with other measurements for Antarctic ice \(^{64}\) . This then allows Equation 1 to be fit to the earthquake spectrum with only \(\Omega_{0}\) and \(f_{c}\) as variables. We find that the icequake corner frequencies at RIS fall approximately within the range of 40- 100 Hz (see Extended Data Fig. 2c). + +## Fault radius and stress-drop + +One can estimate the fault radius, \(R\) , and stress- drop, \(\Delta \tau\) , of an earthquake from the corner frequency. + +The relationship between corner frequency and fault radius, \(R\) , is given by \(^{65}\) , + +\[\overline{f_{c}} = \mathrm{k_{i}}\frac{\beta}{\mathrm{R}} \quad (3),\] + +where \(\overline{f_{c}}\) is the spherically- averaged corner frequency for the earthquake, \(\beta\) is the shearwave speed near the source and \(k_{i}\) is a constant relating the spherically- averaged corner frequencies for a specific fault model for the seismic phase \(i\) . Here, we use the fault model of \(^{66}\) , which gives \(k_{p} = 0.38\) and \(k_{S} = 0.26\) for a rupture speed of \(0.98 \beta\) . We let \(\beta\) equal the shear velocity of ice (1970 \(m s^{- 1} 29\) ). For clast- on- bedrock slip (Regime I, Fig. 5, main text), this is valid as rupture will propagate through the bedrock and the ice that the clasts are embedded within, with us only observing the rupture propagation through the ice. For till- on- till slip (Regime II, Fig. 5, main text), our assumption of \(\beta\) is likely an overestimate, resulting in an overestimate of fault radius. As the seismic properties of the till are unknown, we are limited in assigning a lower value of \(\beta\) for any Regime II events. We assume a symmetric circular fault for this analysis. We therefore calculate average corner frequencies for each event based on the corner frequencies observed at all receivers. The potential effects of the symmetric circular fault assumption are shown in \(^{67}\) . + +The uniform stress- drop of an earthquake can then be found using the fault radius and the relationship given by \(^{68}\) , + +<--- Page Split ---> + +\[\Delta \tau = \frac{7}{16}\frac{M_0}{R^3} \quad (4),\] + +We now have all the observable parameters required to constrain a friction model at an icequake source. + +## Using a rate- and state- friction law for deriving frictional shear-stress and slip from icequakes + +Calculating shear-stress + +Earthquakes are typically generated as the result of stick- slip frictional instabilities at a fault interface \(^{31}\) . We hypothesize that icequakes associated with sliding at the bed of a glacier can be described by a similar model. For our investigation, we assume that the fault- interface is at or near ( \(< 1\) wavelength) the ice- bed interface. Schematic diagrams describing the model are given in Extended Data Fig. 3. Within this framework, we can apply the following rate- and state- friction law given by \(^{31}\) , + +\[\tau = \left[\mu_0 + a\ln \left(\frac{\nu}{\nu_0}\right) + b\ln \left(\frac{\nu_0\theta}{\mathcal{L}}\right)\right]\bar{\sigma} \quad (5),\] + +where \(\tau\) is the total frictional shear- stress, \(\mu_0\) is the steady- state friction coefficient at \(\nu = \nu_0\) , \(\nu\) is the slip velocity, \(\nu_0\) is a reference velocity, defined in this case to be the background slip- rate, \(a\) and \(b\) are material properties, \(\mathcal{L}\) is the characteristic slip distance over which the system returns to steady- state and renew surface contacts, and \(\theta\) is the state variable. The variation of the state variable, \(\theta\) , through time can be defined by the aging or the slip laws \(^{69}\) , given by, + +\[\frac{\partial\theta}{\partial t} = 1 - \frac{\nu\theta}{\mathcal{L}} \quad (aging law) \quad (6),\] \[\frac{\partial\theta}{\partial t} = -\frac{\nu\theta}{\mathcal{L}}\ln \left(\frac{\nu\theta}{\mathcal{L}}\right) \quad (slip law) \quad (7).\] + +The state variable, \(\theta\) , represents the characteristic contact lifetime of a fault. In order to apply the rate- and state- model to the stick- slip icequake system in a mathematically tractable way, we assume that the state variable of the system is constant over the duration of an icequake cycle, i.e. \(\frac{\partial\theta}{\partial t} = 0\) through all time during a cycle. For a destructive frictional failure process, \(\theta\) likely changes with time during earthquake nucleation and as the fault heals. However, for the icequake generation mechanisms proposed in this study (see Fig. 5, main text), damage at the fault interface that affects the frictional properties is likely less significant than at traditional earthquake fault interfaces. This lack of damage is evidenced to some extent by the highly repetitive nature of the icequakes \(^{20,28}\) . We assume that at least part of the icequake patch is near steady- state, or approximately at steady- state if it slips sufficiently fast. A caveat to this is that some of the icequake patch could have remained below the steady- state sliding limit, which we do not explore this here. Overall, we deem the approximation of \(\frac{\partial\theta}{\partial t} = 0\) between individual icequakes as acceptable in this case. This assumption can be used to find the state variable, \(\theta\) , as an expression of \(\nu\) , \(\mathcal{L}\) from either the aging law (Equation 6) or the slip law (Equation 7), which both yield, + +<--- Page Split ---> + +\[\theta = \frac{\mathcal{L}}{\nu} \quad (8).\] + +Equation 5 can then be reduced to a rate- dependent friction law, given by, + +\[\tau = \left[\mu_{0} + (a - b)\ln \left(\frac{\nu}{\nu_{0}}\right)\right]\bar{\sigma} \quad (9).\] + +The coefficient of friction in Equation 9 can then be thought of as \(\mu = \mu_{0} + \Delta \mu\) , where \(\mu_{0}\) is the static friction component, and the dynamic friction component, \(\Delta \mu\) , is given by, + +\[\Delta \mu = (a - b)\ln \left(\frac{\nu}{\nu_{0}}\right) \quad (10),\] + +which when multiplied by the effective- normal- stress, \(\bar{\sigma}\) , can be assumed as equal to the earthquake stress- drop (see Equation 13). + +One can then parametrize Equation 9 in such a way so that it can be solved for individual icequakes. We take \(\mu_{0} = 0.4\) , \(a = 5 \times 10^{- 3}\) and \(b = 15 \times 10^{- 3}\) from \(^{19}\) . We approximate the ratio of the instantaneous sliding velocity to the reference velocity, \(\frac{\nu}{\nu_{0}}\) , as, + +\[\frac{\nu}{\nu_{0}} = \frac{\left(\frac{d}{T}\right)}{\left(\frac{d}{\mathrm{inter - event}}\right)} \quad (11),\] + +where \(d\) is the slip associated with an event (unknown), \(T\) is the slip duration, which we approximate to be equal to the inverse of the icequake corner frequency, \(f_{c}\) , \(^{70}\) and \(t_{\text{inter - event}}\) is the time between two consecutive icequakes. The correspondence of these parameters to the stick- slip cycle is shown in Extended Data Fig. 3b. With this parametrization, the velocity ratio then becomes, + +\[\frac{\nu}{\nu_{0}} = f_{c} \cdot t_{\text{inter - event}} \quad (12).\] + +Assuming that the friction at the interface is velocity- weakening and therefore unstable, one can then assume that the dynamic part of Equation 9 is equal to the stress- drop measured during an icequake, \(\Delta \tau^{71}\) . One should note that this assumption implies that all the dynamic stress- release during slip is accommodated seismically (see red shaded region of Extended Data Fig. 3b). However, there is also frictional shear- stress present that cannot be measured directly using stress- drop measurements. We also assume a seismic radiation efficiency of 1, which is obviously an approximation, with the actual seismic radiation efficiency unknown. Although the radiation efficiency will in reality be \(< 1\) , due to thermal heating and the generation of additional surface area during abrasion, fracture tip energy, and other phenomena such as off- fault cracking are likely insignificant in comparison to standard earthquakes \(^{72}\) , so we deem our first- order approximation as reasonable in this case. For tectonic earthquakes, the seismic radiation efficiency typically might be of the order of 0.1 (for example, see \(^{73}\) ). If the icequake seismic radiation efficiencies were similarly low, then this would be approximately equivalent to reducing the magnitude of \(M_{0}\) by a factor of 10. Sensitivity analysis in the supplementary text suggests that such a reduction in \(M_{0}\) would reduce the shear- stress, \(\tau\) , by an order of magnitude, but the slip velocity, \(\nu_{slip}\) , would only be reduced by a factor of 3. Assuming velocity- weakening friction and a radiation efficiency of one results in the definition of the effective- normal- stress at the fault interface, given by, + +<--- Page Split ---> + +\[\bar{\sigma} = \frac{\Delta\tau}{(a - b)ln\left(\frac{v}{v_0}\right)} \quad (13).\] + +Once we know the effective- normal- stress, \(\bar{\sigma}\) , we can find the overall shear- stress on the fault, \(\tau\) , from Equation 9. + +We emphasize that the effective- normal- stress, \(\bar{\sigma}\) , is the normal stress on the fault, which is not necessarily equivalent to a traditionally defined glaciological effective pressure, \(P_{eff} = P_{ice} - P_{water}\) . The fault effective- normal- stress, \(\bar{\sigma}\) , is the effective- normal- stress that acts over the fault- area, \(A_{fault}\) , derived from the earthquake corner frequency (Equation 3). The actual normal stress acting through clasts in contact with the underlying contact surface might increase the normal stress acting through these clasts (see sliding regime I Fig. 5). However, fault- average normal stress, \(\bar{\sigma}\) , must be equal to the average glaciological effective pressure, \(P_{eff}\) , over the same area of the bed. + +## Calculating slip + +The second glaciologically important parameter to measure at the bed is the slip, and hence the basal slip- rate. To calculate slip, we assume that while an individual icequake cluster is active, all (or at least the vast majority of) slip is accommodated seismically. This is likely the case for RIS, as evidenced by the close agreement between surface slip- rate and seismically measured basal slip- rates (see Fig. 2f). Calculating the basal slip, \(d\) , from an icequake is challenging because one first has to determine a method of estimating the bed shear- modulus, \(G_{bed}\) , since the slip is given by, + +\[d = \frac{M_0}{G_{bed}\cdot A} \quad (14),\] + +where \(M_0\) is the seismic moment released by an earthquake and \(A\) is the area of the fault. + +The bed shear- modulus, \(G_{bed}\) , is calculated by assuming a further behavior of the rate- and state- friction law. This behavior is that an earthquake can only nucleate if it is in the unstable regime. In this study, we assume that the temporally- averaged driving shear- stress at the fault varies over longer time- scales than the icequake inter- event time, with the shear- stress at which the fault fails governed by the effective- normal- stress acting on the fault, \(\bar{\sigma}\) . The approximately constant inter- event time between individual consecutive icequake pairs (see Fig. 2e) within a single cluster validates this assumption. \(^{31}\) define the effective- normal- stress at which a fault becomes unstable is defined as the critical normal stress, \(\bar{\sigma}_c\) , with velocity- weakening behavior prevailing above this stress. \(\bar{\sigma}_c\) is given by \(^{31}\) , + +\[\bar{\sigma}_c = \frac{\mathrm{k}\mathcal{L}}{b - a} \quad (15),\] + +where \(k\) is the spring constant of the system (see Extended Data Fig. 3a), which is given by, + +\[k = \frac{G^{*}}{R} \quad (16),\] + +where \(G^{*}\) is the effective shear- modulus of the bimaterial interface \(^{19}\) and \(R\) here is the radius of the fault, which can be found from the icequake corner frequency, if assuming a + +<--- Page Split ---> + +symmetric, circular fault \(^{66,67}\) . However, this equation still has two unknowns: \(G^{*}\) , the effective shear-modulus that we require to calculate the slip; and \(\mathcal{L}\) , the critical slip distance, otherwise referred to as the state evolution distance. For the purposes of this study, we approximate \(\mathcal{L}\) to remain constant, but allow \(G^{*}\) to vary with effective-normal-stress, which from granular material theory \(^{74}\) is assumed to take the generic empirical form, + +\[G^{*} = A\bar{\sigma}^{n} + C\quad (17),\] + +where \(A\) , \(n\) and \(C\) are constants to invert for. We use a least squares approach to minimize the function, + +\[f(\bar{\sigma}_{c},R,a,b,A,n,C,L) = l n\left(\frac{(A\bar{\sigma}_{c}^{n} + C)\mathcal{L}}{b - a}\right) - \ln (R\bar{\sigma}_{c}) \quad (18),\] + +where \(\bar{\sigma}_{c}\) and \(R\) vary for each icequake, and \(A\) , \(n\) , \(C\) and \(\mathcal{L}\) are varied to minimize the function. \(\bar{\sigma}_{c}\) is taken to be the effective- normal- stress for the first 100 icequakes when a cluster becomes active, as calculated using Equation 13. These parameters are found to be \(A = 22,000\) , \(n = 0.78\) , \(C = 8,200\) \(Pa\) and \(\mathcal{L} = 7.7 \times 10^{- 5} m\) , with the result of the minimization shown in Extended Data Fig. 4. Now \(\mathcal{L}\) can be substituted into Equation 15 to find the bimaterial shear- modulus, \(G^{*}\) . The shear- modulus of the bed, \(G_{bed}\) can then be found using the Poisson ratios of ice (1/3) and till (0.49), which gives \(G^{*} \approx 3.5 G_{bed}^{19}\) . Granular material theory, or at least the relationship of Equation 17, is thought to still hold for clast- over- bedrock sliding since the shear- modulus will still be related to some exponent, \(n\) , of \(\bar{\sigma}\) , even if that exponent were \(\sim 0\) . + +Equation 14 can then be used to find the slip, \(d\) , associated with a single icequake, for the effective- normal- stress applied to the fault at that particular time. We also calculate the approximate slip- rate associated with these highly repetitive icequakes. If one assumes that all the slip when an icequake cluster is active is accommodated seismically, then one can calculate the slip- rate per day, \(v_{slip}\) , + +\[v_{slip} = \frac{d}{t_{inter - event}} \quad (19),\] + +The methods described above allow us to calculate the total shear- stress, \(\tau\) , and the slip, \(d\) , at the bed. These two parameters can provide observational constraint on ice dynamics models of ice streams. + +## A note on assumptions + +A number of assumptions are made to make the derivation of basal shear- stress and slip from icequake observations and a rate- and- state friction model mathematically tractable. There are several assumptions that warrant particular emphasis. The first is the assumption that all slip at an individual sticky- spot is accommodated seismically while that cluster is active. The highly repetitive nature of the icequakes (see Extended Data Fig. 1 and \(^{28}\) ), with approximately constant inter- event times between consecutive icequakes in a cluster, is indicative of the stability of each sticky- spot (see Fig. 2), justifying this assumption. Secondly, a Brune model \(^{63}\) is assumed to describe the earthquake source characteristics. While such a model is likely an approximation for the complex physics of earthquake rupture, it is a common assumption for other earthquake studies that is likely also a valid approximation for the stick- slip icequakes presented here. Thirdly, we approximate that the time- derivative of the state- variable in the rate- and- state + +<--- Page Split ---> + +friction model, \(\frac{\partial \theta}{\partial t}\) , equals zero during an individual icequake cycle. This approximation is valid if slip on the fault is sufficiently fast and if little damage occurs at the fault, compared to more complex earthquake faults. Obviously, this is only an approximation, as damage does likely occur at the fault, at least for the clasts- over- bedrock slip case (regime I, see Fig. 5). Furthermore, an underestimation bias in slip may be introduced by the assumption of no fault frictional heating. Fault frictional heating would reduce the seismic radiation efficiency from our approximation of one \(^{72}\) . The final assumption we emphasize here is that we assume that the icequakes at the beginning of an icequake cluster nucleate at approximately the critical normal stress for nucleation, \(\sigma_{c}\) , rather than at some arbitrary value above it. The icequake slip calculations are dependent upon this assumption. This assumption would not be valid for sporadic earthquakes on complex faults, as shear- stresses could build to different values before failure for each earthquake, even with constant effective- normal- stresses, due to fault heterogeneity. Nor would it necessarily be valid if the driving shear- stress were perturbed over time- scales shorter than the inter- event time, for example by interactions with other icequake clusters. However, although icequake faults still exhibit a degree of heterogeneity due to an inhomogeneous distribution of clasts, this heterogeneity has negligible impact upon the consistency of both the inter- event times and shear- stresses between consecutive icequakes at a given sticky- spot (see Fig. 2). Furthermore, there are only a small number of active icequake clusters at any given time, which are spatially isolated from one another. The consistency in inter- event times and shear- stresses observed in our data, in agreement with similar, laboratory- generated icequakes \(^{12}\) , provides us with confidence in our assumption of icequakes nucleating at the critical nucleation stress, \(\sigma_{c}\) . + +## Additional references: + +58. Kufner, S. et al. Microseismic icequake catalogue, Rutford ice stream (west Antarctica), November 2018 to February 2019 (version 1.0). UK Polar Data Centre, Nat. Environ. Res. Counc. UK Res. Innov. (2021) doi:10.5285/B809A040-8305-4BC5-BAFF-76AA2B823734. +59. Winder, T. et al. QuakeMigrate v1.0.0. Zenodo (2021) doi:10.5281/zenodo.4442749. +60. Lomax, A. & Virieux, J. Probabilistic earthquake location in 3D and layered models. Adv. Seism. Event Locat. Vol. 18 Ser. Mod. Approaches Geophys. 101–134 (2000). +61. Hudson, T. S. TomSHudson/SeisSrcMoment: First formal release (Version 1.0.0). Zenodo (2020) doi:http://doi.org/10.5281/zenodo.4010325. +62. Aki, K. & Richards, P. G. Quantitative Seismology. (University Science Books, 2002). +63. Brune, J. N. Tectonic Stress and the Spectra of Seismic Shear Waves from Earthquakes. J. Geophys. Res. 75, 4997–5009 (1970). +64. Peters, L. E., Anandakrishnan, S., Alley, R. B. & Voigt, D. E. Seismic attenuation in glacial ice: A proxy for englacial temperature. J. Geophys. Res. Earth Surf. 117, 1–10 (2012). +65. Madariaga, R. Dynamics of an expanding circular fault. Bull. Seismol. Soc. Am. 66, 639–666 (1976). +66. Kaneko, Y. & Shearer, P. M. Seismic source spectra and estimated stress drop derived from cohesive-zone models of circular subsurface rupture. Geophys. J. Int. 197, 1002–1015 (2014). +67. Kaneko, Y. & Shearer, P. M. Variability of seismic source spectra, estimated stress drop, and radiated energy, derived from cohesive-zone models of symmetrical and asymmetrical circular and elliptical ruptures. J. Geophys. Res. Solid Earth 120, 1053–1079 (2015). +68. Eshelby, J. D. The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc. R. Soc. London. Ser. A. Math. Phys. Sci. 241, 376–396 (1957). +69. Ruina, A. Slip instability and state variable friction laws. J. Geophys. Res. 88, 10359–10370 (1983). +70. Hanks, T. C. & McGuire, R. K. The character of high-frequency strong ground motion. Bull. + +<--- Page Split ---> + +767 + +768 + +769 + +770 + +771 + +772 + +773 + +774 + +775 + +776 + +777 + +778 + +779 + +780 + +781 + +782 + +783 + +784 + +785 + +786 + +787 + +788 + +789 + +790 + +791 + +792 + +793 + +794 + +795 + +796 + +797 + +798 + +799 + +795 + +796 + +797 + +798 + +799 + +800 + +801 + +802 + +803 + +<--- Page Split ---> + +## Extended Data Figures + +![](images/Extended_Data_Figure_2.jpg) + + +Extended Data Fig. 1. Examples of icequake waveforms and spectra. (a) 30 minutes of continuous data for the Z component of station R3030. Approximate icequake P- phase arrival times associated with a single cluster are shown by the green lines. (b), (c), (d) Stacked waveform data on the Z-, N- and E- components for 173 events in a cluster at station R3030, located at the center of the network. Red line indicates P- phase arrival. Blue lines indicate S- phase arrivals. Grey shading represents the standard deviation of the stacked data. (e) Spectrum for one event within the cluster at station R3030. Waveform data in (a) to (d) are filtered between 10 Hz and 120 Hz. + +<--- Page Split ---> +![](images/Extended_Data_Figure_4.jpg) + +
Extended Data Fig. 2. Quality factor (Q) and corner frequency \((f_{c})\) distributions for the icequakes in this experiment. (a) Histogram of Q. (b) Histogram of \(f_{c}\) . Values for each icequake are averaged for all individual station observations.
+ +![](images/Extended_Data_Figure_5.jpg) + + +Extended Data Fig. 3. Schematic Fig. describing the rate- and state- frictional model as a block- slider model. (a) Diagram of the block- slider model, showing the driving shear- stress, \(\tau\) , the effective- normal- stress, \(\bar{\sigma}\) , and the system spring constant \(k\) . (b) Accumulated shear- stress vs. time for a series of consecutive icequakes. (c) Shear- stress at the fault at a particular time as + +<--- Page Split ---> + +predicted by the rate-and-state model \(^{31}\) . (d) The stick-slip icequake cycle, with the numbers corresponding to the relevant stress states labelled in (b). + +![PLACEHOLDER_24_0] + +
Extended Data Fig. 4. Results of the least squares inversion of Equation 17. Blue scatter points are the data and red scatter points show the least-squares inversion result.
+ +<--- Page Split ---> +![PLACEHOLDER_25_0] + +
Extended Data Fig. 5. Rate-and-state friction model sensitivity analysis. Plot of the sensitivity in frictional shear-stress at the bed, \(\tau_{bed}\) , and slip-rate at the bed, \(\nu_{slip}\) , with perturbation of the key observational parameters. The reference values used to normalize the variations are the average values of \(\tau_{bed}\) and \(\nu_{slip}\) observed at all the clusters. The magnitude of variation in each parameter are summarized in Table S1. See supplementary text for further details.
+ +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- frictionandslipicequakespaperHudson2021supplementaryinformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98_det.mmd b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..af96065c895f10422c61bc4aa8ac95ec9e910a01 --- /dev/null +++ b/preprint/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98/preprint__17150ed3cdf92ba18747ceec229592e5c148879c6154ff02e0bb777923292e98_det.mmd @@ -0,0 +1,666 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 792, 174]]<|/det|> +# Friction and slip measured at the bed of an Antarctic ice stream + +<|ref|>text<|/ref|><|det|>[[44, 195, 590, 238]]<|/det|> +Thomas Hudson ( \(\boxed{\bullet}\) thomas.hudson@earth.ox.ac.uk) University of Oxford https://orcid.org/0000- 0003- 2944- 883X + +<|ref|>text<|/ref|><|det|>[[44, 243, 857, 285]]<|/det|> +Sofia- Katerina Kufner German Research Centre for Geosciences Potsdam https://orcid.org/0000- 0002- 9687- 5455 + +<|ref|>text<|/ref|><|det|>[[44, 290, 956, 333]]<|/det|> +Alex Brisbourne British Antarctic Survey, Natural Environment Research Council https://orcid.org/0000- 0002- 9887- 7120 + +<|ref|>text<|/ref|><|det|>[[44, 337, 590, 378]]<|/det|> +Michael Kendall University of Oxford https://orcid.org/0000- 0002- 1486- 3945 + +<|ref|>text<|/ref|><|det|>[[44, 383, 619, 424]]<|/det|> +Andrew Smith British Antarctic Survey https://orcid.org/0000- 0001- 8577- 482X + +<|ref|>text<|/ref|><|det|>[[44, 429, 316, 470]]<|/det|> +Richard Alley Pennsylvania State University + +<|ref|>text<|/ref|><|det|>[[44, 475, 334, 516]]<|/det|> +Robert Arthern N.E.R.C. British Antarctic Survey + +<|ref|>text<|/ref|><|det|>[[44, 521, 586, 563]]<|/det|> +Tavi Murray Swansea University https://orcid.org/0000- 0001- 6714- 6512 + +<|ref|>text<|/ref|><|det|>[[44, 603, 102, 620]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 641, 137, 660]]<|/det|> +Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 679, 331, 698]]<|/det|> +Posted Date: January 13th, 2022 + +<|ref|>text<|/ref|><|det|>[[44, 716, 475, 736]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 1214097/v1 + +<|ref|>text<|/ref|><|det|>[[44, 754, 910, 797]]<|/det|> +License: © \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[186, 105, 810, 127]]<|/det|> +# Friction and slip measured at the bed of an Antarctic ice stream + +<|ref|>text<|/ref|><|det|>[[133, 171, 863, 210]]<|/det|> +T.S. Hudson \(^{1*}\) , SK Kufner \(^{2}\) , A.M. Brisbourne \(^{2}\) , JM Kendall \(^{1}\) , A.M. Smith \(^{2}\) , R.B. Alley \(^{3}\) , R.J. Arthern \(^{2}\) , T. Murray \(^{4}\) + +<|ref|>sub_title<|/ref|><|det|>[[115, 231, 216, 248]]<|/det|> +## Affiliations: + +<|ref|>text<|/ref|><|det|>[[144, 264, 870, 300]]<|/det|> +\(^{1}\) Department of Earth Sciences, University of Oxford; 3 South Parks Rd, Oxford, OX1 3AN, UK + +<|ref|>text<|/ref|><|det|>[[144, 317, 840, 338]]<|/det|> +\(^{2}\) UKRI British Antarctic Survey; High Cross, Madingley Rd, Cambridge, CB3 0ET, UK + +<|ref|>text<|/ref|><|det|>[[144, 353, 866, 390]]<|/det|> +\(^{3}\) Department of Geosciences, and Earth and Environmental Systems Institute, Pennsylvania State University, University Park, PA 16802, USA + +<|ref|>text<|/ref|><|det|>[[144, 404, 856, 440]]<|/det|> +\(^{4}\) Department of Geography, Swansea University, Swansea, Singleton Park, Swansea, SA2 8PP, UK + +<|ref|>text<|/ref|><|det|>[[144, 465, 697, 484]]<|/det|> +\*Corresponding author email address: thomas.hudson@earth.ox.ac.uk + +<|ref|>sub_title<|/ref|><|det|>[[115, 515, 196, 531]]<|/det|> +## Abstract: + +<|ref|>text<|/ref|><|det|>[[113, 539, 880, 768]]<|/det|> +The slip of glaciers over the underlying bed is the dominant mechanism governing the migration of ice from land into the oceans, contributing to sea- level rise. Yet glacier slip remains poorly understood or constrained by observations. Here we observe both frictional shear- stress and slip at the bed of an ice stream, using 100,000 repetitive stick- slip icequakes from Rutford Ice Stream, Antarctica. Basal shear- stresses and slip- rates vary from \(10^{4}\) to \(10^{7}Pa\) and 0.2 to \(1.5mday^{- 1}\) , respectively. Friction and slip vary temporally over the order of hours and spatially over 10s of meters, caused by corresponding variations in ice- bed interface material and effective- normal- stress. Our findings also suggest that the bed is substantially more complex than currently assumed in ice stream models and that basal effective- normal- stresses may be significantly higher than previously thought. The observations also provide previously unresolved constraint of the basal boundary conditions of ice dynamics models. This is critical for constraining the primary contribution of ice mass loss in Antarctica, and hence the endeavor to reduce uncertainty in sea- level rise projections. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 81, 210, 98]]<|/det|> +## Main Text: + +<|ref|>text<|/ref|><|det|>[[115, 105, 881, 245]]<|/det|> +Glacier slip is the primary mechanism governing the migration of ice from land into the oceans, providing a major contribution to sea- level rise \(^{1,2}\) . Friction at the bed of a glacier fundamentally limits the speed at which the ice can slip. This friction is controlled by a number of factors, including bed material, the presence of debris in basal ice, and hydrological systems that modulate effective- normal- stresses. However, basal friction and slip remain poorly understood or constrained by observations \(^{1,3,5}\) . Such observational constraint of friction and slip is critical for the verification of ice- bed boundary condition assumptions in ice dynamics models, which are required to reduce uncertainty in corresponding sea- level rise projections \(^{4,6,7}\) . + +<|ref|>text<|/ref|><|det|>[[115, 277, 879, 469]]<|/det|> +Previous contributions to address this critical observational void come from laboratory- based experiments \(^{8 - 13}\) geophysical studies \(^{14 - 23}\) , and borehole measurements \(^{24,25}\) . However, to date there have been challenges with such approaches. Laboratory experiments provide insight into fundamental physical properties of the bed material (till) \(^{10}\) and ice- bed interface interactions \(^{12}\) but are limited by scale and the diversity of natural glacier beds. Geophysical studies have measured the in- situ bed strength, but with sparse spatial and temporal resolution \(^{14}\) . Borehole measurements of slip are not only sparse, but have not been accompanied by measurements of shear- stress, making quantitative interpretations difficult. The ice streams and outlet glaciers that contribute the majority of ice flux into the oceans likely have active, spatially- and temporally- varying hydrological systems \(^{26,27}\) , perturbing basal friction and slip over short time- and length- scales. An observational void therefore remains. + +<|ref|>text<|/ref|><|det|>[[115, 500, 878, 763]]<|/det|> +Here we address this observational void by using icequakes to provide the first spatially- mapped, in- situ observations of both frictional drag and slip- rate at the bed of an ice stream. These icequakes are generated by the sudden release of strain at or near the ice- bed interface. The dataset analyzed comprises 100,000 icequakes \(^{28}\) from Rutford Ice Stream (RIS), Antarctica (see Fig. 1). The icequakes originate approximately at the center of the ice stream, where the dominant source of drag is postulated to originate from the bed rather than from the shear margins. These icequakes nucleate in clusters that are highly repetitive (see Extended Data Fig. 1), with near- constant inter- event times of the order of 100s of seconds and icequakes clusters active for hours to days \(^{28}\) . These icequakes are inferred to be at the bed from: their hypocentral depths; the consistent flow- and bed- parallel orientation of their double- couple focal mechanism slip- vectors; and full- waveform modelling a typical RIS icequake source \(^{20,28 - 30}\) . The tight spatial clustering and repetitive nature motivate our use of a rate- and- state friction law in combination with icequake observations to investigate the glacier sliding process. This rate- and- state friction law \(^{31}\) also enables the calculation of other basal parameters including bed shear moduli and insight into the modulation of glaciological effective- normal- pressures. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[217, 88, 784, 380]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 397, 876, 520]]<|/det|> +
Fig. 1. Seismic network and icequake data at Rutford Ice Stream, Antarctica. (a) Location of Rutford Ice Stream (RIS) relative to the Antarctic continent. Topography is from Bedmap232. (b) Map of network with respect to RIS shear-margins. (c) Map of the experiment and icequake data at Rutford Ice Stream, from November 2018 to February 2019. Red scatter points show icequake locations. All icequakes are approximately at ice stream bed28. Green inverted triangles show geophone locations. Bed topography data are from the literature33. Pink dashed line indicates a bed-character boundary from the literature33.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 552, 189, 569]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 579, 441, 596]]<|/det|> +## Observed ice-bed friction and slip-rate + +<|ref|>text<|/ref|><|det|>[[112, 604, 877, 852]]<|/det|> +The icequake source properties and inter- event times are used in combination with a rate- and- state friction law to calculate: fault effective- normal- stress \((\bar{\sigma})\) ; total frictional shear- stress, or drag per unit area \((\tau)\) ; shear- modulus \((G_{bed})\) ; slip \((d)\) ; and slip- rate \((\nu_{slip})\) at the bed of RIS. Fig. 2 shows these results for a representative subset of icequake clusters. Fault effective- normal- stress, shear- stress and shear- modulus (Fig. 2a- c) vary by orders of magnitude between clusters, even after accounting for uncertainty. However, these parameters are all confined within expected physical limits. Effective- normal- stresses remain below the maximum ice overburden pressure, which is the upper possible limit for the average effective- normal- stress over the entire fault. The observed shear- stress ranges from \(\sim 10^{4}\) to \(10^{7}Pa\) . If the icequake cluster locations, or sticky- spots, contribute more drag than the surrounding bed, then sticky- spot shear- stresses could theoretically have a much higher limit than the average bed shear- stress. Although bed shear moduli vary significantly between clusters, the majority of the clusters' shear moduli agree with one of the only previous seismically- derived in- situ measurements, \(70MPa\) , from Whillans Ice Stream14. Additionally, measurements do not exceed the shear- modulus of ice. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 80, 870, 152]]<|/det|> +Slip-rates show smaller variations in amplitude, from \(\sim 0.2\) to \(1.5 \text{m day}^{- 1}\) , but have higher associated uncertainties due to their dependence on both shear-modulus and fault-area. While a number of clusters exhibit time-averaged slip-rates approximately equal to the steady-state surface velocity of RIS (dashed line, Fig. 2f) \(^{34}\) , other clusters have significantly lower slip-rates. + +<|ref|>image<|/ref|><|det|>[[230, 152, 771, 628]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 637, 881, 814]]<|/det|> +
Fig. 2. All icequake-derived basal sliding parameters through time. Data are a subset of icequake clusters over the period of \(5^{\text{th}}\) to \(15^{\text{th}}\) January 2019. Each colored line represents an individual icequake cluster. Uncertainties are shown by shaded regions. (a) Effective-normal-stress. Red dashed-dotted line indicates the maximum possible ice overburden pressure. (b) Total frictional basal shear-stress. (c) Bed shear-modulus. Previous estimates from literature are indicated by the dashed lines \(^{14,15}\) . (d) Slip associated with individual icequakes. (e) Inter-event time between icequakes in a cluster. (f) Equivalent daily slip-rate calculated from the slip and inter-event times in (d) and (e). All data is smoothed by applying a 100-event moving-average window. All uncertainties are estimated using calculus-derived uncertainty propagation methods. Sensitivity analysis of the rate-and-state model is shown in Extended Data Fig. 5.
+ +<|ref|>text<|/ref|><|det|>[[115, 844, 867, 915]]<|/det|> +Fig. 3a-c show the variation in effective-normal-stress, shear-stress and slip-rate for the entire experiment duration. Histograms of the stress and slip-rate distributions are shown in Fig. 3d-f. The normal and shear-stress histograms show a bimodal distribution, with more than two thirds of the icequakes having effective-normal-stresses lower than \(\sim 5 \times 10^{5} \text{Pa}\) and shear-stresses + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 840, 119]]<|/det|> +lower than \(2 \times 10^{5}\) Pa. Conversely, the slip-rates exhibit a unimodal distribution, tailing off below \(0.2 m \text{day}^{- 1}\) and above \(1.5 m \text{day}^{- 1}\) . + +<|ref|>image<|/ref|><|det|>[[230, 137, 763, 540]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 560, 881, 649]]<|/det|> +
Fig. 3. Basal effective-normal-stress, shear-stress and slip-rate for the entire experiment. Colored lines represent individual icequake clusters. (a) Effective-normal-stress on the fault. (b) Shear-stress through time. (c) Slip-rate through time. (d) to (f) Histograms of the respective time-series data in (a) to (c). Uncertainties in (a) to (c) are given by the shaded regions. Other labels as in Fig. 2. Uncertainties are as defined in Fig. 2.
+ +<|ref|>text<|/ref|><|det|>[[114, 678, 870, 802]]<|/det|> +Individual icequake clusters switch on and off, being active for the order of hours to days (see Fig. 2). Within single clusters, bed friction and slip are modulated by signals with dominant periods of \(\sim 6\) to 12 hours (see Fig. 2). However, although this alludes to tidal modulation of basal friction, and indeed surface velocities are known to be modulated by tidal frequencies \(^{34,35}\) , we cannot decipher a clear relationship between tidal signals propagated 40 km upstream from RIS's grounding line and our signals \(^{28}\) . We therefore do not discuss any link with tidal signals further. + +<|ref|>text<|/ref|><|det|>[[114, 833, 872, 905]]<|/det|> +The spatial distribution of average basal shear- stress, slip- rate and fault radius for each cluster over a \(7 \times 6\) km region are shown in Fig. 4. Shear- stresses are largest at the clusters farthest upstream, approximately where the bed properties are inferred to transition from unconsolidated to consolidated till \(^{33}\) (pink dashed- line, Fig. 1) and where the bed has shorter wavelength + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 80, 876, 257]]<|/det|> +topography than upstream that likely inhibits ice flow. Average slip- rate is spatially consistent across all clusters. This is expected, as our study site is located near the center of the ice stream, with no spatial variation in surface slip- rate34. Fault radius, defining the area of an icequake cluster sticky- spot is also measured (see Fig. 4c). Fault radii indicate that individual seismically active sticky- spots have areas \(< 2800 m^2\) . Only a small number of sticky- spots are active at any instant. This suggests that regions of sufficiently high basal friction to generate seismicity are confined to the minority of the bed at a given point in time, yet invoke significant basal drag. Aseismic regions between icequake clusters likely also contribute to the basal drag, presumably providing the dominant source of aseismic drag upstream of the unconsolidated- consolidated sediment boundary (pink dashed line, Fig. 1). + +<|ref|>image<|/ref|><|det|>[[155, 275, 842, 488]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 508, 850, 598]]<|/det|> +
Fig. 4. Spatial variability in average basal shear-stress, slip-rate and fault radius for the clusters. (a) Average shear-stress for the clusters. (b) Average slip-rate for the clusters. (c) Average fault radius for the clusters. Residual topography data is from ground-penetrating radar33. Size of scatter points indicates fault radius. Green inverted triangles indicate the locations of the network of receivers used in this study.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 630, 220, 648]]<|/det|> +## Discussion + +<|ref|>sub_title<|/ref|><|det|>[[115, 657, 416, 674]]<|/det|> +## Frictional shear-stress and slip-rate + +<|ref|>text<|/ref|><|det|>[[113, 682, 872, 857]]<|/det|> +The most important, immediate finding of this work is the ability to observe in- situ frictional shear- stress and slip- rate, the two critical parameters for constraining the basal drag boundary conditions of ice dynamics models. Our approach could be applied to any glacier that generates icequakes. Most fast- moving glaciers likely generate such icequakes, with the majority of glaciers on which seismometers have been deployed exhibiting at least some basal seismicity16,28,36- 44. Seismic tremor associated with sliding can also occur19,45, thought to initiate at the boundary between the conditionally- stable and unstable regimes of the rate- and- state friction model19,31. Indeed, the premise of this study was inspired by such observations19. However, due to the inability to extract both corner frequency and inter- event time information from tremor, it cannot be used to measure shear- stress and slip using our approach. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 80, 875, 220]]<|/det|> +Our confidence in the frictional shear- stress and slip- rate measurements is founded partially on the uncertainty amplitudes, but also fundamentally on the agreement between the observed basal slip- rates and GNSS- derived surface displacement34. This agreement validates assumptions of slip- dominant rather than deformation- dominant flow at RIS and the use of a rate- and- state model and assumptions of the icequake source properties. The small discrepancy between the surface and basal slip- rates is primarily due to uncertainty, except for a minority of particularly sticky- spots. These sticky- spots exhibit particularly strong frictional drag that significantly inhibits local ice flow, albeit for short durations of the order of hours to days. + +<|ref|>text<|/ref|><|det|>[[115, 251, 874, 377]]<|/det|> +Observed basal shear- stresses are of the order of \(10^{4}\) to \(10^{7}\) Pa, acting at sticky- spots with diameters of the order of 10 to \(60\mathrm{m}\) (see Fig. 4). Basal shear- stresses of the order \(10^{5}\) Pa are typical values used in ice dynamics models46 and laboratory experiments8 for RIS's surface slip- rate of \(\sim 400\mathrm{m / yr^{34}}\) . Basal shear- stresses of \(10^{6}\) to \(10^{7}\) Pa might initially appear inconsistently high compared to models and experiments47. However, these high friction sticky- spots are spatially small compared to the total bed area. Our results therefore imply that certain icequake clusters accommodate a considerable proportion of the total basal drag. + +<|ref|>sub_title<|/ref|><|det|>[[115, 409, 395, 426]]<|/det|> +## Icequake generation mechanisms + +<|ref|>text<|/ref|><|det|>[[114, 433, 874, 695]]<|/det|> +We propose that the icequakes are generated by at least one of two mechanisms, or sliding regimes. The presence of two sliding regimes is motivated by the physical system and the bimodal distributions observed in Fig. 3d,e. The regimes (see Fig. 5) are: regime I, rock- on- rock friction between ice- entrained clasts and bedrock at the fault interface; and regime II, where clasts plough through till, with failure accommodated by a till- on- till fault interface. Clasts are pieces of rock partially entrained into the ice (see Fig. 5). The presence of such clasts is discussed in the literature8,12,13,48,49. The motivations for these clast- based icequake models are that they can explain the rate- weakening friction required to generate icequakes13, that clasts are required to generate icequakes in laboratory environments12, and that such icequakes likely originate within one seismic wavelength of the ice- bed interface20. We suggest that the highest effective- normal- stress icequakes exhibit regime I sliding, since this regime allows for the average effective- normal- stress over the entire fault- area to be concentrated over much smaller clast- bedrock contact areas. Similarly, we postulate that the lower effective- normal- stress icequakes are associated with regime II sliding, although we cannot rule out that all icequakes are generated via regime I. + +<|ref|>sub_title<|/ref|><|det|>[[115, 727, 533, 745]]<|/det|> +## Effective normal-stress vs. effective fluid pressure + +<|ref|>text<|/ref|><|det|>[[115, 751, 867, 912]]<|/det|> +Our results imply significant temporal variation in basal effective- normal- stress. Such increases and decreases in effective- normal- stress are inferred to be caused by corresponding decreases and increases in basal water pressure16,50,51. However, while the icequake- derived effective- normal- stresses, \(\bar{\sigma}\) , averaged over the entire fault are equivalent to the average glaciological effective pressure, \(P_{eff} = P_{ice overburden} - P_{water}\) , within that same fault- area, asperities and bed heterogeneity on length- scales shorter than the fault diameter could significantly perturb local effective- normal- pressures. Although all our measured effective- normal- stresses remain below the ice overburden pressure, current hydrological models cannot reconcile glaciological effective pressures greater than \(\sim 0.5\mathrm{MPa}\) for expected till porosities. Sparse observations of + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 81, 881, 191]]<|/det|> +effective- normal- pressures at RIS from borehole measurements find \(P_{eff} \approx 0.2 MPa^{49}\) , although till acoustic impedance measurements at RIS suggest that dewatering is possible \(^{23}\) . Dewatered till would imply \(P_{eff} = P_{ice overburden}\) . Our highest observed effective- normal- stresses therefore suggest either: that our understanding of bed characteristics and associated physical models may have to be revisited, at the very least for RIS; or that the rate- and- state model does not adequately describe icequake physics. + +<|ref|>text<|/ref|><|det|>[[114, 221, 881, 363]]<|/det|> +We suggest three possible explanations for resolving the discrepancy between observed and theoretical maximum effective- normal- stresses. Firstly, prominent bedrock outcrops could significantly inhibit ice flow, allowing stoss- side effective- normal- stresses of the order of at least \(1 MPa\) to develop \(^{52}\) . A second explanation again lies with bedrock outcrops, whereby impermeable bedrock might inhibit the transport of fluids, facilitating dewatered regions. A third explanation is that till porosities are far lower than conceived in current models of bed properties, possibly supported by till impedance measurements \(^{23}\) . These suggestions are not exhaustive. We only suggest here that our results motivate new models to explain such observations. + +<|ref|>text<|/ref|><|det|>[[115, 393, 880, 450]]<|/det|> +If effective- normal- stresses are related to \(P_{eff}\) at sticky- spots, then they provide an observational foundation for calibrating basal- fluid- pressures assumed in: laboratory- experiments \(^{8}\) ; ice dynamics models \(^{46}\) ; and glacier basal hydrology and tidal forcing \(^{34,35}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 480, 611, 499]]<|/det|> +## Enhanced knowledge of Rutford Ice Stream bed conditions + +<|ref|>text<|/ref|><|det|>[[114, 505, 881, 647]]<|/det|> +The considerable variation in bed properties observed at RIS are presented as an example of the enhanced knowledge of the bed properties that our approach provides. Firstly, for the unconsolidated till (label 1, Fig. 5) and much of the consolidated till (label 2, Fig. 5), the effective- normal- stresses are too low to generate the unstable stick- slip conditions required for icequake nucleation. Within consolidated till regions (label 2, Fig. 5), there are small zones that become seismically active if the effective- normal- stress is sufficiently high (label 3, Fig. 5). These sticky- spots turn on and off, modulated by changes in effective- normal- stress, bed strength and bed material. + +<|ref|>text<|/ref|><|det|>[[114, 678, 878, 852]]<|/det|> +Frictional shear- stress at an individual sticky- spot can vary temporally by up to an order of magnitude and spatially by several orders of magnitude. This variation occurs over the order of hours and 100s meters. This implies that there are both till and bedrock outcrops, combined with an active hydrological system or permeable bed capable of such variable spatial and temporal variations in basal fluid pressure. The regions exhibiting the highest shear- stresses are at the upstream edges of local topographic highs near the unconsolidated- consolidated bed boundary (label 6, Fig. 5). This is likely because resistive stresses of these materially stronger, topographic highs can accommodate more basal drag. Some regions of consolidated till might contain pockets of melt water (label 5, Fig. 5). However, we cannot observe such a phenomenon seismically as these patches would have an approximately zero shear- modulus. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[252, 85, 744, 483]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 498, 876, 587]]<|/det|> +
Fig. 5. Schematic diagram summarizing the findings of this study in relation to basal friction and slip with bed characteristics. Bed properties are labelled in the legend. Numbered points are referred to in the text. Note that features not to scale, but arranged approximately according to spatial trends in Fig. 4. Regime I and regime II are shown schematically, with regime I being clast-on-rock icequake slip behavior and regime II being till-on-till slip behavior.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 617, 279, 635]]<|/det|> +## Wider implications + +<|ref|>text<|/ref|><|det|>[[114, 641, 876, 784]]<|/det|> +Our results show that much of the basal drag at an ice stream can be accommodated within small zones of significantly higher- than- average friction. This could have a profound impact on how sliding is formulated in ice dynamics models. However, although friction varies significantly, average basal slip- rates remain predominantly stable at RIS. This is encouraging for current modelling efforts since if the temporally- and spatially- averaged slip- rates are approximately constant, then perhaps such models are not required to be sensitive to small- scale, rapid variations in bed friction. Our observations quantify the highly variable bed properties over a sufficient duration required to test such a hypothesis. + +<|ref|>text<|/ref|><|det|>[[114, 814, 856, 884]]<|/det|> +Another important question to address is how our approach could be implemented at ice sheet scale. One could deploy temporary seismic arrays on important ice streams and outlet glaciers for short durations. A number of targeted deployments would allow verification of the link between surface- and basal- velocity at ice sheet scales2. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 80, 876, 203]]<|/det|> +A further question that this study raises is could a rate-and-state friction model used for the icequake sliding analysis also be used as a mathematical basis for informing sliding laws used in ice dynamics models more generally. Such a model was recently proposed to describe surging glacier behavior \(^{53}\) and has been validated at laboratory scale \(^{12,13}\) . The rate-and-state model meets the conditional stability requirement, not allowing runaway acceleration of a glacier. A more comprehensive comparison to sliding laws for deformable beds, showing agreement for surface velocities \(>100 \mathrm{m / yr}\) , is provided in the supplementary text. + +<|ref|>text<|/ref|><|det|>[[115, 234, 878, 324]]<|/det|> +Finally, these icequakes observations can aid the understanding of earthquake mechanics more generally. Even the smallest magnitude icequakes \((M_{w} \approx - 1.5)\) have high signal- to- noise- ratios, and so could elucidate any lower limits on the fundamental size of earthquake nucleation for given fault properties \(^{54 - 56}\) . Additionally, icequakes in this study have stress- drops that vary with magnitude, contrary to magnitude- invariant stress- drops observed for larger earthquakes \(^{57}\) . + +<|ref|>text<|/ref|><|det|>[[115, 355, 867, 425]]<|/det|> +Our findings show that icequakes can provide the critical observations required to constrain the highly variable friction at the bed of an Antarctic ice stream. Applying such observational constraint to ice dynamics models would reduce uncertainty in corresponding sea- level rise projections. + +<|ref|>sub_title<|/ref|><|det|>[[115, 458, 208, 474]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[111, 481, 880, 914]]<|/det|> +1. Morlighem, M. et al. Spatial patterns of basal drag inferred using control methods from a full-Stokes and simpler models for Pine Island Glacier, West Antarctica. Geophys. Res. Lett. 37, 1–6 (2010). +2. Rignot, E., Mouginot, J. & Scheuchl, B. Ice Flow of the Antarctic Ice Sheet. Science (80-. ). 333, 1427–1430 (2011). +3. Ritz, C. et al. Potential sea-level rise from Antarctic ice-sheet instability constrained by observations. Nature 528, 115–118 (2015). +4. Tsai, V. C., Stewart, A. L. & Thompson, A. F. Marine ice-sheet profiles and stability under Coulomb basal conditions. J. Glaciol. 61, 205–215 (2015). +5. Joughin, I., Smith, B. E. & Schoof, C. G. Regularized Coulomb Friction Laws for Ice Sheet Sliding: Application to Pine Island Glacier, Antarctica. Geophys. Res. Lett. 46, 4764–4771 (2019). +6. Oppenheimer, M. et al. Sea Level Rise and Implications for Low-Lying Islands, Coasts and Communities. IPCC Spec. Rep. Ocean Cryosph. a Chang. Clim. 321–445 (2019). +7. Schlegel, N. J. et al. Exploration of Antarctic Ice Sheet 100-year contribution to sea level rise and associated model uncertainties using the ISSM framework. Cryosphere 12, 3511–3534 (2018). +8. Zoet, L. K. & Iverson, N. R. A slip law for glaciers on deformable beds. Science (80-. ). 368, 76–78 (2020). +9. Iverson, N., Baker, R., Hooke, R., Hanson, B. & Jansson, P. Coupling between a glacier and a soft bed: I. A relation between effective pressure and local shear stress determined from till elasticity. J. Glaciol. 45, 31–40 (1999). +10. Tulaczyk, S., Kamb, W. B. & Engelhart, H. F. Basal mechanics of Ice Stream B, west Antarctica: 1. Till mechanics. J. Geophys. Reasearch 105, 463–481 (2000). +11. Leeman, J. R., Valdez, R. D., Alley, R. B., Anandakrishnan, S. & Saffer, D. M. Mechanical and hydrologic properties of Whillans Ice Stream till: Implications for basal strength and stick-slip failure. J. Geophys. Res. Earth Surf. 121, 1–17 (2016). +12. Zoet, L. K. et al. Application of Constitutive Friction Laws to Glacier Seismicity. Geophys. Res. Lett. 47, 1–9 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 880, 912]]<|/det|> +13. Lipovsky, B. P. et al. Glacier sliding, seismicity and sediment entrainment. Ann. Glaciol. 60, 182–192 (2019). +14. Blankenship, D. D., Bentley, C. R., Rooney, S. T. & Alley, R. B. Till beneath Ice Stream B 1. Properties derived from seismic travel times. J. Geophys. Res. 92, 8903–8911 (1987). +15. Zoet, L. K., Anandakrishnan, S., Alley, R. B., Nyblade, A. A. & Wiens, D. A. Motion of an Antarctic glacier by repeated tidally modulated earthquakes. Nat. Geosci. 5, 623–626 (2012). +16. Gräff, D. & Walter, F. Changing friction at the base of an Alpine glacier. Sci. Rep. 11, 1–10 (2021). +17. Winberry, J. P., Anandakrishnan, S., Alley, R. B., Bindschadler, R. A. & King, M. A. Basal mechanics of ice streams: Insights from the stick-slip motion of Whillans Ice Stream, West Antarctica. J. Geophys. Res. 114, 1–11 (2009). +18. Anandakrishnan, S. & Bentley, C. R. Micro-earthquakes beneath ice streams B and C, West Antarctica: observations and implications. J. Glaciol. 39, 455–462 (1993). +19. Lipovsky, B. P. & Dunham, E. M. Tremor during ice-stream stick slip. Cryosphere 10, 385–399 (2016). +20. Hudson, T. S. et al. Icequake Source Mechanisms for Studying Glacial Sliding. J. Geophys. Res. Earth Surf. 125, (2020). +21. Wiens, D. A., Anandakrishnan, S., Winberry, J. P. & King, M. A. Simultaneous teleseismic and geodetic observations of the stick-slip motion of an Antarctic ice stream. Nature 453, 770–774 (2008). +22. Barcheck, C. G., Tulaczyk, S., Schwartz, S. Y., Walter, J. I. & Winberry, J. P. Implications of basal micro-earthquakes and tremor for ice stream mechanics: Stick-slip basal sliding and till erosion. Earth Planet. Sci. Lett. 486, 54–60 (2018). +23. Smith, A. M. & Murray, T. Bedform topography and basal conditions beneath a fast-flowing West Antarctic ice stream. Quat. Sci. Rev. 28, 584–596 (2009). +24. Engelhardt, H. Basal sliding of Ice Stream B, West Antarctica. J. Glaciol. 44, 223–230 (1998). +25. Truffer, M., Harrison, W. D. & Echelmeyer, K. A. Glacier motion dominated by processes deep in underlying till. J. Glaciol. 46, 213–221 (2000). +26. Rosier, S. H. R., Gudmundsson, G. H. & Green, J. A. M. Temporal variations in the flow of a large Antarctic ice stream controlled by tidally induced changes in the subglacial water system. Cryosph. 9, 1649–1661 (2015). +27. Damsgaard, A. et al. Ice flow dynamics forced by water pressure variations in subglacial granular beds. Geophys. Res. Lett. 43, 12,165-12,173 (2016). +28. Kufner, S. et al. Not all Icequakes are Created Equal: Basal Icequakes Suggest Diverse Bed Deformation Mechanisms at Rutford Ice Stream, West Antarctica. J. Geophys. Res. Earth Surf. 126, (2021). +29. Smith, E. C., Smith, A. M., White, R. S., Brisbourne, A. M. & Pritchard, H. D. Mapping the ice-bed interface characteristics of Rutford Ice Stream, West Antarctica, using microseismicity. J. Geophys. Res. Earth Surf. 120, 1881–1894 (2015). +30. Smith, A. M. Microearthquakes and subglacial conditions. Geophys. Res. Lett. 33, 1–5 (2006). +31. Scholz, C. H. Earthquakes and friction laws. Nature 391, 37–42 (1998). +32. Fretwell, P. et al. Bedmap2: Improved ice bed, surface and thickness datasets for Antarctica. Cryosphere 7, 375–393 (2013). +33. King, E. C., Pritchard, H. D. & Smith, A. M. Subglacial landforms beneath Rutford Ice Stream, Antarctica: detailed bed topography from ice-penetrating radar. Earth Syst. Sci. Data 8, 151–158 (2016). +34. Minchew, B. M., Simons, M., Riel, B. & Milillo, P. Tidally induced variations in vertical and horizontal motion on Rutford Ice Stream, West Antarctica, inferred from remotely sensed observations. J. Geophys. Res. Earth Surf. 122, 167–190 (2017). +35. Rosier, S. H. R., Gudmundsson, G. H. & Green, J. A. M. Temporal variations in the flow of a large Antarctic ice stream controlled by tidally induced changes in the subglacial water system. Cryosphere 9, 1649–1661 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[110, 78, 884, 916]]<|/det|> +36. Stuart, G., Murray, T., Brisbourne, A., Styles, P. & Toon, S. Seismic emissions from a surging glacier: Bakaninbreen, Svalbard. Ann. Glaciol. 42, 151–157 (2005). +37. Walter, F., Deichmann, N. & Funk, M. Basal icequakes during changing subglacial water pressures beneath Gornergletscher, Switzerland. Mitteilungen der Versuchsanstalt fur Wasserbau, Hydrol. und Glaziologie an der Eidgenoss. Tech. Hochschule Zurich 54, 511–521 (2008). +38. Hudson, T. S., Smith, J., Brisbourne, A. & White, R. Automated detection of basal icequakes and discrimination from surface crevassing. Ann. Glaciol. 60, 1–11 (2019). +39. Roeoesli, C., Helmstetter, A., Walter, F. & Kissling, E. Meltwater influences on deep stick-slip icequakes near the base of the Greenland Ice Sheet. J. Geophys. Res. Earth Surf. 1–18 (2016) doi:10.1002/2015JF003601. +40. McBrearty, I. W., Zoet, L. K. & Anandakrishnan, S. Basal seismicity of the Northeast Greenland Ice Stream. J. Glaciol. 1–17 (2020) doi:10.1017/jog.2020.17. +41. Bindschadler, R. A., King, M. A., Alley, R. B., Anandakrishnan, S. & Padman, L. Tidally controlled stick-slip discharge of a West Antarctic ice stream. Science (80-. ). 301, 1087–1089 (2003). +42. Anandakrishnan, S. & Alley, R. B. Ice Stream C, Antarctica, sticky-spots detected by microearthquake monitoring. Ann. Glaciol. 20, 183–186 (1994). +43. Deichmann, N. et al. Evidence for deep icequakes in an Alpine glacier. Ann. Glaciol. 31, 85–90 (2000). +44. Köhler, A., Maupin, V., Nuth, C. & Pelt, W. V. A. N. Characterization of seasonal glacial seismicity from a single-station on-ice record at Holtedahlfonna, Svalbard. 1–14 (2019) doi:10.1017/aog.2019.15. +45. Winberry, J. P., Anandakrishnan, S., Wiens, D. A. & Alley, R. B. Nucleation and seismic tremor associated with the glacial earthquakes of Whillans Ice Stream, Antarctica. Geophys. Res. Lett. 40, 312–315 (2013). +46. Cornford, S. L. et al. Results of the third Marine Ice Sheet Model Intercomparison Project (MISMIP+). Cryosphere 14, 2283–2301 (2020). +47. Ranganathan, M., Minchew, B., Meyer, C. R. & Gudmundsson, G. H. A new approach to inferring basal drag and ice rheology in ice streams, with applications to West Antarctic Ice Streams. J. Glaciol. 67, 229–242 (2021). +48. Hallet, B. A theoretical model of glacier abrasion. J. Glaciol. 23, 39–50 (1979). +49. Smith, A. M. et al. Ice stream subglacial access for ice-sheet history and fast ice flow: The BEAMISH Project on Rutford Ice Stream, West Antarctica and initial results on basal conditions. Ann. Glaciol. (2020) doi:10.1017/aog.2020.82. +50. Sugiyama, S. & Gudmundsson, G. H. Short-term variations in glacier flow controlled by subglacial water pressure at Lauteraargletscher, Bernese Alps, Switzerland. J. Glaciol. 50, 353–362 (2004). +51. Bindschadler, R. The importance of pressurized subglacial water in separation and sliding at the glacier bed. J. Glaciol. 29, 3–19 (1983). +52. Alley, R. B. et al. Bedforms of Thwaites Glacier, West Antarctica: Character and Origin. J. Geophys. Res. Earth Surf. (in Rev. (2021) doi:10.1029/2021JF006339. +53. Thøgersen, K., Gilbert, A., Schuler, T. V. & Malthe-Sørenssen, A. Rate-and-state friction explains glacier surge propagation. Nat. Commun. 10, 1–8 (2019). +54. Dieterich, J. H. Earthquake nucleation on faults with rate-and state-dependent strength. Tectonophysics 211, 115–134 (1992). +55. Cattania, C. & Segall, P. Crack Models of Repeating Earthquakes Predict Observed Moment-Recurrence Scaling. J. Geophys. Res. Solid Earth 124, 476–503 (2019). +56. Chen, T. & Lapusta, N. Scaling of small repeating earthquakes explained by interaction of seismic and aseismic slip in a rate and state fault model. J. Geophys. Res. Solid Earth 114, 1–12 (2009). +57. Abercrombie, R. E. Resolution and uncertainties in estimates of earthquake stress drop and energy release. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences vol. 379 (2021). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 132, 277, 149]]<|/det|> +## Acknowledgments: + +<|ref|>text<|/ref|><|det|>[[113, 155, 880, 417]]<|/det|> +We thank NERC British Antarctic Survey for logistics and field support, and specifically the BEAMISH field team (2018/2019). We also thank J. Hawthorne, R. Katz and S. Anandakrishnan for valuable discussions and feedback. AMB, AMS, and TM were funded by Natural Environment Research Council grants NE/G014159/1 and NE/G013187/1, Seismic instruments were provided by NERC SEIS- UK (Loans 1017 and 1111), by BAS and by the Incorporated Research Institutions for Seismology (IRIS) through the PASSCAL Instrument Center at New Mexico Tech. The facilities of the IRIS Consortium are supported by the National Science Foundation's Seismological Facilities for the Advancement of Geoscience (SAGE) Award under Cooperative Support Agreement EAR- 1851048. All the seismic data used in this analysis will be deposited on the IRIS seismological data repository. The icequake catalogue used for this analysis is available from the UK Polar Data Centre58, with details on how this catalogue was constructed given in the peer- reviewed publication28. All the fundamental code used in this study is available open source. QuakeMigrate59 was used for icequake detection, NonLinLoc for icequake relocation60 and SeisSrcMoment for the moment magnitude and other source parameter analysis61. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 82, 203, 101]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 123, 294, 140]]<|/det|> +## The icequake dataset + +<|ref|>text<|/ref|><|det|>[[115, 156, 875, 295]]<|/det|> +This study uses 100,000 icequakes at the bed of Rutford Ice Stream (RIS), Antarctica. An example of such an icequake arrival can be found in Extended Data Fig. 1. These data were collected over the period of November 2018 to February 2019. The icequakes are detected using QuakeMigrate \(^{38,59}\) and relocated using NonLinLoc \(^{60}\) . A full description of the detection, location and clustering analysis of this seismicity can be found in \(^{28}\) . The hypocentral depths, orientation of focal mechanisms, and full waveform modelling provide us with confidence that these icequakes are associated with sliding within one seismic wavelength ( \(\sim 10 m\) ) of the bed \(^{20,28,29}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 332, 527, 350]]<|/det|> +## Observable parameters from stick-slip icequakes + +<|ref|>text<|/ref|><|det|>[[115, 366, 863, 454]]<|/det|> +Earthquake source models can be used to calculate the size of the earthquake, its duration, the fault radius and the shear- stress- drop associated with the release of seismic energy. These observable parameters are required for any analysis of frictional behavior at the bed of glaciers using icequakes. The methods we use to obtain these parameters from the icequake signals are described below. + +<|ref|>sub_title<|/ref|><|det|>[[174, 470, 304, 487]]<|/det|> +## Seismic moment + +<|ref|>text<|/ref|><|det|>[[173, 504, 864, 540]]<|/det|> +The seismic moment, \(M_{0}\) , of an earthquake describes the energy released and is defined as \(^{62}\) , + +<|ref|>equation<|/ref|><|det|>[[390, 539, 658, 580]]<|/det|> +\[M_{0} = \frac{4\pi\rho\nu_{1}^{2}r\Omega_{0}}{A_{rad,i}C_{free - surface}} (1),\] + +<|ref|>text<|/ref|><|det|>[[173, 582, 876, 782]]<|/det|> +where \(\rho\) is the density of the medium at the earthquake source, \(\nu_{i}\) is the velocity of the seismic phase \(i\) (P or S), \(r\) is the hypocentre- receiver distance, \(\Omega_{0}\) is the long- period spectral amplitude, \(A_{rad,i}\) is the amplitude of radiation of seismic phase \(i\) for the particular source- receiver azimuth and take- off angle, and \(C_{free - surface}\) is the free surface correction term, which depends upon the angle of inclination of the seismic phase arrival at the surface. For this study, we assume typical ice values of \(\rho = 917 kg m^{- 3}\) , \(\nu_{p,ice} = 3841 m s^{- 1}\) , \(\nu_{s,ice} = 1970 m s^{- 1}\) . \(A_{rad,i}\) is calculated as described in \(^{20}\) , based on the assumption that all the icequakes in this study are double- couple (DC) sources with strikes aligned with the ice flow direction. This assumption is based upon previous observations at Rutford Ice Stream \(^{20,28,29}\) . \(\Omega_{0}\) is calculated by fitting a Brune source model to the noise- removed spectrum of the icequake \(^{63}\) . + +<|ref|>sub_title<|/ref|><|det|>[[175, 798, 315, 815]]<|/det|> +## Corner frequency + +<|ref|>text<|/ref|><|det|>[[174, 832, 861, 905]]<|/det|> +The spectrum of an earthquake contains more information than just the long- period spectral amplitude. If one assumes that an earthquake's spectrum can be described by a Brune model \(^{63}\) then one can also measure the corner frequency, \(f_{c}\) , of the earthquake. However, an earthquake's spectrum is also particularly sensitive to seismic attenuation. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[173, 80, 875, 152]]<|/det|> +Seismic attenuation, often described by the quality factor, \(Q\) , reduces the amplitude of an earthquake spectrum non- linearly across all frequencies. If path attenuation is poorly constrained then it can lead to detrimental uncertainty in the measured corner frequency, as evidenced by the trade- off between \(Q\) and \(f_{c}\) in the Brune model \(^{63}\) , + +<|ref|>equation<|/ref|><|det|>[[429, 152, 625, 220]]<|/det|> +\[\Omega (\mathrm{f}) = \frac{\Omega_{0} e^{-\pi f \frac{t}{Q}}}{1 + \left(\frac{f}{f_{c}}\right)^{2}} \quad (2),\] + +<|ref|>text<|/ref|><|det|>[[173, 220, 875, 255]]<|/det|> +where \(\Omega (f)\) is the amplitude of the spectrum for a certain frequency \(f\) and \(t\) is the traveltime. + +<|ref|>text<|/ref|><|det|>[[173, 268, 878, 427]]<|/det|> +To obtain an accurate measurement of corner frequency, we therefore use a linearized spectral ratios method to constrain \(Q\) . This spectral ratios method isolates the path effects from the source effects. An example of the linearized Brune model fit and the observed spectrum for an example icequake is shown in Extended Data Fig. 1c. We obtain estimates of \(Q\) from this method of the order of 200 to 800 (see Extended Data Fig. 2b), which are in agreement with other measurements for Antarctic ice \(^{64}\) . This then allows Equation 1 to be fit to the earthquake spectrum with only \(\Omega_{0}\) and \(f_{c}\) as variables. We find that the icequake corner frequencies at RIS fall approximately within the range of 40- 100 Hz (see Extended Data Fig. 2c). + +<|ref|>sub_title<|/ref|><|det|>[[174, 461, 401, 479]]<|/det|> +## Fault radius and stress-drop + +<|ref|>text<|/ref|><|det|>[[173, 495, 830, 531]]<|/det|> +One can estimate the fault radius, \(R\) , and stress- drop, \(\Delta \tau\) , of an earthquake from the corner frequency. + +<|ref|>text<|/ref|><|det|>[[174, 547, 781, 566]]<|/det|> +The relationship between corner frequency and fault radius, \(R\) , is given by \(^{65}\) , + +<|ref|>equation<|/ref|><|det|>[[457, 564, 595, 600]]<|/det|> +\[\overline{f_{c}} = \mathrm{k_{i}}\frac{\beta}{\mathrm{R}} \quad (3),\] + +<|ref|>text<|/ref|><|det|>[[173, 600, 872, 848]]<|/det|> +where \(\overline{f_{c}}\) is the spherically- averaged corner frequency for the earthquake, \(\beta\) is the shearwave speed near the source and \(k_{i}\) is a constant relating the spherically- averaged corner frequencies for a specific fault model for the seismic phase \(i\) . Here, we use the fault model of \(^{66}\) , which gives \(k_{p} = 0.38\) and \(k_{S} = 0.26\) for a rupture speed of \(0.98 \beta\) . We let \(\beta\) equal the shear velocity of ice (1970 \(m s^{- 1} 29\) ). For clast- on- bedrock slip (Regime I, Fig. 5, main text), this is valid as rupture will propagate through the bedrock and the ice that the clasts are embedded within, with us only observing the rupture propagation through the ice. For till- on- till slip (Regime II, Fig. 5, main text), our assumption of \(\beta\) is likely an overestimate, resulting in an overestimate of fault radius. As the seismic properties of the till are unknown, we are limited in assigning a lower value of \(\beta\) for any Regime II events. We assume a symmetric circular fault for this analysis. We therefore calculate average corner frequencies for each event based on the corner frequencies observed at all receivers. The potential effects of the symmetric circular fault assumption are shown in \(^{67}\) . + +<|ref|>text<|/ref|><|det|>[[173, 864, 878, 900]]<|/det|> +The uniform stress- drop of an earthquake can then be found using the fault radius and the relationship given by \(^{68}\) , + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[446, 78, 608, 115]]<|/det|> +\[\Delta \tau = \frac{7}{16}\frac{M_0}{R^3} \quad (4),\] + +<|ref|>text<|/ref|><|det|>[[173, 115, 860, 150]]<|/det|> +We now have all the observable parameters required to constrain a friction model at an icequake source. + +<|ref|>sub_title<|/ref|><|det|>[[114, 183, 822, 220]]<|/det|> +## Using a rate- and state- friction law for deriving frictional shear-stress and slip from icequakes + +<|ref|>text<|/ref|><|det|>[[174, 236, 370, 253]]<|/det|> +Calculating shear-stress + +<|ref|>text<|/ref|><|det|>[[173, 270, 864, 377]]<|/det|> +Earthquakes are typically generated as the result of stick- slip frictional instabilities at a fault interface \(^{31}\) . We hypothesize that icequakes associated with sliding at the bed of a glacier can be described by a similar model. For our investigation, we assume that the fault- interface is at or near ( \(< 1\) wavelength) the ice- bed interface. Schematic diagrams describing the model are given in Extended Data Fig. 3. Within this framework, we can apply the following rate- and state- friction law given by \(^{31}\) , + +<|ref|>equation<|/ref|><|det|>[[355, 375, 700, 412]]<|/det|> +\[\tau = \left[\mu_0 + a\ln \left(\frac{\nu}{\nu_0}\right) + b\ln \left(\frac{\nu_0\theta}{\mathcal{L}}\right)\right]\bar{\sigma} \quad (5),\] + +<|ref|>text<|/ref|><|det|>[[173, 413, 876, 521]]<|/det|> +where \(\tau\) is the total frictional shear- stress, \(\mu_0\) is the steady- state friction coefficient at \(\nu = \nu_0\) , \(\nu\) is the slip velocity, \(\nu_0\) is a reference velocity, defined in this case to be the background slip- rate, \(a\) and \(b\) are material properties, \(\mathcal{L}\) is the characteristic slip distance over which the system returns to steady- state and renew surface contacts, and \(\theta\) is the state variable. The variation of the state variable, \(\theta\) , through time can be defined by the aging or the slip laws \(^{69}\) , given by, + +<|ref|>equation<|/ref|><|det|>[[361, 519, 685, 590]]<|/det|> +\[\frac{\partial\theta}{\partial t} = 1 - \frac{\nu\theta}{\mathcal{L}} \quad (aging law) \quad (6),\] \[\frac{\partial\theta}{\partial t} = -\frac{\nu\theta}{\mathcal{L}}\ln \left(\frac{\nu\theta}{\mathcal{L}}\right) \quad (slip law) \quad (7).\] + +<|ref|>text<|/ref|><|det|>[[172, 589, 881, 890]]<|/det|> +The state variable, \(\theta\) , represents the characteristic contact lifetime of a fault. In order to apply the rate- and state- model to the stick- slip icequake system in a mathematically tractable way, we assume that the state variable of the system is constant over the duration of an icequake cycle, i.e. \(\frac{\partial\theta}{\partial t} = 0\) through all time during a cycle. For a destructive frictional failure process, \(\theta\) likely changes with time during earthquake nucleation and as the fault heals. However, for the icequake generation mechanisms proposed in this study (see Fig. 5, main text), damage at the fault interface that affects the frictional properties is likely less significant than at traditional earthquake fault interfaces. This lack of damage is evidenced to some extent by the highly repetitive nature of the icequakes \(^{20,28}\) . We assume that at least part of the icequake patch is near steady- state, or approximately at steady- state if it slips sufficiently fast. A caveat to this is that some of the icequake patch could have remained below the steady- state sliding limit, which we do not explore this here. Overall, we deem the approximation of \(\frac{\partial\theta}{\partial t} = 0\) between individual icequakes as acceptable in this case. This assumption can be used to find the state variable, \(\theta\) , as an expression of \(\nu\) , \(\mathcal{L}\) from either the aging law (Equation 6) or the slip law (Equation 7), which both yield, + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[475, 80, 580, 115]]<|/det|> +\[\theta = \frac{\mathcal{L}}{\nu} \quad (8).\] + +<|ref|>text<|/ref|><|det|>[[172, 114, 757, 134]]<|/det|> +Equation 5 can then be reduced to a rate- dependent friction law, given by, + +<|ref|>equation<|/ref|><|det|>[[384, 131, 670, 170]]<|/det|> +\[\tau = \left[\mu_{0} + (a - b)\ln \left(\frac{\nu}{\nu_{0}}\right)\right]\bar{\sigma} \quad (9).\] + +<|ref|>text<|/ref|><|det|>[[172, 168, 880, 205]]<|/det|> +The coefficient of friction in Equation 9 can then be thought of as \(\mu = \mu_{0} + \Delta \mu\) , where \(\mu_{0}\) is the static friction component, and the dynamic friction component, \(\Delta \mu\) , is given by, + +<|ref|>equation<|/ref|><|det|>[[409, 201, 645, 237]]<|/det|> +\[\Delta \mu = (a - b)\ln \left(\frac{\nu}{\nu_{0}}\right) \quad (10),\] + +<|ref|>text<|/ref|><|det|>[[172, 237, 875, 274]]<|/det|> +which when multiplied by the effective- normal- stress, \(\bar{\sigma}\) , can be assumed as equal to the earthquake stress- drop (see Equation 13). + +<|ref|>text<|/ref|><|det|>[[172, 289, 880, 348]]<|/det|> +One can then parametrize Equation 9 in such a way so that it can be solved for individual icequakes. We take \(\mu_{0} = 0.4\) , \(a = 5 \times 10^{- 3}\) and \(b = 15 \times 10^{- 3}\) from \(^{19}\) . We approximate the ratio of the instantaneous sliding velocity to the reference velocity, \(\frac{\nu}{\nu_{0}}\) , as, + +<|ref|>equation<|/ref|><|det|>[[414, 350, 639, 420]]<|/det|> +\[\frac{\nu}{\nu_{0}} = \frac{\left(\frac{d}{T}\right)}{\left(\frac{d}{\mathrm{inter - event}}\right)} \quad (11),\] + +<|ref|>text<|/ref|><|det|>[[172, 416, 875, 505]]<|/det|> +where \(d\) is the slip associated with an event (unknown), \(T\) is the slip duration, which we approximate to be equal to the inverse of the icequake corner frequency, \(f_{c}\) , \(^{70}\) and \(t_{\text{inter - event}}\) is the time between two consecutive icequakes. The correspondence of these parameters to the stick- slip cycle is shown in Extended Data Fig. 3b. With this parametrization, the velocity ratio then becomes, + +<|ref|>equation<|/ref|><|det|>[[410, 502, 641, 538]]<|/det|> +\[\frac{\nu}{\nu_{0}} = f_{c} \cdot t_{\text{inter - event}} \quad (12).\] + +<|ref|>text<|/ref|><|det|>[[172, 537, 877, 874]]<|/det|> +Assuming that the friction at the interface is velocity- weakening and therefore unstable, one can then assume that the dynamic part of Equation 9 is equal to the stress- drop measured during an icequake, \(\Delta \tau^{71}\) . One should note that this assumption implies that all the dynamic stress- release during slip is accommodated seismically (see red shaded region of Extended Data Fig. 3b). However, there is also frictional shear- stress present that cannot be measured directly using stress- drop measurements. We also assume a seismic radiation efficiency of 1, which is obviously an approximation, with the actual seismic radiation efficiency unknown. Although the radiation efficiency will in reality be \(< 1\) , due to thermal heating and the generation of additional surface area during abrasion, fracture tip energy, and other phenomena such as off- fault cracking are likely insignificant in comparison to standard earthquakes \(^{72}\) , so we deem our first- order approximation as reasonable in this case. For tectonic earthquakes, the seismic radiation efficiency typically might be of the order of 0.1 (for example, see \(^{73}\) ). If the icequake seismic radiation efficiencies were similarly low, then this would be approximately equivalent to reducing the magnitude of \(M_{0}\) by a factor of 10. Sensitivity analysis in the supplementary text suggests that such a reduction in \(M_{0}\) would reduce the shear- stress, \(\tau\) , by an order of magnitude, but the slip velocity, \(\nu_{slip}\) , would only be reduced by a factor of 3. Assuming velocity- weakening friction and a radiation efficiency of one results in the definition of the effective- normal- stress at the fault interface, given by, + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[411, 78, 644, 130]]<|/det|> +\[\bar{\sigma} = \frac{\Delta\tau}{(a - b)ln\left(\frac{v}{v_0}\right)} \quad (13).\] + +<|ref|>text<|/ref|><|det|>[[173, 131, 872, 167]]<|/det|> +Once we know the effective- normal- stress, \(\bar{\sigma}\) , we can find the overall shear- stress on the fault, \(\tau\) , from Equation 9. + +<|ref|>text<|/ref|><|det|>[[173, 182, 876, 330]]<|/det|> +We emphasize that the effective- normal- stress, \(\bar{\sigma}\) , is the normal stress on the fault, which is not necessarily equivalent to a traditionally defined glaciological effective pressure, \(P_{eff} = P_{ice} - P_{water}\) . The fault effective- normal- stress, \(\bar{\sigma}\) , is the effective- normal- stress that acts over the fault- area, \(A_{fault}\) , derived from the earthquake corner frequency (Equation 3). The actual normal stress acting through clasts in contact with the underlying contact surface might increase the normal stress acting through these clasts (see sliding regime I Fig. 5). However, fault- average normal stress, \(\bar{\sigma}\) , must be equal to the average glaciological effective pressure, \(P_{eff}\) , over the same area of the bed. + +<|ref|>sub_title<|/ref|><|det|>[[175, 364, 303, 382]]<|/det|> +## Calculating slip + +<|ref|>text<|/ref|><|det|>[[173, 398, 876, 523]]<|/det|> +The second glaciologically important parameter to measure at the bed is the slip, and hence the basal slip- rate. To calculate slip, we assume that while an individual icequake cluster is active, all (or at least the vast majority of) slip is accommodated seismically. This is likely the case for RIS, as evidenced by the close agreement between surface slip- rate and seismically measured basal slip- rates (see Fig. 2f). Calculating the basal slip, \(d\) , from an icequake is challenging because one first has to determine a method of estimating the bed shear- modulus, \(G_{bed}\) , since the slip is given by, + +<|ref|>equation<|/ref|><|det|>[[446, 521, 608, 559]]<|/det|> +\[d = \frac{M_0}{G_{bed}\cdot A} \quad (14),\] + +<|ref|>text<|/ref|><|det|>[[173, 560, 872, 578]]<|/det|> +where \(M_0\) is the seismic moment released by an earthquake and \(A\) is the area of the fault. + +<|ref|>text<|/ref|><|det|>[[173, 593, 878, 768]]<|/det|> +The bed shear- modulus, \(G_{bed}\) , is calculated by assuming a further behavior of the rate- and state- friction law. This behavior is that an earthquake can only nucleate if it is in the unstable regime. In this study, we assume that the temporally- averaged driving shear- stress at the fault varies over longer time- scales than the icequake inter- event time, with the shear- stress at which the fault fails governed by the effective- normal- stress acting on the fault, \(\bar{\sigma}\) . The approximately constant inter- event time between individual consecutive icequake pairs (see Fig. 2e) within a single cluster validates this assumption. \(^{31}\) define the effective- normal- stress at which a fault becomes unstable is defined as the critical normal stress, \(\bar{\sigma}_c\) , with velocity- weakening behavior prevailing above this stress. \(\bar{\sigma}_c\) is given by \(^{31}\) , + +<|ref|>equation<|/ref|><|det|>[[448, 767, 606, 803]]<|/det|> +\[\bar{\sigma}_c = \frac{\mathrm{k}\mathcal{L}}{b - a} \quad (15),\] + +<|ref|>text<|/ref|><|det|>[[173, 804, 870, 839]]<|/det|> +where \(k\) is the spring constant of the system (see Extended Data Fig. 3a), which is given by, + +<|ref|>equation<|/ref|><|det|>[[464, 838, 590, 872]]<|/det|> +\[k = \frac{G^{*}}{R} \quad (16),\] + +<|ref|>text<|/ref|><|det|>[[173, 873, 879, 909]]<|/det|> +where \(G^{*}\) is the effective shear- modulus of the bimaterial interface \(^{19}\) and \(R\) here is the radius of the fault, which can be found from the icequake corner frequency, if assuming a + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[173, 80, 872, 186]]<|/det|> +symmetric, circular fault \(^{66,67}\) . However, this equation still has two unknowns: \(G^{*}\) , the effective shear-modulus that we require to calculate the slip; and \(\mathcal{L}\) , the critical slip distance, otherwise referred to as the state evolution distance. For the purposes of this study, we approximate \(\mathcal{L}\) to remain constant, but allow \(G^{*}\) to vary with effective-normal-stress, which from granular material theory \(^{74}\) is assumed to take the generic empirical form, + +<|ref|>equation<|/ref|><|det|>[[436, 186, 617, 205]]<|/det|> +\[G^{*} = A\bar{\sigma}^{n} + C\quad (17),\] + +<|ref|>text<|/ref|><|det|>[[173, 204, 880, 239]]<|/det|> +where \(A\) , \(n\) and \(C\) are constants to invert for. We use a least squares approach to minimize the function, + +<|ref|>equation<|/ref|><|det|>[[266, 237, 787, 280]]<|/det|> +\[f(\bar{\sigma}_{c},R,a,b,A,n,C,L) = l n\left(\frac{(A\bar{\sigma}_{c}^{n} + C)\mathcal{L}}{b - a}\right) - \ln (R\bar{\sigma}_{c}) \quad (18),\] + +<|ref|>text<|/ref|><|det|>[[172, 280, 879, 457]]<|/det|> +where \(\bar{\sigma}_{c}\) and \(R\) vary for each icequake, and \(A\) , \(n\) , \(C\) and \(\mathcal{L}\) are varied to minimize the function. \(\bar{\sigma}_{c}\) is taken to be the effective- normal- stress for the first 100 icequakes when a cluster becomes active, as calculated using Equation 13. These parameters are found to be \(A = 22,000\) , \(n = 0.78\) , \(C = 8,200\) \(Pa\) and \(\mathcal{L} = 7.7 \times 10^{- 5} m\) , with the result of the minimization shown in Extended Data Fig. 4. Now \(\mathcal{L}\) can be substituted into Equation 15 to find the bimaterial shear- modulus, \(G^{*}\) . The shear- modulus of the bed, \(G_{bed}\) can then be found using the Poisson ratios of ice (1/3) and till (0.49), which gives \(G^{*} \approx 3.5 G_{bed}^{19}\) . Granular material theory, or at least the relationship of Equation 17, is thought to still hold for clast- over- bedrock sliding since the shear- modulus will still be related to some exponent, \(n\) , of \(\bar{\sigma}\) , even if that exponent were \(\sim 0\) . + +<|ref|>text<|/ref|><|det|>[[172, 475, 880, 565]]<|/det|> +Equation 14 can then be used to find the slip, \(d\) , associated with a single icequake, for the effective- normal- stress applied to the fault at that particular time. We also calculate the approximate slip- rate associated with these highly repetitive icequakes. If one assumes that all the slip when an icequake cluster is active is accommodated seismically, then one can calculate the slip- rate per day, \(v_{slip}\) , + +<|ref|>equation<|/ref|><|det|>[[416, 562, 639, 604]]<|/det|> +\[v_{slip} = \frac{d}{t_{inter - event}} \quad (19),\] + +<|ref|>text<|/ref|><|det|>[[172, 620, 879, 673]]<|/det|> +The methods described above allow us to calculate the total shear- stress, \(\tau\) , and the slip, \(d\) , at the bed. These two parameters can provide observational constraint on ice dynamics models of ice streams. + +<|ref|>sub_title<|/ref|><|det|>[[115, 698, 301, 714]]<|/det|> +## A note on assumptions + +<|ref|>text<|/ref|><|det|>[[112, 722, 880, 916]]<|/det|> +A number of assumptions are made to make the derivation of basal shear- stress and slip from icequake observations and a rate- and- state friction model mathematically tractable. There are several assumptions that warrant particular emphasis. The first is the assumption that all slip at an individual sticky- spot is accommodated seismically while that cluster is active. The highly repetitive nature of the icequakes (see Extended Data Fig. 1 and \(^{28}\) ), with approximately constant inter- event times between consecutive icequakes in a cluster, is indicative of the stability of each sticky- spot (see Fig. 2), justifying this assumption. Secondly, a Brune model \(^{63}\) is assumed to describe the earthquake source characteristics. While such a model is likely an approximation for the complex physics of earthquake rupture, it is a common assumption for other earthquake studies that is likely also a valid approximation for the stick- slip icequakes presented here. Thirdly, we approximate that the time- derivative of the state- variable in the rate- and- state + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 81, 877, 475]]<|/det|> +friction model, \(\frac{\partial \theta}{\partial t}\) , equals zero during an individual icequake cycle. This approximation is valid if slip on the fault is sufficiently fast and if little damage occurs at the fault, compared to more complex earthquake faults. Obviously, this is only an approximation, as damage does likely occur at the fault, at least for the clasts- over- bedrock slip case (regime I, see Fig. 5). Furthermore, an underestimation bias in slip may be introduced by the assumption of no fault frictional heating. Fault frictional heating would reduce the seismic radiation efficiency from our approximation of one \(^{72}\) . The final assumption we emphasize here is that we assume that the icequakes at the beginning of an icequake cluster nucleate at approximately the critical normal stress for nucleation, \(\sigma_{c}\) , rather than at some arbitrary value above it. The icequake slip calculations are dependent upon this assumption. This assumption would not be valid for sporadic earthquakes on complex faults, as shear- stresses could build to different values before failure for each earthquake, even with constant effective- normal- stresses, due to fault heterogeneity. Nor would it necessarily be valid if the driving shear- stress were perturbed over time- scales shorter than the inter- event time, for example by interactions with other icequake clusters. However, although icequake faults still exhibit a degree of heterogeneity due to an inhomogeneous distribution of clasts, this heterogeneity has negligible impact upon the consistency of both the inter- event times and shear- stresses between consecutive icequakes at a given sticky- spot (see Fig. 2). Furthermore, there are only a small number of active icequake clusters at any given time, which are spatially isolated from one another. The consistency in inter- event times and shear- stresses observed in our data, in agreement with similar, laboratory- generated icequakes \(^{12}\) , provides us with confidence in our assumption of icequakes nucleating at the critical nucleation stress, \(\sigma_{c}\) . + +<|ref|>sub_title<|/ref|><|det|>[[115, 500, 288, 515]]<|/det|> +## Additional references: + +<|ref|>text<|/ref|><|det|>[[110, 515, 880, 916]]<|/det|> +58. Kufner, S. et al. Microseismic icequake catalogue, Rutford ice stream (west Antarctica), November 2018 to February 2019 (version 1.0). UK Polar Data Centre, Nat. Environ. Res. Counc. UK Res. Innov. (2021) doi:10.5285/B809A040-8305-4BC5-BAFF-76AA2B823734. +59. Winder, T. et al. QuakeMigrate v1.0.0. Zenodo (2021) doi:10.5281/zenodo.4442749. +60. Lomax, A. & Virieux, J. Probabilistic earthquake location in 3D and layered models. Adv. Seism. Event Locat. Vol. 18 Ser. Mod. Approaches Geophys. 101–134 (2000). +61. Hudson, T. S. TomSHudson/SeisSrcMoment: First formal release (Version 1.0.0). Zenodo (2020) doi:http://doi.org/10.5281/zenodo.4010325. +62. Aki, K. & Richards, P. G. Quantitative Seismology. (University Science Books, 2002). +63. Brune, J. N. Tectonic Stress and the Spectra of Seismic Shear Waves from Earthquakes. J. Geophys. Res. 75, 4997–5009 (1970). +64. Peters, L. E., Anandakrishnan, S., Alley, R. B. & Voigt, D. E. Seismic attenuation in glacial ice: A proxy for englacial temperature. J. Geophys. Res. Earth Surf. 117, 1–10 (2012). +65. Madariaga, R. Dynamics of an expanding circular fault. Bull. Seismol. Soc. Am. 66, 639–666 (1976). +66. Kaneko, Y. & Shearer, P. M. Seismic source spectra and estimated stress drop derived from cohesive-zone models of circular subsurface rupture. Geophys. J. Int. 197, 1002–1015 (2014). +67. Kaneko, Y. & Shearer, P. M. Variability of seismic source spectra, estimated stress drop, and radiated energy, derived from cohesive-zone models of symmetrical and asymmetrical circular and elliptical ruptures. J. Geophys. Res. Solid Earth 120, 1053–1079 (2015). +68. Eshelby, J. D. The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc. R. Soc. London. Ser. A. Math. Phys. Sci. 241, 376–396 (1957). +69. Ruina, A. Slip instability and state variable friction laws. J. Geophys. Res. 88, 10359–10370 (1983). +70. Hanks, T. C. & McGuire, R. K. The character of high-frequency strong ground motion. Bull. + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[0, 78, 88, 92]]<|/det|> +767 + +<|ref|>text<|/ref|><|det|>[[0, 95, 88, 110]]<|/det|> +768 + +<|ref|>text<|/ref|><|det|>[[0, 113, 88, 127]]<|/det|> +769 + +<|ref|>text<|/ref|><|det|>[[0, 130, 88, 144]]<|/det|> +770 + +<|ref|>text<|/ref|><|det|>[[0, 147, 88, 161]]<|/det|> +771 + +<|ref|>text<|/ref|><|det|>[[0, 164, 88, 178]]<|/det|> +772 + +<|ref|>text<|/ref|><|det|>[[0, 181, 88, 195]]<|/det|> +773 + +<|ref|>text<|/ref|><|det|>[[0, 198, 88, 212]]<|/det|> +774 + +<|ref|>text<|/ref|><|det|>[[0, 215, 88, 229]]<|/det|> +775 + +<|ref|>text<|/ref|><|det|>[[0, 232, 88, 246]]<|/det|> +776 + +<|ref|>text<|/ref|><|det|>[[0, 249, 88, 263]]<|/det|> +777 + +<|ref|>text<|/ref|><|det|>[[0, 266, 88, 280]]<|/det|> +778 + +<|ref|>text<|/ref|><|det|>[[0, 284, 88, 298]]<|/det|> +779 + +<|ref|>text<|/ref|><|det|>[[0, 302, 88, 316]]<|/det|> +780 + +<|ref|>text<|/ref|><|det|>[[0, 320, 88, 334]]<|/det|> +781 + +<|ref|>text<|/ref|><|det|>[[0, 338, 88, 352]]<|/det|> +782 + +<|ref|>text<|/ref|><|det|>[[0, 356, 88, 370]]<|/det|> +783 + +<|ref|>text<|/ref|><|det|>[[0, 374, 88, 388]]<|/det|> +784 + +<|ref|>text<|/ref|><|det|>[[0, 392, 88, 406]]<|/det|> +785 + +<|ref|>text<|/ref|><|det|>[[0, 410, 88, 424]]<|/det|> +786 + +<|ref|>text<|/ref|><|det|>[[0, 428, 88, 442]]<|/det|> +787 + +<|ref|>text<|/ref|><|det|>[[0, 446, 88, 460]]<|/det|> +788 + +<|ref|>text<|/ref|><|det|>[[0, 464, 88, 478]]<|/det|> +789 + +<|ref|>text<|/ref|><|det|>[[0, 482, 88, 496]]<|/det|> +790 + +<|ref|>text<|/ref|><|det|>[[0, 500, 88, 514]]<|/det|> +791 + +<|ref|>text<|/ref|><|det|>[[0, 518, 88, 532]]<|/det|> +792 + +<|ref|>text<|/ref|><|det|>[[0, 536, 88, 550]]<|/det|> +793 + +<|ref|>text<|/ref|><|det|>[[0, 554, 88, 568]]<|/det|> +794 + +<|ref|>text<|/ref|><|det|>[[0, 572, 88, 586]]<|/det|> +795 + +<|ref|>text<|/ref|><|det|>[[0, 590, 88, 604]]<|/det|> +796 + +<|ref|>text<|/ref|><|det|>[[0, 608, 88, 622]]<|/det|> +797 + +<|ref|>text<|/ref|><|det|>[[0, 626, 88, 640]]<|/det|> +798 + +<|ref|>text<|/ref|><|det|>[[0, 644, 88, 658]]<|/det|> +799 + +<|ref|>text<|/ref|><|det|>[[0, 662, 88, 676]]<|/det|> +795 + +<|ref|>text<|/ref|><|det|>[[0, 680, 88, 694]]<|/det|> +796 + +<|ref|>text<|/ref|><|det|>[[0, 698, 88, 712]]<|/det|> +797 + +<|ref|>text<|/ref|><|det|>[[0, 716, 88, 730]]<|/det|> +798 + +<|ref|>text<|/ref|><|det|>[[0, 734, 88, 748]]<|/det|> +799 + +<|ref|>text<|/ref|><|det|>[[0, 752, 88, 766]]<|/det|> +800 + +<|ref|>text<|/ref|><|det|>[[0, 770, 88, 784]]<|/det|> +801 + +<|ref|>text<|/ref|><|det|>[[0, 788, 88, 802]]<|/det|> +802 + +<|ref|>text<|/ref|><|det|>[[0, 806, 88, 820]]<|/det|> +803 + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 83, 343, 103]]<|/det|> +## Extended Data Figures + +<|ref|>image<|/ref|><|det|>[[163, 132, 825, 540]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[113, 564, 880, 704]]<|/det|> +Extended Data Fig. 1. Examples of icequake waveforms and spectra. (a) 30 minutes of continuous data for the Z component of station R3030. Approximate icequake P- phase arrival times associated with a single cluster are shown by the green lines. (b), (c), (d) Stacked waveform data on the Z-, N- and E- components for 173 events in a cluster at station R3030, located at the center of the network. Red line indicates P- phase arrival. Blue lines indicate S- phase arrivals. Grey shading represents the standard deviation of the stacked data. (e) Spectrum for one event within the cluster at station R3030. Waveform data in (a) to (d) are filtered between 10 Hz and 120 Hz. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[203, 85, 789, 272]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 292, 852, 348]]<|/det|> +
Extended Data Fig. 2. Quality factor (Q) and corner frequency \((f_{c})\) distributions for the icequakes in this experiment. (a) Histogram of Q. (b) Histogram of \(f_{c}\) . Values for each icequake are averaged for all individual station observations.
+ +<|ref|>image<|/ref|><|det|>[[384, 375, 608, 770]]<|/det|> + +<|ref|>text<|/ref|><|det|>[[113, 787, 868, 858]]<|/det|> +Extended Data Fig. 3. Schematic Fig. describing the rate- and state- frictional model as a block- slider model. (a) Diagram of the block- slider model, showing the driving shear- stress, \(\tau\) , the effective- normal- stress, \(\bar{\sigma}\) , and the system spring constant \(k\) . (b) Accumulated shear- stress vs. time for a series of consecutive icequakes. (c) Shear- stress at the fault at a particular time as + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 80, 839, 117]]<|/det|> +predicted by the rate-and-state model \(^{31}\) . (d) The stick-slip icequake cycle, with the numbers corresponding to the relevant stress states labelled in (b). + +<|ref|>image<|/ref|><|det|>[[244, 180, 723, 456]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 476, 844, 514]]<|/det|> +
Extended Data Fig. 4. Results of the least squares inversion of Equation 17. Blue scatter points are the data and red scatter points show the least-squares inversion result.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[272, 82, 728, 417]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 438, 876, 550]]<|/det|> +
Extended Data Fig. 5. Rate-and-state friction model sensitivity analysis. Plot of the sensitivity in frictional shear-stress at the bed, \(\tau_{bed}\) , and slip-rate at the bed, \(\nu_{slip}\) , with perturbation of the key observational parameters. The reference values used to normalize the variations are the average values of \(\tau_{bed}\) and \(\nu_{slip}\) observed at all the clusters. The magnitude of variation in each parameter are summarized in Table S1. See supplementary text for further details.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[58, 130, 723, 150]]<|/det|> +- frictionandslipicequakespaperHudson2021supplementaryinformation.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/images_list.json b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..8b18ea91485ff92fdcb3fe97bb667eaa3c552107 --- /dev/null +++ b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/images_list.json @@ -0,0 +1,77 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1. General architecture of the Bayesian neural network. a Schematic of the Bayesian neural network used for heart disease (arrhythmia) classification. In Bayesian neural networks, the weights are represented by probability distributions, thus naturally including uncertainty in the model. b Example of output neuron activation distributions, obtained for certain output, uncertain output, due to noisy input data, and unknown data (i.e. out-of-distribution data). c Experimental setup. d Hardware implementation of a Bayesian neural network by combining multiple versions of ANNs.", + "footnote": [], + "bbox": [ + [ + 117, + 245, + 883, + 654 + ] + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2. Fabricated filamentary memristor and phase-change memory based array die. a Scanning electron microscopy image of a phase-change memory in the back end of line of our hybrid memristor/CMOS process. b Scanning electron microscopy image of a phase-change memory. c Optical microscopy photograph of the phase-change memory based 1T1R array. d Scanning electron microscopy image of a filamentary memristor in the back end of line of our hybrid memristor/CMOS process. e Scanning electron microscopy image of a filamentary memristor. f Optical microscopy photograph of the filamentary memristor based 1T1R array.", + "footnote": [], + "bbox": [ + [ + 109, + 76, + 864, + 247 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3. Filamentary memristor and phase-change memories as physical random variables with normal distribution. a Probability densities of 2,048 filamentary memristors programmed with eight different programming current values. b Domain of the Gaussian distributions experimentally achieved exploiting different programming conditions for filamentary memristors (blue) and phase-change memories (green). Triangles represent one-shot programming, dots represent iterative programming and the cross represents the low conductance state. c Probability densities of 2,048 phase-change memories programmed with seven different programming current values. d Schematic of the proposed synaptic circuit. Each sample of a Bayesian probabilistic weight is stored as the difference between the conductance values of two adjacent memory cells. e Domain of the normal distributions (Γ) that can be experimentally obtained exploiting the circuit in d by storing samples on two memory cells.", + "footnote": [], + "bbox": [ + [ + 108, + 258, + 890, + 587 + ] + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4. Domains of normal distribution obtained with classical Variational Inference and the proposed technologically plausible method. a Domain of normal distributions \\(\\theta = (\\mu , \\sigma)\\) obtained after training with the classical VI method and mapping the software values to the conductance range achievable with filamentary memristors (blue) and phase-change memories (green). b Domain of normal distributions \\(\\theta = (\\mu , \\sigma)\\) obtained after training with the proposed method calibrated on filamentary memristors (blue) and phase-change memory experimental data (green).", + "footnote": [], + "bbox": [ + [ + 113, + 264, + 884, + 640 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_5.jpg", + "caption": "Figure 5. Measurements of the fabricated memristor-based Bayesian neural network. a tSNE visualization of input data, different colors representing different classes (diseases). Nearby points correspond to similar data and distant points to dissimilar data. b tSNE visualization of experimental data classification. The different colors represent points correctly or incorrectly predicted and unseen data. c Experimental probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases. d Experimental probability density distribution of the epistemic uncertainty for correct predictions, incorrect predictions and unseen diseases. e Simulated probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases for a conventional neural network with the same architecture and using float32 encoding for the synapses. f ROC curve corresponding to the differentiation between correct prediction and incorrect prediction, based on aleatoric uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). g ROC curve corresponding to the differentiation between known and unknown data, based on epistemic uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). h Measured accuracy, epistemic uncertainty, and aleatoric uncertainty (calculated as the area of the ROC curves) as a function of the number of filamentary memristor devices per synapse.", + "footnote": [], + "bbox": [ + [ + 110, + 100, + 888, + 521 + ] + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f.mmd b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f.mmd new file mode 100644 index 0000000000000000000000000000000000000000..b2e6865453b3577c8e82456d13d88701ea94a00d --- /dev/null +++ b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f.mmd @@ -0,0 +1,412 @@ + +# Bringing uncertainty quantification to the extreme-edge with memristor-based Bayesian neural networks + +Djohan Bonnet Université Grenoble Alpes, CEA, LETI + +Tifenn Hirtzlin CNRS + +Atreya Majumdar Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies + +Thomas Dalgaty CEA- Leti, Université Grenoble Alpes https://orcid.org/0000- 0003- 0326- 2121 + +Eduardo Esmanhotto Université Grenoble Alpes, CEA, LETI + +Valentina Meli Université Grenoble Alpes, CEA, LETI + +Niccolò Castellani CEA, LETI, Minatec Campus, Grenoble + +Simon Martin Université Grenoble Alpes, CEA, LETI + +Jean- Francois Nodin CEA- LETI + +Guillaume Bourgeois CEA, LETI, Minatec Campus, Grenoble + +Jean- Michel Portal + +Aix- Marseille Université, CNRS, Institut Matériaux Microélectronique Nanosciences de Provence + +Damien Querlioz + +Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies, 91120 Palaiseau, France + +Elisa Vianello ( elisa.vianello@cea.fr) + +Université Grenoble Alpes, CEA, LETI https://orcid.org/0000- 0002- 8868- 9951 + +Article + +Keywords: + +<--- Page Split ---> + +Posted Date: January 13th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 2458251/v1 + +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on November 20th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43317- 9. + +<--- Page Split ---> + +# Bringing uncertainty quantification to the extreme-edge with memristor-based Bayesian neural networks + +Djohan Bonnet \(^{1,2*}\) , Tifenn Hirtzlin \(^{1}\) , Atreya Majumdar \(^{2}\) , Thomas Dalgaty \(^{3}\) , Eduardo Esmanhotto \(^{1}\) , Valentina Meli \(^{1}\) , Niccolo Castellani \(^{1}\) , Simon Martin \(^{1}\) , Jean- François Nodin \(^{1}\) , Guillaume Bourgeois \(^{1}\) , Jean- Michel Portal \(^{4}\) , Damien Querlioz \(^{2*}\) , and Elisa Vianello \(^{1*}\) + +\(^{1}\) Université Grenoble Alpes, CEA, LETI, Grenoble, France \(^{2}\) Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies, Palaiseau, France \(^{3}\) Université Grenoble Alpes, CEA, LIST, Grenoble, France \(^{4}\) Aix- Marseille Université, CNRS, Institut Matériaux Microélectronique Nanosciences de Provence, Marseille, France \(^{*}\) djohan.bonnet@cea.fr, damien.querlioz@c2n.upsaclay.fr, elisa.vianello@cea.fr + +## ABSTRACT + +Safety- critical sensory processing applications, like medical diagnosis, require making accurate decisions based on a small amount of noisy input data. For these applications, using Bayesian neural networks, able to quantify the uncertainty of the predictions, is a superior approach to using conventional artificial neural networks. However, because of the probabilistic nature of Bayesian neural networks, they can be computationally intensive to use for inference stage and thus not well suited for extreme- edge applications. An emerging idea to solve this problem is to use the intrinsic probabilistic nature of memristors to efficiently implement Bayesian neural network inference: the variability in the resistance of memristors would represent the probability distribution of weights in Bayesian neural networks. However, when using memristors, statistical effects follow the laws of device physics, whereas in Bayesian neural networks, those effects can take on arbitrary shapes. In this work, we overcome this difficulty by adopting a dedicated synapse architecture based on two memristors, and by training Bayesian neural networks with a dedicated variational inference technique that includes a "technological loss" to take into account specificities of memristor physics. This technique allowed us to program a two- layer Bayesian neural network on 75 physical crossbar arrays of 1,024 memristors, incorporating CMOS periphery circuitry to do in- memory computing, to classify arrhythmia in electrocardiograms. Our experimental neural network classified heartbeats with high accuracy, and estimated the certainty of its predictions. In the case of uncertain predictions, it differentiated between ambivalent heartbeats (aleatoric uncertainty), and heartbeats with never- seen patterns (epistemic uncertainty). We show that our technique can also be used with phase change memories, by employing a different "technological loss" term. The great advantage of this approach is its low energy consumption: we estimate an 800 times improvement in energy efficiency compared to a GPU performing the same task. + +## 1 Introduction + +Hardware neural networks based on emerging non- volatile memories can bring intelligence to the edge at a very low energetic cost. In this context, filamentary memristors and phase change memories can be used in a very elegant and energy- efficient way. These devices can act as analog synaptic weights enabling neural network multiply- and- accumulate operations directly in memory by relying on Ohm's law and Kirchoff's current law \(^{1 - 7}\) . Low- power systems of this kind could provide essential services: for example, medical devices could analyze patient measurements and detect life- threatening emergencies or automatically adjust treatment. However, some features of conventional neural networks are not well suited to such applications. In particular, conventional neural networks are notoriously bad at evaluating uncertainty \(^{8 - 10}\) . When trained with small datasets, as is typically the case in medical applications, conventional neural networks tend to "overfit" training data and to provide highly certain answers in all situations \(^{11,12}\) . More importantly, uncertainty in a machine learning context can have different origins, usually referred to as aleatoric and epistemic, which conventional neural networks cannot tell apart \(^{10,13}\) . To illustrate the impact of this limitation, we can use the example of a medical device trained to recognize different types of arrhythmia (irregular heartbeats) in a patient. Aleatoric uncertainty characterizes situations where a measurement could be consistent with different types of arrhythmia. Epistemic uncertainty, on the other hand, arises when an arrhythmia type that is significantly different from the data used to train the device appears and must be classified. Recognizing and differentiating those two types of uncertainty is essential, as they could be indicating that the patient's condition has evolved \(^{13,14}\) . + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1. General architecture of the Bayesian neural network. a Schematic of the Bayesian neural network used for heart disease (arrhythmia) classification. In Bayesian neural networks, the weights are represented by probability distributions, thus naturally including uncertainty in the model. b Example of output neuron activation distributions, obtained for certain output, uncertain output, due to noisy input data, and unknown data (i.e. out-of-distribution data). c Experimental setup. d Hardware implementation of a Bayesian neural network by combining multiple versions of ANNs.
+ +<--- Page Split ---> + +Bayesian neural networks are an alternative class of neural networks where synaptic weights and neuron activations do not take on unique values but are instead represented by probability distributions \(^{15 - 17}\) . Such networks are trained to provide models that provide plausible explanations for the training data, instead of just fitting this data; for this reason, Bayesian neural networks mitigate overfitting \(^{16}\) . And because they model certainty well, they excel at identifying aleatoric and epistemic uncertainty \(^{10}\) . Since computations based on Bayesian neural networks perform operations with probability distributions instead of numerical values, they are hard to materialize with memristors or phase change memories. + +In our work, we use the variability and imperfections of memory devices (which can be represented by probability distributions) as the actual probability distributions of the neural network. This approach is challenging because the statistical properties of memristors and phase change memories are governed by strict rules of device physics \(^{18 - 20}\) . As a consequence, when filamentary oxide- based memristors are at higher resistance states, they will tend to have broader probability distributions, and when they are in lower resistance states they will tend to have narrow resistance states \(^{21,22}\) . In other words, the mean value and the standard deviation of the resistance are correlated. By contrast, the probability distribution of parameters in a Bayesian neural network can take any shape \(^{15,23}\) . Two recent publications have proposed new devices where the inherent probability distribution of their resistance can be tuned. These solutions, which exploit two- dimensional materials \(^{24}\) and magnetic devices \(^{25}\) , were validated with simulations of Bayesian neural networks. + +In this work, we chose to focus on standard nanodevices (filamentary memristors) to demonstrate the first complete nanodevice- based Bayesian neural network implementation of a real- world task. Our demonstrator is able to classify arrhythmia recordings by types, with accurate aleatoric and epistemic uncertainty, using 75 arrays of fabricated \(32 \times 32\) memristor chips that associate hafnium oxide memristors and CMOS periphery circuitry. This array circuit performs in- memory computations based on Kirchoff's laws. Our work overcomes the problems related to the correlation of statistical properties of memristors, using three insights: + +- The Bayesian neural network is trained using variational inference \(^{16}\) , a method that ensures statistical independence between synapses. A prior simulation study \(^{21}\) suggested Markov Chain Monte Carlo training, leading to statistically-dependent synapses, which is extremely challenging to realize experimentally due to memristor imperfection (see Supplementary Note 2). + +- Each synapse is implemented using two memristors that are programmed independently, allowing the partial decorrelation of mean values and standard deviations of synaptic weights. This idea was presented in simulation studies \(^{24,25}\) . + +- During the off-chip training process, we introduce a "technological loss" term to match the Bayesian neural networks to the statistics that can be achieved physically. This idea is proposed here for the first time, and it has been critical in making our experimental system work. + +Our approach is generic. Using a different expression for the technological loss, we show that the same technique can be applied to phase change memory (by following a hybrid experimental-simulation methodology, employing a fabricated array of 16,384 phase change memories). + +Other works have already pointed out connections between Bayesian concepts and nanodevices. Ref. \(^{18}\) exploited the probabilistic nature of memristors to perform Bayesian learning. This approach can only be applied to small- scale tasks, but unlike Bayesian neural networks, it does not suffer from the limitations imposed by the correlation of mean value and standard deviation of memristors \(^{18}\) . The works of \(^{26 - 30}\) use nanodevices to carry out Bayesian network inference. Bayesian networks differ from Bayesian neural networks. Bayesian networks are constructed using expert knowledge and are fully explainable which makes them ideal for tasks like sensor fusion. Conversely, Bayesian neural networks are trained from the ground up and excel on more data- intensive tasks like electrocardiogram or electroencephalogram classification. Finally, the work of Ref. \(^{31}\) treats memristors as Bayesian variables and uses them to program deterministic neural networks in order to increase hardware resilience. + +In this paper, we introduce and describe the general architecture of our Bayesian neural networks, which are based on memory devices, and our technique to match imperfections in nanodevices with the probability distributions of a Bayesian neural network. We then show the experimental classification of arrhythmia with a proper uncertainty evaluation, using 75 arrays of filamentary memristors. + +## Results + +## Memory devices-based Bayesian neural networks + +For our experiments, we considered a two- layer Bayesian neural network (Fig. 1a,b), trained to differentiate nine classes of heart arrhythmia from electrocardiogram (ECG) recordings \(^{32}\) . While in Artificial Neural Networks (ANNs), the synaptic weights are point estimates, Bayesian neural networks replace them with probability distributions. A natural way to implement a Bayesian + +<--- Page Split ---> + +neural network with memristors or phase change memories is to use a collection of \(M\) distinct memory arrays to represent each layer of the neural network (Fig. 1c,d). We sample \(M\) weight values for each synapse based on its probability distribution in the Bayesian neural network, and we program them to the \(M\) memory array. We will see in the next sections that the inherent probabilistic effects in memory devices allow us to perform the sampling and programming operations simultaneously, transforming the largest drawback of emerging memory devices, their variability, into a feature. This approach leads to a collection of \(M\) independent in- memory neural networks. By presenting the same input to each of these arrays, we obtain a collection of \(M\) different outputs representing the output distribution of these neural networks. + +The benefit of using distributions, instead of deterministic values, is that we can quantify the uncertainty of the neural network's output. Intuitively, each of the \(M\) independent neural networks constitutes a reasonable hypothesis explaining the data used to train the Bayesian neural network. The spread of the output distributions captures the certainty or lack of certainty in the model's predictions. The uncertainty depends on both the mean and the variance values of the output neuron activations, as illustrated in Figs. 1b- d. If an output is highly certain ("clear classification"), all \(M\) neural networks will have the same active output neuron (value close to one), and inactive output neurons (value close to zero), and both the aleatoric and epistemic uncertainties will be low (see Methods for the mathematical definitions of aleatoric and epistemic uncertainties). If measurement imprecision causes ambivalence in the predictions ("unclear classification"), e.g., the features allowing differentiation of the arrhythmia types are lost in noise, none of the \(M\) neural networks will provide a certain prediction: all of them will hesitate between the same classes, with non- zero outputs on these classes. The variance of the output remains low, as all \(M\) networks have a consistent behavior: aleatoric uncertainty increases, whereas epistemic uncertainty does not. But what happens if a new arrhythmia type that was not present during training appears, and the network has to classify it? This is an example of out- of- distribution test data (case "Unknown data" in Figs. 1b- d). All \(M\) neural networks sampled from the Bayesian neural network will tend to make a different interpretation of this unknown data resulting in a high variance in the distribution of output neurons. Both aleatoric and epistemic uncertainty will be high. Classifying out- of- distribution data, therefore, is possible by measuring output distributions. This is of fundamental importance for safety- critical applications like medical diagnoses or autonomous driving. + +Performing inference with our approach requires massive parallel Multiply- and- Accumulate (MAC) operations. These operations are power- hungry when carried out on CMOS- based ASICs and field- programmable gate arrays, due to the shuttling of data between processor and memory. In this work we use crossbars of memristors that naturally implement the multiplication between the input voltage and the probabilistic synaptic weight through Ohm's law, and the accumulation through Kirchhoff's current law \(^{2,7,33,34}\) , to significantly lower power consumption. + +## Filamentary memristor and phase-change memory as normal distributions + +Among the non- volatile memories that can be integrated in advanced commercial processes, phase- change memories (PCM) and filamentary memristors have been widely studied for analog in- memory neural network implementation, because of the possibility of adjusting the conductance level of these devices. In our previous work, we demonstrated that the intrinsic variability in filamentary memristors can be leveraged to store the probabilistic weights of Bayesian neural networks \(^{21}\) . However, the conductance distribution follows strict rules due to device physics: the mean value, \(\mu\) , and the standard deviation, \(\sigma\) , are strongly correlated. Phase- change memories suffer from the same limitation \(^{19}\) . Bayesian neural networks require a larger space of normal distribution with mean values that are uncorrelated with the standard deviations. Here, we come up with a new synaptic circuit and the associated programming strategy to obtain largely unrestricted \(\mu\) and \(\sigma\) values. To illustrate the practicality of the proposed solution, we fabricated and tested arrays of hafnium- oxide- based filamentary memristors and of germanium- antimony- tellurium phase- change memories in a one- transistor- one- resistor (1T1R) configuration (Fig. 2). Both memory technologies have been integrated into the back end of line (BEOL) of a 130- nanometer foundry CMOS process with four metal layers (see Methods). Figs. 3a and c show the distributions of 2,048 filamentary- based memristors and phase- change memories, respectively, programmed in eight conductance levels. + +In both cases, the standard deviation of the distribution is related to its mean value and cannot be chosen independently. The resulting domain of normal distributions that can be achieved by exploiting device variability \((\sigma)\) is thus bounded to a one- dimensional space for both technologies (Fig. 3b). In filamentary memristors, \(\sigma\) decreases for increasing conductance values due to the Poisson- like spread of the number of defects injected during the programming operation \(^{35}\) . In phase- change memories, the trend is inverted: device variability increases with the conductance values, moving from a full to a partial amorphous material \(^{19,20}\) . For both technologies, the standard deviation can be reduced by adopting an iterative programming- and- verify scheme (see Methods). To extend the domain of normal distributions, we store each sample of a probabilistic weight as the difference between the conductance values of two adjacent memory cells, as shown in (Fig. 3d). This method is particularly useful, because the difference between two normal distributions is still a normal distribution. Fig. 3e illustrates the corresponding technologically- plausible domain of normal distributions for both filamentary memristors \((\Gamma_{memristor})\) and phase- change memories \((\Gamma_{PCM})\) . Both filamentary memristors and phase change memories suffer from conductance instability + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2. Fabricated filamentary memristor and phase-change memory based array die. a Scanning electron microscopy image of a phase-change memory in the back end of line of our hybrid memristor/CMOS process. b Scanning electron microscopy image of a phase-change memory. c Optical microscopy photograph of the phase-change memory based 1T1R array. d Scanning electron microscopy image of a filamentary memristor in the back end of line of our hybrid memristor/CMOS process. e Scanning electron microscopy image of a filamentary memristor. f Optical microscopy photograph of the filamentary memristor based 1T1R array.
+ +over time, due to the local recombination of oxygen vacancies and structural relaxation of the material, respectively. However, the shape of the technologically plausible domain of normal distributions is only slightly altered by these effects (see Supplementary Note 1). + +## Hardware-calibrated training + +In Bayesian neural networks, the weights are probability distributions, given by the posterior probability distributions, \(p(\Omega |D)\) , where \(D\) is the training data. The most popular methods to approximate the posterior distributions are Markov Chain Monte Carlo (MCMC) sampling and variational inference (VI). We proposed the transfer of a Bayesian neural network trained by MCMC, an algorithm that samples the posterior exactly, in our previous work. However, MCMC lacks scalability and its training time is orders of magnitude longer than that of variational inference. MCMC methods typically require a huge number of samples to approximate the posterior, involving high memory density to store it, rendering them area and energy inefficient. Moreover, the mapping of the software posterior on hardware causes a loss in accuracy and estimation of both epistemic and aleatoric uncertainties of several percentage points (see Supplementary note 2). Here, we use the variational inference method, which scales better than MCMC. Rather than sampling from the exact posterior, the latter is approximated with normal distributions, \(q(\Omega |\theta)\) , where \(\theta\) represents the mean and standard deviation \((\mu , \sigma)\) . The estimation is performed by minimizing the loss function, the Kullback- Leibler divergence between \(p(\Omega |D)\) and \(q(\Omega |\theta)\) : + +\[Loss_{VI} = KL[q(\Omega |\theta)||p(\Omega |D)]. \quad (1)\] + +During the training phase, for each weight, \(\mu\) and \(\sigma\) are learned using the backpropagation algorithm (see Methods). Fig. 4a illustrates the domain of the normal distributions \(\theta = (\mu , \sigma)\) obtained after software training our reference arrhythmia classification task and mapping the software values to the conductance range achievable with filamentary memristors (blue) and phase- change memories (green). The mapping operation is a linear scaling of \(\theta = (\mu , \sigma)\) by a factor \(\gamma\) calculated to minimize the statistical distance between the normal distributions calculated by software and the available experimental ones (see Methods). However, this operation is not sufficient to meet the technology requirements: the desired domain exceeds the available experimental one for both filamentary memristors ( \(\Gamma_{memristor}\) ) and phase- change memories ( \(\Gamma_{PCM}\) ). To compel the learned normal distributions to match with the hardware experimental electrical characteristics, we imposed that \(\theta\) belong to the experimental \(\Gamma\) domain by adding the "technological loss" term to the loss function: + +\[Loss = Loss_{VI} - log(U_{\Gamma}(\theta)), \quad (2)\] + +where \(U_{\Gamma}(\theta)\) is a uniform distribution over the \(\Gamma = \Gamma_{memristor} / \gamma\) or the \(\Gamma = \Gamma_{PCM} / \gamma\) domain. Fig. 4b illustrates the effectiveness of the proposed hardware- calibrated training method: the normal distributions obtained by software simulations perfectly map on both phase- change memories and filamentary memristors experimental values. We demonstrated that by taking hardware physics into account while developing the training algorithm, it is possible to make variational inference a technologically plausible algorithm. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3. Filamentary memristor and phase-change memories as physical random variables with normal distribution. a Probability densities of 2,048 filamentary memristors programmed with eight different programming current values. b Domain of the Gaussian distributions experimentally achieved exploiting different programming conditions for filamentary memristors (blue) and phase-change memories (green). Triangles represent one-shot programming, dots represent iterative programming and the cross represents the low conductance state. c Probability densities of 2,048 phase-change memories programmed with seven different programming current values. d Schematic of the proposed synaptic circuit. Each sample of a Bayesian probabilistic weight is stored as the difference between the conductance values of two adjacent memory cells. e Domain of the normal distributions (Γ) that can be experimentally obtained exploiting the circuit in d by storing samples on two memory cells.
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4. Domains of normal distribution obtained with classical Variational Inference and the proposed technologically plausible method. a Domain of normal distributions \(\theta = (\mu , \sigma)\) obtained after training with the classical VI method and mapping the software values to the conductance range achievable with filamentary memristors (blue) and phase-change memories (green). b Domain of normal distributions \(\theta = (\mu , \sigma)\) obtained after training with the proposed method calibrated on filamentary memristors (blue) and phase-change memory experimental data (green).
+ +<--- Page Split ---> + +## Experimental uncertainty estimation + +To validate our approach, we programmed a Bayesian neural network, trained to recognize arrhythmia, onto a collection of the filamentary memristor dies (Fig. 2f). We classified ECG diagnosis beats using a two- layer Bayesian neural network featuring 32 inputs, 16 hidden neurons in the first layer, and nine output neurons in the second layer. We trained the Bayesian neural network, using the filamentary memristor technological loss, on nine classes: healthy beat and eight types of arrhythmias. During testing, we added a tenth class corresponding to a non- previously seen type of arrhythmia. Following the architecture presented in Fig. 1, we programmed \(M = 50\) independent realizations of model parameter vector \(\theta\) , representing the learned posterior \(q(\Omega |\theta)\) , i.e., we transferred each model realization into an array of conductance values. Each realization can fit in 1.5 dies presented in Fig. 2f (one die for the first layer of the neural network, and a half die for the second layer, see Methods), and described in detail in Supplementary Note 5; therefore, we needed to program a total of 75 dies. Multiply- and- accumulate operations were performed directly in memory using Ohm's and Kirchoff's law (see Methods and Supplementary Note 5). Activation functions were calculated in software. The array performed all the multiply and accumulate operations needed to classify 1000 beats in the test data set. Supplementary Note 6 recapitulates the different steps of our experiment, from training to inference. + +Fig. 5 presents the electrical characterization results of the memristor- based Bayesian neural network. To visualize the input data, we used the t- distributed stochastic neighbor embedding (t- SNE) statistical method (Figs. 5a- b). This visualization technique represents each high- dimensional input data by a point in a two- dimensional space, in the same way that similar data correspond to nearby points and distant points represent dissimilar data. Fig. 5a illustrates the two- dimensional projections of the input data used during inference on the test dataset. The data belonging to a given class (disease) display a "cluster". The unseen diseases (i.e., beats that do not belong to a class learned in the training phase) are the red points. Fig. 5b uses the same representation, where the colors represent data points correctly (blue) or incorrectly (orange) classified by our experiment, while the unseen disease data points are plotted in black. Our experiment recognizes \(75\%\) of the data points correctly. Most errors concern points that lie at the border between several clusters in the t- SNE plot, suggesting that they might be ambivalent (high aleatoric uncertainty cases). To investigate this idea further, Fig. 5c shows the measured probability density distributions of the measured aleatoric uncertainty, which provides a measure of the confidence of network prediction. The different colors represent correct predictions (blue), incorrect predictions (orange), and unseen data (red). The aleatoric uncertainty is lower than 0.5 for \(62\%\) of all correctly classified data points, while it is higher than 0.5 for \(97\%\) of all incorrectly classified data points and unseen disease data points. This result means that our experiment correctly determined as uncertain all of its errors and the unseen disease. It also flagged as uncertain some of its correct predictions, which is expected, as some of them might be ambivalent cases. + +The situation is quite different when we look at the measured epistemic uncertainty (Fig. 5d). \(97\%\) of all correctly and incorrectly classified data points have an epistemic uncertainty lower than 0.5. Conversely, \(98\%\) of the unseen disease data points have epistemic uncertainty higher than 0.5. These results mean that experiments can differentiate ambivalence between classes from the presentation of new unknown inputs. + +These results come in sharp contrast with those of a simulated conventional neural network with the same architecture. This type of neural network, by construction, has no epistemic uncertainty, and the aleatoric uncertainty tends to be extremely low whatever the input (Fig. 5e). This overconfidence is due to the small size of our dataset, making conventional neural networks particularly prone to overfitting. + +To push the interpretation of our experimental results further and make an in- depth assessment of the capability of our experiment to evaluate uncertainty, we used receiver operating characteristic (ROC) curves, a widely used metric for diagnostic ability, obtained by plotting the true positive rate as a function of the false positive rate for various threshold settings. A perfect classifier would yield the (0, 1) point, i.e., an area under the curve (AUC) of one, corresponding to no false negatives and no false positives. The ROC curve of a random classifier approaches the diagonal line, i.e., an area under the curve of 0.5. Fig. 5f shows the ROC curve corresponding to the differentiation between correct predictions and incorrect predictions, based on aleatoric uncertainty, for our experiment, a purely software version of the Bayesian neural network programmed in our experiment, and a conventional neural network with the same architecture (see Methods). Our experiment leads to a ROC curve close to the software Bayesian neural network, showing the high quality of our transfer. It even slightly outperforms the software version (area under the curve of 0.91 vs. 0.90). The conventional neural network has an inferior area under the curve (0.79), highlighting again the overconfidence of such networks. We should remark that an area under the curve of one is impossible for this graph, because predictions with very low aleatoric certainty are sometimes correct. + +Fig. 5g shows the ROC curve corresponding to the differentiation between known and unknown data, based on epistemic uncertainty (see Methods). Our experiment performs particularly well, with an area under the curve of 0.99, close to the perfect value of one, and which slightly outperforms the one obtained by the software (0.95) Bayesian neural network. This enhanced capability to recognize unknown data comes at the price of a small decrease in the general accuracy of the network (75% vs. 79%). We should point out that these differences between software and hardware are most probably not significant, as our test + +<--- Page Split ---> + +set features only 70 unknown data points (out of a dataset of 1000 points). We repeated the software experiment ten times and the area under the curve varied between 0.89 and 1. By contrast, the conventional neural network exhibited no capacity to recognize unknown data and had the diagonal ROC curve of a random classifier, with an area under the curve of 0.5. Table. 1 summarizes all these results. + +A drawback of our approach with regard to conventional neural networks is that we need several \((M)\) versions of the neural networks. This number, however, does not necessarily need to be high. Fig. 5h shows its effect on the Bayesian neural network accuracy and on its capability to evaluate uncertainty, measured by the area under the curve of the two ROC curves mentioned above, obtained using aleatoric and epistemic uncertainty. The accuracy and aleatoric area under the curve approach their saturation values with ten neural networks. The epistemic area under the curve takes a higher number of implementations to converge; however, with ten neural networks, it reaches 0.96, close to its maximum value (0.99). + +## Discussion + +This work demonstrates experimentally a simple and energy- efficient realization of a Bayesian neural network by directly storing the probabilistic weights into resistive memory- based crossbar arrays. The device variability in both filamentary- based memristors and phase- change memories is used to store physical random variables that sample analog conductance values from normal distributions with re- configurable mean and standard deviation. The Bayesian neural networks are trained following a special variational inference approach, incorporating a "technological loss" to overcome the hardware limitations linked to the device physics. We implemented a whole network using a collection of filamentary memristor arrays allowing in- memory computing. The resulting Bayesian neural network matches software simulations in terms of accuracy, and in terms of aleatoric and epistemic uncertainty evaluations, as evidenced by ROC curves for the identification of misclassified heartbeats and unknown data heartbeats. + +The dies that we used for phase- change memory characterization (Fig. 2c) are conventional memory arrays that do not allow in- memory computing and cannot be implemented as a full in- memory Bayesian neural network, unlike what we achieved for filamentary memristors. Therefore, we used our extensive statistical measurements of phase change memories (Fig. 4) to simulate such a network, using the simulator validated in Supplementary Note 3. The results are presented in Supplementary Note 4 and listed in Table 1; they suggest that the phase change memory network would function almost equivalently to the filamentary memristor- one, with only a slight reduction in terms of accuracy and uncertainty evaluation (expressed by the area under the two ROC curves). Fig. 4 shows that the mean value/standard deviation space that can be programmed on phase change memories is more skewed than that of filamentary memristors. (Indeed, it is impossible to program synapses with low mean value and high standard deviation on phase change memories). The fact that Bayesian neural networks based on the two memory technologies still achieve almost matching performance and uncertainty evaluations demonstrates the power of the "technological loss" term to correct for the constraints of technology. + +The most important limitation of our approach is that it requires the use of multiple devices per synapse to represent a distribution of its synaptic weight. The results of Fig. 5h show that the number of devices per synapse does not need to be large. Bayesian neural networks excel in relatively small- data regimes, where strong uncertainty is present: they are not large networks, making device overhead bearable. Currently- developed resistive memories integrated in three dimensions may be particularly suitable to our architecture, which features multiple devices per synapse40. + +We estimated the energy consumption of a final in- memory Bayesian neural network, based on measurements from a state- of- the- art 22- nanometer platform3. We found a cost of 270 nanojoules per inference, 800 times smaller than on a modern GPU fabricated in a 12- nanometer process (see Methods). This efficiency suggests that Bayesian neural networks can be used at the edge in extremely energy- constrained systems, such as medical devices, where reliable decisions are needed. + +## Methods + +## Filamentary and Phase Change Memory technology and circuits + +The circuits described in the Results section were fabricated using a low- power foundry 130- nanometers process with four metal layers. Both phase- change memories and filamentary memristors were fabricated on tungsten vias in metal layer four. The filamentary memristors consist of a 5- nanometer thick metallic bottom electrode, a 5- nanometer thick \(\mathrm{HfO_x}\) active layer deposited by atomic layer deposition, and a 10- nanometer thick Ti top electrode. The memory element is fabricated as a mesa structure with a 200- nanometer diameter. The phase- change memory architecture is characterized by a strip of chalcogenide material lying on top of a TiN heater element, with a thickness of five nanometers and a width of 100 nanometers. The chalcogenide layer is a germanium- antimony- tellurium alloy deposited by sputtering deposition and is 50- nanometer thick. A fifth layer of metal is deposited on top of both phase- change memories and filamentary memristors. + +Two different integrated circuits were used in this article, one integrating filamentary- based memristors and the other phase- change memories (Fig. 2). In both architectures, each memory cell is accessed by a transistor, giving rise to a one + +<--- Page Split ---> +![](images/Figure_5.jpg) + +
Figure 5. Measurements of the fabricated memristor-based Bayesian neural network. a tSNE visualization of input data, different colors representing different classes (diseases). Nearby points correspond to similar data and distant points to dissimilar data. b tSNE visualization of experimental data classification. The different colors represent points correctly or incorrectly predicted and unseen data. c Experimental probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases. d Experimental probability density distribution of the epistemic uncertainty for correct predictions, incorrect predictions and unseen diseases. e Simulated probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases for a conventional neural network with the same architecture and using float32 encoding for the synapses. f ROC curve corresponding to the differentiation between correct prediction and incorrect prediction, based on aleatoric uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). g ROC curve corresponding to the differentiation between known and unknown data, based on epistemic uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). h Measured accuracy, epistemic uncertainty, and aleatoric uncertainty (calculated as the area of the ROC curves) as a function of the number of filamentary memristor devices per synapse.
+ +Table 1. Comparison of accuracy and uncertainty prediction performances. + +
Conventional ANN (float 32)Bayesian (float 32)Bayesian Hardware (filamentary memristor experimental)Bayesian Hardware (phase-change memory simulation)
Accuracy classification80%79%75%73%
Prediction confidence (aleatoric) [AUC]0.790.900.910.85
Anomaly detection (epistemic) [AUC]0.50.950.990.96
+ +<--- Page Split ---> + +transistor- one- resistor (1T1R) unit cell. The transistor, used as a selector, was essential to control the programming current allowing multi- level programming of filamentary memristors. The phase- change memory chip was an array of 16,384 1T1R structures, only individually accessible. The filamentary- based memristor chip was an array of 1,024 1T1R cells arranged in a \(32\times 32\) configuration. This array enabled the selection of multiple memory points capable of performing parallel multiply and accumulate operations. Digital drivers were used to select multiple cells in parallel controlling the word lines (WLs), source lines (SLs), and bit lines (BLs). This array is described in detail in Supplementary Note 5. + +## Iterative programming + +The iterative programming methods adopted for filamentary- based and phase- change memories are different. For filamentary memristors, each device is re- programmed multiple times, with the same conditions, until its conductance reaches the target value (Algorithm 1). For the phase- change memories the programming voltage is increased or decreased at each cycle depending on the conductance value obtained in the previous cycle (Algorithm 2). + +Algorithm 1 Iterative programming for filamentary memristors + +1: \(G_{max}:\) Target conductance max +2: \(G_{min}:\) Target conductance min +3: \(I_{cc}:\) Compliance current for target distribution +4: \(i_{max}:\) Maximum number of iteration +5: \(G:\) filamentary memristor conductance +6: \(G\gets RESET\) 7: \(i\gets 0\) 8: while \(i< i_{max}\) . 9: \(G_{0}\gets SET(I_{cc})\) 10: \(i\gets i + 1\) 11: if \(G_{min}< G< G_{max}\) 12: end 13: else: 14: \(G_{0}\gets RESET\) 15: end + +Algorithm 2 Iterative programming for phase change memories + +1: \(G_{max}:\) Target conductance max +2: \(G_{min}:\) Target conductance min +3: \(V_{s}:\) Applied voltage +4: \(V_{max}:\) Maximum voltage +5: \(\delta V:\) Voltage increment +6: \(G:\) phase change memory conductance +7: \(G\gets RESET\) 8: \(V_{s}\gets V_{init}\) 9: while \(V_{s}< V_{max}\) and \(G< G_{min}\) . 10: \(G\gets SET(V_{s})\) 11: \(V_{s}\gets V_{s} + \delta V\) 12: while \(V_{s}< V_{max}\) and \(G > G_{max}\) . 13: \(G\gets RESET(V_{s})\) 14: \(V_{s}\gets V_{s} - \delta V\) 15: end + +Before the filamentary based memristors chip can be used, it is necessary to form all the devices. The forming operation consist in the following conditions: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} = 1.6 \mathrm{V}\) , \(V_{bl} \in [1.6,4] \mathrm{V}\) . The standard SET conditions are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [1.4,2.2] \mathrm{V}\) , \(V_{bl} = 1.8 \mathrm{V}\) . The standard RESET conditions used are as follows: \(V_{sl} = 2.6 \mathrm{V}\) , \(V_{wl} = 4.8 \mathrm{V}\) , \(V_{bl} = 0 \mathrm{V}\) . The off- chip generated voltage programming pulses have a pulse width of \(1 \mu \mathrm{s}\) for the SET and \(100 \mathrm{ns}\) for the RESET. For the phase change memory chip, the standard SET conditions are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [2,3] \mathrm{V}\) and \(V_{bl} = 4 \mathrm{V}\) . The standard RESET conditions used are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [0.9,4] \mathrm{V}\) and \(V_{bl} = 4.8 \mathrm{V}\) . The off- chip generated voltage programming pulses have a pulse width of \(300 \mathrm{ns}\) and a rise time of \(20 \mathrm{ns}\) . The fall time is \(1500 \mathrm{ns}\) for the SET and \(20 \mathrm{ns}\) for the RESET. + +## Weight and conductance mapping + +The mapping between the mean and standard deviation of the normal distributions obtained after software training, \(\theta_{s} = (\mu_{s}, \sigma_{s})\) , and the corresponding experimental conductance distributions \(\theta_{e} = (\mu_{e}, \sigma_{e})\) in microsiemens is a critical step. The normal distributions chosen by the training algorithm and reported in Fig. 4 were mapped to conductance values in microsiemens according to: + +\[\mu_{e} = \gamma \cdot \mu_{s} \qquad \sigma_{e} = \gamma \cdot \sigma_{s}. \quad (3)\] + +The \(\gamma\) scaling factor has been calculated by minimizing the Kullback- Leibler divergence between the experimental and simulated normal distributions: + +\[\gamma = \underset {\gamma \in \mathbb{R}}{\mathrm{argmin}}\sum_{j\in [1,S]}\underset {i\in [1,E]}{\mathrm{min}}KL(\theta_{e_{i}},\gamma \times \theta_{s_{j}}), \quad (4)\] + +where \(S\) is the number of the software normal distributions and E is the number of available experimental normal distributions. + +<--- Page Split ---> + +## Training using Bayes by Backprop and technological loss + +The training of a Bayesian neural network consists of computing the most likely models (i.e. the posterior distribution, \(p(\Omega |D))\) underlying the training dataset, \(D\) , and the prior belief, \(p(\Omega)\) : + +\[p(\Omega |D) = \frac{p(D|\Omega)p(\Omega)}{p(D)}. \quad (5)\] + +Here \(\Omega\) represents the neural network parameters, \(p(D|\Omega)\) is the likelihood, and \(p(D)\) is the evidence. This equation is unfortunately intractable. Variational Inference approximates the posterior distribution, \(p(\Omega |D)\) , with a simpler variational distribution, \(q(\Omega |\theta)\) , which structure is easier to evaluate39. Typically the variational distributions are normal distributions, where the variational parameters \(\theta = (\mu ,\sigma)\) represent their mean and standard deviation. The approximation of the \(\theta\) parameters, \(\theta^{*}\) , are calculated minimizing the Kullback- Leibler (KL) divergence between the variational distribution, \(q(\Omega |\theta)\) , and the posterior, \(p(\Omega |D)\) , as shown in Eq. 6. The KL divergence is a measure of the similarity between the two distributions. The calculation of the \(\theta^{*} \in \mathbb{R}\) parameters is achieved by backpropagation15. This combination of variational inference and backpropagation is called Bayes by Backprop and has been proved to be efficient for complex applications10. It identifies + +\[\theta^{*} = \underset {\theta \in \mathbb{R}}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)]). \quad (6)\] + +As illustrated in Fig. 4, resistive memories cannot implement all possible normal distributions, no matter the chosen technology flavour. The search of \(\theta^{*}\) should be limited inside \(\Gamma\) , where \(\Gamma\) represents the ensemble of experimental normal distributions that can be built with a given technology. To impose that \(\theta\) belongs to \(\Gamma\) , a "technological loss" term has been added to Eq. 6. The "technological loss" term is defined as \(- log(U_{\Gamma}(\theta))\) , where \(U_{\Gamma}\) is a uniform distribution over the experimental \(\Gamma\) domain, and Eq. 6 becomes: + +\[\theta^{*} = \underset {\theta \in \Gamma}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)]) = \underset {\theta \in \mathbb{R}}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)] - log(U_{\Gamma}(\theta)). \quad (7)\] + +The cost function resulting from Eq. 7 can be denoted as: + +\[F(D,\theta) = KL[q(\Omega |\theta)||p(\Omega)] - \mathbb{E}_{q(\Omega |\theta)}[log(p(D|\Omega))] - log(U_{\Gamma}(\theta)) \quad (8)\] + +Eq. 8 can be solved with classical Bayes by Backprop, and it ensures that the \(\theta^{*} \in \Gamma\) . + +## Experimental setup for arrhythmia classification + +The considered Bayesian neural network featured 32 inputs, 16 hidden neurons in the first layer and nine output neurons, corresponding to the nine different diseases (classes), in the second layer. Since we use conductance subtraction between two filamentary memristors to store one weight, our \(32 \times 32\) crossbar array could take 32 inputs and produce 16 outputs. To realize one sample of our two- layer neural network one and a half crossbar arrays are required ( \(32 \times 32\) cells for the first layer and \(16 \times 18\) cells for the second one). A Bayesian neural network is the collection of several \((M)\) samples, so when using \(32 \times 32\) crossbar arrays, \(1.5M\) arrays are needed. We fully characterized 15 crossbars arrays to implement a Bayesian neural network with \(M = 10\) samples. To reproduce a Bayesian neural network with more than \(M = 10\) samples, we recycled the 15 crossbar arrays exploiting the fact that the cycle- to- cycle and device- to- device variability are similar in filamentary memristors21. Therefore, by reprogramming the 15 arrays 5 times, which is equivalent to using 75 arrays. We obtain a Bayesian Neural Network with \(M = 50\) samples. + +The input data are ECG recordings32. A single heartbeat is a 700 ms recording, and it is converted into 32 features through a Fast- Fourier Transform (FFT). The 32 extracted features are the input of the \(M\) samples of the Bayesian neural network. Since the digital drivers generate only a single read voltage level, \(V_{read}\) (see Supplementary Note 5), each feature is converted into three- bit binary values, \((X_{j}\) with \(j = 0, \ldots , 2)\) . The three bits are applied sequentially to the input of the first layer of each \(s\) sample, with \(s = 1, \ldots , M\) . Each input voltage vector \(X_{j}\) is applied on the bit lines of the \(32 \times 32\) crossbar array to generate output current vector \((V_{read}\) , is applied to the selected bit lines, which correspond to an input one, the unselected bit lines are floating, which correspond to an input zero). The measured output current at the source lines, is the dot product operation through the first layer, \(W_{s} \cdot X_{j}\) , where \(W_{s}\) are the conductance values of a given sample (model realization) \(s\) . The output current for a given three- bit binary input is + +<--- Page Split ---> + +\[I_{i,s} = \frac{W_{s}\cdot X_{0} + 2\times W_{s}\cdot X_{1} + 4\times W_{s}\cdot X_{2}}{7}. \quad (9)\] + +Using this experimental values, we calculate the activation functions of the hidden neurons + +\[a_{i,s} = \frac{I_{i,s}^{+} - I_{i,s}^{-}}{\gamma\cdot V_{read}}, \quad (10)\] + +where \(\gamma\) is the scaling factor calculated with Eq. 4. Each activation function is converted to three- bit binary values. This operation is equivalent to the calculation of a clipped rectified linear unit (ReLu) activation function. The same method is applied to the second layer, in which the calculated activations are the new input. The probability that the input data \(X\) belongs to a given output class \(c\) for a given sample \(W_{s}\) using a softmax function is + +\[p(y = c|X,W_{s}) = \frac{e^{a_{c,s}}}{\sum_{j = 1}^{N}e^{a_{j,s}}}. \quad (11)\] + +The disease classification (i.e., the probability that the input data belong to a specific class of disease) is the average of the probability values calculated with Eq. 11 over the number of samples. The predicted class is calculated as the argmax of the disease classifications. The aleatoric and epistemic uncertainty are calculated with Eqs. 12, 13 and 14. + +## Uncertainty calculation + +Unlike conventional artificial neural networks, where the output values for predictions are point estimates, Bayesian neural networks provide predictive distributions. The total uncertainty in the prediction, i.e., the predictive uncertainty, can be calculated based on the softmax of the predictive distributions calculated according to Eq. 11: + +\[U_{p} = -\sum_{c = 1}^{N}\left(\frac{1}{M}\sum_{s = 1}^{M}p(y = c|X,W_{s})\right)log\left(\frac{1}{M}\sum_{s = 1}^{M}p(y = c |X,W_{s})\right). \quad (12)\] + +The predictive uncertainty (Eq.12) is the sum of epistemic and aleatoric uncertainties + +\[U_{p} = U_{a} + U_{e}. \quad (13)\] + +Decomposing the predictive uncertainty is important, as epistemic and aleatoric uncertainties give us different information. High epistemic uncertainty suggests that the input data is an outlier relative to the training data set. More training data near can therefore reduce epistemic uncertainty, but does not help aleatoric uncertainty. Aleatoric uncertainty is uncertainty in data, to reduce it more refined input data are required (e.g., more powerful sensors). The aleatoric uncertainty can be obtained as: + +\[U_{a} = -\frac{1}{M}\sum_{s = 1}^{M}\sum_{c = 1}^{N}p(y = c|X,W_{s})logp(y = c|X,W_{s}). \quad (14)\] + +## Energy consumption estimates + +To estimate the energy consumption of the Bayesian neural network we first calculated the number of dot product operations for one inference: + +\[O p e r a t i o n s = 2\cdot I_{l}\cdot H_{l} + 2\cdot H_{l}\cdot O_{l}. \quad (15)\] + +Here \(I_{l}\) is the input length, \(H_{l}\) is the hidden layer length, and \(O_{l}\) is the output length. The factor two is due to fact that each sample of a Bayesian probabilistic weight is stored as the difference between the conductance values stored in two memory cells. One inference costs 1344 operations. The cost of a single analog Multiply- and- Accumulate (MAC) operation in a resistive memory- based analog in- memory computing circuit depends on the input and output size and on the weight precision. The cost of an analog MAC operation using \(22\mathrm{nm}\) CMOS technology with \(4\mathrm{b}\) input signal, \(4\mathrm{b}\) weight, and and \(11\mathrm{b}\) output is evaluated + +<--- Page Split ---> + +27 nJ according to ref.3. We used this value to estimate the dot product operation in our circuit, which has 3 b input signal, 4 b weight, and 8 b output. Assuming that a Bayesian inference requires ten single inference operations the estimated energy consumption for our circuit is \(270~\mathrm{nJ}\) . + +To gain a perspective on the energy efficiency of the proposed approach compared to conventional hardware, we benchmarked this figure to the energy required for running the same Bayesian neural network on a state- of- the- art Tesla V100 GPU, which uses a power consumption of 43 W executing the code provided with the Bayes by backprop paper15. The execution time for ten inference operations is \(5\mu \mathrm{s}\) . Based on these estimations, the proposed resistive memory- based analog circuit achieves a reduction of about a factor 800 in energy consumption relative to a conventional digital GPU. + +## Data availability + +All the measured data are available upon request. + +## Code availability + +All software programs used in the presentation of the Article are available upon request. + +## References + +1. Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521, 61 (2015). + +2. Ambrogio, S. et al. Equivalent-accuracy accelerated neural-network training using analogue memory. Nature 558, 60 (2018). + +3. Xue, C.-X. et al. A cmos-integrated compute-in-memory macro based on resistive random-access memory for ai edge devices. Nat. Electron. 4, 81-90 (2021). + +4. Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641-646 (2020). + +5. Wan, W. et al. 33.1 a 74 tmacs/w cmos-ram neurosynaptic core with dynamically reconfigurable dataflow and in-situ transposable weights for probabilistic graphical models. In 2020 IEEE International Solid-State Circuits Conference- (ISSCC), 498-500 (IEEE, 2020). + +6. Jung, S. et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature 601, 211-216 (2022). + +7. Khaddam-Aljameh, R. et al. Hermes-core—a 1.59-tops/mm 2 pcm on 14-nm cmos in-memory compute core using 300-ps/lsb linearized cco-based adcs. IEEE J. Solid-State Circuits 57, 1027-1038 (2022). + +8. Kabir, H. D., Khosravi, A., Hosen, M. A. & Nahavandi, S. Neural network-based uncertainty quantification: A survey of methodologies and applications. IEEE access 6, 36218-36234 (2018). + +9. Jospin, L. V., Buntine, W., Boussaid, F., Laga, H. & Bennamoun, M. Hands-on bayesian neural networks—a tutorial for deep learning users. arXiv preprint arXiv:2007.06823 (2020). + +10. Kendall, A. & Gal, Y. What uncertainties do we need in bayesian deep learning for computer vision? Adv. neural information processing systems 30 (2017). + +11. Szegedy, C. et al. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199 (2013). + +12. Goodfellow, I., Bengio, Y., Courville, A. & Bengio, Y. Deep learning, vol. 1 (MIT press Cambridge, 2016). + +13. Der Kiureghian, A. & Ditlevsen, O. Aleatory or epistemic? does it matter? Struct. safety 31, 105-112 (2009). + +14. Ghahramani, Z. Probabilistic machine learning and artificial intelligence. Nature 521, 452-9 (2015). + +15. Blundell, C., Cornebise, J., Kavukcuoglu, K. & Wierstra, D. Weight uncertainty in neural network. In International conference on machine learning, 1613-1622 (PMLR, 2015). + +16. Neal, R. M. Bayesian learning for neural networks, vol. 118 (Springer Science & Business Media, 2012). + +17. Gal. Uncertainty in deep learning. PhD thesis, Univ. Camb. (2016). + +18. Dalgaty, T. et al. In situ learning using intrinsic memristor variability via markov chain monte carlo sampling. Nat. Electron. 4, 151-161 (2021). + +19. Joshi, V. et al. Accurate deep neural network inference using computational phase-change memory. Nat. communications 11, 1-13 (2020). + +<--- Page Split ---> + +20. Tsai, H. et al. Inference of long-short term memory networks at software-equivalent accuracy using 2.5 m analog phase change memory devices. In 2019 Symposium on VLSI Technology, T82-T83 (IEEE, 2019). + +21. Dalgaty, T., Esmanhotto, E., Castellani, N., Querlioz, D. & Vianello, E. Ex situ transfer of bayesian neural networks to resistive memory-based inference hardware. Adv. Intell. Syst. 3, 2000103 (2021). + +22. Esmanhotto, E. et al. High-density 3d monolithically integrated multiple 1t1r multi-level-cell for neural networks. In 2020 IEEE International Electron Devices Meeting (IEDM), 36-5 (IEEE, 2020). + +23. Fortunato, M., Blundell, C. & Vinyals, O. Bayesian recurrent neural networks. arXiv preprint arXiv:1704.02798 (2017). + +24. Sebastian, A. et al. Two-dimensional materials-based probabilistic synapses and reconfigurable neurons for measuring inference uncertainty using bayesian neural networks. Nat. communications 13, 1-10 (2022). + +25. Liu, S. et al. Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing. Front. Nanotechnol. 78 (2022). + +26. Faria, R., Camsari, K. Y. & Datta, S. Implementing bayesian networks with embedded stochastic mram. AIP Adv. 8, 045101 (2018). + +27. Vodenicarevic, D. et al. Low-energy truly random number generation with superparamagnetic tunnel junctions for unconventional computing. Phys. Rev. Appl. 8, 054045 (2017). + +28. Friedman, J. S., Calvet, L. E., Bessiere, P., Droulez, J. & Querlioz, D. Bayesian inference with muller c-elements. IEEE Transactions on Circuits Syst. I: Regul. Pap. 63, 895-904 (2016). + +29. Zheng, Y. et al. Hardware implementation of bayesian network based on two-dimensional memtransistors. Nat. communications 13, 1-11 (2022). + +30. Harabi, K.-E. et al. A memristor-based bayesian machine. Nat. Electron. 1-12 (2022). + +31. Gao, D. et al. Bayesian inference based robust computing on memristor crossbar. In 2021 58th ACM/IEEE Design Automation Conference (DAC), 121-126 (IEEE, 2021). + +32. Moody, G. B., Mark, R. G. & Goldberger, A. L. Physionet: a web-based resource for the study of physiologic signals. IEEE Eng. Medicine Biol. Mag. 20, 70-75 (2001). + +33. Wan, W. et al. A compute-in-memory chip based on resistive random-access memory. Nature 608, 504-512 (2022). + +34. Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521 (2014). + +35. Balatti, S., Ambrogio, S., Gilmer, D. C. & Ielmini, D. Set variability and failure induced by complementary switching in bipolar ram. IEEE electron device letters 34, 861-863 (2013). + +36. Esmanhotto, E. et al. Experimental demonstration of multilevel resistive random access memory programming for up to two months stable neural networks inference accuracy. Adv. Intell. Syst. 2200145 (2022). + +37. Le Gallo, M. et al. Precision of bit slicing with in-memory computing based on analog phase-change memory crossbars. Neuromorphic Comput. Eng. 2, 014009 (2022). + +38. Hoffman, M. D. & Gelman, A. The no-u-turn sampler: adaptively setting path lengths in hamiltonian monte carlo. J. Mach. Learn. Res. 15, 1593-1623 (2014). + +39. Blei, D. M., Kucukelbir, A. & McAuliffe, J. D. Variational inference: A review for statisticians. J. Am. statistical Assoc. 112, 859-877 (2017). + +40. Ezzedeen, M. et al. Ultrahigh-density 3-d vertical rram with stacked junctionless nanowires for in-memory-computing applications. IEEE Transactions on Electron Devices 67, 4626-4630, DOI: 10.1109/TED.2020.3020779 (2020). + +## Acknowledgements + +This work was supported by European Research Council consolidator grant DIVERSE (reference: 101043854) and by European Research Council starting grant NANOINFER (reference: 715872). It also benefits from a France 2030 government grant managed by the French National Research Agency (ANR-22-PEEL-0010). In addition, we thank L. Hutin, S. Bonnetier, F. Andrieu, J. Arcamone, J. Grollier, P. Bessiere and J. Droulez for discussing various aspects of the article. + +<--- Page Split ---> + +## Author contributions + +D.B. and T.H. proposed the initial idea of the hardware- calibrated training algorithm. D.B, T.H. D.Q, and E.V. conceived the experiments. D.B. and V.M. performed the experiments with the phase- change memory array. D.B., S.M., and N.C. performed the inference measurements on the two- layer Bayesian neural network. D.B. and T.H. conducted the software experiments and analysed the data. T.D. and A.M. performed preliminary studies concerning Bayesian neural networks and uncertainty evaluation. E.E. designed the circuits, under the supervision of J.M.P. The circuits were fabricated at CEA- Leti under the supervision of J.F.N. and G.B. D.Q. and E.V. supervised the work and wrote the initial version of the manuscript. All authors discussed the results and reviewed the manuscript. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- SupplInfoBringinguncertaintyquantificationtotheedge.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f_det.mmd b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..0d7e75e183a393907764ebe37e819a5f4958d03b --- /dev/null +++ b/preprint/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f/preprint__1750d29ced67b94598803a387ee4566fca4c9c13279e5f1996ad43943b84aa0f_det.mmd @@ -0,0 +1,572 @@ +<|ref|>title<|/ref|><|det|>[[44, 108, 920, 207]]<|/det|> +# Bringing uncertainty quantification to the extreme-edge with memristor-based Bayesian neural networks + +<|ref|>text<|/ref|><|det|>[[44, 230, 371, 272]]<|/det|> +Djohan Bonnet Université Grenoble Alpes, CEA, LETI + +<|ref|>text<|/ref|><|det|>[[44, 277, 120, 315]]<|/det|> +Tifenn Hirtzlin CNRS + +<|ref|>text<|/ref|><|det|>[[44, 323, 738, 365]]<|/det|> +Atreya Majumdar Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies + +<|ref|>text<|/ref|><|det|>[[44, 370, 714, 411]]<|/det|> +Thomas Dalgaty CEA- Leti, Université Grenoble Alpes https://orcid.org/0000- 0003- 0326- 2121 + +<|ref|>text<|/ref|><|det|>[[44, 416, 371, 457]]<|/det|> +Eduardo Esmanhotto Université Grenoble Alpes, CEA, LETI + +<|ref|>text<|/ref|><|det|>[[44, 463, 371, 504]]<|/det|> +Valentina Meli Université Grenoble Alpes, CEA, LETI + +<|ref|>text<|/ref|><|det|>[[44, 509, 384, 550]]<|/det|> +Niccolò Castellani CEA, LETI, Minatec Campus, Grenoble + +<|ref|>text<|/ref|><|det|>[[44, 556, 371, 597]]<|/det|> +Simon Martin Université Grenoble Alpes, CEA, LETI + +<|ref|>text<|/ref|><|det|>[[44, 602, 225, 642]]<|/det|> +Jean- Francois Nodin CEA- LETI + +<|ref|>text<|/ref|><|det|>[[44, 649, 384, 689]]<|/det|> +Guillaume Bourgeois CEA, LETI, Minatec Campus, Grenoble + +<|ref|>text<|/ref|><|det|>[[44, 695, 210, 713]]<|/det|> +Jean- Michel Portal + +<|ref|>text<|/ref|><|det|>[[44, 716, 880, 736]]<|/det|> +Aix- Marseille Université, CNRS, Institut Matériaux Microélectronique Nanosciences de Provence + +<|ref|>text<|/ref|><|det|>[[44, 741, 191, 759]]<|/det|> +Damien Querlioz + +<|ref|>text<|/ref|><|det|>[[44, 762, 897, 803]]<|/det|> +Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies, 91120 Palaiseau, France + +<|ref|>text<|/ref|><|det|>[[44, 808, 400, 828]]<|/det|> +Elisa Vianello ( elisa.vianello@cea.fr) + +<|ref|>text<|/ref|><|det|>[[52, 831, 725, 851]]<|/det|> +Université Grenoble Alpes, CEA, LETI https://orcid.org/0000- 0002- 8868- 9951 + +<|ref|>text<|/ref|><|det|>[[44, 892, 101, 909]]<|/det|> +Article + +<|ref|>text<|/ref|><|det|>[[44, 930, 135, 948]]<|/det|> +Keywords: + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 46, 330, 65]]<|/det|> +Posted Date: January 13th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 84, 474, 103]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 2458251/v1 + +<|ref|>text<|/ref|><|det|>[[42, 120, 911, 164]]<|/det|> +License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[44, 182, 531, 202]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 236, 956, 280]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on November 20th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 43317- 9. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[90, 74, 907, 164]]<|/det|> +# Bringing uncertainty quantification to the extreme-edge with memristor-based Bayesian neural networks + +<|ref|>text<|/ref|><|det|>[[90, 172, 888, 245]]<|/det|> +Djohan Bonnet \(^{1,2*}\) , Tifenn Hirtzlin \(^{1}\) , Atreya Majumdar \(^{2}\) , Thomas Dalgaty \(^{3}\) , Eduardo Esmanhotto \(^{1}\) , Valentina Meli \(^{1}\) , Niccolo Castellani \(^{1}\) , Simon Martin \(^{1}\) , Jean- François Nodin \(^{1}\) , Guillaume Bourgeois \(^{1}\) , Jean- Michel Portal \(^{4}\) , Damien Querlioz \(^{2*}\) , and Elisa Vianello \(^{1*}\) + +<|ref|>text<|/ref|><|det|>[[90, 262, 908, 340]]<|/det|> +\(^{1}\) Université Grenoble Alpes, CEA, LETI, Grenoble, France \(^{2}\) Université Paris- Saclay, CNRS, Centre de Nanosciences et de Nanotechnologies, Palaiseau, France \(^{3}\) Université Grenoble Alpes, CEA, LIST, Grenoble, France \(^{4}\) Aix- Marseille Université, CNRS, Institut Matériaux Microélectronique Nanosciences de Provence, Marseille, France \(^{*}\) djohan.bonnet@cea.fr, damien.querlioz@c2n.upsaclay.fr, elisa.vianello@cea.fr + +<|ref|>sub_title<|/ref|><|det|>[[90, 362, 198, 380]]<|/det|> +## ABSTRACT + +<|ref|>text<|/ref|><|det|>[[90, 397, 904, 635]]<|/det|> +Safety- critical sensory processing applications, like medical diagnosis, require making accurate decisions based on a small amount of noisy input data. For these applications, using Bayesian neural networks, able to quantify the uncertainty of the predictions, is a superior approach to using conventional artificial neural networks. However, because of the probabilistic nature of Bayesian neural networks, they can be computationally intensive to use for inference stage and thus not well suited for extreme- edge applications. An emerging idea to solve this problem is to use the intrinsic probabilistic nature of memristors to efficiently implement Bayesian neural network inference: the variability in the resistance of memristors would represent the probability distribution of weights in Bayesian neural networks. However, when using memristors, statistical effects follow the laws of device physics, whereas in Bayesian neural networks, those effects can take on arbitrary shapes. In this work, we overcome this difficulty by adopting a dedicated synapse architecture based on two memristors, and by training Bayesian neural networks with a dedicated variational inference technique that includes a "technological loss" to take into account specificities of memristor physics. This technique allowed us to program a two- layer Bayesian neural network on 75 physical crossbar arrays of 1,024 memristors, incorporating CMOS periphery circuitry to do in- memory computing, to classify arrhythmia in electrocardiograms. Our experimental neural network classified heartbeats with high accuracy, and estimated the certainty of its predictions. In the case of uncertain predictions, it differentiated between ambivalent heartbeats (aleatoric uncertainty), and heartbeats with never- seen patterns (epistemic uncertainty). We show that our technique can also be used with phase change memories, by employing a different "technological loss" term. The great advantage of this approach is its low energy consumption: we estimate an 800 times improvement in energy efficiency compared to a GPU performing the same task. + +<|ref|>sub_title<|/ref|><|det|>[[70, 670, 206, 687]]<|/det|> +## 1 Introduction + +<|ref|>text<|/ref|><|det|>[[66, 694, 910, 920]]<|/det|> +Hardware neural networks based on emerging non- volatile memories can bring intelligence to the edge at a very low energetic cost. In this context, filamentary memristors and phase change memories can be used in a very elegant and energy- efficient way. These devices can act as analog synaptic weights enabling neural network multiply- and- accumulate operations directly in memory by relying on Ohm's law and Kirchoff's current law \(^{1 - 7}\) . Low- power systems of this kind could provide essential services: for example, medical devices could analyze patient measurements and detect life- threatening emergencies or automatically adjust treatment. However, some features of conventional neural networks are not well suited to such applications. In particular, conventional neural networks are notoriously bad at evaluating uncertainty \(^{8 - 10}\) . When trained with small datasets, as is typically the case in medical applications, conventional neural networks tend to "overfit" training data and to provide highly certain answers in all situations \(^{11,12}\) . More importantly, uncertainty in a machine learning context can have different origins, usually referred to as aleatoric and epistemic, which conventional neural networks cannot tell apart \(^{10,13}\) . To illustrate the impact of this limitation, we can use the example of a medical device trained to recognize different types of arrhythmia (irregular heartbeats) in a patient. Aleatoric uncertainty characterizes situations where a measurement could be consistent with different types of arrhythmia. Epistemic uncertainty, on the other hand, arises when an arrhythmia type that is significantly different from the data used to train the device appears and must be classified. Recognizing and differentiating those two types of uncertainty is essential, as they could be indicating that the patient's condition has evolved \(^{13,14}\) . + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[117, 245, 883, 654]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 666, 910, 744]]<|/det|> +
Figure 1. General architecture of the Bayesian neural network. a Schematic of the Bayesian neural network used for heart disease (arrhythmia) classification. In Bayesian neural networks, the weights are represented by probability distributions, thus naturally including uncertainty in the model. b Example of output neuron activation distributions, obtained for certain output, uncertain output, due to noisy input data, and unknown data (i.e. out-of-distribution data). c Experimental setup. d Hardware implementation of a Bayesian neural network by combining multiple versions of ANNs.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[72, 80, 909, 172]]<|/det|> +Bayesian neural networks are an alternative class of neural networks where synaptic weights and neuron activations do not take on unique values but are instead represented by probability distributions \(^{15 - 17}\) . Such networks are trained to provide models that provide plausible explanations for the training data, instead of just fitting this data; for this reason, Bayesian neural networks mitigate overfitting \(^{16}\) . And because they model certainty well, they excel at identifying aleatoric and epistemic uncertainty \(^{10}\) . Since computations based on Bayesian neural networks perform operations with probability distributions instead of numerical values, they are hard to materialize with memristors or phase change memories. + +<|ref|>text<|/ref|><|det|>[[72, 172, 910, 306]]<|/det|> +In our work, we use the variability and imperfections of memory devices (which can be represented by probability distributions) as the actual probability distributions of the neural network. This approach is challenging because the statistical properties of memristors and phase change memories are governed by strict rules of device physics \(^{18 - 20}\) . As a consequence, when filamentary oxide- based memristors are at higher resistance states, they will tend to have broader probability distributions, and when they are in lower resistance states they will tend to have narrow resistance states \(^{21,22}\) . In other words, the mean value and the standard deviation of the resistance are correlated. By contrast, the probability distribution of parameters in a Bayesian neural network can take any shape \(^{15,23}\) . Two recent publications have proposed new devices where the inherent probability distribution of their resistance can be tuned. These solutions, which exploit two- dimensional materials \(^{24}\) and magnetic devices \(^{25}\) , were validated with simulations of Bayesian neural networks. + +<|ref|>text<|/ref|><|det|>[[72, 307, 909, 398]]<|/det|> +In this work, we chose to focus on standard nanodevices (filamentary memristors) to demonstrate the first complete nanodevice- based Bayesian neural network implementation of a real- world task. Our demonstrator is able to classify arrhythmia recordings by types, with accurate aleatoric and epistemic uncertainty, using 75 arrays of fabricated \(32 \times 32\) memristor chips that associate hafnium oxide memristors and CMOS periphery circuitry. This array circuit performs in- memory computations based on Kirchoff's laws. Our work overcomes the problems related to the correlation of statistical properties of memristors, using three insights: + +<|ref|>text<|/ref|><|det|>[[112, 406, 909, 469]]<|/det|> +- The Bayesian neural network is trained using variational inference \(^{16}\) , a method that ensures statistical independence between synapses. A prior simulation study \(^{21}\) suggested Markov Chain Monte Carlo training, leading to statistically-dependent synapses, which is extremely challenging to realize experimentally due to memristor imperfection (see Supplementary Note 2). + +<|ref|>text<|/ref|><|det|>[[111, 478, 907, 510]]<|/det|> +- Each synapse is implemented using two memristors that are programmed independently, allowing the partial decorrelation of mean values and standard deviations of synaptic weights. This idea was presented in simulation studies \(^{24,25}\) . + +<|ref|>text<|/ref|><|det|>[[112, 519, 909, 565]]<|/det|> +- During the off-chip training process, we introduce a "technological loss" term to match the Bayesian neural networks to the statistics that can be achieved physically. This idea is proposed here for the first time, and it has been critical in making our experimental system work. + +<|ref|>text<|/ref|><|det|>[[72, 574, 909, 620]]<|/det|> +Our approach is generic. Using a different expression for the technological loss, we show that the same technique can be applied to phase change memory (by following a hybrid experimental-simulation methodology, employing a fabricated array of 16,384 phase change memories). + +<|ref|>text<|/ref|><|det|>[[72, 620, 909, 756]]<|/det|> +Other works have already pointed out connections between Bayesian concepts and nanodevices. Ref. \(^{18}\) exploited the probabilistic nature of memristors to perform Bayesian learning. This approach can only be applied to small- scale tasks, but unlike Bayesian neural networks, it does not suffer from the limitations imposed by the correlation of mean value and standard deviation of memristors \(^{18}\) . The works of \(^{26 - 30}\) use nanodevices to carry out Bayesian network inference. Bayesian networks differ from Bayesian neural networks. Bayesian networks are constructed using expert knowledge and are fully explainable which makes them ideal for tasks like sensor fusion. Conversely, Bayesian neural networks are trained from the ground up and excel on more data- intensive tasks like electrocardiogram or electroencephalogram classification. Finally, the work of Ref. \(^{31}\) treats memristors as Bayesian variables and uses them to program deterministic neural networks in order to increase hardware resilience. + +<|ref|>text<|/ref|><|det|>[[72, 756, 909, 817]]<|/det|> +In this paper, we introduce and describe the general architecture of our Bayesian neural networks, which are based on memory devices, and our technique to match imperfections in nanodevices with the probability distributions of a Bayesian neural network. We then show the experimental classification of arrhythmia with a proper uncertainty evaluation, using 75 arrays of filamentary memristors. + +<|ref|>sub_title<|/ref|><|det|>[[72, 835, 163, 852]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[72, 860, 476, 875]]<|/det|> +## Memory devices-based Bayesian neural networks + +<|ref|>text<|/ref|><|det|>[[72, 876, 909, 922]]<|/det|> +For our experiments, we considered a two- layer Bayesian neural network (Fig. 1a,b), trained to differentiate nine classes of heart arrhythmia from electrocardiogram (ECG) recordings \(^{32}\) . While in Artificial Neural Networks (ANNs), the synaptic weights are point estimates, Bayesian neural networks replace them with probability distributions. A natural way to implement a Bayesian + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[85, 80, 909, 187]]<|/det|> +neural network with memristors or phase change memories is to use a collection of \(M\) distinct memory arrays to represent each layer of the neural network (Fig. 1c,d). We sample \(M\) weight values for each synapse based on its probability distribution in the Bayesian neural network, and we program them to the \(M\) memory array. We will see in the next sections that the inherent probabilistic effects in memory devices allow us to perform the sampling and programming operations simultaneously, transforming the largest drawback of emerging memory devices, their variability, into a feature. This approach leads to a collection of \(M\) independent in- memory neural networks. By presenting the same input to each of these arrays, we obtain a collection of \(M\) different outputs representing the output distribution of these neural networks. + +<|ref|>text<|/ref|><|det|>[[85, 188, 910, 444]]<|/det|> +The benefit of using distributions, instead of deterministic values, is that we can quantify the uncertainty of the neural network's output. Intuitively, each of the \(M\) independent neural networks constitutes a reasonable hypothesis explaining the data used to train the Bayesian neural network. The spread of the output distributions captures the certainty or lack of certainty in the model's predictions. The uncertainty depends on both the mean and the variance values of the output neuron activations, as illustrated in Figs. 1b- d. If an output is highly certain ("clear classification"), all \(M\) neural networks will have the same active output neuron (value close to one), and inactive output neurons (value close to zero), and both the aleatoric and epistemic uncertainties will be low (see Methods for the mathematical definitions of aleatoric and epistemic uncertainties). If measurement imprecision causes ambivalence in the predictions ("unclear classification"), e.g., the features allowing differentiation of the arrhythmia types are lost in noise, none of the \(M\) neural networks will provide a certain prediction: all of them will hesitate between the same classes, with non- zero outputs on these classes. The variance of the output remains low, as all \(M\) networks have a consistent behavior: aleatoric uncertainty increases, whereas epistemic uncertainty does not. But what happens if a new arrhythmia type that was not present during training appears, and the network has to classify it? This is an example of out- of- distribution test data (case "Unknown data" in Figs. 1b- d). All \(M\) neural networks sampled from the Bayesian neural network will tend to make a different interpretation of this unknown data resulting in a high variance in the distribution of output neurons. Both aleatoric and epistemic uncertainty will be high. Classifying out- of- distribution data, therefore, is possible by measuring output distributions. This is of fundamental importance for safety- critical applications like medical diagnoses or autonomous driving. + +<|ref|>text<|/ref|><|det|>[[86, 445, 909, 521]]<|/det|> +Performing inference with our approach requires massive parallel Multiply- and- Accumulate (MAC) operations. These operations are power- hungry when carried out on CMOS- based ASICs and field- programmable gate arrays, due to the shuttling of data between processor and memory. In this work we use crossbars of memristors that naturally implement the multiplication between the input voltage and the probabilistic synaptic weight through Ohm's law, and the accumulation through Kirchhoff's current law \(^{2,7,33,34}\) , to significantly lower power consumption. + +<|ref|>sub_title<|/ref|><|det|>[[88, 540, 668, 556]]<|/det|> +## Filamentary memristor and phase-change memory as normal distributions + +<|ref|>text<|/ref|><|det|>[[85, 557, 910, 753]]<|/det|> +Among the non- volatile memories that can be integrated in advanced commercial processes, phase- change memories (PCM) and filamentary memristors have been widely studied for analog in- memory neural network implementation, because of the possibility of adjusting the conductance level of these devices. In our previous work, we demonstrated that the intrinsic variability in filamentary memristors can be leveraged to store the probabilistic weights of Bayesian neural networks \(^{21}\) . However, the conductance distribution follows strict rules due to device physics: the mean value, \(\mu\) , and the standard deviation, \(\sigma\) , are strongly correlated. Phase- change memories suffer from the same limitation \(^{19}\) . Bayesian neural networks require a larger space of normal distribution with mean values that are uncorrelated with the standard deviations. Here, we come up with a new synaptic circuit and the associated programming strategy to obtain largely unrestricted \(\mu\) and \(\sigma\) values. To illustrate the practicality of the proposed solution, we fabricated and tested arrays of hafnium- oxide- based filamentary memristors and of germanium- antimony- tellurium phase- change memories in a one- transistor- one- resistor (1T1R) configuration (Fig. 2). Both memory technologies have been integrated into the back end of line (BEOL) of a 130- nanometer foundry CMOS process with four metal layers (see Methods). Figs. 3a and c show the distributions of 2,048 filamentary- based memristors and phase- change memories, respectively, programmed in eight conductance levels. + +<|ref|>text<|/ref|><|det|>[[85, 755, 910, 920]]<|/det|> +In both cases, the standard deviation of the distribution is related to its mean value and cannot be chosen independently. The resulting domain of normal distributions that can be achieved by exploiting device variability \((\sigma)\) is thus bounded to a one- dimensional space for both technologies (Fig. 3b). In filamentary memristors, \(\sigma\) decreases for increasing conductance values due to the Poisson- like spread of the number of defects injected during the programming operation \(^{35}\) . In phase- change memories, the trend is inverted: device variability increases with the conductance values, moving from a full to a partial amorphous material \(^{19,20}\) . For both technologies, the standard deviation can be reduced by adopting an iterative programming- and- verify scheme (see Methods). To extend the domain of normal distributions, we store each sample of a probabilistic weight as the difference between the conductance values of two adjacent memory cells, as shown in (Fig. 3d). This method is particularly useful, because the difference between two normal distributions is still a normal distribution. Fig. 3e illustrates the corresponding technologically- plausible domain of normal distributions for both filamentary memristors \((\Gamma_{memristor})\) and phase- change memories \((\Gamma_{PCM})\) . Both filamentary memristors and phase change memories suffer from conductance instability + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[109, 76, 864, 247]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 255, 908, 350]]<|/det|> +
Figure 2. Fabricated filamentary memristor and phase-change memory based array die. a Scanning electron microscopy image of a phase-change memory in the back end of line of our hybrid memristor/CMOS process. b Scanning electron microscopy image of a phase-change memory. c Optical microscopy photograph of the phase-change memory based 1T1R array. d Scanning electron microscopy image of a filamentary memristor in the back end of line of our hybrid memristor/CMOS process. e Scanning electron microscopy image of a filamentary memristor. f Optical microscopy photograph of the filamentary memristor based 1T1R array.
+ +<|ref|>text<|/ref|><|det|>[[88, 368, 909, 415]]<|/det|> +over time, due to the local recombination of oxygen vacancies and structural relaxation of the material, respectively. However, the shape of the technologically plausible domain of normal distributions is only slightly altered by these effects (see Supplementary Note 1). + +<|ref|>sub_title<|/ref|><|det|>[[89, 427, 313, 443]]<|/det|> +## Hardware-calibrated training + +<|ref|>text<|/ref|><|det|>[[88, 443, 910, 611]]<|/det|> +In Bayesian neural networks, the weights are probability distributions, given by the posterior probability distributions, \(p(\Omega |D)\) , where \(D\) is the training data. The most popular methods to approximate the posterior distributions are Markov Chain Monte Carlo (MCMC) sampling and variational inference (VI). We proposed the transfer of a Bayesian neural network trained by MCMC, an algorithm that samples the posterior exactly, in our previous work. However, MCMC lacks scalability and its training time is orders of magnitude longer than that of variational inference. MCMC methods typically require a huge number of samples to approximate the posterior, involving high memory density to store it, rendering them area and energy inefficient. Moreover, the mapping of the software posterior on hardware causes a loss in accuracy and estimation of both epistemic and aleatoric uncertainties of several percentage points (see Supplementary note 2). Here, we use the variational inference method, which scales better than MCMC. Rather than sampling from the exact posterior, the latter is approximated with normal distributions, \(q(\Omega |\theta)\) , where \(\theta\) represents the mean and standard deviation \((\mu , \sigma)\) . The estimation is performed by minimizing the loss function, the Kullback- Leibler divergence between \(p(\Omega |D)\) and \(q(\Omega |\theta)\) : + +<|ref|>equation<|/ref|><|det|>[[130, 633, 907, 652]]<|/det|> +\[Loss_{VI} = KL[q(\Omega |\theta)||p(\Omega |D)]. \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[88, 658, 910, 797]]<|/det|> +During the training phase, for each weight, \(\mu\) and \(\sigma\) are learned using the backpropagation algorithm (see Methods). Fig. 4a illustrates the domain of the normal distributions \(\theta = (\mu , \sigma)\) obtained after software training our reference arrhythmia classification task and mapping the software values to the conductance range achievable with filamentary memristors (blue) and phase- change memories (green). The mapping operation is a linear scaling of \(\theta = (\mu , \sigma)\) by a factor \(\gamma\) calculated to minimize the statistical distance between the normal distributions calculated by software and the available experimental ones (see Methods). However, this operation is not sufficient to meet the technology requirements: the desired domain exceeds the available experimental one for both filamentary memristors ( \(\Gamma_{memristor}\) ) and phase- change memories ( \(\Gamma_{PCM}\) ). To compel the learned normal distributions to match with the hardware experimental electrical characteristics, we imposed that \(\theta\) belong to the experimental \(\Gamma\) domain by adding the "technological loss" term to the loss function: + +<|ref|>equation<|/ref|><|det|>[[130, 819, 907, 837]]<|/det|> +\[Loss = Loss_{VI} - log(U_{\Gamma}(\theta)), \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[88, 844, 909, 922]]<|/det|> +where \(U_{\Gamma}(\theta)\) is a uniform distribution over the \(\Gamma = \Gamma_{memristor} / \gamma\) or the \(\Gamma = \Gamma_{PCM} / \gamma\) domain. Fig. 4b illustrates the effectiveness of the proposed hardware- calibrated training method: the normal distributions obtained by software simulations perfectly map on both phase- change memories and filamentary memristors experimental values. We demonstrated that by taking hardware physics into account while developing the training algorithm, it is possible to make variational inference a technologically plausible algorithm. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[108, 258, 890, 587]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 597, 911, 734]]<|/det|> +
Figure 3. Filamentary memristor and phase-change memories as physical random variables with normal distribution. a Probability densities of 2,048 filamentary memristors programmed with eight different programming current values. b Domain of the Gaussian distributions experimentally achieved exploiting different programming conditions for filamentary memristors (blue) and phase-change memories (green). Triangles represent one-shot programming, dots represent iterative programming and the cross represents the low conductance state. c Probability densities of 2,048 phase-change memories programmed with seven different programming current values. d Schematic of the proposed synaptic circuit. Each sample of a Bayesian probabilistic weight is stored as the difference between the conductance values of two adjacent memory cells. e Domain of the normal distributions (Γ) that can be experimentally obtained exploiting the circuit in d by storing samples on two memory cells.
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[113, 264, 884, 640]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 653, 909, 730]]<|/det|> +
Figure 4. Domains of normal distribution obtained with classical Variational Inference and the proposed technologically plausible method. a Domain of normal distributions \(\theta = (\mu , \sigma)\) obtained after training with the classical VI method and mapping the software values to the conductance range achievable with filamentary memristors (blue) and phase-change memories (green). b Domain of normal distributions \(\theta = (\mu , \sigma)\) obtained after training with the proposed method calibrated on filamentary memristors (blue) and phase-change memory experimental data (green).
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 80, 373, 95]]<|/det|> +## Experimental uncertainty estimation + +<|ref|>text<|/ref|><|det|>[[88, 97, 910, 293]]<|/det|> +To validate our approach, we programmed a Bayesian neural network, trained to recognize arrhythmia, onto a collection of the filamentary memristor dies (Fig. 2f). We classified ECG diagnosis beats using a two- layer Bayesian neural network featuring 32 inputs, 16 hidden neurons in the first layer, and nine output neurons in the second layer. We trained the Bayesian neural network, using the filamentary memristor technological loss, on nine classes: healthy beat and eight types of arrhythmias. During testing, we added a tenth class corresponding to a non- previously seen type of arrhythmia. Following the architecture presented in Fig. 1, we programmed \(M = 50\) independent realizations of model parameter vector \(\theta\) , representing the learned posterior \(q(\Omega |\theta)\) , i.e., we transferred each model realization into an array of conductance values. Each realization can fit in 1.5 dies presented in Fig. 2f (one die for the first layer of the neural network, and a half die for the second layer, see Methods), and described in detail in Supplementary Note 5; therefore, we needed to program a total of 75 dies. Multiply- and- accumulate operations were performed directly in memory using Ohm's and Kirchoff's law (see Methods and Supplementary Note 5). Activation functions were calculated in software. The array performed all the multiply and accumulate operations needed to classify 1000 beats in the test data set. Supplementary Note 6 recapitulates the different steps of our experiment, from training to inference. + +<|ref|>text<|/ref|><|det|>[[88, 295, 910, 536]]<|/det|> +Fig. 5 presents the electrical characterization results of the memristor- based Bayesian neural network. To visualize the input data, we used the t- distributed stochastic neighbor embedding (t- SNE) statistical method (Figs. 5a- b). This visualization technique represents each high- dimensional input data by a point in a two- dimensional space, in the same way that similar data correspond to nearby points and distant points represent dissimilar data. Fig. 5a illustrates the two- dimensional projections of the input data used during inference on the test dataset. The data belonging to a given class (disease) display a "cluster". The unseen diseases (i.e., beats that do not belong to a class learned in the training phase) are the red points. Fig. 5b uses the same representation, where the colors represent data points correctly (blue) or incorrectly (orange) classified by our experiment, while the unseen disease data points are plotted in black. Our experiment recognizes \(75\%\) of the data points correctly. Most errors concern points that lie at the border between several clusters in the t- SNE plot, suggesting that they might be ambivalent (high aleatoric uncertainty cases). To investigate this idea further, Fig. 5c shows the measured probability density distributions of the measured aleatoric uncertainty, which provides a measure of the confidence of network prediction. The different colors represent correct predictions (blue), incorrect predictions (orange), and unseen data (red). The aleatoric uncertainty is lower than 0.5 for \(62\%\) of all correctly classified data points, while it is higher than 0.5 for \(97\%\) of all incorrectly classified data points and unseen disease data points. This result means that our experiment correctly determined as uncertain all of its errors and the unseen disease. It also flagged as uncertain some of its correct predictions, which is expected, as some of them might be ambivalent cases. + +<|ref|>text<|/ref|><|det|>[[88, 538, 909, 599]]<|/det|> +The situation is quite different when we look at the measured epistemic uncertainty (Fig. 5d). \(97\%\) of all correctly and incorrectly classified data points have an epistemic uncertainty lower than 0.5. Conversely, \(98\%\) of the unseen disease data points have epistemic uncertainty higher than 0.5. These results mean that experiments can differentiate ambivalence between classes from the presentation of new unknown inputs. + +<|ref|>text<|/ref|><|det|>[[88, 600, 909, 661]]<|/det|> +These results come in sharp contrast with those of a simulated conventional neural network with the same architecture. This type of neural network, by construction, has no epistemic uncertainty, and the aleatoric uncertainty tends to be extremely low whatever the input (Fig. 5e). This overconfidence is due to the small size of our dataset, making conventional neural networks particularly prone to overfitting. + +<|ref|>text<|/ref|><|det|>[[88, 662, 910, 843]]<|/det|> +To push the interpretation of our experimental results further and make an in- depth assessment of the capability of our experiment to evaluate uncertainty, we used receiver operating characteristic (ROC) curves, a widely used metric for diagnostic ability, obtained by plotting the true positive rate as a function of the false positive rate for various threshold settings. A perfect classifier would yield the (0, 1) point, i.e., an area under the curve (AUC) of one, corresponding to no false negatives and no false positives. The ROC curve of a random classifier approaches the diagonal line, i.e., an area under the curve of 0.5. Fig. 5f shows the ROC curve corresponding to the differentiation between correct predictions and incorrect predictions, based on aleatoric uncertainty, for our experiment, a purely software version of the Bayesian neural network programmed in our experiment, and a conventional neural network with the same architecture (see Methods). Our experiment leads to a ROC curve close to the software Bayesian neural network, showing the high quality of our transfer. It even slightly outperforms the software version (area under the curve of 0.91 vs. 0.90). The conventional neural network has an inferior area under the curve (0.79), highlighting again the overconfidence of such networks. We should remark that an area under the curve of one is impossible for this graph, because predictions with very low aleatoric certainty are sometimes correct. + +<|ref|>text<|/ref|><|det|>[[88, 845, 909, 920]]<|/det|> +Fig. 5g shows the ROC curve corresponding to the differentiation between known and unknown data, based on epistemic uncertainty (see Methods). Our experiment performs particularly well, with an area under the curve of 0.99, close to the perfect value of one, and which slightly outperforms the one obtained by the software (0.95) Bayesian neural network. This enhanced capability to recognize unknown data comes at the price of a small decrease in the general accuracy of the network (75% vs. 79%). We should point out that these differences between software and hardware are most probably not significant, as our test + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[90, 80, 908, 141]]<|/det|> +set features only 70 unknown data points (out of a dataset of 1000 points). We repeated the software experiment ten times and the area under the curve varied between 0.89 and 1. By contrast, the conventional neural network exhibited no capacity to recognize unknown data and had the diagonal ROC curve of a random classifier, with an area under the curve of 0.5. Table. 1 summarizes all these results. + +<|ref|>text<|/ref|><|det|>[[90, 142, 908, 232]]<|/det|> +A drawback of our approach with regard to conventional neural networks is that we need several \((M)\) versions of the neural networks. This number, however, does not necessarily need to be high. Fig. 5h shows its effect on the Bayesian neural network accuracy and on its capability to evaluate uncertainty, measured by the area under the curve of the two ROC curves mentioned above, obtained using aleatoric and epistemic uncertainty. The accuracy and aleatoric area under the curve approach their saturation values with ten neural networks. The epistemic area under the curve takes a higher number of implementations to converge; however, with ten neural networks, it reaches 0.96, close to its maximum value (0.99). + +<|ref|>sub_title<|/ref|><|det|>[[90, 250, 197, 266]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[90, 273, 908, 409]]<|/det|> +This work demonstrates experimentally a simple and energy- efficient realization of a Bayesian neural network by directly storing the probabilistic weights into resistive memory- based crossbar arrays. The device variability in both filamentary- based memristors and phase- change memories is used to store physical random variables that sample analog conductance values from normal distributions with re- configurable mean and standard deviation. The Bayesian neural networks are trained following a special variational inference approach, incorporating a "technological loss" to overcome the hardware limitations linked to the device physics. We implemented a whole network using a collection of filamentary memristor arrays allowing in- memory computing. The resulting Bayesian neural network matches software simulations in terms of accuracy, and in terms of aleatoric and epistemic uncertainty evaluations, as evidenced by ROC curves for the identification of misclassified heartbeats and unknown data heartbeats. + +<|ref|>text<|/ref|><|det|>[[90, 409, 908, 576]]<|/det|> +The dies that we used for phase- change memory characterization (Fig. 2c) are conventional memory arrays that do not allow in- memory computing and cannot be implemented as a full in- memory Bayesian neural network, unlike what we achieved for filamentary memristors. Therefore, we used our extensive statistical measurements of phase change memories (Fig. 4) to simulate such a network, using the simulator validated in Supplementary Note 3. The results are presented in Supplementary Note 4 and listed in Table 1; they suggest that the phase change memory network would function almost equivalently to the filamentary memristor- one, with only a slight reduction in terms of accuracy and uncertainty evaluation (expressed by the area under the two ROC curves). Fig. 4 shows that the mean value/standard deviation space that can be programmed on phase change memories is more skewed than that of filamentary memristors. (Indeed, it is impossible to program synapses with low mean value and high standard deviation on phase change memories). The fact that Bayesian neural networks based on the two memory technologies still achieve almost matching performance and uncertainty evaluations demonstrates the power of the "technological loss" term to correct for the constraints of technology. + +<|ref|>text<|/ref|><|det|>[[90, 577, 908, 653]]<|/det|> +The most important limitation of our approach is that it requires the use of multiple devices per synapse to represent a distribution of its synaptic weight. The results of Fig. 5h show that the number of devices per synapse does not need to be large. Bayesian neural networks excel in relatively small- data regimes, where strong uncertainty is present: they are not large networks, making device overhead bearable. Currently- developed resistive memories integrated in three dimensions may be particularly suitable to our architecture, which features multiple devices per synapse40. + +<|ref|>text<|/ref|><|det|>[[90, 653, 908, 713]]<|/det|> +We estimated the energy consumption of a final in- memory Bayesian neural network, based on measurements from a state- of- the- art 22- nanometer platform3. We found a cost of 270 nanojoules per inference, 800 times smaller than on a modern GPU fabricated in a 12- nanometer process (see Methods). This efficiency suggests that Bayesian neural networks can be used at the edge in extremely energy- constrained systems, such as medical devices, where reliable decisions are needed. + +<|ref|>sub_title<|/ref|><|det|>[[90, 731, 172, 747]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[90, 754, 538, 769]]<|/det|> +## Filamentary and Phase Change Memory technology and circuits + +<|ref|>text<|/ref|><|det|>[[90, 770, 908, 891]]<|/det|> +The circuits described in the Results section were fabricated using a low- power foundry 130- nanometers process with four metal layers. Both phase- change memories and filamentary memristors were fabricated on tungsten vias in metal layer four. The filamentary memristors consist of a 5- nanometer thick metallic bottom electrode, a 5- nanometer thick \(\mathrm{HfO_x}\) active layer deposited by atomic layer deposition, and a 10- nanometer thick Ti top electrode. The memory element is fabricated as a mesa structure with a 200- nanometer diameter. The phase- change memory architecture is characterized by a strip of chalcogenide material lying on top of a TiN heater element, with a thickness of five nanometers and a width of 100 nanometers. The chalcogenide layer is a germanium- antimony- tellurium alloy deposited by sputtering deposition and is 50- nanometer thick. A fifth layer of metal is deposited on top of both phase- change memories and filamentary memristors. + +<|ref|>text<|/ref|><|det|>[[90, 892, 907, 922]]<|/det|> +Two different integrated circuits were used in this article, one integrating filamentary- based memristors and the other phase- change memories (Fig. 2). In both architectures, each memory cell is accessed by a transistor, giving rise to a one + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[110, 100, 888, 521]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[88, 535, 911, 762]]<|/det|> +
Figure 5. Measurements of the fabricated memristor-based Bayesian neural network. a tSNE visualization of input data, different colors representing different classes (diseases). Nearby points correspond to similar data and distant points to dissimilar data. b tSNE visualization of experimental data classification. The different colors represent points correctly or incorrectly predicted and unseen data. c Experimental probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases. d Experimental probability density distribution of the epistemic uncertainty for correct predictions, incorrect predictions and unseen diseases. e Simulated probability density distribution of the aleatoric uncertainty for correct predictions, incorrect predictions and unseen diseases for a conventional neural network with the same architecture and using float32 encoding for the synapses. f ROC curve corresponding to the differentiation between correct prediction and incorrect prediction, based on aleatoric uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). g ROC curve corresponding to the differentiation between known and unknown data, based on epistemic uncertainty measured on the proposed memristor-based Bayesian neural network (green) and simulated (weights stored as float32 real numbers) for Bayesian (black) and conventional neural network with the same architecture (red). h Measured accuracy, epistemic uncertainty, and aleatoric uncertainty (calculated as the area of the ROC curves) as a function of the number of filamentary memristor devices per synapse.
+ +<|ref|>table<|/ref|><|det|>[[113, 799, 885, 875]]<|/det|> +<|ref|>table_caption<|/ref|><|det|>[[247, 888, 746, 904]]<|/det|> +Table 1. Comparison of accuracy and uncertainty prediction performances. + +
Conventional ANN (float 32)Bayesian (float 32)Bayesian Hardware (filamentary memristor experimental)Bayesian Hardware (phase-change memory simulation)
Accuracy classification80%79%75%73%
Prediction confidence (aleatoric) [AUC]0.790.900.910.85
Anomaly detection (epistemic) [AUC]0.50.950.990.96
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[57, 78, 909, 171]]<|/det|> +transistor- one- resistor (1T1R) unit cell. The transistor, used as a selector, was essential to control the programming current allowing multi- level programming of filamentary memristors. The phase- change memory chip was an array of 16,384 1T1R structures, only individually accessible. The filamentary- based memristor chip was an array of 1,024 1T1R cells arranged in a \(32\times 32\) configuration. This array enabled the selection of multiple memory points capable of performing parallel multiply and accumulate operations. Digital drivers were used to select multiple cells in parallel controlling the word lines (WLs), source lines (SLs), and bit lines (BLs). This array is described in detail in Supplementary Note 5. + +<|ref|>sub_title<|/ref|><|det|>[[90, 180, 247, 194]]<|/det|> +## Iterative programming + +<|ref|>text<|/ref|><|det|>[[90, 195, 907, 256]]<|/det|> +The iterative programming methods adopted for filamentary- based and phase- change memories are different. For filamentary memristors, each device is re- programmed multiple times, with the same conditions, until its conductance reaches the target value (Algorithm 1). For the phase- change memories the programming voltage is increased or decreased at each cycle depending on the conductance value obtained in the previous cycle (Algorithm 2). + +<|ref|>text<|/ref|><|det|>[[90, 272, 468, 301]]<|/det|> +Algorithm 1 Iterative programming for filamentary memristors + +<|ref|>text<|/ref|><|det|>[[91, 303, 425, 530]]<|/det|> +1: \(G_{max}:\) Target conductance max +2: \(G_{min}:\) Target conductance min +3: \(I_{cc}:\) Compliance current for target distribution +4: \(i_{max}:\) Maximum number of iteration +5: \(G:\) filamentary memristor conductance +6: \(G\gets RESET\) 7: \(i\gets 0\) 8: while \(i< i_{max}\) . 9: \(G_{0}\gets SET(I_{cc})\) 10: \(i\gets i + 1\) 11: if \(G_{min}< G< G_{max}\) 12: end 13: else: 14: \(G_{0}\gets RESET\) 15: end + +<|ref|>text<|/ref|><|det|>[[528, 272, 907, 303]]<|/det|> +Algorithm 2 Iterative programming for phase change memories + +<|ref|>text<|/ref|><|det|>[[533, 304, 816, 530]]<|/det|> +1: \(G_{max}:\) Target conductance max +2: \(G_{min}:\) Target conductance min +3: \(V_{s}:\) Applied voltage +4: \(V_{max}:\) Maximum voltage +5: \(\delta V:\) Voltage increment +6: \(G:\) phase change memory conductance +7: \(G\gets RESET\) 8: \(V_{s}\gets V_{init}\) 9: while \(V_{s}< V_{max}\) and \(G< G_{min}\) . 10: \(G\gets SET(V_{s})\) 11: \(V_{s}\gets V_{s} + \delta V\) 12: while \(V_{s}< V_{max}\) and \(G > G_{max}\) . 13: \(G\gets RESET(V_{s})\) 14: \(V_{s}\gets V_{s} - \delta V\) 15: end + +<|ref|>text<|/ref|><|det|>[[57, 545, 909, 666]]<|/det|> +Before the filamentary based memristors chip can be used, it is necessary to form all the devices. The forming operation consist in the following conditions: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} = 1.6 \mathrm{V}\) , \(V_{bl} \in [1.6,4] \mathrm{V}\) . The standard SET conditions are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [1.4,2.2] \mathrm{V}\) , \(V_{bl} = 1.8 \mathrm{V}\) . The standard RESET conditions used are as follows: \(V_{sl} = 2.6 \mathrm{V}\) , \(V_{wl} = 4.8 \mathrm{V}\) , \(V_{bl} = 0 \mathrm{V}\) . The off- chip generated voltage programming pulses have a pulse width of \(1 \mu \mathrm{s}\) for the SET and \(100 \mathrm{ns}\) for the RESET. For the phase change memory chip, the standard SET conditions are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [2,3] \mathrm{V}\) and \(V_{bl} = 4 \mathrm{V}\) . The standard RESET conditions used are as follows: \(V_{sl} = 0 \mathrm{V}\) , \(V_{wl} \in [0.9,4] \mathrm{V}\) and \(V_{bl} = 4.8 \mathrm{V}\) . The off- chip generated voltage programming pulses have a pulse width of \(300 \mathrm{ns}\) and a rise time of \(20 \mathrm{ns}\) . The fall time is \(1500 \mathrm{ns}\) for the SET and \(20 \mathrm{ns}\) for the RESET. + +<|ref|>sub_title<|/ref|><|det|>[[90, 676, 331, 691]]<|/det|> +## Weight and conductance mapping + +<|ref|>text<|/ref|><|det|>[[90, 692, 907, 752]]<|/det|> +The mapping between the mean and standard deviation of the normal distributions obtained after software training, \(\theta_{s} = (\mu_{s}, \sigma_{s})\) , and the corresponding experimental conductance distributions \(\theta_{e} = (\mu_{e}, \sigma_{e})\) in microsiemens is a critical step. The normal distributions chosen by the training algorithm and reported in Fig. 4 were mapped to conductance values in microsiemens according to: + +<|ref|>equation<|/ref|><|det|>[[130, 780, 907, 797]]<|/det|> +\[\mu_{e} = \gamma \cdot \mu_{s} \qquad \sigma_{e} = \gamma \cdot \sigma_{s}. \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[57, 807, 907, 838]]<|/det|> +The \(\gamma\) scaling factor has been calculated by minimizing the Kullback- Leibler divergence between the experimental and simulated normal distributions: + +<|ref|>equation<|/ref|><|det|>[[130, 864, 907, 896]]<|/det|> +\[\gamma = \underset {\gamma \in \mathbb{R}}{\mathrm{argmin}}\sum_{j\in [1,S]}\underset {i\in [1,E]}{\mathrm{min}}KL(\theta_{e_{i}},\gamma \times \theta_{s_{j}}), \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[57, 905, 909, 922]]<|/det|> +where \(S\) is the number of the software normal distributions and E is the number of available experimental normal distributions. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[58, 78, 494, 95]]<|/det|> +## Training using Bayes by Backprop and technological loss + +<|ref|>text<|/ref|><|det|>[[60, 94, 905, 127]]<|/det|> +The training of a Bayesian neural network consists of computing the most likely models (i.e. the posterior distribution, \(p(\Omega |D))\) underlying the training dataset, \(D\) , and the prior belief, \(p(\Omega)\) : + +<|ref|>equation<|/ref|><|det|>[[130, 148, 907, 185]]<|/det|> +\[p(\Omega |D) = \frac{p(D|\Omega)p(\Omega)}{p(D)}. \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[58, 193, 909, 318]]<|/det|> +Here \(\Omega\) represents the neural network parameters, \(p(D|\Omega)\) is the likelihood, and \(p(D)\) is the evidence. This equation is unfortunately intractable. Variational Inference approximates the posterior distribution, \(p(\Omega |D)\) , with a simpler variational distribution, \(q(\Omega |\theta)\) , which structure is easier to evaluate39. Typically the variational distributions are normal distributions, where the variational parameters \(\theta = (\mu ,\sigma)\) represent their mean and standard deviation. The approximation of the \(\theta\) parameters, \(\theta^{*}\) , are calculated minimizing the Kullback- Leibler (KL) divergence between the variational distribution, \(q(\Omega |\theta)\) , and the posterior, \(p(\Omega |D)\) , as shown in Eq. 6. The KL divergence is a measure of the similarity between the two distributions. The calculation of the \(\theta^{*} \in \mathbb{R}\) parameters is achieved by backpropagation15. This combination of variational inference and backpropagation is called Bayes by Backprop and has been proved to be efficient for complex applications10. It identifies + +<|ref|>equation<|/ref|><|det|>[[130, 343, 907, 370]]<|/det|> +\[\theta^{*} = \underset {\theta \in \mathbb{R}}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)]). \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[58, 379, 909, 457]]<|/det|> +As illustrated in Fig. 4, resistive memories cannot implement all possible normal distributions, no matter the chosen technology flavour. The search of \(\theta^{*}\) should be limited inside \(\Gamma\) , where \(\Gamma\) represents the ensemble of experimental normal distributions that can be built with a given technology. To impose that \(\theta\) belongs to \(\Gamma\) , a "technological loss" term has been added to Eq. 6. The "technological loss" term is defined as \(- log(U_{\Gamma}(\theta))\) , where \(U_{\Gamma}\) is a uniform distribution over the experimental \(\Gamma\) domain, and Eq. 6 becomes: + +<|ref|>equation<|/ref|><|det|>[[130, 483, 907, 509]]<|/det|> +\[\theta^{*} = \underset {\theta \in \Gamma}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)]) = \underset {\theta \in \mathbb{R}}{\arg \min}(KL[q(\Omega |\theta)||p(\Omega |D)] - log(U_{\Gamma}(\theta)). \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[110, 519, 488, 536]]<|/det|> +The cost function resulting from Eq. 7 can be denoted as: + +<|ref|>equation<|/ref|><|det|>[[130, 563, 907, 583]]<|/det|> +\[F(D,\theta) = KL[q(\Omega |\theta)||p(\Omega)] - \mathbb{E}_{q(\Omega |\theta)}[log(p(D|\Omega))] - log(U_{\Gamma}(\theta)) \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[58, 592, 645, 609]]<|/det|> +Eq. 8 can be solved with classical Bayes by Backprop, and it ensures that the \(\theta^{*} \in \Gamma\) . + +<|ref|>sub_title<|/ref|><|det|>[[58, 617, 426, 633]]<|/det|> +## Experimental setup for arrhythmia classification + +<|ref|>text<|/ref|><|det|>[[58, 632, 909, 784]]<|/det|> +The considered Bayesian neural network featured 32 inputs, 16 hidden neurons in the first layer and nine output neurons, corresponding to the nine different diseases (classes), in the second layer. Since we use conductance subtraction between two filamentary memristors to store one weight, our \(32 \times 32\) crossbar array could take 32 inputs and produce 16 outputs. To realize one sample of our two- layer neural network one and a half crossbar arrays are required ( \(32 \times 32\) cells for the first layer and \(16 \times 18\) cells for the second one). A Bayesian neural network is the collection of several \((M)\) samples, so when using \(32 \times 32\) crossbar arrays, \(1.5M\) arrays are needed. We fully characterized 15 crossbars arrays to implement a Bayesian neural network with \(M = 10\) samples. To reproduce a Bayesian neural network with more than \(M = 10\) samples, we recycled the 15 crossbar arrays exploiting the fact that the cycle- to- cycle and device- to- device variability are similar in filamentary memristors21. Therefore, by reprogramming the 15 arrays 5 times, which is equivalent to using 75 arrays. We obtain a Bayesian Neural Network with \(M = 50\) samples. + +<|ref|>text<|/ref|><|det|>[[58, 784, 909, 920]]<|/det|> +The input data are ECG recordings32. A single heartbeat is a 700 ms recording, and it is converted into 32 features through a Fast- Fourier Transform (FFT). The 32 extracted features are the input of the \(M\) samples of the Bayesian neural network. Since the digital drivers generate only a single read voltage level, \(V_{read}\) (see Supplementary Note 5), each feature is converted into three- bit binary values, \((X_{j}\) with \(j = 0, \ldots , 2)\) . The three bits are applied sequentially to the input of the first layer of each \(s\) sample, with \(s = 1, \ldots , M\) . Each input voltage vector \(X_{j}\) is applied on the bit lines of the \(32 \times 32\) crossbar array to generate output current vector \((V_{read}\) , is applied to the selected bit lines, which correspond to an input one, the unselected bit lines are floating, which correspond to an input zero). The measured output current at the source lines, is the dot product operation through the first layer, \(W_{s} \cdot X_{j}\) , where \(W_{s}\) are the conductance values of a given sample (model realization) \(s\) . The output current for a given three- bit binary input is + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[130, 100, 907, 132]]<|/det|> +\[I_{i,s} = \frac{W_{s}\cdot X_{0} + 2\times W_{s}\cdot X_{1} + 4\times W_{s}\cdot X_{2}}{7}. \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[58, 140, 684, 155]]<|/det|> +Using this experimental values, we calculate the activation functions of the hidden neurons + +<|ref|>equation<|/ref|><|det|>[[130, 177, 907, 214]]<|/det|> +\[a_{i,s} = \frac{I_{i,s}^{+} - I_{i,s}^{-}}{\gamma\cdot V_{read}}, \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[58, 222, 907, 283]]<|/det|> +where \(\gamma\) is the scaling factor calculated with Eq. 4. Each activation function is converted to three- bit binary values. This operation is equivalent to the calculation of a clipped rectified linear unit (ReLu) activation function. The same method is applied to the second layer, in which the calculated activations are the new input. The probability that the input data \(X\) belongs to a given output class \(c\) for a given sample \(W_{s}\) using a softmax function is + +<|ref|>equation<|/ref|><|det|>[[130, 301, 907, 355]]<|/det|> +\[p(y = c|X,W_{s}) = \frac{e^{a_{c,s}}}{\sum_{j = 1}^{N}e^{a_{j,s}}}. \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[58, 363, 907, 409]]<|/det|> +The disease classification (i.e., the probability that the input data belong to a specific class of disease) is the average of the probability values calculated with Eq. 11 over the number of samples. The predicted class is calculated as the argmax of the disease classifications. The aleatoric and epistemic uncertainty are calculated with Eqs. 12, 13 and 14. + +<|ref|>sub_title<|/ref|><|det|>[[58, 418, 255, 432]]<|/det|> +## Uncertainty calculation + +<|ref|>text<|/ref|><|det|>[[58, 433, 907, 478]]<|/det|> +Unlike conventional artificial neural networks, where the output values for predictions are point estimates, Bayesian neural networks provide predictive distributions. The total uncertainty in the prediction, i.e., the predictive uncertainty, can be calculated based on the softmax of the predictive distributions calculated according to Eq. 11: + +<|ref|>equation<|/ref|><|det|>[[130, 498, 907, 541]]<|/det|> +\[U_{p} = -\sum_{c = 1}^{N}\left(\frac{1}{M}\sum_{s = 1}^{M}p(y = c|X,W_{s})\right)log\left(\frac{1}{M}\sum_{s = 1}^{M}p(y = c |X,W_{s})\right). \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[58, 550, 670, 565]]<|/det|> +The predictive uncertainty (Eq.12) is the sum of epistemic and aleatoric uncertainties + +<|ref|>equation<|/ref|><|det|>[[130, 590, 907, 607]]<|/det|> +\[U_{p} = U_{a} + U_{e}. \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[58, 618, 907, 678]]<|/det|> +Decomposing the predictive uncertainty is important, as epistemic and aleatoric uncertainties give us different information. High epistemic uncertainty suggests that the input data is an outlier relative to the training data set. More training data near can therefore reduce epistemic uncertainty, but does not help aleatoric uncertainty. Aleatoric uncertainty is uncertainty in data, to reduce it more refined input data are required (e.g., more powerful sensors). The aleatoric uncertainty can be obtained as: + +<|ref|>equation<|/ref|><|det|>[[130, 700, 907, 737]]<|/det|> +\[U_{a} = -\frac{1}{M}\sum_{s = 1}^{M}\sum_{c = 1}^{N}p(y = c|X,W_{s})logp(y = c|X,W_{s}). \quad (14)\] + +<|ref|>sub_title<|/ref|><|det|>[[58, 748, 310, 762]]<|/det|> +## Energy consumption estimates + +<|ref|>text<|/ref|><|det|>[[58, 763, 907, 791]]<|/det|> +To estimate the energy consumption of the Bayesian neural network we first calculated the number of dot product operations for one inference: + +<|ref|>equation<|/ref|><|det|>[[130, 818, 907, 835]]<|/det|> +\[O p e r a t i o n s = 2\cdot I_{l}\cdot H_{l} + 2\cdot H_{l}\cdot O_{l}. \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[58, 845, 907, 920]]<|/det|> +Here \(I_{l}\) is the input length, \(H_{l}\) is the hidden layer length, and \(O_{l}\) is the output length. The factor two is due to fact that each sample of a Bayesian probabilistic weight is stored as the difference between the conductance values stored in two memory cells. One inference costs 1344 operations. The cost of a single analog Multiply- and- Accumulate (MAC) operation in a resistive memory- based analog in- memory computing circuit depends on the input and output size and on the weight precision. The cost of an analog MAC operation using \(22\mathrm{nm}\) CMOS technology with \(4\mathrm{b}\) input signal, \(4\mathrm{b}\) weight, and and \(11\mathrm{b}\) output is evaluated + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 910, 125]]<|/det|> +27 nJ according to ref.3. We used this value to estimate the dot product operation in our circuit, which has 3 b input signal, 4 b weight, and 8 b output. Assuming that a Bayesian inference requires ten single inference operations the estimated energy consumption for our circuit is \(270~\mathrm{nJ}\) . + +<|ref|>text<|/ref|><|det|>[[90, 125, 910, 202]]<|/det|> +To gain a perspective on the energy efficiency of the proposed approach compared to conventional hardware, we benchmarked this figure to the energy required for running the same Bayesian neural network on a state- of- the- art Tesla V100 GPU, which uses a power consumption of 43 W executing the code provided with the Bayes by backprop paper15. The execution time for ten inference operations is \(5\mu \mathrm{s}\) . Based on these estimations, the proposed resistive memory- based analog circuit achieves a reduction of about a factor 800 in energy consumption relative to a conventional digital GPU. + +<|ref|>sub_title<|/ref|><|det|>[[91, 217, 240, 235]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[91, 241, 411, 256]]<|/det|> +All the measured data are available upon request. + +<|ref|>sub_title<|/ref|><|det|>[[91, 272, 246, 290]]<|/det|> +## Code availability + +<|ref|>text<|/ref|><|det|>[[90, 295, 664, 311]]<|/det|> +All software programs used in the presentation of the Article are available upon request. + +<|ref|>sub_title<|/ref|><|det|>[[91, 327, 198, 344]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[90, 350, 910, 384]]<|/det|> +1. Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521, 61 (2015). + +<|ref|>text<|/ref|><|det|>[[90, 386, 910, 418]]<|/det|> +2. Ambrogio, S. et al. Equivalent-accuracy accelerated neural-network training using analogue memory. Nature 558, 60 (2018). + +<|ref|>text<|/ref|><|det|>[[90, 420, 910, 452]]<|/det|> +3. Xue, C.-X. et al. A cmos-integrated compute-in-memory macro based on resistive random-access memory for ai edge devices. Nat. Electron. 4, 81-90 (2021). + +<|ref|>text<|/ref|><|det|>[[90, 455, 864, 472]]<|/det|> +4. Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641-646 (2020). + +<|ref|>text<|/ref|><|det|>[[90, 475, 910, 522]]<|/det|> +5. Wan, W. et al. 33.1 a 74 tmacs/w cmos-ram neurosynaptic core with dynamically reconfigurable dataflow and in-situ transposable weights for probabilistic graphical models. In 2020 IEEE International Solid-State Circuits Conference- (ISSCC), 498-500 (IEEE, 2020). + +<|ref|>text<|/ref|><|det|>[[90, 525, 910, 557]]<|/det|> +6. Jung, S. et al. A crossbar array of magnetoresistive memory devices for in-memory computing. Nature 601, 211-216 (2022). + +<|ref|>text<|/ref|><|det|>[[90, 560, 910, 592]]<|/det|> +7. Khaddam-Aljameh, R. et al. Hermes-core—a 1.59-tops/mm 2 pcm on 14-nm cmos in-memory compute core using 300-ps/lsb linearized cco-based adcs. IEEE J. Solid-State Circuits 57, 1027-1038 (2022). + +<|ref|>text<|/ref|><|det|>[[90, 595, 910, 627]]<|/det|> +8. Kabir, H. D., Khosravi, A., Hosen, M. A. & Nahavandi, S. Neural network-based uncertainty quantification: A survey of methodologies and applications. IEEE access 6, 36218-36234 (2018). + +<|ref|>text<|/ref|><|det|>[[90, 630, 910, 662]]<|/det|> +9. Jospin, L. V., Buntine, W., Boussaid, F., Laga, H. & Bennamoun, M. Hands-on bayesian neural networks—a tutorial for deep learning users. arXiv preprint arXiv:2007.06823 (2020). + +<|ref|>text<|/ref|><|det|>[[90, 665, 910, 697]]<|/det|> +10. Kendall, A. & Gal, Y. What uncertainties do we need in bayesian deep learning for computer vision? Adv. neural information processing systems 30 (2017). + +<|ref|>text<|/ref|><|det|>[[90, 700, 770, 717]]<|/det|> +11. Szegedy, C. et al. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199 (2013). + +<|ref|>text<|/ref|><|det|>[[90, 720, 825, 737]]<|/det|> +12. Goodfellow, I., Bengio, Y., Courville, A. & Bengio, Y. Deep learning, vol. 1 (MIT press Cambridge, 2016). + +<|ref|>text<|/ref|><|det|>[[90, 740, 836, 757]]<|/det|> +13. Der Kiureghian, A. & Ditlevsen, O. Aleatory or epistemic? does it matter? Struct. safety 31, 105-112 (2009). + +<|ref|>text<|/ref|><|det|>[[90, 760, 777, 776]]<|/det|> +14. Ghahramani, Z. Probabilistic machine learning and artificial intelligence. Nature 521, 452-9 (2015). + +<|ref|>text<|/ref|><|det|>[[90, 779, 910, 811]]<|/det|> +15. Blundell, C., Cornebise, J., Kavukcuoglu, K. & Wierstra, D. Weight uncertainty in neural network. In International conference on machine learning, 1613-1622 (PMLR, 2015). + +<|ref|>text<|/ref|><|det|>[[90, 814, 810, 831]]<|/det|> +16. Neal, R. M. Bayesian learning for neural networks, vol. 118 (Springer Science & Business Media, 2012). + +<|ref|>text<|/ref|><|det|>[[90, 835, 562, 851]]<|/det|> +17. Gal. Uncertainty in deep learning. PhD thesis, Univ. Camb. (2016). + +<|ref|>text<|/ref|><|det|>[[90, 854, 910, 886]]<|/det|> +18. Dalgaty, T. et al. In situ learning using intrinsic memristor variability via markov chain monte carlo sampling. Nat. Electron. 4, 151-161 (2021). + +<|ref|>text<|/ref|><|det|>[[90, 889, 910, 921]]<|/det|> +19. Joshi, V. et al. Accurate deep neural network inference using computational phase-change memory. Nat. communications 11, 1-13 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[55, 78, 911, 111]]<|/det|> +20. Tsai, H. et al. Inference of long-short term memory networks at software-equivalent accuracy using 2.5 m analog phase change memory devices. In 2019 Symposium on VLSI Technology, T82-T83 (IEEE, 2019). + +<|ref|>text<|/ref|><|det|>[[55, 116, 911, 149]]<|/det|> +21. Dalgaty, T., Esmanhotto, E., Castellani, N., Querlioz, D. & Vianello, E. Ex situ transfer of bayesian neural networks to resistive memory-based inference hardware. Adv. Intell. Syst. 3, 2000103 (2021). + +<|ref|>text<|/ref|><|det|>[[55, 154, 911, 186]]<|/det|> +22. Esmanhotto, E. et al. High-density 3d monolithically integrated multiple 1t1r multi-level-cell for neural networks. In 2020 IEEE International Electron Devices Meeting (IEDM), 36-5 (IEEE, 2020). + +<|ref|>text<|/ref|><|det|>[[55, 191, 910, 208]]<|/det|> +23. Fortunato, M., Blundell, C. & Vinyals, O. Bayesian recurrent neural networks. arXiv preprint arXiv:1704.02798 (2017). + +<|ref|>text<|/ref|><|det|>[[55, 213, 911, 246]]<|/det|> +24. Sebastian, A. et al. Two-dimensional materials-based probabilistic synapses and reconfigurable neurons for measuring inference uncertainty using bayesian neural networks. Nat. communications 13, 1-10 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 250, 911, 283]]<|/det|> +25. Liu, S. et al. Bayesian neural networks using magnetic tunnel junction-based probabilistic in-memory computing. Front. Nanotechnol. 78 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 288, 911, 320]]<|/det|> +26. Faria, R., Camsari, K. Y. & Datta, S. Implementing bayesian networks with embedded stochastic mram. AIP Adv. 8, 045101 (2018). + +<|ref|>text<|/ref|><|det|>[[55, 325, 911, 358]]<|/det|> +27. Vodenicarevic, D. et al. Low-energy truly random number generation with superparamagnetic tunnel junctions for unconventional computing. Phys. Rev. Appl. 8, 054045 (2017). + +<|ref|>text<|/ref|><|det|>[[55, 362, 911, 395]]<|/det|> +28. Friedman, J. S., Calvet, L. E., Bessiere, P., Droulez, J. & Querlioz, D. Bayesian inference with muller c-elements. IEEE Transactions on Circuits Syst. I: Regul. Pap. 63, 895-904 (2016). + +<|ref|>text<|/ref|><|det|>[[55, 400, 911, 432]]<|/det|> +29. Zheng, Y. et al. Hardware implementation of bayesian network based on two-dimensional memtransistors. Nat. communications 13, 1-11 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 437, 686, 454]]<|/det|> +30. Harabi, K.-E. et al. A memristor-based bayesian machine. Nat. Electron. 1-12 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 459, 911, 492]]<|/det|> +31. Gao, D. et al. Bayesian inference based robust computing on memristor crossbar. In 2021 58th ACM/IEEE Design Automation Conference (DAC), 121-126 (IEEE, 2021). + +<|ref|>text<|/ref|><|det|>[[55, 496, 911, 530]]<|/det|> +32. Moody, G. B., Mark, R. G. & Goldberger, A. L. Physionet: a web-based resource for the study of physiologic signals. IEEE Eng. Medicine Biol. Mag. 20, 70-75 (2001). + +<|ref|>text<|/ref|><|det|>[[55, 534, 879, 552]]<|/det|> +33. Wan, W. et al. A compute-in-memory chip based on resistive random-access memory. Nature 608, 504-512 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 556, 911, 590]]<|/det|> +34. Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521 (2014). + +<|ref|>text<|/ref|><|det|>[[55, 594, 911, 627]]<|/det|> +35. Balatti, S., Ambrogio, S., Gilmer, D. C. & Ielmini, D. Set variability and failure induced by complementary switching in bipolar ram. IEEE electron device letters 34, 861-863 (2013). + +<|ref|>text<|/ref|><|det|>[[55, 631, 911, 665]]<|/det|> +36. Esmanhotto, E. et al. Experimental demonstration of multilevel resistive random access memory programming for up to two months stable neural networks inference accuracy. Adv. Intell. Syst. 2200145 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 669, 911, 702]]<|/det|> +37. Le Gallo, M. et al. Precision of bit slicing with in-memory computing based on analog phase-change memory crossbars. Neuromorphic Comput. Eng. 2, 014009 (2022). + +<|ref|>text<|/ref|><|det|>[[55, 706, 911, 739]]<|/det|> +38. Hoffman, M. D. & Gelman, A. The no-u-turn sampler: adaptively setting path lengths in hamiltonian monte carlo. J. Mach. Learn. Res. 15, 1593-1623 (2014). + +<|ref|>text<|/ref|><|det|>[[55, 743, 911, 777]]<|/det|> +39. Blei, D. M., Kucukelbir, A. & McAuliffe, J. D. Variational inference: A review for statisticians. J. Am. statistical Assoc. 112, 859-877 (2017). + +<|ref|>text<|/ref|><|det|>[[55, 781, 911, 814]]<|/det|> +40. Ezzedeen, M. et al. Ultrahigh-density 3-d vertical rram with stacked junctionless nanowires for in-memory-computing applications. IEEE Transactions on Electron Devices 67, 4626-4630, DOI: 10.1109/TED.2020.3020779 (2020). + +<|ref|>sub_title<|/ref|><|det|>[[57, 835, 275, 854]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[57, 860, 911, 922]]<|/det|> +This work was supported by European Research Council consolidator grant DIVERSE (reference: 101043854) and by European Research Council starting grant NANOINFER (reference: 715872). It also benefits from a France 2030 government grant managed by the French National Research Agency (ANR-22-PEEL-0010). In addition, we thank L. Hutin, S. Bonnetier, F. Andrieu, J. Arcamone, J. Grollier, P. Bessiere and J. Droulez for discussing various aspects of the article. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[90, 77, 285, 95]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[88, 100, 909, 210]]<|/det|> +D.B. and T.H. proposed the initial idea of the hardware- calibrated training algorithm. D.B, T.H. D.Q, and E.V. conceived the experiments. D.B. and V.M. performed the experiments with the phase- change memory array. D.B., S.M., and N.C. performed the inference measurements on the two- layer Bayesian neural network. D.B. and T.H. conducted the software experiments and analysed the data. T.D. and A.M. performed preliminary studies concerning Bayesian neural networks and uncertainty evaluation. E.E. designed the circuits, under the supervision of J.M.P. The circuits were fabricated at CEA- Leti under the supervision of J.F.N. and G.B. D.Q. and E.V. supervised the work and wrote the initial version of the manuscript. All authors discussed the results and reviewed the manuscript. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[60, 130, 580, 150]]<|/det|> +- SupplInfoBringinguncertaintyquantificationtotheedge.pdf + +<--- Page Split ---> diff --git a/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/images_list.json b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..90affe62446c848c0dacee67ff877dd39eed199e --- /dev/null +++ b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/images_list.json @@ -0,0 +1,137 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1: Droplet vs. bubble, device structure and layout, and magnetic characterization. (a) Schematic of dynamical magnetic droplet soliton. (b) Schematic of a static magnetic bubble. (c) Schematic of an all-perpendicular STNO composed of [Co/Pd] (fixed) and [Co/Ni] (free) multilayers with a Cu spacer fabricated on a SiN membrane structure. The insets underneath show optical micrographs of the SiN membrane areas through which the different metal layers of the device can be seen. (d) Hysteresis loops of single Co/Pd and Co/Ni layers. (e) Hysteresis loop of a full [Co/Pd]/Cu/[Co/Ni] stack.", + "footnote": [], + "bbox": [ + [ + 260, + 95, + 728, + 755 + ] + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2: Microwave noise and STNO resistance vs. field. (a)-(d) Color plot of the power spectral density (PSD) of the microwave noise as a function of decreasing (a,c) and increasing (b,d) field, with the STNO resistance (white line) overlayed; the applied current is \\(-5 \\mathrm{mA}\\) . (a,b) Wide field sweep covering full saturation at both positive and negative fields. P/AP indicate the parallel/antiparallel state of the STNO; red arrow indicates the droplet region, and green arrow the bubble region. (c,d) Minor field sweeps showing how the droplet/bubble transition is fully reversible.", + "footnote": [], + "bbox": [ + [ + 250, + 95, + 744, + 725 + ] + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3: Phase diagrams based on the resistance and the microwave noise. (a) STNO resistance and (b) integrated (0–0.5 GHz) microwave noise level as a function of field and current. (c) shows the noise level in (b) overlaid on the resistance in (a) displayed using a gray scale highlighting intermediate resistance levels indicative of droplets/bubbles. The dash-dotted black line corresponds to the field-sweep at \\(I = -5 \\mathrm{mA}\\) given in Fig. 2. The parallel (P) and antiparallel (AP) states are easily discernible in the MR-map (a) as dark blue and dark red, while both the droplet and the bubble are characterized by intermediate resistance in green–yellow. The stark difference between the droplet and the bubble is revealed in the noise spectrum (b), where the stability of the bubble is manifested. Note however that the light-blue flanges in (a) correspond to a different droplet regime not captured in the microwave signal presented in (b).", + "footnote": [], + "bbox": [ + [ + 112, + 270, + 880, + 495 + ] + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4: Scanning Transmission X-ray Microscopy (STXM). (a)–(f) STXM images of the nanocontact region vs. decreasing field for a current of -7 mA. Blue corresponds to magnetization alligned with the applied field, red corresponds to magnetization anti-alligned with the applied field, whereas white indicates in-plane magnetization. The STNO resistance and the microwave noise PSD vs. decreasing field are shown in (g) where the points corresponding to the STXM images have been labelled a–f. The inset in (g) highlights the perimeter of the droplet/bubble as the applied field is decreased from 200 mT (dark blue) to 60 mT (blue), and further reduced to -40 mT (red) in steps of 20 mT.", + "footnote": [], + "bbox": [ + [ + 232, + 95, + 760, + 707 + ] + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/Supplementary_Figure_1.jpg", + "caption": "Figure S1: Supplementary Fig.1. A zoom-in of the phase diagram in Fig. 3(a) of the main text. The color code represents the different states: droplet (green), bubble (gray) and antiparallel (red). A droplet is nucleated at high currents and fields. Below a certain positive field \\((H_{\\mathrm{balance}})\\) the droplet is stabilized as a static bubble due to magnetostatic effects. The bubble is pinned below the nanocontact until the negative field is high enough to let the bubble domain expand throughout the film at \\(H_{\\mathrm{pinning}}\\) . At low currents the magnetic switching is only governed by the coercive field \\((H_{c})\\) of the free layer.", + "footnote": [], + "bbox": [ + [ + 149, + 179, + 832, + 625 + ] + ], + "page_idx": 24 + }, + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Figure 1", + "footnote": [], + "bbox": [ + [ + 42, + 95, + 550, + 831 + ] + ], + "page_idx": 25 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Figure 2", + "footnote": [], + "bbox": [ + [ + 44, + 110, + 620, + 850 + ] + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Figure 3", + "footnote": [], + "bbox": [ + [ + 42, + 137, + 951, + 400 + ] + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Figure 4", + "footnote": [], + "bbox": [ + [ + 50, + 45, + 584, + 666 + ] + ], + "page_idx": 28 + } +] \ No newline at end of file diff --git a/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044.mmd b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044.mmd new file mode 100644 index 0000000000000000000000000000000000000000..8e4fe0390651e56bfa4892bbce886f57c581287b --- /dev/null +++ b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044.mmd @@ -0,0 +1,318 @@ + +# Freezing and thawing magnetic droplet solitons + +M. Ahlberg ( martina.ahlberg@physics.gu.se) University of Gothenburg + +Sunjae Chung University of Gothenburg https://orcid.org/0000- 0002- 9970- 0060 + +Sheng Jiang Northwestern Polytechnical University + +Q. Tuan Le KTH Royal Institute of Technology + +Roman Khymyn University of Gothenburg https://orcid.org/0000- 0002- 9698- 1610 + +Hamid Mazraati KTH Royal Institute of Technology + +Markus Weigand Helmholtz-Zentrum Berlin https://orcid.org/0000- 0002- 0325- 2268 + +Iuliia Bykova Max Planck Institute for Intelligent Systems + +Felix GroB Max Plank Institute for Intelligent Systems https://orcid.org/0000- 0002- 2412- 285X + +Eberhard Goering Max Planck Institute for Intelligent Systems + +Gisela Schutz Max Planck Institute for Intelligent Systems + +Joachim Grafe Max Planck Institute for Intelligent Systems https://orcid.org/0000- 0002- 4597- 5923 + +Johan Akerman University of Gothenburg https://orcid.org/0000- 0002- 3513- 6608 + +## Article + +Keywords: spintronics, magnetic droplets, magnetic devices + +Posted Date: May 17th, 2021 + +DOI: https://doi.org/10.21203/rs.3.rs- 479057/v1 + +<--- Page Split ---> + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Version of Record: A version of this preprint was published at Nature Communications on May 5th, 2022. See the published version at https://doi.org/10.1038/s41467-022-30055-7. + +<--- Page Split ---> + +## Freezing and thawing magnetic droplet solitons + +Martina Ahlberg \(^{1,*}\) , Sunjae Chung \(^{1,3,* + }\) , Sheng Jiang \(^{1,4,5}\) , Q. Tuan Le \(^{1,5}\) , Roman Khymyn \(^{1}\) , Hamid Mazraati \(^{2,5}\) , Markus Weigand \(^{6}\) , Iuliia Bykova \(^{6}\) , Felix Groß \(^{6}\) , Eberhard Goering \(^{6}\) , Gisela Schütz \(^{6}\) , Joachim Gräfe \(^{6}\) , & Johan Åkerman \(^{1,2,5, + }\) + +\(^{1}\) Department of Physics, University of Gothenburg, 412 96 Gothenburg, Sweden \(^{2}\) NanOsc AB, 164 40 Kista, Sweden \(^{3}\) Department of Physics Education, Korea National University of Education, Cheongju 28173, Korea \(^{4}\) School of Microelectronics, Northwestern Polytechnical University, Xi'an 710072, China \(^{5}\) Department of Applied Physics, School of Engineering Sciences, KTH Royal Institute of Technology, 100 44 Stockholm, Sweden \(^{6}\) Max Planck Institute for Intelligent Systems, Stuttgart, Germany + +Magnetic droplets are non- topological magnetodynamical solitons displaying a wide range of complex dynamic phenomena with potential for microwave signal generation. Bubbles, on the other hand, are internally static cylindrical magnetic domains, stabilized by external fields and magnetostatic interactions. In its original theory, the droplet was described as an imminently collapsing bubble stabilized by spin transfer torque and, in its zero- frequency limit, as equivalent to a bubble. Without nanoscale lateral confinement, pinning, or an external applied field, such a nanobubble is unstable, and should collapse. Here, we show that + +<--- Page Split ---> + +we can freeze dynamic droplets into static nanobubbles by decreasing the magnetic field. While the bubble has virtually the same resistance as the droplet, all signs of low- frequency microwave noise disappear. The transition is fully reversible and the bubble can be thawed back into a droplet if the magnetic field is increased under current. Whereas the droplet collapses without a sustaining current, the bubble is highly stable and remains intact for days without external drive. Electrical measurements are complemented by direct observation using scanning transmission x- ray microscopy, which corroborates the analysis and confirms that the bubble is stabilized by pinning. + +Magnetic droplets are intrinsically dynamic, non- topological, magnetodynamical solitons1–15, which can be nucleated and sustained both in spin torque nano- oscillators (STNOs)3,6,8,12,14 and spin Hall nano- oscillators (SHNOs)16, provided the magnetodynamically active layer has sufficient perpendicular magnetic anisotropy (PMA). Magnetic droplets are characterized by a reversed core separated from the surrounding magnetization via a perimeter of precessing spins (See Fig.1(a))2,3. While first predicted over 40 years ago in an ideal zero- damping medium1, their possible experimental realization was later suggested theoretically2 in STNOs with PMA free layers17,18. After the first experimental demonstration of magnetic droplets, reported in STNOs with a PMA Co/Ni free layer and a Co fixed layer3, interest in magnetic droplets continues to increase due to its interesting characteristics, such as a highly nonlinear dynamics2,11,19, large power emission3,10,20,21, and possible applications in microwave- assisted magnetic recording (MAMR)22,23 and neuromorphic chips as nonlinear oscillators24–26. Several theoretical5,11,15,19,27–33 and experimental6–10,12,16,21,34–38 studies on magnetic droplets have since been presented. + +<--- Page Split ---> +![](images/Figure_1.jpg) + +
Figure 1: Droplet vs. bubble, device structure and layout, and magnetic characterization. (a) Schematic of dynamical magnetic droplet soliton. (b) Schematic of a static magnetic bubble. (c) Schematic of an all-perpendicular STNO composed of [Co/Pd] (fixed) and [Co/Ni] (free) multilayers with a Cu spacer fabricated on a SiN membrane structure. The insets underneath show optical micrographs of the SiN membrane areas through which the different metal layers of the device can be seen. (d) Hysteresis loops of single Co/Pd and Co/Ni layers. (e) Hysteresis loop of a full [Co/Pd]/Cu/[Co/Ni] stack.
+ +<--- Page Split ---> + +As pointed out by Hoefer et al., the droplet is reminiscent of a magnetic bubble2 (Fig.1(b)) and they identify a possible zero- frequency droplet with a topologically trivial magnetic bubble39–43. Despite the large number of experimental droplet studies, the low- field/low- frequency behavior of droplets has not yet been explored and the relation between droplets and bubbles — as well as a possible transition between the two — remain unclear. In order to explore these phenomena, we here study magnetic droplets specifically in the low- field regime using both electrical and microwave spectroscopy measurements as well as the direct microscopical observation based on Scanning Transmission X- ray Microscopy (STXM). We find clear experimental evidence for a droplet- to- bubble transition as the field strength, and hence the droplet frequency, is reduced, and a reversible bubble- to- droplet transition as the field is again increased in an attempt to squash the bubble, provided stabilizing spin transfer torque is still present via the STNO current. Our experimental results hence corroborate the picture, first expressed by Hoefer et al., that a magnetic droplet can be viewed “as an imminently collapsing bubble that is critically stabilized by the localized injection of spin torque”. + +Figure 1 shows a schematic of the studied all- perpendicular STNOs, comprised of a [Co/Pd]/Cu/[Co/Ni] GMR stack deposited on a \(\mathrm{Si}_{3}\mathrm{N}_{4}\) membrane (for fabrication details, please see Methods). Underneath the schematic we show two optical microscopy images taken from opposite directions to highlight the optical transmission of the \(\mathrm{Si}_{3}\mathrm{N}_{4}\) membrane. In Fig. 1(d) we show the magnetic properties of the individual free and fixed layers based on calibration samples, and their combined behavior in full STNO stacks in Fig. 1(e). + +<--- Page Split ---> + +Figure 2 presents the resistance and microwave signal as a function of field for an applied current of \(- 5\mathrm{mA}\) . The field is first increased from \(- 0.51\mathrm{T}\) to \(0.51\mathrm{T}\) in Fig. 2(a) and then decreased from positive to negative field in Fig. 2(b). At large negative fields, the STNO is in its lowest resistance state, consistent with a parallel (P) relative orientation of its free and fixed layers. At about \(- 0.49\mathrm{T}\) , the resistance increases about \(20\mathrm{mOhm}\) in a step- like fashion and there is a slight increase in the microwave noise background, both strong indications of the nucleation of a droplet. At about \(- 0.38\mathrm{T}\) , there is a second step- like increase in the STNO resistance and a marked further increase in the microwave noise. We interpret this as a transition into a larger droplet as the opposing applied field is reduced. At yet lower fields the droplet continues to grow in size (the STNO resistance increases), while its stability seems to deteriorate as indicated by the growing intensity of the microwave noise background. At about \(- 0.04\mathrm{T}\) , the microwave noise rapidly reaches a maximum and then suddenly disappears altogether, while the resistance exhibits a small jump of about \(5\mathrm{mOhm}\) . The complete microwave silence indicates that the magnetic state is now static, and we are lead to conclude that the droplet precession has stopped entirely and that the droplet has transitioned into a nanobubble state. + +The nanobubble resistance exhibits jumps reminiscent of Barkhausen noise44–46, indicating pinning possibly at grain boundaries or defects of the sputtered film. When the field is further increased, the bubble resistance increases gradually, indicating a continued growth of its size. At about \(0.06\mathrm{T}\) , the entire free layer switches its magnetization direction and the antiparallel (AP) state is clearly identifiable in the resistance. When the fixed layer switches at \(0.23\mathrm{T}\) , a droplet is immediately nucleated. With further increasing of the opposing field, the droplet again shows + +<--- Page Split ---> + +a gradual transition to a smaller size; the droplet finally disappears as the STNO transitions into a full P state at about 0.47 T. The overall behavior is very similar for decreasing fields (Fig.2b) where the same P/AP/droplet/nanobubble states can be clearly identified via the STNO resistance and the microwave noise. + +As mentioned above, the microwave noise power is far from constant for the whole droplet region and peaks at certain fields. We identify these peaks as marks of mode hopping between different droplet states14. While the details of the spectrum is highly reproducible (cf. increasing and decreasing fields) and serves as a fingerprint for each device, the patterns at negative and positive fields are quite different. The magnetoresistance implies that a relatively small and stable droplet \((\mu_{0}H < - 0.4 \text{T})\) is abruptly followed by a larger but similarly stable mode. In contrast to the symmetric noise patterns around the droplet- to- droplet transitions, the strong increase in microwave noise power around the droplet- to- bubble transition is highly asymmetric. There is first an extended field region of monotonic increase in the noise, which is then abruptly cut off and replaced by a completely silent bubble state. This highlights the very different non- dynamical nature of the nanobubble and suggests that mode hopping out of the nanobubble state and back into a droplet state is negligible, once the nanobubble has formed. + +Figure 2(c) and (d) demonstrate that it is possible to freeze the dynamic droplet into a static bubble and then thaw it back into a droplet using only the magnetic field under constant spin transfer torque. In particular, Fig. 2(d) shows how the nanobubble first is about to collapse at 0.025 T as it is getting squeezed by the opposing pressure from the increasing applied field. There + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Figure 2: Microwave noise and STNO resistance vs. field. (a)-(d) Color plot of the power spectral density (PSD) of the microwave noise as a function of decreasing (a,c) and increasing (b,d) field, with the STNO resistance (white line) overlayed; the applied current is \(-5 \mathrm{mA}\) . (a,b) Wide field sweep covering full saturation at both positive and negative fields. P/AP indicate the parallel/antiparallel state of the STNO; red arrow indicates the droplet region, and green arrow the bubble region. (c,d) Minor field sweeps showing how the droplet/bubble transition is fully reversible.
+ +<--- Page Split ---> + +is some slight Barkhausen noise in the rapidly dropping resistance, but otherwise no measurable microwave noise. However, instead of switching to a P state, the resistance then exhibits a sharp minimum after which it shows a rapid increase, which is accompanied by a high level of microwave noise. The collapsing nanobubble is hence rescued by the stabilizing spin transfer torque, which sets the spins in the bubble perimeter into precessional motion and restores the full dynamics of a magnetic droplet. Judging from the resistance, it is noteworthy that the droplet is slightly larger than the smallest nanobubble. Within the experimental accuracy (a field step of \(2\mathrm{mT}\) ), we do not observe any significant hysteresis in this transition. Hence there is a negligible energy barrier between the two states and the bubble can indeed be viewed as a zero- frequency droplet, albeit still likely affected by pinning. + +Figure 3(a) presents a phase diagram based on a two- dimensional map of the STNO resistance as functions of current and field. All data was acquired in a decreasing field at a constant current level. The parallel (P) and antiparallel (AP) configurations are easily identified by the dark blue and dark red colors, respectively, and for current magnitudes below \(1.8\mathrm{mA}\) , these are the only two available states, as expected for a GMR device. However, even at these weak currents, the \(\mathrm{P} \rightarrow \mathrm{AP}\) switching field is clearly affected by the STT from the nanocontact; in contrast, the \(\mathrm{AP} \rightarrow \mathrm{P}\) switching field is entirely unaffected. In an intermediate current region, from about \(- 1.8\) to \(- 3.5\mathrm{mA}\) , the STT can not yet sustain a droplet but is sufficient to create a nanobubble directly from the P state. As magnetic switching typically involves both domain nucleation and domain propagation, we interpret this current dependent switching in the following way (see Supplementary Materials for a zoom- in of this particular part of the phase diagram). For current magnitudes + +<--- Page Split ---> + +below 1 mA, magnetic switching is limited by the field required for domain nucleation and, in addition, the location of initial domain nucleation is far from the nanocontact region as STT from the current has no discernible impact. However, for current magnitudes above 1 mA, where we observe a strong current dependence of the switching field, we conclude that the domain nucleation has moved to underneath the nanocontact. If we reduce the field magnitude, we need a stronger current to assist in the domain nucleation, but once formed, it propagates through the entire free layer. However, at fields weaker than the field needed for domain propagation, i.e. the pinning field, which we read out as about \(60~\mathrm{mT}\) , the nucleated domain is no longer able to propagate and instead remains as a nanobubble directly underneath the nanocontact. The nanobubble can hence form either from the P state or from a droplet. + +The droplet shows two discernable states, a high- field/low- current mode that exhibits a rather small MR (light blue). This mode moves to higher fields with increasing current and is no longer visible above \(\approx - 6 \mathrm{mA}\) . The other distinguishable droplet mode is characterized by an intermediate resistance (green- yellow). The bubble is almost indiscernible from the latter droplet state, even though a subtle line traces out the transition between the two. Moreover, the bubble resistance is not a smooth function of applied field, but displays notches and steps, indicative of Barkhausen noise due to pinning. + +In contrast to their almost identical resistance, a stark difference between the droplet and the bubble is uncovered in Fig. 3(b), where we show the microwave signal integrated over 0–0.5 GHz. The droplet exhibits non- zero power levels of low frequency microwave noise, while the P, AP, + +<--- Page Split ---> + +and bubble states are definitely static and silent. Figure 3(b) also further unveils the complex relation between the applied field and current, and the particular droplet characteristics. A strong microwave noise signal denotes mode hopping and these events exhibit a strong dependence on both field and current. We can identify three traces of mode hopping for positive fields, while there is only two weak trails at negative fields. There is also regions where the droplet is very stable and the noise level is almost zero. These features act like fingerprints for each measured device and are highly reproducible in consecutive measurements, but differ between STNOs. We then overlay the microwave noise data onto the resistance data, now plotted with a gray scale that highlights intermediate resistance levels (Fig. 3(c)). Parts of the low- field/low- current droplet regime (light blue in Fig. 3(a)) does not exhibit any measurable microwave noise. It is possible that its dynamics is on a slower time scale than the microwave frequencies our set- up is sensitive to. + +We finally turn to the results of the scanning transmission X- ray microscopy results, illustrated in Fig. 4. Images of the droplet/bubble are shown in Fig. 4(a)- (f), and the corresponding magnetoresistance and microwave signal are presented in Fig. 4(g) with the matching field of the images marked by their letter. The STXM and the electrical measurements were performed in separate setups, hence there is a small uncertainty in comparing the field values of the two, although both measurements seem highly consistent with each other. The dashed white or black circles mark the position of the nanocontact. It has been placed by assuming that the droplet/bubble in Fig. 4(d) is centered under the NC and by comparing the non- magnetic contrast of the different images. The method works very well as confirmed by the good overlap of the perimeters in the inset of Fig. 4(g), but it should be remembered that the absolute position is still based on this assumption. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Figure 3: Phase diagrams based on the resistance and the microwave noise. (a) STNO resistance and (b) integrated (0–0.5 GHz) microwave noise level as a function of field and current. (c) shows the noise level in (b) overlaid on the resistance in (a) displayed using a gray scale highlighting intermediate resistance levels indicative of droplets/bubbles. The dash-dotted black line corresponds to the field-sweep at \(I = -5 \mathrm{mA}\) given in Fig. 2. The parallel (P) and antiparallel (AP) states are easily discernible in the MR-map (a) as dark blue and dark red, while both the droplet and the bubble are characterized by intermediate resistance in green–yellow. The stark difference between the droplet and the bubble is revealed in the noise spectrum (b), where the stability of the bubble is manifested. Note however that the light-blue flanges in (a) correspond to a different droplet regime not captured in the microwave signal presented in (b).
+ +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4: Scanning Transmission X-ray Microscopy (STXM). (a)–(f) STXM images of the nanocontact region vs. decreasing field for a current of -7 mA. Blue corresponds to magnetization alligned with the applied field, red corresponds to magnetization anti-alligned with the applied field, whereas white indicates in-plane magnetization. The STNO resistance and the microwave noise PSD vs. decreasing field are shown in (g) where the points corresponding to the STXM images have been labelled a–f. The inset in (g) highlights the perimeter of the droplet/bubble as the applied field is decreased from 200 mT (dark blue) to 60 mT (blue), and further reduced to -40 mT (red) in steps of 20 mT.
+ +<--- Page Split ---> + +Figure 4(a) is measured at \(270~\mathrm{mT}\) , and shows a mode associated with a high noise level in Fig. 4(g). Only a weak and mostly white feature is captured in the STXM image. STXM measures a time averaged image and the droplet is in this highly noisy regime expected to experience large drift instabilities and continuously vanish and renucleate underneath the nanocontact. As a consequence, only a washed out and poorly reversed feature results. In contrast, Fig. 4(b)-(d) display more stable and more clearly reversed droplets. They have approximately the same radius as the nanocontact, although the size definitely increases slightly with decreasing field, as expected. We have in earlier STXM work observed a significant effect of the Zhang- Li torque on the droplet size \(^{21,47}\) . The magnitude of this effect depends on the current density \((j_{\mathrm{dc}})\) and we have performed simulations which confirm that the difference between the droplet diameter here and in our former publication is indeed due to a weaker \(j_{\mathrm{dc}}\) . At zero field, a bubble is clearly formed and it prevails down to \(- 40~\mathrm{mT}\) (Fig. 4(e)-(f)). It is no longer centered on the nanocontact, but has mostly expanded in one direction. It should be noted though, that the field in the microscope is given by rotating permanent magnets and the sample may have been subjected to in-plane fields between two set values. Nevertheless, the inset in Fig. 4(g) presents the perimeter of the droplet/bubble as the field decreases from \(200~\mathrm{mT}\) (dark blue) to \(- 40~\mathrm{mT}\) (red), and the initial bubble at \(40~\mathrm{mT}\) (light blue) grows in distinct steps, which implies that the size is controlled by pinning. + +Returning to the original droplet theory of Hoefer et al, we note that pinning was not included. \(^{2}\) It is clear from our experimental observations that pinning has a strong and immediate impact on the relation between droplets and nanobubbles and must be included in the low- field/low- current regime. Instead of exhibiting a continuous slow- down and frequency decrease to zero with de + +<--- Page Split ---> + +creasing field, there is a minimum droplet precession frequency that spin transfer torque can sustain before pinning overcomes the precession. As this minimum frequency is approached from above, the broad- band microwave noise diverges as the competition between the inertia of the precession and the pinning makes the droplet dynamics increasingly erratic until pinning finally gets complete control of the precession abruptly stops, leaving complete microwave silence in its wake. + +1. Ivanov, B. & Kosevich, A. Bound states of a large number of magnons in a ferromagnet with a single-ion anisotropy. Zh. Eksp. Teor. Fiz. 72, 2000 (1977). + +2. Hoefer, M. A., Silva, T. J. & Keller, M. W. Theory for a dissipative droplet soliton excited by a spin torque nanocontact. Phys. Rev. B 82, 054432 (2010). + +3. Mohseni, S. M. et al. Spin Torque-Generated Magnetic Droplet Solitons. Science 339, 1295-1298 (2013). + +4. Mohseni, S. et al. Magnetic droplet solitons in orthogonal nano-contact spin torque oscillators. Physica B 435, 84-87 (2014). + +5. Iacocca, E. et al. Confined dissipative droplet solitons in spin-valve nanowires with perpendicular magnetic anisotropy. Phys. Rev. Lett. 112, 047201 (2014). + +6. Macia, F., Backes, D. & Kent, A. D. Stable magnetic droplet solitons in spin-transfer nanocontacts. Nat. Nanotechnol 9, 992-996 (2014). + +7. Chung, S. et al. Spin transfer torque generated magnetic droplet solitons (invited). J. Appl. Phys. 115, 172612 (2014). + +<--- Page Split ---> + +8. Lendinez, S., Statuto, N., Backes, D., Kent, A. D. & Macia, F. Observation of droplet soliton drift resonances in a spin-transfer-torque nanocontact to a ferromagnetic thin film. Phys. Rev. B 92, 174426 (2015). + +9. Chung, S. et al. Magnetic droplet solitons in orthogonal spin valves. Low Temp. Phys. 41, 833-837 (2015). + +10. Chung, S. et al. Magnetic droplet nucleation boundary in orthogonal spin-torque nano-oscillators. Nat. Commun. 7, 11209 (2016). + +11. Xiao, D. et al. Parametric autoexcitation of magnetic droplet soliton perimeter modes. Phys. Rev. B 95, 024106 (2017). + +12. Lendinez, S. et al. Effect of Temperature on Magnetic Solitons Induced by Spin-Transfer Torque. Phys. Rev. Appl. 7, 054027 (2017). + +13. Sulymenko, O., Prokopenko, O., Tyberkevych, V., Slavin, A. & Serga, A. Bullets and droplets: Two-dimensional spin-wave solitons in modern magnonics (Review Article). Low Temp. Phys. 44, 775 (2018). + +14. Statuto, N., Hahn, C., Hernandez, J. M., Kent, A. D. & Macia, F. Multiple magnetic droplet soliton modes. Phys. Rev. B 99, 174436 (2019). + +15. Mohseni, M. et al. Chiral excitations of magnetic droplet solitons driven by their own inertia. Phys. Rev. B 101, 20417 (2020). + +<--- Page Split ---> + +16. Divinskiy, B. et al. Magnetic droplet solitons generated by pure spin currents. Phys. Rev. B 96, 224419 (2017). + +17. Mohseni, S. M. et al. High frequency operation of a spin-torque oscillator at low field. Phys. Status Solidi RRL 5, 432-434 (2011). + +18. Rippard, W. H. et al. Spin-transfer dynamics in spin valves with out-of-plane magnetized CoNi free layers. Phys. Rev. B 81, 014426 (2010). + +19. Bookman, L. D. & Hoefer, M. A. Analytical theory of modulated magnetic solitons. Phys. Rev. B 88, 184401 (2013). + +20. Locatelli, N., Cros, V. & Grollier, J. Spin-torque building blocks. Nat. Mater. 13, 11 (2014). + +21. Chung, S. et al. Direct Observation of Zhang-Li Torque Expansion of Magnetic Droplet Solitons. Phys. Rev. Lett. 120, 217204 (2018). + +22. Okamoto, S., Kikuchi, N., Furuta, M., Kitakami, O. & Shimatsu, T. Microwave assisted magnetic recording technologies and related physics. J. Phys. D: Appl. Phys. 48, 353001 (2015). + +23. Bosu, S. et al. High frequency out-of-plane oscillation with large cone angle in mag-flip spin torque oscillators for microwave assisted magnetic recording. Appl. Phys. Lett. 110, 142403 (2017). + +24. Torrejon, J. et al. Neuromorphic computing with nanoscale spintronic oscillators. Nature 547, 428-431 (2017). + +<--- Page Split ---> + +25. Romera, M. et al. Vowel recognition with four coupled spin-torque nano-oscillators. Nature 563, 230-234 (2018). + +26. Macià, F. & Kent, A. D. Magnetic droplet solitons. J. Appl. Phys. 128, 100901 (2020). + +27. Maiden, M. D., Bookman, L. D. & Hoefer, M. a. Attraction, merger, reflection, and annihilation in magnetic droplet soliton scattering. Phys. Rev. B 89, 180409 (2014). + +28. Puliafito, V., Siracusano, G., Azzerboni, B. & Finocchio, G. Self-modulated soliton modes excited in a nanocontact spin-torque oscillator. IEEE Magn. Lett. 5, 3000104 (2014). + +29. Wang, C., Xiao, D., Zhou, Y., Akerman, J. & Liu, Y. Phase-locking of multiple magnetic droplets by a microwave magnetic field. AIP Adv. 7, 56019 (2017). + +30. Mohseni, M. et al. Magnetic droplet soliton nucleation in oblique fields. Phys. Rev. B 97, 184402 (2018). + +31. Mohseni, M. et al. Propagating Magnetic Droplet Solitons as Moveable Nanoscale Spin-Wave Sources with Tunable Direction of Emission. Phys. Rev. Appl. 13, 24040 (2020). + +32. Sisodia, N., Muduli, P. K., Papanicolaou, N. & Komineas, S. Chiral droplets and current-driven motion in ferromagnets. Phys. Rev. B 103, 24431 (2021). + +33. Yazdi, H. F., Ghasemi, G., Mohseni, M. & Mohseni, M. Tuning the dynamics of magnetic droplet solitons using dipolar interactions. Phys. Rev. B 103, 24441 (2021). + +34. Backes, D. et al. Direct Observation of a Localized Magnetic Soliton in a Spin-Transfer Nanocontact. Phys. Rev. Lett. 115, 127205 (2015). + +<--- Page Split ---> + +35. Carpentieri, M., Tomasello, R., Zivieri, R. & Finocchio, G. Topological, non-topological and instanton droplets driven by spin-transfer torque in materials with perpendicular magnetic anisotropy and Dzyaloshinskii-Moriya Interaction. Sci. Rep. 5, 16184 (2015). + +36. Hang, J., Hahn, C., Statuto, N., Macia, F. & Kent, A. D. Generation and annihilation time of magnetic droplet solitons. Sci. Rep. 8, 6847 (2018). + +37. Jiang, S. et al. Impact of the Oersted Field on Droplet Nucleation Boundaries. IEEE Magn. Lett. 9, 3104304 (2018). + +38. Shi, K. et al. Observation of Magnetic Droplets in Magnetic Tunnel Junctions. arXiv: 2012.05596 (2020). + +39. Nielsen, J. Bubble domain memory materials. IEEE Trans. Magn. 12, 327-345 (1976). + +40. Giess, E. A. Magnetic Bubble Materials. Science 208, 938-943 (1980). + +41. De Leeuw, F., Van Den Doel, R. & Enz, U. Dynamic properties of magnetic domain walls and magnetic bubbles. Rep. Prog. Phys. 43, 689 (1980). + +42. Komineas, S. & Papanicolaou, N. Topology and dynamics in ferromagnetic media. Physica D 99, 81-107 (1996). + +43. Moutafis, C., Komineas, S. & Bland, J. A. C. Dynamics and switching processes for magnetic bubbles in nanoelements. Phys. Rev. B 79, 224429 (2009). + +44. Kim, D.-H., Choe, S.-B. & Shin, S.-C. Direct Observation of Barkhausen Avalanche in Co Thin Films. Phys. Rev. Lett. 90, 087203 (2003). + +<--- Page Split ---> + +45. Balk, A. L., Stiles, M. D. & Unguris, J. Critical behavior of zero-field magnetic fluctuations in perpendicularly magnetized thin films. Phys. Rev. B 90, 184404 (2014). + +46. Herranen, T. & Laurson, L. Barkhausen Noise from Precessional Domain Wall Motion. Phys. Rev. Lett. 122, 117205 (2019). + +47. Albert, J., Macia, F. & Hernandez, J. M. Effect of the Zhang-Li torque on spin-torque nano-oscillators. Phys. Rev. B 102, 184421 (2020). + +48. J. Grafe, M. Weigand, B. Van Waeyenberge, A. Gangwar, F. GroB, F. Lisiecki, J. Rychly, H. Stoll, N. Trager, J. Forster, F. Stobiecki, J. Dubowik, J. Klos, M. Krawczyk, C. H. Back, E. J. Goering, G. Schutz, H.-J. M. Drouhin, J.-E. Wegrowe, and M. Razeghi. Visualizing nanoscale spin waves using MAXYMUS. In Proc. SPIE, vol. 11090, 1109025 (Spintronics XII, 2019). + +49. Nolle, D. et al. Note: Unique characterization possibilities in the ultra high vacuum scanning transmission x-ray microscope (UHV-STXM) "MAXYMUS" using a rotatable permanent magnetic field up to 0.22 T. Rev. Sci. Instrum. 83, 046112 (2012). + +## Methods + +Sample Preparation A sample stack is consisted of a Ta (4 nm)/ Cu (14 nm) / Ta (4 nm) / Pd (2 nm) seed layer and an all-perpendicular pseudo-spin valve [Co (0.35 nm) / Pd (0.7 nm)] \(\times 5\) / Co (0.35 nm) / Cu (5 nm) / [Co (0.22 nm) / Ni (0.68 nm)] \(\times 4\) / Co (0.22 nm), capped by a Cu (2 nm) / Pd (2 nm) layer, which was deposited by magnetron sputtering on Si wafer with 300 nm thick LPCVD silicon nitride layer. Using a conventional photo-lithography and metal-etching tech + +<--- Page Split ---> + +niques, \(8 \mu \mathrm{m} \times 16 \mu \mathrm{m}\) mesas were fabricated on above stack wafer and all mesas were insulated by a 30- nm- thick \(\mathrm{SiO}_2\) layer deposited by using chemical vapor deposition (CVD). To pattern nanocontacts (NCs) on the top of each mesa having different diameters from 50 to \(150 \mathrm{nm}\) , electron beam lithography was used. \(\mathrm{SiO}_2\) layer was then etched through by the reactive ion etching (RIE) technique to open NCs. The NC- STO device fabrication was completed by the deposition of \(\mathrm{Cu} 200 \mathrm{nm} / \mathrm{Au} 100 \mathrm{nm}\) top electrode and lift- off processing. For STXM measurements, Si was removed from backside using highly selective RIE process and leave only SiN membrane to allow X- ray transmission underneath NC- STOs. (See, Figure 1(c)) For magnetic and electrical charaterization of NC- STOs, same stack was prepared on Si thermally oxidized Si wafer and then similar fabrication processing were done except a deep etching for a membrane structure. + +Magnetic and Electrical Characterization the magnetization hysteresis loops was measured using Alternating Gradient Magnetometry (AGM) with the unpatterned material stacks. \(dc\) and microwave measurements of the fabricated STOs were carried out using our custom- built setup, where magnetic field strength, polarity, and angle can be controlled. A magnetic field between - 0.5 to \(+0.5\) can be manipulated using electromagnet. The device is connected using GSG probe to a \(dc\) - current source (Keithley 6221), a nanovoltmeter (Keithley 2182A), and a spectrum analyzer (R & S FSQ26). A 0- 40GHz bias- tee is used to separate the bias input and the generated microwave signal. The microwave sinalg is amplified by a low- noise amplifier (operational range: 0.1- 26.5 GHz) before being sent to the spectrum analyzer. + +Scanning transmission x- ray microscopy The STXM measurements were performed at the BESSY II synchrotron, using the MPI IS operated MAXYMUS end station at the UE46- PGM2 beam line. \(^{48}\) + +<--- Page Split ---> + +The out- of- plane component of the magnetization was probed using circularly polarized light at normal incidence. The applied field, with a maximum value of \(300\mathrm{mT}\) , was generated by a set of four rotatable permanent magnets49. An optimal XMCD contrast was achieved by setting the photon energy to the Ni \(L_{3}\) edge, which resulted in clear images. The size of each pixel is \(10 \times 10 \mathrm{nm}^{2}\) , while the nominal resolution of the focusing plate is \(18\mathrm{nm}\) . + +Acknowledgements This work was supported by the Swedish Research Council (VR; 2017- 06711 and 2019- 04229). Helmholtz Zentrum Berlin is acknowledged for allocating beam time at the BESSY II synchrotron radiation facility. M. W., E.G., G.S. and J.G. acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the frame work of DynaMAX (Project No. 05K18EYA). This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. 2020R1F1A1049642) + +Acknowledgements These authors contributed equally: Martina Ahlberg and Sunjae Chung. + +Contributions M.A., S.C., and J.A. conceived the project, S.C. S. J. and T.Q.L. performed the electrical measurements. S.C., T.Q.L., S.J., and A.H. fabricated the devices. M.A., S. J., J.G., M.W., F.G. and I.B. carried out the STXM measurements. J.A. coordinated the project. All authors analyzed the results and co- wrote the manuscript. + +Competing Interests The authors declare that they have no competing financial interests. + +Correspondence Correspondence and requests for materials should be addressed to S. Chung (email: sjchung76@knue.ac.kr) and J. Akerman (email: johan.akerman@physics.gu.se). + +<--- Page Split ---> + +Supplementary information. + +<--- Page Split ---> +![](images/Supplementary_Figure_1.jpg) + +
Figure S1: Supplementary Fig.1. A zoom-in of the phase diagram in Fig. 3(a) of the main text. The color code represents the different states: droplet (green), bubble (gray) and antiparallel (red). A droplet is nucleated at high currents and fields. Below a certain positive field \((H_{\mathrm{balance}})\) the droplet is stabilized as a static bubble due to magnetostatic effects. The bubble is pinned below the nanocontact until the negative field is high enough to let the bubble domain expand throughout the film at \(H_{\mathrm{pinning}}\) . At low currents the magnetic switching is only governed by the coercive field \((H_{c})\) of the free layer.
+ +<--- Page Split ---> + +## Figures + +![](images/Figure_1.jpg) + +
Figure 1
+ +Droplet vs. bubble, device structure and layout, and magnetic characterization. (a) Schematic of dynamical magnetic droplet soliton. (b) Schematic of a static magnetic bubble. (c) Schematic of an all-perpendicular STNO composed of [Co/Pd] (fixed) and [Co/Ni] (free) multilayers with a Cu spacer + +<--- Page Split ---> + +fabricated on a SiN membrane structure. The insets underneath show optical micrographs of the SiN membrane areas through which the different metal layers of the device can be seen. (d) Hysteresis loops of single Co/Pd and Co/Ni layers. (e) Hysteresis loop of a full [Co/Pd]/Cu/[Co/Ni] stack. + +![](images/Figure_2.jpg) + +
Figure 2
+ +Microwave noise and STNO resistance vs. field. (a)-(d) Color plot of the power spectral density (PSD) of the microwave noise as a function of decreasing (a,c) and increasing (b,d) field, with the STNO resistance + +<--- Page Split ---> + +(white line) overlayed; the applied current is 5 mA. (a,b) Wide field sweep covering full saturation at both positive and negative fields. P/AP indicate the parallel/antiparallel state of the STNO; red arrow indicates the droplet region, and green arrow the bubble region. (c,d) Minor field sweeps showing how the droplet/bubble transition is fully reversible. + +![](images/Figure_3.jpg) + +
Figure 3
+ +Phase diagrams based on the resistance and the microwave noise. (a) STNO resistance and (b) integrated (0–0.5 GHz) microwave noise level as a function of field and current. (c) shows the noise level in (b) overlaid on the resistance in (a) displayed using a gray scale highlighting intermediate resistance levels indicative of droplets/bubbles. The dash-dotted black line corresponds to the field-sweep at \(I = - 5\) mA given in Fig. 2. The parallel (P) and antiparallel (AP) states are easily discernible in the MR-map (a) as dark blue and dark red, while both the droplet and the bubble are characterized by intermediate resistance in green–yellow. The stark difference between the droplet and the bubble is revealed in the noise spectrum (b), where the stability of the bubble is manifested. Note however that the light-blue flanges in (a) correspond to a different droplet regime not captured in the microwave signal presented in (b). + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Figure 4
+ +Scanning Transmission X- ray Microscopy (STXM). (a)–(f) STXM images of the nanocontact region vs. decreasing field for a current of - 7 mA. Blue corresponds to magnetization alligned with the applied field, red corresponds to magnetization anti- alligned with the applied field, whereas white indicates in- plane magnetization. The STNO resistance and the microwave noise PSD vs. decreasing field are shown in (g) where the points corresponding to the STXM images have been labelled a–f. The inset in (g) highlights the perimeter of the droplet/bubble as the applied field is decreased from 200 mT (dark blue) to 60 mT (blue), and further reduced to - 40 mT (red) in steps of 20 mT. + +## Supplementary Files + +<--- Page Split ---> + +This is a list of supplementary files associated with this preprint. Click to download. + +- FigureSupMat.jpg + +<--- Page Split ---> diff --git a/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044_det.mmd b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..a43785c963691b59aa31fca0f5508f753eadafc3 --- /dev/null +++ b/preprint/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044/preprint__17587bcada726ad31559b3b0b0e60a8d7b3ad69123102830efcd17d952398044_det.mmd @@ -0,0 +1,408 @@ +<|ref|>title<|/ref|><|det|>[[45, 107, 875, 144]]<|/det|> +# Freezing and thawing magnetic droplet solitons + +<|ref|>text<|/ref|><|det|>[[44, 161, 470, 202]]<|/det|> +M. Ahlberg ( martina.ahlberg@physics.gu.se) University of Gothenburg + +<|ref|>text<|/ref|><|det|>[[44, 208, 630, 248]]<|/det|> +Sunjae Chung University of Gothenburg https://orcid.org/0000- 0002- 9970- 0060 + +<|ref|>text<|/ref|><|det|>[[44, 254, 384, 295]]<|/det|> +Sheng Jiang Northwestern Polytechnical University + +<|ref|>text<|/ref|><|det|>[[44, 301, 354, 342]]<|/det|> +Q. Tuan Le KTH Royal Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 348, 630, 389]]<|/det|> +Roman Khymyn University of Gothenburg https://orcid.org/0000- 0002- 9698- 1610 + +<|ref|>text<|/ref|><|det|>[[44, 394, 354, 435]]<|/det|> +Hamid Mazraati KTH Royal Institute of Technology + +<|ref|>text<|/ref|><|det|>[[44, 440, 636, 481]]<|/det|> +Markus Weigand Helmholtz-Zentrum Berlin https://orcid.org/0000- 0002- 0325- 2268 + +<|ref|>text<|/ref|><|det|>[[44, 486, 434, 527]]<|/det|> +Iuliia Bykova Max Planck Institute for Intelligent Systems + +<|ref|>text<|/ref|><|det|>[[44, 532, 784, 573]]<|/det|> +Felix GroB Max Plank Institute for Intelligent Systems https://orcid.org/0000- 0002- 2412- 285X + +<|ref|>text<|/ref|><|det|>[[44, 578, 434, 619]]<|/det|> +Eberhard Goering Max Planck Institute for Intelligent Systems + +<|ref|>text<|/ref|><|det|>[[44, 624, 434, 665]]<|/det|> +Gisela Schutz Max Planck Institute for Intelligent Systems + +<|ref|>text<|/ref|><|det|>[[44, 670, 790, 712]]<|/det|> +Joachim Grafe Max Planck Institute for Intelligent Systems https://orcid.org/0000- 0002- 4597- 5923 + +<|ref|>text<|/ref|><|det|>[[44, 717, 630, 758]]<|/det|> +Johan Akerman University of Gothenburg https://orcid.org/0000- 0002- 3513- 6608 + +<|ref|>sub_title<|/ref|><|det|>[[44, 800, 101, 818]]<|/det|> +## Article + +<|ref|>text<|/ref|><|det|>[[44, 838, 560, 858]]<|/det|> +Keywords: spintronics, magnetic droplets, magnetic devices + +<|ref|>text<|/ref|><|det|>[[44, 876, 295, 895]]<|/det|> +Posted Date: May 17th, 2021 + +<|ref|>text<|/ref|><|det|>[[44, 914, 462, 933]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 479057/v1 + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 44, 910, 87]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 123, 949, 167]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on May 5th, 2022. See the published version at https://doi.org/10.1038/s41467-022-30055-7. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 85, 755, 113]]<|/det|> +## Freezing and thawing magnetic droplet solitons + +<|ref|>text<|/ref|><|det|>[[113, 140, 884, 234]]<|/det|> +Martina Ahlberg \(^{1,*}\) , Sunjae Chung \(^{1,3,* + }\) , Sheng Jiang \(^{1,4,5}\) , Q. Tuan Le \(^{1,5}\) , Roman Khymyn \(^{1}\) , Hamid Mazraati \(^{2,5}\) , Markus Weigand \(^{6}\) , Iuliia Bykova \(^{6}\) , Felix Groß \(^{6}\) , Eberhard Goering \(^{6}\) , Gisela Schütz \(^{6}\) , Joachim Gräfe \(^{6}\) , & Johan Åkerman \(^{1,2,5, + }\) + +<|ref|>text<|/ref|><|det|>[[113, 277, 885, 590]]<|/det|> +\(^{1}\) Department of Physics, University of Gothenburg, 412 96 Gothenburg, Sweden \(^{2}\) NanOsc AB, 164 40 Kista, Sweden \(^{3}\) Department of Physics Education, Korea National University of Education, Cheongju 28173, Korea \(^{4}\) School of Microelectronics, Northwestern Polytechnical University, Xi'an 710072, China \(^{5}\) Department of Applied Physics, School of Engineering Sciences, KTH Royal Institute of Technology, 100 44 Stockholm, Sweden \(^{6}\) Max Planck Institute for Intelligent Systems, Stuttgart, Germany + +<|ref|>text<|/ref|><|det|>[[113, 624, 885, 864]]<|/det|> +Magnetic droplets are non- topological magnetodynamical solitons displaying a wide range of complex dynamic phenomena with potential for microwave signal generation. Bubbles, on the other hand, are internally static cylindrical magnetic domains, stabilized by external fields and magnetostatic interactions. In its original theory, the droplet was described as an imminently collapsing bubble stabilized by spin transfer torque and, in its zero- frequency limit, as equivalent to a bubble. Without nanoscale lateral confinement, pinning, or an external applied field, such a nanobubble is unstable, and should collapse. Here, we show that + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 885, 367]]<|/det|> +we can freeze dynamic droplets into static nanobubbles by decreasing the magnetic field. While the bubble has virtually the same resistance as the droplet, all signs of low- frequency microwave noise disappear. The transition is fully reversible and the bubble can be thawed back into a droplet if the magnetic field is increased under current. Whereas the droplet collapses without a sustaining current, the bubble is highly stable and remains intact for days without external drive. Electrical measurements are complemented by direct observation using scanning transmission x- ray microscopy, which corroborates the analysis and confirms that the bubble is stabilized by pinning. + +<|ref|>text<|/ref|><|det|>[[112, 395, 886, 856]]<|/det|> +Magnetic droplets are intrinsically dynamic, non- topological, magnetodynamical solitons1–15, which can be nucleated and sustained both in spin torque nano- oscillators (STNOs)3,6,8,12,14 and spin Hall nano- oscillators (SHNOs)16, provided the magnetodynamically active layer has sufficient perpendicular magnetic anisotropy (PMA). Magnetic droplets are characterized by a reversed core separated from the surrounding magnetization via a perimeter of precessing spins (See Fig.1(a))2,3. While first predicted over 40 years ago in an ideal zero- damping medium1, their possible experimental realization was later suggested theoretically2 in STNOs with PMA free layers17,18. After the first experimental demonstration of magnetic droplets, reported in STNOs with a PMA Co/Ni free layer and a Co fixed layer3, interest in magnetic droplets continues to increase due to its interesting characteristics, such as a highly nonlinear dynamics2,11,19, large power emission3,10,20,21, and possible applications in microwave- assisted magnetic recording (MAMR)22,23 and neuromorphic chips as nonlinear oscillators24–26. Several theoretical5,11,15,19,27–33 and experimental6–10,12,16,21,34–38 studies on magnetic droplets have since been presented. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[260, 95, 728, 755]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 767, 883, 895]]<|/det|> +
Figure 1: Droplet vs. bubble, device structure and layout, and magnetic characterization. (a) Schematic of dynamical magnetic droplet soliton. (b) Schematic of a static magnetic bubble. (c) Schematic of an all-perpendicular STNO composed of [Co/Pd] (fixed) and [Co/Ni] (free) multilayers with a Cu spacer fabricated on a SiN membrane structure. The insets underneath show optical micrographs of the SiN membrane areas through which the different metal layers of the device can be seen. (d) Hysteresis loops of single Co/Pd and Co/Ni layers. (e) Hysteresis loop of a full [Co/Pd]/Cu/[Co/Ni] stack.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 587]]<|/det|> +As pointed out by Hoefer et al., the droplet is reminiscent of a magnetic bubble2 (Fig.1(b)) and they identify a possible zero- frequency droplet with a topologically trivial magnetic bubble39–43. Despite the large number of experimental droplet studies, the low- field/low- frequency behavior of droplets has not yet been explored and the relation between droplets and bubbles — as well as a possible transition between the two — remain unclear. In order to explore these phenomena, we here study magnetic droplets specifically in the low- field regime using both electrical and microwave spectroscopy measurements as well as the direct microscopical observation based on Scanning Transmission X- ray Microscopy (STXM). We find clear experimental evidence for a droplet- to- bubble transition as the field strength, and hence the droplet frequency, is reduced, and a reversible bubble- to- droplet transition as the field is again increased in an attempt to squash the bubble, provided stabilizing spin transfer torque is still present via the STNO current. Our experimental results hence corroborate the picture, first expressed by Hoefer et al., that a magnetic droplet can be viewed “as an imminently collapsing bubble that is critically stabilized by the localized injection of spin torque”. + +<|ref|>text<|/ref|><|det|>[[113, 621, 885, 825]]<|/det|> +Figure 1 shows a schematic of the studied all- perpendicular STNOs, comprised of a [Co/Pd]/Cu/[Co/Ni] GMR stack deposited on a \(\mathrm{Si}_{3}\mathrm{N}_{4}\) membrane (for fabrication details, please see Methods). Underneath the schematic we show two optical microscopy images taken from opposite directions to highlight the optical transmission of the \(\mathrm{Si}_{3}\mathrm{N}_{4}\) membrane. In Fig. 1(d) we show the magnetic properties of the individual free and fixed layers based on calibration samples, and their combined behavior in full STNO stacks in Fig. 1(e). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 624]]<|/det|> +Figure 2 presents the resistance and microwave signal as a function of field for an applied current of \(- 5\mathrm{mA}\) . The field is first increased from \(- 0.51\mathrm{T}\) to \(0.51\mathrm{T}\) in Fig. 2(a) and then decreased from positive to negative field in Fig. 2(b). At large negative fields, the STNO is in its lowest resistance state, consistent with a parallel (P) relative orientation of its free and fixed layers. At about \(- 0.49\mathrm{T}\) , the resistance increases about \(20\mathrm{mOhm}\) in a step- like fashion and there is a slight increase in the microwave noise background, both strong indications of the nucleation of a droplet. At about \(- 0.38\mathrm{T}\) , there is a second step- like increase in the STNO resistance and a marked further increase in the microwave noise. We interpret this as a transition into a larger droplet as the opposing applied field is reduced. At yet lower fields the droplet continues to grow in size (the STNO resistance increases), while its stability seems to deteriorate as indicated by the growing intensity of the microwave noise background. At about \(- 0.04\mathrm{T}\) , the microwave noise rapidly reaches a maximum and then suddenly disappears altogether, while the resistance exhibits a small jump of about \(5\mathrm{mOhm}\) . The complete microwave silence indicates that the magnetic state is now static, and we are lead to conclude that the droplet precession has stopped entirely and that the droplet has transitioned into a nanobubble state. + +<|ref|>text<|/ref|><|det|>[[113, 659, 885, 861]]<|/det|> +The nanobubble resistance exhibits jumps reminiscent of Barkhausen noise44–46, indicating pinning possibly at grain boundaries or defects of the sputtered film. When the field is further increased, the bubble resistance increases gradually, indicating a continued growth of its size. At about \(0.06\mathrm{T}\) , the entire free layer switches its magnetization direction and the antiparallel (AP) state is clearly identifiable in the resistance. When the fixed layer switches at \(0.23\mathrm{T}\) , a droplet is immediately nucleated. With further increasing of the opposing field, the droplet again shows + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 219]]<|/det|> +a gradual transition to a smaller size; the droplet finally disappears as the STNO transitions into a full P state at about 0.47 T. The overall behavior is very similar for decreasing fields (Fig.2b) where the same P/AP/droplet/nanobubble states can be clearly identified via the STNO resistance and the microwave noise. + +<|ref|>text<|/ref|><|det|>[[112, 258, 885, 681]]<|/det|> +As mentioned above, the microwave noise power is far from constant for the whole droplet region and peaks at certain fields. We identify these peaks as marks of mode hopping between different droplet states14. While the details of the spectrum is highly reproducible (cf. increasing and decreasing fields) and serves as a fingerprint for each device, the patterns at negative and positive fields are quite different. The magnetoresistance implies that a relatively small and stable droplet \((\mu_{0}H < - 0.4 \text{T})\) is abruptly followed by a larger but similarly stable mode. In contrast to the symmetric noise patterns around the droplet- to- droplet transitions, the strong increase in microwave noise power around the droplet- to- bubble transition is highly asymmetric. There is first an extended field region of monotonic increase in the noise, which is then abruptly cut off and replaced by a completely silent bubble state. This highlights the very different non- dynamical nature of the nanobubble and suggests that mode hopping out of the nanobubble state and back into a droplet state is negligible, once the nanobubble has formed. + +<|ref|>text<|/ref|><|det|>[[113, 718, 884, 848]]<|/det|> +Figure 2(c) and (d) demonstrate that it is possible to freeze the dynamic droplet into a static bubble and then thaw it back into a droplet using only the magnetic field under constant spin transfer torque. In particular, Fig. 2(d) shows how the nanobubble first is about to collapse at 0.025 T as it is getting squeezed by the opposing pressure from the increasing applied field. There + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[250, 95, 744, 725]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 734, 883, 861]]<|/det|> +
Figure 2: Microwave noise and STNO resistance vs. field. (a)-(d) Color plot of the power spectral density (PSD) of the microwave noise as a function of decreasing (a,c) and increasing (b,d) field, with the STNO resistance (white line) overlayed; the applied current is \(-5 \mathrm{mA}\) . (a,b) Wide field sweep covering full saturation at both positive and negative fields. P/AP indicate the parallel/antiparallel state of the STNO; red arrow indicates the droplet region, and green arrow the bubble region. (c,d) Minor field sweeps showing how the droplet/bubble transition is fully reversible.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 440]]<|/det|> +is some slight Barkhausen noise in the rapidly dropping resistance, but otherwise no measurable microwave noise. However, instead of switching to a P state, the resistance then exhibits a sharp minimum after which it shows a rapid increase, which is accompanied by a high level of microwave noise. The collapsing nanobubble is hence rescued by the stabilizing spin transfer torque, which sets the spins in the bubble perimeter into precessional motion and restores the full dynamics of a magnetic droplet. Judging from the resistance, it is noteworthy that the droplet is slightly larger than the smallest nanobubble. Within the experimental accuracy (a field step of \(2\mathrm{mT}\) ), we do not observe any significant hysteresis in this transition. Hence there is a negligible energy barrier between the two states and the bubble can indeed be viewed as a zero- frequency droplet, albeit still likely affected by pinning. + +<|ref|>text<|/ref|><|det|>[[112, 477, 886, 863]]<|/det|> +Figure 3(a) presents a phase diagram based on a two- dimensional map of the STNO resistance as functions of current and field. All data was acquired in a decreasing field at a constant current level. The parallel (P) and antiparallel (AP) configurations are easily identified by the dark blue and dark red colors, respectively, and for current magnitudes below \(1.8\mathrm{mA}\) , these are the only two available states, as expected for a GMR device. However, even at these weak currents, the \(\mathrm{P} \rightarrow \mathrm{AP}\) switching field is clearly affected by the STT from the nanocontact; in contrast, the \(\mathrm{AP} \rightarrow \mathrm{P}\) switching field is entirely unaffected. In an intermediate current region, from about \(- 1.8\) to \(- 3.5\mathrm{mA}\) , the STT can not yet sustain a droplet but is sufficient to create a nanobubble directly from the P state. As magnetic switching typically involves both domain nucleation and domain propagation, we interpret this current dependent switching in the following way (see Supplementary Materials for a zoom- in of this particular part of the phase diagram). For current magnitudes + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 885, 440]]<|/det|> +below 1 mA, magnetic switching is limited by the field required for domain nucleation and, in addition, the location of initial domain nucleation is far from the nanocontact region as STT from the current has no discernible impact. However, for current magnitudes above 1 mA, where we observe a strong current dependence of the switching field, we conclude that the domain nucleation has moved to underneath the nanocontact. If we reduce the field magnitude, we need a stronger current to assist in the domain nucleation, but once formed, it propagates through the entire free layer. However, at fields weaker than the field needed for domain propagation, i.e. the pinning field, which we read out as about \(60~\mathrm{mT}\) , the nucleated domain is no longer able to propagate and instead remains as a nanobubble directly underneath the nanocontact. The nanobubble can hence form either from the P state or from a droplet. + +<|ref|>text<|/ref|><|det|>[[113, 477, 885, 715]]<|/det|> +The droplet shows two discernable states, a high- field/low- current mode that exhibits a rather small MR (light blue). This mode moves to higher fields with increasing current and is no longer visible above \(\approx - 6 \mathrm{mA}\) . The other distinguishable droplet mode is characterized by an intermediate resistance (green- yellow). The bubble is almost indiscernible from the latter droplet state, even though a subtle line traces out the transition between the two. Moreover, the bubble resistance is not a smooth function of applied field, but displays notches and steps, indicative of Barkhausen noise due to pinning. + +<|ref|>text<|/ref|><|det|>[[113, 754, 884, 846]]<|/det|> +In contrast to their almost identical resistance, a stark difference between the droplet and the bubble is uncovered in Fig. 3(b), where we show the microwave signal integrated over 0–0.5 GHz. The droplet exhibits non- zero power levels of low frequency microwave noise, while the P, AP, + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 477]]<|/det|> +and bubble states are definitely static and silent. Figure 3(b) also further unveils the complex relation between the applied field and current, and the particular droplet characteristics. A strong microwave noise signal denotes mode hopping and these events exhibit a strong dependence on both field and current. We can identify three traces of mode hopping for positive fields, while there is only two weak trails at negative fields. There is also regions where the droplet is very stable and the noise level is almost zero. These features act like fingerprints for each measured device and are highly reproducible in consecutive measurements, but differ between STNOs. We then overlay the microwave noise data onto the resistance data, now plotted with a gray scale that highlights intermediate resistance levels (Fig. 3(c)). Parts of the low- field/low- current droplet regime (light blue in Fig. 3(a)) does not exhibit any measurable microwave noise. It is possible that its dynamics is on a slower time scale than the microwave frequencies our set- up is sensitive to. + +<|ref|>text<|/ref|><|det|>[[112, 513, 886, 862]]<|/det|> +We finally turn to the results of the scanning transmission X- ray microscopy results, illustrated in Fig. 4. Images of the droplet/bubble are shown in Fig. 4(a)- (f), and the corresponding magnetoresistance and microwave signal are presented in Fig. 4(g) with the matching field of the images marked by their letter. The STXM and the electrical measurements were performed in separate setups, hence there is a small uncertainty in comparing the field values of the two, although both measurements seem highly consistent with each other. The dashed white or black circles mark the position of the nanocontact. It has been placed by assuming that the droplet/bubble in Fig. 4(d) is centered under the NC and by comparing the non- magnetic contrast of the different images. The method works very well as confirmed by the good overlap of the perimeters in the inset of Fig. 4(g), but it should be remembered that the absolute position is still based on this assumption. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[112, 270, 880, 495]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 504, 883, 687]]<|/det|> +
Figure 3: Phase diagrams based on the resistance and the microwave noise. (a) STNO resistance and (b) integrated (0–0.5 GHz) microwave noise level as a function of field and current. (c) shows the noise level in (b) overlaid on the resistance in (a) displayed using a gray scale highlighting intermediate resistance levels indicative of droplets/bubbles. The dash-dotted black line corresponds to the field-sweep at \(I = -5 \mathrm{mA}\) given in Fig. 2. The parallel (P) and antiparallel (AP) states are easily discernible in the MR-map (a) as dark blue and dark red, while both the droplet and the bubble are characterized by intermediate resistance in green–yellow. The stark difference between the droplet and the bubble is revealed in the noise spectrum (b), where the stability of the bubble is manifested. Note however that the light-blue flanges in (a) correspond to a different droplet regime not captured in the microwave signal presented in (b).
+ +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[232, 95, 760, 707]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 716, 883, 863]]<|/det|> +
Figure 4: Scanning Transmission X-ray Microscopy (STXM). (a)–(f) STXM images of the nanocontact region vs. decreasing field for a current of -7 mA. Blue corresponds to magnetization alligned with the applied field, red corresponds to magnetization anti-alligned with the applied field, whereas white indicates in-plane magnetization. The STNO resistance and the microwave noise PSD vs. decreasing field are shown in (g) where the points corresponding to the STXM images have been labelled a–f. The inset in (g) highlights the perimeter of the droplet/bubble as the applied field is decreased from 200 mT (dark blue) to 60 mT (blue), and further reduced to -40 mT (red) in steps of 20 mT.
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 699]]<|/det|> +Figure 4(a) is measured at \(270~\mathrm{mT}\) , and shows a mode associated with a high noise level in Fig. 4(g). Only a weak and mostly white feature is captured in the STXM image. STXM measures a time averaged image and the droplet is in this highly noisy regime expected to experience large drift instabilities and continuously vanish and renucleate underneath the nanocontact. As a consequence, only a washed out and poorly reversed feature results. In contrast, Fig. 4(b)-(d) display more stable and more clearly reversed droplets. They have approximately the same radius as the nanocontact, although the size definitely increases slightly with decreasing field, as expected. We have in earlier STXM work observed a significant effect of the Zhang- Li torque on the droplet size \(^{21,47}\) . The magnitude of this effect depends on the current density \((j_{\mathrm{dc}})\) and we have performed simulations which confirm that the difference between the droplet diameter here and in our former publication is indeed due to a weaker \(j_{\mathrm{dc}}\) . At zero field, a bubble is clearly formed and it prevails down to \(- 40~\mathrm{mT}\) (Fig. 4(e)-(f)). It is no longer centered on the nanocontact, but has mostly expanded in one direction. It should be noted though, that the field in the microscope is given by rotating permanent magnets and the sample may have been subjected to in-plane fields between two set values. Nevertheless, the inset in Fig. 4(g) presents the perimeter of the droplet/bubble as the field decreases from \(200~\mathrm{mT}\) (dark blue) to \(- 40~\mathrm{mT}\) (red), and the initial bubble at \(40~\mathrm{mT}\) (light blue) grows in distinct steps, which implies that the size is controlled by pinning. + +<|ref|>text<|/ref|><|det|>[[113, 731, 888, 860]]<|/det|> +Returning to the original droplet theory of Hoefer et al, we note that pinning was not included. \(^{2}\) It is clear from our experimental observations that pinning has a strong and immediate impact on the relation between droplets and nanobubbles and must be included in the low- field/low- current regime. Instead of exhibiting a continuous slow- down and frequency decrease to zero with de + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 885, 256]]<|/det|> +creasing field, there is a minimum droplet precession frequency that spin transfer torque can sustain before pinning overcomes the precession. As this minimum frequency is approached from above, the broad- band microwave noise diverges as the competition between the inertia of the precession and the pinning makes the droplet dynamics increasingly erratic until pinning finally gets complete control of the precession abruptly stops, leaving complete microwave silence in its wake. + +<|ref|>text<|/ref|><|det|>[[123, 301, 884, 357]]<|/det|> +1. Ivanov, B. & Kosevich, A. Bound states of a large number of magnons in a ferromagnet with a single-ion anisotropy. Zh. Eksp. Teor. Fiz. 72, 2000 (1977). + +<|ref|>text<|/ref|><|det|>[[123, 386, 883, 442]]<|/det|> +2. Hoefer, M. A., Silva, T. J. & Keller, M. W. Theory for a dissipative droplet soliton excited by a spin torque nanocontact. Phys. Rev. B 82, 054432 (2010). + +<|ref|>text<|/ref|><|det|>[[123, 472, 884, 529]]<|/det|> +3. Mohseni, S. M. et al. Spin Torque-Generated Magnetic Droplet Solitons. Science 339, 1295-1298 (2013). + +<|ref|>text<|/ref|><|det|>[[123, 559, 882, 614]]<|/det|> +4. Mohseni, S. et al. Magnetic droplet solitons in orthogonal nano-contact spin torque oscillators. Physica B 435, 84-87 (2014). + +<|ref|>text<|/ref|><|det|>[[123, 645, 882, 701]]<|/det|> +5. Iacocca, E. et al. Confined dissipative droplet solitons in spin-valve nanowires with perpendicular magnetic anisotropy. Phys. Rev. Lett. 112, 047201 (2014). + +<|ref|>text<|/ref|><|det|>[[123, 731, 882, 786]]<|/det|> +6. Macia, F., Backes, D. & Kent, A. D. Stable magnetic droplet solitons in spin-transfer nanocontacts. Nat. Nanotechnol 9, 992-996 (2014). + +<|ref|>text<|/ref|><|det|>[[123, 817, 882, 872]]<|/det|> +7. Chung, S. et al. Spin transfer torque generated magnetic droplet solitons (invited). J. Appl. Phys. 115, 172612 (2014). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[122, 89, 883, 182]]<|/det|> +8. Lendinez, S., Statuto, N., Backes, D., Kent, A. D. & Macia, F. Observation of droplet soliton drift resonances in a spin-transfer-torque nanocontact to a ferromagnetic thin film. Phys. Rev. B 92, 174426 (2015). + +<|ref|>text<|/ref|><|det|>[[122, 220, 883, 276]]<|/det|> +9. Chung, S. et al. Magnetic droplet solitons in orthogonal spin valves. Low Temp. Phys. 41, 833-837 (2015). + +<|ref|>text<|/ref|><|det|>[[115, 314, 882, 370]]<|/det|> +10. Chung, S. et al. Magnetic droplet nucleation boundary in orthogonal spin-torque nano-oscillators. Nat. Commun. 7, 11209 (2016). + +<|ref|>text<|/ref|><|det|>[[115, 407, 881, 464]]<|/det|> +11. Xiao, D. et al. Parametric autoexcitation of magnetic droplet soliton perimeter modes. Phys. Rev. B 95, 024106 (2017). + +<|ref|>text<|/ref|><|det|>[[115, 501, 883, 558]]<|/det|> +12. Lendinez, S. et al. Effect of Temperature on Magnetic Solitons Induced by Spin-Transfer Torque. Phys. Rev. Appl. 7, 054027 (2017). + +<|ref|>text<|/ref|><|det|>[[115, 594, 883, 687]]<|/det|> +13. Sulymenko, O., Prokopenko, O., Tyberkevych, V., Slavin, A. & Serga, A. Bullets and droplets: Two-dimensional spin-wave solitons in modern magnonics (Review Article). Low Temp. Phys. 44, 775 (2018). + +<|ref|>text<|/ref|><|det|>[[115, 724, 883, 781]]<|/det|> +14. Statuto, N., Hahn, C., Hernandez, J. M., Kent, A. D. & Macia, F. Multiple magnetic droplet soliton modes. Phys. Rev. B 99, 174436 (2019). + +<|ref|>text<|/ref|><|det|>[[115, 818, 881, 874]]<|/det|> +15. Mohseni, M. et al. Chiral excitations of magnetic droplet solitons driven by their own inertia. Phys. Rev. B 101, 20417 (2020). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[115, 89, 883, 146]]<|/det|> +16. Divinskiy, B. et al. Magnetic droplet solitons generated by pure spin currents. Phys. Rev. B 96, 224419 (2017). + +<|ref|>text<|/ref|><|det|>[[115, 176, 881, 233]]<|/det|> +17. Mohseni, S. M. et al. High frequency operation of a spin-torque oscillator at low field. Phys. Status Solidi RRL 5, 432-434 (2011). + +<|ref|>text<|/ref|><|det|>[[115, 262, 883, 319]]<|/det|> +18. Rippard, W. H. et al. Spin-transfer dynamics in spin valves with out-of-plane magnetized CoNi free layers. Phys. Rev. B 81, 014426 (2010). + +<|ref|>text<|/ref|><|det|>[[115, 348, 881, 405]]<|/det|> +19. Bookman, L. D. & Hoefer, M. A. Analytical theory of modulated magnetic solitons. Phys. Rev. B 88, 184401 (2013). + +<|ref|>text<|/ref|><|det|>[[113, 434, 872, 455]]<|/det|> +20. Locatelli, N., Cros, V. & Grollier, J. Spin-torque building blocks. Nat. Mater. 13, 11 (2014). + +<|ref|>text<|/ref|><|det|>[[114, 485, 883, 541]]<|/det|> +21. Chung, S. et al. Direct Observation of Zhang-Li Torque Expansion of Magnetic Droplet Solitons. Phys. Rev. Lett. 120, 217204 (2018). + +<|ref|>text<|/ref|><|det|>[[114, 570, 883, 664]]<|/det|> +22. Okamoto, S., Kikuchi, N., Furuta, M., Kitakami, O. & Shimatsu, T. Microwave assisted magnetic recording technologies and related physics. J. Phys. D: Appl. Phys. 48, 353001 (2015). + +<|ref|>text<|/ref|><|det|>[[114, 694, 883, 788]]<|/det|> +23. Bosu, S. et al. High frequency out-of-plane oscillation with large cone angle in mag-flip spin torque oscillators for microwave assisted magnetic recording. Appl. Phys. Lett. 110, 142403 (2017). + +<|ref|>text<|/ref|><|det|>[[114, 817, 883, 874]]<|/det|> +24. Torrejon, J. et al. Neuromorphic computing with nanoscale spintronic oscillators. Nature 547, 428-431 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 884, 147]]<|/det|> +25. Romera, M. et al. Vowel recognition with four coupled spin-torque nano-oscillators. Nature 563, 230-234 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 175, 830, 197]]<|/det|> +26. Macià, F. & Kent, A. D. Magnetic droplet solitons. J. Appl. Phys. 128, 100901 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 223, 883, 280]]<|/det|> +27. Maiden, M. D., Bookman, L. D. & Hoefer, M. a. Attraction, merger, reflection, and annihilation in magnetic droplet soliton scattering. Phys. Rev. B 89, 180409 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 307, 883, 364]]<|/det|> +28. Puliafito, V., Siracusano, G., Azzerboni, B. & Finocchio, G. Self-modulated soliton modes excited in a nanocontact spin-torque oscillator. IEEE Magn. Lett. 5, 3000104 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 392, 883, 450]]<|/det|> +29. Wang, C., Xiao, D., Zhou, Y., Akerman, J. & Liu, Y. Phase-locking of multiple magnetic droplets by a microwave magnetic field. AIP Adv. 7, 56019 (2017). + +<|ref|>text<|/ref|><|det|>[[113, 477, 883, 534]]<|/det|> +30. Mohseni, M. et al. Magnetic droplet soliton nucleation in oblique fields. Phys. Rev. B 97, 184402 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 562, 883, 620]]<|/det|> +31. Mohseni, M. et al. Propagating Magnetic Droplet Solitons as Moveable Nanoscale Spin-Wave Sources with Tunable Direction of Emission. Phys. Rev. Appl. 13, 24040 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 647, 883, 705]]<|/det|> +32. Sisodia, N., Muduli, P. K., Papanicolaou, N. & Komineas, S. Chiral droplets and current-driven motion in ferromagnets. Phys. Rev. B 103, 24431 (2021). + +<|ref|>text<|/ref|><|det|>[[113, 732, 883, 789]]<|/det|> +33. Yazdi, H. F., Ghasemi, G., Mohseni, M. & Mohseni, M. Tuning the dynamics of magnetic droplet solitons using dipolar interactions. Phys. Rev. B 103, 24441 (2021). + +<|ref|>text<|/ref|><|det|>[[113, 817, 883, 874]]<|/det|> +34. Backes, D. et al. Direct Observation of a Localized Magnetic Soliton in a Spin-Transfer Nanocontact. Phys. Rev. Lett. 115, 127205 (2015). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 884, 184]]<|/det|> +35. Carpentieri, M., Tomasello, R., Zivieri, R. & Finocchio, G. Topological, non-topological and instanton droplets driven by spin-transfer torque in materials with perpendicular magnetic anisotropy and Dzyaloshinskii-Moriya Interaction. Sci. Rep. 5, 16184 (2015). + +<|ref|>text<|/ref|><|det|>[[113, 210, 884, 268]]<|/det|> +36. Hang, J., Hahn, C., Statuto, N., Macia, F. & Kent, A. D. Generation and annihilation time of magnetic droplet solitons. Sci. Rep. 8, 6847 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 295, 881, 353]]<|/det|> +37. Jiang, S. et al. Impact of the Oersted Field on Droplet Nucleation Boundaries. IEEE Magn. Lett. 9, 3104304 (2018). + +<|ref|>text<|/ref|><|det|>[[113, 380, 882, 438]]<|/det|> +38. Shi, K. et al. Observation of Magnetic Droplets in Magnetic Tunnel Junctions. arXiv: 2012.05596 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 465, 835, 486]]<|/det|> +39. Nielsen, J. Bubble domain memory materials. IEEE Trans. Magn. 12, 327-345 (1976). + +<|ref|>text<|/ref|><|det|>[[113, 514, 712, 534]]<|/det|> +40. Giess, E. A. Magnetic Bubble Materials. Science 208, 938-943 (1980). + +<|ref|>text<|/ref|><|det|>[[113, 562, 884, 620]]<|/det|> +41. De Leeuw, F., Van Den Doel, R. & Enz, U. Dynamic properties of magnetic domain walls and magnetic bubbles. Rep. Prog. Phys. 43, 689 (1980). + +<|ref|>text<|/ref|><|det|>[[113, 648, 883, 705]]<|/det|> +42. Komineas, S. & Papanicolaou, N. Topology and dynamics in ferromagnetic media. Physica D 99, 81-107 (1996). + +<|ref|>text<|/ref|><|det|>[[113, 732, 883, 789]]<|/det|> +43. Moutafis, C., Komineas, S. & Bland, J. A. C. Dynamics and switching processes for magnetic bubbles in nanoelements. Phys. Rev. B 79, 224429 (2009). + +<|ref|>text<|/ref|><|det|>[[113, 817, 883, 874]]<|/det|> +44. Kim, D.-H., Choe, S.-B. & Shin, S.-C. Direct Observation of Barkhausen Avalanche in Co Thin Films. Phys. Rev. Lett. 90, 087203 (2003). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 883, 147]]<|/det|> +45. Balk, A. L., Stiles, M. D. & Unguris, J. Critical behavior of zero-field magnetic fluctuations in perpendicularly magnetized thin films. Phys. Rev. B 90, 184404 (2014). + +<|ref|>text<|/ref|><|det|>[[113, 176, 881, 233]]<|/det|> +46. Herranen, T. & Laurson, L. Barkhausen Noise from Precessional Domain Wall Motion. Phys. Rev. Lett. 122, 117205 (2019). + +<|ref|>text<|/ref|><|det|>[[113, 262, 881, 319]]<|/det|> +47. Albert, J., Macia, F. & Hernandez, J. M. Effect of the Zhang-Li torque on spin-torque nano-oscillators. Phys. Rev. B 102, 184421 (2020). + +<|ref|>text<|/ref|><|det|>[[113, 347, 883, 479]]<|/det|> +48. J. Grafe, M. Weigand, B. Van Waeyenberge, A. Gangwar, F. GroB, F. Lisiecki, J. Rychly, H. Stoll, N. Trager, J. Forster, F. Stobiecki, J. Dubowik, J. Klos, M. Krawczyk, C. H. Back, E. J. Goering, G. Schutz, H.-J. M. Drouhin, J.-E. Wegrowe, and M. Razeghi. Visualizing nanoscale spin waves using MAXYMUS. In Proc. SPIE, vol. 11090, 1109025 (Spintronics XII, 2019). + +<|ref|>text<|/ref|><|det|>[[113, 506, 883, 600]]<|/det|> +49. Nolle, D. et al. Note: Unique characterization possibilities in the ultra high vacuum scanning transmission x-ray microscope (UHV-STXM) "MAXYMUS" using a rotatable permanent magnetic field up to 0.22 T. Rev. Sci. Instrum. 83, 046112 (2012). + +<|ref|>sub_title<|/ref|><|det|>[[114, 656, 191, 674]]<|/det|> +## Methods + +<|ref|>text<|/ref|><|det|>[[112, 708, 884, 875]]<|/det|> +Sample Preparation A sample stack is consisted of a Ta (4 nm)/ Cu (14 nm) / Ta (4 nm) / Pd (2 nm) seed layer and an all-perpendicular pseudo-spin valve [Co (0.35 nm) / Pd (0.7 nm)] \(\times 5\) / Co (0.35 nm) / Cu (5 nm) / [Co (0.22 nm) / Ni (0.68 nm)] \(\times 4\) / Co (0.22 nm), capped by a Cu (2 nm) / Pd (2 nm) layer, which was deposited by magnetron sputtering on Si wafer with 300 nm thick LPCVD silicon nitride layer. Using a conventional photo-lithography and metal-etching tech + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 886, 440]]<|/det|> +niques, \(8 \mu \mathrm{m} \times 16 \mu \mathrm{m}\) mesas were fabricated on above stack wafer and all mesas were insulated by a 30- nm- thick \(\mathrm{SiO}_2\) layer deposited by using chemical vapor deposition (CVD). To pattern nanocontacts (NCs) on the top of each mesa having different diameters from 50 to \(150 \mathrm{nm}\) , electron beam lithography was used. \(\mathrm{SiO}_2\) layer was then etched through by the reactive ion etching (RIE) technique to open NCs. The NC- STO device fabrication was completed by the deposition of \(\mathrm{Cu} 200 \mathrm{nm} / \mathrm{Au} 100 \mathrm{nm}\) top electrode and lift- off processing. For STXM measurements, Si was removed from backside using highly selective RIE process and leave only SiN membrane to allow X- ray transmission underneath NC- STOs. (See, Figure 1(c)) For magnetic and electrical charaterization of NC- STOs, same stack was prepared on Si thermally oxidized Si wafer and then similar fabrication processing were done except a deep etching for a membrane structure. + +<|ref|>text<|/ref|><|det|>[[112, 469, 886, 782]]<|/det|> +Magnetic and Electrical Characterization the magnetization hysteresis loops was measured using Alternating Gradient Magnetometry (AGM) with the unpatterned material stacks. \(dc\) and microwave measurements of the fabricated STOs were carried out using our custom- built setup, where magnetic field strength, polarity, and angle can be controlled. A magnetic field between - 0.5 to \(+0.5\) can be manipulated using electromagnet. The device is connected using GSG probe to a \(dc\) - current source (Keithley 6221), a nanovoltmeter (Keithley 2182A), and a spectrum analyzer (R & S FSQ26). A 0- 40GHz bias- tee is used to separate the bias input and the generated microwave signal. The microwave sinalg is amplified by a low- noise amplifier (operational range: 0.1- 26.5 GHz) before being sent to the spectrum analyzer. + +<|ref|>text<|/ref|><|det|>[[113, 812, 901, 867]]<|/det|> +Scanning transmission x- ray microscopy The STXM measurements were performed at the BESSY II synchrotron, using the MPI IS operated MAXYMUS end station at the UE46- PGM2 beam line. \(^{48}\) + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 89, 885, 256]]<|/det|> +The out- of- plane component of the magnetization was probed using circularly polarized light at normal incidence. The applied field, with a maximum value of \(300\mathrm{mT}\) , was generated by a set of four rotatable permanent magnets49. An optimal XMCD contrast was achieved by setting the photon energy to the Ni \(L_{3}\) edge, which resulted in clear images. The size of each pixel is \(10 \times 10 \mathrm{nm}^{2}\) , while the nominal resolution of the focusing plate is \(18\mathrm{nm}\) . + +<|ref|>text<|/ref|><|det|>[[113, 308, 885, 499]]<|/det|> +Acknowledgements This work was supported by the Swedish Research Council (VR; 2017- 06711 and 2019- 04229). Helmholtz Zentrum Berlin is acknowledged for allocating beam time at the BESSY II synchrotron radiation facility. M. W., E.G., G.S. and J.G. acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the frame work of DynaMAX (Project No. 05K18EYA). This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. 2020R1F1A1049642) + +<|ref|>text<|/ref|><|det|>[[115, 535, 793, 554]]<|/det|> +Acknowledgements These authors contributed equally: Martina Ahlberg and Sunjae Chung. + +<|ref|>text<|/ref|><|det|>[[113, 589, 884, 710]]<|/det|> +Contributions M.A., S.C., and J.A. conceived the project, S.C. S. J. and T.Q.L. performed the electrical measurements. S.C., T.Q.L., S.J., and A.H. fabricated the devices. M.A., S. J., J.G., M.W., F.G. and I.B. carried out the STXM measurements. J.A. coordinated the project. All authors analyzed the results and co- wrote the manuscript. + +<|ref|>text<|/ref|><|det|>[[113, 747, 770, 766]]<|/det|> +Competing Interests The authors declare that they have no competing financial interests. + +<|ref|>text<|/ref|><|det|>[[113, 802, 883, 854]]<|/det|> +Correspondence Correspondence and requests for materials should be addressed to S. Chung (email: sjchung76@knue.ac.kr) and J. Akerman (email: johan.akerman@physics.gu.se). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 91, 355, 109]]<|/det|> +Supplementary information. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[149, 179, 832, 625]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 636, 883, 764]]<|/det|> +
Figure S1: Supplementary Fig.1. A zoom-in of the phase diagram in Fig. 3(a) of the main text. The color code represents the different states: droplet (green), bubble (gray) and antiparallel (red). A droplet is nucleated at high currents and fields. Below a certain positive field \((H_{\mathrm{balance}})\) the droplet is stabilized as a static bubble due to magnetostatic effects. The bubble is pinned below the nanocontact until the negative field is high enough to let the bubble domain expand throughout the film at \(H_{\mathrm{pinning}}\) . At low currents the magnetic switching is only governed by the coercive field \((H_{c})\) of the free layer.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 44, 143, 68]]<|/det|> +## Figures + +<|ref|>image<|/ref|><|det|>[[42, 95, 550, 831]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 850, 115, 869]]<|/det|> +
Figure 1
+ +<|ref|>text<|/ref|><|det|>[[42, 891, 936, 956]]<|/det|> +Droplet vs. bubble, device structure and layout, and magnetic characterization. (a) Schematic of dynamical magnetic droplet soliton. (b) Schematic of a static magnetic bubble. (c) Schematic of an all-perpendicular STNO composed of [Co/Pd] (fixed) and [Co/Ni] (free) multilayers with a Cu spacer + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 949, 111]]<|/det|> +fabricated on a SiN membrane structure. The insets underneath show optical micrographs of the SiN membrane areas through which the different metal layers of the device can be seen. (d) Hysteresis loops of single Co/Pd and Co/Ni layers. (e) Hysteresis loop of a full [Co/Pd]/Cu/[Co/Ni] stack. + +<|ref|>image<|/ref|><|det|>[[44, 110, 620, 850]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[44, 870, 117, 890]]<|/det|> +
Figure 2
+ +<|ref|>text<|/ref|><|det|>[[42, 911, 953, 954]]<|/det|> +Microwave noise and STNO resistance vs. field. (a)-(d) Color plot of the power spectral density (PSD) of the microwave noise as a function of decreasing (a,c) and increasing (b,d) field, with the STNO resistance + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[42, 45, 950, 134]]<|/det|> +(white line) overlayed; the applied current is 5 mA. (a,b) Wide field sweep covering full saturation at both positive and negative fields. P/AP indicate the parallel/antiparallel state of the STNO; red arrow indicates the droplet region, and green arrow the bubble region. (c,d) Minor field sweeps showing how the droplet/bubble transition is fully reversible. + +<|ref|>image<|/ref|><|det|>[[42, 137, 951, 400]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 419, 117, 439]]<|/det|> +
Figure 3
+ +<|ref|>text<|/ref|><|det|>[[40, 461, 955, 664]]<|/det|> +Phase diagrams based on the resistance and the microwave noise. (a) STNO resistance and (b) integrated (0–0.5 GHz) microwave noise level as a function of field and current. (c) shows the noise level in (b) overlaid on the resistance in (a) displayed using a gray scale highlighting intermediate resistance levels indicative of droplets/bubbles. The dash-dotted black line corresponds to the field-sweep at \(I = - 5\) mA given in Fig. 2. The parallel (P) and antiparallel (AP) states are easily discernible in the MR-map (a) as dark blue and dark red, while both the droplet and the bubble are characterized by intermediate resistance in green–yellow. The stark difference between the droplet and the bubble is revealed in the noise spectrum (b), where the stability of the bubble is manifested. Note however that the light-blue flanges in (a) correspond to a different droplet regime not captured in the microwave signal presented in (b). + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[50, 45, 584, 666]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[42, 685, 116, 704]]<|/det|> +
Figure 4
+ +<|ref|>text<|/ref|><|det|>[[41, 725, 944, 883]]<|/det|> +Scanning Transmission X- ray Microscopy (STXM). (a)–(f) STXM images of the nanocontact region vs. decreasing field for a current of - 7 mA. Blue corresponds to magnetization alligned with the applied field, red corresponds to magnetization anti- alligned with the applied field, whereas white indicates in- plane magnetization. The STNO resistance and the microwave noise PSD vs. decreasing field are shown in (g) where the points corresponding to the STXM images have been labelled a–f. The inset in (g) highlights the perimeter of the droplet/bubble as the applied field is decreased from 200 mT (dark blue) to 60 mT (blue), and further reduced to - 40 mT (red) in steps of 20 mT. + +<|ref|>sub_title<|/ref|><|det|>[[44, 905, 310, 932]]<|/det|> +## Supplementary Files + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[44, 45, 765, 65]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 84, 240, 103]]<|/det|> +- FigureSupMat.jpg + +<--- Page Split ---> diff --git a/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/images_list.json b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..4b9da3458cd1ef07bf89874e3965f2b02af0c1d7 --- /dev/null +++ b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/images_list.json @@ -0,0 +1,62 @@ +[ + { + "type": "image", + "img_path": "images/Figure_1.jpg", + "caption": "Fig. 1| Ultrastrong coupling in \\(\\mathrm{WS}_2\\) monolayer with a random multi-singular plasmonic metasurface. a, Schematic of a gold multi-singular metasurface with a dense array of nanometre-sized plasmonic gaps. Insets: left: schematic of a \\(\\mathrm{WS}_2\\) monolayer integrated with the multi-singular metasurface where the \\(\\mathrm{WS}_2\\) monolayer was mechanically exfoliated onto a PDMS tape and transferred onto the gold nanopattern using a dry transfer method; middle and right: two main paths for generating nanometre-sized gaps. Middle: the first path, a nanometre-sized crack", + "footnote": [], + "bbox": [ + [ + 120, + 401, + 868, + 735 + ] + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/Figure_2.jpg", + "caption": "Fig. 2| Ultrastrong exciton-plasmon coupling in WS \\(_2\\) monolayer. Dark-field scattering spectra under a 0 and b -2% uniaxial strain on the metasurface over different gold film thicknesses (from bottom to top, the gold film thickness \\(\\mathrm{(t_{Au})}\\) increases from 17 nm to 24 nm). The scattering splits into lower and upper polariton branches, exhibiting level anticrossing. The vertical dashed lines refer to the WS \\(_2\\) exciton energy \\(\\omega_{ex} = 2.019 \\mathrm{eV}\\) . The right panels are calculated scattering spectra using the model of two coupled oscillators \\(^{32}\\) . c, Dispersion plots of the measured dark-field scattering spectra. The lower \\((\\omega_{-})\\) and upper \\((\\omega_{+})\\) polariton branches are extracted from the scattering spectra in a,b and fitted (solid lines) with a coupling strength of \\(165.9 \\mathrm{meV}\\) at 0 strain (blue lines) and \\(240.4 \\mathrm{meV}\\) at -2% strain (purple lines) in the full Hopfield Hamiltonian. The experimental spectral peaks are shown as triangles (0 strain) and circles (-2% strain). The exciton energy \\((\\omega_{ex})\\) is shown as the horizontal dashed black line. The plasmonic mode energy \\((\\omega_{pl})\\) is shown as the diagonal dashed blue (SC) and purple (USC) lines. The plasmonic response shifts to higher energy when compressive strain is applied, and the local field enhancement is largely enhanced due to the reduced gap size, and both effects lead to significant changes in the polaritonic dispersion. d, Normalized second", + "footnote": [], + "bbox": [ + [ + 155, + 87, + 845, + 528 + ] + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/Figure_3.jpg", + "caption": "Fig. 3| Tunable ultrastrong exciton-plasmon coupling in WS \\(_2\\) monolayer. a, Mechanically tunable plasmonic resonance (wavelength at \\(\\lambda_{pl}\\) ) and plasmonic enhancement in photoluminescence peak intensity of the strongly coupled plasmonic system. b, Normalized dark-field scattering intensity as a function of excitation energy and strain. The horizontal purple line marks the onset of USC. The black solid lines are the extracted polariton energies of the lower (LPB) and upper (UPB) polariton branches. The white dashed line is the WS \\(_2\\) exciton energy ( \\(\\omega_{ex} = 2.019 \\mathrm{eV}\\) ). c, Normalized coupling strength as a function of strain. The onset of USC is marked by the horizontal dashed line. d, Ground-state energy modification as a function of strain. The black dashed line is the calculation result from the coupling strength ( \\(g\\) ) and the polariton energies ( \\(\\omega_{+}\\) , \\(\\omega_{- }\\) ) from the fitting to the full Hopfield Hamiltonian. The absolute change in ground-state energy reaches 13.4 meV at -2% compressive strain. Error bars in c and d result from the variation in dark-field scattering from multiple batches of samples and are extracted from the standard error of the fit.", + "footnote": [], + "bbox": [ + [ + 220, + 88, + 772, + 465 + ] + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/Figure_4.jpg", + "caption": "Fig. 4| Ultrastrong coupling in multilayer WS2 flakes. a-d, Dark-field scattering spectra of a multilayer WS2 plasmonic system. Dashed white lines denote the exciton energy ( \\(\\omega_{ex} = 2.01 \\mathrm{eV}\\) for a,b; \\(\\omega_{ex} = 2.003 \\mathrm{eV}\\) for c,d;), and the diagonal dashed orange and purple lines indicate the plasmon energy ( \\(\\omega_{pl}\\) ). The open circles represent lower ( \\(\\omega_{-}\\) ) and upper ( \\(\\omega_{+} \\mathrm{p}\\) ) polariton energies obtained from the dark-field scattering spectra of individual plasmonic systems. The solid lines show the lower and upper polariton dispersions using the full Hopfield Hamiltonian. 3L and 4L represent the trilayer and quadrilayer cases. e, Normalized coupling strength ( \\(g / \\omega_{ex}\\) ) as a function of the WS2 layers under 0 (orange) and -2% compressive strain (purple). The horizontal dashed line marks the onset of USC. Error bars are derived from the variation in dark-field scattering from different batches of samples and are extracted from the standard error of the fit.", + "footnote": [], + "bbox": [ + [ + 172, + 88, + 822, + 457 + ] + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82.mmd b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82.mmd new file mode 100644 index 0000000000000000000000000000000000000000..e6bb4079c5ebe15cb9f1a52fb3afc9f31edc9075 --- /dev/null +++ b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82.mmd @@ -0,0 +1,425 @@ + +# Ultrastrong exciton-plasmon couplings in WS2 multilayers synthesized with a random multi- singular metasurface at room temperature + +Yu Luo + +luoyu@ntu.edu.sg + +Nanyang Technological University https://orcid.org/0000- 0003- 2925- 682X + +Tingting Wu Nanyang Technological University + +Chongwu Wang Nanyang Technological University https://orcid.org/0000- 0002- 8749- 196X + +Guangwei Hu Nanyang Technological University https://orcid.org/0000- 0002- 3023- 9632 + +Zhixun Wang Nanyang Technological University https://orcid.org/0000- 0001- 9918- 9939 + +Jiaxin Zhao + +Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University + +Zhe Wang School of Electrical and Electronic Engineering, Nanyang Technological University + +Ksenia Chaykun Nanyang Technological University + +Lin Liu Nanyang Technological University + +Mengxiao Chen https://orcid.org/0000- 0001- 5853- 4791 + +Dong Li School of Mechanical and Aerospace Engineering, College of Engineering, Nanyang Technological University https://orcid.org/0000- 0002- 4484- 8738 + +Qihua Xiong Tsinghua University https://orcid.org/0000- 0002- 2555- 4363 + +Ze Shen Nanyang Technological University https://orcid.org/0000- 0001- 7432- 7936 + +Huajian Gao + +<--- Page Split ---> + +Nanyang Technological University https://orcid.org/0000- 0002- 8656- 846X + +Francisco Garcia- Vidal Universidad Autónoma de Madrid https://orcid.org/0000- 0003- 4354- 0982 + +Lei Wei School of Electrical and Electronic Engineering, Nanyang Technological University https://orcid.org/0000- 0003- 0819- 8325 + +Qi jie Wang Nanyang Technological University https://orcid.org/0000- 0002- 9910- 1455 + +## Article + +# Keywords: + +Posted Date: October 6th, 2023 + +DOI: https://doi.org/10.21203/rs.3.rs- 3409617/v1 + +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +Additional Declarations: There is NO Competing Interest. + +Version of Record: A version of this preprint was published at Nature Communications on April 17th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 47610- z. + +<--- Page Split ---> + +# Ultrastrong exciton-plasmon couplings in WS₂ multilayers + +# synthesized with a random multi-singular metasurface at room + +## temperature + +Tingting \(\mathrm{Wu}^{1,\#}\) , Chongwu Wang \(^{1,\#}\) , Guangwei \(\mathrm{Hu}^{1,\#}\) , Zhixun Wang \(^{1}\) , Jiaxin Zhao \(^{2}\) , Zhe Wang \(^{1}\) , Ksenia Chaykun \(^{2}\) , Lin Liu \(^{1}\) , Mengxiao Chen \(^{3}\) , Dong Li \(^{4}\) , Qihua Xiong \(^{5}\) , Zexiang Shen \(^{2}\) , Huajian Gao \(^{4}\) , Francisco J. Garcia- Vidal \(^{6,7,*}\) , Lei Wei \(^{1,*}\) , Qi Jie Wang \(^{1,2,*}\) and Yu Luo \(^{1,*}\) + +\(^{1}\) School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, Singapore. + +\(^{2}\) School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore, Singapore. + +\(^{3}\) Zhejiang Provincial Key Laboratory of Cardio- Cerebral Vascular Detection Technology and Medicinal Effectiveness Appraisal, Zhejiang University, Hangzhou, China + +\(^{4}\) School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore, Singapore. + +\(^{5}\) State Key Laboratory of Low- Dimensional Quantum Physics and Department of Physics, Tsinghua University, Beijing, China. + +\(^{6}\) Departamento de Física Teorica de la Materia Condensada and Condensed Matter Physics Center (IFIMAC), Universidad Autónoma de Madrid, 28049 Madrid, Spain + +\(^{7}\) Institute of High Performance Computing, Agency for Science, Technology and Research (A\*STAR), Connexis, 138632, Singapore + +\*These authors contributed equally: Tingting Wu, Chongwu Wang, Guangwei Hu. + +\*emails: fj.garcia@uam.es; wei.lei@ntu.edu.sg; qjwang@ntu.edu.sg; luoyu@ntu.edu.sg + +## Abstract + +Van der Waals semiconductors exemplified by two- dimensional transition- metal dichalcogenides have promised next- generation atomically thin optoelectronics. Boosting their interaction with light is vital for practical + +<--- Page Split ---> + +applications, especially in the quantum regime where ultrastrong coupling is highly demanded but not yet realized. Here we report ultrastrong exciton- plasmon coupling at room temperature in tungsten disulfide ( \(\mathrm{WS}_2\) ) layers loaded with a random multi- singular plasmonic metasurface deposited on a flexible polymer substrate. Different from seeking perfect metals or high- quality resonators, we create a unique type of metasurface with a dense array of singularities that can support nanometre- sized plasmonic hotspots to which several \(\mathrm{WS}_2\) excitons coherently interact. The associated normalized coupling strength is 0.12 for monolayer \(\mathrm{WS}_2\) and can be up to 0.164 for quadrilayers, showcasing the ultrastrong exciton- plasmon coupling and important for practical optoelectronic devices based on low- dimensional semiconductors. + +## Introduction + +Exciton- polaritons owing to strong collective excitations of waves and electron- hole pairs, i.e., excitons have facilitated versatile applications in nonlinear optics, quantum photonics, condensed matter physics, and others. Various excitonic systems have been explored, including bulk III- V semiconductors and organic materials, which however either require cryogenic temperatures \(^{1,2}\) to avoid the ionization of excitons at high temperatures or are susceptible to bleaching effects \(^{3}\) . Recently, van der Waals transition- metal dichalcogenides (TMDs) have emerged as new candidates for stable and atomically thin excitonic systems, offering numerous advantages, including direct bandgaps at visible frequencies, large exciton binding energies, pronounced resonance strengths, narrow linewidths even beyond room temperature \(^{4,5}\) , and strongly reduced structural disorder. In addition, strong coupling (SC) between excitons in TMD monolayers and photonic resonances has been realized in all- dielectric microcavities with Fabry- Perot resonators \(^{6,7}\) , metal- based microcavities with reduced mode volumes \(^{8,9}\) , and plasmonic structures towards further enhancement of wave- matter couplings \(^{10,11}\) . + +In a parallel scientific endeavor, ultrastrong coupling (USC) of light- matter interactions has been recently explored in various systems. Here, the normalized coupling strength, defined as \(\eta = g / \omega_{ex}\) where \(g\) is coupling strength and \(\omega_{ex}\) is bare excitation energy, should be larger than 0.1. Compared to SC, USC is a new regime of quantum light- matter interactions with faster control/response even at shorter device lifetimes and important for several quantum optoelectronic applications. However, since its early proposal \(^{12,13}\) , USC has been only observed + +<--- Page Split ---> + +in semiconductor quantum wells14, superconducting circuits15, Landau polaritons16,17, organic molecules18, phonons19, and plasmons20, most of which rely on coupling many dipoles to a cavity mode at cryogenic temperatures, due to the technical challenges in implementation. Realizing USC under ambient conditions remains a challenge. Meanwhile, USC in monolayer TMDs has never been reached, which impedes their promise for ultrathin quantum devices21. This is because of the weak nanoscale matter excitations and the lack of significant transition dipole moments (typically in a few tens of Debyees) at such atomic thicknesses. + +Here, we report the first observation of USC between excitons in a TMD monolayer and surface plasmons in a random multi- singular flexible plasmonic metasurface at room temperature. The dense singular nanometre- sized gaps are created by the cold etching technique and support strong field concentration. Multiple \(\mathrm{WS}_2\) excitons can interact with these deep- nanoscale plasmonic hotspots, thus promoting ultrastrong exciton- plasmon couplings. We observe a normalized coupling strength of 0.082 in unstrained TMD monolayers, which, via strain engineering in the flexible substrate, can be further tuned over a wide range from 0.075 to 0.12, then entering the USC regime. Furthermore, by increasing the number of \(\mathrm{WS}_2\) layers, the exciton- plasmon interaction shows the increased normalized coupling strengths to 0.147 (trilayer) and 0.164 (quadrilayer) under strain. We believe that our reported strategy towards room- temperature USC in TMD multilayers could find applications in lasing, nonlinear optics, wearable optoelectronics, and other technologies in atomically thin platforms. + +## Results + +## Mechanism of USC in \(\mathbf{WS}_2\) multilayers loaded with multi-singular metasurface + +We start from discussing the fundamental ways to boost light- matter interaction. In general, in the case of collective ensembles, the light- matter coupling strength \(g\) is related to the emitter number \((g \propto \sqrt{N})\) and the highly confined cavity field in a very small mode volume \((g \propto 1 / \sqrt{V})^{22,23}\) . Progress towards room- temperature SC with monolayer TMDs in dielectric- based microcavities has been limited by the inevitable increase in the emitter scattering rate \((\gamma \propto k_B T)\) and by difficulties in reducing the cavity mode volume in dielectric structures due to the diffraction limit. To improve mode confinement with smaller mode volume, surface plasmons have been used23,24. Hence, the pronounced room- temperature plasmon- exciton coupling is observed in periodic + +<--- Page Split ---> + +metallic structures supporting high- quality surface lattice resonances25,26 with collectively excited large numbers of TMD excitons, or in plasmonic resonators (such as dimers, gaps etc.) with strong field enhancement in a small mode volume27,28. However, plasmonic lattices suffer from field delocalization, while localized resonances in small gaps or sharp nanostructures, which are usually sparse (the corresponding number of interacting emitters is reduced), pose a challenge for fine fabrication and have a limited number of plasmonic hotspots. To date, exciton- plasmon coupling strength with monolayer TMDs is typically in the range of 30- 60 meV29- 31, far beyond the regime of USC, where the coupling strength should be greater than \(0.1\omega_{ex}\) ( \(\sim 200 \mathrm{meV}\) ). + +Here we construct a cold- etched random multi- singular plasmonic metasurface grown on a flexible thermoplastic polyurethane (TPU) substrate (inset, left, Fig. 1a) as a platform to trigger USC coupling between excitons in two- dimensional (2D) TMDs and surface plasmons. Our metasurface supports dense nanometre- sized gaps as singularities, which are introduced by applying an appropriate mechanical loading to create additional nanometre- sized cracks in the fragment with saturated transferred stress (Fig. 1b), and/or by using biaxial intergranular fractures to drag the deflected fragment domain towards its neighbors at nanometre- sized spacings (Fig. 1c), see Methods and section 1 in the Supplementary Information (SI) for details. The packing density (number of singularities per unit area) is controlled by adjusting the biaxial elongations, as also shown in section 1 of SI. + +Such a strategy offers several advantages. First, the natural nanometre- sized metasurface singularities, usually difficult to be obtained with traditional top- down nanofabrication methods such as electron- beam lithography, allow stronger field localizations and smaller mode volumes (Fig. 1b- f), boosting the coupling strength. Second, considering these high- density plasmonic hotspots plus the randomness of sharp features (hence the polarization insensitivity, see section 2 in SI for details), the average number of interacting excitons (usually characterized as in- plane dipoles) with the plasmonic mode (see the near field of the gap plasmons in Fig. 1f) is also increased. + +To characterize our sample, statistical analysis of the scanning electron microscopy (SEM) images shows that the average number \((\bar{N})\) of nanometre- sized gaps within each unit area \((1 \mu \mathrm{m}^2)\) reaches 10 (Fig. 1d). We do not analyze the packing of gaps much smaller than 3 nm because SEM has difficulties in accurately capturing + +<--- Page Split ---> + +such small gaps. A representative SEM image of a \(20\mathrm{nm}\) thick gold metasurface at the highest packing condition (with the peak \(\bar{N}\) in Fig. 1d) is shown in Fig. 1e. The fracture morphology also shows the isotropic property, where the average fragment domain size is almost the same in different directions; see detailed geometrical and corresponding optical isotopic properties in section 2 of SI. The large interaction between the fragments adjacent to the singularity produces a maximum electric field enhancement of around 100 (Fig. 1f, the color bar used is to clearly show the electric field at all hotspots) within the small mode volume \((V = 2.03\times 10^{4}\mathrm{nm}^{3})\) . A pronounced Rabi splitting (blue and purple lines, Fig. 1g) is thus observed, with the extracted normalized coupling strength \(\eta = 0.12\) (purple line, Fig. 1g) and an estimated number of 23.6 excitons coherently contributing to the interaction with the surface plasmon, demonstrating USC with the \(\mathrm{WS}_2\) monolayer. + +![](images/Figure_1.jpg) + +
Fig. 1| Ultrastrong coupling in \(\mathrm{WS}_2\) monolayer with a random multi-singular plasmonic metasurface. a, Schematic of a gold multi-singular metasurface with a dense array of nanometre-sized plasmonic gaps. Insets: left: schematic of a \(\mathrm{WS}_2\) monolayer integrated with the multi-singular metasurface where the \(\mathrm{WS}_2\) monolayer was mechanically exfoliated onto a PDMS tape and transferred onto the gold nanopattern using a dry transfer method; middle and right: two main paths for generating nanometre-sized gaps. Middle: the first path, a nanometre-sized crack
+ +<--- Page Split ---> + +generates in the fragment with saturated transferred stresses. Right: the second path, adjacent fragment domains are dragged infinitely close together (at nanometre- sized spacings) by biaxial mechanical loadings. b, c, Transmission electron microscopy (TEM) images of a \(20 \mathrm{nm}\) thick gold multi- singular metasurface showing a dense array of nanometre- sized plasmonic gaps. The darker grey areas correspond to gold, the lighter ones to air. b refers to the first path, and c to the second. d, Average number \((\bar{N})\) of nanometre- sized ( \(\sim\) sub- 3 nm in the scanning electron microscopy image) plasmonic gaps per \(1 \mu \mathrm{m}^2\) in the \(20 \mathrm{nm}\) thick metasurface as a function of the second \((2^{\mathrm{nd}})\) elongation. Error bars are standard errors from multiple samples. e, Scanning electron microscopy images of the \(20 \mathrm{nm}\) - thick metasurface at \(80\%\) \(2^{\mathrm{nd}}\) elongation. The white box corresponds to the simulation area in f. f, The simulated near field of the gap plasmons. g, Dark- field scattering spectra of the \(\mathrm{WS}_2\) monolayer on polymer (thermoplastic polyurethane, black) and the plasmonic metasurface uncoupled (green) and coupled (blue for SC and purple for USC) to \(\mathrm{WS}_2\) excitons in arbitrary units (a.u.). + +## Ultrastrong exciton-plasmon coupling and its tunability + +The normalized dark- field scattering response of our samples with various gold film thicknesses (ranging from 17 to \(24 \mathrm{nm}\) to selectively tune the plasmonic resonance to the \(\mathrm{WS}_2\) exciton energy) are plotted in Fig. 2a,b, where the samples are under no strain and \(- 2\%\) uniaxial strain, respectively. Herein, to quantify the coupling strength, we fit the spectrum to a coupled oscillator model \(^{32}\) to obtain the vacuum Rabi frequency and extract the polariton energies (guided by the blue and purple curves in Fig. 2a,b) from the scattering peaks in the spectra. The extracted characteristic polariton dispersion consists of lower \((\omega_{- })\) and upper \((\omega_{+})\) polariton branches (Fig. 2c) where the polariton energies are fitted as the eigenvalues of the full Hopfield Hamiltonian \(^{33,34}\) , yielding + +\[\left(\omega^{2} - \omega_{e x}^{2}\right)\left(\omega^{2} - \omega_{p l}^{2}\right) - 4g^{2}\omega^{2} = 0, \quad (1)\] + +where counter- rotating and photon self- interaction terms are included. Here, \(\omega_{e x}\) and \(\omega_{p l}\) are the energies of the \(\mathrm{WS}_2\) excitons and the plasmonic mode, respectively, and \(g\) is the exciton- plasmon coupling strength. \(\omega_{p l}\) is calculated from the extracted \(\omega_{\pm}\) as \(\omega_{p l} = \omega_{+}\omega_{- } / \omega_{e x}\) . The fitting leads to the typical anticrossing behaviour, with \(g = 165.9 \mathrm{meV}\) for the no strain case and \(g = 240.4 \mathrm{meV}\) for \(- 2\%\) uniaxial strain, corresponding to a normalized coupling strength of \(\eta = 0.082\) and enhanced \(\eta = 0.12\) , respectively (Fig. 2c). This clearly indicates + +<--- Page Split ---> + +that the system operates in the USC regime when uniaxial strain is applied. Similar normalized coupling strengths are observed in different batches of samples (see Fig. S12 in SI), demonstrating the robustness of our platform for observing USC. The normalized coupling strength in the \(\mathrm{WS}_2\) monolayer coupled to the multisingular metasurface dramatically exceeds that of the reported optimized conventional TMD monolayer in plasmonic systems (Fig. S13 in SI, measured under exactly the same experimental conditions), suggesting that the coherent interaction of multiple excitons with the nanometre- sized plasmonic hotspots supported by the metasurface is the key factor in achieving the observed high coupling strength. Notice that in our system, the SC condition (i.e., coupling strength exceeds loss rates) \(g > (\gamma_{pl} + \gamma_{ex}) / 4\) is satisfied, where the damping losses of plasmonic resonance and exciton emission are \(\gamma_{pl} = 380 \mathrm{meV}\) and \(\gamma_{ex} = 45 \mathrm{meV}\) , respectively. + +Applications of USC to nonlinear optics include low- threshold frequency conversion \(^{35}\) and phase- matched optical amplification \(^{36}\) . To demonstrate the advantages of our system with tunable USC, here we exploit USC towards tunable resonant polariton- enhanced nonlinearity. It is known that the excitonic resonances in monolayer TMDs can facilitate second harmonic generation (SHG) \(^{37}\) . In our system, USC allows spectral splitting of the polaritonic resonance with tunable ground state energies and, consequently, leads to dispersive polariton- enhanced SHG in the \(\mathrm{WS}_2\) monolayer. Strong evidence for the pronounced SHG intensity enhancement and spectral splitting can be seen from the nonlinearity spectrum as a function of the pump wavelength under two different values of the applied strain (Fig. 2d). As can be seen in Fig. S14, the polariton- enhanced SHG from the \(\mathrm{WS}_2\) monolayer is around 15 times stronger than that on PDMS, and polarization- independent polariton- enhanced SHG is observed due to the isotropy of our multi- singular metasurface. + +<--- Page Split ---> +![](images/Figure_2.jpg) + +
Fig. 2| Ultrastrong exciton-plasmon coupling in WS \(_2\) monolayer. Dark-field scattering spectra under a 0 and b -2% uniaxial strain on the metasurface over different gold film thicknesses (from bottom to top, the gold film thickness \(\mathrm{(t_{Au})}\) increases from 17 nm to 24 nm). The scattering splits into lower and upper polariton branches, exhibiting level anticrossing. The vertical dashed lines refer to the WS \(_2\) exciton energy \(\omega_{ex} = 2.019 \mathrm{eV}\) . The right panels are calculated scattering spectra using the model of two coupled oscillators \(^{32}\) . c, Dispersion plots of the measured dark-field scattering spectra. The lower \((\omega_{-})\) and upper \((\omega_{+})\) polariton branches are extracted from the scattering spectra in a,b and fitted (solid lines) with a coupling strength of \(165.9 \mathrm{meV}\) at 0 strain (blue lines) and \(240.4 \mathrm{meV}\) at -2% strain (purple lines) in the full Hopfield Hamiltonian. The experimental spectral peaks are shown as triangles (0 strain) and circles (-2% strain). The exciton energy \((\omega_{ex})\) is shown as the horizontal dashed black line. The plasmonic mode energy \((\omega_{pl})\) is shown as the diagonal dashed blue (SC) and purple (USC) lines. The plasmonic response shifts to higher energy when compressive strain is applied, and the local field enhancement is largely enhanced due to the reduced gap size, and both effects lead to significant changes in the polaritonic dispersion. d, Normalized second
+ +<--- Page Split ---> + +harmonic generation (SHG) intensity compared to scattering in the SC (upper, strain=0) and USC (lower, strain=- 2%) regime, respectively. The simultaneous emergence of energy splitting in both scattering and SHG spectra precludes Fano interference phenomena from being responsible for the observed anticrossing. + +Furthermore, thanks to the flexible nature of the TPU polymer substrate, the mechanical bending in our system can be adjusted to modify the plasmonic gap size and the corresponding packing density for modulating exciton- plasmon coupling strength. Under upward (tensile strain)/downward (compressive strain) bending, both the plasmonic resonance (left, Fig. 3a) and the corresponding plasmonic enhancement in photoluminescence peak intensity (right, Fig. 3a) can be tuned. We notice that the small strain applied in this work has a negligible effect on the exciton energy (Fig. S16 in SI), because the \(\mathrm{WS}_2\) monolayer is loosely contacted with the surface of the plasmonic metasurface. Dark- field scattering spectra at different strains are rendered in Fig. 3b, with two polariton branches well fitted (black solid lines, Fig. 3b) by the full Hopfield Hamiltonian described above. Specifically, the normalized coupling strength varies gradually from 0.075 to 0.12 (over the strain range from \(+0.375\%\) to \(- 2\%\) , Fig. 3c), clearly demonstrating the tunability of USC. We expect even higher coupling strengths by further increasing the compressive strain, but to preserve the resilience and robustness of the flexible substrate, we have avoided attempting to increase the compressive strain beyond \(- 2\%\) . + +One interesting implication of the USC is that the global vacuum energy of the system is modified with respect to the coupling strength (Fig. 3d) by dressing excitons with light33,38,39. In our system, the corresponding change in the ground state is calculated as \(\Delta E_G = \hbar (\omega_+ + \omega_- - \omega_{ex} - \omega_{pl}) / 2\) . Note that the ground state energy of a harmonic oscillator is half that of the transition energy40. The normalized ground- state energy variation \((\Delta E_G / E_G)\) versus strain was calculated and fitted to the extracted coupling strengths (Fig. 3c) and polariton energies \(\omega_{\pm}\) (Fig. 3b), yielding a modification of \(0.67\%\) (Fig. 3d). The absolute ground- state energy modification reaches \(13.4 \mathrm{meV}\) at \(- 2\%\) compressive strain. + +<--- Page Split ---> +![](images/Figure_3.jpg) + +
Fig. 3| Tunable ultrastrong exciton-plasmon coupling in WS \(_2\) monolayer. a, Mechanically tunable plasmonic resonance (wavelength at \(\lambda_{pl}\) ) and plasmonic enhancement in photoluminescence peak intensity of the strongly coupled plasmonic system. b, Normalized dark-field scattering intensity as a function of excitation energy and strain. The horizontal purple line marks the onset of USC. The black solid lines are the extracted polariton energies of the lower (LPB) and upper (UPB) polariton branches. The white dashed line is the WS \(_2\) exciton energy ( \(\omega_{ex} = 2.019 \mathrm{eV}\) ). c, Normalized coupling strength as a function of strain. The onset of USC is marked by the horizontal dashed line. d, Ground-state energy modification as a function of strain. The black dashed line is the calculation result from the coupling strength ( \(g\) ) and the polariton energies ( \(\omega_{+}\) , \(\omega_{- }\) ) from the fitting to the full Hopfield Hamiltonian. The absolute change in ground-state energy reaches 13.4 meV at -2% compressive strain. Error bars in c and d result from the variation in dark-field scattering from multiple batches of samples and are extracted from the standard error of the fit.
+ +<--- Page Split ---> + +## Ultrastrong coupling in \(\mathbf{W}\mathbf{S}_{2}\) multilayers + +Last, we show that the coupling strength in USC is also relevant with the layer numbers of \(\mathbf{W}\mathbf{S}_{2}\) . We transferred trilayer and quadralayer \(\mathbf{W}\mathbf{S}_{2}\) flakes onto several metasurfaces with different gold layer thicknesses (19 nm to 23 nm) on flexible PDMS substrates. The dispersion curves of the scattering spectra are shown in Fig. 4, showing the increased energy splitting between polaritons compared to the monolayer. In trilayer \(\mathbf{W}\mathbf{S}_{2}\) (Fig. 4a,b), the Rabi splitting \((\Omega_{R})\) exceeds \(397.9 \mathrm{meV}\) \((\eta \sim 0.108)\) and \(563.9 \mathrm{meV}\) \((\eta \sim 0.147)\) under 0 and \(- 2\%\) compressive strain, respectively; while in quadralayer \(\mathbf{W}\mathbf{S}_{2}\) (Fig. 4c,d), \(\Omega_{R}\) can be as high as \(449.7 \mathrm{meV}\) \((\eta \sim 0.12)\) and \(634.7 \mathrm{meV}\) \((\eta \sim 0.164)\) accordingly. Moving from the monolayer to the quadralayer cases, the normalized coupling strength increases with the number of layers, but less rapidly than predicted by a square root function, as shown in Fig. 4e, because the coupling strength depends on both the number of excitons and the spatial overlap of the confined electric field with the \(\mathbf{W}\mathbf{S}_{2}\) layers (which decreases as the number of \(\mathbf{W}\mathbf{S}_{2}\) layer increases). Thus, we have successfully combined monolayer and multilayer \(\mathbf{W}\mathbf{S}_{2}\) flakes with multi- singular metasurfaces to create a large set of ultrastrongly coupled exciton- plasmon systems under ambient conditions. Moreover, the cooperativity, defined as \(\mathrm{C} = 4g^{2} / \gamma_{p1}\gamma_{ex}\) (a key figure of merit to characterize the coupling regimes of light- matter interaction) in our samples outperforms all room- temperature TMD- based platforms ( \(\mathrm{C} = 28.4\) in quadralayer \(\mathbf{W}\mathbf{S}_{2}\) ). + +We compare our system with the previous reports of light- matter coupling, as summarized in Table 1 in SI for details. The normalized coupling coefficients of other strongly coupled systems are typically in the range of 0.015- 0.03 (0.015- 0.04) for monolayer \(^{29 - 31}\) (multilayer \(^{27,28,41,42}\) ) TMDs, 0.02- 0.095 for organic molecules (less than \(20 \mathrm{nm}\) thick) \(^{43 - 45}\) , and 0.04- 0.09 for quantum dots \(^{46 - 48}\) . Existing room- temperature USC systems are typically implemented using organic molecules, and the reported thickness must be greater than \(60 \mathrm{nm}^{18,49,50}\) . Landau polaritons and superconducting circuits are not included in this comparison because their transition energies are far below those of TMD excitons. Our platform for USC, with the emitter thickness of less than \(1 \mathrm{nm}\) , is thus unique and holds great promise for next- generation atomically thin optoelectronics in the visible. + +<--- Page Split ---> +![](images/Figure_4.jpg) + +
Fig. 4| Ultrastrong coupling in multilayer WS2 flakes. a-d, Dark-field scattering spectra of a multilayer WS2 plasmonic system. Dashed white lines denote the exciton energy ( \(\omega_{ex} = 2.01 \mathrm{eV}\) for a,b; \(\omega_{ex} = 2.003 \mathrm{eV}\) for c,d;), and the diagonal dashed orange and purple lines indicate the plasmon energy ( \(\omega_{pl}\) ). The open circles represent lower ( \(\omega_{-}\) ) and upper ( \(\omega_{+} \mathrm{p}\) ) polariton energies obtained from the dark-field scattering spectra of individual plasmonic systems. The solid lines show the lower and upper polariton dispersions using the full Hopfield Hamiltonian. 3L and 4L represent the trilayer and quadrilayer cases. e, Normalized coupling strength ( \(g / \omega_{ex}\) ) as a function of the WS2 layers under 0 (orange) and -2% compressive strain (purple). The horizontal dashed line marks the onset of USC. Error bars are derived from the variation in dark-field scattering from different batches of samples and are extracted from the standard error of the fit.
+ +## Discussion + +In summary, we have shown the ultrastrong exciton- plasmon coupling at room temperature in WS2 multilayers coupled to random multi- singular plasmonic metasurfaces formed by cold- etching, which can be further tuned + +<--- Page Split ---> + +by mechanical strain in a flexible substrate. Our results promise the USC in the atomic layer limit under ambient conditions, which could be further extended to scenarios of different metals or doped semiconductors and complex 2D heterostructures for more exotic complex light- matter interactions, such as tunable trion polaritons, van der Waals heterostructure polaritons, and moiré induced optical nonlinearities. Our work could also lead to new insights in fundamental science and potential applications in the fields of ultralow power nonlinear nanophotonics, analytical chemistry, quantum optics, among others. + +## Methods + +## Experimental structure. + +\(\mathrm{WS}_2\) coupled multi- singular plasmonic metasurface: A thin gold film was deposited on an amorphous poly(etherimide) (PEI) polymer substrate ( \(125\mu \mathrm{m}\) thick) using an electron beam evaporator at a deposition rate of \(2\mathrm{\AA}\mathrm{s}^{- 1}\) . During the evaporation, the temperature of the vacuum chamber was maintained below \(60^{\circ}\mathrm{C}\) throughout the process to prevent thermal expansion or deformation of the PEI film due to the build- in stress, which would cause wrinkles or defects in the gold film. For the cold- etching process, the first stretch was performed by stretching the gold/PEI film in the \(x\) - direction and stopping when the necks extended along the length of the film. The second stretch was performed by re- stretching the as- fabricated film in the \(y\) - direction and the elongation extent was controlled to obtain gold nanopatterns with different second elongations. After the cold- etching process, the resulting gold nanopatterns were transferred to a flexible TPU substrate ( \(2\mathrm{mm}\) thick) by a dry peel- off method. \(\mathrm{WS}_2\) monolayers and multilayers were mechanically exfoliated from the commercial bulk crystals onto a PDMS tape and transferred to the gold multi- singular metasurface (on TPU) through a dry transfer method (see Fig.S1 in SI for details). + +## Optical characterizations. + +All optical characterizations were performed in the reflective geometry at room temperature. For all linear polarization measurements, the metasurface system was rotated and the polarization- dependent components were analyzed. For photoluminescence measurements, a \(532\mathrm{nm}\) diode laser was used to excite the sample. A + +<--- Page Split ---> + +\(100\times\) microscope objective lens (numerical aperture (NA) D 0.75) was used and the incident laser power was \(\sim 100\mu \mathrm{W}\) with a laser spot size of \(\sim 4\mu \mathrm{m}\) . Dark- field scattering measurements were carried out using a hyperspectral imaging system with a broadband halogen lamp as the light source. A \(50\times\) objective (NA D 0.55) was used and the incident light power was \(\sim 20\mu \mathrm{W}\) with a laser spot size of \(\sim 5\mu \mathrm{m}\) . + +## Numerical Calculations. + +Commercial finite- difference time- domain software was used to calculate the field enhancement of the gap plasmons. The permittivity \(\mathrm{WS}_2\) is modeled as a Lorentzian oscillator \(\epsilon (\omega) = 1 + \sum_{k = 1}^{N}f_{k} / (\omega_{k}^{2} - \omega^{2} - i\gamma_{k}\omega)\) , with \(f_{k}\) , \(\gamma_{k}\) and \(\omega_{k}\) being the oscillator strength, the linewidth of the \(k\) th oscillator and the oscillation energy, respectively. The permittivity of the gold was from the software database. + +## Hopfield Hamiltonian. + +The Hopfield Hamiltonian of our plasmonic system in the USC regime contains three main blocks: + +\[H = H_{s y s} + H_{\mathrm{int}} + H_{A^{2}} \quad (2)\] + +with the Hamiltonian of the closed- plasmonic system + +\[H_{s y s} = \mathrm{h}\omega_{e x}\hat{a}^{\dagger}\hat{a} +\mathrm{h}\omega_{\rho l}\hat{b}^{\dagger}\hat{b} \quad (3)\] + +and the interaction Hamiltonian + +\[H_{\mathrm{int}} = \mathrm{i}\hbar \mathrm{g}\left(\hat{a}^{\dagger} + \hat{a}\right)\left(\hat{b} -\hat{b}^{\dagger}\right) \quad (4)\] + +and the photon self- interaction \((A^{2})\) Hamiltonian + +\[H_{A^{2}} = \frac{\mathrm{h}\mathrm{g}^{2}}{\omega_{e x}}\left(\hat{a}^{\dagger} + \hat{a}\right)\left(\hat{a}^{\dagger} + \hat{a}\right) \quad (5)\] + +\(\hat{a}^{\dagger}\) and \(\hat{a}\) are the exciton creation and annihilation operators, respectively, and \(\hat{b}^{\dagger}\) and \(\hat{b}\) are those of the localized plasmons. Since \(H\) is invariant under translation we define + +\[\hat{c} = w\hat{a} +x\hat{b} +y\hat{a}^{\dagger} +z\hat{b}^{\dagger} \quad (6)\] + +And we can get + +<--- Page Split ---> + +\[[\hat{c},H] = E\hat{c} \quad (7)\] + +The eigenvalue problem is rewritten in matrix form, + +\[\begin{array}{r}{\left[ \begin{array}{c c c c c c c c c c c c c c c c c c c c c}{{\omega_{p l} + 2g^{2} / \omega_{e x} - i\gamma_{p l} / 2}} & {-i g} & {-2g^{2} / \omega_{e x}} & {-i g} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {i g} & {\omega_{e x} - i\gamma_{e x} / 2} & {-i g} & 0 & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {2g^{2} / \omega_{e x}} & {-i g} & {-a_{p l} - 2g^{2} / \omega_{e x} - i\gamma_{p l}^{*} / 2} & {-i g} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {-i g} & 0 & {i g} & {-a_{e x} - i\gamma_{e x}^{*} / 2} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {} \end{array} \right]\left[ \begin{array}{c}{w}\\ {x}\\ {y}\\ {z} \end{array} \right] = E\left[ \begin{array}{c}{w}\\ {x}\\ {y}\\ {z} \right] \end{array} \right]} \end{array} \quad (8)\] + +Two eigenvalues \(\omega_{\mp}\) of the above matrix are the positive solutions of the equation \(^{30,45,46}\) , + +\[\left(\omega^{2} - \omega_{e x}^{2}\right)\left(\omega^{2} - \omega_{p l}^{2}\right) - 4g^{2}\omega^{2} = 0 \quad (9)\] + +from which we obtain + +\[\omega_{c}\omega_{-} = \omega_{p l}\omega_{e x} \quad (10)\] + +To interpret our experimental data in the SC regime, we obtain the analytical polariton dispersion energies with a coupled oscillator model with the Hamiltonian + +\[H = \left[ \begin{array}{cc}{\omega_{p l} - i\gamma_{p l} / 2} & {g}\\ {g} & {\omega_{e x} - i\gamma_{e x} / 2} \end{array} \right] \quad (11)\] + +as + +\[\omega_{\pm} = \frac{1}{2}\Bigg(\omega_{p l} - i\frac{\gamma_{p l}}{2} +\omega_{e x} - i\frac{\gamma_{e x}}{2}\Bigg)\pm \sqrt{8^{2} + \frac{1}{4}\Bigg(\omega_{p l} - i\frac{\gamma_{p l}}{2} -\omega_{e x} + i\frac{\gamma_{e x}}{2}\Bigg)^{2}} \quad (12)\] + +Note that the complex Rabi splitting can be calculated as \(\Omega_{R} = \omega_{+} - \omega_{- } =\) \(2\sqrt{g^{2} + \frac{1}{4}\big(\omega_{p l} - i\frac{\gamma_{p l}}{2} -\omega_{e x} + i\frac{\gamma_{e x}}{2}\big)^{2}}\) , where for zero detuning and zero damping the rabi splitting is \(\Omega_{R} = 2g\) . + +## Strain model: estimation on gap size in the plasmonic metasurface + +The Young's modulus, ultimate strength and thickness of the film are given by \(E\) , \(\sigma_{\mathrm{s}}\) and \(h\) , respectively. Assume that the interfacial shear stress is a constant \(\tau_{0}\) for any interfacial displacement and the substrate stretching is accompanied by a steady neck propagation with the necked stretch ration of \(\lambda_{n}\) . + +The film fractures sequentially into fragments following the propagating neck. For a fragment of film, its size in the initial configuration (undeformed configuration) can be determined as + +<--- Page Split ---> + +\[L_{0} = \frac{E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (13)\] + +The portion of substrate with the same initial position and length as the film fragment experiences full necking and now has the size + +\[L_{\mathrm{subs}} = \lambda_{\mathrm{n}}L_{0} = \frac{\lambda_{\mathrm{n}}E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (14)\] + +Assume that the residual stress in the film fragment is distributed linearly within the fragment and has the maximum value of \(\sigma_{\mathrm{r}}\) . Then we have + +\[L_{0} = \frac{2E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (15)\] + +With Eqs. (13) and (15), we obtain + +\[\sigma_{\mathrm{r}} = E\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}. \quad (16)\] + +Therefore, the size of the film fragment after rupture is + +\[L_{\mathrm{film}} = \frac{2\sigma_{\mathrm{r}}h}{\tau_{0}} = \frac{2E h}{\tau_{0}}\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}, \quad (17)\] + +and the distance between neighboring film fragments, i.e., the averaged gap size in metasurface, can be expressed as + +\[\overline{{L}}_{\mathrm{gap}} = L_{\mathrm{subs}} - L_{\mathrm{film}} = \frac{E h}{\tau_{0}}\bigg(\lambda_{\mathrm{n}} - \lambda_{\mathrm{n}}e^{-\sigma_{\mathrm{s}} / E} - 2\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}\bigg). \quad (18)\] + +## Data availability + +The data supporting the current study in the paper are included in the paper and/or the Supplementary Materials. Additional data related to this paper can be requested from the corresponding authors. + +## References + +1. Diederichs, C. et al. Parametric oscillation in vertical triple microcavities. Nature 440, 904-907 (2006). + +<--- Page Split ---> + +2. Duh, Y. S. et al. Giant photothermal nonlinearity in a single silicon nanostructure. Nat. Commun. 11, 1-9 (2020). + +3. Ha, T. & Tinnefeld, P. Photophysics of fluorescent probes for single-molecule biophysics and super-resolution imaging. Annu. Rev. Phys. Chem. 63, 595-617 (2012). + +4. Zhang, L., Gogna, R., Burg, W., Tutuc, E. & Deng, H. Photonic-crystal exciton-polaritons in monolayer semiconductors. Nat. Commun. 9, 1-8 (2018). + +5. Gu, J., Chakraborty, B., Khatoniar, M. & Menon, V. M. A room-temperature polariton light-emitting diode based on monolayer \(\mathrm{WS}_2\) . Nat. Nanotechnol. 14, 1024-1028 (2019). + +6. Sidler, M. et al. Fermi polaron-polaritons in charge-tunable atomically thin semiconductors. Nat. Phys. 13, 255-261 (2017). + +7. Dufferwiel, S. et al. Exciton-polaritons in van der Waals heterostructures embedded in tunable microcavities. Nat. Commun. 6, 8579 (2015). + +8. Kaliteevski, M. et al. Tamm plasmon-polaritons: Possible electromagnetic states at the interface of a metal and a dielectric Bragg mirror. Phys. Rev. B 76, 165415 (2007). + +9. Lundt, N. et al. Room-temperature Tamm-plasmon exciton-polaritons with a \(\mathrm{WSe}_2\) monolayer. Nat. Commun. 7, 1-6 (2016). + +10. Liu, X. et al. Nonlinear valley phonon scattering under the strong coupling regime. Nat. Mater. 20, 1210-1215 (2021). + +11. Liu, L. et al. Plasmon-induced thermal tuning of few-exciton strong coupling in 2D atomic crystals. Optica 8, 1416-1423 (2021). + +12. Irish, E. K. Generalized rotating-wave approximation for arbitrarily large coupling. Phys. Rev. Lett. 99, 173601 (2007). + +13. Anappara, A. A. et al. Signatures of the ultrastrong light-matter coupling regime. Phys. Rev. B 79, 201303 (2009). + +14. Günter, G. et al. Sub-cycle switch-on of ultrastrong light-matter interaction. Nature 458, 178-181 (2009). + +<--- Page Split ---> + +15. Niemczyk, T. et al. Circuit quantum electrodynamics in the ultrastrong-coupling regime. Nat. Phys. 6, 772-776 (2010). + +16. Scalari, G. et al. Ultrastrong coupling of the cyclotron transition of a 2D electron gas to a THz metamaterial. Science 335, 1323-1326 (2012). + +17. Zhang, Q. et al. Collective non-perturbative coupling of 2D electrons with high-quality-factor terahertz cavity photons. Nat. Phys. 12, 1005-1011 (2016). + +18. Gambino, S. et al. Exploring light-matter interaction phenomena under ultrastrong coupling regime. ACS Photon. 1, 1042-1048 (2014). + +19. Barra-Burillo, M. et al. Microcavity phonon polaritons from the weak to the ultrastrong phonon-photon coupling regime. Nat. Commun. 12, 6206 (2021). + +20. Mueller, N. S. et al. Deep strong light-matter coupling in plasmonic nanoparticle crystals. Nature 583, 780-784 (2020). + +21. Kurman, Y. & Kaminer, I. Tunable bandgap renormalization by nonlocal ultra-strong coupling in nanophotonics. Nat. Phys. 16, 868-874 (2020). + +22. Shalabney, A. et al. Coherent coupling of molecular resonators with a microcavity mode. Nat. Commun. 6, 5981 (2015). + +23. Chikkaraddy, R. et al. Single-molecule strong coupling at room temperature in plasmonic nanocavities. Nature 535, 127-130 (2016). + +24. Halas, Naomi J. et al. Plasmons in strongly coupled metallic nanostructures. Chem. Rev. 111, 3913-3961 (2011). + +25. Wang, S. et al. Coherent coupling of \(\mathrm{WS}_2\) monolayers with metallic photonic nanostructures at room temperature. Nano Lett. 16, 4368-4374 (2016). + +26. Liu, W. et al. Strong exciton-plasmon coupling in \(\mathrm{MoS}_2\) coupled with plasmonic lattice. Nano Lett. 16, 1262-1269 (2016). + +27. Kleemann, M. E. et al. Strong-coupling of \(\mathrm{WSe}_2\) in ultra-compact plasmonic nanocavities at room temperature. Nat. Commun. 8, 1296 (2017). + +<--- Page Split ---> + +28. Stuhrenberg, M. et al. Strong light-matter coupling between plasmons in individual gold bi-pyramids and excitons in mono-and multilayer WSe₂. Nano Lett. 18, 5938-5945 (2018). + +29. Jiang, Y., Wang, H., Wen, S., Chen, H., & Deng, S. Resonance coupling in an individual gold nanorod-monolayer WS₂ heterostructure: photoluminescence enhancement with spectral broadening. ACS Nano 14, 13841-13851 (2020). + +30. Sang, Y. et al. Tuning of two-dimensional plasmon-exciton coupling in full parameter space: a polaritonic non-Hermitian system. Nano Lett. 21, 2596-2602 (2021). + +31. Geisler, M. et al. Single-crystalline gold nanodisks on WS₂ mono-and multilayers for strong coupling at room temperature. ACS Photon. 6, 994-1001 (2019). + +32. Leng, H., Szychowski, B., Daniel, M. C., & Pelton, M. Strong coupling and induced transparency at room temperature with single quantum dots and gap plasmons. Nat. Commun. 9, 4012 (2018). + +33. Ciuti, C. & Carusotto, I. Input-output theory of cavities in the ultrastrong coupling regime: The case of time-independent cavity parameters. Phys. Rev. A 74, 033811 (2006). + +34. Hopfield, J. Theory of the contribution of excitons to the complex dielectric constant of crystals. Phys. Rev. 112, 1555 (1958). + +35. Sánchez-Burillo, E., Zueco, D., Garcia-Ripoll, J. & Martin-Moreno, L. Scattering in the ultrastrong regime: nonlinear optics with one photon. Phys. Rev. Lett. 113, 263604 (2014). + +36. Mathew, J. P., Patel, R. N., Borah, A., Vijay, R. & Deshmukh, M. M. Dynamical strong coupling and parametric amplification of mechanical modes of graphene drums. Nat. Nanotechnol. 11, 747-751 (2016). + +37. Seyler, K. L. et al. Electrical control of second-harmonic generation in a WSe₂ monolayer transistor. Nat. Nanotechnol. 10, 407-411 (2015). + +38. Ciuti, C., Bastard, G. & Carusotto, I. Quantum vacuum properties of the intersubband cavity polariton field. Phys. Rev. B 72, 115303 (2005). + +39. Mahboob, I. & Yamaguchi, H. Bit storage and bit flip operations in an electromechanical oscillator. Nat. Nanotechnol. 3, 275-279 (2008). + +<--- Page Split ---> + +40. Baranov, D. G. et al. Ultrastrong coupling between nanoparticle plasmons and cavity photons at ambient conditions. Nat. Commun. 11, 2715 (2020). + +41. Wang, S. et al. Limits to strong coupling of excitons in multilayer \(\mathrm{WS}_2\) with collective plasmonic resonances. ACS Photon. 6, 286-293 (2019). + +42. Yankovich, A. B. et al. Visualizing spatial variations of plasmon-exciton polaritons at the nanoscale using electron microscopy. Nano Lett. 19, 8171-8181 (2019). + +43. Zengin, G. et al. Realizing strong light-matter interactions between single-nanoparticle plasmons and molecular excitons at ambient conditions. Phys. Rev. Lett. 114, 157401 (2015). + +44. Wersall, M., Cuadra, J., Antosiewicz, T. J., Balci, S., & Shegai, T. Observation of mode splitting in photoluminescence of individual plasmonic nanoparticles strongly coupled to molecular excitons. Nano Lett., 17, 551-558 (2017). + +45. Liu, R. et al. Strong light-matter interactions in single open plasmonic nanocavities at the quantum optics limit. Phy. Rev. Lett. 118, 237401 (2017). + +46. Santhosh, K. et al. Vacuum Rabi splitting in a plasmonic cavity at the single quantum emitter limit. Nat. Commun. 7, 11823, (2016). + +47. Bitton, O. et al. Vacuum Rabi splitting of a dark plasmonic cavity mode revealed by fast electrons. Nat. Commun. 11, 487, (2020). + +48. Park, K. et al. Tip-enhanced strong coupling spectroscopy, imaging, and control of a single quantum emitter. Sci. Adv. 5, 5931 (2019). + +49. Gubbin, C. R., Maier, S. A., & Kéna-Cohen, S. Low-voltage polariton electroluminescence from an ultrastrongly coupled organic light-emitting diode. Appl. Phys. Lett., 104, 85 (2014). + +50. Kéna-Cohen, S., Maier, S. A., & Bradley, D. D. Ultrastrongly Coupled Exciton-Polaritons in Metal-Clad Organic Semiconductor Microcavities. Adv. Opt. Mat., 1, 827-833 (2013). + +<--- Page Split ---> + +## Acknowledgements + +AcknowledgementsThis work was supported by the Singapore National Research Foundation Competitive Research Program (NRF- CRP22- 2019- 0006, NRF- CRP23- 2019- 0007 and NRF- CRP22- 2019- 0007), the Singapore Ministry of Education Academic Research Fund Tier 2 (MOE2019- T2- 2- 127, MOE- T2EP50120- 0002, MOE- T2EP50120- 0009, MOE- T2EP50220- 0020 and MOE- T2EP50122- 0005), AcRF Tier 1 (RG57/21, RG156/19 (S)), AcRF Tier 3 (MOE2016- T3- 1- 006 (S)), A\*STAR (1720700038, A1883c0002, A18A7b0058, A20E5c0095, A2083c0062 and A2090b0144), Spanish Ministry for Science and Innovation- Agencia Estatal de Investigación (AEI) through grants CEX2018- 000805- M and PID2021- 125894NB- I00, the Autonomous Community of Madrid, the Spanish government and the European Union through grant MRR Advanced Materials (MAD2D- CM), and National Medical Research Council (NMRC) (021528- 00001). This work was supported by A\*STAR under its IAF- ICP Programme I2001E0067 and the Schaeffler Hub for Advanced Research at NTU. + +## Author contributions + +T. Wu and Y. Luo designed the research project. T. Wu and M. Chen fabricated the flexible plasmonic metasurfaces; C. Wang fabricated the \(\mathrm{WS}_2\) flakes and the bowtie/dimer antennas. T. Wu, L. Liu, and J. Zhao conducted the optical experiments. Z. Wang, Z. Wang, and T. Wu took the SEM and TEM images. T. Wu, G. Hu and L. Liu performed the finite-difference time-domain simulations and theoretical analysis. F. J. Garcia-Vidal, L. Wei, Q. Wang, and Y. Luo supervised the research. T. Wu, G. Hu, and L. Liu analyzed the data; T. Wu and G. Hu wrote the manuscript. All authors contributed to data interpretation and editing the manuscript. + +## Competing interests + +The authors declare no competing interests. + +<--- Page Split ---> + +## Supplementary Files + +This is a list of supplementary files associated with this preprint. Click to download. + +- Sl.docx + +<--- Page Split ---> diff --git a/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82_det.mmd b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82_det.mmd new file mode 100644 index 0000000000000000000000000000000000000000..2d9006118ca61477f196418d12a2813a9fa561d4 --- /dev/null +++ b/preprint/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82/preprint__177e7068407aec814b602cec4324047ce9cf091ccd0d9aaa6131a5642405bf82_det.mmd @@ -0,0 +1,586 @@ +<|ref|>title<|/ref|><|det|>[[44, 107, 855, 210]]<|/det|> +# Ultrastrong exciton-plasmon couplings in WS2 multilayers synthesized with a random multi- singular metasurface at room temperature + +<|ref|>text<|/ref|><|det|>[[44, 230, 110, 248]]<|/det|> +Yu Luo + +<|ref|>text<|/ref|><|det|>[[55, 258, 234, 275]]<|/det|> +luoyu@ntu.edu.sg + +<|ref|>text<|/ref|><|det|>[[44, 301, 712, 323]]<|/det|> +Nanyang Technological University https://orcid.org/0000- 0003- 2925- 682X + +<|ref|>text<|/ref|><|det|>[[44, 328, 352, 368]]<|/det|> +Tingting Wu Nanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 373, 712, 415]]<|/det|> +Chongwu Wang Nanyang Technological University https://orcid.org/0000- 0002- 8749- 196X + +<|ref|>text<|/ref|><|det|>[[44, 420, 712, 461]]<|/det|> +Guangwei Hu Nanyang Technological University https://orcid.org/0000- 0002- 3023- 9632 + +<|ref|>text<|/ref|><|det|>[[44, 466, 712, 507]]<|/det|> +Zhixun Wang Nanyang Technological University https://orcid.org/0000- 0001- 9918- 9939 + +<|ref|>text<|/ref|><|det|>[[44, 512, 154, 530]]<|/det|> +Jiaxin Zhao + +<|ref|>text<|/ref|><|det|>[[44, 534, 909, 576]]<|/det|> +Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 581, 765, 623]]<|/det|> +Zhe Wang School of Electrical and Electronic Engineering, Nanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 628, 352, 668]]<|/det|> +Ksenia Chaykun Nanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 674, 352, 715]]<|/det|> +Lin Liu Nanyang Technological University + +<|ref|>text<|/ref|><|det|>[[44, 720, 395, 761]]<|/det|> +Mengxiao Chen https://orcid.org/0000- 0001- 5853- 4791 + +<|ref|>text<|/ref|><|det|>[[44, 767, 907, 830]]<|/det|> +Dong Li School of Mechanical and Aerospace Engineering, College of Engineering, Nanyang Technological University https://orcid.org/0000- 0002- 4484- 8738 + +<|ref|>text<|/ref|><|det|>[[44, 836, 592, 876]]<|/det|> +Qihua Xiong Tsinghua University https://orcid.org/0000- 0002- 2555- 4363 + +<|ref|>text<|/ref|><|det|>[[44, 882, 712, 923]]<|/det|> +Ze Shen Nanyang Technological University https://orcid.org/0000- 0001- 7432- 7936 + +<|ref|>text<|/ref|><|det|>[[44, 928, 155, 946]]<|/det|> +Huajian Gao + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[50, 45, 712, 66]]<|/det|> +Nanyang Technological University https://orcid.org/0000- 0002- 8656- 846X + +<|ref|>text<|/ref|><|det|>[[44, 70, 707, 112]]<|/det|> +Francisco Garcia- Vidal Universidad Autónoma de Madrid https://orcid.org/0000- 0003- 4354- 0982 + +<|ref|>text<|/ref|><|det|>[[44, 117, 763, 181]]<|/det|> +Lei Wei School of Electrical and Electronic Engineering, Nanyang Technological University https://orcid.org/0000- 0003- 0819- 8325 + +<|ref|>text<|/ref|><|det|>[[44, 186, 712, 228]]<|/det|> +Qi jie Wang Nanyang Technological University https://orcid.org/0000- 0002- 9910- 1455 + +<|ref|>sub_title<|/ref|><|det|>[[44, 268, 101, 286]]<|/det|> +## Article + +<|ref|>title<|/ref|><|det|>[[44, 306, 135, 324]]<|/det|> +# Keywords: + +<|ref|>text<|/ref|><|det|>[[44, 343, 315, 363]]<|/det|> +Posted Date: October 6th, 2023 + +<|ref|>text<|/ref|><|det|>[[44, 381, 475, 401]]<|/det|> +DOI: https://doi.org/10.21203/rs.3.rs- 3409617/v1 + +<|ref|>text<|/ref|><|det|>[[42, 418, 909, 461]]<|/det|> +License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License + +<|ref|>text<|/ref|><|det|>[[42, 479, 531, 500]]<|/det|> +Additional Declarations: There is NO Competing Interest. + +<|ref|>text<|/ref|><|det|>[[42, 534, 909, 578]]<|/det|> +Version of Record: A version of this preprint was published at Nature Communications on April 17th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 47610- z. + +<--- Page Split ---> +<|ref|>title<|/ref|><|det|>[[113, 105, 770, 128]]<|/det|> +# Ultrastrong exciton-plasmon couplings in WS₂ multilayers + +<|ref|>title<|/ref|><|det|>[[113, 151, 820, 175]]<|/det|> +# synthesized with a random multi-singular metasurface at room + +<|ref|>sub_title<|/ref|><|det|>[[115, 201, 257, 221]]<|/det|> +## temperature + +<|ref|>text<|/ref|><|det|>[[113, 256, 855, 338]]<|/det|> +Tingting \(\mathrm{Wu}^{1,\#}\) , Chongwu Wang \(^{1,\#}\) , Guangwei \(\mathrm{Hu}^{1,\#}\) , Zhixun Wang \(^{1}\) , Jiaxin Zhao \(^{2}\) , Zhe Wang \(^{1}\) , Ksenia Chaykun \(^{2}\) , Lin Liu \(^{1}\) , Mengxiao Chen \(^{3}\) , Dong Li \(^{4}\) , Qihua Xiong \(^{5}\) , Zexiang Shen \(^{2}\) , Huajian Gao \(^{4}\) , Francisco J. Garcia- Vidal \(^{6,7,*}\) , Lei Wei \(^{1,*}\) , Qi Jie Wang \(^{1,2,*}\) and Yu Luo \(^{1,*}\) + +<|ref|>text<|/ref|><|det|>[[113, 356, 864, 375]]<|/det|> +\(^{1}\) School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, Singapore. + +<|ref|>text<|/ref|><|det|>[[113, 387, 856, 406]]<|/det|> +\(^{2}\) School of Physical and Mathematical Sciences, Nanyang Technological University, Singapore, Singapore. + +<|ref|>text<|/ref|><|det|>[[113, 418, 883, 467]]<|/det|> +\(^{3}\) Zhejiang Provincial Key Laboratory of Cardio- Cerebral Vascular Detection Technology and Medicinal Effectiveness Appraisal, Zhejiang University, Hangzhou, China + +<|ref|>text<|/ref|><|det|>[[113, 479, 880, 498]]<|/det|> +\(^{4}\) School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore, Singapore. + +<|ref|>text<|/ref|><|det|>[[113, 510, 883, 558]]<|/det|> +\(^{5}\) State Key Laboratory of Low- Dimensional Quantum Physics and Department of Physics, Tsinghua University, Beijing, China. + +<|ref|>text<|/ref|><|det|>[[113, 570, 883, 618]]<|/det|> +\(^{6}\) Departamento de Física Teorica de la Materia Condensada and Condensed Matter Physics Center (IFIMAC), Universidad Autónoma de Madrid, 28049 Madrid, Spain + +<|ref|>text<|/ref|><|det|>[[113, 630, 886, 678]]<|/det|> +\(^{7}\) Institute of High Performance Computing, Agency for Science, Technology and Research (A\*STAR), Connexis, 138632, Singapore + +<|ref|>text<|/ref|><|det|>[[113, 692, 682, 710]]<|/det|> +\*These authors contributed equally: Tingting Wu, Chongwu Wang, Guangwei Hu. + +<|ref|>text<|/ref|><|det|>[[113, 722, 720, 740]]<|/det|> +\*emails: fj.garcia@uam.es; wei.lei@ntu.edu.sg; qjwang@ntu.edu.sg; luoyu@ntu.edu.sg + +<|ref|>sub_title<|/ref|><|det|>[[115, 793, 182, 809]]<|/det|> +## Abstract + +<|ref|>text<|/ref|><|det|>[[113, 830, 883, 878]]<|/det|> +Van der Waals semiconductors exemplified by two- dimensional transition- metal dichalcogenides have promised next- generation atomically thin optoelectronics. Boosting their interaction with light is vital for practical + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 320]]<|/det|> +applications, especially in the quantum regime where ultrastrong coupling is highly demanded but not yet realized. Here we report ultrastrong exciton- plasmon coupling at room temperature in tungsten disulfide ( \(\mathrm{WS}_2\) ) layers loaded with a random multi- singular plasmonic metasurface deposited on a flexible polymer substrate. Different from seeking perfect metals or high- quality resonators, we create a unique type of metasurface with a dense array of singularities that can support nanometre- sized plasmonic hotspots to which several \(\mathrm{WS}_2\) excitons coherently interact. The associated normalized coupling strength is 0.12 for monolayer \(\mathrm{WS}_2\) and can be up to 0.164 for quadrilayers, showcasing the ultrastrong exciton- plasmon coupling and important for practical optoelectronic devices based on low- dimensional semiconductors. + +<|ref|>sub_title<|/ref|><|det|>[[114, 371, 211, 387]]<|/det|> +## Introduction + +<|ref|>text<|/ref|><|det|>[[113, 408, 885, 732]]<|/det|> +Exciton- polaritons owing to strong collective excitations of waves and electron- hole pairs, i.e., excitons have facilitated versatile applications in nonlinear optics, quantum photonics, condensed matter physics, and others. Various excitonic systems have been explored, including bulk III- V semiconductors and organic materials, which however either require cryogenic temperatures \(^{1,2}\) to avoid the ionization of excitons at high temperatures or are susceptible to bleaching effects \(^{3}\) . Recently, van der Waals transition- metal dichalcogenides (TMDs) have emerged as new candidates for stable and atomically thin excitonic systems, offering numerous advantages, including direct bandgaps at visible frequencies, large exciton binding energies, pronounced resonance strengths, narrow linewidths even beyond room temperature \(^{4,5}\) , and strongly reduced structural disorder. In addition, strong coupling (SC) between excitons in TMD monolayers and photonic resonances has been realized in all- dielectric microcavities with Fabry- Perot resonators \(^{6,7}\) , metal- based microcavities with reduced mode volumes \(^{8,9}\) , and plasmonic structures towards further enhancement of wave- matter couplings \(^{10,11}\) . + +<|ref|>text<|/ref|><|det|>[[114, 744, 885, 885]]<|/det|> +In a parallel scientific endeavor, ultrastrong coupling (USC) of light- matter interactions has been recently explored in various systems. Here, the normalized coupling strength, defined as \(\eta = g / \omega_{ex}\) where \(g\) is coupling strength and \(\omega_{ex}\) is bare excitation energy, should be larger than 0.1. Compared to SC, USC is a new regime of quantum light- matter interactions with faster control/response even at shorter device lifetimes and important for several quantum optoelectronic applications. However, since its early proposal \(^{12,13}\) , USC has been only observed + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 884, 260]]<|/det|> +in semiconductor quantum wells14, superconducting circuits15, Landau polaritons16,17, organic molecules18, phonons19, and plasmons20, most of which rely on coupling many dipoles to a cavity mode at cryogenic temperatures, due to the technical challenges in implementation. Realizing USC under ambient conditions remains a challenge. Meanwhile, USC in monolayer TMDs has never been reached, which impedes their promise for ultrathin quantum devices21. This is because of the weak nanoscale matter excitations and the lack of significant transition dipole moments (typically in a few tens of Debyees) at such atomic thicknesses. + +<|ref|>text<|/ref|><|det|>[[113, 272, 885, 565]]<|/det|> +Here, we report the first observation of USC between excitons in a TMD monolayer and surface plasmons in a random multi- singular flexible plasmonic metasurface at room temperature. The dense singular nanometre- sized gaps are created by the cold etching technique and support strong field concentration. Multiple \(\mathrm{WS}_2\) excitons can interact with these deep- nanoscale plasmonic hotspots, thus promoting ultrastrong exciton- plasmon couplings. We observe a normalized coupling strength of 0.082 in unstrained TMD monolayers, which, via strain engineering in the flexible substrate, can be further tuned over a wide range from 0.075 to 0.12, then entering the USC regime. Furthermore, by increasing the number of \(\mathrm{WS}_2\) layers, the exciton- plasmon interaction shows the increased normalized coupling strengths to 0.147 (trilayer) and 0.164 (quadrilayer) under strain. We believe that our reported strategy towards room- temperature USC in TMD multilayers could find applications in lasing, nonlinear optics, wearable optoelectronics, and other technologies in atomically thin platforms. + +<|ref|>sub_title<|/ref|><|det|>[[115, 615, 171, 630]]<|/det|> +## Results + +<|ref|>sub_title<|/ref|><|det|>[[115, 652, 695, 670]]<|/det|> +## Mechanism of USC in \(\mathbf{WS}_2\) multilayers loaded with multi-singular metasurface + +<|ref|>text<|/ref|><|det|>[[113, 689, 885, 900]]<|/det|> +We start from discussing the fundamental ways to boost light- matter interaction. In general, in the case of collective ensembles, the light- matter coupling strength \(g\) is related to the emitter number \((g \propto \sqrt{N})\) and the highly confined cavity field in a very small mode volume \((g \propto 1 / \sqrt{V})^{22,23}\) . Progress towards room- temperature SC with monolayer TMDs in dielectric- based microcavities has been limited by the inevitable increase in the emitter scattering rate \((\gamma \propto k_B T)\) and by difficulties in reducing the cavity mode volume in dielectric structures due to the diffraction limit. To improve mode confinement with smaller mode volume, surface plasmons have been used23,24. Hence, the pronounced room- temperature plasmon- exciton coupling is observed in periodic + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 885, 291]]<|/det|> +metallic structures supporting high- quality surface lattice resonances25,26 with collectively excited large numbers of TMD excitons, or in plasmonic resonators (such as dimers, gaps etc.) with strong field enhancement in a small mode volume27,28. However, plasmonic lattices suffer from field delocalization, while localized resonances in small gaps or sharp nanostructures, which are usually sparse (the corresponding number of interacting emitters is reduced), pose a challenge for fine fabrication and have a limited number of plasmonic hotspots. To date, exciton- plasmon coupling strength with monolayer TMDs is typically in the range of 30- 60 meV29- 31, far beyond the regime of USC, where the coupling strength should be greater than \(0.1\omega_{ex}\) ( \(\sim 200 \mathrm{meV}\) ). + +<|ref|>text<|/ref|><|det|>[[113, 303, 885, 565]]<|/det|> +Here we construct a cold- etched random multi- singular plasmonic metasurface grown on a flexible thermoplastic polyurethane (TPU) substrate (inset, left, Fig. 1a) as a platform to trigger USC coupling between excitons in two- dimensional (2D) TMDs and surface plasmons. Our metasurface supports dense nanometre- sized gaps as singularities, which are introduced by applying an appropriate mechanical loading to create additional nanometre- sized cracks in the fragment with saturated transferred stress (Fig. 1b), and/or by using biaxial intergranular fractures to drag the deflected fragment domain towards its neighbors at nanometre- sized spacings (Fig. 1c), see Methods and section 1 in the Supplementary Information (SI) for details. The packing density (number of singularities per unit area) is controlled by adjusting the biaxial elongations, as also shown in section 1 of SI. + +<|ref|>text<|/ref|><|det|>[[113, 577, 885, 777]]<|/det|> +Such a strategy offers several advantages. First, the natural nanometre- sized metasurface singularities, usually difficult to be obtained with traditional top- down nanofabrication methods such as electron- beam lithography, allow stronger field localizations and smaller mode volumes (Fig. 1b- f), boosting the coupling strength. Second, considering these high- density plasmonic hotspots plus the randomness of sharp features (hence the polarization insensitivity, see section 2 in SI for details), the average number of interacting excitons (usually characterized as in- plane dipoles) with the plasmonic mode (see the near field of the gap plasmons in Fig. 1f) is also increased. + +<|ref|>text<|/ref|><|det|>[[113, 790, 884, 871]]<|/det|> +To characterize our sample, statistical analysis of the scanning electron microscopy (SEM) images shows that the average number \((\bar{N})\) of nanometre- sized gaps within each unit area \((1 \mu \mathrm{m}^2)\) reaches 10 (Fig. 1d). We do not analyze the packing of gaps much smaller than 3 nm because SEM has difficulties in accurately capturing + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 355]]<|/det|> +such small gaps. A representative SEM image of a \(20\mathrm{nm}\) thick gold metasurface at the highest packing condition (with the peak \(\bar{N}\) in Fig. 1d) is shown in Fig. 1e. The fracture morphology also shows the isotropic property, where the average fragment domain size is almost the same in different directions; see detailed geometrical and corresponding optical isotopic properties in section 2 of SI. The large interaction between the fragments adjacent to the singularity produces a maximum electric field enhancement of around 100 (Fig. 1f, the color bar used is to clearly show the electric field at all hotspots) within the small mode volume \((V = 2.03\times 10^{4}\mathrm{nm}^{3})\) . A pronounced Rabi splitting (blue and purple lines, Fig. 1g) is thus observed, with the extracted normalized coupling strength \(\eta = 0.12\) (purple line, Fig. 1g) and an estimated number of 23.6 excitons coherently contributing to the interaction with the surface plasmon, demonstrating USC with the \(\mathrm{WS}_2\) monolayer. + +<|ref|>image<|/ref|><|det|>[[120, 401, 868, 735]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[112, 755, 884, 888]]<|/det|> +
Fig. 1| Ultrastrong coupling in \(\mathrm{WS}_2\) monolayer with a random multi-singular plasmonic metasurface. a, Schematic of a gold multi-singular metasurface with a dense array of nanometre-sized plasmonic gaps. Insets: left: schematic of a \(\mathrm{WS}_2\) monolayer integrated with the multi-singular metasurface where the \(\mathrm{WS}_2\) monolayer was mechanically exfoliated onto a PDMS tape and transferred onto the gold nanopattern using a dry transfer method; middle and right: two main paths for generating nanometre-sized gaps. Middle: the first path, a nanometre-sized crack
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 87, 886, 400]]<|/det|> +generates in the fragment with saturated transferred stresses. Right: the second path, adjacent fragment domains are dragged infinitely close together (at nanometre- sized spacings) by biaxial mechanical loadings. b, c, Transmission electron microscopy (TEM) images of a \(20 \mathrm{nm}\) thick gold multi- singular metasurface showing a dense array of nanometre- sized plasmonic gaps. The darker grey areas correspond to gold, the lighter ones to air. b refers to the first path, and c to the second. d, Average number \((\bar{N})\) of nanometre- sized ( \(\sim\) sub- 3 nm in the scanning electron microscopy image) plasmonic gaps per \(1 \mu \mathrm{m}^2\) in the \(20 \mathrm{nm}\) thick metasurface as a function of the second \((2^{\mathrm{nd}})\) elongation. Error bars are standard errors from multiple samples. e, Scanning electron microscopy images of the \(20 \mathrm{nm}\) - thick metasurface at \(80\%\) \(2^{\mathrm{nd}}\) elongation. The white box corresponds to the simulation area in f. f, The simulated near field of the gap plasmons. g, Dark- field scattering spectra of the \(\mathrm{WS}_2\) monolayer on polymer (thermoplastic polyurethane, black) and the plasmonic metasurface uncoupled (green) and coupled (blue for SC and purple for USC) to \(\mathrm{WS}_2\) excitons in arbitrary units (a.u.). + +<|ref|>sub_title<|/ref|><|det|>[[114, 455, 524, 473]]<|/det|> +## Ultrastrong exciton-plasmon coupling and its tunability + +<|ref|>text<|/ref|><|det|>[[113, 492, 886, 696]]<|/det|> +The normalized dark- field scattering response of our samples with various gold film thicknesses (ranging from 17 to \(24 \mathrm{nm}\) to selectively tune the plasmonic resonance to the \(\mathrm{WS}_2\) exciton energy) are plotted in Fig. 2a,b, where the samples are under no strain and \(- 2\%\) uniaxial strain, respectively. Herein, to quantify the coupling strength, we fit the spectrum to a coupled oscillator model \(^{32}\) to obtain the vacuum Rabi frequency and extract the polariton energies (guided by the blue and purple curves in Fig. 2a,b) from the scattering peaks in the spectra. The extracted characteristic polariton dispersion consists of lower \((\omega_{- })\) and upper \((\omega_{+})\) polariton branches (Fig. 2c) where the polariton energies are fitted as the eigenvalues of the full Hopfield Hamiltonian \(^{33,34}\) , yielding + +<|ref|>equation<|/ref|><|det|>[[360, 707, 874, 732]]<|/det|> +\[\left(\omega^{2} - \omega_{e x}^{2}\right)\left(\omega^{2} - \omega_{p l}^{2}\right) - 4g^{2}\omega^{2} = 0, \quad (1)\] + +<|ref|>text<|/ref|><|det|>[[113, 748, 886, 900]]<|/det|> +where counter- rotating and photon self- interaction terms are included. Here, \(\omega_{e x}\) and \(\omega_{p l}\) are the energies of the \(\mathrm{WS}_2\) excitons and the plasmonic mode, respectively, and \(g\) is the exciton- plasmon coupling strength. \(\omega_{p l}\) is calculated from the extracted \(\omega_{\pm}\) as \(\omega_{p l} = \omega_{+}\omega_{- } / \omega_{e x}\) . The fitting leads to the typical anticrossing behaviour, with \(g = 165.9 \mathrm{meV}\) for the no strain case and \(g = 240.4 \mathrm{meV}\) for \(- 2\%\) uniaxial strain, corresponding to a normalized coupling strength of \(\eta = 0.082\) and enhanced \(\eta = 0.12\) , respectively (Fig. 2c). This clearly indicates + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 87, 886, 355]]<|/det|> +that the system operates in the USC regime when uniaxial strain is applied. Similar normalized coupling strengths are observed in different batches of samples (see Fig. S12 in SI), demonstrating the robustness of our platform for observing USC. The normalized coupling strength in the \(\mathrm{WS}_2\) monolayer coupled to the multisingular metasurface dramatically exceeds that of the reported optimized conventional TMD monolayer in plasmonic systems (Fig. S13 in SI, measured under exactly the same experimental conditions), suggesting that the coherent interaction of multiple excitons with the nanometre- sized plasmonic hotspots supported by the metasurface is the key factor in achieving the observed high coupling strength. Notice that in our system, the SC condition (i.e., coupling strength exceeds loss rates) \(g > (\gamma_{pl} + \gamma_{ex}) / 4\) is satisfied, where the damping losses of plasmonic resonance and exciton emission are \(\gamma_{pl} = 380 \mathrm{meV}\) and \(\gamma_{ex} = 45 \mathrm{meV}\) , respectively. + +<|ref|>text<|/ref|><|det|>[[113, 367, 886, 660]]<|/det|> +Applications of USC to nonlinear optics include low- threshold frequency conversion \(^{35}\) and phase- matched optical amplification \(^{36}\) . To demonstrate the advantages of our system with tunable USC, here we exploit USC towards tunable resonant polariton- enhanced nonlinearity. It is known that the excitonic resonances in monolayer TMDs can facilitate second harmonic generation (SHG) \(^{37}\) . In our system, USC allows spectral splitting of the polaritonic resonance with tunable ground state energies and, consequently, leads to dispersive polariton- enhanced SHG in the \(\mathrm{WS}_2\) monolayer. Strong evidence for the pronounced SHG intensity enhancement and spectral splitting can be seen from the nonlinearity spectrum as a function of the pump wavelength under two different values of the applied strain (Fig. 2d). As can be seen in Fig. S14, the polariton- enhanced SHG from the \(\mathrm{WS}_2\) monolayer is around 15 times stronger than that on PDMS, and polarization- independent polariton- enhanced SHG is observed due to the isotropy of our multi- singular metasurface. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[155, 87, 845, 528]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 547, 884, 890]]<|/det|> +
Fig. 2| Ultrastrong exciton-plasmon coupling in WS \(_2\) monolayer. Dark-field scattering spectra under a 0 and b -2% uniaxial strain on the metasurface over different gold film thicknesses (from bottom to top, the gold film thickness \(\mathrm{(t_{Au})}\) increases from 17 nm to 24 nm). The scattering splits into lower and upper polariton branches, exhibiting level anticrossing. The vertical dashed lines refer to the WS \(_2\) exciton energy \(\omega_{ex} = 2.019 \mathrm{eV}\) . The right panels are calculated scattering spectra using the model of two coupled oscillators \(^{32}\) . c, Dispersion plots of the measured dark-field scattering spectra. The lower \((\omega_{-})\) and upper \((\omega_{+})\) polariton branches are extracted from the scattering spectra in a,b and fitted (solid lines) with a coupling strength of \(165.9 \mathrm{meV}\) at 0 strain (blue lines) and \(240.4 \mathrm{meV}\) at -2% strain (purple lines) in the full Hopfield Hamiltonian. The experimental spectral peaks are shown as triangles (0 strain) and circles (-2% strain). The exciton energy \((\omega_{ex})\) is shown as the horizontal dashed black line. The plasmonic mode energy \((\omega_{pl})\) is shown as the diagonal dashed blue (SC) and purple (USC) lines. The plasmonic response shifts to higher energy when compressive strain is applied, and the local field enhancement is largely enhanced due to the reduced gap size, and both effects lead to significant changes in the polaritonic dispersion. d, Normalized second
+ +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 883, 165]]<|/det|> +harmonic generation (SHG) intensity compared to scattering in the SC (upper, strain=0) and USC (lower, strain=- 2%) regime, respectively. The simultaneous emergence of energy splitting in both scattering and SHG spectra precludes Fano interference phenomena from being responsible for the observed anticrossing. + +<|ref|>text<|/ref|><|det|>[[113, 212, 886, 567]]<|/det|> +Furthermore, thanks to the flexible nature of the TPU polymer substrate, the mechanical bending in our system can be adjusted to modify the plasmonic gap size and the corresponding packing density for modulating exciton- plasmon coupling strength. Under upward (tensile strain)/downward (compressive strain) bending, both the plasmonic resonance (left, Fig. 3a) and the corresponding plasmonic enhancement in photoluminescence peak intensity (right, Fig. 3a) can be tuned. We notice that the small strain applied in this work has a negligible effect on the exciton energy (Fig. S16 in SI), because the \(\mathrm{WS}_2\) monolayer is loosely contacted with the surface of the plasmonic metasurface. Dark- field scattering spectra at different strains are rendered in Fig. 3b, with two polariton branches well fitted (black solid lines, Fig. 3b) by the full Hopfield Hamiltonian described above. Specifically, the normalized coupling strength varies gradually from 0.075 to 0.12 (over the strain range from \(+0.375\%\) to \(- 2\%\) , Fig. 3c), clearly demonstrating the tunability of USC. We expect even higher coupling strengths by further increasing the compressive strain, but to preserve the resilience and robustness of the flexible substrate, we have avoided attempting to increase the compressive strain beyond \(- 2\%\) . + +<|ref|>text<|/ref|><|det|>[[113, 577, 886, 784]]<|/det|> +One interesting implication of the USC is that the global vacuum energy of the system is modified with respect to the coupling strength (Fig. 3d) by dressing excitons with light33,38,39. In our system, the corresponding change in the ground state is calculated as \(\Delta E_G = \hbar (\omega_+ + \omega_- - \omega_{ex} - \omega_{pl}) / 2\) . Note that the ground state energy of a harmonic oscillator is half that of the transition energy40. The normalized ground- state energy variation \((\Delta E_G / E_G)\) versus strain was calculated and fitted to the extracted coupling strengths (Fig. 3c) and polariton energies \(\omega_{\pm}\) (Fig. 3b), yielding a modification of \(0.67\%\) (Fig. 3d). The absolute ground- state energy modification reaches \(13.4 \mathrm{meV}\) at \(- 2\%\) compressive strain. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[220, 88, 772, 465]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 483, 885, 797]]<|/det|> +
Fig. 3| Tunable ultrastrong exciton-plasmon coupling in WS \(_2\) monolayer. a, Mechanically tunable plasmonic resonance (wavelength at \(\lambda_{pl}\) ) and plasmonic enhancement in photoluminescence peak intensity of the strongly coupled plasmonic system. b, Normalized dark-field scattering intensity as a function of excitation energy and strain. The horizontal purple line marks the onset of USC. The black solid lines are the extracted polariton energies of the lower (LPB) and upper (UPB) polariton branches. The white dashed line is the WS \(_2\) exciton energy ( \(\omega_{ex} = 2.019 \mathrm{eV}\) ). c, Normalized coupling strength as a function of strain. The onset of USC is marked by the horizontal dashed line. d, Ground-state energy modification as a function of strain. The black dashed line is the calculation result from the coupling strength ( \(g\) ) and the polariton energies ( \(\omega_{+}\) , \(\omega_{- }\) ) from the fitting to the full Hopfield Hamiltonian. The absolute change in ground-state energy reaches 13.4 meV at -2% compressive strain. Error bars in c and d result from the variation in dark-field scattering from multiple batches of samples and are extracted from the standard error of the fit.
+ +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 412, 107]]<|/det|> +## Ultrastrong coupling in \(\mathbf{W}\mathbf{S}_{2}\) multilayers + +<|ref|>text<|/ref|><|det|>[[112, 125, 886, 578]]<|/det|> +Last, we show that the coupling strength in USC is also relevant with the layer numbers of \(\mathbf{W}\mathbf{S}_{2}\) . We transferred trilayer and quadralayer \(\mathbf{W}\mathbf{S}_{2}\) flakes onto several metasurfaces with different gold layer thicknesses (19 nm to 23 nm) on flexible PDMS substrates. The dispersion curves of the scattering spectra are shown in Fig. 4, showing the increased energy splitting between polaritons compared to the monolayer. In trilayer \(\mathbf{W}\mathbf{S}_{2}\) (Fig. 4a,b), the Rabi splitting \((\Omega_{R})\) exceeds \(397.9 \mathrm{meV}\) \((\eta \sim 0.108)\) and \(563.9 \mathrm{meV}\) \((\eta \sim 0.147)\) under 0 and \(- 2\%\) compressive strain, respectively; while in quadralayer \(\mathbf{W}\mathbf{S}_{2}\) (Fig. 4c,d), \(\Omega_{R}\) can be as high as \(449.7 \mathrm{meV}\) \((\eta \sim 0.12)\) and \(634.7 \mathrm{meV}\) \((\eta \sim 0.164)\) accordingly. Moving from the monolayer to the quadralayer cases, the normalized coupling strength increases with the number of layers, but less rapidly than predicted by a square root function, as shown in Fig. 4e, because the coupling strength depends on both the number of excitons and the spatial overlap of the confined electric field with the \(\mathbf{W}\mathbf{S}_{2}\) layers (which decreases as the number of \(\mathbf{W}\mathbf{S}_{2}\) layer increases). Thus, we have successfully combined monolayer and multilayer \(\mathbf{W}\mathbf{S}_{2}\) flakes with multi- singular metasurfaces to create a large set of ultrastrongly coupled exciton- plasmon systems under ambient conditions. Moreover, the cooperativity, defined as \(\mathrm{C} = 4g^{2} / \gamma_{p1}\gamma_{ex}\) (a key figure of merit to characterize the coupling regimes of light- matter interaction) in our samples outperforms all room- temperature TMD- based platforms ( \(\mathrm{C} = 28.4\) in quadralayer \(\mathbf{W}\mathbf{S}_{2}\) ). + +<|ref|>text<|/ref|><|det|>[[113, 588, 886, 821]]<|/det|> +We compare our system with the previous reports of light- matter coupling, as summarized in Table 1 in SI for details. The normalized coupling coefficients of other strongly coupled systems are typically in the range of 0.015- 0.03 (0.015- 0.04) for monolayer \(^{29 - 31}\) (multilayer \(^{27,28,41,42}\) ) TMDs, 0.02- 0.095 for organic molecules (less than \(20 \mathrm{nm}\) thick) \(^{43 - 45}\) , and 0.04- 0.09 for quantum dots \(^{46 - 48}\) . Existing room- temperature USC systems are typically implemented using organic molecules, and the reported thickness must be greater than \(60 \mathrm{nm}^{18,49,50}\) . Landau polaritons and superconducting circuits are not included in this comparison because their transition energies are far below those of TMD excitons. Our platform for USC, with the emitter thickness of less than \(1 \mathrm{nm}\) , is thus unique and holds great promise for next- generation atomically thin optoelectronics in the visible. + +<--- Page Split ---> +<|ref|>image<|/ref|><|det|>[[172, 88, 822, 457]]<|/det|> +<|ref|>image_caption<|/ref|><|det|>[[113, 475, 886, 728]]<|/det|> +
Fig. 4| Ultrastrong coupling in multilayer WS2 flakes. a-d, Dark-field scattering spectra of a multilayer WS2 plasmonic system. Dashed white lines denote the exciton energy ( \(\omega_{ex} = 2.01 \mathrm{eV}\) for a,b; \(\omega_{ex} = 2.003 \mathrm{eV}\) for c,d;), and the diagonal dashed orange and purple lines indicate the plasmon energy ( \(\omega_{pl}\) ). The open circles represent lower ( \(\omega_{-}\) ) and upper ( \(\omega_{+} \mathrm{p}\) ) polariton energies obtained from the dark-field scattering spectra of individual plasmonic systems. The solid lines show the lower and upper polariton dispersions using the full Hopfield Hamiltonian. 3L and 4L represent the trilayer and quadrilayer cases. e, Normalized coupling strength ( \(g / \omega_{ex}\) ) as a function of the WS2 layers under 0 (orange) and -2% compressive strain (purple). The horizontal dashed line marks the onset of USC. Error bars are derived from the variation in dark-field scattering from different batches of samples and are extracted from the standard error of the fit.
+ +<|ref|>sub_title<|/ref|><|det|>[[115, 787, 195, 802]]<|/det|> +## Discussion + +<|ref|>text<|/ref|><|det|>[[115, 824, 883, 872]]<|/det|> +In summary, we have shown the ultrastrong exciton- plasmon coupling at room temperature in WS2 multilayers coupled to random multi- singular plasmonic metasurfaces formed by cold- etching, which can be further tuned + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 885, 260]]<|/det|> +by mechanical strain in a flexible substrate. Our results promise the USC in the atomic layer limit under ambient conditions, which could be further extended to scenarios of different metals or doped semiconductors and complex 2D heterostructures for more exotic complex light- matter interactions, such as tunable trion polaritons, van der Waals heterostructure polaritons, and moiré induced optical nonlinearities. Our work could also lead to new insights in fundamental science and potential applications in the fields of ultralow power nonlinear nanophotonics, analytical chemistry, quantum optics, among others. + +<|ref|>sub_title<|/ref|><|det|>[[114, 310, 182, 326]]<|/det|> +## Methods + +<|ref|>sub_title<|/ref|><|det|>[[115, 349, 293, 365]]<|/det|> +## Experimental structure. + +<|ref|>text<|/ref|><|det|>[[112, 377, 886, 732]]<|/det|> +\(\mathrm{WS}_2\) coupled multi- singular plasmonic metasurface: A thin gold film was deposited on an amorphous poly(etherimide) (PEI) polymer substrate ( \(125\mu \mathrm{m}\) thick) using an electron beam evaporator at a deposition rate of \(2\mathrm{\AA}\mathrm{s}^{- 1}\) . During the evaporation, the temperature of the vacuum chamber was maintained below \(60^{\circ}\mathrm{C}\) throughout the process to prevent thermal expansion or deformation of the PEI film due to the build- in stress, which would cause wrinkles or defects in the gold film. For the cold- etching process, the first stretch was performed by stretching the gold/PEI film in the \(x\) - direction and stopping when the necks extended along the length of the film. The second stretch was performed by re- stretching the as- fabricated film in the \(y\) - direction and the elongation extent was controlled to obtain gold nanopatterns with different second elongations. After the cold- etching process, the resulting gold nanopatterns were transferred to a flexible TPU substrate ( \(2\mathrm{mm}\) thick) by a dry peel- off method. \(\mathrm{WS}_2\) monolayers and multilayers were mechanically exfoliated from the commercial bulk crystals onto a PDMS tape and transferred to the gold multi- singular metasurface (on TPU) through a dry transfer method (see Fig.S1 in SI for details). + +<|ref|>sub_title<|/ref|><|det|>[[115, 775, 307, 791]]<|/det|> +## Optical characterizations. + +<|ref|>text<|/ref|><|det|>[[114, 805, 884, 884]]<|/det|> +All optical characterizations were performed in the reflective geometry at room temperature. For all linear polarization measurements, the metasurface system was rotated and the polarization- dependent components were analyzed. For photoluminescence measurements, a \(532\mathrm{nm}\) diode laser was used to excite the sample. A + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[114, 88, 884, 199]]<|/det|> +\(100\times\) microscope objective lens (numerical aperture (NA) D 0.75) was used and the incident laser power was \(\sim 100\mu \mathrm{W}\) with a laser spot size of \(\sim 4\mu \mathrm{m}\) . Dark- field scattering measurements were carried out using a hyperspectral imaging system with a broadband halogen lamp as the light source. A \(50\times\) objective (NA D 0.55) was used and the incident light power was \(\sim 20\mu \mathrm{W}\) with a laser spot size of \(\sim 5\mu \mathrm{m}\) . + +<|ref|>sub_title<|/ref|><|det|>[[114, 241, 295, 258]]<|/det|> +## Numerical Calculations. + +<|ref|>text<|/ref|><|det|>[[113, 270, 886, 389]]<|/det|> +Commercial finite- difference time- domain software was used to calculate the field enhancement of the gap plasmons. The permittivity \(\mathrm{WS}_2\) is modeled as a Lorentzian oscillator \(\epsilon (\omega) = 1 + \sum_{k = 1}^{N}f_{k} / (\omega_{k}^{2} - \omega^{2} - i\gamma_{k}\omega)\) , with \(f_{k}\) , \(\gamma_{k}\) and \(\omega_{k}\) being the oscillator strength, the linewidth of the \(k\) th oscillator and the oscillation energy, respectively. The permittivity of the gold was from the software database. + +<|ref|>sub_title<|/ref|><|det|>[[114, 431, 282, 448]]<|/det|> +## Hopfield Hamiltonian. + +<|ref|>text<|/ref|><|det|>[[113, 461, 797, 480]]<|/det|> +The Hopfield Hamiltonian of our plasmonic system in the USC regime contains three main blocks: + +<|ref|>equation<|/ref|><|det|>[[408, 494, 881, 514]]<|/det|> +\[H = H_{s y s} + H_{\mathrm{int}} + H_{A^{2}} \quad (2)\] + +<|ref|>text<|/ref|><|det|>[[113, 532, 485, 550]]<|/det|> +with the Hamiltonian of the closed- plasmonic system + +<|ref|>equation<|/ref|><|det|>[[395, 563, 881, 587]]<|/det|> +\[H_{s y s} = \mathrm{h}\omega_{e x}\hat{a}^{\dagger}\hat{a} +\mathrm{h}\omega_{\rho l}\hat{b}^{\dagger}\hat{b} \quad (3)\] + +<|ref|>text<|/ref|><|det|>[[113, 605, 337, 621]]<|/det|> +and the interaction Hamiltonian + +<|ref|>equation<|/ref|><|det|>[[391, 634, 881, 661]]<|/det|> +\[H_{\mathrm{int}} = \mathrm{i}\hbar \mathrm{g}\left(\hat{a}^{\dagger} + \hat{a}\right)\left(\hat{b} -\hat{b}^{\dagger}\right) \quad (4)\] + +<|ref|>text<|/ref|><|det|>[[113, 680, 454, 697]]<|/det|> +and the photon self- interaction \((A^{2})\) Hamiltonian + +<|ref|>equation<|/ref|><|det|>[[387, 710, 881, 747]]<|/det|> +\[H_{A^{2}} = \frac{\mathrm{h}\mathrm{g}^{2}}{\omega_{e x}}\left(\hat{a}^{\dagger} + \hat{a}\right)\left(\hat{a}^{\dagger} + \hat{a}\right) \quad (5)\] + +<|ref|>text<|/ref|><|det|>[[113, 764, 884, 814]]<|/det|> +\(\hat{a}^{\dagger}\) and \(\hat{a}\) are the exciton creation and annihilation operators, respectively, and \(\hat{b}^{\dagger}\) and \(\hat{b}\) are those of the localized plasmons. Since \(H\) is invariant under translation we define + +<|ref|>equation<|/ref|><|det|>[[400, 829, 881, 848]]<|/det|> +\[\hat{c} = w\hat{a} +x\hat{b} +y\hat{a}^{\dagger} +z\hat{b}^{\dagger} \quad (6)\] + +<|ref|>text<|/ref|><|det|>[[113, 865, 225, 881]]<|/det|> +And we can get + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[440, 88, 881, 108]]<|/det|> +\[[\hat{c},H] = E\hat{c} \quad (7)\] + +<|ref|>text<|/ref|><|det|>[[113, 124, 476, 141]]<|/det|> +The eigenvalue problem is rewritten in matrix form, + +<|ref|>equation<|/ref|><|det|>[[154, 153, 881, 237]]<|/det|> +\[\begin{array}{r}{\left[ \begin{array}{c c c c c c c c c c c c c c c c c c c c c}{{\omega_{p l} + 2g^{2} / \omega_{e x} - i\gamma_{p l} / 2}} & {-i g} & {-2g^{2} / \omega_{e x}} & {-i g} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {i g} & {\omega_{e x} - i\gamma_{e x} / 2} & {-i g} & 0 & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {2g^{2} / \omega_{e x}} & {-i g} & {-a_{p l} - 2g^{2} / \omega_{e x} - i\gamma_{p l}^{*} / 2} & {-i g} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {-i g} & 0 & {i g} & {-a_{e x} - i\gamma_{e x}^{*} / 2} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {}\\ {} & {} & {} & {} & {} & {} & {} & {} & {} & {} & {} \end{array} \right]\left[ \begin{array}{c}{w}\\ {x}\\ {y}\\ {z} \end{array} \right] = E\left[ \begin{array}{c}{w}\\ {x}\\ {y}\\ {z} \right] \end{array} \right]} \end{array} \quad (8)\] + +<|ref|>text<|/ref|><|det|>[[113, 252, 727, 270]]<|/det|> +Two eigenvalues \(\omega_{\mp}\) of the above matrix are the positive solutions of the equation \(^{30,45,46}\) , + +<|ref|>equation<|/ref|><|det|>[[362, 282, 881, 308]]<|/det|> +\[\left(\omega^{2} - \omega_{e x}^{2}\right)\left(\omega^{2} - \omega_{p l}^{2}\right) - 4g^{2}\omega^{2} = 0 \quad (9)\] + +<|ref|>text<|/ref|><|det|>[[113, 325, 270, 342]]<|/det|> +from which we obtain + +<|ref|>equation<|/ref|><|det|>[[432, 358, 881, 377]]<|/det|> +\[\omega_{c}\omega_{-} = \omega_{p l}\omega_{e x} \quad (10)\] + +<|ref|>text<|/ref|><|det|>[[113, 394, 883, 443]]<|/det|> +To interpret our experimental data in the SC regime, we obtain the analytical polariton dispersion energies with a coupled oscillator model with the Hamiltonian + +<|ref|>equation<|/ref|><|det|>[[366, 454, 881, 500]]<|/det|> +\[H = \left[ \begin{array}{cc}{\omega_{p l} - i\gamma_{p l} / 2} & {g}\\ {g} & {\omega_{e x} - i\gamma_{e x} / 2} \end{array} \right] \quad (11)\] + +<|ref|>text<|/ref|><|det|>[[113, 515, 133, 529]]<|/det|> +as + +<|ref|>equation<|/ref|><|det|>[[247, 543, 881, 590]]<|/det|> +\[\omega_{\pm} = \frac{1}{2}\Bigg(\omega_{p l} - i\frac{\gamma_{p l}}{2} +\omega_{e x} - i\frac{\gamma_{e x}}{2}\Bigg)\pm \sqrt{8^{2} + \frac{1}{4}\Bigg(\omega_{p l} - i\frac{\gamma_{p l}}{2} -\omega_{e x} + i\frac{\gamma_{e x}}{2}\Bigg)^{2}} \quad (12)\] + +<|ref|>text<|/ref|><|det|>[[113, 607, 883, 675]]<|/det|> +Note that the complex Rabi splitting can be calculated as \(\Omega_{R} = \omega_{+} - \omega_{- } =\) \(2\sqrt{g^{2} + \frac{1}{4}\big(\omega_{p l} - i\frac{\gamma_{p l}}{2} -\omega_{e x} + i\frac{\gamma_{e x}}{2}\big)^{2}}\) , where for zero detuning and zero damping the rabi splitting is \(\Omega_{R} = 2g\) . + +<|ref|>sub_title<|/ref|><|det|>[[113, 718, 602, 736]]<|/det|> +## Strain model: estimation on gap size in the plasmonic metasurface + +<|ref|>text<|/ref|><|det|>[[113, 748, 883, 832]]<|/det|> +The Young's modulus, ultimate strength and thickness of the film are given by \(E\) , \(\sigma_{\mathrm{s}}\) and \(h\) , respectively. Assume that the interfacial shear stress is a constant \(\tau_{0}\) for any interfacial displacement and the substrate stretching is accompanied by a steady neck propagation with the necked stretch ration of \(\lambda_{n}\) . + +<|ref|>text<|/ref|><|det|>[[113, 844, 883, 892]]<|/det|> +The film fractures sequentially into fragments following the propagating neck. For a fragment of film, its size in the initial configuration (undeformed configuration) can be determined as + +<--- Page Split ---> +<|ref|>equation<|/ref|><|det|>[[414, 88, 881, 125]]<|/det|> +\[L_{0} = \frac{E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (13)\] + +<|ref|>text<|/ref|><|det|>[[113, 142, 883, 190]]<|/det|> +The portion of substrate with the same initial position and length as the film fragment experiences full necking and now has the size + +<|ref|>equation<|/ref|><|det|>[[375, 203, 881, 240]]<|/det|> +\[L_{\mathrm{subs}} = \lambda_{\mathrm{n}}L_{0} = \frac{\lambda_{\mathrm{n}}E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (14)\] + +<|ref|>text<|/ref|><|det|>[[113, 257, 883, 306]]<|/det|> +Assume that the residual stress in the film fragment is distributed linearly within the fragment and has the maximum value of \(\sigma_{\mathrm{r}}\) . Then we have + +<|ref|>equation<|/ref|><|det|>[[407, 320, 881, 357]]<|/det|> +\[L_{0} = \frac{2E h}{\tau_{0}}\big(1 - e^{-\sigma_{\mathrm{s}} / E}\big). \quad (15)\] + +<|ref|>text<|/ref|><|det|>[[113, 374, 358, 392]]<|/det|> +With Eqs. (13) and (15), we obtain + +<|ref|>equation<|/ref|><|det|>[[416, 405, 881, 442]]<|/det|> +\[\sigma_{\mathrm{r}} = E\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}. \quad (16)\] + +<|ref|>text<|/ref|><|det|>[[113, 457, 495, 475]]<|/det|> +Therefore, the size of the film fragment after rupture is + +<|ref|>equation<|/ref|><|det|>[[372, 487, 881, 525]]<|/det|> +\[L_{\mathrm{film}} = \frac{2\sigma_{\mathrm{r}}h}{\tau_{0}} = \frac{2E h}{\tau_{0}}\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}, \quad (17)\] + +<|ref|>text<|/ref|><|det|>[[113, 540, 884, 587]]<|/det|> +and the distance between neighboring film fragments, i.e., the averaged gap size in metasurface, can be expressed as + +<|ref|>equation<|/ref|><|det|>[[294, 600, 881, 638]]<|/det|> +\[\overline{{L}}_{\mathrm{gap}} = L_{\mathrm{subs}} - L_{\mathrm{film}} = \frac{E h}{\tau_{0}}\bigg(\lambda_{\mathrm{n}} - \lambda_{\mathrm{n}}e^{-\sigma_{\mathrm{s}} / E} - 2\ln \frac{2}{1 + e^{-\sigma_{\mathrm{s}} / E}}\bigg). \quad (18)\] + +<|ref|>sub_title<|/ref|><|det|>[[114, 696, 239, 712]]<|/det|> +## Data availability + +<|ref|>text<|/ref|><|det|>[[114, 732, 881, 781]]<|/det|> +The data supporting the current study in the paper are included in the paper and/or the Supplementary Materials. Additional data related to this paper can be requested from the corresponding authors. + +<|ref|>sub_title<|/ref|><|det|>[[114, 833, 198, 848]]<|/det|> +## References + +<|ref|>text<|/ref|><|det|>[[112, 870, 855, 888]]<|/det|> +1. Diederichs, C. et al. Parametric oscillation in vertical triple microcavities. Nature 440, 904-907 (2006). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[111, 88, 885, 137]]<|/det|> +2. Duh, Y. S. et al. Giant photothermal nonlinearity in a single silicon nanostructure. Nat. Commun. 11, 1-9 (2020). + +<|ref|>text<|/ref|><|det|>[[112, 149, 884, 199]]<|/det|> +3. Ha, T. & Tinnefeld, P. Photophysics of fluorescent probes for single-molecule biophysics and super-resolution imaging. Annu. Rev. Phys. Chem. 63, 595-617 (2012). + +<|ref|>text<|/ref|><|det|>[[112, 210, 884, 259]]<|/det|> +4. Zhang, L., Gogna, R., Burg, W., Tutuc, E. & Deng, H. Photonic-crystal exciton-polaritons in monolayer semiconductors. Nat. Commun. 9, 1-8 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 270, 884, 320]]<|/det|> +5. Gu, J., Chakraborty, B., Khatoniar, M. & Menon, V. M. A room-temperature polariton light-emitting diode based on monolayer \(\mathrm{WS}_2\) . Nat. Nanotechnol. 14, 1024-1028 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 332, 884, 381]]<|/det|> +6. Sidler, M. et al. Fermi polaron-polaritons in charge-tunable atomically thin semiconductors. Nat. Phys. 13, 255-261 (2017). + +<|ref|>text<|/ref|><|det|>[[112, 393, 886, 443]]<|/det|> +7. Dufferwiel, S. et al. Exciton-polaritons in van der Waals heterostructures embedded in tunable microcavities. Nat. Commun. 6, 8579 (2015). + +<|ref|>text<|/ref|><|det|>[[112, 454, 884, 504]]<|/det|> +8. Kaliteevski, M. et al. Tamm plasmon-polaritons: Possible electromagnetic states at the interface of a metal and a dielectric Bragg mirror. Phys. Rev. B 76, 165415 (2007). + +<|ref|>text<|/ref|><|det|>[[112, 515, 884, 565]]<|/det|> +9. Lundt, N. et al. Room-temperature Tamm-plasmon exciton-polaritons with a \(\mathrm{WSe}_2\) monolayer. Nat. Commun. 7, 1-6 (2016). + +<|ref|>text<|/ref|><|det|>[[112, 576, 884, 626]]<|/det|> +10. Liu, X. et al. Nonlinear valley phonon scattering under the strong coupling regime. Nat. Mater. 20, 1210-1215 (2021). + +<|ref|>text<|/ref|><|det|>[[112, 637, 884, 686]]<|/det|> +11. Liu, L. et al. Plasmon-induced thermal tuning of few-exciton strong coupling in 2D atomic crystals. Optica 8, 1416-1423 (2021). + +<|ref|>text<|/ref|><|det|>[[112, 698, 884, 747]]<|/det|> +12. Irish, E. K. Generalized rotating-wave approximation for arbitrarily large coupling. Phys. Rev. Lett. 99, 173601 (2007). + +<|ref|>text<|/ref|><|det|>[[112, 759, 884, 808]]<|/det|> +13. Anappara, A. A. et al. Signatures of the ultrastrong light-matter coupling regime. Phys. Rev. B 79, 201303 (2009). + +<|ref|>text<|/ref|><|det|>[[112, 820, 872, 840]]<|/det|> +14. Günter, G. et al. Sub-cycle switch-on of ultrastrong light-matter interaction. Nature 458, 178-181 (2009). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[113, 88, 884, 135]]<|/det|> +15. Niemczyk, T. et al. Circuit quantum electrodynamics in the ultrastrong-coupling regime. Nat. Phys. 6, 772-776 (2010). + +<|ref|>text<|/ref|><|det|>[[114, 149, 882, 198]]<|/det|> +16. Scalari, G. et al. Ultrastrong coupling of the cyclotron transition of a 2D electron gas to a THz metamaterial. Science 335, 1323-1326 (2012). + +<|ref|>text<|/ref|><|det|>[[114, 210, 883, 259]]<|/det|> +17. Zhang, Q. et al. Collective non-perturbative coupling of 2D electrons with high-quality-factor terahertz cavity photons. Nat. Phys. 12, 1005-1011 (2016). + +<|ref|>text<|/ref|><|det|>[[114, 271, 883, 320]]<|/det|> +18. Gambino, S. et al. Exploring light-matter interaction phenomena under ultrastrong coupling regime. ACS Photon. 1, 1042-1048 (2014). + +<|ref|>text<|/ref|><|det|>[[114, 333, 883, 381]]<|/det|> +19. Barra-Burillo, M. et al. Microcavity phonon polaritons from the weak to the ultrastrong phonon-photon coupling regime. Nat. Commun. 12, 6206 (2021). + +<|ref|>text<|/ref|><|det|>[[114, 394, 883, 443]]<|/det|> +20. Mueller, N. S. et al. Deep strong light-matter coupling in plasmonic nanoparticle crystals. Nature 583, 780-784 (2020). + +<|ref|>text<|/ref|><|det|>[[114, 455, 883, 504]]<|/det|> +21. Kurman, Y. & Kaminer, I. Tunable bandgap renormalization by nonlocal ultra-strong coupling in nanophotonics. Nat. Phys. 16, 868-874 (2020). + +<|ref|>text<|/ref|><|det|>[[114, 516, 883, 564]]<|/det|> +22. Shalabney, A. et al. Coherent coupling of molecular resonators with a microcavity mode. Nat. Commun. 6, 5981 (2015). + +<|ref|>text<|/ref|><|det|>[[114, 577, 882, 625]]<|/det|> +23. Chikkaraddy, R. et al. Single-molecule strong coupling at room temperature in plasmonic nanocavities. Nature 535, 127-130 (2016). + +<|ref|>text<|/ref|><|det|>[[114, 638, 882, 686]]<|/det|> +24. Halas, Naomi J. et al. Plasmons in strongly coupled metallic nanostructures. Chem. Rev. 111, 3913-3961 (2011). + +<|ref|>text<|/ref|><|det|>[[114, 699, 883, 747]]<|/det|> +25. Wang, S. et al. Coherent coupling of \(\mathrm{WS}_2\) monolayers with metallic photonic nanostructures at room temperature. Nano Lett. 16, 4368-4374 (2016). + +<|ref|>text<|/ref|><|det|>[[114, 760, 883, 808]]<|/det|> +26. Liu, W. et al. Strong exciton-plasmon coupling in \(\mathrm{MoS}_2\) coupled with plasmonic lattice. Nano Lett. 16, 1262-1269 (2016). + +<|ref|>text<|/ref|><|det|>[[114, 821, 883, 869]]<|/det|> +27. Kleemann, M. E. et al. Strong-coupling of \(\mathrm{WSe}_2\) in ultra-compact plasmonic nanocavities at room temperature. Nat. Commun. 8, 1296 (2017). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 137]]<|/det|> +28. Stuhrenberg, M. et al. Strong light-matter coupling between plasmons in individual gold bi-pyramids and excitons in mono-and multilayer WSe₂. Nano Lett. 18, 5938-5945 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 150, 884, 228]]<|/det|> +29. Jiang, Y., Wang, H., Wen, S., Chen, H., & Deng, S. Resonance coupling in an individual gold nanorod-monolayer WS₂ heterostructure: photoluminescence enhancement with spectral broadening. ACS Nano 14, 13841-13851 (2020). + +<|ref|>text<|/ref|><|det|>[[112, 241, 884, 289]]<|/det|> +30. Sang, Y. et al. Tuning of two-dimensional plasmon-exciton coupling in full parameter space: a polaritonic non-Hermitian system. Nano Lett. 21, 2596-2602 (2021). + +<|ref|>text<|/ref|><|det|>[[112, 302, 884, 350]]<|/det|> +31. Geisler, M. et al. Single-crystalline gold nanodisks on WS₂ mono-and multilayers for strong coupling at room temperature. ACS Photon. 6, 994-1001 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 363, 884, 411]]<|/det|> +32. Leng, H., Szychowski, B., Daniel, M. C., & Pelton, M. Strong coupling and induced transparency at room temperature with single quantum dots and gap plasmons. Nat. Commun. 9, 4012 (2018). + +<|ref|>text<|/ref|><|det|>[[112, 424, 884, 472]]<|/det|> +33. Ciuti, C. & Carusotto, I. Input-output theory of cavities in the ultrastrong coupling regime: The case of time-independent cavity parameters. Phys. Rev. A 74, 033811 (2006). + +<|ref|>text<|/ref|><|det|>[[112, 485, 884, 533]]<|/det|> +34. Hopfield, J. Theory of the contribution of excitons to the complex dielectric constant of crystals. Phys. Rev. 112, 1555 (1958). + +<|ref|>text<|/ref|><|det|>[[112, 546, 884, 594]]<|/det|> +35. Sánchez-Burillo, E., Zueco, D., Garcia-Ripoll, J. & Martin-Moreno, L. Scattering in the ultrastrong regime: nonlinear optics with one photon. Phys. Rev. Lett. 113, 263604 (2014). + +<|ref|>text<|/ref|><|det|>[[112, 607, 884, 655]]<|/det|> +36. Mathew, J. P., Patel, R. N., Borah, A., Vijay, R. & Deshmukh, M. M. Dynamical strong coupling and parametric amplification of mechanical modes of graphene drums. Nat. Nanotechnol. 11, 747-751 (2016). + +<|ref|>text<|/ref|><|det|>[[112, 668, 884, 715]]<|/det|> +37. Seyler, K. L. et al. Electrical control of second-harmonic generation in a WSe₂ monolayer transistor. Nat. Nanotechnol. 10, 407-411 (2015). + +<|ref|>text<|/ref|><|det|>[[112, 728, 884, 776]]<|/det|> +38. Ciuti, C., Bastard, G. & Carusotto, I. Quantum vacuum properties of the intersubband cavity polariton field. Phys. Rev. B 72, 115303 (2005). + +<|ref|>text<|/ref|><|det|>[[112, 790, 884, 837]]<|/det|> +39. Mahboob, I. & Yamaguchi, H. Bit storage and bit flip operations in an electromechanical oscillator. Nat. Nanotechnol. 3, 275-279 (2008). + +<--- Page Split ---> +<|ref|>text<|/ref|><|det|>[[112, 88, 884, 137]]<|/det|> +40. Baranov, D. G. et al. Ultrastrong coupling between nanoparticle plasmons and cavity photons at ambient conditions. Nat. Commun. 11, 2715 (2020). + +<|ref|>text<|/ref|><|det|>[[112, 149, 884, 198]]<|/det|> +41. Wang, S. et al. Limits to strong coupling of excitons in multilayer \(\mathrm{WS}_2\) with collective plasmonic resonances. ACS Photon. 6, 286-293 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 210, 884, 259]]<|/det|> +42. Yankovich, A. B. et al. Visualizing spatial variations of plasmon-exciton polaritons at the nanoscale using electron microscopy. Nano Lett. 19, 8171-8181 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 271, 884, 320]]<|/det|> +43. Zengin, G. et al. Realizing strong light-matter interactions between single-nanoparticle plasmons and molecular excitons at ambient conditions. Phys. Rev. Lett. 114, 157401 (2015). + +<|ref|>text<|/ref|><|det|>[[112, 332, 884, 411]]<|/det|> +44. Wersall, M., Cuadra, J., Antosiewicz, T. J., Balci, S., & Shegai, T. Observation of mode splitting in photoluminescence of individual plasmonic nanoparticles strongly coupled to molecular excitons. Nano Lett., 17, 551-558 (2017). + +<|ref|>text<|/ref|><|det|>[[112, 424, 884, 473]]<|/det|> +45. Liu, R. et al. Strong light-matter interactions in single open plasmonic nanocavities at the quantum optics limit. Phy. Rev. Lett. 118, 237401 (2017). + +<|ref|>text<|/ref|><|det|>[[112, 485, 883, 534]]<|/det|> +46. Santhosh, K. et al. Vacuum Rabi splitting in a plasmonic cavity at the single quantum emitter limit. Nat. Commun. 7, 11823, (2016). + +<|ref|>text<|/ref|><|det|>[[112, 546, 883, 595]]<|/det|> +47. Bitton, O. et al. Vacuum Rabi splitting of a dark plasmonic cavity mode revealed by fast electrons. Nat. Commun. 11, 487, (2020). + +<|ref|>text<|/ref|><|det|>[[112, 607, 883, 655]]<|/det|> +48. Park, K. et al. Tip-enhanced strong coupling spectroscopy, imaging, and control of a single quantum emitter. Sci. Adv. 5, 5931 (2019). + +<|ref|>text<|/ref|><|det|>[[112, 667, 884, 717]]<|/det|> +49. Gubbin, C. R., Maier, S. A., & Kéna-Cohen, S. Low-voltage polariton electroluminescence from an ultrastrongly coupled organic light-emitting diode. Appl. Phys. Lett., 104, 85 (2014). + +<|ref|>text<|/ref|><|det|>[[112, 728, 884, 778]]<|/det|> +50. Kéna-Cohen, S., Maier, S. A., & Bradley, D. D. Ultrastrongly Coupled Exciton-Polaritons in Metal-Clad Organic Semiconductor Microcavities. Adv. Opt. Mat., 1, 827-833 (2013). + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[115, 90, 260, 106]]<|/det|> +## Acknowledgements + +<|ref|>text<|/ref|><|det|>[[112, 125, 886, 420]]<|/det|> +AcknowledgementsThis work was supported by the Singapore National Research Foundation Competitive Research Program (NRF- CRP22- 2019- 0006, NRF- CRP23- 2019- 0007 and NRF- CRP22- 2019- 0007), the Singapore Ministry of Education Academic Research Fund Tier 2 (MOE2019- T2- 2- 127, MOE- T2EP50120- 0002, MOE- T2EP50120- 0009, MOE- T2EP50220- 0020 and MOE- T2EP50122- 0005), AcRF Tier 1 (RG57/21, RG156/19 (S)), AcRF Tier 3 (MOE2016- T3- 1- 006 (S)), A\*STAR (1720700038, A1883c0002, A18A7b0058, A20E5c0095, A2083c0062 and A2090b0144), Spanish Ministry for Science and Innovation- Agencia Estatal de Investigación (AEI) through grants CEX2018- 000805- M and PID2021- 125894NB- I00, the Autonomous Community of Madrid, the Spanish government and the European Union through grant MRR Advanced Materials (MAD2D- CM), and National Medical Research Council (NMRC) (021528- 00001). This work was supported by A\*STAR under its IAF- ICP Programme I2001E0067 and the Schaeffler Hub for Advanced Research at NTU. + +<|ref|>sub_title<|/ref|><|det|>[[115, 439, 274, 456]]<|/det|> +## Author contributions + +<|ref|>text<|/ref|><|det|>[[113, 476, 886, 648]]<|/det|> +T. Wu and Y. Luo designed the research project. T. Wu and M. Chen fabricated the flexible plasmonic metasurfaces; C. Wang fabricated the \(\mathrm{WS}_2\) flakes and the bowtie/dimer antennas. T. Wu, L. Liu, and J. Zhao conducted the optical experiments. Z. Wang, Z. Wang, and T. Wu took the SEM and TEM images. T. Wu, G. Hu and L. Liu performed the finite-difference time-domain simulations and theoretical analysis. F. J. Garcia-Vidal, L. Wei, Q. Wang, and Y. Luo supervised the research. T. Wu, G. Hu, and L. Liu analyzed the data; T. Wu and G. Hu wrote the manuscript. All authors contributed to data interpretation and editing the manuscript. + +<|ref|>sub_title<|/ref|><|det|>[[115, 667, 265, 684]]<|/det|> +## Competing interests + +<|ref|>text<|/ref|><|det|>[[115, 705, 416, 722]]<|/det|> +The authors declare no competing interests. + +<--- Page Split ---> +<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|> +## Supplementary Files + +<|ref|>text<|/ref|><|det|>[[44, 93, 765, 112]]<|/det|> +This is a list of supplementary files associated with this preprint. Click to download. + +<|ref|>text<|/ref|><|det|>[[61, 131, 150, 149]]<|/det|> +- Sl.docx + +<--- Page Split ---> diff --git a/preprint/preprint__17898f25cf68f7d8d2b6674d38dfbc329fa67140e161d2e9ae2c314184533b72/images_list.json b/preprint/preprint__17898f25cf68f7d8d2b6674d38dfbc329fa67140e161d2e9ae2c314184533b72/images_list.json new file mode 100644 index 0000000000000000000000000000000000000000..0637a088a01e8ddab3bf3fa98dbe804cbde1a0dc --- /dev/null +++ b/preprint/preprint__17898f25cf68f7d8d2b6674d38dfbc329fa67140e161d2e9ae2c314184533b72/images_list.json @@ -0,0 +1 @@ +[] \ No newline at end of file